2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
28 #include "xfs_mount.h"
29 #include "xfs_da_format.h"
30 #include "xfs_da_btree.h"
32 #include "xfs_inode.h"
33 #include "xfs_btree.h"
34 #include "xfs_trans.h"
35 #include "xfs_inode_item.h"
36 #include "xfs_extfree_item.h"
37 #include "xfs_alloc.h"
39 #include "xfs_bmap_util.h"
40 #include "xfs_bmap_btree.h"
41 #include "xfs_rtalloc.h"
42 #include "xfs_error.h"
43 #include "xfs_quota.h"
44 #include "xfs_trans_space.h"
45 #include "xfs_buf_item.h"
46 #include "xfs_trace.h"
47 #include "xfs_symlink.h"
48 #include "xfs_attr_leaf.h"
49 #include "xfs_dinode.h"
50 #include "xfs_filestream.h"
53 kmem_zone_t
*xfs_bmap_free_item_zone
;
56 * Miscellaneous helper functions
60 * Compute and fill in the value of the maximum depth of a bmap btree
61 * in this filesystem. Done once, during mount.
64 xfs_bmap_compute_maxlevels(
65 xfs_mount_t
*mp
, /* file system mount structure */
66 int whichfork
) /* data or attr fork */
68 int level
; /* btree level */
69 uint maxblocks
; /* max blocks at this level */
70 uint maxleafents
; /* max leaf entries possible */
71 int maxrootrecs
; /* max records in root block */
72 int minleafrecs
; /* min records in leaf block */
73 int minnoderecs
; /* min records in node block */
74 int sz
; /* root block size */
77 * The maximum number of extents in a file, hence the maximum
78 * number of leaf entries, is controlled by the type of di_nextents
79 * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
80 * (a signed 16-bit number, xfs_aextnum_t).
82 * Note that we can no longer assume that if we are in ATTR1 that
83 * the fork offset of all the inodes will be
84 * (xfs_default_attroffset(ip) >> 3) because we could have mounted
85 * with ATTR2 and then mounted back with ATTR1, keeping the
86 * di_forkoff's fixed but probably at various positions. Therefore,
87 * for both ATTR1 and ATTR2 we have to assume the worst case scenario
88 * of a minimum size available.
90 if (whichfork
== XFS_DATA_FORK
) {
91 maxleafents
= MAXEXTNUM
;
92 sz
= XFS_BMDR_SPACE_CALC(MINDBTPTRS
);
94 maxleafents
= MAXAEXTNUM
;
95 sz
= XFS_BMDR_SPACE_CALC(MINABTPTRS
);
97 maxrootrecs
= xfs_bmdr_maxrecs(sz
, 0);
98 minleafrecs
= mp
->m_bmap_dmnr
[0];
99 minnoderecs
= mp
->m_bmap_dmnr
[1];
100 maxblocks
= (maxleafents
+ minleafrecs
- 1) / minleafrecs
;
101 for (level
= 1; maxblocks
> 1; level
++) {
102 if (maxblocks
<= maxrootrecs
)
105 maxblocks
= (maxblocks
+ minnoderecs
- 1) / minnoderecs
;
107 mp
->m_bm_maxlevels
[whichfork
] = level
;
110 STATIC
int /* error */
112 struct xfs_btree_cur
*cur
,
116 int *stat
) /* success/failure */
118 cur
->bc_rec
.b
.br_startoff
= off
;
119 cur
->bc_rec
.b
.br_startblock
= bno
;
120 cur
->bc_rec
.b
.br_blockcount
= len
;
121 return xfs_btree_lookup(cur
, XFS_LOOKUP_EQ
, stat
);
124 STATIC
int /* error */
126 struct xfs_btree_cur
*cur
,
130 int *stat
) /* success/failure */
132 cur
->bc_rec
.b
.br_startoff
= off
;
133 cur
->bc_rec
.b
.br_startblock
= bno
;
134 cur
->bc_rec
.b
.br_blockcount
= len
;
135 return xfs_btree_lookup(cur
, XFS_LOOKUP_GE
, stat
);
139 * Check if the inode needs to be converted to btree format.
141 static inline bool xfs_bmap_needs_btree(struct xfs_inode
*ip
, int whichfork
)
143 return XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_EXTENTS
&&
144 XFS_IFORK_NEXTENTS(ip
, whichfork
) >
145 XFS_IFORK_MAXEXT(ip
, whichfork
);
149 * Check if the inode should be converted to extent format.
151 static inline bool xfs_bmap_wants_extents(struct xfs_inode
*ip
, int whichfork
)
153 return XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_BTREE
&&
154 XFS_IFORK_NEXTENTS(ip
, whichfork
) <=
155 XFS_IFORK_MAXEXT(ip
, whichfork
);
159 * Update the record referred to by cur to the value given
160 * by [off, bno, len, state].
161 * This either works (return 0) or gets an EFSCORRUPTED error.
165 struct xfs_btree_cur
*cur
,
171 union xfs_btree_rec rec
;
173 xfs_bmbt_disk_set_allf(&rec
.bmbt
, off
, bno
, len
, state
);
174 return xfs_btree_update(cur
, &rec
);
178 * Compute the worst-case number of indirect blocks that will be used
179 * for ip's delayed extent of length "len".
182 xfs_bmap_worst_indlen(
183 xfs_inode_t
*ip
, /* incore inode pointer */
184 xfs_filblks_t len
) /* delayed extent length */
186 int level
; /* btree level number */
187 int maxrecs
; /* maximum record count at this level */
188 xfs_mount_t
*mp
; /* mount structure */
189 xfs_filblks_t rval
; /* return value */
192 maxrecs
= mp
->m_bmap_dmxr
[0];
193 for (level
= 0, rval
= 0;
194 level
< XFS_BM_MAXLEVELS(mp
, XFS_DATA_FORK
);
197 do_div(len
, maxrecs
);
200 return rval
+ XFS_BM_MAXLEVELS(mp
, XFS_DATA_FORK
) -
203 maxrecs
= mp
->m_bmap_dmxr
[1];
209 * Calculate the default attribute fork offset for newly created inodes.
212 xfs_default_attroffset(
213 struct xfs_inode
*ip
)
215 struct xfs_mount
*mp
= ip
->i_mount
;
218 if (mp
->m_sb
.sb_inodesize
== 256) {
219 offset
= XFS_LITINO(mp
, ip
->i_d
.di_version
) -
220 XFS_BMDR_SPACE_CALC(MINABTPTRS
);
222 offset
= XFS_BMDR_SPACE_CALC(6 * MINABTPTRS
);
225 ASSERT(offset
< XFS_LITINO(mp
, ip
->i_d
.di_version
));
230 * Helper routine to reset inode di_forkoff field when switching
231 * attribute fork from local to extent format - we reset it where
232 * possible to make space available for inline data fork extents.
235 xfs_bmap_forkoff_reset(
239 if (whichfork
== XFS_ATTR_FORK
&&
240 ip
->i_d
.di_format
!= XFS_DINODE_FMT_DEV
&&
241 ip
->i_d
.di_format
!= XFS_DINODE_FMT_UUID
&&
242 ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
) {
243 uint dfl_forkoff
= xfs_default_attroffset(ip
) >> 3;
245 if (dfl_forkoff
> ip
->i_d
.di_forkoff
)
246 ip
->i_d
.di_forkoff
= dfl_forkoff
;
251 * Debug/sanity checking code
255 xfs_bmap_sanity_check(
256 struct xfs_mount
*mp
,
260 struct xfs_btree_block
*block
= XFS_BUF_TO_BLOCK(bp
);
262 if (block
->bb_magic
!= cpu_to_be32(XFS_BMAP_CRC_MAGIC
) &&
263 block
->bb_magic
!= cpu_to_be32(XFS_BMAP_MAGIC
))
266 if (be16_to_cpu(block
->bb_level
) != level
||
267 be16_to_cpu(block
->bb_numrecs
) == 0 ||
268 be16_to_cpu(block
->bb_numrecs
) > mp
->m_bmap_dmxr
[level
!= 0])
275 STATIC
struct xfs_buf
*
277 struct xfs_btree_cur
*cur
,
280 struct xfs_log_item_desc
*lidp
;
286 for (i
= 0; i
< XFS_BTREE_MAXLEVELS
; i
++) {
287 if (!cur
->bc_bufs
[i
])
289 if (XFS_BUF_ADDR(cur
->bc_bufs
[i
]) == bno
)
290 return cur
->bc_bufs
[i
];
293 /* Chase down all the log items to see if the bp is there */
294 list_for_each_entry(lidp
, &cur
->bc_tp
->t_items
, lid_trans
) {
295 struct xfs_buf_log_item
*bip
;
296 bip
= (struct xfs_buf_log_item
*)lidp
->lid_item
;
297 if (bip
->bli_item
.li_type
== XFS_LI_BUF
&&
298 XFS_BUF_ADDR(bip
->bli_buf
) == bno
)
307 struct xfs_btree_block
*block
,
313 __be64
*pp
, *thispa
; /* pointer to block address */
314 xfs_bmbt_key_t
*prevp
, *keyp
;
316 ASSERT(be16_to_cpu(block
->bb_level
) > 0);
319 for( i
= 1; i
<= xfs_btree_get_numrecs(block
); i
++) {
320 dmxr
= mp
->m_bmap_dmxr
[0];
321 keyp
= XFS_BMBT_KEY_ADDR(mp
, block
, i
);
324 ASSERT(be64_to_cpu(prevp
->br_startoff
) <
325 be64_to_cpu(keyp
->br_startoff
));
330 * Compare the block numbers to see if there are dups.
333 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, i
, sz
);
335 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, i
, dmxr
);
337 for (j
= i
+1; j
<= be16_to_cpu(block
->bb_numrecs
); j
++) {
339 thispa
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, j
, sz
);
341 thispa
= XFS_BMBT_PTR_ADDR(mp
, block
, j
, dmxr
);
342 if (*thispa
== *pp
) {
343 xfs_warn(mp
, "%s: thispa(%d) == pp(%d) %Ld",
345 (unsigned long long)be64_to_cpu(*thispa
));
346 panic("%s: ptrs are equal in node\n",
354 * Check that the extents for the inode ip are in the right order in all
359 xfs_bmap_check_leaf_extents(
360 xfs_btree_cur_t
*cur
, /* btree cursor or null */
361 xfs_inode_t
*ip
, /* incore inode pointer */
362 int whichfork
) /* data or attr fork */
364 struct xfs_btree_block
*block
; /* current btree block */
365 xfs_fsblock_t bno
; /* block # of "block" */
366 xfs_buf_t
*bp
; /* buffer for "block" */
367 int error
; /* error return value */
368 xfs_extnum_t i
=0, j
; /* index into the extents list */
369 xfs_ifork_t
*ifp
; /* fork structure */
370 int level
; /* btree level, for checking */
371 xfs_mount_t
*mp
; /* file system mount structure */
372 __be64
*pp
; /* pointer to block address */
373 xfs_bmbt_rec_t
*ep
; /* pointer to current extent */
374 xfs_bmbt_rec_t last
= {0, 0}; /* last extent in prev block */
375 xfs_bmbt_rec_t
*nextp
; /* pointer to next extent */
378 if (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
) {
384 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
385 block
= ifp
->if_broot
;
387 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
389 level
= be16_to_cpu(block
->bb_level
);
391 xfs_check_block(block
, mp
, 1, ifp
->if_broot_bytes
);
392 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, 1, ifp
->if_broot_bytes
);
393 bno
= be64_to_cpu(*pp
);
395 ASSERT(bno
!= NULLDFSBNO
);
396 ASSERT(XFS_FSB_TO_AGNO(mp
, bno
) < mp
->m_sb
.sb_agcount
);
397 ASSERT(XFS_FSB_TO_AGBNO(mp
, bno
) < mp
->m_sb
.sb_agblocks
);
400 * Go down the tree until leaf level is reached, following the first
401 * pointer (leftmost) at each level.
403 while (level
-- > 0) {
404 /* See if buf is in cur first */
406 bp
= xfs_bmap_get_bp(cur
, XFS_FSB_TO_DADDR(mp
, bno
));
409 error
= xfs_btree_read_bufl(mp
, NULL
, bno
, 0, &bp
,
415 block
= XFS_BUF_TO_BLOCK(bp
);
416 XFS_WANT_CORRUPTED_GOTO(
417 xfs_bmap_sanity_check(mp
, bp
, level
),
423 * Check this block for basic sanity (increasing keys and
424 * no duplicate blocks).
427 xfs_check_block(block
, mp
, 0, 0);
428 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, 1, mp
->m_bmap_dmxr
[1]);
429 bno
= be64_to_cpu(*pp
);
430 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp
, bno
), error0
);
433 xfs_trans_brelse(NULL
, bp
);
438 * Here with bp and block set to the leftmost leaf node in the tree.
443 * Loop over all leaf nodes checking that all extents are in the right order.
446 xfs_fsblock_t nextbno
;
447 xfs_extnum_t num_recs
;
450 num_recs
= xfs_btree_get_numrecs(block
);
453 * Read-ahead the next leaf block, if any.
456 nextbno
= be64_to_cpu(block
->bb_u
.l
.bb_rightsib
);
459 * Check all the extents to make sure they are OK.
460 * If we had a previous block, the last entry should
461 * conform with the first entry in this one.
464 ep
= XFS_BMBT_REC_ADDR(mp
, block
, 1);
466 ASSERT(xfs_bmbt_disk_get_startoff(&last
) +
467 xfs_bmbt_disk_get_blockcount(&last
) <=
468 xfs_bmbt_disk_get_startoff(ep
));
470 for (j
= 1; j
< num_recs
; j
++) {
471 nextp
= XFS_BMBT_REC_ADDR(mp
, block
, j
+ 1);
472 ASSERT(xfs_bmbt_disk_get_startoff(ep
) +
473 xfs_bmbt_disk_get_blockcount(ep
) <=
474 xfs_bmbt_disk_get_startoff(nextp
));
482 xfs_trans_brelse(NULL
, bp
);
486 * If we've reached the end, stop.
488 if (bno
== NULLFSBLOCK
)
492 bp
= xfs_bmap_get_bp(cur
, XFS_FSB_TO_DADDR(mp
, bno
));
495 error
= xfs_btree_read_bufl(mp
, NULL
, bno
, 0, &bp
,
501 block
= XFS_BUF_TO_BLOCK(bp
);
505 xfs_trans_brelse(NULL
, bp
);
510 xfs_warn(mp
, "%s: at error0", __func__
);
512 xfs_trans_brelse(NULL
, bp
);
514 xfs_warn(mp
, "%s: BAD after btree leaves for %d extents",
516 panic("%s: CORRUPTED BTREE OR SOMETHING", __func__
);
521 * Add bmap trace insert entries for all the contents of the extent records.
524 xfs_bmap_trace_exlist(
525 xfs_inode_t
*ip
, /* incore inode pointer */
526 xfs_extnum_t cnt
, /* count of entries in the list */
527 int whichfork
, /* data or attr fork */
528 unsigned long caller_ip
)
530 xfs_extnum_t idx
; /* extent record index */
531 xfs_ifork_t
*ifp
; /* inode fork pointer */
534 if (whichfork
== XFS_ATTR_FORK
)
535 state
|= BMAP_ATTRFORK
;
537 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
538 ASSERT(cnt
== (ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
)));
539 for (idx
= 0; idx
< cnt
; idx
++)
540 trace_xfs_extlist(ip
, idx
, whichfork
, caller_ip
);
544 * Validate that the bmbt_irecs being returned from bmapi are valid
545 * given the caller's original parameters. Specifically check the
546 * ranges of the returned irecs to ensure that they only extend beyond
547 * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
550 xfs_bmap_validate_ret(
554 xfs_bmbt_irec_t
*mval
,
558 int i
; /* index to map values */
560 ASSERT(ret_nmap
<= nmap
);
562 for (i
= 0; i
< ret_nmap
; i
++) {
563 ASSERT(mval
[i
].br_blockcount
> 0);
564 if (!(flags
& XFS_BMAPI_ENTIRE
)) {
565 ASSERT(mval
[i
].br_startoff
>= bno
);
566 ASSERT(mval
[i
].br_blockcount
<= len
);
567 ASSERT(mval
[i
].br_startoff
+ mval
[i
].br_blockcount
<=
570 ASSERT(mval
[i
].br_startoff
< bno
+ len
);
571 ASSERT(mval
[i
].br_startoff
+ mval
[i
].br_blockcount
>
575 mval
[i
- 1].br_startoff
+ mval
[i
- 1].br_blockcount
==
576 mval
[i
].br_startoff
);
577 ASSERT(mval
[i
].br_startblock
!= DELAYSTARTBLOCK
&&
578 mval
[i
].br_startblock
!= HOLESTARTBLOCK
);
579 ASSERT(mval
[i
].br_state
== XFS_EXT_NORM
||
580 mval
[i
].br_state
== XFS_EXT_UNWRITTEN
);
585 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork) do { } while (0)
586 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
590 * bmap free list manipulation functions
594 * Add the extent to the list of extents to be free at transaction end.
595 * The list is maintained sorted (by block number).
599 xfs_fsblock_t bno
, /* fs block number of extent */
600 xfs_filblks_t len
, /* length of extent */
601 xfs_bmap_free_t
*flist
, /* list of extents */
602 xfs_mount_t
*mp
) /* mount point structure */
604 xfs_bmap_free_item_t
*cur
; /* current (next) element */
605 xfs_bmap_free_item_t
*new; /* new element */
606 xfs_bmap_free_item_t
*prev
; /* previous element */
611 ASSERT(bno
!= NULLFSBLOCK
);
613 ASSERT(len
<= MAXEXTLEN
);
614 ASSERT(!isnullstartblock(bno
));
615 agno
= XFS_FSB_TO_AGNO(mp
, bno
);
616 agbno
= XFS_FSB_TO_AGBNO(mp
, bno
);
617 ASSERT(agno
< mp
->m_sb
.sb_agcount
);
618 ASSERT(agbno
< mp
->m_sb
.sb_agblocks
);
619 ASSERT(len
< mp
->m_sb
.sb_agblocks
);
620 ASSERT(agbno
+ len
<= mp
->m_sb
.sb_agblocks
);
622 ASSERT(xfs_bmap_free_item_zone
!= NULL
);
623 new = kmem_zone_alloc(xfs_bmap_free_item_zone
, KM_SLEEP
);
624 new->xbfi_startblock
= bno
;
625 new->xbfi_blockcount
= (xfs_extlen_t
)len
;
626 for (prev
= NULL
, cur
= flist
->xbf_first
;
628 prev
= cur
, cur
= cur
->xbfi_next
) {
629 if (cur
->xbfi_startblock
>= bno
)
633 prev
->xbfi_next
= new;
635 flist
->xbf_first
= new;
636 new->xbfi_next
= cur
;
641 * Remove the entry "free" from the free item list. Prev points to the
642 * previous entry, unless "free" is the head of the list.
646 xfs_bmap_free_t
*flist
, /* free item list header */
647 xfs_bmap_free_item_t
*prev
, /* previous item on list, if any */
648 xfs_bmap_free_item_t
*free
) /* list item to be freed */
651 prev
->xbfi_next
= free
->xbfi_next
;
653 flist
->xbf_first
= free
->xbfi_next
;
655 kmem_zone_free(xfs_bmap_free_item_zone
, free
);
659 * Free up any items left in the list.
663 xfs_bmap_free_t
*flist
) /* list of bmap_free_items */
665 xfs_bmap_free_item_t
*free
; /* free list item */
666 xfs_bmap_free_item_t
*next
;
668 if (flist
->xbf_count
== 0)
670 ASSERT(flist
->xbf_first
!= NULL
);
671 for (free
= flist
->xbf_first
; free
; free
= next
) {
672 next
= free
->xbfi_next
;
673 xfs_bmap_del_free(flist
, NULL
, free
);
675 ASSERT(flist
->xbf_count
== 0);
679 * Inode fork format manipulation functions
683 * Transform a btree format file with only one leaf node, where the
684 * extents list will fit in the inode, into an extents format file.
685 * Since the file extents are already in-core, all we have to do is
686 * give up the space for the btree root and pitch the leaf block.
688 STATIC
int /* error */
689 xfs_bmap_btree_to_extents(
690 xfs_trans_t
*tp
, /* transaction pointer */
691 xfs_inode_t
*ip
, /* incore inode pointer */
692 xfs_btree_cur_t
*cur
, /* btree cursor */
693 int *logflagsp
, /* inode logging flags */
694 int whichfork
) /* data or attr fork */
697 struct xfs_btree_block
*cblock
;/* child btree block */
698 xfs_fsblock_t cbno
; /* child block number */
699 xfs_buf_t
*cbp
; /* child block's buffer */
700 int error
; /* error return value */
701 xfs_ifork_t
*ifp
; /* inode fork data */
702 xfs_mount_t
*mp
; /* mount point structure */
703 __be64
*pp
; /* ptr to block address */
704 struct xfs_btree_block
*rblock
;/* root btree block */
707 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
708 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
709 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_BTREE
);
710 rblock
= ifp
->if_broot
;
711 ASSERT(be16_to_cpu(rblock
->bb_level
) == 1);
712 ASSERT(be16_to_cpu(rblock
->bb_numrecs
) == 1);
713 ASSERT(xfs_bmbt_maxrecs(mp
, ifp
->if_broot_bytes
, 0) == 1);
714 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, rblock
, 1, ifp
->if_broot_bytes
);
715 cbno
= be64_to_cpu(*pp
);
718 if ((error
= xfs_btree_check_lptr(cur
, cbno
, 1)))
721 error
= xfs_btree_read_bufl(mp
, tp
, cbno
, 0, &cbp
, XFS_BMAP_BTREE_REF
,
725 cblock
= XFS_BUF_TO_BLOCK(cbp
);
726 if ((error
= xfs_btree_check_block(cur
, cblock
, 0, cbp
)))
728 xfs_bmap_add_free(cbno
, 1, cur
->bc_private
.b
.flist
, mp
);
729 ip
->i_d
.di_nblocks
--;
730 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_BCOUNT
, -1L);
731 xfs_trans_binval(tp
, cbp
);
732 if (cur
->bc_bufs
[0] == cbp
)
733 cur
->bc_bufs
[0] = NULL
;
734 xfs_iroot_realloc(ip
, -1, whichfork
);
735 ASSERT(ifp
->if_broot
== NULL
);
736 ASSERT((ifp
->if_flags
& XFS_IFBROOT
) == 0);
737 XFS_IFORK_FMT_SET(ip
, whichfork
, XFS_DINODE_FMT_EXTENTS
);
738 *logflagsp
= XFS_ILOG_CORE
| xfs_ilog_fext(whichfork
);
743 * Convert an extents-format file into a btree-format file.
744 * The new file will have a root block (in the inode) and a single child block.
746 STATIC
int /* error */
747 xfs_bmap_extents_to_btree(
748 xfs_trans_t
*tp
, /* transaction pointer */
749 xfs_inode_t
*ip
, /* incore inode pointer */
750 xfs_fsblock_t
*firstblock
, /* first-block-allocated */
751 xfs_bmap_free_t
*flist
, /* blocks freed in xaction */
752 xfs_btree_cur_t
**curp
, /* cursor returned to caller */
753 int wasdel
, /* converting a delayed alloc */
754 int *logflagsp
, /* inode logging flags */
755 int whichfork
) /* data or attr fork */
757 struct xfs_btree_block
*ablock
; /* allocated (child) bt block */
758 xfs_buf_t
*abp
; /* buffer for ablock */
759 xfs_alloc_arg_t args
; /* allocation arguments */
760 xfs_bmbt_rec_t
*arp
; /* child record pointer */
761 struct xfs_btree_block
*block
; /* btree root block */
762 xfs_btree_cur_t
*cur
; /* bmap btree cursor */
763 xfs_bmbt_rec_host_t
*ep
; /* extent record pointer */
764 int error
; /* error return value */
765 xfs_extnum_t i
, cnt
; /* extent record index */
766 xfs_ifork_t
*ifp
; /* inode fork pointer */
767 xfs_bmbt_key_t
*kp
; /* root block key pointer */
768 xfs_mount_t
*mp
; /* mount structure */
769 xfs_extnum_t nextents
; /* number of file extents */
770 xfs_bmbt_ptr_t
*pp
; /* root block address pointer */
773 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
774 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_EXTENTS
);
777 * Make space in the inode incore.
779 xfs_iroot_realloc(ip
, 1, whichfork
);
780 ifp
->if_flags
|= XFS_IFBROOT
;
785 block
= ifp
->if_broot
;
786 if (xfs_sb_version_hascrc(&mp
->m_sb
))
787 xfs_btree_init_block_int(mp
, block
, XFS_BUF_DADDR_NULL
,
788 XFS_BMAP_CRC_MAGIC
, 1, 1, ip
->i_ino
,
789 XFS_BTREE_LONG_PTRS
| XFS_BTREE_CRC_BLOCKS
);
791 xfs_btree_init_block_int(mp
, block
, XFS_BUF_DADDR_NULL
,
792 XFS_BMAP_MAGIC
, 1, 1, ip
->i_ino
,
793 XFS_BTREE_LONG_PTRS
);
796 * Need a cursor. Can't allocate until bb_level is filled in.
798 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
799 cur
->bc_private
.b
.firstblock
= *firstblock
;
800 cur
->bc_private
.b
.flist
= flist
;
801 cur
->bc_private
.b
.flags
= wasdel
? XFS_BTCUR_BPRV_WASDEL
: 0;
803 * Convert to a btree with two levels, one record in root.
805 XFS_IFORK_FMT_SET(ip
, whichfork
, XFS_DINODE_FMT_BTREE
);
806 memset(&args
, 0, sizeof(args
));
809 args
.firstblock
= *firstblock
;
810 if (*firstblock
== NULLFSBLOCK
) {
811 args
.type
= XFS_ALLOCTYPE_START_BNO
;
812 args
.fsbno
= XFS_INO_TO_FSB(mp
, ip
->i_ino
);
813 } else if (flist
->xbf_low
) {
814 args
.type
= XFS_ALLOCTYPE_START_BNO
;
815 args
.fsbno
= *firstblock
;
817 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
818 args
.fsbno
= *firstblock
;
820 args
.minlen
= args
.maxlen
= args
.prod
= 1;
821 args
.wasdel
= wasdel
;
823 if ((error
= xfs_alloc_vextent(&args
))) {
824 xfs_iroot_realloc(ip
, -1, whichfork
);
825 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
829 * Allocation can't fail, the space was reserved.
831 ASSERT(args
.fsbno
!= NULLFSBLOCK
);
832 ASSERT(*firstblock
== NULLFSBLOCK
||
833 args
.agno
== XFS_FSB_TO_AGNO(mp
, *firstblock
) ||
835 args
.agno
> XFS_FSB_TO_AGNO(mp
, *firstblock
)));
836 *firstblock
= cur
->bc_private
.b
.firstblock
= args
.fsbno
;
837 cur
->bc_private
.b
.allocated
++;
838 ip
->i_d
.di_nblocks
++;
839 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_BCOUNT
, 1L);
840 abp
= xfs_btree_get_bufl(mp
, tp
, args
.fsbno
, 0);
842 * Fill in the child block.
844 abp
->b_ops
= &xfs_bmbt_buf_ops
;
845 ablock
= XFS_BUF_TO_BLOCK(abp
);
846 if (xfs_sb_version_hascrc(&mp
->m_sb
))
847 xfs_btree_init_block_int(mp
, ablock
, abp
->b_bn
,
848 XFS_BMAP_CRC_MAGIC
, 0, 0, ip
->i_ino
,
849 XFS_BTREE_LONG_PTRS
| XFS_BTREE_CRC_BLOCKS
);
851 xfs_btree_init_block_int(mp
, ablock
, abp
->b_bn
,
852 XFS_BMAP_MAGIC
, 0, 0, ip
->i_ino
,
853 XFS_BTREE_LONG_PTRS
);
855 arp
= XFS_BMBT_REC_ADDR(mp
, ablock
, 1);
856 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
857 for (cnt
= i
= 0; i
< nextents
; i
++) {
858 ep
= xfs_iext_get_ext(ifp
, i
);
859 if (!isnullstartblock(xfs_bmbt_get_startblock(ep
))) {
860 arp
->l0
= cpu_to_be64(ep
->l0
);
861 arp
->l1
= cpu_to_be64(ep
->l1
);
865 ASSERT(cnt
== XFS_IFORK_NEXTENTS(ip
, whichfork
));
866 xfs_btree_set_numrecs(ablock
, cnt
);
869 * Fill in the root key and pointer.
871 kp
= XFS_BMBT_KEY_ADDR(mp
, block
, 1);
872 arp
= XFS_BMBT_REC_ADDR(mp
, ablock
, 1);
873 kp
->br_startoff
= cpu_to_be64(xfs_bmbt_disk_get_startoff(arp
));
874 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, 1, xfs_bmbt_get_maxrecs(cur
,
875 be16_to_cpu(block
->bb_level
)));
876 *pp
= cpu_to_be64(args
.fsbno
);
879 * Do all this logging at the end so that
880 * the root is at the right level.
882 xfs_btree_log_block(cur
, abp
, XFS_BB_ALL_BITS
);
883 xfs_btree_log_recs(cur
, abp
, 1, be16_to_cpu(ablock
->bb_numrecs
));
884 ASSERT(*curp
== NULL
);
886 *logflagsp
= XFS_ILOG_CORE
| xfs_ilog_fbroot(whichfork
);
891 * Convert a local file to an extents file.
892 * This code is out of bounds for data forks of regular files,
893 * since the file data needs to get logged so things will stay consistent.
894 * (The bmap-level manipulations are ok, though).
897 xfs_bmap_local_to_extents_empty(
898 struct xfs_inode
*ip
,
901 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
903 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
);
904 ASSERT(ifp
->if_bytes
== 0);
905 ASSERT(XFS_IFORK_NEXTENTS(ip
, whichfork
) == 0);
907 xfs_bmap_forkoff_reset(ip
, whichfork
);
908 ifp
->if_flags
&= ~XFS_IFINLINE
;
909 ifp
->if_flags
|= XFS_IFEXTENTS
;
910 XFS_IFORK_FMT_SET(ip
, whichfork
, XFS_DINODE_FMT_EXTENTS
);
914 STATIC
int /* error */
915 xfs_bmap_local_to_extents(
916 xfs_trans_t
*tp
, /* transaction pointer */
917 xfs_inode_t
*ip
, /* incore inode pointer */
918 xfs_fsblock_t
*firstblock
, /* first block allocated in xaction */
919 xfs_extlen_t total
, /* total blocks needed by transaction */
920 int *logflagsp
, /* inode logging flags */
922 void (*init_fn
)(struct xfs_trans
*tp
,
924 struct xfs_inode
*ip
,
925 struct xfs_ifork
*ifp
))
928 int flags
; /* logging flags returned */
929 xfs_ifork_t
*ifp
; /* inode fork pointer */
930 xfs_alloc_arg_t args
; /* allocation arguments */
931 xfs_buf_t
*bp
; /* buffer for extent block */
932 xfs_bmbt_rec_host_t
*ep
; /* extent record pointer */
935 * We don't want to deal with the case of keeping inode data inline yet.
936 * So sending the data fork of a regular inode is invalid.
938 ASSERT(!(S_ISREG(ip
->i_d
.di_mode
) && whichfork
== XFS_DATA_FORK
));
939 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
940 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
);
942 if (!ifp
->if_bytes
) {
943 xfs_bmap_local_to_extents_empty(ip
, whichfork
);
944 flags
= XFS_ILOG_CORE
;
950 ASSERT((ifp
->if_flags
& (XFS_IFINLINE
|XFS_IFEXTENTS
|XFS_IFEXTIREC
)) ==
952 memset(&args
, 0, sizeof(args
));
954 args
.mp
= ip
->i_mount
;
955 args
.firstblock
= *firstblock
;
957 * Allocate a block. We know we need only one, since the
958 * file currently fits in an inode.
960 if (*firstblock
== NULLFSBLOCK
) {
961 args
.fsbno
= XFS_INO_TO_FSB(args
.mp
, ip
->i_ino
);
962 args
.type
= XFS_ALLOCTYPE_START_BNO
;
964 args
.fsbno
= *firstblock
;
965 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
968 args
.minlen
= args
.maxlen
= args
.prod
= 1;
969 error
= xfs_alloc_vextent(&args
);
973 /* Can't fail, the space was reserved. */
974 ASSERT(args
.fsbno
!= NULLFSBLOCK
);
975 ASSERT(args
.len
== 1);
976 *firstblock
= args
.fsbno
;
977 bp
= xfs_btree_get_bufl(args
.mp
, tp
, args
.fsbno
, 0);
979 /* initialise the block and copy the data */
980 init_fn(tp
, bp
, ip
, ifp
);
982 /* account for the change in fork size and log everything */
983 xfs_trans_log_buf(tp
, bp
, 0, ifp
->if_bytes
- 1);
984 xfs_idata_realloc(ip
, -ifp
->if_bytes
, whichfork
);
985 xfs_bmap_local_to_extents_empty(ip
, whichfork
);
986 flags
|= XFS_ILOG_CORE
;
988 xfs_iext_add(ifp
, 0, 1);
989 ep
= xfs_iext_get_ext(ifp
, 0);
990 xfs_bmbt_set_allf(ep
, 0, args
.fsbno
, 1, XFS_EXT_NORM
);
991 trace_xfs_bmap_post_update(ip
, 0,
992 whichfork
== XFS_ATTR_FORK
? BMAP_ATTRFORK
: 0,
994 XFS_IFORK_NEXT_SET(ip
, whichfork
, 1);
995 ip
->i_d
.di_nblocks
= 1;
996 xfs_trans_mod_dquot_byino(tp
, ip
,
997 XFS_TRANS_DQ_BCOUNT
, 1L);
998 flags
|= xfs_ilog_fext(whichfork
);
1006 * Called from xfs_bmap_add_attrfork to handle btree format files.
1008 STATIC
int /* error */
1009 xfs_bmap_add_attrfork_btree(
1010 xfs_trans_t
*tp
, /* transaction pointer */
1011 xfs_inode_t
*ip
, /* incore inode pointer */
1012 xfs_fsblock_t
*firstblock
, /* first block allocated */
1013 xfs_bmap_free_t
*flist
, /* blocks to free at commit */
1014 int *flags
) /* inode logging flags */
1016 xfs_btree_cur_t
*cur
; /* btree cursor */
1017 int error
; /* error return value */
1018 xfs_mount_t
*mp
; /* file system mount struct */
1019 int stat
; /* newroot status */
1022 if (ip
->i_df
.if_broot_bytes
<= XFS_IFORK_DSIZE(ip
))
1023 *flags
|= XFS_ILOG_DBROOT
;
1025 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, XFS_DATA_FORK
);
1026 cur
->bc_private
.b
.flist
= flist
;
1027 cur
->bc_private
.b
.firstblock
= *firstblock
;
1028 if ((error
= xfs_bmbt_lookup_ge(cur
, 0, 0, 0, &stat
)))
1030 /* must be at least one entry */
1031 XFS_WANT_CORRUPTED_GOTO(stat
== 1, error0
);
1032 if ((error
= xfs_btree_new_iroot(cur
, flags
, &stat
)))
1035 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
1036 return XFS_ERROR(ENOSPC
);
1038 *firstblock
= cur
->bc_private
.b
.firstblock
;
1039 cur
->bc_private
.b
.allocated
= 0;
1040 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
1044 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
1049 * Called from xfs_bmap_add_attrfork to handle extents format files.
1051 STATIC
int /* error */
1052 xfs_bmap_add_attrfork_extents(
1053 xfs_trans_t
*tp
, /* transaction pointer */
1054 xfs_inode_t
*ip
, /* incore inode pointer */
1055 xfs_fsblock_t
*firstblock
, /* first block allocated */
1056 xfs_bmap_free_t
*flist
, /* blocks to free at commit */
1057 int *flags
) /* inode logging flags */
1059 xfs_btree_cur_t
*cur
; /* bmap btree cursor */
1060 int error
; /* error return value */
1062 if (ip
->i_d
.di_nextents
* sizeof(xfs_bmbt_rec_t
) <= XFS_IFORK_DSIZE(ip
))
1065 error
= xfs_bmap_extents_to_btree(tp
, ip
, firstblock
, flist
, &cur
, 0,
1066 flags
, XFS_DATA_FORK
);
1068 cur
->bc_private
.b
.allocated
= 0;
1069 xfs_btree_del_cursor(cur
,
1070 error
? XFS_BTREE_ERROR
: XFS_BTREE_NOERROR
);
1076 * Called from xfs_bmap_add_attrfork to handle local format files. Each
1077 * different data fork content type needs a different callout to do the
1078 * conversion. Some are basic and only require special block initialisation
1079 * callouts for the data formating, others (directories) are so specialised they
1080 * handle everything themselves.
1082 * XXX (dgc): investigate whether directory conversion can use the generic
1083 * formatting callout. It should be possible - it's just a very complex
1086 STATIC
int /* error */
1087 xfs_bmap_add_attrfork_local(
1088 xfs_trans_t
*tp
, /* transaction pointer */
1089 xfs_inode_t
*ip
, /* incore inode pointer */
1090 xfs_fsblock_t
*firstblock
, /* first block allocated */
1091 xfs_bmap_free_t
*flist
, /* blocks to free at commit */
1092 int *flags
) /* inode logging flags */
1094 xfs_da_args_t dargs
; /* args for dir/attr code */
1096 if (ip
->i_df
.if_bytes
<= XFS_IFORK_DSIZE(ip
))
1099 if (S_ISDIR(ip
->i_d
.di_mode
)) {
1100 memset(&dargs
, 0, sizeof(dargs
));
1102 dargs
.firstblock
= firstblock
;
1103 dargs
.flist
= flist
;
1104 dargs
.total
= ip
->i_mount
->m_dirblkfsbs
;
1105 dargs
.whichfork
= XFS_DATA_FORK
;
1107 return xfs_dir2_sf_to_block(&dargs
);
1110 if (S_ISLNK(ip
->i_d
.di_mode
))
1111 return xfs_bmap_local_to_extents(tp
, ip
, firstblock
, 1,
1112 flags
, XFS_DATA_FORK
,
1113 xfs_symlink_local_to_remote
);
1115 /* should only be called for types that support local format data */
1117 return EFSCORRUPTED
;
1121 * Convert inode from non-attributed to attributed.
1122 * Must not be in a transaction, ip must not be locked.
1124 int /* error code */
1125 xfs_bmap_add_attrfork(
1126 xfs_inode_t
*ip
, /* incore inode pointer */
1127 int size
, /* space new attribute needs */
1128 int rsvd
) /* xact may use reserved blks */
1130 xfs_fsblock_t firstblock
; /* 1st block/ag allocated */
1131 xfs_bmap_free_t flist
; /* freed extent records */
1132 xfs_mount_t
*mp
; /* mount structure */
1133 xfs_trans_t
*tp
; /* transaction pointer */
1134 int blks
; /* space reservation */
1135 int version
= 1; /* superblock attr version */
1136 int committed
; /* xaction was committed */
1137 int logflags
; /* logging flags */
1138 int error
; /* error return value */
1139 int cancel_flags
= 0;
1141 ASSERT(XFS_IFORK_Q(ip
) == 0);
1144 ASSERT(!XFS_NOT_DQATTACHED(mp
, ip
));
1145 tp
= xfs_trans_alloc(mp
, XFS_TRANS_ADDAFORK
);
1146 blks
= XFS_ADDAFORK_SPACE_RES(mp
);
1148 tp
->t_flags
|= XFS_TRANS_RESERVE
;
1149 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_addafork
, blks
, 0);
1151 xfs_trans_cancel(tp
, 0);
1154 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
1155 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1156 error
= xfs_trans_reserve_quota_nblks(tp
, ip
, blks
, 0, rsvd
?
1157 XFS_QMOPT_RES_REGBLKS
| XFS_QMOPT_FORCE_RES
:
1158 XFS_QMOPT_RES_REGBLKS
);
1161 cancel_flags
|= XFS_TRANS_ABORT
;
1162 if (XFS_IFORK_Q(ip
))
1164 if (ip
->i_d
.di_aformat
!= XFS_DINODE_FMT_EXTENTS
) {
1166 * For inodes coming from pre-6.2 filesystems.
1168 ASSERT(ip
->i_d
.di_aformat
== 0);
1169 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
1171 ASSERT(ip
->i_d
.di_anextents
== 0);
1173 xfs_trans_ijoin(tp
, ip
, 0);
1174 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1176 switch (ip
->i_d
.di_format
) {
1177 case XFS_DINODE_FMT_DEV
:
1178 ip
->i_d
.di_forkoff
= roundup(sizeof(xfs_dev_t
), 8) >> 3;
1180 case XFS_DINODE_FMT_UUID
:
1181 ip
->i_d
.di_forkoff
= roundup(sizeof(uuid_t
), 8) >> 3;
1183 case XFS_DINODE_FMT_LOCAL
:
1184 case XFS_DINODE_FMT_EXTENTS
:
1185 case XFS_DINODE_FMT_BTREE
:
1186 ip
->i_d
.di_forkoff
= xfs_attr_shortform_bytesfit(ip
, size
);
1187 if (!ip
->i_d
.di_forkoff
)
1188 ip
->i_d
.di_forkoff
= xfs_default_attroffset(ip
) >> 3;
1189 else if (mp
->m_flags
& XFS_MOUNT_ATTR2
)
1194 error
= XFS_ERROR(EINVAL
);
1198 ASSERT(ip
->i_afp
== NULL
);
1199 ip
->i_afp
= kmem_zone_zalloc(xfs_ifork_zone
, KM_SLEEP
);
1200 ip
->i_afp
->if_flags
= XFS_IFEXTENTS
;
1202 xfs_bmap_init(&flist
, &firstblock
);
1203 switch (ip
->i_d
.di_format
) {
1204 case XFS_DINODE_FMT_LOCAL
:
1205 error
= xfs_bmap_add_attrfork_local(tp
, ip
, &firstblock
, &flist
,
1208 case XFS_DINODE_FMT_EXTENTS
:
1209 error
= xfs_bmap_add_attrfork_extents(tp
, ip
, &firstblock
,
1212 case XFS_DINODE_FMT_BTREE
:
1213 error
= xfs_bmap_add_attrfork_btree(tp
, ip
, &firstblock
, &flist
,
1221 xfs_trans_log_inode(tp
, ip
, logflags
);
1224 if (!xfs_sb_version_hasattr(&mp
->m_sb
) ||
1225 (!xfs_sb_version_hasattr2(&mp
->m_sb
) && version
== 2)) {
1226 __int64_t sbfields
= 0;
1228 spin_lock(&mp
->m_sb_lock
);
1229 if (!xfs_sb_version_hasattr(&mp
->m_sb
)) {
1230 xfs_sb_version_addattr(&mp
->m_sb
);
1231 sbfields
|= XFS_SB_VERSIONNUM
;
1233 if (!xfs_sb_version_hasattr2(&mp
->m_sb
) && version
== 2) {
1234 xfs_sb_version_addattr2(&mp
->m_sb
);
1235 sbfields
|= (XFS_SB_VERSIONNUM
| XFS_SB_FEATURES2
);
1238 spin_unlock(&mp
->m_sb_lock
);
1239 xfs_mod_sb(tp
, sbfields
);
1241 spin_unlock(&mp
->m_sb_lock
);
1244 error
= xfs_bmap_finish(&tp
, &flist
, &committed
);
1247 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1248 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1252 xfs_bmap_cancel(&flist
);
1254 xfs_trans_cancel(tp
, cancel_flags
);
1255 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1260 * Internal and external extent tree search functions.
1264 * Read in the extents to if_extents.
1265 * All inode fields are set up by caller, we just traverse the btree
1266 * and copy the records in. If the file system cannot contain unwritten
1267 * extents, the records are checked for no "state" flags.
1270 xfs_bmap_read_extents(
1271 xfs_trans_t
*tp
, /* transaction pointer */
1272 xfs_inode_t
*ip
, /* incore inode */
1273 int whichfork
) /* data or attr fork */
1275 struct xfs_btree_block
*block
; /* current btree block */
1276 xfs_fsblock_t bno
; /* block # of "block" */
1277 xfs_buf_t
*bp
; /* buffer for "block" */
1278 int error
; /* error return value */
1279 xfs_exntfmt_t exntf
; /* XFS_EXTFMT_NOSTATE, if checking */
1280 xfs_extnum_t i
, j
; /* index into the extents list */
1281 xfs_ifork_t
*ifp
; /* fork structure */
1282 int level
; /* btree level, for checking */
1283 xfs_mount_t
*mp
; /* file system mount structure */
1284 __be64
*pp
; /* pointer to block address */
1286 xfs_extnum_t room
; /* number of entries there's room for */
1290 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1291 exntf
= (whichfork
!= XFS_DATA_FORK
) ? XFS_EXTFMT_NOSTATE
:
1292 XFS_EXTFMT_INODE(ip
);
1293 block
= ifp
->if_broot
;
1295 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1297 level
= be16_to_cpu(block
->bb_level
);
1299 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, 1, ifp
->if_broot_bytes
);
1300 bno
= be64_to_cpu(*pp
);
1301 ASSERT(bno
!= NULLDFSBNO
);
1302 ASSERT(XFS_FSB_TO_AGNO(mp
, bno
) < mp
->m_sb
.sb_agcount
);
1303 ASSERT(XFS_FSB_TO_AGBNO(mp
, bno
) < mp
->m_sb
.sb_agblocks
);
1305 * Go down the tree until leaf level is reached, following the first
1306 * pointer (leftmost) at each level.
1308 while (level
-- > 0) {
1309 error
= xfs_btree_read_bufl(mp
, tp
, bno
, 0, &bp
,
1310 XFS_BMAP_BTREE_REF
, &xfs_bmbt_buf_ops
);
1313 block
= XFS_BUF_TO_BLOCK(bp
);
1314 XFS_WANT_CORRUPTED_GOTO(
1315 xfs_bmap_sanity_check(mp
, bp
, level
),
1319 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, 1, mp
->m_bmap_dmxr
[1]);
1320 bno
= be64_to_cpu(*pp
);
1321 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp
, bno
), error0
);
1322 xfs_trans_brelse(tp
, bp
);
1325 * Here with bp and block set to the leftmost leaf node in the tree.
1327 room
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
1330 * Loop over all leaf nodes. Copy information to the extent records.
1333 xfs_bmbt_rec_t
*frp
;
1334 xfs_fsblock_t nextbno
;
1335 xfs_extnum_t num_recs
;
1338 num_recs
= xfs_btree_get_numrecs(block
);
1339 if (unlikely(i
+ num_recs
> room
)) {
1340 ASSERT(i
+ num_recs
<= room
);
1341 xfs_warn(ip
->i_mount
,
1342 "corrupt dinode %Lu, (btree extents).",
1343 (unsigned long long) ip
->i_ino
);
1344 XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
1345 XFS_ERRLEVEL_LOW
, ip
->i_mount
, block
);
1348 XFS_WANT_CORRUPTED_GOTO(
1349 xfs_bmap_sanity_check(mp
, bp
, 0),
1352 * Read-ahead the next leaf block, if any.
1354 nextbno
= be64_to_cpu(block
->bb_u
.l
.bb_rightsib
);
1355 if (nextbno
!= NULLFSBLOCK
)
1356 xfs_btree_reada_bufl(mp
, nextbno
, 1,
1359 * Copy records into the extent records.
1361 frp
= XFS_BMBT_REC_ADDR(mp
, block
, 1);
1363 for (j
= 0; j
< num_recs
; j
++, i
++, frp
++) {
1364 xfs_bmbt_rec_host_t
*trp
= xfs_iext_get_ext(ifp
, i
);
1365 trp
->l0
= be64_to_cpu(frp
->l0
);
1366 trp
->l1
= be64_to_cpu(frp
->l1
);
1368 if (exntf
== XFS_EXTFMT_NOSTATE
) {
1370 * Check all attribute bmap btree records and
1371 * any "older" data bmap btree records for a
1372 * set bit in the "extent flag" position.
1374 if (unlikely(xfs_check_nostate_extents(ifp
,
1375 start
, num_recs
))) {
1376 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
1382 xfs_trans_brelse(tp
, bp
);
1385 * If we've reached the end, stop.
1387 if (bno
== NULLFSBLOCK
)
1389 error
= xfs_btree_read_bufl(mp
, tp
, bno
, 0, &bp
,
1390 XFS_BMAP_BTREE_REF
, &xfs_bmbt_buf_ops
);
1393 block
= XFS_BUF_TO_BLOCK(bp
);
1395 ASSERT(i
== (ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
)));
1396 ASSERT(i
== XFS_IFORK_NEXTENTS(ip
, whichfork
));
1397 XFS_BMAP_TRACE_EXLIST(ip
, i
, whichfork
);
1400 xfs_trans_brelse(tp
, bp
);
1401 return XFS_ERROR(EFSCORRUPTED
);
1406 * Search the extent records for the entry containing block bno.
1407 * If bno lies in a hole, point to the next entry. If bno lies
1408 * past eof, *eofp will be set, and *prevp will contain the last
1409 * entry (null if none). Else, *lastxp will be set to the index
1410 * of the found entry; *gotp will contain the entry.
1412 STATIC xfs_bmbt_rec_host_t
* /* pointer to found extent entry */
1413 xfs_bmap_search_multi_extents(
1414 xfs_ifork_t
*ifp
, /* inode fork pointer */
1415 xfs_fileoff_t bno
, /* block number searched for */
1416 int *eofp
, /* out: end of file found */
1417 xfs_extnum_t
*lastxp
, /* out: last extent index */
1418 xfs_bmbt_irec_t
*gotp
, /* out: extent entry found */
1419 xfs_bmbt_irec_t
*prevp
) /* out: previous extent entry found */
1421 xfs_bmbt_rec_host_t
*ep
; /* extent record pointer */
1422 xfs_extnum_t lastx
; /* last extent index */
1425 * Initialize the extent entry structure to catch access to
1426 * uninitialized br_startblock field.
1428 gotp
->br_startoff
= 0xffa5a5a5a5a5a5a5LL
;
1429 gotp
->br_blockcount
= 0xa55a5a5a5a5a5a5aLL
;
1430 gotp
->br_state
= XFS_EXT_INVALID
;
1432 gotp
->br_startblock
= 0xffffa5a5a5a5a5a5LL
;
1434 gotp
->br_startblock
= 0xffffa5a5;
1436 prevp
->br_startoff
= NULLFILEOFF
;
1438 ep
= xfs_iext_bno_to_ext(ifp
, bno
, &lastx
);
1440 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, lastx
- 1), prevp
);
1442 if (lastx
< (ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
))) {
1443 xfs_bmbt_get_all(ep
, gotp
);
1457 * Search the extents list for the inode, for the extent containing bno.
1458 * If bno lies in a hole, point to the next entry. If bno lies past eof,
1459 * *eofp will be set, and *prevp will contain the last entry (null if none).
1460 * Else, *lastxp will be set to the index of the found
1461 * entry; *gotp will contain the entry.
1463 STATIC xfs_bmbt_rec_host_t
* /* pointer to found extent entry */
1464 xfs_bmap_search_extents(
1465 xfs_inode_t
*ip
, /* incore inode pointer */
1466 xfs_fileoff_t bno
, /* block number searched for */
1467 int fork
, /* data or attr fork */
1468 int *eofp
, /* out: end of file found */
1469 xfs_extnum_t
*lastxp
, /* out: last extent index */
1470 xfs_bmbt_irec_t
*gotp
, /* out: extent entry found */
1471 xfs_bmbt_irec_t
*prevp
) /* out: previous extent entry found */
1473 xfs_ifork_t
*ifp
; /* inode fork pointer */
1474 xfs_bmbt_rec_host_t
*ep
; /* extent record pointer */
1476 XFS_STATS_INC(xs_look_exlist
);
1477 ifp
= XFS_IFORK_PTR(ip
, fork
);
1479 ep
= xfs_bmap_search_multi_extents(ifp
, bno
, eofp
, lastxp
, gotp
, prevp
);
1481 if (unlikely(!(gotp
->br_startblock
) && (*lastxp
!= NULLEXTNUM
) &&
1482 !(XFS_IS_REALTIME_INODE(ip
) && fork
== XFS_DATA_FORK
))) {
1483 xfs_alert_tag(ip
->i_mount
, XFS_PTAG_FSBLOCK_ZERO
,
1484 "Access to block zero in inode %llu "
1485 "start_block: %llx start_off: %llx "
1486 "blkcnt: %llx extent-state: %x lastx: %x",
1487 (unsigned long long)ip
->i_ino
,
1488 (unsigned long long)gotp
->br_startblock
,
1489 (unsigned long long)gotp
->br_startoff
,
1490 (unsigned long long)gotp
->br_blockcount
,
1491 gotp
->br_state
, *lastxp
);
1492 *lastxp
= NULLEXTNUM
;
1500 * Returns the file-relative block number of the first unused block(s)
1501 * in the file with at least "len" logically contiguous blocks free.
1502 * This is the lowest-address hole if the file has holes, else the first block
1503 * past the end of file.
1504 * Return 0 if the file is currently local (in-inode).
1507 xfs_bmap_first_unused(
1508 xfs_trans_t
*tp
, /* transaction pointer */
1509 xfs_inode_t
*ip
, /* incore inode */
1510 xfs_extlen_t len
, /* size of hole to find */
1511 xfs_fileoff_t
*first_unused
, /* unused block */
1512 int whichfork
) /* data or attr fork */
1514 int error
; /* error return value */
1515 int idx
; /* extent record index */
1516 xfs_ifork_t
*ifp
; /* inode fork pointer */
1517 xfs_fileoff_t lastaddr
; /* last block number seen */
1518 xfs_fileoff_t lowest
; /* lowest useful block */
1519 xfs_fileoff_t max
; /* starting useful block */
1520 xfs_fileoff_t off
; /* offset for this block */
1521 xfs_extnum_t nextents
; /* number of extent entries */
1523 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_BTREE
||
1524 XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_EXTENTS
||
1525 XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
);
1526 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
) {
1530 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1531 if (!(ifp
->if_flags
& XFS_IFEXTENTS
) &&
1532 (error
= xfs_iread_extents(tp
, ip
, whichfork
)))
1534 lowest
= *first_unused
;
1535 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
1536 for (idx
= 0, lastaddr
= 0, max
= lowest
; idx
< nextents
; idx
++) {
1537 xfs_bmbt_rec_host_t
*ep
= xfs_iext_get_ext(ifp
, idx
);
1538 off
= xfs_bmbt_get_startoff(ep
);
1540 * See if the hole before this extent will work.
1542 if (off
>= lowest
+ len
&& off
- max
>= len
) {
1543 *first_unused
= max
;
1546 lastaddr
= off
+ xfs_bmbt_get_blockcount(ep
);
1547 max
= XFS_FILEOFF_MAX(lastaddr
, lowest
);
1549 *first_unused
= max
;
1554 * Returns the file-relative block number of the last block - 1 before
1555 * last_block (input value) in the file.
1556 * This is not based on i_size, it is based on the extent records.
1557 * Returns 0 for local files, as they do not have extent records.
1560 xfs_bmap_last_before(
1561 xfs_trans_t
*tp
, /* transaction pointer */
1562 xfs_inode_t
*ip
, /* incore inode */
1563 xfs_fileoff_t
*last_block
, /* last block */
1564 int whichfork
) /* data or attr fork */
1566 xfs_fileoff_t bno
; /* input file offset */
1567 int eof
; /* hit end of file */
1568 xfs_bmbt_rec_host_t
*ep
; /* pointer to last extent */
1569 int error
; /* error return value */
1570 xfs_bmbt_irec_t got
; /* current extent value */
1571 xfs_ifork_t
*ifp
; /* inode fork pointer */
1572 xfs_extnum_t lastx
; /* last extent used */
1573 xfs_bmbt_irec_t prev
; /* previous extent value */
1575 if (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
&&
1576 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
1577 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_LOCAL
)
1578 return XFS_ERROR(EIO
);
1579 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
) {
1583 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1584 if (!(ifp
->if_flags
& XFS_IFEXTENTS
) &&
1585 (error
= xfs_iread_extents(tp
, ip
, whichfork
)))
1587 bno
= *last_block
- 1;
1588 ep
= xfs_bmap_search_extents(ip
, bno
, whichfork
, &eof
, &lastx
, &got
,
1590 if (eof
|| xfs_bmbt_get_startoff(ep
) > bno
) {
1591 if (prev
.br_startoff
== NULLFILEOFF
)
1594 *last_block
= prev
.br_startoff
+ prev
.br_blockcount
;
1597 * Otherwise *last_block is already the right answer.
1603 xfs_bmap_last_extent(
1604 struct xfs_trans
*tp
,
1605 struct xfs_inode
*ip
,
1607 struct xfs_bmbt_irec
*rec
,
1610 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1614 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
1615 error
= xfs_iread_extents(tp
, ip
, whichfork
);
1620 nextents
= ifp
->if_bytes
/ sizeof(xfs_bmbt_rec_t
);
1621 if (nextents
== 0) {
1626 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, nextents
- 1), rec
);
1632 * Check the last inode extent to determine whether this allocation will result
1633 * in blocks being allocated at the end of the file. When we allocate new data
1634 * blocks at the end of the file which do not start at the previous data block,
1635 * we will try to align the new blocks at stripe unit boundaries.
1637 * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1638 * at, or past the EOF.
1642 struct xfs_bmalloca
*bma
,
1645 struct xfs_bmbt_irec rec
;
1650 error
= xfs_bmap_last_extent(NULL
, bma
->ip
, whichfork
, &rec
,
1661 * Check if we are allocation or past the last extent, or at least into
1662 * the last delayed allocated extent.
1664 bma
->aeof
= bma
->offset
>= rec
.br_startoff
+ rec
.br_blockcount
||
1665 (bma
->offset
>= rec
.br_startoff
&&
1666 isnullstartblock(rec
.br_startblock
));
1671 * Returns the file-relative block number of the first block past eof in
1672 * the file. This is not based on i_size, it is based on the extent records.
1673 * Returns 0 for local files, as they do not have extent records.
1676 xfs_bmap_last_offset(
1677 struct xfs_inode
*ip
,
1678 xfs_fileoff_t
*last_block
,
1681 struct xfs_bmbt_irec rec
;
1687 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
)
1690 if (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
&&
1691 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
)
1692 return XFS_ERROR(EIO
);
1694 error
= xfs_bmap_last_extent(NULL
, ip
, whichfork
, &rec
, &is_empty
);
1695 if (error
|| is_empty
)
1698 *last_block
= rec
.br_startoff
+ rec
.br_blockcount
;
1703 * Returns whether the selected fork of the inode has exactly one
1704 * block or not. For the data fork we check this matches di_size,
1705 * implying the file's range is 0..bsize-1.
1707 int /* 1=>1 block, 0=>otherwise */
1709 xfs_inode_t
*ip
, /* incore inode */
1710 int whichfork
) /* data or attr fork */
1712 xfs_bmbt_rec_host_t
*ep
; /* ptr to fork's extent */
1713 xfs_ifork_t
*ifp
; /* inode fork pointer */
1714 int rval
; /* return value */
1715 xfs_bmbt_irec_t s
; /* internal version of extent */
1718 if (whichfork
== XFS_DATA_FORK
)
1719 return XFS_ISIZE(ip
) == ip
->i_mount
->m_sb
.sb_blocksize
;
1721 if (XFS_IFORK_NEXTENTS(ip
, whichfork
) != 1)
1723 if (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
)
1725 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1726 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
1727 ep
= xfs_iext_get_ext(ifp
, 0);
1728 xfs_bmbt_get_all(ep
, &s
);
1729 rval
= s
.br_startoff
== 0 && s
.br_blockcount
== 1;
1730 if (rval
&& whichfork
== XFS_DATA_FORK
)
1731 ASSERT(XFS_ISIZE(ip
) == ip
->i_mount
->m_sb
.sb_blocksize
);
1736 * Extent tree manipulation functions used during allocation.
1740 * Convert a delayed allocation to a real allocation.
1742 STATIC
int /* error */
1743 xfs_bmap_add_extent_delay_real(
1744 struct xfs_bmalloca
*bma
)
1746 struct xfs_bmbt_irec
*new = &bma
->got
;
1747 int diff
; /* temp value */
1748 xfs_bmbt_rec_host_t
*ep
; /* extent entry for idx */
1749 int error
; /* error return value */
1750 int i
; /* temp state */
1751 xfs_ifork_t
*ifp
; /* inode fork pointer */
1752 xfs_fileoff_t new_endoff
; /* end offset of new entry */
1753 xfs_bmbt_irec_t r
[3]; /* neighbor extent entries */
1754 /* left is 0, right is 1, prev is 2 */
1755 int rval
=0; /* return value (logging flags) */
1756 int state
= 0;/* state bits, accessed thru macros */
1757 xfs_filblks_t da_new
; /* new count del alloc blocks used */
1758 xfs_filblks_t da_old
; /* old count del alloc blocks used */
1759 xfs_filblks_t temp
=0; /* value for da_new calculations */
1760 xfs_filblks_t temp2
=0;/* value for da_new calculations */
1761 int tmp_rval
; /* partial logging flags */
1763 ifp
= XFS_IFORK_PTR(bma
->ip
, XFS_DATA_FORK
);
1765 ASSERT(bma
->idx
>= 0);
1766 ASSERT(bma
->idx
<= ifp
->if_bytes
/ sizeof(struct xfs_bmbt_rec
));
1767 ASSERT(!isnullstartblock(new->br_startblock
));
1769 (bma
->cur
->bc_private
.b
.flags
& XFS_BTCUR_BPRV_WASDEL
));
1771 XFS_STATS_INC(xs_add_exlist
);
1778 * Set up a bunch of variables to make the tests simpler.
1780 ep
= xfs_iext_get_ext(ifp
, bma
->idx
);
1781 xfs_bmbt_get_all(ep
, &PREV
);
1782 new_endoff
= new->br_startoff
+ new->br_blockcount
;
1783 ASSERT(PREV
.br_startoff
<= new->br_startoff
);
1784 ASSERT(PREV
.br_startoff
+ PREV
.br_blockcount
>= new_endoff
);
1786 da_old
= startblockval(PREV
.br_startblock
);
1790 * Set flags determining what part of the previous delayed allocation
1791 * extent is being replaced by a real allocation.
1793 if (PREV
.br_startoff
== new->br_startoff
)
1794 state
|= BMAP_LEFT_FILLING
;
1795 if (PREV
.br_startoff
+ PREV
.br_blockcount
== new_endoff
)
1796 state
|= BMAP_RIGHT_FILLING
;
1799 * Check and set flags if this segment has a left neighbor.
1800 * Don't set contiguous if the combined extent would be too large.
1803 state
|= BMAP_LEFT_VALID
;
1804 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, bma
->idx
- 1), &LEFT
);
1806 if (isnullstartblock(LEFT
.br_startblock
))
1807 state
|= BMAP_LEFT_DELAY
;
1810 if ((state
& BMAP_LEFT_VALID
) && !(state
& BMAP_LEFT_DELAY
) &&
1811 LEFT
.br_startoff
+ LEFT
.br_blockcount
== new->br_startoff
&&
1812 LEFT
.br_startblock
+ LEFT
.br_blockcount
== new->br_startblock
&&
1813 LEFT
.br_state
== new->br_state
&&
1814 LEFT
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
1815 state
|= BMAP_LEFT_CONTIG
;
1818 * Check and set flags if this segment has a right neighbor.
1819 * Don't set contiguous if the combined extent would be too large.
1820 * Also check for all-three-contiguous being too large.
1822 if (bma
->idx
< bma
->ip
->i_df
.if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
) - 1) {
1823 state
|= BMAP_RIGHT_VALID
;
1824 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, bma
->idx
+ 1), &RIGHT
);
1826 if (isnullstartblock(RIGHT
.br_startblock
))
1827 state
|= BMAP_RIGHT_DELAY
;
1830 if ((state
& BMAP_RIGHT_VALID
) && !(state
& BMAP_RIGHT_DELAY
) &&
1831 new_endoff
== RIGHT
.br_startoff
&&
1832 new->br_startblock
+ new->br_blockcount
== RIGHT
.br_startblock
&&
1833 new->br_state
== RIGHT
.br_state
&&
1834 new->br_blockcount
+ RIGHT
.br_blockcount
<= MAXEXTLEN
&&
1835 ((state
& (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
1836 BMAP_RIGHT_FILLING
)) !=
1837 (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
1838 BMAP_RIGHT_FILLING
) ||
1839 LEFT
.br_blockcount
+ new->br_blockcount
+ RIGHT
.br_blockcount
1841 state
|= BMAP_RIGHT_CONTIG
;
1845 * Switch out based on the FILLING and CONTIG state bits.
1847 switch (state
& (BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
1848 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
)) {
1849 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
1850 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
1852 * Filling in all of a previously delayed allocation extent.
1853 * The left and right neighbors are both contiguous with new.
1856 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1857 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, bma
->idx
),
1858 LEFT
.br_blockcount
+ PREV
.br_blockcount
+
1859 RIGHT
.br_blockcount
);
1860 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1862 xfs_iext_remove(bma
->ip
, bma
->idx
+ 1, 2, state
);
1863 bma
->ip
->i_d
.di_nextents
--;
1864 if (bma
->cur
== NULL
)
1865 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1867 rval
= XFS_ILOG_CORE
;
1868 error
= xfs_bmbt_lookup_eq(bma
->cur
, RIGHT
.br_startoff
,
1869 RIGHT
.br_startblock
,
1870 RIGHT
.br_blockcount
, &i
);
1873 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1874 error
= xfs_btree_delete(bma
->cur
, &i
);
1877 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1878 error
= xfs_btree_decrement(bma
->cur
, 0, &i
);
1881 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1882 error
= xfs_bmbt_update(bma
->cur
, LEFT
.br_startoff
,
1884 LEFT
.br_blockcount
+
1885 PREV
.br_blockcount
+
1886 RIGHT
.br_blockcount
, LEFT
.br_state
);
1892 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
1894 * Filling in all of a previously delayed allocation extent.
1895 * The left neighbor is contiguous, the right is not.
1899 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1900 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, bma
->idx
),
1901 LEFT
.br_blockcount
+ PREV
.br_blockcount
);
1902 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1904 xfs_iext_remove(bma
->ip
, bma
->idx
+ 1, 1, state
);
1905 if (bma
->cur
== NULL
)
1906 rval
= XFS_ILOG_DEXT
;
1909 error
= xfs_bmbt_lookup_eq(bma
->cur
, LEFT
.br_startoff
,
1910 LEFT
.br_startblock
, LEFT
.br_blockcount
,
1914 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1915 error
= xfs_bmbt_update(bma
->cur
, LEFT
.br_startoff
,
1917 LEFT
.br_blockcount
+
1918 PREV
.br_blockcount
, LEFT
.br_state
);
1924 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
1926 * Filling in all of a previously delayed allocation extent.
1927 * The right neighbor is contiguous, the left is not.
1929 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1930 xfs_bmbt_set_startblock(ep
, new->br_startblock
);
1931 xfs_bmbt_set_blockcount(ep
,
1932 PREV
.br_blockcount
+ RIGHT
.br_blockcount
);
1933 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1935 xfs_iext_remove(bma
->ip
, bma
->idx
+ 1, 1, state
);
1936 if (bma
->cur
== NULL
)
1937 rval
= XFS_ILOG_DEXT
;
1940 error
= xfs_bmbt_lookup_eq(bma
->cur
, RIGHT
.br_startoff
,
1941 RIGHT
.br_startblock
,
1942 RIGHT
.br_blockcount
, &i
);
1945 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1946 error
= xfs_bmbt_update(bma
->cur
, PREV
.br_startoff
,
1948 PREV
.br_blockcount
+
1949 RIGHT
.br_blockcount
, PREV
.br_state
);
1955 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
:
1957 * Filling in all of a previously delayed allocation extent.
1958 * Neither the left nor right neighbors are contiguous with
1961 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1962 xfs_bmbt_set_startblock(ep
, new->br_startblock
);
1963 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1965 bma
->ip
->i_d
.di_nextents
++;
1966 if (bma
->cur
== NULL
)
1967 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1969 rval
= XFS_ILOG_CORE
;
1970 error
= xfs_bmbt_lookup_eq(bma
->cur
, new->br_startoff
,
1971 new->br_startblock
, new->br_blockcount
,
1975 XFS_WANT_CORRUPTED_GOTO(i
== 0, done
);
1976 bma
->cur
->bc_rec
.b
.br_state
= XFS_EXT_NORM
;
1977 error
= xfs_btree_insert(bma
->cur
, &i
);
1980 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1984 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
:
1986 * Filling in the first part of a previous delayed allocation.
1987 * The left neighbor is contiguous.
1989 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
- 1, state
, _THIS_IP_
);
1990 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, bma
->idx
- 1),
1991 LEFT
.br_blockcount
+ new->br_blockcount
);
1992 xfs_bmbt_set_startoff(ep
,
1993 PREV
.br_startoff
+ new->br_blockcount
);
1994 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
- 1, state
, _THIS_IP_
);
1996 temp
= PREV
.br_blockcount
- new->br_blockcount
;
1997 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1998 xfs_bmbt_set_blockcount(ep
, temp
);
1999 if (bma
->cur
== NULL
)
2000 rval
= XFS_ILOG_DEXT
;
2003 error
= xfs_bmbt_lookup_eq(bma
->cur
, LEFT
.br_startoff
,
2004 LEFT
.br_startblock
, LEFT
.br_blockcount
,
2008 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2009 error
= xfs_bmbt_update(bma
->cur
, LEFT
.br_startoff
,
2011 LEFT
.br_blockcount
+
2017 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma
->ip
, temp
),
2018 startblockval(PREV
.br_startblock
));
2019 xfs_bmbt_set_startblock(ep
, nullstartblock(da_new
));
2020 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
2025 case BMAP_LEFT_FILLING
:
2027 * Filling in the first part of a previous delayed allocation.
2028 * The left neighbor is not contiguous.
2030 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
2031 xfs_bmbt_set_startoff(ep
, new_endoff
);
2032 temp
= PREV
.br_blockcount
- new->br_blockcount
;
2033 xfs_bmbt_set_blockcount(ep
, temp
);
2034 xfs_iext_insert(bma
->ip
, bma
->idx
, 1, new, state
);
2035 bma
->ip
->i_d
.di_nextents
++;
2036 if (bma
->cur
== NULL
)
2037 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2039 rval
= XFS_ILOG_CORE
;
2040 error
= xfs_bmbt_lookup_eq(bma
->cur
, new->br_startoff
,
2041 new->br_startblock
, new->br_blockcount
,
2045 XFS_WANT_CORRUPTED_GOTO(i
== 0, done
);
2046 bma
->cur
->bc_rec
.b
.br_state
= XFS_EXT_NORM
;
2047 error
= xfs_btree_insert(bma
->cur
, &i
);
2050 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2053 if (xfs_bmap_needs_btree(bma
->ip
, XFS_DATA_FORK
)) {
2054 error
= xfs_bmap_extents_to_btree(bma
->tp
, bma
->ip
,
2055 bma
->firstblock
, bma
->flist
,
2056 &bma
->cur
, 1, &tmp_rval
, XFS_DATA_FORK
);
2061 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma
->ip
, temp
),
2062 startblockval(PREV
.br_startblock
) -
2063 (bma
->cur
? bma
->cur
->bc_private
.b
.allocated
: 0));
2064 ep
= xfs_iext_get_ext(ifp
, bma
->idx
+ 1);
2065 xfs_bmbt_set_startblock(ep
, nullstartblock(da_new
));
2066 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
+ 1, state
, _THIS_IP_
);
2069 case BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
2071 * Filling in the last part of a previous delayed allocation.
2072 * The right neighbor is contiguous with the new allocation.
2074 temp
= PREV
.br_blockcount
- new->br_blockcount
;
2075 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
+ 1, state
, _THIS_IP_
);
2076 xfs_bmbt_set_blockcount(ep
, temp
);
2077 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp
, bma
->idx
+ 1),
2078 new->br_startoff
, new->br_startblock
,
2079 new->br_blockcount
+ RIGHT
.br_blockcount
,
2081 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
+ 1, state
, _THIS_IP_
);
2082 if (bma
->cur
== NULL
)
2083 rval
= XFS_ILOG_DEXT
;
2086 error
= xfs_bmbt_lookup_eq(bma
->cur
, RIGHT
.br_startoff
,
2087 RIGHT
.br_startblock
,
2088 RIGHT
.br_blockcount
, &i
);
2091 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2092 error
= xfs_bmbt_update(bma
->cur
, new->br_startoff
,
2094 new->br_blockcount
+
2095 RIGHT
.br_blockcount
,
2101 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma
->ip
, temp
),
2102 startblockval(PREV
.br_startblock
));
2103 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
2104 xfs_bmbt_set_startblock(ep
, nullstartblock(da_new
));
2105 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
2110 case BMAP_RIGHT_FILLING
:
2112 * Filling in the last part of a previous delayed allocation.
2113 * The right neighbor is not contiguous.
2115 temp
= PREV
.br_blockcount
- new->br_blockcount
;
2116 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
2117 xfs_bmbt_set_blockcount(ep
, temp
);
2118 xfs_iext_insert(bma
->ip
, bma
->idx
+ 1, 1, new, state
);
2119 bma
->ip
->i_d
.di_nextents
++;
2120 if (bma
->cur
== NULL
)
2121 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2123 rval
= XFS_ILOG_CORE
;
2124 error
= xfs_bmbt_lookup_eq(bma
->cur
, new->br_startoff
,
2125 new->br_startblock
, new->br_blockcount
,
2129 XFS_WANT_CORRUPTED_GOTO(i
== 0, done
);
2130 bma
->cur
->bc_rec
.b
.br_state
= XFS_EXT_NORM
;
2131 error
= xfs_btree_insert(bma
->cur
, &i
);
2134 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2137 if (xfs_bmap_needs_btree(bma
->ip
, XFS_DATA_FORK
)) {
2138 error
= xfs_bmap_extents_to_btree(bma
->tp
, bma
->ip
,
2139 bma
->firstblock
, bma
->flist
, &bma
->cur
, 1,
2140 &tmp_rval
, XFS_DATA_FORK
);
2145 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma
->ip
, temp
),
2146 startblockval(PREV
.br_startblock
) -
2147 (bma
->cur
? bma
->cur
->bc_private
.b
.allocated
: 0));
2148 ep
= xfs_iext_get_ext(ifp
, bma
->idx
);
2149 xfs_bmbt_set_startblock(ep
, nullstartblock(da_new
));
2150 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
2157 * Filling in the middle part of a previous delayed allocation.
2158 * Contiguity is impossible here.
2159 * This case is avoided almost all the time.
2161 * We start with a delayed allocation:
2163 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
2166 * and we are allocating:
2167 * +rrrrrrrrrrrrrrrrr+
2170 * and we set it up for insertion as:
2171 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
2173 * PREV @ idx LEFT RIGHT
2174 * inserted at idx + 1
2176 temp
= new->br_startoff
- PREV
.br_startoff
;
2177 temp2
= PREV
.br_startoff
+ PREV
.br_blockcount
- new_endoff
;
2178 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, 0, _THIS_IP_
);
2179 xfs_bmbt_set_blockcount(ep
, temp
); /* truncate PREV */
2181 RIGHT
.br_state
= PREV
.br_state
;
2182 RIGHT
.br_startblock
= nullstartblock(
2183 (int)xfs_bmap_worst_indlen(bma
->ip
, temp2
));
2184 RIGHT
.br_startoff
= new_endoff
;
2185 RIGHT
.br_blockcount
= temp2
;
2186 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
2187 xfs_iext_insert(bma
->ip
, bma
->idx
+ 1, 2, &LEFT
, state
);
2188 bma
->ip
->i_d
.di_nextents
++;
2189 if (bma
->cur
== NULL
)
2190 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2192 rval
= XFS_ILOG_CORE
;
2193 error
= xfs_bmbt_lookup_eq(bma
->cur
, new->br_startoff
,
2194 new->br_startblock
, new->br_blockcount
,
2198 XFS_WANT_CORRUPTED_GOTO(i
== 0, done
);
2199 bma
->cur
->bc_rec
.b
.br_state
= XFS_EXT_NORM
;
2200 error
= xfs_btree_insert(bma
->cur
, &i
);
2203 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2206 if (xfs_bmap_needs_btree(bma
->ip
, XFS_DATA_FORK
)) {
2207 error
= xfs_bmap_extents_to_btree(bma
->tp
, bma
->ip
,
2208 bma
->firstblock
, bma
->flist
, &bma
->cur
,
2209 1, &tmp_rval
, XFS_DATA_FORK
);
2214 temp
= xfs_bmap_worst_indlen(bma
->ip
, temp
);
2215 temp2
= xfs_bmap_worst_indlen(bma
->ip
, temp2
);
2216 diff
= (int)(temp
+ temp2
- startblockval(PREV
.br_startblock
) -
2217 (bma
->cur
? bma
->cur
->bc_private
.b
.allocated
: 0));
2219 error
= xfs_icsb_modify_counters(bma
->ip
->i_mount
,
2221 -((int64_t)diff
), 0);
2227 ep
= xfs_iext_get_ext(ifp
, bma
->idx
);
2228 xfs_bmbt_set_startblock(ep
, nullstartblock((int)temp
));
2229 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
2230 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
+ 2, state
, _THIS_IP_
);
2231 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp
, bma
->idx
+ 2),
2232 nullstartblock((int)temp2
));
2233 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
+ 2, state
, _THIS_IP_
);
2236 da_new
= temp
+ temp2
;
2239 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2240 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2241 case BMAP_LEFT_FILLING
| BMAP_RIGHT_CONTIG
:
2242 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
2243 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2244 case BMAP_LEFT_CONTIG
:
2245 case BMAP_RIGHT_CONTIG
:
2247 * These cases are all impossible.
2252 /* convert to a btree if necessary */
2253 if (xfs_bmap_needs_btree(bma
->ip
, XFS_DATA_FORK
)) {
2254 int tmp_logflags
; /* partial log flag return val */
2256 ASSERT(bma
->cur
== NULL
);
2257 error
= xfs_bmap_extents_to_btree(bma
->tp
, bma
->ip
,
2258 bma
->firstblock
, bma
->flist
, &bma
->cur
,
2259 da_old
> 0, &tmp_logflags
, XFS_DATA_FORK
);
2260 bma
->logflags
|= tmp_logflags
;
2265 /* adjust for changes in reserved delayed indirect blocks */
2266 if (da_old
|| da_new
) {
2269 temp
+= bma
->cur
->bc_private
.b
.allocated
;
2270 ASSERT(temp
<= da_old
);
2272 xfs_icsb_modify_counters(bma
->ip
->i_mount
,
2274 (int64_t)(da_old
- temp
), 0);
2277 /* clear out the allocated field, done with it now in any case. */
2279 bma
->cur
->bc_private
.b
.allocated
= 0;
2281 xfs_bmap_check_leaf_extents(bma
->cur
, bma
->ip
, XFS_DATA_FORK
);
2283 bma
->logflags
|= rval
;
2291 * Convert an unwritten allocation to a real allocation or vice versa.
2293 STATIC
int /* error */
2294 xfs_bmap_add_extent_unwritten_real(
2295 struct xfs_trans
*tp
,
2296 xfs_inode_t
*ip
, /* incore inode pointer */
2297 xfs_extnum_t
*idx
, /* extent number to update/insert */
2298 xfs_btree_cur_t
**curp
, /* if *curp is null, not a btree */
2299 xfs_bmbt_irec_t
*new, /* new data to add to file extents */
2300 xfs_fsblock_t
*first
, /* pointer to firstblock variable */
2301 xfs_bmap_free_t
*flist
, /* list of extents to be freed */
2302 int *logflagsp
) /* inode logging flags */
2304 xfs_btree_cur_t
*cur
; /* btree cursor */
2305 xfs_bmbt_rec_host_t
*ep
; /* extent entry for idx */
2306 int error
; /* error return value */
2307 int i
; /* temp state */
2308 xfs_ifork_t
*ifp
; /* inode fork pointer */
2309 xfs_fileoff_t new_endoff
; /* end offset of new entry */
2310 xfs_exntst_t newext
; /* new extent state */
2311 xfs_exntst_t oldext
; /* old extent state */
2312 xfs_bmbt_irec_t r
[3]; /* neighbor extent entries */
2313 /* left is 0, right is 1, prev is 2 */
2314 int rval
=0; /* return value (logging flags) */
2315 int state
= 0;/* state bits, accessed thru macros */
2320 ifp
= XFS_IFORK_PTR(ip
, XFS_DATA_FORK
);
2323 ASSERT(*idx
<= ifp
->if_bytes
/ sizeof(struct xfs_bmbt_rec
));
2324 ASSERT(!isnullstartblock(new->br_startblock
));
2326 XFS_STATS_INC(xs_add_exlist
);
2333 * Set up a bunch of variables to make the tests simpler.
2336 ep
= xfs_iext_get_ext(ifp
, *idx
);
2337 xfs_bmbt_get_all(ep
, &PREV
);
2338 newext
= new->br_state
;
2339 oldext
= (newext
== XFS_EXT_UNWRITTEN
) ?
2340 XFS_EXT_NORM
: XFS_EXT_UNWRITTEN
;
2341 ASSERT(PREV
.br_state
== oldext
);
2342 new_endoff
= new->br_startoff
+ new->br_blockcount
;
2343 ASSERT(PREV
.br_startoff
<= new->br_startoff
);
2344 ASSERT(PREV
.br_startoff
+ PREV
.br_blockcount
>= new_endoff
);
2347 * Set flags determining what part of the previous oldext allocation
2348 * extent is being replaced by a newext allocation.
2350 if (PREV
.br_startoff
== new->br_startoff
)
2351 state
|= BMAP_LEFT_FILLING
;
2352 if (PREV
.br_startoff
+ PREV
.br_blockcount
== new_endoff
)
2353 state
|= BMAP_RIGHT_FILLING
;
2356 * Check and set flags if this segment has a left neighbor.
2357 * Don't set contiguous if the combined extent would be too large.
2360 state
|= BMAP_LEFT_VALID
;
2361 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *idx
- 1), &LEFT
);
2363 if (isnullstartblock(LEFT
.br_startblock
))
2364 state
|= BMAP_LEFT_DELAY
;
2367 if ((state
& BMAP_LEFT_VALID
) && !(state
& BMAP_LEFT_DELAY
) &&
2368 LEFT
.br_startoff
+ LEFT
.br_blockcount
== new->br_startoff
&&
2369 LEFT
.br_startblock
+ LEFT
.br_blockcount
== new->br_startblock
&&
2370 LEFT
.br_state
== newext
&&
2371 LEFT
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
2372 state
|= BMAP_LEFT_CONTIG
;
2375 * Check and set flags if this segment has a right neighbor.
2376 * Don't set contiguous if the combined extent would be too large.
2377 * Also check for all-three-contiguous being too large.
2379 if (*idx
< ip
->i_df
.if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
) - 1) {
2380 state
|= BMAP_RIGHT_VALID
;
2381 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *idx
+ 1), &RIGHT
);
2382 if (isnullstartblock(RIGHT
.br_startblock
))
2383 state
|= BMAP_RIGHT_DELAY
;
2386 if ((state
& BMAP_RIGHT_VALID
) && !(state
& BMAP_RIGHT_DELAY
) &&
2387 new_endoff
== RIGHT
.br_startoff
&&
2388 new->br_startblock
+ new->br_blockcount
== RIGHT
.br_startblock
&&
2389 newext
== RIGHT
.br_state
&&
2390 new->br_blockcount
+ RIGHT
.br_blockcount
<= MAXEXTLEN
&&
2391 ((state
& (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
2392 BMAP_RIGHT_FILLING
)) !=
2393 (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
2394 BMAP_RIGHT_FILLING
) ||
2395 LEFT
.br_blockcount
+ new->br_blockcount
+ RIGHT
.br_blockcount
2397 state
|= BMAP_RIGHT_CONTIG
;
2400 * Switch out based on the FILLING and CONTIG state bits.
2402 switch (state
& (BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
2403 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
)) {
2404 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
2405 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
2407 * Setting all of a previous oldext extent to newext.
2408 * The left and right neighbors are both contiguous with new.
2412 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2413 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
),
2414 LEFT
.br_blockcount
+ PREV
.br_blockcount
+
2415 RIGHT
.br_blockcount
);
2416 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2418 xfs_iext_remove(ip
, *idx
+ 1, 2, state
);
2419 ip
->i_d
.di_nextents
-= 2;
2421 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2423 rval
= XFS_ILOG_CORE
;
2424 if ((error
= xfs_bmbt_lookup_eq(cur
, RIGHT
.br_startoff
,
2425 RIGHT
.br_startblock
,
2426 RIGHT
.br_blockcount
, &i
)))
2428 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2429 if ((error
= xfs_btree_delete(cur
, &i
)))
2431 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2432 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
2434 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2435 if ((error
= xfs_btree_delete(cur
, &i
)))
2437 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2438 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
2440 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2441 if ((error
= xfs_bmbt_update(cur
, LEFT
.br_startoff
,
2443 LEFT
.br_blockcount
+ PREV
.br_blockcount
+
2444 RIGHT
.br_blockcount
, LEFT
.br_state
)))
2449 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
2451 * Setting all of a previous oldext extent to newext.
2452 * The left neighbor is contiguous, the right is not.
2456 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2457 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
),
2458 LEFT
.br_blockcount
+ PREV
.br_blockcount
);
2459 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2461 xfs_iext_remove(ip
, *idx
+ 1, 1, state
);
2462 ip
->i_d
.di_nextents
--;
2464 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2466 rval
= XFS_ILOG_CORE
;
2467 if ((error
= xfs_bmbt_lookup_eq(cur
, PREV
.br_startoff
,
2468 PREV
.br_startblock
, PREV
.br_blockcount
,
2471 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2472 if ((error
= xfs_btree_delete(cur
, &i
)))
2474 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2475 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
2477 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2478 if ((error
= xfs_bmbt_update(cur
, LEFT
.br_startoff
,
2480 LEFT
.br_blockcount
+ PREV
.br_blockcount
,
2486 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
2488 * Setting all of a previous oldext extent to newext.
2489 * The right neighbor is contiguous, the left is not.
2491 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2492 xfs_bmbt_set_blockcount(ep
,
2493 PREV
.br_blockcount
+ RIGHT
.br_blockcount
);
2494 xfs_bmbt_set_state(ep
, newext
);
2495 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2496 xfs_iext_remove(ip
, *idx
+ 1, 1, state
);
2497 ip
->i_d
.di_nextents
--;
2499 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2501 rval
= XFS_ILOG_CORE
;
2502 if ((error
= xfs_bmbt_lookup_eq(cur
, RIGHT
.br_startoff
,
2503 RIGHT
.br_startblock
,
2504 RIGHT
.br_blockcount
, &i
)))
2506 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2507 if ((error
= xfs_btree_delete(cur
, &i
)))
2509 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2510 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
2512 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2513 if ((error
= xfs_bmbt_update(cur
, new->br_startoff
,
2515 new->br_blockcount
+ RIGHT
.br_blockcount
,
2521 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
:
2523 * Setting all of a previous oldext extent to newext.
2524 * Neither the left nor right neighbors are contiguous with
2527 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2528 xfs_bmbt_set_state(ep
, newext
);
2529 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2532 rval
= XFS_ILOG_DEXT
;
2535 if ((error
= xfs_bmbt_lookup_eq(cur
, new->br_startoff
,
2536 new->br_startblock
, new->br_blockcount
,
2539 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2540 if ((error
= xfs_bmbt_update(cur
, new->br_startoff
,
2541 new->br_startblock
, new->br_blockcount
,
2547 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
:
2549 * Setting the first part of a previous oldext extent to newext.
2550 * The left neighbor is contiguous.
2552 trace_xfs_bmap_pre_update(ip
, *idx
- 1, state
, _THIS_IP_
);
2553 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
- 1),
2554 LEFT
.br_blockcount
+ new->br_blockcount
);
2555 xfs_bmbt_set_startoff(ep
,
2556 PREV
.br_startoff
+ new->br_blockcount
);
2557 trace_xfs_bmap_post_update(ip
, *idx
- 1, state
, _THIS_IP_
);
2559 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2560 xfs_bmbt_set_startblock(ep
,
2561 new->br_startblock
+ new->br_blockcount
);
2562 xfs_bmbt_set_blockcount(ep
,
2563 PREV
.br_blockcount
- new->br_blockcount
);
2564 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2569 rval
= XFS_ILOG_DEXT
;
2572 if ((error
= xfs_bmbt_lookup_eq(cur
, PREV
.br_startoff
,
2573 PREV
.br_startblock
, PREV
.br_blockcount
,
2576 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2577 if ((error
= xfs_bmbt_update(cur
,
2578 PREV
.br_startoff
+ new->br_blockcount
,
2579 PREV
.br_startblock
+ new->br_blockcount
,
2580 PREV
.br_blockcount
- new->br_blockcount
,
2583 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
2585 error
= xfs_bmbt_update(cur
, LEFT
.br_startoff
,
2587 LEFT
.br_blockcount
+ new->br_blockcount
,
2594 case BMAP_LEFT_FILLING
:
2596 * Setting the first part of a previous oldext extent to newext.
2597 * The left neighbor is not contiguous.
2599 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2600 ASSERT(ep
&& xfs_bmbt_get_state(ep
) == oldext
);
2601 xfs_bmbt_set_startoff(ep
, new_endoff
);
2602 xfs_bmbt_set_blockcount(ep
,
2603 PREV
.br_blockcount
- new->br_blockcount
);
2604 xfs_bmbt_set_startblock(ep
,
2605 new->br_startblock
+ new->br_blockcount
);
2606 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2608 xfs_iext_insert(ip
, *idx
, 1, new, state
);
2609 ip
->i_d
.di_nextents
++;
2611 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2613 rval
= XFS_ILOG_CORE
;
2614 if ((error
= xfs_bmbt_lookup_eq(cur
, PREV
.br_startoff
,
2615 PREV
.br_startblock
, PREV
.br_blockcount
,
2618 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2619 if ((error
= xfs_bmbt_update(cur
,
2620 PREV
.br_startoff
+ new->br_blockcount
,
2621 PREV
.br_startblock
+ new->br_blockcount
,
2622 PREV
.br_blockcount
- new->br_blockcount
,
2625 cur
->bc_rec
.b
= *new;
2626 if ((error
= xfs_btree_insert(cur
, &i
)))
2628 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2632 case BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
2634 * Setting the last part of a previous oldext extent to newext.
2635 * The right neighbor is contiguous with the new allocation.
2637 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2638 xfs_bmbt_set_blockcount(ep
,
2639 PREV
.br_blockcount
- new->br_blockcount
);
2640 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2644 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2645 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp
, *idx
),
2646 new->br_startoff
, new->br_startblock
,
2647 new->br_blockcount
+ RIGHT
.br_blockcount
, newext
);
2648 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2651 rval
= XFS_ILOG_DEXT
;
2654 if ((error
= xfs_bmbt_lookup_eq(cur
, PREV
.br_startoff
,
2656 PREV
.br_blockcount
, &i
)))
2658 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2659 if ((error
= xfs_bmbt_update(cur
, PREV
.br_startoff
,
2661 PREV
.br_blockcount
- new->br_blockcount
,
2664 if ((error
= xfs_btree_increment(cur
, 0, &i
)))
2666 if ((error
= xfs_bmbt_update(cur
, new->br_startoff
,
2668 new->br_blockcount
+ RIGHT
.br_blockcount
,
2674 case BMAP_RIGHT_FILLING
:
2676 * Setting the last part of a previous oldext extent to newext.
2677 * The right neighbor is not contiguous.
2679 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2680 xfs_bmbt_set_blockcount(ep
,
2681 PREV
.br_blockcount
- new->br_blockcount
);
2682 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2685 xfs_iext_insert(ip
, *idx
, 1, new, state
);
2687 ip
->i_d
.di_nextents
++;
2689 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2691 rval
= XFS_ILOG_CORE
;
2692 if ((error
= xfs_bmbt_lookup_eq(cur
, PREV
.br_startoff
,
2693 PREV
.br_startblock
, PREV
.br_blockcount
,
2696 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2697 if ((error
= xfs_bmbt_update(cur
, PREV
.br_startoff
,
2699 PREV
.br_blockcount
- new->br_blockcount
,
2702 if ((error
= xfs_bmbt_lookup_eq(cur
, new->br_startoff
,
2703 new->br_startblock
, new->br_blockcount
,
2706 XFS_WANT_CORRUPTED_GOTO(i
== 0, done
);
2707 cur
->bc_rec
.b
.br_state
= XFS_EXT_NORM
;
2708 if ((error
= xfs_btree_insert(cur
, &i
)))
2710 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2716 * Setting the middle part of a previous oldext extent to
2717 * newext. Contiguity is impossible here.
2718 * One extent becomes three extents.
2720 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2721 xfs_bmbt_set_blockcount(ep
,
2722 new->br_startoff
- PREV
.br_startoff
);
2723 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2726 r
[1].br_startoff
= new_endoff
;
2727 r
[1].br_blockcount
=
2728 PREV
.br_startoff
+ PREV
.br_blockcount
- new_endoff
;
2729 r
[1].br_startblock
= new->br_startblock
+ new->br_blockcount
;
2730 r
[1].br_state
= oldext
;
2733 xfs_iext_insert(ip
, *idx
, 2, &r
[0], state
);
2735 ip
->i_d
.di_nextents
+= 2;
2737 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2739 rval
= XFS_ILOG_CORE
;
2740 if ((error
= xfs_bmbt_lookup_eq(cur
, PREV
.br_startoff
,
2741 PREV
.br_startblock
, PREV
.br_blockcount
,
2744 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2745 /* new right extent - oldext */
2746 if ((error
= xfs_bmbt_update(cur
, r
[1].br_startoff
,
2747 r
[1].br_startblock
, r
[1].br_blockcount
,
2750 /* new left extent - oldext */
2751 cur
->bc_rec
.b
= PREV
;
2752 cur
->bc_rec
.b
.br_blockcount
=
2753 new->br_startoff
- PREV
.br_startoff
;
2754 if ((error
= xfs_btree_insert(cur
, &i
)))
2756 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2758 * Reset the cursor to the position of the new extent
2759 * we are about to insert as we can't trust it after
2760 * the previous insert.
2762 if ((error
= xfs_bmbt_lookup_eq(cur
, new->br_startoff
,
2763 new->br_startblock
, new->br_blockcount
,
2766 XFS_WANT_CORRUPTED_GOTO(i
== 0, done
);
2767 /* new middle extent - newext */
2768 cur
->bc_rec
.b
.br_state
= new->br_state
;
2769 if ((error
= xfs_btree_insert(cur
, &i
)))
2771 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2775 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2776 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2777 case BMAP_LEFT_FILLING
| BMAP_RIGHT_CONTIG
:
2778 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
2779 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2780 case BMAP_LEFT_CONTIG
:
2781 case BMAP_RIGHT_CONTIG
:
2783 * These cases are all impossible.
2788 /* convert to a btree if necessary */
2789 if (xfs_bmap_needs_btree(ip
, XFS_DATA_FORK
)) {
2790 int tmp_logflags
; /* partial log flag return val */
2792 ASSERT(cur
== NULL
);
2793 error
= xfs_bmap_extents_to_btree(tp
, ip
, first
, flist
, &cur
,
2794 0, &tmp_logflags
, XFS_DATA_FORK
);
2795 *logflagsp
|= tmp_logflags
;
2800 /* clear out the allocated field, done with it now in any case. */
2802 cur
->bc_private
.b
.allocated
= 0;
2806 xfs_bmap_check_leaf_extents(*curp
, ip
, XFS_DATA_FORK
);
2816 * Convert a hole to a delayed allocation.
2819 xfs_bmap_add_extent_hole_delay(
2820 xfs_inode_t
*ip
, /* incore inode pointer */
2821 xfs_extnum_t
*idx
, /* extent number to update/insert */
2822 xfs_bmbt_irec_t
*new) /* new data to add to file extents */
2824 xfs_ifork_t
*ifp
; /* inode fork pointer */
2825 xfs_bmbt_irec_t left
; /* left neighbor extent entry */
2826 xfs_filblks_t newlen
=0; /* new indirect size */
2827 xfs_filblks_t oldlen
=0; /* old indirect size */
2828 xfs_bmbt_irec_t right
; /* right neighbor extent entry */
2829 int state
; /* state bits, accessed thru macros */
2830 xfs_filblks_t temp
=0; /* temp for indirect calculations */
2832 ifp
= XFS_IFORK_PTR(ip
, XFS_DATA_FORK
);
2834 ASSERT(isnullstartblock(new->br_startblock
));
2837 * Check and set flags if this segment has a left neighbor
2840 state
|= BMAP_LEFT_VALID
;
2841 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *idx
- 1), &left
);
2843 if (isnullstartblock(left
.br_startblock
))
2844 state
|= BMAP_LEFT_DELAY
;
2848 * Check and set flags if the current (right) segment exists.
2849 * If it doesn't exist, we're converting the hole at end-of-file.
2851 if (*idx
< ip
->i_df
.if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
)) {
2852 state
|= BMAP_RIGHT_VALID
;
2853 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *idx
), &right
);
2855 if (isnullstartblock(right
.br_startblock
))
2856 state
|= BMAP_RIGHT_DELAY
;
2860 * Set contiguity flags on the left and right neighbors.
2861 * Don't let extents get too large, even if the pieces are contiguous.
2863 if ((state
& BMAP_LEFT_VALID
) && (state
& BMAP_LEFT_DELAY
) &&
2864 left
.br_startoff
+ left
.br_blockcount
== new->br_startoff
&&
2865 left
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
2866 state
|= BMAP_LEFT_CONTIG
;
2868 if ((state
& BMAP_RIGHT_VALID
) && (state
& BMAP_RIGHT_DELAY
) &&
2869 new->br_startoff
+ new->br_blockcount
== right
.br_startoff
&&
2870 new->br_blockcount
+ right
.br_blockcount
<= MAXEXTLEN
&&
2871 (!(state
& BMAP_LEFT_CONTIG
) ||
2872 (left
.br_blockcount
+ new->br_blockcount
+
2873 right
.br_blockcount
<= MAXEXTLEN
)))
2874 state
|= BMAP_RIGHT_CONTIG
;
2877 * Switch out based on the contiguity flags.
2879 switch (state
& (BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
)) {
2880 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2882 * New allocation is contiguous with delayed allocations
2883 * on the left and on the right.
2884 * Merge all three into a single extent record.
2887 temp
= left
.br_blockcount
+ new->br_blockcount
+
2888 right
.br_blockcount
;
2890 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2891 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
), temp
);
2892 oldlen
= startblockval(left
.br_startblock
) +
2893 startblockval(new->br_startblock
) +
2894 startblockval(right
.br_startblock
);
2895 newlen
= xfs_bmap_worst_indlen(ip
, temp
);
2896 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp
, *idx
),
2897 nullstartblock((int)newlen
));
2898 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2900 xfs_iext_remove(ip
, *idx
+ 1, 1, state
);
2903 case BMAP_LEFT_CONTIG
:
2905 * New allocation is contiguous with a delayed allocation
2907 * Merge the new allocation with the left neighbor.
2910 temp
= left
.br_blockcount
+ new->br_blockcount
;
2912 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2913 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
), temp
);
2914 oldlen
= startblockval(left
.br_startblock
) +
2915 startblockval(new->br_startblock
);
2916 newlen
= xfs_bmap_worst_indlen(ip
, temp
);
2917 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp
, *idx
),
2918 nullstartblock((int)newlen
));
2919 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2922 case BMAP_RIGHT_CONTIG
:
2924 * New allocation is contiguous with a delayed allocation
2926 * Merge the new allocation with the right neighbor.
2928 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2929 temp
= new->br_blockcount
+ right
.br_blockcount
;
2930 oldlen
= startblockval(new->br_startblock
) +
2931 startblockval(right
.br_startblock
);
2932 newlen
= xfs_bmap_worst_indlen(ip
, temp
);
2933 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp
, *idx
),
2935 nullstartblock((int)newlen
), temp
, right
.br_state
);
2936 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2941 * New allocation is not contiguous with another
2942 * delayed allocation.
2943 * Insert a new entry.
2945 oldlen
= newlen
= 0;
2946 xfs_iext_insert(ip
, *idx
, 1, new, state
);
2949 if (oldlen
!= newlen
) {
2950 ASSERT(oldlen
> newlen
);
2951 xfs_icsb_modify_counters(ip
->i_mount
, XFS_SBS_FDBLOCKS
,
2952 (int64_t)(oldlen
- newlen
), 0);
2954 * Nothing to do for disk quota accounting here.
2960 * Convert a hole to a real allocation.
2962 STATIC
int /* error */
2963 xfs_bmap_add_extent_hole_real(
2964 struct xfs_bmalloca
*bma
,
2967 struct xfs_bmbt_irec
*new = &bma
->got
;
2968 int error
; /* error return value */
2969 int i
; /* temp state */
2970 xfs_ifork_t
*ifp
; /* inode fork pointer */
2971 xfs_bmbt_irec_t left
; /* left neighbor extent entry */
2972 xfs_bmbt_irec_t right
; /* right neighbor extent entry */
2973 int rval
=0; /* return value (logging flags) */
2974 int state
; /* state bits, accessed thru macros */
2976 ifp
= XFS_IFORK_PTR(bma
->ip
, whichfork
);
2978 ASSERT(bma
->idx
>= 0);
2979 ASSERT(bma
->idx
<= ifp
->if_bytes
/ sizeof(struct xfs_bmbt_rec
));
2980 ASSERT(!isnullstartblock(new->br_startblock
));
2982 !(bma
->cur
->bc_private
.b
.flags
& XFS_BTCUR_BPRV_WASDEL
));
2984 XFS_STATS_INC(xs_add_exlist
);
2987 if (whichfork
== XFS_ATTR_FORK
)
2988 state
|= BMAP_ATTRFORK
;
2991 * Check and set flags if this segment has a left neighbor.
2994 state
|= BMAP_LEFT_VALID
;
2995 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, bma
->idx
- 1), &left
);
2996 if (isnullstartblock(left
.br_startblock
))
2997 state
|= BMAP_LEFT_DELAY
;
3001 * Check and set flags if this segment has a current value.
3002 * Not true if we're inserting into the "hole" at eof.
3004 if (bma
->idx
< ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
)) {
3005 state
|= BMAP_RIGHT_VALID
;
3006 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, bma
->idx
), &right
);
3007 if (isnullstartblock(right
.br_startblock
))
3008 state
|= BMAP_RIGHT_DELAY
;
3012 * We're inserting a real allocation between "left" and "right".
3013 * Set the contiguity flags. Don't let extents get too large.
3015 if ((state
& BMAP_LEFT_VALID
) && !(state
& BMAP_LEFT_DELAY
) &&
3016 left
.br_startoff
+ left
.br_blockcount
== new->br_startoff
&&
3017 left
.br_startblock
+ left
.br_blockcount
== new->br_startblock
&&
3018 left
.br_state
== new->br_state
&&
3019 left
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
3020 state
|= BMAP_LEFT_CONTIG
;
3022 if ((state
& BMAP_RIGHT_VALID
) && !(state
& BMAP_RIGHT_DELAY
) &&
3023 new->br_startoff
+ new->br_blockcount
== right
.br_startoff
&&
3024 new->br_startblock
+ new->br_blockcount
== right
.br_startblock
&&
3025 new->br_state
== right
.br_state
&&
3026 new->br_blockcount
+ right
.br_blockcount
<= MAXEXTLEN
&&
3027 (!(state
& BMAP_LEFT_CONTIG
) ||
3028 left
.br_blockcount
+ new->br_blockcount
+
3029 right
.br_blockcount
<= MAXEXTLEN
))
3030 state
|= BMAP_RIGHT_CONTIG
;
3034 * Select which case we're in here, and implement it.
3036 switch (state
& (BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
)) {
3037 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
3039 * New allocation is contiguous with real allocations on the
3040 * left and on the right.
3041 * Merge all three into a single extent record.
3044 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
3045 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, bma
->idx
),
3046 left
.br_blockcount
+ new->br_blockcount
+
3047 right
.br_blockcount
);
3048 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
3050 xfs_iext_remove(bma
->ip
, bma
->idx
+ 1, 1, state
);
3052 XFS_IFORK_NEXT_SET(bma
->ip
, whichfork
,
3053 XFS_IFORK_NEXTENTS(bma
->ip
, whichfork
) - 1);
3054 if (bma
->cur
== NULL
) {
3055 rval
= XFS_ILOG_CORE
| xfs_ilog_fext(whichfork
);
3057 rval
= XFS_ILOG_CORE
;
3058 error
= xfs_bmbt_lookup_eq(bma
->cur
, right
.br_startoff
,
3059 right
.br_startblock
, right
.br_blockcount
,
3063 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
3064 error
= xfs_btree_delete(bma
->cur
, &i
);
3067 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
3068 error
= xfs_btree_decrement(bma
->cur
, 0, &i
);
3071 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
3072 error
= xfs_bmbt_update(bma
->cur
, left
.br_startoff
,
3074 left
.br_blockcount
+
3075 new->br_blockcount
+
3076 right
.br_blockcount
,
3083 case BMAP_LEFT_CONTIG
:
3085 * New allocation is contiguous with a real allocation
3087 * Merge the new allocation with the left neighbor.
3090 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
3091 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, bma
->idx
),
3092 left
.br_blockcount
+ new->br_blockcount
);
3093 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
3095 if (bma
->cur
== NULL
) {
3096 rval
= xfs_ilog_fext(whichfork
);
3099 error
= xfs_bmbt_lookup_eq(bma
->cur
, left
.br_startoff
,
3100 left
.br_startblock
, left
.br_blockcount
,
3104 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
3105 error
= xfs_bmbt_update(bma
->cur
, left
.br_startoff
,
3107 left
.br_blockcount
+
3115 case BMAP_RIGHT_CONTIG
:
3117 * New allocation is contiguous with a real allocation
3119 * Merge the new allocation with the right neighbor.
3121 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
3122 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp
, bma
->idx
),
3123 new->br_startoff
, new->br_startblock
,
3124 new->br_blockcount
+ right
.br_blockcount
,
3126 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
3128 if (bma
->cur
== NULL
) {
3129 rval
= xfs_ilog_fext(whichfork
);
3132 error
= xfs_bmbt_lookup_eq(bma
->cur
,
3134 right
.br_startblock
,
3135 right
.br_blockcount
, &i
);
3138 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
3139 error
= xfs_bmbt_update(bma
->cur
, new->br_startoff
,
3141 new->br_blockcount
+
3142 right
.br_blockcount
,
3151 * New allocation is not contiguous with another
3153 * Insert a new entry.
3155 xfs_iext_insert(bma
->ip
, bma
->idx
, 1, new, state
);
3156 XFS_IFORK_NEXT_SET(bma
->ip
, whichfork
,
3157 XFS_IFORK_NEXTENTS(bma
->ip
, whichfork
) + 1);
3158 if (bma
->cur
== NULL
) {
3159 rval
= XFS_ILOG_CORE
| xfs_ilog_fext(whichfork
);
3161 rval
= XFS_ILOG_CORE
;
3162 error
= xfs_bmbt_lookup_eq(bma
->cur
,
3165 new->br_blockcount
, &i
);
3168 XFS_WANT_CORRUPTED_GOTO(i
== 0, done
);
3169 bma
->cur
->bc_rec
.b
.br_state
= new->br_state
;
3170 error
= xfs_btree_insert(bma
->cur
, &i
);
3173 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
3178 /* convert to a btree if necessary */
3179 if (xfs_bmap_needs_btree(bma
->ip
, whichfork
)) {
3180 int tmp_logflags
; /* partial log flag return val */
3182 ASSERT(bma
->cur
== NULL
);
3183 error
= xfs_bmap_extents_to_btree(bma
->tp
, bma
->ip
,
3184 bma
->firstblock
, bma
->flist
, &bma
->cur
,
3185 0, &tmp_logflags
, whichfork
);
3186 bma
->logflags
|= tmp_logflags
;
3191 /* clear out the allocated field, done with it now in any case. */
3193 bma
->cur
->bc_private
.b
.allocated
= 0;
3195 xfs_bmap_check_leaf_extents(bma
->cur
, bma
->ip
, whichfork
);
3197 bma
->logflags
|= rval
;
3202 * Functions used in the extent read, allocate and remove paths
3206 * Adjust the size of the new extent based on di_extsize and rt extsize.
3209 xfs_bmap_extsize_align(
3211 xfs_bmbt_irec_t
*gotp
, /* next extent pointer */
3212 xfs_bmbt_irec_t
*prevp
, /* previous extent pointer */
3213 xfs_extlen_t extsz
, /* align to this extent size */
3214 int rt
, /* is this a realtime inode? */
3215 int eof
, /* is extent at end-of-file? */
3216 int delay
, /* creating delalloc extent? */
3217 int convert
, /* overwriting unwritten extent? */
3218 xfs_fileoff_t
*offp
, /* in/out: aligned offset */
3219 xfs_extlen_t
*lenp
) /* in/out: aligned length */
3221 xfs_fileoff_t orig_off
; /* original offset */
3222 xfs_extlen_t orig_alen
; /* original length */
3223 xfs_fileoff_t orig_end
; /* original off+len */
3224 xfs_fileoff_t nexto
; /* next file offset */
3225 xfs_fileoff_t prevo
; /* previous file offset */
3226 xfs_fileoff_t align_off
; /* temp for offset */
3227 xfs_extlen_t align_alen
; /* temp for length */
3228 xfs_extlen_t temp
; /* temp for calculations */
3233 orig_off
= align_off
= *offp
;
3234 orig_alen
= align_alen
= *lenp
;
3235 orig_end
= orig_off
+ orig_alen
;
3238 * If this request overlaps an existing extent, then don't
3239 * attempt to perform any additional alignment.
3241 if (!delay
&& !eof
&&
3242 (orig_off
>= gotp
->br_startoff
) &&
3243 (orig_end
<= gotp
->br_startoff
+ gotp
->br_blockcount
)) {
3248 * If the file offset is unaligned vs. the extent size
3249 * we need to align it. This will be possible unless
3250 * the file was previously written with a kernel that didn't
3251 * perform this alignment, or if a truncate shot us in the
3254 temp
= do_mod(orig_off
, extsz
);
3260 * Same adjustment for the end of the requested area.
3262 if ((temp
= (align_alen
% extsz
))) {
3263 align_alen
+= extsz
- temp
;
3266 * If the previous block overlaps with this proposed allocation
3267 * then move the start forward without adjusting the length.
3269 if (prevp
->br_startoff
!= NULLFILEOFF
) {
3270 if (prevp
->br_startblock
== HOLESTARTBLOCK
)
3271 prevo
= prevp
->br_startoff
;
3273 prevo
= prevp
->br_startoff
+ prevp
->br_blockcount
;
3276 if (align_off
!= orig_off
&& align_off
< prevo
)
3279 * If the next block overlaps with this proposed allocation
3280 * then move the start back without adjusting the length,
3281 * but not before offset 0.
3282 * This may of course make the start overlap previous block,
3283 * and if we hit the offset 0 limit then the next block
3284 * can still overlap too.
3286 if (!eof
&& gotp
->br_startoff
!= NULLFILEOFF
) {
3287 if ((delay
&& gotp
->br_startblock
== HOLESTARTBLOCK
) ||
3288 (!delay
&& gotp
->br_startblock
== DELAYSTARTBLOCK
))
3289 nexto
= gotp
->br_startoff
+ gotp
->br_blockcount
;
3291 nexto
= gotp
->br_startoff
;
3293 nexto
= NULLFILEOFF
;
3295 align_off
+ align_alen
!= orig_end
&&
3296 align_off
+ align_alen
> nexto
)
3297 align_off
= nexto
> align_alen
? nexto
- align_alen
: 0;
3299 * If we're now overlapping the next or previous extent that
3300 * means we can't fit an extsz piece in this hole. Just move
3301 * the start forward to the first valid spot and set
3302 * the length so we hit the end.
3304 if (align_off
!= orig_off
&& align_off
< prevo
)
3306 if (align_off
+ align_alen
!= orig_end
&&
3307 align_off
+ align_alen
> nexto
&&
3308 nexto
!= NULLFILEOFF
) {
3309 ASSERT(nexto
> prevo
);
3310 align_alen
= nexto
- align_off
;
3314 * If realtime, and the result isn't a multiple of the realtime
3315 * extent size we need to remove blocks until it is.
3317 if (rt
&& (temp
= (align_alen
% mp
->m_sb
.sb_rextsize
))) {
3319 * We're not covering the original request, or
3320 * we won't be able to once we fix the length.
3322 if (orig_off
< align_off
||
3323 orig_end
> align_off
+ align_alen
||
3324 align_alen
- temp
< orig_alen
)
3325 return XFS_ERROR(EINVAL
);
3327 * Try to fix it by moving the start up.
3329 if (align_off
+ temp
<= orig_off
) {
3334 * Try to fix it by moving the end in.
3336 else if (align_off
+ align_alen
- temp
>= orig_end
)
3339 * Set the start to the minimum then trim the length.
3342 align_alen
-= orig_off
- align_off
;
3343 align_off
= orig_off
;
3344 align_alen
-= align_alen
% mp
->m_sb
.sb_rextsize
;
3347 * Result doesn't cover the request, fail it.
3349 if (orig_off
< align_off
|| orig_end
> align_off
+ align_alen
)
3350 return XFS_ERROR(EINVAL
);
3352 ASSERT(orig_off
>= align_off
);
3353 ASSERT(orig_end
<= align_off
+ align_alen
);
3357 if (!eof
&& gotp
->br_startoff
!= NULLFILEOFF
)
3358 ASSERT(align_off
+ align_alen
<= gotp
->br_startoff
);
3359 if (prevp
->br_startoff
!= NULLFILEOFF
)
3360 ASSERT(align_off
>= prevp
->br_startoff
+ prevp
->br_blockcount
);
3368 #define XFS_ALLOC_GAP_UNITS 4
3372 struct xfs_bmalloca
*ap
) /* bmap alloc argument struct */
3374 xfs_fsblock_t adjust
; /* adjustment to block numbers */
3375 xfs_agnumber_t fb_agno
; /* ag number of ap->firstblock */
3376 xfs_mount_t
*mp
; /* mount point structure */
3377 int nullfb
; /* true if ap->firstblock isn't set */
3378 int rt
; /* true if inode is realtime */
3380 #define ISVALID(x,y) \
3382 (x) < mp->m_sb.sb_rblocks : \
3383 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3384 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3385 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3387 mp
= ap
->ip
->i_mount
;
3388 nullfb
= *ap
->firstblock
== NULLFSBLOCK
;
3389 rt
= XFS_IS_REALTIME_INODE(ap
->ip
) && ap
->userdata
;
3390 fb_agno
= nullfb
? NULLAGNUMBER
: XFS_FSB_TO_AGNO(mp
, *ap
->firstblock
);
3392 * If allocating at eof, and there's a previous real block,
3393 * try to use its last block as our starting point.
3395 if (ap
->eof
&& ap
->prev
.br_startoff
!= NULLFILEOFF
&&
3396 !isnullstartblock(ap
->prev
.br_startblock
) &&
3397 ISVALID(ap
->prev
.br_startblock
+ ap
->prev
.br_blockcount
,
3398 ap
->prev
.br_startblock
)) {
3399 ap
->blkno
= ap
->prev
.br_startblock
+ ap
->prev
.br_blockcount
;
3401 * Adjust for the gap between prevp and us.
3403 adjust
= ap
->offset
-
3404 (ap
->prev
.br_startoff
+ ap
->prev
.br_blockcount
);
3406 ISVALID(ap
->blkno
+ adjust
, ap
->prev
.br_startblock
))
3407 ap
->blkno
+= adjust
;
3410 * If not at eof, then compare the two neighbor blocks.
3411 * Figure out whether either one gives us a good starting point,
3412 * and pick the better one.
3414 else if (!ap
->eof
) {
3415 xfs_fsblock_t gotbno
; /* right side block number */
3416 xfs_fsblock_t gotdiff
=0; /* right side difference */
3417 xfs_fsblock_t prevbno
; /* left side block number */
3418 xfs_fsblock_t prevdiff
=0; /* left side difference */
3421 * If there's a previous (left) block, select a requested
3422 * start block based on it.
3424 if (ap
->prev
.br_startoff
!= NULLFILEOFF
&&
3425 !isnullstartblock(ap
->prev
.br_startblock
) &&
3426 (prevbno
= ap
->prev
.br_startblock
+
3427 ap
->prev
.br_blockcount
) &&
3428 ISVALID(prevbno
, ap
->prev
.br_startblock
)) {
3430 * Calculate gap to end of previous block.
3432 adjust
= prevdiff
= ap
->offset
-
3433 (ap
->prev
.br_startoff
+
3434 ap
->prev
.br_blockcount
);
3436 * Figure the startblock based on the previous block's
3437 * end and the gap size.
3439 * If the gap is large relative to the piece we're
3440 * allocating, or using it gives us an invalid block
3441 * number, then just use the end of the previous block.
3443 if (prevdiff
<= XFS_ALLOC_GAP_UNITS
* ap
->length
&&
3444 ISVALID(prevbno
+ prevdiff
,
3445 ap
->prev
.br_startblock
))
3450 * If the firstblock forbids it, can't use it,
3453 if (!rt
&& !nullfb
&&
3454 XFS_FSB_TO_AGNO(mp
, prevbno
) != fb_agno
)
3455 prevbno
= NULLFSBLOCK
;
3458 * No previous block or can't follow it, just default.
3461 prevbno
= NULLFSBLOCK
;
3463 * If there's a following (right) block, select a requested
3464 * start block based on it.
3466 if (!isnullstartblock(ap
->got
.br_startblock
)) {
3468 * Calculate gap to start of next block.
3470 adjust
= gotdiff
= ap
->got
.br_startoff
- ap
->offset
;
3472 * Figure the startblock based on the next block's
3473 * start and the gap size.
3475 gotbno
= ap
->got
.br_startblock
;
3478 * If the gap is large relative to the piece we're
3479 * allocating, or using it gives us an invalid block
3480 * number, then just use the start of the next block
3481 * offset by our length.
3483 if (gotdiff
<= XFS_ALLOC_GAP_UNITS
* ap
->length
&&
3484 ISVALID(gotbno
- gotdiff
, gotbno
))
3486 else if (ISVALID(gotbno
- ap
->length
, gotbno
)) {
3487 gotbno
-= ap
->length
;
3488 gotdiff
+= adjust
- ap
->length
;
3492 * If the firstblock forbids it, can't use it,
3495 if (!rt
&& !nullfb
&&
3496 XFS_FSB_TO_AGNO(mp
, gotbno
) != fb_agno
)
3497 gotbno
= NULLFSBLOCK
;
3500 * No next block, just default.
3503 gotbno
= NULLFSBLOCK
;
3505 * If both valid, pick the better one, else the only good
3506 * one, else ap->blkno is already set (to 0 or the inode block).
3508 if (prevbno
!= NULLFSBLOCK
&& gotbno
!= NULLFSBLOCK
)
3509 ap
->blkno
= prevdiff
<= gotdiff
? prevbno
: gotbno
;
3510 else if (prevbno
!= NULLFSBLOCK
)
3511 ap
->blkno
= prevbno
;
3512 else if (gotbno
!= NULLFSBLOCK
)
3519 xfs_bmap_longest_free_extent(
3520 struct xfs_trans
*tp
,
3525 struct xfs_mount
*mp
= tp
->t_mountp
;
3526 struct xfs_perag
*pag
;
3527 xfs_extlen_t longest
;
3530 pag
= xfs_perag_get(mp
, ag
);
3531 if (!pag
->pagf_init
) {
3532 error
= xfs_alloc_pagf_init(mp
, tp
, ag
, XFS_ALLOC_FLAG_TRYLOCK
);
3536 if (!pag
->pagf_init
) {
3542 longest
= xfs_alloc_longest_free_extent(mp
, pag
);
3543 if (*blen
< longest
)
3552 xfs_bmap_select_minlen(
3553 struct xfs_bmalloca
*ap
,
3554 struct xfs_alloc_arg
*args
,
3558 if (notinit
|| *blen
< ap
->minlen
) {
3560 * Since we did a BUF_TRYLOCK above, it is possible that
3561 * there is space for this request.
3563 args
->minlen
= ap
->minlen
;
3564 } else if (*blen
< args
->maxlen
) {
3566 * If the best seen length is less than the request length,
3567 * use the best as the minimum.
3569 args
->minlen
= *blen
;
3572 * Otherwise we've seen an extent as big as maxlen, use that
3575 args
->minlen
= args
->maxlen
;
3580 xfs_bmap_btalloc_nullfb(
3581 struct xfs_bmalloca
*ap
,
3582 struct xfs_alloc_arg
*args
,
3585 struct xfs_mount
*mp
= ap
->ip
->i_mount
;
3586 xfs_agnumber_t ag
, startag
;
3590 args
->type
= XFS_ALLOCTYPE_START_BNO
;
3591 args
->total
= ap
->total
;
3593 startag
= ag
= XFS_FSB_TO_AGNO(mp
, args
->fsbno
);
3594 if (startag
== NULLAGNUMBER
)
3597 while (*blen
< args
->maxlen
) {
3598 error
= xfs_bmap_longest_free_extent(args
->tp
, ag
, blen
,
3603 if (++ag
== mp
->m_sb
.sb_agcount
)
3609 xfs_bmap_select_minlen(ap
, args
, blen
, notinit
);
3614 xfs_bmap_btalloc_filestreams(
3615 struct xfs_bmalloca
*ap
,
3616 struct xfs_alloc_arg
*args
,
3619 struct xfs_mount
*mp
= ap
->ip
->i_mount
;
3624 args
->type
= XFS_ALLOCTYPE_NEAR_BNO
;
3625 args
->total
= ap
->total
;
3627 ag
= XFS_FSB_TO_AGNO(mp
, args
->fsbno
);
3628 if (ag
== NULLAGNUMBER
)
3631 error
= xfs_bmap_longest_free_extent(args
->tp
, ag
, blen
, ¬init
);
3635 if (*blen
< args
->maxlen
) {
3636 error
= xfs_filestream_new_ag(ap
, &ag
);
3640 error
= xfs_bmap_longest_free_extent(args
->tp
, ag
, blen
,
3647 xfs_bmap_select_minlen(ap
, args
, blen
, notinit
);
3650 * Set the failure fallback case to look in the selected AG as stream
3653 ap
->blkno
= args
->fsbno
= XFS_AGB_TO_FSB(mp
, ag
, 0);
3659 struct xfs_bmalloca
*ap
) /* bmap alloc argument struct */
3661 xfs_mount_t
*mp
; /* mount point structure */
3662 xfs_alloctype_t atype
= 0; /* type for allocation routines */
3663 xfs_extlen_t align
; /* minimum allocation alignment */
3664 xfs_agnumber_t fb_agno
; /* ag number of ap->firstblock */
3666 xfs_alloc_arg_t args
;
3668 xfs_extlen_t nextminlen
= 0;
3669 int nullfb
; /* true if ap->firstblock isn't set */
3677 mp
= ap
->ip
->i_mount
;
3679 /* stripe alignment for allocation is determined by mount parameters */
3681 if (mp
->m_swidth
&& (mp
->m_flags
& XFS_MOUNT_SWALLOC
))
3682 stripe_align
= mp
->m_swidth
;
3683 else if (mp
->m_dalign
)
3684 stripe_align
= mp
->m_dalign
;
3686 align
= ap
->userdata
? xfs_get_extsz_hint(ap
->ip
) : 0;
3687 if (unlikely(align
)) {
3688 error
= xfs_bmap_extsize_align(mp
, &ap
->got
, &ap
->prev
,
3689 align
, 0, ap
->eof
, 0, ap
->conv
,
3690 &ap
->offset
, &ap
->length
);
3696 nullfb
= *ap
->firstblock
== NULLFSBLOCK
;
3697 fb_agno
= nullfb
? NULLAGNUMBER
: XFS_FSB_TO_AGNO(mp
, *ap
->firstblock
);
3699 if (ap
->userdata
&& xfs_inode_is_filestream(ap
->ip
)) {
3700 ag
= xfs_filestream_lookup_ag(ap
->ip
);
3701 ag
= (ag
!= NULLAGNUMBER
) ? ag
: 0;
3702 ap
->blkno
= XFS_AGB_TO_FSB(mp
, ag
, 0);
3704 ap
->blkno
= XFS_INO_TO_FSB(mp
, ap
->ip
->i_ino
);
3707 ap
->blkno
= *ap
->firstblock
;
3709 xfs_bmap_adjacent(ap
);
3712 * If allowed, use ap->blkno; otherwise must use firstblock since
3713 * it's in the right allocation group.
3715 if (nullfb
|| XFS_FSB_TO_AGNO(mp
, ap
->blkno
) == fb_agno
)
3718 ap
->blkno
= *ap
->firstblock
;
3720 * Normal allocation, done through xfs_alloc_vextent.
3722 tryagain
= isaligned
= 0;
3723 memset(&args
, 0, sizeof(args
));
3726 args
.fsbno
= ap
->blkno
;
3728 /* Trim the allocation back to the maximum an AG can fit. */
3729 args
.maxlen
= MIN(ap
->length
, XFS_ALLOC_AG_MAX_USABLE(mp
));
3730 args
.firstblock
= *ap
->firstblock
;
3734 * Search for an allocation group with a single extent large
3735 * enough for the request. If one isn't found, then adjust
3736 * the minimum allocation size to the largest space found.
3738 if (ap
->userdata
&& xfs_inode_is_filestream(ap
->ip
))
3739 error
= xfs_bmap_btalloc_filestreams(ap
, &args
, &blen
);
3741 error
= xfs_bmap_btalloc_nullfb(ap
, &args
, &blen
);
3744 } else if (ap
->flist
->xbf_low
) {
3745 if (xfs_inode_is_filestream(ap
->ip
))
3746 args
.type
= XFS_ALLOCTYPE_FIRST_AG
;
3748 args
.type
= XFS_ALLOCTYPE_START_BNO
;
3749 args
.total
= args
.minlen
= ap
->minlen
;
3751 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
3752 args
.total
= ap
->total
;
3753 args
.minlen
= ap
->minlen
;
3755 /* apply extent size hints if obtained earlier */
3756 if (unlikely(align
)) {
3758 if ((args
.mod
= (xfs_extlen_t
)do_mod(ap
->offset
, args
.prod
)))
3759 args
.mod
= (xfs_extlen_t
)(args
.prod
- args
.mod
);
3760 } else if (mp
->m_sb
.sb_blocksize
>= PAGE_CACHE_SIZE
) {
3764 args
.prod
= PAGE_CACHE_SIZE
>> mp
->m_sb
.sb_blocklog
;
3765 if ((args
.mod
= (xfs_extlen_t
)(do_mod(ap
->offset
, args
.prod
))))
3766 args
.mod
= (xfs_extlen_t
)(args
.prod
- args
.mod
);
3769 * If we are not low on available data blocks, and the
3770 * underlying logical volume manager is a stripe, and
3771 * the file offset is zero then try to allocate data
3772 * blocks on stripe unit boundary.
3773 * NOTE: ap->aeof is only set if the allocation length
3774 * is >= the stripe unit and the allocation offset is
3775 * at the end of file.
3777 if (!ap
->flist
->xbf_low
&& ap
->aeof
) {
3779 args
.alignment
= stripe_align
;
3783 * Adjust for alignment
3785 if (blen
> args
.alignment
&& blen
<= args
.maxlen
)
3786 args
.minlen
= blen
- args
.alignment
;
3787 args
.minalignslop
= 0;
3790 * First try an exact bno allocation.
3791 * If it fails then do a near or start bno
3792 * allocation with alignment turned on.
3796 args
.type
= XFS_ALLOCTYPE_THIS_BNO
;
3799 * Compute the minlen+alignment for the
3800 * next case. Set slop so that the value
3801 * of minlen+alignment+slop doesn't go up
3802 * between the calls.
3804 if (blen
> stripe_align
&& blen
<= args
.maxlen
)
3805 nextminlen
= blen
- stripe_align
;
3807 nextminlen
= args
.minlen
;
3808 if (nextminlen
+ stripe_align
> args
.minlen
+ 1)
3810 nextminlen
+ stripe_align
-
3813 args
.minalignslop
= 0;
3817 args
.minalignslop
= 0;
3819 args
.minleft
= ap
->minleft
;
3820 args
.wasdel
= ap
->wasdel
;
3822 args
.userdata
= ap
->userdata
;
3823 if ((error
= xfs_alloc_vextent(&args
)))
3825 if (tryagain
&& args
.fsbno
== NULLFSBLOCK
) {
3827 * Exact allocation failed. Now try with alignment
3831 args
.fsbno
= ap
->blkno
;
3832 args
.alignment
= stripe_align
;
3833 args
.minlen
= nextminlen
;
3834 args
.minalignslop
= 0;
3836 if ((error
= xfs_alloc_vextent(&args
)))
3839 if (isaligned
&& args
.fsbno
== NULLFSBLOCK
) {
3841 * allocation failed, so turn off alignment and
3845 args
.fsbno
= ap
->blkno
;
3847 if ((error
= xfs_alloc_vextent(&args
)))
3850 if (args
.fsbno
== NULLFSBLOCK
&& nullfb
&&
3851 args
.minlen
> ap
->minlen
) {
3852 args
.minlen
= ap
->minlen
;
3853 args
.type
= XFS_ALLOCTYPE_START_BNO
;
3854 args
.fsbno
= ap
->blkno
;
3855 if ((error
= xfs_alloc_vextent(&args
)))
3858 if (args
.fsbno
== NULLFSBLOCK
&& nullfb
) {
3860 args
.type
= XFS_ALLOCTYPE_FIRST_AG
;
3861 args
.total
= ap
->minlen
;
3863 if ((error
= xfs_alloc_vextent(&args
)))
3865 ap
->flist
->xbf_low
= 1;
3867 if (args
.fsbno
!= NULLFSBLOCK
) {
3869 * check the allocation happened at the same or higher AG than
3870 * the first block that was allocated.
3872 ASSERT(*ap
->firstblock
== NULLFSBLOCK
||
3873 XFS_FSB_TO_AGNO(mp
, *ap
->firstblock
) ==
3874 XFS_FSB_TO_AGNO(mp
, args
.fsbno
) ||
3875 (ap
->flist
->xbf_low
&&
3876 XFS_FSB_TO_AGNO(mp
, *ap
->firstblock
) <
3877 XFS_FSB_TO_AGNO(mp
, args
.fsbno
)));
3879 ap
->blkno
= args
.fsbno
;
3880 if (*ap
->firstblock
== NULLFSBLOCK
)
3881 *ap
->firstblock
= args
.fsbno
;
3882 ASSERT(nullfb
|| fb_agno
== args
.agno
||
3883 (ap
->flist
->xbf_low
&& fb_agno
< args
.agno
));
3884 ap
->length
= args
.len
;
3885 ap
->ip
->i_d
.di_nblocks
+= args
.len
;
3886 xfs_trans_log_inode(ap
->tp
, ap
->ip
, XFS_ILOG_CORE
);
3888 ap
->ip
->i_delayed_blks
-= args
.len
;
3890 * Adjust the disk quota also. This was reserved
3893 xfs_trans_mod_dquot_byino(ap
->tp
, ap
->ip
,
3894 ap
->wasdel
? XFS_TRANS_DQ_DELBCOUNT
:
3895 XFS_TRANS_DQ_BCOUNT
,
3898 ap
->blkno
= NULLFSBLOCK
;
3905 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
3906 * It figures out where to ask the underlying allocator to put the new extent.
3910 struct xfs_bmalloca
*ap
) /* bmap alloc argument struct */
3912 if (XFS_IS_REALTIME_INODE(ap
->ip
) && ap
->userdata
)
3913 return xfs_bmap_rtalloc(ap
);
3914 return xfs_bmap_btalloc(ap
);
3918 * Trim the returned map to the required bounds
3922 struct xfs_bmbt_irec
*mval
,
3923 struct xfs_bmbt_irec
*got
,
3931 if ((flags
& XFS_BMAPI_ENTIRE
) ||
3932 got
->br_startoff
+ got
->br_blockcount
<= obno
) {
3934 if (isnullstartblock(got
->br_startblock
))
3935 mval
->br_startblock
= DELAYSTARTBLOCK
;
3941 ASSERT((*bno
>= obno
) || (n
== 0));
3943 mval
->br_startoff
= *bno
;
3944 if (isnullstartblock(got
->br_startblock
))
3945 mval
->br_startblock
= DELAYSTARTBLOCK
;
3947 mval
->br_startblock
= got
->br_startblock
+
3948 (*bno
- got
->br_startoff
);
3950 * Return the minimum of what we got and what we asked for for
3951 * the length. We can use the len variable here because it is
3952 * modified below and we could have been there before coming
3953 * here if the first part of the allocation didn't overlap what
3956 mval
->br_blockcount
= XFS_FILBLKS_MIN(end
- *bno
,
3957 got
->br_blockcount
- (*bno
- got
->br_startoff
));
3958 mval
->br_state
= got
->br_state
;
3959 ASSERT(mval
->br_blockcount
<= len
);
3964 * Update and validate the extent map to return
3967 xfs_bmapi_update_map(
3968 struct xfs_bmbt_irec
**map
,
3976 xfs_bmbt_irec_t
*mval
= *map
;
3978 ASSERT((flags
& XFS_BMAPI_ENTIRE
) ||
3979 ((mval
->br_startoff
+ mval
->br_blockcount
) <= end
));
3980 ASSERT((flags
& XFS_BMAPI_ENTIRE
) || (mval
->br_blockcount
<= *len
) ||
3981 (mval
->br_startoff
< obno
));
3983 *bno
= mval
->br_startoff
+ mval
->br_blockcount
;
3985 if (*n
> 0 && mval
->br_startoff
== mval
[-1].br_startoff
) {
3986 /* update previous map with new information */
3987 ASSERT(mval
->br_startblock
== mval
[-1].br_startblock
);
3988 ASSERT(mval
->br_blockcount
> mval
[-1].br_blockcount
);
3989 ASSERT(mval
->br_state
== mval
[-1].br_state
);
3990 mval
[-1].br_blockcount
= mval
->br_blockcount
;
3991 mval
[-1].br_state
= mval
->br_state
;
3992 } else if (*n
> 0 && mval
->br_startblock
!= DELAYSTARTBLOCK
&&
3993 mval
[-1].br_startblock
!= DELAYSTARTBLOCK
&&
3994 mval
[-1].br_startblock
!= HOLESTARTBLOCK
&&
3995 mval
->br_startblock
== mval
[-1].br_startblock
+
3996 mval
[-1].br_blockcount
&&
3997 ((flags
& XFS_BMAPI_IGSTATE
) ||
3998 mval
[-1].br_state
== mval
->br_state
)) {
3999 ASSERT(mval
->br_startoff
==
4000 mval
[-1].br_startoff
+ mval
[-1].br_blockcount
);
4001 mval
[-1].br_blockcount
+= mval
->br_blockcount
;
4002 } else if (*n
> 0 &&
4003 mval
->br_startblock
== DELAYSTARTBLOCK
&&
4004 mval
[-1].br_startblock
== DELAYSTARTBLOCK
&&
4005 mval
->br_startoff
==
4006 mval
[-1].br_startoff
+ mval
[-1].br_blockcount
) {
4007 mval
[-1].br_blockcount
+= mval
->br_blockcount
;
4008 mval
[-1].br_state
= mval
->br_state
;
4009 } else if (!((*n
== 0) &&
4010 ((mval
->br_startoff
+ mval
->br_blockcount
) <=
4019 * Map file blocks to filesystem blocks without allocation.
4023 struct xfs_inode
*ip
,
4026 struct xfs_bmbt_irec
*mval
,
4030 struct xfs_mount
*mp
= ip
->i_mount
;
4031 struct xfs_ifork
*ifp
;
4032 struct xfs_bmbt_irec got
;
4033 struct xfs_bmbt_irec prev
;
4040 int whichfork
= (flags
& XFS_BMAPI_ATTRFORK
) ?
4041 XFS_ATTR_FORK
: XFS_DATA_FORK
;
4044 ASSERT(!(flags
& ~(XFS_BMAPI_ATTRFORK
|XFS_BMAPI_ENTIRE
|
4045 XFS_BMAPI_IGSTATE
)));
4046 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
));
4048 if (unlikely(XFS_TEST_ERROR(
4049 (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
4050 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
),
4051 mp
, XFS_ERRTAG_BMAPIFORMAT
, XFS_RANDOM_BMAPIFORMAT
))) {
4052 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW
, mp
);
4053 return XFS_ERROR(EFSCORRUPTED
);
4056 if (XFS_FORCED_SHUTDOWN(mp
))
4057 return XFS_ERROR(EIO
);
4059 XFS_STATS_INC(xs_blk_mapr
);
4061 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4063 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
4064 error
= xfs_iread_extents(NULL
, ip
, whichfork
);
4069 xfs_bmap_search_extents(ip
, bno
, whichfork
, &eof
, &lastx
, &got
, &prev
);
4073 while (bno
< end
&& n
< *nmap
) {
4074 /* Reading past eof, act as though there's a hole up to end. */
4076 got
.br_startoff
= end
;
4077 if (got
.br_startoff
> bno
) {
4078 /* Reading in a hole. */
4079 mval
->br_startoff
= bno
;
4080 mval
->br_startblock
= HOLESTARTBLOCK
;
4081 mval
->br_blockcount
=
4082 XFS_FILBLKS_MIN(len
, got
.br_startoff
- bno
);
4083 mval
->br_state
= XFS_EXT_NORM
;
4084 bno
+= mval
->br_blockcount
;
4085 len
-= mval
->br_blockcount
;
4091 /* set up the extent map to return. */
4092 xfs_bmapi_trim_map(mval
, &got
, &bno
, len
, obno
, end
, n
, flags
);
4093 xfs_bmapi_update_map(&mval
, &bno
, &len
, obno
, end
, &n
, flags
);
4095 /* If we're done, stop now. */
4096 if (bno
>= end
|| n
>= *nmap
)
4099 /* Else go on to the next record. */
4100 if (++lastx
< ifp
->if_bytes
/ sizeof(xfs_bmbt_rec_t
))
4101 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, lastx
), &got
);
4110 xfs_bmapi_reserve_delalloc(
4111 struct xfs_inode
*ip
,
4114 struct xfs_bmbt_irec
*got
,
4115 struct xfs_bmbt_irec
*prev
,
4116 xfs_extnum_t
*lastx
,
4119 struct xfs_mount
*mp
= ip
->i_mount
;
4120 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, XFS_DATA_FORK
);
4122 xfs_extlen_t indlen
;
4123 char rt
= XFS_IS_REALTIME_INODE(ip
);
4127 alen
= XFS_FILBLKS_MIN(len
, MAXEXTLEN
);
4129 alen
= XFS_FILBLKS_MIN(alen
, got
->br_startoff
- aoff
);
4131 /* Figure out the extent size, adjust alen */
4132 extsz
= xfs_get_extsz_hint(ip
);
4135 * Make sure we don't exceed a single extent length when we
4136 * align the extent by reducing length we are going to
4137 * allocate by the maximum amount extent size aligment may
4140 alen
= XFS_FILBLKS_MIN(len
, MAXEXTLEN
- (2 * extsz
- 1));
4141 error
= xfs_bmap_extsize_align(mp
, got
, prev
, extsz
, rt
, eof
,
4142 1, 0, &aoff
, &alen
);
4147 extsz
= alen
/ mp
->m_sb
.sb_rextsize
;
4150 * Make a transaction-less quota reservation for delayed allocation
4151 * blocks. This number gets adjusted later. We return if we haven't
4152 * allocated blocks already inside this loop.
4154 error
= xfs_trans_reserve_quota_nblks(NULL
, ip
, (long)alen
, 0,
4155 rt
? XFS_QMOPT_RES_RTBLKS
: XFS_QMOPT_RES_REGBLKS
);
4160 * Split changing sb for alen and indlen since they could be coming
4161 * from different places.
4163 indlen
= (xfs_extlen_t
)xfs_bmap_worst_indlen(ip
, alen
);
4167 error
= xfs_mod_incore_sb(mp
, XFS_SBS_FREXTENTS
,
4168 -((int64_t)extsz
), 0);
4170 error
= xfs_icsb_modify_counters(mp
, XFS_SBS_FDBLOCKS
,
4171 -((int64_t)alen
), 0);
4175 goto out_unreserve_quota
;
4177 error
= xfs_icsb_modify_counters(mp
, XFS_SBS_FDBLOCKS
,
4178 -((int64_t)indlen
), 0);
4180 goto out_unreserve_blocks
;
4183 ip
->i_delayed_blks
+= alen
;
4185 got
->br_startoff
= aoff
;
4186 got
->br_startblock
= nullstartblock(indlen
);
4187 got
->br_blockcount
= alen
;
4188 got
->br_state
= XFS_EXT_NORM
;
4189 xfs_bmap_add_extent_hole_delay(ip
, lastx
, got
);
4192 * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4193 * might have merged it into one of the neighbouring ones.
4195 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *lastx
), got
);
4197 ASSERT(got
->br_startoff
<= aoff
);
4198 ASSERT(got
->br_startoff
+ got
->br_blockcount
>= aoff
+ alen
);
4199 ASSERT(isnullstartblock(got
->br_startblock
));
4200 ASSERT(got
->br_state
== XFS_EXT_NORM
);
4203 out_unreserve_blocks
:
4205 xfs_mod_incore_sb(mp
, XFS_SBS_FREXTENTS
, extsz
, 0);
4207 xfs_icsb_modify_counters(mp
, XFS_SBS_FDBLOCKS
, alen
, 0);
4208 out_unreserve_quota
:
4209 if (XFS_IS_QUOTA_ON(mp
))
4210 xfs_trans_unreserve_quota_nblks(NULL
, ip
, (long)alen
, 0, rt
?
4211 XFS_QMOPT_RES_RTBLKS
: XFS_QMOPT_RES_REGBLKS
);
4216 * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4220 struct xfs_inode
*ip
, /* incore inode */
4221 xfs_fileoff_t bno
, /* starting file offs. mapped */
4222 xfs_filblks_t len
, /* length to map in file */
4223 struct xfs_bmbt_irec
*mval
, /* output: map values */
4224 int *nmap
, /* i/o: mval size/count */
4225 int flags
) /* XFS_BMAPI_... */
4227 struct xfs_mount
*mp
= ip
->i_mount
;
4228 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, XFS_DATA_FORK
);
4229 struct xfs_bmbt_irec got
; /* current file extent record */
4230 struct xfs_bmbt_irec prev
; /* previous file extent record */
4231 xfs_fileoff_t obno
; /* old block number (offset) */
4232 xfs_fileoff_t end
; /* end of mapped file region */
4233 xfs_extnum_t lastx
; /* last useful extent number */
4234 int eof
; /* we've hit the end of extents */
4235 int n
= 0; /* current extent index */
4239 ASSERT(*nmap
<= XFS_BMAP_MAX_NMAP
);
4240 ASSERT(!(flags
& ~XFS_BMAPI_ENTIRE
));
4241 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
4243 if (unlikely(XFS_TEST_ERROR(
4244 (XFS_IFORK_FORMAT(ip
, XFS_DATA_FORK
) != XFS_DINODE_FMT_EXTENTS
&&
4245 XFS_IFORK_FORMAT(ip
, XFS_DATA_FORK
) != XFS_DINODE_FMT_BTREE
),
4246 mp
, XFS_ERRTAG_BMAPIFORMAT
, XFS_RANDOM_BMAPIFORMAT
))) {
4247 XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW
, mp
);
4248 return XFS_ERROR(EFSCORRUPTED
);
4251 if (XFS_FORCED_SHUTDOWN(mp
))
4252 return XFS_ERROR(EIO
);
4254 XFS_STATS_INC(xs_blk_mapw
);
4256 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
4257 error
= xfs_iread_extents(NULL
, ip
, XFS_DATA_FORK
);
4262 xfs_bmap_search_extents(ip
, bno
, XFS_DATA_FORK
, &eof
, &lastx
, &got
, &prev
);
4266 while (bno
< end
&& n
< *nmap
) {
4267 if (eof
|| got
.br_startoff
> bno
) {
4268 error
= xfs_bmapi_reserve_delalloc(ip
, bno
, len
, &got
,
4269 &prev
, &lastx
, eof
);
4279 /* set up the extent map to return. */
4280 xfs_bmapi_trim_map(mval
, &got
, &bno
, len
, obno
, end
, n
, flags
);
4281 xfs_bmapi_update_map(&mval
, &bno
, &len
, obno
, end
, &n
, flags
);
4283 /* If we're done, stop now. */
4284 if (bno
>= end
|| n
>= *nmap
)
4287 /* Else go on to the next record. */
4289 if (++lastx
< ifp
->if_bytes
/ sizeof(xfs_bmbt_rec_t
))
4290 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, lastx
), &got
);
4301 __xfs_bmapi_allocate(
4302 struct xfs_bmalloca
*bma
)
4304 struct xfs_mount
*mp
= bma
->ip
->i_mount
;
4305 int whichfork
= (bma
->flags
& XFS_BMAPI_ATTRFORK
) ?
4306 XFS_ATTR_FORK
: XFS_DATA_FORK
;
4307 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(bma
->ip
, whichfork
);
4308 int tmp_logflags
= 0;
4311 ASSERT(bma
->length
> 0);
4314 * For the wasdelay case, we could also just allocate the stuff asked
4315 * for in this bmap call but that wouldn't be as good.
4318 bma
->length
= (xfs_extlen_t
)bma
->got
.br_blockcount
;
4319 bma
->offset
= bma
->got
.br_startoff
;
4320 if (bma
->idx
!= NULLEXTNUM
&& bma
->idx
) {
4321 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, bma
->idx
- 1),
4325 bma
->length
= XFS_FILBLKS_MIN(bma
->length
, MAXEXTLEN
);
4327 bma
->length
= XFS_FILBLKS_MIN(bma
->length
,
4328 bma
->got
.br_startoff
- bma
->offset
);
4332 * Indicate if this is the first user data in the file, or just any
4335 if (!(bma
->flags
& XFS_BMAPI_METADATA
)) {
4336 bma
->userdata
= (bma
->offset
== 0) ?
4337 XFS_ALLOC_INITIAL_USER_DATA
: XFS_ALLOC_USERDATA
;
4340 bma
->minlen
= (bma
->flags
& XFS_BMAPI_CONTIG
) ? bma
->length
: 1;
4343 * Only want to do the alignment at the eof if it is userdata and
4344 * allocation length is larger than a stripe unit.
4346 if (mp
->m_dalign
&& bma
->length
>= mp
->m_dalign
&&
4347 !(bma
->flags
& XFS_BMAPI_METADATA
) && whichfork
== XFS_DATA_FORK
) {
4348 error
= xfs_bmap_isaeof(bma
, whichfork
);
4353 error
= xfs_bmap_alloc(bma
);
4357 if (bma
->flist
->xbf_low
)
4360 bma
->cur
->bc_private
.b
.firstblock
= *bma
->firstblock
;
4361 if (bma
->blkno
== NULLFSBLOCK
)
4363 if ((ifp
->if_flags
& XFS_IFBROOT
) && !bma
->cur
) {
4364 bma
->cur
= xfs_bmbt_init_cursor(mp
, bma
->tp
, bma
->ip
, whichfork
);
4365 bma
->cur
->bc_private
.b
.firstblock
= *bma
->firstblock
;
4366 bma
->cur
->bc_private
.b
.flist
= bma
->flist
;
4369 * Bump the number of extents we've allocated
4375 bma
->cur
->bc_private
.b
.flags
=
4376 bma
->wasdel
? XFS_BTCUR_BPRV_WASDEL
: 0;
4378 bma
->got
.br_startoff
= bma
->offset
;
4379 bma
->got
.br_startblock
= bma
->blkno
;
4380 bma
->got
.br_blockcount
= bma
->length
;
4381 bma
->got
.br_state
= XFS_EXT_NORM
;
4384 * A wasdelay extent has been initialized, so shouldn't be flagged
4387 if (!bma
->wasdel
&& (bma
->flags
& XFS_BMAPI_PREALLOC
) &&
4388 xfs_sb_version_hasextflgbit(&mp
->m_sb
))
4389 bma
->got
.br_state
= XFS_EXT_UNWRITTEN
;
4392 error
= xfs_bmap_add_extent_delay_real(bma
);
4394 error
= xfs_bmap_add_extent_hole_real(bma
, whichfork
);
4396 bma
->logflags
|= tmp_logflags
;
4401 * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4402 * or xfs_bmap_add_extent_hole_real might have merged it into one of
4403 * the neighbouring ones.
4405 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, bma
->idx
), &bma
->got
);
4407 ASSERT(bma
->got
.br_startoff
<= bma
->offset
);
4408 ASSERT(bma
->got
.br_startoff
+ bma
->got
.br_blockcount
>=
4409 bma
->offset
+ bma
->length
);
4410 ASSERT(bma
->got
.br_state
== XFS_EXT_NORM
||
4411 bma
->got
.br_state
== XFS_EXT_UNWRITTEN
);
4416 xfs_bmapi_convert_unwritten(
4417 struct xfs_bmalloca
*bma
,
4418 struct xfs_bmbt_irec
*mval
,
4422 int whichfork
= (flags
& XFS_BMAPI_ATTRFORK
) ?
4423 XFS_ATTR_FORK
: XFS_DATA_FORK
;
4424 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(bma
->ip
, whichfork
);
4425 int tmp_logflags
= 0;
4428 /* check if we need to do unwritten->real conversion */
4429 if (mval
->br_state
== XFS_EXT_UNWRITTEN
&&
4430 (flags
& XFS_BMAPI_PREALLOC
))
4433 /* check if we need to do real->unwritten conversion */
4434 if (mval
->br_state
== XFS_EXT_NORM
&&
4435 (flags
& (XFS_BMAPI_PREALLOC
| XFS_BMAPI_CONVERT
)) !=
4436 (XFS_BMAPI_PREALLOC
| XFS_BMAPI_CONVERT
))
4440 * Modify (by adding) the state flag, if writing.
4442 ASSERT(mval
->br_blockcount
<= len
);
4443 if ((ifp
->if_flags
& XFS_IFBROOT
) && !bma
->cur
) {
4444 bma
->cur
= xfs_bmbt_init_cursor(bma
->ip
->i_mount
, bma
->tp
,
4445 bma
->ip
, whichfork
);
4446 bma
->cur
->bc_private
.b
.firstblock
= *bma
->firstblock
;
4447 bma
->cur
->bc_private
.b
.flist
= bma
->flist
;
4449 mval
->br_state
= (mval
->br_state
== XFS_EXT_UNWRITTEN
)
4450 ? XFS_EXT_NORM
: XFS_EXT_UNWRITTEN
;
4452 error
= xfs_bmap_add_extent_unwritten_real(bma
->tp
, bma
->ip
, &bma
->idx
,
4453 &bma
->cur
, mval
, bma
->firstblock
, bma
->flist
,
4455 bma
->logflags
|= tmp_logflags
;
4460 * Update our extent pointer, given that
4461 * xfs_bmap_add_extent_unwritten_real might have merged it into one
4462 * of the neighbouring ones.
4464 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, bma
->idx
), &bma
->got
);
4467 * We may have combined previously unwritten space with written space,
4468 * so generate another request.
4470 if (mval
->br_blockcount
< len
)
4476 * Map file blocks to filesystem blocks, and allocate blocks or convert the
4477 * extent state if necessary. Details behaviour is controlled by the flags
4478 * parameter. Only allocates blocks from a single allocation group, to avoid
4481 * The returned value in "firstblock" from the first call in a transaction
4482 * must be remembered and presented to subsequent calls in "firstblock".
4483 * An upper bound for the number of blocks to be allocated is supplied to
4484 * the first call in "total"; if no allocation group has that many free
4485 * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4489 struct xfs_trans
*tp
, /* transaction pointer */
4490 struct xfs_inode
*ip
, /* incore inode */
4491 xfs_fileoff_t bno
, /* starting file offs. mapped */
4492 xfs_filblks_t len
, /* length to map in file */
4493 int flags
, /* XFS_BMAPI_... */
4494 xfs_fsblock_t
*firstblock
, /* first allocated block
4495 controls a.g. for allocs */
4496 xfs_extlen_t total
, /* total blocks needed */
4497 struct xfs_bmbt_irec
*mval
, /* output: map values */
4498 int *nmap
, /* i/o: mval size/count */
4499 struct xfs_bmap_free
*flist
) /* i/o: list extents to free */
4501 struct xfs_mount
*mp
= ip
->i_mount
;
4502 struct xfs_ifork
*ifp
;
4503 struct xfs_bmalloca bma
= { NULL
}; /* args for xfs_bmap_alloc */
4504 xfs_fileoff_t end
; /* end of mapped file region */
4505 int eof
; /* after the end of extents */
4506 int error
; /* error return */
4507 int n
; /* current extent index */
4508 xfs_fileoff_t obno
; /* old block number (offset) */
4509 int whichfork
; /* data or attr fork */
4510 char inhole
; /* current location is hole in file */
4511 char wasdelay
; /* old extent was delayed */
4514 xfs_fileoff_t orig_bno
; /* original block number value */
4515 int orig_flags
; /* original flags arg value */
4516 xfs_filblks_t orig_len
; /* original value of len arg */
4517 struct xfs_bmbt_irec
*orig_mval
; /* original value of mval */
4518 int orig_nmap
; /* original value of *nmap */
4526 whichfork
= (flags
& XFS_BMAPI_ATTRFORK
) ?
4527 XFS_ATTR_FORK
: XFS_DATA_FORK
;
4530 ASSERT(*nmap
<= XFS_BMAP_MAX_NMAP
);
4531 ASSERT(!(flags
& XFS_BMAPI_IGSTATE
));
4534 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_LOCAL
);
4535 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
4537 if (unlikely(XFS_TEST_ERROR(
4538 (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
4539 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
),
4540 mp
, XFS_ERRTAG_BMAPIFORMAT
, XFS_RANDOM_BMAPIFORMAT
))) {
4541 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW
, mp
);
4542 return XFS_ERROR(EFSCORRUPTED
);
4545 if (XFS_FORCED_SHUTDOWN(mp
))
4546 return XFS_ERROR(EIO
);
4548 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4550 XFS_STATS_INC(xs_blk_mapw
);
4552 if (*firstblock
== NULLFSBLOCK
) {
4553 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_BTREE
)
4554 bma
.minleft
= be16_to_cpu(ifp
->if_broot
->bb_level
) + 1;
4561 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
4562 error
= xfs_iread_extents(tp
, ip
, whichfork
);
4567 xfs_bmap_search_extents(ip
, bno
, whichfork
, &eof
, &bma
.idx
, &bma
.got
,
4578 bma
.firstblock
= firstblock
;
4580 if (flags
& XFS_BMAPI_STACK_SWITCH
)
4581 bma
.stack_switch
= 1;
4583 while (bno
< end
&& n
< *nmap
) {
4584 inhole
= eof
|| bma
.got
.br_startoff
> bno
;
4585 wasdelay
= !inhole
&& isnullstartblock(bma
.got
.br_startblock
);
4588 * First, deal with the hole before the allocated space
4589 * that we found, if any.
4591 if (inhole
|| wasdelay
) {
4593 bma
.conv
= !!(flags
& XFS_BMAPI_CONVERT
);
4594 bma
.wasdel
= wasdelay
;
4599 * There's a 32/64 bit type mismatch between the
4600 * allocation length request (which can be 64 bits in
4601 * length) and the bma length request, which is
4602 * xfs_extlen_t and therefore 32 bits. Hence we have to
4603 * check for 32-bit overflows and handle them here.
4605 if (len
> (xfs_filblks_t
)MAXEXTLEN
)
4606 bma
.length
= MAXEXTLEN
;
4611 ASSERT(bma
.length
> 0);
4612 error
= xfs_bmapi_allocate(&bma
);
4615 if (bma
.blkno
== NULLFSBLOCK
)
4619 /* Deal with the allocated space we found. */
4620 xfs_bmapi_trim_map(mval
, &bma
.got
, &bno
, len
, obno
,
4623 /* Execute unwritten extent conversion if necessary */
4624 error
= xfs_bmapi_convert_unwritten(&bma
, mval
, len
, flags
);
4625 if (error
== EAGAIN
)
4630 /* update the extent map to return */
4631 xfs_bmapi_update_map(&mval
, &bno
, &len
, obno
, end
, &n
, flags
);
4634 * If we're done, stop now. Stop when we've allocated
4635 * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise
4636 * the transaction may get too big.
4638 if (bno
>= end
|| n
>= *nmap
|| bma
.nallocs
>= *nmap
)
4641 /* Else go on to the next record. */
4643 if (++bma
.idx
< ifp
->if_bytes
/ sizeof(xfs_bmbt_rec_t
)) {
4644 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, bma
.idx
),
4652 * Transform from btree to extents, give it cur.
4654 if (xfs_bmap_wants_extents(ip
, whichfork
)) {
4655 int tmp_logflags
= 0;
4658 error
= xfs_bmap_btree_to_extents(tp
, ip
, bma
.cur
,
4659 &tmp_logflags
, whichfork
);
4660 bma
.logflags
|= tmp_logflags
;
4665 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
||
4666 XFS_IFORK_NEXTENTS(ip
, whichfork
) >
4667 XFS_IFORK_MAXEXT(ip
, whichfork
));
4671 * Log everything. Do this after conversion, there's no point in
4672 * logging the extent records if we've converted to btree format.
4674 if ((bma
.logflags
& xfs_ilog_fext(whichfork
)) &&
4675 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
)
4676 bma
.logflags
&= ~xfs_ilog_fext(whichfork
);
4677 else if ((bma
.logflags
& xfs_ilog_fbroot(whichfork
)) &&
4678 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
)
4679 bma
.logflags
&= ~xfs_ilog_fbroot(whichfork
);
4681 * Log whatever the flags say, even if error. Otherwise we might miss
4682 * detecting a case where the data is changed, there's an error,
4683 * and it's not logged so we don't shutdown when we should.
4686 xfs_trans_log_inode(tp
, ip
, bma
.logflags
);
4690 ASSERT(*firstblock
== NULLFSBLOCK
||
4691 XFS_FSB_TO_AGNO(mp
, *firstblock
) ==
4693 bma
.cur
->bc_private
.b
.firstblock
) ||
4695 XFS_FSB_TO_AGNO(mp
, *firstblock
) <
4697 bma
.cur
->bc_private
.b
.firstblock
)));
4698 *firstblock
= bma
.cur
->bc_private
.b
.firstblock
;
4700 xfs_btree_del_cursor(bma
.cur
,
4701 error
? XFS_BTREE_ERROR
: XFS_BTREE_NOERROR
);
4704 xfs_bmap_validate_ret(orig_bno
, orig_len
, orig_flags
, orig_mval
,
4710 * Called by xfs_bmapi to update file extent records and the btree
4711 * after removing space (or undoing a delayed allocation).
4713 STATIC
int /* error */
4714 xfs_bmap_del_extent(
4715 xfs_inode_t
*ip
, /* incore inode pointer */
4716 xfs_trans_t
*tp
, /* current transaction pointer */
4717 xfs_extnum_t
*idx
, /* extent number to update/delete */
4718 xfs_bmap_free_t
*flist
, /* list of extents to be freed */
4719 xfs_btree_cur_t
*cur
, /* if null, not a btree */
4720 xfs_bmbt_irec_t
*del
, /* data to remove from extents */
4721 int *logflagsp
, /* inode logging flags */
4722 int whichfork
) /* data or attr fork */
4724 xfs_filblks_t da_new
; /* new delay-alloc indirect blocks */
4725 xfs_filblks_t da_old
; /* old delay-alloc indirect blocks */
4726 xfs_fsblock_t del_endblock
=0; /* first block past del */
4727 xfs_fileoff_t del_endoff
; /* first offset past del */
4728 int delay
; /* current block is delayed allocated */
4729 int do_fx
; /* free extent at end of routine */
4730 xfs_bmbt_rec_host_t
*ep
; /* current extent entry pointer */
4731 int error
; /* error return value */
4732 int flags
; /* inode logging flags */
4733 xfs_bmbt_irec_t got
; /* current extent entry */
4734 xfs_fileoff_t got_endoff
; /* first offset past got */
4735 int i
; /* temp state */
4736 xfs_ifork_t
*ifp
; /* inode fork pointer */
4737 xfs_mount_t
*mp
; /* mount structure */
4738 xfs_filblks_t nblks
; /* quota/sb block count */
4739 xfs_bmbt_irec_t
new; /* new record to be inserted */
4741 uint qfield
; /* quota field to update */
4742 xfs_filblks_t temp
; /* for indirect length calculations */
4743 xfs_filblks_t temp2
; /* for indirect length calculations */
4746 XFS_STATS_INC(xs_del_exlist
);
4748 if (whichfork
== XFS_ATTR_FORK
)
4749 state
|= BMAP_ATTRFORK
;
4752 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4753 ASSERT((*idx
>= 0) && (*idx
< ifp
->if_bytes
/
4754 (uint
)sizeof(xfs_bmbt_rec_t
)));
4755 ASSERT(del
->br_blockcount
> 0);
4756 ep
= xfs_iext_get_ext(ifp
, *idx
);
4757 xfs_bmbt_get_all(ep
, &got
);
4758 ASSERT(got
.br_startoff
<= del
->br_startoff
);
4759 del_endoff
= del
->br_startoff
+ del
->br_blockcount
;
4760 got_endoff
= got
.br_startoff
+ got
.br_blockcount
;
4761 ASSERT(got_endoff
>= del_endoff
);
4762 delay
= isnullstartblock(got
.br_startblock
);
4763 ASSERT(isnullstartblock(del
->br_startblock
) == delay
);
4768 * If deleting a real allocation, must free up the disk space.
4771 flags
= XFS_ILOG_CORE
;
4773 * Realtime allocation. Free it and record di_nblocks update.
4775 if (whichfork
== XFS_DATA_FORK
&& XFS_IS_REALTIME_INODE(ip
)) {
4779 ASSERT(do_mod(del
->br_blockcount
,
4780 mp
->m_sb
.sb_rextsize
) == 0);
4781 ASSERT(do_mod(del
->br_startblock
,
4782 mp
->m_sb
.sb_rextsize
) == 0);
4783 bno
= del
->br_startblock
;
4784 len
= del
->br_blockcount
;
4785 do_div(bno
, mp
->m_sb
.sb_rextsize
);
4786 do_div(len
, mp
->m_sb
.sb_rextsize
);
4787 error
= xfs_rtfree_extent(tp
, bno
, (xfs_extlen_t
)len
);
4791 nblks
= len
* mp
->m_sb
.sb_rextsize
;
4792 qfield
= XFS_TRANS_DQ_RTBCOUNT
;
4795 * Ordinary allocation.
4799 nblks
= del
->br_blockcount
;
4800 qfield
= XFS_TRANS_DQ_BCOUNT
;
4803 * Set up del_endblock and cur for later.
4805 del_endblock
= del
->br_startblock
+ del
->br_blockcount
;
4807 if ((error
= xfs_bmbt_lookup_eq(cur
, got
.br_startoff
,
4808 got
.br_startblock
, got
.br_blockcount
,
4811 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
4813 da_old
= da_new
= 0;
4815 da_old
= startblockval(got
.br_startblock
);
4821 * Set flag value to use in switch statement.
4822 * Left-contig is 2, right-contig is 1.
4824 switch (((got
.br_startoff
== del
->br_startoff
) << 1) |
4825 (got_endoff
== del_endoff
)) {
4828 * Matches the whole extent. Delete the entry.
4830 xfs_iext_remove(ip
, *idx
, 1,
4831 whichfork
== XFS_ATTR_FORK
? BMAP_ATTRFORK
: 0);
4836 XFS_IFORK_NEXT_SET(ip
, whichfork
,
4837 XFS_IFORK_NEXTENTS(ip
, whichfork
) - 1);
4838 flags
|= XFS_ILOG_CORE
;
4840 flags
|= xfs_ilog_fext(whichfork
);
4843 if ((error
= xfs_btree_delete(cur
, &i
)))
4845 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
4850 * Deleting the first part of the extent.
4852 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
4853 xfs_bmbt_set_startoff(ep
, del_endoff
);
4854 temp
= got
.br_blockcount
- del
->br_blockcount
;
4855 xfs_bmbt_set_blockcount(ep
, temp
);
4857 temp
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
, temp
),
4859 xfs_bmbt_set_startblock(ep
, nullstartblock((int)temp
));
4860 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
4864 xfs_bmbt_set_startblock(ep
, del_endblock
);
4865 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
4867 flags
|= xfs_ilog_fext(whichfork
);
4870 if ((error
= xfs_bmbt_update(cur
, del_endoff
, del_endblock
,
4871 got
.br_blockcount
- del
->br_blockcount
,
4878 * Deleting the last part of the extent.
4880 temp
= got
.br_blockcount
- del
->br_blockcount
;
4881 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
4882 xfs_bmbt_set_blockcount(ep
, temp
);
4884 temp
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
, temp
),
4886 xfs_bmbt_set_startblock(ep
, nullstartblock((int)temp
));
4887 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
4891 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
4893 flags
|= xfs_ilog_fext(whichfork
);
4896 if ((error
= xfs_bmbt_update(cur
, got
.br_startoff
,
4898 got
.br_blockcount
- del
->br_blockcount
,
4905 * Deleting the middle of the extent.
4907 temp
= del
->br_startoff
- got
.br_startoff
;
4908 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
4909 xfs_bmbt_set_blockcount(ep
, temp
);
4910 new.br_startoff
= del_endoff
;
4911 temp2
= got_endoff
- del_endoff
;
4912 new.br_blockcount
= temp2
;
4913 new.br_state
= got
.br_state
;
4915 new.br_startblock
= del_endblock
;
4916 flags
|= XFS_ILOG_CORE
;
4918 if ((error
= xfs_bmbt_update(cur
,
4920 got
.br_startblock
, temp
,
4923 if ((error
= xfs_btree_increment(cur
, 0, &i
)))
4925 cur
->bc_rec
.b
= new;
4926 error
= xfs_btree_insert(cur
, &i
);
4927 if (error
&& error
!= ENOSPC
)
4930 * If get no-space back from btree insert,
4931 * it tried a split, and we have a zero
4932 * block reservation.
4933 * Fix up our state and return the error.
4935 if (error
== ENOSPC
) {
4937 * Reset the cursor, don't trust
4938 * it after any insert operation.
4940 if ((error
= xfs_bmbt_lookup_eq(cur
,
4945 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
4947 * Update the btree record back
4948 * to the original value.
4950 if ((error
= xfs_bmbt_update(cur
,
4957 * Reset the extent record back
4958 * to the original value.
4960 xfs_bmbt_set_blockcount(ep
,
4963 error
= XFS_ERROR(ENOSPC
);
4966 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
4968 flags
|= xfs_ilog_fext(whichfork
);
4969 XFS_IFORK_NEXT_SET(ip
, whichfork
,
4970 XFS_IFORK_NEXTENTS(ip
, whichfork
) + 1);
4972 ASSERT(whichfork
== XFS_DATA_FORK
);
4973 temp
= xfs_bmap_worst_indlen(ip
, temp
);
4974 xfs_bmbt_set_startblock(ep
, nullstartblock((int)temp
));
4975 temp2
= xfs_bmap_worst_indlen(ip
, temp2
);
4976 new.br_startblock
= nullstartblock((int)temp2
);
4977 da_new
= temp
+ temp2
;
4978 while (da_new
> da_old
) {
4982 xfs_bmbt_set_startblock(ep
,
4983 nullstartblock((int)temp
));
4985 if (da_new
== da_old
)
4991 nullstartblock((int)temp2
);
4995 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
4996 xfs_iext_insert(ip
, *idx
+ 1, 1, &new, state
);
5001 * If we need to, add to list of extents to delete.
5004 xfs_bmap_add_free(del
->br_startblock
, del
->br_blockcount
, flist
,
5007 * Adjust inode # blocks in the file.
5010 ip
->i_d
.di_nblocks
-= nblks
;
5012 * Adjust quota data.
5015 xfs_trans_mod_dquot_byino(tp
, ip
, qfield
, (long)-nblks
);
5018 * Account for change in delayed indirect blocks.
5019 * Nothing to do for disk quota accounting here.
5021 ASSERT(da_old
>= da_new
);
5022 if (da_old
> da_new
) {
5023 xfs_icsb_modify_counters(mp
, XFS_SBS_FDBLOCKS
,
5024 (int64_t)(da_old
- da_new
), 0);
5032 * Unmap (remove) blocks from a file.
5033 * If nexts is nonzero then the number of extents to remove is limited to
5034 * that value. If not all extents in the block range can be removed then
5039 xfs_trans_t
*tp
, /* transaction pointer */
5040 struct xfs_inode
*ip
, /* incore inode */
5041 xfs_fileoff_t bno
, /* starting offset to unmap */
5042 xfs_filblks_t len
, /* length to unmap in file */
5043 int flags
, /* misc flags */
5044 xfs_extnum_t nexts
, /* number of extents max */
5045 xfs_fsblock_t
*firstblock
, /* first allocated block
5046 controls a.g. for allocs */
5047 xfs_bmap_free_t
*flist
, /* i/o: list extents to free */
5048 int *done
) /* set if not done yet */
5050 xfs_btree_cur_t
*cur
; /* bmap btree cursor */
5051 xfs_bmbt_irec_t del
; /* extent being deleted */
5052 int eof
; /* is deleting at eof */
5053 xfs_bmbt_rec_host_t
*ep
; /* extent record pointer */
5054 int error
; /* error return value */
5055 xfs_extnum_t extno
; /* extent number in list */
5056 xfs_bmbt_irec_t got
; /* current extent record */
5057 xfs_ifork_t
*ifp
; /* inode fork pointer */
5058 int isrt
; /* freeing in rt area */
5059 xfs_extnum_t lastx
; /* last extent index used */
5060 int logflags
; /* transaction logging flags */
5061 xfs_extlen_t mod
; /* rt extent offset */
5062 xfs_mount_t
*mp
; /* mount structure */
5063 xfs_extnum_t nextents
; /* number of file extents */
5064 xfs_bmbt_irec_t prev
; /* previous extent record */
5065 xfs_fileoff_t start
; /* first file offset deleted */
5066 int tmp_logflags
; /* partial logging flags */
5067 int wasdel
; /* was a delayed alloc extent */
5068 int whichfork
; /* data or attribute fork */
5071 trace_xfs_bunmap(ip
, bno
, len
, flags
, _RET_IP_
);
5073 whichfork
= (flags
& XFS_BMAPI_ATTRFORK
) ?
5074 XFS_ATTR_FORK
: XFS_DATA_FORK
;
5075 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
5077 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
5078 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
)) {
5079 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW
,
5081 return XFS_ERROR(EFSCORRUPTED
);
5084 if (XFS_FORCED_SHUTDOWN(mp
))
5085 return XFS_ERROR(EIO
);
5087 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
5091 if (!(ifp
->if_flags
& XFS_IFEXTENTS
) &&
5092 (error
= xfs_iread_extents(tp
, ip
, whichfork
)))
5094 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
5095 if (nextents
== 0) {
5099 XFS_STATS_INC(xs_blk_unmap
);
5100 isrt
= (whichfork
== XFS_DATA_FORK
) && XFS_IS_REALTIME_INODE(ip
);
5102 bno
= start
+ len
- 1;
5103 ep
= xfs_bmap_search_extents(ip
, bno
, whichfork
, &eof
, &lastx
, &got
,
5107 * Check to see if the given block number is past the end of the
5108 * file, back up to the last block if so...
5111 ep
= xfs_iext_get_ext(ifp
, --lastx
);
5112 xfs_bmbt_get_all(ep
, &got
);
5113 bno
= got
.br_startoff
+ got
.br_blockcount
- 1;
5116 if (ifp
->if_flags
& XFS_IFBROOT
) {
5117 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_BTREE
);
5118 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
5119 cur
->bc_private
.b
.firstblock
= *firstblock
;
5120 cur
->bc_private
.b
.flist
= flist
;
5121 cur
->bc_private
.b
.flags
= 0;
5127 * Synchronize by locking the bitmap inode.
5129 xfs_ilock(mp
->m_rbmip
, XFS_ILOCK_EXCL
);
5130 xfs_trans_ijoin(tp
, mp
->m_rbmip
, XFS_ILOCK_EXCL
);
5134 while (bno
!= (xfs_fileoff_t
)-1 && bno
>= start
&& lastx
>= 0 &&
5135 (nexts
== 0 || extno
< nexts
)) {
5137 * Is the found extent after a hole in which bno lives?
5138 * Just back up to the previous extent, if so.
5140 if (got
.br_startoff
> bno
) {
5143 ep
= xfs_iext_get_ext(ifp
, lastx
);
5144 xfs_bmbt_get_all(ep
, &got
);
5147 * Is the last block of this extent before the range
5148 * we're supposed to delete? If so, we're done.
5150 bno
= XFS_FILEOFF_MIN(bno
,
5151 got
.br_startoff
+ got
.br_blockcount
- 1);
5155 * Then deal with the (possibly delayed) allocated space
5160 wasdel
= isnullstartblock(del
.br_startblock
);
5161 if (got
.br_startoff
< start
) {
5162 del
.br_startoff
= start
;
5163 del
.br_blockcount
-= start
- got
.br_startoff
;
5165 del
.br_startblock
+= start
- got
.br_startoff
;
5167 if (del
.br_startoff
+ del
.br_blockcount
> bno
+ 1)
5168 del
.br_blockcount
= bno
+ 1 - del
.br_startoff
;
5169 sum
= del
.br_startblock
+ del
.br_blockcount
;
5171 (mod
= do_mod(sum
, mp
->m_sb
.sb_rextsize
))) {
5173 * Realtime extent not lined up at the end.
5174 * The extent could have been split into written
5175 * and unwritten pieces, or we could just be
5176 * unmapping part of it. But we can't really
5177 * get rid of part of a realtime extent.
5179 if (del
.br_state
== XFS_EXT_UNWRITTEN
||
5180 !xfs_sb_version_hasextflgbit(&mp
->m_sb
)) {
5182 * This piece is unwritten, or we're not
5183 * using unwritten extents. Skip over it.
5186 bno
-= mod
> del
.br_blockcount
?
5187 del
.br_blockcount
: mod
;
5188 if (bno
< got
.br_startoff
) {
5190 xfs_bmbt_get_all(xfs_iext_get_ext(
5196 * It's written, turn it unwritten.
5197 * This is better than zeroing it.
5199 ASSERT(del
.br_state
== XFS_EXT_NORM
);
5200 ASSERT(xfs_trans_get_block_res(tp
) > 0);
5202 * If this spans a realtime extent boundary,
5203 * chop it back to the start of the one we end at.
5205 if (del
.br_blockcount
> mod
) {
5206 del
.br_startoff
+= del
.br_blockcount
- mod
;
5207 del
.br_startblock
+= del
.br_blockcount
- mod
;
5208 del
.br_blockcount
= mod
;
5210 del
.br_state
= XFS_EXT_UNWRITTEN
;
5211 error
= xfs_bmap_add_extent_unwritten_real(tp
, ip
,
5212 &lastx
, &cur
, &del
, firstblock
, flist
,
5218 if (isrt
&& (mod
= do_mod(del
.br_startblock
, mp
->m_sb
.sb_rextsize
))) {
5220 * Realtime extent is lined up at the end but not
5221 * at the front. We'll get rid of full extents if
5224 mod
= mp
->m_sb
.sb_rextsize
- mod
;
5225 if (del
.br_blockcount
> mod
) {
5226 del
.br_blockcount
-= mod
;
5227 del
.br_startoff
+= mod
;
5228 del
.br_startblock
+= mod
;
5229 } else if ((del
.br_startoff
== start
&&
5230 (del
.br_state
== XFS_EXT_UNWRITTEN
||
5231 xfs_trans_get_block_res(tp
) == 0)) ||
5232 !xfs_sb_version_hasextflgbit(&mp
->m_sb
)) {
5234 * Can't make it unwritten. There isn't
5235 * a full extent here so just skip it.
5237 ASSERT(bno
>= del
.br_blockcount
);
5238 bno
-= del
.br_blockcount
;
5239 if (got
.br_startoff
> bno
) {
5241 ep
= xfs_iext_get_ext(ifp
,
5243 xfs_bmbt_get_all(ep
, &got
);
5247 } else if (del
.br_state
== XFS_EXT_UNWRITTEN
) {
5249 * This one is already unwritten.
5250 * It must have a written left neighbor.
5251 * Unwrite the killed part of that one and
5255 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
,
5257 ASSERT(prev
.br_state
== XFS_EXT_NORM
);
5258 ASSERT(!isnullstartblock(prev
.br_startblock
));
5259 ASSERT(del
.br_startblock
==
5260 prev
.br_startblock
+ prev
.br_blockcount
);
5261 if (prev
.br_startoff
< start
) {
5262 mod
= start
- prev
.br_startoff
;
5263 prev
.br_blockcount
-= mod
;
5264 prev
.br_startblock
+= mod
;
5265 prev
.br_startoff
= start
;
5267 prev
.br_state
= XFS_EXT_UNWRITTEN
;
5269 error
= xfs_bmap_add_extent_unwritten_real(tp
,
5270 ip
, &lastx
, &cur
, &prev
,
5271 firstblock
, flist
, &logflags
);
5276 ASSERT(del
.br_state
== XFS_EXT_NORM
);
5277 del
.br_state
= XFS_EXT_UNWRITTEN
;
5278 error
= xfs_bmap_add_extent_unwritten_real(tp
,
5279 ip
, &lastx
, &cur
, &del
,
5280 firstblock
, flist
, &logflags
);
5287 ASSERT(startblockval(del
.br_startblock
) > 0);
5288 /* Update realtime/data freespace, unreserve quota */
5290 xfs_filblks_t rtexts
;
5292 rtexts
= XFS_FSB_TO_B(mp
, del
.br_blockcount
);
5293 do_div(rtexts
, mp
->m_sb
.sb_rextsize
);
5294 xfs_mod_incore_sb(mp
, XFS_SBS_FREXTENTS
,
5295 (int64_t)rtexts
, 0);
5296 (void)xfs_trans_reserve_quota_nblks(NULL
,
5297 ip
, -((long)del
.br_blockcount
), 0,
5298 XFS_QMOPT_RES_RTBLKS
);
5300 xfs_icsb_modify_counters(mp
, XFS_SBS_FDBLOCKS
,
5301 (int64_t)del
.br_blockcount
, 0);
5302 (void)xfs_trans_reserve_quota_nblks(NULL
,
5303 ip
, -((long)del
.br_blockcount
), 0,
5304 XFS_QMOPT_RES_REGBLKS
);
5306 ip
->i_delayed_blks
-= del
.br_blockcount
;
5308 cur
->bc_private
.b
.flags
|=
5309 XFS_BTCUR_BPRV_WASDEL
;
5311 cur
->bc_private
.b
.flags
&= ~XFS_BTCUR_BPRV_WASDEL
;
5313 * If it's the case where the directory code is running
5314 * with no block reservation, and the deleted block is in
5315 * the middle of its extent, and the resulting insert
5316 * of an extent would cause transformation to btree format,
5317 * then reject it. The calling code will then swap
5318 * blocks around instead.
5319 * We have to do this now, rather than waiting for the
5320 * conversion to btree format, since the transaction
5323 if (!wasdel
&& xfs_trans_get_block_res(tp
) == 0 &&
5324 XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_EXTENTS
&&
5325 XFS_IFORK_NEXTENTS(ip
, whichfork
) >= /* Note the >= */
5326 XFS_IFORK_MAXEXT(ip
, whichfork
) &&
5327 del
.br_startoff
> got
.br_startoff
&&
5328 del
.br_startoff
+ del
.br_blockcount
<
5329 got
.br_startoff
+ got
.br_blockcount
) {
5330 error
= XFS_ERROR(ENOSPC
);
5333 error
= xfs_bmap_del_extent(ip
, tp
, &lastx
, flist
, cur
, &del
,
5334 &tmp_logflags
, whichfork
);
5335 logflags
|= tmp_logflags
;
5338 bno
= del
.br_startoff
- 1;
5341 * If not done go on to the next (previous) record.
5343 if (bno
!= (xfs_fileoff_t
)-1 && bno
>= start
) {
5345 ep
= xfs_iext_get_ext(ifp
, lastx
);
5346 if (xfs_bmbt_get_startoff(ep
) > bno
) {
5348 ep
= xfs_iext_get_ext(ifp
,
5351 xfs_bmbt_get_all(ep
, &got
);
5356 *done
= bno
== (xfs_fileoff_t
)-1 || bno
< start
|| lastx
< 0;
5359 * Convert to a btree if necessary.
5361 if (xfs_bmap_needs_btree(ip
, whichfork
)) {
5362 ASSERT(cur
== NULL
);
5363 error
= xfs_bmap_extents_to_btree(tp
, ip
, firstblock
, flist
,
5364 &cur
, 0, &tmp_logflags
, whichfork
);
5365 logflags
|= tmp_logflags
;
5370 * transform from btree to extents, give it cur
5372 else if (xfs_bmap_wants_extents(ip
, whichfork
)) {
5373 ASSERT(cur
!= NULL
);
5374 error
= xfs_bmap_btree_to_extents(tp
, ip
, cur
, &tmp_logflags
,
5376 logflags
|= tmp_logflags
;
5381 * transform from extents to local?
5386 * Log everything. Do this after conversion, there's no point in
5387 * logging the extent records if we've converted to btree format.
5389 if ((logflags
& xfs_ilog_fext(whichfork
)) &&
5390 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
)
5391 logflags
&= ~xfs_ilog_fext(whichfork
);
5392 else if ((logflags
& xfs_ilog_fbroot(whichfork
)) &&
5393 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
)
5394 logflags
&= ~xfs_ilog_fbroot(whichfork
);
5396 * Log inode even in the error case, if the transaction
5397 * is dirty we'll need to shut down the filesystem.
5400 xfs_trans_log_inode(tp
, ip
, logflags
);
5403 *firstblock
= cur
->bc_private
.b
.firstblock
;
5404 cur
->bc_private
.b
.allocated
= 0;
5406 xfs_btree_del_cursor(cur
,
5407 error
? XFS_BTREE_ERROR
: XFS_BTREE_NOERROR
);
5413 * Shift extent records to the left to cover a hole.
5415 * The maximum number of extents to be shifted in a single operation
5416 * is @num_exts, and @current_ext keeps track of the current extent
5417 * index we have shifted. @offset_shift_fsb is the length by which each
5418 * extent is shifted. If there is no hole to shift the extents
5419 * into, this will be considered invalid operation and we abort immediately.
5422 xfs_bmap_shift_extents(
5423 struct xfs_trans
*tp
,
5424 struct xfs_inode
*ip
,
5426 xfs_fileoff_t start_fsb
,
5427 xfs_fileoff_t offset_shift_fsb
,
5428 xfs_extnum_t
*current_ext
,
5429 xfs_fsblock_t
*firstblock
,
5430 struct xfs_bmap_free
*flist
,
5433 struct xfs_btree_cur
*cur
;
5434 struct xfs_bmbt_rec_host
*gotp
;
5435 struct xfs_bmbt_irec got
;
5436 struct xfs_bmbt_irec left
;
5437 struct xfs_mount
*mp
= ip
->i_mount
;
5438 struct xfs_ifork
*ifp
;
5439 xfs_extnum_t nexts
= 0;
5440 xfs_fileoff_t startoff
;
5443 int whichfork
= XFS_DATA_FORK
;
5445 xfs_filblks_t blockcount
= 0;
5448 if (unlikely(XFS_TEST_ERROR(
5449 (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
5450 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
),
5451 mp
, XFS_ERRTAG_BMAPIFORMAT
, XFS_RANDOM_BMAPIFORMAT
))) {
5452 XFS_ERROR_REPORT("xfs_bmap_shift_extents",
5453 XFS_ERRLEVEL_LOW
, mp
);
5454 return XFS_ERROR(EFSCORRUPTED
);
5457 if (XFS_FORCED_SHUTDOWN(mp
))
5458 return XFS_ERROR(EIO
);
5460 ASSERT(current_ext
!= NULL
);
5462 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
5463 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
5464 /* Read in all the extents */
5465 error
= xfs_iread_extents(tp
, ip
, whichfork
);
5471 * If *current_ext is 0, we would need to lookup the extent
5472 * from where we would start shifting and store it in gotp.
5474 if (!*current_ext
) {
5475 gotp
= xfs_iext_bno_to_ext(ifp
, start_fsb
, current_ext
);
5477 * gotp can be null in 2 cases: 1) if there are no extents
5478 * or 2) start_fsb lies in a hole beyond which there are
5479 * no extents. Either way, we are done.
5487 /* We are going to change core inode */
5488 logflags
= XFS_ILOG_CORE
;
5489 if (ifp
->if_flags
& XFS_IFBROOT
) {
5490 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
5491 cur
->bc_private
.b
.firstblock
= *firstblock
;
5492 cur
->bc_private
.b
.flist
= flist
;
5493 cur
->bc_private
.b
.flags
= 0;
5496 logflags
|= XFS_ILOG_DEXT
;
5500 * There may be delalloc extents in the data fork before the range we
5501 * are collapsing out, so we cannot
5502 * use the count of real extents here. Instead we have to calculate it
5503 * from the incore fork.
5505 total_extents
= ifp
->if_bytes
/ sizeof(xfs_bmbt_rec_t
);
5506 while (nexts
++ < num_exts
&& *current_ext
< total_extents
) {
5508 gotp
= xfs_iext_get_ext(ifp
, *current_ext
);
5509 xfs_bmbt_get_all(gotp
, &got
);
5510 startoff
= got
.br_startoff
- offset_shift_fsb
;
5513 * Before shifting extent into hole, make sure that the hole
5514 * is large enough to accomodate the shift.
5517 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
,
5518 *current_ext
- 1), &left
);
5520 if (startoff
< left
.br_startoff
+ left
.br_blockcount
)
5521 error
= XFS_ERROR(EINVAL
);
5522 } else if (offset_shift_fsb
> got
.br_startoff
) {
5524 * When first extent is shifted, offset_shift_fsb
5525 * should be less than the stating offset of
5528 error
= XFS_ERROR(EINVAL
);
5535 error
= xfs_bmbt_lookup_eq(cur
, got
.br_startoff
,
5541 XFS_WANT_CORRUPTED_GOTO(i
== 1, del_cursor
);
5544 /* Check if we can merge 2 adjacent extents */
5546 left
.br_startoff
+ left
.br_blockcount
== startoff
&&
5547 left
.br_startblock
+ left
.br_blockcount
==
5548 got
.br_startblock
&&
5549 left
.br_state
== got
.br_state
&&
5550 left
.br_blockcount
+ got
.br_blockcount
<= MAXEXTLEN
) {
5551 blockcount
= left
.br_blockcount
+
5553 xfs_iext_remove(ip
, *current_ext
, 1, 0);
5555 error
= xfs_btree_delete(cur
, &i
);
5558 XFS_WANT_CORRUPTED_GOTO(i
== 1, del_cursor
);
5560 XFS_IFORK_NEXT_SET(ip
, whichfork
,
5561 XFS_IFORK_NEXTENTS(ip
, whichfork
) - 1);
5562 gotp
= xfs_iext_get_ext(ifp
, --*current_ext
);
5563 xfs_bmbt_get_all(gotp
, &got
);
5565 /* Make cursor point to the extent we will update */
5567 error
= xfs_bmbt_lookup_eq(cur
, got
.br_startoff
,
5573 XFS_WANT_CORRUPTED_GOTO(i
== 1, del_cursor
);
5576 xfs_bmbt_set_blockcount(gotp
, blockcount
);
5577 got
.br_blockcount
= blockcount
;
5579 /* We have to update the startoff */
5580 xfs_bmbt_set_startoff(gotp
, startoff
);
5581 got
.br_startoff
= startoff
;
5585 error
= xfs_bmbt_update(cur
, got
.br_startoff
,
5594 total_extents
= ifp
->if_bytes
/ sizeof(xfs_bmbt_rec_t
);
5597 /* Check if we are done */
5598 if (*current_ext
== total_extents
)
5603 xfs_btree_del_cursor(cur
,
5604 error
? XFS_BTREE_ERROR
: XFS_BTREE_NOERROR
);
5606 xfs_trans_log_inode(tp
, ip
, logflags
);