2 * Copyright (c) 2000-2003 Silicon Graphics, Inc. All Rights Reserved.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it would be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
12 * Further, this software is distributed without any warranty that it is
13 * free of the rightful claim of any third person regarding infringement
14 * or the like. Any license provided herein, whether implied or
15 * otherwise, applies only to this software file. Patent licenses, if
16 * any, provided herein do not apply to combinations of this program with
17 * other software, or any other product whatsoever.
19 * You should have received a copy of the GNU General Public License along
20 * with this program; if not, write the Free Software Foundation, Inc., 59
21 * Temple Place - Suite 330, Boston MA 02111-1307, USA.
23 * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24 * Mountain View, CA 94043, or:
28 * For further information regarding this notice, see:
30 * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
34 #include "xfs_macros.h"
35 #include "xfs_types.h"
38 #include "xfs_trans.h"
39 #include "xfs_trans_priv.h"
44 #include "xfs_dmapi.h"
45 #include "xfs_mount.h"
46 #include "xfs_alloc_btree.h"
47 #include "xfs_bmap_btree.h"
48 #include "xfs_ialloc_btree.h"
49 #include "xfs_btree.h"
51 #include "xfs_alloc.h"
52 #include "xfs_ialloc.h"
53 #include "xfs_attr_sf.h"
54 #include "xfs_dir_sf.h"
55 #include "xfs_dir2_sf.h"
56 #include "xfs_dinode.h"
57 #include "xfs_inode_item.h"
58 #include "xfs_inode.h"
60 #include "xfs_buf_item.h"
62 #include "xfs_error.h"
64 #include "xfs_utils.h"
65 #include "xfs_dir2_trace.h"
66 #include "xfs_quota.h"
71 kmem_zone_t *xfs_ifork_zone;
72 kmem_zone_t *xfs_inode_zone;
73 kmem_zone_t *xfs_chashlist_zone;
76 * Used in xfs_itruncate(). This is the maximum number of extents
77 * freed from a file in a single transaction.
79 #define XFS_ITRUNC_MAX_EXTENTS 2
81 STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
82 STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
83 STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
84 STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
89 * Make sure that the extents in the given memory buffer
103 for (i = 0; i < nrecs; i++) {
104 rec.l0 = get_unaligned((__uint64_t*)&ep->l0);
105 rec.l1 = get_unaligned((__uint64_t*)&ep->l1);
107 xfs_bmbt_disk_get_all(&rec, &irec);
109 xfs_bmbt_get_all(&rec, &irec);
110 if (fmt == XFS_EXTFMT_NOSTATE)
111 ASSERT(irec.br_state == XFS_EXT_NORM);
116 #define xfs_validate_extents(ep, nrecs, disk, fmt)
120 * Check that none of the inode's in the buffer have a next
121 * unlinked field of 0.
133 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
135 for (i = 0; i < j; i++) {
136 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
137 i * mp->m_sb.sb_inodesize);
138 if (INT_ISZERO(dip->di_next_unlinked, ARCH_CONVERT)) {
139 xfs_fs_cmn_err(CE_ALERT, mp,
140 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
142 ASSERT(!INT_ISZERO(dip->di_next_unlinked, ARCH_CONVERT));
149 * called from bwrite on xfs inode buffers
152 xfs_inobp_bwcheck(xfs_buf_t *bp)
159 ASSERT(XFS_BUF_FSPRIVATE3(bp, void *) != NULL);
161 mp = XFS_BUF_FSPRIVATE3(bp, xfs_mount_t *);
164 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
166 for (i = 0; i < j; i++) {
167 dip = (xfs_dinode_t *) xfs_buf_offset(bp,
168 i * mp->m_sb.sb_inodesize);
169 if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC) {
171 "Bad magic # 0x%x in XFS inode buffer 0x%Lx, starting blockno %Ld, offset 0x%x",
172 INT_GET(dip->di_core.di_magic, ARCH_CONVERT),
173 (__uint64_t)(__psunsigned_t) bp,
174 (__int64_t) XFS_BUF_ADDR(bp),
175 xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize));
176 xfs_fs_cmn_err(CE_WARN, mp,
177 "corrupt, unmount and run xfs_repair");
179 if (INT_ISZERO(dip->di_next_unlinked, ARCH_CONVERT)) {
181 "Bad next_unlinked field (0) in XFS inode buffer 0x%p, starting blockno %Ld, offset 0x%x",
182 (__uint64_t)(__psunsigned_t) bp,
183 (__int64_t) XFS_BUF_ADDR(bp),
184 xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize));
185 xfs_fs_cmn_err(CE_WARN, mp,
186 "corrupt, unmount and run xfs_repair");
194 * This routine is called to map an inode number within a file
195 * system to the buffer containing the on-disk version of the
196 * inode. It returns a pointer to the buffer containing the
197 * on-disk inode in the bpp parameter, and in the dip parameter
198 * it returns a pointer to the on-disk inode within that buffer.
200 * If a non-zero error is returned, then the contents of bpp and
201 * dipp are undefined.
203 * Use xfs_imap() to determine the size and location of the
204 * buffer to read from disk.
222 * Call the space managment code to find the location of the
226 error = xfs_imap(mp, tp, ino, &imap, XFS_IMAP_LOOKUP);
229 "xfs_inotobp: xfs_imap() returned an "
230 "error %d on %s. Returning error.", error, mp->m_fsname);
235 * If the inode number maps to a block outside the bounds of the
236 * file system then return NULL rather than calling read_buf
237 * and panicing when we get an error from the driver.
239 if ((imap.im_blkno + imap.im_len) >
240 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
242 "xfs_inotobp: inode number (%d + %d) maps to a block outside the bounds "
243 "of the file system %s. Returning EINVAL.",
244 imap.im_blkno, imap.im_len,mp->m_fsname);
245 return XFS_ERROR(EINVAL);
249 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
250 * default to just a read_buf() call.
252 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
253 (int)imap.im_len, XFS_BUF_LOCK, &bp);
257 "xfs_inotobp: xfs_trans_read_buf() returned an "
258 "error %d on %s. Returning error.", error, mp->m_fsname);
261 dip = (xfs_dinode_t *)xfs_buf_offset(bp, 0);
263 INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
264 XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
265 if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP,
266 XFS_RANDOM_ITOBP_INOTOBP))) {
267 XFS_CORRUPTION_ERROR("xfs_inotobp", XFS_ERRLEVEL_LOW, mp, dip);
268 xfs_trans_brelse(tp, bp);
270 "xfs_inotobp: XFS_TEST_ERROR() returned an "
271 "error on %s. Returning EFSCORRUPTED.", mp->m_fsname);
272 return XFS_ERROR(EFSCORRUPTED);
275 xfs_inobp_check(mp, bp);
278 * Set *dipp to point to the on-disk inode in the buffer.
280 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
282 *offset = imap.im_boffset;
288 * This routine is called to map an inode to the buffer containing
289 * the on-disk version of the inode. It returns a pointer to the
290 * buffer containing the on-disk inode in the bpp parameter, and in
291 * the dip parameter it returns a pointer to the on-disk inode within
294 * If a non-zero error is returned, then the contents of bpp and
295 * dipp are undefined.
297 * If the inode is new and has not yet been initialized, use xfs_imap()
298 * to determine the size and location of the buffer to read from disk.
299 * If the inode has already been mapped to its buffer and read in once,
300 * then use the mapping information stored in the inode rather than
301 * calling xfs_imap(). This allows us to avoid the overhead of looking
302 * at the inode btree for small block file systems (see xfs_dilocate()).
303 * We can tell whether the inode has been mapped in before by comparing
304 * its disk block address to 0. Only uninitialized inodes will have
305 * 0 for the disk block address.
324 if (ip->i_blkno == (xfs_daddr_t)0) {
326 * Call the space management code to find the location of the
330 error = xfs_imap(mp, tp, ip->i_ino, &imap, XFS_IMAP_LOOKUP);
336 * If the inode number maps to a block outside the bounds
337 * of the file system then return NULL rather than calling
338 * read_buf and panicing when we get an error from the
341 if ((imap.im_blkno + imap.im_len) >
342 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
344 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
345 "(imap.im_blkno (0x%llx) "
346 "+ imap.im_len (0x%llx)) > "
347 " XFS_FSB_TO_BB(mp, "
348 "mp->m_sb.sb_dblocks) (0x%llx)",
349 (unsigned long long) imap.im_blkno,
350 (unsigned long long) imap.im_len,
351 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
353 return XFS_ERROR(EINVAL);
357 * Fill in the fields in the inode that will be used to
358 * map the inode to its buffer from now on.
360 ip->i_blkno = imap.im_blkno;
361 ip->i_len = imap.im_len;
362 ip->i_boffset = imap.im_boffset;
365 * We've already mapped the inode once, so just use the
366 * mapping that we saved the first time.
368 imap.im_blkno = ip->i_blkno;
369 imap.im_len = ip->i_len;
370 imap.im_boffset = ip->i_boffset;
372 ASSERT(bno == 0 || bno == imap.im_blkno);
375 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
376 * default to just a read_buf() call.
378 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
379 (int)imap.im_len, XFS_BUF_LOCK, &bp);
383 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
384 "xfs_trans_read_buf() returned error %d, "
385 "imap.im_blkno 0x%llx, imap.im_len 0x%llx",
386 error, (unsigned long long) imap.im_blkno,
387 (unsigned long long) imap.im_len);
393 * Validate the magic number and version of every inode in the buffer
394 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
397 ni = BBTOB(imap.im_len) >> mp->m_sb.sb_inodelog;
401 for (i = 0; i < ni; i++) {
405 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
406 (i << mp->m_sb.sb_inodelog));
407 di_ok = INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
408 XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
409 if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP,
410 XFS_RANDOM_ITOBP_INOTOBP))) {
412 prdev("bad inode magic/vsn daddr %lld #%d (magic=%x)",
414 (unsigned long long)imap.im_blkno, i,
415 INT_GET(dip->di_core.di_magic, ARCH_CONVERT));
417 XFS_CORRUPTION_ERROR("xfs_itobp", XFS_ERRLEVEL_HIGH,
419 xfs_trans_brelse(tp, bp);
420 return XFS_ERROR(EFSCORRUPTED);
423 #endif /* __KERNEL__ */
425 xfs_inobp_check(mp, bp);
428 * Mark the buffer as an inode buffer now that it looks good
430 XFS_BUF_SET_VTYPE(bp, B_FS_INO);
433 * Set *dipp to point to the on-disk inode in the buffer.
435 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
441 * Move inode type and inode format specific information from the
442 * on-disk inode to the in-core inode. For fifos, devs, and sockets
443 * this means set if_rdev to the proper value. For files, directories,
444 * and symlinks this means to bring in the in-line data or extent
445 * pointers. For a file in B-tree format, only the root is immediately
446 * brought in-core. The rest will be in-lined in if_extents when it
447 * is first referenced (see xfs_iread_extents()).
454 xfs_attr_shortform_t *atp;
458 ip->i_df.if_ext_max =
459 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
463 INT_GET(dip->di_core.di_nextents, ARCH_CONVERT) +
464 INT_GET(dip->di_core.di_anextents, ARCH_CONVERT) >
465 INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT))) {
466 xfs_fs_cmn_err(CE_WARN, ip->i_mount,
467 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu."
468 " Unmount and run xfs_repair.",
469 (unsigned long long)ip->i_ino,
470 (int)(INT_GET(dip->di_core.di_nextents, ARCH_CONVERT)
471 + INT_GET(dip->di_core.di_anextents, ARCH_CONVERT)),
473 INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT));
474 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
476 return XFS_ERROR(EFSCORRUPTED);
479 if (unlikely(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT) > ip->i_mount->m_sb.sb_inodesize)) {
480 xfs_fs_cmn_err(CE_WARN, ip->i_mount,
481 "corrupt dinode %Lu, forkoff = 0x%x."
482 " Unmount and run xfs_repair.",
483 (unsigned long long)ip->i_ino,
484 (int)(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT)));
485 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
487 return XFS_ERROR(EFSCORRUPTED);
490 switch (ip->i_d.di_mode & S_IFMT) {
495 if (unlikely(INT_GET(dip->di_core.di_format, ARCH_CONVERT) != XFS_DINODE_FMT_DEV)) {
496 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
498 return XFS_ERROR(EFSCORRUPTED);
501 ip->i_df.if_u2.if_rdev = INT_GET(dip->di_u.di_dev, ARCH_CONVERT);
507 switch (INT_GET(dip->di_core.di_format, ARCH_CONVERT)) {
508 case XFS_DINODE_FMT_LOCAL:
510 * no local regular files yet
512 if (unlikely((INT_GET(dip->di_core.di_mode, ARCH_CONVERT) & S_IFMT) == S_IFREG)) {
513 xfs_fs_cmn_err(CE_WARN, ip->i_mount,
514 "corrupt inode (local format for regular file) %Lu. Unmount and run xfs_repair.",
515 (unsigned long long) ip->i_ino);
516 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
519 return XFS_ERROR(EFSCORRUPTED);
522 di_size = INT_GET(dip->di_core.di_size, ARCH_CONVERT);
523 if (unlikely(di_size >
524 XFS_DFORK_DSIZE_ARCH(dip, ip->i_mount, ARCH_CONVERT))) {
525 xfs_fs_cmn_err(CE_WARN, ip->i_mount,
526 "corrupt inode %Lu (bad size %Ld for local inode). Unmount and run xfs_repair.",
527 (unsigned long long) ip->i_ino,
528 (long long) di_size);
529 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
532 return XFS_ERROR(EFSCORRUPTED);
536 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
538 case XFS_DINODE_FMT_EXTENTS:
539 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
541 case XFS_DINODE_FMT_BTREE:
542 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
545 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
547 return XFS_ERROR(EFSCORRUPTED);
552 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
553 return XFS_ERROR(EFSCORRUPTED);
558 if (!XFS_DFORK_Q_ARCH(dip, ARCH_CONVERT))
560 ASSERT(ip->i_afp == NULL);
561 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
562 ip->i_afp->if_ext_max =
563 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
564 switch (INT_GET(dip->di_core.di_aformat, ARCH_CONVERT)) {
565 case XFS_DINODE_FMT_LOCAL:
566 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR_ARCH(dip, ARCH_CONVERT);
567 size = (int)INT_GET(atp->hdr.totsize, ARCH_CONVERT);
568 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
570 case XFS_DINODE_FMT_EXTENTS:
571 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
573 case XFS_DINODE_FMT_BTREE:
574 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
577 error = XFS_ERROR(EFSCORRUPTED);
581 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
583 xfs_idestroy_fork(ip, XFS_DATA_FORK);
589 * The file is in-lined in the on-disk inode.
590 * If it fits into if_inline_data, then copy
591 * it there, otherwise allocate a buffer for it
592 * and copy the data there. Either way, set
593 * if_data to point at the data.
594 * If we allocate a buffer for the data, make
595 * sure that its size is a multiple of 4 and
596 * record the real size in i_real_bytes.
609 * If the size is unreasonable, then something
610 * is wrong and we just bail out rather than crash in
611 * kmem_alloc() or memcpy() below.
613 if (unlikely(size > XFS_DFORK_SIZE_ARCH(dip, ip->i_mount, whichfork, ARCH_CONVERT))) {
614 xfs_fs_cmn_err(CE_WARN, ip->i_mount,
615 "corrupt inode %Lu (bad size %d for local fork, size = %d). Unmount and run xfs_repair.",
616 (unsigned long long) ip->i_ino, size,
617 XFS_DFORK_SIZE_ARCH(dip, ip->i_mount, whichfork, ARCH_CONVERT));
618 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
620 return XFS_ERROR(EFSCORRUPTED);
622 ifp = XFS_IFORK_PTR(ip, whichfork);
625 ifp->if_u1.if_data = NULL;
626 else if (size <= sizeof(ifp->if_u2.if_inline_data))
627 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
629 real_size = roundup(size, 4);
630 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
632 ifp->if_bytes = size;
633 ifp->if_real_bytes = real_size;
635 memcpy(ifp->if_u1.if_data,
636 XFS_DFORK_PTR_ARCH(dip, whichfork, ARCH_CONVERT), size);
637 ifp->if_flags &= ~XFS_IFEXTENTS;
638 ifp->if_flags |= XFS_IFINLINE;
643 * The file consists of a set of extents all
644 * of which fit into the on-disk inode.
645 * If there are few enough extents to fit into
646 * the if_inline_ext, then copy them there.
647 * Otherwise allocate a buffer for them and copy
648 * them into it. Either way, set if_extents
649 * to point at the extents.
657 xfs_bmbt_rec_t *ep, *dp;
664 ifp = XFS_IFORK_PTR(ip, whichfork);
665 nex = XFS_DFORK_NEXTENTS_ARCH(dip, whichfork, ARCH_CONVERT);
666 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
669 * If the number of extents is unreasonable, then something
670 * is wrong and we just bail out rather than crash in
671 * kmem_alloc() or memcpy() below.
673 if (unlikely(size < 0 || size > XFS_DFORK_SIZE_ARCH(dip, ip->i_mount, whichfork, ARCH_CONVERT))) {
674 xfs_fs_cmn_err(CE_WARN, ip->i_mount,
675 "corrupt inode %Lu ((a)extents = %d). Unmount and run xfs_repair.",
676 (unsigned long long) ip->i_ino, nex);
677 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
679 return XFS_ERROR(EFSCORRUPTED);
684 ifp->if_u1.if_extents = NULL;
685 else if (nex <= XFS_INLINE_EXTS)
686 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
688 ifp->if_u1.if_extents = kmem_alloc(size, KM_SLEEP);
689 ASSERT(ifp->if_u1.if_extents != NULL);
692 ifp->if_bytes = size;
693 ifp->if_real_bytes = real_size;
695 dp = (xfs_bmbt_rec_t *)
696 XFS_DFORK_PTR_ARCH(dip, whichfork, ARCH_CONVERT);
697 xfs_validate_extents(dp, nex, 1, XFS_EXTFMT_INODE(ip));
698 ep = ifp->if_u1.if_extents;
699 for (i = 0; i < nex; i++, ep++, dp++) {
700 ep->l0 = INT_GET(get_unaligned((__uint64_t*)&dp->l0),
702 ep->l1 = INT_GET(get_unaligned((__uint64_t*)&dp->l1),
705 xfs_bmap_trace_exlist("xfs_iformat_extents", ip, nex,
707 if (whichfork != XFS_DATA_FORK ||
708 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
709 if (unlikely(xfs_check_nostate_extents(
710 ifp->if_u1.if_extents, nex))) {
711 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
714 return XFS_ERROR(EFSCORRUPTED);
717 ifp->if_flags |= XFS_IFEXTENTS;
722 * The file has too many extents to fit into
723 * the inode, so they are in B-tree format.
724 * Allocate a buffer for the root of the B-tree
725 * and copy the root into it. The i_extents
726 * field will remain NULL until all of the
727 * extents are read in (when they are needed).
735 xfs_bmdr_block_t *dfp;
741 ifp = XFS_IFORK_PTR(ip, whichfork);
742 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR_ARCH(dip, whichfork, ARCH_CONVERT);
743 size = XFS_BMAP_BROOT_SPACE(dfp);
744 nrecs = XFS_BMAP_BROOT_NUMRECS(dfp);
747 * blow out if -- fork has less extents than can fit in
748 * fork (fork shouldn't be a btree format), root btree
749 * block has more records than can fit into the fork,
750 * or the number of extents is greater than the number of
753 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
754 || XFS_BMDR_SPACE_CALC(nrecs) >
755 XFS_DFORK_SIZE_ARCH(dip, ip->i_mount, whichfork, ARCH_CONVERT)
756 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
757 xfs_fs_cmn_err(CE_WARN, ip->i_mount,
758 "corrupt inode %Lu (btree). Unmount and run xfs_repair.",
759 (unsigned long long) ip->i_ino);
760 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
762 return XFS_ERROR(EFSCORRUPTED);
765 ifp->if_broot_bytes = size;
766 ifp->if_broot = kmem_alloc(size, KM_SLEEP);
767 ASSERT(ifp->if_broot != NULL);
769 * Copy and convert from the on-disk structure
770 * to the in-memory structure.
772 xfs_bmdr_to_bmbt(dfp, XFS_DFORK_SIZE_ARCH(dip, ip->i_mount, whichfork, ARCH_CONVERT),
773 ifp->if_broot, size);
774 ifp->if_flags &= ~XFS_IFEXTENTS;
775 ifp->if_flags |= XFS_IFBROOT;
781 * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk
784 * buf = on-disk representation
785 * dip = native representation
786 * dir = direction - +ve -> disk to native
787 * -ve -> native to disk
788 * arch = on-disk architecture
791 xfs_xlate_dinode_core(
793 xfs_dinode_core_t *dip,
797 xfs_dinode_core_t *buf_core = (xfs_dinode_core_t *)buf;
798 xfs_dinode_core_t *mem_core = (xfs_dinode_core_t *)dip;
801 if (arch == ARCH_NOCONVERT) {
803 memcpy((xfs_caddr_t)mem_core, (xfs_caddr_t)buf_core,
804 sizeof(xfs_dinode_core_t));
806 memcpy((xfs_caddr_t)buf_core, (xfs_caddr_t)mem_core,
807 sizeof(xfs_dinode_core_t));
812 INT_XLATE(buf_core->di_magic, mem_core->di_magic, dir, arch);
813 INT_XLATE(buf_core->di_mode, mem_core->di_mode, dir, arch);
814 INT_XLATE(buf_core->di_version, mem_core->di_version, dir, arch);
815 INT_XLATE(buf_core->di_format, mem_core->di_format, dir, arch);
816 INT_XLATE(buf_core->di_onlink, mem_core->di_onlink, dir, arch);
817 INT_XLATE(buf_core->di_uid, mem_core->di_uid, dir, arch);
818 INT_XLATE(buf_core->di_gid, mem_core->di_gid, dir, arch);
819 INT_XLATE(buf_core->di_nlink, mem_core->di_nlink, dir, arch);
820 INT_XLATE(buf_core->di_projid, mem_core->di_projid, dir, arch);
823 memcpy(mem_core->di_pad, buf_core->di_pad,
824 sizeof(buf_core->di_pad));
826 memcpy(buf_core->di_pad, mem_core->di_pad,
827 sizeof(buf_core->di_pad));
830 INT_XLATE(buf_core->di_flushiter, mem_core->di_flushiter, dir, arch);
832 INT_XLATE(buf_core->di_atime.t_sec, mem_core->di_atime.t_sec,
834 INT_XLATE(buf_core->di_atime.t_nsec, mem_core->di_atime.t_nsec,
836 INT_XLATE(buf_core->di_mtime.t_sec, mem_core->di_mtime.t_sec,
838 INT_XLATE(buf_core->di_mtime.t_nsec, mem_core->di_mtime.t_nsec,
840 INT_XLATE(buf_core->di_ctime.t_sec, mem_core->di_ctime.t_sec,
842 INT_XLATE(buf_core->di_ctime.t_nsec, mem_core->di_ctime.t_nsec,
844 INT_XLATE(buf_core->di_size, mem_core->di_size, dir, arch);
845 INT_XLATE(buf_core->di_nblocks, mem_core->di_nblocks, dir, arch);
846 INT_XLATE(buf_core->di_extsize, mem_core->di_extsize, dir, arch);
847 INT_XLATE(buf_core->di_nextents, mem_core->di_nextents, dir, arch);
848 INT_XLATE(buf_core->di_anextents, mem_core->di_anextents, dir, arch);
849 INT_XLATE(buf_core->di_forkoff, mem_core->di_forkoff, dir, arch);
850 INT_XLATE(buf_core->di_aformat, mem_core->di_aformat, dir, arch);
851 INT_XLATE(buf_core->di_dmevmask, mem_core->di_dmevmask, dir, arch);
852 INT_XLATE(buf_core->di_dmstate, mem_core->di_dmstate, dir, arch);
853 INT_XLATE(buf_core->di_flags, mem_core->di_flags, dir, arch);
854 INT_XLATE(buf_core->di_gen, mem_core->di_gen, dir, arch);
858 * Given a mount structure and an inode number, return a pointer
859 * to a newly allocated in-core inode coresponding to the given
862 * Initialize the inode's attributes and extent pointers if it
863 * already has them (it will not if the inode has no links).
878 ASSERT(xfs_inode_zone != NULL);
880 ip = kmem_zone_zalloc(xfs_inode_zone, KM_SLEEP);
885 * Get pointer's to the on-disk inode and the buffer containing it.
886 * If the inode number refers to a block outside the file system
887 * then xfs_itobp() will return NULL. In this case we should
888 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
889 * know that this is a new incore inode.
891 error = xfs_itobp(mp, tp, ip, &dip, &bp, bno);
894 kmem_zone_free(xfs_inode_zone, ip);
899 * Initialize inode's trace buffers.
900 * Do this before xfs_iformat in case it adds entries.
902 #ifdef XFS_BMAP_TRACE
903 ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_SLEEP);
905 #ifdef XFS_BMBT_TRACE
906 ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_SLEEP);
909 ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_SLEEP);
911 #ifdef XFS_ILOCK_TRACE
912 ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_SLEEP);
914 #ifdef XFS_DIR2_TRACE
915 ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_SLEEP);
919 * If we got something that isn't an inode it means someone
920 * (nfs or dmi) has a stale handle.
922 if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC) {
923 kmem_zone_free(xfs_inode_zone, ip);
924 xfs_trans_brelse(tp, bp);
926 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
927 "dip->di_core.di_magic (0x%x) != "
928 "XFS_DINODE_MAGIC (0x%x)",
929 INT_GET(dip->di_core.di_magic, ARCH_CONVERT),
932 return XFS_ERROR(EINVAL);
936 * If the on-disk inode is already linked to a directory
937 * entry, copy all of the inode into the in-core inode.
938 * xfs_iformat() handles copying in the inode format
939 * specific information.
940 * Otherwise, just get the truly permanent information.
942 if (!INT_ISZERO(dip->di_core.di_mode, ARCH_CONVERT)) {
943 xfs_xlate_dinode_core((xfs_caddr_t)&dip->di_core,
944 &(ip->i_d), 1, ARCH_CONVERT);
945 error = xfs_iformat(ip, dip);
947 kmem_zone_free(xfs_inode_zone, ip);
948 xfs_trans_brelse(tp, bp);
950 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
951 "xfs_iformat() returned error %d",
957 ip->i_d.di_magic = INT_GET(dip->di_core.di_magic, ARCH_CONVERT);
958 ip->i_d.di_version = INT_GET(dip->di_core.di_version, ARCH_CONVERT);
959 ip->i_d.di_gen = INT_GET(dip->di_core.di_gen, ARCH_CONVERT);
960 ip->i_d.di_flushiter = INT_GET(dip->di_core.di_flushiter, ARCH_CONVERT);
962 * Make sure to pull in the mode here as well in
963 * case the inode is released without being used.
964 * This ensures that xfs_inactive() will see that
965 * the inode is already free and not try to mess
966 * with the uninitialized part of it.
970 * Initialize the per-fork minima and maxima for a new
971 * inode here. xfs_iformat will do it for old inodes.
973 ip->i_df.if_ext_max =
974 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
977 INIT_LIST_HEAD(&ip->i_reclaim);
980 * The inode format changed when we moved the link count and
981 * made it 32 bits long. If this is an old format inode,
982 * convert it in memory to look like a new one. If it gets
983 * flushed to disk we will convert back before flushing or
984 * logging it. We zero out the new projid field and the old link
985 * count field. We'll handle clearing the pad field (the remains
986 * of the old uuid field) when we actually convert the inode to
987 * the new format. We don't change the version number so that we
988 * can distinguish this from a real new format inode.
990 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
991 ip->i_d.di_nlink = ip->i_d.di_onlink;
992 ip->i_d.di_onlink = 0;
993 ip->i_d.di_projid = 0;
996 ip->i_delayed_blks = 0;
999 * Mark the buffer containing the inode as something to keep
1000 * around for a while. This helps to keep recently accessed
1001 * meta-data in-core longer.
1003 XFS_BUF_SET_REF(bp, XFS_INO_REF);
1006 * Use xfs_trans_brelse() to release the buffer containing the
1007 * on-disk inode, because it was acquired with xfs_trans_read_buf()
1008 * in xfs_itobp() above. If tp is NULL, this is just a normal
1009 * brelse(). If we're within a transaction, then xfs_trans_brelse()
1010 * will only release the buffer if it is not dirty within the
1011 * transaction. It will be OK to release the buffer in this case,
1012 * because inodes on disk are never destroyed and we will be
1013 * locking the new in-core inode before putting it in the hash
1014 * table where other processes can find it. Thus we don't have
1015 * to worry about the inode being changed just because we released
1018 xfs_trans_brelse(tp, bp);
1024 * Read in extents from a btree-format inode.
1025 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1037 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
1038 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
1040 return XFS_ERROR(EFSCORRUPTED);
1042 size = XFS_IFORK_NEXTENTS(ip, whichfork) * (uint)sizeof(xfs_bmbt_rec_t);
1043 ifp = XFS_IFORK_PTR(ip, whichfork);
1045 * We know that the size is valid (it's checked in iformat_btree)
1047 ifp->if_u1.if_extents = kmem_alloc(size, KM_SLEEP);
1048 ASSERT(ifp->if_u1.if_extents != NULL);
1049 ifp->if_lastex = NULLEXTNUM;
1050 ifp->if_bytes = ifp->if_real_bytes = (int)size;
1051 ifp->if_flags |= XFS_IFEXTENTS;
1052 error = xfs_bmap_read_extents(tp, ip, whichfork);
1054 kmem_free(ifp->if_u1.if_extents, size);
1055 ifp->if_u1.if_extents = NULL;
1056 ifp->if_bytes = ifp->if_real_bytes = 0;
1057 ifp->if_flags &= ~XFS_IFEXTENTS;
1060 xfs_validate_extents((xfs_bmbt_rec_t *)ifp->if_u1.if_extents,
1061 XFS_IFORK_NEXTENTS(ip, whichfork), 0, XFS_EXTFMT_INODE(ip));
1066 * Allocate an inode on disk and return a copy of its in-core version.
1067 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1068 * appropriately within the inode. The uid and gid for the inode are
1069 * set according to the contents of the given cred structure.
1071 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1072 * has a free inode available, call xfs_iget()
1073 * to obtain the in-core version of the allocated inode. Finally,
1074 * fill in the inode and log its initial contents. In this case,
1075 * ialloc_context would be set to NULL and call_again set to false.
1077 * If xfs_dialloc() does not have an available inode,
1078 * it will replenish its supply by doing an allocation. Since we can
1079 * only do one allocation within a transaction without deadlocks, we
1080 * must commit the current transaction before returning the inode itself.
1081 * In this case, therefore, we will set call_again to true and return.
1082 * The caller should then commit the current transaction, start a new
1083 * transaction, and call xfs_ialloc() again to actually get the inode.
1085 * To ensure that some other process does not grab the inode that
1086 * was allocated during the first call to xfs_ialloc(), this routine
1087 * also returns the [locked] bp pointing to the head of the freelist
1088 * as ialloc_context. The caller should hold this buffer across
1089 * the commit and pass it back into this routine on the second call.
1101 xfs_buf_t **ialloc_context,
1102 boolean_t *call_again,
1112 * Call the space management code to pick
1113 * the on-disk inode to be allocated.
1115 ASSERT(pip != NULL);
1116 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
1117 ialloc_context, call_again, &ino);
1121 if (*call_again || ino == NULLFSINO) {
1125 ASSERT(*ialloc_context == NULL);
1128 * Get the in-core inode with the lock held exclusively.
1129 * This is because we're setting fields here we need
1130 * to prevent others from looking at until we're done.
1132 error = xfs_trans_iget(tp->t_mountp, tp, ino, XFS_ILOCK_EXCL, &ip);
1139 vp->v_type = IFTOVT(mode);
1140 ip->i_d.di_mode = (__uint16_t)mode;
1141 ip->i_d.di_onlink = 0;
1142 ip->i_d.di_nlink = nlink;
1143 ASSERT(ip->i_d.di_nlink == nlink);
1144 ip->i_d.di_uid = current_fsuid(cr);
1145 ip->i_d.di_gid = current_fsgid(cr);
1146 ip->i_d.di_projid = prid;
1147 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1150 * If the superblock version is up to where we support new format
1151 * inodes and this is currently an old format inode, then change
1152 * the inode version number now. This way we only do the conversion
1153 * here rather than here and in the flush/logging code.
1155 if (XFS_SB_VERSION_HASNLINK(&tp->t_mountp->m_sb) &&
1156 ip->i_d.di_version == XFS_DINODE_VERSION_1) {
1157 ip->i_d.di_version = XFS_DINODE_VERSION_2;
1159 * We've already zeroed the old link count, the projid field,
1160 * and the pad field.
1165 * Project ids won't be stored on disk if we are using a version 1 inode.
1167 if ( (prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1))
1168 xfs_bump_ino_vers2(tp, ip);
1170 if (XFS_INHERIT_GID(pip, vp->v_vfsp)) {
1171 ip->i_d.di_gid = pip->i_d.di_gid;
1172 if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
1173 ip->i_d.di_mode |= S_ISGID;
1178 * If the group ID of the new file does not match the effective group
1179 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1180 * (and only if the irix_sgid_inherit compatibility variable is set).
1182 if ((irix_sgid_inherit) &&
1183 (ip->i_d.di_mode & S_ISGID) &&
1184 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1185 ip->i_d.di_mode &= ~S_ISGID;
1188 ip->i_d.di_size = 0;
1189 ip->i_d.di_nextents = 0;
1190 ASSERT(ip->i_d.di_nblocks == 0);
1191 xfs_ichgtime(ip, XFS_ICHGTIME_CHG|XFS_ICHGTIME_ACC|XFS_ICHGTIME_MOD);
1193 * di_gen will have been taken care of in xfs_iread.
1195 ip->i_d.di_extsize = 0;
1196 ip->i_d.di_dmevmask = 0;
1197 ip->i_d.di_dmstate = 0;
1198 ip->i_d.di_flags = 0;
1199 flags = XFS_ILOG_CORE;
1200 switch (mode & S_IFMT) {
1205 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1206 ip->i_df.if_u2.if_rdev = rdev;
1207 ip->i_df.if_flags = 0;
1208 flags |= XFS_ILOG_DEV;
1212 if (pip->i_d.di_flags &
1213 (XFS_DIFLAG_NOATIME|XFS_DIFLAG_NODUMP|XFS_DIFLAG_SYNC)) {
1214 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1215 xfs_inherit_noatime)
1216 ip->i_d.di_flags |= XFS_DIFLAG_NOATIME;
1217 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1219 ip->i_d.di_flags |= XFS_DIFLAG_NODUMP;
1220 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1222 ip->i_d.di_flags |= XFS_DIFLAG_SYNC;
1225 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1226 ip->i_df.if_flags = XFS_IFEXTENTS;
1227 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1228 ip->i_df.if_u1.if_extents = NULL;
1234 * Attribute fork settings for new inode.
1236 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1237 ip->i_d.di_anextents = 0;
1240 * Log the new values stuffed into the inode.
1242 xfs_trans_log_inode(tp, ip, flags);
1244 /* now that we have a v_type we can set Linux inode ops (& unlock) */
1245 VFS_INIT_VNODE(XFS_MTOVFS(tp->t_mountp), vp, XFS_ITOBHV(ip), 1);
1252 * Check to make sure that there are no blocks allocated to the
1253 * file beyond the size of the file. We don't check this for
1254 * files with fixed size extents or real time extents, but we
1255 * at least do it for regular files.
1264 xfs_fileoff_t map_first;
1266 xfs_bmbt_irec_t imaps[2];
1268 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1271 if ( ip->i_d.di_flags & XFS_DIFLAG_REALTIME )
1275 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1277 * The filesystem could be shutting down, so bmapi may return
1280 if (xfs_bmapi(NULL, ip, map_first,
1282 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1284 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
1287 ASSERT(nimaps == 1);
1288 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1293 * Calculate the last possible buffered byte in a file. This must
1294 * include data that was buffered beyond the EOF by the write code.
1295 * This also needs to deal with overflowing the xfs_fsize_t type
1296 * which can happen for sizes near the limit.
1298 * We also need to take into account any blocks beyond the EOF. It
1299 * may be the case that they were buffered by a write which failed.
1300 * In that case the pages will still be in memory, but the inode size
1301 * will never have been updated.
1308 xfs_fsize_t last_byte;
1309 xfs_fileoff_t last_block;
1310 xfs_fileoff_t size_last_block;
1313 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE | MR_ACCESS));
1317 * Only check for blocks beyond the EOF if the extents have
1318 * been read in. This eliminates the need for the inode lock,
1319 * and it also saves us from looking when it really isn't
1322 if (ip->i_df.if_flags & XFS_IFEXTENTS) {
1323 error = xfs_bmap_last_offset(NULL, ip, &last_block,
1331 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_d.di_size);
1332 last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1334 last_byte = XFS_FSB_TO_B(mp, last_block);
1335 if (last_byte < 0) {
1336 return XFS_MAXIOFFSET(mp);
1338 last_byte += (1 << mp->m_writeio_log);
1339 if (last_byte < 0) {
1340 return XFS_MAXIOFFSET(mp);
1345 #if defined(XFS_RW_TRACE)
1351 xfs_fsize_t new_size,
1352 xfs_off_t toss_start,
1353 xfs_off_t toss_finish)
1355 if (ip->i_rwtrace == NULL) {
1359 ktrace_enter(ip->i_rwtrace,
1362 (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff),
1363 (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff),
1364 (void*)((long)flag),
1365 (void*)(unsigned long)((new_size >> 32) & 0xffffffff),
1366 (void*)(unsigned long)(new_size & 0xffffffff),
1367 (void*)(unsigned long)((toss_start >> 32) & 0xffffffff),
1368 (void*)(unsigned long)(toss_start & 0xffffffff),
1369 (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff),
1370 (void*)(unsigned long)(toss_finish & 0xffffffff),
1371 (void*)(unsigned long)current_cpu(),
1378 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1382 * Start the truncation of the file to new_size. The new size
1383 * must be smaller than the current size. This routine will
1384 * clear the buffer and page caches of file data in the removed
1385 * range, and xfs_itruncate_finish() will remove the underlying
1388 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1389 * must NOT have the inode lock held at all. This is because we're
1390 * calling into the buffer/page cache code and we can't hold the
1391 * inode lock when we do so.
1393 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1394 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1395 * in the case that the caller is locking things out of order and
1396 * may not be able to call xfs_itruncate_finish() with the inode lock
1397 * held without dropping the I/O lock. If the caller must drop the
1398 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1399 * must be called again with all the same restrictions as the initial
1403 xfs_itruncate_start(
1406 xfs_fsize_t new_size)
1408 xfs_fsize_t last_byte;
1409 xfs_off_t toss_start;
1413 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1414 ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size));
1415 ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1416 (flags == XFS_ITRUNC_MAYBE));
1421 * Call VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES() to get rid of pages and buffers
1422 * overlapping the region being removed. We have to use
1423 * the less efficient VOP_FLUSHINVAL_PAGES() in the case that the
1424 * caller may not be able to finish the truncate without
1425 * dropping the inode's I/O lock. Make sure
1426 * to catch any pages brought in by buffers overlapping
1427 * the EOF by searching out beyond the isize by our
1428 * block size. We round new_size up to a block boundary
1429 * so that we don't toss things on the same block as
1430 * new_size but before it.
1432 * Before calling VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES(), make sure to
1433 * call remapf() over the same region if the file is mapped.
1434 * This frees up mapped file references to the pages in the
1435 * given range and for the VOP_FLUSHINVAL_PAGES() case it ensures
1436 * that we get the latest mapped changes flushed out.
1438 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1439 toss_start = XFS_FSB_TO_B(mp, toss_start);
1440 if (toss_start < 0) {
1442 * The place to start tossing is beyond our maximum
1443 * file size, so there is no way that the data extended
1448 last_byte = xfs_file_last_byte(ip);
1449 xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
1451 if (last_byte > toss_start) {
1452 if (flags & XFS_ITRUNC_DEFINITE) {
1453 VOP_TOSS_PAGES(vp, toss_start, -1, FI_REMAPF_LOCKED);
1455 VOP_FLUSHINVAL_PAGES(vp, toss_start, -1, FI_REMAPF_LOCKED);
1460 if (new_size == 0) {
1461 ASSERT(VN_CACHED(vp) == 0);
1467 * Shrink the file to the given new_size. The new
1468 * size must be smaller than the current size.
1469 * This will free up the underlying blocks
1470 * in the removed range after a call to xfs_itruncate_start()
1471 * or xfs_atruncate_start().
1473 * The transaction passed to this routine must have made
1474 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1475 * This routine may commit the given transaction and
1476 * start new ones, so make sure everything involved in
1477 * the transaction is tidy before calling here.
1478 * Some transaction will be returned to the caller to be
1479 * committed. The incoming transaction must already include
1480 * the inode, and both inode locks must be held exclusively.
1481 * The inode must also be "held" within the transaction. On
1482 * return the inode will be "held" within the returned transaction.
1483 * This routine does NOT require any disk space to be reserved
1484 * for it within the transaction.
1486 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1487 * and it indicates the fork which is to be truncated. For the
1488 * attribute fork we only support truncation to size 0.
1490 * We use the sync parameter to indicate whether or not the first
1491 * transaction we perform might have to be synchronous. For the attr fork,
1492 * it needs to be so if the unlink of the inode is not yet known to be
1493 * permanent in the log. This keeps us from freeing and reusing the
1494 * blocks of the attribute fork before the unlink of the inode becomes
1497 * For the data fork, we normally have to run synchronously if we're
1498 * being called out of the inactive path or we're being called
1499 * out of the create path where we're truncating an existing file.
1500 * Either way, the truncate needs to be sync so blocks don't reappear
1501 * in the file with altered data in case of a crash. wsync filesystems
1502 * can run the first case async because anything that shrinks the inode
1503 * has to run sync so by the time we're called here from inactive, the
1504 * inode size is permanently set to 0.
1506 * Calls from the truncate path always need to be sync unless we're
1507 * in a wsync filesystem and the file has already been unlinked.
1509 * The caller is responsible for correctly setting the sync parameter.
1510 * It gets too hard for us to guess here which path we're being called
1511 * out of just based on inode state.
1514 xfs_itruncate_finish(
1517 xfs_fsize_t new_size,
1521 xfs_fsblock_t first_block;
1522 xfs_fileoff_t first_unmap_block;
1523 xfs_fileoff_t last_block;
1524 xfs_filblks_t unmap_len=0;
1529 xfs_bmap_free_t free_list;
1532 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1533 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
1534 ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size));
1535 ASSERT(*tp != NULL);
1536 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1537 ASSERT(ip->i_transp == *tp);
1538 ASSERT(ip->i_itemp != NULL);
1539 ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
1543 mp = (ntp)->t_mountp;
1544 ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1547 * We only support truncating the entire attribute fork.
1549 if (fork == XFS_ATTR_FORK) {
1552 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1553 xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0);
1555 * The first thing we do is set the size to new_size permanently
1556 * on disk. This way we don't have to worry about anyone ever
1557 * being able to look at the data being freed even in the face
1558 * of a crash. What we're getting around here is the case where
1559 * we free a block, it is allocated to another file, it is written
1560 * to, and then we crash. If the new data gets written to the
1561 * file but the log buffers containing the free and reallocation
1562 * don't, then we'd end up with garbage in the blocks being freed.
1563 * As long as we make the new_size permanent before actually
1564 * freeing any blocks it doesn't matter if they get writtten to.
1566 * The callers must signal into us whether or not the size
1567 * setting here must be synchronous. There are a few cases
1568 * where it doesn't have to be synchronous. Those cases
1569 * occur if the file is unlinked and we know the unlink is
1570 * permanent or if the blocks being truncated are guaranteed
1571 * to be beyond the inode eof (regardless of the link count)
1572 * and the eof value is permanent. Both of these cases occur
1573 * only on wsync-mounted filesystems. In those cases, we're
1574 * guaranteed that no user will ever see the data in the blocks
1575 * that are being truncated so the truncate can run async.
1576 * In the free beyond eof case, the file may wind up with
1577 * more blocks allocated to it than it needs if we crash
1578 * and that won't get fixed until the next time the file
1579 * is re-opened and closed but that's ok as that shouldn't
1580 * be too many blocks.
1582 * However, we can't just make all wsync xactions run async
1583 * because there's one call out of the create path that needs
1584 * to run sync where it's truncating an existing file to size
1585 * 0 whose size is > 0.
1587 * It's probably possible to come up with a test in this
1588 * routine that would correctly distinguish all the above
1589 * cases from the values of the function parameters and the
1590 * inode state but for sanity's sake, I've decided to let the
1591 * layers above just tell us. It's simpler to correctly figure
1592 * out in the layer above exactly under what conditions we
1593 * can run async and I think it's easier for others read and
1594 * follow the logic in case something has to be changed.
1595 * cscope is your friend -- rcc.
1597 * The attribute fork is much simpler.
1599 * For the attribute fork we allow the caller to tell us whether
1600 * the unlink of the inode that led to this call is yet permanent
1601 * in the on disk log. If it is not and we will be freeing extents
1602 * in this inode then we make the first transaction synchronous
1603 * to make sure that the unlink is permanent by the time we free
1606 if (fork == XFS_DATA_FORK) {
1607 if (ip->i_d.di_nextents > 0) {
1608 ip->i_d.di_size = new_size;
1609 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1612 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1613 if (ip->i_d.di_anextents > 0)
1614 xfs_trans_set_sync(ntp);
1616 ASSERT(fork == XFS_DATA_FORK ||
1617 (fork == XFS_ATTR_FORK &&
1618 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1619 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1622 * Since it is possible for space to become allocated beyond
1623 * the end of the file (in a crash where the space is allocated
1624 * but the inode size is not yet updated), simply remove any
1625 * blocks which show up between the new EOF and the maximum
1626 * possible file size. If the first block to be removed is
1627 * beyond the maximum file size (ie it is the same as last_block),
1628 * then there is nothing to do.
1630 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1631 ASSERT(first_unmap_block <= last_block);
1633 if (last_block == first_unmap_block) {
1636 unmap_len = last_block - first_unmap_block + 1;
1640 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1641 * will tell us whether it freed the entire range or
1642 * not. If this is a synchronous mount (wsync),
1643 * then we can tell bunmapi to keep all the
1644 * transactions asynchronous since the unlink
1645 * transaction that made this inode inactive has
1646 * already hit the disk. There's no danger of
1647 * the freed blocks being reused, there being a
1648 * crash, and the reused blocks suddenly reappearing
1649 * in this file with garbage in them once recovery
1652 XFS_BMAP_INIT(&free_list, &first_block);
1653 error = xfs_bunmapi(ntp, ip, first_unmap_block,
1655 XFS_BMAPI_AFLAG(fork) |
1656 (sync ? 0 : XFS_BMAPI_ASYNC),
1657 XFS_ITRUNC_MAX_EXTENTS,
1658 &first_block, &free_list, &done);
1661 * If the bunmapi call encounters an error,
1662 * return to the caller where the transaction
1663 * can be properly aborted. We just need to
1664 * make sure we're not holding any resources
1665 * that we were not when we came in.
1667 xfs_bmap_cancel(&free_list);
1672 * Duplicate the transaction that has the permanent
1673 * reservation and commit the old transaction.
1675 error = xfs_bmap_finish(tp, &free_list, first_block,
1680 * If the bmap finish call encounters an error,
1681 * return to the caller where the transaction
1682 * can be properly aborted. We just need to
1683 * make sure we're not holding any resources
1684 * that we were not when we came in.
1686 * Aborting from this point might lose some
1687 * blocks in the file system, but oh well.
1689 xfs_bmap_cancel(&free_list);
1692 * If the passed in transaction committed
1693 * in xfs_bmap_finish(), then we want to
1694 * add the inode to this one before returning.
1695 * This keeps things simple for the higher
1696 * level code, because it always knows that
1697 * the inode is locked and held in the
1698 * transaction that returns to it whether
1699 * errors occur or not. We don't mark the
1700 * inode dirty so that this transaction can
1701 * be easily aborted if possible.
1703 xfs_trans_ijoin(ntp, ip,
1704 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1705 xfs_trans_ihold(ntp, ip);
1712 * The first xact was committed,
1713 * so add the inode to the new one.
1714 * Mark it dirty so it will be logged
1715 * and moved forward in the log as
1716 * part of every commit.
1718 xfs_trans_ijoin(ntp, ip,
1719 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1720 xfs_trans_ihold(ntp, ip);
1721 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1723 ntp = xfs_trans_dup(ntp);
1724 (void) xfs_trans_commit(*tp, 0, NULL);
1726 error = xfs_trans_reserve(ntp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1727 XFS_TRANS_PERM_LOG_RES,
1728 XFS_ITRUNCATE_LOG_COUNT);
1730 * Add the inode being truncated to the next chained
1733 xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1734 xfs_trans_ihold(ntp, ip);
1739 * Only update the size in the case of the data fork, but
1740 * always re-log the inode so that our permanent transaction
1741 * can keep on rolling it forward in the log.
1743 if (fork == XFS_DATA_FORK) {
1744 xfs_isize_check(mp, ip, new_size);
1745 ip->i_d.di_size = new_size;
1747 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1748 ASSERT((new_size != 0) ||
1749 (fork == XFS_ATTR_FORK) ||
1750 (ip->i_delayed_blks == 0));
1751 ASSERT((new_size != 0) ||
1752 (fork == XFS_ATTR_FORK) ||
1753 (ip->i_d.di_nextents == 0));
1754 xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0);
1762 * Do the first part of growing a file: zero any data in the last
1763 * block that is beyond the old EOF. We need to do this before
1764 * the inode is joined to the transaction to modify the i_size.
1765 * That way we can drop the inode lock and call into the buffer
1766 * cache to get the buffer mapping the EOF.
1771 xfs_fsize_t new_size,
1777 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1778 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1779 ASSERT(new_size > ip->i_d.di_size);
1782 isize = ip->i_d.di_size;
1784 * Zero any pages that may have been created by
1785 * xfs_write_file() beyond the end of the file
1786 * and any blocks between the old and new file sizes.
1788 error = xfs_zero_eof(XFS_ITOV(ip), &ip->i_iocore, new_size, isize,
1796 * This routine is called to extend the size of a file.
1797 * The inode must have both the iolock and the ilock locked
1798 * for update and it must be a part of the current transaction.
1799 * The xfs_igrow_start() function must have been called previously.
1800 * If the change_flag is not zero, the inode change timestamp will
1807 xfs_fsize_t new_size,
1810 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1811 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1812 ASSERT(ip->i_transp == tp);
1813 ASSERT(new_size > ip->i_d.di_size);
1816 * Update the file size. Update the inode change timestamp
1817 * if change_flag set.
1819 ip->i_d.di_size = new_size;
1821 xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
1822 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1828 * This is called when the inode's link count goes to 0.
1829 * We place the on-disk inode on a list in the AGI. It
1830 * will be pulled from this list when the inode is freed.
1842 xfs_agnumber_t agno;
1843 xfs_daddr_t agdaddr;
1850 ASSERT(ip->i_d.di_nlink == 0);
1851 ASSERT(ip->i_d.di_mode != 0);
1852 ASSERT(ip->i_transp == tp);
1856 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1857 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1860 * Get the agi buffer first. It ensures lock ordering
1863 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1864 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1869 * Validate the magic number of the agi block.
1871 agi = XFS_BUF_TO_AGI(agibp);
1873 INT_GET(agi->agi_magicnum, ARCH_CONVERT) == XFS_AGI_MAGIC &&
1874 XFS_AGI_GOOD_VERSION(INT_GET(agi->agi_versionnum, ARCH_CONVERT));
1875 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK,
1876 XFS_RANDOM_IUNLINK))) {
1877 XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi);
1878 xfs_trans_brelse(tp, agibp);
1879 return XFS_ERROR(EFSCORRUPTED);
1882 * Get the index into the agi hash table for the
1883 * list this inode will go on.
1885 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1887 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1888 ASSERT(!INT_ISZERO(agi->agi_unlinked[bucket_index], ARCH_CONVERT));
1889 ASSERT(INT_GET(agi->agi_unlinked[bucket_index], ARCH_CONVERT) != agino);
1891 if (INT_GET(agi->agi_unlinked[bucket_index], ARCH_CONVERT) != NULLAGINO) {
1893 * There is already another inode in the bucket we need
1894 * to add ourselves to. Add us at the front of the list.
1895 * Here we put the head pointer into our next pointer,
1896 * and then we fall through to point the head at us.
1898 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0);
1902 ASSERT(INT_GET(dip->di_next_unlinked, ARCH_CONVERT) == NULLAGINO);
1903 ASSERT(!INT_ISZERO(dip->di_next_unlinked, ARCH_CONVERT));
1904 /* both on-disk, don't endian flip twice */
1905 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1906 offset = ip->i_boffset +
1907 offsetof(xfs_dinode_t, di_next_unlinked);
1908 xfs_trans_inode_buf(tp, ibp);
1909 xfs_trans_log_buf(tp, ibp, offset,
1910 (offset + sizeof(xfs_agino_t) - 1));
1911 xfs_inobp_check(mp, ibp);
1915 * Point the bucket head pointer at the inode being inserted.
1918 INT_SET(agi->agi_unlinked[bucket_index], ARCH_CONVERT, agino);
1919 offset = offsetof(xfs_agi_t, agi_unlinked) +
1920 (sizeof(xfs_agino_t) * bucket_index);
1921 xfs_trans_log_buf(tp, agibp, offset,
1922 (offset + sizeof(xfs_agino_t) - 1));
1927 * Pull the on-disk inode from the AGI unlinked list.
1940 xfs_agnumber_t agno;
1941 xfs_daddr_t agdaddr;
1943 xfs_agino_t next_agino;
1944 xfs_buf_t *last_ibp;
1945 xfs_dinode_t *last_dip;
1947 int offset, last_offset;
1952 * First pull the on-disk inode from the AGI unlinked list.
1956 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1957 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1960 * Get the agi buffer first. It ensures lock ordering
1963 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1964 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1967 "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
1968 error, mp->m_fsname);
1972 * Validate the magic number of the agi block.
1974 agi = XFS_BUF_TO_AGI(agibp);
1976 INT_GET(agi->agi_magicnum, ARCH_CONVERT) == XFS_AGI_MAGIC &&
1977 XFS_AGI_GOOD_VERSION(INT_GET(agi->agi_versionnum, ARCH_CONVERT));
1978 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE,
1979 XFS_RANDOM_IUNLINK_REMOVE))) {
1980 XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW,
1982 xfs_trans_brelse(tp, agibp);
1984 "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
1986 return XFS_ERROR(EFSCORRUPTED);
1989 * Get the index into the agi hash table for the
1990 * list this inode will go on.
1992 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1994 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1995 ASSERT(INT_GET(agi->agi_unlinked[bucket_index], ARCH_CONVERT) != NULLAGINO);
1996 ASSERT(!INT_ISZERO(agi->agi_unlinked[bucket_index], ARCH_CONVERT));
1998 if (INT_GET(agi->agi_unlinked[bucket_index], ARCH_CONVERT) == agino) {
2000 * We're at the head of the list. Get the inode's
2001 * on-disk buffer to see if there is anyone after us
2002 * on the list. Only modify our next pointer if it
2003 * is not already NULLAGINO. This saves us the overhead
2004 * of dealing with the buffer when there is no need to
2007 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0);
2010 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2011 error, mp->m_fsname);
2014 next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
2015 ASSERT(next_agino != 0);
2016 if (next_agino != NULLAGINO) {
2017 INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
2018 offset = ip->i_boffset +
2019 offsetof(xfs_dinode_t, di_next_unlinked);
2020 xfs_trans_inode_buf(tp, ibp);
2021 xfs_trans_log_buf(tp, ibp, offset,
2022 (offset + sizeof(xfs_agino_t) - 1));
2023 xfs_inobp_check(mp, ibp);
2025 xfs_trans_brelse(tp, ibp);
2028 * Point the bucket head pointer at the next inode.
2030 ASSERT(next_agino != 0);
2031 ASSERT(next_agino != agino);
2032 INT_SET(agi->agi_unlinked[bucket_index], ARCH_CONVERT, next_agino);
2033 offset = offsetof(xfs_agi_t, agi_unlinked) +
2034 (sizeof(xfs_agino_t) * bucket_index);
2035 xfs_trans_log_buf(tp, agibp, offset,
2036 (offset + sizeof(xfs_agino_t) - 1));
2039 * We need to search the list for the inode being freed.
2041 next_agino = INT_GET(agi->agi_unlinked[bucket_index], ARCH_CONVERT);
2043 while (next_agino != agino) {
2045 * If the last inode wasn't the one pointing to
2046 * us, then release its buffer since we're not
2047 * going to do anything with it.
2049 if (last_ibp != NULL) {
2050 xfs_trans_brelse(tp, last_ibp);
2052 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2053 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
2054 &last_ibp, &last_offset);
2057 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
2058 error, mp->m_fsname);
2061 next_agino = INT_GET(last_dip->di_next_unlinked, ARCH_CONVERT);
2062 ASSERT(next_agino != NULLAGINO);
2063 ASSERT(next_agino != 0);
2066 * Now last_ibp points to the buffer previous to us on
2067 * the unlinked list. Pull us from the list.
2069 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0);
2072 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2073 error, mp->m_fsname);
2076 next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
2077 ASSERT(next_agino != 0);
2078 ASSERT(next_agino != agino);
2079 if (next_agino != NULLAGINO) {
2080 INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
2081 offset = ip->i_boffset +
2082 offsetof(xfs_dinode_t, di_next_unlinked);
2083 xfs_trans_inode_buf(tp, ibp);
2084 xfs_trans_log_buf(tp, ibp, offset,
2085 (offset + sizeof(xfs_agino_t) - 1));
2086 xfs_inobp_check(mp, ibp);
2088 xfs_trans_brelse(tp, ibp);
2091 * Point the previous inode on the list to the next inode.
2093 INT_SET(last_dip->di_next_unlinked, ARCH_CONVERT, next_agino);
2094 ASSERT(next_agino != 0);
2095 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2096 xfs_trans_inode_buf(tp, last_ibp);
2097 xfs_trans_log_buf(tp, last_ibp, offset,
2098 (offset + sizeof(xfs_agino_t) - 1));
2099 xfs_inobp_check(mp, last_ibp);
2104 static __inline__ int xfs_inode_clean(xfs_inode_t *ip)
2106 return (((ip->i_itemp == NULL) ||
2107 !(ip->i_itemp->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
2108 (ip->i_update_core == 0));
2113 xfs_inode_t *free_ip,
2117 xfs_mount_t *mp = free_ip->i_mount;
2118 int blks_per_cluster;
2121 int i, j, found, pre_flushed;
2125 xfs_inode_t *ip, **ip_found;
2126 xfs_inode_log_item_t *iip;
2127 xfs_log_item_t *lip;
2130 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
2131 blks_per_cluster = 1;
2132 ninodes = mp->m_sb.sb_inopblock;
2133 nbufs = XFS_IALLOC_BLOCKS(mp);
2135 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
2136 mp->m_sb.sb_blocksize;
2137 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
2138 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
2141 ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
2143 for (j = 0; j < nbufs; j++, inum += ninodes) {
2144 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2145 XFS_INO_TO_AGBNO(mp, inum));
2149 * Look for each inode in memory and attempt to lock it,
2150 * we can be racing with flush and tail pushing here.
2151 * any inode we get the locks on, add to an array of
2152 * inode items to process later.
2154 * The get the buffer lock, we could beat a flush
2155 * or tail pushing thread to the lock here, in which
2156 * case they will go looking for the inode buffer
2157 * and fail, we need some other form of interlock
2161 for (i = 0; i < ninodes; i++) {
2162 ih = XFS_IHASH(mp, inum + i);
2163 read_lock(&ih->ih_lock);
2164 for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
2165 if (ip->i_ino == inum + i)
2169 /* Inode not in memory or we found it already,
2172 if (!ip || (ip->i_flags & XFS_ISTALE)) {
2173 read_unlock(&ih->ih_lock);
2177 if (xfs_inode_clean(ip)) {
2178 read_unlock(&ih->ih_lock);
2182 /* If we can get the locks then add it to the
2183 * list, otherwise by the time we get the bp lock
2184 * below it will already be attached to the
2188 /* This inode will already be locked - by us, lets
2192 if (ip == free_ip) {
2193 if (xfs_iflock_nowait(ip)) {
2194 ip->i_flags |= XFS_ISTALE;
2196 if (xfs_inode_clean(ip)) {
2199 ip_found[found++] = ip;
2202 read_unlock(&ih->ih_lock);
2206 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2207 if (xfs_iflock_nowait(ip)) {
2208 ip->i_flags |= XFS_ISTALE;
2210 if (xfs_inode_clean(ip)) {
2212 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2214 ip_found[found++] = ip;
2217 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2221 read_unlock(&ih->ih_lock);
2224 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2225 mp->m_bsize * blks_per_cluster,
2229 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
2231 if (lip->li_type == XFS_LI_INODE) {
2232 iip = (xfs_inode_log_item_t *)lip;
2233 ASSERT(iip->ili_logged == 1);
2234 lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
2236 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2238 iip->ili_inode->i_flags |= XFS_ISTALE;
2241 lip = lip->li_bio_list;
2244 for (i = 0; i < found; i++) {
2249 ip->i_update_core = 0;
2251 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2255 iip->ili_last_fields = iip->ili_format.ilf_fields;
2256 iip->ili_format.ilf_fields = 0;
2257 iip->ili_logged = 1;
2259 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2262 xfs_buf_attach_iodone(bp,
2263 (void(*)(xfs_buf_t*,xfs_log_item_t*))
2264 xfs_istale_done, (xfs_log_item_t *)iip);
2265 if (ip != free_ip) {
2266 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2270 if (found || pre_flushed)
2271 xfs_trans_stale_inode_buf(tp, bp);
2272 xfs_trans_binval(tp, bp);
2275 kmem_free(ip_found, ninodes * sizeof(xfs_inode_t *));
2279 * This is called to return an inode to the inode free list.
2280 * The inode should already be truncated to 0 length and have
2281 * no pages associated with it. This routine also assumes that
2282 * the inode is already a part of the transaction.
2284 * The on-disk copy of the inode will have been added to the list
2285 * of unlinked inodes in the AGI. We need to remove the inode from
2286 * that list atomically with respect to freeing it here.
2292 xfs_bmap_free_t *flist)
2296 xfs_ino_t first_ino;
2298 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2299 ASSERT(ip->i_transp == tp);
2300 ASSERT(ip->i_d.di_nlink == 0);
2301 ASSERT(ip->i_d.di_nextents == 0);
2302 ASSERT(ip->i_d.di_anextents == 0);
2303 ASSERT((ip->i_d.di_size == 0) ||
2304 ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2305 ASSERT(ip->i_d.di_nblocks == 0);
2308 * Pull the on-disk inode from the AGI unlinked list.
2310 error = xfs_iunlink_remove(tp, ip);
2315 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2319 ip->i_d.di_mode = 0; /* mark incore inode as free */
2320 ip->i_d.di_flags = 0;
2321 ip->i_d.di_dmevmask = 0;
2322 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2323 ip->i_df.if_ext_max =
2324 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2325 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2326 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2328 * Bump the generation count so no one will be confused
2329 * by reincarnations of this inode.
2332 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2335 xfs_ifree_cluster(ip, tp, first_ino);
2342 * Reallocate the space for if_broot based on the number of records
2343 * being added or deleted as indicated in rec_diff. Move the records
2344 * and pointers in if_broot to fit the new size. When shrinking this
2345 * will eliminate holes between the records and pointers created by
2346 * the caller. When growing this will create holes to be filled in
2349 * The caller must not request to add more records than would fit in
2350 * the on-disk inode root. If the if_broot is currently NULL, then
2351 * if we adding records one will be allocated. The caller must also
2352 * not request that the number of records go below zero, although
2353 * it can go to zero.
2355 * ip -- the inode whose if_broot area is changing
2356 * ext_diff -- the change in the number of records, positive or negative,
2357 * requested for the if_broot array.
2367 xfs_bmbt_block_t *new_broot;
2374 * Handle the degenerate case quietly.
2376 if (rec_diff == 0) {
2380 ifp = XFS_IFORK_PTR(ip, whichfork);
2383 * If there wasn't any memory allocated before, just
2384 * allocate it now and get out.
2386 if (ifp->if_broot_bytes == 0) {
2387 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
2388 ifp->if_broot = (xfs_bmbt_block_t*)kmem_alloc(new_size,
2390 ifp->if_broot_bytes = (int)new_size;
2395 * If there is already an existing if_broot, then we need
2396 * to realloc() it and shift the pointers to their new
2397 * location. The records don't change location because
2398 * they are kept butted up against the btree block header.
2400 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2401 new_max = cur_max + rec_diff;
2402 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2403 ifp->if_broot = (xfs_bmbt_block_t *)
2404 kmem_realloc(ifp->if_broot,
2406 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
2408 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2409 ifp->if_broot_bytes);
2410 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2412 ifp->if_broot_bytes = (int)new_size;
2413 ASSERT(ifp->if_broot_bytes <=
2414 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2415 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2420 * rec_diff is less than 0. In this case, we are shrinking the
2421 * if_broot buffer. It must already exist. If we go to zero
2422 * records, just get rid of the root and clear the status bit.
2424 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
2425 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2426 new_max = cur_max + rec_diff;
2427 ASSERT(new_max >= 0);
2429 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2433 new_broot = (xfs_bmbt_block_t *)kmem_alloc(new_size, KM_SLEEP);
2435 * First copy over the btree block header.
2437 memcpy(new_broot, ifp->if_broot, sizeof(xfs_bmbt_block_t));
2440 ifp->if_flags &= ~XFS_IFBROOT;
2444 * Only copy the records and pointers if there are any.
2448 * First copy the records.
2450 op = (char *)XFS_BMAP_BROOT_REC_ADDR(ifp->if_broot, 1,
2451 ifp->if_broot_bytes);
2452 np = (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot, 1,
2454 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2457 * Then copy the pointers.
2459 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2460 ifp->if_broot_bytes);
2461 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot, 1,
2463 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2465 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2466 ifp->if_broot = new_broot;
2467 ifp->if_broot_bytes = (int)new_size;
2468 ASSERT(ifp->if_broot_bytes <=
2469 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2475 * This is called when the amount of space needed for if_extents
2476 * is increased or decreased. The change in size is indicated by
2477 * the number of extents that need to be added or deleted in the
2478 * ext_diff parameter.
2480 * If the amount of space needed has decreased below the size of the
2481 * inline buffer, then switch to using the inline buffer. Otherwise,
2482 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2483 * to what is needed.
2485 * ip -- the inode whose if_extents area is changing
2486 * ext_diff -- the change in the number of extents, positive or negative,
2487 * requested for the if_extents array.
2500 if (ext_diff == 0) {
2504 ifp = XFS_IFORK_PTR(ip, whichfork);
2505 byte_diff = ext_diff * (uint)sizeof(xfs_bmbt_rec_t);
2506 new_size = (int)ifp->if_bytes + byte_diff;
2507 ASSERT(new_size >= 0);
2509 if (new_size == 0) {
2510 if (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext) {
2511 ASSERT(ifp->if_real_bytes != 0);
2512 kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
2514 ifp->if_u1.if_extents = NULL;
2516 } else if (new_size <= sizeof(ifp->if_u2.if_inline_ext)) {
2518 * If the valid extents can fit in if_inline_ext,
2519 * copy them from the malloc'd vector and free it.
2521 if (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext) {
2523 * For now, empty files are format EXTENTS,
2524 * so the if_extents pointer is null.
2526 if (ifp->if_u1.if_extents) {
2527 memcpy(ifp->if_u2.if_inline_ext,
2528 ifp->if_u1.if_extents, new_size);
2529 kmem_free(ifp->if_u1.if_extents,
2530 ifp->if_real_bytes);
2532 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
2536 rnew_size = new_size;
2537 if ((rnew_size & (rnew_size - 1)) != 0)
2538 rnew_size = xfs_iroundup(rnew_size);
2540 * Stuck with malloc/realloc.
2542 if (ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext) {
2543 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
2544 kmem_alloc(rnew_size, KM_SLEEP);
2545 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
2546 sizeof(ifp->if_u2.if_inline_ext));
2547 } else if (rnew_size != ifp->if_real_bytes) {
2548 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
2549 kmem_realloc(ifp->if_u1.if_extents,
2555 ifp->if_real_bytes = rnew_size;
2556 ifp->if_bytes = new_size;
2561 * This is called when the amount of space needed for if_data
2562 * is increased or decreased. The change in size is indicated by
2563 * the number of bytes that need to be added or deleted in the
2564 * byte_diff parameter.
2566 * If the amount of space needed has decreased below the size of the
2567 * inline buffer, then switch to using the inline buffer. Otherwise,
2568 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2569 * to what is needed.
2571 * ip -- the inode whose if_data area is changing
2572 * byte_diff -- the change in the number of bytes, positive or negative,
2573 * requested for the if_data array.
2585 if (byte_diff == 0) {
2589 ifp = XFS_IFORK_PTR(ip, whichfork);
2590 new_size = (int)ifp->if_bytes + byte_diff;
2591 ASSERT(new_size >= 0);
2593 if (new_size == 0) {
2594 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2595 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2597 ifp->if_u1.if_data = NULL;
2599 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2601 * If the valid extents/data can fit in if_inline_ext/data,
2602 * copy them from the malloc'd vector and free it.
2604 if (ifp->if_u1.if_data == NULL) {
2605 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2606 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2607 ASSERT(ifp->if_real_bytes != 0);
2608 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2610 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2611 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2616 * Stuck with malloc/realloc.
2617 * For inline data, the underlying buffer must be
2618 * a multiple of 4 bytes in size so that it can be
2619 * logged and stay on word boundaries. We enforce
2622 real_size = roundup(new_size, 4);
2623 if (ifp->if_u1.if_data == NULL) {
2624 ASSERT(ifp->if_real_bytes == 0);
2625 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2626 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2628 * Only do the realloc if the underlying size
2629 * is really changing.
2631 if (ifp->if_real_bytes != real_size) {
2632 ifp->if_u1.if_data =
2633 kmem_realloc(ifp->if_u1.if_data,
2639 ASSERT(ifp->if_real_bytes == 0);
2640 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2641 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2645 ifp->if_real_bytes = real_size;
2646 ifp->if_bytes = new_size;
2647 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2654 * Map inode to disk block and offset.
2656 * mp -- the mount point structure for the current file system
2657 * tp -- the current transaction
2658 * ino -- the inode number of the inode to be located
2659 * imap -- this structure is filled in with the information necessary
2660 * to retrieve the given inode from disk
2661 * flags -- flags to pass to xfs_dilocate indicating whether or not
2662 * lookups in the inode btree were OK or not
2672 xfs_fsblock_t fsbno;
2677 fsbno = imap->im_blkno ?
2678 XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK;
2679 error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags);
2683 imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno);
2684 imap->im_len = XFS_FSB_TO_BB(mp, len);
2685 imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno);
2686 imap->im_ioffset = (ushort)off;
2687 imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog);
2698 ifp = XFS_IFORK_PTR(ip, whichfork);
2699 if (ifp->if_broot != NULL) {
2700 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2701 ifp->if_broot = NULL;
2705 * If the format is local, then we can't have an extents
2706 * array so just look for an inline data array. If we're
2707 * not local then we may or may not have an extents list,
2708 * so check and free it up if we do.
2710 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2711 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2712 (ifp->if_u1.if_data != NULL)) {
2713 ASSERT(ifp->if_real_bytes != 0);
2714 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2715 ifp->if_u1.if_data = NULL;
2716 ifp->if_real_bytes = 0;
2718 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2719 (ifp->if_u1.if_extents != NULL) &&
2720 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)) {
2721 ASSERT(ifp->if_real_bytes != 0);
2722 kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
2723 ifp->if_u1.if_extents = NULL;
2724 ifp->if_real_bytes = 0;
2726 ASSERT(ifp->if_u1.if_extents == NULL ||
2727 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2728 ASSERT(ifp->if_real_bytes == 0);
2729 if (whichfork == XFS_ATTR_FORK) {
2730 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2736 * This is called free all the memory associated with an inode.
2737 * It must free the inode itself and any buffers allocated for
2738 * if_extents/if_data and if_broot. It must also free the lock
2739 * associated with the inode.
2746 switch (ip->i_d.di_mode & S_IFMT) {
2750 xfs_idestroy_fork(ip, XFS_DATA_FORK);
2754 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
2755 mrfree(&ip->i_lock);
2756 mrfree(&ip->i_iolock);
2757 freesema(&ip->i_flock);
2758 #ifdef XFS_BMAP_TRACE
2759 ktrace_free(ip->i_xtrace);
2761 #ifdef XFS_BMBT_TRACE
2762 ktrace_free(ip->i_btrace);
2765 ktrace_free(ip->i_rwtrace);
2767 #ifdef XFS_ILOCK_TRACE
2768 ktrace_free(ip->i_lock_trace);
2770 #ifdef XFS_DIR2_TRACE
2771 ktrace_free(ip->i_dir_trace);
2774 /* XXXdpd should be able to assert this but shutdown
2775 * is leaving the AIL behind. */
2776 ASSERT(((ip->i_itemp->ili_item.li_flags & XFS_LI_IN_AIL) == 0) ||
2777 XFS_FORCED_SHUTDOWN(ip->i_mount));
2778 xfs_inode_item_destroy(ip);
2780 kmem_zone_free(xfs_inode_zone, ip);
2785 * Increment the pin count of the given buffer.
2786 * This value is protected by ipinlock spinlock in the mount structure.
2792 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2794 atomic_inc(&ip->i_pincount);
2798 * Decrement the pin count of the given inode, and wake up
2799 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2800 * inode must have been previoulsy pinned with a call to xfs_ipin().
2806 ASSERT(atomic_read(&ip->i_pincount) > 0);
2808 if (atomic_dec_and_test(&ip->i_pincount)) {
2809 vnode_t *vp = XFS_ITOV_NULL(ip);
2811 /* make sync come back and flush this inode */
2813 struct inode *inode = LINVFS_GET_IP(vp);
2815 if (!(inode->i_state & I_NEW))
2816 mark_inode_dirty_sync(inode);
2819 wake_up(&ip->i_ipin_wait);
2824 * This is called to wait for the given inode to be unpinned.
2825 * It will sleep until this happens. The caller must have the
2826 * inode locked in at least shared mode so that the buffer cannot
2827 * be subsequently pinned once someone is waiting for it to be
2834 xfs_inode_log_item_t *iip;
2837 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE | MR_ACCESS));
2839 if (atomic_read(&ip->i_pincount) == 0) {
2844 if (iip && iip->ili_last_lsn) {
2845 lsn = iip->ili_last_lsn;
2851 * Give the log a push so we don't wait here too long.
2853 xfs_log_force(ip->i_mount, lsn, XFS_LOG_FORCE);
2855 wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0));
2860 * xfs_iextents_copy()
2862 * This is called to copy the REAL extents (as opposed to the delayed
2863 * allocation extents) from the inode into the given buffer. It
2864 * returns the number of bytes copied into the buffer.
2866 * If there are no delayed allocation extents, then we can just
2867 * memcpy() the extents into the buffer. Otherwise, we need to
2868 * examine each extent in turn and skip those which are delayed.
2873 xfs_bmbt_rec_t *buffer,
2877 xfs_bmbt_rec_t *dest_ep;
2879 #ifdef XFS_BMAP_TRACE
2880 static char fname[] = "xfs_iextents_copy";
2885 xfs_fsblock_t start_block;
2887 ifp = XFS_IFORK_PTR(ip, whichfork);
2888 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
2889 ASSERT(ifp->if_bytes > 0);
2891 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2892 xfs_bmap_trace_exlist(fname, ip, nrecs, whichfork);
2896 * There are some delayed allocation extents in the
2897 * inode, so copy the extents one at a time and skip
2898 * the delayed ones. There must be at least one
2899 * non-delayed extent.
2901 ep = ifp->if_u1.if_extents;
2904 for (i = 0; i < nrecs; i++) {
2905 start_block = xfs_bmbt_get_startblock(ep);
2906 if (ISNULLSTARTBLOCK(start_block)) {
2908 * It's a delayed allocation extent, so skip it.
2914 /* Translate to on disk format */
2915 put_unaligned(INT_GET(ep->l0, ARCH_CONVERT),
2916 (__uint64_t*)&dest_ep->l0);
2917 put_unaligned(INT_GET(ep->l1, ARCH_CONVERT),
2918 (__uint64_t*)&dest_ep->l1);
2923 ASSERT(copied != 0);
2924 xfs_validate_extents(buffer, copied, 1, XFS_EXTFMT_INODE(ip));
2926 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2930 * Each of the following cases stores data into the same region
2931 * of the on-disk inode, so only one of them can be valid at
2932 * any given time. While it is possible to have conflicting formats
2933 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2934 * in EXTENTS format, this can only happen when the fork has
2935 * changed formats after being modified but before being flushed.
2936 * In these cases, the format always takes precedence, because the
2937 * format indicates the current state of the fork.
2944 xfs_inode_log_item_t *iip,
2951 #ifdef XFS_TRANS_DEBUG
2954 static const short brootflag[2] =
2955 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2956 static const short dataflag[2] =
2957 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2958 static const short extflag[2] =
2959 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2963 ifp = XFS_IFORK_PTR(ip, whichfork);
2965 * This can happen if we gave up in iformat in an error path,
2966 * for the attribute fork.
2969 ASSERT(whichfork == XFS_ATTR_FORK);
2972 cp = XFS_DFORK_PTR_ARCH(dip, whichfork, ARCH_CONVERT);
2974 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2975 case XFS_DINODE_FMT_LOCAL:
2976 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2977 (ifp->if_bytes > 0)) {
2978 ASSERT(ifp->if_u1.if_data != NULL);
2979 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2980 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2982 if (whichfork == XFS_DATA_FORK) {
2983 if (unlikely(XFS_DIR_SHORTFORM_VALIDATE_ONDISK(mp, dip))) {
2984 XFS_ERROR_REPORT("xfs_iflush_fork",
2985 XFS_ERRLEVEL_LOW, mp);
2986 return XFS_ERROR(EFSCORRUPTED);
2991 case XFS_DINODE_FMT_EXTENTS:
2992 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2993 !(iip->ili_format.ilf_fields & extflag[whichfork]));
2994 ASSERT((ifp->if_u1.if_extents != NULL) || (ifp->if_bytes == 0));
2995 ASSERT((ifp->if_u1.if_extents == NULL) || (ifp->if_bytes > 0));
2996 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
2997 (ifp->if_bytes > 0)) {
2998 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2999 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
3004 case XFS_DINODE_FMT_BTREE:
3005 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
3006 (ifp->if_broot_bytes > 0)) {
3007 ASSERT(ifp->if_broot != NULL);
3008 ASSERT(ifp->if_broot_bytes <=
3009 (XFS_IFORK_SIZE(ip, whichfork) +
3010 XFS_BROOT_SIZE_ADJ));
3011 xfs_bmbt_to_bmdr(ifp->if_broot, ifp->if_broot_bytes,
3012 (xfs_bmdr_block_t *)cp,
3013 XFS_DFORK_SIZE_ARCH(dip, mp, whichfork, ARCH_CONVERT));
3017 case XFS_DINODE_FMT_DEV:
3018 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
3019 ASSERT(whichfork == XFS_DATA_FORK);
3020 INT_SET(dip->di_u.di_dev, ARCH_CONVERT, ip->i_df.if_u2.if_rdev);
3024 case XFS_DINODE_FMT_UUID:
3025 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
3026 ASSERT(whichfork == XFS_DATA_FORK);
3027 memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid,
3041 * xfs_iflush() will write a modified inode's changes out to the
3042 * inode's on disk home. The caller must have the inode lock held
3043 * in at least shared mode and the inode flush semaphore must be
3044 * held as well. The inode lock will still be held upon return from
3045 * the call and the caller is free to unlock it.
3046 * The inode flush lock will be unlocked when the inode reaches the disk.
3047 * The flags indicate how the inode's buffer should be written out.
3054 xfs_inode_log_item_t *iip;
3062 int clcount; /* count of inodes clustered */
3064 enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
3067 XFS_STATS_INC(xs_iflush_count);
3069 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
3070 ASSERT(valusema(&ip->i_flock) <= 0);
3071 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3072 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3078 * If the inode isn't dirty, then just release the inode
3079 * flush lock and do nothing.
3081 if ((ip->i_update_core == 0) &&
3082 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3083 ASSERT((iip != NULL) ?
3084 !(iip->ili_item.li_flags & XFS_LI_IN_AIL) : 1);
3090 * We can't flush the inode until it is unpinned, so
3091 * wait for it. We know noone new can pin it, because
3092 * we are holding the inode lock shared and you need
3093 * to hold it exclusively to pin the inode.
3095 xfs_iunpin_wait(ip);
3098 * This may have been unpinned because the filesystem is shutting
3099 * down forcibly. If that's the case we must not write this inode
3100 * to disk, because the log record didn't make it to disk!
3102 if (XFS_FORCED_SHUTDOWN(mp)) {
3103 ip->i_update_core = 0;
3105 iip->ili_format.ilf_fields = 0;
3107 return XFS_ERROR(EIO);
3111 * Get the buffer containing the on-disk inode.
3113 error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0);
3120 * Decide how buffer will be flushed out. This is done before
3121 * the call to xfs_iflush_int because this field is zeroed by it.
3123 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3125 * Flush out the inode buffer according to the directions
3126 * of the caller. In the cases where the caller has given
3127 * us a choice choose the non-delwri case. This is because
3128 * the inode is in the AIL and we need to get it out soon.
3131 case XFS_IFLUSH_SYNC:
3132 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3135 case XFS_IFLUSH_ASYNC:
3136 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3139 case XFS_IFLUSH_DELWRI:
3149 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3150 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3151 case XFS_IFLUSH_DELWRI:
3154 case XFS_IFLUSH_ASYNC:
3157 case XFS_IFLUSH_SYNC:
3168 * First flush out the inode that xfs_iflush was called with.
3170 error = xfs_iflush_int(ip, bp);
3177 * see if other inodes can be gathered into this write
3180 ip->i_chash->chl_buf = bp;
3182 ch = XFS_CHASH(mp, ip->i_blkno);
3183 s = mutex_spinlock(&ch->ch_lock);
3186 for (iq = ip->i_cnext; iq != ip; iq = iq->i_cnext) {
3188 * Do an un-protected check to see if the inode is dirty and
3189 * is a candidate for flushing. These checks will be repeated
3190 * later after the appropriate locks are acquired.
3193 if ((iq->i_update_core == 0) &&
3195 !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
3196 xfs_ipincount(iq) == 0) {
3201 * Try to get locks. If any are unavailable,
3202 * then this inode cannot be flushed and is skipped.
3205 /* get inode locks (just i_lock) */
3206 if (xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) {
3207 /* get inode flush lock */
3208 if (xfs_iflock_nowait(iq)) {
3209 /* check if pinned */
3210 if (xfs_ipincount(iq) == 0) {
3211 /* arriving here means that
3212 * this inode can be flushed.
3213 * first re-check that it's
3217 if ((iq->i_update_core != 0)||
3219 (iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3221 error = xfs_iflush_int(iq, bp);
3225 goto cluster_corrupt_out;
3234 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3237 mutex_spinunlock(&ch->ch_lock, s);
3240 XFS_STATS_INC(xs_icluster_flushcnt);
3241 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
3245 * If the buffer is pinned then push on the log so we won't
3246 * get stuck waiting in the write for too long.
3248 if (XFS_BUF_ISPINNED(bp)){
3249 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
3252 if (flags & INT_DELWRI) {
3253 xfs_bdwrite(mp, bp);
3254 } else if (flags & INT_ASYNC) {
3255 xfs_bawrite(mp, bp);
3257 error = xfs_bwrite(mp, bp);
3263 xfs_force_shutdown(mp, XFS_CORRUPT_INCORE);
3264 xfs_iflush_abort(ip);
3266 * Unlocks the flush lock
3268 return XFS_ERROR(EFSCORRUPTED);
3270 cluster_corrupt_out:
3271 /* Corruption detected in the clustering loop. Invalidate the
3272 * inode buffer and shut down the filesystem.
3274 mutex_spinunlock(&ch->ch_lock, s);
3277 * Clean up the buffer. If it was B_DELWRI, just release it --
3278 * brelse can handle it with no problems. If not, shut down the
3279 * filesystem before releasing the buffer.
3281 if ((bufwasdelwri= XFS_BUF_ISDELAYWRITE(bp))) {
3285 xfs_force_shutdown(mp, XFS_CORRUPT_INCORE);
3289 * Just like incore_relse: if we have b_iodone functions,
3290 * mark the buffer as an error and call them. Otherwise
3291 * mark it as stale and brelse.
3293 if (XFS_BUF_IODONE_FUNC(bp)) {
3294 XFS_BUF_CLR_BDSTRAT_FUNC(bp);
3298 XFS_BUF_ERROR(bp,EIO);
3306 xfs_iflush_abort(iq);
3308 * Unlocks the flush lock
3310 return XFS_ERROR(EFSCORRUPTED);
3319 xfs_inode_log_item_t *iip;
3322 #ifdef XFS_TRANS_DEBUG
3327 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
3328 ASSERT(valusema(&ip->i_flock) <= 0);
3329 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3330 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3337 * If the inode isn't dirty, then just release the inode
3338 * flush lock and do nothing.
3340 if ((ip->i_update_core == 0) &&
3341 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3346 /* set *dip = inode's place in the buffer */
3347 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset);
3350 * Clear i_update_core before copying out the data.
3351 * This is for coordination with our timestamp updates
3352 * that don't hold the inode lock. They will always
3353 * update the timestamps BEFORE setting i_update_core,
3354 * so if we clear i_update_core after they set it we
3355 * are guaranteed to see their updates to the timestamps.
3356 * I believe that this depends on strongly ordered memory
3357 * semantics, but we have that. We use the SYNCHRONIZE
3358 * macro to make sure that the compiler does not reorder
3359 * the i_update_core access below the data copy below.
3361 ip->i_update_core = 0;
3364 if (XFS_TEST_ERROR(INT_GET(dip->di_core.di_magic,ARCH_CONVERT) != XFS_DINODE_MAGIC,
3365 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3366 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3367 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3368 ip->i_ino, (int) INT_GET(dip->di_core.di_magic, ARCH_CONVERT), dip);
3371 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3372 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3373 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3374 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3375 ip->i_ino, ip, ip->i_d.di_magic);
3378 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3380 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3381 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3382 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3383 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3384 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3388 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
3390 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3391 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3392 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3393 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3394 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3395 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3400 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3401 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3402 XFS_RANDOM_IFLUSH_5)) {
3403 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3404 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3406 ip->i_d.di_nextents + ip->i_d.di_anextents,
3411 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3412 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3413 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3414 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3415 ip->i_ino, ip->i_d.di_forkoff, ip);
3419 * bump the flush iteration count, used to detect flushes which
3420 * postdate a log record during recovery.
3423 ip->i_d.di_flushiter++;
3426 * Copy the dirty parts of the inode into the on-disk
3427 * inode. We always copy out the core of the inode,
3428 * because if the inode is dirty at all the core must
3431 xfs_xlate_dinode_core((xfs_caddr_t)&(dip->di_core), &(ip->i_d),
3434 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3435 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3436 ip->i_d.di_flushiter = 0;
3439 * If this is really an old format inode and the superblock version
3440 * has not been updated to support only new format inodes, then
3441 * convert back to the old inode format. If the superblock version
3442 * has been updated, then make the conversion permanent.
3444 ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 ||
3445 XFS_SB_VERSION_HASNLINK(&mp->m_sb));
3446 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
3447 if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) {
3451 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3452 INT_SET(dip->di_core.di_onlink, ARCH_CONVERT, ip->i_d.di_nlink);
3455 * The superblock version has already been bumped,
3456 * so just make the conversion to the new inode
3459 ip->i_d.di_version = XFS_DINODE_VERSION_2;
3460 INT_SET(dip->di_core.di_version, ARCH_CONVERT, XFS_DINODE_VERSION_2);
3461 ip->i_d.di_onlink = 0;
3462 INT_ZERO(dip->di_core.di_onlink, ARCH_CONVERT);
3463 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3464 memset(&(dip->di_core.di_pad[0]), 0,
3465 sizeof(dip->di_core.di_pad));
3466 ASSERT(ip->i_d.di_projid == 0);
3470 if (xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp) == EFSCORRUPTED) {
3474 if (XFS_IFORK_Q(ip)) {
3476 * The only error from xfs_iflush_fork is on the data fork.
3478 (void) xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3480 xfs_inobp_check(mp, bp);
3483 * We've recorded everything logged in the inode, so we'd
3484 * like to clear the ilf_fields bits so we don't log and
3485 * flush things unnecessarily. However, we can't stop
3486 * logging all this information until the data we've copied
3487 * into the disk buffer is written to disk. If we did we might
3488 * overwrite the copy of the inode in the log with all the
3489 * data after re-logging only part of it, and in the face of
3490 * a crash we wouldn't have all the data we need to recover.
3492 * What we do is move the bits to the ili_last_fields field.
3493 * When logging the inode, these bits are moved back to the
3494 * ilf_fields field. In the xfs_iflush_done() routine we
3495 * clear ili_last_fields, since we know that the information
3496 * those bits represent is permanently on disk. As long as
3497 * the flush completes before the inode is logged again, then
3498 * both ilf_fields and ili_last_fields will be cleared.
3500 * We can play with the ilf_fields bits here, because the inode
3501 * lock must be held exclusively in order to set bits there
3502 * and the flush lock protects the ili_last_fields bits.
3503 * Set ili_logged so the flush done
3504 * routine can tell whether or not to look in the AIL.
3505 * Also, store the current LSN of the inode so that we can tell
3506 * whether the item has moved in the AIL from xfs_iflush_done().
3507 * In order to read the lsn we need the AIL lock, because
3508 * it is a 64 bit value that cannot be read atomically.
3510 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3511 iip->ili_last_fields = iip->ili_format.ilf_fields;
3512 iip->ili_format.ilf_fields = 0;
3513 iip->ili_logged = 1;
3515 ASSERT(sizeof(xfs_lsn_t) == 8); /* don't lock if it shrinks */
3517 iip->ili_flush_lsn = iip->ili_item.li_lsn;
3521 * Attach the function xfs_iflush_done to the inode's
3522 * buffer. This will remove the inode from the AIL
3523 * and unlock the inode's flush lock when the inode is
3524 * completely written to disk.
3526 xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
3527 xfs_iflush_done, (xfs_log_item_t *)iip);
3529 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3530 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3533 * We're flushing an inode which is not in the AIL and has
3534 * not been logged but has i_update_core set. For this
3535 * case we can use a B_DELWRI flush and immediately drop
3536 * the inode flush lock because we can avoid the whole
3537 * AIL state thing. It's OK to drop the flush lock now,
3538 * because we've already locked the buffer and to do anything
3539 * you really need both.
3542 ASSERT(iip->ili_logged == 0);
3543 ASSERT(iip->ili_last_fields == 0);
3544 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3552 return XFS_ERROR(EFSCORRUPTED);
3556 * Flush all inactive inodes in mp. Return true if no user references
3557 * were found, false otherwise.
3574 XFS_MOUNT_ILOCK(mp);
3580 /* Make sure we skip markers inserted by sync */
3581 if (ip->i_mount == NULL) {
3587 * It's up to our caller to purge the root
3588 * and quota vnodes later.
3590 vp = XFS_ITOV_NULL(ip);
3593 XFS_MOUNT_IUNLOCK(mp);
3594 xfs_finish_reclaim(ip, 0, XFS_IFLUSH_ASYNC);
3599 if (vn_count(vp) != 0) {
3600 if (vn_count(vp) == 1 &&
3601 (ip == mp->m_rootip ||
3603 (ip->i_ino == mp->m_sb.sb_uquotino ||
3604 ip->i_ino == mp->m_sb.sb_gquotino)))) {
3609 if (!(flag & XFS_FLUSH_ALL)) {
3616 * Ignore busy inodes but continue flushing
3623 * Sample vp mapping while holding mp locked on MP
3624 * systems, so we don't purge a reclaimed or
3625 * nonexistent vnode. We break from the loop
3626 * since we know that we modify
3627 * it by pulling ourselves from it in xfs_reclaim()
3628 * called via vn_purge() below. Set ip to the next
3629 * entry in the list anyway so we'll know below
3630 * whether we reached the end or not.
3633 XFS_MOUNT_IUNLOCK(mp);
3635 vn_purge(vp, &vmap);
3639 } while (ip != mp->m_inodes);
3641 * We need to distinguish between when we exit the loop
3642 * after a purge and when we simply hit the end of the
3643 * list. We can't use the (ip == mp->m_inodes) test,
3644 * because when we purge an inode at the start of the list
3645 * the next inode on the list becomes mp->m_inodes. That
3646 * would cause such a test to bail out early. The purged
3647 * variable tells us how we got out of the loop.
3653 XFS_MOUNT_IUNLOCK(mp);
3659 * xfs_iaccess: check accessibility of inode for mode.
3668 mode_t orgmode = mode;
3669 struct inode *inode = LINVFS_GET_IP(XFS_ITOV(ip));
3672 * Verify that the MAC policy allows the requested access.
3674 if ((error = _MAC_XFS_IACCESS(ip, mode, cr)))
3675 return XFS_ERROR(error);
3677 if (mode & S_IWUSR) {
3678 umode_t imode = inode->i_mode;
3680 if (IS_RDONLY(inode) &&
3681 (S_ISREG(imode) || S_ISDIR(imode) || S_ISLNK(imode)))
3682 return XFS_ERROR(EROFS);
3684 if (IS_IMMUTABLE(inode))
3685 return XFS_ERROR(EACCES);
3689 * If there's an Access Control List it's used instead of
3692 if ((error = _ACL_XFS_IACCESS(ip, mode, cr)) != -1)
3693 return error ? XFS_ERROR(error) : 0;
3695 if (current_fsuid(cr) != ip->i_d.di_uid) {
3697 if (!in_group_p((gid_t)ip->i_d.di_gid))
3702 * If the DACs are ok we don't need any capability check.
3704 if ((ip->i_d.di_mode & mode) == mode)
3707 * Read/write DACs are always overridable.
3708 * Executable DACs are overridable if at least one exec bit is set.
3710 if (!(orgmode & S_IXUSR) ||
3711 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
3712 if (capable_cred(cr, CAP_DAC_OVERRIDE))
3715 if ((orgmode == S_IRUSR) ||
3716 (S_ISDIR(inode->i_mode) && (!(orgmode & S_IWUSR)))) {
3717 if (capable_cred(cr, CAP_DAC_READ_SEARCH))
3720 cmn_err(CE_NOTE, "Ick: mode=%o, orgmode=%o", mode, orgmode);
3722 return XFS_ERROR(EACCES);
3724 return XFS_ERROR(EACCES);
3728 * Return whether or not it is OK to swap to the given file in the
3729 * given range. Return 0 for OK and otherwise return the error.
3731 * It is only OK to swap to a file if it has no holes, and all
3732 * extents have been initialized.
3734 * We use the vnode behavior chain prevent and allow primitives
3735 * to ensure that the vnode chain stays coherent while we do this.
3736 * This allows us to walk the chain down to the bottom where XFS
3737 * lives without worrying about it changing out from under us.
3745 ip = XFS_BHVTOI(bdp);
3747 * Verify that the file does not have any
3748 * holes or unwritten exents.
3750 return xfs_bmap_check_swappable(ip);
3754 * xfs_iroundup: round up argument to next power of two
3763 if ((v & (v - 1)) == 0)
3765 ASSERT((v & 0x80000000) == 0);
3766 if ((v & (v + 1)) == 0)
3768 for (i = 0, m = 1; i < 31; i++, m <<= 1) {
3772 if ((v & (v + 1)) == 0)
3780 * Change the requested timestamp in the given inode.
3781 * We don't lock across timestamp updates, and we don't log them but
3782 * we do record the fact that there is dirty information in core.
3784 * NOTE -- callers MUST combine XFS_ICHGTIME_MOD or XFS_ICHGTIME_CHG
3785 * with XFS_ICHGTIME_ACC to be sure that access time
3786 * update will take. Calling first with XFS_ICHGTIME_ACC
3787 * and then XFS_ICHGTIME_MOD may fail to modify the access
3788 * timestamp if the filesystem is mounted noacctm.
3791 xfs_ichgtime(xfs_inode_t *ip,
3795 vnode_t *vp = XFS_ITOV(ip);
3796 struct inode *inode = LINVFS_GET_IP(vp);
3799 * We're not supposed to change timestamps in readonly-mounted
3800 * filesystems. Throw it away if anyone asks us.
3802 if (unlikely(vp->v_vfsp->vfs_flag & VFS_RDONLY))
3806 * Don't update access timestamps on reads if mounted "noatime"
3807 * Throw it away if anyone asks us.
3809 if ((ip->i_mount->m_flags & XFS_MOUNT_NOATIME || IS_NOATIME(inode)) &&
3810 ((flags & (XFS_ICHGTIME_ACC|XFS_ICHGTIME_MOD|XFS_ICHGTIME_CHG))
3811 == XFS_ICHGTIME_ACC))
3815 if (flags & XFS_ICHGTIME_MOD) {
3816 VN_MTIMESET(vp, &tv);
3817 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
3818 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
3820 if (flags & XFS_ICHGTIME_ACC) {
3821 VN_ATIMESET(vp, &tv);
3822 ip->i_d.di_atime.t_sec = (__int32_t)tv.tv_sec;
3823 ip->i_d.di_atime.t_nsec = (__int32_t)tv.tv_nsec;
3825 if (flags & XFS_ICHGTIME_CHG) {
3826 VN_CTIMESET(vp, &tv);
3827 ip->i_d.di_ctime.t_sec = (__int32_t)tv.tv_sec;
3828 ip->i_d.di_ctime.t_nsec = (__int32_t)tv.tv_nsec;
3832 * We update the i_update_core field _after_ changing
3833 * the timestamps in order to coordinate properly with
3834 * xfs_iflush() so that we don't lose timestamp updates.
3835 * This keeps us from having to hold the inode lock
3836 * while doing this. We use the SYNCHRONIZE macro to
3837 * ensure that the compiler does not reorder the update
3838 * of i_update_core above the timestamp updates above.
3841 ip->i_update_core = 1;
3842 if (!(inode->i_state & I_LOCK))
3843 mark_inode_dirty_sync(inode);
3846 #ifdef XFS_ILOCK_TRACE
3847 ktrace_t *xfs_ilock_trace_buf;
3850 xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
3852 ktrace_enter(ip->i_lock_trace,
3854 (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */
3855 (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */
3856 (void *)ra, /* caller of ilock */
3857 (void *)(unsigned long)current_cpu(),
3858 (void *)(unsigned long)current_pid(),
3859 0,0,0,0,0,0,0,0,0,0);