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"
25 #include "xfs_trans.h"
26 #include "xfs_trans_priv.h"
29 #include "xfs_mount.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_alloc_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_attr_sf.h"
34 #include "xfs_dinode.h"
35 #include "xfs_inode.h"
36 #include "xfs_buf_item.h"
37 #include "xfs_inode_item.h"
38 #include "xfs_btree.h"
39 #include "xfs_alloc.h"
40 #include "xfs_ialloc.h"
42 #include "xfs_error.h"
43 #include "xfs_utils.h"
44 #include "xfs_quota.h"
45 #include "xfs_filestream.h"
46 #include "xfs_vnodeops.h"
47 #include "xfs_trace.h"
49 kmem_zone_t
*xfs_ifork_zone
;
50 kmem_zone_t
*xfs_inode_zone
;
53 * Used in xfs_itruncate_extents(). This is the maximum number of extents
54 * freed from a file in a single transaction.
56 #define XFS_ITRUNC_MAX_EXTENTS 2
58 STATIC
int xfs_iflush_int(xfs_inode_t
*, xfs_buf_t
*);
59 STATIC
int xfs_iformat_local(xfs_inode_t
*, xfs_dinode_t
*, int, int);
60 STATIC
int xfs_iformat_extents(xfs_inode_t
*, xfs_dinode_t
*, int);
61 STATIC
int xfs_iformat_btree(xfs_inode_t
*, xfs_dinode_t
*, int);
64 * helper function to extract extent size hint from inode
70 if ((ip
->i_d
.di_flags
& XFS_DIFLAG_EXTSIZE
) && ip
->i_d
.di_extsize
)
71 return ip
->i_d
.di_extsize
;
72 if (XFS_IS_REALTIME_INODE(ip
))
73 return ip
->i_mount
->m_sb
.sb_rextsize
;
79 * Make sure that the extents in the given memory buffer
89 xfs_bmbt_rec_host_t rec
;
92 for (i
= 0; i
< nrecs
; i
++) {
93 xfs_bmbt_rec_host_t
*ep
= xfs_iext_get_ext(ifp
, i
);
94 rec
.l0
= get_unaligned(&ep
->l0
);
95 rec
.l1
= get_unaligned(&ep
->l1
);
96 xfs_bmbt_get_all(&rec
, &irec
);
97 if (fmt
== XFS_EXTFMT_NOSTATE
)
98 ASSERT(irec
.br_state
== XFS_EXT_NORM
);
102 #define xfs_validate_extents(ifp, nrecs, fmt)
106 * Check that none of the inode's in the buffer have a next
107 * unlinked field of 0.
119 j
= mp
->m_inode_cluster_size
>> mp
->m_sb
.sb_inodelog
;
121 for (i
= 0; i
< j
; i
++) {
122 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
,
123 i
* mp
->m_sb
.sb_inodesize
);
124 if (!dip
->di_next_unlinked
) {
126 "Detected bogus zero next_unlinked field in incore inode buffer 0x%p.",
128 ASSERT(dip
->di_next_unlinked
);
135 * Find the buffer associated with the given inode map
136 * We do basic validation checks on the buffer once it has been
137 * retrieved from disk.
143 struct xfs_imap
*imap
,
153 buf_flags
|= XBF_UNMAPPED
;
154 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, imap
->im_blkno
,
155 (int)imap
->im_len
, buf_flags
, &bp
);
157 if (error
!= EAGAIN
) {
159 "%s: xfs_trans_read_buf() returned error %d.",
162 ASSERT(buf_flags
& XBF_TRYLOCK
);
168 * Validate the magic number and version of every inode in the buffer
169 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
172 ni
= BBTOB(imap
->im_len
) >> mp
->m_sb
.sb_inodelog
;
173 #else /* usual case */
177 for (i
= 0; i
< ni
; i
++) {
181 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
,
182 (i
<< mp
->m_sb
.sb_inodelog
));
183 di_ok
= dip
->di_magic
== cpu_to_be16(XFS_DINODE_MAGIC
) &&
184 XFS_DINODE_GOOD_VERSION(dip
->di_version
);
185 if (unlikely(XFS_TEST_ERROR(!di_ok
, mp
,
186 XFS_ERRTAG_ITOBP_INOTOBP
,
187 XFS_RANDOM_ITOBP_INOTOBP
))) {
188 if (iget_flags
& XFS_IGET_UNTRUSTED
) {
189 xfs_trans_brelse(tp
, bp
);
190 return XFS_ERROR(EINVAL
);
192 XFS_CORRUPTION_ERROR("xfs_imap_to_bp",
193 XFS_ERRLEVEL_HIGH
, mp
, dip
);
196 "bad inode magic/vsn daddr %lld #%d (magic=%x)",
197 (unsigned long long)imap
->im_blkno
, i
,
198 be16_to_cpu(dip
->di_magic
));
201 xfs_trans_brelse(tp
, bp
);
202 return XFS_ERROR(EFSCORRUPTED
);
206 xfs_inobp_check(mp
, bp
);
212 * This routine is called to map an inode number within a file
213 * system to the buffer containing the on-disk version of the
214 * inode. It returns a pointer to the buffer containing the
215 * on-disk inode in the bpp parameter, and in the dip parameter
216 * it returns a pointer to the on-disk inode within that buffer.
218 * If a non-zero error is returned, then the contents of bpp and
219 * dipp are undefined.
221 * Use xfs_imap() to determine the size and location of the
222 * buffer to read from disk.
234 struct xfs_imap imap
;
239 error
= xfs_imap(mp
, tp
, ino
, &imap
, imap_flags
);
243 error
= xfs_imap_to_bp(mp
, tp
, &imap
, &bp
, 0, imap_flags
);
247 *dipp
= (xfs_dinode_t
*)xfs_buf_offset(bp
, imap
.im_boffset
);
249 *offset
= imap
.im_boffset
;
255 * This routine is called to map an inode to the buffer containing
256 * the on-disk version of the inode. It returns a pointer to the
257 * buffer containing the on-disk inode in the bpp parameter, and in
258 * the dip parameter it returns a pointer to the on-disk inode within
261 * If a non-zero error is returned, then the contents of bpp and
262 * dipp are undefined.
264 * The inode is expected to already been mapped to its buffer and read
265 * in once, thus we can use the mapping information stored in the inode
266 * rather than calling xfs_imap(). This allows us to avoid the overhead
267 * of looking at the inode btree for small block file systems
282 ASSERT(ip
->i_imap
.im_blkno
!= 0);
284 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &bp
, buf_flags
, 0);
289 ASSERT(buf_flags
& XBF_TRYLOCK
);
295 *dipp
= (xfs_dinode_t
*)xfs_buf_offset(bp
, ip
->i_imap
.im_boffset
);
301 * Move inode type and inode format specific information from the
302 * on-disk inode to the in-core inode. For fifos, devs, and sockets
303 * this means set if_rdev to the proper value. For files, directories,
304 * and symlinks this means to bring in the in-line data or extent
305 * pointers. For a file in B-tree format, only the root is immediately
306 * brought in-core. The rest will be in-lined in if_extents when it
307 * is first referenced (see xfs_iread_extents()).
314 xfs_attr_shortform_t
*atp
;
319 if (unlikely(be32_to_cpu(dip
->di_nextents
) +
320 be16_to_cpu(dip
->di_anextents
) >
321 be64_to_cpu(dip
->di_nblocks
))) {
322 xfs_warn(ip
->i_mount
,
323 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
324 (unsigned long long)ip
->i_ino
,
325 (int)(be32_to_cpu(dip
->di_nextents
) +
326 be16_to_cpu(dip
->di_anextents
)),
328 be64_to_cpu(dip
->di_nblocks
));
329 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW
,
331 return XFS_ERROR(EFSCORRUPTED
);
334 if (unlikely(dip
->di_forkoff
> ip
->i_mount
->m_sb
.sb_inodesize
)) {
335 xfs_warn(ip
->i_mount
, "corrupt dinode %Lu, forkoff = 0x%x.",
336 (unsigned long long)ip
->i_ino
,
338 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW
,
340 return XFS_ERROR(EFSCORRUPTED
);
343 if (unlikely((ip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
) &&
344 !ip
->i_mount
->m_rtdev_targp
)) {
345 xfs_warn(ip
->i_mount
,
346 "corrupt dinode %Lu, has realtime flag set.",
348 XFS_CORRUPTION_ERROR("xfs_iformat(realtime)",
349 XFS_ERRLEVEL_LOW
, ip
->i_mount
, dip
);
350 return XFS_ERROR(EFSCORRUPTED
);
353 switch (ip
->i_d
.di_mode
& S_IFMT
) {
358 if (unlikely(dip
->di_format
!= XFS_DINODE_FMT_DEV
)) {
359 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW
,
361 return XFS_ERROR(EFSCORRUPTED
);
364 ip
->i_df
.if_u2
.if_rdev
= xfs_dinode_get_rdev(dip
);
370 switch (dip
->di_format
) {
371 case XFS_DINODE_FMT_LOCAL
:
373 * no local regular files yet
375 if (unlikely(S_ISREG(be16_to_cpu(dip
->di_mode
)))) {
376 xfs_warn(ip
->i_mount
,
377 "corrupt inode %Lu (local format for regular file).",
378 (unsigned long long) ip
->i_ino
);
379 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
382 return XFS_ERROR(EFSCORRUPTED
);
385 di_size
= be64_to_cpu(dip
->di_size
);
386 if (unlikely(di_size
> XFS_DFORK_DSIZE(dip
, ip
->i_mount
))) {
387 xfs_warn(ip
->i_mount
,
388 "corrupt inode %Lu (bad size %Ld for local inode).",
389 (unsigned long long) ip
->i_ino
,
390 (long long) di_size
);
391 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
394 return XFS_ERROR(EFSCORRUPTED
);
398 error
= xfs_iformat_local(ip
, dip
, XFS_DATA_FORK
, size
);
400 case XFS_DINODE_FMT_EXTENTS
:
401 error
= xfs_iformat_extents(ip
, dip
, XFS_DATA_FORK
);
403 case XFS_DINODE_FMT_BTREE
:
404 error
= xfs_iformat_btree(ip
, dip
, XFS_DATA_FORK
);
407 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW
,
409 return XFS_ERROR(EFSCORRUPTED
);
414 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW
, ip
->i_mount
);
415 return XFS_ERROR(EFSCORRUPTED
);
420 if (!XFS_DFORK_Q(dip
))
423 ASSERT(ip
->i_afp
== NULL
);
424 ip
->i_afp
= kmem_zone_zalloc(xfs_ifork_zone
, KM_SLEEP
| KM_NOFS
);
426 switch (dip
->di_aformat
) {
427 case XFS_DINODE_FMT_LOCAL
:
428 atp
= (xfs_attr_shortform_t
*)XFS_DFORK_APTR(dip
);
429 size
= be16_to_cpu(atp
->hdr
.totsize
);
431 if (unlikely(size
< sizeof(struct xfs_attr_sf_hdr
))) {
432 xfs_warn(ip
->i_mount
,
433 "corrupt inode %Lu (bad attr fork size %Ld).",
434 (unsigned long long) ip
->i_ino
,
436 XFS_CORRUPTION_ERROR("xfs_iformat(8)",
439 return XFS_ERROR(EFSCORRUPTED
);
442 error
= xfs_iformat_local(ip
, dip
, XFS_ATTR_FORK
, size
);
444 case XFS_DINODE_FMT_EXTENTS
:
445 error
= xfs_iformat_extents(ip
, dip
, XFS_ATTR_FORK
);
447 case XFS_DINODE_FMT_BTREE
:
448 error
= xfs_iformat_btree(ip
, dip
, XFS_ATTR_FORK
);
451 error
= XFS_ERROR(EFSCORRUPTED
);
455 kmem_zone_free(xfs_ifork_zone
, ip
->i_afp
);
457 xfs_idestroy_fork(ip
, XFS_DATA_FORK
);
463 * The file is in-lined in the on-disk inode.
464 * If it fits into if_inline_data, then copy
465 * it there, otherwise allocate a buffer for it
466 * and copy the data there. Either way, set
467 * if_data to point at the data.
468 * If we allocate a buffer for the data, make
469 * sure that its size is a multiple of 4 and
470 * record the real size in i_real_bytes.
483 * If the size is unreasonable, then something
484 * is wrong and we just bail out rather than crash in
485 * kmem_alloc() or memcpy() below.
487 if (unlikely(size
> XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
))) {
488 xfs_warn(ip
->i_mount
,
489 "corrupt inode %Lu (bad size %d for local fork, size = %d).",
490 (unsigned long long) ip
->i_ino
, size
,
491 XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
));
492 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW
,
494 return XFS_ERROR(EFSCORRUPTED
);
496 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
499 ifp
->if_u1
.if_data
= NULL
;
500 else if (size
<= sizeof(ifp
->if_u2
.if_inline_data
))
501 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
503 real_size
= roundup(size
, 4);
504 ifp
->if_u1
.if_data
= kmem_alloc(real_size
, KM_SLEEP
| KM_NOFS
);
506 ifp
->if_bytes
= size
;
507 ifp
->if_real_bytes
= real_size
;
509 memcpy(ifp
->if_u1
.if_data
, XFS_DFORK_PTR(dip
, whichfork
), size
);
510 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
511 ifp
->if_flags
|= XFS_IFINLINE
;
516 * The file consists of a set of extents all
517 * of which fit into the on-disk inode.
518 * If there are few enough extents to fit into
519 * the if_inline_ext, then copy them there.
520 * Otherwise allocate a buffer for them and copy
521 * them into it. Either way, set if_extents
522 * to point at the extents.
536 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
537 nex
= XFS_DFORK_NEXTENTS(dip
, whichfork
);
538 size
= nex
* (uint
)sizeof(xfs_bmbt_rec_t
);
541 * If the number of extents is unreasonable, then something
542 * is wrong and we just bail out rather than crash in
543 * kmem_alloc() or memcpy() below.
545 if (unlikely(size
< 0 || size
> XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
))) {
546 xfs_warn(ip
->i_mount
, "corrupt inode %Lu ((a)extents = %d).",
547 (unsigned long long) ip
->i_ino
, nex
);
548 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW
,
550 return XFS_ERROR(EFSCORRUPTED
);
553 ifp
->if_real_bytes
= 0;
555 ifp
->if_u1
.if_extents
= NULL
;
556 else if (nex
<= XFS_INLINE_EXTS
)
557 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
559 xfs_iext_add(ifp
, 0, nex
);
561 ifp
->if_bytes
= size
;
563 dp
= (xfs_bmbt_rec_t
*) XFS_DFORK_PTR(dip
, whichfork
);
564 xfs_validate_extents(ifp
, nex
, XFS_EXTFMT_INODE(ip
));
565 for (i
= 0; i
< nex
; i
++, dp
++) {
566 xfs_bmbt_rec_host_t
*ep
= xfs_iext_get_ext(ifp
, i
);
567 ep
->l0
= get_unaligned_be64(&dp
->l0
);
568 ep
->l1
= get_unaligned_be64(&dp
->l1
);
570 XFS_BMAP_TRACE_EXLIST(ip
, nex
, whichfork
);
571 if (whichfork
!= XFS_DATA_FORK
||
572 XFS_EXTFMT_INODE(ip
) == XFS_EXTFMT_NOSTATE
)
573 if (unlikely(xfs_check_nostate_extents(
575 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
578 return XFS_ERROR(EFSCORRUPTED
);
581 ifp
->if_flags
|= XFS_IFEXTENTS
;
586 * The file has too many extents to fit into
587 * the inode, so they are in B-tree format.
588 * Allocate a buffer for the root of the B-tree
589 * and copy the root into it. The i_extents
590 * field will remain NULL until all of the
591 * extents are read in (when they are needed).
599 xfs_bmdr_block_t
*dfp
;
605 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
606 dfp
= (xfs_bmdr_block_t
*)XFS_DFORK_PTR(dip
, whichfork
);
607 size
= XFS_BMAP_BROOT_SPACE(dfp
);
608 nrecs
= be16_to_cpu(dfp
->bb_numrecs
);
611 * blow out if -- fork has less extents than can fit in
612 * fork (fork shouldn't be a btree format), root btree
613 * block has more records than can fit into the fork,
614 * or the number of extents is greater than the number of
617 if (unlikely(XFS_IFORK_NEXTENTS(ip
, whichfork
) <=
618 XFS_IFORK_MAXEXT(ip
, whichfork
) ||
619 XFS_BMDR_SPACE_CALC(nrecs
) >
620 XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
) ||
621 XFS_IFORK_NEXTENTS(ip
, whichfork
) > ip
->i_d
.di_nblocks
)) {
622 xfs_warn(ip
->i_mount
, "corrupt inode %Lu (btree).",
623 (unsigned long long) ip
->i_ino
);
624 XFS_CORRUPTION_ERROR("xfs_iformat_btree", XFS_ERRLEVEL_LOW
,
626 return XFS_ERROR(EFSCORRUPTED
);
629 ifp
->if_broot_bytes
= size
;
630 ifp
->if_broot
= kmem_alloc(size
, KM_SLEEP
| KM_NOFS
);
631 ASSERT(ifp
->if_broot
!= NULL
);
633 * Copy and convert from the on-disk structure
634 * to the in-memory structure.
636 xfs_bmdr_to_bmbt(ip
->i_mount
, dfp
,
637 XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
),
638 ifp
->if_broot
, size
);
639 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
640 ifp
->if_flags
|= XFS_IFBROOT
;
646 xfs_dinode_from_disk(
650 to
->di_magic
= be16_to_cpu(from
->di_magic
);
651 to
->di_mode
= be16_to_cpu(from
->di_mode
);
652 to
->di_version
= from
->di_version
;
653 to
->di_format
= from
->di_format
;
654 to
->di_onlink
= be16_to_cpu(from
->di_onlink
);
655 to
->di_uid
= be32_to_cpu(from
->di_uid
);
656 to
->di_gid
= be32_to_cpu(from
->di_gid
);
657 to
->di_nlink
= be32_to_cpu(from
->di_nlink
);
658 to
->di_projid_lo
= be16_to_cpu(from
->di_projid_lo
);
659 to
->di_projid_hi
= be16_to_cpu(from
->di_projid_hi
);
660 memcpy(to
->di_pad
, from
->di_pad
, sizeof(to
->di_pad
));
661 to
->di_flushiter
= be16_to_cpu(from
->di_flushiter
);
662 to
->di_atime
.t_sec
= be32_to_cpu(from
->di_atime
.t_sec
);
663 to
->di_atime
.t_nsec
= be32_to_cpu(from
->di_atime
.t_nsec
);
664 to
->di_mtime
.t_sec
= be32_to_cpu(from
->di_mtime
.t_sec
);
665 to
->di_mtime
.t_nsec
= be32_to_cpu(from
->di_mtime
.t_nsec
);
666 to
->di_ctime
.t_sec
= be32_to_cpu(from
->di_ctime
.t_sec
);
667 to
->di_ctime
.t_nsec
= be32_to_cpu(from
->di_ctime
.t_nsec
);
668 to
->di_size
= be64_to_cpu(from
->di_size
);
669 to
->di_nblocks
= be64_to_cpu(from
->di_nblocks
);
670 to
->di_extsize
= be32_to_cpu(from
->di_extsize
);
671 to
->di_nextents
= be32_to_cpu(from
->di_nextents
);
672 to
->di_anextents
= be16_to_cpu(from
->di_anextents
);
673 to
->di_forkoff
= from
->di_forkoff
;
674 to
->di_aformat
= from
->di_aformat
;
675 to
->di_dmevmask
= be32_to_cpu(from
->di_dmevmask
);
676 to
->di_dmstate
= be16_to_cpu(from
->di_dmstate
);
677 to
->di_flags
= be16_to_cpu(from
->di_flags
);
678 to
->di_gen
= be32_to_cpu(from
->di_gen
);
684 xfs_icdinode_t
*from
)
686 to
->di_magic
= cpu_to_be16(from
->di_magic
);
687 to
->di_mode
= cpu_to_be16(from
->di_mode
);
688 to
->di_version
= from
->di_version
;
689 to
->di_format
= from
->di_format
;
690 to
->di_onlink
= cpu_to_be16(from
->di_onlink
);
691 to
->di_uid
= cpu_to_be32(from
->di_uid
);
692 to
->di_gid
= cpu_to_be32(from
->di_gid
);
693 to
->di_nlink
= cpu_to_be32(from
->di_nlink
);
694 to
->di_projid_lo
= cpu_to_be16(from
->di_projid_lo
);
695 to
->di_projid_hi
= cpu_to_be16(from
->di_projid_hi
);
696 memcpy(to
->di_pad
, from
->di_pad
, sizeof(to
->di_pad
));
697 to
->di_flushiter
= cpu_to_be16(from
->di_flushiter
);
698 to
->di_atime
.t_sec
= cpu_to_be32(from
->di_atime
.t_sec
);
699 to
->di_atime
.t_nsec
= cpu_to_be32(from
->di_atime
.t_nsec
);
700 to
->di_mtime
.t_sec
= cpu_to_be32(from
->di_mtime
.t_sec
);
701 to
->di_mtime
.t_nsec
= cpu_to_be32(from
->di_mtime
.t_nsec
);
702 to
->di_ctime
.t_sec
= cpu_to_be32(from
->di_ctime
.t_sec
);
703 to
->di_ctime
.t_nsec
= cpu_to_be32(from
->di_ctime
.t_nsec
);
704 to
->di_size
= cpu_to_be64(from
->di_size
);
705 to
->di_nblocks
= cpu_to_be64(from
->di_nblocks
);
706 to
->di_extsize
= cpu_to_be32(from
->di_extsize
);
707 to
->di_nextents
= cpu_to_be32(from
->di_nextents
);
708 to
->di_anextents
= cpu_to_be16(from
->di_anextents
);
709 to
->di_forkoff
= from
->di_forkoff
;
710 to
->di_aformat
= from
->di_aformat
;
711 to
->di_dmevmask
= cpu_to_be32(from
->di_dmevmask
);
712 to
->di_dmstate
= cpu_to_be16(from
->di_dmstate
);
713 to
->di_flags
= cpu_to_be16(from
->di_flags
);
714 to
->di_gen
= cpu_to_be32(from
->di_gen
);
723 if (di_flags
& XFS_DIFLAG_ANY
) {
724 if (di_flags
& XFS_DIFLAG_REALTIME
)
725 flags
|= XFS_XFLAG_REALTIME
;
726 if (di_flags
& XFS_DIFLAG_PREALLOC
)
727 flags
|= XFS_XFLAG_PREALLOC
;
728 if (di_flags
& XFS_DIFLAG_IMMUTABLE
)
729 flags
|= XFS_XFLAG_IMMUTABLE
;
730 if (di_flags
& XFS_DIFLAG_APPEND
)
731 flags
|= XFS_XFLAG_APPEND
;
732 if (di_flags
& XFS_DIFLAG_SYNC
)
733 flags
|= XFS_XFLAG_SYNC
;
734 if (di_flags
& XFS_DIFLAG_NOATIME
)
735 flags
|= XFS_XFLAG_NOATIME
;
736 if (di_flags
& XFS_DIFLAG_NODUMP
)
737 flags
|= XFS_XFLAG_NODUMP
;
738 if (di_flags
& XFS_DIFLAG_RTINHERIT
)
739 flags
|= XFS_XFLAG_RTINHERIT
;
740 if (di_flags
& XFS_DIFLAG_PROJINHERIT
)
741 flags
|= XFS_XFLAG_PROJINHERIT
;
742 if (di_flags
& XFS_DIFLAG_NOSYMLINKS
)
743 flags
|= XFS_XFLAG_NOSYMLINKS
;
744 if (di_flags
& XFS_DIFLAG_EXTSIZE
)
745 flags
|= XFS_XFLAG_EXTSIZE
;
746 if (di_flags
& XFS_DIFLAG_EXTSZINHERIT
)
747 flags
|= XFS_XFLAG_EXTSZINHERIT
;
748 if (di_flags
& XFS_DIFLAG_NODEFRAG
)
749 flags
|= XFS_XFLAG_NODEFRAG
;
750 if (di_flags
& XFS_DIFLAG_FILESTREAM
)
751 flags
|= XFS_XFLAG_FILESTREAM
;
761 xfs_icdinode_t
*dic
= &ip
->i_d
;
763 return _xfs_dic2xflags(dic
->di_flags
) |
764 (XFS_IFORK_Q(ip
) ? XFS_XFLAG_HASATTR
: 0);
771 return _xfs_dic2xflags(be16_to_cpu(dip
->di_flags
)) |
772 (XFS_DFORK_Q(dip
) ? XFS_XFLAG_HASATTR
: 0);
776 * Read the disk inode attributes into the in-core inode structure.
790 * Fill in the location information in the in-core inode.
792 error
= xfs_imap(mp
, tp
, ip
->i_ino
, &ip
->i_imap
, iget_flags
);
797 * Get pointers to the on-disk inode and the buffer containing it.
799 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &bp
, 0, iget_flags
);
802 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, ip
->i_imap
.im_boffset
);
805 * If we got something that isn't an inode it means someone
806 * (nfs or dmi) has a stale handle.
808 if (dip
->di_magic
!= cpu_to_be16(XFS_DINODE_MAGIC
)) {
811 "%s: dip->di_magic (0x%x) != XFS_DINODE_MAGIC (0x%x)",
812 __func__
, be16_to_cpu(dip
->di_magic
), XFS_DINODE_MAGIC
);
814 error
= XFS_ERROR(EINVAL
);
819 * If the on-disk inode is already linked to a directory
820 * entry, copy all of the inode into the in-core inode.
821 * xfs_iformat() handles copying in the inode format
822 * specific information.
823 * Otherwise, just get the truly permanent information.
826 xfs_dinode_from_disk(&ip
->i_d
, dip
);
827 error
= xfs_iformat(ip
, dip
);
830 xfs_alert(mp
, "%s: xfs_iformat() returned error %d",
836 ip
->i_d
.di_magic
= be16_to_cpu(dip
->di_magic
);
837 ip
->i_d
.di_version
= dip
->di_version
;
838 ip
->i_d
.di_gen
= be32_to_cpu(dip
->di_gen
);
839 ip
->i_d
.di_flushiter
= be16_to_cpu(dip
->di_flushiter
);
841 * Make sure to pull in the mode here as well in
842 * case the inode is released without being used.
843 * This ensures that xfs_inactive() will see that
844 * the inode is already free and not try to mess
845 * with the uninitialized part of it.
851 * The inode format changed when we moved the link count and
852 * made it 32 bits long. If this is an old format inode,
853 * convert it in memory to look like a new one. If it gets
854 * flushed to disk we will convert back before flushing or
855 * logging it. We zero out the new projid field and the old link
856 * count field. We'll handle clearing the pad field (the remains
857 * of the old uuid field) when we actually convert the inode to
858 * the new format. We don't change the version number so that we
859 * can distinguish this from a real new format inode.
861 if (ip
->i_d
.di_version
== 1) {
862 ip
->i_d
.di_nlink
= ip
->i_d
.di_onlink
;
863 ip
->i_d
.di_onlink
= 0;
864 xfs_set_projid(ip
, 0);
867 ip
->i_delayed_blks
= 0;
870 * Mark the buffer containing the inode as something to keep
871 * around for a while. This helps to keep recently accessed
872 * meta-data in-core longer.
874 xfs_buf_set_ref(bp
, XFS_INO_REF
);
877 * Use xfs_trans_brelse() to release the buffer containing the
878 * on-disk inode, because it was acquired with xfs_trans_read_buf()
879 * in xfs_itobp() above. If tp is NULL, this is just a normal
880 * brelse(). If we're within a transaction, then xfs_trans_brelse()
881 * will only release the buffer if it is not dirty within the
882 * transaction. It will be OK to release the buffer in this case,
883 * because inodes on disk are never destroyed and we will be
884 * locking the new in-core inode before putting it in the hash
885 * table where other processes can find it. Thus we don't have
886 * to worry about the inode being changed just because we released
890 xfs_trans_brelse(tp
, bp
);
895 * Read in extents from a btree-format inode.
896 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
906 xfs_extnum_t nextents
;
908 if (unlikely(XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
)) {
909 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW
,
911 return XFS_ERROR(EFSCORRUPTED
);
913 nextents
= XFS_IFORK_NEXTENTS(ip
, whichfork
);
914 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
917 * We know that the size is valid (it's checked in iformat_btree)
919 ifp
->if_bytes
= ifp
->if_real_bytes
= 0;
920 ifp
->if_flags
|= XFS_IFEXTENTS
;
921 xfs_iext_add(ifp
, 0, nextents
);
922 error
= xfs_bmap_read_extents(tp
, ip
, whichfork
);
924 xfs_iext_destroy(ifp
);
925 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
928 xfs_validate_extents(ifp
, nextents
, XFS_EXTFMT_INODE(ip
));
933 * Allocate an inode on disk and return a copy of its in-core version.
934 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
935 * appropriately within the inode. The uid and gid for the inode are
936 * set according to the contents of the given cred structure.
938 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
939 * has a free inode available, call xfs_iget()
940 * to obtain the in-core version of the allocated inode. Finally,
941 * fill in the inode and log its initial contents. In this case,
942 * ialloc_context would be set to NULL and call_again set to false.
944 * If xfs_dialloc() does not have an available inode,
945 * it will replenish its supply by doing an allocation. Since we can
946 * only do one allocation within a transaction without deadlocks, we
947 * must commit the current transaction before returning the inode itself.
948 * In this case, therefore, we will set call_again to true and return.
949 * The caller should then commit the current transaction, start a new
950 * transaction, and call xfs_ialloc() again to actually get the inode.
952 * To ensure that some other process does not grab the inode that
953 * was allocated during the first call to xfs_ialloc(), this routine
954 * also returns the [locked] bp pointing to the head of the freelist
955 * as ialloc_context. The caller should hold this buffer across
956 * the commit and pass it back into this routine on the second call.
958 * If we are allocating quota inodes, we do not have a parent inode
959 * to attach to or associate with (i.e. pip == NULL) because they
960 * are not linked into the directory structure - they are attached
961 * directly to the superblock - and so have no parent.
972 xfs_buf_t
**ialloc_context
,
973 boolean_t
*call_again
,
984 * Call the space management code to pick
985 * the on-disk inode to be allocated.
987 error
= xfs_dialloc(tp
, pip
? pip
->i_ino
: 0, mode
, okalloc
,
988 ialloc_context
, call_again
, &ino
);
991 if (*call_again
|| ino
== NULLFSINO
) {
995 ASSERT(*ialloc_context
== NULL
);
998 * Get the in-core inode with the lock held exclusively.
999 * This is because we're setting fields here we need
1000 * to prevent others from looking at until we're done.
1002 error
= xfs_iget(tp
->t_mountp
, tp
, ino
, XFS_IGET_CREATE
,
1003 XFS_ILOCK_EXCL
, &ip
);
1008 ip
->i_d
.di_mode
= mode
;
1009 ip
->i_d
.di_onlink
= 0;
1010 ip
->i_d
.di_nlink
= nlink
;
1011 ASSERT(ip
->i_d
.di_nlink
== nlink
);
1012 ip
->i_d
.di_uid
= current_fsuid();
1013 ip
->i_d
.di_gid
= current_fsgid();
1014 xfs_set_projid(ip
, prid
);
1015 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
1018 * If the superblock version is up to where we support new format
1019 * inodes and this is currently an old format inode, then change
1020 * the inode version number now. This way we only do the conversion
1021 * here rather than here and in the flush/logging code.
1023 if (xfs_sb_version_hasnlink(&tp
->t_mountp
->m_sb
) &&
1024 ip
->i_d
.di_version
== 1) {
1025 ip
->i_d
.di_version
= 2;
1027 * We've already zeroed the old link count, the projid field,
1028 * and the pad field.
1033 * Project ids won't be stored on disk if we are using a version 1 inode.
1035 if ((prid
!= 0) && (ip
->i_d
.di_version
== 1))
1036 xfs_bump_ino_vers2(tp
, ip
);
1038 if (pip
&& XFS_INHERIT_GID(pip
)) {
1039 ip
->i_d
.di_gid
= pip
->i_d
.di_gid
;
1040 if ((pip
->i_d
.di_mode
& S_ISGID
) && S_ISDIR(mode
)) {
1041 ip
->i_d
.di_mode
|= S_ISGID
;
1046 * If the group ID of the new file does not match the effective group
1047 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1048 * (and only if the irix_sgid_inherit compatibility variable is set).
1050 if ((irix_sgid_inherit
) &&
1051 (ip
->i_d
.di_mode
& S_ISGID
) &&
1052 (!in_group_p((gid_t
)ip
->i_d
.di_gid
))) {
1053 ip
->i_d
.di_mode
&= ~S_ISGID
;
1056 ip
->i_d
.di_size
= 0;
1057 ip
->i_d
.di_nextents
= 0;
1058 ASSERT(ip
->i_d
.di_nblocks
== 0);
1061 ip
->i_d
.di_mtime
.t_sec
= (__int32_t
)tv
.tv_sec
;
1062 ip
->i_d
.di_mtime
.t_nsec
= (__int32_t
)tv
.tv_nsec
;
1063 ip
->i_d
.di_atime
= ip
->i_d
.di_mtime
;
1064 ip
->i_d
.di_ctime
= ip
->i_d
.di_mtime
;
1067 * di_gen will have been taken care of in xfs_iread.
1069 ip
->i_d
.di_extsize
= 0;
1070 ip
->i_d
.di_dmevmask
= 0;
1071 ip
->i_d
.di_dmstate
= 0;
1072 ip
->i_d
.di_flags
= 0;
1073 flags
= XFS_ILOG_CORE
;
1074 switch (mode
& S_IFMT
) {
1079 ip
->i_d
.di_format
= XFS_DINODE_FMT_DEV
;
1080 ip
->i_df
.if_u2
.if_rdev
= rdev
;
1081 ip
->i_df
.if_flags
= 0;
1082 flags
|= XFS_ILOG_DEV
;
1086 * we can't set up filestreams until after the VFS inode
1087 * is set up properly.
1089 if (pip
&& xfs_inode_is_filestream(pip
))
1093 if (pip
&& (pip
->i_d
.di_flags
& XFS_DIFLAG_ANY
)) {
1096 if (S_ISDIR(mode
)) {
1097 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
1098 di_flags
|= XFS_DIFLAG_RTINHERIT
;
1099 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
1100 di_flags
|= XFS_DIFLAG_EXTSZINHERIT
;
1101 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
1103 } else if (S_ISREG(mode
)) {
1104 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
1105 di_flags
|= XFS_DIFLAG_REALTIME
;
1106 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
1107 di_flags
|= XFS_DIFLAG_EXTSIZE
;
1108 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
1111 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOATIME
) &&
1112 xfs_inherit_noatime
)
1113 di_flags
|= XFS_DIFLAG_NOATIME
;
1114 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODUMP
) &&
1116 di_flags
|= XFS_DIFLAG_NODUMP
;
1117 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_SYNC
) &&
1119 di_flags
|= XFS_DIFLAG_SYNC
;
1120 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOSYMLINKS
) &&
1121 xfs_inherit_nosymlinks
)
1122 di_flags
|= XFS_DIFLAG_NOSYMLINKS
;
1123 if (pip
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
1124 di_flags
|= XFS_DIFLAG_PROJINHERIT
;
1125 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODEFRAG
) &&
1126 xfs_inherit_nodefrag
)
1127 di_flags
|= XFS_DIFLAG_NODEFRAG
;
1128 if (pip
->i_d
.di_flags
& XFS_DIFLAG_FILESTREAM
)
1129 di_flags
|= XFS_DIFLAG_FILESTREAM
;
1130 ip
->i_d
.di_flags
|= di_flags
;
1134 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
1135 ip
->i_df
.if_flags
= XFS_IFEXTENTS
;
1136 ip
->i_df
.if_bytes
= ip
->i_df
.if_real_bytes
= 0;
1137 ip
->i_df
.if_u1
.if_extents
= NULL
;
1143 * Attribute fork settings for new inode.
1145 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
1146 ip
->i_d
.di_anextents
= 0;
1149 * Log the new values stuffed into the inode.
1151 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
1152 xfs_trans_log_inode(tp
, ip
, flags
);
1154 /* now that we have an i_mode we can setup inode ops and unlock */
1155 xfs_setup_inode(ip
);
1157 /* now we have set up the vfs inode we can associate the filestream */
1159 error
= xfs_filestream_associate(pip
, ip
);
1163 xfs_iflags_set(ip
, XFS_IFILESTREAM
);
1171 * Free up the underlying blocks past new_size. The new size must be smaller
1172 * than the current size. This routine can be used both for the attribute and
1173 * data fork, and does not modify the inode size, which is left to the caller.
1175 * The transaction passed to this routine must have made a permanent log
1176 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1177 * given transaction and start new ones, so make sure everything involved in
1178 * the transaction is tidy before calling here. Some transaction will be
1179 * returned to the caller to be committed. The incoming transaction must
1180 * already include the inode, and both inode locks must be held exclusively.
1181 * The inode must also be "held" within the transaction. On return the inode
1182 * will be "held" within the returned transaction. This routine does NOT
1183 * require any disk space to be reserved for it within the transaction.
1185 * If we get an error, we must return with the inode locked and linked into the
1186 * current transaction. This keeps things simple for the higher level code,
1187 * because it always knows that the inode is locked and held in the transaction
1188 * that returns to it whether errors occur or not. We don't mark the inode
1189 * dirty on error so that transactions can be easily aborted if possible.
1192 xfs_itruncate_extents(
1193 struct xfs_trans
**tpp
,
1194 struct xfs_inode
*ip
,
1196 xfs_fsize_t new_size
)
1198 struct xfs_mount
*mp
= ip
->i_mount
;
1199 struct xfs_trans
*tp
= *tpp
;
1200 struct xfs_trans
*ntp
;
1201 xfs_bmap_free_t free_list
;
1202 xfs_fsblock_t first_block
;
1203 xfs_fileoff_t first_unmap_block
;
1204 xfs_fileoff_t last_block
;
1205 xfs_filblks_t unmap_len
;
1210 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_IOLOCK_EXCL
));
1211 ASSERT(new_size
<= XFS_ISIZE(ip
));
1212 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
1213 ASSERT(ip
->i_itemp
!= NULL
);
1214 ASSERT(ip
->i_itemp
->ili_lock_flags
== 0);
1215 ASSERT(!XFS_NOT_DQATTACHED(mp
, ip
));
1217 trace_xfs_itruncate_extents_start(ip
, new_size
);
1220 * Since it is possible for space to become allocated beyond
1221 * the end of the file (in a crash where the space is allocated
1222 * but the inode size is not yet updated), simply remove any
1223 * blocks which show up between the new EOF and the maximum
1224 * possible file size. If the first block to be removed is
1225 * beyond the maximum file size (ie it is the same as last_block),
1226 * then there is nothing to do.
1228 first_unmap_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)new_size
);
1229 last_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
));
1230 if (first_unmap_block
== last_block
)
1233 ASSERT(first_unmap_block
< last_block
);
1234 unmap_len
= last_block
- first_unmap_block
+ 1;
1236 xfs_bmap_init(&free_list
, &first_block
);
1237 error
= xfs_bunmapi(tp
, ip
,
1238 first_unmap_block
, unmap_len
,
1239 xfs_bmapi_aflag(whichfork
),
1240 XFS_ITRUNC_MAX_EXTENTS
,
1241 &first_block
, &free_list
,
1244 goto out_bmap_cancel
;
1247 * Duplicate the transaction that has the permanent
1248 * reservation and commit the old transaction.
1250 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
1252 xfs_trans_ijoin(tp
, ip
, 0);
1254 goto out_bmap_cancel
;
1258 * Mark the inode dirty so it will be logged and
1259 * moved forward in the log as part of every commit.
1261 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1264 ntp
= xfs_trans_dup(tp
);
1265 error
= xfs_trans_commit(tp
, 0);
1268 xfs_trans_ijoin(tp
, ip
, 0);
1274 * Transaction commit worked ok so we can drop the extra ticket
1275 * reference that we gained in xfs_trans_dup()
1277 xfs_log_ticket_put(tp
->t_ticket
);
1278 error
= xfs_trans_reserve(tp
, 0,
1279 XFS_ITRUNCATE_LOG_RES(mp
), 0,
1280 XFS_TRANS_PERM_LOG_RES
,
1281 XFS_ITRUNCATE_LOG_COUNT
);
1287 * Always re-log the inode so that our permanent transaction can keep
1288 * on rolling it forward in the log.
1290 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1292 trace_xfs_itruncate_extents_end(ip
, new_size
);
1299 * If the bunmapi call encounters an error, return to the caller where
1300 * the transaction can be properly aborted. We just need to make sure
1301 * we're not holding any resources that we were not when we came in.
1303 xfs_bmap_cancel(&free_list
);
1308 * This is called when the inode's link count goes to 0.
1309 * We place the on-disk inode on a list in the AGI. It
1310 * will be pulled from this list when the inode is freed.
1327 ASSERT(ip
->i_d
.di_nlink
== 0);
1328 ASSERT(ip
->i_d
.di_mode
!= 0);
1333 * Get the agi buffer first. It ensures lock ordering
1336 error
= xfs_read_agi(mp
, tp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
), &agibp
);
1339 agi
= XFS_BUF_TO_AGI(agibp
);
1342 * Get the index into the agi hash table for the
1343 * list this inode will go on.
1345 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
1347 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
1348 ASSERT(agi
->agi_unlinked
[bucket_index
]);
1349 ASSERT(be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != agino
);
1351 if (agi
->agi_unlinked
[bucket_index
] != cpu_to_be32(NULLAGINO
)) {
1353 * There is already another inode in the bucket we need
1354 * to add ourselves to. Add us at the front of the list.
1355 * Here we put the head pointer into our next pointer,
1356 * and then we fall through to point the head at us.
1358 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, 0);
1362 ASSERT(dip
->di_next_unlinked
== cpu_to_be32(NULLAGINO
));
1363 dip
->di_next_unlinked
= agi
->agi_unlinked
[bucket_index
];
1364 offset
= ip
->i_imap
.im_boffset
+
1365 offsetof(xfs_dinode_t
, di_next_unlinked
);
1366 xfs_trans_inode_buf(tp
, ibp
);
1367 xfs_trans_log_buf(tp
, ibp
, offset
,
1368 (offset
+ sizeof(xfs_agino_t
) - 1));
1369 xfs_inobp_check(mp
, ibp
);
1373 * Point the bucket head pointer at the inode being inserted.
1376 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(agino
);
1377 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
1378 (sizeof(xfs_agino_t
) * bucket_index
);
1379 xfs_trans_log_buf(tp
, agibp
, offset
,
1380 (offset
+ sizeof(xfs_agino_t
) - 1));
1385 * Pull the on-disk inode from the AGI unlinked list.
1398 xfs_agnumber_t agno
;
1400 xfs_agino_t next_agino
;
1401 xfs_buf_t
*last_ibp
;
1402 xfs_dinode_t
*last_dip
= NULL
;
1404 int offset
, last_offset
= 0;
1408 agno
= XFS_INO_TO_AGNO(mp
, ip
->i_ino
);
1411 * Get the agi buffer first. It ensures lock ordering
1414 error
= xfs_read_agi(mp
, tp
, agno
, &agibp
);
1418 agi
= XFS_BUF_TO_AGI(agibp
);
1421 * Get the index into the agi hash table for the
1422 * list this inode will go on.
1424 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
1426 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
1427 ASSERT(agi
->agi_unlinked
[bucket_index
] != cpu_to_be32(NULLAGINO
));
1428 ASSERT(agi
->agi_unlinked
[bucket_index
]);
1430 if (be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) == agino
) {
1432 * We're at the head of the list. Get the inode's
1433 * on-disk buffer to see if there is anyone after us
1434 * on the list. Only modify our next pointer if it
1435 * is not already NULLAGINO. This saves us the overhead
1436 * of dealing with the buffer when there is no need to
1439 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, 0);
1441 xfs_warn(mp
, "%s: xfs_itobp() returned error %d.",
1445 next_agino
= be32_to_cpu(dip
->di_next_unlinked
);
1446 ASSERT(next_agino
!= 0);
1447 if (next_agino
!= NULLAGINO
) {
1448 dip
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
1449 offset
= ip
->i_imap
.im_boffset
+
1450 offsetof(xfs_dinode_t
, di_next_unlinked
);
1451 xfs_trans_inode_buf(tp
, ibp
);
1452 xfs_trans_log_buf(tp
, ibp
, offset
,
1453 (offset
+ sizeof(xfs_agino_t
) - 1));
1454 xfs_inobp_check(mp
, ibp
);
1456 xfs_trans_brelse(tp
, ibp
);
1459 * Point the bucket head pointer at the next inode.
1461 ASSERT(next_agino
!= 0);
1462 ASSERT(next_agino
!= agino
);
1463 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(next_agino
);
1464 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
1465 (sizeof(xfs_agino_t
) * bucket_index
);
1466 xfs_trans_log_buf(tp
, agibp
, offset
,
1467 (offset
+ sizeof(xfs_agino_t
) - 1));
1470 * We need to search the list for the inode being freed.
1472 next_agino
= be32_to_cpu(agi
->agi_unlinked
[bucket_index
]);
1474 while (next_agino
!= agino
) {
1476 * If the last inode wasn't the one pointing to
1477 * us, then release its buffer since we're not
1478 * going to do anything with it.
1480 if (last_ibp
!= NULL
) {
1481 xfs_trans_brelse(tp
, last_ibp
);
1483 next_ino
= XFS_AGINO_TO_INO(mp
, agno
, next_agino
);
1484 error
= xfs_inotobp(mp
, tp
, next_ino
, &last_dip
,
1485 &last_ibp
, &last_offset
, 0);
1488 "%s: xfs_inotobp() returned error %d.",
1492 next_agino
= be32_to_cpu(last_dip
->di_next_unlinked
);
1493 ASSERT(next_agino
!= NULLAGINO
);
1494 ASSERT(next_agino
!= 0);
1497 * Now last_ibp points to the buffer previous to us on
1498 * the unlinked list. Pull us from the list.
1500 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, 0);
1502 xfs_warn(mp
, "%s: xfs_itobp(2) returned error %d.",
1506 next_agino
= be32_to_cpu(dip
->di_next_unlinked
);
1507 ASSERT(next_agino
!= 0);
1508 ASSERT(next_agino
!= agino
);
1509 if (next_agino
!= NULLAGINO
) {
1510 dip
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
1511 offset
= ip
->i_imap
.im_boffset
+
1512 offsetof(xfs_dinode_t
, di_next_unlinked
);
1513 xfs_trans_inode_buf(tp
, ibp
);
1514 xfs_trans_log_buf(tp
, ibp
, offset
,
1515 (offset
+ sizeof(xfs_agino_t
) - 1));
1516 xfs_inobp_check(mp
, ibp
);
1518 xfs_trans_brelse(tp
, ibp
);
1521 * Point the previous inode on the list to the next inode.
1523 last_dip
->di_next_unlinked
= cpu_to_be32(next_agino
);
1524 ASSERT(next_agino
!= 0);
1525 offset
= last_offset
+ offsetof(xfs_dinode_t
, di_next_unlinked
);
1526 xfs_trans_inode_buf(tp
, last_ibp
);
1527 xfs_trans_log_buf(tp
, last_ibp
, offset
,
1528 (offset
+ sizeof(xfs_agino_t
) - 1));
1529 xfs_inobp_check(mp
, last_ibp
);
1535 * A big issue when freeing the inode cluster is is that we _cannot_ skip any
1536 * inodes that are in memory - they all must be marked stale and attached to
1537 * the cluster buffer.
1541 xfs_inode_t
*free_ip
,
1545 xfs_mount_t
*mp
= free_ip
->i_mount
;
1546 int blks_per_cluster
;
1553 xfs_inode_log_item_t
*iip
;
1554 xfs_log_item_t
*lip
;
1555 struct xfs_perag
*pag
;
1557 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, inum
));
1558 if (mp
->m_sb
.sb_blocksize
>= XFS_INODE_CLUSTER_SIZE(mp
)) {
1559 blks_per_cluster
= 1;
1560 ninodes
= mp
->m_sb
.sb_inopblock
;
1561 nbufs
= XFS_IALLOC_BLOCKS(mp
);
1563 blks_per_cluster
= XFS_INODE_CLUSTER_SIZE(mp
) /
1564 mp
->m_sb
.sb_blocksize
;
1565 ninodes
= blks_per_cluster
* mp
->m_sb
.sb_inopblock
;
1566 nbufs
= XFS_IALLOC_BLOCKS(mp
) / blks_per_cluster
;
1569 for (j
= 0; j
< nbufs
; j
++, inum
+= ninodes
) {
1570 blkno
= XFS_AGB_TO_DADDR(mp
, XFS_INO_TO_AGNO(mp
, inum
),
1571 XFS_INO_TO_AGBNO(mp
, inum
));
1574 * We obtain and lock the backing buffer first in the process
1575 * here, as we have to ensure that any dirty inode that we
1576 * can't get the flush lock on is attached to the buffer.
1577 * If we scan the in-memory inodes first, then buffer IO can
1578 * complete before we get a lock on it, and hence we may fail
1579 * to mark all the active inodes on the buffer stale.
1581 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, blkno
,
1582 mp
->m_bsize
* blks_per_cluster
, 0);
1587 * Walk the inodes already attached to the buffer and mark them
1588 * stale. These will all have the flush locks held, so an
1589 * in-memory inode walk can't lock them. By marking them all
1590 * stale first, we will not attempt to lock them in the loop
1591 * below as the XFS_ISTALE flag will be set.
1595 if (lip
->li_type
== XFS_LI_INODE
) {
1596 iip
= (xfs_inode_log_item_t
*)lip
;
1597 ASSERT(iip
->ili_logged
== 1);
1598 lip
->li_cb
= xfs_istale_done
;
1599 xfs_trans_ail_copy_lsn(mp
->m_ail
,
1600 &iip
->ili_flush_lsn
,
1601 &iip
->ili_item
.li_lsn
);
1602 xfs_iflags_set(iip
->ili_inode
, XFS_ISTALE
);
1604 lip
= lip
->li_bio_list
;
1609 * For each inode in memory attempt to add it to the inode
1610 * buffer and set it up for being staled on buffer IO
1611 * completion. This is safe as we've locked out tail pushing
1612 * and flushing by locking the buffer.
1614 * We have already marked every inode that was part of a
1615 * transaction stale above, which means there is no point in
1616 * even trying to lock them.
1618 for (i
= 0; i
< ninodes
; i
++) {
1621 ip
= radix_tree_lookup(&pag
->pag_ici_root
,
1622 XFS_INO_TO_AGINO(mp
, (inum
+ i
)));
1624 /* Inode not in memory, nothing to do */
1631 * because this is an RCU protected lookup, we could
1632 * find a recently freed or even reallocated inode
1633 * during the lookup. We need to check under the
1634 * i_flags_lock for a valid inode here. Skip it if it
1635 * is not valid, the wrong inode or stale.
1637 spin_lock(&ip
->i_flags_lock
);
1638 if (ip
->i_ino
!= inum
+ i
||
1639 __xfs_iflags_test(ip
, XFS_ISTALE
)) {
1640 spin_unlock(&ip
->i_flags_lock
);
1644 spin_unlock(&ip
->i_flags_lock
);
1647 * Don't try to lock/unlock the current inode, but we
1648 * _cannot_ skip the other inodes that we did not find
1649 * in the list attached to the buffer and are not
1650 * already marked stale. If we can't lock it, back off
1653 if (ip
!= free_ip
&&
1654 !xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
)) {
1662 xfs_iflags_set(ip
, XFS_ISTALE
);
1665 * we don't need to attach clean inodes or those only
1666 * with unlogged changes (which we throw away, anyway).
1669 if (!iip
|| xfs_inode_clean(ip
)) {
1670 ASSERT(ip
!= free_ip
);
1672 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1676 iip
->ili_last_fields
= iip
->ili_fields
;
1677 iip
->ili_fields
= 0;
1678 iip
->ili_logged
= 1;
1679 xfs_trans_ail_copy_lsn(mp
->m_ail
, &iip
->ili_flush_lsn
,
1680 &iip
->ili_item
.li_lsn
);
1682 xfs_buf_attach_iodone(bp
, xfs_istale_done
,
1686 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1689 xfs_trans_stale_inode_buf(tp
, bp
);
1690 xfs_trans_binval(tp
, bp
);
1698 * This is called to return an inode to the inode free list.
1699 * The inode should already be truncated to 0 length and have
1700 * no pages associated with it. This routine also assumes that
1701 * the inode is already a part of the transaction.
1703 * The on-disk copy of the inode will have been added to the list
1704 * of unlinked inodes in the AGI. We need to remove the inode from
1705 * that list atomically with respect to freeing it here.
1711 xfs_bmap_free_t
*flist
)
1715 xfs_ino_t first_ino
;
1719 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
1720 ASSERT(ip
->i_d
.di_nlink
== 0);
1721 ASSERT(ip
->i_d
.di_nextents
== 0);
1722 ASSERT(ip
->i_d
.di_anextents
== 0);
1723 ASSERT(ip
->i_d
.di_size
== 0 || !S_ISREG(ip
->i_d
.di_mode
));
1724 ASSERT(ip
->i_d
.di_nblocks
== 0);
1727 * Pull the on-disk inode from the AGI unlinked list.
1729 error
= xfs_iunlink_remove(tp
, ip
);
1734 error
= xfs_difree(tp
, ip
->i_ino
, flist
, &delete, &first_ino
);
1738 ip
->i_d
.di_mode
= 0; /* mark incore inode as free */
1739 ip
->i_d
.di_flags
= 0;
1740 ip
->i_d
.di_dmevmask
= 0;
1741 ip
->i_d
.di_forkoff
= 0; /* mark the attr fork not in use */
1742 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
1743 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
1745 * Bump the generation count so no one will be confused
1746 * by reincarnations of this inode.
1750 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1752 error
= xfs_itobp(ip
->i_mount
, tp
, ip
, &dip
, &ibp
, 0);
1757 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
1758 * from picking up this inode when it is reclaimed (its incore state
1759 * initialzed but not flushed to disk yet). The in-core di_mode is
1760 * already cleared and a corresponding transaction logged.
1761 * The hack here just synchronizes the in-core to on-disk
1762 * di_mode value in advance before the actual inode sync to disk.
1763 * This is OK because the inode is already unlinked and would never
1764 * change its di_mode again for this inode generation.
1765 * This is a temporary hack that would require a proper fix
1771 error
= xfs_ifree_cluster(ip
, tp
, first_ino
);
1778 * Reallocate the space for if_broot based on the number of records
1779 * being added or deleted as indicated in rec_diff. Move the records
1780 * and pointers in if_broot to fit the new size. When shrinking this
1781 * will eliminate holes between the records and pointers created by
1782 * the caller. When growing this will create holes to be filled in
1785 * The caller must not request to add more records than would fit in
1786 * the on-disk inode root. If the if_broot is currently NULL, then
1787 * if we adding records one will be allocated. The caller must also
1788 * not request that the number of records go below zero, although
1789 * it can go to zero.
1791 * ip -- the inode whose if_broot area is changing
1792 * ext_diff -- the change in the number of records, positive or negative,
1793 * requested for the if_broot array.
1801 struct xfs_mount
*mp
= ip
->i_mount
;
1804 struct xfs_btree_block
*new_broot
;
1811 * Handle the degenerate case quietly.
1813 if (rec_diff
== 0) {
1817 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1820 * If there wasn't any memory allocated before, just
1821 * allocate it now and get out.
1823 if (ifp
->if_broot_bytes
== 0) {
1824 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff
);
1825 ifp
->if_broot
= kmem_alloc(new_size
, KM_SLEEP
| KM_NOFS
);
1826 ifp
->if_broot_bytes
= (int)new_size
;
1831 * If there is already an existing if_broot, then we need
1832 * to realloc() it and shift the pointers to their new
1833 * location. The records don't change location because
1834 * they are kept butted up against the btree block header.
1836 cur_max
= xfs_bmbt_maxrecs(mp
, ifp
->if_broot_bytes
, 0);
1837 new_max
= cur_max
+ rec_diff
;
1838 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max
);
1839 ifp
->if_broot
= kmem_realloc(ifp
->if_broot
, new_size
,
1840 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max
), /* old size */
1841 KM_SLEEP
| KM_NOFS
);
1842 op
= (char *)XFS_BMAP_BROOT_PTR_ADDR(mp
, ifp
->if_broot
, 1,
1843 ifp
->if_broot_bytes
);
1844 np
= (char *)XFS_BMAP_BROOT_PTR_ADDR(mp
, ifp
->if_broot
, 1,
1846 ifp
->if_broot_bytes
= (int)new_size
;
1847 ASSERT(ifp
->if_broot_bytes
<=
1848 XFS_IFORK_SIZE(ip
, whichfork
) + XFS_BROOT_SIZE_ADJ
);
1849 memmove(np
, op
, cur_max
* (uint
)sizeof(xfs_dfsbno_t
));
1854 * rec_diff is less than 0. In this case, we are shrinking the
1855 * if_broot buffer. It must already exist. If we go to zero
1856 * records, just get rid of the root and clear the status bit.
1858 ASSERT((ifp
->if_broot
!= NULL
) && (ifp
->if_broot_bytes
> 0));
1859 cur_max
= xfs_bmbt_maxrecs(mp
, ifp
->if_broot_bytes
, 0);
1860 new_max
= cur_max
+ rec_diff
;
1861 ASSERT(new_max
>= 0);
1863 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max
);
1867 new_broot
= kmem_alloc(new_size
, KM_SLEEP
| KM_NOFS
);
1869 * First copy over the btree block header.
1871 memcpy(new_broot
, ifp
->if_broot
, XFS_BTREE_LBLOCK_LEN
);
1874 ifp
->if_flags
&= ~XFS_IFBROOT
;
1878 * Only copy the records and pointers if there are any.
1882 * First copy the records.
1884 op
= (char *)XFS_BMBT_REC_ADDR(mp
, ifp
->if_broot
, 1);
1885 np
= (char *)XFS_BMBT_REC_ADDR(mp
, new_broot
, 1);
1886 memcpy(np
, op
, new_max
* (uint
)sizeof(xfs_bmbt_rec_t
));
1889 * Then copy the pointers.
1891 op
= (char *)XFS_BMAP_BROOT_PTR_ADDR(mp
, ifp
->if_broot
, 1,
1892 ifp
->if_broot_bytes
);
1893 np
= (char *)XFS_BMAP_BROOT_PTR_ADDR(mp
, new_broot
, 1,
1895 memcpy(np
, op
, new_max
* (uint
)sizeof(xfs_dfsbno_t
));
1897 kmem_free(ifp
->if_broot
);
1898 ifp
->if_broot
= new_broot
;
1899 ifp
->if_broot_bytes
= (int)new_size
;
1900 ASSERT(ifp
->if_broot_bytes
<=
1901 XFS_IFORK_SIZE(ip
, whichfork
) + XFS_BROOT_SIZE_ADJ
);
1907 * This is called when the amount of space needed for if_data
1908 * is increased or decreased. The change in size is indicated by
1909 * the number of bytes that need to be added or deleted in the
1910 * byte_diff parameter.
1912 * If the amount of space needed has decreased below the size of the
1913 * inline buffer, then switch to using the inline buffer. Otherwise,
1914 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
1915 * to what is needed.
1917 * ip -- the inode whose if_data area is changing
1918 * byte_diff -- the change in the number of bytes, positive or negative,
1919 * requested for the if_data array.
1931 if (byte_diff
== 0) {
1935 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1936 new_size
= (int)ifp
->if_bytes
+ byte_diff
;
1937 ASSERT(new_size
>= 0);
1939 if (new_size
== 0) {
1940 if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
1941 kmem_free(ifp
->if_u1
.if_data
);
1943 ifp
->if_u1
.if_data
= NULL
;
1945 } else if (new_size
<= sizeof(ifp
->if_u2
.if_inline_data
)) {
1947 * If the valid extents/data can fit in if_inline_ext/data,
1948 * copy them from the malloc'd vector and free it.
1950 if (ifp
->if_u1
.if_data
== NULL
) {
1951 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
1952 } else if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
1953 ASSERT(ifp
->if_real_bytes
!= 0);
1954 memcpy(ifp
->if_u2
.if_inline_data
, ifp
->if_u1
.if_data
,
1956 kmem_free(ifp
->if_u1
.if_data
);
1957 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
1962 * Stuck with malloc/realloc.
1963 * For inline data, the underlying buffer must be
1964 * a multiple of 4 bytes in size so that it can be
1965 * logged and stay on word boundaries. We enforce
1968 real_size
= roundup(new_size
, 4);
1969 if (ifp
->if_u1
.if_data
== NULL
) {
1970 ASSERT(ifp
->if_real_bytes
== 0);
1971 ifp
->if_u1
.if_data
= kmem_alloc(real_size
,
1972 KM_SLEEP
| KM_NOFS
);
1973 } else if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
1975 * Only do the realloc if the underlying size
1976 * is really changing.
1978 if (ifp
->if_real_bytes
!= real_size
) {
1979 ifp
->if_u1
.if_data
=
1980 kmem_realloc(ifp
->if_u1
.if_data
,
1983 KM_SLEEP
| KM_NOFS
);
1986 ASSERT(ifp
->if_real_bytes
== 0);
1987 ifp
->if_u1
.if_data
= kmem_alloc(real_size
,
1988 KM_SLEEP
| KM_NOFS
);
1989 memcpy(ifp
->if_u1
.if_data
, ifp
->if_u2
.if_inline_data
,
1993 ifp
->if_real_bytes
= real_size
;
1994 ifp
->if_bytes
= new_size
;
1995 ASSERT(ifp
->if_bytes
<= XFS_IFORK_SIZE(ip
, whichfork
));
2005 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2006 if (ifp
->if_broot
!= NULL
) {
2007 kmem_free(ifp
->if_broot
);
2008 ifp
->if_broot
= NULL
;
2012 * If the format is local, then we can't have an extents
2013 * array so just look for an inline data array. If we're
2014 * not local then we may or may not have an extents list,
2015 * so check and free it up if we do.
2017 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
) {
2018 if ((ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) &&
2019 (ifp
->if_u1
.if_data
!= NULL
)) {
2020 ASSERT(ifp
->if_real_bytes
!= 0);
2021 kmem_free(ifp
->if_u1
.if_data
);
2022 ifp
->if_u1
.if_data
= NULL
;
2023 ifp
->if_real_bytes
= 0;
2025 } else if ((ifp
->if_flags
& XFS_IFEXTENTS
) &&
2026 ((ifp
->if_flags
& XFS_IFEXTIREC
) ||
2027 ((ifp
->if_u1
.if_extents
!= NULL
) &&
2028 (ifp
->if_u1
.if_extents
!= ifp
->if_u2
.if_inline_ext
)))) {
2029 ASSERT(ifp
->if_real_bytes
!= 0);
2030 xfs_iext_destroy(ifp
);
2032 ASSERT(ifp
->if_u1
.if_extents
== NULL
||
2033 ifp
->if_u1
.if_extents
== ifp
->if_u2
.if_inline_ext
);
2034 ASSERT(ifp
->if_real_bytes
== 0);
2035 if (whichfork
== XFS_ATTR_FORK
) {
2036 kmem_zone_free(xfs_ifork_zone
, ip
->i_afp
);
2042 * This is called to unpin an inode. The caller must have the inode locked
2043 * in at least shared mode so that the buffer cannot be subsequently pinned
2044 * once someone is waiting for it to be unpinned.
2048 struct xfs_inode
*ip
)
2050 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2052 trace_xfs_inode_unpin_nowait(ip
, _RET_IP_
);
2054 /* Give the log a push to start the unpinning I/O */
2055 xfs_log_force_lsn(ip
->i_mount
, ip
->i_itemp
->ili_last_lsn
, 0);
2061 struct xfs_inode
*ip
)
2063 wait_queue_head_t
*wq
= bit_waitqueue(&ip
->i_flags
, __XFS_IPINNED_BIT
);
2064 DEFINE_WAIT_BIT(wait
, &ip
->i_flags
, __XFS_IPINNED_BIT
);
2069 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
2070 if (xfs_ipincount(ip
))
2072 } while (xfs_ipincount(ip
));
2073 finish_wait(wq
, &wait
.wait
);
2078 struct xfs_inode
*ip
)
2080 if (xfs_ipincount(ip
))
2081 __xfs_iunpin_wait(ip
);
2085 * xfs_iextents_copy()
2087 * This is called to copy the REAL extents (as opposed to the delayed
2088 * allocation extents) from the inode into the given buffer. It
2089 * returns the number of bytes copied into the buffer.
2091 * If there are no delayed allocation extents, then we can just
2092 * memcpy() the extents into the buffer. Otherwise, we need to
2093 * examine each extent in turn and skip those which are delayed.
2105 xfs_fsblock_t start_block
;
2107 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2108 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2109 ASSERT(ifp
->if_bytes
> 0);
2111 nrecs
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
2112 XFS_BMAP_TRACE_EXLIST(ip
, nrecs
, whichfork
);
2116 * There are some delayed allocation extents in the
2117 * inode, so copy the extents one at a time and skip
2118 * the delayed ones. There must be at least one
2119 * non-delayed extent.
2122 for (i
= 0; i
< nrecs
; i
++) {
2123 xfs_bmbt_rec_host_t
*ep
= xfs_iext_get_ext(ifp
, i
);
2124 start_block
= xfs_bmbt_get_startblock(ep
);
2125 if (isnullstartblock(start_block
)) {
2127 * It's a delayed allocation extent, so skip it.
2132 /* Translate to on disk format */
2133 put_unaligned(cpu_to_be64(ep
->l0
), &dp
->l0
);
2134 put_unaligned(cpu_to_be64(ep
->l1
), &dp
->l1
);
2138 ASSERT(copied
!= 0);
2139 xfs_validate_extents(ifp
, copied
, XFS_EXTFMT_INODE(ip
));
2141 return (copied
* (uint
)sizeof(xfs_bmbt_rec_t
));
2145 * Each of the following cases stores data into the same region
2146 * of the on-disk inode, so only one of them can be valid at
2147 * any given time. While it is possible to have conflicting formats
2148 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2149 * in EXTENTS format, this can only happen when the fork has
2150 * changed formats after being modified but before being flushed.
2151 * In these cases, the format always takes precedence, because the
2152 * format indicates the current state of the fork.
2159 xfs_inode_log_item_t
*iip
,
2166 #ifdef XFS_TRANS_DEBUG
2169 static const short brootflag
[2] =
2170 { XFS_ILOG_DBROOT
, XFS_ILOG_ABROOT
};
2171 static const short dataflag
[2] =
2172 { XFS_ILOG_DDATA
, XFS_ILOG_ADATA
};
2173 static const short extflag
[2] =
2174 { XFS_ILOG_DEXT
, XFS_ILOG_AEXT
};
2178 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2180 * This can happen if we gave up in iformat in an error path,
2181 * for the attribute fork.
2184 ASSERT(whichfork
== XFS_ATTR_FORK
);
2187 cp
= XFS_DFORK_PTR(dip
, whichfork
);
2189 switch (XFS_IFORK_FORMAT(ip
, whichfork
)) {
2190 case XFS_DINODE_FMT_LOCAL
:
2191 if ((iip
->ili_fields
& dataflag
[whichfork
]) &&
2192 (ifp
->if_bytes
> 0)) {
2193 ASSERT(ifp
->if_u1
.if_data
!= NULL
);
2194 ASSERT(ifp
->if_bytes
<= XFS_IFORK_SIZE(ip
, whichfork
));
2195 memcpy(cp
, ifp
->if_u1
.if_data
, ifp
->if_bytes
);
2199 case XFS_DINODE_FMT_EXTENTS
:
2200 ASSERT((ifp
->if_flags
& XFS_IFEXTENTS
) ||
2201 !(iip
->ili_fields
& extflag
[whichfork
]));
2202 if ((iip
->ili_fields
& extflag
[whichfork
]) &&
2203 (ifp
->if_bytes
> 0)) {
2204 ASSERT(xfs_iext_get_ext(ifp
, 0));
2205 ASSERT(XFS_IFORK_NEXTENTS(ip
, whichfork
) > 0);
2206 (void)xfs_iextents_copy(ip
, (xfs_bmbt_rec_t
*)cp
,
2211 case XFS_DINODE_FMT_BTREE
:
2212 if ((iip
->ili_fields
& brootflag
[whichfork
]) &&
2213 (ifp
->if_broot_bytes
> 0)) {
2214 ASSERT(ifp
->if_broot
!= NULL
);
2215 ASSERT(ifp
->if_broot_bytes
<=
2216 (XFS_IFORK_SIZE(ip
, whichfork
) +
2217 XFS_BROOT_SIZE_ADJ
));
2218 xfs_bmbt_to_bmdr(mp
, ifp
->if_broot
, ifp
->if_broot_bytes
,
2219 (xfs_bmdr_block_t
*)cp
,
2220 XFS_DFORK_SIZE(dip
, mp
, whichfork
));
2224 case XFS_DINODE_FMT_DEV
:
2225 if (iip
->ili_fields
& XFS_ILOG_DEV
) {
2226 ASSERT(whichfork
== XFS_DATA_FORK
);
2227 xfs_dinode_put_rdev(dip
, ip
->i_df
.if_u2
.if_rdev
);
2231 case XFS_DINODE_FMT_UUID
:
2232 if (iip
->ili_fields
& XFS_ILOG_UUID
) {
2233 ASSERT(whichfork
== XFS_DATA_FORK
);
2234 memcpy(XFS_DFORK_DPTR(dip
),
2235 &ip
->i_df
.if_u2
.if_uuid
,
2251 xfs_mount_t
*mp
= ip
->i_mount
;
2252 struct xfs_perag
*pag
;
2253 unsigned long first_index
, mask
;
2254 unsigned long inodes_per_cluster
;
2256 xfs_inode_t
**ilist
;
2263 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
2265 inodes_per_cluster
= XFS_INODE_CLUSTER_SIZE(mp
) >> mp
->m_sb
.sb_inodelog
;
2266 ilist_size
= inodes_per_cluster
* sizeof(xfs_inode_t
*);
2267 ilist
= kmem_alloc(ilist_size
, KM_MAYFAIL
|KM_NOFS
);
2271 mask
= ~(((XFS_INODE_CLUSTER_SIZE(mp
) >> mp
->m_sb
.sb_inodelog
)) - 1);
2272 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
) & mask
;
2274 /* really need a gang lookup range call here */
2275 nr_found
= radix_tree_gang_lookup(&pag
->pag_ici_root
, (void**)ilist
,
2276 first_index
, inodes_per_cluster
);
2280 for (i
= 0; i
< nr_found
; i
++) {
2286 * because this is an RCU protected lookup, we could find a
2287 * recently freed or even reallocated inode during the lookup.
2288 * We need to check under the i_flags_lock for a valid inode
2289 * here. Skip it if it is not valid or the wrong inode.
2291 spin_lock(&ip
->i_flags_lock
);
2293 (XFS_INO_TO_AGINO(mp
, iq
->i_ino
) & mask
) != first_index
) {
2294 spin_unlock(&ip
->i_flags_lock
);
2297 spin_unlock(&ip
->i_flags_lock
);
2300 * Do an un-protected check to see if the inode is dirty and
2301 * is a candidate for flushing. These checks will be repeated
2302 * later after the appropriate locks are acquired.
2304 if (xfs_inode_clean(iq
) && xfs_ipincount(iq
) == 0)
2308 * Try to get locks. If any are unavailable or it is pinned,
2309 * then this inode cannot be flushed and is skipped.
2312 if (!xfs_ilock_nowait(iq
, XFS_ILOCK_SHARED
))
2314 if (!xfs_iflock_nowait(iq
)) {
2315 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
2318 if (xfs_ipincount(iq
)) {
2320 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
2325 * arriving here means that this inode can be flushed. First
2326 * re-check that it's dirty before flushing.
2328 if (!xfs_inode_clean(iq
)) {
2330 error
= xfs_iflush_int(iq
, bp
);
2332 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
2333 goto cluster_corrupt_out
;
2339 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
2343 XFS_STATS_INC(xs_icluster_flushcnt
);
2344 XFS_STATS_ADD(xs_icluster_flushinode
, clcount
);
2355 cluster_corrupt_out
:
2357 * Corruption detected in the clustering loop. Invalidate the
2358 * inode buffer and shut down the filesystem.
2362 * Clean up the buffer. If it was delwri, just release it --
2363 * brelse can handle it with no problems. If not, shut down the
2364 * filesystem before releasing the buffer.
2366 bufwasdelwri
= (bp
->b_flags
& _XBF_DELWRI_Q
);
2370 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
2372 if (!bufwasdelwri
) {
2374 * Just like incore_relse: if we have b_iodone functions,
2375 * mark the buffer as an error and call them. Otherwise
2376 * mark it as stale and brelse.
2381 xfs_buf_ioerror(bp
, EIO
);
2382 xfs_buf_ioend(bp
, 0);
2390 * Unlocks the flush lock
2392 xfs_iflush_abort(iq
, false);
2395 return XFS_ERROR(EFSCORRUPTED
);
2399 * Flush dirty inode metadata into the backing buffer.
2401 * The caller must have the inode lock and the inode flush lock held. The
2402 * inode lock will still be held upon return to the caller, and the inode
2403 * flush lock will be released after the inode has reached the disk.
2405 * The caller must write out the buffer returned in *bpp and release it.
2409 struct xfs_inode
*ip
,
2410 struct xfs_buf
**bpp
)
2412 struct xfs_mount
*mp
= ip
->i_mount
;
2414 struct xfs_dinode
*dip
;
2417 XFS_STATS_INC(xs_iflush_count
);
2419 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2420 ASSERT(xfs_isiflocked(ip
));
2421 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
2422 ip
->i_d
.di_nextents
> XFS_IFORK_MAXEXT(ip
, XFS_DATA_FORK
));
2426 xfs_iunpin_wait(ip
);
2429 * For stale inodes we cannot rely on the backing buffer remaining
2430 * stale in cache for the remaining life of the stale inode and so
2431 * xfs_itobp() below may give us a buffer that no longer contains
2432 * inodes below. We have to check this after ensuring the inode is
2433 * unpinned so that it is safe to reclaim the stale inode after the
2436 if (xfs_iflags_test(ip
, XFS_ISTALE
)) {
2442 * This may have been unpinned because the filesystem is shutting
2443 * down forcibly. If that's the case we must not write this inode
2444 * to disk, because the log record didn't make it to disk.
2446 * We also have to remove the log item from the AIL in this case,
2447 * as we wait for an empty AIL as part of the unmount process.
2449 if (XFS_FORCED_SHUTDOWN(mp
)) {
2450 error
= XFS_ERROR(EIO
);
2455 * Get the buffer containing the on-disk inode.
2457 error
= xfs_itobp(mp
, NULL
, ip
, &dip
, &bp
, XBF_TRYLOCK
);
2464 * First flush out the inode that xfs_iflush was called with.
2466 error
= xfs_iflush_int(ip
, bp
);
2471 * If the buffer is pinned then push on the log now so we won't
2472 * get stuck waiting in the write for too long.
2474 if (xfs_buf_ispinned(bp
))
2475 xfs_log_force(mp
, 0);
2479 * see if other inodes can be gathered into this write
2481 error
= xfs_iflush_cluster(ip
, bp
);
2483 goto cluster_corrupt_out
;
2490 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
2491 cluster_corrupt_out
:
2492 error
= XFS_ERROR(EFSCORRUPTED
);
2495 * Unlocks the flush lock
2497 xfs_iflush_abort(ip
, false);
2507 xfs_inode_log_item_t
*iip
;
2510 #ifdef XFS_TRANS_DEBUG
2514 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2515 ASSERT(xfs_isiflocked(ip
));
2516 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
2517 ip
->i_d
.di_nextents
> XFS_IFORK_MAXEXT(ip
, XFS_DATA_FORK
));
2522 /* set *dip = inode's place in the buffer */
2523 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, ip
->i_imap
.im_boffset
);
2525 if (XFS_TEST_ERROR(dip
->di_magic
!= cpu_to_be16(XFS_DINODE_MAGIC
),
2526 mp
, XFS_ERRTAG_IFLUSH_1
, XFS_RANDOM_IFLUSH_1
)) {
2527 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
2528 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
2529 __func__
, ip
->i_ino
, be16_to_cpu(dip
->di_magic
), dip
);
2532 if (XFS_TEST_ERROR(ip
->i_d
.di_magic
!= XFS_DINODE_MAGIC
,
2533 mp
, XFS_ERRTAG_IFLUSH_2
, XFS_RANDOM_IFLUSH_2
)) {
2534 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
2535 "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
2536 __func__
, ip
->i_ino
, ip
, ip
->i_d
.di_magic
);
2539 if (S_ISREG(ip
->i_d
.di_mode
)) {
2541 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2542 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
),
2543 mp
, XFS_ERRTAG_IFLUSH_3
, XFS_RANDOM_IFLUSH_3
)) {
2544 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
2545 "%s: Bad regular inode %Lu, ptr 0x%p",
2546 __func__
, ip
->i_ino
, ip
);
2549 } else if (S_ISDIR(ip
->i_d
.di_mode
)) {
2551 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2552 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
) &&
2553 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_LOCAL
),
2554 mp
, XFS_ERRTAG_IFLUSH_4
, XFS_RANDOM_IFLUSH_4
)) {
2555 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
2556 "%s: Bad directory inode %Lu, ptr 0x%p",
2557 __func__
, ip
->i_ino
, ip
);
2561 if (XFS_TEST_ERROR(ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
>
2562 ip
->i_d
.di_nblocks
, mp
, XFS_ERRTAG_IFLUSH_5
,
2563 XFS_RANDOM_IFLUSH_5
)) {
2564 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
2565 "%s: detected corrupt incore inode %Lu, "
2566 "total extents = %d, nblocks = %Ld, ptr 0x%p",
2567 __func__
, ip
->i_ino
,
2568 ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
,
2569 ip
->i_d
.di_nblocks
, ip
);
2572 if (XFS_TEST_ERROR(ip
->i_d
.di_forkoff
> mp
->m_sb
.sb_inodesize
,
2573 mp
, XFS_ERRTAG_IFLUSH_6
, XFS_RANDOM_IFLUSH_6
)) {
2574 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
2575 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
2576 __func__
, ip
->i_ino
, ip
->i_d
.di_forkoff
, ip
);
2580 * bump the flush iteration count, used to detect flushes which
2581 * postdate a log record during recovery.
2584 ip
->i_d
.di_flushiter
++;
2587 * Copy the dirty parts of the inode into the on-disk
2588 * inode. We always copy out the core of the inode,
2589 * because if the inode is dirty at all the core must
2592 xfs_dinode_to_disk(dip
, &ip
->i_d
);
2594 /* Wrap, we never let the log put out DI_MAX_FLUSH */
2595 if (ip
->i_d
.di_flushiter
== DI_MAX_FLUSH
)
2596 ip
->i_d
.di_flushiter
= 0;
2599 * If this is really an old format inode and the superblock version
2600 * has not been updated to support only new format inodes, then
2601 * convert back to the old inode format. If the superblock version
2602 * has been updated, then make the conversion permanent.
2604 ASSERT(ip
->i_d
.di_version
== 1 || xfs_sb_version_hasnlink(&mp
->m_sb
));
2605 if (ip
->i_d
.di_version
== 1) {
2606 if (!xfs_sb_version_hasnlink(&mp
->m_sb
)) {
2610 ASSERT(ip
->i_d
.di_nlink
<= XFS_MAXLINK_1
);
2611 dip
->di_onlink
= cpu_to_be16(ip
->i_d
.di_nlink
);
2614 * The superblock version has already been bumped,
2615 * so just make the conversion to the new inode
2618 ip
->i_d
.di_version
= 2;
2619 dip
->di_version
= 2;
2620 ip
->i_d
.di_onlink
= 0;
2622 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
2623 memset(&(dip
->di_pad
[0]), 0,
2624 sizeof(dip
->di_pad
));
2625 ASSERT(xfs_get_projid(ip
) == 0);
2629 xfs_iflush_fork(ip
, dip
, iip
, XFS_DATA_FORK
, bp
);
2630 if (XFS_IFORK_Q(ip
))
2631 xfs_iflush_fork(ip
, dip
, iip
, XFS_ATTR_FORK
, bp
);
2632 xfs_inobp_check(mp
, bp
);
2635 * We've recorded everything logged in the inode, so we'd like to clear
2636 * the ili_fields bits so we don't log and flush things unnecessarily.
2637 * However, we can't stop logging all this information until the data
2638 * we've copied into the disk buffer is written to disk. If we did we
2639 * might overwrite the copy of the inode in the log with all the data
2640 * after re-logging only part of it, and in the face of a crash we
2641 * wouldn't have all the data we need to recover.
2643 * What we do is move the bits to the ili_last_fields field. When
2644 * logging the inode, these bits are moved back to the ili_fields field.
2645 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
2646 * know that the information those bits represent is permanently on
2647 * disk. As long as the flush completes before the inode is logged
2648 * again, then both ili_fields and ili_last_fields will be cleared.
2650 * We can play with the ili_fields bits here, because the inode lock
2651 * must be held exclusively in order to set bits there and the flush
2652 * lock protects the ili_last_fields bits. Set ili_logged so the flush
2653 * done routine can tell whether or not to look in the AIL. Also, store
2654 * the current LSN of the inode so that we can tell whether the item has
2655 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
2656 * need the AIL lock, because it is a 64 bit value that cannot be read
2659 if (iip
!= NULL
&& iip
->ili_fields
!= 0) {
2660 iip
->ili_last_fields
= iip
->ili_fields
;
2661 iip
->ili_fields
= 0;
2662 iip
->ili_logged
= 1;
2664 xfs_trans_ail_copy_lsn(mp
->m_ail
, &iip
->ili_flush_lsn
,
2665 &iip
->ili_item
.li_lsn
);
2668 * Attach the function xfs_iflush_done to the inode's
2669 * buffer. This will remove the inode from the AIL
2670 * and unlock the inode's flush lock when the inode is
2671 * completely written to disk.
2673 xfs_buf_attach_iodone(bp
, xfs_iflush_done
, &iip
->ili_item
);
2675 ASSERT(bp
->b_fspriv
!= NULL
);
2676 ASSERT(bp
->b_iodone
!= NULL
);
2679 * We're flushing an inode which is not in the AIL and has
2680 * not been logged. For this case we can immediately drop
2681 * the inode flush lock because we can avoid the whole
2682 * AIL state thing. It's OK to drop the flush lock now,
2683 * because we've already locked the buffer and to do anything
2684 * you really need both.
2687 ASSERT(iip
->ili_logged
== 0);
2688 ASSERT(iip
->ili_last_fields
== 0);
2689 ASSERT((iip
->ili_item
.li_flags
& XFS_LI_IN_AIL
) == 0);
2697 return XFS_ERROR(EFSCORRUPTED
);
2701 * Return a pointer to the extent record at file index idx.
2703 xfs_bmbt_rec_host_t
*
2705 xfs_ifork_t
*ifp
, /* inode fork pointer */
2706 xfs_extnum_t idx
) /* index of target extent */
2709 ASSERT(idx
< ifp
->if_bytes
/ sizeof(xfs_bmbt_rec_t
));
2711 if ((ifp
->if_flags
& XFS_IFEXTIREC
) && (idx
== 0)) {
2712 return ifp
->if_u1
.if_ext_irec
->er_extbuf
;
2713 } else if (ifp
->if_flags
& XFS_IFEXTIREC
) {
2714 xfs_ext_irec_t
*erp
; /* irec pointer */
2715 int erp_idx
= 0; /* irec index */
2716 xfs_extnum_t page_idx
= idx
; /* ext index in target list */
2718 erp
= xfs_iext_idx_to_irec(ifp
, &page_idx
, &erp_idx
, 0);
2719 return &erp
->er_extbuf
[page_idx
];
2720 } else if (ifp
->if_bytes
) {
2721 return &ifp
->if_u1
.if_extents
[idx
];
2728 * Insert new item(s) into the extent records for incore inode
2729 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
2733 xfs_inode_t
*ip
, /* incore inode pointer */
2734 xfs_extnum_t idx
, /* starting index of new items */
2735 xfs_extnum_t count
, /* number of inserted items */
2736 xfs_bmbt_irec_t
*new, /* items to insert */
2737 int state
) /* type of extent conversion */
2739 xfs_ifork_t
*ifp
= (state
& BMAP_ATTRFORK
) ? ip
->i_afp
: &ip
->i_df
;
2740 xfs_extnum_t i
; /* extent record index */
2742 trace_xfs_iext_insert(ip
, idx
, new, state
, _RET_IP_
);
2744 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
2745 xfs_iext_add(ifp
, idx
, count
);
2746 for (i
= idx
; i
< idx
+ count
; i
++, new++)
2747 xfs_bmbt_set_all(xfs_iext_get_ext(ifp
, i
), new);
2751 * This is called when the amount of space required for incore file
2752 * extents needs to be increased. The ext_diff parameter stores the
2753 * number of new extents being added and the idx parameter contains
2754 * the extent index where the new extents will be added. If the new
2755 * extents are being appended, then we just need to (re)allocate and
2756 * initialize the space. Otherwise, if the new extents are being
2757 * inserted into the middle of the existing entries, a bit more work
2758 * is required to make room for the new extents to be inserted. The
2759 * caller is responsible for filling in the new extent entries upon
2764 xfs_ifork_t
*ifp
, /* inode fork pointer */
2765 xfs_extnum_t idx
, /* index to begin adding exts */
2766 int ext_diff
) /* number of extents to add */
2768 int byte_diff
; /* new bytes being added */
2769 int new_size
; /* size of extents after adding */
2770 xfs_extnum_t nextents
; /* number of extents in file */
2772 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
2773 ASSERT((idx
>= 0) && (idx
<= nextents
));
2774 byte_diff
= ext_diff
* sizeof(xfs_bmbt_rec_t
);
2775 new_size
= ifp
->if_bytes
+ byte_diff
;
2777 * If the new number of extents (nextents + ext_diff)
2778 * fits inside the inode, then continue to use the inline
2781 if (nextents
+ ext_diff
<= XFS_INLINE_EXTS
) {
2782 if (idx
< nextents
) {
2783 memmove(&ifp
->if_u2
.if_inline_ext
[idx
+ ext_diff
],
2784 &ifp
->if_u2
.if_inline_ext
[idx
],
2785 (nextents
- idx
) * sizeof(xfs_bmbt_rec_t
));
2786 memset(&ifp
->if_u2
.if_inline_ext
[idx
], 0, byte_diff
);
2788 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
2789 ifp
->if_real_bytes
= 0;
2792 * Otherwise use a linear (direct) extent list.
2793 * If the extents are currently inside the inode,
2794 * xfs_iext_realloc_direct will switch us from
2795 * inline to direct extent allocation mode.
2797 else if (nextents
+ ext_diff
<= XFS_LINEAR_EXTS
) {
2798 xfs_iext_realloc_direct(ifp
, new_size
);
2799 if (idx
< nextents
) {
2800 memmove(&ifp
->if_u1
.if_extents
[idx
+ ext_diff
],
2801 &ifp
->if_u1
.if_extents
[idx
],
2802 (nextents
- idx
) * sizeof(xfs_bmbt_rec_t
));
2803 memset(&ifp
->if_u1
.if_extents
[idx
], 0, byte_diff
);
2806 /* Indirection array */
2808 xfs_ext_irec_t
*erp
;
2812 ASSERT(nextents
+ ext_diff
> XFS_LINEAR_EXTS
);
2813 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
2814 erp
= xfs_iext_idx_to_irec(ifp
, &page_idx
, &erp_idx
, 1);
2816 xfs_iext_irec_init(ifp
);
2817 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
2818 erp
= ifp
->if_u1
.if_ext_irec
;
2820 /* Extents fit in target extent page */
2821 if (erp
&& erp
->er_extcount
+ ext_diff
<= XFS_LINEAR_EXTS
) {
2822 if (page_idx
< erp
->er_extcount
) {
2823 memmove(&erp
->er_extbuf
[page_idx
+ ext_diff
],
2824 &erp
->er_extbuf
[page_idx
],
2825 (erp
->er_extcount
- page_idx
) *
2826 sizeof(xfs_bmbt_rec_t
));
2827 memset(&erp
->er_extbuf
[page_idx
], 0, byte_diff
);
2829 erp
->er_extcount
+= ext_diff
;
2830 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, ext_diff
);
2832 /* Insert a new extent page */
2834 xfs_iext_add_indirect_multi(ifp
,
2835 erp_idx
, page_idx
, ext_diff
);
2838 * If extent(s) are being appended to the last page in
2839 * the indirection array and the new extent(s) don't fit
2840 * in the page, then erp is NULL and erp_idx is set to
2841 * the next index needed in the indirection array.
2844 int count
= ext_diff
;
2847 erp
= xfs_iext_irec_new(ifp
, erp_idx
);
2848 erp
->er_extcount
= count
;
2849 count
-= MIN(count
, (int)XFS_LINEAR_EXTS
);
2856 ifp
->if_bytes
= new_size
;
2860 * This is called when incore extents are being added to the indirection
2861 * array and the new extents do not fit in the target extent list. The
2862 * erp_idx parameter contains the irec index for the target extent list
2863 * in the indirection array, and the idx parameter contains the extent
2864 * index within the list. The number of extents being added is stored
2865 * in the count parameter.
2867 * |-------| |-------|
2868 * | | | | idx - number of extents before idx
2870 * | | | | count - number of extents being inserted at idx
2871 * |-------| |-------|
2872 * | count | | nex2 | nex2 - number of extents after idx + count
2873 * |-------| |-------|
2876 xfs_iext_add_indirect_multi(
2877 xfs_ifork_t
*ifp
, /* inode fork pointer */
2878 int erp_idx
, /* target extent irec index */
2879 xfs_extnum_t idx
, /* index within target list */
2880 int count
) /* new extents being added */
2882 int byte_diff
; /* new bytes being added */
2883 xfs_ext_irec_t
*erp
; /* pointer to irec entry */
2884 xfs_extnum_t ext_diff
; /* number of extents to add */
2885 xfs_extnum_t ext_cnt
; /* new extents still needed */
2886 xfs_extnum_t nex2
; /* extents after idx + count */
2887 xfs_bmbt_rec_t
*nex2_ep
= NULL
; /* temp list for nex2 extents */
2888 int nlists
; /* number of irec's (lists) */
2890 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
2891 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
2892 nex2
= erp
->er_extcount
- idx
;
2893 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
2896 * Save second part of target extent list
2897 * (all extents past */
2899 byte_diff
= nex2
* sizeof(xfs_bmbt_rec_t
);
2900 nex2_ep
= (xfs_bmbt_rec_t
*) kmem_alloc(byte_diff
, KM_NOFS
);
2901 memmove(nex2_ep
, &erp
->er_extbuf
[idx
], byte_diff
);
2902 erp
->er_extcount
-= nex2
;
2903 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, -nex2
);
2904 memset(&erp
->er_extbuf
[idx
], 0, byte_diff
);
2908 * Add the new extents to the end of the target
2909 * list, then allocate new irec record(s) and
2910 * extent buffer(s) as needed to store the rest
2911 * of the new extents.
2914 ext_diff
= MIN(ext_cnt
, (int)XFS_LINEAR_EXTS
- erp
->er_extcount
);
2916 erp
->er_extcount
+= ext_diff
;
2917 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, ext_diff
);
2918 ext_cnt
-= ext_diff
;
2922 erp
= xfs_iext_irec_new(ifp
, erp_idx
);
2923 ext_diff
= MIN(ext_cnt
, (int)XFS_LINEAR_EXTS
);
2924 erp
->er_extcount
= ext_diff
;
2925 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, ext_diff
);
2926 ext_cnt
-= ext_diff
;
2929 /* Add nex2 extents back to indirection array */
2931 xfs_extnum_t ext_avail
;
2934 byte_diff
= nex2
* sizeof(xfs_bmbt_rec_t
);
2935 ext_avail
= XFS_LINEAR_EXTS
- erp
->er_extcount
;
2938 * If nex2 extents fit in the current page, append
2939 * nex2_ep after the new extents.
2941 if (nex2
<= ext_avail
) {
2942 i
= erp
->er_extcount
;
2945 * Otherwise, check if space is available in the
2948 else if ((erp_idx
< nlists
- 1) &&
2949 (nex2
<= (ext_avail
= XFS_LINEAR_EXTS
-
2950 ifp
->if_u1
.if_ext_irec
[erp_idx
+1].er_extcount
))) {
2953 /* Create a hole for nex2 extents */
2954 memmove(&erp
->er_extbuf
[nex2
], erp
->er_extbuf
,
2955 erp
->er_extcount
* sizeof(xfs_bmbt_rec_t
));
2958 * Final choice, create a new extent page for
2963 erp
= xfs_iext_irec_new(ifp
, erp_idx
);
2965 memmove(&erp
->er_extbuf
[i
], nex2_ep
, byte_diff
);
2967 erp
->er_extcount
+= nex2
;
2968 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, nex2
);
2973 * This is called when the amount of space required for incore file
2974 * extents needs to be decreased. The ext_diff parameter stores the
2975 * number of extents to be removed and the idx parameter contains
2976 * the extent index where the extents will be removed from.
2978 * If the amount of space needed has decreased below the linear
2979 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
2980 * extent array. Otherwise, use kmem_realloc() to adjust the
2981 * size to what is needed.
2985 xfs_inode_t
*ip
, /* incore inode pointer */
2986 xfs_extnum_t idx
, /* index to begin removing exts */
2987 int ext_diff
, /* number of extents to remove */
2988 int state
) /* type of extent conversion */
2990 xfs_ifork_t
*ifp
= (state
& BMAP_ATTRFORK
) ? ip
->i_afp
: &ip
->i_df
;
2991 xfs_extnum_t nextents
; /* number of extents in file */
2992 int new_size
; /* size of extents after removal */
2994 trace_xfs_iext_remove(ip
, idx
, state
, _RET_IP_
);
2996 ASSERT(ext_diff
> 0);
2997 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
2998 new_size
= (nextents
- ext_diff
) * sizeof(xfs_bmbt_rec_t
);
3000 if (new_size
== 0) {
3001 xfs_iext_destroy(ifp
);
3002 } else if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3003 xfs_iext_remove_indirect(ifp
, idx
, ext_diff
);
3004 } else if (ifp
->if_real_bytes
) {
3005 xfs_iext_remove_direct(ifp
, idx
, ext_diff
);
3007 xfs_iext_remove_inline(ifp
, idx
, ext_diff
);
3009 ifp
->if_bytes
= new_size
;
3013 * This removes ext_diff extents from the inline buffer, beginning
3014 * at extent index idx.
3017 xfs_iext_remove_inline(
3018 xfs_ifork_t
*ifp
, /* inode fork pointer */
3019 xfs_extnum_t idx
, /* index to begin removing exts */
3020 int ext_diff
) /* number of extents to remove */
3022 int nextents
; /* number of extents in file */
3024 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
));
3025 ASSERT(idx
< XFS_INLINE_EXTS
);
3026 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3027 ASSERT(((nextents
- ext_diff
) > 0) &&
3028 (nextents
- ext_diff
) < XFS_INLINE_EXTS
);
3030 if (idx
+ ext_diff
< nextents
) {
3031 memmove(&ifp
->if_u2
.if_inline_ext
[idx
],
3032 &ifp
->if_u2
.if_inline_ext
[idx
+ ext_diff
],
3033 (nextents
- (idx
+ ext_diff
)) *
3034 sizeof(xfs_bmbt_rec_t
));
3035 memset(&ifp
->if_u2
.if_inline_ext
[nextents
- ext_diff
],
3036 0, ext_diff
* sizeof(xfs_bmbt_rec_t
));
3038 memset(&ifp
->if_u2
.if_inline_ext
[idx
], 0,
3039 ext_diff
* sizeof(xfs_bmbt_rec_t
));
3044 * This removes ext_diff extents from a linear (direct) extent list,
3045 * beginning at extent index idx. If the extents are being removed
3046 * from the end of the list (ie. truncate) then we just need to re-
3047 * allocate the list to remove the extra space. Otherwise, if the
3048 * extents are being removed from the middle of the existing extent
3049 * entries, then we first need to move the extent records beginning
3050 * at idx + ext_diff up in the list to overwrite the records being
3051 * removed, then remove the extra space via kmem_realloc.
3054 xfs_iext_remove_direct(
3055 xfs_ifork_t
*ifp
, /* inode fork pointer */
3056 xfs_extnum_t idx
, /* index to begin removing exts */
3057 int ext_diff
) /* number of extents to remove */
3059 xfs_extnum_t nextents
; /* number of extents in file */
3060 int new_size
; /* size of extents after removal */
3062 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
));
3063 new_size
= ifp
->if_bytes
-
3064 (ext_diff
* sizeof(xfs_bmbt_rec_t
));
3065 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3067 if (new_size
== 0) {
3068 xfs_iext_destroy(ifp
);
3071 /* Move extents up in the list (if needed) */
3072 if (idx
+ ext_diff
< nextents
) {
3073 memmove(&ifp
->if_u1
.if_extents
[idx
],
3074 &ifp
->if_u1
.if_extents
[idx
+ ext_diff
],
3075 (nextents
- (idx
+ ext_diff
)) *
3076 sizeof(xfs_bmbt_rec_t
));
3078 memset(&ifp
->if_u1
.if_extents
[nextents
- ext_diff
],
3079 0, ext_diff
* sizeof(xfs_bmbt_rec_t
));
3081 * Reallocate the direct extent list. If the extents
3082 * will fit inside the inode then xfs_iext_realloc_direct
3083 * will switch from direct to inline extent allocation
3086 xfs_iext_realloc_direct(ifp
, new_size
);
3087 ifp
->if_bytes
= new_size
;
3091 * This is called when incore extents are being removed from the
3092 * indirection array and the extents being removed span multiple extent
3093 * buffers. The idx parameter contains the file extent index where we
3094 * want to begin removing extents, and the count parameter contains
3095 * how many extents need to be removed.
3097 * |-------| |-------|
3098 * | nex1 | | | nex1 - number of extents before idx
3099 * |-------| | count |
3100 * | | | | count - number of extents being removed at idx
3101 * | count | |-------|
3102 * | | | nex2 | nex2 - number of extents after idx + count
3103 * |-------| |-------|
3106 xfs_iext_remove_indirect(
3107 xfs_ifork_t
*ifp
, /* inode fork pointer */
3108 xfs_extnum_t idx
, /* index to begin removing extents */
3109 int count
) /* number of extents to remove */
3111 xfs_ext_irec_t
*erp
; /* indirection array pointer */
3112 int erp_idx
= 0; /* indirection array index */
3113 xfs_extnum_t ext_cnt
; /* extents left to remove */
3114 xfs_extnum_t ext_diff
; /* extents to remove in current list */
3115 xfs_extnum_t nex1
; /* number of extents before idx */
3116 xfs_extnum_t nex2
; /* extents after idx + count */
3117 int page_idx
= idx
; /* index in target extent list */
3119 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3120 erp
= xfs_iext_idx_to_irec(ifp
, &page_idx
, &erp_idx
, 0);
3121 ASSERT(erp
!= NULL
);
3125 nex2
= MAX((erp
->er_extcount
- (nex1
+ ext_cnt
)), 0);
3126 ext_diff
= MIN(ext_cnt
, (erp
->er_extcount
- nex1
));
3128 * Check for deletion of entire list;
3129 * xfs_iext_irec_remove() updates extent offsets.
3131 if (ext_diff
== erp
->er_extcount
) {
3132 xfs_iext_irec_remove(ifp
, erp_idx
);
3133 ext_cnt
-= ext_diff
;
3136 ASSERT(erp_idx
< ifp
->if_real_bytes
/
3138 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
3145 /* Move extents up (if needed) */
3147 memmove(&erp
->er_extbuf
[nex1
],
3148 &erp
->er_extbuf
[nex1
+ ext_diff
],
3149 nex2
* sizeof(xfs_bmbt_rec_t
));
3151 /* Zero out rest of page */
3152 memset(&erp
->er_extbuf
[nex1
+ nex2
], 0, (XFS_IEXT_BUFSZ
-
3153 ((nex1
+ nex2
) * sizeof(xfs_bmbt_rec_t
))));
3154 /* Update remaining counters */
3155 erp
->er_extcount
-= ext_diff
;
3156 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, -ext_diff
);
3157 ext_cnt
-= ext_diff
;
3162 ifp
->if_bytes
-= count
* sizeof(xfs_bmbt_rec_t
);
3163 xfs_iext_irec_compact(ifp
);
3167 * Create, destroy, or resize a linear (direct) block of extents.
3170 xfs_iext_realloc_direct(
3171 xfs_ifork_t
*ifp
, /* inode fork pointer */
3172 int new_size
) /* new size of extents */
3174 int rnew_size
; /* real new size of extents */
3176 rnew_size
= new_size
;
3178 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
) ||
3179 ((new_size
>= 0) && (new_size
<= XFS_IEXT_BUFSZ
) &&
3180 (new_size
!= ifp
->if_real_bytes
)));
3182 /* Free extent records */
3183 if (new_size
== 0) {
3184 xfs_iext_destroy(ifp
);
3186 /* Resize direct extent list and zero any new bytes */
3187 else if (ifp
->if_real_bytes
) {
3188 /* Check if extents will fit inside the inode */
3189 if (new_size
<= XFS_INLINE_EXTS
* sizeof(xfs_bmbt_rec_t
)) {
3190 xfs_iext_direct_to_inline(ifp
, new_size
/
3191 (uint
)sizeof(xfs_bmbt_rec_t
));
3192 ifp
->if_bytes
= new_size
;
3195 if (!is_power_of_2(new_size
)){
3196 rnew_size
= roundup_pow_of_two(new_size
);
3198 if (rnew_size
!= ifp
->if_real_bytes
) {
3199 ifp
->if_u1
.if_extents
=
3200 kmem_realloc(ifp
->if_u1
.if_extents
,
3202 ifp
->if_real_bytes
, KM_NOFS
);
3204 if (rnew_size
> ifp
->if_real_bytes
) {
3205 memset(&ifp
->if_u1
.if_extents
[ifp
->if_bytes
/
3206 (uint
)sizeof(xfs_bmbt_rec_t
)], 0,
3207 rnew_size
- ifp
->if_real_bytes
);
3211 * Switch from the inline extent buffer to a direct
3212 * extent list. Be sure to include the inline extent
3213 * bytes in new_size.
3216 new_size
+= ifp
->if_bytes
;
3217 if (!is_power_of_2(new_size
)) {
3218 rnew_size
= roundup_pow_of_two(new_size
);
3220 xfs_iext_inline_to_direct(ifp
, rnew_size
);
3222 ifp
->if_real_bytes
= rnew_size
;
3223 ifp
->if_bytes
= new_size
;
3227 * Switch from linear (direct) extent records to inline buffer.
3230 xfs_iext_direct_to_inline(
3231 xfs_ifork_t
*ifp
, /* inode fork pointer */
3232 xfs_extnum_t nextents
) /* number of extents in file */
3234 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
3235 ASSERT(nextents
<= XFS_INLINE_EXTS
);
3237 * The inline buffer was zeroed when we switched
3238 * from inline to direct extent allocation mode,
3239 * so we don't need to clear it here.
3241 memcpy(ifp
->if_u2
.if_inline_ext
, ifp
->if_u1
.if_extents
,
3242 nextents
* sizeof(xfs_bmbt_rec_t
));
3243 kmem_free(ifp
->if_u1
.if_extents
);
3244 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
3245 ifp
->if_real_bytes
= 0;
3249 * Switch from inline buffer to linear (direct) extent records.
3250 * new_size should already be rounded up to the next power of 2
3251 * by the caller (when appropriate), so use new_size as it is.
3252 * However, since new_size may be rounded up, we can't update
3253 * if_bytes here. It is the caller's responsibility to update
3254 * if_bytes upon return.
3257 xfs_iext_inline_to_direct(
3258 xfs_ifork_t
*ifp
, /* inode fork pointer */
3259 int new_size
) /* number of extents in file */
3261 ifp
->if_u1
.if_extents
= kmem_alloc(new_size
, KM_NOFS
);
3262 memset(ifp
->if_u1
.if_extents
, 0, new_size
);
3263 if (ifp
->if_bytes
) {
3264 memcpy(ifp
->if_u1
.if_extents
, ifp
->if_u2
.if_inline_ext
,
3266 memset(ifp
->if_u2
.if_inline_ext
, 0, XFS_INLINE_EXTS
*
3267 sizeof(xfs_bmbt_rec_t
));
3269 ifp
->if_real_bytes
= new_size
;
3273 * Resize an extent indirection array to new_size bytes.
3276 xfs_iext_realloc_indirect(
3277 xfs_ifork_t
*ifp
, /* inode fork pointer */
3278 int new_size
) /* new indirection array size */
3280 int nlists
; /* number of irec's (ex lists) */
3281 int size
; /* current indirection array size */
3283 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3284 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3285 size
= nlists
* sizeof(xfs_ext_irec_t
);
3286 ASSERT(ifp
->if_real_bytes
);
3287 ASSERT((new_size
>= 0) && (new_size
!= size
));
3288 if (new_size
== 0) {
3289 xfs_iext_destroy(ifp
);
3291 ifp
->if_u1
.if_ext_irec
= (xfs_ext_irec_t
*)
3292 kmem_realloc(ifp
->if_u1
.if_ext_irec
,
3293 new_size
, size
, KM_NOFS
);
3298 * Switch from indirection array to linear (direct) extent allocations.
3301 xfs_iext_indirect_to_direct(
3302 xfs_ifork_t
*ifp
) /* inode fork pointer */
3304 xfs_bmbt_rec_host_t
*ep
; /* extent record pointer */
3305 xfs_extnum_t nextents
; /* number of extents in file */
3306 int size
; /* size of file extents */
3308 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3309 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3310 ASSERT(nextents
<= XFS_LINEAR_EXTS
);
3311 size
= nextents
* sizeof(xfs_bmbt_rec_t
);
3313 xfs_iext_irec_compact_pages(ifp
);
3314 ASSERT(ifp
->if_real_bytes
== XFS_IEXT_BUFSZ
);
3316 ep
= ifp
->if_u1
.if_ext_irec
->er_extbuf
;
3317 kmem_free(ifp
->if_u1
.if_ext_irec
);
3318 ifp
->if_flags
&= ~XFS_IFEXTIREC
;
3319 ifp
->if_u1
.if_extents
= ep
;
3320 ifp
->if_bytes
= size
;
3321 if (nextents
< XFS_LINEAR_EXTS
) {
3322 xfs_iext_realloc_direct(ifp
, size
);
3327 * Free incore file extents.
3331 xfs_ifork_t
*ifp
) /* inode fork pointer */
3333 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3337 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3338 for (erp_idx
= nlists
- 1; erp_idx
>= 0 ; erp_idx
--) {
3339 xfs_iext_irec_remove(ifp
, erp_idx
);
3341 ifp
->if_flags
&= ~XFS_IFEXTIREC
;
3342 } else if (ifp
->if_real_bytes
) {
3343 kmem_free(ifp
->if_u1
.if_extents
);
3344 } else if (ifp
->if_bytes
) {
3345 memset(ifp
->if_u2
.if_inline_ext
, 0, XFS_INLINE_EXTS
*
3346 sizeof(xfs_bmbt_rec_t
));
3348 ifp
->if_u1
.if_extents
= NULL
;
3349 ifp
->if_real_bytes
= 0;
3354 * Return a pointer to the extent record for file system block bno.
3356 xfs_bmbt_rec_host_t
* /* pointer to found extent record */
3357 xfs_iext_bno_to_ext(
3358 xfs_ifork_t
*ifp
, /* inode fork pointer */
3359 xfs_fileoff_t bno
, /* block number to search for */
3360 xfs_extnum_t
*idxp
) /* index of target extent */
3362 xfs_bmbt_rec_host_t
*base
; /* pointer to first extent */
3363 xfs_filblks_t blockcount
= 0; /* number of blocks in extent */
3364 xfs_bmbt_rec_host_t
*ep
= NULL
; /* pointer to target extent */
3365 xfs_ext_irec_t
*erp
= NULL
; /* indirection array pointer */
3366 int high
; /* upper boundary in search */
3367 xfs_extnum_t idx
= 0; /* index of target extent */
3368 int low
; /* lower boundary in search */
3369 xfs_extnum_t nextents
; /* number of file extents */
3370 xfs_fileoff_t startoff
= 0; /* start offset of extent */
3372 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3373 if (nextents
== 0) {
3378 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3379 /* Find target extent list */
3381 erp
= xfs_iext_bno_to_irec(ifp
, bno
, &erp_idx
);
3382 base
= erp
->er_extbuf
;
3383 high
= erp
->er_extcount
- 1;
3385 base
= ifp
->if_u1
.if_extents
;
3386 high
= nextents
- 1;
3388 /* Binary search extent records */
3389 while (low
<= high
) {
3390 idx
= (low
+ high
) >> 1;
3392 startoff
= xfs_bmbt_get_startoff(ep
);
3393 blockcount
= xfs_bmbt_get_blockcount(ep
);
3394 if (bno
< startoff
) {
3396 } else if (bno
>= startoff
+ blockcount
) {
3399 /* Convert back to file-based extent index */
3400 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3401 idx
+= erp
->er_extoff
;
3407 /* Convert back to file-based extent index */
3408 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3409 idx
+= erp
->er_extoff
;
3411 if (bno
>= startoff
+ blockcount
) {
3412 if (++idx
== nextents
) {
3415 ep
= xfs_iext_get_ext(ifp
, idx
);
3423 * Return a pointer to the indirection array entry containing the
3424 * extent record for filesystem block bno. Store the index of the
3425 * target irec in *erp_idxp.
3427 xfs_ext_irec_t
* /* pointer to found extent record */
3428 xfs_iext_bno_to_irec(
3429 xfs_ifork_t
*ifp
, /* inode fork pointer */
3430 xfs_fileoff_t bno
, /* block number to search for */
3431 int *erp_idxp
) /* irec index of target ext list */
3433 xfs_ext_irec_t
*erp
= NULL
; /* indirection array pointer */
3434 xfs_ext_irec_t
*erp_next
; /* next indirection array entry */
3435 int erp_idx
; /* indirection array index */
3436 int nlists
; /* number of extent irec's (lists) */
3437 int high
; /* binary search upper limit */
3438 int low
; /* binary search lower limit */
3440 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3441 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3445 while (low
<= high
) {
3446 erp_idx
= (low
+ high
) >> 1;
3447 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
3448 erp_next
= erp_idx
< nlists
- 1 ? erp
+ 1 : NULL
;
3449 if (bno
< xfs_bmbt_get_startoff(erp
->er_extbuf
)) {
3451 } else if (erp_next
&& bno
>=
3452 xfs_bmbt_get_startoff(erp_next
->er_extbuf
)) {
3458 *erp_idxp
= erp_idx
;
3463 * Return a pointer to the indirection array entry containing the
3464 * extent record at file extent index *idxp. Store the index of the
3465 * target irec in *erp_idxp and store the page index of the target
3466 * extent record in *idxp.
3469 xfs_iext_idx_to_irec(
3470 xfs_ifork_t
*ifp
, /* inode fork pointer */
3471 xfs_extnum_t
*idxp
, /* extent index (file -> page) */
3472 int *erp_idxp
, /* pointer to target irec */
3473 int realloc
) /* new bytes were just added */
3475 xfs_ext_irec_t
*prev
; /* pointer to previous irec */
3476 xfs_ext_irec_t
*erp
= NULL
; /* pointer to current irec */
3477 int erp_idx
; /* indirection array index */
3478 int nlists
; /* number of irec's (ex lists) */
3479 int high
; /* binary search upper limit */
3480 int low
; /* binary search lower limit */
3481 xfs_extnum_t page_idx
= *idxp
; /* extent index in target list */
3483 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3484 ASSERT(page_idx
>= 0);
3485 ASSERT(page_idx
<= ifp
->if_bytes
/ sizeof(xfs_bmbt_rec_t
));
3486 ASSERT(page_idx
< ifp
->if_bytes
/ sizeof(xfs_bmbt_rec_t
) || realloc
);
3488 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3493 /* Binary search extent irec's */
3494 while (low
<= high
) {
3495 erp_idx
= (low
+ high
) >> 1;
3496 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
3497 prev
= erp_idx
> 0 ? erp
- 1 : NULL
;
3498 if (page_idx
< erp
->er_extoff
|| (page_idx
== erp
->er_extoff
&&
3499 realloc
&& prev
&& prev
->er_extcount
< XFS_LINEAR_EXTS
)) {
3501 } else if (page_idx
> erp
->er_extoff
+ erp
->er_extcount
||
3502 (page_idx
== erp
->er_extoff
+ erp
->er_extcount
&&
3505 } else if (page_idx
== erp
->er_extoff
+ erp
->er_extcount
&&
3506 erp
->er_extcount
== XFS_LINEAR_EXTS
) {
3510 erp
= erp_idx
< nlists
? erp
+ 1 : NULL
;
3513 page_idx
-= erp
->er_extoff
;
3518 *erp_idxp
= erp_idx
;
3523 * Allocate and initialize an indirection array once the space needed
3524 * for incore extents increases above XFS_IEXT_BUFSZ.
3528 xfs_ifork_t
*ifp
) /* inode fork pointer */
3530 xfs_ext_irec_t
*erp
; /* indirection array pointer */
3531 xfs_extnum_t nextents
; /* number of extents in file */
3533 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
));
3534 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3535 ASSERT(nextents
<= XFS_LINEAR_EXTS
);
3537 erp
= kmem_alloc(sizeof(xfs_ext_irec_t
), KM_NOFS
);
3539 if (nextents
== 0) {
3540 ifp
->if_u1
.if_extents
= kmem_alloc(XFS_IEXT_BUFSZ
, KM_NOFS
);
3541 } else if (!ifp
->if_real_bytes
) {
3542 xfs_iext_inline_to_direct(ifp
, XFS_IEXT_BUFSZ
);
3543 } else if (ifp
->if_real_bytes
< XFS_IEXT_BUFSZ
) {
3544 xfs_iext_realloc_direct(ifp
, XFS_IEXT_BUFSZ
);
3546 erp
->er_extbuf
= ifp
->if_u1
.if_extents
;
3547 erp
->er_extcount
= nextents
;
3550 ifp
->if_flags
|= XFS_IFEXTIREC
;
3551 ifp
->if_real_bytes
= XFS_IEXT_BUFSZ
;
3552 ifp
->if_bytes
= nextents
* sizeof(xfs_bmbt_rec_t
);
3553 ifp
->if_u1
.if_ext_irec
= erp
;
3559 * Allocate and initialize a new entry in the indirection array.
3563 xfs_ifork_t
*ifp
, /* inode fork pointer */
3564 int erp_idx
) /* index for new irec */
3566 xfs_ext_irec_t
*erp
; /* indirection array pointer */
3567 int i
; /* loop counter */
3568 int nlists
; /* number of irec's (ex lists) */
3570 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3571 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3573 /* Resize indirection array */
3574 xfs_iext_realloc_indirect(ifp
, ++nlists
*
3575 sizeof(xfs_ext_irec_t
));
3577 * Move records down in the array so the
3578 * new page can use erp_idx.
3580 erp
= ifp
->if_u1
.if_ext_irec
;
3581 for (i
= nlists
- 1; i
> erp_idx
; i
--) {
3582 memmove(&erp
[i
], &erp
[i
-1], sizeof(xfs_ext_irec_t
));
3584 ASSERT(i
== erp_idx
);
3586 /* Initialize new extent record */
3587 erp
= ifp
->if_u1
.if_ext_irec
;
3588 erp
[erp_idx
].er_extbuf
= kmem_alloc(XFS_IEXT_BUFSZ
, KM_NOFS
);
3589 ifp
->if_real_bytes
= nlists
* XFS_IEXT_BUFSZ
;
3590 memset(erp
[erp_idx
].er_extbuf
, 0, XFS_IEXT_BUFSZ
);
3591 erp
[erp_idx
].er_extcount
= 0;
3592 erp
[erp_idx
].er_extoff
= erp_idx
> 0 ?
3593 erp
[erp_idx
-1].er_extoff
+ erp
[erp_idx
-1].er_extcount
: 0;
3594 return (&erp
[erp_idx
]);
3598 * Remove a record from the indirection array.
3601 xfs_iext_irec_remove(
3602 xfs_ifork_t
*ifp
, /* inode fork pointer */
3603 int erp_idx
) /* irec index to remove */
3605 xfs_ext_irec_t
*erp
; /* indirection array pointer */
3606 int i
; /* loop counter */
3607 int nlists
; /* number of irec's (ex lists) */
3609 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3610 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3611 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
3612 if (erp
->er_extbuf
) {
3613 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1,
3615 kmem_free(erp
->er_extbuf
);
3617 /* Compact extent records */
3618 erp
= ifp
->if_u1
.if_ext_irec
;
3619 for (i
= erp_idx
; i
< nlists
- 1; i
++) {
3620 memmove(&erp
[i
], &erp
[i
+1], sizeof(xfs_ext_irec_t
));
3623 * Manually free the last extent record from the indirection
3624 * array. A call to xfs_iext_realloc_indirect() with a size
3625 * of zero would result in a call to xfs_iext_destroy() which
3626 * would in turn call this function again, creating a nasty
3630 xfs_iext_realloc_indirect(ifp
,
3631 nlists
* sizeof(xfs_ext_irec_t
));
3633 kmem_free(ifp
->if_u1
.if_ext_irec
);
3635 ifp
->if_real_bytes
= nlists
* XFS_IEXT_BUFSZ
;
3639 * This is called to clean up large amounts of unused memory allocated
3640 * by the indirection array. Before compacting anything though, verify
3641 * that the indirection array is still needed and switch back to the
3642 * linear extent list (or even the inline buffer) if possible. The
3643 * compaction policy is as follows:
3645 * Full Compaction: Extents fit into a single page (or inline buffer)
3646 * Partial Compaction: Extents occupy less than 50% of allocated space
3647 * No Compaction: Extents occupy at least 50% of allocated space
3650 xfs_iext_irec_compact(
3651 xfs_ifork_t
*ifp
) /* inode fork pointer */
3653 xfs_extnum_t nextents
; /* number of extents in file */
3654 int nlists
; /* number of irec's (ex lists) */
3656 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3657 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3658 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3660 if (nextents
== 0) {
3661 xfs_iext_destroy(ifp
);
3662 } else if (nextents
<= XFS_INLINE_EXTS
) {
3663 xfs_iext_indirect_to_direct(ifp
);
3664 xfs_iext_direct_to_inline(ifp
, nextents
);
3665 } else if (nextents
<= XFS_LINEAR_EXTS
) {
3666 xfs_iext_indirect_to_direct(ifp
);
3667 } else if (nextents
< (nlists
* XFS_LINEAR_EXTS
) >> 1) {
3668 xfs_iext_irec_compact_pages(ifp
);
3673 * Combine extents from neighboring extent pages.
3676 xfs_iext_irec_compact_pages(
3677 xfs_ifork_t
*ifp
) /* inode fork pointer */
3679 xfs_ext_irec_t
*erp
, *erp_next
;/* pointers to irec entries */
3680 int erp_idx
= 0; /* indirection array index */
3681 int nlists
; /* number of irec's (ex lists) */
3683 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3684 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3685 while (erp_idx
< nlists
- 1) {
3686 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
3688 if (erp_next
->er_extcount
<=
3689 (XFS_LINEAR_EXTS
- erp
->er_extcount
)) {
3690 memcpy(&erp
->er_extbuf
[erp
->er_extcount
],
3691 erp_next
->er_extbuf
, erp_next
->er_extcount
*
3692 sizeof(xfs_bmbt_rec_t
));
3693 erp
->er_extcount
+= erp_next
->er_extcount
;
3695 * Free page before removing extent record
3696 * so er_extoffs don't get modified in
3697 * xfs_iext_irec_remove.
3699 kmem_free(erp_next
->er_extbuf
);
3700 erp_next
->er_extbuf
= NULL
;
3701 xfs_iext_irec_remove(ifp
, erp_idx
+ 1);
3702 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3710 * This is called to update the er_extoff field in the indirection
3711 * array when extents have been added or removed from one of the
3712 * extent lists. erp_idx contains the irec index to begin updating
3713 * at and ext_diff contains the number of extents that were added
3717 xfs_iext_irec_update_extoffs(
3718 xfs_ifork_t
*ifp
, /* inode fork pointer */
3719 int erp_idx
, /* irec index to update */
3720 int ext_diff
) /* number of new extents */
3722 int i
; /* loop counter */
3723 int nlists
; /* number of irec's (ex lists */
3725 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3726 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3727 for (i
= erp_idx
; i
< nlists
; i
++) {
3728 ifp
->if_u1
.if_ext_irec
[i
].er_extoff
+= ext_diff
;