1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
14 #include "xfs_mount.h"
15 #include "xfs_defer.h"
17 #include "xfs_inode.h"
18 #include "xfs_btree.h"
19 #include "xfs_trans.h"
20 #include "xfs_alloc.h"
22 #include "xfs_bmap_util.h"
23 #include "xfs_bmap_btree.h"
24 #include "xfs_rtalloc.h"
25 #include "xfs_errortag.h"
26 #include "xfs_error.h"
27 #include "xfs_quota.h"
28 #include "xfs_trans_space.h"
29 #include "xfs_buf_item.h"
30 #include "xfs_trace.h"
31 #include "xfs_attr_leaf.h"
32 #include "xfs_filestream.h"
34 #include "xfs_ag_resv.h"
35 #include "xfs_refcount.h"
36 #include "xfs_icache.h"
37 #include "xfs_iomap.h"
40 kmem_zone_t
*xfs_bmap_free_item_zone
;
43 * Miscellaneous helper functions
47 * Compute and fill in the value of the maximum depth of a bmap btree
48 * in this filesystem. Done once, during mount.
51 xfs_bmap_compute_maxlevels(
52 xfs_mount_t
*mp
, /* file system mount structure */
53 int whichfork
) /* data or attr fork */
55 int level
; /* btree level */
56 uint maxblocks
; /* max blocks at this level */
57 uint maxleafents
; /* max leaf entries possible */
58 int maxrootrecs
; /* max records in root block */
59 int minleafrecs
; /* min records in leaf block */
60 int minnoderecs
; /* min records in node block */
61 int sz
; /* root block size */
64 * The maximum number of extents in a file, hence the maximum number of
65 * leaf entries, is controlled by the size of the on-disk extent count,
66 * either a signed 32-bit number for the data fork, or a signed 16-bit
67 * number for the attr fork.
69 * Note that we can no longer assume that if we are in ATTR1 that
70 * the fork offset of all the inodes will be
71 * (xfs_default_attroffset(ip) >> 3) because we could have mounted
72 * with ATTR2 and then mounted back with ATTR1, keeping the
73 * di_forkoff's fixed but probably at various positions. Therefore,
74 * for both ATTR1 and ATTR2 we have to assume the worst case scenario
75 * of a minimum size available.
77 if (whichfork
== XFS_DATA_FORK
) {
78 maxleafents
= MAXEXTNUM
;
79 sz
= XFS_BMDR_SPACE_CALC(MINDBTPTRS
);
81 maxleafents
= MAXAEXTNUM
;
82 sz
= XFS_BMDR_SPACE_CALC(MINABTPTRS
);
84 maxrootrecs
= xfs_bmdr_maxrecs(sz
, 0);
85 minleafrecs
= mp
->m_bmap_dmnr
[0];
86 minnoderecs
= mp
->m_bmap_dmnr
[1];
87 maxblocks
= (maxleafents
+ minleafrecs
- 1) / minleafrecs
;
88 for (level
= 1; maxblocks
> 1; level
++) {
89 if (maxblocks
<= maxrootrecs
)
92 maxblocks
= (maxblocks
+ minnoderecs
- 1) / minnoderecs
;
94 mp
->m_bm_maxlevels
[whichfork
] = level
;
97 STATIC
int /* error */
99 struct xfs_btree_cur
*cur
,
100 struct xfs_bmbt_irec
*irec
,
101 int *stat
) /* success/failure */
103 cur
->bc_rec
.b
= *irec
;
104 return xfs_btree_lookup(cur
, XFS_LOOKUP_EQ
, stat
);
107 STATIC
int /* error */
108 xfs_bmbt_lookup_first(
109 struct xfs_btree_cur
*cur
,
110 int *stat
) /* success/failure */
112 cur
->bc_rec
.b
.br_startoff
= 0;
113 cur
->bc_rec
.b
.br_startblock
= 0;
114 cur
->bc_rec
.b
.br_blockcount
= 0;
115 return xfs_btree_lookup(cur
, XFS_LOOKUP_GE
, stat
);
119 * Check if the inode needs to be converted to btree format.
121 static inline bool xfs_bmap_needs_btree(struct xfs_inode
*ip
, int whichfork
)
123 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
125 return whichfork
!= XFS_COW_FORK
&&
126 ifp
->if_format
== XFS_DINODE_FMT_EXTENTS
&&
127 ifp
->if_nextents
> XFS_IFORK_MAXEXT(ip
, whichfork
);
131 * Check if the inode should be converted to extent format.
133 static inline bool xfs_bmap_wants_extents(struct xfs_inode
*ip
, int whichfork
)
135 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
137 return whichfork
!= XFS_COW_FORK
&&
138 ifp
->if_format
== XFS_DINODE_FMT_BTREE
&&
139 ifp
->if_nextents
<= XFS_IFORK_MAXEXT(ip
, whichfork
);
143 * Update the record referred to by cur to the value given by irec
144 * This either works (return 0) or gets an EFSCORRUPTED error.
148 struct xfs_btree_cur
*cur
,
149 struct xfs_bmbt_irec
*irec
)
151 union xfs_btree_rec rec
;
153 xfs_bmbt_disk_set_all(&rec
.bmbt
, irec
);
154 return xfs_btree_update(cur
, &rec
);
158 * Compute the worst-case number of indirect blocks that will be used
159 * for ip's delayed extent of length "len".
162 xfs_bmap_worst_indlen(
163 xfs_inode_t
*ip
, /* incore inode pointer */
164 xfs_filblks_t len
) /* delayed extent length */
166 int level
; /* btree level number */
167 int maxrecs
; /* maximum record count at this level */
168 xfs_mount_t
*mp
; /* mount structure */
169 xfs_filblks_t rval
; /* return value */
172 maxrecs
= mp
->m_bmap_dmxr
[0];
173 for (level
= 0, rval
= 0;
174 level
< XFS_BM_MAXLEVELS(mp
, XFS_DATA_FORK
);
177 do_div(len
, maxrecs
);
180 return rval
+ XFS_BM_MAXLEVELS(mp
, XFS_DATA_FORK
) -
183 maxrecs
= mp
->m_bmap_dmxr
[1];
189 * Calculate the default attribute fork offset for newly created inodes.
192 xfs_default_attroffset(
193 struct xfs_inode
*ip
)
195 struct xfs_mount
*mp
= ip
->i_mount
;
198 if (mp
->m_sb
.sb_inodesize
== 256)
199 offset
= XFS_LITINO(mp
) - XFS_BMDR_SPACE_CALC(MINABTPTRS
);
201 offset
= XFS_BMDR_SPACE_CALC(6 * MINABTPTRS
);
203 ASSERT(offset
< XFS_LITINO(mp
));
208 * Helper routine to reset inode di_forkoff field when switching
209 * attribute fork from local to extent format - we reset it where
210 * possible to make space available for inline data fork extents.
213 xfs_bmap_forkoff_reset(
217 if (whichfork
== XFS_ATTR_FORK
&&
218 ip
->i_df
.if_format
!= XFS_DINODE_FMT_DEV
&&
219 ip
->i_df
.if_format
!= XFS_DINODE_FMT_BTREE
) {
220 uint dfl_forkoff
= xfs_default_attroffset(ip
) >> 3;
222 if (dfl_forkoff
> ip
->i_d
.di_forkoff
)
223 ip
->i_d
.di_forkoff
= dfl_forkoff
;
228 STATIC
struct xfs_buf
*
230 struct xfs_btree_cur
*cur
,
233 struct xfs_log_item
*lip
;
239 for (i
= 0; i
< XFS_BTREE_MAXLEVELS
; i
++) {
240 if (!cur
->bc_bufs
[i
])
242 if (XFS_BUF_ADDR(cur
->bc_bufs
[i
]) == bno
)
243 return cur
->bc_bufs
[i
];
246 /* Chase down all the log items to see if the bp is there */
247 list_for_each_entry(lip
, &cur
->bc_tp
->t_items
, li_trans
) {
248 struct xfs_buf_log_item
*bip
= (struct xfs_buf_log_item
*)lip
;
250 if (bip
->bli_item
.li_type
== XFS_LI_BUF
&&
251 XFS_BUF_ADDR(bip
->bli_buf
) == bno
)
260 struct xfs_btree_block
*block
,
266 __be64
*pp
, *thispa
; /* pointer to block address */
267 xfs_bmbt_key_t
*prevp
, *keyp
;
269 ASSERT(be16_to_cpu(block
->bb_level
) > 0);
272 for( i
= 1; i
<= xfs_btree_get_numrecs(block
); i
++) {
273 dmxr
= mp
->m_bmap_dmxr
[0];
274 keyp
= XFS_BMBT_KEY_ADDR(mp
, block
, i
);
277 ASSERT(be64_to_cpu(prevp
->br_startoff
) <
278 be64_to_cpu(keyp
->br_startoff
));
283 * Compare the block numbers to see if there are dups.
286 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, i
, sz
);
288 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, i
, dmxr
);
290 for (j
= i
+1; j
<= be16_to_cpu(block
->bb_numrecs
); j
++) {
292 thispa
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, j
, sz
);
294 thispa
= XFS_BMBT_PTR_ADDR(mp
, block
, j
, dmxr
);
295 if (*thispa
== *pp
) {
296 xfs_warn(mp
, "%s: thispa(%d) == pp(%d) %Ld",
298 (unsigned long long)be64_to_cpu(*thispa
));
299 xfs_err(mp
, "%s: ptrs are equal in node\n",
301 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
308 * Check that the extents for the inode ip are in the right order in all
309 * btree leaves. THis becomes prohibitively expensive for large extent count
310 * files, so don't bother with inodes that have more than 10,000 extents in
311 * them. The btree record ordering checks will still be done, so for such large
312 * bmapbt constructs that is going to catch most corruptions.
315 xfs_bmap_check_leaf_extents(
316 xfs_btree_cur_t
*cur
, /* btree cursor or null */
317 xfs_inode_t
*ip
, /* incore inode pointer */
318 int whichfork
) /* data or attr fork */
320 struct xfs_mount
*mp
= ip
->i_mount
;
321 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
322 struct xfs_btree_block
*block
; /* current btree block */
323 xfs_fsblock_t bno
; /* block # of "block" */
324 struct xfs_buf
*bp
; /* buffer for "block" */
325 int error
; /* error return value */
326 xfs_extnum_t i
=0, j
; /* index into the extents list */
327 int level
; /* btree level, for checking */
328 __be64
*pp
; /* pointer to block address */
329 xfs_bmbt_rec_t
*ep
; /* pointer to current extent */
330 xfs_bmbt_rec_t last
= {0, 0}; /* last extent in prev block */
331 xfs_bmbt_rec_t
*nextp
; /* pointer to next extent */
334 if (ifp
->if_format
!= XFS_DINODE_FMT_BTREE
)
337 /* skip large extent count inodes */
338 if (ip
->i_df
.if_nextents
> 10000)
342 block
= ifp
->if_broot
;
344 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
346 level
= be16_to_cpu(block
->bb_level
);
348 xfs_check_block(block
, mp
, 1, ifp
->if_broot_bytes
);
349 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, 1, ifp
->if_broot_bytes
);
350 bno
= be64_to_cpu(*pp
);
352 ASSERT(bno
!= NULLFSBLOCK
);
353 ASSERT(XFS_FSB_TO_AGNO(mp
, bno
) < mp
->m_sb
.sb_agcount
);
354 ASSERT(XFS_FSB_TO_AGBNO(mp
, bno
) < mp
->m_sb
.sb_agblocks
);
357 * Go down the tree until leaf level is reached, following the first
358 * pointer (leftmost) at each level.
360 while (level
-- > 0) {
361 /* See if buf is in cur first */
363 bp
= xfs_bmap_get_bp(cur
, XFS_FSB_TO_DADDR(mp
, bno
));
366 error
= xfs_btree_read_bufl(mp
, NULL
, bno
, &bp
,
372 block
= XFS_BUF_TO_BLOCK(bp
);
377 * Check this block for basic sanity (increasing keys and
378 * no duplicate blocks).
381 xfs_check_block(block
, mp
, 0, 0);
382 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, 1, mp
->m_bmap_dmxr
[1]);
383 bno
= be64_to_cpu(*pp
);
384 if (XFS_IS_CORRUPT(mp
, !xfs_verify_fsbno(mp
, bno
))) {
385 error
= -EFSCORRUPTED
;
390 xfs_trans_brelse(NULL
, bp
);
395 * Here with bp and block set to the leftmost leaf node in the tree.
400 * Loop over all leaf nodes checking that all extents are in the right order.
403 xfs_fsblock_t nextbno
;
404 xfs_extnum_t num_recs
;
407 num_recs
= xfs_btree_get_numrecs(block
);
410 * Read-ahead the next leaf block, if any.
413 nextbno
= be64_to_cpu(block
->bb_u
.l
.bb_rightsib
);
416 * Check all the extents to make sure they are OK.
417 * If we had a previous block, the last entry should
418 * conform with the first entry in this one.
421 ep
= XFS_BMBT_REC_ADDR(mp
, block
, 1);
423 ASSERT(xfs_bmbt_disk_get_startoff(&last
) +
424 xfs_bmbt_disk_get_blockcount(&last
) <=
425 xfs_bmbt_disk_get_startoff(ep
));
427 for (j
= 1; j
< num_recs
; j
++) {
428 nextp
= XFS_BMBT_REC_ADDR(mp
, block
, j
+ 1);
429 ASSERT(xfs_bmbt_disk_get_startoff(ep
) +
430 xfs_bmbt_disk_get_blockcount(ep
) <=
431 xfs_bmbt_disk_get_startoff(nextp
));
439 xfs_trans_brelse(NULL
, bp
);
443 * If we've reached the end, stop.
445 if (bno
== NULLFSBLOCK
)
449 bp
= xfs_bmap_get_bp(cur
, XFS_FSB_TO_DADDR(mp
, bno
));
452 error
= xfs_btree_read_bufl(mp
, NULL
, bno
, &bp
,
458 block
= XFS_BUF_TO_BLOCK(bp
);
464 xfs_warn(mp
, "%s: at error0", __func__
);
466 xfs_trans_brelse(NULL
, bp
);
468 xfs_warn(mp
, "%s: BAD after btree leaves for %d extents",
470 xfs_err(mp
, "%s: CORRUPTED BTREE OR SOMETHING", __func__
);
471 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
476 * Validate that the bmbt_irecs being returned from bmapi are valid
477 * given the caller's original parameters. Specifically check the
478 * ranges of the returned irecs to ensure that they only extend beyond
479 * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
482 xfs_bmap_validate_ret(
486 xfs_bmbt_irec_t
*mval
,
490 int i
; /* index to map values */
492 ASSERT(ret_nmap
<= nmap
);
494 for (i
= 0; i
< ret_nmap
; i
++) {
495 ASSERT(mval
[i
].br_blockcount
> 0);
496 if (!(flags
& XFS_BMAPI_ENTIRE
)) {
497 ASSERT(mval
[i
].br_startoff
>= bno
);
498 ASSERT(mval
[i
].br_blockcount
<= len
);
499 ASSERT(mval
[i
].br_startoff
+ mval
[i
].br_blockcount
<=
502 ASSERT(mval
[i
].br_startoff
< bno
+ len
);
503 ASSERT(mval
[i
].br_startoff
+ mval
[i
].br_blockcount
>
507 mval
[i
- 1].br_startoff
+ mval
[i
- 1].br_blockcount
==
508 mval
[i
].br_startoff
);
509 ASSERT(mval
[i
].br_startblock
!= DELAYSTARTBLOCK
&&
510 mval
[i
].br_startblock
!= HOLESTARTBLOCK
);
511 ASSERT(mval
[i
].br_state
== XFS_EXT_NORM
||
512 mval
[i
].br_state
== XFS_EXT_UNWRITTEN
);
517 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork) do { } while (0)
518 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap) do { } while (0)
522 * bmap free list manipulation functions
526 * Add the extent to the list of extents to be free at transaction end.
527 * The list is maintained sorted (by block number).
531 struct xfs_trans
*tp
,
534 const struct xfs_owner_info
*oinfo
,
537 struct xfs_extent_free_item
*new; /* new element */
539 struct xfs_mount
*mp
= tp
->t_mountp
;
543 ASSERT(bno
!= NULLFSBLOCK
);
545 ASSERT(len
<= MAXEXTLEN
);
546 ASSERT(!isnullstartblock(bno
));
547 agno
= XFS_FSB_TO_AGNO(mp
, bno
);
548 agbno
= XFS_FSB_TO_AGBNO(mp
, bno
);
549 ASSERT(agno
< mp
->m_sb
.sb_agcount
);
550 ASSERT(agbno
< mp
->m_sb
.sb_agblocks
);
551 ASSERT(len
< mp
->m_sb
.sb_agblocks
);
552 ASSERT(agbno
+ len
<= mp
->m_sb
.sb_agblocks
);
554 ASSERT(xfs_bmap_free_item_zone
!= NULL
);
556 new = kmem_cache_alloc(xfs_bmap_free_item_zone
,
557 GFP_KERNEL
| __GFP_NOFAIL
);
558 new->xefi_startblock
= bno
;
559 new->xefi_blockcount
= (xfs_extlen_t
)len
;
561 new->xefi_oinfo
= *oinfo
;
563 new->xefi_oinfo
= XFS_RMAP_OINFO_SKIP_UPDATE
;
564 new->xefi_skip_discard
= skip_discard
;
565 trace_xfs_bmap_free_defer(tp
->t_mountp
,
566 XFS_FSB_TO_AGNO(tp
->t_mountp
, bno
), 0,
567 XFS_FSB_TO_AGBNO(tp
->t_mountp
, bno
), len
);
568 xfs_defer_add(tp
, XFS_DEFER_OPS_TYPE_FREE
, &new->xefi_list
);
572 * Inode fork format manipulation functions
576 * Convert the inode format to extent format if it currently is in btree format,
577 * but the extent list is small enough that it fits into the extent format.
579 * Since the extents are already in-core, all we have to do is give up the space
580 * for the btree root and pitch the leaf block.
582 STATIC
int /* error */
583 xfs_bmap_btree_to_extents(
584 struct xfs_trans
*tp
, /* transaction pointer */
585 struct xfs_inode
*ip
, /* incore inode pointer */
586 struct xfs_btree_cur
*cur
, /* btree cursor */
587 int *logflagsp
, /* inode logging flags */
588 int whichfork
) /* data or attr fork */
590 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
591 struct xfs_mount
*mp
= ip
->i_mount
;
592 struct xfs_btree_block
*rblock
= ifp
->if_broot
;
593 struct xfs_btree_block
*cblock
;/* child btree block */
594 xfs_fsblock_t cbno
; /* child block number */
595 struct xfs_buf
*cbp
; /* child block's buffer */
596 int error
; /* error return value */
597 __be64
*pp
; /* ptr to block address */
598 struct xfs_owner_info oinfo
;
600 /* check if we actually need the extent format first: */
601 if (!xfs_bmap_wants_extents(ip
, whichfork
))
605 ASSERT(whichfork
!= XFS_COW_FORK
);
606 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
607 ASSERT(ifp
->if_format
== XFS_DINODE_FMT_BTREE
);
608 ASSERT(be16_to_cpu(rblock
->bb_level
) == 1);
609 ASSERT(be16_to_cpu(rblock
->bb_numrecs
) == 1);
610 ASSERT(xfs_bmbt_maxrecs(mp
, ifp
->if_broot_bytes
, 0) == 1);
612 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, rblock
, 1, ifp
->if_broot_bytes
);
613 cbno
= be64_to_cpu(*pp
);
615 if (XFS_IS_CORRUPT(cur
->bc_mp
, !xfs_btree_check_lptr(cur
, cbno
, 1)))
616 return -EFSCORRUPTED
;
618 error
= xfs_btree_read_bufl(mp
, tp
, cbno
, &cbp
, XFS_BMAP_BTREE_REF
,
622 cblock
= XFS_BUF_TO_BLOCK(cbp
);
623 if ((error
= xfs_btree_check_block(cur
, cblock
, 0, cbp
)))
625 xfs_rmap_ino_bmbt_owner(&oinfo
, ip
->i_ino
, whichfork
);
626 xfs_bmap_add_free(cur
->bc_tp
, cbno
, 1, &oinfo
);
627 ip
->i_d
.di_nblocks
--;
628 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_BCOUNT
, -1L);
629 xfs_trans_binval(tp
, cbp
);
630 if (cur
->bc_bufs
[0] == cbp
)
631 cur
->bc_bufs
[0] = NULL
;
632 xfs_iroot_realloc(ip
, -1, whichfork
);
633 ASSERT(ifp
->if_broot
== NULL
);
634 ASSERT((ifp
->if_flags
& XFS_IFBROOT
) == 0);
635 ifp
->if_format
= XFS_DINODE_FMT_EXTENTS
;
636 *logflagsp
|= XFS_ILOG_CORE
| xfs_ilog_fext(whichfork
);
641 * Convert an extents-format file into a btree-format file.
642 * The new file will have a root block (in the inode) and a single child block.
644 STATIC
int /* error */
645 xfs_bmap_extents_to_btree(
646 struct xfs_trans
*tp
, /* transaction pointer */
647 struct xfs_inode
*ip
, /* incore inode pointer */
648 struct xfs_btree_cur
**curp
, /* cursor returned to caller */
649 int wasdel
, /* converting a delayed alloc */
650 int *logflagsp
, /* inode logging flags */
651 int whichfork
) /* data or attr fork */
653 struct xfs_btree_block
*ablock
; /* allocated (child) bt block */
654 struct xfs_buf
*abp
; /* buffer for ablock */
655 struct xfs_alloc_arg args
; /* allocation arguments */
656 struct xfs_bmbt_rec
*arp
; /* child record pointer */
657 struct xfs_btree_block
*block
; /* btree root block */
658 struct xfs_btree_cur
*cur
; /* bmap btree cursor */
659 int error
; /* error return value */
660 struct xfs_ifork
*ifp
; /* inode fork pointer */
661 struct xfs_bmbt_key
*kp
; /* root block key pointer */
662 struct xfs_mount
*mp
; /* mount structure */
663 xfs_bmbt_ptr_t
*pp
; /* root block address pointer */
664 struct xfs_iext_cursor icur
;
665 struct xfs_bmbt_irec rec
;
666 xfs_extnum_t cnt
= 0;
669 ASSERT(whichfork
!= XFS_COW_FORK
);
670 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
671 ASSERT(ifp
->if_format
== XFS_DINODE_FMT_EXTENTS
);
674 * Make space in the inode incore. This needs to be undone if we fail
675 * to expand the root.
677 xfs_iroot_realloc(ip
, 1, whichfork
);
678 ifp
->if_flags
|= XFS_IFBROOT
;
683 block
= ifp
->if_broot
;
684 xfs_btree_init_block_int(mp
, block
, XFS_BUF_DADDR_NULL
,
685 XFS_BTNUM_BMAP
, 1, 1, ip
->i_ino
,
686 XFS_BTREE_LONG_PTRS
);
688 * Need a cursor. Can't allocate until bb_level is filled in.
690 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
691 cur
->bc_ino
.flags
= wasdel
? XFS_BTCUR_BMBT_WASDEL
: 0;
693 * Convert to a btree with two levels, one record in root.
695 ifp
->if_format
= XFS_DINODE_FMT_BTREE
;
696 memset(&args
, 0, sizeof(args
));
699 xfs_rmap_ino_bmbt_owner(&args
.oinfo
, ip
->i_ino
, whichfork
);
700 if (tp
->t_firstblock
== NULLFSBLOCK
) {
701 args
.type
= XFS_ALLOCTYPE_START_BNO
;
702 args
.fsbno
= XFS_INO_TO_FSB(mp
, ip
->i_ino
);
703 } else if (tp
->t_flags
& XFS_TRANS_LOWMODE
) {
704 args
.type
= XFS_ALLOCTYPE_START_BNO
;
705 args
.fsbno
= tp
->t_firstblock
;
707 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
708 args
.fsbno
= tp
->t_firstblock
;
710 args
.minlen
= args
.maxlen
= args
.prod
= 1;
711 args
.wasdel
= wasdel
;
713 error
= xfs_alloc_vextent(&args
);
715 goto out_root_realloc
;
717 if (WARN_ON_ONCE(args
.fsbno
== NULLFSBLOCK
)) {
719 goto out_root_realloc
;
723 * Allocation can't fail, the space was reserved.
725 ASSERT(tp
->t_firstblock
== NULLFSBLOCK
||
726 args
.agno
>= XFS_FSB_TO_AGNO(mp
, tp
->t_firstblock
));
727 tp
->t_firstblock
= args
.fsbno
;
728 cur
->bc_ino
.allocated
++;
729 ip
->i_d
.di_nblocks
++;
730 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_BCOUNT
, 1L);
731 error
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
,
732 XFS_FSB_TO_DADDR(mp
, args
.fsbno
),
733 mp
->m_bsize
, 0, &abp
);
735 goto out_unreserve_dquot
;
738 * Fill in the child block.
740 abp
->b_ops
= &xfs_bmbt_buf_ops
;
741 ablock
= XFS_BUF_TO_BLOCK(abp
);
742 xfs_btree_init_block_int(mp
, ablock
, abp
->b_bn
,
743 XFS_BTNUM_BMAP
, 0, 0, ip
->i_ino
,
744 XFS_BTREE_LONG_PTRS
);
746 for_each_xfs_iext(ifp
, &icur
, &rec
) {
747 if (isnullstartblock(rec
.br_startblock
))
749 arp
= XFS_BMBT_REC_ADDR(mp
, ablock
, 1 + cnt
);
750 xfs_bmbt_disk_set_all(arp
, &rec
);
753 ASSERT(cnt
== ifp
->if_nextents
);
754 xfs_btree_set_numrecs(ablock
, cnt
);
757 * Fill in the root key and pointer.
759 kp
= XFS_BMBT_KEY_ADDR(mp
, block
, 1);
760 arp
= XFS_BMBT_REC_ADDR(mp
, ablock
, 1);
761 kp
->br_startoff
= cpu_to_be64(xfs_bmbt_disk_get_startoff(arp
));
762 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, 1, xfs_bmbt_get_maxrecs(cur
,
763 be16_to_cpu(block
->bb_level
)));
764 *pp
= cpu_to_be64(args
.fsbno
);
767 * Do all this logging at the end so that
768 * the root is at the right level.
770 xfs_btree_log_block(cur
, abp
, XFS_BB_ALL_BITS
);
771 xfs_btree_log_recs(cur
, abp
, 1, be16_to_cpu(ablock
->bb_numrecs
));
772 ASSERT(*curp
== NULL
);
774 *logflagsp
= XFS_ILOG_CORE
| xfs_ilog_fbroot(whichfork
);
778 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_BCOUNT
, -1L);
780 xfs_iroot_realloc(ip
, -1, whichfork
);
781 ifp
->if_format
= XFS_DINODE_FMT_EXTENTS
;
782 ASSERT(ifp
->if_broot
== NULL
);
783 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
789 * Convert a local file to an extents file.
790 * This code is out of bounds for data forks of regular files,
791 * since the file data needs to get logged so things will stay consistent.
792 * (The bmap-level manipulations are ok, though).
795 xfs_bmap_local_to_extents_empty(
796 struct xfs_trans
*tp
,
797 struct xfs_inode
*ip
,
800 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
802 ASSERT(whichfork
!= XFS_COW_FORK
);
803 ASSERT(ifp
->if_format
== XFS_DINODE_FMT_LOCAL
);
804 ASSERT(ifp
->if_bytes
== 0);
805 ASSERT(ifp
->if_nextents
== 0);
807 xfs_bmap_forkoff_reset(ip
, whichfork
);
808 ifp
->if_flags
&= ~XFS_IFINLINE
;
809 ifp
->if_flags
|= XFS_IFEXTENTS
;
810 ifp
->if_u1
.if_root
= NULL
;
812 ifp
->if_format
= XFS_DINODE_FMT_EXTENTS
;
813 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
817 STATIC
int /* error */
818 xfs_bmap_local_to_extents(
819 xfs_trans_t
*tp
, /* transaction pointer */
820 xfs_inode_t
*ip
, /* incore inode pointer */
821 xfs_extlen_t total
, /* total blocks needed by transaction */
822 int *logflagsp
, /* inode logging flags */
824 void (*init_fn
)(struct xfs_trans
*tp
,
826 struct xfs_inode
*ip
,
827 struct xfs_ifork
*ifp
))
830 int flags
; /* logging flags returned */
831 struct xfs_ifork
*ifp
; /* inode fork pointer */
832 xfs_alloc_arg_t args
; /* allocation arguments */
833 struct xfs_buf
*bp
; /* buffer for extent block */
834 struct xfs_bmbt_irec rec
;
835 struct xfs_iext_cursor icur
;
838 * We don't want to deal with the case of keeping inode data inline yet.
839 * So sending the data fork of a regular inode is invalid.
841 ASSERT(!(S_ISREG(VFS_I(ip
)->i_mode
) && whichfork
== XFS_DATA_FORK
));
842 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
843 ASSERT(ifp
->if_format
== XFS_DINODE_FMT_LOCAL
);
845 if (!ifp
->if_bytes
) {
846 xfs_bmap_local_to_extents_empty(tp
, ip
, whichfork
);
847 flags
= XFS_ILOG_CORE
;
853 ASSERT((ifp
->if_flags
& (XFS_IFINLINE
|XFS_IFEXTENTS
)) == XFS_IFINLINE
);
854 memset(&args
, 0, sizeof(args
));
856 args
.mp
= ip
->i_mount
;
857 xfs_rmap_ino_owner(&args
.oinfo
, ip
->i_ino
, whichfork
, 0);
859 * Allocate a block. We know we need only one, since the
860 * file currently fits in an inode.
862 if (tp
->t_firstblock
== NULLFSBLOCK
) {
863 args
.fsbno
= XFS_INO_TO_FSB(args
.mp
, ip
->i_ino
);
864 args
.type
= XFS_ALLOCTYPE_START_BNO
;
866 args
.fsbno
= tp
->t_firstblock
;
867 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
870 args
.minlen
= args
.maxlen
= args
.prod
= 1;
871 error
= xfs_alloc_vextent(&args
);
875 /* Can't fail, the space was reserved. */
876 ASSERT(args
.fsbno
!= NULLFSBLOCK
);
877 ASSERT(args
.len
== 1);
878 tp
->t_firstblock
= args
.fsbno
;
879 error
= xfs_trans_get_buf(tp
, args
.mp
->m_ddev_targp
,
880 XFS_FSB_TO_DADDR(args
.mp
, args
.fsbno
),
881 args
.mp
->m_bsize
, 0, &bp
);
886 * Initialize the block, copy the data and log the remote buffer.
888 * The callout is responsible for logging because the remote format
889 * might differ from the local format and thus we don't know how much to
890 * log here. Note that init_fn must also set the buffer log item type
893 init_fn(tp
, bp
, ip
, ifp
);
895 /* account for the change in fork size */
896 xfs_idata_realloc(ip
, -ifp
->if_bytes
, whichfork
);
897 xfs_bmap_local_to_extents_empty(tp
, ip
, whichfork
);
898 flags
|= XFS_ILOG_CORE
;
900 ifp
->if_u1
.if_root
= NULL
;
904 rec
.br_startblock
= args
.fsbno
;
905 rec
.br_blockcount
= 1;
906 rec
.br_state
= XFS_EXT_NORM
;
907 xfs_iext_first(ifp
, &icur
);
908 xfs_iext_insert(ip
, &icur
, &rec
, 0);
910 ifp
->if_nextents
= 1;
911 ip
->i_d
.di_nblocks
= 1;
912 xfs_trans_mod_dquot_byino(tp
, ip
,
913 XFS_TRANS_DQ_BCOUNT
, 1L);
914 flags
|= xfs_ilog_fext(whichfork
);
922 * Called from xfs_bmap_add_attrfork to handle btree format files.
924 STATIC
int /* error */
925 xfs_bmap_add_attrfork_btree(
926 xfs_trans_t
*tp
, /* transaction pointer */
927 xfs_inode_t
*ip
, /* incore inode pointer */
928 int *flags
) /* inode logging flags */
930 xfs_btree_cur_t
*cur
; /* btree cursor */
931 int error
; /* error return value */
932 xfs_mount_t
*mp
; /* file system mount struct */
933 int stat
; /* newroot status */
936 if (ip
->i_df
.if_broot_bytes
<= XFS_IFORK_DSIZE(ip
))
937 *flags
|= XFS_ILOG_DBROOT
;
939 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, XFS_DATA_FORK
);
940 error
= xfs_bmbt_lookup_first(cur
, &stat
);
943 /* must be at least one entry */
944 if (XFS_IS_CORRUPT(mp
, stat
!= 1)) {
945 error
= -EFSCORRUPTED
;
948 if ((error
= xfs_btree_new_iroot(cur
, flags
, &stat
)))
951 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
954 cur
->bc_ino
.allocated
= 0;
955 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
959 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
964 * Called from xfs_bmap_add_attrfork to handle extents format files.
966 STATIC
int /* error */
967 xfs_bmap_add_attrfork_extents(
968 struct xfs_trans
*tp
, /* transaction pointer */
969 struct xfs_inode
*ip
, /* incore inode pointer */
970 int *flags
) /* inode logging flags */
972 xfs_btree_cur_t
*cur
; /* bmap btree cursor */
973 int error
; /* error return value */
975 if (ip
->i_df
.if_nextents
* sizeof(struct xfs_bmbt_rec
) <=
979 error
= xfs_bmap_extents_to_btree(tp
, ip
, &cur
, 0, flags
,
982 cur
->bc_ino
.allocated
= 0;
983 xfs_btree_del_cursor(cur
, error
);
989 * Called from xfs_bmap_add_attrfork to handle local format files. Each
990 * different data fork content type needs a different callout to do the
991 * conversion. Some are basic and only require special block initialisation
992 * callouts for the data formating, others (directories) are so specialised they
993 * handle everything themselves.
995 * XXX (dgc): investigate whether directory conversion can use the generic
996 * formatting callout. It should be possible - it's just a very complex
999 STATIC
int /* error */
1000 xfs_bmap_add_attrfork_local(
1001 struct xfs_trans
*tp
, /* transaction pointer */
1002 struct xfs_inode
*ip
, /* incore inode pointer */
1003 int *flags
) /* inode logging flags */
1005 struct xfs_da_args dargs
; /* args for dir/attr code */
1007 if (ip
->i_df
.if_bytes
<= XFS_IFORK_DSIZE(ip
))
1010 if (S_ISDIR(VFS_I(ip
)->i_mode
)) {
1011 memset(&dargs
, 0, sizeof(dargs
));
1012 dargs
.geo
= ip
->i_mount
->m_dir_geo
;
1014 dargs
.total
= dargs
.geo
->fsbcount
;
1015 dargs
.whichfork
= XFS_DATA_FORK
;
1017 return xfs_dir2_sf_to_block(&dargs
);
1020 if (S_ISLNK(VFS_I(ip
)->i_mode
))
1021 return xfs_bmap_local_to_extents(tp
, ip
, 1, flags
,
1023 xfs_symlink_local_to_remote
);
1025 /* should only be called for types that support local format data */
1027 return -EFSCORRUPTED
;
1030 /* Set an inode attr fork off based on the format */
1032 xfs_bmap_set_attrforkoff(
1033 struct xfs_inode
*ip
,
1037 switch (ip
->i_df
.if_format
) {
1038 case XFS_DINODE_FMT_DEV
:
1039 ip
->i_d
.di_forkoff
= roundup(sizeof(xfs_dev_t
), 8) >> 3;
1041 case XFS_DINODE_FMT_LOCAL
:
1042 case XFS_DINODE_FMT_EXTENTS
:
1043 case XFS_DINODE_FMT_BTREE
:
1044 ip
->i_d
.di_forkoff
= xfs_attr_shortform_bytesfit(ip
, size
);
1045 if (!ip
->i_d
.di_forkoff
)
1046 ip
->i_d
.di_forkoff
= xfs_default_attroffset(ip
) >> 3;
1047 else if ((ip
->i_mount
->m_flags
& XFS_MOUNT_ATTR2
) && version
)
1059 * Convert inode from non-attributed to attributed.
1060 * Must not be in a transaction, ip must not be locked.
1062 int /* error code */
1063 xfs_bmap_add_attrfork(
1064 xfs_inode_t
*ip
, /* incore inode pointer */
1065 int size
, /* space new attribute needs */
1066 int rsvd
) /* xact may use reserved blks */
1068 xfs_mount_t
*mp
; /* mount structure */
1069 xfs_trans_t
*tp
; /* transaction pointer */
1070 int blks
; /* space reservation */
1071 int version
= 1; /* superblock attr version */
1072 int logflags
; /* logging flags */
1073 int error
; /* error return value */
1075 ASSERT(XFS_IFORK_Q(ip
) == 0);
1078 ASSERT(!XFS_NOT_DQATTACHED(mp
, ip
));
1080 blks
= XFS_ADDAFORK_SPACE_RES(mp
);
1082 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_addafork
, blks
, 0,
1083 rsvd
? XFS_TRANS_RESERVE
: 0, &tp
);
1087 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1088 error
= xfs_trans_reserve_quota_nblks(tp
, ip
, blks
, 0, rsvd
?
1089 XFS_QMOPT_RES_REGBLKS
| XFS_QMOPT_FORCE_RES
:
1090 XFS_QMOPT_RES_REGBLKS
);
1093 if (XFS_IFORK_Q(ip
))
1096 xfs_trans_ijoin(tp
, ip
, 0);
1097 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1098 error
= xfs_bmap_set_attrforkoff(ip
, size
, &version
);
1101 ASSERT(ip
->i_afp
== NULL
);
1103 ip
->i_afp
= kmem_cache_zalloc(xfs_ifork_zone
,
1104 GFP_KERNEL
| __GFP_NOFAIL
);
1106 ip
->i_afp
->if_format
= XFS_DINODE_FMT_EXTENTS
;
1107 ip
->i_afp
->if_flags
= XFS_IFEXTENTS
;
1109 switch (ip
->i_df
.if_format
) {
1110 case XFS_DINODE_FMT_LOCAL
:
1111 error
= xfs_bmap_add_attrfork_local(tp
, ip
, &logflags
);
1113 case XFS_DINODE_FMT_EXTENTS
:
1114 error
= xfs_bmap_add_attrfork_extents(tp
, ip
, &logflags
);
1116 case XFS_DINODE_FMT_BTREE
:
1117 error
= xfs_bmap_add_attrfork_btree(tp
, ip
, &logflags
);
1124 xfs_trans_log_inode(tp
, ip
, logflags
);
1127 if (!xfs_sb_version_hasattr(&mp
->m_sb
) ||
1128 (!xfs_sb_version_hasattr2(&mp
->m_sb
) && version
== 2)) {
1129 bool log_sb
= false;
1131 spin_lock(&mp
->m_sb_lock
);
1132 if (!xfs_sb_version_hasattr(&mp
->m_sb
)) {
1133 xfs_sb_version_addattr(&mp
->m_sb
);
1136 if (!xfs_sb_version_hasattr2(&mp
->m_sb
) && version
== 2) {
1137 xfs_sb_version_addattr2(&mp
->m_sb
);
1140 spin_unlock(&mp
->m_sb_lock
);
1145 error
= xfs_trans_commit(tp
);
1146 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1150 xfs_trans_cancel(tp
);
1151 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1156 * Internal and external extent tree search functions.
1159 struct xfs_iread_state
{
1160 struct xfs_iext_cursor icur
;
1161 xfs_extnum_t loaded
;
1164 /* Stuff every bmbt record from this block into the incore extent map. */
1166 xfs_iread_bmbt_block(
1167 struct xfs_btree_cur
*cur
,
1171 struct xfs_iread_state
*ir
= priv
;
1172 struct xfs_mount
*mp
= cur
->bc_mp
;
1173 struct xfs_inode
*ip
= cur
->bc_ino
.ip
;
1174 struct xfs_btree_block
*block
;
1176 struct xfs_bmbt_rec
*frp
;
1177 xfs_extnum_t num_recs
;
1179 int whichfork
= cur
->bc_ino
.whichfork
;
1180 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1182 block
= xfs_btree_get_block(cur
, level
, &bp
);
1184 /* Abort if we find more records than nextents. */
1185 num_recs
= xfs_btree_get_numrecs(block
);
1186 if (unlikely(ir
->loaded
+ num_recs
> ifp
->if_nextents
)) {
1187 xfs_warn(ip
->i_mount
, "corrupt dinode %llu, (btree extents).",
1188 (unsigned long long)ip
->i_ino
);
1189 xfs_inode_verifier_error(ip
, -EFSCORRUPTED
, __func__
, block
,
1190 sizeof(*block
), __this_address
);
1191 return -EFSCORRUPTED
;
1194 /* Copy records into the incore cache. */
1195 frp
= XFS_BMBT_REC_ADDR(mp
, block
, 1);
1196 for (j
= 0; j
< num_recs
; j
++, frp
++, ir
->loaded
++) {
1197 struct xfs_bmbt_irec
new;
1200 xfs_bmbt_disk_get_all(frp
, &new);
1201 fa
= xfs_bmap_validate_extent(ip
, whichfork
, &new);
1203 xfs_inode_verifier_error(ip
, -EFSCORRUPTED
,
1204 "xfs_iread_extents(2)", frp
,
1206 return -EFSCORRUPTED
;
1208 xfs_iext_insert(ip
, &ir
->icur
, &new,
1209 xfs_bmap_fork_to_state(whichfork
));
1210 trace_xfs_read_extent(ip
, &ir
->icur
,
1211 xfs_bmap_fork_to_state(whichfork
), _THIS_IP_
);
1212 xfs_iext_next(ifp
, &ir
->icur
);
1219 * Read in extents from a btree-format inode.
1223 struct xfs_trans
*tp
,
1224 struct xfs_inode
*ip
,
1227 struct xfs_iread_state ir
;
1228 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1229 struct xfs_mount
*mp
= ip
->i_mount
;
1230 struct xfs_btree_cur
*cur
;
1233 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
1235 if (XFS_IS_CORRUPT(mp
, ifp
->if_format
!= XFS_DINODE_FMT_BTREE
)) {
1236 error
= -EFSCORRUPTED
;
1241 xfs_iext_first(ifp
, &ir
.icur
);
1242 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
1243 error
= xfs_btree_visit_blocks(cur
, xfs_iread_bmbt_block
,
1244 XFS_BTREE_VISIT_RECORDS
, &ir
);
1245 xfs_btree_del_cursor(cur
, error
);
1249 if (XFS_IS_CORRUPT(mp
, ir
.loaded
!= ifp
->if_nextents
)) {
1250 error
= -EFSCORRUPTED
;
1253 ASSERT(ir
.loaded
== xfs_iext_count(ifp
));
1255 ifp
->if_flags
|= XFS_IFEXTENTS
;
1258 xfs_iext_destroy(ifp
);
1263 * Returns the relative block number of the first unused block(s) in the given
1264 * fork with at least "len" logically contiguous blocks free. This is the
1265 * lowest-address hole if the fork has holes, else the first block past the end
1266 * of fork. Return 0 if the fork is currently local (in-inode).
1269 xfs_bmap_first_unused(
1270 struct xfs_trans
*tp
, /* transaction pointer */
1271 struct xfs_inode
*ip
, /* incore inode */
1272 xfs_extlen_t len
, /* size of hole to find */
1273 xfs_fileoff_t
*first_unused
, /* unused block */
1274 int whichfork
) /* data or attr fork */
1276 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1277 struct xfs_bmbt_irec got
;
1278 struct xfs_iext_cursor icur
;
1279 xfs_fileoff_t lastaddr
= 0;
1280 xfs_fileoff_t lowest
, max
;
1283 if (ifp
->if_format
== XFS_DINODE_FMT_LOCAL
) {
1288 ASSERT(xfs_ifork_has_extents(ifp
));
1290 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
1291 error
= xfs_iread_extents(tp
, ip
, whichfork
);
1296 lowest
= max
= *first_unused
;
1297 for_each_xfs_iext(ifp
, &icur
, &got
) {
1299 * See if the hole before this extent will work.
1301 if (got
.br_startoff
>= lowest
+ len
&&
1302 got
.br_startoff
- max
>= len
)
1304 lastaddr
= got
.br_startoff
+ got
.br_blockcount
;
1305 max
= XFS_FILEOFF_MAX(lastaddr
, lowest
);
1308 *first_unused
= max
;
1313 * Returns the file-relative block number of the last block - 1 before
1314 * last_block (input value) in the file.
1315 * This is not based on i_size, it is based on the extent records.
1316 * Returns 0 for local files, as they do not have extent records.
1319 xfs_bmap_last_before(
1320 struct xfs_trans
*tp
, /* transaction pointer */
1321 struct xfs_inode
*ip
, /* incore inode */
1322 xfs_fileoff_t
*last_block
, /* last block */
1323 int whichfork
) /* data or attr fork */
1325 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1326 struct xfs_bmbt_irec got
;
1327 struct xfs_iext_cursor icur
;
1330 switch (ifp
->if_format
) {
1331 case XFS_DINODE_FMT_LOCAL
:
1334 case XFS_DINODE_FMT_BTREE
:
1335 case XFS_DINODE_FMT_EXTENTS
:
1339 return -EFSCORRUPTED
;
1342 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
1343 error
= xfs_iread_extents(tp
, ip
, whichfork
);
1348 if (!xfs_iext_lookup_extent_before(ip
, ifp
, last_block
, &icur
, &got
))
1354 xfs_bmap_last_extent(
1355 struct xfs_trans
*tp
,
1356 struct xfs_inode
*ip
,
1358 struct xfs_bmbt_irec
*rec
,
1361 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1362 struct xfs_iext_cursor icur
;
1365 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
1366 error
= xfs_iread_extents(tp
, ip
, whichfork
);
1371 xfs_iext_last(ifp
, &icur
);
1372 if (!xfs_iext_get_extent(ifp
, &icur
, rec
))
1380 * Check the last inode extent to determine whether this allocation will result
1381 * in blocks being allocated at the end of the file. When we allocate new data
1382 * blocks at the end of the file which do not start at the previous data block,
1383 * we will try to align the new blocks at stripe unit boundaries.
1385 * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1386 * at, or past the EOF.
1390 struct xfs_bmalloca
*bma
,
1393 struct xfs_bmbt_irec rec
;
1398 error
= xfs_bmap_last_extent(NULL
, bma
->ip
, whichfork
, &rec
,
1409 * Check if we are allocation or past the last extent, or at least into
1410 * the last delayed allocated extent.
1412 bma
->aeof
= bma
->offset
>= rec
.br_startoff
+ rec
.br_blockcount
||
1413 (bma
->offset
>= rec
.br_startoff
&&
1414 isnullstartblock(rec
.br_startblock
));
1419 * Returns the file-relative block number of the first block past eof in
1420 * the file. This is not based on i_size, it is based on the extent records.
1421 * Returns 0 for local files, as they do not have extent records.
1424 xfs_bmap_last_offset(
1425 struct xfs_inode
*ip
,
1426 xfs_fileoff_t
*last_block
,
1429 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1430 struct xfs_bmbt_irec rec
;
1436 if (ifp
->if_format
== XFS_DINODE_FMT_LOCAL
)
1439 if (XFS_IS_CORRUPT(ip
->i_mount
, !xfs_ifork_has_extents(ifp
)))
1440 return -EFSCORRUPTED
;
1442 error
= xfs_bmap_last_extent(NULL
, ip
, whichfork
, &rec
, &is_empty
);
1443 if (error
|| is_empty
)
1446 *last_block
= rec
.br_startoff
+ rec
.br_blockcount
;
1451 * Returns whether the selected fork of the inode has exactly one
1452 * block or not. For the data fork we check this matches di_size,
1453 * implying the file's range is 0..bsize-1.
1455 int /* 1=>1 block, 0=>otherwise */
1457 struct xfs_inode
*ip
, /* incore inode */
1458 int whichfork
) /* data or attr fork */
1460 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1461 int rval
; /* return value */
1462 struct xfs_bmbt_irec s
; /* internal version of extent */
1463 struct xfs_iext_cursor icur
;
1466 if (whichfork
== XFS_DATA_FORK
)
1467 return XFS_ISIZE(ip
) == ip
->i_mount
->m_sb
.sb_blocksize
;
1469 if (ifp
->if_nextents
!= 1)
1471 if (ifp
->if_format
!= XFS_DINODE_FMT_EXTENTS
)
1473 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
1474 xfs_iext_first(ifp
, &icur
);
1475 xfs_iext_get_extent(ifp
, &icur
, &s
);
1476 rval
= s
.br_startoff
== 0 && s
.br_blockcount
== 1;
1477 if (rval
&& whichfork
== XFS_DATA_FORK
)
1478 ASSERT(XFS_ISIZE(ip
) == ip
->i_mount
->m_sb
.sb_blocksize
);
1483 * Extent tree manipulation functions used during allocation.
1487 * Convert a delayed allocation to a real allocation.
1489 STATIC
int /* error */
1490 xfs_bmap_add_extent_delay_real(
1491 struct xfs_bmalloca
*bma
,
1494 struct xfs_mount
*mp
= bma
->ip
->i_mount
;
1495 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(bma
->ip
, whichfork
);
1496 struct xfs_bmbt_irec
*new = &bma
->got
;
1497 int error
; /* error return value */
1498 int i
; /* temp state */
1499 xfs_fileoff_t new_endoff
; /* end offset of new entry */
1500 xfs_bmbt_irec_t r
[3]; /* neighbor extent entries */
1501 /* left is 0, right is 1, prev is 2 */
1502 int rval
=0; /* return value (logging flags) */
1503 int state
= xfs_bmap_fork_to_state(whichfork
);
1504 xfs_filblks_t da_new
; /* new count del alloc blocks used */
1505 xfs_filblks_t da_old
; /* old count del alloc blocks used */
1506 xfs_filblks_t temp
=0; /* value for da_new calculations */
1507 int tmp_rval
; /* partial logging flags */
1508 struct xfs_bmbt_irec old
;
1510 ASSERT(whichfork
!= XFS_ATTR_FORK
);
1511 ASSERT(!isnullstartblock(new->br_startblock
));
1513 (bma
->cur
->bc_ino
.flags
& XFS_BTCUR_BMBT_WASDEL
));
1515 XFS_STATS_INC(mp
, xs_add_exlist
);
1522 * Set up a bunch of variables to make the tests simpler.
1524 xfs_iext_get_extent(ifp
, &bma
->icur
, &PREV
);
1525 new_endoff
= new->br_startoff
+ new->br_blockcount
;
1526 ASSERT(isnullstartblock(PREV
.br_startblock
));
1527 ASSERT(PREV
.br_startoff
<= new->br_startoff
);
1528 ASSERT(PREV
.br_startoff
+ PREV
.br_blockcount
>= new_endoff
);
1530 da_old
= startblockval(PREV
.br_startblock
);
1534 * Set flags determining what part of the previous delayed allocation
1535 * extent is being replaced by a real allocation.
1537 if (PREV
.br_startoff
== new->br_startoff
)
1538 state
|= BMAP_LEFT_FILLING
;
1539 if (PREV
.br_startoff
+ PREV
.br_blockcount
== new_endoff
)
1540 state
|= BMAP_RIGHT_FILLING
;
1543 * Check and set flags if this segment has a left neighbor.
1544 * Don't set contiguous if the combined extent would be too large.
1546 if (xfs_iext_peek_prev_extent(ifp
, &bma
->icur
, &LEFT
)) {
1547 state
|= BMAP_LEFT_VALID
;
1548 if (isnullstartblock(LEFT
.br_startblock
))
1549 state
|= BMAP_LEFT_DELAY
;
1552 if ((state
& BMAP_LEFT_VALID
) && !(state
& BMAP_LEFT_DELAY
) &&
1553 LEFT
.br_startoff
+ LEFT
.br_blockcount
== new->br_startoff
&&
1554 LEFT
.br_startblock
+ LEFT
.br_blockcount
== new->br_startblock
&&
1555 LEFT
.br_state
== new->br_state
&&
1556 LEFT
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
1557 state
|= BMAP_LEFT_CONTIG
;
1560 * Check and set flags if this segment has a right neighbor.
1561 * Don't set contiguous if the combined extent would be too large.
1562 * Also check for all-three-contiguous being too large.
1564 if (xfs_iext_peek_next_extent(ifp
, &bma
->icur
, &RIGHT
)) {
1565 state
|= BMAP_RIGHT_VALID
;
1566 if (isnullstartblock(RIGHT
.br_startblock
))
1567 state
|= BMAP_RIGHT_DELAY
;
1570 if ((state
& BMAP_RIGHT_VALID
) && !(state
& BMAP_RIGHT_DELAY
) &&
1571 new_endoff
== RIGHT
.br_startoff
&&
1572 new->br_startblock
+ new->br_blockcount
== RIGHT
.br_startblock
&&
1573 new->br_state
== RIGHT
.br_state
&&
1574 new->br_blockcount
+ RIGHT
.br_blockcount
<= MAXEXTLEN
&&
1575 ((state
& (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
1576 BMAP_RIGHT_FILLING
)) !=
1577 (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
1578 BMAP_RIGHT_FILLING
) ||
1579 LEFT
.br_blockcount
+ new->br_blockcount
+ RIGHT
.br_blockcount
1581 state
|= BMAP_RIGHT_CONTIG
;
1585 * Switch out based on the FILLING and CONTIG state bits.
1587 switch (state
& (BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
1588 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
)) {
1589 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
1590 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
1592 * Filling in all of a previously delayed allocation extent.
1593 * The left and right neighbors are both contiguous with new.
1595 LEFT
.br_blockcount
+= PREV
.br_blockcount
+ RIGHT
.br_blockcount
;
1597 xfs_iext_remove(bma
->ip
, &bma
->icur
, state
);
1598 xfs_iext_remove(bma
->ip
, &bma
->icur
, state
);
1599 xfs_iext_prev(ifp
, &bma
->icur
);
1600 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, &LEFT
);
1603 if (bma
->cur
== NULL
)
1604 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1606 rval
= XFS_ILOG_CORE
;
1607 error
= xfs_bmbt_lookup_eq(bma
->cur
, &RIGHT
, &i
);
1610 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
1611 error
= -EFSCORRUPTED
;
1614 error
= xfs_btree_delete(bma
->cur
, &i
);
1617 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
1618 error
= -EFSCORRUPTED
;
1621 error
= xfs_btree_decrement(bma
->cur
, 0, &i
);
1624 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
1625 error
= -EFSCORRUPTED
;
1628 error
= xfs_bmbt_update(bma
->cur
, &LEFT
);
1634 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
1636 * Filling in all of a previously delayed allocation extent.
1637 * The left neighbor is contiguous, the right is not.
1640 LEFT
.br_blockcount
+= PREV
.br_blockcount
;
1642 xfs_iext_remove(bma
->ip
, &bma
->icur
, state
);
1643 xfs_iext_prev(ifp
, &bma
->icur
);
1644 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, &LEFT
);
1646 if (bma
->cur
== NULL
)
1647 rval
= XFS_ILOG_DEXT
;
1650 error
= xfs_bmbt_lookup_eq(bma
->cur
, &old
, &i
);
1653 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
1654 error
= -EFSCORRUPTED
;
1657 error
= xfs_bmbt_update(bma
->cur
, &LEFT
);
1663 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
1665 * Filling in all of a previously delayed allocation extent.
1666 * The right neighbor is contiguous, the left is not. Take care
1667 * with delay -> unwritten extent allocation here because the
1668 * delalloc record we are overwriting is always written.
1670 PREV
.br_startblock
= new->br_startblock
;
1671 PREV
.br_blockcount
+= RIGHT
.br_blockcount
;
1672 PREV
.br_state
= new->br_state
;
1674 xfs_iext_next(ifp
, &bma
->icur
);
1675 xfs_iext_remove(bma
->ip
, &bma
->icur
, state
);
1676 xfs_iext_prev(ifp
, &bma
->icur
);
1677 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, &PREV
);
1679 if (bma
->cur
== NULL
)
1680 rval
= XFS_ILOG_DEXT
;
1683 error
= xfs_bmbt_lookup_eq(bma
->cur
, &RIGHT
, &i
);
1686 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
1687 error
= -EFSCORRUPTED
;
1690 error
= xfs_bmbt_update(bma
->cur
, &PREV
);
1696 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
:
1698 * Filling in all of a previously delayed allocation extent.
1699 * Neither the left nor right neighbors are contiguous with
1702 PREV
.br_startblock
= new->br_startblock
;
1703 PREV
.br_state
= new->br_state
;
1704 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, &PREV
);
1707 if (bma
->cur
== NULL
)
1708 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1710 rval
= XFS_ILOG_CORE
;
1711 error
= xfs_bmbt_lookup_eq(bma
->cur
, new, &i
);
1714 if (XFS_IS_CORRUPT(mp
, i
!= 0)) {
1715 error
= -EFSCORRUPTED
;
1718 error
= xfs_btree_insert(bma
->cur
, &i
);
1721 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
1722 error
= -EFSCORRUPTED
;
1728 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
:
1730 * Filling in the first part of a previous delayed allocation.
1731 * The left neighbor is contiguous.
1734 temp
= PREV
.br_blockcount
- new->br_blockcount
;
1735 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma
->ip
, temp
),
1736 startblockval(PREV
.br_startblock
));
1738 LEFT
.br_blockcount
+= new->br_blockcount
;
1740 PREV
.br_blockcount
= temp
;
1741 PREV
.br_startoff
+= new->br_blockcount
;
1742 PREV
.br_startblock
= nullstartblock(da_new
);
1744 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, &PREV
);
1745 xfs_iext_prev(ifp
, &bma
->icur
);
1746 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, &LEFT
);
1748 if (bma
->cur
== NULL
)
1749 rval
= XFS_ILOG_DEXT
;
1752 error
= xfs_bmbt_lookup_eq(bma
->cur
, &old
, &i
);
1755 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
1756 error
= -EFSCORRUPTED
;
1759 error
= xfs_bmbt_update(bma
->cur
, &LEFT
);
1765 case BMAP_LEFT_FILLING
:
1767 * Filling in the first part of a previous delayed allocation.
1768 * The left neighbor is not contiguous.
1770 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, new);
1773 if (bma
->cur
== NULL
)
1774 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1776 rval
= XFS_ILOG_CORE
;
1777 error
= xfs_bmbt_lookup_eq(bma
->cur
, new, &i
);
1780 if (XFS_IS_CORRUPT(mp
, i
!= 0)) {
1781 error
= -EFSCORRUPTED
;
1784 error
= xfs_btree_insert(bma
->cur
, &i
);
1787 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
1788 error
= -EFSCORRUPTED
;
1793 if (xfs_bmap_needs_btree(bma
->ip
, whichfork
)) {
1794 error
= xfs_bmap_extents_to_btree(bma
->tp
, bma
->ip
,
1795 &bma
->cur
, 1, &tmp_rval
, whichfork
);
1801 temp
= PREV
.br_blockcount
- new->br_blockcount
;
1802 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma
->ip
, temp
),
1803 startblockval(PREV
.br_startblock
) -
1804 (bma
->cur
? bma
->cur
->bc_ino
.allocated
: 0));
1806 PREV
.br_startoff
= new_endoff
;
1807 PREV
.br_blockcount
= temp
;
1808 PREV
.br_startblock
= nullstartblock(da_new
);
1809 xfs_iext_next(ifp
, &bma
->icur
);
1810 xfs_iext_insert(bma
->ip
, &bma
->icur
, &PREV
, state
);
1811 xfs_iext_prev(ifp
, &bma
->icur
);
1814 case BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
1816 * Filling in the last part of a previous delayed allocation.
1817 * The right neighbor is contiguous with the new allocation.
1820 RIGHT
.br_startoff
= new->br_startoff
;
1821 RIGHT
.br_startblock
= new->br_startblock
;
1822 RIGHT
.br_blockcount
+= new->br_blockcount
;
1824 if (bma
->cur
== NULL
)
1825 rval
= XFS_ILOG_DEXT
;
1828 error
= xfs_bmbt_lookup_eq(bma
->cur
, &old
, &i
);
1831 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
1832 error
= -EFSCORRUPTED
;
1835 error
= xfs_bmbt_update(bma
->cur
, &RIGHT
);
1840 temp
= PREV
.br_blockcount
- new->br_blockcount
;
1841 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma
->ip
, temp
),
1842 startblockval(PREV
.br_startblock
));
1844 PREV
.br_blockcount
= temp
;
1845 PREV
.br_startblock
= nullstartblock(da_new
);
1847 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, &PREV
);
1848 xfs_iext_next(ifp
, &bma
->icur
);
1849 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, &RIGHT
);
1852 case BMAP_RIGHT_FILLING
:
1854 * Filling in the last part of a previous delayed allocation.
1855 * The right neighbor is not contiguous.
1857 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, new);
1860 if (bma
->cur
== NULL
)
1861 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1863 rval
= XFS_ILOG_CORE
;
1864 error
= xfs_bmbt_lookup_eq(bma
->cur
, new, &i
);
1867 if (XFS_IS_CORRUPT(mp
, i
!= 0)) {
1868 error
= -EFSCORRUPTED
;
1871 error
= xfs_btree_insert(bma
->cur
, &i
);
1874 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
1875 error
= -EFSCORRUPTED
;
1880 if (xfs_bmap_needs_btree(bma
->ip
, whichfork
)) {
1881 error
= xfs_bmap_extents_to_btree(bma
->tp
, bma
->ip
,
1882 &bma
->cur
, 1, &tmp_rval
, whichfork
);
1888 temp
= PREV
.br_blockcount
- new->br_blockcount
;
1889 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma
->ip
, temp
),
1890 startblockval(PREV
.br_startblock
) -
1891 (bma
->cur
? bma
->cur
->bc_ino
.allocated
: 0));
1893 PREV
.br_startblock
= nullstartblock(da_new
);
1894 PREV
.br_blockcount
= temp
;
1895 xfs_iext_insert(bma
->ip
, &bma
->icur
, &PREV
, state
);
1896 xfs_iext_next(ifp
, &bma
->icur
);
1901 * Filling in the middle part of a previous delayed allocation.
1902 * Contiguity is impossible here.
1903 * This case is avoided almost all the time.
1905 * We start with a delayed allocation:
1907 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
1910 * and we are allocating:
1911 * +rrrrrrrrrrrrrrrrr+
1914 * and we set it up for insertion as:
1915 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
1917 * PREV @ idx LEFT RIGHT
1918 * inserted at idx + 1
1922 /* LEFT is the new middle */
1925 /* RIGHT is the new right */
1926 RIGHT
.br_state
= PREV
.br_state
;
1927 RIGHT
.br_startoff
= new_endoff
;
1928 RIGHT
.br_blockcount
=
1929 PREV
.br_startoff
+ PREV
.br_blockcount
- new_endoff
;
1930 RIGHT
.br_startblock
=
1931 nullstartblock(xfs_bmap_worst_indlen(bma
->ip
,
1932 RIGHT
.br_blockcount
));
1935 PREV
.br_blockcount
= new->br_startoff
- PREV
.br_startoff
;
1936 PREV
.br_startblock
=
1937 nullstartblock(xfs_bmap_worst_indlen(bma
->ip
,
1938 PREV
.br_blockcount
));
1939 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, &PREV
);
1941 xfs_iext_next(ifp
, &bma
->icur
);
1942 xfs_iext_insert(bma
->ip
, &bma
->icur
, &RIGHT
, state
);
1943 xfs_iext_insert(bma
->ip
, &bma
->icur
, &LEFT
, state
);
1946 if (bma
->cur
== NULL
)
1947 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1949 rval
= XFS_ILOG_CORE
;
1950 error
= xfs_bmbt_lookup_eq(bma
->cur
, new, &i
);
1953 if (XFS_IS_CORRUPT(mp
, i
!= 0)) {
1954 error
= -EFSCORRUPTED
;
1957 error
= xfs_btree_insert(bma
->cur
, &i
);
1960 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
1961 error
= -EFSCORRUPTED
;
1966 if (xfs_bmap_needs_btree(bma
->ip
, whichfork
)) {
1967 error
= xfs_bmap_extents_to_btree(bma
->tp
, bma
->ip
,
1968 &bma
->cur
, 1, &tmp_rval
, whichfork
);
1974 da_new
= startblockval(PREV
.br_startblock
) +
1975 startblockval(RIGHT
.br_startblock
);
1978 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
1979 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
1980 case BMAP_LEFT_FILLING
| BMAP_RIGHT_CONTIG
:
1981 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
1982 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
1983 case BMAP_LEFT_CONTIG
:
1984 case BMAP_RIGHT_CONTIG
:
1986 * These cases are all impossible.
1991 /* add reverse mapping unless caller opted out */
1992 if (!(bma
->flags
& XFS_BMAPI_NORMAP
))
1993 xfs_rmap_map_extent(bma
->tp
, bma
->ip
, whichfork
, new);
1995 /* convert to a btree if necessary */
1996 if (xfs_bmap_needs_btree(bma
->ip
, whichfork
)) {
1997 int tmp_logflags
; /* partial log flag return val */
1999 ASSERT(bma
->cur
== NULL
);
2000 error
= xfs_bmap_extents_to_btree(bma
->tp
, bma
->ip
,
2001 &bma
->cur
, da_old
> 0, &tmp_logflags
,
2003 bma
->logflags
|= tmp_logflags
;
2008 if (da_new
!= da_old
)
2009 xfs_mod_delalloc(mp
, (int64_t)da_new
- da_old
);
2012 da_new
+= bma
->cur
->bc_ino
.allocated
;
2013 bma
->cur
->bc_ino
.allocated
= 0;
2016 /* adjust for changes in reserved delayed indirect blocks */
2017 if (da_new
!= da_old
) {
2018 ASSERT(state
== 0 || da_new
< da_old
);
2019 error
= xfs_mod_fdblocks(mp
, (int64_t)(da_old
- da_new
),
2023 xfs_bmap_check_leaf_extents(bma
->cur
, bma
->ip
, whichfork
);
2025 if (whichfork
!= XFS_COW_FORK
)
2026 bma
->logflags
|= rval
;
2034 * Convert an unwritten allocation to a real allocation or vice versa.
2037 xfs_bmap_add_extent_unwritten_real(
2038 struct xfs_trans
*tp
,
2039 xfs_inode_t
*ip
, /* incore inode pointer */
2041 struct xfs_iext_cursor
*icur
,
2042 xfs_btree_cur_t
**curp
, /* if *curp is null, not a btree */
2043 xfs_bmbt_irec_t
*new, /* new data to add to file extents */
2044 int *logflagsp
) /* inode logging flags */
2046 xfs_btree_cur_t
*cur
; /* btree cursor */
2047 int error
; /* error return value */
2048 int i
; /* temp state */
2049 struct xfs_ifork
*ifp
; /* inode fork pointer */
2050 xfs_fileoff_t new_endoff
; /* end offset of new entry */
2051 xfs_bmbt_irec_t r
[3]; /* neighbor extent entries */
2052 /* left is 0, right is 1, prev is 2 */
2053 int rval
=0; /* return value (logging flags) */
2054 int state
= xfs_bmap_fork_to_state(whichfork
);
2055 struct xfs_mount
*mp
= ip
->i_mount
;
2056 struct xfs_bmbt_irec old
;
2061 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2063 ASSERT(!isnullstartblock(new->br_startblock
));
2065 XFS_STATS_INC(mp
, xs_add_exlist
);
2072 * Set up a bunch of variables to make the tests simpler.
2075 xfs_iext_get_extent(ifp
, icur
, &PREV
);
2076 ASSERT(new->br_state
!= PREV
.br_state
);
2077 new_endoff
= new->br_startoff
+ new->br_blockcount
;
2078 ASSERT(PREV
.br_startoff
<= new->br_startoff
);
2079 ASSERT(PREV
.br_startoff
+ PREV
.br_blockcount
>= new_endoff
);
2082 * Set flags determining what part of the previous oldext allocation
2083 * extent is being replaced by a newext allocation.
2085 if (PREV
.br_startoff
== new->br_startoff
)
2086 state
|= BMAP_LEFT_FILLING
;
2087 if (PREV
.br_startoff
+ PREV
.br_blockcount
== new_endoff
)
2088 state
|= BMAP_RIGHT_FILLING
;
2091 * Check and set flags if this segment has a left neighbor.
2092 * Don't set contiguous if the combined extent would be too large.
2094 if (xfs_iext_peek_prev_extent(ifp
, icur
, &LEFT
)) {
2095 state
|= BMAP_LEFT_VALID
;
2096 if (isnullstartblock(LEFT
.br_startblock
))
2097 state
|= BMAP_LEFT_DELAY
;
2100 if ((state
& BMAP_LEFT_VALID
) && !(state
& BMAP_LEFT_DELAY
) &&
2101 LEFT
.br_startoff
+ LEFT
.br_blockcount
== new->br_startoff
&&
2102 LEFT
.br_startblock
+ LEFT
.br_blockcount
== new->br_startblock
&&
2103 LEFT
.br_state
== new->br_state
&&
2104 LEFT
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
2105 state
|= BMAP_LEFT_CONTIG
;
2108 * Check and set flags if this segment has a right neighbor.
2109 * Don't set contiguous if the combined extent would be too large.
2110 * Also check for all-three-contiguous being too large.
2112 if (xfs_iext_peek_next_extent(ifp
, icur
, &RIGHT
)) {
2113 state
|= BMAP_RIGHT_VALID
;
2114 if (isnullstartblock(RIGHT
.br_startblock
))
2115 state
|= BMAP_RIGHT_DELAY
;
2118 if ((state
& BMAP_RIGHT_VALID
) && !(state
& BMAP_RIGHT_DELAY
) &&
2119 new_endoff
== RIGHT
.br_startoff
&&
2120 new->br_startblock
+ new->br_blockcount
== RIGHT
.br_startblock
&&
2121 new->br_state
== RIGHT
.br_state
&&
2122 new->br_blockcount
+ RIGHT
.br_blockcount
<= MAXEXTLEN
&&
2123 ((state
& (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
2124 BMAP_RIGHT_FILLING
)) !=
2125 (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
2126 BMAP_RIGHT_FILLING
) ||
2127 LEFT
.br_blockcount
+ new->br_blockcount
+ RIGHT
.br_blockcount
2129 state
|= BMAP_RIGHT_CONTIG
;
2132 * Switch out based on the FILLING and CONTIG state bits.
2134 switch (state
& (BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
2135 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
)) {
2136 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
2137 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
2139 * Setting all of a previous oldext extent to newext.
2140 * The left and right neighbors are both contiguous with new.
2142 LEFT
.br_blockcount
+= PREV
.br_blockcount
+ RIGHT
.br_blockcount
;
2144 xfs_iext_remove(ip
, icur
, state
);
2145 xfs_iext_remove(ip
, icur
, state
);
2146 xfs_iext_prev(ifp
, icur
);
2147 xfs_iext_update_extent(ip
, state
, icur
, &LEFT
);
2148 ifp
->if_nextents
-= 2;
2150 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2152 rval
= XFS_ILOG_CORE
;
2153 error
= xfs_bmbt_lookup_eq(cur
, &RIGHT
, &i
);
2156 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2157 error
= -EFSCORRUPTED
;
2160 if ((error
= xfs_btree_delete(cur
, &i
)))
2162 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2163 error
= -EFSCORRUPTED
;
2166 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
2168 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2169 error
= -EFSCORRUPTED
;
2172 if ((error
= xfs_btree_delete(cur
, &i
)))
2174 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2175 error
= -EFSCORRUPTED
;
2178 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
2180 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2181 error
= -EFSCORRUPTED
;
2184 error
= xfs_bmbt_update(cur
, &LEFT
);
2190 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
2192 * Setting all of a previous oldext extent to newext.
2193 * The left neighbor is contiguous, the right is not.
2195 LEFT
.br_blockcount
+= PREV
.br_blockcount
;
2197 xfs_iext_remove(ip
, icur
, state
);
2198 xfs_iext_prev(ifp
, icur
);
2199 xfs_iext_update_extent(ip
, state
, icur
, &LEFT
);
2202 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2204 rval
= XFS_ILOG_CORE
;
2205 error
= xfs_bmbt_lookup_eq(cur
, &PREV
, &i
);
2208 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2209 error
= -EFSCORRUPTED
;
2212 if ((error
= xfs_btree_delete(cur
, &i
)))
2214 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2215 error
= -EFSCORRUPTED
;
2218 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
2220 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2221 error
= -EFSCORRUPTED
;
2224 error
= xfs_bmbt_update(cur
, &LEFT
);
2230 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
2232 * Setting all of a previous oldext extent to newext.
2233 * The right neighbor is contiguous, the left is not.
2235 PREV
.br_blockcount
+= RIGHT
.br_blockcount
;
2236 PREV
.br_state
= new->br_state
;
2238 xfs_iext_next(ifp
, icur
);
2239 xfs_iext_remove(ip
, icur
, state
);
2240 xfs_iext_prev(ifp
, icur
);
2241 xfs_iext_update_extent(ip
, state
, icur
, &PREV
);
2245 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2247 rval
= XFS_ILOG_CORE
;
2248 error
= xfs_bmbt_lookup_eq(cur
, &RIGHT
, &i
);
2251 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2252 error
= -EFSCORRUPTED
;
2255 if ((error
= xfs_btree_delete(cur
, &i
)))
2257 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2258 error
= -EFSCORRUPTED
;
2261 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
2263 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2264 error
= -EFSCORRUPTED
;
2267 error
= xfs_bmbt_update(cur
, &PREV
);
2273 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
:
2275 * Setting all of a previous oldext extent to newext.
2276 * Neither the left nor right neighbors are contiguous with
2279 PREV
.br_state
= new->br_state
;
2280 xfs_iext_update_extent(ip
, state
, icur
, &PREV
);
2283 rval
= XFS_ILOG_DEXT
;
2286 error
= xfs_bmbt_lookup_eq(cur
, new, &i
);
2289 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2290 error
= -EFSCORRUPTED
;
2293 error
= xfs_bmbt_update(cur
, &PREV
);
2299 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
:
2301 * Setting the first part of a previous oldext extent to newext.
2302 * The left neighbor is contiguous.
2304 LEFT
.br_blockcount
+= new->br_blockcount
;
2307 PREV
.br_startoff
+= new->br_blockcount
;
2308 PREV
.br_startblock
+= new->br_blockcount
;
2309 PREV
.br_blockcount
-= new->br_blockcount
;
2311 xfs_iext_update_extent(ip
, state
, icur
, &PREV
);
2312 xfs_iext_prev(ifp
, icur
);
2313 xfs_iext_update_extent(ip
, state
, icur
, &LEFT
);
2316 rval
= XFS_ILOG_DEXT
;
2319 error
= xfs_bmbt_lookup_eq(cur
, &old
, &i
);
2322 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2323 error
= -EFSCORRUPTED
;
2326 error
= xfs_bmbt_update(cur
, &PREV
);
2329 error
= xfs_btree_decrement(cur
, 0, &i
);
2332 error
= xfs_bmbt_update(cur
, &LEFT
);
2338 case BMAP_LEFT_FILLING
:
2340 * Setting the first part of a previous oldext extent to newext.
2341 * The left neighbor is not contiguous.
2344 PREV
.br_startoff
+= new->br_blockcount
;
2345 PREV
.br_startblock
+= new->br_blockcount
;
2346 PREV
.br_blockcount
-= new->br_blockcount
;
2348 xfs_iext_update_extent(ip
, state
, icur
, &PREV
);
2349 xfs_iext_insert(ip
, icur
, new, state
);
2353 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2355 rval
= XFS_ILOG_CORE
;
2356 error
= xfs_bmbt_lookup_eq(cur
, &old
, &i
);
2359 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2360 error
= -EFSCORRUPTED
;
2363 error
= xfs_bmbt_update(cur
, &PREV
);
2366 cur
->bc_rec
.b
= *new;
2367 if ((error
= xfs_btree_insert(cur
, &i
)))
2369 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2370 error
= -EFSCORRUPTED
;
2376 case BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
2378 * Setting the last part of a previous oldext extent to newext.
2379 * The right neighbor is contiguous with the new allocation.
2382 PREV
.br_blockcount
-= new->br_blockcount
;
2384 RIGHT
.br_startoff
= new->br_startoff
;
2385 RIGHT
.br_startblock
= new->br_startblock
;
2386 RIGHT
.br_blockcount
+= new->br_blockcount
;
2388 xfs_iext_update_extent(ip
, state
, icur
, &PREV
);
2389 xfs_iext_next(ifp
, icur
);
2390 xfs_iext_update_extent(ip
, state
, icur
, &RIGHT
);
2393 rval
= XFS_ILOG_DEXT
;
2396 error
= xfs_bmbt_lookup_eq(cur
, &old
, &i
);
2399 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2400 error
= -EFSCORRUPTED
;
2403 error
= xfs_bmbt_update(cur
, &PREV
);
2406 error
= xfs_btree_increment(cur
, 0, &i
);
2409 error
= xfs_bmbt_update(cur
, &RIGHT
);
2415 case BMAP_RIGHT_FILLING
:
2417 * Setting the last part of a previous oldext extent to newext.
2418 * The right neighbor is not contiguous.
2421 PREV
.br_blockcount
-= new->br_blockcount
;
2423 xfs_iext_update_extent(ip
, state
, icur
, &PREV
);
2424 xfs_iext_next(ifp
, icur
);
2425 xfs_iext_insert(ip
, icur
, new, state
);
2429 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2431 rval
= XFS_ILOG_CORE
;
2432 error
= xfs_bmbt_lookup_eq(cur
, &old
, &i
);
2435 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2436 error
= -EFSCORRUPTED
;
2439 error
= xfs_bmbt_update(cur
, &PREV
);
2442 error
= xfs_bmbt_lookup_eq(cur
, new, &i
);
2445 if (XFS_IS_CORRUPT(mp
, i
!= 0)) {
2446 error
= -EFSCORRUPTED
;
2449 if ((error
= xfs_btree_insert(cur
, &i
)))
2451 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2452 error
= -EFSCORRUPTED
;
2460 * Setting the middle part of a previous oldext extent to
2461 * newext. Contiguity is impossible here.
2462 * One extent becomes three extents.
2465 PREV
.br_blockcount
= new->br_startoff
- PREV
.br_startoff
;
2468 r
[1].br_startoff
= new_endoff
;
2469 r
[1].br_blockcount
=
2470 old
.br_startoff
+ old
.br_blockcount
- new_endoff
;
2471 r
[1].br_startblock
= new->br_startblock
+ new->br_blockcount
;
2472 r
[1].br_state
= PREV
.br_state
;
2474 xfs_iext_update_extent(ip
, state
, icur
, &PREV
);
2475 xfs_iext_next(ifp
, icur
);
2476 xfs_iext_insert(ip
, icur
, &r
[1], state
);
2477 xfs_iext_insert(ip
, icur
, &r
[0], state
);
2478 ifp
->if_nextents
+= 2;
2481 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2483 rval
= XFS_ILOG_CORE
;
2484 error
= xfs_bmbt_lookup_eq(cur
, &old
, &i
);
2487 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2488 error
= -EFSCORRUPTED
;
2491 /* new right extent - oldext */
2492 error
= xfs_bmbt_update(cur
, &r
[1]);
2495 /* new left extent - oldext */
2496 cur
->bc_rec
.b
= PREV
;
2497 if ((error
= xfs_btree_insert(cur
, &i
)))
2499 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2500 error
= -EFSCORRUPTED
;
2504 * Reset the cursor to the position of the new extent
2505 * we are about to insert as we can't trust it after
2506 * the previous insert.
2508 error
= xfs_bmbt_lookup_eq(cur
, new, &i
);
2511 if (XFS_IS_CORRUPT(mp
, i
!= 0)) {
2512 error
= -EFSCORRUPTED
;
2515 /* new middle extent - newext */
2516 if ((error
= xfs_btree_insert(cur
, &i
)))
2518 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2519 error
= -EFSCORRUPTED
;
2525 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2526 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2527 case BMAP_LEFT_FILLING
| BMAP_RIGHT_CONTIG
:
2528 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
2529 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2530 case BMAP_LEFT_CONTIG
:
2531 case BMAP_RIGHT_CONTIG
:
2533 * These cases are all impossible.
2538 /* update reverse mappings */
2539 xfs_rmap_convert_extent(mp
, tp
, ip
, whichfork
, new);
2541 /* convert to a btree if necessary */
2542 if (xfs_bmap_needs_btree(ip
, whichfork
)) {
2543 int tmp_logflags
; /* partial log flag return val */
2545 ASSERT(cur
== NULL
);
2546 error
= xfs_bmap_extents_to_btree(tp
, ip
, &cur
, 0,
2547 &tmp_logflags
, whichfork
);
2548 *logflagsp
|= tmp_logflags
;
2553 /* clear out the allocated field, done with it now in any case. */
2555 cur
->bc_ino
.allocated
= 0;
2559 xfs_bmap_check_leaf_extents(*curp
, ip
, whichfork
);
2569 * Convert a hole to a delayed allocation.
2572 xfs_bmap_add_extent_hole_delay(
2573 xfs_inode_t
*ip
, /* incore inode pointer */
2575 struct xfs_iext_cursor
*icur
,
2576 xfs_bmbt_irec_t
*new) /* new data to add to file extents */
2578 struct xfs_ifork
*ifp
; /* inode fork pointer */
2579 xfs_bmbt_irec_t left
; /* left neighbor extent entry */
2580 xfs_filblks_t newlen
=0; /* new indirect size */
2581 xfs_filblks_t oldlen
=0; /* old indirect size */
2582 xfs_bmbt_irec_t right
; /* right neighbor extent entry */
2583 int state
= xfs_bmap_fork_to_state(whichfork
);
2584 xfs_filblks_t temp
; /* temp for indirect calculations */
2586 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2587 ASSERT(isnullstartblock(new->br_startblock
));
2590 * Check and set flags if this segment has a left neighbor
2592 if (xfs_iext_peek_prev_extent(ifp
, icur
, &left
)) {
2593 state
|= BMAP_LEFT_VALID
;
2594 if (isnullstartblock(left
.br_startblock
))
2595 state
|= BMAP_LEFT_DELAY
;
2599 * Check and set flags if the current (right) segment exists.
2600 * If it doesn't exist, we're converting the hole at end-of-file.
2602 if (xfs_iext_get_extent(ifp
, icur
, &right
)) {
2603 state
|= BMAP_RIGHT_VALID
;
2604 if (isnullstartblock(right
.br_startblock
))
2605 state
|= BMAP_RIGHT_DELAY
;
2609 * Set contiguity flags on the left and right neighbors.
2610 * Don't let extents get too large, even if the pieces are contiguous.
2612 if ((state
& BMAP_LEFT_VALID
) && (state
& BMAP_LEFT_DELAY
) &&
2613 left
.br_startoff
+ left
.br_blockcount
== new->br_startoff
&&
2614 left
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
2615 state
|= BMAP_LEFT_CONTIG
;
2617 if ((state
& BMAP_RIGHT_VALID
) && (state
& BMAP_RIGHT_DELAY
) &&
2618 new->br_startoff
+ new->br_blockcount
== right
.br_startoff
&&
2619 new->br_blockcount
+ right
.br_blockcount
<= MAXEXTLEN
&&
2620 (!(state
& BMAP_LEFT_CONTIG
) ||
2621 (left
.br_blockcount
+ new->br_blockcount
+
2622 right
.br_blockcount
<= MAXEXTLEN
)))
2623 state
|= BMAP_RIGHT_CONTIG
;
2626 * Switch out based on the contiguity flags.
2628 switch (state
& (BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
)) {
2629 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2631 * New allocation is contiguous with delayed allocations
2632 * on the left and on the right.
2633 * Merge all three into a single extent record.
2635 temp
= left
.br_blockcount
+ new->br_blockcount
+
2636 right
.br_blockcount
;
2638 oldlen
= startblockval(left
.br_startblock
) +
2639 startblockval(new->br_startblock
) +
2640 startblockval(right
.br_startblock
);
2641 newlen
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
, temp
),
2643 left
.br_startblock
= nullstartblock(newlen
);
2644 left
.br_blockcount
= temp
;
2646 xfs_iext_remove(ip
, icur
, state
);
2647 xfs_iext_prev(ifp
, icur
);
2648 xfs_iext_update_extent(ip
, state
, icur
, &left
);
2651 case BMAP_LEFT_CONTIG
:
2653 * New allocation is contiguous with a delayed allocation
2655 * Merge the new allocation with the left neighbor.
2657 temp
= left
.br_blockcount
+ new->br_blockcount
;
2659 oldlen
= startblockval(left
.br_startblock
) +
2660 startblockval(new->br_startblock
);
2661 newlen
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
, temp
),
2663 left
.br_blockcount
= temp
;
2664 left
.br_startblock
= nullstartblock(newlen
);
2666 xfs_iext_prev(ifp
, icur
);
2667 xfs_iext_update_extent(ip
, state
, icur
, &left
);
2670 case BMAP_RIGHT_CONTIG
:
2672 * New allocation is contiguous with a delayed allocation
2674 * Merge the new allocation with the right neighbor.
2676 temp
= new->br_blockcount
+ right
.br_blockcount
;
2677 oldlen
= startblockval(new->br_startblock
) +
2678 startblockval(right
.br_startblock
);
2679 newlen
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
, temp
),
2681 right
.br_startoff
= new->br_startoff
;
2682 right
.br_startblock
= nullstartblock(newlen
);
2683 right
.br_blockcount
= temp
;
2684 xfs_iext_update_extent(ip
, state
, icur
, &right
);
2689 * New allocation is not contiguous with another
2690 * delayed allocation.
2691 * Insert a new entry.
2693 oldlen
= newlen
= 0;
2694 xfs_iext_insert(ip
, icur
, new, state
);
2697 if (oldlen
!= newlen
) {
2698 ASSERT(oldlen
> newlen
);
2699 xfs_mod_fdblocks(ip
->i_mount
, (int64_t)(oldlen
- newlen
),
2702 * Nothing to do for disk quota accounting here.
2704 xfs_mod_delalloc(ip
->i_mount
, (int64_t)newlen
- oldlen
);
2709 * Convert a hole to a real allocation.
2711 STATIC
int /* error */
2712 xfs_bmap_add_extent_hole_real(
2713 struct xfs_trans
*tp
,
2714 struct xfs_inode
*ip
,
2716 struct xfs_iext_cursor
*icur
,
2717 struct xfs_btree_cur
**curp
,
2718 struct xfs_bmbt_irec
*new,
2722 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2723 struct xfs_mount
*mp
= ip
->i_mount
;
2724 struct xfs_btree_cur
*cur
= *curp
;
2725 int error
; /* error return value */
2726 int i
; /* temp state */
2727 xfs_bmbt_irec_t left
; /* left neighbor extent entry */
2728 xfs_bmbt_irec_t right
; /* right neighbor extent entry */
2729 int rval
=0; /* return value (logging flags) */
2730 int state
= xfs_bmap_fork_to_state(whichfork
);
2731 struct xfs_bmbt_irec old
;
2733 ASSERT(!isnullstartblock(new->br_startblock
));
2734 ASSERT(!cur
|| !(cur
->bc_ino
.flags
& XFS_BTCUR_BMBT_WASDEL
));
2736 XFS_STATS_INC(mp
, xs_add_exlist
);
2739 * Check and set flags if this segment has a left neighbor.
2741 if (xfs_iext_peek_prev_extent(ifp
, icur
, &left
)) {
2742 state
|= BMAP_LEFT_VALID
;
2743 if (isnullstartblock(left
.br_startblock
))
2744 state
|= BMAP_LEFT_DELAY
;
2748 * Check and set flags if this segment has a current value.
2749 * Not true if we're inserting into the "hole" at eof.
2751 if (xfs_iext_get_extent(ifp
, icur
, &right
)) {
2752 state
|= BMAP_RIGHT_VALID
;
2753 if (isnullstartblock(right
.br_startblock
))
2754 state
|= BMAP_RIGHT_DELAY
;
2758 * We're inserting a real allocation between "left" and "right".
2759 * Set the contiguity flags. Don't let extents get too large.
2761 if ((state
& BMAP_LEFT_VALID
) && !(state
& BMAP_LEFT_DELAY
) &&
2762 left
.br_startoff
+ left
.br_blockcount
== new->br_startoff
&&
2763 left
.br_startblock
+ left
.br_blockcount
== new->br_startblock
&&
2764 left
.br_state
== new->br_state
&&
2765 left
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
2766 state
|= BMAP_LEFT_CONTIG
;
2768 if ((state
& BMAP_RIGHT_VALID
) && !(state
& BMAP_RIGHT_DELAY
) &&
2769 new->br_startoff
+ new->br_blockcount
== right
.br_startoff
&&
2770 new->br_startblock
+ new->br_blockcount
== right
.br_startblock
&&
2771 new->br_state
== right
.br_state
&&
2772 new->br_blockcount
+ right
.br_blockcount
<= MAXEXTLEN
&&
2773 (!(state
& BMAP_LEFT_CONTIG
) ||
2774 left
.br_blockcount
+ new->br_blockcount
+
2775 right
.br_blockcount
<= MAXEXTLEN
))
2776 state
|= BMAP_RIGHT_CONTIG
;
2780 * Select which case we're in here, and implement it.
2782 switch (state
& (BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
)) {
2783 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2785 * New allocation is contiguous with real allocations on the
2786 * left and on the right.
2787 * Merge all three into a single extent record.
2789 left
.br_blockcount
+= new->br_blockcount
+ right
.br_blockcount
;
2791 xfs_iext_remove(ip
, icur
, state
);
2792 xfs_iext_prev(ifp
, icur
);
2793 xfs_iext_update_extent(ip
, state
, icur
, &left
);
2797 rval
= XFS_ILOG_CORE
| xfs_ilog_fext(whichfork
);
2799 rval
= XFS_ILOG_CORE
;
2800 error
= xfs_bmbt_lookup_eq(cur
, &right
, &i
);
2803 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2804 error
= -EFSCORRUPTED
;
2807 error
= xfs_btree_delete(cur
, &i
);
2810 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2811 error
= -EFSCORRUPTED
;
2814 error
= xfs_btree_decrement(cur
, 0, &i
);
2817 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2818 error
= -EFSCORRUPTED
;
2821 error
= xfs_bmbt_update(cur
, &left
);
2827 case BMAP_LEFT_CONTIG
:
2829 * New allocation is contiguous with a real allocation
2831 * Merge the new allocation with the left neighbor.
2834 left
.br_blockcount
+= new->br_blockcount
;
2836 xfs_iext_prev(ifp
, icur
);
2837 xfs_iext_update_extent(ip
, state
, icur
, &left
);
2840 rval
= xfs_ilog_fext(whichfork
);
2843 error
= xfs_bmbt_lookup_eq(cur
, &old
, &i
);
2846 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2847 error
= -EFSCORRUPTED
;
2850 error
= xfs_bmbt_update(cur
, &left
);
2856 case BMAP_RIGHT_CONTIG
:
2858 * New allocation is contiguous with a real allocation
2860 * Merge the new allocation with the right neighbor.
2864 right
.br_startoff
= new->br_startoff
;
2865 right
.br_startblock
= new->br_startblock
;
2866 right
.br_blockcount
+= new->br_blockcount
;
2867 xfs_iext_update_extent(ip
, state
, icur
, &right
);
2870 rval
= xfs_ilog_fext(whichfork
);
2873 error
= xfs_bmbt_lookup_eq(cur
, &old
, &i
);
2876 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2877 error
= -EFSCORRUPTED
;
2880 error
= xfs_bmbt_update(cur
, &right
);
2888 * New allocation is not contiguous with another
2890 * Insert a new entry.
2892 xfs_iext_insert(ip
, icur
, new, state
);
2896 rval
= XFS_ILOG_CORE
| xfs_ilog_fext(whichfork
);
2898 rval
= XFS_ILOG_CORE
;
2899 error
= xfs_bmbt_lookup_eq(cur
, new, &i
);
2902 if (XFS_IS_CORRUPT(mp
, i
!= 0)) {
2903 error
= -EFSCORRUPTED
;
2906 error
= xfs_btree_insert(cur
, &i
);
2909 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2910 error
= -EFSCORRUPTED
;
2917 /* add reverse mapping unless caller opted out */
2918 if (!(flags
& XFS_BMAPI_NORMAP
))
2919 xfs_rmap_map_extent(tp
, ip
, whichfork
, new);
2921 /* convert to a btree if necessary */
2922 if (xfs_bmap_needs_btree(ip
, whichfork
)) {
2923 int tmp_logflags
; /* partial log flag return val */
2925 ASSERT(cur
== NULL
);
2926 error
= xfs_bmap_extents_to_btree(tp
, ip
, curp
, 0,
2927 &tmp_logflags
, whichfork
);
2928 *logflagsp
|= tmp_logflags
;
2934 /* clear out the allocated field, done with it now in any case. */
2936 cur
->bc_ino
.allocated
= 0;
2938 xfs_bmap_check_leaf_extents(cur
, ip
, whichfork
);
2945 * Functions used in the extent read, allocate and remove paths
2949 * Adjust the size of the new extent based on di_extsize and rt extsize.
2952 xfs_bmap_extsize_align(
2954 xfs_bmbt_irec_t
*gotp
, /* next extent pointer */
2955 xfs_bmbt_irec_t
*prevp
, /* previous extent pointer */
2956 xfs_extlen_t extsz
, /* align to this extent size */
2957 int rt
, /* is this a realtime inode? */
2958 int eof
, /* is extent at end-of-file? */
2959 int delay
, /* creating delalloc extent? */
2960 int convert
, /* overwriting unwritten extent? */
2961 xfs_fileoff_t
*offp
, /* in/out: aligned offset */
2962 xfs_extlen_t
*lenp
) /* in/out: aligned length */
2964 xfs_fileoff_t orig_off
; /* original offset */
2965 xfs_extlen_t orig_alen
; /* original length */
2966 xfs_fileoff_t orig_end
; /* original off+len */
2967 xfs_fileoff_t nexto
; /* next file offset */
2968 xfs_fileoff_t prevo
; /* previous file offset */
2969 xfs_fileoff_t align_off
; /* temp for offset */
2970 xfs_extlen_t align_alen
; /* temp for length */
2971 xfs_extlen_t temp
; /* temp for calculations */
2976 orig_off
= align_off
= *offp
;
2977 orig_alen
= align_alen
= *lenp
;
2978 orig_end
= orig_off
+ orig_alen
;
2981 * If this request overlaps an existing extent, then don't
2982 * attempt to perform any additional alignment.
2984 if (!delay
&& !eof
&&
2985 (orig_off
>= gotp
->br_startoff
) &&
2986 (orig_end
<= gotp
->br_startoff
+ gotp
->br_blockcount
)) {
2991 * If the file offset is unaligned vs. the extent size
2992 * we need to align it. This will be possible unless
2993 * the file was previously written with a kernel that didn't
2994 * perform this alignment, or if a truncate shot us in the
2997 div_u64_rem(orig_off
, extsz
, &temp
);
3003 /* Same adjustment for the end of the requested area. */
3004 temp
= (align_alen
% extsz
);
3006 align_alen
+= extsz
- temp
;
3009 * For large extent hint sizes, the aligned extent might be larger than
3010 * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls
3011 * the length back under MAXEXTLEN. The outer allocation loops handle
3012 * short allocation just fine, so it is safe to do this. We only want to
3013 * do it when we are forced to, though, because it means more allocation
3014 * operations are required.
3016 while (align_alen
> MAXEXTLEN
)
3017 align_alen
-= extsz
;
3018 ASSERT(align_alen
<= MAXEXTLEN
);
3021 * If the previous block overlaps with this proposed allocation
3022 * then move the start forward without adjusting the length.
3024 if (prevp
->br_startoff
!= NULLFILEOFF
) {
3025 if (prevp
->br_startblock
== HOLESTARTBLOCK
)
3026 prevo
= prevp
->br_startoff
;
3028 prevo
= prevp
->br_startoff
+ prevp
->br_blockcount
;
3031 if (align_off
!= orig_off
&& align_off
< prevo
)
3034 * If the next block overlaps with this proposed allocation
3035 * then move the start back without adjusting the length,
3036 * but not before offset 0.
3037 * This may of course make the start overlap previous block,
3038 * and if we hit the offset 0 limit then the next block
3039 * can still overlap too.
3041 if (!eof
&& gotp
->br_startoff
!= NULLFILEOFF
) {
3042 if ((delay
&& gotp
->br_startblock
== HOLESTARTBLOCK
) ||
3043 (!delay
&& gotp
->br_startblock
== DELAYSTARTBLOCK
))
3044 nexto
= gotp
->br_startoff
+ gotp
->br_blockcount
;
3046 nexto
= gotp
->br_startoff
;
3048 nexto
= NULLFILEOFF
;
3050 align_off
+ align_alen
!= orig_end
&&
3051 align_off
+ align_alen
> nexto
)
3052 align_off
= nexto
> align_alen
? nexto
- align_alen
: 0;
3054 * If we're now overlapping the next or previous extent that
3055 * means we can't fit an extsz piece in this hole. Just move
3056 * the start forward to the first valid spot and set
3057 * the length so we hit the end.
3059 if (align_off
!= orig_off
&& align_off
< prevo
)
3061 if (align_off
+ align_alen
!= orig_end
&&
3062 align_off
+ align_alen
> nexto
&&
3063 nexto
!= NULLFILEOFF
) {
3064 ASSERT(nexto
> prevo
);
3065 align_alen
= nexto
- align_off
;
3069 * If realtime, and the result isn't a multiple of the realtime
3070 * extent size we need to remove blocks until it is.
3072 if (rt
&& (temp
= (align_alen
% mp
->m_sb
.sb_rextsize
))) {
3074 * We're not covering the original request, or
3075 * we won't be able to once we fix the length.
3077 if (orig_off
< align_off
||
3078 orig_end
> align_off
+ align_alen
||
3079 align_alen
- temp
< orig_alen
)
3082 * Try to fix it by moving the start up.
3084 if (align_off
+ temp
<= orig_off
) {
3089 * Try to fix it by moving the end in.
3091 else if (align_off
+ align_alen
- temp
>= orig_end
)
3094 * Set the start to the minimum then trim the length.
3097 align_alen
-= orig_off
- align_off
;
3098 align_off
= orig_off
;
3099 align_alen
-= align_alen
% mp
->m_sb
.sb_rextsize
;
3102 * Result doesn't cover the request, fail it.
3104 if (orig_off
< align_off
|| orig_end
> align_off
+ align_alen
)
3107 ASSERT(orig_off
>= align_off
);
3108 /* see MAXEXTLEN handling above */
3109 ASSERT(orig_end
<= align_off
+ align_alen
||
3110 align_alen
+ extsz
> MAXEXTLEN
);
3114 if (!eof
&& gotp
->br_startoff
!= NULLFILEOFF
)
3115 ASSERT(align_off
+ align_alen
<= gotp
->br_startoff
);
3116 if (prevp
->br_startoff
!= NULLFILEOFF
)
3117 ASSERT(align_off
>= prevp
->br_startoff
+ prevp
->br_blockcount
);
3125 #define XFS_ALLOC_GAP_UNITS 4
3129 struct xfs_bmalloca
*ap
) /* bmap alloc argument struct */
3131 xfs_fsblock_t adjust
; /* adjustment to block numbers */
3132 xfs_agnumber_t fb_agno
; /* ag number of ap->firstblock */
3133 xfs_mount_t
*mp
; /* mount point structure */
3134 int nullfb
; /* true if ap->firstblock isn't set */
3135 int rt
; /* true if inode is realtime */
3137 #define ISVALID(x,y) \
3139 (x) < mp->m_sb.sb_rblocks : \
3140 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3141 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3142 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3144 mp
= ap
->ip
->i_mount
;
3145 nullfb
= ap
->tp
->t_firstblock
== NULLFSBLOCK
;
3146 rt
= XFS_IS_REALTIME_INODE(ap
->ip
) &&
3147 (ap
->datatype
& XFS_ALLOC_USERDATA
);
3148 fb_agno
= nullfb
? NULLAGNUMBER
: XFS_FSB_TO_AGNO(mp
,
3149 ap
->tp
->t_firstblock
);
3151 * If allocating at eof, and there's a previous real block,
3152 * try to use its last block as our starting point.
3154 if (ap
->eof
&& ap
->prev
.br_startoff
!= NULLFILEOFF
&&
3155 !isnullstartblock(ap
->prev
.br_startblock
) &&
3156 ISVALID(ap
->prev
.br_startblock
+ ap
->prev
.br_blockcount
,
3157 ap
->prev
.br_startblock
)) {
3158 ap
->blkno
= ap
->prev
.br_startblock
+ ap
->prev
.br_blockcount
;
3160 * Adjust for the gap between prevp and us.
3162 adjust
= ap
->offset
-
3163 (ap
->prev
.br_startoff
+ ap
->prev
.br_blockcount
);
3165 ISVALID(ap
->blkno
+ adjust
, ap
->prev
.br_startblock
))
3166 ap
->blkno
+= adjust
;
3169 * If not at eof, then compare the two neighbor blocks.
3170 * Figure out whether either one gives us a good starting point,
3171 * and pick the better one.
3173 else if (!ap
->eof
) {
3174 xfs_fsblock_t gotbno
; /* right side block number */
3175 xfs_fsblock_t gotdiff
=0; /* right side difference */
3176 xfs_fsblock_t prevbno
; /* left side block number */
3177 xfs_fsblock_t prevdiff
=0; /* left side difference */
3180 * If there's a previous (left) block, select a requested
3181 * start block based on it.
3183 if (ap
->prev
.br_startoff
!= NULLFILEOFF
&&
3184 !isnullstartblock(ap
->prev
.br_startblock
) &&
3185 (prevbno
= ap
->prev
.br_startblock
+
3186 ap
->prev
.br_blockcount
) &&
3187 ISVALID(prevbno
, ap
->prev
.br_startblock
)) {
3189 * Calculate gap to end of previous block.
3191 adjust
= prevdiff
= ap
->offset
-
3192 (ap
->prev
.br_startoff
+
3193 ap
->prev
.br_blockcount
);
3195 * Figure the startblock based on the previous block's
3196 * end and the gap size.
3198 * If the gap is large relative to the piece we're
3199 * allocating, or using it gives us an invalid block
3200 * number, then just use the end of the previous block.
3202 if (prevdiff
<= XFS_ALLOC_GAP_UNITS
* ap
->length
&&
3203 ISVALID(prevbno
+ prevdiff
,
3204 ap
->prev
.br_startblock
))
3209 * If the firstblock forbids it, can't use it,
3212 if (!rt
&& !nullfb
&&
3213 XFS_FSB_TO_AGNO(mp
, prevbno
) != fb_agno
)
3214 prevbno
= NULLFSBLOCK
;
3217 * No previous block or can't follow it, just default.
3220 prevbno
= NULLFSBLOCK
;
3222 * If there's a following (right) block, select a requested
3223 * start block based on it.
3225 if (!isnullstartblock(ap
->got
.br_startblock
)) {
3227 * Calculate gap to start of next block.
3229 adjust
= gotdiff
= ap
->got
.br_startoff
- ap
->offset
;
3231 * Figure the startblock based on the next block's
3232 * start and the gap size.
3234 gotbno
= ap
->got
.br_startblock
;
3237 * If the gap is large relative to the piece we're
3238 * allocating, or using it gives us an invalid block
3239 * number, then just use the start of the next block
3240 * offset by our length.
3242 if (gotdiff
<= XFS_ALLOC_GAP_UNITS
* ap
->length
&&
3243 ISVALID(gotbno
- gotdiff
, gotbno
))
3245 else if (ISVALID(gotbno
- ap
->length
, gotbno
)) {
3246 gotbno
-= ap
->length
;
3247 gotdiff
+= adjust
- ap
->length
;
3251 * If the firstblock forbids it, can't use it,
3254 if (!rt
&& !nullfb
&&
3255 XFS_FSB_TO_AGNO(mp
, gotbno
) != fb_agno
)
3256 gotbno
= NULLFSBLOCK
;
3259 * No next block, just default.
3262 gotbno
= NULLFSBLOCK
;
3264 * If both valid, pick the better one, else the only good
3265 * one, else ap->blkno is already set (to 0 or the inode block).
3267 if (prevbno
!= NULLFSBLOCK
&& gotbno
!= NULLFSBLOCK
)
3268 ap
->blkno
= prevdiff
<= gotdiff
? prevbno
: gotbno
;
3269 else if (prevbno
!= NULLFSBLOCK
)
3270 ap
->blkno
= prevbno
;
3271 else if (gotbno
!= NULLFSBLOCK
)
3278 xfs_bmap_longest_free_extent(
3279 struct xfs_trans
*tp
,
3284 struct xfs_mount
*mp
= tp
->t_mountp
;
3285 struct xfs_perag
*pag
;
3286 xfs_extlen_t longest
;
3289 pag
= xfs_perag_get(mp
, ag
);
3290 if (!pag
->pagf_init
) {
3291 error
= xfs_alloc_pagf_init(mp
, tp
, ag
, XFS_ALLOC_FLAG_TRYLOCK
);
3293 /* Couldn't lock the AGF, so skip this AG. */
3294 if (error
== -EAGAIN
) {
3302 longest
= xfs_alloc_longest_free_extent(pag
,
3303 xfs_alloc_min_freelist(mp
, pag
),
3304 xfs_ag_resv_needed(pag
, XFS_AG_RESV_NONE
));
3305 if (*blen
< longest
)
3314 xfs_bmap_select_minlen(
3315 struct xfs_bmalloca
*ap
,
3316 struct xfs_alloc_arg
*args
,
3320 if (notinit
|| *blen
< ap
->minlen
) {
3322 * Since we did a BUF_TRYLOCK above, it is possible that
3323 * there is space for this request.
3325 args
->minlen
= ap
->minlen
;
3326 } else if (*blen
< args
->maxlen
) {
3328 * If the best seen length is less than the request length,
3329 * use the best as the minimum.
3331 args
->minlen
= *blen
;
3334 * Otherwise we've seen an extent as big as maxlen, use that
3337 args
->minlen
= args
->maxlen
;
3342 xfs_bmap_btalloc_nullfb(
3343 struct xfs_bmalloca
*ap
,
3344 struct xfs_alloc_arg
*args
,
3347 struct xfs_mount
*mp
= ap
->ip
->i_mount
;
3348 xfs_agnumber_t ag
, startag
;
3352 args
->type
= XFS_ALLOCTYPE_START_BNO
;
3353 args
->total
= ap
->total
;
3355 startag
= ag
= XFS_FSB_TO_AGNO(mp
, args
->fsbno
);
3356 if (startag
== NULLAGNUMBER
)
3359 while (*blen
< args
->maxlen
) {
3360 error
= xfs_bmap_longest_free_extent(args
->tp
, ag
, blen
,
3365 if (++ag
== mp
->m_sb
.sb_agcount
)
3371 xfs_bmap_select_minlen(ap
, args
, blen
, notinit
);
3376 xfs_bmap_btalloc_filestreams(
3377 struct xfs_bmalloca
*ap
,
3378 struct xfs_alloc_arg
*args
,
3381 struct xfs_mount
*mp
= ap
->ip
->i_mount
;
3386 args
->type
= XFS_ALLOCTYPE_NEAR_BNO
;
3387 args
->total
= ap
->total
;
3389 ag
= XFS_FSB_TO_AGNO(mp
, args
->fsbno
);
3390 if (ag
== NULLAGNUMBER
)
3393 error
= xfs_bmap_longest_free_extent(args
->tp
, ag
, blen
, ¬init
);
3397 if (*blen
< args
->maxlen
) {
3398 error
= xfs_filestream_new_ag(ap
, &ag
);
3402 error
= xfs_bmap_longest_free_extent(args
->tp
, ag
, blen
,
3409 xfs_bmap_select_minlen(ap
, args
, blen
, notinit
);
3412 * Set the failure fallback case to look in the selected AG as stream
3415 ap
->blkno
= args
->fsbno
= XFS_AGB_TO_FSB(mp
, ag
, 0);
3419 /* Update all inode and quota accounting for the allocation we just did. */
3421 xfs_bmap_btalloc_accounting(
3422 struct xfs_bmalloca
*ap
,
3423 struct xfs_alloc_arg
*args
)
3425 if (ap
->flags
& XFS_BMAPI_COWFORK
) {
3427 * COW fork blocks are in-core only and thus are treated as
3428 * in-core quota reservation (like delalloc blocks) even when
3429 * converted to real blocks. The quota reservation is not
3430 * accounted to disk until blocks are remapped to the data
3431 * fork. So if these blocks were previously delalloc, we
3432 * already have quota reservation and there's nothing to do
3436 xfs_mod_delalloc(ap
->ip
->i_mount
, -(int64_t)args
->len
);
3441 * Otherwise, we've allocated blocks in a hole. The transaction
3442 * has acquired in-core quota reservation for this extent.
3443 * Rather than account these as real blocks, however, we reduce
3444 * the transaction quota reservation based on the allocation.
3445 * This essentially transfers the transaction quota reservation
3446 * to that of a delalloc extent.
3448 ap
->ip
->i_delayed_blks
+= args
->len
;
3449 xfs_trans_mod_dquot_byino(ap
->tp
, ap
->ip
, XFS_TRANS_DQ_RES_BLKS
,
3454 /* data/attr fork only */
3455 ap
->ip
->i_d
.di_nblocks
+= args
->len
;
3456 xfs_trans_log_inode(ap
->tp
, ap
->ip
, XFS_ILOG_CORE
);
3458 ap
->ip
->i_delayed_blks
-= args
->len
;
3459 xfs_mod_delalloc(ap
->ip
->i_mount
, -(int64_t)args
->len
);
3461 xfs_trans_mod_dquot_byino(ap
->tp
, ap
->ip
,
3462 ap
->wasdel
? XFS_TRANS_DQ_DELBCOUNT
: XFS_TRANS_DQ_BCOUNT
,
3468 struct xfs_bmalloca
*ap
) /* bmap alloc argument struct */
3470 xfs_mount_t
*mp
; /* mount point structure */
3471 xfs_alloctype_t atype
= 0; /* type for allocation routines */
3472 xfs_extlen_t align
= 0; /* minimum allocation alignment */
3473 xfs_agnumber_t fb_agno
; /* ag number of ap->firstblock */
3475 xfs_alloc_arg_t args
;
3476 xfs_fileoff_t orig_offset
;
3477 xfs_extlen_t orig_length
;
3479 xfs_extlen_t nextminlen
= 0;
3480 int nullfb
; /* true if ap->firstblock isn't set */
3487 orig_offset
= ap
->offset
;
3488 orig_length
= ap
->length
;
3490 mp
= ap
->ip
->i_mount
;
3492 /* stripe alignment for allocation is determined by mount parameters */
3494 if (mp
->m_swidth
&& (mp
->m_flags
& XFS_MOUNT_SWALLOC
))
3495 stripe_align
= mp
->m_swidth
;
3496 else if (mp
->m_dalign
)
3497 stripe_align
= mp
->m_dalign
;
3499 if (ap
->flags
& XFS_BMAPI_COWFORK
)
3500 align
= xfs_get_cowextsz_hint(ap
->ip
);
3501 else if (ap
->datatype
& XFS_ALLOC_USERDATA
)
3502 align
= xfs_get_extsz_hint(ap
->ip
);
3504 error
= xfs_bmap_extsize_align(mp
, &ap
->got
, &ap
->prev
,
3505 align
, 0, ap
->eof
, 0, ap
->conv
,
3506 &ap
->offset
, &ap
->length
);
3512 nullfb
= ap
->tp
->t_firstblock
== NULLFSBLOCK
;
3513 fb_agno
= nullfb
? NULLAGNUMBER
: XFS_FSB_TO_AGNO(mp
,
3514 ap
->tp
->t_firstblock
);
3516 if ((ap
->datatype
& XFS_ALLOC_USERDATA
) &&
3517 xfs_inode_is_filestream(ap
->ip
)) {
3518 ag
= xfs_filestream_lookup_ag(ap
->ip
);
3519 ag
= (ag
!= NULLAGNUMBER
) ? ag
: 0;
3520 ap
->blkno
= XFS_AGB_TO_FSB(mp
, ag
, 0);
3522 ap
->blkno
= XFS_INO_TO_FSB(mp
, ap
->ip
->i_ino
);
3525 ap
->blkno
= ap
->tp
->t_firstblock
;
3527 xfs_bmap_adjacent(ap
);
3530 * If allowed, use ap->blkno; otherwise must use firstblock since
3531 * it's in the right allocation group.
3533 if (nullfb
|| XFS_FSB_TO_AGNO(mp
, ap
->blkno
) == fb_agno
)
3536 ap
->blkno
= ap
->tp
->t_firstblock
;
3538 * Normal allocation, done through xfs_alloc_vextent.
3540 tryagain
= isaligned
= 0;
3541 memset(&args
, 0, sizeof(args
));
3544 args
.fsbno
= ap
->blkno
;
3545 args
.oinfo
= XFS_RMAP_OINFO_SKIP_UPDATE
;
3547 /* Trim the allocation back to the maximum an AG can fit. */
3548 args
.maxlen
= min(ap
->length
, mp
->m_ag_max_usable
);
3552 * Search for an allocation group with a single extent large
3553 * enough for the request. If one isn't found, then adjust
3554 * the minimum allocation size to the largest space found.
3556 if ((ap
->datatype
& XFS_ALLOC_USERDATA
) &&
3557 xfs_inode_is_filestream(ap
->ip
))
3558 error
= xfs_bmap_btalloc_filestreams(ap
, &args
, &blen
);
3560 error
= xfs_bmap_btalloc_nullfb(ap
, &args
, &blen
);
3563 } else if (ap
->tp
->t_flags
& XFS_TRANS_LOWMODE
) {
3564 if (xfs_inode_is_filestream(ap
->ip
))
3565 args
.type
= XFS_ALLOCTYPE_FIRST_AG
;
3567 args
.type
= XFS_ALLOCTYPE_START_BNO
;
3568 args
.total
= args
.minlen
= ap
->minlen
;
3570 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
3571 args
.total
= ap
->total
;
3572 args
.minlen
= ap
->minlen
;
3574 /* apply extent size hints if obtained earlier */
3577 div_u64_rem(ap
->offset
, args
.prod
, &args
.mod
);
3579 args
.mod
= args
.prod
- args
.mod
;
3580 } else if (mp
->m_sb
.sb_blocksize
>= PAGE_SIZE
) {
3584 args
.prod
= PAGE_SIZE
>> mp
->m_sb
.sb_blocklog
;
3585 div_u64_rem(ap
->offset
, args
.prod
, &args
.mod
);
3587 args
.mod
= args
.prod
- args
.mod
;
3590 * If we are not low on available data blocks, and the underlying
3591 * logical volume manager is a stripe, and the file offset is zero then
3592 * try to allocate data blocks on stripe unit boundary. NOTE: ap->aeof
3593 * is only set if the allocation length is >= the stripe unit and the
3594 * allocation offset is at the end of file.
3596 if (!(ap
->tp
->t_flags
& XFS_TRANS_LOWMODE
) && ap
->aeof
) {
3598 args
.alignment
= stripe_align
;
3602 * Adjust minlen to try and preserve alignment if we
3603 * can't guarantee an aligned maxlen extent.
3605 if (blen
> args
.alignment
&&
3606 blen
<= args
.maxlen
+ args
.alignment
)
3607 args
.minlen
= blen
- args
.alignment
;
3608 args
.minalignslop
= 0;
3611 * First try an exact bno allocation.
3612 * If it fails then do a near or start bno
3613 * allocation with alignment turned on.
3617 args
.type
= XFS_ALLOCTYPE_THIS_BNO
;
3620 * Compute the minlen+alignment for the
3621 * next case. Set slop so that the value
3622 * of minlen+alignment+slop doesn't go up
3623 * between the calls.
3625 if (blen
> stripe_align
&& blen
<= args
.maxlen
)
3626 nextminlen
= blen
- stripe_align
;
3628 nextminlen
= args
.minlen
;
3629 if (nextminlen
+ stripe_align
> args
.minlen
+ 1)
3631 nextminlen
+ stripe_align
-
3634 args
.minalignslop
= 0;
3638 args
.minalignslop
= 0;
3640 args
.minleft
= ap
->minleft
;
3641 args
.wasdel
= ap
->wasdel
;
3642 args
.resv
= XFS_AG_RESV_NONE
;
3643 args
.datatype
= ap
->datatype
;
3645 error
= xfs_alloc_vextent(&args
);
3649 if (tryagain
&& args
.fsbno
== NULLFSBLOCK
) {
3651 * Exact allocation failed. Now try with alignment
3655 args
.fsbno
= ap
->blkno
;
3656 args
.alignment
= stripe_align
;
3657 args
.minlen
= nextminlen
;
3658 args
.minalignslop
= 0;
3660 if ((error
= xfs_alloc_vextent(&args
)))
3663 if (isaligned
&& args
.fsbno
== NULLFSBLOCK
) {
3665 * allocation failed, so turn off alignment and
3669 args
.fsbno
= ap
->blkno
;
3671 if ((error
= xfs_alloc_vextent(&args
)))
3674 if (args
.fsbno
== NULLFSBLOCK
&& nullfb
&&
3675 args
.minlen
> ap
->minlen
) {
3676 args
.minlen
= ap
->minlen
;
3677 args
.type
= XFS_ALLOCTYPE_START_BNO
;
3678 args
.fsbno
= ap
->blkno
;
3679 if ((error
= xfs_alloc_vextent(&args
)))
3682 if (args
.fsbno
== NULLFSBLOCK
&& nullfb
) {
3684 args
.type
= XFS_ALLOCTYPE_FIRST_AG
;
3685 args
.total
= ap
->minlen
;
3686 if ((error
= xfs_alloc_vextent(&args
)))
3688 ap
->tp
->t_flags
|= XFS_TRANS_LOWMODE
;
3690 if (args
.fsbno
!= NULLFSBLOCK
) {
3692 * check the allocation happened at the same or higher AG than
3693 * the first block that was allocated.
3695 ASSERT(ap
->tp
->t_firstblock
== NULLFSBLOCK
||
3696 XFS_FSB_TO_AGNO(mp
, ap
->tp
->t_firstblock
) <=
3697 XFS_FSB_TO_AGNO(mp
, args
.fsbno
));
3699 ap
->blkno
= args
.fsbno
;
3700 if (ap
->tp
->t_firstblock
== NULLFSBLOCK
)
3701 ap
->tp
->t_firstblock
= args
.fsbno
;
3702 ASSERT(nullfb
|| fb_agno
<= args
.agno
);
3703 ap
->length
= args
.len
;
3705 * If the extent size hint is active, we tried to round the
3706 * caller's allocation request offset down to extsz and the
3707 * length up to another extsz boundary. If we found a free
3708 * extent we mapped it in starting at this new offset. If the
3709 * newly mapped space isn't long enough to cover any of the
3710 * range of offsets that was originally requested, move the
3711 * mapping up so that we can fill as much of the caller's
3712 * original request as possible. Free space is apparently
3713 * very fragmented so we're unlikely to be able to satisfy the
3716 if (ap
->length
<= orig_length
)
3717 ap
->offset
= orig_offset
;
3718 else if (ap
->offset
+ ap
->length
< orig_offset
+ orig_length
)
3719 ap
->offset
= orig_offset
+ orig_length
- ap
->length
;
3720 xfs_bmap_btalloc_accounting(ap
, &args
);
3722 ap
->blkno
= NULLFSBLOCK
;
3728 /* Trim extent to fit a logical block range. */
3731 struct xfs_bmbt_irec
*irec
,
3735 xfs_fileoff_t distance
;
3736 xfs_fileoff_t end
= bno
+ len
;
3738 if (irec
->br_startoff
+ irec
->br_blockcount
<= bno
||
3739 irec
->br_startoff
>= end
) {
3740 irec
->br_blockcount
= 0;
3744 if (irec
->br_startoff
< bno
) {
3745 distance
= bno
- irec
->br_startoff
;
3746 if (isnullstartblock(irec
->br_startblock
))
3747 irec
->br_startblock
= DELAYSTARTBLOCK
;
3748 if (irec
->br_startblock
!= DELAYSTARTBLOCK
&&
3749 irec
->br_startblock
!= HOLESTARTBLOCK
)
3750 irec
->br_startblock
+= distance
;
3751 irec
->br_startoff
+= distance
;
3752 irec
->br_blockcount
-= distance
;
3755 if (end
< irec
->br_startoff
+ irec
->br_blockcount
) {
3756 distance
= irec
->br_startoff
+ irec
->br_blockcount
- end
;
3757 irec
->br_blockcount
-= distance
;
3762 * Trim the returned map to the required bounds
3766 struct xfs_bmbt_irec
*mval
,
3767 struct xfs_bmbt_irec
*got
,
3775 if ((flags
& XFS_BMAPI_ENTIRE
) ||
3776 got
->br_startoff
+ got
->br_blockcount
<= obno
) {
3778 if (isnullstartblock(got
->br_startblock
))
3779 mval
->br_startblock
= DELAYSTARTBLOCK
;
3785 ASSERT((*bno
>= obno
) || (n
== 0));
3787 mval
->br_startoff
= *bno
;
3788 if (isnullstartblock(got
->br_startblock
))
3789 mval
->br_startblock
= DELAYSTARTBLOCK
;
3791 mval
->br_startblock
= got
->br_startblock
+
3792 (*bno
- got
->br_startoff
);
3794 * Return the minimum of what we got and what we asked for for
3795 * the length. We can use the len variable here because it is
3796 * modified below and we could have been there before coming
3797 * here if the first part of the allocation didn't overlap what
3800 mval
->br_blockcount
= XFS_FILBLKS_MIN(end
- *bno
,
3801 got
->br_blockcount
- (*bno
- got
->br_startoff
));
3802 mval
->br_state
= got
->br_state
;
3803 ASSERT(mval
->br_blockcount
<= len
);
3808 * Update and validate the extent map to return
3811 xfs_bmapi_update_map(
3812 struct xfs_bmbt_irec
**map
,
3820 xfs_bmbt_irec_t
*mval
= *map
;
3822 ASSERT((flags
& XFS_BMAPI_ENTIRE
) ||
3823 ((mval
->br_startoff
+ mval
->br_blockcount
) <= end
));
3824 ASSERT((flags
& XFS_BMAPI_ENTIRE
) || (mval
->br_blockcount
<= *len
) ||
3825 (mval
->br_startoff
< obno
));
3827 *bno
= mval
->br_startoff
+ mval
->br_blockcount
;
3829 if (*n
> 0 && mval
->br_startoff
== mval
[-1].br_startoff
) {
3830 /* update previous map with new information */
3831 ASSERT(mval
->br_startblock
== mval
[-1].br_startblock
);
3832 ASSERT(mval
->br_blockcount
> mval
[-1].br_blockcount
);
3833 ASSERT(mval
->br_state
== mval
[-1].br_state
);
3834 mval
[-1].br_blockcount
= mval
->br_blockcount
;
3835 mval
[-1].br_state
= mval
->br_state
;
3836 } else if (*n
> 0 && mval
->br_startblock
!= DELAYSTARTBLOCK
&&
3837 mval
[-1].br_startblock
!= DELAYSTARTBLOCK
&&
3838 mval
[-1].br_startblock
!= HOLESTARTBLOCK
&&
3839 mval
->br_startblock
== mval
[-1].br_startblock
+
3840 mval
[-1].br_blockcount
&&
3841 mval
[-1].br_state
== mval
->br_state
) {
3842 ASSERT(mval
->br_startoff
==
3843 mval
[-1].br_startoff
+ mval
[-1].br_blockcount
);
3844 mval
[-1].br_blockcount
+= mval
->br_blockcount
;
3845 } else if (*n
> 0 &&
3846 mval
->br_startblock
== DELAYSTARTBLOCK
&&
3847 mval
[-1].br_startblock
== DELAYSTARTBLOCK
&&
3848 mval
->br_startoff
==
3849 mval
[-1].br_startoff
+ mval
[-1].br_blockcount
) {
3850 mval
[-1].br_blockcount
+= mval
->br_blockcount
;
3851 mval
[-1].br_state
= mval
->br_state
;
3852 } else if (!((*n
== 0) &&
3853 ((mval
->br_startoff
+ mval
->br_blockcount
) <=
3862 * Map file blocks to filesystem blocks without allocation.
3866 struct xfs_inode
*ip
,
3869 struct xfs_bmbt_irec
*mval
,
3873 struct xfs_mount
*mp
= ip
->i_mount
;
3874 int whichfork
= xfs_bmapi_whichfork(flags
);
3875 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
3876 struct xfs_bmbt_irec got
;
3879 struct xfs_iext_cursor icur
;
3885 ASSERT(!(flags
& ~(XFS_BMAPI_ATTRFORK
| XFS_BMAPI_ENTIRE
)));
3886 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
));
3888 if (WARN_ON_ONCE(!ifp
))
3889 return -EFSCORRUPTED
;
3891 if (XFS_IS_CORRUPT(mp
, !xfs_ifork_has_extents(ifp
)) ||
3892 XFS_TEST_ERROR(false, mp
, XFS_ERRTAG_BMAPIFORMAT
))
3893 return -EFSCORRUPTED
;
3895 if (XFS_FORCED_SHUTDOWN(mp
))
3898 XFS_STATS_INC(mp
, xs_blk_mapr
);
3900 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
3901 error
= xfs_iread_extents(NULL
, ip
, whichfork
);
3906 if (!xfs_iext_lookup_extent(ip
, ifp
, bno
, &icur
, &got
))
3911 while (bno
< end
&& n
< *nmap
) {
3912 /* Reading past eof, act as though there's a hole up to end. */
3914 got
.br_startoff
= end
;
3915 if (got
.br_startoff
> bno
) {
3916 /* Reading in a hole. */
3917 mval
->br_startoff
= bno
;
3918 mval
->br_startblock
= HOLESTARTBLOCK
;
3919 mval
->br_blockcount
=
3920 XFS_FILBLKS_MIN(len
, got
.br_startoff
- bno
);
3921 mval
->br_state
= XFS_EXT_NORM
;
3922 bno
+= mval
->br_blockcount
;
3923 len
-= mval
->br_blockcount
;
3929 /* set up the extent map to return. */
3930 xfs_bmapi_trim_map(mval
, &got
, &bno
, len
, obno
, end
, n
, flags
);
3931 xfs_bmapi_update_map(&mval
, &bno
, &len
, obno
, end
, &n
, flags
);
3933 /* If we're done, stop now. */
3934 if (bno
>= end
|| n
>= *nmap
)
3937 /* Else go on to the next record. */
3938 if (!xfs_iext_next_extent(ifp
, &icur
, &got
))
3946 * Add a delayed allocation extent to an inode. Blocks are reserved from the
3947 * global pool and the extent inserted into the inode in-core extent tree.
3949 * On entry, got refers to the first extent beyond the offset of the extent to
3950 * allocate or eof is specified if no such extent exists. On return, got refers
3951 * to the extent record that was inserted to the inode fork.
3953 * Note that the allocated extent may have been merged with contiguous extents
3954 * during insertion into the inode fork. Thus, got does not reflect the current
3955 * state of the inode fork on return. If necessary, the caller can use lastx to
3956 * look up the updated record in the inode fork.
3959 xfs_bmapi_reserve_delalloc(
3960 struct xfs_inode
*ip
,
3964 xfs_filblks_t prealloc
,
3965 struct xfs_bmbt_irec
*got
,
3966 struct xfs_iext_cursor
*icur
,
3969 struct xfs_mount
*mp
= ip
->i_mount
;
3970 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
3972 xfs_extlen_t indlen
;
3974 xfs_fileoff_t aoff
= off
;
3977 * Cap the alloc length. Keep track of prealloc so we know whether to
3978 * tag the inode before we return.
3980 alen
= XFS_FILBLKS_MIN(len
+ prealloc
, MAXEXTLEN
);
3982 alen
= XFS_FILBLKS_MIN(alen
, got
->br_startoff
- aoff
);
3983 if (prealloc
&& alen
>= len
)
3984 prealloc
= alen
- len
;
3986 /* Figure out the extent size, adjust alen */
3987 if (whichfork
== XFS_COW_FORK
) {
3988 struct xfs_bmbt_irec prev
;
3989 xfs_extlen_t extsz
= xfs_get_cowextsz_hint(ip
);
3991 if (!xfs_iext_peek_prev_extent(ifp
, icur
, &prev
))
3992 prev
.br_startoff
= NULLFILEOFF
;
3994 error
= xfs_bmap_extsize_align(mp
, got
, &prev
, extsz
, 0, eof
,
3995 1, 0, &aoff
, &alen
);
4000 * Make a transaction-less quota reservation for delayed allocation
4001 * blocks. This number gets adjusted later. We return if we haven't
4002 * allocated blocks already inside this loop.
4004 error
= xfs_trans_reserve_quota_nblks(NULL
, ip
, (long)alen
, 0,
4005 XFS_QMOPT_RES_REGBLKS
);
4010 * Split changing sb for alen and indlen since they could be coming
4011 * from different places.
4013 indlen
= (xfs_extlen_t
)xfs_bmap_worst_indlen(ip
, alen
);
4016 error
= xfs_mod_fdblocks(mp
, -((int64_t)alen
), false);
4018 goto out_unreserve_quota
;
4020 error
= xfs_mod_fdblocks(mp
, -((int64_t)indlen
), false);
4022 goto out_unreserve_blocks
;
4025 ip
->i_delayed_blks
+= alen
;
4026 xfs_mod_delalloc(ip
->i_mount
, alen
+ indlen
);
4028 got
->br_startoff
= aoff
;
4029 got
->br_startblock
= nullstartblock(indlen
);
4030 got
->br_blockcount
= alen
;
4031 got
->br_state
= XFS_EXT_NORM
;
4033 xfs_bmap_add_extent_hole_delay(ip
, whichfork
, icur
, got
);
4036 * Tag the inode if blocks were preallocated. Note that COW fork
4037 * preallocation can occur at the start or end of the extent, even when
4038 * prealloc == 0, so we must also check the aligned offset and length.
4040 if (whichfork
== XFS_DATA_FORK
&& prealloc
)
4041 xfs_inode_set_eofblocks_tag(ip
);
4042 if (whichfork
== XFS_COW_FORK
&& (prealloc
|| aoff
< off
|| alen
> len
))
4043 xfs_inode_set_cowblocks_tag(ip
);
4047 out_unreserve_blocks
:
4048 xfs_mod_fdblocks(mp
, alen
, false);
4049 out_unreserve_quota
:
4050 if (XFS_IS_QUOTA_ON(mp
))
4051 xfs_trans_unreserve_quota_nblks(NULL
, ip
, (long)alen
, 0,
4052 XFS_QMOPT_RES_REGBLKS
);
4057 xfs_bmap_alloc_userdata(
4058 struct xfs_bmalloca
*bma
)
4060 struct xfs_mount
*mp
= bma
->ip
->i_mount
;
4061 int whichfork
= xfs_bmapi_whichfork(bma
->flags
);
4065 * Set the data type being allocated. For the data fork, the first data
4066 * in the file is treated differently to all other allocations. For the
4067 * attribute fork, we only need to ensure the allocated range is not on
4070 bma
->datatype
= XFS_ALLOC_NOBUSY
;
4071 if (whichfork
== XFS_DATA_FORK
) {
4072 bma
->datatype
|= XFS_ALLOC_USERDATA
;
4073 if (bma
->offset
== 0)
4074 bma
->datatype
|= XFS_ALLOC_INITIAL_USER_DATA
;
4076 if (mp
->m_dalign
&& bma
->length
>= mp
->m_dalign
) {
4077 error
= xfs_bmap_isaeof(bma
, whichfork
);
4082 if (XFS_IS_REALTIME_INODE(bma
->ip
))
4083 return xfs_bmap_rtalloc(bma
);
4086 return xfs_bmap_btalloc(bma
);
4091 struct xfs_bmalloca
*bma
)
4093 struct xfs_mount
*mp
= bma
->ip
->i_mount
;
4094 int whichfork
= xfs_bmapi_whichfork(bma
->flags
);
4095 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(bma
->ip
, whichfork
);
4096 int tmp_logflags
= 0;
4099 ASSERT(bma
->length
> 0);
4102 * For the wasdelay case, we could also just allocate the stuff asked
4103 * for in this bmap call but that wouldn't be as good.
4106 bma
->length
= (xfs_extlen_t
)bma
->got
.br_blockcount
;
4107 bma
->offset
= bma
->got
.br_startoff
;
4108 if (!xfs_iext_peek_prev_extent(ifp
, &bma
->icur
, &bma
->prev
))
4109 bma
->prev
.br_startoff
= NULLFILEOFF
;
4111 bma
->length
= XFS_FILBLKS_MIN(bma
->length
, MAXEXTLEN
);
4113 bma
->length
= XFS_FILBLKS_MIN(bma
->length
,
4114 bma
->got
.br_startoff
- bma
->offset
);
4117 if (bma
->flags
& XFS_BMAPI_CONTIG
)
4118 bma
->minlen
= bma
->length
;
4122 if (bma
->flags
& XFS_BMAPI_METADATA
)
4123 error
= xfs_bmap_btalloc(bma
);
4125 error
= xfs_bmap_alloc_userdata(bma
);
4126 if (error
|| bma
->blkno
== NULLFSBLOCK
)
4129 if (bma
->flags
& XFS_BMAPI_ZERO
) {
4130 error
= xfs_zero_extent(bma
->ip
, bma
->blkno
, bma
->length
);
4135 if ((ifp
->if_flags
& XFS_IFBROOT
) && !bma
->cur
)
4136 bma
->cur
= xfs_bmbt_init_cursor(mp
, bma
->tp
, bma
->ip
, whichfork
);
4138 * Bump the number of extents we've allocated
4144 bma
->cur
->bc_ino
.flags
=
4145 bma
->wasdel
? XFS_BTCUR_BMBT_WASDEL
: 0;
4147 bma
->got
.br_startoff
= bma
->offset
;
4148 bma
->got
.br_startblock
= bma
->blkno
;
4149 bma
->got
.br_blockcount
= bma
->length
;
4150 bma
->got
.br_state
= XFS_EXT_NORM
;
4152 if (bma
->flags
& XFS_BMAPI_PREALLOC
)
4153 bma
->got
.br_state
= XFS_EXT_UNWRITTEN
;
4156 error
= xfs_bmap_add_extent_delay_real(bma
, whichfork
);
4158 error
= xfs_bmap_add_extent_hole_real(bma
->tp
, bma
->ip
,
4159 whichfork
, &bma
->icur
, &bma
->cur
, &bma
->got
,
4160 &bma
->logflags
, bma
->flags
);
4162 bma
->logflags
|= tmp_logflags
;
4167 * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4168 * or xfs_bmap_add_extent_hole_real might have merged it into one of
4169 * the neighbouring ones.
4171 xfs_iext_get_extent(ifp
, &bma
->icur
, &bma
->got
);
4173 ASSERT(bma
->got
.br_startoff
<= bma
->offset
);
4174 ASSERT(bma
->got
.br_startoff
+ bma
->got
.br_blockcount
>=
4175 bma
->offset
+ bma
->length
);
4176 ASSERT(bma
->got
.br_state
== XFS_EXT_NORM
||
4177 bma
->got
.br_state
== XFS_EXT_UNWRITTEN
);
4182 xfs_bmapi_convert_unwritten(
4183 struct xfs_bmalloca
*bma
,
4184 struct xfs_bmbt_irec
*mval
,
4188 int whichfork
= xfs_bmapi_whichfork(flags
);
4189 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(bma
->ip
, whichfork
);
4190 int tmp_logflags
= 0;
4193 /* check if we need to do unwritten->real conversion */
4194 if (mval
->br_state
== XFS_EXT_UNWRITTEN
&&
4195 (flags
& XFS_BMAPI_PREALLOC
))
4198 /* check if we need to do real->unwritten conversion */
4199 if (mval
->br_state
== XFS_EXT_NORM
&&
4200 (flags
& (XFS_BMAPI_PREALLOC
| XFS_BMAPI_CONVERT
)) !=
4201 (XFS_BMAPI_PREALLOC
| XFS_BMAPI_CONVERT
))
4205 * Modify (by adding) the state flag, if writing.
4207 ASSERT(mval
->br_blockcount
<= len
);
4208 if ((ifp
->if_flags
& XFS_IFBROOT
) && !bma
->cur
) {
4209 bma
->cur
= xfs_bmbt_init_cursor(bma
->ip
->i_mount
, bma
->tp
,
4210 bma
->ip
, whichfork
);
4212 mval
->br_state
= (mval
->br_state
== XFS_EXT_UNWRITTEN
)
4213 ? XFS_EXT_NORM
: XFS_EXT_UNWRITTEN
;
4216 * Before insertion into the bmbt, zero the range being converted
4219 if (flags
& XFS_BMAPI_ZERO
) {
4220 error
= xfs_zero_extent(bma
->ip
, mval
->br_startblock
,
4221 mval
->br_blockcount
);
4226 error
= xfs_bmap_add_extent_unwritten_real(bma
->tp
, bma
->ip
, whichfork
,
4227 &bma
->icur
, &bma
->cur
, mval
, &tmp_logflags
);
4229 * Log the inode core unconditionally in the unwritten extent conversion
4230 * path because the conversion might not have done so (e.g., if the
4231 * extent count hasn't changed). We need to make sure the inode is dirty
4232 * in the transaction for the sake of fsync(), even if nothing has
4233 * changed, because fsync() will not force the log for this transaction
4234 * unless it sees the inode pinned.
4236 * Note: If we're only converting cow fork extents, there aren't
4237 * any on-disk updates to make, so we don't need to log anything.
4239 if (whichfork
!= XFS_COW_FORK
)
4240 bma
->logflags
|= tmp_logflags
| XFS_ILOG_CORE
;
4245 * Update our extent pointer, given that
4246 * xfs_bmap_add_extent_unwritten_real might have merged it into one
4247 * of the neighbouring ones.
4249 xfs_iext_get_extent(ifp
, &bma
->icur
, &bma
->got
);
4252 * We may have combined previously unwritten space with written space,
4253 * so generate another request.
4255 if (mval
->br_blockcount
< len
)
4260 static inline xfs_extlen_t
4262 struct xfs_trans
*tp
,
4263 struct xfs_inode
*ip
,
4266 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, fork
);
4268 if (tp
&& tp
->t_firstblock
!= NULLFSBLOCK
)
4270 if (ifp
->if_format
!= XFS_DINODE_FMT_BTREE
)
4272 return be16_to_cpu(ifp
->if_broot
->bb_level
) + 1;
4276 * Log whatever the flags say, even if error. Otherwise we might miss detecting
4277 * a case where the data is changed, there's an error, and it's not logged so we
4278 * don't shutdown when we should. Don't bother logging extents/btree changes if
4279 * we converted to the other format.
4283 struct xfs_bmalloca
*bma
,
4287 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(bma
->ip
, whichfork
);
4289 if ((bma
->logflags
& xfs_ilog_fext(whichfork
)) &&
4290 ifp
->if_format
!= XFS_DINODE_FMT_EXTENTS
)
4291 bma
->logflags
&= ~xfs_ilog_fext(whichfork
);
4292 else if ((bma
->logflags
& xfs_ilog_fbroot(whichfork
)) &&
4293 ifp
->if_format
!= XFS_DINODE_FMT_BTREE
)
4294 bma
->logflags
&= ~xfs_ilog_fbroot(whichfork
);
4297 xfs_trans_log_inode(bma
->tp
, bma
->ip
, bma
->logflags
);
4299 xfs_btree_del_cursor(bma
->cur
, error
);
4303 * Map file blocks to filesystem blocks, and allocate blocks or convert the
4304 * extent state if necessary. Details behaviour is controlled by the flags
4305 * parameter. Only allocates blocks from a single allocation group, to avoid
4310 struct xfs_trans
*tp
, /* transaction pointer */
4311 struct xfs_inode
*ip
, /* incore inode */
4312 xfs_fileoff_t bno
, /* starting file offs. mapped */
4313 xfs_filblks_t len
, /* length to map in file */
4314 int flags
, /* XFS_BMAPI_... */
4315 xfs_extlen_t total
, /* total blocks needed */
4316 struct xfs_bmbt_irec
*mval
, /* output: map values */
4317 int *nmap
) /* i/o: mval size/count */
4319 struct xfs_bmalloca bma
= {
4324 struct xfs_mount
*mp
= ip
->i_mount
;
4325 int whichfork
= xfs_bmapi_whichfork(flags
);
4326 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4327 xfs_fileoff_t end
; /* end of mapped file region */
4328 bool eof
= false; /* after the end of extents */
4329 int error
; /* error return */
4330 int n
; /* current extent index */
4331 xfs_fileoff_t obno
; /* old block number (offset) */
4334 xfs_fileoff_t orig_bno
; /* original block number value */
4335 int orig_flags
; /* original flags arg value */
4336 xfs_filblks_t orig_len
; /* original value of len arg */
4337 struct xfs_bmbt_irec
*orig_mval
; /* original value of mval */
4338 int orig_nmap
; /* original value of *nmap */
4348 ASSERT(*nmap
<= XFS_BMAP_MAX_NMAP
);
4351 ASSERT(ifp
->if_format
!= XFS_DINODE_FMT_LOCAL
);
4352 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
4353 ASSERT(!(flags
& XFS_BMAPI_REMAP
));
4355 /* zeroing is for currently only for data extents, not metadata */
4356 ASSERT((flags
& (XFS_BMAPI_METADATA
| XFS_BMAPI_ZERO
)) !=
4357 (XFS_BMAPI_METADATA
| XFS_BMAPI_ZERO
));
4359 * we can allocate unwritten extents or pre-zero allocated blocks,
4360 * but it makes no sense to do both at once. This would result in
4361 * zeroing the unwritten extent twice, but it still being an
4362 * unwritten extent....
4364 ASSERT((flags
& (XFS_BMAPI_PREALLOC
| XFS_BMAPI_ZERO
)) !=
4365 (XFS_BMAPI_PREALLOC
| XFS_BMAPI_ZERO
));
4367 if (XFS_IS_CORRUPT(mp
, !xfs_ifork_has_extents(ifp
)) ||
4368 XFS_TEST_ERROR(false, mp
, XFS_ERRTAG_BMAPIFORMAT
)) {
4369 return -EFSCORRUPTED
;
4372 if (XFS_FORCED_SHUTDOWN(mp
))
4375 XFS_STATS_INC(mp
, xs_blk_mapw
);
4377 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
4378 error
= xfs_iread_extents(tp
, ip
, whichfork
);
4383 if (!xfs_iext_lookup_extent(ip
, ifp
, bno
, &bma
.icur
, &bma
.got
))
4385 if (!xfs_iext_peek_prev_extent(ifp
, &bma
.icur
, &bma
.prev
))
4386 bma
.prev
.br_startoff
= NULLFILEOFF
;
4387 bma
.minleft
= xfs_bmapi_minleft(tp
, ip
, whichfork
);
4392 while (bno
< end
&& n
< *nmap
) {
4393 bool need_alloc
= false, wasdelay
= false;
4395 /* in hole or beyond EOF? */
4396 if (eof
|| bma
.got
.br_startoff
> bno
) {
4398 * CoW fork conversions should /never/ hit EOF or
4399 * holes. There should always be something for us
4402 ASSERT(!((flags
& XFS_BMAPI_CONVERT
) &&
4403 (flags
& XFS_BMAPI_COWFORK
)));
4406 } else if (isnullstartblock(bma
.got
.br_startblock
)) {
4411 * First, deal with the hole before the allocated space
4412 * that we found, if any.
4414 if (need_alloc
|| wasdelay
) {
4416 bma
.conv
= !!(flags
& XFS_BMAPI_CONVERT
);
4417 bma
.wasdel
= wasdelay
;
4422 * There's a 32/64 bit type mismatch between the
4423 * allocation length request (which can be 64 bits in
4424 * length) and the bma length request, which is
4425 * xfs_extlen_t and therefore 32 bits. Hence we have to
4426 * check for 32-bit overflows and handle them here.
4428 if (len
> (xfs_filblks_t
)MAXEXTLEN
)
4429 bma
.length
= MAXEXTLEN
;
4434 ASSERT(bma
.length
> 0);
4435 error
= xfs_bmapi_allocate(&bma
);
4438 if (bma
.blkno
== NULLFSBLOCK
)
4442 * If this is a CoW allocation, record the data in
4443 * the refcount btree for orphan recovery.
4445 if (whichfork
== XFS_COW_FORK
)
4446 xfs_refcount_alloc_cow_extent(tp
, bma
.blkno
,
4450 /* Deal with the allocated space we found. */
4451 xfs_bmapi_trim_map(mval
, &bma
.got
, &bno
, len
, obno
,
4454 /* Execute unwritten extent conversion if necessary */
4455 error
= xfs_bmapi_convert_unwritten(&bma
, mval
, len
, flags
);
4456 if (error
== -EAGAIN
)
4461 /* update the extent map to return */
4462 xfs_bmapi_update_map(&mval
, &bno
, &len
, obno
, end
, &n
, flags
);
4465 * If we're done, stop now. Stop when we've allocated
4466 * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise
4467 * the transaction may get too big.
4469 if (bno
>= end
|| n
>= *nmap
|| bma
.nallocs
>= *nmap
)
4472 /* Else go on to the next record. */
4474 if (!xfs_iext_next_extent(ifp
, &bma
.icur
, &bma
.got
))
4479 error
= xfs_bmap_btree_to_extents(tp
, ip
, bma
.cur
, &bma
.logflags
,
4484 ASSERT(ifp
->if_format
!= XFS_DINODE_FMT_BTREE
||
4485 ifp
->if_nextents
> XFS_IFORK_MAXEXT(ip
, whichfork
));
4486 xfs_bmapi_finish(&bma
, whichfork
, 0);
4487 xfs_bmap_validate_ret(orig_bno
, orig_len
, orig_flags
, orig_mval
,
4491 xfs_bmapi_finish(&bma
, whichfork
, error
);
4496 * Convert an existing delalloc extent to real blocks based on file offset. This
4497 * attempts to allocate the entire delalloc extent and may require multiple
4498 * invocations to allocate the target offset if a large enough physical extent
4502 xfs_bmapi_convert_delalloc(
4503 struct xfs_inode
*ip
,
4506 struct iomap
*iomap
,
4509 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4510 struct xfs_mount
*mp
= ip
->i_mount
;
4511 xfs_fileoff_t offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
4512 struct xfs_bmalloca bma
= { NULL
};
4514 struct xfs_trans
*tp
;
4517 if (whichfork
== XFS_COW_FORK
)
4518 flags
|= IOMAP_F_SHARED
;
4521 * Space for the extent and indirect blocks was reserved when the
4522 * delalloc extent was created so there's no need to do so here.
4524 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_write
, 0, 0,
4525 XFS_TRANS_RESERVE
, &tp
);
4529 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
4530 xfs_trans_ijoin(tp
, ip
, 0);
4532 if (!xfs_iext_lookup_extent(ip
, ifp
, offset_fsb
, &bma
.icur
, &bma
.got
) ||
4533 bma
.got
.br_startoff
> offset_fsb
) {
4535 * No extent found in the range we are trying to convert. This
4536 * should only happen for the COW fork, where another thread
4537 * might have moved the extent to the data fork in the meantime.
4539 WARN_ON_ONCE(whichfork
!= XFS_COW_FORK
);
4541 goto out_trans_cancel
;
4545 * If we find a real extent here we raced with another thread converting
4546 * the extent. Just return the real extent at this offset.
4548 if (!isnullstartblock(bma
.got
.br_startblock
)) {
4549 xfs_bmbt_to_iomap(ip
, iomap
, &bma
.got
, flags
);
4550 *seq
= READ_ONCE(ifp
->if_seq
);
4551 goto out_trans_cancel
;
4557 bma
.offset
= bma
.got
.br_startoff
;
4558 bma
.length
= max_t(xfs_filblks_t
, bma
.got
.br_blockcount
, MAXEXTLEN
);
4559 bma
.minleft
= xfs_bmapi_minleft(tp
, ip
, whichfork
);
4562 * When we're converting the delalloc reservations backing dirty pages
4563 * in the page cache, we must be careful about how we create the new
4566 * New CoW fork extents are created unwritten, turned into real extents
4567 * when we're about to write the data to disk, and mapped into the data
4568 * fork after the write finishes. End of story.
4570 * New data fork extents must be mapped in as unwritten and converted
4571 * to real extents after the write succeeds to avoid exposing stale
4572 * disk contents if we crash.
4574 bma
.flags
= XFS_BMAPI_PREALLOC
;
4575 if (whichfork
== XFS_COW_FORK
)
4576 bma
.flags
|= XFS_BMAPI_COWFORK
;
4578 if (!xfs_iext_peek_prev_extent(ifp
, &bma
.icur
, &bma
.prev
))
4579 bma
.prev
.br_startoff
= NULLFILEOFF
;
4581 error
= xfs_bmapi_allocate(&bma
);
4586 if (WARN_ON_ONCE(bma
.blkno
== NULLFSBLOCK
))
4588 error
= -EFSCORRUPTED
;
4589 if (WARN_ON_ONCE(!xfs_valid_startblock(ip
, bma
.got
.br_startblock
)))
4592 XFS_STATS_ADD(mp
, xs_xstrat_bytes
, XFS_FSB_TO_B(mp
, bma
.length
));
4593 XFS_STATS_INC(mp
, xs_xstrat_quick
);
4595 ASSERT(!isnullstartblock(bma
.got
.br_startblock
));
4596 xfs_bmbt_to_iomap(ip
, iomap
, &bma
.got
, flags
);
4597 *seq
= READ_ONCE(ifp
->if_seq
);
4599 if (whichfork
== XFS_COW_FORK
)
4600 xfs_refcount_alloc_cow_extent(tp
, bma
.blkno
, bma
.length
);
4602 error
= xfs_bmap_btree_to_extents(tp
, ip
, bma
.cur
, &bma
.logflags
,
4607 xfs_bmapi_finish(&bma
, whichfork
, 0);
4608 error
= xfs_trans_commit(tp
);
4609 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
4613 xfs_bmapi_finish(&bma
, whichfork
, error
);
4615 xfs_trans_cancel(tp
);
4616 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
4622 struct xfs_trans
*tp
,
4623 struct xfs_inode
*ip
,
4626 xfs_fsblock_t startblock
,
4629 struct xfs_mount
*mp
= ip
->i_mount
;
4630 struct xfs_ifork
*ifp
;
4631 struct xfs_btree_cur
*cur
= NULL
;
4632 struct xfs_bmbt_irec got
;
4633 struct xfs_iext_cursor icur
;
4634 int whichfork
= xfs_bmapi_whichfork(flags
);
4635 int logflags
= 0, error
;
4637 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4639 ASSERT(len
<= (xfs_filblks_t
)MAXEXTLEN
);
4640 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
4641 ASSERT(!(flags
& ~(XFS_BMAPI_ATTRFORK
| XFS_BMAPI_PREALLOC
|
4642 XFS_BMAPI_NORMAP
)));
4643 ASSERT((flags
& (XFS_BMAPI_ATTRFORK
| XFS_BMAPI_PREALLOC
)) !=
4644 (XFS_BMAPI_ATTRFORK
| XFS_BMAPI_PREALLOC
));
4646 if (XFS_IS_CORRUPT(mp
, !xfs_ifork_has_extents(ifp
)) ||
4647 XFS_TEST_ERROR(false, mp
, XFS_ERRTAG_BMAPIFORMAT
)) {
4648 return -EFSCORRUPTED
;
4651 if (XFS_FORCED_SHUTDOWN(mp
))
4654 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
4655 error
= xfs_iread_extents(tp
, ip
, whichfork
);
4660 if (xfs_iext_lookup_extent(ip
, ifp
, bno
, &icur
, &got
)) {
4661 /* make sure we only reflink into a hole. */
4662 ASSERT(got
.br_startoff
> bno
);
4663 ASSERT(got
.br_startoff
- bno
>= len
);
4666 ip
->i_d
.di_nblocks
+= len
;
4667 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
4669 if (ifp
->if_flags
& XFS_IFBROOT
) {
4670 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
4671 cur
->bc_ino
.flags
= 0;
4674 got
.br_startoff
= bno
;
4675 got
.br_startblock
= startblock
;
4676 got
.br_blockcount
= len
;
4677 if (flags
& XFS_BMAPI_PREALLOC
)
4678 got
.br_state
= XFS_EXT_UNWRITTEN
;
4680 got
.br_state
= XFS_EXT_NORM
;
4682 error
= xfs_bmap_add_extent_hole_real(tp
, ip
, whichfork
, &icur
,
4683 &cur
, &got
, &logflags
, flags
);
4687 error
= xfs_bmap_btree_to_extents(tp
, ip
, cur
, &logflags
, whichfork
);
4690 if (ip
->i_df
.if_format
!= XFS_DINODE_FMT_EXTENTS
)
4691 logflags
&= ~XFS_ILOG_DEXT
;
4692 else if (ip
->i_df
.if_format
!= XFS_DINODE_FMT_BTREE
)
4693 logflags
&= ~XFS_ILOG_DBROOT
;
4696 xfs_trans_log_inode(tp
, ip
, logflags
);
4698 xfs_btree_del_cursor(cur
, error
);
4703 * When a delalloc extent is split (e.g., due to a hole punch), the original
4704 * indlen reservation must be shared across the two new extents that are left
4707 * Given the original reservation and the worst case indlen for the two new
4708 * extents (as calculated by xfs_bmap_worst_indlen()), split the original
4709 * reservation fairly across the two new extents. If necessary, steal available
4710 * blocks from a deleted extent to make up a reservation deficiency (e.g., if
4711 * ores == 1). The number of stolen blocks is returned. The availability and
4712 * subsequent accounting of stolen blocks is the responsibility of the caller.
4714 static xfs_filblks_t
4715 xfs_bmap_split_indlen(
4716 xfs_filblks_t ores
, /* original res. */
4717 xfs_filblks_t
*indlen1
, /* ext1 worst indlen */
4718 xfs_filblks_t
*indlen2
, /* ext2 worst indlen */
4719 xfs_filblks_t avail
) /* stealable blocks */
4721 xfs_filblks_t len1
= *indlen1
;
4722 xfs_filblks_t len2
= *indlen2
;
4723 xfs_filblks_t nres
= len1
+ len2
; /* new total res. */
4724 xfs_filblks_t stolen
= 0;
4725 xfs_filblks_t resfactor
;
4728 * Steal as many blocks as we can to try and satisfy the worst case
4729 * indlen for both new extents.
4731 if (ores
< nres
&& avail
)
4732 stolen
= XFS_FILBLKS_MIN(nres
- ores
, avail
);
4735 /* nothing else to do if we've satisfied the new reservation */
4740 * We can't meet the total required reservation for the two extents.
4741 * Calculate the percent of the overall shortage between both extents
4742 * and apply this percentage to each of the requested indlen values.
4743 * This distributes the shortage fairly and reduces the chances that one
4744 * of the two extents is left with nothing when extents are repeatedly
4747 resfactor
= (ores
* 100);
4748 do_div(resfactor
, nres
);
4753 ASSERT(len1
+ len2
<= ores
);
4754 ASSERT(len1
< *indlen1
&& len2
< *indlen2
);
4757 * Hand out the remainder to each extent. If one of the two reservations
4758 * is zero, we want to make sure that one gets a block first. The loop
4759 * below starts with len1, so hand len2 a block right off the bat if it
4762 ores
-= (len1
+ len2
);
4763 ASSERT((*indlen1
- len1
) + (*indlen2
- len2
) >= ores
);
4764 if (ores
&& !len2
&& *indlen2
) {
4769 if (len1
< *indlen1
) {
4775 if (len2
< *indlen2
) {
4788 xfs_bmap_del_extent_delay(
4789 struct xfs_inode
*ip
,
4791 struct xfs_iext_cursor
*icur
,
4792 struct xfs_bmbt_irec
*got
,
4793 struct xfs_bmbt_irec
*del
)
4795 struct xfs_mount
*mp
= ip
->i_mount
;
4796 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4797 struct xfs_bmbt_irec
new;
4798 int64_t da_old
, da_new
, da_diff
= 0;
4799 xfs_fileoff_t del_endoff
, got_endoff
;
4800 xfs_filblks_t got_indlen
, new_indlen
, stolen
;
4801 int state
= xfs_bmap_fork_to_state(whichfork
);
4805 XFS_STATS_INC(mp
, xs_del_exlist
);
4807 isrt
= (whichfork
== XFS_DATA_FORK
) && XFS_IS_REALTIME_INODE(ip
);
4808 del_endoff
= del
->br_startoff
+ del
->br_blockcount
;
4809 got_endoff
= got
->br_startoff
+ got
->br_blockcount
;
4810 da_old
= startblockval(got
->br_startblock
);
4813 ASSERT(del
->br_blockcount
> 0);
4814 ASSERT(got
->br_startoff
<= del
->br_startoff
);
4815 ASSERT(got_endoff
>= del_endoff
);
4818 uint64_t rtexts
= XFS_FSB_TO_B(mp
, del
->br_blockcount
);
4820 do_div(rtexts
, mp
->m_sb
.sb_rextsize
);
4821 xfs_mod_frextents(mp
, rtexts
);
4825 * Update the inode delalloc counter now and wait to update the
4826 * sb counters as we might have to borrow some blocks for the
4827 * indirect block accounting.
4829 error
= xfs_trans_reserve_quota_nblks(NULL
, ip
,
4830 -((long)del
->br_blockcount
), 0,
4831 isrt
? XFS_QMOPT_RES_RTBLKS
: XFS_QMOPT_RES_REGBLKS
);
4834 ip
->i_delayed_blks
-= del
->br_blockcount
;
4836 if (got
->br_startoff
== del
->br_startoff
)
4837 state
|= BMAP_LEFT_FILLING
;
4838 if (got_endoff
== del_endoff
)
4839 state
|= BMAP_RIGHT_FILLING
;
4841 switch (state
& (BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
)) {
4842 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
:
4844 * Matches the whole extent. Delete the entry.
4846 xfs_iext_remove(ip
, icur
, state
);
4847 xfs_iext_prev(ifp
, icur
);
4849 case BMAP_LEFT_FILLING
:
4851 * Deleting the first part of the extent.
4853 got
->br_startoff
= del_endoff
;
4854 got
->br_blockcount
-= del
->br_blockcount
;
4855 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
,
4856 got
->br_blockcount
), da_old
);
4857 got
->br_startblock
= nullstartblock((int)da_new
);
4858 xfs_iext_update_extent(ip
, state
, icur
, got
);
4860 case BMAP_RIGHT_FILLING
:
4862 * Deleting the last part of the extent.
4864 got
->br_blockcount
= got
->br_blockcount
- del
->br_blockcount
;
4865 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
,
4866 got
->br_blockcount
), da_old
);
4867 got
->br_startblock
= nullstartblock((int)da_new
);
4868 xfs_iext_update_extent(ip
, state
, icur
, got
);
4872 * Deleting the middle of the extent.
4874 * Distribute the original indlen reservation across the two new
4875 * extents. Steal blocks from the deleted extent if necessary.
4876 * Stealing blocks simply fudges the fdblocks accounting below.
4877 * Warn if either of the new indlen reservations is zero as this
4878 * can lead to delalloc problems.
4880 got
->br_blockcount
= del
->br_startoff
- got
->br_startoff
;
4881 got_indlen
= xfs_bmap_worst_indlen(ip
, got
->br_blockcount
);
4883 new.br_blockcount
= got_endoff
- del_endoff
;
4884 new_indlen
= xfs_bmap_worst_indlen(ip
, new.br_blockcount
);
4886 WARN_ON_ONCE(!got_indlen
|| !new_indlen
);
4887 stolen
= xfs_bmap_split_indlen(da_old
, &got_indlen
, &new_indlen
,
4888 del
->br_blockcount
);
4890 got
->br_startblock
= nullstartblock((int)got_indlen
);
4892 new.br_startoff
= del_endoff
;
4893 new.br_state
= got
->br_state
;
4894 new.br_startblock
= nullstartblock((int)new_indlen
);
4896 xfs_iext_update_extent(ip
, state
, icur
, got
);
4897 xfs_iext_next(ifp
, icur
);
4898 xfs_iext_insert(ip
, icur
, &new, state
);
4900 da_new
= got_indlen
+ new_indlen
- stolen
;
4901 del
->br_blockcount
-= stolen
;
4905 ASSERT(da_old
>= da_new
);
4906 da_diff
= da_old
- da_new
;
4908 da_diff
+= del
->br_blockcount
;
4910 xfs_mod_fdblocks(mp
, da_diff
, false);
4911 xfs_mod_delalloc(mp
, -da_diff
);
4917 xfs_bmap_del_extent_cow(
4918 struct xfs_inode
*ip
,
4919 struct xfs_iext_cursor
*icur
,
4920 struct xfs_bmbt_irec
*got
,
4921 struct xfs_bmbt_irec
*del
)
4923 struct xfs_mount
*mp
= ip
->i_mount
;
4924 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, XFS_COW_FORK
);
4925 struct xfs_bmbt_irec
new;
4926 xfs_fileoff_t del_endoff
, got_endoff
;
4927 int state
= BMAP_COWFORK
;
4929 XFS_STATS_INC(mp
, xs_del_exlist
);
4931 del_endoff
= del
->br_startoff
+ del
->br_blockcount
;
4932 got_endoff
= got
->br_startoff
+ got
->br_blockcount
;
4934 ASSERT(del
->br_blockcount
> 0);
4935 ASSERT(got
->br_startoff
<= del
->br_startoff
);
4936 ASSERT(got_endoff
>= del_endoff
);
4937 ASSERT(!isnullstartblock(got
->br_startblock
));
4939 if (got
->br_startoff
== del
->br_startoff
)
4940 state
|= BMAP_LEFT_FILLING
;
4941 if (got_endoff
== del_endoff
)
4942 state
|= BMAP_RIGHT_FILLING
;
4944 switch (state
& (BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
)) {
4945 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
:
4947 * Matches the whole extent. Delete the entry.
4949 xfs_iext_remove(ip
, icur
, state
);
4950 xfs_iext_prev(ifp
, icur
);
4952 case BMAP_LEFT_FILLING
:
4954 * Deleting the first part of the extent.
4956 got
->br_startoff
= del_endoff
;
4957 got
->br_blockcount
-= del
->br_blockcount
;
4958 got
->br_startblock
= del
->br_startblock
+ del
->br_blockcount
;
4959 xfs_iext_update_extent(ip
, state
, icur
, got
);
4961 case BMAP_RIGHT_FILLING
:
4963 * Deleting the last part of the extent.
4965 got
->br_blockcount
-= del
->br_blockcount
;
4966 xfs_iext_update_extent(ip
, state
, icur
, got
);
4970 * Deleting the middle of the extent.
4972 got
->br_blockcount
= del
->br_startoff
- got
->br_startoff
;
4974 new.br_startoff
= del_endoff
;
4975 new.br_blockcount
= got_endoff
- del_endoff
;
4976 new.br_state
= got
->br_state
;
4977 new.br_startblock
= del
->br_startblock
+ del
->br_blockcount
;
4979 xfs_iext_update_extent(ip
, state
, icur
, got
);
4980 xfs_iext_next(ifp
, icur
);
4981 xfs_iext_insert(ip
, icur
, &new, state
);
4984 ip
->i_delayed_blks
-= del
->br_blockcount
;
4988 * Called by xfs_bmapi to update file extent records and the btree
4989 * after removing space.
4991 STATIC
int /* error */
4992 xfs_bmap_del_extent_real(
4993 xfs_inode_t
*ip
, /* incore inode pointer */
4994 xfs_trans_t
*tp
, /* current transaction pointer */
4995 struct xfs_iext_cursor
*icur
,
4996 xfs_btree_cur_t
*cur
, /* if null, not a btree */
4997 xfs_bmbt_irec_t
*del
, /* data to remove from extents */
4998 int *logflagsp
, /* inode logging flags */
4999 int whichfork
, /* data or attr fork */
5000 int bflags
) /* bmapi flags */
5002 xfs_fsblock_t del_endblock
=0; /* first block past del */
5003 xfs_fileoff_t del_endoff
; /* first offset past del */
5004 int do_fx
; /* free extent at end of routine */
5005 int error
; /* error return value */
5006 int flags
= 0;/* inode logging flags */
5007 struct xfs_bmbt_irec got
; /* current extent entry */
5008 xfs_fileoff_t got_endoff
; /* first offset past got */
5009 int i
; /* temp state */
5010 struct xfs_ifork
*ifp
; /* inode fork pointer */
5011 xfs_mount_t
*mp
; /* mount structure */
5012 xfs_filblks_t nblks
; /* quota/sb block count */
5013 xfs_bmbt_irec_t
new; /* new record to be inserted */
5015 uint qfield
; /* quota field to update */
5016 int state
= xfs_bmap_fork_to_state(whichfork
);
5017 struct xfs_bmbt_irec old
;
5020 XFS_STATS_INC(mp
, xs_del_exlist
);
5022 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
5023 ASSERT(del
->br_blockcount
> 0);
5024 xfs_iext_get_extent(ifp
, icur
, &got
);
5025 ASSERT(got
.br_startoff
<= del
->br_startoff
);
5026 del_endoff
= del
->br_startoff
+ del
->br_blockcount
;
5027 got_endoff
= got
.br_startoff
+ got
.br_blockcount
;
5028 ASSERT(got_endoff
>= del_endoff
);
5029 ASSERT(!isnullstartblock(got
.br_startblock
));
5034 * If it's the case where the directory code is running with no block
5035 * reservation, and the deleted block is in the middle of its extent,
5036 * and the resulting insert of an extent would cause transformation to
5037 * btree format, then reject it. The calling code will then swap blocks
5038 * around instead. We have to do this now, rather than waiting for the
5039 * conversion to btree format, since the transaction will be dirty then.
5041 if (tp
->t_blk_res
== 0 &&
5042 ifp
->if_format
== XFS_DINODE_FMT_EXTENTS
&&
5043 ifp
->if_nextents
>= XFS_IFORK_MAXEXT(ip
, whichfork
) &&
5044 del
->br_startoff
> got
.br_startoff
&& del_endoff
< got_endoff
)
5047 flags
= XFS_ILOG_CORE
;
5048 if (whichfork
== XFS_DATA_FORK
&& XFS_IS_REALTIME_INODE(ip
)) {
5052 len
= div_u64_rem(del
->br_blockcount
, mp
->m_sb
.sb_rextsize
,
5056 if (!(bflags
& XFS_BMAPI_REMAP
)) {
5059 bno
= div_u64_rem(del
->br_startblock
,
5060 mp
->m_sb
.sb_rextsize
, &mod
);
5063 error
= xfs_rtfree_extent(tp
, bno
, (xfs_extlen_t
)len
);
5069 nblks
= len
* mp
->m_sb
.sb_rextsize
;
5070 qfield
= XFS_TRANS_DQ_RTBCOUNT
;
5073 nblks
= del
->br_blockcount
;
5074 qfield
= XFS_TRANS_DQ_BCOUNT
;
5077 del_endblock
= del
->br_startblock
+ del
->br_blockcount
;
5079 error
= xfs_bmbt_lookup_eq(cur
, &got
, &i
);
5082 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
5083 error
= -EFSCORRUPTED
;
5088 if (got
.br_startoff
== del
->br_startoff
)
5089 state
|= BMAP_LEFT_FILLING
;
5090 if (got_endoff
== del_endoff
)
5091 state
|= BMAP_RIGHT_FILLING
;
5093 switch (state
& (BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
)) {
5094 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
:
5096 * Matches the whole extent. Delete the entry.
5098 xfs_iext_remove(ip
, icur
, state
);
5099 xfs_iext_prev(ifp
, icur
);
5102 flags
|= XFS_ILOG_CORE
;
5104 flags
|= xfs_ilog_fext(whichfork
);
5107 if ((error
= xfs_btree_delete(cur
, &i
)))
5109 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
5110 error
= -EFSCORRUPTED
;
5114 case BMAP_LEFT_FILLING
:
5116 * Deleting the first part of the extent.
5118 got
.br_startoff
= del_endoff
;
5119 got
.br_startblock
= del_endblock
;
5120 got
.br_blockcount
-= del
->br_blockcount
;
5121 xfs_iext_update_extent(ip
, state
, icur
, &got
);
5123 flags
|= xfs_ilog_fext(whichfork
);
5126 error
= xfs_bmbt_update(cur
, &got
);
5130 case BMAP_RIGHT_FILLING
:
5132 * Deleting the last part of the extent.
5134 got
.br_blockcount
-= del
->br_blockcount
;
5135 xfs_iext_update_extent(ip
, state
, icur
, &got
);
5137 flags
|= xfs_ilog_fext(whichfork
);
5140 error
= xfs_bmbt_update(cur
, &got
);
5146 * Deleting the middle of the extent.
5150 got
.br_blockcount
= del
->br_startoff
- got
.br_startoff
;
5151 xfs_iext_update_extent(ip
, state
, icur
, &got
);
5153 new.br_startoff
= del_endoff
;
5154 new.br_blockcount
= got_endoff
- del_endoff
;
5155 new.br_state
= got
.br_state
;
5156 new.br_startblock
= del_endblock
;
5158 flags
|= XFS_ILOG_CORE
;
5160 error
= xfs_bmbt_update(cur
, &got
);
5163 error
= xfs_btree_increment(cur
, 0, &i
);
5166 cur
->bc_rec
.b
= new;
5167 error
= xfs_btree_insert(cur
, &i
);
5168 if (error
&& error
!= -ENOSPC
)
5171 * If get no-space back from btree insert, it tried a
5172 * split, and we have a zero block reservation. Fix up
5173 * our state and return the error.
5175 if (error
== -ENOSPC
) {
5177 * Reset the cursor, don't trust it after any
5180 error
= xfs_bmbt_lookup_eq(cur
, &got
, &i
);
5183 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
5184 error
= -EFSCORRUPTED
;
5188 * Update the btree record back
5189 * to the original value.
5191 error
= xfs_bmbt_update(cur
, &old
);
5195 * Reset the extent record back
5196 * to the original value.
5198 xfs_iext_update_extent(ip
, state
, icur
, &old
);
5203 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
5204 error
= -EFSCORRUPTED
;
5208 flags
|= xfs_ilog_fext(whichfork
);
5211 xfs_iext_next(ifp
, icur
);
5212 xfs_iext_insert(ip
, icur
, &new, state
);
5216 /* remove reverse mapping */
5217 xfs_rmap_unmap_extent(tp
, ip
, whichfork
, del
);
5220 * If we need to, add to list of extents to delete.
5222 if (do_fx
&& !(bflags
& XFS_BMAPI_REMAP
)) {
5223 if (xfs_is_reflink_inode(ip
) && whichfork
== XFS_DATA_FORK
) {
5224 xfs_refcount_decrease_extent(tp
, del
);
5226 __xfs_bmap_add_free(tp
, del
->br_startblock
,
5227 del
->br_blockcount
, NULL
,
5228 (bflags
& XFS_BMAPI_NODISCARD
) ||
5229 del
->br_state
== XFS_EXT_UNWRITTEN
);
5234 * Adjust inode # blocks in the file.
5237 ip
->i_d
.di_nblocks
-= nblks
;
5239 * Adjust quota data.
5241 if (qfield
&& !(bflags
& XFS_BMAPI_REMAP
))
5242 xfs_trans_mod_dquot_byino(tp
, ip
, qfield
, (long)-nblks
);
5250 * Unmap (remove) blocks from a file.
5251 * If nexts is nonzero then the number of extents to remove is limited to
5252 * that value. If not all extents in the block range can be removed then
5257 struct xfs_trans
*tp
, /* transaction pointer */
5258 struct xfs_inode
*ip
, /* incore inode */
5259 xfs_fileoff_t start
, /* first file offset deleted */
5260 xfs_filblks_t
*rlen
, /* i/o: amount remaining */
5261 int flags
, /* misc flags */
5262 xfs_extnum_t nexts
) /* number of extents max */
5264 struct xfs_btree_cur
*cur
; /* bmap btree cursor */
5265 struct xfs_bmbt_irec del
; /* extent being deleted */
5266 int error
; /* error return value */
5267 xfs_extnum_t extno
; /* extent number in list */
5268 struct xfs_bmbt_irec got
; /* current extent record */
5269 struct xfs_ifork
*ifp
; /* inode fork pointer */
5270 int isrt
; /* freeing in rt area */
5271 int logflags
; /* transaction logging flags */
5272 xfs_extlen_t mod
; /* rt extent offset */
5273 struct xfs_mount
*mp
= ip
->i_mount
;
5274 int tmp_logflags
; /* partial logging flags */
5275 int wasdel
; /* was a delayed alloc extent */
5276 int whichfork
; /* data or attribute fork */
5278 xfs_filblks_t len
= *rlen
; /* length to unmap in file */
5279 xfs_fileoff_t max_len
;
5280 xfs_agnumber_t prev_agno
= NULLAGNUMBER
, agno
;
5282 struct xfs_iext_cursor icur
;
5285 trace_xfs_bunmap(ip
, start
, len
, flags
, _RET_IP_
);
5287 whichfork
= xfs_bmapi_whichfork(flags
);
5288 ASSERT(whichfork
!= XFS_COW_FORK
);
5289 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
5290 if (XFS_IS_CORRUPT(mp
, !xfs_ifork_has_extents(ifp
)))
5291 return -EFSCORRUPTED
;
5292 if (XFS_FORCED_SHUTDOWN(mp
))
5295 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
5300 * Guesstimate how many blocks we can unmap without running the risk of
5301 * blowing out the transaction with a mix of EFIs and reflink
5304 if (tp
&& xfs_is_reflink_inode(ip
) && whichfork
== XFS_DATA_FORK
)
5305 max_len
= min(len
, xfs_refcount_max_unmap(tp
->t_log_res
));
5309 if (!(ifp
->if_flags
& XFS_IFEXTENTS
) &&
5310 (error
= xfs_iread_extents(tp
, ip
, whichfork
)))
5312 if (xfs_iext_count(ifp
) == 0) {
5316 XFS_STATS_INC(mp
, xs_blk_unmap
);
5317 isrt
= (whichfork
== XFS_DATA_FORK
) && XFS_IS_REALTIME_INODE(ip
);
5320 if (!xfs_iext_lookup_extent_before(ip
, ifp
, &end
, &icur
, &got
)) {
5327 if (ifp
->if_flags
& XFS_IFBROOT
) {
5328 ASSERT(ifp
->if_format
== XFS_DINODE_FMT_BTREE
);
5329 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
5330 cur
->bc_ino
.flags
= 0;
5336 * Synchronize by locking the bitmap inode.
5338 xfs_ilock(mp
->m_rbmip
, XFS_ILOCK_EXCL
|XFS_ILOCK_RTBITMAP
);
5339 xfs_trans_ijoin(tp
, mp
->m_rbmip
, XFS_ILOCK_EXCL
);
5340 xfs_ilock(mp
->m_rsumip
, XFS_ILOCK_EXCL
|XFS_ILOCK_RTSUM
);
5341 xfs_trans_ijoin(tp
, mp
->m_rsumip
, XFS_ILOCK_EXCL
);
5345 while (end
!= (xfs_fileoff_t
)-1 && end
>= start
&&
5346 (nexts
== 0 || extno
< nexts
) && max_len
> 0) {
5348 * Is the found extent after a hole in which end lives?
5349 * Just back up to the previous extent, if so.
5351 if (got
.br_startoff
> end
&&
5352 !xfs_iext_prev_extent(ifp
, &icur
, &got
)) {
5357 * Is the last block of this extent before the range
5358 * we're supposed to delete? If so, we're done.
5360 end
= XFS_FILEOFF_MIN(end
,
5361 got
.br_startoff
+ got
.br_blockcount
- 1);
5365 * Then deal with the (possibly delayed) allocated space
5369 wasdel
= isnullstartblock(del
.br_startblock
);
5372 * Make sure we don't touch multiple AGF headers out of order
5373 * in a single transaction, as that could cause AB-BA deadlocks.
5375 if (!wasdel
&& !isrt
) {
5376 agno
= XFS_FSB_TO_AGNO(mp
, del
.br_startblock
);
5377 if (prev_agno
!= NULLAGNUMBER
&& prev_agno
> agno
)
5381 if (got
.br_startoff
< start
) {
5382 del
.br_startoff
= start
;
5383 del
.br_blockcount
-= start
- got
.br_startoff
;
5385 del
.br_startblock
+= start
- got
.br_startoff
;
5387 if (del
.br_startoff
+ del
.br_blockcount
> end
+ 1)
5388 del
.br_blockcount
= end
+ 1 - del
.br_startoff
;
5390 /* How much can we safely unmap? */
5391 if (max_len
< del
.br_blockcount
) {
5392 del
.br_startoff
+= del
.br_blockcount
- max_len
;
5394 del
.br_startblock
+= del
.br_blockcount
- max_len
;
5395 del
.br_blockcount
= max_len
;
5401 sum
= del
.br_startblock
+ del
.br_blockcount
;
5402 div_u64_rem(sum
, mp
->m_sb
.sb_rextsize
, &mod
);
5405 * Realtime extent not lined up at the end.
5406 * The extent could have been split into written
5407 * and unwritten pieces, or we could just be
5408 * unmapping part of it. But we can't really
5409 * get rid of part of a realtime extent.
5411 if (del
.br_state
== XFS_EXT_UNWRITTEN
) {
5413 * This piece is unwritten, or we're not
5414 * using unwritten extents. Skip over it.
5417 end
-= mod
> del
.br_blockcount
?
5418 del
.br_blockcount
: mod
;
5419 if (end
< got
.br_startoff
&&
5420 !xfs_iext_prev_extent(ifp
, &icur
, &got
)) {
5427 * It's written, turn it unwritten.
5428 * This is better than zeroing it.
5430 ASSERT(del
.br_state
== XFS_EXT_NORM
);
5431 ASSERT(tp
->t_blk_res
> 0);
5433 * If this spans a realtime extent boundary,
5434 * chop it back to the start of the one we end at.
5436 if (del
.br_blockcount
> mod
) {
5437 del
.br_startoff
+= del
.br_blockcount
- mod
;
5438 del
.br_startblock
+= del
.br_blockcount
- mod
;
5439 del
.br_blockcount
= mod
;
5441 del
.br_state
= XFS_EXT_UNWRITTEN
;
5442 error
= xfs_bmap_add_extent_unwritten_real(tp
, ip
,
5443 whichfork
, &icur
, &cur
, &del
,
5449 div_u64_rem(del
.br_startblock
, mp
->m_sb
.sb_rextsize
, &mod
);
5451 xfs_extlen_t off
= mp
->m_sb
.sb_rextsize
- mod
;
5454 * Realtime extent is lined up at the end but not
5455 * at the front. We'll get rid of full extents if
5458 if (del
.br_blockcount
> off
) {
5459 del
.br_blockcount
-= off
;
5460 del
.br_startoff
+= off
;
5461 del
.br_startblock
+= off
;
5462 } else if (del
.br_startoff
== start
&&
5463 (del
.br_state
== XFS_EXT_UNWRITTEN
||
5464 tp
->t_blk_res
== 0)) {
5466 * Can't make it unwritten. There isn't
5467 * a full extent here so just skip it.
5469 ASSERT(end
>= del
.br_blockcount
);
5470 end
-= del
.br_blockcount
;
5471 if (got
.br_startoff
> end
&&
5472 !xfs_iext_prev_extent(ifp
, &icur
, &got
)) {
5477 } else if (del
.br_state
== XFS_EXT_UNWRITTEN
) {
5478 struct xfs_bmbt_irec prev
;
5479 xfs_fileoff_t unwrite_start
;
5482 * This one is already unwritten.
5483 * It must have a written left neighbor.
5484 * Unwrite the killed part of that one and
5487 if (!xfs_iext_prev_extent(ifp
, &icur
, &prev
))
5489 ASSERT(prev
.br_state
== XFS_EXT_NORM
);
5490 ASSERT(!isnullstartblock(prev
.br_startblock
));
5491 ASSERT(del
.br_startblock
==
5492 prev
.br_startblock
+ prev
.br_blockcount
);
5493 unwrite_start
= max3(start
,
5494 del
.br_startoff
- mod
,
5496 mod
= unwrite_start
- prev
.br_startoff
;
5497 prev
.br_startoff
= unwrite_start
;
5498 prev
.br_startblock
+= mod
;
5499 prev
.br_blockcount
-= mod
;
5500 prev
.br_state
= XFS_EXT_UNWRITTEN
;
5501 error
= xfs_bmap_add_extent_unwritten_real(tp
,
5502 ip
, whichfork
, &icur
, &cur
,
5508 ASSERT(del
.br_state
== XFS_EXT_NORM
);
5509 del
.br_state
= XFS_EXT_UNWRITTEN
;
5510 error
= xfs_bmap_add_extent_unwritten_real(tp
,
5511 ip
, whichfork
, &icur
, &cur
,
5521 error
= xfs_bmap_del_extent_delay(ip
, whichfork
, &icur
,
5524 error
= xfs_bmap_del_extent_real(ip
, tp
, &icur
, cur
,
5525 &del
, &tmp_logflags
, whichfork
,
5527 logflags
|= tmp_logflags
;
5533 max_len
-= del
.br_blockcount
;
5534 end
= del
.br_startoff
- 1;
5537 * If not done go on to the next (previous) record.
5539 if (end
!= (xfs_fileoff_t
)-1 && end
>= start
) {
5540 if (!xfs_iext_get_extent(ifp
, &icur
, &got
) ||
5541 (got
.br_startoff
> end
&&
5542 !xfs_iext_prev_extent(ifp
, &icur
, &got
))) {
5549 if (done
|| end
== (xfs_fileoff_t
)-1 || end
< start
)
5552 *rlen
= end
- start
+ 1;
5555 * Convert to a btree if necessary.
5557 if (xfs_bmap_needs_btree(ip
, whichfork
)) {
5558 ASSERT(cur
== NULL
);
5559 error
= xfs_bmap_extents_to_btree(tp
, ip
, &cur
, 0,
5560 &tmp_logflags
, whichfork
);
5561 logflags
|= tmp_logflags
;
5563 error
= xfs_bmap_btree_to_extents(tp
, ip
, cur
, &logflags
,
5569 * Log everything. Do this after conversion, there's no point in
5570 * logging the extent records if we've converted to btree format.
5572 if ((logflags
& xfs_ilog_fext(whichfork
)) &&
5573 ifp
->if_format
!= XFS_DINODE_FMT_EXTENTS
)
5574 logflags
&= ~xfs_ilog_fext(whichfork
);
5575 else if ((logflags
& xfs_ilog_fbroot(whichfork
)) &&
5576 ifp
->if_format
!= XFS_DINODE_FMT_BTREE
)
5577 logflags
&= ~xfs_ilog_fbroot(whichfork
);
5579 * Log inode even in the error case, if the transaction
5580 * is dirty we'll need to shut down the filesystem.
5583 xfs_trans_log_inode(tp
, ip
, logflags
);
5586 cur
->bc_ino
.allocated
= 0;
5587 xfs_btree_del_cursor(cur
, error
);
5592 /* Unmap a range of a file. */
5596 struct xfs_inode
*ip
,
5605 error
= __xfs_bunmapi(tp
, ip
, bno
, &len
, flags
, nexts
);
5611 * Determine whether an extent shift can be accomplished by a merge with the
5612 * extent that precedes the target hole of the shift.
5616 struct xfs_bmbt_irec
*left
, /* preceding extent */
5617 struct xfs_bmbt_irec
*got
, /* current extent to shift */
5618 xfs_fileoff_t shift
) /* shift fsb */
5620 xfs_fileoff_t startoff
;
5622 startoff
= got
->br_startoff
- shift
;
5625 * The extent, once shifted, must be adjacent in-file and on-disk with
5626 * the preceding extent.
5628 if ((left
->br_startoff
+ left
->br_blockcount
!= startoff
) ||
5629 (left
->br_startblock
+ left
->br_blockcount
!= got
->br_startblock
) ||
5630 (left
->br_state
!= got
->br_state
) ||
5631 (left
->br_blockcount
+ got
->br_blockcount
> MAXEXTLEN
))
5638 * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5639 * hole in the file. If an extent shift would result in the extent being fully
5640 * adjacent to the extent that currently precedes the hole, we can merge with
5641 * the preceding extent rather than do the shift.
5643 * This function assumes the caller has verified a shift-by-merge is possible
5644 * with the provided extents via xfs_bmse_can_merge().
5648 struct xfs_trans
*tp
,
5649 struct xfs_inode
*ip
,
5651 xfs_fileoff_t shift
, /* shift fsb */
5652 struct xfs_iext_cursor
*icur
,
5653 struct xfs_bmbt_irec
*got
, /* extent to shift */
5654 struct xfs_bmbt_irec
*left
, /* preceding extent */
5655 struct xfs_btree_cur
*cur
,
5656 int *logflags
) /* output */
5658 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
5659 struct xfs_bmbt_irec
new;
5660 xfs_filblks_t blockcount
;
5662 struct xfs_mount
*mp
= ip
->i_mount
;
5664 blockcount
= left
->br_blockcount
+ got
->br_blockcount
;
5666 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
5667 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
5668 ASSERT(xfs_bmse_can_merge(left
, got
, shift
));
5671 new.br_blockcount
= blockcount
;
5674 * Update the on-disk extent count, the btree if necessary and log the
5678 *logflags
|= XFS_ILOG_CORE
;
5680 *logflags
|= XFS_ILOG_DEXT
;
5684 /* lookup and remove the extent to merge */
5685 error
= xfs_bmbt_lookup_eq(cur
, got
, &i
);
5688 if (XFS_IS_CORRUPT(mp
, i
!= 1))
5689 return -EFSCORRUPTED
;
5691 error
= xfs_btree_delete(cur
, &i
);
5694 if (XFS_IS_CORRUPT(mp
, i
!= 1))
5695 return -EFSCORRUPTED
;
5697 /* lookup and update size of the previous extent */
5698 error
= xfs_bmbt_lookup_eq(cur
, left
, &i
);
5701 if (XFS_IS_CORRUPT(mp
, i
!= 1))
5702 return -EFSCORRUPTED
;
5704 error
= xfs_bmbt_update(cur
, &new);
5708 /* change to extent format if required after extent removal */
5709 error
= xfs_bmap_btree_to_extents(tp
, ip
, cur
, logflags
, whichfork
);
5714 xfs_iext_remove(ip
, icur
, 0);
5715 xfs_iext_prev(ifp
, icur
);
5716 xfs_iext_update_extent(ip
, xfs_bmap_fork_to_state(whichfork
), icur
,
5719 /* update reverse mapping. rmap functions merge the rmaps for us */
5720 xfs_rmap_unmap_extent(tp
, ip
, whichfork
, got
);
5721 memcpy(&new, got
, sizeof(new));
5722 new.br_startoff
= left
->br_startoff
+ left
->br_blockcount
;
5723 xfs_rmap_map_extent(tp
, ip
, whichfork
, &new);
5728 xfs_bmap_shift_update_extent(
5729 struct xfs_trans
*tp
,
5730 struct xfs_inode
*ip
,
5732 struct xfs_iext_cursor
*icur
,
5733 struct xfs_bmbt_irec
*got
,
5734 struct xfs_btree_cur
*cur
,
5736 xfs_fileoff_t startoff
)
5738 struct xfs_mount
*mp
= ip
->i_mount
;
5739 struct xfs_bmbt_irec prev
= *got
;
5742 *logflags
|= XFS_ILOG_CORE
;
5744 got
->br_startoff
= startoff
;
5747 error
= xfs_bmbt_lookup_eq(cur
, &prev
, &i
);
5750 if (XFS_IS_CORRUPT(mp
, i
!= 1))
5751 return -EFSCORRUPTED
;
5753 error
= xfs_bmbt_update(cur
, got
);
5757 *logflags
|= XFS_ILOG_DEXT
;
5760 xfs_iext_update_extent(ip
, xfs_bmap_fork_to_state(whichfork
), icur
,
5763 /* update reverse mapping */
5764 xfs_rmap_unmap_extent(tp
, ip
, whichfork
, &prev
);
5765 xfs_rmap_map_extent(tp
, ip
, whichfork
, got
);
5770 xfs_bmap_collapse_extents(
5771 struct xfs_trans
*tp
,
5772 struct xfs_inode
*ip
,
5773 xfs_fileoff_t
*next_fsb
,
5774 xfs_fileoff_t offset_shift_fsb
,
5777 int whichfork
= XFS_DATA_FORK
;
5778 struct xfs_mount
*mp
= ip
->i_mount
;
5779 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
5780 struct xfs_btree_cur
*cur
= NULL
;
5781 struct xfs_bmbt_irec got
, prev
;
5782 struct xfs_iext_cursor icur
;
5783 xfs_fileoff_t new_startoff
;
5787 if (XFS_IS_CORRUPT(mp
, !xfs_ifork_has_extents(ifp
)) ||
5788 XFS_TEST_ERROR(false, mp
, XFS_ERRTAG_BMAPIFORMAT
)) {
5789 return -EFSCORRUPTED
;
5792 if (XFS_FORCED_SHUTDOWN(mp
))
5795 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
));
5797 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
5798 error
= xfs_iread_extents(tp
, ip
, whichfork
);
5803 if (ifp
->if_flags
& XFS_IFBROOT
) {
5804 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
5805 cur
->bc_ino
.flags
= 0;
5808 if (!xfs_iext_lookup_extent(ip
, ifp
, *next_fsb
, &icur
, &got
)) {
5812 if (XFS_IS_CORRUPT(mp
, isnullstartblock(got
.br_startblock
))) {
5813 error
= -EFSCORRUPTED
;
5817 new_startoff
= got
.br_startoff
- offset_shift_fsb
;
5818 if (xfs_iext_peek_prev_extent(ifp
, &icur
, &prev
)) {
5819 if (new_startoff
< prev
.br_startoff
+ prev
.br_blockcount
) {
5824 if (xfs_bmse_can_merge(&prev
, &got
, offset_shift_fsb
)) {
5825 error
= xfs_bmse_merge(tp
, ip
, whichfork
,
5826 offset_shift_fsb
, &icur
, &got
, &prev
,
5833 if (got
.br_startoff
< offset_shift_fsb
) {
5839 error
= xfs_bmap_shift_update_extent(tp
, ip
, whichfork
, &icur
, &got
,
5840 cur
, &logflags
, new_startoff
);
5845 if (!xfs_iext_next_extent(ifp
, &icur
, &got
)) {
5850 *next_fsb
= got
.br_startoff
;
5853 xfs_btree_del_cursor(cur
, error
);
5855 xfs_trans_log_inode(tp
, ip
, logflags
);
5859 /* Make sure we won't be right-shifting an extent past the maximum bound. */
5861 xfs_bmap_can_insert_extents(
5862 struct xfs_inode
*ip
,
5864 xfs_fileoff_t shift
)
5866 struct xfs_bmbt_irec got
;
5870 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
5872 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
5875 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
5876 error
= xfs_bmap_last_extent(NULL
, ip
, XFS_DATA_FORK
, &got
, &is_empty
);
5877 if (!error
&& !is_empty
&& got
.br_startoff
>= off
&&
5878 ((got
.br_startoff
+ shift
) & BMBT_STARTOFF_MASK
) < got
.br_startoff
)
5880 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
5886 xfs_bmap_insert_extents(
5887 struct xfs_trans
*tp
,
5888 struct xfs_inode
*ip
,
5889 xfs_fileoff_t
*next_fsb
,
5890 xfs_fileoff_t offset_shift_fsb
,
5892 xfs_fileoff_t stop_fsb
)
5894 int whichfork
= XFS_DATA_FORK
;
5895 struct xfs_mount
*mp
= ip
->i_mount
;
5896 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
5897 struct xfs_btree_cur
*cur
= NULL
;
5898 struct xfs_bmbt_irec got
, next
;
5899 struct xfs_iext_cursor icur
;
5900 xfs_fileoff_t new_startoff
;
5904 if (XFS_IS_CORRUPT(mp
, !xfs_ifork_has_extents(ifp
)) ||
5905 XFS_TEST_ERROR(false, mp
, XFS_ERRTAG_BMAPIFORMAT
)) {
5906 return -EFSCORRUPTED
;
5909 if (XFS_FORCED_SHUTDOWN(mp
))
5912 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
));
5914 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
5915 error
= xfs_iread_extents(tp
, ip
, whichfork
);
5920 if (ifp
->if_flags
& XFS_IFBROOT
) {
5921 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
5922 cur
->bc_ino
.flags
= 0;
5925 if (*next_fsb
== NULLFSBLOCK
) {
5926 xfs_iext_last(ifp
, &icur
);
5927 if (!xfs_iext_get_extent(ifp
, &icur
, &got
) ||
5928 stop_fsb
> got
.br_startoff
) {
5933 if (!xfs_iext_lookup_extent(ip
, ifp
, *next_fsb
, &icur
, &got
)) {
5938 if (XFS_IS_CORRUPT(mp
, isnullstartblock(got
.br_startblock
))) {
5939 error
= -EFSCORRUPTED
;
5943 if (XFS_IS_CORRUPT(mp
, stop_fsb
> got
.br_startoff
)) {
5944 error
= -EFSCORRUPTED
;
5948 new_startoff
= got
.br_startoff
+ offset_shift_fsb
;
5949 if (xfs_iext_peek_next_extent(ifp
, &icur
, &next
)) {
5950 if (new_startoff
+ got
.br_blockcount
> next
.br_startoff
) {
5956 * Unlike a left shift (which involves a hole punch), a right
5957 * shift does not modify extent neighbors in any way. We should
5958 * never find mergeable extents in this scenario. Check anyways
5959 * and warn if we encounter two extents that could be one.
5961 if (xfs_bmse_can_merge(&got
, &next
, offset_shift_fsb
))
5965 error
= xfs_bmap_shift_update_extent(tp
, ip
, whichfork
, &icur
, &got
,
5966 cur
, &logflags
, new_startoff
);
5970 if (!xfs_iext_prev_extent(ifp
, &icur
, &got
) ||
5971 stop_fsb
>= got
.br_startoff
+ got
.br_blockcount
) {
5976 *next_fsb
= got
.br_startoff
;
5979 xfs_btree_del_cursor(cur
, error
);
5981 xfs_trans_log_inode(tp
, ip
, logflags
);
5986 * Splits an extent into two extents at split_fsb block such that it is the
5987 * first block of the current_ext. @ext is a target extent to be split.
5988 * @split_fsb is a block where the extents is split. If split_fsb lies in a
5989 * hole or the first block of extents, just return 0.
5992 xfs_bmap_split_extent(
5993 struct xfs_trans
*tp
,
5994 struct xfs_inode
*ip
,
5995 xfs_fileoff_t split_fsb
)
5997 int whichfork
= XFS_DATA_FORK
;
5998 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
5999 struct xfs_btree_cur
*cur
= NULL
;
6000 struct xfs_bmbt_irec got
;
6001 struct xfs_bmbt_irec
new; /* split extent */
6002 struct xfs_mount
*mp
= ip
->i_mount
;
6003 xfs_fsblock_t gotblkcnt
; /* new block count for got */
6004 struct xfs_iext_cursor icur
;
6009 if (XFS_IS_CORRUPT(mp
, !xfs_ifork_has_extents(ifp
)) ||
6010 XFS_TEST_ERROR(false, mp
, XFS_ERRTAG_BMAPIFORMAT
)) {
6011 return -EFSCORRUPTED
;
6014 if (XFS_FORCED_SHUTDOWN(mp
))
6017 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
6018 /* Read in all the extents */
6019 error
= xfs_iread_extents(tp
, ip
, whichfork
);
6025 * If there are not extents, or split_fsb lies in a hole we are done.
6027 if (!xfs_iext_lookup_extent(ip
, ifp
, split_fsb
, &icur
, &got
) ||
6028 got
.br_startoff
>= split_fsb
)
6031 gotblkcnt
= split_fsb
- got
.br_startoff
;
6032 new.br_startoff
= split_fsb
;
6033 new.br_startblock
= got
.br_startblock
+ gotblkcnt
;
6034 new.br_blockcount
= got
.br_blockcount
- gotblkcnt
;
6035 new.br_state
= got
.br_state
;
6037 if (ifp
->if_flags
& XFS_IFBROOT
) {
6038 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
6039 cur
->bc_ino
.flags
= 0;
6040 error
= xfs_bmbt_lookup_eq(cur
, &got
, &i
);
6043 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
6044 error
= -EFSCORRUPTED
;
6049 got
.br_blockcount
= gotblkcnt
;
6050 xfs_iext_update_extent(ip
, xfs_bmap_fork_to_state(whichfork
), &icur
,
6053 logflags
= XFS_ILOG_CORE
;
6055 error
= xfs_bmbt_update(cur
, &got
);
6059 logflags
|= XFS_ILOG_DEXT
;
6061 /* Add new extent */
6062 xfs_iext_next(ifp
, &icur
);
6063 xfs_iext_insert(ip
, &icur
, &new, 0);
6067 error
= xfs_bmbt_lookup_eq(cur
, &new, &i
);
6070 if (XFS_IS_CORRUPT(mp
, i
!= 0)) {
6071 error
= -EFSCORRUPTED
;
6074 error
= xfs_btree_insert(cur
, &i
);
6077 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
6078 error
= -EFSCORRUPTED
;
6084 * Convert to a btree if necessary.
6086 if (xfs_bmap_needs_btree(ip
, whichfork
)) {
6087 int tmp_logflags
; /* partial log flag return val */
6089 ASSERT(cur
== NULL
);
6090 error
= xfs_bmap_extents_to_btree(tp
, ip
, &cur
, 0,
6091 &tmp_logflags
, whichfork
);
6092 logflags
|= tmp_logflags
;
6097 cur
->bc_ino
.allocated
= 0;
6098 xfs_btree_del_cursor(cur
, error
);
6102 xfs_trans_log_inode(tp
, ip
, logflags
);
6106 /* Deferred mapping is only for real extents in the data fork. */
6108 xfs_bmap_is_update_needed(
6109 struct xfs_bmbt_irec
*bmap
)
6111 return bmap
->br_startblock
!= HOLESTARTBLOCK
&&
6112 bmap
->br_startblock
!= DELAYSTARTBLOCK
;
6115 /* Record a bmap intent. */
6118 struct xfs_trans
*tp
,
6119 enum xfs_bmap_intent_type type
,
6120 struct xfs_inode
*ip
,
6122 struct xfs_bmbt_irec
*bmap
)
6124 struct xfs_bmap_intent
*bi
;
6126 trace_xfs_bmap_defer(tp
->t_mountp
,
6127 XFS_FSB_TO_AGNO(tp
->t_mountp
, bmap
->br_startblock
),
6129 XFS_FSB_TO_AGBNO(tp
->t_mountp
, bmap
->br_startblock
),
6130 ip
->i_ino
, whichfork
,
6132 bmap
->br_blockcount
,
6135 bi
= kmem_alloc(sizeof(struct xfs_bmap_intent
), KM_NOFS
);
6136 INIT_LIST_HEAD(&bi
->bi_list
);
6139 bi
->bi_whichfork
= whichfork
;
6140 bi
->bi_bmap
= *bmap
;
6142 xfs_defer_add(tp
, XFS_DEFER_OPS_TYPE_BMAP
, &bi
->bi_list
);
6146 /* Map an extent into a file. */
6148 xfs_bmap_map_extent(
6149 struct xfs_trans
*tp
,
6150 struct xfs_inode
*ip
,
6151 struct xfs_bmbt_irec
*PREV
)
6153 if (!xfs_bmap_is_update_needed(PREV
))
6156 __xfs_bmap_add(tp
, XFS_BMAP_MAP
, ip
, XFS_DATA_FORK
, PREV
);
6159 /* Unmap an extent out of a file. */
6161 xfs_bmap_unmap_extent(
6162 struct xfs_trans
*tp
,
6163 struct xfs_inode
*ip
,
6164 struct xfs_bmbt_irec
*PREV
)
6166 if (!xfs_bmap_is_update_needed(PREV
))
6169 __xfs_bmap_add(tp
, XFS_BMAP_UNMAP
, ip
, XFS_DATA_FORK
, PREV
);
6173 * Process one of the deferred bmap operations. We pass back the
6174 * btree cursor to maintain our lock on the bmapbt between calls.
6177 xfs_bmap_finish_one(
6178 struct xfs_trans
*tp
,
6179 struct xfs_inode
*ip
,
6180 enum xfs_bmap_intent_type type
,
6182 xfs_fileoff_t startoff
,
6183 xfs_fsblock_t startblock
,
6184 xfs_filblks_t
*blockcount
,
6189 ASSERT(tp
->t_firstblock
== NULLFSBLOCK
);
6191 trace_xfs_bmap_deferred(tp
->t_mountp
,
6192 XFS_FSB_TO_AGNO(tp
->t_mountp
, startblock
), type
,
6193 XFS_FSB_TO_AGBNO(tp
->t_mountp
, startblock
),
6194 ip
->i_ino
, whichfork
, startoff
, *blockcount
, state
);
6196 if (WARN_ON_ONCE(whichfork
!= XFS_DATA_FORK
))
6197 return -EFSCORRUPTED
;
6199 if (XFS_TEST_ERROR(false, tp
->t_mountp
,
6200 XFS_ERRTAG_BMAP_FINISH_ONE
))
6205 error
= xfs_bmapi_remap(tp
, ip
, startoff
, *blockcount
,
6209 case XFS_BMAP_UNMAP
:
6210 error
= __xfs_bunmapi(tp
, ip
, startoff
, blockcount
,
6211 XFS_BMAPI_REMAP
, 1);
6215 error
= -EFSCORRUPTED
;
6221 /* Check that an inode's extent does not have invalid flags or bad ranges. */
6223 xfs_bmap_validate_extent(
6224 struct xfs_inode
*ip
,
6226 struct xfs_bmbt_irec
*irec
)
6228 struct xfs_mount
*mp
= ip
->i_mount
;
6230 if (!xfs_verify_fileext(mp
, irec
->br_startoff
, irec
->br_blockcount
))
6231 return __this_address
;
6233 if (XFS_IS_REALTIME_INODE(ip
) && whichfork
== XFS_DATA_FORK
) {
6234 if (!xfs_verify_rtext(mp
, irec
->br_startblock
,
6235 irec
->br_blockcount
))
6236 return __this_address
;
6238 if (!xfs_verify_fsbext(mp
, irec
->br_startblock
,
6239 irec
->br_blockcount
))
6240 return __this_address
;
6242 if (irec
->br_state
!= XFS_EXT_NORM
&& whichfork
!= XFS_DATA_FORK
)
6243 return __this_address
;