2 * Copyright (c) 2000-2005 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"
29 #include "xfs_da_btree.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_alloc_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_dir_sf.h"
34 #include "xfs_dir2_sf.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
38 #include "xfs_btree.h"
39 #include "xfs_dmapi.h"
40 #include "xfs_mount.h"
41 #include "xfs_ialloc.h"
42 #include "xfs_itable.h"
43 #include "xfs_inode_item.h"
44 #include "xfs_extfree_item.h"
45 #include "xfs_alloc.h"
47 #include "xfs_rtalloc.h"
48 #include "xfs_error.h"
49 #include "xfs_dir_leaf.h"
50 #include "xfs_attr_leaf.h"
52 #include "xfs_quota.h"
53 #include "xfs_trans_space.h"
54 #include "xfs_buf_item.h"
59 xfs_bmap_check_leaf_extents(xfs_btree_cur_t *cur, xfs_inode_t *ip, int whichfork);
62 kmem_zone_t *xfs_bmap_free_item_zone;
65 * Prototypes for internal bmap routines.
70 * Called from xfs_bmap_add_attrfork to handle extents format files.
72 STATIC int /* error */
73 xfs_bmap_add_attrfork_extents(
74 xfs_trans_t *tp, /* transaction pointer */
75 xfs_inode_t *ip, /* incore inode pointer */
76 xfs_fsblock_t *firstblock, /* first block allocated */
77 xfs_bmap_free_t *flist, /* blocks to free at commit */
78 int *flags); /* inode logging flags */
81 * Called from xfs_bmap_add_attrfork to handle local format files.
83 STATIC int /* error */
84 xfs_bmap_add_attrfork_local(
85 xfs_trans_t *tp, /* transaction pointer */
86 xfs_inode_t *ip, /* incore inode pointer */
87 xfs_fsblock_t *firstblock, /* first block allocated */
88 xfs_bmap_free_t *flist, /* blocks to free at commit */
89 int *flags); /* inode logging flags */
92 * Called by xfs_bmapi to update extent list structure and the btree
93 * after allocating space (or doing a delayed allocation).
95 STATIC int /* error */
97 xfs_inode_t *ip, /* incore inode pointer */
98 xfs_extnum_t idx, /* extent number to update/insert */
99 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
100 xfs_bmbt_irec_t *new, /* new data to put in extent list */
101 xfs_fsblock_t *first, /* pointer to firstblock variable */
102 xfs_bmap_free_t *flist, /* list of extents to be freed */
103 int *logflagsp, /* inode logging flags */
104 int whichfork, /* data or attr fork */
105 int rsvd); /* OK to allocate reserved blocks */
108 * Called by xfs_bmap_add_extent to handle cases converting a delayed
109 * allocation to a real allocation.
111 STATIC int /* error */
112 xfs_bmap_add_extent_delay_real(
113 xfs_inode_t *ip, /* incore inode pointer */
114 xfs_extnum_t idx, /* extent number to update/insert */
115 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
116 xfs_bmbt_irec_t *new, /* new data to put in extent list */
117 xfs_filblks_t *dnew, /* new delayed-alloc indirect blocks */
118 xfs_fsblock_t *first, /* pointer to firstblock variable */
119 xfs_bmap_free_t *flist, /* list of extents to be freed */
120 int *logflagsp, /* inode logging flags */
121 int rsvd); /* OK to allocate reserved blocks */
124 * Called by xfs_bmap_add_extent to handle cases converting a hole
125 * to a delayed allocation.
127 STATIC int /* error */
128 xfs_bmap_add_extent_hole_delay(
129 xfs_inode_t *ip, /* incore inode pointer */
130 xfs_extnum_t idx, /* extent number to update/insert */
131 xfs_btree_cur_t *cur, /* if null, not a btree */
132 xfs_bmbt_irec_t *new, /* new data to put in extent list */
133 int *logflagsp,/* inode logging flags */
134 int rsvd); /* OK to allocate reserved blocks */
137 * Called by xfs_bmap_add_extent to handle cases converting a hole
138 * to a real allocation.
140 STATIC int /* error */
141 xfs_bmap_add_extent_hole_real(
142 xfs_inode_t *ip, /* incore inode pointer */
143 xfs_extnum_t idx, /* extent number to update/insert */
144 xfs_btree_cur_t *cur, /* if null, not a btree */
145 xfs_bmbt_irec_t *new, /* new data to put in extent list */
146 int *logflagsp, /* inode logging flags */
147 int whichfork); /* data or attr fork */
150 * Called by xfs_bmap_add_extent to handle cases converting an unwritten
151 * allocation to a real allocation or vice versa.
153 STATIC int /* error */
154 xfs_bmap_add_extent_unwritten_real(
155 xfs_inode_t *ip, /* incore inode pointer */
156 xfs_extnum_t idx, /* extent number to update/insert */
157 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
158 xfs_bmbt_irec_t *new, /* new data to put in extent list */
159 int *logflagsp); /* inode logging flags */
162 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
163 * It figures out where to ask the underlying allocator to put the new extent.
165 STATIC int /* error */
167 xfs_bmalloca_t *ap); /* bmap alloc argument struct */
170 * Transform a btree format file with only one leaf node, where the
171 * extents list will fit in the inode, into an extents format file.
172 * Since the extent list is already in-core, all we have to do is
173 * give up the space for the btree root and pitch the leaf block.
175 STATIC int /* error */
176 xfs_bmap_btree_to_extents(
177 xfs_trans_t *tp, /* transaction pointer */
178 xfs_inode_t *ip, /* incore inode pointer */
179 xfs_btree_cur_t *cur, /* btree cursor */
180 int *logflagsp, /* inode logging flags */
181 int whichfork); /* data or attr fork */
185 * Check that the extents list for the inode ip is in the right order.
188 xfs_bmap_check_extents(
189 xfs_inode_t *ip, /* incore inode pointer */
190 int whichfork); /* data or attr fork */
194 * Called by xfs_bmapi to update extent list structure and the btree
195 * after removing space (or undoing a delayed allocation).
197 STATIC int /* error */
199 xfs_inode_t *ip, /* incore inode pointer */
200 xfs_trans_t *tp, /* current trans pointer */
201 xfs_extnum_t idx, /* extent number to update/insert */
202 xfs_bmap_free_t *flist, /* list of extents to be freed */
203 xfs_btree_cur_t *cur, /* if null, not a btree */
204 xfs_bmbt_irec_t *new, /* new data to put in extent list */
205 int *logflagsp,/* inode logging flags */
206 int whichfork, /* data or attr fork */
207 int rsvd); /* OK to allocate reserved blocks */
210 * Remove the entry "free" from the free item list. Prev points to the
211 * previous entry, unless "free" is the head of the list.
215 xfs_bmap_free_t *flist, /* free item list header */
216 xfs_bmap_free_item_t *prev, /* previous item on list, if any */
217 xfs_bmap_free_item_t *free); /* list item to be freed */
220 * Remove count entries from the extents array for inode "ip", starting
221 * at index "idx". Copies the remaining items down over the deleted ones,
222 * and gives back the excess memory.
225 xfs_bmap_delete_exlist(
226 xfs_inode_t *ip, /* incode inode pointer */
227 xfs_extnum_t idx, /* starting delete index */
228 xfs_extnum_t count, /* count of items to delete */
229 int whichfork); /* data or attr fork */
232 * Convert an extents-format file into a btree-format file.
233 * The new file will have a root block (in the inode) and a single child block.
235 STATIC int /* error */
236 xfs_bmap_extents_to_btree(
237 xfs_trans_t *tp, /* transaction pointer */
238 xfs_inode_t *ip, /* incore inode pointer */
239 xfs_fsblock_t *firstblock, /* first-block-allocated */
240 xfs_bmap_free_t *flist, /* blocks freed in xaction */
241 xfs_btree_cur_t **curp, /* cursor returned to caller */
242 int wasdel, /* converting a delayed alloc */
243 int *logflagsp, /* inode logging flags */
244 int whichfork); /* data or attr fork */
247 * Insert new item(s) in the extent list for inode "ip".
248 * Count new items are inserted at offset idx.
251 xfs_bmap_insert_exlist(
252 xfs_inode_t *ip, /* incore inode pointer */
253 xfs_extnum_t idx, /* starting index of new items */
254 xfs_extnum_t count, /* number of inserted items */
255 xfs_bmbt_irec_t *new, /* items to insert */
256 int whichfork); /* data or attr fork */
259 * Convert a local file to an extents file.
260 * This code is sort of bogus, since the file data needs to get
261 * logged so it won't be lost. The bmap-level manipulations are ok, though.
263 STATIC int /* error */
264 xfs_bmap_local_to_extents(
265 xfs_trans_t *tp, /* transaction pointer */
266 xfs_inode_t *ip, /* incore inode pointer */
267 xfs_fsblock_t *firstblock, /* first block allocated in xaction */
268 xfs_extlen_t total, /* total blocks needed by transaction */
269 int *logflagsp, /* inode logging flags */
270 int whichfork); /* data or attr fork */
273 * Search the extents list for the inode, for the extent containing bno.
274 * If bno lies in a hole, point to the next entry. If bno lies past eof,
275 * *eofp will be set, and *prevp will contain the last entry (null if none).
276 * Else, *lastxp will be set to the index of the found
277 * entry; *gotp will contain the entry.
279 STATIC xfs_bmbt_rec_t * /* pointer to found extent entry */
280 xfs_bmap_search_extents(
281 xfs_inode_t *ip, /* incore inode pointer */
282 xfs_fileoff_t bno, /* block number searched for */
283 int whichfork, /* data or attr fork */
284 int *eofp, /* out: end of file found */
285 xfs_extnum_t *lastxp, /* out: last extent index */
286 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
287 xfs_bmbt_irec_t *prevp); /* out: previous extent entry found */
290 * Check the last inode extent to determine whether this allocation will result
291 * in blocks being allocated at the end of the file. When we allocate new data
292 * blocks at the end of the file which do not start at the previous data block,
293 * we will try to align the new blocks at stripe unit boundaries.
295 STATIC int /* error */
297 xfs_inode_t *ip, /* incore inode pointer */
298 xfs_fileoff_t off, /* file offset in fsblocks */
299 int whichfork, /* data or attribute fork */
300 char *aeof); /* return value */
302 #ifdef XFS_BMAP_TRACE
304 * Add a bmap trace buffer entry. Base routine for the others.
307 xfs_bmap_trace_addentry(
308 int opcode, /* operation */
309 char *fname, /* function name */
310 char *desc, /* operation description */
311 xfs_inode_t *ip, /* incore inode pointer */
312 xfs_extnum_t idx, /* index of entry(ies) */
313 xfs_extnum_t cnt, /* count of entries, 1 or 2 */
314 xfs_bmbt_rec_t *r1, /* first record */
315 xfs_bmbt_rec_t *r2, /* second record or null */
316 int whichfork); /* data or attr fork */
319 * Add bmap trace entry prior to a call to xfs_bmap_delete_exlist.
322 xfs_bmap_trace_delete(
323 char *fname, /* function name */
324 char *desc, /* operation description */
325 xfs_inode_t *ip, /* incore inode pointer */
326 xfs_extnum_t idx, /* index of entry(entries) deleted */
327 xfs_extnum_t cnt, /* count of entries deleted, 1 or 2 */
328 int whichfork); /* data or attr fork */
331 * Add bmap trace entry prior to a call to xfs_bmap_insert_exlist, or
332 * reading in the extents list from the disk (in the btree).
335 xfs_bmap_trace_insert(
336 char *fname, /* function name */
337 char *desc, /* operation description */
338 xfs_inode_t *ip, /* incore inode pointer */
339 xfs_extnum_t idx, /* index of entry(entries) inserted */
340 xfs_extnum_t cnt, /* count of entries inserted, 1 or 2 */
341 xfs_bmbt_irec_t *r1, /* inserted record 1 */
342 xfs_bmbt_irec_t *r2, /* inserted record 2 or null */
343 int whichfork); /* data or attr fork */
346 * Add bmap trace entry after updating an extent list entry in place.
349 xfs_bmap_trace_post_update(
350 char *fname, /* function name */
351 char *desc, /* operation description */
352 xfs_inode_t *ip, /* incore inode pointer */
353 xfs_extnum_t idx, /* index of entry updated */
354 int whichfork); /* data or attr fork */
357 * Add bmap trace entry prior to updating an extent list entry in place.
360 xfs_bmap_trace_pre_update(
361 char *fname, /* function name */
362 char *desc, /* operation description */
363 xfs_inode_t *ip, /* incore inode pointer */
364 xfs_extnum_t idx, /* index of entry to be updated */
365 int whichfork); /* data or attr fork */
368 #define xfs_bmap_trace_delete(f,d,ip,i,c,w)
369 #define xfs_bmap_trace_insert(f,d,ip,i,c,r1,r2,w)
370 #define xfs_bmap_trace_post_update(f,d,ip,i,w)
371 #define xfs_bmap_trace_pre_update(f,d,ip,i,w)
372 #endif /* XFS_BMAP_TRACE */
375 * Compute the worst-case number of indirect blocks that will be used
376 * for ip's delayed extent of length "len".
379 xfs_bmap_worst_indlen(
380 xfs_inode_t *ip, /* incore inode pointer */
381 xfs_filblks_t len); /* delayed extent length */
385 * Perform various validation checks on the values being returned
389 xfs_bmap_validate_ret(
393 xfs_bmbt_irec_t *mval,
397 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
400 #if defined(XFS_RW_TRACE)
409 #define xfs_bunmap_trace(ip, bno, len, flags, ra)
410 #endif /* XFS_RW_TRACE */
416 xfs_fsblock_t blockno,
421 xfs_bmap_count_leaves(
427 xfs_bmap_disk_count_leaves(
433 * Bmap internal routines.
437 * Called from xfs_bmap_add_attrfork to handle btree format files.
439 STATIC int /* error */
440 xfs_bmap_add_attrfork_btree(
441 xfs_trans_t *tp, /* transaction pointer */
442 xfs_inode_t *ip, /* incore inode pointer */
443 xfs_fsblock_t *firstblock, /* first block allocated */
444 xfs_bmap_free_t *flist, /* blocks to free at commit */
445 int *flags) /* inode logging flags */
447 xfs_btree_cur_t *cur; /* btree cursor */
448 int error; /* error return value */
449 xfs_mount_t *mp; /* file system mount struct */
450 int stat; /* newroot status */
453 if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
454 *flags |= XFS_ILOG_DBROOT;
456 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
458 cur->bc_private.b.flist = flist;
459 cur->bc_private.b.firstblock = *firstblock;
460 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
462 ASSERT(stat == 1); /* must be at least one entry */
463 if ((error = xfs_bmbt_newroot(cur, flags, &stat)))
466 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
467 return XFS_ERROR(ENOSPC);
469 *firstblock = cur->bc_private.b.firstblock;
470 cur->bc_private.b.allocated = 0;
471 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
475 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
480 * Called from xfs_bmap_add_attrfork to handle extents format files.
482 STATIC int /* error */
483 xfs_bmap_add_attrfork_extents(
484 xfs_trans_t *tp, /* transaction pointer */
485 xfs_inode_t *ip, /* incore inode pointer */
486 xfs_fsblock_t *firstblock, /* first block allocated */
487 xfs_bmap_free_t *flist, /* blocks to free at commit */
488 int *flags) /* inode logging flags */
490 xfs_btree_cur_t *cur; /* bmap btree cursor */
491 int error; /* error return value */
493 if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
496 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
497 flags, XFS_DATA_FORK);
499 cur->bc_private.b.allocated = 0;
500 xfs_btree_del_cursor(cur,
501 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
507 * Called from xfs_bmap_add_attrfork to handle local format files.
509 STATIC int /* error */
510 xfs_bmap_add_attrfork_local(
511 xfs_trans_t *tp, /* transaction pointer */
512 xfs_inode_t *ip, /* incore inode pointer */
513 xfs_fsblock_t *firstblock, /* first block allocated */
514 xfs_bmap_free_t *flist, /* blocks to free at commit */
515 int *flags) /* inode logging flags */
517 xfs_da_args_t dargs; /* args for dir/attr code */
518 int error; /* error return value */
519 xfs_mount_t *mp; /* mount structure pointer */
521 if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
523 if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
525 memset(&dargs, 0, sizeof(dargs));
527 dargs.firstblock = firstblock;
529 dargs.total = mp->m_dirblkfsbs;
530 dargs.whichfork = XFS_DATA_FORK;
532 error = XFS_DIR_SHORTFORM_TO_SINGLE(mp, &dargs);
534 error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
540 * Called by xfs_bmapi to update extent list structure and the btree
541 * after allocating space (or doing a delayed allocation).
543 STATIC int /* error */
545 xfs_inode_t *ip, /* incore inode pointer */
546 xfs_extnum_t idx, /* extent number to update/insert */
547 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
548 xfs_bmbt_irec_t *new, /* new data to put in extent list */
549 xfs_fsblock_t *first, /* pointer to firstblock variable */
550 xfs_bmap_free_t *flist, /* list of extents to be freed */
551 int *logflagsp, /* inode logging flags */
552 int whichfork, /* data or attr fork */
553 int rsvd) /* OK to use reserved data blocks */
555 xfs_btree_cur_t *cur; /* btree cursor or null */
556 xfs_filblks_t da_new; /* new count del alloc blocks used */
557 xfs_filblks_t da_old; /* old count del alloc blocks used */
558 int error; /* error return value */
559 #ifdef XFS_BMAP_TRACE
560 static char fname[] = "xfs_bmap_add_extent";
562 xfs_ifork_t *ifp; /* inode fork ptr */
563 int logflags; /* returned value */
564 xfs_extnum_t nextents; /* number of extents in file now */
566 XFS_STATS_INC(xs_add_exlist);
568 ifp = XFS_IFORK_PTR(ip, whichfork);
569 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
570 ASSERT(idx <= nextents);
574 * This is the first extent added to a new/empty file.
575 * Special case this one, so other routines get to assume there are
576 * already extents in the list.
579 xfs_bmap_trace_insert(fname, "insert empty", ip, 0, 1, new,
581 xfs_bmap_insert_exlist(ip, 0, 1, new, whichfork);
584 if (!ISNULLSTARTBLOCK(new->br_startblock)) {
585 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
586 logflags = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
591 * Any kind of new delayed allocation goes here.
593 else if (ISNULLSTARTBLOCK(new->br_startblock)) {
595 ASSERT((cur->bc_private.b.flags &
596 XFS_BTCUR_BPRV_WASDEL) == 0);
597 if ((error = xfs_bmap_add_extent_hole_delay(ip, idx, cur, new,
602 * Real allocation off the end of the file.
604 else if (idx == nextents) {
606 ASSERT((cur->bc_private.b.flags &
607 XFS_BTCUR_BPRV_WASDEL) == 0);
608 if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur, new,
609 &logflags, whichfork)))
612 xfs_bmbt_irec_t prev; /* old extent at offset idx */
615 * Get the record referred to by idx.
617 xfs_bmbt_get_all(&ifp->if_u1.if_extents[idx], &prev);
619 * If it's a real allocation record, and the new allocation ends
620 * after the start of the referred to record, then we're filling
621 * in a delayed or unwritten allocation with a real one, or
622 * converting real back to unwritten.
624 if (!ISNULLSTARTBLOCK(new->br_startblock) &&
625 new->br_startoff + new->br_blockcount > prev.br_startoff) {
626 if (prev.br_state != XFS_EXT_UNWRITTEN &&
627 ISNULLSTARTBLOCK(prev.br_startblock)) {
628 da_old = STARTBLOCKVAL(prev.br_startblock);
630 ASSERT(cur->bc_private.b.flags &
631 XFS_BTCUR_BPRV_WASDEL);
632 if ((error = xfs_bmap_add_extent_delay_real(ip,
633 idx, &cur, new, &da_new, first, flist,
636 } else if (new->br_state == XFS_EXT_NORM) {
637 ASSERT(new->br_state == XFS_EXT_NORM);
638 if ((error = xfs_bmap_add_extent_unwritten_real(
639 ip, idx, &cur, new, &logflags)))
642 ASSERT(new->br_state == XFS_EXT_UNWRITTEN);
643 if ((error = xfs_bmap_add_extent_unwritten_real(
644 ip, idx, &cur, new, &logflags)))
647 ASSERT(*curp == cur || *curp == NULL);
650 * Otherwise we're filling in a hole with an allocation.
654 ASSERT((cur->bc_private.b.flags &
655 XFS_BTCUR_BPRV_WASDEL) == 0);
656 if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur,
657 new, &logflags, whichfork)))
662 ASSERT(*curp == cur || *curp == NULL);
664 * Convert to a btree if necessary.
666 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
667 XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
668 int tmp_logflags; /* partial log flag return val */
671 error = xfs_bmap_extents_to_btree(ip->i_transp, ip, first,
672 flist, &cur, da_old > 0, &tmp_logflags, whichfork);
673 logflags |= tmp_logflags;
678 * Adjust for changes in reserved delayed indirect blocks.
679 * Nothing to do for disk quotas here.
681 if (da_old || da_new) {
686 nblks += cur->bc_private.b.allocated;
687 ASSERT(nblks <= da_old);
689 xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
690 (int)(da_old - nblks), rsvd);
693 * Clear out the allocated field, done with it now in any case.
696 cur->bc_private.b.allocated = 0;
702 xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
704 *logflagsp = logflags;
709 * Called by xfs_bmap_add_extent to handle cases converting a delayed
710 * allocation to a real allocation.
712 STATIC int /* error */
713 xfs_bmap_add_extent_delay_real(
714 xfs_inode_t *ip, /* incore inode pointer */
715 xfs_extnum_t idx, /* extent number to update/insert */
716 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
717 xfs_bmbt_irec_t *new, /* new data to put in extent list */
718 xfs_filblks_t *dnew, /* new delayed-alloc indirect blocks */
719 xfs_fsblock_t *first, /* pointer to firstblock variable */
720 xfs_bmap_free_t *flist, /* list of extents to be freed */
721 int *logflagsp, /* inode logging flags */
722 int rsvd) /* OK to use reserved data block allocation */
724 xfs_bmbt_rec_t *base; /* base of extent entry list */
725 xfs_btree_cur_t *cur; /* btree cursor */
726 int diff; /* temp value */
727 xfs_bmbt_rec_t *ep; /* extent entry for idx */
728 int error; /* error return value */
729 #ifdef XFS_BMAP_TRACE
730 static char fname[] = "xfs_bmap_add_extent_delay_real";
732 int i; /* temp state */
733 xfs_fileoff_t new_endoff; /* end offset of new entry */
734 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
735 /* left is 0, right is 1, prev is 2 */
736 int rval=0; /* return value (logging flags) */
737 int state = 0;/* state bits, accessed thru macros */
738 xfs_filblks_t temp; /* value for dnew calculations */
739 xfs_filblks_t temp2; /* value for dnew calculations */
740 int tmp_rval; /* partial logging flags */
741 enum { /* bit number definitions for state */
742 LEFT_CONTIG, RIGHT_CONTIG,
743 LEFT_FILLING, RIGHT_FILLING,
744 LEFT_DELAY, RIGHT_DELAY,
745 LEFT_VALID, RIGHT_VALID
751 #define MASK(b) (1 << (b))
752 #define MASK2(a,b) (MASK(a) | MASK(b))
753 #define MASK3(a,b,c) (MASK2(a,b) | MASK(c))
754 #define MASK4(a,b,c,d) (MASK3(a,b,c) | MASK(d))
755 #define STATE_SET(b,v) ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
756 #define STATE_TEST(b) (state & MASK(b))
757 #define STATE_SET_TEST(b,v) ((v) ? ((state |= MASK(b)), 1) : \
758 ((state &= ~MASK(b)), 0))
759 #define SWITCH_STATE \
760 (state & MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG))
763 * Set up a bunch of variables to make the tests simpler.
766 base = ip->i_df.if_u1.if_extents;
768 xfs_bmbt_get_all(ep, &PREV);
769 new_endoff = new->br_startoff + new->br_blockcount;
770 ASSERT(PREV.br_startoff <= new->br_startoff);
771 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
773 * Set flags determining what part of the previous delayed allocation
774 * extent is being replaced by a real allocation.
776 STATE_SET(LEFT_FILLING, PREV.br_startoff == new->br_startoff);
777 STATE_SET(RIGHT_FILLING,
778 PREV.br_startoff + PREV.br_blockcount == new_endoff);
780 * Check and set flags if this segment has a left neighbor.
781 * Don't set contiguous if the combined extent would be too large.
783 if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
784 xfs_bmbt_get_all(ep - 1, &LEFT);
785 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(LEFT.br_startblock));
787 STATE_SET(LEFT_CONTIG,
788 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
789 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
790 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
791 LEFT.br_state == new->br_state &&
792 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN);
794 * Check and set flags if this segment has a right neighbor.
795 * Don't set contiguous if the combined extent would be too large.
796 * Also check for all-three-contiguous being too large.
798 if (STATE_SET_TEST(RIGHT_VALID,
800 ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1)) {
801 xfs_bmbt_get_all(ep + 1, &RIGHT);
802 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(RIGHT.br_startblock));
804 STATE_SET(RIGHT_CONTIG,
805 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
806 new_endoff == RIGHT.br_startoff &&
807 new->br_startblock + new->br_blockcount ==
808 RIGHT.br_startblock &&
809 new->br_state == RIGHT.br_state &&
810 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
811 ((state & MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING)) !=
812 MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING) ||
813 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
817 * Switch out based on the FILLING and CONTIG state bits.
819 switch (SWITCH_STATE) {
821 case MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
823 * Filling in all of a previously delayed allocation extent.
824 * The left and right neighbors are both contiguous with new.
826 xfs_bmap_trace_pre_update(fname, "LF|RF|LC|RC", ip, idx - 1,
828 xfs_bmbt_set_blockcount(ep - 1,
829 LEFT.br_blockcount + PREV.br_blockcount +
830 RIGHT.br_blockcount);
831 xfs_bmap_trace_post_update(fname, "LF|RF|LC|RC", ip, idx - 1,
833 xfs_bmap_trace_delete(fname, "LF|RF|LC|RC", ip, idx, 2,
835 xfs_bmap_delete_exlist(ip, idx, 2, XFS_DATA_FORK);
836 ip->i_df.if_lastex = idx - 1;
837 ip->i_d.di_nextents--;
839 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
841 rval = XFS_ILOG_CORE;
842 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
844 RIGHT.br_blockcount, &i)))
847 if ((error = xfs_bmbt_delete(cur, &i)))
850 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
853 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
857 RIGHT.br_blockcount, LEFT.br_state)))
863 case MASK3(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG):
865 * Filling in all of a previously delayed allocation extent.
866 * The left neighbor is contiguous, the right is not.
868 xfs_bmap_trace_pre_update(fname, "LF|RF|LC", ip, idx - 1,
870 xfs_bmbt_set_blockcount(ep - 1,
871 LEFT.br_blockcount + PREV.br_blockcount);
872 xfs_bmap_trace_post_update(fname, "LF|RF|LC", ip, idx - 1,
874 ip->i_df.if_lastex = idx - 1;
875 xfs_bmap_trace_delete(fname, "LF|RF|LC", ip, idx, 1,
877 xfs_bmap_delete_exlist(ip, idx, 1, XFS_DATA_FORK);
879 rval = XFS_ILOG_DEXT;
882 if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
883 LEFT.br_startblock, LEFT.br_blockcount,
887 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
890 PREV.br_blockcount, LEFT.br_state)))
896 case MASK3(LEFT_FILLING, RIGHT_FILLING, RIGHT_CONTIG):
898 * Filling in all of a previously delayed allocation extent.
899 * The right neighbor is contiguous, the left is not.
901 xfs_bmap_trace_pre_update(fname, "LF|RF|RC", ip, idx,
903 xfs_bmbt_set_startblock(ep, new->br_startblock);
904 xfs_bmbt_set_blockcount(ep,
905 PREV.br_blockcount + RIGHT.br_blockcount);
906 xfs_bmap_trace_post_update(fname, "LF|RF|RC", ip, idx,
908 ip->i_df.if_lastex = idx;
909 xfs_bmap_trace_delete(fname, "LF|RF|RC", ip, idx + 1, 1,
911 xfs_bmap_delete_exlist(ip, idx + 1, 1, XFS_DATA_FORK);
913 rval = XFS_ILOG_DEXT;
916 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
918 RIGHT.br_blockcount, &i)))
921 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
924 RIGHT.br_blockcount, PREV.br_state)))
930 case MASK2(LEFT_FILLING, RIGHT_FILLING):
932 * Filling in all of a previously delayed allocation extent.
933 * Neither the left nor right neighbors are contiguous with
936 xfs_bmap_trace_pre_update(fname, "LF|RF", ip, idx,
938 xfs_bmbt_set_startblock(ep, new->br_startblock);
939 xfs_bmap_trace_post_update(fname, "LF|RF", ip, idx,
941 ip->i_df.if_lastex = idx;
942 ip->i_d.di_nextents++;
944 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
946 rval = XFS_ILOG_CORE;
947 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
948 new->br_startblock, new->br_blockcount,
952 cur->bc_rec.b.br_state = XFS_EXT_NORM;
953 if ((error = xfs_bmbt_insert(cur, &i)))
960 case MASK2(LEFT_FILLING, LEFT_CONTIG):
962 * Filling in the first part of a previous delayed allocation.
963 * The left neighbor is contiguous.
965 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx - 1,
967 xfs_bmbt_set_blockcount(ep - 1,
968 LEFT.br_blockcount + new->br_blockcount);
969 xfs_bmbt_set_startoff(ep,
970 PREV.br_startoff + new->br_blockcount);
971 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx - 1,
973 temp = PREV.br_blockcount - new->br_blockcount;
974 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx,
976 xfs_bmbt_set_blockcount(ep, temp);
977 ip->i_df.if_lastex = idx - 1;
979 rval = XFS_ILOG_DEXT;
982 if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
983 LEFT.br_startblock, LEFT.br_blockcount,
987 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
994 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
995 STARTBLOCKVAL(PREV.br_startblock));
996 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
997 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx,
1002 case MASK(LEFT_FILLING):
1004 * Filling in the first part of a previous delayed allocation.
1005 * The left neighbor is not contiguous.
1007 xfs_bmap_trace_pre_update(fname, "LF", ip, idx, XFS_DATA_FORK);
1008 xfs_bmbt_set_startoff(ep, new_endoff);
1009 temp = PREV.br_blockcount - new->br_blockcount;
1010 xfs_bmbt_set_blockcount(ep, temp);
1011 xfs_bmap_trace_insert(fname, "LF", ip, idx, 1, new, NULL,
1013 xfs_bmap_insert_exlist(ip, idx, 1, new, XFS_DATA_FORK);
1014 ip->i_df.if_lastex = idx;
1015 ip->i_d.di_nextents++;
1017 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1019 rval = XFS_ILOG_CORE;
1020 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1021 new->br_startblock, new->br_blockcount,
1025 cur->bc_rec.b.br_state = XFS_EXT_NORM;
1026 if ((error = xfs_bmbt_insert(cur, &i)))
1030 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1031 ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1032 error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1033 first, flist, &cur, 1, &tmp_rval,
1039 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1040 STARTBLOCKVAL(PREV.br_startblock) -
1041 (cur ? cur->bc_private.b.allocated : 0));
1042 base = ip->i_df.if_u1.if_extents;
1043 ep = &base[idx + 1];
1044 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1045 xfs_bmap_trace_post_update(fname, "LF", ip, idx + 1,
1050 case MASK2(RIGHT_FILLING, RIGHT_CONTIG):
1052 * Filling in the last part of a previous delayed allocation.
1053 * The right neighbor is contiguous with the new allocation.
1055 temp = PREV.br_blockcount - new->br_blockcount;
1056 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx,
1058 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx + 1,
1060 xfs_bmbt_set_blockcount(ep, temp);
1061 xfs_bmbt_set_allf(ep + 1, new->br_startoff, new->br_startblock,
1062 new->br_blockcount + RIGHT.br_blockcount,
1064 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx + 1,
1066 ip->i_df.if_lastex = idx + 1;
1068 rval = XFS_ILOG_DEXT;
1071 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1072 RIGHT.br_startblock,
1073 RIGHT.br_blockcount, &i)))
1076 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1078 new->br_blockcount +
1079 RIGHT.br_blockcount,
1083 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1084 STARTBLOCKVAL(PREV.br_startblock));
1085 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1086 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx,
1091 case MASK(RIGHT_FILLING):
1093 * Filling in the last part of a previous delayed allocation.
1094 * The right neighbor is not contiguous.
1096 temp = PREV.br_blockcount - new->br_blockcount;
1097 xfs_bmap_trace_pre_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1098 xfs_bmbt_set_blockcount(ep, temp);
1099 xfs_bmap_trace_insert(fname, "RF", ip, idx + 1, 1,
1100 new, NULL, XFS_DATA_FORK);
1101 xfs_bmap_insert_exlist(ip, idx + 1, 1, new, XFS_DATA_FORK);
1102 ip->i_df.if_lastex = idx + 1;
1103 ip->i_d.di_nextents++;
1105 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1107 rval = XFS_ILOG_CORE;
1108 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1109 new->br_startblock, new->br_blockcount,
1113 cur->bc_rec.b.br_state = XFS_EXT_NORM;
1114 if ((error = xfs_bmbt_insert(cur, &i)))
1118 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1119 ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1120 error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1121 first, flist, &cur, 1, &tmp_rval,
1127 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1128 STARTBLOCKVAL(PREV.br_startblock) -
1129 (cur ? cur->bc_private.b.allocated : 0));
1130 base = ip->i_df.if_u1.if_extents;
1132 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1133 xfs_bmap_trace_post_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1139 * Filling in the middle part of a previous delayed allocation.
1140 * Contiguity is impossible here.
1141 * This case is avoided almost all the time.
1143 temp = new->br_startoff - PREV.br_startoff;
1144 xfs_bmap_trace_pre_update(fname, "0", ip, idx, XFS_DATA_FORK);
1145 xfs_bmbt_set_blockcount(ep, temp);
1147 r[1].br_startoff = new_endoff;
1148 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
1149 r[1].br_blockcount = temp2;
1150 xfs_bmap_trace_insert(fname, "0", ip, idx + 1, 2, &r[0], &r[1],
1152 xfs_bmap_insert_exlist(ip, idx + 1, 2, &r[0], XFS_DATA_FORK);
1153 ip->i_df.if_lastex = idx + 1;
1154 ip->i_d.di_nextents++;
1156 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1158 rval = XFS_ILOG_CORE;
1159 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1160 new->br_startblock, new->br_blockcount,
1164 cur->bc_rec.b.br_state = XFS_EXT_NORM;
1165 if ((error = xfs_bmbt_insert(cur, &i)))
1169 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1170 ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1171 error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1172 first, flist, &cur, 1, &tmp_rval,
1178 temp = xfs_bmap_worst_indlen(ip, temp);
1179 temp2 = xfs_bmap_worst_indlen(ip, temp2);
1180 diff = (int)(temp + temp2 - STARTBLOCKVAL(PREV.br_startblock) -
1181 (cur ? cur->bc_private.b.allocated : 0));
1183 xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS, -diff, rsvd)) {
1185 * Ick gross gag me with a spoon.
1187 ASSERT(0); /* want to see if this ever happens! */
1193 !xfs_mod_incore_sb(ip->i_mount,
1194 XFS_SBS_FDBLOCKS, -diff, rsvd))
1201 !xfs_mod_incore_sb(ip->i_mount,
1202 XFS_SBS_FDBLOCKS, -diff, rsvd))
1207 base = ip->i_df.if_u1.if_extents;
1209 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1210 xfs_bmap_trace_post_update(fname, "0", ip, idx, XFS_DATA_FORK);
1211 xfs_bmap_trace_pre_update(fname, "0", ip, idx + 2,
1213 xfs_bmbt_set_startblock(ep + 2, NULLSTARTBLOCK((int)temp2));
1214 xfs_bmap_trace_post_update(fname, "0", ip, idx + 2,
1216 *dnew = temp + temp2;
1219 case MASK3(LEFT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1220 case MASK3(RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1221 case MASK2(LEFT_FILLING, RIGHT_CONTIG):
1222 case MASK2(RIGHT_FILLING, LEFT_CONTIG):
1223 case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1224 case MASK(LEFT_CONTIG):
1225 case MASK(RIGHT_CONTIG):
1227 * These cases are all impossible.
1244 #undef STATE_SET_TEST
1249 * Called by xfs_bmap_add_extent to handle cases converting an unwritten
1250 * allocation to a real allocation or vice versa.
1252 STATIC int /* error */
1253 xfs_bmap_add_extent_unwritten_real(
1254 xfs_inode_t *ip, /* incore inode pointer */
1255 xfs_extnum_t idx, /* extent number to update/insert */
1256 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
1257 xfs_bmbt_irec_t *new, /* new data to put in extent list */
1258 int *logflagsp) /* inode logging flags */
1260 xfs_bmbt_rec_t *base; /* base of extent entry list */
1261 xfs_btree_cur_t *cur; /* btree cursor */
1262 xfs_bmbt_rec_t *ep; /* extent entry for idx */
1263 int error; /* error return value */
1264 #ifdef XFS_BMAP_TRACE
1265 static char fname[] = "xfs_bmap_add_extent_unwritten_real";
1267 int i; /* temp state */
1268 xfs_fileoff_t new_endoff; /* end offset of new entry */
1269 xfs_exntst_t newext; /* new extent state */
1270 xfs_exntst_t oldext; /* old extent state */
1271 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
1272 /* left is 0, right is 1, prev is 2 */
1273 int rval=0; /* return value (logging flags) */
1274 int state = 0;/* state bits, accessed thru macros */
1275 enum { /* bit number definitions for state */
1276 LEFT_CONTIG, RIGHT_CONTIG,
1277 LEFT_FILLING, RIGHT_FILLING,
1278 LEFT_DELAY, RIGHT_DELAY,
1279 LEFT_VALID, RIGHT_VALID
1285 #define MASK(b) (1 << (b))
1286 #define MASK2(a,b) (MASK(a) | MASK(b))
1287 #define MASK3(a,b,c) (MASK2(a,b) | MASK(c))
1288 #define MASK4(a,b,c,d) (MASK3(a,b,c) | MASK(d))
1289 #define STATE_SET(b,v) ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1290 #define STATE_TEST(b) (state & MASK(b))
1291 #define STATE_SET_TEST(b,v) ((v) ? ((state |= MASK(b)), 1) : \
1292 ((state &= ~MASK(b)), 0))
1293 #define SWITCH_STATE \
1294 (state & MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG))
1297 * Set up a bunch of variables to make the tests simpler.
1301 base = ip->i_df.if_u1.if_extents;
1303 xfs_bmbt_get_all(ep, &PREV);
1304 newext = new->br_state;
1305 oldext = (newext == XFS_EXT_UNWRITTEN) ?
1306 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
1307 ASSERT(PREV.br_state == oldext);
1308 new_endoff = new->br_startoff + new->br_blockcount;
1309 ASSERT(PREV.br_startoff <= new->br_startoff);
1310 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1312 * Set flags determining what part of the previous oldext allocation
1313 * extent is being replaced by a newext allocation.
1315 STATE_SET(LEFT_FILLING, PREV.br_startoff == new->br_startoff);
1316 STATE_SET(RIGHT_FILLING,
1317 PREV.br_startoff + PREV.br_blockcount == new_endoff);
1319 * Check and set flags if this segment has a left neighbor.
1320 * Don't set contiguous if the combined extent would be too large.
1322 if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1323 xfs_bmbt_get_all(ep - 1, &LEFT);
1324 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(LEFT.br_startblock));
1326 STATE_SET(LEFT_CONTIG,
1327 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
1328 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1329 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1330 LEFT.br_state == newext &&
1331 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1333 * Check and set flags if this segment has a right neighbor.
1334 * Don't set contiguous if the combined extent would be too large.
1335 * Also check for all-three-contiguous being too large.
1337 if (STATE_SET_TEST(RIGHT_VALID,
1339 ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1)) {
1340 xfs_bmbt_get_all(ep + 1, &RIGHT);
1341 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(RIGHT.br_startblock));
1343 STATE_SET(RIGHT_CONTIG,
1344 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
1345 new_endoff == RIGHT.br_startoff &&
1346 new->br_startblock + new->br_blockcount ==
1347 RIGHT.br_startblock &&
1348 newext == RIGHT.br_state &&
1349 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1350 ((state & MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING)) !=
1351 MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING) ||
1352 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1355 * Switch out based on the FILLING and CONTIG state bits.
1357 switch (SWITCH_STATE) {
1359 case MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1361 * Setting all of a previous oldext extent to newext.
1362 * The left and right neighbors are both contiguous with new.
1364 xfs_bmap_trace_pre_update(fname, "LF|RF|LC|RC", ip, idx - 1,
1366 xfs_bmbt_set_blockcount(ep - 1,
1367 LEFT.br_blockcount + PREV.br_blockcount +
1368 RIGHT.br_blockcount);
1369 xfs_bmap_trace_post_update(fname, "LF|RF|LC|RC", ip, idx - 1,
1371 xfs_bmap_trace_delete(fname, "LF|RF|LC|RC", ip, idx, 2,
1373 xfs_bmap_delete_exlist(ip, idx, 2, XFS_DATA_FORK);
1374 ip->i_df.if_lastex = idx - 1;
1375 ip->i_d.di_nextents -= 2;
1377 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1379 rval = XFS_ILOG_CORE;
1380 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1381 RIGHT.br_startblock,
1382 RIGHT.br_blockcount, &i)))
1385 if ((error = xfs_bmbt_delete(cur, &i)))
1388 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1391 if ((error = xfs_bmbt_delete(cur, &i)))
1394 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1397 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1399 LEFT.br_blockcount + PREV.br_blockcount +
1400 RIGHT.br_blockcount, LEFT.br_state)))
1405 case MASK3(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG):
1407 * Setting all of a previous oldext extent to newext.
1408 * The left neighbor is contiguous, the right is not.
1410 xfs_bmap_trace_pre_update(fname, "LF|RF|LC", ip, idx - 1,
1412 xfs_bmbt_set_blockcount(ep - 1,
1413 LEFT.br_blockcount + PREV.br_blockcount);
1414 xfs_bmap_trace_post_update(fname, "LF|RF|LC", ip, idx - 1,
1416 ip->i_df.if_lastex = idx - 1;
1417 xfs_bmap_trace_delete(fname, "LF|RF|LC", ip, idx, 1,
1419 xfs_bmap_delete_exlist(ip, idx, 1, XFS_DATA_FORK);
1420 ip->i_d.di_nextents--;
1422 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1424 rval = XFS_ILOG_CORE;
1425 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1426 PREV.br_startblock, PREV.br_blockcount,
1430 if ((error = xfs_bmbt_delete(cur, &i)))
1433 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1436 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1438 LEFT.br_blockcount + PREV.br_blockcount,
1444 case MASK3(LEFT_FILLING, RIGHT_FILLING, RIGHT_CONTIG):
1446 * Setting all of a previous oldext extent to newext.
1447 * The right neighbor is contiguous, the left is not.
1449 xfs_bmap_trace_pre_update(fname, "LF|RF|RC", ip, idx,
1451 xfs_bmbt_set_blockcount(ep,
1452 PREV.br_blockcount + RIGHT.br_blockcount);
1453 xfs_bmbt_set_state(ep, newext);
1454 xfs_bmap_trace_post_update(fname, "LF|RF|RC", ip, idx,
1456 ip->i_df.if_lastex = idx;
1457 xfs_bmap_trace_delete(fname, "LF|RF|RC", ip, idx + 1, 1,
1459 xfs_bmap_delete_exlist(ip, idx + 1, 1, XFS_DATA_FORK);
1460 ip->i_d.di_nextents--;
1462 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1464 rval = XFS_ILOG_CORE;
1465 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1466 RIGHT.br_startblock,
1467 RIGHT.br_blockcount, &i)))
1470 if ((error = xfs_bmbt_delete(cur, &i)))
1473 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1476 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1478 new->br_blockcount + RIGHT.br_blockcount,
1484 case MASK2(LEFT_FILLING, RIGHT_FILLING):
1486 * Setting all of a previous oldext extent to newext.
1487 * Neither the left nor right neighbors are contiguous with
1490 xfs_bmap_trace_pre_update(fname, "LF|RF", ip, idx,
1492 xfs_bmbt_set_state(ep, newext);
1493 xfs_bmap_trace_post_update(fname, "LF|RF", ip, idx,
1495 ip->i_df.if_lastex = idx;
1497 rval = XFS_ILOG_DEXT;
1500 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1501 new->br_startblock, new->br_blockcount,
1505 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1506 new->br_startblock, new->br_blockcount,
1512 case MASK2(LEFT_FILLING, LEFT_CONTIG):
1514 * Setting the first part of a previous oldext extent to newext.
1515 * The left neighbor is contiguous.
1517 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx - 1,
1519 xfs_bmbt_set_blockcount(ep - 1,
1520 LEFT.br_blockcount + new->br_blockcount);
1521 xfs_bmbt_set_startoff(ep,
1522 PREV.br_startoff + new->br_blockcount);
1523 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx - 1,
1525 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx,
1527 xfs_bmbt_set_startblock(ep,
1528 new->br_startblock + new->br_blockcount);
1529 xfs_bmbt_set_blockcount(ep,
1530 PREV.br_blockcount - new->br_blockcount);
1531 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx,
1533 ip->i_df.if_lastex = idx - 1;
1535 rval = XFS_ILOG_DEXT;
1538 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1539 PREV.br_startblock, PREV.br_blockcount,
1543 if ((error = xfs_bmbt_update(cur,
1544 PREV.br_startoff + new->br_blockcount,
1545 PREV.br_startblock + new->br_blockcount,
1546 PREV.br_blockcount - new->br_blockcount,
1549 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1551 if (xfs_bmbt_update(cur, LEFT.br_startoff,
1553 LEFT.br_blockcount + new->br_blockcount,
1559 case MASK(LEFT_FILLING):
1561 * Setting the first part of a previous oldext extent to newext.
1562 * The left neighbor is not contiguous.
1564 xfs_bmap_trace_pre_update(fname, "LF", ip, idx, XFS_DATA_FORK);
1565 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1566 xfs_bmbt_set_startoff(ep, new_endoff);
1567 xfs_bmbt_set_blockcount(ep,
1568 PREV.br_blockcount - new->br_blockcount);
1569 xfs_bmbt_set_startblock(ep,
1570 new->br_startblock + new->br_blockcount);
1571 xfs_bmap_trace_post_update(fname, "LF", ip, idx, XFS_DATA_FORK);
1572 xfs_bmap_trace_insert(fname, "LF", ip, idx, 1, new, NULL,
1574 xfs_bmap_insert_exlist(ip, idx, 1, new, XFS_DATA_FORK);
1575 ip->i_df.if_lastex = idx;
1576 ip->i_d.di_nextents++;
1578 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1580 rval = XFS_ILOG_CORE;
1581 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1582 PREV.br_startblock, PREV.br_blockcount,
1586 if ((error = xfs_bmbt_update(cur,
1587 PREV.br_startoff + new->br_blockcount,
1588 PREV.br_startblock + new->br_blockcount,
1589 PREV.br_blockcount - new->br_blockcount,
1592 cur->bc_rec.b = *new;
1593 if ((error = xfs_bmbt_insert(cur, &i)))
1599 case MASK2(RIGHT_FILLING, RIGHT_CONTIG):
1601 * Setting the last part of a previous oldext extent to newext.
1602 * The right neighbor is contiguous with the new allocation.
1604 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx,
1606 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx + 1,
1608 xfs_bmbt_set_blockcount(ep,
1609 PREV.br_blockcount - new->br_blockcount);
1610 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx,
1612 xfs_bmbt_set_allf(ep + 1, new->br_startoff, new->br_startblock,
1613 new->br_blockcount + RIGHT.br_blockcount, newext);
1614 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx + 1,
1616 ip->i_df.if_lastex = idx + 1;
1618 rval = XFS_ILOG_DEXT;
1621 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1623 PREV.br_blockcount, &i)))
1626 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1628 PREV.br_blockcount - new->br_blockcount,
1631 if ((error = xfs_bmbt_increment(cur, 0, &i)))
1633 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1635 new->br_blockcount + RIGHT.br_blockcount,
1641 case MASK(RIGHT_FILLING):
1643 * Setting the last part of a previous oldext extent to newext.
1644 * The right neighbor is not contiguous.
1646 xfs_bmap_trace_pre_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1647 xfs_bmbt_set_blockcount(ep,
1648 PREV.br_blockcount - new->br_blockcount);
1649 xfs_bmap_trace_post_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1650 xfs_bmap_trace_insert(fname, "RF", ip, idx + 1, 1,
1651 new, NULL, XFS_DATA_FORK);
1652 xfs_bmap_insert_exlist(ip, idx + 1, 1, new, XFS_DATA_FORK);
1653 ip->i_df.if_lastex = idx + 1;
1654 ip->i_d.di_nextents++;
1656 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1658 rval = XFS_ILOG_CORE;
1659 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1660 PREV.br_startblock, PREV.br_blockcount,
1664 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1666 PREV.br_blockcount - new->br_blockcount,
1669 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1670 new->br_startblock, new->br_blockcount,
1674 cur->bc_rec.b.br_state = XFS_EXT_NORM;
1675 if ((error = xfs_bmbt_insert(cur, &i)))
1683 * Setting the middle part of a previous oldext extent to
1684 * newext. Contiguity is impossible here.
1685 * One extent becomes three extents.
1687 xfs_bmap_trace_pre_update(fname, "0", ip, idx, XFS_DATA_FORK);
1688 xfs_bmbt_set_blockcount(ep,
1689 new->br_startoff - PREV.br_startoff);
1690 xfs_bmap_trace_post_update(fname, "0", ip, idx, XFS_DATA_FORK);
1692 r[1].br_startoff = new_endoff;
1693 r[1].br_blockcount =
1694 PREV.br_startoff + PREV.br_blockcount - new_endoff;
1695 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1696 r[1].br_state = oldext;
1697 xfs_bmap_trace_insert(fname, "0", ip, idx + 1, 2, &r[0], &r[1],
1699 xfs_bmap_insert_exlist(ip, idx + 1, 2, &r[0], XFS_DATA_FORK);
1700 ip->i_df.if_lastex = idx + 1;
1701 ip->i_d.di_nextents += 2;
1703 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1705 rval = XFS_ILOG_CORE;
1706 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1707 PREV.br_startblock, PREV.br_blockcount,
1711 /* new right extent - oldext */
1712 if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1713 r[1].br_startblock, r[1].br_blockcount,
1716 /* new left extent - oldext */
1717 PREV.br_blockcount =
1718 new->br_startoff - PREV.br_startoff;
1719 cur->bc_rec.b = PREV;
1720 if ((error = xfs_bmbt_insert(cur, &i)))
1723 if ((error = xfs_bmbt_increment(cur, 0, &i)))
1726 /* new middle extent - newext */
1727 cur->bc_rec.b = *new;
1728 if ((error = xfs_bmbt_insert(cur, &i)))
1734 case MASK3(LEFT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1735 case MASK3(RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1736 case MASK2(LEFT_FILLING, RIGHT_CONTIG):
1737 case MASK2(RIGHT_FILLING, LEFT_CONTIG):
1738 case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1739 case MASK(LEFT_CONTIG):
1740 case MASK(RIGHT_CONTIG):
1742 * These cases are all impossible.
1759 #undef STATE_SET_TEST
1764 * Called by xfs_bmap_add_extent to handle cases converting a hole
1765 * to a delayed allocation.
1768 STATIC int /* error */
1769 xfs_bmap_add_extent_hole_delay(
1770 xfs_inode_t *ip, /* incore inode pointer */
1771 xfs_extnum_t idx, /* extent number to update/insert */
1772 xfs_btree_cur_t *cur, /* if null, not a btree */
1773 xfs_bmbt_irec_t *new, /* new data to put in extent list */
1774 int *logflagsp, /* inode logging flags */
1775 int rsvd) /* OK to allocate reserved blocks */
1777 xfs_bmbt_rec_t *base; /* base of extent entry list */
1778 xfs_bmbt_rec_t *ep; /* extent list entry for idx */
1779 #ifdef XFS_BMAP_TRACE
1780 static char fname[] = "xfs_bmap_add_extent_hole_delay";
1782 xfs_bmbt_irec_t left; /* left neighbor extent entry */
1783 xfs_filblks_t newlen=0; /* new indirect size */
1784 xfs_filblks_t oldlen=0; /* old indirect size */
1785 xfs_bmbt_irec_t right; /* right neighbor extent entry */
1786 int state; /* state bits, accessed thru macros */
1787 xfs_filblks_t temp; /* temp for indirect calculations */
1788 enum { /* bit number definitions for state */
1789 LEFT_CONTIG, RIGHT_CONTIG,
1790 LEFT_DELAY, RIGHT_DELAY,
1791 LEFT_VALID, RIGHT_VALID
1794 #define MASK(b) (1 << (b))
1795 #define MASK2(a,b) (MASK(a) | MASK(b))
1796 #define STATE_SET(b,v) ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1797 #define STATE_TEST(b) (state & MASK(b))
1798 #define STATE_SET_TEST(b,v) ((v) ? ((state |= MASK(b)), 1) : \
1799 ((state &= ~MASK(b)), 0))
1800 #define SWITCH_STATE (state & MASK2(LEFT_CONTIG, RIGHT_CONTIG))
1802 base = ip->i_df.if_u1.if_extents;
1805 ASSERT(ISNULLSTARTBLOCK(new->br_startblock));
1807 * Check and set flags if this segment has a left neighbor
1809 if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1810 xfs_bmbt_get_all(ep - 1, &left);
1811 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(left.br_startblock));
1814 * Check and set flags if the current (right) segment exists.
1815 * If it doesn't exist, we're converting the hole at end-of-file.
1817 if (STATE_SET_TEST(RIGHT_VALID,
1819 ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1820 xfs_bmbt_get_all(ep, &right);
1821 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(right.br_startblock));
1824 * Set contiguity flags on the left and right neighbors.
1825 * Don't let extents get too large, even if the pieces are contiguous.
1827 STATE_SET(LEFT_CONTIG,
1828 STATE_TEST(LEFT_VALID) && STATE_TEST(LEFT_DELAY) &&
1829 left.br_startoff + left.br_blockcount == new->br_startoff &&
1830 left.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1831 STATE_SET(RIGHT_CONTIG,
1832 STATE_TEST(RIGHT_VALID) && STATE_TEST(RIGHT_DELAY) &&
1833 new->br_startoff + new->br_blockcount == right.br_startoff &&
1834 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1835 (!STATE_TEST(LEFT_CONTIG) ||
1836 (left.br_blockcount + new->br_blockcount +
1837 right.br_blockcount <= MAXEXTLEN)));
1839 * Switch out based on the contiguity flags.
1841 switch (SWITCH_STATE) {
1843 case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1845 * New allocation is contiguous with delayed allocations
1846 * on the left and on the right.
1847 * Merge all three into a single extent list entry.
1849 temp = left.br_blockcount + new->br_blockcount +
1850 right.br_blockcount;
1851 xfs_bmap_trace_pre_update(fname, "LC|RC", ip, idx - 1,
1853 xfs_bmbt_set_blockcount(ep - 1, temp);
1854 oldlen = STARTBLOCKVAL(left.br_startblock) +
1855 STARTBLOCKVAL(new->br_startblock) +
1856 STARTBLOCKVAL(right.br_startblock);
1857 newlen = xfs_bmap_worst_indlen(ip, temp);
1858 xfs_bmbt_set_startblock(ep - 1, NULLSTARTBLOCK((int)newlen));
1859 xfs_bmap_trace_post_update(fname, "LC|RC", ip, idx - 1,
1861 xfs_bmap_trace_delete(fname, "LC|RC", ip, idx, 1,
1863 xfs_bmap_delete_exlist(ip, idx, 1, XFS_DATA_FORK);
1864 ip->i_df.if_lastex = idx - 1;
1867 case MASK(LEFT_CONTIG):
1869 * New allocation is contiguous with a delayed allocation
1871 * Merge the new allocation with the left neighbor.
1873 temp = left.br_blockcount + new->br_blockcount;
1874 xfs_bmap_trace_pre_update(fname, "LC", ip, idx - 1,
1876 xfs_bmbt_set_blockcount(ep - 1, temp);
1877 oldlen = STARTBLOCKVAL(left.br_startblock) +
1878 STARTBLOCKVAL(new->br_startblock);
1879 newlen = xfs_bmap_worst_indlen(ip, temp);
1880 xfs_bmbt_set_startblock(ep - 1, NULLSTARTBLOCK((int)newlen));
1881 xfs_bmap_trace_post_update(fname, "LC", ip, idx - 1,
1883 ip->i_df.if_lastex = idx - 1;
1886 case MASK(RIGHT_CONTIG):
1888 * New allocation is contiguous with a delayed allocation
1890 * Merge the new allocation with the right neighbor.
1892 xfs_bmap_trace_pre_update(fname, "RC", ip, idx, XFS_DATA_FORK);
1893 temp = new->br_blockcount + right.br_blockcount;
1894 oldlen = STARTBLOCKVAL(new->br_startblock) +
1895 STARTBLOCKVAL(right.br_startblock);
1896 newlen = xfs_bmap_worst_indlen(ip, temp);
1897 xfs_bmbt_set_allf(ep, new->br_startoff,
1898 NULLSTARTBLOCK((int)newlen), temp, right.br_state);
1899 xfs_bmap_trace_post_update(fname, "RC", ip, idx, XFS_DATA_FORK);
1900 ip->i_df.if_lastex = idx;
1905 * New allocation is not contiguous with another
1906 * delayed allocation.
1907 * Insert a new entry.
1909 oldlen = newlen = 0;
1910 xfs_bmap_trace_insert(fname, "0", ip, idx, 1, new, NULL,
1912 xfs_bmap_insert_exlist(ip, idx, 1, new, XFS_DATA_FORK);
1913 ip->i_df.if_lastex = idx;
1916 if (oldlen != newlen) {
1917 ASSERT(oldlen > newlen);
1918 xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
1919 (int)(oldlen - newlen), rsvd);
1921 * Nothing to do for disk quota accounting here.
1930 #undef STATE_SET_TEST
1935 * Called by xfs_bmap_add_extent to handle cases converting a hole
1936 * to a real allocation.
1938 STATIC int /* error */
1939 xfs_bmap_add_extent_hole_real(
1940 xfs_inode_t *ip, /* incore inode pointer */
1941 xfs_extnum_t idx, /* extent number to update/insert */
1942 xfs_btree_cur_t *cur, /* if null, not a btree */
1943 xfs_bmbt_irec_t *new, /* new data to put in extent list */
1944 int *logflagsp, /* inode logging flags */
1945 int whichfork) /* data or attr fork */
1947 xfs_bmbt_rec_t *ep; /* pointer to extent entry ins. point */
1948 int error; /* error return value */
1949 #ifdef XFS_BMAP_TRACE
1950 static char fname[] = "xfs_bmap_add_extent_hole_real";
1952 int i; /* temp state */
1953 xfs_ifork_t *ifp; /* inode fork pointer */
1954 xfs_bmbt_irec_t left; /* left neighbor extent entry */
1955 xfs_bmbt_irec_t right; /* right neighbor extent entry */
1956 int state; /* state bits, accessed thru macros */
1957 enum { /* bit number definitions for state */
1958 LEFT_CONTIG, RIGHT_CONTIG,
1959 LEFT_DELAY, RIGHT_DELAY,
1960 LEFT_VALID, RIGHT_VALID
1963 #define MASK(b) (1 << (b))
1964 #define MASK2(a,b) (MASK(a) | MASK(b))
1965 #define STATE_SET(b,v) ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1966 #define STATE_TEST(b) (state & MASK(b))
1967 #define STATE_SET_TEST(b,v) ((v) ? ((state |= MASK(b)), 1) : \
1968 ((state &= ~MASK(b)), 0))
1969 #define SWITCH_STATE (state & MASK2(LEFT_CONTIG, RIGHT_CONTIG))
1971 ifp = XFS_IFORK_PTR(ip, whichfork);
1972 ASSERT(idx <= ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
1973 ep = &ifp->if_u1.if_extents[idx];
1976 * Check and set flags if this segment has a left neighbor.
1978 if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1979 xfs_bmbt_get_all(ep - 1, &left);
1980 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(left.br_startblock));
1983 * Check and set flags if this segment has a current value.
1984 * Not true if we're inserting into the "hole" at eof.
1986 if (STATE_SET_TEST(RIGHT_VALID,
1988 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1989 xfs_bmbt_get_all(ep, &right);
1990 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(right.br_startblock));
1993 * We're inserting a real allocation between "left" and "right".
1994 * Set the contiguity flags. Don't let extents get too large.
1996 STATE_SET(LEFT_CONTIG,
1997 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
1998 left.br_startoff + left.br_blockcount == new->br_startoff &&
1999 left.br_startblock + left.br_blockcount == new->br_startblock &&
2000 left.br_state == new->br_state &&
2001 left.br_blockcount + new->br_blockcount <= MAXEXTLEN);
2002 STATE_SET(RIGHT_CONTIG,
2003 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
2004 new->br_startoff + new->br_blockcount == right.br_startoff &&
2005 new->br_startblock + new->br_blockcount ==
2006 right.br_startblock &&
2007 new->br_state == right.br_state &&
2008 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2009 (!STATE_TEST(LEFT_CONTIG) ||
2010 left.br_blockcount + new->br_blockcount +
2011 right.br_blockcount <= MAXEXTLEN));
2014 * Select which case we're in here, and implement it.
2016 switch (SWITCH_STATE) {
2018 case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
2020 * New allocation is contiguous with real allocations on the
2021 * left and on the right.
2022 * Merge all three into a single extent list entry.
2024 xfs_bmap_trace_pre_update(fname, "LC|RC", ip, idx - 1,
2026 xfs_bmbt_set_blockcount(ep - 1,
2027 left.br_blockcount + new->br_blockcount +
2028 right.br_blockcount);
2029 xfs_bmap_trace_post_update(fname, "LC|RC", ip, idx - 1,
2031 xfs_bmap_trace_delete(fname, "LC|RC", ip,
2033 xfs_bmap_delete_exlist(ip, idx, 1, whichfork);
2034 ifp->if_lastex = idx - 1;
2035 XFS_IFORK_NEXT_SET(ip, whichfork,
2036 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2038 *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2041 *logflagsp = XFS_ILOG_CORE;
2042 if ((error = xfs_bmbt_lookup_eq(cur, right.br_startoff,
2043 right.br_startblock, right.br_blockcount, &i)))
2046 if ((error = xfs_bmbt_delete(cur, &i)))
2049 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
2052 error = xfs_bmbt_update(cur, left.br_startoff,
2054 left.br_blockcount + new->br_blockcount +
2055 right.br_blockcount, left.br_state);
2058 case MASK(LEFT_CONTIG):
2060 * New allocation is contiguous with a real allocation
2062 * Merge the new allocation with the left neighbor.
2064 xfs_bmap_trace_pre_update(fname, "LC", ip, idx - 1, whichfork);
2065 xfs_bmbt_set_blockcount(ep - 1,
2066 left.br_blockcount + new->br_blockcount);
2067 xfs_bmap_trace_post_update(fname, "LC", ip, idx - 1, whichfork);
2068 ifp->if_lastex = idx - 1;
2070 *logflagsp = XFS_ILOG_FEXT(whichfork);
2074 if ((error = xfs_bmbt_lookup_eq(cur, left.br_startoff,
2075 left.br_startblock, left.br_blockcount, &i)))
2078 error = xfs_bmbt_update(cur, left.br_startoff,
2080 left.br_blockcount + new->br_blockcount,
2084 case MASK(RIGHT_CONTIG):
2086 * New allocation is contiguous with a real allocation
2088 * Merge the new allocation with the right neighbor.
2090 xfs_bmap_trace_pre_update(fname, "RC", ip, idx, whichfork);
2091 xfs_bmbt_set_allf(ep, new->br_startoff, new->br_startblock,
2092 new->br_blockcount + right.br_blockcount,
2094 xfs_bmap_trace_post_update(fname, "RC", ip, idx, whichfork);
2095 ifp->if_lastex = idx;
2097 *logflagsp = XFS_ILOG_FEXT(whichfork);
2101 if ((error = xfs_bmbt_lookup_eq(cur, right.br_startoff,
2102 right.br_startblock, right.br_blockcount, &i)))
2105 error = xfs_bmbt_update(cur, new->br_startoff,
2107 new->br_blockcount + right.br_blockcount,
2113 * New allocation is not contiguous with another
2115 * Insert a new entry.
2117 xfs_bmap_trace_insert(fname, "0", ip, idx, 1, new, NULL,
2119 xfs_bmap_insert_exlist(ip, idx, 1, new, whichfork);
2120 ifp->if_lastex = idx;
2121 XFS_IFORK_NEXT_SET(ip, whichfork,
2122 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2124 *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2127 *logflagsp = XFS_ILOG_CORE;
2128 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2129 new->br_startblock, new->br_blockcount, &i)))
2132 cur->bc_rec.b.br_state = new->br_state;
2133 if ((error = xfs_bmbt_insert(cur, &i)))
2142 #undef STATE_SET_TEST
2146 return 0; /* keep gcc quite */
2149 #define XFS_ALLOC_GAP_UNITS 4
2152 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2153 * It figures out where to ask the underlying allocator to put the new extent.
2155 STATIC int /* error */
2157 xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2159 xfs_fsblock_t adjust; /* adjustment to block numbers */
2160 xfs_alloctype_t atype=0; /* type for allocation routines */
2161 int error; /* error return value */
2162 xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
2163 xfs_mount_t *mp; /* mount point structure */
2164 int nullfb; /* true if ap->firstblock isn't set */
2165 int rt; /* true if inode is realtime */
2167 xfs_extlen_t prod=0; /* product factor for allocators */
2168 xfs_extlen_t ralen=0; /* realtime allocation length */
2171 #define ISVALID(x,y) \
2173 (x) < mp->m_sb.sb_rblocks : \
2174 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2175 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2176 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2181 mp = ap->ip->i_mount;
2182 nullfb = ap->firstblock == NULLFSBLOCK;
2183 rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2184 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2187 xfs_extlen_t extsz; /* file extent size for rt */
2188 xfs_fileoff_t nexto; /* next file offset */
2189 xfs_extlen_t orig_alen; /* original ap->alen */
2190 xfs_fileoff_t orig_end; /* original off+len */
2191 xfs_fileoff_t orig_off; /* original ap->off */
2192 xfs_extlen_t mod_off; /* modulus calculations */
2193 xfs_fileoff_t prevo; /* previous file offset */
2194 xfs_rtblock_t rtx; /* realtime extent number */
2195 xfs_extlen_t temp; /* temp for rt calculations */
2198 * Set prod to match the realtime extent size.
2200 if (!(extsz = ap->ip->i_d.di_extsize))
2201 extsz = mp->m_sb.sb_rextsize;
2202 prod = extsz / mp->m_sb.sb_rextsize;
2204 orig_alen = ap->alen;
2205 orig_end = orig_off + orig_alen;
2207 * If the file offset is unaligned vs. the extent size
2208 * we need to align it. This will be possible unless
2209 * the file was previously written with a kernel that didn't
2210 * perform this alignment.
2212 mod_off = do_mod(orig_off, extsz);
2214 ap->alen += mod_off;
2218 * Same adjustment for the end of the requested area.
2220 if ((temp = (ap->alen % extsz)))
2221 ap->alen += extsz - temp;
2223 * If the previous block overlaps with this proposed allocation
2224 * then move the start forward without adjusting the length.
2227 ap->prevp->br_startoff == NULLFILEOFF ?
2229 (ap->prevp->br_startoff +
2230 ap->prevp->br_blockcount);
2231 if (ap->off != orig_off && ap->off < prevo)
2234 * If the next block overlaps with this proposed allocation
2235 * then move the start back without adjusting the length,
2236 * but not before offset 0.
2237 * This may of course make the start overlap previous block,
2238 * and if we hit the offset 0 limit then the next block
2239 * can still overlap too.
2241 nexto = (ap->eof || ap->gotp->br_startoff == NULLFILEOFF) ?
2242 NULLFILEOFF : ap->gotp->br_startoff;
2244 ap->off + ap->alen != orig_end &&
2245 ap->off + ap->alen > nexto)
2246 ap->off = nexto > ap->alen ? nexto - ap->alen : 0;
2248 * If we're now overlapping the next or previous extent that
2249 * means we can't fit an extsz piece in this hole. Just move
2250 * the start forward to the first valid spot and set
2251 * the length so we hit the end.
2253 if ((ap->off != orig_off && ap->off < prevo) ||
2254 (ap->off + ap->alen != orig_end &&
2255 ap->off + ap->alen > nexto)) {
2257 ap->alen = nexto - prevo;
2260 * If the result isn't a multiple of rtextents we need to
2261 * remove blocks until it is.
2263 if ((temp = (ap->alen % mp->m_sb.sb_rextsize))) {
2265 * We're not covering the original request, or
2266 * we won't be able to once we fix the length.
2268 if (orig_off < ap->off ||
2269 orig_end > ap->off + ap->alen ||
2270 ap->alen - temp < orig_alen)
2271 return XFS_ERROR(EINVAL);
2273 * Try to fix it by moving the start up.
2275 if (ap->off + temp <= orig_off) {
2280 * Try to fix it by moving the end in.
2282 else if (ap->off + ap->alen - temp >= orig_end)
2285 * Set the start to the minimum then trim the length.
2288 ap->alen -= orig_off - ap->off;
2290 ap->alen -= ap->alen % mp->m_sb.sb_rextsize;
2293 * Result doesn't cover the request, fail it.
2295 if (orig_off < ap->off || orig_end > ap->off + ap->alen)
2296 return XFS_ERROR(EINVAL);
2298 ASSERT(ap->alen % mp->m_sb.sb_rextsize == 0);
2300 * If the offset & length are not perfectly aligned
2301 * then kill prod, it will just get us in trouble.
2303 if (do_mod(ap->off, extsz) || ap->alen % extsz)
2306 * Set ralen to be the actual requested length in rtextents.
2308 ralen = ap->alen / mp->m_sb.sb_rextsize;
2310 * If the old value was close enough to MAXEXTLEN that
2311 * we rounded up to it, cut it back so it's valid again.
2312 * Note that if it's a really large request (bigger than
2313 * MAXEXTLEN), we don't hear about that number, and can't
2314 * adjust the starting point to match it.
2316 if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2317 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2319 * If it's an allocation to an empty file at offset 0,
2320 * pick an extent that will space things out in the rt area.
2322 if (ap->eof && ap->off == 0) {
2323 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2326 ap->rval = rtx * mp->m_sb.sb_rextsize;
2333 #endif /* __KERNEL__ */
2335 ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2337 ap->rval = ap->firstblock;
2339 * If allocating at eof, and there's a previous real block,
2340 * try to use it's last block as our starting point.
2342 if (ap->eof && ap->prevp->br_startoff != NULLFILEOFF &&
2343 !ISNULLSTARTBLOCK(ap->prevp->br_startblock) &&
2344 ISVALID(ap->prevp->br_startblock + ap->prevp->br_blockcount,
2345 ap->prevp->br_startblock)) {
2346 ap->rval = ap->prevp->br_startblock + ap->prevp->br_blockcount;
2348 * Adjust for the gap between prevp and us.
2351 (ap->prevp->br_startoff + ap->prevp->br_blockcount);
2353 ISVALID(ap->rval + adjust, ap->prevp->br_startblock))
2357 * If not at eof, then compare the two neighbor blocks.
2358 * Figure out whether either one gives us a good starting point,
2359 * and pick the better one.
2361 else if (!ap->eof) {
2362 xfs_fsblock_t gotbno; /* right side block number */
2363 xfs_fsblock_t gotdiff=0; /* right side difference */
2364 xfs_fsblock_t prevbno; /* left side block number */
2365 xfs_fsblock_t prevdiff=0; /* left side difference */
2368 * If there's a previous (left) block, select a requested
2369 * start block based on it.
2371 if (ap->prevp->br_startoff != NULLFILEOFF &&
2372 !ISNULLSTARTBLOCK(ap->prevp->br_startblock) &&
2373 (prevbno = ap->prevp->br_startblock +
2374 ap->prevp->br_blockcount) &&
2375 ISVALID(prevbno, ap->prevp->br_startblock)) {
2377 * Calculate gap to end of previous block.
2379 adjust = prevdiff = ap->off -
2380 (ap->prevp->br_startoff +
2381 ap->prevp->br_blockcount);
2383 * Figure the startblock based on the previous block's
2384 * end and the gap size.
2386 * If the gap is large relative to the piece we're
2387 * allocating, or using it gives us an invalid block
2388 * number, then just use the end of the previous block.
2390 if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2391 ISVALID(prevbno + prevdiff,
2392 ap->prevp->br_startblock))
2397 * If the firstblock forbids it, can't use it,
2400 if (!rt && !nullfb &&
2401 XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2402 prevbno = NULLFSBLOCK;
2405 * No previous block or can't follow it, just default.
2408 prevbno = NULLFSBLOCK;
2410 * If there's a following (right) block, select a requested
2411 * start block based on it.
2413 if (!ISNULLSTARTBLOCK(ap->gotp->br_startblock)) {
2415 * Calculate gap to start of next block.
2417 adjust = gotdiff = ap->gotp->br_startoff - ap->off;
2419 * Figure the startblock based on the next block's
2420 * start and the gap size.
2422 gotbno = ap->gotp->br_startblock;
2425 * If the gap is large relative to the piece we're
2426 * allocating, or using it gives us an invalid block
2427 * number, then just use the start of the next block
2428 * offset by our length.
2430 if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2431 ISVALID(gotbno - gotdiff, gotbno))
2433 else if (ISVALID(gotbno - ap->alen, gotbno)) {
2435 gotdiff += adjust - ap->alen;
2439 * If the firstblock forbids it, can't use it,
2442 if (!rt && !nullfb &&
2443 XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2444 gotbno = NULLFSBLOCK;
2447 * No next block, just default.
2450 gotbno = NULLFSBLOCK;
2452 * If both valid, pick the better one, else the only good
2453 * one, else ap->rval is already set (to 0 or the inode block).
2455 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2456 ap->rval = prevdiff <= gotdiff ? prevbno : gotbno;
2457 else if (prevbno != NULLFSBLOCK)
2459 else if (gotbno != NULLFSBLOCK)
2463 * If allowed, use ap->rval; otherwise must use firstblock since
2464 * it's in the right allocation group.
2466 if (nullfb || rt || XFS_FSB_TO_AGNO(mp, ap->rval) == fb_agno)
2469 ap->rval = ap->firstblock;
2471 * Realtime allocation, done through xfs_rtallocate_extent.
2479 atype = ap->rval == 0 ?
2480 XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2481 do_div(ap->rval, mp->m_sb.sb_rextsize);
2484 if ((error = xfs_rtallocate_extent(ap->tp, ap->rval, 1, ap->alen,
2485 &ralen, atype, ap->wasdel, prod, &rtb)))
2487 if (rtb == NULLFSBLOCK && prod > 1 &&
2488 (error = xfs_rtallocate_extent(ap->tp, ap->rval, 1,
2489 ap->alen, &ralen, atype,
2490 ap->wasdel, 1, &rtb)))
2493 if (ap->rval != NULLFSBLOCK) {
2494 ap->rval *= mp->m_sb.sb_rextsize;
2495 ralen *= mp->m_sb.sb_rextsize;
2497 ap->ip->i_d.di_nblocks += ralen;
2498 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2500 ap->ip->i_delayed_blks -= ralen;
2502 * Adjust the disk quota also. This was reserved
2505 XFS_TRANS_MOD_DQUOT_BYINO(mp, ap->tp, ap->ip,
2506 ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2507 XFS_TRANS_DQ_RTBCOUNT,
2511 #endif /* __KERNEL__ */
2514 * Normal allocation, done through xfs_alloc_vextent.
2518 xfs_alloc_arg_t args;
2522 xfs_extlen_t longest;
2524 xfs_extlen_t nextminlen=0;
2527 xfs_agnumber_t startag;
2530 tryagain = isaligned = 0;
2533 args.fsbno = ap->rval;
2534 args.maxlen = MIN(ap->alen, mp->m_sb.sb_agblocks);
2537 args.type = XFS_ALLOCTYPE_START_BNO;
2538 args.total = ap->total;
2540 * Find the longest available space.
2541 * We're going to try for the whole allocation at once.
2543 startag = ag = XFS_FSB_TO_AGNO(mp, args.fsbno);
2545 down_read(&mp->m_peraglock);
2546 while (blen < ap->alen) {
2547 pag = &mp->m_perag[ag];
2548 if (!pag->pagf_init &&
2549 (error = xfs_alloc_pagf_init(mp, args.tp,
2550 ag, XFS_ALLOC_FLAG_TRYLOCK))) {
2551 up_read(&mp->m_peraglock);
2555 * See xfs_alloc_fix_freelist...
2557 if (pag->pagf_init) {
2558 need = XFS_MIN_FREELIST_PAG(pag, mp);
2559 delta = need > pag->pagf_flcount ?
2560 need - pag->pagf_flcount : 0;
2561 longest = (pag->pagf_longest > delta) ?
2562 (pag->pagf_longest - delta) :
2563 (pag->pagf_flcount > 0 ||
2564 pag->pagf_longest > 0);
2569 if (++ag == mp->m_sb.sb_agcount)
2574 up_read(&mp->m_peraglock);
2576 * Since the above loop did a BUF_TRYLOCK, it is
2577 * possible that there is space for this request.
2579 if (notinit || blen < ap->minlen)
2580 args.minlen = ap->minlen;
2582 * If the best seen length is less than the request
2583 * length, use the best as the minimum.
2585 else if (blen < ap->alen)
2588 * Otherwise we've seen an extent as big as alen,
2589 * use that as the minimum.
2592 args.minlen = ap->alen;
2593 } else if (ap->low) {
2594 args.type = XFS_ALLOCTYPE_FIRST_AG;
2595 args.total = args.minlen = ap->minlen;
2597 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2598 args.total = ap->total;
2599 args.minlen = ap->minlen;
2601 if (ap->ip->i_d.di_extsize) {
2602 args.prod = ap->ip->i_d.di_extsize;
2603 if ((args.mod = (xfs_extlen_t)do_mod(ap->off, args.prod)))
2604 args.mod = (xfs_extlen_t)(args.prod - args.mod);
2605 } else if (mp->m_sb.sb_blocksize >= NBPP) {
2609 args.prod = NBPP >> mp->m_sb.sb_blocklog;
2610 if ((args.mod = (xfs_extlen_t)(do_mod(ap->off, args.prod))))
2611 args.mod = (xfs_extlen_t)(args.prod - args.mod);
2614 * If we are not low on available data blocks, and the
2615 * underlying logical volume manager is a stripe, and
2616 * the file offset is zero then try to allocate data
2617 * blocks on stripe unit boundary.
2618 * NOTE: ap->aeof is only set if the allocation length
2619 * is >= the stripe unit and the allocation offset is
2620 * at the end of file.
2622 if (!ap->low && ap->aeof) {
2624 args.alignment = mp->m_dalign;
2628 * Adjust for alignment
2630 if (blen > args.alignment && blen <= ap->alen)
2631 args.minlen = blen - args.alignment;
2632 args.minalignslop = 0;
2635 * First try an exact bno allocation.
2636 * If it fails then do a near or start bno
2637 * allocation with alignment turned on.
2641 args.type = XFS_ALLOCTYPE_THIS_BNO;
2644 * Compute the minlen+alignment for the
2645 * next case. Set slop so that the value
2646 * of minlen+alignment+slop doesn't go up
2647 * between the calls.
2649 if (blen > mp->m_dalign && blen <= ap->alen)
2650 nextminlen = blen - mp->m_dalign;
2652 nextminlen = args.minlen;
2653 if (nextminlen + mp->m_dalign > args.minlen + 1)
2655 nextminlen + mp->m_dalign -
2658 args.minalignslop = 0;
2662 args.minalignslop = 0;
2664 args.minleft = ap->minleft;
2665 args.wasdel = ap->wasdel;
2667 args.userdata = ap->userdata;
2668 if ((error = xfs_alloc_vextent(&args)))
2670 if (tryagain && args.fsbno == NULLFSBLOCK) {
2672 * Exact allocation failed. Now try with alignment
2676 args.fsbno = ap->rval;
2677 args.alignment = mp->m_dalign;
2678 args.minlen = nextminlen;
2679 args.minalignslop = 0;
2681 if ((error = xfs_alloc_vextent(&args)))
2684 if (isaligned && args.fsbno == NULLFSBLOCK) {
2686 * allocation failed, so turn off alignment and
2690 args.fsbno = ap->rval;
2692 if ((error = xfs_alloc_vextent(&args)))
2695 if (args.fsbno == NULLFSBLOCK && nullfb &&
2696 args.minlen > ap->minlen) {
2697 args.minlen = ap->minlen;
2698 args.type = XFS_ALLOCTYPE_START_BNO;
2699 args.fsbno = ap->rval;
2700 if ((error = xfs_alloc_vextent(&args)))
2703 if (args.fsbno == NULLFSBLOCK && nullfb) {
2705 args.type = XFS_ALLOCTYPE_FIRST_AG;
2706 args.total = ap->minlen;
2708 if ((error = xfs_alloc_vextent(&args)))
2712 if (args.fsbno != NULLFSBLOCK) {
2713 ap->firstblock = ap->rval = args.fsbno;
2714 ASSERT(nullfb || fb_agno == args.agno ||
2715 (ap->low && fb_agno < args.agno));
2716 ap->alen = args.len;
2717 ap->ip->i_d.di_nblocks += args.len;
2718 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2720 ap->ip->i_delayed_blks -= args.len;
2722 * Adjust the disk quota also. This was reserved
2725 XFS_TRANS_MOD_DQUOT_BYINO(mp, ap->tp, ap->ip,
2726 ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2727 XFS_TRANS_DQ_BCOUNT,
2730 ap->rval = NULLFSBLOCK;
2739 * Transform a btree format file with only one leaf node, where the
2740 * extents list will fit in the inode, into an extents format file.
2741 * Since the extent list is already in-core, all we have to do is
2742 * give up the space for the btree root and pitch the leaf block.
2744 STATIC int /* error */
2745 xfs_bmap_btree_to_extents(
2746 xfs_trans_t *tp, /* transaction pointer */
2747 xfs_inode_t *ip, /* incore inode pointer */
2748 xfs_btree_cur_t *cur, /* btree cursor */
2749 int *logflagsp, /* inode logging flags */
2750 int whichfork) /* data or attr fork */
2753 xfs_bmbt_block_t *cblock;/* child btree block */
2754 xfs_fsblock_t cbno; /* child block number */
2755 xfs_buf_t *cbp; /* child block's buffer */
2756 int error; /* error return value */
2757 xfs_ifork_t *ifp; /* inode fork data */
2758 xfs_mount_t *mp; /* mount point structure */
2759 xfs_bmbt_ptr_t *pp; /* ptr to block address */
2760 xfs_bmbt_block_t *rblock;/* root btree block */
2762 ifp = XFS_IFORK_PTR(ip, whichfork);
2763 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2764 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
2765 rblock = ifp->if_broot;
2766 ASSERT(be16_to_cpu(rblock->bb_level) == 1);
2767 ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
2768 ASSERT(XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes) == 1);
2770 pp = XFS_BMAP_BROOT_PTR_ADDR(rblock, 1, ifp->if_broot_bytes);
2773 if ((error = xfs_btree_check_lptr(cur, INT_GET(*pp, ARCH_CONVERT), 1)))
2776 cbno = INT_GET(*pp, ARCH_CONVERT);
2777 if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
2778 XFS_BMAP_BTREE_REF)))
2780 cblock = XFS_BUF_TO_BMBT_BLOCK(cbp);
2781 if ((error = xfs_btree_check_lblock(cur, cblock, 0, cbp)))
2783 xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
2784 ip->i_d.di_nblocks--;
2785 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
2786 xfs_trans_binval(tp, cbp);
2787 if (cur->bc_bufs[0] == cbp)
2788 cur->bc_bufs[0] = NULL;
2789 xfs_iroot_realloc(ip, -1, whichfork);
2790 ASSERT(ifp->if_broot == NULL);
2791 ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
2792 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
2793 *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2798 * Called by xfs_bmapi to update extent list structure and the btree
2799 * after removing space (or undoing a delayed allocation).
2801 STATIC int /* error */
2802 xfs_bmap_del_extent(
2803 xfs_inode_t *ip, /* incore inode pointer */
2804 xfs_trans_t *tp, /* current transaction pointer */
2805 xfs_extnum_t idx, /* extent number to update/delete */
2806 xfs_bmap_free_t *flist, /* list of extents to be freed */
2807 xfs_btree_cur_t *cur, /* if null, not a btree */
2808 xfs_bmbt_irec_t *del, /* data to remove from extent list */
2809 int *logflagsp, /* inode logging flags */
2810 int whichfork, /* data or attr fork */
2811 int rsvd) /* OK to allocate reserved blocks */
2813 xfs_filblks_t da_new; /* new delay-alloc indirect blocks */
2814 xfs_filblks_t da_old; /* old delay-alloc indirect blocks */
2815 xfs_fsblock_t del_endblock=0; /* first block past del */
2816 xfs_fileoff_t del_endoff; /* first offset past del */
2817 int delay; /* current block is delayed allocated */
2818 int do_fx; /* free extent at end of routine */
2819 xfs_bmbt_rec_t *ep; /* current extent entry pointer */
2820 int error; /* error return value */
2821 int flags; /* inode logging flags */
2822 #ifdef XFS_BMAP_TRACE
2823 static char fname[] = "xfs_bmap_del_extent";
2825 xfs_bmbt_irec_t got; /* current extent entry */
2826 xfs_fileoff_t got_endoff; /* first offset past got */
2827 int i; /* temp state */
2828 xfs_ifork_t *ifp; /* inode fork pointer */
2829 xfs_mount_t *mp; /* mount structure */
2830 xfs_filblks_t nblks; /* quota/sb block count */
2831 xfs_bmbt_irec_t new; /* new record to be inserted */
2833 xfs_extnum_t nextents; /* number of extents in list */
2834 uint qfield; /* quota field to update */
2835 xfs_filblks_t temp; /* for indirect length calculations */
2836 xfs_filblks_t temp2; /* for indirect length calculations */
2838 XFS_STATS_INC(xs_del_exlist);
2840 ifp = XFS_IFORK_PTR(ip, whichfork);
2841 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2842 ASSERT(idx >= 0 && idx < nextents);
2843 ASSERT(del->br_blockcount > 0);
2844 ep = &ifp->if_u1.if_extents[idx];
2845 xfs_bmbt_get_all(ep, &got);
2846 ASSERT(got.br_startoff <= del->br_startoff);
2847 del_endoff = del->br_startoff + del->br_blockcount;
2848 got_endoff = got.br_startoff + got.br_blockcount;
2849 ASSERT(got_endoff >= del_endoff);
2850 delay = ISNULLSTARTBLOCK(got.br_startblock);
2851 ASSERT(ISNULLSTARTBLOCK(del->br_startblock) == delay);
2856 * If deleting a real allocation, must free up the disk space.
2859 flags = XFS_ILOG_CORE;
2861 * Realtime allocation. Free it and record di_nblocks update.
2863 if (whichfork == XFS_DATA_FORK &&
2864 (ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
2868 ASSERT(do_mod(del->br_blockcount,
2869 mp->m_sb.sb_rextsize) == 0);
2870 ASSERT(do_mod(del->br_startblock,
2871 mp->m_sb.sb_rextsize) == 0);
2872 bno = del->br_startblock;
2873 len = del->br_blockcount;
2874 do_div(bno, mp->m_sb.sb_rextsize);
2875 do_div(len, mp->m_sb.sb_rextsize);
2876 if ((error = xfs_rtfree_extent(ip->i_transp, bno,
2877 (xfs_extlen_t)len)))
2880 nblks = len * mp->m_sb.sb_rextsize;
2881 qfield = XFS_TRANS_DQ_RTBCOUNT;
2884 * Ordinary allocation.
2888 nblks = del->br_blockcount;
2889 qfield = XFS_TRANS_DQ_BCOUNT;
2892 * Set up del_endblock and cur for later.
2894 del_endblock = del->br_startblock + del->br_blockcount;
2896 if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
2897 got.br_startblock, got.br_blockcount,
2902 da_old = da_new = 0;
2904 da_old = STARTBLOCKVAL(got.br_startblock);
2910 * Set flag value to use in switch statement.
2911 * Left-contig is 2, right-contig is 1.
2913 switch (((got.br_startoff == del->br_startoff) << 1) |
2914 (got_endoff == del_endoff)) {
2917 * Matches the whole extent. Delete the entry.
2919 xfs_bmap_trace_delete(fname, "3", ip, idx, 1, whichfork);
2920 xfs_bmap_delete_exlist(ip, idx, 1, whichfork);
2921 ifp->if_lastex = idx;
2924 XFS_IFORK_NEXT_SET(ip, whichfork,
2925 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2926 flags |= XFS_ILOG_CORE;
2928 flags |= XFS_ILOG_FEXT(whichfork);
2931 if ((error = xfs_bmbt_delete(cur, &i)))
2938 * Deleting the first part of the extent.
2940 xfs_bmap_trace_pre_update(fname, "2", ip, idx, whichfork);
2941 xfs_bmbt_set_startoff(ep, del_endoff);
2942 temp = got.br_blockcount - del->br_blockcount;
2943 xfs_bmbt_set_blockcount(ep, temp);
2944 ifp->if_lastex = idx;
2946 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2948 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
2949 xfs_bmap_trace_post_update(fname, "2", ip, idx,
2954 xfs_bmbt_set_startblock(ep, del_endblock);
2955 xfs_bmap_trace_post_update(fname, "2", ip, idx, whichfork);
2957 flags |= XFS_ILOG_FEXT(whichfork);
2960 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
2961 got.br_blockcount - del->br_blockcount,
2968 * Deleting the last part of the extent.
2970 temp = got.br_blockcount - del->br_blockcount;
2971 xfs_bmap_trace_pre_update(fname, "1", ip, idx, whichfork);
2972 xfs_bmbt_set_blockcount(ep, temp);
2973 ifp->if_lastex = idx;
2975 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2977 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
2978 xfs_bmap_trace_post_update(fname, "1", ip, idx,
2983 xfs_bmap_trace_post_update(fname, "1", ip, idx, whichfork);
2985 flags |= XFS_ILOG_FEXT(whichfork);
2988 if ((error = xfs_bmbt_update(cur, got.br_startoff,
2990 got.br_blockcount - del->br_blockcount,
2997 * Deleting the middle of the extent.
2999 temp = del->br_startoff - got.br_startoff;
3000 xfs_bmap_trace_pre_update(fname, "0", ip, idx, whichfork);
3001 xfs_bmbt_set_blockcount(ep, temp);
3002 new.br_startoff = del_endoff;
3003 temp2 = got_endoff - del_endoff;
3004 new.br_blockcount = temp2;
3005 new.br_state = got.br_state;
3007 new.br_startblock = del_endblock;
3008 flags |= XFS_ILOG_CORE;
3010 if ((error = xfs_bmbt_update(cur,
3012 got.br_startblock, temp,
3015 if ((error = xfs_bmbt_increment(cur, 0, &i)))
3017 cur->bc_rec.b = new;
3018 error = xfs_bmbt_insert(cur, &i);
3019 if (error && error != ENOSPC)
3022 * If get no-space back from btree insert,
3023 * it tried a split, and we have a zero
3024 * block reservation.
3025 * Fix up our state and return the error.
3027 if (error == ENOSPC) {
3029 * Reset the cursor, don't trust
3030 * it after any insert operation.
3032 if ((error = xfs_bmbt_lookup_eq(cur,
3039 * Update the btree record back
3040 * to the original value.
3042 if ((error = xfs_bmbt_update(cur,
3049 * Reset the extent record back
3050 * to the original value.
3052 xfs_bmbt_set_blockcount(ep,
3055 error = XFS_ERROR(ENOSPC);
3060 flags |= XFS_ILOG_FEXT(whichfork);
3061 XFS_IFORK_NEXT_SET(ip, whichfork,
3062 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
3064 ASSERT(whichfork == XFS_DATA_FORK);
3065 temp = xfs_bmap_worst_indlen(ip, temp);
3066 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3067 temp2 = xfs_bmap_worst_indlen(ip, temp2);
3068 new.br_startblock = NULLSTARTBLOCK((int)temp2);
3069 da_new = temp + temp2;
3070 while (da_new > da_old) {
3074 xfs_bmbt_set_startblock(ep,
3075 NULLSTARTBLOCK((int)temp));
3077 if (da_new == da_old)
3083 NULLSTARTBLOCK((int)temp2);
3087 xfs_bmap_trace_post_update(fname, "0", ip, idx, whichfork);
3088 xfs_bmap_trace_insert(fname, "0", ip, idx + 1, 1, &new, NULL,
3090 xfs_bmap_insert_exlist(ip, idx + 1, 1, &new, whichfork);
3091 ifp->if_lastex = idx + 1;
3095 * If we need to, add to list of extents to delete.
3098 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
3101 * Adjust inode # blocks in the file.
3104 ip->i_d.di_nblocks -= nblks;
3106 * Adjust quota data.
3109 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, qfield, (long)-nblks);
3112 * Account for change in delayed indirect blocks.
3113 * Nothing to do for disk quota accounting here.
3115 ASSERT(da_old >= da_new);
3116 if (da_old > da_new)
3117 xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS, (int)(da_old - da_new),
3125 * Remove the entry "free" from the free item list. Prev points to the
3126 * previous entry, unless "free" is the head of the list.
3130 xfs_bmap_free_t *flist, /* free item list header */
3131 xfs_bmap_free_item_t *prev, /* previous item on list, if any */
3132 xfs_bmap_free_item_t *free) /* list item to be freed */
3135 prev->xbfi_next = free->xbfi_next;
3137 flist->xbf_first = free->xbfi_next;
3139 kmem_zone_free(xfs_bmap_free_item_zone, free);
3143 * Remove count entries from the extents array for inode "ip", starting
3144 * at index "idx". Copies the remaining items down over the deleted ones,
3145 * and gives back the excess memory.
3148 xfs_bmap_delete_exlist(
3149 xfs_inode_t *ip, /* incore inode pointer */
3150 xfs_extnum_t idx, /* starting delete index */
3151 xfs_extnum_t count, /* count of items to delete */
3152 int whichfork) /* data or attr fork */
3154 xfs_bmbt_rec_t *base; /* base of extent list */
3155 xfs_ifork_t *ifp; /* inode fork pointer */
3156 xfs_extnum_t nextents; /* number of extents in list after */
3158 ifp = XFS_IFORK_PTR(ip, whichfork);
3159 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3160 base = ifp->if_u1.if_extents;
3161 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - count;
3162 memmove(&base[idx], &base[idx + count],
3163 (nextents - idx) * sizeof(*base));
3164 xfs_iext_realloc(ip, -count, whichfork);
3168 * Convert an extents-format file into a btree-format file.
3169 * The new file will have a root block (in the inode) and a single child block.
3171 STATIC int /* error */
3172 xfs_bmap_extents_to_btree(
3173 xfs_trans_t *tp, /* transaction pointer */
3174 xfs_inode_t *ip, /* incore inode pointer */
3175 xfs_fsblock_t *firstblock, /* first-block-allocated */
3176 xfs_bmap_free_t *flist, /* blocks freed in xaction */
3177 xfs_btree_cur_t **curp, /* cursor returned to caller */
3178 int wasdel, /* converting a delayed alloc */
3179 int *logflagsp, /* inode logging flags */
3180 int whichfork) /* data or attr fork */
3182 xfs_bmbt_block_t *ablock; /* allocated (child) bt block */
3183 xfs_buf_t *abp; /* buffer for ablock */
3184 xfs_alloc_arg_t args; /* allocation arguments */
3185 xfs_bmbt_rec_t *arp; /* child record pointer */
3186 xfs_bmbt_block_t *block; /* btree root block */
3187 xfs_btree_cur_t *cur; /* bmap btree cursor */
3188 xfs_bmbt_rec_t *ep; /* extent list pointer */
3189 int error; /* error return value */
3190 xfs_extnum_t i, cnt; /* extent list index */
3191 xfs_ifork_t *ifp; /* inode fork pointer */
3192 xfs_bmbt_key_t *kp; /* root block key pointer */
3193 xfs_mount_t *mp; /* mount structure */
3194 xfs_extnum_t nextents; /* extent list size */
3195 xfs_bmbt_ptr_t *pp; /* root block address pointer */
3197 ifp = XFS_IFORK_PTR(ip, whichfork);
3198 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3199 ASSERT(ifp->if_ext_max ==
3200 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
3202 * Make space in the inode incore.
3204 xfs_iroot_realloc(ip, 1, whichfork);
3205 ifp->if_flags |= XFS_IFBROOT;
3209 block = ifp->if_broot;
3210 block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3211 block->bb_level = cpu_to_be16(1);
3212 block->bb_numrecs = cpu_to_be16(1);
3213 block->bb_leftsib = cpu_to_be64(NULLDFSBNO);
3214 block->bb_rightsib = cpu_to_be64(NULLDFSBNO);
3216 * Need a cursor. Can't allocate until bb_level is filled in.
3219 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
3221 cur->bc_private.b.firstblock = *firstblock;
3222 cur->bc_private.b.flist = flist;
3223 cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3225 * Convert to a btree with two levels, one record in root.
3227 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3230 if (*firstblock == NULLFSBLOCK) {
3231 args.type = XFS_ALLOCTYPE_START_BNO;
3232 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3233 } else if (flist->xbf_low) {
3234 args.type = XFS_ALLOCTYPE_START_BNO;
3235 args.fsbno = *firstblock;
3237 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3238 args.fsbno = *firstblock;
3240 args.minlen = args.maxlen = args.prod = 1;
3241 args.total = args.minleft = args.alignment = args.mod = args.isfl =
3242 args.minalignslop = 0;
3243 args.wasdel = wasdel;
3245 if ((error = xfs_alloc_vextent(&args))) {
3246 xfs_iroot_realloc(ip, -1, whichfork);
3247 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3251 * Allocation can't fail, the space was reserved.
3253 ASSERT(args.fsbno != NULLFSBLOCK);
3254 ASSERT(*firstblock == NULLFSBLOCK ||
3255 args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3257 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3258 *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3259 cur->bc_private.b.allocated++;
3260 ip->i_d.di_nblocks++;
3261 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3262 abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3264 * Fill in the child block.
3266 ablock = XFS_BUF_TO_BMBT_BLOCK(abp);
3267 ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3268 ablock->bb_level = 0;
3269 ablock->bb_leftsib = cpu_to_be64(NULLDFSBNO);
3270 ablock->bb_rightsib = cpu_to_be64(NULLDFSBNO);
3271 arp = XFS_BMAP_REC_IADDR(ablock, 1, cur);
3272 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3273 for (ep = ifp->if_u1.if_extents, cnt = i = 0; i < nextents; i++, ep++) {
3274 if (!ISNULLSTARTBLOCK(xfs_bmbt_get_startblock(ep))) {
3275 arp->l0 = INT_GET(ep->l0, ARCH_CONVERT);
3276 arp->l1 = INT_GET(ep->l1, ARCH_CONVERT);
3280 ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3281 ablock->bb_numrecs = cpu_to_be16(cnt);
3283 * Fill in the root key and pointer.
3285 kp = XFS_BMAP_KEY_IADDR(block, 1, cur);
3286 arp = XFS_BMAP_REC_IADDR(ablock, 1, cur);
3287 INT_SET(kp->br_startoff, ARCH_CONVERT, xfs_bmbt_disk_get_startoff(arp));
3288 pp = XFS_BMAP_PTR_IADDR(block, 1, cur);
3289 INT_SET(*pp, ARCH_CONVERT, args.fsbno);
3291 * Do all this logging at the end so that
3292 * the root is at the right level.
3294 xfs_bmbt_log_block(cur, abp, XFS_BB_ALL_BITS);
3295 xfs_bmbt_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3296 ASSERT(*curp == NULL);
3298 *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FBROOT(whichfork);
3303 * Insert new item(s) in the extent list for inode "ip".
3304 * Count new items are inserted at offset idx.
3307 xfs_bmap_insert_exlist(
3308 xfs_inode_t *ip, /* incore inode pointer */
3309 xfs_extnum_t idx, /* starting index of new items */
3310 xfs_extnum_t count, /* number of inserted items */
3311 xfs_bmbt_irec_t *new, /* items to insert */
3312 int whichfork) /* data or attr fork */
3314 xfs_bmbt_rec_t *base; /* extent list base */
3315 xfs_ifork_t *ifp; /* inode fork pointer */
3316 xfs_extnum_t nextents; /* extent list size */
3317 xfs_extnum_t to; /* extent list index */
3319 ifp = XFS_IFORK_PTR(ip, whichfork);
3320 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3321 xfs_iext_realloc(ip, count, whichfork);
3322 base = ifp->if_u1.if_extents;
3323 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3324 memmove(&base[idx + count], &base[idx],
3325 (nextents - (idx + count)) * sizeof(*base));
3326 for (to = idx; to < idx + count; to++, new++)
3327 xfs_bmbt_set_all(&base[to], new);
3331 * Helper routine to reset inode di_forkoff field when switching
3332 * attribute fork from local to extent format - we reset it where
3333 * possible to make space available for inline data fork extents.
3336 xfs_bmap_forkoff_reset(
3341 if (whichfork == XFS_ATTR_FORK &&
3342 (ip->i_d.di_format != XFS_DINODE_FMT_DEV) &&
3343 (ip->i_d.di_format != XFS_DINODE_FMT_UUID) &&
3344 ((mp->m_attroffset >> 3) > ip->i_d.di_forkoff)) {
3345 ip->i_d.di_forkoff = mp->m_attroffset >> 3;
3346 ip->i_df.if_ext_max = XFS_IFORK_DSIZE(ip) /
3347 (uint)sizeof(xfs_bmbt_rec_t);
3348 ip->i_afp->if_ext_max = XFS_IFORK_ASIZE(ip) /
3349 (uint)sizeof(xfs_bmbt_rec_t);
3354 * Convert a local file to an extents file.
3355 * This code is out of bounds for data forks of regular files,
3356 * since the file data needs to get logged so things will stay consistent.
3357 * (The bmap-level manipulations are ok, though).
3359 STATIC int /* error */
3360 xfs_bmap_local_to_extents(
3361 xfs_trans_t *tp, /* transaction pointer */
3362 xfs_inode_t *ip, /* incore inode pointer */
3363 xfs_fsblock_t *firstblock, /* first block allocated in xaction */
3364 xfs_extlen_t total, /* total blocks needed by transaction */
3365 int *logflagsp, /* inode logging flags */
3366 int whichfork) /* data or attr fork */
3368 int error; /* error return value */
3369 int flags; /* logging flags returned */
3370 #ifdef XFS_BMAP_TRACE
3371 static char fname[] = "xfs_bmap_local_to_extents";
3373 xfs_ifork_t *ifp; /* inode fork pointer */
3376 * We don't want to deal with the case of keeping inode data inline yet.
3377 * So sending the data fork of a regular inode is invalid.
3379 ASSERT(!((ip->i_d.di_mode & S_IFMT) == S_IFREG &&
3380 whichfork == XFS_DATA_FORK));
3381 ifp = XFS_IFORK_PTR(ip, whichfork);
3382 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3385 if (ifp->if_bytes) {
3386 xfs_alloc_arg_t args; /* allocation arguments */
3387 xfs_buf_t *bp; /* buffer for extent list block */
3388 xfs_bmbt_rec_t *ep; /* extent list pointer */
3391 args.mp = ip->i_mount;
3392 ASSERT(ifp->if_flags & XFS_IFINLINE);
3394 * Allocate a block. We know we need only one, since the
3395 * file currently fits in an inode.
3397 if (*firstblock == NULLFSBLOCK) {
3398 args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3399 args.type = XFS_ALLOCTYPE_START_BNO;
3401 args.fsbno = *firstblock;
3402 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3405 args.mod = args.minleft = args.alignment = args.wasdel =
3406 args.isfl = args.minalignslop = 0;
3407 args.minlen = args.maxlen = args.prod = 1;
3408 if ((error = xfs_alloc_vextent(&args)))
3411 * Can't fail, the space was reserved.
3413 ASSERT(args.fsbno != NULLFSBLOCK);
3414 ASSERT(args.len == 1);
3415 *firstblock = args.fsbno;
3416 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3417 memcpy((char *)XFS_BUF_PTR(bp), ifp->if_u1.if_data,
3419 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3420 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3421 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3422 xfs_iext_realloc(ip, 1, whichfork);
3423 ep = ifp->if_u1.if_extents;
3424 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3425 xfs_bmap_trace_post_update(fname, "new", ip, 0, whichfork);
3426 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3427 ip->i_d.di_nblocks = 1;
3428 XFS_TRANS_MOD_DQUOT_BYINO(args.mp, tp, ip,
3429 XFS_TRANS_DQ_BCOUNT, 1L);
3430 flags |= XFS_ILOG_FEXT(whichfork);
3432 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3433 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3435 ifp->if_flags &= ~XFS_IFINLINE;
3436 ifp->if_flags |= XFS_IFEXTENTS;
3437 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3438 flags |= XFS_ILOG_CORE;
3444 xfs_bmbt_rec_t * /* pointer to found extent entry */
3445 xfs_bmap_do_search_extents(
3446 xfs_bmbt_rec_t *base, /* base of extent list */
3447 xfs_extnum_t lastx, /* last extent index used */
3448 xfs_extnum_t nextents, /* extent list size */
3449 xfs_fileoff_t bno, /* block number searched for */
3450 int *eofp, /* out: end of file found */
3451 xfs_extnum_t *lastxp, /* out: last extent index */
3452 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
3453 xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
3455 xfs_bmbt_rec_t *ep; /* extent list entry pointer */
3456 xfs_bmbt_irec_t got; /* extent list entry, decoded */
3457 int high; /* high index of binary search */
3458 int low; /* low index of binary search */
3461 * Initialize the extent entry structure to catch access to
3462 * uninitialized br_startblock field.
3464 got.br_startoff = 0xffa5a5a5a5a5a5a5LL;
3465 got.br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3466 got.br_state = XFS_EXT_INVALID;
3469 got.br_startblock = 0xffffa5a5a5a5a5a5LL;
3471 got.br_startblock = 0xffffa5a5;
3474 if (lastx != NULLEXTNUM && lastx < nextents)
3478 prevp->br_startoff = NULLFILEOFF;
3479 if (ep && bno >= (got.br_startoff = xfs_bmbt_get_startoff(ep)) &&
3480 bno < got.br_startoff +
3481 (got.br_blockcount = xfs_bmbt_get_blockcount(ep)))
3483 else if (ep && lastx < nextents - 1 &&
3484 bno >= (got.br_startoff = xfs_bmbt_get_startoff(ep + 1)) &&
3485 bno < got.br_startoff +
3486 (got.br_blockcount = xfs_bmbt_get_blockcount(ep + 1))) {
3490 } else if (nextents == 0)
3492 else if (bno == 0 &&
3493 (got.br_startoff = xfs_bmbt_get_startoff(base)) == 0) {
3496 got.br_blockcount = xfs_bmbt_get_blockcount(ep);
3499 /* binary search the extents array */
3501 high = nextents - 1;
3502 while (low <= high) {
3503 XFS_STATS_INC(xs_cmp_exlist);
3504 lastx = (low + high) >> 1;
3506 got.br_startoff = xfs_bmbt_get_startoff(ep);
3507 got.br_blockcount = xfs_bmbt_get_blockcount(ep);
3508 if (bno < got.br_startoff)
3510 else if (bno >= got.br_startoff + got.br_blockcount)
3513 got.br_startblock = xfs_bmbt_get_startblock(ep);
3514 got.br_state = xfs_bmbt_get_state(ep);
3521 if (bno >= got.br_startoff + got.br_blockcount) {
3523 if (lastx == nextents) {
3525 got.br_startblock = xfs_bmbt_get_startblock(ep);
3526 got.br_state = xfs_bmbt_get_state(ep);
3531 xfs_bmbt_get_all(ep, prevp);
3533 got.br_startoff = xfs_bmbt_get_startoff(ep);
3534 got.br_blockcount = xfs_bmbt_get_blockcount(ep);
3539 xfs_bmbt_get_all(ep - 1, prevp);
3543 got.br_startblock = xfs_bmbt_get_startblock(ep);
3544 got.br_state = xfs_bmbt_get_state(ep);
3552 * Search the extents list for the inode, for the extent containing bno.
3553 * If bno lies in a hole, point to the next entry. If bno lies past eof,
3554 * *eofp will be set, and *prevp will contain the last entry (null if none).
3555 * Else, *lastxp will be set to the index of the found
3556 * entry; *gotp will contain the entry.
3558 STATIC xfs_bmbt_rec_t * /* pointer to found extent entry */
3559 xfs_bmap_search_extents(
3560 xfs_inode_t *ip, /* incore inode pointer */
3561 xfs_fileoff_t bno, /* block number searched for */
3562 int whichfork, /* data or attr fork */
3563 int *eofp, /* out: end of file found */
3564 xfs_extnum_t *lastxp, /* out: last extent index */
3565 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
3566 xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
3568 xfs_ifork_t *ifp; /* inode fork pointer */
3569 xfs_bmbt_rec_t *base; /* base of extent list */
3570 xfs_extnum_t lastx; /* last extent index used */
3571 xfs_extnum_t nextents; /* extent list size */
3572 xfs_bmbt_rec_t *ep; /* extent list entry pointer */
3573 int rt; /* realtime flag */
3575 XFS_STATS_INC(xs_look_exlist);
3576 ifp = XFS_IFORK_PTR(ip, whichfork);
3577 lastx = ifp->if_lastex;
3578 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3579 base = &ifp->if_u1.if_extents[0];
3581 ep = xfs_bmap_do_search_extents(base, lastx, nextents, bno, eofp,
3582 lastxp, gotp, prevp);
3583 rt = ip->i_d.di_flags & XFS_DIFLAG_REALTIME;
3584 if(!rt && !gotp->br_startblock && (*lastxp != NULLEXTNUM)) {
3585 cmn_err(CE_PANIC,"Access to block zero: fs: <%s> inode: %lld "
3586 "start_block : %llx start_off : %llx blkcnt : %llx "
3587 "extent-state : %x \n",
3588 (ip->i_mount)->m_fsname,(long long)ip->i_ino,
3589 gotp->br_startblock, gotp->br_startoff,
3590 gotp->br_blockcount,gotp->br_state);
3596 #ifdef XFS_BMAP_TRACE
3597 ktrace_t *xfs_bmap_trace_buf;
3600 * Add a bmap trace buffer entry. Base routine for the others.
3603 xfs_bmap_trace_addentry(
3604 int opcode, /* operation */
3605 char *fname, /* function name */
3606 char *desc, /* operation description */
3607 xfs_inode_t *ip, /* incore inode pointer */
3608 xfs_extnum_t idx, /* index of entry(ies) */
3609 xfs_extnum_t cnt, /* count of entries, 1 or 2 */
3610 xfs_bmbt_rec_t *r1, /* first record */
3611 xfs_bmbt_rec_t *r2, /* second record or null */
3612 int whichfork) /* data or attr fork */
3616 ASSERT(cnt == 1 || cnt == 2);
3621 memset(&tr2, 0, sizeof(tr2));
3624 ktrace_enter(xfs_bmap_trace_buf,
3625 (void *)(__psint_t)(opcode | (whichfork << 16)),
3626 (void *)fname, (void *)desc, (void *)ip,
3627 (void *)(__psint_t)idx,
3628 (void *)(__psint_t)cnt,
3629 (void *)(__psunsigned_t)(ip->i_ino >> 32),
3630 (void *)(__psunsigned_t)(unsigned)ip->i_ino,
3631 (void *)(__psunsigned_t)(r1->l0 >> 32),
3632 (void *)(__psunsigned_t)(unsigned)(r1->l0),
3633 (void *)(__psunsigned_t)(r1->l1 >> 32),
3634 (void *)(__psunsigned_t)(unsigned)(r1->l1),
3635 (void *)(__psunsigned_t)(r2->l0 >> 32),
3636 (void *)(__psunsigned_t)(unsigned)(r2->l0),
3637 (void *)(__psunsigned_t)(r2->l1 >> 32),
3638 (void *)(__psunsigned_t)(unsigned)(r2->l1)
3640 ASSERT(ip->i_xtrace);
3641 ktrace_enter(ip->i_xtrace,
3642 (void *)(__psint_t)(opcode | (whichfork << 16)),
3643 (void *)fname, (void *)desc, (void *)ip,
3644 (void *)(__psint_t)idx,
3645 (void *)(__psint_t)cnt,
3646 (void *)(__psunsigned_t)(ip->i_ino >> 32),
3647 (void *)(__psunsigned_t)(unsigned)ip->i_ino,
3648 (void *)(__psunsigned_t)(r1->l0 >> 32),
3649 (void *)(__psunsigned_t)(unsigned)(r1->l0),
3650 (void *)(__psunsigned_t)(r1->l1 >> 32),
3651 (void *)(__psunsigned_t)(unsigned)(r1->l1),
3652 (void *)(__psunsigned_t)(r2->l0 >> 32),
3653 (void *)(__psunsigned_t)(unsigned)(r2->l0),
3654 (void *)(__psunsigned_t)(r2->l1 >> 32),
3655 (void *)(__psunsigned_t)(unsigned)(r2->l1)
3660 * Add bmap trace entry prior to a call to xfs_bmap_delete_exlist.
3663 xfs_bmap_trace_delete(
3664 char *fname, /* function name */
3665 char *desc, /* operation description */
3666 xfs_inode_t *ip, /* incore inode pointer */
3667 xfs_extnum_t idx, /* index of entry(entries) deleted */
3668 xfs_extnum_t cnt, /* count of entries deleted, 1 or 2 */
3669 int whichfork) /* data or attr fork */
3671 xfs_ifork_t *ifp; /* inode fork pointer */
3673 ifp = XFS_IFORK_PTR(ip, whichfork);
3674 xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_DELETE, fname, desc, ip, idx,
3675 cnt, &ifp->if_u1.if_extents[idx],
3676 cnt == 2 ? &ifp->if_u1.if_extents[idx + 1] : NULL,
3681 * Add bmap trace entry prior to a call to xfs_bmap_insert_exlist, or
3682 * reading in the extents list from the disk (in the btree).
3685 xfs_bmap_trace_insert(
3686 char *fname, /* function name */
3687 char *desc, /* operation description */
3688 xfs_inode_t *ip, /* incore inode pointer */
3689 xfs_extnum_t idx, /* index of entry(entries) inserted */
3690 xfs_extnum_t cnt, /* count of entries inserted, 1 or 2 */
3691 xfs_bmbt_irec_t *r1, /* inserted record 1 */
3692 xfs_bmbt_irec_t *r2, /* inserted record 2 or null */
3693 int whichfork) /* data or attr fork */
3695 xfs_bmbt_rec_t tr1; /* compressed record 1 */
3696 xfs_bmbt_rec_t tr2; /* compressed record 2 if needed */
3698 xfs_bmbt_set_all(&tr1, r1);
3701 xfs_bmbt_set_all(&tr2, r2);
3706 xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_INSERT, fname, desc, ip, idx,
3707 cnt, &tr1, cnt == 2 ? &tr2 : NULL, whichfork);
3711 * Add bmap trace entry after updating an extent list entry in place.
3714 xfs_bmap_trace_post_update(
3715 char *fname, /* function name */
3716 char *desc, /* operation description */
3717 xfs_inode_t *ip, /* incore inode pointer */
3718 xfs_extnum_t idx, /* index of entry updated */
3719 int whichfork) /* data or attr fork */
3721 xfs_ifork_t *ifp; /* inode fork pointer */
3723 ifp = XFS_IFORK_PTR(ip, whichfork);
3724 xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_POST_UP, fname, desc, ip, idx,
3725 1, &ifp->if_u1.if_extents[idx], NULL, whichfork);
3729 * Add bmap trace entry prior to updating an extent list entry in place.
3732 xfs_bmap_trace_pre_update(
3733 char *fname, /* function name */
3734 char *desc, /* operation description */
3735 xfs_inode_t *ip, /* incore inode pointer */
3736 xfs_extnum_t idx, /* index of entry to be updated */
3737 int whichfork) /* data or attr fork */
3739 xfs_ifork_t *ifp; /* inode fork pointer */
3741 ifp = XFS_IFORK_PTR(ip, whichfork);
3742 xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_PRE_UP, fname, desc, ip, idx, 1,
3743 &ifp->if_u1.if_extents[idx], NULL, whichfork);
3745 #endif /* XFS_BMAP_TRACE */
3748 * Compute the worst-case number of indirect blocks that will be used
3749 * for ip's delayed extent of length "len".
3751 STATIC xfs_filblks_t
3752 xfs_bmap_worst_indlen(
3753 xfs_inode_t *ip, /* incore inode pointer */
3754 xfs_filblks_t len) /* delayed extent length */
3756 int level; /* btree level number */
3757 int maxrecs; /* maximum record count at this level */
3758 xfs_mount_t *mp; /* mount structure */
3759 xfs_filblks_t rval; /* return value */
3762 maxrecs = mp->m_bmap_dmxr[0];
3763 for (level = 0, rval = 0;
3764 level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3767 do_div(len, maxrecs);
3770 return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3773 maxrecs = mp->m_bmap_dmxr[1];
3778 #if defined(XFS_RW_TRACE)
3787 if (ip->i_rwtrace == NULL)
3789 ktrace_enter(ip->i_rwtrace,
3790 (void *)(__psint_t)XFS_BUNMAPI,
3792 (void *)(__psint_t)((ip->i_d.di_size >> 32) & 0xffffffff),
3793 (void *)(__psint_t)(ip->i_d.di_size & 0xffffffff),
3794 (void *)(__psint_t)(((xfs_dfiloff_t)bno >> 32) & 0xffffffff),
3795 (void *)(__psint_t)((xfs_dfiloff_t)bno & 0xffffffff),
3796 (void *)(__psint_t)len,
3797 (void *)(__psint_t)flags,
3798 (void *)(unsigned long)current_cpu(),
3810 * Convert inode from non-attributed to attributed.
3811 * Must not be in a transaction, ip must not be locked.
3813 int /* error code */
3814 xfs_bmap_add_attrfork(
3815 xfs_inode_t *ip, /* incore inode pointer */
3816 int size, /* space new attribute needs */
3817 int rsvd) /* xact may use reserved blks */
3819 xfs_fsblock_t firstblock; /* 1st block/ag allocated */
3820 xfs_bmap_free_t flist; /* freed extent list */
3821 xfs_mount_t *mp; /* mount structure */
3822 xfs_trans_t *tp; /* transaction pointer */
3823 unsigned long s; /* spinlock spl value */
3824 int blks; /* space reservation */
3825 int version = 1; /* superblock attr version */
3826 int committed; /* xaction was committed */
3827 int logflags; /* logging flags */
3828 int error; /* error return value */
3830 ASSERT(XFS_IFORK_Q(ip) == 0);
3831 ASSERT(ip->i_df.if_ext_max ==
3832 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3835 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3836 tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3837 blks = XFS_ADDAFORK_SPACE_RES(mp);
3839 tp->t_flags |= XFS_TRANS_RESERVE;
3840 if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3841 XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3843 xfs_ilock(ip, XFS_ILOCK_EXCL);
3844 error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip, blks, 0, rsvd ?
3845 XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3846 XFS_QMOPT_RES_REGBLKS);
3848 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3849 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3852 if (XFS_IFORK_Q(ip))
3854 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3856 * For inodes coming from pre-6.2 filesystems.
3858 ASSERT(ip->i_d.di_aformat == 0);
3859 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
3861 ASSERT(ip->i_d.di_anextents == 0);
3862 VN_HOLD(XFS_ITOV(ip));
3863 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3864 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3865 switch (ip->i_d.di_format) {
3866 case XFS_DINODE_FMT_DEV:
3867 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
3869 case XFS_DINODE_FMT_UUID:
3870 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
3872 case XFS_DINODE_FMT_LOCAL:
3873 case XFS_DINODE_FMT_EXTENTS:
3874 case XFS_DINODE_FMT_BTREE:
3875 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
3876 if (!ip->i_d.di_forkoff)
3877 ip->i_d.di_forkoff = mp->m_attroffset >> 3;
3878 else if (!(mp->m_flags & XFS_MOUNT_COMPAT_ATTR))
3883 error = XFS_ERROR(EINVAL);
3886 ip->i_df.if_ext_max =
3887 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3888 ASSERT(ip->i_afp == NULL);
3889 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
3890 ip->i_afp->if_ext_max =
3891 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3892 ip->i_afp->if_flags = XFS_IFEXTENTS;
3894 XFS_BMAP_INIT(&flist, &firstblock);
3895 switch (ip->i_d.di_format) {
3896 case XFS_DINODE_FMT_LOCAL:
3897 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
3900 case XFS_DINODE_FMT_EXTENTS:
3901 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
3904 case XFS_DINODE_FMT_BTREE:
3905 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
3913 xfs_trans_log_inode(tp, ip, logflags);
3916 if (!XFS_SB_VERSION_HASATTR(&mp->m_sb) ||
3917 (!XFS_SB_VERSION_HASATTR2(&mp->m_sb) && version == 2)) {
3918 __int64_t sbfields = 0;
3920 s = XFS_SB_LOCK(mp);
3921 if (!XFS_SB_VERSION_HASATTR(&mp->m_sb)) {
3922 XFS_SB_VERSION_ADDATTR(&mp->m_sb);
3923 sbfields |= XFS_SB_VERSIONNUM;
3925 if (!XFS_SB_VERSION_HASATTR2(&mp->m_sb) && version == 2) {
3926 XFS_SB_VERSION_ADDATTR2(&mp->m_sb);
3927 sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
3930 XFS_SB_UNLOCK(mp, s);
3931 xfs_mod_sb(tp, sbfields);
3933 XFS_SB_UNLOCK(mp, s);
3935 if ((error = xfs_bmap_finish(&tp, &flist, firstblock, &committed)))
3937 error = xfs_trans_commit(tp, XFS_TRANS_PERM_LOG_RES, NULL);
3938 ASSERT(ip->i_df.if_ext_max ==
3939 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3942 xfs_bmap_cancel(&flist);
3944 ASSERT(ismrlocked(&ip->i_lock,MR_UPDATE));
3945 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3947 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
3948 ASSERT(ip->i_df.if_ext_max ==
3949 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3954 * Add the extent to the list of extents to be free at transaction end.
3955 * The list is maintained sorted (by block number).
3960 xfs_fsblock_t bno, /* fs block number of extent */
3961 xfs_filblks_t len, /* length of extent */
3962 xfs_bmap_free_t *flist, /* list of extents */
3963 xfs_mount_t *mp) /* mount point structure */
3965 xfs_bmap_free_item_t *cur; /* current (next) element */
3966 xfs_bmap_free_item_t *new; /* new element */
3967 xfs_bmap_free_item_t *prev; /* previous element */
3969 xfs_agnumber_t agno;
3970 xfs_agblock_t agbno;
3972 ASSERT(bno != NULLFSBLOCK);
3974 ASSERT(len <= MAXEXTLEN);
3975 ASSERT(!ISNULLSTARTBLOCK(bno));
3976 agno = XFS_FSB_TO_AGNO(mp, bno);
3977 agbno = XFS_FSB_TO_AGBNO(mp, bno);
3978 ASSERT(agno < mp->m_sb.sb_agcount);
3979 ASSERT(agbno < mp->m_sb.sb_agblocks);
3980 ASSERT(len < mp->m_sb.sb_agblocks);
3981 ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
3983 ASSERT(xfs_bmap_free_item_zone != NULL);
3984 new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
3985 new->xbfi_startblock = bno;
3986 new->xbfi_blockcount = (xfs_extlen_t)len;
3987 for (prev = NULL, cur = flist->xbf_first;
3989 prev = cur, cur = cur->xbfi_next) {
3990 if (cur->xbfi_startblock >= bno)
3994 prev->xbfi_next = new;
3996 flist->xbf_first = new;
3997 new->xbfi_next = cur;
4002 * Compute and fill in the value of the maximum depth of a bmap btree
4003 * in this filesystem. Done once, during mount.
4006 xfs_bmap_compute_maxlevels(
4007 xfs_mount_t *mp, /* file system mount structure */
4008 int whichfork) /* data or attr fork */
4010 int level; /* btree level */
4011 uint maxblocks; /* max blocks at this level */
4012 uint maxleafents; /* max leaf entries possible */
4013 int maxrootrecs; /* max records in root block */
4014 int minleafrecs; /* min records in leaf block */
4015 int minnoderecs; /* min records in node block */
4016 int sz; /* root block size */
4019 * The maximum number of extents in a file, hence the maximum
4020 * number of leaf entries, is controlled by the type of di_nextents
4021 * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
4022 * (a signed 16-bit number, xfs_aextnum_t).
4024 if (whichfork == XFS_DATA_FORK) {
4025 maxleafents = MAXEXTNUM;
4026 sz = (mp->m_flags & XFS_MOUNT_COMPAT_ATTR) ?
4027 mp->m_attroffset : XFS_BMDR_SPACE_CALC(MINDBTPTRS);
4029 maxleafents = MAXAEXTNUM;
4030 sz = (mp->m_flags & XFS_MOUNT_COMPAT_ATTR) ?
4031 mp->m_sb.sb_inodesize - mp->m_attroffset :
4032 XFS_BMDR_SPACE_CALC(MINABTPTRS);
4034 maxrootrecs = (int)XFS_BTREE_BLOCK_MAXRECS(sz, xfs_bmdr, 0);
4035 minleafrecs = mp->m_bmap_dmnr[0];
4036 minnoderecs = mp->m_bmap_dmnr[1];
4037 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
4038 for (level = 1; maxblocks > 1; level++) {
4039 if (maxblocks <= maxrootrecs)
4042 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
4044 mp->m_bm_maxlevels[whichfork] = level;
4048 * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
4049 * caller. Frees all the extents that need freeing, which must be done
4050 * last due to locking considerations. We never free any extents in
4051 * the first transaction. This is to allow the caller to make the first
4052 * transaction a synchronous one so that the pointers to the data being
4053 * broken in this transaction will be permanent before the data is actually
4054 * freed. This is necessary to prevent blocks from being reallocated
4055 * and written to before the free and reallocation are actually permanent.
4056 * We do not just make the first transaction synchronous here, because
4057 * there are more efficient ways to gain the same protection in some cases
4058 * (see the file truncation code).
4060 * Return 1 if the given transaction was committed and a new one
4061 * started, and 0 otherwise in the committed parameter.
4066 xfs_trans_t **tp, /* transaction pointer addr */
4067 xfs_bmap_free_t *flist, /* i/o: list extents to free */
4068 xfs_fsblock_t firstblock, /* controlled ag for allocs */
4069 int *committed) /* xact committed or not */
4071 xfs_efd_log_item_t *efd; /* extent free data */
4072 xfs_efi_log_item_t *efi; /* extent free intention */
4073 int error; /* error return value */
4074 xfs_bmap_free_item_t *free; /* free extent list item */
4075 unsigned int logres; /* new log reservation */
4076 unsigned int logcount; /* new log count */
4077 xfs_mount_t *mp; /* filesystem mount structure */
4078 xfs_bmap_free_item_t *next; /* next item on free list */
4079 xfs_trans_t *ntp; /* new transaction pointer */
4081 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
4082 if (flist->xbf_count == 0) {
4087 efi = xfs_trans_get_efi(ntp, flist->xbf_count);
4088 for (free = flist->xbf_first; free; free = free->xbfi_next)
4089 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
4090 free->xbfi_blockcount);
4091 logres = ntp->t_log_res;
4092 logcount = ntp->t_log_count;
4093 ntp = xfs_trans_dup(*tp);
4094 error = xfs_trans_commit(*tp, 0, NULL);
4098 * We have a new transaction, so we should return committed=1,
4099 * even though we're returning an error.
4104 if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
4107 efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
4108 for (free = flist->xbf_first; free != NULL; free = next) {
4109 next = free->xbfi_next;
4110 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
4111 free->xbfi_blockcount))) {
4113 * The bmap free list will be cleaned up at a
4114 * higher level. The EFI will be canceled when
4115 * this transaction is aborted.
4116 * Need to force shutdown here to make sure it
4117 * happens, since this transaction may not be
4121 if (!XFS_FORCED_SHUTDOWN(mp))
4122 xfs_force_shutdown(mp,
4123 (error == EFSCORRUPTED) ?
4124 XFS_CORRUPT_INCORE :
4125 XFS_METADATA_IO_ERROR);
4128 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
4129 free->xbfi_blockcount);
4130 xfs_bmap_del_free(flist, NULL, free);
4136 * Free up any items left in the list.
4140 xfs_bmap_free_t *flist) /* list of bmap_free_items */
4142 xfs_bmap_free_item_t *free; /* free list item */
4143 xfs_bmap_free_item_t *next;
4145 if (flist->xbf_count == 0)
4147 ASSERT(flist->xbf_first != NULL);
4148 for (free = flist->xbf_first; free; free = next) {
4149 next = free->xbfi_next;
4150 xfs_bmap_del_free(flist, NULL, free);
4152 ASSERT(flist->xbf_count == 0);
4156 * Returns the file-relative block number of the first unused block(s)
4157 * in the file with at least "len" logically contiguous blocks free.
4158 * This is the lowest-address hole if the file has holes, else the first block
4159 * past the end of file.
4160 * Return 0 if the file is currently local (in-inode).
4163 xfs_bmap_first_unused(
4164 xfs_trans_t *tp, /* transaction pointer */
4165 xfs_inode_t *ip, /* incore inode */
4166 xfs_extlen_t len, /* size of hole to find */
4167 xfs_fileoff_t *first_unused, /* unused block */
4168 int whichfork) /* data or attr fork */
4170 xfs_bmbt_rec_t *base; /* base of extent array */
4171 xfs_bmbt_rec_t *ep; /* pointer to an extent entry */
4172 int error; /* error return value */
4173 xfs_ifork_t *ifp; /* inode fork pointer */
4174 xfs_fileoff_t lastaddr; /* last block number seen */
4175 xfs_fileoff_t lowest; /* lowest useful block */
4176 xfs_fileoff_t max; /* starting useful block */
4177 xfs_fileoff_t off; /* offset for this block */
4178 xfs_extnum_t nextents; /* number of extent entries */
4180 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
4181 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
4182 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
4183 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4187 ifp = XFS_IFORK_PTR(ip, whichfork);
4188 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4189 (error = xfs_iread_extents(tp, ip, whichfork)))
4191 lowest = *first_unused;
4192 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4193 base = &ifp->if_u1.if_extents[0];
4194 for (lastaddr = 0, max = lowest, ep = base;
4195 ep < &base[nextents];
4197 off = xfs_bmbt_get_startoff(ep);
4199 * See if the hole before this extent will work.
4201 if (off >= lowest + len && off - max >= len) {
4202 *first_unused = max;
4205 lastaddr = off + xfs_bmbt_get_blockcount(ep);
4206 max = XFS_FILEOFF_MAX(lastaddr, lowest);
4208 *first_unused = max;
4213 * Returns the file-relative block number of the last block + 1 before
4214 * last_block (input value) in the file.
4215 * This is not based on i_size, it is based on the extent list.
4216 * Returns 0 for local files, as they do not have an extent list.
4219 xfs_bmap_last_before(
4220 xfs_trans_t *tp, /* transaction pointer */
4221 xfs_inode_t *ip, /* incore inode */
4222 xfs_fileoff_t *last_block, /* last block */
4223 int whichfork) /* data or attr fork */
4225 xfs_fileoff_t bno; /* input file offset */
4226 int eof; /* hit end of file */
4227 xfs_bmbt_rec_t *ep; /* pointer to last extent */
4228 int error; /* error return value */
4229 xfs_bmbt_irec_t got; /* current extent value */
4230 xfs_ifork_t *ifp; /* inode fork pointer */
4231 xfs_extnum_t lastx; /* last extent used */
4232 xfs_bmbt_irec_t prev; /* previous extent value */
4234 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4235 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4236 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4237 return XFS_ERROR(EIO);
4238 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4242 ifp = XFS_IFORK_PTR(ip, whichfork);
4243 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4244 (error = xfs_iread_extents(tp, ip, whichfork)))
4246 bno = *last_block - 1;
4247 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4249 if (eof || xfs_bmbt_get_startoff(ep) > bno) {
4250 if (prev.br_startoff == NULLFILEOFF)
4253 *last_block = prev.br_startoff + prev.br_blockcount;
4256 * Otherwise *last_block is already the right answer.
4262 * Returns the file-relative block number of the first block past eof in
4263 * the file. This is not based on i_size, it is based on the extent list.
4264 * Returns 0 for local files, as they do not have an extent list.
4267 xfs_bmap_last_offset(
4268 xfs_trans_t *tp, /* transaction pointer */
4269 xfs_inode_t *ip, /* incore inode */
4270 xfs_fileoff_t *last_block, /* last block */
4271 int whichfork) /* data or attr fork */
4273 xfs_bmbt_rec_t *base; /* base of extent array */
4274 xfs_bmbt_rec_t *ep; /* pointer to last extent */
4275 int error; /* error return value */
4276 xfs_ifork_t *ifp; /* inode fork pointer */
4277 xfs_extnum_t nextents; /* number of extent entries */
4279 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4280 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4281 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4282 return XFS_ERROR(EIO);
4283 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4287 ifp = XFS_IFORK_PTR(ip, whichfork);
4288 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4289 (error = xfs_iread_extents(tp, ip, whichfork)))
4291 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4296 base = &ifp->if_u1.if_extents[0];
4297 ASSERT(base != NULL);
4298 ep = &base[nextents - 1];
4299 *last_block = xfs_bmbt_get_startoff(ep) + xfs_bmbt_get_blockcount(ep);
4304 * Returns whether the selected fork of the inode has exactly one
4305 * block or not. For the data fork we check this matches di_size,
4306 * implying the file's range is 0..bsize-1.
4308 int /* 1=>1 block, 0=>otherwise */
4310 xfs_inode_t *ip, /* incore inode */
4311 int whichfork) /* data or attr fork */
4313 xfs_bmbt_rec_t *ep; /* ptr to fork's extent */
4314 xfs_ifork_t *ifp; /* inode fork pointer */
4315 int rval; /* return value */
4316 xfs_bmbt_irec_t s; /* internal version of extent */
4319 if (whichfork == XFS_DATA_FORK)
4320 return ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize;
4322 if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4324 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4326 ifp = XFS_IFORK_PTR(ip, whichfork);
4327 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4328 ep = ifp->if_u1.if_extents;
4329 xfs_bmbt_get_all(ep, &s);
4330 rval = s.br_startoff == 0 && s.br_blockcount == 1;
4331 if (rval && whichfork == XFS_DATA_FORK)
4332 ASSERT(ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize);
4337 * Read in the extents to if_extents.
4338 * All inode fields are set up by caller, we just traverse the btree
4339 * and copy the records in. If the file system cannot contain unwritten
4340 * extents, the records are checked for no "state" flags.
4343 xfs_bmap_read_extents(
4344 xfs_trans_t *tp, /* transaction pointer */
4345 xfs_inode_t *ip, /* incore inode */
4346 int whichfork) /* data or attr fork */
4348 xfs_bmbt_block_t *block; /* current btree block */
4349 xfs_fsblock_t bno; /* block # of "block" */
4350 xfs_buf_t *bp; /* buffer for "block" */
4351 int error; /* error return value */
4352 xfs_exntfmt_t exntf; /* XFS_EXTFMT_NOSTATE, if checking */
4353 #ifdef XFS_BMAP_TRACE
4354 static char fname[] = "xfs_bmap_read_extents";
4356 xfs_extnum_t i, j; /* index into the extents list */
4357 xfs_ifork_t *ifp; /* fork structure */
4358 int level; /* btree level, for checking */
4359 xfs_mount_t *mp; /* file system mount structure */
4360 xfs_bmbt_ptr_t *pp; /* pointer to block address */
4362 xfs_extnum_t room; /* number of entries there's room for */
4363 xfs_bmbt_rec_t *trp; /* target record pointer */
4367 ifp = XFS_IFORK_PTR(ip, whichfork);
4368 exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4369 XFS_EXTFMT_INODE(ip);
4370 block = ifp->if_broot;
4372 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4374 level = be16_to_cpu(block->bb_level);
4376 pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
4377 ASSERT(INT_GET(*pp, ARCH_CONVERT) != NULLDFSBNO);
4378 ASSERT(XFS_FSB_TO_AGNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agcount);
4379 ASSERT(XFS_FSB_TO_AGBNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agblocks);
4380 bno = INT_GET(*pp, ARCH_CONVERT);
4382 * Go down the tree until leaf level is reached, following the first
4383 * pointer (leftmost) at each level.
4385 while (level-- > 0) {
4386 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4387 XFS_BMAP_BTREE_REF)))
4389 block = XFS_BUF_TO_BMBT_BLOCK(bp);
4390 XFS_WANT_CORRUPTED_GOTO(
4391 XFS_BMAP_SANITY_CHECK(mp, block, level),
4395 pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt, block,
4396 1, mp->m_bmap_dmxr[1]);
4397 XFS_WANT_CORRUPTED_GOTO(
4398 XFS_FSB_SANITY_CHECK(mp, INT_GET(*pp, ARCH_CONVERT)),
4400 bno = INT_GET(*pp, ARCH_CONVERT);
4401 xfs_trans_brelse(tp, bp);
4404 * Here with bp and block set to the leftmost leaf node in the tree.
4406 room = ifp->if_bytes / (uint)sizeof(*trp);
4407 trp = ifp->if_u1.if_extents;
4410 * Loop over all leaf nodes. Copy information to the extent list.
4413 xfs_bmbt_rec_t *frp, *temp;
4414 xfs_fsblock_t nextbno;
4415 xfs_extnum_t num_recs;
4418 num_recs = be16_to_cpu(block->bb_numrecs);
4419 if (unlikely(i + num_recs > room)) {
4420 ASSERT(i + num_recs <= room);
4421 xfs_fs_cmn_err(CE_WARN, ip->i_mount,
4422 "corrupt dinode %Lu, (btree extents). Unmount and run xfs_repair.",
4423 (unsigned long long) ip->i_ino);
4424 XFS_ERROR_REPORT("xfs_bmap_read_extents(1)",
4429 XFS_WANT_CORRUPTED_GOTO(
4430 XFS_BMAP_SANITY_CHECK(mp, block, 0),
4433 * Read-ahead the next leaf block, if any.
4435 nextbno = be64_to_cpu(block->bb_rightsib);
4436 if (nextbno != NULLFSBLOCK)
4437 xfs_btree_reada_bufl(mp, nextbno, 1);
4439 * Copy records into the extent list.
4441 frp = XFS_BTREE_REC_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt,
4442 block, 1, mp->m_bmap_dmxr[0]);
4444 for (j = 0; j < num_recs; j++, frp++, trp++) {
4445 trp->l0 = INT_GET(frp->l0, ARCH_CONVERT);
4446 trp->l1 = INT_GET(frp->l1, ARCH_CONVERT);
4448 if (exntf == XFS_EXTFMT_NOSTATE) {
4450 * Check all attribute bmap btree records and
4451 * any "older" data bmap btree records for a
4452 * set bit in the "extent flag" position.
4454 if (unlikely(xfs_check_nostate_extents(temp, num_recs))) {
4455 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4462 xfs_trans_brelse(tp, bp);
4465 * If we've reached the end, stop.
4467 if (bno == NULLFSBLOCK)
4469 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4470 XFS_BMAP_BTREE_REF)))
4472 block = XFS_BUF_TO_BMBT_BLOCK(bp);
4474 ASSERT(i == ifp->if_bytes / (uint)sizeof(*trp));
4475 ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4476 xfs_bmap_trace_exlist(fname, ip, i, whichfork);
4479 xfs_trans_brelse(tp, bp);
4480 return XFS_ERROR(EFSCORRUPTED);
4483 #ifdef XFS_BMAP_TRACE
4485 * Add bmap trace insert entries for all the contents of the extent list.
4488 xfs_bmap_trace_exlist(
4489 char *fname, /* function name */
4490 xfs_inode_t *ip, /* incore inode pointer */
4491 xfs_extnum_t cnt, /* count of entries in the list */
4492 int whichfork) /* data or attr fork */
4494 xfs_bmbt_rec_t *base; /* base of extent list */
4495 xfs_bmbt_rec_t *ep; /* current entry in extent list */
4496 xfs_extnum_t idx; /* extent list entry number */
4497 xfs_ifork_t *ifp; /* inode fork pointer */
4498 xfs_bmbt_irec_t s; /* extent list record */
4500 ifp = XFS_IFORK_PTR(ip, whichfork);
4501 ASSERT(cnt == ifp->if_bytes / (uint)sizeof(*base));
4502 base = ifp->if_u1.if_extents;
4503 for (idx = 0, ep = base; idx < cnt; idx++, ep++) {
4504 xfs_bmbt_get_all(ep, &s);
4505 xfs_bmap_trace_insert(fname, "exlist", ip, idx, 1, &s, NULL,
4513 * Validate that the bmbt_irecs being returned from bmapi are valid
4514 * given the callers original parameters. Specifically check the
4515 * ranges of the returned irecs to ensure that they only extent beyond
4516 * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4519 xfs_bmap_validate_ret(
4523 xfs_bmbt_irec_t *mval,
4527 int i; /* index to map values */
4529 ASSERT(ret_nmap <= nmap);
4531 for (i = 0; i < ret_nmap; i++) {
4532 ASSERT(mval[i].br_blockcount > 0);
4533 if (!(flags & XFS_BMAPI_ENTIRE)) {
4534 ASSERT(mval[i].br_startoff >= bno);
4535 ASSERT(mval[i].br_blockcount <= len);
4536 ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4539 ASSERT(mval[i].br_startoff < bno + len);
4540 ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4544 mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4545 mval[i].br_startoff);
4546 if ((flags & XFS_BMAPI_WRITE) && !(flags & XFS_BMAPI_DELAY))
4547 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4548 mval[i].br_startblock != HOLESTARTBLOCK);
4549 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4550 mval[i].br_state == XFS_EXT_UNWRITTEN);
4557 * Map file blocks to filesystem blocks.
4558 * File range is given by the bno/len pair.
4559 * Adds blocks to file if a write ("flags & XFS_BMAPI_WRITE" set)
4560 * into a hole or past eof.
4561 * Only allocates blocks from a single allocation group,
4562 * to avoid locking problems.
4563 * The returned value in "firstblock" from the first call in a transaction
4564 * must be remembered and presented to subsequent calls in "firstblock".
4565 * An upper bound for the number of blocks to be allocated is supplied to
4566 * the first call in "total"; if no allocation group has that many free
4567 * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4571 xfs_trans_t *tp, /* transaction pointer */
4572 xfs_inode_t *ip, /* incore inode */
4573 xfs_fileoff_t bno, /* starting file offs. mapped */
4574 xfs_filblks_t len, /* length to map in file */
4575 int flags, /* XFS_BMAPI_... */
4576 xfs_fsblock_t *firstblock, /* first allocated block
4577 controls a.g. for allocs */
4578 xfs_extlen_t total, /* total blocks needed */
4579 xfs_bmbt_irec_t *mval, /* output: map values */
4580 int *nmap, /* i/o: mval size/count */
4581 xfs_bmap_free_t *flist) /* i/o: list extents to free */
4583 xfs_fsblock_t abno; /* allocated block number */
4584 xfs_extlen_t alen; /* allocated extent length */
4585 xfs_fileoff_t aoff; /* allocated file offset */
4586 xfs_bmalloca_t bma; /* args for xfs_bmap_alloc */
4587 xfs_btree_cur_t *cur; /* bmap btree cursor */
4588 xfs_fileoff_t end; /* end of mapped file region */
4589 int eof; /* we've hit the end of extent list */
4590 char contig; /* allocation must be one extent */
4591 char delay; /* this request is for delayed alloc */
4592 char exact; /* don't do all of wasdelayed extent */
4593 xfs_bmbt_rec_t *ep; /* extent list entry pointer */
4594 int error; /* error return */
4595 xfs_bmbt_irec_t got; /* current extent list record */
4596 xfs_ifork_t *ifp; /* inode fork pointer */
4597 xfs_extlen_t indlen; /* indirect blocks length */
4598 xfs_extnum_t lastx; /* last useful extent number */
4599 int logflags; /* flags for transaction logging */
4600 xfs_extlen_t minleft; /* min blocks left after allocation */
4601 xfs_extlen_t minlen; /* min allocation size */
4602 xfs_mount_t *mp; /* xfs mount structure */
4603 int n; /* current extent index */
4604 int nallocs; /* number of extents alloc\'d */
4605 xfs_extnum_t nextents; /* number of extents in file */
4606 xfs_fileoff_t obno; /* old block number (offset) */
4607 xfs_bmbt_irec_t prev; /* previous extent list record */
4608 int tmp_logflags; /* temp flags holder */
4609 int whichfork; /* data or attr fork */
4610 char inhole; /* current location is hole in file */
4611 char stateless; /* ignore state flag set */
4612 char trim; /* output trimmed to match range */
4613 char userdata; /* allocating non-metadata */
4614 char wasdelay; /* old extent was delayed */
4615 char wr; /* this is a write request */
4616 char rt; /* this is a realtime file */
4617 char rsvd; /* OK to allocate reserved blocks */
4619 xfs_fileoff_t orig_bno; /* original block number value */
4620 int orig_flags; /* original flags arg value */
4621 xfs_filblks_t orig_len; /* original value of len arg */
4622 xfs_bmbt_irec_t *orig_mval; /* original value of mval */
4623 int orig_nmap; /* original value of *nmap */
4632 ASSERT(*nmap <= XFS_BMAP_MAX_NMAP || !(flags & XFS_BMAPI_WRITE));
4633 whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4634 XFS_ATTR_FORK : XFS_DATA_FORK;
4636 if (unlikely(XFS_TEST_ERROR(
4637 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4638 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4639 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4640 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4641 XFS_ERROR_REPORT("xfs_bmapi", XFS_ERRLEVEL_LOW, mp);
4642 return XFS_ERROR(EFSCORRUPTED);
4644 if (XFS_FORCED_SHUTDOWN(mp))
4645 return XFS_ERROR(EIO);
4646 rt = XFS_IS_REALTIME_INODE(ip);
4647 ifp = XFS_IFORK_PTR(ip, whichfork);
4648 ASSERT(ifp->if_ext_max ==
4649 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4650 if ((wr = (flags & XFS_BMAPI_WRITE)) != 0)
4651 XFS_STATS_INC(xs_blk_mapw);
4653 XFS_STATS_INC(xs_blk_mapr);
4654 delay = (flags & XFS_BMAPI_DELAY) != 0;
4655 trim = (flags & XFS_BMAPI_ENTIRE) == 0;
4656 userdata = (flags & XFS_BMAPI_METADATA) == 0;
4657 exact = (flags & XFS_BMAPI_EXACT) != 0;
4658 rsvd = (flags & XFS_BMAPI_RSVBLOCKS) != 0;
4659 contig = (flags & XFS_BMAPI_CONTIG) != 0;
4661 * stateless is used to combine extents which
4662 * differ only due to the state of the extents.
4663 * This technique is used from xfs_getbmap()
4664 * when the caller does not wish to see the
4665 * separation (which is the default).
4667 * This technique is also used when writing a
4668 * buffer which has been partially written,
4669 * (usually by being flushed during a chunkread),
4670 * to ensure one write takes place. This also
4671 * prevents a change in the xfs inode extents at
4672 * this time, intentionally. This change occurs
4673 * on completion of the write operation, in
4674 * xfs_strat_comp(), where the xfs_bmapi() call
4675 * is transactioned, and the extents combined.
4677 stateless = (flags & XFS_BMAPI_IGSTATE) != 0;
4678 if (stateless && wr) /* if writing unwritten space, no */
4679 wr = 0; /* allocations are allowed */
4680 ASSERT(wr || !delay);
4684 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4686 if ((error = xfs_bmap_local_to_extents(tp, ip,
4687 firstblock, total, &logflags, whichfork)))
4690 if (wr && *firstblock == NULLFSBLOCK) {
4691 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4692 minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4697 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4698 (error = xfs_iread_extents(tp, ip, whichfork)))
4700 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4702 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4707 while (bno < end && n < *nmap) {
4709 * Reading past eof, act as though there's a hole
4713 got.br_startoff = end;
4714 inhole = eof || got.br_startoff > bno;
4715 wasdelay = wr && !inhole && !delay &&
4716 ISNULLSTARTBLOCK(got.br_startblock);
4718 * First, deal with the hole before the allocated space
4719 * that we found, if any.
4721 if (wr && (inhole || wasdelay)) {
4723 * For the wasdelay case, we could also just
4724 * allocate the stuff asked for in this bmap call
4725 * but that wouldn't be as good.
4727 if (wasdelay && !exact) {
4728 alen = (xfs_extlen_t)got.br_blockcount;
4729 aoff = got.br_startoff;
4730 if (lastx != NULLEXTNUM && lastx) {
4731 ep = &ifp->if_u1.if_extents[lastx - 1];
4732 xfs_bmbt_get_all(ep, &prev);
4734 } else if (wasdelay) {
4735 alen = (xfs_extlen_t)
4736 XFS_FILBLKS_MIN(len,
4738 got.br_blockcount) - bno);
4741 alen = (xfs_extlen_t)
4742 XFS_FILBLKS_MIN(len, MAXEXTLEN);
4744 alen = (xfs_extlen_t)
4745 XFS_FILBLKS_MIN(alen,
4746 got.br_startoff - bno);
4749 minlen = contig ? alen : 1;
4751 xfs_extlen_t extsz = 0;
4753 /* Figure out the extent size, adjust alen */
4755 if (!(extsz = ip->i_d.di_extsize))
4756 extsz = mp->m_sb.sb_rextsize;
4757 alen = roundup(alen, extsz);
4758 extsz = alen / mp->m_sb.sb_rextsize;
4762 * Make a transaction-less quota reservation for
4763 * delayed allocation blocks. This number gets
4765 * We return EDQUOT if we haven't allocated
4766 * blks already inside this loop;
4768 if (XFS_TRANS_RESERVE_QUOTA_NBLKS(
4769 mp, NULL, ip, (long)alen, 0,
4770 rt ? XFS_QMOPT_RES_RTBLKS :
4771 XFS_QMOPT_RES_REGBLKS)) {
4774 ASSERT(cur == NULL);
4775 return XFS_ERROR(EDQUOT);
4781 * Split changing sb for alen and indlen since
4782 * they could be coming from different places.
4784 indlen = (xfs_extlen_t)
4785 xfs_bmap_worst_indlen(ip, alen);
4789 error = xfs_mod_incore_sb(mp,
4793 error = xfs_mod_incore_sb(mp,
4797 error = xfs_mod_incore_sb(mp,
4801 xfs_mod_incore_sb(ip->i_mount,
4805 xfs_mod_incore_sb(ip->i_mount,
4812 if (XFS_IS_QUOTA_ON(ip->i_mount))
4813 /* unreserve the blocks now */
4814 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(
4817 XFS_QMOPT_RES_RTBLKS :
4818 XFS_QMOPT_RES_REGBLKS);
4822 ip->i_delayed_blks += alen;
4823 abno = NULLSTARTBLOCK(indlen);
4826 * If first time, allocate and fill in
4827 * once-only bma fields.
4829 if (bma.ip == NULL) {
4837 /* Indicate if this is the first user data
4838 * in the file, or just any user data.
4841 bma.userdata = (aoff == 0) ?
4842 XFS_ALLOC_INITIAL_USER_DATA :
4846 * Fill in changeable bma fields.
4849 bma.firstblock = *firstblock;
4852 bma.wasdel = wasdelay;
4853 bma.minlen = minlen;
4854 bma.low = flist->xbf_low;
4855 bma.minleft = minleft;
4857 * Only want to do the alignment at the
4858 * eof if it is userdata and allocation length
4859 * is larger than a stripe unit.
4861 if (mp->m_dalign && alen >= mp->m_dalign &&
4862 userdata && whichfork == XFS_DATA_FORK) {
4863 if ((error = xfs_bmap_isaeof(ip, aoff,
4864 whichfork, &bma.aeof)))
4871 if ((error = xfs_bmap_alloc(&bma)))
4874 * Copy out result fields.
4877 if ((flist->xbf_low = bma.low))
4881 ASSERT(*firstblock == NULLFSBLOCK ||
4882 XFS_FSB_TO_AGNO(mp, *firstblock) ==
4883 XFS_FSB_TO_AGNO(mp, bma.firstblock) ||
4885 XFS_FSB_TO_AGNO(mp, *firstblock) <
4886 XFS_FSB_TO_AGNO(mp, bma.firstblock)));
4887 *firstblock = bma.firstblock;
4889 cur->bc_private.b.firstblock =
4891 if (abno == NULLFSBLOCK)
4893 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
4894 cur = xfs_btree_init_cursor(mp,
4895 tp, NULL, 0, XFS_BTNUM_BMAP,
4897 cur->bc_private.b.firstblock =
4899 cur->bc_private.b.flist = flist;
4902 * Bump the number of extents we've allocated
4908 cur->bc_private.b.flags =
4909 wasdelay ? XFS_BTCUR_BPRV_WASDEL : 0;
4910 got.br_startoff = aoff;
4911 got.br_startblock = abno;
4912 got.br_blockcount = alen;
4913 got.br_state = XFS_EXT_NORM; /* assume normal */
4915 * Determine state of extent, and the filesystem.
4916 * A wasdelay extent has been initialized, so
4917 * shouldn't be flagged as unwritten.
4919 if (wr && XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4920 if (!wasdelay && (flags & XFS_BMAPI_PREALLOC))
4921 got.br_state = XFS_EXT_UNWRITTEN;
4923 error = xfs_bmap_add_extent(ip, lastx, &cur, &got,
4924 firstblock, flist, &tmp_logflags, whichfork,
4926 logflags |= tmp_logflags;
4929 lastx = ifp->if_lastex;
4930 ep = &ifp->if_u1.if_extents[lastx];
4931 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4932 xfs_bmbt_get_all(ep, &got);
4933 ASSERT(got.br_startoff <= aoff);
4934 ASSERT(got.br_startoff + got.br_blockcount >=
4938 ASSERT(ISNULLSTARTBLOCK(got.br_startblock));
4939 ASSERT(STARTBLOCKVAL(got.br_startblock) > 0);
4941 ASSERT(got.br_state == XFS_EXT_NORM ||
4942 got.br_state == XFS_EXT_UNWRITTEN);
4945 * Fall down into the found allocated space case.
4947 } else if (inhole) {
4949 * Reading in a hole.
4951 mval->br_startoff = bno;
4952 mval->br_startblock = HOLESTARTBLOCK;
4953 mval->br_blockcount =
4954 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4955 mval->br_state = XFS_EXT_NORM;
4956 bno += mval->br_blockcount;
4957 len -= mval->br_blockcount;
4963 * Then deal with the allocated space we found.
4966 if (trim && (got.br_startoff + got.br_blockcount > obno)) {
4969 ASSERT((bno >= obno) || (n == 0));
4971 mval->br_startoff = bno;
4972 if (ISNULLSTARTBLOCK(got.br_startblock)) {
4973 ASSERT(!wr || delay);
4974 mval->br_startblock = DELAYSTARTBLOCK;
4976 mval->br_startblock =
4978 (bno - got.br_startoff);
4980 * Return the minimum of what we got and what we
4981 * asked for for the length. We can use the len
4982 * variable here because it is modified below
4983 * and we could have been there before coming
4984 * here if the first part of the allocation
4985 * didn't overlap what was asked for.
4987 mval->br_blockcount =
4988 XFS_FILBLKS_MIN(end - bno, got.br_blockcount -
4989 (bno - got.br_startoff));
4990 mval->br_state = got.br_state;
4991 ASSERT(mval->br_blockcount <= len);
4994 if (ISNULLSTARTBLOCK(mval->br_startblock)) {
4995 ASSERT(!wr || delay);
4996 mval->br_startblock = DELAYSTARTBLOCK;
5001 * Check if writing previously allocated but
5002 * unwritten extents.
5004 if (wr && mval->br_state == XFS_EXT_UNWRITTEN &&
5005 ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_DELAY)) == 0)) {
5007 * Modify (by adding) the state flag, if writing.
5009 ASSERT(mval->br_blockcount <= len);
5010 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
5011 cur = xfs_btree_init_cursor(mp,
5012 tp, NULL, 0, XFS_BTNUM_BMAP,
5014 cur->bc_private.b.firstblock =
5016 cur->bc_private.b.flist = flist;
5018 mval->br_state = XFS_EXT_NORM;
5019 error = xfs_bmap_add_extent(ip, lastx, &cur, mval,
5020 firstblock, flist, &tmp_logflags, whichfork,
5022 logflags |= tmp_logflags;
5025 lastx = ifp->if_lastex;
5026 ep = &ifp->if_u1.if_extents[lastx];
5027 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5028 xfs_bmbt_get_all(ep, &got);
5030 * We may have combined previously unwritten
5031 * space with written space, so generate
5034 if (mval->br_blockcount < len)
5039 ((mval->br_startoff + mval->br_blockcount) <= end));
5040 ASSERT(!trim || (mval->br_blockcount <= len) ||
5041 (mval->br_startoff < obno));
5042 bno = mval->br_startoff + mval->br_blockcount;
5044 if (n > 0 && mval->br_startoff == mval[-1].br_startoff) {
5045 ASSERT(mval->br_startblock == mval[-1].br_startblock);
5046 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
5047 ASSERT(mval->br_state == mval[-1].br_state);
5048 mval[-1].br_blockcount = mval->br_blockcount;
5049 mval[-1].br_state = mval->br_state;
5050 } else if (n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
5051 mval[-1].br_startblock != DELAYSTARTBLOCK &&
5052 mval[-1].br_startblock != HOLESTARTBLOCK &&
5053 mval->br_startblock ==
5054 mval[-1].br_startblock + mval[-1].br_blockcount &&
5055 (stateless || mval[-1].br_state == mval->br_state)) {
5056 ASSERT(mval->br_startoff ==
5057 mval[-1].br_startoff + mval[-1].br_blockcount);
5058 mval[-1].br_blockcount += mval->br_blockcount;
5060 mval->br_startblock == DELAYSTARTBLOCK &&
5061 mval[-1].br_startblock == DELAYSTARTBLOCK &&
5062 mval->br_startoff ==
5063 mval[-1].br_startoff + mval[-1].br_blockcount) {
5064 mval[-1].br_blockcount += mval->br_blockcount;
5065 mval[-1].br_state = mval->br_state;
5066 } else if (!((n == 0) &&
5067 ((mval->br_startoff + mval->br_blockcount) <=
5073 * If we're done, stop now. Stop when we've allocated
5074 * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise
5075 * the transaction may get too big.
5077 if (bno >= end || n >= *nmap || nallocs >= *nmap)
5080 * Else go on to the next record.
5084 if (lastx >= nextents) {
5088 xfs_bmbt_get_all(ep, &got);
5090 ifp->if_lastex = lastx;
5093 * Transform from btree to extents, give it cur.
5095 if (tp && XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5096 XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5098 error = xfs_bmap_btree_to_extents(tp, ip, cur,
5099 &tmp_logflags, whichfork);
5100 logflags |= tmp_logflags;
5104 ASSERT(ifp->if_ext_max ==
5105 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5106 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
5107 XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max);
5112 * Log everything. Do this after conversion, there's no point in
5113 * logging the extent list if we've converted to btree format.
5115 if ((logflags & XFS_ILOG_FEXT(whichfork)) &&
5116 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5117 logflags &= ~XFS_ILOG_FEXT(whichfork);
5118 else if ((logflags & XFS_ILOG_FBROOT(whichfork)) &&
5119 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5120 logflags &= ~XFS_ILOG_FBROOT(whichfork);
5122 * Log whatever the flags say, even if error. Otherwise we might miss
5123 * detecting a case where the data is changed, there's an error,
5124 * and it's not logged so we don't shutdown when we should.
5128 xfs_trans_log_inode(tp, ip, logflags);
5132 ASSERT(*firstblock == NULLFSBLOCK ||
5133 XFS_FSB_TO_AGNO(mp, *firstblock) ==
5135 cur->bc_private.b.firstblock) ||
5137 XFS_FSB_TO_AGNO(mp, *firstblock) <
5139 cur->bc_private.b.firstblock)));
5140 *firstblock = cur->bc_private.b.firstblock;
5142 xfs_btree_del_cursor(cur,
5143 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5146 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
5152 * Map file blocks to filesystem blocks, simple version.
5153 * One block (extent) only, read-only.
5154 * For flags, only the XFS_BMAPI_ATTRFORK flag is examined.
5155 * For the other flag values, the effect is as if XFS_BMAPI_METADATA
5156 * was set and all the others were clear.
5160 xfs_trans_t *tp, /* transaction pointer */
5161 xfs_inode_t *ip, /* incore inode */
5162 int whichfork, /* data or attr fork */
5163 xfs_fsblock_t *fsb, /* output: mapped block */
5164 xfs_fileoff_t bno) /* starting file offs. mapped */
5166 int eof; /* we've hit the end of extent list */
5167 int error; /* error return */
5168 xfs_bmbt_irec_t got; /* current extent list record */
5169 xfs_ifork_t *ifp; /* inode fork pointer */
5170 xfs_extnum_t lastx; /* last useful extent number */
5171 xfs_bmbt_irec_t prev; /* previous extent list record */
5173 ifp = XFS_IFORK_PTR(ip, whichfork);
5175 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
5176 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)) {
5177 XFS_ERROR_REPORT("xfs_bmapi_single", XFS_ERRLEVEL_LOW,
5179 return XFS_ERROR(EFSCORRUPTED);
5181 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
5182 return XFS_ERROR(EIO);
5183 XFS_STATS_INC(xs_blk_mapr);
5184 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5185 (error = xfs_iread_extents(tp, ip, whichfork)))
5187 (void)xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5190 * Reading past eof, act as though there's a hole
5193 if (eof || got.br_startoff > bno) {
5197 ASSERT(!ISNULLSTARTBLOCK(got.br_startblock));
5198 ASSERT(bno < got.br_startoff + got.br_blockcount);
5199 *fsb = got.br_startblock + (bno - got.br_startoff);
5200 ifp->if_lastex = lastx;
5205 * Unmap (remove) blocks from a file.
5206 * If nexts is nonzero then the number of extents to remove is limited to
5207 * that value. If not all extents in the block range can be removed then
5212 xfs_trans_t *tp, /* transaction pointer */
5213 struct xfs_inode *ip, /* incore inode */
5214 xfs_fileoff_t bno, /* starting offset to unmap */
5215 xfs_filblks_t len, /* length to unmap in file */
5216 int flags, /* misc flags */
5217 xfs_extnum_t nexts, /* number of extents max */
5218 xfs_fsblock_t *firstblock, /* first allocated block
5219 controls a.g. for allocs */
5220 xfs_bmap_free_t *flist, /* i/o: list extents to free */
5221 int *done) /* set if not done yet */
5223 xfs_btree_cur_t *cur; /* bmap btree cursor */
5224 xfs_bmbt_irec_t del; /* extent being deleted */
5225 int eof; /* is deleting at eof */
5226 xfs_bmbt_rec_t *ep; /* extent list entry pointer */
5227 int error; /* error return value */
5228 xfs_extnum_t extno; /* extent number in list */
5229 xfs_bmbt_irec_t got; /* current extent list entry */
5230 xfs_ifork_t *ifp; /* inode fork pointer */
5231 int isrt; /* freeing in rt area */
5232 xfs_extnum_t lastx; /* last extent index used */
5233 int logflags; /* transaction logging flags */
5234 xfs_extlen_t mod; /* rt extent offset */
5235 xfs_mount_t *mp; /* mount structure */
5236 xfs_extnum_t nextents; /* size of extent list */
5237 xfs_bmbt_irec_t prev; /* previous extent list entry */
5238 xfs_fileoff_t start; /* first file offset deleted */
5239 int tmp_logflags; /* partial logging flags */
5240 int wasdel; /* was a delayed alloc extent */
5241 int whichfork; /* data or attribute fork */
5242 int rsvd; /* OK to allocate reserved blocks */
5245 xfs_bunmap_trace(ip, bno, len, flags, (inst_t *)__return_address);
5246 whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5247 XFS_ATTR_FORK : XFS_DATA_FORK;
5248 ifp = XFS_IFORK_PTR(ip, whichfork);
5250 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5251 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5252 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5254 return XFS_ERROR(EFSCORRUPTED);
5257 if (XFS_FORCED_SHUTDOWN(mp))
5258 return XFS_ERROR(EIO);
5259 rsvd = (flags & XFS_BMAPI_RSVBLOCKS) != 0;
5262 ASSERT(ifp->if_ext_max ==
5263 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5264 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5265 (error = xfs_iread_extents(tp, ip, whichfork)))
5267 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5268 if (nextents == 0) {
5272 XFS_STATS_INC(xs_blk_unmap);
5273 isrt = (whichfork == XFS_DATA_FORK) &&
5274 (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
5276 bno = start + len - 1;
5277 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5280 * Check to see if the given block number is past the end of the
5281 * file, back up to the last block if so...
5284 ep = &ifp->if_u1.if_extents[--lastx];
5285 xfs_bmbt_get_all(ep, &got);
5286 bno = got.br_startoff + got.br_blockcount - 1;
5289 if (ifp->if_flags & XFS_IFBROOT) {
5290 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5291 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
5293 cur->bc_private.b.firstblock = *firstblock;
5294 cur->bc_private.b.flist = flist;
5295 cur->bc_private.b.flags = 0;
5299 while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5300 (nexts == 0 || extno < nexts)) {
5302 * Is the found extent after a hole in which bno lives?
5303 * Just back up to the previous extent, if so.
5305 if (got.br_startoff > bno) {
5309 xfs_bmbt_get_all(ep, &got);
5312 * Is the last block of this extent before the range
5313 * we're supposed to delete? If so, we're done.
5315 bno = XFS_FILEOFF_MIN(bno,
5316 got.br_startoff + got.br_blockcount - 1);
5320 * Then deal with the (possibly delayed) allocated space
5325 wasdel = ISNULLSTARTBLOCK(del.br_startblock);
5326 if (got.br_startoff < start) {
5327 del.br_startoff = start;
5328 del.br_blockcount -= start - got.br_startoff;
5330 del.br_startblock += start - got.br_startoff;
5332 if (del.br_startoff + del.br_blockcount > bno + 1)
5333 del.br_blockcount = bno + 1 - del.br_startoff;
5334 sum = del.br_startblock + del.br_blockcount;
5336 (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5338 * Realtime extent not lined up at the end.
5339 * The extent could have been split into written
5340 * and unwritten pieces, or we could just be
5341 * unmapping part of it. But we can't really
5342 * get rid of part of a realtime extent.
5344 if (del.br_state == XFS_EXT_UNWRITTEN ||
5345 !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
5347 * This piece is unwritten, or we're not
5348 * using unwritten extents. Skip over it.
5351 bno -= mod > del.br_blockcount ?
5352 del.br_blockcount : mod;
5353 if (bno < got.br_startoff) {
5355 xfs_bmbt_get_all(--ep, &got);
5360 * It's written, turn it unwritten.
5361 * This is better than zeroing it.
5363 ASSERT(del.br_state == XFS_EXT_NORM);
5364 ASSERT(xfs_trans_get_block_res(tp) > 0);
5366 * If this spans a realtime extent boundary,
5367 * chop it back to the start of the one we end at.
5369 if (del.br_blockcount > mod) {
5370 del.br_startoff += del.br_blockcount - mod;
5371 del.br_startblock += del.br_blockcount - mod;
5372 del.br_blockcount = mod;
5374 del.br_state = XFS_EXT_UNWRITTEN;
5375 error = xfs_bmap_add_extent(ip, lastx, &cur, &del,
5376 firstblock, flist, &logflags, XFS_DATA_FORK, 0);
5381 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5383 * Realtime extent is lined up at the end but not
5384 * at the front. We'll get rid of full extents if
5387 mod = mp->m_sb.sb_rextsize - mod;
5388 if (del.br_blockcount > mod) {
5389 del.br_blockcount -= mod;
5390 del.br_startoff += mod;
5391 del.br_startblock += mod;
5392 } else if ((del.br_startoff == start &&
5393 (del.br_state == XFS_EXT_UNWRITTEN ||
5394 xfs_trans_get_block_res(tp) == 0)) ||
5395 !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
5397 * Can't make it unwritten. There isn't
5398 * a full extent here so just skip it.
5400 ASSERT(bno >= del.br_blockcount);
5401 bno -= del.br_blockcount;
5402 if (bno < got.br_startoff) {
5404 xfs_bmbt_get_all(--ep, &got);
5407 } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5409 * This one is already unwritten.
5410 * It must have a written left neighbor.
5411 * Unwrite the killed part of that one and
5415 xfs_bmbt_get_all(ep - 1, &prev);
5416 ASSERT(prev.br_state == XFS_EXT_NORM);
5417 ASSERT(!ISNULLSTARTBLOCK(prev.br_startblock));
5418 ASSERT(del.br_startblock ==
5419 prev.br_startblock + prev.br_blockcount);
5420 if (prev.br_startoff < start) {
5421 mod = start - prev.br_startoff;
5422 prev.br_blockcount -= mod;
5423 prev.br_startblock += mod;
5424 prev.br_startoff = start;
5426 prev.br_state = XFS_EXT_UNWRITTEN;
5427 error = xfs_bmap_add_extent(ip, lastx - 1, &cur,
5428 &prev, firstblock, flist, &logflags,
5434 ASSERT(del.br_state == XFS_EXT_NORM);
5435 del.br_state = XFS_EXT_UNWRITTEN;
5436 error = xfs_bmap_add_extent(ip, lastx, &cur,
5437 &del, firstblock, flist, &logflags,
5445 ASSERT(STARTBLOCKVAL(del.br_startblock) > 0);
5446 /* Update realtim/data freespace, unreserve quota */
5448 xfs_filblks_t rtexts;
5450 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5451 do_div(rtexts, mp->m_sb.sb_rextsize);
5452 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5454 XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, NULL, ip,
5455 -((long)del.br_blockcount), 0,
5456 XFS_QMOPT_RES_RTBLKS);
5458 xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS,
5459 (int)del.br_blockcount, rsvd);
5460 XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, NULL, ip,
5461 -((long)del.br_blockcount), 0,
5462 XFS_QMOPT_RES_REGBLKS);
5464 ip->i_delayed_blks -= del.br_blockcount;
5466 cur->bc_private.b.flags |=
5467 XFS_BTCUR_BPRV_WASDEL;
5469 cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5471 * If it's the case where the directory code is running
5472 * with no block reservation, and the deleted block is in
5473 * the middle of its extent, and the resulting insert
5474 * of an extent would cause transformation to btree format,
5475 * then reject it. The calling code will then swap
5476 * blocks around instead.
5477 * We have to do this now, rather than waiting for the
5478 * conversion to btree format, since the transaction
5481 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5482 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5483 XFS_IFORK_NEXTENTS(ip, whichfork) >= ifp->if_ext_max &&
5484 del.br_startoff > got.br_startoff &&
5485 del.br_startoff + del.br_blockcount <
5486 got.br_startoff + got.br_blockcount) {
5487 error = XFS_ERROR(ENOSPC);
5490 error = xfs_bmap_del_extent(ip, tp, lastx, flist, cur, &del,
5491 &tmp_logflags, whichfork, rsvd);
5492 logflags |= tmp_logflags;
5495 bno = del.br_startoff - 1;
5497 lastx = ifp->if_lastex;
5499 * If not done go on to the next (previous) record.
5500 * Reset ep in case the extents array was re-alloced.
5502 ep = &ifp->if_u1.if_extents[lastx];
5503 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5504 if (lastx >= XFS_IFORK_NEXTENTS(ip, whichfork) ||
5505 xfs_bmbt_get_startoff(ep) > bno) {
5510 xfs_bmbt_get_all(ep, &got);
5514 ifp->if_lastex = lastx;
5515 *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5516 ASSERT(ifp->if_ext_max ==
5517 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5519 * Convert to a btree if necessary.
5521 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5522 XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
5523 ASSERT(cur == NULL);
5524 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5525 &cur, 0, &tmp_logflags, whichfork);
5526 logflags |= tmp_logflags;
5531 * transform from btree to extents, give it cur
5533 else if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5534 XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5535 ASSERT(cur != NULL);
5536 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5538 logflags |= tmp_logflags;
5543 * transform from extents to local?
5545 ASSERT(ifp->if_ext_max ==
5546 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5550 * Log everything. Do this after conversion, there's no point in
5551 * logging the extent list if we've converted to btree format.
5553 if ((logflags & XFS_ILOG_FEXT(whichfork)) &&
5554 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5555 logflags &= ~XFS_ILOG_FEXT(whichfork);
5556 else if ((logflags & XFS_ILOG_FBROOT(whichfork)) &&
5557 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5558 logflags &= ~XFS_ILOG_FBROOT(whichfork);
5560 * Log inode even in the error case, if the transaction
5561 * is dirty we'll need to shut down the filesystem.
5564 xfs_trans_log_inode(tp, ip, logflags);
5567 *firstblock = cur->bc_private.b.firstblock;
5568 cur->bc_private.b.allocated = 0;
5570 xfs_btree_del_cursor(cur,
5571 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5577 * Fcntl interface to xfs_bmapi.
5579 int /* error code */
5581 bhv_desc_t *bdp, /* XFS behavior descriptor*/
5582 struct getbmap *bmv, /* user bmap structure */
5583 void __user *ap, /* pointer to user's array */
5584 int interface) /* interface flags */
5586 __int64_t bmvend; /* last block requested */
5587 int error; /* return value */
5588 __int64_t fixlen; /* length for -1 case */
5589 int i; /* extent number */
5590 xfs_inode_t *ip; /* xfs incore inode pointer */
5591 vnode_t *vp; /* corresponding vnode */
5592 int lock; /* lock state */
5593 xfs_bmbt_irec_t *map; /* buffer for user's data */
5594 xfs_mount_t *mp; /* file system mount point */
5595 int nex; /* # of user extents can do */
5596 int nexleft; /* # of user extents left */
5597 int subnex; /* # of bmapi's can do */
5598 int nmap; /* number of map entries */
5599 struct getbmap out; /* output structure */
5600 int whichfork; /* data or attr fork */
5601 int prealloced; /* this is a file with
5602 * preallocated data space */
5603 int sh_unwritten; /* true, if unwritten */
5604 /* extents listed separately */
5605 int bmapi_flags; /* flags for xfs_bmapi */
5606 __int32_t oflags; /* getbmapx bmv_oflags field */
5608 vp = BHV_TO_VNODE(bdp);
5609 ip = XFS_BHVTOI(bdp);
5612 whichfork = interface & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5613 sh_unwritten = (interface & BMV_IF_PREALLOC) != 0;
5615 /* If the BMV_IF_NO_DMAPI_READ interface bit specified, do not
5616 * generate a DMAPI read event. Otherwise, if the DM_EVENT_READ
5617 * bit is set for the file, generate a read event in order
5618 * that the DMAPI application may do its thing before we return
5619 * the extents. Usually this means restoring user file data to
5620 * regions of the file that look like holes.
5622 * The "old behavior" (from XFS_IOC_GETBMAP) is to not specify
5623 * BMV_IF_NO_DMAPI_READ so that read events are generated.
5624 * If this were not true, callers of ioctl( XFS_IOC_GETBMAP )
5625 * could misinterpret holes in a DMAPI file as true holes,
5626 * when in fact they may represent offline user data.
5628 if ( (interface & BMV_IF_NO_DMAPI_READ) == 0
5629 && DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_READ)
5630 && whichfork == XFS_DATA_FORK) {
5632 error = XFS_SEND_DATA(mp, DM_EVENT_READ, vp, 0, 0, 0, NULL);
5634 return XFS_ERROR(error);
5637 if (whichfork == XFS_ATTR_FORK) {
5638 if (XFS_IFORK_Q(ip)) {
5639 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5640 ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5641 ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5642 return XFS_ERROR(EINVAL);
5643 } else if (unlikely(
5644 ip->i_d.di_aformat != 0 &&
5645 ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5646 XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5648 return XFS_ERROR(EFSCORRUPTED);
5650 } else if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5651 ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5652 ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5653 return XFS_ERROR(EINVAL);
5654 if (whichfork == XFS_DATA_FORK) {
5655 if (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC) {
5657 fixlen = XFS_MAXIOFFSET(mp);
5660 fixlen = ip->i_d.di_size;
5667 if (bmv->bmv_length == -1) {
5668 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5669 bmv->bmv_length = MAX( (__int64_t)(fixlen - bmv->bmv_offset),
5671 } else if (bmv->bmv_length < 0)
5672 return XFS_ERROR(EINVAL);
5673 if (bmv->bmv_length == 0) {
5674 bmv->bmv_entries = 0;
5677 nex = bmv->bmv_count - 1;
5679 return XFS_ERROR(EINVAL);
5680 bmvend = bmv->bmv_offset + bmv->bmv_length;
5682 xfs_ilock(ip, XFS_IOLOCK_SHARED);
5684 if (whichfork == XFS_DATA_FORK && ip->i_delayed_blks) {
5685 /* xfs_fsize_t last_byte = xfs_file_last_byte(ip); */
5686 VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1, 0, FI_REMAPF, error);
5689 ASSERT(whichfork == XFS_ATTR_FORK || ip->i_delayed_blks == 0);
5691 lock = xfs_ilock_map_shared(ip);
5694 * Don't let nex be bigger than the number of extents
5695 * we can have assuming alternating holes and real extents.
5697 if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5698 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5700 bmapi_flags = XFS_BMAPI_AFLAG(whichfork) |
5701 ((sh_unwritten) ? 0 : XFS_BMAPI_IGSTATE);
5704 * Allocate enough space to handle "subnex" maps at a time.
5707 map = kmem_alloc(subnex * sizeof(*map), KM_SLEEP);
5709 bmv->bmv_entries = 0;
5711 if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0) {
5713 goto unlock_and_return;
5719 nmap = (nexleft > subnex) ? subnex : nexleft;
5720 error = xfs_bmapi(NULL, ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5721 XFS_BB_TO_FSB(mp, bmv->bmv_length),
5722 bmapi_flags, NULL, 0, map, &nmap, NULL);
5724 goto unlock_and_return;
5725 ASSERT(nmap <= subnex);
5727 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5729 oflags = (map[i].br_state == XFS_EXT_UNWRITTEN) ?
5730 BMV_OF_PREALLOC : 0;
5731 out.bmv_offset = XFS_FSB_TO_BB(mp, map[i].br_startoff);
5732 out.bmv_length = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5733 ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
5734 if (map[i].br_startblock == HOLESTARTBLOCK &&
5735 ((prealloced && out.bmv_offset + out.bmv_length == bmvend) ||
5736 whichfork == XFS_ATTR_FORK )) {
5738 * came to hole at end of file or the end of
5741 goto unlock_and_return;
5744 (map[i].br_startblock == HOLESTARTBLOCK) ?
5746 XFS_FSB_TO_DB(ip, map[i].br_startblock);
5748 /* return either getbmap/getbmapx structure. */
5749 if (interface & BMV_IF_EXTENDED) {
5750 struct getbmapx outx;
5752 GETBMAP_CONVERT(out,outx);
5753 outx.bmv_oflags = oflags;
5754 outx.bmv_unused1 = outx.bmv_unused2 = 0;
5755 if (copy_to_user(ap, &outx,
5757 error = XFS_ERROR(EFAULT);
5758 goto unlock_and_return;
5761 if (copy_to_user(ap, &out,
5763 error = XFS_ERROR(EFAULT);
5764 goto unlock_and_return;
5768 out.bmv_offset + out.bmv_length;
5769 bmv->bmv_length = MAX((__int64_t)0,
5770 (__int64_t)(bmvend - bmv->bmv_offset));
5772 ap = (interface & BMV_IF_EXTENDED) ?
5774 ((struct getbmapx __user *)ap + 1) :
5776 ((struct getbmap __user *)ap + 1);
5779 } while (nmap && nexleft && bmv->bmv_length);
5782 xfs_iunlock_map_shared(ip, lock);
5783 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5785 kmem_free(map, subnex * sizeof(*map));
5791 * Check the last inode extent to determine whether this allocation will result
5792 * in blocks being allocated at the end of the file. When we allocate new data
5793 * blocks at the end of the file which do not start at the previous data block,
5794 * we will try to align the new blocks at stripe unit boundaries.
5796 STATIC int /* error */
5798 xfs_inode_t *ip, /* incore inode pointer */
5799 xfs_fileoff_t off, /* file offset in fsblocks */
5800 int whichfork, /* data or attribute fork */
5801 char *aeof) /* return value */
5803 int error; /* error return value */
5804 xfs_ifork_t *ifp; /* inode fork pointer */
5805 xfs_bmbt_rec_t *lastrec; /* extent list entry pointer */
5806 xfs_extnum_t nextents; /* size of extent list */
5807 xfs_bmbt_irec_t s; /* expanded extent list entry */
5809 ASSERT(whichfork == XFS_DATA_FORK);
5810 ifp = XFS_IFORK_PTR(ip, whichfork);
5811 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5812 (error = xfs_iread_extents(NULL, ip, whichfork)))
5814 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5815 if (nextents == 0) {
5820 * Go to the last extent
5822 lastrec = &ifp->if_u1.if_extents[nextents - 1];
5823 xfs_bmbt_get_all(lastrec, &s);
5825 * Check we are allocating in the last extent (for delayed allocations)
5826 * or past the last extent for non-delayed allocations.
5828 *aeof = (off >= s.br_startoff &&
5829 off < s.br_startoff + s.br_blockcount &&
5830 ISNULLSTARTBLOCK(s.br_startblock)) ||
5831 off >= s.br_startoff + s.br_blockcount;
5836 * Check if the endoff is outside the last extent. If so the caller will grow
5837 * the allocation to a stripe unit boundary.
5841 xfs_inode_t *ip, /* incore inode pointer */
5842 xfs_fileoff_t endoff, /* file offset in fsblocks */
5843 int whichfork, /* data or attribute fork */
5844 int *eof) /* result value */
5846 xfs_fsblock_t blockcount; /* extent block count */
5847 int error; /* error return value */
5848 xfs_ifork_t *ifp; /* inode fork pointer */
5849 xfs_bmbt_rec_t *lastrec; /* extent list entry pointer */
5850 xfs_extnum_t nextents; /* size of extent list */
5851 xfs_fileoff_t startoff; /* extent starting file offset */
5853 ASSERT(whichfork == XFS_DATA_FORK);
5854 ifp = XFS_IFORK_PTR(ip, whichfork);
5855 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5856 (error = xfs_iread_extents(NULL, ip, whichfork)))
5858 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5859 if (nextents == 0) {
5864 * Go to the last extent
5866 lastrec = &ifp->if_u1.if_extents[nextents - 1];
5867 startoff = xfs_bmbt_get_startoff(lastrec);
5868 blockcount = xfs_bmbt_get_blockcount(lastrec);
5869 *eof = endoff >= startoff + blockcount;
5875 * Check that the extents list for the inode ip is in the right order.
5878 xfs_bmap_check_extents(
5879 xfs_inode_t *ip, /* incore inode pointer */
5880 int whichfork) /* data or attr fork */
5882 xfs_bmbt_rec_t *base; /* base of extents list */
5883 xfs_bmbt_rec_t *ep; /* current extent entry */
5884 xfs_ifork_t *ifp; /* inode fork pointer */
5885 xfs_extnum_t nextents; /* number of extents in list */
5887 ifp = XFS_IFORK_PTR(ip, whichfork);
5888 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
5889 base = ifp->if_u1.if_extents;
5890 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5891 for (ep = base; ep < &base[nextents - 1]; ep++) {
5892 xfs_btree_check_rec(XFS_BTNUM_BMAP, (void *)ep,
5900 xfs_btree_cur_t *cur,
5910 for(i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
5911 bp = cur->bc_bufs[i];
5913 if (XFS_BUF_ADDR(bp) == bno)
5914 break; /* Found it */
5916 if (i == XFS_BTREE_MAXLEVELS)
5919 if (!bp) { /* Chase down all the log items to see if the bp is there */
5920 xfs_log_item_chunk_t *licp;
5924 licp = &tp->t_items;
5925 while (!bp && licp != NULL) {
5926 if (XFS_LIC_ARE_ALL_FREE(licp)) {
5927 licp = licp->lic_next;
5930 for (i = 0; i < licp->lic_unused; i++) {
5931 xfs_log_item_desc_t *lidp;
5932 xfs_log_item_t *lip;
5933 xfs_buf_log_item_t *bip;
5936 if (XFS_LIC_ISFREE(licp, i)) {
5940 lidp = XFS_LIC_SLOT(licp, i);
5941 lip = lidp->lid_item;
5942 if (lip->li_type != XFS_LI_BUF)
5945 bip = (xfs_buf_log_item_t *)lip;
5948 if (XFS_BUF_ADDR(lbp) == bno) {
5950 break; /* Found it */
5953 licp = licp->lic_next;
5961 xfs_bmbt_block_t *block,
5967 xfs_bmbt_ptr_t *pp, *thispa; /* pointer to block address */
5968 xfs_bmbt_key_t *prevp, *keyp;
5970 ASSERT(be16_to_cpu(block->bb_level) > 0);
5973 for( i = 1; i <= be16_to_cpu(block->bb_numrecs); i++) {
5974 dmxr = mp->m_bmap_dmxr[0];
5977 keyp = XFS_BMAP_BROOT_KEY_ADDR(block, i, sz);
5979 keyp = XFS_BTREE_KEY_ADDR(mp->m_sb.sb_blocksize,
5980 xfs_bmbt, block, i, dmxr);
5984 xfs_btree_check_key(XFS_BTNUM_BMAP, prevp, keyp);
5989 * Compare the block numbers to see if there are dups.
5993 pp = XFS_BMAP_BROOT_PTR_ADDR(block, i, sz);
5995 pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize,
5996 xfs_bmbt, block, i, dmxr);
5998 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
6000 thispa = XFS_BMAP_BROOT_PTR_ADDR(block, j, sz);
6002 thispa = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize,
6003 xfs_bmbt, block, j, dmxr);
6005 if (INT_GET(*thispa, ARCH_CONVERT) ==
6006 INT_GET(*pp, ARCH_CONVERT)) {
6007 cmn_err(CE_WARN, "%s: thispa(%d) == pp(%d) %Ld",
6009 INT_GET(*thispa, ARCH_CONVERT));
6010 panic("%s: ptrs are equal in node\n",
6018 * Check that the extents for the inode ip are in the right order in all
6023 xfs_bmap_check_leaf_extents(
6024 xfs_btree_cur_t *cur, /* btree cursor or null */
6025 xfs_inode_t *ip, /* incore inode pointer */
6026 int whichfork) /* data or attr fork */
6028 xfs_bmbt_block_t *block; /* current btree block */
6029 xfs_fsblock_t bno; /* block # of "block" */
6030 xfs_buf_t *bp; /* buffer for "block" */
6031 int error; /* error return value */
6032 xfs_extnum_t i=0; /* index into the extents list */
6033 xfs_ifork_t *ifp; /* fork structure */
6034 int level; /* btree level, for checking */
6035 xfs_mount_t *mp; /* file system mount structure */
6036 xfs_bmbt_ptr_t *pp; /* pointer to block address */
6037 xfs_bmbt_rec_t *ep, *lastp; /* extent pointers in block entry */
6040 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
6046 ifp = XFS_IFORK_PTR(ip, whichfork);
6047 block = ifp->if_broot;
6049 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6051 level = be16_to_cpu(block->bb_level);
6053 xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
6054 pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
6055 ASSERT(INT_GET(*pp, ARCH_CONVERT) != NULLDFSBNO);
6056 ASSERT(XFS_FSB_TO_AGNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agcount);
6057 ASSERT(XFS_FSB_TO_AGBNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agblocks);
6058 bno = INT_GET(*pp, ARCH_CONVERT);
6060 * Go down the tree until leaf level is reached, following the first
6061 * pointer (leftmost) at each level.
6063 while (level-- > 0) {
6064 /* See if buf is in cur first */
6065 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
6071 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
6072 XFS_BMAP_BTREE_REF)))
6074 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6075 XFS_WANT_CORRUPTED_GOTO(
6076 XFS_BMAP_SANITY_CHECK(mp, block, level),
6082 * Check this block for basic sanity (increasing keys and
6083 * no duplicate blocks).
6086 xfs_check_block(block, mp, 0, 0);
6087 pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt, block,
6088 1, mp->m_bmap_dmxr[1]);
6089 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, INT_GET(*pp, ARCH_CONVERT)), error0);
6090 bno = INT_GET(*pp, ARCH_CONVERT);
6093 xfs_trans_brelse(NULL, bp);
6098 * Here with bp and block set to the leftmost leaf node in the tree.
6103 * Loop over all leaf nodes checking that all extents are in the right order.
6107 xfs_bmbt_rec_t *frp;
6108 xfs_fsblock_t nextbno;
6109 xfs_extnum_t num_recs;
6112 num_recs = be16_to_cpu(block->bb_numrecs);
6115 * Read-ahead the next leaf block, if any.
6118 nextbno = be64_to_cpu(block->bb_rightsib);
6121 * Check all the extents to make sure they are OK.
6122 * If we had a previous block, the last entry should
6123 * conform with the first entry in this one.
6126 frp = XFS_BTREE_REC_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt,
6127 block, 1, mp->m_bmap_dmxr[0]);
6129 for (ep = frp;ep < frp + (num_recs - 1); ep++) {
6131 xfs_btree_check_rec(XFS_BTNUM_BMAP,
6132 (void *)lastp, (void *)ep);
6134 xfs_btree_check_rec(XFS_BTNUM_BMAP, (void *)ep,
6137 lastp = frp + num_recs - 1; /* For the next iteration */
6142 xfs_trans_brelse(NULL, bp);
6146 * If we've reached the end, stop.
6148 if (bno == NULLFSBLOCK)
6151 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
6157 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
6158 XFS_BMAP_BTREE_REF)))
6160 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6164 xfs_trans_brelse(NULL, bp);
6169 cmn_err(CE_WARN, "%s: at error0", __FUNCTION__);
6171 xfs_trans_brelse(NULL, bp);
6173 cmn_err(CE_WARN, "%s: BAD after btree leaves for %d extents",
6175 panic("%s: CORRUPTED BTREE OR SOMETHING", __FUNCTION__);
6181 * Count fsblocks of the given fork.
6184 xfs_bmap_count_blocks(
6185 xfs_trans_t *tp, /* transaction pointer */
6186 xfs_inode_t *ip, /* incore inode */
6187 int whichfork, /* data or attr fork */
6188 int *count) /* out: count of blocks */
6190 xfs_bmbt_block_t *block; /* current btree block */
6191 xfs_fsblock_t bno; /* block # of "block" */
6192 xfs_ifork_t *ifp; /* fork structure */
6193 int level; /* btree level, for checking */
6194 xfs_mount_t *mp; /* file system mount structure */
6195 xfs_bmbt_ptr_t *pp; /* pointer to block address */
6199 ifp = XFS_IFORK_PTR(ip, whichfork);
6200 if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
6201 if (unlikely(xfs_bmap_count_leaves(ifp->if_u1.if_extents,
6202 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
6204 XFS_ERROR_REPORT("xfs_bmap_count_blocks(1)",
6205 XFS_ERRLEVEL_LOW, mp);
6206 return XFS_ERROR(EFSCORRUPTED);
6212 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6214 block = ifp->if_broot;
6215 level = be16_to_cpu(block->bb_level);
6217 pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
6218 ASSERT(INT_GET(*pp, ARCH_CONVERT) != NULLDFSBNO);
6219 ASSERT(XFS_FSB_TO_AGNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agcount);
6220 ASSERT(XFS_FSB_TO_AGBNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agblocks);
6221 bno = INT_GET(*pp, ARCH_CONVERT);
6223 if (unlikely(xfs_bmap_count_tree(mp, tp, bno, level, count) < 0)) {
6224 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
6226 return XFS_ERROR(EFSCORRUPTED);
6233 * Recursively walks each level of a btree
6234 * to count total fsblocks is use.
6237 xfs_bmap_count_tree(
6238 xfs_mount_t *mp, /* file system mount point */
6239 xfs_trans_t *tp, /* transaction pointer */
6240 xfs_fsblock_t blockno, /* file system block number */
6241 int levelin, /* level in btree */
6242 int *count) /* Count of blocks */
6245 xfs_buf_t *bp, *nbp;
6246 int level = levelin;
6248 xfs_fsblock_t bno = blockno;
6249 xfs_fsblock_t nextbno;
6250 xfs_bmbt_block_t *block, *nextblock;
6252 xfs_bmbt_rec_t *frp;
6254 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
6257 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6260 /* Not at node above leafs, count this level of nodes */
6261 nextbno = be64_to_cpu(block->bb_rightsib);
6262 while (nextbno != NULLFSBLOCK) {
6263 if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
6264 0, &nbp, XFS_BMAP_BTREE_REF)))
6267 nextblock = XFS_BUF_TO_BMBT_BLOCK(nbp);
6268 nextbno = be64_to_cpu(nextblock->bb_rightsib);
6269 xfs_trans_brelse(tp, nbp);
6272 /* Dive to the next level */
6273 pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize,
6274 xfs_bmbt, block, 1, mp->m_bmap_dmxr[1]);
6275 bno = INT_GET(*pp, ARCH_CONVERT);
6276 if (unlikely((error =
6277 xfs_bmap_count_tree(mp, tp, bno, level, count)) < 0)) {
6278 xfs_trans_brelse(tp, bp);
6279 XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6280 XFS_ERRLEVEL_LOW, mp);
6281 return XFS_ERROR(EFSCORRUPTED);
6283 xfs_trans_brelse(tp, bp);
6285 /* count all level 1 nodes and their leaves */
6287 nextbno = be64_to_cpu(block->bb_rightsib);
6288 numrecs = be16_to_cpu(block->bb_numrecs);
6289 frp = XFS_BTREE_REC_ADDR(mp->m_sb.sb_blocksize,
6290 xfs_bmbt, block, 1, mp->m_bmap_dmxr[0]);
6291 if (unlikely(xfs_bmap_disk_count_leaves(frp, numrecs, count) < 0)) {
6292 xfs_trans_brelse(tp, bp);
6293 XFS_ERROR_REPORT("xfs_bmap_count_tree(2)",
6294 XFS_ERRLEVEL_LOW, mp);
6295 return XFS_ERROR(EFSCORRUPTED);
6297 xfs_trans_brelse(tp, bp);
6298 if (nextbno == NULLFSBLOCK)
6301 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6302 XFS_BMAP_BTREE_REF)))
6305 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6312 * Count leaf blocks given a pointer to an extent list.
6315 xfs_bmap_count_leaves(
6316 xfs_bmbt_rec_t *frp,
6322 for ( b = 1; b <= numrecs; b++, frp++)
6323 *count += xfs_bmbt_get_blockcount(frp);
6328 * Count leaf blocks given a pointer to an extent list originally in btree format.
6331 xfs_bmap_disk_count_leaves(
6332 xfs_bmbt_rec_t *frp,
6338 for ( b = 1; b <= numrecs; b++, frp++)
6339 *count += xfs_bmbt_disk_get_blockcount(frp);
git://ftp.safe.ca / safe / jmp / linux-2.6 / blob