2 * Copyright (c) 2000-2003 Silicon Graphics, Inc. All Rights Reserved.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it would be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
12 * Further, this software is distributed without any warranty that it is
13 * free of the rightful claim of any third person regarding infringement
14 * or the like. Any license provided herein, whether implied or
15 * otherwise, applies only to this software file. Patent licenses, if
16 * any, provided herein do not apply to combinations of this program with
17 * other software, or any other product whatsoever.
19 * You should have received a copy of the GNU General Public License along
20 * with this program; if not, write the Free Software Foundation, Inc., 59
21 * Temple Place - Suite 330, Boston MA 02111-1307, USA.
23 * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24 * Mountain View, CA 94043, or:
28 * For further information regarding this notice, see:
30 * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
34 #include "xfs_macros.h"
35 #include "xfs_types.h"
38 #include "xfs_trans.h"
39 #include "xfs_trans_priv.h"
44 #include "xfs_dmapi.h"
45 #include "xfs_mount.h"
46 #include "xfs_alloc_btree.h"
47 #include "xfs_bmap_btree.h"
48 #include "xfs_ialloc_btree.h"
49 #include "xfs_btree.h"
51 #include "xfs_alloc.h"
52 #include "xfs_ialloc.h"
53 #include "xfs_attr_sf.h"
54 #include "xfs_dir_sf.h"
55 #include "xfs_dir2_sf.h"
56 #include "xfs_dinode.h"
57 #include "xfs_inode_item.h"
58 #include "xfs_inode.h"
60 #include "xfs_buf_item.h"
62 #include "xfs_error.h"
64 #include "xfs_utils.h"
65 #include "xfs_dir2_trace.h"
66 #include "xfs_quota.h"
71 kmem_zone_t
*xfs_ifork_zone
;
72 kmem_zone_t
*xfs_inode_zone
;
73 kmem_zone_t
*xfs_chashlist_zone
;
76 * Used in xfs_itruncate(). This is the maximum number of extents
77 * freed from a file in a single transaction.
79 #define XFS_ITRUNC_MAX_EXTENTS 2
81 STATIC
int xfs_iflush_int(xfs_inode_t
*, xfs_buf_t
*);
82 STATIC
int xfs_iformat_local(xfs_inode_t
*, xfs_dinode_t
*, int, int);
83 STATIC
int xfs_iformat_extents(xfs_inode_t
*, xfs_dinode_t
*, int);
84 STATIC
int xfs_iformat_btree(xfs_inode_t
*, xfs_dinode_t
*, int);
89 * Make sure that the extents in the given memory buffer
103 for (i
= 0; i
< nrecs
; i
++) {
104 rec
.l0
= get_unaligned((__uint64_t
*)&ep
->l0
);
105 rec
.l1
= get_unaligned((__uint64_t
*)&ep
->l1
);
107 xfs_bmbt_disk_get_all(&rec
, &irec
);
109 xfs_bmbt_get_all(&rec
, &irec
);
110 if (fmt
== XFS_EXTFMT_NOSTATE
)
111 ASSERT(irec
.br_state
== XFS_EXT_NORM
);
116 #define xfs_validate_extents(ep, nrecs, disk, fmt)
120 * Check that none of the inode's in the buffer have a next
121 * unlinked field of 0.
133 j
= mp
->m_inode_cluster_size
>> mp
->m_sb
.sb_inodelog
;
135 for (i
= 0; i
< j
; i
++) {
136 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
,
137 i
* mp
->m_sb
.sb_inodesize
);
138 if (INT_ISZERO(dip
->di_next_unlinked
, ARCH_CONVERT
)) {
139 xfs_fs_cmn_err(CE_ALERT
, mp
,
140 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
142 ASSERT(!INT_ISZERO(dip
->di_next_unlinked
, ARCH_CONVERT
));
149 * called from bwrite on xfs inode buffers
152 xfs_inobp_bwcheck(xfs_buf_t
*bp
)
159 ASSERT(XFS_BUF_FSPRIVATE3(bp
, void *) != NULL
);
161 mp
= XFS_BUF_FSPRIVATE3(bp
, xfs_mount_t
*);
164 j
= mp
->m_inode_cluster_size
>> mp
->m_sb
.sb_inodelog
;
166 for (i
= 0; i
< j
; i
++) {
167 dip
= (xfs_dinode_t
*) xfs_buf_offset(bp
,
168 i
* mp
->m_sb
.sb_inodesize
);
169 if (INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
) != XFS_DINODE_MAGIC
) {
171 "Bad magic # 0x%x in XFS inode buffer 0x%Lx, starting blockno %Ld, offset 0x%x",
172 INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
),
173 (__uint64_t
)(__psunsigned_t
) bp
,
174 (__int64_t
) XFS_BUF_ADDR(bp
),
175 xfs_buf_offset(bp
, i
* mp
->m_sb
.sb_inodesize
));
176 xfs_fs_cmn_err(CE_WARN
, mp
,
177 "corrupt, unmount and run xfs_repair");
179 if (INT_ISZERO(dip
->di_next_unlinked
, ARCH_CONVERT
)) {
181 "Bad next_unlinked field (0) in XFS inode buffer 0x%p, starting blockno %Ld, offset 0x%x",
182 (__uint64_t
)(__psunsigned_t
) bp
,
183 (__int64_t
) XFS_BUF_ADDR(bp
),
184 xfs_buf_offset(bp
, i
* mp
->m_sb
.sb_inodesize
));
185 xfs_fs_cmn_err(CE_WARN
, mp
,
186 "corrupt, unmount and run xfs_repair");
194 * This routine is called to map an inode number within a file
195 * system to the buffer containing the on-disk version of the
196 * inode. It returns a pointer to the buffer containing the
197 * on-disk inode in the bpp parameter, and in the dip parameter
198 * it returns a pointer to the on-disk inode within that buffer.
200 * If a non-zero error is returned, then the contents of bpp and
201 * dipp are undefined.
203 * Use xfs_imap() to determine the size and location of the
204 * buffer to read from disk.
222 * Call the space managment code to find the location of the
226 error
= xfs_imap(mp
, tp
, ino
, &imap
, XFS_IMAP_LOOKUP
);
229 "xfs_inotobp: xfs_imap() returned an "
230 "error %d on %s. Returning error.", error
, mp
->m_fsname
);
235 * If the inode number maps to a block outside the bounds of the
236 * file system then return NULL rather than calling read_buf
237 * and panicing when we get an error from the driver.
239 if ((imap
.im_blkno
+ imap
.im_len
) >
240 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
)) {
242 "xfs_inotobp: inode number (%d + %d) maps to a block outside the bounds "
243 "of the file system %s. Returning EINVAL.",
244 imap
.im_blkno
, imap
.im_len
,mp
->m_fsname
);
245 return XFS_ERROR(EINVAL
);
249 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
250 * default to just a read_buf() call.
252 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, imap
.im_blkno
,
253 (int)imap
.im_len
, XFS_BUF_LOCK
, &bp
);
257 "xfs_inotobp: xfs_trans_read_buf() returned an "
258 "error %d on %s. Returning error.", error
, mp
->m_fsname
);
261 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, 0);
263 INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
) == XFS_DINODE_MAGIC
&&
264 XFS_DINODE_GOOD_VERSION(INT_GET(dip
->di_core
.di_version
, ARCH_CONVERT
));
265 if (unlikely(XFS_TEST_ERROR(!di_ok
, mp
, XFS_ERRTAG_ITOBP_INOTOBP
,
266 XFS_RANDOM_ITOBP_INOTOBP
))) {
267 XFS_CORRUPTION_ERROR("xfs_inotobp", XFS_ERRLEVEL_LOW
, mp
, dip
);
268 xfs_trans_brelse(tp
, bp
);
270 "xfs_inotobp: XFS_TEST_ERROR() returned an "
271 "error on %s. Returning EFSCORRUPTED.", mp
->m_fsname
);
272 return XFS_ERROR(EFSCORRUPTED
);
275 xfs_inobp_check(mp
, bp
);
278 * Set *dipp to point to the on-disk inode in the buffer.
280 *dipp
= (xfs_dinode_t
*)xfs_buf_offset(bp
, imap
.im_boffset
);
282 *offset
= imap
.im_boffset
;
288 * This routine is called to map an inode to the buffer containing
289 * the on-disk version of the inode. It returns a pointer to the
290 * buffer containing the on-disk inode in the bpp parameter, and in
291 * the dip parameter it returns a pointer to the on-disk inode within
294 * If a non-zero error is returned, then the contents of bpp and
295 * dipp are undefined.
297 * If the inode is new and has not yet been initialized, use xfs_imap()
298 * to determine the size and location of the buffer to read from disk.
299 * If the inode has already been mapped to its buffer and read in once,
300 * then use the mapping information stored in the inode rather than
301 * calling xfs_imap(). This allows us to avoid the overhead of looking
302 * at the inode btree for small block file systems (see xfs_dilocate()).
303 * We can tell whether the inode has been mapped in before by comparing
304 * its disk block address to 0. Only uninitialized inodes will have
305 * 0 for the disk block address.
324 if (ip
->i_blkno
== (xfs_daddr_t
)0) {
326 * Call the space management code to find the location of the
330 error
= xfs_imap(mp
, tp
, ip
->i_ino
, &imap
, XFS_IMAP_LOOKUP
);
336 * If the inode number maps to a block outside the bounds
337 * of the file system then return NULL rather than calling
338 * read_buf and panicing when we get an error from the
341 if ((imap
.im_blkno
+ imap
.im_len
) >
342 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
)) {
344 xfs_fs_cmn_err(CE_ALERT
, mp
, "xfs_itobp: "
345 "(imap.im_blkno (0x%llx) "
346 "+ imap.im_len (0x%llx)) > "
347 " XFS_FSB_TO_BB(mp, "
348 "mp->m_sb.sb_dblocks) (0x%llx)",
349 (unsigned long long) imap
.im_blkno
,
350 (unsigned long long) imap
.im_len
,
351 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
));
353 return XFS_ERROR(EINVAL
);
357 * Fill in the fields in the inode that will be used to
358 * map the inode to its buffer from now on.
360 ip
->i_blkno
= imap
.im_blkno
;
361 ip
->i_len
= imap
.im_len
;
362 ip
->i_boffset
= imap
.im_boffset
;
365 * We've already mapped the inode once, so just use the
366 * mapping that we saved the first time.
368 imap
.im_blkno
= ip
->i_blkno
;
369 imap
.im_len
= ip
->i_len
;
370 imap
.im_boffset
= ip
->i_boffset
;
372 ASSERT(bno
== 0 || bno
== imap
.im_blkno
);
375 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
376 * default to just a read_buf() call.
378 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, imap
.im_blkno
,
379 (int)imap
.im_len
, XFS_BUF_LOCK
, &bp
);
383 xfs_fs_cmn_err(CE_ALERT
, mp
, "xfs_itobp: "
384 "xfs_trans_read_buf() returned error %d, "
385 "imap.im_blkno 0x%llx, imap.im_len 0x%llx",
386 error
, (unsigned long long) imap
.im_blkno
,
387 (unsigned long long) imap
.im_len
);
393 * Validate the magic number and version of every inode in the buffer
394 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
397 ni
= BBTOB(imap
.im_len
) >> mp
->m_sb
.sb_inodelog
;
401 for (i
= 0; i
< ni
; i
++) {
405 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
,
406 (i
<< mp
->m_sb
.sb_inodelog
));
407 di_ok
= INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
) == XFS_DINODE_MAGIC
&&
408 XFS_DINODE_GOOD_VERSION(INT_GET(dip
->di_core
.di_version
, ARCH_CONVERT
));
409 if (unlikely(XFS_TEST_ERROR(!di_ok
, mp
, XFS_ERRTAG_ITOBP_INOTOBP
,
410 XFS_RANDOM_ITOBP_INOTOBP
))) {
412 prdev("bad inode magic/vsn daddr %lld #%d (magic=%x)",
414 (unsigned long long)imap
.im_blkno
, i
,
415 INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
));
417 XFS_CORRUPTION_ERROR("xfs_itobp", XFS_ERRLEVEL_HIGH
,
419 xfs_trans_brelse(tp
, bp
);
420 return XFS_ERROR(EFSCORRUPTED
);
423 #endif /* __KERNEL__ */
425 xfs_inobp_check(mp
, bp
);
428 * Mark the buffer as an inode buffer now that it looks good
430 XFS_BUF_SET_VTYPE(bp
, B_FS_INO
);
433 * Set *dipp to point to the on-disk inode in the buffer.
435 *dipp
= (xfs_dinode_t
*)xfs_buf_offset(bp
, imap
.im_boffset
);
441 * Move inode type and inode format specific information from the
442 * on-disk inode to the in-core inode. For fifos, devs, and sockets
443 * this means set if_rdev to the proper value. For files, directories,
444 * and symlinks this means to bring in the in-line data or extent
445 * pointers. For a file in B-tree format, only the root is immediately
446 * brought in-core. The rest will be in-lined in if_extents when it
447 * is first referenced (see xfs_iread_extents()).
454 xfs_attr_shortform_t
*atp
;
458 ip
->i_df
.if_ext_max
=
459 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
463 INT_GET(dip
->di_core
.di_nextents
, ARCH_CONVERT
) +
464 INT_GET(dip
->di_core
.di_anextents
, ARCH_CONVERT
) >
465 INT_GET(dip
->di_core
.di_nblocks
, ARCH_CONVERT
))) {
466 xfs_fs_cmn_err(CE_WARN
, ip
->i_mount
,
467 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu."
468 " Unmount and run xfs_repair.",
469 (unsigned long long)ip
->i_ino
,
470 (int)(INT_GET(dip
->di_core
.di_nextents
, ARCH_CONVERT
)
471 + INT_GET(dip
->di_core
.di_anextents
, ARCH_CONVERT
)),
473 INT_GET(dip
->di_core
.di_nblocks
, ARCH_CONVERT
));
474 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW
,
476 return XFS_ERROR(EFSCORRUPTED
);
479 if (unlikely(INT_GET(dip
->di_core
.di_forkoff
, ARCH_CONVERT
) > ip
->i_mount
->m_sb
.sb_inodesize
)) {
480 xfs_fs_cmn_err(CE_WARN
, ip
->i_mount
,
481 "corrupt dinode %Lu, forkoff = 0x%x."
482 " Unmount and run xfs_repair.",
483 (unsigned long long)ip
->i_ino
,
484 (int)(INT_GET(dip
->di_core
.di_forkoff
, ARCH_CONVERT
)));
485 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW
,
487 return XFS_ERROR(EFSCORRUPTED
);
490 switch (ip
->i_d
.di_mode
& S_IFMT
) {
495 if (unlikely(INT_GET(dip
->di_core
.di_format
, ARCH_CONVERT
) != XFS_DINODE_FMT_DEV
)) {
496 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW
,
498 return XFS_ERROR(EFSCORRUPTED
);
501 ip
->i_df
.if_u2
.if_rdev
= INT_GET(dip
->di_u
.di_dev
, ARCH_CONVERT
);
507 switch (INT_GET(dip
->di_core
.di_format
, ARCH_CONVERT
)) {
508 case XFS_DINODE_FMT_LOCAL
:
510 * no local regular files yet
512 if (unlikely((INT_GET(dip
->di_core
.di_mode
, ARCH_CONVERT
) & S_IFMT
) == S_IFREG
)) {
513 xfs_fs_cmn_err(CE_WARN
, ip
->i_mount
,
514 "corrupt inode (local format for regular file) %Lu. Unmount and run xfs_repair.",
515 (unsigned long long) ip
->i_ino
);
516 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
519 return XFS_ERROR(EFSCORRUPTED
);
522 di_size
= INT_GET(dip
->di_core
.di_size
, ARCH_CONVERT
);
523 if (unlikely(di_size
>
524 XFS_DFORK_DSIZE_ARCH(dip
, ip
->i_mount
, ARCH_CONVERT
))) {
525 xfs_fs_cmn_err(CE_WARN
, ip
->i_mount
,
526 "corrupt inode %Lu (bad size %Ld for local inode). Unmount and run xfs_repair.",
527 (unsigned long long) ip
->i_ino
,
528 (long long) di_size
);
529 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
532 return XFS_ERROR(EFSCORRUPTED
);
536 error
= xfs_iformat_local(ip
, dip
, XFS_DATA_FORK
, size
);
538 case XFS_DINODE_FMT_EXTENTS
:
539 error
= xfs_iformat_extents(ip
, dip
, XFS_DATA_FORK
);
541 case XFS_DINODE_FMT_BTREE
:
542 error
= xfs_iformat_btree(ip
, dip
, XFS_DATA_FORK
);
545 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW
,
547 return XFS_ERROR(EFSCORRUPTED
);
552 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW
, ip
->i_mount
);
553 return XFS_ERROR(EFSCORRUPTED
);
558 if (!XFS_DFORK_Q_ARCH(dip
, ARCH_CONVERT
))
560 ASSERT(ip
->i_afp
== NULL
);
561 ip
->i_afp
= kmem_zone_zalloc(xfs_ifork_zone
, KM_SLEEP
);
562 ip
->i_afp
->if_ext_max
=
563 XFS_IFORK_ASIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
564 switch (INT_GET(dip
->di_core
.di_aformat
, ARCH_CONVERT
)) {
565 case XFS_DINODE_FMT_LOCAL
:
566 atp
= (xfs_attr_shortform_t
*)XFS_DFORK_APTR_ARCH(dip
, ARCH_CONVERT
);
567 size
= (int)INT_GET(atp
->hdr
.totsize
, ARCH_CONVERT
);
568 error
= xfs_iformat_local(ip
, dip
, XFS_ATTR_FORK
, size
);
570 case XFS_DINODE_FMT_EXTENTS
:
571 error
= xfs_iformat_extents(ip
, dip
, XFS_ATTR_FORK
);
573 case XFS_DINODE_FMT_BTREE
:
574 error
= xfs_iformat_btree(ip
, dip
, XFS_ATTR_FORK
);
577 error
= XFS_ERROR(EFSCORRUPTED
);
581 kmem_zone_free(xfs_ifork_zone
, ip
->i_afp
);
583 xfs_idestroy_fork(ip
, XFS_DATA_FORK
);
589 * The file is in-lined in the on-disk inode.
590 * If it fits into if_inline_data, then copy
591 * it there, otherwise allocate a buffer for it
592 * and copy the data there. Either way, set
593 * if_data to point at the data.
594 * If we allocate a buffer for the data, make
595 * sure that its size is a multiple of 4 and
596 * record the real size in i_real_bytes.
609 * If the size 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
> XFS_DFORK_SIZE_ARCH(dip
, ip
->i_mount
, whichfork
, ARCH_CONVERT
))) {
614 xfs_fs_cmn_err(CE_WARN
, ip
->i_mount
,
615 "corrupt inode %Lu (bad size %d for local fork, size = %d). Unmount and run xfs_repair.",
616 (unsigned long long) ip
->i_ino
, size
,
617 XFS_DFORK_SIZE_ARCH(dip
, ip
->i_mount
, whichfork
, ARCH_CONVERT
));
618 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW
,
620 return XFS_ERROR(EFSCORRUPTED
);
622 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
625 ifp
->if_u1
.if_data
= NULL
;
626 else if (size
<= sizeof(ifp
->if_u2
.if_inline_data
))
627 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
629 real_size
= roundup(size
, 4);
630 ifp
->if_u1
.if_data
= kmem_alloc(real_size
, KM_SLEEP
);
632 ifp
->if_bytes
= size
;
633 ifp
->if_real_bytes
= real_size
;
635 memcpy(ifp
->if_u1
.if_data
,
636 XFS_DFORK_PTR_ARCH(dip
, whichfork
, ARCH_CONVERT
), size
);
637 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
638 ifp
->if_flags
|= XFS_IFINLINE
;
643 * The file consists of a set of extents all
644 * of which fit into the on-disk inode.
645 * If there are few enough extents to fit into
646 * the if_inline_ext, then copy them there.
647 * Otherwise allocate a buffer for them and copy
648 * them into it. Either way, set if_extents
649 * to point at the extents.
657 xfs_bmbt_rec_t
*ep
, *dp
;
664 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
665 nex
= XFS_DFORK_NEXTENTS_ARCH(dip
, whichfork
, ARCH_CONVERT
);
666 size
= nex
* (uint
)sizeof(xfs_bmbt_rec_t
);
669 * If the number of extents is unreasonable, then something
670 * is wrong and we just bail out rather than crash in
671 * kmem_alloc() or memcpy() below.
673 if (unlikely(size
< 0 || size
> XFS_DFORK_SIZE_ARCH(dip
, ip
->i_mount
, whichfork
, ARCH_CONVERT
))) {
674 xfs_fs_cmn_err(CE_WARN
, ip
->i_mount
,
675 "corrupt inode %Lu ((a)extents = %d). Unmount and run xfs_repair.",
676 (unsigned long long) ip
->i_ino
, nex
);
677 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW
,
679 return XFS_ERROR(EFSCORRUPTED
);
684 ifp
->if_u1
.if_extents
= NULL
;
685 else if (nex
<= XFS_INLINE_EXTS
)
686 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
688 ifp
->if_u1
.if_extents
= kmem_alloc(size
, KM_SLEEP
);
689 ASSERT(ifp
->if_u1
.if_extents
!= NULL
);
692 ifp
->if_bytes
= size
;
693 ifp
->if_real_bytes
= real_size
;
695 dp
= (xfs_bmbt_rec_t
*)
696 XFS_DFORK_PTR_ARCH(dip
, whichfork
, ARCH_CONVERT
);
697 xfs_validate_extents(dp
, nex
, 1, XFS_EXTFMT_INODE(ip
));
698 ep
= ifp
->if_u1
.if_extents
;
699 for (i
= 0; i
< nex
; i
++, ep
++, dp
++) {
700 ep
->l0
= INT_GET(get_unaligned((__uint64_t
*)&dp
->l0
),
702 ep
->l1
= INT_GET(get_unaligned((__uint64_t
*)&dp
->l1
),
705 xfs_bmap_trace_exlist("xfs_iformat_extents", ip
, nex
,
707 if (whichfork
!= XFS_DATA_FORK
||
708 XFS_EXTFMT_INODE(ip
) == XFS_EXTFMT_NOSTATE
)
709 if (unlikely(xfs_check_nostate_extents(
710 ifp
->if_u1
.if_extents
, nex
))) {
711 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
714 return XFS_ERROR(EFSCORRUPTED
);
717 ifp
->if_flags
|= XFS_IFEXTENTS
;
722 * The file has too many extents to fit into
723 * the inode, so they are in B-tree format.
724 * Allocate a buffer for the root of the B-tree
725 * and copy the root into it. The i_extents
726 * field will remain NULL until all of the
727 * extents are read in (when they are needed).
735 xfs_bmdr_block_t
*dfp
;
741 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
742 dfp
= (xfs_bmdr_block_t
*)XFS_DFORK_PTR_ARCH(dip
, whichfork
, ARCH_CONVERT
);
743 size
= XFS_BMAP_BROOT_SPACE(dfp
);
744 nrecs
= XFS_BMAP_BROOT_NUMRECS(dfp
);
747 * blow out if -- fork has less extents than can fit in
748 * fork (fork shouldn't be a btree format), root btree
749 * block has more records than can fit into the fork,
750 * or the number of extents is greater than the number of
753 if (unlikely(XFS_IFORK_NEXTENTS(ip
, whichfork
) <= ifp
->if_ext_max
754 || XFS_BMDR_SPACE_CALC(nrecs
) >
755 XFS_DFORK_SIZE_ARCH(dip
, ip
->i_mount
, whichfork
, ARCH_CONVERT
)
756 || XFS_IFORK_NEXTENTS(ip
, whichfork
) > ip
->i_d
.di_nblocks
)) {
757 xfs_fs_cmn_err(CE_WARN
, ip
->i_mount
,
758 "corrupt inode %Lu (btree). Unmount and run xfs_repair.",
759 (unsigned long long) ip
->i_ino
);
760 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW
,
762 return XFS_ERROR(EFSCORRUPTED
);
765 ifp
->if_broot_bytes
= size
;
766 ifp
->if_broot
= kmem_alloc(size
, KM_SLEEP
);
767 ASSERT(ifp
->if_broot
!= NULL
);
769 * Copy and convert from the on-disk structure
770 * to the in-memory structure.
772 xfs_bmdr_to_bmbt(dfp
, XFS_DFORK_SIZE_ARCH(dip
, ip
->i_mount
, whichfork
, ARCH_CONVERT
),
773 ifp
->if_broot
, size
);
774 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
775 ifp
->if_flags
|= XFS_IFBROOT
;
781 * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk
784 * buf = on-disk representation
785 * dip = native representation
786 * dir = direction - +ve -> disk to native
787 * -ve -> native to disk
788 * arch = on-disk architecture
791 xfs_xlate_dinode_core(
793 xfs_dinode_core_t
*dip
,
797 xfs_dinode_core_t
*buf_core
= (xfs_dinode_core_t
*)buf
;
798 xfs_dinode_core_t
*mem_core
= (xfs_dinode_core_t
*)dip
;
801 if (arch
== ARCH_NOCONVERT
) {
803 memcpy((xfs_caddr_t
)mem_core
, (xfs_caddr_t
)buf_core
,
804 sizeof(xfs_dinode_core_t
));
806 memcpy((xfs_caddr_t
)buf_core
, (xfs_caddr_t
)mem_core
,
807 sizeof(xfs_dinode_core_t
));
812 INT_XLATE(buf_core
->di_magic
, mem_core
->di_magic
, dir
, arch
);
813 INT_XLATE(buf_core
->di_mode
, mem_core
->di_mode
, dir
, arch
);
814 INT_XLATE(buf_core
->di_version
, mem_core
->di_version
, dir
, arch
);
815 INT_XLATE(buf_core
->di_format
, mem_core
->di_format
, dir
, arch
);
816 INT_XLATE(buf_core
->di_onlink
, mem_core
->di_onlink
, dir
, arch
);
817 INT_XLATE(buf_core
->di_uid
, mem_core
->di_uid
, dir
, arch
);
818 INT_XLATE(buf_core
->di_gid
, mem_core
->di_gid
, dir
, arch
);
819 INT_XLATE(buf_core
->di_nlink
, mem_core
->di_nlink
, dir
, arch
);
820 INT_XLATE(buf_core
->di_projid
, mem_core
->di_projid
, dir
, arch
);
823 memcpy(mem_core
->di_pad
, buf_core
->di_pad
,
824 sizeof(buf_core
->di_pad
));
826 memcpy(buf_core
->di_pad
, mem_core
->di_pad
,
827 sizeof(buf_core
->di_pad
));
830 INT_XLATE(buf_core
->di_flushiter
, mem_core
->di_flushiter
, dir
, arch
);
832 INT_XLATE(buf_core
->di_atime
.t_sec
, mem_core
->di_atime
.t_sec
,
834 INT_XLATE(buf_core
->di_atime
.t_nsec
, mem_core
->di_atime
.t_nsec
,
836 INT_XLATE(buf_core
->di_mtime
.t_sec
, mem_core
->di_mtime
.t_sec
,
838 INT_XLATE(buf_core
->di_mtime
.t_nsec
, mem_core
->di_mtime
.t_nsec
,
840 INT_XLATE(buf_core
->di_ctime
.t_sec
, mem_core
->di_ctime
.t_sec
,
842 INT_XLATE(buf_core
->di_ctime
.t_nsec
, mem_core
->di_ctime
.t_nsec
,
844 INT_XLATE(buf_core
->di_size
, mem_core
->di_size
, dir
, arch
);
845 INT_XLATE(buf_core
->di_nblocks
, mem_core
->di_nblocks
, dir
, arch
);
846 INT_XLATE(buf_core
->di_extsize
, mem_core
->di_extsize
, dir
, arch
);
847 INT_XLATE(buf_core
->di_nextents
, mem_core
->di_nextents
, dir
, arch
);
848 INT_XLATE(buf_core
->di_anextents
, mem_core
->di_anextents
, dir
, arch
);
849 INT_XLATE(buf_core
->di_forkoff
, mem_core
->di_forkoff
, dir
, arch
);
850 INT_XLATE(buf_core
->di_aformat
, mem_core
->di_aformat
, dir
, arch
);
851 INT_XLATE(buf_core
->di_dmevmask
, mem_core
->di_dmevmask
, dir
, arch
);
852 INT_XLATE(buf_core
->di_dmstate
, mem_core
->di_dmstate
, dir
, arch
);
853 INT_XLATE(buf_core
->di_flags
, mem_core
->di_flags
, dir
, arch
);
854 INT_XLATE(buf_core
->di_gen
, mem_core
->di_gen
, dir
, arch
);
859 xfs_dinode_core_t
*dic
,
865 di_flags
= INT_GET(dic
->di_flags
, arch
);
866 flags
= XFS_CFORK_Q_ARCH(dic
, arch
) ? XFS_XFLAG_HASATTR
: 0;
867 if (di_flags
& XFS_DIFLAG_ANY
) {
868 if (di_flags
& XFS_DIFLAG_REALTIME
)
869 flags
|= XFS_XFLAG_REALTIME
;
870 if (di_flags
& XFS_DIFLAG_PREALLOC
)
871 flags
|= XFS_XFLAG_PREALLOC
;
872 if (di_flags
& XFS_DIFLAG_IMMUTABLE
)
873 flags
|= XFS_XFLAG_IMMUTABLE
;
874 if (di_flags
& XFS_DIFLAG_APPEND
)
875 flags
|= XFS_XFLAG_APPEND
;
876 if (di_flags
& XFS_DIFLAG_SYNC
)
877 flags
|= XFS_XFLAG_SYNC
;
878 if (di_flags
& XFS_DIFLAG_NOATIME
)
879 flags
|= XFS_XFLAG_NOATIME
;
880 if (di_flags
& XFS_DIFLAG_NODUMP
)
881 flags
|= XFS_XFLAG_NODUMP
;
882 if (di_flags
& XFS_DIFLAG_RTINHERIT
)
883 flags
|= XFS_XFLAG_RTINHERIT
;
884 if (di_flags
& XFS_DIFLAG_PROJINHERIT
)
885 flags
|= XFS_XFLAG_PROJINHERIT
;
886 if (di_flags
& XFS_DIFLAG_NOSYMLINKS
)
887 flags
|= XFS_XFLAG_NOSYMLINKS
;
893 * Given a mount structure and an inode number, return a pointer
894 * to a newly allocated in-core inode coresponding to the given
897 * Initialize the inode's attributes and extent pointers if it
898 * already has them (it will not if the inode has no links).
913 ASSERT(xfs_inode_zone
!= NULL
);
915 ip
= kmem_zone_zalloc(xfs_inode_zone
, KM_SLEEP
);
920 * Get pointer's to the on-disk inode and the buffer containing it.
921 * If the inode number refers to a block outside the file system
922 * then xfs_itobp() will return NULL. In this case we should
923 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
924 * know that this is a new incore inode.
926 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &bp
, bno
);
929 kmem_zone_free(xfs_inode_zone
, ip
);
934 * Initialize inode's trace buffers.
935 * Do this before xfs_iformat in case it adds entries.
937 #ifdef XFS_BMAP_TRACE
938 ip
->i_xtrace
= ktrace_alloc(XFS_BMAP_KTRACE_SIZE
, KM_SLEEP
);
940 #ifdef XFS_BMBT_TRACE
941 ip
->i_btrace
= ktrace_alloc(XFS_BMBT_KTRACE_SIZE
, KM_SLEEP
);
944 ip
->i_rwtrace
= ktrace_alloc(XFS_RW_KTRACE_SIZE
, KM_SLEEP
);
946 #ifdef XFS_ILOCK_TRACE
947 ip
->i_lock_trace
= ktrace_alloc(XFS_ILOCK_KTRACE_SIZE
, KM_SLEEP
);
949 #ifdef XFS_DIR2_TRACE
950 ip
->i_dir_trace
= ktrace_alloc(XFS_DIR2_KTRACE_SIZE
, KM_SLEEP
);
954 * If we got something that isn't an inode it means someone
955 * (nfs or dmi) has a stale handle.
957 if (INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
) != XFS_DINODE_MAGIC
) {
958 kmem_zone_free(xfs_inode_zone
, ip
);
959 xfs_trans_brelse(tp
, bp
);
961 xfs_fs_cmn_err(CE_ALERT
, mp
, "xfs_iread: "
962 "dip->di_core.di_magic (0x%x) != "
963 "XFS_DINODE_MAGIC (0x%x)",
964 INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
),
967 return XFS_ERROR(EINVAL
);
971 * If the on-disk inode is already linked to a directory
972 * entry, copy all of the inode into the in-core inode.
973 * xfs_iformat() handles copying in the inode format
974 * specific information.
975 * Otherwise, just get the truly permanent information.
977 if (!INT_ISZERO(dip
->di_core
.di_mode
, ARCH_CONVERT
)) {
978 xfs_xlate_dinode_core((xfs_caddr_t
)&dip
->di_core
,
979 &(ip
->i_d
), 1, ARCH_CONVERT
);
980 error
= xfs_iformat(ip
, dip
);
982 kmem_zone_free(xfs_inode_zone
, ip
);
983 xfs_trans_brelse(tp
, bp
);
985 xfs_fs_cmn_err(CE_ALERT
, mp
, "xfs_iread: "
986 "xfs_iformat() returned error %d",
992 ip
->i_d
.di_magic
= INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
);
993 ip
->i_d
.di_version
= INT_GET(dip
->di_core
.di_version
, ARCH_CONVERT
);
994 ip
->i_d
.di_gen
= INT_GET(dip
->di_core
.di_gen
, ARCH_CONVERT
);
995 ip
->i_d
.di_flushiter
= INT_GET(dip
->di_core
.di_flushiter
, ARCH_CONVERT
);
997 * Make sure to pull in the mode here as well in
998 * case the inode is released without being used.
999 * This ensures that xfs_inactive() will see that
1000 * the inode is already free and not try to mess
1001 * with the uninitialized part of it.
1003 ip
->i_d
.di_mode
= 0;
1005 * Initialize the per-fork minima and maxima for a new
1006 * inode here. xfs_iformat will do it for old inodes.
1008 ip
->i_df
.if_ext_max
=
1009 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
1012 INIT_LIST_HEAD(&ip
->i_reclaim
);
1015 * The inode format changed when we moved the link count and
1016 * made it 32 bits long. If this is an old format inode,
1017 * convert it in memory to look like a new one. If it gets
1018 * flushed to disk we will convert back before flushing or
1019 * logging it. We zero out the new projid field and the old link
1020 * count field. We'll handle clearing the pad field (the remains
1021 * of the old uuid field) when we actually convert the inode to
1022 * the new format. We don't change the version number so that we
1023 * can distinguish this from a real new format inode.
1025 if (ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
) {
1026 ip
->i_d
.di_nlink
= ip
->i_d
.di_onlink
;
1027 ip
->i_d
.di_onlink
= 0;
1028 ip
->i_d
.di_projid
= 0;
1031 ip
->i_delayed_blks
= 0;
1034 * Mark the buffer containing the inode as something to keep
1035 * around for a while. This helps to keep recently accessed
1036 * meta-data in-core longer.
1038 XFS_BUF_SET_REF(bp
, XFS_INO_REF
);
1041 * Use xfs_trans_brelse() to release the buffer containing the
1042 * on-disk inode, because it was acquired with xfs_trans_read_buf()
1043 * in xfs_itobp() above. If tp is NULL, this is just a normal
1044 * brelse(). If we're within a transaction, then xfs_trans_brelse()
1045 * will only release the buffer if it is not dirty within the
1046 * transaction. It will be OK to release the buffer in this case,
1047 * because inodes on disk are never destroyed and we will be
1048 * locking the new in-core inode before putting it in the hash
1049 * table where other processes can find it. Thus we don't have
1050 * to worry about the inode being changed just because we released
1053 xfs_trans_brelse(tp
, bp
);
1059 * Read in extents from a btree-format inode.
1060 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1072 if (unlikely(XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
)) {
1073 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW
,
1075 return XFS_ERROR(EFSCORRUPTED
);
1077 size
= XFS_IFORK_NEXTENTS(ip
, whichfork
) * (uint
)sizeof(xfs_bmbt_rec_t
);
1078 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1080 * We know that the size is valid (it's checked in iformat_btree)
1082 ifp
->if_u1
.if_extents
= kmem_alloc(size
, KM_SLEEP
);
1083 ASSERT(ifp
->if_u1
.if_extents
!= NULL
);
1084 ifp
->if_lastex
= NULLEXTNUM
;
1085 ifp
->if_bytes
= ifp
->if_real_bytes
= (int)size
;
1086 ifp
->if_flags
|= XFS_IFEXTENTS
;
1087 error
= xfs_bmap_read_extents(tp
, ip
, whichfork
);
1089 kmem_free(ifp
->if_u1
.if_extents
, size
);
1090 ifp
->if_u1
.if_extents
= NULL
;
1091 ifp
->if_bytes
= ifp
->if_real_bytes
= 0;
1092 ifp
->if_flags
&= ~XFS_IFEXTENTS
;
1095 xfs_validate_extents((xfs_bmbt_rec_t
*)ifp
->if_u1
.if_extents
,
1096 XFS_IFORK_NEXTENTS(ip
, whichfork
), 0, XFS_EXTFMT_INODE(ip
));
1101 * Allocate an inode on disk and return a copy of its in-core version.
1102 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1103 * appropriately within the inode. The uid and gid for the inode are
1104 * set according to the contents of the given cred structure.
1106 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1107 * has a free inode available, call xfs_iget()
1108 * to obtain the in-core version of the allocated inode. Finally,
1109 * fill in the inode and log its initial contents. In this case,
1110 * ialloc_context would be set to NULL and call_again set to false.
1112 * If xfs_dialloc() does not have an available inode,
1113 * it will replenish its supply by doing an allocation. Since we can
1114 * only do one allocation within a transaction without deadlocks, we
1115 * must commit the current transaction before returning the inode itself.
1116 * In this case, therefore, we will set call_again to true and return.
1117 * The caller should then commit the current transaction, start a new
1118 * transaction, and call xfs_ialloc() again to actually get the inode.
1120 * To ensure that some other process does not grab the inode that
1121 * was allocated during the first call to xfs_ialloc(), this routine
1122 * also returns the [locked] bp pointing to the head of the freelist
1123 * as ialloc_context. The caller should hold this buffer across
1124 * the commit and pass it back into this routine on the second call.
1136 xfs_buf_t
**ialloc_context
,
1137 boolean_t
*call_again
,
1147 * Call the space management code to pick
1148 * the on-disk inode to be allocated.
1150 error
= xfs_dialloc(tp
, pip
->i_ino
, mode
, okalloc
,
1151 ialloc_context
, call_again
, &ino
);
1155 if (*call_again
|| ino
== NULLFSINO
) {
1159 ASSERT(*ialloc_context
== NULL
);
1162 * Get the in-core inode with the lock held exclusively.
1163 * This is because we're setting fields here we need
1164 * to prevent others from looking at until we're done.
1166 error
= xfs_trans_iget(tp
->t_mountp
, tp
, ino
, XFS_ILOCK_EXCL
, &ip
);
1173 vp
->v_type
= IFTOVT(mode
);
1174 ip
->i_d
.di_mode
= (__uint16_t
)mode
;
1175 ip
->i_d
.di_onlink
= 0;
1176 ip
->i_d
.di_nlink
= nlink
;
1177 ASSERT(ip
->i_d
.di_nlink
== nlink
);
1178 ip
->i_d
.di_uid
= current_fsuid(cr
);
1179 ip
->i_d
.di_gid
= current_fsgid(cr
);
1180 ip
->i_d
.di_projid
= prid
;
1181 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
1184 * If the superblock version is up to where we support new format
1185 * inodes and this is currently an old format inode, then change
1186 * the inode version number now. This way we only do the conversion
1187 * here rather than here and in the flush/logging code.
1189 if (XFS_SB_VERSION_HASNLINK(&tp
->t_mountp
->m_sb
) &&
1190 ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
) {
1191 ip
->i_d
.di_version
= XFS_DINODE_VERSION_2
;
1193 * We've already zeroed the old link count, the projid field,
1194 * and the pad field.
1199 * Project ids won't be stored on disk if we are using a version 1 inode.
1201 if ( (prid
!= 0) && (ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
))
1202 xfs_bump_ino_vers2(tp
, ip
);
1204 if (XFS_INHERIT_GID(pip
, vp
->v_vfsp
)) {
1205 ip
->i_d
.di_gid
= pip
->i_d
.di_gid
;
1206 if ((pip
->i_d
.di_mode
& S_ISGID
) && (mode
& S_IFMT
) == S_IFDIR
) {
1207 ip
->i_d
.di_mode
|= S_ISGID
;
1212 * If the group ID of the new file does not match the effective group
1213 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1214 * (and only if the irix_sgid_inherit compatibility variable is set).
1216 if ((irix_sgid_inherit
) &&
1217 (ip
->i_d
.di_mode
& S_ISGID
) &&
1218 (!in_group_p((gid_t
)ip
->i_d
.di_gid
))) {
1219 ip
->i_d
.di_mode
&= ~S_ISGID
;
1222 ip
->i_d
.di_size
= 0;
1223 ip
->i_d
.di_nextents
= 0;
1224 ASSERT(ip
->i_d
.di_nblocks
== 0);
1225 xfs_ichgtime(ip
, XFS_ICHGTIME_CHG
|XFS_ICHGTIME_ACC
|XFS_ICHGTIME_MOD
);
1227 * di_gen will have been taken care of in xfs_iread.
1229 ip
->i_d
.di_extsize
= 0;
1230 ip
->i_d
.di_dmevmask
= 0;
1231 ip
->i_d
.di_dmstate
= 0;
1232 ip
->i_d
.di_flags
= 0;
1233 flags
= XFS_ILOG_CORE
;
1234 switch (mode
& S_IFMT
) {
1239 ip
->i_d
.di_format
= XFS_DINODE_FMT_DEV
;
1240 ip
->i_df
.if_u2
.if_rdev
= rdev
;
1241 ip
->i_df
.if_flags
= 0;
1242 flags
|= XFS_ILOG_DEV
;
1246 if (unlikely(pip
->i_d
.di_flags
& XFS_DIFLAG_ANY
)) {
1247 if (pip
->i_d
.di_flags
& XFS_DIFLAG_RTINHERIT
) {
1248 if ((mode
& S_IFMT
) == S_IFDIR
) {
1249 ip
->i_d
.di_flags
|= XFS_DIFLAG_RTINHERIT
;
1251 ip
->i_d
.di_flags
|= XFS_DIFLAG_REALTIME
;
1252 ip
->i_iocore
.io_flags
|= XFS_IOCORE_RT
;
1255 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOATIME
) &&
1256 xfs_inherit_noatime
)
1257 ip
->i_d
.di_flags
|= XFS_DIFLAG_NOATIME
;
1258 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NODUMP
) &&
1260 ip
->i_d
.di_flags
|= XFS_DIFLAG_NODUMP
;
1261 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_SYNC
) &&
1263 ip
->i_d
.di_flags
|= XFS_DIFLAG_SYNC
;
1264 if ((pip
->i_d
.di_flags
& XFS_DIFLAG_NOSYMLINKS
) &&
1265 xfs_inherit_nosymlinks
)
1266 ip
->i_d
.di_flags
|= XFS_DIFLAG_NOSYMLINKS
;
1270 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
1271 ip
->i_df
.if_flags
= XFS_IFEXTENTS
;
1272 ip
->i_df
.if_bytes
= ip
->i_df
.if_real_bytes
= 0;
1273 ip
->i_df
.if_u1
.if_extents
= NULL
;
1279 * Attribute fork settings for new inode.
1281 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
1282 ip
->i_d
.di_anextents
= 0;
1285 * Log the new values stuffed into the inode.
1287 xfs_trans_log_inode(tp
, ip
, flags
);
1289 /* now that we have a v_type we can set Linux inode ops (& unlock) */
1290 VFS_INIT_VNODE(XFS_MTOVFS(tp
->t_mountp
), vp
, XFS_ITOBHV(ip
), 1);
1297 * Check to make sure that there are no blocks allocated to the
1298 * file beyond the size of the file. We don't check this for
1299 * files with fixed size extents or real time extents, but we
1300 * at least do it for regular files.
1309 xfs_fileoff_t map_first
;
1311 xfs_bmbt_irec_t imaps
[2];
1313 if ((ip
->i_d
.di_mode
& S_IFMT
) != S_IFREG
)
1316 if ( ip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
)
1320 map_first
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)isize
);
1322 * The filesystem could be shutting down, so bmapi may return
1325 if (xfs_bmapi(NULL
, ip
, map_first
,
1327 (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
)) -
1329 XFS_BMAPI_ENTIRE
, NULL
, 0, imaps
, &nimaps
,
1332 ASSERT(nimaps
== 1);
1333 ASSERT(imaps
[0].br_startblock
== HOLESTARTBLOCK
);
1338 * Calculate the last possible buffered byte in a file. This must
1339 * include data that was buffered beyond the EOF by the write code.
1340 * This also needs to deal with overflowing the xfs_fsize_t type
1341 * which can happen for sizes near the limit.
1343 * We also need to take into account any blocks beyond the EOF. It
1344 * may be the case that they were buffered by a write which failed.
1345 * In that case the pages will still be in memory, but the inode size
1346 * will never have been updated.
1353 xfs_fsize_t last_byte
;
1354 xfs_fileoff_t last_block
;
1355 xfs_fileoff_t size_last_block
;
1358 ASSERT(ismrlocked(&(ip
->i_iolock
), MR_UPDATE
| MR_ACCESS
));
1362 * Only check for blocks beyond the EOF if the extents have
1363 * been read in. This eliminates the need for the inode lock,
1364 * and it also saves us from looking when it really isn't
1367 if (ip
->i_df
.if_flags
& XFS_IFEXTENTS
) {
1368 error
= xfs_bmap_last_offset(NULL
, ip
, &last_block
,
1376 size_last_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)ip
->i_d
.di_size
);
1377 last_block
= XFS_FILEOFF_MAX(last_block
, size_last_block
);
1379 last_byte
= XFS_FSB_TO_B(mp
, last_block
);
1380 if (last_byte
< 0) {
1381 return XFS_MAXIOFFSET(mp
);
1383 last_byte
+= (1 << mp
->m_writeio_log
);
1384 if (last_byte
< 0) {
1385 return XFS_MAXIOFFSET(mp
);
1390 #if defined(XFS_RW_TRACE)
1396 xfs_fsize_t new_size
,
1397 xfs_off_t toss_start
,
1398 xfs_off_t toss_finish
)
1400 if (ip
->i_rwtrace
== NULL
) {
1404 ktrace_enter(ip
->i_rwtrace
,
1407 (void*)(unsigned long)((ip
->i_d
.di_size
>> 32) & 0xffffffff),
1408 (void*)(unsigned long)(ip
->i_d
.di_size
& 0xffffffff),
1409 (void*)((long)flag
),
1410 (void*)(unsigned long)((new_size
>> 32) & 0xffffffff),
1411 (void*)(unsigned long)(new_size
& 0xffffffff),
1412 (void*)(unsigned long)((toss_start
>> 32) & 0xffffffff),
1413 (void*)(unsigned long)(toss_start
& 0xffffffff),
1414 (void*)(unsigned long)((toss_finish
>> 32) & 0xffffffff),
1415 (void*)(unsigned long)(toss_finish
& 0xffffffff),
1416 (void*)(unsigned long)current_cpu(),
1423 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1427 * Start the truncation of the file to new_size. The new size
1428 * must be smaller than the current size. This routine will
1429 * clear the buffer and page caches of file data in the removed
1430 * range, and xfs_itruncate_finish() will remove the underlying
1433 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1434 * must NOT have the inode lock held at all. This is because we're
1435 * calling into the buffer/page cache code and we can't hold the
1436 * inode lock when we do so.
1438 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1439 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1440 * in the case that the caller is locking things out of order and
1441 * may not be able to call xfs_itruncate_finish() with the inode lock
1442 * held without dropping the I/O lock. If the caller must drop the
1443 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1444 * must be called again with all the same restrictions as the initial
1448 xfs_itruncate_start(
1451 xfs_fsize_t new_size
)
1453 xfs_fsize_t last_byte
;
1454 xfs_off_t toss_start
;
1458 ASSERT(ismrlocked(&ip
->i_iolock
, MR_UPDATE
) != 0);
1459 ASSERT((new_size
== 0) || (new_size
<= ip
->i_d
.di_size
));
1460 ASSERT((flags
== XFS_ITRUNC_DEFINITE
) ||
1461 (flags
== XFS_ITRUNC_MAYBE
));
1466 * Call VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES() to get rid of pages and buffers
1467 * overlapping the region being removed. We have to use
1468 * the less efficient VOP_FLUSHINVAL_PAGES() in the case that the
1469 * caller may not be able to finish the truncate without
1470 * dropping the inode's I/O lock. Make sure
1471 * to catch any pages brought in by buffers overlapping
1472 * the EOF by searching out beyond the isize by our
1473 * block size. We round new_size up to a block boundary
1474 * so that we don't toss things on the same block as
1475 * new_size but before it.
1477 * Before calling VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES(), make sure to
1478 * call remapf() over the same region if the file is mapped.
1479 * This frees up mapped file references to the pages in the
1480 * given range and for the VOP_FLUSHINVAL_PAGES() case it ensures
1481 * that we get the latest mapped changes flushed out.
1483 toss_start
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)new_size
);
1484 toss_start
= XFS_FSB_TO_B(mp
, toss_start
);
1485 if (toss_start
< 0) {
1487 * The place to start tossing is beyond our maximum
1488 * file size, so there is no way that the data extended
1493 last_byte
= xfs_file_last_byte(ip
);
1494 xfs_itrunc_trace(XFS_ITRUNC_START
, ip
, flags
, new_size
, toss_start
,
1496 if (last_byte
> toss_start
) {
1497 if (flags
& XFS_ITRUNC_DEFINITE
) {
1498 VOP_TOSS_PAGES(vp
, toss_start
, -1, FI_REMAPF_LOCKED
);
1500 VOP_FLUSHINVAL_PAGES(vp
, toss_start
, -1, FI_REMAPF_LOCKED
);
1505 if (new_size
== 0) {
1506 ASSERT(VN_CACHED(vp
) == 0);
1512 * Shrink the file to the given new_size. The new
1513 * size must be smaller than the current size.
1514 * This will free up the underlying blocks
1515 * in the removed range after a call to xfs_itruncate_start()
1516 * or xfs_atruncate_start().
1518 * The transaction passed to this routine must have made
1519 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1520 * This routine may commit the given transaction and
1521 * start new ones, so make sure everything involved in
1522 * the transaction is tidy before calling here.
1523 * Some transaction will be returned to the caller to be
1524 * committed. The incoming transaction must already include
1525 * the inode, and both inode locks must be held exclusively.
1526 * The inode must also be "held" within the transaction. On
1527 * return the inode will be "held" within the returned transaction.
1528 * This routine does NOT require any disk space to be reserved
1529 * for it within the transaction.
1531 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1532 * and it indicates the fork which is to be truncated. For the
1533 * attribute fork we only support truncation to size 0.
1535 * We use the sync parameter to indicate whether or not the first
1536 * transaction we perform might have to be synchronous. For the attr fork,
1537 * it needs to be so if the unlink of the inode is not yet known to be
1538 * permanent in the log. This keeps us from freeing and reusing the
1539 * blocks of the attribute fork before the unlink of the inode becomes
1542 * For the data fork, we normally have to run synchronously if we're
1543 * being called out of the inactive path or we're being called
1544 * out of the create path where we're truncating an existing file.
1545 * Either way, the truncate needs to be sync so blocks don't reappear
1546 * in the file with altered data in case of a crash. wsync filesystems
1547 * can run the first case async because anything that shrinks the inode
1548 * has to run sync so by the time we're called here from inactive, the
1549 * inode size is permanently set to 0.
1551 * Calls from the truncate path always need to be sync unless we're
1552 * in a wsync filesystem and the file has already been unlinked.
1554 * The caller is responsible for correctly setting the sync parameter.
1555 * It gets too hard for us to guess here which path we're being called
1556 * out of just based on inode state.
1559 xfs_itruncate_finish(
1562 xfs_fsize_t new_size
,
1566 xfs_fsblock_t first_block
;
1567 xfs_fileoff_t first_unmap_block
;
1568 xfs_fileoff_t last_block
;
1569 xfs_filblks_t unmap_len
=0;
1574 xfs_bmap_free_t free_list
;
1577 ASSERT(ismrlocked(&ip
->i_iolock
, MR_UPDATE
) != 0);
1578 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
) != 0);
1579 ASSERT((new_size
== 0) || (new_size
<= ip
->i_d
.di_size
));
1580 ASSERT(*tp
!= NULL
);
1581 ASSERT((*tp
)->t_flags
& XFS_TRANS_PERM_LOG_RES
);
1582 ASSERT(ip
->i_transp
== *tp
);
1583 ASSERT(ip
->i_itemp
!= NULL
);
1584 ASSERT(ip
->i_itemp
->ili_flags
& XFS_ILI_HOLD
);
1588 mp
= (ntp
)->t_mountp
;
1589 ASSERT(! XFS_NOT_DQATTACHED(mp
, ip
));
1592 * We only support truncating the entire attribute fork.
1594 if (fork
== XFS_ATTR_FORK
) {
1597 first_unmap_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)new_size
);
1598 xfs_itrunc_trace(XFS_ITRUNC_FINISH1
, ip
, 0, new_size
, 0, 0);
1600 * The first thing we do is set the size to new_size permanently
1601 * on disk. This way we don't have to worry about anyone ever
1602 * being able to look at the data being freed even in the face
1603 * of a crash. What we're getting around here is the case where
1604 * we free a block, it is allocated to another file, it is written
1605 * to, and then we crash. If the new data gets written to the
1606 * file but the log buffers containing the free and reallocation
1607 * don't, then we'd end up with garbage in the blocks being freed.
1608 * As long as we make the new_size permanent before actually
1609 * freeing any blocks it doesn't matter if they get writtten to.
1611 * The callers must signal into us whether or not the size
1612 * setting here must be synchronous. There are a few cases
1613 * where it doesn't have to be synchronous. Those cases
1614 * occur if the file is unlinked and we know the unlink is
1615 * permanent or if the blocks being truncated are guaranteed
1616 * to be beyond the inode eof (regardless of the link count)
1617 * and the eof value is permanent. Both of these cases occur
1618 * only on wsync-mounted filesystems. In those cases, we're
1619 * guaranteed that no user will ever see the data in the blocks
1620 * that are being truncated so the truncate can run async.
1621 * In the free beyond eof case, the file may wind up with
1622 * more blocks allocated to it than it needs if we crash
1623 * and that won't get fixed until the next time the file
1624 * is re-opened and closed but that's ok as that shouldn't
1625 * be too many blocks.
1627 * However, we can't just make all wsync xactions run async
1628 * because there's one call out of the create path that needs
1629 * to run sync where it's truncating an existing file to size
1630 * 0 whose size is > 0.
1632 * It's probably possible to come up with a test in this
1633 * routine that would correctly distinguish all the above
1634 * cases from the values of the function parameters and the
1635 * inode state but for sanity's sake, I've decided to let the
1636 * layers above just tell us. It's simpler to correctly figure
1637 * out in the layer above exactly under what conditions we
1638 * can run async and I think it's easier for others read and
1639 * follow the logic in case something has to be changed.
1640 * cscope is your friend -- rcc.
1642 * The attribute fork is much simpler.
1644 * For the attribute fork we allow the caller to tell us whether
1645 * the unlink of the inode that led to this call is yet permanent
1646 * in the on disk log. If it is not and we will be freeing extents
1647 * in this inode then we make the first transaction synchronous
1648 * to make sure that the unlink is permanent by the time we free
1651 if (fork
== XFS_DATA_FORK
) {
1652 if (ip
->i_d
.di_nextents
> 0) {
1653 ip
->i_d
.di_size
= new_size
;
1654 xfs_trans_log_inode(ntp
, ip
, XFS_ILOG_CORE
);
1657 ASSERT(!(mp
->m_flags
& XFS_MOUNT_WSYNC
));
1658 if (ip
->i_d
.di_anextents
> 0)
1659 xfs_trans_set_sync(ntp
);
1661 ASSERT(fork
== XFS_DATA_FORK
||
1662 (fork
== XFS_ATTR_FORK
&&
1663 ((sync
&& !(mp
->m_flags
& XFS_MOUNT_WSYNC
)) ||
1664 (sync
== 0 && (mp
->m_flags
& XFS_MOUNT_WSYNC
)))));
1667 * Since it is possible for space to become allocated beyond
1668 * the end of the file (in a crash where the space is allocated
1669 * but the inode size is not yet updated), simply remove any
1670 * blocks which show up between the new EOF and the maximum
1671 * possible file size. If the first block to be removed is
1672 * beyond the maximum file size (ie it is the same as last_block),
1673 * then there is nothing to do.
1675 last_block
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
));
1676 ASSERT(first_unmap_block
<= last_block
);
1678 if (last_block
== first_unmap_block
) {
1681 unmap_len
= last_block
- first_unmap_block
+ 1;
1685 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1686 * will tell us whether it freed the entire range or
1687 * not. If this is a synchronous mount (wsync),
1688 * then we can tell bunmapi to keep all the
1689 * transactions asynchronous since the unlink
1690 * transaction that made this inode inactive has
1691 * already hit the disk. There's no danger of
1692 * the freed blocks being reused, there being a
1693 * crash, and the reused blocks suddenly reappearing
1694 * in this file with garbage in them once recovery
1697 XFS_BMAP_INIT(&free_list
, &first_block
);
1698 error
= xfs_bunmapi(ntp
, ip
, first_unmap_block
,
1700 XFS_BMAPI_AFLAG(fork
) |
1701 (sync
? 0 : XFS_BMAPI_ASYNC
),
1702 XFS_ITRUNC_MAX_EXTENTS
,
1703 &first_block
, &free_list
, &done
);
1706 * If the bunmapi call encounters an error,
1707 * return to the caller where the transaction
1708 * can be properly aborted. We just need to
1709 * make sure we're not holding any resources
1710 * that we were not when we came in.
1712 xfs_bmap_cancel(&free_list
);
1717 * Duplicate the transaction that has the permanent
1718 * reservation and commit the old transaction.
1720 error
= xfs_bmap_finish(tp
, &free_list
, first_block
,
1725 * If the bmap finish call encounters an error,
1726 * return to the caller where the transaction
1727 * can be properly aborted. We just need to
1728 * make sure we're not holding any resources
1729 * that we were not when we came in.
1731 * Aborting from this point might lose some
1732 * blocks in the file system, but oh well.
1734 xfs_bmap_cancel(&free_list
);
1737 * If the passed in transaction committed
1738 * in xfs_bmap_finish(), then we want to
1739 * add the inode to this one before returning.
1740 * This keeps things simple for the higher
1741 * level code, because it always knows that
1742 * the inode is locked and held in the
1743 * transaction that returns to it whether
1744 * errors occur or not. We don't mark the
1745 * inode dirty so that this transaction can
1746 * be easily aborted if possible.
1748 xfs_trans_ijoin(ntp
, ip
,
1749 XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1750 xfs_trans_ihold(ntp
, ip
);
1757 * The first xact was committed,
1758 * so add the inode to the new one.
1759 * Mark it dirty so it will be logged
1760 * and moved forward in the log as
1761 * part of every commit.
1763 xfs_trans_ijoin(ntp
, ip
,
1764 XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1765 xfs_trans_ihold(ntp
, ip
);
1766 xfs_trans_log_inode(ntp
, ip
, XFS_ILOG_CORE
);
1768 ntp
= xfs_trans_dup(ntp
);
1769 (void) xfs_trans_commit(*tp
, 0, NULL
);
1771 error
= xfs_trans_reserve(ntp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0,
1772 XFS_TRANS_PERM_LOG_RES
,
1773 XFS_ITRUNCATE_LOG_COUNT
);
1775 * Add the inode being truncated to the next chained
1778 xfs_trans_ijoin(ntp
, ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1779 xfs_trans_ihold(ntp
, ip
);
1784 * Only update the size in the case of the data fork, but
1785 * always re-log the inode so that our permanent transaction
1786 * can keep on rolling it forward in the log.
1788 if (fork
== XFS_DATA_FORK
) {
1789 xfs_isize_check(mp
, ip
, new_size
);
1790 ip
->i_d
.di_size
= new_size
;
1792 xfs_trans_log_inode(ntp
, ip
, XFS_ILOG_CORE
);
1793 ASSERT((new_size
!= 0) ||
1794 (fork
== XFS_ATTR_FORK
) ||
1795 (ip
->i_delayed_blks
== 0));
1796 ASSERT((new_size
!= 0) ||
1797 (fork
== XFS_ATTR_FORK
) ||
1798 (ip
->i_d
.di_nextents
== 0));
1799 xfs_itrunc_trace(XFS_ITRUNC_FINISH2
, ip
, 0, new_size
, 0, 0);
1807 * Do the first part of growing a file: zero any data in the last
1808 * block that is beyond the old EOF. We need to do this before
1809 * the inode is joined to the transaction to modify the i_size.
1810 * That way we can drop the inode lock and call into the buffer
1811 * cache to get the buffer mapping the EOF.
1816 xfs_fsize_t new_size
,
1822 ASSERT(ismrlocked(&(ip
->i_lock
), MR_UPDATE
) != 0);
1823 ASSERT(ismrlocked(&(ip
->i_iolock
), MR_UPDATE
) != 0);
1824 ASSERT(new_size
> ip
->i_d
.di_size
);
1827 isize
= ip
->i_d
.di_size
;
1829 * Zero any pages that may have been created by
1830 * xfs_write_file() beyond the end of the file
1831 * and any blocks between the old and new file sizes.
1833 error
= xfs_zero_eof(XFS_ITOV(ip
), &ip
->i_iocore
, new_size
, isize
,
1841 * This routine is called to extend the size of a file.
1842 * The inode must have both the iolock and the ilock locked
1843 * for update and it must be a part of the current transaction.
1844 * The xfs_igrow_start() function must have been called previously.
1845 * If the change_flag is not zero, the inode change timestamp will
1852 xfs_fsize_t new_size
,
1855 ASSERT(ismrlocked(&(ip
->i_lock
), MR_UPDATE
) != 0);
1856 ASSERT(ismrlocked(&(ip
->i_iolock
), MR_UPDATE
) != 0);
1857 ASSERT(ip
->i_transp
== tp
);
1858 ASSERT(new_size
> ip
->i_d
.di_size
);
1861 * Update the file size. Update the inode change timestamp
1862 * if change_flag set.
1864 ip
->i_d
.di_size
= new_size
;
1866 xfs_ichgtime(ip
, XFS_ICHGTIME_CHG
);
1867 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1873 * This is called when the inode's link count goes to 0.
1874 * We place the on-disk inode on a list in the AGI. It
1875 * will be pulled from this list when the inode is freed.
1887 xfs_agnumber_t agno
;
1888 xfs_daddr_t agdaddr
;
1895 ASSERT(ip
->i_d
.di_nlink
== 0);
1896 ASSERT(ip
->i_d
.di_mode
!= 0);
1897 ASSERT(ip
->i_transp
== tp
);
1901 agno
= XFS_INO_TO_AGNO(mp
, ip
->i_ino
);
1902 agdaddr
= XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
));
1905 * Get the agi buffer first. It ensures lock ordering
1908 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, agdaddr
,
1909 XFS_FSS_TO_BB(mp
, 1), 0, &agibp
);
1914 * Validate the magic number of the agi block.
1916 agi
= XFS_BUF_TO_AGI(agibp
);
1918 INT_GET(agi
->agi_magicnum
, ARCH_CONVERT
) == XFS_AGI_MAGIC
&&
1919 XFS_AGI_GOOD_VERSION(INT_GET(agi
->agi_versionnum
, ARCH_CONVERT
));
1920 if (unlikely(XFS_TEST_ERROR(!agi_ok
, mp
, XFS_ERRTAG_IUNLINK
,
1921 XFS_RANDOM_IUNLINK
))) {
1922 XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW
, mp
, agi
);
1923 xfs_trans_brelse(tp
, agibp
);
1924 return XFS_ERROR(EFSCORRUPTED
);
1927 * Get the index into the agi hash table for the
1928 * list this inode will go on.
1930 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
1932 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
1933 ASSERT(!INT_ISZERO(agi
->agi_unlinked
[bucket_index
], ARCH_CONVERT
));
1934 ASSERT(INT_GET(agi
->agi_unlinked
[bucket_index
], ARCH_CONVERT
) != agino
);
1936 if (INT_GET(agi
->agi_unlinked
[bucket_index
], ARCH_CONVERT
) != NULLAGINO
) {
1938 * There is already another inode in the bucket we need
1939 * to add ourselves to. Add us at the front of the list.
1940 * Here we put the head pointer into our next pointer,
1941 * and then we fall through to point the head at us.
1943 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, 0);
1947 ASSERT(INT_GET(dip
->di_next_unlinked
, ARCH_CONVERT
) == NULLAGINO
);
1948 ASSERT(!INT_ISZERO(dip
->di_next_unlinked
, ARCH_CONVERT
));
1949 /* both on-disk, don't endian flip twice */
1950 dip
->di_next_unlinked
= agi
->agi_unlinked
[bucket_index
];
1951 offset
= ip
->i_boffset
+
1952 offsetof(xfs_dinode_t
, di_next_unlinked
);
1953 xfs_trans_inode_buf(tp
, ibp
);
1954 xfs_trans_log_buf(tp
, ibp
, offset
,
1955 (offset
+ sizeof(xfs_agino_t
) - 1));
1956 xfs_inobp_check(mp
, ibp
);
1960 * Point the bucket head pointer at the inode being inserted.
1963 INT_SET(agi
->agi_unlinked
[bucket_index
], ARCH_CONVERT
, agino
);
1964 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
1965 (sizeof(xfs_agino_t
) * bucket_index
);
1966 xfs_trans_log_buf(tp
, agibp
, offset
,
1967 (offset
+ sizeof(xfs_agino_t
) - 1));
1972 * Pull the on-disk inode from the AGI unlinked list.
1985 xfs_agnumber_t agno
;
1986 xfs_daddr_t agdaddr
;
1988 xfs_agino_t next_agino
;
1989 xfs_buf_t
*last_ibp
;
1990 xfs_dinode_t
*last_dip
;
1992 int offset
, last_offset
;
1997 * First pull the on-disk inode from the AGI unlinked list.
2001 agno
= XFS_INO_TO_AGNO(mp
, ip
->i_ino
);
2002 agdaddr
= XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
));
2005 * Get the agi buffer first. It ensures lock ordering
2008 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, agdaddr
,
2009 XFS_FSS_TO_BB(mp
, 1), 0, &agibp
);
2012 "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
2013 error
, mp
->m_fsname
);
2017 * Validate the magic number of the agi block.
2019 agi
= XFS_BUF_TO_AGI(agibp
);
2021 INT_GET(agi
->agi_magicnum
, ARCH_CONVERT
) == XFS_AGI_MAGIC
&&
2022 XFS_AGI_GOOD_VERSION(INT_GET(agi
->agi_versionnum
, ARCH_CONVERT
));
2023 if (unlikely(XFS_TEST_ERROR(!agi_ok
, mp
, XFS_ERRTAG_IUNLINK_REMOVE
,
2024 XFS_RANDOM_IUNLINK_REMOVE
))) {
2025 XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW
,
2027 xfs_trans_brelse(tp
, agibp
);
2029 "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
2031 return XFS_ERROR(EFSCORRUPTED
);
2034 * Get the index into the agi hash table for the
2035 * list this inode will go on.
2037 agino
= XFS_INO_TO_AGINO(mp
, ip
->i_ino
);
2039 bucket_index
= agino
% XFS_AGI_UNLINKED_BUCKETS
;
2040 ASSERT(INT_GET(agi
->agi_unlinked
[bucket_index
], ARCH_CONVERT
) != NULLAGINO
);
2041 ASSERT(!INT_ISZERO(agi
->agi_unlinked
[bucket_index
], ARCH_CONVERT
));
2043 if (INT_GET(agi
->agi_unlinked
[bucket_index
], ARCH_CONVERT
) == agino
) {
2045 * We're at the head of the list. Get the inode's
2046 * on-disk buffer to see if there is anyone after us
2047 * on the list. Only modify our next pointer if it
2048 * is not already NULLAGINO. This saves us the overhead
2049 * of dealing with the buffer when there is no need to
2052 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, 0);
2055 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2056 error
, mp
->m_fsname
);
2059 next_agino
= INT_GET(dip
->di_next_unlinked
, ARCH_CONVERT
);
2060 ASSERT(next_agino
!= 0);
2061 if (next_agino
!= NULLAGINO
) {
2062 INT_SET(dip
->di_next_unlinked
, ARCH_CONVERT
, NULLAGINO
);
2063 offset
= ip
->i_boffset
+
2064 offsetof(xfs_dinode_t
, di_next_unlinked
);
2065 xfs_trans_inode_buf(tp
, ibp
);
2066 xfs_trans_log_buf(tp
, ibp
, offset
,
2067 (offset
+ sizeof(xfs_agino_t
) - 1));
2068 xfs_inobp_check(mp
, ibp
);
2070 xfs_trans_brelse(tp
, ibp
);
2073 * Point the bucket head pointer at the next inode.
2075 ASSERT(next_agino
!= 0);
2076 ASSERT(next_agino
!= agino
);
2077 INT_SET(agi
->agi_unlinked
[bucket_index
], ARCH_CONVERT
, next_agino
);
2078 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
2079 (sizeof(xfs_agino_t
) * bucket_index
);
2080 xfs_trans_log_buf(tp
, agibp
, offset
,
2081 (offset
+ sizeof(xfs_agino_t
) - 1));
2084 * We need to search the list for the inode being freed.
2086 next_agino
= INT_GET(agi
->agi_unlinked
[bucket_index
], ARCH_CONVERT
);
2088 while (next_agino
!= agino
) {
2090 * If the last inode wasn't the one pointing to
2091 * us, then release its buffer since we're not
2092 * going to do anything with it.
2094 if (last_ibp
!= NULL
) {
2095 xfs_trans_brelse(tp
, last_ibp
);
2097 next_ino
= XFS_AGINO_TO_INO(mp
, agno
, next_agino
);
2098 error
= xfs_inotobp(mp
, tp
, next_ino
, &last_dip
,
2099 &last_ibp
, &last_offset
);
2102 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
2103 error
, mp
->m_fsname
);
2106 next_agino
= INT_GET(last_dip
->di_next_unlinked
, ARCH_CONVERT
);
2107 ASSERT(next_agino
!= NULLAGINO
);
2108 ASSERT(next_agino
!= 0);
2111 * Now last_ibp points to the buffer previous to us on
2112 * the unlinked list. Pull us from the list.
2114 error
= xfs_itobp(mp
, tp
, ip
, &dip
, &ibp
, 0);
2117 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2118 error
, mp
->m_fsname
);
2121 next_agino
= INT_GET(dip
->di_next_unlinked
, ARCH_CONVERT
);
2122 ASSERT(next_agino
!= 0);
2123 ASSERT(next_agino
!= agino
);
2124 if (next_agino
!= NULLAGINO
) {
2125 INT_SET(dip
->di_next_unlinked
, ARCH_CONVERT
, NULLAGINO
);
2126 offset
= ip
->i_boffset
+
2127 offsetof(xfs_dinode_t
, di_next_unlinked
);
2128 xfs_trans_inode_buf(tp
, ibp
);
2129 xfs_trans_log_buf(tp
, ibp
, offset
,
2130 (offset
+ sizeof(xfs_agino_t
) - 1));
2131 xfs_inobp_check(mp
, ibp
);
2133 xfs_trans_brelse(tp
, ibp
);
2136 * Point the previous inode on the list to the next inode.
2138 INT_SET(last_dip
->di_next_unlinked
, ARCH_CONVERT
, next_agino
);
2139 ASSERT(next_agino
!= 0);
2140 offset
= last_offset
+ offsetof(xfs_dinode_t
, di_next_unlinked
);
2141 xfs_trans_inode_buf(tp
, last_ibp
);
2142 xfs_trans_log_buf(tp
, last_ibp
, offset
,
2143 (offset
+ sizeof(xfs_agino_t
) - 1));
2144 xfs_inobp_check(mp
, last_ibp
);
2149 static __inline__
int xfs_inode_clean(xfs_inode_t
*ip
)
2151 return (((ip
->i_itemp
== NULL
) ||
2152 !(ip
->i_itemp
->ili_format
.ilf_fields
& XFS_ILOG_ALL
)) &&
2153 (ip
->i_update_core
== 0));
2158 xfs_inode_t
*free_ip
,
2162 xfs_mount_t
*mp
= free_ip
->i_mount
;
2163 int blks_per_cluster
;
2166 int i
, j
, found
, pre_flushed
;
2170 xfs_inode_t
*ip
, **ip_found
;
2171 xfs_inode_log_item_t
*iip
;
2172 xfs_log_item_t
*lip
;
2175 if (mp
->m_sb
.sb_blocksize
>= XFS_INODE_CLUSTER_SIZE(mp
)) {
2176 blks_per_cluster
= 1;
2177 ninodes
= mp
->m_sb
.sb_inopblock
;
2178 nbufs
= XFS_IALLOC_BLOCKS(mp
);
2180 blks_per_cluster
= XFS_INODE_CLUSTER_SIZE(mp
) /
2181 mp
->m_sb
.sb_blocksize
;
2182 ninodes
= blks_per_cluster
* mp
->m_sb
.sb_inopblock
;
2183 nbufs
= XFS_IALLOC_BLOCKS(mp
) / blks_per_cluster
;
2186 ip_found
= kmem_alloc(ninodes
* sizeof(xfs_inode_t
*), KM_NOFS
);
2188 for (j
= 0; j
< nbufs
; j
++, inum
+= ninodes
) {
2189 blkno
= XFS_AGB_TO_DADDR(mp
, XFS_INO_TO_AGNO(mp
, inum
),
2190 XFS_INO_TO_AGBNO(mp
, inum
));
2194 * Look for each inode in memory and attempt to lock it,
2195 * we can be racing with flush and tail pushing here.
2196 * any inode we get the locks on, add to an array of
2197 * inode items to process later.
2199 * The get the buffer lock, we could beat a flush
2200 * or tail pushing thread to the lock here, in which
2201 * case they will go looking for the inode buffer
2202 * and fail, we need some other form of interlock
2206 for (i
= 0; i
< ninodes
; i
++) {
2207 ih
= XFS_IHASH(mp
, inum
+ i
);
2208 read_lock(&ih
->ih_lock
);
2209 for (ip
= ih
->ih_next
; ip
!= NULL
; ip
= ip
->i_next
) {
2210 if (ip
->i_ino
== inum
+ i
)
2214 /* Inode not in memory or we found it already,
2217 if (!ip
|| (ip
->i_flags
& XFS_ISTALE
)) {
2218 read_unlock(&ih
->ih_lock
);
2222 if (xfs_inode_clean(ip
)) {
2223 read_unlock(&ih
->ih_lock
);
2227 /* If we can get the locks then add it to the
2228 * list, otherwise by the time we get the bp lock
2229 * below it will already be attached to the
2233 /* This inode will already be locked - by us, lets
2237 if (ip
== free_ip
) {
2238 if (xfs_iflock_nowait(ip
)) {
2239 ip
->i_flags
|= XFS_ISTALE
;
2241 if (xfs_inode_clean(ip
)) {
2244 ip_found
[found
++] = ip
;
2247 read_unlock(&ih
->ih_lock
);
2251 if (xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
)) {
2252 if (xfs_iflock_nowait(ip
)) {
2253 ip
->i_flags
|= XFS_ISTALE
;
2255 if (xfs_inode_clean(ip
)) {
2257 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2259 ip_found
[found
++] = ip
;
2262 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2266 read_unlock(&ih
->ih_lock
);
2269 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, blkno
,
2270 mp
->m_bsize
* blks_per_cluster
,
2274 lip
= XFS_BUF_FSPRIVATE(bp
, xfs_log_item_t
*);
2276 if (lip
->li_type
== XFS_LI_INODE
) {
2277 iip
= (xfs_inode_log_item_t
*)lip
;
2278 ASSERT(iip
->ili_logged
== 1);
2279 lip
->li_cb
= (void(*)(xfs_buf_t
*,xfs_log_item_t
*)) xfs_istale_done
;
2281 iip
->ili_flush_lsn
= iip
->ili_item
.li_lsn
;
2283 iip
->ili_inode
->i_flags
|= XFS_ISTALE
;
2286 lip
= lip
->li_bio_list
;
2289 for (i
= 0; i
< found
; i
++) {
2294 ip
->i_update_core
= 0;
2296 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2300 iip
->ili_last_fields
= iip
->ili_format
.ilf_fields
;
2301 iip
->ili_format
.ilf_fields
= 0;
2302 iip
->ili_logged
= 1;
2304 iip
->ili_flush_lsn
= iip
->ili_item
.li_lsn
;
2307 xfs_buf_attach_iodone(bp
,
2308 (void(*)(xfs_buf_t
*,xfs_log_item_t
*))
2309 xfs_istale_done
, (xfs_log_item_t
*)iip
);
2310 if (ip
!= free_ip
) {
2311 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2315 if (found
|| pre_flushed
)
2316 xfs_trans_stale_inode_buf(tp
, bp
);
2317 xfs_trans_binval(tp
, bp
);
2320 kmem_free(ip_found
, ninodes
* sizeof(xfs_inode_t
*));
2324 * This is called to return an inode to the inode free list.
2325 * The inode should already be truncated to 0 length and have
2326 * no pages associated with it. This routine also assumes that
2327 * the inode is already a part of the transaction.
2329 * The on-disk copy of the inode will have been added to the list
2330 * of unlinked inodes in the AGI. We need to remove the inode from
2331 * that list atomically with respect to freeing it here.
2337 xfs_bmap_free_t
*flist
)
2341 xfs_ino_t first_ino
;
2343 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
));
2344 ASSERT(ip
->i_transp
== tp
);
2345 ASSERT(ip
->i_d
.di_nlink
== 0);
2346 ASSERT(ip
->i_d
.di_nextents
== 0);
2347 ASSERT(ip
->i_d
.di_anextents
== 0);
2348 ASSERT((ip
->i_d
.di_size
== 0) ||
2349 ((ip
->i_d
.di_mode
& S_IFMT
) != S_IFREG
));
2350 ASSERT(ip
->i_d
.di_nblocks
== 0);
2353 * Pull the on-disk inode from the AGI unlinked list.
2355 error
= xfs_iunlink_remove(tp
, ip
);
2360 error
= xfs_difree(tp
, ip
->i_ino
, flist
, &delete, &first_ino
);
2364 ip
->i_d
.di_mode
= 0; /* mark incore inode as free */
2365 ip
->i_d
.di_flags
= 0;
2366 ip
->i_d
.di_dmevmask
= 0;
2367 ip
->i_d
.di_forkoff
= 0; /* mark the attr fork not in use */
2368 ip
->i_df
.if_ext_max
=
2369 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
2370 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
2371 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
2373 * Bump the generation count so no one will be confused
2374 * by reincarnations of this inode.
2377 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2380 xfs_ifree_cluster(ip
, tp
, first_ino
);
2387 * Reallocate the space for if_broot based on the number of records
2388 * being added or deleted as indicated in rec_diff. Move the records
2389 * and pointers in if_broot to fit the new size. When shrinking this
2390 * will eliminate holes between the records and pointers created by
2391 * the caller. When growing this will create holes to be filled in
2394 * The caller must not request to add more records than would fit in
2395 * the on-disk inode root. If the if_broot is currently NULL, then
2396 * if we adding records one will be allocated. The caller must also
2397 * not request that the number of records go below zero, although
2398 * it can go to zero.
2400 * ip -- the inode whose if_broot area is changing
2401 * ext_diff -- the change in the number of records, positive or negative,
2402 * requested for the if_broot array.
2412 xfs_bmbt_block_t
*new_broot
;
2419 * Handle the degenerate case quietly.
2421 if (rec_diff
== 0) {
2425 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2428 * If there wasn't any memory allocated before, just
2429 * allocate it now and get out.
2431 if (ifp
->if_broot_bytes
== 0) {
2432 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff
);
2433 ifp
->if_broot
= (xfs_bmbt_block_t
*)kmem_alloc(new_size
,
2435 ifp
->if_broot_bytes
= (int)new_size
;
2440 * If there is already an existing if_broot, then we need
2441 * to realloc() it and shift the pointers to their new
2442 * location. The records don't change location because
2443 * they are kept butted up against the btree block header.
2445 cur_max
= XFS_BMAP_BROOT_MAXRECS(ifp
->if_broot_bytes
);
2446 new_max
= cur_max
+ rec_diff
;
2447 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max
);
2448 ifp
->if_broot
= (xfs_bmbt_block_t
*)
2449 kmem_realloc(ifp
->if_broot
,
2451 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max
), /* old size */
2453 op
= (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp
->if_broot
, 1,
2454 ifp
->if_broot_bytes
);
2455 np
= (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp
->if_broot
, 1,
2457 ifp
->if_broot_bytes
= (int)new_size
;
2458 ASSERT(ifp
->if_broot_bytes
<=
2459 XFS_IFORK_SIZE(ip
, whichfork
) + XFS_BROOT_SIZE_ADJ
);
2460 memmove(np
, op
, cur_max
* (uint
)sizeof(xfs_dfsbno_t
));
2465 * rec_diff is less than 0. In this case, we are shrinking the
2466 * if_broot buffer. It must already exist. If we go to zero
2467 * records, just get rid of the root and clear the status bit.
2469 ASSERT((ifp
->if_broot
!= NULL
) && (ifp
->if_broot_bytes
> 0));
2470 cur_max
= XFS_BMAP_BROOT_MAXRECS(ifp
->if_broot_bytes
);
2471 new_max
= cur_max
+ rec_diff
;
2472 ASSERT(new_max
>= 0);
2474 new_size
= (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max
);
2478 new_broot
= (xfs_bmbt_block_t
*)kmem_alloc(new_size
, KM_SLEEP
);
2480 * First copy over the btree block header.
2482 memcpy(new_broot
, ifp
->if_broot
, sizeof(xfs_bmbt_block_t
));
2485 ifp
->if_flags
&= ~XFS_IFBROOT
;
2489 * Only copy the records and pointers if there are any.
2493 * First copy the records.
2495 op
= (char *)XFS_BMAP_BROOT_REC_ADDR(ifp
->if_broot
, 1,
2496 ifp
->if_broot_bytes
);
2497 np
= (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot
, 1,
2499 memcpy(np
, op
, new_max
* (uint
)sizeof(xfs_bmbt_rec_t
));
2502 * Then copy the pointers.
2504 op
= (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp
->if_broot
, 1,
2505 ifp
->if_broot_bytes
);
2506 np
= (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot
, 1,
2508 memcpy(np
, op
, new_max
* (uint
)sizeof(xfs_dfsbno_t
));
2510 kmem_free(ifp
->if_broot
, ifp
->if_broot_bytes
);
2511 ifp
->if_broot
= new_broot
;
2512 ifp
->if_broot_bytes
= (int)new_size
;
2513 ASSERT(ifp
->if_broot_bytes
<=
2514 XFS_IFORK_SIZE(ip
, whichfork
) + XFS_BROOT_SIZE_ADJ
);
2520 * This is called when the amount of space needed for if_extents
2521 * is increased or decreased. The change in size is indicated by
2522 * the number of extents that need to be added or deleted in the
2523 * ext_diff parameter.
2525 * If the amount of space needed has decreased below the size of the
2526 * inline buffer, then switch to using the inline buffer. Otherwise,
2527 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2528 * to what is needed.
2530 * ip -- the inode whose if_extents area is changing
2531 * ext_diff -- the change in the number of extents, positive or negative,
2532 * requested for the if_extents array.
2545 if (ext_diff
== 0) {
2549 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2550 byte_diff
= ext_diff
* (uint
)sizeof(xfs_bmbt_rec_t
);
2551 new_size
= (int)ifp
->if_bytes
+ byte_diff
;
2552 ASSERT(new_size
>= 0);
2554 if (new_size
== 0) {
2555 if (ifp
->if_u1
.if_extents
!= ifp
->if_u2
.if_inline_ext
) {
2556 ASSERT(ifp
->if_real_bytes
!= 0);
2557 kmem_free(ifp
->if_u1
.if_extents
, ifp
->if_real_bytes
);
2559 ifp
->if_u1
.if_extents
= NULL
;
2561 } else if (new_size
<= sizeof(ifp
->if_u2
.if_inline_ext
)) {
2563 * If the valid extents can fit in if_inline_ext,
2564 * copy them from the malloc'd vector and free it.
2566 if (ifp
->if_u1
.if_extents
!= ifp
->if_u2
.if_inline_ext
) {
2568 * For now, empty files are format EXTENTS,
2569 * so the if_extents pointer is null.
2571 if (ifp
->if_u1
.if_extents
) {
2572 memcpy(ifp
->if_u2
.if_inline_ext
,
2573 ifp
->if_u1
.if_extents
, new_size
);
2574 kmem_free(ifp
->if_u1
.if_extents
,
2575 ifp
->if_real_bytes
);
2577 ifp
->if_u1
.if_extents
= ifp
->if_u2
.if_inline_ext
;
2581 rnew_size
= new_size
;
2582 if ((rnew_size
& (rnew_size
- 1)) != 0)
2583 rnew_size
= xfs_iroundup(rnew_size
);
2585 * Stuck with malloc/realloc.
2587 if (ifp
->if_u1
.if_extents
== ifp
->if_u2
.if_inline_ext
) {
2588 ifp
->if_u1
.if_extents
= (xfs_bmbt_rec_t
*)
2589 kmem_alloc(rnew_size
, KM_SLEEP
);
2590 memcpy(ifp
->if_u1
.if_extents
, ifp
->if_u2
.if_inline_ext
,
2591 sizeof(ifp
->if_u2
.if_inline_ext
));
2592 } else if (rnew_size
!= ifp
->if_real_bytes
) {
2593 ifp
->if_u1
.if_extents
= (xfs_bmbt_rec_t
*)
2594 kmem_realloc(ifp
->if_u1
.if_extents
,
2600 ifp
->if_real_bytes
= rnew_size
;
2601 ifp
->if_bytes
= new_size
;
2606 * This is called when the amount of space needed for if_data
2607 * is increased or decreased. The change in size is indicated by
2608 * the number of bytes that need to be added or deleted in the
2609 * byte_diff parameter.
2611 * If the amount of space needed has decreased below the size of the
2612 * inline buffer, then switch to using the inline buffer. Otherwise,
2613 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2614 * to what is needed.
2616 * ip -- the inode whose if_data area is changing
2617 * byte_diff -- the change in the number of bytes, positive or negative,
2618 * requested for the if_data array.
2630 if (byte_diff
== 0) {
2634 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2635 new_size
= (int)ifp
->if_bytes
+ byte_diff
;
2636 ASSERT(new_size
>= 0);
2638 if (new_size
== 0) {
2639 if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
2640 kmem_free(ifp
->if_u1
.if_data
, ifp
->if_real_bytes
);
2642 ifp
->if_u1
.if_data
= NULL
;
2644 } else if (new_size
<= sizeof(ifp
->if_u2
.if_inline_data
)) {
2646 * If the valid extents/data can fit in if_inline_ext/data,
2647 * copy them from the malloc'd vector and free it.
2649 if (ifp
->if_u1
.if_data
== NULL
) {
2650 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
2651 } else if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
2652 ASSERT(ifp
->if_real_bytes
!= 0);
2653 memcpy(ifp
->if_u2
.if_inline_data
, ifp
->if_u1
.if_data
,
2655 kmem_free(ifp
->if_u1
.if_data
, ifp
->if_real_bytes
);
2656 ifp
->if_u1
.if_data
= ifp
->if_u2
.if_inline_data
;
2661 * Stuck with malloc/realloc.
2662 * For inline data, the underlying buffer must be
2663 * a multiple of 4 bytes in size so that it can be
2664 * logged and stay on word boundaries. We enforce
2667 real_size
= roundup(new_size
, 4);
2668 if (ifp
->if_u1
.if_data
== NULL
) {
2669 ASSERT(ifp
->if_real_bytes
== 0);
2670 ifp
->if_u1
.if_data
= kmem_alloc(real_size
, KM_SLEEP
);
2671 } else if (ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) {
2673 * Only do the realloc if the underlying size
2674 * is really changing.
2676 if (ifp
->if_real_bytes
!= real_size
) {
2677 ifp
->if_u1
.if_data
=
2678 kmem_realloc(ifp
->if_u1
.if_data
,
2684 ASSERT(ifp
->if_real_bytes
== 0);
2685 ifp
->if_u1
.if_data
= kmem_alloc(real_size
, KM_SLEEP
);
2686 memcpy(ifp
->if_u1
.if_data
, ifp
->if_u2
.if_inline_data
,
2690 ifp
->if_real_bytes
= real_size
;
2691 ifp
->if_bytes
= new_size
;
2692 ASSERT(ifp
->if_bytes
<= XFS_IFORK_SIZE(ip
, whichfork
));
2699 * Map inode to disk block and offset.
2701 * mp -- the mount point structure for the current file system
2702 * tp -- the current transaction
2703 * ino -- the inode number of the inode to be located
2704 * imap -- this structure is filled in with the information necessary
2705 * to retrieve the given inode from disk
2706 * flags -- flags to pass to xfs_dilocate indicating whether or not
2707 * lookups in the inode btree were OK or not
2717 xfs_fsblock_t fsbno
;
2722 fsbno
= imap
->im_blkno
?
2723 XFS_DADDR_TO_FSB(mp
, imap
->im_blkno
) : NULLFSBLOCK
;
2724 error
= xfs_dilocate(mp
, tp
, ino
, &fsbno
, &len
, &off
, flags
);
2728 imap
->im_blkno
= XFS_FSB_TO_DADDR(mp
, fsbno
);
2729 imap
->im_len
= XFS_FSB_TO_BB(mp
, len
);
2730 imap
->im_agblkno
= XFS_FSB_TO_AGBNO(mp
, fsbno
);
2731 imap
->im_ioffset
= (ushort
)off
;
2732 imap
->im_boffset
= (ushort
)(off
<< mp
->m_sb
.sb_inodelog
);
2743 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2744 if (ifp
->if_broot
!= NULL
) {
2745 kmem_free(ifp
->if_broot
, ifp
->if_broot_bytes
);
2746 ifp
->if_broot
= NULL
;
2750 * If the format is local, then we can't have an extents
2751 * array so just look for an inline data array. If we're
2752 * not local then we may or may not have an extents list,
2753 * so check and free it up if we do.
2755 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
) {
2756 if ((ifp
->if_u1
.if_data
!= ifp
->if_u2
.if_inline_data
) &&
2757 (ifp
->if_u1
.if_data
!= NULL
)) {
2758 ASSERT(ifp
->if_real_bytes
!= 0);
2759 kmem_free(ifp
->if_u1
.if_data
, ifp
->if_real_bytes
);
2760 ifp
->if_u1
.if_data
= NULL
;
2761 ifp
->if_real_bytes
= 0;
2763 } else if ((ifp
->if_flags
& XFS_IFEXTENTS
) &&
2764 (ifp
->if_u1
.if_extents
!= NULL
) &&
2765 (ifp
->if_u1
.if_extents
!= ifp
->if_u2
.if_inline_ext
)) {
2766 ASSERT(ifp
->if_real_bytes
!= 0);
2767 kmem_free(ifp
->if_u1
.if_extents
, ifp
->if_real_bytes
);
2768 ifp
->if_u1
.if_extents
= NULL
;
2769 ifp
->if_real_bytes
= 0;
2771 ASSERT(ifp
->if_u1
.if_extents
== NULL
||
2772 ifp
->if_u1
.if_extents
== ifp
->if_u2
.if_inline_ext
);
2773 ASSERT(ifp
->if_real_bytes
== 0);
2774 if (whichfork
== XFS_ATTR_FORK
) {
2775 kmem_zone_free(xfs_ifork_zone
, ip
->i_afp
);
2781 * This is called free all the memory associated with an inode.
2782 * It must free the inode itself and any buffers allocated for
2783 * if_extents/if_data and if_broot. It must also free the lock
2784 * associated with the inode.
2791 switch (ip
->i_d
.di_mode
& S_IFMT
) {
2795 xfs_idestroy_fork(ip
, XFS_DATA_FORK
);
2799 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
2800 mrfree(&ip
->i_lock
);
2801 mrfree(&ip
->i_iolock
);
2802 freesema(&ip
->i_flock
);
2803 #ifdef XFS_BMAP_TRACE
2804 ktrace_free(ip
->i_xtrace
);
2806 #ifdef XFS_BMBT_TRACE
2807 ktrace_free(ip
->i_btrace
);
2810 ktrace_free(ip
->i_rwtrace
);
2812 #ifdef XFS_ILOCK_TRACE
2813 ktrace_free(ip
->i_lock_trace
);
2815 #ifdef XFS_DIR2_TRACE
2816 ktrace_free(ip
->i_dir_trace
);
2819 /* XXXdpd should be able to assert this but shutdown
2820 * is leaving the AIL behind. */
2821 ASSERT(((ip
->i_itemp
->ili_item
.li_flags
& XFS_LI_IN_AIL
) == 0) ||
2822 XFS_FORCED_SHUTDOWN(ip
->i_mount
));
2823 xfs_inode_item_destroy(ip
);
2825 kmem_zone_free(xfs_inode_zone
, ip
);
2830 * Increment the pin count of the given buffer.
2831 * This value is protected by ipinlock spinlock in the mount structure.
2837 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
));
2839 atomic_inc(&ip
->i_pincount
);
2843 * Decrement the pin count of the given inode, and wake up
2844 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2845 * inode must have been previoulsy pinned with a call to xfs_ipin().
2851 ASSERT(atomic_read(&ip
->i_pincount
) > 0);
2853 if (atomic_dec_and_test(&ip
->i_pincount
)) {
2854 vnode_t
*vp
= XFS_ITOV_NULL(ip
);
2856 /* make sync come back and flush this inode */
2858 struct inode
*inode
= LINVFS_GET_IP(vp
);
2860 if (!(inode
->i_state
& I_NEW
))
2861 mark_inode_dirty_sync(inode
);
2864 wake_up(&ip
->i_ipin_wait
);
2869 * This is called to wait for the given inode to be unpinned.
2870 * It will sleep until this happens. The caller must have the
2871 * inode locked in at least shared mode so that the buffer cannot
2872 * be subsequently pinned once someone is waiting for it to be
2879 xfs_inode_log_item_t
*iip
;
2882 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
| MR_ACCESS
));
2884 if (atomic_read(&ip
->i_pincount
) == 0) {
2889 if (iip
&& iip
->ili_last_lsn
) {
2890 lsn
= iip
->ili_last_lsn
;
2896 * Give the log a push so we don't wait here too long.
2898 xfs_log_force(ip
->i_mount
, lsn
, XFS_LOG_FORCE
);
2900 wait_event(ip
->i_ipin_wait
, (atomic_read(&ip
->i_pincount
) == 0));
2905 * xfs_iextents_copy()
2907 * This is called to copy the REAL extents (as opposed to the delayed
2908 * allocation extents) from the inode into the given buffer. It
2909 * returns the number of bytes copied into the buffer.
2911 * If there are no delayed allocation extents, then we can just
2912 * memcpy() the extents into the buffer. Otherwise, we need to
2913 * examine each extent in turn and skip those which are delayed.
2918 xfs_bmbt_rec_t
*buffer
,
2922 xfs_bmbt_rec_t
*dest_ep
;
2924 #ifdef XFS_BMAP_TRACE
2925 static char fname
[] = "xfs_iextents_copy";
2930 xfs_fsblock_t start_block
;
2932 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2933 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
|MR_ACCESS
));
2934 ASSERT(ifp
->if_bytes
> 0);
2936 nrecs
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
2937 xfs_bmap_trace_exlist(fname
, ip
, nrecs
, whichfork
);
2941 * There are some delayed allocation extents in the
2942 * inode, so copy the extents one at a time and skip
2943 * the delayed ones. There must be at least one
2944 * non-delayed extent.
2946 ep
= ifp
->if_u1
.if_extents
;
2949 for (i
= 0; i
< nrecs
; i
++) {
2950 start_block
= xfs_bmbt_get_startblock(ep
);
2951 if (ISNULLSTARTBLOCK(start_block
)) {
2953 * It's a delayed allocation extent, so skip it.
2959 /* Translate to on disk format */
2960 put_unaligned(INT_GET(ep
->l0
, ARCH_CONVERT
),
2961 (__uint64_t
*)&dest_ep
->l0
);
2962 put_unaligned(INT_GET(ep
->l1
, ARCH_CONVERT
),
2963 (__uint64_t
*)&dest_ep
->l1
);
2968 ASSERT(copied
!= 0);
2969 xfs_validate_extents(buffer
, copied
, 1, XFS_EXTFMT_INODE(ip
));
2971 return (copied
* (uint
)sizeof(xfs_bmbt_rec_t
));
2975 * Each of the following cases stores data into the same region
2976 * of the on-disk inode, so only one of them can be valid at
2977 * any given time. While it is possible to have conflicting formats
2978 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2979 * in EXTENTS format, this can only happen when the fork has
2980 * changed formats after being modified but before being flushed.
2981 * In these cases, the format always takes precedence, because the
2982 * format indicates the current state of the fork.
2989 xfs_inode_log_item_t
*iip
,
2996 #ifdef XFS_TRANS_DEBUG
2999 static const short brootflag
[2] =
3000 { XFS_ILOG_DBROOT
, XFS_ILOG_ABROOT
};
3001 static const short dataflag
[2] =
3002 { XFS_ILOG_DDATA
, XFS_ILOG_ADATA
};
3003 static const short extflag
[2] =
3004 { XFS_ILOG_DEXT
, XFS_ILOG_AEXT
};
3008 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
3010 * This can happen if we gave up in iformat in an error path,
3011 * for the attribute fork.
3014 ASSERT(whichfork
== XFS_ATTR_FORK
);
3017 cp
= XFS_DFORK_PTR_ARCH(dip
, whichfork
, ARCH_CONVERT
);
3019 switch (XFS_IFORK_FORMAT(ip
, whichfork
)) {
3020 case XFS_DINODE_FMT_LOCAL
:
3021 if ((iip
->ili_format
.ilf_fields
& dataflag
[whichfork
]) &&
3022 (ifp
->if_bytes
> 0)) {
3023 ASSERT(ifp
->if_u1
.if_data
!= NULL
);
3024 ASSERT(ifp
->if_bytes
<= XFS_IFORK_SIZE(ip
, whichfork
));
3025 memcpy(cp
, ifp
->if_u1
.if_data
, ifp
->if_bytes
);
3027 if (whichfork
== XFS_DATA_FORK
) {
3028 if (unlikely(XFS_DIR_SHORTFORM_VALIDATE_ONDISK(mp
, dip
))) {
3029 XFS_ERROR_REPORT("xfs_iflush_fork",
3030 XFS_ERRLEVEL_LOW
, mp
);
3031 return XFS_ERROR(EFSCORRUPTED
);
3036 case XFS_DINODE_FMT_EXTENTS
:
3037 ASSERT((ifp
->if_flags
& XFS_IFEXTENTS
) ||
3038 !(iip
->ili_format
.ilf_fields
& extflag
[whichfork
]));
3039 ASSERT((ifp
->if_u1
.if_extents
!= NULL
) || (ifp
->if_bytes
== 0));
3040 ASSERT((ifp
->if_u1
.if_extents
== NULL
) || (ifp
->if_bytes
> 0));
3041 if ((iip
->ili_format
.ilf_fields
& extflag
[whichfork
]) &&
3042 (ifp
->if_bytes
> 0)) {
3043 ASSERT(XFS_IFORK_NEXTENTS(ip
, whichfork
) > 0);
3044 (void)xfs_iextents_copy(ip
, (xfs_bmbt_rec_t
*)cp
,
3049 case XFS_DINODE_FMT_BTREE
:
3050 if ((iip
->ili_format
.ilf_fields
& brootflag
[whichfork
]) &&
3051 (ifp
->if_broot_bytes
> 0)) {
3052 ASSERT(ifp
->if_broot
!= NULL
);
3053 ASSERT(ifp
->if_broot_bytes
<=
3054 (XFS_IFORK_SIZE(ip
, whichfork
) +
3055 XFS_BROOT_SIZE_ADJ
));
3056 xfs_bmbt_to_bmdr(ifp
->if_broot
, ifp
->if_broot_bytes
,
3057 (xfs_bmdr_block_t
*)cp
,
3058 XFS_DFORK_SIZE_ARCH(dip
, mp
, whichfork
, ARCH_CONVERT
));
3062 case XFS_DINODE_FMT_DEV
:
3063 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_DEV
) {
3064 ASSERT(whichfork
== XFS_DATA_FORK
);
3065 INT_SET(dip
->di_u
.di_dev
, ARCH_CONVERT
, ip
->i_df
.if_u2
.if_rdev
);
3069 case XFS_DINODE_FMT_UUID
:
3070 if (iip
->ili_format
.ilf_fields
& XFS_ILOG_UUID
) {
3071 ASSERT(whichfork
== XFS_DATA_FORK
);
3072 memcpy(&dip
->di_u
.di_muuid
, &ip
->i_df
.if_u2
.if_uuid
,
3086 * xfs_iflush() will write a modified inode's changes out to the
3087 * inode's on disk home. The caller must have the inode lock held
3088 * in at least shared mode and the inode flush semaphore must be
3089 * held as well. The inode lock will still be held upon return from
3090 * the call and the caller is free to unlock it.
3091 * The inode flush lock will be unlocked when the inode reaches the disk.
3092 * The flags indicate how the inode's buffer should be written out.
3099 xfs_inode_log_item_t
*iip
;
3107 int clcount
; /* count of inodes clustered */
3109 enum { INT_DELWRI
= (1 << 0), INT_ASYNC
= (1 << 1) };
3112 XFS_STATS_INC(xs_iflush_count
);
3114 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
|MR_ACCESS
));
3115 ASSERT(valusema(&ip
->i_flock
) <= 0);
3116 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
3117 ip
->i_d
.di_nextents
> ip
->i_df
.if_ext_max
);
3123 * If the inode isn't dirty, then just release the inode
3124 * flush lock and do nothing.
3126 if ((ip
->i_update_core
== 0) &&
3127 ((iip
== NULL
) || !(iip
->ili_format
.ilf_fields
& XFS_ILOG_ALL
))) {
3128 ASSERT((iip
!= NULL
) ?
3129 !(iip
->ili_item
.li_flags
& XFS_LI_IN_AIL
) : 1);
3135 * We can't flush the inode until it is unpinned, so
3136 * wait for it. We know noone new can pin it, because
3137 * we are holding the inode lock shared and you need
3138 * to hold it exclusively to pin the inode.
3140 xfs_iunpin_wait(ip
);
3143 * This may have been unpinned because the filesystem is shutting
3144 * down forcibly. If that's the case we must not write this inode
3145 * to disk, because the log record didn't make it to disk!
3147 if (XFS_FORCED_SHUTDOWN(mp
)) {
3148 ip
->i_update_core
= 0;
3150 iip
->ili_format
.ilf_fields
= 0;
3152 return XFS_ERROR(EIO
);
3156 * Get the buffer containing the on-disk inode.
3158 error
= xfs_itobp(mp
, NULL
, ip
, &dip
, &bp
, 0);
3165 * Decide how buffer will be flushed out. This is done before
3166 * the call to xfs_iflush_int because this field is zeroed by it.
3168 if (iip
!= NULL
&& iip
->ili_format
.ilf_fields
!= 0) {
3170 * Flush out the inode buffer according to the directions
3171 * of the caller. In the cases where the caller has given
3172 * us a choice choose the non-delwri case. This is because
3173 * the inode is in the AIL and we need to get it out soon.
3176 case XFS_IFLUSH_SYNC
:
3177 case XFS_IFLUSH_DELWRI_ELSE_SYNC
:
3180 case XFS_IFLUSH_ASYNC
:
3181 case XFS_IFLUSH_DELWRI_ELSE_ASYNC
:
3184 case XFS_IFLUSH_DELWRI
:
3194 case XFS_IFLUSH_DELWRI_ELSE_SYNC
:
3195 case XFS_IFLUSH_DELWRI_ELSE_ASYNC
:
3196 case XFS_IFLUSH_DELWRI
:
3199 case XFS_IFLUSH_ASYNC
:
3202 case XFS_IFLUSH_SYNC
:
3213 * First flush out the inode that xfs_iflush was called with.
3215 error
= xfs_iflush_int(ip
, bp
);
3222 * see if other inodes can be gathered into this write
3225 ip
->i_chash
->chl_buf
= bp
;
3227 ch
= XFS_CHASH(mp
, ip
->i_blkno
);
3228 s
= mutex_spinlock(&ch
->ch_lock
);
3231 for (iq
= ip
->i_cnext
; iq
!= ip
; iq
= iq
->i_cnext
) {
3233 * Do an un-protected check to see if the inode is dirty and
3234 * is a candidate for flushing. These checks will be repeated
3235 * later after the appropriate locks are acquired.
3238 if ((iq
->i_update_core
== 0) &&
3240 !(iip
->ili_format
.ilf_fields
& XFS_ILOG_ALL
)) &&
3241 xfs_ipincount(iq
) == 0) {
3246 * Try to get locks. If any are unavailable,
3247 * then this inode cannot be flushed and is skipped.
3250 /* get inode locks (just i_lock) */
3251 if (xfs_ilock_nowait(iq
, XFS_ILOCK_SHARED
)) {
3252 /* get inode flush lock */
3253 if (xfs_iflock_nowait(iq
)) {
3254 /* check if pinned */
3255 if (xfs_ipincount(iq
) == 0) {
3256 /* arriving here means that
3257 * this inode can be flushed.
3258 * first re-check that it's
3262 if ((iq
->i_update_core
!= 0)||
3264 (iip
->ili_format
.ilf_fields
& XFS_ILOG_ALL
))) {
3266 error
= xfs_iflush_int(iq
, bp
);
3270 goto cluster_corrupt_out
;
3279 xfs_iunlock(iq
, XFS_ILOCK_SHARED
);
3282 mutex_spinunlock(&ch
->ch_lock
, s
);
3285 XFS_STATS_INC(xs_icluster_flushcnt
);
3286 XFS_STATS_ADD(xs_icluster_flushinode
, clcount
);
3290 * If the buffer is pinned then push on the log so we won't
3291 * get stuck waiting in the write for too long.
3293 if (XFS_BUF_ISPINNED(bp
)){
3294 xfs_log_force(mp
, (xfs_lsn_t
)0, XFS_LOG_FORCE
);
3297 if (flags
& INT_DELWRI
) {
3298 xfs_bdwrite(mp
, bp
);
3299 } else if (flags
& INT_ASYNC
) {
3300 xfs_bawrite(mp
, bp
);
3302 error
= xfs_bwrite(mp
, bp
);
3308 xfs_force_shutdown(mp
, XFS_CORRUPT_INCORE
);
3309 xfs_iflush_abort(ip
);
3311 * Unlocks the flush lock
3313 return XFS_ERROR(EFSCORRUPTED
);
3315 cluster_corrupt_out
:
3316 /* Corruption detected in the clustering loop. Invalidate the
3317 * inode buffer and shut down the filesystem.
3319 mutex_spinunlock(&ch
->ch_lock
, s
);
3322 * Clean up the buffer. If it was B_DELWRI, just release it --
3323 * brelse can handle it with no problems. If not, shut down the
3324 * filesystem before releasing the buffer.
3326 if ((bufwasdelwri
= XFS_BUF_ISDELAYWRITE(bp
))) {
3330 xfs_force_shutdown(mp
, XFS_CORRUPT_INCORE
);
3334 * Just like incore_relse: if we have b_iodone functions,
3335 * mark the buffer as an error and call them. Otherwise
3336 * mark it as stale and brelse.
3338 if (XFS_BUF_IODONE_FUNC(bp
)) {
3339 XFS_BUF_CLR_BDSTRAT_FUNC(bp
);
3343 XFS_BUF_ERROR(bp
,EIO
);
3351 xfs_iflush_abort(iq
);
3353 * Unlocks the flush lock
3355 return XFS_ERROR(EFSCORRUPTED
);
3364 xfs_inode_log_item_t
*iip
;
3367 #ifdef XFS_TRANS_DEBUG
3372 ASSERT(ismrlocked(&ip
->i_lock
, MR_UPDATE
|MR_ACCESS
));
3373 ASSERT(valusema(&ip
->i_flock
) <= 0);
3374 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
3375 ip
->i_d
.di_nextents
> ip
->i_df
.if_ext_max
);
3382 * If the inode isn't dirty, then just release the inode
3383 * flush lock and do nothing.
3385 if ((ip
->i_update_core
== 0) &&
3386 ((iip
== NULL
) || !(iip
->ili_format
.ilf_fields
& XFS_ILOG_ALL
))) {
3391 /* set *dip = inode's place in the buffer */
3392 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, ip
->i_boffset
);
3395 * Clear i_update_core before copying out the data.
3396 * This is for coordination with our timestamp updates
3397 * that don't hold the inode lock. They will always
3398 * update the timestamps BEFORE setting i_update_core,
3399 * so if we clear i_update_core after they set it we
3400 * are guaranteed to see their updates to the timestamps.
3401 * I believe that this depends on strongly ordered memory
3402 * semantics, but we have that. We use the SYNCHRONIZE
3403 * macro to make sure that the compiler does not reorder
3404 * the i_update_core access below the data copy below.
3406 ip
->i_update_core
= 0;
3409 if (XFS_TEST_ERROR(INT_GET(dip
->di_core
.di_magic
,ARCH_CONVERT
) != XFS_DINODE_MAGIC
,
3410 mp
, XFS_ERRTAG_IFLUSH_1
, XFS_RANDOM_IFLUSH_1
)) {
3411 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3412 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3413 ip
->i_ino
, (int) INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
), dip
);
3416 if (XFS_TEST_ERROR(ip
->i_d
.di_magic
!= XFS_DINODE_MAGIC
,
3417 mp
, XFS_ERRTAG_IFLUSH_2
, XFS_RANDOM_IFLUSH_2
)) {
3418 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3419 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3420 ip
->i_ino
, ip
, ip
->i_d
.di_magic
);
3423 if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
) {
3425 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3426 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
),
3427 mp
, XFS_ERRTAG_IFLUSH_3
, XFS_RANDOM_IFLUSH_3
)) {
3428 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3429 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3433 } else if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFDIR
) {
3435 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3436 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
) &&
3437 (ip
->i_d
.di_format
!= XFS_DINODE_FMT_LOCAL
),
3438 mp
, XFS_ERRTAG_IFLUSH_4
, XFS_RANDOM_IFLUSH_4
)) {
3439 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3440 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3445 if (XFS_TEST_ERROR(ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
>
3446 ip
->i_d
.di_nblocks
, mp
, XFS_ERRTAG_IFLUSH_5
,
3447 XFS_RANDOM_IFLUSH_5
)) {
3448 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3449 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3451 ip
->i_d
.di_nextents
+ ip
->i_d
.di_anextents
,
3456 if (XFS_TEST_ERROR(ip
->i_d
.di_forkoff
> mp
->m_sb
.sb_inodesize
,
3457 mp
, XFS_ERRTAG_IFLUSH_6
, XFS_RANDOM_IFLUSH_6
)) {
3458 xfs_cmn_err(XFS_PTAG_IFLUSH
, CE_ALERT
, mp
,
3459 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3460 ip
->i_ino
, ip
->i_d
.di_forkoff
, ip
);
3464 * bump the flush iteration count, used to detect flushes which
3465 * postdate a log record during recovery.
3468 ip
->i_d
.di_flushiter
++;
3471 * Copy the dirty parts of the inode into the on-disk
3472 * inode. We always copy out the core of the inode,
3473 * because if the inode is dirty at all the core must
3476 xfs_xlate_dinode_core((xfs_caddr_t
)&(dip
->di_core
), &(ip
->i_d
),
3479 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3480 if (ip
->i_d
.di_flushiter
== DI_MAX_FLUSH
)
3481 ip
->i_d
.di_flushiter
= 0;
3484 * If this is really an old format inode and the superblock version
3485 * has not been updated to support only new format inodes, then
3486 * convert back to the old inode format. If the superblock version
3487 * has been updated, then make the conversion permanent.
3489 ASSERT(ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
||
3490 XFS_SB_VERSION_HASNLINK(&mp
->m_sb
));
3491 if (ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
) {
3492 if (!XFS_SB_VERSION_HASNLINK(&mp
->m_sb
)) {
3496 ASSERT(ip
->i_d
.di_nlink
<= XFS_MAXLINK_1
);
3497 INT_SET(dip
->di_core
.di_onlink
, ARCH_CONVERT
, ip
->i_d
.di_nlink
);
3500 * The superblock version has already been bumped,
3501 * so just make the conversion to the new inode
3504 ip
->i_d
.di_version
= XFS_DINODE_VERSION_2
;
3505 INT_SET(dip
->di_core
.di_version
, ARCH_CONVERT
, XFS_DINODE_VERSION_2
);
3506 ip
->i_d
.di_onlink
= 0;
3507 INT_ZERO(dip
->di_core
.di_onlink
, ARCH_CONVERT
);
3508 memset(&(ip
->i_d
.di_pad
[0]), 0, sizeof(ip
->i_d
.di_pad
));
3509 memset(&(dip
->di_core
.di_pad
[0]), 0,
3510 sizeof(dip
->di_core
.di_pad
));
3511 ASSERT(ip
->i_d
.di_projid
== 0);
3515 if (xfs_iflush_fork(ip
, dip
, iip
, XFS_DATA_FORK
, bp
) == EFSCORRUPTED
) {
3519 if (XFS_IFORK_Q(ip
)) {
3521 * The only error from xfs_iflush_fork is on the data fork.
3523 (void) xfs_iflush_fork(ip
, dip
, iip
, XFS_ATTR_FORK
, bp
);
3525 xfs_inobp_check(mp
, bp
);
3528 * We've recorded everything logged in the inode, so we'd
3529 * like to clear the ilf_fields bits so we don't log and
3530 * flush things unnecessarily. However, we can't stop
3531 * logging all this information until the data we've copied
3532 * into the disk buffer is written to disk. If we did we might
3533 * overwrite the copy of the inode in the log with all the
3534 * data after re-logging only part of it, and in the face of
3535 * a crash we wouldn't have all the data we need to recover.
3537 * What we do is move the bits to the ili_last_fields field.
3538 * When logging the inode, these bits are moved back to the
3539 * ilf_fields field. In the xfs_iflush_done() routine we
3540 * clear ili_last_fields, since we know that the information
3541 * those bits represent is permanently on disk. As long as
3542 * the flush completes before the inode is logged again, then
3543 * both ilf_fields and ili_last_fields will be cleared.
3545 * We can play with the ilf_fields bits here, because the inode
3546 * lock must be held exclusively in order to set bits there
3547 * and the flush lock protects the ili_last_fields bits.
3548 * Set ili_logged so the flush done
3549 * routine can tell whether or not to look in the AIL.
3550 * Also, store the current LSN of the inode so that we can tell
3551 * whether the item has moved in the AIL from xfs_iflush_done().
3552 * In order to read the lsn we need the AIL lock, because
3553 * it is a 64 bit value that cannot be read atomically.
3555 if (iip
!= NULL
&& iip
->ili_format
.ilf_fields
!= 0) {
3556 iip
->ili_last_fields
= iip
->ili_format
.ilf_fields
;
3557 iip
->ili_format
.ilf_fields
= 0;
3558 iip
->ili_logged
= 1;
3560 ASSERT(sizeof(xfs_lsn_t
) == 8); /* don't lock if it shrinks */
3562 iip
->ili_flush_lsn
= iip
->ili_item
.li_lsn
;
3566 * Attach the function xfs_iflush_done to the inode's
3567 * buffer. This will remove the inode from the AIL
3568 * and unlock the inode's flush lock when the inode is
3569 * completely written to disk.
3571 xfs_buf_attach_iodone(bp
, (void(*)(xfs_buf_t
*,xfs_log_item_t
*))
3572 xfs_iflush_done
, (xfs_log_item_t
*)iip
);
3574 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) != NULL
);
3575 ASSERT(XFS_BUF_IODONE_FUNC(bp
) != NULL
);
3578 * We're flushing an inode which is not in the AIL and has
3579 * not been logged but has i_update_core set. For this
3580 * case we can use a B_DELWRI flush and immediately drop
3581 * the inode flush lock because we can avoid the whole
3582 * AIL state thing. It's OK to drop the flush lock now,
3583 * because we've already locked the buffer and to do anything
3584 * you really need both.
3587 ASSERT(iip
->ili_logged
== 0);
3588 ASSERT(iip
->ili_last_fields
== 0);
3589 ASSERT((iip
->ili_item
.li_flags
& XFS_LI_IN_AIL
) == 0);
3597 return XFS_ERROR(EFSCORRUPTED
);
3602 * Flush all inactive inodes in mp. Return true if no user references
3603 * were found, false otherwise.
3620 XFS_MOUNT_ILOCK(mp
);
3626 /* Make sure we skip markers inserted by sync */
3627 if (ip
->i_mount
== NULL
) {
3633 * It's up to our caller to purge the root
3634 * and quota vnodes later.
3636 vp
= XFS_ITOV_NULL(ip
);
3639 XFS_MOUNT_IUNLOCK(mp
);
3640 xfs_finish_reclaim(ip
, 0, XFS_IFLUSH_ASYNC
);
3645 if (vn_count(vp
) != 0) {
3646 if (vn_count(vp
) == 1 &&
3647 (ip
== mp
->m_rootip
||
3649 (ip
->i_ino
== mp
->m_sb
.sb_uquotino
||
3650 ip
->i_ino
== mp
->m_sb
.sb_gquotino
)))) {
3655 if (!(flag
& XFS_FLUSH_ALL
)) {
3661 * Ignore busy inodes but continue flushing
3668 * Sample vp mapping while holding mp locked on MP
3669 * systems, so we don't purge a reclaimed or
3670 * nonexistent vnode. We break from the loop
3671 * since we know that we modify
3672 * it by pulling ourselves from it in xfs_reclaim()
3673 * called via vn_purge() below. Set ip to the next
3674 * entry in the list anyway so we'll know below
3675 * whether we reached the end or not.
3678 XFS_MOUNT_IUNLOCK(mp
);
3680 vn_purge(vp
, &vmap
);
3684 } while (ip
!= mp
->m_inodes
);
3686 * We need to distinguish between when we exit the loop
3687 * after a purge and when we simply hit the end of the
3688 * list. We can't use the (ip == mp->m_inodes) test,
3689 * because when we purge an inode at the start of the list
3690 * the next inode on the list becomes mp->m_inodes. That
3691 * would cause such a test to bail out early. The purged
3692 * variable tells us how we got out of the loop.
3698 XFS_MOUNT_IUNLOCK(mp
);
3704 * xfs_iaccess: check accessibility of inode for mode.
3713 mode_t orgmode
= mode
;
3714 struct inode
*inode
= LINVFS_GET_IP(XFS_ITOV(ip
));
3716 if (mode
& S_IWUSR
) {
3717 umode_t imode
= inode
->i_mode
;
3719 if (IS_RDONLY(inode
) &&
3720 (S_ISREG(imode
) || S_ISDIR(imode
) || S_ISLNK(imode
)))
3721 return XFS_ERROR(EROFS
);
3723 if (IS_IMMUTABLE(inode
))
3724 return XFS_ERROR(EACCES
);
3728 * If there's an Access Control List it's used instead of
3731 if ((error
= _ACL_XFS_IACCESS(ip
, mode
, cr
)) != -1)
3732 return error
? XFS_ERROR(error
) : 0;
3734 if (current_fsuid(cr
) != ip
->i_d
.di_uid
) {
3736 if (!in_group_p((gid_t
)ip
->i_d
.di_gid
))
3741 * If the DACs are ok we don't need any capability check.
3743 if ((ip
->i_d
.di_mode
& mode
) == mode
)
3746 * Read/write DACs are always overridable.
3747 * Executable DACs are overridable if at least one exec bit is set.
3749 if (!(orgmode
& S_IXUSR
) ||
3750 (inode
->i_mode
& S_IXUGO
) || S_ISDIR(inode
->i_mode
))
3751 if (capable_cred(cr
, CAP_DAC_OVERRIDE
))
3754 if ((orgmode
== S_IRUSR
) ||
3755 (S_ISDIR(inode
->i_mode
) && (!(orgmode
& S_IWUSR
)))) {
3756 if (capable_cred(cr
, CAP_DAC_READ_SEARCH
))
3759 cmn_err(CE_NOTE
, "Ick: mode=%o, orgmode=%o", mode
, orgmode
);
3761 return XFS_ERROR(EACCES
);
3763 return XFS_ERROR(EACCES
);
3767 * xfs_iroundup: round up argument to next power of two
3776 if ((v
& (v
- 1)) == 0)
3778 ASSERT((v
& 0x80000000) == 0);
3779 if ((v
& (v
+ 1)) == 0)
3781 for (i
= 0, m
= 1; i
< 31; i
++, m
<<= 1) {
3785 if ((v
& (v
+ 1)) == 0)
3793 * Change the requested timestamp in the given inode.
3794 * We don't lock across timestamp updates, and we don't log them but
3795 * we do record the fact that there is dirty information in core.
3797 * NOTE -- callers MUST combine XFS_ICHGTIME_MOD or XFS_ICHGTIME_CHG
3798 * with XFS_ICHGTIME_ACC to be sure that access time
3799 * update will take. Calling first with XFS_ICHGTIME_ACC
3800 * and then XFS_ICHGTIME_MOD may fail to modify the access
3801 * timestamp if the filesystem is mounted noacctm.
3804 xfs_ichgtime(xfs_inode_t
*ip
,
3808 vnode_t
*vp
= XFS_ITOV(ip
);
3809 struct inode
*inode
= LINVFS_GET_IP(vp
);
3812 * We're not supposed to change timestamps in readonly-mounted
3813 * filesystems. Throw it away if anyone asks us.
3815 if (unlikely(vp
->v_vfsp
->vfs_flag
& VFS_RDONLY
))
3819 * Don't update access timestamps on reads if mounted "noatime"
3820 * Throw it away if anyone asks us.
3822 if ((ip
->i_mount
->m_flags
& XFS_MOUNT_NOATIME
|| IS_NOATIME(inode
)) &&
3823 ((flags
& (XFS_ICHGTIME_ACC
|XFS_ICHGTIME_MOD
|XFS_ICHGTIME_CHG
))
3824 == XFS_ICHGTIME_ACC
))
3828 if (flags
& XFS_ICHGTIME_MOD
) {
3829 VN_MTIMESET(vp
, &tv
);
3830 ip
->i_d
.di_mtime
.t_sec
= (__int32_t
)tv
.tv_sec
;
3831 ip
->i_d
.di_mtime
.t_nsec
= (__int32_t
)tv
.tv_nsec
;
3833 if (flags
& XFS_ICHGTIME_ACC
) {
3834 VN_ATIMESET(vp
, &tv
);
3835 ip
->i_d
.di_atime
.t_sec
= (__int32_t
)tv
.tv_sec
;
3836 ip
->i_d
.di_atime
.t_nsec
= (__int32_t
)tv
.tv_nsec
;
3838 if (flags
& XFS_ICHGTIME_CHG
) {
3839 VN_CTIMESET(vp
, &tv
);
3840 ip
->i_d
.di_ctime
.t_sec
= (__int32_t
)tv
.tv_sec
;
3841 ip
->i_d
.di_ctime
.t_nsec
= (__int32_t
)tv
.tv_nsec
;
3845 * We update the i_update_core field _after_ changing
3846 * the timestamps in order to coordinate properly with
3847 * xfs_iflush() so that we don't lose timestamp updates.
3848 * This keeps us from having to hold the inode lock
3849 * while doing this. We use the SYNCHRONIZE macro to
3850 * ensure that the compiler does not reorder the update
3851 * of i_update_core above the timestamp updates above.
3854 ip
->i_update_core
= 1;
3855 if (!(inode
->i_state
& I_LOCK
))
3856 mark_inode_dirty_sync(inode
);
3859 #ifdef XFS_ILOCK_TRACE
3860 ktrace_t
*xfs_ilock_trace_buf
;
3863 xfs_ilock_trace(xfs_inode_t
*ip
, int lock
, unsigned int lockflags
, inst_t
*ra
)
3865 ktrace_enter(ip
->i_lock_trace
,
3867 (void *)(unsigned long)lock
, /* 1 = LOCK, 3=UNLOCK, etc */
3868 (void *)(unsigned long)lockflags
, /* XFS_ILOCK_EXCL etc */
3869 (void *)ra
, /* caller of ilock */
3870 (void *)(unsigned long)current_cpu(),
3871 (void *)(unsigned long)current_pid(),
3872 NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
,NULL
);