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"
30 #include "xfs_mount.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_attr_sf.h"
35 #include "xfs_dinode.h"
36 #include "xfs_inode.h"
37 #include "xfs_buf_item.h"
38 #include "xfs_inode_item.h"
39 #include "xfs_btree.h"
40 #include "xfs_btree_trace.h"
41 #include "xfs_alloc.h"
42 #include "xfs_ialloc.h"
44 #include "xfs_error.h"
45 #include "xfs_utils.h"
46 #include "xfs_quota.h"
47 #include "xfs_filestream.h"
48 #include "xfs_vnodeops.h"
49 #include "xfs_trace.h"
51 kmem_zone_t
*xfs_ifork_zone
;
52 kmem_zone_t
*xfs_inode_zone
;
55 * Used in xfs_itruncate(). This is the maximum number of extents
56 * freed from a file in a single transaction.
58 #define XFS_ITRUNC_MAX_EXTENTS 2
60 STATIC
int xfs_iflush_int(xfs_inode_t
*, xfs_buf_t
*);
61 STATIC
int xfs_iformat_local(xfs_inode_t
*, xfs_dinode_t
*, int, int);
62 STATIC
int xfs_iformat_extents(xfs_inode_t
*, xfs_dinode_t
*, int);
63 STATIC
int xfs_iformat_btree(xfs_inode_t
*, xfs_dinode_t
*, int);
67 * Make sure that the extents in the given memory buffer
77 xfs_bmbt_rec_host_t rec
;
80 for (i
= 0; i
< nrecs
; i
++) {
81 xfs_bmbt_rec_host_t
*ep
= xfs_iext_get_ext(ifp
, i
);
82 rec
.l0
= get_unaligned(&ep
->l0
);
83 rec
.l1
= get_unaligned(&ep
->l1
);
84 xfs_bmbt_get_all(&rec
, &irec
);
85 if (fmt
== XFS_EXTFMT_NOSTATE
)
86 ASSERT(irec
.br_state
== XFS_EXT_NORM
);
90 #define xfs_validate_extents(ifp, nrecs, fmt)
94 * Check that none of the inode's in the buffer have a next
95 * unlinked field of 0.
107 j
= mp
->m_inode_cluster_size
>> mp
->m_sb
.sb_inodelog
;
109 for (i
= 0; i
< j
; i
++) {
110 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
,
111 i
* mp
->m_sb
.sb_inodesize
);
112 if (!dip
->di_next_unlinked
) {
114 "Detected bogus zero next_unlinked field in incore inode buffer 0x%p.",
116 ASSERT(dip
->di_next_unlinked
);
123 * Find the buffer associated with the given inode map
124 * We do basic validation checks on the buffer once it has been
125 * retrieved from disk.
131 struct xfs_imap
*imap
,
141 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, imap
->im_blkno
,
142 (int)imap
->im_len
, buf_flags
, &bp
);
144 if (error
!= EAGAIN
) {
146 "xfs_imap_to_bp: xfs_trans_read_buf()returned "
147 "an error %d on %s. Returning error.",
148 error
, mp
->m_fsname
);
150 ASSERT(buf_flags
& XBF_TRYLOCK
);
156 * Validate the magic number and version of every inode in the buffer
157 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
160 ni
= BBTOB(imap
->im_len
) >> mp
->m_sb
.sb_inodelog
;
161 #else /* usual case */
165 for (i
= 0; i
< ni
; i
++) {
169 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
,
170 (i
<< mp
->m_sb
.sb_inodelog
));
171 di_ok
= be16_to_cpu(dip
->di_magic
) == XFS_DINODE_MAGIC
&&
172 XFS_DINODE_GOOD_VERSION(dip
->di_version
);
173 if (unlikely(XFS_TEST_ERROR(!di_ok
, mp
,
174 XFS_ERRTAG_ITOBP_INOTOBP
,
175 XFS_RANDOM_ITOBP_INOTOBP
))) {
176 if (iget_flags
& XFS_IGET_UNTRUSTED
) {
177 xfs_trans_brelse(tp
, bp
);
178 return XFS_ERROR(EINVAL
);
180 XFS_CORRUPTION_ERROR("xfs_imap_to_bp",
181 XFS_ERRLEVEL_HIGH
, mp
, dip
);
184 "Device %s - bad inode magic/vsn "
185 "daddr %lld #%d (magic=%x)",
186 XFS_BUFTARG_NAME(mp
->m_ddev_targp
),
187 (unsigned long long)imap
->im_blkno
, i
,
188 be16_to_cpu(dip
->di_magic
));
190 xfs_trans_brelse(tp
, bp
);
191 return XFS_ERROR(EFSCORRUPTED
);
195 xfs_inobp_check(mp
, bp
);
198 * Mark the buffer as an inode buffer now that it looks good
200 XFS_BUF_SET_VTYPE(bp
, B_FS_INO
);
207 * This routine is called to map an inode number within a file
208 * system to the buffer containing the on-disk version of the
209 * inode. It returns a pointer to the buffer containing the
210 * on-disk inode in the bpp parameter, and in the dip parameter
211 * it returns a pointer to the on-disk inode within that buffer.
213 * If a non-zero error is returned, then the contents of bpp and
214 * dipp are undefined.
216 * Use xfs_imap() to determine the size and location of the
217 * buffer to read from disk.
229 struct xfs_imap imap
;
234 error
= xfs_imap(mp
, tp
, ino
, &imap
, imap_flags
);
238 error
= xfs_imap_to_bp(mp
, tp
, &imap
, &bp
, XBF_LOCK
, imap_flags
);
242 *dipp
= (xfs_dinode_t
*)xfs_buf_offset(bp
, imap
.im_boffset
);
244 *offset
= imap
.im_boffset
;
250 * This routine is called to map an inode to the buffer containing
251 * the on-disk version of the inode. It returns a pointer to the
252 * buffer containing the on-disk inode in the bpp parameter, and in
253 * the dip parameter it returns a pointer to the on-disk inode within
256 * If a non-zero error is returned, then the contents of bpp and
257 * dipp are undefined.
259 * The inode is expected to already been mapped to its buffer and read
260 * in once, thus we can use the mapping information stored in the inode
261 * rather than calling xfs_imap(). This allows us to avoid the overhead
262 * of looking at the inode btree for small block file systems
277 ASSERT(ip
->i_imap
.im_blkno
!= 0);
279 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &bp
, buf_flags
, 0);
284 ASSERT(buf_flags
& XBF_TRYLOCK
);
290 *dipp
= (xfs_dinode_t
*)xfs_buf_offset(bp
, ip
->i_imap
.im_boffset
);
296 * Move inode type and inode format specific information from the
297 * on-disk inode to the in-core inode. For fifos, devs, and sockets
298 * this means set if_rdev to the proper value. For files, directories,
299 * and symlinks this means to bring in the in-line data or extent
300 * pointers. For a file in B-tree format, only the root is immediately
301 * brought in-core. The rest will be in-lined in if_extents when it
302 * is first referenced (see xfs_iread_extents()).
309 xfs_attr_shortform_t
*atp
;
313 ip
->i_df
.if_ext_max
=
314 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
317 if (unlikely(be32_to_cpu(dip
->di_nextents
) +
318 be16_to_cpu(dip
->di_anextents
) >
319 be64_to_cpu(dip
->di_nblocks
))) {
320 xfs_warn(ip
->i_mount
,
321 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
322 (unsigned long long)ip
->i_ino
,
323 (int)(be32_to_cpu(dip
->di_nextents
) +
324 be16_to_cpu(dip
->di_anextents
)),
326 be64_to_cpu(dip
->di_nblocks
));
327 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW
,
329 return XFS_ERROR(EFSCORRUPTED
);
332 if (unlikely(dip
->di_forkoff
> ip
->i_mount
->m_sb
.sb_inodesize
)) {
333 xfs_warn(ip
->i_mount
, "corrupt dinode %Lu, forkoff = 0x%x.",
334 (unsigned long long)ip
->i_ino
,
336 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW
,
338 return XFS_ERROR(EFSCORRUPTED
);
341 if (unlikely((ip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
) &&
342 !ip
->i_mount
->m_rtdev_targp
)) {
343 xfs_warn(ip
->i_mount
,
344 "corrupt dinode %Lu, has realtime flag set.",
346 XFS_CORRUPTION_ERROR("xfs_iformat(realtime)",
347 XFS_ERRLEVEL_LOW
, ip
->i_mount
, dip
);
348 return XFS_ERROR(EFSCORRUPTED
);
351 switch (ip
->i_d
.di_mode
& S_IFMT
) {
356 if (unlikely(dip
->di_format
!= XFS_DINODE_FMT_DEV
)) {
357 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW
,
359 return XFS_ERROR(EFSCORRUPTED
);
363 ip
->i_df
.if_u2
.if_rdev
= xfs_dinode_get_rdev(dip
);
369 switch (dip
->di_format
) {
370 case XFS_DINODE_FMT_LOCAL
:
372 * no local regular files yet
374 if (unlikely((be16_to_cpu(dip
->di_mode
) & S_IFMT
) == S_IFREG
)) {
375 xfs_warn(ip
->i_mount
,
376 "corrupt inode %Lu (local format for regular file).",
377 (unsigned long long) ip
->i_ino
);
378 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
381 return XFS_ERROR(EFSCORRUPTED
);
384 di_size
= be64_to_cpu(dip
->di_size
);
385 if (unlikely(di_size
> XFS_DFORK_DSIZE(dip
, ip
->i_mount
))) {
386 xfs_warn(ip
->i_mount
,
387 "corrupt inode %Lu (bad size %Ld for local inode).",
388 (unsigned long long) ip
->i_ino
,
389 (long long) di_size
);
390 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
393 return XFS_ERROR(EFSCORRUPTED
);
397 error
= xfs_iformat_local(ip
, dip
, XFS_DATA_FORK
, size
);
399 case XFS_DINODE_FMT_EXTENTS
:
400 error
= xfs_iformat_extents(ip
, dip
, XFS_DATA_FORK
);
402 case XFS_DINODE_FMT_BTREE
:
403 error
= xfs_iformat_btree(ip
, dip
, XFS_DATA_FORK
);
406 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW
,
408 return XFS_ERROR(EFSCORRUPTED
);
413 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW
, ip
->i_mount
);
414 return XFS_ERROR(EFSCORRUPTED
);
419 if (!XFS_DFORK_Q(dip
))
421 ASSERT(ip
->i_afp
== NULL
);
422 ip
->i_afp
= kmem_zone_zalloc(xfs_ifork_zone
, KM_SLEEP
| KM_NOFS
);
423 ip
->i_afp
->if_ext_max
=
424 XFS_IFORK_ASIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
425 switch (dip
->di_aformat
) {
426 case XFS_DINODE_FMT_LOCAL
:
427 atp
= (xfs_attr_shortform_t
*)XFS_DFORK_APTR(dip
);
428 size
= be16_to_cpu(atp
->hdr
.totsize
);
430 if (unlikely(size
< sizeof(struct xfs_attr_sf_hdr
))) {
431 xfs_warn(ip
->i_mount
,
432 "corrupt inode %Lu (bad attr fork size %Ld).",
433 (unsigned long long) ip
->i_ino
,
435 XFS_CORRUPTION_ERROR("xfs_iformat(8)",
438 return XFS_ERROR(EFSCORRUPTED
);
441 error
= xfs_iformat_local(ip
, dip
, XFS_ATTR_FORK
, size
);
443 case XFS_DINODE_FMT_EXTENTS
:
444 error
= xfs_iformat_extents(ip
, dip
, XFS_ATTR_FORK
);
446 case XFS_DINODE_FMT_BTREE
:
447 error
= xfs_iformat_btree(ip
, dip
, XFS_ATTR_FORK
);
450 error
= XFS_ERROR(EFSCORRUPTED
);
454 kmem_zone_free(xfs_ifork_zone
, ip
->i_afp
);
456 xfs_idestroy_fork(ip
, XFS_DATA_FORK
);
462 * The file is in-lined in the on-disk inode.
463 * If it fits into if_inline_data, then copy
464 * it there, otherwise allocate a buffer for it
465 * and copy the data there. Either way, set
466 * if_data to point at the data.
467 * If we allocate a buffer for the data, make
468 * sure that its size is a multiple of 4 and
469 * record the real size in i_real_bytes.
482 * If the size is unreasonable, then something
483 * is wrong and we just bail out rather than crash in
484 * kmem_alloc() or memcpy() below.
486 if (unlikely(size
> XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
))) {
487 xfs_warn(ip
->i_mount
,
488 "corrupt inode %Lu (bad size %d for local fork, size = %d).",
489 (unsigned long long) ip
->i_ino
, size
,
490 XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
));
491 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW
,
493 return XFS_ERROR(EFSCORRUPTED
);
495 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
498 ifp
->if_u1
.if_data
= NULL
;
499 else if (size
<= sizeof(ifp
->if_u2
.if_inline_data
))
500 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
502 real_size
= roundup(size
, 4);
503 ifp
->if_u1
.if_data
= kmem_alloc(real_size
, KM_SLEEP
| KM_NOFS
);
505 ifp
->if_bytes
= size
;
506 ifp
->if_real_bytes
= real_size
;
508 memcpy(ifp
->if_u1
.if_data
, XFS_DFORK_PTR(dip
, whichfork
), size
);
509 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
510 ifp
->if_flags
|= XFS_IFINLINE
;
515 * The file consists of a set of extents all
516 * of which fit into the on-disk inode.
517 * If there are few enough extents to fit into
518 * the if_inline_ext, then copy them there.
519 * Otherwise allocate a buffer for them and copy
520 * them into it. Either way, set if_extents
521 * to point at the extents.
535 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
536 nex
= XFS_DFORK_NEXTENTS(dip
, whichfork
);
537 size
= nex
* (uint
)sizeof(xfs_bmbt_rec_t
);
540 * If the number of extents is unreasonable, then something
541 * is wrong and we just bail out rather than crash in
542 * kmem_alloc() or memcpy() below.
544 if (unlikely(size
< 0 || size
> XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
))) {
545 xfs_warn(ip
->i_mount
, "corrupt inode %Lu ((a)extents = %d).",
546 (unsigned long long) ip
->i_ino
, nex
);
547 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW
,
549 return XFS_ERROR(EFSCORRUPTED
);
552 ifp
->if_real_bytes
= 0;
554 ifp
->if_u1
.if_extents
= NULL
;
555 else if (nex
<= XFS_INLINE_EXTS
)
556 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
558 xfs_iext_add(ifp
, 0, nex
);
560 ifp
->if_bytes
= size
;
562 dp
= (xfs_bmbt_rec_t
*) XFS_DFORK_PTR(dip
, whichfork
);
563 xfs_validate_extents(ifp
, nex
, XFS_EXTFMT_INODE(ip
));
564 for (i
= 0; i
< nex
; i
++, dp
++) {
565 xfs_bmbt_rec_host_t
*ep
= xfs_iext_get_ext(ifp
, i
);
566 ep
->l0
= get_unaligned_be64(&dp
->l0
);
567 ep
->l1
= get_unaligned_be64(&dp
->l1
);
569 XFS_BMAP_TRACE_EXLIST(ip
, nex
, whichfork
);
570 if (whichfork
!= XFS_DATA_FORK
||
571 XFS_EXTFMT_INODE(ip
) == XFS_EXTFMT_NOSTATE
)
572 if (unlikely(xfs_check_nostate_extents(
574 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
577 return XFS_ERROR(EFSCORRUPTED
);
580 ifp
->if_flags
|= XFS_IFEXTENTS
;
585 * The file has too many extents to fit into
586 * the inode, so they are in B-tree format.
587 * Allocate a buffer for the root of the B-tree
588 * and copy the root into it. The i_extents
589 * field will remain NULL until all of the
590 * extents are read in (when they are needed).
598 xfs_bmdr_block_t
*dfp
;
604 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
605 dfp
= (xfs_bmdr_block_t
*)XFS_DFORK_PTR(dip
, whichfork
);
606 size
= XFS_BMAP_BROOT_SPACE(dfp
);
607 nrecs
= be16_to_cpu(dfp
->bb_numrecs
);
610 * blow out if -- fork has less extents than can fit in
611 * fork (fork shouldn't be a btree format), root btree
612 * block has more records than can fit into the fork,
613 * or the number of extents is greater than the number of
616 if (unlikely(XFS_IFORK_NEXTENTS(ip
, whichfork
) <= ifp
->if_ext_max
617 || XFS_BMDR_SPACE_CALC(nrecs
) >
618 XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
)
619 || XFS_IFORK_NEXTENTS(ip
, whichfork
) > ip
->i_d
.di_nblocks
)) {
620 xfs_warn(ip
->i_mount
, "corrupt inode %Lu (btree).",
621 (unsigned long long) ip
->i_ino
);
622 XFS_CORRUPTION_ERROR("xfs_iformat_btree", XFS_ERRLEVEL_LOW
,
624 return XFS_ERROR(EFSCORRUPTED
);
627 ifp
->if_broot_bytes
= size
;
628 ifp
->if_broot
= kmem_alloc(size
, KM_SLEEP
| KM_NOFS
);
629 ASSERT(ifp
->if_broot
!= NULL
);
631 * Copy and convert from the on-disk structure
632 * to the in-memory structure.
634 xfs_bmdr_to_bmbt(ip
->i_mount
, dfp
,
635 XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
),
636 ifp
->if_broot
, size
);
637 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
638 ifp
->if_flags
|= XFS_IFBROOT
;
644 xfs_dinode_from_disk(
648 to
->di_magic
= be16_to_cpu(from
->di_magic
);
649 to
->di_mode
= be16_to_cpu(from
->di_mode
);
650 to
->di_version
= from
->di_version
;
651 to
->di_format
= from
->di_format
;
652 to
->di_onlink
= be16_to_cpu(from
->di_onlink
);
653 to
->di_uid
= be32_to_cpu(from
->di_uid
);
654 to
->di_gid
= be32_to_cpu(from
->di_gid
);
655 to
->di_nlink
= be32_to_cpu(from
->di_nlink
);
656 to
->di_projid_lo
= be16_to_cpu(from
->di_projid_lo
);
657 to
->di_projid_hi
= be16_to_cpu(from
->di_projid_hi
);
658 memcpy(to
->di_pad
, from
->di_pad
, sizeof(to
->di_pad
));
659 to
->di_flushiter
= be16_to_cpu(from
->di_flushiter
);
660 to
->di_atime
.t_sec
= be32_to_cpu(from
->di_atime
.t_sec
);
661 to
->di_atime
.t_nsec
= be32_to_cpu(from
->di_atime
.t_nsec
);
662 to
->di_mtime
.t_sec
= be32_to_cpu(from
->di_mtime
.t_sec
);
663 to
->di_mtime
.t_nsec
= be32_to_cpu(from
->di_mtime
.t_nsec
);
664 to
->di_ctime
.t_sec
= be32_to_cpu(from
->di_ctime
.t_sec
);
665 to
->di_ctime
.t_nsec
= be32_to_cpu(from
->di_ctime
.t_nsec
);
666 to
->di_size
= be64_to_cpu(from
->di_size
);
667 to
->di_nblocks
= be64_to_cpu(from
->di_nblocks
);
668 to
->di_extsize
= be32_to_cpu(from
->di_extsize
);
669 to
->di_nextents
= be32_to_cpu(from
->di_nextents
);
670 to
->di_anextents
= be16_to_cpu(from
->di_anextents
);
671 to
->di_forkoff
= from
->di_forkoff
;
672 to
->di_aformat
= from
->di_aformat
;
673 to
->di_dmevmask
= be32_to_cpu(from
->di_dmevmask
);
674 to
->di_dmstate
= be16_to_cpu(from
->di_dmstate
);
675 to
->di_flags
= be16_to_cpu(from
->di_flags
);
676 to
->di_gen
= be32_to_cpu(from
->di_gen
);
682 xfs_icdinode_t
*from
)
684 to
->di_magic
= cpu_to_be16(from
->di_magic
);
685 to
->di_mode
= cpu_to_be16(from
->di_mode
);
686 to
->di_version
= from
->di_version
;
687 to
->di_format
= from
->di_format
;
688 to
->di_onlink
= cpu_to_be16(from
->di_onlink
);
689 to
->di_uid
= cpu_to_be32(from
->di_uid
);
690 to
->di_gid
= cpu_to_be32(from
->di_gid
);
691 to
->di_nlink
= cpu_to_be32(from
->di_nlink
);
692 to
->di_projid_lo
= cpu_to_be16(from
->di_projid_lo
);
693 to
->di_projid_hi
= cpu_to_be16(from
->di_projid_hi
);
694 memcpy(to
->di_pad
, from
->di_pad
, sizeof(to
->di_pad
));
695 to
->di_flushiter
= cpu_to_be16(from
->di_flushiter
);
696 to
->di_atime
.t_sec
= cpu_to_be32(from
->di_atime
.t_sec
);
697 to
->di_atime
.t_nsec
= cpu_to_be32(from
->di_atime
.t_nsec
);
698 to
->di_mtime
.t_sec
= cpu_to_be32(from
->di_mtime
.t_sec
);
699 to
->di_mtime
.t_nsec
= cpu_to_be32(from
->di_mtime
.t_nsec
);
700 to
->di_ctime
.t_sec
= cpu_to_be32(from
->di_ctime
.t_sec
);
701 to
->di_ctime
.t_nsec
= cpu_to_be32(from
->di_ctime
.t_nsec
);
702 to
->di_size
= cpu_to_be64(from
->di_size
);
703 to
->di_nblocks
= cpu_to_be64(from
->di_nblocks
);
704 to
->di_extsize
= cpu_to_be32(from
->di_extsize
);
705 to
->di_nextents
= cpu_to_be32(from
->di_nextents
);
706 to
->di_anextents
= cpu_to_be16(from
->di_anextents
);
707 to
->di_forkoff
= from
->di_forkoff
;
708 to
->di_aformat
= from
->di_aformat
;
709 to
->di_dmevmask
= cpu_to_be32(from
->di_dmevmask
);
710 to
->di_dmstate
= cpu_to_be16(from
->di_dmstate
);
711 to
->di_flags
= cpu_to_be16(from
->di_flags
);
712 to
->di_gen
= cpu_to_be32(from
->di_gen
);
721 if (di_flags
& XFS_DIFLAG_ANY
) {
722 if (di_flags
& XFS_DIFLAG_REALTIME
)
723 flags
|= XFS_XFLAG_REALTIME
;
724 if (di_flags
& XFS_DIFLAG_PREALLOC
)
725 flags
|= XFS_XFLAG_PREALLOC
;
726 if (di_flags
& XFS_DIFLAG_IMMUTABLE
)
727 flags
|= XFS_XFLAG_IMMUTABLE
;
728 if (di_flags
& XFS_DIFLAG_APPEND
)
729 flags
|= XFS_XFLAG_APPEND
;
730 if (di_flags
& XFS_DIFLAG_SYNC
)
731 flags
|= XFS_XFLAG_SYNC
;
732 if (di_flags
& XFS_DIFLAG_NOATIME
)
733 flags
|= XFS_XFLAG_NOATIME
;
734 if (di_flags
& XFS_DIFLAG_NODUMP
)
735 flags
|= XFS_XFLAG_NODUMP
;
736 if (di_flags
& XFS_DIFLAG_RTINHERIT
)
737 flags
|= XFS_XFLAG_RTINHERIT
;
738 if (di_flags
& XFS_DIFLAG_PROJINHERIT
)
739 flags
|= XFS_XFLAG_PROJINHERIT
;
740 if (di_flags
& XFS_DIFLAG_NOSYMLINKS
)
741 flags
|= XFS_XFLAG_NOSYMLINKS
;
742 if (di_flags
& XFS_DIFLAG_EXTSIZE
)
743 flags
|= XFS_XFLAG_EXTSIZE
;
744 if (di_flags
& XFS_DIFLAG_EXTSZINHERIT
)
745 flags
|= XFS_XFLAG_EXTSZINHERIT
;
746 if (di_flags
& XFS_DIFLAG_NODEFRAG
)
747 flags
|= XFS_XFLAG_NODEFRAG
;
748 if (di_flags
& XFS_DIFLAG_FILESTREAM
)
749 flags
|= XFS_XFLAG_FILESTREAM
;
759 xfs_icdinode_t
*dic
= &ip
->i_d
;
761 return _xfs_dic2xflags(dic
->di_flags
) |
762 (XFS_IFORK_Q(ip
) ? XFS_XFLAG_HASATTR
: 0);
769 return _xfs_dic2xflags(be16_to_cpu(dip
->di_flags
)) |
770 (XFS_DFORK_Q(dip
) ? XFS_XFLAG_HASATTR
: 0);
774 * Read the disk inode attributes into the in-core inode structure.
788 * Fill in the location information in the in-core inode.
790 error
= xfs_imap(mp
, tp
, ip
->i_ino
, &ip
->i_imap
, iget_flags
);
795 * Get pointers to the on-disk inode and the buffer containing it.
797 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &bp
,
798 XBF_LOCK
, iget_flags
);
801 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, ip
->i_imap
.im_boffset
);
804 * If we got something that isn't an inode it means someone
805 * (nfs or dmi) has a stale handle.
807 if (be16_to_cpu(dip
->di_magic
) != XFS_DINODE_MAGIC
) {
810 "%s: dip->di_magic (0x%x) != XFS_DINODE_MAGIC (0x%x)",
811 __func__
, be16_to_cpu(dip
->di_magic
), XFS_DINODE_MAGIC
);
813 error
= XFS_ERROR(EINVAL
);
818 * If the on-disk inode is already linked to a directory
819 * entry, copy all of the inode into the in-core inode.
820 * xfs_iformat() handles copying in the inode format
821 * specific information.
822 * Otherwise, just get the truly permanent information.
825 xfs_dinode_from_disk(&ip
->i_d
, dip
);
826 error
= xfs_iformat(ip
, dip
);
829 xfs_alert(mp
, "%s: xfs_iformat() returned error %d",
835 ip
->i_d
.di_magic
= be16_to_cpu(dip
->di_magic
);
836 ip
->i_d
.di_version
= dip
->di_version
;
837 ip
->i_d
.di_gen
= be32_to_cpu(dip
->di_gen
);
838 ip
->i_d
.di_flushiter
= be16_to_cpu(dip
->di_flushiter
);
840 * Make sure to pull in the mode here as well in
841 * case the inode is released without being used.
842 * This ensures that xfs_inactive() will see that
843 * the inode is already free and not try to mess
844 * with the uninitialized part of it.
848 * Initialize the per-fork minima and maxima for a new
849 * inode here. xfs_iformat will do it for old inodes.
851 ip
->i_df
.if_ext_max
=
852 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
856 * The inode format changed when we moved the link count and
857 * made it 32 bits long. If this is an old format inode,
858 * convert it in memory to look like a new one. If it gets
859 * flushed to disk we will convert back before flushing or
860 * logging it. We zero out the new projid field and the old link
861 * count field. We'll handle clearing the pad field (the remains
862 * of the old uuid field) when we actually convert the inode to
863 * the new format. We don't change the version number so that we
864 * can distinguish this from a real new format inode.
866 if (ip
->i_d
.di_version
== 1) {
867 ip
->i_d
.di_nlink
= ip
->i_d
.di_onlink
;
868 ip
->i_d
.di_onlink
= 0;
869 xfs_set_projid(ip
, 0);
872 ip
->i_delayed_blks
= 0;
873 ip
->i_size
= ip
->i_d
.di_size
;
876 * Mark the buffer containing the inode as something to keep
877 * around for a while. This helps to keep recently accessed
878 * meta-data in-core longer.
880 xfs_buf_set_ref(bp
, XFS_INO_REF
);
883 * Use xfs_trans_brelse() to release the buffer containing the
884 * on-disk inode, because it was acquired with xfs_trans_read_buf()
885 * in xfs_itobp() above. If tp is NULL, this is just a normal
886 * brelse(). If we're within a transaction, then xfs_trans_brelse()
887 * will only release the buffer if it is not dirty within the
888 * transaction. It will be OK to release the buffer in this case,
889 * because inodes on disk are never destroyed and we will be
890 * locking the new in-core inode before putting it in the hash
891 * table where other processes can find it. Thus we don't have
892 * to worry about the inode being changed just because we released
896 xfs_trans_brelse(tp
, bp
);
901 * Read in extents from a btree-format inode.
902 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
912 xfs_extnum_t nextents
;
914 if (unlikely(XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
)) {
915 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW
,
917 return XFS_ERROR(EFSCORRUPTED
);
919 nextents
= XFS_IFORK_NEXTENTS(ip
, whichfork
);
920 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
923 * We know that the size is valid (it's checked in iformat_btree)
925 ifp
->if_lastex
= NULLEXTNUM
;
926 ifp
->if_bytes
= ifp
->if_real_bytes
= 0;
927 ifp
->if_flags
|= XFS_IFEXTENTS
;
928 xfs_iext_add(ifp
, 0, nextents
);
929 error
= xfs_bmap_read_extents(tp
, ip
, whichfork
);
931 xfs_iext_destroy(ifp
);
932 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
935 xfs_validate_extents(ifp
, nextents
, XFS_EXTFMT_INODE(ip
));
940 * Allocate an inode on disk and return a copy of its in-core version.
941 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
942 * appropriately within the inode. The uid and gid for the inode are
943 * set according to the contents of the given cred structure.
945 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
946 * has a free inode available, call xfs_iget()
947 * to obtain the in-core version of the allocated inode. Finally,
948 * fill in the inode and log its initial contents. In this case,
949 * ialloc_context would be set to NULL and call_again set to false.
951 * If xfs_dialloc() does not have an available inode,
952 * it will replenish its supply by doing an allocation. Since we can
953 * only do one allocation within a transaction without deadlocks, we
954 * must commit the current transaction before returning the inode itself.
955 * In this case, therefore, we will set call_again to true and return.
956 * The caller should then commit the current transaction, start a new
957 * transaction, and call xfs_ialloc() again to actually get the inode.
959 * To ensure that some other process does not grab the inode that
960 * was allocated during the first call to xfs_ialloc(), this routine
961 * also returns the [locked] bp pointing to the head of the freelist
962 * as ialloc_context. The caller should hold this buffer across
963 * the commit and pass it back into this routine on the second call.
965 * If we are allocating quota inodes, we do not have a parent inode
966 * to attach to or associate with (i.e. pip == NULL) because they
967 * are not linked into the directory structure - they are attached
968 * directly to the superblock - and so have no parent.
979 xfs_buf_t
**ialloc_context
,
980 boolean_t
*call_again
,
991 * Call the space management code to pick
992 * the on-disk inode to be allocated.
994 error
= xfs_dialloc(tp
, pip
? pip
->i_ino
: 0, mode
, okalloc
,
995 ialloc_context
, call_again
, &ino
);
998 if (*call_again
|| ino
== NULLFSINO
) {
1002 ASSERT(*ialloc_context
== NULL
);
1005 * Get the in-core inode with the lock held exclusively.
1006 * This is because we're setting fields here we need
1007 * to prevent others from looking at until we're done.
1009 error
= xfs_iget(tp
->t_mountp
, tp
, ino
, XFS_IGET_CREATE
,
1010 XFS_ILOCK_EXCL
, &ip
);
1015 ip
->i_d
.di_mode
= (__uint16_t
)mode
;
1016 ip
->i_d
.di_onlink
= 0;
1017 ip
->i_d
.di_nlink
= nlink
;
1018 ASSERT(ip
->i_d
.di_nlink
== nlink
);
1019 ip
->i_d
.di_uid
= current_fsuid();
1020 ip
->i_d
.di_gid
= current_fsgid();
1021 xfs_set_projid(ip
, prid
);
1022 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
1025 * If the superblock version is up to where we support new format
1026 * inodes and this is currently an old format inode, then change
1027 * the inode version number now. This way we only do the conversion
1028 * here rather than here and in the flush/logging code.
1030 if (xfs_sb_version_hasnlink(&tp
->t_mountp
->m_sb
) &&
1031 ip
->i_d
.di_version
== 1) {
1032 ip
->i_d
.di_version
= 2;
1034 * We've already zeroed the old link count, the projid field,
1035 * and the pad field.
1040 * Project ids won't be stored on disk if we are using a version 1 inode.
1042 if ((prid
!= 0) && (ip
->i_d
.di_version
== 1))
1043 xfs_bump_ino_vers2(tp
, ip
);
1045 if (pip
&& XFS_INHERIT_GID(pip
)) {
1046 ip
->i_d
.di_gid
= pip
->i_d
.di_gid
;
1047 if ((pip
->i_d
.di_mode
& S_ISGID
) && (mode
& S_IFMT
) == S_IFDIR
) {
1048 ip
->i_d
.di_mode
|= S_ISGID
;
1053 * If the group ID of the new file does not match the effective group
1054 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1055 * (and only if the irix_sgid_inherit compatibility variable is set).
1057 if ((irix_sgid_inherit
) &&
1058 (ip
->i_d
.di_mode
& S_ISGID
) &&
1059 (!in_group_p((gid_t
)ip
->i_d
.di_gid
))) {
1060 ip
->i_d
.di_mode
&= ~S_ISGID
;
1063 ip
->i_d
.di_size
= 0;
1065 ip
->i_d
.di_nextents
= 0;
1066 ASSERT(ip
->i_d
.di_nblocks
== 0);
1069 ip
->i_d
.di_mtime
.t_sec
= (__int32_t
)tv
.tv_sec
;
1070 ip
->i_d
.di_mtime
.t_nsec
= (__int32_t
)tv
.tv_nsec
;
1071 ip
->i_d
.di_atime
= ip
->i_d
.di_mtime
;
1072 ip
->i_d
.di_ctime
= ip
->i_d
.di_mtime
;
1075 * di_gen will have been taken care of in xfs_iread.
1077 ip
->i_d
.di_extsize
= 0;
1078 ip
->i_d
.di_dmevmask
= 0;
1079 ip
->i_d
.di_dmstate
= 0;
1080 ip
->i_d
.di_flags
= 0;
1081 flags
= XFS_ILOG_CORE
;
1082 switch (mode
& S_IFMT
) {
1087 ip
->i_d
.di_format
= XFS_DINODE_FMT_DEV
;
1088 ip
->i_df
.if_u2
.if_rdev
= rdev
;
1089 ip
->i_df
.if_flags
= 0;
1090 flags
|= XFS_ILOG_DEV
;
1094 * we can't set up filestreams until after the VFS inode
1095 * is set up properly.
1097 if (pip
&& xfs_inode_is_filestream(pip
))
1101 if (pip
&& (pip
->i_d
.di_flags
& XFS_DIFLAG_ANY
)) {
1104 if ((mode
& S_IFMT
) == S_IFDIR
) {
1105 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
1106 di_flags
|= XFS_DIFLAG_RTINHERIT
;
1107 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
1108 di_flags
|= XFS_DIFLAG_EXTSZINHERIT
;
1109 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
1111 } else if ((mode
& S_IFMT
) == S_IFREG
) {
1112 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
1113 di_flags
|= XFS_DIFLAG_REALTIME
;
1114 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
1115 di_flags
|= XFS_DIFLAG_EXTSIZE
;
1116 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
1119 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOATIME
) &&
1120 xfs_inherit_noatime
)
1121 di_flags
|= XFS_DIFLAG_NOATIME
;
1122 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODUMP
) &&
1124 di_flags
|= XFS_DIFLAG_NODUMP
;
1125 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_SYNC
) &&
1127 di_flags
|= XFS_DIFLAG_SYNC
;
1128 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOSYMLINKS
) &&
1129 xfs_inherit_nosymlinks
)
1130 di_flags
|= XFS_DIFLAG_NOSYMLINKS
;
1131 if (pip
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
1132 di_flags
|= XFS_DIFLAG_PROJINHERIT
;
1133 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODEFRAG
) &&
1134 xfs_inherit_nodefrag
)
1135 di_flags
|= XFS_DIFLAG_NODEFRAG
;
1136 if (pip
->i_d
.di_flags
& XFS_DIFLAG_FILESTREAM
)
1137 di_flags
|= XFS_DIFLAG_FILESTREAM
;
1138 ip
->i_d
.di_flags
|= di_flags
;
1142 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
1143 ip
->i_df
.if_flags
= XFS_IFEXTENTS
;
1144 ip
->i_df
.if_bytes
= ip
->i_df
.if_real_bytes
= 0;
1145 ip
->i_df
.if_u1
.if_extents
= NULL
;
1151 * Attribute fork settings for new inode.
1153 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
1154 ip
->i_d
.di_anextents
= 0;
1157 * Log the new values stuffed into the inode.
1159 xfs_trans_ijoin_ref(tp
, ip
, XFS_ILOCK_EXCL
);
1160 xfs_trans_log_inode(tp
, ip
, flags
);
1162 /* now that we have an i_mode we can setup inode ops and unlock */
1163 xfs_setup_inode(ip
);
1165 /* now we have set up the vfs inode we can associate the filestream */
1167 error
= xfs_filestream_associate(pip
, ip
);
1171 xfs_iflags_set(ip
, XFS_IFILESTREAM
);
1179 * Check to make sure that there are no blocks allocated to the
1180 * file beyond the size of the file. We don't check this for
1181 * files with fixed size extents or real time extents, but we
1182 * at least do it for regular files.
1191 xfs_fileoff_t map_first
;
1193 xfs_bmbt_irec_t imaps
[2];
1195 if ((ip
->i_d
.di_mode
& S_IFMT
) != S_IFREG
)
1198 if (XFS_IS_REALTIME_INODE(ip
))
1201 if (ip
->i_d
.di_flags
& XFS_DIFLAG_EXTSIZE
)
1205 map_first
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)isize
);
1207 * The filesystem could be shutting down, so bmapi may return
1210 if (xfs_bmapi(NULL
, ip
, map_first
,
1212 (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
)) -
1214 XFS_BMAPI_ENTIRE
, NULL
, 0, imaps
, &nimaps
,
1217 ASSERT(nimaps
== 1);
1218 ASSERT(imaps
[0].br_startblock
== HOLESTARTBLOCK
);
1223 * Calculate the last possible buffered byte in a file. This must
1224 * include data that was buffered beyond the EOF by the write code.
1225 * This also needs to deal with overflowing the xfs_fsize_t type
1226 * which can happen for sizes near the limit.
1228 * We also need to take into account any blocks beyond the EOF. It
1229 * may be the case that they were buffered by a write which failed.
1230 * In that case the pages will still be in memory, but the inode size
1231 * will never have been updated.
1238 xfs_fsize_t last_byte
;
1239 xfs_fileoff_t last_block
;
1240 xfs_fileoff_t size_last_block
;
1243 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
|XFS_IOLOCK_SHARED
));
1247 * Only check for blocks beyond the EOF if the extents have
1248 * been read in. This eliminates the need for the inode lock,
1249 * and it also saves us from looking when it really isn't
1252 if (ip
->i_df
.if_flags
& XFS_IFEXTENTS
) {
1253 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
1254 error
= xfs_bmap_last_offset(NULL
, ip
, &last_block
,
1256 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
1263 size_last_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)ip
->i_size
);
1264 last_block
= XFS_FILEOFF_MAX(last_block
, size_last_block
);
1266 last_byte
= XFS_FSB_TO_B(mp
, last_block
);
1267 if (last_byte
< 0) {
1268 return XFS_MAXIOFFSET(mp
);
1270 last_byte
+= (1 << mp
->m_writeio_log
);
1271 if (last_byte
< 0) {
1272 return XFS_MAXIOFFSET(mp
);
1278 * Start the truncation of the file to new_size. The new size
1279 * must be smaller than the current size. This routine will
1280 * clear the buffer and page caches of file data in the removed
1281 * range, and xfs_itruncate_finish() will remove the underlying
1284 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1285 * must NOT have the inode lock held at all. This is because we're
1286 * calling into the buffer/page cache code and we can't hold the
1287 * inode lock when we do so.
1289 * We need to wait for any direct I/Os in flight to complete before we
1290 * proceed with the truncate. This is needed to prevent the extents
1291 * being read or written by the direct I/Os from being removed while the
1292 * I/O is in flight as there is no other method of synchronising
1293 * direct I/O with the truncate operation. Also, because we hold
1294 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1295 * started until the truncate completes and drops the lock. Essentially,
1296 * the xfs_ioend_wait() call forms an I/O barrier that provides strict
1297 * ordering between direct I/Os and the truncate operation.
1299 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1300 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1301 * in the case that the caller is locking things out of order and
1302 * may not be able to call xfs_itruncate_finish() with the inode lock
1303 * held without dropping the I/O lock. If the caller must drop the
1304 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1305 * must be called again with all the same restrictions as the initial
1309 xfs_itruncate_start(
1312 xfs_fsize_t new_size
)
1314 xfs_fsize_t last_byte
;
1315 xfs_off_t toss_start
;
1319 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
1320 ASSERT((new_size
== 0) || (new_size
<= ip
->i_size
));
1321 ASSERT((flags
== XFS_ITRUNC_DEFINITE
) ||
1322 (flags
== XFS_ITRUNC_MAYBE
));
1326 /* wait for the completion of any pending DIOs */
1327 if (new_size
== 0 || new_size
< ip
->i_size
)
1331 * Call toss_pages or flushinval_pages to get rid of pages
1332 * overlapping the region being removed. We have to use
1333 * the less efficient flushinval_pages in the case that the
1334 * caller may not be able to finish the truncate without
1335 * dropping the inode's I/O lock. Make sure
1336 * to catch any pages brought in by buffers overlapping
1337 * the EOF by searching out beyond the isize by our
1338 * block size. We round new_size up to a block boundary
1339 * so that we don't toss things on the same block as
1340 * new_size but before it.
1342 * Before calling toss_page or flushinval_pages, make sure to
1343 * call remapf() over the same region if the file is mapped.
1344 * This frees up mapped file references to the pages in the
1345 * given range and for the flushinval_pages case it ensures
1346 * that we get the latest mapped changes flushed out.
1348 toss_start
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)new_size
);
1349 toss_start
= XFS_FSB_TO_B(mp
, toss_start
);
1350 if (toss_start
< 0) {
1352 * The place to start tossing is beyond our maximum
1353 * file size, so there is no way that the data extended
1358 last_byte
= xfs_file_last_byte(ip
);
1359 trace_xfs_itruncate_start(ip
, flags
, new_size
, toss_start
, last_byte
);
1360 if (last_byte
> toss_start
) {
1361 if (flags
& XFS_ITRUNC_DEFINITE
) {
1362 xfs_tosspages(ip
, toss_start
,
1363 -1, FI_REMAPF_LOCKED
);
1365 error
= xfs_flushinval_pages(ip
, toss_start
,
1366 -1, FI_REMAPF_LOCKED
);
1371 if (new_size
== 0) {
1372 ASSERT(VN_CACHED(VFS_I(ip
)) == 0);
1379 * Shrink the file to the given new_size. The new size must be smaller than
1380 * the current size. This will free up the underlying blocks in the removed
1381 * range after a call to xfs_itruncate_start() or xfs_atruncate_start().
1383 * The transaction passed to this routine must have made a permanent log
1384 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1385 * given transaction and start new ones, so make sure everything involved in
1386 * the transaction is tidy before calling here. Some transaction will be
1387 * returned to the caller to be committed. The incoming transaction must
1388 * already include the inode, and both inode locks must be held exclusively.
1389 * The inode must also be "held" within the transaction. On return the inode
1390 * will be "held" within the returned transaction. This routine does NOT
1391 * require any disk space to be reserved for it within the transaction.
1393 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it
1394 * indicates the fork which is to be truncated. For the attribute fork we only
1395 * support truncation to size 0.
1397 * We use the sync parameter to indicate whether or not the first transaction
1398 * we perform might have to be synchronous. For the attr fork, it needs to be
1399 * so if the unlink of the inode is not yet known to be permanent in the log.
1400 * This keeps us from freeing and reusing the blocks of the attribute fork
1401 * before the unlink of the inode becomes permanent.
1403 * For the data fork, we normally have to run synchronously if we're being
1404 * called out of the inactive path or we're being called out of the create path
1405 * where we're truncating an existing file. Either way, the truncate needs to
1406 * be sync so blocks don't reappear in the file with altered data in case of a
1407 * crash. wsync filesystems can run the first case async because anything that
1408 * shrinks the inode has to run sync so by the time we're called here from
1409 * inactive, the inode size is permanently set to 0.
1411 * Calls from the truncate path always need to be sync unless we're in a wsync
1412 * filesystem and the file has already been unlinked.
1414 * The caller is responsible for correctly setting the sync parameter. It gets
1415 * too hard for us to guess here which path we're being called out of just
1416 * based on inode state.
1418 * If we get an error, we must return with the inode locked and linked into the
1419 * current transaction. This keeps things simple for the higher level code,
1420 * because it always knows that the inode is locked and held in the transaction
1421 * that returns to it whether errors occur or not. We don't mark the inode
1422 * dirty on error so that transactions can be easily aborted if possible.
1425 xfs_itruncate_finish(
1428 xfs_fsize_t new_size
,
1432 xfs_fsblock_t first_block
;
1433 xfs_fileoff_t first_unmap_block
;
1434 xfs_fileoff_t last_block
;
1435 xfs_filblks_t unmap_len
=0;
1440 xfs_bmap_free_t free_list
;
1443 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_IOLOCK_EXCL
));
1444 ASSERT((new_size
== 0) || (new_size
<= ip
->i_size
));
1445 ASSERT(*tp
!= NULL
);
1446 ASSERT((*tp
)->t_flags
& XFS_TRANS_PERM_LOG_RES
);
1447 ASSERT(ip
->i_transp
== *tp
);
1448 ASSERT(ip
->i_itemp
!= NULL
);
1449 ASSERT(ip
->i_itemp
->ili_lock_flags
== 0);
1453 mp
= (ntp
)->t_mountp
;
1454 ASSERT(! XFS_NOT_DQATTACHED(mp
, ip
));
1457 * We only support truncating the entire attribute fork.
1459 if (fork
== XFS_ATTR_FORK
) {
1462 first_unmap_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)new_size
);
1463 trace_xfs_itruncate_finish_start(ip
, new_size
);
1466 * The first thing we do is set the size to new_size permanently
1467 * on disk. This way we don't have to worry about anyone ever
1468 * being able to look at the data being freed even in the face
1469 * of a crash. What we're getting around here is the case where
1470 * we free a block, it is allocated to another file, it is written
1471 * to, and then we crash. If the new data gets written to the
1472 * file but the log buffers containing the free and reallocation
1473 * don't, then we'd end up with garbage in the blocks being freed.
1474 * As long as we make the new_size permanent before actually
1475 * freeing any blocks it doesn't matter if they get writtten to.
1477 * The callers must signal into us whether or not the size
1478 * setting here must be synchronous. There are a few cases
1479 * where it doesn't have to be synchronous. Those cases
1480 * occur if the file is unlinked and we know the unlink is
1481 * permanent or if the blocks being truncated are guaranteed
1482 * to be beyond the inode eof (regardless of the link count)
1483 * and the eof value is permanent. Both of these cases occur
1484 * only on wsync-mounted filesystems. In those cases, we're
1485 * guaranteed that no user will ever see the data in the blocks
1486 * that are being truncated so the truncate can run async.
1487 * In the free beyond eof case, the file may wind up with
1488 * more blocks allocated to it than it needs if we crash
1489 * and that won't get fixed until the next time the file
1490 * is re-opened and closed but that's ok as that shouldn't
1491 * be too many blocks.
1493 * However, we can't just make all wsync xactions run async
1494 * because there's one call out of the create path that needs
1495 * to run sync where it's truncating an existing file to size
1496 * 0 whose size is > 0.
1498 * It's probably possible to come up with a test in this
1499 * routine that would correctly distinguish all the above
1500 * cases from the values of the function parameters and the
1501 * inode state but for sanity's sake, I've decided to let the
1502 * layers above just tell us. It's simpler to correctly figure
1503 * out in the layer above exactly under what conditions we
1504 * can run async and I think it's easier for others read and
1505 * follow the logic in case something has to be changed.
1506 * cscope is your friend -- rcc.
1508 * The attribute fork is much simpler.
1510 * For the attribute fork we allow the caller to tell us whether
1511 * the unlink of the inode that led to this call is yet permanent
1512 * in the on disk log. If it is not and we will be freeing extents
1513 * in this inode then we make the first transaction synchronous
1514 * to make sure that the unlink is permanent by the time we free
1517 if (fork
== XFS_DATA_FORK
) {
1518 if (ip
->i_d
.di_nextents
> 0) {
1520 * If we are not changing the file size then do
1521 * not update the on-disk file size - we may be
1522 * called from xfs_inactive_free_eofblocks(). If we
1523 * update the on-disk file size and then the system
1524 * crashes before the contents of the file are
1525 * flushed to disk then the files may be full of
1526 * holes (ie NULL files bug).
1528 if (ip
->i_size
!= new_size
) {
1529 ip
->i_d
.di_size
= new_size
;
1530 ip
->i_size
= new_size
;
1531 xfs_trans_log_inode(ntp
, ip
, XFS_ILOG_CORE
);
1535 ASSERT(!(mp
->m_flags
& XFS_MOUNT_WSYNC
));
1536 if (ip
->i_d
.di_anextents
> 0)
1537 xfs_trans_set_sync(ntp
);
1539 ASSERT(fork
== XFS_DATA_FORK
||
1540 (fork
== XFS_ATTR_FORK
&&
1541 ((sync
&& !(mp
->m_flags
& XFS_MOUNT_WSYNC
)) ||
1542 (sync
== 0 && (mp
->m_flags
& XFS_MOUNT_WSYNC
)))));
1545 * Since it is possible for space to become allocated beyond
1546 * the end of the file (in a crash where the space is allocated
1547 * but the inode size is not yet updated), simply remove any
1548 * blocks which show up between the new EOF and the maximum
1549 * possible file size. If the first block to be removed is
1550 * beyond the maximum file size (ie it is the same as last_block),
1551 * then there is nothing to do.
1553 last_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
));
1554 ASSERT(first_unmap_block
<= last_block
);
1556 if (last_block
== first_unmap_block
) {
1559 unmap_len
= last_block
- first_unmap_block
+ 1;
1563 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1564 * will tell us whether it freed the entire range or
1565 * not. If this is a synchronous mount (wsync),
1566 * then we can tell bunmapi to keep all the
1567 * transactions asynchronous since the unlink
1568 * transaction that made this inode inactive has
1569 * already hit the disk. There's no danger of
1570 * the freed blocks being reused, there being a
1571 * crash, and the reused blocks suddenly reappearing
1572 * in this file with garbage in them once recovery
1575 xfs_bmap_init(&free_list
, &first_block
);
1576 error
= xfs_bunmapi(ntp
, ip
,
1577 first_unmap_block
, unmap_len
,
1578 xfs_bmapi_aflag(fork
),
1579 XFS_ITRUNC_MAX_EXTENTS
,
1580 &first_block
, &free_list
,
1584 * If the bunmapi call encounters an error,
1585 * return to the caller where the transaction
1586 * can be properly aborted. We just need to
1587 * make sure we're not holding any resources
1588 * that we were not when we came in.
1590 xfs_bmap_cancel(&free_list
);
1595 * Duplicate the transaction that has the permanent
1596 * reservation and commit the old transaction.
1598 error
= xfs_bmap_finish(tp
, &free_list
, &committed
);
1601 xfs_trans_ijoin(ntp
, ip
);
1605 * If the bmap finish call encounters an error, return
1606 * to the caller where the transaction can be properly
1607 * aborted. We just need to make sure we're not
1608 * holding any resources that we were not when we came
1611 * Aborting from this point might lose some blocks in
1612 * the file system, but oh well.
1614 xfs_bmap_cancel(&free_list
);
1620 * Mark the inode dirty so it will be logged and
1621 * moved forward in the log as part of every commit.
1623 xfs_trans_log_inode(ntp
, ip
, XFS_ILOG_CORE
);
1626 ntp
= xfs_trans_dup(ntp
);
1627 error
= xfs_trans_commit(*tp
, 0);
1630 xfs_trans_ijoin(ntp
, ip
);
1635 * transaction commit worked ok so we can drop the extra ticket
1636 * reference that we gained in xfs_trans_dup()
1638 xfs_log_ticket_put(ntp
->t_ticket
);
1639 error
= xfs_trans_reserve(ntp
, 0,
1640 XFS_ITRUNCATE_LOG_RES(mp
), 0,
1641 XFS_TRANS_PERM_LOG_RES
,
1642 XFS_ITRUNCATE_LOG_COUNT
);
1647 * Only update the size in the case of the data fork, but
1648 * always re-log the inode so that our permanent transaction
1649 * can keep on rolling it forward in the log.
1651 if (fork
== XFS_DATA_FORK
) {
1652 xfs_isize_check(mp
, ip
, new_size
);
1654 * If we are not changing the file size then do
1655 * not update the on-disk file size - we may be
1656 * called from xfs_inactive_free_eofblocks(). If we
1657 * update the on-disk file size and then the system
1658 * crashes before the contents of the file are
1659 * flushed to disk then the files may be full of
1660 * holes (ie NULL files bug).
1662 if (ip
->i_size
!= new_size
) {
1663 ip
->i_d
.di_size
= new_size
;
1664 ip
->i_size
= new_size
;
1667 xfs_trans_log_inode(ntp
, ip
, XFS_ILOG_CORE
);
1668 ASSERT((new_size
!= 0) ||
1669 (fork
== XFS_ATTR_FORK
) ||
1670 (ip
->i_delayed_blks
== 0));
1671 ASSERT((new_size
!= 0) ||
1672 (fork
== XFS_ATTR_FORK
) ||
1673 (ip
->i_d
.di_nextents
== 0));
1674 trace_xfs_itruncate_finish_end(ip
, new_size
);
1679 * This is called when the inode's link count goes to 0.
1680 * We place the on-disk inode on a list in the AGI. It
1681 * will be pulled from this list when the inode is freed.
1698 ASSERT(ip
->i_d
.di_nlink
== 0);
1699 ASSERT(ip
->i_d
.di_mode
!= 0);
1700 ASSERT(ip
->i_transp
== tp
);
1705 * Get the agi buffer first. It ensures lock ordering
1708 error
= xfs_read_agi(mp
, tp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
), &agibp
);
1711 agi
= XFS_BUF_TO_AGI(agibp
);
1714 * Get the index into the agi hash table for the
1715 * list this inode will go on.
1717 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
1719 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
1720 ASSERT(agi
->agi_unlinked
[bucket_index
]);
1721 ASSERT(be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != agino
);
1723 if (be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != NULLAGINO
) {
1725 * There is already another inode in the bucket we need
1726 * to add ourselves to. Add us at the front of the list.
1727 * Here we put the head pointer into our next pointer,
1728 * and then we fall through to point the head at us.
1730 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, XBF_LOCK
);
1734 ASSERT(be32_to_cpu(dip
->di_next_unlinked
) == NULLAGINO
);
1735 /* both on-disk, don't endian flip twice */
1736 dip
->di_next_unlinked
= agi
->agi_unlinked
[bucket_index
];
1737 offset
= ip
->i_imap
.im_boffset
+
1738 offsetof(xfs_dinode_t
, di_next_unlinked
);
1739 xfs_trans_inode_buf(tp
, ibp
);
1740 xfs_trans_log_buf(tp
, ibp
, offset
,
1741 (offset
+ sizeof(xfs_agino_t
) - 1));
1742 xfs_inobp_check(mp
, ibp
);
1746 * Point the bucket head pointer at the inode being inserted.
1749 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(agino
);
1750 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
1751 (sizeof(xfs_agino_t
) * bucket_index
);
1752 xfs_trans_log_buf(tp
, agibp
, offset
,
1753 (offset
+ sizeof(xfs_agino_t
) - 1));
1758 * Pull the on-disk inode from the AGI unlinked list.
1771 xfs_agnumber_t agno
;
1773 xfs_agino_t next_agino
;
1774 xfs_buf_t
*last_ibp
;
1775 xfs_dinode_t
*last_dip
= NULL
;
1777 int offset
, last_offset
= 0;
1781 agno
= XFS_INO_TO_AGNO(mp
, ip
->i_ino
);
1784 * Get the agi buffer first. It ensures lock ordering
1787 error
= xfs_read_agi(mp
, tp
, agno
, &agibp
);
1791 agi
= XFS_BUF_TO_AGI(agibp
);
1794 * Get the index into the agi hash table for the
1795 * list this inode will go on.
1797 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
1799 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
1800 ASSERT(be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != NULLAGINO
);
1801 ASSERT(agi
->agi_unlinked
[bucket_index
]);
1803 if (be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) == agino
) {
1805 * We're at the head of the list. Get the inode's
1806 * on-disk buffer to see if there is anyone after us
1807 * on the list. Only modify our next pointer if it
1808 * is not already NULLAGINO. This saves us the overhead
1809 * of dealing with the buffer when there is no need to
1812 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, XBF_LOCK
);
1815 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
1816 error
, mp
->m_fsname
);
1819 next_agino
= be32_to_cpu(dip
->di_next_unlinked
);
1820 ASSERT(next_agino
!= 0);
1821 if (next_agino
!= NULLAGINO
) {
1822 dip
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
1823 offset
= ip
->i_imap
.im_boffset
+
1824 offsetof(xfs_dinode_t
, di_next_unlinked
);
1825 xfs_trans_inode_buf(tp
, ibp
);
1826 xfs_trans_log_buf(tp
, ibp
, offset
,
1827 (offset
+ sizeof(xfs_agino_t
) - 1));
1828 xfs_inobp_check(mp
, ibp
);
1830 xfs_trans_brelse(tp
, ibp
);
1833 * Point the bucket head pointer at the next inode.
1835 ASSERT(next_agino
!= 0);
1836 ASSERT(next_agino
!= agino
);
1837 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(next_agino
);
1838 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
1839 (sizeof(xfs_agino_t
) * bucket_index
);
1840 xfs_trans_log_buf(tp
, agibp
, offset
,
1841 (offset
+ sizeof(xfs_agino_t
) - 1));
1844 * We need to search the list for the inode being freed.
1846 next_agino
= be32_to_cpu(agi
->agi_unlinked
[bucket_index
]);
1848 while (next_agino
!= agino
) {
1850 * If the last inode wasn't the one pointing to
1851 * us, then release its buffer since we're not
1852 * going to do anything with it.
1854 if (last_ibp
!= NULL
) {
1855 xfs_trans_brelse(tp
, last_ibp
);
1857 next_ino
= XFS_AGINO_TO_INO(mp
, agno
, next_agino
);
1858 error
= xfs_inotobp(mp
, tp
, next_ino
, &last_dip
,
1859 &last_ibp
, &last_offset
, 0);
1862 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
1863 error
, mp
->m_fsname
);
1866 next_agino
= be32_to_cpu(last_dip
->di_next_unlinked
);
1867 ASSERT(next_agino
!= NULLAGINO
);
1868 ASSERT(next_agino
!= 0);
1871 * Now last_ibp points to the buffer previous to us on
1872 * the unlinked list. Pull us from the list.
1874 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, XBF_LOCK
);
1877 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
1878 error
, mp
->m_fsname
);
1881 next_agino
= be32_to_cpu(dip
->di_next_unlinked
);
1882 ASSERT(next_agino
!= 0);
1883 ASSERT(next_agino
!= agino
);
1884 if (next_agino
!= NULLAGINO
) {
1885 dip
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
1886 offset
= ip
->i_imap
.im_boffset
+
1887 offsetof(xfs_dinode_t
, di_next_unlinked
);
1888 xfs_trans_inode_buf(tp
, ibp
);
1889 xfs_trans_log_buf(tp
, ibp
, offset
,
1890 (offset
+ sizeof(xfs_agino_t
) - 1));
1891 xfs_inobp_check(mp
, ibp
);
1893 xfs_trans_brelse(tp
, ibp
);
1896 * Point the previous inode on the list to the next inode.
1898 last_dip
->di_next_unlinked
= cpu_to_be32(next_agino
);
1899 ASSERT(next_agino
!= 0);
1900 offset
= last_offset
+ offsetof(xfs_dinode_t
, di_next_unlinked
);
1901 xfs_trans_inode_buf(tp
, last_ibp
);
1902 xfs_trans_log_buf(tp
, last_ibp
, offset
,
1903 (offset
+ sizeof(xfs_agino_t
) - 1));
1904 xfs_inobp_check(mp
, last_ibp
);
1910 * A big issue when freeing the inode cluster is is that we _cannot_ skip any
1911 * inodes that are in memory - they all must be marked stale and attached to
1912 * the cluster buffer.
1916 xfs_inode_t
*free_ip
,
1920 xfs_mount_t
*mp
= free_ip
->i_mount
;
1921 int blks_per_cluster
;
1928 xfs_inode_log_item_t
*iip
;
1929 xfs_log_item_t
*lip
;
1930 struct xfs_perag
*pag
;
1932 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, inum
));
1933 if (mp
->m_sb
.sb_blocksize
>= XFS_INODE_CLUSTER_SIZE(mp
)) {
1934 blks_per_cluster
= 1;
1935 ninodes
= mp
->m_sb
.sb_inopblock
;
1936 nbufs
= XFS_IALLOC_BLOCKS(mp
);
1938 blks_per_cluster
= XFS_INODE_CLUSTER_SIZE(mp
) /
1939 mp
->m_sb
.sb_blocksize
;
1940 ninodes
= blks_per_cluster
* mp
->m_sb
.sb_inopblock
;
1941 nbufs
= XFS_IALLOC_BLOCKS(mp
) / blks_per_cluster
;
1944 for (j
= 0; j
< nbufs
; j
++, inum
+= ninodes
) {
1945 blkno
= XFS_AGB_TO_DADDR(mp
, XFS_INO_TO_AGNO(mp
, inum
),
1946 XFS_INO_TO_AGBNO(mp
, inum
));
1949 * We obtain and lock the backing buffer first in the process
1950 * here, as we have to ensure that any dirty inode that we
1951 * can't get the flush lock on is attached to the buffer.
1952 * If we scan the in-memory inodes first, then buffer IO can
1953 * complete before we get a lock on it, and hence we may fail
1954 * to mark all the active inodes on the buffer stale.
1956 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, blkno
,
1957 mp
->m_bsize
* blks_per_cluster
,
1961 * Walk the inodes already attached to the buffer and mark them
1962 * stale. These will all have the flush locks held, so an
1963 * in-memory inode walk can't lock them. By marking them all
1964 * stale first, we will not attempt to lock them in the loop
1965 * below as the XFS_ISTALE flag will be set.
1967 lip
= XFS_BUF_FSPRIVATE(bp
, xfs_log_item_t
*);
1969 if (lip
->li_type
== XFS_LI_INODE
) {
1970 iip
= (xfs_inode_log_item_t
*)lip
;
1971 ASSERT(iip
->ili_logged
== 1);
1972 lip
->li_cb
= xfs_istale_done
;
1973 xfs_trans_ail_copy_lsn(mp
->m_ail
,
1974 &iip
->ili_flush_lsn
,
1975 &iip
->ili_item
.li_lsn
);
1976 xfs_iflags_set(iip
->ili_inode
, XFS_ISTALE
);
1978 lip
= lip
->li_bio_list
;
1983 * For each inode in memory attempt to add it to the inode
1984 * buffer and set it up for being staled on buffer IO
1985 * completion. This is safe as we've locked out tail pushing
1986 * and flushing by locking the buffer.
1988 * We have already marked every inode that was part of a
1989 * transaction stale above, which means there is no point in
1990 * even trying to lock them.
1992 for (i
= 0; i
< ninodes
; i
++) {
1995 ip
= radix_tree_lookup(&pag
->pag_ici_root
,
1996 XFS_INO_TO_AGINO(mp
, (inum
+ i
)));
1998 /* Inode not in memory, nothing to do */
2005 * because this is an RCU protected lookup, we could
2006 * find a recently freed or even reallocated inode
2007 * during the lookup. We need to check under the
2008 * i_flags_lock for a valid inode here. Skip it if it
2009 * is not valid, the wrong inode or stale.
2011 spin_lock(&ip
->i_flags_lock
);
2012 if (ip
->i_ino
!= inum
+ i
||
2013 __xfs_iflags_test(ip
, XFS_ISTALE
)) {
2014 spin_unlock(&ip
->i_flags_lock
);
2018 spin_unlock(&ip
->i_flags_lock
);
2021 * Don't try to lock/unlock the current inode, but we
2022 * _cannot_ skip the other inodes that we did not find
2023 * in the list attached to the buffer and are not
2024 * already marked stale. If we can't lock it, back off
2027 if (ip
!= free_ip
&&
2028 !xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
)) {
2036 xfs_iflags_set(ip
, XFS_ISTALE
);
2039 * we don't need to attach clean inodes or those only
2040 * with unlogged changes (which we throw away, anyway).
2043 if (!iip
|| xfs_inode_clean(ip
)) {
2044 ASSERT(ip
!= free_ip
);
2045 ip
->i_update_core
= 0;
2047 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2051 iip
->ili_last_fields
= iip
->ili_format
.ilf_fields
;
2052 iip
->ili_format
.ilf_fields
= 0;
2053 iip
->ili_logged
= 1;
2054 xfs_trans_ail_copy_lsn(mp
->m_ail
, &iip
->ili_flush_lsn
,
2055 &iip
->ili_item
.li_lsn
);
2057 xfs_buf_attach_iodone(bp
, xfs_istale_done
,
2061 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2064 xfs_trans_stale_inode_buf(tp
, bp
);
2065 xfs_trans_binval(tp
, bp
);
2072 * This is called to return an inode to the inode free list.
2073 * The inode should already be truncated to 0 length and have
2074 * no pages associated with it. This routine also assumes that
2075 * the inode is already a part of the transaction.
2077 * The on-disk copy of the inode will have been added to the list
2078 * of unlinked inodes in the AGI. We need to remove the inode from
2079 * that list atomically with respect to freeing it here.
2085 xfs_bmap_free_t
*flist
)
2089 xfs_ino_t first_ino
;
2093 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
2094 ASSERT(ip
->i_transp
== tp
);
2095 ASSERT(ip
->i_d
.di_nlink
== 0);
2096 ASSERT(ip
->i_d
.di_nextents
== 0);
2097 ASSERT(ip
->i_d
.di_anextents
== 0);
2098 ASSERT((ip
->i_d
.di_size
== 0 && ip
->i_size
== 0) ||
2099 ((ip
->i_d
.di_mode
& S_IFMT
) != S_IFREG
));
2100 ASSERT(ip
->i_d
.di_nblocks
== 0);
2103 * Pull the on-disk inode from the AGI unlinked list.
2105 error
= xfs_iunlink_remove(tp
, ip
);
2110 error
= xfs_difree(tp
, ip
->i_ino
, flist
, &delete, &first_ino
);
2114 ip
->i_d
.di_mode
= 0; /* mark incore inode as free */
2115 ip
->i_d
.di_flags
= 0;
2116 ip
->i_d
.di_dmevmask
= 0;
2117 ip
->i_d
.di_forkoff
= 0; /* mark the attr fork not in use */
2118 ip
->i_df
.if_ext_max
=
2119 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
2120 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
2121 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
2123 * Bump the generation count so no one will be confused
2124 * by reincarnations of this inode.
2128 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2130 error
= xfs_itobp(ip
->i_mount
, tp
, ip
, &dip
, &ibp
, XBF_LOCK
);
2135 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2136 * from picking up this inode when it is reclaimed (its incore state
2137 * initialzed but not flushed to disk yet). The in-core di_mode is
2138 * already cleared and a corresponding transaction logged.
2139 * The hack here just synchronizes the in-core to on-disk
2140 * di_mode value in advance before the actual inode sync to disk.
2141 * This is OK because the inode is already unlinked and would never
2142 * change its di_mode again for this inode generation.
2143 * This is a temporary hack that would require a proper fix
2149 xfs_ifree_cluster(ip
, tp
, first_ino
);
2156 * Reallocate the space for if_broot based on the number of records
2157 * being added or deleted as indicated in rec_diff. Move the records
2158 * and pointers in if_broot to fit the new size. When shrinking this
2159 * will eliminate holes between the records and pointers created by
2160 * the caller. When growing this will create holes to be filled in
2163 * The caller must not request to add more records than would fit in
2164 * the on-disk inode root. If the if_broot is currently NULL, then
2165 * if we adding records one will be allocated. The caller must also
2166 * not request that the number of records go below zero, although
2167 * it can go to zero.
2169 * ip -- the inode whose if_broot area is changing
2170 * ext_diff -- the change in the number of records, positive or negative,
2171 * requested for the if_broot array.
2179 struct xfs_mount
*mp
= ip
->i_mount
;
2182 struct xfs_btree_block
*new_broot
;
2189 * Handle the degenerate case quietly.
2191 if (rec_diff
== 0) {
2195 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2198 * If there wasn't any memory allocated before, just
2199 * allocate it now and get out.
2201 if (ifp
->if_broot_bytes
== 0) {
2202 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff
);
2203 ifp
->if_broot
= kmem_alloc(new_size
, KM_SLEEP
| KM_NOFS
);
2204 ifp
->if_broot_bytes
= (int)new_size
;
2209 * If there is already an existing if_broot, then we need
2210 * to realloc() it and shift the pointers to their new
2211 * location. The records don't change location because
2212 * they are kept butted up against the btree block header.
2214 cur_max
= xfs_bmbt_maxrecs(mp
, ifp
->if_broot_bytes
, 0);
2215 new_max
= cur_max
+ rec_diff
;
2216 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max
);
2217 ifp
->if_broot
= kmem_realloc(ifp
->if_broot
, new_size
,
2218 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max
), /* old size */
2219 KM_SLEEP
| KM_NOFS
);
2220 op
= (char *)XFS_BMAP_BROOT_PTR_ADDR(mp
, ifp
->if_broot
, 1,
2221 ifp
->if_broot_bytes
);
2222 np
= (char *)XFS_BMAP_BROOT_PTR_ADDR(mp
, ifp
->if_broot
, 1,
2224 ifp
->if_broot_bytes
= (int)new_size
;
2225 ASSERT(ifp
->if_broot_bytes
<=
2226 XFS_IFORK_SIZE(ip
, whichfork
) + XFS_BROOT_SIZE_ADJ
);
2227 memmove(np
, op
, cur_max
* (uint
)sizeof(xfs_dfsbno_t
));
2232 * rec_diff is less than 0. In this case, we are shrinking the
2233 * if_broot buffer. It must already exist. If we go to zero
2234 * records, just get rid of the root and clear the status bit.
2236 ASSERT((ifp
->if_broot
!= NULL
) && (ifp
->if_broot_bytes
> 0));
2237 cur_max
= xfs_bmbt_maxrecs(mp
, ifp
->if_broot_bytes
, 0);
2238 new_max
= cur_max
+ rec_diff
;
2239 ASSERT(new_max
>= 0);
2241 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max
);
2245 new_broot
= kmem_alloc(new_size
, KM_SLEEP
| KM_NOFS
);
2247 * First copy over the btree block header.
2249 memcpy(new_broot
, ifp
->if_broot
, XFS_BTREE_LBLOCK_LEN
);
2252 ifp
->if_flags
&= ~XFS_IFBROOT
;
2256 * Only copy the records and pointers if there are any.
2260 * First copy the records.
2262 op
= (char *)XFS_BMBT_REC_ADDR(mp
, ifp
->if_broot
, 1);
2263 np
= (char *)XFS_BMBT_REC_ADDR(mp
, new_broot
, 1);
2264 memcpy(np
, op
, new_max
* (uint
)sizeof(xfs_bmbt_rec_t
));
2267 * Then copy the pointers.
2269 op
= (char *)XFS_BMAP_BROOT_PTR_ADDR(mp
, ifp
->if_broot
, 1,
2270 ifp
->if_broot_bytes
);
2271 np
= (char *)XFS_BMAP_BROOT_PTR_ADDR(mp
, new_broot
, 1,
2273 memcpy(np
, op
, new_max
* (uint
)sizeof(xfs_dfsbno_t
));
2275 kmem_free(ifp
->if_broot
);
2276 ifp
->if_broot
= new_broot
;
2277 ifp
->if_broot_bytes
= (int)new_size
;
2278 ASSERT(ifp
->if_broot_bytes
<=
2279 XFS_IFORK_SIZE(ip
, whichfork
) + XFS_BROOT_SIZE_ADJ
);
2285 * This is called when the amount of space needed for if_data
2286 * is increased or decreased. The change in size is indicated by
2287 * the number of bytes that need to be added or deleted in the
2288 * byte_diff parameter.
2290 * If the amount of space needed has decreased below the size of the
2291 * inline buffer, then switch to using the inline buffer. Otherwise,
2292 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2293 * to what is needed.
2295 * ip -- the inode whose if_data area is changing
2296 * byte_diff -- the change in the number of bytes, positive or negative,
2297 * requested for the if_data array.
2309 if (byte_diff
== 0) {
2313 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2314 new_size
= (int)ifp
->if_bytes
+ byte_diff
;
2315 ASSERT(new_size
>= 0);
2317 if (new_size
== 0) {
2318 if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
2319 kmem_free(ifp
->if_u1
.if_data
);
2321 ifp
->if_u1
.if_data
= NULL
;
2323 } else if (new_size
<= sizeof(ifp
->if_u2
.if_inline_data
)) {
2325 * If the valid extents/data can fit in if_inline_ext/data,
2326 * copy them from the malloc'd vector and free it.
2328 if (ifp
->if_u1
.if_data
== NULL
) {
2329 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
2330 } else if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
2331 ASSERT(ifp
->if_real_bytes
!= 0);
2332 memcpy(ifp
->if_u2
.if_inline_data
, ifp
->if_u1
.if_data
,
2334 kmem_free(ifp
->if_u1
.if_data
);
2335 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
2340 * Stuck with malloc/realloc.
2341 * For inline data, the underlying buffer must be
2342 * a multiple of 4 bytes in size so that it can be
2343 * logged and stay on word boundaries. We enforce
2346 real_size
= roundup(new_size
, 4);
2347 if (ifp
->if_u1
.if_data
== NULL
) {
2348 ASSERT(ifp
->if_real_bytes
== 0);
2349 ifp
->if_u1
.if_data
= kmem_alloc(real_size
,
2350 KM_SLEEP
| KM_NOFS
);
2351 } else if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
2353 * Only do the realloc if the underlying size
2354 * is really changing.
2356 if (ifp
->if_real_bytes
!= real_size
) {
2357 ifp
->if_u1
.if_data
=
2358 kmem_realloc(ifp
->if_u1
.if_data
,
2361 KM_SLEEP
| KM_NOFS
);
2364 ASSERT(ifp
->if_real_bytes
== 0);
2365 ifp
->if_u1
.if_data
= kmem_alloc(real_size
,
2366 KM_SLEEP
| KM_NOFS
);
2367 memcpy(ifp
->if_u1
.if_data
, ifp
->if_u2
.if_inline_data
,
2371 ifp
->if_real_bytes
= real_size
;
2372 ifp
->if_bytes
= new_size
;
2373 ASSERT(ifp
->if_bytes
<= XFS_IFORK_SIZE(ip
, whichfork
));
2383 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2384 if (ifp
->if_broot
!= NULL
) {
2385 kmem_free(ifp
->if_broot
);
2386 ifp
->if_broot
= NULL
;
2390 * If the format is local, then we can't have an extents
2391 * array so just look for an inline data array. If we're
2392 * not local then we may or may not have an extents list,
2393 * so check and free it up if we do.
2395 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
) {
2396 if ((ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) &&
2397 (ifp
->if_u1
.if_data
!= NULL
)) {
2398 ASSERT(ifp
->if_real_bytes
!= 0);
2399 kmem_free(ifp
->if_u1
.if_data
);
2400 ifp
->if_u1
.if_data
= NULL
;
2401 ifp
->if_real_bytes
= 0;
2403 } else if ((ifp
->if_flags
& XFS_IFEXTENTS
) &&
2404 ((ifp
->if_flags
& XFS_IFEXTIREC
) ||
2405 ((ifp
->if_u1
.if_extents
!= NULL
) &&
2406 (ifp
->if_u1
.if_extents
!= ifp
->if_u2
.if_inline_ext
)))) {
2407 ASSERT(ifp
->if_real_bytes
!= 0);
2408 xfs_iext_destroy(ifp
);
2410 ASSERT(ifp
->if_u1
.if_extents
== NULL
||
2411 ifp
->if_u1
.if_extents
== ifp
->if_u2
.if_inline_ext
);
2412 ASSERT(ifp
->if_real_bytes
== 0);
2413 if (whichfork
== XFS_ATTR_FORK
) {
2414 kmem_zone_free(xfs_ifork_zone
, ip
->i_afp
);
2420 * This is called to unpin an inode. The caller must have the inode locked
2421 * in at least shared mode so that the buffer cannot be subsequently pinned
2422 * once someone is waiting for it to be unpinned.
2426 struct xfs_inode
*ip
)
2428 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2430 trace_xfs_inode_unpin_nowait(ip
, _RET_IP_
);
2432 /* Give the log a push to start the unpinning I/O */
2433 xfs_log_force_lsn(ip
->i_mount
, ip
->i_itemp
->ili_last_lsn
, 0);
2439 struct xfs_inode
*ip
)
2441 if (xfs_ipincount(ip
)) {
2442 xfs_iunpin_nowait(ip
);
2443 wait_event(ip
->i_ipin_wait
, (xfs_ipincount(ip
) == 0));
2448 * xfs_iextents_copy()
2450 * This is called to copy the REAL extents (as opposed to the delayed
2451 * allocation extents) from the inode into the given buffer. It
2452 * returns the number of bytes copied into the buffer.
2454 * If there are no delayed allocation extents, then we can just
2455 * memcpy() the extents into the buffer. Otherwise, we need to
2456 * examine each extent in turn and skip those which are delayed.
2468 xfs_fsblock_t start_block
;
2470 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2471 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2472 ASSERT(ifp
->if_bytes
> 0);
2474 nrecs
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
2475 XFS_BMAP_TRACE_EXLIST(ip
, nrecs
, whichfork
);
2479 * There are some delayed allocation extents in the
2480 * inode, so copy the extents one at a time and skip
2481 * the delayed ones. There must be at least one
2482 * non-delayed extent.
2485 for (i
= 0; i
< nrecs
; i
++) {
2486 xfs_bmbt_rec_host_t
*ep
= xfs_iext_get_ext(ifp
, i
);
2487 start_block
= xfs_bmbt_get_startblock(ep
);
2488 if (isnullstartblock(start_block
)) {
2490 * It's a delayed allocation extent, so skip it.
2495 /* Translate to on disk format */
2496 put_unaligned(cpu_to_be64(ep
->l0
), &dp
->l0
);
2497 put_unaligned(cpu_to_be64(ep
->l1
), &dp
->l1
);
2501 ASSERT(copied
!= 0);
2502 xfs_validate_extents(ifp
, copied
, XFS_EXTFMT_INODE(ip
));
2504 return (copied
* (uint
)sizeof(xfs_bmbt_rec_t
));
2508 * Each of the following cases stores data into the same region
2509 * of the on-disk inode, so only one of them can be valid at
2510 * any given time. While it is possible to have conflicting formats
2511 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2512 * in EXTENTS format, this can only happen when the fork has
2513 * changed formats after being modified but before being flushed.
2514 * In these cases, the format always takes precedence, because the
2515 * format indicates the current state of the fork.
2522 xfs_inode_log_item_t
*iip
,
2529 #ifdef XFS_TRANS_DEBUG
2532 static const short brootflag
[2] =
2533 { XFS_ILOG_DBROOT
, XFS_ILOG_ABROOT
};
2534 static const short dataflag
[2] =
2535 { XFS_ILOG_DDATA
, XFS_ILOG_ADATA
};
2536 static const short extflag
[2] =
2537 { XFS_ILOG_DEXT
, XFS_ILOG_AEXT
};
2541 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2543 * This can happen if we gave up in iformat in an error path,
2544 * for the attribute fork.
2547 ASSERT(whichfork
== XFS_ATTR_FORK
);
2550 cp
= XFS_DFORK_PTR(dip
, whichfork
);
2552 switch (XFS_IFORK_FORMAT(ip
, whichfork
)) {
2553 case XFS_DINODE_FMT_LOCAL
:
2554 if ((iip
->ili_format
.ilf_fields
& dataflag
[whichfork
]) &&
2555 (ifp
->if_bytes
> 0)) {
2556 ASSERT(ifp
->if_u1
.if_data
!= NULL
);
2557 ASSERT(ifp
->if_bytes
<= XFS_IFORK_SIZE(ip
, whichfork
));
2558 memcpy(cp
, ifp
->if_u1
.if_data
, ifp
->if_bytes
);
2562 case XFS_DINODE_FMT_EXTENTS
:
2563 ASSERT((ifp
->if_flags
& XFS_IFEXTENTS
) ||
2564 !(iip
->ili_format
.ilf_fields
& extflag
[whichfork
]));
2565 ASSERT((xfs_iext_get_ext(ifp
, 0) != NULL
) ||
2566 (ifp
->if_bytes
== 0));
2567 ASSERT((xfs_iext_get_ext(ifp
, 0) == NULL
) ||
2568 (ifp
->if_bytes
> 0));
2569 if ((iip
->ili_format
.ilf_fields
& extflag
[whichfork
]) &&
2570 (ifp
->if_bytes
> 0)) {
2571 ASSERT(XFS_IFORK_NEXTENTS(ip
, whichfork
) > 0);
2572 (void)xfs_iextents_copy(ip
, (xfs_bmbt_rec_t
*)cp
,
2577 case XFS_DINODE_FMT_BTREE
:
2578 if ((iip
->ili_format
.ilf_fields
& brootflag
[whichfork
]) &&
2579 (ifp
->if_broot_bytes
> 0)) {
2580 ASSERT(ifp
->if_broot
!= NULL
);
2581 ASSERT(ifp
->if_broot_bytes
<=
2582 (XFS_IFORK_SIZE(ip
, whichfork
) +
2583 XFS_BROOT_SIZE_ADJ
));
2584 xfs_bmbt_to_bmdr(mp
, ifp
->if_broot
, ifp
->if_broot_bytes
,
2585 (xfs_bmdr_block_t
*)cp
,
2586 XFS_DFORK_SIZE(dip
, mp
, whichfork
));
2590 case XFS_DINODE_FMT_DEV
:
2591 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_DEV
) {
2592 ASSERT(whichfork
== XFS_DATA_FORK
);
2593 xfs_dinode_put_rdev(dip
, ip
->i_df
.if_u2
.if_rdev
);
2597 case XFS_DINODE_FMT_UUID
:
2598 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_UUID
) {
2599 ASSERT(whichfork
== XFS_DATA_FORK
);
2600 memcpy(XFS_DFORK_DPTR(dip
),
2601 &ip
->i_df
.if_u2
.if_uuid
,
2617 xfs_mount_t
*mp
= ip
->i_mount
;
2618 struct xfs_perag
*pag
;
2619 unsigned long first_index
, mask
;
2620 unsigned long inodes_per_cluster
;
2622 xfs_inode_t
**ilist
;
2629 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
2631 inodes_per_cluster
= XFS_INODE_CLUSTER_SIZE(mp
) >> mp
->m_sb
.sb_inodelog
;
2632 ilist_size
= inodes_per_cluster
* sizeof(xfs_inode_t
*);
2633 ilist
= kmem_alloc(ilist_size
, KM_MAYFAIL
|KM_NOFS
);
2637 mask
= ~(((XFS_INODE_CLUSTER_SIZE(mp
) >> mp
->m_sb
.sb_inodelog
)) - 1);
2638 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
) & mask
;
2640 /* really need a gang lookup range call here */
2641 nr_found
= radix_tree_gang_lookup(&pag
->pag_ici_root
, (void**)ilist
,
2642 first_index
, inodes_per_cluster
);
2646 for (i
= 0; i
< nr_found
; i
++) {
2652 * because this is an RCU protected lookup, we could find a
2653 * recently freed or even reallocated inode during the lookup.
2654 * We need to check under the i_flags_lock for a valid inode
2655 * here. Skip it if it is not valid or the wrong inode.
2657 spin_lock(&ip
->i_flags_lock
);
2659 (XFS_INO_TO_AGINO(mp
, iq
->i_ino
) & mask
) != first_index
) {
2660 spin_unlock(&ip
->i_flags_lock
);
2663 spin_unlock(&ip
->i_flags_lock
);
2666 * Do an un-protected check to see if the inode is dirty and
2667 * is a candidate for flushing. These checks will be repeated
2668 * later after the appropriate locks are acquired.
2670 if (xfs_inode_clean(iq
) && xfs_ipincount(iq
) == 0)
2674 * Try to get locks. If any are unavailable or it is pinned,
2675 * then this inode cannot be flushed and is skipped.
2678 if (!xfs_ilock_nowait(iq
, XFS_ILOCK_SHARED
))
2680 if (!xfs_iflock_nowait(iq
)) {
2681 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
2684 if (xfs_ipincount(iq
)) {
2686 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
2691 * arriving here means that this inode can be flushed. First
2692 * re-check that it's dirty before flushing.
2694 if (!xfs_inode_clean(iq
)) {
2696 error
= xfs_iflush_int(iq
, bp
);
2698 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
2699 goto cluster_corrupt_out
;
2705 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
2709 XFS_STATS_INC(xs_icluster_flushcnt
);
2710 XFS_STATS_ADD(xs_icluster_flushinode
, clcount
);
2721 cluster_corrupt_out
:
2723 * Corruption detected in the clustering loop. Invalidate the
2724 * inode buffer and shut down the filesystem.
2728 * Clean up the buffer. If it was B_DELWRI, just release it --
2729 * brelse can handle it with no problems. If not, shut down the
2730 * filesystem before releasing the buffer.
2732 bufwasdelwri
= XFS_BUF_ISDELAYWRITE(bp
);
2736 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
2738 if (!bufwasdelwri
) {
2740 * Just like incore_relse: if we have b_iodone functions,
2741 * mark the buffer as an error and call them. Otherwise
2742 * mark it as stale and brelse.
2744 if (XFS_BUF_IODONE_FUNC(bp
)) {
2747 XFS_BUF_ERROR(bp
,EIO
);
2748 xfs_buf_ioend(bp
, 0);
2756 * Unlocks the flush lock
2758 xfs_iflush_abort(iq
);
2761 return XFS_ERROR(EFSCORRUPTED
);
2765 * xfs_iflush() will write a modified inode's changes out to the
2766 * inode's on disk home. The caller must have the inode lock held
2767 * in at least shared mode and the inode flush completion must be
2768 * active as well. The inode lock will still be held upon return from
2769 * the call and the caller is free to unlock it.
2770 * The inode flush will be completed when the inode reaches the disk.
2771 * The flags indicate how the inode's buffer should be written out.
2778 xfs_inode_log_item_t
*iip
;
2784 XFS_STATS_INC(xs_iflush_count
);
2786 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2787 ASSERT(!completion_done(&ip
->i_flush
));
2788 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
2789 ip
->i_d
.di_nextents
> ip
->i_df
.if_ext_max
);
2795 * We can't flush the inode until it is unpinned, so wait for it if we
2796 * are allowed to block. We know noone new can pin it, because we are
2797 * holding the inode lock shared and you need to hold it exclusively to
2800 * If we are not allowed to block, force the log out asynchronously so
2801 * that when we come back the inode will be unpinned. If other inodes
2802 * in the same cluster are dirty, they will probably write the inode
2803 * out for us if they occur after the log force completes.
2805 if (!(flags
& SYNC_WAIT
) && xfs_ipincount(ip
)) {
2806 xfs_iunpin_nowait(ip
);
2810 xfs_iunpin_wait(ip
);
2813 * For stale inodes we cannot rely on the backing buffer remaining
2814 * stale in cache for the remaining life of the stale inode and so
2815 * xfs_itobp() below may give us a buffer that no longer contains
2816 * inodes below. We have to check this after ensuring the inode is
2817 * unpinned so that it is safe to reclaim the stale inode after the
2820 if (xfs_iflags_test(ip
, XFS_ISTALE
)) {
2826 * This may have been unpinned because the filesystem is shutting
2827 * down forcibly. If that's the case we must not write this inode
2828 * to disk, because the log record didn't make it to disk!
2830 if (XFS_FORCED_SHUTDOWN(mp
)) {
2831 ip
->i_update_core
= 0;
2833 iip
->ili_format
.ilf_fields
= 0;
2835 return XFS_ERROR(EIO
);
2839 * Get the buffer containing the on-disk inode.
2841 error
= xfs_itobp(mp
, NULL
, ip
, &dip
, &bp
,
2842 (flags
& SYNC_WAIT
) ? XBF_LOCK
: XBF_TRYLOCK
);
2849 * First flush out the inode that xfs_iflush was called with.
2851 error
= xfs_iflush_int(ip
, bp
);
2856 * If the buffer is pinned then push on the log now so we won't
2857 * get stuck waiting in the write for too long.
2859 if (XFS_BUF_ISPINNED(bp
))
2860 xfs_log_force(mp
, 0);
2864 * see if other inodes can be gathered into this write
2866 error
= xfs_iflush_cluster(ip
, bp
);
2868 goto cluster_corrupt_out
;
2870 if (flags
& SYNC_WAIT
)
2871 error
= xfs_bwrite(mp
, bp
);
2873 xfs_bdwrite(mp
, bp
);
2878 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
2879 cluster_corrupt_out
:
2881 * Unlocks the flush lock
2883 xfs_iflush_abort(ip
);
2884 return XFS_ERROR(EFSCORRUPTED
);
2893 xfs_inode_log_item_t
*iip
;
2896 #ifdef XFS_TRANS_DEBUG
2900 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2901 ASSERT(!completion_done(&ip
->i_flush
));
2902 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
2903 ip
->i_d
.di_nextents
> ip
->i_df
.if_ext_max
);
2908 /* set *dip = inode's place in the buffer */
2909 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, ip
->i_imap
.im_boffset
);
2912 * Clear i_update_core before copying out the data.
2913 * This is for coordination with our timestamp updates
2914 * that don't hold the inode lock. They will always
2915 * update the timestamps BEFORE setting i_update_core,
2916 * so if we clear i_update_core after they set it we
2917 * are guaranteed to see their updates to the timestamps.
2918 * I believe that this depends on strongly ordered memory
2919 * semantics, but we have that. We use the SYNCHRONIZE
2920 * macro to make sure that the compiler does not reorder
2921 * the i_update_core access below the data copy below.
2923 ip
->i_update_core
= 0;
2927 * Make sure to get the latest timestamps from the Linux inode.
2929 xfs_synchronize_times(ip
);
2931 if (XFS_TEST_ERROR(be16_to_cpu(dip
->di_magic
) != XFS_DINODE_MAGIC
,
2932 mp
, XFS_ERRTAG_IFLUSH_1
, XFS_RANDOM_IFLUSH_1
)) {
2933 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
2934 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
2935 __func__
, ip
->i_ino
, be16_to_cpu(dip
->di_magic
), dip
);
2938 if (XFS_TEST_ERROR(ip
->i_d
.di_magic
!= XFS_DINODE_MAGIC
,
2939 mp
, XFS_ERRTAG_IFLUSH_2
, XFS_RANDOM_IFLUSH_2
)) {
2940 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
2941 "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
2942 __func__
, ip
->i_ino
, ip
, ip
->i_d
.di_magic
);
2945 if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
) {
2947 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2948 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
),
2949 mp
, XFS_ERRTAG_IFLUSH_3
, XFS_RANDOM_IFLUSH_3
)) {
2950 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
2951 "%s: Bad regular inode %Lu, ptr 0x%p",
2952 __func__
, ip
->i_ino
, ip
);
2955 } else if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFDIR
) {
2957 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2958 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
) &&
2959 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_LOCAL
),
2960 mp
, XFS_ERRTAG_IFLUSH_4
, XFS_RANDOM_IFLUSH_4
)) {
2961 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
2962 "%s: Bad directory inode %Lu, ptr 0x%p",
2963 __func__
, ip
->i_ino
, ip
);
2967 if (XFS_TEST_ERROR(ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
>
2968 ip
->i_d
.di_nblocks
, mp
, XFS_ERRTAG_IFLUSH_5
,
2969 XFS_RANDOM_IFLUSH_5
)) {
2970 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
2971 "%s: detected corrupt incore inode %Lu, "
2972 "total extents = %d, nblocks = %Ld, ptr 0x%p",
2973 __func__
, ip
->i_ino
,
2974 ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
,
2975 ip
->i_d
.di_nblocks
, ip
);
2978 if (XFS_TEST_ERROR(ip
->i_d
.di_forkoff
> mp
->m_sb
.sb_inodesize
,
2979 mp
, XFS_ERRTAG_IFLUSH_6
, XFS_RANDOM_IFLUSH_6
)) {
2980 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
2981 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
2982 __func__
, ip
->i_ino
, ip
->i_d
.di_forkoff
, ip
);
2986 * bump the flush iteration count, used to detect flushes which
2987 * postdate a log record during recovery.
2990 ip
->i_d
.di_flushiter
++;
2993 * Copy the dirty parts of the inode into the on-disk
2994 * inode. We always copy out the core of the inode,
2995 * because if the inode is dirty at all the core must
2998 xfs_dinode_to_disk(dip
, &ip
->i_d
);
3000 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3001 if (ip
->i_d
.di_flushiter
== DI_MAX_FLUSH
)
3002 ip
->i_d
.di_flushiter
= 0;
3005 * If this is really an old format inode and the superblock version
3006 * has not been updated to support only new format inodes, then
3007 * convert back to the old inode format. If the superblock version
3008 * has been updated, then make the conversion permanent.
3010 ASSERT(ip
->i_d
.di_version
== 1 || xfs_sb_version_hasnlink(&mp
->m_sb
));
3011 if (ip
->i_d
.di_version
== 1) {
3012 if (!xfs_sb_version_hasnlink(&mp
->m_sb
)) {
3016 ASSERT(ip
->i_d
.di_nlink
<= XFS_MAXLINK_1
);
3017 dip
->di_onlink
= cpu_to_be16(ip
->i_d
.di_nlink
);
3020 * The superblock version has already been bumped,
3021 * so just make the conversion to the new inode
3024 ip
->i_d
.di_version
= 2;
3025 dip
->di_version
= 2;
3026 ip
->i_d
.di_onlink
= 0;
3028 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
3029 memset(&(dip
->di_pad
[0]), 0,
3030 sizeof(dip
->di_pad
));
3031 ASSERT(xfs_get_projid(ip
) == 0);
3035 xfs_iflush_fork(ip
, dip
, iip
, XFS_DATA_FORK
, bp
);
3036 if (XFS_IFORK_Q(ip
))
3037 xfs_iflush_fork(ip
, dip
, iip
, XFS_ATTR_FORK
, bp
);
3038 xfs_inobp_check(mp
, bp
);
3041 * We've recorded everything logged in the inode, so we'd
3042 * like to clear the ilf_fields bits so we don't log and
3043 * flush things unnecessarily. However, we can't stop
3044 * logging all this information until the data we've copied
3045 * into the disk buffer is written to disk. If we did we might
3046 * overwrite the copy of the inode in the log with all the
3047 * data after re-logging only part of it, and in the face of
3048 * a crash we wouldn't have all the data we need to recover.
3050 * What we do is move the bits to the ili_last_fields field.
3051 * When logging the inode, these bits are moved back to the
3052 * ilf_fields field. In the xfs_iflush_done() routine we
3053 * clear ili_last_fields, since we know that the information
3054 * those bits represent is permanently on disk. As long as
3055 * the flush completes before the inode is logged again, then
3056 * both ilf_fields and ili_last_fields will be cleared.
3058 * We can play with the ilf_fields bits here, because the inode
3059 * lock must be held exclusively in order to set bits there
3060 * and the flush lock protects the ili_last_fields bits.
3061 * Set ili_logged so the flush done
3062 * routine can tell whether or not to look in the AIL.
3063 * Also, store the current LSN of the inode so that we can tell
3064 * whether the item has moved in the AIL from xfs_iflush_done().
3065 * In order to read the lsn we need the AIL lock, because
3066 * it is a 64 bit value that cannot be read atomically.
3068 if (iip
!= NULL
&& iip
->ili_format
.ilf_fields
!= 0) {
3069 iip
->ili_last_fields
= iip
->ili_format
.ilf_fields
;
3070 iip
->ili_format
.ilf_fields
= 0;
3071 iip
->ili_logged
= 1;
3073 xfs_trans_ail_copy_lsn(mp
->m_ail
, &iip
->ili_flush_lsn
,
3074 &iip
->ili_item
.li_lsn
);
3077 * Attach the function xfs_iflush_done to the inode's
3078 * buffer. This will remove the inode from the AIL
3079 * and unlock the inode's flush lock when the inode is
3080 * completely written to disk.
3082 xfs_buf_attach_iodone(bp
, xfs_iflush_done
, &iip
->ili_item
);
3084 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) != NULL
);
3085 ASSERT(XFS_BUF_IODONE_FUNC(bp
) != NULL
);
3088 * We're flushing an inode which is not in the AIL and has
3089 * not been logged but has i_update_core set. For this
3090 * case we can use a B_DELWRI flush and immediately drop
3091 * the inode flush lock because we can avoid the whole
3092 * AIL state thing. It's OK to drop the flush lock now,
3093 * because we've already locked the buffer and to do anything
3094 * you really need both.
3097 ASSERT(iip
->ili_logged
== 0);
3098 ASSERT(iip
->ili_last_fields
== 0);
3099 ASSERT((iip
->ili_item
.li_flags
& XFS_LI_IN_AIL
) == 0);
3107 return XFS_ERROR(EFSCORRUPTED
);
3111 * Return a pointer to the extent record at file index idx.
3113 xfs_bmbt_rec_host_t
*
3115 xfs_ifork_t
*ifp
, /* inode fork pointer */
3116 xfs_extnum_t idx
) /* index of target extent */
3119 if ((ifp
->if_flags
& XFS_IFEXTIREC
) && (idx
== 0)) {
3120 return ifp
->if_u1
.if_ext_irec
->er_extbuf
;
3121 } else if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3122 xfs_ext_irec_t
*erp
; /* irec pointer */
3123 int erp_idx
= 0; /* irec index */
3124 xfs_extnum_t page_idx
= idx
; /* ext index in target list */
3126 erp
= xfs_iext_idx_to_irec(ifp
, &page_idx
, &erp_idx
, 0);
3127 return &erp
->er_extbuf
[page_idx
];
3128 } else if (ifp
->if_bytes
) {
3129 return &ifp
->if_u1
.if_extents
[idx
];
3136 * Insert new item(s) into the extent records for incore inode
3137 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3141 xfs_inode_t
*ip
, /* incore inode pointer */
3142 xfs_extnum_t idx
, /* starting index of new items */
3143 xfs_extnum_t count
, /* number of inserted items */
3144 xfs_bmbt_irec_t
*new, /* items to insert */
3145 int state
) /* type of extent conversion */
3147 xfs_ifork_t
*ifp
= (state
& BMAP_ATTRFORK
) ? ip
->i_afp
: &ip
->i_df
;
3148 xfs_extnum_t i
; /* extent record index */
3150 trace_xfs_iext_insert(ip
, idx
, new, state
, _RET_IP_
);
3152 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
3153 xfs_iext_add(ifp
, idx
, count
);
3154 for (i
= idx
; i
< idx
+ count
; i
++, new++)
3155 xfs_bmbt_set_all(xfs_iext_get_ext(ifp
, i
), new);
3159 * This is called when the amount of space required for incore file
3160 * extents needs to be increased. The ext_diff parameter stores the
3161 * number of new extents being added and the idx parameter contains
3162 * the extent index where the new extents will be added. If the new
3163 * extents are being appended, then we just need to (re)allocate and
3164 * initialize the space. Otherwise, if the new extents are being
3165 * inserted into the middle of the existing entries, a bit more work
3166 * is required to make room for the new extents to be inserted. The
3167 * caller is responsible for filling in the new extent entries upon
3172 xfs_ifork_t
*ifp
, /* inode fork pointer */
3173 xfs_extnum_t idx
, /* index to begin adding exts */
3174 int ext_diff
) /* number of extents to add */
3176 int byte_diff
; /* new bytes being added */
3177 int new_size
; /* size of extents after adding */
3178 xfs_extnum_t nextents
; /* number of extents in file */
3180 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3181 ASSERT((idx
>= 0) && (idx
<= nextents
));
3182 byte_diff
= ext_diff
* sizeof(xfs_bmbt_rec_t
);
3183 new_size
= ifp
->if_bytes
+ byte_diff
;
3185 * If the new number of extents (nextents + ext_diff)
3186 * fits inside the inode, then continue to use the inline
3189 if (nextents
+ ext_diff
<= XFS_INLINE_EXTS
) {
3190 if (idx
< nextents
) {
3191 memmove(&ifp
->if_u2
.if_inline_ext
[idx
+ ext_diff
],
3192 &ifp
->if_u2
.if_inline_ext
[idx
],
3193 (nextents
- idx
) * sizeof(xfs_bmbt_rec_t
));
3194 memset(&ifp
->if_u2
.if_inline_ext
[idx
], 0, byte_diff
);
3196 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
3197 ifp
->if_real_bytes
= 0;
3198 ifp
->if_lastex
= nextents
+ ext_diff
;
3201 * Otherwise use a linear (direct) extent list.
3202 * If the extents are currently inside the inode,
3203 * xfs_iext_realloc_direct will switch us from
3204 * inline to direct extent allocation mode.
3206 else if (nextents
+ ext_diff
<= XFS_LINEAR_EXTS
) {
3207 xfs_iext_realloc_direct(ifp
, new_size
);
3208 if (idx
< nextents
) {
3209 memmove(&ifp
->if_u1
.if_extents
[idx
+ ext_diff
],
3210 &ifp
->if_u1
.if_extents
[idx
],
3211 (nextents
- idx
) * sizeof(xfs_bmbt_rec_t
));
3212 memset(&ifp
->if_u1
.if_extents
[idx
], 0, byte_diff
);
3215 /* Indirection array */
3217 xfs_ext_irec_t
*erp
;
3221 ASSERT(nextents
+ ext_diff
> XFS_LINEAR_EXTS
);
3222 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3223 erp
= xfs_iext_idx_to_irec(ifp
, &page_idx
, &erp_idx
, 1);
3225 xfs_iext_irec_init(ifp
);
3226 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3227 erp
= ifp
->if_u1
.if_ext_irec
;
3229 /* Extents fit in target extent page */
3230 if (erp
&& erp
->er_extcount
+ ext_diff
<= XFS_LINEAR_EXTS
) {
3231 if (page_idx
< erp
->er_extcount
) {
3232 memmove(&erp
->er_extbuf
[page_idx
+ ext_diff
],
3233 &erp
->er_extbuf
[page_idx
],
3234 (erp
->er_extcount
- page_idx
) *
3235 sizeof(xfs_bmbt_rec_t
));
3236 memset(&erp
->er_extbuf
[page_idx
], 0, byte_diff
);
3238 erp
->er_extcount
+= ext_diff
;
3239 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, ext_diff
);
3241 /* Insert a new extent page */
3243 xfs_iext_add_indirect_multi(ifp
,
3244 erp_idx
, page_idx
, ext_diff
);
3247 * If extent(s) are being appended to the last page in
3248 * the indirection array and the new extent(s) don't fit
3249 * in the page, then erp is NULL and erp_idx is set to
3250 * the next index needed in the indirection array.
3253 int count
= ext_diff
;
3256 erp
= xfs_iext_irec_new(ifp
, erp_idx
);
3257 erp
->er_extcount
= count
;
3258 count
-= MIN(count
, (int)XFS_LINEAR_EXTS
);
3265 ifp
->if_bytes
= new_size
;
3269 * This is called when incore extents are being added to the indirection
3270 * array and the new extents do not fit in the target extent list. The
3271 * erp_idx parameter contains the irec index for the target extent list
3272 * in the indirection array, and the idx parameter contains the extent
3273 * index within the list. The number of extents being added is stored
3274 * in the count parameter.
3276 * |-------| |-------|
3277 * | | | | idx - number of extents before idx
3279 * | | | | count - number of extents being inserted at idx
3280 * |-------| |-------|
3281 * | count | | nex2 | nex2 - number of extents after idx + count
3282 * |-------| |-------|
3285 xfs_iext_add_indirect_multi(
3286 xfs_ifork_t
*ifp
, /* inode fork pointer */
3287 int erp_idx
, /* target extent irec index */
3288 xfs_extnum_t idx
, /* index within target list */
3289 int count
) /* new extents being added */
3291 int byte_diff
; /* new bytes being added */
3292 xfs_ext_irec_t
*erp
; /* pointer to irec entry */
3293 xfs_extnum_t ext_diff
; /* number of extents to add */
3294 xfs_extnum_t ext_cnt
; /* new extents still needed */
3295 xfs_extnum_t nex2
; /* extents after idx + count */
3296 xfs_bmbt_rec_t
*nex2_ep
= NULL
; /* temp list for nex2 extents */
3297 int nlists
; /* number of irec's (lists) */
3299 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3300 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
3301 nex2
= erp
->er_extcount
- idx
;
3302 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3305 * Save second part of target extent list
3306 * (all extents past */
3308 byte_diff
= nex2
* sizeof(xfs_bmbt_rec_t
);
3309 nex2_ep
= (xfs_bmbt_rec_t
*) kmem_alloc(byte_diff
, KM_NOFS
);
3310 memmove(nex2_ep
, &erp
->er_extbuf
[idx
], byte_diff
);
3311 erp
->er_extcount
-= nex2
;
3312 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, -nex2
);
3313 memset(&erp
->er_extbuf
[idx
], 0, byte_diff
);
3317 * Add the new extents to the end of the target
3318 * list, then allocate new irec record(s) and
3319 * extent buffer(s) as needed to store the rest
3320 * of the new extents.
3323 ext_diff
= MIN(ext_cnt
, (int)XFS_LINEAR_EXTS
- erp
->er_extcount
);
3325 erp
->er_extcount
+= ext_diff
;
3326 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, ext_diff
);
3327 ext_cnt
-= ext_diff
;
3331 erp
= xfs_iext_irec_new(ifp
, erp_idx
);
3332 ext_diff
= MIN(ext_cnt
, (int)XFS_LINEAR_EXTS
);
3333 erp
->er_extcount
= ext_diff
;
3334 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, ext_diff
);
3335 ext_cnt
-= ext_diff
;
3338 /* Add nex2 extents back to indirection array */
3340 xfs_extnum_t ext_avail
;
3343 byte_diff
= nex2
* sizeof(xfs_bmbt_rec_t
);
3344 ext_avail
= XFS_LINEAR_EXTS
- erp
->er_extcount
;
3347 * If nex2 extents fit in the current page, append
3348 * nex2_ep after the new extents.
3350 if (nex2
<= ext_avail
) {
3351 i
= erp
->er_extcount
;
3354 * Otherwise, check if space is available in the
3357 else if ((erp_idx
< nlists
- 1) &&
3358 (nex2
<= (ext_avail
= XFS_LINEAR_EXTS
-
3359 ifp
->if_u1
.if_ext_irec
[erp_idx
+1].er_extcount
))) {
3362 /* Create a hole for nex2 extents */
3363 memmove(&erp
->er_extbuf
[nex2
], erp
->er_extbuf
,
3364 erp
->er_extcount
* sizeof(xfs_bmbt_rec_t
));
3367 * Final choice, create a new extent page for
3372 erp
= xfs_iext_irec_new(ifp
, erp_idx
);
3374 memmove(&erp
->er_extbuf
[i
], nex2_ep
, byte_diff
);
3376 erp
->er_extcount
+= nex2
;
3377 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, nex2
);
3382 * This is called when the amount of space required for incore file
3383 * extents needs to be decreased. The ext_diff parameter stores the
3384 * number of extents to be removed and the idx parameter contains
3385 * the extent index where the extents will be removed from.
3387 * If the amount of space needed has decreased below the linear
3388 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3389 * extent array. Otherwise, use kmem_realloc() to adjust the
3390 * size to what is needed.
3394 xfs_inode_t
*ip
, /* incore inode pointer */
3395 xfs_extnum_t idx
, /* index to begin removing exts */
3396 int ext_diff
, /* number of extents to remove */
3397 int state
) /* type of extent conversion */
3399 xfs_ifork_t
*ifp
= (state
& BMAP_ATTRFORK
) ? ip
->i_afp
: &ip
->i_df
;
3400 xfs_extnum_t nextents
; /* number of extents in file */
3401 int new_size
; /* size of extents after removal */
3403 trace_xfs_iext_remove(ip
, idx
, state
, _RET_IP_
);
3405 ASSERT(ext_diff
> 0);
3406 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3407 new_size
= (nextents
- ext_diff
) * sizeof(xfs_bmbt_rec_t
);
3409 if (new_size
== 0) {
3410 xfs_iext_destroy(ifp
);
3411 } else if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3412 xfs_iext_remove_indirect(ifp
, idx
, ext_diff
);
3413 } else if (ifp
->if_real_bytes
) {
3414 xfs_iext_remove_direct(ifp
, idx
, ext_diff
);
3416 xfs_iext_remove_inline(ifp
, idx
, ext_diff
);
3418 ifp
->if_bytes
= new_size
;
3422 * This removes ext_diff extents from the inline buffer, beginning
3423 * at extent index idx.
3426 xfs_iext_remove_inline(
3427 xfs_ifork_t
*ifp
, /* inode fork pointer */
3428 xfs_extnum_t idx
, /* index to begin removing exts */
3429 int ext_diff
) /* number of extents to remove */
3431 int nextents
; /* number of extents in file */
3433 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
));
3434 ASSERT(idx
< XFS_INLINE_EXTS
);
3435 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3436 ASSERT(((nextents
- ext_diff
) > 0) &&
3437 (nextents
- ext_diff
) < XFS_INLINE_EXTS
);
3439 if (idx
+ ext_diff
< nextents
) {
3440 memmove(&ifp
->if_u2
.if_inline_ext
[idx
],
3441 &ifp
->if_u2
.if_inline_ext
[idx
+ ext_diff
],
3442 (nextents
- (idx
+ ext_diff
)) *
3443 sizeof(xfs_bmbt_rec_t
));
3444 memset(&ifp
->if_u2
.if_inline_ext
[nextents
- ext_diff
],
3445 0, ext_diff
* sizeof(xfs_bmbt_rec_t
));
3447 memset(&ifp
->if_u2
.if_inline_ext
[idx
], 0,
3448 ext_diff
* sizeof(xfs_bmbt_rec_t
));
3453 * This removes ext_diff extents from a linear (direct) extent list,
3454 * beginning at extent index idx. If the extents are being removed
3455 * from the end of the list (ie. truncate) then we just need to re-
3456 * allocate the list to remove the extra space. Otherwise, if the
3457 * extents are being removed from the middle of the existing extent
3458 * entries, then we first need to move the extent records beginning
3459 * at idx + ext_diff up in the list to overwrite the records being
3460 * removed, then remove the extra space via kmem_realloc.
3463 xfs_iext_remove_direct(
3464 xfs_ifork_t
*ifp
, /* inode fork pointer */
3465 xfs_extnum_t idx
, /* index to begin removing exts */
3466 int ext_diff
) /* number of extents to remove */
3468 xfs_extnum_t nextents
; /* number of extents in file */
3469 int new_size
; /* size of extents after removal */
3471 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
));
3472 new_size
= ifp
->if_bytes
-
3473 (ext_diff
* sizeof(xfs_bmbt_rec_t
));
3474 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3476 if (new_size
== 0) {
3477 xfs_iext_destroy(ifp
);
3480 /* Move extents up in the list (if needed) */
3481 if (idx
+ ext_diff
< nextents
) {
3482 memmove(&ifp
->if_u1
.if_extents
[idx
],
3483 &ifp
->if_u1
.if_extents
[idx
+ ext_diff
],
3484 (nextents
- (idx
+ ext_diff
)) *
3485 sizeof(xfs_bmbt_rec_t
));
3487 memset(&ifp
->if_u1
.if_extents
[nextents
- ext_diff
],
3488 0, ext_diff
* sizeof(xfs_bmbt_rec_t
));
3490 * Reallocate the direct extent list. If the extents
3491 * will fit inside the inode then xfs_iext_realloc_direct
3492 * will switch from direct to inline extent allocation
3495 xfs_iext_realloc_direct(ifp
, new_size
);
3496 ifp
->if_bytes
= new_size
;
3500 * This is called when incore extents are being removed from the
3501 * indirection array and the extents being removed span multiple extent
3502 * buffers. The idx parameter contains the file extent index where we
3503 * want to begin removing extents, and the count parameter contains
3504 * how many extents need to be removed.
3506 * |-------| |-------|
3507 * | nex1 | | | nex1 - number of extents before idx
3508 * |-------| | count |
3509 * | | | | count - number of extents being removed at idx
3510 * | count | |-------|
3511 * | | | nex2 | nex2 - number of extents after idx + count
3512 * |-------| |-------|
3515 xfs_iext_remove_indirect(
3516 xfs_ifork_t
*ifp
, /* inode fork pointer */
3517 xfs_extnum_t idx
, /* index to begin removing extents */
3518 int count
) /* number of extents to remove */
3520 xfs_ext_irec_t
*erp
; /* indirection array pointer */
3521 int erp_idx
= 0; /* indirection array index */
3522 xfs_extnum_t ext_cnt
; /* extents left to remove */
3523 xfs_extnum_t ext_diff
; /* extents to remove in current list */
3524 xfs_extnum_t nex1
; /* number of extents before idx */
3525 xfs_extnum_t nex2
; /* extents after idx + count */
3526 int page_idx
= idx
; /* index in target extent list */
3528 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3529 erp
= xfs_iext_idx_to_irec(ifp
, &page_idx
, &erp_idx
, 0);
3530 ASSERT(erp
!= NULL
);
3534 nex2
= MAX((erp
->er_extcount
- (nex1
+ ext_cnt
)), 0);
3535 ext_diff
= MIN(ext_cnt
, (erp
->er_extcount
- nex1
));
3537 * Check for deletion of entire list;
3538 * xfs_iext_irec_remove() updates extent offsets.
3540 if (ext_diff
== erp
->er_extcount
) {
3541 xfs_iext_irec_remove(ifp
, erp_idx
);
3542 ext_cnt
-= ext_diff
;
3545 ASSERT(erp_idx
< ifp
->if_real_bytes
/
3547 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
3554 /* Move extents up (if needed) */
3556 memmove(&erp
->er_extbuf
[nex1
],
3557 &erp
->er_extbuf
[nex1
+ ext_diff
],
3558 nex2
* sizeof(xfs_bmbt_rec_t
));
3560 /* Zero out rest of page */
3561 memset(&erp
->er_extbuf
[nex1
+ nex2
], 0, (XFS_IEXT_BUFSZ
-
3562 ((nex1
+ nex2
) * sizeof(xfs_bmbt_rec_t
))));
3563 /* Update remaining counters */
3564 erp
->er_extcount
-= ext_diff
;
3565 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, -ext_diff
);
3566 ext_cnt
-= ext_diff
;
3571 ifp
->if_bytes
-= count
* sizeof(xfs_bmbt_rec_t
);
3572 xfs_iext_irec_compact(ifp
);
3576 * Create, destroy, or resize a linear (direct) block of extents.
3579 xfs_iext_realloc_direct(
3580 xfs_ifork_t
*ifp
, /* inode fork pointer */
3581 int new_size
) /* new size of extents */
3583 int rnew_size
; /* real new size of extents */
3585 rnew_size
= new_size
;
3587 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
) ||
3588 ((new_size
>= 0) && (new_size
<= XFS_IEXT_BUFSZ
) &&
3589 (new_size
!= ifp
->if_real_bytes
)));
3591 /* Free extent records */
3592 if (new_size
== 0) {
3593 xfs_iext_destroy(ifp
);
3595 /* Resize direct extent list and zero any new bytes */
3596 else if (ifp
->if_real_bytes
) {
3597 /* Check if extents will fit inside the inode */
3598 if (new_size
<= XFS_INLINE_EXTS
* sizeof(xfs_bmbt_rec_t
)) {
3599 xfs_iext_direct_to_inline(ifp
, new_size
/
3600 (uint
)sizeof(xfs_bmbt_rec_t
));
3601 ifp
->if_bytes
= new_size
;
3604 if (!is_power_of_2(new_size
)){
3605 rnew_size
= roundup_pow_of_two(new_size
);
3607 if (rnew_size
!= ifp
->if_real_bytes
) {
3608 ifp
->if_u1
.if_extents
=
3609 kmem_realloc(ifp
->if_u1
.if_extents
,
3611 ifp
->if_real_bytes
, KM_NOFS
);
3613 if (rnew_size
> ifp
->if_real_bytes
) {
3614 memset(&ifp
->if_u1
.if_extents
[ifp
->if_bytes
/
3615 (uint
)sizeof(xfs_bmbt_rec_t
)], 0,
3616 rnew_size
- ifp
->if_real_bytes
);
3620 * Switch from the inline extent buffer to a direct
3621 * extent list. Be sure to include the inline extent
3622 * bytes in new_size.
3625 new_size
+= ifp
->if_bytes
;
3626 if (!is_power_of_2(new_size
)) {
3627 rnew_size
= roundup_pow_of_two(new_size
);
3629 xfs_iext_inline_to_direct(ifp
, rnew_size
);
3631 ifp
->if_real_bytes
= rnew_size
;
3632 ifp
->if_bytes
= new_size
;
3636 * Switch from linear (direct) extent records to inline buffer.
3639 xfs_iext_direct_to_inline(
3640 xfs_ifork_t
*ifp
, /* inode fork pointer */
3641 xfs_extnum_t nextents
) /* number of extents in file */
3643 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
3644 ASSERT(nextents
<= XFS_INLINE_EXTS
);
3646 * The inline buffer was zeroed when we switched
3647 * from inline to direct extent allocation mode,
3648 * so we don't need to clear it here.
3650 memcpy(ifp
->if_u2
.if_inline_ext
, ifp
->if_u1
.if_extents
,
3651 nextents
* sizeof(xfs_bmbt_rec_t
));
3652 kmem_free(ifp
->if_u1
.if_extents
);
3653 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
3654 ifp
->if_real_bytes
= 0;
3658 * Switch from inline buffer to linear (direct) extent records.
3659 * new_size should already be rounded up to the next power of 2
3660 * by the caller (when appropriate), so use new_size as it is.
3661 * However, since new_size may be rounded up, we can't update
3662 * if_bytes here. It is the caller's responsibility to update
3663 * if_bytes upon return.
3666 xfs_iext_inline_to_direct(
3667 xfs_ifork_t
*ifp
, /* inode fork pointer */
3668 int new_size
) /* number of extents in file */
3670 ifp
->if_u1
.if_extents
= kmem_alloc(new_size
, KM_NOFS
);
3671 memset(ifp
->if_u1
.if_extents
, 0, new_size
);
3672 if (ifp
->if_bytes
) {
3673 memcpy(ifp
->if_u1
.if_extents
, ifp
->if_u2
.if_inline_ext
,
3675 memset(ifp
->if_u2
.if_inline_ext
, 0, XFS_INLINE_EXTS
*
3676 sizeof(xfs_bmbt_rec_t
));
3678 ifp
->if_real_bytes
= new_size
;
3682 * Resize an extent indirection array to new_size bytes.
3685 xfs_iext_realloc_indirect(
3686 xfs_ifork_t
*ifp
, /* inode fork pointer */
3687 int new_size
) /* new indirection array size */
3689 int nlists
; /* number of irec's (ex lists) */
3690 int size
; /* current indirection array size */
3692 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3693 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3694 size
= nlists
* sizeof(xfs_ext_irec_t
);
3695 ASSERT(ifp
->if_real_bytes
);
3696 ASSERT((new_size
>= 0) && (new_size
!= size
));
3697 if (new_size
== 0) {
3698 xfs_iext_destroy(ifp
);
3700 ifp
->if_u1
.if_ext_irec
= (xfs_ext_irec_t
*)
3701 kmem_realloc(ifp
->if_u1
.if_ext_irec
,
3702 new_size
, size
, KM_NOFS
);
3707 * Switch from indirection array to linear (direct) extent allocations.
3710 xfs_iext_indirect_to_direct(
3711 xfs_ifork_t
*ifp
) /* inode fork pointer */
3713 xfs_bmbt_rec_host_t
*ep
; /* extent record pointer */
3714 xfs_extnum_t nextents
; /* number of extents in file */
3715 int size
; /* size of file extents */
3717 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3718 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3719 ASSERT(nextents
<= XFS_LINEAR_EXTS
);
3720 size
= nextents
* sizeof(xfs_bmbt_rec_t
);
3722 xfs_iext_irec_compact_pages(ifp
);
3723 ASSERT(ifp
->if_real_bytes
== XFS_IEXT_BUFSZ
);
3725 ep
= ifp
->if_u1
.if_ext_irec
->er_extbuf
;
3726 kmem_free(ifp
->if_u1
.if_ext_irec
);
3727 ifp
->if_flags
&= ~XFS_IFEXTIREC
;
3728 ifp
->if_u1
.if_extents
= ep
;
3729 ifp
->if_bytes
= size
;
3730 if (nextents
< XFS_LINEAR_EXTS
) {
3731 xfs_iext_realloc_direct(ifp
, size
);
3736 * Free incore file extents.
3740 xfs_ifork_t
*ifp
) /* inode fork pointer */
3742 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3746 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3747 for (erp_idx
= nlists
- 1; erp_idx
>= 0 ; erp_idx
--) {
3748 xfs_iext_irec_remove(ifp
, erp_idx
);
3750 ifp
->if_flags
&= ~XFS_IFEXTIREC
;
3751 } else if (ifp
->if_real_bytes
) {
3752 kmem_free(ifp
->if_u1
.if_extents
);
3753 } else if (ifp
->if_bytes
) {
3754 memset(ifp
->if_u2
.if_inline_ext
, 0, XFS_INLINE_EXTS
*
3755 sizeof(xfs_bmbt_rec_t
));
3757 ifp
->if_u1
.if_extents
= NULL
;
3758 ifp
->if_real_bytes
= 0;
3763 * Return a pointer to the extent record for file system block bno.
3765 xfs_bmbt_rec_host_t
* /* pointer to found extent record */
3766 xfs_iext_bno_to_ext(
3767 xfs_ifork_t
*ifp
, /* inode fork pointer */
3768 xfs_fileoff_t bno
, /* block number to search for */
3769 xfs_extnum_t
*idxp
) /* index of target extent */
3771 xfs_bmbt_rec_host_t
*base
; /* pointer to first extent */
3772 xfs_filblks_t blockcount
= 0; /* number of blocks in extent */
3773 xfs_bmbt_rec_host_t
*ep
= NULL
; /* pointer to target extent */
3774 xfs_ext_irec_t
*erp
= NULL
; /* indirection array pointer */
3775 int high
; /* upper boundary in search */
3776 xfs_extnum_t idx
= 0; /* index of target extent */
3777 int low
; /* lower boundary in search */
3778 xfs_extnum_t nextents
; /* number of file extents */
3779 xfs_fileoff_t startoff
= 0; /* start offset of extent */
3781 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3782 if (nextents
== 0) {
3787 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3788 /* Find target extent list */
3790 erp
= xfs_iext_bno_to_irec(ifp
, bno
, &erp_idx
);
3791 base
= erp
->er_extbuf
;
3792 high
= erp
->er_extcount
- 1;
3794 base
= ifp
->if_u1
.if_extents
;
3795 high
= nextents
- 1;
3797 /* Binary search extent records */
3798 while (low
<= high
) {
3799 idx
= (low
+ high
) >> 1;
3801 startoff
= xfs_bmbt_get_startoff(ep
);
3802 blockcount
= xfs_bmbt_get_blockcount(ep
);
3803 if (bno
< startoff
) {
3805 } else if (bno
>= startoff
+ blockcount
) {
3808 /* Convert back to file-based extent index */
3809 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3810 idx
+= erp
->er_extoff
;
3816 /* Convert back to file-based extent index */
3817 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3818 idx
+= erp
->er_extoff
;
3820 if (bno
>= startoff
+ blockcount
) {
3821 if (++idx
== nextents
) {
3824 ep
= xfs_iext_get_ext(ifp
, idx
);
3832 * Return a pointer to the indirection array entry containing the
3833 * extent record for filesystem block bno. Store the index of the
3834 * target irec in *erp_idxp.
3836 xfs_ext_irec_t
* /* pointer to found extent record */
3837 xfs_iext_bno_to_irec(
3838 xfs_ifork_t
*ifp
, /* inode fork pointer */
3839 xfs_fileoff_t bno
, /* block number to search for */
3840 int *erp_idxp
) /* irec index of target ext list */
3842 xfs_ext_irec_t
*erp
= NULL
; /* indirection array pointer */
3843 xfs_ext_irec_t
*erp_next
; /* next indirection array entry */
3844 int erp_idx
; /* indirection array index */
3845 int nlists
; /* number of extent irec's (lists) */
3846 int high
; /* binary search upper limit */
3847 int low
; /* binary search lower limit */
3849 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3850 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3854 while (low
<= high
) {
3855 erp_idx
= (low
+ high
) >> 1;
3856 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
3857 erp_next
= erp_idx
< nlists
- 1 ? erp
+ 1 : NULL
;
3858 if (bno
< xfs_bmbt_get_startoff(erp
->er_extbuf
)) {
3860 } else if (erp_next
&& bno
>=
3861 xfs_bmbt_get_startoff(erp_next
->er_extbuf
)) {
3867 *erp_idxp
= erp_idx
;
3872 * Return a pointer to the indirection array entry containing the
3873 * extent record at file extent index *idxp. Store the index of the
3874 * target irec in *erp_idxp and store the page index of the target
3875 * extent record in *idxp.
3878 xfs_iext_idx_to_irec(
3879 xfs_ifork_t
*ifp
, /* inode fork pointer */
3880 xfs_extnum_t
*idxp
, /* extent index (file -> page) */
3881 int *erp_idxp
, /* pointer to target irec */
3882 int realloc
) /* new bytes were just added */
3884 xfs_ext_irec_t
*prev
; /* pointer to previous irec */
3885 xfs_ext_irec_t
*erp
= NULL
; /* pointer to current irec */
3886 int erp_idx
; /* indirection array index */
3887 int nlists
; /* number of irec's (ex lists) */
3888 int high
; /* binary search upper limit */
3889 int low
; /* binary search lower limit */
3890 xfs_extnum_t page_idx
= *idxp
; /* extent index in target list */
3892 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3893 ASSERT(page_idx
>= 0 && page_idx
<=
3894 ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
));
3895 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3900 /* Binary search extent irec's */
3901 while (low
<= high
) {
3902 erp_idx
= (low
+ high
) >> 1;
3903 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
3904 prev
= erp_idx
> 0 ? erp
- 1 : NULL
;
3905 if (page_idx
< erp
->er_extoff
|| (page_idx
== erp
->er_extoff
&&
3906 realloc
&& prev
&& prev
->er_extcount
< XFS_LINEAR_EXTS
)) {
3908 } else if (page_idx
> erp
->er_extoff
+ erp
->er_extcount
||
3909 (page_idx
== erp
->er_extoff
+ erp
->er_extcount
&&
3912 } else if (page_idx
== erp
->er_extoff
+ erp
->er_extcount
&&
3913 erp
->er_extcount
== XFS_LINEAR_EXTS
) {
3917 erp
= erp_idx
< nlists
? erp
+ 1 : NULL
;
3920 page_idx
-= erp
->er_extoff
;
3925 *erp_idxp
= erp_idx
;
3930 * Allocate and initialize an indirection array once the space needed
3931 * for incore extents increases above XFS_IEXT_BUFSZ.
3935 xfs_ifork_t
*ifp
) /* inode fork pointer */
3937 xfs_ext_irec_t
*erp
; /* indirection array pointer */
3938 xfs_extnum_t nextents
; /* number of extents in file */
3940 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
));
3941 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3942 ASSERT(nextents
<= XFS_LINEAR_EXTS
);
3944 erp
= kmem_alloc(sizeof(xfs_ext_irec_t
), KM_NOFS
);
3946 if (nextents
== 0) {
3947 ifp
->if_u1
.if_extents
= kmem_alloc(XFS_IEXT_BUFSZ
, KM_NOFS
);
3948 } else if (!ifp
->if_real_bytes
) {
3949 xfs_iext_inline_to_direct(ifp
, XFS_IEXT_BUFSZ
);
3950 } else if (ifp
->if_real_bytes
< XFS_IEXT_BUFSZ
) {
3951 xfs_iext_realloc_direct(ifp
, XFS_IEXT_BUFSZ
);
3953 erp
->er_extbuf
= ifp
->if_u1
.if_extents
;
3954 erp
->er_extcount
= nextents
;
3957 ifp
->if_flags
|= XFS_IFEXTIREC
;
3958 ifp
->if_real_bytes
= XFS_IEXT_BUFSZ
;
3959 ifp
->if_bytes
= nextents
* sizeof(xfs_bmbt_rec_t
);
3960 ifp
->if_u1
.if_ext_irec
= erp
;
3966 * Allocate and initialize a new entry in the indirection array.
3970 xfs_ifork_t
*ifp
, /* inode fork pointer */
3971 int erp_idx
) /* index for new irec */
3973 xfs_ext_irec_t
*erp
; /* indirection array pointer */
3974 int i
; /* loop counter */
3975 int nlists
; /* number of irec's (ex lists) */
3977 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3978 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3980 /* Resize indirection array */
3981 xfs_iext_realloc_indirect(ifp
, ++nlists
*
3982 sizeof(xfs_ext_irec_t
));
3984 * Move records down in the array so the
3985 * new page can use erp_idx.
3987 erp
= ifp
->if_u1
.if_ext_irec
;
3988 for (i
= nlists
- 1; i
> erp_idx
; i
--) {
3989 memmove(&erp
[i
], &erp
[i
-1], sizeof(xfs_ext_irec_t
));
3991 ASSERT(i
== erp_idx
);
3993 /* Initialize new extent record */
3994 erp
= ifp
->if_u1
.if_ext_irec
;
3995 erp
[erp_idx
].er_extbuf
= kmem_alloc(XFS_IEXT_BUFSZ
, KM_NOFS
);
3996 ifp
->if_real_bytes
= nlists
* XFS_IEXT_BUFSZ
;
3997 memset(erp
[erp_idx
].er_extbuf
, 0, XFS_IEXT_BUFSZ
);
3998 erp
[erp_idx
].er_extcount
= 0;
3999 erp
[erp_idx
].er_extoff
= erp_idx
> 0 ?
4000 erp
[erp_idx
-1].er_extoff
+ erp
[erp_idx
-1].er_extcount
: 0;
4001 return (&erp
[erp_idx
]);
4005 * Remove a record from the indirection array.
4008 xfs_iext_irec_remove(
4009 xfs_ifork_t
*ifp
, /* inode fork pointer */
4010 int erp_idx
) /* irec index to remove */
4012 xfs_ext_irec_t
*erp
; /* indirection array pointer */
4013 int i
; /* loop counter */
4014 int nlists
; /* number of irec's (ex lists) */
4016 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4017 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4018 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4019 if (erp
->er_extbuf
) {
4020 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1,
4022 kmem_free(erp
->er_extbuf
);
4024 /* Compact extent records */
4025 erp
= ifp
->if_u1
.if_ext_irec
;
4026 for (i
= erp_idx
; i
< nlists
- 1; i
++) {
4027 memmove(&erp
[i
], &erp
[i
+1], sizeof(xfs_ext_irec_t
));
4030 * Manually free the last extent record from the indirection
4031 * array. A call to xfs_iext_realloc_indirect() with a size
4032 * of zero would result in a call to xfs_iext_destroy() which
4033 * would in turn call this function again, creating a nasty
4037 xfs_iext_realloc_indirect(ifp
,
4038 nlists
* sizeof(xfs_ext_irec_t
));
4040 kmem_free(ifp
->if_u1
.if_ext_irec
);
4042 ifp
->if_real_bytes
= nlists
* XFS_IEXT_BUFSZ
;
4046 * This is called to clean up large amounts of unused memory allocated
4047 * by the indirection array. Before compacting anything though, verify
4048 * that the indirection array is still needed and switch back to the
4049 * linear extent list (or even the inline buffer) if possible. The
4050 * compaction policy is as follows:
4052 * Full Compaction: Extents fit into a single page (or inline buffer)
4053 * Partial Compaction: Extents occupy less than 50% of allocated space
4054 * No Compaction: Extents occupy at least 50% of allocated space
4057 xfs_iext_irec_compact(
4058 xfs_ifork_t
*ifp
) /* inode fork pointer */
4060 xfs_extnum_t nextents
; /* number of extents in file */
4061 int nlists
; /* number of irec's (ex lists) */
4063 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4064 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4065 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4067 if (nextents
== 0) {
4068 xfs_iext_destroy(ifp
);
4069 } else if (nextents
<= XFS_INLINE_EXTS
) {
4070 xfs_iext_indirect_to_direct(ifp
);
4071 xfs_iext_direct_to_inline(ifp
, nextents
);
4072 } else if (nextents
<= XFS_LINEAR_EXTS
) {
4073 xfs_iext_indirect_to_direct(ifp
);
4074 } else if (nextents
< (nlists
* XFS_LINEAR_EXTS
) >> 1) {
4075 xfs_iext_irec_compact_pages(ifp
);
4080 * Combine extents from neighboring extent pages.
4083 xfs_iext_irec_compact_pages(
4084 xfs_ifork_t
*ifp
) /* inode fork pointer */
4086 xfs_ext_irec_t
*erp
, *erp_next
;/* pointers to irec entries */
4087 int erp_idx
= 0; /* indirection array index */
4088 int nlists
; /* number of irec's (ex lists) */
4090 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4091 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4092 while (erp_idx
< nlists
- 1) {
4093 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4095 if (erp_next
->er_extcount
<=
4096 (XFS_LINEAR_EXTS
- erp
->er_extcount
)) {
4097 memcpy(&erp
->er_extbuf
[erp
->er_extcount
],
4098 erp_next
->er_extbuf
, erp_next
->er_extcount
*
4099 sizeof(xfs_bmbt_rec_t
));
4100 erp
->er_extcount
+= erp_next
->er_extcount
;
4102 * Free page before removing extent record
4103 * so er_extoffs don't get modified in
4104 * xfs_iext_irec_remove.
4106 kmem_free(erp_next
->er_extbuf
);
4107 erp_next
->er_extbuf
= NULL
;
4108 xfs_iext_irec_remove(ifp
, erp_idx
+ 1);
4109 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4117 * This is called to update the er_extoff field in the indirection
4118 * array when extents have been added or removed from one of the
4119 * extent lists. erp_idx contains the irec index to begin updating
4120 * at and ext_diff contains the number of extents that were added
4124 xfs_iext_irec_update_extoffs(
4125 xfs_ifork_t
*ifp
, /* inode fork pointer */
4126 int erp_idx
, /* irec index to update */
4127 int ext_diff
) /* number of new extents */
4129 int i
; /* loop counter */
4130 int nlists
; /* number of irec's (ex lists */
4132 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4133 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4134 for (i
= erp_idx
; i
< nlists
; i
++) {
4135 ifp
->if_u1
.if_ext_irec
[i
].er_extoff
+= ext_diff
;