2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
25 #include "xfs_trans.h"
26 #include "xfs_trans_priv.h"
31 #include "xfs_dmapi.h"
32 #include "xfs_mount.h"
33 #include "xfs_bmap_btree.h"
34 #include "xfs_alloc_btree.h"
35 #include "xfs_ialloc_btree.h"
36 #include "xfs_dir_sf.h"
37 #include "xfs_dir2_sf.h"
38 #include "xfs_attr_sf.h"
39 #include "xfs_dinode.h"
40 #include "xfs_inode.h"
41 #include "xfs_buf_item.h"
42 #include "xfs_inode_item.h"
43 #include "xfs_btree.h"
44 #include "xfs_alloc.h"
45 #include "xfs_ialloc.h"
48 #include "xfs_error.h"
49 #include "xfs_utils.h"
50 #include "xfs_dir2_trace.h"
51 #include "xfs_quota.h"
56 kmem_zone_t
*xfs_ifork_zone
;
57 kmem_zone_t
*xfs_inode_zone
;
58 kmem_zone_t
*xfs_chashlist_zone
;
61 * Used in xfs_itruncate(). This is the maximum number of extents
62 * freed from a file in a single transaction.
64 #define XFS_ITRUNC_MAX_EXTENTS 2
66 STATIC
int xfs_iflush_int(xfs_inode_t
*, xfs_buf_t
*);
67 STATIC
int xfs_iformat_local(xfs_inode_t
*, xfs_dinode_t
*, int, int);
68 STATIC
int xfs_iformat_extents(xfs_inode_t
*, xfs_dinode_t
*, int);
69 STATIC
int xfs_iformat_btree(xfs_inode_t
*, xfs_dinode_t
*, int);
74 * Make sure that the extents in the given memory buffer
88 for (i
= 0; i
< nrecs
; i
++) {
89 rec
.l0
= get_unaligned((__uint64_t
*)&ep
->l0
);
90 rec
.l1
= get_unaligned((__uint64_t
*)&ep
->l1
);
92 xfs_bmbt_disk_get_all(&rec
, &irec
);
94 xfs_bmbt_get_all(&rec
, &irec
);
95 if (fmt
== XFS_EXTFMT_NOSTATE
)
96 ASSERT(irec
.br_state
== XFS_EXT_NORM
);
101 #define xfs_validate_extents(ep, nrecs, disk, fmt)
105 * Check that none of the inode's in the buffer have a next
106 * unlinked field of 0.
118 j
= mp
->m_inode_cluster_size
>> mp
->m_sb
.sb_inodelog
;
120 for (i
= 0; i
< j
; i
++) {
121 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
,
122 i
* mp
->m_sb
.sb_inodesize
);
123 if (!dip
->di_next_unlinked
) {
124 xfs_fs_cmn_err(CE_ALERT
, mp
,
125 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
127 ASSERT(dip
->di_next_unlinked
);
134 * This routine is called to map an inode number within a file
135 * system to the buffer containing the on-disk version of the
136 * inode. It returns a pointer to the buffer containing the
137 * on-disk inode in the bpp parameter, and in the dip parameter
138 * it returns a pointer to the on-disk inode within that buffer.
140 * If a non-zero error is returned, then the contents of bpp and
141 * dipp are undefined.
143 * Use xfs_imap() to determine the size and location of the
144 * buffer to read from disk.
162 * Call the space managment code to find the location of the
166 error
= xfs_imap(mp
, tp
, ino
, &imap
, XFS_IMAP_LOOKUP
);
169 "xfs_inotobp: xfs_imap() returned an "
170 "error %d on %s. Returning error.", error
, mp
->m_fsname
);
175 * If the inode number maps to a block outside the bounds of the
176 * file system then return NULL rather than calling read_buf
177 * and panicing when we get an error from the driver.
179 if ((imap
.im_blkno
+ imap
.im_len
) >
180 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
)) {
182 "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds "
183 "of the file system %s. Returning EINVAL.",
184 (unsigned long long)imap
.im_blkno
,
185 imap
.im_len
, mp
->m_fsname
);
186 return XFS_ERROR(EINVAL
);
190 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
191 * default to just a read_buf() call.
193 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, imap
.im_blkno
,
194 (int)imap
.im_len
, XFS_BUF_LOCK
, &bp
);
198 "xfs_inotobp: xfs_trans_read_buf() returned an "
199 "error %d on %s. Returning error.", error
, mp
->m_fsname
);
202 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, 0);
204 INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
) == XFS_DINODE_MAGIC
&&
205 XFS_DINODE_GOOD_VERSION(INT_GET(dip
->di_core
.di_version
, ARCH_CONVERT
));
206 if (unlikely(XFS_TEST_ERROR(!di_ok
, mp
, XFS_ERRTAG_ITOBP_INOTOBP
,
207 XFS_RANDOM_ITOBP_INOTOBP
))) {
208 XFS_CORRUPTION_ERROR("xfs_inotobp", XFS_ERRLEVEL_LOW
, mp
, dip
);
209 xfs_trans_brelse(tp
, bp
);
211 "xfs_inotobp: XFS_TEST_ERROR() returned an "
212 "error on %s. Returning EFSCORRUPTED.", mp
->m_fsname
);
213 return XFS_ERROR(EFSCORRUPTED
);
216 xfs_inobp_check(mp
, bp
);
219 * Set *dipp to point to the on-disk inode in the buffer.
221 *dipp
= (xfs_dinode_t
*)xfs_buf_offset(bp
, imap
.im_boffset
);
223 *offset
= imap
.im_boffset
;
229 * This routine is called to map an inode to the buffer containing
230 * the on-disk version of the inode. It returns a pointer to the
231 * buffer containing the on-disk inode in the bpp parameter, and in
232 * the dip parameter it returns a pointer to the on-disk inode within
235 * If a non-zero error is returned, then the contents of bpp and
236 * dipp are undefined.
238 * If the inode is new and has not yet been initialized, use xfs_imap()
239 * to determine the size and location of the buffer to read from disk.
240 * If the inode has already been mapped to its buffer and read in once,
241 * then use the mapping information stored in the inode rather than
242 * calling xfs_imap(). This allows us to avoid the overhead of looking
243 * at the inode btree for small block file systems (see xfs_dilocate()).
244 * We can tell whether the inode has been mapped in before by comparing
245 * its disk block address to 0. Only uninitialized inodes will have
246 * 0 for the disk block address.
265 if (ip
->i_blkno
== (xfs_daddr_t
)0) {
267 * Call the space management code to find the location of the
271 error
= xfs_imap(mp
, tp
, ip
->i_ino
, &imap
, XFS_IMAP_LOOKUP
);
277 * If the inode number maps to a block outside the bounds
278 * of the file system then return NULL rather than calling
279 * read_buf and panicing when we get an error from the
282 if ((imap
.im_blkno
+ imap
.im_len
) >
283 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
)) {
285 xfs_fs_cmn_err(CE_ALERT
, mp
, "xfs_itobp: "
286 "(imap.im_blkno (0x%llx) "
287 "+ imap.im_len (0x%llx)) > "
288 " XFS_FSB_TO_BB(mp, "
289 "mp->m_sb.sb_dblocks) (0x%llx)",
290 (unsigned long long) imap
.im_blkno
,
291 (unsigned long long) imap
.im_len
,
292 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
));
294 return XFS_ERROR(EINVAL
);
298 * Fill in the fields in the inode that will be used to
299 * map the inode to its buffer from now on.
301 ip
->i_blkno
= imap
.im_blkno
;
302 ip
->i_len
= imap
.im_len
;
303 ip
->i_boffset
= imap
.im_boffset
;
306 * We've already mapped the inode once, so just use the
307 * mapping that we saved the first time.
309 imap
.im_blkno
= ip
->i_blkno
;
310 imap
.im_len
= ip
->i_len
;
311 imap
.im_boffset
= ip
->i_boffset
;
313 ASSERT(bno
== 0 || bno
== imap
.im_blkno
);
316 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
317 * default to just a read_buf() call.
319 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, imap
.im_blkno
,
320 (int)imap
.im_len
, XFS_BUF_LOCK
, &bp
);
324 xfs_fs_cmn_err(CE_ALERT
, mp
, "xfs_itobp: "
325 "xfs_trans_read_buf() returned error %d, "
326 "imap.im_blkno 0x%llx, imap.im_len 0x%llx",
327 error
, (unsigned long long) imap
.im_blkno
,
328 (unsigned long long) imap
.im_len
);
334 * Validate the magic number and version of every inode in the buffer
335 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
338 ni
= BBTOB(imap
.im_len
) >> mp
->m_sb
.sb_inodelog
;
342 for (i
= 0; i
< ni
; i
++) {
346 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
,
347 (i
<< mp
->m_sb
.sb_inodelog
));
348 di_ok
= INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
) == XFS_DINODE_MAGIC
&&
349 XFS_DINODE_GOOD_VERSION(INT_GET(dip
->di_core
.di_version
, ARCH_CONVERT
));
350 if (unlikely(XFS_TEST_ERROR(!di_ok
, mp
, XFS_ERRTAG_ITOBP_INOTOBP
,
351 XFS_RANDOM_ITOBP_INOTOBP
))) {
353 prdev("bad inode magic/vsn daddr %lld #%d (magic=%x)",
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
);
364 #endif /* __KERNEL__ */
366 xfs_inobp_check(mp
, bp
);
369 * Mark the buffer as an inode buffer now that it looks good
371 XFS_BUF_SET_VTYPE(bp
, B_FS_INO
);
374 * Set *dipp to point to the on-disk inode in the buffer.
376 *dipp
= (xfs_dinode_t
*)xfs_buf_offset(bp
, imap
.im_boffset
);
382 * Move inode type and inode format specific information from the
383 * on-disk inode to the in-core inode. For fifos, devs, and sockets
384 * this means set if_rdev to the proper value. For files, directories,
385 * and symlinks this means to bring in the in-line data or extent
386 * pointers. For a file in B-tree format, only the root is immediately
387 * brought in-core. The rest will be in-lined in if_extents when it
388 * is first referenced (see xfs_iread_extents()).
395 xfs_attr_shortform_t
*atp
;
399 ip
->i_df
.if_ext_max
=
400 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
404 INT_GET(dip
->di_core
.di_nextents
, ARCH_CONVERT
) +
405 INT_GET(dip
->di_core
.di_anextents
, ARCH_CONVERT
) >
406 INT_GET(dip
->di_core
.di_nblocks
, ARCH_CONVERT
))) {
407 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
408 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
409 (unsigned long long)ip
->i_ino
,
410 (int)(INT_GET(dip
->di_core
.di_nextents
, ARCH_CONVERT
)
411 + INT_GET(dip
->di_core
.di_anextents
, ARCH_CONVERT
)),
413 INT_GET(dip
->di_core
.di_nblocks
, ARCH_CONVERT
));
414 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW
,
416 return XFS_ERROR(EFSCORRUPTED
);
419 if (unlikely(INT_GET(dip
->di_core
.di_forkoff
, ARCH_CONVERT
) > ip
->i_mount
->m_sb
.sb_inodesize
)) {
420 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
421 "corrupt dinode %Lu, forkoff = 0x%x.",
422 (unsigned long long)ip
->i_ino
,
423 (int)(INT_GET(dip
->di_core
.di_forkoff
, ARCH_CONVERT
)));
424 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW
,
426 return XFS_ERROR(EFSCORRUPTED
);
429 switch (ip
->i_d
.di_mode
& S_IFMT
) {
434 if (unlikely(INT_GET(dip
->di_core
.di_format
, ARCH_CONVERT
) != XFS_DINODE_FMT_DEV
)) {
435 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW
,
437 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
= (int)INT_GET(atp
->hdr
.totsize
, ARCH_CONVERT
);
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.
597 xfs_bmbt_rec_t
*ep
, *dp
;
604 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
605 nex
= XFS_DFORK_NEXTENTS(dip
, whichfork
);
606 size
= nex
* (uint
)sizeof(xfs_bmbt_rec_t
);
609 * If the number of extents is unreasonable, then something
610 * is wrong and we just bail out rather than crash in
611 * kmem_alloc() or memcpy() below.
613 if (unlikely(size
< 0 || size
> XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
))) {
614 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
615 "corrupt inode %Lu ((a)extents = %d).",
616 (unsigned long long) ip
->i_ino
, nex
);
617 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW
,
619 return XFS_ERROR(EFSCORRUPTED
);
624 ifp
->if_u1
.if_extents
= NULL
;
625 else if (nex
<= XFS_INLINE_EXTS
)
626 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
628 ifp
->if_u1
.if_extents
= kmem_alloc(size
, KM_SLEEP
);
629 ASSERT(ifp
->if_u1
.if_extents
!= NULL
);
632 ifp
->if_bytes
= size
;
633 ifp
->if_real_bytes
= real_size
;
635 dp
= (xfs_bmbt_rec_t
*) XFS_DFORK_PTR(dip
, whichfork
);
636 xfs_validate_extents(dp
, nex
, 1, XFS_EXTFMT_INODE(ip
));
637 ep
= ifp
->if_u1
.if_extents
;
638 for (i
= 0; i
< nex
; i
++, ep
++, dp
++) {
639 ep
->l0
= INT_GET(get_unaligned((__uint64_t
*)&dp
->l0
),
641 ep
->l1
= INT_GET(get_unaligned((__uint64_t
*)&dp
->l1
),
644 xfs_bmap_trace_exlist("xfs_iformat_extents", ip
, nex
,
646 if (whichfork
!= XFS_DATA_FORK
||
647 XFS_EXTFMT_INODE(ip
) == XFS_EXTFMT_NOSTATE
)
648 if (unlikely(xfs_check_nostate_extents(
649 ifp
->if_u1
.if_extents
, nex
))) {
650 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
653 return XFS_ERROR(EFSCORRUPTED
);
656 ifp
->if_flags
|= XFS_IFEXTENTS
;
661 * The file has too many extents to fit into
662 * the inode, so they are in B-tree format.
663 * Allocate a buffer for the root of the B-tree
664 * and copy the root into it. The i_extents
665 * field will remain NULL until all of the
666 * extents are read in (when they are needed).
674 xfs_bmdr_block_t
*dfp
;
680 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
681 dfp
= (xfs_bmdr_block_t
*)XFS_DFORK_PTR(dip
, whichfork
);
682 size
= XFS_BMAP_BROOT_SPACE(dfp
);
683 nrecs
= XFS_BMAP_BROOT_NUMRECS(dfp
);
686 * blow out if -- fork has less extents than can fit in
687 * fork (fork shouldn't be a btree format), root btree
688 * block has more records than can fit into the fork,
689 * or the number of extents is greater than the number of
692 if (unlikely(XFS_IFORK_NEXTENTS(ip
, whichfork
) <= ifp
->if_ext_max
693 || XFS_BMDR_SPACE_CALC(nrecs
) >
694 XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
)
695 || XFS_IFORK_NEXTENTS(ip
, whichfork
) > ip
->i_d
.di_nblocks
)) {
696 xfs_fs_repair_cmn_err(CE_WARN
, ip
->i_mount
,
697 "corrupt inode %Lu (btree).",
698 (unsigned long long) ip
->i_ino
);
699 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW
,
701 return XFS_ERROR(EFSCORRUPTED
);
704 ifp
->if_broot_bytes
= size
;
705 ifp
->if_broot
= kmem_alloc(size
, KM_SLEEP
);
706 ASSERT(ifp
->if_broot
!= NULL
);
708 * Copy and convert from the on-disk structure
709 * to the in-memory structure.
711 xfs_bmdr_to_bmbt(dfp
, XFS_DFORK_SIZE(dip
, ip
->i_mount
, whichfork
),
712 ifp
->if_broot
, size
);
713 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
714 ifp
->if_flags
|= XFS_IFBROOT
;
720 * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk
723 * buf = on-disk representation
724 * dip = native representation
725 * dir = direction - +ve -> disk to native
726 * -ve -> native to disk
729 xfs_xlate_dinode_core(
731 xfs_dinode_core_t
*dip
,
734 xfs_dinode_core_t
*buf_core
= (xfs_dinode_core_t
*)buf
;
735 xfs_dinode_core_t
*mem_core
= (xfs_dinode_core_t
*)dip
;
736 xfs_arch_t arch
= ARCH_CONVERT
;
740 INT_XLATE(buf_core
->di_magic
, mem_core
->di_magic
, dir
, arch
);
741 INT_XLATE(buf_core
->di_mode
, mem_core
->di_mode
, dir
, arch
);
742 INT_XLATE(buf_core
->di_version
, mem_core
->di_version
, dir
, arch
);
743 INT_XLATE(buf_core
->di_format
, mem_core
->di_format
, dir
, arch
);
744 INT_XLATE(buf_core
->di_onlink
, mem_core
->di_onlink
, dir
, arch
);
745 INT_XLATE(buf_core
->di_uid
, mem_core
->di_uid
, dir
, arch
);
746 INT_XLATE(buf_core
->di_gid
, mem_core
->di_gid
, dir
, arch
);
747 INT_XLATE(buf_core
->di_nlink
, mem_core
->di_nlink
, dir
, arch
);
748 INT_XLATE(buf_core
->di_projid
, mem_core
->di_projid
, dir
, arch
);
751 memcpy(mem_core
->di_pad
, buf_core
->di_pad
,
752 sizeof(buf_core
->di_pad
));
754 memcpy(buf_core
->di_pad
, mem_core
->di_pad
,
755 sizeof(buf_core
->di_pad
));
758 INT_XLATE(buf_core
->di_flushiter
, mem_core
->di_flushiter
, dir
, arch
);
760 INT_XLATE(buf_core
->di_atime
.t_sec
, mem_core
->di_atime
.t_sec
,
762 INT_XLATE(buf_core
->di_atime
.t_nsec
, mem_core
->di_atime
.t_nsec
,
764 INT_XLATE(buf_core
->di_mtime
.t_sec
, mem_core
->di_mtime
.t_sec
,
766 INT_XLATE(buf_core
->di_mtime
.t_nsec
, mem_core
->di_mtime
.t_nsec
,
768 INT_XLATE(buf_core
->di_ctime
.t_sec
, mem_core
->di_ctime
.t_sec
,
770 INT_XLATE(buf_core
->di_ctime
.t_nsec
, mem_core
->di_ctime
.t_nsec
,
772 INT_XLATE(buf_core
->di_size
, mem_core
->di_size
, dir
, arch
);
773 INT_XLATE(buf_core
->di_nblocks
, mem_core
->di_nblocks
, dir
, arch
);
774 INT_XLATE(buf_core
->di_extsize
, mem_core
->di_extsize
, dir
, arch
);
775 INT_XLATE(buf_core
->di_nextents
, mem_core
->di_nextents
, dir
, arch
);
776 INT_XLATE(buf_core
->di_anextents
, mem_core
->di_anextents
, dir
, arch
);
777 INT_XLATE(buf_core
->di_forkoff
, mem_core
->di_forkoff
, dir
, arch
);
778 INT_XLATE(buf_core
->di_aformat
, mem_core
->di_aformat
, dir
, arch
);
779 INT_XLATE(buf_core
->di_dmevmask
, mem_core
->di_dmevmask
, dir
, arch
);
780 INT_XLATE(buf_core
->di_dmstate
, mem_core
->di_dmstate
, dir
, arch
);
781 INT_XLATE(buf_core
->di_flags
, mem_core
->di_flags
, dir
, arch
);
782 INT_XLATE(buf_core
->di_gen
, mem_core
->di_gen
, dir
, arch
);
787 xfs_dinode_core_t
*dic
,
792 if (di_flags
& XFS_DIFLAG_ANY
) {
793 if (di_flags
& XFS_DIFLAG_REALTIME
)
794 flags
|= XFS_XFLAG_REALTIME
;
795 if (di_flags
& XFS_DIFLAG_PREALLOC
)
796 flags
|= XFS_XFLAG_PREALLOC
;
797 if (di_flags
& XFS_DIFLAG_IMMUTABLE
)
798 flags
|= XFS_XFLAG_IMMUTABLE
;
799 if (di_flags
& XFS_DIFLAG_APPEND
)
800 flags
|= XFS_XFLAG_APPEND
;
801 if (di_flags
& XFS_DIFLAG_SYNC
)
802 flags
|= XFS_XFLAG_SYNC
;
803 if (di_flags
& XFS_DIFLAG_NOATIME
)
804 flags
|= XFS_XFLAG_NOATIME
;
805 if (di_flags
& XFS_DIFLAG_NODUMP
)
806 flags
|= XFS_XFLAG_NODUMP
;
807 if (di_flags
& XFS_DIFLAG_RTINHERIT
)
808 flags
|= XFS_XFLAG_RTINHERIT
;
809 if (di_flags
& XFS_DIFLAG_PROJINHERIT
)
810 flags
|= XFS_XFLAG_PROJINHERIT
;
811 if (di_flags
& XFS_DIFLAG_NOSYMLINKS
)
812 flags
|= XFS_XFLAG_NOSYMLINKS
;
813 if (di_flags
& XFS_DIFLAG_EXTSIZE
)
814 flags
|= XFS_XFLAG_EXTSIZE
;
815 if (di_flags
& XFS_DIFLAG_EXTSZINHERIT
)
816 flags
|= XFS_XFLAG_EXTSZINHERIT
;
826 xfs_dinode_core_t
*dic
= &ip
->i_d
;
828 return _xfs_dic2xflags(dic
, dic
->di_flags
) |
829 (XFS_CFORK_Q(dic
) ? XFS_XFLAG_HASATTR
: 0);
834 xfs_dinode_core_t
*dic
)
836 return _xfs_dic2xflags(dic
, INT_GET(dic
->di_flags
, ARCH_CONVERT
)) |
837 (XFS_CFORK_Q_DISK(dic
) ? XFS_XFLAG_HASATTR
: 0);
841 * Given a mount structure and an inode number, return a pointer
842 * to a newly allocated in-core inode coresponding to the given
845 * Initialize the inode's attributes and extent pointers if it
846 * already has them (it will not if the inode has no links).
861 ASSERT(xfs_inode_zone
!= NULL
);
863 ip
= kmem_zone_zalloc(xfs_inode_zone
, KM_SLEEP
);
868 * Get pointer's to the on-disk inode and the buffer containing it.
869 * If the inode number refers to a block outside the file system
870 * then xfs_itobp() will return NULL. In this case we should
871 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
872 * know that this is a new incore inode.
874 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &bp
, bno
);
877 kmem_zone_free(xfs_inode_zone
, ip
);
882 * Initialize inode's trace buffers.
883 * Do this before xfs_iformat in case it adds entries.
885 #ifdef XFS_BMAP_TRACE
886 ip
->i_xtrace
= ktrace_alloc(XFS_BMAP_KTRACE_SIZE
, KM_SLEEP
);
888 #ifdef XFS_BMBT_TRACE
889 ip
->i_btrace
= ktrace_alloc(XFS_BMBT_KTRACE_SIZE
, KM_SLEEP
);
892 ip
->i_rwtrace
= ktrace_alloc(XFS_RW_KTRACE_SIZE
, KM_SLEEP
);
894 #ifdef XFS_ILOCK_TRACE
895 ip
->i_lock_trace
= ktrace_alloc(XFS_ILOCK_KTRACE_SIZE
, KM_SLEEP
);
897 #ifdef XFS_DIR2_TRACE
898 ip
->i_dir_trace
= ktrace_alloc(XFS_DIR2_KTRACE_SIZE
, KM_SLEEP
);
902 * If we got something that isn't an inode it means someone
903 * (nfs or dmi) has a stale handle.
905 if (INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
) != XFS_DINODE_MAGIC
) {
906 kmem_zone_free(xfs_inode_zone
, ip
);
907 xfs_trans_brelse(tp
, bp
);
909 xfs_fs_cmn_err(CE_ALERT
, mp
, "xfs_iread: "
910 "dip->di_core.di_magic (0x%x) != "
911 "XFS_DINODE_MAGIC (0x%x)",
912 INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
),
915 return XFS_ERROR(EINVAL
);
919 * If the on-disk inode is already linked to a directory
920 * entry, copy all of the inode into the in-core inode.
921 * xfs_iformat() handles copying in the inode format
922 * specific information.
923 * Otherwise, just get the truly permanent information.
925 if (dip
->di_core
.di_mode
) {
926 xfs_xlate_dinode_core((xfs_caddr_t
)&dip
->di_core
,
928 error
= xfs_iformat(ip
, dip
);
930 kmem_zone_free(xfs_inode_zone
, ip
);
931 xfs_trans_brelse(tp
, bp
);
933 xfs_fs_cmn_err(CE_ALERT
, mp
, "xfs_iread: "
934 "xfs_iformat() returned error %d",
940 ip
->i_d
.di_magic
= INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
);
941 ip
->i_d
.di_version
= INT_GET(dip
->di_core
.di_version
, ARCH_CONVERT
);
942 ip
->i_d
.di_gen
= INT_GET(dip
->di_core
.di_gen
, ARCH_CONVERT
);
943 ip
->i_d
.di_flushiter
= INT_GET(dip
->di_core
.di_flushiter
, ARCH_CONVERT
);
945 * Make sure to pull in the mode here as well in
946 * case the inode is released without being used.
947 * This ensures that xfs_inactive() will see that
948 * the inode is already free and not try to mess
949 * with the uninitialized part of it.
953 * Initialize the per-fork minima and maxima for a new
954 * inode here. xfs_iformat will do it for old inodes.
956 ip
->i_df
.if_ext_max
=
957 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
960 INIT_LIST_HEAD(&ip
->i_reclaim
);
963 * The inode format changed when we moved the link count and
964 * made it 32 bits long. If this is an old format inode,
965 * convert it in memory to look like a new one. If it gets
966 * flushed to disk we will convert back before flushing or
967 * logging it. We zero out the new projid field and the old link
968 * count field. We'll handle clearing the pad field (the remains
969 * of the old uuid field) when we actually convert the inode to
970 * the new format. We don't change the version number so that we
971 * can distinguish this from a real new format inode.
973 if (ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
) {
974 ip
->i_d
.di_nlink
= ip
->i_d
.di_onlink
;
975 ip
->i_d
.di_onlink
= 0;
976 ip
->i_d
.di_projid
= 0;
979 ip
->i_delayed_blks
= 0;
982 * Mark the buffer containing the inode as something to keep
983 * around for a while. This helps to keep recently accessed
984 * meta-data in-core longer.
986 XFS_BUF_SET_REF(bp
, XFS_INO_REF
);
989 * Use xfs_trans_brelse() to release the buffer containing the
990 * on-disk inode, because it was acquired with xfs_trans_read_buf()
991 * in xfs_itobp() above. If tp is NULL, this is just a normal
992 * brelse(). If we're within a transaction, then xfs_trans_brelse()
993 * will only release the buffer if it is not dirty within the
994 * transaction. It will be OK to release the buffer in this case,
995 * because inodes on disk are never destroyed and we will be
996 * locking the new in-core inode before putting it in the hash
997 * table where other processes can find it. Thus we don't have
998 * to worry about the inode being changed just because we released
1001 xfs_trans_brelse(tp
, bp
);
1007 * Read in extents from a btree-format inode.
1008 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1020 if (unlikely(XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
)) {
1021 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW
,
1023 return XFS_ERROR(EFSCORRUPTED
);
1025 size
= XFS_IFORK_NEXTENTS(ip
, whichfork
) * (uint
)sizeof(xfs_bmbt_rec_t
);
1026 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1028 * We know that the size is valid (it's checked in iformat_btree)
1030 ifp
->if_u1
.if_extents
= kmem_alloc(size
, KM_SLEEP
);
1031 ASSERT(ifp
->if_u1
.if_extents
!= NULL
);
1032 ifp
->if_lastex
= NULLEXTNUM
;
1033 ifp
->if_bytes
= ifp
->if_real_bytes
= (int)size
;
1034 ifp
->if_flags
|= XFS_IFEXTENTS
;
1035 error
= xfs_bmap_read_extents(tp
, ip
, whichfork
);
1037 kmem_free(ifp
->if_u1
.if_extents
, size
);
1038 ifp
->if_u1
.if_extents
= NULL
;
1039 ifp
->if_bytes
= ifp
->if_real_bytes
= 0;
1040 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
1043 xfs_validate_extents((xfs_bmbt_rec_t
*)ifp
->if_u1
.if_extents
,
1044 XFS_IFORK_NEXTENTS(ip
, whichfork
), 0, XFS_EXTFMT_INODE(ip
));
1049 * Allocate an inode on disk and return a copy of its in-core version.
1050 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1051 * appropriately within the inode. The uid and gid for the inode are
1052 * set according to the contents of the given cred structure.
1054 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1055 * has a free inode available, call xfs_iget()
1056 * to obtain the in-core version of the allocated inode. Finally,
1057 * fill in the inode and log its initial contents. In this case,
1058 * ialloc_context would be set to NULL and call_again set to false.
1060 * If xfs_dialloc() does not have an available inode,
1061 * it will replenish its supply by doing an allocation. Since we can
1062 * only do one allocation within a transaction without deadlocks, we
1063 * must commit the current transaction before returning the inode itself.
1064 * In this case, therefore, we will set call_again to true and return.
1065 * The caller should then commit the current transaction, start a new
1066 * transaction, and call xfs_ialloc() again to actually get the inode.
1068 * To ensure that some other process does not grab the inode that
1069 * was allocated during the first call to xfs_ialloc(), this routine
1070 * also returns the [locked] bp pointing to the head of the freelist
1071 * as ialloc_context. The caller should hold this buffer across
1072 * the commit and pass it back into this routine on the second call.
1084 xfs_buf_t
**ialloc_context
,
1085 boolean_t
*call_again
,
1095 * Call the space management code to pick
1096 * the on-disk inode to be allocated.
1098 error
= xfs_dialloc(tp
, pip
->i_ino
, mode
, okalloc
,
1099 ialloc_context
, call_again
, &ino
);
1103 if (*call_again
|| ino
== NULLFSINO
) {
1107 ASSERT(*ialloc_context
== NULL
);
1110 * Get the in-core inode with the lock held exclusively.
1111 * This is because we're setting fields here we need
1112 * to prevent others from looking at until we're done.
1114 error
= xfs_trans_iget(tp
->t_mountp
, tp
, ino
,
1115 IGET_CREATE
, XFS_ILOCK_EXCL
, &ip
);
1122 ip
->i_d
.di_mode
= (__uint16_t
)mode
;
1123 ip
->i_d
.di_onlink
= 0;
1124 ip
->i_d
.di_nlink
= nlink
;
1125 ASSERT(ip
->i_d
.di_nlink
== nlink
);
1126 ip
->i_d
.di_uid
= current_fsuid(cr
);
1127 ip
->i_d
.di_gid
= current_fsgid(cr
);
1128 ip
->i_d
.di_projid
= prid
;
1129 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
1132 * If the superblock version is up to where we support new format
1133 * inodes and this is currently an old format inode, then change
1134 * the inode version number now. This way we only do the conversion
1135 * here rather than here and in the flush/logging code.
1137 if (XFS_SB_VERSION_HASNLINK(&tp
->t_mountp
->m_sb
) &&
1138 ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
) {
1139 ip
->i_d
.di_version
= XFS_DINODE_VERSION_2
;
1141 * We've already zeroed the old link count, the projid field,
1142 * and the pad field.
1147 * Project ids won't be stored on disk if we are using a version 1 inode.
1149 if ( (prid
!= 0) && (ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
))
1150 xfs_bump_ino_vers2(tp
, ip
);
1152 if (XFS_INHERIT_GID(pip
, vp
->v_vfsp
)) {
1153 ip
->i_d
.di_gid
= pip
->i_d
.di_gid
;
1154 if ((pip
->i_d
.di_mode
& S_ISGID
) && (mode
& S_IFMT
) == S_IFDIR
) {
1155 ip
->i_d
.di_mode
|= S_ISGID
;
1160 * If the group ID of the new file does not match the effective group
1161 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1162 * (and only if the irix_sgid_inherit compatibility variable is set).
1164 if ((irix_sgid_inherit
) &&
1165 (ip
->i_d
.di_mode
& S_ISGID
) &&
1166 (!in_group_p((gid_t
)ip
->i_d
.di_gid
))) {
1167 ip
->i_d
.di_mode
&= ~S_ISGID
;
1170 ip
->i_d
.di_size
= 0;
1171 ip
->i_d
.di_nextents
= 0;
1172 ASSERT(ip
->i_d
.di_nblocks
== 0);
1173 xfs_ichgtime(ip
, XFS_ICHGTIME_CHG
|XFS_ICHGTIME_ACC
|XFS_ICHGTIME_MOD
);
1175 * di_gen will have been taken care of in xfs_iread.
1177 ip
->i_d
.di_extsize
= 0;
1178 ip
->i_d
.di_dmevmask
= 0;
1179 ip
->i_d
.di_dmstate
= 0;
1180 ip
->i_d
.di_flags
= 0;
1181 flags
= XFS_ILOG_CORE
;
1182 switch (mode
& S_IFMT
) {
1187 ip
->i_d
.di_format
= XFS_DINODE_FMT_DEV
;
1188 ip
->i_df
.if_u2
.if_rdev
= rdev
;
1189 ip
->i_df
.if_flags
= 0;
1190 flags
|= XFS_ILOG_DEV
;
1194 if (unlikely(pip
->i_d
.di_flags
& XFS_DIFLAG_ANY
)) {
1197 if ((mode
& S_IFMT
) == S_IFDIR
) {
1198 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
)
1199 di_flags
|= XFS_DIFLAG_RTINHERIT
;
1200 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
1201 di_flags
|= XFS_DIFLAG_EXTSZINHERIT
;
1202 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
1204 } else if ((mode
& S_IFMT
) == S_IFREG
) {
1205 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
) {
1206 di_flags
|= XFS_DIFLAG_REALTIME
;
1207 ip
->i_iocore
.io_flags
|= XFS_IOCORE_RT
;
1209 if (pip
->i_d
.di_flags
& XFS_DIFLAG_EXTSZINHERIT
) {
1210 di_flags
|= XFS_DIFLAG_EXTSIZE
;
1211 ip
->i_d
.di_extsize
= pip
->i_d
.di_extsize
;
1214 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOATIME
) &&
1215 xfs_inherit_noatime
)
1216 di_flags
|= XFS_DIFLAG_NOATIME
;
1217 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODUMP
) &&
1219 di_flags
|= XFS_DIFLAG_NODUMP
;
1220 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_SYNC
) &&
1222 di_flags
|= XFS_DIFLAG_SYNC
;
1223 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOSYMLINKS
) &&
1224 xfs_inherit_nosymlinks
)
1225 di_flags
|= XFS_DIFLAG_NOSYMLINKS
;
1226 if (pip
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
1227 di_flags
|= XFS_DIFLAG_PROJINHERIT
;
1228 ip
->i_d
.di_flags
|= di_flags
;
1232 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
1233 ip
->i_df
.if_flags
= XFS_IFEXTENTS
;
1234 ip
->i_df
.if_bytes
= ip
->i_df
.if_real_bytes
= 0;
1235 ip
->i_df
.if_u1
.if_extents
= NULL
;
1241 * Attribute fork settings for new inode.
1243 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
1244 ip
->i_d
.di_anextents
= 0;
1247 * Log the new values stuffed into the inode.
1249 xfs_trans_log_inode(tp
, ip
, flags
);
1251 /* now that we have an i_mode we can set Linux inode ops (& unlock) */
1252 VFS_INIT_VNODE(XFS_MTOVFS(tp
->t_mountp
), vp
, XFS_ITOBHV(ip
), 1);
1259 * Check to make sure that there are no blocks allocated to the
1260 * file beyond the size of the file. We don't check this for
1261 * files with fixed size extents or real time extents, but we
1262 * at least do it for regular files.
1271 xfs_fileoff_t map_first
;
1273 xfs_bmbt_irec_t imaps
[2];
1275 if ((ip
->i_d
.di_mode
& S_IFMT
) != S_IFREG
)
1278 if (ip
->i_d
.di_flags
& (XFS_DIFLAG_REALTIME
| XFS_DIFLAG_EXTSIZE
))
1282 map_first
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)isize
);
1284 * The filesystem could be shutting down, so bmapi may return
1287 if (xfs_bmapi(NULL
, ip
, map_first
,
1289 (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
)) -
1291 XFS_BMAPI_ENTIRE
, NULL
, 0, imaps
, &nimaps
,
1294 ASSERT(nimaps
== 1);
1295 ASSERT(imaps
[0].br_startblock
== HOLESTARTBLOCK
);
1300 * Calculate the last possible buffered byte in a file. This must
1301 * include data that was buffered beyond the EOF by the write code.
1302 * This also needs to deal with overflowing the xfs_fsize_t type
1303 * which can happen for sizes near the limit.
1305 * We also need to take into account any blocks beyond the EOF. It
1306 * may be the case that they were buffered by a write which failed.
1307 * In that case the pages will still be in memory, but the inode size
1308 * will never have been updated.
1315 xfs_fsize_t last_byte
;
1316 xfs_fileoff_t last_block
;
1317 xfs_fileoff_t size_last_block
;
1320 ASSERT(ismrlocked(&(ip
->i_iolock
), MR_UPDATE
| MR_ACCESS
));
1324 * Only check for blocks beyond the EOF if the extents have
1325 * been read in. This eliminates the need for the inode lock,
1326 * and it also saves us from looking when it really isn't
1329 if (ip
->i_df
.if_flags
& XFS_IFEXTENTS
) {
1330 error
= xfs_bmap_last_offset(NULL
, ip
, &last_block
,
1338 size_last_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)ip
->i_d
.di_size
);
1339 last_block
= XFS_FILEOFF_MAX(last_block
, size_last_block
);
1341 last_byte
= XFS_FSB_TO_B(mp
, last_block
);
1342 if (last_byte
< 0) {
1343 return XFS_MAXIOFFSET(mp
);
1345 last_byte
+= (1 << mp
->m_writeio_log
);
1346 if (last_byte
< 0) {
1347 return XFS_MAXIOFFSET(mp
);
1352 #if defined(XFS_RW_TRACE)
1358 xfs_fsize_t new_size
,
1359 xfs_off_t toss_start
,
1360 xfs_off_t toss_finish
)
1362 if (ip
->i_rwtrace
== NULL
) {
1366 ktrace_enter(ip
->i_rwtrace
,
1369 (void*)(unsigned long)((ip
->i_d
.di_size
>> 32) & 0xffffffff),
1370 (void*)(unsigned long)(ip
->i_d
.di_size
& 0xffffffff),
1371 (void*)((long)flag
),
1372 (void*)(unsigned long)((new_size
>> 32) & 0xffffffff),
1373 (void*)(unsigned long)(new_size
& 0xffffffff),
1374 (void*)(unsigned long)((toss_start
>> 32) & 0xffffffff),
1375 (void*)(unsigned long)(toss_start
& 0xffffffff),
1376 (void*)(unsigned long)((toss_finish
>> 32) & 0xffffffff),
1377 (void*)(unsigned long)(toss_finish
& 0xffffffff),
1378 (void*)(unsigned long)current_cpu(),
1385 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1389 * Start the truncation of the file to new_size. The new size
1390 * must be smaller than the current size. This routine will
1391 * clear the buffer and page caches of file data in the removed
1392 * range, and xfs_itruncate_finish() will remove the underlying
1395 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1396 * must NOT have the inode lock held at all. This is because we're
1397 * calling into the buffer/page cache code and we can't hold the
1398 * inode lock when we do so.
1400 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1401 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1402 * in the case that the caller is locking things out of order and
1403 * may not be able to call xfs_itruncate_finish() with the inode lock
1404 * held without dropping the I/O lock. If the caller must drop the
1405 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1406 * must be called again with all the same restrictions as the initial
1410 xfs_itruncate_start(
1413 xfs_fsize_t new_size
)
1415 xfs_fsize_t last_byte
;
1416 xfs_off_t toss_start
;
1420 ASSERT(ismrlocked(&ip
->i_iolock
, MR_UPDATE
) != 0);
1421 ASSERT((new_size
== 0) || (new_size
<= ip
->i_d
.di_size
));
1422 ASSERT((flags
== XFS_ITRUNC_DEFINITE
) ||
1423 (flags
== XFS_ITRUNC_MAYBE
));
1428 * Call VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES() to get rid of pages and buffers
1429 * overlapping the region being removed. We have to use
1430 * the less efficient VOP_FLUSHINVAL_PAGES() in the case that the
1431 * caller may not be able to finish the truncate without
1432 * dropping the inode's I/O lock. Make sure
1433 * to catch any pages brought in by buffers overlapping
1434 * the EOF by searching out beyond the isize by our
1435 * block size. We round new_size up to a block boundary
1436 * so that we don't toss things on the same block as
1437 * new_size but before it.
1439 * Before calling VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES(), make sure to
1440 * call remapf() over the same region if the file is mapped.
1441 * This frees up mapped file references to the pages in the
1442 * given range and for the VOP_FLUSHINVAL_PAGES() case it ensures
1443 * that we get the latest mapped changes flushed out.
1445 toss_start
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)new_size
);
1446 toss_start
= XFS_FSB_TO_B(mp
, toss_start
);
1447 if (toss_start
< 0) {
1449 * The place to start tossing is beyond our maximum
1450 * file size, so there is no way that the data extended
1455 last_byte
= xfs_file_last_byte(ip
);
1456 xfs_itrunc_trace(XFS_ITRUNC_START
, ip
, flags
, new_size
, toss_start
,
1458 if (last_byte
> toss_start
) {
1459 if (flags
& XFS_ITRUNC_DEFINITE
) {
1460 VOP_TOSS_PAGES(vp
, toss_start
, -1, FI_REMAPF_LOCKED
);
1462 VOP_FLUSHINVAL_PAGES(vp
, toss_start
, -1, FI_REMAPF_LOCKED
);
1467 if (new_size
== 0) {
1468 ASSERT(VN_CACHED(vp
) == 0);
1474 * Shrink the file to the given new_size. The new
1475 * size must be smaller than the current size.
1476 * This will free up the underlying blocks
1477 * in the removed range after a call to xfs_itruncate_start()
1478 * or xfs_atruncate_start().
1480 * The transaction passed to this routine must have made
1481 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1482 * This routine may commit the given transaction and
1483 * start new ones, so make sure everything involved in
1484 * the transaction is tidy before calling here.
1485 * Some transaction will be returned to the caller to be
1486 * committed. The incoming transaction must already include
1487 * the inode, and both inode locks must be held exclusively.
1488 * The inode must also be "held" within the transaction. On
1489 * return the inode will be "held" within the returned transaction.
1490 * This routine does NOT require any disk space to be reserved
1491 * for it within the transaction.
1493 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1494 * and it indicates the fork which is to be truncated. For the
1495 * attribute fork we only support truncation to size 0.
1497 * We use the sync parameter to indicate whether or not the first
1498 * transaction we perform might have to be synchronous. For the attr fork,
1499 * it needs to be so if the unlink of the inode is not yet known to be
1500 * permanent in the log. This keeps us from freeing and reusing the
1501 * blocks of the attribute fork before the unlink of the inode becomes
1504 * For the data fork, we normally have to run synchronously if we're
1505 * being called out of the inactive path or we're being called
1506 * out of the create path where we're truncating an existing file.
1507 * Either way, the truncate needs to be sync so blocks don't reappear
1508 * in the file with altered data in case of a crash. wsync filesystems
1509 * can run the first case async because anything that shrinks the inode
1510 * has to run sync so by the time we're called here from inactive, the
1511 * inode size is permanently set to 0.
1513 * Calls from the truncate path always need to be sync unless we're
1514 * in a wsync filesystem and the file has already been unlinked.
1516 * The caller is responsible for correctly setting the sync parameter.
1517 * It gets too hard for us to guess here which path we're being called
1518 * out of just based on inode state.
1521 xfs_itruncate_finish(
1524 xfs_fsize_t new_size
,
1528 xfs_fsblock_t first_block
;
1529 xfs_fileoff_t first_unmap_block
;
1530 xfs_fileoff_t last_block
;
1531 xfs_filblks_t unmap_len
=0;
1536 xfs_bmap_free_t free_list
;
1539 ASSERT(ismrlocked(&ip
->i_iolock
, MR_UPDATE
) != 0);
1540 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
) != 0);
1541 ASSERT((new_size
== 0) || (new_size
<= ip
->i_d
.di_size
));
1542 ASSERT(*tp
!= NULL
);
1543 ASSERT((*tp
)->t_flags
& XFS_TRANS_PERM_LOG_RES
);
1544 ASSERT(ip
->i_transp
== *tp
);
1545 ASSERT(ip
->i_itemp
!= NULL
);
1546 ASSERT(ip
->i_itemp
->ili_flags
& XFS_ILI_HOLD
);
1550 mp
= (ntp
)->t_mountp
;
1551 ASSERT(! XFS_NOT_DQATTACHED(mp
, ip
));
1554 * We only support truncating the entire attribute fork.
1556 if (fork
== XFS_ATTR_FORK
) {
1559 first_unmap_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)new_size
);
1560 xfs_itrunc_trace(XFS_ITRUNC_FINISH1
, ip
, 0, new_size
, 0, 0);
1562 * The first thing we do is set the size to new_size permanently
1563 * on disk. This way we don't have to worry about anyone ever
1564 * being able to look at the data being freed even in the face
1565 * of a crash. What we're getting around here is the case where
1566 * we free a block, it is allocated to another file, it is written
1567 * to, and then we crash. If the new data gets written to the
1568 * file but the log buffers containing the free and reallocation
1569 * don't, then we'd end up with garbage in the blocks being freed.
1570 * As long as we make the new_size permanent before actually
1571 * freeing any blocks it doesn't matter if they get writtten to.
1573 * The callers must signal into us whether or not the size
1574 * setting here must be synchronous. There are a few cases
1575 * where it doesn't have to be synchronous. Those cases
1576 * occur if the file is unlinked and we know the unlink is
1577 * permanent or if the blocks being truncated are guaranteed
1578 * to be beyond the inode eof (regardless of the link count)
1579 * and the eof value is permanent. Both of these cases occur
1580 * only on wsync-mounted filesystems. In those cases, we're
1581 * guaranteed that no user will ever see the data in the blocks
1582 * that are being truncated so the truncate can run async.
1583 * In the free beyond eof case, the file may wind up with
1584 * more blocks allocated to it than it needs if we crash
1585 * and that won't get fixed until the next time the file
1586 * is re-opened and closed but that's ok as that shouldn't
1587 * be too many blocks.
1589 * However, we can't just make all wsync xactions run async
1590 * because there's one call out of the create path that needs
1591 * to run sync where it's truncating an existing file to size
1592 * 0 whose size is > 0.
1594 * It's probably possible to come up with a test in this
1595 * routine that would correctly distinguish all the above
1596 * cases from the values of the function parameters and the
1597 * inode state but for sanity's sake, I've decided to let the
1598 * layers above just tell us. It's simpler to correctly figure
1599 * out in the layer above exactly under what conditions we
1600 * can run async and I think it's easier for others read and
1601 * follow the logic in case something has to be changed.
1602 * cscope is your friend -- rcc.
1604 * The attribute fork is much simpler.
1606 * For the attribute fork we allow the caller to tell us whether
1607 * the unlink of the inode that led to this call is yet permanent
1608 * in the on disk log. If it is not and we will be freeing extents
1609 * in this inode then we make the first transaction synchronous
1610 * to make sure that the unlink is permanent by the time we free
1613 if (fork
== XFS_DATA_FORK
) {
1614 if (ip
->i_d
.di_nextents
> 0) {
1615 ip
->i_d
.di_size
= new_size
;
1616 xfs_trans_log_inode(ntp
, ip
, XFS_ILOG_CORE
);
1619 ASSERT(!(mp
->m_flags
& XFS_MOUNT_WSYNC
));
1620 if (ip
->i_d
.di_anextents
> 0)
1621 xfs_trans_set_sync(ntp
);
1623 ASSERT(fork
== XFS_DATA_FORK
||
1624 (fork
== XFS_ATTR_FORK
&&
1625 ((sync
&& !(mp
->m_flags
& XFS_MOUNT_WSYNC
)) ||
1626 (sync
== 0 && (mp
->m_flags
& XFS_MOUNT_WSYNC
)))));
1629 * Since it is possible for space to become allocated beyond
1630 * the end of the file (in a crash where the space is allocated
1631 * but the inode size is not yet updated), simply remove any
1632 * blocks which show up between the new EOF and the maximum
1633 * possible file size. If the first block to be removed is
1634 * beyond the maximum file size (ie it is the same as last_block),
1635 * then there is nothing to do.
1637 last_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
));
1638 ASSERT(first_unmap_block
<= last_block
);
1640 if (last_block
== first_unmap_block
) {
1643 unmap_len
= last_block
- first_unmap_block
+ 1;
1647 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1648 * will tell us whether it freed the entire range or
1649 * not. If this is a synchronous mount (wsync),
1650 * then we can tell bunmapi to keep all the
1651 * transactions asynchronous since the unlink
1652 * transaction that made this inode inactive has
1653 * already hit the disk. There's no danger of
1654 * the freed blocks being reused, there being a
1655 * crash, and the reused blocks suddenly reappearing
1656 * in this file with garbage in them once recovery
1659 XFS_BMAP_INIT(&free_list
, &first_block
);
1660 error
= xfs_bunmapi(ntp
, ip
, first_unmap_block
,
1662 XFS_BMAPI_AFLAG(fork
) |
1663 (sync
? 0 : XFS_BMAPI_ASYNC
),
1664 XFS_ITRUNC_MAX_EXTENTS
,
1665 &first_block
, &free_list
, &done
);
1668 * If the bunmapi call encounters an error,
1669 * return to the caller where the transaction
1670 * can be properly aborted. We just need to
1671 * make sure we're not holding any resources
1672 * that we were not when we came in.
1674 xfs_bmap_cancel(&free_list
);
1679 * Duplicate the transaction that has the permanent
1680 * reservation and commit the old transaction.
1682 error
= xfs_bmap_finish(tp
, &free_list
, first_block
,
1687 * If the bmap finish call encounters an error,
1688 * return to the caller where the transaction
1689 * can be properly aborted. We just need to
1690 * make sure we're not holding any resources
1691 * that we were not when we came in.
1693 * Aborting from this point might lose some
1694 * blocks in the file system, but oh well.
1696 xfs_bmap_cancel(&free_list
);
1699 * If the passed in transaction committed
1700 * in xfs_bmap_finish(), then we want to
1701 * add the inode to this one before returning.
1702 * This keeps things simple for the higher
1703 * level code, because it always knows that
1704 * the inode is locked and held in the
1705 * transaction that returns to it whether
1706 * errors occur or not. We don't mark the
1707 * inode dirty so that this transaction can
1708 * be easily aborted if possible.
1710 xfs_trans_ijoin(ntp
, ip
,
1711 XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1712 xfs_trans_ihold(ntp
, ip
);
1719 * The first xact was committed,
1720 * so add the inode to the new one.
1721 * Mark it dirty so it will be logged
1722 * and moved forward in the log as
1723 * part of every commit.
1725 xfs_trans_ijoin(ntp
, ip
,
1726 XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1727 xfs_trans_ihold(ntp
, ip
);
1728 xfs_trans_log_inode(ntp
, ip
, XFS_ILOG_CORE
);
1730 ntp
= xfs_trans_dup(ntp
);
1731 (void) xfs_trans_commit(*tp
, 0, NULL
);
1733 error
= xfs_trans_reserve(ntp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0,
1734 XFS_TRANS_PERM_LOG_RES
,
1735 XFS_ITRUNCATE_LOG_COUNT
);
1737 * Add the inode being truncated to the next chained
1740 xfs_trans_ijoin(ntp
, ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1741 xfs_trans_ihold(ntp
, ip
);
1746 * Only update the size in the case of the data fork, but
1747 * always re-log the inode so that our permanent transaction
1748 * can keep on rolling it forward in the log.
1750 if (fork
== XFS_DATA_FORK
) {
1751 xfs_isize_check(mp
, ip
, new_size
);
1752 ip
->i_d
.di_size
= new_size
;
1754 xfs_trans_log_inode(ntp
, ip
, XFS_ILOG_CORE
);
1755 ASSERT((new_size
!= 0) ||
1756 (fork
== XFS_ATTR_FORK
) ||
1757 (ip
->i_delayed_blks
== 0));
1758 ASSERT((new_size
!= 0) ||
1759 (fork
== XFS_ATTR_FORK
) ||
1760 (ip
->i_d
.di_nextents
== 0));
1761 xfs_itrunc_trace(XFS_ITRUNC_FINISH2
, ip
, 0, new_size
, 0, 0);
1769 * Do the first part of growing a file: zero any data in the last
1770 * block that is beyond the old EOF. We need to do this before
1771 * the inode is joined to the transaction to modify the i_size.
1772 * That way we can drop the inode lock and call into the buffer
1773 * cache to get the buffer mapping the EOF.
1778 xfs_fsize_t new_size
,
1783 ASSERT(ismrlocked(&(ip
->i_lock
), MR_UPDATE
) != 0);
1784 ASSERT(ismrlocked(&(ip
->i_iolock
), MR_UPDATE
) != 0);
1785 ASSERT(new_size
> ip
->i_d
.di_size
);
1788 * Zero any pages that may have been created by
1789 * xfs_write_file() beyond the end of the file
1790 * and any blocks between the old and new file sizes.
1792 error
= xfs_zero_eof(XFS_ITOV(ip
), &ip
->i_iocore
, new_size
,
1793 ip
->i_d
.di_size
, new_size
);
1800 * This routine is called to extend the size of a file.
1801 * The inode must have both the iolock and the ilock locked
1802 * for update and it must be a part of the current transaction.
1803 * The xfs_igrow_start() function must have been called previously.
1804 * If the change_flag is not zero, the inode change timestamp will
1811 xfs_fsize_t new_size
,
1814 ASSERT(ismrlocked(&(ip
->i_lock
), MR_UPDATE
) != 0);
1815 ASSERT(ismrlocked(&(ip
->i_iolock
), MR_UPDATE
) != 0);
1816 ASSERT(ip
->i_transp
== tp
);
1817 ASSERT(new_size
> ip
->i_d
.di_size
);
1820 * Update the file size. Update the inode change timestamp
1821 * if change_flag set.
1823 ip
->i_d
.di_size
= new_size
;
1825 xfs_ichgtime(ip
, XFS_ICHGTIME_CHG
);
1826 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1832 * This is called when the inode's link count goes to 0.
1833 * We place the on-disk inode on a list in the AGI. It
1834 * will be pulled from this list when the inode is freed.
1846 xfs_agnumber_t agno
;
1847 xfs_daddr_t agdaddr
;
1854 ASSERT(ip
->i_d
.di_nlink
== 0);
1855 ASSERT(ip
->i_d
.di_mode
!= 0);
1856 ASSERT(ip
->i_transp
== tp
);
1860 agno
= XFS_INO_TO_AGNO(mp
, ip
->i_ino
);
1861 agdaddr
= XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
));
1864 * Get the agi buffer first. It ensures lock ordering
1867 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, agdaddr
,
1868 XFS_FSS_TO_BB(mp
, 1), 0, &agibp
);
1873 * Validate the magic number of the agi block.
1875 agi
= XFS_BUF_TO_AGI(agibp
);
1877 be32_to_cpu(agi
->agi_magicnum
) == XFS_AGI_MAGIC
&&
1878 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi
->agi_versionnum
));
1879 if (unlikely(XFS_TEST_ERROR(!agi_ok
, mp
, XFS_ERRTAG_IUNLINK
,
1880 XFS_RANDOM_IUNLINK
))) {
1881 XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW
, mp
, agi
);
1882 xfs_trans_brelse(tp
, agibp
);
1883 return XFS_ERROR(EFSCORRUPTED
);
1886 * Get the index into the agi hash table for the
1887 * list this inode will go on.
1889 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
1891 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
1892 ASSERT(agi
->agi_unlinked
[bucket_index
]);
1893 ASSERT(be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != agino
);
1895 if (be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != NULLAGINO
) {
1897 * There is already another inode in the bucket we need
1898 * to add ourselves to. Add us at the front of the list.
1899 * Here we put the head pointer into our next pointer,
1900 * and then we fall through to point the head at us.
1902 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, 0);
1906 ASSERT(INT_GET(dip
->di_next_unlinked
, ARCH_CONVERT
) == NULLAGINO
);
1907 ASSERT(dip
->di_next_unlinked
);
1908 /* both on-disk, don't endian flip twice */
1909 dip
->di_next_unlinked
= agi
->agi_unlinked
[bucket_index
];
1910 offset
= ip
->i_boffset
+
1911 offsetof(xfs_dinode_t
, di_next_unlinked
);
1912 xfs_trans_inode_buf(tp
, ibp
);
1913 xfs_trans_log_buf(tp
, ibp
, offset
,
1914 (offset
+ sizeof(xfs_agino_t
) - 1));
1915 xfs_inobp_check(mp
, ibp
);
1919 * Point the bucket head pointer at the inode being inserted.
1922 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(agino
);
1923 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
1924 (sizeof(xfs_agino_t
) * bucket_index
);
1925 xfs_trans_log_buf(tp
, agibp
, offset
,
1926 (offset
+ sizeof(xfs_agino_t
) - 1));
1931 * Pull the on-disk inode from the AGI unlinked list.
1944 xfs_agnumber_t agno
;
1945 xfs_daddr_t agdaddr
;
1947 xfs_agino_t next_agino
;
1948 xfs_buf_t
*last_ibp
;
1949 xfs_dinode_t
*last_dip
;
1951 int offset
, last_offset
;
1956 * First pull the on-disk inode from the AGI unlinked list.
1960 agno
= XFS_INO_TO_AGNO(mp
, ip
->i_ino
);
1961 agdaddr
= XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
));
1964 * Get the agi buffer first. It ensures lock ordering
1967 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, agdaddr
,
1968 XFS_FSS_TO_BB(mp
, 1), 0, &agibp
);
1971 "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
1972 error
, mp
->m_fsname
);
1976 * Validate the magic number of the agi block.
1978 agi
= XFS_BUF_TO_AGI(agibp
);
1980 be32_to_cpu(agi
->agi_magicnum
) == XFS_AGI_MAGIC
&&
1981 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi
->agi_versionnum
));
1982 if (unlikely(XFS_TEST_ERROR(!agi_ok
, mp
, XFS_ERRTAG_IUNLINK_REMOVE
,
1983 XFS_RANDOM_IUNLINK_REMOVE
))) {
1984 XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW
,
1986 xfs_trans_brelse(tp
, agibp
);
1988 "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
1990 return XFS_ERROR(EFSCORRUPTED
);
1993 * Get the index into the agi hash table for the
1994 * list this inode will go on.
1996 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
1998 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
1999 ASSERT(be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) != NULLAGINO
);
2000 ASSERT(agi
->agi_unlinked
[bucket_index
]);
2002 if (be32_to_cpu(agi
->agi_unlinked
[bucket_index
]) == agino
) {
2004 * We're at the head of the list. Get the inode's
2005 * on-disk buffer to see if there is anyone after us
2006 * on the list. Only modify our next pointer if it
2007 * is not already NULLAGINO. This saves us the overhead
2008 * of dealing with the buffer when there is no need to
2011 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, 0);
2014 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2015 error
, mp
->m_fsname
);
2018 next_agino
= INT_GET(dip
->di_next_unlinked
, ARCH_CONVERT
);
2019 ASSERT(next_agino
!= 0);
2020 if (next_agino
!= NULLAGINO
) {
2021 INT_SET(dip
->di_next_unlinked
, ARCH_CONVERT
, NULLAGINO
);
2022 offset
= ip
->i_boffset
+
2023 offsetof(xfs_dinode_t
, di_next_unlinked
);
2024 xfs_trans_inode_buf(tp
, ibp
);
2025 xfs_trans_log_buf(tp
, ibp
, offset
,
2026 (offset
+ sizeof(xfs_agino_t
) - 1));
2027 xfs_inobp_check(mp
, ibp
);
2029 xfs_trans_brelse(tp
, ibp
);
2032 * Point the bucket head pointer at the next inode.
2034 ASSERT(next_agino
!= 0);
2035 ASSERT(next_agino
!= agino
);
2036 agi
->agi_unlinked
[bucket_index
] = cpu_to_be32(next_agino
);
2037 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
2038 (sizeof(xfs_agino_t
) * bucket_index
);
2039 xfs_trans_log_buf(tp
, agibp
, offset
,
2040 (offset
+ sizeof(xfs_agino_t
) - 1));
2043 * We need to search the list for the inode being freed.
2045 next_agino
= be32_to_cpu(agi
->agi_unlinked
[bucket_index
]);
2047 while (next_agino
!= agino
) {
2049 * If the last inode wasn't the one pointing to
2050 * us, then release its buffer since we're not
2051 * going to do anything with it.
2053 if (last_ibp
!= NULL
) {
2054 xfs_trans_brelse(tp
, last_ibp
);
2056 next_ino
= XFS_AGINO_TO_INO(mp
, agno
, next_agino
);
2057 error
= xfs_inotobp(mp
, tp
, next_ino
, &last_dip
,
2058 &last_ibp
, &last_offset
);
2061 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
2062 error
, mp
->m_fsname
);
2065 next_agino
= INT_GET(last_dip
->di_next_unlinked
, ARCH_CONVERT
);
2066 ASSERT(next_agino
!= NULLAGINO
);
2067 ASSERT(next_agino
!= 0);
2070 * Now last_ibp points to the buffer previous to us on
2071 * the unlinked list. Pull us from the list.
2073 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, 0);
2076 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2077 error
, mp
->m_fsname
);
2080 next_agino
= INT_GET(dip
->di_next_unlinked
, ARCH_CONVERT
);
2081 ASSERT(next_agino
!= 0);
2082 ASSERT(next_agino
!= agino
);
2083 if (next_agino
!= NULLAGINO
) {
2084 INT_SET(dip
->di_next_unlinked
, ARCH_CONVERT
, NULLAGINO
);
2085 offset
= ip
->i_boffset
+
2086 offsetof(xfs_dinode_t
, di_next_unlinked
);
2087 xfs_trans_inode_buf(tp
, ibp
);
2088 xfs_trans_log_buf(tp
, ibp
, offset
,
2089 (offset
+ sizeof(xfs_agino_t
) - 1));
2090 xfs_inobp_check(mp
, ibp
);
2092 xfs_trans_brelse(tp
, ibp
);
2095 * Point the previous inode on the list to the next inode.
2097 INT_SET(last_dip
->di_next_unlinked
, ARCH_CONVERT
, next_agino
);
2098 ASSERT(next_agino
!= 0);
2099 offset
= last_offset
+ offsetof(xfs_dinode_t
, di_next_unlinked
);
2100 xfs_trans_inode_buf(tp
, last_ibp
);
2101 xfs_trans_log_buf(tp
, last_ibp
, offset
,
2102 (offset
+ sizeof(xfs_agino_t
) - 1));
2103 xfs_inobp_check(mp
, last_ibp
);
2108 static __inline__
int xfs_inode_clean(xfs_inode_t
*ip
)
2110 return (((ip
->i_itemp
== NULL
) ||
2111 !(ip
->i_itemp
->ili_format
.ilf_fields
& XFS_ILOG_ALL
)) &&
2112 (ip
->i_update_core
== 0));
2117 xfs_inode_t
*free_ip
,
2121 xfs_mount_t
*mp
= free_ip
->i_mount
;
2122 int blks_per_cluster
;
2125 int i
, j
, found
, pre_flushed
;
2129 xfs_inode_t
*ip
, **ip_found
;
2130 xfs_inode_log_item_t
*iip
;
2131 xfs_log_item_t
*lip
;
2134 if (mp
->m_sb
.sb_blocksize
>= XFS_INODE_CLUSTER_SIZE(mp
)) {
2135 blks_per_cluster
= 1;
2136 ninodes
= mp
->m_sb
.sb_inopblock
;
2137 nbufs
= XFS_IALLOC_BLOCKS(mp
);
2139 blks_per_cluster
= XFS_INODE_CLUSTER_SIZE(mp
) /
2140 mp
->m_sb
.sb_blocksize
;
2141 ninodes
= blks_per_cluster
* mp
->m_sb
.sb_inopblock
;
2142 nbufs
= XFS_IALLOC_BLOCKS(mp
) / blks_per_cluster
;
2145 ip_found
= kmem_alloc(ninodes
* sizeof(xfs_inode_t
*), KM_NOFS
);
2147 for (j
= 0; j
< nbufs
; j
++, inum
+= ninodes
) {
2148 blkno
= XFS_AGB_TO_DADDR(mp
, XFS_INO_TO_AGNO(mp
, inum
),
2149 XFS_INO_TO_AGBNO(mp
, inum
));
2153 * Look for each inode in memory and attempt to lock it,
2154 * we can be racing with flush and tail pushing here.
2155 * any inode we get the locks on, add to an array of
2156 * inode items to process later.
2158 * The get the buffer lock, we could beat a flush
2159 * or tail pushing thread to the lock here, in which
2160 * case they will go looking for the inode buffer
2161 * and fail, we need some other form of interlock
2165 for (i
= 0; i
< ninodes
; i
++) {
2166 ih
= XFS_IHASH(mp
, inum
+ i
);
2167 read_lock(&ih
->ih_lock
);
2168 for (ip
= ih
->ih_next
; ip
!= NULL
; ip
= ip
->i_next
) {
2169 if (ip
->i_ino
== inum
+ i
)
2173 /* Inode not in memory or we found it already,
2176 if (!ip
|| (ip
->i_flags
& XFS_ISTALE
)) {
2177 read_unlock(&ih
->ih_lock
);
2181 if (xfs_inode_clean(ip
)) {
2182 read_unlock(&ih
->ih_lock
);
2186 /* If we can get the locks then add it to the
2187 * list, otherwise by the time we get the bp lock
2188 * below it will already be attached to the
2192 /* This inode will already be locked - by us, lets
2196 if (ip
== free_ip
) {
2197 if (xfs_iflock_nowait(ip
)) {
2198 ip
->i_flags
|= XFS_ISTALE
;
2200 if (xfs_inode_clean(ip
)) {
2203 ip_found
[found
++] = ip
;
2206 read_unlock(&ih
->ih_lock
);
2210 if (xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
)) {
2211 if (xfs_iflock_nowait(ip
)) {
2212 ip
->i_flags
|= XFS_ISTALE
;
2214 if (xfs_inode_clean(ip
)) {
2216 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2218 ip_found
[found
++] = ip
;
2221 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2225 read_unlock(&ih
->ih_lock
);
2228 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, blkno
,
2229 mp
->m_bsize
* blks_per_cluster
,
2233 lip
= XFS_BUF_FSPRIVATE(bp
, xfs_log_item_t
*);
2235 if (lip
->li_type
== XFS_LI_INODE
) {
2236 iip
= (xfs_inode_log_item_t
*)lip
;
2237 ASSERT(iip
->ili_logged
== 1);
2238 lip
->li_cb
= (void(*)(xfs_buf_t
*,xfs_log_item_t
*)) xfs_istale_done
;
2240 iip
->ili_flush_lsn
= iip
->ili_item
.li_lsn
;
2242 iip
->ili_inode
->i_flags
|= XFS_ISTALE
;
2245 lip
= lip
->li_bio_list
;
2248 for (i
= 0; i
< found
; i
++) {
2253 ip
->i_update_core
= 0;
2255 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2259 iip
->ili_last_fields
= iip
->ili_format
.ilf_fields
;
2260 iip
->ili_format
.ilf_fields
= 0;
2261 iip
->ili_logged
= 1;
2263 iip
->ili_flush_lsn
= iip
->ili_item
.li_lsn
;
2266 xfs_buf_attach_iodone(bp
,
2267 (void(*)(xfs_buf_t
*,xfs_log_item_t
*))
2268 xfs_istale_done
, (xfs_log_item_t
*)iip
);
2269 if (ip
!= free_ip
) {
2270 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2274 if (found
|| pre_flushed
)
2275 xfs_trans_stale_inode_buf(tp
, bp
);
2276 xfs_trans_binval(tp
, bp
);
2279 kmem_free(ip_found
, ninodes
* sizeof(xfs_inode_t
*));
2283 * This is called to return an inode to the inode free list.
2284 * The inode should already be truncated to 0 length and have
2285 * no pages associated with it. This routine also assumes that
2286 * the inode is already a part of the transaction.
2288 * The on-disk copy of the inode will have been added to the list
2289 * of unlinked inodes in the AGI. We need to remove the inode from
2290 * that list atomically with respect to freeing it here.
2296 xfs_bmap_free_t
*flist
)
2300 xfs_ino_t first_ino
;
2302 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
));
2303 ASSERT(ip
->i_transp
== tp
);
2304 ASSERT(ip
->i_d
.di_nlink
== 0);
2305 ASSERT(ip
->i_d
.di_nextents
== 0);
2306 ASSERT(ip
->i_d
.di_anextents
== 0);
2307 ASSERT((ip
->i_d
.di_size
== 0) ||
2308 ((ip
->i_d
.di_mode
& S_IFMT
) != S_IFREG
));
2309 ASSERT(ip
->i_d
.di_nblocks
== 0);
2312 * Pull the on-disk inode from the AGI unlinked list.
2314 error
= xfs_iunlink_remove(tp
, ip
);
2319 error
= xfs_difree(tp
, ip
->i_ino
, flist
, &delete, &first_ino
);
2323 ip
->i_d
.di_mode
= 0; /* mark incore inode as free */
2324 ip
->i_d
.di_flags
= 0;
2325 ip
->i_d
.di_dmevmask
= 0;
2326 ip
->i_d
.di_forkoff
= 0; /* mark the attr fork not in use */
2327 ip
->i_df
.if_ext_max
=
2328 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
2329 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
2330 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
2332 * Bump the generation count so no one will be confused
2333 * by reincarnations of this inode.
2336 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2339 xfs_ifree_cluster(ip
, tp
, first_ino
);
2346 * Reallocate the space for if_broot based on the number of records
2347 * being added or deleted as indicated in rec_diff. Move the records
2348 * and pointers in if_broot to fit the new size. When shrinking this
2349 * will eliminate holes between the records and pointers created by
2350 * the caller. When growing this will create holes to be filled in
2353 * The caller must not request to add more records than would fit in
2354 * the on-disk inode root. If the if_broot is currently NULL, then
2355 * if we adding records one will be allocated. The caller must also
2356 * not request that the number of records go below zero, although
2357 * it can go to zero.
2359 * ip -- the inode whose if_broot area is changing
2360 * ext_diff -- the change in the number of records, positive or negative,
2361 * requested for the if_broot array.
2371 xfs_bmbt_block_t
*new_broot
;
2378 * Handle the degenerate case quietly.
2380 if (rec_diff
== 0) {
2384 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2387 * If there wasn't any memory allocated before, just
2388 * allocate it now and get out.
2390 if (ifp
->if_broot_bytes
== 0) {
2391 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff
);
2392 ifp
->if_broot
= (xfs_bmbt_block_t
*)kmem_alloc(new_size
,
2394 ifp
->if_broot_bytes
= (int)new_size
;
2399 * If there is already an existing if_broot, then we need
2400 * to realloc() it and shift the pointers to their new
2401 * location. The records don't change location because
2402 * they are kept butted up against the btree block header.
2404 cur_max
= XFS_BMAP_BROOT_MAXRECS(ifp
->if_broot_bytes
);
2405 new_max
= cur_max
+ rec_diff
;
2406 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max
);
2407 ifp
->if_broot
= (xfs_bmbt_block_t
*)
2408 kmem_realloc(ifp
->if_broot
,
2410 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max
), /* old size */
2412 op
= (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp
->if_broot
, 1,
2413 ifp
->if_broot_bytes
);
2414 np
= (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp
->if_broot
, 1,
2416 ifp
->if_broot_bytes
= (int)new_size
;
2417 ASSERT(ifp
->if_broot_bytes
<=
2418 XFS_IFORK_SIZE(ip
, whichfork
) + XFS_BROOT_SIZE_ADJ
);
2419 memmove(np
, op
, cur_max
* (uint
)sizeof(xfs_dfsbno_t
));
2424 * rec_diff is less than 0. In this case, we are shrinking the
2425 * if_broot buffer. It must already exist. If we go to zero
2426 * records, just get rid of the root and clear the status bit.
2428 ASSERT((ifp
->if_broot
!= NULL
) && (ifp
->if_broot_bytes
> 0));
2429 cur_max
= XFS_BMAP_BROOT_MAXRECS(ifp
->if_broot_bytes
);
2430 new_max
= cur_max
+ rec_diff
;
2431 ASSERT(new_max
>= 0);
2433 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max
);
2437 new_broot
= (xfs_bmbt_block_t
*)kmem_alloc(new_size
, KM_SLEEP
);
2439 * First copy over the btree block header.
2441 memcpy(new_broot
, ifp
->if_broot
, sizeof(xfs_bmbt_block_t
));
2444 ifp
->if_flags
&= ~XFS_IFBROOT
;
2448 * Only copy the records and pointers if there are any.
2452 * First copy the records.
2454 op
= (char *)XFS_BMAP_BROOT_REC_ADDR(ifp
->if_broot
, 1,
2455 ifp
->if_broot_bytes
);
2456 np
= (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot
, 1,
2458 memcpy(np
, op
, new_max
* (uint
)sizeof(xfs_bmbt_rec_t
));
2461 * Then copy the pointers.
2463 op
= (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp
->if_broot
, 1,
2464 ifp
->if_broot_bytes
);
2465 np
= (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot
, 1,
2467 memcpy(np
, op
, new_max
* (uint
)sizeof(xfs_dfsbno_t
));
2469 kmem_free(ifp
->if_broot
, ifp
->if_broot_bytes
);
2470 ifp
->if_broot
= new_broot
;
2471 ifp
->if_broot_bytes
= (int)new_size
;
2472 ASSERT(ifp
->if_broot_bytes
<=
2473 XFS_IFORK_SIZE(ip
, whichfork
) + XFS_BROOT_SIZE_ADJ
);
2479 * This is called when the amount of space needed for if_extents
2480 * is increased or decreased. The change in size is indicated by
2481 * the number of extents that need to be added or deleted in the
2482 * ext_diff parameter.
2484 * If the amount of space needed has decreased below the size of the
2485 * inline buffer, then switch to using the inline buffer. Otherwise,
2486 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2487 * to what is needed.
2489 * ip -- the inode whose if_extents area is changing
2490 * ext_diff -- the change in the number of extents, positive or negative,
2491 * requested for the if_extents array.
2504 if (ext_diff
== 0) {
2508 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2509 byte_diff
= ext_diff
* (uint
)sizeof(xfs_bmbt_rec_t
);
2510 new_size
= (int)ifp
->if_bytes
+ byte_diff
;
2511 ASSERT(new_size
>= 0);
2513 if (new_size
== 0) {
2514 if (ifp
->if_u1
.if_extents
!= ifp
->if_u2
.if_inline_ext
) {
2515 ASSERT(ifp
->if_real_bytes
!= 0);
2516 kmem_free(ifp
->if_u1
.if_extents
, ifp
->if_real_bytes
);
2518 ifp
->if_u1
.if_extents
= NULL
;
2520 } else if (new_size
<= sizeof(ifp
->if_u2
.if_inline_ext
)) {
2522 * If the valid extents can fit in if_inline_ext,
2523 * copy them from the malloc'd vector and free it.
2525 if (ifp
->if_u1
.if_extents
!= ifp
->if_u2
.if_inline_ext
) {
2527 * For now, empty files are format EXTENTS,
2528 * so the if_extents pointer is null.
2530 if (ifp
->if_u1
.if_extents
) {
2531 memcpy(ifp
->if_u2
.if_inline_ext
,
2532 ifp
->if_u1
.if_extents
, new_size
);
2533 kmem_free(ifp
->if_u1
.if_extents
,
2534 ifp
->if_real_bytes
);
2536 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
2540 rnew_size
= new_size
;
2541 if ((rnew_size
& (rnew_size
- 1)) != 0)
2542 rnew_size
= xfs_iroundup(rnew_size
);
2544 * Stuck with malloc/realloc.
2546 if (ifp
->if_u1
.if_extents
== ifp
->if_u2
.if_inline_ext
) {
2547 ifp
->if_u1
.if_extents
= (xfs_bmbt_rec_t
*)
2548 kmem_alloc(rnew_size
, KM_SLEEP
);
2549 memcpy(ifp
->if_u1
.if_extents
, ifp
->if_u2
.if_inline_ext
,
2550 sizeof(ifp
->if_u2
.if_inline_ext
));
2551 } else if (rnew_size
!= ifp
->if_real_bytes
) {
2552 ifp
->if_u1
.if_extents
= (xfs_bmbt_rec_t
*)
2553 kmem_realloc(ifp
->if_u1
.if_extents
,
2559 ifp
->if_real_bytes
= rnew_size
;
2560 ifp
->if_bytes
= new_size
;
2565 * This is called when the amount of space needed for if_data
2566 * is increased or decreased. The change in size is indicated by
2567 * the number of bytes that need to be added or deleted in the
2568 * byte_diff parameter.
2570 * If the amount of space needed has decreased below the size of the
2571 * inline buffer, then switch to using the inline buffer. Otherwise,
2572 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2573 * to what is needed.
2575 * ip -- the inode whose if_data area is changing
2576 * byte_diff -- the change in the number of bytes, positive or negative,
2577 * requested for the if_data array.
2589 if (byte_diff
== 0) {
2593 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2594 new_size
= (int)ifp
->if_bytes
+ byte_diff
;
2595 ASSERT(new_size
>= 0);
2597 if (new_size
== 0) {
2598 if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
2599 kmem_free(ifp
->if_u1
.if_data
, ifp
->if_real_bytes
);
2601 ifp
->if_u1
.if_data
= NULL
;
2603 } else if (new_size
<= sizeof(ifp
->if_u2
.if_inline_data
)) {
2605 * If the valid extents/data can fit in if_inline_ext/data,
2606 * copy them from the malloc'd vector and free it.
2608 if (ifp
->if_u1
.if_data
== NULL
) {
2609 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
2610 } else if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
2611 ASSERT(ifp
->if_real_bytes
!= 0);
2612 memcpy(ifp
->if_u2
.if_inline_data
, ifp
->if_u1
.if_data
,
2614 kmem_free(ifp
->if_u1
.if_data
, ifp
->if_real_bytes
);
2615 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
2620 * Stuck with malloc/realloc.
2621 * For inline data, the underlying buffer must be
2622 * a multiple of 4 bytes in size so that it can be
2623 * logged and stay on word boundaries. We enforce
2626 real_size
= roundup(new_size
, 4);
2627 if (ifp
->if_u1
.if_data
== NULL
) {
2628 ASSERT(ifp
->if_real_bytes
== 0);
2629 ifp
->if_u1
.if_data
= kmem_alloc(real_size
, KM_SLEEP
);
2630 } else if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
2632 * Only do the realloc if the underlying size
2633 * is really changing.
2635 if (ifp
->if_real_bytes
!= real_size
) {
2636 ifp
->if_u1
.if_data
=
2637 kmem_realloc(ifp
->if_u1
.if_data
,
2643 ASSERT(ifp
->if_real_bytes
== 0);
2644 ifp
->if_u1
.if_data
= kmem_alloc(real_size
, KM_SLEEP
);
2645 memcpy(ifp
->if_u1
.if_data
, ifp
->if_u2
.if_inline_data
,
2649 ifp
->if_real_bytes
= real_size
;
2650 ifp
->if_bytes
= new_size
;
2651 ASSERT(ifp
->if_bytes
<= XFS_IFORK_SIZE(ip
, whichfork
));
2658 * Map inode to disk block and offset.
2660 * mp -- the mount point structure for the current file system
2661 * tp -- the current transaction
2662 * ino -- the inode number of the inode to be located
2663 * imap -- this structure is filled in with the information necessary
2664 * to retrieve the given inode from disk
2665 * flags -- flags to pass to xfs_dilocate indicating whether or not
2666 * lookups in the inode btree were OK or not
2676 xfs_fsblock_t fsbno
;
2681 fsbno
= imap
->im_blkno
?
2682 XFS_DADDR_TO_FSB(mp
, imap
->im_blkno
) : NULLFSBLOCK
;
2683 error
= xfs_dilocate(mp
, tp
, ino
, &fsbno
, &len
, &off
, flags
);
2687 imap
->im_blkno
= XFS_FSB_TO_DADDR(mp
, fsbno
);
2688 imap
->im_len
= XFS_FSB_TO_BB(mp
, len
);
2689 imap
->im_agblkno
= XFS_FSB_TO_AGBNO(mp
, fsbno
);
2690 imap
->im_ioffset
= (ushort
)off
;
2691 imap
->im_boffset
= (ushort
)(off
<< mp
->m_sb
.sb_inodelog
);
2702 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2703 if (ifp
->if_broot
!= NULL
) {
2704 kmem_free(ifp
->if_broot
, ifp
->if_broot_bytes
);
2705 ifp
->if_broot
= NULL
;
2709 * If the format is local, then we can't have an extents
2710 * array so just look for an inline data array. If we're
2711 * not local then we may or may not have an extents list,
2712 * so check and free it up if we do.
2714 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
) {
2715 if ((ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) &&
2716 (ifp
->if_u1
.if_data
!= NULL
)) {
2717 ASSERT(ifp
->if_real_bytes
!= 0);
2718 kmem_free(ifp
->if_u1
.if_data
, ifp
->if_real_bytes
);
2719 ifp
->if_u1
.if_data
= NULL
;
2720 ifp
->if_real_bytes
= 0;
2722 } else if ((ifp
->if_flags
& XFS_IFEXTENTS
) &&
2723 (ifp
->if_u1
.if_extents
!= NULL
) &&
2724 (ifp
->if_u1
.if_extents
!= ifp
->if_u2
.if_inline_ext
)) {
2725 ASSERT(ifp
->if_real_bytes
!= 0);
2726 kmem_free(ifp
->if_u1
.if_extents
, ifp
->if_real_bytes
);
2727 ifp
->if_u1
.if_extents
= NULL
;
2728 ifp
->if_real_bytes
= 0;
2730 ASSERT(ifp
->if_u1
.if_extents
== NULL
||
2731 ifp
->if_u1
.if_extents
== ifp
->if_u2
.if_inline_ext
);
2732 ASSERT(ifp
->if_real_bytes
== 0);
2733 if (whichfork
== XFS_ATTR_FORK
) {
2734 kmem_zone_free(xfs_ifork_zone
, ip
->i_afp
);
2740 * This is called free all the memory associated with an inode.
2741 * It must free the inode itself and any buffers allocated for
2742 * if_extents/if_data and if_broot. It must also free the lock
2743 * associated with the inode.
2750 switch (ip
->i_d
.di_mode
& S_IFMT
) {
2754 xfs_idestroy_fork(ip
, XFS_DATA_FORK
);
2758 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
2759 mrfree(&ip
->i_lock
);
2760 mrfree(&ip
->i_iolock
);
2761 freesema(&ip
->i_flock
);
2762 #ifdef XFS_BMAP_TRACE
2763 ktrace_free(ip
->i_xtrace
);
2765 #ifdef XFS_BMBT_TRACE
2766 ktrace_free(ip
->i_btrace
);
2769 ktrace_free(ip
->i_rwtrace
);
2771 #ifdef XFS_ILOCK_TRACE
2772 ktrace_free(ip
->i_lock_trace
);
2774 #ifdef XFS_DIR2_TRACE
2775 ktrace_free(ip
->i_dir_trace
);
2778 /* XXXdpd should be able to assert this but shutdown
2779 * is leaving the AIL behind. */
2780 ASSERT(((ip
->i_itemp
->ili_item
.li_flags
& XFS_LI_IN_AIL
) == 0) ||
2781 XFS_FORCED_SHUTDOWN(ip
->i_mount
));
2782 xfs_inode_item_destroy(ip
);
2784 kmem_zone_free(xfs_inode_zone
, ip
);
2789 * Increment the pin count of the given buffer.
2790 * This value is protected by ipinlock spinlock in the mount structure.
2796 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
));
2798 atomic_inc(&ip
->i_pincount
);
2802 * Decrement the pin count of the given inode, and wake up
2803 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2804 * inode must have been previoulsy pinned with a call to xfs_ipin().
2810 ASSERT(atomic_read(&ip
->i_pincount
) > 0);
2812 if (atomic_dec_and_test(&ip
->i_pincount
)) {
2813 vnode_t
*vp
= XFS_ITOV_NULL(ip
);
2815 /* make sync come back and flush this inode */
2817 struct inode
*inode
= LINVFS_GET_IP(vp
);
2819 if (!(inode
->i_state
& I_NEW
))
2820 mark_inode_dirty_sync(inode
);
2823 wake_up(&ip
->i_ipin_wait
);
2828 * This is called to wait for the given inode to be unpinned.
2829 * It will sleep until this happens. The caller must have the
2830 * inode locked in at least shared mode so that the buffer cannot
2831 * be subsequently pinned once someone is waiting for it to be
2838 xfs_inode_log_item_t
*iip
;
2841 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
| MR_ACCESS
));
2843 if (atomic_read(&ip
->i_pincount
) == 0) {
2848 if (iip
&& iip
->ili_last_lsn
) {
2849 lsn
= iip
->ili_last_lsn
;
2855 * Give the log a push so we don't wait here too long.
2857 xfs_log_force(ip
->i_mount
, lsn
, XFS_LOG_FORCE
);
2859 wait_event(ip
->i_ipin_wait
, (atomic_read(&ip
->i_pincount
) == 0));
2864 * xfs_iextents_copy()
2866 * This is called to copy the REAL extents (as opposed to the delayed
2867 * allocation extents) from the inode into the given buffer. It
2868 * returns the number of bytes copied into the buffer.
2870 * If there are no delayed allocation extents, then we can just
2871 * memcpy() the extents into the buffer. Otherwise, we need to
2872 * examine each extent in turn and skip those which are delayed.
2877 xfs_bmbt_rec_t
*buffer
,
2881 xfs_bmbt_rec_t
*dest_ep
;
2883 #ifdef XFS_BMAP_TRACE
2884 static char fname
[] = "xfs_iextents_copy";
2889 xfs_fsblock_t start_block
;
2891 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2892 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
|MR_ACCESS
));
2893 ASSERT(ifp
->if_bytes
> 0);
2895 nrecs
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
2896 xfs_bmap_trace_exlist(fname
, ip
, nrecs
, whichfork
);
2900 * There are some delayed allocation extents in the
2901 * inode, so copy the extents one at a time and skip
2902 * the delayed ones. There must be at least one
2903 * non-delayed extent.
2905 ep
= ifp
->if_u1
.if_extents
;
2908 for (i
= 0; i
< nrecs
; i
++) {
2909 start_block
= xfs_bmbt_get_startblock(ep
);
2910 if (ISNULLSTARTBLOCK(start_block
)) {
2912 * It's a delayed allocation extent, so skip it.
2918 /* Translate to on disk format */
2919 put_unaligned(INT_GET(ep
->l0
, ARCH_CONVERT
),
2920 (__uint64_t
*)&dest_ep
->l0
);
2921 put_unaligned(INT_GET(ep
->l1
, ARCH_CONVERT
),
2922 (__uint64_t
*)&dest_ep
->l1
);
2927 ASSERT(copied
!= 0);
2928 xfs_validate_extents(buffer
, copied
, 1, XFS_EXTFMT_INODE(ip
));
2930 return (copied
* (uint
)sizeof(xfs_bmbt_rec_t
));
2934 * Each of the following cases stores data into the same region
2935 * of the on-disk inode, so only one of them can be valid at
2936 * any given time. While it is possible to have conflicting formats
2937 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2938 * in EXTENTS format, this can only happen when the fork has
2939 * changed formats after being modified but before being flushed.
2940 * In these cases, the format always takes precedence, because the
2941 * format indicates the current state of the fork.
2948 xfs_inode_log_item_t
*iip
,
2955 #ifdef XFS_TRANS_DEBUG
2958 static const short brootflag
[2] =
2959 { XFS_ILOG_DBROOT
, XFS_ILOG_ABROOT
};
2960 static const short dataflag
[2] =
2961 { XFS_ILOG_DDATA
, XFS_ILOG_ADATA
};
2962 static const short extflag
[2] =
2963 { XFS_ILOG_DEXT
, XFS_ILOG_AEXT
};
2967 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2969 * This can happen if we gave up in iformat in an error path,
2970 * for the attribute fork.
2973 ASSERT(whichfork
== XFS_ATTR_FORK
);
2976 cp
= XFS_DFORK_PTR(dip
, whichfork
);
2978 switch (XFS_IFORK_FORMAT(ip
, whichfork
)) {
2979 case XFS_DINODE_FMT_LOCAL
:
2980 if ((iip
->ili_format
.ilf_fields
& dataflag
[whichfork
]) &&
2981 (ifp
->if_bytes
> 0)) {
2982 ASSERT(ifp
->if_u1
.if_data
!= NULL
);
2983 ASSERT(ifp
->if_bytes
<= XFS_IFORK_SIZE(ip
, whichfork
));
2984 memcpy(cp
, ifp
->if_u1
.if_data
, ifp
->if_bytes
);
2986 if (whichfork
== XFS_DATA_FORK
) {
2987 if (unlikely(XFS_DIR_SHORTFORM_VALIDATE_ONDISK(mp
, dip
))) {
2988 XFS_ERROR_REPORT("xfs_iflush_fork",
2989 XFS_ERRLEVEL_LOW
, mp
);
2990 return XFS_ERROR(EFSCORRUPTED
);
2995 case XFS_DINODE_FMT_EXTENTS
:
2996 ASSERT((ifp
->if_flags
& XFS_IFEXTENTS
) ||
2997 !(iip
->ili_format
.ilf_fields
& extflag
[whichfork
]));
2998 ASSERT((ifp
->if_u1
.if_extents
!= NULL
) || (ifp
->if_bytes
== 0));
2999 ASSERT((ifp
->if_u1
.if_extents
== NULL
) || (ifp
->if_bytes
> 0));
3000 if ((iip
->ili_format
.ilf_fields
& extflag
[whichfork
]) &&
3001 (ifp
->if_bytes
> 0)) {
3002 ASSERT(XFS_IFORK_NEXTENTS(ip
, whichfork
) > 0);
3003 (void)xfs_iextents_copy(ip
, (xfs_bmbt_rec_t
*)cp
,
3008 case XFS_DINODE_FMT_BTREE
:
3009 if ((iip
->ili_format
.ilf_fields
& brootflag
[whichfork
]) &&
3010 (ifp
->if_broot_bytes
> 0)) {
3011 ASSERT(ifp
->if_broot
!= NULL
);
3012 ASSERT(ifp
->if_broot_bytes
<=
3013 (XFS_IFORK_SIZE(ip
, whichfork
) +
3014 XFS_BROOT_SIZE_ADJ
));
3015 xfs_bmbt_to_bmdr(ifp
->if_broot
, ifp
->if_broot_bytes
,
3016 (xfs_bmdr_block_t
*)cp
,
3017 XFS_DFORK_SIZE(dip
, mp
, whichfork
));
3021 case XFS_DINODE_FMT_DEV
:
3022 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_DEV
) {
3023 ASSERT(whichfork
== XFS_DATA_FORK
);
3024 INT_SET(dip
->di_u
.di_dev
, ARCH_CONVERT
, ip
->i_df
.if_u2
.if_rdev
);
3028 case XFS_DINODE_FMT_UUID
:
3029 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_UUID
) {
3030 ASSERT(whichfork
== XFS_DATA_FORK
);
3031 memcpy(&dip
->di_u
.di_muuid
, &ip
->i_df
.if_u2
.if_uuid
,
3045 * xfs_iflush() will write a modified inode's changes out to the
3046 * inode's on disk home. The caller must have the inode lock held
3047 * in at least shared mode and the inode flush semaphore must be
3048 * held as well. The inode lock will still be held upon return from
3049 * the call and the caller is free to unlock it.
3050 * The inode flush lock will be unlocked when the inode reaches the disk.
3051 * The flags indicate how the inode's buffer should be written out.
3058 xfs_inode_log_item_t
*iip
;
3066 int clcount
; /* count of inodes clustered */
3068 enum { INT_DELWRI
= (1 << 0), INT_ASYNC
= (1 << 1) };
3071 XFS_STATS_INC(xs_iflush_count
);
3073 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
|MR_ACCESS
));
3074 ASSERT(valusema(&ip
->i_flock
) <= 0);
3075 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
3076 ip
->i_d
.di_nextents
> ip
->i_df
.if_ext_max
);
3082 * If the inode isn't dirty, then just release the inode
3083 * flush lock and do nothing.
3085 if ((ip
->i_update_core
== 0) &&
3086 ((iip
== NULL
) || !(iip
->ili_format
.ilf_fields
& XFS_ILOG_ALL
))) {
3087 ASSERT((iip
!= NULL
) ?
3088 !(iip
->ili_item
.li_flags
& XFS_LI_IN_AIL
) : 1);
3094 * We can't flush the inode until it is unpinned, so
3095 * wait for it. We know noone new can pin it, because
3096 * we are holding the inode lock shared and you need
3097 * to hold it exclusively to pin the inode.
3099 xfs_iunpin_wait(ip
);
3102 * This may have been unpinned because the filesystem is shutting
3103 * down forcibly. If that's the case we must not write this inode
3104 * to disk, because the log record didn't make it to disk!
3106 if (XFS_FORCED_SHUTDOWN(mp
)) {
3107 ip
->i_update_core
= 0;
3109 iip
->ili_format
.ilf_fields
= 0;
3111 return XFS_ERROR(EIO
);
3115 * Get the buffer containing the on-disk inode.
3117 error
= xfs_itobp(mp
, NULL
, ip
, &dip
, &bp
, 0);
3124 * Decide how buffer will be flushed out. This is done before
3125 * the call to xfs_iflush_int because this field is zeroed by it.
3127 if (iip
!= NULL
&& iip
->ili_format
.ilf_fields
!= 0) {
3129 * Flush out the inode buffer according to the directions
3130 * of the caller. In the cases where the caller has given
3131 * us a choice choose the non-delwri case. This is because
3132 * the inode is in the AIL and we need to get it out soon.
3135 case XFS_IFLUSH_SYNC
:
3136 case XFS_IFLUSH_DELWRI_ELSE_SYNC
:
3139 case XFS_IFLUSH_ASYNC
:
3140 case XFS_IFLUSH_DELWRI_ELSE_ASYNC
:
3143 case XFS_IFLUSH_DELWRI
:
3153 case XFS_IFLUSH_DELWRI_ELSE_SYNC
:
3154 case XFS_IFLUSH_DELWRI_ELSE_ASYNC
:
3155 case XFS_IFLUSH_DELWRI
:
3158 case XFS_IFLUSH_ASYNC
:
3161 case XFS_IFLUSH_SYNC
:
3172 * First flush out the inode that xfs_iflush was called with.
3174 error
= xfs_iflush_int(ip
, bp
);
3181 * see if other inodes can be gathered into this write
3184 ip
->i_chash
->chl_buf
= bp
;
3186 ch
= XFS_CHASH(mp
, ip
->i_blkno
);
3187 s
= mutex_spinlock(&ch
->ch_lock
);
3190 for (iq
= ip
->i_cnext
; iq
!= ip
; iq
= iq
->i_cnext
) {
3192 * Do an un-protected check to see if the inode is dirty and
3193 * is a candidate for flushing. These checks will be repeated
3194 * later after the appropriate locks are acquired.
3197 if ((iq
->i_update_core
== 0) &&
3199 !(iip
->ili_format
.ilf_fields
& XFS_ILOG_ALL
)) &&
3200 xfs_ipincount(iq
) == 0) {
3205 * Try to get locks. If any are unavailable,
3206 * then this inode cannot be flushed and is skipped.
3209 /* get inode locks (just i_lock) */
3210 if (xfs_ilock_nowait(iq
, XFS_ILOCK_SHARED
)) {
3211 /* get inode flush lock */
3212 if (xfs_iflock_nowait(iq
)) {
3213 /* check if pinned */
3214 if (xfs_ipincount(iq
) == 0) {
3215 /* arriving here means that
3216 * this inode can be flushed.
3217 * first re-check that it's
3221 if ((iq
->i_update_core
!= 0)||
3223 (iip
->ili_format
.ilf_fields
& XFS_ILOG_ALL
))) {
3225 error
= xfs_iflush_int(iq
, bp
);
3229 goto cluster_corrupt_out
;
3238 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
3241 mutex_spinunlock(&ch
->ch_lock
, s
);
3244 XFS_STATS_INC(xs_icluster_flushcnt
);
3245 XFS_STATS_ADD(xs_icluster_flushinode
, clcount
);
3249 * If the buffer is pinned then push on the log so we won't
3250 * get stuck waiting in the write for too long.
3252 if (XFS_BUF_ISPINNED(bp
)){
3253 xfs_log_force(mp
, (xfs_lsn_t
)0, XFS_LOG_FORCE
);
3256 if (flags
& INT_DELWRI
) {
3257 xfs_bdwrite(mp
, bp
);
3258 } else if (flags
& INT_ASYNC
) {
3259 xfs_bawrite(mp
, bp
);
3261 error
= xfs_bwrite(mp
, bp
);
3267 xfs_force_shutdown(mp
, XFS_CORRUPT_INCORE
);
3268 xfs_iflush_abort(ip
);
3270 * Unlocks the flush lock
3272 return XFS_ERROR(EFSCORRUPTED
);
3274 cluster_corrupt_out
:
3275 /* Corruption detected in the clustering loop. Invalidate the
3276 * inode buffer and shut down the filesystem.
3278 mutex_spinunlock(&ch
->ch_lock
, s
);
3281 * Clean up the buffer. If it was B_DELWRI, just release it --
3282 * brelse can handle it with no problems. If not, shut down the
3283 * filesystem before releasing the buffer.
3285 if ((bufwasdelwri
= XFS_BUF_ISDELAYWRITE(bp
))) {
3289 xfs_force_shutdown(mp
, XFS_CORRUPT_INCORE
);
3293 * Just like incore_relse: if we have b_iodone functions,
3294 * mark the buffer as an error and call them. Otherwise
3295 * mark it as stale and brelse.
3297 if (XFS_BUF_IODONE_FUNC(bp
)) {
3298 XFS_BUF_CLR_BDSTRAT_FUNC(bp
);
3302 XFS_BUF_ERROR(bp
,EIO
);
3310 xfs_iflush_abort(iq
);
3312 * Unlocks the flush lock
3314 return XFS_ERROR(EFSCORRUPTED
);
3323 xfs_inode_log_item_t
*iip
;
3326 #ifdef XFS_TRANS_DEBUG
3331 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
|MR_ACCESS
));
3332 ASSERT(valusema(&ip
->i_flock
) <= 0);
3333 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
3334 ip
->i_d
.di_nextents
> ip
->i_df
.if_ext_max
);
3341 * If the inode isn't dirty, then just release the inode
3342 * flush lock and do nothing.
3344 if ((ip
->i_update_core
== 0) &&
3345 ((iip
== NULL
) || !(iip
->ili_format
.ilf_fields
& XFS_ILOG_ALL
))) {
3350 /* set *dip = inode's place in the buffer */
3351 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, ip
->i_boffset
);
3354 * Clear i_update_core before copying out the data.
3355 * This is for coordination with our timestamp updates
3356 * that don't hold the inode lock. They will always
3357 * update the timestamps BEFORE setting i_update_core,
3358 * so if we clear i_update_core after they set it we
3359 * are guaranteed to see their updates to the timestamps.
3360 * I believe that this depends on strongly ordered memory
3361 * semantics, but we have that. We use the SYNCHRONIZE
3362 * macro to make sure that the compiler does not reorder
3363 * the i_update_core access below the data copy below.
3365 ip
->i_update_core
= 0;
3369 * Make sure to get the latest atime from the Linux inode.
3371 xfs_synchronize_atime(ip
);
3373 if (XFS_TEST_ERROR(INT_GET(dip
->di_core
.di_magic
,ARCH_CONVERT
) != XFS_DINODE_MAGIC
,
3374 mp
, XFS_ERRTAG_IFLUSH_1
, XFS_RANDOM_IFLUSH_1
)) {
3375 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3376 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3377 ip
->i_ino
, (int) INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
), dip
);
3380 if (XFS_TEST_ERROR(ip
->i_d
.di_magic
!= XFS_DINODE_MAGIC
,
3381 mp
, XFS_ERRTAG_IFLUSH_2
, XFS_RANDOM_IFLUSH_2
)) {
3382 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3383 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3384 ip
->i_ino
, ip
, ip
->i_d
.di_magic
);
3387 if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
) {
3389 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3390 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
),
3391 mp
, XFS_ERRTAG_IFLUSH_3
, XFS_RANDOM_IFLUSH_3
)) {
3392 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3393 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3397 } else if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFDIR
) {
3399 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3400 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
) &&
3401 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_LOCAL
),
3402 mp
, XFS_ERRTAG_IFLUSH_4
, XFS_RANDOM_IFLUSH_4
)) {
3403 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3404 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3409 if (XFS_TEST_ERROR(ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
>
3410 ip
->i_d
.di_nblocks
, mp
, XFS_ERRTAG_IFLUSH_5
,
3411 XFS_RANDOM_IFLUSH_5
)) {
3412 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3413 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3415 ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
,
3420 if (XFS_TEST_ERROR(ip
->i_d
.di_forkoff
> mp
->m_sb
.sb_inodesize
,
3421 mp
, XFS_ERRTAG_IFLUSH_6
, XFS_RANDOM_IFLUSH_6
)) {
3422 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3423 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3424 ip
->i_ino
, ip
->i_d
.di_forkoff
, ip
);
3428 * bump the flush iteration count, used to detect flushes which
3429 * postdate a log record during recovery.
3432 ip
->i_d
.di_flushiter
++;
3435 * Copy the dirty parts of the inode into the on-disk
3436 * inode. We always copy out the core of the inode,
3437 * because if the inode is dirty at all the core must
3440 xfs_xlate_dinode_core((xfs_caddr_t
)&(dip
->di_core
), &(ip
->i_d
), -1);
3442 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3443 if (ip
->i_d
.di_flushiter
== DI_MAX_FLUSH
)
3444 ip
->i_d
.di_flushiter
= 0;
3447 * If this is really an old format inode and the superblock version
3448 * has not been updated to support only new format inodes, then
3449 * convert back to the old inode format. If the superblock version
3450 * has been updated, then make the conversion permanent.
3452 ASSERT(ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
||
3453 XFS_SB_VERSION_HASNLINK(&mp
->m_sb
));
3454 if (ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
) {
3455 if (!XFS_SB_VERSION_HASNLINK(&mp
->m_sb
)) {
3459 ASSERT(ip
->i_d
.di_nlink
<= XFS_MAXLINK_1
);
3460 INT_SET(dip
->di_core
.di_onlink
, ARCH_CONVERT
, ip
->i_d
.di_nlink
);
3463 * The superblock version has already been bumped,
3464 * so just make the conversion to the new inode
3467 ip
->i_d
.di_version
= XFS_DINODE_VERSION_2
;
3468 INT_SET(dip
->di_core
.di_version
, ARCH_CONVERT
, XFS_DINODE_VERSION_2
);
3469 ip
->i_d
.di_onlink
= 0;
3470 dip
->di_core
.di_onlink
= 0;
3471 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
3472 memset(&(dip
->di_core
.di_pad
[0]), 0,
3473 sizeof(dip
->di_core
.di_pad
));
3474 ASSERT(ip
->i_d
.di_projid
== 0);
3478 if (xfs_iflush_fork(ip
, dip
, iip
, XFS_DATA_FORK
, bp
) == EFSCORRUPTED
) {
3482 if (XFS_IFORK_Q(ip
)) {
3484 * The only error from xfs_iflush_fork is on the data fork.
3486 (void) xfs_iflush_fork(ip
, dip
, iip
, XFS_ATTR_FORK
, bp
);
3488 xfs_inobp_check(mp
, bp
);
3491 * We've recorded everything logged in the inode, so we'd
3492 * like to clear the ilf_fields bits so we don't log and
3493 * flush things unnecessarily. However, we can't stop
3494 * logging all this information until the data we've copied
3495 * into the disk buffer is written to disk. If we did we might
3496 * overwrite the copy of the inode in the log with all the
3497 * data after re-logging only part of it, and in the face of
3498 * a crash we wouldn't have all the data we need to recover.
3500 * What we do is move the bits to the ili_last_fields field.
3501 * When logging the inode, these bits are moved back to the
3502 * ilf_fields field. In the xfs_iflush_done() routine we
3503 * clear ili_last_fields, since we know that the information
3504 * those bits represent is permanently on disk. As long as
3505 * the flush completes before the inode is logged again, then
3506 * both ilf_fields and ili_last_fields will be cleared.
3508 * We can play with the ilf_fields bits here, because the inode
3509 * lock must be held exclusively in order to set bits there
3510 * and the flush lock protects the ili_last_fields bits.
3511 * Set ili_logged so the flush done
3512 * routine can tell whether or not to look in the AIL.
3513 * Also, store the current LSN of the inode so that we can tell
3514 * whether the item has moved in the AIL from xfs_iflush_done().
3515 * In order to read the lsn we need the AIL lock, because
3516 * it is a 64 bit value that cannot be read atomically.
3518 if (iip
!= NULL
&& iip
->ili_format
.ilf_fields
!= 0) {
3519 iip
->ili_last_fields
= iip
->ili_format
.ilf_fields
;
3520 iip
->ili_format
.ilf_fields
= 0;
3521 iip
->ili_logged
= 1;
3523 ASSERT(sizeof(xfs_lsn_t
) == 8); /* don't lock if it shrinks */
3525 iip
->ili_flush_lsn
= iip
->ili_item
.li_lsn
;
3529 * Attach the function xfs_iflush_done to the inode's
3530 * buffer. This will remove the inode from the AIL
3531 * and unlock the inode's flush lock when the inode is
3532 * completely written to disk.
3534 xfs_buf_attach_iodone(bp
, (void(*)(xfs_buf_t
*,xfs_log_item_t
*))
3535 xfs_iflush_done
, (xfs_log_item_t
*)iip
);
3537 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) != NULL
);
3538 ASSERT(XFS_BUF_IODONE_FUNC(bp
) != NULL
);
3541 * We're flushing an inode which is not in the AIL and has
3542 * not been logged but has i_update_core set. For this
3543 * case we can use a B_DELWRI flush and immediately drop
3544 * the inode flush lock because we can avoid the whole
3545 * AIL state thing. It's OK to drop the flush lock now,
3546 * because we've already locked the buffer and to do anything
3547 * you really need both.
3550 ASSERT(iip
->ili_logged
== 0);
3551 ASSERT(iip
->ili_last_fields
== 0);
3552 ASSERT((iip
->ili_item
.li_flags
& XFS_LI_IN_AIL
) == 0);
3560 return XFS_ERROR(EFSCORRUPTED
);
3565 * Flush all inactive inodes in mp.
3575 XFS_MOUNT_ILOCK(mp
);
3581 /* Make sure we skip markers inserted by sync */
3582 if (ip
->i_mount
== NULL
) {
3587 vp
= XFS_ITOV_NULL(ip
);
3589 XFS_MOUNT_IUNLOCK(mp
);
3590 xfs_finish_reclaim(ip
, 0, XFS_IFLUSH_ASYNC
);
3594 ASSERT(vn_count(vp
) == 0);
3597 } while (ip
!= mp
->m_inodes
);
3599 XFS_MOUNT_IUNLOCK(mp
);
3603 * xfs_iaccess: check accessibility of inode for mode.
3612 mode_t orgmode
= mode
;
3613 struct inode
*inode
= LINVFS_GET_IP(XFS_ITOV(ip
));
3615 if (mode
& S_IWUSR
) {
3616 umode_t imode
= inode
->i_mode
;
3618 if (IS_RDONLY(inode
) &&
3619 (S_ISREG(imode
) || S_ISDIR(imode
) || S_ISLNK(imode
)))
3620 return XFS_ERROR(EROFS
);
3622 if (IS_IMMUTABLE(inode
))
3623 return XFS_ERROR(EACCES
);
3627 * If there's an Access Control List it's used instead of
3630 if ((error
= _ACL_XFS_IACCESS(ip
, mode
, cr
)) != -1)
3631 return error
? XFS_ERROR(error
) : 0;
3633 if (current_fsuid(cr
) != ip
->i_d
.di_uid
) {
3635 if (!in_group_p((gid_t
)ip
->i_d
.di_gid
))
3640 * If the DACs are ok we don't need any capability check.
3642 if ((ip
->i_d
.di_mode
& mode
) == mode
)
3645 * Read/write DACs are always overridable.
3646 * Executable DACs are overridable if at least one exec bit is set.
3648 if (!(orgmode
& S_IXUSR
) ||
3649 (inode
->i_mode
& S_IXUGO
) || S_ISDIR(inode
->i_mode
))
3650 if (capable_cred(cr
, CAP_DAC_OVERRIDE
))
3653 if ((orgmode
== S_IRUSR
) ||
3654 (S_ISDIR(inode
->i_mode
) && (!(orgmode
& S_IWUSR
)))) {
3655 if (capable_cred(cr
, CAP_DAC_READ_SEARCH
))
3658 cmn_err(CE_NOTE
, "Ick: mode=%o, orgmode=%o", mode
, orgmode
);
3660 return XFS_ERROR(EACCES
);
3662 return XFS_ERROR(EACCES
);
3666 * xfs_iroundup: round up argument to next power of two
3675 if ((v
& (v
- 1)) == 0)
3677 ASSERT((v
& 0x80000000) == 0);
3678 if ((v
& (v
+ 1)) == 0)
3680 for (i
= 0, m
= 1; i
< 31; i
++, m
<<= 1) {
3684 if ((v
& (v
+ 1)) == 0)
3691 #ifdef XFS_ILOCK_TRACE
3692 ktrace_t
*xfs_ilock_trace_buf
;
3695 xfs_ilock_trace(xfs_inode_t
*ip
, int lock
, unsigned int lockflags
, inst_t
*ra
)
3697 ktrace_enter(ip
->i_lock_trace
,
3699 (void *)(unsigned long)lock
, /* 1 = LOCK, 3=UNLOCK, etc */
3700 (void *)(unsigned long)lockflags
, /* XFS_ILOCK_EXCL etc */
3701 (void *)ra
, /* caller of ilock */
3702 (void *)(unsigned long)current_cpu(),
3703 (void *)(unsigned long)current_pid(),
3704 NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
);