2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
25 #include "xfs_trans.h"
26 #include "xfs_trans_priv.h"
30 #include "xfs_dmapi.h"
31 #include "xfs_mount.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_ialloc_btree.h"
35 #include "xfs_dir2_sf.h"
36 #include "xfs_attr_sf.h"
37 #include "xfs_dinode.h"
38 #include "xfs_inode.h"
39 #include "xfs_buf_item.h"
40 #include "xfs_inode_item.h"
41 #include "xfs_btree.h"
42 #include "xfs_alloc.h"
43 #include "xfs_ialloc.h"
46 #include "xfs_error.h"
47 #include "xfs_utils.h"
48 #include "xfs_dir2_trace.h"
49 #include "xfs_quota.h"
51 #include "xfs_filestream.h"
53 kmem_zone_t
*xfs_ifork_zone
;
54 kmem_zone_t
*xfs_inode_zone
;
55 kmem_zone_t
*xfs_chashlist_zone
;
58 * Used in xfs_itruncate(). This is the maximum number of extents
59 * freed from a file in a single transaction.
61 #define XFS_ITRUNC_MAX_EXTENTS 2
63 STATIC
int xfs_iflush_int(xfs_inode_t
*, xfs_buf_t
*);
64 STATIC
int xfs_iformat_local(xfs_inode_t
*, xfs_dinode_t
*, int, int);
65 STATIC
int xfs_iformat_extents(xfs_inode_t
*, xfs_dinode_t
*, int);
66 STATIC
int xfs_iformat_btree(xfs_inode_t
*, xfs_dinode_t
*, int);
70 * Make sure that the extents in the given memory buffer
80 xfs_bmbt_rec_host_t rec
;
83 for (i
= 0; i
< nrecs
; i
++) {
84 xfs_bmbt_rec_host_t
*ep
= xfs_iext_get_ext(ifp
, i
);
85 rec
.l0
= get_unaligned(&ep
->l0
);
86 rec
.l1
= get_unaligned(&ep
->l1
);
87 xfs_bmbt_get_all(&rec
, &irec
);
88 if (fmt
== XFS_EXTFMT_NOSTATE
)
89 ASSERT(irec
.br_state
== XFS_EXT_NORM
);
93 #define xfs_validate_extents(ifp, nrecs, fmt)
97 * Check that none of the inode's in the buffer have a next
98 * unlinked field of 0.
110 j
= mp
->m_inode_cluster_size
>> mp
->m_sb
.sb_inodelog
;
112 for (i
= 0; i
< j
; i
++) {
113 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
,
114 i
* mp
->m_sb
.sb_inodesize
);
115 if (!dip
->di_next_unlinked
) {
116 xfs_fs_cmn_err(CE_ALERT
, mp
,
117 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
119 ASSERT(dip
->di_next_unlinked
);
126 * This routine is called to map an inode number within a file
127 * system to the buffer containing the on-disk version of the
128 * inode. It returns a pointer to the buffer containing the
129 * on-disk inode in the bpp parameter, and in the dip parameter
130 * it returns a pointer to the on-disk inode within that buffer.
132 * If a non-zero error is returned, then the contents of bpp and
133 * dipp are undefined.
135 * Use xfs_imap() to determine the size and location of the
136 * buffer to read from disk.
154 * Call the space management code to find the location of the
158 error
= xfs_imap(mp
, tp
, ino
, &imap
, XFS_IMAP_LOOKUP
);
161 "xfs_inotobp: xfs_imap() returned an "
162 "error %d on %s. Returning error.", error
, mp
->m_fsname
);
167 * If the inode number maps to a block outside the bounds of the
168 * file system then return NULL rather than calling read_buf
169 * and panicing when we get an error from the driver.
171 if ((imap
.im_blkno
+ imap
.im_len
) >
172 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
)) {
174 "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds "
175 "of the file system %s. Returning EINVAL.",
176 (unsigned long long)imap
.im_blkno
,
177 imap
.im_len
, mp
->m_fsname
);
178 return XFS_ERROR(EINVAL
);
182 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
183 * default to just a read_buf() call.
185 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, imap
.im_blkno
,
186 (int)imap
.im_len
, XFS_BUF_LOCK
, &bp
);
190 "xfs_inotobp: xfs_trans_read_buf() returned an "
191 "error %d on %s. Returning error.", error
, mp
->m_fsname
);
194 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, 0);
196 INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
) == XFS_DINODE_MAGIC
&&
197 XFS_DINODE_GOOD_VERSION(INT_GET(dip
->di_core
.di_version
, ARCH_CONVERT
));
198 if (unlikely(XFS_TEST_ERROR(!di_ok
, mp
, XFS_ERRTAG_ITOBP_INOTOBP
,
199 XFS_RANDOM_ITOBP_INOTOBP
))) {
200 XFS_CORRUPTION_ERROR("xfs_inotobp", XFS_ERRLEVEL_LOW
, mp
, dip
);
201 xfs_trans_brelse(tp
, bp
);
203 "xfs_inotobp: XFS_TEST_ERROR() returned an "
204 "error on %s. Returning EFSCORRUPTED.", mp
->m_fsname
);
205 return XFS_ERROR(EFSCORRUPTED
);
208 xfs_inobp_check(mp
, bp
);
211 * Set *dipp to point to the on-disk inode in the buffer.
213 *dipp
= (xfs_dinode_t
*)xfs_buf_offset(bp
, imap
.im_boffset
);
215 *offset
= imap
.im_boffset
;
221 * This routine is called to map an inode to the buffer containing
222 * the on-disk version of the inode. It returns a pointer to the
223 * buffer containing the on-disk inode in the bpp parameter, and in
224 * the dip parameter it returns a pointer to the on-disk inode within
227 * If a non-zero error is returned, then the contents of bpp and
228 * dipp are undefined.
230 * If the inode is new and has not yet been initialized, use xfs_imap()
231 * to determine the size and location of the buffer to read from disk.
232 * If the inode has already been mapped to its buffer and read in once,
233 * then use the mapping information stored in the inode rather than
234 * calling xfs_imap(). This allows us to avoid the overhead of looking
235 * at the inode btree for small block file systems (see xfs_dilocate()).
236 * We can tell whether the inode has been mapped in before by comparing
237 * its disk block address to 0. Only uninitialized inodes will have
238 * 0 for the disk block address.
256 if (ip
->i_blkno
== (xfs_daddr_t
)0) {
258 * Call the space management code to find the location of the
262 if ((error
= xfs_imap(mp
, tp
, ip
->i_ino
, &imap
,
263 XFS_IMAP_LOOKUP
| imap_flags
)))
267 * If the inode number maps to a block outside the bounds
268 * of the file system then return NULL rather than calling
269 * read_buf and panicing when we get an error from the
272 if ((imap
.im_blkno
+ imap
.im_len
) >
273 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
)) {
275 xfs_fs_cmn_err(CE_ALERT
, mp
, "xfs_itobp: "
276 "(imap.im_blkno (0x%llx) "
277 "+ imap.im_len (0x%llx)) > "
278 " XFS_FSB_TO_BB(mp, "
279 "mp->m_sb.sb_dblocks) (0x%llx)",
280 (unsigned long long) imap
.im_blkno
,
281 (unsigned long long) imap
.im_len
,
282 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
));
284 return XFS_ERROR(EINVAL
);
288 * Fill in the fields in the inode that will be used to
289 * map the inode to its buffer from now on.
291 ip
->i_blkno
= imap
.im_blkno
;
292 ip
->i_len
= imap
.im_len
;
293 ip
->i_boffset
= imap
.im_boffset
;
296 * We've already mapped the inode once, so just use the
297 * mapping that we saved the first time.
299 imap
.im_blkno
= ip
->i_blkno
;
300 imap
.im_len
= ip
->i_len
;
301 imap
.im_boffset
= ip
->i_boffset
;
303 ASSERT(bno
== 0 || bno
== imap
.im_blkno
);
306 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
307 * default to just a read_buf() call.
309 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, imap
.im_blkno
,
310 (int)imap
.im_len
, XFS_BUF_LOCK
, &bp
);
313 xfs_fs_cmn_err(CE_ALERT
, mp
, "xfs_itobp: "
314 "xfs_trans_read_buf() returned error %d, "
315 "imap.im_blkno 0x%llx, imap.im_len 0x%llx",
316 error
, (unsigned long long) imap
.im_blkno
,
317 (unsigned long long) imap
.im_len
);
323 * Validate the magic number and version of every inode in the buffer
324 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
325 * No validation is done here in userspace (xfs_repair).
327 #if !defined(__KERNEL__)
330 ni
= BBTOB(imap
.im_len
) >> mp
->m_sb
.sb_inodelog
;
331 #else /* usual case */
335 for (i
= 0; i
< ni
; i
++) {
339 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
,
340 (i
<< mp
->m_sb
.sb_inodelog
));
341 di_ok
= INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
) == XFS_DINODE_MAGIC
&&
342 XFS_DINODE_GOOD_VERSION(INT_GET(dip
->di_core
.di_version
, ARCH_CONVERT
));
343 if (unlikely(XFS_TEST_ERROR(!di_ok
, mp
,
344 XFS_ERRTAG_ITOBP_INOTOBP
,
345 XFS_RANDOM_ITOBP_INOTOBP
))) {
346 if (imap_flags
& XFS_IMAP_BULKSTAT
) {
347 xfs_trans_brelse(tp
, bp
);
348 return XFS_ERROR(EINVAL
);
352 "Device %s - bad inode magic/vsn "
353 "daddr %lld #%d (magic=%x)",
354 XFS_BUFTARG_NAME(mp
->m_ddev_targp
),
355 (unsigned long long)imap
.im_blkno
, i
,
356 INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
));
358 XFS_CORRUPTION_ERROR("xfs_itobp", XFS_ERRLEVEL_HIGH
,
360 xfs_trans_brelse(tp
, bp
);
361 return XFS_ERROR(EFSCORRUPTED
);
365 xfs_inobp_check(mp
, bp
);
368 * Mark the buffer as an inode buffer now that it looks good
370 XFS_BUF_SET_VTYPE(bp
, B_FS_INO
);
373 * Set *dipp to point to the on-disk inode in the buffer.
375 *dipp
= (xfs_dinode_t
*)xfs_buf_offset(bp
, imap
.im_boffset
);
381 * Move inode type and inode format specific information from the
382 * on-disk inode to the in-core inode. For fifos, devs, and sockets
383 * this means set if_rdev to the proper value. For files, directories,
384 * and symlinks this means to bring in the in-line data or extent
385 * pointers. For a file in B-tree format, only the root is immediately
386 * brought in-core. The rest will be in-lined in if_extents when it
387 * is first referenced (see xfs_iread_extents()).
394 xfs_attr_shortform_t
*atp
;
398 ip
->i_df
.if_ext_max
=
399 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
403 INT_GET(dip
->di_core
.di_nextents
, ARCH_CONVERT
) +
404 INT_GET(dip
->di_core
.di_anextents
, ARCH_CONVERT
) >
405 INT_GET(dip
->di_core
.di_nblocks
, ARCH_CONVERT
))) {
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)(INT_GET(dip
->di_core
.di_nextents
, ARCH_CONVERT
)
410 + INT_GET(dip
->di_core
.di_anextents
, ARCH_CONVERT
)),
412 INT_GET(dip
->di_core
.di_nblocks
, ARCH_CONVERT
));
413 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW
,
415 return XFS_ERROR(EFSCORRUPTED
);
418 if (unlikely(INT_GET(dip
->di_core
.di_forkoff
, ARCH_CONVERT
) > 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 (int)(INT_GET(dip
->di_core
.di_forkoff
, ARCH_CONVERT
)));
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(INT_GET(dip
->di_core
.di_format
, ARCH_CONVERT
) != 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
= INT_GET(dip
->di_u
.di_dev
, ARCH_CONVERT
);
446 switch (INT_GET(dip
->di_core
.di_format
, ARCH_CONVERT
)) {
447 case XFS_DINODE_FMT_LOCAL
:
449 * no local regular files yet
451 if (unlikely((INT_GET(dip
->di_core
.di_mode
, ARCH_CONVERT
) & 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
= INT_GET(dip
->di_core
.di_size
, ARCH_CONVERT
);
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 (INT_GET(dip
->di_core
.di_aformat
, ARCH_CONVERT
)) {
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_xlate_dinode_core - translate an xfs_inode_core_t between ondisk
716 * buf = on-disk representation
717 * dip = native representation
718 * dir = direction - +ve -> disk to native
719 * -ve -> native to disk
722 xfs_xlate_dinode_core(
724 xfs_dinode_core_t
*dip
,
727 xfs_dinode_core_t
*buf_core
= (xfs_dinode_core_t
*)buf
;
728 xfs_dinode_core_t
*mem_core
= (xfs_dinode_core_t
*)dip
;
729 xfs_arch_t arch
= ARCH_CONVERT
;
733 INT_XLATE(buf_core
->di_magic
, mem_core
->di_magic
, dir
, arch
);
734 INT_XLATE(buf_core
->di_mode
, mem_core
->di_mode
, dir
, arch
);
735 INT_XLATE(buf_core
->di_version
, mem_core
->di_version
, dir
, arch
);
736 INT_XLATE(buf_core
->di_format
, mem_core
->di_format
, dir
, arch
);
737 INT_XLATE(buf_core
->di_onlink
, mem_core
->di_onlink
, dir
, arch
);
738 INT_XLATE(buf_core
->di_uid
, mem_core
->di_uid
, dir
, arch
);
739 INT_XLATE(buf_core
->di_gid
, mem_core
->di_gid
, dir
, arch
);
740 INT_XLATE(buf_core
->di_nlink
, mem_core
->di_nlink
, dir
, arch
);
741 INT_XLATE(buf_core
->di_projid
, mem_core
->di_projid
, dir
, arch
);
744 memcpy(mem_core
->di_pad
, buf_core
->di_pad
,
745 sizeof(buf_core
->di_pad
));
747 memcpy(buf_core
->di_pad
, mem_core
->di_pad
,
748 sizeof(buf_core
->di_pad
));
751 INT_XLATE(buf_core
->di_flushiter
, mem_core
->di_flushiter
, dir
, arch
);
753 INT_XLATE(buf_core
->di_atime
.t_sec
, mem_core
->di_atime
.t_sec
,
755 INT_XLATE(buf_core
->di_atime
.t_nsec
, mem_core
->di_atime
.t_nsec
,
757 INT_XLATE(buf_core
->di_mtime
.t_sec
, mem_core
->di_mtime
.t_sec
,
759 INT_XLATE(buf_core
->di_mtime
.t_nsec
, mem_core
->di_mtime
.t_nsec
,
761 INT_XLATE(buf_core
->di_ctime
.t_sec
, mem_core
->di_ctime
.t_sec
,
763 INT_XLATE(buf_core
->di_ctime
.t_nsec
, mem_core
->di_ctime
.t_nsec
,
765 INT_XLATE(buf_core
->di_size
, mem_core
->di_size
, dir
, arch
);
766 INT_XLATE(buf_core
->di_nblocks
, mem_core
->di_nblocks
, dir
, arch
);
767 INT_XLATE(buf_core
->di_extsize
, mem_core
->di_extsize
, dir
, arch
);
768 INT_XLATE(buf_core
->di_nextents
, mem_core
->di_nextents
, dir
, arch
);
769 INT_XLATE(buf_core
->di_anextents
, mem_core
->di_anextents
, dir
, arch
);
770 INT_XLATE(buf_core
->di_forkoff
, mem_core
->di_forkoff
, dir
, arch
);
771 INT_XLATE(buf_core
->di_aformat
, mem_core
->di_aformat
, dir
, arch
);
772 INT_XLATE(buf_core
->di_dmevmask
, mem_core
->di_dmevmask
, dir
, arch
);
773 INT_XLATE(buf_core
->di_dmstate
, mem_core
->di_dmstate
, dir
, arch
);
774 INT_XLATE(buf_core
->di_flags
, mem_core
->di_flags
, dir
, arch
);
775 INT_XLATE(buf_core
->di_gen
, mem_core
->di_gen
, dir
, arch
);
784 if (di_flags
& XFS_DIFLAG_ANY
) {
785 if (di_flags
& XFS_DIFLAG_REALTIME
)
786 flags
|= XFS_XFLAG_REALTIME
;
787 if (di_flags
& XFS_DIFLAG_PREALLOC
)
788 flags
|= XFS_XFLAG_PREALLOC
;
789 if (di_flags
& XFS_DIFLAG_IMMUTABLE
)
790 flags
|= XFS_XFLAG_IMMUTABLE
;
791 if (di_flags
& XFS_DIFLAG_APPEND
)
792 flags
|= XFS_XFLAG_APPEND
;
793 if (di_flags
& XFS_DIFLAG_SYNC
)
794 flags
|= XFS_XFLAG_SYNC
;
795 if (di_flags
& XFS_DIFLAG_NOATIME
)
796 flags
|= XFS_XFLAG_NOATIME
;
797 if (di_flags
& XFS_DIFLAG_NODUMP
)
798 flags
|= XFS_XFLAG_NODUMP
;
799 if (di_flags
& XFS_DIFLAG_RTINHERIT
)
800 flags
|= XFS_XFLAG_RTINHERIT
;
801 if (di_flags
& XFS_DIFLAG_PROJINHERIT
)
802 flags
|= XFS_XFLAG_PROJINHERIT
;
803 if (di_flags
& XFS_DIFLAG_NOSYMLINKS
)
804 flags
|= XFS_XFLAG_NOSYMLINKS
;
805 if (di_flags
& XFS_DIFLAG_EXTSIZE
)
806 flags
|= XFS_XFLAG_EXTSIZE
;
807 if (di_flags
& XFS_DIFLAG_EXTSZINHERIT
)
808 flags
|= XFS_XFLAG_EXTSZINHERIT
;
809 if (di_flags
& XFS_DIFLAG_NODEFRAG
)
810 flags
|= XFS_XFLAG_NODEFRAG
;
811 if (di_flags
& XFS_DIFLAG_FILESTREAM
)
812 flags
|= XFS_XFLAG_FILESTREAM
;
822 xfs_dinode_core_t
*dic
= &ip
->i_d
;
824 return _xfs_dic2xflags(dic
->di_flags
) |
825 (XFS_CFORK_Q(dic
) ? XFS_XFLAG_HASATTR
: 0);
830 xfs_dinode_core_t
*dic
)
832 return _xfs_dic2xflags(INT_GET(dic
->di_flags
, ARCH_CONVERT
)) |
833 (XFS_CFORK_Q_DISK(dic
) ? XFS_XFLAG_HASATTR
: 0);
837 * Given a mount structure and an inode number, return a pointer
838 * to a newly allocated in-core inode corresponding to the given
841 * Initialize the inode's attributes and extent pointers if it
842 * already has them (it will not if the inode has no links).
858 ASSERT(xfs_inode_zone
!= NULL
);
860 ip
= kmem_zone_zalloc(xfs_inode_zone
, KM_SLEEP
);
863 spin_lock_init(&ip
->i_flags_lock
);
866 * Get pointer's to the on-disk inode and the buffer containing it.
867 * If the inode number refers to a block outside the file system
868 * then xfs_itobp() will return NULL. In this case we should
869 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
870 * know that this is a new incore inode.
872 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &bp
, bno
, imap_flags
);
874 kmem_zone_free(xfs_inode_zone
, ip
);
879 * Initialize inode's trace buffers.
880 * Do this before xfs_iformat in case it adds entries.
882 #ifdef XFS_BMAP_TRACE
883 ip
->i_xtrace
= ktrace_alloc(XFS_BMAP_KTRACE_SIZE
, KM_SLEEP
);
885 #ifdef XFS_BMBT_TRACE
886 ip
->i_btrace
= ktrace_alloc(XFS_BMBT_KTRACE_SIZE
, KM_SLEEP
);
889 ip
->i_rwtrace
= ktrace_alloc(XFS_RW_KTRACE_SIZE
, KM_SLEEP
);
891 #ifdef XFS_ILOCK_TRACE
892 ip
->i_lock_trace
= ktrace_alloc(XFS_ILOCK_KTRACE_SIZE
, KM_SLEEP
);
894 #ifdef XFS_DIR2_TRACE
895 ip
->i_dir_trace
= ktrace_alloc(XFS_DIR2_KTRACE_SIZE
, KM_SLEEP
);
899 * If we got something that isn't an inode it means someone
900 * (nfs or dmi) has a stale handle.
902 if (INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
) != XFS_DINODE_MAGIC
) {
903 kmem_zone_free(xfs_inode_zone
, ip
);
904 xfs_trans_brelse(tp
, bp
);
906 xfs_fs_cmn_err(CE_ALERT
, mp
, "xfs_iread: "
907 "dip->di_core.di_magic (0x%x) != "
908 "XFS_DINODE_MAGIC (0x%x)",
909 INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
),
912 return XFS_ERROR(EINVAL
);
916 * If the on-disk inode is already linked to a directory
917 * entry, copy all of the inode into the in-core inode.
918 * xfs_iformat() handles copying in the inode format
919 * specific information.
920 * Otherwise, just get the truly permanent information.
922 if (dip
->di_core
.di_mode
) {
923 xfs_xlate_dinode_core((xfs_caddr_t
)&dip
->di_core
,
925 error
= xfs_iformat(ip
, dip
);
927 kmem_zone_free(xfs_inode_zone
, ip
);
928 xfs_trans_brelse(tp
, bp
);
930 xfs_fs_cmn_err(CE_ALERT
, mp
, "xfs_iread: "
931 "xfs_iformat() returned error %d",
937 ip
->i_d
.di_magic
= INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
);
938 ip
->i_d
.di_version
= INT_GET(dip
->di_core
.di_version
, ARCH_CONVERT
);
939 ip
->i_d
.di_gen
= INT_GET(dip
->di_core
.di_gen
, ARCH_CONVERT
);
940 ip
->i_d
.di_flushiter
= INT_GET(dip
->di_core
.di_flushiter
, ARCH_CONVERT
);
942 * Make sure to pull in the mode here as well in
943 * case the inode is released without being used.
944 * This ensures that xfs_inactive() will see that
945 * the inode is already free and not try to mess
946 * with the uninitialized part of it.
950 * Initialize the per-fork minima and maxima for a new
951 * inode here. xfs_iformat will do it for old inodes.
953 ip
->i_df
.if_ext_max
=
954 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
957 INIT_LIST_HEAD(&ip
->i_reclaim
);
960 * The inode format changed when we moved the link count and
961 * made it 32 bits long. If this is an old format inode,
962 * convert it in memory to look like a new one. If it gets
963 * flushed to disk we will convert back before flushing or
964 * logging it. We zero out the new projid field and the old link
965 * count field. We'll handle clearing the pad field (the remains
966 * of the old uuid field) when we actually convert the inode to
967 * the new format. We don't change the version number so that we
968 * can distinguish this from a real new format inode.
970 if (ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
) {
971 ip
->i_d
.di_nlink
= ip
->i_d
.di_onlink
;
972 ip
->i_d
.di_onlink
= 0;
973 ip
->i_d
.di_projid
= 0;
976 ip
->i_delayed_blks
= 0;
977 ip
->i_size
= ip
->i_d
.di_size
;
980 * Mark the buffer containing the inode as something to keep
981 * around for a while. This helps to keep recently accessed
982 * meta-data in-core longer.
984 XFS_BUF_SET_REF(bp
, XFS_INO_REF
);
987 * Use xfs_trans_brelse() to release the buffer containing the
988 * on-disk inode, because it was acquired with xfs_trans_read_buf()
989 * in xfs_itobp() above. If tp is NULL, this is just a normal
990 * brelse(). If we're within a transaction, then xfs_trans_brelse()
991 * will only release the buffer if it is not dirty within the
992 * transaction. It will be OK to release the buffer in this case,
993 * because inodes on disk are never destroyed and we will be
994 * locking the new in-core inode before putting it in the hash
995 * table where other processes can find it. Thus we don't have
996 * to worry about the inode being changed just because we released
999 xfs_trans_brelse(tp
, bp
);
1005 * Read in extents from a btree-format inode.
1006 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1016 xfs_extnum_t nextents
;
1019 if (unlikely(XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
)) {
1020 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW
,
1022 return XFS_ERROR(EFSCORRUPTED
);
1024 nextents
= XFS_IFORK_NEXTENTS(ip
, whichfork
);
1025 size
= nextents
* sizeof(xfs_bmbt_rec_t
);
1026 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1029 * We know that the size is valid (it's checked in iformat_btree)
1031 ifp
->if_lastex
= NULLEXTNUM
;
1032 ifp
->if_bytes
= ifp
->if_real_bytes
= 0;
1033 ifp
->if_flags
|= XFS_IFEXTENTS
;
1034 xfs_iext_add(ifp
, 0, nextents
);
1035 error
= xfs_bmap_read_extents(tp
, ip
, whichfork
);
1037 xfs_iext_destroy(ifp
);
1038 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
1041 xfs_validate_extents(ifp
, nextents
, XFS_EXTFMT_INODE(ip
));
1046 * Allocate an inode on disk and return a copy of its in-core version.
1047 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1048 * appropriately within the inode. The uid and gid for the inode are
1049 * set according to the contents of the given cred structure.
1051 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1052 * has a free inode available, call xfs_iget()
1053 * to obtain the in-core version of the allocated inode. Finally,
1054 * fill in the inode and log its initial contents. In this case,
1055 * ialloc_context would be set to NULL and call_again set to false.
1057 * If xfs_dialloc() does not have an available inode,
1058 * it will replenish its supply by doing an allocation. Since we can
1059 * only do one allocation within a transaction without deadlocks, we
1060 * must commit the current transaction before returning the inode itself.
1061 * In this case, therefore, we will set call_again to true and return.
1062 * The caller should then commit the current transaction, start a new
1063 * transaction, and call xfs_ialloc() again to actually get the inode.
1065 * To ensure that some other process does not grab the inode that
1066 * was allocated during the first call to xfs_ialloc(), this routine
1067 * also returns the [locked] bp pointing to the head of the freelist
1068 * as ialloc_context. The caller should hold this buffer across
1069 * the commit and pass it back into this routine on the second call.
1071 * If we are allocating quota inodes, we do not have a parent inode
1072 * to attach to or associate with (i.e. pip == NULL) because they
1073 * are not linked into the directory structure - they are attached
1074 * directly to the superblock - and so have no parent.
1086 xfs_buf_t
**ialloc_context
,
1087 boolean_t
*call_again
,
1097 * Call the space management code to pick
1098 * the on-disk inode to be allocated.
1100 error
= xfs_dialloc(tp
, pip
? pip
->i_ino
: 0, mode
, okalloc
,
1101 ialloc_context
, call_again
, &ino
);
1105 if (*call_again
|| ino
== NULLFSINO
) {
1109 ASSERT(*ialloc_context
== NULL
);
1112 * Get the in-core inode with the lock held exclusively.
1113 * This is because we're setting fields here we need
1114 * to prevent others from looking at until we're done.
1116 error
= xfs_trans_iget(tp
->t_mountp
, tp
, ino
,
1117 XFS_IGET_CREATE
, XFS_ILOCK_EXCL
, &ip
);
1124 ip
->i_d
.di_mode
= (__uint16_t
)mode
;
1125 ip
->i_d
.di_onlink
= 0;
1126 ip
->i_d
.di_nlink
= nlink
;
1127 ASSERT(ip
->i_d
.di_nlink
== nlink
);
1128 ip
->i_d
.di_uid
= current_fsuid(cr
);
1129 ip
->i_d
.di_gid
= current_fsgid(cr
);
1130 ip
->i_d
.di_projid
= prid
;
1131 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
1134 * If the superblock version is up to where we support new format
1135 * inodes and this is currently an old format inode, then change
1136 * the inode version number now. This way we only do the conversion
1137 * here rather than here and in the flush/logging code.
1139 if (XFS_SB_VERSION_HASNLINK(&tp
->t_mountp
->m_sb
) &&
1140 ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
) {
1141 ip
->i_d
.di_version
= XFS_DINODE_VERSION_2
;
1143 * We've already zeroed the old link count, the projid field,
1144 * and the pad field.
1149 * Project ids won't be stored on disk if we are using a version 1 inode.
1151 if ((prid
!= 0) && (ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
))
1152 xfs_bump_ino_vers2(tp
, ip
);
1154 if (pip
&& XFS_INHERIT_GID(pip
, vp
->v_vfsp
)) {
1155 ip
->i_d
.di_gid
= pip
->i_d
.di_gid
;
1156 if ((pip
->i_d
.di_mode
& S_ISGID
) && (mode
& S_IFMT
) == S_IFDIR
) {
1157 ip
->i_d
.di_mode
|= S_ISGID
;
1162 * If the group ID of the new file does not match the effective group
1163 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1164 * (and only if the irix_sgid_inherit compatibility variable is set).
1166 if ((irix_sgid_inherit
) &&
1167 (ip
->i_d
.di_mode
& S_ISGID
) &&
1168 (!in_group_p((gid_t
)ip
->i_d
.di_gid
))) {
1169 ip
->i_d
.di_mode
&= ~S_ISGID
;
1172 ip
->i_d
.di_size
= 0;
1174 ip
->i_d
.di_nextents
= 0;
1175 ASSERT(ip
->i_d
.di_nblocks
== 0);
1176 xfs_ichgtime(ip
, XFS_ICHGTIME_CHG
|XFS_ICHGTIME_ACC
|XFS_ICHGTIME_MOD
);
1178 * di_gen will have been taken care of in xfs_iread.
1180 ip
->i_d
.di_extsize
= 0;
1181 ip
->i_d
.di_dmevmask
= 0;
1182 ip
->i_d
.di_dmstate
= 0;
1183 ip
->i_d
.di_flags
= 0;
1184 flags
= XFS_ILOG_CORE
;
1185 switch (mode
& S_IFMT
) {
1190 ip
->i_d
.di_format
= XFS_DINODE_FMT_DEV
;
1191 ip
->i_df
.if_u2
.if_rdev
= rdev
;
1192 ip
->i_df
.if_flags
= 0;
1193 flags
|= XFS_ILOG_DEV
;
1196 if (pip
&& xfs_inode_is_filestream(pip
)) {
1197 error
= xfs_filestream_associate(pip
, ip
);
1201 xfs_iflags_set(ip
, XFS_IFILESTREAM
);
1205 if (pip
&& (pip
->i_d
.di_flags
& XFS_DIFLAG_ANY
)) {
1208 if ((mode
& S_IFMT
) == S_IFDIR
) {
1209 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
1210 di_flags
|= XFS_DIFLAG_RTINHERIT
;
1211 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
1212 di_flags
|= XFS_DIFLAG_EXTSZINHERIT
;
1213 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
1215 } else if ((mode
& S_IFMT
) == S_IFREG
) {
1216 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
) {
1217 di_flags
|= XFS_DIFLAG_REALTIME
;
1218 ip
->i_iocore
.io_flags
|= XFS_IOCORE_RT
;
1220 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
1221 di_flags
|= XFS_DIFLAG_EXTSIZE
;
1222 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
1225 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOATIME
) &&
1226 xfs_inherit_noatime
)
1227 di_flags
|= XFS_DIFLAG_NOATIME
;
1228 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODUMP
) &&
1230 di_flags
|= XFS_DIFLAG_NODUMP
;
1231 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_SYNC
) &&
1233 di_flags
|= XFS_DIFLAG_SYNC
;
1234 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOSYMLINKS
) &&
1235 xfs_inherit_nosymlinks
)
1236 di_flags
|= XFS_DIFLAG_NOSYMLINKS
;
1237 if (pip
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
1238 di_flags
|= XFS_DIFLAG_PROJINHERIT
;
1239 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODEFRAG
) &&
1240 xfs_inherit_nodefrag
)
1241 di_flags
|= XFS_DIFLAG_NODEFRAG
;
1242 if (pip
->i_d
.di_flags
& XFS_DIFLAG_FILESTREAM
)
1243 di_flags
|= XFS_DIFLAG_FILESTREAM
;
1244 ip
->i_d
.di_flags
|= di_flags
;
1248 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
1249 ip
->i_df
.if_flags
= XFS_IFEXTENTS
;
1250 ip
->i_df
.if_bytes
= ip
->i_df
.if_real_bytes
= 0;
1251 ip
->i_df
.if_u1
.if_extents
= NULL
;
1257 * Attribute fork settings for new inode.
1259 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
1260 ip
->i_d
.di_anextents
= 0;
1263 * Log the new values stuffed into the inode.
1265 xfs_trans_log_inode(tp
, ip
, flags
);
1267 /* now that we have an i_mode we can setup inode ops and unlock */
1268 bhv_vfs_init_vnode(XFS_MTOVFS(tp
->t_mountp
), vp
, XFS_ITOBHV(ip
), 1);
1275 * Check to make sure that there are no blocks allocated to the
1276 * file beyond the size of the file. We don't check this for
1277 * files with fixed size extents or real time extents, but we
1278 * at least do it for regular files.
1287 xfs_fileoff_t map_first
;
1289 xfs_bmbt_irec_t imaps
[2];
1291 if ((ip
->i_d
.di_mode
& S_IFMT
) != S_IFREG
)
1294 if (ip
->i_d
.di_flags
& (XFS_DIFLAG_REALTIME
| XFS_DIFLAG_EXTSIZE
))
1298 map_first
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)isize
);
1300 * The filesystem could be shutting down, so bmapi may return
1303 if (xfs_bmapi(NULL
, ip
, map_first
,
1305 (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
)) -
1307 XFS_BMAPI_ENTIRE
, NULL
, 0, imaps
, &nimaps
,
1310 ASSERT(nimaps
== 1);
1311 ASSERT(imaps
[0].br_startblock
== HOLESTARTBLOCK
);
1316 * Calculate the last possible buffered byte in a file. This must
1317 * include data that was buffered beyond the EOF by the write code.
1318 * This also needs to deal with overflowing the xfs_fsize_t type
1319 * which can happen for sizes near the limit.
1321 * We also need to take into account any blocks beyond the EOF. It
1322 * may be the case that they were buffered by a write which failed.
1323 * In that case the pages will still be in memory, but the inode size
1324 * will never have been updated.
1331 xfs_fsize_t last_byte
;
1332 xfs_fileoff_t last_block
;
1333 xfs_fileoff_t size_last_block
;
1336 ASSERT(ismrlocked(&(ip
->i_iolock
), MR_UPDATE
| MR_ACCESS
));
1340 * Only check for blocks beyond the EOF if the extents have
1341 * been read in. This eliminates the need for the inode lock,
1342 * and it also saves us from looking when it really isn't
1345 if (ip
->i_df
.if_flags
& XFS_IFEXTENTS
) {
1346 error
= xfs_bmap_last_offset(NULL
, ip
, &last_block
,
1354 size_last_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)ip
->i_size
);
1355 last_block
= XFS_FILEOFF_MAX(last_block
, size_last_block
);
1357 last_byte
= XFS_FSB_TO_B(mp
, last_block
);
1358 if (last_byte
< 0) {
1359 return XFS_MAXIOFFSET(mp
);
1361 last_byte
+= (1 << mp
->m_writeio_log
);
1362 if (last_byte
< 0) {
1363 return XFS_MAXIOFFSET(mp
);
1368 #if defined(XFS_RW_TRACE)
1374 xfs_fsize_t new_size
,
1375 xfs_off_t toss_start
,
1376 xfs_off_t toss_finish
)
1378 if (ip
->i_rwtrace
== NULL
) {
1382 ktrace_enter(ip
->i_rwtrace
,
1385 (void*)(unsigned long)((ip
->i_d
.di_size
>> 32) & 0xffffffff),
1386 (void*)(unsigned long)(ip
->i_d
.di_size
& 0xffffffff),
1387 (void*)((long)flag
),
1388 (void*)(unsigned long)((new_size
>> 32) & 0xffffffff),
1389 (void*)(unsigned long)(new_size
& 0xffffffff),
1390 (void*)(unsigned long)((toss_start
>> 32) & 0xffffffff),
1391 (void*)(unsigned long)(toss_start
& 0xffffffff),
1392 (void*)(unsigned long)((toss_finish
>> 32) & 0xffffffff),
1393 (void*)(unsigned long)(toss_finish
& 0xffffffff),
1394 (void*)(unsigned long)current_cpu(),
1395 (void*)(unsigned long)current_pid(),
1401 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1405 * Start the truncation of the file to new_size. The new size
1406 * must be smaller than the current size. This routine will
1407 * clear the buffer and page caches of file data in the removed
1408 * range, and xfs_itruncate_finish() will remove the underlying
1411 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1412 * must NOT have the inode lock held at all. This is because we're
1413 * calling into the buffer/page cache code and we can't hold the
1414 * inode lock when we do so.
1416 * We need to wait for any direct I/Os in flight to complete before we
1417 * proceed with the truncate. This is needed to prevent the extents
1418 * being read or written by the direct I/Os from being removed while the
1419 * I/O is in flight as there is no other method of synchronising
1420 * direct I/O with the truncate operation. Also, because we hold
1421 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1422 * started until the truncate completes and drops the lock. Essentially,
1423 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1424 * between direct I/Os and the truncate operation.
1426 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1427 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1428 * in the case that the caller is locking things out of order and
1429 * may not be able to call xfs_itruncate_finish() with the inode lock
1430 * held without dropping the I/O lock. If the caller must drop the
1431 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1432 * must be called again with all the same restrictions as the initial
1436 xfs_itruncate_start(
1439 xfs_fsize_t new_size
)
1441 xfs_fsize_t last_byte
;
1442 xfs_off_t toss_start
;
1447 ASSERT(ismrlocked(&ip
->i_iolock
, MR_UPDATE
) != 0);
1448 ASSERT((new_size
== 0) || (new_size
<= ip
->i_size
));
1449 ASSERT((flags
== XFS_ITRUNC_DEFINITE
) ||
1450 (flags
== XFS_ITRUNC_MAYBE
));
1455 vn_iowait(vp
); /* wait for the completion of any pending DIOs */
1458 * Call toss_pages or flushinval_pages to get rid of pages
1459 * overlapping the region being removed. We have to use
1460 * the less efficient flushinval_pages in the case that the
1461 * caller may not be able to finish the truncate without
1462 * dropping the inode's I/O lock. Make sure
1463 * to catch any pages brought in by buffers overlapping
1464 * the EOF by searching out beyond the isize by our
1465 * block size. We round new_size up to a block boundary
1466 * so that we don't toss things on the same block as
1467 * new_size but before it.
1469 * Before calling toss_page or flushinval_pages, make sure to
1470 * call remapf() over the same region if the file is mapped.
1471 * This frees up mapped file references to the pages in the
1472 * given range and for the flushinval_pages case it ensures
1473 * that we get the latest mapped changes flushed out.
1475 toss_start
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)new_size
);
1476 toss_start
= XFS_FSB_TO_B(mp
, toss_start
);
1477 if (toss_start
< 0) {
1479 * The place to start tossing is beyond our maximum
1480 * file size, so there is no way that the data extended
1485 last_byte
= xfs_file_last_byte(ip
);
1486 xfs_itrunc_trace(XFS_ITRUNC_START
, ip
, flags
, new_size
, toss_start
,
1488 if (last_byte
> toss_start
) {
1489 if (flags
& XFS_ITRUNC_DEFINITE
) {
1490 bhv_vop_toss_pages(vp
, toss_start
, -1, FI_REMAPF_LOCKED
);
1492 error
= bhv_vop_flushinval_pages(vp
, toss_start
, -1, FI_REMAPF_LOCKED
);
1497 if (new_size
== 0) {
1498 ASSERT(VN_CACHED(vp
) == 0);
1505 * Shrink the file to the given new_size. The new
1506 * size must be smaller than the current size.
1507 * This will free up the underlying blocks
1508 * in the removed range after a call to xfs_itruncate_start()
1509 * or xfs_atruncate_start().
1511 * The transaction passed to this routine must have made
1512 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1513 * This routine may commit the given transaction and
1514 * start new ones, so make sure everything involved in
1515 * the transaction is tidy before calling here.
1516 * Some transaction will be returned to the caller to be
1517 * committed. The incoming transaction must already include
1518 * the inode, and both inode locks must be held exclusively.
1519 * The inode must also be "held" within the transaction. On
1520 * return the inode will be "held" within the returned transaction.
1521 * This routine does NOT require any disk space to be reserved
1522 * for it within the transaction.
1524 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1525 * and it indicates the fork which is to be truncated. For the
1526 * attribute fork we only support truncation to size 0.
1528 * We use the sync parameter to indicate whether or not the first
1529 * transaction we perform might have to be synchronous. For the attr fork,
1530 * it needs to be so if the unlink of the inode is not yet known to be
1531 * permanent in the log. This keeps us from freeing and reusing the
1532 * blocks of the attribute fork before the unlink of the inode becomes
1535 * For the data fork, we normally have to run synchronously if we're
1536 * being called out of the inactive path or we're being called
1537 * out of the create path where we're truncating an existing file.
1538 * Either way, the truncate needs to be sync so blocks don't reappear
1539 * in the file with altered data in case of a crash. wsync filesystems
1540 * can run the first case async because anything that shrinks the inode
1541 * has to run sync so by the time we're called here from inactive, the
1542 * inode size is permanently set to 0.
1544 * Calls from the truncate path always need to be sync unless we're
1545 * in a wsync filesystem and the file has already been unlinked.
1547 * The caller is responsible for correctly setting the sync parameter.
1548 * It gets too hard for us to guess here which path we're being called
1549 * out of just based on inode state.
1552 xfs_itruncate_finish(
1555 xfs_fsize_t new_size
,
1559 xfs_fsblock_t first_block
;
1560 xfs_fileoff_t first_unmap_block
;
1561 xfs_fileoff_t last_block
;
1562 xfs_filblks_t unmap_len
=0;
1567 xfs_bmap_free_t free_list
;
1570 ASSERT(ismrlocked(&ip
->i_iolock
, MR_UPDATE
) != 0);
1571 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
) != 0);
1572 ASSERT((new_size
== 0) || (new_size
<= ip
->i_size
));
1573 ASSERT(*tp
!= NULL
);
1574 ASSERT((*tp
)->t_flags
& XFS_TRANS_PERM_LOG_RES
);
1575 ASSERT(ip
->i_transp
== *tp
);
1576 ASSERT(ip
->i_itemp
!= NULL
);
1577 ASSERT(ip
->i_itemp
->ili_flags
& XFS_ILI_HOLD
);
1581 mp
= (ntp
)->t_mountp
;
1582 ASSERT(! XFS_NOT_DQATTACHED(mp
, ip
));
1585 * We only support truncating the entire attribute fork.
1587 if (fork
== XFS_ATTR_FORK
) {
1590 first_unmap_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)new_size
);
1591 xfs_itrunc_trace(XFS_ITRUNC_FINISH1
, ip
, 0, new_size
, 0, 0);
1593 * The first thing we do is set the size to new_size permanently
1594 * on disk. This way we don't have to worry about anyone ever
1595 * being able to look at the data being freed even in the face
1596 * of a crash. What we're getting around here is the case where
1597 * we free a block, it is allocated to another file, it is written
1598 * to, and then we crash. If the new data gets written to the
1599 * file but the log buffers containing the free and reallocation
1600 * don't, then we'd end up with garbage in the blocks being freed.
1601 * As long as we make the new_size permanent before actually
1602 * freeing any blocks it doesn't matter if they get writtten to.
1604 * The callers must signal into us whether or not the size
1605 * setting here must be synchronous. There are a few cases
1606 * where it doesn't have to be synchronous. Those cases
1607 * occur if the file is unlinked and we know the unlink is
1608 * permanent or if the blocks being truncated are guaranteed
1609 * to be beyond the inode eof (regardless of the link count)
1610 * and the eof value is permanent. Both of these cases occur
1611 * only on wsync-mounted filesystems. In those cases, we're
1612 * guaranteed that no user will ever see the data in the blocks
1613 * that are being truncated so the truncate can run async.
1614 * In the free beyond eof case, the file may wind up with
1615 * more blocks allocated to it than it needs if we crash
1616 * and that won't get fixed until the next time the file
1617 * is re-opened and closed but that's ok as that shouldn't
1618 * be too many blocks.
1620 * However, we can't just make all wsync xactions run async
1621 * because there's one call out of the create path that needs
1622 * to run sync where it's truncating an existing file to size
1623 * 0 whose size is > 0.
1625 * It's probably possible to come up with a test in this
1626 * routine that would correctly distinguish all the above
1627 * cases from the values of the function parameters and the
1628 * inode state but for sanity's sake, I've decided to let the
1629 * layers above just tell us. It's simpler to correctly figure
1630 * out in the layer above exactly under what conditions we
1631 * can run async and I think it's easier for others read and
1632 * follow the logic in case something has to be changed.
1633 * cscope is your friend -- rcc.
1635 * The attribute fork is much simpler.
1637 * For the attribute fork we allow the caller to tell us whether
1638 * the unlink of the inode that led to this call is yet permanent
1639 * in the on disk log. If it is not and we will be freeing extents
1640 * in this inode then we make the first transaction synchronous
1641 * to make sure that the unlink is permanent by the time we free
1644 if (fork
== XFS_DATA_FORK
) {
1645 if (ip
->i_d
.di_nextents
> 0) {
1647 * If we are not changing the file size then do
1648 * not update the on-disk file size - we may be
1649 * called from xfs_inactive_free_eofblocks(). If we
1650 * update the on-disk file size and then the system
1651 * crashes before the contents of the file are
1652 * flushed to disk then the files may be full of
1653 * holes (ie NULL files bug).
1655 if (ip
->i_size
!= new_size
) {
1656 ip
->i_d
.di_size
= new_size
;
1657 ip
->i_size
= new_size
;
1658 xfs_trans_log_inode(ntp
, ip
, XFS_ILOG_CORE
);
1662 ASSERT(!(mp
->m_flags
& XFS_MOUNT_WSYNC
));
1663 if (ip
->i_d
.di_anextents
> 0)
1664 xfs_trans_set_sync(ntp
);
1666 ASSERT(fork
== XFS_DATA_FORK
||
1667 (fork
== XFS_ATTR_FORK
&&
1668 ((sync
&& !(mp
->m_flags
& XFS_MOUNT_WSYNC
)) ||
1669 (sync
== 0 && (mp
->m_flags
& XFS_MOUNT_WSYNC
)))));
1672 * Since it is possible for space to become allocated beyond
1673 * the end of the file (in a crash where the space is allocated
1674 * but the inode size is not yet updated), simply remove any
1675 * blocks which show up between the new EOF and the maximum
1676 * possible file size. If the first block to be removed is
1677 * beyond the maximum file size (ie it is the same as last_block),
1678 * then there is nothing to do.
1680 last_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
));
1681 ASSERT(first_unmap_block
<= last_block
);
1683 if (last_block
== first_unmap_block
) {
1686 unmap_len
= last_block
- first_unmap_block
+ 1;
1690 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1691 * will tell us whether it freed the entire range or
1692 * not. If this is a synchronous mount (wsync),
1693 * then we can tell bunmapi to keep all the
1694 * transactions asynchronous since the unlink
1695 * transaction that made this inode inactive has
1696 * already hit the disk. There's no danger of
1697 * the freed blocks being reused, there being a
1698 * crash, and the reused blocks suddenly reappearing
1699 * in this file with garbage in them once recovery
1702 XFS_BMAP_INIT(&free_list
, &first_block
);
1703 error
= XFS_BUNMAPI(mp
, ntp
, &ip
->i_iocore
,
1704 first_unmap_block
, unmap_len
,
1705 XFS_BMAPI_AFLAG(fork
) |
1706 (sync
? 0 : XFS_BMAPI_ASYNC
),
1707 XFS_ITRUNC_MAX_EXTENTS
,
1708 &first_block
, &free_list
,
1712 * If the bunmapi call encounters an error,
1713 * return to the caller where the transaction
1714 * can be properly aborted. We just need to
1715 * make sure we're not holding any resources
1716 * that we were not when we came in.
1718 xfs_bmap_cancel(&free_list
);
1723 * Duplicate the transaction that has the permanent
1724 * reservation and commit the old transaction.
1726 error
= xfs_bmap_finish(tp
, &free_list
, &committed
);
1730 * If the bmap finish call encounters an error,
1731 * return to the caller where the transaction
1732 * can be properly aborted. We just need to
1733 * make sure we're not holding any resources
1734 * that we were not when we came in.
1736 * Aborting from this point might lose some
1737 * blocks in the file system, but oh well.
1739 xfs_bmap_cancel(&free_list
);
1742 * If the passed in transaction committed
1743 * in xfs_bmap_finish(), then we want to
1744 * add the inode to this one before returning.
1745 * This keeps things simple for the higher
1746 * level code, because it always knows that
1747 * the inode is locked and held in the
1748 * transaction that returns to it whether
1749 * errors occur or not. We don't mark the
1750 * inode dirty so that this transaction can
1751 * be easily aborted if possible.
1753 xfs_trans_ijoin(ntp
, ip
,
1754 XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1755 xfs_trans_ihold(ntp
, ip
);
1762 * The first xact was committed,
1763 * so add the inode to the new one.
1764 * Mark it dirty so it will be logged
1765 * and moved forward in the log as
1766 * part of every commit.
1768 xfs_trans_ijoin(ntp
, ip
,
1769 XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1770 xfs_trans_ihold(ntp
, ip
);
1771 xfs_trans_log_inode(ntp
, ip
, XFS_ILOG_CORE
);
1773 ntp
= xfs_trans_dup(ntp
);
1774 (void) xfs_trans_commit(*tp
, 0);
1776 error
= xfs_trans_reserve(ntp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0,
1777 XFS_TRANS_PERM_LOG_RES
,
1778 XFS_ITRUNCATE_LOG_COUNT
);
1780 * Add the inode being truncated to the next chained
1783 xfs_trans_ijoin(ntp
, ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1784 xfs_trans_ihold(ntp
, ip
);
1789 * Only update the size in the case of the data fork, but
1790 * always re-log the inode so that our permanent transaction
1791 * can keep on rolling it forward in the log.
1793 if (fork
== XFS_DATA_FORK
) {
1794 xfs_isize_check(mp
, ip
, new_size
);
1796 * If we are not changing the file size then do
1797 * not update the on-disk file size - we may be
1798 * called from xfs_inactive_free_eofblocks(). If we
1799 * update the on-disk file size and then the system
1800 * crashes before the contents of the file are
1801 * flushed to disk then the files may be full of
1802 * holes (ie NULL files bug).
1804 if (ip
->i_size
!= new_size
) {
1805 ip
->i_d
.di_size
= new_size
;
1806 ip
->i_size
= new_size
;
1809 xfs_trans_log_inode(ntp
, ip
, XFS_ILOG_CORE
);
1810 ASSERT((new_size
!= 0) ||
1811 (fork
== XFS_ATTR_FORK
) ||
1812 (ip
->i_delayed_blks
== 0));
1813 ASSERT((new_size
!= 0) ||
1814 (fork
== XFS_ATTR_FORK
) ||
1815 (ip
->i_d
.di_nextents
== 0));
1816 xfs_itrunc_trace(XFS_ITRUNC_FINISH2
, ip
, 0, new_size
, 0, 0);
1824 * Do the first part of growing a file: zero any data in the last
1825 * block that is beyond the old EOF. We need to do this before
1826 * the inode is joined to the transaction to modify the i_size.
1827 * That way we can drop the inode lock and call into the buffer
1828 * cache to get the buffer mapping the EOF.
1833 xfs_fsize_t new_size
,
1838 ASSERT(ismrlocked(&(ip
->i_lock
), MR_UPDATE
) != 0);
1839 ASSERT(ismrlocked(&(ip
->i_iolock
), MR_UPDATE
) != 0);
1840 ASSERT(new_size
> ip
->i_size
);
1843 * Zero any pages that may have been created by
1844 * xfs_write_file() beyond the end of the file
1845 * and any blocks between the old and new file sizes.
1847 error
= xfs_zero_eof(XFS_ITOV(ip
), &ip
->i_iocore
, new_size
,
1855 * This routine is called to extend the size of a file.
1856 * The inode must have both the iolock and the ilock locked
1857 * for update and it must be a part of the current transaction.
1858 * The xfs_igrow_start() function must have been called previously.
1859 * If the change_flag is not zero, the inode change timestamp will
1866 xfs_fsize_t new_size
,
1869 ASSERT(ismrlocked(&(ip
->i_lock
), MR_UPDATE
) != 0);
1870 ASSERT(ismrlocked(&(ip
->i_iolock
), MR_UPDATE
) != 0);
1871 ASSERT(ip
->i_transp
== tp
);
1872 ASSERT(new_size
> ip
->i_size
);
1875 * Update the file size. Update the inode change timestamp
1876 * if change_flag set.
1878 ip
->i_d
.di_size
= new_size
;
1879 ip
->i_size
= new_size
;
1881 xfs_ichgtime(ip
, XFS_ICHGTIME_CHG
);
1882 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1888 * This is called when the inode's link count goes to 0.
1889 * We place the on-disk inode on a list in the AGI. It
1890 * will be pulled from this list when the inode is freed.
1902 xfs_agnumber_t agno
;
1903 xfs_daddr_t agdaddr
;
1910 ASSERT(ip
->i_d
.di_nlink
== 0);
1911 ASSERT(ip
->i_d
.di_mode
!= 0);
1912 ASSERT(ip
->i_transp
== tp
);
1916 agno
= XFS_INO_TO_AGNO(mp
, ip
->i_ino
);
1917 agdaddr
= XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
));
1920 * Get the agi buffer first. It ensures lock ordering
1923 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, agdaddr
,
1924 XFS_FSS_TO_BB(mp
, 1), 0, &agibp
);
1929 * Validate the magic number of the agi block.
1931 agi
= XFS_BUF_TO_AGI(agibp
);
1933 be32_to_cpu(agi
->agi_magicnum
) == XFS_AGI_MAGIC
&&
1934 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi
->agi_versionnum
));
1935 if (unlikely(XFS_TEST_ERROR(!agi_ok
, mp
, XFS_ERRTAG_IUNLINK
,
1936 XFS_RANDOM_IUNLINK
))) {
1937 XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW
, mp
, agi
);
1938 xfs_trans_brelse(tp
, agibp
);
1939 return XFS_ERROR(EFSCORRUPTED
);
1942 * Get the index into the agi hash table for the
1943 * list this inode will go on.
1945 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
1947 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
1948 ASSERT(agi
->agi_unlinked
[bucket_index
]);
1949 ASSERT(be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != agino
);
1951 if (be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != NULLAGINO
) {
1953 * There is already another inode in the bucket we need
1954 * to add ourselves to. Add us at the front of the list.
1955 * Here we put the head pointer into our next pointer,
1956 * and then we fall through to point the head at us.
1958 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, 0, 0);
1962 ASSERT(INT_GET(dip
->di_next_unlinked
, ARCH_CONVERT
) == NULLAGINO
);
1963 ASSERT(dip
->di_next_unlinked
);
1964 /* both on-disk, don't endian flip twice */
1965 dip
->di_next_unlinked
= agi
->agi_unlinked
[bucket_index
];
1966 offset
= ip
->i_boffset
+
1967 offsetof(xfs_dinode_t
, di_next_unlinked
);
1968 xfs_trans_inode_buf(tp
, ibp
);
1969 xfs_trans_log_buf(tp
, ibp
, offset
,
1970 (offset
+ sizeof(xfs_agino_t
) - 1));
1971 xfs_inobp_check(mp
, ibp
);
1975 * Point the bucket head pointer at the inode being inserted.
1978 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(agino
);
1979 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
1980 (sizeof(xfs_agino_t
) * bucket_index
);
1981 xfs_trans_log_buf(tp
, agibp
, offset
,
1982 (offset
+ sizeof(xfs_agino_t
) - 1));
1987 * Pull the on-disk inode from the AGI unlinked list.
2000 xfs_agnumber_t agno
;
2001 xfs_daddr_t agdaddr
;
2003 xfs_agino_t next_agino
;
2004 xfs_buf_t
*last_ibp
;
2005 xfs_dinode_t
*last_dip
= NULL
;
2007 int offset
, last_offset
= 0;
2012 * First pull the on-disk inode from the AGI unlinked list.
2016 agno
= XFS_INO_TO_AGNO(mp
, ip
->i_ino
);
2017 agdaddr
= XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
));
2020 * Get the agi buffer first. It ensures lock ordering
2023 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, agdaddr
,
2024 XFS_FSS_TO_BB(mp
, 1), 0, &agibp
);
2027 "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
2028 error
, mp
->m_fsname
);
2032 * Validate the magic number of the agi block.
2034 agi
= XFS_BUF_TO_AGI(agibp
);
2036 be32_to_cpu(agi
->agi_magicnum
) == XFS_AGI_MAGIC
&&
2037 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi
->agi_versionnum
));
2038 if (unlikely(XFS_TEST_ERROR(!agi_ok
, mp
, XFS_ERRTAG_IUNLINK_REMOVE
,
2039 XFS_RANDOM_IUNLINK_REMOVE
))) {
2040 XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW
,
2042 xfs_trans_brelse(tp
, agibp
);
2044 "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
2046 return XFS_ERROR(EFSCORRUPTED
);
2049 * Get the index into the agi hash table for the
2050 * list this inode will go on.
2052 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
2054 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
2055 ASSERT(be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != NULLAGINO
);
2056 ASSERT(agi
->agi_unlinked
[bucket_index
]);
2058 if (be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) == agino
) {
2060 * We're at the head of the list. Get the inode's
2061 * on-disk buffer to see if there is anyone after us
2062 * on the list. Only modify our next pointer if it
2063 * is not already NULLAGINO. This saves us the overhead
2064 * of dealing with the buffer when there is no need to
2067 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, 0, 0);
2070 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2071 error
, mp
->m_fsname
);
2074 next_agino
= INT_GET(dip
->di_next_unlinked
, ARCH_CONVERT
);
2075 ASSERT(next_agino
!= 0);
2076 if (next_agino
!= NULLAGINO
) {
2077 INT_SET(dip
->di_next_unlinked
, ARCH_CONVERT
, NULLAGINO
);
2078 offset
= ip
->i_boffset
+
2079 offsetof(xfs_dinode_t
, di_next_unlinked
);
2080 xfs_trans_inode_buf(tp
, ibp
);
2081 xfs_trans_log_buf(tp
, ibp
, offset
,
2082 (offset
+ sizeof(xfs_agino_t
) - 1));
2083 xfs_inobp_check(mp
, ibp
);
2085 xfs_trans_brelse(tp
, ibp
);
2088 * Point the bucket head pointer at the next inode.
2090 ASSERT(next_agino
!= 0);
2091 ASSERT(next_agino
!= agino
);
2092 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(next_agino
);
2093 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
2094 (sizeof(xfs_agino_t
) * bucket_index
);
2095 xfs_trans_log_buf(tp
, agibp
, offset
,
2096 (offset
+ sizeof(xfs_agino_t
) - 1));
2099 * We need to search the list for the inode being freed.
2101 next_agino
= be32_to_cpu(agi
->agi_unlinked
[bucket_index
]);
2103 while (next_agino
!= agino
) {
2105 * If the last inode wasn't the one pointing to
2106 * us, then release its buffer since we're not
2107 * going to do anything with it.
2109 if (last_ibp
!= NULL
) {
2110 xfs_trans_brelse(tp
, last_ibp
);
2112 next_ino
= XFS_AGINO_TO_INO(mp
, agno
, next_agino
);
2113 error
= xfs_inotobp(mp
, tp
, next_ino
, &last_dip
,
2114 &last_ibp
, &last_offset
);
2117 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
2118 error
, mp
->m_fsname
);
2121 next_agino
= INT_GET(last_dip
->di_next_unlinked
, ARCH_CONVERT
);
2122 ASSERT(next_agino
!= NULLAGINO
);
2123 ASSERT(next_agino
!= 0);
2126 * Now last_ibp points to the buffer previous to us on
2127 * the unlinked list. Pull us from the list.
2129 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, 0, 0);
2132 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2133 error
, mp
->m_fsname
);
2136 next_agino
= INT_GET(dip
->di_next_unlinked
, ARCH_CONVERT
);
2137 ASSERT(next_agino
!= 0);
2138 ASSERT(next_agino
!= agino
);
2139 if (next_agino
!= NULLAGINO
) {
2140 INT_SET(dip
->di_next_unlinked
, ARCH_CONVERT
, NULLAGINO
);
2141 offset
= ip
->i_boffset
+
2142 offsetof(xfs_dinode_t
, di_next_unlinked
);
2143 xfs_trans_inode_buf(tp
, ibp
);
2144 xfs_trans_log_buf(tp
, ibp
, offset
,
2145 (offset
+ sizeof(xfs_agino_t
) - 1));
2146 xfs_inobp_check(mp
, ibp
);
2148 xfs_trans_brelse(tp
, ibp
);
2151 * Point the previous inode on the list to the next inode.
2153 INT_SET(last_dip
->di_next_unlinked
, ARCH_CONVERT
, next_agino
);
2154 ASSERT(next_agino
!= 0);
2155 offset
= last_offset
+ offsetof(xfs_dinode_t
, di_next_unlinked
);
2156 xfs_trans_inode_buf(tp
, last_ibp
);
2157 xfs_trans_log_buf(tp
, last_ibp
, offset
,
2158 (offset
+ sizeof(xfs_agino_t
) - 1));
2159 xfs_inobp_check(mp
, last_ibp
);
2164 STATIC_INLINE
int xfs_inode_clean(xfs_inode_t
*ip
)
2166 return (((ip
->i_itemp
== NULL
) ||
2167 !(ip
->i_itemp
->ili_format
.ilf_fields
& XFS_ILOG_ALL
)) &&
2168 (ip
->i_update_core
== 0));
2173 xfs_inode_t
*free_ip
,
2177 xfs_mount_t
*mp
= free_ip
->i_mount
;
2178 int blks_per_cluster
;
2181 int i
, j
, found
, pre_flushed
;
2185 xfs_inode_t
*ip
, **ip_found
;
2186 xfs_inode_log_item_t
*iip
;
2187 xfs_log_item_t
*lip
;
2190 if (mp
->m_sb
.sb_blocksize
>= XFS_INODE_CLUSTER_SIZE(mp
)) {
2191 blks_per_cluster
= 1;
2192 ninodes
= mp
->m_sb
.sb_inopblock
;
2193 nbufs
= XFS_IALLOC_BLOCKS(mp
);
2195 blks_per_cluster
= XFS_INODE_CLUSTER_SIZE(mp
) /
2196 mp
->m_sb
.sb_blocksize
;
2197 ninodes
= blks_per_cluster
* mp
->m_sb
.sb_inopblock
;
2198 nbufs
= XFS_IALLOC_BLOCKS(mp
) / blks_per_cluster
;
2201 ip_found
= kmem_alloc(ninodes
* sizeof(xfs_inode_t
*), KM_NOFS
);
2203 for (j
= 0; j
< nbufs
; j
++, inum
+= ninodes
) {
2204 blkno
= XFS_AGB_TO_DADDR(mp
, XFS_INO_TO_AGNO(mp
, inum
),
2205 XFS_INO_TO_AGBNO(mp
, inum
));
2209 * Look for each inode in memory and attempt to lock it,
2210 * we can be racing with flush and tail pushing here.
2211 * any inode we get the locks on, add to an array of
2212 * inode items to process later.
2214 * The get the buffer lock, we could beat a flush
2215 * or tail pushing thread to the lock here, in which
2216 * case they will go looking for the inode buffer
2217 * and fail, we need some other form of interlock
2221 for (i
= 0; i
< ninodes
; i
++) {
2222 ih
= XFS_IHASH(mp
, inum
+ i
);
2223 read_lock(&ih
->ih_lock
);
2224 for (ip
= ih
->ih_next
; ip
!= NULL
; ip
= ip
->i_next
) {
2225 if (ip
->i_ino
== inum
+ i
)
2229 /* Inode not in memory or we found it already,
2232 if (!ip
|| xfs_iflags_test(ip
, XFS_ISTALE
)) {
2233 read_unlock(&ih
->ih_lock
);
2237 if (xfs_inode_clean(ip
)) {
2238 read_unlock(&ih
->ih_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(&ih
->ih_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
);
2280 read_unlock(&ih
->ih_lock
);
2283 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, blkno
,
2284 mp
->m_bsize
* blks_per_cluster
,
2288 lip
= XFS_BUF_FSPRIVATE(bp
, xfs_log_item_t
*);
2290 if (lip
->li_type
== XFS_LI_INODE
) {
2291 iip
= (xfs_inode_log_item_t
*)lip
;
2292 ASSERT(iip
->ili_logged
== 1);
2293 lip
->li_cb
= (void(*)(xfs_buf_t
*,xfs_log_item_t
*)) xfs_istale_done
;
2295 iip
->ili_flush_lsn
= iip
->ili_item
.li_lsn
;
2297 xfs_iflags_set(iip
->ili_inode
, XFS_ISTALE
);
2300 lip
= lip
->li_bio_list
;
2303 for (i
= 0; i
< found
; i
++) {
2308 ip
->i_update_core
= 0;
2310 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2314 iip
->ili_last_fields
= iip
->ili_format
.ilf_fields
;
2315 iip
->ili_format
.ilf_fields
= 0;
2316 iip
->ili_logged
= 1;
2318 iip
->ili_flush_lsn
= iip
->ili_item
.li_lsn
;
2321 xfs_buf_attach_iodone(bp
,
2322 (void(*)(xfs_buf_t
*,xfs_log_item_t
*))
2323 xfs_istale_done
, (xfs_log_item_t
*)iip
);
2324 if (ip
!= free_ip
) {
2325 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2329 if (found
|| pre_flushed
)
2330 xfs_trans_stale_inode_buf(tp
, bp
);
2331 xfs_trans_binval(tp
, bp
);
2334 kmem_free(ip_found
, ninodes
* sizeof(xfs_inode_t
*));
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
;
2357 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
));
2358 ASSERT(ip
->i_transp
== tp
);
2359 ASSERT(ip
->i_d
.di_nlink
== 0);
2360 ASSERT(ip
->i_d
.di_nextents
== 0);
2361 ASSERT(ip
->i_d
.di_anextents
== 0);
2362 ASSERT((ip
->i_d
.di_size
== 0 && ip
->i_size
== 0) ||
2363 ((ip
->i_d
.di_mode
& S_IFMT
) != S_IFREG
));
2364 ASSERT(ip
->i_d
.di_nblocks
== 0);
2367 * Pull the on-disk inode from the AGI unlinked list.
2369 error
= xfs_iunlink_remove(tp
, ip
);
2374 error
= xfs_difree(tp
, ip
->i_ino
, flist
, &delete, &first_ino
);
2378 ip
->i_d
.di_mode
= 0; /* mark incore inode as free */
2379 ip
->i_d
.di_flags
= 0;
2380 ip
->i_d
.di_dmevmask
= 0;
2381 ip
->i_d
.di_forkoff
= 0; /* mark the attr fork not in use */
2382 ip
->i_df
.if_ext_max
=
2383 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
2384 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
2385 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
2387 * Bump the generation count so no one will be confused
2388 * by reincarnations of this inode.
2391 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2394 xfs_ifree_cluster(ip
, tp
, first_ino
);
2401 * Reallocate the space for if_broot based on the number of records
2402 * being added or deleted as indicated in rec_diff. Move the records
2403 * and pointers in if_broot to fit the new size. When shrinking this
2404 * will eliminate holes between the records and pointers created by
2405 * the caller. When growing this will create holes to be filled in
2408 * The caller must not request to add more records than would fit in
2409 * the on-disk inode root. If the if_broot is currently NULL, then
2410 * if we adding records one will be allocated. The caller must also
2411 * not request that the number of records go below zero, although
2412 * it can go to zero.
2414 * ip -- the inode whose if_broot area is changing
2415 * ext_diff -- the change in the number of records, positive or negative,
2416 * requested for the if_broot array.
2426 xfs_bmbt_block_t
*new_broot
;
2433 * Handle the degenerate case quietly.
2435 if (rec_diff
== 0) {
2439 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2442 * If there wasn't any memory allocated before, just
2443 * allocate it now and get out.
2445 if (ifp
->if_broot_bytes
== 0) {
2446 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff
);
2447 ifp
->if_broot
= (xfs_bmbt_block_t
*)kmem_alloc(new_size
,
2449 ifp
->if_broot_bytes
= (int)new_size
;
2454 * If there is already an existing if_broot, then we need
2455 * to realloc() it and shift the pointers to their new
2456 * location. The records don't change location because
2457 * they are kept butted up against the btree block header.
2459 cur_max
= XFS_BMAP_BROOT_MAXRECS(ifp
->if_broot_bytes
);
2460 new_max
= cur_max
+ rec_diff
;
2461 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max
);
2462 ifp
->if_broot
= (xfs_bmbt_block_t
*)
2463 kmem_realloc(ifp
->if_broot
,
2465 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max
), /* old size */
2467 op
= (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp
->if_broot
, 1,
2468 ifp
->if_broot_bytes
);
2469 np
= (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp
->if_broot
, 1,
2471 ifp
->if_broot_bytes
= (int)new_size
;
2472 ASSERT(ifp
->if_broot_bytes
<=
2473 XFS_IFORK_SIZE(ip
, whichfork
) + XFS_BROOT_SIZE_ADJ
);
2474 memmove(np
, op
, cur_max
* (uint
)sizeof(xfs_dfsbno_t
));
2479 * rec_diff is less than 0. In this case, we are shrinking the
2480 * if_broot buffer. It must already exist. If we go to zero
2481 * records, just get rid of the root and clear the status bit.
2483 ASSERT((ifp
->if_broot
!= NULL
) && (ifp
->if_broot_bytes
> 0));
2484 cur_max
= XFS_BMAP_BROOT_MAXRECS(ifp
->if_broot_bytes
);
2485 new_max
= cur_max
+ rec_diff
;
2486 ASSERT(new_max
>= 0);
2488 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max
);
2492 new_broot
= (xfs_bmbt_block_t
*)kmem_alloc(new_size
, KM_SLEEP
);
2494 * First copy over the btree block header.
2496 memcpy(new_broot
, ifp
->if_broot
, sizeof(xfs_bmbt_block_t
));
2499 ifp
->if_flags
&= ~XFS_IFBROOT
;
2503 * Only copy the records and pointers if there are any.
2507 * First copy the records.
2509 op
= (char *)XFS_BMAP_BROOT_REC_ADDR(ifp
->if_broot
, 1,
2510 ifp
->if_broot_bytes
);
2511 np
= (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot
, 1,
2513 memcpy(np
, op
, new_max
* (uint
)sizeof(xfs_bmbt_rec_t
));
2516 * Then copy the pointers.
2518 op
= (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp
->if_broot
, 1,
2519 ifp
->if_broot_bytes
);
2520 np
= (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot
, 1,
2522 memcpy(np
, op
, new_max
* (uint
)sizeof(xfs_dfsbno_t
));
2524 kmem_free(ifp
->if_broot
, ifp
->if_broot_bytes
);
2525 ifp
->if_broot
= new_broot
;
2526 ifp
->if_broot_bytes
= (int)new_size
;
2527 ASSERT(ifp
->if_broot_bytes
<=
2528 XFS_IFORK_SIZE(ip
, whichfork
) + XFS_BROOT_SIZE_ADJ
);
2534 * This is called when the amount of space needed for if_data
2535 * is increased or decreased. The change in size is indicated by
2536 * the number of bytes that need to be added or deleted in the
2537 * byte_diff parameter.
2539 * If the amount of space needed has decreased below the size of the
2540 * inline buffer, then switch to using the inline buffer. Otherwise,
2541 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2542 * to what is needed.
2544 * ip -- the inode whose if_data area is changing
2545 * byte_diff -- the change in the number of bytes, positive or negative,
2546 * requested for the if_data array.
2558 if (byte_diff
== 0) {
2562 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2563 new_size
= (int)ifp
->if_bytes
+ byte_diff
;
2564 ASSERT(new_size
>= 0);
2566 if (new_size
== 0) {
2567 if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
2568 kmem_free(ifp
->if_u1
.if_data
, ifp
->if_real_bytes
);
2570 ifp
->if_u1
.if_data
= NULL
;
2572 } else if (new_size
<= sizeof(ifp
->if_u2
.if_inline_data
)) {
2574 * If the valid extents/data can fit in if_inline_ext/data,
2575 * copy them from the malloc'd vector and free it.
2577 if (ifp
->if_u1
.if_data
== NULL
) {
2578 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
2579 } else if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
2580 ASSERT(ifp
->if_real_bytes
!= 0);
2581 memcpy(ifp
->if_u2
.if_inline_data
, ifp
->if_u1
.if_data
,
2583 kmem_free(ifp
->if_u1
.if_data
, ifp
->if_real_bytes
);
2584 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
2589 * Stuck with malloc/realloc.
2590 * For inline data, the underlying buffer must be
2591 * a multiple of 4 bytes in size so that it can be
2592 * logged and stay on word boundaries. We enforce
2595 real_size
= roundup(new_size
, 4);
2596 if (ifp
->if_u1
.if_data
== NULL
) {
2597 ASSERT(ifp
->if_real_bytes
== 0);
2598 ifp
->if_u1
.if_data
= kmem_alloc(real_size
, KM_SLEEP
);
2599 } else if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
2601 * Only do the realloc if the underlying size
2602 * is really changing.
2604 if (ifp
->if_real_bytes
!= real_size
) {
2605 ifp
->if_u1
.if_data
=
2606 kmem_realloc(ifp
->if_u1
.if_data
,
2612 ASSERT(ifp
->if_real_bytes
== 0);
2613 ifp
->if_u1
.if_data
= kmem_alloc(real_size
, KM_SLEEP
);
2614 memcpy(ifp
->if_u1
.if_data
, ifp
->if_u2
.if_inline_data
,
2618 ifp
->if_real_bytes
= real_size
;
2619 ifp
->if_bytes
= new_size
;
2620 ASSERT(ifp
->if_bytes
<= XFS_IFORK_SIZE(ip
, whichfork
));
2627 * Map inode to disk block and offset.
2629 * mp -- the mount point structure for the current file system
2630 * tp -- the current transaction
2631 * ino -- the inode number of the inode to be located
2632 * imap -- this structure is filled in with the information necessary
2633 * to retrieve the given inode from disk
2634 * flags -- flags to pass to xfs_dilocate indicating whether or not
2635 * lookups in the inode btree were OK or not
2645 xfs_fsblock_t fsbno
;
2650 fsbno
= imap
->im_blkno
?
2651 XFS_DADDR_TO_FSB(mp
, imap
->im_blkno
) : NULLFSBLOCK
;
2652 error
= xfs_dilocate(mp
, tp
, ino
, &fsbno
, &len
, &off
, flags
);
2656 imap
->im_blkno
= XFS_FSB_TO_DADDR(mp
, fsbno
);
2657 imap
->im_len
= XFS_FSB_TO_BB(mp
, len
);
2658 imap
->im_agblkno
= XFS_FSB_TO_AGBNO(mp
, fsbno
);
2659 imap
->im_ioffset
= (ushort
)off
;
2660 imap
->im_boffset
= (ushort
)(off
<< mp
->m_sb
.sb_inodelog
);
2671 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2672 if (ifp
->if_broot
!= NULL
) {
2673 kmem_free(ifp
->if_broot
, ifp
->if_broot_bytes
);
2674 ifp
->if_broot
= NULL
;
2678 * If the format is local, then we can't have an extents
2679 * array so just look for an inline data array. If we're
2680 * not local then we may or may not have an extents list,
2681 * so check and free it up if we do.
2683 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
) {
2684 if ((ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) &&
2685 (ifp
->if_u1
.if_data
!= NULL
)) {
2686 ASSERT(ifp
->if_real_bytes
!= 0);
2687 kmem_free(ifp
->if_u1
.if_data
, ifp
->if_real_bytes
);
2688 ifp
->if_u1
.if_data
= NULL
;
2689 ifp
->if_real_bytes
= 0;
2691 } else if ((ifp
->if_flags
& XFS_IFEXTENTS
) &&
2692 ((ifp
->if_flags
& XFS_IFEXTIREC
) ||
2693 ((ifp
->if_u1
.if_extents
!= NULL
) &&
2694 (ifp
->if_u1
.if_extents
!= ifp
->if_u2
.if_inline_ext
)))) {
2695 ASSERT(ifp
->if_real_bytes
!= 0);
2696 xfs_iext_destroy(ifp
);
2698 ASSERT(ifp
->if_u1
.if_extents
== NULL
||
2699 ifp
->if_u1
.if_extents
== ifp
->if_u2
.if_inline_ext
);
2700 ASSERT(ifp
->if_real_bytes
== 0);
2701 if (whichfork
== XFS_ATTR_FORK
) {
2702 kmem_zone_free(xfs_ifork_zone
, ip
->i_afp
);
2708 * This is called free all the memory associated with an inode.
2709 * It must free the inode itself and any buffers allocated for
2710 * if_extents/if_data and if_broot. It must also free the lock
2711 * associated with the inode.
2718 switch (ip
->i_d
.di_mode
& S_IFMT
) {
2722 xfs_idestroy_fork(ip
, XFS_DATA_FORK
);
2726 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
2727 mrfree(&ip
->i_lock
);
2728 mrfree(&ip
->i_iolock
);
2729 freesema(&ip
->i_flock
);
2730 #ifdef XFS_BMAP_TRACE
2731 ktrace_free(ip
->i_xtrace
);
2733 #ifdef XFS_BMBT_TRACE
2734 ktrace_free(ip
->i_btrace
);
2737 ktrace_free(ip
->i_rwtrace
);
2739 #ifdef XFS_ILOCK_TRACE
2740 ktrace_free(ip
->i_lock_trace
);
2742 #ifdef XFS_DIR2_TRACE
2743 ktrace_free(ip
->i_dir_trace
);
2747 * Only if we are shutting down the fs will we see an
2748 * inode still in the AIL. If it is there, we should remove
2749 * it to prevent a use-after-free from occurring.
2751 xfs_mount_t
*mp
= ip
->i_mount
;
2752 xfs_log_item_t
*lip
= &ip
->i_itemp
->ili_item
;
2755 ASSERT(((lip
->li_flags
& XFS_LI_IN_AIL
) == 0) ||
2756 XFS_FORCED_SHUTDOWN(ip
->i_mount
));
2757 if (lip
->li_flags
& XFS_LI_IN_AIL
) {
2759 if (lip
->li_flags
& XFS_LI_IN_AIL
)
2760 xfs_trans_delete_ail(mp
, lip
, s
);
2764 xfs_inode_item_destroy(ip
);
2766 kmem_zone_free(xfs_inode_zone
, ip
);
2771 * Increment the pin count of the given buffer.
2772 * This value is protected by ipinlock spinlock in the mount structure.
2778 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
));
2780 atomic_inc(&ip
->i_pincount
);
2784 * Decrement the pin count of the given inode, and wake up
2785 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2786 * inode must have been previously pinned with a call to xfs_ipin().
2792 ASSERT(atomic_read(&ip
->i_pincount
) > 0);
2794 if (atomic_dec_and_lock(&ip
->i_pincount
, &ip
->i_flags_lock
)) {
2797 * If the inode is currently being reclaimed, the link between
2798 * the bhv_vnode and the xfs_inode will be broken after the
2799 * XFS_IRECLAIM* flag is set. Hence, if these flags are not
2800 * set, then we can move forward and mark the linux inode dirty
2801 * knowing that it is still valid as it won't freed until after
2802 * the bhv_vnode<->xfs_inode link is broken in xfs_reclaim. The
2803 * i_flags_lock is used to synchronise the setting of the
2804 * XFS_IRECLAIM* flags and the breaking of the link, and so we
2805 * can execute atomically w.r.t to reclaim by holding this lock
2808 * However, we still need to issue the unpin wakeup call as the
2809 * inode reclaim may be blocked waiting for the inode to become
2813 if (!__xfs_iflags_test(ip
, XFS_IRECLAIM
|XFS_IRECLAIMABLE
)) {
2814 bhv_vnode_t
*vp
= XFS_ITOV_NULL(ip
);
2815 struct inode
*inode
= NULL
;
2818 inode
= vn_to_inode(vp
);
2819 BUG_ON(inode
->i_state
& I_CLEAR
);
2821 /* make sync come back and flush this inode */
2822 if (!(inode
->i_state
& (I_NEW
|I_FREEING
)))
2823 mark_inode_dirty_sync(inode
);
2825 spin_unlock(&ip
->i_flags_lock
);
2826 wake_up(&ip
->i_ipin_wait
);
2831 * This is called to wait for the given inode to be unpinned.
2832 * It will sleep until this happens. The caller must have the
2833 * inode locked in at least shared mode so that the buffer cannot
2834 * be subsequently pinned once someone is waiting for it to be
2841 xfs_inode_log_item_t
*iip
;
2844 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
| MR_ACCESS
));
2846 if (atomic_read(&ip
->i_pincount
) == 0) {
2851 if (iip
&& iip
->ili_last_lsn
) {
2852 lsn
= iip
->ili_last_lsn
;
2858 * Give the log a push so we don't wait here too long.
2860 xfs_log_force(ip
->i_mount
, lsn
, XFS_LOG_FORCE
);
2862 wait_event(ip
->i_ipin_wait
, (atomic_read(&ip
->i_pincount
) == 0));
2867 * xfs_iextents_copy()
2869 * This is called to copy the REAL extents (as opposed to the delayed
2870 * allocation extents) from the inode into the given buffer. It
2871 * returns the number of bytes copied into the buffer.
2873 * If there are no delayed allocation extents, then we can just
2874 * memcpy() the extents into the buffer. Otherwise, we need to
2875 * examine each extent in turn and skip those which are delayed.
2887 xfs_fsblock_t start_block
;
2889 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2890 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
|MR_ACCESS
));
2891 ASSERT(ifp
->if_bytes
> 0);
2893 nrecs
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
2894 XFS_BMAP_TRACE_EXLIST(ip
, nrecs
, whichfork
);
2898 * There are some delayed allocation extents in the
2899 * inode, so copy the extents one at a time and skip
2900 * the delayed ones. There must be at least one
2901 * non-delayed extent.
2904 for (i
= 0; i
< nrecs
; i
++) {
2905 xfs_bmbt_rec_host_t
*ep
= xfs_iext_get_ext(ifp
, i
);
2906 start_block
= xfs_bmbt_get_startblock(ep
);
2907 if (ISNULLSTARTBLOCK(start_block
)) {
2909 * It's a delayed allocation extent, so skip it.
2914 /* Translate to on disk format */
2915 put_unaligned(cpu_to_be64(ep
->l0
), &dp
->l0
);
2916 put_unaligned(cpu_to_be64(ep
->l1
), &dp
->l1
);
2920 ASSERT(copied
!= 0);
2921 xfs_validate_extents(ifp
, copied
, XFS_EXTFMT_INODE(ip
));
2923 return (copied
* (uint
)sizeof(xfs_bmbt_rec_t
));
2927 * Each of the following cases stores data into the same region
2928 * of the on-disk inode, so only one of them can be valid at
2929 * any given time. While it is possible to have conflicting formats
2930 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2931 * in EXTENTS format, this can only happen when the fork has
2932 * changed formats after being modified but before being flushed.
2933 * In these cases, the format always takes precedence, because the
2934 * format indicates the current state of the fork.
2941 xfs_inode_log_item_t
*iip
,
2948 #ifdef XFS_TRANS_DEBUG
2951 static const short brootflag
[2] =
2952 { XFS_ILOG_DBROOT
, XFS_ILOG_ABROOT
};
2953 static const short dataflag
[2] =
2954 { XFS_ILOG_DDATA
, XFS_ILOG_ADATA
};
2955 static const short extflag
[2] =
2956 { XFS_ILOG_DEXT
, XFS_ILOG_AEXT
};
2960 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2962 * This can happen if we gave up in iformat in an error path,
2963 * for the attribute fork.
2966 ASSERT(whichfork
== XFS_ATTR_FORK
);
2969 cp
= XFS_DFORK_PTR(dip
, whichfork
);
2971 switch (XFS_IFORK_FORMAT(ip
, whichfork
)) {
2972 case XFS_DINODE_FMT_LOCAL
:
2973 if ((iip
->ili_format
.ilf_fields
& dataflag
[whichfork
]) &&
2974 (ifp
->if_bytes
> 0)) {
2975 ASSERT(ifp
->if_u1
.if_data
!= NULL
);
2976 ASSERT(ifp
->if_bytes
<= XFS_IFORK_SIZE(ip
, whichfork
));
2977 memcpy(cp
, ifp
->if_u1
.if_data
, ifp
->if_bytes
);
2981 case XFS_DINODE_FMT_EXTENTS
:
2982 ASSERT((ifp
->if_flags
& XFS_IFEXTENTS
) ||
2983 !(iip
->ili_format
.ilf_fields
& extflag
[whichfork
]));
2984 ASSERT((xfs_iext_get_ext(ifp
, 0) != NULL
) ||
2985 (ifp
->if_bytes
== 0));
2986 ASSERT((xfs_iext_get_ext(ifp
, 0) == NULL
) ||
2987 (ifp
->if_bytes
> 0));
2988 if ((iip
->ili_format
.ilf_fields
& extflag
[whichfork
]) &&
2989 (ifp
->if_bytes
> 0)) {
2990 ASSERT(XFS_IFORK_NEXTENTS(ip
, whichfork
) > 0);
2991 (void)xfs_iextents_copy(ip
, (xfs_bmbt_rec_t
*)cp
,
2996 case XFS_DINODE_FMT_BTREE
:
2997 if ((iip
->ili_format
.ilf_fields
& brootflag
[whichfork
]) &&
2998 (ifp
->if_broot_bytes
> 0)) {
2999 ASSERT(ifp
->if_broot
!= NULL
);
3000 ASSERT(ifp
->if_broot_bytes
<=
3001 (XFS_IFORK_SIZE(ip
, whichfork
) +
3002 XFS_BROOT_SIZE_ADJ
));
3003 xfs_bmbt_to_bmdr(ifp
->if_broot
, ifp
->if_broot_bytes
,
3004 (xfs_bmdr_block_t
*)cp
,
3005 XFS_DFORK_SIZE(dip
, mp
, whichfork
));
3009 case XFS_DINODE_FMT_DEV
:
3010 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_DEV
) {
3011 ASSERT(whichfork
== XFS_DATA_FORK
);
3012 INT_SET(dip
->di_u
.di_dev
, ARCH_CONVERT
, ip
->i_df
.if_u2
.if_rdev
);
3016 case XFS_DINODE_FMT_UUID
:
3017 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_UUID
) {
3018 ASSERT(whichfork
== XFS_DATA_FORK
);
3019 memcpy(&dip
->di_u
.di_muuid
, &ip
->i_df
.if_u2
.if_uuid
,
3033 * xfs_iflush() will write a modified inode's changes out to the
3034 * inode's on disk home. The caller must have the inode lock held
3035 * in at least shared mode and the inode flush semaphore must be
3036 * held as well. The inode lock will still be held upon return from
3037 * the call and the caller is free to unlock it.
3038 * The inode flush lock will be unlocked when the inode reaches the disk.
3039 * The flags indicate how the inode's buffer should be written out.
3046 xfs_inode_log_item_t
*iip
;
3054 int clcount
; /* count of inodes clustered */
3056 enum { INT_DELWRI
= (1 << 0), INT_ASYNC
= (1 << 1) };
3059 XFS_STATS_INC(xs_iflush_count
);
3061 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
|MR_ACCESS
));
3062 ASSERT(issemalocked(&(ip
->i_flock
)));
3063 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
3064 ip
->i_d
.di_nextents
> ip
->i_df
.if_ext_max
);
3070 * If the inode isn't dirty, then just release the inode
3071 * flush lock and do nothing.
3073 if ((ip
->i_update_core
== 0) &&
3074 ((iip
== NULL
) || !(iip
->ili_format
.ilf_fields
& XFS_ILOG_ALL
))) {
3075 ASSERT((iip
!= NULL
) ?
3076 !(iip
->ili_item
.li_flags
& XFS_LI_IN_AIL
) : 1);
3082 * We can't flush the inode until it is unpinned, so
3083 * wait for it. We know noone new can pin it, because
3084 * we are holding the inode lock shared and you need
3085 * to hold it exclusively to pin the inode.
3087 xfs_iunpin_wait(ip
);
3090 * This may have been unpinned because the filesystem is shutting
3091 * down forcibly. If that's the case we must not write this inode
3092 * to disk, because the log record didn't make it to disk!
3094 if (XFS_FORCED_SHUTDOWN(mp
)) {
3095 ip
->i_update_core
= 0;
3097 iip
->ili_format
.ilf_fields
= 0;
3099 return XFS_ERROR(EIO
);
3103 * Get the buffer containing the on-disk inode.
3105 error
= xfs_itobp(mp
, NULL
, ip
, &dip
, &bp
, 0, 0);
3112 * Decide how buffer will be flushed out. This is done before
3113 * the call to xfs_iflush_int because this field is zeroed by it.
3115 if (iip
!= NULL
&& iip
->ili_format
.ilf_fields
!= 0) {
3117 * Flush out the inode buffer according to the directions
3118 * of the caller. In the cases where the caller has given
3119 * us a choice choose the non-delwri case. This is because
3120 * the inode is in the AIL and we need to get it out soon.
3123 case XFS_IFLUSH_SYNC
:
3124 case XFS_IFLUSH_DELWRI_ELSE_SYNC
:
3127 case XFS_IFLUSH_ASYNC
:
3128 case XFS_IFLUSH_DELWRI_ELSE_ASYNC
:
3131 case XFS_IFLUSH_DELWRI
:
3141 case XFS_IFLUSH_DELWRI_ELSE_SYNC
:
3142 case XFS_IFLUSH_DELWRI_ELSE_ASYNC
:
3143 case XFS_IFLUSH_DELWRI
:
3146 case XFS_IFLUSH_ASYNC
:
3149 case XFS_IFLUSH_SYNC
:
3160 * First flush out the inode that xfs_iflush was called with.
3162 error
= xfs_iflush_int(ip
, bp
);
3169 * see if other inodes can be gathered into this write
3172 ip
->i_chash
->chl_buf
= bp
;
3174 ch
= XFS_CHASH(mp
, ip
->i_blkno
);
3175 s
= mutex_spinlock(&ch
->ch_lock
);
3178 for (iq
= ip
->i_cnext
; iq
!= ip
; iq
= iq
->i_cnext
) {
3180 * Do an un-protected check to see if the inode is dirty and
3181 * is a candidate for flushing. These checks will be repeated
3182 * later after the appropriate locks are acquired.
3185 if ((iq
->i_update_core
== 0) &&
3187 !(iip
->ili_format
.ilf_fields
& XFS_ILOG_ALL
)) &&
3188 xfs_ipincount(iq
) == 0) {
3193 * Try to get locks. If any are unavailable,
3194 * then this inode cannot be flushed and is skipped.
3197 /* get inode locks (just i_lock) */
3198 if (xfs_ilock_nowait(iq
, XFS_ILOCK_SHARED
)) {
3199 /* get inode flush lock */
3200 if (xfs_iflock_nowait(iq
)) {
3201 /* check if pinned */
3202 if (xfs_ipincount(iq
) == 0) {
3203 /* arriving here means that
3204 * this inode can be flushed.
3205 * first re-check that it's
3209 if ((iq
->i_update_core
!= 0)||
3211 (iip
->ili_format
.ilf_fields
& XFS_ILOG_ALL
))) {
3213 error
= xfs_iflush_int(iq
, bp
);
3217 goto cluster_corrupt_out
;
3226 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
3229 mutex_spinunlock(&ch
->ch_lock
, s
);
3232 XFS_STATS_INC(xs_icluster_flushcnt
);
3233 XFS_STATS_ADD(xs_icluster_flushinode
, clcount
);
3237 * If the buffer is pinned then push on the log so we won't
3238 * get stuck waiting in the write for too long.
3240 if (XFS_BUF_ISPINNED(bp
)){
3241 xfs_log_force(mp
, (xfs_lsn_t
)0, XFS_LOG_FORCE
);
3244 if (flags
& INT_DELWRI
) {
3245 xfs_bdwrite(mp
, bp
);
3246 } else if (flags
& INT_ASYNC
) {
3247 xfs_bawrite(mp
, bp
);
3249 error
= xfs_bwrite(mp
, bp
);
3255 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
3256 xfs_iflush_abort(ip
);
3258 * Unlocks the flush lock
3260 return XFS_ERROR(EFSCORRUPTED
);
3262 cluster_corrupt_out
:
3263 /* Corruption detected in the clustering loop. Invalidate the
3264 * inode buffer and shut down the filesystem.
3266 mutex_spinunlock(&ch
->ch_lock
, s
);
3269 * Clean up the buffer. If it was B_DELWRI, just release it --
3270 * brelse can handle it with no problems. If not, shut down the
3271 * filesystem before releasing the buffer.
3273 if ((bufwasdelwri
= XFS_BUF_ISDELAYWRITE(bp
))) {
3277 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
3281 * Just like incore_relse: if we have b_iodone functions,
3282 * mark the buffer as an error and call them. Otherwise
3283 * mark it as stale and brelse.
3285 if (XFS_BUF_IODONE_FUNC(bp
)) {
3286 XFS_BUF_CLR_BDSTRAT_FUNC(bp
);
3290 XFS_BUF_ERROR(bp
,EIO
);
3298 xfs_iflush_abort(iq
);
3300 * Unlocks the flush lock
3302 return XFS_ERROR(EFSCORRUPTED
);
3311 xfs_inode_log_item_t
*iip
;
3314 #ifdef XFS_TRANS_DEBUG
3319 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
|MR_ACCESS
));
3320 ASSERT(issemalocked(&(ip
->i_flock
)));
3321 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
3322 ip
->i_d
.di_nextents
> ip
->i_df
.if_ext_max
);
3329 * If the inode isn't dirty, then just release the inode
3330 * flush lock and do nothing.
3332 if ((ip
->i_update_core
== 0) &&
3333 ((iip
== NULL
) || !(iip
->ili_format
.ilf_fields
& XFS_ILOG_ALL
))) {
3338 /* set *dip = inode's place in the buffer */
3339 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, ip
->i_boffset
);
3342 * Clear i_update_core before copying out the data.
3343 * This is for coordination with our timestamp updates
3344 * that don't hold the inode lock. They will always
3345 * update the timestamps BEFORE setting i_update_core,
3346 * so if we clear i_update_core after they set it we
3347 * are guaranteed to see their updates to the timestamps.
3348 * I believe that this depends on strongly ordered memory
3349 * semantics, but we have that. We use the SYNCHRONIZE
3350 * macro to make sure that the compiler does not reorder
3351 * the i_update_core access below the data copy below.
3353 ip
->i_update_core
= 0;
3357 * Make sure to get the latest atime from the Linux inode.
3359 xfs_synchronize_atime(ip
);
3361 if (XFS_TEST_ERROR(INT_GET(dip
->di_core
.di_magic
,ARCH_CONVERT
) != XFS_DINODE_MAGIC
,
3362 mp
, XFS_ERRTAG_IFLUSH_1
, XFS_RANDOM_IFLUSH_1
)) {
3363 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3364 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3365 ip
->i_ino
, (int) INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
), dip
);
3368 if (XFS_TEST_ERROR(ip
->i_d
.di_magic
!= XFS_DINODE_MAGIC
,
3369 mp
, XFS_ERRTAG_IFLUSH_2
, XFS_RANDOM_IFLUSH_2
)) {
3370 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3371 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3372 ip
->i_ino
, ip
, ip
->i_d
.di_magic
);
3375 if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
) {
3377 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3378 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
),
3379 mp
, XFS_ERRTAG_IFLUSH_3
, XFS_RANDOM_IFLUSH_3
)) {
3380 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3381 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3385 } else if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFDIR
) {
3387 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3388 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
) &&
3389 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_LOCAL
),
3390 mp
, XFS_ERRTAG_IFLUSH_4
, XFS_RANDOM_IFLUSH_4
)) {
3391 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3392 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3397 if (XFS_TEST_ERROR(ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
>
3398 ip
->i_d
.di_nblocks
, mp
, XFS_ERRTAG_IFLUSH_5
,
3399 XFS_RANDOM_IFLUSH_5
)) {
3400 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3401 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3403 ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
,
3408 if (XFS_TEST_ERROR(ip
->i_d
.di_forkoff
> mp
->m_sb
.sb_inodesize
,
3409 mp
, XFS_ERRTAG_IFLUSH_6
, XFS_RANDOM_IFLUSH_6
)) {
3410 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3411 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3412 ip
->i_ino
, ip
->i_d
.di_forkoff
, ip
);
3416 * bump the flush iteration count, used to detect flushes which
3417 * postdate a log record during recovery.
3420 ip
->i_d
.di_flushiter
++;
3423 * Copy the dirty parts of the inode into the on-disk
3424 * inode. We always copy out the core of the inode,
3425 * because if the inode is dirty at all the core must
3428 xfs_xlate_dinode_core((xfs_caddr_t
)&(dip
->di_core
), &(ip
->i_d
), -1);
3430 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3431 if (ip
->i_d
.di_flushiter
== DI_MAX_FLUSH
)
3432 ip
->i_d
.di_flushiter
= 0;
3435 * If this is really an old format inode and the superblock version
3436 * has not been updated to support only new format inodes, then
3437 * convert back to the old inode format. If the superblock version
3438 * has been updated, then make the conversion permanent.
3440 ASSERT(ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
||
3441 XFS_SB_VERSION_HASNLINK(&mp
->m_sb
));
3442 if (ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
) {
3443 if (!XFS_SB_VERSION_HASNLINK(&mp
->m_sb
)) {
3447 ASSERT(ip
->i_d
.di_nlink
<= XFS_MAXLINK_1
);
3448 INT_SET(dip
->di_core
.di_onlink
, ARCH_CONVERT
, ip
->i_d
.di_nlink
);
3451 * The superblock version has already been bumped,
3452 * so just make the conversion to the new inode
3455 ip
->i_d
.di_version
= XFS_DINODE_VERSION_2
;
3456 INT_SET(dip
->di_core
.di_version
, ARCH_CONVERT
, XFS_DINODE_VERSION_2
);
3457 ip
->i_d
.di_onlink
= 0;
3458 dip
->di_core
.di_onlink
= 0;
3459 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
3460 memset(&(dip
->di_core
.di_pad
[0]), 0,
3461 sizeof(dip
->di_core
.di_pad
));
3462 ASSERT(ip
->i_d
.di_projid
== 0);
3466 if (xfs_iflush_fork(ip
, dip
, iip
, XFS_DATA_FORK
, bp
) == EFSCORRUPTED
) {
3470 if (XFS_IFORK_Q(ip
)) {
3472 * The only error from xfs_iflush_fork is on the data fork.
3474 (void) xfs_iflush_fork(ip
, dip
, iip
, XFS_ATTR_FORK
, bp
);
3476 xfs_inobp_check(mp
, bp
);
3479 * We've recorded everything logged in the inode, so we'd
3480 * like to clear the ilf_fields bits so we don't log and
3481 * flush things unnecessarily. However, we can't stop
3482 * logging all this information until the data we've copied
3483 * into the disk buffer is written to disk. If we did we might
3484 * overwrite the copy of the inode in the log with all the
3485 * data after re-logging only part of it, and in the face of
3486 * a crash we wouldn't have all the data we need to recover.
3488 * What we do is move the bits to the ili_last_fields field.
3489 * When logging the inode, these bits are moved back to the
3490 * ilf_fields field. In the xfs_iflush_done() routine we
3491 * clear ili_last_fields, since we know that the information
3492 * those bits represent is permanently on disk. As long as
3493 * the flush completes before the inode is logged again, then
3494 * both ilf_fields and ili_last_fields will be cleared.
3496 * We can play with the ilf_fields bits here, because the inode
3497 * lock must be held exclusively in order to set bits there
3498 * and the flush lock protects the ili_last_fields bits.
3499 * Set ili_logged so the flush done
3500 * routine can tell whether or not to look in the AIL.
3501 * Also, store the current LSN of the inode so that we can tell
3502 * whether the item has moved in the AIL from xfs_iflush_done().
3503 * In order to read the lsn we need the AIL lock, because
3504 * it is a 64 bit value that cannot be read atomically.
3506 if (iip
!= NULL
&& iip
->ili_format
.ilf_fields
!= 0) {
3507 iip
->ili_last_fields
= iip
->ili_format
.ilf_fields
;
3508 iip
->ili_format
.ilf_fields
= 0;
3509 iip
->ili_logged
= 1;
3511 ASSERT(sizeof(xfs_lsn_t
) == 8); /* don't lock if it shrinks */
3513 iip
->ili_flush_lsn
= iip
->ili_item
.li_lsn
;
3517 * Attach the function xfs_iflush_done to the inode's
3518 * buffer. This will remove the inode from the AIL
3519 * and unlock the inode's flush lock when the inode is
3520 * completely written to disk.
3522 xfs_buf_attach_iodone(bp
, (void(*)(xfs_buf_t
*,xfs_log_item_t
*))
3523 xfs_iflush_done
, (xfs_log_item_t
*)iip
);
3525 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) != NULL
);
3526 ASSERT(XFS_BUF_IODONE_FUNC(bp
) != NULL
);
3529 * We're flushing an inode which is not in the AIL and has
3530 * not been logged but has i_update_core set. For this
3531 * case we can use a B_DELWRI flush and immediately drop
3532 * the inode flush lock because we can avoid the whole
3533 * AIL state thing. It's OK to drop the flush lock now,
3534 * because we've already locked the buffer and to do anything
3535 * you really need both.
3538 ASSERT(iip
->ili_logged
== 0);
3539 ASSERT(iip
->ili_last_fields
== 0);
3540 ASSERT((iip
->ili_item
.li_flags
& XFS_LI_IN_AIL
) == 0);
3548 return XFS_ERROR(EFSCORRUPTED
);
3553 * Flush all inactive inodes in mp.
3563 XFS_MOUNT_ILOCK(mp
);
3569 /* Make sure we skip markers inserted by sync */
3570 if (ip
->i_mount
== NULL
) {
3575 vp
= XFS_ITOV_NULL(ip
);
3577 XFS_MOUNT_IUNLOCK(mp
);
3578 xfs_finish_reclaim(ip
, 0, XFS_IFLUSH_ASYNC
);
3582 ASSERT(vn_count(vp
) == 0);
3585 } while (ip
!= mp
->m_inodes
);
3587 XFS_MOUNT_IUNLOCK(mp
);
3591 * xfs_iaccess: check accessibility of inode for mode.
3600 mode_t orgmode
= mode
;
3601 struct inode
*inode
= vn_to_inode(XFS_ITOV(ip
));
3603 if (mode
& S_IWUSR
) {
3604 umode_t imode
= inode
->i_mode
;
3606 if (IS_RDONLY(inode
) &&
3607 (S_ISREG(imode
) || S_ISDIR(imode
) || S_ISLNK(imode
)))
3608 return XFS_ERROR(EROFS
);
3610 if (IS_IMMUTABLE(inode
))
3611 return XFS_ERROR(EACCES
);
3615 * If there's an Access Control List it's used instead of
3618 if ((error
= _ACL_XFS_IACCESS(ip
, mode
, cr
)) != -1)
3619 return error
? XFS_ERROR(error
) : 0;
3621 if (current_fsuid(cr
) != ip
->i_d
.di_uid
) {
3623 if (!in_group_p((gid_t
)ip
->i_d
.di_gid
))
3628 * If the DACs are ok we don't need any capability check.
3630 if ((ip
->i_d
.di_mode
& mode
) == mode
)
3633 * Read/write DACs are always overridable.
3634 * Executable DACs are overridable if at least one exec bit is set.
3636 if (!(orgmode
& S_IXUSR
) ||
3637 (inode
->i_mode
& S_IXUGO
) || S_ISDIR(inode
->i_mode
))
3638 if (capable_cred(cr
, CAP_DAC_OVERRIDE
))
3641 if ((orgmode
== S_IRUSR
) ||
3642 (S_ISDIR(inode
->i_mode
) && (!(orgmode
& S_IWUSR
)))) {
3643 if (capable_cred(cr
, CAP_DAC_READ_SEARCH
))
3646 cmn_err(CE_NOTE
, "Ick: mode=%o, orgmode=%o", mode
, orgmode
);
3648 return XFS_ERROR(EACCES
);
3650 return XFS_ERROR(EACCES
);
3654 * xfs_iroundup: round up argument to next power of two
3663 if ((v
& (v
- 1)) == 0)
3665 ASSERT((v
& 0x80000000) == 0);
3666 if ((v
& (v
+ 1)) == 0)
3668 for (i
= 0, m
= 1; i
< 31; i
++, m
<<= 1) {
3672 if ((v
& (v
+ 1)) == 0)
3679 #ifdef XFS_ILOCK_TRACE
3680 ktrace_t
*xfs_ilock_trace_buf
;
3683 xfs_ilock_trace(xfs_inode_t
*ip
, int lock
, unsigned int lockflags
, inst_t
*ra
)
3685 ktrace_enter(ip
->i_lock_trace
,
3687 (void *)(unsigned long)lock
, /* 1 = LOCK, 3=UNLOCK, etc */
3688 (void *)(unsigned long)lockflags
, /* XFS_ILOCK_EXCL etc */
3689 (void *)ra
, /* caller of ilock */
3690 (void *)(unsigned long)current_cpu(),
3691 (void *)(unsigned long)current_pid(),
3692 NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
);
3697 * Return a pointer to the extent record at file index idx.
3699 xfs_bmbt_rec_host_t
*
3701 xfs_ifork_t
*ifp
, /* inode fork pointer */
3702 xfs_extnum_t idx
) /* index of target extent */
3705 if ((ifp
->if_flags
& XFS_IFEXTIREC
) && (idx
== 0)) {
3706 return ifp
->if_u1
.if_ext_irec
->er_extbuf
;
3707 } else if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3708 xfs_ext_irec_t
*erp
; /* irec pointer */
3709 int erp_idx
= 0; /* irec index */
3710 xfs_extnum_t page_idx
= idx
; /* ext index in target list */
3712 erp
= xfs_iext_idx_to_irec(ifp
, &page_idx
, &erp_idx
, 0);
3713 return &erp
->er_extbuf
[page_idx
];
3714 } else if (ifp
->if_bytes
) {
3715 return &ifp
->if_u1
.if_extents
[idx
];
3722 * Insert new item(s) into the extent records for incore inode
3723 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3727 xfs_ifork_t
*ifp
, /* inode fork pointer */
3728 xfs_extnum_t idx
, /* starting index of new items */
3729 xfs_extnum_t count
, /* number of inserted items */
3730 xfs_bmbt_irec_t
*new) /* items to insert */
3732 xfs_extnum_t i
; /* extent record index */
3734 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
3735 xfs_iext_add(ifp
, idx
, count
);
3736 for (i
= idx
; i
< idx
+ count
; i
++, new++)
3737 xfs_bmbt_set_all(xfs_iext_get_ext(ifp
, i
), new);
3741 * This is called when the amount of space required for incore file
3742 * extents needs to be increased. The ext_diff parameter stores the
3743 * number of new extents being added and the idx parameter contains
3744 * the extent index where the new extents will be added. If the new
3745 * extents are being appended, then we just need to (re)allocate and
3746 * initialize the space. Otherwise, if the new extents are being
3747 * inserted into the middle of the existing entries, a bit more work
3748 * is required to make room for the new extents to be inserted. The
3749 * caller is responsible for filling in the new extent entries upon
3754 xfs_ifork_t
*ifp
, /* inode fork pointer */
3755 xfs_extnum_t idx
, /* index to begin adding exts */
3756 int ext_diff
) /* number of extents to add */
3758 int byte_diff
; /* new bytes being added */
3759 int new_size
; /* size of extents after adding */
3760 xfs_extnum_t nextents
; /* number of extents in file */
3762 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3763 ASSERT((idx
>= 0) && (idx
<= nextents
));
3764 byte_diff
= ext_diff
* sizeof(xfs_bmbt_rec_t
);
3765 new_size
= ifp
->if_bytes
+ byte_diff
;
3767 * If the new number of extents (nextents + ext_diff)
3768 * fits inside the inode, then continue to use the inline
3771 if (nextents
+ ext_diff
<= XFS_INLINE_EXTS
) {
3772 if (idx
< nextents
) {
3773 memmove(&ifp
->if_u2
.if_inline_ext
[idx
+ ext_diff
],
3774 &ifp
->if_u2
.if_inline_ext
[idx
],
3775 (nextents
- idx
) * sizeof(xfs_bmbt_rec_t
));
3776 memset(&ifp
->if_u2
.if_inline_ext
[idx
], 0, byte_diff
);
3778 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
3779 ifp
->if_real_bytes
= 0;
3780 ifp
->if_lastex
= nextents
+ ext_diff
;
3783 * Otherwise use a linear (direct) extent list.
3784 * If the extents are currently inside the inode,
3785 * xfs_iext_realloc_direct will switch us from
3786 * inline to direct extent allocation mode.
3788 else if (nextents
+ ext_diff
<= XFS_LINEAR_EXTS
) {
3789 xfs_iext_realloc_direct(ifp
, new_size
);
3790 if (idx
< nextents
) {
3791 memmove(&ifp
->if_u1
.if_extents
[idx
+ ext_diff
],
3792 &ifp
->if_u1
.if_extents
[idx
],
3793 (nextents
- idx
) * sizeof(xfs_bmbt_rec_t
));
3794 memset(&ifp
->if_u1
.if_extents
[idx
], 0, byte_diff
);
3797 /* Indirection array */
3799 xfs_ext_irec_t
*erp
;
3803 ASSERT(nextents
+ ext_diff
> XFS_LINEAR_EXTS
);
3804 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3805 erp
= xfs_iext_idx_to_irec(ifp
, &page_idx
, &erp_idx
, 1);
3807 xfs_iext_irec_init(ifp
);
3808 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3809 erp
= ifp
->if_u1
.if_ext_irec
;
3811 /* Extents fit in target extent page */
3812 if (erp
&& erp
->er_extcount
+ ext_diff
<= XFS_LINEAR_EXTS
) {
3813 if (page_idx
< erp
->er_extcount
) {
3814 memmove(&erp
->er_extbuf
[page_idx
+ ext_diff
],
3815 &erp
->er_extbuf
[page_idx
],
3816 (erp
->er_extcount
- page_idx
) *
3817 sizeof(xfs_bmbt_rec_t
));
3818 memset(&erp
->er_extbuf
[page_idx
], 0, byte_diff
);
3820 erp
->er_extcount
+= ext_diff
;
3821 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, ext_diff
);
3823 /* Insert a new extent page */
3825 xfs_iext_add_indirect_multi(ifp
,
3826 erp_idx
, page_idx
, ext_diff
);
3829 * If extent(s) are being appended to the last page in
3830 * the indirection array and the new extent(s) don't fit
3831 * in the page, then erp is NULL and erp_idx is set to
3832 * the next index needed in the indirection array.
3835 int count
= ext_diff
;
3838 erp
= xfs_iext_irec_new(ifp
, erp_idx
);
3839 erp
->er_extcount
= count
;
3840 count
-= MIN(count
, (int)XFS_LINEAR_EXTS
);
3847 ifp
->if_bytes
= new_size
;
3851 * This is called when incore extents are being added to the indirection
3852 * array and the new extents do not fit in the target extent list. The
3853 * erp_idx parameter contains the irec index for the target extent list
3854 * in the indirection array, and the idx parameter contains the extent
3855 * index within the list. The number of extents being added is stored
3856 * in the count parameter.
3858 * |-------| |-------|
3859 * | | | | idx - number of extents before idx
3861 * | | | | count - number of extents being inserted at idx
3862 * |-------| |-------|
3863 * | count | | nex2 | nex2 - number of extents after idx + count
3864 * |-------| |-------|
3867 xfs_iext_add_indirect_multi(
3868 xfs_ifork_t
*ifp
, /* inode fork pointer */
3869 int erp_idx
, /* target extent irec index */
3870 xfs_extnum_t idx
, /* index within target list */
3871 int count
) /* new extents being added */
3873 int byte_diff
; /* new bytes being added */
3874 xfs_ext_irec_t
*erp
; /* pointer to irec entry */
3875 xfs_extnum_t ext_diff
; /* number of extents to add */
3876 xfs_extnum_t ext_cnt
; /* new extents still needed */
3877 xfs_extnum_t nex2
; /* extents after idx + count */
3878 xfs_bmbt_rec_t
*nex2_ep
= NULL
; /* temp list for nex2 extents */
3879 int nlists
; /* number of irec's (lists) */
3881 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
3882 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
3883 nex2
= erp
->er_extcount
- idx
;
3884 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
3887 * Save second part of target extent list
3888 * (all extents past */
3890 byte_diff
= nex2
* sizeof(xfs_bmbt_rec_t
);
3891 nex2_ep
= (xfs_bmbt_rec_t
*) kmem_alloc(byte_diff
, KM_SLEEP
);
3892 memmove(nex2_ep
, &erp
->er_extbuf
[idx
], byte_diff
);
3893 erp
->er_extcount
-= nex2
;
3894 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, -nex2
);
3895 memset(&erp
->er_extbuf
[idx
], 0, byte_diff
);
3899 * Add the new extents to the end of the target
3900 * list, then allocate new irec record(s) and
3901 * extent buffer(s) as needed to store the rest
3902 * of the new extents.
3905 ext_diff
= MIN(ext_cnt
, (int)XFS_LINEAR_EXTS
- erp
->er_extcount
);
3907 erp
->er_extcount
+= ext_diff
;
3908 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, ext_diff
);
3909 ext_cnt
-= ext_diff
;
3913 erp
= xfs_iext_irec_new(ifp
, erp_idx
);
3914 ext_diff
= MIN(ext_cnt
, (int)XFS_LINEAR_EXTS
);
3915 erp
->er_extcount
= ext_diff
;
3916 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, ext_diff
);
3917 ext_cnt
-= ext_diff
;
3920 /* Add nex2 extents back to indirection array */
3922 xfs_extnum_t ext_avail
;
3925 byte_diff
= nex2
* sizeof(xfs_bmbt_rec_t
);
3926 ext_avail
= XFS_LINEAR_EXTS
- erp
->er_extcount
;
3929 * If nex2 extents fit in the current page, append
3930 * nex2_ep after the new extents.
3932 if (nex2
<= ext_avail
) {
3933 i
= erp
->er_extcount
;
3936 * Otherwise, check if space is available in the
3939 else if ((erp_idx
< nlists
- 1) &&
3940 (nex2
<= (ext_avail
= XFS_LINEAR_EXTS
-
3941 ifp
->if_u1
.if_ext_irec
[erp_idx
+1].er_extcount
))) {
3944 /* Create a hole for nex2 extents */
3945 memmove(&erp
->er_extbuf
[nex2
], erp
->er_extbuf
,
3946 erp
->er_extcount
* sizeof(xfs_bmbt_rec_t
));
3949 * Final choice, create a new extent page for
3954 erp
= xfs_iext_irec_new(ifp
, erp_idx
);
3956 memmove(&erp
->er_extbuf
[i
], nex2_ep
, byte_diff
);
3957 kmem_free(nex2_ep
, byte_diff
);
3958 erp
->er_extcount
+= nex2
;
3959 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, nex2
);
3964 * This is called when the amount of space required for incore file
3965 * extents needs to be decreased. The ext_diff parameter stores the
3966 * number of extents to be removed and the idx parameter contains
3967 * the extent index where the extents will be removed from.
3969 * If the amount of space needed has decreased below the linear
3970 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3971 * extent array. Otherwise, use kmem_realloc() to adjust the
3972 * size to what is needed.
3976 xfs_ifork_t
*ifp
, /* inode fork pointer */
3977 xfs_extnum_t idx
, /* index to begin removing exts */
3978 int ext_diff
) /* number of extents to remove */
3980 xfs_extnum_t nextents
; /* number of extents in file */
3981 int new_size
; /* size of extents after removal */
3983 ASSERT(ext_diff
> 0);
3984 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3985 new_size
= (nextents
- ext_diff
) * sizeof(xfs_bmbt_rec_t
);
3987 if (new_size
== 0) {
3988 xfs_iext_destroy(ifp
);
3989 } else if (ifp
->if_flags
& XFS_IFEXTIREC
) {
3990 xfs_iext_remove_indirect(ifp
, idx
, ext_diff
);
3991 } else if (ifp
->if_real_bytes
) {
3992 xfs_iext_remove_direct(ifp
, idx
, ext_diff
);
3994 xfs_iext_remove_inline(ifp
, idx
, ext_diff
);
3996 ifp
->if_bytes
= new_size
;
4000 * This removes ext_diff extents from the inline buffer, beginning
4001 * at extent index idx.
4004 xfs_iext_remove_inline(
4005 xfs_ifork_t
*ifp
, /* inode fork pointer */
4006 xfs_extnum_t idx
, /* index to begin removing exts */
4007 int ext_diff
) /* number of extents to remove */
4009 int nextents
; /* number of extents in file */
4011 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
));
4012 ASSERT(idx
< XFS_INLINE_EXTS
);
4013 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4014 ASSERT(((nextents
- ext_diff
) > 0) &&
4015 (nextents
- ext_diff
) < XFS_INLINE_EXTS
);
4017 if (idx
+ ext_diff
< nextents
) {
4018 memmove(&ifp
->if_u2
.if_inline_ext
[idx
],
4019 &ifp
->if_u2
.if_inline_ext
[idx
+ ext_diff
],
4020 (nextents
- (idx
+ ext_diff
)) *
4021 sizeof(xfs_bmbt_rec_t
));
4022 memset(&ifp
->if_u2
.if_inline_ext
[nextents
- ext_diff
],
4023 0, ext_diff
* sizeof(xfs_bmbt_rec_t
));
4025 memset(&ifp
->if_u2
.if_inline_ext
[idx
], 0,
4026 ext_diff
* sizeof(xfs_bmbt_rec_t
));
4031 * This removes ext_diff extents from a linear (direct) extent list,
4032 * beginning at extent index idx. If the extents are being removed
4033 * from the end of the list (ie. truncate) then we just need to re-
4034 * allocate the list to remove the extra space. Otherwise, if the
4035 * extents are being removed from the middle of the existing extent
4036 * entries, then we first need to move the extent records beginning
4037 * at idx + ext_diff up in the list to overwrite the records being
4038 * removed, then remove the extra space via kmem_realloc.
4041 xfs_iext_remove_direct(
4042 xfs_ifork_t
*ifp
, /* inode fork pointer */
4043 xfs_extnum_t idx
, /* index to begin removing exts */
4044 int ext_diff
) /* number of extents to remove */
4046 xfs_extnum_t nextents
; /* number of extents in file */
4047 int new_size
; /* size of extents after removal */
4049 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
));
4050 new_size
= ifp
->if_bytes
-
4051 (ext_diff
* sizeof(xfs_bmbt_rec_t
));
4052 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4054 if (new_size
== 0) {
4055 xfs_iext_destroy(ifp
);
4058 /* Move extents up in the list (if needed) */
4059 if (idx
+ ext_diff
< nextents
) {
4060 memmove(&ifp
->if_u1
.if_extents
[idx
],
4061 &ifp
->if_u1
.if_extents
[idx
+ ext_diff
],
4062 (nextents
- (idx
+ ext_diff
)) *
4063 sizeof(xfs_bmbt_rec_t
));
4065 memset(&ifp
->if_u1
.if_extents
[nextents
- ext_diff
],
4066 0, ext_diff
* sizeof(xfs_bmbt_rec_t
));
4068 * Reallocate the direct extent list. If the extents
4069 * will fit inside the inode then xfs_iext_realloc_direct
4070 * will switch from direct to inline extent allocation
4073 xfs_iext_realloc_direct(ifp
, new_size
);
4074 ifp
->if_bytes
= new_size
;
4078 * This is called when incore extents are being removed from the
4079 * indirection array and the extents being removed span multiple extent
4080 * buffers. The idx parameter contains the file extent index where we
4081 * want to begin removing extents, and the count parameter contains
4082 * how many extents need to be removed.
4084 * |-------| |-------|
4085 * | nex1 | | | nex1 - number of extents before idx
4086 * |-------| | count |
4087 * | | | | count - number of extents being removed at idx
4088 * | count | |-------|
4089 * | | | nex2 | nex2 - number of extents after idx + count
4090 * |-------| |-------|
4093 xfs_iext_remove_indirect(
4094 xfs_ifork_t
*ifp
, /* inode fork pointer */
4095 xfs_extnum_t idx
, /* index to begin removing extents */
4096 int count
) /* number of extents to remove */
4098 xfs_ext_irec_t
*erp
; /* indirection array pointer */
4099 int erp_idx
= 0; /* indirection array index */
4100 xfs_extnum_t ext_cnt
; /* extents left to remove */
4101 xfs_extnum_t ext_diff
; /* extents to remove in current list */
4102 xfs_extnum_t nex1
; /* number of extents before idx */
4103 xfs_extnum_t nex2
; /* extents after idx + count */
4104 int nlists
; /* entries in indirection array */
4105 int page_idx
= idx
; /* index in target extent list */
4107 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4108 erp
= xfs_iext_idx_to_irec(ifp
, &page_idx
, &erp_idx
, 0);
4109 ASSERT(erp
!= NULL
);
4110 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4114 nex2
= MAX((erp
->er_extcount
- (nex1
+ ext_cnt
)), 0);
4115 ext_diff
= MIN(ext_cnt
, (erp
->er_extcount
- nex1
));
4117 * Check for deletion of entire list;
4118 * xfs_iext_irec_remove() updates extent offsets.
4120 if (ext_diff
== erp
->er_extcount
) {
4121 xfs_iext_irec_remove(ifp
, erp_idx
);
4122 ext_cnt
-= ext_diff
;
4125 ASSERT(erp_idx
< ifp
->if_real_bytes
/
4127 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4134 /* Move extents up (if needed) */
4136 memmove(&erp
->er_extbuf
[nex1
],
4137 &erp
->er_extbuf
[nex1
+ ext_diff
],
4138 nex2
* sizeof(xfs_bmbt_rec_t
));
4140 /* Zero out rest of page */
4141 memset(&erp
->er_extbuf
[nex1
+ nex2
], 0, (XFS_IEXT_BUFSZ
-
4142 ((nex1
+ nex2
) * sizeof(xfs_bmbt_rec_t
))));
4143 /* Update remaining counters */
4144 erp
->er_extcount
-= ext_diff
;
4145 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1, -ext_diff
);
4146 ext_cnt
-= ext_diff
;
4151 ifp
->if_bytes
-= count
* sizeof(xfs_bmbt_rec_t
);
4152 xfs_iext_irec_compact(ifp
);
4156 * Create, destroy, or resize a linear (direct) block of extents.
4159 xfs_iext_realloc_direct(
4160 xfs_ifork_t
*ifp
, /* inode fork pointer */
4161 int new_size
) /* new size of extents */
4163 int rnew_size
; /* real new size of extents */
4165 rnew_size
= new_size
;
4167 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
) ||
4168 ((new_size
>= 0) && (new_size
<= XFS_IEXT_BUFSZ
) &&
4169 (new_size
!= ifp
->if_real_bytes
)));
4171 /* Free extent records */
4172 if (new_size
== 0) {
4173 xfs_iext_destroy(ifp
);
4175 /* Resize direct extent list and zero any new bytes */
4176 else if (ifp
->if_real_bytes
) {
4177 /* Check if extents will fit inside the inode */
4178 if (new_size
<= XFS_INLINE_EXTS
* sizeof(xfs_bmbt_rec_t
)) {
4179 xfs_iext_direct_to_inline(ifp
, new_size
/
4180 (uint
)sizeof(xfs_bmbt_rec_t
));
4181 ifp
->if_bytes
= new_size
;
4184 if (!is_power_of_2(new_size
)){
4185 rnew_size
= xfs_iroundup(new_size
);
4187 if (rnew_size
!= ifp
->if_real_bytes
) {
4188 ifp
->if_u1
.if_extents
=
4189 kmem_realloc(ifp
->if_u1
.if_extents
,
4194 if (rnew_size
> ifp
->if_real_bytes
) {
4195 memset(&ifp
->if_u1
.if_extents
[ifp
->if_bytes
/
4196 (uint
)sizeof(xfs_bmbt_rec_t
)], 0,
4197 rnew_size
- ifp
->if_real_bytes
);
4201 * Switch from the inline extent buffer to a direct
4202 * extent list. Be sure to include the inline extent
4203 * bytes in new_size.
4206 new_size
+= ifp
->if_bytes
;
4207 if (!is_power_of_2(new_size
)) {
4208 rnew_size
= xfs_iroundup(new_size
);
4210 xfs_iext_inline_to_direct(ifp
, rnew_size
);
4212 ifp
->if_real_bytes
= rnew_size
;
4213 ifp
->if_bytes
= new_size
;
4217 * Switch from linear (direct) extent records to inline buffer.
4220 xfs_iext_direct_to_inline(
4221 xfs_ifork_t
*ifp
, /* inode fork pointer */
4222 xfs_extnum_t nextents
) /* number of extents in file */
4224 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
4225 ASSERT(nextents
<= XFS_INLINE_EXTS
);
4227 * The inline buffer was zeroed when we switched
4228 * from inline to direct extent allocation mode,
4229 * so we don't need to clear it here.
4231 memcpy(ifp
->if_u2
.if_inline_ext
, ifp
->if_u1
.if_extents
,
4232 nextents
* sizeof(xfs_bmbt_rec_t
));
4233 kmem_free(ifp
->if_u1
.if_extents
, ifp
->if_real_bytes
);
4234 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
4235 ifp
->if_real_bytes
= 0;
4239 * Switch from inline buffer to linear (direct) extent records.
4240 * new_size should already be rounded up to the next power of 2
4241 * by the caller (when appropriate), so use new_size as it is.
4242 * However, since new_size may be rounded up, we can't update
4243 * if_bytes here. It is the caller's responsibility to update
4244 * if_bytes upon return.
4247 xfs_iext_inline_to_direct(
4248 xfs_ifork_t
*ifp
, /* inode fork pointer */
4249 int new_size
) /* number of extents in file */
4251 ifp
->if_u1
.if_extents
= kmem_alloc(new_size
, KM_SLEEP
);
4252 memset(ifp
->if_u1
.if_extents
, 0, new_size
);
4253 if (ifp
->if_bytes
) {
4254 memcpy(ifp
->if_u1
.if_extents
, ifp
->if_u2
.if_inline_ext
,
4256 memset(ifp
->if_u2
.if_inline_ext
, 0, XFS_INLINE_EXTS
*
4257 sizeof(xfs_bmbt_rec_t
));
4259 ifp
->if_real_bytes
= new_size
;
4263 * Resize an extent indirection array to new_size bytes.
4266 xfs_iext_realloc_indirect(
4267 xfs_ifork_t
*ifp
, /* inode fork pointer */
4268 int new_size
) /* new indirection array size */
4270 int nlists
; /* number of irec's (ex lists) */
4271 int size
; /* current indirection array size */
4273 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4274 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4275 size
= nlists
* sizeof(xfs_ext_irec_t
);
4276 ASSERT(ifp
->if_real_bytes
);
4277 ASSERT((new_size
>= 0) && (new_size
!= size
));
4278 if (new_size
== 0) {
4279 xfs_iext_destroy(ifp
);
4281 ifp
->if_u1
.if_ext_irec
= (xfs_ext_irec_t
*)
4282 kmem_realloc(ifp
->if_u1
.if_ext_irec
,
4283 new_size
, size
, KM_SLEEP
);
4288 * Switch from indirection array to linear (direct) extent allocations.
4291 xfs_iext_indirect_to_direct(
4292 xfs_ifork_t
*ifp
) /* inode fork pointer */
4294 xfs_bmbt_rec_host_t
*ep
; /* extent record pointer */
4295 xfs_extnum_t nextents
; /* number of extents in file */
4296 int size
; /* size of file extents */
4298 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4299 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4300 ASSERT(nextents
<= XFS_LINEAR_EXTS
);
4301 size
= nextents
* sizeof(xfs_bmbt_rec_t
);
4303 xfs_iext_irec_compact_full(ifp
);
4304 ASSERT(ifp
->if_real_bytes
== XFS_IEXT_BUFSZ
);
4306 ep
= ifp
->if_u1
.if_ext_irec
->er_extbuf
;
4307 kmem_free(ifp
->if_u1
.if_ext_irec
, sizeof(xfs_ext_irec_t
));
4308 ifp
->if_flags
&= ~XFS_IFEXTIREC
;
4309 ifp
->if_u1
.if_extents
= ep
;
4310 ifp
->if_bytes
= size
;
4311 if (nextents
< XFS_LINEAR_EXTS
) {
4312 xfs_iext_realloc_direct(ifp
, size
);
4317 * Free incore file extents.
4321 xfs_ifork_t
*ifp
) /* inode fork pointer */
4323 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
4327 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4328 for (erp_idx
= nlists
- 1; erp_idx
>= 0 ; erp_idx
--) {
4329 xfs_iext_irec_remove(ifp
, erp_idx
);
4331 ifp
->if_flags
&= ~XFS_IFEXTIREC
;
4332 } else if (ifp
->if_real_bytes
) {
4333 kmem_free(ifp
->if_u1
.if_extents
, ifp
->if_real_bytes
);
4334 } else if (ifp
->if_bytes
) {
4335 memset(ifp
->if_u2
.if_inline_ext
, 0, XFS_INLINE_EXTS
*
4336 sizeof(xfs_bmbt_rec_t
));
4338 ifp
->if_u1
.if_extents
= NULL
;
4339 ifp
->if_real_bytes
= 0;
4344 * Return a pointer to the extent record for file system block bno.
4346 xfs_bmbt_rec_host_t
* /* pointer to found extent record */
4347 xfs_iext_bno_to_ext(
4348 xfs_ifork_t
*ifp
, /* inode fork pointer */
4349 xfs_fileoff_t bno
, /* block number to search for */
4350 xfs_extnum_t
*idxp
) /* index of target extent */
4352 xfs_bmbt_rec_host_t
*base
; /* pointer to first extent */
4353 xfs_filblks_t blockcount
= 0; /* number of blocks in extent */
4354 xfs_bmbt_rec_host_t
*ep
= NULL
; /* pointer to target extent */
4355 xfs_ext_irec_t
*erp
= NULL
; /* indirection array pointer */
4356 int high
; /* upper boundary in search */
4357 xfs_extnum_t idx
= 0; /* index of target extent */
4358 int low
; /* lower boundary in search */
4359 xfs_extnum_t nextents
; /* number of file extents */
4360 xfs_fileoff_t startoff
= 0; /* start offset of extent */
4362 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4363 if (nextents
== 0) {
4368 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
4369 /* Find target extent list */
4371 erp
= xfs_iext_bno_to_irec(ifp
, bno
, &erp_idx
);
4372 base
= erp
->er_extbuf
;
4373 high
= erp
->er_extcount
- 1;
4375 base
= ifp
->if_u1
.if_extents
;
4376 high
= nextents
- 1;
4378 /* Binary search extent records */
4379 while (low
<= high
) {
4380 idx
= (low
+ high
) >> 1;
4382 startoff
= xfs_bmbt_get_startoff(ep
);
4383 blockcount
= xfs_bmbt_get_blockcount(ep
);
4384 if (bno
< startoff
) {
4386 } else if (bno
>= startoff
+ blockcount
) {
4389 /* Convert back to file-based extent index */
4390 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
4391 idx
+= erp
->er_extoff
;
4397 /* Convert back to file-based extent index */
4398 if (ifp
->if_flags
& XFS_IFEXTIREC
) {
4399 idx
+= erp
->er_extoff
;
4401 if (bno
>= startoff
+ blockcount
) {
4402 if (++idx
== nextents
) {
4405 ep
= xfs_iext_get_ext(ifp
, idx
);
4413 * Return a pointer to the indirection array entry containing the
4414 * extent record for filesystem block bno. Store the index of the
4415 * target irec in *erp_idxp.
4417 xfs_ext_irec_t
* /* pointer to found extent record */
4418 xfs_iext_bno_to_irec(
4419 xfs_ifork_t
*ifp
, /* inode fork pointer */
4420 xfs_fileoff_t bno
, /* block number to search for */
4421 int *erp_idxp
) /* irec index of target ext list */
4423 xfs_ext_irec_t
*erp
= NULL
; /* indirection array pointer */
4424 xfs_ext_irec_t
*erp_next
; /* next indirection array entry */
4425 int erp_idx
; /* indirection array index */
4426 int nlists
; /* number of extent irec's (lists) */
4427 int high
; /* binary search upper limit */
4428 int low
; /* binary search lower limit */
4430 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4431 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4435 while (low
<= high
) {
4436 erp_idx
= (low
+ high
) >> 1;
4437 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4438 erp_next
= erp_idx
< nlists
- 1 ? erp
+ 1 : NULL
;
4439 if (bno
< xfs_bmbt_get_startoff(erp
->er_extbuf
)) {
4441 } else if (erp_next
&& bno
>=
4442 xfs_bmbt_get_startoff(erp_next
->er_extbuf
)) {
4448 *erp_idxp
= erp_idx
;
4453 * Return a pointer to the indirection array entry containing the
4454 * extent record at file extent index *idxp. Store the index of the
4455 * target irec in *erp_idxp and store the page index of the target
4456 * extent record in *idxp.
4459 xfs_iext_idx_to_irec(
4460 xfs_ifork_t
*ifp
, /* inode fork pointer */
4461 xfs_extnum_t
*idxp
, /* extent index (file -> page) */
4462 int *erp_idxp
, /* pointer to target irec */
4463 int realloc
) /* new bytes were just added */
4465 xfs_ext_irec_t
*prev
; /* pointer to previous irec */
4466 xfs_ext_irec_t
*erp
= NULL
; /* pointer to current irec */
4467 int erp_idx
; /* indirection array index */
4468 int nlists
; /* number of irec's (ex lists) */
4469 int high
; /* binary search upper limit */
4470 int low
; /* binary search lower limit */
4471 xfs_extnum_t page_idx
= *idxp
; /* extent index in target list */
4473 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4474 ASSERT(page_idx
>= 0 && page_idx
<=
4475 ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
));
4476 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4481 /* Binary search extent irec's */
4482 while (low
<= high
) {
4483 erp_idx
= (low
+ high
) >> 1;
4484 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4485 prev
= erp_idx
> 0 ? erp
- 1 : NULL
;
4486 if (page_idx
< erp
->er_extoff
|| (page_idx
== erp
->er_extoff
&&
4487 realloc
&& prev
&& prev
->er_extcount
< XFS_LINEAR_EXTS
)) {
4489 } else if (page_idx
> erp
->er_extoff
+ erp
->er_extcount
||
4490 (page_idx
== erp
->er_extoff
+ erp
->er_extcount
&&
4493 } else if (page_idx
== erp
->er_extoff
+ erp
->er_extcount
&&
4494 erp
->er_extcount
== XFS_LINEAR_EXTS
) {
4498 erp
= erp_idx
< nlists
? erp
+ 1 : NULL
;
4501 page_idx
-= erp
->er_extoff
;
4506 *erp_idxp
= erp_idx
;
4511 * Allocate and initialize an indirection array once the space needed
4512 * for incore extents increases above XFS_IEXT_BUFSZ.
4516 xfs_ifork_t
*ifp
) /* inode fork pointer */
4518 xfs_ext_irec_t
*erp
; /* indirection array pointer */
4519 xfs_extnum_t nextents
; /* number of extents in file */
4521 ASSERT(!(ifp
->if_flags
& XFS_IFEXTIREC
));
4522 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4523 ASSERT(nextents
<= XFS_LINEAR_EXTS
);
4525 erp
= (xfs_ext_irec_t
*)
4526 kmem_alloc(sizeof(xfs_ext_irec_t
), KM_SLEEP
);
4528 if (nextents
== 0) {
4529 ifp
->if_u1
.if_extents
= kmem_alloc(XFS_IEXT_BUFSZ
, KM_SLEEP
);
4530 } else if (!ifp
->if_real_bytes
) {
4531 xfs_iext_inline_to_direct(ifp
, XFS_IEXT_BUFSZ
);
4532 } else if (ifp
->if_real_bytes
< XFS_IEXT_BUFSZ
) {
4533 xfs_iext_realloc_direct(ifp
, XFS_IEXT_BUFSZ
);
4535 erp
->er_extbuf
= ifp
->if_u1
.if_extents
;
4536 erp
->er_extcount
= nextents
;
4539 ifp
->if_flags
|= XFS_IFEXTIREC
;
4540 ifp
->if_real_bytes
= XFS_IEXT_BUFSZ
;
4541 ifp
->if_bytes
= nextents
* sizeof(xfs_bmbt_rec_t
);
4542 ifp
->if_u1
.if_ext_irec
= erp
;
4548 * Allocate and initialize a new entry in the indirection array.
4552 xfs_ifork_t
*ifp
, /* inode fork pointer */
4553 int erp_idx
) /* index for new irec */
4555 xfs_ext_irec_t
*erp
; /* indirection array pointer */
4556 int i
; /* loop counter */
4557 int nlists
; /* number of irec's (ex lists) */
4559 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4560 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4562 /* Resize indirection array */
4563 xfs_iext_realloc_indirect(ifp
, ++nlists
*
4564 sizeof(xfs_ext_irec_t
));
4566 * Move records down in the array so the
4567 * new page can use erp_idx.
4569 erp
= ifp
->if_u1
.if_ext_irec
;
4570 for (i
= nlists
- 1; i
> erp_idx
; i
--) {
4571 memmove(&erp
[i
], &erp
[i
-1], sizeof(xfs_ext_irec_t
));
4573 ASSERT(i
== erp_idx
);
4575 /* Initialize new extent record */
4576 erp
= ifp
->if_u1
.if_ext_irec
;
4577 erp
[erp_idx
].er_extbuf
= kmem_alloc(XFS_IEXT_BUFSZ
, KM_SLEEP
);
4578 ifp
->if_real_bytes
= nlists
* XFS_IEXT_BUFSZ
;
4579 memset(erp
[erp_idx
].er_extbuf
, 0, XFS_IEXT_BUFSZ
);
4580 erp
[erp_idx
].er_extcount
= 0;
4581 erp
[erp_idx
].er_extoff
= erp_idx
> 0 ?
4582 erp
[erp_idx
-1].er_extoff
+ erp
[erp_idx
-1].er_extcount
: 0;
4583 return (&erp
[erp_idx
]);
4587 * Remove a record from the indirection array.
4590 xfs_iext_irec_remove(
4591 xfs_ifork_t
*ifp
, /* inode fork pointer */
4592 int erp_idx
) /* irec index to remove */
4594 xfs_ext_irec_t
*erp
; /* indirection array pointer */
4595 int i
; /* loop counter */
4596 int nlists
; /* number of irec's (ex lists) */
4598 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4599 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4600 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4601 if (erp
->er_extbuf
) {
4602 xfs_iext_irec_update_extoffs(ifp
, erp_idx
+ 1,
4604 kmem_free(erp
->er_extbuf
, XFS_IEXT_BUFSZ
);
4606 /* Compact extent records */
4607 erp
= ifp
->if_u1
.if_ext_irec
;
4608 for (i
= erp_idx
; i
< nlists
- 1; i
++) {
4609 memmove(&erp
[i
], &erp
[i
+1], sizeof(xfs_ext_irec_t
));
4612 * Manually free the last extent record from the indirection
4613 * array. A call to xfs_iext_realloc_indirect() with a size
4614 * of zero would result in a call to xfs_iext_destroy() which
4615 * would in turn call this function again, creating a nasty
4619 xfs_iext_realloc_indirect(ifp
,
4620 nlists
* sizeof(xfs_ext_irec_t
));
4622 kmem_free(ifp
->if_u1
.if_ext_irec
,
4623 sizeof(xfs_ext_irec_t
));
4625 ifp
->if_real_bytes
= nlists
* XFS_IEXT_BUFSZ
;
4629 * This is called to clean up large amounts of unused memory allocated
4630 * by the indirection array. Before compacting anything though, verify
4631 * that the indirection array is still needed and switch back to the
4632 * linear extent list (or even the inline buffer) if possible. The
4633 * compaction policy is as follows:
4635 * Full Compaction: Extents fit into a single page (or inline buffer)
4636 * Full Compaction: Extents occupy less than 10% of allocated space
4637 * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
4638 * No Compaction: Extents occupy at least 50% of allocated space
4641 xfs_iext_irec_compact(
4642 xfs_ifork_t
*ifp
) /* inode fork pointer */
4644 xfs_extnum_t nextents
; /* number of extents in file */
4645 int nlists
; /* number of irec's (ex lists) */
4647 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4648 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4649 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4651 if (nextents
== 0) {
4652 xfs_iext_destroy(ifp
);
4653 } else if (nextents
<= XFS_INLINE_EXTS
) {
4654 xfs_iext_indirect_to_direct(ifp
);
4655 xfs_iext_direct_to_inline(ifp
, nextents
);
4656 } else if (nextents
<= XFS_LINEAR_EXTS
) {
4657 xfs_iext_indirect_to_direct(ifp
);
4658 } else if (nextents
< (nlists
* XFS_LINEAR_EXTS
) >> 3) {
4659 xfs_iext_irec_compact_full(ifp
);
4660 } else if (nextents
< (nlists
* XFS_LINEAR_EXTS
) >> 1) {
4661 xfs_iext_irec_compact_pages(ifp
);
4666 * Combine extents from neighboring extent pages.
4669 xfs_iext_irec_compact_pages(
4670 xfs_ifork_t
*ifp
) /* inode fork pointer */
4672 xfs_ext_irec_t
*erp
, *erp_next
;/* pointers to irec entries */
4673 int erp_idx
= 0; /* indirection array index */
4674 int nlists
; /* number of irec's (ex lists) */
4676 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4677 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4678 while (erp_idx
< nlists
- 1) {
4679 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4681 if (erp_next
->er_extcount
<=
4682 (XFS_LINEAR_EXTS
- erp
->er_extcount
)) {
4683 memmove(&erp
->er_extbuf
[erp
->er_extcount
],
4684 erp_next
->er_extbuf
, erp_next
->er_extcount
*
4685 sizeof(xfs_bmbt_rec_t
));
4686 erp
->er_extcount
+= erp_next
->er_extcount
;
4688 * Free page before removing extent record
4689 * so er_extoffs don't get modified in
4690 * xfs_iext_irec_remove.
4692 kmem_free(erp_next
->er_extbuf
, XFS_IEXT_BUFSZ
);
4693 erp_next
->er_extbuf
= NULL
;
4694 xfs_iext_irec_remove(ifp
, erp_idx
+ 1);
4695 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4703 * Fully compact the extent records managed by the indirection array.
4706 xfs_iext_irec_compact_full(
4707 xfs_ifork_t
*ifp
) /* inode fork pointer */
4709 xfs_bmbt_rec_host_t
*ep
, *ep_next
; /* extent record pointers */
4710 xfs_ext_irec_t
*erp
, *erp_next
; /* extent irec pointers */
4711 int erp_idx
= 0; /* extent irec index */
4712 int ext_avail
; /* empty entries in ex list */
4713 int ext_diff
; /* number of exts to add */
4714 int nlists
; /* number of irec's (ex lists) */
4716 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4717 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4718 erp
= ifp
->if_u1
.if_ext_irec
;
4719 ep
= &erp
->er_extbuf
[erp
->er_extcount
];
4721 ep_next
= erp_next
->er_extbuf
;
4722 while (erp_idx
< nlists
- 1) {
4723 ext_avail
= XFS_LINEAR_EXTS
- erp
->er_extcount
;
4724 ext_diff
= MIN(ext_avail
, erp_next
->er_extcount
);
4725 memcpy(ep
, ep_next
, ext_diff
* sizeof(xfs_bmbt_rec_t
));
4726 erp
->er_extcount
+= ext_diff
;
4727 erp_next
->er_extcount
-= ext_diff
;
4728 /* Remove next page */
4729 if (erp_next
->er_extcount
== 0) {
4731 * Free page before removing extent record
4732 * so er_extoffs don't get modified in
4733 * xfs_iext_irec_remove.
4735 kmem_free(erp_next
->er_extbuf
,
4736 erp_next
->er_extcount
* sizeof(xfs_bmbt_rec_t
));
4737 erp_next
->er_extbuf
= NULL
;
4738 xfs_iext_irec_remove(ifp
, erp_idx
+ 1);
4739 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4740 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4741 /* Update next page */
4743 /* Move rest of page up to become next new page */
4744 memmove(erp_next
->er_extbuf
, ep_next
,
4745 erp_next
->er_extcount
* sizeof(xfs_bmbt_rec_t
));
4746 ep_next
= erp_next
->er_extbuf
;
4747 memset(&ep_next
[erp_next
->er_extcount
], 0,
4748 (XFS_LINEAR_EXTS
- erp_next
->er_extcount
) *
4749 sizeof(xfs_bmbt_rec_t
));
4751 if (erp
->er_extcount
== XFS_LINEAR_EXTS
) {
4753 if (erp_idx
< nlists
)
4754 erp
= &ifp
->if_u1
.if_ext_irec
[erp_idx
];
4758 ep
= &erp
->er_extbuf
[erp
->er_extcount
];
4760 ep_next
= erp_next
->er_extbuf
;
4765 * This is called to update the er_extoff field in the indirection
4766 * array when extents have been added or removed from one of the
4767 * extent lists. erp_idx contains the irec index to begin updating
4768 * at and ext_diff contains the number of extents that were added
4772 xfs_iext_irec_update_extoffs(
4773 xfs_ifork_t
*ifp
, /* inode fork pointer */
4774 int erp_idx
, /* irec index to update */
4775 int ext_diff
) /* number of new extents */
4777 int i
; /* loop counter */
4778 int nlists
; /* number of irec's (ex lists */
4780 ASSERT(ifp
->if_flags
& XFS_IFEXTIREC
);
4781 nlists
= ifp
->if_real_bytes
/ XFS_IEXT_BUFSZ
;
4782 for (i
= erp_idx
; i
< nlists
; i
++) {
4783 ifp
->if_u1
.if_ext_irec
[i
].er_extoff
+= ext_diff
;