?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ARM: 7409/1: Do not call flush_cache_user_range with mmap_sem held
[linux/fpc-iii.git] / fs / xfs / xfs_bmap.c
bloba175933a7f48ddb46cde2011793af073284148c0
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_bmap_add_extent to handle cases converting a delayed
93 * allocation to a real allocation.
94 */
95 STATIC int /* error */
96 xfs_bmap_add_extent_delay_real(
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_filblks_t *dnew, /* new delayed-alloc indirect blocks */
102 xfs_fsblock_t *first, /* pointer to firstblock variable */
103 xfs_bmap_free_t *flist, /* list of extents to be freed */
104 int *logflagsp); /* inode logging flags */
105
106 /*
107 * Called by xfs_bmap_add_extent to handle cases converting a hole
108 * to a delayed allocation.
109 */
110 STATIC int /* error */
111 xfs_bmap_add_extent_hole_delay(
112 xfs_inode_t *ip, /* incore inode pointer */
113 xfs_extnum_t *idx, /* extent number to update/insert */
114 xfs_bmbt_irec_t *new, /* new data to add to file extents */
115 int *logflagsp); /* inode logging flags */
116
117 /*
118 * Called by xfs_bmap_add_extent to handle cases converting a hole
119 * to a real allocation.
120 */
121 STATIC int /* error */
122 xfs_bmap_add_extent_hole_real(
123 xfs_inode_t *ip, /* incore inode pointer */
124 xfs_extnum_t *idx, /* extent number to update/insert */
125 xfs_btree_cur_t *cur, /* if null, not a btree */
126 xfs_bmbt_irec_t *new, /* new data to add to file extents */
127 int *logflagsp, /* inode logging flags */
128 int whichfork); /* data or attr fork */
129
130 /*
131 * Called by xfs_bmap_add_extent to handle cases converting an unwritten
132 * allocation to a real allocation or vice versa.
133 */
134 STATIC int /* error */
135 xfs_bmap_add_extent_unwritten_real(
136 xfs_inode_t *ip, /* incore inode pointer */
137 xfs_extnum_t *idx, /* extent number to update/insert */
138 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
139 xfs_bmbt_irec_t *new, /* new data to add to file extents */
140 int *logflagsp); /* inode logging flags */
141
142 /*
143 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
144 * It figures out where to ask the underlying allocator to put the new extent.
145 */
146 STATIC int /* error */
147 xfs_bmap_alloc(
148 xfs_bmalloca_t *ap); /* bmap alloc argument struct */
149
150 /*
151 * Transform a btree format file with only one leaf node, where the
152 * extents list will fit in the inode, into an extents format file.
153 * Since the file extents are already in-core, all we have to do is
154 * give up the space for the btree root and pitch the leaf block.
155 */
156 STATIC int /* error */
157 xfs_bmap_btree_to_extents(
158 xfs_trans_t *tp, /* transaction pointer */
159 xfs_inode_t *ip, /* incore inode pointer */
160 xfs_btree_cur_t *cur, /* btree cursor */
161 int *logflagsp, /* inode logging flags */
162 int whichfork); /* data or attr fork */
163
164 /*
165 * Remove the entry "free" from the free item list. Prev points to the
166 * previous entry, unless "free" is the head of the list.
167 */
168 STATIC void
169 xfs_bmap_del_free(
170 xfs_bmap_free_t *flist, /* free item list header */
171 xfs_bmap_free_item_t *prev, /* previous item on list, if any */
172 xfs_bmap_free_item_t *free); /* list item to be freed */
173
174 /*
175 * Convert an extents-format file into a btree-format file.
176 * The new file will have a root block (in the inode) and a single child block.
177 */
178 STATIC int /* error */
179 xfs_bmap_extents_to_btree(
180 xfs_trans_t *tp, /* transaction pointer */
181 xfs_inode_t *ip, /* incore inode pointer */
182 xfs_fsblock_t *firstblock, /* first-block-allocated */
183 xfs_bmap_free_t *flist, /* blocks freed in xaction */
184 xfs_btree_cur_t **curp, /* cursor returned to caller */
185 int wasdel, /* converting a delayed alloc */
186 int *logflagsp, /* inode logging flags */
187 int whichfork); /* data or attr fork */
188
189 /*
190 * Convert a local file to an extents file.
191 * This code is sort of bogus, since the file data needs to get
192 * logged so it won't be lost. The bmap-level manipulations are ok, though.
193 */
194 STATIC int /* error */
195 xfs_bmap_local_to_extents(
196 xfs_trans_t *tp, /* transaction pointer */
197 xfs_inode_t *ip, /* incore inode pointer */
198 xfs_fsblock_t *firstblock, /* first block allocated in xaction */
199 xfs_extlen_t total, /* total blocks needed by transaction */
200 int *logflagsp, /* inode logging flags */
201 int whichfork); /* data or attr fork */
202
203 /*
204 * Search the extents list for the inode, for the extent containing bno.
205 * If bno lies in a hole, point to the next entry. If bno lies past eof,
206 * *eofp will be set, and *prevp will contain the last entry (null if none).
207 * Else, *lastxp will be set to the index of the found
208 * entry; *gotp will contain the entry.
209 */
210 STATIC xfs_bmbt_rec_host_t * /* pointer to found extent entry */
211 xfs_bmap_search_extents(
212 xfs_inode_t *ip, /* incore inode pointer */
213 xfs_fileoff_t bno, /* block number searched for */
214 int whichfork, /* data or attr fork */
215 int *eofp, /* out: end of file found */
216 xfs_extnum_t *lastxp, /* out: last extent index */
217 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
218 xfs_bmbt_irec_t *prevp); /* out: previous extent entry found */
219
220 /*
221 * Check the last inode extent to determine whether this allocation will result
222 * in blocks being allocated at the end of the file. When we allocate new data
223 * blocks at the end of the file which do not start at the previous data block,
224 * we will try to align the new blocks at stripe unit boundaries.
225 */
226 STATIC int /* error */
227 xfs_bmap_isaeof(
228 xfs_inode_t *ip, /* incore inode pointer */
229 xfs_fileoff_t off, /* file offset in fsblocks */
230 int whichfork, /* data or attribute fork */
231 char *aeof); /* return value */
232
233 /*
234 * Compute the worst-case number of indirect blocks that will be used
235 * for ip's delayed extent of length "len".
236 */
237 STATIC xfs_filblks_t
238 xfs_bmap_worst_indlen(
239 xfs_inode_t *ip, /* incore inode pointer */
240 xfs_filblks_t len); /* delayed extent length */
241
242 #ifdef DEBUG
243 /*
244 * Perform various validation checks on the values being returned
245 * from xfs_bmapi().
246 */
247 STATIC void
248 xfs_bmap_validate_ret(
249 xfs_fileoff_t bno,
250 xfs_filblks_t len,
251 int flags,
252 xfs_bmbt_irec_t *mval,
253 int nmap,
254 int ret_nmap);
255 #else
256 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
257 #endif /* DEBUG */
258
259 STATIC int
260 xfs_bmap_count_tree(
261 xfs_mount_t *mp,
262 xfs_trans_t *tp,
263 xfs_ifork_t *ifp,
264 xfs_fsblock_t blockno,
265 int levelin,
266 int *count);
267
268 STATIC void
269 xfs_bmap_count_leaves(
270 xfs_ifork_t *ifp,
271 xfs_extnum_t idx,
272 int numrecs,
273 int *count);
274
275 STATIC void
276 xfs_bmap_disk_count_leaves(
277 struct xfs_mount *mp,
278 struct xfs_btree_block *block,
279 int numrecs,
280 int *count);
281
282 /*
283 * Bmap internal routines.
284 */
285
286 STATIC int /* error */
287 xfs_bmbt_lookup_eq(
288 struct xfs_btree_cur *cur,
289 xfs_fileoff_t off,
290 xfs_fsblock_t bno,
291 xfs_filblks_t len,
292 int *stat) /* success/failure */
293 {
294 cur->bc_rec.b.br_startoff = off;
295 cur->bc_rec.b.br_startblock = bno;
296 cur->bc_rec.b.br_blockcount = len;
297 return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
298 }
299
300 STATIC int /* error */
301 xfs_bmbt_lookup_ge(
302 struct xfs_btree_cur *cur,
303 xfs_fileoff_t off,
304 xfs_fsblock_t bno,
305 xfs_filblks_t len,
306 int *stat) /* success/failure */
307 {
308 cur->bc_rec.b.br_startoff = off;
309 cur->bc_rec.b.br_startblock = bno;
310 cur->bc_rec.b.br_blockcount = len;
311 return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
312 }
313
314 /*
315 * Update the record referred to by cur to the value given
316 * by [off, bno, len, state].
317 * This either works (return 0) or gets an EFSCORRUPTED error.
318 */
319 STATIC int
320 xfs_bmbt_update(
321 struct xfs_btree_cur *cur,
322 xfs_fileoff_t off,
323 xfs_fsblock_t bno,
324 xfs_filblks_t len,
325 xfs_exntst_t state)
326 {
327 union xfs_btree_rec rec;
328
329 xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
330 return xfs_btree_update(cur, &rec);
331 }
332
333 /*
334 * Called from xfs_bmap_add_attrfork to handle btree format files.
335 */
336 STATIC int /* error */
337 xfs_bmap_add_attrfork_btree(
338 xfs_trans_t *tp, /* transaction pointer */
339 xfs_inode_t *ip, /* incore inode pointer */
340 xfs_fsblock_t *firstblock, /* first block allocated */
341 xfs_bmap_free_t *flist, /* blocks to free at commit */
342 int *flags) /* inode logging flags */
343 {
344 xfs_btree_cur_t *cur; /* btree cursor */
345 int error; /* error return value */
346 xfs_mount_t *mp; /* file system mount struct */
347 int stat; /* newroot status */
348
349 mp = ip->i_mount;
350 if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
351 *flags |= XFS_ILOG_DBROOT;
352 else {
353 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
354 cur->bc_private.b.flist = flist;
355 cur->bc_private.b.firstblock = *firstblock;
356 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
357 goto error0;
358 /* must be at least one entry */
359 XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
360 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
361 goto error0;
362 if (stat == 0) {
363 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
364 return XFS_ERROR(ENOSPC);
365 }
366 *firstblock = cur->bc_private.b.firstblock;
367 cur->bc_private.b.allocated = 0;
368 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
369 }
370 return 0;
371 error0:
372 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
373 return error;
374 }
375
376 /*
377 * Called from xfs_bmap_add_attrfork to handle extents format files.
378 */
379 STATIC int /* error */
380 xfs_bmap_add_attrfork_extents(
381 xfs_trans_t *tp, /* transaction pointer */
382 xfs_inode_t *ip, /* incore inode pointer */
383 xfs_fsblock_t *firstblock, /* first block allocated */
384 xfs_bmap_free_t *flist, /* blocks to free at commit */
385 int *flags) /* inode logging flags */
386 {
387 xfs_btree_cur_t *cur; /* bmap btree cursor */
388 int error; /* error return value */
389
390 if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
391 return 0;
392 cur = NULL;
393 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
394 flags, XFS_DATA_FORK);
395 if (cur) {
396 cur->bc_private.b.allocated = 0;
397 xfs_btree_del_cursor(cur,
398 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
399 }
400 return error;
401 }
402
403 /*
404 * Called from xfs_bmap_add_attrfork to handle local format files.
405 */
406 STATIC int /* error */
407 xfs_bmap_add_attrfork_local(
408 xfs_trans_t *tp, /* transaction pointer */
409 xfs_inode_t *ip, /* incore inode pointer */
410 xfs_fsblock_t *firstblock, /* first block allocated */
411 xfs_bmap_free_t *flist, /* blocks to free at commit */
412 int *flags) /* inode logging flags */
413 {
414 xfs_da_args_t dargs; /* args for dir/attr code */
415 int error; /* error return value */
416 xfs_mount_t *mp; /* mount structure pointer */
417
418 if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
419 return 0;
420 if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
421 mp = ip->i_mount;
422 memset(&dargs, 0, sizeof(dargs));
423 dargs.dp = ip;
424 dargs.firstblock = firstblock;
425 dargs.flist = flist;
426 dargs.total = mp->m_dirblkfsbs;
427 dargs.whichfork = XFS_DATA_FORK;
428 dargs.trans = tp;
429 error = xfs_dir2_sf_to_block(&dargs);
430 } else
431 error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
432 XFS_DATA_FORK);
433 return error;
434 }
435
436 /*
437 * Called by xfs_bmapi to update file extent records and the btree
438 * after allocating space (or doing a delayed allocation).
439 */
440 STATIC int /* error */
441 xfs_bmap_add_extent(
442 xfs_inode_t *ip, /* incore inode pointer */
443 xfs_extnum_t *idx, /* extent number to update/insert */
444 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
445 xfs_bmbt_irec_t *new, /* new data to add to file extents */
446 xfs_fsblock_t *first, /* pointer to firstblock variable */
447 xfs_bmap_free_t *flist, /* list of extents to be freed */
448 int *logflagsp, /* inode logging flags */
449 int whichfork) /* data or attr fork */
450 {
451 xfs_btree_cur_t *cur; /* btree cursor or null */
452 xfs_filblks_t da_new; /* new count del alloc blocks used */
453 xfs_filblks_t da_old; /* old count del alloc blocks used */
454 int error; /* error return value */
455 xfs_ifork_t *ifp; /* inode fork ptr */
456 int logflags; /* returned value */
457 xfs_extnum_t nextents; /* number of extents in file now */
458
459 XFS_STATS_INC(xs_add_exlist);
460
461 cur = *curp;
462 ifp = XFS_IFORK_PTR(ip, whichfork);
463 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
464 da_old = da_new = 0;
465 error = 0;
466
467 ASSERT(*idx >= 0);
468 ASSERT(*idx <= nextents);
469
470 /*
471 * This is the first extent added to a new/empty file.
472 * Special case this one, so other routines get to assume there are
473 * already extents in the list.
474 */
475 if (nextents == 0) {
476 xfs_iext_insert(ip, *idx, 1, new,
477 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
478
479 ASSERT(cur == NULL);
480
481 if (!isnullstartblock(new->br_startblock)) {
482 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
483 logflags = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
484 } else
485 logflags = 0;
486 }
487 /*
488 * Any kind of new delayed allocation goes here.
489 */
490 else if (isnullstartblock(new->br_startblock)) {
491 if (cur)
492 ASSERT((cur->bc_private.b.flags &
493 XFS_BTCUR_BPRV_WASDEL) == 0);
494 error = xfs_bmap_add_extent_hole_delay(ip, idx, new,
495 &logflags);
496 }
497 /*
498 * Real allocation off the end of the file.
499 */
500 else if (*idx == nextents) {
501 if (cur)
502 ASSERT((cur->bc_private.b.flags &
503 XFS_BTCUR_BPRV_WASDEL) == 0);
504 error = xfs_bmap_add_extent_hole_real(ip, idx, cur, new,
505 &logflags, whichfork);
506 } else {
507 xfs_bmbt_irec_t prev; /* old extent at offset idx */
508
509 /*
510 * Get the record referred to by idx.
511 */
512 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &prev);
513 /*
514 * If it's a real allocation record, and the new allocation ends
515 * after the start of the referred to record, then we're filling
516 * in a delayed or unwritten allocation with a real one, or
517 * converting real back to unwritten.
518 */
519 if (!isnullstartblock(new->br_startblock) &&
520 new->br_startoff + new->br_blockcount > prev.br_startoff) {
521 if (prev.br_state != XFS_EXT_UNWRITTEN &&
522 isnullstartblock(prev.br_startblock)) {
523 da_old = startblockval(prev.br_startblock);
524 if (cur)
525 ASSERT(cur->bc_private.b.flags &
526 XFS_BTCUR_BPRV_WASDEL);
527 error = xfs_bmap_add_extent_delay_real(ip,
528 idx, &cur, new, &da_new,
529 first, flist, &logflags);
530 } else {
531 ASSERT(new->br_state == XFS_EXT_NORM ||
532 new->br_state == XFS_EXT_UNWRITTEN);
533
534 error = xfs_bmap_add_extent_unwritten_real(ip,
535 idx, &cur, new, &logflags);
536 if (error)
537 goto done;
538 }
539 }
540 /*
541 * Otherwise we're filling in a hole with an allocation.
542 */
543 else {
544 if (cur)
545 ASSERT((cur->bc_private.b.flags &
546 XFS_BTCUR_BPRV_WASDEL) == 0);
547 error = xfs_bmap_add_extent_hole_real(ip, idx, cur,
548 new, &logflags, whichfork);
549 }
550 }
551
552 if (error)
553 goto done;
554 ASSERT(*curp == cur || *curp == NULL);
555
556 /*
557 * Convert to a btree if necessary.
558 */
559 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
560 XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
561 int tmp_logflags; /* partial log flag return val */
562
563 ASSERT(cur == NULL);
564 error = xfs_bmap_extents_to_btree(ip->i_transp, ip, first,
565 flist, &cur, da_old > 0, &tmp_logflags, whichfork);
566 logflags |= tmp_logflags;
567 if (error)
568 goto done;
569 }
570 /*
571 * Adjust for changes in reserved delayed indirect blocks.
572 * Nothing to do for disk quotas here.
573 */
574 if (da_old || da_new) {
575 xfs_filblks_t nblks;
576
577 nblks = da_new;
578 if (cur)
579 nblks += cur->bc_private.b.allocated;
580 ASSERT(nblks <= da_old);
581 if (nblks < da_old)
582 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
583 (int64_t)(da_old - nblks), 0);
584 }
585 /*
586 * Clear out the allocated field, done with it now in any case.
587 */
588 if (cur) {
589 cur->bc_private.b.allocated = 0;
590 *curp = cur;
591 }
592 done:
593 #ifdef DEBUG
594 if (!error)
595 xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
596 #endif
597 *logflagsp = logflags;
598 return error;
599 }
600
601 /*
602 * Called by xfs_bmap_add_extent to handle cases converting a delayed
603 * allocation to a real allocation.
604 */
605 STATIC int /* error */
606 xfs_bmap_add_extent_delay_real(
607 xfs_inode_t *ip, /* incore inode pointer */
608 xfs_extnum_t *idx, /* extent number to update/insert */
609 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
610 xfs_bmbt_irec_t *new, /* new data to add to file extents */
611 xfs_filblks_t *dnew, /* new delayed-alloc indirect blocks */
612 xfs_fsblock_t *first, /* pointer to firstblock variable */
613 xfs_bmap_free_t *flist, /* list of extents to be freed */
614 int *logflagsp) /* inode logging flags */
615 {
616 xfs_btree_cur_t *cur; /* btree cursor */
617 int diff; /* temp value */
618 xfs_bmbt_rec_host_t *ep; /* extent entry for idx */
619 int error; /* error return value */
620 int i; /* temp state */
621 xfs_ifork_t *ifp; /* inode fork pointer */
622 xfs_fileoff_t new_endoff; /* end offset of new entry */
623 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
624 /* left is 0, right is 1, prev is 2 */
625 int rval=0; /* return value (logging flags) */
626 int state = 0;/* state bits, accessed thru macros */
627 xfs_filblks_t temp=0; /* value for dnew calculations */
628 xfs_filblks_t temp2=0;/* value for dnew calculations */
629 int tmp_rval; /* partial logging flags */
630
631 #define LEFT r[0]
632 #define RIGHT r[1]
633 #define PREV r[2]
634
635 /*
636 * Set up a bunch of variables to make the tests simpler.
637 */
638 cur = *curp;
639 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
640 ep = xfs_iext_get_ext(ifp, *idx);
641 xfs_bmbt_get_all(ep, &PREV);
642 new_endoff = new->br_startoff + new->br_blockcount;
643 ASSERT(PREV.br_startoff <= new->br_startoff);
644 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
645
646 /*
647 * Set flags determining what part of the previous delayed allocation
648 * extent is being replaced by a real allocation.
649 */
650 if (PREV.br_startoff == new->br_startoff)
651 state |= BMAP_LEFT_FILLING;
652 if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
653 state |= BMAP_RIGHT_FILLING;
654
655 /*
656 * Check and set flags if this segment has a left neighbor.
657 * Don't set contiguous if the combined extent would be too large.
658 */
659 if (*idx > 0) {
660 state |= BMAP_LEFT_VALID;
661 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
662
663 if (isnullstartblock(LEFT.br_startblock))
664 state |= BMAP_LEFT_DELAY;
665 }
666
667 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
668 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
669 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
670 LEFT.br_state == new->br_state &&
671 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
672 state |= BMAP_LEFT_CONTIG;
673
674 /*
675 * Check and set flags if this segment has a right neighbor.
676 * Don't set contiguous if the combined extent would be too large.
677 * Also check for all-three-contiguous being too large.
678 */
679 if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
680 state |= BMAP_RIGHT_VALID;
681 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
682
683 if (isnullstartblock(RIGHT.br_startblock))
684 state |= BMAP_RIGHT_DELAY;
685 }
686
687 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
688 new_endoff == RIGHT.br_startoff &&
689 new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
690 new->br_state == RIGHT.br_state &&
691 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
692 ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
693 BMAP_RIGHT_FILLING)) !=
694 (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
695 BMAP_RIGHT_FILLING) ||
696 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
697 <= MAXEXTLEN))
698 state |= BMAP_RIGHT_CONTIG;
699
700 error = 0;
701 /*
702 * Switch out based on the FILLING and CONTIG state bits.
703 */
704 switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
705 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
706 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
707 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
708 /*
709 * Filling in all of a previously delayed allocation extent.
710 * The left and right neighbors are both contiguous with new.
711 */
712 --*idx;
713 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
714 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
715 LEFT.br_blockcount + PREV.br_blockcount +
716 RIGHT.br_blockcount);
717 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
718
719 xfs_iext_remove(ip, *idx + 1, 2, state);
720 ip->i_d.di_nextents--;
721 if (cur == NULL)
722 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
723 else {
724 rval = XFS_ILOG_CORE;
725 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
726 RIGHT.br_startblock,
727 RIGHT.br_blockcount, &i)))
728 goto done;
729 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
730 if ((error = xfs_btree_delete(cur, &i)))
731 goto done;
732 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
733 if ((error = xfs_btree_decrement(cur, 0, &i)))
734 goto done;
735 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
736 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
737 LEFT.br_startblock,
738 LEFT.br_blockcount +
739 PREV.br_blockcount +
740 RIGHT.br_blockcount, LEFT.br_state)))
741 goto done;
742 }
743 *dnew = 0;
744 break;
745
746 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
747 /*
748 * Filling in all of a previously delayed allocation extent.
749 * The left neighbor is contiguous, the right is not.
750 */
751 --*idx;
752
753 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
754 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
755 LEFT.br_blockcount + PREV.br_blockcount);
756 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
757
758 xfs_iext_remove(ip, *idx + 1, 1, state);
759 if (cur == NULL)
760 rval = XFS_ILOG_DEXT;
761 else {
762 rval = 0;
763 if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
764 LEFT.br_startblock, LEFT.br_blockcount,
765 &i)))
766 goto done;
767 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
768 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
769 LEFT.br_startblock,
770 LEFT.br_blockcount +
771 PREV.br_blockcount, LEFT.br_state)))
772 goto done;
773 }
774 *dnew = 0;
775 break;
776
777 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
778 /*
779 * Filling in all of a previously delayed allocation extent.
780 * The right neighbor is contiguous, the left is not.
781 */
782 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
783 xfs_bmbt_set_startblock(ep, new->br_startblock);
784 xfs_bmbt_set_blockcount(ep,
785 PREV.br_blockcount + RIGHT.br_blockcount);
786 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
787
788 xfs_iext_remove(ip, *idx + 1, 1, state);
789 if (cur == NULL)
790 rval = XFS_ILOG_DEXT;
791 else {
792 rval = 0;
793 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
794 RIGHT.br_startblock,
795 RIGHT.br_blockcount, &i)))
796 goto done;
797 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
798 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
799 new->br_startblock,
800 PREV.br_blockcount +
801 RIGHT.br_blockcount, PREV.br_state)))
802 goto done;
803 }
804
805 *dnew = 0;
806 break;
807
808 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
809 /*
810 * Filling in all of a previously delayed allocation extent.
811 * Neither the left nor right neighbors are contiguous with
812 * the new one.
813 */
814 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
815 xfs_bmbt_set_startblock(ep, new->br_startblock);
816 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
817
818 ip->i_d.di_nextents++;
819 if (cur == NULL)
820 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
821 else {
822 rval = XFS_ILOG_CORE;
823 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
824 new->br_startblock, new->br_blockcount,
825 &i)))
826 goto done;
827 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
828 cur->bc_rec.b.br_state = XFS_EXT_NORM;
829 if ((error = xfs_btree_insert(cur, &i)))
830 goto done;
831 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
832 }
833
834 *dnew = 0;
835 break;
836
837 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
838 /*
839 * Filling in the first part of a previous delayed allocation.
840 * The left neighbor is contiguous.
841 */
842 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
843 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
844 LEFT.br_blockcount + new->br_blockcount);
845 xfs_bmbt_set_startoff(ep,
846 PREV.br_startoff + new->br_blockcount);
847 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
848
849 temp = PREV.br_blockcount - new->br_blockcount;
850 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
851 xfs_bmbt_set_blockcount(ep, temp);
852 if (cur == NULL)
853 rval = XFS_ILOG_DEXT;
854 else {
855 rval = 0;
856 if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
857 LEFT.br_startblock, LEFT.br_blockcount,
858 &i)))
859 goto done;
860 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
861 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
862 LEFT.br_startblock,
863 LEFT.br_blockcount +
864 new->br_blockcount,
865 LEFT.br_state)))
866 goto done;
867 }
868 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
869 startblockval(PREV.br_startblock));
870 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
871 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
872
873 --*idx;
874 *dnew = temp;
875 break;
876
877 case BMAP_LEFT_FILLING:
878 /*
879 * Filling in the first part of a previous delayed allocation.
880 * The left neighbor is not contiguous.
881 */
882 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
883 xfs_bmbt_set_startoff(ep, new_endoff);
884 temp = PREV.br_blockcount - new->br_blockcount;
885 xfs_bmbt_set_blockcount(ep, temp);
886 xfs_iext_insert(ip, *idx, 1, new, state);
887 ip->i_d.di_nextents++;
888 if (cur == NULL)
889 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
890 else {
891 rval = XFS_ILOG_CORE;
892 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
893 new->br_startblock, new->br_blockcount,
894 &i)))
895 goto done;
896 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
897 cur->bc_rec.b.br_state = XFS_EXT_NORM;
898 if ((error = xfs_btree_insert(cur, &i)))
899 goto done;
900 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
901 }
902 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
903 ip->i_d.di_nextents > ip->i_df.if_ext_max) {
904 error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
905 first, flist, &cur, 1, &tmp_rval,
906 XFS_DATA_FORK);
907 rval |= tmp_rval;
908 if (error)
909 goto done;
910 }
911 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
912 startblockval(PREV.br_startblock) -
913 (cur ? cur->bc_private.b.allocated : 0));
914 ep = xfs_iext_get_ext(ifp, *idx + 1);
915 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
916 trace_xfs_bmap_post_update(ip, *idx + 1, state, _THIS_IP_);
917
918 *dnew = temp;
919 break;
920
921 case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
922 /*
923 * Filling in the last part of a previous delayed allocation.
924 * The right neighbor is contiguous with the new allocation.
925 */
926 temp = PREV.br_blockcount - new->br_blockcount;
927 trace_xfs_bmap_pre_update(ip, *idx + 1, state, _THIS_IP_);
928 xfs_bmbt_set_blockcount(ep, temp);
929 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx + 1),
930 new->br_startoff, new->br_startblock,
931 new->br_blockcount + RIGHT.br_blockcount,
932 RIGHT.br_state);
933 trace_xfs_bmap_post_update(ip, *idx + 1, state, _THIS_IP_);
934 if (cur == NULL)
935 rval = XFS_ILOG_DEXT;
936 else {
937 rval = 0;
938 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
939 RIGHT.br_startblock,
940 RIGHT.br_blockcount, &i)))
941 goto done;
942 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
943 if ((error = xfs_bmbt_update(cur, new->br_startoff,
944 new->br_startblock,
945 new->br_blockcount +
946 RIGHT.br_blockcount,
947 RIGHT.br_state)))
948 goto done;
949 }
950
951 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
952 startblockval(PREV.br_startblock));
953 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
954 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
955 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
956
957 ++*idx;
958 *dnew = temp;
959 break;
960
961 case BMAP_RIGHT_FILLING:
962 /*
963 * Filling in the last part of a previous delayed allocation.
964 * The right neighbor is not contiguous.
965 */
966 temp = PREV.br_blockcount - new->br_blockcount;
967 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
968 xfs_bmbt_set_blockcount(ep, temp);
969 xfs_iext_insert(ip, *idx + 1, 1, new, state);
970 ip->i_d.di_nextents++;
971 if (cur == NULL)
972 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
973 else {
974 rval = XFS_ILOG_CORE;
975 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
976 new->br_startblock, new->br_blockcount,
977 &i)))
978 goto done;
979 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
980 cur->bc_rec.b.br_state = XFS_EXT_NORM;
981 if ((error = xfs_btree_insert(cur, &i)))
982 goto done;
983 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
984 }
985 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
986 ip->i_d.di_nextents > ip->i_df.if_ext_max) {
987 error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
988 first, flist, &cur, 1, &tmp_rval,
989 XFS_DATA_FORK);
990 rval |= tmp_rval;
991 if (error)
992 goto done;
993 }
994 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
995 startblockval(PREV.br_startblock) -
996 (cur ? cur->bc_private.b.allocated : 0));
997 ep = xfs_iext_get_ext(ifp, *idx);
998 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
999 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1000
1001 ++*idx;
1002 *dnew = temp;
1003 break;
1004
1005 case 0:
1006 /*
1007 * Filling in the middle part of a previous delayed allocation.
1008 * Contiguity is impossible here.
1009 * This case is avoided almost all the time.
1010 *
1011 * We start with a delayed allocation:
1012 *
1013 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
1014 * PREV @ idx
1015 *
1016 * and we are allocating:
1017 * +rrrrrrrrrrrrrrrrr+
1018 * new
1019 *
1020 * and we set it up for insertion as:
1021 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
1022 * new
1023 * PREV @ idx LEFT RIGHT
1024 * inserted at idx + 1
1025 */
1026 temp = new->br_startoff - PREV.br_startoff;
1027 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
1028 trace_xfs_bmap_pre_update(ip, *idx, 0, _THIS_IP_);
1029 xfs_bmbt_set_blockcount(ep, temp); /* truncate PREV */
1030 LEFT = *new;
1031 RIGHT.br_state = PREV.br_state;
1032 RIGHT.br_startblock = nullstartblock(
1033 (int)xfs_bmap_worst_indlen(ip, temp2));
1034 RIGHT.br_startoff = new_endoff;
1035 RIGHT.br_blockcount = temp2;
1036 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
1037 xfs_iext_insert(ip, *idx + 1, 2, &LEFT, state);
1038 ip->i_d.di_nextents++;
1039 if (cur == NULL)
1040 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1041 else {
1042 rval = XFS_ILOG_CORE;
1043 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1044 new->br_startblock, new->br_blockcount,
1045 &i)))
1046 goto done;
1047 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1048 cur->bc_rec.b.br_state = XFS_EXT_NORM;
1049 if ((error = xfs_btree_insert(cur, &i)))
1050 goto done;
1051 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1052 }
1053 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1054 ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1055 error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1056 first, flist, &cur, 1, &tmp_rval,
1057 XFS_DATA_FORK);
1058 rval |= tmp_rval;
1059 if (error)
1060 goto done;
1061 }
1062 temp = xfs_bmap_worst_indlen(ip, temp);
1063 temp2 = xfs_bmap_worst_indlen(ip, temp2);
1064 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
1065 (cur ? cur->bc_private.b.allocated : 0));
1066 if (diff > 0 &&
1067 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
1068 -((int64_t)diff), 0)) {
1069 /*
1070 * Ick gross gag me with a spoon.
1071 */
1072 ASSERT(0); /* want to see if this ever happens! */
1073 while (diff > 0) {
1074 if (temp) {
1075 temp--;
1076 diff--;
1077 if (!diff ||
1078 !xfs_icsb_modify_counters(ip->i_mount,
1079 XFS_SBS_FDBLOCKS,
1080 -((int64_t)diff), 0))
1081 break;
1082 }
1083 if (temp2) {
1084 temp2--;
1085 diff--;
1086 if (!diff ||
1087 !xfs_icsb_modify_counters(ip->i_mount,
1088 XFS_SBS_FDBLOCKS,
1089 -((int64_t)diff), 0))
1090 break;
1091 }
1092 }
1093 }
1094 ep = xfs_iext_get_ext(ifp, *idx);
1095 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
1096 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1097 trace_xfs_bmap_pre_update(ip, *idx + 2, state, _THIS_IP_);
1098 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx + 2),
1099 nullstartblock((int)temp2));
1100 trace_xfs_bmap_post_update(ip, *idx + 2, state, _THIS_IP_);
1101
1102 ++*idx;
1103 *dnew = temp + temp2;
1104 break;
1105
1106 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1107 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1108 case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1109 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1110 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1111 case BMAP_LEFT_CONTIG:
1112 case BMAP_RIGHT_CONTIG:
1113 /*
1114 * These cases are all impossible.
1115 */
1116 ASSERT(0);
1117 }
1118 *curp = cur;
1119 done:
1120 *logflagsp = rval;
1121 return error;
1122 #undef LEFT
1123 #undef RIGHT
1124 #undef PREV
1125 }
1126
1127 /*
1128 * Called by xfs_bmap_add_extent to handle cases converting an unwritten
1129 * allocation to a real allocation or vice versa.
1130 */
1131 STATIC int /* error */
1132 xfs_bmap_add_extent_unwritten_real(
1133 xfs_inode_t *ip, /* incore inode pointer */
1134 xfs_extnum_t *idx, /* extent number to update/insert */
1135 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
1136 xfs_bmbt_irec_t *new, /* new data to add to file extents */
1137 int *logflagsp) /* inode logging flags */
1138 {
1139 xfs_btree_cur_t *cur; /* btree cursor */
1140 xfs_bmbt_rec_host_t *ep; /* extent entry for idx */
1141 int error; /* error return value */
1142 int i; /* temp state */
1143 xfs_ifork_t *ifp; /* inode fork pointer */
1144 xfs_fileoff_t new_endoff; /* end offset of new entry */
1145 xfs_exntst_t newext; /* new extent state */
1146 xfs_exntst_t oldext; /* old extent state */
1147 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
1148 /* left is 0, right is 1, prev is 2 */
1149 int rval=0; /* return value (logging flags) */
1150 int state = 0;/* state bits, accessed thru macros */
1151
1152 #define LEFT r[0]
1153 #define RIGHT r[1]
1154 #define PREV r[2]
1155 /*
1156 * Set up a bunch of variables to make the tests simpler.
1157 */
1158 error = 0;
1159 cur = *curp;
1160 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1161 ep = xfs_iext_get_ext(ifp, *idx);
1162 xfs_bmbt_get_all(ep, &PREV);
1163 newext = new->br_state;
1164 oldext = (newext == XFS_EXT_UNWRITTEN) ?
1165 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
1166 ASSERT(PREV.br_state == oldext);
1167 new_endoff = new->br_startoff + new->br_blockcount;
1168 ASSERT(PREV.br_startoff <= new->br_startoff);
1169 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1170
1171 /*
1172 * Set flags determining what part of the previous oldext allocation
1173 * extent is being replaced by a newext allocation.
1174 */
1175 if (PREV.br_startoff == new->br_startoff)
1176 state |= BMAP_LEFT_FILLING;
1177 if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1178 state |= BMAP_RIGHT_FILLING;
1179
1180 /*
1181 * Check and set flags if this segment has a left neighbor.
1182 * Don't set contiguous if the combined extent would be too large.
1183 */
1184 if (*idx > 0) {
1185 state |= BMAP_LEFT_VALID;
1186 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
1187
1188 if (isnullstartblock(LEFT.br_startblock))
1189 state |= BMAP_LEFT_DELAY;
1190 }
1191
1192 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1193 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1194 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1195 LEFT.br_state == newext &&
1196 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1197 state |= BMAP_LEFT_CONTIG;
1198
1199 /*
1200 * Check and set flags if this segment has a right neighbor.
1201 * Don't set contiguous if the combined extent would be too large.
1202 * Also check for all-three-contiguous being too large.
1203 */
1204 if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1205 state |= BMAP_RIGHT_VALID;
1206 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
1207 if (isnullstartblock(RIGHT.br_startblock))
1208 state |= BMAP_RIGHT_DELAY;
1209 }
1210
1211 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1212 new_endoff == RIGHT.br_startoff &&
1213 new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1214 newext == RIGHT.br_state &&
1215 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1216 ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1217 BMAP_RIGHT_FILLING)) !=
1218 (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1219 BMAP_RIGHT_FILLING) ||
1220 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1221 <= MAXEXTLEN))
1222 state |= BMAP_RIGHT_CONTIG;
1223
1224 /*
1225 * Switch out based on the FILLING and CONTIG state bits.
1226 */
1227 switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1228 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1229 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1230 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1231 /*
1232 * Setting all of a previous oldext extent to newext.
1233 * The left and right neighbors are both contiguous with new.
1234 */
1235 --*idx;
1236
1237 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1238 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1239 LEFT.br_blockcount + PREV.br_blockcount +
1240 RIGHT.br_blockcount);
1241 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1242
1243 xfs_iext_remove(ip, *idx + 1, 2, state);
1244 ip->i_d.di_nextents -= 2;
1245 if (cur == NULL)
1246 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1247 else {
1248 rval = XFS_ILOG_CORE;
1249 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1250 RIGHT.br_startblock,
1251 RIGHT.br_blockcount, &i)))
1252 goto done;
1253 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1254 if ((error = xfs_btree_delete(cur, &i)))
1255 goto done;
1256 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1257 if ((error = xfs_btree_decrement(cur, 0, &i)))
1258 goto done;
1259 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1260 if ((error = xfs_btree_delete(cur, &i)))
1261 goto done;
1262 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1263 if ((error = xfs_btree_decrement(cur, 0, &i)))
1264 goto done;
1265 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1266 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1267 LEFT.br_startblock,
1268 LEFT.br_blockcount + PREV.br_blockcount +
1269 RIGHT.br_blockcount, LEFT.br_state)))
1270 goto done;
1271 }
1272 break;
1273
1274 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1275 /*
1276 * Setting all of a previous oldext extent to newext.
1277 * The left neighbor is contiguous, the right is not.
1278 */
1279 --*idx;
1280
1281 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1282 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1283 LEFT.br_blockcount + PREV.br_blockcount);
1284 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1285
1286 xfs_iext_remove(ip, *idx + 1, 1, state);
1287 ip->i_d.di_nextents--;
1288 if (cur == NULL)
1289 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1290 else {
1291 rval = XFS_ILOG_CORE;
1292 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1293 PREV.br_startblock, PREV.br_blockcount,
1294 &i)))
1295 goto done;
1296 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1297 if ((error = xfs_btree_delete(cur, &i)))
1298 goto done;
1299 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1300 if ((error = xfs_btree_decrement(cur, 0, &i)))
1301 goto done;
1302 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1303 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1304 LEFT.br_startblock,
1305 LEFT.br_blockcount + PREV.br_blockcount,
1306 LEFT.br_state)))
1307 goto done;
1308 }
1309 break;
1310
1311 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1312 /*
1313 * Setting all of a previous oldext extent to newext.
1314 * The right neighbor is contiguous, the left is not.
1315 */
1316 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1317 xfs_bmbt_set_blockcount(ep,
1318 PREV.br_blockcount + RIGHT.br_blockcount);
1319 xfs_bmbt_set_state(ep, newext);
1320 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1321 xfs_iext_remove(ip, *idx + 1, 1, state);
1322 ip->i_d.di_nextents--;
1323 if (cur == NULL)
1324 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1325 else {
1326 rval = XFS_ILOG_CORE;
1327 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1328 RIGHT.br_startblock,
1329 RIGHT.br_blockcount, &i)))
1330 goto done;
1331 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1332 if ((error = xfs_btree_delete(cur, &i)))
1333 goto done;
1334 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1335 if ((error = xfs_btree_decrement(cur, 0, &i)))
1336 goto done;
1337 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1338 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1339 new->br_startblock,
1340 new->br_blockcount + RIGHT.br_blockcount,
1341 newext)))
1342 goto done;
1343 }
1344 break;
1345
1346 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1347 /*
1348 * Setting all of a previous oldext extent to newext.
1349 * Neither the left nor right neighbors are contiguous with
1350 * the new one.
1351 */
1352 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1353 xfs_bmbt_set_state(ep, newext);
1354 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1355
1356 if (cur == NULL)
1357 rval = XFS_ILOG_DEXT;
1358 else {
1359 rval = 0;
1360 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1361 new->br_startblock, new->br_blockcount,
1362 &i)))
1363 goto done;
1364 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1365 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1366 new->br_startblock, new->br_blockcount,
1367 newext)))
1368 goto done;
1369 }
1370 break;
1371
1372 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1373 /*
1374 * Setting the first part of a previous oldext extent to newext.
1375 * The left neighbor is contiguous.
1376 */
1377 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
1378 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
1379 LEFT.br_blockcount + new->br_blockcount);
1380 xfs_bmbt_set_startoff(ep,
1381 PREV.br_startoff + new->br_blockcount);
1382 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
1383
1384 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1385 xfs_bmbt_set_startblock(ep,
1386 new->br_startblock + new->br_blockcount);
1387 xfs_bmbt_set_blockcount(ep,
1388 PREV.br_blockcount - new->br_blockcount);
1389 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1390
1391 --*idx;
1392
1393 if (cur == NULL)
1394 rval = XFS_ILOG_DEXT;
1395 else {
1396 rval = 0;
1397 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1398 PREV.br_startblock, PREV.br_blockcount,
1399 &i)))
1400 goto done;
1401 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1402 if ((error = xfs_bmbt_update(cur,
1403 PREV.br_startoff + new->br_blockcount,
1404 PREV.br_startblock + new->br_blockcount,
1405 PREV.br_blockcount - new->br_blockcount,
1406 oldext)))
1407 goto done;
1408 if ((error = xfs_btree_decrement(cur, 0, &i)))
1409 goto done;
1410 if (xfs_bmbt_update(cur, LEFT.br_startoff,
1411 LEFT.br_startblock,
1412 LEFT.br_blockcount + new->br_blockcount,
1413 LEFT.br_state))
1414 goto done;
1415 }
1416 break;
1417
1418 case BMAP_LEFT_FILLING:
1419 /*
1420 * Setting the first part of a previous oldext extent to newext.
1421 * The left neighbor is not contiguous.
1422 */
1423 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1424 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1425 xfs_bmbt_set_startoff(ep, new_endoff);
1426 xfs_bmbt_set_blockcount(ep,
1427 PREV.br_blockcount - new->br_blockcount);
1428 xfs_bmbt_set_startblock(ep,
1429 new->br_startblock + new->br_blockcount);
1430 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1431
1432 xfs_iext_insert(ip, *idx, 1, new, state);
1433 ip->i_d.di_nextents++;
1434 if (cur == NULL)
1435 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1436 else {
1437 rval = XFS_ILOG_CORE;
1438 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1439 PREV.br_startblock, PREV.br_blockcount,
1440 &i)))
1441 goto done;
1442 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1443 if ((error = xfs_bmbt_update(cur,
1444 PREV.br_startoff + new->br_blockcount,
1445 PREV.br_startblock + new->br_blockcount,
1446 PREV.br_blockcount - new->br_blockcount,
1447 oldext)))
1448 goto done;
1449 cur->bc_rec.b = *new;
1450 if ((error = xfs_btree_insert(cur, &i)))
1451 goto done;
1452 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1453 }
1454 break;
1455
1456 case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1457 /*
1458 * Setting the last part of a previous oldext extent to newext.
1459 * The right neighbor is contiguous with the new allocation.
1460 */
1461 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1462 xfs_bmbt_set_blockcount(ep,
1463 PREV.br_blockcount - new->br_blockcount);
1464 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1465
1466 ++*idx;
1467
1468 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1469 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1470 new->br_startoff, new->br_startblock,
1471 new->br_blockcount + RIGHT.br_blockcount, newext);
1472 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1473
1474 if (cur == NULL)
1475 rval = XFS_ILOG_DEXT;
1476 else {
1477 rval = 0;
1478 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1479 PREV.br_startblock,
1480 PREV.br_blockcount, &i)))
1481 goto done;
1482 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1483 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1484 PREV.br_startblock,
1485 PREV.br_blockcount - new->br_blockcount,
1486 oldext)))
1487 goto done;
1488 if ((error = xfs_btree_increment(cur, 0, &i)))
1489 goto done;
1490 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1491 new->br_startblock,
1492 new->br_blockcount + RIGHT.br_blockcount,
1493 newext)))
1494 goto done;
1495 }
1496 break;
1497
1498 case BMAP_RIGHT_FILLING:
1499 /*
1500 * Setting the last part of a previous oldext extent to newext.
1501 * The right neighbor is not contiguous.
1502 */
1503 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1504 xfs_bmbt_set_blockcount(ep,
1505 PREV.br_blockcount - new->br_blockcount);
1506 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1507
1508 ++*idx;
1509 xfs_iext_insert(ip, *idx, 1, new, state);
1510
1511 ip->i_d.di_nextents++;
1512 if (cur == NULL)
1513 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1514 else {
1515 rval = XFS_ILOG_CORE;
1516 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1517 PREV.br_startblock, PREV.br_blockcount,
1518 &i)))
1519 goto done;
1520 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1521 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1522 PREV.br_startblock,
1523 PREV.br_blockcount - new->br_blockcount,
1524 oldext)))
1525 goto done;
1526 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1527 new->br_startblock, new->br_blockcount,
1528 &i)))
1529 goto done;
1530 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1531 cur->bc_rec.b.br_state = XFS_EXT_NORM;
1532 if ((error = xfs_btree_insert(cur, &i)))
1533 goto done;
1534 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1535 }
1536 break;
1537
1538 case 0:
1539 /*
1540 * Setting the middle part of a previous oldext extent to
1541 * newext. Contiguity is impossible here.
1542 * One extent becomes three extents.
1543 */
1544 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1545 xfs_bmbt_set_blockcount(ep,
1546 new->br_startoff - PREV.br_startoff);
1547 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1548
1549 r[0] = *new;
1550 r[1].br_startoff = new_endoff;
1551 r[1].br_blockcount =
1552 PREV.br_startoff + PREV.br_blockcount - new_endoff;
1553 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1554 r[1].br_state = oldext;
1555
1556 ++*idx;
1557 xfs_iext_insert(ip, *idx, 2, &r[0], state);
1558
1559 ip->i_d.di_nextents += 2;
1560 if (cur == NULL)
1561 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1562 else {
1563 rval = XFS_ILOG_CORE;
1564 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1565 PREV.br_startblock, PREV.br_blockcount,
1566 &i)))
1567 goto done;
1568 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1569 /* new right extent - oldext */
1570 if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1571 r[1].br_startblock, r[1].br_blockcount,
1572 r[1].br_state)))
1573 goto done;
1574 /* new left extent - oldext */
1575 cur->bc_rec.b = PREV;
1576 cur->bc_rec.b.br_blockcount =
1577 new->br_startoff - PREV.br_startoff;
1578 if ((error = xfs_btree_insert(cur, &i)))
1579 goto done;
1580 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1581 /*
1582 * Reset the cursor to the position of the new extent
1583 * we are about to insert as we can't trust it after
1584 * the previous insert.
1585 */
1586 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1587 new->br_startblock, new->br_blockcount,
1588 &i)))
1589 goto done;
1590 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1591 /* new middle extent - newext */
1592 cur->bc_rec.b.br_state = new->br_state;
1593 if ((error = xfs_btree_insert(cur, &i)))
1594 goto done;
1595 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1596 }
1597 break;
1598
1599 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1600 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1601 case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1602 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1603 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1604 case BMAP_LEFT_CONTIG:
1605 case BMAP_RIGHT_CONTIG:
1606 /*
1607 * These cases are all impossible.
1608 */
1609 ASSERT(0);
1610 }
1611 *curp = cur;
1612 done:
1613 *logflagsp = rval;
1614 return error;
1615 #undef LEFT
1616 #undef RIGHT
1617 #undef PREV
1618 }
1619
1620 /*
1621 * Called by xfs_bmap_add_extent to handle cases converting a hole
1622 * to a delayed allocation.
1623 */
1624 /*ARGSUSED*/
1625 STATIC int /* error */
1626 xfs_bmap_add_extent_hole_delay(
1627 xfs_inode_t *ip, /* incore inode pointer */
1628 xfs_extnum_t *idx, /* extent number to update/insert */
1629 xfs_bmbt_irec_t *new, /* new data to add to file extents */
1630 int *logflagsp) /* inode logging flags */
1631 {
1632 xfs_ifork_t *ifp; /* inode fork pointer */
1633 xfs_bmbt_irec_t left; /* left neighbor extent entry */
1634 xfs_filblks_t newlen=0; /* new indirect size */
1635 xfs_filblks_t oldlen=0; /* old indirect size */
1636 xfs_bmbt_irec_t right; /* right neighbor extent entry */
1637 int state; /* state bits, accessed thru macros */
1638 xfs_filblks_t temp=0; /* temp for indirect calculations */
1639
1640 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1641 state = 0;
1642 ASSERT(isnullstartblock(new->br_startblock));
1643
1644 /*
1645 * Check and set flags if this segment has a left neighbor
1646 */
1647 if (*idx > 0) {
1648 state |= BMAP_LEFT_VALID;
1649 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
1650
1651 if (isnullstartblock(left.br_startblock))
1652 state |= BMAP_LEFT_DELAY;
1653 }
1654
1655 /*
1656 * Check and set flags if the current (right) segment exists.
1657 * If it doesn't exist, we're converting the hole at end-of-file.
1658 */
1659 if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1660 state |= BMAP_RIGHT_VALID;
1661 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
1662
1663 if (isnullstartblock(right.br_startblock))
1664 state |= BMAP_RIGHT_DELAY;
1665 }
1666
1667 /*
1668 * Set contiguity flags on the left and right neighbors.
1669 * Don't let extents get too large, even if the pieces are contiguous.
1670 */
1671 if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
1672 left.br_startoff + left.br_blockcount == new->br_startoff &&
1673 left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1674 state |= BMAP_LEFT_CONTIG;
1675
1676 if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
1677 new->br_startoff + new->br_blockcount == right.br_startoff &&
1678 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1679 (!(state & BMAP_LEFT_CONTIG) ||
1680 (left.br_blockcount + new->br_blockcount +
1681 right.br_blockcount <= MAXEXTLEN)))
1682 state |= BMAP_RIGHT_CONTIG;
1683
1684 /*
1685 * Switch out based on the contiguity flags.
1686 */
1687 switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1688 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1689 /*
1690 * New allocation is contiguous with delayed allocations
1691 * on the left and on the right.
1692 * Merge all three into a single extent record.
1693 */
1694 --*idx;
1695 temp = left.br_blockcount + new->br_blockcount +
1696 right.br_blockcount;
1697
1698 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1699 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1700 oldlen = startblockval(left.br_startblock) +
1701 startblockval(new->br_startblock) +
1702 startblockval(right.br_startblock);
1703 newlen = xfs_bmap_worst_indlen(ip, temp);
1704 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1705 nullstartblock((int)newlen));
1706 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1707
1708 xfs_iext_remove(ip, *idx + 1, 1, state);
1709 break;
1710
1711 case BMAP_LEFT_CONTIG:
1712 /*
1713 * New allocation is contiguous with a delayed allocation
1714 * on the left.
1715 * Merge the new allocation with the left neighbor.
1716 */
1717 --*idx;
1718 temp = left.br_blockcount + new->br_blockcount;
1719
1720 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1721 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1722 oldlen = startblockval(left.br_startblock) +
1723 startblockval(new->br_startblock);
1724 newlen = xfs_bmap_worst_indlen(ip, temp);
1725 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1726 nullstartblock((int)newlen));
1727 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1728 break;
1729
1730 case BMAP_RIGHT_CONTIG:
1731 /*
1732 * New allocation is contiguous with a delayed allocation
1733 * on the right.
1734 * Merge the new allocation with the right neighbor.
1735 */
1736 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1737 temp = new->br_blockcount + right.br_blockcount;
1738 oldlen = startblockval(new->br_startblock) +
1739 startblockval(right.br_startblock);
1740 newlen = xfs_bmap_worst_indlen(ip, temp);
1741 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1742 new->br_startoff,
1743 nullstartblock((int)newlen), temp, right.br_state);
1744 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1745 break;
1746
1747 case 0:
1748 /*
1749 * New allocation is not contiguous with another
1750 * delayed allocation.
1751 * Insert a new entry.
1752 */
1753 oldlen = newlen = 0;
1754 xfs_iext_insert(ip, *idx, 1, new, state);
1755 break;
1756 }
1757 if (oldlen != newlen) {
1758 ASSERT(oldlen > newlen);
1759 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
1760 (int64_t)(oldlen - newlen), 0);
1761 /*
1762 * Nothing to do for disk quota accounting here.
1763 */
1764 }
1765 *logflagsp = 0;
1766 return 0;
1767 }
1768
1769 /*
1770 * Called by xfs_bmap_add_extent to handle cases converting a hole
1771 * to a real allocation.
1772 */
1773 STATIC int /* error */
1774 xfs_bmap_add_extent_hole_real(
1775 xfs_inode_t *ip, /* incore inode pointer */
1776 xfs_extnum_t *idx, /* extent number to update/insert */
1777 xfs_btree_cur_t *cur, /* if null, not a btree */
1778 xfs_bmbt_irec_t *new, /* new data to add to file extents */
1779 int *logflagsp, /* inode logging flags */
1780 int whichfork) /* data or attr fork */
1781 {
1782 int error; /* error return value */
1783 int i; /* temp state */
1784 xfs_ifork_t *ifp; /* inode fork pointer */
1785 xfs_bmbt_irec_t left; /* left neighbor extent entry */
1786 xfs_bmbt_irec_t right; /* right neighbor extent entry */
1787 int rval=0; /* return value (logging flags) */
1788 int state; /* state bits, accessed thru macros */
1789
1790 ifp = XFS_IFORK_PTR(ip, whichfork);
1791 ASSERT(*idx <= ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
1792 state = 0;
1793
1794 if (whichfork == XFS_ATTR_FORK)
1795 state |= BMAP_ATTRFORK;
1796
1797 /*
1798 * Check and set flags if this segment has a left neighbor.
1799 */
1800 if (*idx > 0) {
1801 state |= BMAP_LEFT_VALID;
1802 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
1803 if (isnullstartblock(left.br_startblock))
1804 state |= BMAP_LEFT_DELAY;
1805 }
1806
1807 /*
1808 * Check and set flags if this segment has a current value.
1809 * Not true if we're inserting into the "hole" at eof.
1810 */
1811 if (*idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1812 state |= BMAP_RIGHT_VALID;
1813 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
1814 if (isnullstartblock(right.br_startblock))
1815 state |= BMAP_RIGHT_DELAY;
1816 }
1817
1818 /*
1819 * We're inserting a real allocation between "left" and "right".
1820 * Set the contiguity flags. Don't let extents get too large.
1821 */
1822 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1823 left.br_startoff + left.br_blockcount == new->br_startoff &&
1824 left.br_startblock + left.br_blockcount == new->br_startblock &&
1825 left.br_state == new->br_state &&
1826 left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1827 state |= BMAP_LEFT_CONTIG;
1828
1829 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1830 new->br_startoff + new->br_blockcount == right.br_startoff &&
1831 new->br_startblock + new->br_blockcount == right.br_startblock &&
1832 new->br_state == right.br_state &&
1833 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1834 (!(state & BMAP_LEFT_CONTIG) ||
1835 left.br_blockcount + new->br_blockcount +
1836 right.br_blockcount <= MAXEXTLEN))
1837 state |= BMAP_RIGHT_CONTIG;
1838
1839 error = 0;
1840 /*
1841 * Select which case we're in here, and implement it.
1842 */
1843 switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1844 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1845 /*
1846 * New allocation is contiguous with real allocations on the
1847 * left and on the right.
1848 * Merge all three into a single extent record.
1849 */
1850 --*idx;
1851 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1852 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1853 left.br_blockcount + new->br_blockcount +
1854 right.br_blockcount);
1855 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1856
1857 xfs_iext_remove(ip, *idx + 1, 1, state);
1858
1859 XFS_IFORK_NEXT_SET(ip, whichfork,
1860 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
1861 if (cur == NULL) {
1862 rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1863 } else {
1864 rval = XFS_ILOG_CORE;
1865 if ((error = xfs_bmbt_lookup_eq(cur,
1866 right.br_startoff,
1867 right.br_startblock,
1868 right.br_blockcount, &i)))
1869 goto done;
1870 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1871 if ((error = xfs_btree_delete(cur, &i)))
1872 goto done;
1873 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1874 if ((error = xfs_btree_decrement(cur, 0, &i)))
1875 goto done;
1876 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1877 if ((error = xfs_bmbt_update(cur, left.br_startoff,
1878 left.br_startblock,
1879 left.br_blockcount +
1880 new->br_blockcount +
1881 right.br_blockcount,
1882 left.br_state)))
1883 goto done;
1884 }
1885 break;
1886
1887 case BMAP_LEFT_CONTIG:
1888 /*
1889 * New allocation is contiguous with a real allocation
1890 * on the left.
1891 * Merge the new allocation with the left neighbor.
1892 */
1893 --*idx;
1894 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1895 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1896 left.br_blockcount + new->br_blockcount);
1897 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1898
1899 if (cur == NULL) {
1900 rval = xfs_ilog_fext(whichfork);
1901 } else {
1902 rval = 0;
1903 if ((error = xfs_bmbt_lookup_eq(cur,
1904 left.br_startoff,
1905 left.br_startblock,
1906 left.br_blockcount, &i)))
1907 goto done;
1908 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1909 if ((error = xfs_bmbt_update(cur, left.br_startoff,
1910 left.br_startblock,
1911 left.br_blockcount +
1912 new->br_blockcount,
1913 left.br_state)))
1914 goto done;
1915 }
1916 break;
1917
1918 case BMAP_RIGHT_CONTIG:
1919 /*
1920 * New allocation is contiguous with a real allocation
1921 * on the right.
1922 * Merge the new allocation with the right neighbor.
1923 */
1924 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1925 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1926 new->br_startoff, new->br_startblock,
1927 new->br_blockcount + right.br_blockcount,
1928 right.br_state);
1929 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1930
1931 if (cur == NULL) {
1932 rval = xfs_ilog_fext(whichfork);
1933 } else {
1934 rval = 0;
1935 if ((error = xfs_bmbt_lookup_eq(cur,
1936 right.br_startoff,
1937 right.br_startblock,
1938 right.br_blockcount, &i)))
1939 goto done;
1940 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1941 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1942 new->br_startblock,
1943 new->br_blockcount +
1944 right.br_blockcount,
1945 right.br_state)))
1946 goto done;
1947 }
1948 break;
1949
1950 case 0:
1951 /*
1952 * New allocation is not contiguous with another
1953 * real allocation.
1954 * Insert a new entry.
1955 */
1956 xfs_iext_insert(ip, *idx, 1, new, state);
1957 XFS_IFORK_NEXT_SET(ip, whichfork,
1958 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
1959 if (cur == NULL) {
1960 rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1961 } else {
1962 rval = XFS_ILOG_CORE;
1963 if ((error = xfs_bmbt_lookup_eq(cur,
1964 new->br_startoff,
1965 new->br_startblock,
1966 new->br_blockcount, &i)))
1967 goto done;
1968 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1969 cur->bc_rec.b.br_state = new->br_state;
1970 if ((error = xfs_btree_insert(cur, &i)))
1971 goto done;
1972 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1973 }
1974 break;
1975 }
1976 done:
1977 *logflagsp = rval;
1978 return error;
1979 }
1980
1981 /*
1982 * Adjust the size of the new extent based on di_extsize and rt extsize.
1983 */
1984 STATIC int
1985 xfs_bmap_extsize_align(
1986 xfs_mount_t *mp,
1987 xfs_bmbt_irec_t *gotp, /* next extent pointer */
1988 xfs_bmbt_irec_t *prevp, /* previous extent pointer */
1989 xfs_extlen_t extsz, /* align to this extent size */
1990 int rt, /* is this a realtime inode? */
1991 int eof, /* is extent at end-of-file? */
1992 int delay, /* creating delalloc extent? */
1993 int convert, /* overwriting unwritten extent? */
1994 xfs_fileoff_t *offp, /* in/out: aligned offset */
1995 xfs_extlen_t *lenp) /* in/out: aligned length */
1996 {
1997 xfs_fileoff_t orig_off; /* original offset */
1998 xfs_extlen_t orig_alen; /* original length */
1999 xfs_fileoff_t orig_end; /* original off+len */
2000 xfs_fileoff_t nexto; /* next file offset */
2001 xfs_fileoff_t prevo; /* previous file offset */
2002 xfs_fileoff_t align_off; /* temp for offset */
2003 xfs_extlen_t align_alen; /* temp for length */
2004 xfs_extlen_t temp; /* temp for calculations */
2005
2006 if (convert)
2007 return 0;
2008
2009 orig_off = align_off = *offp;
2010 orig_alen = align_alen = *lenp;
2011 orig_end = orig_off + orig_alen;
2012
2013 /*
2014 * If this request overlaps an existing extent, then don't
2015 * attempt to perform any additional alignment.
2016 */
2017 if (!delay && !eof &&
2018 (orig_off >= gotp->br_startoff) &&
2019 (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2020 return 0;
2021 }
2022
2023 /*
2024 * If the file offset is unaligned vs. the extent size
2025 * we need to align it. This will be possible unless
2026 * the file was previously written with a kernel that didn't
2027 * perform this alignment, or if a truncate shot us in the
2028 * foot.
2029 */
2030 temp = do_mod(orig_off, extsz);
2031 if (temp) {
2032 align_alen += temp;
2033 align_off -= temp;
2034 }
2035 /*
2036 * Same adjustment for the end of the requested area.
2037 */
2038 if ((temp = (align_alen % extsz))) {
2039 align_alen += extsz - temp;
2040 }
2041 /*
2042 * If the previous block overlaps with this proposed allocation
2043 * then move the start forward without adjusting the length.
2044 */
2045 if (prevp->br_startoff != NULLFILEOFF) {
2046 if (prevp->br_startblock == HOLESTARTBLOCK)
2047 prevo = prevp->br_startoff;
2048 else
2049 prevo = prevp->br_startoff + prevp->br_blockcount;
2050 } else
2051 prevo = 0;
2052 if (align_off != orig_off && align_off < prevo)
2053 align_off = prevo;
2054 /*
2055 * If the next block overlaps with this proposed allocation
2056 * then move the start back without adjusting the length,
2057 * but not before offset 0.
2058 * This may of course make the start overlap previous block,
2059 * and if we hit the offset 0 limit then the next block
2060 * can still overlap too.
2061 */
2062 if (!eof && gotp->br_startoff != NULLFILEOFF) {
2063 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
2064 (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
2065 nexto = gotp->br_startoff + gotp->br_blockcount;
2066 else
2067 nexto = gotp->br_startoff;
2068 } else
2069 nexto = NULLFILEOFF;
2070 if (!eof &&
2071 align_off + align_alen != orig_end &&
2072 align_off + align_alen > nexto)
2073 align_off = nexto > align_alen ? nexto - align_alen : 0;
2074 /*
2075 * If we're now overlapping the next or previous extent that
2076 * means we can't fit an extsz piece in this hole. Just move
2077 * the start forward to the first valid spot and set
2078 * the length so we hit the end.
2079 */
2080 if (align_off != orig_off && align_off < prevo)
2081 align_off = prevo;
2082 if (align_off + align_alen != orig_end &&
2083 align_off + align_alen > nexto &&
2084 nexto != NULLFILEOFF) {
2085 ASSERT(nexto > prevo);
2086 align_alen = nexto - align_off;
2087 }
2088
2089 /*
2090 * If realtime, and the result isn't a multiple of the realtime
2091 * extent size we need to remove blocks until it is.
2092 */
2093 if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
2094 /*
2095 * We're not covering the original request, or
2096 * we won't be able to once we fix the length.
2097 */
2098 if (orig_off < align_off ||
2099 orig_end > align_off + align_alen ||
2100 align_alen - temp < orig_alen)
2101 return XFS_ERROR(EINVAL);
2102 /*
2103 * Try to fix it by moving the start up.
2104 */
2105 if (align_off + temp <= orig_off) {
2106 align_alen -= temp;
2107 align_off += temp;
2108 }
2109 /*
2110 * Try to fix it by moving the end in.
2111 */
2112 else if (align_off + align_alen - temp >= orig_end)
2113 align_alen -= temp;
2114 /*
2115 * Set the start to the minimum then trim the length.
2116 */
2117 else {
2118 align_alen -= orig_off - align_off;
2119 align_off = orig_off;
2120 align_alen -= align_alen % mp->m_sb.sb_rextsize;
2121 }
2122 /*
2123 * Result doesn't cover the request, fail it.
2124 */
2125 if (orig_off < align_off || orig_end > align_off + align_alen)
2126 return XFS_ERROR(EINVAL);
2127 } else {
2128 ASSERT(orig_off >= align_off);
2129 ASSERT(orig_end <= align_off + align_alen);
2130 }
2131
2132 #ifdef DEBUG
2133 if (!eof && gotp->br_startoff != NULLFILEOFF)
2134 ASSERT(align_off + align_alen <= gotp->br_startoff);
2135 if (prevp->br_startoff != NULLFILEOFF)
2136 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
2137 #endif
2138
2139 *lenp = align_alen;
2140 *offp = align_off;
2141 return 0;
2142 }
2143
2144 #define XFS_ALLOC_GAP_UNITS 4
2145
2146 STATIC void
2147 xfs_bmap_adjacent(
2148 xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2149 {
2150 xfs_fsblock_t adjust; /* adjustment to block numbers */
2151 xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
2152 xfs_mount_t *mp; /* mount point structure */
2153 int nullfb; /* true if ap->firstblock isn't set */
2154 int rt; /* true if inode is realtime */
2155
2156 #define ISVALID(x,y) \
2157 (rt ? \
2158 (x) < mp->m_sb.sb_rblocks : \
2159 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2160 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2161 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2162
2163 mp = ap->ip->i_mount;
2164 nullfb = ap->firstblock == NULLFSBLOCK;
2165 rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2166 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2167 /*
2168 * If allocating at eof, and there's a previous real block,
2169 * try to use its last block as our starting point.
2170 */
2171 if (ap->eof && ap->prevp->br_startoff != NULLFILEOFF &&
2172 !isnullstartblock(ap->prevp->br_startblock) &&
2173 ISVALID(ap->prevp->br_startblock + ap->prevp->br_blockcount,
2174 ap->prevp->br_startblock)) {
2175 ap->rval = ap->prevp->br_startblock + ap->prevp->br_blockcount;
2176 /*
2177 * Adjust for the gap between prevp and us.
2178 */
2179 adjust = ap->off -
2180 (ap->prevp->br_startoff + ap->prevp->br_blockcount);
2181 if (adjust &&
2182 ISVALID(ap->rval + adjust, ap->prevp->br_startblock))
2183 ap->rval += adjust;
2184 }
2185 /*
2186 * If not at eof, then compare the two neighbor blocks.
2187 * Figure out whether either one gives us a good starting point,
2188 * and pick the better one.
2189 */
2190 else if (!ap->eof) {
2191 xfs_fsblock_t gotbno; /* right side block number */
2192 xfs_fsblock_t gotdiff=0; /* right side difference */
2193 xfs_fsblock_t prevbno; /* left side block number */
2194 xfs_fsblock_t prevdiff=0; /* left side difference */
2195
2196 /*
2197 * If there's a previous (left) block, select a requested
2198 * start block based on it.
2199 */
2200 if (ap->prevp->br_startoff != NULLFILEOFF &&
2201 !isnullstartblock(ap->prevp->br_startblock) &&
2202 (prevbno = ap->prevp->br_startblock +
2203 ap->prevp->br_blockcount) &&
2204 ISVALID(prevbno, ap->prevp->br_startblock)) {
2205 /*
2206 * Calculate gap to end of previous block.
2207 */
2208 adjust = prevdiff = ap->off -
2209 (ap->prevp->br_startoff +
2210 ap->prevp->br_blockcount);
2211 /*
2212 * Figure the startblock based on the previous block's
2213 * end and the gap size.
2214 * Heuristic!
2215 * If the gap is large relative to the piece we're
2216 * allocating, or using it gives us an invalid block
2217 * number, then just use the end of the previous block.
2218 */
2219 if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2220 ISVALID(prevbno + prevdiff,
2221 ap->prevp->br_startblock))
2222 prevbno += adjust;
2223 else
2224 prevdiff += adjust;
2225 /*
2226 * If the firstblock forbids it, can't use it,
2227 * must use default.
2228 */
2229 if (!rt && !nullfb &&
2230 XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2231 prevbno = NULLFSBLOCK;
2232 }
2233 /*
2234 * No previous block or can't follow it, just default.
2235 */
2236 else
2237 prevbno = NULLFSBLOCK;
2238 /*
2239 * If there's a following (right) block, select a requested
2240 * start block based on it.
2241 */
2242 if (!isnullstartblock(ap->gotp->br_startblock)) {
2243 /*
2244 * Calculate gap to start of next block.
2245 */
2246 adjust = gotdiff = ap->gotp->br_startoff - ap->off;
2247 /*
2248 * Figure the startblock based on the next block's
2249 * start and the gap size.
2250 */
2251 gotbno = ap->gotp->br_startblock;
2252 /*
2253 * Heuristic!
2254 * If the gap is large relative to the piece we're
2255 * allocating, or using it gives us an invalid block
2256 * number, then just use the start of the next block
2257 * offset by our length.
2258 */
2259 if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2260 ISVALID(gotbno - gotdiff, gotbno))
2261 gotbno -= adjust;
2262 else if (ISVALID(gotbno - ap->alen, gotbno)) {
2263 gotbno -= ap->alen;
2264 gotdiff += adjust - ap->alen;
2265 } else
2266 gotdiff += adjust;
2267 /*
2268 * If the firstblock forbids it, can't use it,
2269 * must use default.
2270 */
2271 if (!rt && !nullfb &&
2272 XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2273 gotbno = NULLFSBLOCK;
2274 }
2275 /*
2276 * No next block, just default.
2277 */
2278 else
2279 gotbno = NULLFSBLOCK;
2280 /*
2281 * If both valid, pick the better one, else the only good
2282 * one, else ap->rval is already set (to 0 or the inode block).
2283 */
2284 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2285 ap->rval = prevdiff <= gotdiff ? prevbno : gotbno;
2286 else if (prevbno != NULLFSBLOCK)
2287 ap->rval = prevbno;
2288 else if (gotbno != NULLFSBLOCK)
2289 ap->rval = gotbno;
2290 }
2291 #undef ISVALID
2292 }
2293
2294 STATIC int
2295 xfs_bmap_rtalloc(
2296 xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2297 {
2298 xfs_alloctype_t atype = 0; /* type for allocation routines */
2299 int error; /* error return value */
2300 xfs_mount_t *mp; /* mount point structure */
2301 xfs_extlen_t prod = 0; /* product factor for allocators */
2302 xfs_extlen_t ralen = 0; /* realtime allocation length */
2303 xfs_extlen_t align; /* minimum allocation alignment */
2304 xfs_rtblock_t rtb;
2305
2306 mp = ap->ip->i_mount;
2307 align = xfs_get_extsz_hint(ap->ip);
2308 prod = align / mp->m_sb.sb_rextsize;
2309 error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2310 align, 1, ap->eof, 0,
2311 ap->conv, &ap->off, &ap->alen);
2312 if (error)
2313 return error;
2314 ASSERT(ap->alen);
2315 ASSERT(ap->alen % mp->m_sb.sb_rextsize == 0);
2316
2317 /*
2318 * If the offset & length are not perfectly aligned
2319 * then kill prod, it will just get us in trouble.
2320 */
2321 if (do_mod(ap->off, align) || ap->alen % align)
2322 prod = 1;
2323 /*
2324 * Set ralen to be the actual requested length in rtextents.
2325 */
2326 ralen = ap->alen / mp->m_sb.sb_rextsize;
2327 /*
2328 * If the old value was close enough to MAXEXTLEN that
2329 * we rounded up to it, cut it back so it's valid again.
2330 * Note that if it's a really large request (bigger than
2331 * MAXEXTLEN), we don't hear about that number, and can't
2332 * adjust the starting point to match it.
2333 */
2334 if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2335 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2336
2337 /*
2338 * Lock out other modifications to the RT bitmap inode.
2339 */
2340 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
2341 xfs_trans_ijoin_ref(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
2342
2343 /*
2344 * If it's an allocation to an empty file at offset 0,
2345 * pick an extent that will space things out in the rt area.
2346 */
2347 if (ap->eof && ap->off == 0) {
2348 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
2349
2350 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2351 if (error)
2352 return error;
2353 ap->rval = rtx * mp->m_sb.sb_rextsize;
2354 } else {
2355 ap->rval = 0;
2356 }
2357
2358 xfs_bmap_adjacent(ap);
2359
2360 /*
2361 * Realtime allocation, done through xfs_rtallocate_extent.
2362 */
2363 atype = ap->rval == 0 ? XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2364 do_div(ap->rval, mp->m_sb.sb_rextsize);
2365 rtb = ap->rval;
2366 ap->alen = ralen;
2367 if ((error = xfs_rtallocate_extent(ap->tp, ap->rval, 1, ap->alen,
2368 &ralen, atype, ap->wasdel, prod, &rtb)))
2369 return error;
2370 if (rtb == NULLFSBLOCK && prod > 1 &&
2371 (error = xfs_rtallocate_extent(ap->tp, ap->rval, 1,
2372 ap->alen, &ralen, atype,
2373 ap->wasdel, 1, &rtb)))
2374 return error;
2375 ap->rval = rtb;
2376 if (ap->rval != NULLFSBLOCK) {
2377 ap->rval *= mp->m_sb.sb_rextsize;
2378 ralen *= mp->m_sb.sb_rextsize;
2379 ap->alen = ralen;
2380 ap->ip->i_d.di_nblocks += ralen;
2381 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2382 if (ap->wasdel)
2383 ap->ip->i_delayed_blks -= ralen;
2384 /*
2385 * Adjust the disk quota also. This was reserved
2386 * earlier.
2387 */
2388 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2389 ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2390 XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2391 } else {
2392 ap->alen = 0;
2393 }
2394 return 0;
2395 }
2396
2397 STATIC int
2398 xfs_bmap_btalloc_nullfb(
2399 struct xfs_bmalloca *ap,
2400 struct xfs_alloc_arg *args,
2401 xfs_extlen_t *blen)
2402 {
2403 struct xfs_mount *mp = ap->ip->i_mount;
2404 struct xfs_perag *pag;
2405 xfs_agnumber_t ag, startag;
2406 int notinit = 0;
2407 int error;
2408
2409 if (ap->userdata && xfs_inode_is_filestream(ap->ip))
2410 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2411 else
2412 args->type = XFS_ALLOCTYPE_START_BNO;
2413 args->total = ap->total;
2414
2415 /*
2416 * Search for an allocation group with a single extent large enough
2417 * for the request. If one isn't found, then adjust the minimum
2418 * allocation size to the largest space found.
2419 */
2420 startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
2421 if (startag == NULLAGNUMBER)
2422 startag = ag = 0;
2423
2424 pag = xfs_perag_get(mp, ag);
2425 while (*blen < args->maxlen) {
2426 if (!pag->pagf_init) {
2427 error = xfs_alloc_pagf_init(mp, args->tp, ag,
2428 XFS_ALLOC_FLAG_TRYLOCK);
2429 if (error) {
2430 xfs_perag_put(pag);
2431 return error;
2432 }
2433 }
2434
2435 /*
2436 * See xfs_alloc_fix_freelist...
2437 */
2438 if (pag->pagf_init) {
2439 xfs_extlen_t longest;
2440 longest = xfs_alloc_longest_free_extent(mp, pag);
2441 if (*blen < longest)
2442 *blen = longest;
2443 } else
2444 notinit = 1;
2445
2446 if (xfs_inode_is_filestream(ap->ip)) {
2447 if (*blen >= args->maxlen)
2448 break;
2449
2450 if (ap->userdata) {
2451 /*
2452 * If startag is an invalid AG, we've
2453 * come here once before and
2454 * xfs_filestream_new_ag picked the
2455 * best currently available.
2456 *
2457 * Don't continue looping, since we
2458 * could loop forever.
2459 */
2460 if (startag == NULLAGNUMBER)
2461 break;
2462
2463 error = xfs_filestream_new_ag(ap, &ag);
2464 xfs_perag_put(pag);
2465 if (error)
2466 return error;
2467
2468 /* loop again to set 'blen'*/
2469 startag = NULLAGNUMBER;
2470 pag = xfs_perag_get(mp, ag);
2471 continue;
2472 }
2473 }
2474 if (++ag == mp->m_sb.sb_agcount)
2475 ag = 0;
2476 if (ag == startag)
2477 break;
2478 xfs_perag_put(pag);
2479 pag = xfs_perag_get(mp, ag);
2480 }
2481 xfs_perag_put(pag);
2482
2483 /*
2484 * Since the above loop did a BUF_TRYLOCK, it is
2485 * possible that there is space for this request.
2486 */
2487 if (notinit || *blen < ap->minlen)
2488 args->minlen = ap->minlen;
2489 /*
2490 * If the best seen length is less than the request
2491 * length, use the best as the minimum.
2492 */
2493 else if (*blen < args->maxlen)
2494 args->minlen = *blen;
2495 /*
2496 * Otherwise we've seen an extent as big as maxlen,
2497 * use that as the minimum.
2498 */
2499 else
2500 args->minlen = args->maxlen;
2501
2502 /*
2503 * set the failure fallback case to look in the selected
2504 * AG as the stream may have moved.
2505 */
2506 if (xfs_inode_is_filestream(ap->ip))
2507 ap->rval = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
2508
2509 return 0;
2510 }
2511
2512 STATIC int
2513 xfs_bmap_btalloc(
2514 xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2515 {
2516 xfs_mount_t *mp; /* mount point structure */
2517 xfs_alloctype_t atype = 0; /* type for allocation routines */
2518 xfs_extlen_t align; /* minimum allocation alignment */
2519 xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
2520 xfs_agnumber_t ag;
2521 xfs_alloc_arg_t args;
2522 xfs_extlen_t blen;
2523 xfs_extlen_t nextminlen = 0;
2524 int nullfb; /* true if ap->firstblock isn't set */
2525 int isaligned;
2526 int tryagain;
2527 int error;
2528
2529 mp = ap->ip->i_mount;
2530 align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
2531 if (unlikely(align)) {
2532 error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2533 align, 0, ap->eof, 0, ap->conv,
2534 &ap->off, &ap->alen);
2535 ASSERT(!error);
2536 ASSERT(ap->alen);
2537 }
2538 nullfb = ap->firstblock == NULLFSBLOCK;
2539 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2540 if (nullfb) {
2541 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
2542 ag = xfs_filestream_lookup_ag(ap->ip);
2543 ag = (ag != NULLAGNUMBER) ? ag : 0;
2544 ap->rval = XFS_AGB_TO_FSB(mp, ag, 0);
2545 } else {
2546 ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2547 }
2548 } else
2549 ap->rval = ap->firstblock;
2550
2551 xfs_bmap_adjacent(ap);
2552
2553 /*
2554 * If allowed, use ap->rval; otherwise must use firstblock since
2555 * it's in the right allocation group.
2556 */
2557 if (nullfb || XFS_FSB_TO_AGNO(mp, ap->rval) == fb_agno)
2558 ;
2559 else
2560 ap->rval = ap->firstblock;
2561 /*
2562 * Normal allocation, done through xfs_alloc_vextent.
2563 */
2564 tryagain = isaligned = 0;
2565 args.tp = ap->tp;
2566 args.mp = mp;
2567 args.fsbno = ap->rval;
2568
2569 /* Trim the allocation back to the maximum an AG can fit. */
2570 args.maxlen = MIN(ap->alen, XFS_ALLOC_AG_MAX_USABLE(mp));
2571 args.firstblock = ap->firstblock;
2572 blen = 0;
2573 if (nullfb) {
2574 error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
2575 if (error)
2576 return error;
2577 } else if (ap->low) {
2578 if (xfs_inode_is_filestream(ap->ip))
2579 args.type = XFS_ALLOCTYPE_FIRST_AG;
2580 else
2581 args.type = XFS_ALLOCTYPE_START_BNO;
2582 args.total = args.minlen = ap->minlen;
2583 } else {
2584 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2585 args.total = ap->total;
2586 args.minlen = ap->minlen;
2587 }
2588 /* apply extent size hints if obtained earlier */
2589 if (unlikely(align)) {
2590 args.prod = align;
2591 if ((args.mod = (xfs_extlen_t)do_mod(ap->off, args.prod)))
2592 args.mod = (xfs_extlen_t)(args.prod - args.mod);
2593 } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
2594 args.prod = 1;
2595 args.mod = 0;
2596 } else {
2597 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
2598 if ((args.mod = (xfs_extlen_t)(do_mod(ap->off, args.prod))))
2599 args.mod = (xfs_extlen_t)(args.prod - args.mod);
2600 }
2601 /*
2602 * If we are not low on available data blocks, and the
2603 * underlying logical volume manager is a stripe, and
2604 * the file offset is zero then try to allocate data
2605 * blocks on stripe unit boundary.
2606 * NOTE: ap->aeof is only set if the allocation length
2607 * is >= the stripe unit and the allocation offset is
2608 * at the end of file.
2609 */
2610 if (!ap->low && ap->aeof) {
2611 if (!ap->off) {
2612 args.alignment = mp->m_dalign;
2613 atype = args.type;
2614 isaligned = 1;
2615 /*
2616 * Adjust for alignment
2617 */
2618 if (blen > args.alignment && blen <= args.maxlen)
2619 args.minlen = blen - args.alignment;
2620 args.minalignslop = 0;
2621 } else {
2622 /*
2623 * First try an exact bno allocation.
2624 * If it fails then do a near or start bno
2625 * allocation with alignment turned on.
2626 */
2627 atype = args.type;
2628 tryagain = 1;
2629 args.type = XFS_ALLOCTYPE_THIS_BNO;
2630 args.alignment = 1;
2631 /*
2632 * Compute the minlen+alignment for the
2633 * next case. Set slop so that the value
2634 * of minlen+alignment+slop doesn't go up
2635 * between the calls.
2636 */
2637 if (blen > mp->m_dalign && blen <= args.maxlen)
2638 nextminlen = blen - mp->m_dalign;
2639 else
2640 nextminlen = args.minlen;
2641 if (nextminlen + mp->m_dalign > args.minlen + 1)
2642 args.minalignslop =
2643 nextminlen + mp->m_dalign -
2644 args.minlen - 1;
2645 else
2646 args.minalignslop = 0;
2647 }
2648 } else {
2649 args.alignment = 1;
2650 args.minalignslop = 0;
2651 }
2652 args.minleft = ap->minleft;
2653 args.wasdel = ap->wasdel;
2654 args.isfl = 0;
2655 args.userdata = ap->userdata;
2656 if ((error = xfs_alloc_vextent(&args)))
2657 return error;
2658 if (tryagain && args.fsbno == NULLFSBLOCK) {
2659 /*
2660 * Exact allocation failed. Now try with alignment
2661 * turned on.
2662 */
2663 args.type = atype;
2664 args.fsbno = ap->rval;
2665 args.alignment = mp->m_dalign;
2666 args.minlen = nextminlen;
2667 args.minalignslop = 0;
2668 isaligned = 1;
2669 if ((error = xfs_alloc_vextent(&args)))
2670 return error;
2671 }
2672 if (isaligned && args.fsbno == NULLFSBLOCK) {
2673 /*
2674 * allocation failed, so turn off alignment and
2675 * try again.
2676 */
2677 args.type = atype;
2678 args.fsbno = ap->rval;
2679 args.alignment = 0;
2680 if ((error = xfs_alloc_vextent(&args)))
2681 return error;
2682 }
2683 if (args.fsbno == NULLFSBLOCK && nullfb &&
2684 args.minlen > ap->minlen) {
2685 args.minlen = ap->minlen;
2686 args.type = XFS_ALLOCTYPE_START_BNO;
2687 args.fsbno = ap->rval;
2688 if ((error = xfs_alloc_vextent(&args)))
2689 return error;
2690 }
2691 if (args.fsbno == NULLFSBLOCK && nullfb) {
2692 args.fsbno = 0;
2693 args.type = XFS_ALLOCTYPE_FIRST_AG;
2694 args.total = ap->minlen;
2695 args.minleft = 0;
2696 if ((error = xfs_alloc_vextent(&args)))
2697 return error;
2698 ap->low = 1;
2699 }
2700 if (args.fsbno != NULLFSBLOCK) {
2701 ap->firstblock = ap->rval = args.fsbno;
2702 ASSERT(nullfb || fb_agno == args.agno ||
2703 (ap->low && fb_agno < args.agno));
2704 ap->alen = args.len;
2705 ap->ip->i_d.di_nblocks += args.len;
2706 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2707 if (ap->wasdel)
2708 ap->ip->i_delayed_blks -= args.len;
2709 /*
2710 * Adjust the disk quota also. This was reserved
2711 * earlier.
2712 */
2713 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2714 ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2715 XFS_TRANS_DQ_BCOUNT,
2716 (long) args.len);
2717 } else {
2718 ap->rval = NULLFSBLOCK;
2719 ap->alen = 0;
2720 }
2721 return 0;
2722 }
2723
2724 /*
2725 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2726 * It figures out where to ask the underlying allocator to put the new extent.
2727 */
2728 STATIC int
2729 xfs_bmap_alloc(
2730 xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2731 {
2732 if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
2733 return xfs_bmap_rtalloc(ap);
2734 return xfs_bmap_btalloc(ap);
2735 }
2736
2737 /*
2738 * Transform a btree format file with only one leaf node, where the
2739 * extents list will fit in the inode, into an extents format file.
2740 * Since the file extents are already in-core, all we have to do is
2741 * give up the space for the btree root and pitch the leaf block.
2742 */
2743 STATIC int /* error */
2744 xfs_bmap_btree_to_extents(
2745 xfs_trans_t *tp, /* transaction pointer */
2746 xfs_inode_t *ip, /* incore inode pointer */
2747 xfs_btree_cur_t *cur, /* btree cursor */
2748 int *logflagsp, /* inode logging flags */
2749 int whichfork) /* data or attr fork */
2750 {
2751 /* REFERENCED */
2752 struct xfs_btree_block *cblock;/* child btree block */
2753 xfs_fsblock_t cbno; /* child block number */
2754 xfs_buf_t *cbp; /* child block's buffer */
2755 int error; /* error return value */
2756 xfs_ifork_t *ifp; /* inode fork data */
2757 xfs_mount_t *mp; /* mount point structure */
2758 __be64 *pp; /* ptr to block address */
2759 struct xfs_btree_block *rblock;/* root btree block */
2760
2761 mp = ip->i_mount;
2762 ifp = XFS_IFORK_PTR(ip, whichfork);
2763 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2764 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
2765 rblock = ifp->if_broot;
2766 ASSERT(be16_to_cpu(rblock->bb_level) == 1);
2767 ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
2768 ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
2769 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
2770 cbno = be64_to_cpu(*pp);
2771 *logflagsp = 0;
2772 #ifdef DEBUG
2773 if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
2774 return error;
2775 #endif
2776 if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
2777 XFS_BMAP_BTREE_REF)))
2778 return error;
2779 cblock = XFS_BUF_TO_BLOCK(cbp);
2780 if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
2781 return error;
2782 xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
2783 ip->i_d.di_nblocks--;
2784 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
2785 xfs_trans_binval(tp, cbp);
2786 if (cur->bc_bufs[0] == cbp)
2787 cur->bc_bufs[0] = NULL;
2788 xfs_iroot_realloc(ip, -1, whichfork);
2789 ASSERT(ifp->if_broot == NULL);
2790 ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
2791 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
2792 *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2793 return 0;
2794 }
2795
2796 /*
2797 * Called by xfs_bmapi to update file extent records and the btree
2798 * after removing space (or undoing a delayed allocation).
2799 */
2800 STATIC int /* error */
2801 xfs_bmap_del_extent(
2802 xfs_inode_t *ip, /* incore inode pointer */
2803 xfs_trans_t *tp, /* current transaction pointer */
2804 xfs_extnum_t *idx, /* extent number to update/delete */
2805 xfs_bmap_free_t *flist, /* list of extents to be freed */
2806 xfs_btree_cur_t *cur, /* if null, not a btree */
2807 xfs_bmbt_irec_t *del, /* data to remove from extents */
2808 int *logflagsp, /* inode logging flags */
2809 int whichfork) /* data or attr fork */
2810 {
2811 xfs_filblks_t da_new; /* new delay-alloc indirect blocks */
2812 xfs_filblks_t da_old; /* old delay-alloc indirect blocks */
2813 xfs_fsblock_t del_endblock=0; /* first block past del */
2814 xfs_fileoff_t del_endoff; /* first offset past del */
2815 int delay; /* current block is delayed allocated */
2816 int do_fx; /* free extent at end of routine */
2817 xfs_bmbt_rec_host_t *ep; /* current extent entry pointer */
2818 int error; /* error return value */
2819 int flags; /* inode logging flags */
2820 xfs_bmbt_irec_t got; /* current extent entry */
2821 xfs_fileoff_t got_endoff; /* first offset past got */
2822 int i; /* temp state */
2823 xfs_ifork_t *ifp; /* inode fork pointer */
2824 xfs_mount_t *mp; /* mount structure */
2825 xfs_filblks_t nblks; /* quota/sb block count */
2826 xfs_bmbt_irec_t new; /* new record to be inserted */
2827 /* REFERENCED */
2828 uint qfield; /* quota field to update */
2829 xfs_filblks_t temp; /* for indirect length calculations */
2830 xfs_filblks_t temp2; /* for indirect length calculations */
2831 int state = 0;
2832
2833 XFS_STATS_INC(xs_del_exlist);
2834
2835 if (whichfork == XFS_ATTR_FORK)
2836 state |= BMAP_ATTRFORK;
2837
2838 mp = ip->i_mount;
2839 ifp = XFS_IFORK_PTR(ip, whichfork);
2840 ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
2841 (uint)sizeof(xfs_bmbt_rec_t)));
2842 ASSERT(del->br_blockcount > 0);
2843 ep = xfs_iext_get_ext(ifp, *idx);
2844 xfs_bmbt_get_all(ep, &got);
2845 ASSERT(got.br_startoff <= del->br_startoff);
2846 del_endoff = del->br_startoff + del->br_blockcount;
2847 got_endoff = got.br_startoff + got.br_blockcount;
2848 ASSERT(got_endoff >= del_endoff);
2849 delay = isnullstartblock(got.br_startblock);
2850 ASSERT(isnullstartblock(del->br_startblock) == delay);
2851 flags = 0;
2852 qfield = 0;
2853 error = 0;
2854 /*
2855 * If deleting a real allocation, must free up the disk space.
2856 */
2857 if (!delay) {
2858 flags = XFS_ILOG_CORE;
2859 /*
2860 * Realtime allocation. Free it and record di_nblocks update.
2861 */
2862 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
2863 xfs_fsblock_t bno;
2864 xfs_filblks_t len;
2865
2866 ASSERT(do_mod(del->br_blockcount,
2867 mp->m_sb.sb_rextsize) == 0);
2868 ASSERT(do_mod(del->br_startblock,
2869 mp->m_sb.sb_rextsize) == 0);
2870 bno = del->br_startblock;
2871 len = del->br_blockcount;
2872 do_div(bno, mp->m_sb.sb_rextsize);
2873 do_div(len, mp->m_sb.sb_rextsize);
2874 if ((error = xfs_rtfree_extent(ip->i_transp, bno,
2875 (xfs_extlen_t)len)))
2876 goto done;
2877 do_fx = 0;
2878 nblks = len * mp->m_sb.sb_rextsize;
2879 qfield = XFS_TRANS_DQ_RTBCOUNT;
2880 }
2881 /*
2882 * Ordinary allocation.
2883 */
2884 else {
2885 do_fx = 1;
2886 nblks = del->br_blockcount;
2887 qfield = XFS_TRANS_DQ_BCOUNT;
2888 }
2889 /*
2890 * Set up del_endblock and cur for later.
2891 */
2892 del_endblock = del->br_startblock + del->br_blockcount;
2893 if (cur) {
2894 if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
2895 got.br_startblock, got.br_blockcount,
2896 &i)))
2897 goto done;
2898 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2899 }
2900 da_old = da_new = 0;
2901 } else {
2902 da_old = startblockval(got.br_startblock);
2903 da_new = 0;
2904 nblks = 0;
2905 do_fx = 0;
2906 }
2907 /*
2908 * Set flag value to use in switch statement.
2909 * Left-contig is 2, right-contig is 1.
2910 */
2911 switch (((got.br_startoff == del->br_startoff) << 1) |
2912 (got_endoff == del_endoff)) {
2913 case 3:
2914 /*
2915 * Matches the whole extent. Delete the entry.
2916 */
2917 xfs_iext_remove(ip, *idx, 1,
2918 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
2919 --*idx;
2920 if (delay)
2921 break;
2922
2923 XFS_IFORK_NEXT_SET(ip, whichfork,
2924 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2925 flags |= XFS_ILOG_CORE;
2926 if (!cur) {
2927 flags |= xfs_ilog_fext(whichfork);
2928 break;
2929 }
2930 if ((error = xfs_btree_delete(cur, &i)))
2931 goto done;
2932 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2933 break;
2934
2935 case 2:
2936 /*
2937 * Deleting the first part of the extent.
2938 */
2939 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2940 xfs_bmbt_set_startoff(ep, del_endoff);
2941 temp = got.br_blockcount - del->br_blockcount;
2942 xfs_bmbt_set_blockcount(ep, temp);
2943 if (delay) {
2944 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2945 da_old);
2946 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2947 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2948 da_new = temp;
2949 break;
2950 }
2951 xfs_bmbt_set_startblock(ep, del_endblock);
2952 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2953 if (!cur) {
2954 flags |= xfs_ilog_fext(whichfork);
2955 break;
2956 }
2957 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
2958 got.br_blockcount - del->br_blockcount,
2959 got.br_state)))
2960 goto done;
2961 break;
2962
2963 case 1:
2964 /*
2965 * Deleting the last part of the extent.
2966 */
2967 temp = got.br_blockcount - del->br_blockcount;
2968 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2969 xfs_bmbt_set_blockcount(ep, temp);
2970 if (delay) {
2971 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2972 da_old);
2973 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2974 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2975 da_new = temp;
2976 break;
2977 }
2978 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2979 if (!cur) {
2980 flags |= xfs_ilog_fext(whichfork);
2981 break;
2982 }
2983 if ((error = xfs_bmbt_update(cur, got.br_startoff,
2984 got.br_startblock,
2985 got.br_blockcount - del->br_blockcount,
2986 got.br_state)))
2987 goto done;
2988 break;
2989
2990 case 0:
2991 /*
2992 * Deleting the middle of the extent.
2993 */
2994 temp = del->br_startoff - got.br_startoff;
2995 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2996 xfs_bmbt_set_blockcount(ep, temp);
2997 new.br_startoff = del_endoff;
2998 temp2 = got_endoff - del_endoff;
2999 new.br_blockcount = temp2;
3000 new.br_state = got.br_state;
3001 if (!delay) {
3002 new.br_startblock = del_endblock;
3003 flags |= XFS_ILOG_CORE;
3004 if (cur) {
3005 if ((error = xfs_bmbt_update(cur,
3006 got.br_startoff,
3007 got.br_startblock, temp,
3008 got.br_state)))
3009 goto done;
3010 if ((error = xfs_btree_increment(cur, 0, &i)))
3011 goto done;
3012 cur->bc_rec.b = new;
3013 error = xfs_btree_insert(cur, &i);
3014 if (error && error != ENOSPC)
3015 goto done;
3016 /*
3017 * If get no-space back from btree insert,
3018 * it tried a split, and we have a zero
3019 * block reservation.
3020 * Fix up our state and return the error.
3021 */
3022 if (error == ENOSPC) {
3023 /*
3024 * Reset the cursor, don't trust
3025 * it after any insert operation.
3026 */
3027 if ((error = xfs_bmbt_lookup_eq(cur,
3028 got.br_startoff,
3029 got.br_startblock,
3030 temp, &i)))
3031 goto done;
3032 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3033 /*
3034 * Update the btree record back
3035 * to the original value.
3036 */
3037 if ((error = xfs_bmbt_update(cur,
3038 got.br_startoff,
3039 got.br_startblock,
3040 got.br_blockcount,
3041 got.br_state)))
3042 goto done;
3043 /*
3044 * Reset the extent record back
3045 * to the original value.
3046 */
3047 xfs_bmbt_set_blockcount(ep,
3048 got.br_blockcount);
3049 flags = 0;
3050 error = XFS_ERROR(ENOSPC);
3051 goto done;
3052 }
3053 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3054 } else
3055 flags |= xfs_ilog_fext(whichfork);
3056 XFS_IFORK_NEXT_SET(ip, whichfork,
3057 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
3058 } else {
3059 ASSERT(whichfork == XFS_DATA_FORK);
3060 temp = xfs_bmap_worst_indlen(ip, temp);
3061 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
3062 temp2 = xfs_bmap_worst_indlen(ip, temp2);
3063 new.br_startblock = nullstartblock((int)temp2);
3064 da_new = temp + temp2;
3065 while (da_new > da_old) {
3066 if (temp) {
3067 temp--;
3068 da_new--;
3069 xfs_bmbt_set_startblock(ep,
3070 nullstartblock((int)temp));
3071 }
3072 if (da_new == da_old)
3073 break;
3074 if (temp2) {
3075 temp2--;
3076 da_new--;
3077 new.br_startblock =
3078 nullstartblock((int)temp2);
3079 }
3080 }
3081 }
3082 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
3083 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
3084 ++*idx;
3085 break;
3086 }
3087 /*
3088 * If we need to, add to list of extents to delete.
3089 */
3090 if (do_fx)
3091 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
3092 mp);
3093 /*
3094 * Adjust inode # blocks in the file.
3095 */
3096 if (nblks)
3097 ip->i_d.di_nblocks -= nblks;
3098 /*
3099 * Adjust quota data.
3100 */
3101 if (qfield)
3102 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
3103
3104 /*
3105 * Account for change in delayed indirect blocks.
3106 * Nothing to do for disk quota accounting here.
3107 */
3108 ASSERT(da_old >= da_new);
3109 if (da_old > da_new) {
3110 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
3111 (int64_t)(da_old - da_new), 0);
3112 }
3113 done:
3114 *logflagsp = flags;
3115 return error;
3116 }
3117
3118 /*
3119 * Remove the entry "free" from the free item list. Prev points to the
3120 * previous entry, unless "free" is the head of the list.
3121 */
3122 STATIC void
3123 xfs_bmap_del_free(
3124 xfs_bmap_free_t *flist, /* free item list header */
3125 xfs_bmap_free_item_t *prev, /* previous item on list, if any */
3126 xfs_bmap_free_item_t *free) /* list item to be freed */
3127 {
3128 if (prev)
3129 prev->xbfi_next = free->xbfi_next;
3130 else
3131 flist->xbf_first = free->xbfi_next;
3132 flist->xbf_count--;
3133 kmem_zone_free(xfs_bmap_free_item_zone, free);
3134 }
3135
3136 /*
3137 * Convert an extents-format file into a btree-format file.
3138 * The new file will have a root block (in the inode) and a single child block.
3139 */
3140 STATIC int /* error */
3141 xfs_bmap_extents_to_btree(
3142 xfs_trans_t *tp, /* transaction pointer */
3143 xfs_inode_t *ip, /* incore inode pointer */
3144 xfs_fsblock_t *firstblock, /* first-block-allocated */
3145 xfs_bmap_free_t *flist, /* blocks freed in xaction */
3146 xfs_btree_cur_t **curp, /* cursor returned to caller */
3147 int wasdel, /* converting a delayed alloc */
3148 int *logflagsp, /* inode logging flags */
3149 int whichfork) /* data or attr fork */
3150 {
3151 struct xfs_btree_block *ablock; /* allocated (child) bt block */
3152 xfs_buf_t *abp; /* buffer for ablock */
3153 xfs_alloc_arg_t args; /* allocation arguments */
3154 xfs_bmbt_rec_t *arp; /* child record pointer */
3155 struct xfs_btree_block *block; /* btree root block */
3156 xfs_btree_cur_t *cur; /* bmap btree cursor */
3157 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
3158 int error; /* error return value */
3159 xfs_extnum_t i, cnt; /* extent record index */
3160 xfs_ifork_t *ifp; /* inode fork pointer */
3161 xfs_bmbt_key_t *kp; /* root block key pointer */
3162 xfs_mount_t *mp; /* mount structure */
3163 xfs_extnum_t nextents; /* number of file extents */
3164 xfs_bmbt_ptr_t *pp; /* root block address pointer */
3165
3166 ifp = XFS_IFORK_PTR(ip, whichfork);
3167 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3168 ASSERT(ifp->if_ext_max ==
3169 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
3170 /*
3171 * Make space in the inode incore.
3172 */
3173 xfs_iroot_realloc(ip, 1, whichfork);
3174 ifp->if_flags |= XFS_IFBROOT;
3175
3176 /*
3177 * Fill in the root.
3178 */
3179 block = ifp->if_broot;
3180 block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3181 block->bb_level = cpu_to_be16(1);
3182 block->bb_numrecs = cpu_to_be16(1);
3183 block->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3184 block->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3185
3186 /*
3187 * Need a cursor. Can't allocate until bb_level is filled in.
3188 */
3189 mp = ip->i_mount;
3190 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
3191 cur->bc_private.b.firstblock = *firstblock;
3192 cur->bc_private.b.flist = flist;
3193 cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3194 /*
3195 * Convert to a btree with two levels, one record in root.
3196 */
3197 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3198 args.tp = tp;
3199 args.mp = mp;
3200 args.firstblock = *firstblock;
3201 if (*firstblock == NULLFSBLOCK) {
3202 args.type = XFS_ALLOCTYPE_START_BNO;
3203 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3204 } else if (flist->xbf_low) {
3205 args.type = XFS_ALLOCTYPE_START_BNO;
3206 args.fsbno = *firstblock;
3207 } else {
3208 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3209 args.fsbno = *firstblock;
3210 }
3211 args.minlen = args.maxlen = args.prod = 1;
3212 args.total = args.minleft = args.alignment = args.mod = args.isfl =
3213 args.minalignslop = 0;
3214 args.wasdel = wasdel;
3215 *logflagsp = 0;
3216 if ((error = xfs_alloc_vextent(&args))) {
3217 xfs_iroot_realloc(ip, -1, whichfork);
3218 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3219 return error;
3220 }
3221 /*
3222 * Allocation can't fail, the space was reserved.
3223 */
3224 ASSERT(args.fsbno != NULLFSBLOCK);
3225 ASSERT(*firstblock == NULLFSBLOCK ||
3226 args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3227 (flist->xbf_low &&
3228 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3229 *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3230 cur->bc_private.b.allocated++;
3231 ip->i_d.di_nblocks++;
3232 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3233 abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3234 /*
3235 * Fill in the child block.
3236 */
3237 ablock = XFS_BUF_TO_BLOCK(abp);
3238 ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3239 ablock->bb_level = 0;
3240 ablock->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3241 ablock->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3242 arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3243 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3244 for (cnt = i = 0; i < nextents; i++) {
3245 ep = xfs_iext_get_ext(ifp, i);
3246 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
3247 arp->l0 = cpu_to_be64(ep->l0);
3248 arp->l1 = cpu_to_be64(ep->l1);
3249 arp++; cnt++;
3250 }
3251 }
3252 ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3253 xfs_btree_set_numrecs(ablock, cnt);
3254
3255 /*
3256 * Fill in the root key and pointer.
3257 */
3258 kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
3259 arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3260 kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
3261 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
3262 be16_to_cpu(block->bb_level)));
3263 *pp = cpu_to_be64(args.fsbno);
3264
3265 /*
3266 * Do all this logging at the end so that
3267 * the root is at the right level.
3268 */
3269 xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
3270 xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3271 ASSERT(*curp == NULL);
3272 *curp = cur;
3273 *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
3274 return 0;
3275 }
3276
3277 /*
3278 * Calculate the default attribute fork offset for newly created inodes.
3279 */
3280 uint
3281 xfs_default_attroffset(
3282 struct xfs_inode *ip)
3283 {
3284 struct xfs_mount *mp = ip->i_mount;
3285 uint offset;
3286
3287 if (mp->m_sb.sb_inodesize == 256) {
3288 offset = XFS_LITINO(mp) -
3289 XFS_BMDR_SPACE_CALC(MINABTPTRS);
3290 } else {
3291 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
3292 }
3293
3294 ASSERT(offset < XFS_LITINO(mp));
3295 return offset;
3296 }
3297
3298 /*
3299 * Helper routine to reset inode di_forkoff field when switching
3300 * attribute fork from local to extent format - we reset it where
3301 * possible to make space available for inline data fork extents.
3302 */
3303 STATIC void
3304 xfs_bmap_forkoff_reset(
3305 xfs_mount_t *mp,
3306 xfs_inode_t *ip,
3307 int whichfork)
3308 {
3309 if (whichfork == XFS_ATTR_FORK &&
3310 ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
3311 ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
3312 ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
3313 uint dfl_forkoff = xfs_default_attroffset(ip) >> 3;
3314
3315 if (dfl_forkoff > ip->i_d.di_forkoff) {
3316 ip->i_d.di_forkoff = dfl_forkoff;
3317 ip->i_df.if_ext_max =
3318 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t);
3319 ip->i_afp->if_ext_max =
3320 XFS_IFORK_ASIZE(ip) / sizeof(xfs_bmbt_rec_t);
3321 }
3322 }
3323 }
3324
3325 /*
3326 * Convert a local file to an extents file.
3327 * This code is out of bounds for data forks of regular files,
3328 * since the file data needs to get logged so things will stay consistent.
3329 * (The bmap-level manipulations are ok, though).
3330 */
3331 STATIC int /* error */
3332 xfs_bmap_local_to_extents(
3333 xfs_trans_t *tp, /* transaction pointer */
3334 xfs_inode_t *ip, /* incore inode pointer */
3335 xfs_fsblock_t *firstblock, /* first block allocated in xaction */
3336 xfs_extlen_t total, /* total blocks needed by transaction */
3337 int *logflagsp, /* inode logging flags */
3338 int whichfork) /* data or attr fork */
3339 {
3340 int error; /* error return value */
3341 int flags; /* logging flags returned */
3342 xfs_ifork_t *ifp; /* inode fork pointer */
3343
3344 /*
3345 * We don't want to deal with the case of keeping inode data inline yet.
3346 * So sending the data fork of a regular inode is invalid.
3347 */
3348 ASSERT(!((ip->i_d.di_mode & S_IFMT) == S_IFREG &&
3349 whichfork == XFS_DATA_FORK));
3350 ifp = XFS_IFORK_PTR(ip, whichfork);
3351 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3352 flags = 0;
3353 error = 0;
3354 if (ifp->if_bytes) {
3355 xfs_alloc_arg_t args; /* allocation arguments */
3356 xfs_buf_t *bp; /* buffer for extent block */
3357 xfs_bmbt_rec_host_t *ep;/* extent record pointer */
3358
3359 args.tp = tp;
3360 args.mp = ip->i_mount;
3361 args.firstblock = *firstblock;
3362 ASSERT((ifp->if_flags &
3363 (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == XFS_IFINLINE);
3364 /*
3365 * Allocate a block. We know we need only one, since the
3366 * file currently fits in an inode.
3367 */
3368 if (*firstblock == NULLFSBLOCK) {
3369 args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3370 args.type = XFS_ALLOCTYPE_START_BNO;
3371 } else {
3372 args.fsbno = *firstblock;
3373 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3374 }
3375 args.total = total;
3376 args.mod = args.minleft = args.alignment = args.wasdel =
3377 args.isfl = args.minalignslop = 0;
3378 args.minlen = args.maxlen = args.prod = 1;
3379 if ((error = xfs_alloc_vextent(&args)))
3380 goto done;
3381 /*
3382 * Can't fail, the space was reserved.
3383 */
3384 ASSERT(args.fsbno != NULLFSBLOCK);
3385 ASSERT(args.len == 1);
3386 *firstblock = args.fsbno;
3387 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3388 memcpy((char *)XFS_BUF_PTR(bp), ifp->if_u1.if_data,
3389 ifp->if_bytes);
3390 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3391 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3392 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3393 xfs_iext_add(ifp, 0, 1);
3394 ep = xfs_iext_get_ext(ifp, 0);
3395 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3396 trace_xfs_bmap_post_update(ip, 0,
3397 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
3398 _THIS_IP_);
3399 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3400 ip->i_d.di_nblocks = 1;
3401 xfs_trans_mod_dquot_byino(tp, ip,
3402 XFS_TRANS_DQ_BCOUNT, 1L);
3403 flags |= xfs_ilog_fext(whichfork);
3404 } else {
3405 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3406 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3407 }
3408 ifp->if_flags &= ~XFS_IFINLINE;
3409 ifp->if_flags |= XFS_IFEXTENTS;
3410 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3411 flags |= XFS_ILOG_CORE;
3412 done:
3413 *logflagsp = flags;
3414 return error;
3415 }
3416
3417 /*
3418 * Search the extent records for the entry containing block bno.
3419 * If bno lies in a hole, point to the next entry. If bno lies
3420 * past eof, *eofp will be set, and *prevp will contain the last
3421 * entry (null if none). Else, *lastxp will be set to the index
3422 * of the found entry; *gotp will contain the entry.
3423 */
3424 STATIC xfs_bmbt_rec_host_t * /* pointer to found extent entry */
3425 xfs_bmap_search_multi_extents(
3426 xfs_ifork_t *ifp, /* inode fork pointer */
3427 xfs_fileoff_t bno, /* block number searched for */
3428 int *eofp, /* out: end of file found */
3429 xfs_extnum_t *lastxp, /* out: last extent index */
3430 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
3431 xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
3432 {
3433 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
3434 xfs_extnum_t lastx; /* last extent index */
3435
3436 /*
3437 * Initialize the extent entry structure to catch access to
3438 * uninitialized br_startblock field.
3439 */
3440 gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
3441 gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3442 gotp->br_state = XFS_EXT_INVALID;
3443 #if XFS_BIG_BLKNOS
3444 gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
3445 #else
3446 gotp->br_startblock = 0xffffa5a5;
3447 #endif
3448 prevp->br_startoff = NULLFILEOFF;
3449
3450 ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
3451 if (lastx > 0) {
3452 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
3453 }
3454 if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
3455 xfs_bmbt_get_all(ep, gotp);
3456 *eofp = 0;
3457 } else {
3458 if (lastx > 0) {
3459 *gotp = *prevp;
3460 }
3461 *eofp = 1;
3462 ep = NULL;
3463 }
3464 *lastxp = lastx;
3465 return ep;
3466 }
3467
3468 /*
3469 * Search the extents list for the inode, for the extent containing bno.
3470 * If bno lies in a hole, point to the next entry. If bno lies past eof,
3471 * *eofp will be set, and *prevp will contain the last entry (null if none).
3472 * Else, *lastxp will be set to the index of the found
3473 * entry; *gotp will contain the entry.
3474 */
3475 STATIC xfs_bmbt_rec_host_t * /* pointer to found extent entry */
3476 xfs_bmap_search_extents(
3477 xfs_inode_t *ip, /* incore inode pointer */
3478 xfs_fileoff_t bno, /* block number searched for */
3479 int fork, /* data or attr fork */
3480 int *eofp, /* out: end of file found */
3481 xfs_extnum_t *lastxp, /* out: last extent index */
3482 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
3483 xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
3484 {
3485 xfs_ifork_t *ifp; /* inode fork pointer */
3486 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
3487
3488 XFS_STATS_INC(xs_look_exlist);
3489 ifp = XFS_IFORK_PTR(ip, fork);
3490
3491 ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
3492
3493 if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
3494 !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
3495 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
3496 "Access to block zero in inode %llu "
3497 "start_block: %llx start_off: %llx "
3498 "blkcnt: %llx extent-state: %x lastx: %x\n",
3499 (unsigned long long)ip->i_ino,
3500 (unsigned long long)gotp->br_startblock,
3501 (unsigned long long)gotp->br_startoff,
3502 (unsigned long long)gotp->br_blockcount,
3503 gotp->br_state, *lastxp);
3504 *lastxp = NULLEXTNUM;
3505 *eofp = 1;
3506 return NULL;
3507 }
3508 return ep;
3509 }
3510
3511 /*
3512 * Compute the worst-case number of indirect blocks that will be used
3513 * for ip's delayed extent of length "len".
3514 */
3515 STATIC xfs_filblks_t
3516 xfs_bmap_worst_indlen(
3517 xfs_inode_t *ip, /* incore inode pointer */
3518 xfs_filblks_t len) /* delayed extent length */
3519 {
3520 int level; /* btree level number */
3521 int maxrecs; /* maximum record count at this level */
3522 xfs_mount_t *mp; /* mount structure */
3523 xfs_filblks_t rval; /* return value */
3524
3525 mp = ip->i_mount;
3526 maxrecs = mp->m_bmap_dmxr[0];
3527 for (level = 0, rval = 0;
3528 level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3529 level++) {
3530 len += maxrecs - 1;
3531 do_div(len, maxrecs);
3532 rval += len;
3533 if (len == 1)
3534 return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3535 level - 1;
3536 if (level == 0)
3537 maxrecs = mp->m_bmap_dmxr[1];
3538 }
3539 return rval;
3540 }
3541
3542 /*
3543 * Convert inode from non-attributed to attributed.
3544 * Must not be in a transaction, ip must not be locked.
3545 */
3546 int /* error code */
3547 xfs_bmap_add_attrfork(
3548 xfs_inode_t *ip, /* incore inode pointer */
3549 int size, /* space new attribute needs */
3550 int rsvd) /* xact may use reserved blks */
3551 {
3552 xfs_fsblock_t firstblock; /* 1st block/ag allocated */
3553 xfs_bmap_free_t flist; /* freed extent records */
3554 xfs_mount_t *mp; /* mount structure */
3555 xfs_trans_t *tp; /* transaction pointer */
3556 int blks; /* space reservation */
3557 int version = 1; /* superblock attr version */
3558 int committed; /* xaction was committed */
3559 int logflags; /* logging flags */
3560 int error; /* error return value */
3561
3562 ASSERT(XFS_IFORK_Q(ip) == 0);
3563 ASSERT(ip->i_df.if_ext_max ==
3564 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3565
3566 mp = ip->i_mount;
3567 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3568 tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3569 blks = XFS_ADDAFORK_SPACE_RES(mp);
3570 if (rsvd)
3571 tp->t_flags |= XFS_TRANS_RESERVE;
3572 if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3573 XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3574 goto error0;
3575 xfs_ilock(ip, XFS_ILOCK_EXCL);
3576 error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
3577 XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3578 XFS_QMOPT_RES_REGBLKS);
3579 if (error) {
3580 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3581 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3582 return error;
3583 }
3584 if (XFS_IFORK_Q(ip))
3585 goto error1;
3586 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3587 /*
3588 * For inodes coming from pre-6.2 filesystems.
3589 */
3590 ASSERT(ip->i_d.di_aformat == 0);
3591 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
3592 }
3593 ASSERT(ip->i_d.di_anextents == 0);
3594
3595 xfs_trans_ijoin_ref(tp, ip, XFS_ILOCK_EXCL);
3596 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3597
3598 switch (ip->i_d.di_format) {
3599 case XFS_DINODE_FMT_DEV:
3600 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
3601 break;
3602 case XFS_DINODE_FMT_UUID:
3603 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
3604 break;
3605 case XFS_DINODE_FMT_LOCAL:
3606 case XFS_DINODE_FMT_EXTENTS:
3607 case XFS_DINODE_FMT_BTREE:
3608 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
3609 if (!ip->i_d.di_forkoff)
3610 ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
3611 else if (mp->m_flags & XFS_MOUNT_ATTR2)
3612 version = 2;
3613 break;
3614 default:
3615 ASSERT(0);
3616 error = XFS_ERROR(EINVAL);
3617 goto error1;
3618 }
3619 ip->i_df.if_ext_max =
3620 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3621 ASSERT(ip->i_afp == NULL);
3622 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
3623 ip->i_afp->if_ext_max =
3624 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3625 ip->i_afp->if_flags = XFS_IFEXTENTS;
3626 logflags = 0;
3627 xfs_bmap_init(&flist, &firstblock);
3628 switch (ip->i_d.di_format) {
3629 case XFS_DINODE_FMT_LOCAL:
3630 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
3631 &logflags);
3632 break;
3633 case XFS_DINODE_FMT_EXTENTS:
3634 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
3635 &flist, &logflags);
3636 break;
3637 case XFS_DINODE_FMT_BTREE:
3638 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
3639 &logflags);
3640 break;
3641 default:
3642 error = 0;
3643 break;
3644 }
3645 if (logflags)
3646 xfs_trans_log_inode(tp, ip, logflags);
3647 if (error)
3648 goto error2;
3649 if (!xfs_sb_version_hasattr(&mp->m_sb) ||
3650 (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
3651 __int64_t sbfields = 0;
3652
3653 spin_lock(&mp->m_sb_lock);
3654 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
3655 xfs_sb_version_addattr(&mp->m_sb);
3656 sbfields |= XFS_SB_VERSIONNUM;
3657 }
3658 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
3659 xfs_sb_version_addattr2(&mp->m_sb);
3660 sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
3661 }
3662 if (sbfields) {
3663 spin_unlock(&mp->m_sb_lock);
3664 xfs_mod_sb(tp, sbfields);
3665 } else
3666 spin_unlock(&mp->m_sb_lock);
3667 }
3668 if ((error = xfs_bmap_finish(&tp, &flist, &committed)))
3669 goto error2;
3670 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3671 ASSERT(ip->i_df.if_ext_max ==
3672 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3673 return error;
3674 error2:
3675 xfs_bmap_cancel(&flist);
3676 error1:
3677 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3678 error0:
3679 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
3680 ASSERT(ip->i_df.if_ext_max ==
3681 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3682 return error;
3683 }
3684
3685 /*
3686 * Add the extent to the list of extents to be free at transaction end.
3687 * The list is maintained sorted (by block number).
3688 */
3689 /* ARGSUSED */
3690 void
3691 xfs_bmap_add_free(
3692 xfs_fsblock_t bno, /* fs block number of extent */
3693 xfs_filblks_t len, /* length of extent */
3694 xfs_bmap_free_t *flist, /* list of extents */
3695 xfs_mount_t *mp) /* mount point structure */
3696 {
3697 xfs_bmap_free_item_t *cur; /* current (next) element */
3698 xfs_bmap_free_item_t *new; /* new element */
3699 xfs_bmap_free_item_t *prev; /* previous element */
3700 #ifdef DEBUG
3701 xfs_agnumber_t agno;
3702 xfs_agblock_t agbno;
3703
3704 ASSERT(bno != NULLFSBLOCK);
3705 ASSERT(len > 0);
3706 ASSERT(len <= MAXEXTLEN);
3707 ASSERT(!isnullstartblock(bno));
3708 agno = XFS_FSB_TO_AGNO(mp, bno);
3709 agbno = XFS_FSB_TO_AGBNO(mp, bno);
3710 ASSERT(agno < mp->m_sb.sb_agcount);
3711 ASSERT(agbno < mp->m_sb.sb_agblocks);
3712 ASSERT(len < mp->m_sb.sb_agblocks);
3713 ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
3714 #endif
3715 ASSERT(xfs_bmap_free_item_zone != NULL);
3716 new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
3717 new->xbfi_startblock = bno;
3718 new->xbfi_blockcount = (xfs_extlen_t)len;
3719 for (prev = NULL, cur = flist->xbf_first;
3720 cur != NULL;
3721 prev = cur, cur = cur->xbfi_next) {
3722 if (cur->xbfi_startblock >= bno)
3723 break;
3724 }
3725 if (prev)
3726 prev->xbfi_next = new;
3727 else
3728 flist->xbf_first = new;
3729 new->xbfi_next = cur;
3730 flist->xbf_count++;
3731 }
3732
3733 /*
3734 * Compute and fill in the value of the maximum depth of a bmap btree
3735 * in this filesystem. Done once, during mount.
3736 */
3737 void
3738 xfs_bmap_compute_maxlevels(
3739 xfs_mount_t *mp, /* file system mount structure */
3740 int whichfork) /* data or attr fork */
3741 {
3742 int level; /* btree level */
3743 uint maxblocks; /* max blocks at this level */
3744 uint maxleafents; /* max leaf entries possible */
3745 int maxrootrecs; /* max records in root block */
3746 int minleafrecs; /* min records in leaf block */
3747 int minnoderecs; /* min records in node block */
3748 int sz; /* root block size */
3749
3750 /*
3751 * The maximum number of extents in a file, hence the maximum
3752 * number of leaf entries, is controlled by the type of di_nextents
3753 * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
3754 * (a signed 16-bit number, xfs_aextnum_t).
3755 *
3756 * Note that we can no longer assume that if we are in ATTR1 that
3757 * the fork offset of all the inodes will be
3758 * (xfs_default_attroffset(ip) >> 3) because we could have mounted
3759 * with ATTR2 and then mounted back with ATTR1, keeping the
3760 * di_forkoff's fixed but probably at various positions. Therefore,
3761 * for both ATTR1 and ATTR2 we have to assume the worst case scenario
3762 * of a minimum size available.
3763 */
3764 if (whichfork == XFS_DATA_FORK) {
3765 maxleafents = MAXEXTNUM;
3766 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
3767 } else {
3768 maxleafents = MAXAEXTNUM;
3769 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
3770 }
3771 maxrootrecs = xfs_bmdr_maxrecs(mp, sz, 0);
3772 minleafrecs = mp->m_bmap_dmnr[0];
3773 minnoderecs = mp->m_bmap_dmnr[1];
3774 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
3775 for (level = 1; maxblocks > 1; level++) {
3776 if (maxblocks <= maxrootrecs)
3777 maxblocks = 1;
3778 else
3779 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
3780 }
3781 mp->m_bm_maxlevels[whichfork] = level;
3782 }
3783
3784 /*
3785 * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
3786 * caller. Frees all the extents that need freeing, which must be done
3787 * last due to locking considerations. We never free any extents in
3788 * the first transaction.
3789 *
3790 * Return 1 if the given transaction was committed and a new one
3791 * started, and 0 otherwise in the committed parameter.
3792 */
3793 int /* error */
3794 xfs_bmap_finish(
3795 xfs_trans_t **tp, /* transaction pointer addr */
3796 xfs_bmap_free_t *flist, /* i/o: list extents to free */
3797 int *committed) /* xact committed or not */
3798 {
3799 xfs_efd_log_item_t *efd; /* extent free data */
3800 xfs_efi_log_item_t *efi; /* extent free intention */
3801 int error; /* error return value */
3802 xfs_bmap_free_item_t *free; /* free extent item */
3803 unsigned int logres; /* new log reservation */
3804 unsigned int logcount; /* new log count */
3805 xfs_mount_t *mp; /* filesystem mount structure */
3806 xfs_bmap_free_item_t *next; /* next item on free list */
3807 xfs_trans_t *ntp; /* new transaction pointer */
3808
3809 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
3810 if (flist->xbf_count == 0) {
3811 *committed = 0;
3812 return 0;
3813 }
3814 ntp = *tp;
3815 efi = xfs_trans_get_efi(ntp, flist->xbf_count);
3816 for (free = flist->xbf_first; free; free = free->xbfi_next)
3817 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
3818 free->xbfi_blockcount);
3819 logres = ntp->t_log_res;
3820 logcount = ntp->t_log_count;
3821 ntp = xfs_trans_dup(*tp);
3822 error = xfs_trans_commit(*tp, 0);
3823 *tp = ntp;
3824 *committed = 1;
3825 /*
3826 * We have a new transaction, so we should return committed=1,
3827 * even though we're returning an error.
3828 */
3829 if (error)
3830 return error;
3831
3832 /*
3833 * transaction commit worked ok so we can drop the extra ticket
3834 * reference that we gained in xfs_trans_dup()
3835 */
3836 xfs_log_ticket_put(ntp->t_ticket);
3837
3838 if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
3839 logcount)))
3840 return error;
3841 efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
3842 for (free = flist->xbf_first; free != NULL; free = next) {
3843 next = free->xbfi_next;
3844 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
3845 free->xbfi_blockcount))) {
3846 /*
3847 * The bmap free list will be cleaned up at a
3848 * higher level. The EFI will be canceled when
3849 * this transaction is aborted.
3850 * Need to force shutdown here to make sure it
3851 * happens, since this transaction may not be
3852 * dirty yet.
3853 */
3854 mp = ntp->t_mountp;
3855 if (!XFS_FORCED_SHUTDOWN(mp))
3856 xfs_force_shutdown(mp,
3857 (error == EFSCORRUPTED) ?
3858 SHUTDOWN_CORRUPT_INCORE :
3859 SHUTDOWN_META_IO_ERROR);
3860 return error;
3861 }
3862 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
3863 free->xbfi_blockcount);
3864 xfs_bmap_del_free(flist, NULL, free);
3865 }
3866 return 0;
3867 }
3868
3869 /*
3870 * Free up any items left in the list.
3871 */
3872 void
3873 xfs_bmap_cancel(
3874 xfs_bmap_free_t *flist) /* list of bmap_free_items */
3875 {
3876 xfs_bmap_free_item_t *free; /* free list item */
3877 xfs_bmap_free_item_t *next;
3878
3879 if (flist->xbf_count == 0)
3880 return;
3881 ASSERT(flist->xbf_first != NULL);
3882 for (free = flist->xbf_first; free; free = next) {
3883 next = free->xbfi_next;
3884 xfs_bmap_del_free(flist, NULL, free);
3885 }
3886 ASSERT(flist->xbf_count == 0);
3887 }
3888
3889 /*
3890 * Returns the file-relative block number of the first unused block(s)
3891 * in the file with at least "len" logically contiguous blocks free.
3892 * This is the lowest-address hole if the file has holes, else the first block
3893 * past the end of file.
3894 * Return 0 if the file is currently local (in-inode).
3895 */
3896 int /* error */
3897 xfs_bmap_first_unused(
3898 xfs_trans_t *tp, /* transaction pointer */
3899 xfs_inode_t *ip, /* incore inode */
3900 xfs_extlen_t len, /* size of hole to find */
3901 xfs_fileoff_t *first_unused, /* unused block */
3902 int whichfork) /* data or attr fork */
3903 {
3904 int error; /* error return value */
3905 int idx; /* extent record index */
3906 xfs_ifork_t *ifp; /* inode fork pointer */
3907 xfs_fileoff_t lastaddr; /* last block number seen */
3908 xfs_fileoff_t lowest; /* lowest useful block */
3909 xfs_fileoff_t max; /* starting useful block */
3910 xfs_fileoff_t off; /* offset for this block */
3911 xfs_extnum_t nextents; /* number of extent entries */
3912
3913 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
3914 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
3915 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3916 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3917 *first_unused = 0;
3918 return 0;
3919 }
3920 ifp = XFS_IFORK_PTR(ip, whichfork);
3921 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3922 (error = xfs_iread_extents(tp, ip, whichfork)))
3923 return error;
3924 lowest = *first_unused;
3925 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3926 for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
3927 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
3928 off = xfs_bmbt_get_startoff(ep);
3929 /*
3930 * See if the hole before this extent will work.
3931 */
3932 if (off >= lowest + len && off - max >= len) {
3933 *first_unused = max;
3934 return 0;
3935 }
3936 lastaddr = off + xfs_bmbt_get_blockcount(ep);
3937 max = XFS_FILEOFF_MAX(lastaddr, lowest);
3938 }
3939 *first_unused = max;
3940 return 0;
3941 }
3942
3943 /*
3944 * Returns the file-relative block number of the last block + 1 before
3945 * last_block (input value) in the file.
3946 * This is not based on i_size, it is based on the extent records.
3947 * Returns 0 for local files, as they do not have extent records.
3948 */
3949 int /* error */
3950 xfs_bmap_last_before(
3951 xfs_trans_t *tp, /* transaction pointer */
3952 xfs_inode_t *ip, /* incore inode */
3953 xfs_fileoff_t *last_block, /* last block */
3954 int whichfork) /* data or attr fork */
3955 {
3956 xfs_fileoff_t bno; /* input file offset */
3957 int eof; /* hit end of file */
3958 xfs_bmbt_rec_host_t *ep; /* pointer to last extent */
3959 int error; /* error return value */
3960 xfs_bmbt_irec_t got; /* current extent value */
3961 xfs_ifork_t *ifp; /* inode fork pointer */
3962 xfs_extnum_t lastx; /* last extent used */
3963 xfs_bmbt_irec_t prev; /* previous extent value */
3964
3965 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
3966 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
3967 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
3968 return XFS_ERROR(EIO);
3969 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3970 *last_block = 0;
3971 return 0;
3972 }
3973 ifp = XFS_IFORK_PTR(ip, whichfork);
3974 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3975 (error = xfs_iread_extents(tp, ip, whichfork)))
3976 return error;
3977 bno = *last_block - 1;
3978 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
3979 &prev);
3980 if (eof || xfs_bmbt_get_startoff(ep) > bno) {
3981 if (prev.br_startoff == NULLFILEOFF)
3982 *last_block = 0;
3983 else
3984 *last_block = prev.br_startoff + prev.br_blockcount;
3985 }
3986 /*
3987 * Otherwise *last_block is already the right answer.
3988 */
3989 return 0;
3990 }
3991
3992 /*
3993 * Returns the file-relative block number of the first block past eof in
3994 * the file. This is not based on i_size, it is based on the extent records.
3995 * Returns 0 for local files, as they do not have extent records.
3996 */
3997 int /* error */
3998 xfs_bmap_last_offset(
3999 xfs_trans_t *tp, /* transaction pointer */
4000 xfs_inode_t *ip, /* incore inode */
4001 xfs_fileoff_t *last_block, /* last block */
4002 int whichfork) /* data or attr fork */
4003 {
4004 xfs_bmbt_rec_host_t *ep; /* pointer to last extent */
4005 int error; /* error return value */
4006 xfs_ifork_t *ifp; /* inode fork pointer */
4007 xfs_extnum_t nextents; /* number of extent entries */
4008
4009 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4010 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4011 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4012 return XFS_ERROR(EIO);
4013 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4014 *last_block = 0;
4015 return 0;
4016 }
4017 ifp = XFS_IFORK_PTR(ip, whichfork);
4018 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4019 (error = xfs_iread_extents(tp, ip, whichfork)))
4020 return error;
4021 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4022 if (!nextents) {
4023 *last_block = 0;
4024 return 0;
4025 }
4026 ep = xfs_iext_get_ext(ifp, nextents - 1);
4027 *last_block = xfs_bmbt_get_startoff(ep) + xfs_bmbt_get_blockcount(ep);
4028 return 0;
4029 }
4030
4031 /*
4032 * Returns whether the selected fork of the inode has exactly one
4033 * block or not. For the data fork we check this matches di_size,
4034 * implying the file's range is 0..bsize-1.
4035 */
4036 int /* 1=>1 block, 0=>otherwise */
4037 xfs_bmap_one_block(
4038 xfs_inode_t *ip, /* incore inode */
4039 int whichfork) /* data or attr fork */
4040 {
4041 xfs_bmbt_rec_host_t *ep; /* ptr to fork's extent */
4042 xfs_ifork_t *ifp; /* inode fork pointer */
4043 int rval; /* return value */
4044 xfs_bmbt_irec_t s; /* internal version of extent */
4045
4046 #ifndef DEBUG
4047 if (whichfork == XFS_DATA_FORK) {
4048 return ((ip->i_d.di_mode & S_IFMT) == S_IFREG) ?
4049 (ip->i_size == ip->i_mount->m_sb.sb_blocksize) :
4050 (ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize);
4051 }
4052 #endif /* !DEBUG */
4053 if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4054 return 0;
4055 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4056 return 0;
4057 ifp = XFS_IFORK_PTR(ip, whichfork);
4058 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4059 ep = xfs_iext_get_ext(ifp, 0);
4060 xfs_bmbt_get_all(ep, &s);
4061 rval = s.br_startoff == 0 && s.br_blockcount == 1;
4062 if (rval && whichfork == XFS_DATA_FORK)
4063 ASSERT(ip->i_size == ip->i_mount->m_sb.sb_blocksize);
4064 return rval;
4065 }
4066
4067 STATIC int
4068 xfs_bmap_sanity_check(
4069 struct xfs_mount *mp,
4070 struct xfs_buf *bp,
4071 int level)
4072 {
4073 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
4074
4075 if (be32_to_cpu(block->bb_magic) != XFS_BMAP_MAGIC ||
4076 be16_to_cpu(block->bb_level) != level ||
4077 be16_to_cpu(block->bb_numrecs) == 0 ||
4078 be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
4079 return 0;
4080 return 1;
4081 }
4082
4083 /*
4084 * Read in the extents to if_extents.
4085 * All inode fields are set up by caller, we just traverse the btree
4086 * and copy the records in. If the file system cannot contain unwritten
4087 * extents, the records are checked for no "state" flags.
4088 */
4089 int /* error */
4090 xfs_bmap_read_extents(
4091 xfs_trans_t *tp, /* transaction pointer */
4092 xfs_inode_t *ip, /* incore inode */
4093 int whichfork) /* data or attr fork */
4094 {
4095 struct xfs_btree_block *block; /* current btree block */
4096 xfs_fsblock_t bno; /* block # of "block" */
4097 xfs_buf_t *bp; /* buffer for "block" */
4098 int error; /* error return value */
4099 xfs_exntfmt_t exntf; /* XFS_EXTFMT_NOSTATE, if checking */
4100 xfs_extnum_t i, j; /* index into the extents list */
4101 xfs_ifork_t *ifp; /* fork structure */
4102 int level; /* btree level, for checking */
4103 xfs_mount_t *mp; /* file system mount structure */
4104 __be64 *pp; /* pointer to block address */
4105 /* REFERENCED */
4106 xfs_extnum_t room; /* number of entries there's room for */
4107
4108 bno = NULLFSBLOCK;
4109 mp = ip->i_mount;
4110 ifp = XFS_IFORK_PTR(ip, whichfork);
4111 exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4112 XFS_EXTFMT_INODE(ip);
4113 block = ifp->if_broot;
4114 /*
4115 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4116 */
4117 level = be16_to_cpu(block->bb_level);
4118 ASSERT(level > 0);
4119 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
4120 bno = be64_to_cpu(*pp);
4121 ASSERT(bno != NULLDFSBNO);
4122 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
4123 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
4124 /*
4125 * Go down the tree until leaf level is reached, following the first
4126 * pointer (leftmost) at each level.
4127 */
4128 while (level-- > 0) {
4129 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4130 XFS_BMAP_BTREE_REF)))
4131 return error;
4132 block = XFS_BUF_TO_BLOCK(bp);
4133 XFS_WANT_CORRUPTED_GOTO(
4134 xfs_bmap_sanity_check(mp, bp, level),
4135 error0);
4136 if (level == 0)
4137 break;
4138 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
4139 bno = be64_to_cpu(*pp);
4140 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
4141 xfs_trans_brelse(tp, bp);
4142 }
4143 /*
4144 * Here with bp and block set to the leftmost leaf node in the tree.
4145 */
4146 room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4147 i = 0;
4148 /*
4149 * Loop over all leaf nodes. Copy information to the extent records.
4150 */
4151 for (;;) {
4152 xfs_bmbt_rec_t *frp;
4153 xfs_fsblock_t nextbno;
4154 xfs_extnum_t num_recs;
4155 xfs_extnum_t start;
4156
4157
4158 num_recs = xfs_btree_get_numrecs(block);
4159 if (unlikely(i + num_recs > room)) {
4160 ASSERT(i + num_recs <= room);
4161 xfs_warn(ip->i_mount,
4162 "corrupt dinode %Lu, (btree extents).",
4163 (unsigned long long) ip->i_ino);
4164 XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
4165 XFS_ERRLEVEL_LOW, ip->i_mount, block);
4166 goto error0;
4167 }
4168 XFS_WANT_CORRUPTED_GOTO(
4169 xfs_bmap_sanity_check(mp, bp, 0),
4170 error0);
4171 /*
4172 * Read-ahead the next leaf block, if any.
4173 */
4174 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
4175 if (nextbno != NULLFSBLOCK)
4176 xfs_btree_reada_bufl(mp, nextbno, 1);
4177 /*
4178 * Copy records into the extent records.
4179 */
4180 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
4181 start = i;
4182 for (j = 0; j < num_recs; j++, i++, frp++) {
4183 xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
4184 trp->l0 = be64_to_cpu(frp->l0);
4185 trp->l1 = be64_to_cpu(frp->l1);
4186 }
4187 if (exntf == XFS_EXTFMT_NOSTATE) {
4188 /*
4189 * Check all attribute bmap btree records and
4190 * any "older" data bmap btree records for a
4191 * set bit in the "extent flag" position.
4192 */
4193 if (unlikely(xfs_check_nostate_extents(ifp,
4194 start, num_recs))) {
4195 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4196 XFS_ERRLEVEL_LOW,
4197 ip->i_mount);
4198 goto error0;
4199 }
4200 }
4201 xfs_trans_brelse(tp, bp);
4202 bno = nextbno;
4203 /*
4204 * If we've reached the end, stop.
4205 */
4206 if (bno == NULLFSBLOCK)
4207 break;
4208 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4209 XFS_BMAP_BTREE_REF)))
4210 return error;
4211 block = XFS_BUF_TO_BLOCK(bp);
4212 }
4213 ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4214 ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4215 XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
4216 return 0;
4217 error0:
4218 xfs_trans_brelse(tp, bp);
4219 return XFS_ERROR(EFSCORRUPTED);
4220 }
4221
4222 #ifdef DEBUG
4223 /*
4224 * Add bmap trace insert entries for all the contents of the extent records.
4225 */
4226 void
4227 xfs_bmap_trace_exlist(
4228 xfs_inode_t *ip, /* incore inode pointer */
4229 xfs_extnum_t cnt, /* count of entries in the list */
4230 int whichfork, /* data or attr fork */
4231 unsigned long caller_ip)
4232 {
4233 xfs_extnum_t idx; /* extent record index */
4234 xfs_ifork_t *ifp; /* inode fork pointer */
4235 int state = 0;
4236
4237 if (whichfork == XFS_ATTR_FORK)
4238 state |= BMAP_ATTRFORK;
4239
4240 ifp = XFS_IFORK_PTR(ip, whichfork);
4241 ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4242 for (idx = 0; idx < cnt; idx++)
4243 trace_xfs_extlist(ip, idx, whichfork, caller_ip);
4244 }
4245
4246 /*
4247 * Validate that the bmbt_irecs being returned from bmapi are valid
4248 * given the callers original parameters. Specifically check the
4249 * ranges of the returned irecs to ensure that they only extent beyond
4250 * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4251 */
4252 STATIC void
4253 xfs_bmap_validate_ret(
4254 xfs_fileoff_t bno,
4255 xfs_filblks_t len,
4256 int flags,
4257 xfs_bmbt_irec_t *mval,
4258 int nmap,
4259 int ret_nmap)
4260 {
4261 int i; /* index to map values */
4262
4263 ASSERT(ret_nmap <= nmap);
4264
4265 for (i = 0; i < ret_nmap; i++) {
4266 ASSERT(mval[i].br_blockcount > 0);
4267 if (!(flags & XFS_BMAPI_ENTIRE)) {
4268 ASSERT(mval[i].br_startoff >= bno);
4269 ASSERT(mval[i].br_blockcount <= len);
4270 ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4271 bno + len);
4272 } else {
4273 ASSERT(mval[i].br_startoff < bno + len);
4274 ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4275 bno);
4276 }
4277 ASSERT(i == 0 ||
4278 mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4279 mval[i].br_startoff);
4280 if ((flags & XFS_BMAPI_WRITE) && !(flags & XFS_BMAPI_DELAY))
4281 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4282 mval[i].br_startblock != HOLESTARTBLOCK);
4283 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4284 mval[i].br_state == XFS_EXT_UNWRITTEN);
4285 }
4286 }
4287 #endif /* DEBUG */
4288
4289
4290 /*
4291 * Map file blocks to filesystem blocks.
4292 * File range is given by the bno/len pair.
4293 * Adds blocks to file if a write ("flags & XFS_BMAPI_WRITE" set)
4294 * into a hole or past eof.
4295 * Only allocates blocks from a single allocation group,
4296 * to avoid locking problems.
4297 * The returned value in "firstblock" from the first call in a transaction
4298 * must be remembered and presented to subsequent calls in "firstblock".
4299 * An upper bound for the number of blocks to be allocated is supplied to
4300 * the first call in "total"; if no allocation group has that many free
4301 * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4302 */
4303 int /* error */
4304 xfs_bmapi(
4305 xfs_trans_t *tp, /* transaction pointer */
4306 xfs_inode_t *ip, /* incore inode */
4307 xfs_fileoff_t bno, /* starting file offs. mapped */
4308 xfs_filblks_t len, /* length to map in file */
4309 int flags, /* XFS_BMAPI_... */
4310 xfs_fsblock_t *firstblock, /* first allocated block
4311 controls a.g. for allocs */
4312 xfs_extlen_t total, /* total blocks needed */
4313 xfs_bmbt_irec_t *mval, /* output: map values */
4314 int *nmap, /* i/o: mval size/count */
4315 xfs_bmap_free_t *flist) /* i/o: list extents to free */
4316 {
4317 xfs_fsblock_t abno; /* allocated block number */
4318 xfs_extlen_t alen; /* allocated extent length */
4319 xfs_fileoff_t aoff; /* allocated file offset */
4320 xfs_bmalloca_t bma = { 0 }; /* args for xfs_bmap_alloc */
4321 xfs_btree_cur_t *cur; /* bmap btree cursor */
4322 xfs_fileoff_t end; /* end of mapped file region */
4323 int eof; /* we've hit the end of extents */
4324 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
4325 int error; /* error return */
4326 xfs_bmbt_irec_t got; /* current file extent record */
4327 xfs_ifork_t *ifp; /* inode fork pointer */
4328 xfs_extlen_t indlen; /* indirect blocks length */
4329 xfs_extnum_t lastx; /* last useful extent number */
4330 int logflags; /* flags for transaction logging */
4331 xfs_extlen_t minleft; /* min blocks left after allocation */
4332 xfs_extlen_t minlen; /* min allocation size */
4333 xfs_mount_t *mp; /* xfs mount structure */
4334 int n; /* current extent index */
4335 int nallocs; /* number of extents alloc'd */
4336 xfs_extnum_t nextents; /* number of extents in file */
4337 xfs_fileoff_t obno; /* old block number (offset) */
4338 xfs_bmbt_irec_t prev; /* previous file extent record */
4339 int tmp_logflags; /* temp flags holder */
4340 int whichfork; /* data or attr fork */
4341 char inhole; /* current location is hole in file */
4342 char wasdelay; /* old extent was delayed */
4343 char wr; /* this is a write request */
4344 char rt; /* this is a realtime file */
4345 #ifdef DEBUG
4346 xfs_fileoff_t orig_bno; /* original block number value */
4347 int orig_flags; /* original flags arg value */
4348 xfs_filblks_t orig_len; /* original value of len arg */
4349 xfs_bmbt_irec_t *orig_mval; /* original value of mval */
4350 int orig_nmap; /* original value of *nmap */
4351
4352 orig_bno = bno;
4353 orig_len = len;
4354 orig_flags = flags;
4355 orig_mval = mval;
4356 orig_nmap = *nmap;
4357 #endif
4358 ASSERT(*nmap >= 1);
4359 ASSERT(*nmap <= XFS_BMAP_MAX_NMAP || !(flags & XFS_BMAPI_WRITE));
4360 whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4361 XFS_ATTR_FORK : XFS_DATA_FORK;
4362 mp = ip->i_mount;
4363 if (unlikely(XFS_TEST_ERROR(
4364 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4365 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4366 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4367 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4368 XFS_ERROR_REPORT("xfs_bmapi", XFS_ERRLEVEL_LOW, mp);
4369 return XFS_ERROR(EFSCORRUPTED);
4370 }
4371 if (XFS_FORCED_SHUTDOWN(mp))
4372 return XFS_ERROR(EIO);
4373 rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4374 ifp = XFS_IFORK_PTR(ip, whichfork);
4375 ASSERT(ifp->if_ext_max ==
4376 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4377 if ((wr = (flags & XFS_BMAPI_WRITE)) != 0)
4378 XFS_STATS_INC(xs_blk_mapw);
4379 else
4380 XFS_STATS_INC(xs_blk_mapr);
4381 /*
4382 * IGSTATE flag is used to combine extents which
4383 * differ only due to the state of the extents.
4384 * This technique is used from xfs_getbmap()
4385 * when the caller does not wish to see the
4386 * separation (which is the default).
4387 *
4388 * This technique is also used when writing a
4389 * buffer which has been partially written,
4390 * (usually by being flushed during a chunkread),
4391 * to ensure one write takes place. This also
4392 * prevents a change in the xfs inode extents at
4393 * this time, intentionally. This change occurs
4394 * on completion of the write operation, in
4395 * xfs_strat_comp(), where the xfs_bmapi() call
4396 * is transactioned, and the extents combined.
4397 */
4398 if ((flags & XFS_BMAPI_IGSTATE) && wr) /* if writing unwritten space */
4399 wr = 0; /* no allocations are allowed */
4400 ASSERT(wr || !(flags & XFS_BMAPI_DELAY));
4401 logflags = 0;
4402 nallocs = 0;
4403 cur = NULL;
4404 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4405 ASSERT(wr && tp);
4406 if ((error = xfs_bmap_local_to_extents(tp, ip,
4407 firstblock, total, &logflags, whichfork)))
4408 goto error0;
4409 }
4410 if (wr && *firstblock == NULLFSBLOCK) {
4411 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4412 minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4413 else
4414 minleft = 1;
4415 } else
4416 minleft = 0;
4417 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4418 (error = xfs_iread_extents(tp, ip, whichfork)))
4419 goto error0;
4420 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4421 &prev);
4422 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4423 n = 0;
4424 end = bno + len;
4425 obno = bno;
4426 bma.ip = NULL;
4427
4428 while (bno < end && n < *nmap) {
4429 /*
4430 * Reading past eof, act as though there's a hole
4431 * up to end.
4432 */
4433 if (eof && !wr)
4434 got.br_startoff = end;
4435 inhole = eof || got.br_startoff > bno;
4436 wasdelay = wr && !inhole && !(flags & XFS_BMAPI_DELAY) &&
4437 isnullstartblock(got.br_startblock);
4438 /*
4439 * First, deal with the hole before the allocated space
4440 * that we found, if any.
4441 */
4442 if (wr && (inhole || wasdelay)) {
4443 /*
4444 * For the wasdelay case, we could also just
4445 * allocate the stuff asked for in this bmap call
4446 * but that wouldn't be as good.
4447 */
4448 if (wasdelay) {
4449 alen = (xfs_extlen_t)got.br_blockcount;
4450 aoff = got.br_startoff;
4451 if (lastx != NULLEXTNUM && lastx) {
4452 ep = xfs_iext_get_ext(ifp, lastx - 1);
4453 xfs_bmbt_get_all(ep, &prev);
4454 }
4455 } else {
4456 alen = (xfs_extlen_t)
4457 XFS_FILBLKS_MIN(len, MAXEXTLEN);
4458 if (!eof)
4459 alen = (xfs_extlen_t)
4460 XFS_FILBLKS_MIN(alen,
4461 got.br_startoff - bno);
4462 aoff = bno;
4463 }
4464 minlen = (flags & XFS_BMAPI_CONTIG) ? alen : 1;
4465 if (flags & XFS_BMAPI_DELAY) {
4466 xfs_extlen_t extsz;
4467
4468 /* Figure out the extent size, adjust alen */
4469 extsz = xfs_get_extsz_hint(ip);
4470 if (extsz) {
4471 /*
4472 * make sure we don't exceed a single
4473 * extent length when we align the
4474 * extent by reducing length we are
4475 * going to allocate by the maximum
4476 * amount extent size aligment may
4477 * require.
4478 */
4479 alen = XFS_FILBLKS_MIN(len,
4480 MAXEXTLEN - (2 * extsz - 1));
4481 error = xfs_bmap_extsize_align(mp,
4482 &got, &prev, extsz,
4483 rt, eof,
4484 flags&XFS_BMAPI_DELAY,
4485 flags&XFS_BMAPI_CONVERT,
4486 &aoff, &alen);
4487 ASSERT(!error);
4488 }
4489
4490 if (rt)
4491 extsz = alen / mp->m_sb.sb_rextsize;
4492
4493 /*
4494 * Make a transaction-less quota reservation for
4495 * delayed allocation blocks. This number gets
4496 * adjusted later. We return if we haven't
4497 * allocated blocks already inside this loop.
4498 */
4499 error = xfs_trans_reserve_quota_nblks(
4500 NULL, ip, (long)alen, 0,
4501 rt ? XFS_QMOPT_RES_RTBLKS :
4502 XFS_QMOPT_RES_REGBLKS);
4503 if (error) {
4504 if (n == 0) {
4505 *nmap = 0;
4506 ASSERT(cur == NULL);
4507 return error;
4508 }
4509 break;
4510 }
4511
4512 /*
4513 * Split changing sb for alen and indlen since
4514 * they could be coming from different places.
4515 */
4516 indlen = (xfs_extlen_t)
4517 xfs_bmap_worst_indlen(ip, alen);
4518 ASSERT(indlen > 0);
4519
4520 if (rt) {
4521 error = xfs_mod_incore_sb(mp,
4522 XFS_SBS_FREXTENTS,
4523 -((int64_t)extsz), 0);
4524 } else {
4525 error = xfs_icsb_modify_counters(mp,
4526 XFS_SBS_FDBLOCKS,
4527 -((int64_t)alen), 0);
4528 }
4529 if (!error) {
4530 error = xfs_icsb_modify_counters(mp,
4531 XFS_SBS_FDBLOCKS,
4532 -((int64_t)indlen), 0);
4533 if (error && rt)
4534 xfs_mod_incore_sb(mp,
4535 XFS_SBS_FREXTENTS,
4536 (int64_t)extsz, 0);
4537 else if (error)
4538 xfs_icsb_modify_counters(mp,
4539 XFS_SBS_FDBLOCKS,
4540 (int64_t)alen, 0);
4541 }
4542
4543 if (error) {
4544 if (XFS_IS_QUOTA_ON(mp))
4545 /* unreserve the blocks now */
4546 (void)
4547 xfs_trans_unreserve_quota_nblks(
4548 NULL, ip,
4549 (long)alen, 0, rt ?
4550 XFS_QMOPT_RES_RTBLKS :
4551 XFS_QMOPT_RES_REGBLKS);
4552 break;
4553 }
4554
4555 ip->i_delayed_blks += alen;
4556 abno = nullstartblock(indlen);
4557 } else {
4558 /*
4559 * If first time, allocate and fill in
4560 * once-only bma fields.
4561 */
4562 if (bma.ip == NULL) {
4563 bma.tp = tp;
4564 bma.ip = ip;
4565 bma.prevp = &prev;
4566 bma.gotp = &got;
4567 bma.total = total;
4568 bma.userdata = 0;
4569 }
4570 /* Indicate if this is the first user data
4571 * in the file, or just any user data.
4572 */
4573 if (!(flags & XFS_BMAPI_METADATA)) {
4574 bma.userdata = (aoff == 0) ?
4575 XFS_ALLOC_INITIAL_USER_DATA :
4576 XFS_ALLOC_USERDATA;
4577 }
4578 /*
4579 * Fill in changeable bma fields.
4580 */
4581 bma.eof = eof;
4582 bma.firstblock = *firstblock;
4583 bma.alen = alen;
4584 bma.off = aoff;
4585 bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4586 bma.wasdel = wasdelay;
4587 bma.minlen = minlen;
4588 bma.low = flist->xbf_low;
4589 bma.minleft = minleft;
4590 /*
4591 * Only want to do the alignment at the
4592 * eof if it is userdata and allocation length
4593 * is larger than a stripe unit.
4594 */
4595 if (mp->m_dalign && alen >= mp->m_dalign &&
4596 (!(flags & XFS_BMAPI_METADATA)) &&
4597 (whichfork == XFS_DATA_FORK)) {
4598 if ((error = xfs_bmap_isaeof(ip, aoff,
4599 whichfork, &bma.aeof)))
4600 goto error0;
4601 } else
4602 bma.aeof = 0;
4603 /*
4604 * Call allocator.
4605 */
4606 if ((error = xfs_bmap_alloc(&bma)))
4607 goto error0;
4608 /*
4609 * Copy out result fields.
4610 */
4611 abno = bma.rval;
4612 if ((flist->xbf_low = bma.low))
4613 minleft = 0;
4614 alen = bma.alen;
4615 aoff = bma.off;
4616 ASSERT(*firstblock == NULLFSBLOCK ||
4617 XFS_FSB_TO_AGNO(mp, *firstblock) ==
4618 XFS_FSB_TO_AGNO(mp, bma.firstblock) ||
4619 (flist->xbf_low &&
4620 XFS_FSB_TO_AGNO(mp, *firstblock) <
4621 XFS_FSB_TO_AGNO(mp, bma.firstblock)));
4622 *firstblock = bma.firstblock;
4623 if (cur)
4624 cur->bc_private.b.firstblock =
4625 *firstblock;
4626 if (abno == NULLFSBLOCK)
4627 break;
4628 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
4629 cur = xfs_bmbt_init_cursor(mp, tp,
4630 ip, whichfork);
4631 cur->bc_private.b.firstblock =
4632 *firstblock;
4633 cur->bc_private.b.flist = flist;
4634 }
4635 /*
4636 * Bump the number of extents we've allocated
4637 * in this call.
4638 */
4639 nallocs++;
4640 }
4641 if (cur)
4642 cur->bc_private.b.flags =
4643 wasdelay ? XFS_BTCUR_BPRV_WASDEL : 0;
4644 got.br_startoff = aoff;
4645 got.br_startblock = abno;
4646 got.br_blockcount = alen;
4647 got.br_state = XFS_EXT_NORM; /* assume normal */
4648 /*
4649 * Determine state of extent, and the filesystem.
4650 * A wasdelay extent has been initialized, so
4651 * shouldn't be flagged as unwritten.
4652 */
4653 if (wr && xfs_sb_version_hasextflgbit(&mp->m_sb)) {
4654 if (!wasdelay && (flags & XFS_BMAPI_PREALLOC))
4655 got.br_state = XFS_EXT_UNWRITTEN;
4656 }
4657 error = xfs_bmap_add_extent(ip, &lastx, &cur, &got,
4658 firstblock, flist, &tmp_logflags,
4659 whichfork);
4660 logflags |= tmp_logflags;
4661 if (error)
4662 goto error0;
4663 ep = xfs_iext_get_ext(ifp, lastx);
4664 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4665 xfs_bmbt_get_all(ep, &got);
4666 ASSERT(got.br_startoff <= aoff);
4667 ASSERT(got.br_startoff + got.br_blockcount >=
4668 aoff + alen);
4669 #ifdef DEBUG
4670 if (flags & XFS_BMAPI_DELAY) {
4671 ASSERT(isnullstartblock(got.br_startblock));
4672 ASSERT(startblockval(got.br_startblock) > 0);
4673 }
4674 ASSERT(got.br_state == XFS_EXT_NORM ||
4675 got.br_state == XFS_EXT_UNWRITTEN);
4676 #endif
4677 /*
4678 * Fall down into the found allocated space case.
4679 */
4680 } else if (inhole) {
4681 /*
4682 * Reading in a hole.
4683 */
4684 mval->br_startoff = bno;
4685 mval->br_startblock = HOLESTARTBLOCK;
4686 mval->br_blockcount =
4687 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4688 mval->br_state = XFS_EXT_NORM;
4689 bno += mval->br_blockcount;
4690 len -= mval->br_blockcount;
4691 mval++;
4692 n++;
4693 continue;
4694 }
4695 /*
4696 * Then deal with the allocated space we found.
4697 */
4698 ASSERT(ep != NULL);
4699 if (!(flags & XFS_BMAPI_ENTIRE) &&
4700 (got.br_startoff + got.br_blockcount > obno)) {
4701 if (obno > bno)
4702 bno = obno;
4703 ASSERT((bno >= obno) || (n == 0));
4704 ASSERT(bno < end);
4705 mval->br_startoff = bno;
4706 if (isnullstartblock(got.br_startblock)) {
4707 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
4708 mval->br_startblock = DELAYSTARTBLOCK;
4709 } else
4710 mval->br_startblock =
4711 got.br_startblock +
4712 (bno - got.br_startoff);
4713 /*
4714 * Return the minimum of what we got and what we
4715 * asked for for the length. We can use the len
4716 * variable here because it is modified below
4717 * and we could have been there before coming
4718 * here if the first part of the allocation
4719 * didn't overlap what was asked for.
4720 */
4721 mval->br_blockcount =
4722 XFS_FILBLKS_MIN(end - bno, got.br_blockcount -
4723 (bno - got.br_startoff));
4724 mval->br_state = got.br_state;
4725 ASSERT(mval->br_blockcount <= len);
4726 } else {
4727 *mval = got;
4728 if (isnullstartblock(mval->br_startblock)) {
4729 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
4730 mval->br_startblock = DELAYSTARTBLOCK;
4731 }
4732 }
4733
4734 /*
4735 * Check if writing previously allocated but
4736 * unwritten extents.
4737 */
4738 if (wr &&
4739 ((mval->br_state == XFS_EXT_UNWRITTEN &&
4740 ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_DELAY)) == 0)) ||
4741 (mval->br_state == XFS_EXT_NORM &&
4742 ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_CONVERT)) ==
4743 (XFS_BMAPI_PREALLOC|XFS_BMAPI_CONVERT))))) {
4744 /*
4745 * Modify (by adding) the state flag, if writing.
4746 */
4747 ASSERT(mval->br_blockcount <= len);
4748 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
4749 cur = xfs_bmbt_init_cursor(mp,
4750 tp, ip, whichfork);
4751 cur->bc_private.b.firstblock =
4752 *firstblock;
4753 cur->bc_private.b.flist = flist;
4754 }
4755 mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4756 ? XFS_EXT_NORM
4757 : XFS_EXT_UNWRITTEN;
4758 error = xfs_bmap_add_extent(ip, &lastx, &cur, mval,
4759 firstblock, flist, &tmp_logflags,
4760 whichfork);
4761 logflags |= tmp_logflags;
4762 if (error)
4763 goto error0;
4764 ep = xfs_iext_get_ext(ifp, lastx);
4765 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4766 xfs_bmbt_get_all(ep, &got);
4767 /*
4768 * We may have combined previously unwritten
4769 * space with written space, so generate
4770 * another request.
4771 */
4772 if (mval->br_blockcount < len)
4773 continue;
4774 }
4775
4776 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4777 ((mval->br_startoff + mval->br_blockcount) <= end));
4778 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4779 (mval->br_blockcount <= len) ||
4780 (mval->br_startoff < obno));
4781 bno = mval->br_startoff + mval->br_blockcount;
4782 len = end - bno;
4783 if (n > 0 && mval->br_startoff == mval[-1].br_startoff) {
4784 ASSERT(mval->br_startblock == mval[-1].br_startblock);
4785 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
4786 ASSERT(mval->br_state == mval[-1].br_state);
4787 mval[-1].br_blockcount = mval->br_blockcount;
4788 mval[-1].br_state = mval->br_state;
4789 } else if (n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
4790 mval[-1].br_startblock != DELAYSTARTBLOCK &&
4791 mval[-1].br_startblock != HOLESTARTBLOCK &&
4792 mval->br_startblock ==
4793 mval[-1].br_startblock + mval[-1].br_blockcount &&
4794 ((flags & XFS_BMAPI_IGSTATE) ||
4795 mval[-1].br_state == mval->br_state)) {
4796 ASSERT(mval->br_startoff ==
4797 mval[-1].br_startoff + mval[-1].br_blockcount);
4798 mval[-1].br_blockcount += mval->br_blockcount;
4799 } else if (n > 0 &&
4800 mval->br_startblock == DELAYSTARTBLOCK &&
4801 mval[-1].br_startblock == DELAYSTARTBLOCK &&
4802 mval->br_startoff ==
4803 mval[-1].br_startoff + mval[-1].br_blockcount) {
4804 mval[-1].br_blockcount += mval->br_blockcount;
4805 mval[-1].br_state = mval->br_state;
4806 } else if (!((n == 0) &&
4807 ((mval->br_startoff + mval->br_blockcount) <=
4808 obno))) {
4809 mval++;
4810 n++;
4811 }
4812 /*
4813 * If we're done, stop now. Stop when we've allocated
4814 * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise
4815 * the transaction may get too big.
4816 */
4817 if (bno >= end || n >= *nmap || nallocs >= *nmap)
4818 break;
4819 /*
4820 * Else go on to the next record.
4821 */
4822 prev = got;
4823 if (++lastx < nextents) {
4824 ep = xfs_iext_get_ext(ifp, lastx);
4825 xfs_bmbt_get_all(ep, &got);
4826 } else {
4827 eof = 1;
4828 }
4829 }
4830 *nmap = n;
4831 /*
4832 * Transform from btree to extents, give it cur.
4833 */
4834 if (tp && XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
4835 XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
4836 ASSERT(wr && cur);
4837 error = xfs_bmap_btree_to_extents(tp, ip, cur,
4838 &tmp_logflags, whichfork);
4839 logflags |= tmp_logflags;
4840 if (error)
4841 goto error0;
4842 }
4843 ASSERT(ifp->if_ext_max ==
4844 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4845 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4846 XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max);
4847 error = 0;
4848 error0:
4849 /*
4850 * Log everything. Do this after conversion, there's no point in
4851 * logging the extent records if we've converted to btree format.
4852 */
4853 if ((logflags & xfs_ilog_fext(whichfork)) &&
4854 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4855 logflags &= ~xfs_ilog_fext(whichfork);
4856 else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
4857 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4858 logflags &= ~xfs_ilog_fbroot(whichfork);
4859 /*
4860 * Log whatever the flags say, even if error. Otherwise we might miss
4861 * detecting a case where the data is changed, there's an error,
4862 * and it's not logged so we don't shutdown when we should.
4863 */
4864 if (logflags) {
4865 ASSERT(tp && wr);
4866 xfs_trans_log_inode(tp, ip, logflags);
4867 }
4868 if (cur) {
4869 if (!error) {
4870 ASSERT(*firstblock == NULLFSBLOCK ||
4871 XFS_FSB_TO_AGNO(mp, *firstblock) ==
4872 XFS_FSB_TO_AGNO(mp,
4873 cur->bc_private.b.firstblock) ||
4874 (flist->xbf_low &&
4875 XFS_FSB_TO_AGNO(mp, *firstblock) <
4876 XFS_FSB_TO_AGNO(mp,
4877 cur->bc_private.b.firstblock)));
4878 *firstblock = cur->bc_private.b.firstblock;
4879 }
4880 xfs_btree_del_cursor(cur,
4881 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4882 }
4883 if (!error)
4884 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4885 orig_nmap, *nmap);
4886 return error;
4887 }
4888
4889 /*
4890 * Map file blocks to filesystem blocks, simple version.
4891 * One block (extent) only, read-only.
4892 * For flags, only the XFS_BMAPI_ATTRFORK flag is examined.
4893 * For the other flag values, the effect is as if XFS_BMAPI_METADATA
4894 * was set and all the others were clear.
4895 */
4896 int /* error */
4897 xfs_bmapi_single(
4898 xfs_trans_t *tp, /* transaction pointer */
4899 xfs_inode_t *ip, /* incore inode */
4900 int whichfork, /* data or attr fork */
4901 xfs_fsblock_t *fsb, /* output: mapped block */
4902 xfs_fileoff_t bno) /* starting file offs. mapped */
4903 {
4904 int eof; /* we've hit the end of extents */
4905 int error; /* error return */
4906 xfs_bmbt_irec_t got; /* current file extent record */
4907 xfs_ifork_t *ifp; /* inode fork pointer */
4908 xfs_extnum_t lastx; /* last useful extent number */
4909 xfs_bmbt_irec_t prev; /* previous file extent record */
4910
4911 ifp = XFS_IFORK_PTR(ip, whichfork);
4912 if (unlikely(
4913 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4914 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)) {
4915 XFS_ERROR_REPORT("xfs_bmapi_single", XFS_ERRLEVEL_LOW,
4916 ip->i_mount);
4917 return XFS_ERROR(EFSCORRUPTED);
4918 }
4919 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
4920 return XFS_ERROR(EIO);
4921 XFS_STATS_INC(xs_blk_mapr);
4922 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4923 (error = xfs_iread_extents(tp, ip, whichfork)))
4924 return error;
4925 (void)xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4926 &prev);
4927 /*
4928 * Reading past eof, act as though there's a hole
4929 * up to end.
4930 */
4931 if (eof || got.br_startoff > bno) {
4932 *fsb = NULLFSBLOCK;
4933 return 0;
4934 }
4935 ASSERT(!isnullstartblock(got.br_startblock));
4936 ASSERT(bno < got.br_startoff + got.br_blockcount);
4937 *fsb = got.br_startblock + (bno - got.br_startoff);
4938 return 0;
4939 }
4940
4941 /*
4942 * Unmap (remove) blocks from a file.
4943 * If nexts is nonzero then the number of extents to remove is limited to
4944 * that value. If not all extents in the block range can be removed then
4945 * *done is set.
4946 */
4947 int /* error */
4948 xfs_bunmapi(
4949 xfs_trans_t *tp, /* transaction pointer */
4950 struct xfs_inode *ip, /* incore inode */
4951 xfs_fileoff_t bno, /* starting offset to unmap */
4952 xfs_filblks_t len, /* length to unmap in file */
4953 int flags, /* misc flags */
4954 xfs_extnum_t nexts, /* number of extents max */
4955 xfs_fsblock_t *firstblock, /* first allocated block
4956 controls a.g. for allocs */
4957 xfs_bmap_free_t *flist, /* i/o: list extents to free */
4958 int *done) /* set if not done yet */
4959 {
4960 xfs_btree_cur_t *cur; /* bmap btree cursor */
4961 xfs_bmbt_irec_t del; /* extent being deleted */
4962 int eof; /* is deleting at eof */
4963 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
4964 int error; /* error return value */
4965 xfs_extnum_t extno; /* extent number in list */
4966 xfs_bmbt_irec_t got; /* current extent record */
4967 xfs_ifork_t *ifp; /* inode fork pointer */
4968 int isrt; /* freeing in rt area */
4969 xfs_extnum_t lastx; /* last extent index used */
4970 int logflags; /* transaction logging flags */
4971 xfs_extlen_t mod; /* rt extent offset */
4972 xfs_mount_t *mp; /* mount structure */
4973 xfs_extnum_t nextents; /* number of file extents */
4974 xfs_bmbt_irec_t prev; /* previous extent record */
4975 xfs_fileoff_t start; /* first file offset deleted */
4976 int tmp_logflags; /* partial logging flags */
4977 int wasdel; /* was a delayed alloc extent */
4978 int whichfork; /* data or attribute fork */
4979 xfs_fsblock_t sum;
4980
4981 trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
4982
4983 whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4984 XFS_ATTR_FORK : XFS_DATA_FORK;
4985 ifp = XFS_IFORK_PTR(ip, whichfork);
4986 if (unlikely(
4987 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4988 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
4989 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
4990 ip->i_mount);
4991 return XFS_ERROR(EFSCORRUPTED);
4992 }
4993 mp = ip->i_mount;
4994 if (XFS_FORCED_SHUTDOWN(mp))
4995 return XFS_ERROR(EIO);
4996
4997 ASSERT(len > 0);
4998 ASSERT(nexts >= 0);
4999 ASSERT(ifp->if_ext_max ==
5000 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5001 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5002 (error = xfs_iread_extents(tp, ip, whichfork)))
5003 return error;
5004 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5005 if (nextents == 0) {
5006 *done = 1;
5007 return 0;
5008 }
5009 XFS_STATS_INC(xs_blk_unmap);
5010 isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5011 start = bno;
5012 bno = start + len - 1;
5013 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5014 &prev);
5015
5016 /*
5017 * Check to see if the given block number is past the end of the
5018 * file, back up to the last block if so...
5019 */
5020 if (eof) {
5021 ep = xfs_iext_get_ext(ifp, --lastx);
5022 xfs_bmbt_get_all(ep, &got);
5023 bno = got.br_startoff + got.br_blockcount - 1;
5024 }
5025 logflags = 0;
5026 if (ifp->if_flags & XFS_IFBROOT) {
5027 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5028 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5029 cur->bc_private.b.firstblock = *firstblock;
5030 cur->bc_private.b.flist = flist;
5031 cur->bc_private.b.flags = 0;
5032 } else
5033 cur = NULL;
5034 extno = 0;
5035 while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5036 (nexts == 0 || extno < nexts)) {
5037 /*
5038 * Is the found extent after a hole in which bno lives?
5039 * Just back up to the previous extent, if so.
5040 */
5041 if (got.br_startoff > bno) {
5042 if (--lastx < 0)
5043 break;
5044 ep = xfs_iext_get_ext(ifp, lastx);
5045 xfs_bmbt_get_all(ep, &got);
5046 }
5047 /*
5048 * Is the last block of this extent before the range
5049 * we're supposed to delete? If so, we're done.
5050 */
5051 bno = XFS_FILEOFF_MIN(bno,
5052 got.br_startoff + got.br_blockcount - 1);
5053 if (bno < start)
5054 break;
5055 /*
5056 * Then deal with the (possibly delayed) allocated space
5057 * we found.
5058 */
5059 ASSERT(ep != NULL);
5060 del = got;
5061 wasdel = isnullstartblock(del.br_startblock);
5062 if (got.br_startoff < start) {
5063 del.br_startoff = start;
5064 del.br_blockcount -= start - got.br_startoff;
5065 if (!wasdel)
5066 del.br_startblock += start - got.br_startoff;
5067 }
5068 if (del.br_startoff + del.br_blockcount > bno + 1)
5069 del.br_blockcount = bno + 1 - del.br_startoff;
5070 sum = del.br_startblock + del.br_blockcount;
5071 if (isrt &&
5072 (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5073 /*
5074 * Realtime extent not lined up at the end.
5075 * The extent could have been split into written
5076 * and unwritten pieces, or we could just be
5077 * unmapping part of it. But we can't really
5078 * get rid of part of a realtime extent.
5079 */
5080 if (del.br_state == XFS_EXT_UNWRITTEN ||
5081 !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5082 /*
5083 * This piece is unwritten, or we're not
5084 * using unwritten extents. Skip over it.
5085 */
5086 ASSERT(bno >= mod);
5087 bno -= mod > del.br_blockcount ?
5088 del.br_blockcount : mod;
5089 if (bno < got.br_startoff) {
5090 if (--lastx >= 0)
5091 xfs_bmbt_get_all(xfs_iext_get_ext(
5092 ifp, lastx), &got);
5093 }
5094 continue;
5095 }
5096 /*
5097 * It's written, turn it unwritten.
5098 * This is better than zeroing it.
5099 */
5100 ASSERT(del.br_state == XFS_EXT_NORM);
5101 ASSERT(xfs_trans_get_block_res(tp) > 0);
5102 /*
5103 * If this spans a realtime extent boundary,
5104 * chop it back to the start of the one we end at.
5105 */
5106 if (del.br_blockcount > mod) {
5107 del.br_startoff += del.br_blockcount - mod;
5108 del.br_startblock += del.br_blockcount - mod;
5109 del.br_blockcount = mod;
5110 }
5111 del.br_state = XFS_EXT_UNWRITTEN;
5112 error = xfs_bmap_add_extent(ip, &lastx, &cur, &del,
5113 firstblock, flist, &logflags,
5114 XFS_DATA_FORK);
5115 if (error)
5116 goto error0;
5117 goto nodelete;
5118 }
5119 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5120 /*
5121 * Realtime extent is lined up at the end but not
5122 * at the front. We'll get rid of full extents if
5123 * we can.
5124 */
5125 mod = mp->m_sb.sb_rextsize - mod;
5126 if (del.br_blockcount > mod) {
5127 del.br_blockcount -= mod;
5128 del.br_startoff += mod;
5129 del.br_startblock += mod;
5130 } else if ((del.br_startoff == start &&
5131 (del.br_state == XFS_EXT_UNWRITTEN ||
5132 xfs_trans_get_block_res(tp) == 0)) ||
5133 !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5134 /*
5135 * Can't make it unwritten. There isn't
5136 * a full extent here so just skip it.
5137 */
5138 ASSERT(bno >= del.br_blockcount);
5139 bno -= del.br_blockcount;
5140 if (got.br_startoff > bno) {
5141 if (--lastx >= 0) {
5142 ep = xfs_iext_get_ext(ifp,
5143 lastx);
5144 xfs_bmbt_get_all(ep, &got);
5145 }
5146 }
5147 continue;
5148 } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5149 /*
5150 * This one is already unwritten.
5151 * It must have a written left neighbor.
5152 * Unwrite the killed part of that one and
5153 * try again.
5154 */
5155 ASSERT(lastx > 0);
5156 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5157 lastx - 1), &prev);
5158 ASSERT(prev.br_state == XFS_EXT_NORM);
5159 ASSERT(!isnullstartblock(prev.br_startblock));
5160 ASSERT(del.br_startblock ==
5161 prev.br_startblock + prev.br_blockcount);
5162 if (prev.br_startoff < start) {
5163 mod = start - prev.br_startoff;
5164 prev.br_blockcount -= mod;
5165 prev.br_startblock += mod;
5166 prev.br_startoff = start;
5167 }
5168 prev.br_state = XFS_EXT_UNWRITTEN;
5169 lastx--;
5170 error = xfs_bmap_add_extent(ip, &lastx, &cur,
5171 &prev, firstblock, flist, &logflags,
5172 XFS_DATA_FORK);
5173 if (error)
5174 goto error0;
5175 goto nodelete;
5176 } else {
5177 ASSERT(del.br_state == XFS_EXT_NORM);
5178 del.br_state = XFS_EXT_UNWRITTEN;
5179 error = xfs_bmap_add_extent(ip, &lastx, &cur,
5180 &del, firstblock, flist, &logflags,
5181 XFS_DATA_FORK);
5182 if (error)
5183 goto error0;
5184 goto nodelete;
5185 }
5186 }
5187 if (wasdel) {
5188 ASSERT(startblockval(del.br_startblock) > 0);
5189 /* Update realtime/data freespace, unreserve quota */
5190 if (isrt) {
5191 xfs_filblks_t rtexts;
5192
5193 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5194 do_div(rtexts, mp->m_sb.sb_rextsize);
5195 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5196 (int64_t)rtexts, 0);
5197 (void)xfs_trans_reserve_quota_nblks(NULL,
5198 ip, -((long)del.br_blockcount), 0,
5199 XFS_QMOPT_RES_RTBLKS);
5200 } else {
5201 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5202 (int64_t)del.br_blockcount, 0);
5203 (void)xfs_trans_reserve_quota_nblks(NULL,
5204 ip, -((long)del.br_blockcount), 0,
5205 XFS_QMOPT_RES_REGBLKS);
5206 }
5207 ip->i_delayed_blks -= del.br_blockcount;
5208 if (cur)
5209 cur->bc_private.b.flags |=
5210 XFS_BTCUR_BPRV_WASDEL;
5211 } else if (cur)
5212 cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5213 /*
5214 * If it's the case where the directory code is running
5215 * with no block reservation, and the deleted block is in
5216 * the middle of its extent, and the resulting insert
5217 * of an extent would cause transformation to btree format,
5218 * then reject it. The calling code will then swap
5219 * blocks around instead.
5220 * We have to do this now, rather than waiting for the
5221 * conversion to btree format, since the transaction
5222 * will be dirty.
5223 */
5224 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5225 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5226 XFS_IFORK_NEXTENTS(ip, whichfork) >= ifp->if_ext_max &&
5227 del.br_startoff > got.br_startoff &&
5228 del.br_startoff + del.br_blockcount <
5229 got.br_startoff + got.br_blockcount) {
5230 error = XFS_ERROR(ENOSPC);
5231 goto error0;
5232 }
5233 error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5234 &tmp_logflags, whichfork);
5235 logflags |= tmp_logflags;
5236 if (error)
5237 goto error0;
5238 bno = del.br_startoff - 1;
5239 nodelete:
5240 /*
5241 * If not done go on to the next (previous) record.
5242 */
5243 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5244 if (lastx >= 0) {
5245 ep = xfs_iext_get_ext(ifp, lastx);
5246 if (xfs_bmbt_get_startoff(ep) > bno) {
5247 if (--lastx >= 0)
5248 ep = xfs_iext_get_ext(ifp,
5249 lastx);
5250 }
5251 xfs_bmbt_get_all(ep, &got);
5252 }
5253 extno++;
5254 }
5255 }
5256 *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5257 ASSERT(ifp->if_ext_max ==
5258 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5259 /*
5260 * Convert to a btree if necessary.
5261 */
5262 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5263 XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
5264 ASSERT(cur == NULL);
5265 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5266 &cur, 0, &tmp_logflags, whichfork);
5267 logflags |= tmp_logflags;
5268 if (error)
5269 goto error0;
5270 }
5271 /*
5272 * transform from btree to extents, give it cur
5273 */
5274 else if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5275 XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5276 ASSERT(cur != NULL);
5277 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5278 whichfork);
5279 logflags |= tmp_logflags;
5280 if (error)
5281 goto error0;
5282 }
5283 /*
5284 * transform from extents to local?
5285 */
5286 ASSERT(ifp->if_ext_max ==
5287 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5288 error = 0;
5289 error0:
5290 /*
5291 * Log everything. Do this after conversion, there's no point in
5292 * logging the extent records if we've converted to btree format.
5293 */
5294 if ((logflags & xfs_ilog_fext(whichfork)) &&
5295 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5296 logflags &= ~xfs_ilog_fext(whichfork);
5297 else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5298 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5299 logflags &= ~xfs_ilog_fbroot(whichfork);
5300 /*
5301 * Log inode even in the error case, if the transaction
5302 * is dirty we'll need to shut down the filesystem.
5303 */
5304 if (logflags)
5305 xfs_trans_log_inode(tp, ip, logflags);
5306 if (cur) {
5307 if (!error) {
5308 *firstblock = cur->bc_private.b.firstblock;
5309 cur->bc_private.b.allocated = 0;
5310 }
5311 xfs_btree_del_cursor(cur,
5312 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5313 }
5314 return error;
5315 }
5316
5317 /*
5318 * returns 1 for success, 0 if we failed to map the extent.
5319 */
5320 STATIC int
5321 xfs_getbmapx_fix_eof_hole(
5322 xfs_inode_t *ip, /* xfs incore inode pointer */
5323 struct getbmapx *out, /* output structure */
5324 int prealloced, /* this is a file with
5325 * preallocated data space */
5326 __int64_t end, /* last block requested */
5327 xfs_fsblock_t startblock)
5328 {
5329 __int64_t fixlen;
5330 xfs_mount_t *mp; /* file system mount point */
5331 xfs_ifork_t *ifp; /* inode fork pointer */
5332 xfs_extnum_t lastx; /* last extent pointer */
5333 xfs_fileoff_t fileblock;
5334
5335 if (startblock == HOLESTARTBLOCK) {
5336 mp = ip->i_mount;
5337 out->bmv_block = -1;
5338 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, ip->i_size));
5339 fixlen -= out->bmv_offset;
5340 if (prealloced && out->bmv_offset + out->bmv_length == end) {
5341 /* Came to hole at EOF. Trim it. */
5342 if (fixlen <= 0)
5343 return 0;
5344 out->bmv_length = fixlen;
5345 }
5346 } else {
5347 if (startblock == DELAYSTARTBLOCK)
5348 out->bmv_block = -2;
5349 else
5350 out->bmv_block = xfs_fsb_to_db(ip, startblock);
5351 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
5352 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
5353 if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
5354 (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
5355 out->bmv_oflags |= BMV_OF_LAST;
5356 }
5357
5358 return 1;
5359 }
5360
5361 /*
5362 * Get inode's extents as described in bmv, and format for output.
5363 * Calls formatter to fill the user's buffer until all extents
5364 * are mapped, until the passed-in bmv->bmv_count slots have
5365 * been filled, or until the formatter short-circuits the loop,
5366 * if it is tracking filled-in extents on its own.
5367 */
5368 int /* error code */
5369 xfs_getbmap(
5370 xfs_inode_t *ip,
5371 struct getbmapx *bmv, /* user bmap structure */
5372 xfs_bmap_format_t formatter, /* format to user */
5373 void *arg) /* formatter arg */
5374 {
5375 __int64_t bmvend; /* last block requested */
5376 int error = 0; /* return value */
5377 __int64_t fixlen; /* length for -1 case */
5378 int i; /* extent number */
5379 int lock; /* lock state */
5380 xfs_bmbt_irec_t *map; /* buffer for user's data */
5381 xfs_mount_t *mp; /* file system mount point */
5382 int nex; /* # of user extents can do */
5383 int nexleft; /* # of user extents left */
5384 int subnex; /* # of bmapi's can do */
5385 int nmap; /* number of map entries */
5386 struct getbmapx *out; /* output structure */
5387 int whichfork; /* data or attr fork */
5388 int prealloced; /* this is a file with
5389 * preallocated data space */
5390 int iflags; /* interface flags */
5391 int bmapi_flags; /* flags for xfs_bmapi */
5392 int cur_ext = 0;
5393
5394 mp = ip->i_mount;
5395 iflags = bmv->bmv_iflags;
5396 whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5397
5398 if (whichfork == XFS_ATTR_FORK) {
5399 if (XFS_IFORK_Q(ip)) {
5400 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5401 ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5402 ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5403 return XFS_ERROR(EINVAL);
5404 } else if (unlikely(
5405 ip->i_d.di_aformat != 0 &&
5406 ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5407 XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5408 ip->i_mount);
5409 return XFS_ERROR(EFSCORRUPTED);
5410 }
5411
5412 prealloced = 0;
5413 fixlen = 1LL << 32;
5414 } else {
5415 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5416 ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5417 ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5418 return XFS_ERROR(EINVAL);
5419
5420 if (xfs_get_extsz_hint(ip) ||
5421 ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5422 prealloced = 1;
5423 fixlen = XFS_MAXIOFFSET(mp);
5424 } else {
5425 prealloced = 0;
5426 fixlen = ip->i_size;
5427 }
5428 }
5429
5430 if (bmv->bmv_length == -1) {
5431 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5432 bmv->bmv_length =
5433 max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
5434 } else if (bmv->bmv_length == 0) {
5435 bmv->bmv_entries = 0;
5436 return 0;
5437 } else if (bmv->bmv_length < 0) {
5438 return XFS_ERROR(EINVAL);
5439 }
5440
5441 nex = bmv->bmv_count - 1;
5442 if (nex <= 0)
5443 return XFS_ERROR(EINVAL);
5444 bmvend = bmv->bmv_offset + bmv->bmv_length;
5445
5446
5447 if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
5448 return XFS_ERROR(ENOMEM);
5449 out = kmem_zalloc(bmv->bmv_count * sizeof(struct getbmapx), KM_MAYFAIL);
5450 if (!out)
5451 return XFS_ERROR(ENOMEM);
5452
5453 xfs_ilock(ip, XFS_IOLOCK_SHARED);
5454 if (whichfork == XFS_DATA_FORK && !(iflags & BMV_IF_DELALLOC)) {
5455 if (ip->i_delayed_blks || ip->i_size > ip->i_d.di_size) {
5456 error = xfs_flush_pages(ip, 0, -1, 0, FI_REMAPF);
5457 if (error)
5458 goto out_unlock_iolock;
5459 }
5460 /*
5461 * even after flushing the inode, there can still be delalloc
5462 * blocks on the inode beyond EOF due to speculative
5463 * preallocation. These are not removed until the release
5464 * function is called or the inode is inactivated. Hence we
5465 * cannot assert here that ip->i_delayed_blks == 0.
5466 */
5467 }
5468
5469 lock = xfs_ilock_map_shared(ip);
5470
5471 /*
5472 * Don't let nex be bigger than the number of extents
5473 * we can have assuming alternating holes and real extents.
5474 */
5475 if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5476 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5477
5478 bmapi_flags = xfs_bmapi_aflag(whichfork);
5479 if (!(iflags & BMV_IF_PREALLOC))
5480 bmapi_flags |= XFS_BMAPI_IGSTATE;
5481
5482 /*
5483 * Allocate enough space to handle "subnex" maps at a time.
5484 */
5485 error = ENOMEM;
5486 subnex = 16;
5487 map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
5488 if (!map)
5489 goto out_unlock_ilock;
5490
5491 bmv->bmv_entries = 0;
5492
5493 if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
5494 (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
5495 error = 0;
5496 goto out_free_map;
5497 }
5498
5499 nexleft = nex;
5500
5501 do {
5502 nmap = (nexleft > subnex) ? subnex : nexleft;
5503 error = xfs_bmapi(NULL, ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5504 XFS_BB_TO_FSB(mp, bmv->bmv_length),
5505 bmapi_flags, NULL, 0, map, &nmap,
5506 NULL);
5507 if (error)
5508 goto out_free_map;
5509 ASSERT(nmap <= subnex);
5510
5511 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5512 out[cur_ext].bmv_oflags = 0;
5513 if (map[i].br_state == XFS_EXT_UNWRITTEN)
5514 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
5515 else if (map[i].br_startblock == DELAYSTARTBLOCK)
5516 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
5517 out[cur_ext].bmv_offset =
5518 XFS_FSB_TO_BB(mp, map[i].br_startoff);
5519 out[cur_ext].bmv_length =
5520 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5521 out[cur_ext].bmv_unused1 = 0;
5522 out[cur_ext].bmv_unused2 = 0;
5523 ASSERT(((iflags & BMV_IF_DELALLOC) != 0) ||
5524 (map[i].br_startblock != DELAYSTARTBLOCK));
5525 if (map[i].br_startblock == HOLESTARTBLOCK &&
5526 whichfork == XFS_ATTR_FORK) {
5527 /* came to the end of attribute fork */
5528 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
5529 goto out_free_map;
5530 }
5531
5532 if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
5533 prealloced, bmvend,
5534 map[i].br_startblock))
5535 goto out_free_map;
5536
5537 bmv->bmv_offset =
5538 out[cur_ext].bmv_offset +
5539 out[cur_ext].bmv_length;
5540 bmv->bmv_length =
5541 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
5542
5543 /*
5544 * In case we don't want to return the hole,
5545 * don't increase cur_ext so that we can reuse
5546 * it in the next loop.
5547 */
5548 if ((iflags & BMV_IF_NO_HOLES) &&
5549 map[i].br_startblock == HOLESTARTBLOCK) {
5550 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
5551 continue;
5552 }
5553
5554 nexleft--;
5555 bmv->bmv_entries++;
5556 cur_ext++;
5557 }
5558 } while (nmap && nexleft && bmv->bmv_length);
5559
5560 out_free_map:
5561 kmem_free(map);
5562 out_unlock_ilock:
5563 xfs_iunlock_map_shared(ip, lock);
5564 out_unlock_iolock:
5565 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5566
5567 for (i = 0; i < cur_ext; i++) {
5568 int full = 0; /* user array is full */
5569
5570 /* format results & advance arg */
5571 error = formatter(&arg, &out[i], &full);
5572 if (error || full)
5573 break;
5574 }
5575
5576 kmem_free(out);
5577 return error;
5578 }
5579
5580 /*
5581 * Check the last inode extent to determine whether this allocation will result
5582 * in blocks being allocated at the end of the file. When we allocate new data
5583 * blocks at the end of the file which do not start at the previous data block,
5584 * we will try to align the new blocks at stripe unit boundaries.
5585 */
5586 STATIC int /* error */
5587 xfs_bmap_isaeof(
5588 xfs_inode_t *ip, /* incore inode pointer */
5589 xfs_fileoff_t off, /* file offset in fsblocks */
5590 int whichfork, /* data or attribute fork */
5591 char *aeof) /* return value */
5592 {
5593 int error; /* error return value */
5594 xfs_ifork_t *ifp; /* inode fork pointer */
5595 xfs_bmbt_rec_host_t *lastrec; /* extent record pointer */
5596 xfs_extnum_t nextents; /* number of file extents */
5597 xfs_bmbt_irec_t s; /* expanded extent record */
5598
5599 ASSERT(whichfork == XFS_DATA_FORK);
5600 ifp = XFS_IFORK_PTR(ip, whichfork);
5601 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5602 (error = xfs_iread_extents(NULL, ip, whichfork)))
5603 return error;
5604 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5605 if (nextents == 0) {
5606 *aeof = 1;
5607 return 0;
5608 }
5609 /*
5610 * Go to the last extent
5611 */
5612 lastrec = xfs_iext_get_ext(ifp, nextents - 1);
5613 xfs_bmbt_get_all(lastrec, &s);
5614 /*
5615 * Check we are allocating in the last extent (for delayed allocations)
5616 * or past the last extent for non-delayed allocations.
5617 */
5618 *aeof = (off >= s.br_startoff &&
5619 off < s.br_startoff + s.br_blockcount &&
5620 isnullstartblock(s.br_startblock)) ||
5621 off >= s.br_startoff + s.br_blockcount;
5622 return 0;
5623 }
5624
5625 /*
5626 * Check if the endoff is outside the last extent. If so the caller will grow
5627 * the allocation to a stripe unit boundary.
5628 */
5629 int /* error */
5630 xfs_bmap_eof(
5631 xfs_inode_t *ip, /* incore inode pointer */
5632 xfs_fileoff_t endoff, /* file offset in fsblocks */
5633 int whichfork, /* data or attribute fork */
5634 int *eof) /* result value */
5635 {
5636 xfs_fsblock_t blockcount; /* extent block count */
5637 int error; /* error return value */
5638 xfs_ifork_t *ifp; /* inode fork pointer */
5639 xfs_bmbt_rec_host_t *lastrec; /* extent record pointer */
5640 xfs_extnum_t nextents; /* number of file extents */
5641 xfs_fileoff_t startoff; /* extent starting file offset */
5642
5643 ASSERT(whichfork == XFS_DATA_FORK);
5644 ifp = XFS_IFORK_PTR(ip, whichfork);
5645 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5646 (error = xfs_iread_extents(NULL, ip, whichfork)))
5647 return error;
5648 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5649 if (nextents == 0) {
5650 *eof = 1;
5651 return 0;
5652 }
5653 /*
5654 * Go to the last extent
5655 */
5656 lastrec = xfs_iext_get_ext(ifp, nextents - 1);
5657 startoff = xfs_bmbt_get_startoff(lastrec);
5658 blockcount = xfs_bmbt_get_blockcount(lastrec);
5659 *eof = endoff >= startoff + blockcount;
5660 return 0;
5661 }
5662
5663 #ifdef DEBUG
5664 STATIC struct xfs_buf *
5665 xfs_bmap_get_bp(
5666 struct xfs_btree_cur *cur,
5667 xfs_fsblock_t bno)
5668 {
5669 struct xfs_log_item_desc *lidp;
5670 int i;
5671
5672 if (!cur)
5673 return NULL;
5674
5675 for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
5676 if (!cur->bc_bufs[i])
5677 break;
5678 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
5679 return cur->bc_bufs[i];
5680 }
5681
5682 /* Chase down all the log items to see if the bp is there */
5683 list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
5684 struct xfs_buf_log_item *bip;
5685 bip = (struct xfs_buf_log_item *)lidp->lid_item;
5686 if (bip->bli_item.li_type == XFS_LI_BUF &&
5687 XFS_BUF_ADDR(bip->bli_buf) == bno)
5688 return bip->bli_buf;
5689 }
5690
5691 return NULL;
5692 }
5693
5694 STATIC void
5695 xfs_check_block(
5696 struct xfs_btree_block *block,
5697 xfs_mount_t *mp,
5698 int root,
5699 short sz)
5700 {
5701 int i, j, dmxr;
5702 __be64 *pp, *thispa; /* pointer to block address */
5703 xfs_bmbt_key_t *prevp, *keyp;
5704
5705 ASSERT(be16_to_cpu(block->bb_level) > 0);
5706
5707 prevp = NULL;
5708 for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
5709 dmxr = mp->m_bmap_dmxr[0];
5710 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
5711
5712 if (prevp) {
5713 ASSERT(be64_to_cpu(prevp->br_startoff) <
5714 be64_to_cpu(keyp->br_startoff));
5715 }
5716 prevp = keyp;
5717
5718 /*
5719 * Compare the block numbers to see if there are dups.
5720 */
5721 if (root)
5722 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
5723 else
5724 pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
5725
5726 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
5727 if (root)
5728 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
5729 else
5730 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
5731 if (*thispa == *pp) {
5732 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
5733 __func__, j, i,
5734 (unsigned long long)be64_to_cpu(*thispa));
5735 panic("%s: ptrs are equal in node\n",
5736 __func__);
5737 }
5738 }
5739 }
5740 }
5741
5742 /*
5743 * Check that the extents for the inode ip are in the right order in all
5744 * btree leaves.
5745 */
5746
5747 STATIC void
5748 xfs_bmap_check_leaf_extents(
5749 xfs_btree_cur_t *cur, /* btree cursor or null */
5750 xfs_inode_t *ip, /* incore inode pointer */
5751 int whichfork) /* data or attr fork */
5752 {
5753 struct xfs_btree_block *block; /* current btree block */
5754 xfs_fsblock_t bno; /* block # of "block" */
5755 xfs_buf_t *bp; /* buffer for "block" */
5756 int error; /* error return value */
5757 xfs_extnum_t i=0, j; /* index into the extents list */
5758 xfs_ifork_t *ifp; /* fork structure */
5759 int level; /* btree level, for checking */
5760 xfs_mount_t *mp; /* file system mount structure */
5761 __be64 *pp; /* pointer to block address */
5762 xfs_bmbt_rec_t *ep; /* pointer to current extent */
5763 xfs_bmbt_rec_t last = {0, 0}; /* last extent in prev block */
5764 xfs_bmbt_rec_t *nextp; /* pointer to next extent */
5765 int bp_release = 0;
5766
5767 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
5768 return;
5769 }
5770
5771 bno = NULLFSBLOCK;
5772 mp = ip->i_mount;
5773 ifp = XFS_IFORK_PTR(ip, whichfork);
5774 block = ifp->if_broot;
5775 /*
5776 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5777 */
5778 level = be16_to_cpu(block->bb_level);
5779 ASSERT(level > 0);
5780 xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
5781 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5782 bno = be64_to_cpu(*pp);
5783
5784 ASSERT(bno != NULLDFSBNO);
5785 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5786 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5787
5788 /*
5789 * Go down the tree until leaf level is reached, following the first
5790 * pointer (leftmost) at each level.
5791 */
5792 while (level-- > 0) {
5793 /* See if buf is in cur first */
5794 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5795 if (bp) {
5796 bp_release = 0;
5797 } else {
5798 bp_release = 1;
5799 }
5800 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5801 XFS_BMAP_BTREE_REF)))
5802 goto error_norelse;
5803 block = XFS_BUF_TO_BLOCK(bp);
5804 XFS_WANT_CORRUPTED_GOTO(
5805 xfs_bmap_sanity_check(mp, bp, level),
5806 error0);
5807 if (level == 0)
5808 break;
5809
5810 /*
5811 * Check this block for basic sanity (increasing keys and
5812 * no duplicate blocks).
5813 */
5814
5815 xfs_check_block(block, mp, 0, 0);
5816 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
5817 bno = be64_to_cpu(*pp);
5818 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
5819 if (bp_release) {
5820 bp_release = 0;
5821 xfs_trans_brelse(NULL, bp);
5822 }
5823 }
5824
5825 /*
5826 * Here with bp and block set to the leftmost leaf node in the tree.
5827 */
5828 i = 0;
5829
5830 /*
5831 * Loop over all leaf nodes checking that all extents are in the right order.
5832 */
5833 for (;;) {
5834 xfs_fsblock_t nextbno;
5835 xfs_extnum_t num_recs;
5836
5837
5838 num_recs = xfs_btree_get_numrecs(block);
5839
5840 /*
5841 * Read-ahead the next leaf block, if any.
5842 */
5843
5844 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
5845
5846 /*
5847 * Check all the extents to make sure they are OK.
5848 * If we had a previous block, the last entry should
5849 * conform with the first entry in this one.
5850 */
5851
5852 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
5853 if (i) {
5854 ASSERT(xfs_bmbt_disk_get_startoff(&last) +
5855 xfs_bmbt_disk_get_blockcount(&last) <=
5856 xfs_bmbt_disk_get_startoff(ep));
5857 }
5858 for (j = 1; j < num_recs; j++) {
5859 nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
5860 ASSERT(xfs_bmbt_disk_get_startoff(ep) +
5861 xfs_bmbt_disk_get_blockcount(ep) <=
5862 xfs_bmbt_disk_get_startoff(nextp));
5863 ep = nextp;
5864 }
5865
5866 last = *ep;
5867 i += num_recs;
5868 if (bp_release) {
5869 bp_release = 0;
5870 xfs_trans_brelse(NULL, bp);
5871 }
5872 bno = nextbno;
5873 /*
5874 * If we've reached the end, stop.
5875 */
5876 if (bno == NULLFSBLOCK)
5877 break;
5878
5879 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5880 if (bp) {
5881 bp_release = 0;
5882 } else {
5883 bp_release = 1;
5884 }
5885 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5886 XFS_BMAP_BTREE_REF)))
5887 goto error_norelse;
5888 block = XFS_BUF_TO_BLOCK(bp);
5889 }
5890 if (bp_release) {
5891 bp_release = 0;
5892 xfs_trans_brelse(NULL, bp);
5893 }
5894 return;
5895
5896 error0:
5897 xfs_warn(mp, "%s: at error0", __func__);
5898 if (bp_release)
5899 xfs_trans_brelse(NULL, bp);
5900 error_norelse:
5901 xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
5902 __func__, i);
5903 panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
5904 return;
5905 }
5906 #endif
5907
5908 /*
5909 * Count fsblocks of the given fork.
5910 */
5911 int /* error */
5912 xfs_bmap_count_blocks(
5913 xfs_trans_t *tp, /* transaction pointer */
5914 xfs_inode_t *ip, /* incore inode */
5915 int whichfork, /* data or attr fork */
5916 int *count) /* out: count of blocks */
5917 {
5918 struct xfs_btree_block *block; /* current btree block */
5919 xfs_fsblock_t bno; /* block # of "block" */
5920 xfs_ifork_t *ifp; /* fork structure */
5921 int level; /* btree level, for checking */
5922 xfs_mount_t *mp; /* file system mount structure */
5923 __be64 *pp; /* pointer to block address */
5924
5925 bno = NULLFSBLOCK;
5926 mp = ip->i_mount;
5927 ifp = XFS_IFORK_PTR(ip, whichfork);
5928 if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
5929 xfs_bmap_count_leaves(ifp, 0,
5930 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
5931 count);
5932 return 0;
5933 }
5934
5935 /*
5936 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5937 */
5938 block = ifp->if_broot;
5939 level = be16_to_cpu(block->bb_level);
5940 ASSERT(level > 0);
5941 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5942 bno = be64_to_cpu(*pp);
5943 ASSERT(bno != NULLDFSBNO);
5944 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5945 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5946
5947 if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
5948 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
5949 mp);
5950 return XFS_ERROR(EFSCORRUPTED);
5951 }
5952
5953 return 0;
5954 }
5955
5956 /*
5957 * Recursively walks each level of a btree
5958 * to count total fsblocks is use.
5959 */
5960 STATIC int /* error */
5961 xfs_bmap_count_tree(
5962 xfs_mount_t *mp, /* file system mount point */
5963 xfs_trans_t *tp, /* transaction pointer */
5964 xfs_ifork_t *ifp, /* inode fork pointer */
5965 xfs_fsblock_t blockno, /* file system block number */
5966 int levelin, /* level in btree */
5967 int *count) /* Count of blocks */
5968 {
5969 int error;
5970 xfs_buf_t *bp, *nbp;
5971 int level = levelin;
5972 __be64 *pp;
5973 xfs_fsblock_t bno = blockno;
5974 xfs_fsblock_t nextbno;
5975 struct xfs_btree_block *block, *nextblock;
5976 int numrecs;
5977
5978 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
5979 return error;
5980 *count += 1;
5981 block = XFS_BUF_TO_BLOCK(bp);
5982
5983 if (--level) {
5984 /* Not at node above leaves, count this level of nodes */
5985 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
5986 while (nextbno != NULLFSBLOCK) {
5987 if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
5988 0, &nbp, XFS_BMAP_BTREE_REF)))
5989 return error;
5990 *count += 1;
5991 nextblock = XFS_BUF_TO_BLOCK(nbp);
5992 nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
5993 xfs_trans_brelse(tp, nbp);
5994 }
5995
5996 /* Dive to the next level */
5997 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
5998 bno = be64_to_cpu(*pp);
5999 if (unlikely((error =
6000 xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
6001 xfs_trans_brelse(tp, bp);
6002 XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6003 XFS_ERRLEVEL_LOW, mp);
6004 return XFS_ERROR(EFSCORRUPTED);
6005 }
6006 xfs_trans_brelse(tp, bp);
6007 } else {
6008 /* count all level 1 nodes and their leaves */
6009 for (;;) {
6010 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
6011 numrecs = be16_to_cpu(block->bb_numrecs);
6012 xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
6013 xfs_trans_brelse(tp, bp);
6014 if (nextbno == NULLFSBLOCK)
6015 break;
6016 bno = nextbno;
6017 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6018 XFS_BMAP_BTREE_REF)))
6019 return error;
6020 *count += 1;
6021 block = XFS_BUF_TO_BLOCK(bp);
6022 }
6023 }
6024 return 0;
6025 }
6026
6027 /*
6028 * Count leaf blocks given a range of extent records.
6029 */
6030 STATIC void
6031 xfs_bmap_count_leaves(
6032 xfs_ifork_t *ifp,
6033 xfs_extnum_t idx,
6034 int numrecs,
6035 int *count)
6036 {
6037 int b;
6038
6039 for (b = 0; b < numrecs; b++) {
6040 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
6041 *count += xfs_bmbt_get_blockcount(frp);
6042 }
6043 }
6044
6045 /*
6046 * Count leaf blocks given a range of extent records originally
6047 * in btree format.
6048 */
6049 STATIC void
6050 xfs_bmap_disk_count_leaves(
6051 struct xfs_mount *mp,
6052 struct xfs_btree_block *block,
6053 int numrecs,
6054 int *count)
6055 {
6056 int b;
6057 xfs_bmbt_rec_t *frp;
6058
6059 for (b = 1; b <= numrecs; b++) {
6060 frp = XFS_BMBT_REC_ADDR(mp, block, b);
6061 *count += xfs_bmbt_disk_get_blockcount(frp);
6062 }
6063 }
6064
6065 /*
6066 * dead simple method of punching delalyed allocation blocks from a range in
6067 * the inode. Walks a block at a time so will be slow, but is only executed in
6068 * rare error cases so the overhead is not critical. This will alays punch out
6069 * both the start and end blocks, even if the ranges only partially overlap
6070 * them, so it is up to the caller to ensure that partial blocks are not
6071 * passed in.
6072 */
6073 int
6074 xfs_bmap_punch_delalloc_range(
6075 struct xfs_inode *ip,
6076 xfs_fileoff_t start_fsb,
6077 xfs_fileoff_t length)
6078 {
6079 xfs_fileoff_t remaining = length;
6080 int error = 0;
6081
6082 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
6083
6084 do {
6085 int done;
6086 xfs_bmbt_irec_t imap;
6087 int nimaps = 1;
6088 xfs_fsblock_t firstblock;
6089 xfs_bmap_free_t flist;
6090
6091 /*
6092 * Map the range first and check that it is a delalloc extent
6093 * before trying to unmap the range. Otherwise we will be
6094 * trying to remove a real extent (which requires a
6095 * transaction) or a hole, which is probably a bad idea...
6096 */
6097 error = xfs_bmapi(NULL, ip, start_fsb, 1,
6098 XFS_BMAPI_ENTIRE, NULL, 0, &imap,
6099 &nimaps, NULL);
6100
6101 if (error) {
6102 /* something screwed, just bail */
6103 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
6104 xfs_alert(ip->i_mount,
6105 "Failed delalloc mapping lookup ino %lld fsb %lld.",
6106 ip->i_ino, start_fsb);
6107 }
6108 break;
6109 }
6110 if (!nimaps) {
6111 /* nothing there */
6112 goto next_block;
6113 }
6114 if (imap.br_startblock != DELAYSTARTBLOCK) {
6115 /* been converted, ignore */
6116 goto next_block;
6117 }
6118 WARN_ON(imap.br_blockcount == 0);
6119
6120 /*
6121 * Note: while we initialise the firstblock/flist pair, they
6122 * should never be used because blocks should never be
6123 * allocated or freed for a delalloc extent and hence we need
6124 * don't cancel or finish them after the xfs_bunmapi() call.
6125 */
6126 xfs_bmap_init(&flist, &firstblock);
6127 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
6128 &flist, &done);
6129 if (error)
6130 break;
6131
6132 ASSERT(!flist.xbf_count && !flist.xbf_first);
6133 next_block:
6134 start_fsb++;
6135 remaining--;
6136 } while(remaining > 0);
6137
6138 return error;
6139 }
Linux with added FPC-III support