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_dinode.h"
33 #include "xfs_inode.h"
34 #include "xfs_btree.h"
35 #include "xfs_mount.h"
36 #include "xfs_itable.h"
37 #include "xfs_inode_item.h"
38 #include "xfs_extfree_item.h"
39 #include "xfs_alloc.h"
41 #include "xfs_rtalloc.h"
42 #include "xfs_error.h"
43 #include "xfs_attr_leaf.h"
45 #include "xfs_quota.h"
46 #include "xfs_trans_space.h"
47 #include "xfs_buf_item.h"
48 #include "xfs_filestream.h"
49 #include "xfs_vnodeops.h"
50 #include "xfs_trace.h"
55 xfs_bmap_check_leaf_extents(xfs_btree_cur_t *cur, xfs_inode_t *ip, int whichfork);
58 kmem_zone_t *xfs_bmap_free_item_zone;
61 * Prototypes for internal bmap routines.
66 * Called from xfs_bmap_add_attrfork to handle extents format files.
68 STATIC int /* error */
69 xfs_bmap_add_attrfork_extents(
70 xfs_trans_t *tp, /* transaction pointer */
71 xfs_inode_t *ip, /* incore inode pointer */
72 xfs_fsblock_t *firstblock, /* first block allocated */
73 xfs_bmap_free_t *flist, /* blocks to free at commit */
74 int *flags); /* inode logging flags */
77 * Called from xfs_bmap_add_attrfork to handle local format files.
79 STATIC int /* error */
80 xfs_bmap_add_attrfork_local(
81 xfs_trans_t *tp, /* transaction pointer */
82 xfs_inode_t *ip, /* incore inode pointer */
83 xfs_fsblock_t *firstblock, /* first block allocated */
84 xfs_bmap_free_t *flist, /* blocks to free at commit */
85 int *flags); /* inode logging flags */
88 * Called by xfs_bmap_add_extent to handle cases converting a delayed
89 * allocation to a real allocation.
91 STATIC int /* error */
92 xfs_bmap_add_extent_delay_real(
93 struct xfs_trans *tp, /* transaction pointer */
94 xfs_inode_t *ip, /* incore inode pointer */
95 xfs_extnum_t *idx, /* extent number to update/insert */
96 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
97 xfs_bmbt_irec_t *new, /* new data to add to file extents */
98 xfs_filblks_t *dnew, /* new delayed-alloc indirect blocks */
99 xfs_fsblock_t *first, /* pointer to firstblock variable */
100 xfs_bmap_free_t *flist, /* list of extents to be freed */
101 int *logflagsp); /* inode logging flags */
104 * Called by xfs_bmap_add_extent to handle cases converting a hole
105 * to a delayed allocation.
107 STATIC int /* error */
108 xfs_bmap_add_extent_hole_delay(
109 xfs_inode_t *ip, /* incore inode pointer */
110 xfs_extnum_t *idx, /* extent number to update/insert */
111 xfs_bmbt_irec_t *new, /* new data to add to file extents */
112 int *logflagsp); /* inode logging flags */
115 * Called by xfs_bmap_add_extent to handle cases converting a hole
116 * to a real allocation.
118 STATIC int /* error */
119 xfs_bmap_add_extent_hole_real(
120 xfs_inode_t *ip, /* incore inode pointer */
121 xfs_extnum_t *idx, /* extent number to update/insert */
122 xfs_btree_cur_t *cur, /* if null, not a btree */
123 xfs_bmbt_irec_t *new, /* new data to add to file extents */
124 int *logflagsp, /* inode logging flags */
125 int whichfork); /* data or attr fork */
128 * Called by xfs_bmap_add_extent to handle cases converting an unwritten
129 * allocation to a real allocation or vice versa.
131 STATIC int /* error */
132 xfs_bmap_add_extent_unwritten_real(
133 xfs_inode_t *ip, /* incore inode pointer */
134 xfs_extnum_t *idx, /* extent number to update/insert */
135 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
136 xfs_bmbt_irec_t *new, /* new data to add to file extents */
137 int *logflagsp); /* inode logging flags */
140 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
141 * It figures out where to ask the underlying allocator to put the new extent.
143 STATIC int /* error */
145 xfs_bmalloca_t *ap); /* bmap alloc argument struct */
148 * Transform a btree format file with only one leaf node, where the
149 * extents list will fit in the inode, into an extents format file.
150 * Since the file extents are already in-core, all we have to do is
151 * give up the space for the btree root and pitch the leaf block.
153 STATIC int /* error */
154 xfs_bmap_btree_to_extents(
155 xfs_trans_t *tp, /* transaction pointer */
156 xfs_inode_t *ip, /* incore inode pointer */
157 xfs_btree_cur_t *cur, /* btree cursor */
158 int *logflagsp, /* inode logging flags */
159 int whichfork); /* data or attr fork */
162 * Remove the entry "free" from the free item list. Prev points to the
163 * previous entry, unless "free" is the head of the list.
167 xfs_bmap_free_t *flist, /* free item list header */
168 xfs_bmap_free_item_t *prev, /* previous item on list, if any */
169 xfs_bmap_free_item_t *free); /* list item to be freed */
172 * Convert an extents-format file into a btree-format file.
173 * The new file will have a root block (in the inode) and a single child block.
175 STATIC int /* error */
176 xfs_bmap_extents_to_btree(
177 xfs_trans_t *tp, /* transaction pointer */
178 xfs_inode_t *ip, /* incore inode pointer */
179 xfs_fsblock_t *firstblock, /* first-block-allocated */
180 xfs_bmap_free_t *flist, /* blocks freed in xaction */
181 xfs_btree_cur_t **curp, /* cursor returned to caller */
182 int wasdel, /* converting a delayed alloc */
183 int *logflagsp, /* inode logging flags */
184 int whichfork); /* data or attr fork */
187 * Convert a local file to an extents file.
188 * This code is sort of bogus, since the file data needs to get
189 * logged so it won't be lost. The bmap-level manipulations are ok, though.
191 STATIC int /* error */
192 xfs_bmap_local_to_extents(
193 xfs_trans_t *tp, /* transaction pointer */
194 xfs_inode_t *ip, /* incore inode pointer */
195 xfs_fsblock_t *firstblock, /* first block allocated in xaction */
196 xfs_extlen_t total, /* total blocks needed by transaction */
197 int *logflagsp, /* inode logging flags */
198 int whichfork); /* data or attr fork */
201 * Search the extents list for the inode, for the extent containing bno.
202 * If bno lies in a hole, point to the next entry. If bno lies past eof,
203 * *eofp will be set, and *prevp will contain the last entry (null if none).
204 * Else, *lastxp will be set to the index of the found
205 * entry; *gotp will contain the entry.
207 STATIC xfs_bmbt_rec_host_t * /* pointer to found extent entry */
208 xfs_bmap_search_extents(
209 xfs_inode_t *ip, /* incore inode pointer */
210 xfs_fileoff_t bno, /* block number searched for */
211 int whichfork, /* data or attr fork */
212 int *eofp, /* out: end of file found */
213 xfs_extnum_t *lastxp, /* out: last extent index */
214 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
215 xfs_bmbt_irec_t *prevp); /* out: previous extent entry found */
218 * Check the last inode extent to determine whether this allocation will result
219 * in blocks being allocated at the end of the file. When we allocate new data
220 * blocks at the end of the file which do not start at the previous data block,
221 * we will try to align the new blocks at stripe unit boundaries.
223 STATIC int /* error */
225 xfs_inode_t *ip, /* incore inode pointer */
226 xfs_fileoff_t off, /* file offset in fsblocks */
227 int whichfork, /* data or attribute fork */
228 char *aeof); /* return value */
231 * Compute the worst-case number of indirect blocks that will be used
232 * for ip's delayed extent of length "len".
235 xfs_bmap_worst_indlen(
236 xfs_inode_t *ip, /* incore inode pointer */
237 xfs_filblks_t len); /* delayed extent length */
241 * Perform various validation checks on the values being returned
245 xfs_bmap_validate_ret(
249 xfs_bmbt_irec_t *mval,
253 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
261 xfs_fsblock_t blockno,
266 xfs_bmap_count_leaves(
273 xfs_bmap_disk_count_leaves(
274 struct xfs_mount *mp,
275 struct xfs_btree_block *block,
280 * Bmap internal routines.
283 STATIC int /* error */
285 struct xfs_btree_cur *cur,
289 int *stat) /* success/failure */
291 cur->bc_rec.b.br_startoff = off;
292 cur->bc_rec.b.br_startblock = bno;
293 cur->bc_rec.b.br_blockcount = len;
294 return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
297 STATIC int /* error */
299 struct xfs_btree_cur *cur,
303 int *stat) /* success/failure */
305 cur->bc_rec.b.br_startoff = off;
306 cur->bc_rec.b.br_startblock = bno;
307 cur->bc_rec.b.br_blockcount = len;
308 return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
312 * Update the record referred to by cur to the value given
313 * by [off, bno, len, state].
314 * This either works (return 0) or gets an EFSCORRUPTED error.
318 struct xfs_btree_cur *cur,
324 union xfs_btree_rec rec;
326 xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
327 return xfs_btree_update(cur, &rec);
331 * Called from xfs_bmap_add_attrfork to handle btree format files.
333 STATIC int /* error */
334 xfs_bmap_add_attrfork_btree(
335 xfs_trans_t *tp, /* transaction pointer */
336 xfs_inode_t *ip, /* incore inode pointer */
337 xfs_fsblock_t *firstblock, /* first block allocated */
338 xfs_bmap_free_t *flist, /* blocks to free at commit */
339 int *flags) /* inode logging flags */
341 xfs_btree_cur_t *cur; /* btree cursor */
342 int error; /* error return value */
343 xfs_mount_t *mp; /* file system mount struct */
344 int stat; /* newroot status */
347 if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
348 *flags |= XFS_ILOG_DBROOT;
350 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
351 cur->bc_private.b.flist = flist;
352 cur->bc_private.b.firstblock = *firstblock;
353 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
355 /* must be at least one entry */
356 XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
357 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
360 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
361 return XFS_ERROR(ENOSPC);
363 *firstblock = cur->bc_private.b.firstblock;
364 cur->bc_private.b.allocated = 0;
365 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
369 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
374 * Called from xfs_bmap_add_attrfork to handle extents format files.
376 STATIC int /* error */
377 xfs_bmap_add_attrfork_extents(
378 xfs_trans_t *tp, /* transaction pointer */
379 xfs_inode_t *ip, /* incore inode pointer */
380 xfs_fsblock_t *firstblock, /* first block allocated */
381 xfs_bmap_free_t *flist, /* blocks to free at commit */
382 int *flags) /* inode logging flags */
384 xfs_btree_cur_t *cur; /* bmap btree cursor */
385 int error; /* error return value */
387 if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
390 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
391 flags, XFS_DATA_FORK);
393 cur->bc_private.b.allocated = 0;
394 xfs_btree_del_cursor(cur,
395 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
401 * Called from xfs_bmap_add_attrfork to handle local format files.
403 STATIC int /* error */
404 xfs_bmap_add_attrfork_local(
405 xfs_trans_t *tp, /* transaction pointer */
406 xfs_inode_t *ip, /* incore inode pointer */
407 xfs_fsblock_t *firstblock, /* first block allocated */
408 xfs_bmap_free_t *flist, /* blocks to free at commit */
409 int *flags) /* inode logging flags */
411 xfs_da_args_t dargs; /* args for dir/attr code */
412 int error; /* error return value */
413 xfs_mount_t *mp; /* mount structure pointer */
415 if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
417 if (S_ISDIR(ip->i_d.di_mode)) {
419 memset(&dargs, 0, sizeof(dargs));
421 dargs.firstblock = firstblock;
423 dargs.total = mp->m_dirblkfsbs;
424 dargs.whichfork = XFS_DATA_FORK;
426 error = xfs_dir2_sf_to_block(&dargs);
428 error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
434 * Called by xfs_bmapi to update file extent records and the btree
435 * after allocating space (or doing a delayed allocation).
437 STATIC int /* error */
439 struct xfs_trans *tp, /* transaction pointer */
440 xfs_inode_t *ip, /* incore inode pointer */
441 xfs_extnum_t *idx, /* extent number to update/insert */
442 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
443 xfs_bmbt_irec_t *new, /* new data to add to file extents */
444 xfs_fsblock_t *first, /* pointer to firstblock variable */
445 xfs_bmap_free_t *flist, /* list of extents to be freed */
446 int *logflagsp, /* inode logging flags */
447 int whichfork) /* data or attr fork */
449 xfs_btree_cur_t *cur; /* btree cursor or null */
450 xfs_filblks_t da_new; /* new count del alloc blocks used */
451 xfs_filblks_t da_old; /* old count del alloc blocks used */
452 int error; /* error return value */
453 xfs_ifork_t *ifp; /* inode fork ptr */
454 int logflags; /* returned value */
455 xfs_extnum_t nextents; /* number of extents in file now */
457 XFS_STATS_INC(xs_add_exlist);
460 ifp = XFS_IFORK_PTR(ip, whichfork);
461 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
466 ASSERT(*idx <= nextents);
469 * Any kind of new delayed allocation goes here.
471 if (isnullstartblock(new->br_startblock)) {
473 ASSERT((cur->bc_private.b.flags &
474 XFS_BTCUR_BPRV_WASDEL) == 0);
475 error = xfs_bmap_add_extent_hole_delay(ip, idx, new,
479 * Real allocation off the end of the file.
481 else if (*idx == nextents) {
483 ASSERT((cur->bc_private.b.flags &
484 XFS_BTCUR_BPRV_WASDEL) == 0);
485 error = xfs_bmap_add_extent_hole_real(ip, idx, cur, new,
486 &logflags, whichfork);
488 xfs_bmbt_irec_t prev; /* old extent at offset idx */
491 * Get the record referred to by idx.
493 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &prev);
495 * If it's a real allocation record, and the new allocation ends
496 * after the start of the referred to record, then we're filling
497 * in a delayed or unwritten allocation with a real one, or
498 * converting real back to unwritten.
500 if (!isnullstartblock(new->br_startblock) &&
501 new->br_startoff + new->br_blockcount > prev.br_startoff) {
502 if (prev.br_state != XFS_EXT_UNWRITTEN &&
503 isnullstartblock(prev.br_startblock)) {
504 da_old = startblockval(prev.br_startblock);
506 ASSERT(cur->bc_private.b.flags &
507 XFS_BTCUR_BPRV_WASDEL);
508 error = xfs_bmap_add_extent_delay_real(tp, ip,
509 idx, &cur, new, &da_new,
510 first, flist, &logflags);
512 ASSERT(new->br_state == XFS_EXT_NORM ||
513 new->br_state == XFS_EXT_UNWRITTEN);
515 error = xfs_bmap_add_extent_unwritten_real(ip,
516 idx, &cur, new, &logflags);
522 * Otherwise we're filling in a hole with an allocation.
526 ASSERT((cur->bc_private.b.flags &
527 XFS_BTCUR_BPRV_WASDEL) == 0);
528 error = xfs_bmap_add_extent_hole_real(ip, idx, cur,
529 new, &logflags, whichfork);
535 ASSERT(*curp == cur || *curp == NULL);
538 * Convert to a btree if necessary.
540 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
541 XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
542 int tmp_logflags; /* partial log flag return val */
545 error = xfs_bmap_extents_to_btree(tp, ip, first,
546 flist, &cur, da_old > 0, &tmp_logflags, whichfork);
547 logflags |= tmp_logflags;
552 * Adjust for changes in reserved delayed indirect blocks.
553 * Nothing to do for disk quotas here.
555 if (da_old || da_new) {
560 nblks += cur->bc_private.b.allocated;
561 ASSERT(nblks <= da_old);
563 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
564 (int64_t)(da_old - nblks), 0);
567 * Clear out the allocated field, done with it now in any case.
570 cur->bc_private.b.allocated = 0;
576 xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
578 *logflagsp = logflags;
583 * Called by xfs_bmap_add_extent to handle cases converting a delayed
584 * allocation to a real allocation.
586 STATIC int /* error */
587 xfs_bmap_add_extent_delay_real(
588 struct xfs_trans *tp, /* transaction pointer */
589 xfs_inode_t *ip, /* incore inode pointer */
590 xfs_extnum_t *idx, /* extent number to update/insert */
591 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
592 xfs_bmbt_irec_t *new, /* new data to add to file extents */
593 xfs_filblks_t *dnew, /* new delayed-alloc indirect blocks */
594 xfs_fsblock_t *first, /* pointer to firstblock variable */
595 xfs_bmap_free_t *flist, /* list of extents to be freed */
596 int *logflagsp) /* inode logging flags */
598 xfs_btree_cur_t *cur; /* btree cursor */
599 int diff; /* temp value */
600 xfs_bmbt_rec_host_t *ep; /* extent entry for idx */
601 int error; /* error return value */
602 int i; /* temp state */
603 xfs_ifork_t *ifp; /* inode fork pointer */
604 xfs_fileoff_t new_endoff; /* end offset of new entry */
605 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
606 /* left is 0, right is 1, prev is 2 */
607 int rval=0; /* return value (logging flags) */
608 int state = 0;/* state bits, accessed thru macros */
609 xfs_filblks_t temp=0; /* value for dnew calculations */
610 xfs_filblks_t temp2=0;/* value for dnew calculations */
611 int tmp_rval; /* partial logging flags */
618 * Set up a bunch of variables to make the tests simpler.
621 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
622 ep = xfs_iext_get_ext(ifp, *idx);
623 xfs_bmbt_get_all(ep, &PREV);
624 new_endoff = new->br_startoff + new->br_blockcount;
625 ASSERT(PREV.br_startoff <= new->br_startoff);
626 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
629 * Set flags determining what part of the previous delayed allocation
630 * extent is being replaced by a real allocation.
632 if (PREV.br_startoff == new->br_startoff)
633 state |= BMAP_LEFT_FILLING;
634 if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
635 state |= BMAP_RIGHT_FILLING;
638 * Check and set flags if this segment has a left neighbor.
639 * Don't set contiguous if the combined extent would be too large.
642 state |= BMAP_LEFT_VALID;
643 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
645 if (isnullstartblock(LEFT.br_startblock))
646 state |= BMAP_LEFT_DELAY;
649 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
650 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
651 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
652 LEFT.br_state == new->br_state &&
653 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
654 state |= BMAP_LEFT_CONTIG;
657 * Check and set flags if this segment has a right neighbor.
658 * Don't set contiguous if the combined extent would be too large.
659 * Also check for all-three-contiguous being too large.
661 if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
662 state |= BMAP_RIGHT_VALID;
663 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
665 if (isnullstartblock(RIGHT.br_startblock))
666 state |= BMAP_RIGHT_DELAY;
669 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
670 new_endoff == RIGHT.br_startoff &&
671 new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
672 new->br_state == RIGHT.br_state &&
673 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
674 ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
675 BMAP_RIGHT_FILLING)) !=
676 (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
677 BMAP_RIGHT_FILLING) ||
678 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
680 state |= BMAP_RIGHT_CONTIG;
684 * Switch out based on the FILLING and CONTIG state bits.
686 switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
687 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
688 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
689 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
691 * Filling in all of a previously delayed allocation extent.
692 * The left and right neighbors are both contiguous with new.
695 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
696 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
697 LEFT.br_blockcount + PREV.br_blockcount +
698 RIGHT.br_blockcount);
699 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
701 xfs_iext_remove(ip, *idx + 1, 2, state);
702 ip->i_d.di_nextents--;
704 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
706 rval = XFS_ILOG_CORE;
707 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
709 RIGHT.br_blockcount, &i)))
711 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
712 if ((error = xfs_btree_delete(cur, &i)))
714 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
715 if ((error = xfs_btree_decrement(cur, 0, &i)))
717 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
718 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
722 RIGHT.br_blockcount, LEFT.br_state)))
728 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
730 * Filling in all of a previously delayed allocation extent.
731 * The left neighbor is contiguous, the right is not.
735 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
736 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
737 LEFT.br_blockcount + PREV.br_blockcount);
738 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
740 xfs_iext_remove(ip, *idx + 1, 1, state);
742 rval = XFS_ILOG_DEXT;
745 if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
746 LEFT.br_startblock, LEFT.br_blockcount,
749 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
750 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
753 PREV.br_blockcount, LEFT.br_state)))
759 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
761 * Filling in all of a previously delayed allocation extent.
762 * The right neighbor is contiguous, the left is not.
764 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
765 xfs_bmbt_set_startblock(ep, new->br_startblock);
766 xfs_bmbt_set_blockcount(ep,
767 PREV.br_blockcount + RIGHT.br_blockcount);
768 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
770 xfs_iext_remove(ip, *idx + 1, 1, state);
772 rval = XFS_ILOG_DEXT;
775 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
777 RIGHT.br_blockcount, &i)))
779 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
780 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
783 RIGHT.br_blockcount, PREV.br_state)))
790 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
792 * Filling in all of a previously delayed allocation extent.
793 * Neither the left nor right neighbors are contiguous with
796 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
797 xfs_bmbt_set_startblock(ep, new->br_startblock);
798 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
800 ip->i_d.di_nextents++;
802 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
804 rval = XFS_ILOG_CORE;
805 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
806 new->br_startblock, new->br_blockcount,
809 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
810 cur->bc_rec.b.br_state = XFS_EXT_NORM;
811 if ((error = xfs_btree_insert(cur, &i)))
813 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
819 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
821 * Filling in the first part of a previous delayed allocation.
822 * The left neighbor is contiguous.
824 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
825 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
826 LEFT.br_blockcount + new->br_blockcount);
827 xfs_bmbt_set_startoff(ep,
828 PREV.br_startoff + new->br_blockcount);
829 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
831 temp = PREV.br_blockcount - new->br_blockcount;
832 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
833 xfs_bmbt_set_blockcount(ep, temp);
835 rval = XFS_ILOG_DEXT;
838 if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
839 LEFT.br_startblock, LEFT.br_blockcount,
842 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
843 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
850 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
851 startblockval(PREV.br_startblock));
852 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
853 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
859 case BMAP_LEFT_FILLING:
861 * Filling in the first part of a previous delayed allocation.
862 * The left neighbor is not contiguous.
864 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
865 xfs_bmbt_set_startoff(ep, new_endoff);
866 temp = PREV.br_blockcount - new->br_blockcount;
867 xfs_bmbt_set_blockcount(ep, temp);
868 xfs_iext_insert(ip, *idx, 1, new, state);
869 ip->i_d.di_nextents++;
871 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
873 rval = XFS_ILOG_CORE;
874 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
875 new->br_startblock, new->br_blockcount,
878 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
879 cur->bc_rec.b.br_state = XFS_EXT_NORM;
880 if ((error = xfs_btree_insert(cur, &i)))
882 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
884 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
885 ip->i_d.di_nextents > ip->i_df.if_ext_max) {
886 error = xfs_bmap_extents_to_btree(tp, ip,
887 first, flist, &cur, 1, &tmp_rval,
893 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
894 startblockval(PREV.br_startblock) -
895 (cur ? cur->bc_private.b.allocated : 0));
896 ep = xfs_iext_get_ext(ifp, *idx + 1);
897 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
898 trace_xfs_bmap_post_update(ip, *idx + 1, state, _THIS_IP_);
903 case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
905 * Filling in the last part of a previous delayed allocation.
906 * The right neighbor is contiguous with the new allocation.
908 temp = PREV.br_blockcount - new->br_blockcount;
909 trace_xfs_bmap_pre_update(ip, *idx + 1, state, _THIS_IP_);
910 xfs_bmbt_set_blockcount(ep, temp);
911 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx + 1),
912 new->br_startoff, new->br_startblock,
913 new->br_blockcount + RIGHT.br_blockcount,
915 trace_xfs_bmap_post_update(ip, *idx + 1, state, _THIS_IP_);
917 rval = XFS_ILOG_DEXT;
920 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
922 RIGHT.br_blockcount, &i)))
924 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
925 if ((error = xfs_bmbt_update(cur, new->br_startoff,
933 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
934 startblockval(PREV.br_startblock));
935 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
936 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
937 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
943 case BMAP_RIGHT_FILLING:
945 * Filling in the last part of a previous delayed allocation.
946 * The right neighbor is not contiguous.
948 temp = PREV.br_blockcount - new->br_blockcount;
949 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
950 xfs_bmbt_set_blockcount(ep, temp);
951 xfs_iext_insert(ip, *idx + 1, 1, new, state);
952 ip->i_d.di_nextents++;
954 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
956 rval = XFS_ILOG_CORE;
957 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
958 new->br_startblock, new->br_blockcount,
961 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
962 cur->bc_rec.b.br_state = XFS_EXT_NORM;
963 if ((error = xfs_btree_insert(cur, &i)))
965 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
967 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
968 ip->i_d.di_nextents > ip->i_df.if_ext_max) {
969 error = xfs_bmap_extents_to_btree(tp, ip,
970 first, flist, &cur, 1, &tmp_rval,
976 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
977 startblockval(PREV.br_startblock) -
978 (cur ? cur->bc_private.b.allocated : 0));
979 ep = xfs_iext_get_ext(ifp, *idx);
980 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
981 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
989 * Filling in the middle part of a previous delayed allocation.
990 * Contiguity is impossible here.
991 * This case is avoided almost all the time.
993 * We start with a delayed allocation:
995 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
998 * and we are allocating:
999 * +rrrrrrrrrrrrrrrrr+
1002 * and we set it up for insertion as:
1003 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
1005 * PREV @ idx LEFT RIGHT
1006 * inserted at idx + 1
1008 temp = new->br_startoff - PREV.br_startoff;
1009 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
1010 trace_xfs_bmap_pre_update(ip, *idx, 0, _THIS_IP_);
1011 xfs_bmbt_set_blockcount(ep, temp); /* truncate PREV */
1013 RIGHT.br_state = PREV.br_state;
1014 RIGHT.br_startblock = nullstartblock(
1015 (int)xfs_bmap_worst_indlen(ip, temp2));
1016 RIGHT.br_startoff = new_endoff;
1017 RIGHT.br_blockcount = temp2;
1018 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
1019 xfs_iext_insert(ip, *idx + 1, 2, &LEFT, state);
1020 ip->i_d.di_nextents++;
1022 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1024 rval = XFS_ILOG_CORE;
1025 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1026 new->br_startblock, new->br_blockcount,
1029 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1030 cur->bc_rec.b.br_state = XFS_EXT_NORM;
1031 if ((error = xfs_btree_insert(cur, &i)))
1033 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1035 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1036 ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1037 error = xfs_bmap_extents_to_btree(tp, ip,
1038 first, flist, &cur, 1, &tmp_rval,
1044 temp = xfs_bmap_worst_indlen(ip, temp);
1045 temp2 = xfs_bmap_worst_indlen(ip, temp2);
1046 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
1047 (cur ? cur->bc_private.b.allocated : 0));
1049 error = xfs_icsb_modify_counters(ip->i_mount,
1051 -((int64_t)diff), 0);
1057 ep = xfs_iext_get_ext(ifp, *idx);
1058 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
1059 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1060 trace_xfs_bmap_pre_update(ip, *idx + 2, state, _THIS_IP_);
1061 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx + 2),
1062 nullstartblock((int)temp2));
1063 trace_xfs_bmap_post_update(ip, *idx + 2, state, _THIS_IP_);
1066 *dnew = temp + temp2;
1069 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1070 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1071 case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1072 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1073 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1074 case BMAP_LEFT_CONTIG:
1075 case BMAP_RIGHT_CONTIG:
1077 * These cases are all impossible.
1091 * Called by xfs_bmap_add_extent to handle cases converting an unwritten
1092 * allocation to a real allocation or vice versa.
1094 STATIC int /* error */
1095 xfs_bmap_add_extent_unwritten_real(
1096 xfs_inode_t *ip, /* incore inode pointer */
1097 xfs_extnum_t *idx, /* extent number to update/insert */
1098 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
1099 xfs_bmbt_irec_t *new, /* new data to add to file extents */
1100 int *logflagsp) /* inode logging flags */
1102 xfs_btree_cur_t *cur; /* btree cursor */
1103 xfs_bmbt_rec_host_t *ep; /* extent entry for idx */
1104 int error; /* error return value */
1105 int i; /* temp state */
1106 xfs_ifork_t *ifp; /* inode fork pointer */
1107 xfs_fileoff_t new_endoff; /* end offset of new entry */
1108 xfs_exntst_t newext; /* new extent state */
1109 xfs_exntst_t oldext; /* old extent state */
1110 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
1111 /* left is 0, right is 1, prev is 2 */
1112 int rval=0; /* return value (logging flags) */
1113 int state = 0;/* state bits, accessed thru macros */
1119 * Set up a bunch of variables to make the tests simpler.
1123 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1124 ep = xfs_iext_get_ext(ifp, *idx);
1125 xfs_bmbt_get_all(ep, &PREV);
1126 newext = new->br_state;
1127 oldext = (newext == XFS_EXT_UNWRITTEN) ?
1128 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
1129 ASSERT(PREV.br_state == oldext);
1130 new_endoff = new->br_startoff + new->br_blockcount;
1131 ASSERT(PREV.br_startoff <= new->br_startoff);
1132 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1135 * Set flags determining what part of the previous oldext allocation
1136 * extent is being replaced by a newext allocation.
1138 if (PREV.br_startoff == new->br_startoff)
1139 state |= BMAP_LEFT_FILLING;
1140 if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1141 state |= BMAP_RIGHT_FILLING;
1144 * Check and set flags if this segment has a left neighbor.
1145 * Don't set contiguous if the combined extent would be too large.
1148 state |= BMAP_LEFT_VALID;
1149 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
1151 if (isnullstartblock(LEFT.br_startblock))
1152 state |= BMAP_LEFT_DELAY;
1155 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1156 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1157 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1158 LEFT.br_state == newext &&
1159 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1160 state |= BMAP_LEFT_CONTIG;
1163 * Check and set flags if this segment has a right neighbor.
1164 * Don't set contiguous if the combined extent would be too large.
1165 * Also check for all-three-contiguous being too large.
1167 if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1168 state |= BMAP_RIGHT_VALID;
1169 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
1170 if (isnullstartblock(RIGHT.br_startblock))
1171 state |= BMAP_RIGHT_DELAY;
1174 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1175 new_endoff == RIGHT.br_startoff &&
1176 new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1177 newext == RIGHT.br_state &&
1178 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1179 ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1180 BMAP_RIGHT_FILLING)) !=
1181 (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1182 BMAP_RIGHT_FILLING) ||
1183 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1185 state |= BMAP_RIGHT_CONTIG;
1188 * Switch out based on the FILLING and CONTIG state bits.
1190 switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1191 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1192 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1193 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1195 * Setting all of a previous oldext extent to newext.
1196 * The left and right neighbors are both contiguous with new.
1200 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1201 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1202 LEFT.br_blockcount + PREV.br_blockcount +
1203 RIGHT.br_blockcount);
1204 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1206 xfs_iext_remove(ip, *idx + 1, 2, state);
1207 ip->i_d.di_nextents -= 2;
1209 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1211 rval = XFS_ILOG_CORE;
1212 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1213 RIGHT.br_startblock,
1214 RIGHT.br_blockcount, &i)))
1216 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1217 if ((error = xfs_btree_delete(cur, &i)))
1219 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1220 if ((error = xfs_btree_decrement(cur, 0, &i)))
1222 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1223 if ((error = xfs_btree_delete(cur, &i)))
1225 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1226 if ((error = xfs_btree_decrement(cur, 0, &i)))
1228 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1229 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1231 LEFT.br_blockcount + PREV.br_blockcount +
1232 RIGHT.br_blockcount, LEFT.br_state)))
1237 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1239 * Setting all of a previous oldext extent to newext.
1240 * The left neighbor is contiguous, the right is not.
1244 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1245 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1246 LEFT.br_blockcount + PREV.br_blockcount);
1247 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1249 xfs_iext_remove(ip, *idx + 1, 1, state);
1250 ip->i_d.di_nextents--;
1252 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1254 rval = XFS_ILOG_CORE;
1255 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1256 PREV.br_startblock, PREV.br_blockcount,
1259 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1260 if ((error = xfs_btree_delete(cur, &i)))
1262 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1263 if ((error = xfs_btree_decrement(cur, 0, &i)))
1265 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1266 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1268 LEFT.br_blockcount + PREV.br_blockcount,
1274 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1276 * Setting all of a previous oldext extent to newext.
1277 * The right neighbor is contiguous, the left is not.
1279 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1280 xfs_bmbt_set_blockcount(ep,
1281 PREV.br_blockcount + RIGHT.br_blockcount);
1282 xfs_bmbt_set_state(ep, newext);
1283 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1284 xfs_iext_remove(ip, *idx + 1, 1, state);
1285 ip->i_d.di_nextents--;
1287 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1289 rval = XFS_ILOG_CORE;
1290 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1291 RIGHT.br_startblock,
1292 RIGHT.br_blockcount, &i)))
1294 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1295 if ((error = xfs_btree_delete(cur, &i)))
1297 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1298 if ((error = xfs_btree_decrement(cur, 0, &i)))
1300 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1301 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1303 new->br_blockcount + RIGHT.br_blockcount,
1309 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1311 * Setting all of a previous oldext extent to newext.
1312 * Neither the left nor right neighbors are contiguous with
1315 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1316 xfs_bmbt_set_state(ep, newext);
1317 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1320 rval = XFS_ILOG_DEXT;
1323 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1324 new->br_startblock, new->br_blockcount,
1327 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1328 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1329 new->br_startblock, new->br_blockcount,
1335 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1337 * Setting the first part of a previous oldext extent to newext.
1338 * The left neighbor is contiguous.
1340 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
1341 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
1342 LEFT.br_blockcount + new->br_blockcount);
1343 xfs_bmbt_set_startoff(ep,
1344 PREV.br_startoff + new->br_blockcount);
1345 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
1347 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1348 xfs_bmbt_set_startblock(ep,
1349 new->br_startblock + new->br_blockcount);
1350 xfs_bmbt_set_blockcount(ep,
1351 PREV.br_blockcount - new->br_blockcount);
1352 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1357 rval = XFS_ILOG_DEXT;
1360 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1361 PREV.br_startblock, PREV.br_blockcount,
1364 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1365 if ((error = xfs_bmbt_update(cur,
1366 PREV.br_startoff + new->br_blockcount,
1367 PREV.br_startblock + new->br_blockcount,
1368 PREV.br_blockcount - new->br_blockcount,
1371 if ((error = xfs_btree_decrement(cur, 0, &i)))
1373 if (xfs_bmbt_update(cur, LEFT.br_startoff,
1375 LEFT.br_blockcount + new->br_blockcount,
1381 case BMAP_LEFT_FILLING:
1383 * Setting the first part of a previous oldext extent to newext.
1384 * The left neighbor is not contiguous.
1386 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1387 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1388 xfs_bmbt_set_startoff(ep, new_endoff);
1389 xfs_bmbt_set_blockcount(ep,
1390 PREV.br_blockcount - new->br_blockcount);
1391 xfs_bmbt_set_startblock(ep,
1392 new->br_startblock + new->br_blockcount);
1393 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1395 xfs_iext_insert(ip, *idx, 1, new, state);
1396 ip->i_d.di_nextents++;
1398 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1400 rval = XFS_ILOG_CORE;
1401 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1402 PREV.br_startblock, PREV.br_blockcount,
1405 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1406 if ((error = xfs_bmbt_update(cur,
1407 PREV.br_startoff + new->br_blockcount,
1408 PREV.br_startblock + new->br_blockcount,
1409 PREV.br_blockcount - new->br_blockcount,
1412 cur->bc_rec.b = *new;
1413 if ((error = xfs_btree_insert(cur, &i)))
1415 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1419 case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1421 * Setting the last part of a previous oldext extent to newext.
1422 * The right neighbor is contiguous with the new allocation.
1424 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1425 xfs_bmbt_set_blockcount(ep,
1426 PREV.br_blockcount - new->br_blockcount);
1427 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1431 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1432 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1433 new->br_startoff, new->br_startblock,
1434 new->br_blockcount + RIGHT.br_blockcount, newext);
1435 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1438 rval = XFS_ILOG_DEXT;
1441 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1443 PREV.br_blockcount, &i)))
1445 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1446 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1448 PREV.br_blockcount - new->br_blockcount,
1451 if ((error = xfs_btree_increment(cur, 0, &i)))
1453 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1455 new->br_blockcount + RIGHT.br_blockcount,
1461 case BMAP_RIGHT_FILLING:
1463 * Setting the last part of a previous oldext extent to newext.
1464 * The right neighbor is not contiguous.
1466 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1467 xfs_bmbt_set_blockcount(ep,
1468 PREV.br_blockcount - new->br_blockcount);
1469 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1472 xfs_iext_insert(ip, *idx, 1, new, state);
1474 ip->i_d.di_nextents++;
1476 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1478 rval = XFS_ILOG_CORE;
1479 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1480 PREV.br_startblock, PREV.br_blockcount,
1483 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1484 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1486 PREV.br_blockcount - new->br_blockcount,
1489 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1490 new->br_startblock, new->br_blockcount,
1493 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1494 cur->bc_rec.b.br_state = XFS_EXT_NORM;
1495 if ((error = xfs_btree_insert(cur, &i)))
1497 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1503 * Setting the middle part of a previous oldext extent to
1504 * newext. Contiguity is impossible here.
1505 * One extent becomes three extents.
1507 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1508 xfs_bmbt_set_blockcount(ep,
1509 new->br_startoff - PREV.br_startoff);
1510 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1513 r[1].br_startoff = new_endoff;
1514 r[1].br_blockcount =
1515 PREV.br_startoff + PREV.br_blockcount - new_endoff;
1516 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1517 r[1].br_state = oldext;
1520 xfs_iext_insert(ip, *idx, 2, &r[0], state);
1522 ip->i_d.di_nextents += 2;
1524 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1526 rval = XFS_ILOG_CORE;
1527 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1528 PREV.br_startblock, PREV.br_blockcount,
1531 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1532 /* new right extent - oldext */
1533 if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1534 r[1].br_startblock, r[1].br_blockcount,
1537 /* new left extent - oldext */
1538 cur->bc_rec.b = PREV;
1539 cur->bc_rec.b.br_blockcount =
1540 new->br_startoff - PREV.br_startoff;
1541 if ((error = xfs_btree_insert(cur, &i)))
1543 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1545 * Reset the cursor to the position of the new extent
1546 * we are about to insert as we can't trust it after
1547 * the previous insert.
1549 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1550 new->br_startblock, new->br_blockcount,
1553 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1554 /* new middle extent - newext */
1555 cur->bc_rec.b.br_state = new->br_state;
1556 if ((error = xfs_btree_insert(cur, &i)))
1558 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1562 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1563 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1564 case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1565 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1566 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1567 case BMAP_LEFT_CONTIG:
1568 case BMAP_RIGHT_CONTIG:
1570 * These cases are all impossible.
1584 * Called by xfs_bmap_add_extent to handle cases converting a hole
1585 * to a delayed allocation.
1588 STATIC int /* error */
1589 xfs_bmap_add_extent_hole_delay(
1590 xfs_inode_t *ip, /* incore inode pointer */
1591 xfs_extnum_t *idx, /* extent number to update/insert */
1592 xfs_bmbt_irec_t *new, /* new data to add to file extents */
1593 int *logflagsp) /* inode logging flags */
1595 xfs_ifork_t *ifp; /* inode fork pointer */
1596 xfs_bmbt_irec_t left; /* left neighbor extent entry */
1597 xfs_filblks_t newlen=0; /* new indirect size */
1598 xfs_filblks_t oldlen=0; /* old indirect size */
1599 xfs_bmbt_irec_t right; /* right neighbor extent entry */
1600 int state; /* state bits, accessed thru macros */
1601 xfs_filblks_t temp=0; /* temp for indirect calculations */
1603 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1605 ASSERT(isnullstartblock(new->br_startblock));
1608 * Check and set flags if this segment has a left neighbor
1611 state |= BMAP_LEFT_VALID;
1612 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
1614 if (isnullstartblock(left.br_startblock))
1615 state |= BMAP_LEFT_DELAY;
1619 * Check and set flags if the current (right) segment exists.
1620 * If it doesn't exist, we're converting the hole at end-of-file.
1622 if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1623 state |= BMAP_RIGHT_VALID;
1624 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
1626 if (isnullstartblock(right.br_startblock))
1627 state |= BMAP_RIGHT_DELAY;
1631 * Set contiguity flags on the left and right neighbors.
1632 * Don't let extents get too large, even if the pieces are contiguous.
1634 if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
1635 left.br_startoff + left.br_blockcount == new->br_startoff &&
1636 left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1637 state |= BMAP_LEFT_CONTIG;
1639 if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
1640 new->br_startoff + new->br_blockcount == right.br_startoff &&
1641 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1642 (!(state & BMAP_LEFT_CONTIG) ||
1643 (left.br_blockcount + new->br_blockcount +
1644 right.br_blockcount <= MAXEXTLEN)))
1645 state |= BMAP_RIGHT_CONTIG;
1648 * Switch out based on the contiguity flags.
1650 switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1651 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1653 * New allocation is contiguous with delayed allocations
1654 * on the left and on the right.
1655 * Merge all three into a single extent record.
1658 temp = left.br_blockcount + new->br_blockcount +
1659 right.br_blockcount;
1661 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1662 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1663 oldlen = startblockval(left.br_startblock) +
1664 startblockval(new->br_startblock) +
1665 startblockval(right.br_startblock);
1666 newlen = xfs_bmap_worst_indlen(ip, temp);
1667 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1668 nullstartblock((int)newlen));
1669 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1671 xfs_iext_remove(ip, *idx + 1, 1, state);
1674 case BMAP_LEFT_CONTIG:
1676 * New allocation is contiguous with a delayed allocation
1678 * Merge the new allocation with the left neighbor.
1681 temp = left.br_blockcount + new->br_blockcount;
1683 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1684 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1685 oldlen = startblockval(left.br_startblock) +
1686 startblockval(new->br_startblock);
1687 newlen = xfs_bmap_worst_indlen(ip, temp);
1688 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1689 nullstartblock((int)newlen));
1690 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1693 case BMAP_RIGHT_CONTIG:
1695 * New allocation is contiguous with a delayed allocation
1697 * Merge the new allocation with the right neighbor.
1699 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1700 temp = new->br_blockcount + right.br_blockcount;
1701 oldlen = startblockval(new->br_startblock) +
1702 startblockval(right.br_startblock);
1703 newlen = xfs_bmap_worst_indlen(ip, temp);
1704 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1706 nullstartblock((int)newlen), temp, right.br_state);
1707 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1712 * New allocation is not contiguous with another
1713 * delayed allocation.
1714 * Insert a new entry.
1716 oldlen = newlen = 0;
1717 xfs_iext_insert(ip, *idx, 1, new, state);
1720 if (oldlen != newlen) {
1721 ASSERT(oldlen > newlen);
1722 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
1723 (int64_t)(oldlen - newlen), 0);
1725 * Nothing to do for disk quota accounting here.
1733 * Called by xfs_bmap_add_extent to handle cases converting a hole
1734 * to a real allocation.
1736 STATIC int /* error */
1737 xfs_bmap_add_extent_hole_real(
1738 xfs_inode_t *ip, /* incore inode pointer */
1739 xfs_extnum_t *idx, /* extent number to update/insert */
1740 xfs_btree_cur_t *cur, /* if null, not a btree */
1741 xfs_bmbt_irec_t *new, /* new data to add to file extents */
1742 int *logflagsp, /* inode logging flags */
1743 int whichfork) /* data or attr fork */
1745 int error; /* error return value */
1746 int i; /* temp state */
1747 xfs_ifork_t *ifp; /* inode fork pointer */
1748 xfs_bmbt_irec_t left; /* left neighbor extent entry */
1749 xfs_bmbt_irec_t right; /* right neighbor extent entry */
1750 int rval=0; /* return value (logging flags) */
1751 int state; /* state bits, accessed thru macros */
1753 ifp = XFS_IFORK_PTR(ip, whichfork);
1754 ASSERT(*idx <= ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
1757 if (whichfork == XFS_ATTR_FORK)
1758 state |= BMAP_ATTRFORK;
1761 * Check and set flags if this segment has a left neighbor.
1764 state |= BMAP_LEFT_VALID;
1765 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
1766 if (isnullstartblock(left.br_startblock))
1767 state |= BMAP_LEFT_DELAY;
1771 * Check and set flags if this segment has a current value.
1772 * Not true if we're inserting into the "hole" at eof.
1774 if (*idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1775 state |= BMAP_RIGHT_VALID;
1776 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
1777 if (isnullstartblock(right.br_startblock))
1778 state |= BMAP_RIGHT_DELAY;
1782 * We're inserting a real allocation between "left" and "right".
1783 * Set the contiguity flags. Don't let extents get too large.
1785 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1786 left.br_startoff + left.br_blockcount == new->br_startoff &&
1787 left.br_startblock + left.br_blockcount == new->br_startblock &&
1788 left.br_state == new->br_state &&
1789 left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1790 state |= BMAP_LEFT_CONTIG;
1792 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1793 new->br_startoff + new->br_blockcount == right.br_startoff &&
1794 new->br_startblock + new->br_blockcount == right.br_startblock &&
1795 new->br_state == right.br_state &&
1796 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1797 (!(state & BMAP_LEFT_CONTIG) ||
1798 left.br_blockcount + new->br_blockcount +
1799 right.br_blockcount <= MAXEXTLEN))
1800 state |= BMAP_RIGHT_CONTIG;
1804 * Select which case we're in here, and implement it.
1806 switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1807 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1809 * New allocation is contiguous with real allocations on the
1810 * left and on the right.
1811 * Merge all three into a single extent record.
1814 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1815 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1816 left.br_blockcount + new->br_blockcount +
1817 right.br_blockcount);
1818 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1820 xfs_iext_remove(ip, *idx + 1, 1, state);
1822 XFS_IFORK_NEXT_SET(ip, whichfork,
1823 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
1825 rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1827 rval = XFS_ILOG_CORE;
1828 if ((error = xfs_bmbt_lookup_eq(cur,
1830 right.br_startblock,
1831 right.br_blockcount, &i)))
1833 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1834 if ((error = xfs_btree_delete(cur, &i)))
1836 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1837 if ((error = xfs_btree_decrement(cur, 0, &i)))
1839 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1840 if ((error = xfs_bmbt_update(cur, left.br_startoff,
1842 left.br_blockcount +
1843 new->br_blockcount +
1844 right.br_blockcount,
1850 case BMAP_LEFT_CONTIG:
1852 * New allocation is contiguous with a real allocation
1854 * Merge the new allocation with the left neighbor.
1857 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1858 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1859 left.br_blockcount + new->br_blockcount);
1860 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1863 rval = xfs_ilog_fext(whichfork);
1866 if ((error = xfs_bmbt_lookup_eq(cur,
1869 left.br_blockcount, &i)))
1871 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1872 if ((error = xfs_bmbt_update(cur, left.br_startoff,
1874 left.br_blockcount +
1881 case BMAP_RIGHT_CONTIG:
1883 * New allocation is contiguous with a real allocation
1885 * Merge the new allocation with the right neighbor.
1887 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1888 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1889 new->br_startoff, new->br_startblock,
1890 new->br_blockcount + right.br_blockcount,
1892 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1895 rval = xfs_ilog_fext(whichfork);
1898 if ((error = xfs_bmbt_lookup_eq(cur,
1900 right.br_startblock,
1901 right.br_blockcount, &i)))
1903 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1904 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1906 new->br_blockcount +
1907 right.br_blockcount,
1915 * New allocation is not contiguous with another
1917 * Insert a new entry.
1919 xfs_iext_insert(ip, *idx, 1, new, state);
1920 XFS_IFORK_NEXT_SET(ip, whichfork,
1921 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
1923 rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1925 rval = XFS_ILOG_CORE;
1926 if ((error = xfs_bmbt_lookup_eq(cur,
1929 new->br_blockcount, &i)))
1931 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1932 cur->bc_rec.b.br_state = new->br_state;
1933 if ((error = xfs_btree_insert(cur, &i)))
1935 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1945 * Adjust the size of the new extent based on di_extsize and rt extsize.
1948 xfs_bmap_extsize_align(
1950 xfs_bmbt_irec_t *gotp, /* next extent pointer */
1951 xfs_bmbt_irec_t *prevp, /* previous extent pointer */
1952 xfs_extlen_t extsz, /* align to this extent size */
1953 int rt, /* is this a realtime inode? */
1954 int eof, /* is extent at end-of-file? */
1955 int delay, /* creating delalloc extent? */
1956 int convert, /* overwriting unwritten extent? */
1957 xfs_fileoff_t *offp, /* in/out: aligned offset */
1958 xfs_extlen_t *lenp) /* in/out: aligned length */
1960 xfs_fileoff_t orig_off; /* original offset */
1961 xfs_extlen_t orig_alen; /* original length */
1962 xfs_fileoff_t orig_end; /* original off+len */
1963 xfs_fileoff_t nexto; /* next file offset */
1964 xfs_fileoff_t prevo; /* previous file offset */
1965 xfs_fileoff_t align_off; /* temp for offset */
1966 xfs_extlen_t align_alen; /* temp for length */
1967 xfs_extlen_t temp; /* temp for calculations */
1972 orig_off = align_off = *offp;
1973 orig_alen = align_alen = *lenp;
1974 orig_end = orig_off + orig_alen;
1977 * If this request overlaps an existing extent, then don't
1978 * attempt to perform any additional alignment.
1980 if (!delay && !eof &&
1981 (orig_off >= gotp->br_startoff) &&
1982 (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
1987 * If the file offset is unaligned vs. the extent size
1988 * we need to align it. This will be possible unless
1989 * the file was previously written with a kernel that didn't
1990 * perform this alignment, or if a truncate shot us in the
1993 temp = do_mod(orig_off, extsz);
1999 * Same adjustment for the end of the requested area.
2001 if ((temp = (align_alen % extsz))) {
2002 align_alen += extsz - temp;
2005 * If the previous block overlaps with this proposed allocation
2006 * then move the start forward without adjusting the length.
2008 if (prevp->br_startoff != NULLFILEOFF) {
2009 if (prevp->br_startblock == HOLESTARTBLOCK)
2010 prevo = prevp->br_startoff;
2012 prevo = prevp->br_startoff + prevp->br_blockcount;
2015 if (align_off != orig_off && align_off < prevo)
2018 * If the next block overlaps with this proposed allocation
2019 * then move the start back without adjusting the length,
2020 * but not before offset 0.
2021 * This may of course make the start overlap previous block,
2022 * and if we hit the offset 0 limit then the next block
2023 * can still overlap too.
2025 if (!eof && gotp->br_startoff != NULLFILEOFF) {
2026 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
2027 (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
2028 nexto = gotp->br_startoff + gotp->br_blockcount;
2030 nexto = gotp->br_startoff;
2032 nexto = NULLFILEOFF;
2034 align_off + align_alen != orig_end &&
2035 align_off + align_alen > nexto)
2036 align_off = nexto > align_alen ? nexto - align_alen : 0;
2038 * If we're now overlapping the next or previous extent that
2039 * means we can't fit an extsz piece in this hole. Just move
2040 * the start forward to the first valid spot and set
2041 * the length so we hit the end.
2043 if (align_off != orig_off && align_off < prevo)
2045 if (align_off + align_alen != orig_end &&
2046 align_off + align_alen > nexto &&
2047 nexto != NULLFILEOFF) {
2048 ASSERT(nexto > prevo);
2049 align_alen = nexto - align_off;
2053 * If realtime, and the result isn't a multiple of the realtime
2054 * extent size we need to remove blocks until it is.
2056 if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
2058 * We're not covering the original request, or
2059 * we won't be able to once we fix the length.
2061 if (orig_off < align_off ||
2062 orig_end > align_off + align_alen ||
2063 align_alen - temp < orig_alen)
2064 return XFS_ERROR(EINVAL);
2066 * Try to fix it by moving the start up.
2068 if (align_off + temp <= orig_off) {
2073 * Try to fix it by moving the end in.
2075 else if (align_off + align_alen - temp >= orig_end)
2078 * Set the start to the minimum then trim the length.
2081 align_alen -= orig_off - align_off;
2082 align_off = orig_off;
2083 align_alen -= align_alen % mp->m_sb.sb_rextsize;
2086 * Result doesn't cover the request, fail it.
2088 if (orig_off < align_off || orig_end > align_off + align_alen)
2089 return XFS_ERROR(EINVAL);
2091 ASSERT(orig_off >= align_off);
2092 ASSERT(orig_end <= align_off + align_alen);
2096 if (!eof && gotp->br_startoff != NULLFILEOFF)
2097 ASSERT(align_off + align_alen <= gotp->br_startoff);
2098 if (prevp->br_startoff != NULLFILEOFF)
2099 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
2107 #define XFS_ALLOC_GAP_UNITS 4
2111 xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2113 xfs_fsblock_t adjust; /* adjustment to block numbers */
2114 xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
2115 xfs_mount_t *mp; /* mount point structure */
2116 int nullfb; /* true if ap->firstblock isn't set */
2117 int rt; /* true if inode is realtime */
2119 #define ISVALID(x,y) \
2121 (x) < mp->m_sb.sb_rblocks : \
2122 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2123 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2124 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2126 mp = ap->ip->i_mount;
2127 nullfb = ap->firstblock == NULLFSBLOCK;
2128 rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2129 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2131 * If allocating at eof, and there's a previous real block,
2132 * try to use its last block as our starting point.
2134 if (ap->eof && ap->prevp->br_startoff != NULLFILEOFF &&
2135 !isnullstartblock(ap->prevp->br_startblock) &&
2136 ISVALID(ap->prevp->br_startblock + ap->prevp->br_blockcount,
2137 ap->prevp->br_startblock)) {
2138 ap->rval = ap->prevp->br_startblock + ap->prevp->br_blockcount;
2140 * Adjust for the gap between prevp and us.
2143 (ap->prevp->br_startoff + ap->prevp->br_blockcount);
2145 ISVALID(ap->rval + adjust, ap->prevp->br_startblock))
2149 * If not at eof, then compare the two neighbor blocks.
2150 * Figure out whether either one gives us a good starting point,
2151 * and pick the better one.
2153 else if (!ap->eof) {
2154 xfs_fsblock_t gotbno; /* right side block number */
2155 xfs_fsblock_t gotdiff=0; /* right side difference */
2156 xfs_fsblock_t prevbno; /* left side block number */
2157 xfs_fsblock_t prevdiff=0; /* left side difference */
2160 * If there's a previous (left) block, select a requested
2161 * start block based on it.
2163 if (ap->prevp->br_startoff != NULLFILEOFF &&
2164 !isnullstartblock(ap->prevp->br_startblock) &&
2165 (prevbno = ap->prevp->br_startblock +
2166 ap->prevp->br_blockcount) &&
2167 ISVALID(prevbno, ap->prevp->br_startblock)) {
2169 * Calculate gap to end of previous block.
2171 adjust = prevdiff = ap->off -
2172 (ap->prevp->br_startoff +
2173 ap->prevp->br_blockcount);
2175 * Figure the startblock based on the previous block's
2176 * end and the gap size.
2178 * If the gap is large relative to the piece we're
2179 * allocating, or using it gives us an invalid block
2180 * number, then just use the end of the previous block.
2182 if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2183 ISVALID(prevbno + prevdiff,
2184 ap->prevp->br_startblock))
2189 * If the firstblock forbids it, can't use it,
2192 if (!rt && !nullfb &&
2193 XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2194 prevbno = NULLFSBLOCK;
2197 * No previous block or can't follow it, just default.
2200 prevbno = NULLFSBLOCK;
2202 * If there's a following (right) block, select a requested
2203 * start block based on it.
2205 if (!isnullstartblock(ap->gotp->br_startblock)) {
2207 * Calculate gap to start of next block.
2209 adjust = gotdiff = ap->gotp->br_startoff - ap->off;
2211 * Figure the startblock based on the next block's
2212 * start and the gap size.
2214 gotbno = ap->gotp->br_startblock;
2217 * If the gap is large relative to the piece we're
2218 * allocating, or using it gives us an invalid block
2219 * number, then just use the start of the next block
2220 * offset by our length.
2222 if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2223 ISVALID(gotbno - gotdiff, gotbno))
2225 else if (ISVALID(gotbno - ap->alen, gotbno)) {
2227 gotdiff += adjust - ap->alen;
2231 * If the firstblock forbids it, can't use it,
2234 if (!rt && !nullfb &&
2235 XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2236 gotbno = NULLFSBLOCK;
2239 * No next block, just default.
2242 gotbno = NULLFSBLOCK;
2244 * If both valid, pick the better one, else the only good
2245 * one, else ap->rval is already set (to 0 or the inode block).
2247 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2248 ap->rval = prevdiff <= gotdiff ? prevbno : gotbno;
2249 else if (prevbno != NULLFSBLOCK)
2251 else if (gotbno != NULLFSBLOCK)
2259 xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2261 xfs_alloctype_t atype = 0; /* type for allocation routines */
2262 int error; /* error return value */
2263 xfs_mount_t *mp; /* mount point structure */
2264 xfs_extlen_t prod = 0; /* product factor for allocators */
2265 xfs_extlen_t ralen = 0; /* realtime allocation length */
2266 xfs_extlen_t align; /* minimum allocation alignment */
2269 mp = ap->ip->i_mount;
2270 align = xfs_get_extsz_hint(ap->ip);
2271 prod = align / mp->m_sb.sb_rextsize;
2272 error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2273 align, 1, ap->eof, 0,
2274 ap->conv, &ap->off, &ap->alen);
2278 ASSERT(ap->alen % mp->m_sb.sb_rextsize == 0);
2281 * If the offset & length are not perfectly aligned
2282 * then kill prod, it will just get us in trouble.
2284 if (do_mod(ap->off, align) || ap->alen % align)
2287 * Set ralen to be the actual requested length in rtextents.
2289 ralen = ap->alen / mp->m_sb.sb_rextsize;
2291 * If the old value was close enough to MAXEXTLEN that
2292 * we rounded up to it, cut it back so it's valid again.
2293 * Note that if it's a really large request (bigger than
2294 * MAXEXTLEN), we don't hear about that number, and can't
2295 * adjust the starting point to match it.
2297 if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2298 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2301 * Lock out other modifications to the RT bitmap inode.
2303 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
2304 xfs_trans_ijoin_ref(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
2307 * If it's an allocation to an empty file at offset 0,
2308 * pick an extent that will space things out in the rt area.
2310 if (ap->eof && ap->off == 0) {
2311 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
2313 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2316 ap->rval = rtx * mp->m_sb.sb_rextsize;
2321 xfs_bmap_adjacent(ap);
2324 * Realtime allocation, done through xfs_rtallocate_extent.
2326 atype = ap->rval == 0 ? XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2327 do_div(ap->rval, mp->m_sb.sb_rextsize);
2330 if ((error = xfs_rtallocate_extent(ap->tp, ap->rval, 1, ap->alen,
2331 &ralen, atype, ap->wasdel, prod, &rtb)))
2333 if (rtb == NULLFSBLOCK && prod > 1 &&
2334 (error = xfs_rtallocate_extent(ap->tp, ap->rval, 1,
2335 ap->alen, &ralen, atype,
2336 ap->wasdel, 1, &rtb)))
2339 if (ap->rval != NULLFSBLOCK) {
2340 ap->rval *= mp->m_sb.sb_rextsize;
2341 ralen *= mp->m_sb.sb_rextsize;
2343 ap->ip->i_d.di_nblocks += ralen;
2344 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2346 ap->ip->i_delayed_blks -= ralen;
2348 * Adjust the disk quota also. This was reserved
2351 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2352 ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2353 XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2361 xfs_bmap_btalloc_nullfb(
2362 struct xfs_bmalloca *ap,
2363 struct xfs_alloc_arg *args,
2366 struct xfs_mount *mp = ap->ip->i_mount;
2367 struct xfs_perag *pag;
2368 xfs_agnumber_t ag, startag;
2372 if (ap->userdata && xfs_inode_is_filestream(ap->ip))
2373 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2375 args->type = XFS_ALLOCTYPE_START_BNO;
2376 args->total = ap->total;
2379 * Search for an allocation group with a single extent large enough
2380 * for the request. If one isn't found, then adjust the minimum
2381 * allocation size to the largest space found.
2383 startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
2384 if (startag == NULLAGNUMBER)
2387 pag = xfs_perag_get(mp, ag);
2388 while (*blen < args->maxlen) {
2389 if (!pag->pagf_init) {
2390 error = xfs_alloc_pagf_init(mp, args->tp, ag,
2391 XFS_ALLOC_FLAG_TRYLOCK);
2399 * See xfs_alloc_fix_freelist...
2401 if (pag->pagf_init) {
2402 xfs_extlen_t longest;
2403 longest = xfs_alloc_longest_free_extent(mp, pag);
2404 if (*blen < longest)
2409 if (xfs_inode_is_filestream(ap->ip)) {
2410 if (*blen >= args->maxlen)
2415 * If startag is an invalid AG, we've
2416 * come here once before and
2417 * xfs_filestream_new_ag picked the
2418 * best currently available.
2420 * Don't continue looping, since we
2421 * could loop forever.
2423 if (startag == NULLAGNUMBER)
2426 error = xfs_filestream_new_ag(ap, &ag);
2431 /* loop again to set 'blen'*/
2432 startag = NULLAGNUMBER;
2433 pag = xfs_perag_get(mp, ag);
2437 if (++ag == mp->m_sb.sb_agcount)
2442 pag = xfs_perag_get(mp, ag);
2447 * Since the above loop did a BUF_TRYLOCK, it is
2448 * possible that there is space for this request.
2450 if (notinit || *blen < ap->minlen)
2451 args->minlen = ap->minlen;
2453 * If the best seen length is less than the request
2454 * length, use the best as the minimum.
2456 else if (*blen < args->maxlen)
2457 args->minlen = *blen;
2459 * Otherwise we've seen an extent as big as maxlen,
2460 * use that as the minimum.
2463 args->minlen = args->maxlen;
2466 * set the failure fallback case to look in the selected
2467 * AG as the stream may have moved.
2469 if (xfs_inode_is_filestream(ap->ip))
2470 ap->rval = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
2477 xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2479 xfs_mount_t *mp; /* mount point structure */
2480 xfs_alloctype_t atype = 0; /* type for allocation routines */
2481 xfs_extlen_t align; /* minimum allocation alignment */
2482 xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
2484 xfs_alloc_arg_t args;
2486 xfs_extlen_t nextminlen = 0;
2487 int nullfb; /* true if ap->firstblock isn't set */
2492 mp = ap->ip->i_mount;
2493 align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
2494 if (unlikely(align)) {
2495 error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2496 align, 0, ap->eof, 0, ap->conv,
2497 &ap->off, &ap->alen);
2501 nullfb = ap->firstblock == NULLFSBLOCK;
2502 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2504 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
2505 ag = xfs_filestream_lookup_ag(ap->ip);
2506 ag = (ag != NULLAGNUMBER) ? ag : 0;
2507 ap->rval = XFS_AGB_TO_FSB(mp, ag, 0);
2509 ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2512 ap->rval = ap->firstblock;
2514 xfs_bmap_adjacent(ap);
2517 * If allowed, use ap->rval; otherwise must use firstblock since
2518 * it's in the right allocation group.
2520 if (nullfb || XFS_FSB_TO_AGNO(mp, ap->rval) == fb_agno)
2523 ap->rval = ap->firstblock;
2525 * Normal allocation, done through xfs_alloc_vextent.
2527 tryagain = isaligned = 0;
2530 args.fsbno = ap->rval;
2532 /* Trim the allocation back to the maximum an AG can fit. */
2533 args.maxlen = MIN(ap->alen, XFS_ALLOC_AG_MAX_USABLE(mp));
2534 args.firstblock = ap->firstblock;
2537 error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
2540 } else if (ap->low) {
2541 if (xfs_inode_is_filestream(ap->ip))
2542 args.type = XFS_ALLOCTYPE_FIRST_AG;
2544 args.type = XFS_ALLOCTYPE_START_BNO;
2545 args.total = args.minlen = ap->minlen;
2547 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2548 args.total = ap->total;
2549 args.minlen = ap->minlen;
2551 /* apply extent size hints if obtained earlier */
2552 if (unlikely(align)) {
2554 if ((args.mod = (xfs_extlen_t)do_mod(ap->off, args.prod)))
2555 args.mod = (xfs_extlen_t)(args.prod - args.mod);
2556 } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
2560 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
2561 if ((args.mod = (xfs_extlen_t)(do_mod(ap->off, args.prod))))
2562 args.mod = (xfs_extlen_t)(args.prod - args.mod);
2565 * If we are not low on available data blocks, and the
2566 * underlying logical volume manager is a stripe, and
2567 * the file offset is zero then try to allocate data
2568 * blocks on stripe unit boundary.
2569 * NOTE: ap->aeof is only set if the allocation length
2570 * is >= the stripe unit and the allocation offset is
2571 * at the end of file.
2573 if (!ap->low && ap->aeof) {
2575 args.alignment = mp->m_dalign;
2579 * Adjust for alignment
2581 if (blen > args.alignment && blen <= args.maxlen)
2582 args.minlen = blen - args.alignment;
2583 args.minalignslop = 0;
2586 * First try an exact bno allocation.
2587 * If it fails then do a near or start bno
2588 * allocation with alignment turned on.
2592 args.type = XFS_ALLOCTYPE_THIS_BNO;
2595 * Compute the minlen+alignment for the
2596 * next case. Set slop so that the value
2597 * of minlen+alignment+slop doesn't go up
2598 * between the calls.
2600 if (blen > mp->m_dalign && blen <= args.maxlen)
2601 nextminlen = blen - mp->m_dalign;
2603 nextminlen = args.minlen;
2604 if (nextminlen + mp->m_dalign > args.minlen + 1)
2606 nextminlen + mp->m_dalign -
2609 args.minalignslop = 0;
2613 args.minalignslop = 0;
2615 args.minleft = ap->minleft;
2616 args.wasdel = ap->wasdel;
2618 args.userdata = ap->userdata;
2619 if ((error = xfs_alloc_vextent(&args)))
2621 if (tryagain && args.fsbno == NULLFSBLOCK) {
2623 * Exact allocation failed. Now try with alignment
2627 args.fsbno = ap->rval;
2628 args.alignment = mp->m_dalign;
2629 args.minlen = nextminlen;
2630 args.minalignslop = 0;
2632 if ((error = xfs_alloc_vextent(&args)))
2635 if (isaligned && args.fsbno == NULLFSBLOCK) {
2637 * allocation failed, so turn off alignment and
2641 args.fsbno = ap->rval;
2643 if ((error = xfs_alloc_vextent(&args)))
2646 if (args.fsbno == NULLFSBLOCK && nullfb &&
2647 args.minlen > ap->minlen) {
2648 args.minlen = ap->minlen;
2649 args.type = XFS_ALLOCTYPE_START_BNO;
2650 args.fsbno = ap->rval;
2651 if ((error = xfs_alloc_vextent(&args)))
2654 if (args.fsbno == NULLFSBLOCK && nullfb) {
2656 args.type = XFS_ALLOCTYPE_FIRST_AG;
2657 args.total = ap->minlen;
2659 if ((error = xfs_alloc_vextent(&args)))
2663 if (args.fsbno != NULLFSBLOCK) {
2664 ap->firstblock = ap->rval = args.fsbno;
2665 ASSERT(nullfb || fb_agno == args.agno ||
2666 (ap->low && fb_agno < args.agno));
2667 ap->alen = args.len;
2668 ap->ip->i_d.di_nblocks += args.len;
2669 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2671 ap->ip->i_delayed_blks -= args.len;
2673 * Adjust the disk quota also. This was reserved
2676 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2677 ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2678 XFS_TRANS_DQ_BCOUNT,
2681 ap->rval = NULLFSBLOCK;
2688 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2689 * It figures out where to ask the underlying allocator to put the new extent.
2693 xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2695 if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
2696 return xfs_bmap_rtalloc(ap);
2697 return xfs_bmap_btalloc(ap);
2701 * Transform a btree format file with only one leaf node, where the
2702 * extents list will fit in the inode, into an extents format file.
2703 * Since the file extents are already in-core, all we have to do is
2704 * give up the space for the btree root and pitch the leaf block.
2706 STATIC int /* error */
2707 xfs_bmap_btree_to_extents(
2708 xfs_trans_t *tp, /* transaction pointer */
2709 xfs_inode_t *ip, /* incore inode pointer */
2710 xfs_btree_cur_t *cur, /* btree cursor */
2711 int *logflagsp, /* inode logging flags */
2712 int whichfork) /* data or attr fork */
2715 struct xfs_btree_block *cblock;/* child btree block */
2716 xfs_fsblock_t cbno; /* child block number */
2717 xfs_buf_t *cbp; /* child block's buffer */
2718 int error; /* error return value */
2719 xfs_ifork_t *ifp; /* inode fork data */
2720 xfs_mount_t *mp; /* mount point structure */
2721 __be64 *pp; /* ptr to block address */
2722 struct xfs_btree_block *rblock;/* root btree block */
2725 ifp = XFS_IFORK_PTR(ip, whichfork);
2726 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2727 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
2728 rblock = ifp->if_broot;
2729 ASSERT(be16_to_cpu(rblock->bb_level) == 1);
2730 ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
2731 ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
2732 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
2733 cbno = be64_to_cpu(*pp);
2736 if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
2739 if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
2740 XFS_BMAP_BTREE_REF)))
2742 cblock = XFS_BUF_TO_BLOCK(cbp);
2743 if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
2745 xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
2746 ip->i_d.di_nblocks--;
2747 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
2748 xfs_trans_binval(tp, cbp);
2749 if (cur->bc_bufs[0] == cbp)
2750 cur->bc_bufs[0] = NULL;
2751 xfs_iroot_realloc(ip, -1, whichfork);
2752 ASSERT(ifp->if_broot == NULL);
2753 ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
2754 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
2755 *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2760 * Called by xfs_bmapi to update file extent records and the btree
2761 * after removing space (or undoing a delayed allocation).
2763 STATIC int /* error */
2764 xfs_bmap_del_extent(
2765 xfs_inode_t *ip, /* incore inode pointer */
2766 xfs_trans_t *tp, /* current transaction pointer */
2767 xfs_extnum_t *idx, /* extent number to update/delete */
2768 xfs_bmap_free_t *flist, /* list of extents to be freed */
2769 xfs_btree_cur_t *cur, /* if null, not a btree */
2770 xfs_bmbt_irec_t *del, /* data to remove from extents */
2771 int *logflagsp, /* inode logging flags */
2772 int whichfork) /* data or attr fork */
2774 xfs_filblks_t da_new; /* new delay-alloc indirect blocks */
2775 xfs_filblks_t da_old; /* old delay-alloc indirect blocks */
2776 xfs_fsblock_t del_endblock=0; /* first block past del */
2777 xfs_fileoff_t del_endoff; /* first offset past del */
2778 int delay; /* current block is delayed allocated */
2779 int do_fx; /* free extent at end of routine */
2780 xfs_bmbt_rec_host_t *ep; /* current extent entry pointer */
2781 int error; /* error return value */
2782 int flags; /* inode logging flags */
2783 xfs_bmbt_irec_t got; /* current extent entry */
2784 xfs_fileoff_t got_endoff; /* first offset past got */
2785 int i; /* temp state */
2786 xfs_ifork_t *ifp; /* inode fork pointer */
2787 xfs_mount_t *mp; /* mount structure */
2788 xfs_filblks_t nblks; /* quota/sb block count */
2789 xfs_bmbt_irec_t new; /* new record to be inserted */
2791 uint qfield; /* quota field to update */
2792 xfs_filblks_t temp; /* for indirect length calculations */
2793 xfs_filblks_t temp2; /* for indirect length calculations */
2796 XFS_STATS_INC(xs_del_exlist);
2798 if (whichfork == XFS_ATTR_FORK)
2799 state |= BMAP_ATTRFORK;
2802 ifp = XFS_IFORK_PTR(ip, whichfork);
2803 ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
2804 (uint)sizeof(xfs_bmbt_rec_t)));
2805 ASSERT(del->br_blockcount > 0);
2806 ep = xfs_iext_get_ext(ifp, *idx);
2807 xfs_bmbt_get_all(ep, &got);
2808 ASSERT(got.br_startoff <= del->br_startoff);
2809 del_endoff = del->br_startoff + del->br_blockcount;
2810 got_endoff = got.br_startoff + got.br_blockcount;
2811 ASSERT(got_endoff >= del_endoff);
2812 delay = isnullstartblock(got.br_startblock);
2813 ASSERT(isnullstartblock(del->br_startblock) == delay);
2818 * If deleting a real allocation, must free up the disk space.
2821 flags = XFS_ILOG_CORE;
2823 * Realtime allocation. Free it and record di_nblocks update.
2825 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
2829 ASSERT(do_mod(del->br_blockcount,
2830 mp->m_sb.sb_rextsize) == 0);
2831 ASSERT(do_mod(del->br_startblock,
2832 mp->m_sb.sb_rextsize) == 0);
2833 bno = del->br_startblock;
2834 len = del->br_blockcount;
2835 do_div(bno, mp->m_sb.sb_rextsize);
2836 do_div(len, mp->m_sb.sb_rextsize);
2837 error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
2841 nblks = len * mp->m_sb.sb_rextsize;
2842 qfield = XFS_TRANS_DQ_RTBCOUNT;
2845 * Ordinary allocation.
2849 nblks = del->br_blockcount;
2850 qfield = XFS_TRANS_DQ_BCOUNT;
2853 * Set up del_endblock and cur for later.
2855 del_endblock = del->br_startblock + del->br_blockcount;
2857 if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
2858 got.br_startblock, got.br_blockcount,
2861 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2863 da_old = da_new = 0;
2865 da_old = startblockval(got.br_startblock);
2871 * Set flag value to use in switch statement.
2872 * Left-contig is 2, right-contig is 1.
2874 switch (((got.br_startoff == del->br_startoff) << 1) |
2875 (got_endoff == del_endoff)) {
2878 * Matches the whole extent. Delete the entry.
2880 xfs_iext_remove(ip, *idx, 1,
2881 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
2886 XFS_IFORK_NEXT_SET(ip, whichfork,
2887 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2888 flags |= XFS_ILOG_CORE;
2890 flags |= xfs_ilog_fext(whichfork);
2893 if ((error = xfs_btree_delete(cur, &i)))
2895 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2900 * Deleting the first part of the extent.
2902 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2903 xfs_bmbt_set_startoff(ep, del_endoff);
2904 temp = got.br_blockcount - del->br_blockcount;
2905 xfs_bmbt_set_blockcount(ep, temp);
2907 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2909 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2910 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2914 xfs_bmbt_set_startblock(ep, del_endblock);
2915 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2917 flags |= xfs_ilog_fext(whichfork);
2920 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
2921 got.br_blockcount - del->br_blockcount,
2928 * Deleting the last part of the extent.
2930 temp = got.br_blockcount - del->br_blockcount;
2931 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2932 xfs_bmbt_set_blockcount(ep, temp);
2934 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2936 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2937 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2941 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2943 flags |= xfs_ilog_fext(whichfork);
2946 if ((error = xfs_bmbt_update(cur, got.br_startoff,
2948 got.br_blockcount - del->br_blockcount,
2955 * Deleting the middle of the extent.
2957 temp = del->br_startoff - got.br_startoff;
2958 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2959 xfs_bmbt_set_blockcount(ep, temp);
2960 new.br_startoff = del_endoff;
2961 temp2 = got_endoff - del_endoff;
2962 new.br_blockcount = temp2;
2963 new.br_state = got.br_state;
2965 new.br_startblock = del_endblock;
2966 flags |= XFS_ILOG_CORE;
2968 if ((error = xfs_bmbt_update(cur,
2970 got.br_startblock, temp,
2973 if ((error = xfs_btree_increment(cur, 0, &i)))
2975 cur->bc_rec.b = new;
2976 error = xfs_btree_insert(cur, &i);
2977 if (error && error != ENOSPC)
2980 * If get no-space back from btree insert,
2981 * it tried a split, and we have a zero
2982 * block reservation.
2983 * Fix up our state and return the error.
2985 if (error == ENOSPC) {
2987 * Reset the cursor, don't trust
2988 * it after any insert operation.
2990 if ((error = xfs_bmbt_lookup_eq(cur,
2995 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2997 * Update the btree record back
2998 * to the original value.
3000 if ((error = xfs_bmbt_update(cur,
3007 * Reset the extent record back
3008 * to the original value.
3010 xfs_bmbt_set_blockcount(ep,
3013 error = XFS_ERROR(ENOSPC);
3016 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3018 flags |= xfs_ilog_fext(whichfork);
3019 XFS_IFORK_NEXT_SET(ip, whichfork,
3020 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
3022 ASSERT(whichfork == XFS_DATA_FORK);
3023 temp = xfs_bmap_worst_indlen(ip, temp);
3024 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
3025 temp2 = xfs_bmap_worst_indlen(ip, temp2);
3026 new.br_startblock = nullstartblock((int)temp2);
3027 da_new = temp + temp2;
3028 while (da_new > da_old) {
3032 xfs_bmbt_set_startblock(ep,
3033 nullstartblock((int)temp));
3035 if (da_new == da_old)
3041 nullstartblock((int)temp2);
3045 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
3046 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
3051 * If we need to, add to list of extents to delete.
3054 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
3057 * Adjust inode # blocks in the file.
3060 ip->i_d.di_nblocks -= nblks;
3062 * Adjust quota data.
3065 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
3068 * Account for change in delayed indirect blocks.
3069 * Nothing to do for disk quota accounting here.
3071 ASSERT(da_old >= da_new);
3072 if (da_old > da_new) {
3073 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
3074 (int64_t)(da_old - da_new), 0);
3082 * Remove the entry "free" from the free item list. Prev points to the
3083 * previous entry, unless "free" is the head of the list.
3087 xfs_bmap_free_t *flist, /* free item list header */
3088 xfs_bmap_free_item_t *prev, /* previous item on list, if any */
3089 xfs_bmap_free_item_t *free) /* list item to be freed */
3092 prev->xbfi_next = free->xbfi_next;
3094 flist->xbf_first = free->xbfi_next;
3096 kmem_zone_free(xfs_bmap_free_item_zone, free);
3100 * Convert an extents-format file into a btree-format file.
3101 * The new file will have a root block (in the inode) and a single child block.
3103 STATIC int /* error */
3104 xfs_bmap_extents_to_btree(
3105 xfs_trans_t *tp, /* transaction pointer */
3106 xfs_inode_t *ip, /* incore inode pointer */
3107 xfs_fsblock_t *firstblock, /* first-block-allocated */
3108 xfs_bmap_free_t *flist, /* blocks freed in xaction */
3109 xfs_btree_cur_t **curp, /* cursor returned to caller */
3110 int wasdel, /* converting a delayed alloc */
3111 int *logflagsp, /* inode logging flags */
3112 int whichfork) /* data or attr fork */
3114 struct xfs_btree_block *ablock; /* allocated (child) bt block */
3115 xfs_buf_t *abp; /* buffer for ablock */
3116 xfs_alloc_arg_t args; /* allocation arguments */
3117 xfs_bmbt_rec_t *arp; /* child record pointer */
3118 struct xfs_btree_block *block; /* btree root block */
3119 xfs_btree_cur_t *cur; /* bmap btree cursor */
3120 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
3121 int error; /* error return value */
3122 xfs_extnum_t i, cnt; /* extent record index */
3123 xfs_ifork_t *ifp; /* inode fork pointer */
3124 xfs_bmbt_key_t *kp; /* root block key pointer */
3125 xfs_mount_t *mp; /* mount structure */
3126 xfs_extnum_t nextents; /* number of file extents */
3127 xfs_bmbt_ptr_t *pp; /* root block address pointer */
3129 ifp = XFS_IFORK_PTR(ip, whichfork);
3130 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3131 ASSERT(ifp->if_ext_max ==
3132 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
3134 * Make space in the inode incore.
3136 xfs_iroot_realloc(ip, 1, whichfork);
3137 ifp->if_flags |= XFS_IFBROOT;
3142 block = ifp->if_broot;
3143 block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3144 block->bb_level = cpu_to_be16(1);
3145 block->bb_numrecs = cpu_to_be16(1);
3146 block->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3147 block->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3150 * Need a cursor. Can't allocate until bb_level is filled in.
3153 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
3154 cur->bc_private.b.firstblock = *firstblock;
3155 cur->bc_private.b.flist = flist;
3156 cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3158 * Convert to a btree with two levels, one record in root.
3160 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3163 args.firstblock = *firstblock;
3164 if (*firstblock == NULLFSBLOCK) {
3165 args.type = XFS_ALLOCTYPE_START_BNO;
3166 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3167 } else if (flist->xbf_low) {
3168 args.type = XFS_ALLOCTYPE_START_BNO;
3169 args.fsbno = *firstblock;
3171 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3172 args.fsbno = *firstblock;
3174 args.minlen = args.maxlen = args.prod = 1;
3175 args.total = args.minleft = args.alignment = args.mod = args.isfl =
3176 args.minalignslop = 0;
3177 args.wasdel = wasdel;
3179 if ((error = xfs_alloc_vextent(&args))) {
3180 xfs_iroot_realloc(ip, -1, whichfork);
3181 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3185 * Allocation can't fail, the space was reserved.
3187 ASSERT(args.fsbno != NULLFSBLOCK);
3188 ASSERT(*firstblock == NULLFSBLOCK ||
3189 args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3191 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3192 *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3193 cur->bc_private.b.allocated++;
3194 ip->i_d.di_nblocks++;
3195 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3196 abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3198 * Fill in the child block.
3200 ablock = XFS_BUF_TO_BLOCK(abp);
3201 ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3202 ablock->bb_level = 0;
3203 ablock->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3204 ablock->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3205 arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3206 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3207 for (cnt = i = 0; i < nextents; i++) {
3208 ep = xfs_iext_get_ext(ifp, i);
3209 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
3210 arp->l0 = cpu_to_be64(ep->l0);
3211 arp->l1 = cpu_to_be64(ep->l1);
3215 ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3216 xfs_btree_set_numrecs(ablock, cnt);
3219 * Fill in the root key and pointer.
3221 kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
3222 arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3223 kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
3224 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
3225 be16_to_cpu(block->bb_level)));
3226 *pp = cpu_to_be64(args.fsbno);
3229 * Do all this logging at the end so that
3230 * the root is at the right level.
3232 xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
3233 xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3234 ASSERT(*curp == NULL);
3236 *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
3241 * Calculate the default attribute fork offset for newly created inodes.
3244 xfs_default_attroffset(
3245 struct xfs_inode *ip)
3247 struct xfs_mount *mp = ip->i_mount;
3250 if (mp->m_sb.sb_inodesize == 256) {
3251 offset = XFS_LITINO(mp) -
3252 XFS_BMDR_SPACE_CALC(MINABTPTRS);
3254 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
3257 ASSERT(offset < XFS_LITINO(mp));
3262 * Helper routine to reset inode di_forkoff field when switching
3263 * attribute fork from local to extent format - we reset it where
3264 * possible to make space available for inline data fork extents.
3267 xfs_bmap_forkoff_reset(
3272 if (whichfork == XFS_ATTR_FORK &&
3273 ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
3274 ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
3275 ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
3276 uint dfl_forkoff = xfs_default_attroffset(ip) >> 3;
3278 if (dfl_forkoff > ip->i_d.di_forkoff) {
3279 ip->i_d.di_forkoff = dfl_forkoff;
3280 ip->i_df.if_ext_max =
3281 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t);
3282 ip->i_afp->if_ext_max =
3283 XFS_IFORK_ASIZE(ip) / sizeof(xfs_bmbt_rec_t);
3289 * Convert a local file to an extents file.
3290 * This code is out of bounds for data forks of regular files,
3291 * since the file data needs to get logged so things will stay consistent.
3292 * (The bmap-level manipulations are ok, though).
3294 STATIC int /* error */
3295 xfs_bmap_local_to_extents(
3296 xfs_trans_t *tp, /* transaction pointer */
3297 xfs_inode_t *ip, /* incore inode pointer */
3298 xfs_fsblock_t *firstblock, /* first block allocated in xaction */
3299 xfs_extlen_t total, /* total blocks needed by transaction */
3300 int *logflagsp, /* inode logging flags */
3301 int whichfork) /* data or attr fork */
3303 int error; /* error return value */
3304 int flags; /* logging flags returned */
3305 xfs_ifork_t *ifp; /* inode fork pointer */
3308 * We don't want to deal with the case of keeping inode data inline yet.
3309 * So sending the data fork of a regular inode is invalid.
3311 ASSERT(!(S_ISREG(ip->i_d.di_mode) && whichfork == XFS_DATA_FORK));
3312 ifp = XFS_IFORK_PTR(ip, whichfork);
3313 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3316 if (ifp->if_bytes) {
3317 xfs_alloc_arg_t args; /* allocation arguments */
3318 xfs_buf_t *bp; /* buffer for extent block */
3319 xfs_bmbt_rec_host_t *ep;/* extent record pointer */
3322 args.mp = ip->i_mount;
3323 args.firstblock = *firstblock;
3324 ASSERT((ifp->if_flags &
3325 (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == XFS_IFINLINE);
3327 * Allocate a block. We know we need only one, since the
3328 * file currently fits in an inode.
3330 if (*firstblock == NULLFSBLOCK) {
3331 args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3332 args.type = XFS_ALLOCTYPE_START_BNO;
3334 args.fsbno = *firstblock;
3335 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3338 args.mod = args.minleft = args.alignment = args.wasdel =
3339 args.isfl = args.minalignslop = 0;
3340 args.minlen = args.maxlen = args.prod = 1;
3341 if ((error = xfs_alloc_vextent(&args)))
3344 * Can't fail, the space was reserved.
3346 ASSERT(args.fsbno != NULLFSBLOCK);
3347 ASSERT(args.len == 1);
3348 *firstblock = args.fsbno;
3349 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3350 memcpy(bp->b_addr, ifp->if_u1.if_data, ifp->if_bytes);
3351 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3352 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3353 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3354 xfs_iext_add(ifp, 0, 1);
3355 ep = xfs_iext_get_ext(ifp, 0);
3356 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3357 trace_xfs_bmap_post_update(ip, 0,
3358 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
3360 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3361 ip->i_d.di_nblocks = 1;
3362 xfs_trans_mod_dquot_byino(tp, ip,
3363 XFS_TRANS_DQ_BCOUNT, 1L);
3364 flags |= xfs_ilog_fext(whichfork);
3366 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3367 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3369 ifp->if_flags &= ~XFS_IFINLINE;
3370 ifp->if_flags |= XFS_IFEXTENTS;
3371 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3372 flags |= XFS_ILOG_CORE;
3379 * Search the extent records for the entry containing block bno.
3380 * If bno lies in a hole, point to the next entry. If bno lies
3381 * past eof, *eofp will be set, and *prevp will contain the last
3382 * entry (null if none). Else, *lastxp will be set to the index
3383 * of the found entry; *gotp will contain the entry.
3385 STATIC xfs_bmbt_rec_host_t * /* pointer to found extent entry */
3386 xfs_bmap_search_multi_extents(
3387 xfs_ifork_t *ifp, /* inode fork pointer */
3388 xfs_fileoff_t bno, /* block number searched for */
3389 int *eofp, /* out: end of file found */
3390 xfs_extnum_t *lastxp, /* out: last extent index */
3391 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
3392 xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
3394 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
3395 xfs_extnum_t lastx; /* last extent index */
3398 * Initialize the extent entry structure to catch access to
3399 * uninitialized br_startblock field.
3401 gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
3402 gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3403 gotp->br_state = XFS_EXT_INVALID;
3405 gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
3407 gotp->br_startblock = 0xffffa5a5;
3409 prevp->br_startoff = NULLFILEOFF;
3411 ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
3413 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
3415 if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
3416 xfs_bmbt_get_all(ep, gotp);
3430 * Search the extents list for the inode, for the extent containing bno.
3431 * If bno lies in a hole, point to the next entry. If bno lies past eof,
3432 * *eofp will be set, and *prevp will contain the last entry (null if none).
3433 * Else, *lastxp will be set to the index of the found
3434 * entry; *gotp will contain the entry.
3436 STATIC xfs_bmbt_rec_host_t * /* pointer to found extent entry */
3437 xfs_bmap_search_extents(
3438 xfs_inode_t *ip, /* incore inode pointer */
3439 xfs_fileoff_t bno, /* block number searched for */
3440 int fork, /* data or attr fork */
3441 int *eofp, /* out: end of file found */
3442 xfs_extnum_t *lastxp, /* out: last extent index */
3443 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
3444 xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
3446 xfs_ifork_t *ifp; /* inode fork pointer */
3447 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
3449 XFS_STATS_INC(xs_look_exlist);
3450 ifp = XFS_IFORK_PTR(ip, fork);
3452 ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
3454 if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
3455 !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
3456 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
3457 "Access to block zero in inode %llu "
3458 "start_block: %llx start_off: %llx "
3459 "blkcnt: %llx extent-state: %x lastx: %x\n",
3460 (unsigned long long)ip->i_ino,
3461 (unsigned long long)gotp->br_startblock,
3462 (unsigned long long)gotp->br_startoff,
3463 (unsigned long long)gotp->br_blockcount,
3464 gotp->br_state, *lastxp);
3465 *lastxp = NULLEXTNUM;
3473 * Compute the worst-case number of indirect blocks that will be used
3474 * for ip's delayed extent of length "len".
3476 STATIC xfs_filblks_t
3477 xfs_bmap_worst_indlen(
3478 xfs_inode_t *ip, /* incore inode pointer */
3479 xfs_filblks_t len) /* delayed extent length */
3481 int level; /* btree level number */
3482 int maxrecs; /* maximum record count at this level */
3483 xfs_mount_t *mp; /* mount structure */
3484 xfs_filblks_t rval; /* return value */
3487 maxrecs = mp->m_bmap_dmxr[0];
3488 for (level = 0, rval = 0;
3489 level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3492 do_div(len, maxrecs);
3495 return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3498 maxrecs = mp->m_bmap_dmxr[1];
3504 * Convert inode from non-attributed to attributed.
3505 * Must not be in a transaction, ip must not be locked.
3507 int /* error code */
3508 xfs_bmap_add_attrfork(
3509 xfs_inode_t *ip, /* incore inode pointer */
3510 int size, /* space new attribute needs */
3511 int rsvd) /* xact may use reserved blks */
3513 xfs_fsblock_t firstblock; /* 1st block/ag allocated */
3514 xfs_bmap_free_t flist; /* freed extent records */
3515 xfs_mount_t *mp; /* mount structure */
3516 xfs_trans_t *tp; /* transaction pointer */
3517 int blks; /* space reservation */
3518 int version = 1; /* superblock attr version */
3519 int committed; /* xaction was committed */
3520 int logflags; /* logging flags */
3521 int error; /* error return value */
3523 ASSERT(XFS_IFORK_Q(ip) == 0);
3524 ASSERT(ip->i_df.if_ext_max ==
3525 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3528 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3529 tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3530 blks = XFS_ADDAFORK_SPACE_RES(mp);
3532 tp->t_flags |= XFS_TRANS_RESERVE;
3533 if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3534 XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3536 xfs_ilock(ip, XFS_ILOCK_EXCL);
3537 error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
3538 XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3539 XFS_QMOPT_RES_REGBLKS);
3541 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3542 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3545 if (XFS_IFORK_Q(ip))
3547 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3549 * For inodes coming from pre-6.2 filesystems.
3551 ASSERT(ip->i_d.di_aformat == 0);
3552 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
3554 ASSERT(ip->i_d.di_anextents == 0);
3556 xfs_trans_ijoin_ref(tp, ip, XFS_ILOCK_EXCL);
3557 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3559 switch (ip->i_d.di_format) {
3560 case XFS_DINODE_FMT_DEV:
3561 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
3563 case XFS_DINODE_FMT_UUID:
3564 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
3566 case XFS_DINODE_FMT_LOCAL:
3567 case XFS_DINODE_FMT_EXTENTS:
3568 case XFS_DINODE_FMT_BTREE:
3569 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
3570 if (!ip->i_d.di_forkoff)
3571 ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
3572 else if (mp->m_flags & XFS_MOUNT_ATTR2)
3577 error = XFS_ERROR(EINVAL);
3580 ip->i_df.if_ext_max =
3581 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3582 ASSERT(ip->i_afp == NULL);
3583 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
3584 ip->i_afp->if_ext_max =
3585 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3586 ip->i_afp->if_flags = XFS_IFEXTENTS;
3588 xfs_bmap_init(&flist, &firstblock);
3589 switch (ip->i_d.di_format) {
3590 case XFS_DINODE_FMT_LOCAL:
3591 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
3594 case XFS_DINODE_FMT_EXTENTS:
3595 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
3598 case XFS_DINODE_FMT_BTREE:
3599 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
3607 xfs_trans_log_inode(tp, ip, logflags);
3610 if (!xfs_sb_version_hasattr(&mp->m_sb) ||
3611 (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
3612 __int64_t sbfields = 0;
3614 spin_lock(&mp->m_sb_lock);
3615 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
3616 xfs_sb_version_addattr(&mp->m_sb);
3617 sbfields |= XFS_SB_VERSIONNUM;
3619 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
3620 xfs_sb_version_addattr2(&mp->m_sb);
3621 sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
3624 spin_unlock(&mp->m_sb_lock);
3625 xfs_mod_sb(tp, sbfields);
3627 spin_unlock(&mp->m_sb_lock);
3629 if ((error = xfs_bmap_finish(&tp, &flist, &committed)))
3631 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3632 ASSERT(ip->i_df.if_ext_max ==
3633 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3636 xfs_bmap_cancel(&flist);
3638 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3640 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
3641 ASSERT(ip->i_df.if_ext_max ==
3642 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3647 * Add the extent to the list of extents to be free at transaction end.
3648 * The list is maintained sorted (by block number).
3653 xfs_fsblock_t bno, /* fs block number of extent */
3654 xfs_filblks_t len, /* length of extent */
3655 xfs_bmap_free_t *flist, /* list of extents */
3656 xfs_mount_t *mp) /* mount point structure */
3658 xfs_bmap_free_item_t *cur; /* current (next) element */
3659 xfs_bmap_free_item_t *new; /* new element */
3660 xfs_bmap_free_item_t *prev; /* previous element */
3662 xfs_agnumber_t agno;
3663 xfs_agblock_t agbno;
3665 ASSERT(bno != NULLFSBLOCK);
3667 ASSERT(len <= MAXEXTLEN);
3668 ASSERT(!isnullstartblock(bno));
3669 agno = XFS_FSB_TO_AGNO(mp, bno);
3670 agbno = XFS_FSB_TO_AGBNO(mp, bno);
3671 ASSERT(agno < mp->m_sb.sb_agcount);
3672 ASSERT(agbno < mp->m_sb.sb_agblocks);
3673 ASSERT(len < mp->m_sb.sb_agblocks);
3674 ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
3676 ASSERT(xfs_bmap_free_item_zone != NULL);
3677 new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
3678 new->xbfi_startblock = bno;
3679 new->xbfi_blockcount = (xfs_extlen_t)len;
3680 for (prev = NULL, cur = flist->xbf_first;
3682 prev = cur, cur = cur->xbfi_next) {
3683 if (cur->xbfi_startblock >= bno)
3687 prev->xbfi_next = new;
3689 flist->xbf_first = new;
3690 new->xbfi_next = cur;
3695 * Compute and fill in the value of the maximum depth of a bmap btree
3696 * in this filesystem. Done once, during mount.
3699 xfs_bmap_compute_maxlevels(
3700 xfs_mount_t *mp, /* file system mount structure */
3701 int whichfork) /* data or attr fork */
3703 int level; /* btree level */
3704 uint maxblocks; /* max blocks at this level */
3705 uint maxleafents; /* max leaf entries possible */
3706 int maxrootrecs; /* max records in root block */
3707 int minleafrecs; /* min records in leaf block */
3708 int minnoderecs; /* min records in node block */
3709 int sz; /* root block size */
3712 * The maximum number of extents in a file, hence the maximum
3713 * number of leaf entries, is controlled by the type of di_nextents
3714 * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
3715 * (a signed 16-bit number, xfs_aextnum_t).
3717 * Note that we can no longer assume that if we are in ATTR1 that
3718 * the fork offset of all the inodes will be
3719 * (xfs_default_attroffset(ip) >> 3) because we could have mounted
3720 * with ATTR2 and then mounted back with ATTR1, keeping the
3721 * di_forkoff's fixed but probably at various positions. Therefore,
3722 * for both ATTR1 and ATTR2 we have to assume the worst case scenario
3723 * of a minimum size available.
3725 if (whichfork == XFS_DATA_FORK) {
3726 maxleafents = MAXEXTNUM;
3727 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
3729 maxleafents = MAXAEXTNUM;
3730 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
3732 maxrootrecs = xfs_bmdr_maxrecs(mp, sz, 0);
3733 minleafrecs = mp->m_bmap_dmnr[0];
3734 minnoderecs = mp->m_bmap_dmnr[1];
3735 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
3736 for (level = 1; maxblocks > 1; level++) {
3737 if (maxblocks <= maxrootrecs)
3740 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
3742 mp->m_bm_maxlevels[whichfork] = level;
3746 * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
3747 * caller. Frees all the extents that need freeing, which must be done
3748 * last due to locking considerations. We never free any extents in
3749 * the first transaction.
3751 * Return 1 if the given transaction was committed and a new one
3752 * started, and 0 otherwise in the committed parameter.
3756 xfs_trans_t **tp, /* transaction pointer addr */
3757 xfs_bmap_free_t *flist, /* i/o: list extents to free */
3758 int *committed) /* xact committed or not */
3760 xfs_efd_log_item_t *efd; /* extent free data */
3761 xfs_efi_log_item_t *efi; /* extent free intention */
3762 int error; /* error return value */
3763 xfs_bmap_free_item_t *free; /* free extent item */
3764 unsigned int logres; /* new log reservation */
3765 unsigned int logcount; /* new log count */
3766 xfs_mount_t *mp; /* filesystem mount structure */
3767 xfs_bmap_free_item_t *next; /* next item on free list */
3768 xfs_trans_t *ntp; /* new transaction pointer */
3770 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
3771 if (flist->xbf_count == 0) {
3776 efi = xfs_trans_get_efi(ntp, flist->xbf_count);
3777 for (free = flist->xbf_first; free; free = free->xbfi_next)
3778 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
3779 free->xbfi_blockcount);
3780 logres = ntp->t_log_res;
3781 logcount = ntp->t_log_count;
3782 ntp = xfs_trans_dup(*tp);
3783 error = xfs_trans_commit(*tp, 0);
3787 * We have a new transaction, so we should return committed=1,
3788 * even though we're returning an error.
3794 * transaction commit worked ok so we can drop the extra ticket
3795 * reference that we gained in xfs_trans_dup()
3797 xfs_log_ticket_put(ntp->t_ticket);
3799 if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
3802 efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
3803 for (free = flist->xbf_first; free != NULL; free = next) {
3804 next = free->xbfi_next;
3805 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
3806 free->xbfi_blockcount))) {
3808 * The bmap free list will be cleaned up at a
3809 * higher level. The EFI will be canceled when
3810 * this transaction is aborted.
3811 * Need to force shutdown here to make sure it
3812 * happens, since this transaction may not be
3816 if (!XFS_FORCED_SHUTDOWN(mp))
3817 xfs_force_shutdown(mp,
3818 (error == EFSCORRUPTED) ?
3819 SHUTDOWN_CORRUPT_INCORE :
3820 SHUTDOWN_META_IO_ERROR);
3823 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
3824 free->xbfi_blockcount);
3825 xfs_bmap_del_free(flist, NULL, free);
3831 * Free up any items left in the list.
3835 xfs_bmap_free_t *flist) /* list of bmap_free_items */
3837 xfs_bmap_free_item_t *free; /* free list item */
3838 xfs_bmap_free_item_t *next;
3840 if (flist->xbf_count == 0)
3842 ASSERT(flist->xbf_first != NULL);
3843 for (free = flist->xbf_first; free; free = next) {
3844 next = free->xbfi_next;
3845 xfs_bmap_del_free(flist, NULL, free);
3847 ASSERT(flist->xbf_count == 0);
3851 * Returns the file-relative block number of the first unused block(s)
3852 * in the file with at least "len" logically contiguous blocks free.
3853 * This is the lowest-address hole if the file has holes, else the first block
3854 * past the end of file.
3855 * Return 0 if the file is currently local (in-inode).
3858 xfs_bmap_first_unused(
3859 xfs_trans_t *tp, /* transaction pointer */
3860 xfs_inode_t *ip, /* incore inode */
3861 xfs_extlen_t len, /* size of hole to find */
3862 xfs_fileoff_t *first_unused, /* unused block */
3863 int whichfork) /* data or attr fork */
3865 int error; /* error return value */
3866 int idx; /* extent record index */
3867 xfs_ifork_t *ifp; /* inode fork pointer */
3868 xfs_fileoff_t lastaddr; /* last block number seen */
3869 xfs_fileoff_t lowest; /* lowest useful block */
3870 xfs_fileoff_t max; /* starting useful block */
3871 xfs_fileoff_t off; /* offset for this block */
3872 xfs_extnum_t nextents; /* number of extent entries */
3874 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
3875 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
3876 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3877 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3881 ifp = XFS_IFORK_PTR(ip, whichfork);
3882 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3883 (error = xfs_iread_extents(tp, ip, whichfork)))
3885 lowest = *first_unused;
3886 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3887 for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
3888 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
3889 off = xfs_bmbt_get_startoff(ep);
3891 * See if the hole before this extent will work.
3893 if (off >= lowest + len && off - max >= len) {
3894 *first_unused = max;
3897 lastaddr = off + xfs_bmbt_get_blockcount(ep);
3898 max = XFS_FILEOFF_MAX(lastaddr, lowest);
3900 *first_unused = max;
3905 * Returns the file-relative block number of the last block + 1 before
3906 * last_block (input value) in the file.
3907 * This is not based on i_size, it is based on the extent records.
3908 * Returns 0 for local files, as they do not have extent records.
3911 xfs_bmap_last_before(
3912 xfs_trans_t *tp, /* transaction pointer */
3913 xfs_inode_t *ip, /* incore inode */
3914 xfs_fileoff_t *last_block, /* last block */
3915 int whichfork) /* data or attr fork */
3917 xfs_fileoff_t bno; /* input file offset */
3918 int eof; /* hit end of file */
3919 xfs_bmbt_rec_host_t *ep; /* pointer to last extent */
3920 int error; /* error return value */
3921 xfs_bmbt_irec_t got; /* current extent value */
3922 xfs_ifork_t *ifp; /* inode fork pointer */
3923 xfs_extnum_t lastx; /* last extent used */
3924 xfs_bmbt_irec_t prev; /* previous extent value */
3926 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
3927 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
3928 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
3929 return XFS_ERROR(EIO);
3930 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3934 ifp = XFS_IFORK_PTR(ip, whichfork);
3935 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3936 (error = xfs_iread_extents(tp, ip, whichfork)))
3938 bno = *last_block - 1;
3939 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
3941 if (eof || xfs_bmbt_get_startoff(ep) > bno) {
3942 if (prev.br_startoff == NULLFILEOFF)
3945 *last_block = prev.br_startoff + prev.br_blockcount;
3948 * Otherwise *last_block is already the right answer.
3954 * Returns the file-relative block number of the first block past eof in
3955 * the file. This is not based on i_size, it is based on the extent records.
3956 * Returns 0 for local files, as they do not have extent records.
3959 xfs_bmap_last_offset(
3960 xfs_trans_t *tp, /* transaction pointer */
3961 xfs_inode_t *ip, /* incore inode */
3962 xfs_fileoff_t *last_block, /* last block */
3963 int whichfork) /* data or attr fork */
3965 xfs_bmbt_rec_host_t *ep; /* pointer to last extent */
3966 int error; /* error return value */
3967 xfs_ifork_t *ifp; /* inode fork pointer */
3968 xfs_extnum_t nextents; /* number of extent entries */
3970 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
3971 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
3972 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
3973 return XFS_ERROR(EIO);
3974 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3978 ifp = XFS_IFORK_PTR(ip, whichfork);
3979 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3980 (error = xfs_iread_extents(tp, ip, whichfork)))
3982 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3987 ep = xfs_iext_get_ext(ifp, nextents - 1);
3988 *last_block = xfs_bmbt_get_startoff(ep) + xfs_bmbt_get_blockcount(ep);
3993 * Returns whether the selected fork of the inode has exactly one
3994 * block or not. For the data fork we check this matches di_size,
3995 * implying the file's range is 0..bsize-1.
3997 int /* 1=>1 block, 0=>otherwise */
3999 xfs_inode_t *ip, /* incore inode */
4000 int whichfork) /* data or attr fork */
4002 xfs_bmbt_rec_host_t *ep; /* ptr to fork's extent */
4003 xfs_ifork_t *ifp; /* inode fork pointer */
4004 int rval; /* return value */
4005 xfs_bmbt_irec_t s; /* internal version of extent */
4008 if (whichfork == XFS_DATA_FORK) {
4009 return S_ISREG(ip->i_d.di_mode) ?
4010 (ip->i_size == ip->i_mount->m_sb.sb_blocksize) :
4011 (ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize);
4014 if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4016 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4018 ifp = XFS_IFORK_PTR(ip, whichfork);
4019 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4020 ep = xfs_iext_get_ext(ifp, 0);
4021 xfs_bmbt_get_all(ep, &s);
4022 rval = s.br_startoff == 0 && s.br_blockcount == 1;
4023 if (rval && whichfork == XFS_DATA_FORK)
4024 ASSERT(ip->i_size == ip->i_mount->m_sb.sb_blocksize);
4029 xfs_bmap_sanity_check(
4030 struct xfs_mount *mp,
4034 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
4036 if (block->bb_magic != cpu_to_be32(XFS_BMAP_MAGIC) ||
4037 be16_to_cpu(block->bb_level) != level ||
4038 be16_to_cpu(block->bb_numrecs) == 0 ||
4039 be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
4045 * Read in the extents to if_extents.
4046 * All inode fields are set up by caller, we just traverse the btree
4047 * and copy the records in. If the file system cannot contain unwritten
4048 * extents, the records are checked for no "state" flags.
4051 xfs_bmap_read_extents(
4052 xfs_trans_t *tp, /* transaction pointer */
4053 xfs_inode_t *ip, /* incore inode */
4054 int whichfork) /* data or attr fork */
4056 struct xfs_btree_block *block; /* current btree block */
4057 xfs_fsblock_t bno; /* block # of "block" */
4058 xfs_buf_t *bp; /* buffer for "block" */
4059 int error; /* error return value */
4060 xfs_exntfmt_t exntf; /* XFS_EXTFMT_NOSTATE, if checking */
4061 xfs_extnum_t i, j; /* index into the extents list */
4062 xfs_ifork_t *ifp; /* fork structure */
4063 int level; /* btree level, for checking */
4064 xfs_mount_t *mp; /* file system mount structure */
4065 __be64 *pp; /* pointer to block address */
4067 xfs_extnum_t room; /* number of entries there's room for */
4071 ifp = XFS_IFORK_PTR(ip, whichfork);
4072 exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4073 XFS_EXTFMT_INODE(ip);
4074 block = ifp->if_broot;
4076 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4078 level = be16_to_cpu(block->bb_level);
4080 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
4081 bno = be64_to_cpu(*pp);
4082 ASSERT(bno != NULLDFSBNO);
4083 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
4084 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
4086 * Go down the tree until leaf level is reached, following the first
4087 * pointer (leftmost) at each level.
4089 while (level-- > 0) {
4090 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4091 XFS_BMAP_BTREE_REF)))
4093 block = XFS_BUF_TO_BLOCK(bp);
4094 XFS_WANT_CORRUPTED_GOTO(
4095 xfs_bmap_sanity_check(mp, bp, level),
4099 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
4100 bno = be64_to_cpu(*pp);
4101 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
4102 xfs_trans_brelse(tp, bp);
4105 * Here with bp and block set to the leftmost leaf node in the tree.
4107 room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4110 * Loop over all leaf nodes. Copy information to the extent records.
4113 xfs_bmbt_rec_t *frp;
4114 xfs_fsblock_t nextbno;
4115 xfs_extnum_t num_recs;
4119 num_recs = xfs_btree_get_numrecs(block);
4120 if (unlikely(i + num_recs > room)) {
4121 ASSERT(i + num_recs <= room);
4122 xfs_warn(ip->i_mount,
4123 "corrupt dinode %Lu, (btree extents).",
4124 (unsigned long long) ip->i_ino);
4125 XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
4126 XFS_ERRLEVEL_LOW, ip->i_mount, block);
4129 XFS_WANT_CORRUPTED_GOTO(
4130 xfs_bmap_sanity_check(mp, bp, 0),
4133 * Read-ahead the next leaf block, if any.
4135 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
4136 if (nextbno != NULLFSBLOCK)
4137 xfs_btree_reada_bufl(mp, nextbno, 1);
4139 * Copy records into the extent records.
4141 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
4143 for (j = 0; j < num_recs; j++, i++, frp++) {
4144 xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
4145 trp->l0 = be64_to_cpu(frp->l0);
4146 trp->l1 = be64_to_cpu(frp->l1);
4148 if (exntf == XFS_EXTFMT_NOSTATE) {
4150 * Check all attribute bmap btree records and
4151 * any "older" data bmap btree records for a
4152 * set bit in the "extent flag" position.
4154 if (unlikely(xfs_check_nostate_extents(ifp,
4155 start, num_recs))) {
4156 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4162 xfs_trans_brelse(tp, bp);
4165 * If we've reached the end, stop.
4167 if (bno == NULLFSBLOCK)
4169 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4170 XFS_BMAP_BTREE_REF)))
4172 block = XFS_BUF_TO_BLOCK(bp);
4174 ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4175 ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4176 XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
4179 xfs_trans_brelse(tp, bp);
4180 return XFS_ERROR(EFSCORRUPTED);
4185 * Add bmap trace insert entries for all the contents of the extent records.
4188 xfs_bmap_trace_exlist(
4189 xfs_inode_t *ip, /* incore inode pointer */
4190 xfs_extnum_t cnt, /* count of entries in the list */
4191 int whichfork, /* data or attr fork */
4192 unsigned long caller_ip)
4194 xfs_extnum_t idx; /* extent record index */
4195 xfs_ifork_t *ifp; /* inode fork pointer */
4198 if (whichfork == XFS_ATTR_FORK)
4199 state |= BMAP_ATTRFORK;
4201 ifp = XFS_IFORK_PTR(ip, whichfork);
4202 ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4203 for (idx = 0; idx < cnt; idx++)
4204 trace_xfs_extlist(ip, idx, whichfork, caller_ip);
4208 * Validate that the bmbt_irecs being returned from bmapi are valid
4209 * given the callers original parameters. Specifically check the
4210 * ranges of the returned irecs to ensure that they only extent beyond
4211 * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4214 xfs_bmap_validate_ret(
4218 xfs_bmbt_irec_t *mval,
4222 int i; /* index to map values */
4224 ASSERT(ret_nmap <= nmap);
4226 for (i = 0; i < ret_nmap; i++) {
4227 ASSERT(mval[i].br_blockcount > 0);
4228 if (!(flags & XFS_BMAPI_ENTIRE)) {
4229 ASSERT(mval[i].br_startoff >= bno);
4230 ASSERT(mval[i].br_blockcount <= len);
4231 ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4234 ASSERT(mval[i].br_startoff < bno + len);
4235 ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4239 mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4240 mval[i].br_startoff);
4241 if (flags & XFS_BMAPI_WRITE)
4242 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4243 mval[i].br_startblock != HOLESTARTBLOCK);
4244 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4245 mval[i].br_state == XFS_EXT_UNWRITTEN);
4252 * Trim the returned map to the required bounds
4256 struct xfs_bmbt_irec *mval,
4257 struct xfs_bmbt_irec *got,
4265 if ((flags & XFS_BMAPI_ENTIRE) ||
4266 got->br_startoff + got->br_blockcount <= obno) {
4268 if (isnullstartblock(got->br_startblock))
4269 mval->br_startblock = DELAYSTARTBLOCK;
4275 ASSERT((*bno >= obno) || (n == 0));
4277 mval->br_startoff = *bno;
4278 if (isnullstartblock(got->br_startblock))
4279 mval->br_startblock = DELAYSTARTBLOCK;
4281 mval->br_startblock = got->br_startblock +
4282 (*bno - got->br_startoff);
4284 * Return the minimum of what we got and what we asked for for
4285 * the length. We can use the len variable here because it is
4286 * modified below and we could have been there before coming
4287 * here if the first part of the allocation didn't overlap what
4290 mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
4291 got->br_blockcount - (*bno - got->br_startoff));
4292 mval->br_state = got->br_state;
4293 ASSERT(mval->br_blockcount <= len);
4298 * Update and validate the extent map to return
4301 xfs_bmapi_update_map(
4302 struct xfs_bmbt_irec **map,
4310 xfs_bmbt_irec_t *mval = *map;
4312 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4313 ((mval->br_startoff + mval->br_blockcount) <= end));
4314 ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
4315 (mval->br_startoff < obno));
4317 *bno = mval->br_startoff + mval->br_blockcount;
4319 if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
4320 /* update previous map with new information */
4321 ASSERT(mval->br_startblock == mval[-1].br_startblock);
4322 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
4323 ASSERT(mval->br_state == mval[-1].br_state);
4324 mval[-1].br_blockcount = mval->br_blockcount;
4325 mval[-1].br_state = mval->br_state;
4326 } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
4327 mval[-1].br_startblock != DELAYSTARTBLOCK &&
4328 mval[-1].br_startblock != HOLESTARTBLOCK &&
4329 mval->br_startblock == mval[-1].br_startblock +
4330 mval[-1].br_blockcount &&
4331 ((flags & XFS_BMAPI_IGSTATE) ||
4332 mval[-1].br_state == mval->br_state)) {
4333 ASSERT(mval->br_startoff ==
4334 mval[-1].br_startoff + mval[-1].br_blockcount);
4335 mval[-1].br_blockcount += mval->br_blockcount;
4336 } else if (*n > 0 &&
4337 mval->br_startblock == DELAYSTARTBLOCK &&
4338 mval[-1].br_startblock == DELAYSTARTBLOCK &&
4339 mval->br_startoff ==
4340 mval[-1].br_startoff + mval[-1].br_blockcount) {
4341 mval[-1].br_blockcount += mval->br_blockcount;
4342 mval[-1].br_state = mval->br_state;
4343 } else if (!((*n == 0) &&
4344 ((mval->br_startoff + mval->br_blockcount) <=
4353 * Map file blocks to filesystem blocks without allocation.
4357 struct xfs_inode *ip,
4360 struct xfs_bmbt_irec *mval,
4364 struct xfs_mount *mp = ip->i_mount;
4365 struct xfs_ifork *ifp;
4366 struct xfs_bmbt_irec got;
4367 struct xfs_bmbt_irec prev;
4374 int whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4375 XFS_ATTR_FORK : XFS_DATA_FORK;
4378 ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
4379 XFS_BMAPI_IGSTATE)));
4381 if (unlikely(XFS_TEST_ERROR(
4382 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4383 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4384 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4385 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
4386 return XFS_ERROR(EFSCORRUPTED);
4389 if (XFS_FORCED_SHUTDOWN(mp))
4390 return XFS_ERROR(EIO);
4392 XFS_STATS_INC(xs_blk_mapr);
4394 ifp = XFS_IFORK_PTR(ip, whichfork);
4395 ASSERT(ifp->if_ext_max ==
4396 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4398 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4399 error = xfs_iread_extents(NULL, ip, whichfork);
4404 xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got, &prev);
4408 while (bno < end && n < *nmap) {
4409 /* Reading past eof, act as though there's a hole up to end. */
4411 got.br_startoff = end;
4412 if (got.br_startoff > bno) {
4413 /* Reading in a hole. */
4414 mval->br_startoff = bno;
4415 mval->br_startblock = HOLESTARTBLOCK;
4416 mval->br_blockcount =
4417 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4418 mval->br_state = XFS_EXT_NORM;
4419 bno += mval->br_blockcount;
4420 len -= mval->br_blockcount;
4426 /* set up the extent map to return. */
4427 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4428 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4430 /* If we're done, stop now. */
4431 if (bno >= end || n >= *nmap)
4434 /* Else go on to the next record. */
4435 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4436 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4445 xfs_bmapi_reserve_delalloc(
4446 struct xfs_inode *ip,
4449 struct xfs_bmbt_irec *got,
4450 struct xfs_bmbt_irec *prev,
4451 xfs_extnum_t *lastx,
4454 struct xfs_mount *mp = ip->i_mount;
4455 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4457 xfs_extlen_t indlen;
4458 xfs_fsblock_t firstblock = NULLFSBLOCK;
4459 struct xfs_btree_cur *cur = NULL;
4460 int tmp_logflags = 0;
4461 char rt = XFS_IS_REALTIME_INODE(ip);
4465 alen = XFS_FILBLKS_MIN(len, MAXEXTLEN);
4467 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4469 /* Figure out the extent size, adjust alen */
4470 extsz = xfs_get_extsz_hint(ip);
4473 * Make sure we don't exceed a single extent length when we
4474 * align the extent by reducing length we are going to
4475 * allocate by the maximum amount extent size aligment may
4478 alen = XFS_FILBLKS_MIN(len, MAXEXTLEN - (2 * extsz - 1));
4479 error = xfs_bmap_extsize_align(mp, got, prev, extsz, rt, eof,
4480 1, 0, &aoff, &alen);
4485 extsz = alen / mp->m_sb.sb_rextsize;
4488 * Make a transaction-less quota reservation for delayed allocation
4489 * blocks. This number gets adjusted later. We return if we haven't
4490 * allocated blocks already inside this loop.
4492 error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4493 rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4498 * Split changing sb for alen and indlen since they could be coming
4499 * from different places.
4501 indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4505 error = xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
4506 -((int64_t)extsz), 0);
4508 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4509 -((int64_t)alen), 0);
4513 goto out_unreserve_quota;
4515 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4516 -((int64_t)indlen), 0);
4518 goto out_unreserve_blocks;
4521 ip->i_delayed_blks += alen;
4523 got->br_startoff = aoff;
4524 got->br_startblock = nullstartblock(indlen);
4525 got->br_blockcount = alen;
4526 got->br_state = XFS_EXT_NORM;
4528 error = xfs_bmap_add_extent(NULL, ip, lastx, &cur, got, &firstblock,
4529 NULL, &tmp_logflags, XFS_DATA_FORK);
4531 ASSERT(!tmp_logflags);
4535 * Update our extent pointer, given that xfs_bmap_add_extent might
4536 * have merged it into one of the neighbouring ones.
4538 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), got);
4540 ASSERT(got->br_startoff <= aoff);
4541 ASSERT(got->br_startoff + got->br_blockcount >= aoff + alen);
4542 ASSERT(isnullstartblock(got->br_startblock));
4543 ASSERT(got->br_state == XFS_EXT_NORM);
4546 out_unreserve_blocks:
4548 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS, extsz, 0);
4550 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, alen, 0);
4551 out_unreserve_quota:
4552 if (XFS_IS_QUOTA_ON(mp))
4553 xfs_trans_unreserve_quota_nblks(NULL, ip, alen, 0, rt ?
4554 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4559 * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4563 struct xfs_inode *ip, /* incore inode */
4564 xfs_fileoff_t bno, /* starting file offs. mapped */
4565 xfs_filblks_t len, /* length to map in file */
4566 struct xfs_bmbt_irec *mval, /* output: map values */
4567 int *nmap, /* i/o: mval size/count */
4568 int flags) /* XFS_BMAPI_... */
4570 struct xfs_mount *mp = ip->i_mount;
4571 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4572 struct xfs_bmbt_irec got; /* current file extent record */
4573 struct xfs_bmbt_irec prev; /* previous file extent record */
4574 xfs_fileoff_t obno; /* old block number (offset) */
4575 xfs_fileoff_t end; /* end of mapped file region */
4576 xfs_extnum_t lastx; /* last useful extent number */
4577 int eof; /* we've hit the end of extents */
4578 int n = 0; /* current extent index */
4582 ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4583 ASSERT(!(flags & ~XFS_BMAPI_ENTIRE));
4585 if (unlikely(XFS_TEST_ERROR(
4586 (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4587 XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4588 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4589 XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW, mp);
4590 return XFS_ERROR(EFSCORRUPTED);
4593 if (XFS_FORCED_SHUTDOWN(mp))
4594 return XFS_ERROR(EIO);
4596 XFS_STATS_INC(xs_blk_mapw);
4598 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4599 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4604 xfs_bmap_search_extents(ip, bno, XFS_DATA_FORK, &eof, &lastx, &got, &prev);
4608 while (bno < end && n < *nmap) {
4609 if (eof || got.br_startoff > bno) {
4610 error = xfs_bmapi_reserve_delalloc(ip, bno, len, &got,
4611 &prev, &lastx, eof);
4621 /* set up the extent map to return. */
4622 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4623 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4625 /* If we're done, stop now. */
4626 if (bno >= end || n >= *nmap)
4629 /* Else go on to the next record. */
4631 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4632 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4643 * Map file blocks to filesystem blocks.
4644 * File range is given by the bno/len pair.
4645 * Adds blocks to file if a write ("flags & XFS_BMAPI_WRITE" set)
4646 * into a hole or past eof.
4647 * Only allocates blocks from a single allocation group,
4648 * to avoid locking problems.
4649 * The returned value in "firstblock" from the first call in a transaction
4650 * must be remembered and presented to subsequent calls in "firstblock".
4651 * An upper bound for the number of blocks to be allocated is supplied to
4652 * the first call in "total"; if no allocation group has that many free
4653 * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4657 xfs_trans_t *tp, /* transaction pointer */
4658 xfs_inode_t *ip, /* incore inode */
4659 xfs_fileoff_t bno, /* starting file offs. mapped */
4660 xfs_filblks_t len, /* length to map in file */
4661 int flags, /* XFS_BMAPI_... */
4662 xfs_fsblock_t *firstblock, /* first allocated block
4663 controls a.g. for allocs */
4664 xfs_extlen_t total, /* total blocks needed */
4665 xfs_bmbt_irec_t *mval, /* output: map values */
4666 int *nmap, /* i/o: mval size/count */
4667 xfs_bmap_free_t *flist) /* i/o: list extents to free */
4669 xfs_fsblock_t abno; /* allocated block number */
4670 xfs_extlen_t alen; /* allocated extent length */
4671 xfs_fileoff_t aoff; /* allocated file offset */
4672 xfs_bmalloca_t bma = { 0 }; /* args for xfs_bmap_alloc */
4673 xfs_btree_cur_t *cur; /* bmap btree cursor */
4674 xfs_fileoff_t end; /* end of mapped file region */
4675 int eof; /* we've hit the end of extents */
4676 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
4677 int error; /* error return */
4678 xfs_bmbt_irec_t got; /* current file extent record */
4679 xfs_ifork_t *ifp; /* inode fork pointer */
4680 xfs_extnum_t lastx; /* last useful extent number */
4681 int logflags; /* flags for transaction logging */
4682 xfs_extlen_t minleft; /* min blocks left after allocation */
4683 xfs_extlen_t minlen; /* min allocation size */
4684 xfs_mount_t *mp; /* xfs mount structure */
4685 int n; /* current extent index */
4686 int nallocs; /* number of extents alloc'd */
4687 xfs_fileoff_t obno; /* old block number (offset) */
4688 xfs_bmbt_irec_t prev; /* previous file extent record */
4689 int tmp_logflags; /* temp flags holder */
4690 int whichfork; /* data or attr fork */
4691 char inhole; /* current location is hole in file */
4692 char wasdelay; /* old extent was delayed */
4693 char wr; /* this is a write request */
4694 char rt; /* this is a realtime file */
4696 xfs_fileoff_t orig_bno; /* original block number value */
4697 int orig_flags; /* original flags arg value */
4698 xfs_filblks_t orig_len; /* original value of len arg */
4699 xfs_bmbt_irec_t *orig_mval; /* original value of mval */
4700 int orig_nmap; /* original value of *nmap */
4709 ASSERT(*nmap <= XFS_BMAP_MAX_NMAP || !(flags & XFS_BMAPI_WRITE));
4710 whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4711 XFS_ATTR_FORK : XFS_DATA_FORK;
4713 if (unlikely(XFS_TEST_ERROR(
4714 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4715 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4716 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4717 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4718 XFS_ERROR_REPORT("xfs_bmapi", XFS_ERRLEVEL_LOW, mp);
4719 return XFS_ERROR(EFSCORRUPTED);
4721 if (XFS_FORCED_SHUTDOWN(mp))
4722 return XFS_ERROR(EIO);
4723 rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4724 ifp = XFS_IFORK_PTR(ip, whichfork);
4725 ASSERT(ifp->if_ext_max ==
4726 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4727 if ((wr = (flags & XFS_BMAPI_WRITE)) != 0)
4728 XFS_STATS_INC(xs_blk_mapw);
4730 XFS_STATS_INC(xs_blk_mapr);
4732 * IGSTATE flag is used to combine extents which
4733 * differ only due to the state of the extents.
4734 * This technique is used from xfs_getbmap()
4735 * when the caller does not wish to see the
4736 * separation (which is the default).
4738 * This technique is also used when writing a
4739 * buffer which has been partially written,
4740 * (usually by being flushed during a chunkread),
4741 * to ensure one write takes place. This also
4742 * prevents a change in the xfs inode extents at
4743 * this time, intentionally. This change occurs
4744 * on completion of the write operation, in
4745 * xfs_strat_comp(), where the xfs_bmapi() call
4746 * is transactioned, and the extents combined.
4748 if ((flags & XFS_BMAPI_IGSTATE) && wr) /* if writing unwritten space */
4749 wr = 0; /* no allocations are allowed */
4753 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4755 if ((error = xfs_bmap_local_to_extents(tp, ip,
4756 firstblock, total, &logflags, whichfork)))
4759 if (wr && *firstblock == NULLFSBLOCK) {
4760 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4761 minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4766 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4767 (error = xfs_iread_extents(tp, ip, whichfork)))
4769 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4776 while (bno < end && n < *nmap) {
4778 * Reading past eof, act as though there's a hole
4782 got.br_startoff = end;
4783 inhole = eof || got.br_startoff > bno;
4784 wasdelay = wr && !inhole && isnullstartblock(got.br_startblock);
4786 * First, deal with the hole before the allocated space
4787 * that we found, if any.
4789 if (wr && (inhole || wasdelay)) {
4791 * For the wasdelay case, we could also just
4792 * allocate the stuff asked for in this bmap call
4793 * but that wouldn't be as good.
4796 alen = (xfs_extlen_t)got.br_blockcount;
4797 aoff = got.br_startoff;
4798 if (lastx != NULLEXTNUM && lastx) {
4799 ep = xfs_iext_get_ext(ifp, lastx - 1);
4800 xfs_bmbt_get_all(ep, &prev);
4803 alen = (xfs_extlen_t)
4804 XFS_FILBLKS_MIN(len, MAXEXTLEN);
4806 alen = (xfs_extlen_t)
4807 XFS_FILBLKS_MIN(alen,
4808 got.br_startoff - bno);
4811 minlen = (flags & XFS_BMAPI_CONTIG) ? alen : 1;
4814 * If first time, allocate and fill in
4815 * once-only bma fields.
4817 if (bma.ip == NULL) {
4825 /* Indicate if this is the first user data
4826 * in the file, or just any user data.
4828 if (!(flags & XFS_BMAPI_METADATA)) {
4829 bma.userdata = (aoff == 0) ?
4830 XFS_ALLOC_INITIAL_USER_DATA :
4834 * Fill in changeable bma fields.
4837 bma.firstblock = *firstblock;
4840 bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4841 bma.wasdel = wasdelay;
4842 bma.minlen = minlen;
4843 bma.low = flist->xbf_low;
4844 bma.minleft = minleft;
4846 * Only want to do the alignment at the
4847 * eof if it is userdata and allocation length
4848 * is larger than a stripe unit.
4850 if (mp->m_dalign && alen >= mp->m_dalign &&
4851 (!(flags & XFS_BMAPI_METADATA)) &&
4852 (whichfork == XFS_DATA_FORK)) {
4853 if ((error = xfs_bmap_isaeof(ip, aoff,
4854 whichfork, &bma.aeof)))
4861 if ((error = xfs_bmap_alloc(&bma)))
4864 * Copy out result fields.
4867 if ((flist->xbf_low = bma.low))
4871 ASSERT(*firstblock == NULLFSBLOCK ||
4872 XFS_FSB_TO_AGNO(mp, *firstblock) ==
4873 XFS_FSB_TO_AGNO(mp, bma.firstblock) ||
4875 XFS_FSB_TO_AGNO(mp, *firstblock) <
4876 XFS_FSB_TO_AGNO(mp, bma.firstblock)));
4877 *firstblock = bma.firstblock;
4879 cur->bc_private.b.firstblock =
4881 if (abno == NULLFSBLOCK)
4883 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
4884 cur = xfs_bmbt_init_cursor(mp, tp,
4886 cur->bc_private.b.firstblock =
4888 cur->bc_private.b.flist = flist;
4891 * Bump the number of extents we've allocated
4897 cur->bc_private.b.flags =
4898 wasdelay ? XFS_BTCUR_BPRV_WASDEL : 0;
4899 got.br_startoff = aoff;
4900 got.br_startblock = abno;
4901 got.br_blockcount = alen;
4902 got.br_state = XFS_EXT_NORM; /* assume normal */
4904 * Determine state of extent, and the filesystem.
4905 * A wasdelay extent has been initialized, so
4906 * shouldn't be flagged as unwritten.
4908 if (wr && xfs_sb_version_hasextflgbit(&mp->m_sb)) {
4909 if (!wasdelay && (flags & XFS_BMAPI_PREALLOC))
4910 got.br_state = XFS_EXT_UNWRITTEN;
4912 error = xfs_bmap_add_extent(tp, ip, &lastx, &cur, &got,
4913 firstblock, flist, &tmp_logflags,
4915 logflags |= tmp_logflags;
4918 ep = xfs_iext_get_ext(ifp, lastx);
4919 xfs_bmbt_get_all(ep, &got);
4920 ASSERT(got.br_startoff <= aoff);
4921 ASSERT(got.br_startoff + got.br_blockcount >=
4923 ASSERT(got.br_state == XFS_EXT_NORM ||
4924 got.br_state == XFS_EXT_UNWRITTEN);
4926 * Fall down into the found allocated space case.
4928 } else if (inhole) {
4930 * Reading in a hole.
4932 mval->br_startoff = bno;
4933 mval->br_startblock = HOLESTARTBLOCK;
4934 mval->br_blockcount =
4935 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4936 mval->br_state = XFS_EXT_NORM;
4937 bno += mval->br_blockcount;
4938 len -= mval->br_blockcount;
4944 /* Deal with the allocated space we found. */
4945 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4948 * Check if writing previously allocated but
4949 * unwritten extents.
4952 ((mval->br_state == XFS_EXT_UNWRITTEN &&
4953 ((flags & XFS_BMAPI_PREALLOC) == 0)) ||
4954 (mval->br_state == XFS_EXT_NORM &&
4955 ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_CONVERT)) ==
4956 (XFS_BMAPI_PREALLOC|XFS_BMAPI_CONVERT))))) {
4958 * Modify (by adding) the state flag, if writing.
4960 ASSERT(mval->br_blockcount <= len);
4961 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
4962 cur = xfs_bmbt_init_cursor(mp,
4964 cur->bc_private.b.firstblock =
4966 cur->bc_private.b.flist = flist;
4968 mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4970 : XFS_EXT_UNWRITTEN;
4971 error = xfs_bmap_add_extent(tp, ip, &lastx, &cur, mval,
4972 firstblock, flist, &tmp_logflags,
4974 logflags |= tmp_logflags;
4977 ep = xfs_iext_get_ext(ifp, lastx);
4978 xfs_bmbt_get_all(ep, &got);
4980 * We may have combined previously unwritten
4981 * space with written space, so generate
4984 if (mval->br_blockcount < len)
4988 /* update the extent map to return */
4989 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4992 * If we're done, stop now. Stop when we've allocated
4993 * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise
4994 * the transaction may get too big.
4996 if (bno >= end || n >= *nmap || nallocs >= *nmap)
4999 * Else go on to the next record.
5002 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t)) {
5003 ep = xfs_iext_get_ext(ifp, lastx);
5004 xfs_bmbt_get_all(ep, &got);
5011 * Transform from btree to extents, give it cur.
5013 if (tp && XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5014 XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5016 error = xfs_bmap_btree_to_extents(tp, ip, cur,
5017 &tmp_logflags, whichfork);
5018 logflags |= tmp_logflags;
5022 ASSERT(ifp->if_ext_max ==
5023 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5024 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
5025 XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max);
5029 * Log everything. Do this after conversion, there's no point in
5030 * logging the extent records if we've converted to btree format.
5032 if ((logflags & xfs_ilog_fext(whichfork)) &&
5033 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5034 logflags &= ~xfs_ilog_fext(whichfork);
5035 else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5036 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5037 logflags &= ~xfs_ilog_fbroot(whichfork);
5039 * Log whatever the flags say, even if error. Otherwise we might miss
5040 * detecting a case where the data is changed, there's an error,
5041 * and it's not logged so we don't shutdown when we should.
5045 xfs_trans_log_inode(tp, ip, logflags);
5049 ASSERT(*firstblock == NULLFSBLOCK ||
5050 XFS_FSB_TO_AGNO(mp, *firstblock) ==
5052 cur->bc_private.b.firstblock) ||
5054 XFS_FSB_TO_AGNO(mp, *firstblock) <
5056 cur->bc_private.b.firstblock)));
5057 *firstblock = cur->bc_private.b.firstblock;
5059 xfs_btree_del_cursor(cur,
5060 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5063 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
5069 * Unmap (remove) blocks from a file.
5070 * If nexts is nonzero then the number of extents to remove is limited to
5071 * that value. If not all extents in the block range can be removed then
5076 xfs_trans_t *tp, /* transaction pointer */
5077 struct xfs_inode *ip, /* incore inode */
5078 xfs_fileoff_t bno, /* starting offset to unmap */
5079 xfs_filblks_t len, /* length to unmap in file */
5080 int flags, /* misc flags */
5081 xfs_extnum_t nexts, /* number of extents max */
5082 xfs_fsblock_t *firstblock, /* first allocated block
5083 controls a.g. for allocs */
5084 xfs_bmap_free_t *flist, /* i/o: list extents to free */
5085 int *done) /* set if not done yet */
5087 xfs_btree_cur_t *cur; /* bmap btree cursor */
5088 xfs_bmbt_irec_t del; /* extent being deleted */
5089 int eof; /* is deleting at eof */
5090 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
5091 int error; /* error return value */
5092 xfs_extnum_t extno; /* extent number in list */
5093 xfs_bmbt_irec_t got; /* current extent record */
5094 xfs_ifork_t *ifp; /* inode fork pointer */
5095 int isrt; /* freeing in rt area */
5096 xfs_extnum_t lastx; /* last extent index used */
5097 int logflags; /* transaction logging flags */
5098 xfs_extlen_t mod; /* rt extent offset */
5099 xfs_mount_t *mp; /* mount structure */
5100 xfs_extnum_t nextents; /* number of file extents */
5101 xfs_bmbt_irec_t prev; /* previous extent record */
5102 xfs_fileoff_t start; /* first file offset deleted */
5103 int tmp_logflags; /* partial logging flags */
5104 int wasdel; /* was a delayed alloc extent */
5105 int whichfork; /* data or attribute fork */
5108 trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5110 whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5111 XFS_ATTR_FORK : XFS_DATA_FORK;
5112 ifp = XFS_IFORK_PTR(ip, whichfork);
5114 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5115 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5116 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5118 return XFS_ERROR(EFSCORRUPTED);
5121 if (XFS_FORCED_SHUTDOWN(mp))
5122 return XFS_ERROR(EIO);
5126 ASSERT(ifp->if_ext_max ==
5127 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5128 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5129 (error = xfs_iread_extents(tp, ip, whichfork)))
5131 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5132 if (nextents == 0) {
5136 XFS_STATS_INC(xs_blk_unmap);
5137 isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5139 bno = start + len - 1;
5140 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5144 * Check to see if the given block number is past the end of the
5145 * file, back up to the last block if so...
5148 ep = xfs_iext_get_ext(ifp, --lastx);
5149 xfs_bmbt_get_all(ep, &got);
5150 bno = got.br_startoff + got.br_blockcount - 1;
5153 if (ifp->if_flags & XFS_IFBROOT) {
5154 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5155 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5156 cur->bc_private.b.firstblock = *firstblock;
5157 cur->bc_private.b.flist = flist;
5158 cur->bc_private.b.flags = 0;
5162 while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5163 (nexts == 0 || extno < nexts)) {
5165 * Is the found extent after a hole in which bno lives?
5166 * Just back up to the previous extent, if so.
5168 if (got.br_startoff > bno) {
5171 ep = xfs_iext_get_ext(ifp, lastx);
5172 xfs_bmbt_get_all(ep, &got);
5175 * Is the last block of this extent before the range
5176 * we're supposed to delete? If so, we're done.
5178 bno = XFS_FILEOFF_MIN(bno,
5179 got.br_startoff + got.br_blockcount - 1);
5183 * Then deal with the (possibly delayed) allocated space
5188 wasdel = isnullstartblock(del.br_startblock);
5189 if (got.br_startoff < start) {
5190 del.br_startoff = start;
5191 del.br_blockcount -= start - got.br_startoff;
5193 del.br_startblock += start - got.br_startoff;
5195 if (del.br_startoff + del.br_blockcount > bno + 1)
5196 del.br_blockcount = bno + 1 - del.br_startoff;
5197 sum = del.br_startblock + del.br_blockcount;
5199 (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5201 * Realtime extent not lined up at the end.
5202 * The extent could have been split into written
5203 * and unwritten pieces, or we could just be
5204 * unmapping part of it. But we can't really
5205 * get rid of part of a realtime extent.
5207 if (del.br_state == XFS_EXT_UNWRITTEN ||
5208 !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5210 * This piece is unwritten, or we're not
5211 * using unwritten extents. Skip over it.
5214 bno -= mod > del.br_blockcount ?
5215 del.br_blockcount : mod;
5216 if (bno < got.br_startoff) {
5218 xfs_bmbt_get_all(xfs_iext_get_ext(
5224 * It's written, turn it unwritten.
5225 * This is better than zeroing it.
5227 ASSERT(del.br_state == XFS_EXT_NORM);
5228 ASSERT(xfs_trans_get_block_res(tp) > 0);
5230 * If this spans a realtime extent boundary,
5231 * chop it back to the start of the one we end at.
5233 if (del.br_blockcount > mod) {
5234 del.br_startoff += del.br_blockcount - mod;
5235 del.br_startblock += del.br_blockcount - mod;
5236 del.br_blockcount = mod;
5238 del.br_state = XFS_EXT_UNWRITTEN;
5239 error = xfs_bmap_add_extent(tp, ip, &lastx, &cur, &del,
5240 firstblock, flist, &logflags,
5246 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5248 * Realtime extent is lined up at the end but not
5249 * at the front. We'll get rid of full extents if
5252 mod = mp->m_sb.sb_rextsize - mod;
5253 if (del.br_blockcount > mod) {
5254 del.br_blockcount -= mod;
5255 del.br_startoff += mod;
5256 del.br_startblock += mod;
5257 } else if ((del.br_startoff == start &&
5258 (del.br_state == XFS_EXT_UNWRITTEN ||
5259 xfs_trans_get_block_res(tp) == 0)) ||
5260 !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5262 * Can't make it unwritten. There isn't
5263 * a full extent here so just skip it.
5265 ASSERT(bno >= del.br_blockcount);
5266 bno -= del.br_blockcount;
5267 if (got.br_startoff > bno) {
5269 ep = xfs_iext_get_ext(ifp,
5271 xfs_bmbt_get_all(ep, &got);
5275 } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5277 * This one is already unwritten.
5278 * It must have a written left neighbor.
5279 * Unwrite the killed part of that one and
5283 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5285 ASSERT(prev.br_state == XFS_EXT_NORM);
5286 ASSERT(!isnullstartblock(prev.br_startblock));
5287 ASSERT(del.br_startblock ==
5288 prev.br_startblock + prev.br_blockcount);
5289 if (prev.br_startoff < start) {
5290 mod = start - prev.br_startoff;
5291 prev.br_blockcount -= mod;
5292 prev.br_startblock += mod;
5293 prev.br_startoff = start;
5295 prev.br_state = XFS_EXT_UNWRITTEN;
5297 error = xfs_bmap_add_extent(tp, ip, &lastx,
5298 &cur, &prev, firstblock, flist,
5299 &logflags, XFS_DATA_FORK);
5304 ASSERT(del.br_state == XFS_EXT_NORM);
5305 del.br_state = XFS_EXT_UNWRITTEN;
5306 error = xfs_bmap_add_extent(tp, ip, &lastx,
5307 &cur, &del, firstblock, flist,
5308 &logflags, XFS_DATA_FORK);
5315 ASSERT(startblockval(del.br_startblock) > 0);
5316 /* Update realtime/data freespace, unreserve quota */
5318 xfs_filblks_t rtexts;
5320 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5321 do_div(rtexts, mp->m_sb.sb_rextsize);
5322 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5323 (int64_t)rtexts, 0);
5324 (void)xfs_trans_reserve_quota_nblks(NULL,
5325 ip, -((long)del.br_blockcount), 0,
5326 XFS_QMOPT_RES_RTBLKS);
5328 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5329 (int64_t)del.br_blockcount, 0);
5330 (void)xfs_trans_reserve_quota_nblks(NULL,
5331 ip, -((long)del.br_blockcount), 0,
5332 XFS_QMOPT_RES_REGBLKS);
5334 ip->i_delayed_blks -= del.br_blockcount;
5336 cur->bc_private.b.flags |=
5337 XFS_BTCUR_BPRV_WASDEL;
5339 cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5341 * If it's the case where the directory code is running
5342 * with no block reservation, and the deleted block is in
5343 * the middle of its extent, and the resulting insert
5344 * of an extent would cause transformation to btree format,
5345 * then reject it. The calling code will then swap
5346 * blocks around instead.
5347 * We have to do this now, rather than waiting for the
5348 * conversion to btree format, since the transaction
5351 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5352 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5353 XFS_IFORK_NEXTENTS(ip, whichfork) >= ifp->if_ext_max &&
5354 del.br_startoff > got.br_startoff &&
5355 del.br_startoff + del.br_blockcount <
5356 got.br_startoff + got.br_blockcount) {
5357 error = XFS_ERROR(ENOSPC);
5360 error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5361 &tmp_logflags, whichfork);
5362 logflags |= tmp_logflags;
5365 bno = del.br_startoff - 1;
5368 * If not done go on to the next (previous) record.
5370 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5372 ep = xfs_iext_get_ext(ifp, lastx);
5373 if (xfs_bmbt_get_startoff(ep) > bno) {
5375 ep = xfs_iext_get_ext(ifp,
5378 xfs_bmbt_get_all(ep, &got);
5383 *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5384 ASSERT(ifp->if_ext_max ==
5385 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5387 * Convert to a btree if necessary.
5389 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5390 XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
5391 ASSERT(cur == NULL);
5392 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5393 &cur, 0, &tmp_logflags, whichfork);
5394 logflags |= tmp_logflags;
5399 * transform from btree to extents, give it cur
5401 else if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5402 XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5403 ASSERT(cur != NULL);
5404 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5406 logflags |= tmp_logflags;
5411 * transform from extents to local?
5413 ASSERT(ifp->if_ext_max ==
5414 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5418 * Log everything. Do this after conversion, there's no point in
5419 * logging the extent records if we've converted to btree format.
5421 if ((logflags & xfs_ilog_fext(whichfork)) &&
5422 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5423 logflags &= ~xfs_ilog_fext(whichfork);
5424 else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5425 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5426 logflags &= ~xfs_ilog_fbroot(whichfork);
5428 * Log inode even in the error case, if the transaction
5429 * is dirty we'll need to shut down the filesystem.
5432 xfs_trans_log_inode(tp, ip, logflags);
5435 *firstblock = cur->bc_private.b.firstblock;
5436 cur->bc_private.b.allocated = 0;
5438 xfs_btree_del_cursor(cur,
5439 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5445 * returns 1 for success, 0 if we failed to map the extent.
5448 xfs_getbmapx_fix_eof_hole(
5449 xfs_inode_t *ip, /* xfs incore inode pointer */
5450 struct getbmapx *out, /* output structure */
5451 int prealloced, /* this is a file with
5452 * preallocated data space */
5453 __int64_t end, /* last block requested */
5454 xfs_fsblock_t startblock)
5457 xfs_mount_t *mp; /* file system mount point */
5458 xfs_ifork_t *ifp; /* inode fork pointer */
5459 xfs_extnum_t lastx; /* last extent pointer */
5460 xfs_fileoff_t fileblock;
5462 if (startblock == HOLESTARTBLOCK) {
5464 out->bmv_block = -1;
5465 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, ip->i_size));
5466 fixlen -= out->bmv_offset;
5467 if (prealloced && out->bmv_offset + out->bmv_length == end) {
5468 /* Came to hole at EOF. Trim it. */
5471 out->bmv_length = fixlen;
5474 if (startblock == DELAYSTARTBLOCK)
5475 out->bmv_block = -2;
5477 out->bmv_block = xfs_fsb_to_db(ip, startblock);
5478 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
5479 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
5480 if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
5481 (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
5482 out->bmv_oflags |= BMV_OF_LAST;
5489 * Get inode's extents as described in bmv, and format for output.
5490 * Calls formatter to fill the user's buffer until all extents
5491 * are mapped, until the passed-in bmv->bmv_count slots have
5492 * been filled, or until the formatter short-circuits the loop,
5493 * if it is tracking filled-in extents on its own.
5495 int /* error code */
5498 struct getbmapx *bmv, /* user bmap structure */
5499 xfs_bmap_format_t formatter, /* format to user */
5500 void *arg) /* formatter arg */
5502 __int64_t bmvend; /* last block requested */
5503 int error = 0; /* return value */
5504 __int64_t fixlen; /* length for -1 case */
5505 int i; /* extent number */
5506 int lock; /* lock state */
5507 xfs_bmbt_irec_t *map; /* buffer for user's data */
5508 xfs_mount_t *mp; /* file system mount point */
5509 int nex; /* # of user extents can do */
5510 int nexleft; /* # of user extents left */
5511 int subnex; /* # of bmapi's can do */
5512 int nmap; /* number of map entries */
5513 struct getbmapx *out; /* output structure */
5514 int whichfork; /* data or attr fork */
5515 int prealloced; /* this is a file with
5516 * preallocated data space */
5517 int iflags; /* interface flags */
5518 int bmapi_flags; /* flags for xfs_bmapi */
5522 iflags = bmv->bmv_iflags;
5523 whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5525 if (whichfork == XFS_ATTR_FORK) {
5526 if (XFS_IFORK_Q(ip)) {
5527 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5528 ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5529 ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5530 return XFS_ERROR(EINVAL);
5531 } else if (unlikely(
5532 ip->i_d.di_aformat != 0 &&
5533 ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5534 XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5536 return XFS_ERROR(EFSCORRUPTED);
5542 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5543 ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5544 ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5545 return XFS_ERROR(EINVAL);
5547 if (xfs_get_extsz_hint(ip) ||
5548 ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5550 fixlen = XFS_MAXIOFFSET(mp);
5553 fixlen = ip->i_size;
5557 if (bmv->bmv_length == -1) {
5558 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5560 max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
5561 } else if (bmv->bmv_length == 0) {
5562 bmv->bmv_entries = 0;
5564 } else if (bmv->bmv_length < 0) {
5565 return XFS_ERROR(EINVAL);
5568 nex = bmv->bmv_count - 1;
5570 return XFS_ERROR(EINVAL);
5571 bmvend = bmv->bmv_offset + bmv->bmv_length;
5574 if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
5575 return XFS_ERROR(ENOMEM);
5576 out = kmem_zalloc(bmv->bmv_count * sizeof(struct getbmapx), KM_MAYFAIL);
5578 return XFS_ERROR(ENOMEM);
5580 xfs_ilock(ip, XFS_IOLOCK_SHARED);
5581 if (whichfork == XFS_DATA_FORK && !(iflags & BMV_IF_DELALLOC)) {
5582 if (ip->i_delayed_blks || ip->i_size > ip->i_d.di_size) {
5583 error = xfs_flush_pages(ip, 0, -1, 0, FI_REMAPF);
5585 goto out_unlock_iolock;
5588 * even after flushing the inode, there can still be delalloc
5589 * blocks on the inode beyond EOF due to speculative
5590 * preallocation. These are not removed until the release
5591 * function is called or the inode is inactivated. Hence we
5592 * cannot assert here that ip->i_delayed_blks == 0.
5596 lock = xfs_ilock_map_shared(ip);
5599 * Don't let nex be bigger than the number of extents
5600 * we can have assuming alternating holes and real extents.
5602 if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5603 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5605 bmapi_flags = xfs_bmapi_aflag(whichfork);
5606 if (!(iflags & BMV_IF_PREALLOC))
5607 bmapi_flags |= XFS_BMAPI_IGSTATE;
5610 * Allocate enough space to handle "subnex" maps at a time.
5614 map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
5616 goto out_unlock_ilock;
5618 bmv->bmv_entries = 0;
5620 if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
5621 (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
5629 nmap = (nexleft > subnex) ? subnex : nexleft;
5630 error = xfs_bmapi_read(ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5631 XFS_BB_TO_FSB(mp, bmv->bmv_length),
5632 map, &nmap, bmapi_flags);
5635 ASSERT(nmap <= subnex);
5637 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5638 out[cur_ext].bmv_oflags = 0;
5639 if (map[i].br_state == XFS_EXT_UNWRITTEN)
5640 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
5641 else if (map[i].br_startblock == DELAYSTARTBLOCK)
5642 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
5643 out[cur_ext].bmv_offset =
5644 XFS_FSB_TO_BB(mp, map[i].br_startoff);
5645 out[cur_ext].bmv_length =
5646 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5647 out[cur_ext].bmv_unused1 = 0;
5648 out[cur_ext].bmv_unused2 = 0;
5649 ASSERT(((iflags & BMV_IF_DELALLOC) != 0) ||
5650 (map[i].br_startblock != DELAYSTARTBLOCK));
5651 if (map[i].br_startblock == HOLESTARTBLOCK &&
5652 whichfork == XFS_ATTR_FORK) {
5653 /* came to the end of attribute fork */
5654 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
5658 if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
5660 map[i].br_startblock))
5664 out[cur_ext].bmv_offset +
5665 out[cur_ext].bmv_length;
5667 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
5670 * In case we don't want to return the hole,
5671 * don't increase cur_ext so that we can reuse
5672 * it in the next loop.
5674 if ((iflags & BMV_IF_NO_HOLES) &&
5675 map[i].br_startblock == HOLESTARTBLOCK) {
5676 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
5684 } while (nmap && nexleft && bmv->bmv_length);
5689 xfs_iunlock_map_shared(ip, lock);
5691 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5693 for (i = 0; i < cur_ext; i++) {
5694 int full = 0; /* user array is full */
5696 /* format results & advance arg */
5697 error = formatter(&arg, &out[i], &full);
5707 * Check the last inode extent to determine whether this allocation will result
5708 * in blocks being allocated at the end of the file. When we allocate new data
5709 * blocks at the end of the file which do not start at the previous data block,
5710 * we will try to align the new blocks at stripe unit boundaries.
5712 STATIC int /* error */
5714 xfs_inode_t *ip, /* incore inode pointer */
5715 xfs_fileoff_t off, /* file offset in fsblocks */
5716 int whichfork, /* data or attribute fork */
5717 char *aeof) /* return value */
5719 int error; /* error return value */
5720 xfs_ifork_t *ifp; /* inode fork pointer */
5721 xfs_bmbt_rec_host_t *lastrec; /* extent record pointer */
5722 xfs_extnum_t nextents; /* number of file extents */
5723 xfs_bmbt_irec_t s; /* expanded extent record */
5725 ASSERT(whichfork == XFS_DATA_FORK);
5726 ifp = XFS_IFORK_PTR(ip, whichfork);
5727 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5728 (error = xfs_iread_extents(NULL, ip, whichfork)))
5730 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5731 if (nextents == 0) {
5736 * Go to the last extent
5738 lastrec = xfs_iext_get_ext(ifp, nextents - 1);
5739 xfs_bmbt_get_all(lastrec, &s);
5741 * Check we are allocating in the last extent (for delayed allocations)
5742 * or past the last extent for non-delayed allocations.
5744 *aeof = (off >= s.br_startoff &&
5745 off < s.br_startoff + s.br_blockcount &&
5746 isnullstartblock(s.br_startblock)) ||
5747 off >= s.br_startoff + s.br_blockcount;
5752 * Check if the endoff is outside the last extent. If so the caller will grow
5753 * the allocation to a stripe unit boundary.
5757 xfs_inode_t *ip, /* incore inode pointer */
5758 xfs_fileoff_t endoff, /* file offset in fsblocks */
5759 int whichfork, /* data or attribute fork */
5760 int *eof) /* result value */
5762 xfs_fsblock_t blockcount; /* extent block count */
5763 int error; /* error return value */
5764 xfs_ifork_t *ifp; /* inode fork pointer */
5765 xfs_bmbt_rec_host_t *lastrec; /* extent record pointer */
5766 xfs_extnum_t nextents; /* number of file extents */
5767 xfs_fileoff_t startoff; /* extent starting file offset */
5769 ASSERT(whichfork == XFS_DATA_FORK);
5770 ifp = XFS_IFORK_PTR(ip, whichfork);
5771 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5772 (error = xfs_iread_extents(NULL, ip, whichfork)))
5774 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5775 if (nextents == 0) {
5780 * Go to the last extent
5782 lastrec = xfs_iext_get_ext(ifp, nextents - 1);
5783 startoff = xfs_bmbt_get_startoff(lastrec);
5784 blockcount = xfs_bmbt_get_blockcount(lastrec);
5785 *eof = endoff >= startoff + blockcount;
5790 STATIC struct xfs_buf *
5792 struct xfs_btree_cur *cur,
5795 struct xfs_log_item_desc *lidp;
5801 for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
5802 if (!cur->bc_bufs[i])
5804 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
5805 return cur->bc_bufs[i];
5808 /* Chase down all the log items to see if the bp is there */
5809 list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
5810 struct xfs_buf_log_item *bip;
5811 bip = (struct xfs_buf_log_item *)lidp->lid_item;
5812 if (bip->bli_item.li_type == XFS_LI_BUF &&
5813 XFS_BUF_ADDR(bip->bli_buf) == bno)
5814 return bip->bli_buf;
5822 struct xfs_btree_block *block,
5828 __be64 *pp, *thispa; /* pointer to block address */
5829 xfs_bmbt_key_t *prevp, *keyp;
5831 ASSERT(be16_to_cpu(block->bb_level) > 0);
5834 for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
5835 dmxr = mp->m_bmap_dmxr[0];
5836 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
5839 ASSERT(be64_to_cpu(prevp->br_startoff) <
5840 be64_to_cpu(keyp->br_startoff));
5845 * Compare the block numbers to see if there are dups.
5848 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
5850 pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
5852 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
5854 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
5856 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
5857 if (*thispa == *pp) {
5858 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
5860 (unsigned long long)be64_to_cpu(*thispa));
5861 panic("%s: ptrs are equal in node\n",
5869 * Check that the extents for the inode ip are in the right order in all
5874 xfs_bmap_check_leaf_extents(
5875 xfs_btree_cur_t *cur, /* btree cursor or null */
5876 xfs_inode_t *ip, /* incore inode pointer */
5877 int whichfork) /* data or attr fork */
5879 struct xfs_btree_block *block; /* current btree block */
5880 xfs_fsblock_t bno; /* block # of "block" */
5881 xfs_buf_t *bp; /* buffer for "block" */
5882 int error; /* error return value */
5883 xfs_extnum_t i=0, j; /* index into the extents list */
5884 xfs_ifork_t *ifp; /* fork structure */
5885 int level; /* btree level, for checking */
5886 xfs_mount_t *mp; /* file system mount structure */
5887 __be64 *pp; /* pointer to block address */
5888 xfs_bmbt_rec_t *ep; /* pointer to current extent */
5889 xfs_bmbt_rec_t last = {0, 0}; /* last extent in prev block */
5890 xfs_bmbt_rec_t *nextp; /* pointer to next extent */
5893 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
5899 ifp = XFS_IFORK_PTR(ip, whichfork);
5900 block = ifp->if_broot;
5902 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5904 level = be16_to_cpu(block->bb_level);
5906 xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
5907 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5908 bno = be64_to_cpu(*pp);
5910 ASSERT(bno != NULLDFSBNO);
5911 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5912 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5915 * Go down the tree until leaf level is reached, following the first
5916 * pointer (leftmost) at each level.
5918 while (level-- > 0) {
5919 /* See if buf is in cur first */
5920 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5926 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5927 XFS_BMAP_BTREE_REF)))
5929 block = XFS_BUF_TO_BLOCK(bp);
5930 XFS_WANT_CORRUPTED_GOTO(
5931 xfs_bmap_sanity_check(mp, bp, level),
5937 * Check this block for basic sanity (increasing keys and
5938 * no duplicate blocks).
5941 xfs_check_block(block, mp, 0, 0);
5942 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
5943 bno = be64_to_cpu(*pp);
5944 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
5947 xfs_trans_brelse(NULL, bp);
5952 * Here with bp and block set to the leftmost leaf node in the tree.
5957 * Loop over all leaf nodes checking that all extents are in the right order.
5960 xfs_fsblock_t nextbno;
5961 xfs_extnum_t num_recs;
5964 num_recs = xfs_btree_get_numrecs(block);
5967 * Read-ahead the next leaf block, if any.
5970 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
5973 * Check all the extents to make sure they are OK.
5974 * If we had a previous block, the last entry should
5975 * conform with the first entry in this one.
5978 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
5980 ASSERT(xfs_bmbt_disk_get_startoff(&last) +
5981 xfs_bmbt_disk_get_blockcount(&last) <=
5982 xfs_bmbt_disk_get_startoff(ep));
5984 for (j = 1; j < num_recs; j++) {
5985 nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
5986 ASSERT(xfs_bmbt_disk_get_startoff(ep) +
5987 xfs_bmbt_disk_get_blockcount(ep) <=
5988 xfs_bmbt_disk_get_startoff(nextp));
5996 xfs_trans_brelse(NULL, bp);
6000 * If we've reached the end, stop.
6002 if (bno == NULLFSBLOCK)
6005 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
6011 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
6012 XFS_BMAP_BTREE_REF)))
6014 block = XFS_BUF_TO_BLOCK(bp);
6018 xfs_trans_brelse(NULL, bp);
6023 xfs_warn(mp, "%s: at error0", __func__);
6025 xfs_trans_brelse(NULL, bp);
6027 xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
6029 panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
6035 * Count fsblocks of the given fork.
6038 xfs_bmap_count_blocks(
6039 xfs_trans_t *tp, /* transaction pointer */
6040 xfs_inode_t *ip, /* incore inode */
6041 int whichfork, /* data or attr fork */
6042 int *count) /* out: count of blocks */
6044 struct xfs_btree_block *block; /* current btree block */
6045 xfs_fsblock_t bno; /* block # of "block" */
6046 xfs_ifork_t *ifp; /* fork structure */
6047 int level; /* btree level, for checking */
6048 xfs_mount_t *mp; /* file system mount structure */
6049 __be64 *pp; /* pointer to block address */
6053 ifp = XFS_IFORK_PTR(ip, whichfork);
6054 if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
6055 xfs_bmap_count_leaves(ifp, 0,
6056 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
6062 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6064 block = ifp->if_broot;
6065 level = be16_to_cpu(block->bb_level);
6067 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
6068 bno = be64_to_cpu(*pp);
6069 ASSERT(bno != NULLDFSBNO);
6070 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
6071 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
6073 if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
6074 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
6076 return XFS_ERROR(EFSCORRUPTED);
6083 * Recursively walks each level of a btree
6084 * to count total fsblocks is use.
6086 STATIC int /* error */
6087 xfs_bmap_count_tree(
6088 xfs_mount_t *mp, /* file system mount point */
6089 xfs_trans_t *tp, /* transaction pointer */
6090 xfs_ifork_t *ifp, /* inode fork pointer */
6091 xfs_fsblock_t blockno, /* file system block number */
6092 int levelin, /* level in btree */
6093 int *count) /* Count of blocks */
6096 xfs_buf_t *bp, *nbp;
6097 int level = levelin;
6099 xfs_fsblock_t bno = blockno;
6100 xfs_fsblock_t nextbno;
6101 struct xfs_btree_block *block, *nextblock;
6104 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
6107 block = XFS_BUF_TO_BLOCK(bp);
6110 /* Not at node above leaves, count this level of nodes */
6111 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
6112 while (nextbno != NULLFSBLOCK) {
6113 if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
6114 0, &nbp, XFS_BMAP_BTREE_REF)))
6117 nextblock = XFS_BUF_TO_BLOCK(nbp);
6118 nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
6119 xfs_trans_brelse(tp, nbp);
6122 /* Dive to the next level */
6123 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
6124 bno = be64_to_cpu(*pp);
6125 if (unlikely((error =
6126 xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
6127 xfs_trans_brelse(tp, bp);
6128 XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6129 XFS_ERRLEVEL_LOW, mp);
6130 return XFS_ERROR(EFSCORRUPTED);
6132 xfs_trans_brelse(tp, bp);
6134 /* count all level 1 nodes and their leaves */
6136 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
6137 numrecs = be16_to_cpu(block->bb_numrecs);
6138 xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
6139 xfs_trans_brelse(tp, bp);
6140 if (nextbno == NULLFSBLOCK)
6143 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6144 XFS_BMAP_BTREE_REF)))
6147 block = XFS_BUF_TO_BLOCK(bp);
6154 * Count leaf blocks given a range of extent records.
6157 xfs_bmap_count_leaves(
6165 for (b = 0; b < numrecs; b++) {
6166 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
6167 *count += xfs_bmbt_get_blockcount(frp);
6172 * Count leaf blocks given a range of extent records originally
6176 xfs_bmap_disk_count_leaves(
6177 struct xfs_mount *mp,
6178 struct xfs_btree_block *block,
6183 xfs_bmbt_rec_t *frp;
6185 for (b = 1; b <= numrecs; b++) {
6186 frp = XFS_BMBT_REC_ADDR(mp, block, b);
6187 *count += xfs_bmbt_disk_get_blockcount(frp);
6192 * dead simple method of punching delalyed allocation blocks from a range in
6193 * the inode. Walks a block at a time so will be slow, but is only executed in
6194 * rare error cases so the overhead is not critical. This will alays punch out
6195 * both the start and end blocks, even if the ranges only partially overlap
6196 * them, so it is up to the caller to ensure that partial blocks are not
6200 xfs_bmap_punch_delalloc_range(
6201 struct xfs_inode *ip,
6202 xfs_fileoff_t start_fsb,
6203 xfs_fileoff_t length)
6205 xfs_fileoff_t remaining = length;
6208 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
6212 xfs_bmbt_irec_t imap;
6214 xfs_fsblock_t firstblock;
6215 xfs_bmap_free_t flist;
6218 * Map the range first and check that it is a delalloc extent
6219 * before trying to unmap the range. Otherwise we will be
6220 * trying to remove a real extent (which requires a
6221 * transaction) or a hole, which is probably a bad idea...
6223 error = xfs_bmapi_read(ip, start_fsb, 1, &imap, &nimaps,
6227 /* something screwed, just bail */
6228 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
6229 xfs_alert(ip->i_mount,
6230 "Failed delalloc mapping lookup ino %lld fsb %lld.",
6231 ip->i_ino, start_fsb);
6239 if (imap.br_startblock != DELAYSTARTBLOCK) {
6240 /* been converted, ignore */
6243 WARN_ON(imap.br_blockcount == 0);
6246 * Note: while we initialise the firstblock/flist pair, they
6247 * should never be used because blocks should never be
6248 * allocated or freed for a delalloc extent and hence we need
6249 * don't cancel or finish them after the xfs_bunmapi() call.
6251 xfs_bmap_init(&flist, &firstblock);
6252 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
6257 ASSERT(!flist.xbf_count && !flist.xbf_first);
6261 } while(remaining > 0);