2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 #include <linux/log2.h>
22 #include "xfs_types.h"
25 #include "xfs_trans.h"
26 #include "xfs_trans_priv.h"
29 #include "xfs_mount.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_alloc_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_attr_sf.h"
34 #include "xfs_dinode.h"
35 #include "xfs_inode.h"
36 #include "xfs_buf_item.h"
37 #include "xfs_inode_item.h"
38 #include "xfs_btree.h"
39 #include "xfs_alloc.h"
40 #include "xfs_ialloc.h"
42 #include "xfs_error.h"
43 #include "xfs_utils.h"
44 #include "xfs_quota.h"
45 #include "xfs_filestream.h"
46 #include "xfs_vnodeops.h"
47 #include "xfs_trace.h"
49 kmem_zone_t
*xfs_ifork_zone
;
50 kmem_zone_t
*xfs_inode_zone
;
53 * Used in xfs_itruncate_extents(). This is the maximum number of extents
54 * freed from a file in a single transaction.
56 #define XFS_ITRUNC_MAX_EXTENTS 2
58 STATIC
int xfs_iflush_int(xfs_inode_t
*, xfs_buf_t
*);
59 STATIC
int xfs_iformat_local(xfs_inode_t
*, xfs_dinode_t
*, int, int);
60 STATIC
int xfs_iformat_extents(xfs_inode_t
*, xfs_dinode_t
*, int);
61 STATIC
int xfs_iformat_btree(xfs_inode_t
*, xfs_dinode_t
*, int);
64 * helper function to extract extent size hint from inode
70 if ((ip
->i_d
.di_flags
& XFS_DIFLAG_EXTSIZE
) && ip
->i_d
.di_extsize
)
71 return ip
->i_d
.di_extsize
;
72 if (XFS_IS_REALTIME_INODE(ip
))
73 return ip
->i_mount
->m_sb
.sb_rextsize
;
79 * Make sure that the extents in the given memory buffer
89 xfs_bmbt_rec_host_t rec
;
92 for (i
= 0; i
< nrecs
; i
++) {
93 xfs_bmbt_rec_host_t
*ep
= xfs_iext_get_ext(ifp
, i
);
94 rec
.l0
= get_unaligned(&ep
->l0
);
95 rec
.l1
= get_unaligned(&ep
->l1
);
96 xfs_bmbt_get_all(&rec
, &irec
);
97 if (fmt
== XFS_EXTFMT_NOSTATE
)
98 ASSERT(irec
.br_state
== XFS_EXT_NORM
);
102 #define xfs_validate_extents(ifp, nrecs, fmt)
106 * Check that none of the inode's in the buffer have a next
107 * unlinked field of 0.
119 j
= mp
->m_inode_cluster_size
>> mp
->m_sb
.sb_inodelog
;
121 for (i
= 0; i
< j
; i
++) {
122 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
,
123 i
* mp
->m_sb
.sb_inodesize
);
124 if (!dip
->di_next_unlinked
) {
126 "Detected bogus zero next_unlinked field in incore inode buffer 0x%p.",
128 ASSERT(dip
->di_next_unlinked
);
135 * This routine is called to map an inode to the buffer containing the on-disk
136 * version of the inode. It returns a pointer to the buffer containing the
137 * on-disk inode in the bpp parameter, and in the dipp parameter it returns a
138 * pointer to the on-disk inode within that buffer.
140 * If a non-zero error is returned, then the contents of bpp and dipp are
145 struct xfs_mount
*mp
,
146 struct xfs_trans
*tp
,
147 struct xfs_imap
*imap
,
148 struct xfs_dinode
**dipp
,
149 struct xfs_buf
**bpp
,
158 buf_flags
|= XBF_UNMAPPED
;
159 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, imap
->im_blkno
,
160 (int)imap
->im_len
, buf_flags
, &bp
);
162 if (error
!= EAGAIN
) {
164 "%s: xfs_trans_read_buf() returned error %d.",
167 ASSERT(buf_flags
& XBF_TRYLOCK
);
173 * Validate the magic number and version of every inode in the buffer
174 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
177 ni
= BBTOB(imap
->im_len
) >> mp
->m_sb
.sb_inodelog
;
178 #else /* usual case */
182 for (i
= 0; i
< ni
; i
++) {
186 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
,
187 (i
<< mp
->m_sb
.sb_inodelog
));
188 di_ok
= dip
->di_magic
== cpu_to_be16(XFS_DINODE_MAGIC
) &&
189 XFS_DINODE_GOOD_VERSION(dip
->di_version
);
190 if (unlikely(XFS_TEST_ERROR(!di_ok
, mp
,
191 XFS_ERRTAG_ITOBP_INOTOBP
,
192 XFS_RANDOM_ITOBP_INOTOBP
))) {
193 if (iget_flags
& XFS_IGET_UNTRUSTED
) {
194 xfs_trans_brelse(tp
, bp
);
195 return XFS_ERROR(EINVAL
);
197 XFS_CORRUPTION_ERROR(__func__
, XFS_ERRLEVEL_HIGH
,
201 "bad inode magic/vsn daddr %lld #%d (magic=%x)",
202 (unsigned long long)imap
->im_blkno
, i
,
203 be16_to_cpu(dip
->di_magic
));
206 xfs_trans_brelse(tp
, bp
);
207 return XFS_ERROR(EFSCORRUPTED
);
211 xfs_inobp_check(mp
, bp
);
214 *dipp
= (struct xfs_dinode
*)xfs_buf_offset(bp
, imap
->im_boffset
);
219 * Move inode type and inode format specific information from the
220 * on-disk inode to the in-core inode. For fifos, devs, and sockets
221 * this means set if_rdev to the proper value. For files, directories,
222 * and symlinks this means to bring in the in-line data or extent
223 * pointers. For a file in B-tree format, only the root is immediately
224 * brought in-core. The rest will be in-lined in if_extents when it
225 * is first referenced (see xfs_iread_extents()).
232 xfs_attr_shortform_t
*atp
;
237 if (unlikely(be32_to_cpu(dip
->di_nextents
) +
238 be16_to_cpu(dip
->di_anextents
) >
239 be64_to_cpu(dip
->di_nblocks
))) {
240 xfs_warn(ip
->i_mount
,
241 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
242 (unsigned long long)ip
->i_ino
,
243 (int)(be32_to_cpu(dip
->di_nextents
) +
244 be16_to_cpu(dip
->di_anextents
)),
246 be64_to_cpu(dip
->di_nblocks
));
247 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW
,
249 return XFS_ERROR(EFSCORRUPTED
);
252 if (unlikely(dip
->di_forkoff
> ip
->i_mount
->m_sb
.sb_inodesize
)) {
253 xfs_warn(ip
->i_mount
, "corrupt dinode %Lu, forkoff = 0x%x.",
254 (unsigned long long)ip
->i_ino
,
256 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW
,
258 return XFS_ERROR(EFSCORRUPTED
);
261 if (unlikely((ip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
) &&
262 !ip
->i_mount
->m_rtdev_targp
)) {
263 xfs_warn(ip
->i_mount
,
264 "corrupt dinode %Lu, has realtime flag set.",
266 XFS_CORRUPTION_ERROR("xfs_iformat(realtime)",
267 XFS_ERRLEVEL_LOW
, ip
->i_mount
, dip
);
268 return XFS_ERROR(EFSCORRUPTED
);
271 switch (ip
->i_d
.di_mode
& S_IFMT
) {
276 if (unlikely(dip
->di_format
!= XFS_DINODE_FMT_DEV
)) {
277 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW
,
279 return XFS_ERROR(EFSCORRUPTED
);
282 ip
->i_df
.if_u2
.if_rdev
= xfs_dinode_get_rdev(dip
);
288 switch (dip
->di_format
) {
289 case XFS_DINODE_FMT_LOCAL
:
291 * no local regular files yet
293 if (unlikely(S_ISREG(be16_to_cpu(dip
->di_mode
)))) {
294 xfs_warn(ip
->i_mount
,
295 "corrupt inode %Lu (local format for regular file).",
296 (unsigned long long) ip
->i_ino
);
297 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
300 return XFS_ERROR(EFSCORRUPTED
);
303 di_size
= be64_to_cpu(dip
->di_size
);
304 if (unlikely(di_size
> XFS_DFORK_DSIZE(dip
, ip
->i_mount
))) {
305 xfs_warn(ip
->i_mount
,
306 "corrupt inode %Lu (bad size %Ld for local inode).",
307 (unsigned long long) ip
->i_ino
,
308 (long long) di_size
);
309 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
312 return XFS_ERROR(EFSCORRUPTED
);
316 error
= xfs_iformat_local(ip
, dip
, XFS_DATA_FORK
, size
);
318 case XFS_DINODE_FMT_EXTENTS
:
319 error
= xfs_iformat_extents(ip
, dip
, XFS_DATA_FORK
);
321 case XFS_DINODE_FMT_BTREE
:
322 error
= xfs_iformat_btree(ip
, dip
, XFS_DATA_FORK
);
325 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW
,
327 return XFS_ERROR(EFSCORRUPTED
);
332 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW
, ip
->i_mount
);
333 return XFS_ERROR(EFSCORRUPTED
);
338 if (!XFS_DFORK_Q(dip
))
341 ASSERT(ip
->i_afp
== NULL
);
342 ip
->i_afp
= kmem_zone_zalloc(xfs_ifork_zone
, KM_SLEEP
| KM_NOFS
);
344 switch (dip
->di_aformat
) {
345 case XFS_DINODE_FMT_LOCAL
:
346 atp
= (xfs_attr_shortform_t
*)XFS_DFORK_APTR(dip
);
347 size
= be16_to_cpu(atp
->hdr
.totsize
);
349 if (unlikely(size
< sizeof(struct xfs_attr_sf_hdr
))) {
350 xfs_warn(ip
->i_mount
,
351 "corrupt inode %Lu (bad attr fork size %Ld).",
352 (unsigned long long) ip
->i_ino
,
354 XFS_CORRUPTION_ERROR("xfs_iformat(8)",
357 return XFS_ERROR(EFSCORRUPTED
);
360 error
= xfs_iformat_local(ip
, dip
, XFS_ATTR_FORK
, size
);
362 case XFS_DINODE_FMT_EXTENTS
:
363 error
= xfs_iformat_extents(ip
, dip
, XFS_ATTR_FORK
);
365 case XFS_DINODE_FMT_BTREE
:
366 error
= xfs_iformat_btree(ip
, dip
, XFS_ATTR_FORK
);
369 error
= XFS_ERROR(EFSCORRUPTED
);
373 kmem_zone_free(xfs_ifork_zone
, ip
->i_afp
);
375 xfs_idestroy_fork(ip
, XFS_DATA_FORK
);
381 * The file is in-lined in the on-disk inode.
382 * If it fits into if_inline_data, then copy
383 * it there, otherwise allocate a buffer for it
384 * and copy the data there. Either way, set
385 * if_data to point at the data.
386 * If we allocate a buffer for the data, make
387 * sure that its size is a multiple of 4 and
388 * record the real size in i_real_bytes.
401 * If the size is unreasonable, then something
402 * is wrong and we just bail out rather than crash in
403 * kmem_alloc() or memcpy() below.
405 if (unlikely(size
> XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
))) {
406 xfs_warn(ip
->i_mount
,
407 "corrupt inode %Lu (bad size %d for local fork, size = %d).",
408 (unsigned long long) ip
->i_ino
, size
,
409 XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
));
410 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW
,
412 return XFS_ERROR(EFSCORRUPTED
);
414 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
417 ifp
->if_u1
.if_data
= NULL
;
418 else if (size
<= sizeof(ifp
->if_u2
.if_inline_data
))
419 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
421 real_size
= roundup(size
, 4);
422 ifp
->if_u1
.if_data
= kmem_alloc(real_size
, KM_SLEEP
| KM_NOFS
);
424 ifp
->if_bytes
= size
;
425 ifp
->if_real_bytes
= real_size
;
427 memcpy(ifp
->if_u1
.if_data
, XFS_DFORK_PTR(dip
, whichfork
), size
);
428 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
429 ifp
->if_flags
|= XFS_IFINLINE
;
434 * The file consists of a set of extents all
435 * of which fit into the on-disk inode.
436 * If there are few enough extents to fit into
437 * the if_inline_ext, then copy them there.
438 * Otherwise allocate a buffer for them and copy
439 * them into it. Either way, set if_extents
440 * to point at the extents.
454 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
455 nex
= XFS_DFORK_NEXTENTS(dip
, whichfork
);
456 size
= nex
* (uint
)sizeof(xfs_bmbt_rec_t
);
459 * If the number of extents is unreasonable, then something
460 * is wrong and we just bail out rather than crash in
461 * kmem_alloc() or memcpy() below.
463 if (unlikely(size
< 0 || size
> XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
))) {
464 xfs_warn(ip
->i_mount
, "corrupt inode %Lu ((a)extents = %d).",
465 (unsigned long long) ip
->i_ino
, nex
);
466 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW
,
468 return XFS_ERROR(EFSCORRUPTED
);
471 ifp
->if_real_bytes
= 0;
473 ifp
->if_u1
.if_extents
= NULL
;
474 else if (nex
<= XFS_INLINE_EXTS
)
475 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
477 xfs_iext_add(ifp
, 0, nex
);
479 ifp
->if_bytes
= size
;
481 dp
= (xfs_bmbt_rec_t
*) XFS_DFORK_PTR(dip
, whichfork
);
482 xfs_validate_extents(ifp
, nex
, XFS_EXTFMT_INODE(ip
));
483 for (i
= 0; i
< nex
; i
++, dp
++) {
484 xfs_bmbt_rec_host_t
*ep
= xfs_iext_get_ext(ifp
, i
);
485 ep
->l0
= get_unaligned_be64(&dp
->l0
);
486 ep
->l1
= get_unaligned_be64(&dp
->l1
);
488 XFS_BMAP_TRACE_EXLIST(ip
, nex
, whichfork
);
489 if (whichfork
!= XFS_DATA_FORK
||
490 XFS_EXTFMT_INODE(ip
) == XFS_EXTFMT_NOSTATE
)
491 if (unlikely(xfs_check_nostate_extents(
493 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
496 return XFS_ERROR(EFSCORRUPTED
);
499 ifp
->if_flags
|= XFS_IFEXTENTS
;
504 * The file has too many extents to fit into
505 * the inode, so they are in B-tree format.
506 * Allocate a buffer for the root of the B-tree
507 * and copy the root into it. The i_extents
508 * field will remain NULL until all of the
509 * extents are read in (when they are needed).
517 xfs_bmdr_block_t
*dfp
;
523 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
524 dfp
= (xfs_bmdr_block_t
*)XFS_DFORK_PTR(dip
, whichfork
);
525 size
= XFS_BMAP_BROOT_SPACE(dfp
);
526 nrecs
= be16_to_cpu(dfp
->bb_numrecs
);
529 * blow out if -- fork has less extents than can fit in
530 * fork (fork shouldn't be a btree format), root btree
531 * block has more records than can fit into the fork,
532 * or the number of extents is greater than the number of
535 if (unlikely(XFS_IFORK_NEXTENTS(ip
, whichfork
) <=
536 XFS_IFORK_MAXEXT(ip
, whichfork
) ||
537 XFS_BMDR_SPACE_CALC(nrecs
) >
538 XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
) ||
539 XFS_IFORK_NEXTENTS(ip
, whichfork
) > ip
->i_d
.di_nblocks
)) {
540 xfs_warn(ip
->i_mount
, "corrupt inode %Lu (btree).",
541 (unsigned long long) ip
->i_ino
);
542 XFS_CORRUPTION_ERROR("xfs_iformat_btree", XFS_ERRLEVEL_LOW
,
544 return XFS_ERROR(EFSCORRUPTED
);
547 ifp
->if_broot_bytes
= size
;
548 ifp
->if_broot
= kmem_alloc(size
, KM_SLEEP
| KM_NOFS
);
549 ASSERT(ifp
->if_broot
!= NULL
);
551 * Copy and convert from the on-disk structure
552 * to the in-memory structure.
554 xfs_bmdr_to_bmbt(ip
->i_mount
, dfp
,
555 XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
),
556 ifp
->if_broot
, size
);
557 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
558 ifp
->if_flags
|= XFS_IFBROOT
;
564 xfs_dinode_from_disk(
568 to
->di_magic
= be16_to_cpu(from
->di_magic
);
569 to
->di_mode
= be16_to_cpu(from
->di_mode
);
570 to
->di_version
= from
->di_version
;
571 to
->di_format
= from
->di_format
;
572 to
->di_onlink
= be16_to_cpu(from
->di_onlink
);
573 to
->di_uid
= be32_to_cpu(from
->di_uid
);
574 to
->di_gid
= be32_to_cpu(from
->di_gid
);
575 to
->di_nlink
= be32_to_cpu(from
->di_nlink
);
576 to
->di_projid_lo
= be16_to_cpu(from
->di_projid_lo
);
577 to
->di_projid_hi
= be16_to_cpu(from
->di_projid_hi
);
578 memcpy(to
->di_pad
, from
->di_pad
, sizeof(to
->di_pad
));
579 to
->di_flushiter
= be16_to_cpu(from
->di_flushiter
);
580 to
->di_atime
.t_sec
= be32_to_cpu(from
->di_atime
.t_sec
);
581 to
->di_atime
.t_nsec
= be32_to_cpu(from
->di_atime
.t_nsec
);
582 to
->di_mtime
.t_sec
= be32_to_cpu(from
->di_mtime
.t_sec
);
583 to
->di_mtime
.t_nsec
= be32_to_cpu(from
->di_mtime
.t_nsec
);
584 to
->di_ctime
.t_sec
= be32_to_cpu(from
->di_ctime
.t_sec
);
585 to
->di_ctime
.t_nsec
= be32_to_cpu(from
->di_ctime
.t_nsec
);
586 to
->di_size
= be64_to_cpu(from
->di_size
);
587 to
->di_nblocks
= be64_to_cpu(from
->di_nblocks
);
588 to
->di_extsize
= be32_to_cpu(from
->di_extsize
);
589 to
->di_nextents
= be32_to_cpu(from
->di_nextents
);
590 to
->di_anextents
= be16_to_cpu(from
->di_anextents
);
591 to
->di_forkoff
= from
->di_forkoff
;
592 to
->di_aformat
= from
->di_aformat
;
593 to
->di_dmevmask
= be32_to_cpu(from
->di_dmevmask
);
594 to
->di_dmstate
= be16_to_cpu(from
->di_dmstate
);
595 to
->di_flags
= be16_to_cpu(from
->di_flags
);
596 to
->di_gen
= be32_to_cpu(from
->di_gen
);
602 xfs_icdinode_t
*from
)
604 to
->di_magic
= cpu_to_be16(from
->di_magic
);
605 to
->di_mode
= cpu_to_be16(from
->di_mode
);
606 to
->di_version
= from
->di_version
;
607 to
->di_format
= from
->di_format
;
608 to
->di_onlink
= cpu_to_be16(from
->di_onlink
);
609 to
->di_uid
= cpu_to_be32(from
->di_uid
);
610 to
->di_gid
= cpu_to_be32(from
->di_gid
);
611 to
->di_nlink
= cpu_to_be32(from
->di_nlink
);
612 to
->di_projid_lo
= cpu_to_be16(from
->di_projid_lo
);
613 to
->di_projid_hi
= cpu_to_be16(from
->di_projid_hi
);
614 memcpy(to
->di_pad
, from
->di_pad
, sizeof(to
->di_pad
));
615 to
->di_flushiter
= cpu_to_be16(from
->di_flushiter
);
616 to
->di_atime
.t_sec
= cpu_to_be32(from
->di_atime
.t_sec
);
617 to
->di_atime
.t_nsec
= cpu_to_be32(from
->di_atime
.t_nsec
);
618 to
->di_mtime
.t_sec
= cpu_to_be32(from
->di_mtime
.t_sec
);
619 to
->di_mtime
.t_nsec
= cpu_to_be32(from
->di_mtime
.t_nsec
);
620 to
->di_ctime
.t_sec
= cpu_to_be32(from
->di_ctime
.t_sec
);
621 to
->di_ctime
.t_nsec
= cpu_to_be32(from
->di_ctime
.t_nsec
);
622 to
->di_size
= cpu_to_be64(from
->di_size
);
623 to
->di_nblocks
= cpu_to_be64(from
->di_nblocks
);
624 to
->di_extsize
= cpu_to_be32(from
->di_extsize
);
625 to
->di_nextents
= cpu_to_be32(from
->di_nextents
);
626 to
->di_anextents
= cpu_to_be16(from
->di_anextents
);
627 to
->di_forkoff
= from
->di_forkoff
;
628 to
->di_aformat
= from
->di_aformat
;
629 to
->di_dmevmask
= cpu_to_be32(from
->di_dmevmask
);
630 to
->di_dmstate
= cpu_to_be16(from
->di_dmstate
);
631 to
->di_flags
= cpu_to_be16(from
->di_flags
);
632 to
->di_gen
= cpu_to_be32(from
->di_gen
);
641 if (di_flags
& XFS_DIFLAG_ANY
) {
642 if (di_flags
& XFS_DIFLAG_REALTIME
)
643 flags
|= XFS_XFLAG_REALTIME
;
644 if (di_flags
& XFS_DIFLAG_PREALLOC
)
645 flags
|= XFS_XFLAG_PREALLOC
;
646 if (di_flags
& XFS_DIFLAG_IMMUTABLE
)
647 flags
|= XFS_XFLAG_IMMUTABLE
;
648 if (di_flags
& XFS_DIFLAG_APPEND
)
649 flags
|= XFS_XFLAG_APPEND
;
650 if (di_flags
& XFS_DIFLAG_SYNC
)
651 flags
|= XFS_XFLAG_SYNC
;
652 if (di_flags
& XFS_DIFLAG_NOATIME
)
653 flags
|= XFS_XFLAG_NOATIME
;
654 if (di_flags
& XFS_DIFLAG_NODUMP
)
655 flags
|= XFS_XFLAG_NODUMP
;
656 if (di_flags
& XFS_DIFLAG_RTINHERIT
)
657 flags
|= XFS_XFLAG_RTINHERIT
;
658 if (di_flags
& XFS_DIFLAG_PROJINHERIT
)
659 flags
|= XFS_XFLAG_PROJINHERIT
;
660 if (di_flags
& XFS_DIFLAG_NOSYMLINKS
)
661 flags
|= XFS_XFLAG_NOSYMLINKS
;
662 if (di_flags
& XFS_DIFLAG_EXTSIZE
)
663 flags
|= XFS_XFLAG_EXTSIZE
;
664 if (di_flags
& XFS_DIFLAG_EXTSZINHERIT
)
665 flags
|= XFS_XFLAG_EXTSZINHERIT
;
666 if (di_flags
& XFS_DIFLAG_NODEFRAG
)
667 flags
|= XFS_XFLAG_NODEFRAG
;
668 if (di_flags
& XFS_DIFLAG_FILESTREAM
)
669 flags
|= XFS_XFLAG_FILESTREAM
;
679 xfs_icdinode_t
*dic
= &ip
->i_d
;
681 return _xfs_dic2xflags(dic
->di_flags
) |
682 (XFS_IFORK_Q(ip
) ? XFS_XFLAG_HASATTR
: 0);
689 return _xfs_dic2xflags(be16_to_cpu(dip
->di_flags
)) |
690 (XFS_DFORK_Q(dip
) ? XFS_XFLAG_HASATTR
: 0);
694 * Read the disk inode attributes into the in-core inode structure.
708 * Fill in the location information in the in-core inode.
710 error
= xfs_imap(mp
, tp
, ip
->i_ino
, &ip
->i_imap
, iget_flags
);
715 * Get pointers to the on-disk inode and the buffer containing it.
717 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &dip
, &bp
, 0, iget_flags
);
722 * If we got something that isn't an inode it means someone
723 * (nfs or dmi) has a stale handle.
725 if (dip
->di_magic
!= cpu_to_be16(XFS_DINODE_MAGIC
)) {
728 "%s: dip->di_magic (0x%x) != XFS_DINODE_MAGIC (0x%x)",
729 __func__
, be16_to_cpu(dip
->di_magic
), XFS_DINODE_MAGIC
);
731 error
= XFS_ERROR(EINVAL
);
736 * If the on-disk inode is already linked to a directory
737 * entry, copy all of the inode into the in-core inode.
738 * xfs_iformat() handles copying in the inode format
739 * specific information.
740 * Otherwise, just get the truly permanent information.
743 xfs_dinode_from_disk(&ip
->i_d
, dip
);
744 error
= xfs_iformat(ip
, dip
);
747 xfs_alert(mp
, "%s: xfs_iformat() returned error %d",
753 ip
->i_d
.di_magic
= be16_to_cpu(dip
->di_magic
);
754 ip
->i_d
.di_version
= dip
->di_version
;
755 ip
->i_d
.di_gen
= be32_to_cpu(dip
->di_gen
);
756 ip
->i_d
.di_flushiter
= be16_to_cpu(dip
->di_flushiter
);
758 * Make sure to pull in the mode here as well in
759 * case the inode is released without being used.
760 * This ensures that xfs_inactive() will see that
761 * the inode is already free and not try to mess
762 * with the uninitialized part of it.
768 * The inode format changed when we moved the link count and
769 * made it 32 bits long. If this is an old format inode,
770 * convert it in memory to look like a new one. If it gets
771 * flushed to disk we will convert back before flushing or
772 * logging it. We zero out the new projid field and the old link
773 * count field. We'll handle clearing the pad field (the remains
774 * of the old uuid field) when we actually convert the inode to
775 * the new format. We don't change the version number so that we
776 * can distinguish this from a real new format inode.
778 if (ip
->i_d
.di_version
== 1) {
779 ip
->i_d
.di_nlink
= ip
->i_d
.di_onlink
;
780 ip
->i_d
.di_onlink
= 0;
781 xfs_set_projid(ip
, 0);
784 ip
->i_delayed_blks
= 0;
787 * Mark the buffer containing the inode as something to keep
788 * around for a while. This helps to keep recently accessed
789 * meta-data in-core longer.
791 xfs_buf_set_ref(bp
, XFS_INO_REF
);
794 * Use xfs_trans_brelse() to release the buffer containing the
795 * on-disk inode, because it was acquired with xfs_trans_read_buf()
796 * in xfs_imap_to_bp() above. If tp is NULL, this is just a normal
797 * brelse(). If we're within a transaction, then xfs_trans_brelse()
798 * will only release the buffer if it is not dirty within the
799 * transaction. It will be OK to release the buffer in this case,
800 * because inodes on disk are never destroyed and we will be
801 * locking the new in-core inode before putting it in the hash
802 * table where other processes can find it. Thus we don't have
803 * to worry about the inode being changed just because we released
807 xfs_trans_brelse(tp
, bp
);
812 * Read in extents from a btree-format inode.
813 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
823 xfs_extnum_t nextents
;
825 if (unlikely(XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
)) {
826 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW
,
828 return XFS_ERROR(EFSCORRUPTED
);
830 nextents
= XFS_IFORK_NEXTENTS(ip
, whichfork
);
831 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
834 * We know that the size is valid (it's checked in iformat_btree)
836 ifp
->if_bytes
= ifp
->if_real_bytes
= 0;
837 ifp
->if_flags
|= XFS_IFEXTENTS
;
838 xfs_iext_add(ifp
, 0, nextents
);
839 error
= xfs_bmap_read_extents(tp
, ip
, whichfork
);
841 xfs_iext_destroy(ifp
);
842 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
845 xfs_validate_extents(ifp
, nextents
, XFS_EXTFMT_INODE(ip
));
850 * Allocate an inode on disk and return a copy of its in-core version.
851 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
852 * appropriately within the inode. The uid and gid for the inode are
853 * set according to the contents of the given cred structure.
855 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
856 * has a free inode available, call xfs_iget()
857 * to obtain the in-core version of the allocated inode. Finally,
858 * fill in the inode and log its initial contents. In this case,
859 * ialloc_context would be set to NULL and call_again set to false.
861 * If xfs_dialloc() does not have an available inode,
862 * it will replenish its supply by doing an allocation. Since we can
863 * only do one allocation within a transaction without deadlocks, we
864 * must commit the current transaction before returning the inode itself.
865 * In this case, therefore, we will set call_again to true and return.
866 * The caller should then commit the current transaction, start a new
867 * transaction, and call xfs_ialloc() again to actually get the inode.
869 * To ensure that some other process does not grab the inode that
870 * was allocated during the first call to xfs_ialloc(), this routine
871 * also returns the [locked] bp pointing to the head of the freelist
872 * as ialloc_context. The caller should hold this buffer across
873 * the commit and pass it back into this routine on the second call.
875 * If we are allocating quota inodes, we do not have a parent inode
876 * to attach to or associate with (i.e. pip == NULL) because they
877 * are not linked into the directory structure - they are attached
878 * directly to the superblock - and so have no parent.
889 xfs_buf_t
**ialloc_context
,
890 boolean_t
*call_again
,
901 * Call the space management code to pick
902 * the on-disk inode to be allocated.
904 error
= xfs_dialloc(tp
, pip
? pip
->i_ino
: 0, mode
, okalloc
,
905 ialloc_context
, call_again
, &ino
);
908 if (*call_again
|| ino
== NULLFSINO
) {
912 ASSERT(*ialloc_context
== NULL
);
915 * Get the in-core inode with the lock held exclusively.
916 * This is because we're setting fields here we need
917 * to prevent others from looking at until we're done.
919 error
= xfs_iget(tp
->t_mountp
, tp
, ino
, XFS_IGET_CREATE
,
920 XFS_ILOCK_EXCL
, &ip
);
925 ip
->i_d
.di_mode
= mode
;
926 ip
->i_d
.di_onlink
= 0;
927 ip
->i_d
.di_nlink
= nlink
;
928 ASSERT(ip
->i_d
.di_nlink
== nlink
);
929 ip
->i_d
.di_uid
= current_fsuid();
930 ip
->i_d
.di_gid
= current_fsgid();
931 xfs_set_projid(ip
, prid
);
932 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
935 * If the superblock version is up to where we support new format
936 * inodes and this is currently an old format inode, then change
937 * the inode version number now. This way we only do the conversion
938 * here rather than here and in the flush/logging code.
940 if (xfs_sb_version_hasnlink(&tp
->t_mountp
->m_sb
) &&
941 ip
->i_d
.di_version
== 1) {
942 ip
->i_d
.di_version
= 2;
944 * We've already zeroed the old link count, the projid field,
950 * Project ids won't be stored on disk if we are using a version 1 inode.
952 if ((prid
!= 0) && (ip
->i_d
.di_version
== 1))
953 xfs_bump_ino_vers2(tp
, ip
);
955 if (pip
&& XFS_INHERIT_GID(pip
)) {
956 ip
->i_d
.di_gid
= pip
->i_d
.di_gid
;
957 if ((pip
->i_d
.di_mode
& S_ISGID
) && S_ISDIR(mode
)) {
958 ip
->i_d
.di_mode
|= S_ISGID
;
963 * If the group ID of the new file does not match the effective group
964 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
965 * (and only if the irix_sgid_inherit compatibility variable is set).
967 if ((irix_sgid_inherit
) &&
968 (ip
->i_d
.di_mode
& S_ISGID
) &&
969 (!in_group_p((gid_t
)ip
->i_d
.di_gid
))) {
970 ip
->i_d
.di_mode
&= ~S_ISGID
;
974 ip
->i_d
.di_nextents
= 0;
975 ASSERT(ip
->i_d
.di_nblocks
== 0);
978 ip
->i_d
.di_mtime
.t_sec
= (__int32_t
)tv
.tv_sec
;
979 ip
->i_d
.di_mtime
.t_nsec
= (__int32_t
)tv
.tv_nsec
;
980 ip
->i_d
.di_atime
= ip
->i_d
.di_mtime
;
981 ip
->i_d
.di_ctime
= ip
->i_d
.di_mtime
;
984 * di_gen will have been taken care of in xfs_iread.
986 ip
->i_d
.di_extsize
= 0;
987 ip
->i_d
.di_dmevmask
= 0;
988 ip
->i_d
.di_dmstate
= 0;
989 ip
->i_d
.di_flags
= 0;
990 flags
= XFS_ILOG_CORE
;
991 switch (mode
& S_IFMT
) {
996 ip
->i_d
.di_format
= XFS_DINODE_FMT_DEV
;
997 ip
->i_df
.if_u2
.if_rdev
= rdev
;
998 ip
->i_df
.if_flags
= 0;
999 flags
|= XFS_ILOG_DEV
;
1003 * we can't set up filestreams until after the VFS inode
1004 * is set up properly.
1006 if (pip
&& xfs_inode_is_filestream(pip
))
1010 if (pip
&& (pip
->i_d
.di_flags
& XFS_DIFLAG_ANY
)) {
1013 if (S_ISDIR(mode
)) {
1014 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
1015 di_flags
|= XFS_DIFLAG_RTINHERIT
;
1016 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
1017 di_flags
|= XFS_DIFLAG_EXTSZINHERIT
;
1018 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
1020 } else if (S_ISREG(mode
)) {
1021 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
1022 di_flags
|= XFS_DIFLAG_REALTIME
;
1023 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
1024 di_flags
|= XFS_DIFLAG_EXTSIZE
;
1025 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
1028 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOATIME
) &&
1029 xfs_inherit_noatime
)
1030 di_flags
|= XFS_DIFLAG_NOATIME
;
1031 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODUMP
) &&
1033 di_flags
|= XFS_DIFLAG_NODUMP
;
1034 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_SYNC
) &&
1036 di_flags
|= XFS_DIFLAG_SYNC
;
1037 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOSYMLINKS
) &&
1038 xfs_inherit_nosymlinks
)
1039 di_flags
|= XFS_DIFLAG_NOSYMLINKS
;
1040 if (pip
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
1041 di_flags
|= XFS_DIFLAG_PROJINHERIT
;
1042 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODEFRAG
) &&
1043 xfs_inherit_nodefrag
)
1044 di_flags
|= XFS_DIFLAG_NODEFRAG
;
1045 if (pip
->i_d
.di_flags
& XFS_DIFLAG_FILESTREAM
)
1046 di_flags
|= XFS_DIFLAG_FILESTREAM
;
1047 ip
->i_d
.di_flags
|= di_flags
;
1051 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
1052 ip
->i_df
.if_flags
= XFS_IFEXTENTS
;
1053 ip
->i_df
.if_bytes
= ip
->i_df
.if_real_bytes
= 0;
1054 ip
->i_df
.if_u1
.if_extents
= NULL
;
1060 * Attribute fork settings for new inode.
1062 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
1063 ip
->i_d
.di_anextents
= 0;
1066 * Log the new values stuffed into the inode.
1068 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
1069 xfs_trans_log_inode(tp
, ip
, flags
);
1071 /* now that we have an i_mode we can setup inode ops and unlock */
1072 xfs_setup_inode(ip
);
1074 /* now we have set up the vfs inode we can associate the filestream */
1076 error
= xfs_filestream_associate(pip
, ip
);
1080 xfs_iflags_set(ip
, XFS_IFILESTREAM
);
1088 * Free up the underlying blocks past new_size. The new size must be smaller
1089 * than the current size. This routine can be used both for the attribute and
1090 * data fork, and does not modify the inode size, which is left to the caller.
1092 * The transaction passed to this routine must have made a permanent log
1093 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1094 * given transaction and start new ones, so make sure everything involved in
1095 * the transaction is tidy before calling here. Some transaction will be
1096 * returned to the caller to be committed. The incoming transaction must
1097 * already include the inode, and both inode locks must be held exclusively.
1098 * The inode must also be "held" within the transaction. On return the inode
1099 * will be "held" within the returned transaction. This routine does NOT
1100 * require any disk space to be reserved for it within the transaction.
1102 * If we get an error, we must return with the inode locked and linked into the
1103 * current transaction. This keeps things simple for the higher level code,
1104 * because it always knows that the inode is locked and held in the transaction
1105 * that returns to it whether errors occur or not. We don't mark the inode
1106 * dirty on error so that transactions can be easily aborted if possible.
1109 xfs_itruncate_extents(
1110 struct xfs_trans
**tpp
,
1111 struct xfs_inode
*ip
,
1113 xfs_fsize_t new_size
)
1115 struct xfs_mount
*mp
= ip
->i_mount
;
1116 struct xfs_trans
*tp
= *tpp
;
1117 struct xfs_trans
*ntp
;
1118 xfs_bmap_free_t free_list
;
1119 xfs_fsblock_t first_block
;
1120 xfs_fileoff_t first_unmap_block
;
1121 xfs_fileoff_t last_block
;
1122 xfs_filblks_t unmap_len
;
1127 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_IOLOCK_EXCL
));
1128 ASSERT(new_size
<= XFS_ISIZE(ip
));
1129 ASSERT(tp
->t_flags
& XFS_TRANS_PERM_LOG_RES
);
1130 ASSERT(ip
->i_itemp
!= NULL
);
1131 ASSERT(ip
->i_itemp
->ili_lock_flags
== 0);
1132 ASSERT(!XFS_NOT_DQATTACHED(mp
, ip
));
1134 trace_xfs_itruncate_extents_start(ip
, new_size
);
1137 * Since it is possible for space to become allocated beyond
1138 * the end of the file (in a crash where the space is allocated
1139 * but the inode size is not yet updated), simply remove any
1140 * blocks which show up between the new EOF and the maximum
1141 * possible file size. If the first block to be removed is
1142 * beyond the maximum file size (ie it is the same as last_block),
1143 * then there is nothing to do.
1145 first_unmap_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)new_size
);
1146 last_block
= XFS_B_TO_FSB(mp
, mp
->m_super
->s_maxbytes
);
1147 if (first_unmap_block
== last_block
)
1150 ASSERT(first_unmap_block
< last_block
);
1151 unmap_len
= last_block
- first_unmap_block
+ 1;
1153 xfs_bmap_init(&free_list
, &first_block
);
1154 error
= xfs_bunmapi(tp
, ip
,
1155 first_unmap_block
, unmap_len
,
1156 xfs_bmapi_aflag(whichfork
),
1157 XFS_ITRUNC_MAX_EXTENTS
,
1158 &first_block
, &free_list
,
1161 goto out_bmap_cancel
;
1164 * Duplicate the transaction that has the permanent
1165 * reservation and commit the old transaction.
1167 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
1169 xfs_trans_ijoin(tp
, ip
, 0);
1171 goto out_bmap_cancel
;
1175 * Mark the inode dirty so it will be logged and
1176 * moved forward in the log as part of every commit.
1178 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1181 ntp
= xfs_trans_dup(tp
);
1182 error
= xfs_trans_commit(tp
, 0);
1185 xfs_trans_ijoin(tp
, ip
, 0);
1191 * Transaction commit worked ok so we can drop the extra ticket
1192 * reference that we gained in xfs_trans_dup()
1194 xfs_log_ticket_put(tp
->t_ticket
);
1195 error
= xfs_trans_reserve(tp
, 0,
1196 XFS_ITRUNCATE_LOG_RES(mp
), 0,
1197 XFS_TRANS_PERM_LOG_RES
,
1198 XFS_ITRUNCATE_LOG_COUNT
);
1204 * Always re-log the inode so that our permanent transaction can keep
1205 * on rolling it forward in the log.
1207 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1209 trace_xfs_itruncate_extents_end(ip
, new_size
);
1216 * If the bunmapi call encounters an error, return to the caller where
1217 * the transaction can be properly aborted. We just need to make sure
1218 * we're not holding any resources that we were not when we came in.
1220 xfs_bmap_cancel(&free_list
);
1225 * This is called when the inode's link count goes to 0.
1226 * We place the on-disk inode on a list in the AGI. It
1227 * will be pulled from this list when the inode is freed.
1244 ASSERT(ip
->i_d
.di_nlink
== 0);
1245 ASSERT(ip
->i_d
.di_mode
!= 0);
1250 * Get the agi buffer first. It ensures lock ordering
1253 error
= xfs_read_agi(mp
, tp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
), &agibp
);
1256 agi
= XFS_BUF_TO_AGI(agibp
);
1259 * Get the index into the agi hash table for the
1260 * list this inode will go on.
1262 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
1264 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
1265 ASSERT(agi
->agi_unlinked
[bucket_index
]);
1266 ASSERT(be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != agino
);
1268 if (agi
->agi_unlinked
[bucket_index
] != cpu_to_be32(NULLAGINO
)) {
1270 * There is already another inode in the bucket we need
1271 * to add ourselves to. Add us at the front of the list.
1272 * Here we put the head pointer into our next pointer,
1273 * and then we fall through to point the head at us.
1275 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &dip
, &ibp
,
1280 ASSERT(dip
->di_next_unlinked
== cpu_to_be32(NULLAGINO
));
1281 dip
->di_next_unlinked
= agi
->agi_unlinked
[bucket_index
];
1282 offset
= ip
->i_imap
.im_boffset
+
1283 offsetof(xfs_dinode_t
, di_next_unlinked
);
1284 xfs_trans_inode_buf(tp
, ibp
);
1285 xfs_trans_log_buf(tp
, ibp
, offset
,
1286 (offset
+ sizeof(xfs_agino_t
) - 1));
1287 xfs_inobp_check(mp
, ibp
);
1291 * Point the bucket head pointer at the inode being inserted.
1294 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(agino
);
1295 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
1296 (sizeof(xfs_agino_t
) * bucket_index
);
1297 xfs_trans_log_buf(tp
, agibp
, offset
,
1298 (offset
+ sizeof(xfs_agino_t
) - 1));
1303 * Pull the on-disk inode from the AGI unlinked list.
1316 xfs_agnumber_t agno
;
1318 xfs_agino_t next_agino
;
1319 xfs_buf_t
*last_ibp
;
1320 xfs_dinode_t
*last_dip
= NULL
;
1322 int offset
, last_offset
= 0;
1326 agno
= XFS_INO_TO_AGNO(mp
, ip
->i_ino
);
1329 * Get the agi buffer first. It ensures lock ordering
1332 error
= xfs_read_agi(mp
, tp
, agno
, &agibp
);
1336 agi
= XFS_BUF_TO_AGI(agibp
);
1339 * Get the index into the agi hash table for the
1340 * list this inode will go on.
1342 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
1344 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
1345 ASSERT(agi
->agi_unlinked
[bucket_index
] != cpu_to_be32(NULLAGINO
));
1346 ASSERT(agi
->agi_unlinked
[bucket_index
]);
1348 if (be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) == agino
) {
1350 * We're at the head of the list. Get the inode's on-disk
1351 * buffer to see if there is anyone after us on the list.
1352 * Only modify our next pointer if it is not already NULLAGINO.
1353 * This saves us the overhead of dealing with the buffer when
1354 * there is no need to change it.
1356 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &dip
, &ibp
,
1359 xfs_warn(mp
, "%s: xfs_imap_to_bp returned error %d.",
1363 next_agino
= be32_to_cpu(dip
->di_next_unlinked
);
1364 ASSERT(next_agino
!= 0);
1365 if (next_agino
!= NULLAGINO
) {
1366 dip
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
1367 offset
= ip
->i_imap
.im_boffset
+
1368 offsetof(xfs_dinode_t
, di_next_unlinked
);
1369 xfs_trans_inode_buf(tp
, ibp
);
1370 xfs_trans_log_buf(tp
, ibp
, offset
,
1371 (offset
+ sizeof(xfs_agino_t
) - 1));
1372 xfs_inobp_check(mp
, ibp
);
1374 xfs_trans_brelse(tp
, ibp
);
1377 * Point the bucket head pointer at the next inode.
1379 ASSERT(next_agino
!= 0);
1380 ASSERT(next_agino
!= agino
);
1381 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(next_agino
);
1382 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
1383 (sizeof(xfs_agino_t
) * bucket_index
);
1384 xfs_trans_log_buf(tp
, agibp
, offset
,
1385 (offset
+ sizeof(xfs_agino_t
) - 1));
1388 * We need to search the list for the inode being freed.
1390 next_agino
= be32_to_cpu(agi
->agi_unlinked
[bucket_index
]);
1392 while (next_agino
!= agino
) {
1393 struct xfs_imap imap
;
1396 xfs_trans_brelse(tp
, last_ibp
);
1399 next_ino
= XFS_AGINO_TO_INO(mp
, agno
, next_agino
);
1401 error
= xfs_imap(mp
, tp
, next_ino
, &imap
, 0);
1404 "%s: xfs_imap returned error %d.",
1409 error
= xfs_imap_to_bp(mp
, tp
, &imap
, &last_dip
,
1413 "%s: xfs_imap_to_bp returned error %d.",
1418 last_offset
= imap
.im_boffset
;
1419 next_agino
= be32_to_cpu(last_dip
->di_next_unlinked
);
1420 ASSERT(next_agino
!= NULLAGINO
);
1421 ASSERT(next_agino
!= 0);
1425 * Now last_ibp points to the buffer previous to us on the
1426 * unlinked list. Pull us from the list.
1428 error
= xfs_imap_to_bp(mp
, tp
, &ip
->i_imap
, &dip
, &ibp
,
1431 xfs_warn(mp
, "%s: xfs_imap_to_bp(2) returned error %d.",
1435 next_agino
= be32_to_cpu(dip
->di_next_unlinked
);
1436 ASSERT(next_agino
!= 0);
1437 ASSERT(next_agino
!= agino
);
1438 if (next_agino
!= NULLAGINO
) {
1439 dip
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
1440 offset
= ip
->i_imap
.im_boffset
+
1441 offsetof(xfs_dinode_t
, di_next_unlinked
);
1442 xfs_trans_inode_buf(tp
, ibp
);
1443 xfs_trans_log_buf(tp
, ibp
, offset
,
1444 (offset
+ sizeof(xfs_agino_t
) - 1));
1445 xfs_inobp_check(mp
, ibp
);
1447 xfs_trans_brelse(tp
, ibp
);
1450 * Point the previous inode on the list to the next inode.
1452 last_dip
->di_next_unlinked
= cpu_to_be32(next_agino
);
1453 ASSERT(next_agino
!= 0);
1454 offset
= last_offset
+ offsetof(xfs_dinode_t
, di_next_unlinked
);
1455 xfs_trans_inode_buf(tp
, last_ibp
);
1456 xfs_trans_log_buf(tp
, last_ibp
, offset
,
1457 (offset
+ sizeof(xfs_agino_t
) - 1));
1458 xfs_inobp_check(mp
, last_ibp
);
1464 * A big issue when freeing the inode cluster is is that we _cannot_ skip any
1465 * inodes that are in memory - they all must be marked stale and attached to
1466 * the cluster buffer.
1470 xfs_inode_t
*free_ip
,
1474 xfs_mount_t
*mp
= free_ip
->i_mount
;
1475 int blks_per_cluster
;
1482 xfs_inode_log_item_t
*iip
;
1483 xfs_log_item_t
*lip
;
1484 struct xfs_perag
*pag
;
1486 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, inum
));
1487 if (mp
->m_sb
.sb_blocksize
>= XFS_INODE_CLUSTER_SIZE(mp
)) {
1488 blks_per_cluster
= 1;
1489 ninodes
= mp
->m_sb
.sb_inopblock
;
1490 nbufs
= XFS_IALLOC_BLOCKS(mp
);
1492 blks_per_cluster
= XFS_INODE_CLUSTER_SIZE(mp
) /
1493 mp
->m_sb
.sb_blocksize
;
1494 ninodes
= blks_per_cluster
* mp
->m_sb
.sb_inopblock
;
1495 nbufs
= XFS_IALLOC_BLOCKS(mp
) / blks_per_cluster
;
1498 for (j
= 0; j
< nbufs
; j
++, inum
+= ninodes
) {
1499 blkno
= XFS_AGB_TO_DADDR(mp
, XFS_INO_TO_AGNO(mp
, inum
),
1500 XFS_INO_TO_AGBNO(mp
, inum
));
1503 * We obtain and lock the backing buffer first in the process
1504 * here, as we have to ensure that any dirty inode that we
1505 * can't get the flush lock on is attached to the buffer.
1506 * If we scan the in-memory inodes first, then buffer IO can
1507 * complete before we get a lock on it, and hence we may fail
1508 * to mark all the active inodes on the buffer stale.
1510 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, blkno
,
1511 mp
->m_bsize
* blks_per_cluster
, 0);
1516 * Walk the inodes already attached to the buffer and mark them
1517 * stale. These will all have the flush locks held, so an
1518 * in-memory inode walk can't lock them. By marking them all
1519 * stale first, we will not attempt to lock them in the loop
1520 * below as the XFS_ISTALE flag will be set.
1524 if (lip
->li_type
== XFS_LI_INODE
) {
1525 iip
= (xfs_inode_log_item_t
*)lip
;
1526 ASSERT(iip
->ili_logged
== 1);
1527 lip
->li_cb
= xfs_istale_done
;
1528 xfs_trans_ail_copy_lsn(mp
->m_ail
,
1529 &iip
->ili_flush_lsn
,
1530 &iip
->ili_item
.li_lsn
);
1531 xfs_iflags_set(iip
->ili_inode
, XFS_ISTALE
);
1533 lip
= lip
->li_bio_list
;
1538 * For each inode in memory attempt to add it to the inode
1539 * buffer and set it up for being staled on buffer IO
1540 * completion. This is safe as we've locked out tail pushing
1541 * and flushing by locking the buffer.
1543 * We have already marked every inode that was part of a
1544 * transaction stale above, which means there is no point in
1545 * even trying to lock them.
1547 for (i
= 0; i
< ninodes
; i
++) {
1550 ip
= radix_tree_lookup(&pag
->pag_ici_root
,
1551 XFS_INO_TO_AGINO(mp
, (inum
+ i
)));
1553 /* Inode not in memory, nothing to do */
1560 * because this is an RCU protected lookup, we could
1561 * find a recently freed or even reallocated inode
1562 * during the lookup. We need to check under the
1563 * i_flags_lock for a valid inode here. Skip it if it
1564 * is not valid, the wrong inode or stale.
1566 spin_lock(&ip
->i_flags_lock
);
1567 if (ip
->i_ino
!= inum
+ i
||
1568 __xfs_iflags_test(ip
, XFS_ISTALE
)) {
1569 spin_unlock(&ip
->i_flags_lock
);
1573 spin_unlock(&ip
->i_flags_lock
);
1576 * Don't try to lock/unlock the current inode, but we
1577 * _cannot_ skip the other inodes that we did not find
1578 * in the list attached to the buffer and are not
1579 * already marked stale. If we can't lock it, back off
1582 if (ip
!= free_ip
&&
1583 !xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
)) {
1591 xfs_iflags_set(ip
, XFS_ISTALE
);
1594 * we don't need to attach clean inodes or those only
1595 * with unlogged changes (which we throw away, anyway).
1598 if (!iip
|| xfs_inode_clean(ip
)) {
1599 ASSERT(ip
!= free_ip
);
1601 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1605 iip
->ili_last_fields
= iip
->ili_fields
;
1606 iip
->ili_fields
= 0;
1607 iip
->ili_logged
= 1;
1608 xfs_trans_ail_copy_lsn(mp
->m_ail
, &iip
->ili_flush_lsn
,
1609 &iip
->ili_item
.li_lsn
);
1611 xfs_buf_attach_iodone(bp
, xfs_istale_done
,
1615 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1618 xfs_trans_stale_inode_buf(tp
, bp
);
1619 xfs_trans_binval(tp
, bp
);
1627 * This is called to return an inode to the inode free list.
1628 * The inode should already be truncated to 0 length and have
1629 * no pages associated with it. This routine also assumes that
1630 * the inode is already a part of the transaction.
1632 * The on-disk copy of the inode will have been added to the list
1633 * of unlinked inodes in the AGI. We need to remove the inode from
1634 * that list atomically with respect to freeing it here.
1640 xfs_bmap_free_t
*flist
)
1644 xfs_ino_t first_ino
;
1648 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
1649 ASSERT(ip
->i_d
.di_nlink
== 0);
1650 ASSERT(ip
->i_d
.di_nextents
== 0);
1651 ASSERT(ip
->i_d
.di_anextents
== 0);
1652 ASSERT(ip
->i_d
.di_size
== 0 || !S_ISREG(ip
->i_d
.di_mode
));
1653 ASSERT(ip
->i_d
.di_nblocks
== 0);
1656 * Pull the on-disk inode from the AGI unlinked list.
1658 error
= xfs_iunlink_remove(tp
, ip
);
1663 error
= xfs_difree(tp
, ip
->i_ino
, flist
, &delete, &first_ino
);
1667 ip
->i_d
.di_mode
= 0; /* mark incore inode as free */
1668 ip
->i_d
.di_flags
= 0;
1669 ip
->i_d
.di_dmevmask
= 0;
1670 ip
->i_d
.di_forkoff
= 0; /* mark the attr fork not in use */
1671 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
1672 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
1674 * Bump the generation count so no one will be confused
1675 * by reincarnations of this inode.
1679 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1681 error
= xfs_imap_to_bp(ip
->i_mount
, tp
, &ip
->i_imap
, &dip
, &ibp
,
1687 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
1688 * from picking up this inode when it is reclaimed (its incore state
1689 * initialzed but not flushed to disk yet). The in-core di_mode is
1690 * already cleared and a corresponding transaction logged.
1691 * The hack here just synchronizes the in-core to on-disk
1692 * di_mode value in advance before the actual inode sync to disk.
1693 * This is OK because the inode is already unlinked and would never
1694 * change its di_mode again for this inode generation.
1695 * This is a temporary hack that would require a proper fix
1701 error
= xfs_ifree_cluster(ip
, tp
, first_ino
);
1708 * Reallocate the space for if_broot based on the number of records
1709 * being added or deleted as indicated in rec_diff. Move the records
1710 * and pointers in if_broot to fit the new size. When shrinking this
1711 * will eliminate holes between the records and pointers created by
1712 * the caller. When growing this will create holes to be filled in
1715 * The caller must not request to add more records than would fit in
1716 * the on-disk inode root. If the if_broot is currently NULL, then
1717 * if we adding records one will be allocated. The caller must also
1718 * not request that the number of records go below zero, although
1719 * it can go to zero.
1721 * ip -- the inode whose if_broot area is changing
1722 * ext_diff -- the change in the number of records, positive or negative,
1723 * requested for the if_broot array.
1731 struct xfs_mount
*mp
= ip
->i_mount
;
1734 struct xfs_btree_block
*new_broot
;
1741 * Handle the degenerate case quietly.
1743 if (rec_diff
== 0) {
1747 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1750 * If there wasn't any memory allocated before, just
1751 * allocate it now and get out.
1753 if (ifp
->if_broot_bytes
== 0) {
1754 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff
);
1755 ifp
->if_broot
= kmem_alloc(new_size
, KM_SLEEP
| KM_NOFS
);
1756 ifp
->if_broot_bytes
= (int)new_size
;
1761 * If there is already an existing if_broot, then we need
1762 * to realloc() it and shift the pointers to their new
1763 * location. The records don't change location because
1764 * they are kept butted up against the btree block header.
1766 cur_max
= xfs_bmbt_maxrecs(mp
, ifp
->if_broot_bytes
, 0);
1767 new_max
= cur_max
+ rec_diff
;
1768 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max
);
1769 ifp
->if_broot
= kmem_realloc(ifp
->if_broot
, new_size
,
1770 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max
), /* old size */
1771 KM_SLEEP
| KM_NOFS
);
1772 op
= (char *)XFS_BMAP_BROOT_PTR_ADDR(mp
, ifp
->if_broot
, 1,
1773 ifp
->if_broot_bytes
);
1774 np
= (char *)XFS_BMAP_BROOT_PTR_ADDR(mp
, ifp
->if_broot
, 1,
1776 ifp
->if_broot_bytes
= (int)new_size
;
1777 ASSERT(ifp
->if_broot_bytes
<=
1778 XFS_IFORK_SIZE(ip
, whichfork
) + XFS_BROOT_SIZE_ADJ
);
1779 memmove(np
, op
, cur_max
* (uint
)sizeof(xfs_dfsbno_t
));
1784 * rec_diff is less than 0. In this case, we are shrinking the
1785 * if_broot buffer. It must already exist. If we go to zero
1786 * records, just get rid of the root and clear the status bit.
1788 ASSERT((ifp
->if_broot
!= NULL
) && (ifp
->if_broot_bytes
> 0));
1789 cur_max
= xfs_bmbt_maxrecs(mp
, ifp
->if_broot_bytes
, 0);
1790 new_max
= cur_max
+ rec_diff
;
1791 ASSERT(new_max
>= 0);
1793 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max
);
1797 new_broot
= kmem_alloc(new_size
, KM_SLEEP
| KM_NOFS
);
1799 * First copy over the btree block header.
1801 memcpy(new_broot
, ifp
->if_broot
, XFS_BTREE_LBLOCK_LEN
);
1804 ifp
->if_flags
&= ~XFS_IFBROOT
;
1808 * Only copy the records and pointers if there are any.
1812 * First copy the records.
1814 op
= (char *)XFS_BMBT_REC_ADDR(mp
, ifp
->if_broot
, 1);
1815 np
= (char *)XFS_BMBT_REC_ADDR(mp
, new_broot
, 1);
1816 memcpy(np
, op
, new_max
* (uint
)sizeof(xfs_bmbt_rec_t
));
1819 * Then copy the pointers.
1821 op
= (char *)XFS_BMAP_BROOT_PTR_ADDR(mp
, ifp
->if_broot
, 1,
1822 ifp
->if_broot_bytes
);
1823 np
= (char *)XFS_BMAP_BROOT_PTR_ADDR(mp
, new_broot
, 1,
1825 memcpy(np
, op
, new_max
* (uint
)sizeof(xfs_dfsbno_t
));
1827 kmem_free(ifp
->if_broot
);
1828 ifp
->if_broot
= new_broot
;
1829 ifp
->if_broot_bytes
= (int)new_size
;
1830 ASSERT(ifp
->if_broot_bytes
<=
1831 XFS_IFORK_SIZE(ip
, whichfork
) + XFS_BROOT_SIZE_ADJ
);
1837 * This is called when the amount of space needed for if_data
1838 * is increased or decreased. The change in size is indicated by
1839 * the number of bytes that need to be added or deleted in the
1840 * byte_diff parameter.
1842 * If the amount of space needed has decreased below the size of the
1843 * inline buffer, then switch to using the inline buffer. Otherwise,
1844 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
1845 * to what is needed.
1847 * ip -- the inode whose if_data area is changing
1848 * byte_diff -- the change in the number of bytes, positive or negative,
1849 * requested for the if_data array.
1861 if (byte_diff
== 0) {
1865 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1866 new_size
= (int)ifp
->if_bytes
+ byte_diff
;
1867 ASSERT(new_size
>= 0);
1869 if (new_size
== 0) {
1870 if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
1871 kmem_free(ifp
->if_u1
.if_data
);
1873 ifp
->if_u1
.if_data
= NULL
;
1875 } else if (new_size
<= sizeof(ifp
->if_u2
.if_inline_data
)) {
1877 * If the valid extents/data can fit in if_inline_ext/data,
1878 * copy them from the malloc'd vector and free it.
1880 if (ifp
->if_u1
.if_data
== NULL
) {
1881 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
1882 } else if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
1883 ASSERT(ifp
->if_real_bytes
!= 0);
1884 memcpy(ifp
->if_u2
.if_inline_data
, ifp
->if_u1
.if_data
,
1886 kmem_free(ifp
->if_u1
.if_data
);
1887 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
1892 * Stuck with malloc/realloc.
1893 * For inline data, the underlying buffer must be
1894 * a multiple of 4 bytes in size so that it can be
1895 * logged and stay on word boundaries. We enforce
1898 real_size
= roundup(new_size
, 4);
1899 if (ifp
->if_u1
.if_data
== NULL
) {
1900 ASSERT(ifp
->if_real_bytes
== 0);
1901 ifp
->if_u1
.if_data
= kmem_alloc(real_size
,
1902 KM_SLEEP
| KM_NOFS
);
1903 } else if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
1905 * Only do the realloc if the underlying size
1906 * is really changing.
1908 if (ifp
->if_real_bytes
!= real_size
) {
1909 ifp
->if_u1
.if_data
=
1910 kmem_realloc(ifp
->if_u1
.if_data
,
1913 KM_SLEEP
| KM_NOFS
);
1916 ASSERT(ifp
->if_real_bytes
== 0);
1917 ifp
->if_u1
.if_data
= kmem_alloc(real_size
,
1918 KM_SLEEP
| KM_NOFS
);
1919 memcpy(ifp
->if_u1
.if_data
, ifp
->if_u2
.if_inline_data
,
1923 ifp
->if_real_bytes
= real_size
;
1924 ifp
->if_bytes
= new_size
;
1925 ASSERT(ifp
->if_bytes
<= XFS_IFORK_SIZE(ip
, whichfork
));
1935 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1936 if (ifp
->if_broot
!= NULL
) {
1937 kmem_free(ifp
->if_broot
);
1938 ifp
->if_broot
= NULL
;
1942 * If the format is local, then we can't have an extents
1943 * array so just look for an inline data array. If we're
1944 * not local then we may or may not have an extents list,
1945 * so check and free it up if we do.
1947 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
) {
1948 if ((ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) &&
1949 (ifp
->if_u1
.if_data
!= NULL
)) {
1950 ASSERT(ifp
->if_real_bytes
!= 0);
1951 kmem_free(ifp
->if_u1
.if_data
);
1952 ifp
->if_u1
.if_data
= NULL
;
1953 ifp
->if_real_bytes
= 0;
1955 } else if ((ifp
->if_flags
& XFS_IFEXTENTS
) &&
1956 ((ifp
->if_flags
& XFS_IFEXTIREC
) ||
1957 ((ifp
->if_u1
.if_extents
!= NULL
) &&
1958 (ifp
->if_u1
.if_extents
!= ifp
->if_u2
.if_inline_ext
)))) {
1959 ASSERT(ifp
->if_real_bytes
!= 0);
1960 xfs_iext_destroy(ifp
);
1962 ASSERT(ifp
->if_u1
.if_extents
== NULL
||
1963 ifp
->if_u1
.if_extents
== ifp
->if_u2
.if_inline_ext
);
1964 ASSERT(ifp
->if_real_bytes
== 0);
1965 if (whichfork
== XFS_ATTR_FORK
) {
1966 kmem_zone_free(xfs_ifork_zone
, ip
->i_afp
);
1972 * This is called to unpin an inode. The caller must have the inode locked
1973 * in at least shared mode so that the buffer cannot be subsequently pinned
1974 * once someone is waiting for it to be unpinned.
1978 struct xfs_inode
*ip
)
1980 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
1982 trace_xfs_inode_unpin_nowait(ip
, _RET_IP_
);
1984 /* Give the log a push to start the unpinning I/O */
1985 xfs_log_force_lsn(ip
->i_mount
, ip
->i_itemp
->ili_last_lsn
, 0);
1991 struct xfs_inode
*ip
)
1993 wait_queue_head_t
*wq
= bit_waitqueue(&ip
->i_flags
, __XFS_IPINNED_BIT
);
1994 DEFINE_WAIT_BIT(wait
, &ip
->i_flags
, __XFS_IPINNED_BIT
);
1999 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
2000 if (xfs_ipincount(ip
))
2002 } while (xfs_ipincount(ip
));
2003 finish_wait(wq
, &wait
.wait
);
2008 struct xfs_inode
*ip
)
2010 if (xfs_ipincount(ip
))
2011 __xfs_iunpin_wait(ip
);
2015 * xfs_iextents_copy()
2017 * This is called to copy the REAL extents (as opposed to the delayed
2018 * allocation extents) from the inode into the given buffer. It
2019 * returns the number of bytes copied into the buffer.
2021 * If there are no delayed allocation extents, then we can just
2022 * memcpy() the extents into the buffer. Otherwise, we need to
2023 * examine each extent in turn and skip those which are delayed.
2035 xfs_fsblock_t start_block
;
2037 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2038 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2039 ASSERT(ifp
->if_bytes
> 0);
2041 nrecs
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
2042 XFS_BMAP_TRACE_EXLIST(ip
, nrecs
, whichfork
);
2046 * There are some delayed allocation extents in the
2047 * inode, so copy the extents one at a time and skip
2048 * the delayed ones. There must be at least one
2049 * non-delayed extent.
2052 for (i
= 0; i
< nrecs
; i
++) {
2053 xfs_bmbt_rec_host_t
*ep
= xfs_iext_get_ext(ifp
, i
);
2054 start_block
= xfs_bmbt_get_startblock(ep
);
2055 if (isnullstartblock(start_block
)) {
2057 * It's a delayed allocation extent, so skip it.
2062 /* Translate to on disk format */
2063 put_unaligned(cpu_to_be64(ep
->l0
), &dp
->l0
);
2064 put_unaligned(cpu_to_be64(ep
->l1
), &dp
->l1
);
2068 ASSERT(copied
!= 0);
2069 xfs_validate_extents(ifp
, copied
, XFS_EXTFMT_INODE(ip
));
2071 return (copied
* (uint
)sizeof(xfs_bmbt_rec_t
));
2075 * Each of the following cases stores data into the same region
2076 * of the on-disk inode, so only one of them can be valid at
2077 * any given time. While it is possible to have conflicting formats
2078 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2079 * in EXTENTS format, this can only happen when the fork has
2080 * changed formats after being modified but before being flushed.
2081 * In these cases, the format always takes precedence, because the
2082 * format indicates the current state of the fork.
2089 xfs_inode_log_item_t
*iip
,
2096 #ifdef XFS_TRANS_DEBUG
2099 static const short brootflag
[2] =
2100 { XFS_ILOG_DBROOT
, XFS_ILOG_ABROOT
};
2101 static const short dataflag
[2] =
2102 { XFS_ILOG_DDATA
, XFS_ILOG_ADATA
};
2103 static const short extflag
[2] =
2104 { XFS_ILOG_DEXT
, XFS_ILOG_AEXT
};
2108 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2110 * This can happen if we gave up in iformat in an error path,
2111 * for the attribute fork.
2114 ASSERT(whichfork
== XFS_ATTR_FORK
);
2117 cp
= XFS_DFORK_PTR(dip
, whichfork
);
2119 switch (XFS_IFORK_FORMAT(ip
, whichfork
)) {
2120 case XFS_DINODE_FMT_LOCAL
:
2121 if ((iip
->ili_fields
& dataflag
[whichfork
]) &&
2122 (ifp
->if_bytes
> 0)) {
2123 ASSERT(ifp
->if_u1
.if_data
!= NULL
);
2124 ASSERT(ifp
->if_bytes
<= XFS_IFORK_SIZE(ip
, whichfork
));
2125 memcpy(cp
, ifp
->if_u1
.if_data
, ifp
->if_bytes
);
2129 case XFS_DINODE_FMT_EXTENTS
:
2130 ASSERT((ifp
->if_flags
& XFS_IFEXTENTS
) ||
2131 !(iip
->ili_fields
& extflag
[whichfork
]));
2132 if ((iip
->ili_fields
& extflag
[whichfork
]) &&
2133 (ifp
->if_bytes
> 0)) {
2134 ASSERT(xfs_iext_get_ext(ifp
, 0));
2135 ASSERT(XFS_IFORK_NEXTENTS(ip
, whichfork
) > 0);
2136 (void)xfs_iextents_copy(ip
, (xfs_bmbt_rec_t
*)cp
,
2141 case XFS_DINODE_FMT_BTREE
:
2142 if ((iip
->ili_fields
& brootflag
[whichfork
]) &&
2143 (ifp
->if_broot_bytes
> 0)) {
2144 ASSERT(ifp
->if_broot
!= NULL
);
2145 ASSERT(ifp
->if_broot_bytes
<=
2146 (XFS_IFORK_SIZE(ip
, whichfork
) +
2147 XFS_BROOT_SIZE_ADJ
));
2148 xfs_bmbt_to_bmdr(mp
, ifp
->if_broot
, ifp
->if_broot_bytes
,
2149 (xfs_bmdr_block_t
*)cp
,
2150 XFS_DFORK_SIZE(dip
, mp
, whichfork
));
2154 case XFS_DINODE_FMT_DEV
:
2155 if (iip
->ili_fields
& XFS_ILOG_DEV
) {
2156 ASSERT(whichfork
== XFS_DATA_FORK
);
2157 xfs_dinode_put_rdev(dip
, ip
->i_df
.if_u2
.if_rdev
);
2161 case XFS_DINODE_FMT_UUID
:
2162 if (iip
->ili_fields
& XFS_ILOG_UUID
) {
2163 ASSERT(whichfork
== XFS_DATA_FORK
);
2164 memcpy(XFS_DFORK_DPTR(dip
),
2165 &ip
->i_df
.if_u2
.if_uuid
,
2181 xfs_mount_t
*mp
= ip
->i_mount
;
2182 struct xfs_perag
*pag
;
2183 unsigned long first_index
, mask
;
2184 unsigned long inodes_per_cluster
;
2186 xfs_inode_t
**ilist
;
2193 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ip
->i_ino
));
2195 inodes_per_cluster
= XFS_INODE_CLUSTER_SIZE(mp
) >> mp
->m_sb
.sb_inodelog
;
2196 ilist_size
= inodes_per_cluster
* sizeof(xfs_inode_t
*);
2197 ilist
= kmem_alloc(ilist_size
, KM_MAYFAIL
|KM_NOFS
);
2201 mask
= ~(((XFS_INODE_CLUSTER_SIZE(mp
) >> mp
->m_sb
.sb_inodelog
)) - 1);
2202 first_index
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
) & mask
;
2204 /* really need a gang lookup range call here */
2205 nr_found
= radix_tree_gang_lookup(&pag
->pag_ici_root
, (void**)ilist
,
2206 first_index
, inodes_per_cluster
);
2210 for (i
= 0; i
< nr_found
; i
++) {
2216 * because this is an RCU protected lookup, we could find a
2217 * recently freed or even reallocated inode during the lookup.
2218 * We need to check under the i_flags_lock for a valid inode
2219 * here. Skip it if it is not valid or the wrong inode.
2221 spin_lock(&ip
->i_flags_lock
);
2223 (XFS_INO_TO_AGINO(mp
, iq
->i_ino
) & mask
) != first_index
) {
2224 spin_unlock(&ip
->i_flags_lock
);
2227 spin_unlock(&ip
->i_flags_lock
);
2230 * Do an un-protected check to see if the inode is dirty and
2231 * is a candidate for flushing. These checks will be repeated
2232 * later after the appropriate locks are acquired.
2234 if (xfs_inode_clean(iq
) && xfs_ipincount(iq
) == 0)
2238 * Try to get locks. If any are unavailable or it is pinned,
2239 * then this inode cannot be flushed and is skipped.
2242 if (!xfs_ilock_nowait(iq
, XFS_ILOCK_SHARED
))
2244 if (!xfs_iflock_nowait(iq
)) {
2245 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
2248 if (xfs_ipincount(iq
)) {
2250 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
2255 * arriving here means that this inode can be flushed. First
2256 * re-check that it's dirty before flushing.
2258 if (!xfs_inode_clean(iq
)) {
2260 error
= xfs_iflush_int(iq
, bp
);
2262 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
2263 goto cluster_corrupt_out
;
2269 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
2273 XFS_STATS_INC(xs_icluster_flushcnt
);
2274 XFS_STATS_ADD(xs_icluster_flushinode
, clcount
);
2285 cluster_corrupt_out
:
2287 * Corruption detected in the clustering loop. Invalidate the
2288 * inode buffer and shut down the filesystem.
2292 * Clean up the buffer. If it was delwri, just release it --
2293 * brelse can handle it with no problems. If not, shut down the
2294 * filesystem before releasing the buffer.
2296 bufwasdelwri
= (bp
->b_flags
& _XBF_DELWRI_Q
);
2300 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
2302 if (!bufwasdelwri
) {
2304 * Just like incore_relse: if we have b_iodone functions,
2305 * mark the buffer as an error and call them. Otherwise
2306 * mark it as stale and brelse.
2311 xfs_buf_ioerror(bp
, EIO
);
2312 xfs_buf_ioend(bp
, 0);
2320 * Unlocks the flush lock
2322 xfs_iflush_abort(iq
, false);
2325 return XFS_ERROR(EFSCORRUPTED
);
2329 * Flush dirty inode metadata into the backing buffer.
2331 * The caller must have the inode lock and the inode flush lock held. The
2332 * inode lock will still be held upon return to the caller, and the inode
2333 * flush lock will be released after the inode has reached the disk.
2335 * The caller must write out the buffer returned in *bpp and release it.
2339 struct xfs_inode
*ip
,
2340 struct xfs_buf
**bpp
)
2342 struct xfs_mount
*mp
= ip
->i_mount
;
2344 struct xfs_dinode
*dip
;
2347 XFS_STATS_INC(xs_iflush_count
);
2349 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2350 ASSERT(xfs_isiflocked(ip
));
2351 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
2352 ip
->i_d
.di_nextents
> XFS_IFORK_MAXEXT(ip
, XFS_DATA_FORK
));
2356 xfs_iunpin_wait(ip
);
2359 * For stale inodes we cannot rely on the backing buffer remaining
2360 * stale in cache for the remaining life of the stale inode and so
2361 * xfs_imap_to_bp() below may give us a buffer that no longer contains
2362 * inodes below. We have to check this after ensuring the inode is
2363 * unpinned so that it is safe to reclaim the stale inode after the
2366 if (xfs_iflags_test(ip
, XFS_ISTALE
)) {
2372 * This may have been unpinned because the filesystem is shutting
2373 * down forcibly. If that's the case we must not write this inode
2374 * to disk, because the log record didn't make it to disk.
2376 * We also have to remove the log item from the AIL in this case,
2377 * as we wait for an empty AIL as part of the unmount process.
2379 if (XFS_FORCED_SHUTDOWN(mp
)) {
2380 error
= XFS_ERROR(EIO
);
2385 * Get the buffer containing the on-disk inode.
2387 error
= xfs_imap_to_bp(mp
, NULL
, &ip
->i_imap
, &dip
, &bp
, XBF_TRYLOCK
,
2395 * First flush out the inode that xfs_iflush was called with.
2397 error
= xfs_iflush_int(ip
, bp
);
2402 * If the buffer is pinned then push on the log now so we won't
2403 * get stuck waiting in the write for too long.
2405 if (xfs_buf_ispinned(bp
))
2406 xfs_log_force(mp
, 0);
2410 * see if other inodes can be gathered into this write
2412 error
= xfs_iflush_cluster(ip
, bp
);
2414 goto cluster_corrupt_out
;
2421 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
2422 cluster_corrupt_out
:
2423 error
= XFS_ERROR(EFSCORRUPTED
);
2426 * Unlocks the flush lock
2428 xfs_iflush_abort(ip
, false);
2438 xfs_inode_log_item_t
*iip
;
2441 #ifdef XFS_TRANS_DEBUG
2445 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
));
2446 ASSERT(xfs_isiflocked(ip
));
2447 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
2448 ip
->i_d
.di_nextents
> XFS_IFORK_MAXEXT(ip
, XFS_DATA_FORK
));
2453 /* set *dip = inode's place in the buffer */
2454 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, ip
->i_imap
.im_boffset
);
2456 if (XFS_TEST_ERROR(dip
->di_magic
!= cpu_to_be16(XFS_DINODE_MAGIC
),
2457 mp
, XFS_ERRTAG_IFLUSH_1
, XFS_RANDOM_IFLUSH_1
)) {
2458 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
2459 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
2460 __func__
, ip
->i_ino
, be16_to_cpu(dip
->di_magic
), dip
);
2463 if (XFS_TEST_ERROR(ip
->i_d
.di_magic
!= XFS_DINODE_MAGIC
,
2464 mp
, XFS_ERRTAG_IFLUSH_2
, XFS_RANDOM_IFLUSH_2
)) {
2465 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
2466 "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
2467 __func__
, ip
->i_ino
, ip
, ip
->i_d
.di_magic
);
2470 if (S_ISREG(ip
->i_d
.di_mode
)) {
2472 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2473 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
),
2474 mp
, XFS_ERRTAG_IFLUSH_3
, XFS_RANDOM_IFLUSH_3
)) {
2475 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
2476 "%s: Bad regular inode %Lu, ptr 0x%p",
2477 __func__
, ip
->i_ino
, ip
);
2480 } else if (S_ISDIR(ip
->i_d
.di_mode
)) {
2482 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2483 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
) &&
2484 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_LOCAL
),
2485 mp
, XFS_ERRTAG_IFLUSH_4
, XFS_RANDOM_IFLUSH_4
)) {
2486 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
2487 "%s: Bad directory inode %Lu, ptr 0x%p",
2488 __func__
, ip
->i_ino
, ip
);
2492 if (XFS_TEST_ERROR(ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
>
2493 ip
->i_d
.di_nblocks
, mp
, XFS_ERRTAG_IFLUSH_5
,
2494 XFS_RANDOM_IFLUSH_5
)) {
2495 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
2496 "%s: detected corrupt incore inode %Lu, "
2497 "total extents = %d, nblocks = %Ld, ptr 0x%p",
2498 __func__
, ip
->i_ino
,
2499 ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
,
2500 ip
->i_d
.di_nblocks
, ip
);
2503 if (XFS_TEST_ERROR(ip
->i_d
.di_forkoff
> mp
->m_sb
.sb_inodesize
,
2504 mp
, XFS_ERRTAG_IFLUSH_6
, XFS_RANDOM_IFLUSH_6
)) {
2505 xfs_alert_tag(mp
, XFS_PTAG_IFLUSH
,
2506 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
2507 __func__
, ip
->i_ino
, ip
->i_d
.di_forkoff
, ip
);
2511 * bump the flush iteration count, used to detect flushes which
2512 * postdate a log record during recovery.
2515 ip
->i_d
.di_flushiter
++;
2518 * Copy the dirty parts of the inode into the on-disk
2519 * inode. We always copy out the core of the inode,
2520 * because if the inode is dirty at all the core must
2523 xfs_dinode_to_disk(dip
, &ip
->i_d
);
2525 /* Wrap, we never let the log put out DI_MAX_FLUSH */
2526 if (ip
->i_d
.di_flushiter
== DI_MAX_FLUSH
)
2527 ip
->i_d
.di_flushiter
= 0;
2530 * If this is really an old format inode and the superblock version
2531 * has not been updated to support only new format inodes, then
2532 * convert back to the old inode format. If the superblock version
2533 * has been updated, then make the conversion permanent.
2535 ASSERT(ip
->i_d
.di_version
== 1 || xfs_sb_version_hasnlink(&mp
->m_sb
));
2536 if (ip
->i_d
.di_version
== 1) {
2537 if (!xfs_sb_version_hasnlink(&mp
->m_sb
)) {
2541 ASSERT(ip
->i_d
.di_nlink
<= XFS_MAXLINK_1
);
2542 dip
->di_onlink
= cpu_to_be16(ip
->i_d
.di_nlink
);
2545 * The superblock version has already been bumped,
2546 * so just make the conversion to the new inode
2549 ip
->i_d
.di_version
= 2;
2550 dip
->di_version
= 2;
2551 ip
->i_d
.di_onlink
= 0;
2553 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
2554 memset(&(dip
->di_pad
[0]), 0,
2555 sizeof(dip
->di_pad
));
2556 ASSERT(xfs_get_projid(ip
) == 0);
2560 xfs_iflush_fork(ip
, dip
, iip
, XFS_DATA_FORK
, bp
);
2561 if (XFS_IFORK_Q(ip
))
2562 xfs_iflush_fork(ip
, dip
, iip
, XFS_ATTR_FORK
, bp
);
2563 xfs_inobp_check(mp
, bp
);
2566 * We've recorded everything logged in the inode, so we'd like to clear
2567 * the ili_fields bits so we don't log and flush things unnecessarily.
2568 * However, we can't stop logging all this information until the data
2569 * we've copied into the disk buffer is written to disk. If we did we
2570 * might overwrite the copy of the inode in the log with all the data
2571 * after re-logging only part of it, and in the face of a crash we
2572 * wouldn't have all the data we need to recover.
2574 * What we do is move the bits to the ili_last_fields field. When
2575 * logging the inode, these bits are moved back to the ili_fields field.
2576 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
2577 * know that the information those bits represent is permanently on
2578 * disk. As long as the flush completes before the inode is logged
2579 * again, then both ili_fields and ili_last_fields will be cleared.
2581 * We can play with the ili_fields bits here, because the inode lock
2582 * must be held exclusively in order to set bits there and the flush
2583 * lock protects the ili_last_fields bits. Set ili_logged so the flush
2584 * done routine can tell whether or not to look in the AIL. Also, store
2585 * the current LSN of the inode so that we can tell whether the item has
2586 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
2587 * need the AIL lock, because it is a 64 bit value that cannot be read
2590 if (iip
!= NULL
&& iip
->ili_fields
!= 0) {
2591 iip
->ili_last_fields
= iip
->ili_fields
;
2592 iip
->ili_fields
= 0;
2593 iip
->ili_logged
= 1;
2595 xfs_trans_ail_copy_lsn(mp
->m_ail
, &iip
->ili_flush_lsn
,
2596 &iip
->ili_item
.li_lsn
);
2599 * Attach the function xfs_iflush_done to the inode's
2600 * buffer. This will remove the inode from the AIL
2601 * and unlock the inode's flush lock when the inode is
2602 * completely written to disk.
2604 xfs_buf_attach_iodone(bp
, xfs_iflush_done
, &iip
->ili_item
);
2606 ASSERT(bp
->b_fspriv
!= NULL
);
2607 ASSERT(bp
->b_iodone
!= NULL
);
2610 * We're flushing an inode which is not in the AIL and has
2611 * not been logged. For this case we can immediately drop
2612 * the inode flush lock because we can avoid the whole
2613 * AIL state thing. It's OK to drop the flush lock now,
2614 * because we've already locked the buffer and to do anything
2615 * you really need both.
2618 ASSERT(iip
->ili_logged
== 0);
2619 ASSERT(iip
->ili_last_fields
== 0);
2620 ASSERT((iip
->ili_item
.li_flags
& XFS_LI_IN_AIL
) == 0);
2628 return XFS_ERROR(EFSCORRUPTED
);
2632 * Return a pointer to the extent record at file index idx.
2634 xfs_bmbt_rec_host_t
*
2636 xfs_ifork_t
*ifp
, /* inode fork pointer */
2637 xfs_extnum_t idx
) /* index of target extent */
2640 ASSERT(idx
< ifp
->if_bytes
/ sizeof(xfs_bmbt_rec_t
));
2642 if ((ifp
->if_flags
& XFS_IFEXTIREC
) && (idx
== 0)) {
2643 return ifp
->if_u1
.if_ext_irec
->er_extbuf
;
2644 } else if (ifp
->if_flags
& XFS_IFEXTIREC
) {
2645 xfs_ext_irec_t
*erp
; /* irec pointer */
2646 int erp_idx
= 0; /* irec index */
2647 xfs_extnum_t page_idx
= idx
; /* ext index in target list */
2649 erp
= xfs_iext_idx_to_irec(ifp
, &page_idx
, &erp_idx
, 0);
2650 return &erp
->er_extbuf
[page_idx
];
2651 } else if (ifp
->if_bytes
) {
2652 return &ifp
->if_u1
.if_extents
[idx
];
2659 * Insert new item(s) into the extent records for incore inode
2660 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
2664 xfs_inode_t
*ip
, /* incore inode pointer */
2665 xfs_extnum_t idx
, /* starting index of new items */
2666 xfs_extnum_t count
, /* number of inserted items */
2667 xfs_bmbt_irec_t
*new, /* items to insert */
2668 int state
) /* type of extent conversion */
2670 xfs_ifork_t
*ifp
= (state
& BMAP_ATTRFORK
) ? ip
->i_afp
: &ip
->i_df
;
2671 xfs_extnum_t i
; /* extent record index */
2673 trace_xfs_iext_insert(ip
, idx
, new, state
, _RET_IP_
);
2675 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
2676 xfs_iext_add(ifp
, idx
, count
);
2677 for (i
= idx
; i
< idx
+ count
; i
++, new++)
2678 xfs_bmbt_set_all(xfs_iext_get_ext(ifp
, i
), new);
2682 * This is called when the amount of space required for incore file
2683 * extents needs to be increased. The ext_diff parameter stores the
2684 * number of new extents being added and the idx parameter contains
2685 * the extent index where the new extents will be added. If the new
2686 * extents are being appended, then we just need to (re)allocate and
2687 * initialize the space. Otherwise, if the new extents are being
2688 * inserted into the middle of the existing entries, a bit more work
2689 * is required to make room for the new extents to be inserted. The
2690 * caller is responsible for filling in the new extent entries upon
2695 xfs_ifork_t
*ifp
, /* inode fork pointer */
2696 xfs_extnum_t idx
, /* index to begin adding exts */
2697 int ext_diff
) /* number of extents to add */
2699 int byte_diff
; /* new bytes being added */
2700 int new_size
; /* size of extents after adding */
2701 xfs_extnum_t nextents
; /* number of extents in file */
2703 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
2704 ASSERT((idx
>= 0) && (idx
<= nextents
));
2705 byte_diff
= ext_diff
* sizeof(xfs_bmbt_rec_t
);
2706 new_size
= ifp
->if_bytes
+ byte_diff
;
2708 * If the new number of extents (nextents + ext_diff)
2709 * fits inside the inode, then continue to use the inline
2712 if (nextents
+ ext_diff
<= XFS_INLINE_EXTS
) {
2713 if (idx
< nextents
) {
2714 memmove(&ifp
->if_u2
.if_inline_ext
[idx
+ ext_diff
],
2715 &ifp
->if_u2
.if_inline_ext
[idx
],
2716 (nextents
- idx
) * sizeof(xfs_bmbt_rec_t
));
2717 memset(&ifp
->if_u2
.if_inline_ext
[idx
], 0, byte_diff
);
2719 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
2720 ifp
->if_real_bytes
= 0;
2723 * Otherwise use a linear (direct) extent list.
2724 * If the extents are currently inside the inode,
2725 * xfs_iext_realloc_direct will switch us from
2726 * inline to direct extent allocation mode.
2728 else if (nextents
+ ext_diff
<= XFS_LINEAR_EXTS
) {
2729 xfs_iext_realloc_direct(ifp
, new_size
);
2730 if (idx
< nextents
) {
2731 memmove(&ifp
->if_u1
.if_extents
[idx
+ ext_diff
],
2732 &ifp
->if_u1
.if_extents
[idx
],
2733 (nextents
- idx
) * sizeof(xfs_bmbt_rec_t
));
2734 memset(&ifp
->if_u1
.if_extents
[idx
], 0, byte_diff
);
2737 /* Indirection array */
2739 xfs_ext_irec_t
*erp
;
2743 ASSERT(nextents
+ ext_diff
> XFS_LINEAR_EXTS
);
2744 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
2745 erp
= xfs_iext_idx_to_irec(ifp
, &page_idx
, &erp_idx
, 1);
2747 xfs_iext_irec_init(ifp
);
2748 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
2749 erp
= ifp
->if_u1
.if_ext_irec
;
2751 /* Extents fit in target extent page */
2752 if (erp
&& erp
->er_extcount
+ ext_diff
<= XFS_LINEAR_EXTS
) {
2753 if (page_idx
< erp
->er_extcount
) {
2754 memmove(&erp
->er_extbuf
[page_idx
+ ext_diff
],
2755 &erp
->er_extbuf
[page_idx
],
2756 (erp
->er_extcount
- page_idx
) *
2757 sizeof(xfs_bmbt_rec_t
));
2758 memset(&erp
->er_extbuf
[page_idx
], 0, byte_diff
);
2760 erp
->er_extcount
+= ext_diff
;
2761 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, ext_diff
);
2763 /* Insert a new extent page */
2765 xfs_iext_add_indirect_multi(ifp
,
2766 erp_idx
, page_idx
, ext_diff
);
2769 * If extent(s) are being appended to the last page in
2770 * the indirection array and the new extent(s) don't fit
2771 * in the page, then erp is NULL and erp_idx is set to
2772 * the next index needed in the indirection array.
2775 int count
= ext_diff
;
2778 erp
= xfs_iext_irec_new(ifp
, erp_idx
);
2779 erp
->er_extcount
= count
;
2780 count
-= MIN(count
, (int)XFS_LINEAR_EXTS
);
2787 ifp
->if_bytes
= new_size
;
2791 * This is called when incore extents are being added to the indirection
2792 * array and the new extents do not fit in the target extent list. The
2793 * erp_idx parameter contains the irec index for the target extent list
2794 * in the indirection array, and the idx parameter contains the extent
2795 * index within the list. The number of extents being added is stored
2796 * in the count parameter.
2798 * |-------| |-------|
2799 * | | | | idx - number of extents before idx
2801 * | | | | count - number of extents being inserted at idx
2802 * |-------| |-------|
2803 * | count | | nex2 | nex2 - number of extents after idx + count
2804 * |-------| |-------|
2807 xfs_iext_add_indirect_multi(
2808 xfs_ifork_t
*ifp
, /* inode fork pointer */
2809 int erp_idx
, /* target extent irec index */
2810 xfs_extnum_t idx
, /* index within target list */
2811 int count
) /* new extents being added */
2813 int byte_diff
; /* new bytes being added */
2814 xfs_ext_irec_t
*erp
; /* pointer to irec entry */
2815 xfs_extnum_t ext_diff
; /* number of extents to add */
2816 xfs_extnum_t ext_cnt
; /* new extents still needed */
2817 xfs_extnum_t nex2
; /* extents after idx + count */
2818 xfs_bmbt_rec_t
*nex2_ep
= NULL
; /* temp list for nex2 extents */
2819 int nlists
; /* number of irec's (lists) */
2821 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
2822 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
2823 nex2
= erp
->er_extcount
- idx
;
2824 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
2827 * Save second part of target extent list
2828 * (all extents past */
2830 byte_diff
= nex2
* sizeof(xfs_bmbt_rec_t
);
2831 nex2_ep
= (xfs_bmbt_rec_t
*) kmem_alloc(byte_diff
, KM_NOFS
);
2832 memmove(nex2_ep
, &erp
->er_extbuf
[idx
], byte_diff
);
2833 erp
->er_extcount
-= nex2
;
2834 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, -nex2
);
2835 memset(&erp
->er_extbuf
[idx
], 0, byte_diff
);
2839 * Add the new extents to the end of the target
2840 * list, then allocate new irec record(s) and
2841 * extent buffer(s) as needed to store the rest
2842 * of the new extents.
2845 ext_diff
= MIN(ext_cnt
, (int)XFS_LINEAR_EXTS
- erp
->er_extcount
);
2847 erp
->er_extcount
+= ext_diff
;
2848 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, ext_diff
);
2849 ext_cnt
-= ext_diff
;
2853 erp
= xfs_iext_irec_new(ifp
, erp_idx
);
2854 ext_diff
= MIN(ext_cnt
, (int)XFS_LINEAR_EXTS
);
2855 erp
->er_extcount
= ext_diff
;
2856 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, ext_diff
);
2857 ext_cnt
-= ext_diff
;
2860 /* Add nex2 extents back to indirection array */
2862 xfs_extnum_t ext_avail
;
2865 byte_diff
= nex2
* sizeof(xfs_bmbt_rec_t
);
2866 ext_avail
= XFS_LINEAR_EXTS
- erp
->er_extcount
;
2869 * If nex2 extents fit in the current page, append
2870 * nex2_ep after the new extents.
2872 if (nex2
<= ext_avail
) {
2873 i
= erp
->er_extcount
;
2876 * Otherwise, check if space is available in the
2879 else if ((erp_idx
< nlists
- 1) &&
2880 (nex2
<= (ext_avail
= XFS_LINEAR_EXTS
-
2881 ifp
->if_u1
.if_ext_irec
[erp_idx
+1].er_extcount
))) {
2884 /* Create a hole for nex2 extents */
2885 memmove(&erp
->er_extbuf
[nex2
], erp
->er_extbuf
,
2886 erp
->er_extcount
* sizeof(xfs_bmbt_rec_t
));
2889 * Final choice, create a new extent page for
2894 erp
= xfs_iext_irec_new(ifp
, erp_idx
);
2896 memmove(&erp
->er_extbuf
[i
], nex2_ep
, byte_diff
);
2898 erp
->er_extcount
+= nex2
;
2899 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, nex2
);
2904 * This is called when the amount of space required for incore file
2905 * extents needs to be decreased. The ext_diff parameter stores the
2906 * number of extents to be removed and the idx parameter contains
2907 * the extent index where the extents will be removed from.
2909 * If the amount of space needed has decreased below the linear
2910 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
2911 * extent array. Otherwise, use kmem_realloc() to adjust the
2912 * size to what is needed.
2916 xfs_inode_t
*ip
, /* incore inode pointer */
2917 xfs_extnum_t idx
, /* index to begin removing exts */
2918 int ext_diff
, /* number of extents to remove */
2919 int state
) /* type of extent conversion */
2921 xfs_ifork_t
*ifp
= (state
& BMAP_ATTRFORK
) ? ip
->i_afp
: &ip
->i_df
;
2922 xfs_extnum_t nextents
; /* number of extents in file */
2923 int new_size
; /* size of extents after removal */
2925 trace_xfs_iext_remove(ip
, idx
, state
, _RET_IP_
);
2927 ASSERT(ext_diff
> 0);
2928 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
2929 new_size
= (nextents
- ext_diff
) * sizeof(xfs_bmbt_rec_t
);
2931 if (new_size
== 0) {
2932 xfs_iext_destroy(ifp
);
2933 } else if (ifp
->if_flags
& XFS_IFEXTIREC
) {
2934 xfs_iext_remove_indirect(ifp
, idx
, ext_diff
);
2935 } else if (ifp
->if_real_bytes
) {
2936 xfs_iext_remove_direct(ifp
, idx
, ext_diff
);
2938 xfs_iext_remove_inline(ifp
, idx
, ext_diff
);
2940 ifp
->if_bytes
= new_size
;
2944 * This removes ext_diff extents from the inline buffer, beginning
2945 * at extent index idx.
2948 xfs_iext_remove_inline(
2949 xfs_ifork_t
*ifp
, /* inode fork pointer */
2950 xfs_extnum_t idx
, /* index to begin removing exts */
2951 int ext_diff
) /* number of extents to remove */
2953 int nextents
; /* number of extents in file */
2955 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
));
2956 ASSERT(idx
< XFS_INLINE_EXTS
);
2957 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
2958 ASSERT(((nextents
- ext_diff
) > 0) &&
2959 (nextents
- ext_diff
) < XFS_INLINE_EXTS
);
2961 if (idx
+ ext_diff
< nextents
) {
2962 memmove(&ifp
->if_u2
.if_inline_ext
[idx
],
2963 &ifp
->if_u2
.if_inline_ext
[idx
+ ext_diff
],
2964 (nextents
- (idx
+ ext_diff
)) *
2965 sizeof(xfs_bmbt_rec_t
));
2966 memset(&ifp
->if_u2
.if_inline_ext
[nextents
- ext_diff
],
2967 0, ext_diff
* sizeof(xfs_bmbt_rec_t
));
2969 memset(&ifp
->if_u2
.if_inline_ext
[idx
], 0,
2970 ext_diff
* sizeof(xfs_bmbt_rec_t
));
2975 * This removes ext_diff extents from a linear (direct) extent list,
2976 * beginning at extent index idx. If the extents are being removed
2977 * from the end of the list (ie. truncate) then we just need to re-
2978 * allocate the list to remove the extra space. Otherwise, if the
2979 * extents are being removed from the middle of the existing extent
2980 * entries, then we first need to move the extent records beginning
2981 * at idx + ext_diff up in the list to overwrite the records being
2982 * removed, then remove the extra space via kmem_realloc.
2985 xfs_iext_remove_direct(
2986 xfs_ifork_t
*ifp
, /* inode fork pointer */
2987 xfs_extnum_t idx
, /* index to begin removing exts */
2988 int ext_diff
) /* number of extents to remove */
2990 xfs_extnum_t nextents
; /* number of extents in file */
2991 int new_size
; /* size of extents after removal */
2993 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
));
2994 new_size
= ifp
->if_bytes
-
2995 (ext_diff
* sizeof(xfs_bmbt_rec_t
));
2996 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
2998 if (new_size
== 0) {
2999 xfs_iext_destroy(ifp
);
3002 /* Move extents up in the list (if needed) */
3003 if (idx
+ ext_diff
< nextents
) {
3004 memmove(&ifp
->if_u1
.if_extents
[idx
],
3005 &ifp
->if_u1
.if_extents
[idx
+ ext_diff
],
3006 (nextents
- (idx
+ ext_diff
)) *
3007 sizeof(xfs_bmbt_rec_t
));
3009 memset(&ifp
->if_u1
.if_extents
[nextents
- ext_diff
],
3010 0, ext_diff
* sizeof(xfs_bmbt_rec_t
));
3012 * Reallocate the direct extent list. If the extents
3013 * will fit inside the inode then xfs_iext_realloc_direct
3014 * will switch from direct to inline extent allocation
3017 xfs_iext_realloc_direct(ifp
, new_size
);
3018 ifp
->if_bytes
= new_size
;
3022 * This is called when incore extents are being removed from the
3023 * indirection array and the extents being removed span multiple extent
3024 * buffers. The idx parameter contains the file extent index where we
3025 * want to begin removing extents, and the count parameter contains
3026 * how many extents need to be removed.
3028 * |-------| |-------|
3029 * | nex1 | | | nex1 - number of extents before idx
3030 * |-------| | count |
3031 * | | | | count - number of extents being removed at idx
3032 * | count | |-------|
3033 * | | | nex2 | nex2 - number of extents after idx + count
3034 * |-------| |-------|
3037 xfs_iext_remove_indirect(
3038 xfs_ifork_t
*ifp
, /* inode fork pointer */
3039 xfs_extnum_t idx
, /* index to begin removing extents */
3040 int count
) /* number of extents to remove */
3042 xfs_ext_irec_t
*erp
; /* indirection array pointer */
3043 int erp_idx
= 0; /* indirection array index */
3044 xfs_extnum_t ext_cnt
; /* extents left to remove */
3045 xfs_extnum_t ext_diff
; /* extents to remove in current list */
3046 xfs_extnum_t nex1
; /* number of extents before idx */
3047 xfs_extnum_t nex2
; /* extents after idx + count */
3048 int page_idx
= idx
; /* index in target extent list */
3050 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3051 erp
= xfs_iext_idx_to_irec(ifp
, &page_idx
, &erp_idx
, 0);
3052 ASSERT(erp
!= NULL
);
3056 nex2
= MAX((erp
->er_extcount
- (nex1
+ ext_cnt
)), 0);
3057 ext_diff
= MIN(ext_cnt
, (erp
->er_extcount
- nex1
));
3059 * Check for deletion of entire list;
3060 * xfs_iext_irec_remove() updates extent offsets.
3062 if (ext_diff
== erp
->er_extcount
) {
3063 xfs_iext_irec_remove(ifp
, erp_idx
);
3064 ext_cnt
-= ext_diff
;
3067 ASSERT(erp_idx
< ifp
->if_real_bytes
/
3069 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
3076 /* Move extents up (if needed) */
3078 memmove(&erp
->er_extbuf
[nex1
],
3079 &erp
->er_extbuf
[nex1
+ ext_diff
],
3080 nex2
* sizeof(xfs_bmbt_rec_t
));
3082 /* Zero out rest of page */
3083 memset(&erp
->er_extbuf
[nex1
+ nex2
], 0, (XFS_IEXT_BUFSZ
-
3084 ((nex1
+ nex2
) * sizeof(xfs_bmbt_rec_t
))));
3085 /* Update remaining counters */
3086 erp
->er_extcount
-= ext_diff
;
3087 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, -ext_diff
);
3088 ext_cnt
-= ext_diff
;
3093 ifp
->if_bytes
-= count
* sizeof(xfs_bmbt_rec_t
);
3094 xfs_iext_irec_compact(ifp
);
3098 * Create, destroy, or resize a linear (direct) block of extents.
3101 xfs_iext_realloc_direct(
3102 xfs_ifork_t
*ifp
, /* inode fork pointer */
3103 int new_size
) /* new size of extents */
3105 int rnew_size
; /* real new size of extents */
3107 rnew_size
= new_size
;
3109 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
) ||
3110 ((new_size
>= 0) && (new_size
<= XFS_IEXT_BUFSZ
) &&
3111 (new_size
!= ifp
->if_real_bytes
)));
3113 /* Free extent records */
3114 if (new_size
== 0) {
3115 xfs_iext_destroy(ifp
);
3117 /* Resize direct extent list and zero any new bytes */
3118 else if (ifp
->if_real_bytes
) {
3119 /* Check if extents will fit inside the inode */
3120 if (new_size
<= XFS_INLINE_EXTS
* sizeof(xfs_bmbt_rec_t
)) {
3121 xfs_iext_direct_to_inline(ifp
, new_size
/
3122 (uint
)sizeof(xfs_bmbt_rec_t
));
3123 ifp
->if_bytes
= new_size
;
3126 if (!is_power_of_2(new_size
)){
3127 rnew_size
= roundup_pow_of_two(new_size
);
3129 if (rnew_size
!= ifp
->if_real_bytes
) {
3130 ifp
->if_u1
.if_extents
=
3131 kmem_realloc(ifp
->if_u1
.if_extents
,
3133 ifp
->if_real_bytes
, KM_NOFS
);
3135 if (rnew_size
> ifp
->if_real_bytes
) {
3136 memset(&ifp
->if_u1
.if_extents
[ifp
->if_bytes
/
3137 (uint
)sizeof(xfs_bmbt_rec_t
)], 0,
3138 rnew_size
- ifp
->if_real_bytes
);
3142 * Switch from the inline extent buffer to a direct
3143 * extent list. Be sure to include the inline extent
3144 * bytes in new_size.
3147 new_size
+= ifp
->if_bytes
;
3148 if (!is_power_of_2(new_size
)) {
3149 rnew_size
= roundup_pow_of_two(new_size
);
3151 xfs_iext_inline_to_direct(ifp
, rnew_size
);
3153 ifp
->if_real_bytes
= rnew_size
;
3154 ifp
->if_bytes
= new_size
;
3158 * Switch from linear (direct) extent records to inline buffer.
3161 xfs_iext_direct_to_inline(
3162 xfs_ifork_t
*ifp
, /* inode fork pointer */
3163 xfs_extnum_t nextents
) /* number of extents in file */
3165 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
3166 ASSERT(nextents
<= XFS_INLINE_EXTS
);
3168 * The inline buffer was zeroed when we switched
3169 * from inline to direct extent allocation mode,
3170 * so we don't need to clear it here.
3172 memcpy(ifp
->if_u2
.if_inline_ext
, ifp
->if_u1
.if_extents
,
3173 nextents
* sizeof(xfs_bmbt_rec_t
));
3174 kmem_free(ifp
->if_u1
.if_extents
);
3175 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
3176 ifp
->if_real_bytes
= 0;
3180 * Switch from inline buffer to linear (direct) extent records.
3181 * new_size should already be rounded up to the next power of 2
3182 * by the caller (when appropriate), so use new_size as it is.
3183 * However, since new_size may be rounded up, we can't update
3184 * if_bytes here. It is the caller's responsibility to update
3185 * if_bytes upon return.
3188 xfs_iext_inline_to_direct(
3189 xfs_ifork_t
*ifp
, /* inode fork pointer */
3190 int new_size
) /* number of extents in file */
3192 ifp
->if_u1
.if_extents
= kmem_alloc(new_size
, KM_NOFS
);
3193 memset(ifp
->if_u1
.if_extents
, 0, new_size
);
3194 if (ifp
->if_bytes
) {
3195 memcpy(ifp
->if_u1
.if_extents
, ifp
->if_u2
.if_inline_ext
,
3197 memset(ifp
->if_u2
.if_inline_ext
, 0, XFS_INLINE_EXTS
*
3198 sizeof(xfs_bmbt_rec_t
));
3200 ifp
->if_real_bytes
= new_size
;
3204 * Resize an extent indirection array to new_size bytes.
3207 xfs_iext_realloc_indirect(
3208 xfs_ifork_t
*ifp
, /* inode fork pointer */
3209 int new_size
) /* new indirection array size */
3211 int nlists
; /* number of irec's (ex lists) */
3212 int size
; /* current indirection array size */
3214 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3215 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3216 size
= nlists
* sizeof(xfs_ext_irec_t
);
3217 ASSERT(ifp
->if_real_bytes
);
3218 ASSERT((new_size
>= 0) && (new_size
!= size
));
3219 if (new_size
== 0) {
3220 xfs_iext_destroy(ifp
);
3222 ifp
->if_u1
.if_ext_irec
= (xfs_ext_irec_t
*)
3223 kmem_realloc(ifp
->if_u1
.if_ext_irec
,
3224 new_size
, size
, KM_NOFS
);
3229 * Switch from indirection array to linear (direct) extent allocations.
3232 xfs_iext_indirect_to_direct(
3233 xfs_ifork_t
*ifp
) /* inode fork pointer */
3235 xfs_bmbt_rec_host_t
*ep
; /* extent record pointer */
3236 xfs_extnum_t nextents
; /* number of extents in file */
3237 int size
; /* size of file extents */
3239 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3240 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3241 ASSERT(nextents
<= XFS_LINEAR_EXTS
);
3242 size
= nextents
* sizeof(xfs_bmbt_rec_t
);
3244 xfs_iext_irec_compact_pages(ifp
);
3245 ASSERT(ifp
->if_real_bytes
== XFS_IEXT_BUFSZ
);
3247 ep
= ifp
->if_u1
.if_ext_irec
->er_extbuf
;
3248 kmem_free(ifp
->if_u1
.if_ext_irec
);
3249 ifp
->if_flags
&= ~XFS_IFEXTIREC
;
3250 ifp
->if_u1
.if_extents
= ep
;
3251 ifp
->if_bytes
= size
;
3252 if (nextents
< XFS_LINEAR_EXTS
) {
3253 xfs_iext_realloc_direct(ifp
, size
);
3258 * Free incore file extents.
3262 xfs_ifork_t
*ifp
) /* inode fork pointer */
3264 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3268 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3269 for (erp_idx
= nlists
- 1; erp_idx
>= 0 ; erp_idx
--) {
3270 xfs_iext_irec_remove(ifp
, erp_idx
);
3272 ifp
->if_flags
&= ~XFS_IFEXTIREC
;
3273 } else if (ifp
->if_real_bytes
) {
3274 kmem_free(ifp
->if_u1
.if_extents
);
3275 } else if (ifp
->if_bytes
) {
3276 memset(ifp
->if_u2
.if_inline_ext
, 0, XFS_INLINE_EXTS
*
3277 sizeof(xfs_bmbt_rec_t
));
3279 ifp
->if_u1
.if_extents
= NULL
;
3280 ifp
->if_real_bytes
= 0;
3285 * Return a pointer to the extent record for file system block bno.
3287 xfs_bmbt_rec_host_t
* /* pointer to found extent record */
3288 xfs_iext_bno_to_ext(
3289 xfs_ifork_t
*ifp
, /* inode fork pointer */
3290 xfs_fileoff_t bno
, /* block number to search for */
3291 xfs_extnum_t
*idxp
) /* index of target extent */
3293 xfs_bmbt_rec_host_t
*base
; /* pointer to first extent */
3294 xfs_filblks_t blockcount
= 0; /* number of blocks in extent */
3295 xfs_bmbt_rec_host_t
*ep
= NULL
; /* pointer to target extent */
3296 xfs_ext_irec_t
*erp
= NULL
; /* indirection array pointer */
3297 int high
; /* upper boundary in search */
3298 xfs_extnum_t idx
= 0; /* index of target extent */
3299 int low
; /* lower boundary in search */
3300 xfs_extnum_t nextents
; /* number of file extents */
3301 xfs_fileoff_t startoff
= 0; /* start offset of extent */
3303 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3304 if (nextents
== 0) {
3309 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3310 /* Find target extent list */
3312 erp
= xfs_iext_bno_to_irec(ifp
, bno
, &erp_idx
);
3313 base
= erp
->er_extbuf
;
3314 high
= erp
->er_extcount
- 1;
3316 base
= ifp
->if_u1
.if_extents
;
3317 high
= nextents
- 1;
3319 /* Binary search extent records */
3320 while (low
<= high
) {
3321 idx
= (low
+ high
) >> 1;
3323 startoff
= xfs_bmbt_get_startoff(ep
);
3324 blockcount
= xfs_bmbt_get_blockcount(ep
);
3325 if (bno
< startoff
) {
3327 } else if (bno
>= startoff
+ blockcount
) {
3330 /* Convert back to file-based extent index */
3331 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3332 idx
+= erp
->er_extoff
;
3338 /* Convert back to file-based extent index */
3339 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3340 idx
+= erp
->er_extoff
;
3342 if (bno
>= startoff
+ blockcount
) {
3343 if (++idx
== nextents
) {
3346 ep
= xfs_iext_get_ext(ifp
, idx
);
3354 * Return a pointer to the indirection array entry containing the
3355 * extent record for filesystem block bno. Store the index of the
3356 * target irec in *erp_idxp.
3358 xfs_ext_irec_t
* /* pointer to found extent record */
3359 xfs_iext_bno_to_irec(
3360 xfs_ifork_t
*ifp
, /* inode fork pointer */
3361 xfs_fileoff_t bno
, /* block number to search for */
3362 int *erp_idxp
) /* irec index of target ext list */
3364 xfs_ext_irec_t
*erp
= NULL
; /* indirection array pointer */
3365 xfs_ext_irec_t
*erp_next
; /* next indirection array entry */
3366 int erp_idx
; /* indirection array index */
3367 int nlists
; /* number of extent irec's (lists) */
3368 int high
; /* binary search upper limit */
3369 int low
; /* binary search lower limit */
3371 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3372 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3376 while (low
<= high
) {
3377 erp_idx
= (low
+ high
) >> 1;
3378 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
3379 erp_next
= erp_idx
< nlists
- 1 ? erp
+ 1 : NULL
;
3380 if (bno
< xfs_bmbt_get_startoff(erp
->er_extbuf
)) {
3382 } else if (erp_next
&& bno
>=
3383 xfs_bmbt_get_startoff(erp_next
->er_extbuf
)) {
3389 *erp_idxp
= erp_idx
;
3394 * Return a pointer to the indirection array entry containing the
3395 * extent record at file extent index *idxp. Store the index of the
3396 * target irec in *erp_idxp and store the page index of the target
3397 * extent record in *idxp.
3400 xfs_iext_idx_to_irec(
3401 xfs_ifork_t
*ifp
, /* inode fork pointer */
3402 xfs_extnum_t
*idxp
, /* extent index (file -> page) */
3403 int *erp_idxp
, /* pointer to target irec */
3404 int realloc
) /* new bytes were just added */
3406 xfs_ext_irec_t
*prev
; /* pointer to previous irec */
3407 xfs_ext_irec_t
*erp
= NULL
; /* pointer to current irec */
3408 int erp_idx
; /* indirection array index */
3409 int nlists
; /* number of irec's (ex lists) */
3410 int high
; /* binary search upper limit */
3411 int low
; /* binary search lower limit */
3412 xfs_extnum_t page_idx
= *idxp
; /* extent index in target list */
3414 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3415 ASSERT(page_idx
>= 0);
3416 ASSERT(page_idx
<= ifp
->if_bytes
/ sizeof(xfs_bmbt_rec_t
));
3417 ASSERT(page_idx
< ifp
->if_bytes
/ sizeof(xfs_bmbt_rec_t
) || realloc
);
3419 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3424 /* Binary search extent irec's */
3425 while (low
<= high
) {
3426 erp_idx
= (low
+ high
) >> 1;
3427 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
3428 prev
= erp_idx
> 0 ? erp
- 1 : NULL
;
3429 if (page_idx
< erp
->er_extoff
|| (page_idx
== erp
->er_extoff
&&
3430 realloc
&& prev
&& prev
->er_extcount
< XFS_LINEAR_EXTS
)) {
3432 } else if (page_idx
> erp
->er_extoff
+ erp
->er_extcount
||
3433 (page_idx
== erp
->er_extoff
+ erp
->er_extcount
&&
3436 } else if (page_idx
== erp
->er_extoff
+ erp
->er_extcount
&&
3437 erp
->er_extcount
== XFS_LINEAR_EXTS
) {
3441 erp
= erp_idx
< nlists
? erp
+ 1 : NULL
;
3444 page_idx
-= erp
->er_extoff
;
3449 *erp_idxp
= erp_idx
;
3454 * Allocate and initialize an indirection array once the space needed
3455 * for incore extents increases above XFS_IEXT_BUFSZ.
3459 xfs_ifork_t
*ifp
) /* inode fork pointer */
3461 xfs_ext_irec_t
*erp
; /* indirection array pointer */
3462 xfs_extnum_t nextents
; /* number of extents in file */
3464 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
));
3465 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3466 ASSERT(nextents
<= XFS_LINEAR_EXTS
);
3468 erp
= kmem_alloc(sizeof(xfs_ext_irec_t
), KM_NOFS
);
3470 if (nextents
== 0) {
3471 ifp
->if_u1
.if_extents
= kmem_alloc(XFS_IEXT_BUFSZ
, KM_NOFS
);
3472 } else if (!ifp
->if_real_bytes
) {
3473 xfs_iext_inline_to_direct(ifp
, XFS_IEXT_BUFSZ
);
3474 } else if (ifp
->if_real_bytes
< XFS_IEXT_BUFSZ
) {
3475 xfs_iext_realloc_direct(ifp
, XFS_IEXT_BUFSZ
);
3477 erp
->er_extbuf
= ifp
->if_u1
.if_extents
;
3478 erp
->er_extcount
= nextents
;
3481 ifp
->if_flags
|= XFS_IFEXTIREC
;
3482 ifp
->if_real_bytes
= XFS_IEXT_BUFSZ
;
3483 ifp
->if_bytes
= nextents
* sizeof(xfs_bmbt_rec_t
);
3484 ifp
->if_u1
.if_ext_irec
= erp
;
3490 * Allocate and initialize a new entry in the indirection array.
3494 xfs_ifork_t
*ifp
, /* inode fork pointer */
3495 int erp_idx
) /* index for new irec */
3497 xfs_ext_irec_t
*erp
; /* indirection array pointer */
3498 int i
; /* loop counter */
3499 int nlists
; /* number of irec's (ex lists) */
3501 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3502 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3504 /* Resize indirection array */
3505 xfs_iext_realloc_indirect(ifp
, ++nlists
*
3506 sizeof(xfs_ext_irec_t
));
3508 * Move records down in the array so the
3509 * new page can use erp_idx.
3511 erp
= ifp
->if_u1
.if_ext_irec
;
3512 for (i
= nlists
- 1; i
> erp_idx
; i
--) {
3513 memmove(&erp
[i
], &erp
[i
-1], sizeof(xfs_ext_irec_t
));
3515 ASSERT(i
== erp_idx
);
3517 /* Initialize new extent record */
3518 erp
= ifp
->if_u1
.if_ext_irec
;
3519 erp
[erp_idx
].er_extbuf
= kmem_alloc(XFS_IEXT_BUFSZ
, KM_NOFS
);
3520 ifp
->if_real_bytes
= nlists
* XFS_IEXT_BUFSZ
;
3521 memset(erp
[erp_idx
].er_extbuf
, 0, XFS_IEXT_BUFSZ
);
3522 erp
[erp_idx
].er_extcount
= 0;
3523 erp
[erp_idx
].er_extoff
= erp_idx
> 0 ?
3524 erp
[erp_idx
-1].er_extoff
+ erp
[erp_idx
-1].er_extcount
: 0;
3525 return (&erp
[erp_idx
]);
3529 * Remove a record from the indirection array.
3532 xfs_iext_irec_remove(
3533 xfs_ifork_t
*ifp
, /* inode fork pointer */
3534 int erp_idx
) /* irec index to remove */
3536 xfs_ext_irec_t
*erp
; /* indirection array pointer */
3537 int i
; /* loop counter */
3538 int nlists
; /* number of irec's (ex lists) */
3540 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3541 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3542 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
3543 if (erp
->er_extbuf
) {
3544 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1,
3546 kmem_free(erp
->er_extbuf
);
3548 /* Compact extent records */
3549 erp
= ifp
->if_u1
.if_ext_irec
;
3550 for (i
= erp_idx
; i
< nlists
- 1; i
++) {
3551 memmove(&erp
[i
], &erp
[i
+1], sizeof(xfs_ext_irec_t
));
3554 * Manually free the last extent record from the indirection
3555 * array. A call to xfs_iext_realloc_indirect() with a size
3556 * of zero would result in a call to xfs_iext_destroy() which
3557 * would in turn call this function again, creating a nasty
3561 xfs_iext_realloc_indirect(ifp
,
3562 nlists
* sizeof(xfs_ext_irec_t
));
3564 kmem_free(ifp
->if_u1
.if_ext_irec
);
3566 ifp
->if_real_bytes
= nlists
* XFS_IEXT_BUFSZ
;
3570 * This is called to clean up large amounts of unused memory allocated
3571 * by the indirection array. Before compacting anything though, verify
3572 * that the indirection array is still needed and switch back to the
3573 * linear extent list (or even the inline buffer) if possible. The
3574 * compaction policy is as follows:
3576 * Full Compaction: Extents fit into a single page (or inline buffer)
3577 * Partial Compaction: Extents occupy less than 50% of allocated space
3578 * No Compaction: Extents occupy at least 50% of allocated space
3581 xfs_iext_irec_compact(
3582 xfs_ifork_t
*ifp
) /* inode fork pointer */
3584 xfs_extnum_t nextents
; /* number of extents in file */
3585 int nlists
; /* number of irec's (ex lists) */
3587 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3588 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3589 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3591 if (nextents
== 0) {
3592 xfs_iext_destroy(ifp
);
3593 } else if (nextents
<= XFS_INLINE_EXTS
) {
3594 xfs_iext_indirect_to_direct(ifp
);
3595 xfs_iext_direct_to_inline(ifp
, nextents
);
3596 } else if (nextents
<= XFS_LINEAR_EXTS
) {
3597 xfs_iext_indirect_to_direct(ifp
);
3598 } else if (nextents
< (nlists
* XFS_LINEAR_EXTS
) >> 1) {
3599 xfs_iext_irec_compact_pages(ifp
);
3604 * Combine extents from neighboring extent pages.
3607 xfs_iext_irec_compact_pages(
3608 xfs_ifork_t
*ifp
) /* inode fork pointer */
3610 xfs_ext_irec_t
*erp
, *erp_next
;/* pointers to irec entries */
3611 int erp_idx
= 0; /* indirection array index */
3612 int nlists
; /* number of irec's (ex lists) */
3614 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3615 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3616 while (erp_idx
< nlists
- 1) {
3617 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
3619 if (erp_next
->er_extcount
<=
3620 (XFS_LINEAR_EXTS
- erp
->er_extcount
)) {
3621 memcpy(&erp
->er_extbuf
[erp
->er_extcount
],
3622 erp_next
->er_extbuf
, erp_next
->er_extcount
*
3623 sizeof(xfs_bmbt_rec_t
));
3624 erp
->er_extcount
+= erp_next
->er_extcount
;
3626 * Free page before removing extent record
3627 * so er_extoffs don't get modified in
3628 * xfs_iext_irec_remove.
3630 kmem_free(erp_next
->er_extbuf
);
3631 erp_next
->er_extbuf
= NULL
;
3632 xfs_iext_irec_remove(ifp
, erp_idx
+ 1);
3633 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3641 * This is called to update the er_extoff field in the indirection
3642 * array when extents have been added or removed from one of the
3643 * extent lists. erp_idx contains the irec index to begin updating
3644 * at and ext_diff contains the number of extents that were added
3648 xfs_iext_irec_update_extoffs(
3649 xfs_ifork_t
*ifp
, /* inode fork pointer */
3650 int erp_idx
, /* irec index to update */
3651 int ext_diff
) /* number of new extents */
3653 int i
; /* loop counter */
3654 int nlists
; /* number of irec's (ex lists */
3656 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3657 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3658 for (i
= erp_idx
; i
< nlists
; i
++) {
3659 ifp
->if_u1
.if_ext_irec
[i
].er_extoff
+= ext_diff
;