2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 #include <linux/log2.h>
22 #include "xfs_types.h"
27 #include "xfs_trans.h"
28 #include "xfs_trans_priv.h"
32 #include "xfs_dmapi.h"
33 #include "xfs_mount.h"
34 #include "xfs_bmap_btree.h"
35 #include "xfs_alloc_btree.h"
36 #include "xfs_ialloc_btree.h"
37 #include "xfs_dir2_sf.h"
38 #include "xfs_attr_sf.h"
39 #include "xfs_dinode.h"
40 #include "xfs_inode.h"
41 #include "xfs_buf_item.h"
42 #include "xfs_inode_item.h"
43 #include "xfs_btree.h"
44 #include "xfs_btree_trace.h"
45 #include "xfs_alloc.h"
46 #include "xfs_ialloc.h"
49 #include "xfs_error.h"
50 #include "xfs_utils.h"
51 #include "xfs_dir2_trace.h"
52 #include "xfs_quota.h"
54 #include "xfs_filestream.h"
55 #include "xfs_vnodeops.h"
57 kmem_zone_t
*xfs_ifork_zone
;
58 kmem_zone_t
*xfs_inode_zone
;
61 * Used in xfs_itruncate(). This is the maximum number of extents
62 * freed from a file in a single transaction.
64 #define XFS_ITRUNC_MAX_EXTENTS 2
66 STATIC
int xfs_iflush_int(xfs_inode_t
*, xfs_buf_t
*);
67 STATIC
int xfs_iformat_local(xfs_inode_t
*, xfs_dinode_t
*, int, int);
68 STATIC
int xfs_iformat_extents(xfs_inode_t
*, xfs_dinode_t
*, int);
69 STATIC
int xfs_iformat_btree(xfs_inode_t
*, xfs_dinode_t
*, int);
73 * Make sure that the extents in the given memory buffer
83 xfs_bmbt_rec_host_t rec
;
86 for (i
= 0; i
< nrecs
; i
++) {
87 xfs_bmbt_rec_host_t
*ep
= xfs_iext_get_ext(ifp
, i
);
88 rec
.l0
= get_unaligned(&ep
->l0
);
89 rec
.l1
= get_unaligned(&ep
->l1
);
90 xfs_bmbt_get_all(&rec
, &irec
);
91 if (fmt
== XFS_EXTFMT_NOSTATE
)
92 ASSERT(irec
.br_state
== XFS_EXT_NORM
);
96 #define xfs_validate_extents(ifp, nrecs, fmt)
100 * Check that none of the inode's in the buffer have a next
101 * unlinked field of 0.
113 j
= mp
->m_inode_cluster_size
>> mp
->m_sb
.sb_inodelog
;
115 for (i
= 0; i
< j
; i
++) {
116 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
,
117 i
* mp
->m_sb
.sb_inodesize
);
118 if (!dip
->di_next_unlinked
) {
119 xfs_fs_cmn_err(CE_ALERT
, mp
,
120 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
122 ASSERT(dip
->di_next_unlinked
);
129 * Find the buffer associated with the given inode map
130 * We do basic validation checks on the buffer once it has been
131 * retrieved from disk.
147 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, imap
->im_blkno
,
148 (int)imap
->im_len
, buf_flags
, &bp
);
150 if (error
!= EAGAIN
) {
152 "xfs_imap_to_bp: xfs_trans_read_buf()returned "
153 "an error %d on %s. Returning error.",
154 error
, mp
->m_fsname
);
156 ASSERT(buf_flags
& XFS_BUF_TRYLOCK
);
162 * Validate the magic number and version of every inode in the buffer
163 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
166 ni
= BBTOB(imap
->im_len
) >> mp
->m_sb
.sb_inodelog
;
167 #else /* usual case */
171 for (i
= 0; i
< ni
; i
++) {
175 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
,
176 (i
<< mp
->m_sb
.sb_inodelog
));
177 di_ok
= be16_to_cpu(dip
->di_core
.di_magic
) == XFS_DINODE_MAGIC
&&
178 XFS_DINODE_GOOD_VERSION(dip
->di_core
.di_version
);
179 if (unlikely(XFS_TEST_ERROR(!di_ok
, mp
,
180 XFS_ERRTAG_ITOBP_INOTOBP
,
181 XFS_RANDOM_ITOBP_INOTOBP
))) {
182 if (imap_flags
& XFS_IMAP_BULKSTAT
) {
183 xfs_trans_brelse(tp
, bp
);
184 return XFS_ERROR(EINVAL
);
186 XFS_CORRUPTION_ERROR("xfs_imap_to_bp",
187 XFS_ERRLEVEL_HIGH
, mp
, dip
);
190 "Device %s - bad inode magic/vsn "
191 "daddr %lld #%d (magic=%x)",
192 XFS_BUFTARG_NAME(mp
->m_ddev_targp
),
193 (unsigned long long)imap
->im_blkno
, i
,
194 be16_to_cpu(dip
->di_core
.di_magic
));
196 xfs_trans_brelse(tp
, bp
);
197 return XFS_ERROR(EFSCORRUPTED
);
201 xfs_inobp_check(mp
, bp
);
204 * Mark the buffer as an inode buffer now that it looks good
206 XFS_BUF_SET_VTYPE(bp
, B_FS_INO
);
213 * This routine is called to map an inode number within a file
214 * system to the buffer containing the on-disk version of the
215 * inode. It returns a pointer to the buffer containing the
216 * on-disk inode in the bpp parameter, and in the dip parameter
217 * it returns a pointer to the on-disk inode within that buffer.
219 * If a non-zero error is returned, then the contents of bpp and
220 * dipp are undefined.
222 * Use xfs_imap() to determine the size and location of the
223 * buffer to read from disk.
239 error
= xfs_imap(mp
, tp
, ino
, &imap
, XFS_IMAP_LOOKUP
);
243 error
= xfs_imap_to_bp(mp
, tp
, &imap
, &bp
, XFS_BUF_LOCK
, 0);
247 *dipp
= (xfs_dinode_t
*)xfs_buf_offset(bp
, imap
.im_boffset
);
249 *offset
= imap
.im_boffset
;
255 * This routine is called to map an inode to the buffer containing
256 * the on-disk version of the inode. It returns a pointer to the
257 * buffer containing the on-disk inode in the bpp parameter, and in
258 * the dip parameter it returns a pointer to the on-disk inode within
261 * If a non-zero error is returned, then the contents of bpp and
262 * dipp are undefined.
264 * If the inode is new and has not yet been initialized, use xfs_imap()
265 * to determine the size and location of the buffer to read from disk.
266 * If the inode has already been mapped to its buffer and read in once,
267 * then use the mapping information stored in the inode rather than
268 * calling xfs_imap(). This allows us to avoid the overhead of looking
269 * at the inode btree for small block file systems (see xfs_dilocate()).
270 * We can tell whether the inode has been mapped in before by comparing
271 * its disk block address to 0. Only uninitialized inodes will have
272 * 0 for the disk block address.
289 if (ip
->i_blkno
== (xfs_daddr_t
)0) {
291 error
= xfs_imap(mp
, tp
, ip
->i_ino
, &imap
,
292 XFS_IMAP_LOOKUP
| imap_flags
);
297 * Fill in the fields in the inode that will be used to
298 * map the inode to its buffer from now on.
300 ip
->i_blkno
= imap
.im_blkno
;
301 ip
->i_len
= imap
.im_len
;
302 ip
->i_boffset
= imap
.im_boffset
;
305 * We've already mapped the inode once, so just use the
306 * mapping that we saved the first time.
308 imap
.im_blkno
= ip
->i_blkno
;
309 imap
.im_len
= ip
->i_len
;
310 imap
.im_boffset
= ip
->i_boffset
;
312 ASSERT(bno
== 0 || bno
== imap
.im_blkno
);
314 error
= xfs_imap_to_bp(mp
, tp
, &imap
, &bp
, buf_flags
, imap_flags
);
319 ASSERT(buf_flags
& XFS_BUF_TRYLOCK
);
325 *dipp
= (xfs_dinode_t
*)xfs_buf_offset(bp
, imap
.im_boffset
);
331 * Move inode type and inode format specific information from the
332 * on-disk inode to the in-core inode. For fifos, devs, and sockets
333 * this means set if_rdev to the proper value. For files, directories,
334 * and symlinks this means to bring in the in-line data or extent
335 * pointers. For a file in B-tree format, only the root is immediately
336 * brought in-core. The rest will be in-lined in if_extents when it
337 * is first referenced (see xfs_iread_extents()).
344 xfs_attr_shortform_t
*atp
;
348 ip
->i_df
.if_ext_max
=
349 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
352 if (unlikely(be32_to_cpu(dip
->di_core
.di_nextents
) +
353 be16_to_cpu(dip
->di_core
.di_anextents
) >
354 be64_to_cpu(dip
->di_core
.di_nblocks
))) {
355 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
356 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
357 (unsigned long long)ip
->i_ino
,
358 (int)(be32_to_cpu(dip
->di_core
.di_nextents
) +
359 be16_to_cpu(dip
->di_core
.di_anextents
)),
361 be64_to_cpu(dip
->di_core
.di_nblocks
));
362 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW
,
364 return XFS_ERROR(EFSCORRUPTED
);
367 if (unlikely(dip
->di_core
.di_forkoff
> ip
->i_mount
->m_sb
.sb_inodesize
)) {
368 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
369 "corrupt dinode %Lu, forkoff = 0x%x.",
370 (unsigned long long)ip
->i_ino
,
371 dip
->di_core
.di_forkoff
);
372 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW
,
374 return XFS_ERROR(EFSCORRUPTED
);
377 switch (ip
->i_d
.di_mode
& S_IFMT
) {
382 if (unlikely(dip
->di_core
.di_format
!= XFS_DINODE_FMT_DEV
)) {
383 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW
,
385 return XFS_ERROR(EFSCORRUPTED
);
389 ip
->i_df
.if_u2
.if_rdev
= be32_to_cpu(dip
->di_u
.di_dev
);
395 switch (dip
->di_core
.di_format
) {
396 case XFS_DINODE_FMT_LOCAL
:
398 * no local regular files yet
400 if (unlikely((be16_to_cpu(dip
->di_core
.di_mode
) & S_IFMT
) == S_IFREG
)) {
401 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
403 "(local format for regular file).",
404 (unsigned long long) ip
->i_ino
);
405 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
408 return XFS_ERROR(EFSCORRUPTED
);
411 di_size
= be64_to_cpu(dip
->di_core
.di_size
);
412 if (unlikely(di_size
> XFS_DFORK_DSIZE(dip
, ip
->i_mount
))) {
413 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
415 "(bad size %Ld for local inode).",
416 (unsigned long long) ip
->i_ino
,
417 (long long) di_size
);
418 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
421 return XFS_ERROR(EFSCORRUPTED
);
425 error
= xfs_iformat_local(ip
, dip
, XFS_DATA_FORK
, size
);
427 case XFS_DINODE_FMT_EXTENTS
:
428 error
= xfs_iformat_extents(ip
, dip
, XFS_DATA_FORK
);
430 case XFS_DINODE_FMT_BTREE
:
431 error
= xfs_iformat_btree(ip
, dip
, XFS_DATA_FORK
);
434 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW
,
436 return XFS_ERROR(EFSCORRUPTED
);
441 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW
, ip
->i_mount
);
442 return XFS_ERROR(EFSCORRUPTED
);
447 if (!XFS_DFORK_Q(dip
))
449 ASSERT(ip
->i_afp
== NULL
);
450 ip
->i_afp
= kmem_zone_zalloc(xfs_ifork_zone
, KM_SLEEP
);
451 ip
->i_afp
->if_ext_max
=
452 XFS_IFORK_ASIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
453 switch (dip
->di_core
.di_aformat
) {
454 case XFS_DINODE_FMT_LOCAL
:
455 atp
= (xfs_attr_shortform_t
*)XFS_DFORK_APTR(dip
);
456 size
= be16_to_cpu(atp
->hdr
.totsize
);
457 error
= xfs_iformat_local(ip
, dip
, XFS_ATTR_FORK
, size
);
459 case XFS_DINODE_FMT_EXTENTS
:
460 error
= xfs_iformat_extents(ip
, dip
, XFS_ATTR_FORK
);
462 case XFS_DINODE_FMT_BTREE
:
463 error
= xfs_iformat_btree(ip
, dip
, XFS_ATTR_FORK
);
466 error
= XFS_ERROR(EFSCORRUPTED
);
470 kmem_zone_free(xfs_ifork_zone
, ip
->i_afp
);
472 xfs_idestroy_fork(ip
, XFS_DATA_FORK
);
478 * The file is in-lined in the on-disk inode.
479 * If it fits into if_inline_data, then copy
480 * it there, otherwise allocate a buffer for it
481 * and copy the data there. Either way, set
482 * if_data to point at the data.
483 * If we allocate a buffer for the data, make
484 * sure that its size is a multiple of 4 and
485 * record the real size in i_real_bytes.
498 * If the size is unreasonable, then something
499 * is wrong and we just bail out rather than crash in
500 * kmem_alloc() or memcpy() below.
502 if (unlikely(size
> XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
))) {
503 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
505 "(bad size %d for local fork, size = %d).",
506 (unsigned long long) ip
->i_ino
, size
,
507 XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
));
508 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW
,
510 return XFS_ERROR(EFSCORRUPTED
);
512 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
515 ifp
->if_u1
.if_data
= NULL
;
516 else if (size
<= sizeof(ifp
->if_u2
.if_inline_data
))
517 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
519 real_size
= roundup(size
, 4);
520 ifp
->if_u1
.if_data
= kmem_alloc(real_size
, KM_SLEEP
);
522 ifp
->if_bytes
= size
;
523 ifp
->if_real_bytes
= real_size
;
525 memcpy(ifp
->if_u1
.if_data
, XFS_DFORK_PTR(dip
, whichfork
), size
);
526 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
527 ifp
->if_flags
|= XFS_IFINLINE
;
532 * The file consists of a set of extents all
533 * of which fit into the on-disk inode.
534 * If there are few enough extents to fit into
535 * the if_inline_ext, then copy them there.
536 * Otherwise allocate a buffer for them and copy
537 * them into it. Either way, set if_extents
538 * to point at the extents.
552 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
553 nex
= XFS_DFORK_NEXTENTS(dip
, whichfork
);
554 size
= nex
* (uint
)sizeof(xfs_bmbt_rec_t
);
557 * If the number of extents is unreasonable, then something
558 * is wrong and we just bail out rather than crash in
559 * kmem_alloc() or memcpy() below.
561 if (unlikely(size
< 0 || size
> XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
))) {
562 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
563 "corrupt inode %Lu ((a)extents = %d).",
564 (unsigned long long) ip
->i_ino
, nex
);
565 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW
,
567 return XFS_ERROR(EFSCORRUPTED
);
570 ifp
->if_real_bytes
= 0;
572 ifp
->if_u1
.if_extents
= NULL
;
573 else if (nex
<= XFS_INLINE_EXTS
)
574 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
576 xfs_iext_add(ifp
, 0, nex
);
578 ifp
->if_bytes
= size
;
580 dp
= (xfs_bmbt_rec_t
*) XFS_DFORK_PTR(dip
, whichfork
);
581 xfs_validate_extents(ifp
, nex
, XFS_EXTFMT_INODE(ip
));
582 for (i
= 0; i
< nex
; i
++, dp
++) {
583 xfs_bmbt_rec_host_t
*ep
= xfs_iext_get_ext(ifp
, i
);
584 ep
->l0
= get_unaligned_be64(&dp
->l0
);
585 ep
->l1
= get_unaligned_be64(&dp
->l1
);
587 XFS_BMAP_TRACE_EXLIST(ip
, nex
, whichfork
);
588 if (whichfork
!= XFS_DATA_FORK
||
589 XFS_EXTFMT_INODE(ip
) == XFS_EXTFMT_NOSTATE
)
590 if (unlikely(xfs_check_nostate_extents(
592 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
595 return XFS_ERROR(EFSCORRUPTED
);
598 ifp
->if_flags
|= XFS_IFEXTENTS
;
603 * The file has too many extents to fit into
604 * the inode, so they are in B-tree format.
605 * Allocate a buffer for the root of the B-tree
606 * and copy the root into it. The i_extents
607 * field will remain NULL until all of the
608 * extents are read in (when they are needed).
616 xfs_bmdr_block_t
*dfp
;
622 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
623 dfp
= (xfs_bmdr_block_t
*)XFS_DFORK_PTR(dip
, whichfork
);
624 size
= XFS_BMAP_BROOT_SPACE(dfp
);
625 nrecs
= be16_to_cpu(dfp
->bb_numrecs
);
628 * blow out if -- fork has less extents than can fit in
629 * fork (fork shouldn't be a btree format), root btree
630 * block has more records than can fit into the fork,
631 * or the number of extents is greater than the number of
634 if (unlikely(XFS_IFORK_NEXTENTS(ip
, whichfork
) <= ifp
->if_ext_max
635 || XFS_BMDR_SPACE_CALC(nrecs
) >
636 XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
)
637 || XFS_IFORK_NEXTENTS(ip
, whichfork
) > ip
->i_d
.di_nblocks
)) {
638 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
639 "corrupt inode %Lu (btree).",
640 (unsigned long long) ip
->i_ino
);
641 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW
,
643 return XFS_ERROR(EFSCORRUPTED
);
646 ifp
->if_broot_bytes
= size
;
647 ifp
->if_broot
= kmem_alloc(size
, KM_SLEEP
);
648 ASSERT(ifp
->if_broot
!= NULL
);
650 * Copy and convert from the on-disk structure
651 * to the in-memory structure.
653 xfs_bmdr_to_bmbt(ip
->i_mount
, dfp
,
654 XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
),
655 ifp
->if_broot
, size
);
656 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
657 ifp
->if_flags
|= XFS_IFBROOT
;
663 xfs_dinode_from_disk(
665 xfs_dinode_core_t
*from
)
667 to
->di_magic
= be16_to_cpu(from
->di_magic
);
668 to
->di_mode
= be16_to_cpu(from
->di_mode
);
669 to
->di_version
= from
->di_version
;
670 to
->di_format
= from
->di_format
;
671 to
->di_onlink
= be16_to_cpu(from
->di_onlink
);
672 to
->di_uid
= be32_to_cpu(from
->di_uid
);
673 to
->di_gid
= be32_to_cpu(from
->di_gid
);
674 to
->di_nlink
= be32_to_cpu(from
->di_nlink
);
675 to
->di_projid
= be16_to_cpu(from
->di_projid
);
676 memcpy(to
->di_pad
, from
->di_pad
, sizeof(to
->di_pad
));
677 to
->di_flushiter
= be16_to_cpu(from
->di_flushiter
);
678 to
->di_atime
.t_sec
= be32_to_cpu(from
->di_atime
.t_sec
);
679 to
->di_atime
.t_nsec
= be32_to_cpu(from
->di_atime
.t_nsec
);
680 to
->di_mtime
.t_sec
= be32_to_cpu(from
->di_mtime
.t_sec
);
681 to
->di_mtime
.t_nsec
= be32_to_cpu(from
->di_mtime
.t_nsec
);
682 to
->di_ctime
.t_sec
= be32_to_cpu(from
->di_ctime
.t_sec
);
683 to
->di_ctime
.t_nsec
= be32_to_cpu(from
->di_ctime
.t_nsec
);
684 to
->di_size
= be64_to_cpu(from
->di_size
);
685 to
->di_nblocks
= be64_to_cpu(from
->di_nblocks
);
686 to
->di_extsize
= be32_to_cpu(from
->di_extsize
);
687 to
->di_nextents
= be32_to_cpu(from
->di_nextents
);
688 to
->di_anextents
= be16_to_cpu(from
->di_anextents
);
689 to
->di_forkoff
= from
->di_forkoff
;
690 to
->di_aformat
= from
->di_aformat
;
691 to
->di_dmevmask
= be32_to_cpu(from
->di_dmevmask
);
692 to
->di_dmstate
= be16_to_cpu(from
->di_dmstate
);
693 to
->di_flags
= be16_to_cpu(from
->di_flags
);
694 to
->di_gen
= be32_to_cpu(from
->di_gen
);
699 xfs_dinode_core_t
*to
,
700 xfs_icdinode_t
*from
)
702 to
->di_magic
= cpu_to_be16(from
->di_magic
);
703 to
->di_mode
= cpu_to_be16(from
->di_mode
);
704 to
->di_version
= from
->di_version
;
705 to
->di_format
= from
->di_format
;
706 to
->di_onlink
= cpu_to_be16(from
->di_onlink
);
707 to
->di_uid
= cpu_to_be32(from
->di_uid
);
708 to
->di_gid
= cpu_to_be32(from
->di_gid
);
709 to
->di_nlink
= cpu_to_be32(from
->di_nlink
);
710 to
->di_projid
= cpu_to_be16(from
->di_projid
);
711 memcpy(to
->di_pad
, from
->di_pad
, sizeof(to
->di_pad
));
712 to
->di_flushiter
= cpu_to_be16(from
->di_flushiter
);
713 to
->di_atime
.t_sec
= cpu_to_be32(from
->di_atime
.t_sec
);
714 to
->di_atime
.t_nsec
= cpu_to_be32(from
->di_atime
.t_nsec
);
715 to
->di_mtime
.t_sec
= cpu_to_be32(from
->di_mtime
.t_sec
);
716 to
->di_mtime
.t_nsec
= cpu_to_be32(from
->di_mtime
.t_nsec
);
717 to
->di_ctime
.t_sec
= cpu_to_be32(from
->di_ctime
.t_sec
);
718 to
->di_ctime
.t_nsec
= cpu_to_be32(from
->di_ctime
.t_nsec
);
719 to
->di_size
= cpu_to_be64(from
->di_size
);
720 to
->di_nblocks
= cpu_to_be64(from
->di_nblocks
);
721 to
->di_extsize
= cpu_to_be32(from
->di_extsize
);
722 to
->di_nextents
= cpu_to_be32(from
->di_nextents
);
723 to
->di_anextents
= cpu_to_be16(from
->di_anextents
);
724 to
->di_forkoff
= from
->di_forkoff
;
725 to
->di_aformat
= from
->di_aformat
;
726 to
->di_dmevmask
= cpu_to_be32(from
->di_dmevmask
);
727 to
->di_dmstate
= cpu_to_be16(from
->di_dmstate
);
728 to
->di_flags
= cpu_to_be16(from
->di_flags
);
729 to
->di_gen
= cpu_to_be32(from
->di_gen
);
738 if (di_flags
& XFS_DIFLAG_ANY
) {
739 if (di_flags
& XFS_DIFLAG_REALTIME
)
740 flags
|= XFS_XFLAG_REALTIME
;
741 if (di_flags
& XFS_DIFLAG_PREALLOC
)
742 flags
|= XFS_XFLAG_PREALLOC
;
743 if (di_flags
& XFS_DIFLAG_IMMUTABLE
)
744 flags
|= XFS_XFLAG_IMMUTABLE
;
745 if (di_flags
& XFS_DIFLAG_APPEND
)
746 flags
|= XFS_XFLAG_APPEND
;
747 if (di_flags
& XFS_DIFLAG_SYNC
)
748 flags
|= XFS_XFLAG_SYNC
;
749 if (di_flags
& XFS_DIFLAG_NOATIME
)
750 flags
|= XFS_XFLAG_NOATIME
;
751 if (di_flags
& XFS_DIFLAG_NODUMP
)
752 flags
|= XFS_XFLAG_NODUMP
;
753 if (di_flags
& XFS_DIFLAG_RTINHERIT
)
754 flags
|= XFS_XFLAG_RTINHERIT
;
755 if (di_flags
& XFS_DIFLAG_PROJINHERIT
)
756 flags
|= XFS_XFLAG_PROJINHERIT
;
757 if (di_flags
& XFS_DIFLAG_NOSYMLINKS
)
758 flags
|= XFS_XFLAG_NOSYMLINKS
;
759 if (di_flags
& XFS_DIFLAG_EXTSIZE
)
760 flags
|= XFS_XFLAG_EXTSIZE
;
761 if (di_flags
& XFS_DIFLAG_EXTSZINHERIT
)
762 flags
|= XFS_XFLAG_EXTSZINHERIT
;
763 if (di_flags
& XFS_DIFLAG_NODEFRAG
)
764 flags
|= XFS_XFLAG_NODEFRAG
;
765 if (di_flags
& XFS_DIFLAG_FILESTREAM
)
766 flags
|= XFS_XFLAG_FILESTREAM
;
776 xfs_icdinode_t
*dic
= &ip
->i_d
;
778 return _xfs_dic2xflags(dic
->di_flags
) |
779 (XFS_IFORK_Q(ip
) ? XFS_XFLAG_HASATTR
: 0);
786 xfs_dinode_core_t
*dic
= &dip
->di_core
;
788 return _xfs_dic2xflags(be16_to_cpu(dic
->di_flags
)) |
789 (XFS_DFORK_Q(dip
) ? XFS_XFLAG_HASATTR
: 0);
793 * Allocate and initialise an xfs_inode.
797 struct xfs_mount
*mp
,
800 struct xfs_inode
*ip
;
803 * if this didn't occur in transactions, we could use
804 * KM_MAYFAIL and return NULL here on ENOMEM. Set the
805 * code up to do this anyway.
807 ip
= kmem_zone_alloc(xfs_inode_zone
, KM_SLEEP
);
811 ASSERT(atomic_read(&ip
->i_iocount
) == 0);
812 ASSERT(atomic_read(&ip
->i_pincount
) == 0);
813 ASSERT(!spin_is_locked(&ip
->i_flags_lock
));
814 ASSERT(list_empty(&ip
->i_reclaim
));
817 * initialise the VFS inode here to get failures
818 * out of the way early.
820 if (!inode_init_always(mp
->m_super
, VFS_I(ip
))) {
821 kmem_zone_free(xfs_inode_zone
, ip
);
825 /* initialise the xfs inode */
832 memset(&ip
->i_df
, 0, sizeof(xfs_ifork_t
));
834 ip
->i_update_core
= 0;
835 ip
->i_update_size
= 0;
836 ip
->i_delayed_blks
= 0;
837 memset(&ip
->i_d
, 0, sizeof(xfs_icdinode_t
));
842 * Initialize inode's trace buffers.
844 #ifdef XFS_INODE_TRACE
845 ip
->i_trace
= ktrace_alloc(INODE_TRACE_SIZE
, KM_NOFS
);
847 #ifdef XFS_BMAP_TRACE
848 ip
->i_xtrace
= ktrace_alloc(XFS_BMAP_KTRACE_SIZE
, KM_NOFS
);
850 #ifdef XFS_BTREE_TRACE
851 ip
->i_btrace
= ktrace_alloc(XFS_BMBT_KTRACE_SIZE
, KM_NOFS
);
854 ip
->i_rwtrace
= ktrace_alloc(XFS_RW_KTRACE_SIZE
, KM_NOFS
);
856 #ifdef XFS_ILOCK_TRACE
857 ip
->i_lock_trace
= ktrace_alloc(XFS_ILOCK_KTRACE_SIZE
, KM_NOFS
);
859 #ifdef XFS_DIR2_TRACE
860 ip
->i_dir_trace
= ktrace_alloc(XFS_DIR2_KTRACE_SIZE
, KM_NOFS
);
867 * Given a mount structure and an inode number, return a pointer
868 * to a newly allocated in-core inode corresponding to the given
871 * Initialize the inode's attributes and extent pointers if it
872 * already has them (it will not if the inode has no links).
888 ip
= xfs_inode_alloc(mp
, ino
);
893 * Get pointer's to the on-disk inode and the buffer containing it.
894 * If the inode number refers to a block outside the file system
895 * then xfs_itobp() will return NULL. In this case we should
896 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
897 * know that this is a new incore inode.
899 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &bp
, bno
, imap_flags
, XFS_BUF_LOCK
);
906 * If we got something that isn't an inode it means someone
907 * (nfs or dmi) has a stale handle.
909 if (be16_to_cpu(dip
->di_core
.di_magic
) != XFS_DINODE_MAGIC
) {
911 xfs_trans_brelse(tp
, bp
);
913 xfs_fs_cmn_err(CE_ALERT
, mp
, "xfs_iread: "
914 "dip->di_core.di_magic (0x%x) != "
915 "XFS_DINODE_MAGIC (0x%x)",
916 be16_to_cpu(dip
->di_core
.di_magic
),
919 return XFS_ERROR(EINVAL
);
923 * If the on-disk inode is already linked to a directory
924 * entry, copy all of the inode into the in-core inode.
925 * xfs_iformat() handles copying in the inode format
926 * specific information.
927 * Otherwise, just get the truly permanent information.
929 if (dip
->di_core
.di_mode
) {
930 xfs_dinode_from_disk(&ip
->i_d
, &dip
->di_core
);
931 error
= xfs_iformat(ip
, dip
);
934 xfs_trans_brelse(tp
, bp
);
936 xfs_fs_cmn_err(CE_ALERT
, mp
, "xfs_iread: "
937 "xfs_iformat() returned error %d",
943 ip
->i_d
.di_magic
= be16_to_cpu(dip
->di_core
.di_magic
);
944 ip
->i_d
.di_version
= dip
->di_core
.di_version
;
945 ip
->i_d
.di_gen
= be32_to_cpu(dip
->di_core
.di_gen
);
946 ip
->i_d
.di_flushiter
= be16_to_cpu(dip
->di_core
.di_flushiter
);
948 * Make sure to pull in the mode here as well in
949 * case the inode is released without being used.
950 * This ensures that xfs_inactive() will see that
951 * the inode is already free and not try to mess
952 * with the uninitialized part of it.
956 * Initialize the per-fork minima and maxima for a new
957 * inode here. xfs_iformat will do it for old inodes.
959 ip
->i_df
.if_ext_max
=
960 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
964 * The inode format changed when we moved the link count and
965 * made it 32 bits long. If this is an old format inode,
966 * convert it in memory to look like a new one. If it gets
967 * flushed to disk we will convert back before flushing or
968 * logging it. We zero out the new projid field and the old link
969 * count field. We'll handle clearing the pad field (the remains
970 * of the old uuid field) when we actually convert the inode to
971 * the new format. We don't change the version number so that we
972 * can distinguish this from a real new format inode.
974 if (ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
) {
975 ip
->i_d
.di_nlink
= ip
->i_d
.di_onlink
;
976 ip
->i_d
.di_onlink
= 0;
977 ip
->i_d
.di_projid
= 0;
980 ip
->i_delayed_blks
= 0;
981 ip
->i_size
= ip
->i_d
.di_size
;
984 * Mark the buffer containing the inode as something to keep
985 * around for a while. This helps to keep recently accessed
986 * meta-data in-core longer.
988 XFS_BUF_SET_REF(bp
, XFS_INO_REF
);
991 * Use xfs_trans_brelse() to release the buffer containing the
992 * on-disk inode, because it was acquired with xfs_trans_read_buf()
993 * in xfs_itobp() above. If tp is NULL, this is just a normal
994 * brelse(). If we're within a transaction, then xfs_trans_brelse()
995 * will only release the buffer if it is not dirty within the
996 * transaction. It will be OK to release the buffer in this case,
997 * because inodes on disk are never destroyed and we will be
998 * locking the new in-core inode before putting it in the hash
999 * table where other processes can find it. Thus we don't have
1000 * to worry about the inode being changed just because we released
1003 xfs_trans_brelse(tp
, bp
);
1009 * Read in extents from a btree-format inode.
1010 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1020 xfs_extnum_t nextents
;
1023 if (unlikely(XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
)) {
1024 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW
,
1026 return XFS_ERROR(EFSCORRUPTED
);
1028 nextents
= XFS_IFORK_NEXTENTS(ip
, whichfork
);
1029 size
= nextents
* sizeof(xfs_bmbt_rec_t
);
1030 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1033 * We know that the size is valid (it's checked in iformat_btree)
1035 ifp
->if_lastex
= NULLEXTNUM
;
1036 ifp
->if_bytes
= ifp
->if_real_bytes
= 0;
1037 ifp
->if_flags
|= XFS_IFEXTENTS
;
1038 xfs_iext_add(ifp
, 0, nextents
);
1039 error
= xfs_bmap_read_extents(tp
, ip
, whichfork
);
1041 xfs_iext_destroy(ifp
);
1042 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
1045 xfs_validate_extents(ifp
, nextents
, XFS_EXTFMT_INODE(ip
));
1050 * Allocate an inode on disk and return a copy of its in-core version.
1051 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1052 * appropriately within the inode. The uid and gid for the inode are
1053 * set according to the contents of the given cred structure.
1055 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1056 * has a free inode available, call xfs_iget()
1057 * to obtain the in-core version of the allocated inode. Finally,
1058 * fill in the inode and log its initial contents. In this case,
1059 * ialloc_context would be set to NULL and call_again set to false.
1061 * If xfs_dialloc() does not have an available inode,
1062 * it will replenish its supply by doing an allocation. Since we can
1063 * only do one allocation within a transaction without deadlocks, we
1064 * must commit the current transaction before returning the inode itself.
1065 * In this case, therefore, we will set call_again to true and return.
1066 * The caller should then commit the current transaction, start a new
1067 * transaction, and call xfs_ialloc() again to actually get the inode.
1069 * To ensure that some other process does not grab the inode that
1070 * was allocated during the first call to xfs_ialloc(), this routine
1071 * also returns the [locked] bp pointing to the head of the freelist
1072 * as ialloc_context. The caller should hold this buffer across
1073 * the commit and pass it back into this routine on the second call.
1075 * If we are allocating quota inodes, we do not have a parent inode
1076 * to attach to or associate with (i.e. pip == NULL) because they
1077 * are not linked into the directory structure - they are attached
1078 * directly to the superblock - and so have no parent.
1090 xfs_buf_t
**ialloc_context
,
1091 boolean_t
*call_again
,
1099 int filestreams
= 0;
1102 * Call the space management code to pick
1103 * the on-disk inode to be allocated.
1105 error
= xfs_dialloc(tp
, pip
? pip
->i_ino
: 0, mode
, okalloc
,
1106 ialloc_context
, call_again
, &ino
);
1109 if (*call_again
|| ino
== NULLFSINO
) {
1113 ASSERT(*ialloc_context
== NULL
);
1116 * Get the in-core inode with the lock held exclusively.
1117 * This is because we're setting fields here we need
1118 * to prevent others from looking at until we're done.
1120 error
= xfs_trans_iget(tp
->t_mountp
, tp
, ino
,
1121 XFS_IGET_CREATE
, XFS_ILOCK_EXCL
, &ip
);
1126 ip
->i_d
.di_mode
= (__uint16_t
)mode
;
1127 ip
->i_d
.di_onlink
= 0;
1128 ip
->i_d
.di_nlink
= nlink
;
1129 ASSERT(ip
->i_d
.di_nlink
== nlink
);
1130 ip
->i_d
.di_uid
= current_fsuid();
1131 ip
->i_d
.di_gid
= current_fsgid();
1132 ip
->i_d
.di_projid
= prid
;
1133 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
1136 * If the superblock version is up to where we support new format
1137 * inodes and this is currently an old format inode, then change
1138 * the inode version number now. This way we only do the conversion
1139 * here rather than here and in the flush/logging code.
1141 if (xfs_sb_version_hasnlink(&tp
->t_mountp
->m_sb
) &&
1142 ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
) {
1143 ip
->i_d
.di_version
= XFS_DINODE_VERSION_2
;
1145 * We've already zeroed the old link count, the projid field,
1146 * and the pad field.
1151 * Project ids won't be stored on disk if we are using a version 1 inode.
1153 if ((prid
!= 0) && (ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
))
1154 xfs_bump_ino_vers2(tp
, ip
);
1156 if (pip
&& XFS_INHERIT_GID(pip
)) {
1157 ip
->i_d
.di_gid
= pip
->i_d
.di_gid
;
1158 if ((pip
->i_d
.di_mode
& S_ISGID
) && (mode
& S_IFMT
) == S_IFDIR
) {
1159 ip
->i_d
.di_mode
|= S_ISGID
;
1164 * If the group ID of the new file does not match the effective group
1165 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1166 * (and only if the irix_sgid_inherit compatibility variable is set).
1168 if ((irix_sgid_inherit
) &&
1169 (ip
->i_d
.di_mode
& S_ISGID
) &&
1170 (!in_group_p((gid_t
)ip
->i_d
.di_gid
))) {
1171 ip
->i_d
.di_mode
&= ~S_ISGID
;
1174 ip
->i_d
.di_size
= 0;
1176 ip
->i_d
.di_nextents
= 0;
1177 ASSERT(ip
->i_d
.di_nblocks
== 0);
1180 ip
->i_d
.di_mtime
.t_sec
= (__int32_t
)tv
.tv_sec
;
1181 ip
->i_d
.di_mtime
.t_nsec
= (__int32_t
)tv
.tv_nsec
;
1182 ip
->i_d
.di_atime
= ip
->i_d
.di_mtime
;
1183 ip
->i_d
.di_ctime
= ip
->i_d
.di_mtime
;
1186 * di_gen will have been taken care of in xfs_iread.
1188 ip
->i_d
.di_extsize
= 0;
1189 ip
->i_d
.di_dmevmask
= 0;
1190 ip
->i_d
.di_dmstate
= 0;
1191 ip
->i_d
.di_flags
= 0;
1192 flags
= XFS_ILOG_CORE
;
1193 switch (mode
& S_IFMT
) {
1198 ip
->i_d
.di_format
= XFS_DINODE_FMT_DEV
;
1199 ip
->i_df
.if_u2
.if_rdev
= rdev
;
1200 ip
->i_df
.if_flags
= 0;
1201 flags
|= XFS_ILOG_DEV
;
1205 * we can't set up filestreams until after the VFS inode
1206 * is set up properly.
1208 if (pip
&& xfs_inode_is_filestream(pip
))
1212 if (pip
&& (pip
->i_d
.di_flags
& XFS_DIFLAG_ANY
)) {
1215 if ((mode
& S_IFMT
) == S_IFDIR
) {
1216 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
1217 di_flags
|= XFS_DIFLAG_RTINHERIT
;
1218 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
1219 di_flags
|= XFS_DIFLAG_EXTSZINHERIT
;
1220 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
1222 } else if ((mode
& S_IFMT
) == S_IFREG
) {
1223 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
1224 di_flags
|= XFS_DIFLAG_REALTIME
;
1225 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
1226 di_flags
|= XFS_DIFLAG_EXTSIZE
;
1227 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
1230 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOATIME
) &&
1231 xfs_inherit_noatime
)
1232 di_flags
|= XFS_DIFLAG_NOATIME
;
1233 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODUMP
) &&
1235 di_flags
|= XFS_DIFLAG_NODUMP
;
1236 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_SYNC
) &&
1238 di_flags
|= XFS_DIFLAG_SYNC
;
1239 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOSYMLINKS
) &&
1240 xfs_inherit_nosymlinks
)
1241 di_flags
|= XFS_DIFLAG_NOSYMLINKS
;
1242 if (pip
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
1243 di_flags
|= XFS_DIFLAG_PROJINHERIT
;
1244 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODEFRAG
) &&
1245 xfs_inherit_nodefrag
)
1246 di_flags
|= XFS_DIFLAG_NODEFRAG
;
1247 if (pip
->i_d
.di_flags
& XFS_DIFLAG_FILESTREAM
)
1248 di_flags
|= XFS_DIFLAG_FILESTREAM
;
1249 ip
->i_d
.di_flags
|= di_flags
;
1253 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
1254 ip
->i_df
.if_flags
= XFS_IFEXTENTS
;
1255 ip
->i_df
.if_bytes
= ip
->i_df
.if_real_bytes
= 0;
1256 ip
->i_df
.if_u1
.if_extents
= NULL
;
1262 * Attribute fork settings for new inode.
1264 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
1265 ip
->i_d
.di_anextents
= 0;
1268 * Log the new values stuffed into the inode.
1270 xfs_trans_log_inode(tp
, ip
, flags
);
1272 /* now that we have an i_mode we can setup inode ops and unlock */
1273 xfs_setup_inode(ip
);
1275 /* now we have set up the vfs inode we can associate the filestream */
1277 error
= xfs_filestream_associate(pip
, ip
);
1281 xfs_iflags_set(ip
, XFS_IFILESTREAM
);
1289 * Check to make sure that there are no blocks allocated to the
1290 * file beyond the size of the file. We don't check this for
1291 * files with fixed size extents or real time extents, but we
1292 * at least do it for regular files.
1301 xfs_fileoff_t map_first
;
1303 xfs_bmbt_irec_t imaps
[2];
1305 if ((ip
->i_d
.di_mode
& S_IFMT
) != S_IFREG
)
1308 if (XFS_IS_REALTIME_INODE(ip
))
1311 if (ip
->i_d
.di_flags
& XFS_DIFLAG_EXTSIZE
)
1315 map_first
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)isize
);
1317 * The filesystem could be shutting down, so bmapi may return
1320 if (xfs_bmapi(NULL
, ip
, map_first
,
1322 (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
)) -
1324 XFS_BMAPI_ENTIRE
, NULL
, 0, imaps
, &nimaps
,
1327 ASSERT(nimaps
== 1);
1328 ASSERT(imaps
[0].br_startblock
== HOLESTARTBLOCK
);
1333 * Calculate the last possible buffered byte in a file. This must
1334 * include data that was buffered beyond the EOF by the write code.
1335 * This also needs to deal with overflowing the xfs_fsize_t type
1336 * which can happen for sizes near the limit.
1338 * We also need to take into account any blocks beyond the EOF. It
1339 * may be the case that they were buffered by a write which failed.
1340 * In that case the pages will still be in memory, but the inode size
1341 * will never have been updated.
1348 xfs_fsize_t last_byte
;
1349 xfs_fileoff_t last_block
;
1350 xfs_fileoff_t size_last_block
;
1353 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
|XFS_IOLOCK_SHARED
));
1357 * Only check for blocks beyond the EOF if the extents have
1358 * been read in. This eliminates the need for the inode lock,
1359 * and it also saves us from looking when it really isn't
1362 if (ip
->i_df
.if_flags
& XFS_IFEXTENTS
) {
1363 error
= xfs_bmap_last_offset(NULL
, ip
, &last_block
,
1371 size_last_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)ip
->i_size
);
1372 last_block
= XFS_FILEOFF_MAX(last_block
, size_last_block
);
1374 last_byte
= XFS_FSB_TO_B(mp
, last_block
);
1375 if (last_byte
< 0) {
1376 return XFS_MAXIOFFSET(mp
);
1378 last_byte
+= (1 << mp
->m_writeio_log
);
1379 if (last_byte
< 0) {
1380 return XFS_MAXIOFFSET(mp
);
1385 #if defined(XFS_RW_TRACE)
1391 xfs_fsize_t new_size
,
1392 xfs_off_t toss_start
,
1393 xfs_off_t toss_finish
)
1395 if (ip
->i_rwtrace
== NULL
) {
1399 ktrace_enter(ip
->i_rwtrace
,
1402 (void*)(unsigned long)((ip
->i_d
.di_size
>> 32) & 0xffffffff),
1403 (void*)(unsigned long)(ip
->i_d
.di_size
& 0xffffffff),
1404 (void*)((long)flag
),
1405 (void*)(unsigned long)((new_size
>> 32) & 0xffffffff),
1406 (void*)(unsigned long)(new_size
& 0xffffffff),
1407 (void*)(unsigned long)((toss_start
>> 32) & 0xffffffff),
1408 (void*)(unsigned long)(toss_start
& 0xffffffff),
1409 (void*)(unsigned long)((toss_finish
>> 32) & 0xffffffff),
1410 (void*)(unsigned long)(toss_finish
& 0xffffffff),
1411 (void*)(unsigned long)current_cpu(),
1412 (void*)(unsigned long)current_pid(),
1418 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1422 * Start the truncation of the file to new_size. The new size
1423 * must be smaller than the current size. This routine will
1424 * clear the buffer and page caches of file data in the removed
1425 * range, and xfs_itruncate_finish() will remove the underlying
1428 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1429 * must NOT have the inode lock held at all. This is because we're
1430 * calling into the buffer/page cache code and we can't hold the
1431 * inode lock when we do so.
1433 * We need to wait for any direct I/Os in flight to complete before we
1434 * proceed with the truncate. This is needed to prevent the extents
1435 * being read or written by the direct I/Os from being removed while the
1436 * I/O is in flight as there is no other method of synchronising
1437 * direct I/O with the truncate operation. Also, because we hold
1438 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1439 * started until the truncate completes and drops the lock. Essentially,
1440 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1441 * between direct I/Os and the truncate operation.
1443 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1444 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1445 * in the case that the caller is locking things out of order and
1446 * may not be able to call xfs_itruncate_finish() with the inode lock
1447 * held without dropping the I/O lock. If the caller must drop the
1448 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1449 * must be called again with all the same restrictions as the initial
1453 xfs_itruncate_start(
1456 xfs_fsize_t new_size
)
1458 xfs_fsize_t last_byte
;
1459 xfs_off_t toss_start
;
1463 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
1464 ASSERT((new_size
== 0) || (new_size
<= ip
->i_size
));
1465 ASSERT((flags
== XFS_ITRUNC_DEFINITE
) ||
1466 (flags
== XFS_ITRUNC_MAYBE
));
1470 /* wait for the completion of any pending DIOs */
1471 if (new_size
== 0 || new_size
< ip
->i_size
)
1475 * Call toss_pages or flushinval_pages to get rid of pages
1476 * overlapping the region being removed. We have to use
1477 * the less efficient flushinval_pages in the case that the
1478 * caller may not be able to finish the truncate without
1479 * dropping the inode's I/O lock. Make sure
1480 * to catch any pages brought in by buffers overlapping
1481 * the EOF by searching out beyond the isize by our
1482 * block size. We round new_size up to a block boundary
1483 * so that we don't toss things on the same block as
1484 * new_size but before it.
1486 * Before calling toss_page or flushinval_pages, make sure to
1487 * call remapf() over the same region if the file is mapped.
1488 * This frees up mapped file references to the pages in the
1489 * given range and for the flushinval_pages case it ensures
1490 * that we get the latest mapped changes flushed out.
1492 toss_start
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)new_size
);
1493 toss_start
= XFS_FSB_TO_B(mp
, toss_start
);
1494 if (toss_start
< 0) {
1496 * The place to start tossing is beyond our maximum
1497 * file size, so there is no way that the data extended
1502 last_byte
= xfs_file_last_byte(ip
);
1503 xfs_itrunc_trace(XFS_ITRUNC_START
, ip
, flags
, new_size
, toss_start
,
1505 if (last_byte
> toss_start
) {
1506 if (flags
& XFS_ITRUNC_DEFINITE
) {
1507 xfs_tosspages(ip
, toss_start
,
1508 -1, FI_REMAPF_LOCKED
);
1510 error
= xfs_flushinval_pages(ip
, toss_start
,
1511 -1, FI_REMAPF_LOCKED
);
1516 if (new_size
== 0) {
1517 ASSERT(VN_CACHED(VFS_I(ip
)) == 0);
1524 * Shrink the file to the given new_size. The new size must be smaller than
1525 * the current size. This will free up the underlying blocks in the removed
1526 * range after a call to xfs_itruncate_start() or xfs_atruncate_start().
1528 * The transaction passed to this routine must have made a permanent log
1529 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1530 * given transaction and start new ones, so make sure everything involved in
1531 * the transaction is tidy before calling here. Some transaction will be
1532 * returned to the caller to be committed. The incoming transaction must
1533 * already include the inode, and both inode locks must be held exclusively.
1534 * The inode must also be "held" within the transaction. On return the inode
1535 * will be "held" within the returned transaction. This routine does NOT
1536 * require any disk space to be reserved for it within the transaction.
1538 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it
1539 * indicates the fork which is to be truncated. For the attribute fork we only
1540 * support truncation to size 0.
1542 * We use the sync parameter to indicate whether or not the first transaction
1543 * we perform might have to be synchronous. For the attr fork, it needs to be
1544 * so if the unlink of the inode is not yet known to be permanent in the log.
1545 * This keeps us from freeing and reusing the blocks of the attribute fork
1546 * before the unlink of the inode becomes permanent.
1548 * For the data fork, we normally have to run synchronously if we're being
1549 * called out of the inactive path or we're being called out of the create path
1550 * where we're truncating an existing file. Either way, the truncate needs to
1551 * be sync so blocks don't reappear in the file with altered data in case of a
1552 * crash. wsync filesystems can run the first case async because anything that
1553 * shrinks the inode has to run sync so by the time we're called here from
1554 * inactive, the inode size is permanently set to 0.
1556 * Calls from the truncate path always need to be sync unless we're in a wsync
1557 * filesystem and the file has already been unlinked.
1559 * The caller is responsible for correctly setting the sync parameter. It gets
1560 * too hard for us to guess here which path we're being called out of just
1561 * based on inode state.
1563 * If we get an error, we must return with the inode locked and linked into the
1564 * current transaction. This keeps things simple for the higher level code,
1565 * because it always knows that the inode is locked and held in the transaction
1566 * that returns to it whether errors occur or not. We don't mark the inode
1567 * dirty on error so that transactions can be easily aborted if possible.
1570 xfs_itruncate_finish(
1573 xfs_fsize_t new_size
,
1577 xfs_fsblock_t first_block
;
1578 xfs_fileoff_t first_unmap_block
;
1579 xfs_fileoff_t last_block
;
1580 xfs_filblks_t unmap_len
=0;
1585 xfs_bmap_free_t free_list
;
1588 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_IOLOCK_EXCL
));
1589 ASSERT((new_size
== 0) || (new_size
<= ip
->i_size
));
1590 ASSERT(*tp
!= NULL
);
1591 ASSERT((*tp
)->t_flags
& XFS_TRANS_PERM_LOG_RES
);
1592 ASSERT(ip
->i_transp
== *tp
);
1593 ASSERT(ip
->i_itemp
!= NULL
);
1594 ASSERT(ip
->i_itemp
->ili_flags
& XFS_ILI_HOLD
);
1598 mp
= (ntp
)->t_mountp
;
1599 ASSERT(! XFS_NOT_DQATTACHED(mp
, ip
));
1602 * We only support truncating the entire attribute fork.
1604 if (fork
== XFS_ATTR_FORK
) {
1607 first_unmap_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)new_size
);
1608 xfs_itrunc_trace(XFS_ITRUNC_FINISH1
, ip
, 0, new_size
, 0, 0);
1610 * The first thing we do is set the size to new_size permanently
1611 * on disk. This way we don't have to worry about anyone ever
1612 * being able to look at the data being freed even in the face
1613 * of a crash. What we're getting around here is the case where
1614 * we free a block, it is allocated to another file, it is written
1615 * to, and then we crash. If the new data gets written to the
1616 * file but the log buffers containing the free and reallocation
1617 * don't, then we'd end up with garbage in the blocks being freed.
1618 * As long as we make the new_size permanent before actually
1619 * freeing any blocks it doesn't matter if they get writtten to.
1621 * The callers must signal into us whether or not the size
1622 * setting here must be synchronous. There are a few cases
1623 * where it doesn't have to be synchronous. Those cases
1624 * occur if the file is unlinked and we know the unlink is
1625 * permanent or if the blocks being truncated are guaranteed
1626 * to be beyond the inode eof (regardless of the link count)
1627 * and the eof value is permanent. Both of these cases occur
1628 * only on wsync-mounted filesystems. In those cases, we're
1629 * guaranteed that no user will ever see the data in the blocks
1630 * that are being truncated so the truncate can run async.
1631 * In the free beyond eof case, the file may wind up with
1632 * more blocks allocated to it than it needs if we crash
1633 * and that won't get fixed until the next time the file
1634 * is re-opened and closed but that's ok as that shouldn't
1635 * be too many blocks.
1637 * However, we can't just make all wsync xactions run async
1638 * because there's one call out of the create path that needs
1639 * to run sync where it's truncating an existing file to size
1640 * 0 whose size is > 0.
1642 * It's probably possible to come up with a test in this
1643 * routine that would correctly distinguish all the above
1644 * cases from the values of the function parameters and the
1645 * inode state but for sanity's sake, I've decided to let the
1646 * layers above just tell us. It's simpler to correctly figure
1647 * out in the layer above exactly under what conditions we
1648 * can run async and I think it's easier for others read and
1649 * follow the logic in case something has to be changed.
1650 * cscope is your friend -- rcc.
1652 * The attribute fork is much simpler.
1654 * For the attribute fork we allow the caller to tell us whether
1655 * the unlink of the inode that led to this call is yet permanent
1656 * in the on disk log. If it is not and we will be freeing extents
1657 * in this inode then we make the first transaction synchronous
1658 * to make sure that the unlink is permanent by the time we free
1661 if (fork
== XFS_DATA_FORK
) {
1662 if (ip
->i_d
.di_nextents
> 0) {
1664 * If we are not changing the file size then do
1665 * not update the on-disk file size - we may be
1666 * called from xfs_inactive_free_eofblocks(). If we
1667 * update the on-disk file size and then the system
1668 * crashes before the contents of the file are
1669 * flushed to disk then the files may be full of
1670 * holes (ie NULL files bug).
1672 if (ip
->i_size
!= new_size
) {
1673 ip
->i_d
.di_size
= new_size
;
1674 ip
->i_size
= new_size
;
1675 xfs_trans_log_inode(ntp
, ip
, XFS_ILOG_CORE
);
1679 ASSERT(!(mp
->m_flags
& XFS_MOUNT_WSYNC
));
1680 if (ip
->i_d
.di_anextents
> 0)
1681 xfs_trans_set_sync(ntp
);
1683 ASSERT(fork
== XFS_DATA_FORK
||
1684 (fork
== XFS_ATTR_FORK
&&
1685 ((sync
&& !(mp
->m_flags
& XFS_MOUNT_WSYNC
)) ||
1686 (sync
== 0 && (mp
->m_flags
& XFS_MOUNT_WSYNC
)))));
1689 * Since it is possible for space to become allocated beyond
1690 * the end of the file (in a crash where the space is allocated
1691 * but the inode size is not yet updated), simply remove any
1692 * blocks which show up between the new EOF and the maximum
1693 * possible file size. If the first block to be removed is
1694 * beyond the maximum file size (ie it is the same as last_block),
1695 * then there is nothing to do.
1697 last_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
));
1698 ASSERT(first_unmap_block
<= last_block
);
1700 if (last_block
== first_unmap_block
) {
1703 unmap_len
= last_block
- first_unmap_block
+ 1;
1707 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1708 * will tell us whether it freed the entire range or
1709 * not. If this is a synchronous mount (wsync),
1710 * then we can tell bunmapi to keep all the
1711 * transactions asynchronous since the unlink
1712 * transaction that made this inode inactive has
1713 * already hit the disk. There's no danger of
1714 * the freed blocks being reused, there being a
1715 * crash, and the reused blocks suddenly reappearing
1716 * in this file with garbage in them once recovery
1719 XFS_BMAP_INIT(&free_list
, &first_block
);
1720 error
= xfs_bunmapi(ntp
, ip
,
1721 first_unmap_block
, unmap_len
,
1722 XFS_BMAPI_AFLAG(fork
) |
1723 (sync
? 0 : XFS_BMAPI_ASYNC
),
1724 XFS_ITRUNC_MAX_EXTENTS
,
1725 &first_block
, &free_list
,
1729 * If the bunmapi call encounters an error,
1730 * return to the caller where the transaction
1731 * can be properly aborted. We just need to
1732 * make sure we're not holding any resources
1733 * that we were not when we came in.
1735 xfs_bmap_cancel(&free_list
);
1740 * Duplicate the transaction that has the permanent
1741 * reservation and commit the old transaction.
1743 error
= xfs_bmap_finish(tp
, &free_list
, &committed
);
1746 /* link the inode into the next xact in the chain */
1747 xfs_trans_ijoin(ntp
, ip
,
1748 XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1749 xfs_trans_ihold(ntp
, ip
);
1754 * If the bmap finish call encounters an error, return
1755 * to the caller where the transaction can be properly
1756 * aborted. We just need to make sure we're not
1757 * holding any resources that we were not when we came
1760 * Aborting from this point might lose some blocks in
1761 * the file system, but oh well.
1763 xfs_bmap_cancel(&free_list
);
1769 * Mark the inode dirty so it will be logged and
1770 * moved forward in the log as part of every commit.
1772 xfs_trans_log_inode(ntp
, ip
, XFS_ILOG_CORE
);
1775 ntp
= xfs_trans_dup(ntp
);
1776 error
= xfs_trans_commit(*tp
, 0);
1779 /* link the inode into the next transaction in the chain */
1780 xfs_trans_ijoin(ntp
, ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1781 xfs_trans_ihold(ntp
, ip
);
1784 error
= xfs_trans_reserve(ntp
, 0,
1785 XFS_ITRUNCATE_LOG_RES(mp
), 0,
1786 XFS_TRANS_PERM_LOG_RES
,
1787 XFS_ITRUNCATE_LOG_COUNT
);
1792 * Only update the size in the case of the data fork, but
1793 * always re-log the inode so that our permanent transaction
1794 * can keep on rolling it forward in the log.
1796 if (fork
== XFS_DATA_FORK
) {
1797 xfs_isize_check(mp
, ip
, new_size
);
1799 * If we are not changing the file size then do
1800 * not update the on-disk file size - we may be
1801 * called from xfs_inactive_free_eofblocks(). If we
1802 * update the on-disk file size and then the system
1803 * crashes before the contents of the file are
1804 * flushed to disk then the files may be full of
1805 * holes (ie NULL files bug).
1807 if (ip
->i_size
!= new_size
) {
1808 ip
->i_d
.di_size
= new_size
;
1809 ip
->i_size
= new_size
;
1812 xfs_trans_log_inode(ntp
, ip
, XFS_ILOG_CORE
);
1813 ASSERT((new_size
!= 0) ||
1814 (fork
== XFS_ATTR_FORK
) ||
1815 (ip
->i_delayed_blks
== 0));
1816 ASSERT((new_size
!= 0) ||
1817 (fork
== XFS_ATTR_FORK
) ||
1818 (ip
->i_d
.di_nextents
== 0));
1819 xfs_itrunc_trace(XFS_ITRUNC_FINISH2
, ip
, 0, new_size
, 0, 0);
1824 * This is called when the inode's link count goes to 0.
1825 * We place the on-disk inode on a list in the AGI. It
1826 * will be pulled from this list when the inode is freed.
1838 xfs_agnumber_t agno
;
1839 xfs_daddr_t agdaddr
;
1846 ASSERT(ip
->i_d
.di_nlink
== 0);
1847 ASSERT(ip
->i_d
.di_mode
!= 0);
1848 ASSERT(ip
->i_transp
== tp
);
1852 agno
= XFS_INO_TO_AGNO(mp
, ip
->i_ino
);
1853 agdaddr
= XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
));
1856 * Get the agi buffer first. It ensures lock ordering
1859 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, agdaddr
,
1860 XFS_FSS_TO_BB(mp
, 1), 0, &agibp
);
1865 * Validate the magic number of the agi block.
1867 agi
= XFS_BUF_TO_AGI(agibp
);
1869 be32_to_cpu(agi
->agi_magicnum
) == XFS_AGI_MAGIC
&&
1870 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi
->agi_versionnum
));
1871 if (unlikely(XFS_TEST_ERROR(!agi_ok
, mp
, XFS_ERRTAG_IUNLINK
,
1872 XFS_RANDOM_IUNLINK
))) {
1873 XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW
, mp
, agi
);
1874 xfs_trans_brelse(tp
, agibp
);
1875 return XFS_ERROR(EFSCORRUPTED
);
1878 * Get the index into the agi hash table for the
1879 * list this inode will go on.
1881 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
1883 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
1884 ASSERT(agi
->agi_unlinked
[bucket_index
]);
1885 ASSERT(be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != agino
);
1887 if (be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != NULLAGINO
) {
1889 * There is already another inode in the bucket we need
1890 * to add ourselves to. Add us at the front of the list.
1891 * Here we put the head pointer into our next pointer,
1892 * and then we fall through to point the head at us.
1894 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, 0, 0, XFS_BUF_LOCK
);
1898 ASSERT(be32_to_cpu(dip
->di_next_unlinked
) == NULLAGINO
);
1899 /* both on-disk, don't endian flip twice */
1900 dip
->di_next_unlinked
= agi
->agi_unlinked
[bucket_index
];
1901 offset
= ip
->i_boffset
+
1902 offsetof(xfs_dinode_t
, di_next_unlinked
);
1903 xfs_trans_inode_buf(tp
, ibp
);
1904 xfs_trans_log_buf(tp
, ibp
, offset
,
1905 (offset
+ sizeof(xfs_agino_t
) - 1));
1906 xfs_inobp_check(mp
, ibp
);
1910 * Point the bucket head pointer at the inode being inserted.
1913 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(agino
);
1914 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
1915 (sizeof(xfs_agino_t
) * bucket_index
);
1916 xfs_trans_log_buf(tp
, agibp
, offset
,
1917 (offset
+ sizeof(xfs_agino_t
) - 1));
1922 * Pull the on-disk inode from the AGI unlinked list.
1935 xfs_agnumber_t agno
;
1936 xfs_daddr_t agdaddr
;
1938 xfs_agino_t next_agino
;
1939 xfs_buf_t
*last_ibp
;
1940 xfs_dinode_t
*last_dip
= NULL
;
1942 int offset
, last_offset
= 0;
1947 * First pull the on-disk inode from the AGI unlinked list.
1951 agno
= XFS_INO_TO_AGNO(mp
, ip
->i_ino
);
1952 agdaddr
= XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
));
1955 * Get the agi buffer first. It ensures lock ordering
1958 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, agdaddr
,
1959 XFS_FSS_TO_BB(mp
, 1), 0, &agibp
);
1962 "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
1963 error
, mp
->m_fsname
);
1967 * Validate the magic number of the agi block.
1969 agi
= XFS_BUF_TO_AGI(agibp
);
1971 be32_to_cpu(agi
->agi_magicnum
) == XFS_AGI_MAGIC
&&
1972 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi
->agi_versionnum
));
1973 if (unlikely(XFS_TEST_ERROR(!agi_ok
, mp
, XFS_ERRTAG_IUNLINK_REMOVE
,
1974 XFS_RANDOM_IUNLINK_REMOVE
))) {
1975 XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW
,
1977 xfs_trans_brelse(tp
, agibp
);
1979 "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
1981 return XFS_ERROR(EFSCORRUPTED
);
1984 * Get the index into the agi hash table for the
1985 * list this inode will go on.
1987 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
1989 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
1990 ASSERT(be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != NULLAGINO
);
1991 ASSERT(agi
->agi_unlinked
[bucket_index
]);
1993 if (be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) == agino
) {
1995 * We're at the head of the list. Get the inode's
1996 * on-disk buffer to see if there is anyone after us
1997 * on the list. Only modify our next pointer if it
1998 * is not already NULLAGINO. This saves us the overhead
1999 * of dealing with the buffer when there is no need to
2002 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, 0, 0, XFS_BUF_LOCK
);
2005 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2006 error
, mp
->m_fsname
);
2009 next_agino
= be32_to_cpu(dip
->di_next_unlinked
);
2010 ASSERT(next_agino
!= 0);
2011 if (next_agino
!= NULLAGINO
) {
2012 dip
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
2013 offset
= ip
->i_boffset
+
2014 offsetof(xfs_dinode_t
, di_next_unlinked
);
2015 xfs_trans_inode_buf(tp
, ibp
);
2016 xfs_trans_log_buf(tp
, ibp
, offset
,
2017 (offset
+ sizeof(xfs_agino_t
) - 1));
2018 xfs_inobp_check(mp
, ibp
);
2020 xfs_trans_brelse(tp
, ibp
);
2023 * Point the bucket head pointer at the next inode.
2025 ASSERT(next_agino
!= 0);
2026 ASSERT(next_agino
!= agino
);
2027 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(next_agino
);
2028 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
2029 (sizeof(xfs_agino_t
) * bucket_index
);
2030 xfs_trans_log_buf(tp
, agibp
, offset
,
2031 (offset
+ sizeof(xfs_agino_t
) - 1));
2034 * We need to search the list for the inode being freed.
2036 next_agino
= be32_to_cpu(agi
->agi_unlinked
[bucket_index
]);
2038 while (next_agino
!= agino
) {
2040 * If the last inode wasn't the one pointing to
2041 * us, then release its buffer since we're not
2042 * going to do anything with it.
2044 if (last_ibp
!= NULL
) {
2045 xfs_trans_brelse(tp
, last_ibp
);
2047 next_ino
= XFS_AGINO_TO_INO(mp
, agno
, next_agino
);
2048 error
= xfs_inotobp(mp
, tp
, next_ino
, &last_dip
,
2049 &last_ibp
, &last_offset
);
2052 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
2053 error
, mp
->m_fsname
);
2056 next_agino
= be32_to_cpu(last_dip
->di_next_unlinked
);
2057 ASSERT(next_agino
!= NULLAGINO
);
2058 ASSERT(next_agino
!= 0);
2061 * Now last_ibp points to the buffer previous to us on
2062 * the unlinked list. Pull us from the list.
2064 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, 0, 0, XFS_BUF_LOCK
);
2067 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2068 error
, mp
->m_fsname
);
2071 next_agino
= be32_to_cpu(dip
->di_next_unlinked
);
2072 ASSERT(next_agino
!= 0);
2073 ASSERT(next_agino
!= agino
);
2074 if (next_agino
!= NULLAGINO
) {
2075 dip
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
2076 offset
= ip
->i_boffset
+
2077 offsetof(xfs_dinode_t
, di_next_unlinked
);
2078 xfs_trans_inode_buf(tp
, ibp
);
2079 xfs_trans_log_buf(tp
, ibp
, offset
,
2080 (offset
+ sizeof(xfs_agino_t
) - 1));
2081 xfs_inobp_check(mp
, ibp
);
2083 xfs_trans_brelse(tp
, ibp
);
2086 * Point the previous inode on the list to the next inode.
2088 last_dip
->di_next_unlinked
= cpu_to_be32(next_agino
);
2089 ASSERT(next_agino
!= 0);
2090 offset
= last_offset
+ offsetof(xfs_dinode_t
, di_next_unlinked
);
2091 xfs_trans_inode_buf(tp
, last_ibp
);
2092 xfs_trans_log_buf(tp
, last_ibp
, offset
,
2093 (offset
+ sizeof(xfs_agino_t
) - 1));
2094 xfs_inobp_check(mp
, last_ibp
);
2101 xfs_inode_t
*free_ip
,
2105 xfs_mount_t
*mp
= free_ip
->i_mount
;
2106 int blks_per_cluster
;
2109 int i
, j
, found
, pre_flushed
;
2112 xfs_inode_t
*ip
, **ip_found
;
2113 xfs_inode_log_item_t
*iip
;
2114 xfs_log_item_t
*lip
;
2115 xfs_perag_t
*pag
= xfs_get_perag(mp
, inum
);
2117 if (mp
->m_sb
.sb_blocksize
>= XFS_INODE_CLUSTER_SIZE(mp
)) {
2118 blks_per_cluster
= 1;
2119 ninodes
= mp
->m_sb
.sb_inopblock
;
2120 nbufs
= XFS_IALLOC_BLOCKS(mp
);
2122 blks_per_cluster
= XFS_INODE_CLUSTER_SIZE(mp
) /
2123 mp
->m_sb
.sb_blocksize
;
2124 ninodes
= blks_per_cluster
* mp
->m_sb
.sb_inopblock
;
2125 nbufs
= XFS_IALLOC_BLOCKS(mp
) / blks_per_cluster
;
2128 ip_found
= kmem_alloc(ninodes
* sizeof(xfs_inode_t
*), KM_NOFS
);
2130 for (j
= 0; j
< nbufs
; j
++, inum
+= ninodes
) {
2131 blkno
= XFS_AGB_TO_DADDR(mp
, XFS_INO_TO_AGNO(mp
, inum
),
2132 XFS_INO_TO_AGBNO(mp
, inum
));
2136 * Look for each inode in memory and attempt to lock it,
2137 * we can be racing with flush and tail pushing here.
2138 * any inode we get the locks on, add to an array of
2139 * inode items to process later.
2141 * The get the buffer lock, we could beat a flush
2142 * or tail pushing thread to the lock here, in which
2143 * case they will go looking for the inode buffer
2144 * and fail, we need some other form of interlock
2148 for (i
= 0; i
< ninodes
; i
++) {
2149 read_lock(&pag
->pag_ici_lock
);
2150 ip
= radix_tree_lookup(&pag
->pag_ici_root
,
2151 XFS_INO_TO_AGINO(mp
, (inum
+ i
)));
2153 /* Inode not in memory or we found it already,
2156 if (!ip
|| xfs_iflags_test(ip
, XFS_ISTALE
)) {
2157 read_unlock(&pag
->pag_ici_lock
);
2161 if (xfs_inode_clean(ip
)) {
2162 read_unlock(&pag
->pag_ici_lock
);
2166 /* If we can get the locks then add it to the
2167 * list, otherwise by the time we get the bp lock
2168 * below it will already be attached to the
2172 /* This inode will already be locked - by us, lets
2176 if (ip
== free_ip
) {
2177 if (xfs_iflock_nowait(ip
)) {
2178 xfs_iflags_set(ip
, XFS_ISTALE
);
2179 if (xfs_inode_clean(ip
)) {
2182 ip_found
[found
++] = ip
;
2185 read_unlock(&pag
->pag_ici_lock
);
2189 if (xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
)) {
2190 if (xfs_iflock_nowait(ip
)) {
2191 xfs_iflags_set(ip
, XFS_ISTALE
);
2193 if (xfs_inode_clean(ip
)) {
2195 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2197 ip_found
[found
++] = ip
;
2200 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2203 read_unlock(&pag
->pag_ici_lock
);
2206 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, blkno
,
2207 mp
->m_bsize
* blks_per_cluster
,
2211 lip
= XFS_BUF_FSPRIVATE(bp
, xfs_log_item_t
*);
2213 if (lip
->li_type
== XFS_LI_INODE
) {
2214 iip
= (xfs_inode_log_item_t
*)lip
;
2215 ASSERT(iip
->ili_logged
== 1);
2216 lip
->li_cb
= (void(*)(xfs_buf_t
*,xfs_log_item_t
*)) xfs_istale_done
;
2217 spin_lock(&mp
->m_ail_lock
);
2218 iip
->ili_flush_lsn
= iip
->ili_item
.li_lsn
;
2219 spin_unlock(&mp
->m_ail_lock
);
2220 xfs_iflags_set(iip
->ili_inode
, XFS_ISTALE
);
2223 lip
= lip
->li_bio_list
;
2226 for (i
= 0; i
< found
; i
++) {
2231 ip
->i_update_core
= 0;
2233 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2237 iip
->ili_last_fields
= iip
->ili_format
.ilf_fields
;
2238 iip
->ili_format
.ilf_fields
= 0;
2239 iip
->ili_logged
= 1;
2240 spin_lock(&mp
->m_ail_lock
);
2241 iip
->ili_flush_lsn
= iip
->ili_item
.li_lsn
;
2242 spin_unlock(&mp
->m_ail_lock
);
2244 xfs_buf_attach_iodone(bp
,
2245 (void(*)(xfs_buf_t
*,xfs_log_item_t
*))
2246 xfs_istale_done
, (xfs_log_item_t
*)iip
);
2247 if (ip
!= free_ip
) {
2248 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2252 if (found
|| pre_flushed
)
2253 xfs_trans_stale_inode_buf(tp
, bp
);
2254 xfs_trans_binval(tp
, bp
);
2257 kmem_free(ip_found
);
2258 xfs_put_perag(mp
, pag
);
2262 * This is called to return an inode to the inode free list.
2263 * The inode should already be truncated to 0 length and have
2264 * no pages associated with it. This routine also assumes that
2265 * the inode is already a part of the transaction.
2267 * The on-disk copy of the inode will have been added to the list
2268 * of unlinked inodes in the AGI. We need to remove the inode from
2269 * that list atomically with respect to freeing it here.
2275 xfs_bmap_free_t
*flist
)
2279 xfs_ino_t first_ino
;
2283 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
2284 ASSERT(ip
->i_transp
== tp
);
2285 ASSERT(ip
->i_d
.di_nlink
== 0);
2286 ASSERT(ip
->i_d
.di_nextents
== 0);
2287 ASSERT(ip
->i_d
.di_anextents
== 0);
2288 ASSERT((ip
->i_d
.di_size
== 0 && ip
->i_size
== 0) ||
2289 ((ip
->i_d
.di_mode
& S_IFMT
) != S_IFREG
));
2290 ASSERT(ip
->i_d
.di_nblocks
== 0);
2293 * Pull the on-disk inode from the AGI unlinked list.
2295 error
= xfs_iunlink_remove(tp
, ip
);
2300 error
= xfs_difree(tp
, ip
->i_ino
, flist
, &delete, &first_ino
);
2304 ip
->i_d
.di_mode
= 0; /* mark incore inode as free */
2305 ip
->i_d
.di_flags
= 0;
2306 ip
->i_d
.di_dmevmask
= 0;
2307 ip
->i_d
.di_forkoff
= 0; /* mark the attr fork not in use */
2308 ip
->i_df
.if_ext_max
=
2309 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
2310 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
2311 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
2313 * Bump the generation count so no one will be confused
2314 * by reincarnations of this inode.
2318 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2320 error
= xfs_itobp(ip
->i_mount
, tp
, ip
, &dip
, &ibp
, 0, 0, XFS_BUF_LOCK
);
2325 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2326 * from picking up this inode when it is reclaimed (its incore state
2327 * initialzed but not flushed to disk yet). The in-core di_mode is
2328 * already cleared and a corresponding transaction logged.
2329 * The hack here just synchronizes the in-core to on-disk
2330 * di_mode value in advance before the actual inode sync to disk.
2331 * This is OK because the inode is already unlinked and would never
2332 * change its di_mode again for this inode generation.
2333 * This is a temporary hack that would require a proper fix
2336 dip
->di_core
.di_mode
= 0;
2339 xfs_ifree_cluster(ip
, tp
, first_ino
);
2346 * Reallocate the space for if_broot based on the number of records
2347 * being added or deleted as indicated in rec_diff. Move the records
2348 * and pointers in if_broot to fit the new size. When shrinking this
2349 * will eliminate holes between the records and pointers created by
2350 * the caller. When growing this will create holes to be filled in
2353 * The caller must not request to add more records than would fit in
2354 * the on-disk inode root. If the if_broot is currently NULL, then
2355 * if we adding records one will be allocated. The caller must also
2356 * not request that the number of records go below zero, although
2357 * it can go to zero.
2359 * ip -- the inode whose if_broot area is changing
2360 * ext_diff -- the change in the number of records, positive or negative,
2361 * requested for the if_broot array.
2369 struct xfs_mount
*mp
= ip
->i_mount
;
2372 struct xfs_btree_block
*new_broot
;
2379 * Handle the degenerate case quietly.
2381 if (rec_diff
== 0) {
2385 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2388 * If there wasn't any memory allocated before, just
2389 * allocate it now and get out.
2391 if (ifp
->if_broot_bytes
== 0) {
2392 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff
);
2393 ifp
->if_broot
= kmem_alloc(new_size
, KM_SLEEP
);
2394 ifp
->if_broot_bytes
= (int)new_size
;
2399 * If there is already an existing if_broot, then we need
2400 * to realloc() it and shift the pointers to their new
2401 * location. The records don't change location because
2402 * they are kept butted up against the btree block header.
2404 cur_max
= xfs_bmbt_maxrecs(mp
, ifp
->if_broot_bytes
, 0);
2405 new_max
= cur_max
+ rec_diff
;
2406 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max
);
2407 ifp
->if_broot
= kmem_realloc(ifp
->if_broot
, new_size
,
2408 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max
), /* old size */
2410 op
= (char *)XFS_BMAP_BROOT_PTR_ADDR(mp
, ifp
->if_broot
, 1,
2411 ifp
->if_broot_bytes
);
2412 np
= (char *)XFS_BMAP_BROOT_PTR_ADDR(mp
, ifp
->if_broot
, 1,
2414 ifp
->if_broot_bytes
= (int)new_size
;
2415 ASSERT(ifp
->if_broot_bytes
<=
2416 XFS_IFORK_SIZE(ip
, whichfork
) + XFS_BROOT_SIZE_ADJ
);
2417 memmove(np
, op
, cur_max
* (uint
)sizeof(xfs_dfsbno_t
));
2422 * rec_diff is less than 0. In this case, we are shrinking the
2423 * if_broot buffer. It must already exist. If we go to zero
2424 * records, just get rid of the root and clear the status bit.
2426 ASSERT((ifp
->if_broot
!= NULL
) && (ifp
->if_broot_bytes
> 0));
2427 cur_max
= xfs_bmbt_maxrecs(mp
, ifp
->if_broot_bytes
, 0);
2428 new_max
= cur_max
+ rec_diff
;
2429 ASSERT(new_max
>= 0);
2431 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max
);
2435 new_broot
= kmem_alloc(new_size
, KM_SLEEP
);
2437 * First copy over the btree block header.
2439 memcpy(new_broot
, ifp
->if_broot
, XFS_BTREE_LBLOCK_LEN
);
2442 ifp
->if_flags
&= ~XFS_IFBROOT
;
2446 * Only copy the records and pointers if there are any.
2450 * First copy the records.
2452 op
= (char *)XFS_BMBT_REC_ADDR(mp
, ifp
->if_broot
, 1);
2453 np
= (char *)XFS_BMBT_REC_ADDR(mp
, 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(mp
, ifp
->if_broot
, 1,
2460 ifp
->if_broot_bytes
);
2461 np
= (char *)XFS_BMAP_BROOT_PTR_ADDR(mp
, new_broot
, 1,
2463 memcpy(np
, op
, new_max
* (uint
)sizeof(xfs_dfsbno_t
));
2465 kmem_free(ifp
->if_broot
);
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_data
2476 * is increased or decreased. The change in size is indicated by
2477 * the number of bytes that need to be added or deleted in the
2478 * byte_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_data area is changing
2486 * byte_diff -- the change in the number of bytes, positive or negative,
2487 * requested for the if_data array.
2499 if (byte_diff
== 0) {
2503 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2504 new_size
= (int)ifp
->if_bytes
+ byte_diff
;
2505 ASSERT(new_size
>= 0);
2507 if (new_size
== 0) {
2508 if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
2509 kmem_free(ifp
->if_u1
.if_data
);
2511 ifp
->if_u1
.if_data
= NULL
;
2513 } else if (new_size
<= sizeof(ifp
->if_u2
.if_inline_data
)) {
2515 * If the valid extents/data can fit in if_inline_ext/data,
2516 * copy them from the malloc'd vector and free it.
2518 if (ifp
->if_u1
.if_data
== NULL
) {
2519 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
2520 } else if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
2521 ASSERT(ifp
->if_real_bytes
!= 0);
2522 memcpy(ifp
->if_u2
.if_inline_data
, ifp
->if_u1
.if_data
,
2524 kmem_free(ifp
->if_u1
.if_data
);
2525 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
2530 * Stuck with malloc/realloc.
2531 * For inline data, the underlying buffer must be
2532 * a multiple of 4 bytes in size so that it can be
2533 * logged and stay on word boundaries. We enforce
2536 real_size
= roundup(new_size
, 4);
2537 if (ifp
->if_u1
.if_data
== NULL
) {
2538 ASSERT(ifp
->if_real_bytes
== 0);
2539 ifp
->if_u1
.if_data
= kmem_alloc(real_size
, KM_SLEEP
);
2540 } else if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
2542 * Only do the realloc if the underlying size
2543 * is really changing.
2545 if (ifp
->if_real_bytes
!= real_size
) {
2546 ifp
->if_u1
.if_data
=
2547 kmem_realloc(ifp
->if_u1
.if_data
,
2553 ASSERT(ifp
->if_real_bytes
== 0);
2554 ifp
->if_u1
.if_data
= kmem_alloc(real_size
, KM_SLEEP
);
2555 memcpy(ifp
->if_u1
.if_data
, ifp
->if_u2
.if_inline_data
,
2559 ifp
->if_real_bytes
= real_size
;
2560 ifp
->if_bytes
= new_size
;
2561 ASSERT(ifp
->if_bytes
<= XFS_IFORK_SIZE(ip
, whichfork
));
2568 * Map inode to disk block and offset.
2570 * mp -- the mount point structure for the current file system
2571 * tp -- the current transaction
2572 * ino -- the inode number of the inode to be located
2573 * imap -- this structure is filled in with the information necessary
2574 * to retrieve the given inode from disk
2575 * flags -- flags to pass to xfs_dilocate indicating whether or not
2576 * lookups in the inode btree were OK or not
2586 xfs_fsblock_t fsbno
;
2591 fsbno
= imap
->im_blkno
?
2592 XFS_DADDR_TO_FSB(mp
, imap
->im_blkno
) : NULLFSBLOCK
;
2593 error
= xfs_dilocate(mp
, tp
, ino
, &fsbno
, &len
, &off
, flags
);
2597 imap
->im_blkno
= XFS_FSB_TO_DADDR(mp
, fsbno
);
2598 imap
->im_len
= XFS_FSB_TO_BB(mp
, len
);
2599 imap
->im_agblkno
= XFS_FSB_TO_AGBNO(mp
, fsbno
);
2600 imap
->im_ioffset
= (ushort
)off
;
2601 imap
->im_boffset
= (ushort
)(off
<< mp
->m_sb
.sb_inodelog
);
2604 * If the inode number maps to a block outside the bounds
2605 * of the file system then return NULL rather than calling
2606 * read_buf and panicing when we get an error from the
2609 if ((imap
->im_blkno
+ imap
->im_len
) >
2610 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
)) {
2611 xfs_fs_cmn_err(CE_ALERT
, mp
, "xfs_imap: "
2612 "(imap->im_blkno (0x%llx) + imap->im_len (0x%llx)) > "
2613 " XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks) (0x%llx)",
2614 (unsigned long long) imap
->im_blkno
,
2615 (unsigned long long) imap
->im_len
,
2616 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
));
2629 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2630 if (ifp
->if_broot
!= NULL
) {
2631 kmem_free(ifp
->if_broot
);
2632 ifp
->if_broot
= NULL
;
2636 * If the format is local, then we can't have an extents
2637 * array so just look for an inline data array. If we're
2638 * not local then we may or may not have an extents list,
2639 * so check and free it up if we do.
2641 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
) {
2642 if ((ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) &&
2643 (ifp
->if_u1
.if_data
!= NULL
)) {
2644 ASSERT(ifp
->if_real_bytes
!= 0);
2645 kmem_free(ifp
->if_u1
.if_data
);
2646 ifp
->if_u1
.if_data
= NULL
;
2647 ifp
->if_real_bytes
= 0;
2649 } else if ((ifp
->if_flags
& XFS_IFEXTENTS
) &&
2650 ((ifp
->if_flags
& XFS_IFEXTIREC
) ||
2651 ((ifp
->if_u1
.if_extents
!= NULL
) &&
2652 (ifp
->if_u1
.if_extents
!= ifp
->if_u2
.if_inline_ext
)))) {
2653 ASSERT(ifp
->if_real_bytes
!= 0);
2654 xfs_iext_destroy(ifp
);
2656 ASSERT(ifp
->if_u1
.if_extents
== NULL
||
2657 ifp
->if_u1
.if_extents
== ifp
->if_u2
.if_inline_ext
);
2658 ASSERT(ifp
->if_real_bytes
== 0);
2659 if (whichfork
== XFS_ATTR_FORK
) {
2660 kmem_zone_free(xfs_ifork_zone
, ip
->i_afp
);
2666 * This is called free all the memory associated with an inode.
2667 * It must free the inode itself and any buffers allocated for
2668 * if_extents/if_data and if_broot. It must also free the lock
2669 * associated with the inode.
2671 * Note: because we don't initialise everything on reallocation out
2672 * of the zone, we must ensure we nullify everything correctly before
2673 * freeing the structure.
2679 switch (ip
->i_d
.di_mode
& S_IFMT
) {
2683 xfs_idestroy_fork(ip
, XFS_DATA_FORK
);
2687 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
2689 #ifdef XFS_INODE_TRACE
2690 ktrace_free(ip
->i_trace
);
2692 #ifdef XFS_BMAP_TRACE
2693 ktrace_free(ip
->i_xtrace
);
2695 #ifdef XFS_BTREE_TRACE
2696 ktrace_free(ip
->i_btrace
);
2699 ktrace_free(ip
->i_rwtrace
);
2701 #ifdef XFS_ILOCK_TRACE
2702 ktrace_free(ip
->i_lock_trace
);
2704 #ifdef XFS_DIR2_TRACE
2705 ktrace_free(ip
->i_dir_trace
);
2709 * Only if we are shutting down the fs will we see an
2710 * inode still in the AIL. If it is there, we should remove
2711 * it to prevent a use-after-free from occurring.
2713 xfs_mount_t
*mp
= ip
->i_mount
;
2714 xfs_log_item_t
*lip
= &ip
->i_itemp
->ili_item
;
2716 ASSERT(((lip
->li_flags
& XFS_LI_IN_AIL
) == 0) ||
2717 XFS_FORCED_SHUTDOWN(ip
->i_mount
));
2718 if (lip
->li_flags
& XFS_LI_IN_AIL
) {
2719 spin_lock(&mp
->m_ail_lock
);
2720 if (lip
->li_flags
& XFS_LI_IN_AIL
)
2721 xfs_trans_delete_ail(mp
, lip
);
2723 spin_unlock(&mp
->m_ail_lock
);
2725 xfs_inode_item_destroy(ip
);
2728 /* asserts to verify all state is correct here */
2729 ASSERT(atomic_read(&ip
->i_iocount
) == 0);
2730 ASSERT(atomic_read(&ip
->i_pincount
) == 0);
2731 ASSERT(!spin_is_locked(&ip
->i_flags_lock
));
2732 ASSERT(list_empty(&ip
->i_reclaim
));
2733 kmem_zone_free(xfs_inode_zone
, ip
);
2738 * Increment the pin count of the given buffer.
2739 * This value is protected by ipinlock spinlock in the mount structure.
2745 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
2747 atomic_inc(&ip
->i_pincount
);
2751 * Decrement the pin count of the given inode, and wake up
2752 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2753 * inode must have been previously pinned with a call to xfs_ipin().
2759 ASSERT(atomic_read(&ip
->i_pincount
) > 0);
2761 if (atomic_dec_and_test(&ip
->i_pincount
))
2762 wake_up(&ip
->i_ipin_wait
);
2766 * This is called to unpin an inode. It can be directed to wait or to return
2767 * immediately without waiting for the inode to be unpinned. The caller must
2768 * have the inode locked in at least shared mode so that the buffer cannot be
2769 * subsequently pinned once someone is waiting for it to be unpinned.
2776 xfs_inode_log_item_t
*iip
= ip
->i_itemp
;
2778 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2779 if (atomic_read(&ip
->i_pincount
) == 0)
2782 /* Give the log a push to start the unpinning I/O */
2783 xfs_log_force(ip
->i_mount
, (iip
&& iip
->ili_last_lsn
) ?
2784 iip
->ili_last_lsn
: 0, XFS_LOG_FORCE
);
2786 wait_event(ip
->i_ipin_wait
, (atomic_read(&ip
->i_pincount
) == 0));
2793 __xfs_iunpin_wait(ip
, 1);
2800 __xfs_iunpin_wait(ip
, 0);
2805 * xfs_iextents_copy()
2807 * This is called to copy the REAL extents (as opposed to the delayed
2808 * allocation extents) from the inode into the given buffer. It
2809 * returns the number of bytes copied into the buffer.
2811 * If there are no delayed allocation extents, then we can just
2812 * memcpy() the extents into the buffer. Otherwise, we need to
2813 * examine each extent in turn and skip those which are delayed.
2825 xfs_fsblock_t start_block
;
2827 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2828 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2829 ASSERT(ifp
->if_bytes
> 0);
2831 nrecs
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
2832 XFS_BMAP_TRACE_EXLIST(ip
, nrecs
, whichfork
);
2836 * There are some delayed allocation extents in the
2837 * inode, so copy the extents one at a time and skip
2838 * the delayed ones. There must be at least one
2839 * non-delayed extent.
2842 for (i
= 0; i
< nrecs
; i
++) {
2843 xfs_bmbt_rec_host_t
*ep
= xfs_iext_get_ext(ifp
, i
);
2844 start_block
= xfs_bmbt_get_startblock(ep
);
2845 if (ISNULLSTARTBLOCK(start_block
)) {
2847 * It's a delayed allocation extent, so skip it.
2852 /* Translate to on disk format */
2853 put_unaligned(cpu_to_be64(ep
->l0
), &dp
->l0
);
2854 put_unaligned(cpu_to_be64(ep
->l1
), &dp
->l1
);
2858 ASSERT(copied
!= 0);
2859 xfs_validate_extents(ifp
, copied
, XFS_EXTFMT_INODE(ip
));
2861 return (copied
* (uint
)sizeof(xfs_bmbt_rec_t
));
2865 * Each of the following cases stores data into the same region
2866 * of the on-disk inode, so only one of them can be valid at
2867 * any given time. While it is possible to have conflicting formats
2868 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2869 * in EXTENTS format, this can only happen when the fork has
2870 * changed formats after being modified but before being flushed.
2871 * In these cases, the format always takes precedence, because the
2872 * format indicates the current state of the fork.
2879 xfs_inode_log_item_t
*iip
,
2886 #ifdef XFS_TRANS_DEBUG
2889 static const short brootflag
[2] =
2890 { XFS_ILOG_DBROOT
, XFS_ILOG_ABROOT
};
2891 static const short dataflag
[2] =
2892 { XFS_ILOG_DDATA
, XFS_ILOG_ADATA
};
2893 static const short extflag
[2] =
2894 { XFS_ILOG_DEXT
, XFS_ILOG_AEXT
};
2898 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2900 * This can happen if we gave up in iformat in an error path,
2901 * for the attribute fork.
2904 ASSERT(whichfork
== XFS_ATTR_FORK
);
2907 cp
= XFS_DFORK_PTR(dip
, whichfork
);
2909 switch (XFS_IFORK_FORMAT(ip
, whichfork
)) {
2910 case XFS_DINODE_FMT_LOCAL
:
2911 if ((iip
->ili_format
.ilf_fields
& dataflag
[whichfork
]) &&
2912 (ifp
->if_bytes
> 0)) {
2913 ASSERT(ifp
->if_u1
.if_data
!= NULL
);
2914 ASSERT(ifp
->if_bytes
<= XFS_IFORK_SIZE(ip
, whichfork
));
2915 memcpy(cp
, ifp
->if_u1
.if_data
, ifp
->if_bytes
);
2919 case XFS_DINODE_FMT_EXTENTS
:
2920 ASSERT((ifp
->if_flags
& XFS_IFEXTENTS
) ||
2921 !(iip
->ili_format
.ilf_fields
& extflag
[whichfork
]));
2922 ASSERT((xfs_iext_get_ext(ifp
, 0) != NULL
) ||
2923 (ifp
->if_bytes
== 0));
2924 ASSERT((xfs_iext_get_ext(ifp
, 0) == NULL
) ||
2925 (ifp
->if_bytes
> 0));
2926 if ((iip
->ili_format
.ilf_fields
& extflag
[whichfork
]) &&
2927 (ifp
->if_bytes
> 0)) {
2928 ASSERT(XFS_IFORK_NEXTENTS(ip
, whichfork
) > 0);
2929 (void)xfs_iextents_copy(ip
, (xfs_bmbt_rec_t
*)cp
,
2934 case XFS_DINODE_FMT_BTREE
:
2935 if ((iip
->ili_format
.ilf_fields
& brootflag
[whichfork
]) &&
2936 (ifp
->if_broot_bytes
> 0)) {
2937 ASSERT(ifp
->if_broot
!= NULL
);
2938 ASSERT(ifp
->if_broot_bytes
<=
2939 (XFS_IFORK_SIZE(ip
, whichfork
) +
2940 XFS_BROOT_SIZE_ADJ
));
2941 xfs_bmbt_to_bmdr(mp
, ifp
->if_broot
, ifp
->if_broot_bytes
,
2942 (xfs_bmdr_block_t
*)cp
,
2943 XFS_DFORK_SIZE(dip
, mp
, whichfork
));
2947 case XFS_DINODE_FMT_DEV
:
2948 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_DEV
) {
2949 ASSERT(whichfork
== XFS_DATA_FORK
);
2950 dip
->di_u
.di_dev
= cpu_to_be32(ip
->i_df
.if_u2
.if_rdev
);
2954 case XFS_DINODE_FMT_UUID
:
2955 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_UUID
) {
2956 ASSERT(whichfork
== XFS_DATA_FORK
);
2957 memcpy(&dip
->di_u
.di_muuid
, &ip
->i_df
.if_u2
.if_uuid
,
2973 xfs_mount_t
*mp
= ip
->i_mount
;
2974 xfs_perag_t
*pag
= xfs_get_perag(mp
, ip
->i_ino
);
2975 unsigned long first_index
, mask
;
2976 unsigned long inodes_per_cluster
;
2978 xfs_inode_t
**ilist
;
2985 ASSERT(pag
->pagi_inodeok
);
2986 ASSERT(pag
->pag_ici_init
);
2988 inodes_per_cluster
= XFS_INODE_CLUSTER_SIZE(mp
) >> mp
->m_sb
.sb_inodelog
;
2989 ilist_size
= inodes_per_cluster
* sizeof(xfs_inode_t
*);
2990 ilist
= kmem_alloc(ilist_size
, KM_MAYFAIL
|KM_NOFS
);
2994 mask
= ~(((XFS_INODE_CLUSTER_SIZE(mp
) >> mp
->m_sb
.sb_inodelog
)) - 1);
2995 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
) & mask
;
2996 read_lock(&pag
->pag_ici_lock
);
2997 /* really need a gang lookup range call here */
2998 nr_found
= radix_tree_gang_lookup(&pag
->pag_ici_root
, (void**)ilist
,
2999 first_index
, inodes_per_cluster
);
3003 for (i
= 0; i
< nr_found
; i
++) {
3007 /* if the inode lies outside this cluster, we're done. */
3008 if ((XFS_INO_TO_AGINO(mp
, iq
->i_ino
) & mask
) != first_index
)
3011 * Do an un-protected check to see if the inode is dirty and
3012 * is a candidate for flushing. These checks will be repeated
3013 * later after the appropriate locks are acquired.
3015 if (xfs_inode_clean(iq
) && xfs_ipincount(iq
) == 0)
3019 * Try to get locks. If any are unavailable or it is pinned,
3020 * then this inode cannot be flushed and is skipped.
3023 if (!xfs_ilock_nowait(iq
, XFS_ILOCK_SHARED
))
3025 if (!xfs_iflock_nowait(iq
)) {
3026 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
3029 if (xfs_ipincount(iq
)) {
3031 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
3036 * arriving here means that this inode can be flushed. First
3037 * re-check that it's dirty before flushing.
3039 if (!xfs_inode_clean(iq
)) {
3041 error
= xfs_iflush_int(iq
, bp
);
3043 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
3044 goto cluster_corrupt_out
;
3050 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
3054 XFS_STATS_INC(xs_icluster_flushcnt
);
3055 XFS_STATS_ADD(xs_icluster_flushinode
, clcount
);
3059 read_unlock(&pag
->pag_ici_lock
);
3064 cluster_corrupt_out
:
3066 * Corruption detected in the clustering loop. Invalidate the
3067 * inode buffer and shut down the filesystem.
3069 read_unlock(&pag
->pag_ici_lock
);
3071 * Clean up the buffer. If it was B_DELWRI, just release it --
3072 * brelse can handle it with no problems. If not, shut down the
3073 * filesystem before releasing the buffer.
3075 bufwasdelwri
= XFS_BUF_ISDELAYWRITE(bp
);
3079 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
3081 if (!bufwasdelwri
) {
3083 * Just like incore_relse: if we have b_iodone functions,
3084 * mark the buffer as an error and call them. Otherwise
3085 * mark it as stale and brelse.
3087 if (XFS_BUF_IODONE_FUNC(bp
)) {
3088 XFS_BUF_CLR_BDSTRAT_FUNC(bp
);
3092 XFS_BUF_ERROR(bp
,EIO
);
3101 * Unlocks the flush lock
3103 xfs_iflush_abort(iq
);
3105 return XFS_ERROR(EFSCORRUPTED
);
3109 * xfs_iflush() will write a modified inode's changes out to the
3110 * inode's on disk home. The caller must have the inode lock held
3111 * in at least shared mode and the inode flush completion must be
3112 * active as well. The inode lock will still be held upon return from
3113 * the call and the caller is free to unlock it.
3114 * The inode flush will be completed when the inode reaches the disk.
3115 * The flags indicate how the inode's buffer should be written out.
3122 xfs_inode_log_item_t
*iip
;
3127 int noblock
= (flags
== XFS_IFLUSH_ASYNC_NOBLOCK
);
3128 enum { INT_DELWRI
= (1 << 0), INT_ASYNC
= (1 << 1) };
3130 XFS_STATS_INC(xs_iflush_count
);
3132 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
3133 ASSERT(!completion_done(&ip
->i_flush
));
3134 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
3135 ip
->i_d
.di_nextents
> ip
->i_df
.if_ext_max
);
3141 * If the inode isn't dirty, then just release the inode
3142 * flush lock and do nothing.
3144 if (xfs_inode_clean(ip
)) {
3150 * We can't flush the inode until it is unpinned, so wait for it if we
3151 * are allowed to block. We know noone new can pin it, because we are
3152 * holding the inode lock shared and you need to hold it exclusively to
3155 * If we are not allowed to block, force the log out asynchronously so
3156 * that when we come back the inode will be unpinned. If other inodes
3157 * in the same cluster are dirty, they will probably write the inode
3158 * out for us if they occur after the log force completes.
3160 if (noblock
&& xfs_ipincount(ip
)) {
3161 xfs_iunpin_nowait(ip
);
3165 xfs_iunpin_wait(ip
);
3168 * This may have been unpinned because the filesystem is shutting
3169 * down forcibly. If that's the case we must not write this inode
3170 * to disk, because the log record didn't make it to disk!
3172 if (XFS_FORCED_SHUTDOWN(mp
)) {
3173 ip
->i_update_core
= 0;
3175 iip
->ili_format
.ilf_fields
= 0;
3177 return XFS_ERROR(EIO
);
3181 * Decide how buffer will be flushed out. This is done before
3182 * the call to xfs_iflush_int because this field is zeroed by it.
3184 if (iip
!= NULL
&& iip
->ili_format
.ilf_fields
!= 0) {
3186 * Flush out the inode buffer according to the directions
3187 * of the caller. In the cases where the caller has given
3188 * us a choice choose the non-delwri case. This is because
3189 * the inode is in the AIL and we need to get it out soon.
3192 case XFS_IFLUSH_SYNC
:
3193 case XFS_IFLUSH_DELWRI_ELSE_SYNC
:
3196 case XFS_IFLUSH_ASYNC_NOBLOCK
:
3197 case XFS_IFLUSH_ASYNC
:
3198 case XFS_IFLUSH_DELWRI_ELSE_ASYNC
:
3201 case XFS_IFLUSH_DELWRI
:
3211 case XFS_IFLUSH_DELWRI_ELSE_SYNC
:
3212 case XFS_IFLUSH_DELWRI_ELSE_ASYNC
:
3213 case XFS_IFLUSH_DELWRI
:
3216 case XFS_IFLUSH_ASYNC_NOBLOCK
:
3217 case XFS_IFLUSH_ASYNC
:
3220 case XFS_IFLUSH_SYNC
:
3231 * Get the buffer containing the on-disk inode.
3233 error
= xfs_itobp(mp
, NULL
, ip
, &dip
, &bp
, 0, 0,
3234 noblock
? XFS_BUF_TRYLOCK
: XFS_BUF_LOCK
);
3241 * First flush out the inode that xfs_iflush was called with.
3243 error
= xfs_iflush_int(ip
, bp
);
3248 * If the buffer is pinned then push on the log now so we won't
3249 * get stuck waiting in the write for too long.
3251 if (XFS_BUF_ISPINNED(bp
))
3252 xfs_log_force(mp
, (xfs_lsn_t
)0, XFS_LOG_FORCE
);
3256 * see if other inodes can be gathered into this write
3258 error
= xfs_iflush_cluster(ip
, bp
);
3260 goto cluster_corrupt_out
;
3262 if (flags
& INT_DELWRI
) {
3263 xfs_bdwrite(mp
, bp
);
3264 } else if (flags
& INT_ASYNC
) {
3265 error
= xfs_bawrite(mp
, bp
);
3267 error
= xfs_bwrite(mp
, bp
);
3273 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
3274 cluster_corrupt_out
:
3276 * Unlocks the flush lock
3278 xfs_iflush_abort(ip
);
3279 return XFS_ERROR(EFSCORRUPTED
);
3288 xfs_inode_log_item_t
*iip
;
3291 #ifdef XFS_TRANS_DEBUG
3295 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
3296 ASSERT(!completion_done(&ip
->i_flush
));
3297 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
3298 ip
->i_d
.di_nextents
> ip
->i_df
.if_ext_max
);
3305 * If the inode isn't dirty, then just release the inode
3306 * flush lock and do nothing.
3308 if (xfs_inode_clean(ip
)) {
3313 /* set *dip = inode's place in the buffer */
3314 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, ip
->i_boffset
);
3317 * Clear i_update_core before copying out the data.
3318 * This is for coordination with our timestamp updates
3319 * that don't hold the inode lock. They will always
3320 * update the timestamps BEFORE setting i_update_core,
3321 * so if we clear i_update_core after they set it we
3322 * are guaranteed to see their updates to the timestamps.
3323 * I believe that this depends on strongly ordered memory
3324 * semantics, but we have that. We use the SYNCHRONIZE
3325 * macro to make sure that the compiler does not reorder
3326 * the i_update_core access below the data copy below.
3328 ip
->i_update_core
= 0;
3332 * Make sure to get the latest atime from the Linux inode.
3334 xfs_synchronize_atime(ip
);
3336 if (XFS_TEST_ERROR(be16_to_cpu(dip
->di_core
.di_magic
) != XFS_DINODE_MAGIC
,
3337 mp
, XFS_ERRTAG_IFLUSH_1
, XFS_RANDOM_IFLUSH_1
)) {
3338 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3339 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3340 ip
->i_ino
, be16_to_cpu(dip
->di_core
.di_magic
), dip
);
3343 if (XFS_TEST_ERROR(ip
->i_d
.di_magic
!= XFS_DINODE_MAGIC
,
3344 mp
, XFS_ERRTAG_IFLUSH_2
, XFS_RANDOM_IFLUSH_2
)) {
3345 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3346 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3347 ip
->i_ino
, ip
, ip
->i_d
.di_magic
);
3350 if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
) {
3352 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3353 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
),
3354 mp
, XFS_ERRTAG_IFLUSH_3
, XFS_RANDOM_IFLUSH_3
)) {
3355 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3356 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3360 } else if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFDIR
) {
3362 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3363 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
) &&
3364 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_LOCAL
),
3365 mp
, XFS_ERRTAG_IFLUSH_4
, XFS_RANDOM_IFLUSH_4
)) {
3366 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3367 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3372 if (XFS_TEST_ERROR(ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
>
3373 ip
->i_d
.di_nblocks
, mp
, XFS_ERRTAG_IFLUSH_5
,
3374 XFS_RANDOM_IFLUSH_5
)) {
3375 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3376 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3378 ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
,
3383 if (XFS_TEST_ERROR(ip
->i_d
.di_forkoff
> mp
->m_sb
.sb_inodesize
,
3384 mp
, XFS_ERRTAG_IFLUSH_6
, XFS_RANDOM_IFLUSH_6
)) {
3385 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3386 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3387 ip
->i_ino
, ip
->i_d
.di_forkoff
, ip
);
3391 * bump the flush iteration count, used to detect flushes which
3392 * postdate a log record during recovery.
3395 ip
->i_d
.di_flushiter
++;
3398 * Copy the dirty parts of the inode into the on-disk
3399 * inode. We always copy out the core of the inode,
3400 * because if the inode is dirty at all the core must
3403 xfs_dinode_to_disk(&dip
->di_core
, &ip
->i_d
);
3405 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3406 if (ip
->i_d
.di_flushiter
== DI_MAX_FLUSH
)
3407 ip
->i_d
.di_flushiter
= 0;
3410 * If this is really an old format inode and the superblock version
3411 * has not been updated to support only new format inodes, then
3412 * convert back to the old inode format. If the superblock version
3413 * has been updated, then make the conversion permanent.
3415 ASSERT(ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
||
3416 xfs_sb_version_hasnlink(&mp
->m_sb
));
3417 if (ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
) {
3418 if (!xfs_sb_version_hasnlink(&mp
->m_sb
)) {
3422 ASSERT(ip
->i_d
.di_nlink
<= XFS_MAXLINK_1
);
3423 dip
->di_core
.di_onlink
= cpu_to_be16(ip
->i_d
.di_nlink
);
3426 * The superblock version has already been bumped,
3427 * so just make the conversion to the new inode
3430 ip
->i_d
.di_version
= XFS_DINODE_VERSION_2
;
3431 dip
->di_core
.di_version
= XFS_DINODE_VERSION_2
;
3432 ip
->i_d
.di_onlink
= 0;
3433 dip
->di_core
.di_onlink
= 0;
3434 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
3435 memset(&(dip
->di_core
.di_pad
[0]), 0,
3436 sizeof(dip
->di_core
.di_pad
));
3437 ASSERT(ip
->i_d
.di_projid
== 0);
3441 xfs_iflush_fork(ip
, dip
, iip
, XFS_DATA_FORK
, bp
);
3442 if (XFS_IFORK_Q(ip
))
3443 xfs_iflush_fork(ip
, dip
, iip
, XFS_ATTR_FORK
, bp
);
3444 xfs_inobp_check(mp
, bp
);
3447 * We've recorded everything logged in the inode, so we'd
3448 * like to clear the ilf_fields bits so we don't log and
3449 * flush things unnecessarily. However, we can't stop
3450 * logging all this information until the data we've copied
3451 * into the disk buffer is written to disk. If we did we might
3452 * overwrite the copy of the inode in the log with all the
3453 * data after re-logging only part of it, and in the face of
3454 * a crash we wouldn't have all the data we need to recover.
3456 * What we do is move the bits to the ili_last_fields field.
3457 * When logging the inode, these bits are moved back to the
3458 * ilf_fields field. In the xfs_iflush_done() routine we
3459 * clear ili_last_fields, since we know that the information
3460 * those bits represent is permanently on disk. As long as
3461 * the flush completes before the inode is logged again, then
3462 * both ilf_fields and ili_last_fields will be cleared.
3464 * We can play with the ilf_fields bits here, because the inode
3465 * lock must be held exclusively in order to set bits there
3466 * and the flush lock protects the ili_last_fields bits.
3467 * Set ili_logged so the flush done
3468 * routine can tell whether or not to look in the AIL.
3469 * Also, store the current LSN of the inode so that we can tell
3470 * whether the item has moved in the AIL from xfs_iflush_done().
3471 * In order to read the lsn we need the AIL lock, because
3472 * it is a 64 bit value that cannot be read atomically.
3474 if (iip
!= NULL
&& iip
->ili_format
.ilf_fields
!= 0) {
3475 iip
->ili_last_fields
= iip
->ili_format
.ilf_fields
;
3476 iip
->ili_format
.ilf_fields
= 0;
3477 iip
->ili_logged
= 1;
3479 ASSERT(sizeof(xfs_lsn_t
) == 8); /* don't lock if it shrinks */
3480 spin_lock(&mp
->m_ail_lock
);
3481 iip
->ili_flush_lsn
= iip
->ili_item
.li_lsn
;
3482 spin_unlock(&mp
->m_ail_lock
);
3485 * Attach the function xfs_iflush_done to the inode's
3486 * buffer. This will remove the inode from the AIL
3487 * and unlock the inode's flush lock when the inode is
3488 * completely written to disk.
3490 xfs_buf_attach_iodone(bp
, (void(*)(xfs_buf_t
*,xfs_log_item_t
*))
3491 xfs_iflush_done
, (xfs_log_item_t
*)iip
);
3493 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) != NULL
);
3494 ASSERT(XFS_BUF_IODONE_FUNC(bp
) != NULL
);
3497 * We're flushing an inode which is not in the AIL and has
3498 * not been logged but has i_update_core set. For this
3499 * case we can use a B_DELWRI flush and immediately drop
3500 * the inode flush lock because we can avoid the whole
3501 * AIL state thing. It's OK to drop the flush lock now,
3502 * because we've already locked the buffer and to do anything
3503 * you really need both.
3506 ASSERT(iip
->ili_logged
== 0);
3507 ASSERT(iip
->ili_last_fields
== 0);
3508 ASSERT((iip
->ili_item
.li_flags
& XFS_LI_IN_AIL
) == 0);
3516 return XFS_ERROR(EFSCORRUPTED
);
3521 #ifdef XFS_ILOCK_TRACE
3522 ktrace_t
*xfs_ilock_trace_buf
;
3525 xfs_ilock_trace(xfs_inode_t
*ip
, int lock
, unsigned int lockflags
, inst_t
*ra
)
3527 ktrace_enter(ip
->i_lock_trace
,
3529 (void *)(unsigned long)lock
, /* 1 = LOCK, 3=UNLOCK, etc */
3530 (void *)(unsigned long)lockflags
, /* XFS_ILOCK_EXCL etc */
3531 (void *)ra
, /* caller of ilock */
3532 (void *)(unsigned long)current_cpu(),
3533 (void *)(unsigned long)current_pid(),
3534 NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
);
3539 * Return a pointer to the extent record at file index idx.
3541 xfs_bmbt_rec_host_t
*
3543 xfs_ifork_t
*ifp
, /* inode fork pointer */
3544 xfs_extnum_t idx
) /* index of target extent */
3547 if ((ifp
->if_flags
& XFS_IFEXTIREC
) && (idx
== 0)) {
3548 return ifp
->if_u1
.if_ext_irec
->er_extbuf
;
3549 } else if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3550 xfs_ext_irec_t
*erp
; /* irec pointer */
3551 int erp_idx
= 0; /* irec index */
3552 xfs_extnum_t page_idx
= idx
; /* ext index in target list */
3554 erp
= xfs_iext_idx_to_irec(ifp
, &page_idx
, &erp_idx
, 0);
3555 return &erp
->er_extbuf
[page_idx
];
3556 } else if (ifp
->if_bytes
) {
3557 return &ifp
->if_u1
.if_extents
[idx
];
3564 * Insert new item(s) into the extent records for incore inode
3565 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3569 xfs_ifork_t
*ifp
, /* inode fork pointer */
3570 xfs_extnum_t idx
, /* starting index of new items */
3571 xfs_extnum_t count
, /* number of inserted items */
3572 xfs_bmbt_irec_t
*new) /* items to insert */
3574 xfs_extnum_t i
; /* extent record index */
3576 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
3577 xfs_iext_add(ifp
, idx
, count
);
3578 for (i
= idx
; i
< idx
+ count
; i
++, new++)
3579 xfs_bmbt_set_all(xfs_iext_get_ext(ifp
, i
), new);
3583 * This is called when the amount of space required for incore file
3584 * extents needs to be increased. The ext_diff parameter stores the
3585 * number of new extents being added and the idx parameter contains
3586 * the extent index where the new extents will be added. If the new
3587 * extents are being appended, then we just need to (re)allocate and
3588 * initialize the space. Otherwise, if the new extents are being
3589 * inserted into the middle of the existing entries, a bit more work
3590 * is required to make room for the new extents to be inserted. The
3591 * caller is responsible for filling in the new extent entries upon
3596 xfs_ifork_t
*ifp
, /* inode fork pointer */
3597 xfs_extnum_t idx
, /* index to begin adding exts */
3598 int ext_diff
) /* number of extents to add */
3600 int byte_diff
; /* new bytes being added */
3601 int new_size
; /* size of extents after adding */
3602 xfs_extnum_t nextents
; /* number of extents in file */
3604 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3605 ASSERT((idx
>= 0) && (idx
<= nextents
));
3606 byte_diff
= ext_diff
* sizeof(xfs_bmbt_rec_t
);
3607 new_size
= ifp
->if_bytes
+ byte_diff
;
3609 * If the new number of extents (nextents + ext_diff)
3610 * fits inside the inode, then continue to use the inline
3613 if (nextents
+ ext_diff
<= XFS_INLINE_EXTS
) {
3614 if (idx
< nextents
) {
3615 memmove(&ifp
->if_u2
.if_inline_ext
[idx
+ ext_diff
],
3616 &ifp
->if_u2
.if_inline_ext
[idx
],
3617 (nextents
- idx
) * sizeof(xfs_bmbt_rec_t
));
3618 memset(&ifp
->if_u2
.if_inline_ext
[idx
], 0, byte_diff
);
3620 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
3621 ifp
->if_real_bytes
= 0;
3622 ifp
->if_lastex
= nextents
+ ext_diff
;
3625 * Otherwise use a linear (direct) extent list.
3626 * If the extents are currently inside the inode,
3627 * xfs_iext_realloc_direct will switch us from
3628 * inline to direct extent allocation mode.
3630 else if (nextents
+ ext_diff
<= XFS_LINEAR_EXTS
) {
3631 xfs_iext_realloc_direct(ifp
, new_size
);
3632 if (idx
< nextents
) {
3633 memmove(&ifp
->if_u1
.if_extents
[idx
+ ext_diff
],
3634 &ifp
->if_u1
.if_extents
[idx
],
3635 (nextents
- idx
) * sizeof(xfs_bmbt_rec_t
));
3636 memset(&ifp
->if_u1
.if_extents
[idx
], 0, byte_diff
);
3639 /* Indirection array */
3641 xfs_ext_irec_t
*erp
;
3645 ASSERT(nextents
+ ext_diff
> XFS_LINEAR_EXTS
);
3646 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3647 erp
= xfs_iext_idx_to_irec(ifp
, &page_idx
, &erp_idx
, 1);
3649 xfs_iext_irec_init(ifp
);
3650 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3651 erp
= ifp
->if_u1
.if_ext_irec
;
3653 /* Extents fit in target extent page */
3654 if (erp
&& erp
->er_extcount
+ ext_diff
<= XFS_LINEAR_EXTS
) {
3655 if (page_idx
< erp
->er_extcount
) {
3656 memmove(&erp
->er_extbuf
[page_idx
+ ext_diff
],
3657 &erp
->er_extbuf
[page_idx
],
3658 (erp
->er_extcount
- page_idx
) *
3659 sizeof(xfs_bmbt_rec_t
));
3660 memset(&erp
->er_extbuf
[page_idx
], 0, byte_diff
);
3662 erp
->er_extcount
+= ext_diff
;
3663 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, ext_diff
);
3665 /* Insert a new extent page */
3667 xfs_iext_add_indirect_multi(ifp
,
3668 erp_idx
, page_idx
, ext_diff
);
3671 * If extent(s) are being appended to the last page in
3672 * the indirection array and the new extent(s) don't fit
3673 * in the page, then erp is NULL and erp_idx is set to
3674 * the next index needed in the indirection array.
3677 int count
= ext_diff
;
3680 erp
= xfs_iext_irec_new(ifp
, erp_idx
);
3681 erp
->er_extcount
= count
;
3682 count
-= MIN(count
, (int)XFS_LINEAR_EXTS
);
3689 ifp
->if_bytes
= new_size
;
3693 * This is called when incore extents are being added to the indirection
3694 * array and the new extents do not fit in the target extent list. The
3695 * erp_idx parameter contains the irec index for the target extent list
3696 * in the indirection array, and the idx parameter contains the extent
3697 * index within the list. The number of extents being added is stored
3698 * in the count parameter.
3700 * |-------| |-------|
3701 * | | | | idx - number of extents before idx
3703 * | | | | count - number of extents being inserted at idx
3704 * |-------| |-------|
3705 * | count | | nex2 | nex2 - number of extents after idx + count
3706 * |-------| |-------|
3709 xfs_iext_add_indirect_multi(
3710 xfs_ifork_t
*ifp
, /* inode fork pointer */
3711 int erp_idx
, /* target extent irec index */
3712 xfs_extnum_t idx
, /* index within target list */
3713 int count
) /* new extents being added */
3715 int byte_diff
; /* new bytes being added */
3716 xfs_ext_irec_t
*erp
; /* pointer to irec entry */
3717 xfs_extnum_t ext_diff
; /* number of extents to add */
3718 xfs_extnum_t ext_cnt
; /* new extents still needed */
3719 xfs_extnum_t nex2
; /* extents after idx + count */
3720 xfs_bmbt_rec_t
*nex2_ep
= NULL
; /* temp list for nex2 extents */
3721 int nlists
; /* number of irec's (lists) */
3723 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3724 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
3725 nex2
= erp
->er_extcount
- idx
;
3726 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3729 * Save second part of target extent list
3730 * (all extents past */
3732 byte_diff
= nex2
* sizeof(xfs_bmbt_rec_t
);
3733 nex2_ep
= (xfs_bmbt_rec_t
*) kmem_alloc(byte_diff
, KM_NOFS
);
3734 memmove(nex2_ep
, &erp
->er_extbuf
[idx
], byte_diff
);
3735 erp
->er_extcount
-= nex2
;
3736 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, -nex2
);
3737 memset(&erp
->er_extbuf
[idx
], 0, byte_diff
);
3741 * Add the new extents to the end of the target
3742 * list, then allocate new irec record(s) and
3743 * extent buffer(s) as needed to store the rest
3744 * of the new extents.
3747 ext_diff
= MIN(ext_cnt
, (int)XFS_LINEAR_EXTS
- erp
->er_extcount
);
3749 erp
->er_extcount
+= ext_diff
;
3750 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, ext_diff
);
3751 ext_cnt
-= ext_diff
;
3755 erp
= xfs_iext_irec_new(ifp
, erp_idx
);
3756 ext_diff
= MIN(ext_cnt
, (int)XFS_LINEAR_EXTS
);
3757 erp
->er_extcount
= ext_diff
;
3758 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, ext_diff
);
3759 ext_cnt
-= ext_diff
;
3762 /* Add nex2 extents back to indirection array */
3764 xfs_extnum_t ext_avail
;
3767 byte_diff
= nex2
* sizeof(xfs_bmbt_rec_t
);
3768 ext_avail
= XFS_LINEAR_EXTS
- erp
->er_extcount
;
3771 * If nex2 extents fit in the current page, append
3772 * nex2_ep after the new extents.
3774 if (nex2
<= ext_avail
) {
3775 i
= erp
->er_extcount
;
3778 * Otherwise, check if space is available in the
3781 else if ((erp_idx
< nlists
- 1) &&
3782 (nex2
<= (ext_avail
= XFS_LINEAR_EXTS
-
3783 ifp
->if_u1
.if_ext_irec
[erp_idx
+1].er_extcount
))) {
3786 /* Create a hole for nex2 extents */
3787 memmove(&erp
->er_extbuf
[nex2
], erp
->er_extbuf
,
3788 erp
->er_extcount
* sizeof(xfs_bmbt_rec_t
));
3791 * Final choice, create a new extent page for
3796 erp
= xfs_iext_irec_new(ifp
, erp_idx
);
3798 memmove(&erp
->er_extbuf
[i
], nex2_ep
, byte_diff
);
3800 erp
->er_extcount
+= nex2
;
3801 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, nex2
);
3806 * This is called when the amount of space required for incore file
3807 * extents needs to be decreased. The ext_diff parameter stores the
3808 * number of extents to be removed and the idx parameter contains
3809 * the extent index where the extents will be removed from.
3811 * If the amount of space needed has decreased below the linear
3812 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3813 * extent array. Otherwise, use kmem_realloc() to adjust the
3814 * size to what is needed.
3818 xfs_ifork_t
*ifp
, /* inode fork pointer */
3819 xfs_extnum_t idx
, /* index to begin removing exts */
3820 int ext_diff
) /* number of extents to remove */
3822 xfs_extnum_t nextents
; /* number of extents in file */
3823 int new_size
; /* size of extents after removal */
3825 ASSERT(ext_diff
> 0);
3826 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3827 new_size
= (nextents
- ext_diff
) * sizeof(xfs_bmbt_rec_t
);
3829 if (new_size
== 0) {
3830 xfs_iext_destroy(ifp
);
3831 } else if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3832 xfs_iext_remove_indirect(ifp
, idx
, ext_diff
);
3833 } else if (ifp
->if_real_bytes
) {
3834 xfs_iext_remove_direct(ifp
, idx
, ext_diff
);
3836 xfs_iext_remove_inline(ifp
, idx
, ext_diff
);
3838 ifp
->if_bytes
= new_size
;
3842 * This removes ext_diff extents from the inline buffer, beginning
3843 * at extent index idx.
3846 xfs_iext_remove_inline(
3847 xfs_ifork_t
*ifp
, /* inode fork pointer */
3848 xfs_extnum_t idx
, /* index to begin removing exts */
3849 int ext_diff
) /* number of extents to remove */
3851 int nextents
; /* number of extents in file */
3853 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
));
3854 ASSERT(idx
< XFS_INLINE_EXTS
);
3855 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3856 ASSERT(((nextents
- ext_diff
) > 0) &&
3857 (nextents
- ext_diff
) < XFS_INLINE_EXTS
);
3859 if (idx
+ ext_diff
< nextents
) {
3860 memmove(&ifp
->if_u2
.if_inline_ext
[idx
],
3861 &ifp
->if_u2
.if_inline_ext
[idx
+ ext_diff
],
3862 (nextents
- (idx
+ ext_diff
)) *
3863 sizeof(xfs_bmbt_rec_t
));
3864 memset(&ifp
->if_u2
.if_inline_ext
[nextents
- ext_diff
],
3865 0, ext_diff
* sizeof(xfs_bmbt_rec_t
));
3867 memset(&ifp
->if_u2
.if_inline_ext
[idx
], 0,
3868 ext_diff
* sizeof(xfs_bmbt_rec_t
));
3873 * This removes ext_diff extents from a linear (direct) extent list,
3874 * beginning at extent index idx. If the extents are being removed
3875 * from the end of the list (ie. truncate) then we just need to re-
3876 * allocate the list to remove the extra space. Otherwise, if the
3877 * extents are being removed from the middle of the existing extent
3878 * entries, then we first need to move the extent records beginning
3879 * at idx + ext_diff up in the list to overwrite the records being
3880 * removed, then remove the extra space via kmem_realloc.
3883 xfs_iext_remove_direct(
3884 xfs_ifork_t
*ifp
, /* inode fork pointer */
3885 xfs_extnum_t idx
, /* index to begin removing exts */
3886 int ext_diff
) /* number of extents to remove */
3888 xfs_extnum_t nextents
; /* number of extents in file */
3889 int new_size
; /* size of extents after removal */
3891 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
));
3892 new_size
= ifp
->if_bytes
-
3893 (ext_diff
* sizeof(xfs_bmbt_rec_t
));
3894 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3896 if (new_size
== 0) {
3897 xfs_iext_destroy(ifp
);
3900 /* Move extents up in the list (if needed) */
3901 if (idx
+ ext_diff
< nextents
) {
3902 memmove(&ifp
->if_u1
.if_extents
[idx
],
3903 &ifp
->if_u1
.if_extents
[idx
+ ext_diff
],
3904 (nextents
- (idx
+ ext_diff
)) *
3905 sizeof(xfs_bmbt_rec_t
));
3907 memset(&ifp
->if_u1
.if_extents
[nextents
- ext_diff
],
3908 0, ext_diff
* sizeof(xfs_bmbt_rec_t
));
3910 * Reallocate the direct extent list. If the extents
3911 * will fit inside the inode then xfs_iext_realloc_direct
3912 * will switch from direct to inline extent allocation
3915 xfs_iext_realloc_direct(ifp
, new_size
);
3916 ifp
->if_bytes
= new_size
;
3920 * This is called when incore extents are being removed from the
3921 * indirection array and the extents being removed span multiple extent
3922 * buffers. The idx parameter contains the file extent index where we
3923 * want to begin removing extents, and the count parameter contains
3924 * how many extents need to be removed.
3926 * |-------| |-------|
3927 * | nex1 | | | nex1 - number of extents before idx
3928 * |-------| | count |
3929 * | | | | count - number of extents being removed at idx
3930 * | count | |-------|
3931 * | | | nex2 | nex2 - number of extents after idx + count
3932 * |-------| |-------|
3935 xfs_iext_remove_indirect(
3936 xfs_ifork_t
*ifp
, /* inode fork pointer */
3937 xfs_extnum_t idx
, /* index to begin removing extents */
3938 int count
) /* number of extents to remove */
3940 xfs_ext_irec_t
*erp
; /* indirection array pointer */
3941 int erp_idx
= 0; /* indirection array index */
3942 xfs_extnum_t ext_cnt
; /* extents left to remove */
3943 xfs_extnum_t ext_diff
; /* extents to remove in current list */
3944 xfs_extnum_t nex1
; /* number of extents before idx */
3945 xfs_extnum_t nex2
; /* extents after idx + count */
3946 int nlists
; /* entries in indirection array */
3947 int page_idx
= idx
; /* index in target extent list */
3949 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3950 erp
= xfs_iext_idx_to_irec(ifp
, &page_idx
, &erp_idx
, 0);
3951 ASSERT(erp
!= NULL
);
3952 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3956 nex2
= MAX((erp
->er_extcount
- (nex1
+ ext_cnt
)), 0);
3957 ext_diff
= MIN(ext_cnt
, (erp
->er_extcount
- nex1
));
3959 * Check for deletion of entire list;
3960 * xfs_iext_irec_remove() updates extent offsets.
3962 if (ext_diff
== erp
->er_extcount
) {
3963 xfs_iext_irec_remove(ifp
, erp_idx
);
3964 ext_cnt
-= ext_diff
;
3967 ASSERT(erp_idx
< ifp
->if_real_bytes
/
3969 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
3976 /* Move extents up (if needed) */
3978 memmove(&erp
->er_extbuf
[nex1
],
3979 &erp
->er_extbuf
[nex1
+ ext_diff
],
3980 nex2
* sizeof(xfs_bmbt_rec_t
));
3982 /* Zero out rest of page */
3983 memset(&erp
->er_extbuf
[nex1
+ nex2
], 0, (XFS_IEXT_BUFSZ
-
3984 ((nex1
+ nex2
) * sizeof(xfs_bmbt_rec_t
))));
3985 /* Update remaining counters */
3986 erp
->er_extcount
-= ext_diff
;
3987 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, -ext_diff
);
3988 ext_cnt
-= ext_diff
;
3993 ifp
->if_bytes
-= count
* sizeof(xfs_bmbt_rec_t
);
3994 xfs_iext_irec_compact(ifp
);
3998 * Create, destroy, or resize a linear (direct) block of extents.
4001 xfs_iext_realloc_direct(
4002 xfs_ifork_t
*ifp
, /* inode fork pointer */
4003 int new_size
) /* new size of extents */
4005 int rnew_size
; /* real new size of extents */
4007 rnew_size
= new_size
;
4009 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
) ||
4010 ((new_size
>= 0) && (new_size
<= XFS_IEXT_BUFSZ
) &&
4011 (new_size
!= ifp
->if_real_bytes
)));
4013 /* Free extent records */
4014 if (new_size
== 0) {
4015 xfs_iext_destroy(ifp
);
4017 /* Resize direct extent list and zero any new bytes */
4018 else if (ifp
->if_real_bytes
) {
4019 /* Check if extents will fit inside the inode */
4020 if (new_size
<= XFS_INLINE_EXTS
* sizeof(xfs_bmbt_rec_t
)) {
4021 xfs_iext_direct_to_inline(ifp
, new_size
/
4022 (uint
)sizeof(xfs_bmbt_rec_t
));
4023 ifp
->if_bytes
= new_size
;
4026 if (!is_power_of_2(new_size
)){
4027 rnew_size
= roundup_pow_of_two(new_size
);
4029 if (rnew_size
!= ifp
->if_real_bytes
) {
4030 ifp
->if_u1
.if_extents
=
4031 kmem_realloc(ifp
->if_u1
.if_extents
,
4033 ifp
->if_real_bytes
, KM_NOFS
);
4035 if (rnew_size
> ifp
->if_real_bytes
) {
4036 memset(&ifp
->if_u1
.if_extents
[ifp
->if_bytes
/
4037 (uint
)sizeof(xfs_bmbt_rec_t
)], 0,
4038 rnew_size
- ifp
->if_real_bytes
);
4042 * Switch from the inline extent buffer to a direct
4043 * extent list. Be sure to include the inline extent
4044 * bytes in new_size.
4047 new_size
+= ifp
->if_bytes
;
4048 if (!is_power_of_2(new_size
)) {
4049 rnew_size
= roundup_pow_of_two(new_size
);
4051 xfs_iext_inline_to_direct(ifp
, rnew_size
);
4053 ifp
->if_real_bytes
= rnew_size
;
4054 ifp
->if_bytes
= new_size
;
4058 * Switch from linear (direct) extent records to inline buffer.
4061 xfs_iext_direct_to_inline(
4062 xfs_ifork_t
*ifp
, /* inode fork pointer */
4063 xfs_extnum_t nextents
) /* number of extents in file */
4065 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
4066 ASSERT(nextents
<= XFS_INLINE_EXTS
);
4068 * The inline buffer was zeroed when we switched
4069 * from inline to direct extent allocation mode,
4070 * so we don't need to clear it here.
4072 memcpy(ifp
->if_u2
.if_inline_ext
, ifp
->if_u1
.if_extents
,
4073 nextents
* sizeof(xfs_bmbt_rec_t
));
4074 kmem_free(ifp
->if_u1
.if_extents
);
4075 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
4076 ifp
->if_real_bytes
= 0;
4080 * Switch from inline buffer to linear (direct) extent records.
4081 * new_size should already be rounded up to the next power of 2
4082 * by the caller (when appropriate), so use new_size as it is.
4083 * However, since new_size may be rounded up, we can't update
4084 * if_bytes here. It is the caller's responsibility to update
4085 * if_bytes upon return.
4088 xfs_iext_inline_to_direct(
4089 xfs_ifork_t
*ifp
, /* inode fork pointer */
4090 int new_size
) /* number of extents in file */
4092 ifp
->if_u1
.if_extents
= kmem_alloc(new_size
, KM_NOFS
);
4093 memset(ifp
->if_u1
.if_extents
, 0, new_size
);
4094 if (ifp
->if_bytes
) {
4095 memcpy(ifp
->if_u1
.if_extents
, ifp
->if_u2
.if_inline_ext
,
4097 memset(ifp
->if_u2
.if_inline_ext
, 0, XFS_INLINE_EXTS
*
4098 sizeof(xfs_bmbt_rec_t
));
4100 ifp
->if_real_bytes
= new_size
;
4104 * Resize an extent indirection array to new_size bytes.
4107 xfs_iext_realloc_indirect(
4108 xfs_ifork_t
*ifp
, /* inode fork pointer */
4109 int new_size
) /* new indirection array size */
4111 int nlists
; /* number of irec's (ex lists) */
4112 int size
; /* current indirection array size */
4114 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4115 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4116 size
= nlists
* sizeof(xfs_ext_irec_t
);
4117 ASSERT(ifp
->if_real_bytes
);
4118 ASSERT((new_size
>= 0) && (new_size
!= size
));
4119 if (new_size
== 0) {
4120 xfs_iext_destroy(ifp
);
4122 ifp
->if_u1
.if_ext_irec
= (xfs_ext_irec_t
*)
4123 kmem_realloc(ifp
->if_u1
.if_ext_irec
,
4124 new_size
, size
, KM_NOFS
);
4129 * Switch from indirection array to linear (direct) extent allocations.
4132 xfs_iext_indirect_to_direct(
4133 xfs_ifork_t
*ifp
) /* inode fork pointer */
4135 xfs_bmbt_rec_host_t
*ep
; /* extent record pointer */
4136 xfs_extnum_t nextents
; /* number of extents in file */
4137 int size
; /* size of file extents */
4139 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4140 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4141 ASSERT(nextents
<= XFS_LINEAR_EXTS
);
4142 size
= nextents
* sizeof(xfs_bmbt_rec_t
);
4144 xfs_iext_irec_compact_pages(ifp
);
4145 ASSERT(ifp
->if_real_bytes
== XFS_IEXT_BUFSZ
);
4147 ep
= ifp
->if_u1
.if_ext_irec
->er_extbuf
;
4148 kmem_free(ifp
->if_u1
.if_ext_irec
);
4149 ifp
->if_flags
&= ~XFS_IFEXTIREC
;
4150 ifp
->if_u1
.if_extents
= ep
;
4151 ifp
->if_bytes
= size
;
4152 if (nextents
< XFS_LINEAR_EXTS
) {
4153 xfs_iext_realloc_direct(ifp
, size
);
4158 * Free incore file extents.
4162 xfs_ifork_t
*ifp
) /* inode fork pointer */
4164 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
4168 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4169 for (erp_idx
= nlists
- 1; erp_idx
>= 0 ; erp_idx
--) {
4170 xfs_iext_irec_remove(ifp
, erp_idx
);
4172 ifp
->if_flags
&= ~XFS_IFEXTIREC
;
4173 } else if (ifp
->if_real_bytes
) {
4174 kmem_free(ifp
->if_u1
.if_extents
);
4175 } else if (ifp
->if_bytes
) {
4176 memset(ifp
->if_u2
.if_inline_ext
, 0, XFS_INLINE_EXTS
*
4177 sizeof(xfs_bmbt_rec_t
));
4179 ifp
->if_u1
.if_extents
= NULL
;
4180 ifp
->if_real_bytes
= 0;
4185 * Return a pointer to the extent record for file system block bno.
4187 xfs_bmbt_rec_host_t
* /* pointer to found extent record */
4188 xfs_iext_bno_to_ext(
4189 xfs_ifork_t
*ifp
, /* inode fork pointer */
4190 xfs_fileoff_t bno
, /* block number to search for */
4191 xfs_extnum_t
*idxp
) /* index of target extent */
4193 xfs_bmbt_rec_host_t
*base
; /* pointer to first extent */
4194 xfs_filblks_t blockcount
= 0; /* number of blocks in extent */
4195 xfs_bmbt_rec_host_t
*ep
= NULL
; /* pointer to target extent */
4196 xfs_ext_irec_t
*erp
= NULL
; /* indirection array pointer */
4197 int high
; /* upper boundary in search */
4198 xfs_extnum_t idx
= 0; /* index of target extent */
4199 int low
; /* lower boundary in search */
4200 xfs_extnum_t nextents
; /* number of file extents */
4201 xfs_fileoff_t startoff
= 0; /* start offset of extent */
4203 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4204 if (nextents
== 0) {
4209 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
4210 /* Find target extent list */
4212 erp
= xfs_iext_bno_to_irec(ifp
, bno
, &erp_idx
);
4213 base
= erp
->er_extbuf
;
4214 high
= erp
->er_extcount
- 1;
4216 base
= ifp
->if_u1
.if_extents
;
4217 high
= nextents
- 1;
4219 /* Binary search extent records */
4220 while (low
<= high
) {
4221 idx
= (low
+ high
) >> 1;
4223 startoff
= xfs_bmbt_get_startoff(ep
);
4224 blockcount
= xfs_bmbt_get_blockcount(ep
);
4225 if (bno
< startoff
) {
4227 } else if (bno
>= startoff
+ blockcount
) {
4230 /* Convert back to file-based extent index */
4231 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
4232 idx
+= erp
->er_extoff
;
4238 /* Convert back to file-based extent index */
4239 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
4240 idx
+= erp
->er_extoff
;
4242 if (bno
>= startoff
+ blockcount
) {
4243 if (++idx
== nextents
) {
4246 ep
= xfs_iext_get_ext(ifp
, idx
);
4254 * Return a pointer to the indirection array entry containing the
4255 * extent record for filesystem block bno. Store the index of the
4256 * target irec in *erp_idxp.
4258 xfs_ext_irec_t
* /* pointer to found extent record */
4259 xfs_iext_bno_to_irec(
4260 xfs_ifork_t
*ifp
, /* inode fork pointer */
4261 xfs_fileoff_t bno
, /* block number to search for */
4262 int *erp_idxp
) /* irec index of target ext list */
4264 xfs_ext_irec_t
*erp
= NULL
; /* indirection array pointer */
4265 xfs_ext_irec_t
*erp_next
; /* next indirection array entry */
4266 int erp_idx
; /* indirection array index */
4267 int nlists
; /* number of extent irec's (lists) */
4268 int high
; /* binary search upper limit */
4269 int low
; /* binary search lower limit */
4271 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4272 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4276 while (low
<= high
) {
4277 erp_idx
= (low
+ high
) >> 1;
4278 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4279 erp_next
= erp_idx
< nlists
- 1 ? erp
+ 1 : NULL
;
4280 if (bno
< xfs_bmbt_get_startoff(erp
->er_extbuf
)) {
4282 } else if (erp_next
&& bno
>=
4283 xfs_bmbt_get_startoff(erp_next
->er_extbuf
)) {
4289 *erp_idxp
= erp_idx
;
4294 * Return a pointer to the indirection array entry containing the
4295 * extent record at file extent index *idxp. Store the index of the
4296 * target irec in *erp_idxp and store the page index of the target
4297 * extent record in *idxp.
4300 xfs_iext_idx_to_irec(
4301 xfs_ifork_t
*ifp
, /* inode fork pointer */
4302 xfs_extnum_t
*idxp
, /* extent index (file -> page) */
4303 int *erp_idxp
, /* pointer to target irec */
4304 int realloc
) /* new bytes were just added */
4306 xfs_ext_irec_t
*prev
; /* pointer to previous irec */
4307 xfs_ext_irec_t
*erp
= NULL
; /* pointer to current irec */
4308 int erp_idx
; /* indirection array index */
4309 int nlists
; /* number of irec's (ex lists) */
4310 int high
; /* binary search upper limit */
4311 int low
; /* binary search lower limit */
4312 xfs_extnum_t page_idx
= *idxp
; /* extent index in target list */
4314 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4315 ASSERT(page_idx
>= 0 && page_idx
<=
4316 ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
));
4317 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4322 /* Binary search extent irec's */
4323 while (low
<= high
) {
4324 erp_idx
= (low
+ high
) >> 1;
4325 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4326 prev
= erp_idx
> 0 ? erp
- 1 : NULL
;
4327 if (page_idx
< erp
->er_extoff
|| (page_idx
== erp
->er_extoff
&&
4328 realloc
&& prev
&& prev
->er_extcount
< XFS_LINEAR_EXTS
)) {
4330 } else if (page_idx
> erp
->er_extoff
+ erp
->er_extcount
||
4331 (page_idx
== erp
->er_extoff
+ erp
->er_extcount
&&
4334 } else if (page_idx
== erp
->er_extoff
+ erp
->er_extcount
&&
4335 erp
->er_extcount
== XFS_LINEAR_EXTS
) {
4339 erp
= erp_idx
< nlists
? erp
+ 1 : NULL
;
4342 page_idx
-= erp
->er_extoff
;
4347 *erp_idxp
= erp_idx
;
4352 * Allocate and initialize an indirection array once the space needed
4353 * for incore extents increases above XFS_IEXT_BUFSZ.
4357 xfs_ifork_t
*ifp
) /* inode fork pointer */
4359 xfs_ext_irec_t
*erp
; /* indirection array pointer */
4360 xfs_extnum_t nextents
; /* number of extents in file */
4362 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
));
4363 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4364 ASSERT(nextents
<= XFS_LINEAR_EXTS
);
4366 erp
= kmem_alloc(sizeof(xfs_ext_irec_t
), KM_NOFS
);
4368 if (nextents
== 0) {
4369 ifp
->if_u1
.if_extents
= kmem_alloc(XFS_IEXT_BUFSZ
, KM_NOFS
);
4370 } else if (!ifp
->if_real_bytes
) {
4371 xfs_iext_inline_to_direct(ifp
, XFS_IEXT_BUFSZ
);
4372 } else if (ifp
->if_real_bytes
< XFS_IEXT_BUFSZ
) {
4373 xfs_iext_realloc_direct(ifp
, XFS_IEXT_BUFSZ
);
4375 erp
->er_extbuf
= ifp
->if_u1
.if_extents
;
4376 erp
->er_extcount
= nextents
;
4379 ifp
->if_flags
|= XFS_IFEXTIREC
;
4380 ifp
->if_real_bytes
= XFS_IEXT_BUFSZ
;
4381 ifp
->if_bytes
= nextents
* sizeof(xfs_bmbt_rec_t
);
4382 ifp
->if_u1
.if_ext_irec
= erp
;
4388 * Allocate and initialize a new entry in the indirection array.
4392 xfs_ifork_t
*ifp
, /* inode fork pointer */
4393 int erp_idx
) /* index for new irec */
4395 xfs_ext_irec_t
*erp
; /* indirection array pointer */
4396 int i
; /* loop counter */
4397 int nlists
; /* number of irec's (ex lists) */
4399 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4400 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4402 /* Resize indirection array */
4403 xfs_iext_realloc_indirect(ifp
, ++nlists
*
4404 sizeof(xfs_ext_irec_t
));
4406 * Move records down in the array so the
4407 * new page can use erp_idx.
4409 erp
= ifp
->if_u1
.if_ext_irec
;
4410 for (i
= nlists
- 1; i
> erp_idx
; i
--) {
4411 memmove(&erp
[i
], &erp
[i
-1], sizeof(xfs_ext_irec_t
));
4413 ASSERT(i
== erp_idx
);
4415 /* Initialize new extent record */
4416 erp
= ifp
->if_u1
.if_ext_irec
;
4417 erp
[erp_idx
].er_extbuf
= kmem_alloc(XFS_IEXT_BUFSZ
, KM_NOFS
);
4418 ifp
->if_real_bytes
= nlists
* XFS_IEXT_BUFSZ
;
4419 memset(erp
[erp_idx
].er_extbuf
, 0, XFS_IEXT_BUFSZ
);
4420 erp
[erp_idx
].er_extcount
= 0;
4421 erp
[erp_idx
].er_extoff
= erp_idx
> 0 ?
4422 erp
[erp_idx
-1].er_extoff
+ erp
[erp_idx
-1].er_extcount
: 0;
4423 return (&erp
[erp_idx
]);
4427 * Remove a record from the indirection array.
4430 xfs_iext_irec_remove(
4431 xfs_ifork_t
*ifp
, /* inode fork pointer */
4432 int erp_idx
) /* irec index to remove */
4434 xfs_ext_irec_t
*erp
; /* indirection array pointer */
4435 int i
; /* loop counter */
4436 int nlists
; /* number of irec's (ex lists) */
4438 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4439 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4440 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4441 if (erp
->er_extbuf
) {
4442 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1,
4444 kmem_free(erp
->er_extbuf
);
4446 /* Compact extent records */
4447 erp
= ifp
->if_u1
.if_ext_irec
;
4448 for (i
= erp_idx
; i
< nlists
- 1; i
++) {
4449 memmove(&erp
[i
], &erp
[i
+1], sizeof(xfs_ext_irec_t
));
4452 * Manually free the last extent record from the indirection
4453 * array. A call to xfs_iext_realloc_indirect() with a size
4454 * of zero would result in a call to xfs_iext_destroy() which
4455 * would in turn call this function again, creating a nasty
4459 xfs_iext_realloc_indirect(ifp
,
4460 nlists
* sizeof(xfs_ext_irec_t
));
4462 kmem_free(ifp
->if_u1
.if_ext_irec
);
4464 ifp
->if_real_bytes
= nlists
* XFS_IEXT_BUFSZ
;
4468 * This is called to clean up large amounts of unused memory allocated
4469 * by the indirection array. Before compacting anything though, verify
4470 * that the indirection array is still needed and switch back to the
4471 * linear extent list (or even the inline buffer) if possible. The
4472 * compaction policy is as follows:
4474 * Full Compaction: Extents fit into a single page (or inline buffer)
4475 * Partial Compaction: Extents occupy less than 50% of allocated space
4476 * No Compaction: Extents occupy at least 50% of allocated space
4479 xfs_iext_irec_compact(
4480 xfs_ifork_t
*ifp
) /* inode fork pointer */
4482 xfs_extnum_t nextents
; /* number of extents in file */
4483 int nlists
; /* number of irec's (ex lists) */
4485 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4486 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4487 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4489 if (nextents
== 0) {
4490 xfs_iext_destroy(ifp
);
4491 } else if (nextents
<= XFS_INLINE_EXTS
) {
4492 xfs_iext_indirect_to_direct(ifp
);
4493 xfs_iext_direct_to_inline(ifp
, nextents
);
4494 } else if (nextents
<= XFS_LINEAR_EXTS
) {
4495 xfs_iext_indirect_to_direct(ifp
);
4496 } else if (nextents
< (nlists
* XFS_LINEAR_EXTS
) >> 1) {
4497 xfs_iext_irec_compact_pages(ifp
);
4502 * Combine extents from neighboring extent pages.
4505 xfs_iext_irec_compact_pages(
4506 xfs_ifork_t
*ifp
) /* inode fork pointer */
4508 xfs_ext_irec_t
*erp
, *erp_next
;/* pointers to irec entries */
4509 int erp_idx
= 0; /* indirection array index */
4510 int nlists
; /* number of irec's (ex lists) */
4512 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4513 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4514 while (erp_idx
< nlists
- 1) {
4515 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4517 if (erp_next
->er_extcount
<=
4518 (XFS_LINEAR_EXTS
- erp
->er_extcount
)) {
4519 memcpy(&erp
->er_extbuf
[erp
->er_extcount
],
4520 erp_next
->er_extbuf
, erp_next
->er_extcount
*
4521 sizeof(xfs_bmbt_rec_t
));
4522 erp
->er_extcount
+= erp_next
->er_extcount
;
4524 * Free page before removing extent record
4525 * so er_extoffs don't get modified in
4526 * xfs_iext_irec_remove.
4528 kmem_free(erp_next
->er_extbuf
);
4529 erp_next
->er_extbuf
= NULL
;
4530 xfs_iext_irec_remove(ifp
, erp_idx
+ 1);
4531 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4539 * This is called to update the er_extoff field in the indirection
4540 * array when extents have been added or removed from one of the
4541 * extent lists. erp_idx contains the irec index to begin updating
4542 * at and ext_diff contains the number of extents that were added
4546 xfs_iext_irec_update_extoffs(
4547 xfs_ifork_t
*ifp
, /* inode fork pointer */
4548 int erp_idx
, /* irec index to update */
4549 int ext_diff
) /* number of new extents */
4551 int i
; /* loop counter */
4552 int nlists
; /* number of irec's (ex lists */
4554 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4555 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4556 for (i
= erp_idx
; i
< nlists
; i
++) {
4557 ifp
->if_u1
.if_ext_irec
[i
].er_extoff
+= ext_diff
;