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"
44 #include "xfs_quota.h"
45 #include "xfs_trans_space.h"
46 #include "xfs_buf_item.h"
47 #include "xfs_filestream.h"
48 #include "xfs_vnodeops.h"
49 #include "xfs_trace.h"
52 kmem_zone_t *xfs_bmap_free_item_zone;
55 * Prototypes for internal bmap routines.
60 xfs_bmap_check_leaf_extents(
61 struct xfs_btree_cur *cur,
65 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork) do { } while (0)
70 * Called from xfs_bmap_add_attrfork to handle extents format files.
72 STATIC int /* error */
73 xfs_bmap_add_attrfork_extents(
74 xfs_trans_t *tp, /* transaction pointer */
75 xfs_inode_t *ip, /* incore inode pointer */
76 xfs_fsblock_t *firstblock, /* first block allocated */
77 xfs_bmap_free_t *flist, /* blocks to free at commit */
78 int *flags); /* inode logging flags */
81 * Called from xfs_bmap_add_attrfork to handle local format files.
83 STATIC int /* error */
84 xfs_bmap_add_attrfork_local(
85 xfs_trans_t *tp, /* transaction pointer */
86 xfs_inode_t *ip, /* incore inode pointer */
87 xfs_fsblock_t *firstblock, /* first block allocated */
88 xfs_bmap_free_t *flist, /* blocks to free at commit */
89 int *flags); /* inode logging flags */
92 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
93 * It figures out where to ask the underlying allocator to put the new extent.
95 STATIC int /* error */
97 xfs_bmalloca_t *ap); /* bmap alloc argument struct */
100 * Transform a btree format file with only one leaf node, where the
101 * extents list will fit in the inode, into an extents format file.
102 * Since the file extents are already in-core, all we have to do is
103 * give up the space for the btree root and pitch the leaf block.
105 STATIC int /* error */
106 xfs_bmap_btree_to_extents(
107 xfs_trans_t *tp, /* transaction pointer */
108 xfs_inode_t *ip, /* incore inode pointer */
109 xfs_btree_cur_t *cur, /* btree cursor */
110 int *logflagsp, /* inode logging flags */
111 int whichfork); /* data or attr fork */
114 * Remove the entry "free" from the free item list. Prev points to the
115 * previous entry, unless "free" is the head of the list.
119 xfs_bmap_free_t *flist, /* free item list header */
120 xfs_bmap_free_item_t *prev, /* previous item on list, if any */
121 xfs_bmap_free_item_t *free); /* list item to be freed */
124 * Convert an extents-format file into a btree-format file.
125 * The new file will have a root block (in the inode) and a single child block.
127 STATIC int /* error */
128 xfs_bmap_extents_to_btree(
129 xfs_trans_t *tp, /* transaction pointer */
130 xfs_inode_t *ip, /* incore inode pointer */
131 xfs_fsblock_t *firstblock, /* first-block-allocated */
132 xfs_bmap_free_t *flist, /* blocks freed in xaction */
133 xfs_btree_cur_t **curp, /* cursor returned to caller */
134 int wasdel, /* converting a delayed alloc */
135 int *logflagsp, /* inode logging flags */
136 int whichfork); /* data or attr fork */
139 * Convert a local file to an extents file.
140 * This code is sort of bogus, since the file data needs to get
141 * logged so it won't be lost. The bmap-level manipulations are ok, though.
143 STATIC int /* error */
144 xfs_bmap_local_to_extents(
145 xfs_trans_t *tp, /* transaction pointer */
146 xfs_inode_t *ip, /* incore inode pointer */
147 xfs_fsblock_t *firstblock, /* first block allocated in xaction */
148 xfs_extlen_t total, /* total blocks needed by transaction */
149 int *logflagsp, /* inode logging flags */
150 int whichfork); /* data or attr fork */
153 * Search the extents list for the inode, for the extent containing bno.
154 * If bno lies in a hole, point to the next entry. If bno lies past eof,
155 * *eofp will be set, and *prevp will contain the last entry (null if none).
156 * Else, *lastxp will be set to the index of the found
157 * entry; *gotp will contain the entry.
159 STATIC xfs_bmbt_rec_host_t * /* pointer to found extent entry */
160 xfs_bmap_search_extents(
161 xfs_inode_t *ip, /* incore inode pointer */
162 xfs_fileoff_t bno, /* block number searched for */
163 int whichfork, /* data or attr fork */
164 int *eofp, /* out: end of file found */
165 xfs_extnum_t *lastxp, /* out: last extent index */
166 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
167 xfs_bmbt_irec_t *prevp); /* out: previous extent entry found */
170 * Compute the worst-case number of indirect blocks that will be used
171 * for ip's delayed extent of length "len".
174 xfs_bmap_worst_indlen(
175 xfs_inode_t *ip, /* incore inode pointer */
176 xfs_filblks_t len); /* delayed extent length */
180 * Perform various validation checks on the values being returned
184 xfs_bmap_validate_ret(
188 xfs_bmbt_irec_t *mval,
192 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
200 xfs_fsblock_t blockno,
205 xfs_bmap_count_leaves(
212 xfs_bmap_disk_count_leaves(
213 struct xfs_mount *mp,
214 struct xfs_btree_block *block,
219 * Bmap internal routines.
222 STATIC int /* error */
224 struct xfs_btree_cur *cur,
228 int *stat) /* success/failure */
230 cur->bc_rec.b.br_startoff = off;
231 cur->bc_rec.b.br_startblock = bno;
232 cur->bc_rec.b.br_blockcount = len;
233 return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
236 STATIC int /* error */
238 struct xfs_btree_cur *cur,
242 int *stat) /* success/failure */
244 cur->bc_rec.b.br_startoff = off;
245 cur->bc_rec.b.br_startblock = bno;
246 cur->bc_rec.b.br_blockcount = len;
247 return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
251 * Check if the inode needs to be converted to btree format.
253 static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
255 return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
256 XFS_IFORK_NEXTENTS(ip, whichfork) >
257 XFS_IFORK_MAXEXT(ip, whichfork);
261 * Check if the inode should be converted to extent format.
263 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
265 return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
266 XFS_IFORK_NEXTENTS(ip, whichfork) <=
267 XFS_IFORK_MAXEXT(ip, whichfork);
271 * Update the record referred to by cur to the value given
272 * by [off, bno, len, state].
273 * This either works (return 0) or gets an EFSCORRUPTED error.
277 struct xfs_btree_cur *cur,
283 union xfs_btree_rec rec;
285 xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
286 return xfs_btree_update(cur, &rec);
290 * Called from xfs_bmap_add_attrfork to handle btree format files.
292 STATIC int /* error */
293 xfs_bmap_add_attrfork_btree(
294 xfs_trans_t *tp, /* transaction pointer */
295 xfs_inode_t *ip, /* incore inode pointer */
296 xfs_fsblock_t *firstblock, /* first block allocated */
297 xfs_bmap_free_t *flist, /* blocks to free at commit */
298 int *flags) /* inode logging flags */
300 xfs_btree_cur_t *cur; /* btree cursor */
301 int error; /* error return value */
302 xfs_mount_t *mp; /* file system mount struct */
303 int stat; /* newroot status */
306 if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
307 *flags |= XFS_ILOG_DBROOT;
309 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
310 cur->bc_private.b.flist = flist;
311 cur->bc_private.b.firstblock = *firstblock;
312 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
314 /* must be at least one entry */
315 XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
316 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
319 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
320 return XFS_ERROR(ENOSPC);
322 *firstblock = cur->bc_private.b.firstblock;
323 cur->bc_private.b.allocated = 0;
324 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
328 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
333 * Called from xfs_bmap_add_attrfork to handle extents format files.
335 STATIC int /* error */
336 xfs_bmap_add_attrfork_extents(
337 xfs_trans_t *tp, /* transaction pointer */
338 xfs_inode_t *ip, /* incore inode pointer */
339 xfs_fsblock_t *firstblock, /* first block allocated */
340 xfs_bmap_free_t *flist, /* blocks to free at commit */
341 int *flags) /* inode logging flags */
343 xfs_btree_cur_t *cur; /* bmap btree cursor */
344 int error; /* error return value */
346 if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
349 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
350 flags, XFS_DATA_FORK);
352 cur->bc_private.b.allocated = 0;
353 xfs_btree_del_cursor(cur,
354 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
360 * Called from xfs_bmap_add_attrfork to handle local format files.
362 STATIC int /* error */
363 xfs_bmap_add_attrfork_local(
364 xfs_trans_t *tp, /* transaction pointer */
365 xfs_inode_t *ip, /* incore inode pointer */
366 xfs_fsblock_t *firstblock, /* first block allocated */
367 xfs_bmap_free_t *flist, /* blocks to free at commit */
368 int *flags) /* inode logging flags */
370 xfs_da_args_t dargs; /* args for dir/attr code */
371 int error; /* error return value */
372 xfs_mount_t *mp; /* mount structure pointer */
374 if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
376 if (S_ISDIR(ip->i_d.di_mode)) {
378 memset(&dargs, 0, sizeof(dargs));
380 dargs.firstblock = firstblock;
382 dargs.total = mp->m_dirblkfsbs;
383 dargs.whichfork = XFS_DATA_FORK;
385 error = xfs_dir2_sf_to_block(&dargs);
387 error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
393 * Convert a delayed allocation to a real allocation.
395 STATIC int /* error */
396 xfs_bmap_add_extent_delay_real(
397 struct xfs_bmalloca *bma)
399 struct xfs_bmbt_irec *new = &bma->got;
400 int diff; /* temp value */
401 xfs_bmbt_rec_host_t *ep; /* extent entry for idx */
402 int error; /* error return value */
403 int i; /* temp state */
404 xfs_ifork_t *ifp; /* inode fork pointer */
405 xfs_fileoff_t new_endoff; /* end offset of new entry */
406 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
407 /* left is 0, right is 1, prev is 2 */
408 int rval=0; /* return value (logging flags) */
409 int state = 0;/* state bits, accessed thru macros */
410 xfs_filblks_t da_new; /* new count del alloc blocks used */
411 xfs_filblks_t da_old; /* old count del alloc blocks used */
412 xfs_filblks_t temp=0; /* value for da_new calculations */
413 xfs_filblks_t temp2=0;/* value for da_new calculations */
414 int tmp_rval; /* partial logging flags */
416 ifp = XFS_IFORK_PTR(bma->ip, XFS_DATA_FORK);
418 ASSERT(bma->idx >= 0);
419 ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
420 ASSERT(!isnullstartblock(new->br_startblock));
422 (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
424 XFS_STATS_INC(xs_add_exlist);
431 * Set up a bunch of variables to make the tests simpler.
433 ep = xfs_iext_get_ext(ifp, bma->idx);
434 xfs_bmbt_get_all(ep, &PREV);
435 new_endoff = new->br_startoff + new->br_blockcount;
436 ASSERT(PREV.br_startoff <= new->br_startoff);
437 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
439 da_old = startblockval(PREV.br_startblock);
443 * Set flags determining what part of the previous delayed allocation
444 * extent is being replaced by a real allocation.
446 if (PREV.br_startoff == new->br_startoff)
447 state |= BMAP_LEFT_FILLING;
448 if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
449 state |= BMAP_RIGHT_FILLING;
452 * Check and set flags if this segment has a left neighbor.
453 * Don't set contiguous if the combined extent would be too large.
456 state |= BMAP_LEFT_VALID;
457 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
459 if (isnullstartblock(LEFT.br_startblock))
460 state |= BMAP_LEFT_DELAY;
463 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
464 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
465 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
466 LEFT.br_state == new->br_state &&
467 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
468 state |= BMAP_LEFT_CONTIG;
471 * Check and set flags if this segment has a right neighbor.
472 * Don't set contiguous if the combined extent would be too large.
473 * Also check for all-three-contiguous being too large.
475 if (bma->idx < bma->ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
476 state |= BMAP_RIGHT_VALID;
477 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
479 if (isnullstartblock(RIGHT.br_startblock))
480 state |= BMAP_RIGHT_DELAY;
483 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
484 new_endoff == RIGHT.br_startoff &&
485 new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
486 new->br_state == RIGHT.br_state &&
487 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
488 ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
489 BMAP_RIGHT_FILLING)) !=
490 (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
491 BMAP_RIGHT_FILLING) ||
492 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
494 state |= BMAP_RIGHT_CONTIG;
498 * Switch out based on the FILLING and CONTIG state bits.
500 switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
501 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
502 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
503 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
505 * Filling in all of a previously delayed allocation extent.
506 * The left and right neighbors are both contiguous with new.
509 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
510 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
511 LEFT.br_blockcount + PREV.br_blockcount +
512 RIGHT.br_blockcount);
513 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
515 xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
516 bma->ip->i_d.di_nextents--;
517 if (bma->cur == NULL)
518 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
520 rval = XFS_ILOG_CORE;
521 error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
523 RIGHT.br_blockcount, &i);
526 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
527 error = xfs_btree_delete(bma->cur, &i);
530 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
531 error = xfs_btree_decrement(bma->cur, 0, &i);
534 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
535 error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
539 RIGHT.br_blockcount, LEFT.br_state);
545 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
547 * Filling in all of a previously delayed allocation extent.
548 * The left neighbor is contiguous, the right is not.
552 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
553 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
554 LEFT.br_blockcount + PREV.br_blockcount);
555 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
557 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
558 if (bma->cur == NULL)
559 rval = XFS_ILOG_DEXT;
562 error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
563 LEFT.br_startblock, LEFT.br_blockcount,
567 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
568 error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
571 PREV.br_blockcount, LEFT.br_state);
577 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
579 * Filling in all of a previously delayed allocation extent.
580 * The right neighbor is contiguous, the left is not.
582 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
583 xfs_bmbt_set_startblock(ep, new->br_startblock);
584 xfs_bmbt_set_blockcount(ep,
585 PREV.br_blockcount + RIGHT.br_blockcount);
586 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
588 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
589 if (bma->cur == NULL)
590 rval = XFS_ILOG_DEXT;
593 error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
595 RIGHT.br_blockcount, &i);
598 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
599 error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
602 RIGHT.br_blockcount, PREV.br_state);
608 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
610 * Filling in all of a previously delayed allocation extent.
611 * Neither the left nor right neighbors are contiguous with
614 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
615 xfs_bmbt_set_startblock(ep, new->br_startblock);
616 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
618 bma->ip->i_d.di_nextents++;
619 if (bma->cur == NULL)
620 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
622 rval = XFS_ILOG_CORE;
623 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
624 new->br_startblock, new->br_blockcount,
628 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
629 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
630 error = xfs_btree_insert(bma->cur, &i);
633 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
637 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
639 * Filling in the first part of a previous delayed allocation.
640 * The left neighbor is contiguous.
642 trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
643 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
644 LEFT.br_blockcount + new->br_blockcount);
645 xfs_bmbt_set_startoff(ep,
646 PREV.br_startoff + new->br_blockcount);
647 trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
649 temp = PREV.br_blockcount - new->br_blockcount;
650 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
651 xfs_bmbt_set_blockcount(ep, temp);
652 if (bma->cur == NULL)
653 rval = XFS_ILOG_DEXT;
656 error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
657 LEFT.br_startblock, LEFT.br_blockcount,
661 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
662 error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
670 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
671 startblockval(PREV.br_startblock));
672 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
673 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
678 case BMAP_LEFT_FILLING:
680 * Filling in the first part of a previous delayed allocation.
681 * The left neighbor is not contiguous.
683 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
684 xfs_bmbt_set_startoff(ep, new_endoff);
685 temp = PREV.br_blockcount - new->br_blockcount;
686 xfs_bmbt_set_blockcount(ep, temp);
687 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
688 bma->ip->i_d.di_nextents++;
689 if (bma->cur == NULL)
690 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
692 rval = XFS_ILOG_CORE;
693 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
694 new->br_startblock, new->br_blockcount,
698 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
699 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
700 error = xfs_btree_insert(bma->cur, &i);
703 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
706 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
707 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
708 bma->firstblock, bma->flist,
709 &bma->cur, 1, &tmp_rval, XFS_DATA_FORK);
714 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
715 startblockval(PREV.br_startblock) -
716 (bma->cur ? bma->cur->bc_private.b.allocated : 0));
717 ep = xfs_iext_get_ext(ifp, bma->idx + 1);
718 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
719 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
722 case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
724 * Filling in the last part of a previous delayed allocation.
725 * The right neighbor is contiguous with the new allocation.
727 temp = PREV.br_blockcount - new->br_blockcount;
728 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
729 xfs_bmbt_set_blockcount(ep, temp);
730 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
731 new->br_startoff, new->br_startblock,
732 new->br_blockcount + RIGHT.br_blockcount,
734 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
735 if (bma->cur == NULL)
736 rval = XFS_ILOG_DEXT;
739 error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
741 RIGHT.br_blockcount, &i);
744 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
745 error = xfs_bmbt_update(bma->cur, new->br_startoff,
754 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
755 startblockval(PREV.br_startblock));
756 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
757 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
758 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
763 case BMAP_RIGHT_FILLING:
765 * Filling in the last part of a previous delayed allocation.
766 * The right neighbor is not contiguous.
768 temp = PREV.br_blockcount - new->br_blockcount;
769 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
770 xfs_bmbt_set_blockcount(ep, temp);
771 xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
772 bma->ip->i_d.di_nextents++;
773 if (bma->cur == NULL)
774 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
776 rval = XFS_ILOG_CORE;
777 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
778 new->br_startblock, new->br_blockcount,
782 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
783 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
784 error = xfs_btree_insert(bma->cur, &i);
787 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
790 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
791 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
792 bma->firstblock, bma->flist, &bma->cur, 1,
793 &tmp_rval, XFS_DATA_FORK);
798 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
799 startblockval(PREV.br_startblock) -
800 (bma->cur ? bma->cur->bc_private.b.allocated : 0));
801 ep = xfs_iext_get_ext(ifp, bma->idx);
802 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
803 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
810 * Filling in the middle part of a previous delayed allocation.
811 * Contiguity is impossible here.
812 * This case is avoided almost all the time.
814 * We start with a delayed allocation:
816 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
819 * and we are allocating:
820 * +rrrrrrrrrrrrrrrrr+
823 * and we set it up for insertion as:
824 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
826 * PREV @ idx LEFT RIGHT
827 * inserted at idx + 1
829 temp = new->br_startoff - PREV.br_startoff;
830 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
831 trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
832 xfs_bmbt_set_blockcount(ep, temp); /* truncate PREV */
834 RIGHT.br_state = PREV.br_state;
835 RIGHT.br_startblock = nullstartblock(
836 (int)xfs_bmap_worst_indlen(bma->ip, temp2));
837 RIGHT.br_startoff = new_endoff;
838 RIGHT.br_blockcount = temp2;
839 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
840 xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
841 bma->ip->i_d.di_nextents++;
842 if (bma->cur == NULL)
843 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
845 rval = XFS_ILOG_CORE;
846 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
847 new->br_startblock, new->br_blockcount,
851 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
852 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
853 error = xfs_btree_insert(bma->cur, &i);
856 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
859 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
860 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
861 bma->firstblock, bma->flist, &bma->cur,
862 1, &tmp_rval, XFS_DATA_FORK);
867 temp = xfs_bmap_worst_indlen(bma->ip, temp);
868 temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
869 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
870 (bma->cur ? bma->cur->bc_private.b.allocated : 0));
872 error = xfs_icsb_modify_counters(bma->ip->i_mount,
874 -((int64_t)diff), 0);
880 ep = xfs_iext_get_ext(ifp, bma->idx);
881 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
882 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
883 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
884 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
885 nullstartblock((int)temp2));
886 trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
889 da_new = temp + temp2;
892 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
893 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
894 case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
895 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
896 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
897 case BMAP_LEFT_CONTIG:
898 case BMAP_RIGHT_CONTIG:
900 * These cases are all impossible.
905 /* convert to a btree if necessary */
906 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
907 int tmp_logflags; /* partial log flag return val */
909 ASSERT(bma->cur == NULL);
910 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
911 bma->firstblock, bma->flist, &bma->cur,
912 da_old > 0, &tmp_logflags, XFS_DATA_FORK);
913 bma->logflags |= tmp_logflags;
918 /* adjust for changes in reserved delayed indirect blocks */
919 if (da_old || da_new) {
922 temp += bma->cur->bc_private.b.allocated;
923 ASSERT(temp <= da_old);
925 xfs_icsb_modify_counters(bma->ip->i_mount,
927 (int64_t)(da_old - temp), 0);
930 /* clear out the allocated field, done with it now in any case. */
932 bma->cur->bc_private.b.allocated = 0;
934 xfs_bmap_check_leaf_extents(bma->cur, bma->ip, XFS_DATA_FORK);
936 bma->logflags |= rval;
944 * Convert an unwritten allocation to a real allocation or vice versa.
946 STATIC int /* error */
947 xfs_bmap_add_extent_unwritten_real(
948 struct xfs_trans *tp,
949 xfs_inode_t *ip, /* incore inode pointer */
950 xfs_extnum_t *idx, /* extent number to update/insert */
951 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
952 xfs_bmbt_irec_t *new, /* new data to add to file extents */
953 xfs_fsblock_t *first, /* pointer to firstblock variable */
954 xfs_bmap_free_t *flist, /* list of extents to be freed */
955 int *logflagsp) /* inode logging flags */
957 xfs_btree_cur_t *cur; /* btree cursor */
958 xfs_bmbt_rec_host_t *ep; /* extent entry for idx */
959 int error; /* error return value */
960 int i; /* temp state */
961 xfs_ifork_t *ifp; /* inode fork pointer */
962 xfs_fileoff_t new_endoff; /* end offset of new entry */
963 xfs_exntst_t newext; /* new extent state */
964 xfs_exntst_t oldext; /* old extent state */
965 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
966 /* left is 0, right is 1, prev is 2 */
967 int rval=0; /* return value (logging flags) */
968 int state = 0;/* state bits, accessed thru macros */
973 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
976 ASSERT(*idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
977 ASSERT(!isnullstartblock(new->br_startblock));
979 XFS_STATS_INC(xs_add_exlist);
986 * Set up a bunch of variables to make the tests simpler.
989 ep = xfs_iext_get_ext(ifp, *idx);
990 xfs_bmbt_get_all(ep, &PREV);
991 newext = new->br_state;
992 oldext = (newext == XFS_EXT_UNWRITTEN) ?
993 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
994 ASSERT(PREV.br_state == oldext);
995 new_endoff = new->br_startoff + new->br_blockcount;
996 ASSERT(PREV.br_startoff <= new->br_startoff);
997 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1000 * Set flags determining what part of the previous oldext allocation
1001 * extent is being replaced by a newext allocation.
1003 if (PREV.br_startoff == new->br_startoff)
1004 state |= BMAP_LEFT_FILLING;
1005 if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1006 state |= BMAP_RIGHT_FILLING;
1009 * Check and set flags if this segment has a left neighbor.
1010 * Don't set contiguous if the combined extent would be too large.
1013 state |= BMAP_LEFT_VALID;
1014 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
1016 if (isnullstartblock(LEFT.br_startblock))
1017 state |= BMAP_LEFT_DELAY;
1020 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1021 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1022 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1023 LEFT.br_state == newext &&
1024 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1025 state |= BMAP_LEFT_CONTIG;
1028 * Check and set flags if this segment has a right neighbor.
1029 * Don't set contiguous if the combined extent would be too large.
1030 * Also check for all-three-contiguous being too large.
1032 if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1033 state |= BMAP_RIGHT_VALID;
1034 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
1035 if (isnullstartblock(RIGHT.br_startblock))
1036 state |= BMAP_RIGHT_DELAY;
1039 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1040 new_endoff == RIGHT.br_startoff &&
1041 new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1042 newext == RIGHT.br_state &&
1043 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1044 ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1045 BMAP_RIGHT_FILLING)) !=
1046 (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1047 BMAP_RIGHT_FILLING) ||
1048 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1050 state |= BMAP_RIGHT_CONTIG;
1053 * Switch out based on the FILLING and CONTIG state bits.
1055 switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1056 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1057 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1058 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1060 * Setting all of a previous oldext extent to newext.
1061 * The left and right neighbors are both contiguous with new.
1065 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1066 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1067 LEFT.br_blockcount + PREV.br_blockcount +
1068 RIGHT.br_blockcount);
1069 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1071 xfs_iext_remove(ip, *idx + 1, 2, state);
1072 ip->i_d.di_nextents -= 2;
1074 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1076 rval = XFS_ILOG_CORE;
1077 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1078 RIGHT.br_startblock,
1079 RIGHT.br_blockcount, &i)))
1081 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1082 if ((error = xfs_btree_delete(cur, &i)))
1084 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1085 if ((error = xfs_btree_decrement(cur, 0, &i)))
1087 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1088 if ((error = xfs_btree_delete(cur, &i)))
1090 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1091 if ((error = xfs_btree_decrement(cur, 0, &i)))
1093 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1094 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1096 LEFT.br_blockcount + PREV.br_blockcount +
1097 RIGHT.br_blockcount, LEFT.br_state)))
1102 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1104 * Setting all of a previous oldext extent to newext.
1105 * The left neighbor is contiguous, the right is not.
1109 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1110 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1111 LEFT.br_blockcount + PREV.br_blockcount);
1112 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1114 xfs_iext_remove(ip, *idx + 1, 1, state);
1115 ip->i_d.di_nextents--;
1117 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1119 rval = XFS_ILOG_CORE;
1120 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1121 PREV.br_startblock, PREV.br_blockcount,
1124 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1125 if ((error = xfs_btree_delete(cur, &i)))
1127 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1128 if ((error = xfs_btree_decrement(cur, 0, &i)))
1130 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1131 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1133 LEFT.br_blockcount + PREV.br_blockcount,
1139 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1141 * Setting all of a previous oldext extent to newext.
1142 * The right neighbor is contiguous, the left is not.
1144 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1145 xfs_bmbt_set_blockcount(ep,
1146 PREV.br_blockcount + RIGHT.br_blockcount);
1147 xfs_bmbt_set_state(ep, newext);
1148 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1149 xfs_iext_remove(ip, *idx + 1, 1, state);
1150 ip->i_d.di_nextents--;
1152 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1154 rval = XFS_ILOG_CORE;
1155 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1156 RIGHT.br_startblock,
1157 RIGHT.br_blockcount, &i)))
1159 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1160 if ((error = xfs_btree_delete(cur, &i)))
1162 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1163 if ((error = xfs_btree_decrement(cur, 0, &i)))
1165 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1166 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1168 new->br_blockcount + RIGHT.br_blockcount,
1174 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1176 * Setting all of a previous oldext extent to newext.
1177 * Neither the left nor right neighbors are contiguous with
1180 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1181 xfs_bmbt_set_state(ep, newext);
1182 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1185 rval = XFS_ILOG_DEXT;
1188 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1189 new->br_startblock, new->br_blockcount,
1192 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1193 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1194 new->br_startblock, new->br_blockcount,
1200 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1202 * Setting the first part of a previous oldext extent to newext.
1203 * The left neighbor is contiguous.
1205 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
1206 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
1207 LEFT.br_blockcount + new->br_blockcount);
1208 xfs_bmbt_set_startoff(ep,
1209 PREV.br_startoff + new->br_blockcount);
1210 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
1212 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1213 xfs_bmbt_set_startblock(ep,
1214 new->br_startblock + new->br_blockcount);
1215 xfs_bmbt_set_blockcount(ep,
1216 PREV.br_blockcount - new->br_blockcount);
1217 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1222 rval = XFS_ILOG_DEXT;
1225 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1226 PREV.br_startblock, PREV.br_blockcount,
1229 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1230 if ((error = xfs_bmbt_update(cur,
1231 PREV.br_startoff + new->br_blockcount,
1232 PREV.br_startblock + new->br_blockcount,
1233 PREV.br_blockcount - new->br_blockcount,
1236 if ((error = xfs_btree_decrement(cur, 0, &i)))
1238 error = xfs_bmbt_update(cur, LEFT.br_startoff,
1240 LEFT.br_blockcount + new->br_blockcount,
1247 case BMAP_LEFT_FILLING:
1249 * Setting the first part of a previous oldext extent to newext.
1250 * The left neighbor is not contiguous.
1252 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1253 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1254 xfs_bmbt_set_startoff(ep, new_endoff);
1255 xfs_bmbt_set_blockcount(ep,
1256 PREV.br_blockcount - new->br_blockcount);
1257 xfs_bmbt_set_startblock(ep,
1258 new->br_startblock + new->br_blockcount);
1259 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1261 xfs_iext_insert(ip, *idx, 1, new, state);
1262 ip->i_d.di_nextents++;
1264 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1266 rval = XFS_ILOG_CORE;
1267 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1268 PREV.br_startblock, PREV.br_blockcount,
1271 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1272 if ((error = xfs_bmbt_update(cur,
1273 PREV.br_startoff + new->br_blockcount,
1274 PREV.br_startblock + new->br_blockcount,
1275 PREV.br_blockcount - new->br_blockcount,
1278 cur->bc_rec.b = *new;
1279 if ((error = xfs_btree_insert(cur, &i)))
1281 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1285 case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1287 * Setting the last part of a previous oldext extent to newext.
1288 * The right neighbor is contiguous with the new allocation.
1290 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1291 xfs_bmbt_set_blockcount(ep,
1292 PREV.br_blockcount - new->br_blockcount);
1293 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1297 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1298 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1299 new->br_startoff, new->br_startblock,
1300 new->br_blockcount + RIGHT.br_blockcount, newext);
1301 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1304 rval = XFS_ILOG_DEXT;
1307 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1309 PREV.br_blockcount, &i)))
1311 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1312 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1314 PREV.br_blockcount - new->br_blockcount,
1317 if ((error = xfs_btree_increment(cur, 0, &i)))
1319 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1321 new->br_blockcount + RIGHT.br_blockcount,
1327 case BMAP_RIGHT_FILLING:
1329 * Setting the last part of a previous oldext extent to newext.
1330 * The right neighbor is not contiguous.
1332 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1333 xfs_bmbt_set_blockcount(ep,
1334 PREV.br_blockcount - new->br_blockcount);
1335 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1338 xfs_iext_insert(ip, *idx, 1, new, state);
1340 ip->i_d.di_nextents++;
1342 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1344 rval = XFS_ILOG_CORE;
1345 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1346 PREV.br_startblock, PREV.br_blockcount,
1349 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1350 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1352 PREV.br_blockcount - new->br_blockcount,
1355 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1356 new->br_startblock, new->br_blockcount,
1359 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1360 cur->bc_rec.b.br_state = XFS_EXT_NORM;
1361 if ((error = xfs_btree_insert(cur, &i)))
1363 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1369 * Setting the middle part of a previous oldext extent to
1370 * newext. Contiguity is impossible here.
1371 * One extent becomes three extents.
1373 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1374 xfs_bmbt_set_blockcount(ep,
1375 new->br_startoff - PREV.br_startoff);
1376 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1379 r[1].br_startoff = new_endoff;
1380 r[1].br_blockcount =
1381 PREV.br_startoff + PREV.br_blockcount - new_endoff;
1382 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1383 r[1].br_state = oldext;
1386 xfs_iext_insert(ip, *idx, 2, &r[0], state);
1388 ip->i_d.di_nextents += 2;
1390 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1392 rval = XFS_ILOG_CORE;
1393 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1394 PREV.br_startblock, PREV.br_blockcount,
1397 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1398 /* new right extent - oldext */
1399 if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1400 r[1].br_startblock, r[1].br_blockcount,
1403 /* new left extent - oldext */
1404 cur->bc_rec.b = PREV;
1405 cur->bc_rec.b.br_blockcount =
1406 new->br_startoff - PREV.br_startoff;
1407 if ((error = xfs_btree_insert(cur, &i)))
1409 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1411 * Reset the cursor to the position of the new extent
1412 * we are about to insert as we can't trust it after
1413 * the previous insert.
1415 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1416 new->br_startblock, new->br_blockcount,
1419 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1420 /* new middle extent - newext */
1421 cur->bc_rec.b.br_state = new->br_state;
1422 if ((error = xfs_btree_insert(cur, &i)))
1424 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1428 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1429 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1430 case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1431 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1432 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1433 case BMAP_LEFT_CONTIG:
1434 case BMAP_RIGHT_CONTIG:
1436 * These cases are all impossible.
1441 /* convert to a btree if necessary */
1442 if (xfs_bmap_needs_btree(ip, XFS_DATA_FORK)) {
1443 int tmp_logflags; /* partial log flag return val */
1445 ASSERT(cur == NULL);
1446 error = xfs_bmap_extents_to_btree(tp, ip, first, flist, &cur,
1447 0, &tmp_logflags, XFS_DATA_FORK);
1448 *logflagsp |= tmp_logflags;
1453 /* clear out the allocated field, done with it now in any case. */
1455 cur->bc_private.b.allocated = 0;
1459 xfs_bmap_check_leaf_extents(*curp, ip, XFS_DATA_FORK);
1469 * Convert a hole to a delayed allocation.
1472 xfs_bmap_add_extent_hole_delay(
1473 xfs_inode_t *ip, /* incore inode pointer */
1474 xfs_extnum_t *idx, /* extent number to update/insert */
1475 xfs_bmbt_irec_t *new) /* new data to add to file extents */
1477 xfs_ifork_t *ifp; /* inode fork pointer */
1478 xfs_bmbt_irec_t left; /* left neighbor extent entry */
1479 xfs_filblks_t newlen=0; /* new indirect size */
1480 xfs_filblks_t oldlen=0; /* old indirect size */
1481 xfs_bmbt_irec_t right; /* right neighbor extent entry */
1482 int state; /* state bits, accessed thru macros */
1483 xfs_filblks_t temp=0; /* temp for indirect calculations */
1485 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1487 ASSERT(isnullstartblock(new->br_startblock));
1490 * Check and set flags if this segment has a left neighbor
1493 state |= BMAP_LEFT_VALID;
1494 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
1496 if (isnullstartblock(left.br_startblock))
1497 state |= BMAP_LEFT_DELAY;
1501 * Check and set flags if the current (right) segment exists.
1502 * If it doesn't exist, we're converting the hole at end-of-file.
1504 if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1505 state |= BMAP_RIGHT_VALID;
1506 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
1508 if (isnullstartblock(right.br_startblock))
1509 state |= BMAP_RIGHT_DELAY;
1513 * Set contiguity flags on the left and right neighbors.
1514 * Don't let extents get too large, even if the pieces are contiguous.
1516 if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
1517 left.br_startoff + left.br_blockcount == new->br_startoff &&
1518 left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1519 state |= BMAP_LEFT_CONTIG;
1521 if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
1522 new->br_startoff + new->br_blockcount == right.br_startoff &&
1523 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1524 (!(state & BMAP_LEFT_CONTIG) ||
1525 (left.br_blockcount + new->br_blockcount +
1526 right.br_blockcount <= MAXEXTLEN)))
1527 state |= BMAP_RIGHT_CONTIG;
1530 * Switch out based on the contiguity flags.
1532 switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1533 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1535 * New allocation is contiguous with delayed allocations
1536 * on the left and on the right.
1537 * Merge all three into a single extent record.
1540 temp = left.br_blockcount + new->br_blockcount +
1541 right.br_blockcount;
1543 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1544 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1545 oldlen = startblockval(left.br_startblock) +
1546 startblockval(new->br_startblock) +
1547 startblockval(right.br_startblock);
1548 newlen = xfs_bmap_worst_indlen(ip, temp);
1549 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1550 nullstartblock((int)newlen));
1551 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1553 xfs_iext_remove(ip, *idx + 1, 1, state);
1556 case BMAP_LEFT_CONTIG:
1558 * New allocation is contiguous with a delayed allocation
1560 * Merge the new allocation with the left neighbor.
1563 temp = left.br_blockcount + new->br_blockcount;
1565 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1566 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1567 oldlen = startblockval(left.br_startblock) +
1568 startblockval(new->br_startblock);
1569 newlen = xfs_bmap_worst_indlen(ip, temp);
1570 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1571 nullstartblock((int)newlen));
1572 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1575 case BMAP_RIGHT_CONTIG:
1577 * New allocation is contiguous with a delayed allocation
1579 * Merge the new allocation with the right neighbor.
1581 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1582 temp = new->br_blockcount + right.br_blockcount;
1583 oldlen = startblockval(new->br_startblock) +
1584 startblockval(right.br_startblock);
1585 newlen = xfs_bmap_worst_indlen(ip, temp);
1586 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1588 nullstartblock((int)newlen), temp, right.br_state);
1589 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1594 * New allocation is not contiguous with another
1595 * delayed allocation.
1596 * Insert a new entry.
1598 oldlen = newlen = 0;
1599 xfs_iext_insert(ip, *idx, 1, new, state);
1602 if (oldlen != newlen) {
1603 ASSERT(oldlen > newlen);
1604 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
1605 (int64_t)(oldlen - newlen), 0);
1607 * Nothing to do for disk quota accounting here.
1613 * Convert a hole to a real allocation.
1615 STATIC int /* error */
1616 xfs_bmap_add_extent_hole_real(
1617 struct xfs_bmalloca *bma,
1620 struct xfs_bmbt_irec *new = &bma->got;
1621 int error; /* error return value */
1622 int i; /* temp state */
1623 xfs_ifork_t *ifp; /* inode fork pointer */
1624 xfs_bmbt_irec_t left; /* left neighbor extent entry */
1625 xfs_bmbt_irec_t right; /* right neighbor extent entry */
1626 int rval=0; /* return value (logging flags) */
1627 int state; /* state bits, accessed thru macros */
1629 ifp = XFS_IFORK_PTR(bma->ip, whichfork);
1631 ASSERT(bma->idx >= 0);
1632 ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
1633 ASSERT(!isnullstartblock(new->br_startblock));
1635 !(bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1637 XFS_STATS_INC(xs_add_exlist);
1640 if (whichfork == XFS_ATTR_FORK)
1641 state |= BMAP_ATTRFORK;
1644 * Check and set flags if this segment has a left neighbor.
1647 state |= BMAP_LEFT_VALID;
1648 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &left);
1649 if (isnullstartblock(left.br_startblock))
1650 state |= BMAP_LEFT_DELAY;
1654 * Check and set flags if this segment has a current value.
1655 * Not true if we're inserting into the "hole" at eof.
1657 if (bma->idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1658 state |= BMAP_RIGHT_VALID;
1659 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &right);
1660 if (isnullstartblock(right.br_startblock))
1661 state |= BMAP_RIGHT_DELAY;
1665 * We're inserting a real allocation between "left" and "right".
1666 * Set the contiguity flags. Don't let extents get too large.
1668 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1669 left.br_startoff + left.br_blockcount == new->br_startoff &&
1670 left.br_startblock + left.br_blockcount == new->br_startblock &&
1671 left.br_state == new->br_state &&
1672 left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1673 state |= BMAP_LEFT_CONTIG;
1675 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1676 new->br_startoff + new->br_blockcount == right.br_startoff &&
1677 new->br_startblock + new->br_blockcount == right.br_startblock &&
1678 new->br_state == right.br_state &&
1679 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1680 (!(state & BMAP_LEFT_CONTIG) ||
1681 left.br_blockcount + new->br_blockcount +
1682 right.br_blockcount <= MAXEXTLEN))
1683 state |= BMAP_RIGHT_CONTIG;
1687 * Select which case we're in here, and implement it.
1689 switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1690 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1692 * New allocation is contiguous with real allocations on the
1693 * left and on the right.
1694 * Merge all three into a single extent record.
1697 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1698 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1699 left.br_blockcount + new->br_blockcount +
1700 right.br_blockcount);
1701 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1703 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1705 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
1706 XFS_IFORK_NEXTENTS(bma->ip, whichfork) - 1);
1707 if (bma->cur == NULL) {
1708 rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1710 rval = XFS_ILOG_CORE;
1711 error = xfs_bmbt_lookup_eq(bma->cur, right.br_startoff,
1712 right.br_startblock, right.br_blockcount,
1716 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1717 error = xfs_btree_delete(bma->cur, &i);
1720 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1721 error = xfs_btree_decrement(bma->cur, 0, &i);
1724 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1725 error = xfs_bmbt_update(bma->cur, left.br_startoff,
1727 left.br_blockcount +
1728 new->br_blockcount +
1729 right.br_blockcount,
1736 case BMAP_LEFT_CONTIG:
1738 * New allocation is contiguous with a real allocation
1740 * Merge the new allocation with the left neighbor.
1743 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1744 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1745 left.br_blockcount + new->br_blockcount);
1746 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1748 if (bma->cur == NULL) {
1749 rval = xfs_ilog_fext(whichfork);
1752 error = xfs_bmbt_lookup_eq(bma->cur, left.br_startoff,
1753 left.br_startblock, left.br_blockcount,
1757 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1758 error = xfs_bmbt_update(bma->cur, left.br_startoff,
1760 left.br_blockcount +
1768 case BMAP_RIGHT_CONTIG:
1770 * New allocation is contiguous with a real allocation
1772 * Merge the new allocation with the right neighbor.
1774 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1775 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx),
1776 new->br_startoff, new->br_startblock,
1777 new->br_blockcount + right.br_blockcount,
1779 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1781 if (bma->cur == NULL) {
1782 rval = xfs_ilog_fext(whichfork);
1785 error = xfs_bmbt_lookup_eq(bma->cur,
1787 right.br_startblock,
1788 right.br_blockcount, &i);
1791 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1792 error = xfs_bmbt_update(bma->cur, new->br_startoff,
1794 new->br_blockcount +
1795 right.br_blockcount,
1804 * New allocation is not contiguous with another
1806 * Insert a new entry.
1808 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
1809 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
1810 XFS_IFORK_NEXTENTS(bma->ip, whichfork) + 1);
1811 if (bma->cur == NULL) {
1812 rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1814 rval = XFS_ILOG_CORE;
1815 error = xfs_bmbt_lookup_eq(bma->cur,
1818 new->br_blockcount, &i);
1821 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1822 bma->cur->bc_rec.b.br_state = new->br_state;
1823 error = xfs_btree_insert(bma->cur, &i);
1826 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1831 /* convert to a btree if necessary */
1832 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1833 int tmp_logflags; /* partial log flag return val */
1835 ASSERT(bma->cur == NULL);
1836 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1837 bma->firstblock, bma->flist, &bma->cur,
1838 0, &tmp_logflags, whichfork);
1839 bma->logflags |= tmp_logflags;
1844 /* clear out the allocated field, done with it now in any case. */
1846 bma->cur->bc_private.b.allocated = 0;
1848 xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
1850 bma->logflags |= rval;
1855 * Adjust the size of the new extent based on di_extsize and rt extsize.
1858 xfs_bmap_extsize_align(
1860 xfs_bmbt_irec_t *gotp, /* next extent pointer */
1861 xfs_bmbt_irec_t *prevp, /* previous extent pointer */
1862 xfs_extlen_t extsz, /* align to this extent size */
1863 int rt, /* is this a realtime inode? */
1864 int eof, /* is extent at end-of-file? */
1865 int delay, /* creating delalloc extent? */
1866 int convert, /* overwriting unwritten extent? */
1867 xfs_fileoff_t *offp, /* in/out: aligned offset */
1868 xfs_extlen_t *lenp) /* in/out: aligned length */
1870 xfs_fileoff_t orig_off; /* original offset */
1871 xfs_extlen_t orig_alen; /* original length */
1872 xfs_fileoff_t orig_end; /* original off+len */
1873 xfs_fileoff_t nexto; /* next file offset */
1874 xfs_fileoff_t prevo; /* previous file offset */
1875 xfs_fileoff_t align_off; /* temp for offset */
1876 xfs_extlen_t align_alen; /* temp for length */
1877 xfs_extlen_t temp; /* temp for calculations */
1882 orig_off = align_off = *offp;
1883 orig_alen = align_alen = *lenp;
1884 orig_end = orig_off + orig_alen;
1887 * If this request overlaps an existing extent, then don't
1888 * attempt to perform any additional alignment.
1890 if (!delay && !eof &&
1891 (orig_off >= gotp->br_startoff) &&
1892 (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
1897 * If the file offset is unaligned vs. the extent size
1898 * we need to align it. This will be possible unless
1899 * the file was previously written with a kernel that didn't
1900 * perform this alignment, or if a truncate shot us in the
1903 temp = do_mod(orig_off, extsz);
1909 * Same adjustment for the end of the requested area.
1911 if ((temp = (align_alen % extsz))) {
1912 align_alen += extsz - temp;
1915 * If the previous block overlaps with this proposed allocation
1916 * then move the start forward without adjusting the length.
1918 if (prevp->br_startoff != NULLFILEOFF) {
1919 if (prevp->br_startblock == HOLESTARTBLOCK)
1920 prevo = prevp->br_startoff;
1922 prevo = prevp->br_startoff + prevp->br_blockcount;
1925 if (align_off != orig_off && align_off < prevo)
1928 * If the next block overlaps with this proposed allocation
1929 * then move the start back without adjusting the length,
1930 * but not before offset 0.
1931 * This may of course make the start overlap previous block,
1932 * and if we hit the offset 0 limit then the next block
1933 * can still overlap too.
1935 if (!eof && gotp->br_startoff != NULLFILEOFF) {
1936 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
1937 (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
1938 nexto = gotp->br_startoff + gotp->br_blockcount;
1940 nexto = gotp->br_startoff;
1942 nexto = NULLFILEOFF;
1944 align_off + align_alen != orig_end &&
1945 align_off + align_alen > nexto)
1946 align_off = nexto > align_alen ? nexto - align_alen : 0;
1948 * If we're now overlapping the next or previous extent that
1949 * means we can't fit an extsz piece in this hole. Just move
1950 * the start forward to the first valid spot and set
1951 * the length so we hit the end.
1953 if (align_off != orig_off && align_off < prevo)
1955 if (align_off + align_alen != orig_end &&
1956 align_off + align_alen > nexto &&
1957 nexto != NULLFILEOFF) {
1958 ASSERT(nexto > prevo);
1959 align_alen = nexto - align_off;
1963 * If realtime, and the result isn't a multiple of the realtime
1964 * extent size we need to remove blocks until it is.
1966 if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
1968 * We're not covering the original request, or
1969 * we won't be able to once we fix the length.
1971 if (orig_off < align_off ||
1972 orig_end > align_off + align_alen ||
1973 align_alen - temp < orig_alen)
1974 return XFS_ERROR(EINVAL);
1976 * Try to fix it by moving the start up.
1978 if (align_off + temp <= orig_off) {
1983 * Try to fix it by moving the end in.
1985 else if (align_off + align_alen - temp >= orig_end)
1988 * Set the start to the minimum then trim the length.
1991 align_alen -= orig_off - align_off;
1992 align_off = orig_off;
1993 align_alen -= align_alen % mp->m_sb.sb_rextsize;
1996 * Result doesn't cover the request, fail it.
1998 if (orig_off < align_off || orig_end > align_off + align_alen)
1999 return XFS_ERROR(EINVAL);
2001 ASSERT(orig_off >= align_off);
2002 ASSERT(orig_end <= align_off + align_alen);
2006 if (!eof && gotp->br_startoff != NULLFILEOFF)
2007 ASSERT(align_off + align_alen <= gotp->br_startoff);
2008 if (prevp->br_startoff != NULLFILEOFF)
2009 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
2017 #define XFS_ALLOC_GAP_UNITS 4
2021 xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2023 xfs_fsblock_t adjust; /* adjustment to block numbers */
2024 xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
2025 xfs_mount_t *mp; /* mount point structure */
2026 int nullfb; /* true if ap->firstblock isn't set */
2027 int rt; /* true if inode is realtime */
2029 #define ISVALID(x,y) \
2031 (x) < mp->m_sb.sb_rblocks : \
2032 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2033 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2034 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2036 mp = ap->ip->i_mount;
2037 nullfb = *ap->firstblock == NULLFSBLOCK;
2038 rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2039 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
2041 * If allocating at eof, and there's a previous real block,
2042 * try to use its last block as our starting point.
2044 if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
2045 !isnullstartblock(ap->prev.br_startblock) &&
2046 ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
2047 ap->prev.br_startblock)) {
2048 ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
2050 * Adjust for the gap between prevp and us.
2052 adjust = ap->offset -
2053 (ap->prev.br_startoff + ap->prev.br_blockcount);
2055 ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
2056 ap->blkno += adjust;
2059 * If not at eof, then compare the two neighbor blocks.
2060 * Figure out whether either one gives us a good starting point,
2061 * and pick the better one.
2063 else if (!ap->eof) {
2064 xfs_fsblock_t gotbno; /* right side block number */
2065 xfs_fsblock_t gotdiff=0; /* right side difference */
2066 xfs_fsblock_t prevbno; /* left side block number */
2067 xfs_fsblock_t prevdiff=0; /* left side difference */
2070 * If there's a previous (left) block, select a requested
2071 * start block based on it.
2073 if (ap->prev.br_startoff != NULLFILEOFF &&
2074 !isnullstartblock(ap->prev.br_startblock) &&
2075 (prevbno = ap->prev.br_startblock +
2076 ap->prev.br_blockcount) &&
2077 ISVALID(prevbno, ap->prev.br_startblock)) {
2079 * Calculate gap to end of previous block.
2081 adjust = prevdiff = ap->offset -
2082 (ap->prev.br_startoff +
2083 ap->prev.br_blockcount);
2085 * Figure the startblock based on the previous block's
2086 * end and the gap size.
2088 * If the gap is large relative to the piece we're
2089 * allocating, or using it gives us an invalid block
2090 * number, then just use the end of the previous block.
2092 if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
2093 ISVALID(prevbno + prevdiff,
2094 ap->prev.br_startblock))
2099 * If the firstblock forbids it, can't use it,
2102 if (!rt && !nullfb &&
2103 XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2104 prevbno = NULLFSBLOCK;
2107 * No previous block or can't follow it, just default.
2110 prevbno = NULLFSBLOCK;
2112 * If there's a following (right) block, select a requested
2113 * start block based on it.
2115 if (!isnullstartblock(ap->got.br_startblock)) {
2117 * Calculate gap to start of next block.
2119 adjust = gotdiff = ap->got.br_startoff - ap->offset;
2121 * Figure the startblock based on the next block's
2122 * start and the gap size.
2124 gotbno = ap->got.br_startblock;
2127 * If the gap is large relative to the piece we're
2128 * allocating, or using it gives us an invalid block
2129 * number, then just use the start of the next block
2130 * offset by our length.
2132 if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
2133 ISVALID(gotbno - gotdiff, gotbno))
2135 else if (ISVALID(gotbno - ap->length, gotbno)) {
2136 gotbno -= ap->length;
2137 gotdiff += adjust - ap->length;
2141 * If the firstblock forbids it, can't use it,
2144 if (!rt && !nullfb &&
2145 XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2146 gotbno = NULLFSBLOCK;
2149 * No next block, just default.
2152 gotbno = NULLFSBLOCK;
2154 * If both valid, pick the better one, else the only good
2155 * one, else ap->blkno is already set (to 0 or the inode block).
2157 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2158 ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
2159 else if (prevbno != NULLFSBLOCK)
2160 ap->blkno = prevbno;
2161 else if (gotbno != NULLFSBLOCK)
2169 xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2171 xfs_alloctype_t atype = 0; /* type for allocation routines */
2172 int error; /* error return value */
2173 xfs_mount_t *mp; /* mount point structure */
2174 xfs_extlen_t prod = 0; /* product factor for allocators */
2175 xfs_extlen_t ralen = 0; /* realtime allocation length */
2176 xfs_extlen_t align; /* minimum allocation alignment */
2179 mp = ap->ip->i_mount;
2180 align = xfs_get_extsz_hint(ap->ip);
2181 prod = align / mp->m_sb.sb_rextsize;
2182 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
2183 align, 1, ap->eof, 0,
2184 ap->conv, &ap->offset, &ap->length);
2188 ASSERT(ap->length % mp->m_sb.sb_rextsize == 0);
2191 * If the offset & length are not perfectly aligned
2192 * then kill prod, it will just get us in trouble.
2194 if (do_mod(ap->offset, align) || ap->length % align)
2197 * Set ralen to be the actual requested length in rtextents.
2199 ralen = ap->length / mp->m_sb.sb_rextsize;
2201 * If the old value was close enough to MAXEXTLEN that
2202 * we rounded up to it, cut it back so it's valid again.
2203 * Note that if it's a really large request (bigger than
2204 * MAXEXTLEN), we don't hear about that number, and can't
2205 * adjust the starting point to match it.
2207 if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2208 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2211 * Lock out other modifications to the RT bitmap inode.
2213 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
2214 xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
2217 * If it's an allocation to an empty file at offset 0,
2218 * pick an extent that will space things out in the rt area.
2220 if (ap->eof && ap->offset == 0) {
2221 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
2223 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2226 ap->blkno = rtx * mp->m_sb.sb_rextsize;
2231 xfs_bmap_adjacent(ap);
2234 * Realtime allocation, done through xfs_rtallocate_extent.
2236 atype = ap->blkno == 0 ? XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2237 do_div(ap->blkno, mp->m_sb.sb_rextsize);
2240 if ((error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length,
2241 &ralen, atype, ap->wasdel, prod, &rtb)))
2243 if (rtb == NULLFSBLOCK && prod > 1 &&
2244 (error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1,
2245 ap->length, &ralen, atype,
2246 ap->wasdel, 1, &rtb)))
2249 if (ap->blkno != NULLFSBLOCK) {
2250 ap->blkno *= mp->m_sb.sb_rextsize;
2251 ralen *= mp->m_sb.sb_rextsize;
2253 ap->ip->i_d.di_nblocks += ralen;
2254 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2256 ap->ip->i_delayed_blks -= ralen;
2258 * Adjust the disk quota also. This was reserved
2261 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2262 ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2263 XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2271 xfs_bmap_btalloc_nullfb(
2272 struct xfs_bmalloca *ap,
2273 struct xfs_alloc_arg *args,
2276 struct xfs_mount *mp = ap->ip->i_mount;
2277 struct xfs_perag *pag;
2278 xfs_agnumber_t ag, startag;
2282 if (ap->userdata && xfs_inode_is_filestream(ap->ip))
2283 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2285 args->type = XFS_ALLOCTYPE_START_BNO;
2286 args->total = ap->total;
2289 * Search for an allocation group with a single extent large enough
2290 * for the request. If one isn't found, then adjust the minimum
2291 * allocation size to the largest space found.
2293 startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
2294 if (startag == NULLAGNUMBER)
2297 pag = xfs_perag_get(mp, ag);
2298 while (*blen < args->maxlen) {
2299 if (!pag->pagf_init) {
2300 error = xfs_alloc_pagf_init(mp, args->tp, ag,
2301 XFS_ALLOC_FLAG_TRYLOCK);
2309 * See xfs_alloc_fix_freelist...
2311 if (pag->pagf_init) {
2312 xfs_extlen_t longest;
2313 longest = xfs_alloc_longest_free_extent(mp, pag);
2314 if (*blen < longest)
2319 if (xfs_inode_is_filestream(ap->ip)) {
2320 if (*blen >= args->maxlen)
2325 * If startag is an invalid AG, we've
2326 * come here once before and
2327 * xfs_filestream_new_ag picked the
2328 * best currently available.
2330 * Don't continue looping, since we
2331 * could loop forever.
2333 if (startag == NULLAGNUMBER)
2336 error = xfs_filestream_new_ag(ap, &ag);
2341 /* loop again to set 'blen'*/
2342 startag = NULLAGNUMBER;
2343 pag = xfs_perag_get(mp, ag);
2347 if (++ag == mp->m_sb.sb_agcount)
2352 pag = xfs_perag_get(mp, ag);
2357 * Since the above loop did a BUF_TRYLOCK, it is
2358 * possible that there is space for this request.
2360 if (notinit || *blen < ap->minlen)
2361 args->minlen = ap->minlen;
2363 * If the best seen length is less than the request
2364 * length, use the best as the minimum.
2366 else if (*blen < args->maxlen)
2367 args->minlen = *blen;
2369 * Otherwise we've seen an extent as big as maxlen,
2370 * use that as the minimum.
2373 args->minlen = args->maxlen;
2376 * set the failure fallback case to look in the selected
2377 * AG as the stream may have moved.
2379 if (xfs_inode_is_filestream(ap->ip))
2380 ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
2387 xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2389 xfs_mount_t *mp; /* mount point structure */
2390 xfs_alloctype_t atype = 0; /* type for allocation routines */
2391 xfs_extlen_t align; /* minimum allocation alignment */
2392 xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
2394 xfs_alloc_arg_t args;
2396 xfs_extlen_t nextminlen = 0;
2397 int nullfb; /* true if ap->firstblock isn't set */
2404 mp = ap->ip->i_mount;
2405 align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
2406 if (unlikely(align)) {
2407 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
2408 align, 0, ap->eof, 0, ap->conv,
2409 &ap->offset, &ap->length);
2413 nullfb = *ap->firstblock == NULLFSBLOCK;
2414 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
2416 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
2417 ag = xfs_filestream_lookup_ag(ap->ip);
2418 ag = (ag != NULLAGNUMBER) ? ag : 0;
2419 ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
2421 ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2424 ap->blkno = *ap->firstblock;
2426 xfs_bmap_adjacent(ap);
2429 * If allowed, use ap->blkno; otherwise must use firstblock since
2430 * it's in the right allocation group.
2432 if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
2435 ap->blkno = *ap->firstblock;
2437 * Normal allocation, done through xfs_alloc_vextent.
2439 tryagain = isaligned = 0;
2440 memset(&args, 0, sizeof(args));
2443 args.fsbno = ap->blkno;
2445 /* Trim the allocation back to the maximum an AG can fit. */
2446 args.maxlen = MIN(ap->length, XFS_ALLOC_AG_MAX_USABLE(mp));
2447 args.firstblock = *ap->firstblock;
2450 error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
2453 } else if (ap->flist->xbf_low) {
2454 if (xfs_inode_is_filestream(ap->ip))
2455 args.type = XFS_ALLOCTYPE_FIRST_AG;
2457 args.type = XFS_ALLOCTYPE_START_BNO;
2458 args.total = args.minlen = ap->minlen;
2460 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2461 args.total = ap->total;
2462 args.minlen = ap->minlen;
2464 /* apply extent size hints if obtained earlier */
2465 if (unlikely(align)) {
2467 if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
2468 args.mod = (xfs_extlen_t)(args.prod - args.mod);
2469 } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
2473 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
2474 if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
2475 args.mod = (xfs_extlen_t)(args.prod - args.mod);
2478 * If we are not low on available data blocks, and the
2479 * underlying logical volume manager is a stripe, and
2480 * the file offset is zero then try to allocate data
2481 * blocks on stripe unit boundary.
2482 * NOTE: ap->aeof is only set if the allocation length
2483 * is >= the stripe unit and the allocation offset is
2484 * at the end of file.
2486 if (!ap->flist->xbf_low && ap->aeof) {
2488 args.alignment = mp->m_dalign;
2492 * Adjust for alignment
2494 if (blen > args.alignment && blen <= args.maxlen)
2495 args.minlen = blen - args.alignment;
2496 args.minalignslop = 0;
2499 * First try an exact bno allocation.
2500 * If it fails then do a near or start bno
2501 * allocation with alignment turned on.
2505 args.type = XFS_ALLOCTYPE_THIS_BNO;
2508 * Compute the minlen+alignment for the
2509 * next case. Set slop so that the value
2510 * of minlen+alignment+slop doesn't go up
2511 * between the calls.
2513 if (blen > mp->m_dalign && blen <= args.maxlen)
2514 nextminlen = blen - mp->m_dalign;
2516 nextminlen = args.minlen;
2517 if (nextminlen + mp->m_dalign > args.minlen + 1)
2519 nextminlen + mp->m_dalign -
2522 args.minalignslop = 0;
2526 args.minalignslop = 0;
2528 args.minleft = ap->minleft;
2529 args.wasdel = ap->wasdel;
2531 args.userdata = ap->userdata;
2532 if ((error = xfs_alloc_vextent(&args)))
2534 if (tryagain && args.fsbno == NULLFSBLOCK) {
2536 * Exact allocation failed. Now try with alignment
2540 args.fsbno = ap->blkno;
2541 args.alignment = mp->m_dalign;
2542 args.minlen = nextminlen;
2543 args.minalignslop = 0;
2545 if ((error = xfs_alloc_vextent(&args)))
2548 if (isaligned && args.fsbno == NULLFSBLOCK) {
2550 * allocation failed, so turn off alignment and
2554 args.fsbno = ap->blkno;
2556 if ((error = xfs_alloc_vextent(&args)))
2559 if (args.fsbno == NULLFSBLOCK && nullfb &&
2560 args.minlen > ap->minlen) {
2561 args.minlen = ap->minlen;
2562 args.type = XFS_ALLOCTYPE_START_BNO;
2563 args.fsbno = ap->blkno;
2564 if ((error = xfs_alloc_vextent(&args)))
2567 if (args.fsbno == NULLFSBLOCK && nullfb) {
2569 args.type = XFS_ALLOCTYPE_FIRST_AG;
2570 args.total = ap->minlen;
2572 if ((error = xfs_alloc_vextent(&args)))
2574 ap->flist->xbf_low = 1;
2576 if (args.fsbno != NULLFSBLOCK) {
2578 * check the allocation happened at the same or higher AG than
2579 * the first block that was allocated.
2581 ASSERT(*ap->firstblock == NULLFSBLOCK ||
2582 XFS_FSB_TO_AGNO(mp, *ap->firstblock) ==
2583 XFS_FSB_TO_AGNO(mp, args.fsbno) ||
2584 (ap->flist->xbf_low &&
2585 XFS_FSB_TO_AGNO(mp, *ap->firstblock) <
2586 XFS_FSB_TO_AGNO(mp, args.fsbno)));
2588 ap->blkno = args.fsbno;
2589 if (*ap->firstblock == NULLFSBLOCK)
2590 *ap->firstblock = args.fsbno;
2591 ASSERT(nullfb || fb_agno == args.agno ||
2592 (ap->flist->xbf_low && fb_agno < args.agno));
2593 ap->length = args.len;
2594 ap->ip->i_d.di_nblocks += args.len;
2595 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2597 ap->ip->i_delayed_blks -= args.len;
2599 * Adjust the disk quota also. This was reserved
2602 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2603 ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2604 XFS_TRANS_DQ_BCOUNT,
2607 ap->blkno = NULLFSBLOCK;
2614 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2615 * It figures out where to ask the underlying allocator to put the new extent.
2619 xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2621 if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
2622 return xfs_bmap_rtalloc(ap);
2623 return xfs_bmap_btalloc(ap);
2627 * Transform a btree format file with only one leaf node, where the
2628 * extents list will fit in the inode, into an extents format file.
2629 * Since the file extents are already in-core, all we have to do is
2630 * give up the space for the btree root and pitch the leaf block.
2632 STATIC int /* error */
2633 xfs_bmap_btree_to_extents(
2634 xfs_trans_t *tp, /* transaction pointer */
2635 xfs_inode_t *ip, /* incore inode pointer */
2636 xfs_btree_cur_t *cur, /* btree cursor */
2637 int *logflagsp, /* inode logging flags */
2638 int whichfork) /* data or attr fork */
2641 struct xfs_btree_block *cblock;/* child btree block */
2642 xfs_fsblock_t cbno; /* child block number */
2643 xfs_buf_t *cbp; /* child block's buffer */
2644 int error; /* error return value */
2645 xfs_ifork_t *ifp; /* inode fork data */
2646 xfs_mount_t *mp; /* mount point structure */
2647 __be64 *pp; /* ptr to block address */
2648 struct xfs_btree_block *rblock;/* root btree block */
2651 ifp = XFS_IFORK_PTR(ip, whichfork);
2652 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2653 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
2654 rblock = ifp->if_broot;
2655 ASSERT(be16_to_cpu(rblock->bb_level) == 1);
2656 ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
2657 ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
2658 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
2659 cbno = be64_to_cpu(*pp);
2662 if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
2665 error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF,
2669 cblock = XFS_BUF_TO_BLOCK(cbp);
2670 if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
2672 xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
2673 ip->i_d.di_nblocks--;
2674 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
2675 xfs_trans_binval(tp, cbp);
2676 if (cur->bc_bufs[0] == cbp)
2677 cur->bc_bufs[0] = NULL;
2678 xfs_iroot_realloc(ip, -1, whichfork);
2679 ASSERT(ifp->if_broot == NULL);
2680 ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
2681 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
2682 *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2687 * Called by xfs_bmapi to update file extent records and the btree
2688 * after removing space (or undoing a delayed allocation).
2690 STATIC int /* error */
2691 xfs_bmap_del_extent(
2692 xfs_inode_t *ip, /* incore inode pointer */
2693 xfs_trans_t *tp, /* current transaction pointer */
2694 xfs_extnum_t *idx, /* extent number to update/delete */
2695 xfs_bmap_free_t *flist, /* list of extents to be freed */
2696 xfs_btree_cur_t *cur, /* if null, not a btree */
2697 xfs_bmbt_irec_t *del, /* data to remove from extents */
2698 int *logflagsp, /* inode logging flags */
2699 int whichfork) /* data or attr fork */
2701 xfs_filblks_t da_new; /* new delay-alloc indirect blocks */
2702 xfs_filblks_t da_old; /* old delay-alloc indirect blocks */
2703 xfs_fsblock_t del_endblock=0; /* first block past del */
2704 xfs_fileoff_t del_endoff; /* first offset past del */
2705 int delay; /* current block is delayed allocated */
2706 int do_fx; /* free extent at end of routine */
2707 xfs_bmbt_rec_host_t *ep; /* current extent entry pointer */
2708 int error; /* error return value */
2709 int flags; /* inode logging flags */
2710 xfs_bmbt_irec_t got; /* current extent entry */
2711 xfs_fileoff_t got_endoff; /* first offset past got */
2712 int i; /* temp state */
2713 xfs_ifork_t *ifp; /* inode fork pointer */
2714 xfs_mount_t *mp; /* mount structure */
2715 xfs_filblks_t nblks; /* quota/sb block count */
2716 xfs_bmbt_irec_t new; /* new record to be inserted */
2718 uint qfield; /* quota field to update */
2719 xfs_filblks_t temp; /* for indirect length calculations */
2720 xfs_filblks_t temp2; /* for indirect length calculations */
2723 XFS_STATS_INC(xs_del_exlist);
2725 if (whichfork == XFS_ATTR_FORK)
2726 state |= BMAP_ATTRFORK;
2729 ifp = XFS_IFORK_PTR(ip, whichfork);
2730 ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
2731 (uint)sizeof(xfs_bmbt_rec_t)));
2732 ASSERT(del->br_blockcount > 0);
2733 ep = xfs_iext_get_ext(ifp, *idx);
2734 xfs_bmbt_get_all(ep, &got);
2735 ASSERT(got.br_startoff <= del->br_startoff);
2736 del_endoff = del->br_startoff + del->br_blockcount;
2737 got_endoff = got.br_startoff + got.br_blockcount;
2738 ASSERT(got_endoff >= del_endoff);
2739 delay = isnullstartblock(got.br_startblock);
2740 ASSERT(isnullstartblock(del->br_startblock) == delay);
2745 * If deleting a real allocation, must free up the disk space.
2748 flags = XFS_ILOG_CORE;
2750 * Realtime allocation. Free it and record di_nblocks update.
2752 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
2756 ASSERT(do_mod(del->br_blockcount,
2757 mp->m_sb.sb_rextsize) == 0);
2758 ASSERT(do_mod(del->br_startblock,
2759 mp->m_sb.sb_rextsize) == 0);
2760 bno = del->br_startblock;
2761 len = del->br_blockcount;
2762 do_div(bno, mp->m_sb.sb_rextsize);
2763 do_div(len, mp->m_sb.sb_rextsize);
2764 error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
2768 nblks = len * mp->m_sb.sb_rextsize;
2769 qfield = XFS_TRANS_DQ_RTBCOUNT;
2772 * Ordinary allocation.
2776 nblks = del->br_blockcount;
2777 qfield = XFS_TRANS_DQ_BCOUNT;
2780 * Set up del_endblock and cur for later.
2782 del_endblock = del->br_startblock + del->br_blockcount;
2784 if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
2785 got.br_startblock, got.br_blockcount,
2788 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2790 da_old = da_new = 0;
2792 da_old = startblockval(got.br_startblock);
2798 * Set flag value to use in switch statement.
2799 * Left-contig is 2, right-contig is 1.
2801 switch (((got.br_startoff == del->br_startoff) << 1) |
2802 (got_endoff == del_endoff)) {
2805 * Matches the whole extent. Delete the entry.
2807 xfs_iext_remove(ip, *idx, 1,
2808 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
2813 XFS_IFORK_NEXT_SET(ip, whichfork,
2814 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2815 flags |= XFS_ILOG_CORE;
2817 flags |= xfs_ilog_fext(whichfork);
2820 if ((error = xfs_btree_delete(cur, &i)))
2822 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2827 * Deleting the first part of the extent.
2829 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2830 xfs_bmbt_set_startoff(ep, del_endoff);
2831 temp = got.br_blockcount - del->br_blockcount;
2832 xfs_bmbt_set_blockcount(ep, temp);
2834 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2836 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2837 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2841 xfs_bmbt_set_startblock(ep, del_endblock);
2842 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2844 flags |= xfs_ilog_fext(whichfork);
2847 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
2848 got.br_blockcount - del->br_blockcount,
2855 * Deleting the last part of the extent.
2857 temp = got.br_blockcount - del->br_blockcount;
2858 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2859 xfs_bmbt_set_blockcount(ep, temp);
2861 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2863 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2864 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2868 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2870 flags |= xfs_ilog_fext(whichfork);
2873 if ((error = xfs_bmbt_update(cur, got.br_startoff,
2875 got.br_blockcount - del->br_blockcount,
2882 * Deleting the middle of the extent.
2884 temp = del->br_startoff - got.br_startoff;
2885 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2886 xfs_bmbt_set_blockcount(ep, temp);
2887 new.br_startoff = del_endoff;
2888 temp2 = got_endoff - del_endoff;
2889 new.br_blockcount = temp2;
2890 new.br_state = got.br_state;
2892 new.br_startblock = del_endblock;
2893 flags |= XFS_ILOG_CORE;
2895 if ((error = xfs_bmbt_update(cur,
2897 got.br_startblock, temp,
2900 if ((error = xfs_btree_increment(cur, 0, &i)))
2902 cur->bc_rec.b = new;
2903 error = xfs_btree_insert(cur, &i);
2904 if (error && error != ENOSPC)
2907 * If get no-space back from btree insert,
2908 * it tried a split, and we have a zero
2909 * block reservation.
2910 * Fix up our state and return the error.
2912 if (error == ENOSPC) {
2914 * Reset the cursor, don't trust
2915 * it after any insert operation.
2917 if ((error = xfs_bmbt_lookup_eq(cur,
2922 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2924 * Update the btree record back
2925 * to the original value.
2927 if ((error = xfs_bmbt_update(cur,
2934 * Reset the extent record back
2935 * to the original value.
2937 xfs_bmbt_set_blockcount(ep,
2940 error = XFS_ERROR(ENOSPC);
2943 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2945 flags |= xfs_ilog_fext(whichfork);
2946 XFS_IFORK_NEXT_SET(ip, whichfork,
2947 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2949 ASSERT(whichfork == XFS_DATA_FORK);
2950 temp = xfs_bmap_worst_indlen(ip, temp);
2951 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2952 temp2 = xfs_bmap_worst_indlen(ip, temp2);
2953 new.br_startblock = nullstartblock((int)temp2);
2954 da_new = temp + temp2;
2955 while (da_new > da_old) {
2959 xfs_bmbt_set_startblock(ep,
2960 nullstartblock((int)temp));
2962 if (da_new == da_old)
2968 nullstartblock((int)temp2);
2972 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2973 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
2978 * If we need to, add to list of extents to delete.
2981 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
2984 * Adjust inode # blocks in the file.
2987 ip->i_d.di_nblocks -= nblks;
2989 * Adjust quota data.
2992 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
2995 * Account for change in delayed indirect blocks.
2996 * Nothing to do for disk quota accounting here.
2998 ASSERT(da_old >= da_new);
2999 if (da_old > da_new) {
3000 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
3001 (int64_t)(da_old - da_new), 0);
3009 * Remove the entry "free" from the free item list. Prev points to the
3010 * previous entry, unless "free" is the head of the list.
3014 xfs_bmap_free_t *flist, /* free item list header */
3015 xfs_bmap_free_item_t *prev, /* previous item on list, if any */
3016 xfs_bmap_free_item_t *free) /* list item to be freed */
3019 prev->xbfi_next = free->xbfi_next;
3021 flist->xbf_first = free->xbfi_next;
3023 kmem_zone_free(xfs_bmap_free_item_zone, free);
3027 * Convert an extents-format file into a btree-format file.
3028 * The new file will have a root block (in the inode) and a single child block.
3030 STATIC int /* error */
3031 xfs_bmap_extents_to_btree(
3032 xfs_trans_t *tp, /* transaction pointer */
3033 xfs_inode_t *ip, /* incore inode pointer */
3034 xfs_fsblock_t *firstblock, /* first-block-allocated */
3035 xfs_bmap_free_t *flist, /* blocks freed in xaction */
3036 xfs_btree_cur_t **curp, /* cursor returned to caller */
3037 int wasdel, /* converting a delayed alloc */
3038 int *logflagsp, /* inode logging flags */
3039 int whichfork) /* data or attr fork */
3041 struct xfs_btree_block *ablock; /* allocated (child) bt block */
3042 xfs_buf_t *abp; /* buffer for ablock */
3043 xfs_alloc_arg_t args; /* allocation arguments */
3044 xfs_bmbt_rec_t *arp; /* child record pointer */
3045 struct xfs_btree_block *block; /* btree root block */
3046 xfs_btree_cur_t *cur; /* bmap btree cursor */
3047 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
3048 int error; /* error return value */
3049 xfs_extnum_t i, cnt; /* extent record index */
3050 xfs_ifork_t *ifp; /* inode fork pointer */
3051 xfs_bmbt_key_t *kp; /* root block key pointer */
3052 xfs_mount_t *mp; /* mount structure */
3053 xfs_extnum_t nextents; /* number of file extents */
3054 xfs_bmbt_ptr_t *pp; /* root block address pointer */
3056 ifp = XFS_IFORK_PTR(ip, whichfork);
3057 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3060 * Make space in the inode incore.
3062 xfs_iroot_realloc(ip, 1, whichfork);
3063 ifp->if_flags |= XFS_IFBROOT;
3068 block = ifp->if_broot;
3069 block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3070 block->bb_level = cpu_to_be16(1);
3071 block->bb_numrecs = cpu_to_be16(1);
3072 block->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3073 block->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3076 * Need a cursor. Can't allocate until bb_level is filled in.
3079 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
3080 cur->bc_private.b.firstblock = *firstblock;
3081 cur->bc_private.b.flist = flist;
3082 cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3084 * Convert to a btree with two levels, one record in root.
3086 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3087 memset(&args, 0, sizeof(args));
3090 args.firstblock = *firstblock;
3091 if (*firstblock == NULLFSBLOCK) {
3092 args.type = XFS_ALLOCTYPE_START_BNO;
3093 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3094 } else if (flist->xbf_low) {
3095 args.type = XFS_ALLOCTYPE_START_BNO;
3096 args.fsbno = *firstblock;
3098 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3099 args.fsbno = *firstblock;
3101 args.minlen = args.maxlen = args.prod = 1;
3102 args.wasdel = wasdel;
3104 if ((error = xfs_alloc_vextent(&args))) {
3105 xfs_iroot_realloc(ip, -1, whichfork);
3106 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3110 * Allocation can't fail, the space was reserved.
3112 ASSERT(args.fsbno != NULLFSBLOCK);
3113 ASSERT(*firstblock == NULLFSBLOCK ||
3114 args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3116 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3117 *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3118 cur->bc_private.b.allocated++;
3119 ip->i_d.di_nblocks++;
3120 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3121 abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3123 * Fill in the child block.
3125 abp->b_ops = &xfs_bmbt_buf_ops;
3126 ablock = XFS_BUF_TO_BLOCK(abp);
3127 ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3128 ablock->bb_level = 0;
3129 ablock->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3130 ablock->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3131 arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3132 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3133 for (cnt = i = 0; i < nextents; i++) {
3134 ep = xfs_iext_get_ext(ifp, i);
3135 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
3136 arp->l0 = cpu_to_be64(ep->l0);
3137 arp->l1 = cpu_to_be64(ep->l1);
3141 ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3142 xfs_btree_set_numrecs(ablock, cnt);
3145 * Fill in the root key and pointer.
3147 kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
3148 arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3149 kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
3150 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
3151 be16_to_cpu(block->bb_level)));
3152 *pp = cpu_to_be64(args.fsbno);
3155 * Do all this logging at the end so that
3156 * the root is at the right level.
3158 xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
3159 xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3160 ASSERT(*curp == NULL);
3162 *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
3167 * Calculate the default attribute fork offset for newly created inodes.
3170 xfs_default_attroffset(
3171 struct xfs_inode *ip)
3173 struct xfs_mount *mp = ip->i_mount;
3176 if (mp->m_sb.sb_inodesize == 256) {
3177 offset = XFS_LITINO(mp) -
3178 XFS_BMDR_SPACE_CALC(MINABTPTRS);
3180 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
3183 ASSERT(offset < XFS_LITINO(mp));
3188 * Helper routine to reset inode di_forkoff field when switching
3189 * attribute fork from local to extent format - we reset it where
3190 * possible to make space available for inline data fork extents.
3193 xfs_bmap_forkoff_reset(
3198 if (whichfork == XFS_ATTR_FORK &&
3199 ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
3200 ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
3201 ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
3202 uint dfl_forkoff = xfs_default_attroffset(ip) >> 3;
3204 if (dfl_forkoff > ip->i_d.di_forkoff)
3205 ip->i_d.di_forkoff = dfl_forkoff;
3210 * Convert a local file to an extents file.
3211 * This code is out of bounds for data forks of regular files,
3212 * since the file data needs to get logged so things will stay consistent.
3213 * (The bmap-level manipulations are ok, though).
3215 STATIC int /* error */
3216 xfs_bmap_local_to_extents(
3217 xfs_trans_t *tp, /* transaction pointer */
3218 xfs_inode_t *ip, /* incore inode pointer */
3219 xfs_fsblock_t *firstblock, /* first block allocated in xaction */
3220 xfs_extlen_t total, /* total blocks needed by transaction */
3221 int *logflagsp, /* inode logging flags */
3222 int whichfork) /* data or attr fork */
3224 int error; /* error return value */
3225 int flags; /* logging flags returned */
3226 xfs_ifork_t *ifp; /* inode fork pointer */
3229 * We don't want to deal with the case of keeping inode data inline yet.
3230 * So sending the data fork of a regular inode is invalid.
3232 ASSERT(!(S_ISREG(ip->i_d.di_mode) && whichfork == XFS_DATA_FORK));
3233 ifp = XFS_IFORK_PTR(ip, whichfork);
3234 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3237 if (ifp->if_bytes) {
3238 xfs_alloc_arg_t args; /* allocation arguments */
3239 xfs_buf_t *bp; /* buffer for extent block */
3240 xfs_bmbt_rec_host_t *ep;/* extent record pointer */
3242 memset(&args, 0, sizeof(args));
3244 args.mp = ip->i_mount;
3245 args.firstblock = *firstblock;
3246 ASSERT((ifp->if_flags &
3247 (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == XFS_IFINLINE);
3249 * Allocate a block. We know we need only one, since the
3250 * file currently fits in an inode.
3252 if (*firstblock == NULLFSBLOCK) {
3253 args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3254 args.type = XFS_ALLOCTYPE_START_BNO;
3256 args.fsbno = *firstblock;
3257 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3260 args.minlen = args.maxlen = args.prod = 1;
3261 if ((error = xfs_alloc_vextent(&args)))
3264 * Can't fail, the space was reserved.
3266 ASSERT(args.fsbno != NULLFSBLOCK);
3267 ASSERT(args.len == 1);
3268 *firstblock = args.fsbno;
3269 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3270 bp->b_ops = &xfs_bmbt_buf_ops;
3271 memcpy(bp->b_addr, ifp->if_u1.if_data, ifp->if_bytes);
3272 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3273 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3274 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3275 xfs_iext_add(ifp, 0, 1);
3276 ep = xfs_iext_get_ext(ifp, 0);
3277 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3278 trace_xfs_bmap_post_update(ip, 0,
3279 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
3281 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3282 ip->i_d.di_nblocks = 1;
3283 xfs_trans_mod_dquot_byino(tp, ip,
3284 XFS_TRANS_DQ_BCOUNT, 1L);
3285 flags |= xfs_ilog_fext(whichfork);
3287 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3288 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3290 ifp->if_flags &= ~XFS_IFINLINE;
3291 ifp->if_flags |= XFS_IFEXTENTS;
3292 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3293 flags |= XFS_ILOG_CORE;
3300 * Search the extent records for the entry containing block bno.
3301 * If bno lies in a hole, point to the next entry. If bno lies
3302 * past eof, *eofp will be set, and *prevp will contain the last
3303 * entry (null if none). Else, *lastxp will be set to the index
3304 * of the found entry; *gotp will contain the entry.
3306 STATIC xfs_bmbt_rec_host_t * /* pointer to found extent entry */
3307 xfs_bmap_search_multi_extents(
3308 xfs_ifork_t *ifp, /* inode fork pointer */
3309 xfs_fileoff_t bno, /* block number searched for */
3310 int *eofp, /* out: end of file found */
3311 xfs_extnum_t *lastxp, /* out: last extent index */
3312 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
3313 xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
3315 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
3316 xfs_extnum_t lastx; /* last extent index */
3319 * Initialize the extent entry structure to catch access to
3320 * uninitialized br_startblock field.
3322 gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
3323 gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3324 gotp->br_state = XFS_EXT_INVALID;
3326 gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
3328 gotp->br_startblock = 0xffffa5a5;
3330 prevp->br_startoff = NULLFILEOFF;
3332 ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
3334 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
3336 if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
3337 xfs_bmbt_get_all(ep, gotp);
3351 * Search the extents list for the inode, for the extent containing bno.
3352 * If bno lies in a hole, point to the next entry. If bno lies past eof,
3353 * *eofp will be set, and *prevp will contain the last entry (null if none).
3354 * Else, *lastxp will be set to the index of the found
3355 * entry; *gotp will contain the entry.
3357 STATIC xfs_bmbt_rec_host_t * /* pointer to found extent entry */
3358 xfs_bmap_search_extents(
3359 xfs_inode_t *ip, /* incore inode pointer */
3360 xfs_fileoff_t bno, /* block number searched for */
3361 int fork, /* data or attr fork */
3362 int *eofp, /* out: end of file found */
3363 xfs_extnum_t *lastxp, /* out: last extent index */
3364 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
3365 xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
3367 xfs_ifork_t *ifp; /* inode fork pointer */
3368 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
3370 XFS_STATS_INC(xs_look_exlist);
3371 ifp = XFS_IFORK_PTR(ip, fork);
3373 ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
3375 if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
3376 !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
3377 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
3378 "Access to block zero in inode %llu "
3379 "start_block: %llx start_off: %llx "
3380 "blkcnt: %llx extent-state: %x lastx: %x\n",
3381 (unsigned long long)ip->i_ino,
3382 (unsigned long long)gotp->br_startblock,
3383 (unsigned long long)gotp->br_startoff,
3384 (unsigned long long)gotp->br_blockcount,
3385 gotp->br_state, *lastxp);
3386 *lastxp = NULLEXTNUM;
3394 * Compute the worst-case number of indirect blocks that will be used
3395 * for ip's delayed extent of length "len".
3397 STATIC xfs_filblks_t
3398 xfs_bmap_worst_indlen(
3399 xfs_inode_t *ip, /* incore inode pointer */
3400 xfs_filblks_t len) /* delayed extent length */
3402 int level; /* btree level number */
3403 int maxrecs; /* maximum record count at this level */
3404 xfs_mount_t *mp; /* mount structure */
3405 xfs_filblks_t rval; /* return value */
3408 maxrecs = mp->m_bmap_dmxr[0];
3409 for (level = 0, rval = 0;
3410 level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3413 do_div(len, maxrecs);
3416 return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3419 maxrecs = mp->m_bmap_dmxr[1];
3425 * Convert inode from non-attributed to attributed.
3426 * Must not be in a transaction, ip must not be locked.
3428 int /* error code */
3429 xfs_bmap_add_attrfork(
3430 xfs_inode_t *ip, /* incore inode pointer */
3431 int size, /* space new attribute needs */
3432 int rsvd) /* xact may use reserved blks */
3434 xfs_fsblock_t firstblock; /* 1st block/ag allocated */
3435 xfs_bmap_free_t flist; /* freed extent records */
3436 xfs_mount_t *mp; /* mount structure */
3437 xfs_trans_t *tp; /* transaction pointer */
3438 int blks; /* space reservation */
3439 int version = 1; /* superblock attr version */
3440 int committed; /* xaction was committed */
3441 int logflags; /* logging flags */
3442 int error; /* error return value */
3444 ASSERT(XFS_IFORK_Q(ip) == 0);
3447 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3448 tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3449 blks = XFS_ADDAFORK_SPACE_RES(mp);
3451 tp->t_flags |= XFS_TRANS_RESERVE;
3452 if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3453 XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3455 xfs_ilock(ip, XFS_ILOCK_EXCL);
3456 error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
3457 XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3458 XFS_QMOPT_RES_REGBLKS);
3460 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3461 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3464 if (XFS_IFORK_Q(ip))
3466 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3468 * For inodes coming from pre-6.2 filesystems.
3470 ASSERT(ip->i_d.di_aformat == 0);
3471 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
3473 ASSERT(ip->i_d.di_anextents == 0);
3475 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3476 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3478 switch (ip->i_d.di_format) {
3479 case XFS_DINODE_FMT_DEV:
3480 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
3482 case XFS_DINODE_FMT_UUID:
3483 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
3485 case XFS_DINODE_FMT_LOCAL:
3486 case XFS_DINODE_FMT_EXTENTS:
3487 case XFS_DINODE_FMT_BTREE:
3488 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
3489 if (!ip->i_d.di_forkoff)
3490 ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
3491 else if (mp->m_flags & XFS_MOUNT_ATTR2)
3496 error = XFS_ERROR(EINVAL);
3500 ASSERT(ip->i_afp == NULL);
3501 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
3502 ip->i_afp->if_flags = XFS_IFEXTENTS;
3504 xfs_bmap_init(&flist, &firstblock);
3505 switch (ip->i_d.di_format) {
3506 case XFS_DINODE_FMT_LOCAL:
3507 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
3510 case XFS_DINODE_FMT_EXTENTS:
3511 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
3514 case XFS_DINODE_FMT_BTREE:
3515 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
3523 xfs_trans_log_inode(tp, ip, logflags);
3526 if (!xfs_sb_version_hasattr(&mp->m_sb) ||
3527 (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
3528 __int64_t sbfields = 0;
3530 spin_lock(&mp->m_sb_lock);
3531 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
3532 xfs_sb_version_addattr(&mp->m_sb);
3533 sbfields |= XFS_SB_VERSIONNUM;
3535 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
3536 xfs_sb_version_addattr2(&mp->m_sb);
3537 sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
3540 spin_unlock(&mp->m_sb_lock);
3541 xfs_mod_sb(tp, sbfields);
3543 spin_unlock(&mp->m_sb_lock);
3546 error = xfs_bmap_finish(&tp, &flist, &committed);
3549 return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3551 xfs_bmap_cancel(&flist);
3553 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3555 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
3560 * Add the extent to the list of extents to be free at transaction end.
3561 * The list is maintained sorted (by block number).
3566 xfs_fsblock_t bno, /* fs block number of extent */
3567 xfs_filblks_t len, /* length of extent */
3568 xfs_bmap_free_t *flist, /* list of extents */
3569 xfs_mount_t *mp) /* mount point structure */
3571 xfs_bmap_free_item_t *cur; /* current (next) element */
3572 xfs_bmap_free_item_t *new; /* new element */
3573 xfs_bmap_free_item_t *prev; /* previous element */
3575 xfs_agnumber_t agno;
3576 xfs_agblock_t agbno;
3578 ASSERT(bno != NULLFSBLOCK);
3580 ASSERT(len <= MAXEXTLEN);
3581 ASSERT(!isnullstartblock(bno));
3582 agno = XFS_FSB_TO_AGNO(mp, bno);
3583 agbno = XFS_FSB_TO_AGBNO(mp, bno);
3584 ASSERT(agno < mp->m_sb.sb_agcount);
3585 ASSERT(agbno < mp->m_sb.sb_agblocks);
3586 ASSERT(len < mp->m_sb.sb_agblocks);
3587 ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
3589 ASSERT(xfs_bmap_free_item_zone != NULL);
3590 new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
3591 new->xbfi_startblock = bno;
3592 new->xbfi_blockcount = (xfs_extlen_t)len;
3593 for (prev = NULL, cur = flist->xbf_first;
3595 prev = cur, cur = cur->xbfi_next) {
3596 if (cur->xbfi_startblock >= bno)
3600 prev->xbfi_next = new;
3602 flist->xbf_first = new;
3603 new->xbfi_next = cur;
3608 * Compute and fill in the value of the maximum depth of a bmap btree
3609 * in this filesystem. Done once, during mount.
3612 xfs_bmap_compute_maxlevels(
3613 xfs_mount_t *mp, /* file system mount structure */
3614 int whichfork) /* data or attr fork */
3616 int level; /* btree level */
3617 uint maxblocks; /* max blocks at this level */
3618 uint maxleafents; /* max leaf entries possible */
3619 int maxrootrecs; /* max records in root block */
3620 int minleafrecs; /* min records in leaf block */
3621 int minnoderecs; /* min records in node block */
3622 int sz; /* root block size */
3625 * The maximum number of extents in a file, hence the maximum
3626 * number of leaf entries, is controlled by the type of di_nextents
3627 * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
3628 * (a signed 16-bit number, xfs_aextnum_t).
3630 * Note that we can no longer assume that if we are in ATTR1 that
3631 * the fork offset of all the inodes will be
3632 * (xfs_default_attroffset(ip) >> 3) because we could have mounted
3633 * with ATTR2 and then mounted back with ATTR1, keeping the
3634 * di_forkoff's fixed but probably at various positions. Therefore,
3635 * for both ATTR1 and ATTR2 we have to assume the worst case scenario
3636 * of a minimum size available.
3638 if (whichfork == XFS_DATA_FORK) {
3639 maxleafents = MAXEXTNUM;
3640 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
3642 maxleafents = MAXAEXTNUM;
3643 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
3645 maxrootrecs = xfs_bmdr_maxrecs(mp, sz, 0);
3646 minleafrecs = mp->m_bmap_dmnr[0];
3647 minnoderecs = mp->m_bmap_dmnr[1];
3648 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
3649 for (level = 1; maxblocks > 1; level++) {
3650 if (maxblocks <= maxrootrecs)
3653 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
3655 mp->m_bm_maxlevels[whichfork] = level;
3659 * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
3660 * caller. Frees all the extents that need freeing, which must be done
3661 * last due to locking considerations. We never free any extents in
3662 * the first transaction.
3664 * Return 1 if the given transaction was committed and a new one
3665 * started, and 0 otherwise in the committed parameter.
3669 xfs_trans_t **tp, /* transaction pointer addr */
3670 xfs_bmap_free_t *flist, /* i/o: list extents to free */
3671 int *committed) /* xact committed or not */
3673 xfs_efd_log_item_t *efd; /* extent free data */
3674 xfs_efi_log_item_t *efi; /* extent free intention */
3675 int error; /* error return value */
3676 xfs_bmap_free_item_t *free; /* free extent item */
3677 unsigned int logres; /* new log reservation */
3678 unsigned int logcount; /* new log count */
3679 xfs_mount_t *mp; /* filesystem mount structure */
3680 xfs_bmap_free_item_t *next; /* next item on free list */
3681 xfs_trans_t *ntp; /* new transaction pointer */
3683 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
3684 if (flist->xbf_count == 0) {
3689 efi = xfs_trans_get_efi(ntp, flist->xbf_count);
3690 for (free = flist->xbf_first; free; free = free->xbfi_next)
3691 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
3692 free->xbfi_blockcount);
3693 logres = ntp->t_log_res;
3694 logcount = ntp->t_log_count;
3695 ntp = xfs_trans_dup(*tp);
3696 error = xfs_trans_commit(*tp, 0);
3700 * We have a new transaction, so we should return committed=1,
3701 * even though we're returning an error.
3707 * transaction commit worked ok so we can drop the extra ticket
3708 * reference that we gained in xfs_trans_dup()
3710 xfs_log_ticket_put(ntp->t_ticket);
3712 if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
3715 efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
3716 for (free = flist->xbf_first; free != NULL; free = next) {
3717 next = free->xbfi_next;
3718 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
3719 free->xbfi_blockcount))) {
3721 * The bmap free list will be cleaned up at a
3722 * higher level. The EFI will be canceled when
3723 * this transaction is aborted.
3724 * Need to force shutdown here to make sure it
3725 * happens, since this transaction may not be
3729 if (!XFS_FORCED_SHUTDOWN(mp))
3730 xfs_force_shutdown(mp,
3731 (error == EFSCORRUPTED) ?
3732 SHUTDOWN_CORRUPT_INCORE :
3733 SHUTDOWN_META_IO_ERROR);
3736 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
3737 free->xbfi_blockcount);
3738 xfs_bmap_del_free(flist, NULL, free);
3744 * Free up any items left in the list.
3748 xfs_bmap_free_t *flist) /* list of bmap_free_items */
3750 xfs_bmap_free_item_t *free; /* free list item */
3751 xfs_bmap_free_item_t *next;
3753 if (flist->xbf_count == 0)
3755 ASSERT(flist->xbf_first != NULL);
3756 for (free = flist->xbf_first; free; free = next) {
3757 next = free->xbfi_next;
3758 xfs_bmap_del_free(flist, NULL, free);
3760 ASSERT(flist->xbf_count == 0);
3764 * Returns the file-relative block number of the first unused block(s)
3765 * in the file with at least "len" logically contiguous blocks free.
3766 * This is the lowest-address hole if the file has holes, else the first block
3767 * past the end of file.
3768 * Return 0 if the file is currently local (in-inode).
3771 xfs_bmap_first_unused(
3772 xfs_trans_t *tp, /* transaction pointer */
3773 xfs_inode_t *ip, /* incore inode */
3774 xfs_extlen_t len, /* size of hole to find */
3775 xfs_fileoff_t *first_unused, /* unused block */
3776 int whichfork) /* data or attr fork */
3778 int error; /* error return value */
3779 int idx; /* extent record index */
3780 xfs_ifork_t *ifp; /* inode fork pointer */
3781 xfs_fileoff_t lastaddr; /* last block number seen */
3782 xfs_fileoff_t lowest; /* lowest useful block */
3783 xfs_fileoff_t max; /* starting useful block */
3784 xfs_fileoff_t off; /* offset for this block */
3785 xfs_extnum_t nextents; /* number of extent entries */
3787 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
3788 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
3789 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3790 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3794 ifp = XFS_IFORK_PTR(ip, whichfork);
3795 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3796 (error = xfs_iread_extents(tp, ip, whichfork)))
3798 lowest = *first_unused;
3799 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3800 for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
3801 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
3802 off = xfs_bmbt_get_startoff(ep);
3804 * See if the hole before this extent will work.
3806 if (off >= lowest + len && off - max >= len) {
3807 *first_unused = max;
3810 lastaddr = off + xfs_bmbt_get_blockcount(ep);
3811 max = XFS_FILEOFF_MAX(lastaddr, lowest);
3813 *first_unused = max;
3818 * Returns the file-relative block number of the last block + 1 before
3819 * last_block (input value) in the file.
3820 * This is not based on i_size, it is based on the extent records.
3821 * Returns 0 for local files, as they do not have extent records.
3824 xfs_bmap_last_before(
3825 xfs_trans_t *tp, /* transaction pointer */
3826 xfs_inode_t *ip, /* incore inode */
3827 xfs_fileoff_t *last_block, /* last block */
3828 int whichfork) /* data or attr fork */
3830 xfs_fileoff_t bno; /* input file offset */
3831 int eof; /* hit end of file */
3832 xfs_bmbt_rec_host_t *ep; /* pointer to last extent */
3833 int error; /* error return value */
3834 xfs_bmbt_irec_t got; /* current extent value */
3835 xfs_ifork_t *ifp; /* inode fork pointer */
3836 xfs_extnum_t lastx; /* last extent used */
3837 xfs_bmbt_irec_t prev; /* previous extent value */
3839 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
3840 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
3841 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
3842 return XFS_ERROR(EIO);
3843 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3847 ifp = XFS_IFORK_PTR(ip, whichfork);
3848 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3849 (error = xfs_iread_extents(tp, ip, whichfork)))
3851 bno = *last_block - 1;
3852 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
3854 if (eof || xfs_bmbt_get_startoff(ep) > bno) {
3855 if (prev.br_startoff == NULLFILEOFF)
3858 *last_block = prev.br_startoff + prev.br_blockcount;
3861 * Otherwise *last_block is already the right answer.
3867 xfs_bmap_last_extent(
3868 struct xfs_trans *tp,
3869 struct xfs_inode *ip,
3871 struct xfs_bmbt_irec *rec,
3874 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
3878 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
3879 error = xfs_iread_extents(tp, ip, whichfork);
3884 nextents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
3885 if (nextents == 0) {
3890 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
3896 * Check the last inode extent to determine whether this allocation will result
3897 * in blocks being allocated at the end of the file. When we allocate new data
3898 * blocks at the end of the file which do not start at the previous data block,
3899 * we will try to align the new blocks at stripe unit boundaries.
3901 * Returns 0 in bma->aeof if the file (fork) is empty as any new write will be
3902 * at, or past the EOF.
3906 struct xfs_bmalloca *bma,
3909 struct xfs_bmbt_irec rec;
3914 error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
3916 if (error || is_empty)
3920 * Check if we are allocation or past the last extent, or at least into
3921 * the last delayed allocated extent.
3923 bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
3924 (bma->offset >= rec.br_startoff &&
3925 isnullstartblock(rec.br_startblock));
3930 * Check if the endoff is outside the last extent. If so the caller will grow
3931 * the allocation to a stripe unit boundary. All offsets are considered outside
3932 * the end of file for an empty fork, so 1 is returned in *eof in that case.
3936 struct xfs_inode *ip,
3937 xfs_fileoff_t endoff,
3941 struct xfs_bmbt_irec rec;
3944 error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof);
3948 *eof = endoff >= rec.br_startoff + rec.br_blockcount;
3953 * Returns the file-relative block number of the first block past eof in
3954 * the file. This is not based on i_size, it is based on the extent records.
3955 * Returns 0 for local files, as they do not have extent records.
3958 xfs_bmap_last_offset(
3959 struct xfs_trans *tp,
3960 struct xfs_inode *ip,
3961 xfs_fileoff_t *last_block,
3964 struct xfs_bmbt_irec rec;
3970 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
3973 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
3974 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
3975 return XFS_ERROR(EIO);
3977 error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
3978 if (error || is_empty)
3981 *last_block = rec.br_startoff + rec.br_blockcount;
3986 * Returns whether the selected fork of the inode has exactly one
3987 * block or not. For the data fork we check this matches di_size,
3988 * implying the file's range is 0..bsize-1.
3990 int /* 1=>1 block, 0=>otherwise */
3992 xfs_inode_t *ip, /* incore inode */
3993 int whichfork) /* data or attr fork */
3995 xfs_bmbt_rec_host_t *ep; /* ptr to fork's extent */
3996 xfs_ifork_t *ifp; /* inode fork pointer */
3997 int rval; /* return value */
3998 xfs_bmbt_irec_t s; /* internal version of extent */
4001 if (whichfork == XFS_DATA_FORK)
4002 return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
4004 if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4006 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4008 ifp = XFS_IFORK_PTR(ip, whichfork);
4009 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4010 ep = xfs_iext_get_ext(ifp, 0);
4011 xfs_bmbt_get_all(ep, &s);
4012 rval = s.br_startoff == 0 && s.br_blockcount == 1;
4013 if (rval && whichfork == XFS_DATA_FORK)
4014 ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
4019 xfs_bmap_sanity_check(
4020 struct xfs_mount *mp,
4024 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
4026 if (block->bb_magic != cpu_to_be32(XFS_BMAP_MAGIC) ||
4027 be16_to_cpu(block->bb_level) != level ||
4028 be16_to_cpu(block->bb_numrecs) == 0 ||
4029 be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
4035 * Read in the extents to if_extents.
4036 * All inode fields are set up by caller, we just traverse the btree
4037 * and copy the records in. If the file system cannot contain unwritten
4038 * extents, the records are checked for no "state" flags.
4041 xfs_bmap_read_extents(
4042 xfs_trans_t *tp, /* transaction pointer */
4043 xfs_inode_t *ip, /* incore inode */
4044 int whichfork) /* data or attr fork */
4046 struct xfs_btree_block *block; /* current btree block */
4047 xfs_fsblock_t bno; /* block # of "block" */
4048 xfs_buf_t *bp; /* buffer for "block" */
4049 int error; /* error return value */
4050 xfs_exntfmt_t exntf; /* XFS_EXTFMT_NOSTATE, if checking */
4051 xfs_extnum_t i, j; /* index into the extents list */
4052 xfs_ifork_t *ifp; /* fork structure */
4053 int level; /* btree level, for checking */
4054 xfs_mount_t *mp; /* file system mount structure */
4055 __be64 *pp; /* pointer to block address */
4057 xfs_extnum_t room; /* number of entries there's room for */
4061 ifp = XFS_IFORK_PTR(ip, whichfork);
4062 exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4063 XFS_EXTFMT_INODE(ip);
4064 block = ifp->if_broot;
4066 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4068 level = be16_to_cpu(block->bb_level);
4070 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
4071 bno = be64_to_cpu(*pp);
4072 ASSERT(bno != NULLDFSBNO);
4073 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
4074 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
4076 * Go down the tree until leaf level is reached, following the first
4077 * pointer (leftmost) at each level.
4079 while (level-- > 0) {
4080 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4081 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
4084 block = XFS_BUF_TO_BLOCK(bp);
4085 XFS_WANT_CORRUPTED_GOTO(
4086 xfs_bmap_sanity_check(mp, bp, level),
4090 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
4091 bno = be64_to_cpu(*pp);
4092 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
4093 xfs_trans_brelse(tp, bp);
4096 * Here with bp and block set to the leftmost leaf node in the tree.
4098 room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4101 * Loop over all leaf nodes. Copy information to the extent records.
4104 xfs_bmbt_rec_t *frp;
4105 xfs_fsblock_t nextbno;
4106 xfs_extnum_t num_recs;
4109 num_recs = xfs_btree_get_numrecs(block);
4110 if (unlikely(i + num_recs > room)) {
4111 ASSERT(i + num_recs <= room);
4112 xfs_warn(ip->i_mount,
4113 "corrupt dinode %Lu, (btree extents).",
4114 (unsigned long long) ip->i_ino);
4115 XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
4116 XFS_ERRLEVEL_LOW, ip->i_mount, block);
4119 XFS_WANT_CORRUPTED_GOTO(
4120 xfs_bmap_sanity_check(mp, bp, 0),
4123 * Read-ahead the next leaf block, if any.
4125 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
4126 if (nextbno != NULLFSBLOCK)
4127 xfs_btree_reada_bufl(mp, nextbno, 1,
4130 * Copy records into the extent records.
4132 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
4134 for (j = 0; j < num_recs; j++, i++, frp++) {
4135 xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
4136 trp->l0 = be64_to_cpu(frp->l0);
4137 trp->l1 = be64_to_cpu(frp->l1);
4139 if (exntf == XFS_EXTFMT_NOSTATE) {
4141 * Check all attribute bmap btree records and
4142 * any "older" data bmap btree records for a
4143 * set bit in the "extent flag" position.
4145 if (unlikely(xfs_check_nostate_extents(ifp,
4146 start, num_recs))) {
4147 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4153 xfs_trans_brelse(tp, bp);
4156 * If we've reached the end, stop.
4158 if (bno == NULLFSBLOCK)
4160 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4161 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
4164 block = XFS_BUF_TO_BLOCK(bp);
4166 ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4167 ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4168 XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
4171 xfs_trans_brelse(tp, bp);
4172 return XFS_ERROR(EFSCORRUPTED);
4177 * Add bmap trace insert entries for all the contents of the extent records.
4180 xfs_bmap_trace_exlist(
4181 xfs_inode_t *ip, /* incore inode pointer */
4182 xfs_extnum_t cnt, /* count of entries in the list */
4183 int whichfork, /* data or attr fork */
4184 unsigned long caller_ip)
4186 xfs_extnum_t idx; /* extent record index */
4187 xfs_ifork_t *ifp; /* inode fork pointer */
4190 if (whichfork == XFS_ATTR_FORK)
4191 state |= BMAP_ATTRFORK;
4193 ifp = XFS_IFORK_PTR(ip, whichfork);
4194 ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4195 for (idx = 0; idx < cnt; idx++)
4196 trace_xfs_extlist(ip, idx, whichfork, caller_ip);
4200 * Validate that the bmbt_irecs being returned from bmapi are valid
4201 * given the callers original parameters. Specifically check the
4202 * ranges of the returned irecs to ensure that they only extent beyond
4203 * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4206 xfs_bmap_validate_ret(
4210 xfs_bmbt_irec_t *mval,
4214 int i; /* index to map values */
4216 ASSERT(ret_nmap <= nmap);
4218 for (i = 0; i < ret_nmap; i++) {
4219 ASSERT(mval[i].br_blockcount > 0);
4220 if (!(flags & XFS_BMAPI_ENTIRE)) {
4221 ASSERT(mval[i].br_startoff >= bno);
4222 ASSERT(mval[i].br_blockcount <= len);
4223 ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4226 ASSERT(mval[i].br_startoff < bno + len);
4227 ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4231 mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4232 mval[i].br_startoff);
4233 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4234 mval[i].br_startblock != HOLESTARTBLOCK);
4235 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4236 mval[i].br_state == XFS_EXT_UNWRITTEN);
4243 * Trim the returned map to the required bounds
4247 struct xfs_bmbt_irec *mval,
4248 struct xfs_bmbt_irec *got,
4256 if ((flags & XFS_BMAPI_ENTIRE) ||
4257 got->br_startoff + got->br_blockcount <= obno) {
4259 if (isnullstartblock(got->br_startblock))
4260 mval->br_startblock = DELAYSTARTBLOCK;
4266 ASSERT((*bno >= obno) || (n == 0));
4268 mval->br_startoff = *bno;
4269 if (isnullstartblock(got->br_startblock))
4270 mval->br_startblock = DELAYSTARTBLOCK;
4272 mval->br_startblock = got->br_startblock +
4273 (*bno - got->br_startoff);
4275 * Return the minimum of what we got and what we asked for for
4276 * the length. We can use the len variable here because it is
4277 * modified below and we could have been there before coming
4278 * here if the first part of the allocation didn't overlap what
4281 mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
4282 got->br_blockcount - (*bno - got->br_startoff));
4283 mval->br_state = got->br_state;
4284 ASSERT(mval->br_blockcount <= len);
4289 * Update and validate the extent map to return
4292 xfs_bmapi_update_map(
4293 struct xfs_bmbt_irec **map,
4301 xfs_bmbt_irec_t *mval = *map;
4303 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4304 ((mval->br_startoff + mval->br_blockcount) <= end));
4305 ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
4306 (mval->br_startoff < obno));
4308 *bno = mval->br_startoff + mval->br_blockcount;
4310 if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
4311 /* update previous map with new information */
4312 ASSERT(mval->br_startblock == mval[-1].br_startblock);
4313 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
4314 ASSERT(mval->br_state == mval[-1].br_state);
4315 mval[-1].br_blockcount = mval->br_blockcount;
4316 mval[-1].br_state = mval->br_state;
4317 } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
4318 mval[-1].br_startblock != DELAYSTARTBLOCK &&
4319 mval[-1].br_startblock != HOLESTARTBLOCK &&
4320 mval->br_startblock == mval[-1].br_startblock +
4321 mval[-1].br_blockcount &&
4322 ((flags & XFS_BMAPI_IGSTATE) ||
4323 mval[-1].br_state == mval->br_state)) {
4324 ASSERT(mval->br_startoff ==
4325 mval[-1].br_startoff + mval[-1].br_blockcount);
4326 mval[-1].br_blockcount += mval->br_blockcount;
4327 } else if (*n > 0 &&
4328 mval->br_startblock == DELAYSTARTBLOCK &&
4329 mval[-1].br_startblock == DELAYSTARTBLOCK &&
4330 mval->br_startoff ==
4331 mval[-1].br_startoff + mval[-1].br_blockcount) {
4332 mval[-1].br_blockcount += mval->br_blockcount;
4333 mval[-1].br_state = mval->br_state;
4334 } else if (!((*n == 0) &&
4335 ((mval->br_startoff + mval->br_blockcount) <=
4344 * Map file blocks to filesystem blocks without allocation.
4348 struct xfs_inode *ip,
4351 struct xfs_bmbt_irec *mval,
4355 struct xfs_mount *mp = ip->i_mount;
4356 struct xfs_ifork *ifp;
4357 struct xfs_bmbt_irec got;
4358 struct xfs_bmbt_irec prev;
4365 int whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4366 XFS_ATTR_FORK : XFS_DATA_FORK;
4369 ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
4370 XFS_BMAPI_IGSTATE)));
4372 if (unlikely(XFS_TEST_ERROR(
4373 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4374 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4375 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4376 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
4377 return XFS_ERROR(EFSCORRUPTED);
4380 if (XFS_FORCED_SHUTDOWN(mp))
4381 return XFS_ERROR(EIO);
4383 XFS_STATS_INC(xs_blk_mapr);
4385 ifp = XFS_IFORK_PTR(ip, whichfork);
4387 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4388 error = xfs_iread_extents(NULL, ip, whichfork);
4393 xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got, &prev);
4397 while (bno < end && n < *nmap) {
4398 /* Reading past eof, act as though there's a hole up to end. */
4400 got.br_startoff = end;
4401 if (got.br_startoff > bno) {
4402 /* Reading in a hole. */
4403 mval->br_startoff = bno;
4404 mval->br_startblock = HOLESTARTBLOCK;
4405 mval->br_blockcount =
4406 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4407 mval->br_state = XFS_EXT_NORM;
4408 bno += mval->br_blockcount;
4409 len -= mval->br_blockcount;
4415 /* set up the extent map to return. */
4416 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4417 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4419 /* If we're done, stop now. */
4420 if (bno >= end || n >= *nmap)
4423 /* Else go on to the next record. */
4424 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4425 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4434 xfs_bmapi_reserve_delalloc(
4435 struct xfs_inode *ip,
4438 struct xfs_bmbt_irec *got,
4439 struct xfs_bmbt_irec *prev,
4440 xfs_extnum_t *lastx,
4443 struct xfs_mount *mp = ip->i_mount;
4444 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4446 xfs_extlen_t indlen;
4447 char rt = XFS_IS_REALTIME_INODE(ip);
4451 alen = XFS_FILBLKS_MIN(len, MAXEXTLEN);
4453 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4455 /* Figure out the extent size, adjust alen */
4456 extsz = xfs_get_extsz_hint(ip);
4459 * Make sure we don't exceed a single extent length when we
4460 * align the extent by reducing length we are going to
4461 * allocate by the maximum amount extent size aligment may
4464 alen = XFS_FILBLKS_MIN(len, MAXEXTLEN - (2 * extsz - 1));
4465 error = xfs_bmap_extsize_align(mp, got, prev, extsz, rt, eof,
4466 1, 0, &aoff, &alen);
4471 extsz = alen / mp->m_sb.sb_rextsize;
4474 * Make a transaction-less quota reservation for delayed allocation
4475 * blocks. This number gets adjusted later. We return if we haven't
4476 * allocated blocks already inside this loop.
4478 error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4479 rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4484 * Split changing sb for alen and indlen since they could be coming
4485 * from different places.
4487 indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4491 error = xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
4492 -((int64_t)extsz), 0);
4494 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4495 -((int64_t)alen), 0);
4499 goto out_unreserve_quota;
4501 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4502 -((int64_t)indlen), 0);
4504 goto out_unreserve_blocks;
4507 ip->i_delayed_blks += alen;
4509 got->br_startoff = aoff;
4510 got->br_startblock = nullstartblock(indlen);
4511 got->br_blockcount = alen;
4512 got->br_state = XFS_EXT_NORM;
4513 xfs_bmap_add_extent_hole_delay(ip, lastx, got);
4516 * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4517 * might have merged it into one of the neighbouring ones.
4519 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), got);
4521 ASSERT(got->br_startoff <= aoff);
4522 ASSERT(got->br_startoff + got->br_blockcount >= aoff + alen);
4523 ASSERT(isnullstartblock(got->br_startblock));
4524 ASSERT(got->br_state == XFS_EXT_NORM);
4527 out_unreserve_blocks:
4529 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS, extsz, 0);
4531 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, alen, 0);
4532 out_unreserve_quota:
4533 if (XFS_IS_QUOTA_ON(mp))
4534 xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ?
4535 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4540 * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4544 struct xfs_inode *ip, /* incore inode */
4545 xfs_fileoff_t bno, /* starting file offs. mapped */
4546 xfs_filblks_t len, /* length to map in file */
4547 struct xfs_bmbt_irec *mval, /* output: map values */
4548 int *nmap, /* i/o: mval size/count */
4549 int flags) /* XFS_BMAPI_... */
4551 struct xfs_mount *mp = ip->i_mount;
4552 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4553 struct xfs_bmbt_irec got; /* current file extent record */
4554 struct xfs_bmbt_irec prev; /* previous file extent record */
4555 xfs_fileoff_t obno; /* old block number (offset) */
4556 xfs_fileoff_t end; /* end of mapped file region */
4557 xfs_extnum_t lastx; /* last useful extent number */
4558 int eof; /* we've hit the end of extents */
4559 int n = 0; /* current extent index */
4563 ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4564 ASSERT(!(flags & ~XFS_BMAPI_ENTIRE));
4566 if (unlikely(XFS_TEST_ERROR(
4567 (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4568 XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4569 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4570 XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW, mp);
4571 return XFS_ERROR(EFSCORRUPTED);
4574 if (XFS_FORCED_SHUTDOWN(mp))
4575 return XFS_ERROR(EIO);
4577 XFS_STATS_INC(xs_blk_mapw);
4579 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4580 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4585 xfs_bmap_search_extents(ip, bno, XFS_DATA_FORK, &eof, &lastx, &got, &prev);
4589 while (bno < end && n < *nmap) {
4590 if (eof || got.br_startoff > bno) {
4591 error = xfs_bmapi_reserve_delalloc(ip, bno, len, &got,
4592 &prev, &lastx, eof);
4602 /* set up the extent map to return. */
4603 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4604 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4606 /* If we're done, stop now. */
4607 if (bno >= end || n >= *nmap)
4610 /* Else go on to the next record. */
4612 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4613 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4624 __xfs_bmapi_allocate(
4625 struct xfs_bmalloca *bma)
4627 struct xfs_mount *mp = bma->ip->i_mount;
4628 int whichfork = (bma->flags & XFS_BMAPI_ATTRFORK) ?
4629 XFS_ATTR_FORK : XFS_DATA_FORK;
4630 struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4631 int tmp_logflags = 0;
4635 ASSERT(bma->length > 0);
4637 rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(bma->ip);
4640 * For the wasdelay case, we could also just allocate the stuff asked
4641 * for in this bmap call but that wouldn't be as good.
4644 bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4645 bma->offset = bma->got.br_startoff;
4646 if (bma->idx != NULLEXTNUM && bma->idx) {
4647 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4651 bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4653 bma->length = XFS_FILBLKS_MIN(bma->length,
4654 bma->got.br_startoff - bma->offset);
4658 * Indicate if this is the first user data in the file, or just any
4661 if (!(bma->flags & XFS_BMAPI_METADATA)) {
4662 bma->userdata = (bma->offset == 0) ?
4663 XFS_ALLOC_INITIAL_USER_DATA : XFS_ALLOC_USERDATA;
4666 bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4669 * Only want to do the alignment at the eof if it is userdata and
4670 * allocation length is larger than a stripe unit.
4672 if (mp->m_dalign && bma->length >= mp->m_dalign &&
4673 !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4674 error = xfs_bmap_isaeof(bma, whichfork);
4679 if (bma->flags & XFS_BMAPI_STACK_SWITCH)
4680 bma->stack_switch = 1;
4682 error = xfs_bmap_alloc(bma);
4686 if (bma->flist->xbf_low)
4689 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4690 if (bma->blkno == NULLFSBLOCK)
4692 if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4693 bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4694 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4695 bma->cur->bc_private.b.flist = bma->flist;
4698 * Bump the number of extents we've allocated
4704 bma->cur->bc_private.b.flags =
4705 bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4707 bma->got.br_startoff = bma->offset;
4708 bma->got.br_startblock = bma->blkno;
4709 bma->got.br_blockcount = bma->length;
4710 bma->got.br_state = XFS_EXT_NORM;
4713 * A wasdelay extent has been initialized, so shouldn't be flagged
4716 if (!bma->wasdel && (bma->flags & XFS_BMAPI_PREALLOC) &&
4717 xfs_sb_version_hasextflgbit(&mp->m_sb))
4718 bma->got.br_state = XFS_EXT_UNWRITTEN;
4721 error = xfs_bmap_add_extent_delay_real(bma);
4723 error = xfs_bmap_add_extent_hole_real(bma, whichfork);
4725 bma->logflags |= tmp_logflags;
4730 * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4731 * or xfs_bmap_add_extent_hole_real might have merged it into one of
4732 * the neighbouring ones.
4734 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4736 ASSERT(bma->got.br_startoff <= bma->offset);
4737 ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4738 bma->offset + bma->length);
4739 ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4740 bma->got.br_state == XFS_EXT_UNWRITTEN);
4745 xfs_bmapi_allocate_worker(
4746 struct work_struct *work)
4748 struct xfs_bmalloca *args = container_of(work,
4749 struct xfs_bmalloca, work);
4750 unsigned long pflags;
4752 /* we are in a transaction context here */
4753 current_set_flags_nested(&pflags, PF_FSTRANS);
4755 args->result = __xfs_bmapi_allocate(args);
4756 complete(args->done);
4758 current_restore_flags_nested(&pflags, PF_FSTRANS);
4762 * Some allocation requests often come in with little stack to work on. Push
4763 * them off to a worker thread so there is lots of stack to use. Otherwise just
4764 * call directly to avoid the context switch overhead here.
4768 struct xfs_bmalloca *args)
4770 DECLARE_COMPLETION_ONSTACK(done);
4772 if (!args->stack_switch)
4773 return __xfs_bmapi_allocate(args);
4777 INIT_WORK_ONSTACK(&args->work, xfs_bmapi_allocate_worker);
4778 queue_work(xfs_alloc_wq, &args->work);
4779 wait_for_completion(&done);
4780 return args->result;
4784 xfs_bmapi_convert_unwritten(
4785 struct xfs_bmalloca *bma,
4786 struct xfs_bmbt_irec *mval,
4790 int whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4791 XFS_ATTR_FORK : XFS_DATA_FORK;
4792 struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4793 int tmp_logflags = 0;
4796 /* check if we need to do unwritten->real conversion */
4797 if (mval->br_state == XFS_EXT_UNWRITTEN &&
4798 (flags & XFS_BMAPI_PREALLOC))
4801 /* check if we need to do real->unwritten conversion */
4802 if (mval->br_state == XFS_EXT_NORM &&
4803 (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4804 (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4808 * Modify (by adding) the state flag, if writing.
4810 ASSERT(mval->br_blockcount <= len);
4811 if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4812 bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4813 bma->ip, whichfork);
4814 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4815 bma->cur->bc_private.b.flist = bma->flist;
4817 mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4818 ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4820 error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, &bma->idx,
4821 &bma->cur, mval, bma->firstblock, bma->flist,
4823 bma->logflags |= tmp_logflags;
4828 * Update our extent pointer, given that
4829 * xfs_bmap_add_extent_unwritten_real might have merged it into one
4830 * of the neighbouring ones.
4832 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4835 * We may have combined previously unwritten space with written space,
4836 * so generate another request.
4838 if (mval->br_blockcount < len)
4844 * Map file blocks to filesystem blocks, and allocate blocks or convert the
4845 * extent state if necessary. Details behaviour is controlled by the flags
4846 * parameter. Only allocates blocks from a single allocation group, to avoid
4849 * The returned value in "firstblock" from the first call in a transaction
4850 * must be remembered and presented to subsequent calls in "firstblock".
4851 * An upper bound for the number of blocks to be allocated is supplied to
4852 * the first call in "total"; if no allocation group has that many free
4853 * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4857 struct xfs_trans *tp, /* transaction pointer */
4858 struct xfs_inode *ip, /* incore inode */
4859 xfs_fileoff_t bno, /* starting file offs. mapped */
4860 xfs_filblks_t len, /* length to map in file */
4861 int flags, /* XFS_BMAPI_... */
4862 xfs_fsblock_t *firstblock, /* first allocated block
4863 controls a.g. for allocs */
4864 xfs_extlen_t total, /* total blocks needed */
4865 struct xfs_bmbt_irec *mval, /* output: map values */
4866 int *nmap, /* i/o: mval size/count */
4867 struct xfs_bmap_free *flist) /* i/o: list extents to free */
4869 struct xfs_mount *mp = ip->i_mount;
4870 struct xfs_ifork *ifp;
4871 struct xfs_bmalloca bma = { 0 }; /* args for xfs_bmap_alloc */
4872 xfs_fileoff_t end; /* end of mapped file region */
4873 int eof; /* after the end of extents */
4874 int error; /* error return */
4875 int n; /* current extent index */
4876 xfs_fileoff_t obno; /* old block number (offset) */
4877 int whichfork; /* data or attr fork */
4878 char inhole; /* current location is hole in file */
4879 char wasdelay; /* old extent was delayed */
4882 xfs_fileoff_t orig_bno; /* original block number value */
4883 int orig_flags; /* original flags arg value */
4884 xfs_filblks_t orig_len; /* original value of len arg */
4885 struct xfs_bmbt_irec *orig_mval; /* original value of mval */
4886 int orig_nmap; /* original value of *nmap */
4896 ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4897 ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4901 whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4902 XFS_ATTR_FORK : XFS_DATA_FORK;
4904 if (unlikely(XFS_TEST_ERROR(
4905 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4906 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4907 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4908 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4909 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4910 return XFS_ERROR(EFSCORRUPTED);
4913 if (XFS_FORCED_SHUTDOWN(mp))
4914 return XFS_ERROR(EIO);
4916 ifp = XFS_IFORK_PTR(ip, whichfork);
4918 XFS_STATS_INC(xs_blk_mapw);
4920 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4921 error = xfs_bmap_local_to_extents(tp, ip, firstblock, total,
4922 &bma.logflags, whichfork);
4927 if (*firstblock == NULLFSBLOCK) {
4928 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4929 bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4936 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4937 error = xfs_iread_extents(tp, ip, whichfork);
4942 xfs_bmap_search_extents(ip, bno, whichfork, &eof, &bma.idx, &bma.got,
4953 bma.firstblock = firstblock;
4955 while (bno < end && n < *nmap) {
4956 inhole = eof || bma.got.br_startoff > bno;
4957 wasdelay = !inhole && isnullstartblock(bma.got.br_startblock);
4960 * First, deal with the hole before the allocated space
4961 * that we found, if any.
4963 if (inhole || wasdelay) {
4965 bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4966 bma.wasdel = wasdelay;
4971 * There's a 32/64 bit type mismatch between the
4972 * allocation length request (which can be 64 bits in
4973 * length) and the bma length request, which is
4974 * xfs_extlen_t and therefore 32 bits. Hence we have to
4975 * check for 32-bit overflows and handle them here.
4977 if (len > (xfs_filblks_t)MAXEXTLEN)
4978 bma.length = MAXEXTLEN;
4983 ASSERT(bma.length > 0);
4984 error = xfs_bmapi_allocate(&bma);
4987 if (bma.blkno == NULLFSBLOCK)
4991 /* Deal with the allocated space we found. */
4992 xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4995 /* Execute unwritten extent conversion if necessary */
4996 error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4997 if (error == EAGAIN)
5002 /* update the extent map to return */
5003 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
5006 * If we're done, stop now. Stop when we've allocated
5007 * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise
5008 * the transaction may get too big.
5010 if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
5013 /* Else go on to the next record. */
5015 if (++bma.idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t)) {
5016 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma.idx),
5024 * Transform from btree to extents, give it cur.
5026 if (xfs_bmap_wants_extents(ip, whichfork)) {
5027 int tmp_logflags = 0;
5030 error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
5031 &tmp_logflags, whichfork);
5032 bma.logflags |= tmp_logflags;
5037 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
5038 XFS_IFORK_NEXTENTS(ip, whichfork) >
5039 XFS_IFORK_MAXEXT(ip, whichfork));
5043 * Log everything. Do this after conversion, there's no point in
5044 * logging the extent records if we've converted to btree format.
5046 if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
5047 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5048 bma.logflags &= ~xfs_ilog_fext(whichfork);
5049 else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
5050 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5051 bma.logflags &= ~xfs_ilog_fbroot(whichfork);
5053 * Log whatever the flags say, even if error. Otherwise we might miss
5054 * detecting a case where the data is changed, there's an error,
5055 * and it's not logged so we don't shutdown when we should.
5058 xfs_trans_log_inode(tp, ip, bma.logflags);
5062 ASSERT(*firstblock == NULLFSBLOCK ||
5063 XFS_FSB_TO_AGNO(mp, *firstblock) ==
5065 bma.cur->bc_private.b.firstblock) ||
5067 XFS_FSB_TO_AGNO(mp, *firstblock) <
5069 bma.cur->bc_private.b.firstblock)));
5070 *firstblock = bma.cur->bc_private.b.firstblock;
5072 xfs_btree_del_cursor(bma.cur,
5073 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5076 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
5082 * Unmap (remove) blocks from a file.
5083 * If nexts is nonzero then the number of extents to remove is limited to
5084 * that value. If not all extents in the block range can be removed then
5089 xfs_trans_t *tp, /* transaction pointer */
5090 struct xfs_inode *ip, /* incore inode */
5091 xfs_fileoff_t bno, /* starting offset to unmap */
5092 xfs_filblks_t len, /* length to unmap in file */
5093 int flags, /* misc flags */
5094 xfs_extnum_t nexts, /* number of extents max */
5095 xfs_fsblock_t *firstblock, /* first allocated block
5096 controls a.g. for allocs */
5097 xfs_bmap_free_t *flist, /* i/o: list extents to free */
5098 int *done) /* set if not done yet */
5100 xfs_btree_cur_t *cur; /* bmap btree cursor */
5101 xfs_bmbt_irec_t del; /* extent being deleted */
5102 int eof; /* is deleting at eof */
5103 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
5104 int error; /* error return value */
5105 xfs_extnum_t extno; /* extent number in list */
5106 xfs_bmbt_irec_t got; /* current extent record */
5107 xfs_ifork_t *ifp; /* inode fork pointer */
5108 int isrt; /* freeing in rt area */
5109 xfs_extnum_t lastx; /* last extent index used */
5110 int logflags; /* transaction logging flags */
5111 xfs_extlen_t mod; /* rt extent offset */
5112 xfs_mount_t *mp; /* mount structure */
5113 xfs_extnum_t nextents; /* number of file extents */
5114 xfs_bmbt_irec_t prev; /* previous extent record */
5115 xfs_fileoff_t start; /* first file offset deleted */
5116 int tmp_logflags; /* partial logging flags */
5117 int wasdel; /* was a delayed alloc extent */
5118 int whichfork; /* data or attribute fork */
5121 trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5123 whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5124 XFS_ATTR_FORK : XFS_DATA_FORK;
5125 ifp = XFS_IFORK_PTR(ip, whichfork);
5127 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5128 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5129 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5131 return XFS_ERROR(EFSCORRUPTED);
5134 if (XFS_FORCED_SHUTDOWN(mp))
5135 return XFS_ERROR(EIO);
5140 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5141 (error = xfs_iread_extents(tp, ip, whichfork)))
5143 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5144 if (nextents == 0) {
5148 XFS_STATS_INC(xs_blk_unmap);
5149 isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5151 bno = start + len - 1;
5152 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5156 * Check to see if the given block number is past the end of the
5157 * file, back up to the last block if so...
5160 ep = xfs_iext_get_ext(ifp, --lastx);
5161 xfs_bmbt_get_all(ep, &got);
5162 bno = got.br_startoff + got.br_blockcount - 1;
5165 if (ifp->if_flags & XFS_IFBROOT) {
5166 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5167 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5168 cur->bc_private.b.firstblock = *firstblock;
5169 cur->bc_private.b.flist = flist;
5170 cur->bc_private.b.flags = 0;
5176 * Synchronize by locking the bitmap inode.
5178 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
5179 xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5183 while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5184 (nexts == 0 || extno < nexts)) {
5186 * Is the found extent after a hole in which bno lives?
5187 * Just back up to the previous extent, if so.
5189 if (got.br_startoff > bno) {
5192 ep = xfs_iext_get_ext(ifp, lastx);
5193 xfs_bmbt_get_all(ep, &got);
5196 * Is the last block of this extent before the range
5197 * we're supposed to delete? If so, we're done.
5199 bno = XFS_FILEOFF_MIN(bno,
5200 got.br_startoff + got.br_blockcount - 1);
5204 * Then deal with the (possibly delayed) allocated space
5209 wasdel = isnullstartblock(del.br_startblock);
5210 if (got.br_startoff < start) {
5211 del.br_startoff = start;
5212 del.br_blockcount -= start - got.br_startoff;
5214 del.br_startblock += start - got.br_startoff;
5216 if (del.br_startoff + del.br_blockcount > bno + 1)
5217 del.br_blockcount = bno + 1 - del.br_startoff;
5218 sum = del.br_startblock + del.br_blockcount;
5220 (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5222 * Realtime extent not lined up at the end.
5223 * The extent could have been split into written
5224 * and unwritten pieces, or we could just be
5225 * unmapping part of it. But we can't really
5226 * get rid of part of a realtime extent.
5228 if (del.br_state == XFS_EXT_UNWRITTEN ||
5229 !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5231 * This piece is unwritten, or we're not
5232 * using unwritten extents. Skip over it.
5235 bno -= mod > del.br_blockcount ?
5236 del.br_blockcount : mod;
5237 if (bno < got.br_startoff) {
5239 xfs_bmbt_get_all(xfs_iext_get_ext(
5245 * It's written, turn it unwritten.
5246 * This is better than zeroing it.
5248 ASSERT(del.br_state == XFS_EXT_NORM);
5249 ASSERT(xfs_trans_get_block_res(tp) > 0);
5251 * If this spans a realtime extent boundary,
5252 * chop it back to the start of the one we end at.
5254 if (del.br_blockcount > mod) {
5255 del.br_startoff += del.br_blockcount - mod;
5256 del.br_startblock += del.br_blockcount - mod;
5257 del.br_blockcount = mod;
5259 del.br_state = XFS_EXT_UNWRITTEN;
5260 error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5261 &lastx, &cur, &del, firstblock, flist,
5267 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5269 * Realtime extent is lined up at the end but not
5270 * at the front. We'll get rid of full extents if
5273 mod = mp->m_sb.sb_rextsize - mod;
5274 if (del.br_blockcount > mod) {
5275 del.br_blockcount -= mod;
5276 del.br_startoff += mod;
5277 del.br_startblock += mod;
5278 } else if ((del.br_startoff == start &&
5279 (del.br_state == XFS_EXT_UNWRITTEN ||
5280 xfs_trans_get_block_res(tp) == 0)) ||
5281 !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5283 * Can't make it unwritten. There isn't
5284 * a full extent here so just skip it.
5286 ASSERT(bno >= del.br_blockcount);
5287 bno -= del.br_blockcount;
5288 if (got.br_startoff > bno) {
5290 ep = xfs_iext_get_ext(ifp,
5292 xfs_bmbt_get_all(ep, &got);
5296 } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5298 * This one is already unwritten.
5299 * It must have a written left neighbor.
5300 * Unwrite the killed part of that one and
5304 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5306 ASSERT(prev.br_state == XFS_EXT_NORM);
5307 ASSERT(!isnullstartblock(prev.br_startblock));
5308 ASSERT(del.br_startblock ==
5309 prev.br_startblock + prev.br_blockcount);
5310 if (prev.br_startoff < start) {
5311 mod = start - prev.br_startoff;
5312 prev.br_blockcount -= mod;
5313 prev.br_startblock += mod;
5314 prev.br_startoff = start;
5316 prev.br_state = XFS_EXT_UNWRITTEN;
5318 error = xfs_bmap_add_extent_unwritten_real(tp,
5319 ip, &lastx, &cur, &prev,
5320 firstblock, flist, &logflags);
5325 ASSERT(del.br_state == XFS_EXT_NORM);
5326 del.br_state = XFS_EXT_UNWRITTEN;
5327 error = xfs_bmap_add_extent_unwritten_real(tp,
5328 ip, &lastx, &cur, &del,
5329 firstblock, flist, &logflags);
5336 ASSERT(startblockval(del.br_startblock) > 0);
5337 /* Update realtime/data freespace, unreserve quota */
5339 xfs_filblks_t rtexts;
5341 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5342 do_div(rtexts, mp->m_sb.sb_rextsize);
5343 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5344 (int64_t)rtexts, 0);
5345 (void)xfs_trans_reserve_quota_nblks(NULL,
5346 ip, -((long)del.br_blockcount), 0,
5347 XFS_QMOPT_RES_RTBLKS);
5349 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5350 (int64_t)del.br_blockcount, 0);
5351 (void)xfs_trans_reserve_quota_nblks(NULL,
5352 ip, -((long)del.br_blockcount), 0,
5353 XFS_QMOPT_RES_REGBLKS);
5355 ip->i_delayed_blks -= del.br_blockcount;
5357 cur->bc_private.b.flags |=
5358 XFS_BTCUR_BPRV_WASDEL;
5360 cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5362 * If it's the case where the directory code is running
5363 * with no block reservation, and the deleted block is in
5364 * the middle of its extent, and the resulting insert
5365 * of an extent would cause transformation to btree format,
5366 * then reject it. The calling code will then swap
5367 * blocks around instead.
5368 * We have to do this now, rather than waiting for the
5369 * conversion to btree format, since the transaction
5372 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5373 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5374 XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
5375 XFS_IFORK_MAXEXT(ip, whichfork) &&
5376 del.br_startoff > got.br_startoff &&
5377 del.br_startoff + del.br_blockcount <
5378 got.br_startoff + got.br_blockcount) {
5379 error = XFS_ERROR(ENOSPC);
5382 error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5383 &tmp_logflags, whichfork);
5384 logflags |= tmp_logflags;
5387 bno = del.br_startoff - 1;
5390 * If not done go on to the next (previous) record.
5392 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5394 ep = xfs_iext_get_ext(ifp, lastx);
5395 if (xfs_bmbt_get_startoff(ep) > bno) {
5397 ep = xfs_iext_get_ext(ifp,
5400 xfs_bmbt_get_all(ep, &got);
5405 *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5408 * Convert to a btree if necessary.
5410 if (xfs_bmap_needs_btree(ip, whichfork)) {
5411 ASSERT(cur == NULL);
5412 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5413 &cur, 0, &tmp_logflags, whichfork);
5414 logflags |= tmp_logflags;
5419 * transform from btree to extents, give it cur
5421 else if (xfs_bmap_wants_extents(ip, whichfork)) {
5422 ASSERT(cur != NULL);
5423 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5425 logflags |= tmp_logflags;
5430 * transform from extents to local?
5435 * Log everything. Do this after conversion, there's no point in
5436 * logging the extent records if we've converted to btree format.
5438 if ((logflags & xfs_ilog_fext(whichfork)) &&
5439 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5440 logflags &= ~xfs_ilog_fext(whichfork);
5441 else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5442 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5443 logflags &= ~xfs_ilog_fbroot(whichfork);
5445 * Log inode even in the error case, if the transaction
5446 * is dirty we'll need to shut down the filesystem.
5449 xfs_trans_log_inode(tp, ip, logflags);
5452 *firstblock = cur->bc_private.b.firstblock;
5453 cur->bc_private.b.allocated = 0;
5455 xfs_btree_del_cursor(cur,
5456 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5462 * returns 1 for success, 0 if we failed to map the extent.
5465 xfs_getbmapx_fix_eof_hole(
5466 xfs_inode_t *ip, /* xfs incore inode pointer */
5467 struct getbmapx *out, /* output structure */
5468 int prealloced, /* this is a file with
5469 * preallocated data space */
5470 __int64_t end, /* last block requested */
5471 xfs_fsblock_t startblock)
5474 xfs_mount_t *mp; /* file system mount point */
5475 xfs_ifork_t *ifp; /* inode fork pointer */
5476 xfs_extnum_t lastx; /* last extent pointer */
5477 xfs_fileoff_t fileblock;
5479 if (startblock == HOLESTARTBLOCK) {
5481 out->bmv_block = -1;
5482 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
5483 fixlen -= out->bmv_offset;
5484 if (prealloced && out->bmv_offset + out->bmv_length == end) {
5485 /* Came to hole at EOF. Trim it. */
5488 out->bmv_length = fixlen;
5491 if (startblock == DELAYSTARTBLOCK)
5492 out->bmv_block = -2;
5494 out->bmv_block = xfs_fsb_to_db(ip, startblock);
5495 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
5496 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
5497 if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
5498 (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
5499 out->bmv_oflags |= BMV_OF_LAST;
5506 * Get inode's extents as described in bmv, and format for output.
5507 * Calls formatter to fill the user's buffer until all extents
5508 * are mapped, until the passed-in bmv->bmv_count slots have
5509 * been filled, or until the formatter short-circuits the loop,
5510 * if it is tracking filled-in extents on its own.
5512 int /* error code */
5515 struct getbmapx *bmv, /* user bmap structure */
5516 xfs_bmap_format_t formatter, /* format to user */
5517 void *arg) /* formatter arg */
5519 __int64_t bmvend; /* last block requested */
5520 int error = 0; /* return value */
5521 __int64_t fixlen; /* length for -1 case */
5522 int i; /* extent number */
5523 int lock; /* lock state */
5524 xfs_bmbt_irec_t *map; /* buffer for user's data */
5525 xfs_mount_t *mp; /* file system mount point */
5526 int nex; /* # of user extents can do */
5527 int nexleft; /* # of user extents left */
5528 int subnex; /* # of bmapi's can do */
5529 int nmap; /* number of map entries */
5530 struct getbmapx *out; /* output structure */
5531 int whichfork; /* data or attr fork */
5532 int prealloced; /* this is a file with
5533 * preallocated data space */
5534 int iflags; /* interface flags */
5535 int bmapi_flags; /* flags for xfs_bmapi */
5539 iflags = bmv->bmv_iflags;
5540 whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5542 if (whichfork == XFS_ATTR_FORK) {
5543 if (XFS_IFORK_Q(ip)) {
5544 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5545 ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5546 ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5547 return XFS_ERROR(EINVAL);
5548 } else if (unlikely(
5549 ip->i_d.di_aformat != 0 &&
5550 ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5551 XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5553 return XFS_ERROR(EFSCORRUPTED);
5559 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5560 ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5561 ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5562 return XFS_ERROR(EINVAL);
5564 if (xfs_get_extsz_hint(ip) ||
5565 ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5567 fixlen = mp->m_super->s_maxbytes;
5570 fixlen = XFS_ISIZE(ip);
5574 if (bmv->bmv_length == -1) {
5575 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5577 max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
5578 } else if (bmv->bmv_length == 0) {
5579 bmv->bmv_entries = 0;
5581 } else if (bmv->bmv_length < 0) {
5582 return XFS_ERROR(EINVAL);
5585 nex = bmv->bmv_count - 1;
5587 return XFS_ERROR(EINVAL);
5588 bmvend = bmv->bmv_offset + bmv->bmv_length;
5591 if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
5592 return XFS_ERROR(ENOMEM);
5593 out = kmem_zalloc(bmv->bmv_count * sizeof(struct getbmapx), KM_MAYFAIL);
5595 out = kmem_zalloc_large(bmv->bmv_count *
5596 sizeof(struct getbmapx));
5598 return XFS_ERROR(ENOMEM);
5601 xfs_ilock(ip, XFS_IOLOCK_SHARED);
5602 if (whichfork == XFS_DATA_FORK && !(iflags & BMV_IF_DELALLOC)) {
5603 if (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size) {
5604 error = -filemap_write_and_wait(VFS_I(ip)->i_mapping);
5606 goto out_unlock_iolock;
5609 * even after flushing the inode, there can still be delalloc
5610 * blocks on the inode beyond EOF due to speculative
5611 * preallocation. These are not removed until the release
5612 * function is called or the inode is inactivated. Hence we
5613 * cannot assert here that ip->i_delayed_blks == 0.
5617 lock = xfs_ilock_map_shared(ip);
5620 * Don't let nex be bigger than the number of extents
5621 * we can have assuming alternating holes and real extents.
5623 if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5624 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5626 bmapi_flags = xfs_bmapi_aflag(whichfork);
5627 if (!(iflags & BMV_IF_PREALLOC))
5628 bmapi_flags |= XFS_BMAPI_IGSTATE;
5631 * Allocate enough space to handle "subnex" maps at a time.
5635 map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
5637 goto out_unlock_ilock;
5639 bmv->bmv_entries = 0;
5641 if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
5642 (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
5650 nmap = (nexleft > subnex) ? subnex : nexleft;
5651 error = xfs_bmapi_read(ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5652 XFS_BB_TO_FSB(mp, bmv->bmv_length),
5653 map, &nmap, bmapi_flags);
5656 ASSERT(nmap <= subnex);
5658 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5659 out[cur_ext].bmv_oflags = 0;
5660 if (map[i].br_state == XFS_EXT_UNWRITTEN)
5661 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
5662 else if (map[i].br_startblock == DELAYSTARTBLOCK)
5663 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
5664 out[cur_ext].bmv_offset =
5665 XFS_FSB_TO_BB(mp, map[i].br_startoff);
5666 out[cur_ext].bmv_length =
5667 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5668 out[cur_ext].bmv_unused1 = 0;
5669 out[cur_ext].bmv_unused2 = 0;
5672 * delayed allocation extents that start beyond EOF can
5673 * occur due to speculative EOF allocation when the
5674 * delalloc extent is larger than the largest freespace
5675 * extent at conversion time. These extents cannot be
5676 * converted by data writeback, so can exist here even
5677 * if we are not supposed to be finding delalloc
5680 if (map[i].br_startblock == DELAYSTARTBLOCK &&
5681 map[i].br_startoff <= XFS_B_TO_FSB(mp, XFS_ISIZE(ip)))
5682 ASSERT((iflags & BMV_IF_DELALLOC) != 0);
5684 if (map[i].br_startblock == HOLESTARTBLOCK &&
5685 whichfork == XFS_ATTR_FORK) {
5686 /* came to the end of attribute fork */
5687 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
5691 if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
5693 map[i].br_startblock))
5697 out[cur_ext].bmv_offset +
5698 out[cur_ext].bmv_length;
5700 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
5703 * In case we don't want to return the hole,
5704 * don't increase cur_ext so that we can reuse
5705 * it in the next loop.
5707 if ((iflags & BMV_IF_NO_HOLES) &&
5708 map[i].br_startblock == HOLESTARTBLOCK) {
5709 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
5717 } while (nmap && nexleft && bmv->bmv_length);
5722 xfs_iunlock_map_shared(ip, lock);
5724 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5726 for (i = 0; i < cur_ext; i++) {
5727 int full = 0; /* user array is full */
5729 /* format results & advance arg */
5730 error = formatter(&arg, &out[i], &full);
5735 if (is_vmalloc_addr(out))
5736 kmem_free_large(out);
5743 STATIC struct xfs_buf *
5745 struct xfs_btree_cur *cur,
5748 struct xfs_log_item_desc *lidp;
5754 for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
5755 if (!cur->bc_bufs[i])
5757 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
5758 return cur->bc_bufs[i];
5761 /* Chase down all the log items to see if the bp is there */
5762 list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
5763 struct xfs_buf_log_item *bip;
5764 bip = (struct xfs_buf_log_item *)lidp->lid_item;
5765 if (bip->bli_item.li_type == XFS_LI_BUF &&
5766 XFS_BUF_ADDR(bip->bli_buf) == bno)
5767 return bip->bli_buf;
5775 struct xfs_btree_block *block,
5781 __be64 *pp, *thispa; /* pointer to block address */
5782 xfs_bmbt_key_t *prevp, *keyp;
5784 ASSERT(be16_to_cpu(block->bb_level) > 0);
5787 for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
5788 dmxr = mp->m_bmap_dmxr[0];
5789 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
5792 ASSERT(be64_to_cpu(prevp->br_startoff) <
5793 be64_to_cpu(keyp->br_startoff));
5798 * Compare the block numbers to see if there are dups.
5801 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
5803 pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
5805 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
5807 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
5809 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
5810 if (*thispa == *pp) {
5811 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
5813 (unsigned long long)be64_to_cpu(*thispa));
5814 panic("%s: ptrs are equal in node\n",
5822 * Check that the extents for the inode ip are in the right order in all
5827 xfs_bmap_check_leaf_extents(
5828 xfs_btree_cur_t *cur, /* btree cursor or null */
5829 xfs_inode_t *ip, /* incore inode pointer */
5830 int whichfork) /* data or attr fork */
5832 struct xfs_btree_block *block; /* current btree block */
5833 xfs_fsblock_t bno; /* block # of "block" */
5834 xfs_buf_t *bp; /* buffer for "block" */
5835 int error; /* error return value */
5836 xfs_extnum_t i=0, j; /* index into the extents list */
5837 xfs_ifork_t *ifp; /* fork structure */
5838 int level; /* btree level, for checking */
5839 xfs_mount_t *mp; /* file system mount structure */
5840 __be64 *pp; /* pointer to block address */
5841 xfs_bmbt_rec_t *ep; /* pointer to current extent */
5842 xfs_bmbt_rec_t last = {0, 0}; /* last extent in prev block */
5843 xfs_bmbt_rec_t *nextp; /* pointer to next extent */
5846 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
5852 ifp = XFS_IFORK_PTR(ip, whichfork);
5853 block = ifp->if_broot;
5855 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5857 level = be16_to_cpu(block->bb_level);
5859 xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
5860 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5861 bno = be64_to_cpu(*pp);
5863 ASSERT(bno != NULLDFSBNO);
5864 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5865 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5868 * Go down the tree until leaf level is reached, following the first
5869 * pointer (leftmost) at each level.
5871 while (level-- > 0) {
5872 /* See if buf is in cur first */
5874 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5877 error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5883 block = XFS_BUF_TO_BLOCK(bp);
5884 XFS_WANT_CORRUPTED_GOTO(
5885 xfs_bmap_sanity_check(mp, bp, level),
5891 * Check this block for basic sanity (increasing keys and
5892 * no duplicate blocks).
5895 xfs_check_block(block, mp, 0, 0);
5896 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
5897 bno = be64_to_cpu(*pp);
5898 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
5901 xfs_trans_brelse(NULL, bp);
5906 * Here with bp and block set to the leftmost leaf node in the tree.
5911 * Loop over all leaf nodes checking that all extents are in the right order.
5914 xfs_fsblock_t nextbno;
5915 xfs_extnum_t num_recs;
5918 num_recs = xfs_btree_get_numrecs(block);
5921 * Read-ahead the next leaf block, if any.
5924 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
5927 * Check all the extents to make sure they are OK.
5928 * If we had a previous block, the last entry should
5929 * conform with the first entry in this one.
5932 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
5934 ASSERT(xfs_bmbt_disk_get_startoff(&last) +
5935 xfs_bmbt_disk_get_blockcount(&last) <=
5936 xfs_bmbt_disk_get_startoff(ep));
5938 for (j = 1; j < num_recs; j++) {
5939 nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
5940 ASSERT(xfs_bmbt_disk_get_startoff(ep) +
5941 xfs_bmbt_disk_get_blockcount(ep) <=
5942 xfs_bmbt_disk_get_startoff(nextp));
5950 xfs_trans_brelse(NULL, bp);
5954 * If we've reached the end, stop.
5956 if (bno == NULLFSBLOCK)
5960 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5963 error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5969 block = XFS_BUF_TO_BLOCK(bp);
5973 xfs_trans_brelse(NULL, bp);
5978 xfs_warn(mp, "%s: at error0", __func__);
5980 xfs_trans_brelse(NULL, bp);
5982 xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
5984 panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
5990 * Count fsblocks of the given fork.
5993 xfs_bmap_count_blocks(
5994 xfs_trans_t *tp, /* transaction pointer */
5995 xfs_inode_t *ip, /* incore inode */
5996 int whichfork, /* data or attr fork */
5997 int *count) /* out: count of blocks */
5999 struct xfs_btree_block *block; /* current btree block */
6000 xfs_fsblock_t bno; /* block # of "block" */
6001 xfs_ifork_t *ifp; /* fork structure */
6002 int level; /* btree level, for checking */
6003 xfs_mount_t *mp; /* file system mount structure */
6004 __be64 *pp; /* pointer to block address */
6008 ifp = XFS_IFORK_PTR(ip, whichfork);
6009 if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
6010 xfs_bmap_count_leaves(ifp, 0,
6011 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
6017 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6019 block = ifp->if_broot;
6020 level = be16_to_cpu(block->bb_level);
6022 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
6023 bno = be64_to_cpu(*pp);
6024 ASSERT(bno != NULLDFSBNO);
6025 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
6026 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
6028 if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
6029 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
6031 return XFS_ERROR(EFSCORRUPTED);
6038 * Recursively walks each level of a btree
6039 * to count total fsblocks is use.
6041 STATIC int /* error */
6042 xfs_bmap_count_tree(
6043 xfs_mount_t *mp, /* file system mount point */
6044 xfs_trans_t *tp, /* transaction pointer */
6045 xfs_ifork_t *ifp, /* inode fork pointer */
6046 xfs_fsblock_t blockno, /* file system block number */
6047 int levelin, /* level in btree */
6048 int *count) /* Count of blocks */
6051 xfs_buf_t *bp, *nbp;
6052 int level = levelin;
6054 xfs_fsblock_t bno = blockno;
6055 xfs_fsblock_t nextbno;
6056 struct xfs_btree_block *block, *nextblock;
6059 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF,
6064 block = XFS_BUF_TO_BLOCK(bp);
6067 /* Not at node above leaves, count this level of nodes */
6068 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
6069 while (nextbno != NULLFSBLOCK) {
6070 error = xfs_btree_read_bufl(mp, tp, nextbno, 0, &nbp,
6076 nextblock = XFS_BUF_TO_BLOCK(nbp);
6077 nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
6078 xfs_trans_brelse(tp, nbp);
6081 /* Dive to the next level */
6082 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
6083 bno = be64_to_cpu(*pp);
6084 if (unlikely((error =
6085 xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
6086 xfs_trans_brelse(tp, bp);
6087 XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6088 XFS_ERRLEVEL_LOW, mp);
6089 return XFS_ERROR(EFSCORRUPTED);
6091 xfs_trans_brelse(tp, bp);
6093 /* count all level 1 nodes and their leaves */
6095 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
6096 numrecs = be16_to_cpu(block->bb_numrecs);
6097 xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
6098 xfs_trans_brelse(tp, bp);
6099 if (nextbno == NULLFSBLOCK)
6102 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6108 block = XFS_BUF_TO_BLOCK(bp);
6115 * Count leaf blocks given a range of extent records.
6118 xfs_bmap_count_leaves(
6126 for (b = 0; b < numrecs; b++) {
6127 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
6128 *count += xfs_bmbt_get_blockcount(frp);
6133 * Count leaf blocks given a range of extent records originally
6137 xfs_bmap_disk_count_leaves(
6138 struct xfs_mount *mp,
6139 struct xfs_btree_block *block,
6144 xfs_bmbt_rec_t *frp;
6146 for (b = 1; b <= numrecs; b++) {
6147 frp = XFS_BMBT_REC_ADDR(mp, block, b);
6148 *count += xfs_bmbt_disk_get_blockcount(frp);
6153 * dead simple method of punching delalyed allocation blocks from a range in
6154 * the inode. Walks a block at a time so will be slow, but is only executed in
6155 * rare error cases so the overhead is not critical. This will alays punch out
6156 * both the start and end blocks, even if the ranges only partially overlap
6157 * them, so it is up to the caller to ensure that partial blocks are not
6161 xfs_bmap_punch_delalloc_range(
6162 struct xfs_inode *ip,
6163 xfs_fileoff_t start_fsb,
6164 xfs_fileoff_t length)
6166 xfs_fileoff_t remaining = length;
6169 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
6173 xfs_bmbt_irec_t imap;
6175 xfs_fsblock_t firstblock;
6176 xfs_bmap_free_t flist;
6179 * Map the range first and check that it is a delalloc extent
6180 * before trying to unmap the range. Otherwise we will be
6181 * trying to remove a real extent (which requires a
6182 * transaction) or a hole, which is probably a bad idea...
6184 error = xfs_bmapi_read(ip, start_fsb, 1, &imap, &nimaps,
6188 /* something screwed, just bail */
6189 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
6190 xfs_alert(ip->i_mount,
6191 "Failed delalloc mapping lookup ino %lld fsb %lld.",
6192 ip->i_ino, start_fsb);
6200 if (imap.br_startblock != DELAYSTARTBLOCK) {
6201 /* been converted, ignore */
6204 WARN_ON(imap.br_blockcount == 0);
6207 * Note: while we initialise the firstblock/flist pair, they
6208 * should never be used because blocks should never be
6209 * allocated or freed for a delalloc extent and hence we need
6210 * don't cancel or finish them after the xfs_bunmapi() call.
6212 xfs_bmap_init(&flist, &firstblock);
6213 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
6218 ASSERT(!flist.xbf_count && !flist.xbf_first);
6222 } while(remaining > 0);
6228 * Convert the given file system block to a disk block. We have to treat it
6229 * differently based on whether the file is a real time file or not, because the
6233 xfs_fsb_to_db(struct xfs_inode *ip, xfs_fsblock_t fsb)
6235 return (XFS_IS_REALTIME_INODE(ip) ? \
6236 (xfs_daddr_t)XFS_FSB_TO_BB((ip)->i_mount, (fsb)) : \
6237 XFS_FSB_TO_DADDR((ip)->i_mount, (fsb)));