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"
31 #include "xfs_dmapi.h"
32 #include "xfs_mount.h"
33 #include "xfs_bmap_btree.h"
34 #include "xfs_alloc_btree.h"
35 #include "xfs_ialloc_btree.h"
36 #include "xfs_dir_sf.h"
37 #include "xfs_dir2_sf.h"
38 #include "xfs_attr_sf.h"
39 #include "xfs_dinode.h"
40 #include "xfs_inode.h"
41 #include "xfs_buf_item.h"
42 #include "xfs_inode_item.h"
43 #include "xfs_btree.h"
44 #include "xfs_alloc.h"
45 #include "xfs_ialloc.h"
48 #include "xfs_error.h"
49 #include "xfs_utils.h"
50 #include "xfs_dir2_trace.h"
51 #include "xfs_quota.h"
56 kmem_zone_t
*xfs_ifork_zone
;
57 kmem_zone_t
*xfs_inode_zone
;
58 kmem_zone_t
*xfs_chashlist_zone
;
61 * Used in xfs_itruncate(). This is the maximum number of extents
62 * freed from a file in a single transaction.
64 #define XFS_ITRUNC_MAX_EXTENTS 2
66 STATIC
int xfs_iflush_int(xfs_inode_t
*, xfs_buf_t
*);
67 STATIC
int xfs_iformat_local(xfs_inode_t
*, xfs_dinode_t
*, int, int);
68 STATIC
int xfs_iformat_extents(xfs_inode_t
*, xfs_dinode_t
*, int);
69 STATIC
int xfs_iformat_btree(xfs_inode_t
*, xfs_dinode_t
*, int);
74 * Make sure that the extents in the given memory buffer
89 for (i
= 0; i
< nrecs
; i
++) {
90 ep
= xfs_iext_get_ext(ifp
, i
);
91 rec
.l0
= get_unaligned((__uint64_t
*)&ep
->l0
);
92 rec
.l1
= get_unaligned((__uint64_t
*)&ep
->l1
);
94 xfs_bmbt_disk_get_all(&rec
, &irec
);
96 xfs_bmbt_get_all(&rec
, &irec
);
97 if (fmt
== XFS_EXTFMT_NOSTATE
)
98 ASSERT(irec
.br_state
== XFS_EXT_NORM
);
102 #define xfs_validate_extents(ifp, nrecs, disk, fmt)
106 * Check that none of the inode's in the buffer have a next
107 * unlinked field of 0.
119 j
= mp
->m_inode_cluster_size
>> mp
->m_sb
.sb_inodelog
;
121 for (i
= 0; i
< j
; i
++) {
122 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
,
123 i
* mp
->m_sb
.sb_inodesize
);
124 if (!dip
->di_next_unlinked
) {
125 xfs_fs_cmn_err(CE_ALERT
, mp
,
126 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
128 ASSERT(dip
->di_next_unlinked
);
135 * This routine is called to map an inode number within a file
136 * system to the buffer containing the on-disk version of the
137 * inode. It returns a pointer to the buffer containing the
138 * on-disk inode in the bpp parameter, and in the dip parameter
139 * it returns a pointer to the on-disk inode within that buffer.
141 * If a non-zero error is returned, then the contents of bpp and
142 * dipp are undefined.
144 * Use xfs_imap() to determine the size and location of the
145 * buffer to read from disk.
163 * Call the space management code to find the location of the
167 error
= xfs_imap(mp
, tp
, ino
, &imap
, XFS_IMAP_LOOKUP
);
170 "xfs_inotobp: xfs_imap() returned an "
171 "error %d on %s. Returning error.", error
, mp
->m_fsname
);
176 * If the inode number maps to a block outside the bounds of the
177 * file system then return NULL rather than calling read_buf
178 * and panicing when we get an error from the driver.
180 if ((imap
.im_blkno
+ imap
.im_len
) >
181 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
)) {
183 "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds "
184 "of the file system %s. Returning EINVAL.",
185 (unsigned long long)imap
.im_blkno
,
186 imap
.im_len
, mp
->m_fsname
);
187 return XFS_ERROR(EINVAL
);
191 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
192 * default to just a read_buf() call.
194 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, imap
.im_blkno
,
195 (int)imap
.im_len
, XFS_BUF_LOCK
, &bp
);
199 "xfs_inotobp: xfs_trans_read_buf() returned an "
200 "error %d on %s. Returning error.", error
, mp
->m_fsname
);
203 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, 0);
205 INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
) == XFS_DINODE_MAGIC
&&
206 XFS_DINODE_GOOD_VERSION(INT_GET(dip
->di_core
.di_version
, ARCH_CONVERT
));
207 if (unlikely(XFS_TEST_ERROR(!di_ok
, mp
, XFS_ERRTAG_ITOBP_INOTOBP
,
208 XFS_RANDOM_ITOBP_INOTOBP
))) {
209 XFS_CORRUPTION_ERROR("xfs_inotobp", XFS_ERRLEVEL_LOW
, mp
, dip
);
210 xfs_trans_brelse(tp
, bp
);
212 "xfs_inotobp: XFS_TEST_ERROR() returned an "
213 "error on %s. Returning EFSCORRUPTED.", mp
->m_fsname
);
214 return XFS_ERROR(EFSCORRUPTED
);
217 xfs_inobp_check(mp
, bp
);
220 * Set *dipp to point to the on-disk inode in the buffer.
222 *dipp
= (xfs_dinode_t
*)xfs_buf_offset(bp
, imap
.im_boffset
);
224 *offset
= imap
.im_boffset
;
230 * This routine is called to map an inode to the buffer containing
231 * the on-disk version of the inode. It returns a pointer to the
232 * buffer containing the on-disk inode in the bpp parameter, and in
233 * the dip parameter it returns a pointer to the on-disk inode within
236 * If a non-zero error is returned, then the contents of bpp and
237 * dipp are undefined.
239 * If the inode is new and has not yet been initialized, use xfs_imap()
240 * to determine the size and location of the buffer to read from disk.
241 * If the inode has already been mapped to its buffer and read in once,
242 * then use the mapping information stored in the inode rather than
243 * calling xfs_imap(). This allows us to avoid the overhead of looking
244 * at the inode btree for small block file systems (see xfs_dilocate()).
245 * We can tell whether the inode has been mapped in before by comparing
246 * its disk block address to 0. Only uninitialized inodes will have
247 * 0 for the disk block address.
267 if (ip
->i_blkno
== (xfs_daddr_t
)0) {
269 * Call the space management code to find the location of the
273 if ((error
= xfs_imap(mp
, tp
, ip
->i_ino
, &imap
,
274 XFS_IMAP_LOOKUP
| imap_flags
)))
278 * If the inode number maps to a block outside the bounds
279 * of the file system then return NULL rather than calling
280 * read_buf and panicing when we get an error from the
283 if ((imap
.im_blkno
+ imap
.im_len
) >
284 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
)) {
286 xfs_fs_cmn_err(CE_ALERT
, mp
, "xfs_itobp: "
287 "(imap.im_blkno (0x%llx) "
288 "+ imap.im_len (0x%llx)) > "
289 " XFS_FSB_TO_BB(mp, "
290 "mp->m_sb.sb_dblocks) (0x%llx)",
291 (unsigned long long) imap
.im_blkno
,
292 (unsigned long long) imap
.im_len
,
293 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
));
295 return XFS_ERROR(EINVAL
);
299 * Fill in the fields in the inode that will be used to
300 * map the inode to its buffer from now on.
302 ip
->i_blkno
= imap
.im_blkno
;
303 ip
->i_len
= imap
.im_len
;
304 ip
->i_boffset
= imap
.im_boffset
;
307 * We've already mapped the inode once, so just use the
308 * mapping that we saved the first time.
310 imap
.im_blkno
= ip
->i_blkno
;
311 imap
.im_len
= ip
->i_len
;
312 imap
.im_boffset
= ip
->i_boffset
;
314 ASSERT(bno
== 0 || bno
== imap
.im_blkno
);
317 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
318 * default to just a read_buf() call.
320 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, imap
.im_blkno
,
321 (int)imap
.im_len
, XFS_BUF_LOCK
, &bp
);
325 xfs_fs_cmn_err(CE_ALERT
, mp
, "xfs_itobp: "
326 "xfs_trans_read_buf() returned error %d, "
327 "imap.im_blkno 0x%llx, imap.im_len 0x%llx",
328 error
, (unsigned long long) imap
.im_blkno
,
329 (unsigned long long) imap
.im_len
);
335 * Validate the magic number and version of every inode in the buffer
336 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
339 ni
= (imap_flags
& XFS_IMAP_BULKSTAT
) ? 0 :
340 (BBTOB(imap
.im_len
) >> mp
->m_sb
.sb_inodelog
);
342 ni
= (imap_flags
& XFS_IMAP_BULKSTAT
) ? 0 : 1;
344 for (i
= 0; i
< ni
; i
++) {
348 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
,
349 (i
<< mp
->m_sb
.sb_inodelog
));
350 di_ok
= INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
) == XFS_DINODE_MAGIC
&&
351 XFS_DINODE_GOOD_VERSION(INT_GET(dip
->di_core
.di_version
, ARCH_CONVERT
));
352 if (unlikely(XFS_TEST_ERROR(!di_ok
, mp
, XFS_ERRTAG_ITOBP_INOTOBP
,
353 XFS_RANDOM_ITOBP_INOTOBP
))) {
355 cmn_err(CE_ALERT
, "Device %s - bad inode magic/vsn "
356 "daddr %lld #%d (magic=%x)",
357 XFS_BUFTARG_NAME(mp
->m_ddev_targp
),
358 (unsigned long long)imap
.im_blkno
, i
,
359 INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
));
361 XFS_CORRUPTION_ERROR("xfs_itobp", XFS_ERRLEVEL_HIGH
,
363 xfs_trans_brelse(tp
, bp
);
364 return XFS_ERROR(EFSCORRUPTED
);
367 #endif /* __KERNEL__ */
369 xfs_inobp_check(mp
, bp
);
372 * Mark the buffer as an inode buffer now that it looks good
374 XFS_BUF_SET_VTYPE(bp
, B_FS_INO
);
377 * Set *dipp to point to the on-disk inode in the buffer.
379 *dipp
= (xfs_dinode_t
*)xfs_buf_offset(bp
, imap
.im_boffset
);
385 * Move inode type and inode format specific information from the
386 * on-disk inode to the in-core inode. For fifos, devs, and sockets
387 * this means set if_rdev to the proper value. For files, directories,
388 * and symlinks this means to bring in the in-line data or extent
389 * pointers. For a file in B-tree format, only the root is immediately
390 * brought in-core. The rest will be in-lined in if_extents when it
391 * is first referenced (see xfs_iread_extents()).
398 xfs_attr_shortform_t
*atp
;
402 ip
->i_df
.if_ext_max
=
403 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
407 INT_GET(dip
->di_core
.di_nextents
, ARCH_CONVERT
) +
408 INT_GET(dip
->di_core
.di_anextents
, ARCH_CONVERT
) >
409 INT_GET(dip
->di_core
.di_nblocks
, ARCH_CONVERT
))) {
410 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
411 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
412 (unsigned long long)ip
->i_ino
,
413 (int)(INT_GET(dip
->di_core
.di_nextents
, ARCH_CONVERT
)
414 + INT_GET(dip
->di_core
.di_anextents
, ARCH_CONVERT
)),
416 INT_GET(dip
->di_core
.di_nblocks
, ARCH_CONVERT
));
417 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW
,
419 return XFS_ERROR(EFSCORRUPTED
);
422 if (unlikely(INT_GET(dip
->di_core
.di_forkoff
, ARCH_CONVERT
) > ip
->i_mount
->m_sb
.sb_inodesize
)) {
423 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
424 "corrupt dinode %Lu, forkoff = 0x%x.",
425 (unsigned long long)ip
->i_ino
,
426 (int)(INT_GET(dip
->di_core
.di_forkoff
, ARCH_CONVERT
)));
427 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW
,
429 return XFS_ERROR(EFSCORRUPTED
);
432 switch (ip
->i_d
.di_mode
& S_IFMT
) {
437 if (unlikely(INT_GET(dip
->di_core
.di_format
, ARCH_CONVERT
) != XFS_DINODE_FMT_DEV
)) {
438 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW
,
440 return XFS_ERROR(EFSCORRUPTED
);
443 ip
->i_df
.if_u2
.if_rdev
= INT_GET(dip
->di_u
.di_dev
, ARCH_CONVERT
);
449 switch (INT_GET(dip
->di_core
.di_format
, ARCH_CONVERT
)) {
450 case XFS_DINODE_FMT_LOCAL
:
452 * no local regular files yet
454 if (unlikely((INT_GET(dip
->di_core
.di_mode
, ARCH_CONVERT
) & S_IFMT
) == S_IFREG
)) {
455 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
457 "(local format for regular file).",
458 (unsigned long long) ip
->i_ino
);
459 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
462 return XFS_ERROR(EFSCORRUPTED
);
465 di_size
= INT_GET(dip
->di_core
.di_size
, ARCH_CONVERT
);
466 if (unlikely(di_size
> XFS_DFORK_DSIZE(dip
, ip
->i_mount
))) {
467 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
469 "(bad size %Ld for local inode).",
470 (unsigned long long) ip
->i_ino
,
471 (long long) di_size
);
472 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
475 return XFS_ERROR(EFSCORRUPTED
);
479 error
= xfs_iformat_local(ip
, dip
, XFS_DATA_FORK
, size
);
481 case XFS_DINODE_FMT_EXTENTS
:
482 error
= xfs_iformat_extents(ip
, dip
, XFS_DATA_FORK
);
484 case XFS_DINODE_FMT_BTREE
:
485 error
= xfs_iformat_btree(ip
, dip
, XFS_DATA_FORK
);
488 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW
,
490 return XFS_ERROR(EFSCORRUPTED
);
495 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW
, ip
->i_mount
);
496 return XFS_ERROR(EFSCORRUPTED
);
501 if (!XFS_DFORK_Q(dip
))
503 ASSERT(ip
->i_afp
== NULL
);
504 ip
->i_afp
= kmem_zone_zalloc(xfs_ifork_zone
, KM_SLEEP
);
505 ip
->i_afp
->if_ext_max
=
506 XFS_IFORK_ASIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
507 switch (INT_GET(dip
->di_core
.di_aformat
, ARCH_CONVERT
)) {
508 case XFS_DINODE_FMT_LOCAL
:
509 atp
= (xfs_attr_shortform_t
*)XFS_DFORK_APTR(dip
);
510 size
= be16_to_cpu(atp
->hdr
.totsize
);
511 error
= xfs_iformat_local(ip
, dip
, XFS_ATTR_FORK
, size
);
513 case XFS_DINODE_FMT_EXTENTS
:
514 error
= xfs_iformat_extents(ip
, dip
, XFS_ATTR_FORK
);
516 case XFS_DINODE_FMT_BTREE
:
517 error
= xfs_iformat_btree(ip
, dip
, XFS_ATTR_FORK
);
520 error
= XFS_ERROR(EFSCORRUPTED
);
524 kmem_zone_free(xfs_ifork_zone
, ip
->i_afp
);
526 xfs_idestroy_fork(ip
, XFS_DATA_FORK
);
532 * The file is in-lined in the on-disk inode.
533 * If it fits into if_inline_data, then copy
534 * it there, otherwise allocate a buffer for it
535 * and copy the data there. Either way, set
536 * if_data to point at the data.
537 * If we allocate a buffer for the data, make
538 * sure that its size is a multiple of 4 and
539 * record the real size in i_real_bytes.
552 * If the size is unreasonable, then something
553 * is wrong and we just bail out rather than crash in
554 * kmem_alloc() or memcpy() below.
556 if (unlikely(size
> XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
))) {
557 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
559 "(bad size %d for local fork, size = %d).",
560 (unsigned long long) ip
->i_ino
, size
,
561 XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
));
562 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW
,
564 return XFS_ERROR(EFSCORRUPTED
);
566 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
569 ifp
->if_u1
.if_data
= NULL
;
570 else if (size
<= sizeof(ifp
->if_u2
.if_inline_data
))
571 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
573 real_size
= roundup(size
, 4);
574 ifp
->if_u1
.if_data
= kmem_alloc(real_size
, KM_SLEEP
);
576 ifp
->if_bytes
= size
;
577 ifp
->if_real_bytes
= real_size
;
579 memcpy(ifp
->if_u1
.if_data
, XFS_DFORK_PTR(dip
, whichfork
), size
);
580 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
581 ifp
->if_flags
|= XFS_IFINLINE
;
586 * The file consists of a set of extents all
587 * of which fit into the on-disk inode.
588 * If there are few enough extents to fit into
589 * the if_inline_ext, then copy them there.
590 * Otherwise allocate a buffer for them and copy
591 * them into it. Either way, set if_extents
592 * to point at the extents.
600 xfs_bmbt_rec_t
*ep
, *dp
;
606 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
607 nex
= XFS_DFORK_NEXTENTS(dip
, whichfork
);
608 size
= nex
* (uint
)sizeof(xfs_bmbt_rec_t
);
611 * If the number of extents is unreasonable, then something
612 * is wrong and we just bail out rather than crash in
613 * kmem_alloc() or memcpy() below.
615 if (unlikely(size
< 0 || size
> XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
))) {
616 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
617 "corrupt inode %Lu ((a)extents = %d).",
618 (unsigned long long) ip
->i_ino
, nex
);
619 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW
,
621 return XFS_ERROR(EFSCORRUPTED
);
624 ifp
->if_real_bytes
= 0;
626 ifp
->if_u1
.if_extents
= NULL
;
627 else if (nex
<= XFS_INLINE_EXTS
)
628 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
630 xfs_iext_add(ifp
, 0, nex
);
632 ifp
->if_bytes
= size
;
634 dp
= (xfs_bmbt_rec_t
*) XFS_DFORK_PTR(dip
, whichfork
);
635 xfs_validate_extents(ifp
, nex
, 1, XFS_EXTFMT_INODE(ip
));
636 for (i
= 0; i
< nex
; i
++, dp
++) {
637 ep
= xfs_iext_get_ext(ifp
, i
);
638 ep
->l0
= INT_GET(get_unaligned((__uint64_t
*)&dp
->l0
),
640 ep
->l1
= INT_GET(get_unaligned((__uint64_t
*)&dp
->l1
),
643 xfs_bmap_trace_exlist("xfs_iformat_extents", ip
, nex
,
645 if (whichfork
!= XFS_DATA_FORK
||
646 XFS_EXTFMT_INODE(ip
) == XFS_EXTFMT_NOSTATE
)
647 if (unlikely(xfs_check_nostate_extents(
649 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
652 return XFS_ERROR(EFSCORRUPTED
);
655 ifp
->if_flags
|= XFS_IFEXTENTS
;
660 * The file has too many extents to fit into
661 * the inode, so they are in B-tree format.
662 * Allocate a buffer for the root of the B-tree
663 * and copy the root into it. The i_extents
664 * field will remain NULL until all of the
665 * extents are read in (when they are needed).
673 xfs_bmdr_block_t
*dfp
;
679 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
680 dfp
= (xfs_bmdr_block_t
*)XFS_DFORK_PTR(dip
, whichfork
);
681 size
= XFS_BMAP_BROOT_SPACE(dfp
);
682 nrecs
= XFS_BMAP_BROOT_NUMRECS(dfp
);
685 * blow out if -- fork has less extents than can fit in
686 * fork (fork shouldn't be a btree format), root btree
687 * block has more records than can fit into the fork,
688 * or the number of extents is greater than the number of
691 if (unlikely(XFS_IFORK_NEXTENTS(ip
, whichfork
) <= ifp
->if_ext_max
692 || XFS_BMDR_SPACE_CALC(nrecs
) >
693 XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
)
694 || XFS_IFORK_NEXTENTS(ip
, whichfork
) > ip
->i_d
.di_nblocks
)) {
695 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
696 "corrupt inode %Lu (btree).",
697 (unsigned long long) ip
->i_ino
);
698 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW
,
700 return XFS_ERROR(EFSCORRUPTED
);
703 ifp
->if_broot_bytes
= size
;
704 ifp
->if_broot
= kmem_alloc(size
, KM_SLEEP
);
705 ASSERT(ifp
->if_broot
!= NULL
);
707 * Copy and convert from the on-disk structure
708 * to the in-memory structure.
710 xfs_bmdr_to_bmbt(dfp
, XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
),
711 ifp
->if_broot
, size
);
712 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
713 ifp
->if_flags
|= XFS_IFBROOT
;
719 * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk
722 * buf = on-disk representation
723 * dip = native representation
724 * dir = direction - +ve -> disk to native
725 * -ve -> native to disk
728 xfs_xlate_dinode_core(
730 xfs_dinode_core_t
*dip
,
733 xfs_dinode_core_t
*buf_core
= (xfs_dinode_core_t
*)buf
;
734 xfs_dinode_core_t
*mem_core
= (xfs_dinode_core_t
*)dip
;
735 xfs_arch_t arch
= ARCH_CONVERT
;
739 INT_XLATE(buf_core
->di_magic
, mem_core
->di_magic
, dir
, arch
);
740 INT_XLATE(buf_core
->di_mode
, mem_core
->di_mode
, dir
, arch
);
741 INT_XLATE(buf_core
->di_version
, mem_core
->di_version
, dir
, arch
);
742 INT_XLATE(buf_core
->di_format
, mem_core
->di_format
, dir
, arch
);
743 INT_XLATE(buf_core
->di_onlink
, mem_core
->di_onlink
, dir
, arch
);
744 INT_XLATE(buf_core
->di_uid
, mem_core
->di_uid
, dir
, arch
);
745 INT_XLATE(buf_core
->di_gid
, mem_core
->di_gid
, dir
, arch
);
746 INT_XLATE(buf_core
->di_nlink
, mem_core
->di_nlink
, dir
, arch
);
747 INT_XLATE(buf_core
->di_projid
, mem_core
->di_projid
, dir
, arch
);
750 memcpy(mem_core
->di_pad
, buf_core
->di_pad
,
751 sizeof(buf_core
->di_pad
));
753 memcpy(buf_core
->di_pad
, mem_core
->di_pad
,
754 sizeof(buf_core
->di_pad
));
757 INT_XLATE(buf_core
->di_flushiter
, mem_core
->di_flushiter
, dir
, arch
);
759 INT_XLATE(buf_core
->di_atime
.t_sec
, mem_core
->di_atime
.t_sec
,
761 INT_XLATE(buf_core
->di_atime
.t_nsec
, mem_core
->di_atime
.t_nsec
,
763 INT_XLATE(buf_core
->di_mtime
.t_sec
, mem_core
->di_mtime
.t_sec
,
765 INT_XLATE(buf_core
->di_mtime
.t_nsec
, mem_core
->di_mtime
.t_nsec
,
767 INT_XLATE(buf_core
->di_ctime
.t_sec
, mem_core
->di_ctime
.t_sec
,
769 INT_XLATE(buf_core
->di_ctime
.t_nsec
, mem_core
->di_ctime
.t_nsec
,
771 INT_XLATE(buf_core
->di_size
, mem_core
->di_size
, dir
, arch
);
772 INT_XLATE(buf_core
->di_nblocks
, mem_core
->di_nblocks
, dir
, arch
);
773 INT_XLATE(buf_core
->di_extsize
, mem_core
->di_extsize
, dir
, arch
);
774 INT_XLATE(buf_core
->di_nextents
, mem_core
->di_nextents
, dir
, arch
);
775 INT_XLATE(buf_core
->di_anextents
, mem_core
->di_anextents
, dir
, arch
);
776 INT_XLATE(buf_core
->di_forkoff
, mem_core
->di_forkoff
, dir
, arch
);
777 INT_XLATE(buf_core
->di_aformat
, mem_core
->di_aformat
, dir
, arch
);
778 INT_XLATE(buf_core
->di_dmevmask
, mem_core
->di_dmevmask
, dir
, arch
);
779 INT_XLATE(buf_core
->di_dmstate
, mem_core
->di_dmstate
, dir
, arch
);
780 INT_XLATE(buf_core
->di_flags
, mem_core
->di_flags
, dir
, arch
);
781 INT_XLATE(buf_core
->di_gen
, mem_core
->di_gen
, dir
, arch
);
786 xfs_dinode_core_t
*dic
,
791 if (di_flags
& XFS_DIFLAG_ANY
) {
792 if (di_flags
& XFS_DIFLAG_REALTIME
)
793 flags
|= XFS_XFLAG_REALTIME
;
794 if (di_flags
& XFS_DIFLAG_PREALLOC
)
795 flags
|= XFS_XFLAG_PREALLOC
;
796 if (di_flags
& XFS_DIFLAG_IMMUTABLE
)
797 flags
|= XFS_XFLAG_IMMUTABLE
;
798 if (di_flags
& XFS_DIFLAG_APPEND
)
799 flags
|= XFS_XFLAG_APPEND
;
800 if (di_flags
& XFS_DIFLAG_SYNC
)
801 flags
|= XFS_XFLAG_SYNC
;
802 if (di_flags
& XFS_DIFLAG_NOATIME
)
803 flags
|= XFS_XFLAG_NOATIME
;
804 if (di_flags
& XFS_DIFLAG_NODUMP
)
805 flags
|= XFS_XFLAG_NODUMP
;
806 if (di_flags
& XFS_DIFLAG_RTINHERIT
)
807 flags
|= XFS_XFLAG_RTINHERIT
;
808 if (di_flags
& XFS_DIFLAG_PROJINHERIT
)
809 flags
|= XFS_XFLAG_PROJINHERIT
;
810 if (di_flags
& XFS_DIFLAG_NOSYMLINKS
)
811 flags
|= XFS_XFLAG_NOSYMLINKS
;
812 if (di_flags
& XFS_DIFLAG_EXTSIZE
)
813 flags
|= XFS_XFLAG_EXTSIZE
;
814 if (di_flags
& XFS_DIFLAG_EXTSZINHERIT
)
815 flags
|= XFS_XFLAG_EXTSZINHERIT
;
816 if (di_flags
& XFS_DIFLAG_NODEFRAG
)
817 flags
|= XFS_XFLAG_NODEFRAG
;
827 xfs_dinode_core_t
*dic
= &ip
->i_d
;
829 return _xfs_dic2xflags(dic
, dic
->di_flags
) |
830 (XFS_CFORK_Q(dic
) ? XFS_XFLAG_HASATTR
: 0);
835 xfs_dinode_core_t
*dic
)
837 return _xfs_dic2xflags(dic
, INT_GET(dic
->di_flags
, ARCH_CONVERT
)) |
838 (XFS_CFORK_Q_DISK(dic
) ? XFS_XFLAG_HASATTR
: 0);
842 * Given a mount structure and an inode number, return a pointer
843 * to a newly allocated in-core inode corresponding to the given
846 * Initialize the inode's attributes and extent pointers if it
847 * already has them (it will not if the inode has no links).
862 ASSERT(xfs_inode_zone
!= NULL
);
864 ip
= kmem_zone_zalloc(xfs_inode_zone
, KM_SLEEP
);
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
, 0);
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 (INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
) != 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 INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
),
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_xlate_dinode_core((xfs_caddr_t
)&dip
->di_core
,
928 error
= xfs_iformat(ip
, dip
);
930 kmem_zone_free(xfs_inode_zone
, ip
);
931 xfs_trans_brelse(tp
, bp
);
933 xfs_fs_cmn_err(CE_ALERT
, mp
, "xfs_iread: "
934 "xfs_iformat() returned error %d",
940 ip
->i_d
.di_magic
= INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
);
941 ip
->i_d
.di_version
= INT_GET(dip
->di_core
.di_version
, ARCH_CONVERT
);
942 ip
->i_d
.di_gen
= INT_GET(dip
->di_core
.di_gen
, ARCH_CONVERT
);
943 ip
->i_d
.di_flushiter
= INT_GET(dip
->di_core
.di_flushiter
, ARCH_CONVERT
);
945 * Make sure to pull in the mode here as well in
946 * case the inode is released without being used.
947 * This ensures that xfs_inactive() will see that
948 * the inode is already free and not try to mess
949 * with the uninitialized part of it.
953 * Initialize the per-fork minima and maxima for a new
954 * inode here. xfs_iformat will do it for old inodes.
956 ip
->i_df
.if_ext_max
=
957 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
960 INIT_LIST_HEAD(&ip
->i_reclaim
);
963 * The inode format changed when we moved the link count and
964 * made it 32 bits long. If this is an old format inode,
965 * convert it in memory to look like a new one. If it gets
966 * flushed to disk we will convert back before flushing or
967 * logging it. We zero out the new projid field and the old link
968 * count field. We'll handle clearing the pad field (the remains
969 * of the old uuid field) when we actually convert the inode to
970 * the new format. We don't change the version number so that we
971 * can distinguish this from a real new format inode.
973 if (ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
) {
974 ip
->i_d
.di_nlink
= ip
->i_d
.di_onlink
;
975 ip
->i_d
.di_onlink
= 0;
976 ip
->i_d
.di_projid
= 0;
979 ip
->i_delayed_blks
= 0;
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
, 0, 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.
1083 xfs_buf_t
**ialloc_context
,
1084 boolean_t
*call_again
,
1094 * Call the space management code to pick
1095 * the on-disk inode to be allocated.
1097 error
= xfs_dialloc(tp
, pip
->i_ino
, mode
, okalloc
,
1098 ialloc_context
, call_again
, &ino
);
1102 if (*call_again
|| ino
== NULLFSINO
) {
1106 ASSERT(*ialloc_context
== NULL
);
1109 * Get the in-core inode with the lock held exclusively.
1110 * This is because we're setting fields here we need
1111 * to prevent others from looking at until we're done.
1113 error
= xfs_trans_iget(tp
->t_mountp
, tp
, ino
,
1114 IGET_CREATE
, XFS_ILOCK_EXCL
, &ip
);
1121 ip
->i_d
.di_mode
= (__uint16_t
)mode
;
1122 ip
->i_d
.di_onlink
= 0;
1123 ip
->i_d
.di_nlink
= nlink
;
1124 ASSERT(ip
->i_d
.di_nlink
== nlink
);
1125 ip
->i_d
.di_uid
= current_fsuid(cr
);
1126 ip
->i_d
.di_gid
= current_fsgid(cr
);
1127 ip
->i_d
.di_projid
= prid
;
1128 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
1131 * If the superblock version is up to where we support new format
1132 * inodes and this is currently an old format inode, then change
1133 * the inode version number now. This way we only do the conversion
1134 * here rather than here and in the flush/logging code.
1136 if (XFS_SB_VERSION_HASNLINK(&tp
->t_mountp
->m_sb
) &&
1137 ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
) {
1138 ip
->i_d
.di_version
= XFS_DINODE_VERSION_2
;
1140 * We've already zeroed the old link count, the projid field,
1141 * and the pad field.
1146 * Project ids won't be stored on disk if we are using a version 1 inode.
1148 if ( (prid
!= 0) && (ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
))
1149 xfs_bump_ino_vers2(tp
, ip
);
1151 if (XFS_INHERIT_GID(pip
, vp
->v_vfsp
)) {
1152 ip
->i_d
.di_gid
= pip
->i_d
.di_gid
;
1153 if ((pip
->i_d
.di_mode
& S_ISGID
) && (mode
& S_IFMT
) == S_IFDIR
) {
1154 ip
->i_d
.di_mode
|= S_ISGID
;
1159 * If the group ID of the new file does not match the effective group
1160 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1161 * (and only if the irix_sgid_inherit compatibility variable is set).
1163 if ((irix_sgid_inherit
) &&
1164 (ip
->i_d
.di_mode
& S_ISGID
) &&
1165 (!in_group_p((gid_t
)ip
->i_d
.di_gid
))) {
1166 ip
->i_d
.di_mode
&= ~S_ISGID
;
1169 ip
->i_d
.di_size
= 0;
1170 ip
->i_d
.di_nextents
= 0;
1171 ASSERT(ip
->i_d
.di_nblocks
== 0);
1172 xfs_ichgtime(ip
, XFS_ICHGTIME_CHG
|XFS_ICHGTIME_ACC
|XFS_ICHGTIME_MOD
);
1174 * di_gen will have been taken care of in xfs_iread.
1176 ip
->i_d
.di_extsize
= 0;
1177 ip
->i_d
.di_dmevmask
= 0;
1178 ip
->i_d
.di_dmstate
= 0;
1179 ip
->i_d
.di_flags
= 0;
1180 flags
= XFS_ILOG_CORE
;
1181 switch (mode
& S_IFMT
) {
1186 ip
->i_d
.di_format
= XFS_DINODE_FMT_DEV
;
1187 ip
->i_df
.if_u2
.if_rdev
= rdev
;
1188 ip
->i_df
.if_flags
= 0;
1189 flags
|= XFS_ILOG_DEV
;
1193 if (unlikely(pip
->i_d
.di_flags
& XFS_DIFLAG_ANY
)) {
1196 if ((mode
& S_IFMT
) == S_IFDIR
) {
1197 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
1198 di_flags
|= XFS_DIFLAG_RTINHERIT
;
1199 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
1200 di_flags
|= XFS_DIFLAG_EXTSZINHERIT
;
1201 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
1203 } else if ((mode
& S_IFMT
) == S_IFREG
) {
1204 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
) {
1205 di_flags
|= XFS_DIFLAG_REALTIME
;
1206 ip
->i_iocore
.io_flags
|= XFS_IOCORE_RT
;
1208 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
1209 di_flags
|= XFS_DIFLAG_EXTSIZE
;
1210 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
1213 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOATIME
) &&
1214 xfs_inherit_noatime
)
1215 di_flags
|= XFS_DIFLAG_NOATIME
;
1216 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODUMP
) &&
1218 di_flags
|= XFS_DIFLAG_NODUMP
;
1219 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_SYNC
) &&
1221 di_flags
|= XFS_DIFLAG_SYNC
;
1222 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOSYMLINKS
) &&
1223 xfs_inherit_nosymlinks
)
1224 di_flags
|= XFS_DIFLAG_NOSYMLINKS
;
1225 if (pip
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
1226 di_flags
|= XFS_DIFLAG_PROJINHERIT
;
1227 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODEFRAG
) &&
1228 xfs_inherit_nodefrag
)
1229 di_flags
|= XFS_DIFLAG_NODEFRAG
;
1230 ip
->i_d
.di_flags
|= di_flags
;
1234 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
1235 ip
->i_df
.if_flags
= XFS_IFEXTENTS
;
1236 ip
->i_df
.if_bytes
= ip
->i_df
.if_real_bytes
= 0;
1237 ip
->i_df
.if_u1
.if_extents
= NULL
;
1243 * Attribute fork settings for new inode.
1245 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
1246 ip
->i_d
.di_anextents
= 0;
1249 * Log the new values stuffed into the inode.
1251 xfs_trans_log_inode(tp
, ip
, flags
);
1253 /* now that we have an i_mode we can set Linux inode ops (& unlock) */
1254 VFS_INIT_VNODE(XFS_MTOVFS(tp
->t_mountp
), vp
, XFS_ITOBHV(ip
), 1);
1261 * Check to make sure that there are no blocks allocated to the
1262 * file beyond the size of the file. We don't check this for
1263 * files with fixed size extents or real time extents, but we
1264 * at least do it for regular files.
1273 xfs_fileoff_t map_first
;
1275 xfs_bmbt_irec_t imaps
[2];
1277 if ((ip
->i_d
.di_mode
& S_IFMT
) != S_IFREG
)
1280 if (ip
->i_d
.di_flags
& (XFS_DIFLAG_REALTIME
| XFS_DIFLAG_EXTSIZE
))
1284 map_first
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)isize
);
1286 * The filesystem could be shutting down, so bmapi may return
1289 if (xfs_bmapi(NULL
, ip
, map_first
,
1291 (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
)) -
1293 XFS_BMAPI_ENTIRE
, NULL
, 0, imaps
, &nimaps
,
1296 ASSERT(nimaps
== 1);
1297 ASSERT(imaps
[0].br_startblock
== HOLESTARTBLOCK
);
1302 * Calculate the last possible buffered byte in a file. This must
1303 * include data that was buffered beyond the EOF by the write code.
1304 * This also needs to deal with overflowing the xfs_fsize_t type
1305 * which can happen for sizes near the limit.
1307 * We also need to take into account any blocks beyond the EOF. It
1308 * may be the case that they were buffered by a write which failed.
1309 * In that case the pages will still be in memory, but the inode size
1310 * will never have been updated.
1317 xfs_fsize_t last_byte
;
1318 xfs_fileoff_t last_block
;
1319 xfs_fileoff_t size_last_block
;
1322 ASSERT(ismrlocked(&(ip
->i_iolock
), MR_UPDATE
| MR_ACCESS
));
1326 * Only check for blocks beyond the EOF if the extents have
1327 * been read in. This eliminates the need for the inode lock,
1328 * and it also saves us from looking when it really isn't
1331 if (ip
->i_df
.if_flags
& XFS_IFEXTENTS
) {
1332 error
= xfs_bmap_last_offset(NULL
, ip
, &last_block
,
1340 size_last_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)ip
->i_d
.di_size
);
1341 last_block
= XFS_FILEOFF_MAX(last_block
, size_last_block
);
1343 last_byte
= XFS_FSB_TO_B(mp
, last_block
);
1344 if (last_byte
< 0) {
1345 return XFS_MAXIOFFSET(mp
);
1347 last_byte
+= (1 << mp
->m_writeio_log
);
1348 if (last_byte
< 0) {
1349 return XFS_MAXIOFFSET(mp
);
1354 #if defined(XFS_RW_TRACE)
1360 xfs_fsize_t new_size
,
1361 xfs_off_t toss_start
,
1362 xfs_off_t toss_finish
)
1364 if (ip
->i_rwtrace
== NULL
) {
1368 ktrace_enter(ip
->i_rwtrace
,
1371 (void*)(unsigned long)((ip
->i_d
.di_size
>> 32) & 0xffffffff),
1372 (void*)(unsigned long)(ip
->i_d
.di_size
& 0xffffffff),
1373 (void*)((long)flag
),
1374 (void*)(unsigned long)((new_size
>> 32) & 0xffffffff),
1375 (void*)(unsigned long)(new_size
& 0xffffffff),
1376 (void*)(unsigned long)((toss_start
>> 32) & 0xffffffff),
1377 (void*)(unsigned long)(toss_start
& 0xffffffff),
1378 (void*)(unsigned long)((toss_finish
>> 32) & 0xffffffff),
1379 (void*)(unsigned long)(toss_finish
& 0xffffffff),
1380 (void*)(unsigned long)current_cpu(),
1381 (void*)(unsigned long)current_pid(),
1387 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1391 * Start the truncation of the file to new_size. The new size
1392 * must be smaller than the current size. This routine will
1393 * clear the buffer and page caches of file data in the removed
1394 * range, and xfs_itruncate_finish() will remove the underlying
1397 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1398 * must NOT have the inode lock held at all. This is because we're
1399 * calling into the buffer/page cache code and we can't hold the
1400 * inode lock when we do so.
1402 * We need to wait for any direct I/Os in flight to complete before we
1403 * proceed with the truncate. This is needed to prevent the extents
1404 * being read or written by the direct I/Os from being removed while the
1405 * I/O is in flight as there is no other method of synchronising
1406 * direct I/O with the truncate operation. Also, because we hold
1407 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1408 * started until the truncate completes and drops the lock. Essentially,
1409 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1410 * between direct I/Os and the truncate operation.
1412 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1413 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1414 * in the case that the caller is locking things out of order and
1415 * may not be able to call xfs_itruncate_finish() with the inode lock
1416 * held without dropping the I/O lock. If the caller must drop the
1417 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1418 * must be called again with all the same restrictions as the initial
1422 xfs_itruncate_start(
1425 xfs_fsize_t new_size
)
1427 xfs_fsize_t last_byte
;
1428 xfs_off_t toss_start
;
1432 ASSERT(ismrlocked(&ip
->i_iolock
, MR_UPDATE
) != 0);
1433 ASSERT((new_size
== 0) || (new_size
<= ip
->i_d
.di_size
));
1434 ASSERT((flags
== XFS_ITRUNC_DEFINITE
) ||
1435 (flags
== XFS_ITRUNC_MAYBE
));
1440 vn_iowait(vp
); /* wait for the completion of any pending DIOs */
1443 * Call VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES() to get rid of pages and buffers
1444 * overlapping the region being removed. We have to use
1445 * the less efficient VOP_FLUSHINVAL_PAGES() in the case that the
1446 * caller may not be able to finish the truncate without
1447 * dropping the inode's I/O lock. Make sure
1448 * to catch any pages brought in by buffers overlapping
1449 * the EOF by searching out beyond the isize by our
1450 * block size. We round new_size up to a block boundary
1451 * so that we don't toss things on the same block as
1452 * new_size but before it.
1454 * Before calling VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES(), make sure to
1455 * call remapf() over the same region if the file is mapped.
1456 * This frees up mapped file references to the pages in the
1457 * given range and for the VOP_FLUSHINVAL_PAGES() case it ensures
1458 * that we get the latest mapped changes flushed out.
1460 toss_start
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)new_size
);
1461 toss_start
= XFS_FSB_TO_B(mp
, toss_start
);
1462 if (toss_start
< 0) {
1464 * The place to start tossing is beyond our maximum
1465 * file size, so there is no way that the data extended
1470 last_byte
= xfs_file_last_byte(ip
);
1471 xfs_itrunc_trace(XFS_ITRUNC_START
, ip
, flags
, new_size
, toss_start
,
1473 if (last_byte
> toss_start
) {
1474 if (flags
& XFS_ITRUNC_DEFINITE
) {
1475 VOP_TOSS_PAGES(vp
, toss_start
, -1, FI_REMAPF_LOCKED
);
1477 VOP_FLUSHINVAL_PAGES(vp
, toss_start
, -1, FI_REMAPF_LOCKED
);
1482 if (new_size
== 0) {
1483 ASSERT(VN_CACHED(vp
) == 0);
1489 * Shrink the file to the given new_size. The new
1490 * size must be smaller than the current size.
1491 * This will free up the underlying blocks
1492 * in the removed range after a call to xfs_itruncate_start()
1493 * or xfs_atruncate_start().
1495 * The transaction passed to this routine must have made
1496 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1497 * This routine may commit the given transaction and
1498 * start new ones, so make sure everything involved in
1499 * the transaction is tidy before calling here.
1500 * Some transaction will be returned to the caller to be
1501 * committed. The incoming transaction must already include
1502 * the inode, and both inode locks must be held exclusively.
1503 * The inode must also be "held" within the transaction. On
1504 * return the inode will be "held" within the returned transaction.
1505 * This routine does NOT require any disk space to be reserved
1506 * for it within the transaction.
1508 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1509 * and it indicates the fork which is to be truncated. For the
1510 * attribute fork we only support truncation to size 0.
1512 * We use the sync parameter to indicate whether or not the first
1513 * transaction we perform might have to be synchronous. For the attr fork,
1514 * it needs to be so if the unlink of the inode is not yet known to be
1515 * permanent in the log. This keeps us from freeing and reusing the
1516 * blocks of the attribute fork before the unlink of the inode becomes
1519 * For the data fork, we normally have to run synchronously if we're
1520 * being called out of the inactive path or we're being called
1521 * out of the create path where we're truncating an existing file.
1522 * Either way, the truncate needs to be sync so blocks don't reappear
1523 * in the file with altered data in case of a crash. wsync filesystems
1524 * can run the first case async because anything that shrinks the inode
1525 * has to run sync so by the time we're called here from inactive, the
1526 * inode size is permanently set to 0.
1528 * Calls from the truncate path always need to be sync unless we're
1529 * in a wsync filesystem and the file has already been unlinked.
1531 * The caller is responsible for correctly setting the sync parameter.
1532 * It gets too hard for us to guess here which path we're being called
1533 * out of just based on inode state.
1536 xfs_itruncate_finish(
1539 xfs_fsize_t new_size
,
1543 xfs_fsblock_t first_block
;
1544 xfs_fileoff_t first_unmap_block
;
1545 xfs_fileoff_t last_block
;
1546 xfs_filblks_t unmap_len
=0;
1551 xfs_bmap_free_t free_list
;
1554 ASSERT(ismrlocked(&ip
->i_iolock
, MR_UPDATE
) != 0);
1555 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
) != 0);
1556 ASSERT((new_size
== 0) || (new_size
<= ip
->i_d
.di_size
));
1557 ASSERT(*tp
!= NULL
);
1558 ASSERT((*tp
)->t_flags
& XFS_TRANS_PERM_LOG_RES
);
1559 ASSERT(ip
->i_transp
== *tp
);
1560 ASSERT(ip
->i_itemp
!= NULL
);
1561 ASSERT(ip
->i_itemp
->ili_flags
& XFS_ILI_HOLD
);
1565 mp
= (ntp
)->t_mountp
;
1566 ASSERT(! XFS_NOT_DQATTACHED(mp
, ip
));
1569 * We only support truncating the entire attribute fork.
1571 if (fork
== XFS_ATTR_FORK
) {
1574 first_unmap_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)new_size
);
1575 xfs_itrunc_trace(XFS_ITRUNC_FINISH1
, ip
, 0, new_size
, 0, 0);
1577 * The first thing we do is set the size to new_size permanently
1578 * on disk. This way we don't have to worry about anyone ever
1579 * being able to look at the data being freed even in the face
1580 * of a crash. What we're getting around here is the case where
1581 * we free a block, it is allocated to another file, it is written
1582 * to, and then we crash. If the new data gets written to the
1583 * file but the log buffers containing the free and reallocation
1584 * don't, then we'd end up with garbage in the blocks being freed.
1585 * As long as we make the new_size permanent before actually
1586 * freeing any blocks it doesn't matter if they get writtten to.
1588 * The callers must signal into us whether or not the size
1589 * setting here must be synchronous. There are a few cases
1590 * where it doesn't have to be synchronous. Those cases
1591 * occur if the file is unlinked and we know the unlink is
1592 * permanent or if the blocks being truncated are guaranteed
1593 * to be beyond the inode eof (regardless of the link count)
1594 * and the eof value is permanent. Both of these cases occur
1595 * only on wsync-mounted filesystems. In those cases, we're
1596 * guaranteed that no user will ever see the data in the blocks
1597 * that are being truncated so the truncate can run async.
1598 * In the free beyond eof case, the file may wind up with
1599 * more blocks allocated to it than it needs if we crash
1600 * and that won't get fixed until the next time the file
1601 * is re-opened and closed but that's ok as that shouldn't
1602 * be too many blocks.
1604 * However, we can't just make all wsync xactions run async
1605 * because there's one call out of the create path that needs
1606 * to run sync where it's truncating an existing file to size
1607 * 0 whose size is > 0.
1609 * It's probably possible to come up with a test in this
1610 * routine that would correctly distinguish all the above
1611 * cases from the values of the function parameters and the
1612 * inode state but for sanity's sake, I've decided to let the
1613 * layers above just tell us. It's simpler to correctly figure
1614 * out in the layer above exactly under what conditions we
1615 * can run async and I think it's easier for others read and
1616 * follow the logic in case something has to be changed.
1617 * cscope is your friend -- rcc.
1619 * The attribute fork is much simpler.
1621 * For the attribute fork we allow the caller to tell us whether
1622 * the unlink of the inode that led to this call is yet permanent
1623 * in the on disk log. If it is not and we will be freeing extents
1624 * in this inode then we make the first transaction synchronous
1625 * to make sure that the unlink is permanent by the time we free
1628 if (fork
== XFS_DATA_FORK
) {
1629 if (ip
->i_d
.di_nextents
> 0) {
1630 ip
->i_d
.di_size
= new_size
;
1631 xfs_trans_log_inode(ntp
, ip
, XFS_ILOG_CORE
);
1634 ASSERT(!(mp
->m_flags
& XFS_MOUNT_WSYNC
));
1635 if (ip
->i_d
.di_anextents
> 0)
1636 xfs_trans_set_sync(ntp
);
1638 ASSERT(fork
== XFS_DATA_FORK
||
1639 (fork
== XFS_ATTR_FORK
&&
1640 ((sync
&& !(mp
->m_flags
& XFS_MOUNT_WSYNC
)) ||
1641 (sync
== 0 && (mp
->m_flags
& XFS_MOUNT_WSYNC
)))));
1644 * Since it is possible for space to become allocated beyond
1645 * the end of the file (in a crash where the space is allocated
1646 * but the inode size is not yet updated), simply remove any
1647 * blocks which show up between the new EOF and the maximum
1648 * possible file size. If the first block to be removed is
1649 * beyond the maximum file size (ie it is the same as last_block),
1650 * then there is nothing to do.
1652 last_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
));
1653 ASSERT(first_unmap_block
<= last_block
);
1655 if (last_block
== first_unmap_block
) {
1658 unmap_len
= last_block
- first_unmap_block
+ 1;
1662 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1663 * will tell us whether it freed the entire range or
1664 * not. If this is a synchronous mount (wsync),
1665 * then we can tell bunmapi to keep all the
1666 * transactions asynchronous since the unlink
1667 * transaction that made this inode inactive has
1668 * already hit the disk. There's no danger of
1669 * the freed blocks being reused, there being a
1670 * crash, and the reused blocks suddenly reappearing
1671 * in this file with garbage in them once recovery
1674 XFS_BMAP_INIT(&free_list
, &first_block
);
1675 error
= XFS_BUNMAPI(mp
, ntp
, &ip
->i_iocore
,
1676 first_unmap_block
, unmap_len
,
1677 XFS_BMAPI_AFLAG(fork
) |
1678 (sync
? 0 : XFS_BMAPI_ASYNC
),
1679 XFS_ITRUNC_MAX_EXTENTS
,
1680 &first_block
, &free_list
,
1684 * If the bunmapi call encounters an error,
1685 * return to the caller where the transaction
1686 * can be properly aborted. We just need to
1687 * make sure we're not holding any resources
1688 * that we were not when we came in.
1690 xfs_bmap_cancel(&free_list
);
1695 * Duplicate the transaction that has the permanent
1696 * reservation and commit the old transaction.
1698 error
= xfs_bmap_finish(tp
, &free_list
, first_block
,
1703 * If the bmap finish call encounters an error,
1704 * return to the caller where the transaction
1705 * can be properly aborted. We just need to
1706 * make sure we're not holding any resources
1707 * that we were not when we came in.
1709 * Aborting from this point might lose some
1710 * blocks in the file system, but oh well.
1712 xfs_bmap_cancel(&free_list
);
1715 * If the passed in transaction committed
1716 * in xfs_bmap_finish(), then we want to
1717 * add the inode to this one before returning.
1718 * This keeps things simple for the higher
1719 * level code, because it always knows that
1720 * the inode is locked and held in the
1721 * transaction that returns to it whether
1722 * errors occur or not. We don't mark the
1723 * inode dirty so that this transaction can
1724 * be easily aborted if possible.
1726 xfs_trans_ijoin(ntp
, ip
,
1727 XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1728 xfs_trans_ihold(ntp
, ip
);
1735 * The first xact was committed,
1736 * so add the inode to the new one.
1737 * Mark it dirty so it will be logged
1738 * and moved forward in the log as
1739 * part of every commit.
1741 xfs_trans_ijoin(ntp
, ip
,
1742 XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1743 xfs_trans_ihold(ntp
, ip
);
1744 xfs_trans_log_inode(ntp
, ip
, XFS_ILOG_CORE
);
1746 ntp
= xfs_trans_dup(ntp
);
1747 (void) xfs_trans_commit(*tp
, 0, NULL
);
1749 error
= xfs_trans_reserve(ntp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0,
1750 XFS_TRANS_PERM_LOG_RES
,
1751 XFS_ITRUNCATE_LOG_COUNT
);
1753 * Add the inode being truncated to the next chained
1756 xfs_trans_ijoin(ntp
, ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1757 xfs_trans_ihold(ntp
, ip
);
1762 * Only update the size in the case of the data fork, but
1763 * always re-log the inode so that our permanent transaction
1764 * can keep on rolling it forward in the log.
1766 if (fork
== XFS_DATA_FORK
) {
1767 xfs_isize_check(mp
, ip
, new_size
);
1768 ip
->i_d
.di_size
= new_size
;
1770 xfs_trans_log_inode(ntp
, ip
, XFS_ILOG_CORE
);
1771 ASSERT((new_size
!= 0) ||
1772 (fork
== XFS_ATTR_FORK
) ||
1773 (ip
->i_delayed_blks
== 0));
1774 ASSERT((new_size
!= 0) ||
1775 (fork
== XFS_ATTR_FORK
) ||
1776 (ip
->i_d
.di_nextents
== 0));
1777 xfs_itrunc_trace(XFS_ITRUNC_FINISH2
, ip
, 0, new_size
, 0, 0);
1785 * Do the first part of growing a file: zero any data in the last
1786 * block that is beyond the old EOF. We need to do this before
1787 * the inode is joined to the transaction to modify the i_size.
1788 * That way we can drop the inode lock and call into the buffer
1789 * cache to get the buffer mapping the EOF.
1794 xfs_fsize_t new_size
,
1799 ASSERT(ismrlocked(&(ip
->i_lock
), MR_UPDATE
) != 0);
1800 ASSERT(ismrlocked(&(ip
->i_iolock
), MR_UPDATE
) != 0);
1801 ASSERT(new_size
> ip
->i_d
.di_size
);
1804 * Zero any pages that may have been created by
1805 * xfs_write_file() beyond the end of the file
1806 * and any blocks between the old and new file sizes.
1808 error
= xfs_zero_eof(XFS_ITOV(ip
), &ip
->i_iocore
, new_size
,
1809 ip
->i_d
.di_size
, new_size
);
1816 * This routine is called to extend the size of a file.
1817 * The inode must have both the iolock and the ilock locked
1818 * for update and it must be a part of the current transaction.
1819 * The xfs_igrow_start() function must have been called previously.
1820 * If the change_flag is not zero, the inode change timestamp will
1827 xfs_fsize_t new_size
,
1830 ASSERT(ismrlocked(&(ip
->i_lock
), MR_UPDATE
) != 0);
1831 ASSERT(ismrlocked(&(ip
->i_iolock
), MR_UPDATE
) != 0);
1832 ASSERT(ip
->i_transp
== tp
);
1833 ASSERT(new_size
> ip
->i_d
.di_size
);
1836 * Update the file size. Update the inode change timestamp
1837 * if change_flag set.
1839 ip
->i_d
.di_size
= new_size
;
1841 xfs_ichgtime(ip
, XFS_ICHGTIME_CHG
);
1842 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1848 * This is called when the inode's link count goes to 0.
1849 * We place the on-disk inode on a list in the AGI. It
1850 * will be pulled from this list when the inode is freed.
1862 xfs_agnumber_t agno
;
1863 xfs_daddr_t agdaddr
;
1870 ASSERT(ip
->i_d
.di_nlink
== 0);
1871 ASSERT(ip
->i_d
.di_mode
!= 0);
1872 ASSERT(ip
->i_transp
== tp
);
1876 agno
= XFS_INO_TO_AGNO(mp
, ip
->i_ino
);
1877 agdaddr
= XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
));
1880 * Get the agi buffer first. It ensures lock ordering
1883 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, agdaddr
,
1884 XFS_FSS_TO_BB(mp
, 1), 0, &agibp
);
1889 * Validate the magic number of the agi block.
1891 agi
= XFS_BUF_TO_AGI(agibp
);
1893 be32_to_cpu(agi
->agi_magicnum
) == XFS_AGI_MAGIC
&&
1894 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi
->agi_versionnum
));
1895 if (unlikely(XFS_TEST_ERROR(!agi_ok
, mp
, XFS_ERRTAG_IUNLINK
,
1896 XFS_RANDOM_IUNLINK
))) {
1897 XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW
, mp
, agi
);
1898 xfs_trans_brelse(tp
, agibp
);
1899 return XFS_ERROR(EFSCORRUPTED
);
1902 * Get the index into the agi hash table for the
1903 * list this inode will go on.
1905 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
1907 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
1908 ASSERT(agi
->agi_unlinked
[bucket_index
]);
1909 ASSERT(be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != agino
);
1911 if (be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != NULLAGINO
) {
1913 * There is already another inode in the bucket we need
1914 * to add ourselves to. Add us at the front of the list.
1915 * Here we put the head pointer into our next pointer,
1916 * and then we fall through to point the head at us.
1918 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, 0, 0);
1922 ASSERT(INT_GET(dip
->di_next_unlinked
, ARCH_CONVERT
) == NULLAGINO
);
1923 ASSERT(dip
->di_next_unlinked
);
1924 /* both on-disk, don't endian flip twice */
1925 dip
->di_next_unlinked
= agi
->agi_unlinked
[bucket_index
];
1926 offset
= ip
->i_boffset
+
1927 offsetof(xfs_dinode_t
, di_next_unlinked
);
1928 xfs_trans_inode_buf(tp
, ibp
);
1929 xfs_trans_log_buf(tp
, ibp
, offset
,
1930 (offset
+ sizeof(xfs_agino_t
) - 1));
1931 xfs_inobp_check(mp
, ibp
);
1935 * Point the bucket head pointer at the inode being inserted.
1938 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(agino
);
1939 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
1940 (sizeof(xfs_agino_t
) * bucket_index
);
1941 xfs_trans_log_buf(tp
, agibp
, offset
,
1942 (offset
+ sizeof(xfs_agino_t
) - 1));
1947 * Pull the on-disk inode from the AGI unlinked list.
1960 xfs_agnumber_t agno
;
1961 xfs_daddr_t agdaddr
;
1963 xfs_agino_t next_agino
;
1964 xfs_buf_t
*last_ibp
;
1965 xfs_dinode_t
*last_dip
;
1967 int offset
, last_offset
;
1972 * First pull the on-disk inode from the AGI unlinked list.
1976 agno
= XFS_INO_TO_AGNO(mp
, ip
->i_ino
);
1977 agdaddr
= XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
));
1980 * Get the agi buffer first. It ensures lock ordering
1983 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, agdaddr
,
1984 XFS_FSS_TO_BB(mp
, 1), 0, &agibp
);
1987 "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
1988 error
, mp
->m_fsname
);
1992 * Validate the magic number of the agi block.
1994 agi
= XFS_BUF_TO_AGI(agibp
);
1996 be32_to_cpu(agi
->agi_magicnum
) == XFS_AGI_MAGIC
&&
1997 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi
->agi_versionnum
));
1998 if (unlikely(XFS_TEST_ERROR(!agi_ok
, mp
, XFS_ERRTAG_IUNLINK_REMOVE
,
1999 XFS_RANDOM_IUNLINK_REMOVE
))) {
2000 XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW
,
2002 xfs_trans_brelse(tp
, agibp
);
2004 "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
2006 return XFS_ERROR(EFSCORRUPTED
);
2009 * Get the index into the agi hash table for the
2010 * list this inode will go on.
2012 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
2014 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
2015 ASSERT(be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != NULLAGINO
);
2016 ASSERT(agi
->agi_unlinked
[bucket_index
]);
2018 if (be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) == agino
) {
2020 * We're at the head of the list. Get the inode's
2021 * on-disk buffer to see if there is anyone after us
2022 * on the list. Only modify our next pointer if it
2023 * is not already NULLAGINO. This saves us the overhead
2024 * of dealing with the buffer when there is no need to
2027 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, 0, 0);
2030 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2031 error
, mp
->m_fsname
);
2034 next_agino
= INT_GET(dip
->di_next_unlinked
, ARCH_CONVERT
);
2035 ASSERT(next_agino
!= 0);
2036 if (next_agino
!= NULLAGINO
) {
2037 INT_SET(dip
->di_next_unlinked
, ARCH_CONVERT
, NULLAGINO
);
2038 offset
= ip
->i_boffset
+
2039 offsetof(xfs_dinode_t
, di_next_unlinked
);
2040 xfs_trans_inode_buf(tp
, ibp
);
2041 xfs_trans_log_buf(tp
, ibp
, offset
,
2042 (offset
+ sizeof(xfs_agino_t
) - 1));
2043 xfs_inobp_check(mp
, ibp
);
2045 xfs_trans_brelse(tp
, ibp
);
2048 * Point the bucket head pointer at the next inode.
2050 ASSERT(next_agino
!= 0);
2051 ASSERT(next_agino
!= agino
);
2052 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(next_agino
);
2053 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
2054 (sizeof(xfs_agino_t
) * bucket_index
);
2055 xfs_trans_log_buf(tp
, agibp
, offset
,
2056 (offset
+ sizeof(xfs_agino_t
) - 1));
2059 * We need to search the list for the inode being freed.
2061 next_agino
= be32_to_cpu(agi
->agi_unlinked
[bucket_index
]);
2063 while (next_agino
!= agino
) {
2065 * If the last inode wasn't the one pointing to
2066 * us, then release its buffer since we're not
2067 * going to do anything with it.
2069 if (last_ibp
!= NULL
) {
2070 xfs_trans_brelse(tp
, last_ibp
);
2072 next_ino
= XFS_AGINO_TO_INO(mp
, agno
, next_agino
);
2073 error
= xfs_inotobp(mp
, tp
, next_ino
, &last_dip
,
2074 &last_ibp
, &last_offset
);
2077 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
2078 error
, mp
->m_fsname
);
2081 next_agino
= INT_GET(last_dip
->di_next_unlinked
, ARCH_CONVERT
);
2082 ASSERT(next_agino
!= NULLAGINO
);
2083 ASSERT(next_agino
!= 0);
2086 * Now last_ibp points to the buffer previous to us on
2087 * the unlinked list. Pull us from the list.
2089 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, 0, 0);
2092 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2093 error
, mp
->m_fsname
);
2096 next_agino
= INT_GET(dip
->di_next_unlinked
, ARCH_CONVERT
);
2097 ASSERT(next_agino
!= 0);
2098 ASSERT(next_agino
!= agino
);
2099 if (next_agino
!= NULLAGINO
) {
2100 INT_SET(dip
->di_next_unlinked
, ARCH_CONVERT
, NULLAGINO
);
2101 offset
= ip
->i_boffset
+
2102 offsetof(xfs_dinode_t
, di_next_unlinked
);
2103 xfs_trans_inode_buf(tp
, ibp
);
2104 xfs_trans_log_buf(tp
, ibp
, offset
,
2105 (offset
+ sizeof(xfs_agino_t
) - 1));
2106 xfs_inobp_check(mp
, ibp
);
2108 xfs_trans_brelse(tp
, ibp
);
2111 * Point the previous inode on the list to the next inode.
2113 INT_SET(last_dip
->di_next_unlinked
, ARCH_CONVERT
, next_agino
);
2114 ASSERT(next_agino
!= 0);
2115 offset
= last_offset
+ offsetof(xfs_dinode_t
, di_next_unlinked
);
2116 xfs_trans_inode_buf(tp
, last_ibp
);
2117 xfs_trans_log_buf(tp
, last_ibp
, offset
,
2118 (offset
+ sizeof(xfs_agino_t
) - 1));
2119 xfs_inobp_check(mp
, last_ibp
);
2124 static __inline__
int xfs_inode_clean(xfs_inode_t
*ip
)
2126 return (((ip
->i_itemp
== NULL
) ||
2127 !(ip
->i_itemp
->ili_format
.ilf_fields
& XFS_ILOG_ALL
)) &&
2128 (ip
->i_update_core
== 0));
2133 xfs_inode_t
*free_ip
,
2137 xfs_mount_t
*mp
= free_ip
->i_mount
;
2138 int blks_per_cluster
;
2141 int i
, j
, found
, pre_flushed
;
2145 xfs_inode_t
*ip
, **ip_found
;
2146 xfs_inode_log_item_t
*iip
;
2147 xfs_log_item_t
*lip
;
2150 if (mp
->m_sb
.sb_blocksize
>= XFS_INODE_CLUSTER_SIZE(mp
)) {
2151 blks_per_cluster
= 1;
2152 ninodes
= mp
->m_sb
.sb_inopblock
;
2153 nbufs
= XFS_IALLOC_BLOCKS(mp
);
2155 blks_per_cluster
= XFS_INODE_CLUSTER_SIZE(mp
) /
2156 mp
->m_sb
.sb_blocksize
;
2157 ninodes
= blks_per_cluster
* mp
->m_sb
.sb_inopblock
;
2158 nbufs
= XFS_IALLOC_BLOCKS(mp
) / blks_per_cluster
;
2161 ip_found
= kmem_alloc(ninodes
* sizeof(xfs_inode_t
*), KM_NOFS
);
2163 for (j
= 0; j
< nbufs
; j
++, inum
+= ninodes
) {
2164 blkno
= XFS_AGB_TO_DADDR(mp
, XFS_INO_TO_AGNO(mp
, inum
),
2165 XFS_INO_TO_AGBNO(mp
, inum
));
2169 * Look for each inode in memory and attempt to lock it,
2170 * we can be racing with flush and tail pushing here.
2171 * any inode we get the locks on, add to an array of
2172 * inode items to process later.
2174 * The get the buffer lock, we could beat a flush
2175 * or tail pushing thread to the lock here, in which
2176 * case they will go looking for the inode buffer
2177 * and fail, we need some other form of interlock
2181 for (i
= 0; i
< ninodes
; i
++) {
2182 ih
= XFS_IHASH(mp
, inum
+ i
);
2183 read_lock(&ih
->ih_lock
);
2184 for (ip
= ih
->ih_next
; ip
!= NULL
; ip
= ip
->i_next
) {
2185 if (ip
->i_ino
== inum
+ i
)
2189 /* Inode not in memory or we found it already,
2192 if (!ip
|| (ip
->i_flags
& XFS_ISTALE
)) {
2193 read_unlock(&ih
->ih_lock
);
2197 if (xfs_inode_clean(ip
)) {
2198 read_unlock(&ih
->ih_lock
);
2202 /* If we can get the locks then add it to the
2203 * list, otherwise by the time we get the bp lock
2204 * below it will already be attached to the
2208 /* This inode will already be locked - by us, lets
2212 if (ip
== free_ip
) {
2213 if (xfs_iflock_nowait(ip
)) {
2214 ip
->i_flags
|= XFS_ISTALE
;
2216 if (xfs_inode_clean(ip
)) {
2219 ip_found
[found
++] = ip
;
2222 read_unlock(&ih
->ih_lock
);
2226 if (xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
)) {
2227 if (xfs_iflock_nowait(ip
)) {
2228 ip
->i_flags
|= XFS_ISTALE
;
2230 if (xfs_inode_clean(ip
)) {
2232 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2234 ip_found
[found
++] = ip
;
2237 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2241 read_unlock(&ih
->ih_lock
);
2244 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, blkno
,
2245 mp
->m_bsize
* blks_per_cluster
,
2249 lip
= XFS_BUF_FSPRIVATE(bp
, xfs_log_item_t
*);
2251 if (lip
->li_type
== XFS_LI_INODE
) {
2252 iip
= (xfs_inode_log_item_t
*)lip
;
2253 ASSERT(iip
->ili_logged
== 1);
2254 lip
->li_cb
= (void(*)(xfs_buf_t
*,xfs_log_item_t
*)) xfs_istale_done
;
2256 iip
->ili_flush_lsn
= iip
->ili_item
.li_lsn
;
2258 iip
->ili_inode
->i_flags
|= XFS_ISTALE
;
2261 lip
= lip
->li_bio_list
;
2264 for (i
= 0; i
< found
; i
++) {
2269 ip
->i_update_core
= 0;
2271 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2275 iip
->ili_last_fields
= iip
->ili_format
.ilf_fields
;
2276 iip
->ili_format
.ilf_fields
= 0;
2277 iip
->ili_logged
= 1;
2279 iip
->ili_flush_lsn
= iip
->ili_item
.li_lsn
;
2282 xfs_buf_attach_iodone(bp
,
2283 (void(*)(xfs_buf_t
*,xfs_log_item_t
*))
2284 xfs_istale_done
, (xfs_log_item_t
*)iip
);
2285 if (ip
!= free_ip
) {
2286 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2290 if (found
|| pre_flushed
)
2291 xfs_trans_stale_inode_buf(tp
, bp
);
2292 xfs_trans_binval(tp
, bp
);
2295 kmem_free(ip_found
, ninodes
* sizeof(xfs_inode_t
*));
2299 * This is called to return an inode to the inode free list.
2300 * The inode should already be truncated to 0 length and have
2301 * no pages associated with it. This routine also assumes that
2302 * the inode is already a part of the transaction.
2304 * The on-disk copy of the inode will have been added to the list
2305 * of unlinked inodes in the AGI. We need to remove the inode from
2306 * that list atomically with respect to freeing it here.
2312 xfs_bmap_free_t
*flist
)
2316 xfs_ino_t first_ino
;
2318 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
));
2319 ASSERT(ip
->i_transp
== tp
);
2320 ASSERT(ip
->i_d
.di_nlink
== 0);
2321 ASSERT(ip
->i_d
.di_nextents
== 0);
2322 ASSERT(ip
->i_d
.di_anextents
== 0);
2323 ASSERT((ip
->i_d
.di_size
== 0) ||
2324 ((ip
->i_d
.di_mode
& S_IFMT
) != S_IFREG
));
2325 ASSERT(ip
->i_d
.di_nblocks
== 0);
2328 * Pull the on-disk inode from the AGI unlinked list.
2330 error
= xfs_iunlink_remove(tp
, ip
);
2335 error
= xfs_difree(tp
, ip
->i_ino
, flist
, &delete, &first_ino
);
2339 ip
->i_d
.di_mode
= 0; /* mark incore inode as free */
2340 ip
->i_d
.di_flags
= 0;
2341 ip
->i_d
.di_dmevmask
= 0;
2342 ip
->i_d
.di_forkoff
= 0; /* mark the attr fork not in use */
2343 ip
->i_df
.if_ext_max
=
2344 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
2345 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
2346 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
2348 * Bump the generation count so no one will be confused
2349 * by reincarnations of this inode.
2352 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2355 xfs_ifree_cluster(ip
, tp
, first_ino
);
2362 * Reallocate the space for if_broot based on the number of records
2363 * being added or deleted as indicated in rec_diff. Move the records
2364 * and pointers in if_broot to fit the new size. When shrinking this
2365 * will eliminate holes between the records and pointers created by
2366 * the caller. When growing this will create holes to be filled in
2369 * The caller must not request to add more records than would fit in
2370 * the on-disk inode root. If the if_broot is currently NULL, then
2371 * if we adding records one will be allocated. The caller must also
2372 * not request that the number of records go below zero, although
2373 * it can go to zero.
2375 * ip -- the inode whose if_broot area is changing
2376 * ext_diff -- the change in the number of records, positive or negative,
2377 * requested for the if_broot array.
2387 xfs_bmbt_block_t
*new_broot
;
2394 * Handle the degenerate case quietly.
2396 if (rec_diff
== 0) {
2400 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2403 * If there wasn't any memory allocated before, just
2404 * allocate it now and get out.
2406 if (ifp
->if_broot_bytes
== 0) {
2407 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff
);
2408 ifp
->if_broot
= (xfs_bmbt_block_t
*)kmem_alloc(new_size
,
2410 ifp
->if_broot_bytes
= (int)new_size
;
2415 * If there is already an existing if_broot, then we need
2416 * to realloc() it and shift the pointers to their new
2417 * location. The records don't change location because
2418 * they are kept butted up against the btree block header.
2420 cur_max
= XFS_BMAP_BROOT_MAXRECS(ifp
->if_broot_bytes
);
2421 new_max
= cur_max
+ rec_diff
;
2422 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max
);
2423 ifp
->if_broot
= (xfs_bmbt_block_t
*)
2424 kmem_realloc(ifp
->if_broot
,
2426 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max
), /* old size */
2428 op
= (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp
->if_broot
, 1,
2429 ifp
->if_broot_bytes
);
2430 np
= (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp
->if_broot
, 1,
2432 ifp
->if_broot_bytes
= (int)new_size
;
2433 ASSERT(ifp
->if_broot_bytes
<=
2434 XFS_IFORK_SIZE(ip
, whichfork
) + XFS_BROOT_SIZE_ADJ
);
2435 memmove(np
, op
, cur_max
* (uint
)sizeof(xfs_dfsbno_t
));
2440 * rec_diff is less than 0. In this case, we are shrinking the
2441 * if_broot buffer. It must already exist. If we go to zero
2442 * records, just get rid of the root and clear the status bit.
2444 ASSERT((ifp
->if_broot
!= NULL
) && (ifp
->if_broot_bytes
> 0));
2445 cur_max
= XFS_BMAP_BROOT_MAXRECS(ifp
->if_broot_bytes
);
2446 new_max
= cur_max
+ rec_diff
;
2447 ASSERT(new_max
>= 0);
2449 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max
);
2453 new_broot
= (xfs_bmbt_block_t
*)kmem_alloc(new_size
, KM_SLEEP
);
2455 * First copy over the btree block header.
2457 memcpy(new_broot
, ifp
->if_broot
, sizeof(xfs_bmbt_block_t
));
2460 ifp
->if_flags
&= ~XFS_IFBROOT
;
2464 * Only copy the records and pointers if there are any.
2468 * First copy the records.
2470 op
= (char *)XFS_BMAP_BROOT_REC_ADDR(ifp
->if_broot
, 1,
2471 ifp
->if_broot_bytes
);
2472 np
= (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot
, 1,
2474 memcpy(np
, op
, new_max
* (uint
)sizeof(xfs_bmbt_rec_t
));
2477 * Then copy the pointers.
2479 op
= (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp
->if_broot
, 1,
2480 ifp
->if_broot_bytes
);
2481 np
= (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot
, 1,
2483 memcpy(np
, op
, new_max
* (uint
)sizeof(xfs_dfsbno_t
));
2485 kmem_free(ifp
->if_broot
, ifp
->if_broot_bytes
);
2486 ifp
->if_broot
= new_broot
;
2487 ifp
->if_broot_bytes
= (int)new_size
;
2488 ASSERT(ifp
->if_broot_bytes
<=
2489 XFS_IFORK_SIZE(ip
, whichfork
) + XFS_BROOT_SIZE_ADJ
);
2495 * This is called when the amount of space needed for if_data
2496 * is increased or decreased. The change in size is indicated by
2497 * the number of bytes that need to be added or deleted in the
2498 * byte_diff parameter.
2500 * If the amount of space needed has decreased below the size of the
2501 * inline buffer, then switch to using the inline buffer. Otherwise,
2502 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2503 * to what is needed.
2505 * ip -- the inode whose if_data area is changing
2506 * byte_diff -- the change in the number of bytes, positive or negative,
2507 * requested for the if_data array.
2519 if (byte_diff
== 0) {
2523 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2524 new_size
= (int)ifp
->if_bytes
+ byte_diff
;
2525 ASSERT(new_size
>= 0);
2527 if (new_size
== 0) {
2528 if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
2529 kmem_free(ifp
->if_u1
.if_data
, ifp
->if_real_bytes
);
2531 ifp
->if_u1
.if_data
= NULL
;
2533 } else if (new_size
<= sizeof(ifp
->if_u2
.if_inline_data
)) {
2535 * If the valid extents/data can fit in if_inline_ext/data,
2536 * copy them from the malloc'd vector and free it.
2538 if (ifp
->if_u1
.if_data
== NULL
) {
2539 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
2540 } else if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
2541 ASSERT(ifp
->if_real_bytes
!= 0);
2542 memcpy(ifp
->if_u2
.if_inline_data
, ifp
->if_u1
.if_data
,
2544 kmem_free(ifp
->if_u1
.if_data
, ifp
->if_real_bytes
);
2545 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
2550 * Stuck with malloc/realloc.
2551 * For inline data, the underlying buffer must be
2552 * a multiple of 4 bytes in size so that it can be
2553 * logged and stay on word boundaries. We enforce
2556 real_size
= roundup(new_size
, 4);
2557 if (ifp
->if_u1
.if_data
== NULL
) {
2558 ASSERT(ifp
->if_real_bytes
== 0);
2559 ifp
->if_u1
.if_data
= kmem_alloc(real_size
, KM_SLEEP
);
2560 } else if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
2562 * Only do the realloc if the underlying size
2563 * is really changing.
2565 if (ifp
->if_real_bytes
!= real_size
) {
2566 ifp
->if_u1
.if_data
=
2567 kmem_realloc(ifp
->if_u1
.if_data
,
2573 ASSERT(ifp
->if_real_bytes
== 0);
2574 ifp
->if_u1
.if_data
= kmem_alloc(real_size
, KM_SLEEP
);
2575 memcpy(ifp
->if_u1
.if_data
, ifp
->if_u2
.if_inline_data
,
2579 ifp
->if_real_bytes
= real_size
;
2580 ifp
->if_bytes
= new_size
;
2581 ASSERT(ifp
->if_bytes
<= XFS_IFORK_SIZE(ip
, whichfork
));
2588 * Map inode to disk block and offset.
2590 * mp -- the mount point structure for the current file system
2591 * tp -- the current transaction
2592 * ino -- the inode number of the inode to be located
2593 * imap -- this structure is filled in with the information necessary
2594 * to retrieve the given inode from disk
2595 * flags -- flags to pass to xfs_dilocate indicating whether or not
2596 * lookups in the inode btree were OK or not
2606 xfs_fsblock_t fsbno
;
2611 fsbno
= imap
->im_blkno
?
2612 XFS_DADDR_TO_FSB(mp
, imap
->im_blkno
) : NULLFSBLOCK
;
2613 error
= xfs_dilocate(mp
, tp
, ino
, &fsbno
, &len
, &off
, flags
);
2617 imap
->im_blkno
= XFS_FSB_TO_DADDR(mp
, fsbno
);
2618 imap
->im_len
= XFS_FSB_TO_BB(mp
, len
);
2619 imap
->im_agblkno
= XFS_FSB_TO_AGBNO(mp
, fsbno
);
2620 imap
->im_ioffset
= (ushort
)off
;
2621 imap
->im_boffset
= (ushort
)(off
<< mp
->m_sb
.sb_inodelog
);
2632 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2633 if (ifp
->if_broot
!= NULL
) {
2634 kmem_free(ifp
->if_broot
, ifp
->if_broot_bytes
);
2635 ifp
->if_broot
= NULL
;
2639 * If the format is local, then we can't have an extents
2640 * array so just look for an inline data array. If we're
2641 * not local then we may or may not have an extents list,
2642 * so check and free it up if we do.
2644 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
) {
2645 if ((ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) &&
2646 (ifp
->if_u1
.if_data
!= NULL
)) {
2647 ASSERT(ifp
->if_real_bytes
!= 0);
2648 kmem_free(ifp
->if_u1
.if_data
, ifp
->if_real_bytes
);
2649 ifp
->if_u1
.if_data
= NULL
;
2650 ifp
->if_real_bytes
= 0;
2652 } else if ((ifp
->if_flags
& XFS_IFEXTENTS
) &&
2653 ((ifp
->if_flags
& XFS_IFEXTIREC
) ||
2654 ((ifp
->if_u1
.if_extents
!= NULL
) &&
2655 (ifp
->if_u1
.if_extents
!= ifp
->if_u2
.if_inline_ext
)))) {
2656 ASSERT(ifp
->if_real_bytes
!= 0);
2657 xfs_iext_destroy(ifp
);
2659 ASSERT(ifp
->if_u1
.if_extents
== NULL
||
2660 ifp
->if_u1
.if_extents
== ifp
->if_u2
.if_inline_ext
);
2661 ASSERT(ifp
->if_real_bytes
== 0);
2662 if (whichfork
== XFS_ATTR_FORK
) {
2663 kmem_zone_free(xfs_ifork_zone
, ip
->i_afp
);
2669 * This is called free all the memory associated with an inode.
2670 * It must free the inode itself and any buffers allocated for
2671 * if_extents/if_data and if_broot. It must also free the lock
2672 * associated with the inode.
2679 switch (ip
->i_d
.di_mode
& S_IFMT
) {
2683 xfs_idestroy_fork(ip
, XFS_DATA_FORK
);
2687 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
2688 mrfree(&ip
->i_lock
);
2689 mrfree(&ip
->i_iolock
);
2690 freesema(&ip
->i_flock
);
2691 #ifdef XFS_BMAP_TRACE
2692 ktrace_free(ip
->i_xtrace
);
2694 #ifdef XFS_BMBT_TRACE
2695 ktrace_free(ip
->i_btrace
);
2698 ktrace_free(ip
->i_rwtrace
);
2700 #ifdef XFS_ILOCK_TRACE
2701 ktrace_free(ip
->i_lock_trace
);
2703 #ifdef XFS_DIR2_TRACE
2704 ktrace_free(ip
->i_dir_trace
);
2707 /* XXXdpd should be able to assert this but shutdown
2708 * is leaving the AIL behind. */
2709 ASSERT(((ip
->i_itemp
->ili_item
.li_flags
& XFS_LI_IN_AIL
) == 0) ||
2710 XFS_FORCED_SHUTDOWN(ip
->i_mount
));
2711 xfs_inode_item_destroy(ip
);
2713 kmem_zone_free(xfs_inode_zone
, ip
);
2718 * Increment the pin count of the given buffer.
2719 * This value is protected by ipinlock spinlock in the mount structure.
2725 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
));
2727 atomic_inc(&ip
->i_pincount
);
2731 * Decrement the pin count of the given inode, and wake up
2732 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2733 * inode must have been previously pinned with a call to xfs_ipin().
2739 ASSERT(atomic_read(&ip
->i_pincount
) > 0);
2741 if (atomic_dec_and_test(&ip
->i_pincount
)) {
2743 * If the inode is currently being reclaimed, the
2744 * linux inode _and_ the xfs vnode may have been
2745 * freed so we cannot reference either of them safely.
2746 * Hence we should not try to do anything to them
2747 * if the xfs inode is currently in the reclaim
2750 * However, we still need to issue the unpin wakeup
2751 * call as the inode reclaim may be blocked waiting for
2752 * the inode to become unpinned.
2754 if (!(ip
->i_flags
& (XFS_IRECLAIM
|XFS_IRECLAIMABLE
))) {
2755 vnode_t
*vp
= XFS_ITOV_NULL(ip
);
2757 /* make sync come back and flush this inode */
2759 struct inode
*inode
= vn_to_inode(vp
);
2761 if (!(inode
->i_state
&
2762 (I_NEW
|I_FREEING
|I_CLEAR
)))
2763 mark_inode_dirty_sync(inode
);
2766 wake_up(&ip
->i_ipin_wait
);
2771 * This is called to wait for the given inode to be unpinned.
2772 * It will sleep until this happens. The caller must have the
2773 * inode locked in at least shared mode so that the buffer cannot
2774 * be subsequently pinned once someone is waiting for it to be
2781 xfs_inode_log_item_t
*iip
;
2784 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
| MR_ACCESS
));
2786 if (atomic_read(&ip
->i_pincount
) == 0) {
2791 if (iip
&& iip
->ili_last_lsn
) {
2792 lsn
= iip
->ili_last_lsn
;
2798 * Give the log a push so we don't wait here too long.
2800 xfs_log_force(ip
->i_mount
, lsn
, XFS_LOG_FORCE
);
2802 wait_event(ip
->i_ipin_wait
, (atomic_read(&ip
->i_pincount
) == 0));
2807 * xfs_iextents_copy()
2809 * This is called to copy the REAL extents (as opposed to the delayed
2810 * allocation extents) from the inode into the given buffer. It
2811 * returns the number of bytes copied into the buffer.
2813 * If there are no delayed allocation extents, then we can just
2814 * memcpy() the extents into the buffer. Otherwise, we need to
2815 * examine each extent in turn and skip those which are delayed.
2820 xfs_bmbt_rec_t
*buffer
,
2824 xfs_bmbt_rec_t
*dest_ep
;
2826 #ifdef XFS_BMAP_TRACE
2827 static char fname
[] = "xfs_iextents_copy";
2832 xfs_fsblock_t start_block
;
2834 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2835 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
|MR_ACCESS
));
2836 ASSERT(ifp
->if_bytes
> 0);
2838 nrecs
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
2839 xfs_bmap_trace_exlist(fname
, ip
, nrecs
, whichfork
);
2843 * There are some delayed allocation extents in the
2844 * inode, so copy the extents one at a time and skip
2845 * the delayed ones. There must be at least one
2846 * non-delayed extent.
2850 for (i
= 0; i
< nrecs
; i
++) {
2851 ep
= xfs_iext_get_ext(ifp
, i
);
2852 start_block
= xfs_bmbt_get_startblock(ep
);
2853 if (ISNULLSTARTBLOCK(start_block
)) {
2855 * It's a delayed allocation extent, so skip it.
2860 /* Translate to on disk format */
2861 put_unaligned(INT_GET(ep
->l0
, ARCH_CONVERT
),
2862 (__uint64_t
*)&dest_ep
->l0
);
2863 put_unaligned(INT_GET(ep
->l1
, ARCH_CONVERT
),
2864 (__uint64_t
*)&dest_ep
->l1
);
2868 ASSERT(copied
!= 0);
2869 xfs_validate_extents(ifp
, copied
, 1, XFS_EXTFMT_INODE(ip
));
2871 return (copied
* (uint
)sizeof(xfs_bmbt_rec_t
));
2875 * Each of the following cases stores data into the same region
2876 * of the on-disk inode, so only one of them can be valid at
2877 * any given time. While it is possible to have conflicting formats
2878 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2879 * in EXTENTS format, this can only happen when the fork has
2880 * changed formats after being modified but before being flushed.
2881 * In these cases, the format always takes precedence, because the
2882 * format indicates the current state of the fork.
2889 xfs_inode_log_item_t
*iip
,
2896 #ifdef XFS_TRANS_DEBUG
2899 static const short brootflag
[2] =
2900 { XFS_ILOG_DBROOT
, XFS_ILOG_ABROOT
};
2901 static const short dataflag
[2] =
2902 { XFS_ILOG_DDATA
, XFS_ILOG_ADATA
};
2903 static const short extflag
[2] =
2904 { XFS_ILOG_DEXT
, XFS_ILOG_AEXT
};
2908 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2910 * This can happen if we gave up in iformat in an error path,
2911 * for the attribute fork.
2914 ASSERT(whichfork
== XFS_ATTR_FORK
);
2917 cp
= XFS_DFORK_PTR(dip
, whichfork
);
2919 switch (XFS_IFORK_FORMAT(ip
, whichfork
)) {
2920 case XFS_DINODE_FMT_LOCAL
:
2921 if ((iip
->ili_format
.ilf_fields
& dataflag
[whichfork
]) &&
2922 (ifp
->if_bytes
> 0)) {
2923 ASSERT(ifp
->if_u1
.if_data
!= NULL
);
2924 ASSERT(ifp
->if_bytes
<= XFS_IFORK_SIZE(ip
, whichfork
));
2925 memcpy(cp
, ifp
->if_u1
.if_data
, ifp
->if_bytes
);
2927 if (whichfork
== XFS_DATA_FORK
) {
2928 if (unlikely(XFS_DIR_SHORTFORM_VALIDATE_ONDISK(mp
, dip
))) {
2929 XFS_ERROR_REPORT("xfs_iflush_fork",
2930 XFS_ERRLEVEL_LOW
, mp
);
2931 return XFS_ERROR(EFSCORRUPTED
);
2936 case XFS_DINODE_FMT_EXTENTS
:
2937 ASSERT((ifp
->if_flags
& XFS_IFEXTENTS
) ||
2938 !(iip
->ili_format
.ilf_fields
& extflag
[whichfork
]));
2939 ASSERT((xfs_iext_get_ext(ifp
, 0) != NULL
) ||
2940 (ifp
->if_bytes
== 0));
2941 ASSERT((xfs_iext_get_ext(ifp
, 0) == NULL
) ||
2942 (ifp
->if_bytes
> 0));
2943 if ((iip
->ili_format
.ilf_fields
& extflag
[whichfork
]) &&
2944 (ifp
->if_bytes
> 0)) {
2945 ASSERT(XFS_IFORK_NEXTENTS(ip
, whichfork
) > 0);
2946 (void)xfs_iextents_copy(ip
, (xfs_bmbt_rec_t
*)cp
,
2951 case XFS_DINODE_FMT_BTREE
:
2952 if ((iip
->ili_format
.ilf_fields
& brootflag
[whichfork
]) &&
2953 (ifp
->if_broot_bytes
> 0)) {
2954 ASSERT(ifp
->if_broot
!= NULL
);
2955 ASSERT(ifp
->if_broot_bytes
<=
2956 (XFS_IFORK_SIZE(ip
, whichfork
) +
2957 XFS_BROOT_SIZE_ADJ
));
2958 xfs_bmbt_to_bmdr(ifp
->if_broot
, ifp
->if_broot_bytes
,
2959 (xfs_bmdr_block_t
*)cp
,
2960 XFS_DFORK_SIZE(dip
, mp
, whichfork
));
2964 case XFS_DINODE_FMT_DEV
:
2965 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_DEV
) {
2966 ASSERT(whichfork
== XFS_DATA_FORK
);
2967 INT_SET(dip
->di_u
.di_dev
, ARCH_CONVERT
, ip
->i_df
.if_u2
.if_rdev
);
2971 case XFS_DINODE_FMT_UUID
:
2972 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_UUID
) {
2973 ASSERT(whichfork
== XFS_DATA_FORK
);
2974 memcpy(&dip
->di_u
.di_muuid
, &ip
->i_df
.if_u2
.if_uuid
,
2988 * xfs_iflush() will write a modified inode's changes out to the
2989 * inode's on disk home. The caller must have the inode lock held
2990 * in at least shared mode and the inode flush semaphore must be
2991 * held as well. The inode lock will still be held upon return from
2992 * the call and the caller is free to unlock it.
2993 * The inode flush lock will be unlocked when the inode reaches the disk.
2994 * The flags indicate how the inode's buffer should be written out.
3001 xfs_inode_log_item_t
*iip
;
3009 int clcount
; /* count of inodes clustered */
3011 enum { INT_DELWRI
= (1 << 0), INT_ASYNC
= (1 << 1) };
3014 XFS_STATS_INC(xs_iflush_count
);
3016 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
|MR_ACCESS
));
3017 ASSERT(valusema(&ip
->i_flock
) <= 0);
3018 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
3019 ip
->i_d
.di_nextents
> ip
->i_df
.if_ext_max
);
3025 * If the inode isn't dirty, then just release the inode
3026 * flush lock and do nothing.
3028 if ((ip
->i_update_core
== 0) &&
3029 ((iip
== NULL
) || !(iip
->ili_format
.ilf_fields
& XFS_ILOG_ALL
))) {
3030 ASSERT((iip
!= NULL
) ?
3031 !(iip
->ili_item
.li_flags
& XFS_LI_IN_AIL
) : 1);
3037 * We can't flush the inode until it is unpinned, so
3038 * wait for it. We know noone new can pin it, because
3039 * we are holding the inode lock shared and you need
3040 * to hold it exclusively to pin the inode.
3042 xfs_iunpin_wait(ip
);
3045 * This may have been unpinned because the filesystem is shutting
3046 * down forcibly. If that's the case we must not write this inode
3047 * to disk, because the log record didn't make it to disk!
3049 if (XFS_FORCED_SHUTDOWN(mp
)) {
3050 ip
->i_update_core
= 0;
3052 iip
->ili_format
.ilf_fields
= 0;
3054 return XFS_ERROR(EIO
);
3058 * Get the buffer containing the on-disk inode.
3060 error
= xfs_itobp(mp
, NULL
, ip
, &dip
, &bp
, 0, 0);
3067 * Decide how buffer will be flushed out. This is done before
3068 * the call to xfs_iflush_int because this field is zeroed by it.
3070 if (iip
!= NULL
&& iip
->ili_format
.ilf_fields
!= 0) {
3072 * Flush out the inode buffer according to the directions
3073 * of the caller. In the cases where the caller has given
3074 * us a choice choose the non-delwri case. This is because
3075 * the inode is in the AIL and we need to get it out soon.
3078 case XFS_IFLUSH_SYNC
:
3079 case XFS_IFLUSH_DELWRI_ELSE_SYNC
:
3082 case XFS_IFLUSH_ASYNC
:
3083 case XFS_IFLUSH_DELWRI_ELSE_ASYNC
:
3086 case XFS_IFLUSH_DELWRI
:
3096 case XFS_IFLUSH_DELWRI_ELSE_SYNC
:
3097 case XFS_IFLUSH_DELWRI_ELSE_ASYNC
:
3098 case XFS_IFLUSH_DELWRI
:
3101 case XFS_IFLUSH_ASYNC
:
3104 case XFS_IFLUSH_SYNC
:
3115 * First flush out the inode that xfs_iflush was called with.
3117 error
= xfs_iflush_int(ip
, bp
);
3124 * see if other inodes can be gathered into this write
3127 ip
->i_chash
->chl_buf
= bp
;
3129 ch
= XFS_CHASH(mp
, ip
->i_blkno
);
3130 s
= mutex_spinlock(&ch
->ch_lock
);
3133 for (iq
= ip
->i_cnext
; iq
!= ip
; iq
= iq
->i_cnext
) {
3135 * Do an un-protected check to see if the inode is dirty and
3136 * is a candidate for flushing. These checks will be repeated
3137 * later after the appropriate locks are acquired.
3140 if ((iq
->i_update_core
== 0) &&
3142 !(iip
->ili_format
.ilf_fields
& XFS_ILOG_ALL
)) &&
3143 xfs_ipincount(iq
) == 0) {
3148 * Try to get locks. If any are unavailable,
3149 * then this inode cannot be flushed and is skipped.
3152 /* get inode locks (just i_lock) */
3153 if (xfs_ilock_nowait(iq
, XFS_ILOCK_SHARED
)) {
3154 /* get inode flush lock */
3155 if (xfs_iflock_nowait(iq
)) {
3156 /* check if pinned */
3157 if (xfs_ipincount(iq
) == 0) {
3158 /* arriving here means that
3159 * this inode can be flushed.
3160 * first re-check that it's
3164 if ((iq
->i_update_core
!= 0)||
3166 (iip
->ili_format
.ilf_fields
& XFS_ILOG_ALL
))) {
3168 error
= xfs_iflush_int(iq
, bp
);
3172 goto cluster_corrupt_out
;
3181 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
3184 mutex_spinunlock(&ch
->ch_lock
, s
);
3187 XFS_STATS_INC(xs_icluster_flushcnt
);
3188 XFS_STATS_ADD(xs_icluster_flushinode
, clcount
);
3192 * If the buffer is pinned then push on the log so we won't
3193 * get stuck waiting in the write for too long.
3195 if (XFS_BUF_ISPINNED(bp
)){
3196 xfs_log_force(mp
, (xfs_lsn_t
)0, XFS_LOG_FORCE
);
3199 if (flags
& INT_DELWRI
) {
3200 xfs_bdwrite(mp
, bp
);
3201 } else if (flags
& INT_ASYNC
) {
3202 xfs_bawrite(mp
, bp
);
3204 error
= xfs_bwrite(mp
, bp
);
3210 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
3211 xfs_iflush_abort(ip
);
3213 * Unlocks the flush lock
3215 return XFS_ERROR(EFSCORRUPTED
);
3217 cluster_corrupt_out
:
3218 /* Corruption detected in the clustering loop. Invalidate the
3219 * inode buffer and shut down the filesystem.
3221 mutex_spinunlock(&ch
->ch_lock
, s
);
3224 * Clean up the buffer. If it was B_DELWRI, just release it --
3225 * brelse can handle it with no problems. If not, shut down the
3226 * filesystem before releasing the buffer.
3228 if ((bufwasdelwri
= XFS_BUF_ISDELAYWRITE(bp
))) {
3232 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
3236 * Just like incore_relse: if we have b_iodone functions,
3237 * mark the buffer as an error and call them. Otherwise
3238 * mark it as stale and brelse.
3240 if (XFS_BUF_IODONE_FUNC(bp
)) {
3241 XFS_BUF_CLR_BDSTRAT_FUNC(bp
);
3245 XFS_BUF_ERROR(bp
,EIO
);
3253 xfs_iflush_abort(iq
);
3255 * Unlocks the flush lock
3257 return XFS_ERROR(EFSCORRUPTED
);
3266 xfs_inode_log_item_t
*iip
;
3269 #ifdef XFS_TRANS_DEBUG
3274 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
|MR_ACCESS
));
3275 ASSERT(valusema(&ip
->i_flock
) <= 0);
3276 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
3277 ip
->i_d
.di_nextents
> ip
->i_df
.if_ext_max
);
3284 * If the inode isn't dirty, then just release the inode
3285 * flush lock and do nothing.
3287 if ((ip
->i_update_core
== 0) &&
3288 ((iip
== NULL
) || !(iip
->ili_format
.ilf_fields
& XFS_ILOG_ALL
))) {
3293 /* set *dip = inode's place in the buffer */
3294 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, ip
->i_boffset
);
3297 * Clear i_update_core before copying out the data.
3298 * This is for coordination with our timestamp updates
3299 * that don't hold the inode lock. They will always
3300 * update the timestamps BEFORE setting i_update_core,
3301 * so if we clear i_update_core after they set it we
3302 * are guaranteed to see their updates to the timestamps.
3303 * I believe that this depends on strongly ordered memory
3304 * semantics, but we have that. We use the SYNCHRONIZE
3305 * macro to make sure that the compiler does not reorder
3306 * the i_update_core access below the data copy below.
3308 ip
->i_update_core
= 0;
3312 * Make sure to get the latest atime from the Linux inode.
3314 xfs_synchronize_atime(ip
);
3316 if (XFS_TEST_ERROR(INT_GET(dip
->di_core
.di_magic
,ARCH_CONVERT
) != XFS_DINODE_MAGIC
,
3317 mp
, XFS_ERRTAG_IFLUSH_1
, XFS_RANDOM_IFLUSH_1
)) {
3318 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3319 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3320 ip
->i_ino
, (int) INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
), dip
);
3323 if (XFS_TEST_ERROR(ip
->i_d
.di_magic
!= XFS_DINODE_MAGIC
,
3324 mp
, XFS_ERRTAG_IFLUSH_2
, XFS_RANDOM_IFLUSH_2
)) {
3325 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3326 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3327 ip
->i_ino
, ip
, ip
->i_d
.di_magic
);
3330 if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
) {
3332 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3333 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
),
3334 mp
, XFS_ERRTAG_IFLUSH_3
, XFS_RANDOM_IFLUSH_3
)) {
3335 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3336 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3340 } else if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFDIR
) {
3342 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3343 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
) &&
3344 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_LOCAL
),
3345 mp
, XFS_ERRTAG_IFLUSH_4
, XFS_RANDOM_IFLUSH_4
)) {
3346 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3347 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3352 if (XFS_TEST_ERROR(ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
>
3353 ip
->i_d
.di_nblocks
, mp
, XFS_ERRTAG_IFLUSH_5
,
3354 XFS_RANDOM_IFLUSH_5
)) {
3355 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3356 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3358 ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
,
3363 if (XFS_TEST_ERROR(ip
->i_d
.di_forkoff
> mp
->m_sb
.sb_inodesize
,
3364 mp
, XFS_ERRTAG_IFLUSH_6
, XFS_RANDOM_IFLUSH_6
)) {
3365 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3366 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3367 ip
->i_ino
, ip
->i_d
.di_forkoff
, ip
);
3371 * bump the flush iteration count, used to detect flushes which
3372 * postdate a log record during recovery.
3375 ip
->i_d
.di_flushiter
++;
3378 * Copy the dirty parts of the inode into the on-disk
3379 * inode. We always copy out the core of the inode,
3380 * because if the inode is dirty at all the core must
3383 xfs_xlate_dinode_core((xfs_caddr_t
)&(dip
->di_core
), &(ip
->i_d
), -1);
3385 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3386 if (ip
->i_d
.di_flushiter
== DI_MAX_FLUSH
)
3387 ip
->i_d
.di_flushiter
= 0;
3390 * If this is really an old format inode and the superblock version
3391 * has not been updated to support only new format inodes, then
3392 * convert back to the old inode format. If the superblock version
3393 * has been updated, then make the conversion permanent.
3395 ASSERT(ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
||
3396 XFS_SB_VERSION_HASNLINK(&mp
->m_sb
));
3397 if (ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
) {
3398 if (!XFS_SB_VERSION_HASNLINK(&mp
->m_sb
)) {
3402 ASSERT(ip
->i_d
.di_nlink
<= XFS_MAXLINK_1
);
3403 INT_SET(dip
->di_core
.di_onlink
, ARCH_CONVERT
, ip
->i_d
.di_nlink
);
3406 * The superblock version has already been bumped,
3407 * so just make the conversion to the new inode
3410 ip
->i_d
.di_version
= XFS_DINODE_VERSION_2
;
3411 INT_SET(dip
->di_core
.di_version
, ARCH_CONVERT
, XFS_DINODE_VERSION_2
);
3412 ip
->i_d
.di_onlink
= 0;
3413 dip
->di_core
.di_onlink
= 0;
3414 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
3415 memset(&(dip
->di_core
.di_pad
[0]), 0,
3416 sizeof(dip
->di_core
.di_pad
));
3417 ASSERT(ip
->i_d
.di_projid
== 0);
3421 if (xfs_iflush_fork(ip
, dip
, iip
, XFS_DATA_FORK
, bp
) == EFSCORRUPTED
) {
3425 if (XFS_IFORK_Q(ip
)) {
3427 * The only error from xfs_iflush_fork is on the data fork.
3429 (void) xfs_iflush_fork(ip
, dip
, iip
, XFS_ATTR_FORK
, bp
);
3431 xfs_inobp_check(mp
, bp
);
3434 * We've recorded everything logged in the inode, so we'd
3435 * like to clear the ilf_fields bits so we don't log and
3436 * flush things unnecessarily. However, we can't stop
3437 * logging all this information until the data we've copied
3438 * into the disk buffer is written to disk. If we did we might
3439 * overwrite the copy of the inode in the log with all the
3440 * data after re-logging only part of it, and in the face of
3441 * a crash we wouldn't have all the data we need to recover.
3443 * What we do is move the bits to the ili_last_fields field.
3444 * When logging the inode, these bits are moved back to the
3445 * ilf_fields field. In the xfs_iflush_done() routine we
3446 * clear ili_last_fields, since we know that the information
3447 * those bits represent is permanently on disk. As long as
3448 * the flush completes before the inode is logged again, then
3449 * both ilf_fields and ili_last_fields will be cleared.
3451 * We can play with the ilf_fields bits here, because the inode
3452 * lock must be held exclusively in order to set bits there
3453 * and the flush lock protects the ili_last_fields bits.
3454 * Set ili_logged so the flush done
3455 * routine can tell whether or not to look in the AIL.
3456 * Also, store the current LSN of the inode so that we can tell
3457 * whether the item has moved in the AIL from xfs_iflush_done().
3458 * In order to read the lsn we need the AIL lock, because
3459 * it is a 64 bit value that cannot be read atomically.
3461 if (iip
!= NULL
&& iip
->ili_format
.ilf_fields
!= 0) {
3462 iip
->ili_last_fields
= iip
->ili_format
.ilf_fields
;
3463 iip
->ili_format
.ilf_fields
= 0;
3464 iip
->ili_logged
= 1;
3466 ASSERT(sizeof(xfs_lsn_t
) == 8); /* don't lock if it shrinks */
3468 iip
->ili_flush_lsn
= iip
->ili_item
.li_lsn
;
3472 * Attach the function xfs_iflush_done to the inode's
3473 * buffer. This will remove the inode from the AIL
3474 * and unlock the inode's flush lock when the inode is
3475 * completely written to disk.
3477 xfs_buf_attach_iodone(bp
, (void(*)(xfs_buf_t
*,xfs_log_item_t
*))
3478 xfs_iflush_done
, (xfs_log_item_t
*)iip
);
3480 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) != NULL
);
3481 ASSERT(XFS_BUF_IODONE_FUNC(bp
) != NULL
);
3484 * We're flushing an inode which is not in the AIL and has
3485 * not been logged but has i_update_core set. For this
3486 * case we can use a B_DELWRI flush and immediately drop
3487 * the inode flush lock because we can avoid the whole
3488 * AIL state thing. It's OK to drop the flush lock now,
3489 * because we've already locked the buffer and to do anything
3490 * you really need both.
3493 ASSERT(iip
->ili_logged
== 0);
3494 ASSERT(iip
->ili_last_fields
== 0);
3495 ASSERT((iip
->ili_item
.li_flags
& XFS_LI_IN_AIL
) == 0);
3503 return XFS_ERROR(EFSCORRUPTED
);
3508 * Flush all inactive inodes in mp.
3518 XFS_MOUNT_ILOCK(mp
);
3524 /* Make sure we skip markers inserted by sync */
3525 if (ip
->i_mount
== NULL
) {
3530 vp
= XFS_ITOV_NULL(ip
);
3532 XFS_MOUNT_IUNLOCK(mp
);
3533 xfs_finish_reclaim(ip
, 0, XFS_IFLUSH_ASYNC
);
3537 ASSERT(vn_count(vp
) == 0);
3540 } while (ip
!= mp
->m_inodes
);
3542 XFS_MOUNT_IUNLOCK(mp
);
3546 * xfs_iaccess: check accessibility of inode for mode.
3555 mode_t orgmode
= mode
;
3556 struct inode
*inode
= vn_to_inode(XFS_ITOV(ip
));
3558 if (mode
& S_IWUSR
) {
3559 umode_t imode
= inode
->i_mode
;
3561 if (IS_RDONLY(inode
) &&
3562 (S_ISREG(imode
) || S_ISDIR(imode
) || S_ISLNK(imode
)))
3563 return XFS_ERROR(EROFS
);
3565 if (IS_IMMUTABLE(inode
))
3566 return XFS_ERROR(EACCES
);
3570 * If there's an Access Control List it's used instead of
3573 if ((error
= _ACL_XFS_IACCESS(ip
, mode
, cr
)) != -1)
3574 return error
? XFS_ERROR(error
) : 0;
3576 if (current_fsuid(cr
) != ip
->i_d
.di_uid
) {
3578 if (!in_group_p((gid_t
)ip
->i_d
.di_gid
))
3583 * If the DACs are ok we don't need any capability check.
3585 if ((ip
->i_d
.di_mode
& mode
) == mode
)
3588 * Read/write DACs are always overridable.
3589 * Executable DACs are overridable if at least one exec bit is set.
3591 if (!(orgmode
& S_IXUSR
) ||
3592 (inode
->i_mode
& S_IXUGO
) || S_ISDIR(inode
->i_mode
))
3593 if (capable_cred(cr
, CAP_DAC_OVERRIDE
))
3596 if ((orgmode
== S_IRUSR
) ||
3597 (S_ISDIR(inode
->i_mode
) && (!(orgmode
& S_IWUSR
)))) {
3598 if (capable_cred(cr
, CAP_DAC_READ_SEARCH
))
3601 cmn_err(CE_NOTE
, "Ick: mode=%o, orgmode=%o", mode
, orgmode
);
3603 return XFS_ERROR(EACCES
);
3605 return XFS_ERROR(EACCES
);
3609 * xfs_iroundup: round up argument to next power of two
3618 if ((v
& (v
- 1)) == 0)
3620 ASSERT((v
& 0x80000000) == 0);
3621 if ((v
& (v
+ 1)) == 0)
3623 for (i
= 0, m
= 1; i
< 31; i
++, m
<<= 1) {
3627 if ((v
& (v
+ 1)) == 0)
3634 #ifdef XFS_ILOCK_TRACE
3635 ktrace_t
*xfs_ilock_trace_buf
;
3638 xfs_ilock_trace(xfs_inode_t
*ip
, int lock
, unsigned int lockflags
, inst_t
*ra
)
3640 ktrace_enter(ip
->i_lock_trace
,
3642 (void *)(unsigned long)lock
, /* 1 = LOCK, 3=UNLOCK, etc */
3643 (void *)(unsigned long)lockflags
, /* XFS_ILOCK_EXCL etc */
3644 (void *)ra
, /* caller of ilock */
3645 (void *)(unsigned long)current_cpu(),
3646 (void *)(unsigned long)current_pid(),
3647 NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
);
3652 * Return a pointer to the extent record at file index idx.
3656 xfs_ifork_t
*ifp
, /* inode fork pointer */
3657 xfs_extnum_t idx
) /* index of target extent */
3660 if ((ifp
->if_flags
& XFS_IFEXTIREC
) && (idx
== 0)) {
3661 return ifp
->if_u1
.if_ext_irec
->er_extbuf
;
3662 } else if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3663 xfs_ext_irec_t
*erp
; /* irec pointer */
3664 int erp_idx
= 0; /* irec index */
3665 xfs_extnum_t page_idx
= idx
; /* ext index in target list */
3667 erp
= xfs_iext_idx_to_irec(ifp
, &page_idx
, &erp_idx
, 0);
3668 return &erp
->er_extbuf
[page_idx
];
3669 } else if (ifp
->if_bytes
) {
3670 return &ifp
->if_u1
.if_extents
[idx
];
3677 * Insert new item(s) into the extent records for incore inode
3678 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3682 xfs_ifork_t
*ifp
, /* inode fork pointer */
3683 xfs_extnum_t idx
, /* starting index of new items */
3684 xfs_extnum_t count
, /* number of inserted items */
3685 xfs_bmbt_irec_t
*new) /* items to insert */
3687 xfs_bmbt_rec_t
*ep
; /* extent record pointer */
3688 xfs_extnum_t i
; /* extent record index */
3690 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
3691 xfs_iext_add(ifp
, idx
, count
);
3692 for (i
= idx
; i
< idx
+ count
; i
++, new++) {
3693 ep
= xfs_iext_get_ext(ifp
, i
);
3694 xfs_bmbt_set_all(ep
, new);
3699 * This is called when the amount of space required for incore file
3700 * extents needs to be increased. The ext_diff parameter stores the
3701 * number of new extents being added and the idx parameter contains
3702 * the extent index where the new extents will be added. If the new
3703 * extents are being appended, then we just need to (re)allocate and
3704 * initialize the space. Otherwise, if the new extents are being
3705 * inserted into the middle of the existing entries, a bit more work
3706 * is required to make room for the new extents to be inserted. The
3707 * caller is responsible for filling in the new extent entries upon
3712 xfs_ifork_t
*ifp
, /* inode fork pointer */
3713 xfs_extnum_t idx
, /* index to begin adding exts */
3714 int ext_diff
) /* number of extents to add */
3716 int byte_diff
; /* new bytes being added */
3717 int new_size
; /* size of extents after adding */
3718 xfs_extnum_t nextents
; /* number of extents in file */
3720 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3721 ASSERT((idx
>= 0) && (idx
<= nextents
));
3722 byte_diff
= ext_diff
* sizeof(xfs_bmbt_rec_t
);
3723 new_size
= ifp
->if_bytes
+ byte_diff
;
3725 * If the new number of extents (nextents + ext_diff)
3726 * fits inside the inode, then continue to use the inline
3729 if (nextents
+ ext_diff
<= XFS_INLINE_EXTS
) {
3730 if (idx
< nextents
) {
3731 memmove(&ifp
->if_u2
.if_inline_ext
[idx
+ ext_diff
],
3732 &ifp
->if_u2
.if_inline_ext
[idx
],
3733 (nextents
- idx
) * sizeof(xfs_bmbt_rec_t
));
3734 memset(&ifp
->if_u2
.if_inline_ext
[idx
], 0, byte_diff
);
3736 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
3737 ifp
->if_real_bytes
= 0;
3738 ifp
->if_lastex
= nextents
+ ext_diff
;
3741 * Otherwise use a linear (direct) extent list.
3742 * If the extents are currently inside the inode,
3743 * xfs_iext_realloc_direct will switch us from
3744 * inline to direct extent allocation mode.
3746 else if (nextents
+ ext_diff
<= XFS_LINEAR_EXTS
) {
3747 xfs_iext_realloc_direct(ifp
, new_size
);
3748 if (idx
< nextents
) {
3749 memmove(&ifp
->if_u1
.if_extents
[idx
+ ext_diff
],
3750 &ifp
->if_u1
.if_extents
[idx
],
3751 (nextents
- idx
) * sizeof(xfs_bmbt_rec_t
));
3752 memset(&ifp
->if_u1
.if_extents
[idx
], 0, byte_diff
);
3755 /* Indirection array */
3757 xfs_ext_irec_t
*erp
;
3761 ASSERT(nextents
+ ext_diff
> XFS_LINEAR_EXTS
);
3762 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3763 erp
= xfs_iext_idx_to_irec(ifp
, &page_idx
, &erp_idx
, 1);
3765 xfs_iext_irec_init(ifp
);
3766 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3767 erp
= ifp
->if_u1
.if_ext_irec
;
3769 /* Extents fit in target extent page */
3770 if (erp
&& erp
->er_extcount
+ ext_diff
<= XFS_LINEAR_EXTS
) {
3771 if (page_idx
< erp
->er_extcount
) {
3772 memmove(&erp
->er_extbuf
[page_idx
+ ext_diff
],
3773 &erp
->er_extbuf
[page_idx
],
3774 (erp
->er_extcount
- page_idx
) *
3775 sizeof(xfs_bmbt_rec_t
));
3776 memset(&erp
->er_extbuf
[page_idx
], 0, byte_diff
);
3778 erp
->er_extcount
+= ext_diff
;
3779 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, ext_diff
);
3781 /* Insert a new extent page */
3783 xfs_iext_add_indirect_multi(ifp
,
3784 erp_idx
, page_idx
, ext_diff
);
3787 * If extent(s) are being appended to the last page in
3788 * the indirection array and the new extent(s) don't fit
3789 * in the page, then erp is NULL and erp_idx is set to
3790 * the next index needed in the indirection array.
3793 int count
= ext_diff
;
3796 erp
= xfs_iext_irec_new(ifp
, erp_idx
);
3797 erp
->er_extcount
= count
;
3798 count
-= MIN(count
, (int)XFS_LINEAR_EXTS
);
3805 ifp
->if_bytes
= new_size
;
3809 * This is called when incore extents are being added to the indirection
3810 * array and the new extents do not fit in the target extent list. The
3811 * erp_idx parameter contains the irec index for the target extent list
3812 * in the indirection array, and the idx parameter contains the extent
3813 * index within the list. The number of extents being added is stored
3814 * in the count parameter.
3816 * |-------| |-------|
3817 * | | | | idx - number of extents before idx
3819 * | | | | count - number of extents being inserted at idx
3820 * |-------| |-------|
3821 * | count | | nex2 | nex2 - number of extents after idx + count
3822 * |-------| |-------|
3825 xfs_iext_add_indirect_multi(
3826 xfs_ifork_t
*ifp
, /* inode fork pointer */
3827 int erp_idx
, /* target extent irec index */
3828 xfs_extnum_t idx
, /* index within target list */
3829 int count
) /* new extents being added */
3831 int byte_diff
; /* new bytes being added */
3832 xfs_ext_irec_t
*erp
; /* pointer to irec entry */
3833 xfs_extnum_t ext_diff
; /* number of extents to add */
3834 xfs_extnum_t ext_cnt
; /* new extents still needed */
3835 xfs_extnum_t nex2
; /* extents after idx + count */
3836 xfs_bmbt_rec_t
*nex2_ep
= NULL
; /* temp list for nex2 extents */
3837 int nlists
; /* number of irec's (lists) */
3839 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3840 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
3841 nex2
= erp
->er_extcount
- idx
;
3842 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3845 * Save second part of target extent list
3846 * (all extents past */
3848 byte_diff
= nex2
* sizeof(xfs_bmbt_rec_t
);
3849 nex2_ep
= (xfs_bmbt_rec_t
*) kmem_alloc(byte_diff
, KM_SLEEP
);
3850 memmove(nex2_ep
, &erp
->er_extbuf
[idx
], byte_diff
);
3851 erp
->er_extcount
-= nex2
;
3852 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, -nex2
);
3853 memset(&erp
->er_extbuf
[idx
], 0, byte_diff
);
3857 * Add the new extents to the end of the target
3858 * list, then allocate new irec record(s) and
3859 * extent buffer(s) as needed to store the rest
3860 * of the new extents.
3863 ext_diff
= MIN(ext_cnt
, (int)XFS_LINEAR_EXTS
- erp
->er_extcount
);
3865 erp
->er_extcount
+= ext_diff
;
3866 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, ext_diff
);
3867 ext_cnt
-= ext_diff
;
3871 erp
= xfs_iext_irec_new(ifp
, erp_idx
);
3872 ext_diff
= MIN(ext_cnt
, (int)XFS_LINEAR_EXTS
);
3873 erp
->er_extcount
= ext_diff
;
3874 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, ext_diff
);
3875 ext_cnt
-= ext_diff
;
3878 /* Add nex2 extents back to indirection array */
3880 xfs_extnum_t ext_avail
;
3883 byte_diff
= nex2
* sizeof(xfs_bmbt_rec_t
);
3884 ext_avail
= XFS_LINEAR_EXTS
- erp
->er_extcount
;
3887 * If nex2 extents fit in the current page, append
3888 * nex2_ep after the new extents.
3890 if (nex2
<= ext_avail
) {
3891 i
= erp
->er_extcount
;
3894 * Otherwise, check if space is available in the
3897 else if ((erp_idx
< nlists
- 1) &&
3898 (nex2
<= (ext_avail
= XFS_LINEAR_EXTS
-
3899 ifp
->if_u1
.if_ext_irec
[erp_idx
+1].er_extcount
))) {
3902 /* Create a hole for nex2 extents */
3903 memmove(&erp
->er_extbuf
[nex2
], erp
->er_extbuf
,
3904 erp
->er_extcount
* sizeof(xfs_bmbt_rec_t
));
3907 * Final choice, create a new extent page for
3912 erp
= xfs_iext_irec_new(ifp
, erp_idx
);
3914 memmove(&erp
->er_extbuf
[i
], nex2_ep
, byte_diff
);
3915 kmem_free(nex2_ep
, byte_diff
);
3916 erp
->er_extcount
+= nex2
;
3917 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, nex2
);
3922 * This is called when the amount of space required for incore file
3923 * extents needs to be decreased. The ext_diff parameter stores the
3924 * number of extents to be removed and the idx parameter contains
3925 * the extent index where the extents will be removed from.
3927 * If the amount of space needed has decreased below the linear
3928 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3929 * extent array. Otherwise, use kmem_realloc() to adjust the
3930 * size to what is needed.
3934 xfs_ifork_t
*ifp
, /* inode fork pointer */
3935 xfs_extnum_t idx
, /* index to begin removing exts */
3936 int ext_diff
) /* number of extents to remove */
3938 xfs_extnum_t nextents
; /* number of extents in file */
3939 int new_size
; /* size of extents after removal */
3941 ASSERT(ext_diff
> 0);
3942 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3943 new_size
= (nextents
- ext_diff
) * sizeof(xfs_bmbt_rec_t
);
3945 if (new_size
== 0) {
3946 xfs_iext_destroy(ifp
);
3947 } else if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3948 xfs_iext_remove_indirect(ifp
, idx
, ext_diff
);
3949 } else if (ifp
->if_real_bytes
) {
3950 xfs_iext_remove_direct(ifp
, idx
, ext_diff
);
3952 xfs_iext_remove_inline(ifp
, idx
, ext_diff
);
3954 ifp
->if_bytes
= new_size
;
3958 * This removes ext_diff extents from the inline buffer, beginning
3959 * at extent index idx.
3962 xfs_iext_remove_inline(
3963 xfs_ifork_t
*ifp
, /* inode fork pointer */
3964 xfs_extnum_t idx
, /* index to begin removing exts */
3965 int ext_diff
) /* number of extents to remove */
3967 int nextents
; /* number of extents in file */
3969 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
));
3970 ASSERT(idx
< XFS_INLINE_EXTS
);
3971 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3972 ASSERT(((nextents
- ext_diff
) > 0) &&
3973 (nextents
- ext_diff
) < XFS_INLINE_EXTS
);
3975 if (idx
+ ext_diff
< nextents
) {
3976 memmove(&ifp
->if_u2
.if_inline_ext
[idx
],
3977 &ifp
->if_u2
.if_inline_ext
[idx
+ ext_diff
],
3978 (nextents
- (idx
+ ext_diff
)) *
3979 sizeof(xfs_bmbt_rec_t
));
3980 memset(&ifp
->if_u2
.if_inline_ext
[nextents
- ext_diff
],
3981 0, ext_diff
* sizeof(xfs_bmbt_rec_t
));
3983 memset(&ifp
->if_u2
.if_inline_ext
[idx
], 0,
3984 ext_diff
* sizeof(xfs_bmbt_rec_t
));
3989 * This removes ext_diff extents from a linear (direct) extent list,
3990 * beginning at extent index idx. If the extents are being removed
3991 * from the end of the list (ie. truncate) then we just need to re-
3992 * allocate the list to remove the extra space. Otherwise, if the
3993 * extents are being removed from the middle of the existing extent
3994 * entries, then we first need to move the extent records beginning
3995 * at idx + ext_diff up in the list to overwrite the records being
3996 * removed, then remove the extra space via kmem_realloc.
3999 xfs_iext_remove_direct(
4000 xfs_ifork_t
*ifp
, /* inode fork pointer */
4001 xfs_extnum_t idx
, /* index to begin removing exts */
4002 int ext_diff
) /* number of extents to remove */
4004 xfs_extnum_t nextents
; /* number of extents in file */
4005 int new_size
; /* size of extents after removal */
4007 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
));
4008 new_size
= ifp
->if_bytes
-
4009 (ext_diff
* sizeof(xfs_bmbt_rec_t
));
4010 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4012 if (new_size
== 0) {
4013 xfs_iext_destroy(ifp
);
4016 /* Move extents up in the list (if needed) */
4017 if (idx
+ ext_diff
< nextents
) {
4018 memmove(&ifp
->if_u1
.if_extents
[idx
],
4019 &ifp
->if_u1
.if_extents
[idx
+ ext_diff
],
4020 (nextents
- (idx
+ ext_diff
)) *
4021 sizeof(xfs_bmbt_rec_t
));
4023 memset(&ifp
->if_u1
.if_extents
[nextents
- ext_diff
],
4024 0, ext_diff
* sizeof(xfs_bmbt_rec_t
));
4026 * Reallocate the direct extent list. If the extents
4027 * will fit inside the inode then xfs_iext_realloc_direct
4028 * will switch from direct to inline extent allocation
4031 xfs_iext_realloc_direct(ifp
, new_size
);
4032 ifp
->if_bytes
= new_size
;
4036 * This is called when incore extents are being removed from the
4037 * indirection array and the extents being removed span multiple extent
4038 * buffers. The idx parameter contains the file extent index where we
4039 * want to begin removing extents, and the count parameter contains
4040 * how many extents need to be removed.
4042 * |-------| |-------|
4043 * | nex1 | | | nex1 - number of extents before idx
4044 * |-------| | count |
4045 * | | | | count - number of extents being removed at idx
4046 * | count | |-------|
4047 * | | | nex2 | nex2 - number of extents after idx + count
4048 * |-------| |-------|
4051 xfs_iext_remove_indirect(
4052 xfs_ifork_t
*ifp
, /* inode fork pointer */
4053 xfs_extnum_t idx
, /* index to begin removing extents */
4054 int count
) /* number of extents to remove */
4056 xfs_ext_irec_t
*erp
; /* indirection array pointer */
4057 int erp_idx
= 0; /* indirection array index */
4058 xfs_extnum_t ext_cnt
; /* extents left to remove */
4059 xfs_extnum_t ext_diff
; /* extents to remove in current list */
4060 xfs_extnum_t nex1
; /* number of extents before idx */
4061 xfs_extnum_t nex2
; /* extents after idx + count */
4062 int nlists
; /* entries in indirection array */
4063 int page_idx
= idx
; /* index in target extent list */
4065 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4066 erp
= xfs_iext_idx_to_irec(ifp
, &page_idx
, &erp_idx
, 0);
4067 ASSERT(erp
!= NULL
);
4068 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4072 nex2
= MAX((erp
->er_extcount
- (nex1
+ ext_cnt
)), 0);
4073 ext_diff
= MIN(ext_cnt
, (erp
->er_extcount
- nex1
));
4075 * Check for deletion of entire list;
4076 * xfs_iext_irec_remove() updates extent offsets.
4078 if (ext_diff
== erp
->er_extcount
) {
4079 xfs_iext_irec_remove(ifp
, erp_idx
);
4080 ext_cnt
-= ext_diff
;
4083 ASSERT(erp_idx
< ifp
->if_real_bytes
/
4085 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4092 /* Move extents up (if needed) */
4094 memmove(&erp
->er_extbuf
[nex1
],
4095 &erp
->er_extbuf
[nex1
+ ext_diff
],
4096 nex2
* sizeof(xfs_bmbt_rec_t
));
4098 /* Zero out rest of page */
4099 memset(&erp
->er_extbuf
[nex1
+ nex2
], 0, (XFS_IEXT_BUFSZ
-
4100 ((nex1
+ nex2
) * sizeof(xfs_bmbt_rec_t
))));
4101 /* Update remaining counters */
4102 erp
->er_extcount
-= ext_diff
;
4103 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, -ext_diff
);
4104 ext_cnt
-= ext_diff
;
4109 ifp
->if_bytes
-= count
* sizeof(xfs_bmbt_rec_t
);
4110 xfs_iext_irec_compact(ifp
);
4114 * Create, destroy, or resize a linear (direct) block of extents.
4117 xfs_iext_realloc_direct(
4118 xfs_ifork_t
*ifp
, /* inode fork pointer */
4119 int new_size
) /* new size of extents */
4121 int rnew_size
; /* real new size of extents */
4123 rnew_size
= new_size
;
4125 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
) ||
4126 ((new_size
>= 0) && (new_size
<= XFS_IEXT_BUFSZ
) &&
4127 (new_size
!= ifp
->if_real_bytes
)));
4129 /* Free extent records */
4130 if (new_size
== 0) {
4131 xfs_iext_destroy(ifp
);
4133 /* Resize direct extent list and zero any new bytes */
4134 else if (ifp
->if_real_bytes
) {
4135 /* Check if extents will fit inside the inode */
4136 if (new_size
<= XFS_INLINE_EXTS
* sizeof(xfs_bmbt_rec_t
)) {
4137 xfs_iext_direct_to_inline(ifp
, new_size
/
4138 (uint
)sizeof(xfs_bmbt_rec_t
));
4139 ifp
->if_bytes
= new_size
;
4142 if ((new_size
& (new_size
- 1)) != 0) {
4143 rnew_size
= xfs_iroundup(new_size
);
4145 if (rnew_size
!= ifp
->if_real_bytes
) {
4146 ifp
->if_u1
.if_extents
= (xfs_bmbt_rec_t
*)
4147 kmem_realloc(ifp
->if_u1
.if_extents
,
4152 if (rnew_size
> ifp
->if_real_bytes
) {
4153 memset(&ifp
->if_u1
.if_extents
[ifp
->if_bytes
/
4154 (uint
)sizeof(xfs_bmbt_rec_t
)], 0,
4155 rnew_size
- ifp
->if_real_bytes
);
4159 * Switch from the inline extent buffer to a direct
4160 * extent list. Be sure to include the inline extent
4161 * bytes in new_size.
4164 new_size
+= ifp
->if_bytes
;
4165 if ((new_size
& (new_size
- 1)) != 0) {
4166 rnew_size
= xfs_iroundup(new_size
);
4168 xfs_iext_inline_to_direct(ifp
, rnew_size
);
4170 ifp
->if_real_bytes
= rnew_size
;
4171 ifp
->if_bytes
= new_size
;
4175 * Switch from linear (direct) extent records to inline buffer.
4178 xfs_iext_direct_to_inline(
4179 xfs_ifork_t
*ifp
, /* inode fork pointer */
4180 xfs_extnum_t nextents
) /* number of extents in file */
4182 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
4183 ASSERT(nextents
<= XFS_INLINE_EXTS
);
4185 * The inline buffer was zeroed when we switched
4186 * from inline to direct extent allocation mode,
4187 * so we don't need to clear it here.
4189 memcpy(ifp
->if_u2
.if_inline_ext
, ifp
->if_u1
.if_extents
,
4190 nextents
* sizeof(xfs_bmbt_rec_t
));
4191 kmem_free(ifp
->if_u1
.if_extents
, ifp
->if_real_bytes
);
4192 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
4193 ifp
->if_real_bytes
= 0;
4197 * Switch from inline buffer to linear (direct) extent records.
4198 * new_size should already be rounded up to the next power of 2
4199 * by the caller (when appropriate), so use new_size as it is.
4200 * However, since new_size may be rounded up, we can't update
4201 * if_bytes here. It is the caller's responsibility to update
4202 * if_bytes upon return.
4205 xfs_iext_inline_to_direct(
4206 xfs_ifork_t
*ifp
, /* inode fork pointer */
4207 int new_size
) /* number of extents in file */
4209 ifp
->if_u1
.if_extents
= (xfs_bmbt_rec_t
*)
4210 kmem_alloc(new_size
, KM_SLEEP
);
4211 memset(ifp
->if_u1
.if_extents
, 0, new_size
);
4212 if (ifp
->if_bytes
) {
4213 memcpy(ifp
->if_u1
.if_extents
, ifp
->if_u2
.if_inline_ext
,
4215 memset(ifp
->if_u2
.if_inline_ext
, 0, XFS_INLINE_EXTS
*
4216 sizeof(xfs_bmbt_rec_t
));
4218 ifp
->if_real_bytes
= new_size
;
4222 * Resize an extent indirection array to new_size bytes.
4225 xfs_iext_realloc_indirect(
4226 xfs_ifork_t
*ifp
, /* inode fork pointer */
4227 int new_size
) /* new indirection array size */
4229 int nlists
; /* number of irec's (ex lists) */
4230 int size
; /* current indirection array size */
4232 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4233 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4234 size
= nlists
* sizeof(xfs_ext_irec_t
);
4235 ASSERT(ifp
->if_real_bytes
);
4236 ASSERT((new_size
>= 0) && (new_size
!= size
));
4237 if (new_size
== 0) {
4238 xfs_iext_destroy(ifp
);
4240 ifp
->if_u1
.if_ext_irec
= (xfs_ext_irec_t
*)
4241 kmem_realloc(ifp
->if_u1
.if_ext_irec
,
4242 new_size
, size
, KM_SLEEP
);
4247 * Switch from indirection array to linear (direct) extent allocations.
4250 xfs_iext_indirect_to_direct(
4251 xfs_ifork_t
*ifp
) /* inode fork pointer */
4253 xfs_bmbt_rec_t
*ep
; /* extent record pointer */
4254 xfs_extnum_t nextents
; /* number of extents in file */
4255 int size
; /* size of file extents */
4257 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4258 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4259 ASSERT(nextents
<= XFS_LINEAR_EXTS
);
4260 size
= nextents
* sizeof(xfs_bmbt_rec_t
);
4262 xfs_iext_irec_compact_full(ifp
);
4263 ASSERT(ifp
->if_real_bytes
== XFS_IEXT_BUFSZ
);
4265 ep
= ifp
->if_u1
.if_ext_irec
->er_extbuf
;
4266 kmem_free(ifp
->if_u1
.if_ext_irec
, sizeof(xfs_ext_irec_t
));
4267 ifp
->if_flags
&= ~XFS_IFEXTIREC
;
4268 ifp
->if_u1
.if_extents
= ep
;
4269 ifp
->if_bytes
= size
;
4270 if (nextents
< XFS_LINEAR_EXTS
) {
4271 xfs_iext_realloc_direct(ifp
, size
);
4276 * Free incore file extents.
4280 xfs_ifork_t
*ifp
) /* inode fork pointer */
4282 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
4286 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4287 for (erp_idx
= nlists
- 1; erp_idx
>= 0 ; erp_idx
--) {
4288 xfs_iext_irec_remove(ifp
, erp_idx
);
4290 ifp
->if_flags
&= ~XFS_IFEXTIREC
;
4291 } else if (ifp
->if_real_bytes
) {
4292 kmem_free(ifp
->if_u1
.if_extents
, ifp
->if_real_bytes
);
4293 } else if (ifp
->if_bytes
) {
4294 memset(ifp
->if_u2
.if_inline_ext
, 0, XFS_INLINE_EXTS
*
4295 sizeof(xfs_bmbt_rec_t
));
4297 ifp
->if_u1
.if_extents
= NULL
;
4298 ifp
->if_real_bytes
= 0;
4303 * Return a pointer to the extent record for file system block bno.
4305 xfs_bmbt_rec_t
* /* pointer to found extent record */
4306 xfs_iext_bno_to_ext(
4307 xfs_ifork_t
*ifp
, /* inode fork pointer */
4308 xfs_fileoff_t bno
, /* block number to search for */
4309 xfs_extnum_t
*idxp
) /* index of target extent */
4311 xfs_bmbt_rec_t
*base
; /* pointer to first extent */
4312 xfs_filblks_t blockcount
= 0; /* number of blocks in extent */
4313 xfs_bmbt_rec_t
*ep
= NULL
; /* pointer to target extent */
4314 xfs_ext_irec_t
*erp
= NULL
; /* indirection array pointer */
4315 int high
; /* upper boundary in search */
4316 xfs_extnum_t idx
= 0; /* index of target extent */
4317 int low
; /* lower boundary in search */
4318 xfs_extnum_t nextents
; /* number of file extents */
4319 xfs_fileoff_t startoff
= 0; /* start offset of extent */
4321 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4322 if (nextents
== 0) {
4327 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
4328 /* Find target extent list */
4330 erp
= xfs_iext_bno_to_irec(ifp
, bno
, &erp_idx
);
4331 base
= erp
->er_extbuf
;
4332 high
= erp
->er_extcount
- 1;
4334 base
= ifp
->if_u1
.if_extents
;
4335 high
= nextents
- 1;
4337 /* Binary search extent records */
4338 while (low
<= high
) {
4339 idx
= (low
+ high
) >> 1;
4341 startoff
= xfs_bmbt_get_startoff(ep
);
4342 blockcount
= xfs_bmbt_get_blockcount(ep
);
4343 if (bno
< startoff
) {
4345 } else if (bno
>= startoff
+ blockcount
) {
4348 /* Convert back to file-based extent index */
4349 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
4350 idx
+= erp
->er_extoff
;
4356 /* Convert back to file-based extent index */
4357 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
4358 idx
+= erp
->er_extoff
;
4360 if (bno
>= startoff
+ blockcount
) {
4361 if (++idx
== nextents
) {
4364 ep
= xfs_iext_get_ext(ifp
, idx
);
4372 * Return a pointer to the indirection array entry containing the
4373 * extent record for filesystem block bno. Store the index of the
4374 * target irec in *erp_idxp.
4376 xfs_ext_irec_t
* /* pointer to found extent record */
4377 xfs_iext_bno_to_irec(
4378 xfs_ifork_t
*ifp
, /* inode fork pointer */
4379 xfs_fileoff_t bno
, /* block number to search for */
4380 int *erp_idxp
) /* irec index of target ext list */
4382 xfs_ext_irec_t
*erp
= NULL
; /* indirection array pointer */
4383 xfs_ext_irec_t
*erp_next
; /* next indirection array entry */
4384 int erp_idx
; /* indirection array index */
4385 int nlists
; /* number of extent irec's (lists) */
4386 int high
; /* binary search upper limit */
4387 int low
; /* binary search lower limit */
4389 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4390 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4394 while (low
<= high
) {
4395 erp_idx
= (low
+ high
) >> 1;
4396 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4397 erp_next
= erp_idx
< nlists
- 1 ? erp
+ 1 : NULL
;
4398 if (bno
< xfs_bmbt_get_startoff(erp
->er_extbuf
)) {
4400 } else if (erp_next
&& bno
>=
4401 xfs_bmbt_get_startoff(erp_next
->er_extbuf
)) {
4407 *erp_idxp
= erp_idx
;
4412 * Return a pointer to the indirection array entry containing the
4413 * extent record at file extent index *idxp. Store the index of the
4414 * target irec in *erp_idxp and store the page index of the target
4415 * extent record in *idxp.
4418 xfs_iext_idx_to_irec(
4419 xfs_ifork_t
*ifp
, /* inode fork pointer */
4420 xfs_extnum_t
*idxp
, /* extent index (file -> page) */
4421 int *erp_idxp
, /* pointer to target irec */
4422 int realloc
) /* new bytes were just added */
4424 xfs_ext_irec_t
*prev
; /* pointer to previous irec */
4425 xfs_ext_irec_t
*erp
= NULL
; /* pointer to current irec */
4426 int erp_idx
; /* indirection array index */
4427 int nlists
; /* number of irec's (ex lists) */
4428 int high
; /* binary search upper limit */
4429 int low
; /* binary search lower limit */
4430 xfs_extnum_t page_idx
= *idxp
; /* extent index in target list */
4432 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4433 ASSERT(page_idx
>= 0 && page_idx
<=
4434 ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
));
4435 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4440 /* Binary search extent irec's */
4441 while (low
<= high
) {
4442 erp_idx
= (low
+ high
) >> 1;
4443 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4444 prev
= erp_idx
> 0 ? erp
- 1 : NULL
;
4445 if (page_idx
< erp
->er_extoff
|| (page_idx
== erp
->er_extoff
&&
4446 realloc
&& prev
&& prev
->er_extcount
< XFS_LINEAR_EXTS
)) {
4448 } else if (page_idx
> erp
->er_extoff
+ erp
->er_extcount
||
4449 (page_idx
== erp
->er_extoff
+ erp
->er_extcount
&&
4452 } else if (page_idx
== erp
->er_extoff
+ erp
->er_extcount
&&
4453 erp
->er_extcount
== XFS_LINEAR_EXTS
) {
4457 erp
= erp_idx
< nlists
? erp
+ 1 : NULL
;
4460 page_idx
-= erp
->er_extoff
;
4465 *erp_idxp
= erp_idx
;
4470 * Allocate and initialize an indirection array once the space needed
4471 * for incore extents increases above XFS_IEXT_BUFSZ.
4475 xfs_ifork_t
*ifp
) /* inode fork pointer */
4477 xfs_ext_irec_t
*erp
; /* indirection array pointer */
4478 xfs_extnum_t nextents
; /* number of extents in file */
4480 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
));
4481 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4482 ASSERT(nextents
<= XFS_LINEAR_EXTS
);
4484 erp
= (xfs_ext_irec_t
*)
4485 kmem_alloc(sizeof(xfs_ext_irec_t
), KM_SLEEP
);
4487 if (nextents
== 0) {
4488 ifp
->if_u1
.if_extents
= (xfs_bmbt_rec_t
*)
4489 kmem_alloc(XFS_IEXT_BUFSZ
, KM_SLEEP
);
4490 } else if (!ifp
->if_real_bytes
) {
4491 xfs_iext_inline_to_direct(ifp
, XFS_IEXT_BUFSZ
);
4492 } else if (ifp
->if_real_bytes
< XFS_IEXT_BUFSZ
) {
4493 xfs_iext_realloc_direct(ifp
, XFS_IEXT_BUFSZ
);
4495 erp
->er_extbuf
= ifp
->if_u1
.if_extents
;
4496 erp
->er_extcount
= nextents
;
4499 ifp
->if_flags
|= XFS_IFEXTIREC
;
4500 ifp
->if_real_bytes
= XFS_IEXT_BUFSZ
;
4501 ifp
->if_bytes
= nextents
* sizeof(xfs_bmbt_rec_t
);
4502 ifp
->if_u1
.if_ext_irec
= erp
;
4508 * Allocate and initialize a new entry in the indirection array.
4512 xfs_ifork_t
*ifp
, /* inode fork pointer */
4513 int erp_idx
) /* index for new irec */
4515 xfs_ext_irec_t
*erp
; /* indirection array pointer */
4516 int i
; /* loop counter */
4517 int nlists
; /* number of irec's (ex lists) */
4519 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4520 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4522 /* Resize indirection array */
4523 xfs_iext_realloc_indirect(ifp
, ++nlists
*
4524 sizeof(xfs_ext_irec_t
));
4526 * Move records down in the array so the
4527 * new page can use erp_idx.
4529 erp
= ifp
->if_u1
.if_ext_irec
;
4530 for (i
= nlists
- 1; i
> erp_idx
; i
--) {
4531 memmove(&erp
[i
], &erp
[i
-1], sizeof(xfs_ext_irec_t
));
4533 ASSERT(i
== erp_idx
);
4535 /* Initialize new extent record */
4536 erp
= ifp
->if_u1
.if_ext_irec
;
4537 erp
[erp_idx
].er_extbuf
= (xfs_bmbt_rec_t
*)
4538 kmem_alloc(XFS_IEXT_BUFSZ
, KM_SLEEP
);
4539 ifp
->if_real_bytes
= nlists
* XFS_IEXT_BUFSZ
;
4540 memset(erp
[erp_idx
].er_extbuf
, 0, XFS_IEXT_BUFSZ
);
4541 erp
[erp_idx
].er_extcount
= 0;
4542 erp
[erp_idx
].er_extoff
= erp_idx
> 0 ?
4543 erp
[erp_idx
-1].er_extoff
+ erp
[erp_idx
-1].er_extcount
: 0;
4544 return (&erp
[erp_idx
]);
4548 * Remove a record from the indirection array.
4551 xfs_iext_irec_remove(
4552 xfs_ifork_t
*ifp
, /* inode fork pointer */
4553 int erp_idx
) /* irec index to remove */
4555 xfs_ext_irec_t
*erp
; /* indirection array pointer */
4556 int i
; /* loop counter */
4557 int nlists
; /* number of irec's (ex lists) */
4559 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4560 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4561 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4562 if (erp
->er_extbuf
) {
4563 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1,
4565 kmem_free(erp
->er_extbuf
, XFS_IEXT_BUFSZ
);
4567 /* Compact extent records */
4568 erp
= ifp
->if_u1
.if_ext_irec
;
4569 for (i
= erp_idx
; i
< nlists
- 1; i
++) {
4570 memmove(&erp
[i
], &erp
[i
+1], sizeof(xfs_ext_irec_t
));
4573 * Manually free the last extent record from the indirection
4574 * array. A call to xfs_iext_realloc_indirect() with a size
4575 * of zero would result in a call to xfs_iext_destroy() which
4576 * would in turn call this function again, creating a nasty
4580 xfs_iext_realloc_indirect(ifp
,
4581 nlists
* sizeof(xfs_ext_irec_t
));
4583 kmem_free(ifp
->if_u1
.if_ext_irec
,
4584 sizeof(xfs_ext_irec_t
));
4586 ifp
->if_real_bytes
= nlists
* XFS_IEXT_BUFSZ
;
4590 * This is called to clean up large amounts of unused memory allocated
4591 * by the indirection array. Before compacting anything though, verify
4592 * that the indirection array is still needed and switch back to the
4593 * linear extent list (or even the inline buffer) if possible. The
4594 * compaction policy is as follows:
4596 * Full Compaction: Extents fit into a single page (or inline buffer)
4597 * Full Compaction: Extents occupy less than 10% of allocated space
4598 * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
4599 * No Compaction: Extents occupy at least 50% of allocated space
4602 xfs_iext_irec_compact(
4603 xfs_ifork_t
*ifp
) /* inode fork pointer */
4605 xfs_extnum_t nextents
; /* number of extents in file */
4606 int nlists
; /* number of irec's (ex lists) */
4608 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4609 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4610 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4612 if (nextents
== 0) {
4613 xfs_iext_destroy(ifp
);
4614 } else if (nextents
<= XFS_INLINE_EXTS
) {
4615 xfs_iext_indirect_to_direct(ifp
);
4616 xfs_iext_direct_to_inline(ifp
, nextents
);
4617 } else if (nextents
<= XFS_LINEAR_EXTS
) {
4618 xfs_iext_indirect_to_direct(ifp
);
4619 } else if (nextents
< (nlists
* XFS_LINEAR_EXTS
) >> 3) {
4620 xfs_iext_irec_compact_full(ifp
);
4621 } else if (nextents
< (nlists
* XFS_LINEAR_EXTS
) >> 1) {
4622 xfs_iext_irec_compact_pages(ifp
);
4627 * Combine extents from neighboring extent pages.
4630 xfs_iext_irec_compact_pages(
4631 xfs_ifork_t
*ifp
) /* inode fork pointer */
4633 xfs_ext_irec_t
*erp
, *erp_next
;/* pointers to irec entries */
4634 int erp_idx
= 0; /* indirection array index */
4635 int nlists
; /* number of irec's (ex lists) */
4637 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4638 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4639 while (erp_idx
< nlists
- 1) {
4640 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4642 if (erp_next
->er_extcount
<=
4643 (XFS_LINEAR_EXTS
- erp
->er_extcount
)) {
4644 memmove(&erp
->er_extbuf
[erp
->er_extcount
],
4645 erp_next
->er_extbuf
, erp_next
->er_extcount
*
4646 sizeof(xfs_bmbt_rec_t
));
4647 erp
->er_extcount
+= erp_next
->er_extcount
;
4649 * Free page before removing extent record
4650 * so er_extoffs don't get modified in
4651 * xfs_iext_irec_remove.
4653 kmem_free(erp_next
->er_extbuf
, XFS_IEXT_BUFSZ
);
4654 erp_next
->er_extbuf
= NULL
;
4655 xfs_iext_irec_remove(ifp
, erp_idx
+ 1);
4656 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4664 * Fully compact the extent records managed by the indirection array.
4667 xfs_iext_irec_compact_full(
4668 xfs_ifork_t
*ifp
) /* inode fork pointer */
4670 xfs_bmbt_rec_t
*ep
, *ep_next
; /* extent record pointers */
4671 xfs_ext_irec_t
*erp
, *erp_next
; /* extent irec pointers */
4672 int erp_idx
= 0; /* extent irec index */
4673 int ext_avail
; /* empty entries in ex list */
4674 int ext_diff
; /* number of exts to add */
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 erp
= ifp
->if_u1
.if_ext_irec
;
4680 ep
= &erp
->er_extbuf
[erp
->er_extcount
];
4682 ep_next
= erp_next
->er_extbuf
;
4683 while (erp_idx
< nlists
- 1) {
4684 ext_avail
= XFS_LINEAR_EXTS
- erp
->er_extcount
;
4685 ext_diff
= MIN(ext_avail
, erp_next
->er_extcount
);
4686 memcpy(ep
, ep_next
, ext_diff
* sizeof(xfs_bmbt_rec_t
));
4687 erp
->er_extcount
+= ext_diff
;
4688 erp_next
->er_extcount
-= ext_diff
;
4689 /* Remove next page */
4690 if (erp_next
->er_extcount
== 0) {
4692 * Free page before removing extent record
4693 * so er_extoffs don't get modified in
4694 * xfs_iext_irec_remove.
4696 kmem_free(erp_next
->er_extbuf
,
4697 erp_next
->er_extcount
* sizeof(xfs_bmbt_rec_t
));
4698 erp_next
->er_extbuf
= NULL
;
4699 xfs_iext_irec_remove(ifp
, erp_idx
+ 1);
4700 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4701 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4702 /* Update next page */
4704 /* Move rest of page up to become next new page */
4705 memmove(erp_next
->er_extbuf
, ep_next
,
4706 erp_next
->er_extcount
* sizeof(xfs_bmbt_rec_t
));
4707 ep_next
= erp_next
->er_extbuf
;
4708 memset(&ep_next
[erp_next
->er_extcount
], 0,
4709 (XFS_LINEAR_EXTS
- erp_next
->er_extcount
) *
4710 sizeof(xfs_bmbt_rec_t
));
4712 if (erp
->er_extcount
== XFS_LINEAR_EXTS
) {
4714 if (erp_idx
< nlists
)
4715 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4719 ep
= &erp
->er_extbuf
[erp
->er_extcount
];
4721 ep_next
= erp_next
->er_extbuf
;
4726 * This is called to update the er_extoff field in the indirection
4727 * array when extents have been added or removed from one of the
4728 * extent lists. erp_idx contains the irec index to begin updating
4729 * at and ext_diff contains the number of extents that were added
4733 xfs_iext_irec_update_extoffs(
4734 xfs_ifork_t
*ifp
, /* inode fork pointer */
4735 int erp_idx
, /* irec index to update */
4736 int ext_diff
) /* number of new extents */
4738 int i
; /* loop counter */
4739 int nlists
; /* number of irec's (ex lists */
4741 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4742 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4743 for (i
= erp_idx
; i
< nlists
; i
++) {
4744 ifp
->if_u1
.if_ext_irec
[i
].er_extoff
+= ext_diff
;