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
20 #include "xfs_types.h"
25 #include "xfs_trans.h"
26 #include "xfs_trans_priv.h"
30 #include "xfs_dmapi.h"
31 #include "xfs_mount.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_ialloc_btree.h"
35 #include "xfs_dir2_sf.h"
36 #include "xfs_attr_sf.h"
37 #include "xfs_dinode.h"
38 #include "xfs_inode.h"
39 #include "xfs_buf_item.h"
40 #include "xfs_inode_item.h"
41 #include "xfs_btree.h"
42 #include "xfs_alloc.h"
43 #include "xfs_ialloc.h"
46 #include "xfs_error.h"
47 #include "xfs_utils.h"
48 #include "xfs_dir2_trace.h"
49 #include "xfs_quota.h"
51 #include "xfs_filestream.h"
53 kmem_zone_t
*xfs_ifork_zone
;
54 kmem_zone_t
*xfs_inode_zone
;
55 kmem_zone_t
*xfs_chashlist_zone
;
58 * Used in xfs_itruncate(). This is the maximum number of extents
59 * freed from a file in a single transaction.
61 #define XFS_ITRUNC_MAX_EXTENTS 2
63 STATIC
int xfs_iflush_int(xfs_inode_t
*, xfs_buf_t
*);
64 STATIC
int xfs_iformat_local(xfs_inode_t
*, xfs_dinode_t
*, int, int);
65 STATIC
int xfs_iformat_extents(xfs_inode_t
*, xfs_dinode_t
*, int);
66 STATIC
int xfs_iformat_btree(xfs_inode_t
*, xfs_dinode_t
*, int);
70 * Make sure that the extents in the given memory buffer
80 xfs_bmbt_rec_host_t rec
;
83 for (i
= 0; i
< nrecs
; i
++) {
84 xfs_bmbt_rec_host_t
*ep
= xfs_iext_get_ext(ifp
, i
);
85 rec
.l0
= get_unaligned(&ep
->l0
);
86 rec
.l1
= get_unaligned(&ep
->l1
);
87 xfs_bmbt_get_all(&rec
, &irec
);
88 if (fmt
== XFS_EXTFMT_NOSTATE
)
89 ASSERT(irec
.br_state
== XFS_EXT_NORM
);
93 #define xfs_validate_extents(ifp, nrecs, fmt)
97 * Check that none of the inode's in the buffer have a next
98 * unlinked field of 0.
110 j
= mp
->m_inode_cluster_size
>> mp
->m_sb
.sb_inodelog
;
112 for (i
= 0; i
< j
; i
++) {
113 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
,
114 i
* mp
->m_sb
.sb_inodesize
);
115 if (!dip
->di_next_unlinked
) {
116 xfs_fs_cmn_err(CE_ALERT
, mp
,
117 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
119 ASSERT(dip
->di_next_unlinked
);
126 * This routine is called to map an inode number within a file
127 * system to the buffer containing the on-disk version of the
128 * inode. It returns a pointer to the buffer containing the
129 * on-disk inode in the bpp parameter, and in the dip parameter
130 * it returns a pointer to the on-disk inode within that buffer.
132 * If a non-zero error is returned, then the contents of bpp and
133 * dipp are undefined.
135 * Use xfs_imap() to determine the size and location of the
136 * buffer to read from disk.
154 * Call the space management code to find the location of the
158 error
= xfs_imap(mp
, tp
, ino
, &imap
, XFS_IMAP_LOOKUP
);
161 "xfs_inotobp: xfs_imap() returned an "
162 "error %d on %s. Returning error.", error
, mp
->m_fsname
);
167 * If the inode number maps to a block outside the bounds of the
168 * file system then return NULL rather than calling read_buf
169 * and panicing when we get an error from the driver.
171 if ((imap
.im_blkno
+ imap
.im_len
) >
172 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
)) {
174 "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds "
175 "of the file system %s. Returning EINVAL.",
176 (unsigned long long)imap
.im_blkno
,
177 imap
.im_len
, mp
->m_fsname
);
178 return XFS_ERROR(EINVAL
);
182 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
183 * default to just a read_buf() call.
185 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, imap
.im_blkno
,
186 (int)imap
.im_len
, XFS_BUF_LOCK
, &bp
);
190 "xfs_inotobp: xfs_trans_read_buf() returned an "
191 "error %d on %s. Returning error.", error
, mp
->m_fsname
);
194 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, 0);
196 be16_to_cpu(dip
->di_core
.di_magic
) == XFS_DINODE_MAGIC
&&
197 XFS_DINODE_GOOD_VERSION(dip
->di_core
.di_version
);
198 if (unlikely(XFS_TEST_ERROR(!di_ok
, mp
, XFS_ERRTAG_ITOBP_INOTOBP
,
199 XFS_RANDOM_ITOBP_INOTOBP
))) {
200 XFS_CORRUPTION_ERROR("xfs_inotobp", XFS_ERRLEVEL_LOW
, mp
, dip
);
201 xfs_trans_brelse(tp
, bp
);
203 "xfs_inotobp: XFS_TEST_ERROR() returned an "
204 "error on %s. Returning EFSCORRUPTED.", mp
->m_fsname
);
205 return XFS_ERROR(EFSCORRUPTED
);
208 xfs_inobp_check(mp
, bp
);
211 * Set *dipp to point to the on-disk inode in the buffer.
213 *dipp
= (xfs_dinode_t
*)xfs_buf_offset(bp
, imap
.im_boffset
);
215 *offset
= imap
.im_boffset
;
221 * This routine is called to map an inode to the buffer containing
222 * the on-disk version of the inode. It returns a pointer to the
223 * buffer containing the on-disk inode in the bpp parameter, and in
224 * the dip parameter it returns a pointer to the on-disk inode within
227 * If a non-zero error is returned, then the contents of bpp and
228 * dipp are undefined.
230 * If the inode is new and has not yet been initialized, use xfs_imap()
231 * to determine the size and location of the buffer to read from disk.
232 * If the inode has already been mapped to its buffer and read in once,
233 * then use the mapping information stored in the inode rather than
234 * calling xfs_imap(). This allows us to avoid the overhead of looking
235 * at the inode btree for small block file systems (see xfs_dilocate()).
236 * We can tell whether the inode has been mapped in before by comparing
237 * its disk block address to 0. Only uninitialized inodes will have
238 * 0 for the disk block address.
256 if (ip
->i_blkno
== (xfs_daddr_t
)0) {
258 * Call the space management code to find the location of the
262 if ((error
= xfs_imap(mp
, tp
, ip
->i_ino
, &imap
,
263 XFS_IMAP_LOOKUP
| imap_flags
)))
267 * If the inode number maps to a block outside the bounds
268 * of the file system then return NULL rather than calling
269 * read_buf and panicing when we get an error from the
272 if ((imap
.im_blkno
+ imap
.im_len
) >
273 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
)) {
275 xfs_fs_cmn_err(CE_ALERT
, mp
, "xfs_itobp: "
276 "(imap.im_blkno (0x%llx) "
277 "+ imap.im_len (0x%llx)) > "
278 " XFS_FSB_TO_BB(mp, "
279 "mp->m_sb.sb_dblocks) (0x%llx)",
280 (unsigned long long) imap
.im_blkno
,
281 (unsigned long long) imap
.im_len
,
282 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
));
284 return XFS_ERROR(EINVAL
);
288 * Fill in the fields in the inode that will be used to
289 * map the inode to its buffer from now on.
291 ip
->i_blkno
= imap
.im_blkno
;
292 ip
->i_len
= imap
.im_len
;
293 ip
->i_boffset
= imap
.im_boffset
;
296 * We've already mapped the inode once, so just use the
297 * mapping that we saved the first time.
299 imap
.im_blkno
= ip
->i_blkno
;
300 imap
.im_len
= ip
->i_len
;
301 imap
.im_boffset
= ip
->i_boffset
;
303 ASSERT(bno
== 0 || bno
== imap
.im_blkno
);
306 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
307 * default to just a read_buf() call.
309 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, imap
.im_blkno
,
310 (int)imap
.im_len
, XFS_BUF_LOCK
, &bp
);
313 xfs_fs_cmn_err(CE_ALERT
, mp
, "xfs_itobp: "
314 "xfs_trans_read_buf() returned error %d, "
315 "imap.im_blkno 0x%llx, imap.im_len 0x%llx",
316 error
, (unsigned long long) imap
.im_blkno
,
317 (unsigned long long) imap
.im_len
);
323 * Validate the magic number and version of every inode in the buffer
324 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
325 * No validation is done here in userspace (xfs_repair).
327 #if !defined(__KERNEL__)
330 ni
= BBTOB(imap
.im_len
) >> mp
->m_sb
.sb_inodelog
;
331 #else /* usual case */
335 for (i
= 0; i
< ni
; i
++) {
339 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
,
340 (i
<< mp
->m_sb
.sb_inodelog
));
341 di_ok
= be16_to_cpu(dip
->di_core
.di_magic
) == XFS_DINODE_MAGIC
&&
342 XFS_DINODE_GOOD_VERSION(dip
->di_core
.di_version
);
343 if (unlikely(XFS_TEST_ERROR(!di_ok
, mp
,
344 XFS_ERRTAG_ITOBP_INOTOBP
,
345 XFS_RANDOM_ITOBP_INOTOBP
))) {
346 if (imap_flags
& XFS_IMAP_BULKSTAT
) {
347 xfs_trans_brelse(tp
, bp
);
348 return XFS_ERROR(EINVAL
);
352 "Device %s - bad inode magic/vsn "
353 "daddr %lld #%d (magic=%x)",
354 XFS_BUFTARG_NAME(mp
->m_ddev_targp
),
355 (unsigned long long)imap
.im_blkno
, i
,
356 be16_to_cpu(dip
->di_core
.di_magic
));
358 XFS_CORRUPTION_ERROR("xfs_itobp", XFS_ERRLEVEL_HIGH
,
360 xfs_trans_brelse(tp
, bp
);
361 return XFS_ERROR(EFSCORRUPTED
);
365 xfs_inobp_check(mp
, bp
);
368 * Mark the buffer as an inode buffer now that it looks good
370 XFS_BUF_SET_VTYPE(bp
, B_FS_INO
);
373 * Set *dipp to point to the on-disk inode in the buffer.
375 *dipp
= (xfs_dinode_t
*)xfs_buf_offset(bp
, imap
.im_boffset
);
381 * Move inode type and inode format specific information from the
382 * on-disk inode to the in-core inode. For fifos, devs, and sockets
383 * this means set if_rdev to the proper value. For files, directories,
384 * and symlinks this means to bring in the in-line data or extent
385 * pointers. For a file in B-tree format, only the root is immediately
386 * brought in-core. The rest will be in-lined in if_extents when it
387 * is first referenced (see xfs_iread_extents()).
394 xfs_attr_shortform_t
*atp
;
398 ip
->i_df
.if_ext_max
=
399 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
402 if (unlikely(be32_to_cpu(dip
->di_core
.di_nextents
) +
403 be16_to_cpu(dip
->di_core
.di_anextents
) >
404 be64_to_cpu(dip
->di_core
.di_nblocks
))) {
405 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
406 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
407 (unsigned long long)ip
->i_ino
,
408 (int)(be32_to_cpu(dip
->di_core
.di_nextents
) +
409 be16_to_cpu(dip
->di_core
.di_anextents
)),
411 be64_to_cpu(dip
->di_core
.di_nblocks
));
412 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW
,
414 return XFS_ERROR(EFSCORRUPTED
);
417 if (unlikely(dip
->di_core
.di_forkoff
> ip
->i_mount
->m_sb
.sb_inodesize
)) {
418 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
419 "corrupt dinode %Lu, forkoff = 0x%x.",
420 (unsigned long long)ip
->i_ino
,
421 dip
->di_core
.di_forkoff
);
422 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW
,
424 return XFS_ERROR(EFSCORRUPTED
);
427 switch (ip
->i_d
.di_mode
& S_IFMT
) {
432 if (unlikely(dip
->di_core
.di_format
!= XFS_DINODE_FMT_DEV
)) {
433 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW
,
435 return XFS_ERROR(EFSCORRUPTED
);
439 ip
->i_df
.if_u2
.if_rdev
= be32_to_cpu(dip
->di_u
.di_dev
);
445 switch (dip
->di_core
.di_format
) {
446 case XFS_DINODE_FMT_LOCAL
:
448 * no local regular files yet
450 if (unlikely((be16_to_cpu(dip
->di_core
.di_mode
) & S_IFMT
) == S_IFREG
)) {
451 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
453 "(local format for regular file).",
454 (unsigned long long) ip
->i_ino
);
455 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
458 return XFS_ERROR(EFSCORRUPTED
);
461 di_size
= be64_to_cpu(dip
->di_core
.di_size
);
462 if (unlikely(di_size
> XFS_DFORK_DSIZE(dip
, ip
->i_mount
))) {
463 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
465 "(bad size %Ld for local inode).",
466 (unsigned long long) ip
->i_ino
,
467 (long long) di_size
);
468 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
471 return XFS_ERROR(EFSCORRUPTED
);
475 error
= xfs_iformat_local(ip
, dip
, XFS_DATA_FORK
, size
);
477 case XFS_DINODE_FMT_EXTENTS
:
478 error
= xfs_iformat_extents(ip
, dip
, XFS_DATA_FORK
);
480 case XFS_DINODE_FMT_BTREE
:
481 error
= xfs_iformat_btree(ip
, dip
, XFS_DATA_FORK
);
484 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW
,
486 return XFS_ERROR(EFSCORRUPTED
);
491 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW
, ip
->i_mount
);
492 return XFS_ERROR(EFSCORRUPTED
);
497 if (!XFS_DFORK_Q(dip
))
499 ASSERT(ip
->i_afp
== NULL
);
500 ip
->i_afp
= kmem_zone_zalloc(xfs_ifork_zone
, KM_SLEEP
);
501 ip
->i_afp
->if_ext_max
=
502 XFS_IFORK_ASIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
503 switch (dip
->di_core
.di_aformat
) {
504 case XFS_DINODE_FMT_LOCAL
:
505 atp
= (xfs_attr_shortform_t
*)XFS_DFORK_APTR(dip
);
506 size
= be16_to_cpu(atp
->hdr
.totsize
);
507 error
= xfs_iformat_local(ip
, dip
, XFS_ATTR_FORK
, size
);
509 case XFS_DINODE_FMT_EXTENTS
:
510 error
= xfs_iformat_extents(ip
, dip
, XFS_ATTR_FORK
);
512 case XFS_DINODE_FMT_BTREE
:
513 error
= xfs_iformat_btree(ip
, dip
, XFS_ATTR_FORK
);
516 error
= XFS_ERROR(EFSCORRUPTED
);
520 kmem_zone_free(xfs_ifork_zone
, ip
->i_afp
);
522 xfs_idestroy_fork(ip
, XFS_DATA_FORK
);
528 * The file is in-lined in the on-disk inode.
529 * If it fits into if_inline_data, then copy
530 * it there, otherwise allocate a buffer for it
531 * and copy the data there. Either way, set
532 * if_data to point at the data.
533 * If we allocate a buffer for the data, make
534 * sure that its size is a multiple of 4 and
535 * record the real size in i_real_bytes.
548 * If the size is unreasonable, then something
549 * is wrong and we just bail out rather than crash in
550 * kmem_alloc() or memcpy() below.
552 if (unlikely(size
> XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
))) {
553 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
555 "(bad size %d for local fork, size = %d).",
556 (unsigned long long) ip
->i_ino
, size
,
557 XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
));
558 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW
,
560 return XFS_ERROR(EFSCORRUPTED
);
562 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
565 ifp
->if_u1
.if_data
= NULL
;
566 else if (size
<= sizeof(ifp
->if_u2
.if_inline_data
))
567 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
569 real_size
= roundup(size
, 4);
570 ifp
->if_u1
.if_data
= kmem_alloc(real_size
, KM_SLEEP
);
572 ifp
->if_bytes
= size
;
573 ifp
->if_real_bytes
= real_size
;
575 memcpy(ifp
->if_u1
.if_data
, XFS_DFORK_PTR(dip
, whichfork
), size
);
576 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
577 ifp
->if_flags
|= XFS_IFINLINE
;
582 * The file consists of a set of extents all
583 * of which fit into the on-disk inode.
584 * If there are few enough extents to fit into
585 * the if_inline_ext, then copy them there.
586 * Otherwise allocate a buffer for them and copy
587 * them into it. Either way, set if_extents
588 * to point at the extents.
602 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
603 nex
= XFS_DFORK_NEXTENTS(dip
, whichfork
);
604 size
= nex
* (uint
)sizeof(xfs_bmbt_rec_t
);
607 * If the number of extents is unreasonable, then something
608 * is wrong and we just bail out rather than crash in
609 * kmem_alloc() or memcpy() below.
611 if (unlikely(size
< 0 || size
> XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
))) {
612 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
613 "corrupt inode %Lu ((a)extents = %d).",
614 (unsigned long long) ip
->i_ino
, nex
);
615 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW
,
617 return XFS_ERROR(EFSCORRUPTED
);
620 ifp
->if_real_bytes
= 0;
622 ifp
->if_u1
.if_extents
= NULL
;
623 else if (nex
<= XFS_INLINE_EXTS
)
624 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
626 xfs_iext_add(ifp
, 0, nex
);
628 ifp
->if_bytes
= size
;
630 dp
= (xfs_bmbt_rec_t
*) XFS_DFORK_PTR(dip
, whichfork
);
631 xfs_validate_extents(ifp
, nex
, XFS_EXTFMT_INODE(ip
));
632 for (i
= 0; i
< nex
; i
++, dp
++) {
633 xfs_bmbt_rec_host_t
*ep
= xfs_iext_get_ext(ifp
, i
);
634 ep
->l0
= be64_to_cpu(get_unaligned(&dp
->l0
));
635 ep
->l1
= be64_to_cpu(get_unaligned(&dp
->l1
));
637 XFS_BMAP_TRACE_EXLIST(ip
, nex
, whichfork
);
638 if (whichfork
!= XFS_DATA_FORK
||
639 XFS_EXTFMT_INODE(ip
) == XFS_EXTFMT_NOSTATE
)
640 if (unlikely(xfs_check_nostate_extents(
642 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
645 return XFS_ERROR(EFSCORRUPTED
);
648 ifp
->if_flags
|= XFS_IFEXTENTS
;
653 * The file has too many extents to fit into
654 * the inode, so they are in B-tree format.
655 * Allocate a buffer for the root of the B-tree
656 * and copy the root into it. The i_extents
657 * field will remain NULL until all of the
658 * extents are read in (when they are needed).
666 xfs_bmdr_block_t
*dfp
;
672 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
673 dfp
= (xfs_bmdr_block_t
*)XFS_DFORK_PTR(dip
, whichfork
);
674 size
= XFS_BMAP_BROOT_SPACE(dfp
);
675 nrecs
= XFS_BMAP_BROOT_NUMRECS(dfp
);
678 * blow out if -- fork has less extents than can fit in
679 * fork (fork shouldn't be a btree format), root btree
680 * block has more records than can fit into the fork,
681 * or the number of extents is greater than the number of
684 if (unlikely(XFS_IFORK_NEXTENTS(ip
, whichfork
) <= ifp
->if_ext_max
685 || XFS_BMDR_SPACE_CALC(nrecs
) >
686 XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
)
687 || XFS_IFORK_NEXTENTS(ip
, whichfork
) > ip
->i_d
.di_nblocks
)) {
688 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
689 "corrupt inode %Lu (btree).",
690 (unsigned long long) ip
->i_ino
);
691 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW
,
693 return XFS_ERROR(EFSCORRUPTED
);
696 ifp
->if_broot_bytes
= size
;
697 ifp
->if_broot
= kmem_alloc(size
, KM_SLEEP
);
698 ASSERT(ifp
->if_broot
!= NULL
);
700 * Copy and convert from the on-disk structure
701 * to the in-memory structure.
703 xfs_bmdr_to_bmbt(dfp
, XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
),
704 ifp
->if_broot
, size
);
705 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
706 ifp
->if_flags
|= XFS_IFBROOT
;
712 xfs_dinode_from_disk(
714 xfs_dinode_core_t
*from
)
716 to
->di_magic
= be16_to_cpu(from
->di_magic
);
717 to
->di_mode
= be16_to_cpu(from
->di_mode
);
718 to
->di_version
= from
->di_version
;
719 to
->di_format
= from
->di_format
;
720 to
->di_onlink
= be16_to_cpu(from
->di_onlink
);
721 to
->di_uid
= be32_to_cpu(from
->di_uid
);
722 to
->di_gid
= be32_to_cpu(from
->di_gid
);
723 to
->di_nlink
= be32_to_cpu(from
->di_nlink
);
724 to
->di_projid
= be16_to_cpu(from
->di_projid
);
725 memcpy(to
->di_pad
, from
->di_pad
, sizeof(to
->di_pad
));
726 to
->di_flushiter
= be16_to_cpu(from
->di_flushiter
);
727 to
->di_atime
.t_sec
= be32_to_cpu(from
->di_atime
.t_sec
);
728 to
->di_atime
.t_nsec
= be32_to_cpu(from
->di_atime
.t_nsec
);
729 to
->di_mtime
.t_sec
= be32_to_cpu(from
->di_mtime
.t_sec
);
730 to
->di_mtime
.t_nsec
= be32_to_cpu(from
->di_mtime
.t_nsec
);
731 to
->di_ctime
.t_sec
= be32_to_cpu(from
->di_ctime
.t_sec
);
732 to
->di_ctime
.t_nsec
= be32_to_cpu(from
->di_ctime
.t_nsec
);
733 to
->di_size
= be64_to_cpu(from
->di_size
);
734 to
->di_nblocks
= be64_to_cpu(from
->di_nblocks
);
735 to
->di_extsize
= be32_to_cpu(from
->di_extsize
);
736 to
->di_nextents
= be32_to_cpu(from
->di_nextents
);
737 to
->di_anextents
= be16_to_cpu(from
->di_anextents
);
738 to
->di_forkoff
= from
->di_forkoff
;
739 to
->di_aformat
= from
->di_aformat
;
740 to
->di_dmevmask
= be32_to_cpu(from
->di_dmevmask
);
741 to
->di_dmstate
= be16_to_cpu(from
->di_dmstate
);
742 to
->di_flags
= be16_to_cpu(from
->di_flags
);
743 to
->di_gen
= be32_to_cpu(from
->di_gen
);
748 xfs_dinode_core_t
*to
,
749 xfs_icdinode_t
*from
)
751 to
->di_magic
= cpu_to_be16(from
->di_magic
);
752 to
->di_mode
= cpu_to_be16(from
->di_mode
);
753 to
->di_version
= from
->di_version
;
754 to
->di_format
= from
->di_format
;
755 to
->di_onlink
= cpu_to_be16(from
->di_onlink
);
756 to
->di_uid
= cpu_to_be32(from
->di_uid
);
757 to
->di_gid
= cpu_to_be32(from
->di_gid
);
758 to
->di_nlink
= cpu_to_be32(from
->di_nlink
);
759 to
->di_projid
= cpu_to_be16(from
->di_projid
);
760 memcpy(to
->di_pad
, from
->di_pad
, sizeof(to
->di_pad
));
761 to
->di_flushiter
= cpu_to_be16(from
->di_flushiter
);
762 to
->di_atime
.t_sec
= cpu_to_be32(from
->di_atime
.t_sec
);
763 to
->di_atime
.t_nsec
= cpu_to_be32(from
->di_atime
.t_nsec
);
764 to
->di_mtime
.t_sec
= cpu_to_be32(from
->di_mtime
.t_sec
);
765 to
->di_mtime
.t_nsec
= cpu_to_be32(from
->di_mtime
.t_nsec
);
766 to
->di_ctime
.t_sec
= cpu_to_be32(from
->di_ctime
.t_sec
);
767 to
->di_ctime
.t_nsec
= cpu_to_be32(from
->di_ctime
.t_nsec
);
768 to
->di_size
= cpu_to_be64(from
->di_size
);
769 to
->di_nblocks
= cpu_to_be64(from
->di_nblocks
);
770 to
->di_extsize
= cpu_to_be32(from
->di_extsize
);
771 to
->di_nextents
= cpu_to_be32(from
->di_nextents
);
772 to
->di_anextents
= cpu_to_be16(from
->di_anextents
);
773 to
->di_forkoff
= from
->di_forkoff
;
774 to
->di_aformat
= from
->di_aformat
;
775 to
->di_dmevmask
= cpu_to_be32(from
->di_dmevmask
);
776 to
->di_dmstate
= cpu_to_be16(from
->di_dmstate
);
777 to
->di_flags
= cpu_to_be16(from
->di_flags
);
778 to
->di_gen
= cpu_to_be32(from
->di_gen
);
787 if (di_flags
& XFS_DIFLAG_ANY
) {
788 if (di_flags
& XFS_DIFLAG_REALTIME
)
789 flags
|= XFS_XFLAG_REALTIME
;
790 if (di_flags
& XFS_DIFLAG_PREALLOC
)
791 flags
|= XFS_XFLAG_PREALLOC
;
792 if (di_flags
& XFS_DIFLAG_IMMUTABLE
)
793 flags
|= XFS_XFLAG_IMMUTABLE
;
794 if (di_flags
& XFS_DIFLAG_APPEND
)
795 flags
|= XFS_XFLAG_APPEND
;
796 if (di_flags
& XFS_DIFLAG_SYNC
)
797 flags
|= XFS_XFLAG_SYNC
;
798 if (di_flags
& XFS_DIFLAG_NOATIME
)
799 flags
|= XFS_XFLAG_NOATIME
;
800 if (di_flags
& XFS_DIFLAG_NODUMP
)
801 flags
|= XFS_XFLAG_NODUMP
;
802 if (di_flags
& XFS_DIFLAG_RTINHERIT
)
803 flags
|= XFS_XFLAG_RTINHERIT
;
804 if (di_flags
& XFS_DIFLAG_PROJINHERIT
)
805 flags
|= XFS_XFLAG_PROJINHERIT
;
806 if (di_flags
& XFS_DIFLAG_NOSYMLINKS
)
807 flags
|= XFS_XFLAG_NOSYMLINKS
;
808 if (di_flags
& XFS_DIFLAG_EXTSIZE
)
809 flags
|= XFS_XFLAG_EXTSIZE
;
810 if (di_flags
& XFS_DIFLAG_EXTSZINHERIT
)
811 flags
|= XFS_XFLAG_EXTSZINHERIT
;
812 if (di_flags
& XFS_DIFLAG_NODEFRAG
)
813 flags
|= XFS_XFLAG_NODEFRAG
;
814 if (di_flags
& XFS_DIFLAG_FILESTREAM
)
815 flags
|= XFS_XFLAG_FILESTREAM
;
825 xfs_icdinode_t
*dic
= &ip
->i_d
;
827 return _xfs_dic2xflags(dic
->di_flags
) |
828 (XFS_CFORK_Q(dic
) ? XFS_XFLAG_HASATTR
: 0);
833 xfs_dinode_core_t
*dic
)
835 return _xfs_dic2xflags(be16_to_cpu(dic
->di_flags
)) |
836 (XFS_CFORK_Q_DISK(dic
) ? XFS_XFLAG_HASATTR
: 0);
840 * Given a mount structure and an inode number, return a pointer
841 * to a newly allocated in-core inode corresponding to the given
844 * Initialize the inode's attributes and extent pointers if it
845 * already has them (it will not if the inode has no links).
861 ASSERT(xfs_inode_zone
!= NULL
);
863 ip
= kmem_zone_zalloc(xfs_inode_zone
, KM_SLEEP
);
866 spin_lock_init(&ip
->i_flags_lock
);
869 * Get pointer's to the on-disk inode and the buffer containing it.
870 * If the inode number refers to a block outside the file system
871 * then xfs_itobp() will return NULL. In this case we should
872 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
873 * know that this is a new incore inode.
875 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &bp
, bno
, imap_flags
);
877 kmem_zone_free(xfs_inode_zone
, ip
);
882 * Initialize inode's trace buffers.
883 * Do this before xfs_iformat in case it adds entries.
885 #ifdef XFS_BMAP_TRACE
886 ip
->i_xtrace
= ktrace_alloc(XFS_BMAP_KTRACE_SIZE
, KM_SLEEP
);
888 #ifdef XFS_BMBT_TRACE
889 ip
->i_btrace
= ktrace_alloc(XFS_BMBT_KTRACE_SIZE
, KM_SLEEP
);
892 ip
->i_rwtrace
= ktrace_alloc(XFS_RW_KTRACE_SIZE
, KM_SLEEP
);
894 #ifdef XFS_ILOCK_TRACE
895 ip
->i_lock_trace
= ktrace_alloc(XFS_ILOCK_KTRACE_SIZE
, KM_SLEEP
);
897 #ifdef XFS_DIR2_TRACE
898 ip
->i_dir_trace
= ktrace_alloc(XFS_DIR2_KTRACE_SIZE
, KM_SLEEP
);
902 * If we got something that isn't an inode it means someone
903 * (nfs or dmi) has a stale handle.
905 if (be16_to_cpu(dip
->di_core
.di_magic
) != XFS_DINODE_MAGIC
) {
906 kmem_zone_free(xfs_inode_zone
, ip
);
907 xfs_trans_brelse(tp
, bp
);
909 xfs_fs_cmn_err(CE_ALERT
, mp
, "xfs_iread: "
910 "dip->di_core.di_magic (0x%x) != "
911 "XFS_DINODE_MAGIC (0x%x)",
912 be16_to_cpu(dip
->di_core
.di_magic
),
915 return XFS_ERROR(EINVAL
);
919 * If the on-disk inode is already linked to a directory
920 * entry, copy all of the inode into the in-core inode.
921 * xfs_iformat() handles copying in the inode format
922 * specific information.
923 * Otherwise, just get the truly permanent information.
925 if (dip
->di_core
.di_mode
) {
926 xfs_dinode_from_disk(&ip
->i_d
, &dip
->di_core
);
927 error
= xfs_iformat(ip
, dip
);
929 kmem_zone_free(xfs_inode_zone
, ip
);
930 xfs_trans_brelse(tp
, bp
);
932 xfs_fs_cmn_err(CE_ALERT
, mp
, "xfs_iread: "
933 "xfs_iformat() returned error %d",
939 ip
->i_d
.di_magic
= be16_to_cpu(dip
->di_core
.di_magic
);
940 ip
->i_d
.di_version
= dip
->di_core
.di_version
;
941 ip
->i_d
.di_gen
= be32_to_cpu(dip
->di_core
.di_gen
);
942 ip
->i_d
.di_flushiter
= be16_to_cpu(dip
->di_core
.di_flushiter
);
944 * Make sure to pull in the mode here as well in
945 * case the inode is released without being used.
946 * This ensures that xfs_inactive() will see that
947 * the inode is already free and not try to mess
948 * with the uninitialized part of it.
952 * Initialize the per-fork minima and maxima for a new
953 * inode here. xfs_iformat will do it for old inodes.
955 ip
->i_df
.if_ext_max
=
956 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
959 INIT_LIST_HEAD(&ip
->i_reclaim
);
962 * The inode format changed when we moved the link count and
963 * made it 32 bits long. If this is an old format inode,
964 * convert it in memory to look like a new one. If it gets
965 * flushed to disk we will convert back before flushing or
966 * logging it. We zero out the new projid field and the old link
967 * count field. We'll handle clearing the pad field (the remains
968 * of the old uuid field) when we actually convert the inode to
969 * the new format. We don't change the version number so that we
970 * can distinguish this from a real new format inode.
972 if (ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
) {
973 ip
->i_d
.di_nlink
= ip
->i_d
.di_onlink
;
974 ip
->i_d
.di_onlink
= 0;
975 ip
->i_d
.di_projid
= 0;
978 ip
->i_delayed_blks
= 0;
979 ip
->i_size
= ip
->i_d
.di_size
;
982 * Mark the buffer containing the inode as something to keep
983 * around for a while. This helps to keep recently accessed
984 * meta-data in-core longer.
986 XFS_BUF_SET_REF(bp
, XFS_INO_REF
);
989 * Use xfs_trans_brelse() to release the buffer containing the
990 * on-disk inode, because it was acquired with xfs_trans_read_buf()
991 * in xfs_itobp() above. If tp is NULL, this is just a normal
992 * brelse(). If we're within a transaction, then xfs_trans_brelse()
993 * will only release the buffer if it is not dirty within the
994 * transaction. It will be OK to release the buffer in this case,
995 * because inodes on disk are never destroyed and we will be
996 * locking the new in-core inode before putting it in the hash
997 * table where other processes can find it. Thus we don't have
998 * to worry about the inode being changed just because we released
1001 xfs_trans_brelse(tp
, bp
);
1007 * Read in extents from a btree-format inode.
1008 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1018 xfs_extnum_t nextents
;
1021 if (unlikely(XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
)) {
1022 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW
,
1024 return XFS_ERROR(EFSCORRUPTED
);
1026 nextents
= XFS_IFORK_NEXTENTS(ip
, whichfork
);
1027 size
= nextents
* sizeof(xfs_bmbt_rec_t
);
1028 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1031 * We know that the size is valid (it's checked in iformat_btree)
1033 ifp
->if_lastex
= NULLEXTNUM
;
1034 ifp
->if_bytes
= ifp
->if_real_bytes
= 0;
1035 ifp
->if_flags
|= XFS_IFEXTENTS
;
1036 xfs_iext_add(ifp
, 0, nextents
);
1037 error
= xfs_bmap_read_extents(tp
, ip
, whichfork
);
1039 xfs_iext_destroy(ifp
);
1040 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
1043 xfs_validate_extents(ifp
, nextents
, XFS_EXTFMT_INODE(ip
));
1048 * Allocate an inode on disk and return a copy of its in-core version.
1049 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1050 * appropriately within the inode. The uid and gid for the inode are
1051 * set according to the contents of the given cred structure.
1053 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1054 * has a free inode available, call xfs_iget()
1055 * to obtain the in-core version of the allocated inode. Finally,
1056 * fill in the inode and log its initial contents. In this case,
1057 * ialloc_context would be set to NULL and call_again set to false.
1059 * If xfs_dialloc() does not have an available inode,
1060 * it will replenish its supply by doing an allocation. Since we can
1061 * only do one allocation within a transaction without deadlocks, we
1062 * must commit the current transaction before returning the inode itself.
1063 * In this case, therefore, we will set call_again to true and return.
1064 * The caller should then commit the current transaction, start a new
1065 * transaction, and call xfs_ialloc() again to actually get the inode.
1067 * To ensure that some other process does not grab the inode that
1068 * was allocated during the first call to xfs_ialloc(), this routine
1069 * also returns the [locked] bp pointing to the head of the freelist
1070 * as ialloc_context. The caller should hold this buffer across
1071 * the commit and pass it back into this routine on the second call.
1073 * If we are allocating quota inodes, we do not have a parent inode
1074 * to attach to or associate with (i.e. pip == NULL) because they
1075 * are not linked into the directory structure - they are attached
1076 * directly to the superblock - and so have no parent.
1088 xfs_buf_t
**ialloc_context
,
1089 boolean_t
*call_again
,
1099 * Call the space management code to pick
1100 * the on-disk inode to be allocated.
1102 error
= xfs_dialloc(tp
, pip
? pip
->i_ino
: 0, mode
, okalloc
,
1103 ialloc_context
, call_again
, &ino
);
1107 if (*call_again
|| ino
== NULLFSINO
) {
1111 ASSERT(*ialloc_context
== NULL
);
1114 * Get the in-core inode with the lock held exclusively.
1115 * This is because we're setting fields here we need
1116 * to prevent others from looking at until we're done.
1118 error
= xfs_trans_iget(tp
->t_mountp
, tp
, ino
,
1119 XFS_IGET_CREATE
, XFS_ILOCK_EXCL
, &ip
);
1126 ip
->i_d
.di_mode
= (__uint16_t
)mode
;
1127 ip
->i_d
.di_onlink
= 0;
1128 ip
->i_d
.di_nlink
= nlink
;
1129 ASSERT(ip
->i_d
.di_nlink
== nlink
);
1130 ip
->i_d
.di_uid
= current_fsuid(cr
);
1131 ip
->i_d
.di_gid
= current_fsgid(cr
);
1132 ip
->i_d
.di_projid
= prid
;
1133 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
1136 * If the superblock version is up to where we support new format
1137 * inodes and this is currently an old format inode, then change
1138 * the inode version number now. This way we only do the conversion
1139 * here rather than here and in the flush/logging code.
1141 if (XFS_SB_VERSION_HASNLINK(&tp
->t_mountp
->m_sb
) &&
1142 ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
) {
1143 ip
->i_d
.di_version
= XFS_DINODE_VERSION_2
;
1145 * We've already zeroed the old link count, the projid field,
1146 * and the pad field.
1151 * Project ids won't be stored on disk if we are using a version 1 inode.
1153 if ((prid
!= 0) && (ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
))
1154 xfs_bump_ino_vers2(tp
, ip
);
1156 if (pip
&& XFS_INHERIT_GID(pip
, vp
->v_vfsp
)) {
1157 ip
->i_d
.di_gid
= pip
->i_d
.di_gid
;
1158 if ((pip
->i_d
.di_mode
& S_ISGID
) && (mode
& S_IFMT
) == S_IFDIR
) {
1159 ip
->i_d
.di_mode
|= S_ISGID
;
1164 * If the group ID of the new file does not match the effective group
1165 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1166 * (and only if the irix_sgid_inherit compatibility variable is set).
1168 if ((irix_sgid_inherit
) &&
1169 (ip
->i_d
.di_mode
& S_ISGID
) &&
1170 (!in_group_p((gid_t
)ip
->i_d
.di_gid
))) {
1171 ip
->i_d
.di_mode
&= ~S_ISGID
;
1174 ip
->i_d
.di_size
= 0;
1176 ip
->i_d
.di_nextents
= 0;
1177 ASSERT(ip
->i_d
.di_nblocks
== 0);
1178 xfs_ichgtime(ip
, XFS_ICHGTIME_CHG
|XFS_ICHGTIME_ACC
|XFS_ICHGTIME_MOD
);
1180 * di_gen will have been taken care of in xfs_iread.
1182 ip
->i_d
.di_extsize
= 0;
1183 ip
->i_d
.di_dmevmask
= 0;
1184 ip
->i_d
.di_dmstate
= 0;
1185 ip
->i_d
.di_flags
= 0;
1186 flags
= XFS_ILOG_CORE
;
1187 switch (mode
& S_IFMT
) {
1192 ip
->i_d
.di_format
= XFS_DINODE_FMT_DEV
;
1193 ip
->i_df
.if_u2
.if_rdev
= rdev
;
1194 ip
->i_df
.if_flags
= 0;
1195 flags
|= XFS_ILOG_DEV
;
1198 if (pip
&& xfs_inode_is_filestream(pip
)) {
1199 error
= xfs_filestream_associate(pip
, ip
);
1203 xfs_iflags_set(ip
, XFS_IFILESTREAM
);
1207 if (pip
&& (pip
->i_d
.di_flags
& XFS_DIFLAG_ANY
)) {
1210 if ((mode
& S_IFMT
) == S_IFDIR
) {
1211 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
1212 di_flags
|= XFS_DIFLAG_RTINHERIT
;
1213 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
1214 di_flags
|= XFS_DIFLAG_EXTSZINHERIT
;
1215 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
1217 } else if ((mode
& S_IFMT
) == S_IFREG
) {
1218 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
) {
1219 di_flags
|= XFS_DIFLAG_REALTIME
;
1220 ip
->i_iocore
.io_flags
|= XFS_IOCORE_RT
;
1222 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
1223 di_flags
|= XFS_DIFLAG_EXTSIZE
;
1224 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
1227 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOATIME
) &&
1228 xfs_inherit_noatime
)
1229 di_flags
|= XFS_DIFLAG_NOATIME
;
1230 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODUMP
) &&
1232 di_flags
|= XFS_DIFLAG_NODUMP
;
1233 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_SYNC
) &&
1235 di_flags
|= XFS_DIFLAG_SYNC
;
1236 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOSYMLINKS
) &&
1237 xfs_inherit_nosymlinks
)
1238 di_flags
|= XFS_DIFLAG_NOSYMLINKS
;
1239 if (pip
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
1240 di_flags
|= XFS_DIFLAG_PROJINHERIT
;
1241 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODEFRAG
) &&
1242 xfs_inherit_nodefrag
)
1243 di_flags
|= XFS_DIFLAG_NODEFRAG
;
1244 if (pip
->i_d
.di_flags
& XFS_DIFLAG_FILESTREAM
)
1245 di_flags
|= XFS_DIFLAG_FILESTREAM
;
1246 ip
->i_d
.di_flags
|= di_flags
;
1250 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
1251 ip
->i_df
.if_flags
= XFS_IFEXTENTS
;
1252 ip
->i_df
.if_bytes
= ip
->i_df
.if_real_bytes
= 0;
1253 ip
->i_df
.if_u1
.if_extents
= NULL
;
1259 * Attribute fork settings for new inode.
1261 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
1262 ip
->i_d
.di_anextents
= 0;
1265 * Log the new values stuffed into the inode.
1267 xfs_trans_log_inode(tp
, ip
, flags
);
1269 /* now that we have an i_mode we can setup inode ops and unlock */
1270 bhv_vfs_init_vnode(XFS_MTOVFS(tp
->t_mountp
), vp
, XFS_ITOBHV(ip
), 1);
1277 * Check to make sure that there are no blocks allocated to the
1278 * file beyond the size of the file. We don't check this for
1279 * files with fixed size extents or real time extents, but we
1280 * at least do it for regular files.
1289 xfs_fileoff_t map_first
;
1291 xfs_bmbt_irec_t imaps
[2];
1293 if ((ip
->i_d
.di_mode
& S_IFMT
) != S_IFREG
)
1296 if (ip
->i_d
.di_flags
& (XFS_DIFLAG_REALTIME
| XFS_DIFLAG_EXTSIZE
))
1300 map_first
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)isize
);
1302 * The filesystem could be shutting down, so bmapi may return
1305 if (xfs_bmapi(NULL
, ip
, map_first
,
1307 (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
)) -
1309 XFS_BMAPI_ENTIRE
, NULL
, 0, imaps
, &nimaps
,
1312 ASSERT(nimaps
== 1);
1313 ASSERT(imaps
[0].br_startblock
== HOLESTARTBLOCK
);
1318 * Calculate the last possible buffered byte in a file. This must
1319 * include data that was buffered beyond the EOF by the write code.
1320 * This also needs to deal with overflowing the xfs_fsize_t type
1321 * which can happen for sizes near the limit.
1323 * We also need to take into account any blocks beyond the EOF. It
1324 * may be the case that they were buffered by a write which failed.
1325 * In that case the pages will still be in memory, but the inode size
1326 * will never have been updated.
1333 xfs_fsize_t last_byte
;
1334 xfs_fileoff_t last_block
;
1335 xfs_fileoff_t size_last_block
;
1338 ASSERT(ismrlocked(&(ip
->i_iolock
), MR_UPDATE
| MR_ACCESS
));
1342 * Only check for blocks beyond the EOF if the extents have
1343 * been read in. This eliminates the need for the inode lock,
1344 * and it also saves us from looking when it really isn't
1347 if (ip
->i_df
.if_flags
& XFS_IFEXTENTS
) {
1348 error
= xfs_bmap_last_offset(NULL
, ip
, &last_block
,
1356 size_last_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)ip
->i_size
);
1357 last_block
= XFS_FILEOFF_MAX(last_block
, size_last_block
);
1359 last_byte
= XFS_FSB_TO_B(mp
, last_block
);
1360 if (last_byte
< 0) {
1361 return XFS_MAXIOFFSET(mp
);
1363 last_byte
+= (1 << mp
->m_writeio_log
);
1364 if (last_byte
< 0) {
1365 return XFS_MAXIOFFSET(mp
);
1370 #if defined(XFS_RW_TRACE)
1376 xfs_fsize_t new_size
,
1377 xfs_off_t toss_start
,
1378 xfs_off_t toss_finish
)
1380 if (ip
->i_rwtrace
== NULL
) {
1384 ktrace_enter(ip
->i_rwtrace
,
1387 (void*)(unsigned long)((ip
->i_d
.di_size
>> 32) & 0xffffffff),
1388 (void*)(unsigned long)(ip
->i_d
.di_size
& 0xffffffff),
1389 (void*)((long)flag
),
1390 (void*)(unsigned long)((new_size
>> 32) & 0xffffffff),
1391 (void*)(unsigned long)(new_size
& 0xffffffff),
1392 (void*)(unsigned long)((toss_start
>> 32) & 0xffffffff),
1393 (void*)(unsigned long)(toss_start
& 0xffffffff),
1394 (void*)(unsigned long)((toss_finish
>> 32) & 0xffffffff),
1395 (void*)(unsigned long)(toss_finish
& 0xffffffff),
1396 (void*)(unsigned long)current_cpu(),
1397 (void*)(unsigned long)current_pid(),
1403 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1407 * Start the truncation of the file to new_size. The new size
1408 * must be smaller than the current size. This routine will
1409 * clear the buffer and page caches of file data in the removed
1410 * range, and xfs_itruncate_finish() will remove the underlying
1413 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1414 * must NOT have the inode lock held at all. This is because we're
1415 * calling into the buffer/page cache code and we can't hold the
1416 * inode lock when we do so.
1418 * We need to wait for any direct I/Os in flight to complete before we
1419 * proceed with the truncate. This is needed to prevent the extents
1420 * being read or written by the direct I/Os from being removed while the
1421 * I/O is in flight as there is no other method of synchronising
1422 * direct I/O with the truncate operation. Also, because we hold
1423 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1424 * started until the truncate completes and drops the lock. Essentially,
1425 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1426 * between direct I/Os and the truncate operation.
1428 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1429 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1430 * in the case that the caller is locking things out of order and
1431 * may not be able to call xfs_itruncate_finish() with the inode lock
1432 * held without dropping the I/O lock. If the caller must drop the
1433 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1434 * must be called again with all the same restrictions as the initial
1438 xfs_itruncate_start(
1441 xfs_fsize_t new_size
)
1443 xfs_fsize_t last_byte
;
1444 xfs_off_t toss_start
;
1449 ASSERT(ismrlocked(&ip
->i_iolock
, MR_UPDATE
) != 0);
1450 ASSERT((new_size
== 0) || (new_size
<= ip
->i_size
));
1451 ASSERT((flags
== XFS_ITRUNC_DEFINITE
) ||
1452 (flags
== XFS_ITRUNC_MAYBE
));
1457 vn_iowait(vp
); /* wait for the completion of any pending DIOs */
1460 * Call toss_pages or flushinval_pages to get rid of pages
1461 * overlapping the region being removed. We have to use
1462 * the less efficient flushinval_pages in the case that the
1463 * caller may not be able to finish the truncate without
1464 * dropping the inode's I/O lock. Make sure
1465 * to catch any pages brought in by buffers overlapping
1466 * the EOF by searching out beyond the isize by our
1467 * block size. We round new_size up to a block boundary
1468 * so that we don't toss things on the same block as
1469 * new_size but before it.
1471 * Before calling toss_page or flushinval_pages, make sure to
1472 * call remapf() over the same region if the file is mapped.
1473 * This frees up mapped file references to the pages in the
1474 * given range and for the flushinval_pages case it ensures
1475 * that we get the latest mapped changes flushed out.
1477 toss_start
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)new_size
);
1478 toss_start
= XFS_FSB_TO_B(mp
, toss_start
);
1479 if (toss_start
< 0) {
1481 * The place to start tossing is beyond our maximum
1482 * file size, so there is no way that the data extended
1487 last_byte
= xfs_file_last_byte(ip
);
1488 xfs_itrunc_trace(XFS_ITRUNC_START
, ip
, flags
, new_size
, toss_start
,
1490 if (last_byte
> toss_start
) {
1491 if (flags
& XFS_ITRUNC_DEFINITE
) {
1492 bhv_vop_toss_pages(vp
, toss_start
, -1, FI_REMAPF_LOCKED
);
1494 error
= bhv_vop_flushinval_pages(vp
, toss_start
, -1, FI_REMAPF_LOCKED
);
1499 if (new_size
== 0) {
1500 ASSERT(VN_CACHED(vp
) == 0);
1507 * Shrink the file to the given new_size. The new
1508 * size must be smaller than the current size.
1509 * This will free up the underlying blocks
1510 * in the removed range after a call to xfs_itruncate_start()
1511 * or xfs_atruncate_start().
1513 * The transaction passed to this routine must have made
1514 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1515 * This routine may commit the given transaction and
1516 * start new ones, so make sure everything involved in
1517 * the transaction is tidy before calling here.
1518 * Some transaction will be returned to the caller to be
1519 * committed. The incoming transaction must already include
1520 * the inode, and both inode locks must be held exclusively.
1521 * The inode must also be "held" within the transaction. On
1522 * return the inode will be "held" within the returned transaction.
1523 * This routine does NOT require any disk space to be reserved
1524 * for it within the transaction.
1526 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1527 * and it indicates the fork which is to be truncated. For the
1528 * attribute fork we only support truncation to size 0.
1530 * We use the sync parameter to indicate whether or not the first
1531 * transaction we perform might have to be synchronous. For the attr fork,
1532 * it needs to be so if the unlink of the inode is not yet known to be
1533 * permanent in the log. This keeps us from freeing and reusing the
1534 * blocks of the attribute fork before the unlink of the inode becomes
1537 * For the data fork, we normally have to run synchronously if we're
1538 * being called out of the inactive path or we're being called
1539 * out of the create path where we're truncating an existing file.
1540 * Either way, the truncate needs to be sync so blocks don't reappear
1541 * in the file with altered data in case of a crash. wsync filesystems
1542 * can run the first case async because anything that shrinks the inode
1543 * has to run sync so by the time we're called here from inactive, the
1544 * inode size is permanently set to 0.
1546 * Calls from the truncate path always need to be sync unless we're
1547 * in a wsync filesystem and the file has already been unlinked.
1549 * The caller is responsible for correctly setting the sync parameter.
1550 * It gets too hard for us to guess here which path we're being called
1551 * out of just based on inode state.
1554 xfs_itruncate_finish(
1557 xfs_fsize_t new_size
,
1561 xfs_fsblock_t first_block
;
1562 xfs_fileoff_t first_unmap_block
;
1563 xfs_fileoff_t last_block
;
1564 xfs_filblks_t unmap_len
=0;
1569 xfs_bmap_free_t free_list
;
1572 ASSERT(ismrlocked(&ip
->i_iolock
, MR_UPDATE
) != 0);
1573 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
) != 0);
1574 ASSERT((new_size
== 0) || (new_size
<= ip
->i_size
));
1575 ASSERT(*tp
!= NULL
);
1576 ASSERT((*tp
)->t_flags
& XFS_TRANS_PERM_LOG_RES
);
1577 ASSERT(ip
->i_transp
== *tp
);
1578 ASSERT(ip
->i_itemp
!= NULL
);
1579 ASSERT(ip
->i_itemp
->ili_flags
& XFS_ILI_HOLD
);
1583 mp
= (ntp
)->t_mountp
;
1584 ASSERT(! XFS_NOT_DQATTACHED(mp
, ip
));
1587 * We only support truncating the entire attribute fork.
1589 if (fork
== XFS_ATTR_FORK
) {
1592 first_unmap_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)new_size
);
1593 xfs_itrunc_trace(XFS_ITRUNC_FINISH1
, ip
, 0, new_size
, 0, 0);
1595 * The first thing we do is set the size to new_size permanently
1596 * on disk. This way we don't have to worry about anyone ever
1597 * being able to look at the data being freed even in the face
1598 * of a crash. What we're getting around here is the case where
1599 * we free a block, it is allocated to another file, it is written
1600 * to, and then we crash. If the new data gets written to the
1601 * file but the log buffers containing the free and reallocation
1602 * don't, then we'd end up with garbage in the blocks being freed.
1603 * As long as we make the new_size permanent before actually
1604 * freeing any blocks it doesn't matter if they get writtten to.
1606 * The callers must signal into us whether or not the size
1607 * setting here must be synchronous. There are a few cases
1608 * where it doesn't have to be synchronous. Those cases
1609 * occur if the file is unlinked and we know the unlink is
1610 * permanent or if the blocks being truncated are guaranteed
1611 * to be beyond the inode eof (regardless of the link count)
1612 * and the eof value is permanent. Both of these cases occur
1613 * only on wsync-mounted filesystems. In those cases, we're
1614 * guaranteed that no user will ever see the data in the blocks
1615 * that are being truncated so the truncate can run async.
1616 * In the free beyond eof case, the file may wind up with
1617 * more blocks allocated to it than it needs if we crash
1618 * and that won't get fixed until the next time the file
1619 * is re-opened and closed but that's ok as that shouldn't
1620 * be too many blocks.
1622 * However, we can't just make all wsync xactions run async
1623 * because there's one call out of the create path that needs
1624 * to run sync where it's truncating an existing file to size
1625 * 0 whose size is > 0.
1627 * It's probably possible to come up with a test in this
1628 * routine that would correctly distinguish all the above
1629 * cases from the values of the function parameters and the
1630 * inode state but for sanity's sake, I've decided to let the
1631 * layers above just tell us. It's simpler to correctly figure
1632 * out in the layer above exactly under what conditions we
1633 * can run async and I think it's easier for others read and
1634 * follow the logic in case something has to be changed.
1635 * cscope is your friend -- rcc.
1637 * The attribute fork is much simpler.
1639 * For the attribute fork we allow the caller to tell us whether
1640 * the unlink of the inode that led to this call is yet permanent
1641 * in the on disk log. If it is not and we will be freeing extents
1642 * in this inode then we make the first transaction synchronous
1643 * to make sure that the unlink is permanent by the time we free
1646 if (fork
== XFS_DATA_FORK
) {
1647 if (ip
->i_d
.di_nextents
> 0) {
1649 * If we are not changing the file size then do
1650 * not update the on-disk file size - we may be
1651 * called from xfs_inactive_free_eofblocks(). If we
1652 * update the on-disk file size and then the system
1653 * crashes before the contents of the file are
1654 * flushed to disk then the files may be full of
1655 * holes (ie NULL files bug).
1657 if (ip
->i_size
!= new_size
) {
1658 ip
->i_d
.di_size
= new_size
;
1659 ip
->i_size
= new_size
;
1660 xfs_trans_log_inode(ntp
, ip
, XFS_ILOG_CORE
);
1664 ASSERT(!(mp
->m_flags
& XFS_MOUNT_WSYNC
));
1665 if (ip
->i_d
.di_anextents
> 0)
1666 xfs_trans_set_sync(ntp
);
1668 ASSERT(fork
== XFS_DATA_FORK
||
1669 (fork
== XFS_ATTR_FORK
&&
1670 ((sync
&& !(mp
->m_flags
& XFS_MOUNT_WSYNC
)) ||
1671 (sync
== 0 && (mp
->m_flags
& XFS_MOUNT_WSYNC
)))));
1674 * Since it is possible for space to become allocated beyond
1675 * the end of the file (in a crash where the space is allocated
1676 * but the inode size is not yet updated), simply remove any
1677 * blocks which show up between the new EOF and the maximum
1678 * possible file size. If the first block to be removed is
1679 * beyond the maximum file size (ie it is the same as last_block),
1680 * then there is nothing to do.
1682 last_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
));
1683 ASSERT(first_unmap_block
<= last_block
);
1685 if (last_block
== first_unmap_block
) {
1688 unmap_len
= last_block
- first_unmap_block
+ 1;
1692 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1693 * will tell us whether it freed the entire range or
1694 * not. If this is a synchronous mount (wsync),
1695 * then we can tell bunmapi to keep all the
1696 * transactions asynchronous since the unlink
1697 * transaction that made this inode inactive has
1698 * already hit the disk. There's no danger of
1699 * the freed blocks being reused, there being a
1700 * crash, and the reused blocks suddenly reappearing
1701 * in this file with garbage in them once recovery
1704 XFS_BMAP_INIT(&free_list
, &first_block
);
1705 error
= XFS_BUNMAPI(mp
, ntp
, &ip
->i_iocore
,
1706 first_unmap_block
, unmap_len
,
1707 XFS_BMAPI_AFLAG(fork
) |
1708 (sync
? 0 : XFS_BMAPI_ASYNC
),
1709 XFS_ITRUNC_MAX_EXTENTS
,
1710 &first_block
, &free_list
,
1714 * If the bunmapi call encounters an error,
1715 * return to the caller where the transaction
1716 * can be properly aborted. We just need to
1717 * make sure we're not holding any resources
1718 * that we were not when we came in.
1720 xfs_bmap_cancel(&free_list
);
1725 * Duplicate the transaction that has the permanent
1726 * reservation and commit the old transaction.
1728 error
= xfs_bmap_finish(tp
, &free_list
, &committed
);
1732 * If the bmap finish call encounters an error,
1733 * return to the caller where the transaction
1734 * can be properly aborted. We just need to
1735 * make sure we're not holding any resources
1736 * that we were not when we came in.
1738 * Aborting from this point might lose some
1739 * blocks in the file system, but oh well.
1741 xfs_bmap_cancel(&free_list
);
1744 * If the passed in transaction committed
1745 * in xfs_bmap_finish(), then we want to
1746 * add the inode to this one before returning.
1747 * This keeps things simple for the higher
1748 * level code, because it always knows that
1749 * the inode is locked and held in the
1750 * transaction that returns to it whether
1751 * errors occur or not. We don't mark the
1752 * inode dirty so that this transaction can
1753 * be easily aborted if possible.
1755 xfs_trans_ijoin(ntp
, ip
,
1756 XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1757 xfs_trans_ihold(ntp
, ip
);
1764 * The first xact was committed,
1765 * so add the inode to the new one.
1766 * Mark it dirty so it will be logged
1767 * and moved forward in the log as
1768 * part of every commit.
1770 xfs_trans_ijoin(ntp
, ip
,
1771 XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1772 xfs_trans_ihold(ntp
, ip
);
1773 xfs_trans_log_inode(ntp
, ip
, XFS_ILOG_CORE
);
1775 ntp
= xfs_trans_dup(ntp
);
1776 (void) xfs_trans_commit(*tp
, 0);
1778 error
= xfs_trans_reserve(ntp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0,
1779 XFS_TRANS_PERM_LOG_RES
,
1780 XFS_ITRUNCATE_LOG_COUNT
);
1782 * Add the inode being truncated to the next chained
1785 xfs_trans_ijoin(ntp
, ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1786 xfs_trans_ihold(ntp
, ip
);
1791 * Only update the size in the case of the data fork, but
1792 * always re-log the inode so that our permanent transaction
1793 * can keep on rolling it forward in the log.
1795 if (fork
== XFS_DATA_FORK
) {
1796 xfs_isize_check(mp
, ip
, new_size
);
1798 * If we are not changing the file size then do
1799 * not update the on-disk file size - we may be
1800 * called from xfs_inactive_free_eofblocks(). If we
1801 * update the on-disk file size and then the system
1802 * crashes before the contents of the file are
1803 * flushed to disk then the files may be full of
1804 * holes (ie NULL files bug).
1806 if (ip
->i_size
!= new_size
) {
1807 ip
->i_d
.di_size
= new_size
;
1808 ip
->i_size
= new_size
;
1811 xfs_trans_log_inode(ntp
, ip
, XFS_ILOG_CORE
);
1812 ASSERT((new_size
!= 0) ||
1813 (fork
== XFS_ATTR_FORK
) ||
1814 (ip
->i_delayed_blks
== 0));
1815 ASSERT((new_size
!= 0) ||
1816 (fork
== XFS_ATTR_FORK
) ||
1817 (ip
->i_d
.di_nextents
== 0));
1818 xfs_itrunc_trace(XFS_ITRUNC_FINISH2
, ip
, 0, new_size
, 0, 0);
1826 * Do the first part of growing a file: zero any data in the last
1827 * block that is beyond the old EOF. We need to do this before
1828 * the inode is joined to the transaction to modify the i_size.
1829 * That way we can drop the inode lock and call into the buffer
1830 * cache to get the buffer mapping the EOF.
1835 xfs_fsize_t new_size
,
1840 ASSERT(ismrlocked(&(ip
->i_lock
), MR_UPDATE
) != 0);
1841 ASSERT(ismrlocked(&(ip
->i_iolock
), MR_UPDATE
) != 0);
1842 ASSERT(new_size
> ip
->i_size
);
1845 * Zero any pages that may have been created by
1846 * xfs_write_file() beyond the end of the file
1847 * and any blocks between the old and new file sizes.
1849 error
= xfs_zero_eof(XFS_ITOV(ip
), &ip
->i_iocore
, new_size
,
1857 * This routine is called to extend the size of a file.
1858 * The inode must have both the iolock and the ilock locked
1859 * for update and it must be a part of the current transaction.
1860 * The xfs_igrow_start() function must have been called previously.
1861 * If the change_flag is not zero, the inode change timestamp will
1868 xfs_fsize_t new_size
,
1871 ASSERT(ismrlocked(&(ip
->i_lock
), MR_UPDATE
) != 0);
1872 ASSERT(ismrlocked(&(ip
->i_iolock
), MR_UPDATE
) != 0);
1873 ASSERT(ip
->i_transp
== tp
);
1874 ASSERT(new_size
> ip
->i_size
);
1877 * Update the file size. Update the inode change timestamp
1878 * if change_flag set.
1880 ip
->i_d
.di_size
= new_size
;
1881 ip
->i_size
= new_size
;
1883 xfs_ichgtime(ip
, XFS_ICHGTIME_CHG
);
1884 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1890 * This is called when the inode's link count goes to 0.
1891 * We place the on-disk inode on a list in the AGI. It
1892 * will be pulled from this list when the inode is freed.
1904 xfs_agnumber_t agno
;
1905 xfs_daddr_t agdaddr
;
1912 ASSERT(ip
->i_d
.di_nlink
== 0);
1913 ASSERT(ip
->i_d
.di_mode
!= 0);
1914 ASSERT(ip
->i_transp
== tp
);
1918 agno
= XFS_INO_TO_AGNO(mp
, ip
->i_ino
);
1919 agdaddr
= XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
));
1922 * Get the agi buffer first. It ensures lock ordering
1925 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, agdaddr
,
1926 XFS_FSS_TO_BB(mp
, 1), 0, &agibp
);
1931 * Validate the magic number of the agi block.
1933 agi
= XFS_BUF_TO_AGI(agibp
);
1935 be32_to_cpu(agi
->agi_magicnum
) == XFS_AGI_MAGIC
&&
1936 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi
->agi_versionnum
));
1937 if (unlikely(XFS_TEST_ERROR(!agi_ok
, mp
, XFS_ERRTAG_IUNLINK
,
1938 XFS_RANDOM_IUNLINK
))) {
1939 XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW
, mp
, agi
);
1940 xfs_trans_brelse(tp
, agibp
);
1941 return XFS_ERROR(EFSCORRUPTED
);
1944 * Get the index into the agi hash table for the
1945 * list this inode will go on.
1947 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
1949 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
1950 ASSERT(agi
->agi_unlinked
[bucket_index
]);
1951 ASSERT(be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != agino
);
1953 if (be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != NULLAGINO
) {
1955 * There is already another inode in the bucket we need
1956 * to add ourselves to. Add us at the front of the list.
1957 * Here we put the head pointer into our next pointer,
1958 * and then we fall through to point the head at us.
1960 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, 0, 0);
1964 ASSERT(be32_to_cpu(dip
->di_next_unlinked
) == NULLAGINO
);
1965 /* both on-disk, don't endian flip twice */
1966 dip
->di_next_unlinked
= agi
->agi_unlinked
[bucket_index
];
1967 offset
= ip
->i_boffset
+
1968 offsetof(xfs_dinode_t
, di_next_unlinked
);
1969 xfs_trans_inode_buf(tp
, ibp
);
1970 xfs_trans_log_buf(tp
, ibp
, offset
,
1971 (offset
+ sizeof(xfs_agino_t
) - 1));
1972 xfs_inobp_check(mp
, ibp
);
1976 * Point the bucket head pointer at the inode being inserted.
1979 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(agino
);
1980 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
1981 (sizeof(xfs_agino_t
) * bucket_index
);
1982 xfs_trans_log_buf(tp
, agibp
, offset
,
1983 (offset
+ sizeof(xfs_agino_t
) - 1));
1988 * Pull the on-disk inode from the AGI unlinked list.
2001 xfs_agnumber_t agno
;
2002 xfs_daddr_t agdaddr
;
2004 xfs_agino_t next_agino
;
2005 xfs_buf_t
*last_ibp
;
2006 xfs_dinode_t
*last_dip
= NULL
;
2008 int offset
, last_offset
= 0;
2013 * First pull the on-disk inode from the AGI unlinked list.
2017 agno
= XFS_INO_TO_AGNO(mp
, ip
->i_ino
);
2018 agdaddr
= XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
));
2021 * Get the agi buffer first. It ensures lock ordering
2024 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, agdaddr
,
2025 XFS_FSS_TO_BB(mp
, 1), 0, &agibp
);
2028 "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
2029 error
, mp
->m_fsname
);
2033 * Validate the magic number of the agi block.
2035 agi
= XFS_BUF_TO_AGI(agibp
);
2037 be32_to_cpu(agi
->agi_magicnum
) == XFS_AGI_MAGIC
&&
2038 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi
->agi_versionnum
));
2039 if (unlikely(XFS_TEST_ERROR(!agi_ok
, mp
, XFS_ERRTAG_IUNLINK_REMOVE
,
2040 XFS_RANDOM_IUNLINK_REMOVE
))) {
2041 XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW
,
2043 xfs_trans_brelse(tp
, agibp
);
2045 "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
2047 return XFS_ERROR(EFSCORRUPTED
);
2050 * Get the index into the agi hash table for the
2051 * list this inode will go on.
2053 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
2055 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
2056 ASSERT(be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != NULLAGINO
);
2057 ASSERT(agi
->agi_unlinked
[bucket_index
]);
2059 if (be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) == agino
) {
2061 * We're at the head of the list. Get the inode's
2062 * on-disk buffer to see if there is anyone after us
2063 * on the list. Only modify our next pointer if it
2064 * is not already NULLAGINO. This saves us the overhead
2065 * of dealing with the buffer when there is no need to
2068 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, 0, 0);
2071 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2072 error
, mp
->m_fsname
);
2075 next_agino
= be32_to_cpu(dip
->di_next_unlinked
);
2076 ASSERT(next_agino
!= 0);
2077 if (next_agino
!= NULLAGINO
) {
2078 dip
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
2079 offset
= ip
->i_boffset
+
2080 offsetof(xfs_dinode_t
, di_next_unlinked
);
2081 xfs_trans_inode_buf(tp
, ibp
);
2082 xfs_trans_log_buf(tp
, ibp
, offset
,
2083 (offset
+ sizeof(xfs_agino_t
) - 1));
2084 xfs_inobp_check(mp
, ibp
);
2086 xfs_trans_brelse(tp
, ibp
);
2089 * Point the bucket head pointer at the next inode.
2091 ASSERT(next_agino
!= 0);
2092 ASSERT(next_agino
!= agino
);
2093 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(next_agino
);
2094 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
2095 (sizeof(xfs_agino_t
) * bucket_index
);
2096 xfs_trans_log_buf(tp
, agibp
, offset
,
2097 (offset
+ sizeof(xfs_agino_t
) - 1));
2100 * We need to search the list for the inode being freed.
2102 next_agino
= be32_to_cpu(agi
->agi_unlinked
[bucket_index
]);
2104 while (next_agino
!= agino
) {
2106 * If the last inode wasn't the one pointing to
2107 * us, then release its buffer since we're not
2108 * going to do anything with it.
2110 if (last_ibp
!= NULL
) {
2111 xfs_trans_brelse(tp
, last_ibp
);
2113 next_ino
= XFS_AGINO_TO_INO(mp
, agno
, next_agino
);
2114 error
= xfs_inotobp(mp
, tp
, next_ino
, &last_dip
,
2115 &last_ibp
, &last_offset
);
2118 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
2119 error
, mp
->m_fsname
);
2122 next_agino
= be32_to_cpu(last_dip
->di_next_unlinked
);
2123 ASSERT(next_agino
!= NULLAGINO
);
2124 ASSERT(next_agino
!= 0);
2127 * Now last_ibp points to the buffer previous to us on
2128 * the unlinked list. Pull us from the list.
2130 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, 0, 0);
2133 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2134 error
, mp
->m_fsname
);
2137 next_agino
= be32_to_cpu(dip
->di_next_unlinked
);
2138 ASSERT(next_agino
!= 0);
2139 ASSERT(next_agino
!= agino
);
2140 if (next_agino
!= NULLAGINO
) {
2141 dip
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
2142 offset
= ip
->i_boffset
+
2143 offsetof(xfs_dinode_t
, di_next_unlinked
);
2144 xfs_trans_inode_buf(tp
, ibp
);
2145 xfs_trans_log_buf(tp
, ibp
, offset
,
2146 (offset
+ sizeof(xfs_agino_t
) - 1));
2147 xfs_inobp_check(mp
, ibp
);
2149 xfs_trans_brelse(tp
, ibp
);
2152 * Point the previous inode on the list to the next inode.
2154 last_dip
->di_next_unlinked
= cpu_to_be32(next_agino
);
2155 ASSERT(next_agino
!= 0);
2156 offset
= last_offset
+ offsetof(xfs_dinode_t
, di_next_unlinked
);
2157 xfs_trans_inode_buf(tp
, last_ibp
);
2158 xfs_trans_log_buf(tp
, last_ibp
, offset
,
2159 (offset
+ sizeof(xfs_agino_t
) - 1));
2160 xfs_inobp_check(mp
, last_ibp
);
2165 STATIC_INLINE
int xfs_inode_clean(xfs_inode_t
*ip
)
2167 return (((ip
->i_itemp
== NULL
) ||
2168 !(ip
->i_itemp
->ili_format
.ilf_fields
& XFS_ILOG_ALL
)) &&
2169 (ip
->i_update_core
== 0));
2174 xfs_inode_t
*free_ip
,
2178 xfs_mount_t
*mp
= free_ip
->i_mount
;
2179 int blks_per_cluster
;
2182 int i
, j
, found
, pre_flushed
;
2186 xfs_inode_t
*ip
, **ip_found
;
2187 xfs_inode_log_item_t
*iip
;
2188 xfs_log_item_t
*lip
;
2191 if (mp
->m_sb
.sb_blocksize
>= XFS_INODE_CLUSTER_SIZE(mp
)) {
2192 blks_per_cluster
= 1;
2193 ninodes
= mp
->m_sb
.sb_inopblock
;
2194 nbufs
= XFS_IALLOC_BLOCKS(mp
);
2196 blks_per_cluster
= XFS_INODE_CLUSTER_SIZE(mp
) /
2197 mp
->m_sb
.sb_blocksize
;
2198 ninodes
= blks_per_cluster
* mp
->m_sb
.sb_inopblock
;
2199 nbufs
= XFS_IALLOC_BLOCKS(mp
) / blks_per_cluster
;
2202 ip_found
= kmem_alloc(ninodes
* sizeof(xfs_inode_t
*), KM_NOFS
);
2204 for (j
= 0; j
< nbufs
; j
++, inum
+= ninodes
) {
2205 blkno
= XFS_AGB_TO_DADDR(mp
, XFS_INO_TO_AGNO(mp
, inum
),
2206 XFS_INO_TO_AGBNO(mp
, inum
));
2210 * Look for each inode in memory and attempt to lock it,
2211 * we can be racing with flush and tail pushing here.
2212 * any inode we get the locks on, add to an array of
2213 * inode items to process later.
2215 * The get the buffer lock, we could beat a flush
2216 * or tail pushing thread to the lock here, in which
2217 * case they will go looking for the inode buffer
2218 * and fail, we need some other form of interlock
2222 for (i
= 0; i
< ninodes
; i
++) {
2223 ih
= XFS_IHASH(mp
, inum
+ i
);
2224 read_lock(&ih
->ih_lock
);
2225 for (ip
= ih
->ih_next
; ip
!= NULL
; ip
= ip
->i_next
) {
2226 if (ip
->i_ino
== inum
+ i
)
2230 /* Inode not in memory or we found it already,
2233 if (!ip
|| xfs_iflags_test(ip
, XFS_ISTALE
)) {
2234 read_unlock(&ih
->ih_lock
);
2238 if (xfs_inode_clean(ip
)) {
2239 read_unlock(&ih
->ih_lock
);
2243 /* If we can get the locks then add it to the
2244 * list, otherwise by the time we get the bp lock
2245 * below it will already be attached to the
2249 /* This inode will already be locked - by us, lets
2253 if (ip
== free_ip
) {
2254 if (xfs_iflock_nowait(ip
)) {
2255 xfs_iflags_set(ip
, XFS_ISTALE
);
2256 if (xfs_inode_clean(ip
)) {
2259 ip_found
[found
++] = ip
;
2262 read_unlock(&ih
->ih_lock
);
2266 if (xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
)) {
2267 if (xfs_iflock_nowait(ip
)) {
2268 xfs_iflags_set(ip
, XFS_ISTALE
);
2270 if (xfs_inode_clean(ip
)) {
2272 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2274 ip_found
[found
++] = ip
;
2277 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2281 read_unlock(&ih
->ih_lock
);
2284 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, blkno
,
2285 mp
->m_bsize
* blks_per_cluster
,
2289 lip
= XFS_BUF_FSPRIVATE(bp
, xfs_log_item_t
*);
2291 if (lip
->li_type
== XFS_LI_INODE
) {
2292 iip
= (xfs_inode_log_item_t
*)lip
;
2293 ASSERT(iip
->ili_logged
== 1);
2294 lip
->li_cb
= (void(*)(xfs_buf_t
*,xfs_log_item_t
*)) xfs_istale_done
;
2296 iip
->ili_flush_lsn
= iip
->ili_item
.li_lsn
;
2298 xfs_iflags_set(iip
->ili_inode
, XFS_ISTALE
);
2301 lip
= lip
->li_bio_list
;
2304 for (i
= 0; i
< found
; i
++) {
2309 ip
->i_update_core
= 0;
2311 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2315 iip
->ili_last_fields
= iip
->ili_format
.ilf_fields
;
2316 iip
->ili_format
.ilf_fields
= 0;
2317 iip
->ili_logged
= 1;
2319 iip
->ili_flush_lsn
= iip
->ili_item
.li_lsn
;
2322 xfs_buf_attach_iodone(bp
,
2323 (void(*)(xfs_buf_t
*,xfs_log_item_t
*))
2324 xfs_istale_done
, (xfs_log_item_t
*)iip
);
2325 if (ip
!= free_ip
) {
2326 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2330 if (found
|| pre_flushed
)
2331 xfs_trans_stale_inode_buf(tp
, bp
);
2332 xfs_trans_binval(tp
, bp
);
2335 kmem_free(ip_found
, ninodes
* sizeof(xfs_inode_t
*));
2339 * This is called to return an inode to the inode free list.
2340 * The inode should already be truncated to 0 length and have
2341 * no pages associated with it. This routine also assumes that
2342 * the inode is already a part of the transaction.
2344 * The on-disk copy of the inode will have been added to the list
2345 * of unlinked inodes in the AGI. We need to remove the inode from
2346 * that list atomically with respect to freeing it here.
2352 xfs_bmap_free_t
*flist
)
2356 xfs_ino_t first_ino
;
2358 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
));
2359 ASSERT(ip
->i_transp
== tp
);
2360 ASSERT(ip
->i_d
.di_nlink
== 0);
2361 ASSERT(ip
->i_d
.di_nextents
== 0);
2362 ASSERT(ip
->i_d
.di_anextents
== 0);
2363 ASSERT((ip
->i_d
.di_size
== 0 && ip
->i_size
== 0) ||
2364 ((ip
->i_d
.di_mode
& S_IFMT
) != S_IFREG
));
2365 ASSERT(ip
->i_d
.di_nblocks
== 0);
2368 * Pull the on-disk inode from the AGI unlinked list.
2370 error
= xfs_iunlink_remove(tp
, ip
);
2375 error
= xfs_difree(tp
, ip
->i_ino
, flist
, &delete, &first_ino
);
2379 ip
->i_d
.di_mode
= 0; /* mark incore inode as free */
2380 ip
->i_d
.di_flags
= 0;
2381 ip
->i_d
.di_dmevmask
= 0;
2382 ip
->i_d
.di_forkoff
= 0; /* mark the attr fork not in use */
2383 ip
->i_df
.if_ext_max
=
2384 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
2385 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
2386 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
2388 * Bump the generation count so no one will be confused
2389 * by reincarnations of this inode.
2392 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2395 xfs_ifree_cluster(ip
, tp
, first_ino
);
2402 * Reallocate the space for if_broot based on the number of records
2403 * being added or deleted as indicated in rec_diff. Move the records
2404 * and pointers in if_broot to fit the new size. When shrinking this
2405 * will eliminate holes between the records and pointers created by
2406 * the caller. When growing this will create holes to be filled in
2409 * The caller must not request to add more records than would fit in
2410 * the on-disk inode root. If the if_broot is currently NULL, then
2411 * if we adding records one will be allocated. The caller must also
2412 * not request that the number of records go below zero, although
2413 * it can go to zero.
2415 * ip -- the inode whose if_broot area is changing
2416 * ext_diff -- the change in the number of records, positive or negative,
2417 * requested for the if_broot array.
2427 xfs_bmbt_block_t
*new_broot
;
2434 * Handle the degenerate case quietly.
2436 if (rec_diff
== 0) {
2440 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2443 * If there wasn't any memory allocated before, just
2444 * allocate it now and get out.
2446 if (ifp
->if_broot_bytes
== 0) {
2447 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff
);
2448 ifp
->if_broot
= (xfs_bmbt_block_t
*)kmem_alloc(new_size
,
2450 ifp
->if_broot_bytes
= (int)new_size
;
2455 * If there is already an existing if_broot, then we need
2456 * to realloc() it and shift the pointers to their new
2457 * location. The records don't change location because
2458 * they are kept butted up against the btree block header.
2460 cur_max
= XFS_BMAP_BROOT_MAXRECS(ifp
->if_broot_bytes
);
2461 new_max
= cur_max
+ rec_diff
;
2462 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max
);
2463 ifp
->if_broot
= (xfs_bmbt_block_t
*)
2464 kmem_realloc(ifp
->if_broot
,
2466 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max
), /* old size */
2468 op
= (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp
->if_broot
, 1,
2469 ifp
->if_broot_bytes
);
2470 np
= (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp
->if_broot
, 1,
2472 ifp
->if_broot_bytes
= (int)new_size
;
2473 ASSERT(ifp
->if_broot_bytes
<=
2474 XFS_IFORK_SIZE(ip
, whichfork
) + XFS_BROOT_SIZE_ADJ
);
2475 memmove(np
, op
, cur_max
* (uint
)sizeof(xfs_dfsbno_t
));
2480 * rec_diff is less than 0. In this case, we are shrinking the
2481 * if_broot buffer. It must already exist. If we go to zero
2482 * records, just get rid of the root and clear the status bit.
2484 ASSERT((ifp
->if_broot
!= NULL
) && (ifp
->if_broot_bytes
> 0));
2485 cur_max
= XFS_BMAP_BROOT_MAXRECS(ifp
->if_broot_bytes
);
2486 new_max
= cur_max
+ rec_diff
;
2487 ASSERT(new_max
>= 0);
2489 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max
);
2493 new_broot
= (xfs_bmbt_block_t
*)kmem_alloc(new_size
, KM_SLEEP
);
2495 * First copy over the btree block header.
2497 memcpy(new_broot
, ifp
->if_broot
, sizeof(xfs_bmbt_block_t
));
2500 ifp
->if_flags
&= ~XFS_IFBROOT
;
2504 * Only copy the records and pointers if there are any.
2508 * First copy the records.
2510 op
= (char *)XFS_BMAP_BROOT_REC_ADDR(ifp
->if_broot
, 1,
2511 ifp
->if_broot_bytes
);
2512 np
= (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot
, 1,
2514 memcpy(np
, op
, new_max
* (uint
)sizeof(xfs_bmbt_rec_t
));
2517 * Then copy the pointers.
2519 op
= (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp
->if_broot
, 1,
2520 ifp
->if_broot_bytes
);
2521 np
= (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot
, 1,
2523 memcpy(np
, op
, new_max
* (uint
)sizeof(xfs_dfsbno_t
));
2525 kmem_free(ifp
->if_broot
, ifp
->if_broot_bytes
);
2526 ifp
->if_broot
= new_broot
;
2527 ifp
->if_broot_bytes
= (int)new_size
;
2528 ASSERT(ifp
->if_broot_bytes
<=
2529 XFS_IFORK_SIZE(ip
, whichfork
) + XFS_BROOT_SIZE_ADJ
);
2535 * This is called when the amount of space needed for if_data
2536 * is increased or decreased. The change in size is indicated by
2537 * the number of bytes that need to be added or deleted in the
2538 * byte_diff parameter.
2540 * If the amount of space needed has decreased below the size of the
2541 * inline buffer, then switch to using the inline buffer. Otherwise,
2542 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2543 * to what is needed.
2545 * ip -- the inode whose if_data area is changing
2546 * byte_diff -- the change in the number of bytes, positive or negative,
2547 * requested for the if_data array.
2559 if (byte_diff
== 0) {
2563 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2564 new_size
= (int)ifp
->if_bytes
+ byte_diff
;
2565 ASSERT(new_size
>= 0);
2567 if (new_size
== 0) {
2568 if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
2569 kmem_free(ifp
->if_u1
.if_data
, ifp
->if_real_bytes
);
2571 ifp
->if_u1
.if_data
= NULL
;
2573 } else if (new_size
<= sizeof(ifp
->if_u2
.if_inline_data
)) {
2575 * If the valid extents/data can fit in if_inline_ext/data,
2576 * copy them from the malloc'd vector and free it.
2578 if (ifp
->if_u1
.if_data
== NULL
) {
2579 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
2580 } else if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
2581 ASSERT(ifp
->if_real_bytes
!= 0);
2582 memcpy(ifp
->if_u2
.if_inline_data
, ifp
->if_u1
.if_data
,
2584 kmem_free(ifp
->if_u1
.if_data
, ifp
->if_real_bytes
);
2585 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
2590 * Stuck with malloc/realloc.
2591 * For inline data, the underlying buffer must be
2592 * a multiple of 4 bytes in size so that it can be
2593 * logged and stay on word boundaries. We enforce
2596 real_size
= roundup(new_size
, 4);
2597 if (ifp
->if_u1
.if_data
== NULL
) {
2598 ASSERT(ifp
->if_real_bytes
== 0);
2599 ifp
->if_u1
.if_data
= kmem_alloc(real_size
, KM_SLEEP
);
2600 } else if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
2602 * Only do the realloc if the underlying size
2603 * is really changing.
2605 if (ifp
->if_real_bytes
!= real_size
) {
2606 ifp
->if_u1
.if_data
=
2607 kmem_realloc(ifp
->if_u1
.if_data
,
2613 ASSERT(ifp
->if_real_bytes
== 0);
2614 ifp
->if_u1
.if_data
= kmem_alloc(real_size
, KM_SLEEP
);
2615 memcpy(ifp
->if_u1
.if_data
, ifp
->if_u2
.if_inline_data
,
2619 ifp
->if_real_bytes
= real_size
;
2620 ifp
->if_bytes
= new_size
;
2621 ASSERT(ifp
->if_bytes
<= XFS_IFORK_SIZE(ip
, whichfork
));
2628 * Map inode to disk block and offset.
2630 * mp -- the mount point structure for the current file system
2631 * tp -- the current transaction
2632 * ino -- the inode number of the inode to be located
2633 * imap -- this structure is filled in with the information necessary
2634 * to retrieve the given inode from disk
2635 * flags -- flags to pass to xfs_dilocate indicating whether or not
2636 * lookups in the inode btree were OK or not
2646 xfs_fsblock_t fsbno
;
2651 fsbno
= imap
->im_blkno
?
2652 XFS_DADDR_TO_FSB(mp
, imap
->im_blkno
) : NULLFSBLOCK
;
2653 error
= xfs_dilocate(mp
, tp
, ino
, &fsbno
, &len
, &off
, flags
);
2657 imap
->im_blkno
= XFS_FSB_TO_DADDR(mp
, fsbno
);
2658 imap
->im_len
= XFS_FSB_TO_BB(mp
, len
);
2659 imap
->im_agblkno
= XFS_FSB_TO_AGBNO(mp
, fsbno
);
2660 imap
->im_ioffset
= (ushort
)off
;
2661 imap
->im_boffset
= (ushort
)(off
<< mp
->m_sb
.sb_inodelog
);
2672 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2673 if (ifp
->if_broot
!= NULL
) {
2674 kmem_free(ifp
->if_broot
, ifp
->if_broot_bytes
);
2675 ifp
->if_broot
= NULL
;
2679 * If the format is local, then we can't have an extents
2680 * array so just look for an inline data array. If we're
2681 * not local then we may or may not have an extents list,
2682 * so check and free it up if we do.
2684 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
) {
2685 if ((ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) &&
2686 (ifp
->if_u1
.if_data
!= NULL
)) {
2687 ASSERT(ifp
->if_real_bytes
!= 0);
2688 kmem_free(ifp
->if_u1
.if_data
, ifp
->if_real_bytes
);
2689 ifp
->if_u1
.if_data
= NULL
;
2690 ifp
->if_real_bytes
= 0;
2692 } else if ((ifp
->if_flags
& XFS_IFEXTENTS
) &&
2693 ((ifp
->if_flags
& XFS_IFEXTIREC
) ||
2694 ((ifp
->if_u1
.if_extents
!= NULL
) &&
2695 (ifp
->if_u1
.if_extents
!= ifp
->if_u2
.if_inline_ext
)))) {
2696 ASSERT(ifp
->if_real_bytes
!= 0);
2697 xfs_iext_destroy(ifp
);
2699 ASSERT(ifp
->if_u1
.if_extents
== NULL
||
2700 ifp
->if_u1
.if_extents
== ifp
->if_u2
.if_inline_ext
);
2701 ASSERT(ifp
->if_real_bytes
== 0);
2702 if (whichfork
== XFS_ATTR_FORK
) {
2703 kmem_zone_free(xfs_ifork_zone
, ip
->i_afp
);
2709 * This is called free all the memory associated with an inode.
2710 * It must free the inode itself and any buffers allocated for
2711 * if_extents/if_data and if_broot. It must also free the lock
2712 * associated with the inode.
2719 switch (ip
->i_d
.di_mode
& S_IFMT
) {
2723 xfs_idestroy_fork(ip
, XFS_DATA_FORK
);
2727 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
2728 mrfree(&ip
->i_lock
);
2729 mrfree(&ip
->i_iolock
);
2730 freesema(&ip
->i_flock
);
2731 #ifdef XFS_BMAP_TRACE
2732 ktrace_free(ip
->i_xtrace
);
2734 #ifdef XFS_BMBT_TRACE
2735 ktrace_free(ip
->i_btrace
);
2738 ktrace_free(ip
->i_rwtrace
);
2740 #ifdef XFS_ILOCK_TRACE
2741 ktrace_free(ip
->i_lock_trace
);
2743 #ifdef XFS_DIR2_TRACE
2744 ktrace_free(ip
->i_dir_trace
);
2748 * Only if we are shutting down the fs will we see an
2749 * inode still in the AIL. If it is there, we should remove
2750 * it to prevent a use-after-free from occurring.
2752 xfs_mount_t
*mp
= ip
->i_mount
;
2753 xfs_log_item_t
*lip
= &ip
->i_itemp
->ili_item
;
2756 ASSERT(((lip
->li_flags
& XFS_LI_IN_AIL
) == 0) ||
2757 XFS_FORCED_SHUTDOWN(ip
->i_mount
));
2758 if (lip
->li_flags
& XFS_LI_IN_AIL
) {
2760 if (lip
->li_flags
& XFS_LI_IN_AIL
)
2761 xfs_trans_delete_ail(mp
, lip
, s
);
2765 xfs_inode_item_destroy(ip
);
2767 kmem_zone_free(xfs_inode_zone
, ip
);
2772 * Increment the pin count of the given buffer.
2773 * This value is protected by ipinlock spinlock in the mount structure.
2779 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
));
2781 atomic_inc(&ip
->i_pincount
);
2785 * Decrement the pin count of the given inode, and wake up
2786 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2787 * inode must have been previously pinned with a call to xfs_ipin().
2793 ASSERT(atomic_read(&ip
->i_pincount
) > 0);
2795 if (atomic_dec_and_lock(&ip
->i_pincount
, &ip
->i_flags_lock
)) {
2798 * If the inode is currently being reclaimed, the link between
2799 * the bhv_vnode and the xfs_inode will be broken after the
2800 * XFS_IRECLAIM* flag is set. Hence, if these flags are not
2801 * set, then we can move forward and mark the linux inode dirty
2802 * knowing that it is still valid as it won't freed until after
2803 * the bhv_vnode<->xfs_inode link is broken in xfs_reclaim. The
2804 * i_flags_lock is used to synchronise the setting of the
2805 * XFS_IRECLAIM* flags and the breaking of the link, and so we
2806 * can execute atomically w.r.t to reclaim by holding this lock
2809 * However, we still need to issue the unpin wakeup call as the
2810 * inode reclaim may be blocked waiting for the inode to become
2814 if (!__xfs_iflags_test(ip
, XFS_IRECLAIM
|XFS_IRECLAIMABLE
)) {
2815 bhv_vnode_t
*vp
= XFS_ITOV_NULL(ip
);
2816 struct inode
*inode
= NULL
;
2819 inode
= vn_to_inode(vp
);
2820 BUG_ON(inode
->i_state
& I_CLEAR
);
2822 /* make sync come back and flush this inode */
2823 if (!(inode
->i_state
& (I_NEW
|I_FREEING
)))
2824 mark_inode_dirty_sync(inode
);
2826 spin_unlock(&ip
->i_flags_lock
);
2827 wake_up(&ip
->i_ipin_wait
);
2832 * This is called to wait for the given inode to be unpinned.
2833 * It will sleep until this happens. The caller must have the
2834 * inode locked in at least shared mode so that the buffer cannot
2835 * be subsequently pinned once someone is waiting for it to be
2842 xfs_inode_log_item_t
*iip
;
2845 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
| MR_ACCESS
));
2847 if (atomic_read(&ip
->i_pincount
) == 0) {
2852 if (iip
&& iip
->ili_last_lsn
) {
2853 lsn
= iip
->ili_last_lsn
;
2859 * Give the log a push so we don't wait here too long.
2861 xfs_log_force(ip
->i_mount
, lsn
, XFS_LOG_FORCE
);
2863 wait_event(ip
->i_ipin_wait
, (atomic_read(&ip
->i_pincount
) == 0));
2868 * xfs_iextents_copy()
2870 * This is called to copy the REAL extents (as opposed to the delayed
2871 * allocation extents) from the inode into the given buffer. It
2872 * returns the number of bytes copied into the buffer.
2874 * If there are no delayed allocation extents, then we can just
2875 * memcpy() the extents into the buffer. Otherwise, we need to
2876 * examine each extent in turn and skip those which are delayed.
2888 xfs_fsblock_t start_block
;
2890 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2891 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
|MR_ACCESS
));
2892 ASSERT(ifp
->if_bytes
> 0);
2894 nrecs
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
2895 XFS_BMAP_TRACE_EXLIST(ip
, nrecs
, whichfork
);
2899 * There are some delayed allocation extents in the
2900 * inode, so copy the extents one at a time and skip
2901 * the delayed ones. There must be at least one
2902 * non-delayed extent.
2905 for (i
= 0; i
< nrecs
; i
++) {
2906 xfs_bmbt_rec_host_t
*ep
= xfs_iext_get_ext(ifp
, i
);
2907 start_block
= xfs_bmbt_get_startblock(ep
);
2908 if (ISNULLSTARTBLOCK(start_block
)) {
2910 * It's a delayed allocation extent, so skip it.
2915 /* Translate to on disk format */
2916 put_unaligned(cpu_to_be64(ep
->l0
), &dp
->l0
);
2917 put_unaligned(cpu_to_be64(ep
->l1
), &dp
->l1
);
2921 ASSERT(copied
!= 0);
2922 xfs_validate_extents(ifp
, copied
, XFS_EXTFMT_INODE(ip
));
2924 return (copied
* (uint
)sizeof(xfs_bmbt_rec_t
));
2928 * Each of the following cases stores data into the same region
2929 * of the on-disk inode, so only one of them can be valid at
2930 * any given time. While it is possible to have conflicting formats
2931 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2932 * in EXTENTS format, this can only happen when the fork has
2933 * changed formats after being modified but before being flushed.
2934 * In these cases, the format always takes precedence, because the
2935 * format indicates the current state of the fork.
2942 xfs_inode_log_item_t
*iip
,
2949 #ifdef XFS_TRANS_DEBUG
2952 static const short brootflag
[2] =
2953 { XFS_ILOG_DBROOT
, XFS_ILOG_ABROOT
};
2954 static const short dataflag
[2] =
2955 { XFS_ILOG_DDATA
, XFS_ILOG_ADATA
};
2956 static const short extflag
[2] =
2957 { XFS_ILOG_DEXT
, XFS_ILOG_AEXT
};
2961 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2963 * This can happen if we gave up in iformat in an error path,
2964 * for the attribute fork.
2967 ASSERT(whichfork
== XFS_ATTR_FORK
);
2970 cp
= XFS_DFORK_PTR(dip
, whichfork
);
2972 switch (XFS_IFORK_FORMAT(ip
, whichfork
)) {
2973 case XFS_DINODE_FMT_LOCAL
:
2974 if ((iip
->ili_format
.ilf_fields
& dataflag
[whichfork
]) &&
2975 (ifp
->if_bytes
> 0)) {
2976 ASSERT(ifp
->if_u1
.if_data
!= NULL
);
2977 ASSERT(ifp
->if_bytes
<= XFS_IFORK_SIZE(ip
, whichfork
));
2978 memcpy(cp
, ifp
->if_u1
.if_data
, ifp
->if_bytes
);
2982 case XFS_DINODE_FMT_EXTENTS
:
2983 ASSERT((ifp
->if_flags
& XFS_IFEXTENTS
) ||
2984 !(iip
->ili_format
.ilf_fields
& extflag
[whichfork
]));
2985 ASSERT((xfs_iext_get_ext(ifp
, 0) != NULL
) ||
2986 (ifp
->if_bytes
== 0));
2987 ASSERT((xfs_iext_get_ext(ifp
, 0) == NULL
) ||
2988 (ifp
->if_bytes
> 0));
2989 if ((iip
->ili_format
.ilf_fields
& extflag
[whichfork
]) &&
2990 (ifp
->if_bytes
> 0)) {
2991 ASSERT(XFS_IFORK_NEXTENTS(ip
, whichfork
) > 0);
2992 (void)xfs_iextents_copy(ip
, (xfs_bmbt_rec_t
*)cp
,
2997 case XFS_DINODE_FMT_BTREE
:
2998 if ((iip
->ili_format
.ilf_fields
& brootflag
[whichfork
]) &&
2999 (ifp
->if_broot_bytes
> 0)) {
3000 ASSERT(ifp
->if_broot
!= NULL
);
3001 ASSERT(ifp
->if_broot_bytes
<=
3002 (XFS_IFORK_SIZE(ip
, whichfork
) +
3003 XFS_BROOT_SIZE_ADJ
));
3004 xfs_bmbt_to_bmdr(ifp
->if_broot
, ifp
->if_broot_bytes
,
3005 (xfs_bmdr_block_t
*)cp
,
3006 XFS_DFORK_SIZE(dip
, mp
, whichfork
));
3010 case XFS_DINODE_FMT_DEV
:
3011 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_DEV
) {
3012 ASSERT(whichfork
== XFS_DATA_FORK
);
3013 dip
->di_u
.di_dev
= cpu_to_be32(ip
->i_df
.if_u2
.if_rdev
);
3017 case XFS_DINODE_FMT_UUID
:
3018 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_UUID
) {
3019 ASSERT(whichfork
== XFS_DATA_FORK
);
3020 memcpy(&dip
->di_u
.di_muuid
, &ip
->i_df
.if_u2
.if_uuid
,
3034 * xfs_iflush() will write a modified inode's changes out to the
3035 * inode's on disk home. The caller must have the inode lock held
3036 * in at least shared mode and the inode flush semaphore must be
3037 * held as well. The inode lock will still be held upon return from
3038 * the call and the caller is free to unlock it.
3039 * The inode flush lock will be unlocked when the inode reaches the disk.
3040 * The flags indicate how the inode's buffer should be written out.
3047 xfs_inode_log_item_t
*iip
;
3055 int clcount
; /* count of inodes clustered */
3057 enum { INT_DELWRI
= (1 << 0), INT_ASYNC
= (1 << 1) };
3060 XFS_STATS_INC(xs_iflush_count
);
3062 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
|MR_ACCESS
));
3063 ASSERT(issemalocked(&(ip
->i_flock
)));
3064 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
3065 ip
->i_d
.di_nextents
> ip
->i_df
.if_ext_max
);
3071 * If the inode isn't dirty, then just release the inode
3072 * flush lock and do nothing.
3074 if ((ip
->i_update_core
== 0) &&
3075 ((iip
== NULL
) || !(iip
->ili_format
.ilf_fields
& XFS_ILOG_ALL
))) {
3076 ASSERT((iip
!= NULL
) ?
3077 !(iip
->ili_item
.li_flags
& XFS_LI_IN_AIL
) : 1);
3083 * We can't flush the inode until it is unpinned, so
3084 * wait for it. We know noone new can pin it, because
3085 * we are holding the inode lock shared and you need
3086 * to hold it exclusively to pin the inode.
3088 xfs_iunpin_wait(ip
);
3091 * This may have been unpinned because the filesystem is shutting
3092 * down forcibly. If that's the case we must not write this inode
3093 * to disk, because the log record didn't make it to disk!
3095 if (XFS_FORCED_SHUTDOWN(mp
)) {
3096 ip
->i_update_core
= 0;
3098 iip
->ili_format
.ilf_fields
= 0;
3100 return XFS_ERROR(EIO
);
3104 * Get the buffer containing the on-disk inode.
3106 error
= xfs_itobp(mp
, NULL
, ip
, &dip
, &bp
, 0, 0);
3113 * Decide how buffer will be flushed out. This is done before
3114 * the call to xfs_iflush_int because this field is zeroed by it.
3116 if (iip
!= NULL
&& iip
->ili_format
.ilf_fields
!= 0) {
3118 * Flush out the inode buffer according to the directions
3119 * of the caller. In the cases where the caller has given
3120 * us a choice choose the non-delwri case. This is because
3121 * the inode is in the AIL and we need to get it out soon.
3124 case XFS_IFLUSH_SYNC
:
3125 case XFS_IFLUSH_DELWRI_ELSE_SYNC
:
3128 case XFS_IFLUSH_ASYNC
:
3129 case XFS_IFLUSH_DELWRI_ELSE_ASYNC
:
3132 case XFS_IFLUSH_DELWRI
:
3142 case XFS_IFLUSH_DELWRI_ELSE_SYNC
:
3143 case XFS_IFLUSH_DELWRI_ELSE_ASYNC
:
3144 case XFS_IFLUSH_DELWRI
:
3147 case XFS_IFLUSH_ASYNC
:
3150 case XFS_IFLUSH_SYNC
:
3161 * First flush out the inode that xfs_iflush was called with.
3163 error
= xfs_iflush_int(ip
, bp
);
3170 * see if other inodes can be gathered into this write
3173 ip
->i_chash
->chl_buf
= bp
;
3175 ch
= XFS_CHASH(mp
, ip
->i_blkno
);
3176 s
= mutex_spinlock(&ch
->ch_lock
);
3179 for (iq
= ip
->i_cnext
; iq
!= ip
; iq
= iq
->i_cnext
) {
3181 * Do an un-protected check to see if the inode is dirty and
3182 * is a candidate for flushing. These checks will be repeated
3183 * later after the appropriate locks are acquired.
3186 if ((iq
->i_update_core
== 0) &&
3188 !(iip
->ili_format
.ilf_fields
& XFS_ILOG_ALL
)) &&
3189 xfs_ipincount(iq
) == 0) {
3194 * Try to get locks. If any are unavailable,
3195 * then this inode cannot be flushed and is skipped.
3198 /* get inode locks (just i_lock) */
3199 if (xfs_ilock_nowait(iq
, XFS_ILOCK_SHARED
)) {
3200 /* get inode flush lock */
3201 if (xfs_iflock_nowait(iq
)) {
3202 /* check if pinned */
3203 if (xfs_ipincount(iq
) == 0) {
3204 /* arriving here means that
3205 * this inode can be flushed.
3206 * first re-check that it's
3210 if ((iq
->i_update_core
!= 0)||
3212 (iip
->ili_format
.ilf_fields
& XFS_ILOG_ALL
))) {
3214 error
= xfs_iflush_int(iq
, bp
);
3218 goto cluster_corrupt_out
;
3227 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
3230 mutex_spinunlock(&ch
->ch_lock
, s
);
3233 XFS_STATS_INC(xs_icluster_flushcnt
);
3234 XFS_STATS_ADD(xs_icluster_flushinode
, clcount
);
3238 * If the buffer is pinned then push on the log so we won't
3239 * get stuck waiting in the write for too long.
3241 if (XFS_BUF_ISPINNED(bp
)){
3242 xfs_log_force(mp
, (xfs_lsn_t
)0, XFS_LOG_FORCE
);
3245 if (flags
& INT_DELWRI
) {
3246 xfs_bdwrite(mp
, bp
);
3247 } else if (flags
& INT_ASYNC
) {
3248 xfs_bawrite(mp
, bp
);
3250 error
= xfs_bwrite(mp
, bp
);
3256 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
3257 xfs_iflush_abort(ip
);
3259 * Unlocks the flush lock
3261 return XFS_ERROR(EFSCORRUPTED
);
3263 cluster_corrupt_out
:
3264 /* Corruption detected in the clustering loop. Invalidate the
3265 * inode buffer and shut down the filesystem.
3267 mutex_spinunlock(&ch
->ch_lock
, s
);
3270 * Clean up the buffer. If it was B_DELWRI, just release it --
3271 * brelse can handle it with no problems. If not, shut down the
3272 * filesystem before releasing the buffer.
3274 if ((bufwasdelwri
= XFS_BUF_ISDELAYWRITE(bp
))) {
3278 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
3282 * Just like incore_relse: if we have b_iodone functions,
3283 * mark the buffer as an error and call them. Otherwise
3284 * mark it as stale and brelse.
3286 if (XFS_BUF_IODONE_FUNC(bp
)) {
3287 XFS_BUF_CLR_BDSTRAT_FUNC(bp
);
3291 XFS_BUF_ERROR(bp
,EIO
);
3299 xfs_iflush_abort(iq
);
3301 * Unlocks the flush lock
3303 return XFS_ERROR(EFSCORRUPTED
);
3312 xfs_inode_log_item_t
*iip
;
3315 #ifdef XFS_TRANS_DEBUG
3320 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
|MR_ACCESS
));
3321 ASSERT(issemalocked(&(ip
->i_flock
)));
3322 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
3323 ip
->i_d
.di_nextents
> ip
->i_df
.if_ext_max
);
3330 * If the inode isn't dirty, then just release the inode
3331 * flush lock and do nothing.
3333 if ((ip
->i_update_core
== 0) &&
3334 ((iip
== NULL
) || !(iip
->ili_format
.ilf_fields
& XFS_ILOG_ALL
))) {
3339 /* set *dip = inode's place in the buffer */
3340 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, ip
->i_boffset
);
3343 * Clear i_update_core before copying out the data.
3344 * This is for coordination with our timestamp updates
3345 * that don't hold the inode lock. They will always
3346 * update the timestamps BEFORE setting i_update_core,
3347 * so if we clear i_update_core after they set it we
3348 * are guaranteed to see their updates to the timestamps.
3349 * I believe that this depends on strongly ordered memory
3350 * semantics, but we have that. We use the SYNCHRONIZE
3351 * macro to make sure that the compiler does not reorder
3352 * the i_update_core access below the data copy below.
3354 ip
->i_update_core
= 0;
3358 * Make sure to get the latest atime from the Linux inode.
3360 xfs_synchronize_atime(ip
);
3362 if (XFS_TEST_ERROR(be16_to_cpu(dip
->di_core
.di_magic
) != XFS_DINODE_MAGIC
,
3363 mp
, XFS_ERRTAG_IFLUSH_1
, XFS_RANDOM_IFLUSH_1
)) {
3364 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3365 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3366 ip
->i_ino
, be16_to_cpu(dip
->di_core
.di_magic
), dip
);
3369 if (XFS_TEST_ERROR(ip
->i_d
.di_magic
!= XFS_DINODE_MAGIC
,
3370 mp
, XFS_ERRTAG_IFLUSH_2
, XFS_RANDOM_IFLUSH_2
)) {
3371 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3372 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3373 ip
->i_ino
, ip
, ip
->i_d
.di_magic
);
3376 if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
) {
3378 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3379 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
),
3380 mp
, XFS_ERRTAG_IFLUSH_3
, XFS_RANDOM_IFLUSH_3
)) {
3381 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3382 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3386 } else if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFDIR
) {
3388 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3389 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
) &&
3390 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_LOCAL
),
3391 mp
, XFS_ERRTAG_IFLUSH_4
, XFS_RANDOM_IFLUSH_4
)) {
3392 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3393 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3398 if (XFS_TEST_ERROR(ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
>
3399 ip
->i_d
.di_nblocks
, mp
, XFS_ERRTAG_IFLUSH_5
,
3400 XFS_RANDOM_IFLUSH_5
)) {
3401 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3402 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3404 ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
,
3409 if (XFS_TEST_ERROR(ip
->i_d
.di_forkoff
> mp
->m_sb
.sb_inodesize
,
3410 mp
, XFS_ERRTAG_IFLUSH_6
, XFS_RANDOM_IFLUSH_6
)) {
3411 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3412 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3413 ip
->i_ino
, ip
->i_d
.di_forkoff
, ip
);
3417 * bump the flush iteration count, used to detect flushes which
3418 * postdate a log record during recovery.
3421 ip
->i_d
.di_flushiter
++;
3424 * Copy the dirty parts of the inode into the on-disk
3425 * inode. We always copy out the core of the inode,
3426 * because if the inode is dirty at all the core must
3429 xfs_dinode_to_disk(&dip
->di_core
, &ip
->i_d
);
3431 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3432 if (ip
->i_d
.di_flushiter
== DI_MAX_FLUSH
)
3433 ip
->i_d
.di_flushiter
= 0;
3436 * If this is really an old format inode and the superblock version
3437 * has not been updated to support only new format inodes, then
3438 * convert back to the old inode format. If the superblock version
3439 * has been updated, then make the conversion permanent.
3441 ASSERT(ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
||
3442 XFS_SB_VERSION_HASNLINK(&mp
->m_sb
));
3443 if (ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
) {
3444 if (!XFS_SB_VERSION_HASNLINK(&mp
->m_sb
)) {
3448 ASSERT(ip
->i_d
.di_nlink
<= XFS_MAXLINK_1
);
3449 dip
->di_core
.di_onlink
= cpu_to_be16(ip
->i_d
.di_nlink
);
3452 * The superblock version has already been bumped,
3453 * so just make the conversion to the new inode
3456 ip
->i_d
.di_version
= XFS_DINODE_VERSION_2
;
3457 dip
->di_core
.di_version
= XFS_DINODE_VERSION_2
;
3458 ip
->i_d
.di_onlink
= 0;
3459 dip
->di_core
.di_onlink
= 0;
3460 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
3461 memset(&(dip
->di_core
.di_pad
[0]), 0,
3462 sizeof(dip
->di_core
.di_pad
));
3463 ASSERT(ip
->i_d
.di_projid
== 0);
3467 if (xfs_iflush_fork(ip
, dip
, iip
, XFS_DATA_FORK
, bp
) == EFSCORRUPTED
) {
3471 if (XFS_IFORK_Q(ip
)) {
3473 * The only error from xfs_iflush_fork is on the data fork.
3475 (void) xfs_iflush_fork(ip
, dip
, iip
, XFS_ATTR_FORK
, bp
);
3477 xfs_inobp_check(mp
, bp
);
3480 * We've recorded everything logged in the inode, so we'd
3481 * like to clear the ilf_fields bits so we don't log and
3482 * flush things unnecessarily. However, we can't stop
3483 * logging all this information until the data we've copied
3484 * into the disk buffer is written to disk. If we did we might
3485 * overwrite the copy of the inode in the log with all the
3486 * data after re-logging only part of it, and in the face of
3487 * a crash we wouldn't have all the data we need to recover.
3489 * What we do is move the bits to the ili_last_fields field.
3490 * When logging the inode, these bits are moved back to the
3491 * ilf_fields field. In the xfs_iflush_done() routine we
3492 * clear ili_last_fields, since we know that the information
3493 * those bits represent is permanently on disk. As long as
3494 * the flush completes before the inode is logged again, then
3495 * both ilf_fields and ili_last_fields will be cleared.
3497 * We can play with the ilf_fields bits here, because the inode
3498 * lock must be held exclusively in order to set bits there
3499 * and the flush lock protects the ili_last_fields bits.
3500 * Set ili_logged so the flush done
3501 * routine can tell whether or not to look in the AIL.
3502 * Also, store the current LSN of the inode so that we can tell
3503 * whether the item has moved in the AIL from xfs_iflush_done().
3504 * In order to read the lsn we need the AIL lock, because
3505 * it is a 64 bit value that cannot be read atomically.
3507 if (iip
!= NULL
&& iip
->ili_format
.ilf_fields
!= 0) {
3508 iip
->ili_last_fields
= iip
->ili_format
.ilf_fields
;
3509 iip
->ili_format
.ilf_fields
= 0;
3510 iip
->ili_logged
= 1;
3512 ASSERT(sizeof(xfs_lsn_t
) == 8); /* don't lock if it shrinks */
3514 iip
->ili_flush_lsn
= iip
->ili_item
.li_lsn
;
3518 * Attach the function xfs_iflush_done to the inode's
3519 * buffer. This will remove the inode from the AIL
3520 * and unlock the inode's flush lock when the inode is
3521 * completely written to disk.
3523 xfs_buf_attach_iodone(bp
, (void(*)(xfs_buf_t
*,xfs_log_item_t
*))
3524 xfs_iflush_done
, (xfs_log_item_t
*)iip
);
3526 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) != NULL
);
3527 ASSERT(XFS_BUF_IODONE_FUNC(bp
) != NULL
);
3530 * We're flushing an inode which is not in the AIL and has
3531 * not been logged but has i_update_core set. For this
3532 * case we can use a B_DELWRI flush and immediately drop
3533 * the inode flush lock because we can avoid the whole
3534 * AIL state thing. It's OK to drop the flush lock now,
3535 * because we've already locked the buffer and to do anything
3536 * you really need both.
3539 ASSERT(iip
->ili_logged
== 0);
3540 ASSERT(iip
->ili_last_fields
== 0);
3541 ASSERT((iip
->ili_item
.li_flags
& XFS_LI_IN_AIL
) == 0);
3549 return XFS_ERROR(EFSCORRUPTED
);
3554 * Flush all inactive inodes in mp.
3564 XFS_MOUNT_ILOCK(mp
);
3570 /* Make sure we skip markers inserted by sync */
3571 if (ip
->i_mount
== NULL
) {
3576 vp
= XFS_ITOV_NULL(ip
);
3578 XFS_MOUNT_IUNLOCK(mp
);
3579 xfs_finish_reclaim(ip
, 0, XFS_IFLUSH_ASYNC
);
3583 ASSERT(vn_count(vp
) == 0);
3586 } while (ip
!= mp
->m_inodes
);
3588 XFS_MOUNT_IUNLOCK(mp
);
3592 * xfs_iaccess: check accessibility of inode for mode.
3601 mode_t orgmode
= mode
;
3602 struct inode
*inode
= vn_to_inode(XFS_ITOV(ip
));
3604 if (mode
& S_IWUSR
) {
3605 umode_t imode
= inode
->i_mode
;
3607 if (IS_RDONLY(inode
) &&
3608 (S_ISREG(imode
) || S_ISDIR(imode
) || S_ISLNK(imode
)))
3609 return XFS_ERROR(EROFS
);
3611 if (IS_IMMUTABLE(inode
))
3612 return XFS_ERROR(EACCES
);
3616 * If there's an Access Control List it's used instead of
3619 if ((error
= _ACL_XFS_IACCESS(ip
, mode
, cr
)) != -1)
3620 return error
? XFS_ERROR(error
) : 0;
3622 if (current_fsuid(cr
) != ip
->i_d
.di_uid
) {
3624 if (!in_group_p((gid_t
)ip
->i_d
.di_gid
))
3629 * If the DACs are ok we don't need any capability check.
3631 if ((ip
->i_d
.di_mode
& mode
) == mode
)
3634 * Read/write DACs are always overridable.
3635 * Executable DACs are overridable if at least one exec bit is set.
3637 if (!(orgmode
& S_IXUSR
) ||
3638 (inode
->i_mode
& S_IXUGO
) || S_ISDIR(inode
->i_mode
))
3639 if (capable_cred(cr
, CAP_DAC_OVERRIDE
))
3642 if ((orgmode
== S_IRUSR
) ||
3643 (S_ISDIR(inode
->i_mode
) && (!(orgmode
& S_IWUSR
)))) {
3644 if (capable_cred(cr
, CAP_DAC_READ_SEARCH
))
3647 cmn_err(CE_NOTE
, "Ick: mode=%o, orgmode=%o", mode
, orgmode
);
3649 return XFS_ERROR(EACCES
);
3651 return XFS_ERROR(EACCES
);
3655 * xfs_iroundup: round up argument to next power of two
3664 if ((v
& (v
- 1)) == 0)
3666 ASSERT((v
& 0x80000000) == 0);
3667 if ((v
& (v
+ 1)) == 0)
3669 for (i
= 0, m
= 1; i
< 31; i
++, m
<<= 1) {
3673 if ((v
& (v
+ 1)) == 0)
3680 #ifdef XFS_ILOCK_TRACE
3681 ktrace_t
*xfs_ilock_trace_buf
;
3684 xfs_ilock_trace(xfs_inode_t
*ip
, int lock
, unsigned int lockflags
, inst_t
*ra
)
3686 ktrace_enter(ip
->i_lock_trace
,
3688 (void *)(unsigned long)lock
, /* 1 = LOCK, 3=UNLOCK, etc */
3689 (void *)(unsigned long)lockflags
, /* XFS_ILOCK_EXCL etc */
3690 (void *)ra
, /* caller of ilock */
3691 (void *)(unsigned long)current_cpu(),
3692 (void *)(unsigned long)current_pid(),
3693 NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
);
3698 * Return a pointer to the extent record at file index idx.
3700 xfs_bmbt_rec_host_t
*
3702 xfs_ifork_t
*ifp
, /* inode fork pointer */
3703 xfs_extnum_t idx
) /* index of target extent */
3706 if ((ifp
->if_flags
& XFS_IFEXTIREC
) && (idx
== 0)) {
3707 return ifp
->if_u1
.if_ext_irec
->er_extbuf
;
3708 } else if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3709 xfs_ext_irec_t
*erp
; /* irec pointer */
3710 int erp_idx
= 0; /* irec index */
3711 xfs_extnum_t page_idx
= idx
; /* ext index in target list */
3713 erp
= xfs_iext_idx_to_irec(ifp
, &page_idx
, &erp_idx
, 0);
3714 return &erp
->er_extbuf
[page_idx
];
3715 } else if (ifp
->if_bytes
) {
3716 return &ifp
->if_u1
.if_extents
[idx
];
3723 * Insert new item(s) into the extent records for incore inode
3724 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3728 xfs_ifork_t
*ifp
, /* inode fork pointer */
3729 xfs_extnum_t idx
, /* starting index of new items */
3730 xfs_extnum_t count
, /* number of inserted items */
3731 xfs_bmbt_irec_t
*new) /* items to insert */
3733 xfs_extnum_t i
; /* extent record index */
3735 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
3736 xfs_iext_add(ifp
, idx
, count
);
3737 for (i
= idx
; i
< idx
+ count
; i
++, new++)
3738 xfs_bmbt_set_all(xfs_iext_get_ext(ifp
, i
), new);
3742 * This is called when the amount of space required for incore file
3743 * extents needs to be increased. The ext_diff parameter stores the
3744 * number of new extents being added and the idx parameter contains
3745 * the extent index where the new extents will be added. If the new
3746 * extents are being appended, then we just need to (re)allocate and
3747 * initialize the space. Otherwise, if the new extents are being
3748 * inserted into the middle of the existing entries, a bit more work
3749 * is required to make room for the new extents to be inserted. The
3750 * caller is responsible for filling in the new extent entries upon
3755 xfs_ifork_t
*ifp
, /* inode fork pointer */
3756 xfs_extnum_t idx
, /* index to begin adding exts */
3757 int ext_diff
) /* number of extents to add */
3759 int byte_diff
; /* new bytes being added */
3760 int new_size
; /* size of extents after adding */
3761 xfs_extnum_t nextents
; /* number of extents in file */
3763 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3764 ASSERT((idx
>= 0) && (idx
<= nextents
));
3765 byte_diff
= ext_diff
* sizeof(xfs_bmbt_rec_t
);
3766 new_size
= ifp
->if_bytes
+ byte_diff
;
3768 * If the new number of extents (nextents + ext_diff)
3769 * fits inside the inode, then continue to use the inline
3772 if (nextents
+ ext_diff
<= XFS_INLINE_EXTS
) {
3773 if (idx
< nextents
) {
3774 memmove(&ifp
->if_u2
.if_inline_ext
[idx
+ ext_diff
],
3775 &ifp
->if_u2
.if_inline_ext
[idx
],
3776 (nextents
- idx
) * sizeof(xfs_bmbt_rec_t
));
3777 memset(&ifp
->if_u2
.if_inline_ext
[idx
], 0, byte_diff
);
3779 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
3780 ifp
->if_real_bytes
= 0;
3781 ifp
->if_lastex
= nextents
+ ext_diff
;
3784 * Otherwise use a linear (direct) extent list.
3785 * If the extents are currently inside the inode,
3786 * xfs_iext_realloc_direct will switch us from
3787 * inline to direct extent allocation mode.
3789 else if (nextents
+ ext_diff
<= XFS_LINEAR_EXTS
) {
3790 xfs_iext_realloc_direct(ifp
, new_size
);
3791 if (idx
< nextents
) {
3792 memmove(&ifp
->if_u1
.if_extents
[idx
+ ext_diff
],
3793 &ifp
->if_u1
.if_extents
[idx
],
3794 (nextents
- idx
) * sizeof(xfs_bmbt_rec_t
));
3795 memset(&ifp
->if_u1
.if_extents
[idx
], 0, byte_diff
);
3798 /* Indirection array */
3800 xfs_ext_irec_t
*erp
;
3804 ASSERT(nextents
+ ext_diff
> XFS_LINEAR_EXTS
);
3805 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3806 erp
= xfs_iext_idx_to_irec(ifp
, &page_idx
, &erp_idx
, 1);
3808 xfs_iext_irec_init(ifp
);
3809 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3810 erp
= ifp
->if_u1
.if_ext_irec
;
3812 /* Extents fit in target extent page */
3813 if (erp
&& erp
->er_extcount
+ ext_diff
<= XFS_LINEAR_EXTS
) {
3814 if (page_idx
< erp
->er_extcount
) {
3815 memmove(&erp
->er_extbuf
[page_idx
+ ext_diff
],
3816 &erp
->er_extbuf
[page_idx
],
3817 (erp
->er_extcount
- page_idx
) *
3818 sizeof(xfs_bmbt_rec_t
));
3819 memset(&erp
->er_extbuf
[page_idx
], 0, byte_diff
);
3821 erp
->er_extcount
+= ext_diff
;
3822 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, ext_diff
);
3824 /* Insert a new extent page */
3826 xfs_iext_add_indirect_multi(ifp
,
3827 erp_idx
, page_idx
, ext_diff
);
3830 * If extent(s) are being appended to the last page in
3831 * the indirection array and the new extent(s) don't fit
3832 * in the page, then erp is NULL and erp_idx is set to
3833 * the next index needed in the indirection array.
3836 int count
= ext_diff
;
3839 erp
= xfs_iext_irec_new(ifp
, erp_idx
);
3840 erp
->er_extcount
= count
;
3841 count
-= MIN(count
, (int)XFS_LINEAR_EXTS
);
3848 ifp
->if_bytes
= new_size
;
3852 * This is called when incore extents are being added to the indirection
3853 * array and the new extents do not fit in the target extent list. The
3854 * erp_idx parameter contains the irec index for the target extent list
3855 * in the indirection array, and the idx parameter contains the extent
3856 * index within the list. The number of extents being added is stored
3857 * in the count parameter.
3859 * |-------| |-------|
3860 * | | | | idx - number of extents before idx
3862 * | | | | count - number of extents being inserted at idx
3863 * |-------| |-------|
3864 * | count | | nex2 | nex2 - number of extents after idx + count
3865 * |-------| |-------|
3868 xfs_iext_add_indirect_multi(
3869 xfs_ifork_t
*ifp
, /* inode fork pointer */
3870 int erp_idx
, /* target extent irec index */
3871 xfs_extnum_t idx
, /* index within target list */
3872 int count
) /* new extents being added */
3874 int byte_diff
; /* new bytes being added */
3875 xfs_ext_irec_t
*erp
; /* pointer to irec entry */
3876 xfs_extnum_t ext_diff
; /* number of extents to add */
3877 xfs_extnum_t ext_cnt
; /* new extents still needed */
3878 xfs_extnum_t nex2
; /* extents after idx + count */
3879 xfs_bmbt_rec_t
*nex2_ep
= NULL
; /* temp list for nex2 extents */
3880 int nlists
; /* number of irec's (lists) */
3882 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3883 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
3884 nex2
= erp
->er_extcount
- idx
;
3885 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3888 * Save second part of target extent list
3889 * (all extents past */
3891 byte_diff
= nex2
* sizeof(xfs_bmbt_rec_t
);
3892 nex2_ep
= (xfs_bmbt_rec_t
*) kmem_alloc(byte_diff
, KM_SLEEP
);
3893 memmove(nex2_ep
, &erp
->er_extbuf
[idx
], byte_diff
);
3894 erp
->er_extcount
-= nex2
;
3895 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, -nex2
);
3896 memset(&erp
->er_extbuf
[idx
], 0, byte_diff
);
3900 * Add the new extents to the end of the target
3901 * list, then allocate new irec record(s) and
3902 * extent buffer(s) as needed to store the rest
3903 * of the new extents.
3906 ext_diff
= MIN(ext_cnt
, (int)XFS_LINEAR_EXTS
- erp
->er_extcount
);
3908 erp
->er_extcount
+= ext_diff
;
3909 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, ext_diff
);
3910 ext_cnt
-= ext_diff
;
3914 erp
= xfs_iext_irec_new(ifp
, erp_idx
);
3915 ext_diff
= MIN(ext_cnt
, (int)XFS_LINEAR_EXTS
);
3916 erp
->er_extcount
= ext_diff
;
3917 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, ext_diff
);
3918 ext_cnt
-= ext_diff
;
3921 /* Add nex2 extents back to indirection array */
3923 xfs_extnum_t ext_avail
;
3926 byte_diff
= nex2
* sizeof(xfs_bmbt_rec_t
);
3927 ext_avail
= XFS_LINEAR_EXTS
- erp
->er_extcount
;
3930 * If nex2 extents fit in the current page, append
3931 * nex2_ep after the new extents.
3933 if (nex2
<= ext_avail
) {
3934 i
= erp
->er_extcount
;
3937 * Otherwise, check if space is available in the
3940 else if ((erp_idx
< nlists
- 1) &&
3941 (nex2
<= (ext_avail
= XFS_LINEAR_EXTS
-
3942 ifp
->if_u1
.if_ext_irec
[erp_idx
+1].er_extcount
))) {
3945 /* Create a hole for nex2 extents */
3946 memmove(&erp
->er_extbuf
[nex2
], erp
->er_extbuf
,
3947 erp
->er_extcount
* sizeof(xfs_bmbt_rec_t
));
3950 * Final choice, create a new extent page for
3955 erp
= xfs_iext_irec_new(ifp
, erp_idx
);
3957 memmove(&erp
->er_extbuf
[i
], nex2_ep
, byte_diff
);
3958 kmem_free(nex2_ep
, byte_diff
);
3959 erp
->er_extcount
+= nex2
;
3960 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, nex2
);
3965 * This is called when the amount of space required for incore file
3966 * extents needs to be decreased. The ext_diff parameter stores the
3967 * number of extents to be removed and the idx parameter contains
3968 * the extent index where the extents will be removed from.
3970 * If the amount of space needed has decreased below the linear
3971 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3972 * extent array. Otherwise, use kmem_realloc() to adjust the
3973 * size to what is needed.
3977 xfs_ifork_t
*ifp
, /* inode fork pointer */
3978 xfs_extnum_t idx
, /* index to begin removing exts */
3979 int ext_diff
) /* number of extents to remove */
3981 xfs_extnum_t nextents
; /* number of extents in file */
3982 int new_size
; /* size of extents after removal */
3984 ASSERT(ext_diff
> 0);
3985 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3986 new_size
= (nextents
- ext_diff
) * sizeof(xfs_bmbt_rec_t
);
3988 if (new_size
== 0) {
3989 xfs_iext_destroy(ifp
);
3990 } else if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3991 xfs_iext_remove_indirect(ifp
, idx
, ext_diff
);
3992 } else if (ifp
->if_real_bytes
) {
3993 xfs_iext_remove_direct(ifp
, idx
, ext_diff
);
3995 xfs_iext_remove_inline(ifp
, idx
, ext_diff
);
3997 ifp
->if_bytes
= new_size
;
4001 * This removes ext_diff extents from the inline buffer, beginning
4002 * at extent index idx.
4005 xfs_iext_remove_inline(
4006 xfs_ifork_t
*ifp
, /* inode fork pointer */
4007 xfs_extnum_t idx
, /* index to begin removing exts */
4008 int ext_diff
) /* number of extents to remove */
4010 int nextents
; /* number of extents in file */
4012 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
));
4013 ASSERT(idx
< XFS_INLINE_EXTS
);
4014 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4015 ASSERT(((nextents
- ext_diff
) > 0) &&
4016 (nextents
- ext_diff
) < XFS_INLINE_EXTS
);
4018 if (idx
+ ext_diff
< nextents
) {
4019 memmove(&ifp
->if_u2
.if_inline_ext
[idx
],
4020 &ifp
->if_u2
.if_inline_ext
[idx
+ ext_diff
],
4021 (nextents
- (idx
+ ext_diff
)) *
4022 sizeof(xfs_bmbt_rec_t
));
4023 memset(&ifp
->if_u2
.if_inline_ext
[nextents
- ext_diff
],
4024 0, ext_diff
* sizeof(xfs_bmbt_rec_t
));
4026 memset(&ifp
->if_u2
.if_inline_ext
[idx
], 0,
4027 ext_diff
* sizeof(xfs_bmbt_rec_t
));
4032 * This removes ext_diff extents from a linear (direct) extent list,
4033 * beginning at extent index idx. If the extents are being removed
4034 * from the end of the list (ie. truncate) then we just need to re-
4035 * allocate the list to remove the extra space. Otherwise, if the
4036 * extents are being removed from the middle of the existing extent
4037 * entries, then we first need to move the extent records beginning
4038 * at idx + ext_diff up in the list to overwrite the records being
4039 * removed, then remove the extra space via kmem_realloc.
4042 xfs_iext_remove_direct(
4043 xfs_ifork_t
*ifp
, /* inode fork pointer */
4044 xfs_extnum_t idx
, /* index to begin removing exts */
4045 int ext_diff
) /* number of extents to remove */
4047 xfs_extnum_t nextents
; /* number of extents in file */
4048 int new_size
; /* size of extents after removal */
4050 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
));
4051 new_size
= ifp
->if_bytes
-
4052 (ext_diff
* sizeof(xfs_bmbt_rec_t
));
4053 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4055 if (new_size
== 0) {
4056 xfs_iext_destroy(ifp
);
4059 /* Move extents up in the list (if needed) */
4060 if (idx
+ ext_diff
< nextents
) {
4061 memmove(&ifp
->if_u1
.if_extents
[idx
],
4062 &ifp
->if_u1
.if_extents
[idx
+ ext_diff
],
4063 (nextents
- (idx
+ ext_diff
)) *
4064 sizeof(xfs_bmbt_rec_t
));
4066 memset(&ifp
->if_u1
.if_extents
[nextents
- ext_diff
],
4067 0, ext_diff
* sizeof(xfs_bmbt_rec_t
));
4069 * Reallocate the direct extent list. If the extents
4070 * will fit inside the inode then xfs_iext_realloc_direct
4071 * will switch from direct to inline extent allocation
4074 xfs_iext_realloc_direct(ifp
, new_size
);
4075 ifp
->if_bytes
= new_size
;
4079 * This is called when incore extents are being removed from the
4080 * indirection array and the extents being removed span multiple extent
4081 * buffers. The idx parameter contains the file extent index where we
4082 * want to begin removing extents, and the count parameter contains
4083 * how many extents need to be removed.
4085 * |-------| |-------|
4086 * | nex1 | | | nex1 - number of extents before idx
4087 * |-------| | count |
4088 * | | | | count - number of extents being removed at idx
4089 * | count | |-------|
4090 * | | | nex2 | nex2 - number of extents after idx + count
4091 * |-------| |-------|
4094 xfs_iext_remove_indirect(
4095 xfs_ifork_t
*ifp
, /* inode fork pointer */
4096 xfs_extnum_t idx
, /* index to begin removing extents */
4097 int count
) /* number of extents to remove */
4099 xfs_ext_irec_t
*erp
; /* indirection array pointer */
4100 int erp_idx
= 0; /* indirection array index */
4101 xfs_extnum_t ext_cnt
; /* extents left to remove */
4102 xfs_extnum_t ext_diff
; /* extents to remove in current list */
4103 xfs_extnum_t nex1
; /* number of extents before idx */
4104 xfs_extnum_t nex2
; /* extents after idx + count */
4105 int nlists
; /* entries in indirection array */
4106 int page_idx
= idx
; /* index in target extent list */
4108 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4109 erp
= xfs_iext_idx_to_irec(ifp
, &page_idx
, &erp_idx
, 0);
4110 ASSERT(erp
!= NULL
);
4111 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4115 nex2
= MAX((erp
->er_extcount
- (nex1
+ ext_cnt
)), 0);
4116 ext_diff
= MIN(ext_cnt
, (erp
->er_extcount
- nex1
));
4118 * Check for deletion of entire list;
4119 * xfs_iext_irec_remove() updates extent offsets.
4121 if (ext_diff
== erp
->er_extcount
) {
4122 xfs_iext_irec_remove(ifp
, erp_idx
);
4123 ext_cnt
-= ext_diff
;
4126 ASSERT(erp_idx
< ifp
->if_real_bytes
/
4128 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4135 /* Move extents up (if needed) */
4137 memmove(&erp
->er_extbuf
[nex1
],
4138 &erp
->er_extbuf
[nex1
+ ext_diff
],
4139 nex2
* sizeof(xfs_bmbt_rec_t
));
4141 /* Zero out rest of page */
4142 memset(&erp
->er_extbuf
[nex1
+ nex2
], 0, (XFS_IEXT_BUFSZ
-
4143 ((nex1
+ nex2
) * sizeof(xfs_bmbt_rec_t
))));
4144 /* Update remaining counters */
4145 erp
->er_extcount
-= ext_diff
;
4146 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, -ext_diff
);
4147 ext_cnt
-= ext_diff
;
4152 ifp
->if_bytes
-= count
* sizeof(xfs_bmbt_rec_t
);
4153 xfs_iext_irec_compact(ifp
);
4157 * Create, destroy, or resize a linear (direct) block of extents.
4160 xfs_iext_realloc_direct(
4161 xfs_ifork_t
*ifp
, /* inode fork pointer */
4162 int new_size
) /* new size of extents */
4164 int rnew_size
; /* real new size of extents */
4166 rnew_size
= new_size
;
4168 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
) ||
4169 ((new_size
>= 0) && (new_size
<= XFS_IEXT_BUFSZ
) &&
4170 (new_size
!= ifp
->if_real_bytes
)));
4172 /* Free extent records */
4173 if (new_size
== 0) {
4174 xfs_iext_destroy(ifp
);
4176 /* Resize direct extent list and zero any new bytes */
4177 else if (ifp
->if_real_bytes
) {
4178 /* Check if extents will fit inside the inode */
4179 if (new_size
<= XFS_INLINE_EXTS
* sizeof(xfs_bmbt_rec_t
)) {
4180 xfs_iext_direct_to_inline(ifp
, new_size
/
4181 (uint
)sizeof(xfs_bmbt_rec_t
));
4182 ifp
->if_bytes
= new_size
;
4185 if (!is_power_of_2(new_size
)){
4186 rnew_size
= xfs_iroundup(new_size
);
4188 if (rnew_size
!= ifp
->if_real_bytes
) {
4189 ifp
->if_u1
.if_extents
=
4190 kmem_realloc(ifp
->if_u1
.if_extents
,
4195 if (rnew_size
> ifp
->if_real_bytes
) {
4196 memset(&ifp
->if_u1
.if_extents
[ifp
->if_bytes
/
4197 (uint
)sizeof(xfs_bmbt_rec_t
)], 0,
4198 rnew_size
- ifp
->if_real_bytes
);
4202 * Switch from the inline extent buffer to a direct
4203 * extent list. Be sure to include the inline extent
4204 * bytes in new_size.
4207 new_size
+= ifp
->if_bytes
;
4208 if (!is_power_of_2(new_size
)) {
4209 rnew_size
= xfs_iroundup(new_size
);
4211 xfs_iext_inline_to_direct(ifp
, rnew_size
);
4213 ifp
->if_real_bytes
= rnew_size
;
4214 ifp
->if_bytes
= new_size
;
4218 * Switch from linear (direct) extent records to inline buffer.
4221 xfs_iext_direct_to_inline(
4222 xfs_ifork_t
*ifp
, /* inode fork pointer */
4223 xfs_extnum_t nextents
) /* number of extents in file */
4225 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
4226 ASSERT(nextents
<= XFS_INLINE_EXTS
);
4228 * The inline buffer was zeroed when we switched
4229 * from inline to direct extent allocation mode,
4230 * so we don't need to clear it here.
4232 memcpy(ifp
->if_u2
.if_inline_ext
, ifp
->if_u1
.if_extents
,
4233 nextents
* sizeof(xfs_bmbt_rec_t
));
4234 kmem_free(ifp
->if_u1
.if_extents
, ifp
->if_real_bytes
);
4235 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
4236 ifp
->if_real_bytes
= 0;
4240 * Switch from inline buffer to linear (direct) extent records.
4241 * new_size should already be rounded up to the next power of 2
4242 * by the caller (when appropriate), so use new_size as it is.
4243 * However, since new_size may be rounded up, we can't update
4244 * if_bytes here. It is the caller's responsibility to update
4245 * if_bytes upon return.
4248 xfs_iext_inline_to_direct(
4249 xfs_ifork_t
*ifp
, /* inode fork pointer */
4250 int new_size
) /* number of extents in file */
4252 ifp
->if_u1
.if_extents
= kmem_alloc(new_size
, KM_SLEEP
);
4253 memset(ifp
->if_u1
.if_extents
, 0, new_size
);
4254 if (ifp
->if_bytes
) {
4255 memcpy(ifp
->if_u1
.if_extents
, ifp
->if_u2
.if_inline_ext
,
4257 memset(ifp
->if_u2
.if_inline_ext
, 0, XFS_INLINE_EXTS
*
4258 sizeof(xfs_bmbt_rec_t
));
4260 ifp
->if_real_bytes
= new_size
;
4264 * Resize an extent indirection array to new_size bytes.
4267 xfs_iext_realloc_indirect(
4268 xfs_ifork_t
*ifp
, /* inode fork pointer */
4269 int new_size
) /* new indirection array size */
4271 int nlists
; /* number of irec's (ex lists) */
4272 int size
; /* current indirection array size */
4274 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4275 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4276 size
= nlists
* sizeof(xfs_ext_irec_t
);
4277 ASSERT(ifp
->if_real_bytes
);
4278 ASSERT((new_size
>= 0) && (new_size
!= size
));
4279 if (new_size
== 0) {
4280 xfs_iext_destroy(ifp
);
4282 ifp
->if_u1
.if_ext_irec
= (xfs_ext_irec_t
*)
4283 kmem_realloc(ifp
->if_u1
.if_ext_irec
,
4284 new_size
, size
, KM_SLEEP
);
4289 * Switch from indirection array to linear (direct) extent allocations.
4292 xfs_iext_indirect_to_direct(
4293 xfs_ifork_t
*ifp
) /* inode fork pointer */
4295 xfs_bmbt_rec_host_t
*ep
; /* extent record pointer */
4296 xfs_extnum_t nextents
; /* number of extents in file */
4297 int size
; /* size of file extents */
4299 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4300 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4301 ASSERT(nextents
<= XFS_LINEAR_EXTS
);
4302 size
= nextents
* sizeof(xfs_bmbt_rec_t
);
4304 xfs_iext_irec_compact_full(ifp
);
4305 ASSERT(ifp
->if_real_bytes
== XFS_IEXT_BUFSZ
);
4307 ep
= ifp
->if_u1
.if_ext_irec
->er_extbuf
;
4308 kmem_free(ifp
->if_u1
.if_ext_irec
, sizeof(xfs_ext_irec_t
));
4309 ifp
->if_flags
&= ~XFS_IFEXTIREC
;
4310 ifp
->if_u1
.if_extents
= ep
;
4311 ifp
->if_bytes
= size
;
4312 if (nextents
< XFS_LINEAR_EXTS
) {
4313 xfs_iext_realloc_direct(ifp
, size
);
4318 * Free incore file extents.
4322 xfs_ifork_t
*ifp
) /* inode fork pointer */
4324 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
4328 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4329 for (erp_idx
= nlists
- 1; erp_idx
>= 0 ; erp_idx
--) {
4330 xfs_iext_irec_remove(ifp
, erp_idx
);
4332 ifp
->if_flags
&= ~XFS_IFEXTIREC
;
4333 } else if (ifp
->if_real_bytes
) {
4334 kmem_free(ifp
->if_u1
.if_extents
, ifp
->if_real_bytes
);
4335 } else if (ifp
->if_bytes
) {
4336 memset(ifp
->if_u2
.if_inline_ext
, 0, XFS_INLINE_EXTS
*
4337 sizeof(xfs_bmbt_rec_t
));
4339 ifp
->if_u1
.if_extents
= NULL
;
4340 ifp
->if_real_bytes
= 0;
4345 * Return a pointer to the extent record for file system block bno.
4347 xfs_bmbt_rec_host_t
* /* pointer to found extent record */
4348 xfs_iext_bno_to_ext(
4349 xfs_ifork_t
*ifp
, /* inode fork pointer */
4350 xfs_fileoff_t bno
, /* block number to search for */
4351 xfs_extnum_t
*idxp
) /* index of target extent */
4353 xfs_bmbt_rec_host_t
*base
; /* pointer to first extent */
4354 xfs_filblks_t blockcount
= 0; /* number of blocks in extent */
4355 xfs_bmbt_rec_host_t
*ep
= NULL
; /* pointer to target extent */
4356 xfs_ext_irec_t
*erp
= NULL
; /* indirection array pointer */
4357 int high
; /* upper boundary in search */
4358 xfs_extnum_t idx
= 0; /* index of target extent */
4359 int low
; /* lower boundary in search */
4360 xfs_extnum_t nextents
; /* number of file extents */
4361 xfs_fileoff_t startoff
= 0; /* start offset of extent */
4363 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4364 if (nextents
== 0) {
4369 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
4370 /* Find target extent list */
4372 erp
= xfs_iext_bno_to_irec(ifp
, bno
, &erp_idx
);
4373 base
= erp
->er_extbuf
;
4374 high
= erp
->er_extcount
- 1;
4376 base
= ifp
->if_u1
.if_extents
;
4377 high
= nextents
- 1;
4379 /* Binary search extent records */
4380 while (low
<= high
) {
4381 idx
= (low
+ high
) >> 1;
4383 startoff
= xfs_bmbt_get_startoff(ep
);
4384 blockcount
= xfs_bmbt_get_blockcount(ep
);
4385 if (bno
< startoff
) {
4387 } else if (bno
>= startoff
+ blockcount
) {
4390 /* Convert back to file-based extent index */
4391 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
4392 idx
+= erp
->er_extoff
;
4398 /* Convert back to file-based extent index */
4399 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
4400 idx
+= erp
->er_extoff
;
4402 if (bno
>= startoff
+ blockcount
) {
4403 if (++idx
== nextents
) {
4406 ep
= xfs_iext_get_ext(ifp
, idx
);
4414 * Return a pointer to the indirection array entry containing the
4415 * extent record for filesystem block bno. Store the index of the
4416 * target irec in *erp_idxp.
4418 xfs_ext_irec_t
* /* pointer to found extent record */
4419 xfs_iext_bno_to_irec(
4420 xfs_ifork_t
*ifp
, /* inode fork pointer */
4421 xfs_fileoff_t bno
, /* block number to search for */
4422 int *erp_idxp
) /* irec index of target ext list */
4424 xfs_ext_irec_t
*erp
= NULL
; /* indirection array pointer */
4425 xfs_ext_irec_t
*erp_next
; /* next indirection array entry */
4426 int erp_idx
; /* indirection array index */
4427 int nlists
; /* number of extent irec's (lists) */
4428 int high
; /* binary search upper limit */
4429 int low
; /* binary search lower limit */
4431 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4432 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4436 while (low
<= high
) {
4437 erp_idx
= (low
+ high
) >> 1;
4438 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4439 erp_next
= erp_idx
< nlists
- 1 ? erp
+ 1 : NULL
;
4440 if (bno
< xfs_bmbt_get_startoff(erp
->er_extbuf
)) {
4442 } else if (erp_next
&& bno
>=
4443 xfs_bmbt_get_startoff(erp_next
->er_extbuf
)) {
4449 *erp_idxp
= erp_idx
;
4454 * Return a pointer to the indirection array entry containing the
4455 * extent record at file extent index *idxp. Store the index of the
4456 * target irec in *erp_idxp and store the page index of the target
4457 * extent record in *idxp.
4460 xfs_iext_idx_to_irec(
4461 xfs_ifork_t
*ifp
, /* inode fork pointer */
4462 xfs_extnum_t
*idxp
, /* extent index (file -> page) */
4463 int *erp_idxp
, /* pointer to target irec */
4464 int realloc
) /* new bytes were just added */
4466 xfs_ext_irec_t
*prev
; /* pointer to previous irec */
4467 xfs_ext_irec_t
*erp
= NULL
; /* pointer to current irec */
4468 int erp_idx
; /* indirection array index */
4469 int nlists
; /* number of irec's (ex lists) */
4470 int high
; /* binary search upper limit */
4471 int low
; /* binary search lower limit */
4472 xfs_extnum_t page_idx
= *idxp
; /* extent index in target list */
4474 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4475 ASSERT(page_idx
>= 0 && page_idx
<=
4476 ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
));
4477 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4482 /* Binary search extent irec's */
4483 while (low
<= high
) {
4484 erp_idx
= (low
+ high
) >> 1;
4485 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4486 prev
= erp_idx
> 0 ? erp
- 1 : NULL
;
4487 if (page_idx
< erp
->er_extoff
|| (page_idx
== erp
->er_extoff
&&
4488 realloc
&& prev
&& prev
->er_extcount
< XFS_LINEAR_EXTS
)) {
4490 } else if (page_idx
> erp
->er_extoff
+ erp
->er_extcount
||
4491 (page_idx
== erp
->er_extoff
+ erp
->er_extcount
&&
4494 } else if (page_idx
== erp
->er_extoff
+ erp
->er_extcount
&&
4495 erp
->er_extcount
== XFS_LINEAR_EXTS
) {
4499 erp
= erp_idx
< nlists
? erp
+ 1 : NULL
;
4502 page_idx
-= erp
->er_extoff
;
4507 *erp_idxp
= erp_idx
;
4512 * Allocate and initialize an indirection array once the space needed
4513 * for incore extents increases above XFS_IEXT_BUFSZ.
4517 xfs_ifork_t
*ifp
) /* inode fork pointer */
4519 xfs_ext_irec_t
*erp
; /* indirection array pointer */
4520 xfs_extnum_t nextents
; /* number of extents in file */
4522 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
));
4523 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4524 ASSERT(nextents
<= XFS_LINEAR_EXTS
);
4526 erp
= (xfs_ext_irec_t
*)
4527 kmem_alloc(sizeof(xfs_ext_irec_t
), KM_SLEEP
);
4529 if (nextents
== 0) {
4530 ifp
->if_u1
.if_extents
= kmem_alloc(XFS_IEXT_BUFSZ
, KM_SLEEP
);
4531 } else if (!ifp
->if_real_bytes
) {
4532 xfs_iext_inline_to_direct(ifp
, XFS_IEXT_BUFSZ
);
4533 } else if (ifp
->if_real_bytes
< XFS_IEXT_BUFSZ
) {
4534 xfs_iext_realloc_direct(ifp
, XFS_IEXT_BUFSZ
);
4536 erp
->er_extbuf
= ifp
->if_u1
.if_extents
;
4537 erp
->er_extcount
= nextents
;
4540 ifp
->if_flags
|= XFS_IFEXTIREC
;
4541 ifp
->if_real_bytes
= XFS_IEXT_BUFSZ
;
4542 ifp
->if_bytes
= nextents
* sizeof(xfs_bmbt_rec_t
);
4543 ifp
->if_u1
.if_ext_irec
= erp
;
4549 * Allocate and initialize a new entry in the indirection array.
4553 xfs_ifork_t
*ifp
, /* inode fork pointer */
4554 int erp_idx
) /* index for new irec */
4556 xfs_ext_irec_t
*erp
; /* indirection array pointer */
4557 int i
; /* loop counter */
4558 int nlists
; /* number of irec's (ex lists) */
4560 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4561 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4563 /* Resize indirection array */
4564 xfs_iext_realloc_indirect(ifp
, ++nlists
*
4565 sizeof(xfs_ext_irec_t
));
4567 * Move records down in the array so the
4568 * new page can use erp_idx.
4570 erp
= ifp
->if_u1
.if_ext_irec
;
4571 for (i
= nlists
- 1; i
> erp_idx
; i
--) {
4572 memmove(&erp
[i
], &erp
[i
-1], sizeof(xfs_ext_irec_t
));
4574 ASSERT(i
== erp_idx
);
4576 /* Initialize new extent record */
4577 erp
= ifp
->if_u1
.if_ext_irec
;
4578 erp
[erp_idx
].er_extbuf
= kmem_alloc(XFS_IEXT_BUFSZ
, KM_SLEEP
);
4579 ifp
->if_real_bytes
= nlists
* XFS_IEXT_BUFSZ
;
4580 memset(erp
[erp_idx
].er_extbuf
, 0, XFS_IEXT_BUFSZ
);
4581 erp
[erp_idx
].er_extcount
= 0;
4582 erp
[erp_idx
].er_extoff
= erp_idx
> 0 ?
4583 erp
[erp_idx
-1].er_extoff
+ erp
[erp_idx
-1].er_extcount
: 0;
4584 return (&erp
[erp_idx
]);
4588 * Remove a record from the indirection array.
4591 xfs_iext_irec_remove(
4592 xfs_ifork_t
*ifp
, /* inode fork pointer */
4593 int erp_idx
) /* irec index to remove */
4595 xfs_ext_irec_t
*erp
; /* indirection array pointer */
4596 int i
; /* loop counter */
4597 int nlists
; /* number of irec's (ex lists) */
4599 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4600 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4601 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4602 if (erp
->er_extbuf
) {
4603 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1,
4605 kmem_free(erp
->er_extbuf
, XFS_IEXT_BUFSZ
);
4607 /* Compact extent records */
4608 erp
= ifp
->if_u1
.if_ext_irec
;
4609 for (i
= erp_idx
; i
< nlists
- 1; i
++) {
4610 memmove(&erp
[i
], &erp
[i
+1], sizeof(xfs_ext_irec_t
));
4613 * Manually free the last extent record from the indirection
4614 * array. A call to xfs_iext_realloc_indirect() with a size
4615 * of zero would result in a call to xfs_iext_destroy() which
4616 * would in turn call this function again, creating a nasty
4620 xfs_iext_realloc_indirect(ifp
,
4621 nlists
* sizeof(xfs_ext_irec_t
));
4623 kmem_free(ifp
->if_u1
.if_ext_irec
,
4624 sizeof(xfs_ext_irec_t
));
4626 ifp
->if_real_bytes
= nlists
* XFS_IEXT_BUFSZ
;
4630 * This is called to clean up large amounts of unused memory allocated
4631 * by the indirection array. Before compacting anything though, verify
4632 * that the indirection array is still needed and switch back to the
4633 * linear extent list (or even the inline buffer) if possible. The
4634 * compaction policy is as follows:
4636 * Full Compaction: Extents fit into a single page (or inline buffer)
4637 * Full Compaction: Extents occupy less than 10% of allocated space
4638 * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
4639 * No Compaction: Extents occupy at least 50% of allocated space
4642 xfs_iext_irec_compact(
4643 xfs_ifork_t
*ifp
) /* inode fork pointer */
4645 xfs_extnum_t nextents
; /* number of extents in file */
4646 int nlists
; /* number of irec's (ex lists) */
4648 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4649 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4650 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4652 if (nextents
== 0) {
4653 xfs_iext_destroy(ifp
);
4654 } else if (nextents
<= XFS_INLINE_EXTS
) {
4655 xfs_iext_indirect_to_direct(ifp
);
4656 xfs_iext_direct_to_inline(ifp
, nextents
);
4657 } else if (nextents
<= XFS_LINEAR_EXTS
) {
4658 xfs_iext_indirect_to_direct(ifp
);
4659 } else if (nextents
< (nlists
* XFS_LINEAR_EXTS
) >> 3) {
4660 xfs_iext_irec_compact_full(ifp
);
4661 } else if (nextents
< (nlists
* XFS_LINEAR_EXTS
) >> 1) {
4662 xfs_iext_irec_compact_pages(ifp
);
4667 * Combine extents from neighboring extent pages.
4670 xfs_iext_irec_compact_pages(
4671 xfs_ifork_t
*ifp
) /* inode fork pointer */
4673 xfs_ext_irec_t
*erp
, *erp_next
;/* pointers to irec entries */
4674 int erp_idx
= 0; /* indirection array index */
4675 int nlists
; /* number of irec's (ex lists) */
4677 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4678 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4679 while (erp_idx
< nlists
- 1) {
4680 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4682 if (erp_next
->er_extcount
<=
4683 (XFS_LINEAR_EXTS
- erp
->er_extcount
)) {
4684 memmove(&erp
->er_extbuf
[erp
->er_extcount
],
4685 erp_next
->er_extbuf
, erp_next
->er_extcount
*
4686 sizeof(xfs_bmbt_rec_t
));
4687 erp
->er_extcount
+= erp_next
->er_extcount
;
4689 * Free page before removing extent record
4690 * so er_extoffs don't get modified in
4691 * xfs_iext_irec_remove.
4693 kmem_free(erp_next
->er_extbuf
, XFS_IEXT_BUFSZ
);
4694 erp_next
->er_extbuf
= NULL
;
4695 xfs_iext_irec_remove(ifp
, erp_idx
+ 1);
4696 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4704 * Fully compact the extent records managed by the indirection array.
4707 xfs_iext_irec_compact_full(
4708 xfs_ifork_t
*ifp
) /* inode fork pointer */
4710 xfs_bmbt_rec_host_t
*ep
, *ep_next
; /* extent record pointers */
4711 xfs_ext_irec_t
*erp
, *erp_next
; /* extent irec pointers */
4712 int erp_idx
= 0; /* extent irec index */
4713 int ext_avail
; /* empty entries in ex list */
4714 int ext_diff
; /* number of exts to add */
4715 int nlists
; /* number of irec's (ex lists) */
4717 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4718 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4719 erp
= ifp
->if_u1
.if_ext_irec
;
4720 ep
= &erp
->er_extbuf
[erp
->er_extcount
];
4722 ep_next
= erp_next
->er_extbuf
;
4723 while (erp_idx
< nlists
- 1) {
4724 ext_avail
= XFS_LINEAR_EXTS
- erp
->er_extcount
;
4725 ext_diff
= MIN(ext_avail
, erp_next
->er_extcount
);
4726 memcpy(ep
, ep_next
, ext_diff
* sizeof(xfs_bmbt_rec_t
));
4727 erp
->er_extcount
+= ext_diff
;
4728 erp_next
->er_extcount
-= ext_diff
;
4729 /* Remove next page */
4730 if (erp_next
->er_extcount
== 0) {
4732 * Free page before removing extent record
4733 * so er_extoffs don't get modified in
4734 * xfs_iext_irec_remove.
4736 kmem_free(erp_next
->er_extbuf
,
4737 erp_next
->er_extcount
* sizeof(xfs_bmbt_rec_t
));
4738 erp_next
->er_extbuf
= NULL
;
4739 xfs_iext_irec_remove(ifp
, erp_idx
+ 1);
4740 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4741 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4742 /* Update next page */
4744 /* Move rest of page up to become next new page */
4745 memmove(erp_next
->er_extbuf
, ep_next
,
4746 erp_next
->er_extcount
* sizeof(xfs_bmbt_rec_t
));
4747 ep_next
= erp_next
->er_extbuf
;
4748 memset(&ep_next
[erp_next
->er_extcount
], 0,
4749 (XFS_LINEAR_EXTS
- erp_next
->er_extcount
) *
4750 sizeof(xfs_bmbt_rec_t
));
4752 if (erp
->er_extcount
== XFS_LINEAR_EXTS
) {
4754 if (erp_idx
< nlists
)
4755 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4759 ep
= &erp
->er_extbuf
[erp
->er_extcount
];
4761 ep_next
= erp_next
->er_extbuf
;
4766 * This is called to update the er_extoff field in the indirection
4767 * array when extents have been added or removed from one of the
4768 * extent lists. erp_idx contains the irec index to begin updating
4769 * at and ext_diff contains the number of extents that were added
4773 xfs_iext_irec_update_extoffs(
4774 xfs_ifork_t
*ifp
, /* inode fork pointer */
4775 int erp_idx
, /* irec index to update */
4776 int ext_diff
) /* number of new extents */
4778 int i
; /* loop counter */
4779 int nlists
; /* number of irec's (ex lists */
4781 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4782 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4783 for (i
= erp_idx
; i
< nlists
; i
++) {
4784 ifp
->if_u1
.if_ext_irec
[i
].er_extoff
+= ext_diff
;