2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 #include <linux/log2.h>
22 #include "xfs_types.h"
26 #include "xfs_trans.h"
27 #include "xfs_trans_priv.h"
30 #include "xfs_mount.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_attr_sf.h"
35 #include "xfs_dinode.h"
36 #include "xfs_inode.h"
37 #include "xfs_buf_item.h"
38 #include "xfs_inode_item.h"
39 #include "xfs_btree.h"
40 #include "xfs_alloc.h"
41 #include "xfs_ialloc.h"
43 #include "xfs_error.h"
44 #include "xfs_utils.h"
45 #include "xfs_quota.h"
46 #include "xfs_filestream.h"
47 #include "xfs_vnodeops.h"
48 #include "xfs_trace.h"
50 kmem_zone_t
*xfs_ifork_zone
;
51 kmem_zone_t
*xfs_inode_zone
;
54 * Used in xfs_itruncate_extents(). This is the maximum number of extents
55 * freed from a file in a single transaction.
57 #define XFS_ITRUNC_MAX_EXTENTS 2
59 STATIC
int xfs_iflush_int(xfs_inode_t
*, xfs_buf_t
*);
60 STATIC
int xfs_iformat_local(xfs_inode_t
*, xfs_dinode_t
*, int, int);
61 STATIC
int xfs_iformat_extents(xfs_inode_t
*, xfs_dinode_t
*, int);
62 STATIC
int xfs_iformat_btree(xfs_inode_t
*, xfs_dinode_t
*, int);
66 * Make sure that the extents in the given memory buffer
76 xfs_bmbt_rec_host_t rec
;
79 for (i
= 0; i
< nrecs
; i
++) {
80 xfs_bmbt_rec_host_t
*ep
= xfs_iext_get_ext(ifp
, i
);
81 rec
.l0
= get_unaligned(&ep
->l0
);
82 rec
.l1
= get_unaligned(&ep
->l1
);
83 xfs_bmbt_get_all(&rec
, &irec
);
84 if (fmt
== XFS_EXTFMT_NOSTATE
)
85 ASSERT(irec
.br_state
== XFS_EXT_NORM
);
89 #define xfs_validate_extents(ifp, nrecs, fmt)
93 * Check that none of the inode's in the buffer have a next
94 * unlinked field of 0.
106 j
= mp
->m_inode_cluster_size
>> mp
->m_sb
.sb_inodelog
;
108 for (i
= 0; i
< j
; i
++) {
109 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
,
110 i
* mp
->m_sb
.sb_inodesize
);
111 if (!dip
->di_next_unlinked
) {
113 "Detected bogus zero next_unlinked field in incore inode buffer 0x%p.",
115 ASSERT(dip
->di_next_unlinked
);
122 * Find the buffer associated with the given inode map
123 * We do basic validation checks on the buffer once it has been
124 * retrieved from disk.
130 struct xfs_imap
*imap
,
140 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, imap
->im_blkno
,
141 (int)imap
->im_len
, buf_flags
, &bp
);
143 if (error
!= EAGAIN
) {
145 "%s: xfs_trans_read_buf() returned error %d.",
148 ASSERT(buf_flags
& XBF_TRYLOCK
);
154 * Validate the magic number and version of every inode in the buffer
155 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
158 ni
= BBTOB(imap
->im_len
) >> mp
->m_sb
.sb_inodelog
;
159 #else /* usual case */
163 for (i
= 0; i
< ni
; i
++) {
167 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
,
168 (i
<< mp
->m_sb
.sb_inodelog
));
169 di_ok
= dip
->di_magic
== cpu_to_be16(XFS_DINODE_MAGIC
) &&
170 XFS_DINODE_GOOD_VERSION(dip
->di_version
);
171 if (unlikely(XFS_TEST_ERROR(!di_ok
, mp
,
172 XFS_ERRTAG_ITOBP_INOTOBP
,
173 XFS_RANDOM_ITOBP_INOTOBP
))) {
174 if (iget_flags
& XFS_IGET_UNTRUSTED
) {
175 xfs_trans_brelse(tp
, bp
);
176 return XFS_ERROR(EINVAL
);
178 XFS_CORRUPTION_ERROR("xfs_imap_to_bp",
179 XFS_ERRLEVEL_HIGH
, mp
, dip
);
182 "bad inode magic/vsn daddr %lld #%d (magic=%x)",
183 (unsigned long long)imap
->im_blkno
, i
,
184 be16_to_cpu(dip
->di_magic
));
187 xfs_trans_brelse(tp
, bp
);
188 return XFS_ERROR(EFSCORRUPTED
);
192 xfs_inobp_check(mp
, bp
);
198 * This routine is called to map an inode number within a file
199 * system to the buffer containing the on-disk version of the
200 * inode. It returns a pointer to the buffer containing the
201 * on-disk inode in the bpp parameter, and in the dip parameter
202 * it returns a pointer to the on-disk inode within that buffer.
204 * If a non-zero error is returned, then the contents of bpp and
205 * dipp are undefined.
207 * Use xfs_imap() to determine the size and location of the
208 * buffer to read from disk.
220 struct xfs_imap imap
;
225 error
= xfs_imap(mp
, tp
, ino
, &imap
, imap_flags
);
229 error
= xfs_imap_to_bp(mp
, tp
, &imap
, &bp
, XBF_LOCK
, imap_flags
);
233 *dipp
= (xfs_dinode_t
*)xfs_buf_offset(bp
, imap
.im_boffset
);
235 *offset
= imap
.im_boffset
;
241 * This routine is called to map an inode to the buffer containing
242 * the on-disk version of the inode. It returns a pointer to the
243 * buffer containing the on-disk inode in the bpp parameter, and in
244 * the dip parameter it returns a pointer to the on-disk inode within
247 * If a non-zero error is returned, then the contents of bpp and
248 * dipp are undefined.
250 * The inode is expected to already been mapped to its buffer and read
251 * in once, thus we can use the mapping information stored in the inode
252 * rather than calling xfs_imap(). This allows us to avoid the overhead
253 * of looking at the inode btree for small block file systems
268 ASSERT(ip
->i_imap
.im_blkno
!= 0);
270 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &bp
, buf_flags
, 0);
275 ASSERT(buf_flags
& XBF_TRYLOCK
);
281 *dipp
= (xfs_dinode_t
*)xfs_buf_offset(bp
, ip
->i_imap
.im_boffset
);
287 * Move inode type and inode format specific information from the
288 * on-disk inode to the in-core inode. For fifos, devs, and sockets
289 * this means set if_rdev to the proper value. For files, directories,
290 * and symlinks this means to bring in the in-line data or extent
291 * pointers. For a file in B-tree format, only the root is immediately
292 * brought in-core. The rest will be in-lined in if_extents when it
293 * is first referenced (see xfs_iread_extents()).
300 xfs_attr_shortform_t
*atp
;
304 ip
->i_df
.if_ext_max
=
305 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
308 if (unlikely(be32_to_cpu(dip
->di_nextents
) +
309 be16_to_cpu(dip
->di_anextents
) >
310 be64_to_cpu(dip
->di_nblocks
))) {
311 xfs_warn(ip
->i_mount
,
312 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
313 (unsigned long long)ip
->i_ino
,
314 (int)(be32_to_cpu(dip
->di_nextents
) +
315 be16_to_cpu(dip
->di_anextents
)),
317 be64_to_cpu(dip
->di_nblocks
));
318 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW
,
320 return XFS_ERROR(EFSCORRUPTED
);
323 if (unlikely(dip
->di_forkoff
> ip
->i_mount
->m_sb
.sb_inodesize
)) {
324 xfs_warn(ip
->i_mount
, "corrupt dinode %Lu, forkoff = 0x%x.",
325 (unsigned long long)ip
->i_ino
,
327 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW
,
329 return XFS_ERROR(EFSCORRUPTED
);
332 if (unlikely((ip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
) &&
333 !ip
->i_mount
->m_rtdev_targp
)) {
334 xfs_warn(ip
->i_mount
,
335 "corrupt dinode %Lu, has realtime flag set.",
337 XFS_CORRUPTION_ERROR("xfs_iformat(realtime)",
338 XFS_ERRLEVEL_LOW
, ip
->i_mount
, dip
);
339 return XFS_ERROR(EFSCORRUPTED
);
342 switch (ip
->i_d
.di_mode
& S_IFMT
) {
347 if (unlikely(dip
->di_format
!= XFS_DINODE_FMT_DEV
)) {
348 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW
,
350 return XFS_ERROR(EFSCORRUPTED
);
354 ip
->i_df
.if_u2
.if_rdev
= xfs_dinode_get_rdev(dip
);
360 switch (dip
->di_format
) {
361 case XFS_DINODE_FMT_LOCAL
:
363 * no local regular files yet
365 if (unlikely(S_ISREG(be16_to_cpu(dip
->di_mode
)))) {
366 xfs_warn(ip
->i_mount
,
367 "corrupt inode %Lu (local format for regular file).",
368 (unsigned long long) ip
->i_ino
);
369 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
372 return XFS_ERROR(EFSCORRUPTED
);
375 di_size
= be64_to_cpu(dip
->di_size
);
376 if (unlikely(di_size
> XFS_DFORK_DSIZE(dip
, ip
->i_mount
))) {
377 xfs_warn(ip
->i_mount
,
378 "corrupt inode %Lu (bad size %Ld for local inode).",
379 (unsigned long long) ip
->i_ino
,
380 (long long) di_size
);
381 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
384 return XFS_ERROR(EFSCORRUPTED
);
388 error
= xfs_iformat_local(ip
, dip
, XFS_DATA_FORK
, size
);
390 case XFS_DINODE_FMT_EXTENTS
:
391 error
= xfs_iformat_extents(ip
, dip
, XFS_DATA_FORK
);
393 case XFS_DINODE_FMT_BTREE
:
394 error
= xfs_iformat_btree(ip
, dip
, XFS_DATA_FORK
);
397 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW
,
399 return XFS_ERROR(EFSCORRUPTED
);
404 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW
, ip
->i_mount
);
405 return XFS_ERROR(EFSCORRUPTED
);
410 if (!XFS_DFORK_Q(dip
))
412 ASSERT(ip
->i_afp
== NULL
);
413 ip
->i_afp
= kmem_zone_zalloc(xfs_ifork_zone
, KM_SLEEP
| KM_NOFS
);
414 ip
->i_afp
->if_ext_max
=
415 XFS_IFORK_ASIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
416 switch (dip
->di_aformat
) {
417 case XFS_DINODE_FMT_LOCAL
:
418 atp
= (xfs_attr_shortform_t
*)XFS_DFORK_APTR(dip
);
419 size
= be16_to_cpu(atp
->hdr
.totsize
);
421 if (unlikely(size
< sizeof(struct xfs_attr_sf_hdr
))) {
422 xfs_warn(ip
->i_mount
,
423 "corrupt inode %Lu (bad attr fork size %Ld).",
424 (unsigned long long) ip
->i_ino
,
426 XFS_CORRUPTION_ERROR("xfs_iformat(8)",
429 return XFS_ERROR(EFSCORRUPTED
);
432 error
= xfs_iformat_local(ip
, dip
, XFS_ATTR_FORK
, size
);
434 case XFS_DINODE_FMT_EXTENTS
:
435 error
= xfs_iformat_extents(ip
, dip
, XFS_ATTR_FORK
);
437 case XFS_DINODE_FMT_BTREE
:
438 error
= xfs_iformat_btree(ip
, dip
, XFS_ATTR_FORK
);
441 error
= XFS_ERROR(EFSCORRUPTED
);
445 kmem_zone_free(xfs_ifork_zone
, ip
->i_afp
);
447 xfs_idestroy_fork(ip
, XFS_DATA_FORK
);
453 * The file is in-lined in the on-disk inode.
454 * If it fits into if_inline_data, then copy
455 * it there, otherwise allocate a buffer for it
456 * and copy the data there. Either way, set
457 * if_data to point at the data.
458 * If we allocate a buffer for the data, make
459 * sure that its size is a multiple of 4 and
460 * record the real size in i_real_bytes.
473 * If the size is unreasonable, then something
474 * is wrong and we just bail out rather than crash in
475 * kmem_alloc() or memcpy() below.
477 if (unlikely(size
> XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
))) {
478 xfs_warn(ip
->i_mount
,
479 "corrupt inode %Lu (bad size %d for local fork, size = %d).",
480 (unsigned long long) ip
->i_ino
, size
,
481 XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
));
482 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW
,
484 return XFS_ERROR(EFSCORRUPTED
);
486 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
489 ifp
->if_u1
.if_data
= NULL
;
490 else if (size
<= sizeof(ifp
->if_u2
.if_inline_data
))
491 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
493 real_size
= roundup(size
, 4);
494 ifp
->if_u1
.if_data
= kmem_alloc(real_size
, KM_SLEEP
| KM_NOFS
);
496 ifp
->if_bytes
= size
;
497 ifp
->if_real_bytes
= real_size
;
499 memcpy(ifp
->if_u1
.if_data
, XFS_DFORK_PTR(dip
, whichfork
), size
);
500 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
501 ifp
->if_flags
|= XFS_IFINLINE
;
506 * The file consists of a set of extents all
507 * of which fit into the on-disk inode.
508 * If there are few enough extents to fit into
509 * the if_inline_ext, then copy them there.
510 * Otherwise allocate a buffer for them and copy
511 * them into it. Either way, set if_extents
512 * to point at the extents.
526 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
527 nex
= XFS_DFORK_NEXTENTS(dip
, whichfork
);
528 size
= nex
* (uint
)sizeof(xfs_bmbt_rec_t
);
531 * If the number of extents is unreasonable, then something
532 * is wrong and we just bail out rather than crash in
533 * kmem_alloc() or memcpy() below.
535 if (unlikely(size
< 0 || size
> XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
))) {
536 xfs_warn(ip
->i_mount
, "corrupt inode %Lu ((a)extents = %d).",
537 (unsigned long long) ip
->i_ino
, nex
);
538 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW
,
540 return XFS_ERROR(EFSCORRUPTED
);
543 ifp
->if_real_bytes
= 0;
545 ifp
->if_u1
.if_extents
= NULL
;
546 else if (nex
<= XFS_INLINE_EXTS
)
547 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
549 xfs_iext_add(ifp
, 0, nex
);
551 ifp
->if_bytes
= size
;
553 dp
= (xfs_bmbt_rec_t
*) XFS_DFORK_PTR(dip
, whichfork
);
554 xfs_validate_extents(ifp
, nex
, XFS_EXTFMT_INODE(ip
));
555 for (i
= 0; i
< nex
; i
++, dp
++) {
556 xfs_bmbt_rec_host_t
*ep
= xfs_iext_get_ext(ifp
, i
);
557 ep
->l0
= get_unaligned_be64(&dp
->l0
);
558 ep
->l1
= get_unaligned_be64(&dp
->l1
);
560 XFS_BMAP_TRACE_EXLIST(ip
, nex
, whichfork
);
561 if (whichfork
!= XFS_DATA_FORK
||
562 XFS_EXTFMT_INODE(ip
) == XFS_EXTFMT_NOSTATE
)
563 if (unlikely(xfs_check_nostate_extents(
565 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
568 return XFS_ERROR(EFSCORRUPTED
);
571 ifp
->if_flags
|= XFS_IFEXTENTS
;
576 * The file has too many extents to fit into
577 * the inode, so they are in B-tree format.
578 * Allocate a buffer for the root of the B-tree
579 * and copy the root into it. The i_extents
580 * field will remain NULL until all of the
581 * extents are read in (when they are needed).
589 xfs_bmdr_block_t
*dfp
;
595 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
596 dfp
= (xfs_bmdr_block_t
*)XFS_DFORK_PTR(dip
, whichfork
);
597 size
= XFS_BMAP_BROOT_SPACE(dfp
);
598 nrecs
= be16_to_cpu(dfp
->bb_numrecs
);
601 * blow out if -- fork has less extents than can fit in
602 * fork (fork shouldn't be a btree format), root btree
603 * block has more records than can fit into the fork,
604 * or the number of extents is greater than the number of
607 if (unlikely(XFS_IFORK_NEXTENTS(ip
, whichfork
) <= ifp
->if_ext_max
608 || XFS_BMDR_SPACE_CALC(nrecs
) >
609 XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
)
610 || XFS_IFORK_NEXTENTS(ip
, whichfork
) > ip
->i_d
.di_nblocks
)) {
611 xfs_warn(ip
->i_mount
, "corrupt inode %Lu (btree).",
612 (unsigned long long) ip
->i_ino
);
613 XFS_CORRUPTION_ERROR("xfs_iformat_btree", XFS_ERRLEVEL_LOW
,
615 return XFS_ERROR(EFSCORRUPTED
);
618 ifp
->if_broot_bytes
= size
;
619 ifp
->if_broot
= kmem_alloc(size
, KM_SLEEP
| KM_NOFS
);
620 ASSERT(ifp
->if_broot
!= NULL
);
622 * Copy and convert from the on-disk structure
623 * to the in-memory structure.
625 xfs_bmdr_to_bmbt(ip
->i_mount
, dfp
,
626 XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
),
627 ifp
->if_broot
, size
);
628 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
629 ifp
->if_flags
|= XFS_IFBROOT
;
635 xfs_dinode_from_disk(
639 to
->di_magic
= be16_to_cpu(from
->di_magic
);
640 to
->di_mode
= be16_to_cpu(from
->di_mode
);
641 to
->di_version
= from
->di_version
;
642 to
->di_format
= from
->di_format
;
643 to
->di_onlink
= be16_to_cpu(from
->di_onlink
);
644 to
->di_uid
= be32_to_cpu(from
->di_uid
);
645 to
->di_gid
= be32_to_cpu(from
->di_gid
);
646 to
->di_nlink
= be32_to_cpu(from
->di_nlink
);
647 to
->di_projid_lo
= be16_to_cpu(from
->di_projid_lo
);
648 to
->di_projid_hi
= be16_to_cpu(from
->di_projid_hi
);
649 memcpy(to
->di_pad
, from
->di_pad
, sizeof(to
->di_pad
));
650 to
->di_flushiter
= be16_to_cpu(from
->di_flushiter
);
651 to
->di_atime
.t_sec
= be32_to_cpu(from
->di_atime
.t_sec
);
652 to
->di_atime
.t_nsec
= be32_to_cpu(from
->di_atime
.t_nsec
);
653 to
->di_mtime
.t_sec
= be32_to_cpu(from
->di_mtime
.t_sec
);
654 to
->di_mtime
.t_nsec
= be32_to_cpu(from
->di_mtime
.t_nsec
);
655 to
->di_ctime
.t_sec
= be32_to_cpu(from
->di_ctime
.t_sec
);
656 to
->di_ctime
.t_nsec
= be32_to_cpu(from
->di_ctime
.t_nsec
);
657 to
->di_size
= be64_to_cpu(from
->di_size
);
658 to
->di_nblocks
= be64_to_cpu(from
->di_nblocks
);
659 to
->di_extsize
= be32_to_cpu(from
->di_extsize
);
660 to
->di_nextents
= be32_to_cpu(from
->di_nextents
);
661 to
->di_anextents
= be16_to_cpu(from
->di_anextents
);
662 to
->di_forkoff
= from
->di_forkoff
;
663 to
->di_aformat
= from
->di_aformat
;
664 to
->di_dmevmask
= be32_to_cpu(from
->di_dmevmask
);
665 to
->di_dmstate
= be16_to_cpu(from
->di_dmstate
);
666 to
->di_flags
= be16_to_cpu(from
->di_flags
);
667 to
->di_gen
= be32_to_cpu(from
->di_gen
);
673 xfs_icdinode_t
*from
)
675 to
->di_magic
= cpu_to_be16(from
->di_magic
);
676 to
->di_mode
= cpu_to_be16(from
->di_mode
);
677 to
->di_version
= from
->di_version
;
678 to
->di_format
= from
->di_format
;
679 to
->di_onlink
= cpu_to_be16(from
->di_onlink
);
680 to
->di_uid
= cpu_to_be32(from
->di_uid
);
681 to
->di_gid
= cpu_to_be32(from
->di_gid
);
682 to
->di_nlink
= cpu_to_be32(from
->di_nlink
);
683 to
->di_projid_lo
= cpu_to_be16(from
->di_projid_lo
);
684 to
->di_projid_hi
= cpu_to_be16(from
->di_projid_hi
);
685 memcpy(to
->di_pad
, from
->di_pad
, sizeof(to
->di_pad
));
686 to
->di_flushiter
= cpu_to_be16(from
->di_flushiter
);
687 to
->di_atime
.t_sec
= cpu_to_be32(from
->di_atime
.t_sec
);
688 to
->di_atime
.t_nsec
= cpu_to_be32(from
->di_atime
.t_nsec
);
689 to
->di_mtime
.t_sec
= cpu_to_be32(from
->di_mtime
.t_sec
);
690 to
->di_mtime
.t_nsec
= cpu_to_be32(from
->di_mtime
.t_nsec
);
691 to
->di_ctime
.t_sec
= cpu_to_be32(from
->di_ctime
.t_sec
);
692 to
->di_ctime
.t_nsec
= cpu_to_be32(from
->di_ctime
.t_nsec
);
693 to
->di_size
= cpu_to_be64(from
->di_size
);
694 to
->di_nblocks
= cpu_to_be64(from
->di_nblocks
);
695 to
->di_extsize
= cpu_to_be32(from
->di_extsize
);
696 to
->di_nextents
= cpu_to_be32(from
->di_nextents
);
697 to
->di_anextents
= cpu_to_be16(from
->di_anextents
);
698 to
->di_forkoff
= from
->di_forkoff
;
699 to
->di_aformat
= from
->di_aformat
;
700 to
->di_dmevmask
= cpu_to_be32(from
->di_dmevmask
);
701 to
->di_dmstate
= cpu_to_be16(from
->di_dmstate
);
702 to
->di_flags
= cpu_to_be16(from
->di_flags
);
703 to
->di_gen
= cpu_to_be32(from
->di_gen
);
712 if (di_flags
& XFS_DIFLAG_ANY
) {
713 if (di_flags
& XFS_DIFLAG_REALTIME
)
714 flags
|= XFS_XFLAG_REALTIME
;
715 if (di_flags
& XFS_DIFLAG_PREALLOC
)
716 flags
|= XFS_XFLAG_PREALLOC
;
717 if (di_flags
& XFS_DIFLAG_IMMUTABLE
)
718 flags
|= XFS_XFLAG_IMMUTABLE
;
719 if (di_flags
& XFS_DIFLAG_APPEND
)
720 flags
|= XFS_XFLAG_APPEND
;
721 if (di_flags
& XFS_DIFLAG_SYNC
)
722 flags
|= XFS_XFLAG_SYNC
;
723 if (di_flags
& XFS_DIFLAG_NOATIME
)
724 flags
|= XFS_XFLAG_NOATIME
;
725 if (di_flags
& XFS_DIFLAG_NODUMP
)
726 flags
|= XFS_XFLAG_NODUMP
;
727 if (di_flags
& XFS_DIFLAG_RTINHERIT
)
728 flags
|= XFS_XFLAG_RTINHERIT
;
729 if (di_flags
& XFS_DIFLAG_PROJINHERIT
)
730 flags
|= XFS_XFLAG_PROJINHERIT
;
731 if (di_flags
& XFS_DIFLAG_NOSYMLINKS
)
732 flags
|= XFS_XFLAG_NOSYMLINKS
;
733 if (di_flags
& XFS_DIFLAG_EXTSIZE
)
734 flags
|= XFS_XFLAG_EXTSIZE
;
735 if (di_flags
& XFS_DIFLAG_EXTSZINHERIT
)
736 flags
|= XFS_XFLAG_EXTSZINHERIT
;
737 if (di_flags
& XFS_DIFLAG_NODEFRAG
)
738 flags
|= XFS_XFLAG_NODEFRAG
;
739 if (di_flags
& XFS_DIFLAG_FILESTREAM
)
740 flags
|= XFS_XFLAG_FILESTREAM
;
750 xfs_icdinode_t
*dic
= &ip
->i_d
;
752 return _xfs_dic2xflags(dic
->di_flags
) |
753 (XFS_IFORK_Q(ip
) ? XFS_XFLAG_HASATTR
: 0);
760 return _xfs_dic2xflags(be16_to_cpu(dip
->di_flags
)) |
761 (XFS_DFORK_Q(dip
) ? XFS_XFLAG_HASATTR
: 0);
765 * Read the disk inode attributes into the in-core inode structure.
779 * Fill in the location information in the in-core inode.
781 error
= xfs_imap(mp
, tp
, ip
->i_ino
, &ip
->i_imap
, iget_flags
);
786 * Get pointers to the on-disk inode and the buffer containing it.
788 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &bp
,
789 XBF_LOCK
, iget_flags
);
792 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, ip
->i_imap
.im_boffset
);
795 * If we got something that isn't an inode it means someone
796 * (nfs or dmi) has a stale handle.
798 if (dip
->di_magic
!= cpu_to_be16(XFS_DINODE_MAGIC
)) {
801 "%s: dip->di_magic (0x%x) != XFS_DINODE_MAGIC (0x%x)",
802 __func__
, be16_to_cpu(dip
->di_magic
), XFS_DINODE_MAGIC
);
804 error
= XFS_ERROR(EINVAL
);
809 * If the on-disk inode is already linked to a directory
810 * entry, copy all of the inode into the in-core inode.
811 * xfs_iformat() handles copying in the inode format
812 * specific information.
813 * Otherwise, just get the truly permanent information.
816 xfs_dinode_from_disk(&ip
->i_d
, dip
);
817 error
= xfs_iformat(ip
, dip
);
820 xfs_alert(mp
, "%s: xfs_iformat() returned error %d",
826 ip
->i_d
.di_magic
= be16_to_cpu(dip
->di_magic
);
827 ip
->i_d
.di_version
= dip
->di_version
;
828 ip
->i_d
.di_gen
= be32_to_cpu(dip
->di_gen
);
829 ip
->i_d
.di_flushiter
= be16_to_cpu(dip
->di_flushiter
);
831 * Make sure to pull in the mode here as well in
832 * case the inode is released without being used.
833 * This ensures that xfs_inactive() will see that
834 * the inode is already free and not try to mess
835 * with the uninitialized part of it.
839 * Initialize the per-fork minima and maxima for a new
840 * inode here. xfs_iformat will do it for old inodes.
842 ip
->i_df
.if_ext_max
=
843 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
847 * The inode format changed when we moved the link count and
848 * made it 32 bits long. If this is an old format inode,
849 * convert it in memory to look like a new one. If it gets
850 * flushed to disk we will convert back before flushing or
851 * logging it. We zero out the new projid field and the old link
852 * count field. We'll handle clearing the pad field (the remains
853 * of the old uuid field) when we actually convert the inode to
854 * the new format. We don't change the version number so that we
855 * can distinguish this from a real new format inode.
857 if (ip
->i_d
.di_version
== 1) {
858 ip
->i_d
.di_nlink
= ip
->i_d
.di_onlink
;
859 ip
->i_d
.di_onlink
= 0;
860 xfs_set_projid(ip
, 0);
863 ip
->i_delayed_blks
= 0;
864 ip
->i_size
= ip
->i_d
.di_size
;
867 * Mark the buffer containing the inode as something to keep
868 * around for a while. This helps to keep recently accessed
869 * meta-data in-core longer.
871 xfs_buf_set_ref(bp
, XFS_INO_REF
);
874 * Use xfs_trans_brelse() to release the buffer containing the
875 * on-disk inode, because it was acquired with xfs_trans_read_buf()
876 * in xfs_itobp() above. If tp is NULL, this is just a normal
877 * brelse(). If we're within a transaction, then xfs_trans_brelse()
878 * will only release the buffer if it is not dirty within the
879 * transaction. It will be OK to release the buffer in this case,
880 * because inodes on disk are never destroyed and we will be
881 * locking the new in-core inode before putting it in the hash
882 * table where other processes can find it. Thus we don't have
883 * to worry about the inode being changed just because we released
887 xfs_trans_brelse(tp
, bp
);
892 * Read in extents from a btree-format inode.
893 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
903 xfs_extnum_t nextents
;
905 if (unlikely(XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
)) {
906 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW
,
908 return XFS_ERROR(EFSCORRUPTED
);
910 nextents
= XFS_IFORK_NEXTENTS(ip
, whichfork
);
911 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
914 * We know that the size is valid (it's checked in iformat_btree)
916 ifp
->if_bytes
= ifp
->if_real_bytes
= 0;
917 ifp
->if_flags
|= XFS_IFEXTENTS
;
918 xfs_iext_add(ifp
, 0, nextents
);
919 error
= xfs_bmap_read_extents(tp
, ip
, whichfork
);
921 xfs_iext_destroy(ifp
);
922 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
925 xfs_validate_extents(ifp
, nextents
, XFS_EXTFMT_INODE(ip
));
930 * Allocate an inode on disk and return a copy of its in-core version.
931 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
932 * appropriately within the inode. The uid and gid for the inode are
933 * set according to the contents of the given cred structure.
935 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
936 * has a free inode available, call xfs_iget()
937 * to obtain the in-core version of the allocated inode. Finally,
938 * fill in the inode and log its initial contents. In this case,
939 * ialloc_context would be set to NULL and call_again set to false.
941 * If xfs_dialloc() does not have an available inode,
942 * it will replenish its supply by doing an allocation. Since we can
943 * only do one allocation within a transaction without deadlocks, we
944 * must commit the current transaction before returning the inode itself.
945 * In this case, therefore, we will set call_again to true and return.
946 * The caller should then commit the current transaction, start a new
947 * transaction, and call xfs_ialloc() again to actually get the inode.
949 * To ensure that some other process does not grab the inode that
950 * was allocated during the first call to xfs_ialloc(), this routine
951 * also returns the [locked] bp pointing to the head of the freelist
952 * as ialloc_context. The caller should hold this buffer across
953 * the commit and pass it back into this routine on the second call.
955 * If we are allocating quota inodes, we do not have a parent inode
956 * to attach to or associate with (i.e. pip == NULL) because they
957 * are not linked into the directory structure - they are attached
958 * directly to the superblock - and so have no parent.
969 xfs_buf_t
**ialloc_context
,
970 boolean_t
*call_again
,
981 * Call the space management code to pick
982 * the on-disk inode to be allocated.
984 error
= xfs_dialloc(tp
, pip
? pip
->i_ino
: 0, mode
, okalloc
,
985 ialloc_context
, call_again
, &ino
);
988 if (*call_again
|| ino
== NULLFSINO
) {
992 ASSERT(*ialloc_context
== NULL
);
995 * Get the in-core inode with the lock held exclusively.
996 * This is because we're setting fields here we need
997 * to prevent others from looking at until we're done.
999 error
= xfs_iget(tp
->t_mountp
, tp
, ino
, XFS_IGET_CREATE
,
1000 XFS_ILOCK_EXCL
, &ip
);
1005 ip
->i_d
.di_mode
= mode
;
1006 ip
->i_d
.di_onlink
= 0;
1007 ip
->i_d
.di_nlink
= nlink
;
1008 ASSERT(ip
->i_d
.di_nlink
== nlink
);
1009 ip
->i_d
.di_uid
= current_fsuid();
1010 ip
->i_d
.di_gid
= current_fsgid();
1011 xfs_set_projid(ip
, prid
);
1012 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
1015 * If the superblock version is up to where we support new format
1016 * inodes and this is currently an old format inode, then change
1017 * the inode version number now. This way we only do the conversion
1018 * here rather than here and in the flush/logging code.
1020 if (xfs_sb_version_hasnlink(&tp
->t_mountp
->m_sb
) &&
1021 ip
->i_d
.di_version
== 1) {
1022 ip
->i_d
.di_version
= 2;
1024 * We've already zeroed the old link count, the projid field,
1025 * and the pad field.
1030 * Project ids won't be stored on disk if we are using a version 1 inode.
1032 if ((prid
!= 0) && (ip
->i_d
.di_version
== 1))
1033 xfs_bump_ino_vers2(tp
, ip
);
1035 if (pip
&& XFS_INHERIT_GID(pip
)) {
1036 ip
->i_d
.di_gid
= pip
->i_d
.di_gid
;
1037 if ((pip
->i_d
.di_mode
& S_ISGID
) && S_ISDIR(mode
)) {
1038 ip
->i_d
.di_mode
|= S_ISGID
;
1043 * If the group ID of the new file does not match the effective group
1044 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1045 * (and only if the irix_sgid_inherit compatibility variable is set).
1047 if ((irix_sgid_inherit
) &&
1048 (ip
->i_d
.di_mode
& S_ISGID
) &&
1049 (!in_group_p((gid_t
)ip
->i_d
.di_gid
))) {
1050 ip
->i_d
.di_mode
&= ~S_ISGID
;
1053 ip
->i_d
.di_size
= 0;
1055 ip
->i_d
.di_nextents
= 0;
1056 ASSERT(ip
->i_d
.di_nblocks
== 0);
1059 ip
->i_d
.di_mtime
.t_sec
= (__int32_t
)tv
.tv_sec
;
1060 ip
->i_d
.di_mtime
.t_nsec
= (__int32_t
)tv
.tv_nsec
;
1061 ip
->i_d
.di_atime
= ip
->i_d
.di_mtime
;
1062 ip
->i_d
.di_ctime
= ip
->i_d
.di_mtime
;
1065 * di_gen will have been taken care of in xfs_iread.
1067 ip
->i_d
.di_extsize
= 0;
1068 ip
->i_d
.di_dmevmask
= 0;
1069 ip
->i_d
.di_dmstate
= 0;
1070 ip
->i_d
.di_flags
= 0;
1071 flags
= XFS_ILOG_CORE
;
1072 switch (mode
& S_IFMT
) {
1077 ip
->i_d
.di_format
= XFS_DINODE_FMT_DEV
;
1078 ip
->i_df
.if_u2
.if_rdev
= rdev
;
1079 ip
->i_df
.if_flags
= 0;
1080 flags
|= XFS_ILOG_DEV
;
1084 * we can't set up filestreams until after the VFS inode
1085 * is set up properly.
1087 if (pip
&& xfs_inode_is_filestream(pip
))
1091 if (pip
&& (pip
->i_d
.di_flags
& XFS_DIFLAG_ANY
)) {
1094 if (S_ISDIR(mode
)) {
1095 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
1096 di_flags
|= XFS_DIFLAG_RTINHERIT
;
1097 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
1098 di_flags
|= XFS_DIFLAG_EXTSZINHERIT
;
1099 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
1101 } else if (S_ISREG(mode
)) {
1102 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
1103 di_flags
|= XFS_DIFLAG_REALTIME
;
1104 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
1105 di_flags
|= XFS_DIFLAG_EXTSIZE
;
1106 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
1109 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOATIME
) &&
1110 xfs_inherit_noatime
)
1111 di_flags
|= XFS_DIFLAG_NOATIME
;
1112 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODUMP
) &&
1114 di_flags
|= XFS_DIFLAG_NODUMP
;
1115 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_SYNC
) &&
1117 di_flags
|= XFS_DIFLAG_SYNC
;
1118 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOSYMLINKS
) &&
1119 xfs_inherit_nosymlinks
)
1120 di_flags
|= XFS_DIFLAG_NOSYMLINKS
;
1121 if (pip
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
1122 di_flags
|= XFS_DIFLAG_PROJINHERIT
;
1123 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODEFRAG
) &&
1124 xfs_inherit_nodefrag
)
1125 di_flags
|= XFS_DIFLAG_NODEFRAG
;
1126 if (pip
->i_d
.di_flags
& XFS_DIFLAG_FILESTREAM
)
1127 di_flags
|= XFS_DIFLAG_FILESTREAM
;
1128 ip
->i_d
.di_flags
|= di_flags
;
1132 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
1133 ip
->i_df
.if_flags
= XFS_IFEXTENTS
;
1134 ip
->i_df
.if_bytes
= ip
->i_df
.if_real_bytes
= 0;
1135 ip
->i_df
.if_u1
.if_extents
= NULL
;
1141 * Attribute fork settings for new inode.
1143 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
1144 ip
->i_d
.di_anextents
= 0;
1147 * Log the new values stuffed into the inode.
1149 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
1150 xfs_trans_log_inode(tp
, ip
, flags
);
1152 /* now that we have an i_mode we can setup inode ops and unlock */
1153 xfs_setup_inode(ip
);
1155 /* now we have set up the vfs inode we can associate the filestream */
1157 error
= xfs_filestream_associate(pip
, ip
);
1161 xfs_iflags_set(ip
, XFS_IFILESTREAM
);
1169 * Free up the underlying blocks past new_size. The new size must be smaller
1170 * than the current size. This routine can be used both for the attribute and
1171 * data fork, and does not modify the inode size, which is left to the caller.
1173 * The transaction passed to this routine must have made a permanent log
1174 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1175 * given transaction and start new ones, so make sure everything involved in
1176 * the transaction is tidy before calling here. Some transaction will be
1177 * returned to the caller to be committed. The incoming transaction must
1178 * already include the inode, and both inode locks must be held exclusively.
1179 * The inode must also be "held" within the transaction. On return the inode
1180 * will be "held" within the returned transaction. This routine does NOT
1181 * require any disk space to be reserved for it within the transaction.
1183 * If we get an error, we must return with the inode locked and linked into the
1184 * current transaction. This keeps things simple for the higher level code,
1185 * because it always knows that the inode is locked and held in the transaction
1186 * that returns to it whether errors occur or not. We don't mark the inode
1187 * dirty on error so that transactions can be easily aborted if possible.
1190 xfs_itruncate_extents(
1191 struct xfs_trans
**tpp
,
1192 struct xfs_inode
*ip
,
1194 xfs_fsize_t new_size
)
1196 struct xfs_mount
*mp
= ip
->i_mount
;
1197 struct xfs_trans
*tp
= *tpp
;
1198 struct xfs_trans
*ntp
;
1199 xfs_bmap_free_t free_list
;
1200 xfs_fsblock_t first_block
;
1201 xfs_fileoff_t first_unmap_block
;
1202 xfs_fileoff_t last_block
;
1203 xfs_filblks_t unmap_len
;
1208 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_IOLOCK_EXCL
));
1209 ASSERT(new_size
<= ip
->i_size
);
1210 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
1211 ASSERT(ip
->i_itemp
!= NULL
);
1212 ASSERT(ip
->i_itemp
->ili_lock_flags
== 0);
1213 ASSERT(!XFS_NOT_DQATTACHED(mp
, ip
));
1215 trace_xfs_itruncate_extents_start(ip
, new_size
);
1218 * Since it is possible for space to become allocated beyond
1219 * the end of the file (in a crash where the space is allocated
1220 * but the inode size is not yet updated), simply remove any
1221 * blocks which show up between the new EOF and the maximum
1222 * possible file size. If the first block to be removed is
1223 * beyond the maximum file size (ie it is the same as last_block),
1224 * then there is nothing to do.
1226 first_unmap_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)new_size
);
1227 last_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
));
1228 if (first_unmap_block
== last_block
)
1231 ASSERT(first_unmap_block
< last_block
);
1232 unmap_len
= last_block
- first_unmap_block
+ 1;
1234 xfs_bmap_init(&free_list
, &first_block
);
1235 error
= xfs_bunmapi(tp
, ip
,
1236 first_unmap_block
, unmap_len
,
1237 xfs_bmapi_aflag(whichfork
),
1238 XFS_ITRUNC_MAX_EXTENTS
,
1239 &first_block
, &free_list
,
1242 goto out_bmap_cancel
;
1245 * Duplicate the transaction that has the permanent
1246 * reservation and commit the old transaction.
1248 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
1250 xfs_trans_ijoin(tp
, ip
, 0);
1252 goto out_bmap_cancel
;
1256 * Mark the inode dirty so it will be logged and
1257 * moved forward in the log as part of every commit.
1259 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1262 ntp
= xfs_trans_dup(tp
);
1263 error
= xfs_trans_commit(tp
, 0);
1266 xfs_trans_ijoin(tp
, ip
, 0);
1272 * Transaction commit worked ok so we can drop the extra ticket
1273 * reference that we gained in xfs_trans_dup()
1275 xfs_log_ticket_put(tp
->t_ticket
);
1276 error
= xfs_trans_reserve(tp
, 0,
1277 XFS_ITRUNCATE_LOG_RES(mp
), 0,
1278 XFS_TRANS_PERM_LOG_RES
,
1279 XFS_ITRUNCATE_LOG_COUNT
);
1285 * Always re-log the inode so that our permanent transaction can keep
1286 * on rolling it forward in the log.
1288 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1290 trace_xfs_itruncate_extents_end(ip
, new_size
);
1297 * If the bunmapi call encounters an error, return to the caller where
1298 * the transaction can be properly aborted. We just need to make sure
1299 * we're not holding any resources that we were not when we came in.
1301 xfs_bmap_cancel(&free_list
);
1306 * This is called when the inode's link count goes to 0.
1307 * We place the on-disk inode on a list in the AGI. It
1308 * will be pulled from this list when the inode is freed.
1325 ASSERT(ip
->i_d
.di_nlink
== 0);
1326 ASSERT(ip
->i_d
.di_mode
!= 0);
1331 * Get the agi buffer first. It ensures lock ordering
1334 error
= xfs_read_agi(mp
, tp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
), &agibp
);
1337 agi
= XFS_BUF_TO_AGI(agibp
);
1340 * Get the index into the agi hash table for the
1341 * list this inode will go on.
1343 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
1345 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
1346 ASSERT(agi
->agi_unlinked
[bucket_index
]);
1347 ASSERT(be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != agino
);
1349 if (agi
->agi_unlinked
[bucket_index
] != cpu_to_be32(NULLAGINO
)) {
1351 * There is already another inode in the bucket we need
1352 * to add ourselves to. Add us at the front of the list.
1353 * Here we put the head pointer into our next pointer,
1354 * and then we fall through to point the head at us.
1356 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, XBF_LOCK
);
1360 ASSERT(dip
->di_next_unlinked
== cpu_to_be32(NULLAGINO
));
1361 dip
->di_next_unlinked
= agi
->agi_unlinked
[bucket_index
];
1362 offset
= ip
->i_imap
.im_boffset
+
1363 offsetof(xfs_dinode_t
, di_next_unlinked
);
1364 xfs_trans_inode_buf(tp
, ibp
);
1365 xfs_trans_log_buf(tp
, ibp
, offset
,
1366 (offset
+ sizeof(xfs_agino_t
) - 1));
1367 xfs_inobp_check(mp
, ibp
);
1371 * Point the bucket head pointer at the inode being inserted.
1374 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(agino
);
1375 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
1376 (sizeof(xfs_agino_t
) * bucket_index
);
1377 xfs_trans_log_buf(tp
, agibp
, offset
,
1378 (offset
+ sizeof(xfs_agino_t
) - 1));
1383 * Pull the on-disk inode from the AGI unlinked list.
1396 xfs_agnumber_t agno
;
1398 xfs_agino_t next_agino
;
1399 xfs_buf_t
*last_ibp
;
1400 xfs_dinode_t
*last_dip
= NULL
;
1402 int offset
, last_offset
= 0;
1406 agno
= XFS_INO_TO_AGNO(mp
, ip
->i_ino
);
1409 * Get the agi buffer first. It ensures lock ordering
1412 error
= xfs_read_agi(mp
, tp
, agno
, &agibp
);
1416 agi
= XFS_BUF_TO_AGI(agibp
);
1419 * Get the index into the agi hash table for the
1420 * list this inode will go on.
1422 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
1424 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
1425 ASSERT(agi
->agi_unlinked
[bucket_index
] != cpu_to_be32(NULLAGINO
));
1426 ASSERT(agi
->agi_unlinked
[bucket_index
]);
1428 if (be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) == agino
) {
1430 * We're at the head of the list. Get the inode's
1431 * on-disk buffer to see if there is anyone after us
1432 * on the list. Only modify our next pointer if it
1433 * is not already NULLAGINO. This saves us the overhead
1434 * of dealing with the buffer when there is no need to
1437 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, XBF_LOCK
);
1439 xfs_warn(mp
, "%s: xfs_itobp() returned error %d.",
1443 next_agino
= be32_to_cpu(dip
->di_next_unlinked
);
1444 ASSERT(next_agino
!= 0);
1445 if (next_agino
!= NULLAGINO
) {
1446 dip
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
1447 offset
= ip
->i_imap
.im_boffset
+
1448 offsetof(xfs_dinode_t
, di_next_unlinked
);
1449 xfs_trans_inode_buf(tp
, ibp
);
1450 xfs_trans_log_buf(tp
, ibp
, offset
,
1451 (offset
+ sizeof(xfs_agino_t
) - 1));
1452 xfs_inobp_check(mp
, ibp
);
1454 xfs_trans_brelse(tp
, ibp
);
1457 * Point the bucket head pointer at the next inode.
1459 ASSERT(next_agino
!= 0);
1460 ASSERT(next_agino
!= agino
);
1461 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(next_agino
);
1462 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
1463 (sizeof(xfs_agino_t
) * bucket_index
);
1464 xfs_trans_log_buf(tp
, agibp
, offset
,
1465 (offset
+ sizeof(xfs_agino_t
) - 1));
1468 * We need to search the list for the inode being freed.
1470 next_agino
= be32_to_cpu(agi
->agi_unlinked
[bucket_index
]);
1472 while (next_agino
!= agino
) {
1474 * If the last inode wasn't the one pointing to
1475 * us, then release its buffer since we're not
1476 * going to do anything with it.
1478 if (last_ibp
!= NULL
) {
1479 xfs_trans_brelse(tp
, last_ibp
);
1481 next_ino
= XFS_AGINO_TO_INO(mp
, agno
, next_agino
);
1482 error
= xfs_inotobp(mp
, tp
, next_ino
, &last_dip
,
1483 &last_ibp
, &last_offset
, 0);
1486 "%s: xfs_inotobp() returned error %d.",
1490 next_agino
= be32_to_cpu(last_dip
->di_next_unlinked
);
1491 ASSERT(next_agino
!= NULLAGINO
);
1492 ASSERT(next_agino
!= 0);
1495 * Now last_ibp points to the buffer previous to us on
1496 * the unlinked list. Pull us from the list.
1498 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, XBF_LOCK
);
1500 xfs_warn(mp
, "%s: xfs_itobp(2) returned error %d.",
1504 next_agino
= be32_to_cpu(dip
->di_next_unlinked
);
1505 ASSERT(next_agino
!= 0);
1506 ASSERT(next_agino
!= agino
);
1507 if (next_agino
!= NULLAGINO
) {
1508 dip
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
1509 offset
= ip
->i_imap
.im_boffset
+
1510 offsetof(xfs_dinode_t
, di_next_unlinked
);
1511 xfs_trans_inode_buf(tp
, ibp
);
1512 xfs_trans_log_buf(tp
, ibp
, offset
,
1513 (offset
+ sizeof(xfs_agino_t
) - 1));
1514 xfs_inobp_check(mp
, ibp
);
1516 xfs_trans_brelse(tp
, ibp
);
1519 * Point the previous inode on the list to the next inode.
1521 last_dip
->di_next_unlinked
= cpu_to_be32(next_agino
);
1522 ASSERT(next_agino
!= 0);
1523 offset
= last_offset
+ offsetof(xfs_dinode_t
, di_next_unlinked
);
1524 xfs_trans_inode_buf(tp
, last_ibp
);
1525 xfs_trans_log_buf(tp
, last_ibp
, offset
,
1526 (offset
+ sizeof(xfs_agino_t
) - 1));
1527 xfs_inobp_check(mp
, last_ibp
);
1533 * A big issue when freeing the inode cluster is is that we _cannot_ skip any
1534 * inodes that are in memory - they all must be marked stale and attached to
1535 * the cluster buffer.
1539 xfs_inode_t
*free_ip
,
1543 xfs_mount_t
*mp
= free_ip
->i_mount
;
1544 int blks_per_cluster
;
1551 xfs_inode_log_item_t
*iip
;
1552 xfs_log_item_t
*lip
;
1553 struct xfs_perag
*pag
;
1555 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, inum
));
1556 if (mp
->m_sb
.sb_blocksize
>= XFS_INODE_CLUSTER_SIZE(mp
)) {
1557 blks_per_cluster
= 1;
1558 ninodes
= mp
->m_sb
.sb_inopblock
;
1559 nbufs
= XFS_IALLOC_BLOCKS(mp
);
1561 blks_per_cluster
= XFS_INODE_CLUSTER_SIZE(mp
) /
1562 mp
->m_sb
.sb_blocksize
;
1563 ninodes
= blks_per_cluster
* mp
->m_sb
.sb_inopblock
;
1564 nbufs
= XFS_IALLOC_BLOCKS(mp
) / blks_per_cluster
;
1567 for (j
= 0; j
< nbufs
; j
++, inum
+= ninodes
) {
1568 blkno
= XFS_AGB_TO_DADDR(mp
, XFS_INO_TO_AGNO(mp
, inum
),
1569 XFS_INO_TO_AGBNO(mp
, inum
));
1572 * We obtain and lock the backing buffer first in the process
1573 * here, as we have to ensure that any dirty inode that we
1574 * can't get the flush lock on is attached to the buffer.
1575 * If we scan the in-memory inodes first, then buffer IO can
1576 * complete before we get a lock on it, and hence we may fail
1577 * to mark all the active inodes on the buffer stale.
1579 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, blkno
,
1580 mp
->m_bsize
* blks_per_cluster
,
1586 * Walk the inodes already attached to the buffer and mark them
1587 * stale. These will all have the flush locks held, so an
1588 * in-memory inode walk can't lock them. By marking them all
1589 * stale first, we will not attempt to lock them in the loop
1590 * below as the XFS_ISTALE flag will be set.
1594 if (lip
->li_type
== XFS_LI_INODE
) {
1595 iip
= (xfs_inode_log_item_t
*)lip
;
1596 ASSERT(iip
->ili_logged
== 1);
1597 lip
->li_cb
= xfs_istale_done
;
1598 xfs_trans_ail_copy_lsn(mp
->m_ail
,
1599 &iip
->ili_flush_lsn
,
1600 &iip
->ili_item
.li_lsn
);
1601 xfs_iflags_set(iip
->ili_inode
, XFS_ISTALE
);
1603 lip
= lip
->li_bio_list
;
1608 * For each inode in memory attempt to add it to the inode
1609 * buffer and set it up for being staled on buffer IO
1610 * completion. This is safe as we've locked out tail pushing
1611 * and flushing by locking the buffer.
1613 * We have already marked every inode that was part of a
1614 * transaction stale above, which means there is no point in
1615 * even trying to lock them.
1617 for (i
= 0; i
< ninodes
; i
++) {
1620 ip
= radix_tree_lookup(&pag
->pag_ici_root
,
1621 XFS_INO_TO_AGINO(mp
, (inum
+ i
)));
1623 /* Inode not in memory, nothing to do */
1630 * because this is an RCU protected lookup, we could
1631 * find a recently freed or even reallocated inode
1632 * during the lookup. We need to check under the
1633 * i_flags_lock for a valid inode here. Skip it if it
1634 * is not valid, the wrong inode or stale.
1636 spin_lock(&ip
->i_flags_lock
);
1637 if (ip
->i_ino
!= inum
+ i
||
1638 __xfs_iflags_test(ip
, XFS_ISTALE
)) {
1639 spin_unlock(&ip
->i_flags_lock
);
1643 spin_unlock(&ip
->i_flags_lock
);
1646 * Don't try to lock/unlock the current inode, but we
1647 * _cannot_ skip the other inodes that we did not find
1648 * in the list attached to the buffer and are not
1649 * already marked stale. If we can't lock it, back off
1652 if (ip
!= free_ip
&&
1653 !xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
)) {
1661 xfs_iflags_set(ip
, XFS_ISTALE
);
1664 * we don't need to attach clean inodes or those only
1665 * with unlogged changes (which we throw away, anyway).
1668 if (!iip
|| xfs_inode_clean(ip
)) {
1669 ASSERT(ip
!= free_ip
);
1670 ip
->i_update_core
= 0;
1672 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1676 iip
->ili_last_fields
= iip
->ili_format
.ilf_fields
;
1677 iip
->ili_format
.ilf_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 && ip
->i_size
== 0) ||
1724 (!S_ISREG(ip
->i_d
.di_mode
)));
1725 ASSERT(ip
->i_d
.di_nblocks
== 0);
1728 * Pull the on-disk inode from the AGI unlinked list.
1730 error
= xfs_iunlink_remove(tp
, ip
);
1735 error
= xfs_difree(tp
, ip
->i_ino
, flist
, &delete, &first_ino
);
1739 ip
->i_d
.di_mode
= 0; /* mark incore inode as free */
1740 ip
->i_d
.di_flags
= 0;
1741 ip
->i_d
.di_dmevmask
= 0;
1742 ip
->i_d
.di_forkoff
= 0; /* mark the attr fork not in use */
1743 ip
->i_df
.if_ext_max
=
1744 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
1745 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
1746 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
1748 * Bump the generation count so no one will be confused
1749 * by reincarnations of this inode.
1753 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1755 error
= xfs_itobp(ip
->i_mount
, tp
, ip
, &dip
, &ibp
, XBF_LOCK
);
1760 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
1761 * from picking up this inode when it is reclaimed (its incore state
1762 * initialzed but not flushed to disk yet). The in-core di_mode is
1763 * already cleared and a corresponding transaction logged.
1764 * The hack here just synchronizes the in-core to on-disk
1765 * di_mode value in advance before the actual inode sync to disk.
1766 * This is OK because the inode is already unlinked and would never
1767 * change its di_mode again for this inode generation.
1768 * This is a temporary hack that would require a proper fix
1774 error
= xfs_ifree_cluster(ip
, tp
, first_ino
);
1781 * Reallocate the space for if_broot based on the number of records
1782 * being added or deleted as indicated in rec_diff. Move the records
1783 * and pointers in if_broot to fit the new size. When shrinking this
1784 * will eliminate holes between the records and pointers created by
1785 * the caller. When growing this will create holes to be filled in
1788 * The caller must not request to add more records than would fit in
1789 * the on-disk inode root. If the if_broot is currently NULL, then
1790 * if we adding records one will be allocated. The caller must also
1791 * not request that the number of records go below zero, although
1792 * it can go to zero.
1794 * ip -- the inode whose if_broot area is changing
1795 * ext_diff -- the change in the number of records, positive or negative,
1796 * requested for the if_broot array.
1804 struct xfs_mount
*mp
= ip
->i_mount
;
1807 struct xfs_btree_block
*new_broot
;
1814 * Handle the degenerate case quietly.
1816 if (rec_diff
== 0) {
1820 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1823 * If there wasn't any memory allocated before, just
1824 * allocate it now and get out.
1826 if (ifp
->if_broot_bytes
== 0) {
1827 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff
);
1828 ifp
->if_broot
= kmem_alloc(new_size
, KM_SLEEP
| KM_NOFS
);
1829 ifp
->if_broot_bytes
= (int)new_size
;
1834 * If there is already an existing if_broot, then we need
1835 * to realloc() it and shift the pointers to their new
1836 * location. The records don't change location because
1837 * they are kept butted up against the btree block header.
1839 cur_max
= xfs_bmbt_maxrecs(mp
, ifp
->if_broot_bytes
, 0);
1840 new_max
= cur_max
+ rec_diff
;
1841 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max
);
1842 ifp
->if_broot
= kmem_realloc(ifp
->if_broot
, new_size
,
1843 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max
), /* old size */
1844 KM_SLEEP
| KM_NOFS
);
1845 op
= (char *)XFS_BMAP_BROOT_PTR_ADDR(mp
, ifp
->if_broot
, 1,
1846 ifp
->if_broot_bytes
);
1847 np
= (char *)XFS_BMAP_BROOT_PTR_ADDR(mp
, ifp
->if_broot
, 1,
1849 ifp
->if_broot_bytes
= (int)new_size
;
1850 ASSERT(ifp
->if_broot_bytes
<=
1851 XFS_IFORK_SIZE(ip
, whichfork
) + XFS_BROOT_SIZE_ADJ
);
1852 memmove(np
, op
, cur_max
* (uint
)sizeof(xfs_dfsbno_t
));
1857 * rec_diff is less than 0. In this case, we are shrinking the
1858 * if_broot buffer. It must already exist. If we go to zero
1859 * records, just get rid of the root and clear the status bit.
1861 ASSERT((ifp
->if_broot
!= NULL
) && (ifp
->if_broot_bytes
> 0));
1862 cur_max
= xfs_bmbt_maxrecs(mp
, ifp
->if_broot_bytes
, 0);
1863 new_max
= cur_max
+ rec_diff
;
1864 ASSERT(new_max
>= 0);
1866 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max
);
1870 new_broot
= kmem_alloc(new_size
, KM_SLEEP
| KM_NOFS
);
1872 * First copy over the btree block header.
1874 memcpy(new_broot
, ifp
->if_broot
, XFS_BTREE_LBLOCK_LEN
);
1877 ifp
->if_flags
&= ~XFS_IFBROOT
;
1881 * Only copy the records and pointers if there are any.
1885 * First copy the records.
1887 op
= (char *)XFS_BMBT_REC_ADDR(mp
, ifp
->if_broot
, 1);
1888 np
= (char *)XFS_BMBT_REC_ADDR(mp
, new_broot
, 1);
1889 memcpy(np
, op
, new_max
* (uint
)sizeof(xfs_bmbt_rec_t
));
1892 * Then copy the pointers.
1894 op
= (char *)XFS_BMAP_BROOT_PTR_ADDR(mp
, ifp
->if_broot
, 1,
1895 ifp
->if_broot_bytes
);
1896 np
= (char *)XFS_BMAP_BROOT_PTR_ADDR(mp
, new_broot
, 1,
1898 memcpy(np
, op
, new_max
* (uint
)sizeof(xfs_dfsbno_t
));
1900 kmem_free(ifp
->if_broot
);
1901 ifp
->if_broot
= new_broot
;
1902 ifp
->if_broot_bytes
= (int)new_size
;
1903 ASSERT(ifp
->if_broot_bytes
<=
1904 XFS_IFORK_SIZE(ip
, whichfork
) + XFS_BROOT_SIZE_ADJ
);
1910 * This is called when the amount of space needed for if_data
1911 * is increased or decreased. The change in size is indicated by
1912 * the number of bytes that need to be added or deleted in the
1913 * byte_diff parameter.
1915 * If the amount of space needed has decreased below the size of the
1916 * inline buffer, then switch to using the inline buffer. Otherwise,
1917 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
1918 * to what is needed.
1920 * ip -- the inode whose if_data area is changing
1921 * byte_diff -- the change in the number of bytes, positive or negative,
1922 * requested for the if_data array.
1934 if (byte_diff
== 0) {
1938 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1939 new_size
= (int)ifp
->if_bytes
+ byte_diff
;
1940 ASSERT(new_size
>= 0);
1942 if (new_size
== 0) {
1943 if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
1944 kmem_free(ifp
->if_u1
.if_data
);
1946 ifp
->if_u1
.if_data
= NULL
;
1948 } else if (new_size
<= sizeof(ifp
->if_u2
.if_inline_data
)) {
1950 * If the valid extents/data can fit in if_inline_ext/data,
1951 * copy them from the malloc'd vector and free it.
1953 if (ifp
->if_u1
.if_data
== NULL
) {
1954 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
1955 } else if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
1956 ASSERT(ifp
->if_real_bytes
!= 0);
1957 memcpy(ifp
->if_u2
.if_inline_data
, ifp
->if_u1
.if_data
,
1959 kmem_free(ifp
->if_u1
.if_data
);
1960 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
1965 * Stuck with malloc/realloc.
1966 * For inline data, the underlying buffer must be
1967 * a multiple of 4 bytes in size so that it can be
1968 * logged and stay on word boundaries. We enforce
1971 real_size
= roundup(new_size
, 4);
1972 if (ifp
->if_u1
.if_data
== NULL
) {
1973 ASSERT(ifp
->if_real_bytes
== 0);
1974 ifp
->if_u1
.if_data
= kmem_alloc(real_size
,
1975 KM_SLEEP
| KM_NOFS
);
1976 } else if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
1978 * Only do the realloc if the underlying size
1979 * is really changing.
1981 if (ifp
->if_real_bytes
!= real_size
) {
1982 ifp
->if_u1
.if_data
=
1983 kmem_realloc(ifp
->if_u1
.if_data
,
1986 KM_SLEEP
| KM_NOFS
);
1989 ASSERT(ifp
->if_real_bytes
== 0);
1990 ifp
->if_u1
.if_data
= kmem_alloc(real_size
,
1991 KM_SLEEP
| KM_NOFS
);
1992 memcpy(ifp
->if_u1
.if_data
, ifp
->if_u2
.if_inline_data
,
1996 ifp
->if_real_bytes
= real_size
;
1997 ifp
->if_bytes
= new_size
;
1998 ASSERT(ifp
->if_bytes
<= XFS_IFORK_SIZE(ip
, whichfork
));
2008 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2009 if (ifp
->if_broot
!= NULL
) {
2010 kmem_free(ifp
->if_broot
);
2011 ifp
->if_broot
= NULL
;
2015 * If the format is local, then we can't have an extents
2016 * array so just look for an inline data array. If we're
2017 * not local then we may or may not have an extents list,
2018 * so check and free it up if we do.
2020 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
) {
2021 if ((ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) &&
2022 (ifp
->if_u1
.if_data
!= NULL
)) {
2023 ASSERT(ifp
->if_real_bytes
!= 0);
2024 kmem_free(ifp
->if_u1
.if_data
);
2025 ifp
->if_u1
.if_data
= NULL
;
2026 ifp
->if_real_bytes
= 0;
2028 } else if ((ifp
->if_flags
& XFS_IFEXTENTS
) &&
2029 ((ifp
->if_flags
& XFS_IFEXTIREC
) ||
2030 ((ifp
->if_u1
.if_extents
!= NULL
) &&
2031 (ifp
->if_u1
.if_extents
!= ifp
->if_u2
.if_inline_ext
)))) {
2032 ASSERT(ifp
->if_real_bytes
!= 0);
2033 xfs_iext_destroy(ifp
);
2035 ASSERT(ifp
->if_u1
.if_extents
== NULL
||
2036 ifp
->if_u1
.if_extents
== ifp
->if_u2
.if_inline_ext
);
2037 ASSERT(ifp
->if_real_bytes
== 0);
2038 if (whichfork
== XFS_ATTR_FORK
) {
2039 kmem_zone_free(xfs_ifork_zone
, ip
->i_afp
);
2045 * This is called to unpin an inode. The caller must have the inode locked
2046 * in at least shared mode so that the buffer cannot be subsequently pinned
2047 * once someone is waiting for it to be unpinned.
2051 struct xfs_inode
*ip
)
2053 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2055 trace_xfs_inode_unpin_nowait(ip
, _RET_IP_
);
2057 /* Give the log a push to start the unpinning I/O */
2058 xfs_log_force_lsn(ip
->i_mount
, ip
->i_itemp
->ili_last_lsn
, 0);
2064 struct xfs_inode
*ip
)
2066 if (xfs_ipincount(ip
)) {
2067 xfs_iunpin_nowait(ip
);
2068 wait_event(ip
->i_ipin_wait
, (xfs_ipincount(ip
) == 0));
2073 * xfs_iextents_copy()
2075 * This is called to copy the REAL extents (as opposed to the delayed
2076 * allocation extents) from the inode into the given buffer. It
2077 * returns the number of bytes copied into the buffer.
2079 * If there are no delayed allocation extents, then we can just
2080 * memcpy() the extents into the buffer. Otherwise, we need to
2081 * examine each extent in turn and skip those which are delayed.
2093 xfs_fsblock_t start_block
;
2095 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2096 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2097 ASSERT(ifp
->if_bytes
> 0);
2099 nrecs
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
2100 XFS_BMAP_TRACE_EXLIST(ip
, nrecs
, whichfork
);
2104 * There are some delayed allocation extents in the
2105 * inode, so copy the extents one at a time and skip
2106 * the delayed ones. There must be at least one
2107 * non-delayed extent.
2110 for (i
= 0; i
< nrecs
; i
++) {
2111 xfs_bmbt_rec_host_t
*ep
= xfs_iext_get_ext(ifp
, i
);
2112 start_block
= xfs_bmbt_get_startblock(ep
);
2113 if (isnullstartblock(start_block
)) {
2115 * It's a delayed allocation extent, so skip it.
2120 /* Translate to on disk format */
2121 put_unaligned(cpu_to_be64(ep
->l0
), &dp
->l0
);
2122 put_unaligned(cpu_to_be64(ep
->l1
), &dp
->l1
);
2126 ASSERT(copied
!= 0);
2127 xfs_validate_extents(ifp
, copied
, XFS_EXTFMT_INODE(ip
));
2129 return (copied
* (uint
)sizeof(xfs_bmbt_rec_t
));
2133 * Each of the following cases stores data into the same region
2134 * of the on-disk inode, so only one of them can be valid at
2135 * any given time. While it is possible to have conflicting formats
2136 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2137 * in EXTENTS format, this can only happen when the fork has
2138 * changed formats after being modified but before being flushed.
2139 * In these cases, the format always takes precedence, because the
2140 * format indicates the current state of the fork.
2147 xfs_inode_log_item_t
*iip
,
2154 #ifdef XFS_TRANS_DEBUG
2157 static const short brootflag
[2] =
2158 { XFS_ILOG_DBROOT
, XFS_ILOG_ABROOT
};
2159 static const short dataflag
[2] =
2160 { XFS_ILOG_DDATA
, XFS_ILOG_ADATA
};
2161 static const short extflag
[2] =
2162 { XFS_ILOG_DEXT
, XFS_ILOG_AEXT
};
2166 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2168 * This can happen if we gave up in iformat in an error path,
2169 * for the attribute fork.
2172 ASSERT(whichfork
== XFS_ATTR_FORK
);
2175 cp
= XFS_DFORK_PTR(dip
, whichfork
);
2177 switch (XFS_IFORK_FORMAT(ip
, whichfork
)) {
2178 case XFS_DINODE_FMT_LOCAL
:
2179 if ((iip
->ili_format
.ilf_fields
& dataflag
[whichfork
]) &&
2180 (ifp
->if_bytes
> 0)) {
2181 ASSERT(ifp
->if_u1
.if_data
!= NULL
);
2182 ASSERT(ifp
->if_bytes
<= XFS_IFORK_SIZE(ip
, whichfork
));
2183 memcpy(cp
, ifp
->if_u1
.if_data
, ifp
->if_bytes
);
2187 case XFS_DINODE_FMT_EXTENTS
:
2188 ASSERT((ifp
->if_flags
& XFS_IFEXTENTS
) ||
2189 !(iip
->ili_format
.ilf_fields
& extflag
[whichfork
]));
2190 if ((iip
->ili_format
.ilf_fields
& extflag
[whichfork
]) &&
2191 (ifp
->if_bytes
> 0)) {
2192 ASSERT(xfs_iext_get_ext(ifp
, 0));
2193 ASSERT(XFS_IFORK_NEXTENTS(ip
, whichfork
) > 0);
2194 (void)xfs_iextents_copy(ip
, (xfs_bmbt_rec_t
*)cp
,
2199 case XFS_DINODE_FMT_BTREE
:
2200 if ((iip
->ili_format
.ilf_fields
& brootflag
[whichfork
]) &&
2201 (ifp
->if_broot_bytes
> 0)) {
2202 ASSERT(ifp
->if_broot
!= NULL
);
2203 ASSERT(ifp
->if_broot_bytes
<=
2204 (XFS_IFORK_SIZE(ip
, whichfork
) +
2205 XFS_BROOT_SIZE_ADJ
));
2206 xfs_bmbt_to_bmdr(mp
, ifp
->if_broot
, ifp
->if_broot_bytes
,
2207 (xfs_bmdr_block_t
*)cp
,
2208 XFS_DFORK_SIZE(dip
, mp
, whichfork
));
2212 case XFS_DINODE_FMT_DEV
:
2213 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_DEV
) {
2214 ASSERT(whichfork
== XFS_DATA_FORK
);
2215 xfs_dinode_put_rdev(dip
, ip
->i_df
.if_u2
.if_rdev
);
2219 case XFS_DINODE_FMT_UUID
:
2220 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_UUID
) {
2221 ASSERT(whichfork
== XFS_DATA_FORK
);
2222 memcpy(XFS_DFORK_DPTR(dip
),
2223 &ip
->i_df
.if_u2
.if_uuid
,
2239 xfs_mount_t
*mp
= ip
->i_mount
;
2240 struct xfs_perag
*pag
;
2241 unsigned long first_index
, mask
;
2242 unsigned long inodes_per_cluster
;
2244 xfs_inode_t
**ilist
;
2251 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
2253 inodes_per_cluster
= XFS_INODE_CLUSTER_SIZE(mp
) >> mp
->m_sb
.sb_inodelog
;
2254 ilist_size
= inodes_per_cluster
* sizeof(xfs_inode_t
*);
2255 ilist
= kmem_alloc(ilist_size
, KM_MAYFAIL
|KM_NOFS
);
2259 mask
= ~(((XFS_INODE_CLUSTER_SIZE(mp
) >> mp
->m_sb
.sb_inodelog
)) - 1);
2260 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
) & mask
;
2262 /* really need a gang lookup range call here */
2263 nr_found
= radix_tree_gang_lookup(&pag
->pag_ici_root
, (void**)ilist
,
2264 first_index
, inodes_per_cluster
);
2268 for (i
= 0; i
< nr_found
; i
++) {
2274 * because this is an RCU protected lookup, we could find a
2275 * recently freed or even reallocated inode during the lookup.
2276 * We need to check under the i_flags_lock for a valid inode
2277 * here. Skip it if it is not valid or the wrong inode.
2279 spin_lock(&ip
->i_flags_lock
);
2281 (XFS_INO_TO_AGINO(mp
, iq
->i_ino
) & mask
) != first_index
) {
2282 spin_unlock(&ip
->i_flags_lock
);
2285 spin_unlock(&ip
->i_flags_lock
);
2288 * Do an un-protected check to see if the inode is dirty and
2289 * is a candidate for flushing. These checks will be repeated
2290 * later after the appropriate locks are acquired.
2292 if (xfs_inode_clean(iq
) && xfs_ipincount(iq
) == 0)
2296 * Try to get locks. If any are unavailable or it is pinned,
2297 * then this inode cannot be flushed and is skipped.
2300 if (!xfs_ilock_nowait(iq
, XFS_ILOCK_SHARED
))
2302 if (!xfs_iflock_nowait(iq
)) {
2303 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
2306 if (xfs_ipincount(iq
)) {
2308 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
2313 * arriving here means that this inode can be flushed. First
2314 * re-check that it's dirty before flushing.
2316 if (!xfs_inode_clean(iq
)) {
2318 error
= xfs_iflush_int(iq
, bp
);
2320 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
2321 goto cluster_corrupt_out
;
2327 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
2331 XFS_STATS_INC(xs_icluster_flushcnt
);
2332 XFS_STATS_ADD(xs_icluster_flushinode
, clcount
);
2343 cluster_corrupt_out
:
2345 * Corruption detected in the clustering loop. Invalidate the
2346 * inode buffer and shut down the filesystem.
2350 * Clean up the buffer. If it was B_DELWRI, just release it --
2351 * brelse can handle it with no problems. If not, shut down the
2352 * filesystem before releasing the buffer.
2354 bufwasdelwri
= XFS_BUF_ISDELAYWRITE(bp
);
2358 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
2360 if (!bufwasdelwri
) {
2362 * Just like incore_relse: if we have b_iodone functions,
2363 * mark the buffer as an error and call them. Otherwise
2364 * mark it as stale and brelse.
2369 xfs_buf_ioerror(bp
, EIO
);
2370 xfs_buf_ioend(bp
, 0);
2378 * Unlocks the flush lock
2380 xfs_iflush_abort(iq
);
2383 return XFS_ERROR(EFSCORRUPTED
);
2387 * xfs_iflush() will write a modified inode's changes out to the
2388 * inode's on disk home. The caller must have the inode lock held
2389 * in at least shared mode and the inode flush completion must be
2390 * active as well. The inode lock will still be held upon return from
2391 * the call and the caller is free to unlock it.
2392 * The inode flush will be completed when the inode reaches the disk.
2393 * The flags indicate how the inode's buffer should be written out.
2400 xfs_inode_log_item_t
*iip
;
2406 XFS_STATS_INC(xs_iflush_count
);
2408 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2409 ASSERT(!completion_done(&ip
->i_flush
));
2410 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
2411 ip
->i_d
.di_nextents
> ip
->i_df
.if_ext_max
);
2417 * We can't flush the inode until it is unpinned, so wait for it if we
2418 * are allowed to block. We know no one new can pin it, because we are
2419 * holding the inode lock shared and you need to hold it exclusively to
2422 * If we are not allowed to block, force the log out asynchronously so
2423 * that when we come back the inode will be unpinned. If other inodes
2424 * in the same cluster are dirty, they will probably write the inode
2425 * out for us if they occur after the log force completes.
2427 if (!(flags
& SYNC_WAIT
) && xfs_ipincount(ip
)) {
2428 xfs_iunpin_nowait(ip
);
2432 xfs_iunpin_wait(ip
);
2435 * For stale inodes we cannot rely on the backing buffer remaining
2436 * stale in cache for the remaining life of the stale inode and so
2437 * xfs_itobp() below may give us a buffer that no longer contains
2438 * inodes below. We have to check this after ensuring the inode is
2439 * unpinned so that it is safe to reclaim the stale inode after the
2442 if (xfs_iflags_test(ip
, XFS_ISTALE
)) {
2448 * This may have been unpinned because the filesystem is shutting
2449 * down forcibly. If that's the case we must not write this inode
2450 * to disk, because the log record didn't make it to disk!
2452 if (XFS_FORCED_SHUTDOWN(mp
)) {
2453 ip
->i_update_core
= 0;
2455 iip
->ili_format
.ilf_fields
= 0;
2457 return XFS_ERROR(EIO
);
2461 * Get the buffer containing the on-disk inode.
2463 error
= xfs_itobp(mp
, NULL
, ip
, &dip
, &bp
,
2464 (flags
& SYNC_TRYLOCK
) ? XBF_TRYLOCK
: XBF_LOCK
);
2471 * First flush out the inode that xfs_iflush was called with.
2473 error
= xfs_iflush_int(ip
, bp
);
2478 * If the buffer is pinned then push on the log now so we won't
2479 * get stuck waiting in the write for too long.
2481 if (xfs_buf_ispinned(bp
))
2482 xfs_log_force(mp
, 0);
2486 * see if other inodes can be gathered into this write
2488 error
= xfs_iflush_cluster(ip
, bp
);
2490 goto cluster_corrupt_out
;
2492 if (flags
& SYNC_WAIT
)
2493 error
= xfs_bwrite(bp
);
2495 xfs_buf_delwri_queue(bp
);
2502 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
2503 cluster_corrupt_out
:
2505 * Unlocks the flush lock
2507 xfs_iflush_abort(ip
);
2508 return XFS_ERROR(EFSCORRUPTED
);
2517 xfs_inode_log_item_t
*iip
;
2520 #ifdef XFS_TRANS_DEBUG
2524 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2525 ASSERT(!completion_done(&ip
->i_flush
));
2526 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
2527 ip
->i_d
.di_nextents
> ip
->i_df
.if_ext_max
);
2532 /* set *dip = inode's place in the buffer */
2533 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, ip
->i_imap
.im_boffset
);
2536 * Clear i_update_core before copying out the data.
2537 * This is for coordination with our timestamp updates
2538 * that don't hold the inode lock. They will always
2539 * update the timestamps BEFORE setting i_update_core,
2540 * so if we clear i_update_core after they set it we
2541 * are guaranteed to see their updates to the timestamps.
2542 * I believe that this depends on strongly ordered memory
2543 * semantics, but we have that. We use the SYNCHRONIZE
2544 * macro to make sure that the compiler does not reorder
2545 * the i_update_core access below the data copy below.
2547 ip
->i_update_core
= 0;
2551 * Make sure to get the latest timestamps from the Linux inode.
2553 xfs_synchronize_times(ip
);
2555 if (XFS_TEST_ERROR(dip
->di_magic
!= cpu_to_be16(XFS_DINODE_MAGIC
),
2556 mp
, XFS_ERRTAG_IFLUSH_1
, XFS_RANDOM_IFLUSH_1
)) {
2557 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
2558 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
2559 __func__
, ip
->i_ino
, be16_to_cpu(dip
->di_magic
), dip
);
2562 if (XFS_TEST_ERROR(ip
->i_d
.di_magic
!= XFS_DINODE_MAGIC
,
2563 mp
, XFS_ERRTAG_IFLUSH_2
, XFS_RANDOM_IFLUSH_2
)) {
2564 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
2565 "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
2566 __func__
, ip
->i_ino
, ip
, ip
->i_d
.di_magic
);
2569 if (S_ISREG(ip
->i_d
.di_mode
)) {
2571 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2572 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
),
2573 mp
, XFS_ERRTAG_IFLUSH_3
, XFS_RANDOM_IFLUSH_3
)) {
2574 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
2575 "%s: Bad regular inode %Lu, ptr 0x%p",
2576 __func__
, ip
->i_ino
, ip
);
2579 } else if (S_ISDIR(ip
->i_d
.di_mode
)) {
2581 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2582 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
) &&
2583 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_LOCAL
),
2584 mp
, XFS_ERRTAG_IFLUSH_4
, XFS_RANDOM_IFLUSH_4
)) {
2585 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
2586 "%s: Bad directory inode %Lu, ptr 0x%p",
2587 __func__
, ip
->i_ino
, ip
);
2591 if (XFS_TEST_ERROR(ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
>
2592 ip
->i_d
.di_nblocks
, mp
, XFS_ERRTAG_IFLUSH_5
,
2593 XFS_RANDOM_IFLUSH_5
)) {
2594 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
2595 "%s: detected corrupt incore inode %Lu, "
2596 "total extents = %d, nblocks = %Ld, ptr 0x%p",
2597 __func__
, ip
->i_ino
,
2598 ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
,
2599 ip
->i_d
.di_nblocks
, ip
);
2602 if (XFS_TEST_ERROR(ip
->i_d
.di_forkoff
> mp
->m_sb
.sb_inodesize
,
2603 mp
, XFS_ERRTAG_IFLUSH_6
, XFS_RANDOM_IFLUSH_6
)) {
2604 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
2605 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
2606 __func__
, ip
->i_ino
, ip
->i_d
.di_forkoff
, ip
);
2610 * bump the flush iteration count, used to detect flushes which
2611 * postdate a log record during recovery.
2614 ip
->i_d
.di_flushiter
++;
2617 * Copy the dirty parts of the inode into the on-disk
2618 * inode. We always copy out the core of the inode,
2619 * because if the inode is dirty at all the core must
2622 xfs_dinode_to_disk(dip
, &ip
->i_d
);
2624 /* Wrap, we never let the log put out DI_MAX_FLUSH */
2625 if (ip
->i_d
.di_flushiter
== DI_MAX_FLUSH
)
2626 ip
->i_d
.di_flushiter
= 0;
2629 * If this is really an old format inode and the superblock version
2630 * has not been updated to support only new format inodes, then
2631 * convert back to the old inode format. If the superblock version
2632 * has been updated, then make the conversion permanent.
2634 ASSERT(ip
->i_d
.di_version
== 1 || xfs_sb_version_hasnlink(&mp
->m_sb
));
2635 if (ip
->i_d
.di_version
== 1) {
2636 if (!xfs_sb_version_hasnlink(&mp
->m_sb
)) {
2640 ASSERT(ip
->i_d
.di_nlink
<= XFS_MAXLINK_1
);
2641 dip
->di_onlink
= cpu_to_be16(ip
->i_d
.di_nlink
);
2644 * The superblock version has already been bumped,
2645 * so just make the conversion to the new inode
2648 ip
->i_d
.di_version
= 2;
2649 dip
->di_version
= 2;
2650 ip
->i_d
.di_onlink
= 0;
2652 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
2653 memset(&(dip
->di_pad
[0]), 0,
2654 sizeof(dip
->di_pad
));
2655 ASSERT(xfs_get_projid(ip
) == 0);
2659 xfs_iflush_fork(ip
, dip
, iip
, XFS_DATA_FORK
, bp
);
2660 if (XFS_IFORK_Q(ip
))
2661 xfs_iflush_fork(ip
, dip
, iip
, XFS_ATTR_FORK
, bp
);
2662 xfs_inobp_check(mp
, bp
);
2665 * We've recorded everything logged in the inode, so we'd
2666 * like to clear the ilf_fields bits so we don't log and
2667 * flush things unnecessarily. However, we can't stop
2668 * logging all this information until the data we've copied
2669 * into the disk buffer is written to disk. If we did we might
2670 * overwrite the copy of the inode in the log with all the
2671 * data after re-logging only part of it, and in the face of
2672 * a crash we wouldn't have all the data we need to recover.
2674 * What we do is move the bits to the ili_last_fields field.
2675 * When logging the inode, these bits are moved back to the
2676 * ilf_fields field. In the xfs_iflush_done() routine we
2677 * clear ili_last_fields, since we know that the information
2678 * those bits represent is permanently on disk. As long as
2679 * the flush completes before the inode is logged again, then
2680 * both ilf_fields and ili_last_fields will be cleared.
2682 * We can play with the ilf_fields bits here, because the inode
2683 * lock must be held exclusively in order to set bits there
2684 * and the flush lock protects the ili_last_fields bits.
2685 * Set ili_logged so the flush done
2686 * routine can tell whether or not to look in the AIL.
2687 * Also, store the current LSN of the inode so that we can tell
2688 * whether the item has moved in the AIL from xfs_iflush_done().
2689 * In order to read the lsn we need the AIL lock, because
2690 * it is a 64 bit value that cannot be read atomically.
2692 if (iip
!= NULL
&& iip
->ili_format
.ilf_fields
!= 0) {
2693 iip
->ili_last_fields
= iip
->ili_format
.ilf_fields
;
2694 iip
->ili_format
.ilf_fields
= 0;
2695 iip
->ili_logged
= 1;
2697 xfs_trans_ail_copy_lsn(mp
->m_ail
, &iip
->ili_flush_lsn
,
2698 &iip
->ili_item
.li_lsn
);
2701 * Attach the function xfs_iflush_done to the inode's
2702 * buffer. This will remove the inode from the AIL
2703 * and unlock the inode's flush lock when the inode is
2704 * completely written to disk.
2706 xfs_buf_attach_iodone(bp
, xfs_iflush_done
, &iip
->ili_item
);
2708 ASSERT(bp
->b_fspriv
!= NULL
);
2709 ASSERT(bp
->b_iodone
!= NULL
);
2712 * We're flushing an inode which is not in the AIL and has
2713 * not been logged but has i_update_core set. For this
2714 * case we can use a B_DELWRI flush and immediately drop
2715 * the inode flush lock because we can avoid the whole
2716 * AIL state thing. It's OK to drop the flush lock now,
2717 * because we've already locked the buffer and to do anything
2718 * you really need both.
2721 ASSERT(iip
->ili_logged
== 0);
2722 ASSERT(iip
->ili_last_fields
== 0);
2723 ASSERT((iip
->ili_item
.li_flags
& XFS_LI_IN_AIL
) == 0);
2731 return XFS_ERROR(EFSCORRUPTED
);
2736 struct xfs_inode
*ip
)
2740 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2742 bp
= xfs_incore(ip
->i_mount
->m_ddev_targp
, ip
->i_imap
.im_blkno
,
2743 ip
->i_imap
.im_len
, XBF_TRYLOCK
);
2747 if (XFS_BUF_ISDELAYWRITE(bp
)) {
2748 xfs_buf_delwri_promote(bp
);
2749 wake_up_process(ip
->i_mount
->m_ddev_targp
->bt_task
);
2756 * Return a pointer to the extent record at file index idx.
2758 xfs_bmbt_rec_host_t
*
2760 xfs_ifork_t
*ifp
, /* inode fork pointer */
2761 xfs_extnum_t idx
) /* index of target extent */
2764 ASSERT(idx
< ifp
->if_bytes
/ sizeof(xfs_bmbt_rec_t
));
2766 if ((ifp
->if_flags
& XFS_IFEXTIREC
) && (idx
== 0)) {
2767 return ifp
->if_u1
.if_ext_irec
->er_extbuf
;
2768 } else if (ifp
->if_flags
& XFS_IFEXTIREC
) {
2769 xfs_ext_irec_t
*erp
; /* irec pointer */
2770 int erp_idx
= 0; /* irec index */
2771 xfs_extnum_t page_idx
= idx
; /* ext index in target list */
2773 erp
= xfs_iext_idx_to_irec(ifp
, &page_idx
, &erp_idx
, 0);
2774 return &erp
->er_extbuf
[page_idx
];
2775 } else if (ifp
->if_bytes
) {
2776 return &ifp
->if_u1
.if_extents
[idx
];
2783 * Insert new item(s) into the extent records for incore inode
2784 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
2788 xfs_inode_t
*ip
, /* incore inode pointer */
2789 xfs_extnum_t idx
, /* starting index of new items */
2790 xfs_extnum_t count
, /* number of inserted items */
2791 xfs_bmbt_irec_t
*new, /* items to insert */
2792 int state
) /* type of extent conversion */
2794 xfs_ifork_t
*ifp
= (state
& BMAP_ATTRFORK
) ? ip
->i_afp
: &ip
->i_df
;
2795 xfs_extnum_t i
; /* extent record index */
2797 trace_xfs_iext_insert(ip
, idx
, new, state
, _RET_IP_
);
2799 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
2800 xfs_iext_add(ifp
, idx
, count
);
2801 for (i
= idx
; i
< idx
+ count
; i
++, new++)
2802 xfs_bmbt_set_all(xfs_iext_get_ext(ifp
, i
), new);
2806 * This is called when the amount of space required for incore file
2807 * extents needs to be increased. The ext_diff parameter stores the
2808 * number of new extents being added and the idx parameter contains
2809 * the extent index where the new extents will be added. If the new
2810 * extents are being appended, then we just need to (re)allocate and
2811 * initialize the space. Otherwise, if the new extents are being
2812 * inserted into the middle of the existing entries, a bit more work
2813 * is required to make room for the new extents to be inserted. The
2814 * caller is responsible for filling in the new extent entries upon
2819 xfs_ifork_t
*ifp
, /* inode fork pointer */
2820 xfs_extnum_t idx
, /* index to begin adding exts */
2821 int ext_diff
) /* number of extents to add */
2823 int byte_diff
; /* new bytes being added */
2824 int new_size
; /* size of extents after adding */
2825 xfs_extnum_t nextents
; /* number of extents in file */
2827 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
2828 ASSERT((idx
>= 0) && (idx
<= nextents
));
2829 byte_diff
= ext_diff
* sizeof(xfs_bmbt_rec_t
);
2830 new_size
= ifp
->if_bytes
+ byte_diff
;
2832 * If the new number of extents (nextents + ext_diff)
2833 * fits inside the inode, then continue to use the inline
2836 if (nextents
+ ext_diff
<= XFS_INLINE_EXTS
) {
2837 if (idx
< nextents
) {
2838 memmove(&ifp
->if_u2
.if_inline_ext
[idx
+ ext_diff
],
2839 &ifp
->if_u2
.if_inline_ext
[idx
],
2840 (nextents
- idx
) * sizeof(xfs_bmbt_rec_t
));
2841 memset(&ifp
->if_u2
.if_inline_ext
[idx
], 0, byte_diff
);
2843 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
2844 ifp
->if_real_bytes
= 0;
2847 * Otherwise use a linear (direct) extent list.
2848 * If the extents are currently inside the inode,
2849 * xfs_iext_realloc_direct will switch us from
2850 * inline to direct extent allocation mode.
2852 else if (nextents
+ ext_diff
<= XFS_LINEAR_EXTS
) {
2853 xfs_iext_realloc_direct(ifp
, new_size
);
2854 if (idx
< nextents
) {
2855 memmove(&ifp
->if_u1
.if_extents
[idx
+ ext_diff
],
2856 &ifp
->if_u1
.if_extents
[idx
],
2857 (nextents
- idx
) * sizeof(xfs_bmbt_rec_t
));
2858 memset(&ifp
->if_u1
.if_extents
[idx
], 0, byte_diff
);
2861 /* Indirection array */
2863 xfs_ext_irec_t
*erp
;
2867 ASSERT(nextents
+ ext_diff
> XFS_LINEAR_EXTS
);
2868 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
2869 erp
= xfs_iext_idx_to_irec(ifp
, &page_idx
, &erp_idx
, 1);
2871 xfs_iext_irec_init(ifp
);
2872 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
2873 erp
= ifp
->if_u1
.if_ext_irec
;
2875 /* Extents fit in target extent page */
2876 if (erp
&& erp
->er_extcount
+ ext_diff
<= XFS_LINEAR_EXTS
) {
2877 if (page_idx
< erp
->er_extcount
) {
2878 memmove(&erp
->er_extbuf
[page_idx
+ ext_diff
],
2879 &erp
->er_extbuf
[page_idx
],
2880 (erp
->er_extcount
- page_idx
) *
2881 sizeof(xfs_bmbt_rec_t
));
2882 memset(&erp
->er_extbuf
[page_idx
], 0, byte_diff
);
2884 erp
->er_extcount
+= ext_diff
;
2885 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, ext_diff
);
2887 /* Insert a new extent page */
2889 xfs_iext_add_indirect_multi(ifp
,
2890 erp_idx
, page_idx
, ext_diff
);
2893 * If extent(s) are being appended to the last page in
2894 * the indirection array and the new extent(s) don't fit
2895 * in the page, then erp is NULL and erp_idx is set to
2896 * the next index needed in the indirection array.
2899 int count
= ext_diff
;
2902 erp
= xfs_iext_irec_new(ifp
, erp_idx
);
2903 erp
->er_extcount
= count
;
2904 count
-= MIN(count
, (int)XFS_LINEAR_EXTS
);
2911 ifp
->if_bytes
= new_size
;
2915 * This is called when incore extents are being added to the indirection
2916 * array and the new extents do not fit in the target extent list. The
2917 * erp_idx parameter contains the irec index for the target extent list
2918 * in the indirection array, and the idx parameter contains the extent
2919 * index within the list. The number of extents being added is stored
2920 * in the count parameter.
2922 * |-------| |-------|
2923 * | | | | idx - number of extents before idx
2925 * | | | | count - number of extents being inserted at idx
2926 * |-------| |-------|
2927 * | count | | nex2 | nex2 - number of extents after idx + count
2928 * |-------| |-------|
2931 xfs_iext_add_indirect_multi(
2932 xfs_ifork_t
*ifp
, /* inode fork pointer */
2933 int erp_idx
, /* target extent irec index */
2934 xfs_extnum_t idx
, /* index within target list */
2935 int count
) /* new extents being added */
2937 int byte_diff
; /* new bytes being added */
2938 xfs_ext_irec_t
*erp
; /* pointer to irec entry */
2939 xfs_extnum_t ext_diff
; /* number of extents to add */
2940 xfs_extnum_t ext_cnt
; /* new extents still needed */
2941 xfs_extnum_t nex2
; /* extents after idx + count */
2942 xfs_bmbt_rec_t
*nex2_ep
= NULL
; /* temp list for nex2 extents */
2943 int nlists
; /* number of irec's (lists) */
2945 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
2946 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
2947 nex2
= erp
->er_extcount
- idx
;
2948 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
2951 * Save second part of target extent list
2952 * (all extents past */
2954 byte_diff
= nex2
* sizeof(xfs_bmbt_rec_t
);
2955 nex2_ep
= (xfs_bmbt_rec_t
*) kmem_alloc(byte_diff
, KM_NOFS
);
2956 memmove(nex2_ep
, &erp
->er_extbuf
[idx
], byte_diff
);
2957 erp
->er_extcount
-= nex2
;
2958 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, -nex2
);
2959 memset(&erp
->er_extbuf
[idx
], 0, byte_diff
);
2963 * Add the new extents to the end of the target
2964 * list, then allocate new irec record(s) and
2965 * extent buffer(s) as needed to store the rest
2966 * of the new extents.
2969 ext_diff
= MIN(ext_cnt
, (int)XFS_LINEAR_EXTS
- erp
->er_extcount
);
2971 erp
->er_extcount
+= ext_diff
;
2972 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, ext_diff
);
2973 ext_cnt
-= ext_diff
;
2977 erp
= xfs_iext_irec_new(ifp
, erp_idx
);
2978 ext_diff
= MIN(ext_cnt
, (int)XFS_LINEAR_EXTS
);
2979 erp
->er_extcount
= ext_diff
;
2980 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, ext_diff
);
2981 ext_cnt
-= ext_diff
;
2984 /* Add nex2 extents back to indirection array */
2986 xfs_extnum_t ext_avail
;
2989 byte_diff
= nex2
* sizeof(xfs_bmbt_rec_t
);
2990 ext_avail
= XFS_LINEAR_EXTS
- erp
->er_extcount
;
2993 * If nex2 extents fit in the current page, append
2994 * nex2_ep after the new extents.
2996 if (nex2
<= ext_avail
) {
2997 i
= erp
->er_extcount
;
3000 * Otherwise, check if space is available in the
3003 else if ((erp_idx
< nlists
- 1) &&
3004 (nex2
<= (ext_avail
= XFS_LINEAR_EXTS
-
3005 ifp
->if_u1
.if_ext_irec
[erp_idx
+1].er_extcount
))) {
3008 /* Create a hole for nex2 extents */
3009 memmove(&erp
->er_extbuf
[nex2
], erp
->er_extbuf
,
3010 erp
->er_extcount
* sizeof(xfs_bmbt_rec_t
));
3013 * Final choice, create a new extent page for
3018 erp
= xfs_iext_irec_new(ifp
, erp_idx
);
3020 memmove(&erp
->er_extbuf
[i
], nex2_ep
, byte_diff
);
3022 erp
->er_extcount
+= nex2
;
3023 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, nex2
);
3028 * This is called when the amount of space required for incore file
3029 * extents needs to be decreased. The ext_diff parameter stores the
3030 * number of extents to be removed and the idx parameter contains
3031 * the extent index where the extents will be removed from.
3033 * If the amount of space needed has decreased below the linear
3034 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3035 * extent array. Otherwise, use kmem_realloc() to adjust the
3036 * size to what is needed.
3040 xfs_inode_t
*ip
, /* incore inode pointer */
3041 xfs_extnum_t idx
, /* index to begin removing exts */
3042 int ext_diff
, /* number of extents to remove */
3043 int state
) /* type of extent conversion */
3045 xfs_ifork_t
*ifp
= (state
& BMAP_ATTRFORK
) ? ip
->i_afp
: &ip
->i_df
;
3046 xfs_extnum_t nextents
; /* number of extents in file */
3047 int new_size
; /* size of extents after removal */
3049 trace_xfs_iext_remove(ip
, idx
, state
, _RET_IP_
);
3051 ASSERT(ext_diff
> 0);
3052 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3053 new_size
= (nextents
- ext_diff
) * sizeof(xfs_bmbt_rec_t
);
3055 if (new_size
== 0) {
3056 xfs_iext_destroy(ifp
);
3057 } else if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3058 xfs_iext_remove_indirect(ifp
, idx
, ext_diff
);
3059 } else if (ifp
->if_real_bytes
) {
3060 xfs_iext_remove_direct(ifp
, idx
, ext_diff
);
3062 xfs_iext_remove_inline(ifp
, idx
, ext_diff
);
3064 ifp
->if_bytes
= new_size
;
3068 * This removes ext_diff extents from the inline buffer, beginning
3069 * at extent index idx.
3072 xfs_iext_remove_inline(
3073 xfs_ifork_t
*ifp
, /* inode fork pointer */
3074 xfs_extnum_t idx
, /* index to begin removing exts */
3075 int ext_diff
) /* number of extents to remove */
3077 int nextents
; /* number of extents in file */
3079 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
));
3080 ASSERT(idx
< XFS_INLINE_EXTS
);
3081 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3082 ASSERT(((nextents
- ext_diff
) > 0) &&
3083 (nextents
- ext_diff
) < XFS_INLINE_EXTS
);
3085 if (idx
+ ext_diff
< nextents
) {
3086 memmove(&ifp
->if_u2
.if_inline_ext
[idx
],
3087 &ifp
->if_u2
.if_inline_ext
[idx
+ ext_diff
],
3088 (nextents
- (idx
+ ext_diff
)) *
3089 sizeof(xfs_bmbt_rec_t
));
3090 memset(&ifp
->if_u2
.if_inline_ext
[nextents
- ext_diff
],
3091 0, ext_diff
* sizeof(xfs_bmbt_rec_t
));
3093 memset(&ifp
->if_u2
.if_inline_ext
[idx
], 0,
3094 ext_diff
* sizeof(xfs_bmbt_rec_t
));
3099 * This removes ext_diff extents from a linear (direct) extent list,
3100 * beginning at extent index idx. If the extents are being removed
3101 * from the end of the list (ie. truncate) then we just need to re-
3102 * allocate the list to remove the extra space. Otherwise, if the
3103 * extents are being removed from the middle of the existing extent
3104 * entries, then we first need to move the extent records beginning
3105 * at idx + ext_diff up in the list to overwrite the records being
3106 * removed, then remove the extra space via kmem_realloc.
3109 xfs_iext_remove_direct(
3110 xfs_ifork_t
*ifp
, /* inode fork pointer */
3111 xfs_extnum_t idx
, /* index to begin removing exts */
3112 int ext_diff
) /* number of extents to remove */
3114 xfs_extnum_t nextents
; /* number of extents in file */
3115 int new_size
; /* size of extents after removal */
3117 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
));
3118 new_size
= ifp
->if_bytes
-
3119 (ext_diff
* sizeof(xfs_bmbt_rec_t
));
3120 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3122 if (new_size
== 0) {
3123 xfs_iext_destroy(ifp
);
3126 /* Move extents up in the list (if needed) */
3127 if (idx
+ ext_diff
< nextents
) {
3128 memmove(&ifp
->if_u1
.if_extents
[idx
],
3129 &ifp
->if_u1
.if_extents
[idx
+ ext_diff
],
3130 (nextents
- (idx
+ ext_diff
)) *
3131 sizeof(xfs_bmbt_rec_t
));
3133 memset(&ifp
->if_u1
.if_extents
[nextents
- ext_diff
],
3134 0, ext_diff
* sizeof(xfs_bmbt_rec_t
));
3136 * Reallocate the direct extent list. If the extents
3137 * will fit inside the inode then xfs_iext_realloc_direct
3138 * will switch from direct to inline extent allocation
3141 xfs_iext_realloc_direct(ifp
, new_size
);
3142 ifp
->if_bytes
= new_size
;
3146 * This is called when incore extents are being removed from the
3147 * indirection array and the extents being removed span multiple extent
3148 * buffers. The idx parameter contains the file extent index where we
3149 * want to begin removing extents, and the count parameter contains
3150 * how many extents need to be removed.
3152 * |-------| |-------|
3153 * | nex1 | | | nex1 - number of extents before idx
3154 * |-------| | count |
3155 * | | | | count - number of extents being removed at idx
3156 * | count | |-------|
3157 * | | | nex2 | nex2 - number of extents after idx + count
3158 * |-------| |-------|
3161 xfs_iext_remove_indirect(
3162 xfs_ifork_t
*ifp
, /* inode fork pointer */
3163 xfs_extnum_t idx
, /* index to begin removing extents */
3164 int count
) /* number of extents to remove */
3166 xfs_ext_irec_t
*erp
; /* indirection array pointer */
3167 int erp_idx
= 0; /* indirection array index */
3168 xfs_extnum_t ext_cnt
; /* extents left to remove */
3169 xfs_extnum_t ext_diff
; /* extents to remove in current list */
3170 xfs_extnum_t nex1
; /* number of extents before idx */
3171 xfs_extnum_t nex2
; /* extents after idx + count */
3172 int page_idx
= idx
; /* index in target extent list */
3174 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3175 erp
= xfs_iext_idx_to_irec(ifp
, &page_idx
, &erp_idx
, 0);
3176 ASSERT(erp
!= NULL
);
3180 nex2
= MAX((erp
->er_extcount
- (nex1
+ ext_cnt
)), 0);
3181 ext_diff
= MIN(ext_cnt
, (erp
->er_extcount
- nex1
));
3183 * Check for deletion of entire list;
3184 * xfs_iext_irec_remove() updates extent offsets.
3186 if (ext_diff
== erp
->er_extcount
) {
3187 xfs_iext_irec_remove(ifp
, erp_idx
);
3188 ext_cnt
-= ext_diff
;
3191 ASSERT(erp_idx
< ifp
->if_real_bytes
/
3193 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
3200 /* Move extents up (if needed) */
3202 memmove(&erp
->er_extbuf
[nex1
],
3203 &erp
->er_extbuf
[nex1
+ ext_diff
],
3204 nex2
* sizeof(xfs_bmbt_rec_t
));
3206 /* Zero out rest of page */
3207 memset(&erp
->er_extbuf
[nex1
+ nex2
], 0, (XFS_IEXT_BUFSZ
-
3208 ((nex1
+ nex2
) * sizeof(xfs_bmbt_rec_t
))));
3209 /* Update remaining counters */
3210 erp
->er_extcount
-= ext_diff
;
3211 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, -ext_diff
);
3212 ext_cnt
-= ext_diff
;
3217 ifp
->if_bytes
-= count
* sizeof(xfs_bmbt_rec_t
);
3218 xfs_iext_irec_compact(ifp
);
3222 * Create, destroy, or resize a linear (direct) block of extents.
3225 xfs_iext_realloc_direct(
3226 xfs_ifork_t
*ifp
, /* inode fork pointer */
3227 int new_size
) /* new size of extents */
3229 int rnew_size
; /* real new size of extents */
3231 rnew_size
= new_size
;
3233 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
) ||
3234 ((new_size
>= 0) && (new_size
<= XFS_IEXT_BUFSZ
) &&
3235 (new_size
!= ifp
->if_real_bytes
)));
3237 /* Free extent records */
3238 if (new_size
== 0) {
3239 xfs_iext_destroy(ifp
);
3241 /* Resize direct extent list and zero any new bytes */
3242 else if (ifp
->if_real_bytes
) {
3243 /* Check if extents will fit inside the inode */
3244 if (new_size
<= XFS_INLINE_EXTS
* sizeof(xfs_bmbt_rec_t
)) {
3245 xfs_iext_direct_to_inline(ifp
, new_size
/
3246 (uint
)sizeof(xfs_bmbt_rec_t
));
3247 ifp
->if_bytes
= new_size
;
3250 if (!is_power_of_2(new_size
)){
3251 rnew_size
= roundup_pow_of_two(new_size
);
3253 if (rnew_size
!= ifp
->if_real_bytes
) {
3254 ifp
->if_u1
.if_extents
=
3255 kmem_realloc(ifp
->if_u1
.if_extents
,
3257 ifp
->if_real_bytes
, KM_NOFS
);
3259 if (rnew_size
> ifp
->if_real_bytes
) {
3260 memset(&ifp
->if_u1
.if_extents
[ifp
->if_bytes
/
3261 (uint
)sizeof(xfs_bmbt_rec_t
)], 0,
3262 rnew_size
- ifp
->if_real_bytes
);
3266 * Switch from the inline extent buffer to a direct
3267 * extent list. Be sure to include the inline extent
3268 * bytes in new_size.
3271 new_size
+= ifp
->if_bytes
;
3272 if (!is_power_of_2(new_size
)) {
3273 rnew_size
= roundup_pow_of_two(new_size
);
3275 xfs_iext_inline_to_direct(ifp
, rnew_size
);
3277 ifp
->if_real_bytes
= rnew_size
;
3278 ifp
->if_bytes
= new_size
;
3282 * Switch from linear (direct) extent records to inline buffer.
3285 xfs_iext_direct_to_inline(
3286 xfs_ifork_t
*ifp
, /* inode fork pointer */
3287 xfs_extnum_t nextents
) /* number of extents in file */
3289 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
3290 ASSERT(nextents
<= XFS_INLINE_EXTS
);
3292 * The inline buffer was zeroed when we switched
3293 * from inline to direct extent allocation mode,
3294 * so we don't need to clear it here.
3296 memcpy(ifp
->if_u2
.if_inline_ext
, ifp
->if_u1
.if_extents
,
3297 nextents
* sizeof(xfs_bmbt_rec_t
));
3298 kmem_free(ifp
->if_u1
.if_extents
);
3299 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
3300 ifp
->if_real_bytes
= 0;
3304 * Switch from inline buffer to linear (direct) extent records.
3305 * new_size should already be rounded up to the next power of 2
3306 * by the caller (when appropriate), so use new_size as it is.
3307 * However, since new_size may be rounded up, we can't update
3308 * if_bytes here. It is the caller's responsibility to update
3309 * if_bytes upon return.
3312 xfs_iext_inline_to_direct(
3313 xfs_ifork_t
*ifp
, /* inode fork pointer */
3314 int new_size
) /* number of extents in file */
3316 ifp
->if_u1
.if_extents
= kmem_alloc(new_size
, KM_NOFS
);
3317 memset(ifp
->if_u1
.if_extents
, 0, new_size
);
3318 if (ifp
->if_bytes
) {
3319 memcpy(ifp
->if_u1
.if_extents
, ifp
->if_u2
.if_inline_ext
,
3321 memset(ifp
->if_u2
.if_inline_ext
, 0, XFS_INLINE_EXTS
*
3322 sizeof(xfs_bmbt_rec_t
));
3324 ifp
->if_real_bytes
= new_size
;
3328 * Resize an extent indirection array to new_size bytes.
3331 xfs_iext_realloc_indirect(
3332 xfs_ifork_t
*ifp
, /* inode fork pointer */
3333 int new_size
) /* new indirection array size */
3335 int nlists
; /* number of irec's (ex lists) */
3336 int size
; /* current indirection array size */
3338 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3339 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3340 size
= nlists
* sizeof(xfs_ext_irec_t
);
3341 ASSERT(ifp
->if_real_bytes
);
3342 ASSERT((new_size
>= 0) && (new_size
!= size
));
3343 if (new_size
== 0) {
3344 xfs_iext_destroy(ifp
);
3346 ifp
->if_u1
.if_ext_irec
= (xfs_ext_irec_t
*)
3347 kmem_realloc(ifp
->if_u1
.if_ext_irec
,
3348 new_size
, size
, KM_NOFS
);
3353 * Switch from indirection array to linear (direct) extent allocations.
3356 xfs_iext_indirect_to_direct(
3357 xfs_ifork_t
*ifp
) /* inode fork pointer */
3359 xfs_bmbt_rec_host_t
*ep
; /* extent record pointer */
3360 xfs_extnum_t nextents
; /* number of extents in file */
3361 int size
; /* size of file extents */
3363 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3364 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3365 ASSERT(nextents
<= XFS_LINEAR_EXTS
);
3366 size
= nextents
* sizeof(xfs_bmbt_rec_t
);
3368 xfs_iext_irec_compact_pages(ifp
);
3369 ASSERT(ifp
->if_real_bytes
== XFS_IEXT_BUFSZ
);
3371 ep
= ifp
->if_u1
.if_ext_irec
->er_extbuf
;
3372 kmem_free(ifp
->if_u1
.if_ext_irec
);
3373 ifp
->if_flags
&= ~XFS_IFEXTIREC
;
3374 ifp
->if_u1
.if_extents
= ep
;
3375 ifp
->if_bytes
= size
;
3376 if (nextents
< XFS_LINEAR_EXTS
) {
3377 xfs_iext_realloc_direct(ifp
, size
);
3382 * Free incore file extents.
3386 xfs_ifork_t
*ifp
) /* inode fork pointer */
3388 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3392 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3393 for (erp_idx
= nlists
- 1; erp_idx
>= 0 ; erp_idx
--) {
3394 xfs_iext_irec_remove(ifp
, erp_idx
);
3396 ifp
->if_flags
&= ~XFS_IFEXTIREC
;
3397 } else if (ifp
->if_real_bytes
) {
3398 kmem_free(ifp
->if_u1
.if_extents
);
3399 } else if (ifp
->if_bytes
) {
3400 memset(ifp
->if_u2
.if_inline_ext
, 0, XFS_INLINE_EXTS
*
3401 sizeof(xfs_bmbt_rec_t
));
3403 ifp
->if_u1
.if_extents
= NULL
;
3404 ifp
->if_real_bytes
= 0;
3409 * Return a pointer to the extent record for file system block bno.
3411 xfs_bmbt_rec_host_t
* /* pointer to found extent record */
3412 xfs_iext_bno_to_ext(
3413 xfs_ifork_t
*ifp
, /* inode fork pointer */
3414 xfs_fileoff_t bno
, /* block number to search for */
3415 xfs_extnum_t
*idxp
) /* index of target extent */
3417 xfs_bmbt_rec_host_t
*base
; /* pointer to first extent */
3418 xfs_filblks_t blockcount
= 0; /* number of blocks in extent */
3419 xfs_bmbt_rec_host_t
*ep
= NULL
; /* pointer to target extent */
3420 xfs_ext_irec_t
*erp
= NULL
; /* indirection array pointer */
3421 int high
; /* upper boundary in search */
3422 xfs_extnum_t idx
= 0; /* index of target extent */
3423 int low
; /* lower boundary in search */
3424 xfs_extnum_t nextents
; /* number of file extents */
3425 xfs_fileoff_t startoff
= 0; /* start offset of extent */
3427 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3428 if (nextents
== 0) {
3433 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3434 /* Find target extent list */
3436 erp
= xfs_iext_bno_to_irec(ifp
, bno
, &erp_idx
);
3437 base
= erp
->er_extbuf
;
3438 high
= erp
->er_extcount
- 1;
3440 base
= ifp
->if_u1
.if_extents
;
3441 high
= nextents
- 1;
3443 /* Binary search extent records */
3444 while (low
<= high
) {
3445 idx
= (low
+ high
) >> 1;
3447 startoff
= xfs_bmbt_get_startoff(ep
);
3448 blockcount
= xfs_bmbt_get_blockcount(ep
);
3449 if (bno
< startoff
) {
3451 } else if (bno
>= startoff
+ blockcount
) {
3454 /* Convert back to file-based extent index */
3455 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3456 idx
+= erp
->er_extoff
;
3462 /* Convert back to file-based extent index */
3463 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3464 idx
+= erp
->er_extoff
;
3466 if (bno
>= startoff
+ blockcount
) {
3467 if (++idx
== nextents
) {
3470 ep
= xfs_iext_get_ext(ifp
, idx
);
3478 * Return a pointer to the indirection array entry containing the
3479 * extent record for filesystem block bno. Store the index of the
3480 * target irec in *erp_idxp.
3482 xfs_ext_irec_t
* /* pointer to found extent record */
3483 xfs_iext_bno_to_irec(
3484 xfs_ifork_t
*ifp
, /* inode fork pointer */
3485 xfs_fileoff_t bno
, /* block number to search for */
3486 int *erp_idxp
) /* irec index of target ext list */
3488 xfs_ext_irec_t
*erp
= NULL
; /* indirection array pointer */
3489 xfs_ext_irec_t
*erp_next
; /* next indirection array entry */
3490 int erp_idx
; /* indirection array index */
3491 int nlists
; /* number of extent irec's (lists) */
3492 int high
; /* binary search upper limit */
3493 int low
; /* binary search lower limit */
3495 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3496 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3500 while (low
<= high
) {
3501 erp_idx
= (low
+ high
) >> 1;
3502 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
3503 erp_next
= erp_idx
< nlists
- 1 ? erp
+ 1 : NULL
;
3504 if (bno
< xfs_bmbt_get_startoff(erp
->er_extbuf
)) {
3506 } else if (erp_next
&& bno
>=
3507 xfs_bmbt_get_startoff(erp_next
->er_extbuf
)) {
3513 *erp_idxp
= erp_idx
;
3518 * Return a pointer to the indirection array entry containing the
3519 * extent record at file extent index *idxp. Store the index of the
3520 * target irec in *erp_idxp and store the page index of the target
3521 * extent record in *idxp.
3524 xfs_iext_idx_to_irec(
3525 xfs_ifork_t
*ifp
, /* inode fork pointer */
3526 xfs_extnum_t
*idxp
, /* extent index (file -> page) */
3527 int *erp_idxp
, /* pointer to target irec */
3528 int realloc
) /* new bytes were just added */
3530 xfs_ext_irec_t
*prev
; /* pointer to previous irec */
3531 xfs_ext_irec_t
*erp
= NULL
; /* pointer to current irec */
3532 int erp_idx
; /* indirection array index */
3533 int nlists
; /* number of irec's (ex lists) */
3534 int high
; /* binary search upper limit */
3535 int low
; /* binary search lower limit */
3536 xfs_extnum_t page_idx
= *idxp
; /* extent index in target list */
3538 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3539 ASSERT(page_idx
>= 0);
3540 ASSERT(page_idx
<= ifp
->if_bytes
/ sizeof(xfs_bmbt_rec_t
));
3541 ASSERT(page_idx
< ifp
->if_bytes
/ sizeof(xfs_bmbt_rec_t
) || realloc
);
3543 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3548 /* Binary search extent irec's */
3549 while (low
<= high
) {
3550 erp_idx
= (low
+ high
) >> 1;
3551 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
3552 prev
= erp_idx
> 0 ? erp
- 1 : NULL
;
3553 if (page_idx
< erp
->er_extoff
|| (page_idx
== erp
->er_extoff
&&
3554 realloc
&& prev
&& prev
->er_extcount
< XFS_LINEAR_EXTS
)) {
3556 } else if (page_idx
> erp
->er_extoff
+ erp
->er_extcount
||
3557 (page_idx
== erp
->er_extoff
+ erp
->er_extcount
&&
3560 } else if (page_idx
== erp
->er_extoff
+ erp
->er_extcount
&&
3561 erp
->er_extcount
== XFS_LINEAR_EXTS
) {
3565 erp
= erp_idx
< nlists
? erp
+ 1 : NULL
;
3568 page_idx
-= erp
->er_extoff
;
3573 *erp_idxp
= erp_idx
;
3578 * Allocate and initialize an indirection array once the space needed
3579 * for incore extents increases above XFS_IEXT_BUFSZ.
3583 xfs_ifork_t
*ifp
) /* inode fork pointer */
3585 xfs_ext_irec_t
*erp
; /* indirection array pointer */
3586 xfs_extnum_t nextents
; /* number of extents in file */
3588 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
));
3589 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3590 ASSERT(nextents
<= XFS_LINEAR_EXTS
);
3592 erp
= kmem_alloc(sizeof(xfs_ext_irec_t
), KM_NOFS
);
3594 if (nextents
== 0) {
3595 ifp
->if_u1
.if_extents
= kmem_alloc(XFS_IEXT_BUFSZ
, KM_NOFS
);
3596 } else if (!ifp
->if_real_bytes
) {
3597 xfs_iext_inline_to_direct(ifp
, XFS_IEXT_BUFSZ
);
3598 } else if (ifp
->if_real_bytes
< XFS_IEXT_BUFSZ
) {
3599 xfs_iext_realloc_direct(ifp
, XFS_IEXT_BUFSZ
);
3601 erp
->er_extbuf
= ifp
->if_u1
.if_extents
;
3602 erp
->er_extcount
= nextents
;
3605 ifp
->if_flags
|= XFS_IFEXTIREC
;
3606 ifp
->if_real_bytes
= XFS_IEXT_BUFSZ
;
3607 ifp
->if_bytes
= nextents
* sizeof(xfs_bmbt_rec_t
);
3608 ifp
->if_u1
.if_ext_irec
= erp
;
3614 * Allocate and initialize a new entry in the indirection array.
3618 xfs_ifork_t
*ifp
, /* inode fork pointer */
3619 int erp_idx
) /* index for new irec */
3621 xfs_ext_irec_t
*erp
; /* indirection array pointer */
3622 int i
; /* loop counter */
3623 int nlists
; /* number of irec's (ex lists) */
3625 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3626 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3628 /* Resize indirection array */
3629 xfs_iext_realloc_indirect(ifp
, ++nlists
*
3630 sizeof(xfs_ext_irec_t
));
3632 * Move records down in the array so the
3633 * new page can use erp_idx.
3635 erp
= ifp
->if_u1
.if_ext_irec
;
3636 for (i
= nlists
- 1; i
> erp_idx
; i
--) {
3637 memmove(&erp
[i
], &erp
[i
-1], sizeof(xfs_ext_irec_t
));
3639 ASSERT(i
== erp_idx
);
3641 /* Initialize new extent record */
3642 erp
= ifp
->if_u1
.if_ext_irec
;
3643 erp
[erp_idx
].er_extbuf
= kmem_alloc(XFS_IEXT_BUFSZ
, KM_NOFS
);
3644 ifp
->if_real_bytes
= nlists
* XFS_IEXT_BUFSZ
;
3645 memset(erp
[erp_idx
].er_extbuf
, 0, XFS_IEXT_BUFSZ
);
3646 erp
[erp_idx
].er_extcount
= 0;
3647 erp
[erp_idx
].er_extoff
= erp_idx
> 0 ?
3648 erp
[erp_idx
-1].er_extoff
+ erp
[erp_idx
-1].er_extcount
: 0;
3649 return (&erp
[erp_idx
]);
3653 * Remove a record from the indirection array.
3656 xfs_iext_irec_remove(
3657 xfs_ifork_t
*ifp
, /* inode fork pointer */
3658 int erp_idx
) /* irec index to remove */
3660 xfs_ext_irec_t
*erp
; /* indirection array pointer */
3661 int i
; /* loop counter */
3662 int nlists
; /* number of irec's (ex lists) */
3664 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3665 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3666 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
3667 if (erp
->er_extbuf
) {
3668 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1,
3670 kmem_free(erp
->er_extbuf
);
3672 /* Compact extent records */
3673 erp
= ifp
->if_u1
.if_ext_irec
;
3674 for (i
= erp_idx
; i
< nlists
- 1; i
++) {
3675 memmove(&erp
[i
], &erp
[i
+1], sizeof(xfs_ext_irec_t
));
3678 * Manually free the last extent record from the indirection
3679 * array. A call to xfs_iext_realloc_indirect() with a size
3680 * of zero would result in a call to xfs_iext_destroy() which
3681 * would in turn call this function again, creating a nasty
3685 xfs_iext_realloc_indirect(ifp
,
3686 nlists
* sizeof(xfs_ext_irec_t
));
3688 kmem_free(ifp
->if_u1
.if_ext_irec
);
3690 ifp
->if_real_bytes
= nlists
* XFS_IEXT_BUFSZ
;
3694 * This is called to clean up large amounts of unused memory allocated
3695 * by the indirection array. Before compacting anything though, verify
3696 * that the indirection array is still needed and switch back to the
3697 * linear extent list (or even the inline buffer) if possible. The
3698 * compaction policy is as follows:
3700 * Full Compaction: Extents fit into a single page (or inline buffer)
3701 * Partial Compaction: Extents occupy less than 50% of allocated space
3702 * No Compaction: Extents occupy at least 50% of allocated space
3705 xfs_iext_irec_compact(
3706 xfs_ifork_t
*ifp
) /* inode fork pointer */
3708 xfs_extnum_t nextents
; /* number of extents in file */
3709 int nlists
; /* number of irec's (ex lists) */
3711 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3712 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3713 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3715 if (nextents
== 0) {
3716 xfs_iext_destroy(ifp
);
3717 } else if (nextents
<= XFS_INLINE_EXTS
) {
3718 xfs_iext_indirect_to_direct(ifp
);
3719 xfs_iext_direct_to_inline(ifp
, nextents
);
3720 } else if (nextents
<= XFS_LINEAR_EXTS
) {
3721 xfs_iext_indirect_to_direct(ifp
);
3722 } else if (nextents
< (nlists
* XFS_LINEAR_EXTS
) >> 1) {
3723 xfs_iext_irec_compact_pages(ifp
);
3728 * Combine extents from neighboring extent pages.
3731 xfs_iext_irec_compact_pages(
3732 xfs_ifork_t
*ifp
) /* inode fork pointer */
3734 xfs_ext_irec_t
*erp
, *erp_next
;/* pointers to irec entries */
3735 int erp_idx
= 0; /* indirection array index */
3736 int nlists
; /* number of irec's (ex lists) */
3738 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3739 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3740 while (erp_idx
< nlists
- 1) {
3741 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
3743 if (erp_next
->er_extcount
<=
3744 (XFS_LINEAR_EXTS
- erp
->er_extcount
)) {
3745 memcpy(&erp
->er_extbuf
[erp
->er_extcount
],
3746 erp_next
->er_extbuf
, erp_next
->er_extcount
*
3747 sizeof(xfs_bmbt_rec_t
));
3748 erp
->er_extcount
+= erp_next
->er_extcount
;
3750 * Free page before removing extent record
3751 * so er_extoffs don't get modified in
3752 * xfs_iext_irec_remove.
3754 kmem_free(erp_next
->er_extbuf
);
3755 erp_next
->er_extbuf
= NULL
;
3756 xfs_iext_irec_remove(ifp
, erp_idx
+ 1);
3757 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3765 * This is called to update the er_extoff field in the indirection
3766 * array when extents have been added or removed from one of the
3767 * extent lists. erp_idx contains the irec index to begin updating
3768 * at and ext_diff contains the number of extents that were added
3772 xfs_iext_irec_update_extoffs(
3773 xfs_ifork_t
*ifp
, /* inode fork pointer */
3774 int erp_idx
, /* irec index to update */
3775 int ext_diff
) /* number of new extents */
3777 int i
; /* loop counter */
3778 int nlists
; /* number of irec's (ex lists */
3780 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3781 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3782 for (i
= erp_idx
; i
< nlists
; i
++) {
3783 ifp
->if_u1
.if_ext_irec
[i
].er_extoff
+= ext_diff
;