]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - fs/xfs/xfs_bmap.c
projects / mirror_ubuntu-zesty-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
xfs: only lock the rt bitmap inode once per allocation
[mirror_ubuntu-zesty-kernel.git] / fs / xfs / xfs_bmap.c
1 /*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
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.
8 *
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.
13 *
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
17 */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_inum.h"
24 #include "xfs_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_dir2.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_alloc_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_dir2_sf.h"
33 #include "xfs_dinode.h"
34 #include "xfs_inode.h"
35 #include "xfs_btree.h"
36 #include "xfs_mount.h"
37 #include "xfs_itable.h"
38 #include "xfs_dir2_data.h"
39 #include "xfs_dir2_leaf.h"
40 #include "xfs_dir2_block.h"
41 #include "xfs_inode_item.h"
42 #include "xfs_extfree_item.h"
43 #include "xfs_alloc.h"
44 #include "xfs_bmap.h"
45 #include "xfs_rtalloc.h"
46 #include "xfs_error.h"
47 #include "xfs_attr_leaf.h"
48 #include "xfs_rw.h"
49 #include "xfs_quota.h"
50 #include "xfs_trans_space.h"
51 #include "xfs_buf_item.h"
52 #include "xfs_filestream.h"
53 #include "xfs_vnodeops.h"
54 #include "xfs_trace.h"
55
56
57 #ifdef DEBUG
58 STATIC void
59 xfs_bmap_check_leaf_extents(xfs_btree_cur_t *cur, xfs_inode_t *ip, int whichfork);
60 #endif
61
62 kmem_zone_t *xfs_bmap_free_item_zone;
63
64 /*
65 * Prototypes for internal bmap routines.
66 */
67
68
69 /*
70 * Called from xfs_bmap_add_attrfork to handle extents format files.
71 */
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 */
79
80 /*
81 * Called from xfs_bmap_add_attrfork to handle local format files.
82 */
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 */
90
91 /*
92 * Called by xfs_bmapi to update file extent records and the btree
93 * after allocating space (or doing a delayed allocation).
94 */
95 STATIC int /* error */
96 xfs_bmap_add_extent(
97 xfs_inode_t *ip, /* incore inode pointer */
98 xfs_extnum_t idx, /* extent number to update/insert */
99 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
100 xfs_bmbt_irec_t *new, /* new data to add to file extents */
101 xfs_fsblock_t *first, /* pointer to firstblock variable */
102 xfs_bmap_free_t *flist, /* list of extents to be freed */
103 int *logflagsp, /* inode logging flags */
104 int whichfork, /* data or attr fork */
105 int rsvd); /* OK to allocate reserved blocks */
106
107 /*
108 * Called by xfs_bmap_add_extent to handle cases converting a delayed
109 * allocation to a real allocation.
110 */
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 add to file extents */
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 */
122
123 /*
124 * Called by xfs_bmap_add_extent to handle cases converting a hole
125 * to a delayed allocation.
126 */
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_bmbt_irec_t *new, /* new data to add to file extents */
132 int *logflagsp,/* inode logging flags */
133 int rsvd); /* OK to allocate reserved blocks */
134
135 /*
136 * Called by xfs_bmap_add_extent to handle cases converting a hole
137 * to a real allocation.
138 */
139 STATIC int /* error */
140 xfs_bmap_add_extent_hole_real(
141 xfs_inode_t *ip, /* incore inode pointer */
142 xfs_extnum_t idx, /* extent number to update/insert */
143 xfs_btree_cur_t *cur, /* if null, not a btree */
144 xfs_bmbt_irec_t *new, /* new data to add to file extents */
145 int *logflagsp, /* inode logging flags */
146 int whichfork); /* data or attr fork */
147
148 /*
149 * Called by xfs_bmap_add_extent to handle cases converting an unwritten
150 * allocation to a real allocation or vice versa.
151 */
152 STATIC int /* error */
153 xfs_bmap_add_extent_unwritten_real(
154 xfs_inode_t *ip, /* incore inode pointer */
155 xfs_extnum_t idx, /* extent number to update/insert */
156 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
157 xfs_bmbt_irec_t *new, /* new data to add to file extents */
158 int *logflagsp); /* inode logging flags */
159
160 /*
161 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
162 * It figures out where to ask the underlying allocator to put the new extent.
163 */
164 STATIC int /* error */
165 xfs_bmap_alloc(
166 xfs_bmalloca_t *ap); /* bmap alloc argument struct */
167
168 /*
169 * Transform a btree format file with only one leaf node, where the
170 * extents list will fit in the inode, into an extents format file.
171 * Since the file extents are already in-core, all we have to do is
172 * give up the space for the btree root and pitch the leaf block.
173 */
174 STATIC int /* error */
175 xfs_bmap_btree_to_extents(
176 xfs_trans_t *tp, /* transaction pointer */
177 xfs_inode_t *ip, /* incore inode pointer */
178 xfs_btree_cur_t *cur, /* btree cursor */
179 int *logflagsp, /* inode logging flags */
180 int whichfork); /* data or attr fork */
181
182 /*
183 * Called by xfs_bmapi to update file extent records and the btree
184 * after removing space (or undoing a delayed allocation).
185 */
186 STATIC int /* error */
187 xfs_bmap_del_extent(
188 xfs_inode_t *ip, /* incore inode pointer */
189 xfs_trans_t *tp, /* current trans pointer */
190 xfs_extnum_t idx, /* extent number to update/insert */
191 xfs_bmap_free_t *flist, /* list of extents to be freed */
192 xfs_btree_cur_t *cur, /* if null, not a btree */
193 xfs_bmbt_irec_t *new, /* new data to add to file extents */
194 int *logflagsp,/* inode logging flags */
195 int whichfork, /* data or attr fork */
196 int rsvd); /* OK to allocate reserved blocks */
197
198 /*
199 * Remove the entry "free" from the free item list. Prev points to the
200 * previous entry, unless "free" is the head of the list.
201 */
202 STATIC void
203 xfs_bmap_del_free(
204 xfs_bmap_free_t *flist, /* free item list header */
205 xfs_bmap_free_item_t *prev, /* previous item on list, if any */
206 xfs_bmap_free_item_t *free); /* list item to be freed */
207
208 /*
209 * Convert an extents-format file into a btree-format file.
210 * The new file will have a root block (in the inode) and a single child block.
211 */
212 STATIC int /* error */
213 xfs_bmap_extents_to_btree(
214 xfs_trans_t *tp, /* transaction pointer */
215 xfs_inode_t *ip, /* incore inode pointer */
216 xfs_fsblock_t *firstblock, /* first-block-allocated */
217 xfs_bmap_free_t *flist, /* blocks freed in xaction */
218 xfs_btree_cur_t **curp, /* cursor returned to caller */
219 int wasdel, /* converting a delayed alloc */
220 int *logflagsp, /* inode logging flags */
221 int whichfork); /* data or attr fork */
222
223 /*
224 * Convert a local file to an extents file.
225 * This code is sort of bogus, since the file data needs to get
226 * logged so it won't be lost. The bmap-level manipulations are ok, though.
227 */
228 STATIC int /* error */
229 xfs_bmap_local_to_extents(
230 xfs_trans_t *tp, /* transaction pointer */
231 xfs_inode_t *ip, /* incore inode pointer */
232 xfs_fsblock_t *firstblock, /* first block allocated in xaction */
233 xfs_extlen_t total, /* total blocks needed by transaction */
234 int *logflagsp, /* inode logging flags */
235 int whichfork); /* data or attr fork */
236
237 /*
238 * Search the extents list for the inode, for the extent containing bno.
239 * If bno lies in a hole, point to the next entry. If bno lies past eof,
240 * *eofp will be set, and *prevp will contain the last entry (null if none).
241 * Else, *lastxp will be set to the index of the found
242 * entry; *gotp will contain the entry.
243 */
244 STATIC xfs_bmbt_rec_host_t * /* pointer to found extent entry */
245 xfs_bmap_search_extents(
246 xfs_inode_t *ip, /* incore inode pointer */
247 xfs_fileoff_t bno, /* block number searched for */
248 int whichfork, /* data or attr fork */
249 int *eofp, /* out: end of file found */
250 xfs_extnum_t *lastxp, /* out: last extent index */
251 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
252 xfs_bmbt_irec_t *prevp); /* out: previous extent entry found */
253
254 /*
255 * Check the last inode extent to determine whether this allocation will result
256 * in blocks being allocated at the end of the file. When we allocate new data
257 * blocks at the end of the file which do not start at the previous data block,
258 * we will try to align the new blocks at stripe unit boundaries.
259 */
260 STATIC int /* error */
261 xfs_bmap_isaeof(
262 xfs_inode_t *ip, /* incore inode pointer */
263 xfs_fileoff_t off, /* file offset in fsblocks */
264 int whichfork, /* data or attribute fork */
265 char *aeof); /* return value */
266
267 /*
268 * Compute the worst-case number of indirect blocks that will be used
269 * for ip's delayed extent of length "len".
270 */
271 STATIC xfs_filblks_t
272 xfs_bmap_worst_indlen(
273 xfs_inode_t *ip, /* incore inode pointer */
274 xfs_filblks_t len); /* delayed extent length */
275
276 #ifdef DEBUG
277 /*
278 * Perform various validation checks on the values being returned
279 * from xfs_bmapi().
280 */
281 STATIC void
282 xfs_bmap_validate_ret(
283 xfs_fileoff_t bno,
284 xfs_filblks_t len,
285 int flags,
286 xfs_bmbt_irec_t *mval,
287 int nmap,
288 int ret_nmap);
289 #else
290 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
291 #endif /* DEBUG */
292
293 STATIC int
294 xfs_bmap_count_tree(
295 xfs_mount_t *mp,
296 xfs_trans_t *tp,
297 xfs_ifork_t *ifp,
298 xfs_fsblock_t blockno,
299 int levelin,
300 int *count);
301
302 STATIC void
303 xfs_bmap_count_leaves(
304 xfs_ifork_t *ifp,
305 xfs_extnum_t idx,
306 int numrecs,
307 int *count);
308
309 STATIC void
310 xfs_bmap_disk_count_leaves(
311 struct xfs_mount *mp,
312 struct xfs_btree_block *block,
313 int numrecs,
314 int *count);
315
316 /*
317 * Bmap internal routines.
318 */
319
320 STATIC int /* error */
321 xfs_bmbt_lookup_eq(
322 struct xfs_btree_cur *cur,
323 xfs_fileoff_t off,
324 xfs_fsblock_t bno,
325 xfs_filblks_t len,
326 int *stat) /* success/failure */
327 {
328 cur->bc_rec.b.br_startoff = off;
329 cur->bc_rec.b.br_startblock = bno;
330 cur->bc_rec.b.br_blockcount = len;
331 return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
332 }
333
334 STATIC int /* error */
335 xfs_bmbt_lookup_ge(
336 struct xfs_btree_cur *cur,
337 xfs_fileoff_t off,
338 xfs_fsblock_t bno,
339 xfs_filblks_t len,
340 int *stat) /* success/failure */
341 {
342 cur->bc_rec.b.br_startoff = off;
343 cur->bc_rec.b.br_startblock = bno;
344 cur->bc_rec.b.br_blockcount = len;
345 return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
346 }
347
348 /*
349 * Update the record referred to by cur to the value given
350 * by [off, bno, len, state].
351 * This either works (return 0) or gets an EFSCORRUPTED error.
352 */
353 STATIC int
354 xfs_bmbt_update(
355 struct xfs_btree_cur *cur,
356 xfs_fileoff_t off,
357 xfs_fsblock_t bno,
358 xfs_filblks_t len,
359 xfs_exntst_t state)
360 {
361 union xfs_btree_rec rec;
362
363 xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
364 return xfs_btree_update(cur, &rec);
365 }
366
367 /*
368 * Called from xfs_bmap_add_attrfork to handle btree format files.
369 */
370 STATIC int /* error */
371 xfs_bmap_add_attrfork_btree(
372 xfs_trans_t *tp, /* transaction pointer */
373 xfs_inode_t *ip, /* incore inode pointer */
374 xfs_fsblock_t *firstblock, /* first block allocated */
375 xfs_bmap_free_t *flist, /* blocks to free at commit */
376 int *flags) /* inode logging flags */
377 {
378 xfs_btree_cur_t *cur; /* btree cursor */
379 int error; /* error return value */
380 xfs_mount_t *mp; /* file system mount struct */
381 int stat; /* newroot status */
382
383 mp = ip->i_mount;
384 if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
385 *flags |= XFS_ILOG_DBROOT;
386 else {
387 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
388 cur->bc_private.b.flist = flist;
389 cur->bc_private.b.firstblock = *firstblock;
390 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
391 goto error0;
392 /* must be at least one entry */
393 XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
394 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
395 goto error0;
396 if (stat == 0) {
397 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
398 return XFS_ERROR(ENOSPC);
399 }
400 *firstblock = cur->bc_private.b.firstblock;
401 cur->bc_private.b.allocated = 0;
402 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
403 }
404 return 0;
405 error0:
406 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
407 return error;
408 }
409
410 /*
411 * Called from xfs_bmap_add_attrfork to handle extents format files.
412 */
413 STATIC int /* error */
414 xfs_bmap_add_attrfork_extents(
415 xfs_trans_t *tp, /* transaction pointer */
416 xfs_inode_t *ip, /* incore inode pointer */
417 xfs_fsblock_t *firstblock, /* first block allocated */
418 xfs_bmap_free_t *flist, /* blocks to free at commit */
419 int *flags) /* inode logging flags */
420 {
421 xfs_btree_cur_t *cur; /* bmap btree cursor */
422 int error; /* error return value */
423
424 if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
425 return 0;
426 cur = NULL;
427 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
428 flags, XFS_DATA_FORK);
429 if (cur) {
430 cur->bc_private.b.allocated = 0;
431 xfs_btree_del_cursor(cur,
432 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
433 }
434 return error;
435 }
436
437 /*
438 * Called from xfs_bmap_add_attrfork to handle local format files.
439 */
440 STATIC int /* error */
441 xfs_bmap_add_attrfork_local(
442 xfs_trans_t *tp, /* transaction pointer */
443 xfs_inode_t *ip, /* incore inode pointer */
444 xfs_fsblock_t *firstblock, /* first block allocated */
445 xfs_bmap_free_t *flist, /* blocks to free at commit */
446 int *flags) /* inode logging flags */
447 {
448 xfs_da_args_t dargs; /* args for dir/attr code */
449 int error; /* error return value */
450 xfs_mount_t *mp; /* mount structure pointer */
451
452 if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
453 return 0;
454 if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
455 mp = ip->i_mount;
456 memset(&dargs, 0, sizeof(dargs));
457 dargs.dp = ip;
458 dargs.firstblock = firstblock;
459 dargs.flist = flist;
460 dargs.total = mp->m_dirblkfsbs;
461 dargs.whichfork = XFS_DATA_FORK;
462 dargs.trans = tp;
463 error = xfs_dir2_sf_to_block(&dargs);
464 } else
465 error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
466 XFS_DATA_FORK);
467 return error;
468 }
469
470 /*
471 * Called by xfs_bmapi to update file extent records and the btree
472 * after allocating space (or doing a delayed allocation).
473 */
474 STATIC int /* error */
475 xfs_bmap_add_extent(
476 xfs_inode_t *ip, /* incore inode pointer */
477 xfs_extnum_t idx, /* extent number to update/insert */
478 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
479 xfs_bmbt_irec_t *new, /* new data to add to file extents */
480 xfs_fsblock_t *first, /* pointer to firstblock variable */
481 xfs_bmap_free_t *flist, /* list of extents to be freed */
482 int *logflagsp, /* inode logging flags */
483 int whichfork, /* data or attr fork */
484 int rsvd) /* OK to use reserved data blocks */
485 {
486 xfs_btree_cur_t *cur; /* btree cursor or null */
487 xfs_filblks_t da_new; /* new count del alloc blocks used */
488 xfs_filblks_t da_old; /* old count del alloc blocks used */
489 int error; /* error return value */
490 xfs_ifork_t *ifp; /* inode fork ptr */
491 int logflags; /* returned value */
492 xfs_extnum_t nextents; /* number of extents in file now */
493
494 XFS_STATS_INC(xs_add_exlist);
495 cur = *curp;
496 ifp = XFS_IFORK_PTR(ip, whichfork);
497 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
498 ASSERT(idx <= nextents);
499 da_old = da_new = 0;
500 error = 0;
501 /*
502 * This is the first extent added to a new/empty file.
503 * Special case this one, so other routines get to assume there are
504 * already extents in the list.
505 */
506 if (nextents == 0) {
507 xfs_iext_insert(ip, 0, 1, new,
508 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
509
510 ASSERT(cur == NULL);
511 ifp->if_lastex = 0;
512 if (!isnullstartblock(new->br_startblock)) {
513 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
514 logflags = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
515 } else
516 logflags = 0;
517 }
518 /*
519 * Any kind of new delayed allocation goes here.
520 */
521 else if (isnullstartblock(new->br_startblock)) {
522 if (cur)
523 ASSERT((cur->bc_private.b.flags &
524 XFS_BTCUR_BPRV_WASDEL) == 0);
525 if ((error = xfs_bmap_add_extent_hole_delay(ip, idx, new,
526 &logflags, rsvd)))
527 goto done;
528 }
529 /*
530 * Real allocation off the end of the file.
531 */
532 else if (idx == nextents) {
533 if (cur)
534 ASSERT((cur->bc_private.b.flags &
535 XFS_BTCUR_BPRV_WASDEL) == 0);
536 if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur, new,
537 &logflags, whichfork)))
538 goto done;
539 } else {
540 xfs_bmbt_irec_t prev; /* old extent at offset idx */
541
542 /*
543 * Get the record referred to by idx.
544 */
545 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx), &prev);
546 /*
547 * If it's a real allocation record, and the new allocation ends
548 * after the start of the referred to record, then we're filling
549 * in a delayed or unwritten allocation with a real one, or
550 * converting real back to unwritten.
551 */
552 if (!isnullstartblock(new->br_startblock) &&
553 new->br_startoff + new->br_blockcount > prev.br_startoff) {
554 if (prev.br_state != XFS_EXT_UNWRITTEN &&
555 isnullstartblock(prev.br_startblock)) {
556 da_old = startblockval(prev.br_startblock);
557 if (cur)
558 ASSERT(cur->bc_private.b.flags &
559 XFS_BTCUR_BPRV_WASDEL);
560 if ((error = xfs_bmap_add_extent_delay_real(ip,
561 idx, &cur, new, &da_new, first, flist,
562 &logflags, rsvd)))
563 goto done;
564 } else if (new->br_state == XFS_EXT_NORM) {
565 ASSERT(new->br_state == XFS_EXT_NORM);
566 if ((error = xfs_bmap_add_extent_unwritten_real(
567 ip, idx, &cur, new, &logflags)))
568 goto done;
569 } else {
570 ASSERT(new->br_state == XFS_EXT_UNWRITTEN);
571 if ((error = xfs_bmap_add_extent_unwritten_real(
572 ip, idx, &cur, new, &logflags)))
573 goto done;
574 }
575 ASSERT(*curp == cur || *curp == NULL);
576 }
577 /*
578 * Otherwise we're filling in a hole with an allocation.
579 */
580 else {
581 if (cur)
582 ASSERT((cur->bc_private.b.flags &
583 XFS_BTCUR_BPRV_WASDEL) == 0);
584 if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur,
585 new, &logflags, whichfork)))
586 goto done;
587 }
588 }
589
590 ASSERT(*curp == cur || *curp == NULL);
591 /*
592 * Convert to a btree if necessary.
593 */
594 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
595 XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
596 int tmp_logflags; /* partial log flag return val */
597
598 ASSERT(cur == NULL);
599 error = xfs_bmap_extents_to_btree(ip->i_transp, ip, first,
600 flist, &cur, da_old > 0, &tmp_logflags, whichfork);
601 logflags |= tmp_logflags;
602 if (error)
603 goto done;
604 }
605 /*
606 * Adjust for changes in reserved delayed indirect blocks.
607 * Nothing to do for disk quotas here.
608 */
609 if (da_old || da_new) {
610 xfs_filblks_t nblks;
611
612 nblks = da_new;
613 if (cur)
614 nblks += cur->bc_private.b.allocated;
615 ASSERT(nblks <= da_old);
616 if (nblks < da_old)
617 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
618 (int64_t)(da_old - nblks), rsvd);
619 }
620 /*
621 * Clear out the allocated field, done with it now in any case.
622 */
623 if (cur) {
624 cur->bc_private.b.allocated = 0;
625 *curp = cur;
626 }
627 done:
628 #ifdef DEBUG
629 if (!error)
630 xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
631 #endif
632 *logflagsp = logflags;
633 return error;
634 }
635
636 /*
637 * Called by xfs_bmap_add_extent to handle cases converting a delayed
638 * allocation to a real allocation.
639 */
640 STATIC int /* error */
641 xfs_bmap_add_extent_delay_real(
642 xfs_inode_t *ip, /* incore inode pointer */
643 xfs_extnum_t idx, /* extent number to update/insert */
644 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
645 xfs_bmbt_irec_t *new, /* new data to add to file extents */
646 xfs_filblks_t *dnew, /* new delayed-alloc indirect blocks */
647 xfs_fsblock_t *first, /* pointer to firstblock variable */
648 xfs_bmap_free_t *flist, /* list of extents to be freed */
649 int *logflagsp, /* inode logging flags */
650 int rsvd) /* OK to use reserved data block allocation */
651 {
652 xfs_btree_cur_t *cur; /* btree cursor */
653 int diff; /* temp value */
654 xfs_bmbt_rec_host_t *ep; /* extent entry for idx */
655 int error; /* error return value */
656 int i; /* temp state */
657 xfs_ifork_t *ifp; /* inode fork pointer */
658 xfs_fileoff_t new_endoff; /* end offset of new entry */
659 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
660 /* left is 0, right is 1, prev is 2 */
661 int rval=0; /* return value (logging flags) */
662 int state = 0;/* state bits, accessed thru macros */
663 xfs_filblks_t temp=0; /* value for dnew calculations */
664 xfs_filblks_t temp2=0;/* value for dnew calculations */
665 int tmp_rval; /* partial logging flags */
666
667 #define LEFT r[0]
668 #define RIGHT r[1]
669 #define PREV r[2]
670
671 /*
672 * Set up a bunch of variables to make the tests simpler.
673 */
674 cur = *curp;
675 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
676 ep = xfs_iext_get_ext(ifp, idx);
677 xfs_bmbt_get_all(ep, &PREV);
678 new_endoff = new->br_startoff + new->br_blockcount;
679 ASSERT(PREV.br_startoff <= new->br_startoff);
680 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
681
682 /*
683 * Set flags determining what part of the previous delayed allocation
684 * extent is being replaced by a real allocation.
685 */
686 if (PREV.br_startoff == new->br_startoff)
687 state |= BMAP_LEFT_FILLING;
688 if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
689 state |= BMAP_RIGHT_FILLING;
690
691 /*
692 * Check and set flags if this segment has a left neighbor.
693 * Don't set contiguous if the combined extent would be too large.
694 */
695 if (idx > 0) {
696 state |= BMAP_LEFT_VALID;
697 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &LEFT);
698
699 if (isnullstartblock(LEFT.br_startblock))
700 state |= BMAP_LEFT_DELAY;
701 }
702
703 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
704 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
705 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
706 LEFT.br_state == new->br_state &&
707 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
708 state |= BMAP_LEFT_CONTIG;
709
710 /*
711 * Check and set flags if this segment has a right neighbor.
712 * Don't set contiguous if the combined extent would be too large.
713 * Also check for all-three-contiguous being too large.
714 */
715 if (idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
716 state |= BMAP_RIGHT_VALID;
717 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx + 1), &RIGHT);
718
719 if (isnullstartblock(RIGHT.br_startblock))
720 state |= BMAP_RIGHT_DELAY;
721 }
722
723 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
724 new_endoff == RIGHT.br_startoff &&
725 new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
726 new->br_state == RIGHT.br_state &&
727 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
728 ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
729 BMAP_RIGHT_FILLING)) !=
730 (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
731 BMAP_RIGHT_FILLING) ||
732 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
733 <= MAXEXTLEN))
734 state |= BMAP_RIGHT_CONTIG;
735
736 error = 0;
737 /*
738 * Switch out based on the FILLING and CONTIG state bits.
739 */
740 switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
741 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
742 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
743 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
744 /*
745 * Filling in all of a previously delayed allocation extent.
746 * The left and right neighbors are both contiguous with new.
747 */
748 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
749 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
750 LEFT.br_blockcount + PREV.br_blockcount +
751 RIGHT.br_blockcount);
752 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
753
754 xfs_iext_remove(ip, idx, 2, state);
755 ip->i_df.if_lastex = idx - 1;
756 ip->i_d.di_nextents--;
757 if (cur == NULL)
758 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
759 else {
760 rval = XFS_ILOG_CORE;
761 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
762 RIGHT.br_startblock,
763 RIGHT.br_blockcount, &i)))
764 goto done;
765 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
766 if ((error = xfs_btree_delete(cur, &i)))
767 goto done;
768 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
769 if ((error = xfs_btree_decrement(cur, 0, &i)))
770 goto done;
771 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
772 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
773 LEFT.br_startblock,
774 LEFT.br_blockcount +
775 PREV.br_blockcount +
776 RIGHT.br_blockcount, LEFT.br_state)))
777 goto done;
778 }
779 *dnew = 0;
780 break;
781
782 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
783 /*
784 * Filling in all of a previously delayed allocation extent.
785 * The left neighbor is contiguous, the right is not.
786 */
787 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
788 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
789 LEFT.br_blockcount + PREV.br_blockcount);
790 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
791
792 ip->i_df.if_lastex = idx - 1;
793 xfs_iext_remove(ip, idx, 1, state);
794 if (cur == NULL)
795 rval = XFS_ILOG_DEXT;
796 else {
797 rval = 0;
798 if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
799 LEFT.br_startblock, LEFT.br_blockcount,
800 &i)))
801 goto done;
802 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
803 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
804 LEFT.br_startblock,
805 LEFT.br_blockcount +
806 PREV.br_blockcount, LEFT.br_state)))
807 goto done;
808 }
809 *dnew = 0;
810 break;
811
812 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
813 /*
814 * Filling in all of a previously delayed allocation extent.
815 * The right neighbor is contiguous, the left is not.
816 */
817 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
818 xfs_bmbt_set_startblock(ep, new->br_startblock);
819 xfs_bmbt_set_blockcount(ep,
820 PREV.br_blockcount + RIGHT.br_blockcount);
821 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
822
823 ip->i_df.if_lastex = idx;
824 xfs_iext_remove(ip, idx + 1, 1, state);
825 if (cur == NULL)
826 rval = XFS_ILOG_DEXT;
827 else {
828 rval = 0;
829 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
830 RIGHT.br_startblock,
831 RIGHT.br_blockcount, &i)))
832 goto done;
833 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
834 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
835 new->br_startblock,
836 PREV.br_blockcount +
837 RIGHT.br_blockcount, PREV.br_state)))
838 goto done;
839 }
840 *dnew = 0;
841 break;
842
843 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
844 /*
845 * Filling in all of a previously delayed allocation extent.
846 * Neither the left nor right neighbors are contiguous with
847 * the new one.
848 */
849 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
850 xfs_bmbt_set_startblock(ep, new->br_startblock);
851 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
852
853 ip->i_df.if_lastex = idx;
854 ip->i_d.di_nextents++;
855 if (cur == NULL)
856 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
857 else {
858 rval = XFS_ILOG_CORE;
859 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
860 new->br_startblock, new->br_blockcount,
861 &i)))
862 goto done;
863 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
864 cur->bc_rec.b.br_state = XFS_EXT_NORM;
865 if ((error = xfs_btree_insert(cur, &i)))
866 goto done;
867 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
868 }
869 *dnew = 0;
870 break;
871
872 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
873 /*
874 * Filling in the first part of a previous delayed allocation.
875 * The left neighbor is contiguous.
876 */
877 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
878 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
879 LEFT.br_blockcount + new->br_blockcount);
880 xfs_bmbt_set_startoff(ep,
881 PREV.br_startoff + new->br_blockcount);
882 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
883
884 temp = PREV.br_blockcount - new->br_blockcount;
885 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
886 xfs_bmbt_set_blockcount(ep, temp);
887 ip->i_df.if_lastex = idx - 1;
888 if (cur == NULL)
889 rval = XFS_ILOG_DEXT;
890 else {
891 rval = 0;
892 if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
893 LEFT.br_startblock, LEFT.br_blockcount,
894 &i)))
895 goto done;
896 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
897 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
898 LEFT.br_startblock,
899 LEFT.br_blockcount +
900 new->br_blockcount,
901 LEFT.br_state)))
902 goto done;
903 }
904 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
905 startblockval(PREV.br_startblock));
906 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
907 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
908 *dnew = temp;
909 break;
910
911 case BMAP_LEFT_FILLING:
912 /*
913 * Filling in the first part of a previous delayed allocation.
914 * The left neighbor is not contiguous.
915 */
916 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
917 xfs_bmbt_set_startoff(ep, new_endoff);
918 temp = PREV.br_blockcount - new->br_blockcount;
919 xfs_bmbt_set_blockcount(ep, temp);
920 xfs_iext_insert(ip, idx, 1, new, state);
921 ip->i_df.if_lastex = idx;
922 ip->i_d.di_nextents++;
923 if (cur == NULL)
924 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
925 else {
926 rval = XFS_ILOG_CORE;
927 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
928 new->br_startblock, new->br_blockcount,
929 &i)))
930 goto done;
931 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
932 cur->bc_rec.b.br_state = XFS_EXT_NORM;
933 if ((error = xfs_btree_insert(cur, &i)))
934 goto done;
935 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
936 }
937 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
938 ip->i_d.di_nextents > ip->i_df.if_ext_max) {
939 error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
940 first, flist, &cur, 1, &tmp_rval,
941 XFS_DATA_FORK);
942 rval |= tmp_rval;
943 if (error)
944 goto done;
945 }
946 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
947 startblockval(PREV.br_startblock) -
948 (cur ? cur->bc_private.b.allocated : 0));
949 ep = xfs_iext_get_ext(ifp, idx + 1);
950 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
951 trace_xfs_bmap_post_update(ip, idx + 1, state, _THIS_IP_);
952 *dnew = temp;
953 break;
954
955 case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
956 /*
957 * Filling in the last part of a previous delayed allocation.
958 * The right neighbor is contiguous with the new allocation.
959 */
960 temp = PREV.br_blockcount - new->br_blockcount;
961 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
962 trace_xfs_bmap_pre_update(ip, idx + 1, state, _THIS_IP_);
963 xfs_bmbt_set_blockcount(ep, temp);
964 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, idx + 1),
965 new->br_startoff, new->br_startblock,
966 new->br_blockcount + RIGHT.br_blockcount,
967 RIGHT.br_state);
968 trace_xfs_bmap_post_update(ip, idx + 1, state, _THIS_IP_);
969 ip->i_df.if_lastex = idx + 1;
970 if (cur == NULL)
971 rval = XFS_ILOG_DEXT;
972 else {
973 rval = 0;
974 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
975 RIGHT.br_startblock,
976 RIGHT.br_blockcount, &i)))
977 goto done;
978 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
979 if ((error = xfs_bmbt_update(cur, new->br_startoff,
980 new->br_startblock,
981 new->br_blockcount +
982 RIGHT.br_blockcount,
983 RIGHT.br_state)))
984 goto done;
985 }
986 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
987 startblockval(PREV.br_startblock));
988 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
989 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
990 *dnew = temp;
991 break;
992
993 case BMAP_RIGHT_FILLING:
994 /*
995 * Filling in the last part of a previous delayed allocation.
996 * The right neighbor is not contiguous.
997 */
998 temp = PREV.br_blockcount - new->br_blockcount;
999 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1000 xfs_bmbt_set_blockcount(ep, temp);
1001 xfs_iext_insert(ip, idx + 1, 1, new, state);
1002 ip->i_df.if_lastex = idx + 1;
1003 ip->i_d.di_nextents++;
1004 if (cur == NULL)
1005 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1006 else {
1007 rval = XFS_ILOG_CORE;
1008 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1009 new->br_startblock, new->br_blockcount,
1010 &i)))
1011 goto done;
1012 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1013 cur->bc_rec.b.br_state = XFS_EXT_NORM;
1014 if ((error = xfs_btree_insert(cur, &i)))
1015 goto done;
1016 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1017 }
1018 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1019 ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1020 error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1021 first, flist, &cur, 1, &tmp_rval,
1022 XFS_DATA_FORK);
1023 rval |= tmp_rval;
1024 if (error)
1025 goto done;
1026 }
1027 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1028 startblockval(PREV.br_startblock) -
1029 (cur ? cur->bc_private.b.allocated : 0));
1030 ep = xfs_iext_get_ext(ifp, idx);
1031 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
1032 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1033 *dnew = temp;
1034 break;
1035
1036 case 0:
1037 /*
1038 * Filling in the middle part of a previous delayed allocation.
1039 * Contiguity is impossible here.
1040 * This case is avoided almost all the time.
1041 *
1042 * We start with a delayed allocation:
1043 *
1044 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
1045 * PREV @ idx
1046 *
1047 * and we are allocating:
1048 * +rrrrrrrrrrrrrrrrr+
1049 * new
1050 *
1051 * and we set it up for insertion as:
1052 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
1053 * new
1054 * PREV @ idx LEFT RIGHT
1055 * inserted at idx + 1
1056 */
1057 temp = new->br_startoff - PREV.br_startoff;
1058 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
1059 trace_xfs_bmap_pre_update(ip, idx, 0, _THIS_IP_);
1060 xfs_bmbt_set_blockcount(ep, temp); /* truncate PREV */
1061 LEFT = *new;
1062 RIGHT.br_state = PREV.br_state;
1063 RIGHT.br_startblock = nullstartblock(
1064 (int)xfs_bmap_worst_indlen(ip, temp2));
1065 RIGHT.br_startoff = new_endoff;
1066 RIGHT.br_blockcount = temp2;
1067 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
1068 xfs_iext_insert(ip, idx + 1, 2, &LEFT, state);
1069 ip->i_df.if_lastex = idx + 1;
1070 ip->i_d.di_nextents++;
1071 if (cur == NULL)
1072 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1073 else {
1074 rval = XFS_ILOG_CORE;
1075 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1076 new->br_startblock, new->br_blockcount,
1077 &i)))
1078 goto done;
1079 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1080 cur->bc_rec.b.br_state = XFS_EXT_NORM;
1081 if ((error = xfs_btree_insert(cur, &i)))
1082 goto done;
1083 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1084 }
1085 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1086 ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1087 error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1088 first, flist, &cur, 1, &tmp_rval,
1089 XFS_DATA_FORK);
1090 rval |= tmp_rval;
1091 if (error)
1092 goto done;
1093 }
1094 temp = xfs_bmap_worst_indlen(ip, temp);
1095 temp2 = xfs_bmap_worst_indlen(ip, temp2);
1096 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
1097 (cur ? cur->bc_private.b.allocated : 0));
1098 if (diff > 0 &&
1099 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
1100 -((int64_t)diff), rsvd)) {
1101 /*
1102 * Ick gross gag me with a spoon.
1103 */
1104 ASSERT(0); /* want to see if this ever happens! */
1105 while (diff > 0) {
1106 if (temp) {
1107 temp--;
1108 diff--;
1109 if (!diff ||
1110 !xfs_icsb_modify_counters(ip->i_mount,
1111 XFS_SBS_FDBLOCKS,
1112 -((int64_t)diff), rsvd))
1113 break;
1114 }
1115 if (temp2) {
1116 temp2--;
1117 diff--;
1118 if (!diff ||
1119 !xfs_icsb_modify_counters(ip->i_mount,
1120 XFS_SBS_FDBLOCKS,
1121 -((int64_t)diff), rsvd))
1122 break;
1123 }
1124 }
1125 }
1126 ep = xfs_iext_get_ext(ifp, idx);
1127 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
1128 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1129 trace_xfs_bmap_pre_update(ip, idx + 2, state, _THIS_IP_);
1130 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx + 2),
1131 nullstartblock((int)temp2));
1132 trace_xfs_bmap_post_update(ip, idx + 2, state, _THIS_IP_);
1133 *dnew = temp + temp2;
1134 break;
1135
1136 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1137 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1138 case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1139 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1140 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1141 case BMAP_LEFT_CONTIG:
1142 case BMAP_RIGHT_CONTIG:
1143 /*
1144 * These cases are all impossible.
1145 */
1146 ASSERT(0);
1147 }
1148 *curp = cur;
1149 done:
1150 *logflagsp = rval;
1151 return error;
1152 #undef LEFT
1153 #undef RIGHT
1154 #undef PREV
1155 }
1156
1157 /*
1158 * Called by xfs_bmap_add_extent to handle cases converting an unwritten
1159 * allocation to a real allocation or vice versa.
1160 */
1161 STATIC int /* error */
1162 xfs_bmap_add_extent_unwritten_real(
1163 xfs_inode_t *ip, /* incore inode pointer */
1164 xfs_extnum_t idx, /* extent number to update/insert */
1165 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
1166 xfs_bmbt_irec_t *new, /* new data to add to file extents */
1167 int *logflagsp) /* inode logging flags */
1168 {
1169 xfs_btree_cur_t *cur; /* btree cursor */
1170 xfs_bmbt_rec_host_t *ep; /* extent entry for idx */
1171 int error; /* error return value */
1172 int i; /* temp state */
1173 xfs_ifork_t *ifp; /* inode fork pointer */
1174 xfs_fileoff_t new_endoff; /* end offset of new entry */
1175 xfs_exntst_t newext; /* new extent state */
1176 xfs_exntst_t oldext; /* old extent state */
1177 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
1178 /* left is 0, right is 1, prev is 2 */
1179 int rval=0; /* return value (logging flags) */
1180 int state = 0;/* state bits, accessed thru macros */
1181
1182 #define LEFT r[0]
1183 #define RIGHT r[1]
1184 #define PREV r[2]
1185 /*
1186 * Set up a bunch of variables to make the tests simpler.
1187 */
1188 error = 0;
1189 cur = *curp;
1190 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1191 ep = xfs_iext_get_ext(ifp, idx);
1192 xfs_bmbt_get_all(ep, &PREV);
1193 newext = new->br_state;
1194 oldext = (newext == XFS_EXT_UNWRITTEN) ?
1195 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
1196 ASSERT(PREV.br_state == oldext);
1197 new_endoff = new->br_startoff + new->br_blockcount;
1198 ASSERT(PREV.br_startoff <= new->br_startoff);
1199 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1200
1201 /*
1202 * Set flags determining what part of the previous oldext allocation
1203 * extent is being replaced by a newext allocation.
1204 */
1205 if (PREV.br_startoff == new->br_startoff)
1206 state |= BMAP_LEFT_FILLING;
1207 if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1208 state |= BMAP_RIGHT_FILLING;
1209
1210 /*
1211 * Check and set flags if this segment has a left neighbor.
1212 * Don't set contiguous if the combined extent would be too large.
1213 */
1214 if (idx > 0) {
1215 state |= BMAP_LEFT_VALID;
1216 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &LEFT);
1217
1218 if (isnullstartblock(LEFT.br_startblock))
1219 state |= BMAP_LEFT_DELAY;
1220 }
1221
1222 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1223 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1224 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1225 LEFT.br_state == newext &&
1226 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1227 state |= BMAP_LEFT_CONTIG;
1228
1229 /*
1230 * Check and set flags if this segment has a right neighbor.
1231 * Don't set contiguous if the combined extent would be too large.
1232 * Also check for all-three-contiguous being too large.
1233 */
1234 if (idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1235 state |= BMAP_RIGHT_VALID;
1236 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx + 1), &RIGHT);
1237 if (isnullstartblock(RIGHT.br_startblock))
1238 state |= BMAP_RIGHT_DELAY;
1239 }
1240
1241 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1242 new_endoff == RIGHT.br_startoff &&
1243 new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1244 newext == RIGHT.br_state &&
1245 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1246 ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1247 BMAP_RIGHT_FILLING)) !=
1248 (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1249 BMAP_RIGHT_FILLING) ||
1250 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1251 <= MAXEXTLEN))
1252 state |= BMAP_RIGHT_CONTIG;
1253
1254 /*
1255 * Switch out based on the FILLING and CONTIG state bits.
1256 */
1257 switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1258 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1259 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1260 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1261 /*
1262 * Setting all of a previous oldext extent to newext.
1263 * The left and right neighbors are both contiguous with new.
1264 */
1265 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
1266 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1267 LEFT.br_blockcount + PREV.br_blockcount +
1268 RIGHT.br_blockcount);
1269 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
1270
1271 xfs_iext_remove(ip, idx, 2, state);
1272 ip->i_df.if_lastex = idx - 1;
1273 ip->i_d.di_nextents -= 2;
1274 if (cur == NULL)
1275 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1276 else {
1277 rval = XFS_ILOG_CORE;
1278 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1279 RIGHT.br_startblock,
1280 RIGHT.br_blockcount, &i)))
1281 goto done;
1282 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1283 if ((error = xfs_btree_delete(cur, &i)))
1284 goto done;
1285 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1286 if ((error = xfs_btree_decrement(cur, 0, &i)))
1287 goto done;
1288 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1289 if ((error = xfs_btree_delete(cur, &i)))
1290 goto done;
1291 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1292 if ((error = xfs_btree_decrement(cur, 0, &i)))
1293 goto done;
1294 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1295 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1296 LEFT.br_startblock,
1297 LEFT.br_blockcount + PREV.br_blockcount +
1298 RIGHT.br_blockcount, LEFT.br_state)))
1299 goto done;
1300 }
1301 break;
1302
1303 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1304 /*
1305 * Setting all of a previous oldext extent to newext.
1306 * The left neighbor is contiguous, the right is not.
1307 */
1308 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
1309 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1310 LEFT.br_blockcount + PREV.br_blockcount);
1311 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
1312
1313 ip->i_df.if_lastex = idx - 1;
1314 xfs_iext_remove(ip, idx, 1, state);
1315 ip->i_d.di_nextents--;
1316 if (cur == NULL)
1317 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1318 else {
1319 rval = XFS_ILOG_CORE;
1320 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1321 PREV.br_startblock, PREV.br_blockcount,
1322 &i)))
1323 goto done;
1324 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1325 if ((error = xfs_btree_delete(cur, &i)))
1326 goto done;
1327 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1328 if ((error = xfs_btree_decrement(cur, 0, &i)))
1329 goto done;
1330 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1331 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1332 LEFT.br_startblock,
1333 LEFT.br_blockcount + PREV.br_blockcount,
1334 LEFT.br_state)))
1335 goto done;
1336 }
1337 break;
1338
1339 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1340 /*
1341 * Setting all of a previous oldext extent to newext.
1342 * The right neighbor is contiguous, the left is not.
1343 */
1344 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1345 xfs_bmbt_set_blockcount(ep,
1346 PREV.br_blockcount + RIGHT.br_blockcount);
1347 xfs_bmbt_set_state(ep, newext);
1348 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1349 ip->i_df.if_lastex = idx;
1350 xfs_iext_remove(ip, idx + 1, 1, state);
1351 ip->i_d.di_nextents--;
1352 if (cur == NULL)
1353 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1354 else {
1355 rval = XFS_ILOG_CORE;
1356 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1357 RIGHT.br_startblock,
1358 RIGHT.br_blockcount, &i)))
1359 goto done;
1360 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1361 if ((error = xfs_btree_delete(cur, &i)))
1362 goto done;
1363 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1364 if ((error = xfs_btree_decrement(cur, 0, &i)))
1365 goto done;
1366 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1367 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1368 new->br_startblock,
1369 new->br_blockcount + RIGHT.br_blockcount,
1370 newext)))
1371 goto done;
1372 }
1373 break;
1374
1375 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1376 /*
1377 * Setting all of a previous oldext extent to newext.
1378 * Neither the left nor right neighbors are contiguous with
1379 * the new one.
1380 */
1381 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1382 xfs_bmbt_set_state(ep, newext);
1383 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1384
1385 ip->i_df.if_lastex = idx;
1386 if (cur == NULL)
1387 rval = XFS_ILOG_DEXT;
1388 else {
1389 rval = 0;
1390 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1391 new->br_startblock, new->br_blockcount,
1392 &i)))
1393 goto done;
1394 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1395 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1396 new->br_startblock, new->br_blockcount,
1397 newext)))
1398 goto done;
1399 }
1400 break;
1401
1402 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1403 /*
1404 * Setting the first part of a previous oldext extent to newext.
1405 * The left neighbor is contiguous.
1406 */
1407 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
1408 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1409 LEFT.br_blockcount + new->br_blockcount);
1410 xfs_bmbt_set_startoff(ep,
1411 PREV.br_startoff + new->br_blockcount);
1412 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
1413
1414 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1415 xfs_bmbt_set_startblock(ep,
1416 new->br_startblock + new->br_blockcount);
1417 xfs_bmbt_set_blockcount(ep,
1418 PREV.br_blockcount - new->br_blockcount);
1419 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1420
1421 ip->i_df.if_lastex = idx - 1;
1422 if (cur == NULL)
1423 rval = XFS_ILOG_DEXT;
1424 else {
1425 rval = 0;
1426 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1427 PREV.br_startblock, PREV.br_blockcount,
1428 &i)))
1429 goto done;
1430 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1431 if ((error = xfs_bmbt_update(cur,
1432 PREV.br_startoff + new->br_blockcount,
1433 PREV.br_startblock + new->br_blockcount,
1434 PREV.br_blockcount - new->br_blockcount,
1435 oldext)))
1436 goto done;
1437 if ((error = xfs_btree_decrement(cur, 0, &i)))
1438 goto done;
1439 if (xfs_bmbt_update(cur, LEFT.br_startoff,
1440 LEFT.br_startblock,
1441 LEFT.br_blockcount + new->br_blockcount,
1442 LEFT.br_state))
1443 goto done;
1444 }
1445 break;
1446
1447 case BMAP_LEFT_FILLING:
1448 /*
1449 * Setting the first part of a previous oldext extent to newext.
1450 * The left neighbor is not contiguous.
1451 */
1452 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1453 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1454 xfs_bmbt_set_startoff(ep, new_endoff);
1455 xfs_bmbt_set_blockcount(ep,
1456 PREV.br_blockcount - new->br_blockcount);
1457 xfs_bmbt_set_startblock(ep,
1458 new->br_startblock + new->br_blockcount);
1459 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1460
1461 xfs_iext_insert(ip, idx, 1, new, state);
1462 ip->i_df.if_lastex = idx;
1463 ip->i_d.di_nextents++;
1464 if (cur == NULL)
1465 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1466 else {
1467 rval = XFS_ILOG_CORE;
1468 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1469 PREV.br_startblock, PREV.br_blockcount,
1470 &i)))
1471 goto done;
1472 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1473 if ((error = xfs_bmbt_update(cur,
1474 PREV.br_startoff + new->br_blockcount,
1475 PREV.br_startblock + new->br_blockcount,
1476 PREV.br_blockcount - new->br_blockcount,
1477 oldext)))
1478 goto done;
1479 cur->bc_rec.b = *new;
1480 if ((error = xfs_btree_insert(cur, &i)))
1481 goto done;
1482 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1483 }
1484 break;
1485
1486 case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1487 /*
1488 * Setting the last part of a previous oldext extent to newext.
1489 * The right neighbor is contiguous with the new allocation.
1490 */
1491 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1492 trace_xfs_bmap_pre_update(ip, idx + 1, state, _THIS_IP_);
1493 xfs_bmbt_set_blockcount(ep,
1494 PREV.br_blockcount - new->br_blockcount);
1495 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1496 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, idx + 1),
1497 new->br_startoff, new->br_startblock,
1498 new->br_blockcount + RIGHT.br_blockcount, newext);
1499 trace_xfs_bmap_post_update(ip, idx + 1, state, _THIS_IP_);
1500
1501 ip->i_df.if_lastex = idx + 1;
1502 if (cur == NULL)
1503 rval = XFS_ILOG_DEXT;
1504 else {
1505 rval = 0;
1506 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1507 PREV.br_startblock,
1508 PREV.br_blockcount, &i)))
1509 goto done;
1510 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1511 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1512 PREV.br_startblock,
1513 PREV.br_blockcount - new->br_blockcount,
1514 oldext)))
1515 goto done;
1516 if ((error = xfs_btree_increment(cur, 0, &i)))
1517 goto done;
1518 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1519 new->br_startblock,
1520 new->br_blockcount + RIGHT.br_blockcount,
1521 newext)))
1522 goto done;
1523 }
1524 break;
1525
1526 case BMAP_RIGHT_FILLING:
1527 /*
1528 * Setting the last part of a previous oldext extent to newext.
1529 * The right neighbor is not contiguous.
1530 */
1531 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1532 xfs_bmbt_set_blockcount(ep,
1533 PREV.br_blockcount - new->br_blockcount);
1534 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1535
1536 xfs_iext_insert(ip, idx + 1, 1, new, state);
1537 ip->i_df.if_lastex = idx + 1;
1538 ip->i_d.di_nextents++;
1539 if (cur == NULL)
1540 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1541 else {
1542 rval = XFS_ILOG_CORE;
1543 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1544 PREV.br_startblock, PREV.br_blockcount,
1545 &i)))
1546 goto done;
1547 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1548 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1549 PREV.br_startblock,
1550 PREV.br_blockcount - new->br_blockcount,
1551 oldext)))
1552 goto done;
1553 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1554 new->br_startblock, new->br_blockcount,
1555 &i)))
1556 goto done;
1557 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1558 cur->bc_rec.b.br_state = XFS_EXT_NORM;
1559 if ((error = xfs_btree_insert(cur, &i)))
1560 goto done;
1561 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1562 }
1563 break;
1564
1565 case 0:
1566 /*
1567 * Setting the middle part of a previous oldext extent to
1568 * newext. Contiguity is impossible here.
1569 * One extent becomes three extents.
1570 */
1571 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1572 xfs_bmbt_set_blockcount(ep,
1573 new->br_startoff - PREV.br_startoff);
1574 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1575
1576 r[0] = *new;
1577 r[1].br_startoff = new_endoff;
1578 r[1].br_blockcount =
1579 PREV.br_startoff + PREV.br_blockcount - new_endoff;
1580 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1581 r[1].br_state = oldext;
1582 xfs_iext_insert(ip, idx + 1, 2, &r[0], state);
1583 ip->i_df.if_lastex = idx + 1;
1584 ip->i_d.di_nextents += 2;
1585 if (cur == NULL)
1586 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1587 else {
1588 rval = XFS_ILOG_CORE;
1589 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1590 PREV.br_startblock, PREV.br_blockcount,
1591 &i)))
1592 goto done;
1593 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1594 /* new right extent - oldext */
1595 if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1596 r[1].br_startblock, r[1].br_blockcount,
1597 r[1].br_state)))
1598 goto done;
1599 /* new left extent - oldext */
1600 cur->bc_rec.b = PREV;
1601 cur->bc_rec.b.br_blockcount =
1602 new->br_startoff - PREV.br_startoff;
1603 if ((error = xfs_btree_insert(cur, &i)))
1604 goto done;
1605 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1606 /*
1607 * Reset the cursor to the position of the new extent
1608 * we are about to insert as we can't trust it after
1609 * the previous insert.
1610 */
1611 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1612 new->br_startblock, new->br_blockcount,
1613 &i)))
1614 goto done;
1615 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1616 /* new middle extent - newext */
1617 cur->bc_rec.b.br_state = new->br_state;
1618 if ((error = xfs_btree_insert(cur, &i)))
1619 goto done;
1620 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1621 }
1622 break;
1623
1624 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1625 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1626 case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1627 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1628 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1629 case BMAP_LEFT_CONTIG:
1630 case BMAP_RIGHT_CONTIG:
1631 /*
1632 * These cases are all impossible.
1633 */
1634 ASSERT(0);
1635 }
1636 *curp = cur;
1637 done:
1638 *logflagsp = rval;
1639 return error;
1640 #undef LEFT
1641 #undef RIGHT
1642 #undef PREV
1643 }
1644
1645 /*
1646 * Called by xfs_bmap_add_extent to handle cases converting a hole
1647 * to a delayed allocation.
1648 */
1649 /*ARGSUSED*/
1650 STATIC int /* error */
1651 xfs_bmap_add_extent_hole_delay(
1652 xfs_inode_t *ip, /* incore inode pointer */
1653 xfs_extnum_t idx, /* extent number to update/insert */
1654 xfs_bmbt_irec_t *new, /* new data to add to file extents */
1655 int *logflagsp, /* inode logging flags */
1656 int rsvd) /* OK to allocate reserved blocks */
1657 {
1658 xfs_bmbt_rec_host_t *ep; /* extent record for idx */
1659 xfs_ifork_t *ifp; /* inode fork pointer */
1660 xfs_bmbt_irec_t left; /* left neighbor extent entry */
1661 xfs_filblks_t newlen=0; /* new indirect size */
1662 xfs_filblks_t oldlen=0; /* old indirect size */
1663 xfs_bmbt_irec_t right; /* right neighbor extent entry */
1664 int state; /* state bits, accessed thru macros */
1665 xfs_filblks_t temp=0; /* temp for indirect calculations */
1666
1667 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1668 ep = xfs_iext_get_ext(ifp, idx);
1669 state = 0;
1670 ASSERT(isnullstartblock(new->br_startblock));
1671
1672 /*
1673 * Check and set flags if this segment has a left neighbor
1674 */
1675 if (idx > 0) {
1676 state |= BMAP_LEFT_VALID;
1677 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &left);
1678
1679 if (isnullstartblock(left.br_startblock))
1680 state |= BMAP_LEFT_DELAY;
1681 }
1682
1683 /*
1684 * Check and set flags if the current (right) segment exists.
1685 * If it doesn't exist, we're converting the hole at end-of-file.
1686 */
1687 if (idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1688 state |= BMAP_RIGHT_VALID;
1689 xfs_bmbt_get_all(ep, &right);
1690
1691 if (isnullstartblock(right.br_startblock))
1692 state |= BMAP_RIGHT_DELAY;
1693 }
1694
1695 /*
1696 * Set contiguity flags on the left and right neighbors.
1697 * Don't let extents get too large, even if the pieces are contiguous.
1698 */
1699 if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
1700 left.br_startoff + left.br_blockcount == new->br_startoff &&
1701 left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1702 state |= BMAP_LEFT_CONTIG;
1703
1704 if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
1705 new->br_startoff + new->br_blockcount == right.br_startoff &&
1706 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1707 (!(state & BMAP_LEFT_CONTIG) ||
1708 (left.br_blockcount + new->br_blockcount +
1709 right.br_blockcount <= MAXEXTLEN)))
1710 state |= BMAP_RIGHT_CONTIG;
1711
1712 /*
1713 * Switch out based on the contiguity flags.
1714 */
1715 switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1716 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1717 /*
1718 * New allocation is contiguous with delayed allocations
1719 * on the left and on the right.
1720 * Merge all three into a single extent record.
1721 */
1722 temp = left.br_blockcount + new->br_blockcount +
1723 right.br_blockcount;
1724
1725 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
1726 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1), temp);
1727 oldlen = startblockval(left.br_startblock) +
1728 startblockval(new->br_startblock) +
1729 startblockval(right.br_startblock);
1730 newlen = xfs_bmap_worst_indlen(ip, temp);
1731 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx - 1),
1732 nullstartblock((int)newlen));
1733 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
1734
1735 xfs_iext_remove(ip, idx, 1, state);
1736 ip->i_df.if_lastex = idx - 1;
1737 break;
1738
1739 case BMAP_LEFT_CONTIG:
1740 /*
1741 * New allocation is contiguous with a delayed allocation
1742 * on the left.
1743 * Merge the new allocation with the left neighbor.
1744 */
1745 temp = left.br_blockcount + new->br_blockcount;
1746 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
1747 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1), temp);
1748 oldlen = startblockval(left.br_startblock) +
1749 startblockval(new->br_startblock);
1750 newlen = xfs_bmap_worst_indlen(ip, temp);
1751 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx - 1),
1752 nullstartblock((int)newlen));
1753 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
1754
1755 ip->i_df.if_lastex = idx - 1;
1756 break;
1757
1758 case BMAP_RIGHT_CONTIG:
1759 /*
1760 * New allocation is contiguous with a delayed allocation
1761 * on the right.
1762 * Merge the new allocation with the right neighbor.
1763 */
1764 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1765 temp = new->br_blockcount + right.br_blockcount;
1766 oldlen = startblockval(new->br_startblock) +
1767 startblockval(right.br_startblock);
1768 newlen = xfs_bmap_worst_indlen(ip, temp);
1769 xfs_bmbt_set_allf(ep, new->br_startoff,
1770 nullstartblock((int)newlen), temp, right.br_state);
1771 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1772
1773 ip->i_df.if_lastex = idx;
1774 break;
1775
1776 case 0:
1777 /*
1778 * New allocation is not contiguous with another
1779 * delayed allocation.
1780 * Insert a new entry.
1781 */
1782 oldlen = newlen = 0;
1783 xfs_iext_insert(ip, idx, 1, new, state);
1784 ip->i_df.if_lastex = idx;
1785 break;
1786 }
1787 if (oldlen != newlen) {
1788 ASSERT(oldlen > newlen);
1789 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
1790 (int64_t)(oldlen - newlen), rsvd);
1791 /*
1792 * Nothing to do for disk quota accounting here.
1793 */
1794 }
1795 *logflagsp = 0;
1796 return 0;
1797 }
1798
1799 /*
1800 * Called by xfs_bmap_add_extent to handle cases converting a hole
1801 * to a real allocation.
1802 */
1803 STATIC int /* error */
1804 xfs_bmap_add_extent_hole_real(
1805 xfs_inode_t *ip, /* incore inode pointer */
1806 xfs_extnum_t idx, /* extent number to update/insert */
1807 xfs_btree_cur_t *cur, /* if null, not a btree */
1808 xfs_bmbt_irec_t *new, /* new data to add to file extents */
1809 int *logflagsp, /* inode logging flags */
1810 int whichfork) /* data or attr fork */
1811 {
1812 xfs_bmbt_rec_host_t *ep; /* pointer to extent entry ins. point */
1813 int error; /* error return value */
1814 int i; /* temp state */
1815 xfs_ifork_t *ifp; /* inode fork pointer */
1816 xfs_bmbt_irec_t left; /* left neighbor extent entry */
1817 xfs_bmbt_irec_t right; /* right neighbor extent entry */
1818 int rval=0; /* return value (logging flags) */
1819 int state; /* state bits, accessed thru macros */
1820
1821 ifp = XFS_IFORK_PTR(ip, whichfork);
1822 ASSERT(idx <= ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
1823 ep = xfs_iext_get_ext(ifp, idx);
1824 state = 0;
1825
1826 if (whichfork == XFS_ATTR_FORK)
1827 state |= BMAP_ATTRFORK;
1828
1829 /*
1830 * Check and set flags if this segment has a left neighbor.
1831 */
1832 if (idx > 0) {
1833 state |= BMAP_LEFT_VALID;
1834 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &left);
1835 if (isnullstartblock(left.br_startblock))
1836 state |= BMAP_LEFT_DELAY;
1837 }
1838
1839 /*
1840 * Check and set flags if this segment has a current value.
1841 * Not true if we're inserting into the "hole" at eof.
1842 */
1843 if (idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1844 state |= BMAP_RIGHT_VALID;
1845 xfs_bmbt_get_all(ep, &right);
1846 if (isnullstartblock(right.br_startblock))
1847 state |= BMAP_RIGHT_DELAY;
1848 }
1849
1850 /*
1851 * We're inserting a real allocation between "left" and "right".
1852 * Set the contiguity flags. Don't let extents get too large.
1853 */
1854 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1855 left.br_startoff + left.br_blockcount == new->br_startoff &&
1856 left.br_startblock + left.br_blockcount == new->br_startblock &&
1857 left.br_state == new->br_state &&
1858 left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1859 state |= BMAP_LEFT_CONTIG;
1860
1861 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1862 new->br_startoff + new->br_blockcount == right.br_startoff &&
1863 new->br_startblock + new->br_blockcount == right.br_startblock &&
1864 new->br_state == right.br_state &&
1865 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1866 (!(state & BMAP_LEFT_CONTIG) ||
1867 left.br_blockcount + new->br_blockcount +
1868 right.br_blockcount <= MAXEXTLEN))
1869 state |= BMAP_RIGHT_CONTIG;
1870
1871 error = 0;
1872 /*
1873 * Select which case we're in here, and implement it.
1874 */
1875 switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1876 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1877 /*
1878 * New allocation is contiguous with real allocations on the
1879 * left and on the right.
1880 * Merge all three into a single extent record.
1881 */
1882 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
1883 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1884 left.br_blockcount + new->br_blockcount +
1885 right.br_blockcount);
1886 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
1887
1888 xfs_iext_remove(ip, idx, 1, state);
1889 ifp->if_lastex = idx - 1;
1890 XFS_IFORK_NEXT_SET(ip, whichfork,
1891 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
1892 if (cur == NULL) {
1893 rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1894 } else {
1895 rval = XFS_ILOG_CORE;
1896 if ((error = xfs_bmbt_lookup_eq(cur,
1897 right.br_startoff,
1898 right.br_startblock,
1899 right.br_blockcount, &i)))
1900 goto done;
1901 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1902 if ((error = xfs_btree_delete(cur, &i)))
1903 goto done;
1904 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1905 if ((error = xfs_btree_decrement(cur, 0, &i)))
1906 goto done;
1907 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1908 if ((error = xfs_bmbt_update(cur, left.br_startoff,
1909 left.br_startblock,
1910 left.br_blockcount +
1911 new->br_blockcount +
1912 right.br_blockcount,
1913 left.br_state)))
1914 goto done;
1915 }
1916 break;
1917
1918 case BMAP_LEFT_CONTIG:
1919 /*
1920 * New allocation is contiguous with a real allocation
1921 * on the left.
1922 * Merge the new allocation with the left neighbor.
1923 */
1924 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
1925 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1926 left.br_blockcount + new->br_blockcount);
1927 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
1928
1929 ifp->if_lastex = idx - 1;
1930 if (cur == NULL) {
1931 rval = xfs_ilog_fext(whichfork);
1932 } else {
1933 rval = 0;
1934 if ((error = xfs_bmbt_lookup_eq(cur,
1935 left.br_startoff,
1936 left.br_startblock,
1937 left.br_blockcount, &i)))
1938 goto done;
1939 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1940 if ((error = xfs_bmbt_update(cur, left.br_startoff,
1941 left.br_startblock,
1942 left.br_blockcount +
1943 new->br_blockcount,
1944 left.br_state)))
1945 goto done;
1946 }
1947 break;
1948
1949 case BMAP_RIGHT_CONTIG:
1950 /*
1951 * New allocation is contiguous with a real allocation
1952 * on the right.
1953 * Merge the new allocation with the right neighbor.
1954 */
1955 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1956 xfs_bmbt_set_allf(ep, new->br_startoff, new->br_startblock,
1957 new->br_blockcount + right.br_blockcount,
1958 right.br_state);
1959 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1960
1961 ifp->if_lastex = idx;
1962 if (cur == NULL) {
1963 rval = xfs_ilog_fext(whichfork);
1964 } else {
1965 rval = 0;
1966 if ((error = xfs_bmbt_lookup_eq(cur,
1967 right.br_startoff,
1968 right.br_startblock,
1969 right.br_blockcount, &i)))
1970 goto done;
1971 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1972 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1973 new->br_startblock,
1974 new->br_blockcount +
1975 right.br_blockcount,
1976 right.br_state)))
1977 goto done;
1978 }
1979 break;
1980
1981 case 0:
1982 /*
1983 * New allocation is not contiguous with another
1984 * real allocation.
1985 * Insert a new entry.
1986 */
1987 xfs_iext_insert(ip, idx, 1, new, state);
1988 ifp->if_lastex = idx;
1989 XFS_IFORK_NEXT_SET(ip, whichfork,
1990 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
1991 if (cur == NULL) {
1992 rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1993 } else {
1994 rval = XFS_ILOG_CORE;
1995 if ((error = xfs_bmbt_lookup_eq(cur,
1996 new->br_startoff,
1997 new->br_startblock,
1998 new->br_blockcount, &i)))
1999 goto done;
2000 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2001 cur->bc_rec.b.br_state = new->br_state;
2002 if ((error = xfs_btree_insert(cur, &i)))
2003 goto done;
2004 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2005 }
2006 break;
2007 }
2008 done:
2009 *logflagsp = rval;
2010 return error;
2011 }
2012
2013 /*
2014 * Adjust the size of the new extent based on di_extsize and rt extsize.
2015 */
2016 STATIC int
2017 xfs_bmap_extsize_align(
2018 xfs_mount_t *mp,
2019 xfs_bmbt_irec_t *gotp, /* next extent pointer */
2020 xfs_bmbt_irec_t *prevp, /* previous extent pointer */
2021 xfs_extlen_t extsz, /* align to this extent size */
2022 int rt, /* is this a realtime inode? */
2023 int eof, /* is extent at end-of-file? */
2024 int delay, /* creating delalloc extent? */
2025 int convert, /* overwriting unwritten extent? */
2026 xfs_fileoff_t *offp, /* in/out: aligned offset */
2027 xfs_extlen_t *lenp) /* in/out: aligned length */
2028 {
2029 xfs_fileoff_t orig_off; /* original offset */
2030 xfs_extlen_t orig_alen; /* original length */
2031 xfs_fileoff_t orig_end; /* original off+len */
2032 xfs_fileoff_t nexto; /* next file offset */
2033 xfs_fileoff_t prevo; /* previous file offset */
2034 xfs_fileoff_t align_off; /* temp for offset */
2035 xfs_extlen_t align_alen; /* temp for length */
2036 xfs_extlen_t temp; /* temp for calculations */
2037
2038 if (convert)
2039 return 0;
2040
2041 orig_off = align_off = *offp;
2042 orig_alen = align_alen = *lenp;
2043 orig_end = orig_off + orig_alen;
2044
2045 /*
2046 * If this request overlaps an existing extent, then don't
2047 * attempt to perform any additional alignment.
2048 */
2049 if (!delay && !eof &&
2050 (orig_off >= gotp->br_startoff) &&
2051 (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2052 return 0;
2053 }
2054
2055 /*
2056 * If the file offset is unaligned vs. the extent size
2057 * we need to align it. This will be possible unless
2058 * the file was previously written with a kernel that didn't
2059 * perform this alignment, or if a truncate shot us in the
2060 * foot.
2061 */
2062 temp = do_mod(orig_off, extsz);
2063 if (temp) {
2064 align_alen += temp;
2065 align_off -= temp;
2066 }
2067 /*
2068 * Same adjustment for the end of the requested area.
2069 */
2070 if ((temp = (align_alen % extsz))) {
2071 align_alen += extsz - temp;
2072 }
2073 /*
2074 * If the previous block overlaps with this proposed allocation
2075 * then move the start forward without adjusting the length.
2076 */
2077 if (prevp->br_startoff != NULLFILEOFF) {
2078 if (prevp->br_startblock == HOLESTARTBLOCK)
2079 prevo = prevp->br_startoff;
2080 else
2081 prevo = prevp->br_startoff + prevp->br_blockcount;
2082 } else
2083 prevo = 0;
2084 if (align_off != orig_off && align_off < prevo)
2085 align_off = prevo;
2086 /*
2087 * If the next block overlaps with this proposed allocation
2088 * then move the start back without adjusting the length,
2089 * but not before offset 0.
2090 * This may of course make the start overlap previous block,
2091 * and if we hit the offset 0 limit then the next block
2092 * can still overlap too.
2093 */
2094 if (!eof && gotp->br_startoff != NULLFILEOFF) {
2095 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
2096 (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
2097 nexto = gotp->br_startoff + gotp->br_blockcount;
2098 else
2099 nexto = gotp->br_startoff;
2100 } else
2101 nexto = NULLFILEOFF;
2102 if (!eof &&
2103 align_off + align_alen != orig_end &&
2104 align_off + align_alen > nexto)
2105 align_off = nexto > align_alen ? nexto - align_alen : 0;
2106 /*
2107 * If we're now overlapping the next or previous extent that
2108 * means we can't fit an extsz piece in this hole. Just move
2109 * the start forward to the first valid spot and set
2110 * the length so we hit the end.
2111 */
2112 if (align_off != orig_off && align_off < prevo)
2113 align_off = prevo;
2114 if (align_off + align_alen != orig_end &&
2115 align_off + align_alen > nexto &&
2116 nexto != NULLFILEOFF) {
2117 ASSERT(nexto > prevo);
2118 align_alen = nexto - align_off;
2119 }
2120
2121 /*
2122 * If realtime, and the result isn't a multiple of the realtime
2123 * extent size we need to remove blocks until it is.
2124 */
2125 if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
2126 /*
2127 * We're not covering the original request, or
2128 * we won't be able to once we fix the length.
2129 */
2130 if (orig_off < align_off ||
2131 orig_end > align_off + align_alen ||
2132 align_alen - temp < orig_alen)
2133 return XFS_ERROR(EINVAL);
2134 /*
2135 * Try to fix it by moving the start up.
2136 */
2137 if (align_off + temp <= orig_off) {
2138 align_alen -= temp;
2139 align_off += temp;
2140 }
2141 /*
2142 * Try to fix it by moving the end in.
2143 */
2144 else if (align_off + align_alen - temp >= orig_end)
2145 align_alen -= temp;
2146 /*
2147 * Set the start to the minimum then trim the length.
2148 */
2149 else {
2150 align_alen -= orig_off - align_off;
2151 align_off = orig_off;
2152 align_alen -= align_alen % mp->m_sb.sb_rextsize;
2153 }
2154 /*
2155 * Result doesn't cover the request, fail it.
2156 */
2157 if (orig_off < align_off || orig_end > align_off + align_alen)
2158 return XFS_ERROR(EINVAL);
2159 } else {
2160 ASSERT(orig_off >= align_off);
2161 ASSERT(orig_end <= align_off + align_alen);
2162 }
2163
2164 #ifdef DEBUG
2165 if (!eof && gotp->br_startoff != NULLFILEOFF)
2166 ASSERT(align_off + align_alen <= gotp->br_startoff);
2167 if (prevp->br_startoff != NULLFILEOFF)
2168 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
2169 #endif
2170
2171 *lenp = align_alen;
2172 *offp = align_off;
2173 return 0;
2174 }
2175
2176 #define XFS_ALLOC_GAP_UNITS 4
2177
2178 STATIC void
2179 xfs_bmap_adjacent(
2180 xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2181 {
2182 xfs_fsblock_t adjust; /* adjustment to block numbers */
2183 xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
2184 xfs_mount_t *mp; /* mount point structure */
2185 int nullfb; /* true if ap->firstblock isn't set */
2186 int rt; /* true if inode is realtime */
2187
2188 #define ISVALID(x,y) \
2189 (rt ? \
2190 (x) < mp->m_sb.sb_rblocks : \
2191 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2192 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2193 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2194
2195 mp = ap->ip->i_mount;
2196 nullfb = ap->firstblock == NULLFSBLOCK;
2197 rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2198 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2199 /*
2200 * If allocating at eof, and there's a previous real block,
2201 * try to use its last block as our starting point.
2202 */
2203 if (ap->eof && ap->prevp->br_startoff != NULLFILEOFF &&
2204 !isnullstartblock(ap->prevp->br_startblock) &&
2205 ISVALID(ap->prevp->br_startblock + ap->prevp->br_blockcount,
2206 ap->prevp->br_startblock)) {
2207 ap->rval = ap->prevp->br_startblock + ap->prevp->br_blockcount;
2208 /*
2209 * Adjust for the gap between prevp and us.
2210 */
2211 adjust = ap->off -
2212 (ap->prevp->br_startoff + ap->prevp->br_blockcount);
2213 if (adjust &&
2214 ISVALID(ap->rval + adjust, ap->prevp->br_startblock))
2215 ap->rval += adjust;
2216 }
2217 /*
2218 * If not at eof, then compare the two neighbor blocks.
2219 * Figure out whether either one gives us a good starting point,
2220 * and pick the better one.
2221 */
2222 else if (!ap->eof) {
2223 xfs_fsblock_t gotbno; /* right side block number */
2224 xfs_fsblock_t gotdiff=0; /* right side difference */
2225 xfs_fsblock_t prevbno; /* left side block number */
2226 xfs_fsblock_t prevdiff=0; /* left side difference */
2227
2228 /*
2229 * If there's a previous (left) block, select a requested
2230 * start block based on it.
2231 */
2232 if (ap->prevp->br_startoff != NULLFILEOFF &&
2233 !isnullstartblock(ap->prevp->br_startblock) &&
2234 (prevbno = ap->prevp->br_startblock +
2235 ap->prevp->br_blockcount) &&
2236 ISVALID(prevbno, ap->prevp->br_startblock)) {
2237 /*
2238 * Calculate gap to end of previous block.
2239 */
2240 adjust = prevdiff = ap->off -
2241 (ap->prevp->br_startoff +
2242 ap->prevp->br_blockcount);
2243 /*
2244 * Figure the startblock based on the previous block's
2245 * end and the gap size.
2246 * Heuristic!
2247 * If the gap is large relative to the piece we're
2248 * allocating, or using it gives us an invalid block
2249 * number, then just use the end of the previous block.
2250 */
2251 if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2252 ISVALID(prevbno + prevdiff,
2253 ap->prevp->br_startblock))
2254 prevbno += adjust;
2255 else
2256 prevdiff += adjust;
2257 /*
2258 * If the firstblock forbids it, can't use it,
2259 * must use default.
2260 */
2261 if (!rt && !nullfb &&
2262 XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2263 prevbno = NULLFSBLOCK;
2264 }
2265 /*
2266 * No previous block or can't follow it, just default.
2267 */
2268 else
2269 prevbno = NULLFSBLOCK;
2270 /*
2271 * If there's a following (right) block, select a requested
2272 * start block based on it.
2273 */
2274 if (!isnullstartblock(ap->gotp->br_startblock)) {
2275 /*
2276 * Calculate gap to start of next block.
2277 */
2278 adjust = gotdiff = ap->gotp->br_startoff - ap->off;
2279 /*
2280 * Figure the startblock based on the next block's
2281 * start and the gap size.
2282 */
2283 gotbno = ap->gotp->br_startblock;
2284 /*
2285 * Heuristic!
2286 * If the gap is large relative to the piece we're
2287 * allocating, or using it gives us an invalid block
2288 * number, then just use the start of the next block
2289 * offset by our length.
2290 */
2291 if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2292 ISVALID(gotbno - gotdiff, gotbno))
2293 gotbno -= adjust;
2294 else if (ISVALID(gotbno - ap->alen, gotbno)) {
2295 gotbno -= ap->alen;
2296 gotdiff += adjust - ap->alen;
2297 } else
2298 gotdiff += adjust;
2299 /*
2300 * If the firstblock forbids it, can't use it,
2301 * must use default.
2302 */
2303 if (!rt && !nullfb &&
2304 XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2305 gotbno = NULLFSBLOCK;
2306 }
2307 /*
2308 * No next block, just default.
2309 */
2310 else
2311 gotbno = NULLFSBLOCK;
2312 /*
2313 * If both valid, pick the better one, else the only good
2314 * one, else ap->rval is already set (to 0 or the inode block).
2315 */
2316 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2317 ap->rval = prevdiff <= gotdiff ? prevbno : gotbno;
2318 else if (prevbno != NULLFSBLOCK)
2319 ap->rval = prevbno;
2320 else if (gotbno != NULLFSBLOCK)
2321 ap->rval = gotbno;
2322 }
2323 #undef ISVALID
2324 }
2325
2326 STATIC int
2327 xfs_bmap_rtalloc(
2328 xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2329 {
2330 xfs_alloctype_t atype = 0; /* type for allocation routines */
2331 int error; /* error return value */
2332 xfs_mount_t *mp; /* mount point structure */
2333 xfs_extlen_t prod = 0; /* product factor for allocators */
2334 xfs_extlen_t ralen = 0; /* realtime allocation length */
2335 xfs_extlen_t align; /* minimum allocation alignment */
2336 xfs_inode_t *ip; /* bitmap incore inode */
2337 xfs_rtblock_t rtb;
2338
2339 mp = ap->ip->i_mount;
2340 align = xfs_get_extsz_hint(ap->ip);
2341 prod = align / mp->m_sb.sb_rextsize;
2342 error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2343 align, 1, ap->eof, 0,
2344 ap->conv, &ap->off, &ap->alen);
2345 if (error)
2346 return error;
2347 ASSERT(ap->alen);
2348 ASSERT(ap->alen % mp->m_sb.sb_rextsize == 0);
2349
2350 /*
2351 * If the offset & length are not perfectly aligned
2352 * then kill prod, it will just get us in trouble.
2353 */
2354 if (do_mod(ap->off, align) || ap->alen % align)
2355 prod = 1;
2356 /*
2357 * Set ralen to be the actual requested length in rtextents.
2358 */
2359 ralen = ap->alen / mp->m_sb.sb_rextsize;
2360 /*
2361 * If the old value was close enough to MAXEXTLEN that
2362 * we rounded up to it, cut it back so it's valid again.
2363 * Note that if it's a really large request (bigger than
2364 * MAXEXTLEN), we don't hear about that number, and can't
2365 * adjust the starting point to match it.
2366 */
2367 if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2368 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2369
2370 /*
2371 * Lock out other modifications to the RT bitmap inode.
2372 */
2373 error = xfs_trans_iget(mp, ap->tp, mp->m_sb.sb_rbmino, 0,
2374 XFS_ILOCK_EXCL, &ip);
2375 if (error)
2376 return error;
2377 ASSERT(ip == mp->m_rbmip);
2378
2379 /*
2380 * If it's an allocation to an empty file at offset 0,
2381 * pick an extent that will space things out in the rt area.
2382 */
2383 if (ap->eof && ap->off == 0) {
2384 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
2385
2386 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2387 if (error)
2388 return error;
2389 ap->rval = rtx * mp->m_sb.sb_rextsize;
2390 } else {
2391 ap->rval = 0;
2392 }
2393
2394 xfs_bmap_adjacent(ap);
2395
2396 /*
2397 * Realtime allocation, done through xfs_rtallocate_extent.
2398 */
2399 atype = ap->rval == 0 ? XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2400 do_div(ap->rval, mp->m_sb.sb_rextsize);
2401 rtb = ap->rval;
2402 ap->alen = ralen;
2403 if ((error = xfs_rtallocate_extent(ap->tp, ap->rval, 1, ap->alen,
2404 &ralen, atype, ap->wasdel, prod, &rtb)))
2405 return error;
2406 if (rtb == NULLFSBLOCK && prod > 1 &&
2407 (error = xfs_rtallocate_extent(ap->tp, ap->rval, 1,
2408 ap->alen, &ralen, atype,
2409 ap->wasdel, 1, &rtb)))
2410 return error;
2411 ap->rval = rtb;
2412 if (ap->rval != NULLFSBLOCK) {
2413 ap->rval *= mp->m_sb.sb_rextsize;
2414 ralen *= mp->m_sb.sb_rextsize;
2415 ap->alen = ralen;
2416 ap->ip->i_d.di_nblocks += ralen;
2417 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2418 if (ap->wasdel)
2419 ap->ip->i_delayed_blks -= ralen;
2420 /*
2421 * Adjust the disk quota also. This was reserved
2422 * earlier.
2423 */
2424 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2425 ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2426 XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2427 } else {
2428 ap->alen = 0;
2429 }
2430 return 0;
2431 }
2432
2433 STATIC int
2434 xfs_bmap_btalloc_nullfb(
2435 struct xfs_bmalloca *ap,
2436 struct xfs_alloc_arg *args,
2437 xfs_extlen_t *blen)
2438 {
2439 struct xfs_mount *mp = ap->ip->i_mount;
2440 struct xfs_perag *pag;
2441 xfs_agnumber_t ag, startag;
2442 int notinit = 0;
2443 int error;
2444
2445 if (ap->userdata && xfs_inode_is_filestream(ap->ip))
2446 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2447 else
2448 args->type = XFS_ALLOCTYPE_START_BNO;
2449 args->total = ap->total;
2450
2451 /*
2452 * Search for an allocation group with a single extent large enough
2453 * for the request. If one isn't found, then adjust the minimum
2454 * allocation size to the largest space found.
2455 */
2456 startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
2457 if (startag == NULLAGNUMBER)
2458 startag = ag = 0;
2459
2460 pag = xfs_perag_get(mp, ag);
2461 while (*blen < args->maxlen) {
2462 if (!pag->pagf_init) {
2463 error = xfs_alloc_pagf_init(mp, args->tp, ag,
2464 XFS_ALLOC_FLAG_TRYLOCK);
2465 if (error) {
2466 xfs_perag_put(pag);
2467 return error;
2468 }
2469 }
2470
2471 /*
2472 * See xfs_alloc_fix_freelist...
2473 */
2474 if (pag->pagf_init) {
2475 xfs_extlen_t longest;
2476 longest = xfs_alloc_longest_free_extent(mp, pag);
2477 if (*blen < longest)
2478 *blen = longest;
2479 } else
2480 notinit = 1;
2481
2482 if (xfs_inode_is_filestream(ap->ip)) {
2483 if (*blen >= args->maxlen)
2484 break;
2485
2486 if (ap->userdata) {
2487 /*
2488 * If startag is an invalid AG, we've
2489 * come here once before and
2490 * xfs_filestream_new_ag picked the
2491 * best currently available.
2492 *
2493 * Don't continue looping, since we
2494 * could loop forever.
2495 */
2496 if (startag == NULLAGNUMBER)
2497 break;
2498
2499 error = xfs_filestream_new_ag(ap, &ag);
2500 xfs_perag_put(pag);
2501 if (error)
2502 return error;
2503
2504 /* loop again to set 'blen'*/
2505 startag = NULLAGNUMBER;
2506 pag = xfs_perag_get(mp, ag);
2507 continue;
2508 }
2509 }
2510 if (++ag == mp->m_sb.sb_agcount)
2511 ag = 0;
2512 if (ag == startag)
2513 break;
2514 xfs_perag_put(pag);
2515 pag = xfs_perag_get(mp, ag);
2516 }
2517 xfs_perag_put(pag);
2518
2519 /*
2520 * Since the above loop did a BUF_TRYLOCK, it is
2521 * possible that there is space for this request.
2522 */
2523 if (notinit || *blen < ap->minlen)
2524 args->minlen = ap->minlen;
2525 /*
2526 * If the best seen length is less than the request
2527 * length, use the best as the minimum.
2528 */
2529 else if (*blen < args->maxlen)
2530 args->minlen = *blen;
2531 /*
2532 * Otherwise we've seen an extent as big as maxlen,
2533 * use that as the minimum.
2534 */
2535 else
2536 args->minlen = args->maxlen;
2537
2538 /*
2539 * set the failure fallback case to look in the selected
2540 * AG as the stream may have moved.
2541 */
2542 if (xfs_inode_is_filestream(ap->ip))
2543 ap->rval = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
2544
2545 return 0;
2546 }
2547
2548 STATIC int
2549 xfs_bmap_btalloc(
2550 xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2551 {
2552 xfs_mount_t *mp; /* mount point structure */
2553 xfs_alloctype_t atype = 0; /* type for allocation routines */
2554 xfs_extlen_t align; /* minimum allocation alignment */
2555 xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
2556 xfs_agnumber_t ag;
2557 xfs_alloc_arg_t args;
2558 xfs_extlen_t blen;
2559 xfs_extlen_t nextminlen = 0;
2560 int nullfb; /* true if ap->firstblock isn't set */
2561 int isaligned;
2562 int tryagain;
2563 int error;
2564
2565 mp = ap->ip->i_mount;
2566 align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
2567 if (unlikely(align)) {
2568 error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2569 align, 0, ap->eof, 0, ap->conv,
2570 &ap->off, &ap->alen);
2571 ASSERT(!error);
2572 ASSERT(ap->alen);
2573 }
2574 nullfb = ap->firstblock == NULLFSBLOCK;
2575 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2576 if (nullfb) {
2577 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
2578 ag = xfs_filestream_lookup_ag(ap->ip);
2579 ag = (ag != NULLAGNUMBER) ? ag : 0;
2580 ap->rval = XFS_AGB_TO_FSB(mp, ag, 0);
2581 } else {
2582 ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2583 }
2584 } else
2585 ap->rval = ap->firstblock;
2586
2587 xfs_bmap_adjacent(ap);
2588
2589 /*
2590 * If allowed, use ap->rval; otherwise must use firstblock since
2591 * it's in the right allocation group.
2592 */
2593 if (nullfb || XFS_FSB_TO_AGNO(mp, ap->rval) == fb_agno)
2594 ;
2595 else
2596 ap->rval = ap->firstblock;
2597 /*
2598 * Normal allocation, done through xfs_alloc_vextent.
2599 */
2600 tryagain = isaligned = 0;
2601 args.tp = ap->tp;
2602 args.mp = mp;
2603 args.fsbno = ap->rval;
2604
2605 /* Trim the allocation back to the maximum an AG can fit. */
2606 args.maxlen = MIN(ap->alen, XFS_ALLOC_AG_MAX_USABLE(mp));
2607 args.firstblock = ap->firstblock;
2608 blen = 0;
2609 if (nullfb) {
2610 error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
2611 if (error)
2612 return error;
2613 } else if (ap->low) {
2614 if (xfs_inode_is_filestream(ap->ip))
2615 args.type = XFS_ALLOCTYPE_FIRST_AG;
2616 else
2617 args.type = XFS_ALLOCTYPE_START_BNO;
2618 args.total = args.minlen = ap->minlen;
2619 } else {
2620 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2621 args.total = ap->total;
2622 args.minlen = ap->minlen;
2623 }
2624 /* apply extent size hints if obtained earlier */
2625 if (unlikely(align)) {
2626 args.prod = align;
2627 if ((args.mod = (xfs_extlen_t)do_mod(ap->off, args.prod)))
2628 args.mod = (xfs_extlen_t)(args.prod - args.mod);
2629 } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
2630 args.prod = 1;
2631 args.mod = 0;
2632 } else {
2633 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
2634 if ((args.mod = (xfs_extlen_t)(do_mod(ap->off, args.prod))))
2635 args.mod = (xfs_extlen_t)(args.prod - args.mod);
2636 }
2637 /*
2638 * If we are not low on available data blocks, and the
2639 * underlying logical volume manager is a stripe, and
2640 * the file offset is zero then try to allocate data
2641 * blocks on stripe unit boundary.
2642 * NOTE: ap->aeof is only set if the allocation length
2643 * is >= the stripe unit and the allocation offset is
2644 * at the end of file.
2645 */
2646 if (!ap->low && ap->aeof) {
2647 if (!ap->off) {
2648 args.alignment = mp->m_dalign;
2649 atype = args.type;
2650 isaligned = 1;
2651 /*
2652 * Adjust for alignment
2653 */
2654 if (blen > args.alignment && blen <= args.maxlen)
2655 args.minlen = blen - args.alignment;
2656 args.minalignslop = 0;
2657 } else {
2658 /*
2659 * First try an exact bno allocation.
2660 * If it fails then do a near or start bno
2661 * allocation with alignment turned on.
2662 */
2663 atype = args.type;
2664 tryagain = 1;
2665 args.type = XFS_ALLOCTYPE_THIS_BNO;
2666 args.alignment = 1;
2667 /*
2668 * Compute the minlen+alignment for the
2669 * next case. Set slop so that the value
2670 * of minlen+alignment+slop doesn't go up
2671 * between the calls.
2672 */
2673 if (blen > mp->m_dalign && blen <= args.maxlen)
2674 nextminlen = blen - mp->m_dalign;
2675 else
2676 nextminlen = args.minlen;
2677 if (nextminlen + mp->m_dalign > args.minlen + 1)
2678 args.minalignslop =
2679 nextminlen + mp->m_dalign -
2680 args.minlen - 1;
2681 else
2682 args.minalignslop = 0;
2683 }
2684 } else {
2685 args.alignment = 1;
2686 args.minalignslop = 0;
2687 }
2688 args.minleft = ap->minleft;
2689 args.wasdel = ap->wasdel;
2690 args.isfl = 0;
2691 args.userdata = ap->userdata;
2692 if ((error = xfs_alloc_vextent(&args)))
2693 return error;
2694 if (tryagain && args.fsbno == NULLFSBLOCK) {
2695 /*
2696 * Exact allocation failed. Now try with alignment
2697 * turned on.
2698 */
2699 args.type = atype;
2700 args.fsbno = ap->rval;
2701 args.alignment = mp->m_dalign;
2702 args.minlen = nextminlen;
2703 args.minalignslop = 0;
2704 isaligned = 1;
2705 if ((error = xfs_alloc_vextent(&args)))
2706 return error;
2707 }
2708 if (isaligned && args.fsbno == NULLFSBLOCK) {
2709 /*
2710 * allocation failed, so turn off alignment and
2711 * try again.
2712 */
2713 args.type = atype;
2714 args.fsbno = ap->rval;
2715 args.alignment = 0;
2716 if ((error = xfs_alloc_vextent(&args)))
2717 return error;
2718 }
2719 if (args.fsbno == NULLFSBLOCK && nullfb &&
2720 args.minlen > ap->minlen) {
2721 args.minlen = ap->minlen;
2722 args.type = XFS_ALLOCTYPE_START_BNO;
2723 args.fsbno = ap->rval;
2724 if ((error = xfs_alloc_vextent(&args)))
2725 return error;
2726 }
2727 if (args.fsbno == NULLFSBLOCK && nullfb) {
2728 args.fsbno = 0;
2729 args.type = XFS_ALLOCTYPE_FIRST_AG;
2730 args.total = ap->minlen;
2731 args.minleft = 0;
2732 if ((error = xfs_alloc_vextent(&args)))
2733 return error;
2734 ap->low = 1;
2735 }
2736 if (args.fsbno != NULLFSBLOCK) {
2737 ap->firstblock = ap->rval = args.fsbno;
2738 ASSERT(nullfb || fb_agno == args.agno ||
2739 (ap->low && fb_agno < args.agno));
2740 ap->alen = args.len;
2741 ap->ip->i_d.di_nblocks += args.len;
2742 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2743 if (ap->wasdel)
2744 ap->ip->i_delayed_blks -= args.len;
2745 /*
2746 * Adjust the disk quota also. This was reserved
2747 * earlier.
2748 */
2749 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2750 ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2751 XFS_TRANS_DQ_BCOUNT,
2752 (long) args.len);
2753 } else {
2754 ap->rval = NULLFSBLOCK;
2755 ap->alen = 0;
2756 }
2757 return 0;
2758 }
2759
2760 /*
2761 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2762 * It figures out where to ask the underlying allocator to put the new extent.
2763 */
2764 STATIC int
2765 xfs_bmap_alloc(
2766 xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2767 {
2768 if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
2769 return xfs_bmap_rtalloc(ap);
2770 return xfs_bmap_btalloc(ap);
2771 }
2772
2773 /*
2774 * Transform a btree format file with only one leaf node, where the
2775 * extents list will fit in the inode, into an extents format file.
2776 * Since the file extents are already in-core, all we have to do is
2777 * give up the space for the btree root and pitch the leaf block.
2778 */
2779 STATIC int /* error */
2780 xfs_bmap_btree_to_extents(
2781 xfs_trans_t *tp, /* transaction pointer */
2782 xfs_inode_t *ip, /* incore inode pointer */
2783 xfs_btree_cur_t *cur, /* btree cursor */
2784 int *logflagsp, /* inode logging flags */
2785 int whichfork) /* data or attr fork */
2786 {
2787 /* REFERENCED */
2788 struct xfs_btree_block *cblock;/* child btree block */
2789 xfs_fsblock_t cbno; /* child block number */
2790 xfs_buf_t *cbp; /* child block's buffer */
2791 int error; /* error return value */
2792 xfs_ifork_t *ifp; /* inode fork data */
2793 xfs_mount_t *mp; /* mount point structure */
2794 __be64 *pp; /* ptr to block address */
2795 struct xfs_btree_block *rblock;/* root btree block */
2796
2797 mp = ip->i_mount;
2798 ifp = XFS_IFORK_PTR(ip, whichfork);
2799 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2800 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
2801 rblock = ifp->if_broot;
2802 ASSERT(be16_to_cpu(rblock->bb_level) == 1);
2803 ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
2804 ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
2805 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
2806 cbno = be64_to_cpu(*pp);
2807 *logflagsp = 0;
2808 #ifdef DEBUG
2809 if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
2810 return error;
2811 #endif
2812 if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
2813 XFS_BMAP_BTREE_REF)))
2814 return error;
2815 cblock = XFS_BUF_TO_BLOCK(cbp);
2816 if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
2817 return error;
2818 xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
2819 ip->i_d.di_nblocks--;
2820 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
2821 xfs_trans_binval(tp, cbp);
2822 if (cur->bc_bufs[0] == cbp)
2823 cur->bc_bufs[0] = NULL;
2824 xfs_iroot_realloc(ip, -1, whichfork);
2825 ASSERT(ifp->if_broot == NULL);
2826 ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
2827 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
2828 *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2829 return 0;
2830 }
2831
2832 /*
2833 * Called by xfs_bmapi to update file extent records and the btree
2834 * after removing space (or undoing a delayed allocation).
2835 */
2836 STATIC int /* error */
2837 xfs_bmap_del_extent(
2838 xfs_inode_t *ip, /* incore inode pointer */
2839 xfs_trans_t *tp, /* current transaction pointer */
2840 xfs_extnum_t idx, /* extent number to update/delete */
2841 xfs_bmap_free_t *flist, /* list of extents to be freed */
2842 xfs_btree_cur_t *cur, /* if null, not a btree */
2843 xfs_bmbt_irec_t *del, /* data to remove from extents */
2844 int *logflagsp, /* inode logging flags */
2845 int whichfork, /* data or attr fork */
2846 int rsvd) /* OK to allocate reserved blocks */
2847 {
2848 xfs_filblks_t da_new; /* new delay-alloc indirect blocks */
2849 xfs_filblks_t da_old; /* old delay-alloc indirect blocks */
2850 xfs_fsblock_t del_endblock=0; /* first block past del */
2851 xfs_fileoff_t del_endoff; /* first offset past del */
2852 int delay; /* current block is delayed allocated */
2853 int do_fx; /* free extent at end of routine */
2854 xfs_bmbt_rec_host_t *ep; /* current extent entry pointer */
2855 int error; /* error return value */
2856 int flags; /* inode logging flags */
2857 xfs_bmbt_irec_t got; /* current extent entry */
2858 xfs_fileoff_t got_endoff; /* first offset past got */
2859 int i; /* temp state */
2860 xfs_ifork_t *ifp; /* inode fork pointer */
2861 xfs_mount_t *mp; /* mount structure */
2862 xfs_filblks_t nblks; /* quota/sb block count */
2863 xfs_bmbt_irec_t new; /* new record to be inserted */
2864 /* REFERENCED */
2865 uint qfield; /* quota field to update */
2866 xfs_filblks_t temp; /* for indirect length calculations */
2867 xfs_filblks_t temp2; /* for indirect length calculations */
2868 int state = 0;
2869
2870 XFS_STATS_INC(xs_del_exlist);
2871
2872 if (whichfork == XFS_ATTR_FORK)
2873 state |= BMAP_ATTRFORK;
2874
2875 mp = ip->i_mount;
2876 ifp = XFS_IFORK_PTR(ip, whichfork);
2877 ASSERT((idx >= 0) && (idx < ifp->if_bytes /
2878 (uint)sizeof(xfs_bmbt_rec_t)));
2879 ASSERT(del->br_blockcount > 0);
2880 ep = xfs_iext_get_ext(ifp, idx);
2881 xfs_bmbt_get_all(ep, &got);
2882 ASSERT(got.br_startoff <= del->br_startoff);
2883 del_endoff = del->br_startoff + del->br_blockcount;
2884 got_endoff = got.br_startoff + got.br_blockcount;
2885 ASSERT(got_endoff >= del_endoff);
2886 delay = isnullstartblock(got.br_startblock);
2887 ASSERT(isnullstartblock(del->br_startblock) == delay);
2888 flags = 0;
2889 qfield = 0;
2890 error = 0;
2891 /*
2892 * If deleting a real allocation, must free up the disk space.
2893 */
2894 if (!delay) {
2895 flags = XFS_ILOG_CORE;
2896 /*
2897 * Realtime allocation. Free it and record di_nblocks update.
2898 */
2899 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
2900 xfs_fsblock_t bno;
2901 xfs_filblks_t len;
2902
2903 ASSERT(do_mod(del->br_blockcount,
2904 mp->m_sb.sb_rextsize) == 0);
2905 ASSERT(do_mod(del->br_startblock,
2906 mp->m_sb.sb_rextsize) == 0);
2907 bno = del->br_startblock;
2908 len = del->br_blockcount;
2909 do_div(bno, mp->m_sb.sb_rextsize);
2910 do_div(len, mp->m_sb.sb_rextsize);
2911 if ((error = xfs_rtfree_extent(ip->i_transp, bno,
2912 (xfs_extlen_t)len)))
2913 goto done;
2914 do_fx = 0;
2915 nblks = len * mp->m_sb.sb_rextsize;
2916 qfield = XFS_TRANS_DQ_RTBCOUNT;
2917 }
2918 /*
2919 * Ordinary allocation.
2920 */
2921 else {
2922 do_fx = 1;
2923 nblks = del->br_blockcount;
2924 qfield = XFS_TRANS_DQ_BCOUNT;
2925 }
2926 /*
2927 * Set up del_endblock and cur for later.
2928 */
2929 del_endblock = del->br_startblock + del->br_blockcount;
2930 if (cur) {
2931 if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
2932 got.br_startblock, got.br_blockcount,
2933 &i)))
2934 goto done;
2935 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2936 }
2937 da_old = da_new = 0;
2938 } else {
2939 da_old = startblockval(got.br_startblock);
2940 da_new = 0;
2941 nblks = 0;
2942 do_fx = 0;
2943 }
2944 /*
2945 * Set flag value to use in switch statement.
2946 * Left-contig is 2, right-contig is 1.
2947 */
2948 switch (((got.br_startoff == del->br_startoff) << 1) |
2949 (got_endoff == del_endoff)) {
2950 case 3:
2951 /*
2952 * Matches the whole extent. Delete the entry.
2953 */
2954 xfs_iext_remove(ip, idx, 1,
2955 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
2956 ifp->if_lastex = idx;
2957 if (delay)
2958 break;
2959 XFS_IFORK_NEXT_SET(ip, whichfork,
2960 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2961 flags |= XFS_ILOG_CORE;
2962 if (!cur) {
2963 flags |= xfs_ilog_fext(whichfork);
2964 break;
2965 }
2966 if ((error = xfs_btree_delete(cur, &i)))
2967 goto done;
2968 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2969 break;
2970
2971 case 2:
2972 /*
2973 * Deleting the first part of the extent.
2974 */
2975 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
2976 xfs_bmbt_set_startoff(ep, del_endoff);
2977 temp = got.br_blockcount - del->br_blockcount;
2978 xfs_bmbt_set_blockcount(ep, temp);
2979 ifp->if_lastex = idx;
2980 if (delay) {
2981 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2982 da_old);
2983 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2984 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
2985 da_new = temp;
2986 break;
2987 }
2988 xfs_bmbt_set_startblock(ep, del_endblock);
2989 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
2990 if (!cur) {
2991 flags |= xfs_ilog_fext(whichfork);
2992 break;
2993 }
2994 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
2995 got.br_blockcount - del->br_blockcount,
2996 got.br_state)))
2997 goto done;
2998 break;
2999
3000 case 1:
3001 /*
3002 * Deleting the last part of the extent.
3003 */
3004 temp = got.br_blockcount - del->br_blockcount;
3005 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
3006 xfs_bmbt_set_blockcount(ep, temp);
3007 ifp->if_lastex = idx;
3008 if (delay) {
3009 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
3010 da_old);
3011 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
3012 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
3013 da_new = temp;
3014 break;
3015 }
3016 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
3017 if (!cur) {
3018 flags |= xfs_ilog_fext(whichfork);
3019 break;
3020 }
3021 if ((error = xfs_bmbt_update(cur, got.br_startoff,
3022 got.br_startblock,
3023 got.br_blockcount - del->br_blockcount,
3024 got.br_state)))
3025 goto done;
3026 break;
3027
3028 case 0:
3029 /*
3030 * Deleting the middle of the extent.
3031 */
3032 temp = del->br_startoff - got.br_startoff;
3033 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
3034 xfs_bmbt_set_blockcount(ep, temp);
3035 new.br_startoff = del_endoff;
3036 temp2 = got_endoff - del_endoff;
3037 new.br_blockcount = temp2;
3038 new.br_state = got.br_state;
3039 if (!delay) {
3040 new.br_startblock = del_endblock;
3041 flags |= XFS_ILOG_CORE;
3042 if (cur) {
3043 if ((error = xfs_bmbt_update(cur,
3044 got.br_startoff,
3045 got.br_startblock, temp,
3046 got.br_state)))
3047 goto done;
3048 if ((error = xfs_btree_increment(cur, 0, &i)))
3049 goto done;
3050 cur->bc_rec.b = new;
3051 error = xfs_btree_insert(cur, &i);
3052 if (error && error != ENOSPC)
3053 goto done;
3054 /*
3055 * If get no-space back from btree insert,
3056 * it tried a split, and we have a zero
3057 * block reservation.
3058 * Fix up our state and return the error.
3059 */
3060 if (error == ENOSPC) {
3061 /*
3062 * Reset the cursor, don't trust
3063 * it after any insert operation.
3064 */
3065 if ((error = xfs_bmbt_lookup_eq(cur,
3066 got.br_startoff,
3067 got.br_startblock,
3068 temp, &i)))
3069 goto done;
3070 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3071 /*
3072 * Update the btree record back
3073 * to the original value.
3074 */
3075 if ((error = xfs_bmbt_update(cur,
3076 got.br_startoff,
3077 got.br_startblock,
3078 got.br_blockcount,
3079 got.br_state)))
3080 goto done;
3081 /*
3082 * Reset the extent record back
3083 * to the original value.
3084 */
3085 xfs_bmbt_set_blockcount(ep,
3086 got.br_blockcount);
3087 flags = 0;
3088 error = XFS_ERROR(ENOSPC);
3089 goto done;
3090 }
3091 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3092 } else
3093 flags |= xfs_ilog_fext(whichfork);
3094 XFS_IFORK_NEXT_SET(ip, whichfork,
3095 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
3096 } else {
3097 ASSERT(whichfork == XFS_DATA_FORK);
3098 temp = xfs_bmap_worst_indlen(ip, temp);
3099 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
3100 temp2 = xfs_bmap_worst_indlen(ip, temp2);
3101 new.br_startblock = nullstartblock((int)temp2);
3102 da_new = temp + temp2;
3103 while (da_new > da_old) {
3104 if (temp) {
3105 temp--;
3106 da_new--;
3107 xfs_bmbt_set_startblock(ep,
3108 nullstartblock((int)temp));
3109 }
3110 if (da_new == da_old)
3111 break;
3112 if (temp2) {
3113 temp2--;
3114 da_new--;
3115 new.br_startblock =
3116 nullstartblock((int)temp2);
3117 }
3118 }
3119 }
3120 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
3121 xfs_iext_insert(ip, idx + 1, 1, &new, state);
3122 ifp->if_lastex = idx + 1;
3123 break;
3124 }
3125 /*
3126 * If we need to, add to list of extents to delete.
3127 */
3128 if (do_fx)
3129 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
3130 mp);
3131 /*
3132 * Adjust inode # blocks in the file.
3133 */
3134 if (nblks)
3135 ip->i_d.di_nblocks -= nblks;
3136 /*
3137 * Adjust quota data.
3138 */
3139 if (qfield)
3140 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
3141
3142 /*
3143 * Account for change in delayed indirect blocks.
3144 * Nothing to do for disk quota accounting here.
3145 */
3146 ASSERT(da_old >= da_new);
3147 if (da_old > da_new) {
3148 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
3149 (int64_t)(da_old - da_new), rsvd);
3150 }
3151 done:
3152 *logflagsp = flags;
3153 return error;
3154 }
3155
3156 /*
3157 * Remove the entry "free" from the free item list. Prev points to the
3158 * previous entry, unless "free" is the head of the list.
3159 */
3160 STATIC void
3161 xfs_bmap_del_free(
3162 xfs_bmap_free_t *flist, /* free item list header */
3163 xfs_bmap_free_item_t *prev, /* previous item on list, if any */
3164 xfs_bmap_free_item_t *free) /* list item to be freed */
3165 {
3166 if (prev)
3167 prev->xbfi_next = free->xbfi_next;
3168 else
3169 flist->xbf_first = free->xbfi_next;
3170 flist->xbf_count--;
3171 kmem_zone_free(xfs_bmap_free_item_zone, free);
3172 }
3173
3174 /*
3175 * Convert an extents-format file into a btree-format file.
3176 * The new file will have a root block (in the inode) and a single child block.
3177 */
3178 STATIC int /* error */
3179 xfs_bmap_extents_to_btree(
3180 xfs_trans_t *tp, /* transaction pointer */
3181 xfs_inode_t *ip, /* incore inode pointer */
3182 xfs_fsblock_t *firstblock, /* first-block-allocated */
3183 xfs_bmap_free_t *flist, /* blocks freed in xaction */
3184 xfs_btree_cur_t **curp, /* cursor returned to caller */
3185 int wasdel, /* converting a delayed alloc */
3186 int *logflagsp, /* inode logging flags */
3187 int whichfork) /* data or attr fork */
3188 {
3189 struct xfs_btree_block *ablock; /* allocated (child) bt block */
3190 xfs_buf_t *abp; /* buffer for ablock */
3191 xfs_alloc_arg_t args; /* allocation arguments */
3192 xfs_bmbt_rec_t *arp; /* child record pointer */
3193 struct xfs_btree_block *block; /* btree root block */
3194 xfs_btree_cur_t *cur; /* bmap btree cursor */
3195 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
3196 int error; /* error return value */
3197 xfs_extnum_t i, cnt; /* extent record index */
3198 xfs_ifork_t *ifp; /* inode fork pointer */
3199 xfs_bmbt_key_t *kp; /* root block key pointer */
3200 xfs_mount_t *mp; /* mount structure */
3201 xfs_extnum_t nextents; /* number of file extents */
3202 xfs_bmbt_ptr_t *pp; /* root block address pointer */
3203
3204 ifp = XFS_IFORK_PTR(ip, whichfork);
3205 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3206 ASSERT(ifp->if_ext_max ==
3207 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
3208 /*
3209 * Make space in the inode incore.
3210 */
3211 xfs_iroot_realloc(ip, 1, whichfork);
3212 ifp->if_flags |= XFS_IFBROOT;
3213
3214 /*
3215 * Fill in the root.
3216 */
3217 block = ifp->if_broot;
3218 block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3219 block->bb_level = cpu_to_be16(1);
3220 block->bb_numrecs = cpu_to_be16(1);
3221 block->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3222 block->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3223
3224 /*
3225 * Need a cursor. Can't allocate until bb_level is filled in.
3226 */
3227 mp = ip->i_mount;
3228 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
3229 cur->bc_private.b.firstblock = *firstblock;
3230 cur->bc_private.b.flist = flist;
3231 cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3232 /*
3233 * Convert to a btree with two levels, one record in root.
3234 */
3235 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3236 args.tp = tp;
3237 args.mp = mp;
3238 args.firstblock = *firstblock;
3239 if (*firstblock == NULLFSBLOCK) {
3240 args.type = XFS_ALLOCTYPE_START_BNO;
3241 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3242 } else if (flist->xbf_low) {
3243 args.type = XFS_ALLOCTYPE_START_BNO;
3244 args.fsbno = *firstblock;
3245 } else {
3246 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3247 args.fsbno = *firstblock;
3248 }
3249 args.minlen = args.maxlen = args.prod = 1;
3250 args.total = args.minleft = args.alignment = args.mod = args.isfl =
3251 args.minalignslop = 0;
3252 args.wasdel = wasdel;
3253 *logflagsp = 0;
3254 if ((error = xfs_alloc_vextent(&args))) {
3255 xfs_iroot_realloc(ip, -1, whichfork);
3256 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3257 return error;
3258 }
3259 /*
3260 * Allocation can't fail, the space was reserved.
3261 */
3262 ASSERT(args.fsbno != NULLFSBLOCK);
3263 ASSERT(*firstblock == NULLFSBLOCK ||
3264 args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3265 (flist->xbf_low &&
3266 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3267 *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3268 cur->bc_private.b.allocated++;
3269 ip->i_d.di_nblocks++;
3270 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3271 abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3272 /*
3273 * Fill in the child block.
3274 */
3275 ablock = XFS_BUF_TO_BLOCK(abp);
3276 ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3277 ablock->bb_level = 0;
3278 ablock->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3279 ablock->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3280 arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3281 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3282 for (cnt = i = 0; i < nextents; i++) {
3283 ep = xfs_iext_get_ext(ifp, i);
3284 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
3285 arp->l0 = cpu_to_be64(ep->l0);
3286 arp->l1 = cpu_to_be64(ep->l1);
3287 arp++; cnt++;
3288 }
3289 }
3290 ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3291 xfs_btree_set_numrecs(ablock, cnt);
3292
3293 /*
3294 * Fill in the root key and pointer.
3295 */
3296 kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
3297 arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3298 kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
3299 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
3300 be16_to_cpu(block->bb_level)));
3301 *pp = cpu_to_be64(args.fsbno);
3302
3303 /*
3304 * Do all this logging at the end so that
3305 * the root is at the right level.
3306 */
3307 xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
3308 xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3309 ASSERT(*curp == NULL);
3310 *curp = cur;
3311 *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
3312 return 0;
3313 }
3314
3315 /*
3316 * Calculate the default attribute fork offset for newly created inodes.
3317 */
3318 uint
3319 xfs_default_attroffset(
3320 struct xfs_inode *ip)
3321 {
3322 struct xfs_mount *mp = ip->i_mount;
3323 uint offset;
3324
3325 if (mp->m_sb.sb_inodesize == 256) {
3326 offset = XFS_LITINO(mp) -
3327 XFS_BMDR_SPACE_CALC(MINABTPTRS);
3328 } else {
3329 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
3330 }
3331
3332 ASSERT(offset < XFS_LITINO(mp));
3333 return offset;
3334 }
3335
3336 /*
3337 * Helper routine to reset inode di_forkoff field when switching
3338 * attribute fork from local to extent format - we reset it where
3339 * possible to make space available for inline data fork extents.
3340 */
3341 STATIC void
3342 xfs_bmap_forkoff_reset(
3343 xfs_mount_t *mp,
3344 xfs_inode_t *ip,
3345 int whichfork)
3346 {
3347 if (whichfork == XFS_ATTR_FORK &&
3348 ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
3349 ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
3350 ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
3351 uint dfl_forkoff = xfs_default_attroffset(ip) >> 3;
3352
3353 if (dfl_forkoff > ip->i_d.di_forkoff) {
3354 ip->i_d.di_forkoff = dfl_forkoff;
3355 ip->i_df.if_ext_max =
3356 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t);
3357 ip->i_afp->if_ext_max =
3358 XFS_IFORK_ASIZE(ip) / sizeof(xfs_bmbt_rec_t);
3359 }
3360 }
3361 }
3362
3363 /*
3364 * Convert a local file to an extents file.
3365 * This code is out of bounds for data forks of regular files,
3366 * since the file data needs to get logged so things will stay consistent.
3367 * (The bmap-level manipulations are ok, though).
3368 */
3369 STATIC int /* error */
3370 xfs_bmap_local_to_extents(
3371 xfs_trans_t *tp, /* transaction pointer */
3372 xfs_inode_t *ip, /* incore inode pointer */
3373 xfs_fsblock_t *firstblock, /* first block allocated in xaction */
3374 xfs_extlen_t total, /* total blocks needed by transaction */
3375 int *logflagsp, /* inode logging flags */
3376 int whichfork) /* data or attr fork */
3377 {
3378 int error; /* error return value */
3379 int flags; /* logging flags returned */
3380 xfs_ifork_t *ifp; /* inode fork pointer */
3381
3382 /*
3383 * We don't want to deal with the case of keeping inode data inline yet.
3384 * So sending the data fork of a regular inode is invalid.
3385 */
3386 ASSERT(!((ip->i_d.di_mode & S_IFMT) == S_IFREG &&
3387 whichfork == XFS_DATA_FORK));
3388 ifp = XFS_IFORK_PTR(ip, whichfork);
3389 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3390 flags = 0;
3391 error = 0;
3392 if (ifp->if_bytes) {
3393 xfs_alloc_arg_t args; /* allocation arguments */
3394 xfs_buf_t *bp; /* buffer for extent block */
3395 xfs_bmbt_rec_host_t *ep;/* extent record pointer */
3396
3397 args.tp = tp;
3398 args.mp = ip->i_mount;
3399 args.firstblock = *firstblock;
3400 ASSERT((ifp->if_flags &
3401 (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == XFS_IFINLINE);
3402 /*
3403 * Allocate a block. We know we need only one, since the
3404 * file currently fits in an inode.
3405 */
3406 if (*firstblock == NULLFSBLOCK) {
3407 args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3408 args.type = XFS_ALLOCTYPE_START_BNO;
3409 } else {
3410 args.fsbno = *firstblock;
3411 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3412 }
3413 args.total = total;
3414 args.mod = args.minleft = args.alignment = args.wasdel =
3415 args.isfl = args.minalignslop = 0;
3416 args.minlen = args.maxlen = args.prod = 1;
3417 if ((error = xfs_alloc_vextent(&args)))
3418 goto done;
3419 /*
3420 * Can't fail, the space was reserved.
3421 */
3422 ASSERT(args.fsbno != NULLFSBLOCK);
3423 ASSERT(args.len == 1);
3424 *firstblock = args.fsbno;
3425 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3426 memcpy((char *)XFS_BUF_PTR(bp), ifp->if_u1.if_data,
3427 ifp->if_bytes);
3428 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3429 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3430 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3431 xfs_iext_add(ifp, 0, 1);
3432 ep = xfs_iext_get_ext(ifp, 0);
3433 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3434 trace_xfs_bmap_post_update(ip, 0,
3435 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
3436 _THIS_IP_);
3437 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3438 ip->i_d.di_nblocks = 1;
3439 xfs_trans_mod_dquot_byino(tp, ip,
3440 XFS_TRANS_DQ_BCOUNT, 1L);
3441 flags |= xfs_ilog_fext(whichfork);
3442 } else {
3443 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3444 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3445 }
3446 ifp->if_flags &= ~XFS_IFINLINE;
3447 ifp->if_flags |= XFS_IFEXTENTS;
3448 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3449 flags |= XFS_ILOG_CORE;
3450 done:
3451 *logflagsp = flags;
3452 return error;
3453 }
3454
3455 /*
3456 * Search the extent records for the entry containing block bno.
3457 * If bno lies in a hole, point to the next entry. If bno lies
3458 * past eof, *eofp will be set, and *prevp will contain the last
3459 * entry (null if none). Else, *lastxp will be set to the index
3460 * of the found entry; *gotp will contain the entry.
3461 */
3462 STATIC xfs_bmbt_rec_host_t * /* pointer to found extent entry */
3463 xfs_bmap_search_multi_extents(
3464 xfs_ifork_t *ifp, /* inode fork pointer */
3465 xfs_fileoff_t bno, /* block number searched for */
3466 int *eofp, /* out: end of file found */
3467 xfs_extnum_t *lastxp, /* out: last extent index */
3468 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
3469 xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
3470 {
3471 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
3472 xfs_extnum_t lastx; /* last extent index */
3473
3474 /*
3475 * Initialize the extent entry structure to catch access to
3476 * uninitialized br_startblock field.
3477 */
3478 gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
3479 gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3480 gotp->br_state = XFS_EXT_INVALID;
3481 #if XFS_BIG_BLKNOS
3482 gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
3483 #else
3484 gotp->br_startblock = 0xffffa5a5;
3485 #endif
3486 prevp->br_startoff = NULLFILEOFF;
3487
3488 ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
3489 if (lastx > 0) {
3490 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
3491 }
3492 if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
3493 xfs_bmbt_get_all(ep, gotp);
3494 *eofp = 0;
3495 } else {
3496 if (lastx > 0) {
3497 *gotp = *prevp;
3498 }
3499 *eofp = 1;
3500 ep = NULL;
3501 }
3502 *lastxp = lastx;
3503 return ep;
3504 }
3505
3506 /*
3507 * Search the extents list for the inode, for the extent containing bno.
3508 * If bno lies in a hole, point to the next entry. If bno lies past eof,
3509 * *eofp will be set, and *prevp will contain the last entry (null if none).
3510 * Else, *lastxp will be set to the index of the found
3511 * entry; *gotp will contain the entry.
3512 */
3513 STATIC xfs_bmbt_rec_host_t * /* pointer to found extent entry */
3514 xfs_bmap_search_extents(
3515 xfs_inode_t *ip, /* incore inode pointer */
3516 xfs_fileoff_t bno, /* block number searched for */
3517 int fork, /* data or attr fork */
3518 int *eofp, /* out: end of file found */
3519 xfs_extnum_t *lastxp, /* out: last extent index */
3520 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
3521 xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
3522 {
3523 xfs_ifork_t *ifp; /* inode fork pointer */
3524 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
3525
3526 XFS_STATS_INC(xs_look_exlist);
3527 ifp = XFS_IFORK_PTR(ip, fork);
3528
3529 ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
3530
3531 if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
3532 !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
3533 xfs_cmn_err(XFS_PTAG_FSBLOCK_ZERO, CE_ALERT, ip->i_mount,
3534 "Access to block zero in inode %llu "
3535 "start_block: %llx start_off: %llx "
3536 "blkcnt: %llx extent-state: %x lastx: %x\n",
3537 (unsigned long long)ip->i_ino,
3538 (unsigned long long)gotp->br_startblock,
3539 (unsigned long long)gotp->br_startoff,
3540 (unsigned long long)gotp->br_blockcount,
3541 gotp->br_state, *lastxp);
3542 *lastxp = NULLEXTNUM;
3543 *eofp = 1;
3544 return NULL;
3545 }
3546 return ep;
3547 }
3548
3549 /*
3550 * Compute the worst-case number of indirect blocks that will be used
3551 * for ip's delayed extent of length "len".
3552 */
3553 STATIC xfs_filblks_t
3554 xfs_bmap_worst_indlen(
3555 xfs_inode_t *ip, /* incore inode pointer */
3556 xfs_filblks_t len) /* delayed extent length */
3557 {
3558 int level; /* btree level number */
3559 int maxrecs; /* maximum record count at this level */
3560 xfs_mount_t *mp; /* mount structure */
3561 xfs_filblks_t rval; /* return value */
3562
3563 mp = ip->i_mount;
3564 maxrecs = mp->m_bmap_dmxr[0];
3565 for (level = 0, rval = 0;
3566 level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3567 level++) {
3568 len += maxrecs - 1;
3569 do_div(len, maxrecs);
3570 rval += len;
3571 if (len == 1)
3572 return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3573 level - 1;
3574 if (level == 0)
3575 maxrecs = mp->m_bmap_dmxr[1];
3576 }
3577 return rval;
3578 }
3579
3580 /*
3581 * Convert inode from non-attributed to attributed.
3582 * Must not be in a transaction, ip must not be locked.
3583 */
3584 int /* error code */
3585 xfs_bmap_add_attrfork(
3586 xfs_inode_t *ip, /* incore inode pointer */
3587 int size, /* space new attribute needs */
3588 int rsvd) /* xact may use reserved blks */
3589 {
3590 xfs_fsblock_t firstblock; /* 1st block/ag allocated */
3591 xfs_bmap_free_t flist; /* freed extent records */
3592 xfs_mount_t *mp; /* mount structure */
3593 xfs_trans_t *tp; /* transaction pointer */
3594 int blks; /* space reservation */
3595 int version = 1; /* superblock attr version */
3596 int committed; /* xaction was committed */
3597 int logflags; /* logging flags */
3598 int error; /* error return value */
3599
3600 ASSERT(XFS_IFORK_Q(ip) == 0);
3601 ASSERT(ip->i_df.if_ext_max ==
3602 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3603
3604 mp = ip->i_mount;
3605 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3606 tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3607 blks = XFS_ADDAFORK_SPACE_RES(mp);
3608 if (rsvd)
3609 tp->t_flags |= XFS_TRANS_RESERVE;
3610 if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3611 XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3612 goto error0;
3613 xfs_ilock(ip, XFS_ILOCK_EXCL);
3614 error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
3615 XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3616 XFS_QMOPT_RES_REGBLKS);
3617 if (error) {
3618 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3619 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3620 return error;
3621 }
3622 if (XFS_IFORK_Q(ip))
3623 goto error1;
3624 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3625 /*
3626 * For inodes coming from pre-6.2 filesystems.
3627 */
3628 ASSERT(ip->i_d.di_aformat == 0);
3629 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
3630 }
3631 ASSERT(ip->i_d.di_anextents == 0);
3632
3633 xfs_trans_ijoin_ref(tp, ip, XFS_ILOCK_EXCL);
3634 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3635
3636 switch (ip->i_d.di_format) {
3637 case XFS_DINODE_FMT_DEV:
3638 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
3639 break;
3640 case XFS_DINODE_FMT_UUID:
3641 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
3642 break;
3643 case XFS_DINODE_FMT_LOCAL:
3644 case XFS_DINODE_FMT_EXTENTS:
3645 case XFS_DINODE_FMT_BTREE:
3646 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
3647 if (!ip->i_d.di_forkoff)
3648 ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
3649 else if (mp->m_flags & XFS_MOUNT_ATTR2)
3650 version = 2;
3651 break;
3652 default:
3653 ASSERT(0);
3654 error = XFS_ERROR(EINVAL);
3655 goto error1;
3656 }
3657 ip->i_df.if_ext_max =
3658 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3659 ASSERT(ip->i_afp == NULL);
3660 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
3661 ip->i_afp->if_ext_max =
3662 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3663 ip->i_afp->if_flags = XFS_IFEXTENTS;
3664 logflags = 0;
3665 xfs_bmap_init(&flist, &firstblock);
3666 switch (ip->i_d.di_format) {
3667 case XFS_DINODE_FMT_LOCAL:
3668 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
3669 &logflags);
3670 break;
3671 case XFS_DINODE_FMT_EXTENTS:
3672 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
3673 &flist, &logflags);
3674 break;
3675 case XFS_DINODE_FMT_BTREE:
3676 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
3677 &logflags);
3678 break;
3679 default:
3680 error = 0;
3681 break;
3682 }
3683 if (logflags)
3684 xfs_trans_log_inode(tp, ip, logflags);
3685 if (error)
3686 goto error2;
3687 if (!xfs_sb_version_hasattr(&mp->m_sb) ||
3688 (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
3689 __int64_t sbfields = 0;
3690
3691 spin_lock(&mp->m_sb_lock);
3692 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
3693 xfs_sb_version_addattr(&mp->m_sb);
3694 sbfields |= XFS_SB_VERSIONNUM;
3695 }
3696 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
3697 xfs_sb_version_addattr2(&mp->m_sb);
3698 sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
3699 }
3700 if (sbfields) {
3701 spin_unlock(&mp->m_sb_lock);
3702 xfs_mod_sb(tp, sbfields);
3703 } else
3704 spin_unlock(&mp->m_sb_lock);
3705 }
3706 if ((error = xfs_bmap_finish(&tp, &flist, &committed)))
3707 goto error2;
3708 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3709 ASSERT(ip->i_df.if_ext_max ==
3710 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3711 return error;
3712 error2:
3713 xfs_bmap_cancel(&flist);
3714 error1:
3715 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3716 error0:
3717 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
3718 ASSERT(ip->i_df.if_ext_max ==
3719 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3720 return error;
3721 }
3722
3723 /*
3724 * Add the extent to the list of extents to be free at transaction end.
3725 * The list is maintained sorted (by block number).
3726 */
3727 /* ARGSUSED */
3728 void
3729 xfs_bmap_add_free(
3730 xfs_fsblock_t bno, /* fs block number of extent */
3731 xfs_filblks_t len, /* length of extent */
3732 xfs_bmap_free_t *flist, /* list of extents */
3733 xfs_mount_t *mp) /* mount point structure */
3734 {
3735 xfs_bmap_free_item_t *cur; /* current (next) element */
3736 xfs_bmap_free_item_t *new; /* new element */
3737 xfs_bmap_free_item_t *prev; /* previous element */
3738 #ifdef DEBUG
3739 xfs_agnumber_t agno;
3740 xfs_agblock_t agbno;
3741
3742 ASSERT(bno != NULLFSBLOCK);
3743 ASSERT(len > 0);
3744 ASSERT(len <= MAXEXTLEN);
3745 ASSERT(!isnullstartblock(bno));
3746 agno = XFS_FSB_TO_AGNO(mp, bno);
3747 agbno = XFS_FSB_TO_AGBNO(mp, bno);
3748 ASSERT(agno < mp->m_sb.sb_agcount);
3749 ASSERT(agbno < mp->m_sb.sb_agblocks);
3750 ASSERT(len < mp->m_sb.sb_agblocks);
3751 ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
3752 #endif
3753 ASSERT(xfs_bmap_free_item_zone != NULL);
3754 new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
3755 new->xbfi_startblock = bno;
3756 new->xbfi_blockcount = (xfs_extlen_t)len;
3757 for (prev = NULL, cur = flist->xbf_first;
3758 cur != NULL;
3759 prev = cur, cur = cur->xbfi_next) {
3760 if (cur->xbfi_startblock >= bno)
3761 break;
3762 }
3763 if (prev)
3764 prev->xbfi_next = new;
3765 else
3766 flist->xbf_first = new;
3767 new->xbfi_next = cur;
3768 flist->xbf_count++;
3769 }
3770
3771 /*
3772 * Compute and fill in the value of the maximum depth of a bmap btree
3773 * in this filesystem. Done once, during mount.
3774 */
3775 void
3776 xfs_bmap_compute_maxlevels(
3777 xfs_mount_t *mp, /* file system mount structure */
3778 int whichfork) /* data or attr fork */
3779 {
3780 int level; /* btree level */
3781 uint maxblocks; /* max blocks at this level */
3782 uint maxleafents; /* max leaf entries possible */
3783 int maxrootrecs; /* max records in root block */
3784 int minleafrecs; /* min records in leaf block */
3785 int minnoderecs; /* min records in node block */
3786 int sz; /* root block size */
3787
3788 /*
3789 * The maximum number of extents in a file, hence the maximum
3790 * number of leaf entries, is controlled by the type of di_nextents
3791 * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
3792 * (a signed 16-bit number, xfs_aextnum_t).
3793 *
3794 * Note that we can no longer assume that if we are in ATTR1 that
3795 * the fork offset of all the inodes will be
3796 * (xfs_default_attroffset(ip) >> 3) because we could have mounted
3797 * with ATTR2 and then mounted back with ATTR1, keeping the
3798 * di_forkoff's fixed but probably at various positions. Therefore,
3799 * for both ATTR1 and ATTR2 we have to assume the worst case scenario
3800 * of a minimum size available.
3801 */
3802 if (whichfork == XFS_DATA_FORK) {
3803 maxleafents = MAXEXTNUM;
3804 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
3805 } else {
3806 maxleafents = MAXAEXTNUM;
3807 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
3808 }
3809 maxrootrecs = xfs_bmdr_maxrecs(mp, sz, 0);
3810 minleafrecs = mp->m_bmap_dmnr[0];
3811 minnoderecs = mp->m_bmap_dmnr[1];
3812 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
3813 for (level = 1; maxblocks > 1; level++) {
3814 if (maxblocks <= maxrootrecs)
3815 maxblocks = 1;
3816 else
3817 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
3818 }
3819 mp->m_bm_maxlevels[whichfork] = level;
3820 }
3821
3822 /*
3823 * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
3824 * caller. Frees all the extents that need freeing, which must be done
3825 * last due to locking considerations. We never free any extents in
3826 * the first transaction. This is to allow the caller to make the first
3827 * transaction a synchronous one so that the pointers to the data being
3828 * broken in this transaction will be permanent before the data is actually
3829 * freed. This is necessary to prevent blocks from being reallocated
3830 * and written to before the free and reallocation are actually permanent.
3831 * We do not just make the first transaction synchronous here, because
3832 * there are more efficient ways to gain the same protection in some cases
3833 * (see the file truncation code).
3834 *
3835 * Return 1 if the given transaction was committed and a new one
3836 * started, and 0 otherwise in the committed parameter.
3837 */
3838 /*ARGSUSED*/
3839 int /* error */
3840 xfs_bmap_finish(
3841 xfs_trans_t **tp, /* transaction pointer addr */
3842 xfs_bmap_free_t *flist, /* i/o: list extents to free */
3843 int *committed) /* xact committed or not */
3844 {
3845 xfs_efd_log_item_t *efd; /* extent free data */
3846 xfs_efi_log_item_t *efi; /* extent free intention */
3847 int error; /* error return value */
3848 xfs_bmap_free_item_t *free; /* free extent item */
3849 unsigned int logres; /* new log reservation */
3850 unsigned int logcount; /* new log count */
3851 xfs_mount_t *mp; /* filesystem mount structure */
3852 xfs_bmap_free_item_t *next; /* next item on free list */
3853 xfs_trans_t *ntp; /* new transaction pointer */
3854
3855 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
3856 if (flist->xbf_count == 0) {
3857 *committed = 0;
3858 return 0;
3859 }
3860 ntp = *tp;
3861 efi = xfs_trans_get_efi(ntp, flist->xbf_count);
3862 for (free = flist->xbf_first; free; free = free->xbfi_next)
3863 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
3864 free->xbfi_blockcount);
3865 logres = ntp->t_log_res;
3866 logcount = ntp->t_log_count;
3867 ntp = xfs_trans_dup(*tp);
3868 error = xfs_trans_commit(*tp, 0);
3869 *tp = ntp;
3870 *committed = 1;
3871 /*
3872 * We have a new transaction, so we should return committed=1,
3873 * even though we're returning an error.
3874 */
3875 if (error)
3876 return error;
3877
3878 /*
3879 * transaction commit worked ok so we can drop the extra ticket
3880 * reference that we gained in xfs_trans_dup()
3881 */
3882 xfs_log_ticket_put(ntp->t_ticket);
3883
3884 if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
3885 logcount)))
3886 return error;
3887 efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
3888 for (free = flist->xbf_first; free != NULL; free = next) {
3889 next = free->xbfi_next;
3890 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
3891 free->xbfi_blockcount))) {
3892 /*
3893 * The bmap free list will be cleaned up at a
3894 * higher level. The EFI will be canceled when
3895 * this transaction is aborted.
3896 * Need to force shutdown here to make sure it
3897 * happens, since this transaction may not be
3898 * dirty yet.
3899 */
3900 mp = ntp->t_mountp;
3901 if (!XFS_FORCED_SHUTDOWN(mp))
3902 xfs_force_shutdown(mp,
3903 (error == EFSCORRUPTED) ?
3904 SHUTDOWN_CORRUPT_INCORE :
3905 SHUTDOWN_META_IO_ERROR);
3906 return error;
3907 }
3908 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
3909 free->xbfi_blockcount);
3910 xfs_bmap_del_free(flist, NULL, free);
3911 }
3912 return 0;
3913 }
3914
3915 /*
3916 * Free up any items left in the list.
3917 */
3918 void
3919 xfs_bmap_cancel(
3920 xfs_bmap_free_t *flist) /* list of bmap_free_items */
3921 {
3922 xfs_bmap_free_item_t *free; /* free list item */
3923 xfs_bmap_free_item_t *next;
3924
3925 if (flist->xbf_count == 0)
3926 return;
3927 ASSERT(flist->xbf_first != NULL);
3928 for (free = flist->xbf_first; free; free = next) {
3929 next = free->xbfi_next;
3930 xfs_bmap_del_free(flist, NULL, free);
3931 }
3932 ASSERT(flist->xbf_count == 0);
3933 }
3934
3935 /*
3936 * Returns the file-relative block number of the first unused block(s)
3937 * in the file with at least "len" logically contiguous blocks free.
3938 * This is the lowest-address hole if the file has holes, else the first block
3939 * past the end of file.
3940 * Return 0 if the file is currently local (in-inode).
3941 */
3942 int /* error */
3943 xfs_bmap_first_unused(
3944 xfs_trans_t *tp, /* transaction pointer */
3945 xfs_inode_t *ip, /* incore inode */
3946 xfs_extlen_t len, /* size of hole to find */
3947 xfs_fileoff_t *first_unused, /* unused block */
3948 int whichfork) /* data or attr fork */
3949 {
3950 int error; /* error return value */
3951 int idx; /* extent record index */
3952 xfs_ifork_t *ifp; /* inode fork pointer */
3953 xfs_fileoff_t lastaddr; /* last block number seen */
3954 xfs_fileoff_t lowest; /* lowest useful block */
3955 xfs_fileoff_t max; /* starting useful block */
3956 xfs_fileoff_t off; /* offset for this block */
3957 xfs_extnum_t nextents; /* number of extent entries */
3958
3959 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
3960 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
3961 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3962 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3963 *first_unused = 0;
3964 return 0;
3965 }
3966 ifp = XFS_IFORK_PTR(ip, whichfork);
3967 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3968 (error = xfs_iread_extents(tp, ip, whichfork)))
3969 return error;
3970 lowest = *first_unused;
3971 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3972 for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
3973 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
3974 off = xfs_bmbt_get_startoff(ep);
3975 /*
3976 * See if the hole before this extent will work.
3977 */
3978 if (off >= lowest + len && off - max >= len) {
3979 *first_unused = max;
3980 return 0;
3981 }
3982 lastaddr = off + xfs_bmbt_get_blockcount(ep);
3983 max = XFS_FILEOFF_MAX(lastaddr, lowest);
3984 }
3985 *first_unused = max;
3986 return 0;
3987 }
3988
3989 /*
3990 * Returns the file-relative block number of the last block + 1 before
3991 * last_block (input value) in the file.
3992 * This is not based on i_size, it is based on the extent records.
3993 * Returns 0 for local files, as they do not have extent records.
3994 */
3995 int /* error */
3996 xfs_bmap_last_before(
3997 xfs_trans_t *tp, /* transaction pointer */
3998 xfs_inode_t *ip, /* incore inode */
3999 xfs_fileoff_t *last_block, /* last block */
4000 int whichfork) /* data or attr fork */
4001 {
4002 xfs_fileoff_t bno; /* input file offset */
4003 int eof; /* hit end of file */
4004 xfs_bmbt_rec_host_t *ep; /* pointer to last extent */
4005 int error; /* error return value */
4006 xfs_bmbt_irec_t got; /* current extent value */
4007 xfs_ifork_t *ifp; /* inode fork pointer */
4008 xfs_extnum_t lastx; /* last extent used */
4009 xfs_bmbt_irec_t prev; /* previous extent value */
4010
4011 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4012 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4013 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4014 return XFS_ERROR(EIO);
4015 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4016 *last_block = 0;
4017 return 0;
4018 }
4019 ifp = XFS_IFORK_PTR(ip, whichfork);
4020 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4021 (error = xfs_iread_extents(tp, ip, whichfork)))
4022 return error;
4023 bno = *last_block - 1;
4024 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4025 &prev);
4026 if (eof || xfs_bmbt_get_startoff(ep) > bno) {
4027 if (prev.br_startoff == NULLFILEOFF)
4028 *last_block = 0;
4029 else
4030 *last_block = prev.br_startoff + prev.br_blockcount;
4031 }
4032 /*
4033 * Otherwise *last_block is already the right answer.
4034 */
4035 return 0;
4036 }
4037
4038 /*
4039 * Returns the file-relative block number of the first block past eof in
4040 * the file. This is not based on i_size, it is based on the extent records.
4041 * Returns 0 for local files, as they do not have extent records.
4042 */
4043 int /* error */
4044 xfs_bmap_last_offset(
4045 xfs_trans_t *tp, /* transaction pointer */
4046 xfs_inode_t *ip, /* incore inode */
4047 xfs_fileoff_t *last_block, /* last block */
4048 int whichfork) /* data or attr fork */
4049 {
4050 xfs_bmbt_rec_host_t *ep; /* pointer to last extent */
4051 int error; /* error return value */
4052 xfs_ifork_t *ifp; /* inode fork pointer */
4053 xfs_extnum_t nextents; /* number of extent entries */
4054
4055 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4056 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4057 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4058 return XFS_ERROR(EIO);
4059 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4060 *last_block = 0;
4061 return 0;
4062 }
4063 ifp = XFS_IFORK_PTR(ip, whichfork);
4064 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4065 (error = xfs_iread_extents(tp, ip, whichfork)))
4066 return error;
4067 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4068 if (!nextents) {
4069 *last_block = 0;
4070 return 0;
4071 }
4072 ep = xfs_iext_get_ext(ifp, nextents - 1);
4073 *last_block = xfs_bmbt_get_startoff(ep) + xfs_bmbt_get_blockcount(ep);
4074 return 0;
4075 }
4076
4077 /*
4078 * Returns whether the selected fork of the inode has exactly one
4079 * block or not. For the data fork we check this matches di_size,
4080 * implying the file's range is 0..bsize-1.
4081 */
4082 int /* 1=>1 block, 0=>otherwise */
4083 xfs_bmap_one_block(
4084 xfs_inode_t *ip, /* incore inode */
4085 int whichfork) /* data or attr fork */
4086 {
4087 xfs_bmbt_rec_host_t *ep; /* ptr to fork's extent */
4088 xfs_ifork_t *ifp; /* inode fork pointer */
4089 int rval; /* return value */
4090 xfs_bmbt_irec_t s; /* internal version of extent */
4091
4092 #ifndef DEBUG
4093 if (whichfork == XFS_DATA_FORK) {
4094 return ((ip->i_d.di_mode & S_IFMT) == S_IFREG) ?
4095 (ip->i_size == ip->i_mount->m_sb.sb_blocksize) :
4096 (ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize);
4097 }
4098 #endif /* !DEBUG */
4099 if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4100 return 0;
4101 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4102 return 0;
4103 ifp = XFS_IFORK_PTR(ip, whichfork);
4104 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4105 ep = xfs_iext_get_ext(ifp, 0);
4106 xfs_bmbt_get_all(ep, &s);
4107 rval = s.br_startoff == 0 && s.br_blockcount == 1;
4108 if (rval && whichfork == XFS_DATA_FORK)
4109 ASSERT(ip->i_size == ip->i_mount->m_sb.sb_blocksize);
4110 return rval;
4111 }
4112
4113 STATIC int
4114 xfs_bmap_sanity_check(
4115 struct xfs_mount *mp,
4116 struct xfs_buf *bp,
4117 int level)
4118 {
4119 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
4120
4121 if (be32_to_cpu(block->bb_magic) != XFS_BMAP_MAGIC ||
4122 be16_to_cpu(block->bb_level) != level ||
4123 be16_to_cpu(block->bb_numrecs) == 0 ||
4124 be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
4125 return 0;
4126 return 1;
4127 }
4128
4129 /*
4130 * Read in the extents to if_extents.
4131 * All inode fields are set up by caller, we just traverse the btree
4132 * and copy the records in. If the file system cannot contain unwritten
4133 * extents, the records are checked for no "state" flags.
4134 */
4135 int /* error */
4136 xfs_bmap_read_extents(
4137 xfs_trans_t *tp, /* transaction pointer */
4138 xfs_inode_t *ip, /* incore inode */
4139 int whichfork) /* data or attr fork */
4140 {
4141 struct xfs_btree_block *block; /* current btree block */
4142 xfs_fsblock_t bno; /* block # of "block" */
4143 xfs_buf_t *bp; /* buffer for "block" */
4144 int error; /* error return value */
4145 xfs_exntfmt_t exntf; /* XFS_EXTFMT_NOSTATE, if checking */
4146 xfs_extnum_t i, j; /* index into the extents list */
4147 xfs_ifork_t *ifp; /* fork structure */
4148 int level; /* btree level, for checking */
4149 xfs_mount_t *mp; /* file system mount structure */
4150 __be64 *pp; /* pointer to block address */
4151 /* REFERENCED */
4152 xfs_extnum_t room; /* number of entries there's room for */
4153
4154 bno = NULLFSBLOCK;
4155 mp = ip->i_mount;
4156 ifp = XFS_IFORK_PTR(ip, whichfork);
4157 exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4158 XFS_EXTFMT_INODE(ip);
4159 block = ifp->if_broot;
4160 /*
4161 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4162 */
4163 level = be16_to_cpu(block->bb_level);
4164 ASSERT(level > 0);
4165 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
4166 bno = be64_to_cpu(*pp);
4167 ASSERT(bno != NULLDFSBNO);
4168 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
4169 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
4170 /*
4171 * Go down the tree until leaf level is reached, following the first
4172 * pointer (leftmost) at each level.
4173 */
4174 while (level-- > 0) {
4175 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4176 XFS_BMAP_BTREE_REF)))
4177 return error;
4178 block = XFS_BUF_TO_BLOCK(bp);
4179 XFS_WANT_CORRUPTED_GOTO(
4180 xfs_bmap_sanity_check(mp, bp, level),
4181 error0);
4182 if (level == 0)
4183 break;
4184 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
4185 bno = be64_to_cpu(*pp);
4186 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
4187 xfs_trans_brelse(tp, bp);
4188 }
4189 /*
4190 * Here with bp and block set to the leftmost leaf node in the tree.
4191 */
4192 room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4193 i = 0;
4194 /*
4195 * Loop over all leaf nodes. Copy information to the extent records.
4196 */
4197 for (;;) {
4198 xfs_bmbt_rec_t *frp;
4199 xfs_fsblock_t nextbno;
4200 xfs_extnum_t num_recs;
4201 xfs_extnum_t start;
4202
4203
4204 num_recs = xfs_btree_get_numrecs(block);
4205 if (unlikely(i + num_recs > room)) {
4206 ASSERT(i + num_recs <= room);
4207 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
4208 "corrupt dinode %Lu, (btree extents).",
4209 (unsigned long long) ip->i_ino);
4210 XFS_ERROR_REPORT("xfs_bmap_read_extents(1)",
4211 XFS_ERRLEVEL_LOW,
4212 ip->i_mount);
4213 goto error0;
4214 }
4215 XFS_WANT_CORRUPTED_GOTO(
4216 xfs_bmap_sanity_check(mp, bp, 0),
4217 error0);
4218 /*
4219 * Read-ahead the next leaf block, if any.
4220 */
4221 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
4222 if (nextbno != NULLFSBLOCK)
4223 xfs_btree_reada_bufl(mp, nextbno, 1);
4224 /*
4225 * Copy records into the extent records.
4226 */
4227 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
4228 start = i;
4229 for (j = 0; j < num_recs; j++, i++, frp++) {
4230 xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
4231 trp->l0 = be64_to_cpu(frp->l0);
4232 trp->l1 = be64_to_cpu(frp->l1);
4233 }
4234 if (exntf == XFS_EXTFMT_NOSTATE) {
4235 /*
4236 * Check all attribute bmap btree records and
4237 * any "older" data bmap btree records for a
4238 * set bit in the "extent flag" position.
4239 */
4240 if (unlikely(xfs_check_nostate_extents(ifp,
4241 start, num_recs))) {
4242 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4243 XFS_ERRLEVEL_LOW,
4244 ip->i_mount);
4245 goto error0;
4246 }
4247 }
4248 xfs_trans_brelse(tp, bp);
4249 bno = nextbno;
4250 /*
4251 * If we've reached the end, stop.
4252 */
4253 if (bno == NULLFSBLOCK)
4254 break;
4255 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4256 XFS_BMAP_BTREE_REF)))
4257 return error;
4258 block = XFS_BUF_TO_BLOCK(bp);
4259 }
4260 ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4261 ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4262 XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
4263 return 0;
4264 error0:
4265 xfs_trans_brelse(tp, bp);
4266 return XFS_ERROR(EFSCORRUPTED);
4267 }
4268
4269 #ifdef DEBUG
4270 /*
4271 * Add bmap trace insert entries for all the contents of the extent records.
4272 */
4273 void
4274 xfs_bmap_trace_exlist(
4275 xfs_inode_t *ip, /* incore inode pointer */
4276 xfs_extnum_t cnt, /* count of entries in the list */
4277 int whichfork, /* data or attr fork */
4278 unsigned long caller_ip)
4279 {
4280 xfs_extnum_t idx; /* extent record index */
4281 xfs_ifork_t *ifp; /* inode fork pointer */
4282 int state = 0;
4283
4284 if (whichfork == XFS_ATTR_FORK)
4285 state |= BMAP_ATTRFORK;
4286
4287 ifp = XFS_IFORK_PTR(ip, whichfork);
4288 ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4289 for (idx = 0; idx < cnt; idx++)
4290 trace_xfs_extlist(ip, idx, whichfork, caller_ip);
4291 }
4292
4293 /*
4294 * Validate that the bmbt_irecs being returned from bmapi are valid
4295 * given the callers original parameters. Specifically check the
4296 * ranges of the returned irecs to ensure that they only extent beyond
4297 * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4298 */
4299 STATIC void
4300 xfs_bmap_validate_ret(
4301 xfs_fileoff_t bno,
4302 xfs_filblks_t len,
4303 int flags,
4304 xfs_bmbt_irec_t *mval,
4305 int nmap,
4306 int ret_nmap)
4307 {
4308 int i; /* index to map values */
4309
4310 ASSERT(ret_nmap <= nmap);
4311
4312 for (i = 0; i < ret_nmap; i++) {
4313 ASSERT(mval[i].br_blockcount > 0);
4314 if (!(flags & XFS_BMAPI_ENTIRE)) {
4315 ASSERT(mval[i].br_startoff >= bno);
4316 ASSERT(mval[i].br_blockcount <= len);
4317 ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4318 bno + len);
4319 } else {
4320 ASSERT(mval[i].br_startoff < bno + len);
4321 ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4322 bno);
4323 }
4324 ASSERT(i == 0 ||
4325 mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4326 mval[i].br_startoff);
4327 if ((flags & XFS_BMAPI_WRITE) && !(flags & XFS_BMAPI_DELAY))
4328 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4329 mval[i].br_startblock != HOLESTARTBLOCK);
4330 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4331 mval[i].br_state == XFS_EXT_UNWRITTEN);
4332 }
4333 }
4334 #endif /* DEBUG */
4335
4336
4337 /*
4338 * Map file blocks to filesystem blocks.
4339 * File range is given by the bno/len pair.
4340 * Adds blocks to file if a write ("flags & XFS_BMAPI_WRITE" set)
4341 * into a hole or past eof.
4342 * Only allocates blocks from a single allocation group,
4343 * to avoid locking problems.
4344 * The returned value in "firstblock" from the first call in a transaction
4345 * must be remembered and presented to subsequent calls in "firstblock".
4346 * An upper bound for the number of blocks to be allocated is supplied to
4347 * the first call in "total"; if no allocation group has that many free
4348 * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4349 */
4350 int /* error */
4351 xfs_bmapi(
4352 xfs_trans_t *tp, /* transaction pointer */
4353 xfs_inode_t *ip, /* incore inode */
4354 xfs_fileoff_t bno, /* starting file offs. mapped */
4355 xfs_filblks_t len, /* length to map in file */
4356 int flags, /* XFS_BMAPI_... */
4357 xfs_fsblock_t *firstblock, /* first allocated block
4358 controls a.g. for allocs */
4359 xfs_extlen_t total, /* total blocks needed */
4360 xfs_bmbt_irec_t *mval, /* output: map values */
4361 int *nmap, /* i/o: mval size/count */
4362 xfs_bmap_free_t *flist) /* i/o: list extents to free */
4363 {
4364 xfs_fsblock_t abno; /* allocated block number */
4365 xfs_extlen_t alen; /* allocated extent length */
4366 xfs_fileoff_t aoff; /* allocated file offset */
4367 xfs_bmalloca_t bma = { 0 }; /* args for xfs_bmap_alloc */
4368 xfs_btree_cur_t *cur; /* bmap btree cursor */
4369 xfs_fileoff_t end; /* end of mapped file region */
4370 int eof; /* we've hit the end of extents */
4371 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
4372 int error; /* error return */
4373 xfs_bmbt_irec_t got; /* current file extent record */
4374 xfs_ifork_t *ifp; /* inode fork pointer */
4375 xfs_extlen_t indlen; /* indirect blocks length */
4376 xfs_extnum_t lastx; /* last useful extent number */
4377 int logflags; /* flags for transaction logging */
4378 xfs_extlen_t minleft; /* min blocks left after allocation */
4379 xfs_extlen_t minlen; /* min allocation size */
4380 xfs_mount_t *mp; /* xfs mount structure */
4381 int n; /* current extent index */
4382 int nallocs; /* number of extents alloc'd */
4383 xfs_extnum_t nextents; /* number of extents in file */
4384 xfs_fileoff_t obno; /* old block number (offset) */
4385 xfs_bmbt_irec_t prev; /* previous file extent record */
4386 int tmp_logflags; /* temp flags holder */
4387 int whichfork; /* data or attr fork */
4388 char inhole; /* current location is hole in file */
4389 char wasdelay; /* old extent was delayed */
4390 char wr; /* this is a write request */
4391 char rt; /* this is a realtime file */
4392 #ifdef DEBUG
4393 xfs_fileoff_t orig_bno; /* original block number value */
4394 int orig_flags; /* original flags arg value */
4395 xfs_filblks_t orig_len; /* original value of len arg */
4396 xfs_bmbt_irec_t *orig_mval; /* original value of mval */
4397 int orig_nmap; /* original value of *nmap */
4398
4399 orig_bno = bno;
4400 orig_len = len;
4401 orig_flags = flags;
4402 orig_mval = mval;
4403 orig_nmap = *nmap;
4404 #endif
4405 ASSERT(*nmap >= 1);
4406 ASSERT(*nmap <= XFS_BMAP_MAX_NMAP || !(flags & XFS_BMAPI_WRITE));
4407 whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4408 XFS_ATTR_FORK : XFS_DATA_FORK;
4409 mp = ip->i_mount;
4410 if (unlikely(XFS_TEST_ERROR(
4411 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4412 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4413 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4414 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4415 XFS_ERROR_REPORT("xfs_bmapi", XFS_ERRLEVEL_LOW, mp);
4416 return XFS_ERROR(EFSCORRUPTED);
4417 }
4418 if (XFS_FORCED_SHUTDOWN(mp))
4419 return XFS_ERROR(EIO);
4420 rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4421 ifp = XFS_IFORK_PTR(ip, whichfork);
4422 ASSERT(ifp->if_ext_max ==
4423 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4424 if ((wr = (flags & XFS_BMAPI_WRITE)) != 0)
4425 XFS_STATS_INC(xs_blk_mapw);
4426 else
4427 XFS_STATS_INC(xs_blk_mapr);
4428 /*
4429 * IGSTATE flag is used to combine extents which
4430 * differ only due to the state of the extents.
4431 * This technique is used from xfs_getbmap()
4432 * when the caller does not wish to see the
4433 * separation (which is the default).
4434 *
4435 * This technique is also used when writing a
4436 * buffer which has been partially written,
4437 * (usually by being flushed during a chunkread),
4438 * to ensure one write takes place. This also
4439 * prevents a change in the xfs inode extents at
4440 * this time, intentionally. This change occurs
4441 * on completion of the write operation, in
4442 * xfs_strat_comp(), where the xfs_bmapi() call
4443 * is transactioned, and the extents combined.
4444 */
4445 if ((flags & XFS_BMAPI_IGSTATE) && wr) /* if writing unwritten space */
4446 wr = 0; /* no allocations are allowed */
4447 ASSERT(wr || !(flags & XFS_BMAPI_DELAY));
4448 logflags = 0;
4449 nallocs = 0;
4450 cur = NULL;
4451 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4452 ASSERT(wr && tp);
4453 if ((error = xfs_bmap_local_to_extents(tp, ip,
4454 firstblock, total, &logflags, whichfork)))
4455 goto error0;
4456 }
4457 if (wr && *firstblock == NULLFSBLOCK) {
4458 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4459 minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4460 else
4461 minleft = 1;
4462 } else
4463 minleft = 0;
4464 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4465 (error = xfs_iread_extents(tp, ip, whichfork)))
4466 goto error0;
4467 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4468 &prev);
4469 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4470 n = 0;
4471 end = bno + len;
4472 obno = bno;
4473 bma.ip = NULL;
4474
4475 while (bno < end && n < *nmap) {
4476 /*
4477 * Reading past eof, act as though there's a hole
4478 * up to end.
4479 */
4480 if (eof && !wr)
4481 got.br_startoff = end;
4482 inhole = eof || got.br_startoff > bno;
4483 wasdelay = wr && !inhole && !(flags & XFS_BMAPI_DELAY) &&
4484 isnullstartblock(got.br_startblock);
4485 /*
4486 * First, deal with the hole before the allocated space
4487 * that we found, if any.
4488 */
4489 if (wr && (inhole || wasdelay)) {
4490 /*
4491 * For the wasdelay case, we could also just
4492 * allocate the stuff asked for in this bmap call
4493 * but that wouldn't be as good.
4494 */
4495 if (wasdelay) {
4496 alen = (xfs_extlen_t)got.br_blockcount;
4497 aoff = got.br_startoff;
4498 if (lastx != NULLEXTNUM && lastx) {
4499 ep = xfs_iext_get_ext(ifp, lastx - 1);
4500 xfs_bmbt_get_all(ep, &prev);
4501 }
4502 } else {
4503 alen = (xfs_extlen_t)
4504 XFS_FILBLKS_MIN(len, MAXEXTLEN);
4505 if (!eof)
4506 alen = (xfs_extlen_t)
4507 XFS_FILBLKS_MIN(alen,
4508 got.br_startoff - bno);
4509 aoff = bno;
4510 }
4511 minlen = (flags & XFS_BMAPI_CONTIG) ? alen : 1;
4512 if (flags & XFS_BMAPI_DELAY) {
4513 xfs_extlen_t extsz;
4514
4515 /* Figure out the extent size, adjust alen */
4516 extsz = xfs_get_extsz_hint(ip);
4517 if (extsz) {
4518 /*
4519 * make sure we don't exceed a single
4520 * extent length when we align the
4521 * extent by reducing length we are
4522 * going to allocate by the maximum
4523 * amount extent size aligment may
4524 * require.
4525 */
4526 alen = XFS_FILBLKS_MIN(len,
4527 MAXEXTLEN - (2 * extsz - 1));
4528 error = xfs_bmap_extsize_align(mp,
4529 &got, &prev, extsz,
4530 rt, eof,
4531 flags&XFS_BMAPI_DELAY,
4532 flags&XFS_BMAPI_CONVERT,
4533 &aoff, &alen);
4534 ASSERT(!error);
4535 }
4536
4537 if (rt)
4538 extsz = alen / mp->m_sb.sb_rextsize;
4539
4540 /*
4541 * Make a transaction-less quota reservation for
4542 * delayed allocation blocks. This number gets
4543 * adjusted later. We return if we haven't
4544 * allocated blocks already inside this loop.
4545 */
4546 error = xfs_trans_reserve_quota_nblks(
4547 NULL, ip, (long)alen, 0,
4548 rt ? XFS_QMOPT_RES_RTBLKS :
4549 XFS_QMOPT_RES_REGBLKS);
4550 if (error) {
4551 if (n == 0) {
4552 *nmap = 0;
4553 ASSERT(cur == NULL);
4554 return error;
4555 }
4556 break;
4557 }
4558
4559 /*
4560 * Split changing sb for alen and indlen since
4561 * they could be coming from different places.
4562 */
4563 indlen = (xfs_extlen_t)
4564 xfs_bmap_worst_indlen(ip, alen);
4565 ASSERT(indlen > 0);
4566
4567 if (rt) {
4568 error = xfs_mod_incore_sb(mp,
4569 XFS_SBS_FREXTENTS,
4570 -((int64_t)extsz), (flags &
4571 XFS_BMAPI_RSVBLOCKS));
4572 } else {
4573 error = xfs_icsb_modify_counters(mp,
4574 XFS_SBS_FDBLOCKS,
4575 -((int64_t)alen), (flags &
4576 XFS_BMAPI_RSVBLOCKS));
4577 }
4578 if (!error) {
4579 error = xfs_icsb_modify_counters(mp,
4580 XFS_SBS_FDBLOCKS,
4581 -((int64_t)indlen), (flags &
4582 XFS_BMAPI_RSVBLOCKS));
4583 if (error && rt)
4584 xfs_mod_incore_sb(mp,
4585 XFS_SBS_FREXTENTS,
4586 (int64_t)extsz, (flags &
4587 XFS_BMAPI_RSVBLOCKS));
4588 else if (error)
4589 xfs_icsb_modify_counters(mp,
4590 XFS_SBS_FDBLOCKS,
4591 (int64_t)alen, (flags &
4592 XFS_BMAPI_RSVBLOCKS));
4593 }
4594
4595 if (error) {
4596 if (XFS_IS_QUOTA_ON(mp))
4597 /* unreserve the blocks now */
4598 (void)
4599 xfs_trans_unreserve_quota_nblks(
4600 NULL, ip,
4601 (long)alen, 0, rt ?
4602 XFS_QMOPT_RES_RTBLKS :
4603 XFS_QMOPT_RES_REGBLKS);
4604 break;
4605 }
4606
4607 ip->i_delayed_blks += alen;
4608 abno = nullstartblock(indlen);
4609 } else {
4610 /*
4611 * If first time, allocate and fill in
4612 * once-only bma fields.
4613 */
4614 if (bma.ip == NULL) {
4615 bma.tp = tp;
4616 bma.ip = ip;
4617 bma.prevp = &prev;
4618 bma.gotp = &got;
4619 bma.total = total;
4620 bma.userdata = 0;
4621 }
4622 /* Indicate if this is the first user data
4623 * in the file, or just any user data.
4624 */
4625 if (!(flags & XFS_BMAPI_METADATA)) {
4626 bma.userdata = (aoff == 0) ?
4627 XFS_ALLOC_INITIAL_USER_DATA :
4628 XFS_ALLOC_USERDATA;
4629 }
4630 /*
4631 * Fill in changeable bma fields.
4632 */
4633 bma.eof = eof;
4634 bma.firstblock = *firstblock;
4635 bma.alen = alen;
4636 bma.off = aoff;
4637 bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4638 bma.wasdel = wasdelay;
4639 bma.minlen = minlen;
4640 bma.low = flist->xbf_low;
4641 bma.minleft = minleft;
4642 /*
4643 * Only want to do the alignment at the
4644 * eof if it is userdata and allocation length
4645 * is larger than a stripe unit.
4646 */
4647 if (mp->m_dalign && alen >= mp->m_dalign &&
4648 (!(flags & XFS_BMAPI_METADATA)) &&
4649 (whichfork == XFS_DATA_FORK)) {
4650 if ((error = xfs_bmap_isaeof(ip, aoff,
4651 whichfork, &bma.aeof)))
4652 goto error0;
4653 } else
4654 bma.aeof = 0;
4655 /*
4656 * Call allocator.
4657 */
4658 if ((error = xfs_bmap_alloc(&bma)))
4659 goto error0;
4660 /*
4661 * Copy out result fields.
4662 */
4663 abno = bma.rval;
4664 if ((flist->xbf_low = bma.low))
4665 minleft = 0;
4666 alen = bma.alen;
4667 aoff = bma.off;
4668 ASSERT(*firstblock == NULLFSBLOCK ||
4669 XFS_FSB_TO_AGNO(mp, *firstblock) ==
4670 XFS_FSB_TO_AGNO(mp, bma.firstblock) ||
4671 (flist->xbf_low &&
4672 XFS_FSB_TO_AGNO(mp, *firstblock) <
4673 XFS_FSB_TO_AGNO(mp, bma.firstblock)));
4674 *firstblock = bma.firstblock;
4675 if (cur)
4676 cur->bc_private.b.firstblock =
4677 *firstblock;
4678 if (abno == NULLFSBLOCK)
4679 break;
4680 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
4681 cur = xfs_bmbt_init_cursor(mp, tp,
4682 ip, whichfork);
4683 cur->bc_private.b.firstblock =
4684 *firstblock;
4685 cur->bc_private.b.flist = flist;
4686 }
4687 /*
4688 * Bump the number of extents we've allocated
4689 * in this call.
4690 */
4691 nallocs++;
4692 }
4693 if (cur)
4694 cur->bc_private.b.flags =
4695 wasdelay ? XFS_BTCUR_BPRV_WASDEL : 0;
4696 got.br_startoff = aoff;
4697 got.br_startblock = abno;
4698 got.br_blockcount = alen;
4699 got.br_state = XFS_EXT_NORM; /* assume normal */
4700 /*
4701 * Determine state of extent, and the filesystem.
4702 * A wasdelay extent has been initialized, so
4703 * shouldn't be flagged as unwritten.
4704 */
4705 if (wr && xfs_sb_version_hasextflgbit(&mp->m_sb)) {
4706 if (!wasdelay && (flags & XFS_BMAPI_PREALLOC))
4707 got.br_state = XFS_EXT_UNWRITTEN;
4708 }
4709 error = xfs_bmap_add_extent(ip, lastx, &cur, &got,
4710 firstblock, flist, &tmp_logflags,
4711 whichfork, (flags & XFS_BMAPI_RSVBLOCKS));
4712 logflags |= tmp_logflags;
4713 if (error)
4714 goto error0;
4715 lastx = ifp->if_lastex;
4716 ep = xfs_iext_get_ext(ifp, lastx);
4717 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4718 xfs_bmbt_get_all(ep, &got);
4719 ASSERT(got.br_startoff <= aoff);
4720 ASSERT(got.br_startoff + got.br_blockcount >=
4721 aoff + alen);
4722 #ifdef DEBUG
4723 if (flags & XFS_BMAPI_DELAY) {
4724 ASSERT(isnullstartblock(got.br_startblock));
4725 ASSERT(startblockval(got.br_startblock) > 0);
4726 }
4727 ASSERT(got.br_state == XFS_EXT_NORM ||
4728 got.br_state == XFS_EXT_UNWRITTEN);
4729 #endif
4730 /*
4731 * Fall down into the found allocated space case.
4732 */
4733 } else if (inhole) {
4734 /*
4735 * Reading in a hole.
4736 */
4737 mval->br_startoff = bno;
4738 mval->br_startblock = HOLESTARTBLOCK;
4739 mval->br_blockcount =
4740 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4741 mval->br_state = XFS_EXT_NORM;
4742 bno += mval->br_blockcount;
4743 len -= mval->br_blockcount;
4744 mval++;
4745 n++;
4746 continue;
4747 }
4748 /*
4749 * Then deal with the allocated space we found.
4750 */
4751 ASSERT(ep != NULL);
4752 if (!(flags & XFS_BMAPI_ENTIRE) &&
4753 (got.br_startoff + got.br_blockcount > obno)) {
4754 if (obno > bno)
4755 bno = obno;
4756 ASSERT((bno >= obno) || (n == 0));
4757 ASSERT(bno < end);
4758 mval->br_startoff = bno;
4759 if (isnullstartblock(got.br_startblock)) {
4760 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
4761 mval->br_startblock = DELAYSTARTBLOCK;
4762 } else
4763 mval->br_startblock =
4764 got.br_startblock +
4765 (bno - got.br_startoff);
4766 /*
4767 * Return the minimum of what we got and what we
4768 * asked for for the length. We can use the len
4769 * variable here because it is modified below
4770 * and we could have been there before coming
4771 * here if the first part of the allocation
4772 * didn't overlap what was asked for.
4773 */
4774 mval->br_blockcount =
4775 XFS_FILBLKS_MIN(end - bno, got.br_blockcount -
4776 (bno - got.br_startoff));
4777 mval->br_state = got.br_state;
4778 ASSERT(mval->br_blockcount <= len);
4779 } else {
4780 *mval = got;
4781 if (isnullstartblock(mval->br_startblock)) {
4782 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
4783 mval->br_startblock = DELAYSTARTBLOCK;
4784 }
4785 }
4786
4787 /*
4788 * Check if writing previously allocated but
4789 * unwritten extents.
4790 */
4791 if (wr &&
4792 ((mval->br_state == XFS_EXT_UNWRITTEN &&
4793 ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_DELAY)) == 0)) ||
4794 (mval->br_state == XFS_EXT_NORM &&
4795 ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_CONVERT)) ==
4796 (XFS_BMAPI_PREALLOC|XFS_BMAPI_CONVERT))))) {
4797 /*
4798 * Modify (by adding) the state flag, if writing.
4799 */
4800 ASSERT(mval->br_blockcount <= len);
4801 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
4802 cur = xfs_bmbt_init_cursor(mp,
4803 tp, ip, whichfork);
4804 cur->bc_private.b.firstblock =
4805 *firstblock;
4806 cur->bc_private.b.flist = flist;
4807 }
4808 mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4809 ? XFS_EXT_NORM
4810 : XFS_EXT_UNWRITTEN;
4811 error = xfs_bmap_add_extent(ip, lastx, &cur, mval,
4812 firstblock, flist, &tmp_logflags,
4813 whichfork, (flags & XFS_BMAPI_RSVBLOCKS));
4814 logflags |= tmp_logflags;
4815 if (error)
4816 goto error0;
4817 lastx = ifp->if_lastex;
4818 ep = xfs_iext_get_ext(ifp, lastx);
4819 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4820 xfs_bmbt_get_all(ep, &got);
4821 /*
4822 * We may have combined previously unwritten
4823 * space with written space, so generate
4824 * another request.
4825 */
4826 if (mval->br_blockcount < len)
4827 continue;
4828 }
4829
4830 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4831 ((mval->br_startoff + mval->br_blockcount) <= end));
4832 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4833 (mval->br_blockcount <= len) ||
4834 (mval->br_startoff < obno));
4835 bno = mval->br_startoff + mval->br_blockcount;
4836 len = end - bno;
4837 if (n > 0 && mval->br_startoff == mval[-1].br_startoff) {
4838 ASSERT(mval->br_startblock == mval[-1].br_startblock);
4839 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
4840 ASSERT(mval->br_state == mval[-1].br_state);
4841 mval[-1].br_blockcount = mval->br_blockcount;
4842 mval[-1].br_state = mval->br_state;
4843 } else if (n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
4844 mval[-1].br_startblock != DELAYSTARTBLOCK &&
4845 mval[-1].br_startblock != HOLESTARTBLOCK &&
4846 mval->br_startblock ==
4847 mval[-1].br_startblock + mval[-1].br_blockcount &&
4848 ((flags & XFS_BMAPI_IGSTATE) ||
4849 mval[-1].br_state == mval->br_state)) {
4850 ASSERT(mval->br_startoff ==
4851 mval[-1].br_startoff + mval[-1].br_blockcount);
4852 mval[-1].br_blockcount += mval->br_blockcount;
4853 } else if (n > 0 &&
4854 mval->br_startblock == DELAYSTARTBLOCK &&
4855 mval[-1].br_startblock == DELAYSTARTBLOCK &&
4856 mval->br_startoff ==
4857 mval[-1].br_startoff + mval[-1].br_blockcount) {
4858 mval[-1].br_blockcount += mval->br_blockcount;
4859 mval[-1].br_state = mval->br_state;
4860 } else if (!((n == 0) &&
4861 ((mval->br_startoff + mval->br_blockcount) <=
4862 obno))) {
4863 mval++;
4864 n++;
4865 }
4866 /*
4867 * If we're done, stop now. Stop when we've allocated
4868 * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise
4869 * the transaction may get too big.
4870 */
4871 if (bno >= end || n >= *nmap || nallocs >= *nmap)
4872 break;
4873 /*
4874 * Else go on to the next record.
4875 */
4876 ep = xfs_iext_get_ext(ifp, ++lastx);
4877 prev = got;
4878 if (lastx >= nextents)
4879 eof = 1;
4880 else
4881 xfs_bmbt_get_all(ep, &got);
4882 }
4883 ifp->if_lastex = lastx;
4884 *nmap = n;
4885 /*
4886 * Transform from btree to extents, give it cur.
4887 */
4888 if (tp && XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
4889 XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
4890 ASSERT(wr && cur);
4891 error = xfs_bmap_btree_to_extents(tp, ip, cur,
4892 &tmp_logflags, whichfork);
4893 logflags |= tmp_logflags;
4894 if (error)
4895 goto error0;
4896 }
4897 ASSERT(ifp->if_ext_max ==
4898 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4899 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4900 XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max);
4901 error = 0;
4902 error0:
4903 /*
4904 * Log everything. Do this after conversion, there's no point in
4905 * logging the extent records if we've converted to btree format.
4906 */
4907 if ((logflags & xfs_ilog_fext(whichfork)) &&
4908 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4909 logflags &= ~xfs_ilog_fext(whichfork);
4910 else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
4911 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4912 logflags &= ~xfs_ilog_fbroot(whichfork);
4913 /*
4914 * Log whatever the flags say, even if error. Otherwise we might miss
4915 * detecting a case where the data is changed, there's an error,
4916 * and it's not logged so we don't shutdown when we should.
4917 */
4918 if (logflags) {
4919 ASSERT(tp && wr);
4920 xfs_trans_log_inode(tp, ip, logflags);
4921 }
4922 if (cur) {
4923 if (!error) {
4924 ASSERT(*firstblock == NULLFSBLOCK ||
4925 XFS_FSB_TO_AGNO(mp, *firstblock) ==
4926 XFS_FSB_TO_AGNO(mp,
4927 cur->bc_private.b.firstblock) ||
4928 (flist->xbf_low &&
4929 XFS_FSB_TO_AGNO(mp, *firstblock) <
4930 XFS_FSB_TO_AGNO(mp,
4931 cur->bc_private.b.firstblock)));
4932 *firstblock = cur->bc_private.b.firstblock;
4933 }
4934 xfs_btree_del_cursor(cur,
4935 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4936 }
4937 if (!error)
4938 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4939 orig_nmap, *nmap);
4940 return error;
4941 }
4942
4943 /*
4944 * Map file blocks to filesystem blocks, simple version.
4945 * One block (extent) only, read-only.
4946 * For flags, only the XFS_BMAPI_ATTRFORK flag is examined.
4947 * For the other flag values, the effect is as if XFS_BMAPI_METADATA
4948 * was set and all the others were clear.
4949 */
4950 int /* error */
4951 xfs_bmapi_single(
4952 xfs_trans_t *tp, /* transaction pointer */
4953 xfs_inode_t *ip, /* incore inode */
4954 int whichfork, /* data or attr fork */
4955 xfs_fsblock_t *fsb, /* output: mapped block */
4956 xfs_fileoff_t bno) /* starting file offs. mapped */
4957 {
4958 int eof; /* we've hit the end of extents */
4959 int error; /* error return */
4960 xfs_bmbt_irec_t got; /* current file extent record */
4961 xfs_ifork_t *ifp; /* inode fork pointer */
4962 xfs_extnum_t lastx; /* last useful extent number */
4963 xfs_bmbt_irec_t prev; /* previous file extent record */
4964
4965 ifp = XFS_IFORK_PTR(ip, whichfork);
4966 if (unlikely(
4967 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4968 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)) {
4969 XFS_ERROR_REPORT("xfs_bmapi_single", XFS_ERRLEVEL_LOW,
4970 ip->i_mount);
4971 return XFS_ERROR(EFSCORRUPTED);
4972 }
4973 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
4974 return XFS_ERROR(EIO);
4975 XFS_STATS_INC(xs_blk_mapr);
4976 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4977 (error = xfs_iread_extents(tp, ip, whichfork)))
4978 return error;
4979 (void)xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4980 &prev);
4981 /*
4982 * Reading past eof, act as though there's a hole
4983 * up to end.
4984 */
4985 if (eof || got.br_startoff > bno) {
4986 *fsb = NULLFSBLOCK;
4987 return 0;
4988 }
4989 ASSERT(!isnullstartblock(got.br_startblock));
4990 ASSERT(bno < got.br_startoff + got.br_blockcount);
4991 *fsb = got.br_startblock + (bno - got.br_startoff);
4992 ifp->if_lastex = lastx;
4993 return 0;
4994 }
4995
4996 /*
4997 * Unmap (remove) blocks from a file.
4998 * If nexts is nonzero then the number of extents to remove is limited to
4999 * that value. If not all extents in the block range can be removed then
5000 * *done is set.
5001 */
5002 int /* error */
5003 xfs_bunmapi(
5004 xfs_trans_t *tp, /* transaction pointer */
5005 struct xfs_inode *ip, /* incore inode */
5006 xfs_fileoff_t bno, /* starting offset to unmap */
5007 xfs_filblks_t len, /* length to unmap in file */
5008 int flags, /* misc flags */
5009 xfs_extnum_t nexts, /* number of extents max */
5010 xfs_fsblock_t *firstblock, /* first allocated block
5011 controls a.g. for allocs */
5012 xfs_bmap_free_t *flist, /* i/o: list extents to free */
5013 int *done) /* set if not done yet */
5014 {
5015 xfs_btree_cur_t *cur; /* bmap btree cursor */
5016 xfs_bmbt_irec_t del; /* extent being deleted */
5017 int eof; /* is deleting at eof */
5018 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
5019 int error; /* error return value */
5020 xfs_extnum_t extno; /* extent number in list */
5021 xfs_bmbt_irec_t got; /* current extent record */
5022 xfs_ifork_t *ifp; /* inode fork pointer */
5023 int isrt; /* freeing in rt area */
5024 xfs_extnum_t lastx; /* last extent index used */
5025 int logflags; /* transaction logging flags */
5026 xfs_extlen_t mod; /* rt extent offset */
5027 xfs_mount_t *mp; /* mount structure */
5028 xfs_extnum_t nextents; /* number of file extents */
5029 xfs_bmbt_irec_t prev; /* previous extent record */
5030 xfs_fileoff_t start; /* first file offset deleted */
5031 int tmp_logflags; /* partial logging flags */
5032 int wasdel; /* was a delayed alloc extent */
5033 int whichfork; /* data or attribute fork */
5034 int rsvd; /* OK to allocate reserved blocks */
5035 xfs_fsblock_t sum;
5036
5037 trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5038
5039 whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5040 XFS_ATTR_FORK : XFS_DATA_FORK;
5041 ifp = XFS_IFORK_PTR(ip, whichfork);
5042 if (unlikely(
5043 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5044 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5045 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5046 ip->i_mount);
5047 return XFS_ERROR(EFSCORRUPTED);
5048 }
5049 mp = ip->i_mount;
5050 if (XFS_FORCED_SHUTDOWN(mp))
5051 return XFS_ERROR(EIO);
5052 rsvd = (flags & XFS_BMAPI_RSVBLOCKS) != 0;
5053 ASSERT(len > 0);
5054 ASSERT(nexts >= 0);
5055 ASSERT(ifp->if_ext_max ==
5056 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5057 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5058 (error = xfs_iread_extents(tp, ip, whichfork)))
5059 return error;
5060 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5061 if (nextents == 0) {
5062 *done = 1;
5063 return 0;
5064 }
5065 XFS_STATS_INC(xs_blk_unmap);
5066 isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5067 start = bno;
5068 bno = start + len - 1;
5069 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5070 &prev);
5071
5072 /*
5073 * Check to see if the given block number is past the end of the
5074 * file, back up to the last block if so...
5075 */
5076 if (eof) {
5077 ep = xfs_iext_get_ext(ifp, --lastx);
5078 xfs_bmbt_get_all(ep, &got);
5079 bno = got.br_startoff + got.br_blockcount - 1;
5080 }
5081 logflags = 0;
5082 if (ifp->if_flags & XFS_IFBROOT) {
5083 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5084 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5085 cur->bc_private.b.firstblock = *firstblock;
5086 cur->bc_private.b.flist = flist;
5087 cur->bc_private.b.flags = 0;
5088 } else
5089 cur = NULL;
5090 extno = 0;
5091 while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5092 (nexts == 0 || extno < nexts)) {
5093 /*
5094 * Is the found extent after a hole in which bno lives?
5095 * Just back up to the previous extent, if so.
5096 */
5097 if (got.br_startoff > bno) {
5098 if (--lastx < 0)
5099 break;
5100 ep = xfs_iext_get_ext(ifp, lastx);
5101 xfs_bmbt_get_all(ep, &got);
5102 }
5103 /*
5104 * Is the last block of this extent before the range
5105 * we're supposed to delete? If so, we're done.
5106 */
5107 bno = XFS_FILEOFF_MIN(bno,
5108 got.br_startoff + got.br_blockcount - 1);
5109 if (bno < start)
5110 break;
5111 /*
5112 * Then deal with the (possibly delayed) allocated space
5113 * we found.
5114 */
5115 ASSERT(ep != NULL);
5116 del = got;
5117 wasdel = isnullstartblock(del.br_startblock);
5118 if (got.br_startoff < start) {
5119 del.br_startoff = start;
5120 del.br_blockcount -= start - got.br_startoff;
5121 if (!wasdel)
5122 del.br_startblock += start - got.br_startoff;
5123 }
5124 if (del.br_startoff + del.br_blockcount > bno + 1)
5125 del.br_blockcount = bno + 1 - del.br_startoff;
5126 sum = del.br_startblock + del.br_blockcount;
5127 if (isrt &&
5128 (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5129 /*
5130 * Realtime extent not lined up at the end.
5131 * The extent could have been split into written
5132 * and unwritten pieces, or we could just be
5133 * unmapping part of it. But we can't really
5134 * get rid of part of a realtime extent.
5135 */
5136 if (del.br_state == XFS_EXT_UNWRITTEN ||
5137 !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5138 /*
5139 * This piece is unwritten, or we're not
5140 * using unwritten extents. Skip over it.
5141 */
5142 ASSERT(bno >= mod);
5143 bno -= mod > del.br_blockcount ?
5144 del.br_blockcount : mod;
5145 if (bno < got.br_startoff) {
5146 if (--lastx >= 0)
5147 xfs_bmbt_get_all(xfs_iext_get_ext(
5148 ifp, lastx), &got);
5149 }
5150 continue;
5151 }
5152 /*
5153 * It's written, turn it unwritten.
5154 * This is better than zeroing it.
5155 */
5156 ASSERT(del.br_state == XFS_EXT_NORM);
5157 ASSERT(xfs_trans_get_block_res(tp) > 0);
5158 /*
5159 * If this spans a realtime extent boundary,
5160 * chop it back to the start of the one we end at.
5161 */
5162 if (del.br_blockcount > mod) {
5163 del.br_startoff += del.br_blockcount - mod;
5164 del.br_startblock += del.br_blockcount - mod;
5165 del.br_blockcount = mod;
5166 }
5167 del.br_state = XFS_EXT_UNWRITTEN;
5168 error = xfs_bmap_add_extent(ip, lastx, &cur, &del,
5169 firstblock, flist, &logflags,
5170 XFS_DATA_FORK, 0);
5171 if (error)
5172 goto error0;
5173 goto nodelete;
5174 }
5175 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5176 /*
5177 * Realtime extent is lined up at the end but not
5178 * at the front. We'll get rid of full extents if
5179 * we can.
5180 */
5181 mod = mp->m_sb.sb_rextsize - mod;
5182 if (del.br_blockcount > mod) {
5183 del.br_blockcount -= mod;
5184 del.br_startoff += mod;
5185 del.br_startblock += mod;
5186 } else if ((del.br_startoff == start &&
5187 (del.br_state == XFS_EXT_UNWRITTEN ||
5188 xfs_trans_get_block_res(tp) == 0)) ||
5189 !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5190 /*
5191 * Can't make it unwritten. There isn't
5192 * a full extent here so just skip it.
5193 */
5194 ASSERT(bno >= del.br_blockcount);
5195 bno -= del.br_blockcount;
5196 if (bno < got.br_startoff) {
5197 if (--lastx >= 0)
5198 xfs_bmbt_get_all(--ep, &got);
5199 }
5200 continue;
5201 } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5202 /*
5203 * This one is already unwritten.
5204 * It must have a written left neighbor.
5205 * Unwrite the killed part of that one and
5206 * try again.
5207 */
5208 ASSERT(lastx > 0);
5209 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5210 lastx - 1), &prev);
5211 ASSERT(prev.br_state == XFS_EXT_NORM);
5212 ASSERT(!isnullstartblock(prev.br_startblock));
5213 ASSERT(del.br_startblock ==
5214 prev.br_startblock + prev.br_blockcount);
5215 if (prev.br_startoff < start) {
5216 mod = start - prev.br_startoff;
5217 prev.br_blockcount -= mod;
5218 prev.br_startblock += mod;
5219 prev.br_startoff = start;
5220 }
5221 prev.br_state = XFS_EXT_UNWRITTEN;
5222 error = xfs_bmap_add_extent(ip, lastx - 1, &cur,
5223 &prev, firstblock, flist, &logflags,
5224 XFS_DATA_FORK, 0);
5225 if (error)
5226 goto error0;
5227 goto nodelete;
5228 } else {
5229 ASSERT(del.br_state == XFS_EXT_NORM);
5230 del.br_state = XFS_EXT_UNWRITTEN;
5231 error = xfs_bmap_add_extent(ip, lastx, &cur,
5232 &del, firstblock, flist, &logflags,
5233 XFS_DATA_FORK, 0);
5234 if (error)
5235 goto error0;
5236 goto nodelete;
5237 }
5238 }
5239 if (wasdel) {
5240 ASSERT(startblockval(del.br_startblock) > 0);
5241 /* Update realtime/data freespace, unreserve quota */
5242 if (isrt) {
5243 xfs_filblks_t rtexts;
5244
5245 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5246 do_div(rtexts, mp->m_sb.sb_rextsize);
5247 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5248 (int64_t)rtexts, rsvd);
5249 (void)xfs_trans_reserve_quota_nblks(NULL,
5250 ip, -((long)del.br_blockcount), 0,
5251 XFS_QMOPT_RES_RTBLKS);
5252 } else {
5253 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5254 (int64_t)del.br_blockcount, rsvd);
5255 (void)xfs_trans_reserve_quota_nblks(NULL,
5256 ip, -((long)del.br_blockcount), 0,
5257 XFS_QMOPT_RES_REGBLKS);
5258 }
5259 ip->i_delayed_blks -= del.br_blockcount;
5260 if (cur)
5261 cur->bc_private.b.flags |=
5262 XFS_BTCUR_BPRV_WASDEL;
5263 } else if (cur)
5264 cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5265 /*
5266 * If it's the case where the directory code is running
5267 * with no block reservation, and the deleted block is in
5268 * the middle of its extent, and the resulting insert
5269 * of an extent would cause transformation to btree format,
5270 * then reject it. The calling code will then swap
5271 * blocks around instead.
5272 * We have to do this now, rather than waiting for the
5273 * conversion to btree format, since the transaction
5274 * will be dirty.
5275 */
5276 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5277 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5278 XFS_IFORK_NEXTENTS(ip, whichfork) >= ifp->if_ext_max &&
5279 del.br_startoff > got.br_startoff &&
5280 del.br_startoff + del.br_blockcount <
5281 got.br_startoff + got.br_blockcount) {
5282 error = XFS_ERROR(ENOSPC);
5283 goto error0;
5284 }
5285 error = xfs_bmap_del_extent(ip, tp, lastx, flist, cur, &del,
5286 &tmp_logflags, whichfork, rsvd);
5287 logflags |= tmp_logflags;
5288 if (error)
5289 goto error0;
5290 bno = del.br_startoff - 1;
5291 nodelete:
5292 lastx = ifp->if_lastex;
5293 /*
5294 * If not done go on to the next (previous) record.
5295 * Reset ep in case the extents array was re-alloced.
5296 */
5297 ep = xfs_iext_get_ext(ifp, lastx);
5298 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5299 if (lastx >= XFS_IFORK_NEXTENTS(ip, whichfork) ||
5300 xfs_bmbt_get_startoff(ep) > bno) {
5301 if (--lastx >= 0)
5302 ep = xfs_iext_get_ext(ifp, lastx);
5303 }
5304 if (lastx >= 0)
5305 xfs_bmbt_get_all(ep, &got);
5306 extno++;
5307 }
5308 }
5309 ifp->if_lastex = lastx;
5310 *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5311 ASSERT(ifp->if_ext_max ==
5312 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5313 /*
5314 * Convert to a btree if necessary.
5315 */
5316 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5317 XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
5318 ASSERT(cur == NULL);
5319 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5320 &cur, 0, &tmp_logflags, whichfork);
5321 logflags |= tmp_logflags;
5322 if (error)
5323 goto error0;
5324 }
5325 /*
5326 * transform from btree to extents, give it cur
5327 */
5328 else if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5329 XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5330 ASSERT(cur != NULL);
5331 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5332 whichfork);
5333 logflags |= tmp_logflags;
5334 if (error)
5335 goto error0;
5336 }
5337 /*
5338 * transform from extents to local?
5339 */
5340 ASSERT(ifp->if_ext_max ==
5341 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5342 error = 0;
5343 error0:
5344 /*
5345 * Log everything. Do this after conversion, there's no point in
5346 * logging the extent records if we've converted to btree format.
5347 */
5348 if ((logflags & xfs_ilog_fext(whichfork)) &&
5349 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5350 logflags &= ~xfs_ilog_fext(whichfork);
5351 else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5352 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5353 logflags &= ~xfs_ilog_fbroot(whichfork);
5354 /*
5355 * Log inode even in the error case, if the transaction
5356 * is dirty we'll need to shut down the filesystem.
5357 */
5358 if (logflags)
5359 xfs_trans_log_inode(tp, ip, logflags);
5360 if (cur) {
5361 if (!error) {
5362 *firstblock = cur->bc_private.b.firstblock;
5363 cur->bc_private.b.allocated = 0;
5364 }
5365 xfs_btree_del_cursor(cur,
5366 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5367 }
5368 return error;
5369 }
5370
5371 /*
5372 * returns 1 for success, 0 if we failed to map the extent.
5373 */
5374 STATIC int
5375 xfs_getbmapx_fix_eof_hole(
5376 xfs_inode_t *ip, /* xfs incore inode pointer */
5377 struct getbmapx *out, /* output structure */
5378 int prealloced, /* this is a file with
5379 * preallocated data space */
5380 __int64_t end, /* last block requested */
5381 xfs_fsblock_t startblock)
5382 {
5383 __int64_t fixlen;
5384 xfs_mount_t *mp; /* file system mount point */
5385 xfs_ifork_t *ifp; /* inode fork pointer */
5386 xfs_extnum_t lastx; /* last extent pointer */
5387 xfs_fileoff_t fileblock;
5388
5389 if (startblock == HOLESTARTBLOCK) {
5390 mp = ip->i_mount;
5391 out->bmv_block = -1;
5392 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, ip->i_size));
5393 fixlen -= out->bmv_offset;
5394 if (prealloced && out->bmv_offset + out->bmv_length == end) {
5395 /* Came to hole at EOF. Trim it. */
5396 if (fixlen <= 0)
5397 return 0;
5398 out->bmv_length = fixlen;
5399 }
5400 } else {
5401 if (startblock == DELAYSTARTBLOCK)
5402 out->bmv_block = -2;
5403 else
5404 out->bmv_block = xfs_fsb_to_db(ip, startblock);
5405 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
5406 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
5407 if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
5408 (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
5409 out->bmv_oflags |= BMV_OF_LAST;
5410 }
5411
5412 return 1;
5413 }
5414
5415 /*
5416 * Get inode's extents as described in bmv, and format for output.
5417 * Calls formatter to fill the user's buffer until all extents
5418 * are mapped, until the passed-in bmv->bmv_count slots have
5419 * been filled, or until the formatter short-circuits the loop,
5420 * if it is tracking filled-in extents on its own.
5421 */
5422 int /* error code */
5423 xfs_getbmap(
5424 xfs_inode_t *ip,
5425 struct getbmapx *bmv, /* user bmap structure */
5426 xfs_bmap_format_t formatter, /* format to user */
5427 void *arg) /* formatter arg */
5428 {
5429 __int64_t bmvend; /* last block requested */
5430 int error = 0; /* return value */
5431 __int64_t fixlen; /* length for -1 case */
5432 int i; /* extent number */
5433 int lock; /* lock state */
5434 xfs_bmbt_irec_t *map; /* buffer for user's data */
5435 xfs_mount_t *mp; /* file system mount point */
5436 int nex; /* # of user extents can do */
5437 int nexleft; /* # of user extents left */
5438 int subnex; /* # of bmapi's can do */
5439 int nmap; /* number of map entries */
5440 struct getbmapx *out; /* output structure */
5441 int whichfork; /* data or attr fork */
5442 int prealloced; /* this is a file with
5443 * preallocated data space */
5444 int iflags; /* interface flags */
5445 int bmapi_flags; /* flags for xfs_bmapi */
5446 int cur_ext = 0;
5447
5448 mp = ip->i_mount;
5449 iflags = bmv->bmv_iflags;
5450 whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5451
5452 if (whichfork == XFS_ATTR_FORK) {
5453 if (XFS_IFORK_Q(ip)) {
5454 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5455 ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5456 ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5457 return XFS_ERROR(EINVAL);
5458 } else if (unlikely(
5459 ip->i_d.di_aformat != 0 &&
5460 ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5461 XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5462 ip->i_mount);
5463 return XFS_ERROR(EFSCORRUPTED);
5464 }
5465
5466 prealloced = 0;
5467 fixlen = 1LL << 32;
5468 } else {
5469 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5470 ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5471 ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5472 return XFS_ERROR(EINVAL);
5473
5474 if (xfs_get_extsz_hint(ip) ||
5475 ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5476 prealloced = 1;
5477 fixlen = XFS_MAXIOFFSET(mp);
5478 } else {
5479 prealloced = 0;
5480 fixlen = ip->i_size;
5481 }
5482 }
5483
5484 if (bmv->bmv_length == -1) {
5485 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5486 bmv->bmv_length =
5487 max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
5488 } else if (bmv->bmv_length == 0) {
5489 bmv->bmv_entries = 0;
5490 return 0;
5491 } else if (bmv->bmv_length < 0) {
5492 return XFS_ERROR(EINVAL);
5493 }
5494
5495 nex = bmv->bmv_count - 1;
5496 if (nex <= 0)
5497 return XFS_ERROR(EINVAL);
5498 bmvend = bmv->bmv_offset + bmv->bmv_length;
5499
5500
5501 if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
5502 return XFS_ERROR(ENOMEM);
5503 out = kmem_zalloc(bmv->bmv_count * sizeof(struct getbmapx), KM_MAYFAIL);
5504 if (!out)
5505 return XFS_ERROR(ENOMEM);
5506
5507 xfs_ilock(ip, XFS_IOLOCK_SHARED);
5508 if (whichfork == XFS_DATA_FORK && !(iflags & BMV_IF_DELALLOC)) {
5509 if (ip->i_delayed_blks || ip->i_size > ip->i_d.di_size) {
5510 error = xfs_flush_pages(ip, 0, -1, 0, FI_REMAPF);
5511 if (error)
5512 goto out_unlock_iolock;
5513 }
5514 /*
5515 * even after flushing the inode, there can still be delalloc
5516 * blocks on the inode beyond EOF due to speculative
5517 * preallocation. These are not removed until the release
5518 * function is called or the inode is inactivated. Hence we
5519 * cannot assert here that ip->i_delayed_blks == 0.
5520 */
5521 }
5522
5523 lock = xfs_ilock_map_shared(ip);
5524
5525 /*
5526 * Don't let nex be bigger than the number of extents
5527 * we can have assuming alternating holes and real extents.
5528 */
5529 if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5530 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5531
5532 bmapi_flags = xfs_bmapi_aflag(whichfork);
5533 if (!(iflags & BMV_IF_PREALLOC))
5534 bmapi_flags |= XFS_BMAPI_IGSTATE;
5535
5536 /*
5537 * Allocate enough space to handle "subnex" maps at a time.
5538 */
5539 error = ENOMEM;
5540 subnex = 16;
5541 map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
5542 if (!map)
5543 goto out_unlock_ilock;
5544
5545 bmv->bmv_entries = 0;
5546
5547 if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
5548 (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
5549 error = 0;
5550 goto out_free_map;
5551 }
5552
5553 nexleft = nex;
5554
5555 do {
5556 nmap = (nexleft > subnex) ? subnex : nexleft;
5557 error = xfs_bmapi(NULL, ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5558 XFS_BB_TO_FSB(mp, bmv->bmv_length),
5559 bmapi_flags, NULL, 0, map, &nmap,
5560 NULL);
5561 if (error)
5562 goto out_free_map;
5563 ASSERT(nmap <= subnex);
5564
5565 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5566 out[cur_ext].bmv_oflags = 0;
5567 if (map[i].br_state == XFS_EXT_UNWRITTEN)
5568 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
5569 else if (map[i].br_startblock == DELAYSTARTBLOCK)
5570 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
5571 out[cur_ext].bmv_offset =
5572 XFS_FSB_TO_BB(mp, map[i].br_startoff);
5573 out[cur_ext].bmv_length =
5574 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5575 out[cur_ext].bmv_unused1 = 0;
5576 out[cur_ext].bmv_unused2 = 0;
5577 ASSERT(((iflags & BMV_IF_DELALLOC) != 0) ||
5578 (map[i].br_startblock != DELAYSTARTBLOCK));
5579 if (map[i].br_startblock == HOLESTARTBLOCK &&
5580 whichfork == XFS_ATTR_FORK) {
5581 /* came to the end of attribute fork */
5582 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
5583 goto out_free_map;
5584 }
5585
5586 if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
5587 prealloced, bmvend,
5588 map[i].br_startblock))
5589 goto out_free_map;
5590
5591 bmv->bmv_offset =
5592 out[cur_ext].bmv_offset +
5593 out[cur_ext].bmv_length;
5594 bmv->bmv_length =
5595 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
5596
5597 /*
5598 * In case we don't want to return the hole,
5599 * don't increase cur_ext so that we can reuse
5600 * it in the next loop.
5601 */
5602 if ((iflags & BMV_IF_NO_HOLES) &&
5603 map[i].br_startblock == HOLESTARTBLOCK) {
5604 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
5605 continue;
5606 }
5607
5608 nexleft--;
5609 bmv->bmv_entries++;
5610 cur_ext++;
5611 }
5612 } while (nmap && nexleft && bmv->bmv_length);
5613
5614 out_free_map:
5615 kmem_free(map);
5616 out_unlock_ilock:
5617 xfs_iunlock_map_shared(ip, lock);
5618 out_unlock_iolock:
5619 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5620
5621 for (i = 0; i < cur_ext; i++) {
5622 int full = 0; /* user array is full */
5623
5624 /* format results & advance arg */
5625 error = formatter(&arg, &out[i], &full);
5626 if (error || full)
5627 break;
5628 }
5629
5630 kmem_free(out);
5631 return error;
5632 }
5633
5634 /*
5635 * Check the last inode extent to determine whether this allocation will result
5636 * in blocks being allocated at the end of the file. When we allocate new data
5637 * blocks at the end of the file which do not start at the previous data block,
5638 * we will try to align the new blocks at stripe unit boundaries.
5639 */
5640 STATIC int /* error */
5641 xfs_bmap_isaeof(
5642 xfs_inode_t *ip, /* incore inode pointer */
5643 xfs_fileoff_t off, /* file offset in fsblocks */
5644 int whichfork, /* data or attribute fork */
5645 char *aeof) /* return value */
5646 {
5647 int error; /* error return value */
5648 xfs_ifork_t *ifp; /* inode fork pointer */
5649 xfs_bmbt_rec_host_t *lastrec; /* extent record pointer */
5650 xfs_extnum_t nextents; /* number of file extents */
5651 xfs_bmbt_irec_t s; /* expanded extent record */
5652
5653 ASSERT(whichfork == XFS_DATA_FORK);
5654 ifp = XFS_IFORK_PTR(ip, whichfork);
5655 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5656 (error = xfs_iread_extents(NULL, ip, whichfork)))
5657 return error;
5658 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5659 if (nextents == 0) {
5660 *aeof = 1;
5661 return 0;
5662 }
5663 /*
5664 * Go to the last extent
5665 */
5666 lastrec = xfs_iext_get_ext(ifp, nextents - 1);
5667 xfs_bmbt_get_all(lastrec, &s);
5668 /*
5669 * Check we are allocating in the last extent (for delayed allocations)
5670 * or past the last extent for non-delayed allocations.
5671 */
5672 *aeof = (off >= s.br_startoff &&
5673 off < s.br_startoff + s.br_blockcount &&
5674 isnullstartblock(s.br_startblock)) ||
5675 off >= s.br_startoff + s.br_blockcount;
5676 return 0;
5677 }
5678
5679 /*
5680 * Check if the endoff is outside the last extent. If so the caller will grow
5681 * the allocation to a stripe unit boundary.
5682 */
5683 int /* error */
5684 xfs_bmap_eof(
5685 xfs_inode_t *ip, /* incore inode pointer */
5686 xfs_fileoff_t endoff, /* file offset in fsblocks */
5687 int whichfork, /* data or attribute fork */
5688 int *eof) /* result value */
5689 {
5690 xfs_fsblock_t blockcount; /* extent block count */
5691 int error; /* error return value */
5692 xfs_ifork_t *ifp; /* inode fork pointer */
5693 xfs_bmbt_rec_host_t *lastrec; /* extent record pointer */
5694 xfs_extnum_t nextents; /* number of file extents */
5695 xfs_fileoff_t startoff; /* extent starting file offset */
5696
5697 ASSERT(whichfork == XFS_DATA_FORK);
5698 ifp = XFS_IFORK_PTR(ip, whichfork);
5699 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5700 (error = xfs_iread_extents(NULL, ip, whichfork)))
5701 return error;
5702 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5703 if (nextents == 0) {
5704 *eof = 1;
5705 return 0;
5706 }
5707 /*
5708 * Go to the last extent
5709 */
5710 lastrec = xfs_iext_get_ext(ifp, nextents - 1);
5711 startoff = xfs_bmbt_get_startoff(lastrec);
5712 blockcount = xfs_bmbt_get_blockcount(lastrec);
5713 *eof = endoff >= startoff + blockcount;
5714 return 0;
5715 }
5716
5717 #ifdef DEBUG
5718 STATIC struct xfs_buf *
5719 xfs_bmap_get_bp(
5720 struct xfs_btree_cur *cur,
5721 xfs_fsblock_t bno)
5722 {
5723 struct xfs_log_item_desc *lidp;
5724 int i;
5725
5726 if (!cur)
5727 return NULL;
5728
5729 for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
5730 if (!cur->bc_bufs[i])
5731 break;
5732 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
5733 return cur->bc_bufs[i];
5734 }
5735
5736 /* Chase down all the log items to see if the bp is there */
5737 list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
5738 struct xfs_buf_log_item *bip;
5739 bip = (struct xfs_buf_log_item *)lidp->lid_item;
5740 if (bip->bli_item.li_type == XFS_LI_BUF &&
5741 XFS_BUF_ADDR(bip->bli_buf) == bno)
5742 return bip->bli_buf;
5743 }
5744
5745 return NULL;
5746 }
5747
5748 STATIC void
5749 xfs_check_block(
5750 struct xfs_btree_block *block,
5751 xfs_mount_t *mp,
5752 int root,
5753 short sz)
5754 {
5755 int i, j, dmxr;
5756 __be64 *pp, *thispa; /* pointer to block address */
5757 xfs_bmbt_key_t *prevp, *keyp;
5758
5759 ASSERT(be16_to_cpu(block->bb_level) > 0);
5760
5761 prevp = NULL;
5762 for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
5763 dmxr = mp->m_bmap_dmxr[0];
5764 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
5765
5766 if (prevp) {
5767 ASSERT(be64_to_cpu(prevp->br_startoff) <
5768 be64_to_cpu(keyp->br_startoff));
5769 }
5770 prevp = keyp;
5771
5772 /*
5773 * Compare the block numbers to see if there are dups.
5774 */
5775 if (root)
5776 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
5777 else
5778 pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
5779
5780 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
5781 if (root)
5782 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
5783 else
5784 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
5785 if (*thispa == *pp) {
5786 cmn_err(CE_WARN, "%s: thispa(%d) == pp(%d) %Ld",
5787 __func__, j, i,
5788 (unsigned long long)be64_to_cpu(*thispa));
5789 panic("%s: ptrs are equal in node\n",
5790 __func__);
5791 }
5792 }
5793 }
5794 }
5795
5796 /*
5797 * Check that the extents for the inode ip are in the right order in all
5798 * btree leaves.
5799 */
5800
5801 STATIC void
5802 xfs_bmap_check_leaf_extents(
5803 xfs_btree_cur_t *cur, /* btree cursor or null */
5804 xfs_inode_t *ip, /* incore inode pointer */
5805 int whichfork) /* data or attr fork */
5806 {
5807 struct xfs_btree_block *block; /* current btree block */
5808 xfs_fsblock_t bno; /* block # of "block" */
5809 xfs_buf_t *bp; /* buffer for "block" */
5810 int error; /* error return value */
5811 xfs_extnum_t i=0, j; /* index into the extents list */
5812 xfs_ifork_t *ifp; /* fork structure */
5813 int level; /* btree level, for checking */
5814 xfs_mount_t *mp; /* file system mount structure */
5815 __be64 *pp; /* pointer to block address */
5816 xfs_bmbt_rec_t *ep; /* pointer to current extent */
5817 xfs_bmbt_rec_t last = {0, 0}; /* last extent in prev block */
5818 xfs_bmbt_rec_t *nextp; /* pointer to next extent */
5819 int bp_release = 0;
5820
5821 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
5822 return;
5823 }
5824
5825 bno = NULLFSBLOCK;
5826 mp = ip->i_mount;
5827 ifp = XFS_IFORK_PTR(ip, whichfork);
5828 block = ifp->if_broot;
5829 /*
5830 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5831 */
5832 level = be16_to_cpu(block->bb_level);
5833 ASSERT(level > 0);
5834 xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
5835 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5836 bno = be64_to_cpu(*pp);
5837
5838 ASSERT(bno != NULLDFSBNO);
5839 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5840 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5841
5842 /*
5843 * Go down the tree until leaf level is reached, following the first
5844 * pointer (leftmost) at each level.
5845 */
5846 while (level-- > 0) {
5847 /* See if buf is in cur first */
5848 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5849 if (bp) {
5850 bp_release = 0;
5851 } else {
5852 bp_release = 1;
5853 }
5854 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5855 XFS_BMAP_BTREE_REF)))
5856 goto error_norelse;
5857 block = XFS_BUF_TO_BLOCK(bp);
5858 XFS_WANT_CORRUPTED_GOTO(
5859 xfs_bmap_sanity_check(mp, bp, level),
5860 error0);
5861 if (level == 0)
5862 break;
5863
5864 /*
5865 * Check this block for basic sanity (increasing keys and
5866 * no duplicate blocks).
5867 */
5868
5869 xfs_check_block(block, mp, 0, 0);
5870 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
5871 bno = be64_to_cpu(*pp);
5872 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
5873 if (bp_release) {
5874 bp_release = 0;
5875 xfs_trans_brelse(NULL, bp);
5876 }
5877 }
5878
5879 /*
5880 * Here with bp and block set to the leftmost leaf node in the tree.
5881 */
5882 i = 0;
5883
5884 /*
5885 * Loop over all leaf nodes checking that all extents are in the right order.
5886 */
5887 for (;;) {
5888 xfs_fsblock_t nextbno;
5889 xfs_extnum_t num_recs;
5890
5891
5892 num_recs = xfs_btree_get_numrecs(block);
5893
5894 /*
5895 * Read-ahead the next leaf block, if any.
5896 */
5897
5898 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
5899
5900 /*
5901 * Check all the extents to make sure they are OK.
5902 * If we had a previous block, the last entry should
5903 * conform with the first entry in this one.
5904 */
5905
5906 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
5907 if (i) {
5908 ASSERT(xfs_bmbt_disk_get_startoff(&last) +
5909 xfs_bmbt_disk_get_blockcount(&last) <=
5910 xfs_bmbt_disk_get_startoff(ep));
5911 }
5912 for (j = 1; j < num_recs; j++) {
5913 nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
5914 ASSERT(xfs_bmbt_disk_get_startoff(ep) +
5915 xfs_bmbt_disk_get_blockcount(ep) <=
5916 xfs_bmbt_disk_get_startoff(nextp));
5917 ep = nextp;
5918 }
5919
5920 last = *ep;
5921 i += num_recs;
5922 if (bp_release) {
5923 bp_release = 0;
5924 xfs_trans_brelse(NULL, bp);
5925 }
5926 bno = nextbno;
5927 /*
5928 * If we've reached the end, stop.
5929 */
5930 if (bno == NULLFSBLOCK)
5931 break;
5932
5933 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5934 if (bp) {
5935 bp_release = 0;
5936 } else {
5937 bp_release = 1;
5938 }
5939 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5940 XFS_BMAP_BTREE_REF)))
5941 goto error_norelse;
5942 block = XFS_BUF_TO_BLOCK(bp);
5943 }
5944 if (bp_release) {
5945 bp_release = 0;
5946 xfs_trans_brelse(NULL, bp);
5947 }
5948 return;
5949
5950 error0:
5951 cmn_err(CE_WARN, "%s: at error0", __func__);
5952 if (bp_release)
5953 xfs_trans_brelse(NULL, bp);
5954 error_norelse:
5955 cmn_err(CE_WARN, "%s: BAD after btree leaves for %d extents",
5956 __func__, i);
5957 panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
5958 return;
5959 }
5960 #endif
5961
5962 /*
5963 * Count fsblocks of the given fork.
5964 */
5965 int /* error */
5966 xfs_bmap_count_blocks(
5967 xfs_trans_t *tp, /* transaction pointer */
5968 xfs_inode_t *ip, /* incore inode */
5969 int whichfork, /* data or attr fork */
5970 int *count) /* out: count of blocks */
5971 {
5972 struct xfs_btree_block *block; /* current btree block */
5973 xfs_fsblock_t bno; /* block # of "block" */
5974 xfs_ifork_t *ifp; /* fork structure */
5975 int level; /* btree level, for checking */
5976 xfs_mount_t *mp; /* file system mount structure */
5977 __be64 *pp; /* pointer to block address */
5978
5979 bno = NULLFSBLOCK;
5980 mp = ip->i_mount;
5981 ifp = XFS_IFORK_PTR(ip, whichfork);
5982 if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
5983 xfs_bmap_count_leaves(ifp, 0,
5984 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
5985 count);
5986 return 0;
5987 }
5988
5989 /*
5990 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5991 */
5992 block = ifp->if_broot;
5993 level = be16_to_cpu(block->bb_level);
5994 ASSERT(level > 0);
5995 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5996 bno = be64_to_cpu(*pp);
5997 ASSERT(bno != NULLDFSBNO);
5998 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5999 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
6000
6001 if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
6002 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
6003 mp);
6004 return XFS_ERROR(EFSCORRUPTED);
6005 }
6006
6007 return 0;
6008 }
6009
6010 /*
6011 * Recursively walks each level of a btree
6012 * to count total fsblocks is use.
6013 */
6014 STATIC int /* error */
6015 xfs_bmap_count_tree(
6016 xfs_mount_t *mp, /* file system mount point */
6017 xfs_trans_t *tp, /* transaction pointer */
6018 xfs_ifork_t *ifp, /* inode fork pointer */
6019 xfs_fsblock_t blockno, /* file system block number */
6020 int levelin, /* level in btree */
6021 int *count) /* Count of blocks */
6022 {
6023 int error;
6024 xfs_buf_t *bp, *nbp;
6025 int level = levelin;
6026 __be64 *pp;
6027 xfs_fsblock_t bno = blockno;
6028 xfs_fsblock_t nextbno;
6029 struct xfs_btree_block *block, *nextblock;
6030 int numrecs;
6031
6032 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
6033 return error;
6034 *count += 1;
6035 block = XFS_BUF_TO_BLOCK(bp);
6036
6037 if (--level) {
6038 /* Not at node above leaves, count this level of nodes */
6039 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
6040 while (nextbno != NULLFSBLOCK) {
6041 if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
6042 0, &nbp, XFS_BMAP_BTREE_REF)))
6043 return error;
6044 *count += 1;
6045 nextblock = XFS_BUF_TO_BLOCK(nbp);
6046 nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
6047 xfs_trans_brelse(tp, nbp);
6048 }
6049
6050 /* Dive to the next level */
6051 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
6052 bno = be64_to_cpu(*pp);
6053 if (unlikely((error =
6054 xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
6055 xfs_trans_brelse(tp, bp);
6056 XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6057 XFS_ERRLEVEL_LOW, mp);
6058 return XFS_ERROR(EFSCORRUPTED);
6059 }
6060 xfs_trans_brelse(tp, bp);
6061 } else {
6062 /* count all level 1 nodes and their leaves */
6063 for (;;) {
6064 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
6065 numrecs = be16_to_cpu(block->bb_numrecs);
6066 xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
6067 xfs_trans_brelse(tp, bp);
6068 if (nextbno == NULLFSBLOCK)
6069 break;
6070 bno = nextbno;
6071 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6072 XFS_BMAP_BTREE_REF)))
6073 return error;
6074 *count += 1;
6075 block = XFS_BUF_TO_BLOCK(bp);
6076 }
6077 }
6078 return 0;
6079 }
6080
6081 /*
6082 * Count leaf blocks given a range of extent records.
6083 */
6084 STATIC void
6085 xfs_bmap_count_leaves(
6086 xfs_ifork_t *ifp,
6087 xfs_extnum_t idx,
6088 int numrecs,
6089 int *count)
6090 {
6091 int b;
6092
6093 for (b = 0; b < numrecs; b++) {
6094 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
6095 *count += xfs_bmbt_get_blockcount(frp);
6096 }
6097 }
6098
6099 /*
6100 * Count leaf blocks given a range of extent records originally
6101 * in btree format.
6102 */
6103 STATIC void
6104 xfs_bmap_disk_count_leaves(
6105 struct xfs_mount *mp,
6106 struct xfs_btree_block *block,
6107 int numrecs,
6108 int *count)
6109 {
6110 int b;
6111 xfs_bmbt_rec_t *frp;
6112
6113 for (b = 1; b <= numrecs; b++) {
6114 frp = XFS_BMBT_REC_ADDR(mp, block, b);
6115 *count += xfs_bmbt_disk_get_blockcount(frp);
6116 }
6117 }
6118
6119 /*
6120 * dead simple method of punching delalyed allocation blocks from a range in
6121 * the inode. Walks a block at a time so will be slow, but is only executed in
6122 * rare error cases so the overhead is not critical. This will alays punch out
6123 * both the start and end blocks, even if the ranges only partially overlap
6124 * them, so it is up to the caller to ensure that partial blocks are not
6125 * passed in.
6126 */
6127 int
6128 xfs_bmap_punch_delalloc_range(
6129 struct xfs_inode *ip,
6130 xfs_fileoff_t start_fsb,
6131 xfs_fileoff_t length)
6132 {
6133 xfs_fileoff_t remaining = length;
6134 int error = 0;
6135
6136 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
6137
6138 do {
6139 int done;
6140 xfs_bmbt_irec_t imap;
6141 int nimaps = 1;
6142 xfs_fsblock_t firstblock;
6143 xfs_bmap_free_t flist;
6144
6145 /*
6146 * Map the range first and check that it is a delalloc extent
6147 * before trying to unmap the range. Otherwise we will be
6148 * trying to remove a real extent (which requires a
6149 * transaction) or a hole, which is probably a bad idea...
6150 */
6151 error = xfs_bmapi(NULL, ip, start_fsb, 1,
6152 XFS_BMAPI_ENTIRE, NULL, 0, &imap,
6153 &nimaps, NULL);
6154
6155 if (error) {
6156 /* something screwed, just bail */
6157 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
6158 xfs_fs_cmn_err(CE_ALERT, ip->i_mount,
6159 "Failed delalloc mapping lookup ino %lld fsb %lld.",
6160 ip->i_ino, start_fsb);
6161 }
6162 break;
6163 }
6164 if (!nimaps) {
6165 /* nothing there */
6166 goto next_block;
6167 }
6168 if (imap.br_startblock != DELAYSTARTBLOCK) {
6169 /* been converted, ignore */
6170 goto next_block;
6171 }
6172 WARN_ON(imap.br_blockcount == 0);
6173
6174 /*
6175 * Note: while we initialise the firstblock/flist pair, they
6176 * should never be used because blocks should never be
6177 * allocated or freed for a delalloc extent and hence we need
6178 * don't cancel or finish them after the xfs_bunmapi() call.
6179 */
6180 xfs_bmap_init(&flist, &firstblock);
6181 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
6182 &flist, &done);
6183 if (error)
6184 break;
6185
6186 ASSERT(!flist.xbf_count && !flist.xbf_first);
6187 next_block:
6188 start_fsb++;
6189 remaining--;
6190 } while(remaining > 0);
6191
6192 return error;
6193 }
ubuntu-zesty-kernel mirror