2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
25 #include "xfs_trans.h"
26 #include "xfs_trans_priv.h"
30 #include "xfs_dmapi.h"
31 #include "xfs_mount.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_ialloc_btree.h"
35 #include "xfs_dir2_sf.h"
36 #include "xfs_attr_sf.h"
37 #include "xfs_dinode.h"
38 #include "xfs_inode.h"
39 #include "xfs_buf_item.h"
40 #include "xfs_inode_item.h"
41 #include "xfs_btree.h"
42 #include "xfs_alloc.h"
43 #include "xfs_ialloc.h"
46 #include "xfs_error.h"
47 #include "xfs_utils.h"
48 #include "xfs_dir2_trace.h"
49 #include "xfs_quota.h"
51 #include "xfs_filestream.h"
52 #include "xfs_vnodeops.h"
54 kmem_zone_t
*xfs_ifork_zone
;
55 kmem_zone_t
*xfs_inode_zone
;
56 kmem_zone_t
*xfs_icluster_zone
;
59 * Used in xfs_itruncate(). This is the maximum number of extents
60 * freed from a file in a single transaction.
62 #define XFS_ITRUNC_MAX_EXTENTS 2
64 STATIC
int xfs_iflush_int(xfs_inode_t
*, xfs_buf_t
*);
65 STATIC
int xfs_iformat_local(xfs_inode_t
*, xfs_dinode_t
*, int, int);
66 STATIC
int xfs_iformat_extents(xfs_inode_t
*, xfs_dinode_t
*, int);
67 STATIC
int xfs_iformat_btree(xfs_inode_t
*, xfs_dinode_t
*, int);
71 * Make sure that the extents in the given memory buffer
81 xfs_bmbt_rec_host_t rec
;
84 for (i
= 0; i
< nrecs
; i
++) {
85 xfs_bmbt_rec_host_t
*ep
= xfs_iext_get_ext(ifp
, i
);
86 rec
.l0
= get_unaligned(&ep
->l0
);
87 rec
.l1
= get_unaligned(&ep
->l1
);
88 xfs_bmbt_get_all(&rec
, &irec
);
89 if (fmt
== XFS_EXTFMT_NOSTATE
)
90 ASSERT(irec
.br_state
== XFS_EXT_NORM
);
94 #define xfs_validate_extents(ifp, nrecs, fmt)
98 * Check that none of the inode's in the buffer have a next
99 * unlinked field of 0.
111 j
= mp
->m_inode_cluster_size
>> mp
->m_sb
.sb_inodelog
;
113 for (i
= 0; i
< j
; i
++) {
114 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
,
115 i
* mp
->m_sb
.sb_inodesize
);
116 if (!dip
->di_next_unlinked
) {
117 xfs_fs_cmn_err(CE_ALERT
, mp
,
118 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
120 ASSERT(dip
->di_next_unlinked
);
127 * This routine is called to map an inode number within a file
128 * system to the buffer containing the on-disk version of the
129 * inode. It returns a pointer to the buffer containing the
130 * on-disk inode in the bpp parameter, and in the dip parameter
131 * it returns a pointer to the on-disk inode within that buffer.
133 * If a non-zero error is returned, then the contents of bpp and
134 * dipp are undefined.
136 * Use xfs_imap() to determine the size and location of the
137 * buffer to read from disk.
155 * Call the space management code to find the location of the
159 error
= xfs_imap(mp
, tp
, ino
, &imap
, XFS_IMAP_LOOKUP
);
162 "xfs_inotobp: xfs_imap() returned an "
163 "error %d on %s. Returning error.", error
, mp
->m_fsname
);
168 * If the inode number maps to a block outside the bounds of the
169 * file system then return NULL rather than calling read_buf
170 * and panicing when we get an error from the driver.
172 if ((imap
.im_blkno
+ imap
.im_len
) >
173 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
)) {
175 "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds "
176 "of the file system %s. Returning EINVAL.",
177 (unsigned long long)imap
.im_blkno
,
178 imap
.im_len
, mp
->m_fsname
);
179 return XFS_ERROR(EINVAL
);
183 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
184 * default to just a read_buf() call.
186 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, imap
.im_blkno
,
187 (int)imap
.im_len
, XFS_BUF_LOCK
, &bp
);
191 "xfs_inotobp: xfs_trans_read_buf() returned an "
192 "error %d on %s. Returning error.", error
, mp
->m_fsname
);
195 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, 0);
197 be16_to_cpu(dip
->di_core
.di_magic
) == XFS_DINODE_MAGIC
&&
198 XFS_DINODE_GOOD_VERSION(dip
->di_core
.di_version
);
199 if (unlikely(XFS_TEST_ERROR(!di_ok
, mp
, XFS_ERRTAG_ITOBP_INOTOBP
,
200 XFS_RANDOM_ITOBP_INOTOBP
))) {
201 XFS_CORRUPTION_ERROR("xfs_inotobp", XFS_ERRLEVEL_LOW
, mp
, dip
);
202 xfs_trans_brelse(tp
, bp
);
204 "xfs_inotobp: XFS_TEST_ERROR() returned an "
205 "error on %s. Returning EFSCORRUPTED.", mp
->m_fsname
);
206 return XFS_ERROR(EFSCORRUPTED
);
209 xfs_inobp_check(mp
, bp
);
212 * Set *dipp to point to the on-disk inode in the buffer.
214 *dipp
= (xfs_dinode_t
*)xfs_buf_offset(bp
, imap
.im_boffset
);
216 *offset
= imap
.im_boffset
;
222 * This routine is called to map an inode to the buffer containing
223 * the on-disk version of the inode. It returns a pointer to the
224 * buffer containing the on-disk inode in the bpp parameter, and in
225 * the dip parameter it returns a pointer to the on-disk inode within
228 * If a non-zero error is returned, then the contents of bpp and
229 * dipp are undefined.
231 * If the inode is new and has not yet been initialized, use xfs_imap()
232 * to determine the size and location of the buffer to read from disk.
233 * If the inode has already been mapped to its buffer and read in once,
234 * then use the mapping information stored in the inode rather than
235 * calling xfs_imap(). This allows us to avoid the overhead of looking
236 * at the inode btree for small block file systems (see xfs_dilocate()).
237 * We can tell whether the inode has been mapped in before by comparing
238 * its disk block address to 0. Only uninitialized inodes will have
239 * 0 for the disk block address.
257 if (ip
->i_blkno
== (xfs_daddr_t
)0) {
259 * Call the space management code to find the location of the
263 if ((error
= xfs_imap(mp
, tp
, ip
->i_ino
, &imap
,
264 XFS_IMAP_LOOKUP
| imap_flags
)))
268 * If the inode number maps to a block outside the bounds
269 * of the file system then return NULL rather than calling
270 * read_buf and panicing when we get an error from the
273 if ((imap
.im_blkno
+ imap
.im_len
) >
274 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
)) {
276 xfs_fs_cmn_err(CE_ALERT
, mp
, "xfs_itobp: "
277 "(imap.im_blkno (0x%llx) "
278 "+ imap.im_len (0x%llx)) > "
279 " XFS_FSB_TO_BB(mp, "
280 "mp->m_sb.sb_dblocks) (0x%llx)",
281 (unsigned long long) imap
.im_blkno
,
282 (unsigned long long) imap
.im_len
,
283 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
));
285 return XFS_ERROR(EINVAL
);
289 * Fill in the fields in the inode that will be used to
290 * map the inode to its buffer from now on.
292 ip
->i_blkno
= imap
.im_blkno
;
293 ip
->i_len
= imap
.im_len
;
294 ip
->i_boffset
= imap
.im_boffset
;
297 * We've already mapped the inode once, so just use the
298 * mapping that we saved the first time.
300 imap
.im_blkno
= ip
->i_blkno
;
301 imap
.im_len
= ip
->i_len
;
302 imap
.im_boffset
= ip
->i_boffset
;
304 ASSERT(bno
== 0 || bno
== imap
.im_blkno
);
307 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
308 * default to just a read_buf() call.
310 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, imap
.im_blkno
,
311 (int)imap
.im_len
, XFS_BUF_LOCK
, &bp
);
314 xfs_fs_cmn_err(CE_ALERT
, mp
, "xfs_itobp: "
315 "xfs_trans_read_buf() returned error %d, "
316 "imap.im_blkno 0x%llx, imap.im_len 0x%llx",
317 error
, (unsigned long long) imap
.im_blkno
,
318 (unsigned long long) imap
.im_len
);
324 * Validate the magic number and version of every inode in the buffer
325 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
326 * No validation is done here in userspace (xfs_repair).
328 #if !defined(__KERNEL__)
331 ni
= BBTOB(imap
.im_len
) >> mp
->m_sb
.sb_inodelog
;
332 #else /* usual case */
336 for (i
= 0; i
< ni
; i
++) {
340 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
,
341 (i
<< mp
->m_sb
.sb_inodelog
));
342 di_ok
= be16_to_cpu(dip
->di_core
.di_magic
) == XFS_DINODE_MAGIC
&&
343 XFS_DINODE_GOOD_VERSION(dip
->di_core
.di_version
);
344 if (unlikely(XFS_TEST_ERROR(!di_ok
, mp
,
345 XFS_ERRTAG_ITOBP_INOTOBP
,
346 XFS_RANDOM_ITOBP_INOTOBP
))) {
347 if (imap_flags
& XFS_IMAP_BULKSTAT
) {
348 xfs_trans_brelse(tp
, bp
);
349 return XFS_ERROR(EINVAL
);
353 "Device %s - bad inode magic/vsn "
354 "daddr %lld #%d (magic=%x)",
355 XFS_BUFTARG_NAME(mp
->m_ddev_targp
),
356 (unsigned long long)imap
.im_blkno
, i
,
357 be16_to_cpu(dip
->di_core
.di_magic
));
359 XFS_CORRUPTION_ERROR("xfs_itobp", XFS_ERRLEVEL_HIGH
,
361 xfs_trans_brelse(tp
, bp
);
362 return XFS_ERROR(EFSCORRUPTED
);
366 xfs_inobp_check(mp
, bp
);
369 * Mark the buffer as an inode buffer now that it looks good
371 XFS_BUF_SET_VTYPE(bp
, B_FS_INO
);
374 * Set *dipp to point to the on-disk inode in the buffer.
376 *dipp
= (xfs_dinode_t
*)xfs_buf_offset(bp
, imap
.im_boffset
);
382 * Move inode type and inode format specific information from the
383 * on-disk inode to the in-core inode. For fifos, devs, and sockets
384 * this means set if_rdev to the proper value. For files, directories,
385 * and symlinks this means to bring in the in-line data or extent
386 * pointers. For a file in B-tree format, only the root is immediately
387 * brought in-core. The rest will be in-lined in if_extents when it
388 * is first referenced (see xfs_iread_extents()).
395 xfs_attr_shortform_t
*atp
;
399 ip
->i_df
.if_ext_max
=
400 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
403 if (unlikely(be32_to_cpu(dip
->di_core
.di_nextents
) +
404 be16_to_cpu(dip
->di_core
.di_anextents
) >
405 be64_to_cpu(dip
->di_core
.di_nblocks
))) {
406 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
407 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
408 (unsigned long long)ip
->i_ino
,
409 (int)(be32_to_cpu(dip
->di_core
.di_nextents
) +
410 be16_to_cpu(dip
->di_core
.di_anextents
)),
412 be64_to_cpu(dip
->di_core
.di_nblocks
));
413 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW
,
415 return XFS_ERROR(EFSCORRUPTED
);
418 if (unlikely(dip
->di_core
.di_forkoff
> ip
->i_mount
->m_sb
.sb_inodesize
)) {
419 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
420 "corrupt dinode %Lu, forkoff = 0x%x.",
421 (unsigned long long)ip
->i_ino
,
422 dip
->di_core
.di_forkoff
);
423 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW
,
425 return XFS_ERROR(EFSCORRUPTED
);
428 switch (ip
->i_d
.di_mode
& S_IFMT
) {
433 if (unlikely(dip
->di_core
.di_format
!= XFS_DINODE_FMT_DEV
)) {
434 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW
,
436 return XFS_ERROR(EFSCORRUPTED
);
440 ip
->i_df
.if_u2
.if_rdev
= be32_to_cpu(dip
->di_u
.di_dev
);
446 switch (dip
->di_core
.di_format
) {
447 case XFS_DINODE_FMT_LOCAL
:
449 * no local regular files yet
451 if (unlikely((be16_to_cpu(dip
->di_core
.di_mode
) & S_IFMT
) == S_IFREG
)) {
452 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
454 "(local format for regular file).",
455 (unsigned long long) ip
->i_ino
);
456 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
459 return XFS_ERROR(EFSCORRUPTED
);
462 di_size
= be64_to_cpu(dip
->di_core
.di_size
);
463 if (unlikely(di_size
> XFS_DFORK_DSIZE(dip
, ip
->i_mount
))) {
464 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
466 "(bad size %Ld for local inode).",
467 (unsigned long long) ip
->i_ino
,
468 (long long) di_size
);
469 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
472 return XFS_ERROR(EFSCORRUPTED
);
476 error
= xfs_iformat_local(ip
, dip
, XFS_DATA_FORK
, size
);
478 case XFS_DINODE_FMT_EXTENTS
:
479 error
= xfs_iformat_extents(ip
, dip
, XFS_DATA_FORK
);
481 case XFS_DINODE_FMT_BTREE
:
482 error
= xfs_iformat_btree(ip
, dip
, XFS_DATA_FORK
);
485 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW
,
487 return XFS_ERROR(EFSCORRUPTED
);
492 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW
, ip
->i_mount
);
493 return XFS_ERROR(EFSCORRUPTED
);
498 if (!XFS_DFORK_Q(dip
))
500 ASSERT(ip
->i_afp
== NULL
);
501 ip
->i_afp
= kmem_zone_zalloc(xfs_ifork_zone
, KM_SLEEP
);
502 ip
->i_afp
->if_ext_max
=
503 XFS_IFORK_ASIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
504 switch (dip
->di_core
.di_aformat
) {
505 case XFS_DINODE_FMT_LOCAL
:
506 atp
= (xfs_attr_shortform_t
*)XFS_DFORK_APTR(dip
);
507 size
= be16_to_cpu(atp
->hdr
.totsize
);
508 error
= xfs_iformat_local(ip
, dip
, XFS_ATTR_FORK
, size
);
510 case XFS_DINODE_FMT_EXTENTS
:
511 error
= xfs_iformat_extents(ip
, dip
, XFS_ATTR_FORK
);
513 case XFS_DINODE_FMT_BTREE
:
514 error
= xfs_iformat_btree(ip
, dip
, XFS_ATTR_FORK
);
517 error
= XFS_ERROR(EFSCORRUPTED
);
521 kmem_zone_free(xfs_ifork_zone
, ip
->i_afp
);
523 xfs_idestroy_fork(ip
, XFS_DATA_FORK
);
529 * The file is in-lined in the on-disk inode.
530 * If it fits into if_inline_data, then copy
531 * it there, otherwise allocate a buffer for it
532 * and copy the data there. Either way, set
533 * if_data to point at the data.
534 * If we allocate a buffer for the data, make
535 * sure that its size is a multiple of 4 and
536 * record the real size in i_real_bytes.
549 * If the size is unreasonable, then something
550 * is wrong and we just bail out rather than crash in
551 * kmem_alloc() or memcpy() below.
553 if (unlikely(size
> XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
))) {
554 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
556 "(bad size %d for local fork, size = %d).",
557 (unsigned long long) ip
->i_ino
, size
,
558 XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
));
559 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW
,
561 return XFS_ERROR(EFSCORRUPTED
);
563 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
566 ifp
->if_u1
.if_data
= NULL
;
567 else if (size
<= sizeof(ifp
->if_u2
.if_inline_data
))
568 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
570 real_size
= roundup(size
, 4);
571 ifp
->if_u1
.if_data
= kmem_alloc(real_size
, KM_SLEEP
);
573 ifp
->if_bytes
= size
;
574 ifp
->if_real_bytes
= real_size
;
576 memcpy(ifp
->if_u1
.if_data
, XFS_DFORK_PTR(dip
, whichfork
), size
);
577 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
578 ifp
->if_flags
|= XFS_IFINLINE
;
583 * The file consists of a set of extents all
584 * of which fit into the on-disk inode.
585 * If there are few enough extents to fit into
586 * the if_inline_ext, then copy them there.
587 * Otherwise allocate a buffer for them and copy
588 * them into it. Either way, set if_extents
589 * to point at the extents.
603 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
604 nex
= XFS_DFORK_NEXTENTS(dip
, whichfork
);
605 size
= nex
* (uint
)sizeof(xfs_bmbt_rec_t
);
608 * If the number of extents is unreasonable, then something
609 * is wrong and we just bail out rather than crash in
610 * kmem_alloc() or memcpy() below.
612 if (unlikely(size
< 0 || size
> XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
))) {
613 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
614 "corrupt inode %Lu ((a)extents = %d).",
615 (unsigned long long) ip
->i_ino
, nex
);
616 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW
,
618 return XFS_ERROR(EFSCORRUPTED
);
621 ifp
->if_real_bytes
= 0;
623 ifp
->if_u1
.if_extents
= NULL
;
624 else if (nex
<= XFS_INLINE_EXTS
)
625 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
627 xfs_iext_add(ifp
, 0, nex
);
629 ifp
->if_bytes
= size
;
631 dp
= (xfs_bmbt_rec_t
*) XFS_DFORK_PTR(dip
, whichfork
);
632 xfs_validate_extents(ifp
, nex
, XFS_EXTFMT_INODE(ip
));
633 for (i
= 0; i
< nex
; i
++, dp
++) {
634 xfs_bmbt_rec_host_t
*ep
= xfs_iext_get_ext(ifp
, i
);
635 ep
->l0
= be64_to_cpu(get_unaligned(&dp
->l0
));
636 ep
->l1
= be64_to_cpu(get_unaligned(&dp
->l1
));
638 XFS_BMAP_TRACE_EXLIST(ip
, nex
, whichfork
);
639 if (whichfork
!= XFS_DATA_FORK
||
640 XFS_EXTFMT_INODE(ip
) == XFS_EXTFMT_NOSTATE
)
641 if (unlikely(xfs_check_nostate_extents(
643 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
646 return XFS_ERROR(EFSCORRUPTED
);
649 ifp
->if_flags
|= XFS_IFEXTENTS
;
654 * The file has too many extents to fit into
655 * the inode, so they are in B-tree format.
656 * Allocate a buffer for the root of the B-tree
657 * and copy the root into it. The i_extents
658 * field will remain NULL until all of the
659 * extents are read in (when they are needed).
667 xfs_bmdr_block_t
*dfp
;
673 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
674 dfp
= (xfs_bmdr_block_t
*)XFS_DFORK_PTR(dip
, whichfork
);
675 size
= XFS_BMAP_BROOT_SPACE(dfp
);
676 nrecs
= XFS_BMAP_BROOT_NUMRECS(dfp
);
679 * blow out if -- fork has less extents than can fit in
680 * fork (fork shouldn't be a btree format), root btree
681 * block has more records than can fit into the fork,
682 * or the number of extents is greater than the number of
685 if (unlikely(XFS_IFORK_NEXTENTS(ip
, whichfork
) <= ifp
->if_ext_max
686 || XFS_BMDR_SPACE_CALC(nrecs
) >
687 XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
)
688 || XFS_IFORK_NEXTENTS(ip
, whichfork
) > ip
->i_d
.di_nblocks
)) {
689 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
690 "corrupt inode %Lu (btree).",
691 (unsigned long long) ip
->i_ino
);
692 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW
,
694 return XFS_ERROR(EFSCORRUPTED
);
697 ifp
->if_broot_bytes
= size
;
698 ifp
->if_broot
= kmem_alloc(size
, KM_SLEEP
);
699 ASSERT(ifp
->if_broot
!= NULL
);
701 * Copy and convert from the on-disk structure
702 * to the in-memory structure.
704 xfs_bmdr_to_bmbt(dfp
, XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
),
705 ifp
->if_broot
, size
);
706 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
707 ifp
->if_flags
|= XFS_IFBROOT
;
713 xfs_dinode_from_disk(
715 xfs_dinode_core_t
*from
)
717 to
->di_magic
= be16_to_cpu(from
->di_magic
);
718 to
->di_mode
= be16_to_cpu(from
->di_mode
);
719 to
->di_version
= from
->di_version
;
720 to
->di_format
= from
->di_format
;
721 to
->di_onlink
= be16_to_cpu(from
->di_onlink
);
722 to
->di_uid
= be32_to_cpu(from
->di_uid
);
723 to
->di_gid
= be32_to_cpu(from
->di_gid
);
724 to
->di_nlink
= be32_to_cpu(from
->di_nlink
);
725 to
->di_projid
= be16_to_cpu(from
->di_projid
);
726 memcpy(to
->di_pad
, from
->di_pad
, sizeof(to
->di_pad
));
727 to
->di_flushiter
= be16_to_cpu(from
->di_flushiter
);
728 to
->di_atime
.t_sec
= be32_to_cpu(from
->di_atime
.t_sec
);
729 to
->di_atime
.t_nsec
= be32_to_cpu(from
->di_atime
.t_nsec
);
730 to
->di_mtime
.t_sec
= be32_to_cpu(from
->di_mtime
.t_sec
);
731 to
->di_mtime
.t_nsec
= be32_to_cpu(from
->di_mtime
.t_nsec
);
732 to
->di_ctime
.t_sec
= be32_to_cpu(from
->di_ctime
.t_sec
);
733 to
->di_ctime
.t_nsec
= be32_to_cpu(from
->di_ctime
.t_nsec
);
734 to
->di_size
= be64_to_cpu(from
->di_size
);
735 to
->di_nblocks
= be64_to_cpu(from
->di_nblocks
);
736 to
->di_extsize
= be32_to_cpu(from
->di_extsize
);
737 to
->di_nextents
= be32_to_cpu(from
->di_nextents
);
738 to
->di_anextents
= be16_to_cpu(from
->di_anextents
);
739 to
->di_forkoff
= from
->di_forkoff
;
740 to
->di_aformat
= from
->di_aformat
;
741 to
->di_dmevmask
= be32_to_cpu(from
->di_dmevmask
);
742 to
->di_dmstate
= be16_to_cpu(from
->di_dmstate
);
743 to
->di_flags
= be16_to_cpu(from
->di_flags
);
744 to
->di_gen
= be32_to_cpu(from
->di_gen
);
749 xfs_dinode_core_t
*to
,
750 xfs_icdinode_t
*from
)
752 to
->di_magic
= cpu_to_be16(from
->di_magic
);
753 to
->di_mode
= cpu_to_be16(from
->di_mode
);
754 to
->di_version
= from
->di_version
;
755 to
->di_format
= from
->di_format
;
756 to
->di_onlink
= cpu_to_be16(from
->di_onlink
);
757 to
->di_uid
= cpu_to_be32(from
->di_uid
);
758 to
->di_gid
= cpu_to_be32(from
->di_gid
);
759 to
->di_nlink
= cpu_to_be32(from
->di_nlink
);
760 to
->di_projid
= cpu_to_be16(from
->di_projid
);
761 memcpy(to
->di_pad
, from
->di_pad
, sizeof(to
->di_pad
));
762 to
->di_flushiter
= cpu_to_be16(from
->di_flushiter
);
763 to
->di_atime
.t_sec
= cpu_to_be32(from
->di_atime
.t_sec
);
764 to
->di_atime
.t_nsec
= cpu_to_be32(from
->di_atime
.t_nsec
);
765 to
->di_mtime
.t_sec
= cpu_to_be32(from
->di_mtime
.t_sec
);
766 to
->di_mtime
.t_nsec
= cpu_to_be32(from
->di_mtime
.t_nsec
);
767 to
->di_ctime
.t_sec
= cpu_to_be32(from
->di_ctime
.t_sec
);
768 to
->di_ctime
.t_nsec
= cpu_to_be32(from
->di_ctime
.t_nsec
);
769 to
->di_size
= cpu_to_be64(from
->di_size
);
770 to
->di_nblocks
= cpu_to_be64(from
->di_nblocks
);
771 to
->di_extsize
= cpu_to_be32(from
->di_extsize
);
772 to
->di_nextents
= cpu_to_be32(from
->di_nextents
);
773 to
->di_anextents
= cpu_to_be16(from
->di_anextents
);
774 to
->di_forkoff
= from
->di_forkoff
;
775 to
->di_aformat
= from
->di_aformat
;
776 to
->di_dmevmask
= cpu_to_be32(from
->di_dmevmask
);
777 to
->di_dmstate
= cpu_to_be16(from
->di_dmstate
);
778 to
->di_flags
= cpu_to_be16(from
->di_flags
);
779 to
->di_gen
= cpu_to_be32(from
->di_gen
);
788 if (di_flags
& XFS_DIFLAG_ANY
) {
789 if (di_flags
& XFS_DIFLAG_REALTIME
)
790 flags
|= XFS_XFLAG_REALTIME
;
791 if (di_flags
& XFS_DIFLAG_PREALLOC
)
792 flags
|= XFS_XFLAG_PREALLOC
;
793 if (di_flags
& XFS_DIFLAG_IMMUTABLE
)
794 flags
|= XFS_XFLAG_IMMUTABLE
;
795 if (di_flags
& XFS_DIFLAG_APPEND
)
796 flags
|= XFS_XFLAG_APPEND
;
797 if (di_flags
& XFS_DIFLAG_SYNC
)
798 flags
|= XFS_XFLAG_SYNC
;
799 if (di_flags
& XFS_DIFLAG_NOATIME
)
800 flags
|= XFS_XFLAG_NOATIME
;
801 if (di_flags
& XFS_DIFLAG_NODUMP
)
802 flags
|= XFS_XFLAG_NODUMP
;
803 if (di_flags
& XFS_DIFLAG_RTINHERIT
)
804 flags
|= XFS_XFLAG_RTINHERIT
;
805 if (di_flags
& XFS_DIFLAG_PROJINHERIT
)
806 flags
|= XFS_XFLAG_PROJINHERIT
;
807 if (di_flags
& XFS_DIFLAG_NOSYMLINKS
)
808 flags
|= XFS_XFLAG_NOSYMLINKS
;
809 if (di_flags
& XFS_DIFLAG_EXTSIZE
)
810 flags
|= XFS_XFLAG_EXTSIZE
;
811 if (di_flags
& XFS_DIFLAG_EXTSZINHERIT
)
812 flags
|= XFS_XFLAG_EXTSZINHERIT
;
813 if (di_flags
& XFS_DIFLAG_NODEFRAG
)
814 flags
|= XFS_XFLAG_NODEFRAG
;
815 if (di_flags
& XFS_DIFLAG_FILESTREAM
)
816 flags
|= XFS_XFLAG_FILESTREAM
;
826 xfs_icdinode_t
*dic
= &ip
->i_d
;
828 return _xfs_dic2xflags(dic
->di_flags
) |
829 (XFS_CFORK_Q(dic
) ? XFS_XFLAG_HASATTR
: 0);
834 xfs_dinode_core_t
*dic
)
836 return _xfs_dic2xflags(be16_to_cpu(dic
->di_flags
)) |
837 (XFS_CFORK_Q_DISK(dic
) ? XFS_XFLAG_HASATTR
: 0);
841 * Given a mount structure and an inode number, return a pointer
842 * to a newly allocated in-core inode corresponding to the given
845 * Initialize the inode's attributes and extent pointers if it
846 * 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
);
867 atomic_set(&ip
->i_iocount
, 0);
868 spin_lock_init(&ip
->i_flags_lock
);
871 * Get pointer's to the on-disk inode and the buffer containing it.
872 * If the inode number refers to a block outside the file system
873 * then xfs_itobp() will return NULL. In this case we should
874 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
875 * know that this is a new incore inode.
877 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &bp
, bno
, imap_flags
);
879 kmem_zone_free(xfs_inode_zone
, ip
);
884 * Initialize inode's trace buffers.
885 * Do this before xfs_iformat in case it adds entries.
887 #ifdef XFS_INODE_TRACE
888 ip
->i_trace
= ktrace_alloc(INODE_TRACE_SIZE
, KM_SLEEP
);
890 #ifdef XFS_BMAP_TRACE
891 ip
->i_xtrace
= ktrace_alloc(XFS_BMAP_KTRACE_SIZE
, KM_SLEEP
);
893 #ifdef XFS_BMBT_TRACE
894 ip
->i_btrace
= ktrace_alloc(XFS_BMBT_KTRACE_SIZE
, KM_SLEEP
);
897 ip
->i_rwtrace
= ktrace_alloc(XFS_RW_KTRACE_SIZE
, KM_SLEEP
);
899 #ifdef XFS_ILOCK_TRACE
900 ip
->i_lock_trace
= ktrace_alloc(XFS_ILOCK_KTRACE_SIZE
, KM_SLEEP
);
902 #ifdef XFS_DIR2_TRACE
903 ip
->i_dir_trace
= ktrace_alloc(XFS_DIR2_KTRACE_SIZE
, KM_SLEEP
);
907 * If we got something that isn't an inode it means someone
908 * (nfs or dmi) has a stale handle.
910 if (be16_to_cpu(dip
->di_core
.di_magic
) != XFS_DINODE_MAGIC
) {
911 kmem_zone_free(xfs_inode_zone
, ip
);
912 xfs_trans_brelse(tp
, bp
);
914 xfs_fs_cmn_err(CE_ALERT
, mp
, "xfs_iread: "
915 "dip->di_core.di_magic (0x%x) != "
916 "XFS_DINODE_MAGIC (0x%x)",
917 be16_to_cpu(dip
->di_core
.di_magic
),
920 return XFS_ERROR(EINVAL
);
924 * If the on-disk inode is already linked to a directory
925 * entry, copy all of the inode into the in-core inode.
926 * xfs_iformat() handles copying in the inode format
927 * specific information.
928 * Otherwise, just get the truly permanent information.
930 if (dip
->di_core
.di_mode
) {
931 xfs_dinode_from_disk(&ip
->i_d
, &dip
->di_core
);
932 error
= xfs_iformat(ip
, dip
);
934 kmem_zone_free(xfs_inode_zone
, ip
);
935 xfs_trans_brelse(tp
, bp
);
937 xfs_fs_cmn_err(CE_ALERT
, mp
, "xfs_iread: "
938 "xfs_iformat() returned error %d",
944 ip
->i_d
.di_magic
= be16_to_cpu(dip
->di_core
.di_magic
);
945 ip
->i_d
.di_version
= dip
->di_core
.di_version
;
946 ip
->i_d
.di_gen
= be32_to_cpu(dip
->di_core
.di_gen
);
947 ip
->i_d
.di_flushiter
= be16_to_cpu(dip
->di_core
.di_flushiter
);
949 * Make sure to pull in the mode here as well in
950 * case the inode is released without being used.
951 * This ensures that xfs_inactive() will see that
952 * the inode is already free and not try to mess
953 * with the uninitialized part of it.
957 * Initialize the per-fork minima and maxima for a new
958 * inode here. xfs_iformat will do it for old inodes.
960 ip
->i_df
.if_ext_max
=
961 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
964 INIT_LIST_HEAD(&ip
->i_reclaim
);
967 * The inode format changed when we moved the link count and
968 * made it 32 bits long. If this is an old format inode,
969 * convert it in memory to look like a new one. If it gets
970 * flushed to disk we will convert back before flushing or
971 * logging it. We zero out the new projid field and the old link
972 * count field. We'll handle clearing the pad field (the remains
973 * of the old uuid field) when we actually convert the inode to
974 * the new format. We don't change the version number so that we
975 * can distinguish this from a real new format inode.
977 if (ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
) {
978 ip
->i_d
.di_nlink
= ip
->i_d
.di_onlink
;
979 ip
->i_d
.di_onlink
= 0;
980 ip
->i_d
.di_projid
= 0;
983 ip
->i_delayed_blks
= 0;
984 ip
->i_size
= ip
->i_d
.di_size
;
987 * Mark the buffer containing the inode as something to keep
988 * around for a while. This helps to keep recently accessed
989 * meta-data in-core longer.
991 XFS_BUF_SET_REF(bp
, XFS_INO_REF
);
994 * Use xfs_trans_brelse() to release the buffer containing the
995 * on-disk inode, because it was acquired with xfs_trans_read_buf()
996 * in xfs_itobp() above. If tp is NULL, this is just a normal
997 * brelse(). If we're within a transaction, then xfs_trans_brelse()
998 * will only release the buffer if it is not dirty within the
999 * transaction. It will be OK to release the buffer in this case,
1000 * because inodes on disk are never destroyed and we will be
1001 * locking the new in-core inode before putting it in the hash
1002 * table where other processes can find it. Thus we don't have
1003 * to worry about the inode being changed just because we released
1006 xfs_trans_brelse(tp
, bp
);
1012 * Read in extents from a btree-format inode.
1013 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1023 xfs_extnum_t nextents
;
1026 if (unlikely(XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
)) {
1027 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW
,
1029 return XFS_ERROR(EFSCORRUPTED
);
1031 nextents
= XFS_IFORK_NEXTENTS(ip
, whichfork
);
1032 size
= nextents
* sizeof(xfs_bmbt_rec_t
);
1033 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1036 * We know that the size is valid (it's checked in iformat_btree)
1038 ifp
->if_lastex
= NULLEXTNUM
;
1039 ifp
->if_bytes
= ifp
->if_real_bytes
= 0;
1040 ifp
->if_flags
|= XFS_IFEXTENTS
;
1041 xfs_iext_add(ifp
, 0, nextents
);
1042 error
= xfs_bmap_read_extents(tp
, ip
, whichfork
);
1044 xfs_iext_destroy(ifp
);
1045 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
1048 xfs_validate_extents(ifp
, nextents
, XFS_EXTFMT_INODE(ip
));
1053 * Allocate an inode on disk and return a copy of its in-core version.
1054 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1055 * appropriately within the inode. The uid and gid for the inode are
1056 * set according to the contents of the given cred structure.
1058 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1059 * has a free inode available, call xfs_iget()
1060 * to obtain the in-core version of the allocated inode. Finally,
1061 * fill in the inode and log its initial contents. In this case,
1062 * ialloc_context would be set to NULL and call_again set to false.
1064 * If xfs_dialloc() does not have an available inode,
1065 * it will replenish its supply by doing an allocation. Since we can
1066 * only do one allocation within a transaction without deadlocks, we
1067 * must commit the current transaction before returning the inode itself.
1068 * In this case, therefore, we will set call_again to true and return.
1069 * The caller should then commit the current transaction, start a new
1070 * transaction, and call xfs_ialloc() again to actually get the inode.
1072 * To ensure that some other process does not grab the inode that
1073 * was allocated during the first call to xfs_ialloc(), this routine
1074 * also returns the [locked] bp pointing to the head of the freelist
1075 * as ialloc_context. The caller should hold this buffer across
1076 * the commit and pass it back into this routine on the second call.
1078 * If we are allocating quota inodes, we do not have a parent inode
1079 * to attach to or associate with (i.e. pip == NULL) because they
1080 * are not linked into the directory structure - they are attached
1081 * directly to the superblock - and so have no parent.
1093 xfs_buf_t
**ialloc_context
,
1094 boolean_t
*call_again
,
1104 * Call the space management code to pick
1105 * the on-disk inode to be allocated.
1107 error
= xfs_dialloc(tp
, pip
? pip
->i_ino
: 0, mode
, okalloc
,
1108 ialloc_context
, call_again
, &ino
);
1112 if (*call_again
|| ino
== NULLFSINO
) {
1116 ASSERT(*ialloc_context
== NULL
);
1119 * Get the in-core inode with the lock held exclusively.
1120 * This is because we're setting fields here we need
1121 * to prevent others from looking at until we're done.
1123 error
= xfs_trans_iget(tp
->t_mountp
, tp
, ino
,
1124 XFS_IGET_CREATE
, XFS_ILOCK_EXCL
, &ip
);
1131 ip
->i_d
.di_mode
= (__uint16_t
)mode
;
1132 ip
->i_d
.di_onlink
= 0;
1133 ip
->i_d
.di_nlink
= nlink
;
1134 ASSERT(ip
->i_d
.di_nlink
== nlink
);
1135 ip
->i_d
.di_uid
= current_fsuid(cr
);
1136 ip
->i_d
.di_gid
= current_fsgid(cr
);
1137 ip
->i_d
.di_projid
= prid
;
1138 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
1141 * If the superblock version is up to where we support new format
1142 * inodes and this is currently an old format inode, then change
1143 * the inode version number now. This way we only do the conversion
1144 * here rather than here and in the flush/logging code.
1146 if (XFS_SB_VERSION_HASNLINK(&tp
->t_mountp
->m_sb
) &&
1147 ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
) {
1148 ip
->i_d
.di_version
= XFS_DINODE_VERSION_2
;
1150 * We've already zeroed the old link count, the projid field,
1151 * and the pad field.
1156 * Project ids won't be stored on disk if we are using a version 1 inode.
1158 if ((prid
!= 0) && (ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
))
1159 xfs_bump_ino_vers2(tp
, ip
);
1161 if (pip
&& XFS_INHERIT_GID(pip
)) {
1162 ip
->i_d
.di_gid
= pip
->i_d
.di_gid
;
1163 if ((pip
->i_d
.di_mode
& S_ISGID
) && (mode
& S_IFMT
) == S_IFDIR
) {
1164 ip
->i_d
.di_mode
|= S_ISGID
;
1169 * If the group ID of the new file does not match the effective group
1170 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1171 * (and only if the irix_sgid_inherit compatibility variable is set).
1173 if ((irix_sgid_inherit
) &&
1174 (ip
->i_d
.di_mode
& S_ISGID
) &&
1175 (!in_group_p((gid_t
)ip
->i_d
.di_gid
))) {
1176 ip
->i_d
.di_mode
&= ~S_ISGID
;
1179 ip
->i_d
.di_size
= 0;
1181 ip
->i_d
.di_nextents
= 0;
1182 ASSERT(ip
->i_d
.di_nblocks
== 0);
1183 xfs_ichgtime(ip
, XFS_ICHGTIME_CHG
|XFS_ICHGTIME_ACC
|XFS_ICHGTIME_MOD
);
1185 * di_gen will have been taken care of in xfs_iread.
1187 ip
->i_d
.di_extsize
= 0;
1188 ip
->i_d
.di_dmevmask
= 0;
1189 ip
->i_d
.di_dmstate
= 0;
1190 ip
->i_d
.di_flags
= 0;
1191 flags
= XFS_ILOG_CORE
;
1192 switch (mode
& S_IFMT
) {
1197 ip
->i_d
.di_format
= XFS_DINODE_FMT_DEV
;
1198 ip
->i_df
.if_u2
.if_rdev
= rdev
;
1199 ip
->i_df
.if_flags
= 0;
1200 flags
|= XFS_ILOG_DEV
;
1203 if (pip
&& xfs_inode_is_filestream(pip
)) {
1204 error
= xfs_filestream_associate(pip
, ip
);
1208 xfs_iflags_set(ip
, XFS_IFILESTREAM
);
1212 if (pip
&& (pip
->i_d
.di_flags
& XFS_DIFLAG_ANY
)) {
1215 if ((mode
& S_IFMT
) == S_IFDIR
) {
1216 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
1217 di_flags
|= XFS_DIFLAG_RTINHERIT
;
1218 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
1219 di_flags
|= XFS_DIFLAG_EXTSZINHERIT
;
1220 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
1222 } else if ((mode
& S_IFMT
) == S_IFREG
) {
1223 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
1224 di_flags
|= XFS_DIFLAG_REALTIME
;
1225 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
1226 di_flags
|= XFS_DIFLAG_EXTSIZE
;
1227 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
1230 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOATIME
) &&
1231 xfs_inherit_noatime
)
1232 di_flags
|= XFS_DIFLAG_NOATIME
;
1233 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODUMP
) &&
1235 di_flags
|= XFS_DIFLAG_NODUMP
;
1236 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_SYNC
) &&
1238 di_flags
|= XFS_DIFLAG_SYNC
;
1239 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOSYMLINKS
) &&
1240 xfs_inherit_nosymlinks
)
1241 di_flags
|= XFS_DIFLAG_NOSYMLINKS
;
1242 if (pip
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
1243 di_flags
|= XFS_DIFLAG_PROJINHERIT
;
1244 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODEFRAG
) &&
1245 xfs_inherit_nodefrag
)
1246 di_flags
|= XFS_DIFLAG_NODEFRAG
;
1247 if (pip
->i_d
.di_flags
& XFS_DIFLAG_FILESTREAM
)
1248 di_flags
|= XFS_DIFLAG_FILESTREAM
;
1249 ip
->i_d
.di_flags
|= di_flags
;
1253 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
1254 ip
->i_df
.if_flags
= XFS_IFEXTENTS
;
1255 ip
->i_df
.if_bytes
= ip
->i_df
.if_real_bytes
= 0;
1256 ip
->i_df
.if_u1
.if_extents
= NULL
;
1262 * Attribute fork settings for new inode.
1264 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
1265 ip
->i_d
.di_anextents
= 0;
1268 * Log the new values stuffed into the inode.
1270 xfs_trans_log_inode(tp
, ip
, flags
);
1272 /* now that we have an i_mode we can setup inode ops and unlock */
1273 xfs_initialize_vnode(tp
->t_mountp
, vp
, ip
);
1280 * Check to make sure that there are no blocks allocated to the
1281 * file beyond the size of the file. We don't check this for
1282 * files with fixed size extents or real time extents, but we
1283 * at least do it for regular files.
1292 xfs_fileoff_t map_first
;
1294 xfs_bmbt_irec_t imaps
[2];
1296 if ((ip
->i_d
.di_mode
& S_IFMT
) != S_IFREG
)
1299 if (ip
->i_d
.di_flags
& (XFS_DIFLAG_REALTIME
| XFS_DIFLAG_EXTSIZE
))
1303 map_first
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)isize
);
1305 * The filesystem could be shutting down, so bmapi may return
1308 if (xfs_bmapi(NULL
, ip
, map_first
,
1310 (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
)) -
1312 XFS_BMAPI_ENTIRE
, NULL
, 0, imaps
, &nimaps
,
1315 ASSERT(nimaps
== 1);
1316 ASSERT(imaps
[0].br_startblock
== HOLESTARTBLOCK
);
1321 * Calculate the last possible buffered byte in a file. This must
1322 * include data that was buffered beyond the EOF by the write code.
1323 * This also needs to deal with overflowing the xfs_fsize_t type
1324 * which can happen for sizes near the limit.
1326 * We also need to take into account any blocks beyond the EOF. It
1327 * may be the case that they were buffered by a write which failed.
1328 * In that case the pages will still be in memory, but the inode size
1329 * will never have been updated.
1336 xfs_fsize_t last_byte
;
1337 xfs_fileoff_t last_block
;
1338 xfs_fileoff_t size_last_block
;
1341 ASSERT(ismrlocked(&(ip
->i_iolock
), MR_UPDATE
| MR_ACCESS
));
1345 * Only check for blocks beyond the EOF if the extents have
1346 * been read in. This eliminates the need for the inode lock,
1347 * and it also saves us from looking when it really isn't
1350 if (ip
->i_df
.if_flags
& XFS_IFEXTENTS
) {
1351 error
= xfs_bmap_last_offset(NULL
, ip
, &last_block
,
1359 size_last_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)ip
->i_size
);
1360 last_block
= XFS_FILEOFF_MAX(last_block
, size_last_block
);
1362 last_byte
= XFS_FSB_TO_B(mp
, last_block
);
1363 if (last_byte
< 0) {
1364 return XFS_MAXIOFFSET(mp
);
1366 last_byte
+= (1 << mp
->m_writeio_log
);
1367 if (last_byte
< 0) {
1368 return XFS_MAXIOFFSET(mp
);
1373 #if defined(XFS_RW_TRACE)
1379 xfs_fsize_t new_size
,
1380 xfs_off_t toss_start
,
1381 xfs_off_t toss_finish
)
1383 if (ip
->i_rwtrace
== NULL
) {
1387 ktrace_enter(ip
->i_rwtrace
,
1390 (void*)(unsigned long)((ip
->i_d
.di_size
>> 32) & 0xffffffff),
1391 (void*)(unsigned long)(ip
->i_d
.di_size
& 0xffffffff),
1392 (void*)((long)flag
),
1393 (void*)(unsigned long)((new_size
>> 32) & 0xffffffff),
1394 (void*)(unsigned long)(new_size
& 0xffffffff),
1395 (void*)(unsigned long)((toss_start
>> 32) & 0xffffffff),
1396 (void*)(unsigned long)(toss_start
& 0xffffffff),
1397 (void*)(unsigned long)((toss_finish
>> 32) & 0xffffffff),
1398 (void*)(unsigned long)(toss_finish
& 0xffffffff),
1399 (void*)(unsigned long)current_cpu(),
1400 (void*)(unsigned long)current_pid(),
1406 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1410 * Start the truncation of the file to new_size. The new size
1411 * must be smaller than the current size. This routine will
1412 * clear the buffer and page caches of file data in the removed
1413 * range, and xfs_itruncate_finish() will remove the underlying
1416 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1417 * must NOT have the inode lock held at all. This is because we're
1418 * calling into the buffer/page cache code and we can't hold the
1419 * inode lock when we do so.
1421 * We need to wait for any direct I/Os in flight to complete before we
1422 * proceed with the truncate. This is needed to prevent the extents
1423 * being read or written by the direct I/Os from being removed while the
1424 * I/O is in flight as there is no other method of synchronising
1425 * direct I/O with the truncate operation. Also, because we hold
1426 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1427 * started until the truncate completes and drops the lock. Essentially,
1428 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1429 * between direct I/Os and the truncate operation.
1431 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1432 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1433 * in the case that the caller is locking things out of order and
1434 * may not be able to call xfs_itruncate_finish() with the inode lock
1435 * held without dropping the I/O lock. If the caller must drop the
1436 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1437 * must be called again with all the same restrictions as the initial
1441 xfs_itruncate_start(
1444 xfs_fsize_t new_size
)
1446 xfs_fsize_t last_byte
;
1447 xfs_off_t toss_start
;
1452 ASSERT(ismrlocked(&ip
->i_iolock
, MR_UPDATE
) != 0);
1453 ASSERT((new_size
== 0) || (new_size
<= ip
->i_size
));
1454 ASSERT((flags
== XFS_ITRUNC_DEFINITE
) ||
1455 (flags
== XFS_ITRUNC_MAYBE
));
1460 /* wait for the completion of any pending DIOs */
1461 if (new_size
< ip
->i_size
)
1465 * Call toss_pages or flushinval_pages to get rid of pages
1466 * overlapping the region being removed. We have to use
1467 * the less efficient flushinval_pages in the case that the
1468 * caller may not be able to finish the truncate without
1469 * dropping the inode's I/O lock. Make sure
1470 * to catch any pages brought in by buffers overlapping
1471 * the EOF by searching out beyond the isize by our
1472 * block size. We round new_size up to a block boundary
1473 * so that we don't toss things on the same block as
1474 * new_size but before it.
1476 * Before calling toss_page or flushinval_pages, make sure to
1477 * call remapf() over the same region if the file is mapped.
1478 * This frees up mapped file references to the pages in the
1479 * given range and for the flushinval_pages case it ensures
1480 * that we get the latest mapped changes flushed out.
1482 toss_start
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)new_size
);
1483 toss_start
= XFS_FSB_TO_B(mp
, toss_start
);
1484 if (toss_start
< 0) {
1486 * The place to start tossing is beyond our maximum
1487 * file size, so there is no way that the data extended
1492 last_byte
= xfs_file_last_byte(ip
);
1493 xfs_itrunc_trace(XFS_ITRUNC_START
, ip
, flags
, new_size
, toss_start
,
1495 if (last_byte
> toss_start
) {
1496 if (flags
& XFS_ITRUNC_DEFINITE
) {
1497 xfs_tosspages(ip
, toss_start
,
1498 -1, FI_REMAPF_LOCKED
);
1500 error
= xfs_flushinval_pages(ip
, toss_start
,
1501 -1, FI_REMAPF_LOCKED
);
1506 if (new_size
== 0) {
1507 ASSERT(VN_CACHED(vp
) == 0);
1514 * Shrink the file to the given new_size. The new
1515 * size must be smaller than the current size.
1516 * This will free up the underlying blocks
1517 * in the removed range after a call to xfs_itruncate_start()
1518 * or xfs_atruncate_start().
1520 * The transaction passed to this routine must have made
1521 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1522 * This routine may commit the given transaction and
1523 * start new ones, so make sure everything involved in
1524 * the transaction is tidy before calling here.
1525 * Some transaction will be returned to the caller to be
1526 * committed. The incoming transaction must already include
1527 * the inode, and both inode locks must be held exclusively.
1528 * The inode must also be "held" within the transaction. On
1529 * return the inode will be "held" within the returned transaction.
1530 * This routine does NOT require any disk space to be reserved
1531 * for it within the transaction.
1533 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1534 * and it indicates the fork which is to be truncated. For the
1535 * attribute fork we only support truncation to size 0.
1537 * We use the sync parameter to indicate whether or not the first
1538 * transaction we perform might have to be synchronous. For the attr fork,
1539 * it needs to be so if the unlink of the inode is not yet known to be
1540 * permanent in the log. This keeps us from freeing and reusing the
1541 * blocks of the attribute fork before the unlink of the inode becomes
1544 * For the data fork, we normally have to run synchronously if we're
1545 * being called out of the inactive path or we're being called
1546 * out of the create path where we're truncating an existing file.
1547 * Either way, the truncate needs to be sync so blocks don't reappear
1548 * in the file with altered data in case of a crash. wsync filesystems
1549 * can run the first case async because anything that shrinks the inode
1550 * has to run sync so by the time we're called here from inactive, the
1551 * inode size is permanently set to 0.
1553 * Calls from the truncate path always need to be sync unless we're
1554 * in a wsync filesystem and the file has already been unlinked.
1556 * The caller is responsible for correctly setting the sync parameter.
1557 * It gets too hard for us to guess here which path we're being called
1558 * out of just based on inode state.
1561 xfs_itruncate_finish(
1564 xfs_fsize_t new_size
,
1568 xfs_fsblock_t first_block
;
1569 xfs_fileoff_t first_unmap_block
;
1570 xfs_fileoff_t last_block
;
1571 xfs_filblks_t unmap_len
=0;
1576 xfs_bmap_free_t free_list
;
1579 ASSERT(ismrlocked(&ip
->i_iolock
, MR_UPDATE
) != 0);
1580 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
) != 0);
1581 ASSERT((new_size
== 0) || (new_size
<= ip
->i_size
));
1582 ASSERT(*tp
!= NULL
);
1583 ASSERT((*tp
)->t_flags
& XFS_TRANS_PERM_LOG_RES
);
1584 ASSERT(ip
->i_transp
== *tp
);
1585 ASSERT(ip
->i_itemp
!= NULL
);
1586 ASSERT(ip
->i_itemp
->ili_flags
& XFS_ILI_HOLD
);
1590 mp
= (ntp
)->t_mountp
;
1591 ASSERT(! XFS_NOT_DQATTACHED(mp
, ip
));
1594 * We only support truncating the entire attribute fork.
1596 if (fork
== XFS_ATTR_FORK
) {
1599 first_unmap_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)new_size
);
1600 xfs_itrunc_trace(XFS_ITRUNC_FINISH1
, ip
, 0, new_size
, 0, 0);
1602 * The first thing we do is set the size to new_size permanently
1603 * on disk. This way we don't have to worry about anyone ever
1604 * being able to look at the data being freed even in the face
1605 * of a crash. What we're getting around here is the case where
1606 * we free a block, it is allocated to another file, it is written
1607 * to, and then we crash. If the new data gets written to the
1608 * file but the log buffers containing the free and reallocation
1609 * don't, then we'd end up with garbage in the blocks being freed.
1610 * As long as we make the new_size permanent before actually
1611 * freeing any blocks it doesn't matter if they get writtten to.
1613 * The callers must signal into us whether or not the size
1614 * setting here must be synchronous. There are a few cases
1615 * where it doesn't have to be synchronous. Those cases
1616 * occur if the file is unlinked and we know the unlink is
1617 * permanent or if the blocks being truncated are guaranteed
1618 * to be beyond the inode eof (regardless of the link count)
1619 * and the eof value is permanent. Both of these cases occur
1620 * only on wsync-mounted filesystems. In those cases, we're
1621 * guaranteed that no user will ever see the data in the blocks
1622 * that are being truncated so the truncate can run async.
1623 * In the free beyond eof case, the file may wind up with
1624 * more blocks allocated to it than it needs if we crash
1625 * and that won't get fixed until the next time the file
1626 * is re-opened and closed but that's ok as that shouldn't
1627 * be too many blocks.
1629 * However, we can't just make all wsync xactions run async
1630 * because there's one call out of the create path that needs
1631 * to run sync where it's truncating an existing file to size
1632 * 0 whose size is > 0.
1634 * It's probably possible to come up with a test in this
1635 * routine that would correctly distinguish all the above
1636 * cases from the values of the function parameters and the
1637 * inode state but for sanity's sake, I've decided to let the
1638 * layers above just tell us. It's simpler to correctly figure
1639 * out in the layer above exactly under what conditions we
1640 * can run async and I think it's easier for others read and
1641 * follow the logic in case something has to be changed.
1642 * cscope is your friend -- rcc.
1644 * The attribute fork is much simpler.
1646 * For the attribute fork we allow the caller to tell us whether
1647 * the unlink of the inode that led to this call is yet permanent
1648 * in the on disk log. If it is not and we will be freeing extents
1649 * in this inode then we make the first transaction synchronous
1650 * to make sure that the unlink is permanent by the time we free
1653 if (fork
== XFS_DATA_FORK
) {
1654 if (ip
->i_d
.di_nextents
> 0) {
1656 * If we are not changing the file size then do
1657 * not update the on-disk file size - we may be
1658 * called from xfs_inactive_free_eofblocks(). If we
1659 * update the on-disk file size and then the system
1660 * crashes before the contents of the file are
1661 * flushed to disk then the files may be full of
1662 * holes (ie NULL files bug).
1664 if (ip
->i_size
!= new_size
) {
1665 ip
->i_d
.di_size
= new_size
;
1666 ip
->i_size
= new_size
;
1667 xfs_trans_log_inode(ntp
, ip
, XFS_ILOG_CORE
);
1671 ASSERT(!(mp
->m_flags
& XFS_MOUNT_WSYNC
));
1672 if (ip
->i_d
.di_anextents
> 0)
1673 xfs_trans_set_sync(ntp
);
1675 ASSERT(fork
== XFS_DATA_FORK
||
1676 (fork
== XFS_ATTR_FORK
&&
1677 ((sync
&& !(mp
->m_flags
& XFS_MOUNT_WSYNC
)) ||
1678 (sync
== 0 && (mp
->m_flags
& XFS_MOUNT_WSYNC
)))));
1681 * Since it is possible for space to become allocated beyond
1682 * the end of the file (in a crash where the space is allocated
1683 * but the inode size is not yet updated), simply remove any
1684 * blocks which show up between the new EOF and the maximum
1685 * possible file size. If the first block to be removed is
1686 * beyond the maximum file size (ie it is the same as last_block),
1687 * then there is nothing to do.
1689 last_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
));
1690 ASSERT(first_unmap_block
<= last_block
);
1692 if (last_block
== first_unmap_block
) {
1695 unmap_len
= last_block
- first_unmap_block
+ 1;
1699 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1700 * will tell us whether it freed the entire range or
1701 * not. If this is a synchronous mount (wsync),
1702 * then we can tell bunmapi to keep all the
1703 * transactions asynchronous since the unlink
1704 * transaction that made this inode inactive has
1705 * already hit the disk. There's no danger of
1706 * the freed blocks being reused, there being a
1707 * crash, and the reused blocks suddenly reappearing
1708 * in this file with garbage in them once recovery
1711 XFS_BMAP_INIT(&free_list
, &first_block
);
1712 error
= xfs_bunmapi(ntp
, ip
,
1713 first_unmap_block
, unmap_len
,
1714 XFS_BMAPI_AFLAG(fork
) |
1715 (sync
? 0 : XFS_BMAPI_ASYNC
),
1716 XFS_ITRUNC_MAX_EXTENTS
,
1717 &first_block
, &free_list
,
1721 * If the bunmapi call encounters an error,
1722 * return to the caller where the transaction
1723 * can be properly aborted. We just need to
1724 * make sure we're not holding any resources
1725 * that we were not when we came in.
1727 xfs_bmap_cancel(&free_list
);
1732 * Duplicate the transaction that has the permanent
1733 * reservation and commit the old transaction.
1735 error
= xfs_bmap_finish(tp
, &free_list
, &committed
);
1739 * If the bmap finish call encounters an error,
1740 * return to the caller where the transaction
1741 * can be properly aborted. We just need to
1742 * make sure we're not holding any resources
1743 * that we were not when we came in.
1745 * Aborting from this point might lose some
1746 * blocks in the file system, but oh well.
1748 xfs_bmap_cancel(&free_list
);
1751 * If the passed in transaction committed
1752 * in xfs_bmap_finish(), then we want to
1753 * add the inode to this one before returning.
1754 * This keeps things simple for the higher
1755 * level code, because it always knows that
1756 * the inode is locked and held in the
1757 * transaction that returns to it whether
1758 * errors occur or not. We don't mark the
1759 * inode dirty so that this transaction can
1760 * be easily aborted if possible.
1762 xfs_trans_ijoin(ntp
, ip
,
1763 XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1764 xfs_trans_ihold(ntp
, ip
);
1771 * The first xact was committed,
1772 * so add the inode to the new one.
1773 * Mark it dirty so it will be logged
1774 * and moved forward in the log as
1775 * part of every commit.
1777 xfs_trans_ijoin(ntp
, ip
,
1778 XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1779 xfs_trans_ihold(ntp
, ip
);
1780 xfs_trans_log_inode(ntp
, ip
, XFS_ILOG_CORE
);
1782 ntp
= xfs_trans_dup(ntp
);
1783 (void) xfs_trans_commit(*tp
, 0);
1785 error
= xfs_trans_reserve(ntp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0,
1786 XFS_TRANS_PERM_LOG_RES
,
1787 XFS_ITRUNCATE_LOG_COUNT
);
1789 * Add the inode being truncated to the next chained
1792 xfs_trans_ijoin(ntp
, ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1793 xfs_trans_ihold(ntp
, ip
);
1798 * Only update the size in the case of the data fork, but
1799 * always re-log the inode so that our permanent transaction
1800 * can keep on rolling it forward in the log.
1802 if (fork
== XFS_DATA_FORK
) {
1803 xfs_isize_check(mp
, ip
, new_size
);
1805 * If we are not changing the file size then do
1806 * not update the on-disk file size - we may be
1807 * called from xfs_inactive_free_eofblocks(). If we
1808 * update the on-disk file size and then the system
1809 * crashes before the contents of the file are
1810 * flushed to disk then the files may be full of
1811 * holes (ie NULL files bug).
1813 if (ip
->i_size
!= new_size
) {
1814 ip
->i_d
.di_size
= new_size
;
1815 ip
->i_size
= new_size
;
1818 xfs_trans_log_inode(ntp
, ip
, XFS_ILOG_CORE
);
1819 ASSERT((new_size
!= 0) ||
1820 (fork
== XFS_ATTR_FORK
) ||
1821 (ip
->i_delayed_blks
== 0));
1822 ASSERT((new_size
!= 0) ||
1823 (fork
== XFS_ATTR_FORK
) ||
1824 (ip
->i_d
.di_nextents
== 0));
1825 xfs_itrunc_trace(XFS_ITRUNC_FINISH2
, ip
, 0, new_size
, 0, 0);
1833 * Do the first part of growing a file: zero any data in the last
1834 * block that is beyond the old EOF. We need to do this before
1835 * the inode is joined to the transaction to modify the i_size.
1836 * That way we can drop the inode lock and call into the buffer
1837 * cache to get the buffer mapping the EOF.
1842 xfs_fsize_t new_size
,
1845 ASSERT(ismrlocked(&(ip
->i_lock
), MR_UPDATE
) != 0);
1846 ASSERT(ismrlocked(&(ip
->i_iolock
), MR_UPDATE
) != 0);
1847 ASSERT(new_size
> ip
->i_size
);
1850 * Zero any pages that may have been created by
1851 * xfs_write_file() beyond the end of the file
1852 * and any blocks between the old and new file sizes.
1854 return xfs_zero_eof(ip
, new_size
, ip
->i_size
);
1860 * This routine is called to extend the size of a file.
1861 * The inode must have both the iolock and the ilock locked
1862 * for update and it must be a part of the current transaction.
1863 * The xfs_igrow_start() function must have been called previously.
1864 * If the change_flag is not zero, the inode change timestamp will
1871 xfs_fsize_t new_size
,
1874 ASSERT(ismrlocked(&(ip
->i_lock
), MR_UPDATE
) != 0);
1875 ASSERT(ismrlocked(&(ip
->i_iolock
), MR_UPDATE
) != 0);
1876 ASSERT(ip
->i_transp
== tp
);
1877 ASSERT(new_size
> ip
->i_size
);
1880 * Update the file size. Update the inode change timestamp
1881 * if change_flag set.
1883 ip
->i_d
.di_size
= new_size
;
1884 ip
->i_size
= new_size
;
1886 xfs_ichgtime(ip
, XFS_ICHGTIME_CHG
);
1887 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1893 * This is called when the inode's link count goes to 0.
1894 * We place the on-disk inode on a list in the AGI. It
1895 * will be pulled from this list when the inode is freed.
1907 xfs_agnumber_t agno
;
1908 xfs_daddr_t agdaddr
;
1915 ASSERT(ip
->i_d
.di_nlink
== 0);
1916 ASSERT(ip
->i_d
.di_mode
!= 0);
1917 ASSERT(ip
->i_transp
== tp
);
1921 agno
= XFS_INO_TO_AGNO(mp
, ip
->i_ino
);
1922 agdaddr
= XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
));
1925 * Get the agi buffer first. It ensures lock ordering
1928 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, agdaddr
,
1929 XFS_FSS_TO_BB(mp
, 1), 0, &agibp
);
1934 * Validate the magic number of the agi block.
1936 agi
= XFS_BUF_TO_AGI(agibp
);
1938 be32_to_cpu(agi
->agi_magicnum
) == XFS_AGI_MAGIC
&&
1939 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi
->agi_versionnum
));
1940 if (unlikely(XFS_TEST_ERROR(!agi_ok
, mp
, XFS_ERRTAG_IUNLINK
,
1941 XFS_RANDOM_IUNLINK
))) {
1942 XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW
, mp
, agi
);
1943 xfs_trans_brelse(tp
, agibp
);
1944 return XFS_ERROR(EFSCORRUPTED
);
1947 * Get the index into the agi hash table for the
1948 * list this inode will go on.
1950 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
1952 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
1953 ASSERT(agi
->agi_unlinked
[bucket_index
]);
1954 ASSERT(be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != agino
);
1956 if (be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != NULLAGINO
) {
1958 * There is already another inode in the bucket we need
1959 * to add ourselves to. Add us at the front of the list.
1960 * Here we put the head pointer into our next pointer,
1961 * and then we fall through to point the head at us.
1963 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, 0, 0);
1967 ASSERT(be32_to_cpu(dip
->di_next_unlinked
) == NULLAGINO
);
1968 /* both on-disk, don't endian flip twice */
1969 dip
->di_next_unlinked
= agi
->agi_unlinked
[bucket_index
];
1970 offset
= ip
->i_boffset
+
1971 offsetof(xfs_dinode_t
, di_next_unlinked
);
1972 xfs_trans_inode_buf(tp
, ibp
);
1973 xfs_trans_log_buf(tp
, ibp
, offset
,
1974 (offset
+ sizeof(xfs_agino_t
) - 1));
1975 xfs_inobp_check(mp
, ibp
);
1979 * Point the bucket head pointer at the inode being inserted.
1982 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(agino
);
1983 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
1984 (sizeof(xfs_agino_t
) * bucket_index
);
1985 xfs_trans_log_buf(tp
, agibp
, offset
,
1986 (offset
+ sizeof(xfs_agino_t
) - 1));
1991 * Pull the on-disk inode from the AGI unlinked list.
2004 xfs_agnumber_t agno
;
2005 xfs_daddr_t agdaddr
;
2007 xfs_agino_t next_agino
;
2008 xfs_buf_t
*last_ibp
;
2009 xfs_dinode_t
*last_dip
= NULL
;
2011 int offset
, last_offset
= 0;
2016 * First pull the on-disk inode from the AGI unlinked list.
2020 agno
= XFS_INO_TO_AGNO(mp
, ip
->i_ino
);
2021 agdaddr
= XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
));
2024 * Get the agi buffer first. It ensures lock ordering
2027 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, agdaddr
,
2028 XFS_FSS_TO_BB(mp
, 1), 0, &agibp
);
2031 "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
2032 error
, mp
->m_fsname
);
2036 * Validate the magic number of the agi block.
2038 agi
= XFS_BUF_TO_AGI(agibp
);
2040 be32_to_cpu(agi
->agi_magicnum
) == XFS_AGI_MAGIC
&&
2041 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi
->agi_versionnum
));
2042 if (unlikely(XFS_TEST_ERROR(!agi_ok
, mp
, XFS_ERRTAG_IUNLINK_REMOVE
,
2043 XFS_RANDOM_IUNLINK_REMOVE
))) {
2044 XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW
,
2046 xfs_trans_brelse(tp
, agibp
);
2048 "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
2050 return XFS_ERROR(EFSCORRUPTED
);
2053 * Get the index into the agi hash table for the
2054 * list this inode will go on.
2056 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
2058 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
2059 ASSERT(be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != NULLAGINO
);
2060 ASSERT(agi
->agi_unlinked
[bucket_index
]);
2062 if (be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) == agino
) {
2064 * We're at the head of the list. Get the inode's
2065 * on-disk buffer to see if there is anyone after us
2066 * on the list. Only modify our next pointer if it
2067 * is not already NULLAGINO. This saves us the overhead
2068 * of dealing with the buffer when there is no need to
2071 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, 0, 0);
2074 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2075 error
, mp
->m_fsname
);
2078 next_agino
= be32_to_cpu(dip
->di_next_unlinked
);
2079 ASSERT(next_agino
!= 0);
2080 if (next_agino
!= NULLAGINO
) {
2081 dip
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
2082 offset
= ip
->i_boffset
+
2083 offsetof(xfs_dinode_t
, di_next_unlinked
);
2084 xfs_trans_inode_buf(tp
, ibp
);
2085 xfs_trans_log_buf(tp
, ibp
, offset
,
2086 (offset
+ sizeof(xfs_agino_t
) - 1));
2087 xfs_inobp_check(mp
, ibp
);
2089 xfs_trans_brelse(tp
, ibp
);
2092 * Point the bucket head pointer at the next inode.
2094 ASSERT(next_agino
!= 0);
2095 ASSERT(next_agino
!= agino
);
2096 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(next_agino
);
2097 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
2098 (sizeof(xfs_agino_t
) * bucket_index
);
2099 xfs_trans_log_buf(tp
, agibp
, offset
,
2100 (offset
+ sizeof(xfs_agino_t
) - 1));
2103 * We need to search the list for the inode being freed.
2105 next_agino
= be32_to_cpu(agi
->agi_unlinked
[bucket_index
]);
2107 while (next_agino
!= agino
) {
2109 * If the last inode wasn't the one pointing to
2110 * us, then release its buffer since we're not
2111 * going to do anything with it.
2113 if (last_ibp
!= NULL
) {
2114 xfs_trans_brelse(tp
, last_ibp
);
2116 next_ino
= XFS_AGINO_TO_INO(mp
, agno
, next_agino
);
2117 error
= xfs_inotobp(mp
, tp
, next_ino
, &last_dip
,
2118 &last_ibp
, &last_offset
);
2121 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
2122 error
, mp
->m_fsname
);
2125 next_agino
= be32_to_cpu(last_dip
->di_next_unlinked
);
2126 ASSERT(next_agino
!= NULLAGINO
);
2127 ASSERT(next_agino
!= 0);
2130 * Now last_ibp points to the buffer previous to us on
2131 * the unlinked list. Pull us from the list.
2133 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, 0, 0);
2136 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2137 error
, mp
->m_fsname
);
2140 next_agino
= be32_to_cpu(dip
->di_next_unlinked
);
2141 ASSERT(next_agino
!= 0);
2142 ASSERT(next_agino
!= agino
);
2143 if (next_agino
!= NULLAGINO
) {
2144 dip
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
2145 offset
= ip
->i_boffset
+
2146 offsetof(xfs_dinode_t
, di_next_unlinked
);
2147 xfs_trans_inode_buf(tp
, ibp
);
2148 xfs_trans_log_buf(tp
, ibp
, offset
,
2149 (offset
+ sizeof(xfs_agino_t
) - 1));
2150 xfs_inobp_check(mp
, ibp
);
2152 xfs_trans_brelse(tp
, ibp
);
2155 * Point the previous inode on the list to the next inode.
2157 last_dip
->di_next_unlinked
= cpu_to_be32(next_agino
);
2158 ASSERT(next_agino
!= 0);
2159 offset
= last_offset
+ offsetof(xfs_dinode_t
, di_next_unlinked
);
2160 xfs_trans_inode_buf(tp
, last_ibp
);
2161 xfs_trans_log_buf(tp
, last_ibp
, offset
,
2162 (offset
+ sizeof(xfs_agino_t
) - 1));
2163 xfs_inobp_check(mp
, last_ibp
);
2168 STATIC_INLINE
int xfs_inode_clean(xfs_inode_t
*ip
)
2170 return (((ip
->i_itemp
== NULL
) ||
2171 !(ip
->i_itemp
->ili_format
.ilf_fields
& XFS_ILOG_ALL
)) &&
2172 (ip
->i_update_core
== 0));
2177 xfs_inode_t
*free_ip
,
2181 xfs_mount_t
*mp
= free_ip
->i_mount
;
2182 int blks_per_cluster
;
2185 int i
, j
, found
, pre_flushed
;
2188 xfs_inode_t
*ip
, **ip_found
;
2189 xfs_inode_log_item_t
*iip
;
2190 xfs_log_item_t
*lip
;
2191 xfs_perag_t
*pag
= xfs_get_perag(mp
, inum
);
2193 if (mp
->m_sb
.sb_blocksize
>= XFS_INODE_CLUSTER_SIZE(mp
)) {
2194 blks_per_cluster
= 1;
2195 ninodes
= mp
->m_sb
.sb_inopblock
;
2196 nbufs
= XFS_IALLOC_BLOCKS(mp
);
2198 blks_per_cluster
= XFS_INODE_CLUSTER_SIZE(mp
) /
2199 mp
->m_sb
.sb_blocksize
;
2200 ninodes
= blks_per_cluster
* mp
->m_sb
.sb_inopblock
;
2201 nbufs
= XFS_IALLOC_BLOCKS(mp
) / blks_per_cluster
;
2204 ip_found
= kmem_alloc(ninodes
* sizeof(xfs_inode_t
*), KM_NOFS
);
2206 for (j
= 0; j
< nbufs
; j
++, inum
+= ninodes
) {
2207 blkno
= XFS_AGB_TO_DADDR(mp
, XFS_INO_TO_AGNO(mp
, inum
),
2208 XFS_INO_TO_AGBNO(mp
, inum
));
2212 * Look for each inode in memory and attempt to lock it,
2213 * we can be racing with flush and tail pushing here.
2214 * any inode we get the locks on, add to an array of
2215 * inode items to process later.
2217 * The get the buffer lock, we could beat a flush
2218 * or tail pushing thread to the lock here, in which
2219 * case they will go looking for the inode buffer
2220 * and fail, we need some other form of interlock
2224 for (i
= 0; i
< ninodes
; i
++) {
2225 read_lock(&pag
->pag_ici_lock
);
2226 ip
= radix_tree_lookup(&pag
->pag_ici_root
,
2227 XFS_INO_TO_AGINO(mp
, (inum
+ i
)));
2229 /* Inode not in memory or we found it already,
2232 if (!ip
|| xfs_iflags_test(ip
, XFS_ISTALE
)) {
2233 read_unlock(&pag
->pag_ici_lock
);
2237 if (xfs_inode_clean(ip
)) {
2238 read_unlock(&pag
->pag_ici_lock
);
2242 /* If we can get the locks then add it to the
2243 * list, otherwise by the time we get the bp lock
2244 * below it will already be attached to the
2248 /* This inode will already be locked - by us, lets
2252 if (ip
== free_ip
) {
2253 if (xfs_iflock_nowait(ip
)) {
2254 xfs_iflags_set(ip
, XFS_ISTALE
);
2255 if (xfs_inode_clean(ip
)) {
2258 ip_found
[found
++] = ip
;
2261 read_unlock(&pag
->pag_ici_lock
);
2265 if (xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
)) {
2266 if (xfs_iflock_nowait(ip
)) {
2267 xfs_iflags_set(ip
, XFS_ISTALE
);
2269 if (xfs_inode_clean(ip
)) {
2271 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2273 ip_found
[found
++] = ip
;
2276 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2279 read_unlock(&pag
->pag_ici_lock
);
2282 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, blkno
,
2283 mp
->m_bsize
* blks_per_cluster
,
2287 lip
= XFS_BUF_FSPRIVATE(bp
, xfs_log_item_t
*);
2289 if (lip
->li_type
== XFS_LI_INODE
) {
2290 iip
= (xfs_inode_log_item_t
*)lip
;
2291 ASSERT(iip
->ili_logged
== 1);
2292 lip
->li_cb
= (void(*)(xfs_buf_t
*,xfs_log_item_t
*)) xfs_istale_done
;
2293 spin_lock(&mp
->m_ail_lock
);
2294 iip
->ili_flush_lsn
= iip
->ili_item
.li_lsn
;
2295 spin_unlock(&mp
->m_ail_lock
);
2296 xfs_iflags_set(iip
->ili_inode
, XFS_ISTALE
);
2299 lip
= lip
->li_bio_list
;
2302 for (i
= 0; i
< found
; i
++) {
2307 ip
->i_update_core
= 0;
2309 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2313 iip
->ili_last_fields
= iip
->ili_format
.ilf_fields
;
2314 iip
->ili_format
.ilf_fields
= 0;
2315 iip
->ili_logged
= 1;
2316 spin_lock(&mp
->m_ail_lock
);
2317 iip
->ili_flush_lsn
= iip
->ili_item
.li_lsn
;
2318 spin_unlock(&mp
->m_ail_lock
);
2320 xfs_buf_attach_iodone(bp
,
2321 (void(*)(xfs_buf_t
*,xfs_log_item_t
*))
2322 xfs_istale_done
, (xfs_log_item_t
*)iip
);
2323 if (ip
!= free_ip
) {
2324 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2328 if (found
|| pre_flushed
)
2329 xfs_trans_stale_inode_buf(tp
, bp
);
2330 xfs_trans_binval(tp
, bp
);
2333 kmem_free(ip_found
, ninodes
* sizeof(xfs_inode_t
*));
2334 xfs_put_perag(mp
, pag
);
2338 * This is called to return an inode to the inode free list.
2339 * The inode should already be truncated to 0 length and have
2340 * no pages associated with it. This routine also assumes that
2341 * the inode is already a part of the transaction.
2343 * The on-disk copy of the inode will have been added to the list
2344 * of unlinked inodes in the AGI. We need to remove the inode from
2345 * that list atomically with respect to freeing it here.
2351 xfs_bmap_free_t
*flist
)
2355 xfs_ino_t first_ino
;
2359 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
));
2360 ASSERT(ip
->i_transp
== tp
);
2361 ASSERT(ip
->i_d
.di_nlink
== 0);
2362 ASSERT(ip
->i_d
.di_nextents
== 0);
2363 ASSERT(ip
->i_d
.di_anextents
== 0);
2364 ASSERT((ip
->i_d
.di_size
== 0 && ip
->i_size
== 0) ||
2365 ((ip
->i_d
.di_mode
& S_IFMT
) != S_IFREG
));
2366 ASSERT(ip
->i_d
.di_nblocks
== 0);
2369 * Pull the on-disk inode from the AGI unlinked list.
2371 error
= xfs_iunlink_remove(tp
, ip
);
2376 error
= xfs_difree(tp
, ip
->i_ino
, flist
, &delete, &first_ino
);
2380 ip
->i_d
.di_mode
= 0; /* mark incore inode as free */
2381 ip
->i_d
.di_flags
= 0;
2382 ip
->i_d
.di_dmevmask
= 0;
2383 ip
->i_d
.di_forkoff
= 0; /* mark the attr fork not in use */
2384 ip
->i_df
.if_ext_max
=
2385 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
2386 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
2387 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
2389 * Bump the generation count so no one will be confused
2390 * by reincarnations of this inode.
2394 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2396 error
= xfs_itobp(ip
->i_mount
, tp
, ip
, &dip
, &ibp
, 0, 0);
2401 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2402 * from picking up this inode when it is reclaimed (its incore state
2403 * initialzed but not flushed to disk yet). The in-core di_mode is
2404 * already cleared and a corresponding transaction logged.
2405 * The hack here just synchronizes the in-core to on-disk
2406 * di_mode value in advance before the actual inode sync to disk.
2407 * This is OK because the inode is already unlinked and would never
2408 * change its di_mode again for this inode generation.
2409 * This is a temporary hack that would require a proper fix
2412 dip
->di_core
.di_mode
= 0;
2415 xfs_ifree_cluster(ip
, tp
, first_ino
);
2422 * Reallocate the space for if_broot based on the number of records
2423 * being added or deleted as indicated in rec_diff. Move the records
2424 * and pointers in if_broot to fit the new size. When shrinking this
2425 * will eliminate holes between the records and pointers created by
2426 * the caller. When growing this will create holes to be filled in
2429 * The caller must not request to add more records than would fit in
2430 * the on-disk inode root. If the if_broot is currently NULL, then
2431 * if we adding records one will be allocated. The caller must also
2432 * not request that the number of records go below zero, although
2433 * it can go to zero.
2435 * ip -- the inode whose if_broot area is changing
2436 * ext_diff -- the change in the number of records, positive or negative,
2437 * requested for the if_broot array.
2447 xfs_bmbt_block_t
*new_broot
;
2454 * Handle the degenerate case quietly.
2456 if (rec_diff
== 0) {
2460 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2463 * If there wasn't any memory allocated before, just
2464 * allocate it now and get out.
2466 if (ifp
->if_broot_bytes
== 0) {
2467 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff
);
2468 ifp
->if_broot
= (xfs_bmbt_block_t
*)kmem_alloc(new_size
,
2470 ifp
->if_broot_bytes
= (int)new_size
;
2475 * If there is already an existing if_broot, then we need
2476 * to realloc() it and shift the pointers to their new
2477 * location. The records don't change location because
2478 * they are kept butted up against the btree block header.
2480 cur_max
= XFS_BMAP_BROOT_MAXRECS(ifp
->if_broot_bytes
);
2481 new_max
= cur_max
+ rec_diff
;
2482 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max
);
2483 ifp
->if_broot
= (xfs_bmbt_block_t
*)
2484 kmem_realloc(ifp
->if_broot
,
2486 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max
), /* old size */
2488 op
= (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp
->if_broot
, 1,
2489 ifp
->if_broot_bytes
);
2490 np
= (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp
->if_broot
, 1,
2492 ifp
->if_broot_bytes
= (int)new_size
;
2493 ASSERT(ifp
->if_broot_bytes
<=
2494 XFS_IFORK_SIZE(ip
, whichfork
) + XFS_BROOT_SIZE_ADJ
);
2495 memmove(np
, op
, cur_max
* (uint
)sizeof(xfs_dfsbno_t
));
2500 * rec_diff is less than 0. In this case, we are shrinking the
2501 * if_broot buffer. It must already exist. If we go to zero
2502 * records, just get rid of the root and clear the status bit.
2504 ASSERT((ifp
->if_broot
!= NULL
) && (ifp
->if_broot_bytes
> 0));
2505 cur_max
= XFS_BMAP_BROOT_MAXRECS(ifp
->if_broot_bytes
);
2506 new_max
= cur_max
+ rec_diff
;
2507 ASSERT(new_max
>= 0);
2509 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max
);
2513 new_broot
= (xfs_bmbt_block_t
*)kmem_alloc(new_size
, KM_SLEEP
);
2515 * First copy over the btree block header.
2517 memcpy(new_broot
, ifp
->if_broot
, sizeof(xfs_bmbt_block_t
));
2520 ifp
->if_flags
&= ~XFS_IFBROOT
;
2524 * Only copy the records and pointers if there are any.
2528 * First copy the records.
2530 op
= (char *)XFS_BMAP_BROOT_REC_ADDR(ifp
->if_broot
, 1,
2531 ifp
->if_broot_bytes
);
2532 np
= (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot
, 1,
2534 memcpy(np
, op
, new_max
* (uint
)sizeof(xfs_bmbt_rec_t
));
2537 * Then copy the pointers.
2539 op
= (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp
->if_broot
, 1,
2540 ifp
->if_broot_bytes
);
2541 np
= (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot
, 1,
2543 memcpy(np
, op
, new_max
* (uint
)sizeof(xfs_dfsbno_t
));
2545 kmem_free(ifp
->if_broot
, ifp
->if_broot_bytes
);
2546 ifp
->if_broot
= new_broot
;
2547 ifp
->if_broot_bytes
= (int)new_size
;
2548 ASSERT(ifp
->if_broot_bytes
<=
2549 XFS_IFORK_SIZE(ip
, whichfork
) + XFS_BROOT_SIZE_ADJ
);
2555 * This is called when the amount of space needed for if_data
2556 * is increased or decreased. The change in size is indicated by
2557 * the number of bytes that need to be added or deleted in the
2558 * byte_diff parameter.
2560 * If the amount of space needed has decreased below the size of the
2561 * inline buffer, then switch to using the inline buffer. Otherwise,
2562 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2563 * to what is needed.
2565 * ip -- the inode whose if_data area is changing
2566 * byte_diff -- the change in the number of bytes, positive or negative,
2567 * requested for the if_data array.
2579 if (byte_diff
== 0) {
2583 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2584 new_size
= (int)ifp
->if_bytes
+ byte_diff
;
2585 ASSERT(new_size
>= 0);
2587 if (new_size
== 0) {
2588 if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
2589 kmem_free(ifp
->if_u1
.if_data
, ifp
->if_real_bytes
);
2591 ifp
->if_u1
.if_data
= NULL
;
2593 } else if (new_size
<= sizeof(ifp
->if_u2
.if_inline_data
)) {
2595 * If the valid extents/data can fit in if_inline_ext/data,
2596 * copy them from the malloc'd vector and free it.
2598 if (ifp
->if_u1
.if_data
== NULL
) {
2599 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
2600 } else if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
2601 ASSERT(ifp
->if_real_bytes
!= 0);
2602 memcpy(ifp
->if_u2
.if_inline_data
, ifp
->if_u1
.if_data
,
2604 kmem_free(ifp
->if_u1
.if_data
, ifp
->if_real_bytes
);
2605 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
2610 * Stuck with malloc/realloc.
2611 * For inline data, the underlying buffer must be
2612 * a multiple of 4 bytes in size so that it can be
2613 * logged and stay on word boundaries. We enforce
2616 real_size
= roundup(new_size
, 4);
2617 if (ifp
->if_u1
.if_data
== NULL
) {
2618 ASSERT(ifp
->if_real_bytes
== 0);
2619 ifp
->if_u1
.if_data
= kmem_alloc(real_size
, KM_SLEEP
);
2620 } else if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
2622 * Only do the realloc if the underlying size
2623 * is really changing.
2625 if (ifp
->if_real_bytes
!= real_size
) {
2626 ifp
->if_u1
.if_data
=
2627 kmem_realloc(ifp
->if_u1
.if_data
,
2633 ASSERT(ifp
->if_real_bytes
== 0);
2634 ifp
->if_u1
.if_data
= kmem_alloc(real_size
, KM_SLEEP
);
2635 memcpy(ifp
->if_u1
.if_data
, ifp
->if_u2
.if_inline_data
,
2639 ifp
->if_real_bytes
= real_size
;
2640 ifp
->if_bytes
= new_size
;
2641 ASSERT(ifp
->if_bytes
<= XFS_IFORK_SIZE(ip
, whichfork
));
2648 * Map inode to disk block and offset.
2650 * mp -- the mount point structure for the current file system
2651 * tp -- the current transaction
2652 * ino -- the inode number of the inode to be located
2653 * imap -- this structure is filled in with the information necessary
2654 * to retrieve the given inode from disk
2655 * flags -- flags to pass to xfs_dilocate indicating whether or not
2656 * lookups in the inode btree were OK or not
2666 xfs_fsblock_t fsbno
;
2671 fsbno
= imap
->im_blkno
?
2672 XFS_DADDR_TO_FSB(mp
, imap
->im_blkno
) : NULLFSBLOCK
;
2673 error
= xfs_dilocate(mp
, tp
, ino
, &fsbno
, &len
, &off
, flags
);
2677 imap
->im_blkno
= XFS_FSB_TO_DADDR(mp
, fsbno
);
2678 imap
->im_len
= XFS_FSB_TO_BB(mp
, len
);
2679 imap
->im_agblkno
= XFS_FSB_TO_AGBNO(mp
, fsbno
);
2680 imap
->im_ioffset
= (ushort
)off
;
2681 imap
->im_boffset
= (ushort
)(off
<< mp
->m_sb
.sb_inodelog
);
2692 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2693 if (ifp
->if_broot
!= NULL
) {
2694 kmem_free(ifp
->if_broot
, ifp
->if_broot_bytes
);
2695 ifp
->if_broot
= NULL
;
2699 * If the format is local, then we can't have an extents
2700 * array so just look for an inline data array. If we're
2701 * not local then we may or may not have an extents list,
2702 * so check and free it up if we do.
2704 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
) {
2705 if ((ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) &&
2706 (ifp
->if_u1
.if_data
!= NULL
)) {
2707 ASSERT(ifp
->if_real_bytes
!= 0);
2708 kmem_free(ifp
->if_u1
.if_data
, ifp
->if_real_bytes
);
2709 ifp
->if_u1
.if_data
= NULL
;
2710 ifp
->if_real_bytes
= 0;
2712 } else if ((ifp
->if_flags
& XFS_IFEXTENTS
) &&
2713 ((ifp
->if_flags
& XFS_IFEXTIREC
) ||
2714 ((ifp
->if_u1
.if_extents
!= NULL
) &&
2715 (ifp
->if_u1
.if_extents
!= ifp
->if_u2
.if_inline_ext
)))) {
2716 ASSERT(ifp
->if_real_bytes
!= 0);
2717 xfs_iext_destroy(ifp
);
2719 ASSERT(ifp
->if_u1
.if_extents
== NULL
||
2720 ifp
->if_u1
.if_extents
== ifp
->if_u2
.if_inline_ext
);
2721 ASSERT(ifp
->if_real_bytes
== 0);
2722 if (whichfork
== XFS_ATTR_FORK
) {
2723 kmem_zone_free(xfs_ifork_zone
, ip
->i_afp
);
2729 * This is called free all the memory associated with an inode.
2730 * It must free the inode itself and any buffers allocated for
2731 * if_extents/if_data and if_broot. It must also free the lock
2732 * associated with the inode.
2738 switch (ip
->i_d
.di_mode
& S_IFMT
) {
2742 xfs_idestroy_fork(ip
, XFS_DATA_FORK
);
2746 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
2747 mrfree(&ip
->i_lock
);
2748 mrfree(&ip
->i_iolock
);
2749 freesema(&ip
->i_flock
);
2751 #ifdef XFS_INODE_TRACE
2752 ktrace_free(ip
->i_trace
);
2754 #ifdef XFS_BMAP_TRACE
2755 ktrace_free(ip
->i_xtrace
);
2757 #ifdef XFS_BMBT_TRACE
2758 ktrace_free(ip
->i_btrace
);
2761 ktrace_free(ip
->i_rwtrace
);
2763 #ifdef XFS_ILOCK_TRACE
2764 ktrace_free(ip
->i_lock_trace
);
2766 #ifdef XFS_DIR2_TRACE
2767 ktrace_free(ip
->i_dir_trace
);
2771 * Only if we are shutting down the fs will we see an
2772 * inode still in the AIL. If it is there, we should remove
2773 * it to prevent a use-after-free from occurring.
2775 xfs_mount_t
*mp
= ip
->i_mount
;
2776 xfs_log_item_t
*lip
= &ip
->i_itemp
->ili_item
;
2778 ASSERT(((lip
->li_flags
& XFS_LI_IN_AIL
) == 0) ||
2779 XFS_FORCED_SHUTDOWN(ip
->i_mount
));
2780 if (lip
->li_flags
& XFS_LI_IN_AIL
) {
2781 spin_lock(&mp
->m_ail_lock
);
2782 if (lip
->li_flags
& XFS_LI_IN_AIL
)
2783 xfs_trans_delete_ail(mp
, lip
);
2785 spin_unlock(&mp
->m_ail_lock
);
2787 xfs_inode_item_destroy(ip
);
2789 kmem_zone_free(xfs_inode_zone
, ip
);
2794 * Increment the pin count of the given buffer.
2795 * This value is protected by ipinlock spinlock in the mount structure.
2801 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
));
2803 atomic_inc(&ip
->i_pincount
);
2807 * Decrement the pin count of the given inode, and wake up
2808 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2809 * inode must have been previously pinned with a call to xfs_ipin().
2815 ASSERT(atomic_read(&ip
->i_pincount
) > 0);
2817 if (atomic_dec_and_test(&ip
->i_pincount
))
2818 wake_up(&ip
->i_ipin_wait
);
2822 * This is called to wait for the given inode to be unpinned.
2823 * It will sleep until this happens. The caller must have the
2824 * inode locked in at least shared mode so that the buffer cannot
2825 * be subsequently pinned once someone is waiting for it to be
2832 xfs_inode_log_item_t
*iip
;
2835 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
| MR_ACCESS
));
2837 if (atomic_read(&ip
->i_pincount
) == 0) {
2842 if (iip
&& iip
->ili_last_lsn
) {
2843 lsn
= iip
->ili_last_lsn
;
2849 * Give the log a push so we don't wait here too long.
2851 xfs_log_force(ip
->i_mount
, lsn
, XFS_LOG_FORCE
);
2853 wait_event(ip
->i_ipin_wait
, (atomic_read(&ip
->i_pincount
) == 0));
2858 * xfs_iextents_copy()
2860 * This is called to copy the REAL extents (as opposed to the delayed
2861 * allocation extents) from the inode into the given buffer. It
2862 * returns the number of bytes copied into the buffer.
2864 * If there are no delayed allocation extents, then we can just
2865 * memcpy() the extents into the buffer. Otherwise, we need to
2866 * examine each extent in turn and skip those which are delayed.
2878 xfs_fsblock_t start_block
;
2880 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2881 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
|MR_ACCESS
));
2882 ASSERT(ifp
->if_bytes
> 0);
2884 nrecs
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
2885 XFS_BMAP_TRACE_EXLIST(ip
, nrecs
, whichfork
);
2889 * There are some delayed allocation extents in the
2890 * inode, so copy the extents one at a time and skip
2891 * the delayed ones. There must be at least one
2892 * non-delayed extent.
2895 for (i
= 0; i
< nrecs
; i
++) {
2896 xfs_bmbt_rec_host_t
*ep
= xfs_iext_get_ext(ifp
, i
);
2897 start_block
= xfs_bmbt_get_startblock(ep
);
2898 if (ISNULLSTARTBLOCK(start_block
)) {
2900 * It's a delayed allocation extent, so skip it.
2905 /* Translate to on disk format */
2906 put_unaligned(cpu_to_be64(ep
->l0
), &dp
->l0
);
2907 put_unaligned(cpu_to_be64(ep
->l1
), &dp
->l1
);
2911 ASSERT(copied
!= 0);
2912 xfs_validate_extents(ifp
, copied
, XFS_EXTFMT_INODE(ip
));
2914 return (copied
* (uint
)sizeof(xfs_bmbt_rec_t
));
2918 * Each of the following cases stores data into the same region
2919 * of the on-disk inode, so only one of them can be valid at
2920 * any given time. While it is possible to have conflicting formats
2921 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2922 * in EXTENTS format, this can only happen when the fork has
2923 * changed formats after being modified but before being flushed.
2924 * In these cases, the format always takes precedence, because the
2925 * format indicates the current state of the fork.
2932 xfs_inode_log_item_t
*iip
,
2939 #ifdef XFS_TRANS_DEBUG
2942 static const short brootflag
[2] =
2943 { XFS_ILOG_DBROOT
, XFS_ILOG_ABROOT
};
2944 static const short dataflag
[2] =
2945 { XFS_ILOG_DDATA
, XFS_ILOG_ADATA
};
2946 static const short extflag
[2] =
2947 { XFS_ILOG_DEXT
, XFS_ILOG_AEXT
};
2951 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2953 * This can happen if we gave up in iformat in an error path,
2954 * for the attribute fork.
2957 ASSERT(whichfork
== XFS_ATTR_FORK
);
2960 cp
= XFS_DFORK_PTR(dip
, whichfork
);
2962 switch (XFS_IFORK_FORMAT(ip
, whichfork
)) {
2963 case XFS_DINODE_FMT_LOCAL
:
2964 if ((iip
->ili_format
.ilf_fields
& dataflag
[whichfork
]) &&
2965 (ifp
->if_bytes
> 0)) {
2966 ASSERT(ifp
->if_u1
.if_data
!= NULL
);
2967 ASSERT(ifp
->if_bytes
<= XFS_IFORK_SIZE(ip
, whichfork
));
2968 memcpy(cp
, ifp
->if_u1
.if_data
, ifp
->if_bytes
);
2972 case XFS_DINODE_FMT_EXTENTS
:
2973 ASSERT((ifp
->if_flags
& XFS_IFEXTENTS
) ||
2974 !(iip
->ili_format
.ilf_fields
& extflag
[whichfork
]));
2975 ASSERT((xfs_iext_get_ext(ifp
, 0) != NULL
) ||
2976 (ifp
->if_bytes
== 0));
2977 ASSERT((xfs_iext_get_ext(ifp
, 0) == NULL
) ||
2978 (ifp
->if_bytes
> 0));
2979 if ((iip
->ili_format
.ilf_fields
& extflag
[whichfork
]) &&
2980 (ifp
->if_bytes
> 0)) {
2981 ASSERT(XFS_IFORK_NEXTENTS(ip
, whichfork
) > 0);
2982 (void)xfs_iextents_copy(ip
, (xfs_bmbt_rec_t
*)cp
,
2987 case XFS_DINODE_FMT_BTREE
:
2988 if ((iip
->ili_format
.ilf_fields
& brootflag
[whichfork
]) &&
2989 (ifp
->if_broot_bytes
> 0)) {
2990 ASSERT(ifp
->if_broot
!= NULL
);
2991 ASSERT(ifp
->if_broot_bytes
<=
2992 (XFS_IFORK_SIZE(ip
, whichfork
) +
2993 XFS_BROOT_SIZE_ADJ
));
2994 xfs_bmbt_to_bmdr(ifp
->if_broot
, ifp
->if_broot_bytes
,
2995 (xfs_bmdr_block_t
*)cp
,
2996 XFS_DFORK_SIZE(dip
, mp
, whichfork
));
3000 case XFS_DINODE_FMT_DEV
:
3001 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_DEV
) {
3002 ASSERT(whichfork
== XFS_DATA_FORK
);
3003 dip
->di_u
.di_dev
= cpu_to_be32(ip
->i_df
.if_u2
.if_rdev
);
3007 case XFS_DINODE_FMT_UUID
:
3008 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_UUID
) {
3009 ASSERT(whichfork
== XFS_DATA_FORK
);
3010 memcpy(&dip
->di_u
.di_muuid
, &ip
->i_df
.if_u2
.if_uuid
,
3024 * xfs_iflush() will write a modified inode's changes out to the
3025 * inode's on disk home. The caller must have the inode lock held
3026 * in at least shared mode and the inode flush semaphore must be
3027 * held as well. The inode lock will still be held upon return from
3028 * the call and the caller is free to unlock it.
3029 * The inode flush lock will be unlocked when the inode reaches the disk.
3030 * The flags indicate how the inode's buffer should be written out.
3037 xfs_inode_log_item_t
*iip
;
3044 int clcount
; /* count of inodes clustered */
3046 struct hlist_node
*entry
;
3047 enum { INT_DELWRI
= (1 << 0), INT_ASYNC
= (1 << 1) };
3049 XFS_STATS_INC(xs_iflush_count
);
3051 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
|MR_ACCESS
));
3052 ASSERT(issemalocked(&(ip
->i_flock
)));
3053 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
3054 ip
->i_d
.di_nextents
> ip
->i_df
.if_ext_max
);
3060 * If the inode isn't dirty, then just release the inode
3061 * flush lock and do nothing.
3063 if ((ip
->i_update_core
== 0) &&
3064 ((iip
== NULL
) || !(iip
->ili_format
.ilf_fields
& XFS_ILOG_ALL
))) {
3065 ASSERT((iip
!= NULL
) ?
3066 !(iip
->ili_item
.li_flags
& XFS_LI_IN_AIL
) : 1);
3072 * We can't flush the inode until it is unpinned, so
3073 * wait for it. We know noone new can pin it, because
3074 * we are holding the inode lock shared and you need
3075 * to hold it exclusively to pin the inode.
3077 xfs_iunpin_wait(ip
);
3080 * This may have been unpinned because the filesystem is shutting
3081 * down forcibly. If that's the case we must not write this inode
3082 * to disk, because the log record didn't make it to disk!
3084 if (XFS_FORCED_SHUTDOWN(mp
)) {
3085 ip
->i_update_core
= 0;
3087 iip
->ili_format
.ilf_fields
= 0;
3089 return XFS_ERROR(EIO
);
3093 * Get the buffer containing the on-disk inode.
3095 error
= xfs_itobp(mp
, NULL
, ip
, &dip
, &bp
, 0, 0);
3102 * Decide how buffer will be flushed out. This is done before
3103 * the call to xfs_iflush_int because this field is zeroed by it.
3105 if (iip
!= NULL
&& iip
->ili_format
.ilf_fields
!= 0) {
3107 * Flush out the inode buffer according to the directions
3108 * of the caller. In the cases where the caller has given
3109 * us a choice choose the non-delwri case. This is because
3110 * the inode is in the AIL and we need to get it out soon.
3113 case XFS_IFLUSH_SYNC
:
3114 case XFS_IFLUSH_DELWRI_ELSE_SYNC
:
3117 case XFS_IFLUSH_ASYNC
:
3118 case XFS_IFLUSH_DELWRI_ELSE_ASYNC
:
3121 case XFS_IFLUSH_DELWRI
:
3131 case XFS_IFLUSH_DELWRI_ELSE_SYNC
:
3132 case XFS_IFLUSH_DELWRI_ELSE_ASYNC
:
3133 case XFS_IFLUSH_DELWRI
:
3136 case XFS_IFLUSH_ASYNC
:
3139 case XFS_IFLUSH_SYNC
:
3150 * First flush out the inode that xfs_iflush was called with.
3152 error
= xfs_iflush_int(ip
, bp
);
3159 * see if other inodes can be gathered into this write
3161 spin_lock(&ip
->i_cluster
->icl_lock
);
3162 ip
->i_cluster
->icl_buf
= bp
;
3165 hlist_for_each_entry(iq
, entry
, &ip
->i_cluster
->icl_inodes
, i_cnode
) {
3170 * Do an un-protected check to see if the inode is dirty and
3171 * is a candidate for flushing. These checks will be repeated
3172 * later after the appropriate locks are acquired.
3175 if ((iq
->i_update_core
== 0) &&
3177 !(iip
->ili_format
.ilf_fields
& XFS_ILOG_ALL
)) &&
3178 xfs_ipincount(iq
) == 0) {
3183 * Try to get locks. If any are unavailable,
3184 * then this inode cannot be flushed and is skipped.
3187 /* get inode locks (just i_lock) */
3188 if (xfs_ilock_nowait(iq
, XFS_ILOCK_SHARED
)) {
3189 /* get inode flush lock */
3190 if (xfs_iflock_nowait(iq
)) {
3191 /* check if pinned */
3192 if (xfs_ipincount(iq
) == 0) {
3193 /* arriving here means that
3194 * this inode can be flushed.
3195 * first re-check that it's
3199 if ((iq
->i_update_core
!= 0)||
3201 (iip
->ili_format
.ilf_fields
& XFS_ILOG_ALL
))) {
3203 error
= xfs_iflush_int(iq
, bp
);
3207 goto cluster_corrupt_out
;
3216 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
3219 spin_unlock(&ip
->i_cluster
->icl_lock
);
3222 XFS_STATS_INC(xs_icluster_flushcnt
);
3223 XFS_STATS_ADD(xs_icluster_flushinode
, clcount
);
3227 * If the buffer is pinned then push on the log so we won't
3228 * get stuck waiting in the write for too long.
3230 if (XFS_BUF_ISPINNED(bp
)){
3231 xfs_log_force(mp
, (xfs_lsn_t
)0, XFS_LOG_FORCE
);
3234 if (flags
& INT_DELWRI
) {
3235 xfs_bdwrite(mp
, bp
);
3236 } else if (flags
& INT_ASYNC
) {
3237 xfs_bawrite(mp
, bp
);
3239 error
= xfs_bwrite(mp
, bp
);
3245 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
3246 xfs_iflush_abort(ip
);
3248 * Unlocks the flush lock
3250 return XFS_ERROR(EFSCORRUPTED
);
3252 cluster_corrupt_out
:
3253 /* Corruption detected in the clustering loop. Invalidate the
3254 * inode buffer and shut down the filesystem.
3256 spin_unlock(&ip
->i_cluster
->icl_lock
);
3259 * Clean up the buffer. If it was B_DELWRI, just release it --
3260 * brelse can handle it with no problems. If not, shut down the
3261 * filesystem before releasing the buffer.
3263 if ((bufwasdelwri
= XFS_BUF_ISDELAYWRITE(bp
))) {
3267 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
3271 * Just like incore_relse: if we have b_iodone functions,
3272 * mark the buffer as an error and call them. Otherwise
3273 * mark it as stale and brelse.
3275 if (XFS_BUF_IODONE_FUNC(bp
)) {
3276 XFS_BUF_CLR_BDSTRAT_FUNC(bp
);
3280 XFS_BUF_ERROR(bp
,EIO
);
3288 xfs_iflush_abort(iq
);
3290 * Unlocks the flush lock
3292 return XFS_ERROR(EFSCORRUPTED
);
3301 xfs_inode_log_item_t
*iip
;
3304 #ifdef XFS_TRANS_DEBUG
3308 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
|MR_ACCESS
));
3309 ASSERT(issemalocked(&(ip
->i_flock
)));
3310 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
3311 ip
->i_d
.di_nextents
> ip
->i_df
.if_ext_max
);
3318 * If the inode isn't dirty, then just release the inode
3319 * flush lock and do nothing.
3321 if ((ip
->i_update_core
== 0) &&
3322 ((iip
== NULL
) || !(iip
->ili_format
.ilf_fields
& XFS_ILOG_ALL
))) {
3327 /* set *dip = inode's place in the buffer */
3328 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, ip
->i_boffset
);
3331 * Clear i_update_core before copying out the data.
3332 * This is for coordination with our timestamp updates
3333 * that don't hold the inode lock. They will always
3334 * update the timestamps BEFORE setting i_update_core,
3335 * so if we clear i_update_core after they set it we
3336 * are guaranteed to see their updates to the timestamps.
3337 * I believe that this depends on strongly ordered memory
3338 * semantics, but we have that. We use the SYNCHRONIZE
3339 * macro to make sure that the compiler does not reorder
3340 * the i_update_core access below the data copy below.
3342 ip
->i_update_core
= 0;
3346 * Make sure to get the latest atime from the Linux inode.
3348 xfs_synchronize_atime(ip
);
3350 if (XFS_TEST_ERROR(be16_to_cpu(dip
->di_core
.di_magic
) != XFS_DINODE_MAGIC
,
3351 mp
, XFS_ERRTAG_IFLUSH_1
, XFS_RANDOM_IFLUSH_1
)) {
3352 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3353 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3354 ip
->i_ino
, be16_to_cpu(dip
->di_core
.di_magic
), dip
);
3357 if (XFS_TEST_ERROR(ip
->i_d
.di_magic
!= XFS_DINODE_MAGIC
,
3358 mp
, XFS_ERRTAG_IFLUSH_2
, XFS_RANDOM_IFLUSH_2
)) {
3359 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3360 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3361 ip
->i_ino
, ip
, ip
->i_d
.di_magic
);
3364 if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
) {
3366 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3367 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
),
3368 mp
, XFS_ERRTAG_IFLUSH_3
, XFS_RANDOM_IFLUSH_3
)) {
3369 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3370 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3374 } else if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFDIR
) {
3376 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3377 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
) &&
3378 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_LOCAL
),
3379 mp
, XFS_ERRTAG_IFLUSH_4
, XFS_RANDOM_IFLUSH_4
)) {
3380 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3381 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3386 if (XFS_TEST_ERROR(ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
>
3387 ip
->i_d
.di_nblocks
, mp
, XFS_ERRTAG_IFLUSH_5
,
3388 XFS_RANDOM_IFLUSH_5
)) {
3389 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3390 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3392 ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
,
3397 if (XFS_TEST_ERROR(ip
->i_d
.di_forkoff
> mp
->m_sb
.sb_inodesize
,
3398 mp
, XFS_ERRTAG_IFLUSH_6
, XFS_RANDOM_IFLUSH_6
)) {
3399 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3400 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3401 ip
->i_ino
, ip
->i_d
.di_forkoff
, ip
);
3405 * bump the flush iteration count, used to detect flushes which
3406 * postdate a log record during recovery.
3409 ip
->i_d
.di_flushiter
++;
3412 * Copy the dirty parts of the inode into the on-disk
3413 * inode. We always copy out the core of the inode,
3414 * because if the inode is dirty at all the core must
3417 xfs_dinode_to_disk(&dip
->di_core
, &ip
->i_d
);
3419 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3420 if (ip
->i_d
.di_flushiter
== DI_MAX_FLUSH
)
3421 ip
->i_d
.di_flushiter
= 0;
3424 * If this is really an old format inode and the superblock version
3425 * has not been updated to support only new format inodes, then
3426 * convert back to the old inode format. If the superblock version
3427 * has been updated, then make the conversion permanent.
3429 ASSERT(ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
||
3430 XFS_SB_VERSION_HASNLINK(&mp
->m_sb
));
3431 if (ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
) {
3432 if (!XFS_SB_VERSION_HASNLINK(&mp
->m_sb
)) {
3436 ASSERT(ip
->i_d
.di_nlink
<= XFS_MAXLINK_1
);
3437 dip
->di_core
.di_onlink
= cpu_to_be16(ip
->i_d
.di_nlink
);
3440 * The superblock version has already been bumped,
3441 * so just make the conversion to the new inode
3444 ip
->i_d
.di_version
= XFS_DINODE_VERSION_2
;
3445 dip
->di_core
.di_version
= XFS_DINODE_VERSION_2
;
3446 ip
->i_d
.di_onlink
= 0;
3447 dip
->di_core
.di_onlink
= 0;
3448 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
3449 memset(&(dip
->di_core
.di_pad
[0]), 0,
3450 sizeof(dip
->di_core
.di_pad
));
3451 ASSERT(ip
->i_d
.di_projid
== 0);
3455 if (xfs_iflush_fork(ip
, dip
, iip
, XFS_DATA_FORK
, bp
) == EFSCORRUPTED
) {
3459 if (XFS_IFORK_Q(ip
)) {
3461 * The only error from xfs_iflush_fork is on the data fork.
3463 (void) xfs_iflush_fork(ip
, dip
, iip
, XFS_ATTR_FORK
, bp
);
3465 xfs_inobp_check(mp
, bp
);
3468 * We've recorded everything logged in the inode, so we'd
3469 * like to clear the ilf_fields bits so we don't log and
3470 * flush things unnecessarily. However, we can't stop
3471 * logging all this information until the data we've copied
3472 * into the disk buffer is written to disk. If we did we might
3473 * overwrite the copy of the inode in the log with all the
3474 * data after re-logging only part of it, and in the face of
3475 * a crash we wouldn't have all the data we need to recover.
3477 * What we do is move the bits to the ili_last_fields field.
3478 * When logging the inode, these bits are moved back to the
3479 * ilf_fields field. In the xfs_iflush_done() routine we
3480 * clear ili_last_fields, since we know that the information
3481 * those bits represent is permanently on disk. As long as
3482 * the flush completes before the inode is logged again, then
3483 * both ilf_fields and ili_last_fields will be cleared.
3485 * We can play with the ilf_fields bits here, because the inode
3486 * lock must be held exclusively in order to set bits there
3487 * and the flush lock protects the ili_last_fields bits.
3488 * Set ili_logged so the flush done
3489 * routine can tell whether or not to look in the AIL.
3490 * Also, store the current LSN of the inode so that we can tell
3491 * whether the item has moved in the AIL from xfs_iflush_done().
3492 * In order to read the lsn we need the AIL lock, because
3493 * it is a 64 bit value that cannot be read atomically.
3495 if (iip
!= NULL
&& iip
->ili_format
.ilf_fields
!= 0) {
3496 iip
->ili_last_fields
= iip
->ili_format
.ilf_fields
;
3497 iip
->ili_format
.ilf_fields
= 0;
3498 iip
->ili_logged
= 1;
3500 ASSERT(sizeof(xfs_lsn_t
) == 8); /* don't lock if it shrinks */
3501 spin_lock(&mp
->m_ail_lock
);
3502 iip
->ili_flush_lsn
= iip
->ili_item
.li_lsn
;
3503 spin_unlock(&mp
->m_ail_lock
);
3506 * Attach the function xfs_iflush_done to the inode's
3507 * buffer. This will remove the inode from the AIL
3508 * and unlock the inode's flush lock when the inode is
3509 * completely written to disk.
3511 xfs_buf_attach_iodone(bp
, (void(*)(xfs_buf_t
*,xfs_log_item_t
*))
3512 xfs_iflush_done
, (xfs_log_item_t
*)iip
);
3514 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) != NULL
);
3515 ASSERT(XFS_BUF_IODONE_FUNC(bp
) != NULL
);
3518 * We're flushing an inode which is not in the AIL and has
3519 * not been logged but has i_update_core set. For this
3520 * case we can use a B_DELWRI flush and immediately drop
3521 * the inode flush lock because we can avoid the whole
3522 * AIL state thing. It's OK to drop the flush lock now,
3523 * because we've already locked the buffer and to do anything
3524 * you really need both.
3527 ASSERT(iip
->ili_logged
== 0);
3528 ASSERT(iip
->ili_last_fields
== 0);
3529 ASSERT((iip
->ili_item
.li_flags
& XFS_LI_IN_AIL
) == 0);
3537 return XFS_ERROR(EFSCORRUPTED
);
3542 * Flush all inactive inodes in mp.
3552 XFS_MOUNT_ILOCK(mp
);
3558 /* Make sure we skip markers inserted by sync */
3559 if (ip
->i_mount
== NULL
) {
3564 vp
= XFS_ITOV_NULL(ip
);
3566 XFS_MOUNT_IUNLOCK(mp
);
3567 xfs_finish_reclaim(ip
, 0, XFS_IFLUSH_ASYNC
);
3571 ASSERT(vn_count(vp
) == 0);
3574 } while (ip
!= mp
->m_inodes
);
3576 XFS_MOUNT_IUNLOCK(mp
);
3580 * xfs_iaccess: check accessibility of inode for mode.
3589 mode_t orgmode
= mode
;
3590 struct inode
*inode
= vn_to_inode(XFS_ITOV(ip
));
3592 if (mode
& S_IWUSR
) {
3593 umode_t imode
= inode
->i_mode
;
3595 if (IS_RDONLY(inode
) &&
3596 (S_ISREG(imode
) || S_ISDIR(imode
) || S_ISLNK(imode
)))
3597 return XFS_ERROR(EROFS
);
3599 if (IS_IMMUTABLE(inode
))
3600 return XFS_ERROR(EACCES
);
3604 * If there's an Access Control List it's used instead of
3607 if ((error
= _ACL_XFS_IACCESS(ip
, mode
, cr
)) != -1)
3608 return error
? XFS_ERROR(error
) : 0;
3610 if (current_fsuid(cr
) != ip
->i_d
.di_uid
) {
3612 if (!in_group_p((gid_t
)ip
->i_d
.di_gid
))
3617 * If the DACs are ok we don't need any capability check.
3619 if ((ip
->i_d
.di_mode
& mode
) == mode
)
3622 * Read/write DACs are always overridable.
3623 * Executable DACs are overridable if at least one exec bit is set.
3625 if (!(orgmode
& S_IXUSR
) ||
3626 (inode
->i_mode
& S_IXUGO
) || S_ISDIR(inode
->i_mode
))
3627 if (capable_cred(cr
, CAP_DAC_OVERRIDE
))
3630 if ((orgmode
== S_IRUSR
) ||
3631 (S_ISDIR(inode
->i_mode
) && (!(orgmode
& S_IWUSR
)))) {
3632 if (capable_cred(cr
, CAP_DAC_READ_SEARCH
))
3635 cmn_err(CE_NOTE
, "Ick: mode=%o, orgmode=%o", mode
, orgmode
);
3637 return XFS_ERROR(EACCES
);
3639 return XFS_ERROR(EACCES
);
3643 * xfs_iroundup: round up argument to next power of two
3652 if ((v
& (v
- 1)) == 0)
3654 ASSERT((v
& 0x80000000) == 0);
3655 if ((v
& (v
+ 1)) == 0)
3657 for (i
= 0, m
= 1; i
< 31; i
++, m
<<= 1) {
3661 if ((v
& (v
+ 1)) == 0)
3668 #ifdef XFS_ILOCK_TRACE
3669 ktrace_t
*xfs_ilock_trace_buf
;
3672 xfs_ilock_trace(xfs_inode_t
*ip
, int lock
, unsigned int lockflags
, inst_t
*ra
)
3674 ktrace_enter(ip
->i_lock_trace
,
3676 (void *)(unsigned long)lock
, /* 1 = LOCK, 3=UNLOCK, etc */
3677 (void *)(unsigned long)lockflags
, /* XFS_ILOCK_EXCL etc */
3678 (void *)ra
, /* caller of ilock */
3679 (void *)(unsigned long)current_cpu(),
3680 (void *)(unsigned long)current_pid(),
3681 NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
);
3686 * Return a pointer to the extent record at file index idx.
3688 xfs_bmbt_rec_host_t
*
3690 xfs_ifork_t
*ifp
, /* inode fork pointer */
3691 xfs_extnum_t idx
) /* index of target extent */
3694 if ((ifp
->if_flags
& XFS_IFEXTIREC
) && (idx
== 0)) {
3695 return ifp
->if_u1
.if_ext_irec
->er_extbuf
;
3696 } else if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3697 xfs_ext_irec_t
*erp
; /* irec pointer */
3698 int erp_idx
= 0; /* irec index */
3699 xfs_extnum_t page_idx
= idx
; /* ext index in target list */
3701 erp
= xfs_iext_idx_to_irec(ifp
, &page_idx
, &erp_idx
, 0);
3702 return &erp
->er_extbuf
[page_idx
];
3703 } else if (ifp
->if_bytes
) {
3704 return &ifp
->if_u1
.if_extents
[idx
];
3711 * Insert new item(s) into the extent records for incore inode
3712 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3716 xfs_ifork_t
*ifp
, /* inode fork pointer */
3717 xfs_extnum_t idx
, /* starting index of new items */
3718 xfs_extnum_t count
, /* number of inserted items */
3719 xfs_bmbt_irec_t
*new) /* items to insert */
3721 xfs_extnum_t i
; /* extent record index */
3723 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
3724 xfs_iext_add(ifp
, idx
, count
);
3725 for (i
= idx
; i
< idx
+ count
; i
++, new++)
3726 xfs_bmbt_set_all(xfs_iext_get_ext(ifp
, i
), new);
3730 * This is called when the amount of space required for incore file
3731 * extents needs to be increased. The ext_diff parameter stores the
3732 * number of new extents being added and the idx parameter contains
3733 * the extent index where the new extents will be added. If the new
3734 * extents are being appended, then we just need to (re)allocate and
3735 * initialize the space. Otherwise, if the new extents are being
3736 * inserted into the middle of the existing entries, a bit more work
3737 * is required to make room for the new extents to be inserted. The
3738 * caller is responsible for filling in the new extent entries upon
3743 xfs_ifork_t
*ifp
, /* inode fork pointer */
3744 xfs_extnum_t idx
, /* index to begin adding exts */
3745 int ext_diff
) /* number of extents to add */
3747 int byte_diff
; /* new bytes being added */
3748 int new_size
; /* size of extents after adding */
3749 xfs_extnum_t nextents
; /* number of extents in file */
3751 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3752 ASSERT((idx
>= 0) && (idx
<= nextents
));
3753 byte_diff
= ext_diff
* sizeof(xfs_bmbt_rec_t
);
3754 new_size
= ifp
->if_bytes
+ byte_diff
;
3756 * If the new number of extents (nextents + ext_diff)
3757 * fits inside the inode, then continue to use the inline
3760 if (nextents
+ ext_diff
<= XFS_INLINE_EXTS
) {
3761 if (idx
< nextents
) {
3762 memmove(&ifp
->if_u2
.if_inline_ext
[idx
+ ext_diff
],
3763 &ifp
->if_u2
.if_inline_ext
[idx
],
3764 (nextents
- idx
) * sizeof(xfs_bmbt_rec_t
));
3765 memset(&ifp
->if_u2
.if_inline_ext
[idx
], 0, byte_diff
);
3767 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
3768 ifp
->if_real_bytes
= 0;
3769 ifp
->if_lastex
= nextents
+ ext_diff
;
3772 * Otherwise use a linear (direct) extent list.
3773 * If the extents are currently inside the inode,
3774 * xfs_iext_realloc_direct will switch us from
3775 * inline to direct extent allocation mode.
3777 else if (nextents
+ ext_diff
<= XFS_LINEAR_EXTS
) {
3778 xfs_iext_realloc_direct(ifp
, new_size
);
3779 if (idx
< nextents
) {
3780 memmove(&ifp
->if_u1
.if_extents
[idx
+ ext_diff
],
3781 &ifp
->if_u1
.if_extents
[idx
],
3782 (nextents
- idx
) * sizeof(xfs_bmbt_rec_t
));
3783 memset(&ifp
->if_u1
.if_extents
[idx
], 0, byte_diff
);
3786 /* Indirection array */
3788 xfs_ext_irec_t
*erp
;
3792 ASSERT(nextents
+ ext_diff
> XFS_LINEAR_EXTS
);
3793 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3794 erp
= xfs_iext_idx_to_irec(ifp
, &page_idx
, &erp_idx
, 1);
3796 xfs_iext_irec_init(ifp
);
3797 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3798 erp
= ifp
->if_u1
.if_ext_irec
;
3800 /* Extents fit in target extent page */
3801 if (erp
&& erp
->er_extcount
+ ext_diff
<= XFS_LINEAR_EXTS
) {
3802 if (page_idx
< erp
->er_extcount
) {
3803 memmove(&erp
->er_extbuf
[page_idx
+ ext_diff
],
3804 &erp
->er_extbuf
[page_idx
],
3805 (erp
->er_extcount
- page_idx
) *
3806 sizeof(xfs_bmbt_rec_t
));
3807 memset(&erp
->er_extbuf
[page_idx
], 0, byte_diff
);
3809 erp
->er_extcount
+= ext_diff
;
3810 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, ext_diff
);
3812 /* Insert a new extent page */
3814 xfs_iext_add_indirect_multi(ifp
,
3815 erp_idx
, page_idx
, ext_diff
);
3818 * If extent(s) are being appended to the last page in
3819 * the indirection array and the new extent(s) don't fit
3820 * in the page, then erp is NULL and erp_idx is set to
3821 * the next index needed in the indirection array.
3824 int count
= ext_diff
;
3827 erp
= xfs_iext_irec_new(ifp
, erp_idx
);
3828 erp
->er_extcount
= count
;
3829 count
-= MIN(count
, (int)XFS_LINEAR_EXTS
);
3836 ifp
->if_bytes
= new_size
;
3840 * This is called when incore extents are being added to the indirection
3841 * array and the new extents do not fit in the target extent list. The
3842 * erp_idx parameter contains the irec index for the target extent list
3843 * in the indirection array, and the idx parameter contains the extent
3844 * index within the list. The number of extents being added is stored
3845 * in the count parameter.
3847 * |-------| |-------|
3848 * | | | | idx - number of extents before idx
3850 * | | | | count - number of extents being inserted at idx
3851 * |-------| |-------|
3852 * | count | | nex2 | nex2 - number of extents after idx + count
3853 * |-------| |-------|
3856 xfs_iext_add_indirect_multi(
3857 xfs_ifork_t
*ifp
, /* inode fork pointer */
3858 int erp_idx
, /* target extent irec index */
3859 xfs_extnum_t idx
, /* index within target list */
3860 int count
) /* new extents being added */
3862 int byte_diff
; /* new bytes being added */
3863 xfs_ext_irec_t
*erp
; /* pointer to irec entry */
3864 xfs_extnum_t ext_diff
; /* number of extents to add */
3865 xfs_extnum_t ext_cnt
; /* new extents still needed */
3866 xfs_extnum_t nex2
; /* extents after idx + count */
3867 xfs_bmbt_rec_t
*nex2_ep
= NULL
; /* temp list for nex2 extents */
3868 int nlists
; /* number of irec's (lists) */
3870 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3871 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
3872 nex2
= erp
->er_extcount
- idx
;
3873 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3876 * Save second part of target extent list
3877 * (all extents past */
3879 byte_diff
= nex2
* sizeof(xfs_bmbt_rec_t
);
3880 nex2_ep
= (xfs_bmbt_rec_t
*) kmem_alloc(byte_diff
, KM_SLEEP
);
3881 memmove(nex2_ep
, &erp
->er_extbuf
[idx
], byte_diff
);
3882 erp
->er_extcount
-= nex2
;
3883 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, -nex2
);
3884 memset(&erp
->er_extbuf
[idx
], 0, byte_diff
);
3888 * Add the new extents to the end of the target
3889 * list, then allocate new irec record(s) and
3890 * extent buffer(s) as needed to store the rest
3891 * of the new extents.
3894 ext_diff
= MIN(ext_cnt
, (int)XFS_LINEAR_EXTS
- erp
->er_extcount
);
3896 erp
->er_extcount
+= ext_diff
;
3897 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, ext_diff
);
3898 ext_cnt
-= ext_diff
;
3902 erp
= xfs_iext_irec_new(ifp
, erp_idx
);
3903 ext_diff
= MIN(ext_cnt
, (int)XFS_LINEAR_EXTS
);
3904 erp
->er_extcount
= ext_diff
;
3905 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, ext_diff
);
3906 ext_cnt
-= ext_diff
;
3909 /* Add nex2 extents back to indirection array */
3911 xfs_extnum_t ext_avail
;
3914 byte_diff
= nex2
* sizeof(xfs_bmbt_rec_t
);
3915 ext_avail
= XFS_LINEAR_EXTS
- erp
->er_extcount
;
3918 * If nex2 extents fit in the current page, append
3919 * nex2_ep after the new extents.
3921 if (nex2
<= ext_avail
) {
3922 i
= erp
->er_extcount
;
3925 * Otherwise, check if space is available in the
3928 else if ((erp_idx
< nlists
- 1) &&
3929 (nex2
<= (ext_avail
= XFS_LINEAR_EXTS
-
3930 ifp
->if_u1
.if_ext_irec
[erp_idx
+1].er_extcount
))) {
3933 /* Create a hole for nex2 extents */
3934 memmove(&erp
->er_extbuf
[nex2
], erp
->er_extbuf
,
3935 erp
->er_extcount
* sizeof(xfs_bmbt_rec_t
));
3938 * Final choice, create a new extent page for
3943 erp
= xfs_iext_irec_new(ifp
, erp_idx
);
3945 memmove(&erp
->er_extbuf
[i
], nex2_ep
, byte_diff
);
3946 kmem_free(nex2_ep
, byte_diff
);
3947 erp
->er_extcount
+= nex2
;
3948 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, nex2
);
3953 * This is called when the amount of space required for incore file
3954 * extents needs to be decreased. The ext_diff parameter stores the
3955 * number of extents to be removed and the idx parameter contains
3956 * the extent index where the extents will be removed from.
3958 * If the amount of space needed has decreased below the linear
3959 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3960 * extent array. Otherwise, use kmem_realloc() to adjust the
3961 * size to what is needed.
3965 xfs_ifork_t
*ifp
, /* inode fork pointer */
3966 xfs_extnum_t idx
, /* index to begin removing exts */
3967 int ext_diff
) /* number of extents to remove */
3969 xfs_extnum_t nextents
; /* number of extents in file */
3970 int new_size
; /* size of extents after removal */
3972 ASSERT(ext_diff
> 0);
3973 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3974 new_size
= (nextents
- ext_diff
) * sizeof(xfs_bmbt_rec_t
);
3976 if (new_size
== 0) {
3977 xfs_iext_destroy(ifp
);
3978 } else if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3979 xfs_iext_remove_indirect(ifp
, idx
, ext_diff
);
3980 } else if (ifp
->if_real_bytes
) {
3981 xfs_iext_remove_direct(ifp
, idx
, ext_diff
);
3983 xfs_iext_remove_inline(ifp
, idx
, ext_diff
);
3985 ifp
->if_bytes
= new_size
;
3989 * This removes ext_diff extents from the inline buffer, beginning
3990 * at extent index idx.
3993 xfs_iext_remove_inline(
3994 xfs_ifork_t
*ifp
, /* inode fork pointer */
3995 xfs_extnum_t idx
, /* index to begin removing exts */
3996 int ext_diff
) /* number of extents to remove */
3998 int nextents
; /* number of extents in file */
4000 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
));
4001 ASSERT(idx
< XFS_INLINE_EXTS
);
4002 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4003 ASSERT(((nextents
- ext_diff
) > 0) &&
4004 (nextents
- ext_diff
) < XFS_INLINE_EXTS
);
4006 if (idx
+ ext_diff
< nextents
) {
4007 memmove(&ifp
->if_u2
.if_inline_ext
[idx
],
4008 &ifp
->if_u2
.if_inline_ext
[idx
+ ext_diff
],
4009 (nextents
- (idx
+ ext_diff
)) *
4010 sizeof(xfs_bmbt_rec_t
));
4011 memset(&ifp
->if_u2
.if_inline_ext
[nextents
- ext_diff
],
4012 0, ext_diff
* sizeof(xfs_bmbt_rec_t
));
4014 memset(&ifp
->if_u2
.if_inline_ext
[idx
], 0,
4015 ext_diff
* sizeof(xfs_bmbt_rec_t
));
4020 * This removes ext_diff extents from a linear (direct) extent list,
4021 * beginning at extent index idx. If the extents are being removed
4022 * from the end of the list (ie. truncate) then we just need to re-
4023 * allocate the list to remove the extra space. Otherwise, if the
4024 * extents are being removed from the middle of the existing extent
4025 * entries, then we first need to move the extent records beginning
4026 * at idx + ext_diff up in the list to overwrite the records being
4027 * removed, then remove the extra space via kmem_realloc.
4030 xfs_iext_remove_direct(
4031 xfs_ifork_t
*ifp
, /* inode fork pointer */
4032 xfs_extnum_t idx
, /* index to begin removing exts */
4033 int ext_diff
) /* number of extents to remove */
4035 xfs_extnum_t nextents
; /* number of extents in file */
4036 int new_size
; /* size of extents after removal */
4038 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
));
4039 new_size
= ifp
->if_bytes
-
4040 (ext_diff
* sizeof(xfs_bmbt_rec_t
));
4041 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4043 if (new_size
== 0) {
4044 xfs_iext_destroy(ifp
);
4047 /* Move extents up in the list (if needed) */
4048 if (idx
+ ext_diff
< nextents
) {
4049 memmove(&ifp
->if_u1
.if_extents
[idx
],
4050 &ifp
->if_u1
.if_extents
[idx
+ ext_diff
],
4051 (nextents
- (idx
+ ext_diff
)) *
4052 sizeof(xfs_bmbt_rec_t
));
4054 memset(&ifp
->if_u1
.if_extents
[nextents
- ext_diff
],
4055 0, ext_diff
* sizeof(xfs_bmbt_rec_t
));
4057 * Reallocate the direct extent list. If the extents
4058 * will fit inside the inode then xfs_iext_realloc_direct
4059 * will switch from direct to inline extent allocation
4062 xfs_iext_realloc_direct(ifp
, new_size
);
4063 ifp
->if_bytes
= new_size
;
4067 * This is called when incore extents are being removed from the
4068 * indirection array and the extents being removed span multiple extent
4069 * buffers. The idx parameter contains the file extent index where we
4070 * want to begin removing extents, and the count parameter contains
4071 * how many extents need to be removed.
4073 * |-------| |-------|
4074 * | nex1 | | | nex1 - number of extents before idx
4075 * |-------| | count |
4076 * | | | | count - number of extents being removed at idx
4077 * | count | |-------|
4078 * | | | nex2 | nex2 - number of extents after idx + count
4079 * |-------| |-------|
4082 xfs_iext_remove_indirect(
4083 xfs_ifork_t
*ifp
, /* inode fork pointer */
4084 xfs_extnum_t idx
, /* index to begin removing extents */
4085 int count
) /* number of extents to remove */
4087 xfs_ext_irec_t
*erp
; /* indirection array pointer */
4088 int erp_idx
= 0; /* indirection array index */
4089 xfs_extnum_t ext_cnt
; /* extents left to remove */
4090 xfs_extnum_t ext_diff
; /* extents to remove in current list */
4091 xfs_extnum_t nex1
; /* number of extents before idx */
4092 xfs_extnum_t nex2
; /* extents after idx + count */
4093 int nlists
; /* entries in indirection array */
4094 int page_idx
= idx
; /* index in target extent list */
4096 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4097 erp
= xfs_iext_idx_to_irec(ifp
, &page_idx
, &erp_idx
, 0);
4098 ASSERT(erp
!= NULL
);
4099 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4103 nex2
= MAX((erp
->er_extcount
- (nex1
+ ext_cnt
)), 0);
4104 ext_diff
= MIN(ext_cnt
, (erp
->er_extcount
- nex1
));
4106 * Check for deletion of entire list;
4107 * xfs_iext_irec_remove() updates extent offsets.
4109 if (ext_diff
== erp
->er_extcount
) {
4110 xfs_iext_irec_remove(ifp
, erp_idx
);
4111 ext_cnt
-= ext_diff
;
4114 ASSERT(erp_idx
< ifp
->if_real_bytes
/
4116 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4123 /* Move extents up (if needed) */
4125 memmove(&erp
->er_extbuf
[nex1
],
4126 &erp
->er_extbuf
[nex1
+ ext_diff
],
4127 nex2
* sizeof(xfs_bmbt_rec_t
));
4129 /* Zero out rest of page */
4130 memset(&erp
->er_extbuf
[nex1
+ nex2
], 0, (XFS_IEXT_BUFSZ
-
4131 ((nex1
+ nex2
) * sizeof(xfs_bmbt_rec_t
))));
4132 /* Update remaining counters */
4133 erp
->er_extcount
-= ext_diff
;
4134 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, -ext_diff
);
4135 ext_cnt
-= ext_diff
;
4140 ifp
->if_bytes
-= count
* sizeof(xfs_bmbt_rec_t
);
4141 xfs_iext_irec_compact(ifp
);
4145 * Create, destroy, or resize a linear (direct) block of extents.
4148 xfs_iext_realloc_direct(
4149 xfs_ifork_t
*ifp
, /* inode fork pointer */
4150 int new_size
) /* new size of extents */
4152 int rnew_size
; /* real new size of extents */
4154 rnew_size
= new_size
;
4156 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
) ||
4157 ((new_size
>= 0) && (new_size
<= XFS_IEXT_BUFSZ
) &&
4158 (new_size
!= ifp
->if_real_bytes
)));
4160 /* Free extent records */
4161 if (new_size
== 0) {
4162 xfs_iext_destroy(ifp
);
4164 /* Resize direct extent list and zero any new bytes */
4165 else if (ifp
->if_real_bytes
) {
4166 /* Check if extents will fit inside the inode */
4167 if (new_size
<= XFS_INLINE_EXTS
* sizeof(xfs_bmbt_rec_t
)) {
4168 xfs_iext_direct_to_inline(ifp
, new_size
/
4169 (uint
)sizeof(xfs_bmbt_rec_t
));
4170 ifp
->if_bytes
= new_size
;
4173 if (!is_power_of_2(new_size
)){
4174 rnew_size
= xfs_iroundup(new_size
);
4176 if (rnew_size
!= ifp
->if_real_bytes
) {
4177 ifp
->if_u1
.if_extents
=
4178 kmem_realloc(ifp
->if_u1
.if_extents
,
4183 if (rnew_size
> ifp
->if_real_bytes
) {
4184 memset(&ifp
->if_u1
.if_extents
[ifp
->if_bytes
/
4185 (uint
)sizeof(xfs_bmbt_rec_t
)], 0,
4186 rnew_size
- ifp
->if_real_bytes
);
4190 * Switch from the inline extent buffer to a direct
4191 * extent list. Be sure to include the inline extent
4192 * bytes in new_size.
4195 new_size
+= ifp
->if_bytes
;
4196 if (!is_power_of_2(new_size
)) {
4197 rnew_size
= xfs_iroundup(new_size
);
4199 xfs_iext_inline_to_direct(ifp
, rnew_size
);
4201 ifp
->if_real_bytes
= rnew_size
;
4202 ifp
->if_bytes
= new_size
;
4206 * Switch from linear (direct) extent records to inline buffer.
4209 xfs_iext_direct_to_inline(
4210 xfs_ifork_t
*ifp
, /* inode fork pointer */
4211 xfs_extnum_t nextents
) /* number of extents in file */
4213 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
4214 ASSERT(nextents
<= XFS_INLINE_EXTS
);
4216 * The inline buffer was zeroed when we switched
4217 * from inline to direct extent allocation mode,
4218 * so we don't need to clear it here.
4220 memcpy(ifp
->if_u2
.if_inline_ext
, ifp
->if_u1
.if_extents
,
4221 nextents
* sizeof(xfs_bmbt_rec_t
));
4222 kmem_free(ifp
->if_u1
.if_extents
, ifp
->if_real_bytes
);
4223 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
4224 ifp
->if_real_bytes
= 0;
4228 * Switch from inline buffer to linear (direct) extent records.
4229 * new_size should already be rounded up to the next power of 2
4230 * by the caller (when appropriate), so use new_size as it is.
4231 * However, since new_size may be rounded up, we can't update
4232 * if_bytes here. It is the caller's responsibility to update
4233 * if_bytes upon return.
4236 xfs_iext_inline_to_direct(
4237 xfs_ifork_t
*ifp
, /* inode fork pointer */
4238 int new_size
) /* number of extents in file */
4240 ifp
->if_u1
.if_extents
= kmem_alloc(new_size
, KM_SLEEP
);
4241 memset(ifp
->if_u1
.if_extents
, 0, new_size
);
4242 if (ifp
->if_bytes
) {
4243 memcpy(ifp
->if_u1
.if_extents
, ifp
->if_u2
.if_inline_ext
,
4245 memset(ifp
->if_u2
.if_inline_ext
, 0, XFS_INLINE_EXTS
*
4246 sizeof(xfs_bmbt_rec_t
));
4248 ifp
->if_real_bytes
= new_size
;
4252 * Resize an extent indirection array to new_size bytes.
4255 xfs_iext_realloc_indirect(
4256 xfs_ifork_t
*ifp
, /* inode fork pointer */
4257 int new_size
) /* new indirection array size */
4259 int nlists
; /* number of irec's (ex lists) */
4260 int size
; /* current indirection array size */
4262 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4263 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4264 size
= nlists
* sizeof(xfs_ext_irec_t
);
4265 ASSERT(ifp
->if_real_bytes
);
4266 ASSERT((new_size
>= 0) && (new_size
!= size
));
4267 if (new_size
== 0) {
4268 xfs_iext_destroy(ifp
);
4270 ifp
->if_u1
.if_ext_irec
= (xfs_ext_irec_t
*)
4271 kmem_realloc(ifp
->if_u1
.if_ext_irec
,
4272 new_size
, size
, KM_SLEEP
);
4277 * Switch from indirection array to linear (direct) extent allocations.
4280 xfs_iext_indirect_to_direct(
4281 xfs_ifork_t
*ifp
) /* inode fork pointer */
4283 xfs_bmbt_rec_host_t
*ep
; /* extent record pointer */
4284 xfs_extnum_t nextents
; /* number of extents in file */
4285 int size
; /* size of file extents */
4287 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4288 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4289 ASSERT(nextents
<= XFS_LINEAR_EXTS
);
4290 size
= nextents
* sizeof(xfs_bmbt_rec_t
);
4292 xfs_iext_irec_compact_full(ifp
);
4293 ASSERT(ifp
->if_real_bytes
== XFS_IEXT_BUFSZ
);
4295 ep
= ifp
->if_u1
.if_ext_irec
->er_extbuf
;
4296 kmem_free(ifp
->if_u1
.if_ext_irec
, sizeof(xfs_ext_irec_t
));
4297 ifp
->if_flags
&= ~XFS_IFEXTIREC
;
4298 ifp
->if_u1
.if_extents
= ep
;
4299 ifp
->if_bytes
= size
;
4300 if (nextents
< XFS_LINEAR_EXTS
) {
4301 xfs_iext_realloc_direct(ifp
, size
);
4306 * Free incore file extents.
4310 xfs_ifork_t
*ifp
) /* inode fork pointer */
4312 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
4316 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4317 for (erp_idx
= nlists
- 1; erp_idx
>= 0 ; erp_idx
--) {
4318 xfs_iext_irec_remove(ifp
, erp_idx
);
4320 ifp
->if_flags
&= ~XFS_IFEXTIREC
;
4321 } else if (ifp
->if_real_bytes
) {
4322 kmem_free(ifp
->if_u1
.if_extents
, ifp
->if_real_bytes
);
4323 } else if (ifp
->if_bytes
) {
4324 memset(ifp
->if_u2
.if_inline_ext
, 0, XFS_INLINE_EXTS
*
4325 sizeof(xfs_bmbt_rec_t
));
4327 ifp
->if_u1
.if_extents
= NULL
;
4328 ifp
->if_real_bytes
= 0;
4333 * Return a pointer to the extent record for file system block bno.
4335 xfs_bmbt_rec_host_t
* /* pointer to found extent record */
4336 xfs_iext_bno_to_ext(
4337 xfs_ifork_t
*ifp
, /* inode fork pointer */
4338 xfs_fileoff_t bno
, /* block number to search for */
4339 xfs_extnum_t
*idxp
) /* index of target extent */
4341 xfs_bmbt_rec_host_t
*base
; /* pointer to first extent */
4342 xfs_filblks_t blockcount
= 0; /* number of blocks in extent */
4343 xfs_bmbt_rec_host_t
*ep
= NULL
; /* pointer to target extent */
4344 xfs_ext_irec_t
*erp
= NULL
; /* indirection array pointer */
4345 int high
; /* upper boundary in search */
4346 xfs_extnum_t idx
= 0; /* index of target extent */
4347 int low
; /* lower boundary in search */
4348 xfs_extnum_t nextents
; /* number of file extents */
4349 xfs_fileoff_t startoff
= 0; /* start offset of extent */
4351 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4352 if (nextents
== 0) {
4357 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
4358 /* Find target extent list */
4360 erp
= xfs_iext_bno_to_irec(ifp
, bno
, &erp_idx
);
4361 base
= erp
->er_extbuf
;
4362 high
= erp
->er_extcount
- 1;
4364 base
= ifp
->if_u1
.if_extents
;
4365 high
= nextents
- 1;
4367 /* Binary search extent records */
4368 while (low
<= high
) {
4369 idx
= (low
+ high
) >> 1;
4371 startoff
= xfs_bmbt_get_startoff(ep
);
4372 blockcount
= xfs_bmbt_get_blockcount(ep
);
4373 if (bno
< startoff
) {
4375 } else if (bno
>= startoff
+ blockcount
) {
4378 /* Convert back to file-based extent index */
4379 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
4380 idx
+= erp
->er_extoff
;
4386 /* Convert back to file-based extent index */
4387 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
4388 idx
+= erp
->er_extoff
;
4390 if (bno
>= startoff
+ blockcount
) {
4391 if (++idx
== nextents
) {
4394 ep
= xfs_iext_get_ext(ifp
, idx
);
4402 * Return a pointer to the indirection array entry containing the
4403 * extent record for filesystem block bno. Store the index of the
4404 * target irec in *erp_idxp.
4406 xfs_ext_irec_t
* /* pointer to found extent record */
4407 xfs_iext_bno_to_irec(
4408 xfs_ifork_t
*ifp
, /* inode fork pointer */
4409 xfs_fileoff_t bno
, /* block number to search for */
4410 int *erp_idxp
) /* irec index of target ext list */
4412 xfs_ext_irec_t
*erp
= NULL
; /* indirection array pointer */
4413 xfs_ext_irec_t
*erp_next
; /* next indirection array entry */
4414 int erp_idx
; /* indirection array index */
4415 int nlists
; /* number of extent irec's (lists) */
4416 int high
; /* binary search upper limit */
4417 int low
; /* binary search lower limit */
4419 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4420 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4424 while (low
<= high
) {
4425 erp_idx
= (low
+ high
) >> 1;
4426 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4427 erp_next
= erp_idx
< nlists
- 1 ? erp
+ 1 : NULL
;
4428 if (bno
< xfs_bmbt_get_startoff(erp
->er_extbuf
)) {
4430 } else if (erp_next
&& bno
>=
4431 xfs_bmbt_get_startoff(erp_next
->er_extbuf
)) {
4437 *erp_idxp
= erp_idx
;
4442 * Return a pointer to the indirection array entry containing the
4443 * extent record at file extent index *idxp. Store the index of the
4444 * target irec in *erp_idxp and store the page index of the target
4445 * extent record in *idxp.
4448 xfs_iext_idx_to_irec(
4449 xfs_ifork_t
*ifp
, /* inode fork pointer */
4450 xfs_extnum_t
*idxp
, /* extent index (file -> page) */
4451 int *erp_idxp
, /* pointer to target irec */
4452 int realloc
) /* new bytes were just added */
4454 xfs_ext_irec_t
*prev
; /* pointer to previous irec */
4455 xfs_ext_irec_t
*erp
= NULL
; /* pointer to current irec */
4456 int erp_idx
; /* indirection array index */
4457 int nlists
; /* number of irec's (ex lists) */
4458 int high
; /* binary search upper limit */
4459 int low
; /* binary search lower limit */
4460 xfs_extnum_t page_idx
= *idxp
; /* extent index in target list */
4462 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4463 ASSERT(page_idx
>= 0 && page_idx
<=
4464 ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
));
4465 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4470 /* Binary search extent irec's */
4471 while (low
<= high
) {
4472 erp_idx
= (low
+ high
) >> 1;
4473 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4474 prev
= erp_idx
> 0 ? erp
- 1 : NULL
;
4475 if (page_idx
< erp
->er_extoff
|| (page_idx
== erp
->er_extoff
&&
4476 realloc
&& prev
&& prev
->er_extcount
< XFS_LINEAR_EXTS
)) {
4478 } else if (page_idx
> erp
->er_extoff
+ erp
->er_extcount
||
4479 (page_idx
== erp
->er_extoff
+ erp
->er_extcount
&&
4482 } else if (page_idx
== erp
->er_extoff
+ erp
->er_extcount
&&
4483 erp
->er_extcount
== XFS_LINEAR_EXTS
) {
4487 erp
= erp_idx
< nlists
? erp
+ 1 : NULL
;
4490 page_idx
-= erp
->er_extoff
;
4495 *erp_idxp
= erp_idx
;
4500 * Allocate and initialize an indirection array once the space needed
4501 * for incore extents increases above XFS_IEXT_BUFSZ.
4505 xfs_ifork_t
*ifp
) /* inode fork pointer */
4507 xfs_ext_irec_t
*erp
; /* indirection array pointer */
4508 xfs_extnum_t nextents
; /* number of extents in file */
4510 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
));
4511 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4512 ASSERT(nextents
<= XFS_LINEAR_EXTS
);
4514 erp
= (xfs_ext_irec_t
*)
4515 kmem_alloc(sizeof(xfs_ext_irec_t
), KM_SLEEP
);
4517 if (nextents
== 0) {
4518 ifp
->if_u1
.if_extents
= kmem_alloc(XFS_IEXT_BUFSZ
, KM_SLEEP
);
4519 } else if (!ifp
->if_real_bytes
) {
4520 xfs_iext_inline_to_direct(ifp
, XFS_IEXT_BUFSZ
);
4521 } else if (ifp
->if_real_bytes
< XFS_IEXT_BUFSZ
) {
4522 xfs_iext_realloc_direct(ifp
, XFS_IEXT_BUFSZ
);
4524 erp
->er_extbuf
= ifp
->if_u1
.if_extents
;
4525 erp
->er_extcount
= nextents
;
4528 ifp
->if_flags
|= XFS_IFEXTIREC
;
4529 ifp
->if_real_bytes
= XFS_IEXT_BUFSZ
;
4530 ifp
->if_bytes
= nextents
* sizeof(xfs_bmbt_rec_t
);
4531 ifp
->if_u1
.if_ext_irec
= erp
;
4537 * Allocate and initialize a new entry in the indirection array.
4541 xfs_ifork_t
*ifp
, /* inode fork pointer */
4542 int erp_idx
) /* index for new irec */
4544 xfs_ext_irec_t
*erp
; /* indirection array pointer */
4545 int i
; /* loop counter */
4546 int nlists
; /* number of irec's (ex lists) */
4548 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4549 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4551 /* Resize indirection array */
4552 xfs_iext_realloc_indirect(ifp
, ++nlists
*
4553 sizeof(xfs_ext_irec_t
));
4555 * Move records down in the array so the
4556 * new page can use erp_idx.
4558 erp
= ifp
->if_u1
.if_ext_irec
;
4559 for (i
= nlists
- 1; i
> erp_idx
; i
--) {
4560 memmove(&erp
[i
], &erp
[i
-1], sizeof(xfs_ext_irec_t
));
4562 ASSERT(i
== erp_idx
);
4564 /* Initialize new extent record */
4565 erp
= ifp
->if_u1
.if_ext_irec
;
4566 erp
[erp_idx
].er_extbuf
= kmem_alloc(XFS_IEXT_BUFSZ
, KM_SLEEP
);
4567 ifp
->if_real_bytes
= nlists
* XFS_IEXT_BUFSZ
;
4568 memset(erp
[erp_idx
].er_extbuf
, 0, XFS_IEXT_BUFSZ
);
4569 erp
[erp_idx
].er_extcount
= 0;
4570 erp
[erp_idx
].er_extoff
= erp_idx
> 0 ?
4571 erp
[erp_idx
-1].er_extoff
+ erp
[erp_idx
-1].er_extcount
: 0;
4572 return (&erp
[erp_idx
]);
4576 * Remove a record from the indirection array.
4579 xfs_iext_irec_remove(
4580 xfs_ifork_t
*ifp
, /* inode fork pointer */
4581 int erp_idx
) /* irec index to remove */
4583 xfs_ext_irec_t
*erp
; /* indirection array pointer */
4584 int i
; /* loop counter */
4585 int nlists
; /* number of irec's (ex lists) */
4587 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4588 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4589 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4590 if (erp
->er_extbuf
) {
4591 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1,
4593 kmem_free(erp
->er_extbuf
, XFS_IEXT_BUFSZ
);
4595 /* Compact extent records */
4596 erp
= ifp
->if_u1
.if_ext_irec
;
4597 for (i
= erp_idx
; i
< nlists
- 1; i
++) {
4598 memmove(&erp
[i
], &erp
[i
+1], sizeof(xfs_ext_irec_t
));
4601 * Manually free the last extent record from the indirection
4602 * array. A call to xfs_iext_realloc_indirect() with a size
4603 * of zero would result in a call to xfs_iext_destroy() which
4604 * would in turn call this function again, creating a nasty
4608 xfs_iext_realloc_indirect(ifp
,
4609 nlists
* sizeof(xfs_ext_irec_t
));
4611 kmem_free(ifp
->if_u1
.if_ext_irec
,
4612 sizeof(xfs_ext_irec_t
));
4614 ifp
->if_real_bytes
= nlists
* XFS_IEXT_BUFSZ
;
4618 * This is called to clean up large amounts of unused memory allocated
4619 * by the indirection array. Before compacting anything though, verify
4620 * that the indirection array is still needed and switch back to the
4621 * linear extent list (or even the inline buffer) if possible. The
4622 * compaction policy is as follows:
4624 * Full Compaction: Extents fit into a single page (or inline buffer)
4625 * Full Compaction: Extents occupy less than 10% of allocated space
4626 * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
4627 * No Compaction: Extents occupy at least 50% of allocated space
4630 xfs_iext_irec_compact(
4631 xfs_ifork_t
*ifp
) /* inode fork pointer */
4633 xfs_extnum_t nextents
; /* number of extents in file */
4634 int nlists
; /* number of irec's (ex lists) */
4636 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4637 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4638 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4640 if (nextents
== 0) {
4641 xfs_iext_destroy(ifp
);
4642 } else if (nextents
<= XFS_INLINE_EXTS
) {
4643 xfs_iext_indirect_to_direct(ifp
);
4644 xfs_iext_direct_to_inline(ifp
, nextents
);
4645 } else if (nextents
<= XFS_LINEAR_EXTS
) {
4646 xfs_iext_indirect_to_direct(ifp
);
4647 } else if (nextents
< (nlists
* XFS_LINEAR_EXTS
) >> 3) {
4648 xfs_iext_irec_compact_full(ifp
);
4649 } else if (nextents
< (nlists
* XFS_LINEAR_EXTS
) >> 1) {
4650 xfs_iext_irec_compact_pages(ifp
);
4655 * Combine extents from neighboring extent pages.
4658 xfs_iext_irec_compact_pages(
4659 xfs_ifork_t
*ifp
) /* inode fork pointer */
4661 xfs_ext_irec_t
*erp
, *erp_next
;/* pointers to irec entries */
4662 int erp_idx
= 0; /* indirection array index */
4663 int nlists
; /* number of irec's (ex lists) */
4665 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4666 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4667 while (erp_idx
< nlists
- 1) {
4668 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4670 if (erp_next
->er_extcount
<=
4671 (XFS_LINEAR_EXTS
- erp
->er_extcount
)) {
4672 memmove(&erp
->er_extbuf
[erp
->er_extcount
],
4673 erp_next
->er_extbuf
, erp_next
->er_extcount
*
4674 sizeof(xfs_bmbt_rec_t
));
4675 erp
->er_extcount
+= erp_next
->er_extcount
;
4677 * Free page before removing extent record
4678 * so er_extoffs don't get modified in
4679 * xfs_iext_irec_remove.
4681 kmem_free(erp_next
->er_extbuf
, XFS_IEXT_BUFSZ
);
4682 erp_next
->er_extbuf
= NULL
;
4683 xfs_iext_irec_remove(ifp
, erp_idx
+ 1);
4684 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4692 * Fully compact the extent records managed by the indirection array.
4695 xfs_iext_irec_compact_full(
4696 xfs_ifork_t
*ifp
) /* inode fork pointer */
4698 xfs_bmbt_rec_host_t
*ep
, *ep_next
; /* extent record pointers */
4699 xfs_ext_irec_t
*erp
, *erp_next
; /* extent irec pointers */
4700 int erp_idx
= 0; /* extent irec index */
4701 int ext_avail
; /* empty entries in ex list */
4702 int ext_diff
; /* number of exts to add */
4703 int nlists
; /* number of irec's (ex lists) */
4705 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4706 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4707 erp
= ifp
->if_u1
.if_ext_irec
;
4708 ep
= &erp
->er_extbuf
[erp
->er_extcount
];
4710 ep_next
= erp_next
->er_extbuf
;
4711 while (erp_idx
< nlists
- 1) {
4712 ext_avail
= XFS_LINEAR_EXTS
- erp
->er_extcount
;
4713 ext_diff
= MIN(ext_avail
, erp_next
->er_extcount
);
4714 memcpy(ep
, ep_next
, ext_diff
* sizeof(xfs_bmbt_rec_t
));
4715 erp
->er_extcount
+= ext_diff
;
4716 erp_next
->er_extcount
-= ext_diff
;
4717 /* Remove next page */
4718 if (erp_next
->er_extcount
== 0) {
4720 * Free page before removing extent record
4721 * so er_extoffs don't get modified in
4722 * xfs_iext_irec_remove.
4724 kmem_free(erp_next
->er_extbuf
,
4725 erp_next
->er_extcount
* sizeof(xfs_bmbt_rec_t
));
4726 erp_next
->er_extbuf
= NULL
;
4727 xfs_iext_irec_remove(ifp
, erp_idx
+ 1);
4728 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4729 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4730 /* Update next page */
4732 /* Move rest of page up to become next new page */
4733 memmove(erp_next
->er_extbuf
, ep_next
,
4734 erp_next
->er_extcount
* sizeof(xfs_bmbt_rec_t
));
4735 ep_next
= erp_next
->er_extbuf
;
4736 memset(&ep_next
[erp_next
->er_extcount
], 0,
4737 (XFS_LINEAR_EXTS
- erp_next
->er_extcount
) *
4738 sizeof(xfs_bmbt_rec_t
));
4740 if (erp
->er_extcount
== XFS_LINEAR_EXTS
) {
4742 if (erp_idx
< nlists
)
4743 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4747 ep
= &erp
->er_extbuf
[erp
->er_extcount
];
4749 ep_next
= erp_next
->er_extbuf
;
4754 * This is called to update the er_extoff field in the indirection
4755 * array when extents have been added or removed from one of the
4756 * extent lists. erp_idx contains the irec index to begin updating
4757 * at and ext_diff contains the number of extents that were added
4761 xfs_iext_irec_update_extoffs(
4762 xfs_ifork_t
*ifp
, /* inode fork pointer */
4763 int erp_idx
, /* irec index to update */
4764 int ext_diff
) /* number of new extents */
4766 int i
; /* loop counter */
4767 int nlists
; /* number of irec's (ex lists */
4769 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4770 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4771 for (i
= erp_idx
; i
< nlists
; i
++) {
4772 ifp
->if_u1
.if_ext_irec
[i
].er_extoff
+= ext_diff
;