2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_alloc_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_dir2_sf.h"
33 #include "xfs_attr_sf.h"
34 #include "xfs_dinode.h"
35 #include "xfs_inode.h"
36 #include "xfs_btree.h"
37 #include "xfs_dmapi.h"
38 #include "xfs_mount.h"
39 #include "xfs_ialloc.h"
40 #include "xfs_itable.h"
41 #include "xfs_dir2_data.h"
42 #include "xfs_dir2_leaf.h"
43 #include "xfs_dir2_block.h"
44 #include "xfs_inode_item.h"
45 #include "xfs_extfree_item.h"
46 #include "xfs_alloc.h"
48 #include "xfs_rtalloc.h"
49 #include "xfs_error.h"
50 #include "xfs_attr_leaf.h"
52 #include "xfs_quota.h"
53 #include "xfs_trans_space.h"
54 #include "xfs_buf_item.h"
59 xfs_bmap_check_leaf_extents(xfs_btree_cur_t *cur, xfs_inode_t *ip, int whichfork);
62 kmem_zone_t *xfs_bmap_free_item_zone;
65 * Prototypes for internal bmap routines.
70 * Called from xfs_bmap_add_attrfork to handle extents format files.
72 STATIC int /* error */
73 xfs_bmap_add_attrfork_extents(
74 xfs_trans_t *tp, /* transaction pointer */
75 xfs_inode_t *ip, /* incore inode pointer */
76 xfs_fsblock_t *firstblock, /* first block allocated */
77 xfs_bmap_free_t *flist, /* blocks to free at commit */
78 int *flags); /* inode logging flags */
81 * Called from xfs_bmap_add_attrfork to handle local format files.
83 STATIC int /* error */
84 xfs_bmap_add_attrfork_local(
85 xfs_trans_t *tp, /* transaction pointer */
86 xfs_inode_t *ip, /* incore inode pointer */
87 xfs_fsblock_t *firstblock, /* first block allocated */
88 xfs_bmap_free_t *flist, /* blocks to free at commit */
89 int *flags); /* inode logging flags */
92 * Called by xfs_bmapi to update file extent records and the btree
93 * after allocating space (or doing a delayed allocation).
95 STATIC int /* error */
97 xfs_inode_t *ip, /* incore inode pointer */
98 xfs_extnum_t idx, /* extent number to update/insert */
99 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
100 xfs_bmbt_irec_t *new, /* new data to add to file extents */
101 xfs_fsblock_t *first, /* pointer to firstblock variable */
102 xfs_bmap_free_t *flist, /* list of extents to be freed */
103 int *logflagsp, /* inode logging flags */
104 xfs_extdelta_t *delta, /* Change made to incore extents */
105 int whichfork, /* data or attr fork */
106 int rsvd); /* OK to allocate reserved blocks */
109 * Called by xfs_bmap_add_extent to handle cases converting a delayed
110 * allocation to a real allocation.
112 STATIC int /* error */
113 xfs_bmap_add_extent_delay_real(
114 xfs_inode_t *ip, /* incore inode pointer */
115 xfs_extnum_t idx, /* extent number to update/insert */
116 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
117 xfs_bmbt_irec_t *new, /* new data to add to file extents */
118 xfs_filblks_t *dnew, /* new delayed-alloc indirect blocks */
119 xfs_fsblock_t *first, /* pointer to firstblock variable */
120 xfs_bmap_free_t *flist, /* list of extents to be freed */
121 int *logflagsp, /* inode logging flags */
122 xfs_extdelta_t *delta, /* Change made to incore extents */
123 int rsvd); /* OK to allocate reserved blocks */
126 * Called by xfs_bmap_add_extent to handle cases converting a hole
127 * to a delayed allocation.
129 STATIC int /* error */
130 xfs_bmap_add_extent_hole_delay(
131 xfs_inode_t *ip, /* incore inode pointer */
132 xfs_extnum_t idx, /* extent number to update/insert */
133 xfs_btree_cur_t *cur, /* if null, not a btree */
134 xfs_bmbt_irec_t *new, /* new data to add to file extents */
135 int *logflagsp,/* inode logging flags */
136 xfs_extdelta_t *delta, /* Change made to incore extents */
137 int rsvd); /* OK to allocate reserved blocks */
140 * Called by xfs_bmap_add_extent to handle cases converting a hole
141 * to a real allocation.
143 STATIC int /* error */
144 xfs_bmap_add_extent_hole_real(
145 xfs_inode_t *ip, /* incore inode pointer */
146 xfs_extnum_t idx, /* extent number to update/insert */
147 xfs_btree_cur_t *cur, /* if null, not a btree */
148 xfs_bmbt_irec_t *new, /* new data to add to file extents */
149 int *logflagsp, /* inode logging flags */
150 xfs_extdelta_t *delta, /* Change made to incore extents */
151 int whichfork); /* data or attr fork */
154 * Called by xfs_bmap_add_extent to handle cases converting an unwritten
155 * allocation to a real allocation or vice versa.
157 STATIC int /* error */
158 xfs_bmap_add_extent_unwritten_real(
159 xfs_inode_t *ip, /* incore inode pointer */
160 xfs_extnum_t idx, /* extent number to update/insert */
161 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
162 xfs_bmbt_irec_t *new, /* new data to add to file extents */
163 int *logflagsp, /* inode logging flags */
164 xfs_extdelta_t *delta); /* Change made to incore extents */
167 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
168 * It figures out where to ask the underlying allocator to put the new extent.
170 STATIC int /* error */
172 xfs_bmalloca_t *ap); /* bmap alloc argument struct */
175 * Transform a btree format file with only one leaf node, where the
176 * extents list will fit in the inode, into an extents format file.
177 * Since the file extents are already in-core, all we have to do is
178 * give up the space for the btree root and pitch the leaf block.
180 STATIC int /* error */
181 xfs_bmap_btree_to_extents(
182 xfs_trans_t *tp, /* transaction pointer */
183 xfs_inode_t *ip, /* incore inode pointer */
184 xfs_btree_cur_t *cur, /* btree cursor */
185 int *logflagsp, /* inode logging flags */
186 int whichfork); /* data or attr fork */
189 * Called by xfs_bmapi to update file extent records and the btree
190 * after removing space (or undoing a delayed allocation).
192 STATIC int /* error */
194 xfs_inode_t *ip, /* incore inode pointer */
195 xfs_trans_t *tp, /* current trans pointer */
196 xfs_extnum_t idx, /* extent number to update/insert */
197 xfs_bmap_free_t *flist, /* list of extents to be freed */
198 xfs_btree_cur_t *cur, /* if null, not a btree */
199 xfs_bmbt_irec_t *new, /* new data to add to file extents */
200 int *logflagsp,/* inode logging flags */
201 xfs_extdelta_t *delta, /* Change made to incore extents */
202 int whichfork, /* data or attr fork */
203 int rsvd); /* OK to allocate reserved blocks */
206 * Remove the entry "free" from the free item list. Prev points to the
207 * previous entry, unless "free" is the head of the list.
211 xfs_bmap_free_t *flist, /* free item list header */
212 xfs_bmap_free_item_t *prev, /* previous item on list, if any */
213 xfs_bmap_free_item_t *free); /* list item to be freed */
216 * Convert an extents-format file into a btree-format file.
217 * The new file will have a root block (in the inode) and a single child block.
219 STATIC int /* error */
220 xfs_bmap_extents_to_btree(
221 xfs_trans_t *tp, /* transaction pointer */
222 xfs_inode_t *ip, /* incore inode pointer */
223 xfs_fsblock_t *firstblock, /* first-block-allocated */
224 xfs_bmap_free_t *flist, /* blocks freed in xaction */
225 xfs_btree_cur_t **curp, /* cursor returned to caller */
226 int wasdel, /* converting a delayed alloc */
227 int *logflagsp, /* inode logging flags */
228 int whichfork); /* data or attr fork */
231 * Convert a local file to an extents file.
232 * This code is sort of bogus, since the file data needs to get
233 * logged so it won't be lost. The bmap-level manipulations are ok, though.
235 STATIC int /* error */
236 xfs_bmap_local_to_extents(
237 xfs_trans_t *tp, /* transaction pointer */
238 xfs_inode_t *ip, /* incore inode pointer */
239 xfs_fsblock_t *firstblock, /* first block allocated in xaction */
240 xfs_extlen_t total, /* total blocks needed by transaction */
241 int *logflagsp, /* inode logging flags */
242 int whichfork); /* data or attr fork */
245 * Search the extents list for the inode, for the extent containing bno.
246 * If bno lies in a hole, point to the next entry. If bno lies past eof,
247 * *eofp will be set, and *prevp will contain the last entry (null if none).
248 * Else, *lastxp will be set to the index of the found
249 * entry; *gotp will contain the entry.
251 STATIC xfs_bmbt_rec_t * /* pointer to found extent entry */
252 xfs_bmap_search_extents(
253 xfs_inode_t *ip, /* incore inode pointer */
254 xfs_fileoff_t bno, /* block number searched for */
255 int whichfork, /* data or attr fork */
256 int *eofp, /* out: end of file found */
257 xfs_extnum_t *lastxp, /* out: last extent index */
258 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
259 xfs_bmbt_irec_t *prevp); /* out: previous extent entry found */
262 * Check the last inode extent to determine whether this allocation will result
263 * in blocks being allocated at the end of the file. When we allocate new data
264 * blocks at the end of the file which do not start at the previous data block,
265 * we will try to align the new blocks at stripe unit boundaries.
267 STATIC int /* error */
269 xfs_inode_t *ip, /* incore inode pointer */
270 xfs_fileoff_t off, /* file offset in fsblocks */
271 int whichfork, /* data or attribute fork */
272 char *aeof); /* return value */
274 #ifdef XFS_BMAP_TRACE
276 * Add a bmap trace buffer entry. Base routine for the others.
279 xfs_bmap_trace_addentry(
280 int opcode, /* operation */
281 char *fname, /* function name */
282 char *desc, /* operation description */
283 xfs_inode_t *ip, /* incore inode pointer */
284 xfs_extnum_t idx, /* index of entry(ies) */
285 xfs_extnum_t cnt, /* count of entries, 1 or 2 */
286 xfs_bmbt_rec_t *r1, /* first record */
287 xfs_bmbt_rec_t *r2, /* second record or null */
288 int whichfork); /* data or attr fork */
291 * Add bmap trace entry prior to a call to xfs_iext_remove.
294 xfs_bmap_trace_delete(
295 char *fname, /* function name */
296 char *desc, /* operation description */
297 xfs_inode_t *ip, /* incore inode pointer */
298 xfs_extnum_t idx, /* index of entry(entries) deleted */
299 xfs_extnum_t cnt, /* count of entries deleted, 1 or 2 */
300 int whichfork); /* data or attr fork */
303 * Add bmap trace entry prior to a call to xfs_iext_insert, or
304 * reading in the extents list from the disk (in the btree).
307 xfs_bmap_trace_insert(
308 char *fname, /* function name */
309 char *desc, /* operation description */
310 xfs_inode_t *ip, /* incore inode pointer */
311 xfs_extnum_t idx, /* index of entry(entries) inserted */
312 xfs_extnum_t cnt, /* count of entries inserted, 1 or 2 */
313 xfs_bmbt_irec_t *r1, /* inserted record 1 */
314 xfs_bmbt_irec_t *r2, /* inserted record 2 or null */
315 int whichfork); /* data or attr fork */
318 * Add bmap trace entry after updating an extent record in place.
321 xfs_bmap_trace_post_update(
322 char *fname, /* function name */
323 char *desc, /* operation description */
324 xfs_inode_t *ip, /* incore inode pointer */
325 xfs_extnum_t idx, /* index of entry updated */
326 int whichfork); /* data or attr fork */
329 * Add bmap trace entry prior to updating an extent record in place.
332 xfs_bmap_trace_pre_update(
333 char *fname, /* function name */
334 char *desc, /* operation description */
335 xfs_inode_t *ip, /* incore inode pointer */
336 xfs_extnum_t idx, /* index of entry to be updated */
337 int whichfork); /* data or attr fork */
340 #define xfs_bmap_trace_delete(f,d,ip,i,c,w)
341 #define xfs_bmap_trace_insert(f,d,ip,i,c,r1,r2,w)
342 #define xfs_bmap_trace_post_update(f,d,ip,i,w)
343 #define xfs_bmap_trace_pre_update(f,d,ip,i,w)
344 #endif /* XFS_BMAP_TRACE */
347 * Compute the worst-case number of indirect blocks that will be used
348 * for ip's delayed extent of length "len".
351 xfs_bmap_worst_indlen(
352 xfs_inode_t *ip, /* incore inode pointer */
353 xfs_filblks_t len); /* delayed extent length */
357 * Perform various validation checks on the values being returned
361 xfs_bmap_validate_ret(
365 xfs_bmbt_irec_t *mval,
369 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
372 #if defined(XFS_RW_TRACE)
381 #define xfs_bunmap_trace(ip, bno, len, flags, ra)
382 #endif /* XFS_RW_TRACE */
389 xfs_fsblock_t blockno,
394 xfs_bmap_count_leaves(
401 xfs_bmap_disk_count_leaves(
404 xfs_bmbt_block_t *block,
409 * Bmap internal routines.
413 * Called from xfs_bmap_add_attrfork to handle btree format files.
415 STATIC int /* error */
416 xfs_bmap_add_attrfork_btree(
417 xfs_trans_t *tp, /* transaction pointer */
418 xfs_inode_t *ip, /* incore inode pointer */
419 xfs_fsblock_t *firstblock, /* first block allocated */
420 xfs_bmap_free_t *flist, /* blocks to free at commit */
421 int *flags) /* inode logging flags */
423 xfs_btree_cur_t *cur; /* btree cursor */
424 int error; /* error return value */
425 xfs_mount_t *mp; /* file system mount struct */
426 int stat; /* newroot status */
429 if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
430 *flags |= XFS_ILOG_DBROOT;
432 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
434 cur->bc_private.b.flist = flist;
435 cur->bc_private.b.firstblock = *firstblock;
436 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
438 ASSERT(stat == 1); /* must be at least one entry */
439 if ((error = xfs_bmbt_newroot(cur, flags, &stat)))
442 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
443 return XFS_ERROR(ENOSPC);
445 *firstblock = cur->bc_private.b.firstblock;
446 cur->bc_private.b.allocated = 0;
447 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
451 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
456 * Called from xfs_bmap_add_attrfork to handle extents format files.
458 STATIC int /* error */
459 xfs_bmap_add_attrfork_extents(
460 xfs_trans_t *tp, /* transaction pointer */
461 xfs_inode_t *ip, /* incore inode pointer */
462 xfs_fsblock_t *firstblock, /* first block allocated */
463 xfs_bmap_free_t *flist, /* blocks to free at commit */
464 int *flags) /* inode logging flags */
466 xfs_btree_cur_t *cur; /* bmap btree cursor */
467 int error; /* error return value */
469 if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
472 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
473 flags, XFS_DATA_FORK);
475 cur->bc_private.b.allocated = 0;
476 xfs_btree_del_cursor(cur,
477 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
483 * Called from xfs_bmap_add_attrfork to handle local format files.
485 STATIC int /* error */
486 xfs_bmap_add_attrfork_local(
487 xfs_trans_t *tp, /* transaction pointer */
488 xfs_inode_t *ip, /* incore inode pointer */
489 xfs_fsblock_t *firstblock, /* first block allocated */
490 xfs_bmap_free_t *flist, /* blocks to free at commit */
491 int *flags) /* inode logging flags */
493 xfs_da_args_t dargs; /* args for dir/attr code */
494 int error; /* error return value */
495 xfs_mount_t *mp; /* mount structure pointer */
497 if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
499 if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
501 memset(&dargs, 0, sizeof(dargs));
503 dargs.firstblock = firstblock;
505 dargs.total = mp->m_dirblkfsbs;
506 dargs.whichfork = XFS_DATA_FORK;
508 error = xfs_dir2_sf_to_block(&dargs);
510 error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
516 * Called by xfs_bmapi to update file extent records and the btree
517 * after allocating space (or doing a delayed allocation).
519 STATIC int /* error */
521 xfs_inode_t *ip, /* incore inode pointer */
522 xfs_extnum_t idx, /* extent number to update/insert */
523 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
524 xfs_bmbt_irec_t *new, /* new data to add to file extents */
525 xfs_fsblock_t *first, /* pointer to firstblock variable */
526 xfs_bmap_free_t *flist, /* list of extents to be freed */
527 int *logflagsp, /* inode logging flags */
528 xfs_extdelta_t *delta, /* Change made to incore extents */
529 int whichfork, /* data or attr fork */
530 int rsvd) /* OK to use reserved data blocks */
532 xfs_btree_cur_t *cur; /* btree cursor or null */
533 xfs_filblks_t da_new; /* new count del alloc blocks used */
534 xfs_filblks_t da_old; /* old count del alloc blocks used */
535 int error; /* error return value */
536 #ifdef XFS_BMAP_TRACE
537 static char fname[] = "xfs_bmap_add_extent";
539 xfs_ifork_t *ifp; /* inode fork ptr */
540 int logflags; /* returned value */
541 xfs_extnum_t nextents; /* number of extents in file now */
543 XFS_STATS_INC(xs_add_exlist);
545 ifp = XFS_IFORK_PTR(ip, whichfork);
546 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
547 ASSERT(idx <= nextents);
551 * This is the first extent added to a new/empty file.
552 * Special case this one, so other routines get to assume there are
553 * already extents in the list.
556 xfs_bmap_trace_insert(fname, "insert empty", ip, 0, 1, new,
558 xfs_iext_insert(ifp, 0, 1, new);
561 if (!ISNULLSTARTBLOCK(new->br_startblock)) {
562 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
563 logflags = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
566 /* DELTA: single new extent */
568 if (delta->xed_startoff > new->br_startoff)
569 delta->xed_startoff = new->br_startoff;
570 if (delta->xed_blockcount <
571 new->br_startoff + new->br_blockcount)
572 delta->xed_blockcount = new->br_startoff +
577 * Any kind of new delayed allocation goes here.
579 else if (ISNULLSTARTBLOCK(new->br_startblock)) {
581 ASSERT((cur->bc_private.b.flags &
582 XFS_BTCUR_BPRV_WASDEL) == 0);
583 if ((error = xfs_bmap_add_extent_hole_delay(ip, idx, cur, new,
584 &logflags, delta, rsvd)))
588 * Real allocation off the end of the file.
590 else if (idx == nextents) {
592 ASSERT((cur->bc_private.b.flags &
593 XFS_BTCUR_BPRV_WASDEL) == 0);
594 if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur, new,
595 &logflags, delta, whichfork)))
598 xfs_bmbt_irec_t prev; /* old extent at offset idx */
601 * Get the record referred to by idx.
603 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx), &prev);
605 * If it's a real allocation record, and the new allocation ends
606 * after the start of the referred to record, then we're filling
607 * in a delayed or unwritten allocation with a real one, or
608 * converting real back to unwritten.
610 if (!ISNULLSTARTBLOCK(new->br_startblock) &&
611 new->br_startoff + new->br_blockcount > prev.br_startoff) {
612 if (prev.br_state != XFS_EXT_UNWRITTEN &&
613 ISNULLSTARTBLOCK(prev.br_startblock)) {
614 da_old = STARTBLOCKVAL(prev.br_startblock);
616 ASSERT(cur->bc_private.b.flags &
617 XFS_BTCUR_BPRV_WASDEL);
618 if ((error = xfs_bmap_add_extent_delay_real(ip,
619 idx, &cur, new, &da_new, first, flist,
620 &logflags, delta, rsvd)))
622 } else if (new->br_state == XFS_EXT_NORM) {
623 ASSERT(new->br_state == XFS_EXT_NORM);
624 if ((error = xfs_bmap_add_extent_unwritten_real(
625 ip, idx, &cur, new, &logflags, delta)))
628 ASSERT(new->br_state == XFS_EXT_UNWRITTEN);
629 if ((error = xfs_bmap_add_extent_unwritten_real(
630 ip, idx, &cur, new, &logflags, delta)))
633 ASSERT(*curp == cur || *curp == NULL);
636 * Otherwise we're filling in a hole with an allocation.
640 ASSERT((cur->bc_private.b.flags &
641 XFS_BTCUR_BPRV_WASDEL) == 0);
642 if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur,
643 new, &logflags, delta, whichfork)))
648 ASSERT(*curp == cur || *curp == NULL);
650 * Convert to a btree if necessary.
652 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
653 XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
654 int tmp_logflags; /* partial log flag return val */
657 error = xfs_bmap_extents_to_btree(ip->i_transp, ip, first,
658 flist, &cur, da_old > 0, &tmp_logflags, whichfork);
659 logflags |= tmp_logflags;
664 * Adjust for changes in reserved delayed indirect blocks.
665 * Nothing to do for disk quotas here.
667 if (da_old || da_new) {
672 nblks += cur->bc_private.b.allocated;
673 ASSERT(nblks <= da_old);
675 xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
676 (int64_t)(da_old - nblks), rsvd);
679 * Clear out the allocated field, done with it now in any case.
682 cur->bc_private.b.allocated = 0;
688 xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
690 *logflagsp = logflags;
695 * Called by xfs_bmap_add_extent to handle cases converting a delayed
696 * allocation to a real allocation.
698 STATIC int /* error */
699 xfs_bmap_add_extent_delay_real(
700 xfs_inode_t *ip, /* incore inode pointer */
701 xfs_extnum_t idx, /* extent number to update/insert */
702 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
703 xfs_bmbt_irec_t *new, /* new data to add to file extents */
704 xfs_filblks_t *dnew, /* new delayed-alloc indirect blocks */
705 xfs_fsblock_t *first, /* pointer to firstblock variable */
706 xfs_bmap_free_t *flist, /* list of extents to be freed */
707 int *logflagsp, /* inode logging flags */
708 xfs_extdelta_t *delta, /* Change made to incore extents */
709 int rsvd) /* OK to use reserved data block allocation */
711 xfs_btree_cur_t *cur; /* btree cursor */
712 int diff; /* temp value */
713 xfs_bmbt_rec_t *ep; /* extent entry for idx */
714 int error; /* error return value */
715 #ifdef XFS_BMAP_TRACE
716 static char fname[] = "xfs_bmap_add_extent_delay_real";
718 int i; /* temp state */
719 xfs_ifork_t *ifp; /* inode fork pointer */
720 xfs_fileoff_t new_endoff; /* end offset of new entry */
721 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
722 /* left is 0, right is 1, prev is 2 */
723 int rval=0; /* return value (logging flags) */
724 int state = 0;/* state bits, accessed thru macros */
725 xfs_filblks_t temp=0; /* value for dnew calculations */
726 xfs_filblks_t temp2=0;/* value for dnew calculations */
727 int tmp_rval; /* partial logging flags */
728 enum { /* bit number definitions for state */
729 LEFT_CONTIG, RIGHT_CONTIG,
730 LEFT_FILLING, RIGHT_FILLING,
731 LEFT_DELAY, RIGHT_DELAY,
732 LEFT_VALID, RIGHT_VALID
738 #define MASK(b) (1 << (b))
739 #define MASK2(a,b) (MASK(a) | MASK(b))
740 #define MASK3(a,b,c) (MASK2(a,b) | MASK(c))
741 #define MASK4(a,b,c,d) (MASK3(a,b,c) | MASK(d))
742 #define STATE_SET(b,v) ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
743 #define STATE_TEST(b) (state & MASK(b))
744 #define STATE_SET_TEST(b,v) ((v) ? ((state |= MASK(b)), 1) : \
745 ((state &= ~MASK(b)), 0))
746 #define SWITCH_STATE \
747 (state & MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG))
750 * Set up a bunch of variables to make the tests simpler.
753 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
754 ep = xfs_iext_get_ext(ifp, idx);
755 xfs_bmbt_get_all(ep, &PREV);
756 new_endoff = new->br_startoff + new->br_blockcount;
757 ASSERT(PREV.br_startoff <= new->br_startoff);
758 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
760 * Set flags determining what part of the previous delayed allocation
761 * extent is being replaced by a real allocation.
763 STATE_SET(LEFT_FILLING, PREV.br_startoff == new->br_startoff);
764 STATE_SET(RIGHT_FILLING,
765 PREV.br_startoff + PREV.br_blockcount == new_endoff);
767 * Check and set flags if this segment has a left neighbor.
768 * Don't set contiguous if the combined extent would be too large.
770 if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
771 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &LEFT);
772 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(LEFT.br_startblock));
774 STATE_SET(LEFT_CONTIG,
775 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
776 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
777 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
778 LEFT.br_state == new->br_state &&
779 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN);
781 * Check and set flags if this segment has a right neighbor.
782 * Don't set contiguous if the combined extent would be too large.
783 * Also check for all-three-contiguous being too large.
785 if (STATE_SET_TEST(RIGHT_VALID,
787 ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1)) {
788 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx + 1), &RIGHT);
789 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(RIGHT.br_startblock));
791 STATE_SET(RIGHT_CONTIG,
792 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
793 new_endoff == RIGHT.br_startoff &&
794 new->br_startblock + new->br_blockcount ==
795 RIGHT.br_startblock &&
796 new->br_state == RIGHT.br_state &&
797 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
798 ((state & MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING)) !=
799 MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING) ||
800 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
804 * Switch out based on the FILLING and CONTIG state bits.
806 switch (SWITCH_STATE) {
808 case MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
810 * Filling in all of a previously delayed allocation extent.
811 * The left and right neighbors are both contiguous with new.
813 xfs_bmap_trace_pre_update(fname, "LF|RF|LC|RC", ip, idx - 1,
815 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
816 LEFT.br_blockcount + PREV.br_blockcount +
817 RIGHT.br_blockcount);
818 xfs_bmap_trace_post_update(fname, "LF|RF|LC|RC", ip, idx - 1,
820 xfs_bmap_trace_delete(fname, "LF|RF|LC|RC", ip, idx, 2,
822 xfs_iext_remove(ifp, idx, 2);
823 ip->i_df.if_lastex = idx - 1;
824 ip->i_d.di_nextents--;
826 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
828 rval = XFS_ILOG_CORE;
829 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
831 RIGHT.br_blockcount, &i)))
834 if ((error = xfs_bmbt_delete(cur, &i)))
837 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
840 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
844 RIGHT.br_blockcount, LEFT.br_state)))
848 /* DELTA: Three in-core extents are replaced by one. */
849 temp = LEFT.br_startoff;
850 temp2 = LEFT.br_blockcount +
855 case MASK3(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG):
857 * Filling in all of a previously delayed allocation extent.
858 * The left neighbor is contiguous, the right is not.
860 xfs_bmap_trace_pre_update(fname, "LF|RF|LC", ip, idx - 1,
862 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
863 LEFT.br_blockcount + PREV.br_blockcount);
864 xfs_bmap_trace_post_update(fname, "LF|RF|LC", ip, idx - 1,
866 ip->i_df.if_lastex = idx - 1;
867 xfs_bmap_trace_delete(fname, "LF|RF|LC", ip, idx, 1,
869 xfs_iext_remove(ifp, idx, 1);
871 rval = XFS_ILOG_DEXT;
874 if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
875 LEFT.br_startblock, LEFT.br_blockcount,
879 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
882 PREV.br_blockcount, LEFT.br_state)))
886 /* DELTA: Two in-core extents are replaced by one. */
887 temp = LEFT.br_startoff;
888 temp2 = LEFT.br_blockcount +
892 case MASK3(LEFT_FILLING, RIGHT_FILLING, RIGHT_CONTIG):
894 * Filling in all of a previously delayed allocation extent.
895 * The right neighbor is contiguous, the left is not.
897 xfs_bmap_trace_pre_update(fname, "LF|RF|RC", ip, idx,
899 xfs_bmbt_set_startblock(ep, new->br_startblock);
900 xfs_bmbt_set_blockcount(ep,
901 PREV.br_blockcount + RIGHT.br_blockcount);
902 xfs_bmap_trace_post_update(fname, "LF|RF|RC", ip, idx,
904 ip->i_df.if_lastex = idx;
905 xfs_bmap_trace_delete(fname, "LF|RF|RC", ip, idx + 1, 1,
907 xfs_iext_remove(ifp, idx + 1, 1);
909 rval = XFS_ILOG_DEXT;
912 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
914 RIGHT.br_blockcount, &i)))
917 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
920 RIGHT.br_blockcount, PREV.br_state)))
924 /* DELTA: Two in-core extents are replaced by one. */
925 temp = PREV.br_startoff;
926 temp2 = PREV.br_blockcount +
930 case MASK2(LEFT_FILLING, RIGHT_FILLING):
932 * Filling in all of a previously delayed allocation extent.
933 * Neither the left nor right neighbors are contiguous with
936 xfs_bmap_trace_pre_update(fname, "LF|RF", ip, idx,
938 xfs_bmbt_set_startblock(ep, new->br_startblock);
939 xfs_bmap_trace_post_update(fname, "LF|RF", ip, idx,
941 ip->i_df.if_lastex = idx;
942 ip->i_d.di_nextents++;
944 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
946 rval = XFS_ILOG_CORE;
947 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
948 new->br_startblock, new->br_blockcount,
952 cur->bc_rec.b.br_state = XFS_EXT_NORM;
953 if ((error = xfs_bmbt_insert(cur, &i)))
958 /* DELTA: The in-core extent described by new changed type. */
959 temp = new->br_startoff;
960 temp2 = new->br_blockcount;
963 case MASK2(LEFT_FILLING, LEFT_CONTIG):
965 * Filling in the first part of a previous delayed allocation.
966 * The left neighbor is contiguous.
968 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx - 1,
970 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
971 LEFT.br_blockcount + new->br_blockcount);
972 xfs_bmbt_set_startoff(ep,
973 PREV.br_startoff + new->br_blockcount);
974 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx - 1,
976 temp = PREV.br_blockcount - new->br_blockcount;
977 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx,
979 xfs_bmbt_set_blockcount(ep, temp);
980 ip->i_df.if_lastex = idx - 1;
982 rval = XFS_ILOG_DEXT;
985 if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
986 LEFT.br_startblock, LEFT.br_blockcount,
990 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
997 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
998 STARTBLOCKVAL(PREV.br_startblock));
999 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1000 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx,
1003 /* DELTA: The boundary between two in-core extents moved. */
1004 temp = LEFT.br_startoff;
1005 temp2 = LEFT.br_blockcount +
1009 case MASK(LEFT_FILLING):
1011 * Filling in the first part of a previous delayed allocation.
1012 * The left neighbor is not contiguous.
1014 xfs_bmap_trace_pre_update(fname, "LF", ip, idx, XFS_DATA_FORK);
1015 xfs_bmbt_set_startoff(ep, new_endoff);
1016 temp = PREV.br_blockcount - new->br_blockcount;
1017 xfs_bmbt_set_blockcount(ep, temp);
1018 xfs_bmap_trace_insert(fname, "LF", ip, idx, 1, new, NULL,
1020 xfs_iext_insert(ifp, idx, 1, new);
1021 ip->i_df.if_lastex = idx;
1022 ip->i_d.di_nextents++;
1024 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1026 rval = XFS_ILOG_CORE;
1027 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1028 new->br_startblock, new->br_blockcount,
1032 cur->bc_rec.b.br_state = XFS_EXT_NORM;
1033 if ((error = xfs_bmbt_insert(cur, &i)))
1037 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1038 ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1039 error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1040 first, flist, &cur, 1, &tmp_rval,
1046 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1047 STARTBLOCKVAL(PREV.br_startblock) -
1048 (cur ? cur->bc_private.b.allocated : 0));
1049 ep = xfs_iext_get_ext(ifp, idx + 1);
1050 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1051 xfs_bmap_trace_post_update(fname, "LF", ip, idx + 1,
1054 /* DELTA: One in-core extent is split in two. */
1055 temp = PREV.br_startoff;
1056 temp2 = PREV.br_blockcount;
1059 case MASK2(RIGHT_FILLING, RIGHT_CONTIG):
1061 * Filling in the last part of a previous delayed allocation.
1062 * The right neighbor is contiguous with the new allocation.
1064 temp = PREV.br_blockcount - new->br_blockcount;
1065 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx,
1067 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx + 1,
1069 xfs_bmbt_set_blockcount(ep, temp);
1070 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, idx + 1),
1071 new->br_startoff, new->br_startblock,
1072 new->br_blockcount + RIGHT.br_blockcount,
1074 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx + 1,
1076 ip->i_df.if_lastex = idx + 1;
1078 rval = XFS_ILOG_DEXT;
1081 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1082 RIGHT.br_startblock,
1083 RIGHT.br_blockcount, &i)))
1086 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1088 new->br_blockcount +
1089 RIGHT.br_blockcount,
1093 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1094 STARTBLOCKVAL(PREV.br_startblock));
1095 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1096 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx,
1099 /* DELTA: The boundary between two in-core extents moved. */
1100 temp = PREV.br_startoff;
1101 temp2 = PREV.br_blockcount +
1102 RIGHT.br_blockcount;
1105 case MASK(RIGHT_FILLING):
1107 * Filling in the last part of a previous delayed allocation.
1108 * The right neighbor is not contiguous.
1110 temp = PREV.br_blockcount - new->br_blockcount;
1111 xfs_bmap_trace_pre_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1112 xfs_bmbt_set_blockcount(ep, temp);
1113 xfs_bmap_trace_insert(fname, "RF", ip, idx + 1, 1,
1114 new, NULL, XFS_DATA_FORK);
1115 xfs_iext_insert(ifp, idx + 1, 1, new);
1116 ip->i_df.if_lastex = idx + 1;
1117 ip->i_d.di_nextents++;
1119 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1121 rval = XFS_ILOG_CORE;
1122 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1123 new->br_startblock, new->br_blockcount,
1127 cur->bc_rec.b.br_state = XFS_EXT_NORM;
1128 if ((error = xfs_bmbt_insert(cur, &i)))
1132 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1133 ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1134 error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1135 first, flist, &cur, 1, &tmp_rval,
1141 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1142 STARTBLOCKVAL(PREV.br_startblock) -
1143 (cur ? cur->bc_private.b.allocated : 0));
1144 ep = xfs_iext_get_ext(ifp, idx);
1145 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1146 xfs_bmap_trace_post_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1148 /* DELTA: One in-core extent is split in two. */
1149 temp = PREV.br_startoff;
1150 temp2 = PREV.br_blockcount;
1155 * Filling in the middle part of a previous delayed allocation.
1156 * Contiguity is impossible here.
1157 * This case is avoided almost all the time.
1159 temp = new->br_startoff - PREV.br_startoff;
1160 xfs_bmap_trace_pre_update(fname, "0", ip, idx, XFS_DATA_FORK);
1161 xfs_bmbt_set_blockcount(ep, temp);
1163 r[1].br_state = PREV.br_state;
1164 r[1].br_startblock = 0;
1165 r[1].br_startoff = new_endoff;
1166 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
1167 r[1].br_blockcount = temp2;
1168 xfs_bmap_trace_insert(fname, "0", ip, idx + 1, 2, &r[0], &r[1],
1170 xfs_iext_insert(ifp, idx + 1, 2, &r[0]);
1171 ip->i_df.if_lastex = idx + 1;
1172 ip->i_d.di_nextents++;
1174 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1176 rval = XFS_ILOG_CORE;
1177 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1178 new->br_startblock, new->br_blockcount,
1182 cur->bc_rec.b.br_state = XFS_EXT_NORM;
1183 if ((error = xfs_bmbt_insert(cur, &i)))
1187 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1188 ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1189 error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1190 first, flist, &cur, 1, &tmp_rval,
1196 temp = xfs_bmap_worst_indlen(ip, temp);
1197 temp2 = xfs_bmap_worst_indlen(ip, temp2);
1198 diff = (int)(temp + temp2 - STARTBLOCKVAL(PREV.br_startblock) -
1199 (cur ? cur->bc_private.b.allocated : 0));
1201 xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS, -((int64_t)diff), rsvd)) {
1203 * Ick gross gag me with a spoon.
1205 ASSERT(0); /* want to see if this ever happens! */
1211 !xfs_mod_incore_sb(ip->i_mount,
1212 XFS_SBS_FDBLOCKS, -((int64_t)diff), rsvd))
1219 !xfs_mod_incore_sb(ip->i_mount,
1220 XFS_SBS_FDBLOCKS, -((int64_t)diff), rsvd))
1225 ep = xfs_iext_get_ext(ifp, idx);
1226 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1227 xfs_bmap_trace_post_update(fname, "0", ip, idx, XFS_DATA_FORK);
1228 xfs_bmap_trace_pre_update(fname, "0", ip, idx + 2,
1230 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx + 2),
1231 NULLSTARTBLOCK((int)temp2));
1232 xfs_bmap_trace_post_update(fname, "0", ip, idx + 2,
1234 *dnew = temp + temp2;
1235 /* DELTA: One in-core extent is split in three. */
1236 temp = PREV.br_startoff;
1237 temp2 = PREV.br_blockcount;
1240 case MASK3(LEFT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1241 case MASK3(RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1242 case MASK2(LEFT_FILLING, RIGHT_CONTIG):
1243 case MASK2(RIGHT_FILLING, LEFT_CONTIG):
1244 case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1245 case MASK(LEFT_CONTIG):
1246 case MASK(RIGHT_CONTIG):
1248 * These cases are all impossible.
1255 if (delta->xed_startoff > temp)
1256 delta->xed_startoff = temp;
1257 if (delta->xed_blockcount < temp2)
1258 delta->xed_blockcount = temp2;
1272 #undef STATE_SET_TEST
1277 * Called by xfs_bmap_add_extent to handle cases converting an unwritten
1278 * allocation to a real allocation or vice versa.
1280 STATIC int /* error */
1281 xfs_bmap_add_extent_unwritten_real(
1282 xfs_inode_t *ip, /* incore inode pointer */
1283 xfs_extnum_t idx, /* extent number to update/insert */
1284 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
1285 xfs_bmbt_irec_t *new, /* new data to add to file extents */
1286 int *logflagsp, /* inode logging flags */
1287 xfs_extdelta_t *delta) /* Change made to incore extents */
1289 xfs_btree_cur_t *cur; /* btree cursor */
1290 xfs_bmbt_rec_t *ep; /* extent entry for idx */
1291 int error; /* error return value */
1292 #ifdef XFS_BMAP_TRACE
1293 static char fname[] = "xfs_bmap_add_extent_unwritten_real";
1295 int i; /* temp state */
1296 xfs_ifork_t *ifp; /* inode fork pointer */
1297 xfs_fileoff_t new_endoff; /* end offset of new entry */
1298 xfs_exntst_t newext; /* new extent state */
1299 xfs_exntst_t oldext; /* old extent state */
1300 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
1301 /* left is 0, right is 1, prev is 2 */
1302 int rval=0; /* return value (logging flags) */
1303 int state = 0;/* state bits, accessed thru macros */
1304 xfs_filblks_t temp=0;
1305 xfs_filblks_t temp2=0;
1306 enum { /* bit number definitions for state */
1307 LEFT_CONTIG, RIGHT_CONTIG,
1308 LEFT_FILLING, RIGHT_FILLING,
1309 LEFT_DELAY, RIGHT_DELAY,
1310 LEFT_VALID, RIGHT_VALID
1316 #define MASK(b) (1 << (b))
1317 #define MASK2(a,b) (MASK(a) | MASK(b))
1318 #define MASK3(a,b,c) (MASK2(a,b) | MASK(c))
1319 #define MASK4(a,b,c,d) (MASK3(a,b,c) | MASK(d))
1320 #define STATE_SET(b,v) ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1321 #define STATE_TEST(b) (state & MASK(b))
1322 #define STATE_SET_TEST(b,v) ((v) ? ((state |= MASK(b)), 1) : \
1323 ((state &= ~MASK(b)), 0))
1324 #define SWITCH_STATE \
1325 (state & MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG))
1328 * Set up a bunch of variables to make the tests simpler.
1332 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1333 ep = xfs_iext_get_ext(ifp, idx);
1334 xfs_bmbt_get_all(ep, &PREV);
1335 newext = new->br_state;
1336 oldext = (newext == XFS_EXT_UNWRITTEN) ?
1337 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
1338 ASSERT(PREV.br_state == oldext);
1339 new_endoff = new->br_startoff + new->br_blockcount;
1340 ASSERT(PREV.br_startoff <= new->br_startoff);
1341 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1343 * Set flags determining what part of the previous oldext allocation
1344 * extent is being replaced by a newext allocation.
1346 STATE_SET(LEFT_FILLING, PREV.br_startoff == new->br_startoff);
1347 STATE_SET(RIGHT_FILLING,
1348 PREV.br_startoff + PREV.br_blockcount == new_endoff);
1350 * Check and set flags if this segment has a left neighbor.
1351 * Don't set contiguous if the combined extent would be too large.
1353 if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1354 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &LEFT);
1355 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(LEFT.br_startblock));
1357 STATE_SET(LEFT_CONTIG,
1358 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
1359 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1360 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1361 LEFT.br_state == newext &&
1362 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1364 * Check and set flags if this segment has a right neighbor.
1365 * Don't set contiguous if the combined extent would be too large.
1366 * Also check for all-three-contiguous being too large.
1368 if (STATE_SET_TEST(RIGHT_VALID,
1370 ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1)) {
1371 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx + 1), &RIGHT);
1372 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(RIGHT.br_startblock));
1374 STATE_SET(RIGHT_CONTIG,
1375 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
1376 new_endoff == RIGHT.br_startoff &&
1377 new->br_startblock + new->br_blockcount ==
1378 RIGHT.br_startblock &&
1379 newext == RIGHT.br_state &&
1380 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1381 ((state & MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING)) !=
1382 MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING) ||
1383 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1386 * Switch out based on the FILLING and CONTIG state bits.
1388 switch (SWITCH_STATE) {
1390 case MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1392 * Setting all of a previous oldext extent to newext.
1393 * The left and right neighbors are both contiguous with new.
1395 xfs_bmap_trace_pre_update(fname, "LF|RF|LC|RC", ip, idx - 1,
1397 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1398 LEFT.br_blockcount + PREV.br_blockcount +
1399 RIGHT.br_blockcount);
1400 xfs_bmap_trace_post_update(fname, "LF|RF|LC|RC", ip, idx - 1,
1402 xfs_bmap_trace_delete(fname, "LF|RF|LC|RC", ip, idx, 2,
1404 xfs_iext_remove(ifp, idx, 2);
1405 ip->i_df.if_lastex = idx - 1;
1406 ip->i_d.di_nextents -= 2;
1408 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1410 rval = XFS_ILOG_CORE;
1411 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1412 RIGHT.br_startblock,
1413 RIGHT.br_blockcount, &i)))
1416 if ((error = xfs_bmbt_delete(cur, &i)))
1419 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1422 if ((error = xfs_bmbt_delete(cur, &i)))
1425 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1428 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1430 LEFT.br_blockcount + PREV.br_blockcount +
1431 RIGHT.br_blockcount, LEFT.br_state)))
1434 /* DELTA: Three in-core extents are replaced by one. */
1435 temp = LEFT.br_startoff;
1436 temp2 = LEFT.br_blockcount +
1437 PREV.br_blockcount +
1438 RIGHT.br_blockcount;
1441 case MASK3(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG):
1443 * Setting all of a previous oldext extent to newext.
1444 * The left neighbor is contiguous, the right is not.
1446 xfs_bmap_trace_pre_update(fname, "LF|RF|LC", ip, idx - 1,
1448 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1449 LEFT.br_blockcount + PREV.br_blockcount);
1450 xfs_bmap_trace_post_update(fname, "LF|RF|LC", ip, idx - 1,
1452 ip->i_df.if_lastex = idx - 1;
1453 xfs_bmap_trace_delete(fname, "LF|RF|LC", ip, idx, 1,
1455 xfs_iext_remove(ifp, idx, 1);
1456 ip->i_d.di_nextents--;
1458 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1460 rval = XFS_ILOG_CORE;
1461 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1462 PREV.br_startblock, PREV.br_blockcount,
1466 if ((error = xfs_bmbt_delete(cur, &i)))
1469 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1472 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1474 LEFT.br_blockcount + PREV.br_blockcount,
1478 /* DELTA: Two in-core extents are replaced by one. */
1479 temp = LEFT.br_startoff;
1480 temp2 = LEFT.br_blockcount +
1484 case MASK3(LEFT_FILLING, RIGHT_FILLING, RIGHT_CONTIG):
1486 * Setting all of a previous oldext extent to newext.
1487 * The right neighbor is contiguous, the left is not.
1489 xfs_bmap_trace_pre_update(fname, "LF|RF|RC", ip, idx,
1491 xfs_bmbt_set_blockcount(ep,
1492 PREV.br_blockcount + RIGHT.br_blockcount);
1493 xfs_bmbt_set_state(ep, newext);
1494 xfs_bmap_trace_post_update(fname, "LF|RF|RC", ip, idx,
1496 ip->i_df.if_lastex = idx;
1497 xfs_bmap_trace_delete(fname, "LF|RF|RC", ip, idx + 1, 1,
1499 xfs_iext_remove(ifp, idx + 1, 1);
1500 ip->i_d.di_nextents--;
1502 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1504 rval = XFS_ILOG_CORE;
1505 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1506 RIGHT.br_startblock,
1507 RIGHT.br_blockcount, &i)))
1510 if ((error = xfs_bmbt_delete(cur, &i)))
1513 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1516 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1518 new->br_blockcount + RIGHT.br_blockcount,
1522 /* DELTA: Two in-core extents are replaced by one. */
1523 temp = PREV.br_startoff;
1524 temp2 = PREV.br_blockcount +
1525 RIGHT.br_blockcount;
1528 case MASK2(LEFT_FILLING, RIGHT_FILLING):
1530 * Setting all of a previous oldext extent to newext.
1531 * Neither the left nor right neighbors are contiguous with
1534 xfs_bmap_trace_pre_update(fname, "LF|RF", ip, idx,
1536 xfs_bmbt_set_state(ep, newext);
1537 xfs_bmap_trace_post_update(fname, "LF|RF", ip, idx,
1539 ip->i_df.if_lastex = idx;
1541 rval = XFS_ILOG_DEXT;
1544 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1545 new->br_startblock, new->br_blockcount,
1549 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1550 new->br_startblock, new->br_blockcount,
1554 /* DELTA: The in-core extent described by new changed type. */
1555 temp = new->br_startoff;
1556 temp2 = new->br_blockcount;
1559 case MASK2(LEFT_FILLING, LEFT_CONTIG):
1561 * Setting the first part of a previous oldext extent to newext.
1562 * The left neighbor is contiguous.
1564 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx - 1,
1566 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1567 LEFT.br_blockcount + new->br_blockcount);
1568 xfs_bmbt_set_startoff(ep,
1569 PREV.br_startoff + new->br_blockcount);
1570 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx - 1,
1572 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx,
1574 xfs_bmbt_set_startblock(ep,
1575 new->br_startblock + new->br_blockcount);
1576 xfs_bmbt_set_blockcount(ep,
1577 PREV.br_blockcount - new->br_blockcount);
1578 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx,
1580 ip->i_df.if_lastex = idx - 1;
1582 rval = XFS_ILOG_DEXT;
1585 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1586 PREV.br_startblock, PREV.br_blockcount,
1590 if ((error = xfs_bmbt_update(cur,
1591 PREV.br_startoff + new->br_blockcount,
1592 PREV.br_startblock + new->br_blockcount,
1593 PREV.br_blockcount - new->br_blockcount,
1596 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1598 if (xfs_bmbt_update(cur, LEFT.br_startoff,
1600 LEFT.br_blockcount + new->br_blockcount,
1604 /* DELTA: The boundary between two in-core extents moved. */
1605 temp = LEFT.br_startoff;
1606 temp2 = LEFT.br_blockcount +
1610 case MASK(LEFT_FILLING):
1612 * Setting the first part of a previous oldext extent to newext.
1613 * The left neighbor is not contiguous.
1615 xfs_bmap_trace_pre_update(fname, "LF", ip, idx, XFS_DATA_FORK);
1616 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1617 xfs_bmbt_set_startoff(ep, new_endoff);
1618 xfs_bmbt_set_blockcount(ep,
1619 PREV.br_blockcount - new->br_blockcount);
1620 xfs_bmbt_set_startblock(ep,
1621 new->br_startblock + new->br_blockcount);
1622 xfs_bmap_trace_post_update(fname, "LF", ip, idx, XFS_DATA_FORK);
1623 xfs_bmap_trace_insert(fname, "LF", ip, idx, 1, new, NULL,
1625 xfs_iext_insert(ifp, idx, 1, new);
1626 ip->i_df.if_lastex = idx;
1627 ip->i_d.di_nextents++;
1629 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1631 rval = XFS_ILOG_CORE;
1632 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1633 PREV.br_startblock, PREV.br_blockcount,
1637 if ((error = xfs_bmbt_update(cur,
1638 PREV.br_startoff + new->br_blockcount,
1639 PREV.br_startblock + new->br_blockcount,
1640 PREV.br_blockcount - new->br_blockcount,
1643 cur->bc_rec.b = *new;
1644 if ((error = xfs_bmbt_insert(cur, &i)))
1648 /* DELTA: One in-core extent is split in two. */
1649 temp = PREV.br_startoff;
1650 temp2 = PREV.br_blockcount;
1653 case MASK2(RIGHT_FILLING, RIGHT_CONTIG):
1655 * Setting the last part of a previous oldext extent to newext.
1656 * The right neighbor is contiguous with the new allocation.
1658 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx,
1660 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx + 1,
1662 xfs_bmbt_set_blockcount(ep,
1663 PREV.br_blockcount - new->br_blockcount);
1664 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx,
1666 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, idx + 1),
1667 new->br_startoff, new->br_startblock,
1668 new->br_blockcount + RIGHT.br_blockcount, newext);
1669 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx + 1,
1671 ip->i_df.if_lastex = idx + 1;
1673 rval = XFS_ILOG_DEXT;
1676 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1678 PREV.br_blockcount, &i)))
1681 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1683 PREV.br_blockcount - new->br_blockcount,
1686 if ((error = xfs_bmbt_increment(cur, 0, &i)))
1688 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1690 new->br_blockcount + RIGHT.br_blockcount,
1694 /* DELTA: The boundary between two in-core extents moved. */
1695 temp = PREV.br_startoff;
1696 temp2 = PREV.br_blockcount +
1697 RIGHT.br_blockcount;
1700 case MASK(RIGHT_FILLING):
1702 * Setting the last part of a previous oldext extent to newext.
1703 * The right neighbor is not contiguous.
1705 xfs_bmap_trace_pre_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1706 xfs_bmbt_set_blockcount(ep,
1707 PREV.br_blockcount - new->br_blockcount);
1708 xfs_bmap_trace_post_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1709 xfs_bmap_trace_insert(fname, "RF", ip, idx + 1, 1,
1710 new, NULL, XFS_DATA_FORK);
1711 xfs_iext_insert(ifp, idx + 1, 1, new);
1712 ip->i_df.if_lastex = idx + 1;
1713 ip->i_d.di_nextents++;
1715 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1717 rval = XFS_ILOG_CORE;
1718 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1719 PREV.br_startblock, PREV.br_blockcount,
1723 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1725 PREV.br_blockcount - new->br_blockcount,
1728 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1729 new->br_startblock, new->br_blockcount,
1733 cur->bc_rec.b.br_state = XFS_EXT_NORM;
1734 if ((error = xfs_bmbt_insert(cur, &i)))
1738 /* DELTA: One in-core extent is split in two. */
1739 temp = PREV.br_startoff;
1740 temp2 = PREV.br_blockcount;
1745 * Setting the middle part of a previous oldext extent to
1746 * newext. Contiguity is impossible here.
1747 * One extent becomes three extents.
1749 xfs_bmap_trace_pre_update(fname, "0", ip, idx, XFS_DATA_FORK);
1750 xfs_bmbt_set_blockcount(ep,
1751 new->br_startoff - PREV.br_startoff);
1752 xfs_bmap_trace_post_update(fname, "0", ip, idx, XFS_DATA_FORK);
1754 r[1].br_startoff = new_endoff;
1755 r[1].br_blockcount =
1756 PREV.br_startoff + PREV.br_blockcount - new_endoff;
1757 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1758 r[1].br_state = oldext;
1759 xfs_bmap_trace_insert(fname, "0", ip, idx + 1, 2, &r[0], &r[1],
1761 xfs_iext_insert(ifp, idx + 1, 2, &r[0]);
1762 ip->i_df.if_lastex = idx + 1;
1763 ip->i_d.di_nextents += 2;
1765 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1767 rval = XFS_ILOG_CORE;
1768 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1769 PREV.br_startblock, PREV.br_blockcount,
1773 /* new right extent - oldext */
1774 if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1775 r[1].br_startblock, r[1].br_blockcount,
1778 /* new left extent - oldext */
1779 PREV.br_blockcount =
1780 new->br_startoff - PREV.br_startoff;
1781 cur->bc_rec.b = PREV;
1782 if ((error = xfs_bmbt_insert(cur, &i)))
1785 if ((error = xfs_bmbt_increment(cur, 0, &i)))
1788 /* new middle extent - newext */
1789 cur->bc_rec.b = *new;
1790 if ((error = xfs_bmbt_insert(cur, &i)))
1794 /* DELTA: One in-core extent is split in three. */
1795 temp = PREV.br_startoff;
1796 temp2 = PREV.br_blockcount;
1799 case MASK3(LEFT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1800 case MASK3(RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1801 case MASK2(LEFT_FILLING, RIGHT_CONTIG):
1802 case MASK2(RIGHT_FILLING, LEFT_CONTIG):
1803 case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1804 case MASK(LEFT_CONTIG):
1805 case MASK(RIGHT_CONTIG):
1807 * These cases are all impossible.
1814 if (delta->xed_startoff > temp)
1815 delta->xed_startoff = temp;
1816 if (delta->xed_blockcount < temp2)
1817 delta->xed_blockcount = temp2;
1831 #undef STATE_SET_TEST
1836 * Called by xfs_bmap_add_extent to handle cases converting a hole
1837 * to a delayed allocation.
1840 STATIC int /* error */
1841 xfs_bmap_add_extent_hole_delay(
1842 xfs_inode_t *ip, /* incore inode pointer */
1843 xfs_extnum_t idx, /* extent number to update/insert */
1844 xfs_btree_cur_t *cur, /* if null, not a btree */
1845 xfs_bmbt_irec_t *new, /* new data to add to file extents */
1846 int *logflagsp, /* inode logging flags */
1847 xfs_extdelta_t *delta, /* Change made to incore extents */
1848 int rsvd) /* OK to allocate reserved blocks */
1850 xfs_bmbt_rec_t *ep; /* extent record for idx */
1851 #ifdef XFS_BMAP_TRACE
1852 static char fname[] = "xfs_bmap_add_extent_hole_delay";
1854 xfs_ifork_t *ifp; /* inode fork pointer */
1855 xfs_bmbt_irec_t left; /* left neighbor extent entry */
1856 xfs_filblks_t newlen=0; /* new indirect size */
1857 xfs_filblks_t oldlen=0; /* old indirect size */
1858 xfs_bmbt_irec_t right; /* right neighbor extent entry */
1859 int state; /* state bits, accessed thru macros */
1860 xfs_filblks_t temp=0; /* temp for indirect calculations */
1861 xfs_filblks_t temp2=0;
1862 enum { /* bit number definitions for state */
1863 LEFT_CONTIG, RIGHT_CONTIG,
1864 LEFT_DELAY, RIGHT_DELAY,
1865 LEFT_VALID, RIGHT_VALID
1868 #define MASK(b) (1 << (b))
1869 #define MASK2(a,b) (MASK(a) | MASK(b))
1870 #define STATE_SET(b,v) ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1871 #define STATE_TEST(b) (state & MASK(b))
1872 #define STATE_SET_TEST(b,v) ((v) ? ((state |= MASK(b)), 1) : \
1873 ((state &= ~MASK(b)), 0))
1874 #define SWITCH_STATE (state & MASK2(LEFT_CONTIG, RIGHT_CONTIG))
1876 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1877 ep = xfs_iext_get_ext(ifp, idx);
1879 ASSERT(ISNULLSTARTBLOCK(new->br_startblock));
1881 * Check and set flags if this segment has a left neighbor
1883 if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1884 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &left);
1885 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(left.br_startblock));
1888 * Check and set flags if the current (right) segment exists.
1889 * If it doesn't exist, we're converting the hole at end-of-file.
1891 if (STATE_SET_TEST(RIGHT_VALID,
1893 ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1894 xfs_bmbt_get_all(ep, &right);
1895 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(right.br_startblock));
1898 * Set contiguity flags on the left and right neighbors.
1899 * Don't let extents get too large, even if the pieces are contiguous.
1901 STATE_SET(LEFT_CONTIG,
1902 STATE_TEST(LEFT_VALID) && STATE_TEST(LEFT_DELAY) &&
1903 left.br_startoff + left.br_blockcount == new->br_startoff &&
1904 left.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1905 STATE_SET(RIGHT_CONTIG,
1906 STATE_TEST(RIGHT_VALID) && STATE_TEST(RIGHT_DELAY) &&
1907 new->br_startoff + new->br_blockcount == right.br_startoff &&
1908 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1909 (!STATE_TEST(LEFT_CONTIG) ||
1910 (left.br_blockcount + new->br_blockcount +
1911 right.br_blockcount <= MAXEXTLEN)));
1913 * Switch out based on the contiguity flags.
1915 switch (SWITCH_STATE) {
1917 case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1919 * New allocation is contiguous with delayed allocations
1920 * on the left and on the right.
1921 * Merge all three into a single extent record.
1923 temp = left.br_blockcount + new->br_blockcount +
1924 right.br_blockcount;
1925 xfs_bmap_trace_pre_update(fname, "LC|RC", ip, idx - 1,
1927 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1), temp);
1928 oldlen = STARTBLOCKVAL(left.br_startblock) +
1929 STARTBLOCKVAL(new->br_startblock) +
1930 STARTBLOCKVAL(right.br_startblock);
1931 newlen = xfs_bmap_worst_indlen(ip, temp);
1932 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx - 1),
1933 NULLSTARTBLOCK((int)newlen));
1934 xfs_bmap_trace_post_update(fname, "LC|RC", ip, idx - 1,
1936 xfs_bmap_trace_delete(fname, "LC|RC", ip, idx, 1,
1938 xfs_iext_remove(ifp, idx, 1);
1939 ip->i_df.if_lastex = idx - 1;
1940 /* DELTA: Two in-core extents were replaced by one. */
1942 temp = left.br_startoff;
1945 case MASK(LEFT_CONTIG):
1947 * New allocation is contiguous with a delayed allocation
1949 * Merge the new allocation with the left neighbor.
1951 temp = left.br_blockcount + new->br_blockcount;
1952 xfs_bmap_trace_pre_update(fname, "LC", ip, idx - 1,
1954 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1), temp);
1955 oldlen = STARTBLOCKVAL(left.br_startblock) +
1956 STARTBLOCKVAL(new->br_startblock);
1957 newlen = xfs_bmap_worst_indlen(ip, temp);
1958 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx - 1),
1959 NULLSTARTBLOCK((int)newlen));
1960 xfs_bmap_trace_post_update(fname, "LC", ip, idx - 1,
1962 ip->i_df.if_lastex = idx - 1;
1963 /* DELTA: One in-core extent grew into a hole. */
1965 temp = left.br_startoff;
1968 case MASK(RIGHT_CONTIG):
1970 * New allocation is contiguous with a delayed allocation
1972 * Merge the new allocation with the right neighbor.
1974 xfs_bmap_trace_pre_update(fname, "RC", ip, idx, XFS_DATA_FORK);
1975 temp = new->br_blockcount + right.br_blockcount;
1976 oldlen = STARTBLOCKVAL(new->br_startblock) +
1977 STARTBLOCKVAL(right.br_startblock);
1978 newlen = xfs_bmap_worst_indlen(ip, temp);
1979 xfs_bmbt_set_allf(ep, new->br_startoff,
1980 NULLSTARTBLOCK((int)newlen), temp, right.br_state);
1981 xfs_bmap_trace_post_update(fname, "RC", ip, idx, XFS_DATA_FORK);
1982 ip->i_df.if_lastex = idx;
1983 /* DELTA: One in-core extent grew into a hole. */
1985 temp = new->br_startoff;
1990 * New allocation is not contiguous with another
1991 * delayed allocation.
1992 * Insert a new entry.
1994 oldlen = newlen = 0;
1995 xfs_bmap_trace_insert(fname, "0", ip, idx, 1, new, NULL,
1997 xfs_iext_insert(ifp, idx, 1, new);
1998 ip->i_df.if_lastex = idx;
1999 /* DELTA: A new in-core extent was added in a hole. */
2000 temp2 = new->br_blockcount;
2001 temp = new->br_startoff;
2004 if (oldlen != newlen) {
2005 ASSERT(oldlen > newlen);
2006 xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
2007 (int64_t)(oldlen - newlen), rsvd);
2009 * Nothing to do for disk quota accounting here.
2014 if (delta->xed_startoff > temp)
2015 delta->xed_startoff = temp;
2016 if (delta->xed_blockcount < temp2)
2017 delta->xed_blockcount = temp2;
2025 #undef STATE_SET_TEST
2030 * Called by xfs_bmap_add_extent to handle cases converting a hole
2031 * to a real allocation.
2033 STATIC int /* error */
2034 xfs_bmap_add_extent_hole_real(
2035 xfs_inode_t *ip, /* incore inode pointer */
2036 xfs_extnum_t idx, /* extent number to update/insert */
2037 xfs_btree_cur_t *cur, /* if null, not a btree */
2038 xfs_bmbt_irec_t *new, /* new data to add to file extents */
2039 int *logflagsp, /* inode logging flags */
2040 xfs_extdelta_t *delta, /* Change made to incore extents */
2041 int whichfork) /* data or attr fork */
2043 xfs_bmbt_rec_t *ep; /* pointer to extent entry ins. point */
2044 int error; /* error return value */
2045 #ifdef XFS_BMAP_TRACE
2046 static char fname[] = "xfs_bmap_add_extent_hole_real";
2048 int i; /* temp state */
2049 xfs_ifork_t *ifp; /* inode fork pointer */
2050 xfs_bmbt_irec_t left; /* left neighbor extent entry */
2051 xfs_bmbt_irec_t right; /* right neighbor extent entry */
2052 int rval=0; /* return value (logging flags) */
2053 int state; /* state bits, accessed thru macros */
2054 xfs_filblks_t temp=0;
2055 xfs_filblks_t temp2=0;
2056 enum { /* bit number definitions for state */
2057 LEFT_CONTIG, RIGHT_CONTIG,
2058 LEFT_DELAY, RIGHT_DELAY,
2059 LEFT_VALID, RIGHT_VALID
2062 #define MASK(b) (1 << (b))
2063 #define MASK2(a,b) (MASK(a) | MASK(b))
2064 #define STATE_SET(b,v) ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
2065 #define STATE_TEST(b) (state & MASK(b))
2066 #define STATE_SET_TEST(b,v) ((v) ? ((state |= MASK(b)), 1) : \
2067 ((state &= ~MASK(b)), 0))
2068 #define SWITCH_STATE (state & MASK2(LEFT_CONTIG, RIGHT_CONTIG))
2070 ifp = XFS_IFORK_PTR(ip, whichfork);
2071 ASSERT(idx <= ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
2072 ep = xfs_iext_get_ext(ifp, idx);
2075 * Check and set flags if this segment has a left neighbor.
2077 if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
2078 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &left);
2079 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(left.br_startblock));
2082 * Check and set flags if this segment has a current value.
2083 * Not true if we're inserting into the "hole" at eof.
2085 if (STATE_SET_TEST(RIGHT_VALID,
2087 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
2088 xfs_bmbt_get_all(ep, &right);
2089 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(right.br_startblock));
2092 * We're inserting a real allocation between "left" and "right".
2093 * Set the contiguity flags. Don't let extents get too large.
2095 STATE_SET(LEFT_CONTIG,
2096 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
2097 left.br_startoff + left.br_blockcount == new->br_startoff &&
2098 left.br_startblock + left.br_blockcount == new->br_startblock &&
2099 left.br_state == new->br_state &&
2100 left.br_blockcount + new->br_blockcount <= MAXEXTLEN);
2101 STATE_SET(RIGHT_CONTIG,
2102 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
2103 new->br_startoff + new->br_blockcount == right.br_startoff &&
2104 new->br_startblock + new->br_blockcount ==
2105 right.br_startblock &&
2106 new->br_state == right.br_state &&
2107 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2108 (!STATE_TEST(LEFT_CONTIG) ||
2109 left.br_blockcount + new->br_blockcount +
2110 right.br_blockcount <= MAXEXTLEN));
2114 * Select which case we're in here, and implement it.
2116 switch (SWITCH_STATE) {
2118 case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
2120 * New allocation is contiguous with real allocations on the
2121 * left and on the right.
2122 * Merge all three into a single extent record.
2124 xfs_bmap_trace_pre_update(fname, "LC|RC", ip, idx - 1,
2126 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
2127 left.br_blockcount + new->br_blockcount +
2128 right.br_blockcount);
2129 xfs_bmap_trace_post_update(fname, "LC|RC", ip, idx - 1,
2131 xfs_bmap_trace_delete(fname, "LC|RC", ip,
2133 xfs_iext_remove(ifp, idx, 1);
2134 ifp->if_lastex = idx - 1;
2135 XFS_IFORK_NEXT_SET(ip, whichfork,
2136 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2138 rval = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2140 rval = XFS_ILOG_CORE;
2141 if ((error = xfs_bmbt_lookup_eq(cur,
2143 right.br_startblock,
2144 right.br_blockcount, &i)))
2147 if ((error = xfs_bmbt_delete(cur, &i)))
2150 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
2153 if ((error = xfs_bmbt_update(cur, left.br_startoff,
2155 left.br_blockcount +
2156 new->br_blockcount +
2157 right.br_blockcount,
2161 /* DELTA: Two in-core extents were replaced by one. */
2162 temp = left.br_startoff;
2163 temp2 = left.br_blockcount +
2164 new->br_blockcount +
2165 right.br_blockcount;
2168 case MASK(LEFT_CONTIG):
2170 * New allocation is contiguous with a real allocation
2172 * Merge the new allocation with the left neighbor.
2174 xfs_bmap_trace_pre_update(fname, "LC", ip, idx - 1, whichfork);
2175 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
2176 left.br_blockcount + new->br_blockcount);
2177 xfs_bmap_trace_post_update(fname, "LC", ip, idx - 1, whichfork);
2178 ifp->if_lastex = idx - 1;
2180 rval = XFS_ILOG_FEXT(whichfork);
2183 if ((error = xfs_bmbt_lookup_eq(cur,
2186 left.br_blockcount, &i)))
2189 if ((error = xfs_bmbt_update(cur, left.br_startoff,
2191 left.br_blockcount +
2196 /* DELTA: One in-core extent grew. */
2197 temp = left.br_startoff;
2198 temp2 = left.br_blockcount +
2202 case MASK(RIGHT_CONTIG):
2204 * New allocation is contiguous with a real allocation
2206 * Merge the new allocation with the right neighbor.
2208 xfs_bmap_trace_pre_update(fname, "RC", ip, idx, whichfork);
2209 xfs_bmbt_set_allf(ep, new->br_startoff, new->br_startblock,
2210 new->br_blockcount + right.br_blockcount,
2212 xfs_bmap_trace_post_update(fname, "RC", ip, idx, whichfork);
2213 ifp->if_lastex = idx;
2215 rval = XFS_ILOG_FEXT(whichfork);
2218 if ((error = xfs_bmbt_lookup_eq(cur,
2220 right.br_startblock,
2221 right.br_blockcount, &i)))
2224 if ((error = xfs_bmbt_update(cur, new->br_startoff,
2226 new->br_blockcount +
2227 right.br_blockcount,
2231 /* DELTA: One in-core extent grew. */
2232 temp = new->br_startoff;
2233 temp2 = new->br_blockcount +
2234 right.br_blockcount;
2239 * New allocation is not contiguous with another
2241 * Insert a new entry.
2243 xfs_bmap_trace_insert(fname, "0", ip, idx, 1, new, NULL,
2245 xfs_iext_insert(ifp, idx, 1, new);
2246 ifp->if_lastex = idx;
2247 XFS_IFORK_NEXT_SET(ip, whichfork,
2248 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2250 rval = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2252 rval = XFS_ILOG_CORE;
2253 if ((error = xfs_bmbt_lookup_eq(cur,
2256 new->br_blockcount, &i)))
2259 cur->bc_rec.b.br_state = new->br_state;
2260 if ((error = xfs_bmbt_insert(cur, &i)))
2264 /* DELTA: A new extent was added in a hole. */
2265 temp = new->br_startoff;
2266 temp2 = new->br_blockcount;
2271 if (delta->xed_startoff > temp)
2272 delta->xed_startoff = temp;
2273 if (delta->xed_blockcount < temp2)
2274 delta->xed_blockcount = temp2;
2283 #undef STATE_SET_TEST
2288 * Adjust the size of the new extent based on di_extsize and rt extsize.
2291 xfs_bmap_extsize_align(
2293 xfs_bmbt_irec_t *gotp, /* next extent pointer */
2294 xfs_bmbt_irec_t *prevp, /* previous extent pointer */
2295 xfs_extlen_t extsz, /* align to this extent size */
2296 int rt, /* is this a realtime inode? */
2297 int eof, /* is extent at end-of-file? */
2298 int delay, /* creating delalloc extent? */
2299 int convert, /* overwriting unwritten extent? */
2300 xfs_fileoff_t *offp, /* in/out: aligned offset */
2301 xfs_extlen_t *lenp) /* in/out: aligned length */
2303 xfs_fileoff_t orig_off; /* original offset */
2304 xfs_extlen_t orig_alen; /* original length */
2305 xfs_fileoff_t orig_end; /* original off+len */
2306 xfs_fileoff_t nexto; /* next file offset */
2307 xfs_fileoff_t prevo; /* previous file offset */
2308 xfs_fileoff_t align_off; /* temp for offset */
2309 xfs_extlen_t align_alen; /* temp for length */
2310 xfs_extlen_t temp; /* temp for calculations */
2315 orig_off = align_off = *offp;
2316 orig_alen = align_alen = *lenp;
2317 orig_end = orig_off + orig_alen;
2320 * If this request overlaps an existing extent, then don't
2321 * attempt to perform any additional alignment.
2323 if (!delay && !eof &&
2324 (orig_off >= gotp->br_startoff) &&
2325 (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2330 * If the file offset is unaligned vs. the extent size
2331 * we need to align it. This will be possible unless
2332 * the file was previously written with a kernel that didn't
2333 * perform this alignment, or if a truncate shot us in the
2336 temp = do_mod(orig_off, extsz);
2342 * Same adjustment for the end of the requested area.
2344 if ((temp = (align_alen % extsz))) {
2345 align_alen += extsz - temp;
2348 * If the previous block overlaps with this proposed allocation
2349 * then move the start forward without adjusting the length.
2351 if (prevp->br_startoff != NULLFILEOFF) {
2352 if (prevp->br_startblock == HOLESTARTBLOCK)
2353 prevo = prevp->br_startoff;
2355 prevo = prevp->br_startoff + prevp->br_blockcount;
2358 if (align_off != orig_off && align_off < prevo)
2361 * If the next block overlaps with this proposed allocation
2362 * then move the start back without adjusting the length,
2363 * but not before offset 0.
2364 * This may of course make the start overlap previous block,
2365 * and if we hit the offset 0 limit then the next block
2366 * can still overlap too.
2368 if (!eof && gotp->br_startoff != NULLFILEOFF) {
2369 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
2370 (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
2371 nexto = gotp->br_startoff + gotp->br_blockcount;
2373 nexto = gotp->br_startoff;
2375 nexto = NULLFILEOFF;
2377 align_off + align_alen != orig_end &&
2378 align_off + align_alen > nexto)
2379 align_off = nexto > align_alen ? nexto - align_alen : 0;
2381 * If we're now overlapping the next or previous extent that
2382 * means we can't fit an extsz piece in this hole. Just move
2383 * the start forward to the first valid spot and set
2384 * the length so we hit the end.
2386 if (align_off != orig_off && align_off < prevo)
2388 if (align_off + align_alen != orig_end &&
2389 align_off + align_alen > nexto &&
2390 nexto != NULLFILEOFF) {
2391 ASSERT(nexto > prevo);
2392 align_alen = nexto - align_off;
2396 * If realtime, and the result isn't a multiple of the realtime
2397 * extent size we need to remove blocks until it is.
2399 if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
2401 * We're not covering the original request, or
2402 * we won't be able to once we fix the length.
2404 if (orig_off < align_off ||
2405 orig_end > align_off + align_alen ||
2406 align_alen - temp < orig_alen)
2407 return XFS_ERROR(EINVAL);
2409 * Try to fix it by moving the start up.
2411 if (align_off + temp <= orig_off) {
2416 * Try to fix it by moving the end in.
2418 else if (align_off + align_alen - temp >= orig_end)
2421 * Set the start to the minimum then trim the length.
2424 align_alen -= orig_off - align_off;
2425 align_off = orig_off;
2426 align_alen -= align_alen % mp->m_sb.sb_rextsize;
2429 * Result doesn't cover the request, fail it.
2431 if (orig_off < align_off || orig_end > align_off + align_alen)
2432 return XFS_ERROR(EINVAL);
2434 ASSERT(orig_off >= align_off);
2435 ASSERT(orig_end <= align_off + align_alen);
2439 if (!eof && gotp->br_startoff != NULLFILEOFF)
2440 ASSERT(align_off + align_alen <= gotp->br_startoff);
2441 if (prevp->br_startoff != NULLFILEOFF)
2442 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
2450 #define XFS_ALLOC_GAP_UNITS 4
2454 xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2456 xfs_fsblock_t adjust; /* adjustment to block numbers */
2457 xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
2458 xfs_mount_t *mp; /* mount point structure */
2459 int nullfb; /* true if ap->firstblock isn't set */
2460 int rt; /* true if inode is realtime */
2462 #define ISVALID(x,y) \
2464 (x) < mp->m_sb.sb_rblocks : \
2465 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2466 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2467 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2469 mp = ap->ip->i_mount;
2470 nullfb = ap->firstblock == NULLFSBLOCK;
2471 rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2472 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2474 * If allocating at eof, and there's a previous real block,
2475 * try to use it's last block as our starting point.
2477 if (ap->eof && ap->prevp->br_startoff != NULLFILEOFF &&
2478 !ISNULLSTARTBLOCK(ap->prevp->br_startblock) &&
2479 ISVALID(ap->prevp->br_startblock + ap->prevp->br_blockcount,
2480 ap->prevp->br_startblock)) {
2481 ap->rval = ap->prevp->br_startblock + ap->prevp->br_blockcount;
2483 * Adjust for the gap between prevp and us.
2486 (ap->prevp->br_startoff + ap->prevp->br_blockcount);
2488 ISVALID(ap->rval + adjust, ap->prevp->br_startblock))
2492 * If not at eof, then compare the two neighbor blocks.
2493 * Figure out whether either one gives us a good starting point,
2494 * and pick the better one.
2496 else if (!ap->eof) {
2497 xfs_fsblock_t gotbno; /* right side block number */
2498 xfs_fsblock_t gotdiff=0; /* right side difference */
2499 xfs_fsblock_t prevbno; /* left side block number */
2500 xfs_fsblock_t prevdiff=0; /* left side difference */
2503 * If there's a previous (left) block, select a requested
2504 * start block based on it.
2506 if (ap->prevp->br_startoff != NULLFILEOFF &&
2507 !ISNULLSTARTBLOCK(ap->prevp->br_startblock) &&
2508 (prevbno = ap->prevp->br_startblock +
2509 ap->prevp->br_blockcount) &&
2510 ISVALID(prevbno, ap->prevp->br_startblock)) {
2512 * Calculate gap to end of previous block.
2514 adjust = prevdiff = ap->off -
2515 (ap->prevp->br_startoff +
2516 ap->prevp->br_blockcount);
2518 * Figure the startblock based on the previous block's
2519 * end and the gap size.
2521 * If the gap is large relative to the piece we're
2522 * allocating, or using it gives us an invalid block
2523 * number, then just use the end of the previous block.
2525 if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2526 ISVALID(prevbno + prevdiff,
2527 ap->prevp->br_startblock))
2532 * If the firstblock forbids it, can't use it,
2535 if (!rt && !nullfb &&
2536 XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2537 prevbno = NULLFSBLOCK;
2540 * No previous block or can't follow it, just default.
2543 prevbno = NULLFSBLOCK;
2545 * If there's a following (right) block, select a requested
2546 * start block based on it.
2548 if (!ISNULLSTARTBLOCK(ap->gotp->br_startblock)) {
2550 * Calculate gap to start of next block.
2552 adjust = gotdiff = ap->gotp->br_startoff - ap->off;
2554 * Figure the startblock based on the next block's
2555 * start and the gap size.
2557 gotbno = ap->gotp->br_startblock;
2560 * If the gap is large relative to the piece we're
2561 * allocating, or using it gives us an invalid block
2562 * number, then just use the start of the next block
2563 * offset by our length.
2565 if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2566 ISVALID(gotbno - gotdiff, gotbno))
2568 else if (ISVALID(gotbno - ap->alen, gotbno)) {
2570 gotdiff += adjust - ap->alen;
2574 * If the firstblock forbids it, can't use it,
2577 if (!rt && !nullfb &&
2578 XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2579 gotbno = NULLFSBLOCK;
2582 * No next block, just default.
2585 gotbno = NULLFSBLOCK;
2587 * If both valid, pick the better one, else the only good
2588 * one, else ap->rval is already set (to 0 or the inode block).
2590 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2591 ap->rval = prevdiff <= gotdiff ? prevbno : gotbno;
2592 else if (prevbno != NULLFSBLOCK)
2594 else if (gotbno != NULLFSBLOCK)
2603 xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2605 xfs_alloctype_t atype = 0; /* type for allocation routines */
2606 int error; /* error return value */
2607 xfs_mount_t *mp; /* mount point structure */
2608 xfs_extlen_t prod = 0; /* product factor for allocators */
2609 xfs_extlen_t ralen = 0; /* realtime allocation length */
2610 xfs_extlen_t align; /* minimum allocation alignment */
2611 xfs_rtblock_t rtx; /* realtime extent number */
2614 mp = ap->ip->i_mount;
2615 align = ap->ip->i_d.di_extsize ?
2616 ap->ip->i_d.di_extsize : mp->m_sb.sb_rextsize;
2617 prod = align / mp->m_sb.sb_rextsize;
2618 error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2619 align, 1, ap->eof, 0,
2620 ap->conv, &ap->off, &ap->alen);
2624 ASSERT(ap->alen % mp->m_sb.sb_rextsize == 0);
2627 * If the offset & length are not perfectly aligned
2628 * then kill prod, it will just get us in trouble.
2630 if (do_mod(ap->off, align) || ap->alen % align)
2633 * Set ralen to be the actual requested length in rtextents.
2635 ralen = ap->alen / mp->m_sb.sb_rextsize;
2637 * If the old value was close enough to MAXEXTLEN that
2638 * we rounded up to it, cut it back so it's valid again.
2639 * Note that if it's a really large request (bigger than
2640 * MAXEXTLEN), we don't hear about that number, and can't
2641 * adjust the starting point to match it.
2643 if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2644 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2646 * If it's an allocation to an empty file at offset 0,
2647 * pick an extent that will space things out in the rt area.
2649 if (ap->eof && ap->off == 0) {
2650 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2653 ap->rval = rtx * mp->m_sb.sb_rextsize;
2658 xfs_bmap_adjacent(ap);
2661 * Realtime allocation, done through xfs_rtallocate_extent.
2663 atype = ap->rval == 0 ? XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2664 do_div(ap->rval, mp->m_sb.sb_rextsize);
2667 if ((error = xfs_rtallocate_extent(ap->tp, ap->rval, 1, ap->alen,
2668 &ralen, atype, ap->wasdel, prod, &rtb)))
2670 if (rtb == NULLFSBLOCK && prod > 1 &&
2671 (error = xfs_rtallocate_extent(ap->tp, ap->rval, 1,
2672 ap->alen, &ralen, atype,
2673 ap->wasdel, 1, &rtb)))
2676 if (ap->rval != NULLFSBLOCK) {
2677 ap->rval *= mp->m_sb.sb_rextsize;
2678 ralen *= mp->m_sb.sb_rextsize;
2680 ap->ip->i_d.di_nblocks += ralen;
2681 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2683 ap->ip->i_delayed_blks -= ralen;
2685 * Adjust the disk quota also. This was reserved
2688 XFS_TRANS_MOD_DQUOT_BYINO(mp, ap->tp, ap->ip,
2689 ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2690 XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2699 xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2701 xfs_mount_t *mp; /* mount point structure */
2702 xfs_alloctype_t atype = 0; /* type for allocation routines */
2703 xfs_extlen_t align; /* minimum allocation alignment */
2705 xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
2706 xfs_agnumber_t startag;
2707 xfs_alloc_arg_t args;
2710 xfs_extlen_t longest;
2712 xfs_extlen_t nextminlen = 0;
2714 int nullfb; /* true if ap->firstblock isn't set */
2720 mp = ap->ip->i_mount;
2721 align = (ap->userdata && ap->ip->i_d.di_extsize &&
2722 (ap->ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE)) ?
2723 ap->ip->i_d.di_extsize : 0;
2724 if (unlikely(align)) {
2725 error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2726 align, 0, ap->eof, 0, ap->conv,
2727 &ap->off, &ap->alen);
2731 nullfb = ap->firstblock == NULLFSBLOCK;
2732 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2734 ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2736 ap->rval = ap->firstblock;
2738 xfs_bmap_adjacent(ap);
2741 * If allowed, use ap->rval; otherwise must use firstblock since
2742 * it's in the right allocation group.
2744 if (nullfb || XFS_FSB_TO_AGNO(mp, ap->rval) == fb_agno)
2747 ap->rval = ap->firstblock;
2749 * Normal allocation, done through xfs_alloc_vextent.
2751 tryagain = isaligned = 0;
2754 args.fsbno = ap->rval;
2755 args.maxlen = MIN(ap->alen, mp->m_sb.sb_agblocks);
2756 args.firstblock = ap->firstblock;
2759 args.type = XFS_ALLOCTYPE_START_BNO;
2760 args.total = ap->total;
2762 * Find the longest available space.
2763 * We're going to try for the whole allocation at once.
2765 startag = ag = XFS_FSB_TO_AGNO(mp, args.fsbno);
2767 down_read(&mp->m_peraglock);
2768 while (blen < ap->alen) {
2769 pag = &mp->m_perag[ag];
2770 if (!pag->pagf_init &&
2771 (error = xfs_alloc_pagf_init(mp, args.tp,
2772 ag, XFS_ALLOC_FLAG_TRYLOCK))) {
2773 up_read(&mp->m_peraglock);
2777 * See xfs_alloc_fix_freelist...
2779 if (pag->pagf_init) {
2780 need = XFS_MIN_FREELIST_PAG(pag, mp);
2781 delta = need > pag->pagf_flcount ?
2782 need - pag->pagf_flcount : 0;
2783 longest = (pag->pagf_longest > delta) ?
2784 (pag->pagf_longest - delta) :
2785 (pag->pagf_flcount > 0 ||
2786 pag->pagf_longest > 0);
2791 if (++ag == mp->m_sb.sb_agcount)
2796 up_read(&mp->m_peraglock);
2798 * Since the above loop did a BUF_TRYLOCK, it is
2799 * possible that there is space for this request.
2801 if (notinit || blen < ap->minlen)
2802 args.minlen = ap->minlen;
2804 * If the best seen length is less than the request
2805 * length, use the best as the minimum.
2807 else if (blen < ap->alen)
2810 * Otherwise we've seen an extent as big as alen,
2811 * use that as the minimum.
2814 args.minlen = ap->alen;
2815 } else if (ap->low) {
2816 args.type = XFS_ALLOCTYPE_START_BNO;
2817 args.total = args.minlen = ap->minlen;
2819 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2820 args.total = ap->total;
2821 args.minlen = ap->minlen;
2823 if (unlikely(ap->userdata && ap->ip->i_d.di_extsize &&
2824 (ap->ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE))) {
2825 args.prod = ap->ip->i_d.di_extsize;
2826 if ((args.mod = (xfs_extlen_t)do_mod(ap->off, args.prod)))
2827 args.mod = (xfs_extlen_t)(args.prod - args.mod);
2828 } else if (mp->m_sb.sb_blocksize >= NBPP) {
2832 args.prod = NBPP >> mp->m_sb.sb_blocklog;
2833 if ((args.mod = (xfs_extlen_t)(do_mod(ap->off, args.prod))))
2834 args.mod = (xfs_extlen_t)(args.prod - args.mod);
2837 * If we are not low on available data blocks, and the
2838 * underlying logical volume manager is a stripe, and
2839 * the file offset is zero then try to allocate data
2840 * blocks on stripe unit boundary.
2841 * NOTE: ap->aeof is only set if the allocation length
2842 * is >= the stripe unit and the allocation offset is
2843 * at the end of file.
2845 if (!ap->low && ap->aeof) {
2847 args.alignment = mp->m_dalign;
2851 * Adjust for alignment
2853 if (blen > args.alignment && blen <= ap->alen)
2854 args.minlen = blen - args.alignment;
2855 args.minalignslop = 0;
2858 * First try an exact bno allocation.
2859 * If it fails then do a near or start bno
2860 * allocation with alignment turned on.
2864 args.type = XFS_ALLOCTYPE_THIS_BNO;
2867 * Compute the minlen+alignment for the
2868 * next case. Set slop so that the value
2869 * of minlen+alignment+slop doesn't go up
2870 * between the calls.
2872 if (blen > mp->m_dalign && blen <= ap->alen)
2873 nextminlen = blen - mp->m_dalign;
2875 nextminlen = args.minlen;
2876 if (nextminlen + mp->m_dalign > args.minlen + 1)
2878 nextminlen + mp->m_dalign -
2881 args.minalignslop = 0;
2885 args.minalignslop = 0;
2887 args.minleft = ap->minleft;
2888 args.wasdel = ap->wasdel;
2890 args.userdata = ap->userdata;
2891 if ((error = xfs_alloc_vextent(&args)))
2893 if (tryagain && args.fsbno == NULLFSBLOCK) {
2895 * Exact allocation failed. Now try with alignment
2899 args.fsbno = ap->rval;
2900 args.alignment = mp->m_dalign;
2901 args.minlen = nextminlen;
2902 args.minalignslop = 0;
2904 if ((error = xfs_alloc_vextent(&args)))
2907 if (isaligned && args.fsbno == NULLFSBLOCK) {
2909 * allocation failed, so turn off alignment and
2913 args.fsbno = ap->rval;
2915 if ((error = xfs_alloc_vextent(&args)))
2918 if (args.fsbno == NULLFSBLOCK && nullfb &&
2919 args.minlen > ap->minlen) {
2920 args.minlen = ap->minlen;
2921 args.type = XFS_ALLOCTYPE_START_BNO;
2922 args.fsbno = ap->rval;
2923 if ((error = xfs_alloc_vextent(&args)))
2926 if (args.fsbno == NULLFSBLOCK && nullfb) {
2928 args.type = XFS_ALLOCTYPE_FIRST_AG;
2929 args.total = ap->minlen;
2931 if ((error = xfs_alloc_vextent(&args)))
2935 if (args.fsbno != NULLFSBLOCK) {
2936 ap->firstblock = ap->rval = args.fsbno;
2937 ASSERT(nullfb || fb_agno == args.agno ||
2938 (ap->low && fb_agno < args.agno));
2939 ap->alen = args.len;
2940 ap->ip->i_d.di_nblocks += args.len;
2941 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2943 ap->ip->i_delayed_blks -= args.len;
2945 * Adjust the disk quota also. This was reserved
2948 XFS_TRANS_MOD_DQUOT_BYINO(mp, ap->tp, ap->ip,
2949 ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2950 XFS_TRANS_DQ_BCOUNT,
2953 ap->rval = NULLFSBLOCK;
2960 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2961 * It figures out where to ask the underlying allocator to put the new extent.
2965 xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2967 if ((ap->ip->i_d.di_flags & XFS_DIFLAG_REALTIME) && ap->userdata)
2968 return xfs_bmap_rtalloc(ap);
2969 return xfs_bmap_btalloc(ap);
2973 * Transform a btree format file with only one leaf node, where the
2974 * extents list will fit in the inode, into an extents format file.
2975 * Since the file extents are already in-core, all we have to do is
2976 * give up the space for the btree root and pitch the leaf block.
2978 STATIC int /* error */
2979 xfs_bmap_btree_to_extents(
2980 xfs_trans_t *tp, /* transaction pointer */
2981 xfs_inode_t *ip, /* incore inode pointer */
2982 xfs_btree_cur_t *cur, /* btree cursor */
2983 int *logflagsp, /* inode logging flags */
2984 int whichfork) /* data or attr fork */
2987 xfs_bmbt_block_t *cblock;/* child btree block */
2988 xfs_fsblock_t cbno; /* child block number */
2989 xfs_buf_t *cbp; /* child block's buffer */
2990 int error; /* error return value */
2991 xfs_ifork_t *ifp; /* inode fork data */
2992 xfs_mount_t *mp; /* mount point structure */
2993 __be64 *pp; /* ptr to block address */
2994 xfs_bmbt_block_t *rblock;/* root btree block */
2996 ifp = XFS_IFORK_PTR(ip, whichfork);
2997 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2998 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
2999 rblock = ifp->if_broot;
3000 ASSERT(be16_to_cpu(rblock->bb_level) == 1);
3001 ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
3002 ASSERT(XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes) == 1);
3004 pp = XFS_BMAP_BROOT_PTR_ADDR(rblock, 1, ifp->if_broot_bytes);
3005 cbno = be64_to_cpu(*pp);
3008 if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
3011 if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
3012 XFS_BMAP_BTREE_REF)))
3014 cblock = XFS_BUF_TO_BMBT_BLOCK(cbp);
3015 if ((error = xfs_btree_check_lblock(cur, cblock, 0, cbp)))
3017 xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
3018 ip->i_d.di_nblocks--;
3019 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
3020 xfs_trans_binval(tp, cbp);
3021 if (cur->bc_bufs[0] == cbp)
3022 cur->bc_bufs[0] = NULL;
3023 xfs_iroot_realloc(ip, -1, whichfork);
3024 ASSERT(ifp->if_broot == NULL);
3025 ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
3026 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3027 *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
3032 * Called by xfs_bmapi to update file extent records and the btree
3033 * after removing space (or undoing a delayed allocation).
3035 STATIC int /* error */
3036 xfs_bmap_del_extent(
3037 xfs_inode_t *ip, /* incore inode pointer */
3038 xfs_trans_t *tp, /* current transaction pointer */
3039 xfs_extnum_t idx, /* extent number to update/delete */
3040 xfs_bmap_free_t *flist, /* list of extents to be freed */
3041 xfs_btree_cur_t *cur, /* if null, not a btree */
3042 xfs_bmbt_irec_t *del, /* data to remove from extents */
3043 int *logflagsp, /* inode logging flags */
3044 xfs_extdelta_t *delta, /* Change made to incore extents */
3045 int whichfork, /* data or attr fork */
3046 int rsvd) /* OK to allocate reserved blocks */
3048 xfs_filblks_t da_new; /* new delay-alloc indirect blocks */
3049 xfs_filblks_t da_old; /* old delay-alloc indirect blocks */
3050 xfs_fsblock_t del_endblock=0; /* first block past del */
3051 xfs_fileoff_t del_endoff; /* first offset past del */
3052 int delay; /* current block is delayed allocated */
3053 int do_fx; /* free extent at end of routine */
3054 xfs_bmbt_rec_t *ep; /* current extent entry pointer */
3055 int error; /* error return value */
3056 int flags; /* inode logging flags */
3057 #ifdef XFS_BMAP_TRACE
3058 static char fname[] = "xfs_bmap_del_extent";
3060 xfs_bmbt_irec_t got; /* current extent entry */
3061 xfs_fileoff_t got_endoff; /* first offset past got */
3062 int i; /* temp state */
3063 xfs_ifork_t *ifp; /* inode fork pointer */
3064 xfs_mount_t *mp; /* mount structure */
3065 xfs_filblks_t nblks; /* quota/sb block count */
3066 xfs_bmbt_irec_t new; /* new record to be inserted */
3068 uint qfield; /* quota field to update */
3069 xfs_filblks_t temp; /* for indirect length calculations */
3070 xfs_filblks_t temp2; /* for indirect length calculations */
3072 XFS_STATS_INC(xs_del_exlist);
3074 ifp = XFS_IFORK_PTR(ip, whichfork);
3075 ASSERT((idx >= 0) && (idx < ifp->if_bytes /
3076 (uint)sizeof(xfs_bmbt_rec_t)));
3077 ASSERT(del->br_blockcount > 0);
3078 ep = xfs_iext_get_ext(ifp, idx);
3079 xfs_bmbt_get_all(ep, &got);
3080 ASSERT(got.br_startoff <= del->br_startoff);
3081 del_endoff = del->br_startoff + del->br_blockcount;
3082 got_endoff = got.br_startoff + got.br_blockcount;
3083 ASSERT(got_endoff >= del_endoff);
3084 delay = ISNULLSTARTBLOCK(got.br_startblock);
3085 ASSERT(ISNULLSTARTBLOCK(del->br_startblock) == delay);
3090 * If deleting a real allocation, must free up the disk space.
3093 flags = XFS_ILOG_CORE;
3095 * Realtime allocation. Free it and record di_nblocks update.
3097 if (whichfork == XFS_DATA_FORK &&
3098 (ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
3102 ASSERT(do_mod(del->br_blockcount,
3103 mp->m_sb.sb_rextsize) == 0);
3104 ASSERT(do_mod(del->br_startblock,
3105 mp->m_sb.sb_rextsize) == 0);
3106 bno = del->br_startblock;
3107 len = del->br_blockcount;
3108 do_div(bno, mp->m_sb.sb_rextsize);
3109 do_div(len, mp->m_sb.sb_rextsize);
3110 if ((error = xfs_rtfree_extent(ip->i_transp, bno,
3111 (xfs_extlen_t)len)))
3114 nblks = len * mp->m_sb.sb_rextsize;
3115 qfield = XFS_TRANS_DQ_RTBCOUNT;
3118 * Ordinary allocation.
3122 nblks = del->br_blockcount;
3123 qfield = XFS_TRANS_DQ_BCOUNT;
3126 * Set up del_endblock and cur for later.
3128 del_endblock = del->br_startblock + del->br_blockcount;
3130 if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
3131 got.br_startblock, got.br_blockcount,
3136 da_old = da_new = 0;
3138 da_old = STARTBLOCKVAL(got.br_startblock);
3144 * Set flag value to use in switch statement.
3145 * Left-contig is 2, right-contig is 1.
3147 switch (((got.br_startoff == del->br_startoff) << 1) |
3148 (got_endoff == del_endoff)) {
3151 * Matches the whole extent. Delete the entry.
3153 xfs_bmap_trace_delete(fname, "3", ip, idx, 1, whichfork);
3154 xfs_iext_remove(ifp, idx, 1);
3155 ifp->if_lastex = idx;
3158 XFS_IFORK_NEXT_SET(ip, whichfork,
3159 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
3160 flags |= XFS_ILOG_CORE;
3162 flags |= XFS_ILOG_FEXT(whichfork);
3165 if ((error = xfs_bmbt_delete(cur, &i)))
3172 * Deleting the first part of the extent.
3174 xfs_bmap_trace_pre_update(fname, "2", ip, idx, whichfork);
3175 xfs_bmbt_set_startoff(ep, del_endoff);
3176 temp = got.br_blockcount - del->br_blockcount;
3177 xfs_bmbt_set_blockcount(ep, temp);
3178 ifp->if_lastex = idx;
3180 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
3182 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3183 xfs_bmap_trace_post_update(fname, "2", ip, idx,
3188 xfs_bmbt_set_startblock(ep, del_endblock);
3189 xfs_bmap_trace_post_update(fname, "2", ip, idx, whichfork);
3191 flags |= XFS_ILOG_FEXT(whichfork);
3194 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
3195 got.br_blockcount - del->br_blockcount,
3202 * Deleting the last part of the extent.
3204 temp = got.br_blockcount - del->br_blockcount;
3205 xfs_bmap_trace_pre_update(fname, "1", ip, idx, whichfork);
3206 xfs_bmbt_set_blockcount(ep, temp);
3207 ifp->if_lastex = idx;
3209 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
3211 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3212 xfs_bmap_trace_post_update(fname, "1", ip, idx,
3217 xfs_bmap_trace_post_update(fname, "1", ip, idx, whichfork);
3219 flags |= XFS_ILOG_FEXT(whichfork);
3222 if ((error = xfs_bmbt_update(cur, got.br_startoff,
3224 got.br_blockcount - del->br_blockcount,
3231 * Deleting the middle of the extent.
3233 temp = del->br_startoff - got.br_startoff;
3234 xfs_bmap_trace_pre_update(fname, "0", ip, idx, whichfork);
3235 xfs_bmbt_set_blockcount(ep, temp);
3236 new.br_startoff = del_endoff;
3237 temp2 = got_endoff - del_endoff;
3238 new.br_blockcount = temp2;
3239 new.br_state = got.br_state;
3241 new.br_startblock = del_endblock;
3242 flags |= XFS_ILOG_CORE;
3244 if ((error = xfs_bmbt_update(cur,
3246 got.br_startblock, temp,
3249 if ((error = xfs_bmbt_increment(cur, 0, &i)))
3251 cur->bc_rec.b = new;
3252 error = xfs_bmbt_insert(cur, &i);
3253 if (error && error != ENOSPC)
3256 * If get no-space back from btree insert,
3257 * it tried a split, and we have a zero
3258 * block reservation.
3259 * Fix up our state and return the error.
3261 if (error == ENOSPC) {
3263 * Reset the cursor, don't trust
3264 * it after any insert operation.
3266 if ((error = xfs_bmbt_lookup_eq(cur,
3273 * Update the btree record back
3274 * to the original value.
3276 if ((error = xfs_bmbt_update(cur,
3283 * Reset the extent record back
3284 * to the original value.
3286 xfs_bmbt_set_blockcount(ep,
3289 error = XFS_ERROR(ENOSPC);
3294 flags |= XFS_ILOG_FEXT(whichfork);
3295 XFS_IFORK_NEXT_SET(ip, whichfork,
3296 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
3298 ASSERT(whichfork == XFS_DATA_FORK);
3299 temp = xfs_bmap_worst_indlen(ip, temp);
3300 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3301 temp2 = xfs_bmap_worst_indlen(ip, temp2);
3302 new.br_startblock = NULLSTARTBLOCK((int)temp2);
3303 da_new = temp + temp2;
3304 while (da_new > da_old) {
3308 xfs_bmbt_set_startblock(ep,
3309 NULLSTARTBLOCK((int)temp));
3311 if (da_new == da_old)
3317 NULLSTARTBLOCK((int)temp2);
3321 xfs_bmap_trace_post_update(fname, "0", ip, idx, whichfork);
3322 xfs_bmap_trace_insert(fname, "0", ip, idx + 1, 1, &new, NULL,
3324 xfs_iext_insert(ifp, idx + 1, 1, &new);
3325 ifp->if_lastex = idx + 1;
3329 * If we need to, add to list of extents to delete.
3332 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
3335 * Adjust inode # blocks in the file.
3338 ip->i_d.di_nblocks -= nblks;
3340 * Adjust quota data.
3343 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, qfield, (long)-nblks);
3346 * Account for change in delayed indirect blocks.
3347 * Nothing to do for disk quota accounting here.
3349 ASSERT(da_old >= da_new);
3350 if (da_old > da_new)
3351 xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS, (int64_t)(da_old - da_new),
3354 /* DELTA: report the original extent. */
3355 if (delta->xed_startoff > got.br_startoff)
3356 delta->xed_startoff = got.br_startoff;
3357 if (delta->xed_blockcount < got.br_startoff+got.br_blockcount)
3358 delta->xed_blockcount = got.br_startoff +
3367 * Remove the entry "free" from the free item list. Prev points to the
3368 * previous entry, unless "free" is the head of the list.
3372 xfs_bmap_free_t *flist, /* free item list header */
3373 xfs_bmap_free_item_t *prev, /* previous item on list, if any */
3374 xfs_bmap_free_item_t *free) /* list item to be freed */
3377 prev->xbfi_next = free->xbfi_next;
3379 flist->xbf_first = free->xbfi_next;
3381 kmem_zone_free(xfs_bmap_free_item_zone, free);
3385 * Convert an extents-format file into a btree-format file.
3386 * The new file will have a root block (in the inode) and a single child block.
3388 STATIC int /* error */
3389 xfs_bmap_extents_to_btree(
3390 xfs_trans_t *tp, /* transaction pointer */
3391 xfs_inode_t *ip, /* incore inode pointer */
3392 xfs_fsblock_t *firstblock, /* first-block-allocated */
3393 xfs_bmap_free_t *flist, /* blocks freed in xaction */
3394 xfs_btree_cur_t **curp, /* cursor returned to caller */
3395 int wasdel, /* converting a delayed alloc */
3396 int *logflagsp, /* inode logging flags */
3397 int whichfork) /* data or attr fork */
3399 xfs_bmbt_block_t *ablock; /* allocated (child) bt block */
3400 xfs_buf_t *abp; /* buffer for ablock */
3401 xfs_alloc_arg_t args; /* allocation arguments */
3402 xfs_bmbt_rec_t *arp; /* child record pointer */
3403 xfs_bmbt_block_t *block; /* btree root block */
3404 xfs_btree_cur_t *cur; /* bmap btree cursor */
3405 xfs_bmbt_rec_t *ep; /* extent record pointer */
3406 int error; /* error return value */
3407 xfs_extnum_t i, cnt; /* extent record index */
3408 xfs_ifork_t *ifp; /* inode fork pointer */
3409 xfs_bmbt_key_t *kp; /* root block key pointer */
3410 xfs_mount_t *mp; /* mount structure */
3411 xfs_extnum_t nextents; /* number of file extents */
3412 xfs_bmbt_ptr_t *pp; /* root block address pointer */
3414 ifp = XFS_IFORK_PTR(ip, whichfork);
3415 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3416 ASSERT(ifp->if_ext_max ==
3417 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
3419 * Make space in the inode incore.
3421 xfs_iroot_realloc(ip, 1, whichfork);
3422 ifp->if_flags |= XFS_IFBROOT;
3426 block = ifp->if_broot;
3427 block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3428 block->bb_level = cpu_to_be16(1);
3429 block->bb_numrecs = cpu_to_be16(1);
3430 block->bb_leftsib = cpu_to_be64(NULLDFSBNO);
3431 block->bb_rightsib = cpu_to_be64(NULLDFSBNO);
3433 * Need a cursor. Can't allocate until bb_level is filled in.
3436 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
3438 cur->bc_private.b.firstblock = *firstblock;
3439 cur->bc_private.b.flist = flist;
3440 cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3442 * Convert to a btree with two levels, one record in root.
3444 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3447 args.firstblock = *firstblock;
3448 if (*firstblock == NULLFSBLOCK) {
3449 args.type = XFS_ALLOCTYPE_START_BNO;
3450 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3451 } else if (flist->xbf_low) {
3452 args.type = XFS_ALLOCTYPE_START_BNO;
3453 args.fsbno = *firstblock;
3455 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3456 args.fsbno = *firstblock;
3458 args.minlen = args.maxlen = args.prod = 1;
3459 args.total = args.minleft = args.alignment = args.mod = args.isfl =
3460 args.minalignslop = 0;
3461 args.wasdel = wasdel;
3463 if ((error = xfs_alloc_vextent(&args))) {
3464 xfs_iroot_realloc(ip, -1, whichfork);
3465 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3469 * Allocation can't fail, the space was reserved.
3471 ASSERT(args.fsbno != NULLFSBLOCK);
3472 ASSERT(*firstblock == NULLFSBLOCK ||
3473 args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3475 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3476 *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3477 cur->bc_private.b.allocated++;
3478 ip->i_d.di_nblocks++;
3479 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3480 abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3482 * Fill in the child block.
3484 ablock = XFS_BUF_TO_BMBT_BLOCK(abp);
3485 ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3486 ablock->bb_level = 0;
3487 ablock->bb_leftsib = cpu_to_be64(NULLDFSBNO);
3488 ablock->bb_rightsib = cpu_to_be64(NULLDFSBNO);
3489 arp = XFS_BMAP_REC_IADDR(ablock, 1, cur);
3490 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3491 for (cnt = i = 0; i < nextents; i++) {
3492 ep = xfs_iext_get_ext(ifp, i);
3493 if (!ISNULLSTARTBLOCK(xfs_bmbt_get_startblock(ep))) {
3494 arp->l0 = INT_GET(ep->l0, ARCH_CONVERT);
3495 arp->l1 = INT_GET(ep->l1, ARCH_CONVERT);
3499 ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3500 ablock->bb_numrecs = cpu_to_be16(cnt);
3502 * Fill in the root key and pointer.
3504 kp = XFS_BMAP_KEY_IADDR(block, 1, cur);
3505 arp = XFS_BMAP_REC_IADDR(ablock, 1, cur);
3506 kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
3507 pp = XFS_BMAP_PTR_IADDR(block, 1, cur);
3508 *pp = cpu_to_be64(args.fsbno);
3510 * Do all this logging at the end so that
3511 * the root is at the right level.
3513 xfs_bmbt_log_block(cur, abp, XFS_BB_ALL_BITS);
3514 xfs_bmbt_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3515 ASSERT(*curp == NULL);
3517 *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FBROOT(whichfork);
3522 * Helper routine to reset inode di_forkoff field when switching
3523 * attribute fork from local to extent format - we reset it where
3524 * possible to make space available for inline data fork extents.
3527 xfs_bmap_forkoff_reset(
3532 if (whichfork == XFS_ATTR_FORK &&
3533 (ip->i_d.di_format != XFS_DINODE_FMT_DEV) &&
3534 (ip->i_d.di_format != XFS_DINODE_FMT_UUID) &&
3535 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3536 ((mp->m_attroffset >> 3) > ip->i_d.di_forkoff)) {
3537 ip->i_d.di_forkoff = mp->m_attroffset >> 3;
3538 ip->i_df.if_ext_max = XFS_IFORK_DSIZE(ip) /
3539 (uint)sizeof(xfs_bmbt_rec_t);
3540 ip->i_afp->if_ext_max = XFS_IFORK_ASIZE(ip) /
3541 (uint)sizeof(xfs_bmbt_rec_t);
3546 * Convert a local file to an extents file.
3547 * This code is out of bounds for data forks of regular files,
3548 * since the file data needs to get logged so things will stay consistent.
3549 * (The bmap-level manipulations are ok, though).
3551 STATIC int /* error */
3552 xfs_bmap_local_to_extents(
3553 xfs_trans_t *tp, /* transaction pointer */
3554 xfs_inode_t *ip, /* incore inode pointer */
3555 xfs_fsblock_t *firstblock, /* first block allocated in xaction */
3556 xfs_extlen_t total, /* total blocks needed by transaction */
3557 int *logflagsp, /* inode logging flags */
3558 int whichfork) /* data or attr fork */
3560 int error; /* error return value */
3561 int flags; /* logging flags returned */
3562 #ifdef XFS_BMAP_TRACE
3563 static char fname[] = "xfs_bmap_local_to_extents";
3565 xfs_ifork_t *ifp; /* inode fork pointer */
3568 * We don't want to deal with the case of keeping inode data inline yet.
3569 * So sending the data fork of a regular inode is invalid.
3571 ASSERT(!((ip->i_d.di_mode & S_IFMT) == S_IFREG &&
3572 whichfork == XFS_DATA_FORK));
3573 ifp = XFS_IFORK_PTR(ip, whichfork);
3574 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3577 if (ifp->if_bytes) {
3578 xfs_alloc_arg_t args; /* allocation arguments */
3579 xfs_buf_t *bp; /* buffer for extent block */
3580 xfs_bmbt_rec_t *ep; /* extent record pointer */
3583 args.mp = ip->i_mount;
3584 args.firstblock = *firstblock;
3585 ASSERT((ifp->if_flags &
3586 (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == XFS_IFINLINE);
3588 * Allocate a block. We know we need only one, since the
3589 * file currently fits in an inode.
3591 if (*firstblock == NULLFSBLOCK) {
3592 args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3593 args.type = XFS_ALLOCTYPE_START_BNO;
3595 args.fsbno = *firstblock;
3596 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3599 args.mod = args.minleft = args.alignment = args.wasdel =
3600 args.isfl = args.minalignslop = 0;
3601 args.minlen = args.maxlen = args.prod = 1;
3602 if ((error = xfs_alloc_vextent(&args)))
3605 * Can't fail, the space was reserved.
3607 ASSERT(args.fsbno != NULLFSBLOCK);
3608 ASSERT(args.len == 1);
3609 *firstblock = args.fsbno;
3610 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3611 memcpy((char *)XFS_BUF_PTR(bp), ifp->if_u1.if_data,
3613 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3614 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3615 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3616 xfs_iext_add(ifp, 0, 1);
3617 ep = xfs_iext_get_ext(ifp, 0);
3618 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3619 xfs_bmap_trace_post_update(fname, "new", ip, 0, whichfork);
3620 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3621 ip->i_d.di_nblocks = 1;
3622 XFS_TRANS_MOD_DQUOT_BYINO(args.mp, tp, ip,
3623 XFS_TRANS_DQ_BCOUNT, 1L);
3624 flags |= XFS_ILOG_FEXT(whichfork);
3626 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3627 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3629 ifp->if_flags &= ~XFS_IFINLINE;
3630 ifp->if_flags |= XFS_IFEXTENTS;
3631 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3632 flags |= XFS_ILOG_CORE;
3639 * Search the extent records for the entry containing block bno.
3640 * If bno lies in a hole, point to the next entry. If bno lies
3641 * past eof, *eofp will be set, and *prevp will contain the last
3642 * entry (null if none). Else, *lastxp will be set to the index
3643 * of the found entry; *gotp will contain the entry.
3645 xfs_bmbt_rec_t * /* pointer to found extent entry */
3646 xfs_bmap_search_multi_extents(
3647 xfs_ifork_t *ifp, /* inode fork pointer */
3648 xfs_fileoff_t bno, /* block number searched for */
3649 int *eofp, /* out: end of file found */
3650 xfs_extnum_t *lastxp, /* out: last extent index */
3651 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
3652 xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
3654 xfs_bmbt_rec_t *ep; /* extent record pointer */
3655 xfs_extnum_t lastx; /* last extent index */
3658 * Initialize the extent entry structure to catch access to
3659 * uninitialized br_startblock field.
3661 gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
3662 gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3663 gotp->br_state = XFS_EXT_INVALID;
3665 gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
3667 gotp->br_startblock = 0xffffa5a5;
3669 prevp->br_startoff = NULLFILEOFF;
3671 ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
3673 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
3675 if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
3676 xfs_bmbt_get_all(ep, gotp);
3690 * Search the extents list for the inode, for the extent containing bno.
3691 * If bno lies in a hole, point to the next entry. If bno lies past eof,
3692 * *eofp will be set, and *prevp will contain the last entry (null if none).
3693 * Else, *lastxp will be set to the index of the found
3694 * entry; *gotp will contain the entry.
3696 STATIC xfs_bmbt_rec_t * /* pointer to found extent entry */
3697 xfs_bmap_search_extents(
3698 xfs_inode_t *ip, /* incore inode pointer */
3699 xfs_fileoff_t bno, /* block number searched for */
3700 int fork, /* data or attr fork */
3701 int *eofp, /* out: end of file found */
3702 xfs_extnum_t *lastxp, /* out: last extent index */
3703 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
3704 xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
3706 xfs_ifork_t *ifp; /* inode fork pointer */
3707 xfs_bmbt_rec_t *ep; /* extent record pointer */
3709 XFS_STATS_INC(xs_look_exlist);
3710 ifp = XFS_IFORK_PTR(ip, fork);
3712 ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
3714 if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
3715 !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
3716 xfs_cmn_err(XFS_PTAG_FSBLOCK_ZERO, CE_ALERT, ip->i_mount,
3717 "Access to block zero in inode %llu "
3718 "start_block: %llx start_off: %llx "
3719 "blkcnt: %llx extent-state: %x lastx: %x\n",
3720 (unsigned long long)ip->i_ino,
3721 (unsigned long long)gotp->br_startblock,
3722 (unsigned long long)gotp->br_startoff,
3723 (unsigned long long)gotp->br_blockcount,
3724 gotp->br_state, *lastxp);
3725 *lastxp = NULLEXTNUM;
3733 #ifdef XFS_BMAP_TRACE
3734 ktrace_t *xfs_bmap_trace_buf;
3737 * Add a bmap trace buffer entry. Base routine for the others.
3740 xfs_bmap_trace_addentry(
3741 int opcode, /* operation */
3742 char *fname, /* function name */
3743 char *desc, /* operation description */
3744 xfs_inode_t *ip, /* incore inode pointer */
3745 xfs_extnum_t idx, /* index of entry(ies) */
3746 xfs_extnum_t cnt, /* count of entries, 1 or 2 */
3747 xfs_bmbt_rec_t *r1, /* first record */
3748 xfs_bmbt_rec_t *r2, /* second record or null */
3749 int whichfork) /* data or attr fork */
3753 ASSERT(cnt == 1 || cnt == 2);
3758 memset(&tr2, 0, sizeof(tr2));
3761 ktrace_enter(xfs_bmap_trace_buf,
3762 (void *)(__psint_t)(opcode | (whichfork << 16)),
3763 (void *)fname, (void *)desc, (void *)ip,
3764 (void *)(__psint_t)idx,
3765 (void *)(__psint_t)cnt,
3766 (void *)(__psunsigned_t)(ip->i_ino >> 32),
3767 (void *)(__psunsigned_t)(unsigned)ip->i_ino,
3768 (void *)(__psunsigned_t)(r1->l0 >> 32),
3769 (void *)(__psunsigned_t)(unsigned)(r1->l0),
3770 (void *)(__psunsigned_t)(r1->l1 >> 32),
3771 (void *)(__psunsigned_t)(unsigned)(r1->l1),
3772 (void *)(__psunsigned_t)(r2->l0 >> 32),
3773 (void *)(__psunsigned_t)(unsigned)(r2->l0),
3774 (void *)(__psunsigned_t)(r2->l1 >> 32),
3775 (void *)(__psunsigned_t)(unsigned)(r2->l1)
3777 ASSERT(ip->i_xtrace);
3778 ktrace_enter(ip->i_xtrace,
3779 (void *)(__psint_t)(opcode | (whichfork << 16)),
3780 (void *)fname, (void *)desc, (void *)ip,
3781 (void *)(__psint_t)idx,
3782 (void *)(__psint_t)cnt,
3783 (void *)(__psunsigned_t)(ip->i_ino >> 32),
3784 (void *)(__psunsigned_t)(unsigned)ip->i_ino,
3785 (void *)(__psunsigned_t)(r1->l0 >> 32),
3786 (void *)(__psunsigned_t)(unsigned)(r1->l0),
3787 (void *)(__psunsigned_t)(r1->l1 >> 32),
3788 (void *)(__psunsigned_t)(unsigned)(r1->l1),
3789 (void *)(__psunsigned_t)(r2->l0 >> 32),
3790 (void *)(__psunsigned_t)(unsigned)(r2->l0),
3791 (void *)(__psunsigned_t)(r2->l1 >> 32),
3792 (void *)(__psunsigned_t)(unsigned)(r2->l1)
3797 * Add bmap trace entry prior to a call to xfs_iext_remove.
3800 xfs_bmap_trace_delete(
3801 char *fname, /* function name */
3802 char *desc, /* operation description */
3803 xfs_inode_t *ip, /* incore inode pointer */
3804 xfs_extnum_t idx, /* index of entry(entries) deleted */
3805 xfs_extnum_t cnt, /* count of entries deleted, 1 or 2 */
3806 int whichfork) /* data or attr fork */
3808 xfs_ifork_t *ifp; /* inode fork pointer */
3810 ifp = XFS_IFORK_PTR(ip, whichfork);
3811 xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_DELETE, fname, desc, ip, idx,
3812 cnt, xfs_iext_get_ext(ifp, idx),
3813 cnt == 2 ? xfs_iext_get_ext(ifp, idx + 1) : NULL,
3818 * Add bmap trace entry prior to a call to xfs_iext_insert, or
3819 * reading in the extents list from the disk (in the btree).
3822 xfs_bmap_trace_insert(
3823 char *fname, /* function name */
3824 char *desc, /* operation description */
3825 xfs_inode_t *ip, /* incore inode pointer */
3826 xfs_extnum_t idx, /* index of entry(entries) inserted */
3827 xfs_extnum_t cnt, /* count of entries inserted, 1 or 2 */
3828 xfs_bmbt_irec_t *r1, /* inserted record 1 */
3829 xfs_bmbt_irec_t *r2, /* inserted record 2 or null */
3830 int whichfork) /* data or attr fork */
3832 xfs_bmbt_rec_t tr1; /* compressed record 1 */
3833 xfs_bmbt_rec_t tr2; /* compressed record 2 if needed */
3835 xfs_bmbt_set_all(&tr1, r1);
3838 xfs_bmbt_set_all(&tr2, r2);
3843 xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_INSERT, fname, desc, ip, idx,
3844 cnt, &tr1, cnt == 2 ? &tr2 : NULL, whichfork);
3848 * Add bmap trace entry after updating an extent record in place.
3851 xfs_bmap_trace_post_update(
3852 char *fname, /* function name */
3853 char *desc, /* operation description */
3854 xfs_inode_t *ip, /* incore inode pointer */
3855 xfs_extnum_t idx, /* index of entry updated */
3856 int whichfork) /* data or attr fork */
3858 xfs_ifork_t *ifp; /* inode fork pointer */
3860 ifp = XFS_IFORK_PTR(ip, whichfork);
3861 xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_POST_UP, fname, desc, ip, idx,
3862 1, xfs_iext_get_ext(ifp, idx), NULL, whichfork);
3866 * Add bmap trace entry prior to updating an extent record in place.
3869 xfs_bmap_trace_pre_update(
3870 char *fname, /* function name */
3871 char *desc, /* operation description */
3872 xfs_inode_t *ip, /* incore inode pointer */
3873 xfs_extnum_t idx, /* index of entry to be updated */
3874 int whichfork) /* data or attr fork */
3876 xfs_ifork_t *ifp; /* inode fork pointer */
3878 ifp = XFS_IFORK_PTR(ip, whichfork);
3879 xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_PRE_UP, fname, desc, ip, idx, 1,
3880 xfs_iext_get_ext(ifp, idx), NULL, whichfork);
3882 #endif /* XFS_BMAP_TRACE */
3885 * Compute the worst-case number of indirect blocks that will be used
3886 * for ip's delayed extent of length "len".
3888 STATIC xfs_filblks_t
3889 xfs_bmap_worst_indlen(
3890 xfs_inode_t *ip, /* incore inode pointer */
3891 xfs_filblks_t len) /* delayed extent length */
3893 int level; /* btree level number */
3894 int maxrecs; /* maximum record count at this level */
3895 xfs_mount_t *mp; /* mount structure */
3896 xfs_filblks_t rval; /* return value */
3899 maxrecs = mp->m_bmap_dmxr[0];
3900 for (level = 0, rval = 0;
3901 level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3904 do_div(len, maxrecs);
3907 return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3910 maxrecs = mp->m_bmap_dmxr[1];
3915 #if defined(XFS_RW_TRACE)
3924 if (ip->i_rwtrace == NULL)
3926 ktrace_enter(ip->i_rwtrace,
3927 (void *)(__psint_t)XFS_BUNMAP,
3929 (void *)(__psint_t)((ip->i_d.di_size >> 32) & 0xffffffff),
3930 (void *)(__psint_t)(ip->i_d.di_size & 0xffffffff),
3931 (void *)(__psint_t)(((xfs_dfiloff_t)bno >> 32) & 0xffffffff),
3932 (void *)(__psint_t)((xfs_dfiloff_t)bno & 0xffffffff),
3933 (void *)(__psint_t)len,
3934 (void *)(__psint_t)flags,
3935 (void *)(unsigned long)current_cpu(),
3947 * Convert inode from non-attributed to attributed.
3948 * Must not be in a transaction, ip must not be locked.
3950 int /* error code */
3951 xfs_bmap_add_attrfork(
3952 xfs_inode_t *ip, /* incore inode pointer */
3953 int size, /* space new attribute needs */
3954 int rsvd) /* xact may use reserved blks */
3956 xfs_fsblock_t firstblock; /* 1st block/ag allocated */
3957 xfs_bmap_free_t flist; /* freed extent records */
3958 xfs_mount_t *mp; /* mount structure */
3959 xfs_trans_t *tp; /* transaction pointer */
3960 unsigned long s; /* spinlock spl value */
3961 int blks; /* space reservation */
3962 int version = 1; /* superblock attr version */
3963 int committed; /* xaction was committed */
3964 int logflags; /* logging flags */
3965 int error; /* error return value */
3967 ASSERT(XFS_IFORK_Q(ip) == 0);
3968 ASSERT(ip->i_df.if_ext_max ==
3969 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3972 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3973 tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3974 blks = XFS_ADDAFORK_SPACE_RES(mp);
3976 tp->t_flags |= XFS_TRANS_RESERVE;
3977 if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3978 XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3980 xfs_ilock(ip, XFS_ILOCK_EXCL);
3981 error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip, blks, 0, rsvd ?
3982 XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3983 XFS_QMOPT_RES_REGBLKS);
3985 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3986 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3989 if (XFS_IFORK_Q(ip))
3991 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3993 * For inodes coming from pre-6.2 filesystems.
3995 ASSERT(ip->i_d.di_aformat == 0);
3996 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
3998 ASSERT(ip->i_d.di_anextents == 0);
3999 VN_HOLD(XFS_ITOV(ip));
4000 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4001 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4002 switch (ip->i_d.di_format) {
4003 case XFS_DINODE_FMT_DEV:
4004 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
4006 case XFS_DINODE_FMT_UUID:
4007 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
4009 case XFS_DINODE_FMT_LOCAL:
4010 case XFS_DINODE_FMT_EXTENTS:
4011 case XFS_DINODE_FMT_BTREE:
4012 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
4013 if (!ip->i_d.di_forkoff)
4014 ip->i_d.di_forkoff = mp->m_attroffset >> 3;
4015 else if (mp->m_flags & XFS_MOUNT_ATTR2)
4020 error = XFS_ERROR(EINVAL);
4023 ip->i_df.if_ext_max =
4024 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
4025 ASSERT(ip->i_afp == NULL);
4026 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
4027 ip->i_afp->if_ext_max =
4028 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
4029 ip->i_afp->if_flags = XFS_IFEXTENTS;
4031 XFS_BMAP_INIT(&flist, &firstblock);
4032 switch (ip->i_d.di_format) {
4033 case XFS_DINODE_FMT_LOCAL:
4034 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
4037 case XFS_DINODE_FMT_EXTENTS:
4038 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
4041 case XFS_DINODE_FMT_BTREE:
4042 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
4050 xfs_trans_log_inode(tp, ip, logflags);
4053 if (!XFS_SB_VERSION_HASATTR(&mp->m_sb) ||
4054 (!XFS_SB_VERSION_HASATTR2(&mp->m_sb) && version == 2)) {
4055 __int64_t sbfields = 0;
4057 s = XFS_SB_LOCK(mp);
4058 if (!XFS_SB_VERSION_HASATTR(&mp->m_sb)) {
4059 XFS_SB_VERSION_ADDATTR(&mp->m_sb);
4060 sbfields |= XFS_SB_VERSIONNUM;
4062 if (!XFS_SB_VERSION_HASATTR2(&mp->m_sb) && version == 2) {
4063 XFS_SB_VERSION_ADDATTR2(&mp->m_sb);
4064 sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
4067 XFS_SB_UNLOCK(mp, s);
4068 xfs_mod_sb(tp, sbfields);
4070 XFS_SB_UNLOCK(mp, s);
4072 if ((error = xfs_bmap_finish(&tp, &flist, &committed)))
4074 error = xfs_trans_commit(tp, XFS_TRANS_PERM_LOG_RES, NULL);
4075 ASSERT(ip->i_df.if_ext_max ==
4076 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
4079 xfs_bmap_cancel(&flist);
4081 ASSERT(ismrlocked(&ip->i_lock,MR_UPDATE));
4082 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4084 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
4085 ASSERT(ip->i_df.if_ext_max ==
4086 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
4091 * Add the extent to the list of extents to be free at transaction end.
4092 * The list is maintained sorted (by block number).
4097 xfs_fsblock_t bno, /* fs block number of extent */
4098 xfs_filblks_t len, /* length of extent */
4099 xfs_bmap_free_t *flist, /* list of extents */
4100 xfs_mount_t *mp) /* mount point structure */
4102 xfs_bmap_free_item_t *cur; /* current (next) element */
4103 xfs_bmap_free_item_t *new; /* new element */
4104 xfs_bmap_free_item_t *prev; /* previous element */
4106 xfs_agnumber_t agno;
4107 xfs_agblock_t agbno;
4109 ASSERT(bno != NULLFSBLOCK);
4111 ASSERT(len <= MAXEXTLEN);
4112 ASSERT(!ISNULLSTARTBLOCK(bno));
4113 agno = XFS_FSB_TO_AGNO(mp, bno);
4114 agbno = XFS_FSB_TO_AGBNO(mp, bno);
4115 ASSERT(agno < mp->m_sb.sb_agcount);
4116 ASSERT(agbno < mp->m_sb.sb_agblocks);
4117 ASSERT(len < mp->m_sb.sb_agblocks);
4118 ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
4120 ASSERT(xfs_bmap_free_item_zone != NULL);
4121 new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
4122 new->xbfi_startblock = bno;
4123 new->xbfi_blockcount = (xfs_extlen_t)len;
4124 for (prev = NULL, cur = flist->xbf_first;
4126 prev = cur, cur = cur->xbfi_next) {
4127 if (cur->xbfi_startblock >= bno)
4131 prev->xbfi_next = new;
4133 flist->xbf_first = new;
4134 new->xbfi_next = cur;
4139 * Compute and fill in the value of the maximum depth of a bmap btree
4140 * in this filesystem. Done once, during mount.
4143 xfs_bmap_compute_maxlevels(
4144 xfs_mount_t *mp, /* file system mount structure */
4145 int whichfork) /* data or attr fork */
4147 int level; /* btree level */
4148 uint maxblocks; /* max blocks at this level */
4149 uint maxleafents; /* max leaf entries possible */
4150 int maxrootrecs; /* max records in root block */
4151 int minleafrecs; /* min records in leaf block */
4152 int minnoderecs; /* min records in node block */
4153 int sz; /* root block size */
4156 * The maximum number of extents in a file, hence the maximum
4157 * number of leaf entries, is controlled by the type of di_nextents
4158 * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
4159 * (a signed 16-bit number, xfs_aextnum_t).
4161 if (whichfork == XFS_DATA_FORK) {
4162 maxleafents = MAXEXTNUM;
4163 sz = (mp->m_flags & XFS_MOUNT_ATTR2) ?
4164 XFS_BMDR_SPACE_CALC(MINDBTPTRS) : mp->m_attroffset;
4166 maxleafents = MAXAEXTNUM;
4167 sz = (mp->m_flags & XFS_MOUNT_ATTR2) ?
4168 XFS_BMDR_SPACE_CALC(MINABTPTRS) :
4169 mp->m_sb.sb_inodesize - mp->m_attroffset;
4171 maxrootrecs = (int)XFS_BTREE_BLOCK_MAXRECS(sz, xfs_bmdr, 0);
4172 minleafrecs = mp->m_bmap_dmnr[0];
4173 minnoderecs = mp->m_bmap_dmnr[1];
4174 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
4175 for (level = 1; maxblocks > 1; level++) {
4176 if (maxblocks <= maxrootrecs)
4179 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
4181 mp->m_bm_maxlevels[whichfork] = level;
4185 * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
4186 * caller. Frees all the extents that need freeing, which must be done
4187 * last due to locking considerations. We never free any extents in
4188 * the first transaction. This is to allow the caller to make the first
4189 * transaction a synchronous one so that the pointers to the data being
4190 * broken in this transaction will be permanent before the data is actually
4191 * freed. This is necessary to prevent blocks from being reallocated
4192 * and written to before the free and reallocation are actually permanent.
4193 * We do not just make the first transaction synchronous here, because
4194 * there are more efficient ways to gain the same protection in some cases
4195 * (see the file truncation code).
4197 * Return 1 if the given transaction was committed and a new one
4198 * started, and 0 otherwise in the committed parameter.
4203 xfs_trans_t **tp, /* transaction pointer addr */
4204 xfs_bmap_free_t *flist, /* i/o: list extents to free */
4205 int *committed) /* xact committed or not */
4207 xfs_efd_log_item_t *efd; /* extent free data */
4208 xfs_efi_log_item_t *efi; /* extent free intention */
4209 int error; /* error return value */
4210 xfs_bmap_free_item_t *free; /* free extent item */
4211 unsigned int logres; /* new log reservation */
4212 unsigned int logcount; /* new log count */
4213 xfs_mount_t *mp; /* filesystem mount structure */
4214 xfs_bmap_free_item_t *next; /* next item on free list */
4215 xfs_trans_t *ntp; /* new transaction pointer */
4217 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
4218 if (flist->xbf_count == 0) {
4223 efi = xfs_trans_get_efi(ntp, flist->xbf_count);
4224 for (free = flist->xbf_first; free; free = free->xbfi_next)
4225 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
4226 free->xbfi_blockcount);
4227 logres = ntp->t_log_res;
4228 logcount = ntp->t_log_count;
4229 ntp = xfs_trans_dup(*tp);
4230 error = xfs_trans_commit(*tp, 0, NULL);
4234 * We have a new transaction, so we should return committed=1,
4235 * even though we're returning an error.
4240 if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
4243 efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
4244 for (free = flist->xbf_first; free != NULL; free = next) {
4245 next = free->xbfi_next;
4246 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
4247 free->xbfi_blockcount))) {
4249 * The bmap free list will be cleaned up at a
4250 * higher level. The EFI will be canceled when
4251 * this transaction is aborted.
4252 * Need to force shutdown here to make sure it
4253 * happens, since this transaction may not be
4257 if (!XFS_FORCED_SHUTDOWN(mp))
4258 xfs_force_shutdown(mp,
4259 (error == EFSCORRUPTED) ?
4260 SHUTDOWN_CORRUPT_INCORE :
4261 SHUTDOWN_META_IO_ERROR);
4264 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
4265 free->xbfi_blockcount);
4266 xfs_bmap_del_free(flist, NULL, free);
4272 * Free up any items left in the list.
4276 xfs_bmap_free_t *flist) /* list of bmap_free_items */
4278 xfs_bmap_free_item_t *free; /* free list item */
4279 xfs_bmap_free_item_t *next;
4281 if (flist->xbf_count == 0)
4283 ASSERT(flist->xbf_first != NULL);
4284 for (free = flist->xbf_first; free; free = next) {
4285 next = free->xbfi_next;
4286 xfs_bmap_del_free(flist, NULL, free);
4288 ASSERT(flist->xbf_count == 0);
4292 * Returns the file-relative block number of the first unused block(s)
4293 * in the file with at least "len" logically contiguous blocks free.
4294 * This is the lowest-address hole if the file has holes, else the first block
4295 * past the end of file.
4296 * Return 0 if the file is currently local (in-inode).
4299 xfs_bmap_first_unused(
4300 xfs_trans_t *tp, /* transaction pointer */
4301 xfs_inode_t *ip, /* incore inode */
4302 xfs_extlen_t len, /* size of hole to find */
4303 xfs_fileoff_t *first_unused, /* unused block */
4304 int whichfork) /* data or attr fork */
4306 xfs_bmbt_rec_t *ep; /* pointer to an extent entry */
4307 int error; /* error return value */
4308 int idx; /* extent record index */
4309 xfs_ifork_t *ifp; /* inode fork pointer */
4310 xfs_fileoff_t lastaddr; /* last block number seen */
4311 xfs_fileoff_t lowest; /* lowest useful block */
4312 xfs_fileoff_t max; /* starting useful block */
4313 xfs_fileoff_t off; /* offset for this block */
4314 xfs_extnum_t nextents; /* number of extent entries */
4316 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
4317 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
4318 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
4319 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4323 ifp = XFS_IFORK_PTR(ip, whichfork);
4324 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4325 (error = xfs_iread_extents(tp, ip, whichfork)))
4327 lowest = *first_unused;
4328 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4329 for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
4330 ep = xfs_iext_get_ext(ifp, idx);
4331 off = xfs_bmbt_get_startoff(ep);
4333 * See if the hole before this extent will work.
4335 if (off >= lowest + len && off - max >= len) {
4336 *first_unused = max;
4339 lastaddr = off + xfs_bmbt_get_blockcount(ep);
4340 max = XFS_FILEOFF_MAX(lastaddr, lowest);
4342 *first_unused = max;
4347 * Returns the file-relative block number of the last block + 1 before
4348 * last_block (input value) in the file.
4349 * This is not based on i_size, it is based on the extent records.
4350 * Returns 0 for local files, as they do not have extent records.
4353 xfs_bmap_last_before(
4354 xfs_trans_t *tp, /* transaction pointer */
4355 xfs_inode_t *ip, /* incore inode */
4356 xfs_fileoff_t *last_block, /* last block */
4357 int whichfork) /* data or attr fork */
4359 xfs_fileoff_t bno; /* input file offset */
4360 int eof; /* hit end of file */
4361 xfs_bmbt_rec_t *ep; /* pointer to last extent */
4362 int error; /* error return value */
4363 xfs_bmbt_irec_t got; /* current extent value */
4364 xfs_ifork_t *ifp; /* inode fork pointer */
4365 xfs_extnum_t lastx; /* last extent used */
4366 xfs_bmbt_irec_t prev; /* previous extent value */
4368 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4369 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4370 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4371 return XFS_ERROR(EIO);
4372 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4376 ifp = XFS_IFORK_PTR(ip, whichfork);
4377 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4378 (error = xfs_iread_extents(tp, ip, whichfork)))
4380 bno = *last_block - 1;
4381 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4383 if (eof || xfs_bmbt_get_startoff(ep) > bno) {
4384 if (prev.br_startoff == NULLFILEOFF)
4387 *last_block = prev.br_startoff + prev.br_blockcount;
4390 * Otherwise *last_block is already the right answer.
4396 * Returns the file-relative block number of the first block past eof in
4397 * the file. This is not based on i_size, it is based on the extent records.
4398 * Returns 0 for local files, as they do not have extent records.
4401 xfs_bmap_last_offset(
4402 xfs_trans_t *tp, /* transaction pointer */
4403 xfs_inode_t *ip, /* incore inode */
4404 xfs_fileoff_t *last_block, /* last block */
4405 int whichfork) /* data or attr fork */
4407 xfs_bmbt_rec_t *ep; /* pointer to last extent */
4408 int error; /* error return value */
4409 xfs_ifork_t *ifp; /* inode fork pointer */
4410 xfs_extnum_t nextents; /* number of extent entries */
4412 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4413 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4414 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4415 return XFS_ERROR(EIO);
4416 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4420 ifp = XFS_IFORK_PTR(ip, whichfork);
4421 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4422 (error = xfs_iread_extents(tp, ip, whichfork)))
4424 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4429 ep = xfs_iext_get_ext(ifp, nextents - 1);
4430 *last_block = xfs_bmbt_get_startoff(ep) + xfs_bmbt_get_blockcount(ep);
4435 * Returns whether the selected fork of the inode has exactly one
4436 * block or not. For the data fork we check this matches di_size,
4437 * implying the file's range is 0..bsize-1.
4439 int /* 1=>1 block, 0=>otherwise */
4441 xfs_inode_t *ip, /* incore inode */
4442 int whichfork) /* data or attr fork */
4444 xfs_bmbt_rec_t *ep; /* ptr to fork's extent */
4445 xfs_ifork_t *ifp; /* inode fork pointer */
4446 int rval; /* return value */
4447 xfs_bmbt_irec_t s; /* internal version of extent */
4450 if (whichfork == XFS_DATA_FORK)
4451 return ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize;
4453 if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4455 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4457 ifp = XFS_IFORK_PTR(ip, whichfork);
4458 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4459 ep = xfs_iext_get_ext(ifp, 0);
4460 xfs_bmbt_get_all(ep, &s);
4461 rval = s.br_startoff == 0 && s.br_blockcount == 1;
4462 if (rval && whichfork == XFS_DATA_FORK)
4463 ASSERT(ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize);
4468 * Read in the extents to if_extents.
4469 * All inode fields are set up by caller, we just traverse the btree
4470 * and copy the records in. If the file system cannot contain unwritten
4471 * extents, the records are checked for no "state" flags.
4474 xfs_bmap_read_extents(
4475 xfs_trans_t *tp, /* transaction pointer */
4476 xfs_inode_t *ip, /* incore inode */
4477 int whichfork) /* data or attr fork */
4479 xfs_bmbt_block_t *block; /* current btree block */
4480 xfs_fsblock_t bno; /* block # of "block" */
4481 xfs_buf_t *bp; /* buffer for "block" */
4482 int error; /* error return value */
4483 xfs_exntfmt_t exntf; /* XFS_EXTFMT_NOSTATE, if checking */
4484 #ifdef XFS_BMAP_TRACE
4485 static char fname[] = "xfs_bmap_read_extents";
4487 xfs_extnum_t i, j; /* index into the extents list */
4488 xfs_ifork_t *ifp; /* fork structure */
4489 int level; /* btree level, for checking */
4490 xfs_mount_t *mp; /* file system mount structure */
4491 __be64 *pp; /* pointer to block address */
4493 xfs_extnum_t room; /* number of entries there's room for */
4497 ifp = XFS_IFORK_PTR(ip, whichfork);
4498 exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4499 XFS_EXTFMT_INODE(ip);
4500 block = ifp->if_broot;
4502 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4504 level = be16_to_cpu(block->bb_level);
4506 pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
4507 bno = be64_to_cpu(*pp);
4508 ASSERT(bno != NULLDFSBNO);
4509 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
4510 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
4512 * Go down the tree until leaf level is reached, following the first
4513 * pointer (leftmost) at each level.
4515 while (level-- > 0) {
4516 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4517 XFS_BMAP_BTREE_REF)))
4519 block = XFS_BUF_TO_BMBT_BLOCK(bp);
4520 XFS_WANT_CORRUPTED_GOTO(
4521 XFS_BMAP_SANITY_CHECK(mp, block, level),
4525 pp = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, 1, mp->m_bmap_dmxr[1]);
4526 bno = be64_to_cpu(*pp);
4527 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
4528 xfs_trans_brelse(tp, bp);
4531 * Here with bp and block set to the leftmost leaf node in the tree.
4533 room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4536 * Loop over all leaf nodes. Copy information to the extent records.
4539 xfs_bmbt_rec_t *frp, *trp;
4540 xfs_fsblock_t nextbno;
4541 xfs_extnum_t num_recs;
4545 num_recs = be16_to_cpu(block->bb_numrecs);
4546 if (unlikely(i + num_recs > room)) {
4547 ASSERT(i + num_recs <= room);
4548 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
4549 "corrupt dinode %Lu, (btree extents).",
4550 (unsigned long long) ip->i_ino);
4551 XFS_ERROR_REPORT("xfs_bmap_read_extents(1)",
4556 XFS_WANT_CORRUPTED_GOTO(
4557 XFS_BMAP_SANITY_CHECK(mp, block, 0),
4560 * Read-ahead the next leaf block, if any.
4562 nextbno = be64_to_cpu(block->bb_rightsib);
4563 if (nextbno != NULLFSBLOCK)
4564 xfs_btree_reada_bufl(mp, nextbno, 1);
4566 * Copy records into the extent records.
4568 frp = XFS_BTREE_REC_ADDR(xfs_bmbt, block, 1);
4570 for (j = 0; j < num_recs; j++, i++, frp++) {
4571 trp = xfs_iext_get_ext(ifp, i);
4572 trp->l0 = INT_GET(frp->l0, ARCH_CONVERT);
4573 trp->l1 = INT_GET(frp->l1, ARCH_CONVERT);
4575 if (exntf == XFS_EXTFMT_NOSTATE) {
4577 * Check all attribute bmap btree records and
4578 * any "older" data bmap btree records for a
4579 * set bit in the "extent flag" position.
4581 if (unlikely(xfs_check_nostate_extents(ifp,
4582 start, num_recs))) {
4583 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4589 xfs_trans_brelse(tp, bp);
4592 * If we've reached the end, stop.
4594 if (bno == NULLFSBLOCK)
4596 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4597 XFS_BMAP_BTREE_REF)))
4599 block = XFS_BUF_TO_BMBT_BLOCK(bp);
4601 ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4602 ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4603 xfs_bmap_trace_exlist(fname, ip, i, whichfork);
4606 xfs_trans_brelse(tp, bp);
4607 return XFS_ERROR(EFSCORRUPTED);
4610 #ifdef XFS_BMAP_TRACE
4612 * Add bmap trace insert entries for all the contents of the extent records.
4615 xfs_bmap_trace_exlist(
4616 char *fname, /* function name */
4617 xfs_inode_t *ip, /* incore inode pointer */
4618 xfs_extnum_t cnt, /* count of entries in the list */
4619 int whichfork) /* data or attr fork */
4621 xfs_bmbt_rec_t *ep; /* current extent record */
4622 xfs_extnum_t idx; /* extent record index */
4623 xfs_ifork_t *ifp; /* inode fork pointer */
4624 xfs_bmbt_irec_t s; /* file extent record */
4626 ifp = XFS_IFORK_PTR(ip, whichfork);
4627 ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4628 for (idx = 0; idx < cnt; idx++) {
4629 ep = xfs_iext_get_ext(ifp, idx);
4630 xfs_bmbt_get_all(ep, &s);
4631 xfs_bmap_trace_insert(fname, "exlist", ip, idx, 1, &s, NULL,
4639 * Validate that the bmbt_irecs being returned from bmapi are valid
4640 * given the callers original parameters. Specifically check the
4641 * ranges of the returned irecs to ensure that they only extent beyond
4642 * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4645 xfs_bmap_validate_ret(
4649 xfs_bmbt_irec_t *mval,
4653 int i; /* index to map values */
4655 ASSERT(ret_nmap <= nmap);
4657 for (i = 0; i < ret_nmap; i++) {
4658 ASSERT(mval[i].br_blockcount > 0);
4659 if (!(flags & XFS_BMAPI_ENTIRE)) {
4660 ASSERT(mval[i].br_startoff >= bno);
4661 ASSERT(mval[i].br_blockcount <= len);
4662 ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4665 ASSERT(mval[i].br_startoff < bno + len);
4666 ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4670 mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4671 mval[i].br_startoff);
4672 if ((flags & XFS_BMAPI_WRITE) && !(flags & XFS_BMAPI_DELAY))
4673 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4674 mval[i].br_startblock != HOLESTARTBLOCK);
4675 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4676 mval[i].br_state == XFS_EXT_UNWRITTEN);
4683 * Map file blocks to filesystem blocks.
4684 * File range is given by the bno/len pair.
4685 * Adds blocks to file if a write ("flags & XFS_BMAPI_WRITE" set)
4686 * into a hole or past eof.
4687 * Only allocates blocks from a single allocation group,
4688 * to avoid locking problems.
4689 * The returned value in "firstblock" from the first call in a transaction
4690 * must be remembered and presented to subsequent calls in "firstblock".
4691 * An upper bound for the number of blocks to be allocated is supplied to
4692 * the first call in "total"; if no allocation group has that many free
4693 * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4697 xfs_trans_t *tp, /* transaction pointer */
4698 xfs_inode_t *ip, /* incore inode */
4699 xfs_fileoff_t bno, /* starting file offs. mapped */
4700 xfs_filblks_t len, /* length to map in file */
4701 int flags, /* XFS_BMAPI_... */
4702 xfs_fsblock_t *firstblock, /* first allocated block
4703 controls a.g. for allocs */
4704 xfs_extlen_t total, /* total blocks needed */
4705 xfs_bmbt_irec_t *mval, /* output: map values */
4706 int *nmap, /* i/o: mval size/count */
4707 xfs_bmap_free_t *flist, /* i/o: list extents to free */
4708 xfs_extdelta_t *delta) /* o: change made to incore extents */
4710 xfs_fsblock_t abno; /* allocated block number */
4711 xfs_extlen_t alen; /* allocated extent length */
4712 xfs_fileoff_t aoff; /* allocated file offset */
4713 xfs_bmalloca_t bma; /* args for xfs_bmap_alloc */
4714 xfs_btree_cur_t *cur; /* bmap btree cursor */
4715 xfs_fileoff_t end; /* end of mapped file region */
4716 int eof; /* we've hit the end of extents */
4717 xfs_bmbt_rec_t *ep; /* extent record pointer */
4718 int error; /* error return */
4719 xfs_bmbt_irec_t got; /* current file extent record */
4720 xfs_ifork_t *ifp; /* inode fork pointer */
4721 xfs_extlen_t indlen; /* indirect blocks length */
4722 xfs_extnum_t lastx; /* last useful extent number */
4723 int logflags; /* flags for transaction logging */
4724 xfs_extlen_t minleft; /* min blocks left after allocation */
4725 xfs_extlen_t minlen; /* min allocation size */
4726 xfs_mount_t *mp; /* xfs mount structure */
4727 int n; /* current extent index */
4728 int nallocs; /* number of extents alloc\'d */
4729 xfs_extnum_t nextents; /* number of extents in file */
4730 xfs_fileoff_t obno; /* old block number (offset) */
4731 xfs_bmbt_irec_t prev; /* previous file extent record */
4732 int tmp_logflags; /* temp flags holder */
4733 int whichfork; /* data or attr fork */
4734 char inhole; /* current location is hole in file */
4735 char wasdelay; /* old extent was delayed */
4736 char wr; /* this is a write request */
4737 char rt; /* this is a realtime file */
4739 xfs_fileoff_t orig_bno; /* original block number value */
4740 int orig_flags; /* original flags arg value */
4741 xfs_filblks_t orig_len; /* original value of len arg */
4742 xfs_bmbt_irec_t *orig_mval; /* original value of mval */
4743 int orig_nmap; /* original value of *nmap */
4752 ASSERT(*nmap <= XFS_BMAP_MAX_NMAP || !(flags & XFS_BMAPI_WRITE));
4753 whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4754 XFS_ATTR_FORK : XFS_DATA_FORK;
4756 if (unlikely(XFS_TEST_ERROR(
4757 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4758 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4759 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4760 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4761 XFS_ERROR_REPORT("xfs_bmapi", XFS_ERRLEVEL_LOW, mp);
4762 return XFS_ERROR(EFSCORRUPTED);
4764 if (XFS_FORCED_SHUTDOWN(mp))
4765 return XFS_ERROR(EIO);
4766 rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4767 ifp = XFS_IFORK_PTR(ip, whichfork);
4768 ASSERT(ifp->if_ext_max ==
4769 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4770 if ((wr = (flags & XFS_BMAPI_WRITE)) != 0)
4771 XFS_STATS_INC(xs_blk_mapw);
4773 XFS_STATS_INC(xs_blk_mapr);
4775 * IGSTATE flag is used to combine extents which
4776 * differ only due to the state of the extents.
4777 * This technique is used from xfs_getbmap()
4778 * when the caller does not wish to see the
4779 * separation (which is the default).
4781 * This technique is also used when writing a
4782 * buffer which has been partially written,
4783 * (usually by being flushed during a chunkread),
4784 * to ensure one write takes place. This also
4785 * prevents a change in the xfs inode extents at
4786 * this time, intentionally. This change occurs
4787 * on completion of the write operation, in
4788 * xfs_strat_comp(), where the xfs_bmapi() call
4789 * is transactioned, and the extents combined.
4791 if ((flags & XFS_BMAPI_IGSTATE) && wr) /* if writing unwritten space */
4792 wr = 0; /* no allocations are allowed */
4793 ASSERT(wr || !(flags & XFS_BMAPI_DELAY));
4797 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4799 if ((error = xfs_bmap_local_to_extents(tp, ip,
4800 firstblock, total, &logflags, whichfork)))
4803 if (wr && *firstblock == NULLFSBLOCK) {
4804 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4805 minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4810 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4811 (error = xfs_iread_extents(tp, ip, whichfork)))
4813 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4815 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4821 delta->xed_startoff = NULLFILEOFF;
4822 delta->xed_blockcount = 0;
4824 while (bno < end && n < *nmap) {
4826 * Reading past eof, act as though there's a hole
4830 got.br_startoff = end;
4831 inhole = eof || got.br_startoff > bno;
4832 wasdelay = wr && !inhole && !(flags & XFS_BMAPI_DELAY) &&
4833 ISNULLSTARTBLOCK(got.br_startblock);
4835 * First, deal with the hole before the allocated space
4836 * that we found, if any.
4838 if (wr && (inhole || wasdelay)) {
4840 * For the wasdelay case, we could also just
4841 * allocate the stuff asked for in this bmap call
4842 * but that wouldn't be as good.
4844 if (wasdelay && !(flags & XFS_BMAPI_EXACT)) {
4845 alen = (xfs_extlen_t)got.br_blockcount;
4846 aoff = got.br_startoff;
4847 if (lastx != NULLEXTNUM && lastx) {
4848 ep = xfs_iext_get_ext(ifp, lastx - 1);
4849 xfs_bmbt_get_all(ep, &prev);
4851 } else if (wasdelay) {
4852 alen = (xfs_extlen_t)
4853 XFS_FILBLKS_MIN(len,
4855 got.br_blockcount) - bno);
4858 alen = (xfs_extlen_t)
4859 XFS_FILBLKS_MIN(len, MAXEXTLEN);
4861 alen = (xfs_extlen_t)
4862 XFS_FILBLKS_MIN(alen,
4863 got.br_startoff - bno);
4866 minlen = (flags & XFS_BMAPI_CONTIG) ? alen : 1;
4867 if (flags & XFS_BMAPI_DELAY) {
4870 /* Figure out the extent size, adjust alen */
4872 if (!(extsz = ip->i_d.di_extsize))
4873 extsz = mp->m_sb.sb_rextsize;
4875 extsz = ip->i_d.di_extsize;
4878 error = xfs_bmap_extsize_align(mp,
4881 flags&XFS_BMAPI_DELAY,
4882 flags&XFS_BMAPI_CONVERT,
4888 extsz = alen / mp->m_sb.sb_rextsize;
4891 * Make a transaction-less quota reservation for
4892 * delayed allocation blocks. This number gets
4893 * adjusted later. We return if we haven't
4894 * allocated blocks already inside this loop.
4896 if ((error = XFS_TRANS_RESERVE_QUOTA_NBLKS(
4897 mp, NULL, ip, (long)alen, 0,
4898 rt ? XFS_QMOPT_RES_RTBLKS :
4899 XFS_QMOPT_RES_REGBLKS))) {
4902 ASSERT(cur == NULL);
4909 * Split changing sb for alen and indlen since
4910 * they could be coming from different places.
4912 indlen = (xfs_extlen_t)
4913 xfs_bmap_worst_indlen(ip, alen);
4917 error = xfs_mod_incore_sb(mp,
4919 -((int64_t)extsz), (flags &
4920 XFS_BMAPI_RSVBLOCKS));
4922 error = xfs_mod_incore_sb(mp,
4924 -((int64_t)alen), (flags &
4925 XFS_BMAPI_RSVBLOCKS));
4928 error = xfs_mod_incore_sb(mp,
4930 -((int64_t)indlen), (flags &
4931 XFS_BMAPI_RSVBLOCKS));
4933 xfs_mod_incore_sb(mp,
4935 (int64_t)extsz, (flags &
4936 XFS_BMAPI_RSVBLOCKS));
4938 xfs_mod_incore_sb(mp,
4940 (int64_t)alen, (flags &
4941 XFS_BMAPI_RSVBLOCKS));
4945 if (XFS_IS_QUOTA_ON(mp))
4946 /* unreserve the blocks now */
4948 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(
4951 XFS_QMOPT_RES_RTBLKS :
4952 XFS_QMOPT_RES_REGBLKS);
4956 ip->i_delayed_blks += alen;
4957 abno = NULLSTARTBLOCK(indlen);
4960 * If first time, allocate and fill in
4961 * once-only bma fields.
4963 if (bma.ip == NULL) {
4971 /* Indicate if this is the first user data
4972 * in the file, or just any user data.
4974 if (!(flags & XFS_BMAPI_METADATA)) {
4975 bma.userdata = (aoff == 0) ?
4976 XFS_ALLOC_INITIAL_USER_DATA :
4980 * Fill in changeable bma fields.
4983 bma.firstblock = *firstblock;
4986 bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4987 bma.wasdel = wasdelay;
4988 bma.minlen = minlen;
4989 bma.low = flist->xbf_low;
4990 bma.minleft = minleft;
4992 * Only want to do the alignment at the
4993 * eof if it is userdata and allocation length
4994 * is larger than a stripe unit.
4996 if (mp->m_dalign && alen >= mp->m_dalign &&
4997 (!(flags & XFS_BMAPI_METADATA)) &&
4998 (whichfork == XFS_DATA_FORK)) {
4999 if ((error = xfs_bmap_isaeof(ip, aoff,
5000 whichfork, &bma.aeof)))
5007 if ((error = xfs_bmap_alloc(&bma)))
5010 * Copy out result fields.
5013 if ((flist->xbf_low = bma.low))
5017 ASSERT(*firstblock == NULLFSBLOCK ||
5018 XFS_FSB_TO_AGNO(mp, *firstblock) ==
5019 XFS_FSB_TO_AGNO(mp, bma.firstblock) ||
5021 XFS_FSB_TO_AGNO(mp, *firstblock) <
5022 XFS_FSB_TO_AGNO(mp, bma.firstblock)));
5023 *firstblock = bma.firstblock;
5025 cur->bc_private.b.firstblock =
5027 if (abno == NULLFSBLOCK)
5029 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
5030 cur = xfs_btree_init_cursor(mp,
5031 tp, NULL, 0, XFS_BTNUM_BMAP,
5033 cur->bc_private.b.firstblock =
5035 cur->bc_private.b.flist = flist;
5038 * Bump the number of extents we've allocated
5044 cur->bc_private.b.flags =
5045 wasdelay ? XFS_BTCUR_BPRV_WASDEL : 0;
5046 got.br_startoff = aoff;
5047 got.br_startblock = abno;
5048 got.br_blockcount = alen;
5049 got.br_state = XFS_EXT_NORM; /* assume normal */
5051 * Determine state of extent, and the filesystem.
5052 * A wasdelay extent has been initialized, so
5053 * shouldn't be flagged as unwritten.
5055 if (wr && XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
5056 if (!wasdelay && (flags & XFS_BMAPI_PREALLOC))
5057 got.br_state = XFS_EXT_UNWRITTEN;
5059 error = xfs_bmap_add_extent(ip, lastx, &cur, &got,
5060 firstblock, flist, &tmp_logflags, delta,
5061 whichfork, (flags & XFS_BMAPI_RSVBLOCKS));
5062 logflags |= tmp_logflags;
5065 lastx = ifp->if_lastex;
5066 ep = xfs_iext_get_ext(ifp, lastx);
5067 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5068 xfs_bmbt_get_all(ep, &got);
5069 ASSERT(got.br_startoff <= aoff);
5070 ASSERT(got.br_startoff + got.br_blockcount >=
5073 if (flags & XFS_BMAPI_DELAY) {
5074 ASSERT(ISNULLSTARTBLOCK(got.br_startblock));
5075 ASSERT(STARTBLOCKVAL(got.br_startblock) > 0);
5077 ASSERT(got.br_state == XFS_EXT_NORM ||
5078 got.br_state == XFS_EXT_UNWRITTEN);
5081 * Fall down into the found allocated space case.
5083 } else if (inhole) {
5085 * Reading in a hole.
5087 mval->br_startoff = bno;
5088 mval->br_startblock = HOLESTARTBLOCK;
5089 mval->br_blockcount =
5090 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
5091 mval->br_state = XFS_EXT_NORM;
5092 bno += mval->br_blockcount;
5093 len -= mval->br_blockcount;
5099 * Then deal with the allocated space we found.
5102 if (!(flags & XFS_BMAPI_ENTIRE) &&
5103 (got.br_startoff + got.br_blockcount > obno)) {
5106 ASSERT((bno >= obno) || (n == 0));
5108 mval->br_startoff = bno;
5109 if (ISNULLSTARTBLOCK(got.br_startblock)) {
5110 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
5111 mval->br_startblock = DELAYSTARTBLOCK;
5113 mval->br_startblock =
5115 (bno - got.br_startoff);
5117 * Return the minimum of what we got and what we
5118 * asked for for the length. We can use the len
5119 * variable here because it is modified below
5120 * and we could have been there before coming
5121 * here if the first part of the allocation
5122 * didn't overlap what was asked for.
5124 mval->br_blockcount =
5125 XFS_FILBLKS_MIN(end - bno, got.br_blockcount -
5126 (bno - got.br_startoff));
5127 mval->br_state = got.br_state;
5128 ASSERT(mval->br_blockcount <= len);
5131 if (ISNULLSTARTBLOCK(mval->br_startblock)) {
5132 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
5133 mval->br_startblock = DELAYSTARTBLOCK;
5138 * Check if writing previously allocated but
5139 * unwritten extents.
5141 if (wr && mval->br_state == XFS_EXT_UNWRITTEN &&
5142 ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_DELAY)) == 0)) {
5144 * Modify (by adding) the state flag, if writing.
5146 ASSERT(mval->br_blockcount <= len);
5147 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
5148 cur = xfs_btree_init_cursor(mp,
5149 tp, NULL, 0, XFS_BTNUM_BMAP,
5151 cur->bc_private.b.firstblock =
5153 cur->bc_private.b.flist = flist;
5155 mval->br_state = XFS_EXT_NORM;
5156 error = xfs_bmap_add_extent(ip, lastx, &cur, mval,
5157 firstblock, flist, &tmp_logflags, delta,
5158 whichfork, (flags & XFS_BMAPI_RSVBLOCKS));
5159 logflags |= tmp_logflags;
5162 lastx = ifp->if_lastex;
5163 ep = xfs_iext_get_ext(ifp, lastx);
5164 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5165 xfs_bmbt_get_all(ep, &got);
5167 * We may have combined previously unwritten
5168 * space with written space, so generate
5171 if (mval->br_blockcount < len)
5175 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
5176 ((mval->br_startoff + mval->br_blockcount) <= end));
5177 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
5178 (mval->br_blockcount <= len) ||
5179 (mval->br_startoff < obno));
5180 bno = mval->br_startoff + mval->br_blockcount;
5182 if (n > 0 && mval->br_startoff == mval[-1].br_startoff) {
5183 ASSERT(mval->br_startblock == mval[-1].br_startblock);
5184 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
5185 ASSERT(mval->br_state == mval[-1].br_state);
5186 mval[-1].br_blockcount = mval->br_blockcount;
5187 mval[-1].br_state = mval->br_state;
5188 } else if (n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
5189 mval[-1].br_startblock != DELAYSTARTBLOCK &&
5190 mval[-1].br_startblock != HOLESTARTBLOCK &&
5191 mval->br_startblock ==
5192 mval[-1].br_startblock + mval[-1].br_blockcount &&
5193 ((flags & XFS_BMAPI_IGSTATE) ||
5194 mval[-1].br_state == mval->br_state)) {
5195 ASSERT(mval->br_startoff ==
5196 mval[-1].br_startoff + mval[-1].br_blockcount);
5197 mval[-1].br_blockcount += mval->br_blockcount;
5199 mval->br_startblock == DELAYSTARTBLOCK &&
5200 mval[-1].br_startblock == DELAYSTARTBLOCK &&
5201 mval->br_startoff ==
5202 mval[-1].br_startoff + mval[-1].br_blockcount) {
5203 mval[-1].br_blockcount += mval->br_blockcount;
5204 mval[-1].br_state = mval->br_state;
5205 } else if (!((n == 0) &&
5206 ((mval->br_startoff + mval->br_blockcount) <=
5212 * If we're done, stop now. Stop when we've allocated
5213 * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise
5214 * the transaction may get too big.
5216 if (bno >= end || n >= *nmap || nallocs >= *nmap)
5219 * Else go on to the next record.
5221 ep = xfs_iext_get_ext(ifp, ++lastx);
5222 if (lastx >= nextents) {
5226 xfs_bmbt_get_all(ep, &got);
5228 ifp->if_lastex = lastx;
5231 * Transform from btree to extents, give it cur.
5233 if (tp && XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5234 XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5236 error = xfs_bmap_btree_to_extents(tp, ip, cur,
5237 &tmp_logflags, whichfork);
5238 logflags |= tmp_logflags;
5242 ASSERT(ifp->if_ext_max ==
5243 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5244 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
5245 XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max);
5247 if (delta && delta->xed_startoff != NULLFILEOFF) {
5248 /* A change was actually made.
5249 * Note that delta->xed_blockount is an offset at this
5250 * point and needs to be converted to a block count.
5252 ASSERT(delta->xed_blockcount > delta->xed_startoff);
5253 delta->xed_blockcount -= delta->xed_startoff;
5257 * Log everything. Do this after conversion, there's no point in
5258 * logging the extent records if we've converted to btree format.
5260 if ((logflags & XFS_ILOG_FEXT(whichfork)) &&
5261 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5262 logflags &= ~XFS_ILOG_FEXT(whichfork);
5263 else if ((logflags & XFS_ILOG_FBROOT(whichfork)) &&
5264 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5265 logflags &= ~XFS_ILOG_FBROOT(whichfork);
5267 * Log whatever the flags say, even if error. Otherwise we might miss
5268 * detecting a case where the data is changed, there's an error,
5269 * and it's not logged so we don't shutdown when we should.
5273 xfs_trans_log_inode(tp, ip, logflags);
5277 ASSERT(*firstblock == NULLFSBLOCK ||
5278 XFS_FSB_TO_AGNO(mp, *firstblock) ==
5280 cur->bc_private.b.firstblock) ||
5282 XFS_FSB_TO_AGNO(mp, *firstblock) <
5284 cur->bc_private.b.firstblock)));
5285 *firstblock = cur->bc_private.b.firstblock;
5287 xfs_btree_del_cursor(cur,
5288 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5291 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
5297 * Map file blocks to filesystem blocks, simple version.
5298 * One block (extent) only, read-only.
5299 * For flags, only the XFS_BMAPI_ATTRFORK flag is examined.
5300 * For the other flag values, the effect is as if XFS_BMAPI_METADATA
5301 * was set and all the others were clear.
5305 xfs_trans_t *tp, /* transaction pointer */
5306 xfs_inode_t *ip, /* incore inode */
5307 int whichfork, /* data or attr fork */
5308 xfs_fsblock_t *fsb, /* output: mapped block */
5309 xfs_fileoff_t bno) /* starting file offs. mapped */
5311 int eof; /* we've hit the end of extents */
5312 int error; /* error return */
5313 xfs_bmbt_irec_t got; /* current file extent record */
5314 xfs_ifork_t *ifp; /* inode fork pointer */
5315 xfs_extnum_t lastx; /* last useful extent number */
5316 xfs_bmbt_irec_t prev; /* previous file extent record */
5318 ifp = XFS_IFORK_PTR(ip, whichfork);
5320 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
5321 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)) {
5322 XFS_ERROR_REPORT("xfs_bmapi_single", XFS_ERRLEVEL_LOW,
5324 return XFS_ERROR(EFSCORRUPTED);
5326 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
5327 return XFS_ERROR(EIO);
5328 XFS_STATS_INC(xs_blk_mapr);
5329 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5330 (error = xfs_iread_extents(tp, ip, whichfork)))
5332 (void)xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5335 * Reading past eof, act as though there's a hole
5338 if (eof || got.br_startoff > bno) {
5342 ASSERT(!ISNULLSTARTBLOCK(got.br_startblock));
5343 ASSERT(bno < got.br_startoff + got.br_blockcount);
5344 *fsb = got.br_startblock + (bno - got.br_startoff);
5345 ifp->if_lastex = lastx;
5350 * Unmap (remove) blocks from a file.
5351 * If nexts is nonzero then the number of extents to remove is limited to
5352 * that value. If not all extents in the block range can be removed then
5357 xfs_trans_t *tp, /* transaction pointer */
5358 struct xfs_inode *ip, /* incore inode */
5359 xfs_fileoff_t bno, /* starting offset to unmap */
5360 xfs_filblks_t len, /* length to unmap in file */
5361 int flags, /* misc flags */
5362 xfs_extnum_t nexts, /* number of extents max */
5363 xfs_fsblock_t *firstblock, /* first allocated block
5364 controls a.g. for allocs */
5365 xfs_bmap_free_t *flist, /* i/o: list extents to free */
5366 xfs_extdelta_t *delta, /* o: change made to incore
5368 int *done) /* set if not done yet */
5370 xfs_btree_cur_t *cur; /* bmap btree cursor */
5371 xfs_bmbt_irec_t del; /* extent being deleted */
5372 int eof; /* is deleting at eof */
5373 xfs_bmbt_rec_t *ep; /* extent record pointer */
5374 int error; /* error return value */
5375 xfs_extnum_t extno; /* extent number in list */
5376 xfs_bmbt_irec_t got; /* current extent record */
5377 xfs_ifork_t *ifp; /* inode fork pointer */
5378 int isrt; /* freeing in rt area */
5379 xfs_extnum_t lastx; /* last extent index used */
5380 int logflags; /* transaction logging flags */
5381 xfs_extlen_t mod; /* rt extent offset */
5382 xfs_mount_t *mp; /* mount structure */
5383 xfs_extnum_t nextents; /* number of file extents */
5384 xfs_bmbt_irec_t prev; /* previous extent record */
5385 xfs_fileoff_t start; /* first file offset deleted */
5386 int tmp_logflags; /* partial logging flags */
5387 int wasdel; /* was a delayed alloc extent */
5388 int whichfork; /* data or attribute fork */
5389 int rsvd; /* OK to allocate reserved blocks */
5392 xfs_bunmap_trace(ip, bno, len, flags, (inst_t *)__return_address);
5393 whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5394 XFS_ATTR_FORK : XFS_DATA_FORK;
5395 ifp = XFS_IFORK_PTR(ip, whichfork);
5397 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5398 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5399 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5401 return XFS_ERROR(EFSCORRUPTED);
5404 if (XFS_FORCED_SHUTDOWN(mp))
5405 return XFS_ERROR(EIO);
5406 rsvd = (flags & XFS_BMAPI_RSVBLOCKS) != 0;
5409 ASSERT(ifp->if_ext_max ==
5410 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5411 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5412 (error = xfs_iread_extents(tp, ip, whichfork)))
5414 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5415 if (nextents == 0) {
5419 XFS_STATS_INC(xs_blk_unmap);
5420 isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5422 bno = start + len - 1;
5423 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5426 delta->xed_startoff = NULLFILEOFF;
5427 delta->xed_blockcount = 0;
5430 * Check to see if the given block number is past the end of the
5431 * file, back up to the last block if so...
5434 ep = xfs_iext_get_ext(ifp, --lastx);
5435 xfs_bmbt_get_all(ep, &got);
5436 bno = got.br_startoff + got.br_blockcount - 1;
5439 if (ifp->if_flags & XFS_IFBROOT) {
5440 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5441 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
5443 cur->bc_private.b.firstblock = *firstblock;
5444 cur->bc_private.b.flist = flist;
5445 cur->bc_private.b.flags = 0;
5449 while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5450 (nexts == 0 || extno < nexts)) {
5452 * Is the found extent after a hole in which bno lives?
5453 * Just back up to the previous extent, if so.
5455 if (got.br_startoff > bno) {
5458 ep = xfs_iext_get_ext(ifp, lastx);
5459 xfs_bmbt_get_all(ep, &got);
5462 * Is the last block of this extent before the range
5463 * we're supposed to delete? If so, we're done.
5465 bno = XFS_FILEOFF_MIN(bno,
5466 got.br_startoff + got.br_blockcount - 1);
5470 * Then deal with the (possibly delayed) allocated space
5475 wasdel = ISNULLSTARTBLOCK(del.br_startblock);
5476 if (got.br_startoff < start) {
5477 del.br_startoff = start;
5478 del.br_blockcount -= start - got.br_startoff;
5480 del.br_startblock += start - got.br_startoff;
5482 if (del.br_startoff + del.br_blockcount > bno + 1)
5483 del.br_blockcount = bno + 1 - del.br_startoff;
5484 sum = del.br_startblock + del.br_blockcount;
5486 (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5488 * Realtime extent not lined up at the end.
5489 * The extent could have been split into written
5490 * and unwritten pieces, or we could just be
5491 * unmapping part of it. But we can't really
5492 * get rid of part of a realtime extent.
5494 if (del.br_state == XFS_EXT_UNWRITTEN ||
5495 !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
5497 * This piece is unwritten, or we're not
5498 * using unwritten extents. Skip over it.
5501 bno -= mod > del.br_blockcount ?
5502 del.br_blockcount : mod;
5503 if (bno < got.br_startoff) {
5505 xfs_bmbt_get_all(xfs_iext_get_ext(
5511 * It's written, turn it unwritten.
5512 * This is better than zeroing it.
5514 ASSERT(del.br_state == XFS_EXT_NORM);
5515 ASSERT(xfs_trans_get_block_res(tp) > 0);
5517 * If this spans a realtime extent boundary,
5518 * chop it back to the start of the one we end at.
5520 if (del.br_blockcount > mod) {
5521 del.br_startoff += del.br_blockcount - mod;
5522 del.br_startblock += del.br_blockcount - mod;
5523 del.br_blockcount = mod;
5525 del.br_state = XFS_EXT_UNWRITTEN;
5526 error = xfs_bmap_add_extent(ip, lastx, &cur, &del,
5527 firstblock, flist, &logflags, delta,
5533 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5535 * Realtime extent is lined up at the end but not
5536 * at the front. We'll get rid of full extents if
5539 mod = mp->m_sb.sb_rextsize - mod;
5540 if (del.br_blockcount > mod) {
5541 del.br_blockcount -= mod;
5542 del.br_startoff += mod;
5543 del.br_startblock += mod;
5544 } else if ((del.br_startoff == start &&
5545 (del.br_state == XFS_EXT_UNWRITTEN ||
5546 xfs_trans_get_block_res(tp) == 0)) ||
5547 !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
5549 * Can't make it unwritten. There isn't
5550 * a full extent here so just skip it.
5552 ASSERT(bno >= del.br_blockcount);
5553 bno -= del.br_blockcount;
5554 if (bno < got.br_startoff) {
5556 xfs_bmbt_get_all(--ep, &got);
5559 } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5561 * This one is already unwritten.
5562 * It must have a written left neighbor.
5563 * Unwrite the killed part of that one and
5567 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5569 ASSERT(prev.br_state == XFS_EXT_NORM);
5570 ASSERT(!ISNULLSTARTBLOCK(prev.br_startblock));
5571 ASSERT(del.br_startblock ==
5572 prev.br_startblock + prev.br_blockcount);
5573 if (prev.br_startoff < start) {
5574 mod = start - prev.br_startoff;
5575 prev.br_blockcount -= mod;
5576 prev.br_startblock += mod;
5577 prev.br_startoff = start;
5579 prev.br_state = XFS_EXT_UNWRITTEN;
5580 error = xfs_bmap_add_extent(ip, lastx - 1, &cur,
5581 &prev, firstblock, flist, &logflags,
5582 delta, XFS_DATA_FORK, 0);
5587 ASSERT(del.br_state == XFS_EXT_NORM);
5588 del.br_state = XFS_EXT_UNWRITTEN;
5589 error = xfs_bmap_add_extent(ip, lastx, &cur,
5590 &del, firstblock, flist, &logflags,
5591 delta, XFS_DATA_FORK, 0);
5598 ASSERT(STARTBLOCKVAL(del.br_startblock) > 0);
5599 /* Update realtime/data freespace, unreserve quota */
5601 xfs_filblks_t rtexts;
5603 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5604 do_div(rtexts, mp->m_sb.sb_rextsize);
5605 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5606 (int64_t)rtexts, rsvd);
5607 (void)XFS_TRANS_RESERVE_QUOTA_NBLKS(mp,
5608 NULL, ip, -((long)del.br_blockcount), 0,
5609 XFS_QMOPT_RES_RTBLKS);
5611 xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS,
5612 (int64_t)del.br_blockcount, rsvd);
5613 (void)XFS_TRANS_RESERVE_QUOTA_NBLKS(mp,
5614 NULL, ip, -((long)del.br_blockcount), 0,
5615 XFS_QMOPT_RES_REGBLKS);
5617 ip->i_delayed_blks -= del.br_blockcount;
5619 cur->bc_private.b.flags |=
5620 XFS_BTCUR_BPRV_WASDEL;
5622 cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5624 * If it's the case where the directory code is running
5625 * with no block reservation, and the deleted block is in
5626 * the middle of its extent, and the resulting insert
5627 * of an extent would cause transformation to btree format,
5628 * then reject it. The calling code will then swap
5629 * blocks around instead.
5630 * We have to do this now, rather than waiting for the
5631 * conversion to btree format, since the transaction
5634 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5635 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5636 XFS_IFORK_NEXTENTS(ip, whichfork) >= ifp->if_ext_max &&
5637 del.br_startoff > got.br_startoff &&
5638 del.br_startoff + del.br_blockcount <
5639 got.br_startoff + got.br_blockcount) {
5640 error = XFS_ERROR(ENOSPC);
5643 error = xfs_bmap_del_extent(ip, tp, lastx, flist, cur, &del,
5644 &tmp_logflags, delta, whichfork, rsvd);
5645 logflags |= tmp_logflags;
5648 bno = del.br_startoff - 1;
5650 lastx = ifp->if_lastex;
5652 * If not done go on to the next (previous) record.
5653 * Reset ep in case the extents array was re-alloced.
5655 ep = xfs_iext_get_ext(ifp, lastx);
5656 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5657 if (lastx >= XFS_IFORK_NEXTENTS(ip, whichfork) ||
5658 xfs_bmbt_get_startoff(ep) > bno) {
5660 ep = xfs_iext_get_ext(ifp, lastx);
5663 xfs_bmbt_get_all(ep, &got);
5667 ifp->if_lastex = lastx;
5668 *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5669 ASSERT(ifp->if_ext_max ==
5670 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5672 * Convert to a btree if necessary.
5674 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5675 XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
5676 ASSERT(cur == NULL);
5677 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5678 &cur, 0, &tmp_logflags, whichfork);
5679 logflags |= tmp_logflags;
5684 * transform from btree to extents, give it cur
5686 else if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5687 XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5688 ASSERT(cur != NULL);
5689 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5691 logflags |= tmp_logflags;
5696 * transform from extents to local?
5698 ASSERT(ifp->if_ext_max ==
5699 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5701 if (delta && delta->xed_startoff != NULLFILEOFF) {
5702 /* A change was actually made.
5703 * Note that delta->xed_blockount is an offset at this
5704 * point and needs to be converted to a block count.
5706 ASSERT(delta->xed_blockcount > delta->xed_startoff);
5707 delta->xed_blockcount -= delta->xed_startoff;
5711 * Log everything. Do this after conversion, there's no point in
5712 * logging the extent records if we've converted to btree format.
5714 if ((logflags & XFS_ILOG_FEXT(whichfork)) &&
5715 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5716 logflags &= ~XFS_ILOG_FEXT(whichfork);
5717 else if ((logflags & XFS_ILOG_FBROOT(whichfork)) &&
5718 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5719 logflags &= ~XFS_ILOG_FBROOT(whichfork);
5721 * Log inode even in the error case, if the transaction
5722 * is dirty we'll need to shut down the filesystem.
5725 xfs_trans_log_inode(tp, ip, logflags);
5728 *firstblock = cur->bc_private.b.firstblock;
5729 cur->bc_private.b.allocated = 0;
5731 xfs_btree_del_cursor(cur,
5732 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5738 * Fcntl interface to xfs_bmapi.
5740 int /* error code */
5742 bhv_desc_t *bdp, /* XFS behavior descriptor*/
5743 struct getbmap *bmv, /* user bmap structure */
5744 void __user *ap, /* pointer to user's array */
5745 int interface) /* interface flags */
5747 __int64_t bmvend; /* last block requested */
5748 int error; /* return value */
5749 __int64_t fixlen; /* length for -1 case */
5750 int i; /* extent number */
5751 xfs_inode_t *ip; /* xfs incore inode pointer */
5752 bhv_vnode_t *vp; /* corresponding vnode */
5753 int lock; /* lock state */
5754 xfs_bmbt_irec_t *map; /* buffer for user's data */
5755 xfs_mount_t *mp; /* file system mount point */
5756 int nex; /* # of user extents can do */
5757 int nexleft; /* # of user extents left */
5758 int subnex; /* # of bmapi's can do */
5759 int nmap; /* number of map entries */
5760 struct getbmap out; /* output structure */
5761 int whichfork; /* data or attr fork */
5762 int prealloced; /* this is a file with
5763 * preallocated data space */
5764 int sh_unwritten; /* true, if unwritten */
5765 /* extents listed separately */
5766 int bmapi_flags; /* flags for xfs_bmapi */
5767 __int32_t oflags; /* getbmapx bmv_oflags field */
5769 vp = BHV_TO_VNODE(bdp);
5770 ip = XFS_BHVTOI(bdp);
5773 whichfork = interface & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5774 sh_unwritten = (interface & BMV_IF_PREALLOC) != 0;
5776 /* If the BMV_IF_NO_DMAPI_READ interface bit specified, do not
5777 * generate a DMAPI read event. Otherwise, if the DM_EVENT_READ
5778 * bit is set for the file, generate a read event in order
5779 * that the DMAPI application may do its thing before we return
5780 * the extents. Usually this means restoring user file data to
5781 * regions of the file that look like holes.
5783 * The "old behavior" (from XFS_IOC_GETBMAP) is to not specify
5784 * BMV_IF_NO_DMAPI_READ so that read events are generated.
5785 * If this were not true, callers of ioctl( XFS_IOC_GETBMAP )
5786 * could misinterpret holes in a DMAPI file as true holes,
5787 * when in fact they may represent offline user data.
5789 if ( (interface & BMV_IF_NO_DMAPI_READ) == 0
5790 && DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_READ)
5791 && whichfork == XFS_DATA_FORK) {
5793 error = XFS_SEND_DATA(mp, DM_EVENT_READ, vp, 0, 0, 0, NULL);
5795 return XFS_ERROR(error);
5798 if (whichfork == XFS_ATTR_FORK) {
5799 if (XFS_IFORK_Q(ip)) {
5800 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5801 ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5802 ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5803 return XFS_ERROR(EINVAL);
5804 } else if (unlikely(
5805 ip->i_d.di_aformat != 0 &&
5806 ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5807 XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5809 return XFS_ERROR(EFSCORRUPTED);
5811 } else if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5812 ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5813 ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5814 return XFS_ERROR(EINVAL);
5815 if (whichfork == XFS_DATA_FORK) {
5816 if ((ip->i_d.di_extsize && (ip->i_d.di_flags &
5817 (XFS_DIFLAG_REALTIME|XFS_DIFLAG_EXTSIZE))) ||
5818 ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5820 fixlen = XFS_MAXIOFFSET(mp);
5823 fixlen = ip->i_d.di_size;
5830 if (bmv->bmv_length == -1) {
5831 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5832 bmv->bmv_length = MAX( (__int64_t)(fixlen - bmv->bmv_offset),
5834 } else if (bmv->bmv_length < 0)
5835 return XFS_ERROR(EINVAL);
5836 if (bmv->bmv_length == 0) {
5837 bmv->bmv_entries = 0;
5840 nex = bmv->bmv_count - 1;
5842 return XFS_ERROR(EINVAL);
5843 bmvend = bmv->bmv_offset + bmv->bmv_length;
5845 xfs_ilock(ip, XFS_IOLOCK_SHARED);
5847 if (whichfork == XFS_DATA_FORK && ip->i_delayed_blks) {
5848 /* xfs_fsize_t last_byte = xfs_file_last_byte(ip); */
5849 error = bhv_vop_flush_pages(vp, (xfs_off_t)0, -1, 0, FI_REMAPF);
5852 ASSERT(whichfork == XFS_ATTR_FORK || ip->i_delayed_blks == 0);
5854 lock = xfs_ilock_map_shared(ip);
5857 * Don't let nex be bigger than the number of extents
5858 * we can have assuming alternating holes and real extents.
5860 if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5861 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5863 bmapi_flags = XFS_BMAPI_AFLAG(whichfork) |
5864 ((sh_unwritten) ? 0 : XFS_BMAPI_IGSTATE);
5867 * Allocate enough space to handle "subnex" maps at a time.
5870 map = kmem_alloc(subnex * sizeof(*map), KM_SLEEP);
5872 bmv->bmv_entries = 0;
5874 if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0) {
5876 goto unlock_and_return;
5882 nmap = (nexleft > subnex) ? subnex : nexleft;
5883 error = xfs_bmapi(NULL, ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5884 XFS_BB_TO_FSB(mp, bmv->bmv_length),
5885 bmapi_flags, NULL, 0, map, &nmap,
5888 goto unlock_and_return;
5889 ASSERT(nmap <= subnex);
5891 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5893 oflags = (map[i].br_state == XFS_EXT_UNWRITTEN) ?
5894 BMV_OF_PREALLOC : 0;
5895 out.bmv_offset = XFS_FSB_TO_BB(mp, map[i].br_startoff);
5896 out.bmv_length = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5897 ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
5898 if (map[i].br_startblock == HOLESTARTBLOCK &&
5899 ((prealloced && out.bmv_offset + out.bmv_length == bmvend) ||
5900 whichfork == XFS_ATTR_FORK )) {
5902 * came to hole at end of file or the end of
5905 goto unlock_and_return;
5908 (map[i].br_startblock == HOLESTARTBLOCK) ?
5910 XFS_FSB_TO_DB(ip, map[i].br_startblock);
5912 /* return either getbmap/getbmapx structure. */
5913 if (interface & BMV_IF_EXTENDED) {
5914 struct getbmapx outx;
5916 GETBMAP_CONVERT(out,outx);
5917 outx.bmv_oflags = oflags;
5918 outx.bmv_unused1 = outx.bmv_unused2 = 0;
5919 if (copy_to_user(ap, &outx,
5921 error = XFS_ERROR(EFAULT);
5922 goto unlock_and_return;
5925 if (copy_to_user(ap, &out,
5927 error = XFS_ERROR(EFAULT);
5928 goto unlock_and_return;
5932 out.bmv_offset + out.bmv_length;
5933 bmv->bmv_length = MAX((__int64_t)0,
5934 (__int64_t)(bmvend - bmv->bmv_offset));
5936 ap = (interface & BMV_IF_EXTENDED) ?
5938 ((struct getbmapx __user *)ap + 1) :
5940 ((struct getbmap __user *)ap + 1);
5943 } while (nmap && nexleft && bmv->bmv_length);
5946 xfs_iunlock_map_shared(ip, lock);
5947 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5949 kmem_free(map, subnex * sizeof(*map));
5955 * Check the last inode extent to determine whether this allocation will result
5956 * in blocks being allocated at the end of the file. When we allocate new data
5957 * blocks at the end of the file which do not start at the previous data block,
5958 * we will try to align the new blocks at stripe unit boundaries.
5960 STATIC int /* error */
5962 xfs_inode_t *ip, /* incore inode pointer */
5963 xfs_fileoff_t off, /* file offset in fsblocks */
5964 int whichfork, /* data or attribute fork */
5965 char *aeof) /* return value */
5967 int error; /* error return value */
5968 xfs_ifork_t *ifp; /* inode fork pointer */
5969 xfs_bmbt_rec_t *lastrec; /* extent record pointer */
5970 xfs_extnum_t nextents; /* number of file extents */
5971 xfs_bmbt_irec_t s; /* expanded extent record */
5973 ASSERT(whichfork == XFS_DATA_FORK);
5974 ifp = XFS_IFORK_PTR(ip, whichfork);
5975 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5976 (error = xfs_iread_extents(NULL, ip, whichfork)))
5978 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5979 if (nextents == 0) {
5984 * Go to the last extent
5986 lastrec = xfs_iext_get_ext(ifp, nextents - 1);
5987 xfs_bmbt_get_all(lastrec, &s);
5989 * Check we are allocating in the last extent (for delayed allocations)
5990 * or past the last extent for non-delayed allocations.
5992 *aeof = (off >= s.br_startoff &&
5993 off < s.br_startoff + s.br_blockcount &&
5994 ISNULLSTARTBLOCK(s.br_startblock)) ||
5995 off >= s.br_startoff + s.br_blockcount;
6000 * Check if the endoff is outside the last extent. If so the caller will grow
6001 * the allocation to a stripe unit boundary.
6005 xfs_inode_t *ip, /* incore inode pointer */
6006 xfs_fileoff_t endoff, /* file offset in fsblocks */
6007 int whichfork, /* data or attribute fork */
6008 int *eof) /* result value */
6010 xfs_fsblock_t blockcount; /* extent block count */
6011 int error; /* error return value */
6012 xfs_ifork_t *ifp; /* inode fork pointer */
6013 xfs_bmbt_rec_t *lastrec; /* extent record pointer */
6014 xfs_extnum_t nextents; /* number of file extents */
6015 xfs_fileoff_t startoff; /* extent starting file offset */
6017 ASSERT(whichfork == XFS_DATA_FORK);
6018 ifp = XFS_IFORK_PTR(ip, whichfork);
6019 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
6020 (error = xfs_iread_extents(NULL, ip, whichfork)))
6022 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
6023 if (nextents == 0) {
6028 * Go to the last extent
6030 lastrec = xfs_iext_get_ext(ifp, nextents - 1);
6031 startoff = xfs_bmbt_get_startoff(lastrec);
6032 blockcount = xfs_bmbt_get_blockcount(lastrec);
6033 *eof = endoff >= startoff + blockcount;
6041 xfs_btree_cur_t *cur,
6051 for(i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
6052 bp = cur->bc_bufs[i];
6054 if (XFS_BUF_ADDR(bp) == bno)
6055 break; /* Found it */
6057 if (i == XFS_BTREE_MAXLEVELS)
6060 if (!bp) { /* Chase down all the log items to see if the bp is there */
6061 xfs_log_item_chunk_t *licp;
6065 licp = &tp->t_items;
6066 while (!bp && licp != NULL) {
6067 if (XFS_LIC_ARE_ALL_FREE(licp)) {
6068 licp = licp->lic_next;
6071 for (i = 0; i < licp->lic_unused; i++) {
6072 xfs_log_item_desc_t *lidp;
6073 xfs_log_item_t *lip;
6074 xfs_buf_log_item_t *bip;
6077 if (XFS_LIC_ISFREE(licp, i)) {
6081 lidp = XFS_LIC_SLOT(licp, i);
6082 lip = lidp->lid_item;
6083 if (lip->li_type != XFS_LI_BUF)
6086 bip = (xfs_buf_log_item_t *)lip;
6089 if (XFS_BUF_ADDR(lbp) == bno) {
6091 break; /* Found it */
6094 licp = licp->lic_next;
6102 xfs_bmbt_block_t *block,
6108 __be64 *pp, *thispa; /* pointer to block address */
6109 xfs_bmbt_key_t *prevp, *keyp;
6111 ASSERT(be16_to_cpu(block->bb_level) > 0);
6114 for( i = 1; i <= be16_to_cpu(block->bb_numrecs); i++) {
6115 dmxr = mp->m_bmap_dmxr[0];
6118 keyp = XFS_BMAP_BROOT_KEY_ADDR(block, i, sz);
6120 keyp = XFS_BTREE_KEY_ADDR(xfs_bmbt, block, i);
6124 xfs_btree_check_key(XFS_BTNUM_BMAP, prevp, keyp);
6129 * Compare the block numbers to see if there are dups.
6133 pp = XFS_BMAP_BROOT_PTR_ADDR(block, i, sz);
6135 pp = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, i, dmxr);
6137 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
6139 thispa = XFS_BMAP_BROOT_PTR_ADDR(block, j, sz);
6141 thispa = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, j,
6144 if (*thispa == *pp) {
6145 cmn_err(CE_WARN, "%s: thispa(%d) == pp(%d) %Ld",
6147 (unsigned long long)be64_to_cpu(*thispa));
6148 panic("%s: ptrs are equal in node\n",
6156 * Check that the extents for the inode ip are in the right order in all
6161 xfs_bmap_check_leaf_extents(
6162 xfs_btree_cur_t *cur, /* btree cursor or null */
6163 xfs_inode_t *ip, /* incore inode pointer */
6164 int whichfork) /* data or attr fork */
6166 xfs_bmbt_block_t *block; /* current btree block */
6167 xfs_fsblock_t bno; /* block # of "block" */
6168 xfs_buf_t *bp; /* buffer for "block" */
6169 int error; /* error return value */
6170 xfs_extnum_t i=0, j; /* index into the extents list */
6171 xfs_ifork_t *ifp; /* fork structure */
6172 int level; /* btree level, for checking */
6173 xfs_mount_t *mp; /* file system mount structure */
6174 __be64 *pp; /* pointer to block address */
6175 xfs_bmbt_rec_t *ep; /* pointer to current extent */
6176 xfs_bmbt_rec_t *lastp; /* pointer to previous extent */
6177 xfs_bmbt_rec_t *nextp; /* pointer to next extent */
6180 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
6186 ifp = XFS_IFORK_PTR(ip, whichfork);
6187 block = ifp->if_broot;
6189 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6191 level = be16_to_cpu(block->bb_level);
6193 xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
6194 pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
6195 bno = be64_to_cpu(*pp);
6197 ASSERT(bno != NULLDFSBNO);
6198 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
6199 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
6202 * Go down the tree until leaf level is reached, following the first
6203 * pointer (leftmost) at each level.
6205 while (level-- > 0) {
6206 /* See if buf is in cur first */
6207 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
6213 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
6214 XFS_BMAP_BTREE_REF)))
6216 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6217 XFS_WANT_CORRUPTED_GOTO(
6218 XFS_BMAP_SANITY_CHECK(mp, block, level),
6224 * Check this block for basic sanity (increasing keys and
6225 * no duplicate blocks).
6228 xfs_check_block(block, mp, 0, 0);
6229 pp = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, 1, mp->m_bmap_dmxr[1]);
6230 bno = be64_to_cpu(*pp);
6231 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
6234 xfs_trans_brelse(NULL, bp);
6239 * Here with bp and block set to the leftmost leaf node in the tree.
6244 * Loop over all leaf nodes checking that all extents are in the right order.
6248 xfs_fsblock_t nextbno;
6249 xfs_extnum_t num_recs;
6252 num_recs = be16_to_cpu(block->bb_numrecs);
6255 * Read-ahead the next leaf block, if any.
6258 nextbno = be64_to_cpu(block->bb_rightsib);
6261 * Check all the extents to make sure they are OK.
6262 * If we had a previous block, the last entry should
6263 * conform with the first entry in this one.
6266 ep = XFS_BTREE_REC_ADDR(xfs_bmbt, block, 1);
6267 for (j = 1; j < num_recs; j++) {
6268 nextp = XFS_BTREE_REC_ADDR(xfs_bmbt, block, j + 1);
6270 xfs_btree_check_rec(XFS_BTNUM_BMAP,
6271 (void *)lastp, (void *)ep);
6273 xfs_btree_check_rec(XFS_BTNUM_BMAP, (void *)ep,
6282 xfs_trans_brelse(NULL, bp);
6286 * If we've reached the end, stop.
6288 if (bno == NULLFSBLOCK)
6291 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
6297 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
6298 XFS_BMAP_BTREE_REF)))
6300 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6304 xfs_trans_brelse(NULL, bp);
6309 cmn_err(CE_WARN, "%s: at error0", __FUNCTION__);
6311 xfs_trans_brelse(NULL, bp);
6313 cmn_err(CE_WARN, "%s: BAD after btree leaves for %d extents",
6315 panic("%s: CORRUPTED BTREE OR SOMETHING", __FUNCTION__);
6321 * Count fsblocks of the given fork.
6324 xfs_bmap_count_blocks(
6325 xfs_trans_t *tp, /* transaction pointer */
6326 xfs_inode_t *ip, /* incore inode */
6327 int whichfork, /* data or attr fork */
6328 int *count) /* out: count of blocks */
6330 xfs_bmbt_block_t *block; /* current btree block */
6331 xfs_fsblock_t bno; /* block # of "block" */
6332 xfs_ifork_t *ifp; /* fork structure */
6333 int level; /* btree level, for checking */
6334 xfs_mount_t *mp; /* file system mount structure */
6335 __be64 *pp; /* pointer to block address */
6339 ifp = XFS_IFORK_PTR(ip, whichfork);
6340 if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
6341 if (unlikely(xfs_bmap_count_leaves(ifp, 0,
6342 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
6344 XFS_ERROR_REPORT("xfs_bmap_count_blocks(1)",
6345 XFS_ERRLEVEL_LOW, mp);
6346 return XFS_ERROR(EFSCORRUPTED);
6352 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6354 block = ifp->if_broot;
6355 level = be16_to_cpu(block->bb_level);
6357 pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
6358 bno = be64_to_cpu(*pp);
6359 ASSERT(bno != NULLDFSBNO);
6360 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
6361 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
6363 if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
6364 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
6366 return XFS_ERROR(EFSCORRUPTED);
6373 * Recursively walks each level of a btree
6374 * to count total fsblocks is use.
6377 xfs_bmap_count_tree(
6378 xfs_mount_t *mp, /* file system mount point */
6379 xfs_trans_t *tp, /* transaction pointer */
6380 xfs_ifork_t *ifp, /* inode fork pointer */
6381 xfs_fsblock_t blockno, /* file system block number */
6382 int levelin, /* level in btree */
6383 int *count) /* Count of blocks */
6386 xfs_buf_t *bp, *nbp;
6387 int level = levelin;
6389 xfs_fsblock_t bno = blockno;
6390 xfs_fsblock_t nextbno;
6391 xfs_bmbt_block_t *block, *nextblock;
6394 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
6397 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6400 /* Not at node above leafs, count this level of nodes */
6401 nextbno = be64_to_cpu(block->bb_rightsib);
6402 while (nextbno != NULLFSBLOCK) {
6403 if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
6404 0, &nbp, XFS_BMAP_BTREE_REF)))
6407 nextblock = XFS_BUF_TO_BMBT_BLOCK(nbp);
6408 nextbno = be64_to_cpu(nextblock->bb_rightsib);
6409 xfs_trans_brelse(tp, nbp);
6412 /* Dive to the next level */
6413 pp = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, 1, mp->m_bmap_dmxr[1]);
6414 bno = be64_to_cpu(*pp);
6415 if (unlikely((error =
6416 xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
6417 xfs_trans_brelse(tp, bp);
6418 XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6419 XFS_ERRLEVEL_LOW, mp);
6420 return XFS_ERROR(EFSCORRUPTED);
6422 xfs_trans_brelse(tp, bp);
6424 /* count all level 1 nodes and their leaves */
6426 nextbno = be64_to_cpu(block->bb_rightsib);
6427 numrecs = be16_to_cpu(block->bb_numrecs);
6428 if (unlikely(xfs_bmap_disk_count_leaves(ifp,
6429 0, block, numrecs, count) < 0)) {
6430 xfs_trans_brelse(tp, bp);
6431 XFS_ERROR_REPORT("xfs_bmap_count_tree(2)",
6432 XFS_ERRLEVEL_LOW, mp);
6433 return XFS_ERROR(EFSCORRUPTED);
6435 xfs_trans_brelse(tp, bp);
6436 if (nextbno == NULLFSBLOCK)
6439 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6440 XFS_BMAP_BTREE_REF)))
6443 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6450 * Count leaf blocks given a range of extent records.
6453 xfs_bmap_count_leaves(
6460 xfs_bmbt_rec_t *frp;
6462 for (b = 0; b < numrecs; b++) {
6463 frp = xfs_iext_get_ext(ifp, idx + b);
6464 *count += xfs_bmbt_get_blockcount(frp);
6470 * Count leaf blocks given a range of extent records originally
6474 xfs_bmap_disk_count_leaves(
6477 xfs_bmbt_block_t *block,
6482 xfs_bmbt_rec_t *frp;
6484 for (b = 1; b <= numrecs; b++) {
6485 frp = XFS_BTREE_REC_ADDR(xfs_bmbt, block, idx + b);
6486 *count += xfs_bmbt_disk_get_blockcount(frp);
projects / linux-drm-fsl-dcu.git / blob