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"
26 #include "xfs_trans.h"
27 #include "xfs_trans_priv.h"
31 #include "xfs_dmapi.h"
32 #include "xfs_mount.h"
33 #include "xfs_bmap_btree.h"
34 #include "xfs_alloc_btree.h"
35 #include "xfs_ialloc_btree.h"
36 #include "xfs_dir2_sf.h"
37 #include "xfs_attr_sf.h"
38 #include "xfs_dinode.h"
39 #include "xfs_inode.h"
40 #include "xfs_buf_item.h"
41 #include "xfs_inode_item.h"
42 #include "xfs_btree.h"
43 #include "xfs_btree_trace.h"
44 #include "xfs_alloc.h"
45 #include "xfs_ialloc.h"
48 #include "xfs_error.h"
49 #include "xfs_utils.h"
50 #include "xfs_dir2_trace.h"
51 #include "xfs_quota.h"
52 #include "xfs_filestream.h"
53 #include "xfs_vnodeops.h"
55 kmem_zone_t
*xfs_ifork_zone
;
56 kmem_zone_t
*xfs_inode_zone
;
59 * Used in xfs_itruncate(). This is the maximum number of extents
60 * freed from a file in a single transaction.
62 #define XFS_ITRUNC_MAX_EXTENTS 2
64 STATIC
int xfs_iflush_int(xfs_inode_t
*, xfs_buf_t
*);
65 STATIC
int xfs_iformat_local(xfs_inode_t
*, xfs_dinode_t
*, int, int);
66 STATIC
int xfs_iformat_extents(xfs_inode_t
*, xfs_dinode_t
*, int);
67 STATIC
int xfs_iformat_btree(xfs_inode_t
*, xfs_dinode_t
*, int);
71 * Make sure that the extents in the given memory buffer
81 xfs_bmbt_rec_host_t rec
;
84 for (i
= 0; i
< nrecs
; i
++) {
85 xfs_bmbt_rec_host_t
*ep
= xfs_iext_get_ext(ifp
, i
);
86 rec
.l0
= get_unaligned(&ep
->l0
);
87 rec
.l1
= get_unaligned(&ep
->l1
);
88 xfs_bmbt_get_all(&rec
, &irec
);
89 if (fmt
== XFS_EXTFMT_NOSTATE
)
90 ASSERT(irec
.br_state
== XFS_EXT_NORM
);
94 #define xfs_validate_extents(ifp, nrecs, fmt)
98 * Check that none of the inode's in the buffer have a next
99 * unlinked field of 0.
111 j
= mp
->m_inode_cluster_size
>> mp
->m_sb
.sb_inodelog
;
113 for (i
= 0; i
< j
; i
++) {
114 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
,
115 i
* mp
->m_sb
.sb_inodesize
);
116 if (!dip
->di_next_unlinked
) {
117 xfs_fs_cmn_err(CE_ALERT
, mp
,
118 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
120 ASSERT(dip
->di_next_unlinked
);
127 * Find the buffer associated with the given inode map
128 * We do basic validation checks on the buffer once it has been
129 * retrieved from disk.
135 struct xfs_imap
*imap
,
145 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, imap
->im_blkno
,
146 (int)imap
->im_len
, buf_flags
, &bp
);
148 if (error
!= EAGAIN
) {
150 "xfs_imap_to_bp: xfs_trans_read_buf()returned "
151 "an error %d on %s. Returning error.",
152 error
, mp
->m_fsname
);
154 ASSERT(buf_flags
& XFS_BUF_TRYLOCK
);
160 * Validate the magic number and version of every inode in the buffer
161 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
164 ni
= BBTOB(imap
->im_len
) >> mp
->m_sb
.sb_inodelog
;
165 #else /* usual case */
169 for (i
= 0; i
< ni
; i
++) {
173 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
,
174 (i
<< mp
->m_sb
.sb_inodelog
));
175 di_ok
= be16_to_cpu(dip
->di_magic
) == XFS_DINODE_MAGIC
&&
176 XFS_DINODE_GOOD_VERSION(dip
->di_version
);
177 if (unlikely(XFS_TEST_ERROR(!di_ok
, mp
,
178 XFS_ERRTAG_ITOBP_INOTOBP
,
179 XFS_RANDOM_ITOBP_INOTOBP
))) {
180 if (iget_flags
& XFS_IGET_UNTRUSTED
) {
181 xfs_trans_brelse(tp
, bp
);
182 return XFS_ERROR(EINVAL
);
184 XFS_CORRUPTION_ERROR("xfs_imap_to_bp",
185 XFS_ERRLEVEL_HIGH
, mp
, dip
);
188 "Device %s - bad inode magic/vsn "
189 "daddr %lld #%d (magic=%x)",
190 XFS_BUFTARG_NAME(mp
->m_ddev_targp
),
191 (unsigned long long)imap
->im_blkno
, i
,
192 be16_to_cpu(dip
->di_magic
));
194 xfs_trans_brelse(tp
, bp
);
195 return XFS_ERROR(EFSCORRUPTED
);
199 xfs_inobp_check(mp
, bp
);
202 * Mark the buffer as an inode buffer now that it looks good
204 XFS_BUF_SET_VTYPE(bp
, B_FS_INO
);
211 * This routine is called to map an inode number within a file
212 * system to the buffer containing the on-disk version of the
213 * inode. It returns a pointer to the buffer containing the
214 * on-disk inode in the bpp parameter, and in the dip parameter
215 * it returns a pointer to the on-disk inode within that buffer.
217 * If a non-zero error is returned, then the contents of bpp and
218 * dipp are undefined.
220 * Use xfs_imap() to determine the size and location of the
221 * buffer to read from disk.
233 struct xfs_imap imap
;
238 error
= xfs_imap(mp
, tp
, ino
, &imap
, imap_flags
);
242 error
= xfs_imap_to_bp(mp
, tp
, &imap
, &bp
, XFS_BUF_LOCK
, imap_flags
);
246 *dipp
= (xfs_dinode_t
*)xfs_buf_offset(bp
, imap
.im_boffset
);
248 *offset
= imap
.im_boffset
;
254 * This routine is called to map an inode to the buffer containing
255 * the on-disk version of the inode. It returns a pointer to the
256 * buffer containing the on-disk inode in the bpp parameter, and in
257 * the dip parameter it returns a pointer to the on-disk inode within
260 * If a non-zero error is returned, then the contents of bpp and
261 * dipp are undefined.
263 * The inode is expected to already been mapped to its buffer and read
264 * in once, thus we can use the mapping information stored in the inode
265 * rather than calling xfs_imap(). This allows us to avoid the overhead
266 * of looking at the inode btree for small block file systems
281 ASSERT(ip
->i_imap
.im_blkno
!= 0);
283 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &bp
, buf_flags
, 0);
288 ASSERT(buf_flags
& XFS_BUF_TRYLOCK
);
294 *dipp
= (xfs_dinode_t
*)xfs_buf_offset(bp
, ip
->i_imap
.im_boffset
);
300 * Move inode type and inode format specific information from the
301 * on-disk inode to the in-core inode. For fifos, devs, and sockets
302 * this means set if_rdev to the proper value. For files, directories,
303 * and symlinks this means to bring in the in-line data or extent
304 * pointers. For a file in B-tree format, only the root is immediately
305 * brought in-core. The rest will be in-lined in if_extents when it
306 * is first referenced (see xfs_iread_extents()).
313 xfs_attr_shortform_t
*atp
;
317 ip
->i_df
.if_ext_max
=
318 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
321 if (unlikely(be32_to_cpu(dip
->di_nextents
) +
322 be16_to_cpu(dip
->di_anextents
) >
323 be64_to_cpu(dip
->di_nblocks
))) {
324 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
325 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
326 (unsigned long long)ip
->i_ino
,
327 (int)(be32_to_cpu(dip
->di_nextents
) +
328 be16_to_cpu(dip
->di_anextents
)),
330 be64_to_cpu(dip
->di_nblocks
));
331 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW
,
333 return XFS_ERROR(EFSCORRUPTED
);
336 if (unlikely(dip
->di_forkoff
> ip
->i_mount
->m_sb
.sb_inodesize
)) {
337 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
338 "corrupt dinode %Lu, forkoff = 0x%x.",
339 (unsigned long long)ip
->i_ino
,
341 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW
,
343 return XFS_ERROR(EFSCORRUPTED
);
346 if (unlikely((ip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
) &&
347 !ip
->i_mount
->m_rtdev_targp
)) {
348 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
349 "corrupt dinode %Lu, has realtime flag set.",
351 XFS_CORRUPTION_ERROR("xfs_iformat(realtime)",
352 XFS_ERRLEVEL_LOW
, ip
->i_mount
, dip
);
353 return XFS_ERROR(EFSCORRUPTED
);
356 switch (ip
->i_d
.di_mode
& S_IFMT
) {
361 if (unlikely(dip
->di_format
!= XFS_DINODE_FMT_DEV
)) {
362 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW
,
364 return XFS_ERROR(EFSCORRUPTED
);
368 ip
->i_df
.if_u2
.if_rdev
= xfs_dinode_get_rdev(dip
);
374 switch (dip
->di_format
) {
375 case XFS_DINODE_FMT_LOCAL
:
377 * no local regular files yet
379 if (unlikely((be16_to_cpu(dip
->di_mode
) & S_IFMT
) == S_IFREG
)) {
380 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
382 "(local format for regular file).",
383 (unsigned long long) ip
->i_ino
);
384 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
387 return XFS_ERROR(EFSCORRUPTED
);
390 di_size
= be64_to_cpu(dip
->di_size
);
391 if (unlikely(di_size
> XFS_DFORK_DSIZE(dip
, ip
->i_mount
))) {
392 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
394 "(bad size %Ld for local inode).",
395 (unsigned long long) ip
->i_ino
,
396 (long long) di_size
);
397 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
400 return XFS_ERROR(EFSCORRUPTED
);
404 error
= xfs_iformat_local(ip
, dip
, XFS_DATA_FORK
, size
);
406 case XFS_DINODE_FMT_EXTENTS
:
407 error
= xfs_iformat_extents(ip
, dip
, XFS_DATA_FORK
);
409 case XFS_DINODE_FMT_BTREE
:
410 error
= xfs_iformat_btree(ip
, dip
, XFS_DATA_FORK
);
413 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW
,
415 return XFS_ERROR(EFSCORRUPTED
);
420 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW
, ip
->i_mount
);
421 return XFS_ERROR(EFSCORRUPTED
);
426 if (!XFS_DFORK_Q(dip
))
428 ASSERT(ip
->i_afp
== NULL
);
429 ip
->i_afp
= kmem_zone_zalloc(xfs_ifork_zone
, KM_SLEEP
);
430 ip
->i_afp
->if_ext_max
=
431 XFS_IFORK_ASIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
432 switch (dip
->di_aformat
) {
433 case XFS_DINODE_FMT_LOCAL
:
434 atp
= (xfs_attr_shortform_t
*)XFS_DFORK_APTR(dip
);
435 size
= be16_to_cpu(atp
->hdr
.totsize
);
437 if (unlikely(size
< sizeof(struct xfs_attr_sf_hdr
))) {
438 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
440 "(bad attr fork size %Ld).",
441 (unsigned long long) ip
->i_ino
,
443 XFS_CORRUPTION_ERROR("xfs_iformat(8)",
446 return XFS_ERROR(EFSCORRUPTED
);
449 error
= xfs_iformat_local(ip
, dip
, XFS_ATTR_FORK
, size
);
451 case XFS_DINODE_FMT_EXTENTS
:
452 error
= xfs_iformat_extents(ip
, dip
, XFS_ATTR_FORK
);
454 case XFS_DINODE_FMT_BTREE
:
455 error
= xfs_iformat_btree(ip
, dip
, XFS_ATTR_FORK
);
458 error
= XFS_ERROR(EFSCORRUPTED
);
462 kmem_zone_free(xfs_ifork_zone
, ip
->i_afp
);
464 xfs_idestroy_fork(ip
, XFS_DATA_FORK
);
470 * The file is in-lined in the on-disk inode.
471 * If it fits into if_inline_data, then copy
472 * it there, otherwise allocate a buffer for it
473 * and copy the data there. Either way, set
474 * if_data to point at the data.
475 * If we allocate a buffer for the data, make
476 * sure that its size is a multiple of 4 and
477 * record the real size in i_real_bytes.
490 * If the size is unreasonable, then something
491 * is wrong and we just bail out rather than crash in
492 * kmem_alloc() or memcpy() below.
494 if (unlikely(size
> XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
))) {
495 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
497 "(bad size %d for local fork, size = %d).",
498 (unsigned long long) ip
->i_ino
, size
,
499 XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
));
500 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW
,
502 return XFS_ERROR(EFSCORRUPTED
);
504 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
507 ifp
->if_u1
.if_data
= NULL
;
508 else if (size
<= sizeof(ifp
->if_u2
.if_inline_data
))
509 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
511 real_size
= roundup(size
, 4);
512 ifp
->if_u1
.if_data
= kmem_alloc(real_size
, KM_SLEEP
);
514 ifp
->if_bytes
= size
;
515 ifp
->if_real_bytes
= real_size
;
517 memcpy(ifp
->if_u1
.if_data
, XFS_DFORK_PTR(dip
, whichfork
), size
);
518 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
519 ifp
->if_flags
|= XFS_IFINLINE
;
524 * The file consists of a set of extents all
525 * of which fit into the on-disk inode.
526 * If there are few enough extents to fit into
527 * the if_inline_ext, then copy them there.
528 * Otherwise allocate a buffer for them and copy
529 * them into it. Either way, set if_extents
530 * to point at the extents.
544 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
545 nex
= XFS_DFORK_NEXTENTS(dip
, whichfork
);
546 size
= nex
* (uint
)sizeof(xfs_bmbt_rec_t
);
549 * If the number of extents is unreasonable, then something
550 * is wrong and we just bail out rather than crash in
551 * kmem_alloc() or memcpy() below.
553 if (unlikely(size
< 0 || size
> XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
))) {
554 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
555 "corrupt inode %Lu ((a)extents = %d).",
556 (unsigned long long) ip
->i_ino
, nex
);
557 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW
,
559 return XFS_ERROR(EFSCORRUPTED
);
562 ifp
->if_real_bytes
= 0;
564 ifp
->if_u1
.if_extents
= NULL
;
565 else if (nex
<= XFS_INLINE_EXTS
)
566 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
568 xfs_iext_add(ifp
, 0, nex
);
570 ifp
->if_bytes
= size
;
572 dp
= (xfs_bmbt_rec_t
*) XFS_DFORK_PTR(dip
, whichfork
);
573 xfs_validate_extents(ifp
, nex
, XFS_EXTFMT_INODE(ip
));
574 for (i
= 0; i
< nex
; i
++, dp
++) {
575 xfs_bmbt_rec_host_t
*ep
= xfs_iext_get_ext(ifp
, i
);
576 ep
->l0
= get_unaligned_be64(&dp
->l0
);
577 ep
->l1
= get_unaligned_be64(&dp
->l1
);
579 XFS_BMAP_TRACE_EXLIST(ip
, nex
, whichfork
);
580 if (whichfork
!= XFS_DATA_FORK
||
581 XFS_EXTFMT_INODE(ip
) == XFS_EXTFMT_NOSTATE
)
582 if (unlikely(xfs_check_nostate_extents(
584 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
587 return XFS_ERROR(EFSCORRUPTED
);
590 ifp
->if_flags
|= XFS_IFEXTENTS
;
595 * The file has too many extents to fit into
596 * the inode, so they are in B-tree format.
597 * Allocate a buffer for the root of the B-tree
598 * and copy the root into it. The i_extents
599 * field will remain NULL until all of the
600 * extents are read in (when they are needed).
608 xfs_bmdr_block_t
*dfp
;
614 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
615 dfp
= (xfs_bmdr_block_t
*)XFS_DFORK_PTR(dip
, whichfork
);
616 size
= XFS_BMAP_BROOT_SPACE(dfp
);
617 nrecs
= be16_to_cpu(dfp
->bb_numrecs
);
620 * blow out if -- fork has less extents than can fit in
621 * fork (fork shouldn't be a btree format), root btree
622 * block has more records than can fit into the fork,
623 * or the number of extents is greater than the number of
626 if (unlikely(XFS_IFORK_NEXTENTS(ip
, whichfork
) <= ifp
->if_ext_max
627 || XFS_BMDR_SPACE_CALC(nrecs
) >
628 XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
)
629 || XFS_IFORK_NEXTENTS(ip
, whichfork
) > ip
->i_d
.di_nblocks
)) {
630 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
631 "corrupt inode %Lu (btree).",
632 (unsigned long long) ip
->i_ino
);
633 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW
,
635 return XFS_ERROR(EFSCORRUPTED
);
638 ifp
->if_broot_bytes
= size
;
639 ifp
->if_broot
= kmem_alloc(size
, KM_SLEEP
);
640 ASSERT(ifp
->if_broot
!= NULL
);
642 * Copy and convert from the on-disk structure
643 * to the in-memory structure.
645 xfs_bmdr_to_bmbt(ip
->i_mount
, dfp
,
646 XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
),
647 ifp
->if_broot
, size
);
648 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
649 ifp
->if_flags
|= XFS_IFBROOT
;
655 xfs_dinode_from_disk(
659 to
->di_magic
= be16_to_cpu(from
->di_magic
);
660 to
->di_mode
= be16_to_cpu(from
->di_mode
);
661 to
->di_version
= from
->di_version
;
662 to
->di_format
= from
->di_format
;
663 to
->di_onlink
= be16_to_cpu(from
->di_onlink
);
664 to
->di_uid
= be32_to_cpu(from
->di_uid
);
665 to
->di_gid
= be32_to_cpu(from
->di_gid
);
666 to
->di_nlink
= be32_to_cpu(from
->di_nlink
);
667 to
->di_projid
= be16_to_cpu(from
->di_projid
);
668 memcpy(to
->di_pad
, from
->di_pad
, sizeof(to
->di_pad
));
669 to
->di_flushiter
= be16_to_cpu(from
->di_flushiter
);
670 to
->di_atime
.t_sec
= be32_to_cpu(from
->di_atime
.t_sec
);
671 to
->di_atime
.t_nsec
= be32_to_cpu(from
->di_atime
.t_nsec
);
672 to
->di_mtime
.t_sec
= be32_to_cpu(from
->di_mtime
.t_sec
);
673 to
->di_mtime
.t_nsec
= be32_to_cpu(from
->di_mtime
.t_nsec
);
674 to
->di_ctime
.t_sec
= be32_to_cpu(from
->di_ctime
.t_sec
);
675 to
->di_ctime
.t_nsec
= be32_to_cpu(from
->di_ctime
.t_nsec
);
676 to
->di_size
= be64_to_cpu(from
->di_size
);
677 to
->di_nblocks
= be64_to_cpu(from
->di_nblocks
);
678 to
->di_extsize
= be32_to_cpu(from
->di_extsize
);
679 to
->di_nextents
= be32_to_cpu(from
->di_nextents
);
680 to
->di_anextents
= be16_to_cpu(from
->di_anextents
);
681 to
->di_forkoff
= from
->di_forkoff
;
682 to
->di_aformat
= from
->di_aformat
;
683 to
->di_dmevmask
= be32_to_cpu(from
->di_dmevmask
);
684 to
->di_dmstate
= be16_to_cpu(from
->di_dmstate
);
685 to
->di_flags
= be16_to_cpu(from
->di_flags
);
686 to
->di_gen
= be32_to_cpu(from
->di_gen
);
692 xfs_icdinode_t
*from
)
694 to
->di_magic
= cpu_to_be16(from
->di_magic
);
695 to
->di_mode
= cpu_to_be16(from
->di_mode
);
696 to
->di_version
= from
->di_version
;
697 to
->di_format
= from
->di_format
;
698 to
->di_onlink
= cpu_to_be16(from
->di_onlink
);
699 to
->di_uid
= cpu_to_be32(from
->di_uid
);
700 to
->di_gid
= cpu_to_be32(from
->di_gid
);
701 to
->di_nlink
= cpu_to_be32(from
->di_nlink
);
702 to
->di_projid
= cpu_to_be16(from
->di_projid
);
703 memcpy(to
->di_pad
, from
->di_pad
, sizeof(to
->di_pad
));
704 to
->di_flushiter
= cpu_to_be16(from
->di_flushiter
);
705 to
->di_atime
.t_sec
= cpu_to_be32(from
->di_atime
.t_sec
);
706 to
->di_atime
.t_nsec
= cpu_to_be32(from
->di_atime
.t_nsec
);
707 to
->di_mtime
.t_sec
= cpu_to_be32(from
->di_mtime
.t_sec
);
708 to
->di_mtime
.t_nsec
= cpu_to_be32(from
->di_mtime
.t_nsec
);
709 to
->di_ctime
.t_sec
= cpu_to_be32(from
->di_ctime
.t_sec
);
710 to
->di_ctime
.t_nsec
= cpu_to_be32(from
->di_ctime
.t_nsec
);
711 to
->di_size
= cpu_to_be64(from
->di_size
);
712 to
->di_nblocks
= cpu_to_be64(from
->di_nblocks
);
713 to
->di_extsize
= cpu_to_be32(from
->di_extsize
);
714 to
->di_nextents
= cpu_to_be32(from
->di_nextents
);
715 to
->di_anextents
= cpu_to_be16(from
->di_anextents
);
716 to
->di_forkoff
= from
->di_forkoff
;
717 to
->di_aformat
= from
->di_aformat
;
718 to
->di_dmevmask
= cpu_to_be32(from
->di_dmevmask
);
719 to
->di_dmstate
= cpu_to_be16(from
->di_dmstate
);
720 to
->di_flags
= cpu_to_be16(from
->di_flags
);
721 to
->di_gen
= cpu_to_be32(from
->di_gen
);
730 if (di_flags
& XFS_DIFLAG_ANY
) {
731 if (di_flags
& XFS_DIFLAG_REALTIME
)
732 flags
|= XFS_XFLAG_REALTIME
;
733 if (di_flags
& XFS_DIFLAG_PREALLOC
)
734 flags
|= XFS_XFLAG_PREALLOC
;
735 if (di_flags
& XFS_DIFLAG_IMMUTABLE
)
736 flags
|= XFS_XFLAG_IMMUTABLE
;
737 if (di_flags
& XFS_DIFLAG_APPEND
)
738 flags
|= XFS_XFLAG_APPEND
;
739 if (di_flags
& XFS_DIFLAG_SYNC
)
740 flags
|= XFS_XFLAG_SYNC
;
741 if (di_flags
& XFS_DIFLAG_NOATIME
)
742 flags
|= XFS_XFLAG_NOATIME
;
743 if (di_flags
& XFS_DIFLAG_NODUMP
)
744 flags
|= XFS_XFLAG_NODUMP
;
745 if (di_flags
& XFS_DIFLAG_RTINHERIT
)
746 flags
|= XFS_XFLAG_RTINHERIT
;
747 if (di_flags
& XFS_DIFLAG_PROJINHERIT
)
748 flags
|= XFS_XFLAG_PROJINHERIT
;
749 if (di_flags
& XFS_DIFLAG_NOSYMLINKS
)
750 flags
|= XFS_XFLAG_NOSYMLINKS
;
751 if (di_flags
& XFS_DIFLAG_EXTSIZE
)
752 flags
|= XFS_XFLAG_EXTSIZE
;
753 if (di_flags
& XFS_DIFLAG_EXTSZINHERIT
)
754 flags
|= XFS_XFLAG_EXTSZINHERIT
;
755 if (di_flags
& XFS_DIFLAG_NODEFRAG
)
756 flags
|= XFS_XFLAG_NODEFRAG
;
757 if (di_flags
& XFS_DIFLAG_FILESTREAM
)
758 flags
|= XFS_XFLAG_FILESTREAM
;
768 xfs_icdinode_t
*dic
= &ip
->i_d
;
770 return _xfs_dic2xflags(dic
->di_flags
) |
771 (XFS_IFORK_Q(ip
) ? XFS_XFLAG_HASATTR
: 0);
778 return _xfs_dic2xflags(be16_to_cpu(dip
->di_flags
)) |
779 (XFS_DFORK_Q(dip
) ? XFS_XFLAG_HASATTR
: 0);
783 * Read the disk inode attributes into the in-core inode structure.
797 * Fill in the location information in the in-core inode.
799 error
= xfs_imap(mp
, tp
, ip
->i_ino
, &ip
->i_imap
, iget_flags
);
804 * Get pointers to the on-disk inode and the buffer containing it.
806 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &bp
,
807 XFS_BUF_LOCK
, iget_flags
);
810 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, ip
->i_imap
.im_boffset
);
813 * If we got something that isn't an inode it means someone
814 * (nfs or dmi) has a stale handle.
816 if (be16_to_cpu(dip
->di_magic
) != XFS_DINODE_MAGIC
) {
818 xfs_fs_cmn_err(CE_ALERT
, mp
, "xfs_iread: "
819 "dip->di_magic (0x%x) != "
820 "XFS_DINODE_MAGIC (0x%x)",
821 be16_to_cpu(dip
->di_magic
),
824 error
= XFS_ERROR(EINVAL
);
829 * If the on-disk inode is already linked to a directory
830 * entry, copy all of the inode into the in-core inode.
831 * xfs_iformat() handles copying in the inode format
832 * specific information.
833 * Otherwise, just get the truly permanent information.
836 xfs_dinode_from_disk(&ip
->i_d
, dip
);
837 error
= xfs_iformat(ip
, dip
);
840 xfs_fs_cmn_err(CE_ALERT
, mp
, "xfs_iread: "
841 "xfs_iformat() returned error %d",
847 ip
->i_d
.di_magic
= be16_to_cpu(dip
->di_magic
);
848 ip
->i_d
.di_version
= dip
->di_version
;
849 ip
->i_d
.di_gen
= be32_to_cpu(dip
->di_gen
);
850 ip
->i_d
.di_flushiter
= be16_to_cpu(dip
->di_flushiter
);
852 * Make sure to pull in the mode here as well in
853 * case the inode is released without being used.
854 * This ensures that xfs_inactive() will see that
855 * the inode is already free and not try to mess
856 * with the uninitialized part of it.
860 * Initialize the per-fork minima and maxima for a new
861 * inode here. xfs_iformat will do it for old inodes.
863 ip
->i_df
.if_ext_max
=
864 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
868 * The inode format changed when we moved the link count and
869 * made it 32 bits long. If this is an old format inode,
870 * convert it in memory to look like a new one. If it gets
871 * flushed to disk we will convert back before flushing or
872 * logging it. We zero out the new projid field and the old link
873 * count field. We'll handle clearing the pad field (the remains
874 * of the old uuid field) when we actually convert the inode to
875 * the new format. We don't change the version number so that we
876 * can distinguish this from a real new format inode.
878 if (ip
->i_d
.di_version
== 1) {
879 ip
->i_d
.di_nlink
= ip
->i_d
.di_onlink
;
880 ip
->i_d
.di_onlink
= 0;
881 ip
->i_d
.di_projid
= 0;
884 ip
->i_delayed_blks
= 0;
885 ip
->i_size
= ip
->i_d
.di_size
;
888 * Mark the buffer containing the inode as something to keep
889 * around for a while. This helps to keep recently accessed
890 * meta-data in-core longer.
892 XFS_BUF_SET_REF(bp
, XFS_INO_REF
);
895 * Use xfs_trans_brelse() to release the buffer containing the
896 * on-disk inode, because it was acquired with xfs_trans_read_buf()
897 * in xfs_itobp() above. If tp is NULL, this is just a normal
898 * brelse(). If we're within a transaction, then xfs_trans_brelse()
899 * will only release the buffer if it is not dirty within the
900 * transaction. It will be OK to release the buffer in this case,
901 * because inodes on disk are never destroyed and we will be
902 * locking the new in-core inode before putting it in the hash
903 * table where other processes can find it. Thus we don't have
904 * to worry about the inode being changed just because we released
908 xfs_trans_brelse(tp
, bp
);
913 * Read in extents from a btree-format inode.
914 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
924 xfs_extnum_t nextents
;
927 if (unlikely(XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
)) {
928 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW
,
930 return XFS_ERROR(EFSCORRUPTED
);
932 nextents
= XFS_IFORK_NEXTENTS(ip
, whichfork
);
933 size
= nextents
* sizeof(xfs_bmbt_rec_t
);
934 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
937 * We know that the size is valid (it's checked in iformat_btree)
939 ifp
->if_lastex
= NULLEXTNUM
;
940 ifp
->if_bytes
= ifp
->if_real_bytes
= 0;
941 ifp
->if_flags
|= XFS_IFEXTENTS
;
942 xfs_iext_add(ifp
, 0, nextents
);
943 error
= xfs_bmap_read_extents(tp
, ip
, whichfork
);
945 xfs_iext_destroy(ifp
);
946 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
949 xfs_validate_extents(ifp
, nextents
, XFS_EXTFMT_INODE(ip
));
954 * Allocate an inode on disk and return a copy of its in-core version.
955 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
956 * appropriately within the inode. The uid and gid for the inode are
957 * set according to the contents of the given cred structure.
959 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
960 * has a free inode available, call xfs_iget()
961 * to obtain the in-core version of the allocated inode. Finally,
962 * fill in the inode and log its initial contents. In this case,
963 * ialloc_context would be set to NULL and call_again set to false.
965 * If xfs_dialloc() does not have an available inode,
966 * it will replenish its supply by doing an allocation. Since we can
967 * only do one allocation within a transaction without deadlocks, we
968 * must commit the current transaction before returning the inode itself.
969 * In this case, therefore, we will set call_again to true and return.
970 * The caller should then commit the current transaction, start a new
971 * transaction, and call xfs_ialloc() again to actually get the inode.
973 * To ensure that some other process does not grab the inode that
974 * was allocated during the first call to xfs_ialloc(), this routine
975 * also returns the [locked] bp pointing to the head of the freelist
976 * as ialloc_context. The caller should hold this buffer across
977 * the commit and pass it back into this routine on the second call.
979 * If we are allocating quota inodes, we do not have a parent inode
980 * to attach to or associate with (i.e. pip == NULL) because they
981 * are not linked into the directory structure - they are attached
982 * directly to the superblock - and so have no parent.
994 xfs_buf_t
**ialloc_context
,
995 boolean_t
*call_again
,
1003 int filestreams
= 0;
1006 * Call the space management code to pick
1007 * the on-disk inode to be allocated.
1009 error
= xfs_dialloc(tp
, pip
? pip
->i_ino
: 0, mode
, okalloc
,
1010 ialloc_context
, call_again
, &ino
);
1013 if (*call_again
|| ino
== NULLFSINO
) {
1017 ASSERT(*ialloc_context
== NULL
);
1020 * Get the in-core inode with the lock held exclusively.
1021 * This is because we're setting fields here we need
1022 * to prevent others from looking at until we're done.
1024 error
= xfs_trans_iget(tp
->t_mountp
, tp
, ino
,
1025 XFS_IGET_CREATE
, XFS_ILOCK_EXCL
, &ip
);
1030 ip
->i_d
.di_mode
= (__uint16_t
)mode
;
1031 ip
->i_d
.di_onlink
= 0;
1032 ip
->i_d
.di_nlink
= nlink
;
1033 ASSERT(ip
->i_d
.di_nlink
== nlink
);
1034 ip
->i_d
.di_uid
= current_fsuid();
1035 ip
->i_d
.di_gid
= current_fsgid();
1036 ip
->i_d
.di_projid
= prid
;
1037 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
1040 * If the superblock version is up to where we support new format
1041 * inodes and this is currently an old format inode, then change
1042 * the inode version number now. This way we only do the conversion
1043 * here rather than here and in the flush/logging code.
1045 if (xfs_sb_version_hasnlink(&tp
->t_mountp
->m_sb
) &&
1046 ip
->i_d
.di_version
== 1) {
1047 ip
->i_d
.di_version
= 2;
1049 * We've already zeroed the old link count, the projid field,
1050 * and the pad field.
1055 * Project ids won't be stored on disk if we are using a version 1 inode.
1057 if ((prid
!= 0) && (ip
->i_d
.di_version
== 1))
1058 xfs_bump_ino_vers2(tp
, ip
);
1060 if (pip
&& XFS_INHERIT_GID(pip
)) {
1061 ip
->i_d
.di_gid
= pip
->i_d
.di_gid
;
1062 if ((pip
->i_d
.di_mode
& S_ISGID
) && (mode
& S_IFMT
) == S_IFDIR
) {
1063 ip
->i_d
.di_mode
|= S_ISGID
;
1068 * If the group ID of the new file does not match the effective group
1069 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1070 * (and only if the irix_sgid_inherit compatibility variable is set).
1072 if ((irix_sgid_inherit
) &&
1073 (ip
->i_d
.di_mode
& S_ISGID
) &&
1074 (!in_group_p((gid_t
)ip
->i_d
.di_gid
))) {
1075 ip
->i_d
.di_mode
&= ~S_ISGID
;
1078 ip
->i_d
.di_size
= 0;
1080 ip
->i_d
.di_nextents
= 0;
1081 ASSERT(ip
->i_d
.di_nblocks
== 0);
1084 ip
->i_d
.di_mtime
.t_sec
= (__int32_t
)tv
.tv_sec
;
1085 ip
->i_d
.di_mtime
.t_nsec
= (__int32_t
)tv
.tv_nsec
;
1086 ip
->i_d
.di_atime
= ip
->i_d
.di_mtime
;
1087 ip
->i_d
.di_ctime
= ip
->i_d
.di_mtime
;
1090 * di_gen will have been taken care of in xfs_iread.
1092 ip
->i_d
.di_extsize
= 0;
1093 ip
->i_d
.di_dmevmask
= 0;
1094 ip
->i_d
.di_dmstate
= 0;
1095 ip
->i_d
.di_flags
= 0;
1096 flags
= XFS_ILOG_CORE
;
1097 switch (mode
& S_IFMT
) {
1102 ip
->i_d
.di_format
= XFS_DINODE_FMT_DEV
;
1103 ip
->i_df
.if_u2
.if_rdev
= rdev
;
1104 ip
->i_df
.if_flags
= 0;
1105 flags
|= XFS_ILOG_DEV
;
1109 * we can't set up filestreams until after the VFS inode
1110 * is set up properly.
1112 if (pip
&& xfs_inode_is_filestream(pip
))
1116 if (pip
&& (pip
->i_d
.di_flags
& XFS_DIFLAG_ANY
)) {
1119 if ((mode
& S_IFMT
) == S_IFDIR
) {
1120 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
1121 di_flags
|= XFS_DIFLAG_RTINHERIT
;
1122 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
1123 di_flags
|= XFS_DIFLAG_EXTSZINHERIT
;
1124 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
1126 } else if ((mode
& S_IFMT
) == S_IFREG
) {
1127 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
1128 di_flags
|= XFS_DIFLAG_REALTIME
;
1129 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
1130 di_flags
|= XFS_DIFLAG_EXTSIZE
;
1131 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
1134 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOATIME
) &&
1135 xfs_inherit_noatime
)
1136 di_flags
|= XFS_DIFLAG_NOATIME
;
1137 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODUMP
) &&
1139 di_flags
|= XFS_DIFLAG_NODUMP
;
1140 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_SYNC
) &&
1142 di_flags
|= XFS_DIFLAG_SYNC
;
1143 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOSYMLINKS
) &&
1144 xfs_inherit_nosymlinks
)
1145 di_flags
|= XFS_DIFLAG_NOSYMLINKS
;
1146 if (pip
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
1147 di_flags
|= XFS_DIFLAG_PROJINHERIT
;
1148 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODEFRAG
) &&
1149 xfs_inherit_nodefrag
)
1150 di_flags
|= XFS_DIFLAG_NODEFRAG
;
1151 if (pip
->i_d
.di_flags
& XFS_DIFLAG_FILESTREAM
)
1152 di_flags
|= XFS_DIFLAG_FILESTREAM
;
1153 ip
->i_d
.di_flags
|= di_flags
;
1157 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
1158 ip
->i_df
.if_flags
= XFS_IFEXTENTS
;
1159 ip
->i_df
.if_bytes
= ip
->i_df
.if_real_bytes
= 0;
1160 ip
->i_df
.if_u1
.if_extents
= NULL
;
1166 * Attribute fork settings for new inode.
1168 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
1169 ip
->i_d
.di_anextents
= 0;
1172 * Log the new values stuffed into the inode.
1174 xfs_trans_log_inode(tp
, ip
, flags
);
1176 /* now that we have an i_mode we can setup inode ops and unlock */
1177 xfs_setup_inode(ip
);
1179 /* now we have set up the vfs inode we can associate the filestream */
1181 error
= xfs_filestream_associate(pip
, ip
);
1185 xfs_iflags_set(ip
, XFS_IFILESTREAM
);
1193 * Check to make sure that there are no blocks allocated to the
1194 * file beyond the size of the file. We don't check this for
1195 * files with fixed size extents or real time extents, but we
1196 * at least do it for regular files.
1205 xfs_fileoff_t map_first
;
1207 xfs_bmbt_irec_t imaps
[2];
1209 if ((ip
->i_d
.di_mode
& S_IFMT
) != S_IFREG
)
1212 if (XFS_IS_REALTIME_INODE(ip
))
1215 if (ip
->i_d
.di_flags
& XFS_DIFLAG_EXTSIZE
)
1219 map_first
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)isize
);
1221 * The filesystem could be shutting down, so bmapi may return
1224 if (xfs_bmapi(NULL
, ip
, map_first
,
1226 (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
)) -
1228 XFS_BMAPI_ENTIRE
, NULL
, 0, imaps
, &nimaps
,
1231 ASSERT(nimaps
== 1);
1232 ASSERT(imaps
[0].br_startblock
== HOLESTARTBLOCK
);
1237 * Calculate the last possible buffered byte in a file. This must
1238 * include data that was buffered beyond the EOF by the write code.
1239 * This also needs to deal with overflowing the xfs_fsize_t type
1240 * which can happen for sizes near the limit.
1242 * We also need to take into account any blocks beyond the EOF. It
1243 * may be the case that they were buffered by a write which failed.
1244 * In that case the pages will still be in memory, but the inode size
1245 * will never have been updated.
1252 xfs_fsize_t last_byte
;
1253 xfs_fileoff_t last_block
;
1254 xfs_fileoff_t size_last_block
;
1257 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
|XFS_IOLOCK_SHARED
));
1261 * Only check for blocks beyond the EOF if the extents have
1262 * been read in. This eliminates the need for the inode lock,
1263 * and it also saves us from looking when it really isn't
1266 if (ip
->i_df
.if_flags
& XFS_IFEXTENTS
) {
1267 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
1268 error
= xfs_bmap_last_offset(NULL
, ip
, &last_block
,
1270 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
1277 size_last_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)ip
->i_size
);
1278 last_block
= XFS_FILEOFF_MAX(last_block
, size_last_block
);
1280 last_byte
= XFS_FSB_TO_B(mp
, last_block
);
1281 if (last_byte
< 0) {
1282 return XFS_MAXIOFFSET(mp
);
1284 last_byte
+= (1 << mp
->m_writeio_log
);
1285 if (last_byte
< 0) {
1286 return XFS_MAXIOFFSET(mp
);
1291 #if defined(XFS_RW_TRACE)
1297 xfs_fsize_t new_size
,
1298 xfs_off_t toss_start
,
1299 xfs_off_t toss_finish
)
1301 if (ip
->i_rwtrace
== NULL
) {
1305 ktrace_enter(ip
->i_rwtrace
,
1308 (void*)(unsigned long)((ip
->i_d
.di_size
>> 32) & 0xffffffff),
1309 (void*)(unsigned long)(ip
->i_d
.di_size
& 0xffffffff),
1310 (void*)((long)flag
),
1311 (void*)(unsigned long)((new_size
>> 32) & 0xffffffff),
1312 (void*)(unsigned long)(new_size
& 0xffffffff),
1313 (void*)(unsigned long)((toss_start
>> 32) & 0xffffffff),
1314 (void*)(unsigned long)(toss_start
& 0xffffffff),
1315 (void*)(unsigned long)((toss_finish
>> 32) & 0xffffffff),
1316 (void*)(unsigned long)(toss_finish
& 0xffffffff),
1317 (void*)(unsigned long)current_cpu(),
1318 (void*)(unsigned long)current_pid(),
1324 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1328 * Start the truncation of the file to new_size. The new size
1329 * must be smaller than the current size. This routine will
1330 * clear the buffer and page caches of file data in the removed
1331 * range, and xfs_itruncate_finish() will remove the underlying
1334 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1335 * must NOT have the inode lock held at all. This is because we're
1336 * calling into the buffer/page cache code and we can't hold the
1337 * inode lock when we do so.
1339 * We need to wait for any direct I/Os in flight to complete before we
1340 * proceed with the truncate. This is needed to prevent the extents
1341 * being read or written by the direct I/Os from being removed while the
1342 * I/O is in flight as there is no other method of synchronising
1343 * direct I/O with the truncate operation. Also, because we hold
1344 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1345 * started until the truncate completes and drops the lock. Essentially,
1346 * the xfs_ioend_wait() call forms an I/O barrier that provides strict
1347 * ordering between direct I/Os and the truncate operation.
1349 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1350 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1351 * in the case that the caller is locking things out of order and
1352 * may not be able to call xfs_itruncate_finish() with the inode lock
1353 * held without dropping the I/O lock. If the caller must drop the
1354 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1355 * must be called again with all the same restrictions as the initial
1359 xfs_itruncate_start(
1362 xfs_fsize_t new_size
)
1364 xfs_fsize_t last_byte
;
1365 xfs_off_t toss_start
;
1369 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
1370 ASSERT((new_size
== 0) || (new_size
<= ip
->i_size
));
1371 ASSERT((flags
== XFS_ITRUNC_DEFINITE
) ||
1372 (flags
== XFS_ITRUNC_MAYBE
));
1376 /* wait for the completion of any pending DIOs */
1377 if (new_size
== 0 || new_size
< ip
->i_size
)
1381 * Call toss_pages or flushinval_pages to get rid of pages
1382 * overlapping the region being removed. We have to use
1383 * the less efficient flushinval_pages in the case that the
1384 * caller may not be able to finish the truncate without
1385 * dropping the inode's I/O lock. Make sure
1386 * to catch any pages brought in by buffers overlapping
1387 * the EOF by searching out beyond the isize by our
1388 * block size. We round new_size up to a block boundary
1389 * so that we don't toss things on the same block as
1390 * new_size but before it.
1392 * Before calling toss_page or flushinval_pages, make sure to
1393 * call remapf() over the same region if the file is mapped.
1394 * This frees up mapped file references to the pages in the
1395 * given range and for the flushinval_pages case it ensures
1396 * that we get the latest mapped changes flushed out.
1398 toss_start
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)new_size
);
1399 toss_start
= XFS_FSB_TO_B(mp
, toss_start
);
1400 if (toss_start
< 0) {
1402 * The place to start tossing is beyond our maximum
1403 * file size, so there is no way that the data extended
1408 last_byte
= xfs_file_last_byte(ip
);
1409 xfs_itrunc_trace(XFS_ITRUNC_START
, ip
, flags
, new_size
, toss_start
,
1411 if (last_byte
> toss_start
) {
1412 if (flags
& XFS_ITRUNC_DEFINITE
) {
1413 xfs_tosspages(ip
, toss_start
,
1414 -1, FI_REMAPF_LOCKED
);
1416 error
= xfs_flushinval_pages(ip
, toss_start
,
1417 -1, FI_REMAPF_LOCKED
);
1422 if (new_size
== 0) {
1423 ASSERT(VN_CACHED(VFS_I(ip
)) == 0);
1430 * Shrink the file to the given new_size. The new size must be smaller than
1431 * the current size. This will free up the underlying blocks in the removed
1432 * range after a call to xfs_itruncate_start() or xfs_atruncate_start().
1434 * The transaction passed to this routine must have made a permanent log
1435 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1436 * given transaction and start new ones, so make sure everything involved in
1437 * the transaction is tidy before calling here. Some transaction will be
1438 * returned to the caller to be committed. The incoming transaction must
1439 * already include the inode, and both inode locks must be held exclusively.
1440 * The inode must also be "held" within the transaction. On return the inode
1441 * will be "held" within the returned transaction. This routine does NOT
1442 * require any disk space to be reserved for it within the transaction.
1444 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it
1445 * indicates the fork which is to be truncated. For the attribute fork we only
1446 * support truncation to size 0.
1448 * We use the sync parameter to indicate whether or not the first transaction
1449 * we perform might have to be synchronous. For the attr fork, it needs to be
1450 * so if the unlink of the inode is not yet known to be permanent in the log.
1451 * This keeps us from freeing and reusing the blocks of the attribute fork
1452 * before the unlink of the inode becomes permanent.
1454 * For the data fork, we normally have to run synchronously if we're being
1455 * called out of the inactive path or we're being called out of the create path
1456 * where we're truncating an existing file. Either way, the truncate needs to
1457 * be sync so blocks don't reappear in the file with altered data in case of a
1458 * crash. wsync filesystems can run the first case async because anything that
1459 * shrinks the inode has to run sync so by the time we're called here from
1460 * inactive, the inode size is permanently set to 0.
1462 * Calls from the truncate path always need to be sync unless we're in a wsync
1463 * filesystem and the file has already been unlinked.
1465 * The caller is responsible for correctly setting the sync parameter. It gets
1466 * too hard for us to guess here which path we're being called out of just
1467 * based on inode state.
1469 * If we get an error, we must return with the inode locked and linked into the
1470 * current transaction. This keeps things simple for the higher level code,
1471 * because it always knows that the inode is locked and held in the transaction
1472 * that returns to it whether errors occur or not. We don't mark the inode
1473 * dirty on error so that transactions can be easily aborted if possible.
1476 xfs_itruncate_finish(
1479 xfs_fsize_t new_size
,
1483 xfs_fsblock_t first_block
;
1484 xfs_fileoff_t first_unmap_block
;
1485 xfs_fileoff_t last_block
;
1486 xfs_filblks_t unmap_len
=0;
1491 xfs_bmap_free_t free_list
;
1494 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_IOLOCK_EXCL
));
1495 ASSERT((new_size
== 0) || (new_size
<= ip
->i_size
));
1496 ASSERT(*tp
!= NULL
);
1497 ASSERT((*tp
)->t_flags
& XFS_TRANS_PERM_LOG_RES
);
1498 ASSERT(ip
->i_transp
== *tp
);
1499 ASSERT(ip
->i_itemp
!= NULL
);
1500 ASSERT(ip
->i_itemp
->ili_flags
& XFS_ILI_HOLD
);
1504 mp
= (ntp
)->t_mountp
;
1505 ASSERT(! XFS_NOT_DQATTACHED(mp
, ip
));
1508 * We only support truncating the entire attribute fork.
1510 if (fork
== XFS_ATTR_FORK
) {
1513 first_unmap_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)new_size
);
1514 xfs_itrunc_trace(XFS_ITRUNC_FINISH1
, ip
, 0, new_size
, 0, 0);
1516 * The first thing we do is set the size to new_size permanently
1517 * on disk. This way we don't have to worry about anyone ever
1518 * being able to look at the data being freed even in the face
1519 * of a crash. What we're getting around here is the case where
1520 * we free a block, it is allocated to another file, it is written
1521 * to, and then we crash. If the new data gets written to the
1522 * file but the log buffers containing the free and reallocation
1523 * don't, then we'd end up with garbage in the blocks being freed.
1524 * As long as we make the new_size permanent before actually
1525 * freeing any blocks it doesn't matter if they get writtten to.
1527 * The callers must signal into us whether or not the size
1528 * setting here must be synchronous. There are a few cases
1529 * where it doesn't have to be synchronous. Those cases
1530 * occur if the file is unlinked and we know the unlink is
1531 * permanent or if the blocks being truncated are guaranteed
1532 * to be beyond the inode eof (regardless of the link count)
1533 * and the eof value is permanent. Both of these cases occur
1534 * only on wsync-mounted filesystems. In those cases, we're
1535 * guaranteed that no user will ever see the data in the blocks
1536 * that are being truncated so the truncate can run async.
1537 * In the free beyond eof case, the file may wind up with
1538 * more blocks allocated to it than it needs if we crash
1539 * and that won't get fixed until the next time the file
1540 * is re-opened and closed but that's ok as that shouldn't
1541 * be too many blocks.
1543 * However, we can't just make all wsync xactions run async
1544 * because there's one call out of the create path that needs
1545 * to run sync where it's truncating an existing file to size
1546 * 0 whose size is > 0.
1548 * It's probably possible to come up with a test in this
1549 * routine that would correctly distinguish all the above
1550 * cases from the values of the function parameters and the
1551 * inode state but for sanity's sake, I've decided to let the
1552 * layers above just tell us. It's simpler to correctly figure
1553 * out in the layer above exactly under what conditions we
1554 * can run async and I think it's easier for others read and
1555 * follow the logic in case something has to be changed.
1556 * cscope is your friend -- rcc.
1558 * The attribute fork is much simpler.
1560 * For the attribute fork we allow the caller to tell us whether
1561 * the unlink of the inode that led to this call is yet permanent
1562 * in the on disk log. If it is not and we will be freeing extents
1563 * in this inode then we make the first transaction synchronous
1564 * to make sure that the unlink is permanent by the time we free
1567 if (fork
== XFS_DATA_FORK
) {
1568 if (ip
->i_d
.di_nextents
> 0) {
1570 * If we are not changing the file size then do
1571 * not update the on-disk file size - we may be
1572 * called from xfs_inactive_free_eofblocks(). If we
1573 * update the on-disk file size and then the system
1574 * crashes before the contents of the file are
1575 * flushed to disk then the files may be full of
1576 * holes (ie NULL files bug).
1578 if (ip
->i_size
!= new_size
) {
1579 ip
->i_d
.di_size
= new_size
;
1580 ip
->i_size
= new_size
;
1581 xfs_trans_log_inode(ntp
, ip
, XFS_ILOG_CORE
);
1585 ASSERT(!(mp
->m_flags
& XFS_MOUNT_WSYNC
));
1586 if (ip
->i_d
.di_anextents
> 0)
1587 xfs_trans_set_sync(ntp
);
1589 ASSERT(fork
== XFS_DATA_FORK
||
1590 (fork
== XFS_ATTR_FORK
&&
1591 ((sync
&& !(mp
->m_flags
& XFS_MOUNT_WSYNC
)) ||
1592 (sync
== 0 && (mp
->m_flags
& XFS_MOUNT_WSYNC
)))));
1595 * Since it is possible for space to become allocated beyond
1596 * the end of the file (in a crash where the space is allocated
1597 * but the inode size is not yet updated), simply remove any
1598 * blocks which show up between the new EOF and the maximum
1599 * possible file size. If the first block to be removed is
1600 * beyond the maximum file size (ie it is the same as last_block),
1601 * then there is nothing to do.
1603 last_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
));
1604 ASSERT(first_unmap_block
<= last_block
);
1606 if (last_block
== first_unmap_block
) {
1609 unmap_len
= last_block
- first_unmap_block
+ 1;
1613 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1614 * will tell us whether it freed the entire range or
1615 * not. If this is a synchronous mount (wsync),
1616 * then we can tell bunmapi to keep all the
1617 * transactions asynchronous since the unlink
1618 * transaction that made this inode inactive has
1619 * already hit the disk. There's no danger of
1620 * the freed blocks being reused, there being a
1621 * crash, and the reused blocks suddenly reappearing
1622 * in this file with garbage in them once recovery
1625 xfs_bmap_init(&free_list
, &first_block
);
1626 error
= xfs_bunmapi(ntp
, ip
,
1627 first_unmap_block
, unmap_len
,
1628 xfs_bmapi_aflag(fork
) |
1629 (sync
? 0 : XFS_BMAPI_ASYNC
),
1630 XFS_ITRUNC_MAX_EXTENTS
,
1631 &first_block
, &free_list
,
1635 * If the bunmapi call encounters an error,
1636 * return to the caller where the transaction
1637 * can be properly aborted. We just need to
1638 * make sure we're not holding any resources
1639 * that we were not when we came in.
1641 xfs_bmap_cancel(&free_list
);
1646 * Duplicate the transaction that has the permanent
1647 * reservation and commit the old transaction.
1649 error
= xfs_bmap_finish(tp
, &free_list
, &committed
);
1652 /* link the inode into the next xact in the chain */
1653 xfs_trans_ijoin(ntp
, ip
,
1654 XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1655 xfs_trans_ihold(ntp
, ip
);
1660 * If the bmap finish call encounters an error, return
1661 * to the caller where the transaction can be properly
1662 * aborted. We just need to make sure we're not
1663 * holding any resources that we were not when we came
1666 * Aborting from this point might lose some blocks in
1667 * the file system, but oh well.
1669 xfs_bmap_cancel(&free_list
);
1675 * Mark the inode dirty so it will be logged and
1676 * moved forward in the log as part of every commit.
1678 xfs_trans_log_inode(ntp
, ip
, XFS_ILOG_CORE
);
1681 ntp
= xfs_trans_dup(ntp
);
1682 error
= xfs_trans_commit(*tp
, 0);
1685 /* link the inode into the next transaction in the chain */
1686 xfs_trans_ijoin(ntp
, ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1687 xfs_trans_ihold(ntp
, ip
);
1692 * transaction commit worked ok so we can drop the extra ticket
1693 * reference that we gained in xfs_trans_dup()
1695 xfs_log_ticket_put(ntp
->t_ticket
);
1696 error
= xfs_trans_reserve(ntp
, 0,
1697 XFS_ITRUNCATE_LOG_RES(mp
), 0,
1698 XFS_TRANS_PERM_LOG_RES
,
1699 XFS_ITRUNCATE_LOG_COUNT
);
1704 * Only update the size in the case of the data fork, but
1705 * always re-log the inode so that our permanent transaction
1706 * can keep on rolling it forward in the log.
1708 if (fork
== XFS_DATA_FORK
) {
1709 xfs_isize_check(mp
, ip
, new_size
);
1711 * If we are not changing the file size then do
1712 * not update the on-disk file size - we may be
1713 * called from xfs_inactive_free_eofblocks(). If we
1714 * update the on-disk file size and then the system
1715 * crashes before the contents of the file are
1716 * flushed to disk then the files may be full of
1717 * holes (ie NULL files bug).
1719 if (ip
->i_size
!= new_size
) {
1720 ip
->i_d
.di_size
= new_size
;
1721 ip
->i_size
= new_size
;
1724 xfs_trans_log_inode(ntp
, ip
, XFS_ILOG_CORE
);
1725 ASSERT((new_size
!= 0) ||
1726 (fork
== XFS_ATTR_FORK
) ||
1727 (ip
->i_delayed_blks
== 0));
1728 ASSERT((new_size
!= 0) ||
1729 (fork
== XFS_ATTR_FORK
) ||
1730 (ip
->i_d
.di_nextents
== 0));
1731 xfs_itrunc_trace(XFS_ITRUNC_FINISH2
, ip
, 0, new_size
, 0, 0);
1736 * This is called when the inode's link count goes to 0.
1737 * We place the on-disk inode on a list in the AGI. It
1738 * will be pulled from this list when the inode is freed.
1755 ASSERT(ip
->i_d
.di_nlink
== 0);
1756 ASSERT(ip
->i_d
.di_mode
!= 0);
1757 ASSERT(ip
->i_transp
== tp
);
1762 * Get the agi buffer first. It ensures lock ordering
1765 error
= xfs_read_agi(mp
, tp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
), &agibp
);
1768 agi
= XFS_BUF_TO_AGI(agibp
);
1771 * Get the index into the agi hash table for the
1772 * list this inode will go on.
1774 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
1776 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
1777 ASSERT(agi
->agi_unlinked
[bucket_index
]);
1778 ASSERT(be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != agino
);
1780 if (be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != NULLAGINO
) {
1782 * There is already another inode in the bucket we need
1783 * to add ourselves to. Add us at the front of the list.
1784 * Here we put the head pointer into our next pointer,
1785 * and then we fall through to point the head at us.
1787 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, XFS_BUF_LOCK
);
1791 ASSERT(be32_to_cpu(dip
->di_next_unlinked
) == NULLAGINO
);
1792 /* both on-disk, don't endian flip twice */
1793 dip
->di_next_unlinked
= agi
->agi_unlinked
[bucket_index
];
1794 offset
= ip
->i_imap
.im_boffset
+
1795 offsetof(xfs_dinode_t
, di_next_unlinked
);
1796 xfs_trans_inode_buf(tp
, ibp
);
1797 xfs_trans_log_buf(tp
, ibp
, offset
,
1798 (offset
+ sizeof(xfs_agino_t
) - 1));
1799 xfs_inobp_check(mp
, ibp
);
1803 * Point the bucket head pointer at the inode being inserted.
1806 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(agino
);
1807 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
1808 (sizeof(xfs_agino_t
) * bucket_index
);
1809 xfs_trans_log_buf(tp
, agibp
, offset
,
1810 (offset
+ sizeof(xfs_agino_t
) - 1));
1815 * Pull the on-disk inode from the AGI unlinked list.
1828 xfs_agnumber_t agno
;
1830 xfs_agino_t next_agino
;
1831 xfs_buf_t
*last_ibp
;
1832 xfs_dinode_t
*last_dip
= NULL
;
1834 int offset
, last_offset
= 0;
1838 agno
= XFS_INO_TO_AGNO(mp
, ip
->i_ino
);
1841 * Get the agi buffer first. It ensures lock ordering
1844 error
= xfs_read_agi(mp
, tp
, agno
, &agibp
);
1848 agi
= XFS_BUF_TO_AGI(agibp
);
1851 * Get the index into the agi hash table for the
1852 * list this inode will go on.
1854 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
1856 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
1857 ASSERT(be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != NULLAGINO
);
1858 ASSERT(agi
->agi_unlinked
[bucket_index
]);
1860 if (be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) == agino
) {
1862 * We're at the head of the list. Get the inode's
1863 * on-disk buffer to see if there is anyone after us
1864 * on the list. Only modify our next pointer if it
1865 * is not already NULLAGINO. This saves us the overhead
1866 * of dealing with the buffer when there is no need to
1869 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, XFS_BUF_LOCK
);
1872 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
1873 error
, mp
->m_fsname
);
1876 next_agino
= be32_to_cpu(dip
->di_next_unlinked
);
1877 ASSERT(next_agino
!= 0);
1878 if (next_agino
!= NULLAGINO
) {
1879 dip
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
1880 offset
= ip
->i_imap
.im_boffset
+
1881 offsetof(xfs_dinode_t
, di_next_unlinked
);
1882 xfs_trans_inode_buf(tp
, ibp
);
1883 xfs_trans_log_buf(tp
, ibp
, offset
,
1884 (offset
+ sizeof(xfs_agino_t
) - 1));
1885 xfs_inobp_check(mp
, ibp
);
1887 xfs_trans_brelse(tp
, ibp
);
1890 * Point the bucket head pointer at the next inode.
1892 ASSERT(next_agino
!= 0);
1893 ASSERT(next_agino
!= agino
);
1894 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(next_agino
);
1895 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
1896 (sizeof(xfs_agino_t
) * bucket_index
);
1897 xfs_trans_log_buf(tp
, agibp
, offset
,
1898 (offset
+ sizeof(xfs_agino_t
) - 1));
1901 * We need to search the list for the inode being freed.
1903 next_agino
= be32_to_cpu(agi
->agi_unlinked
[bucket_index
]);
1905 while (next_agino
!= agino
) {
1907 * If the last inode wasn't the one pointing to
1908 * us, then release its buffer since we're not
1909 * going to do anything with it.
1911 if (last_ibp
!= NULL
) {
1912 xfs_trans_brelse(tp
, last_ibp
);
1914 next_ino
= XFS_AGINO_TO_INO(mp
, agno
, next_agino
);
1915 error
= xfs_inotobp(mp
, tp
, next_ino
, &last_dip
,
1916 &last_ibp
, &last_offset
, 0);
1919 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
1920 error
, mp
->m_fsname
);
1923 next_agino
= be32_to_cpu(last_dip
->di_next_unlinked
);
1924 ASSERT(next_agino
!= NULLAGINO
);
1925 ASSERT(next_agino
!= 0);
1928 * Now last_ibp points to the buffer previous to us on
1929 * the unlinked list. Pull us from the list.
1931 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, XFS_BUF_LOCK
);
1934 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
1935 error
, mp
->m_fsname
);
1938 next_agino
= be32_to_cpu(dip
->di_next_unlinked
);
1939 ASSERT(next_agino
!= 0);
1940 ASSERT(next_agino
!= agino
);
1941 if (next_agino
!= NULLAGINO
) {
1942 dip
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
1943 offset
= ip
->i_imap
.im_boffset
+
1944 offsetof(xfs_dinode_t
, di_next_unlinked
);
1945 xfs_trans_inode_buf(tp
, ibp
);
1946 xfs_trans_log_buf(tp
, ibp
, offset
,
1947 (offset
+ sizeof(xfs_agino_t
) - 1));
1948 xfs_inobp_check(mp
, ibp
);
1950 xfs_trans_brelse(tp
, ibp
);
1953 * Point the previous inode on the list to the next inode.
1955 last_dip
->di_next_unlinked
= cpu_to_be32(next_agino
);
1956 ASSERT(next_agino
!= 0);
1957 offset
= last_offset
+ offsetof(xfs_dinode_t
, di_next_unlinked
);
1958 xfs_trans_inode_buf(tp
, last_ibp
);
1959 xfs_trans_log_buf(tp
, last_ibp
, offset
,
1960 (offset
+ sizeof(xfs_agino_t
) - 1));
1961 xfs_inobp_check(mp
, last_ibp
);
1968 xfs_inode_t
*free_ip
,
1972 xfs_mount_t
*mp
= free_ip
->i_mount
;
1973 int blks_per_cluster
;
1976 int i
, j
, found
, pre_flushed
;
1979 xfs_inode_t
*ip
, **ip_found
;
1980 xfs_inode_log_item_t
*iip
;
1981 xfs_log_item_t
*lip
;
1982 xfs_perag_t
*pag
= xfs_get_perag(mp
, inum
);
1984 if (mp
->m_sb
.sb_blocksize
>= XFS_INODE_CLUSTER_SIZE(mp
)) {
1985 blks_per_cluster
= 1;
1986 ninodes
= mp
->m_sb
.sb_inopblock
;
1987 nbufs
= XFS_IALLOC_BLOCKS(mp
);
1989 blks_per_cluster
= XFS_INODE_CLUSTER_SIZE(mp
) /
1990 mp
->m_sb
.sb_blocksize
;
1991 ninodes
= blks_per_cluster
* mp
->m_sb
.sb_inopblock
;
1992 nbufs
= XFS_IALLOC_BLOCKS(mp
) / blks_per_cluster
;
1995 ip_found
= kmem_alloc(ninodes
* sizeof(xfs_inode_t
*), KM_NOFS
);
1997 for (j
= 0; j
< nbufs
; j
++, inum
+= ninodes
) {
1998 blkno
= XFS_AGB_TO_DADDR(mp
, XFS_INO_TO_AGNO(mp
, inum
),
1999 XFS_INO_TO_AGBNO(mp
, inum
));
2003 * Look for each inode in memory and attempt to lock it,
2004 * we can be racing with flush and tail pushing here.
2005 * any inode we get the locks on, add to an array of
2006 * inode items to process later.
2008 * The get the buffer lock, we could beat a flush
2009 * or tail pushing thread to the lock here, in which
2010 * case they will go looking for the inode buffer
2011 * and fail, we need some other form of interlock
2015 for (i
= 0; i
< ninodes
; i
++) {
2016 read_lock(&pag
->pag_ici_lock
);
2017 ip
= radix_tree_lookup(&pag
->pag_ici_root
,
2018 XFS_INO_TO_AGINO(mp
, (inum
+ i
)));
2020 /* Inode not in memory or we found it already,
2023 if (!ip
|| xfs_iflags_test(ip
, XFS_ISTALE
)) {
2024 read_unlock(&pag
->pag_ici_lock
);
2028 if (xfs_inode_clean(ip
)) {
2029 read_unlock(&pag
->pag_ici_lock
);
2033 /* If we can get the locks then add it to the
2034 * list, otherwise by the time we get the bp lock
2035 * below it will already be attached to the
2039 /* This inode will already be locked - by us, lets
2043 if (ip
== free_ip
) {
2044 if (xfs_iflock_nowait(ip
)) {
2045 xfs_iflags_set(ip
, XFS_ISTALE
);
2046 if (xfs_inode_clean(ip
)) {
2049 ip_found
[found
++] = ip
;
2052 read_unlock(&pag
->pag_ici_lock
);
2056 if (xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
)) {
2057 if (xfs_iflock_nowait(ip
)) {
2058 xfs_iflags_set(ip
, XFS_ISTALE
);
2060 if (xfs_inode_clean(ip
)) {
2062 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2064 ip_found
[found
++] = ip
;
2067 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2070 read_unlock(&pag
->pag_ici_lock
);
2073 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, blkno
,
2074 mp
->m_bsize
* blks_per_cluster
,
2078 lip
= XFS_BUF_FSPRIVATE(bp
, xfs_log_item_t
*);
2080 if (lip
->li_type
== XFS_LI_INODE
) {
2081 iip
= (xfs_inode_log_item_t
*)lip
;
2082 ASSERT(iip
->ili_logged
== 1);
2083 lip
->li_cb
= (void(*)(xfs_buf_t
*,xfs_log_item_t
*)) xfs_istale_done
;
2084 xfs_trans_ail_copy_lsn(mp
->m_ail
,
2085 &iip
->ili_flush_lsn
,
2086 &iip
->ili_item
.li_lsn
);
2087 xfs_iflags_set(iip
->ili_inode
, XFS_ISTALE
);
2090 lip
= lip
->li_bio_list
;
2093 for (i
= 0; i
< found
; i
++) {
2098 ip
->i_update_core
= 0;
2100 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2104 iip
->ili_last_fields
= iip
->ili_format
.ilf_fields
;
2105 iip
->ili_format
.ilf_fields
= 0;
2106 iip
->ili_logged
= 1;
2107 xfs_trans_ail_copy_lsn(mp
->m_ail
, &iip
->ili_flush_lsn
,
2108 &iip
->ili_item
.li_lsn
);
2110 xfs_buf_attach_iodone(bp
,
2111 (void(*)(xfs_buf_t
*,xfs_log_item_t
*))
2112 xfs_istale_done
, (xfs_log_item_t
*)iip
);
2113 if (ip
!= free_ip
) {
2114 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2118 if (found
|| pre_flushed
)
2119 xfs_trans_stale_inode_buf(tp
, bp
);
2120 xfs_trans_binval(tp
, bp
);
2123 kmem_free(ip_found
);
2124 xfs_put_perag(mp
, pag
);
2128 * This is called to return an inode to the inode free list.
2129 * The inode should already be truncated to 0 length and have
2130 * no pages associated with it. This routine also assumes that
2131 * the inode is already a part of the transaction.
2133 * The on-disk copy of the inode will have been added to the list
2134 * of unlinked inodes in the AGI. We need to remove the inode from
2135 * that list atomically with respect to freeing it here.
2141 xfs_bmap_free_t
*flist
)
2145 xfs_ino_t first_ino
;
2149 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
2150 ASSERT(ip
->i_transp
== tp
);
2151 ASSERT(ip
->i_d
.di_nlink
== 0);
2152 ASSERT(ip
->i_d
.di_nextents
== 0);
2153 ASSERT(ip
->i_d
.di_anextents
== 0);
2154 ASSERT((ip
->i_d
.di_size
== 0 && ip
->i_size
== 0) ||
2155 ((ip
->i_d
.di_mode
& S_IFMT
) != S_IFREG
));
2156 ASSERT(ip
->i_d
.di_nblocks
== 0);
2159 * Pull the on-disk inode from the AGI unlinked list.
2161 error
= xfs_iunlink_remove(tp
, ip
);
2166 error
= xfs_difree(tp
, ip
->i_ino
, flist
, &delete, &first_ino
);
2170 ip
->i_d
.di_mode
= 0; /* mark incore inode as free */
2171 ip
->i_d
.di_flags
= 0;
2172 ip
->i_d
.di_dmevmask
= 0;
2173 ip
->i_d
.di_forkoff
= 0; /* mark the attr fork not in use */
2174 ip
->i_df
.if_ext_max
=
2175 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
2176 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
2177 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
2179 * Bump the generation count so no one will be confused
2180 * by reincarnations of this inode.
2184 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2186 error
= xfs_itobp(ip
->i_mount
, tp
, ip
, &dip
, &ibp
, XFS_BUF_LOCK
);
2191 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2192 * from picking up this inode when it is reclaimed (its incore state
2193 * initialzed but not flushed to disk yet). The in-core di_mode is
2194 * already cleared and a corresponding transaction logged.
2195 * The hack here just synchronizes the in-core to on-disk
2196 * di_mode value in advance before the actual inode sync to disk.
2197 * This is OK because the inode is already unlinked and would never
2198 * change its di_mode again for this inode generation.
2199 * This is a temporary hack that would require a proper fix
2205 xfs_ifree_cluster(ip
, tp
, first_ino
);
2212 * Reallocate the space for if_broot based on the number of records
2213 * being added or deleted as indicated in rec_diff. Move the records
2214 * and pointers in if_broot to fit the new size. When shrinking this
2215 * will eliminate holes between the records and pointers created by
2216 * the caller. When growing this will create holes to be filled in
2219 * The caller must not request to add more records than would fit in
2220 * the on-disk inode root. If the if_broot is currently NULL, then
2221 * if we adding records one will be allocated. The caller must also
2222 * not request that the number of records go below zero, although
2223 * it can go to zero.
2225 * ip -- the inode whose if_broot area is changing
2226 * ext_diff -- the change in the number of records, positive or negative,
2227 * requested for the if_broot array.
2235 struct xfs_mount
*mp
= ip
->i_mount
;
2238 struct xfs_btree_block
*new_broot
;
2245 * Handle the degenerate case quietly.
2247 if (rec_diff
== 0) {
2251 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2254 * If there wasn't any memory allocated before, just
2255 * allocate it now and get out.
2257 if (ifp
->if_broot_bytes
== 0) {
2258 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff
);
2259 ifp
->if_broot
= kmem_alloc(new_size
, KM_SLEEP
);
2260 ifp
->if_broot_bytes
= (int)new_size
;
2265 * If there is already an existing if_broot, then we need
2266 * to realloc() it and shift the pointers to their new
2267 * location. The records don't change location because
2268 * they are kept butted up against the btree block header.
2270 cur_max
= xfs_bmbt_maxrecs(mp
, ifp
->if_broot_bytes
, 0);
2271 new_max
= cur_max
+ rec_diff
;
2272 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max
);
2273 ifp
->if_broot
= kmem_realloc(ifp
->if_broot
, new_size
,
2274 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max
), /* old size */
2276 op
= (char *)XFS_BMAP_BROOT_PTR_ADDR(mp
, ifp
->if_broot
, 1,
2277 ifp
->if_broot_bytes
);
2278 np
= (char *)XFS_BMAP_BROOT_PTR_ADDR(mp
, ifp
->if_broot
, 1,
2280 ifp
->if_broot_bytes
= (int)new_size
;
2281 ASSERT(ifp
->if_broot_bytes
<=
2282 XFS_IFORK_SIZE(ip
, whichfork
) + XFS_BROOT_SIZE_ADJ
);
2283 memmove(np
, op
, cur_max
* (uint
)sizeof(xfs_dfsbno_t
));
2288 * rec_diff is less than 0. In this case, we are shrinking the
2289 * if_broot buffer. It must already exist. If we go to zero
2290 * records, just get rid of the root and clear the status bit.
2292 ASSERT((ifp
->if_broot
!= NULL
) && (ifp
->if_broot_bytes
> 0));
2293 cur_max
= xfs_bmbt_maxrecs(mp
, ifp
->if_broot_bytes
, 0);
2294 new_max
= cur_max
+ rec_diff
;
2295 ASSERT(new_max
>= 0);
2297 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max
);
2301 new_broot
= kmem_alloc(new_size
, KM_SLEEP
);
2303 * First copy over the btree block header.
2305 memcpy(new_broot
, ifp
->if_broot
, XFS_BTREE_LBLOCK_LEN
);
2308 ifp
->if_flags
&= ~XFS_IFBROOT
;
2312 * Only copy the records and pointers if there are any.
2316 * First copy the records.
2318 op
= (char *)XFS_BMBT_REC_ADDR(mp
, ifp
->if_broot
, 1);
2319 np
= (char *)XFS_BMBT_REC_ADDR(mp
, new_broot
, 1);
2320 memcpy(np
, op
, new_max
* (uint
)sizeof(xfs_bmbt_rec_t
));
2323 * Then copy the pointers.
2325 op
= (char *)XFS_BMAP_BROOT_PTR_ADDR(mp
, ifp
->if_broot
, 1,
2326 ifp
->if_broot_bytes
);
2327 np
= (char *)XFS_BMAP_BROOT_PTR_ADDR(mp
, new_broot
, 1,
2329 memcpy(np
, op
, new_max
* (uint
)sizeof(xfs_dfsbno_t
));
2331 kmem_free(ifp
->if_broot
);
2332 ifp
->if_broot
= new_broot
;
2333 ifp
->if_broot_bytes
= (int)new_size
;
2334 ASSERT(ifp
->if_broot_bytes
<=
2335 XFS_IFORK_SIZE(ip
, whichfork
) + XFS_BROOT_SIZE_ADJ
);
2341 * This is called when the amount of space needed for if_data
2342 * is increased or decreased. The change in size is indicated by
2343 * the number of bytes that need to be added or deleted in the
2344 * byte_diff parameter.
2346 * If the amount of space needed has decreased below the size of the
2347 * inline buffer, then switch to using the inline buffer. Otherwise,
2348 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2349 * to what is needed.
2351 * ip -- the inode whose if_data area is changing
2352 * byte_diff -- the change in the number of bytes, positive or negative,
2353 * requested for the if_data array.
2365 if (byte_diff
== 0) {
2369 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2370 new_size
= (int)ifp
->if_bytes
+ byte_diff
;
2371 ASSERT(new_size
>= 0);
2373 if (new_size
== 0) {
2374 if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
2375 kmem_free(ifp
->if_u1
.if_data
);
2377 ifp
->if_u1
.if_data
= NULL
;
2379 } else if (new_size
<= sizeof(ifp
->if_u2
.if_inline_data
)) {
2381 * If the valid extents/data can fit in if_inline_ext/data,
2382 * copy them from the malloc'd vector and free it.
2384 if (ifp
->if_u1
.if_data
== NULL
) {
2385 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
2386 } else if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
2387 ASSERT(ifp
->if_real_bytes
!= 0);
2388 memcpy(ifp
->if_u2
.if_inline_data
, ifp
->if_u1
.if_data
,
2390 kmem_free(ifp
->if_u1
.if_data
);
2391 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
2396 * Stuck with malloc/realloc.
2397 * For inline data, the underlying buffer must be
2398 * a multiple of 4 bytes in size so that it can be
2399 * logged and stay on word boundaries. We enforce
2402 real_size
= roundup(new_size
, 4);
2403 if (ifp
->if_u1
.if_data
== NULL
) {
2404 ASSERT(ifp
->if_real_bytes
== 0);
2405 ifp
->if_u1
.if_data
= kmem_alloc(real_size
, KM_SLEEP
);
2406 } else if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
2408 * Only do the realloc if the underlying size
2409 * is really changing.
2411 if (ifp
->if_real_bytes
!= real_size
) {
2412 ifp
->if_u1
.if_data
=
2413 kmem_realloc(ifp
->if_u1
.if_data
,
2419 ASSERT(ifp
->if_real_bytes
== 0);
2420 ifp
->if_u1
.if_data
= kmem_alloc(real_size
, KM_SLEEP
);
2421 memcpy(ifp
->if_u1
.if_data
, ifp
->if_u2
.if_inline_data
,
2425 ifp
->if_real_bytes
= real_size
;
2426 ifp
->if_bytes
= new_size
;
2427 ASSERT(ifp
->if_bytes
<= XFS_IFORK_SIZE(ip
, whichfork
));
2437 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2438 if (ifp
->if_broot
!= NULL
) {
2439 kmem_free(ifp
->if_broot
);
2440 ifp
->if_broot
= NULL
;
2444 * If the format is local, then we can't have an extents
2445 * array so just look for an inline data array. If we're
2446 * not local then we may or may not have an extents list,
2447 * so check and free it up if we do.
2449 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
) {
2450 if ((ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) &&
2451 (ifp
->if_u1
.if_data
!= NULL
)) {
2452 ASSERT(ifp
->if_real_bytes
!= 0);
2453 kmem_free(ifp
->if_u1
.if_data
);
2454 ifp
->if_u1
.if_data
= NULL
;
2455 ifp
->if_real_bytes
= 0;
2457 } else if ((ifp
->if_flags
& XFS_IFEXTENTS
) &&
2458 ((ifp
->if_flags
& XFS_IFEXTIREC
) ||
2459 ((ifp
->if_u1
.if_extents
!= NULL
) &&
2460 (ifp
->if_u1
.if_extents
!= ifp
->if_u2
.if_inline_ext
)))) {
2461 ASSERT(ifp
->if_real_bytes
!= 0);
2462 xfs_iext_destroy(ifp
);
2464 ASSERT(ifp
->if_u1
.if_extents
== NULL
||
2465 ifp
->if_u1
.if_extents
== ifp
->if_u2
.if_inline_ext
);
2466 ASSERT(ifp
->if_real_bytes
== 0);
2467 if (whichfork
== XFS_ATTR_FORK
) {
2468 kmem_zone_free(xfs_ifork_zone
, ip
->i_afp
);
2474 * Increment the pin count of the given buffer.
2475 * This value is protected by ipinlock spinlock in the mount structure.
2481 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
2483 atomic_inc(&ip
->i_pincount
);
2487 * Decrement the pin count of the given inode, and wake up
2488 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2489 * inode must have been previously pinned with a call to xfs_ipin().
2495 ASSERT(atomic_read(&ip
->i_pincount
) > 0);
2497 if (atomic_dec_and_test(&ip
->i_pincount
))
2498 wake_up(&ip
->i_ipin_wait
);
2502 * This is called to unpin an inode. It can be directed to wait or to return
2503 * immediately without waiting for the inode to be unpinned. The caller must
2504 * have the inode locked in at least shared mode so that the buffer cannot be
2505 * subsequently pinned once someone is waiting for it to be unpinned.
2512 xfs_inode_log_item_t
*iip
= ip
->i_itemp
;
2514 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2515 if (atomic_read(&ip
->i_pincount
) == 0)
2518 /* Give the log a push to start the unpinning I/O */
2519 xfs_log_force(ip
->i_mount
, (iip
&& iip
->ili_last_lsn
) ?
2520 iip
->ili_last_lsn
: 0, XFS_LOG_FORCE
);
2522 wait_event(ip
->i_ipin_wait
, (atomic_read(&ip
->i_pincount
) == 0));
2529 __xfs_iunpin_wait(ip
, 1);
2536 __xfs_iunpin_wait(ip
, 0);
2541 * xfs_iextents_copy()
2543 * This is called to copy the REAL extents (as opposed to the delayed
2544 * allocation extents) from the inode into the given buffer. It
2545 * returns the number of bytes copied into the buffer.
2547 * If there are no delayed allocation extents, then we can just
2548 * memcpy() the extents into the buffer. Otherwise, we need to
2549 * examine each extent in turn and skip those which are delayed.
2561 xfs_fsblock_t start_block
;
2563 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2564 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2565 ASSERT(ifp
->if_bytes
> 0);
2567 nrecs
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
2568 XFS_BMAP_TRACE_EXLIST(ip
, nrecs
, whichfork
);
2572 * There are some delayed allocation extents in the
2573 * inode, so copy the extents one at a time and skip
2574 * the delayed ones. There must be at least one
2575 * non-delayed extent.
2578 for (i
= 0; i
< nrecs
; i
++) {
2579 xfs_bmbt_rec_host_t
*ep
= xfs_iext_get_ext(ifp
, i
);
2580 start_block
= xfs_bmbt_get_startblock(ep
);
2581 if (isnullstartblock(start_block
)) {
2583 * It's a delayed allocation extent, so skip it.
2588 /* Translate to on disk format */
2589 put_unaligned(cpu_to_be64(ep
->l0
), &dp
->l0
);
2590 put_unaligned(cpu_to_be64(ep
->l1
), &dp
->l1
);
2594 ASSERT(copied
!= 0);
2595 xfs_validate_extents(ifp
, copied
, XFS_EXTFMT_INODE(ip
));
2597 return (copied
* (uint
)sizeof(xfs_bmbt_rec_t
));
2601 * Each of the following cases stores data into the same region
2602 * of the on-disk inode, so only one of them can be valid at
2603 * any given time. While it is possible to have conflicting formats
2604 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2605 * in EXTENTS format, this can only happen when the fork has
2606 * changed formats after being modified but before being flushed.
2607 * In these cases, the format always takes precedence, because the
2608 * format indicates the current state of the fork.
2615 xfs_inode_log_item_t
*iip
,
2622 #ifdef XFS_TRANS_DEBUG
2625 static const short brootflag
[2] =
2626 { XFS_ILOG_DBROOT
, XFS_ILOG_ABROOT
};
2627 static const short dataflag
[2] =
2628 { XFS_ILOG_DDATA
, XFS_ILOG_ADATA
};
2629 static const short extflag
[2] =
2630 { XFS_ILOG_DEXT
, XFS_ILOG_AEXT
};
2634 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2636 * This can happen if we gave up in iformat in an error path,
2637 * for the attribute fork.
2640 ASSERT(whichfork
== XFS_ATTR_FORK
);
2643 cp
= XFS_DFORK_PTR(dip
, whichfork
);
2645 switch (XFS_IFORK_FORMAT(ip
, whichfork
)) {
2646 case XFS_DINODE_FMT_LOCAL
:
2647 if ((iip
->ili_format
.ilf_fields
& dataflag
[whichfork
]) &&
2648 (ifp
->if_bytes
> 0)) {
2649 ASSERT(ifp
->if_u1
.if_data
!= NULL
);
2650 ASSERT(ifp
->if_bytes
<= XFS_IFORK_SIZE(ip
, whichfork
));
2651 memcpy(cp
, ifp
->if_u1
.if_data
, ifp
->if_bytes
);
2655 case XFS_DINODE_FMT_EXTENTS
:
2656 ASSERT((ifp
->if_flags
& XFS_IFEXTENTS
) ||
2657 !(iip
->ili_format
.ilf_fields
& extflag
[whichfork
]));
2658 ASSERT((xfs_iext_get_ext(ifp
, 0) != NULL
) ||
2659 (ifp
->if_bytes
== 0));
2660 ASSERT((xfs_iext_get_ext(ifp
, 0) == NULL
) ||
2661 (ifp
->if_bytes
> 0));
2662 if ((iip
->ili_format
.ilf_fields
& extflag
[whichfork
]) &&
2663 (ifp
->if_bytes
> 0)) {
2664 ASSERT(XFS_IFORK_NEXTENTS(ip
, whichfork
) > 0);
2665 (void)xfs_iextents_copy(ip
, (xfs_bmbt_rec_t
*)cp
,
2670 case XFS_DINODE_FMT_BTREE
:
2671 if ((iip
->ili_format
.ilf_fields
& brootflag
[whichfork
]) &&
2672 (ifp
->if_broot_bytes
> 0)) {
2673 ASSERT(ifp
->if_broot
!= NULL
);
2674 ASSERT(ifp
->if_broot_bytes
<=
2675 (XFS_IFORK_SIZE(ip
, whichfork
) +
2676 XFS_BROOT_SIZE_ADJ
));
2677 xfs_bmbt_to_bmdr(mp
, ifp
->if_broot
, ifp
->if_broot_bytes
,
2678 (xfs_bmdr_block_t
*)cp
,
2679 XFS_DFORK_SIZE(dip
, mp
, whichfork
));
2683 case XFS_DINODE_FMT_DEV
:
2684 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_DEV
) {
2685 ASSERT(whichfork
== XFS_DATA_FORK
);
2686 xfs_dinode_put_rdev(dip
, ip
->i_df
.if_u2
.if_rdev
);
2690 case XFS_DINODE_FMT_UUID
:
2691 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_UUID
) {
2692 ASSERT(whichfork
== XFS_DATA_FORK
);
2693 memcpy(XFS_DFORK_DPTR(dip
),
2694 &ip
->i_df
.if_u2
.if_uuid
,
2710 xfs_mount_t
*mp
= ip
->i_mount
;
2711 xfs_perag_t
*pag
= xfs_get_perag(mp
, ip
->i_ino
);
2712 unsigned long first_index
, mask
;
2713 unsigned long inodes_per_cluster
;
2715 xfs_inode_t
**ilist
;
2722 ASSERT(pag
->pagi_inodeok
);
2723 ASSERT(pag
->pag_ici_init
);
2725 inodes_per_cluster
= XFS_INODE_CLUSTER_SIZE(mp
) >> mp
->m_sb
.sb_inodelog
;
2726 ilist_size
= inodes_per_cluster
* sizeof(xfs_inode_t
*);
2727 ilist
= kmem_alloc(ilist_size
, KM_MAYFAIL
|KM_NOFS
);
2731 mask
= ~(((XFS_INODE_CLUSTER_SIZE(mp
) >> mp
->m_sb
.sb_inodelog
)) - 1);
2732 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
) & mask
;
2733 read_lock(&pag
->pag_ici_lock
);
2734 /* really need a gang lookup range call here */
2735 nr_found
= radix_tree_gang_lookup(&pag
->pag_ici_root
, (void**)ilist
,
2736 first_index
, inodes_per_cluster
);
2740 for (i
= 0; i
< nr_found
; i
++) {
2744 /* if the inode lies outside this cluster, we're done. */
2745 if ((XFS_INO_TO_AGINO(mp
, iq
->i_ino
) & mask
) != first_index
)
2748 * Do an un-protected check to see if the inode is dirty and
2749 * is a candidate for flushing. These checks will be repeated
2750 * later after the appropriate locks are acquired.
2752 if (xfs_inode_clean(iq
) && xfs_ipincount(iq
) == 0)
2756 * Try to get locks. If any are unavailable or it is pinned,
2757 * then this inode cannot be flushed and is skipped.
2760 if (!xfs_ilock_nowait(iq
, XFS_ILOCK_SHARED
))
2762 if (!xfs_iflock_nowait(iq
)) {
2763 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
2766 if (xfs_ipincount(iq
)) {
2768 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
2773 * arriving here means that this inode can be flushed. First
2774 * re-check that it's dirty before flushing.
2776 if (!xfs_inode_clean(iq
)) {
2778 error
= xfs_iflush_int(iq
, bp
);
2780 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
2781 goto cluster_corrupt_out
;
2787 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
2791 XFS_STATS_INC(xs_icluster_flushcnt
);
2792 XFS_STATS_ADD(xs_icluster_flushinode
, clcount
);
2796 read_unlock(&pag
->pag_ici_lock
);
2801 cluster_corrupt_out
:
2803 * Corruption detected in the clustering loop. Invalidate the
2804 * inode buffer and shut down the filesystem.
2806 read_unlock(&pag
->pag_ici_lock
);
2808 * Clean up the buffer. If it was B_DELWRI, just release it --
2809 * brelse can handle it with no problems. If not, shut down the
2810 * filesystem before releasing the buffer.
2812 bufwasdelwri
= XFS_BUF_ISDELAYWRITE(bp
);
2816 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
2818 if (!bufwasdelwri
) {
2820 * Just like incore_relse: if we have b_iodone functions,
2821 * mark the buffer as an error and call them. Otherwise
2822 * mark it as stale and brelse.
2824 if (XFS_BUF_IODONE_FUNC(bp
)) {
2825 XFS_BUF_CLR_BDSTRAT_FUNC(bp
);
2828 XFS_BUF_ERROR(bp
,EIO
);
2837 * Unlocks the flush lock
2839 xfs_iflush_abort(iq
);
2841 return XFS_ERROR(EFSCORRUPTED
);
2845 * xfs_iflush() will write a modified inode's changes out to the
2846 * inode's on disk home. The caller must have the inode lock held
2847 * in at least shared mode and the inode flush completion must be
2848 * active as well. The inode lock will still be held upon return from
2849 * the call and the caller is free to unlock it.
2850 * The inode flush will be completed when the inode reaches the disk.
2851 * The flags indicate how the inode's buffer should be written out.
2858 xfs_inode_log_item_t
*iip
;
2863 int noblock
= (flags
== XFS_IFLUSH_ASYNC_NOBLOCK
);
2864 enum { INT_DELWRI
= (1 << 0), INT_ASYNC
= (1 << 1) };
2866 XFS_STATS_INC(xs_iflush_count
);
2868 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2869 ASSERT(!completion_done(&ip
->i_flush
));
2870 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
2871 ip
->i_d
.di_nextents
> ip
->i_df
.if_ext_max
);
2877 * If the inode isn't dirty, then just release the inode flush lock and
2880 if (xfs_inode_clean(ip
)) {
2886 * We can't flush the inode until it is unpinned, so wait for it if we
2887 * are allowed to block. We know noone new can pin it, because we are
2888 * holding the inode lock shared and you need to hold it exclusively to
2891 * If we are not allowed to block, force the log out asynchronously so
2892 * that when we come back the inode will be unpinned. If other inodes
2893 * in the same cluster are dirty, they will probably write the inode
2894 * out for us if they occur after the log force completes.
2896 if (noblock
&& xfs_ipincount(ip
)) {
2897 xfs_iunpin_nowait(ip
);
2901 xfs_iunpin_wait(ip
);
2904 * For stale inodes we cannot rely on the backing buffer remaining
2905 * stale in cache for the remaining life of the stale inode and so
2906 * xfs_itobp() below may give us a buffer that no longer contains
2907 * inodes below. We have to check this after ensuring the inode is
2908 * unpinned so that it is safe to reclaim the stale inode after the
2911 if (xfs_iflags_test(ip
, XFS_ISTALE
)) {
2917 * This may have been unpinned because the filesystem is shutting
2918 * down forcibly. If that's the case we must not write this inode
2919 * to disk, because the log record didn't make it to disk!
2921 if (XFS_FORCED_SHUTDOWN(mp
)) {
2922 ip
->i_update_core
= 0;
2924 iip
->ili_format
.ilf_fields
= 0;
2926 return XFS_ERROR(EIO
);
2930 * Decide how buffer will be flushed out. This is done before
2931 * the call to xfs_iflush_int because this field is zeroed by it.
2933 if (iip
!= NULL
&& iip
->ili_format
.ilf_fields
!= 0) {
2935 * Flush out the inode buffer according to the directions
2936 * of the caller. In the cases where the caller has given
2937 * us a choice choose the non-delwri case. This is because
2938 * the inode is in the AIL and we need to get it out soon.
2941 case XFS_IFLUSH_SYNC
:
2942 case XFS_IFLUSH_DELWRI_ELSE_SYNC
:
2945 case XFS_IFLUSH_ASYNC_NOBLOCK
:
2946 case XFS_IFLUSH_ASYNC
:
2947 case XFS_IFLUSH_DELWRI_ELSE_ASYNC
:
2950 case XFS_IFLUSH_DELWRI
:
2960 case XFS_IFLUSH_DELWRI_ELSE_SYNC
:
2961 case XFS_IFLUSH_DELWRI_ELSE_ASYNC
:
2962 case XFS_IFLUSH_DELWRI
:
2965 case XFS_IFLUSH_ASYNC_NOBLOCK
:
2966 case XFS_IFLUSH_ASYNC
:
2969 case XFS_IFLUSH_SYNC
:
2980 * Get the buffer containing the on-disk inode.
2982 error
= xfs_itobp(mp
, NULL
, ip
, &dip
, &bp
,
2983 noblock
? XFS_BUF_TRYLOCK
: XFS_BUF_LOCK
);
2990 * First flush out the inode that xfs_iflush was called with.
2992 error
= xfs_iflush_int(ip
, bp
);
2997 * If the buffer is pinned then push on the log now so we won't
2998 * get stuck waiting in the write for too long.
3000 if (XFS_BUF_ISPINNED(bp
))
3001 xfs_log_force(mp
, (xfs_lsn_t
)0, XFS_LOG_FORCE
);
3005 * see if other inodes can be gathered into this write
3007 error
= xfs_iflush_cluster(ip
, bp
);
3009 goto cluster_corrupt_out
;
3011 if (flags
& INT_DELWRI
) {
3012 xfs_bdwrite(mp
, bp
);
3013 } else if (flags
& INT_ASYNC
) {
3014 error
= xfs_bawrite(mp
, bp
);
3016 error
= xfs_bwrite(mp
, bp
);
3022 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
3023 cluster_corrupt_out
:
3025 * Unlocks the flush lock
3027 xfs_iflush_abort(ip
);
3028 return XFS_ERROR(EFSCORRUPTED
);
3037 xfs_inode_log_item_t
*iip
;
3040 #ifdef XFS_TRANS_DEBUG
3044 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
3045 ASSERT(!completion_done(&ip
->i_flush
));
3046 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
3047 ip
->i_d
.di_nextents
> ip
->i_df
.if_ext_max
);
3054 * If the inode isn't dirty, then just release the inode
3055 * flush lock and do nothing.
3057 if (xfs_inode_clean(ip
)) {
3062 /* set *dip = inode's place in the buffer */
3063 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, ip
->i_imap
.im_boffset
);
3066 * Clear i_update_core before copying out the data.
3067 * This is for coordination with our timestamp updates
3068 * that don't hold the inode lock. They will always
3069 * update the timestamps BEFORE setting i_update_core,
3070 * so if we clear i_update_core after they set it we
3071 * are guaranteed to see their updates to the timestamps.
3072 * I believe that this depends on strongly ordered memory
3073 * semantics, but we have that. We use the SYNCHRONIZE
3074 * macro to make sure that the compiler does not reorder
3075 * the i_update_core access below the data copy below.
3077 ip
->i_update_core
= 0;
3081 * Make sure to get the latest timestamps from the Linux inode.
3083 xfs_synchronize_times(ip
);
3085 if (XFS_TEST_ERROR(be16_to_cpu(dip
->di_magic
) != XFS_DINODE_MAGIC
,
3086 mp
, XFS_ERRTAG_IFLUSH_1
, XFS_RANDOM_IFLUSH_1
)) {
3087 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3088 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3089 ip
->i_ino
, be16_to_cpu(dip
->di_magic
), dip
);
3092 if (XFS_TEST_ERROR(ip
->i_d
.di_magic
!= XFS_DINODE_MAGIC
,
3093 mp
, XFS_ERRTAG_IFLUSH_2
, XFS_RANDOM_IFLUSH_2
)) {
3094 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3095 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3096 ip
->i_ino
, ip
, ip
->i_d
.di_magic
);
3099 if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
) {
3101 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3102 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
),
3103 mp
, XFS_ERRTAG_IFLUSH_3
, XFS_RANDOM_IFLUSH_3
)) {
3104 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3105 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3109 } else if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFDIR
) {
3111 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3112 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
) &&
3113 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_LOCAL
),
3114 mp
, XFS_ERRTAG_IFLUSH_4
, XFS_RANDOM_IFLUSH_4
)) {
3115 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3116 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3121 if (XFS_TEST_ERROR(ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
>
3122 ip
->i_d
.di_nblocks
, mp
, XFS_ERRTAG_IFLUSH_5
,
3123 XFS_RANDOM_IFLUSH_5
)) {
3124 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3125 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3127 ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
,
3132 if (XFS_TEST_ERROR(ip
->i_d
.di_forkoff
> mp
->m_sb
.sb_inodesize
,
3133 mp
, XFS_ERRTAG_IFLUSH_6
, XFS_RANDOM_IFLUSH_6
)) {
3134 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3135 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3136 ip
->i_ino
, ip
->i_d
.di_forkoff
, ip
);
3140 * bump the flush iteration count, used to detect flushes which
3141 * postdate a log record during recovery.
3144 ip
->i_d
.di_flushiter
++;
3147 * Copy the dirty parts of the inode into the on-disk
3148 * inode. We always copy out the core of the inode,
3149 * because if the inode is dirty at all the core must
3152 xfs_dinode_to_disk(dip
, &ip
->i_d
);
3154 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3155 if (ip
->i_d
.di_flushiter
== DI_MAX_FLUSH
)
3156 ip
->i_d
.di_flushiter
= 0;
3159 * If this is really an old format inode and the superblock version
3160 * has not been updated to support only new format inodes, then
3161 * convert back to the old inode format. If the superblock version
3162 * has been updated, then make the conversion permanent.
3164 ASSERT(ip
->i_d
.di_version
== 1 || xfs_sb_version_hasnlink(&mp
->m_sb
));
3165 if (ip
->i_d
.di_version
== 1) {
3166 if (!xfs_sb_version_hasnlink(&mp
->m_sb
)) {
3170 ASSERT(ip
->i_d
.di_nlink
<= XFS_MAXLINK_1
);
3171 dip
->di_onlink
= cpu_to_be16(ip
->i_d
.di_nlink
);
3174 * The superblock version has already been bumped,
3175 * so just make the conversion to the new inode
3178 ip
->i_d
.di_version
= 2;
3179 dip
->di_version
= 2;
3180 ip
->i_d
.di_onlink
= 0;
3182 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
3183 memset(&(dip
->di_pad
[0]), 0,
3184 sizeof(dip
->di_pad
));
3185 ASSERT(ip
->i_d
.di_projid
== 0);
3189 xfs_iflush_fork(ip
, dip
, iip
, XFS_DATA_FORK
, bp
);
3190 if (XFS_IFORK_Q(ip
))
3191 xfs_iflush_fork(ip
, dip
, iip
, XFS_ATTR_FORK
, bp
);
3192 xfs_inobp_check(mp
, bp
);
3195 * We've recorded everything logged in the inode, so we'd
3196 * like to clear the ilf_fields bits so we don't log and
3197 * flush things unnecessarily. However, we can't stop
3198 * logging all this information until the data we've copied
3199 * into the disk buffer is written to disk. If we did we might
3200 * overwrite the copy of the inode in the log with all the
3201 * data after re-logging only part of it, and in the face of
3202 * a crash we wouldn't have all the data we need to recover.
3204 * What we do is move the bits to the ili_last_fields field.
3205 * When logging the inode, these bits are moved back to the
3206 * ilf_fields field. In the xfs_iflush_done() routine we
3207 * clear ili_last_fields, since we know that the information
3208 * those bits represent is permanently on disk. As long as
3209 * the flush completes before the inode is logged again, then
3210 * both ilf_fields and ili_last_fields will be cleared.
3212 * We can play with the ilf_fields bits here, because the inode
3213 * lock must be held exclusively in order to set bits there
3214 * and the flush lock protects the ili_last_fields bits.
3215 * Set ili_logged so the flush done
3216 * routine can tell whether or not to look in the AIL.
3217 * Also, store the current LSN of the inode so that we can tell
3218 * whether the item has moved in the AIL from xfs_iflush_done().
3219 * In order to read the lsn we need the AIL lock, because
3220 * it is a 64 bit value that cannot be read atomically.
3222 if (iip
!= NULL
&& iip
->ili_format
.ilf_fields
!= 0) {
3223 iip
->ili_last_fields
= iip
->ili_format
.ilf_fields
;
3224 iip
->ili_format
.ilf_fields
= 0;
3225 iip
->ili_logged
= 1;
3227 xfs_trans_ail_copy_lsn(mp
->m_ail
, &iip
->ili_flush_lsn
,
3228 &iip
->ili_item
.li_lsn
);
3231 * Attach the function xfs_iflush_done to the inode's
3232 * buffer. This will remove the inode from the AIL
3233 * and unlock the inode's flush lock when the inode is
3234 * completely written to disk.
3236 xfs_buf_attach_iodone(bp
, (void(*)(xfs_buf_t
*,xfs_log_item_t
*))
3237 xfs_iflush_done
, (xfs_log_item_t
*)iip
);
3239 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) != NULL
);
3240 ASSERT(XFS_BUF_IODONE_FUNC(bp
) != NULL
);
3243 * We're flushing an inode which is not in the AIL and has
3244 * not been logged but has i_update_core set. For this
3245 * case we can use a B_DELWRI flush and immediately drop
3246 * the inode flush lock because we can avoid the whole
3247 * AIL state thing. It's OK to drop the flush lock now,
3248 * because we've already locked the buffer and to do anything
3249 * you really need both.
3252 ASSERT(iip
->ili_logged
== 0);
3253 ASSERT(iip
->ili_last_fields
== 0);
3254 ASSERT((iip
->ili_item
.li_flags
& XFS_LI_IN_AIL
) == 0);
3262 return XFS_ERROR(EFSCORRUPTED
);
3267 #ifdef XFS_ILOCK_TRACE
3269 xfs_ilock_trace(xfs_inode_t
*ip
, int lock
, unsigned int lockflags
, inst_t
*ra
)
3271 ktrace_enter(ip
->i_lock_trace
,
3273 (void *)(unsigned long)lock
, /* 1 = LOCK, 3=UNLOCK, etc */
3274 (void *)(unsigned long)lockflags
, /* XFS_ILOCK_EXCL etc */
3275 (void *)ra
, /* caller of ilock */
3276 (void *)(unsigned long)current_cpu(),
3277 (void *)(unsigned long)current_pid(),
3278 NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
);
3283 * Return a pointer to the extent record at file index idx.
3285 xfs_bmbt_rec_host_t
*
3287 xfs_ifork_t
*ifp
, /* inode fork pointer */
3288 xfs_extnum_t idx
) /* index of target extent */
3291 if ((ifp
->if_flags
& XFS_IFEXTIREC
) && (idx
== 0)) {
3292 return ifp
->if_u1
.if_ext_irec
->er_extbuf
;
3293 } else if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3294 xfs_ext_irec_t
*erp
; /* irec pointer */
3295 int erp_idx
= 0; /* irec index */
3296 xfs_extnum_t page_idx
= idx
; /* ext index in target list */
3298 erp
= xfs_iext_idx_to_irec(ifp
, &page_idx
, &erp_idx
, 0);
3299 return &erp
->er_extbuf
[page_idx
];
3300 } else if (ifp
->if_bytes
) {
3301 return &ifp
->if_u1
.if_extents
[idx
];
3308 * Insert new item(s) into the extent records for incore inode
3309 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3313 xfs_ifork_t
*ifp
, /* inode fork pointer */
3314 xfs_extnum_t idx
, /* starting index of new items */
3315 xfs_extnum_t count
, /* number of inserted items */
3316 xfs_bmbt_irec_t
*new) /* items to insert */
3318 xfs_extnum_t i
; /* extent record index */
3320 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
3321 xfs_iext_add(ifp
, idx
, count
);
3322 for (i
= idx
; i
< idx
+ count
; i
++, new++)
3323 xfs_bmbt_set_all(xfs_iext_get_ext(ifp
, i
), new);
3327 * This is called when the amount of space required for incore file
3328 * extents needs to be increased. The ext_diff parameter stores the
3329 * number of new extents being added and the idx parameter contains
3330 * the extent index where the new extents will be added. If the new
3331 * extents are being appended, then we just need to (re)allocate and
3332 * initialize the space. Otherwise, if the new extents are being
3333 * inserted into the middle of the existing entries, a bit more work
3334 * is required to make room for the new extents to be inserted. The
3335 * caller is responsible for filling in the new extent entries upon
3340 xfs_ifork_t
*ifp
, /* inode fork pointer */
3341 xfs_extnum_t idx
, /* index to begin adding exts */
3342 int ext_diff
) /* number of extents to add */
3344 int byte_diff
; /* new bytes being added */
3345 int new_size
; /* size of extents after adding */
3346 xfs_extnum_t nextents
; /* number of extents in file */
3348 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3349 ASSERT((idx
>= 0) && (idx
<= nextents
));
3350 byte_diff
= ext_diff
* sizeof(xfs_bmbt_rec_t
);
3351 new_size
= ifp
->if_bytes
+ byte_diff
;
3353 * If the new number of extents (nextents + ext_diff)
3354 * fits inside the inode, then continue to use the inline
3357 if (nextents
+ ext_diff
<= XFS_INLINE_EXTS
) {
3358 if (idx
< nextents
) {
3359 memmove(&ifp
->if_u2
.if_inline_ext
[idx
+ ext_diff
],
3360 &ifp
->if_u2
.if_inline_ext
[idx
],
3361 (nextents
- idx
) * sizeof(xfs_bmbt_rec_t
));
3362 memset(&ifp
->if_u2
.if_inline_ext
[idx
], 0, byte_diff
);
3364 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
3365 ifp
->if_real_bytes
= 0;
3366 ifp
->if_lastex
= nextents
+ ext_diff
;
3369 * Otherwise use a linear (direct) extent list.
3370 * If the extents are currently inside the inode,
3371 * xfs_iext_realloc_direct will switch us from
3372 * inline to direct extent allocation mode.
3374 else if (nextents
+ ext_diff
<= XFS_LINEAR_EXTS
) {
3375 xfs_iext_realloc_direct(ifp
, new_size
);
3376 if (idx
< nextents
) {
3377 memmove(&ifp
->if_u1
.if_extents
[idx
+ ext_diff
],
3378 &ifp
->if_u1
.if_extents
[idx
],
3379 (nextents
- idx
) * sizeof(xfs_bmbt_rec_t
));
3380 memset(&ifp
->if_u1
.if_extents
[idx
], 0, byte_diff
);
3383 /* Indirection array */
3385 xfs_ext_irec_t
*erp
;
3389 ASSERT(nextents
+ ext_diff
> XFS_LINEAR_EXTS
);
3390 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3391 erp
= xfs_iext_idx_to_irec(ifp
, &page_idx
, &erp_idx
, 1);
3393 xfs_iext_irec_init(ifp
);
3394 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3395 erp
= ifp
->if_u1
.if_ext_irec
;
3397 /* Extents fit in target extent page */
3398 if (erp
&& erp
->er_extcount
+ ext_diff
<= XFS_LINEAR_EXTS
) {
3399 if (page_idx
< erp
->er_extcount
) {
3400 memmove(&erp
->er_extbuf
[page_idx
+ ext_diff
],
3401 &erp
->er_extbuf
[page_idx
],
3402 (erp
->er_extcount
- page_idx
) *
3403 sizeof(xfs_bmbt_rec_t
));
3404 memset(&erp
->er_extbuf
[page_idx
], 0, byte_diff
);
3406 erp
->er_extcount
+= ext_diff
;
3407 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, ext_diff
);
3409 /* Insert a new extent page */
3411 xfs_iext_add_indirect_multi(ifp
,
3412 erp_idx
, page_idx
, ext_diff
);
3415 * If extent(s) are being appended to the last page in
3416 * the indirection array and the new extent(s) don't fit
3417 * in the page, then erp is NULL and erp_idx is set to
3418 * the next index needed in the indirection array.
3421 int count
= ext_diff
;
3424 erp
= xfs_iext_irec_new(ifp
, erp_idx
);
3425 erp
->er_extcount
= count
;
3426 count
-= MIN(count
, (int)XFS_LINEAR_EXTS
);
3433 ifp
->if_bytes
= new_size
;
3437 * This is called when incore extents are being added to the indirection
3438 * array and the new extents do not fit in the target extent list. The
3439 * erp_idx parameter contains the irec index for the target extent list
3440 * in the indirection array, and the idx parameter contains the extent
3441 * index within the list. The number of extents being added is stored
3442 * in the count parameter.
3444 * |-------| |-------|
3445 * | | | | idx - number of extents before idx
3447 * | | | | count - number of extents being inserted at idx
3448 * |-------| |-------|
3449 * | count | | nex2 | nex2 - number of extents after idx + count
3450 * |-------| |-------|
3453 xfs_iext_add_indirect_multi(
3454 xfs_ifork_t
*ifp
, /* inode fork pointer */
3455 int erp_idx
, /* target extent irec index */
3456 xfs_extnum_t idx
, /* index within target list */
3457 int count
) /* new extents being added */
3459 int byte_diff
; /* new bytes being added */
3460 xfs_ext_irec_t
*erp
; /* pointer to irec entry */
3461 xfs_extnum_t ext_diff
; /* number of extents to add */
3462 xfs_extnum_t ext_cnt
; /* new extents still needed */
3463 xfs_extnum_t nex2
; /* extents after idx + count */
3464 xfs_bmbt_rec_t
*nex2_ep
= NULL
; /* temp list for nex2 extents */
3465 int nlists
; /* number of irec's (lists) */
3467 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3468 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
3469 nex2
= erp
->er_extcount
- idx
;
3470 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3473 * Save second part of target extent list
3474 * (all extents past */
3476 byte_diff
= nex2
* sizeof(xfs_bmbt_rec_t
);
3477 nex2_ep
= (xfs_bmbt_rec_t
*) kmem_alloc(byte_diff
, KM_NOFS
);
3478 memmove(nex2_ep
, &erp
->er_extbuf
[idx
], byte_diff
);
3479 erp
->er_extcount
-= nex2
;
3480 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, -nex2
);
3481 memset(&erp
->er_extbuf
[idx
], 0, byte_diff
);
3485 * Add the new extents to the end of the target
3486 * list, then allocate new irec record(s) and
3487 * extent buffer(s) as needed to store the rest
3488 * of the new extents.
3491 ext_diff
= MIN(ext_cnt
, (int)XFS_LINEAR_EXTS
- erp
->er_extcount
);
3493 erp
->er_extcount
+= ext_diff
;
3494 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, ext_diff
);
3495 ext_cnt
-= ext_diff
;
3499 erp
= xfs_iext_irec_new(ifp
, erp_idx
);
3500 ext_diff
= MIN(ext_cnt
, (int)XFS_LINEAR_EXTS
);
3501 erp
->er_extcount
= ext_diff
;
3502 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, ext_diff
);
3503 ext_cnt
-= ext_diff
;
3506 /* Add nex2 extents back to indirection array */
3508 xfs_extnum_t ext_avail
;
3511 byte_diff
= nex2
* sizeof(xfs_bmbt_rec_t
);
3512 ext_avail
= XFS_LINEAR_EXTS
- erp
->er_extcount
;
3515 * If nex2 extents fit in the current page, append
3516 * nex2_ep after the new extents.
3518 if (nex2
<= ext_avail
) {
3519 i
= erp
->er_extcount
;
3522 * Otherwise, check if space is available in the
3525 else if ((erp_idx
< nlists
- 1) &&
3526 (nex2
<= (ext_avail
= XFS_LINEAR_EXTS
-
3527 ifp
->if_u1
.if_ext_irec
[erp_idx
+1].er_extcount
))) {
3530 /* Create a hole for nex2 extents */
3531 memmove(&erp
->er_extbuf
[nex2
], erp
->er_extbuf
,
3532 erp
->er_extcount
* sizeof(xfs_bmbt_rec_t
));
3535 * Final choice, create a new extent page for
3540 erp
= xfs_iext_irec_new(ifp
, erp_idx
);
3542 memmove(&erp
->er_extbuf
[i
], nex2_ep
, byte_diff
);
3544 erp
->er_extcount
+= nex2
;
3545 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, nex2
);
3550 * This is called when the amount of space required for incore file
3551 * extents needs to be decreased. The ext_diff parameter stores the
3552 * number of extents to be removed and the idx parameter contains
3553 * the extent index where the extents will be removed from.
3555 * If the amount of space needed has decreased below the linear
3556 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3557 * extent array. Otherwise, use kmem_realloc() to adjust the
3558 * size to what is needed.
3562 xfs_ifork_t
*ifp
, /* inode fork pointer */
3563 xfs_extnum_t idx
, /* index to begin removing exts */
3564 int ext_diff
) /* number of extents to remove */
3566 xfs_extnum_t nextents
; /* number of extents in file */
3567 int new_size
; /* size of extents after removal */
3569 ASSERT(ext_diff
> 0);
3570 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3571 new_size
= (nextents
- ext_diff
) * sizeof(xfs_bmbt_rec_t
);
3573 if (new_size
== 0) {
3574 xfs_iext_destroy(ifp
);
3575 } else if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3576 xfs_iext_remove_indirect(ifp
, idx
, ext_diff
);
3577 } else if (ifp
->if_real_bytes
) {
3578 xfs_iext_remove_direct(ifp
, idx
, ext_diff
);
3580 xfs_iext_remove_inline(ifp
, idx
, ext_diff
);
3582 ifp
->if_bytes
= new_size
;
3586 * This removes ext_diff extents from the inline buffer, beginning
3587 * at extent index idx.
3590 xfs_iext_remove_inline(
3591 xfs_ifork_t
*ifp
, /* inode fork pointer */
3592 xfs_extnum_t idx
, /* index to begin removing exts */
3593 int ext_diff
) /* number of extents to remove */
3595 int nextents
; /* number of extents in file */
3597 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
));
3598 ASSERT(idx
< XFS_INLINE_EXTS
);
3599 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3600 ASSERT(((nextents
- ext_diff
) > 0) &&
3601 (nextents
- ext_diff
) < XFS_INLINE_EXTS
);
3603 if (idx
+ ext_diff
< nextents
) {
3604 memmove(&ifp
->if_u2
.if_inline_ext
[idx
],
3605 &ifp
->if_u2
.if_inline_ext
[idx
+ ext_diff
],
3606 (nextents
- (idx
+ ext_diff
)) *
3607 sizeof(xfs_bmbt_rec_t
));
3608 memset(&ifp
->if_u2
.if_inline_ext
[nextents
- ext_diff
],
3609 0, ext_diff
* sizeof(xfs_bmbt_rec_t
));
3611 memset(&ifp
->if_u2
.if_inline_ext
[idx
], 0,
3612 ext_diff
* sizeof(xfs_bmbt_rec_t
));
3617 * This removes ext_diff extents from a linear (direct) extent list,
3618 * beginning at extent index idx. If the extents are being removed
3619 * from the end of the list (ie. truncate) then we just need to re-
3620 * allocate the list to remove the extra space. Otherwise, if the
3621 * extents are being removed from the middle of the existing extent
3622 * entries, then we first need to move the extent records beginning
3623 * at idx + ext_diff up in the list to overwrite the records being
3624 * removed, then remove the extra space via kmem_realloc.
3627 xfs_iext_remove_direct(
3628 xfs_ifork_t
*ifp
, /* inode fork pointer */
3629 xfs_extnum_t idx
, /* index to begin removing exts */
3630 int ext_diff
) /* number of extents to remove */
3632 xfs_extnum_t nextents
; /* number of extents in file */
3633 int new_size
; /* size of extents after removal */
3635 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
));
3636 new_size
= ifp
->if_bytes
-
3637 (ext_diff
* sizeof(xfs_bmbt_rec_t
));
3638 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3640 if (new_size
== 0) {
3641 xfs_iext_destroy(ifp
);
3644 /* Move extents up in the list (if needed) */
3645 if (idx
+ ext_diff
< nextents
) {
3646 memmove(&ifp
->if_u1
.if_extents
[idx
],
3647 &ifp
->if_u1
.if_extents
[idx
+ ext_diff
],
3648 (nextents
- (idx
+ ext_diff
)) *
3649 sizeof(xfs_bmbt_rec_t
));
3651 memset(&ifp
->if_u1
.if_extents
[nextents
- ext_diff
],
3652 0, ext_diff
* sizeof(xfs_bmbt_rec_t
));
3654 * Reallocate the direct extent list. If the extents
3655 * will fit inside the inode then xfs_iext_realloc_direct
3656 * will switch from direct to inline extent allocation
3659 xfs_iext_realloc_direct(ifp
, new_size
);
3660 ifp
->if_bytes
= new_size
;
3664 * This is called when incore extents are being removed from the
3665 * indirection array and the extents being removed span multiple extent
3666 * buffers. The idx parameter contains the file extent index where we
3667 * want to begin removing extents, and the count parameter contains
3668 * how many extents need to be removed.
3670 * |-------| |-------|
3671 * | nex1 | | | nex1 - number of extents before idx
3672 * |-------| | count |
3673 * | | | | count - number of extents being removed at idx
3674 * | count | |-------|
3675 * | | | nex2 | nex2 - number of extents after idx + count
3676 * |-------| |-------|
3679 xfs_iext_remove_indirect(
3680 xfs_ifork_t
*ifp
, /* inode fork pointer */
3681 xfs_extnum_t idx
, /* index to begin removing extents */
3682 int count
) /* number of extents to remove */
3684 xfs_ext_irec_t
*erp
; /* indirection array pointer */
3685 int erp_idx
= 0; /* indirection array index */
3686 xfs_extnum_t ext_cnt
; /* extents left to remove */
3687 xfs_extnum_t ext_diff
; /* extents to remove in current list */
3688 xfs_extnum_t nex1
; /* number of extents before idx */
3689 xfs_extnum_t nex2
; /* extents after idx + count */
3690 int nlists
; /* entries in indirection array */
3691 int page_idx
= idx
; /* index in target extent list */
3693 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3694 erp
= xfs_iext_idx_to_irec(ifp
, &page_idx
, &erp_idx
, 0);
3695 ASSERT(erp
!= NULL
);
3696 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3700 nex2
= MAX((erp
->er_extcount
- (nex1
+ ext_cnt
)), 0);
3701 ext_diff
= MIN(ext_cnt
, (erp
->er_extcount
- nex1
));
3703 * Check for deletion of entire list;
3704 * xfs_iext_irec_remove() updates extent offsets.
3706 if (ext_diff
== erp
->er_extcount
) {
3707 xfs_iext_irec_remove(ifp
, erp_idx
);
3708 ext_cnt
-= ext_diff
;
3711 ASSERT(erp_idx
< ifp
->if_real_bytes
/
3713 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
3720 /* Move extents up (if needed) */
3722 memmove(&erp
->er_extbuf
[nex1
],
3723 &erp
->er_extbuf
[nex1
+ ext_diff
],
3724 nex2
* sizeof(xfs_bmbt_rec_t
));
3726 /* Zero out rest of page */
3727 memset(&erp
->er_extbuf
[nex1
+ nex2
], 0, (XFS_IEXT_BUFSZ
-
3728 ((nex1
+ nex2
) * sizeof(xfs_bmbt_rec_t
))));
3729 /* Update remaining counters */
3730 erp
->er_extcount
-= ext_diff
;
3731 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, -ext_diff
);
3732 ext_cnt
-= ext_diff
;
3737 ifp
->if_bytes
-= count
* sizeof(xfs_bmbt_rec_t
);
3738 xfs_iext_irec_compact(ifp
);
3742 * Create, destroy, or resize a linear (direct) block of extents.
3745 xfs_iext_realloc_direct(
3746 xfs_ifork_t
*ifp
, /* inode fork pointer */
3747 int new_size
) /* new size of extents */
3749 int rnew_size
; /* real new size of extents */
3751 rnew_size
= new_size
;
3753 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
) ||
3754 ((new_size
>= 0) && (new_size
<= XFS_IEXT_BUFSZ
) &&
3755 (new_size
!= ifp
->if_real_bytes
)));
3757 /* Free extent records */
3758 if (new_size
== 0) {
3759 xfs_iext_destroy(ifp
);
3761 /* Resize direct extent list and zero any new bytes */
3762 else if (ifp
->if_real_bytes
) {
3763 /* Check if extents will fit inside the inode */
3764 if (new_size
<= XFS_INLINE_EXTS
* sizeof(xfs_bmbt_rec_t
)) {
3765 xfs_iext_direct_to_inline(ifp
, new_size
/
3766 (uint
)sizeof(xfs_bmbt_rec_t
));
3767 ifp
->if_bytes
= new_size
;
3770 if (!is_power_of_2(new_size
)){
3771 rnew_size
= roundup_pow_of_two(new_size
);
3773 if (rnew_size
!= ifp
->if_real_bytes
) {
3774 ifp
->if_u1
.if_extents
=
3775 kmem_realloc(ifp
->if_u1
.if_extents
,
3777 ifp
->if_real_bytes
, KM_NOFS
);
3779 if (rnew_size
> ifp
->if_real_bytes
) {
3780 memset(&ifp
->if_u1
.if_extents
[ifp
->if_bytes
/
3781 (uint
)sizeof(xfs_bmbt_rec_t
)], 0,
3782 rnew_size
- ifp
->if_real_bytes
);
3786 * Switch from the inline extent buffer to a direct
3787 * extent list. Be sure to include the inline extent
3788 * bytes in new_size.
3791 new_size
+= ifp
->if_bytes
;
3792 if (!is_power_of_2(new_size
)) {
3793 rnew_size
= roundup_pow_of_two(new_size
);
3795 xfs_iext_inline_to_direct(ifp
, rnew_size
);
3797 ifp
->if_real_bytes
= rnew_size
;
3798 ifp
->if_bytes
= new_size
;
3802 * Switch from linear (direct) extent records to inline buffer.
3805 xfs_iext_direct_to_inline(
3806 xfs_ifork_t
*ifp
, /* inode fork pointer */
3807 xfs_extnum_t nextents
) /* number of extents in file */
3809 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
3810 ASSERT(nextents
<= XFS_INLINE_EXTS
);
3812 * The inline buffer was zeroed when we switched
3813 * from inline to direct extent allocation mode,
3814 * so we don't need to clear it here.
3816 memcpy(ifp
->if_u2
.if_inline_ext
, ifp
->if_u1
.if_extents
,
3817 nextents
* sizeof(xfs_bmbt_rec_t
));
3818 kmem_free(ifp
->if_u1
.if_extents
);
3819 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
3820 ifp
->if_real_bytes
= 0;
3824 * Switch from inline buffer to linear (direct) extent records.
3825 * new_size should already be rounded up to the next power of 2
3826 * by the caller (when appropriate), so use new_size as it is.
3827 * However, since new_size may be rounded up, we can't update
3828 * if_bytes here. It is the caller's responsibility to update
3829 * if_bytes upon return.
3832 xfs_iext_inline_to_direct(
3833 xfs_ifork_t
*ifp
, /* inode fork pointer */
3834 int new_size
) /* number of extents in file */
3836 ifp
->if_u1
.if_extents
= kmem_alloc(new_size
, KM_NOFS
);
3837 memset(ifp
->if_u1
.if_extents
, 0, new_size
);
3838 if (ifp
->if_bytes
) {
3839 memcpy(ifp
->if_u1
.if_extents
, ifp
->if_u2
.if_inline_ext
,
3841 memset(ifp
->if_u2
.if_inline_ext
, 0, XFS_INLINE_EXTS
*
3842 sizeof(xfs_bmbt_rec_t
));
3844 ifp
->if_real_bytes
= new_size
;
3848 * Resize an extent indirection array to new_size bytes.
3851 xfs_iext_realloc_indirect(
3852 xfs_ifork_t
*ifp
, /* inode fork pointer */
3853 int new_size
) /* new indirection array size */
3855 int nlists
; /* number of irec's (ex lists) */
3856 int size
; /* current indirection array size */
3858 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3859 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3860 size
= nlists
* sizeof(xfs_ext_irec_t
);
3861 ASSERT(ifp
->if_real_bytes
);
3862 ASSERT((new_size
>= 0) && (new_size
!= size
));
3863 if (new_size
== 0) {
3864 xfs_iext_destroy(ifp
);
3866 ifp
->if_u1
.if_ext_irec
= (xfs_ext_irec_t
*)
3867 kmem_realloc(ifp
->if_u1
.if_ext_irec
,
3868 new_size
, size
, KM_NOFS
);
3873 * Switch from indirection array to linear (direct) extent allocations.
3876 xfs_iext_indirect_to_direct(
3877 xfs_ifork_t
*ifp
) /* inode fork pointer */
3879 xfs_bmbt_rec_host_t
*ep
; /* extent record pointer */
3880 xfs_extnum_t nextents
; /* number of extents in file */
3881 int size
; /* size of file extents */
3883 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3884 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3885 ASSERT(nextents
<= XFS_LINEAR_EXTS
);
3886 size
= nextents
* sizeof(xfs_bmbt_rec_t
);
3888 xfs_iext_irec_compact_pages(ifp
);
3889 ASSERT(ifp
->if_real_bytes
== XFS_IEXT_BUFSZ
);
3891 ep
= ifp
->if_u1
.if_ext_irec
->er_extbuf
;
3892 kmem_free(ifp
->if_u1
.if_ext_irec
);
3893 ifp
->if_flags
&= ~XFS_IFEXTIREC
;
3894 ifp
->if_u1
.if_extents
= ep
;
3895 ifp
->if_bytes
= size
;
3896 if (nextents
< XFS_LINEAR_EXTS
) {
3897 xfs_iext_realloc_direct(ifp
, size
);
3902 * Free incore file extents.
3906 xfs_ifork_t
*ifp
) /* inode fork pointer */
3908 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3912 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3913 for (erp_idx
= nlists
- 1; erp_idx
>= 0 ; erp_idx
--) {
3914 xfs_iext_irec_remove(ifp
, erp_idx
);
3916 ifp
->if_flags
&= ~XFS_IFEXTIREC
;
3917 } else if (ifp
->if_real_bytes
) {
3918 kmem_free(ifp
->if_u1
.if_extents
);
3919 } else if (ifp
->if_bytes
) {
3920 memset(ifp
->if_u2
.if_inline_ext
, 0, XFS_INLINE_EXTS
*
3921 sizeof(xfs_bmbt_rec_t
));
3923 ifp
->if_u1
.if_extents
= NULL
;
3924 ifp
->if_real_bytes
= 0;
3929 * Return a pointer to the extent record for file system block bno.
3931 xfs_bmbt_rec_host_t
* /* pointer to found extent record */
3932 xfs_iext_bno_to_ext(
3933 xfs_ifork_t
*ifp
, /* inode fork pointer */
3934 xfs_fileoff_t bno
, /* block number to search for */
3935 xfs_extnum_t
*idxp
) /* index of target extent */
3937 xfs_bmbt_rec_host_t
*base
; /* pointer to first extent */
3938 xfs_filblks_t blockcount
= 0; /* number of blocks in extent */
3939 xfs_bmbt_rec_host_t
*ep
= NULL
; /* pointer to target extent */
3940 xfs_ext_irec_t
*erp
= NULL
; /* indirection array pointer */
3941 int high
; /* upper boundary in search */
3942 xfs_extnum_t idx
= 0; /* index of target extent */
3943 int low
; /* lower boundary in search */
3944 xfs_extnum_t nextents
; /* number of file extents */
3945 xfs_fileoff_t startoff
= 0; /* start offset of extent */
3947 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3948 if (nextents
== 0) {
3953 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3954 /* Find target extent list */
3956 erp
= xfs_iext_bno_to_irec(ifp
, bno
, &erp_idx
);
3957 base
= erp
->er_extbuf
;
3958 high
= erp
->er_extcount
- 1;
3960 base
= ifp
->if_u1
.if_extents
;
3961 high
= nextents
- 1;
3963 /* Binary search extent records */
3964 while (low
<= high
) {
3965 idx
= (low
+ high
) >> 1;
3967 startoff
= xfs_bmbt_get_startoff(ep
);
3968 blockcount
= xfs_bmbt_get_blockcount(ep
);
3969 if (bno
< startoff
) {
3971 } else if (bno
>= startoff
+ blockcount
) {
3974 /* Convert back to file-based extent index */
3975 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3976 idx
+= erp
->er_extoff
;
3982 /* Convert back to file-based extent index */
3983 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3984 idx
+= erp
->er_extoff
;
3986 if (bno
>= startoff
+ blockcount
) {
3987 if (++idx
== nextents
) {
3990 ep
= xfs_iext_get_ext(ifp
, idx
);
3998 * Return a pointer to the indirection array entry containing the
3999 * extent record for filesystem block bno. Store the index of the
4000 * target irec in *erp_idxp.
4002 xfs_ext_irec_t
* /* pointer to found extent record */
4003 xfs_iext_bno_to_irec(
4004 xfs_ifork_t
*ifp
, /* inode fork pointer */
4005 xfs_fileoff_t bno
, /* block number to search for */
4006 int *erp_idxp
) /* irec index of target ext list */
4008 xfs_ext_irec_t
*erp
= NULL
; /* indirection array pointer */
4009 xfs_ext_irec_t
*erp_next
; /* next indirection array entry */
4010 int erp_idx
; /* indirection array index */
4011 int nlists
; /* number of extent irec's (lists) */
4012 int high
; /* binary search upper limit */
4013 int low
; /* binary search lower limit */
4015 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4016 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4020 while (low
<= high
) {
4021 erp_idx
= (low
+ high
) >> 1;
4022 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4023 erp_next
= erp_idx
< nlists
- 1 ? erp
+ 1 : NULL
;
4024 if (bno
< xfs_bmbt_get_startoff(erp
->er_extbuf
)) {
4026 } else if (erp_next
&& bno
>=
4027 xfs_bmbt_get_startoff(erp_next
->er_extbuf
)) {
4033 *erp_idxp
= erp_idx
;
4038 * Return a pointer to the indirection array entry containing the
4039 * extent record at file extent index *idxp. Store the index of the
4040 * target irec in *erp_idxp and store the page index of the target
4041 * extent record in *idxp.
4044 xfs_iext_idx_to_irec(
4045 xfs_ifork_t
*ifp
, /* inode fork pointer */
4046 xfs_extnum_t
*idxp
, /* extent index (file -> page) */
4047 int *erp_idxp
, /* pointer to target irec */
4048 int realloc
) /* new bytes were just added */
4050 xfs_ext_irec_t
*prev
; /* pointer to previous irec */
4051 xfs_ext_irec_t
*erp
= NULL
; /* pointer to current irec */
4052 int erp_idx
; /* indirection array index */
4053 int nlists
; /* number of irec's (ex lists) */
4054 int high
; /* binary search upper limit */
4055 int low
; /* binary search lower limit */
4056 xfs_extnum_t page_idx
= *idxp
; /* extent index in target list */
4058 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4059 ASSERT(page_idx
>= 0 && page_idx
<=
4060 ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
));
4061 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4066 /* Binary search extent irec's */
4067 while (low
<= high
) {
4068 erp_idx
= (low
+ high
) >> 1;
4069 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4070 prev
= erp_idx
> 0 ? erp
- 1 : NULL
;
4071 if (page_idx
< erp
->er_extoff
|| (page_idx
== erp
->er_extoff
&&
4072 realloc
&& prev
&& prev
->er_extcount
< XFS_LINEAR_EXTS
)) {
4074 } else if (page_idx
> erp
->er_extoff
+ erp
->er_extcount
||
4075 (page_idx
== erp
->er_extoff
+ erp
->er_extcount
&&
4078 } else if (page_idx
== erp
->er_extoff
+ erp
->er_extcount
&&
4079 erp
->er_extcount
== XFS_LINEAR_EXTS
) {
4083 erp
= erp_idx
< nlists
? erp
+ 1 : NULL
;
4086 page_idx
-= erp
->er_extoff
;
4091 *erp_idxp
= erp_idx
;
4096 * Allocate and initialize an indirection array once the space needed
4097 * for incore extents increases above XFS_IEXT_BUFSZ.
4101 xfs_ifork_t
*ifp
) /* inode fork pointer */
4103 xfs_ext_irec_t
*erp
; /* indirection array pointer */
4104 xfs_extnum_t nextents
; /* number of extents in file */
4106 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
));
4107 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4108 ASSERT(nextents
<= XFS_LINEAR_EXTS
);
4110 erp
= kmem_alloc(sizeof(xfs_ext_irec_t
), KM_NOFS
);
4112 if (nextents
== 0) {
4113 ifp
->if_u1
.if_extents
= kmem_alloc(XFS_IEXT_BUFSZ
, KM_NOFS
);
4114 } else if (!ifp
->if_real_bytes
) {
4115 xfs_iext_inline_to_direct(ifp
, XFS_IEXT_BUFSZ
);
4116 } else if (ifp
->if_real_bytes
< XFS_IEXT_BUFSZ
) {
4117 xfs_iext_realloc_direct(ifp
, XFS_IEXT_BUFSZ
);
4119 erp
->er_extbuf
= ifp
->if_u1
.if_extents
;
4120 erp
->er_extcount
= nextents
;
4123 ifp
->if_flags
|= XFS_IFEXTIREC
;
4124 ifp
->if_real_bytes
= XFS_IEXT_BUFSZ
;
4125 ifp
->if_bytes
= nextents
* sizeof(xfs_bmbt_rec_t
);
4126 ifp
->if_u1
.if_ext_irec
= erp
;
4132 * Allocate and initialize a new entry in the indirection array.
4136 xfs_ifork_t
*ifp
, /* inode fork pointer */
4137 int erp_idx
) /* index for new irec */
4139 xfs_ext_irec_t
*erp
; /* indirection array pointer */
4140 int i
; /* loop counter */
4141 int nlists
; /* number of irec's (ex lists) */
4143 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4144 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4146 /* Resize indirection array */
4147 xfs_iext_realloc_indirect(ifp
, ++nlists
*
4148 sizeof(xfs_ext_irec_t
));
4150 * Move records down in the array so the
4151 * new page can use erp_idx.
4153 erp
= ifp
->if_u1
.if_ext_irec
;
4154 for (i
= nlists
- 1; i
> erp_idx
; i
--) {
4155 memmove(&erp
[i
], &erp
[i
-1], sizeof(xfs_ext_irec_t
));
4157 ASSERT(i
== erp_idx
);
4159 /* Initialize new extent record */
4160 erp
= ifp
->if_u1
.if_ext_irec
;
4161 erp
[erp_idx
].er_extbuf
= kmem_alloc(XFS_IEXT_BUFSZ
, KM_NOFS
);
4162 ifp
->if_real_bytes
= nlists
* XFS_IEXT_BUFSZ
;
4163 memset(erp
[erp_idx
].er_extbuf
, 0, XFS_IEXT_BUFSZ
);
4164 erp
[erp_idx
].er_extcount
= 0;
4165 erp
[erp_idx
].er_extoff
= erp_idx
> 0 ?
4166 erp
[erp_idx
-1].er_extoff
+ erp
[erp_idx
-1].er_extcount
: 0;
4167 return (&erp
[erp_idx
]);
4171 * Remove a record from the indirection array.
4174 xfs_iext_irec_remove(
4175 xfs_ifork_t
*ifp
, /* inode fork pointer */
4176 int erp_idx
) /* irec index to remove */
4178 xfs_ext_irec_t
*erp
; /* indirection array pointer */
4179 int i
; /* loop counter */
4180 int nlists
; /* number of irec's (ex lists) */
4182 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4183 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4184 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4185 if (erp
->er_extbuf
) {
4186 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1,
4188 kmem_free(erp
->er_extbuf
);
4190 /* Compact extent records */
4191 erp
= ifp
->if_u1
.if_ext_irec
;
4192 for (i
= erp_idx
; i
< nlists
- 1; i
++) {
4193 memmove(&erp
[i
], &erp
[i
+1], sizeof(xfs_ext_irec_t
));
4196 * Manually free the last extent record from the indirection
4197 * array. A call to xfs_iext_realloc_indirect() with a size
4198 * of zero would result in a call to xfs_iext_destroy() which
4199 * would in turn call this function again, creating a nasty
4203 xfs_iext_realloc_indirect(ifp
,
4204 nlists
* sizeof(xfs_ext_irec_t
));
4206 kmem_free(ifp
->if_u1
.if_ext_irec
);
4208 ifp
->if_real_bytes
= nlists
* XFS_IEXT_BUFSZ
;
4212 * This is called to clean up large amounts of unused memory allocated
4213 * by the indirection array. Before compacting anything though, verify
4214 * that the indirection array is still needed and switch back to the
4215 * linear extent list (or even the inline buffer) if possible. The
4216 * compaction policy is as follows:
4218 * Full Compaction: Extents fit into a single page (or inline buffer)
4219 * Partial Compaction: Extents occupy less than 50% of allocated space
4220 * No Compaction: Extents occupy at least 50% of allocated space
4223 xfs_iext_irec_compact(
4224 xfs_ifork_t
*ifp
) /* inode fork pointer */
4226 xfs_extnum_t nextents
; /* number of extents in file */
4227 int nlists
; /* number of irec's (ex lists) */
4229 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4230 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4231 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4233 if (nextents
== 0) {
4234 xfs_iext_destroy(ifp
);
4235 } else if (nextents
<= XFS_INLINE_EXTS
) {
4236 xfs_iext_indirect_to_direct(ifp
);
4237 xfs_iext_direct_to_inline(ifp
, nextents
);
4238 } else if (nextents
<= XFS_LINEAR_EXTS
) {
4239 xfs_iext_indirect_to_direct(ifp
);
4240 } else if (nextents
< (nlists
* XFS_LINEAR_EXTS
) >> 1) {
4241 xfs_iext_irec_compact_pages(ifp
);
4246 * Combine extents from neighboring extent pages.
4249 xfs_iext_irec_compact_pages(
4250 xfs_ifork_t
*ifp
) /* inode fork pointer */
4252 xfs_ext_irec_t
*erp
, *erp_next
;/* pointers to irec entries */
4253 int erp_idx
= 0; /* indirection array index */
4254 int nlists
; /* number of irec's (ex lists) */
4256 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4257 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4258 while (erp_idx
< nlists
- 1) {
4259 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4261 if (erp_next
->er_extcount
<=
4262 (XFS_LINEAR_EXTS
- erp
->er_extcount
)) {
4263 memcpy(&erp
->er_extbuf
[erp
->er_extcount
],
4264 erp_next
->er_extbuf
, erp_next
->er_extcount
*
4265 sizeof(xfs_bmbt_rec_t
));
4266 erp
->er_extcount
+= erp_next
->er_extcount
;
4268 * Free page before removing extent record
4269 * so er_extoffs don't get modified in
4270 * xfs_iext_irec_remove.
4272 kmem_free(erp_next
->er_extbuf
);
4273 erp_next
->er_extbuf
= NULL
;
4274 xfs_iext_irec_remove(ifp
, erp_idx
+ 1);
4275 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4283 * This is called to update the er_extoff field in the indirection
4284 * array when extents have been added or removed from one of the
4285 * extent lists. erp_idx contains the irec index to begin updating
4286 * at and ext_diff contains the number of extents that were added
4290 xfs_iext_irec_update_extoffs(
4291 xfs_ifork_t
*ifp
, /* inode fork pointer */
4292 int erp_idx
, /* irec index to update */
4293 int ext_diff
) /* number of new extents */
4295 int i
; /* loop counter */
4296 int nlists
; /* number of irec's (ex lists */
4298 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4299 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4300 for (i
= erp_idx
; i
< nlists
; i
++) {
4301 ifp
->if_u1
.if_ext_irec
[i
].er_extoff
+= ext_diff
;