?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
[TCP]: TCP_CONG_YEAH requires TCP_CONG_VEGAS
[linux-2.6/verdex.git] / fs / xfs / xfs_bmap.c
blobb1ea26e40aaf217815cca6a6a8f3ef9340750bab
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_attr_sf.h"
34 #include "xfs_dinode.h"
35 #include "xfs_inode.h"
36 #include "xfs_btree.h"
37 #include "xfs_dmapi.h"
38 #include "xfs_mount.h"
39 #include "xfs_ialloc.h"
40 #include "xfs_itable.h"
41 #include "xfs_dir2_data.h"
42 #include "xfs_dir2_leaf.h"
43 #include "xfs_dir2_block.h"
44 #include "xfs_inode_item.h"
45 #include "xfs_extfree_item.h"
46 #include "xfs_alloc.h"
47 #include "xfs_bmap.h"
48 #include "xfs_rtalloc.h"
49 #include "xfs_error.h"
50 #include "xfs_attr_leaf.h"
51 #include "xfs_rw.h"
52 #include "xfs_quota.h"
53 #include "xfs_trans_space.h"
54 #include "xfs_buf_item.h"
55
56
57 #ifdef DEBUG
58 STATIC void
59 xfs_bmap_check_leaf_extents(xfs_btree_cur_t *cur, xfs_inode_t *ip, int whichfork);
60 #endif
61
62 kmem_zone_t *xfs_bmap_free_item_zone;
63
64 /*
65 * Prototypes for internal bmap routines.
66 */
67
68
69 /*
70 * Called from xfs_bmap_add_attrfork to handle extents format files.
71 */
72 STATIC int /* error */
73 xfs_bmap_add_attrfork_extents(
74 xfs_trans_t *tp, /* transaction pointer */
75 xfs_inode_t *ip, /* incore inode pointer */
76 xfs_fsblock_t *firstblock, /* first block allocated */
77 xfs_bmap_free_t *flist, /* blocks to free at commit */
78 int *flags); /* inode logging flags */
79
80 /*
81 * Called from xfs_bmap_add_attrfork to handle local format files.
82 */
83 STATIC int /* error */
84 xfs_bmap_add_attrfork_local(
85 xfs_trans_t *tp, /* transaction pointer */
86 xfs_inode_t *ip, /* incore inode pointer */
87 xfs_fsblock_t *firstblock, /* first block allocated */
88 xfs_bmap_free_t *flist, /* blocks to free at commit */
89 int *flags); /* inode logging flags */
90
91 /*
92 * Called by xfs_bmapi to update file extent records and the btree
93 * after allocating space (or doing a delayed allocation).
94 */
95 STATIC int /* error */
96 xfs_bmap_add_extent(
97 xfs_inode_t *ip, /* incore inode pointer */
98 xfs_extnum_t idx, /* extent number to update/insert */
99 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
100 xfs_bmbt_irec_t *new, /* new data to add to file extents */
101 xfs_fsblock_t *first, /* pointer to firstblock variable */
102 xfs_bmap_free_t *flist, /* list of extents to be freed */
103 int *logflagsp, /* inode logging flags */
104 xfs_extdelta_t *delta, /* Change made to incore extents */
105 int whichfork, /* data or attr fork */
106 int rsvd); /* OK to allocate reserved blocks */
107
108 /*
109 * Called by xfs_bmap_add_extent to handle cases converting a delayed
110 * allocation to a real allocation.
111 */
112 STATIC int /* error */
113 xfs_bmap_add_extent_delay_real(
114 xfs_inode_t *ip, /* incore inode pointer */
115 xfs_extnum_t idx, /* extent number to update/insert */
116 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
117 xfs_bmbt_irec_t *new, /* new data to add to file extents */
118 xfs_filblks_t *dnew, /* new delayed-alloc indirect blocks */
119 xfs_fsblock_t *first, /* pointer to firstblock variable */
120 xfs_bmap_free_t *flist, /* list of extents to be freed */
121 int *logflagsp, /* inode logging flags */
122 xfs_extdelta_t *delta, /* Change made to incore extents */
123 int rsvd); /* OK to allocate reserved blocks */
124
125 /*
126 * Called by xfs_bmap_add_extent to handle cases converting a hole
127 * to a delayed allocation.
128 */
129 STATIC int /* error */
130 xfs_bmap_add_extent_hole_delay(
131 xfs_inode_t *ip, /* incore inode pointer */
132 xfs_extnum_t idx, /* extent number to update/insert */
133 xfs_bmbt_irec_t *new, /* new data to add to file extents */
134 int *logflagsp,/* inode logging flags */
135 xfs_extdelta_t *delta, /* Change made to incore extents */
136 int rsvd); /* OK to allocate reserved blocks */
137
138 /*
139 * Called by xfs_bmap_add_extent to handle cases converting a hole
140 * to a real allocation.
141 */
142 STATIC int /* error */
143 xfs_bmap_add_extent_hole_real(
144 xfs_inode_t *ip, /* incore inode pointer */
145 xfs_extnum_t idx, /* extent number to update/insert */
146 xfs_btree_cur_t *cur, /* if null, not a btree */
147 xfs_bmbt_irec_t *new, /* new data to add to file extents */
148 int *logflagsp, /* inode logging flags */
149 xfs_extdelta_t *delta, /* Change made to incore extents */
150 int whichfork); /* data or attr fork */
151
152 /*
153 * Called by xfs_bmap_add_extent to handle cases converting an unwritten
154 * allocation to a real allocation or vice versa.
155 */
156 STATIC int /* error */
157 xfs_bmap_add_extent_unwritten_real(
158 xfs_inode_t *ip, /* incore inode pointer */
159 xfs_extnum_t idx, /* extent number to update/insert */
160 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
161 xfs_bmbt_irec_t *new, /* new data to add to file extents */
162 int *logflagsp, /* inode logging flags */
163 xfs_extdelta_t *delta); /* Change made to incore extents */
164
165 /*
166 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
167 * It figures out where to ask the underlying allocator to put the new extent.
168 */
169 STATIC int /* error */
170 xfs_bmap_alloc(
171 xfs_bmalloca_t *ap); /* bmap alloc argument struct */
172
173 /*
174 * Transform a btree format file with only one leaf node, where the
175 * extents list will fit in the inode, into an extents format file.
176 * Since the file extents are already in-core, all we have to do is
177 * give up the space for the btree root and pitch the leaf block.
178 */
179 STATIC int /* error */
180 xfs_bmap_btree_to_extents(
181 xfs_trans_t *tp, /* transaction pointer */
182 xfs_inode_t *ip, /* incore inode pointer */
183 xfs_btree_cur_t *cur, /* btree cursor */
184 int *logflagsp, /* inode logging flags */
185 int whichfork); /* data or attr fork */
186
187 /*
188 * Called by xfs_bmapi to update file extent records and the btree
189 * after removing space (or undoing a delayed allocation).
190 */
191 STATIC int /* error */
192 xfs_bmap_del_extent(
193 xfs_inode_t *ip, /* incore inode pointer */
194 xfs_trans_t *tp, /* current trans pointer */
195 xfs_extnum_t idx, /* extent number to update/insert */
196 xfs_bmap_free_t *flist, /* list of extents to be freed */
197 xfs_btree_cur_t *cur, /* if null, not a btree */
198 xfs_bmbt_irec_t *new, /* new data to add to file extents */
199 int *logflagsp,/* inode logging flags */
200 xfs_extdelta_t *delta, /* Change made to incore extents */
201 int whichfork, /* data or attr fork */
202 int rsvd); /* OK to allocate reserved blocks */
203
204 /*
205 * Remove the entry "free" from the free item list. Prev points to the
206 * previous entry, unless "free" is the head of the list.
207 */
208 STATIC void
209 xfs_bmap_del_free(
210 xfs_bmap_free_t *flist, /* free item list header */
211 xfs_bmap_free_item_t *prev, /* previous item on list, if any */
212 xfs_bmap_free_item_t *free); /* list item to be freed */
213
214 /*
215 * Convert an extents-format file into a btree-format file.
216 * The new file will have a root block (in the inode) and a single child block.
217 */
218 STATIC int /* error */
219 xfs_bmap_extents_to_btree(
220 xfs_trans_t *tp, /* transaction pointer */
221 xfs_inode_t *ip, /* incore inode pointer */
222 xfs_fsblock_t *firstblock, /* first-block-allocated */
223 xfs_bmap_free_t *flist, /* blocks freed in xaction */
224 xfs_btree_cur_t **curp, /* cursor returned to caller */
225 int wasdel, /* converting a delayed alloc */
226 int *logflagsp, /* inode logging flags */
227 int whichfork); /* data or attr fork */
228
229 /*
230 * Convert a local file to an extents file.
231 * This code is sort of bogus, since the file data needs to get
232 * logged so it won't be lost. The bmap-level manipulations are ok, though.
233 */
234 STATIC int /* error */
235 xfs_bmap_local_to_extents(
236 xfs_trans_t *tp, /* transaction pointer */
237 xfs_inode_t *ip, /* incore inode pointer */
238 xfs_fsblock_t *firstblock, /* first block allocated in xaction */
239 xfs_extlen_t total, /* total blocks needed by transaction */
240 int *logflagsp, /* inode logging flags */
241 int whichfork); /* data or attr fork */
242
243 /*
244 * Search the extents list for the inode, for the extent containing bno.
245 * If bno lies in a hole, point to the next entry. If bno lies past eof,
246 * *eofp will be set, and *prevp will contain the last entry (null if none).
247 * Else, *lastxp will be set to the index of the found
248 * entry; *gotp will contain the entry.
249 */
250 STATIC xfs_bmbt_rec_t * /* pointer to found extent entry */
251 xfs_bmap_search_extents(
252 xfs_inode_t *ip, /* incore inode pointer */
253 xfs_fileoff_t bno, /* block number searched for */
254 int whichfork, /* data or attr fork */
255 int *eofp, /* out: end of file found */
256 xfs_extnum_t *lastxp, /* out: last extent index */
257 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
258 xfs_bmbt_irec_t *prevp); /* out: previous extent entry found */
259
260 /*
261 * Check the last inode extent to determine whether this allocation will result
262 * in blocks being allocated at the end of the file. When we allocate new data
263 * blocks at the end of the file which do not start at the previous data block,
264 * we will try to align the new blocks at stripe unit boundaries.
265 */
266 STATIC int /* error */
267 xfs_bmap_isaeof(
268 xfs_inode_t *ip, /* incore inode pointer */
269 xfs_fileoff_t off, /* file offset in fsblocks */
270 int whichfork, /* data or attribute fork */
271 char *aeof); /* return value */
272
273 #ifdef XFS_BMAP_TRACE
274 /*
275 * Add a bmap trace buffer entry. Base routine for the others.
276 */
277 STATIC void
278 xfs_bmap_trace_addentry(
279 int opcode, /* operation */
280 char *fname, /* function name */
281 char *desc, /* operation description */
282 xfs_inode_t *ip, /* incore inode pointer */
283 xfs_extnum_t idx, /* index of entry(ies) */
284 xfs_extnum_t cnt, /* count of entries, 1 or 2 */
285 xfs_bmbt_rec_t *r1, /* first record */
286 xfs_bmbt_rec_t *r2, /* second record or null */
287 int whichfork); /* data or attr fork */
288
289 /*
290 * Add bmap trace entry prior to a call to xfs_iext_remove.
291 */
292 STATIC void
293 xfs_bmap_trace_delete(
294 char *fname, /* function name */
295 char *desc, /* operation description */
296 xfs_inode_t *ip, /* incore inode pointer */
297 xfs_extnum_t idx, /* index of entry(entries) deleted */
298 xfs_extnum_t cnt, /* count of entries deleted, 1 or 2 */
299 int whichfork); /* data or attr fork */
300
301 /*
302 * Add bmap trace entry prior to a call to xfs_iext_insert, or
303 * reading in the extents list from the disk (in the btree).
304 */
305 STATIC void
306 xfs_bmap_trace_insert(
307 char *fname, /* function name */
308 char *desc, /* operation description */
309 xfs_inode_t *ip, /* incore inode pointer */
310 xfs_extnum_t idx, /* index of entry(entries) inserted */
311 xfs_extnum_t cnt, /* count of entries inserted, 1 or 2 */
312 xfs_bmbt_irec_t *r1, /* inserted record 1 */
313 xfs_bmbt_irec_t *r2, /* inserted record 2 or null */
314 int whichfork); /* data or attr fork */
315
316 /*
317 * Add bmap trace entry after updating an extent record in place.
318 */
319 STATIC void
320 xfs_bmap_trace_post_update(
321 char *fname, /* function name */
322 char *desc, /* operation description */
323 xfs_inode_t *ip, /* incore inode pointer */
324 xfs_extnum_t idx, /* index of entry updated */
325 int whichfork); /* data or attr fork */
326
327 /*
328 * Add bmap trace entry prior to updating an extent record in place.
329 */
330 STATIC void
331 xfs_bmap_trace_pre_update(
332 char *fname, /* function name */
333 char *desc, /* operation description */
334 xfs_inode_t *ip, /* incore inode pointer */
335 xfs_extnum_t idx, /* index of entry to be updated */
336 int whichfork); /* data or attr fork */
337
338 #else
339 #define xfs_bmap_trace_delete(f,d,ip,i,c,w)
340 #define xfs_bmap_trace_insert(f,d,ip,i,c,r1,r2,w)
341 #define xfs_bmap_trace_post_update(f,d,ip,i,w)
342 #define xfs_bmap_trace_pre_update(f,d,ip,i,w)
343 #endif /* XFS_BMAP_TRACE */
344
345 /*
346 * Compute the worst-case number of indirect blocks that will be used
347 * for ip's delayed extent of length "len".
348 */
349 STATIC xfs_filblks_t
350 xfs_bmap_worst_indlen(
351 xfs_inode_t *ip, /* incore inode pointer */
352 xfs_filblks_t len); /* delayed extent length */
353
354 #ifdef DEBUG
355 /*
356 * Perform various validation checks on the values being returned
357 * from xfs_bmapi().
358 */
359 STATIC void
360 xfs_bmap_validate_ret(
361 xfs_fileoff_t bno,
362 xfs_filblks_t len,
363 int flags,
364 xfs_bmbt_irec_t *mval,
365 int nmap,
366 int ret_nmap);
367 #else
368 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
369 #endif /* DEBUG */
370
371 #if defined(XFS_RW_TRACE)
372 STATIC void
373 xfs_bunmap_trace(
374 xfs_inode_t *ip,
375 xfs_fileoff_t bno,
376 xfs_filblks_t len,
377 int flags,
378 inst_t *ra);
379 #else
380 #define xfs_bunmap_trace(ip, bno, len, flags, ra)
381 #endif /* XFS_RW_TRACE */
382
383 STATIC int
384 xfs_bmap_count_tree(
385 xfs_mount_t *mp,
386 xfs_trans_t *tp,
387 xfs_ifork_t *ifp,
388 xfs_fsblock_t blockno,
389 int levelin,
390 int *count);
391
392 STATIC int
393 xfs_bmap_count_leaves(
394 xfs_ifork_t *ifp,
395 xfs_extnum_t idx,
396 int numrecs,
397 int *count);
398
399 STATIC int
400 xfs_bmap_disk_count_leaves(
401 xfs_extnum_t idx,
402 xfs_bmbt_block_t *block,
403 int numrecs,
404 int *count);
405
406 /*
407 * Bmap internal routines.
408 */
409
410 /*
411 * Called from xfs_bmap_add_attrfork to handle btree format files.
412 */
413 STATIC int /* error */
414 xfs_bmap_add_attrfork_btree(
415 xfs_trans_t *tp, /* transaction pointer */
416 xfs_inode_t *ip, /* incore inode pointer */
417 xfs_fsblock_t *firstblock, /* first block allocated */
418 xfs_bmap_free_t *flist, /* blocks to free at commit */
419 int *flags) /* inode logging flags */
420 {
421 xfs_btree_cur_t *cur; /* btree cursor */
422 int error; /* error return value */
423 xfs_mount_t *mp; /* file system mount struct */
424 int stat; /* newroot status */
425
426 mp = ip->i_mount;
427 if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
428 *flags |= XFS_ILOG_DBROOT;
429 else {
430 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
431 XFS_DATA_FORK);
432 cur->bc_private.b.flist = flist;
433 cur->bc_private.b.firstblock = *firstblock;
434 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
435 goto error0;
436 ASSERT(stat == 1); /* must be at least one entry */
437 if ((error = xfs_bmbt_newroot(cur, flags, &stat)))
438 goto error0;
439 if (stat == 0) {
440 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
441 return XFS_ERROR(ENOSPC);
442 }
443 *firstblock = cur->bc_private.b.firstblock;
444 cur->bc_private.b.allocated = 0;
445 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
446 }
447 return 0;
448 error0:
449 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
450 return error;
451 }
452
453 /*
454 * Called from xfs_bmap_add_attrfork to handle extents format files.
455 */
456 STATIC int /* error */
457 xfs_bmap_add_attrfork_extents(
458 xfs_trans_t *tp, /* transaction pointer */
459 xfs_inode_t *ip, /* incore inode pointer */
460 xfs_fsblock_t *firstblock, /* first block allocated */
461 xfs_bmap_free_t *flist, /* blocks to free at commit */
462 int *flags) /* inode logging flags */
463 {
464 xfs_btree_cur_t *cur; /* bmap btree cursor */
465 int error; /* error return value */
466
467 if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
468 return 0;
469 cur = NULL;
470 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
471 flags, XFS_DATA_FORK);
472 if (cur) {
473 cur->bc_private.b.allocated = 0;
474 xfs_btree_del_cursor(cur,
475 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
476 }
477 return error;
478 }
479
480 /*
481 * Called from xfs_bmap_add_attrfork to handle local format files.
482 */
483 STATIC int /* error */
484 xfs_bmap_add_attrfork_local(
485 xfs_trans_t *tp, /* transaction pointer */
486 xfs_inode_t *ip, /* incore inode pointer */
487 xfs_fsblock_t *firstblock, /* first block allocated */
488 xfs_bmap_free_t *flist, /* blocks to free at commit */
489 int *flags) /* inode logging flags */
490 {
491 xfs_da_args_t dargs; /* args for dir/attr code */
492 int error; /* error return value */
493 xfs_mount_t *mp; /* mount structure pointer */
494
495 if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
496 return 0;
497 if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
498 mp = ip->i_mount;
499 memset(&dargs, 0, sizeof(dargs));
500 dargs.dp = ip;
501 dargs.firstblock = firstblock;
502 dargs.flist = flist;
503 dargs.total = mp->m_dirblkfsbs;
504 dargs.whichfork = XFS_DATA_FORK;
505 dargs.trans = tp;
506 error = xfs_dir2_sf_to_block(&dargs);
507 } else
508 error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
509 XFS_DATA_FORK);
510 return error;
511 }
512
513 /*
514 * Called by xfs_bmapi to update file extent records and the btree
515 * after allocating space (or doing a delayed allocation).
516 */
517 STATIC int /* error */
518 xfs_bmap_add_extent(
519 xfs_inode_t *ip, /* incore inode pointer */
520 xfs_extnum_t idx, /* extent number to update/insert */
521 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
522 xfs_bmbt_irec_t *new, /* new data to add to file extents */
523 xfs_fsblock_t *first, /* pointer to firstblock variable */
524 xfs_bmap_free_t *flist, /* list of extents to be freed */
525 int *logflagsp, /* inode logging flags */
526 xfs_extdelta_t *delta, /* Change made to incore extents */
527 int whichfork, /* data or attr fork */
528 int rsvd) /* OK to use reserved data blocks */
529 {
530 xfs_btree_cur_t *cur; /* btree cursor or null */
531 xfs_filblks_t da_new; /* new count del alloc blocks used */
532 xfs_filblks_t da_old; /* old count del alloc blocks used */
533 int error; /* error return value */
534 #ifdef XFS_BMAP_TRACE
535 static char fname[] = "xfs_bmap_add_extent";
536 #endif
537 xfs_ifork_t *ifp; /* inode fork ptr */
538 int logflags; /* returned value */
539 xfs_extnum_t nextents; /* number of extents in file now */
540
541 XFS_STATS_INC(xs_add_exlist);
542 cur = *curp;
543 ifp = XFS_IFORK_PTR(ip, whichfork);
544 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
545 ASSERT(idx <= nextents);
546 da_old = da_new = 0;
547 error = 0;
548 /*
549 * This is the first extent added to a new/empty file.
550 * Special case this one, so other routines get to assume there are
551 * already extents in the list.
552 */
553 if (nextents == 0) {
554 xfs_bmap_trace_insert(fname, "insert empty", ip, 0, 1, new,
555 NULL, whichfork);
556 xfs_iext_insert(ifp, 0, 1, new);
557 ASSERT(cur == NULL);
558 ifp->if_lastex = 0;
559 if (!ISNULLSTARTBLOCK(new->br_startblock)) {
560 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
561 logflags = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
562 } else
563 logflags = 0;
564 /* DELTA: single new extent */
565 if (delta) {
566 if (delta->xed_startoff > new->br_startoff)
567 delta->xed_startoff = new->br_startoff;
568 if (delta->xed_blockcount <
569 new->br_startoff + new->br_blockcount)
570 delta->xed_blockcount = new->br_startoff +
571 new->br_blockcount;
572 }
573 }
574 /*
575 * Any kind of new delayed allocation goes here.
576 */
577 else if (ISNULLSTARTBLOCK(new->br_startblock)) {
578 if (cur)
579 ASSERT((cur->bc_private.b.flags &
580 XFS_BTCUR_BPRV_WASDEL) == 0);
581 if ((error = xfs_bmap_add_extent_hole_delay(ip, idx, new,
582 &logflags, delta, rsvd)))
583 goto done;
584 }
585 /*
586 * Real allocation off the end of the file.
587 */
588 else if (idx == nextents) {
589 if (cur)
590 ASSERT((cur->bc_private.b.flags &
591 XFS_BTCUR_BPRV_WASDEL) == 0);
592 if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur, new,
593 &logflags, delta, whichfork)))
594 goto done;
595 } else {
596 xfs_bmbt_irec_t prev; /* old extent at offset idx */
597
598 /*
599 * Get the record referred to by idx.
600 */
601 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx), &prev);
602 /*
603 * If it's a real allocation record, and the new allocation ends
604 * after the start of the referred to record, then we're filling
605 * in a delayed or unwritten allocation with a real one, or
606 * converting real back to unwritten.
607 */
608 if (!ISNULLSTARTBLOCK(new->br_startblock) &&
609 new->br_startoff + new->br_blockcount > prev.br_startoff) {
610 if (prev.br_state != XFS_EXT_UNWRITTEN &&
611 ISNULLSTARTBLOCK(prev.br_startblock)) {
612 da_old = STARTBLOCKVAL(prev.br_startblock);
613 if (cur)
614 ASSERT(cur->bc_private.b.flags &
615 XFS_BTCUR_BPRV_WASDEL);
616 if ((error = xfs_bmap_add_extent_delay_real(ip,
617 idx, &cur, new, &da_new, first, flist,
618 &logflags, delta, rsvd)))
619 goto done;
620 } else if (new->br_state == XFS_EXT_NORM) {
621 ASSERT(new->br_state == XFS_EXT_NORM);
622 if ((error = xfs_bmap_add_extent_unwritten_real(
623 ip, idx, &cur, new, &logflags, delta)))
624 goto done;
625 } else {
626 ASSERT(new->br_state == XFS_EXT_UNWRITTEN);
627 if ((error = xfs_bmap_add_extent_unwritten_real(
628 ip, idx, &cur, new, &logflags, delta)))
629 goto done;
630 }
631 ASSERT(*curp == cur || *curp == NULL);
632 }
633 /*
634 * Otherwise we're filling in a hole with an allocation.
635 */
636 else {
637 if (cur)
638 ASSERT((cur->bc_private.b.flags &
639 XFS_BTCUR_BPRV_WASDEL) == 0);
640 if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur,
641 new, &logflags, delta, whichfork)))
642 goto done;
643 }
644 }
645
646 ASSERT(*curp == cur || *curp == NULL);
647 /*
648 * Convert to a btree if necessary.
649 */
650 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
651 XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
652 int tmp_logflags; /* partial log flag return val */
653
654 ASSERT(cur == NULL);
655 error = xfs_bmap_extents_to_btree(ip->i_transp, ip, first,
656 flist, &cur, da_old > 0, &tmp_logflags, whichfork);
657 logflags |= tmp_logflags;
658 if (error)
659 goto done;
660 }
661 /*
662 * Adjust for changes in reserved delayed indirect blocks.
663 * Nothing to do for disk quotas here.
664 */
665 if (da_old || da_new) {
666 xfs_filblks_t nblks;
667
668 nblks = da_new;
669 if (cur)
670 nblks += cur->bc_private.b.allocated;
671 ASSERT(nblks <= da_old);
672 if (nblks < da_old)
673 xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
674 (int64_t)(da_old - nblks), rsvd);
675 }
676 /*
677 * Clear out the allocated field, done with it now in any case.
678 */
679 if (cur) {
680 cur->bc_private.b.allocated = 0;
681 *curp = cur;
682 }
683 done:
684 #ifdef DEBUG
685 if (!error)
686 xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
687 #endif
688 *logflagsp = logflags;
689 return error;
690 }
691
692 /*
693 * Called by xfs_bmap_add_extent to handle cases converting a delayed
694 * allocation to a real allocation.
695 */
696 STATIC int /* error */
697 xfs_bmap_add_extent_delay_real(
698 xfs_inode_t *ip, /* incore inode pointer */
699 xfs_extnum_t idx, /* extent number to update/insert */
700 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
701 xfs_bmbt_irec_t *new, /* new data to add to file extents */
702 xfs_filblks_t *dnew, /* new delayed-alloc indirect blocks */
703 xfs_fsblock_t *first, /* pointer to firstblock variable */
704 xfs_bmap_free_t *flist, /* list of extents to be freed */
705 int *logflagsp, /* inode logging flags */
706 xfs_extdelta_t *delta, /* Change made to incore extents */
707 int rsvd) /* OK to use reserved data block allocation */
708 {
709 xfs_btree_cur_t *cur; /* btree cursor */
710 int diff; /* temp value */
711 xfs_bmbt_rec_t *ep; /* extent entry for idx */
712 int error; /* error return value */
713 #ifdef XFS_BMAP_TRACE
714 static char fname[] = "xfs_bmap_add_extent_delay_real";
715 #endif
716 int i; /* temp state */
717 xfs_ifork_t *ifp; /* inode fork pointer */
718 xfs_fileoff_t new_endoff; /* end offset of new entry */
719 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
720 /* left is 0, right is 1, prev is 2 */
721 int rval=0; /* return value (logging flags) */
722 int state = 0;/* state bits, accessed thru macros */
723 xfs_filblks_t temp=0; /* value for dnew calculations */
724 xfs_filblks_t temp2=0;/* value for dnew calculations */
725 int tmp_rval; /* partial logging flags */
726 enum { /* bit number definitions for state */
727 LEFT_CONTIG, RIGHT_CONTIG,
728 LEFT_FILLING, RIGHT_FILLING,
729 LEFT_DELAY, RIGHT_DELAY,
730 LEFT_VALID, RIGHT_VALID
731 };
732
733 #define LEFT r[0]
734 #define RIGHT r[1]
735 #define PREV r[2]
736 #define MASK(b) (1 << (b))
737 #define MASK2(a,b) (MASK(a) | MASK(b))
738 #define MASK3(a,b,c) (MASK2(a,b) | MASK(c))
739 #define MASK4(a,b,c,d) (MASK3(a,b,c) | MASK(d))
740 #define STATE_SET(b,v) ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
741 #define STATE_TEST(b) (state & MASK(b))
742 #define STATE_SET_TEST(b,v) ((v) ? ((state |= MASK(b)), 1) : \
743 ((state &= ~MASK(b)), 0))
744 #define SWITCH_STATE \
745 (state & MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG))
746
747 /*
748 * Set up a bunch of variables to make the tests simpler.
749 */
750 cur = *curp;
751 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
752 ep = xfs_iext_get_ext(ifp, idx);
753 xfs_bmbt_get_all(ep, &PREV);
754 new_endoff = new->br_startoff + new->br_blockcount;
755 ASSERT(PREV.br_startoff <= new->br_startoff);
756 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
757 /*
758 * Set flags determining what part of the previous delayed allocation
759 * extent is being replaced by a real allocation.
760 */
761 STATE_SET(LEFT_FILLING, PREV.br_startoff == new->br_startoff);
762 STATE_SET(RIGHT_FILLING,
763 PREV.br_startoff + PREV.br_blockcount == new_endoff);
764 /*
765 * Check and set flags if this segment has a left neighbor.
766 * Don't set contiguous if the combined extent would be too large.
767 */
768 if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
769 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &LEFT);
770 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(LEFT.br_startblock));
771 }
772 STATE_SET(LEFT_CONTIG,
773 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
774 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
775 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
776 LEFT.br_state == new->br_state &&
777 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN);
778 /*
779 * Check and set flags if this segment has a right neighbor.
780 * Don't set contiguous if the combined extent would be too large.
781 * Also check for all-three-contiguous being too large.
782 */
783 if (STATE_SET_TEST(RIGHT_VALID,
784 idx <
785 ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1)) {
786 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx + 1), &RIGHT);
787 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(RIGHT.br_startblock));
788 }
789 STATE_SET(RIGHT_CONTIG,
790 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
791 new_endoff == RIGHT.br_startoff &&
792 new->br_startblock + new->br_blockcount ==
793 RIGHT.br_startblock &&
794 new->br_state == RIGHT.br_state &&
795 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
796 ((state & MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING)) !=
797 MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING) ||
798 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
799 <= MAXEXTLEN));
800 error = 0;
801 /*
802 * Switch out based on the FILLING and CONTIG state bits.
803 */
804 switch (SWITCH_STATE) {
805
806 case MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
807 /*
808 * Filling in all of a previously delayed allocation extent.
809 * The left and right neighbors are both contiguous with new.
810 */
811 xfs_bmap_trace_pre_update(fname, "LF|RF|LC|RC", ip, idx - 1,
812 XFS_DATA_FORK);
813 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
814 LEFT.br_blockcount + PREV.br_blockcount +
815 RIGHT.br_blockcount);
816 xfs_bmap_trace_post_update(fname, "LF|RF|LC|RC", ip, idx - 1,
817 XFS_DATA_FORK);
818 xfs_bmap_trace_delete(fname, "LF|RF|LC|RC", ip, idx, 2,
819 XFS_DATA_FORK);
820 xfs_iext_remove(ifp, idx, 2);
821 ip->i_df.if_lastex = idx - 1;
822 ip->i_d.di_nextents--;
823 if (cur == NULL)
824 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
825 else {
826 rval = XFS_ILOG_CORE;
827 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
828 RIGHT.br_startblock,
829 RIGHT.br_blockcount, &i)))
830 goto done;
831 ASSERT(i == 1);
832 if ((error = xfs_bmbt_delete(cur, &i)))
833 goto done;
834 ASSERT(i == 1);
835 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
836 goto done;
837 ASSERT(i == 1);
838 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
839 LEFT.br_startblock,
840 LEFT.br_blockcount +
841 PREV.br_blockcount +
842 RIGHT.br_blockcount, LEFT.br_state)))
843 goto done;
844 }
845 *dnew = 0;
846 /* DELTA: Three in-core extents are replaced by one. */
847 temp = LEFT.br_startoff;
848 temp2 = LEFT.br_blockcount +
849 PREV.br_blockcount +
850 RIGHT.br_blockcount;
851 break;
852
853 case MASK3(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG):
854 /*
855 * Filling in all of a previously delayed allocation extent.
856 * The left neighbor is contiguous, the right is not.
857 */
858 xfs_bmap_trace_pre_update(fname, "LF|RF|LC", ip, idx - 1,
859 XFS_DATA_FORK);
860 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
861 LEFT.br_blockcount + PREV.br_blockcount);
862 xfs_bmap_trace_post_update(fname, "LF|RF|LC", ip, idx - 1,
863 XFS_DATA_FORK);
864 ip->i_df.if_lastex = idx - 1;
865 xfs_bmap_trace_delete(fname, "LF|RF|LC", ip, idx, 1,
866 XFS_DATA_FORK);
867 xfs_iext_remove(ifp, idx, 1);
868 if (cur == NULL)
869 rval = XFS_ILOG_DEXT;
870 else {
871 rval = 0;
872 if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
873 LEFT.br_startblock, LEFT.br_blockcount,
874 &i)))
875 goto done;
876 ASSERT(i == 1);
877 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
878 LEFT.br_startblock,
879 LEFT.br_blockcount +
880 PREV.br_blockcount, LEFT.br_state)))
881 goto done;
882 }
883 *dnew = 0;
884 /* DELTA: Two in-core extents are replaced by one. */
885 temp = LEFT.br_startoff;
886 temp2 = LEFT.br_blockcount +
887 PREV.br_blockcount;
888 break;
889
890 case MASK3(LEFT_FILLING, RIGHT_FILLING, RIGHT_CONTIG):
891 /*
892 * Filling in all of a previously delayed allocation extent.
893 * The right neighbor is contiguous, the left is not.
894 */
895 xfs_bmap_trace_pre_update(fname, "LF|RF|RC", ip, idx,
896 XFS_DATA_FORK);
897 xfs_bmbt_set_startblock(ep, new->br_startblock);
898 xfs_bmbt_set_blockcount(ep,
899 PREV.br_blockcount + RIGHT.br_blockcount);
900 xfs_bmap_trace_post_update(fname, "LF|RF|RC", ip, idx,
901 XFS_DATA_FORK);
902 ip->i_df.if_lastex = idx;
903 xfs_bmap_trace_delete(fname, "LF|RF|RC", ip, idx + 1, 1,
904 XFS_DATA_FORK);
905 xfs_iext_remove(ifp, idx + 1, 1);
906 if (cur == NULL)
907 rval = XFS_ILOG_DEXT;
908 else {
909 rval = 0;
910 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
911 RIGHT.br_startblock,
912 RIGHT.br_blockcount, &i)))
913 goto done;
914 ASSERT(i == 1);
915 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
916 new->br_startblock,
917 PREV.br_blockcount +
918 RIGHT.br_blockcount, PREV.br_state)))
919 goto done;
920 }
921 *dnew = 0;
922 /* DELTA: Two in-core extents are replaced by one. */
923 temp = PREV.br_startoff;
924 temp2 = PREV.br_blockcount +
925 RIGHT.br_blockcount;
926 break;
927
928 case MASK2(LEFT_FILLING, RIGHT_FILLING):
929 /*
930 * Filling in all of a previously delayed allocation extent.
931 * Neither the left nor right neighbors are contiguous with
932 * the new one.
933 */
934 xfs_bmap_trace_pre_update(fname, "LF|RF", ip, idx,
935 XFS_DATA_FORK);
936 xfs_bmbt_set_startblock(ep, new->br_startblock);
937 xfs_bmap_trace_post_update(fname, "LF|RF", ip, idx,
938 XFS_DATA_FORK);
939 ip->i_df.if_lastex = idx;
940 ip->i_d.di_nextents++;
941 if (cur == NULL)
942 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
943 else {
944 rval = XFS_ILOG_CORE;
945 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
946 new->br_startblock, new->br_blockcount,
947 &i)))
948 goto done;
949 ASSERT(i == 0);
950 cur->bc_rec.b.br_state = XFS_EXT_NORM;
951 if ((error = xfs_bmbt_insert(cur, &i)))
952 goto done;
953 ASSERT(i == 1);
954 }
955 *dnew = 0;
956 /* DELTA: The in-core extent described by new changed type. */
957 temp = new->br_startoff;
958 temp2 = new->br_blockcount;
959 break;
960
961 case MASK2(LEFT_FILLING, LEFT_CONTIG):
962 /*
963 * Filling in the first part of a previous delayed allocation.
964 * The left neighbor is contiguous.
965 */
966 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx - 1,
967 XFS_DATA_FORK);
968 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
969 LEFT.br_blockcount + new->br_blockcount);
970 xfs_bmbt_set_startoff(ep,
971 PREV.br_startoff + new->br_blockcount);
972 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx - 1,
973 XFS_DATA_FORK);
974 temp = PREV.br_blockcount - new->br_blockcount;
975 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx,
976 XFS_DATA_FORK);
977 xfs_bmbt_set_blockcount(ep, temp);
978 ip->i_df.if_lastex = idx - 1;
979 if (cur == NULL)
980 rval = XFS_ILOG_DEXT;
981 else {
982 rval = 0;
983 if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
984 LEFT.br_startblock, LEFT.br_blockcount,
985 &i)))
986 goto done;
987 ASSERT(i == 1);
988 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
989 LEFT.br_startblock,
990 LEFT.br_blockcount +
991 new->br_blockcount,
992 LEFT.br_state)))
993 goto done;
994 }
995 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
996 STARTBLOCKVAL(PREV.br_startblock));
997 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
998 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx,
999 XFS_DATA_FORK);
1000 *dnew = temp;
1001 /* DELTA: The boundary between two in-core extents moved. */
1002 temp = LEFT.br_startoff;
1003 temp2 = LEFT.br_blockcount +
1004 PREV.br_blockcount;
1005 break;
1006
1007 case MASK(LEFT_FILLING):
1008 /*
1009 * Filling in the first part of a previous delayed allocation.
1010 * The left neighbor is not contiguous.
1011 */
1012 xfs_bmap_trace_pre_update(fname, "LF", ip, idx, XFS_DATA_FORK);
1013 xfs_bmbt_set_startoff(ep, new_endoff);
1014 temp = PREV.br_blockcount - new->br_blockcount;
1015 xfs_bmbt_set_blockcount(ep, temp);
1016 xfs_bmap_trace_insert(fname, "LF", ip, idx, 1, new, NULL,
1017 XFS_DATA_FORK);
1018 xfs_iext_insert(ifp, idx, 1, new);
1019 ip->i_df.if_lastex = idx;
1020 ip->i_d.di_nextents++;
1021 if (cur == NULL)
1022 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1023 else {
1024 rval = XFS_ILOG_CORE;
1025 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1026 new->br_startblock, new->br_blockcount,
1027 &i)))
1028 goto done;
1029 ASSERT(i == 0);
1030 cur->bc_rec.b.br_state = XFS_EXT_NORM;
1031 if ((error = xfs_bmbt_insert(cur, &i)))
1032 goto done;
1033 ASSERT(i == 1);
1034 }
1035 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1036 ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1037 error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1038 first, flist, &cur, 1, &tmp_rval,
1039 XFS_DATA_FORK);
1040 rval |= tmp_rval;
1041 if (error)
1042 goto done;
1043 }
1044 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1045 STARTBLOCKVAL(PREV.br_startblock) -
1046 (cur ? cur->bc_private.b.allocated : 0));
1047 ep = xfs_iext_get_ext(ifp, idx + 1);
1048 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1049 xfs_bmap_trace_post_update(fname, "LF", ip, idx + 1,
1050 XFS_DATA_FORK);
1051 *dnew = temp;
1052 /* DELTA: One in-core extent is split in two. */
1053 temp = PREV.br_startoff;
1054 temp2 = PREV.br_blockcount;
1055 break;
1056
1057 case MASK2(RIGHT_FILLING, RIGHT_CONTIG):
1058 /*
1059 * Filling in the last part of a previous delayed allocation.
1060 * The right neighbor is contiguous with the new allocation.
1061 */
1062 temp = PREV.br_blockcount - new->br_blockcount;
1063 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx,
1064 XFS_DATA_FORK);
1065 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx + 1,
1066 XFS_DATA_FORK);
1067 xfs_bmbt_set_blockcount(ep, temp);
1068 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, idx + 1),
1069 new->br_startoff, new->br_startblock,
1070 new->br_blockcount + RIGHT.br_blockcount,
1071 RIGHT.br_state);
1072 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx + 1,
1073 XFS_DATA_FORK);
1074 ip->i_df.if_lastex = idx + 1;
1075 if (cur == NULL)
1076 rval = XFS_ILOG_DEXT;
1077 else {
1078 rval = 0;
1079 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1080 RIGHT.br_startblock,
1081 RIGHT.br_blockcount, &i)))
1082 goto done;
1083 ASSERT(i == 1);
1084 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1085 new->br_startblock,
1086 new->br_blockcount +
1087 RIGHT.br_blockcount,
1088 RIGHT.br_state)))
1089 goto done;
1090 }
1091 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1092 STARTBLOCKVAL(PREV.br_startblock));
1093 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1094 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx,
1095 XFS_DATA_FORK);
1096 *dnew = temp;
1097 /* DELTA: The boundary between two in-core extents moved. */
1098 temp = PREV.br_startoff;
1099 temp2 = PREV.br_blockcount +
1100 RIGHT.br_blockcount;
1101 break;
1102
1103 case MASK(RIGHT_FILLING):
1104 /*
1105 * Filling in the last part of a previous delayed allocation.
1106 * The right neighbor is not contiguous.
1107 */
1108 temp = PREV.br_blockcount - new->br_blockcount;
1109 xfs_bmap_trace_pre_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1110 xfs_bmbt_set_blockcount(ep, temp);
1111 xfs_bmap_trace_insert(fname, "RF", ip, idx + 1, 1,
1112 new, NULL, XFS_DATA_FORK);
1113 xfs_iext_insert(ifp, idx + 1, 1, new);
1114 ip->i_df.if_lastex = idx + 1;
1115 ip->i_d.di_nextents++;
1116 if (cur == NULL)
1117 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1118 else {
1119 rval = XFS_ILOG_CORE;
1120 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1121 new->br_startblock, new->br_blockcount,
1122 &i)))
1123 goto done;
1124 ASSERT(i == 0);
1125 cur->bc_rec.b.br_state = XFS_EXT_NORM;
1126 if ((error = xfs_bmbt_insert(cur, &i)))
1127 goto done;
1128 ASSERT(i == 1);
1129 }
1130 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1131 ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1132 error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1133 first, flist, &cur, 1, &tmp_rval,
1134 XFS_DATA_FORK);
1135 rval |= tmp_rval;
1136 if (error)
1137 goto done;
1138 }
1139 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1140 STARTBLOCKVAL(PREV.br_startblock) -
1141 (cur ? cur->bc_private.b.allocated : 0));
1142 ep = xfs_iext_get_ext(ifp, idx);
1143 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1144 xfs_bmap_trace_post_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1145 *dnew = temp;
1146 /* DELTA: One in-core extent is split in two. */
1147 temp = PREV.br_startoff;
1148 temp2 = PREV.br_blockcount;
1149 break;
1150
1151 case 0:
1152 /*
1153 * Filling in the middle part of a previous delayed allocation.
1154 * Contiguity is impossible here.
1155 * This case is avoided almost all the time.
1156 */
1157 temp = new->br_startoff - PREV.br_startoff;
1158 xfs_bmap_trace_pre_update(fname, "0", ip, idx, XFS_DATA_FORK);
1159 xfs_bmbt_set_blockcount(ep, temp);
1160 r[0] = *new;
1161 r[1].br_state = PREV.br_state;
1162 r[1].br_startblock = 0;
1163 r[1].br_startoff = new_endoff;
1164 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
1165 r[1].br_blockcount = temp2;
1166 xfs_bmap_trace_insert(fname, "0", ip, idx + 1, 2, &r[0], &r[1],
1167 XFS_DATA_FORK);
1168 xfs_iext_insert(ifp, idx + 1, 2, &r[0]);
1169 ip->i_df.if_lastex = idx + 1;
1170 ip->i_d.di_nextents++;
1171 if (cur == NULL)
1172 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1173 else {
1174 rval = XFS_ILOG_CORE;
1175 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1176 new->br_startblock, new->br_blockcount,
1177 &i)))
1178 goto done;
1179 ASSERT(i == 0);
1180 cur->bc_rec.b.br_state = XFS_EXT_NORM;
1181 if ((error = xfs_bmbt_insert(cur, &i)))
1182 goto done;
1183 ASSERT(i == 1);
1184 }
1185 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1186 ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1187 error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1188 first, flist, &cur, 1, &tmp_rval,
1189 XFS_DATA_FORK);
1190 rval |= tmp_rval;
1191 if (error)
1192 goto done;
1193 }
1194 temp = xfs_bmap_worst_indlen(ip, temp);
1195 temp2 = xfs_bmap_worst_indlen(ip, temp2);
1196 diff = (int)(temp + temp2 - STARTBLOCKVAL(PREV.br_startblock) -
1197 (cur ? cur->bc_private.b.allocated : 0));
1198 if (diff > 0 &&
1199 xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS, -((int64_t)diff), rsvd)) {
1200 /*
1201 * Ick gross gag me with a spoon.
1202 */
1203 ASSERT(0); /* want to see if this ever happens! */
1204 while (diff > 0) {
1205 if (temp) {
1206 temp--;
1207 diff--;
1208 if (!diff ||
1209 !xfs_mod_incore_sb(ip->i_mount,
1210 XFS_SBS_FDBLOCKS, -((int64_t)diff), rsvd))
1211 break;
1212 }
1213 if (temp2) {
1214 temp2--;
1215 diff--;
1216 if (!diff ||
1217 !xfs_mod_incore_sb(ip->i_mount,
1218 XFS_SBS_FDBLOCKS, -((int64_t)diff), rsvd))
1219 break;
1220 }
1221 }
1222 }
1223 ep = xfs_iext_get_ext(ifp, idx);
1224 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1225 xfs_bmap_trace_post_update(fname, "0", ip, idx, XFS_DATA_FORK);
1226 xfs_bmap_trace_pre_update(fname, "0", ip, idx + 2,
1227 XFS_DATA_FORK);
1228 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx + 2),
1229 NULLSTARTBLOCK((int)temp2));
1230 xfs_bmap_trace_post_update(fname, "0", ip, idx + 2,
1231 XFS_DATA_FORK);
1232 *dnew = temp + temp2;
1233 /* DELTA: One in-core extent is split in three. */
1234 temp = PREV.br_startoff;
1235 temp2 = PREV.br_blockcount;
1236 break;
1237
1238 case MASK3(LEFT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1239 case MASK3(RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1240 case MASK2(LEFT_FILLING, RIGHT_CONTIG):
1241 case MASK2(RIGHT_FILLING, LEFT_CONTIG):
1242 case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1243 case MASK(LEFT_CONTIG):
1244 case MASK(RIGHT_CONTIG):
1245 /*
1246 * These cases are all impossible.
1247 */
1248 ASSERT(0);
1249 }
1250 *curp = cur;
1251 if (delta) {
1252 temp2 += temp;
1253 if (delta->xed_startoff > temp)
1254 delta->xed_startoff = temp;
1255 if (delta->xed_blockcount < temp2)
1256 delta->xed_blockcount = temp2;
1257 }
1258 done:
1259 *logflagsp = rval;
1260 return error;
1261 #undef LEFT
1262 #undef RIGHT
1263 #undef PREV
1264 #undef MASK
1265 #undef MASK2
1266 #undef MASK3
1267 #undef MASK4
1268 #undef STATE_SET
1269 #undef STATE_TEST
1270 #undef STATE_SET_TEST
1271 #undef SWITCH_STATE
1272 }
1273
1274 /*
1275 * Called by xfs_bmap_add_extent to handle cases converting an unwritten
1276 * allocation to a real allocation or vice versa.
1277 */
1278 STATIC int /* error */
1279 xfs_bmap_add_extent_unwritten_real(
1280 xfs_inode_t *ip, /* incore inode pointer */
1281 xfs_extnum_t idx, /* extent number to update/insert */
1282 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
1283 xfs_bmbt_irec_t *new, /* new data to add to file extents */
1284 int *logflagsp, /* inode logging flags */
1285 xfs_extdelta_t *delta) /* Change made to incore extents */
1286 {
1287 xfs_btree_cur_t *cur; /* btree cursor */
1288 xfs_bmbt_rec_t *ep; /* extent entry for idx */
1289 int error; /* error return value */
1290 #ifdef XFS_BMAP_TRACE
1291 static char fname[] = "xfs_bmap_add_extent_unwritten_real";
1292 #endif
1293 int i; /* temp state */
1294 xfs_ifork_t *ifp; /* inode fork pointer */
1295 xfs_fileoff_t new_endoff; /* end offset of new entry */
1296 xfs_exntst_t newext; /* new extent state */
1297 xfs_exntst_t oldext; /* old extent state */
1298 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
1299 /* left is 0, right is 1, prev is 2 */
1300 int rval=0; /* return value (logging flags) */
1301 int state = 0;/* state bits, accessed thru macros */
1302 xfs_filblks_t temp=0;
1303 xfs_filblks_t temp2=0;
1304 enum { /* bit number definitions for state */
1305 LEFT_CONTIG, RIGHT_CONTIG,
1306 LEFT_FILLING, RIGHT_FILLING,
1307 LEFT_DELAY, RIGHT_DELAY,
1308 LEFT_VALID, RIGHT_VALID
1309 };
1310
1311 #define LEFT r[0]
1312 #define RIGHT r[1]
1313 #define PREV r[2]
1314 #define MASK(b) (1 << (b))
1315 #define MASK2(a,b) (MASK(a) | MASK(b))
1316 #define MASK3(a,b,c) (MASK2(a,b) | MASK(c))
1317 #define MASK4(a,b,c,d) (MASK3(a,b,c) | MASK(d))
1318 #define STATE_SET(b,v) ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1319 #define STATE_TEST(b) (state & MASK(b))
1320 #define STATE_SET_TEST(b,v) ((v) ? ((state |= MASK(b)), 1) : \
1321 ((state &= ~MASK(b)), 0))
1322 #define SWITCH_STATE \
1323 (state & MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG))
1324
1325 /*
1326 * Set up a bunch of variables to make the tests simpler.
1327 */
1328 error = 0;
1329 cur = *curp;
1330 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1331 ep = xfs_iext_get_ext(ifp, idx);
1332 xfs_bmbt_get_all(ep, &PREV);
1333 newext = new->br_state;
1334 oldext = (newext == XFS_EXT_UNWRITTEN) ?
1335 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
1336 ASSERT(PREV.br_state == oldext);
1337 new_endoff = new->br_startoff + new->br_blockcount;
1338 ASSERT(PREV.br_startoff <= new->br_startoff);
1339 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1340 /*
1341 * Set flags determining what part of the previous oldext allocation
1342 * extent is being replaced by a newext allocation.
1343 */
1344 STATE_SET(LEFT_FILLING, PREV.br_startoff == new->br_startoff);
1345 STATE_SET(RIGHT_FILLING,
1346 PREV.br_startoff + PREV.br_blockcount == new_endoff);
1347 /*
1348 * Check and set flags if this segment has a left neighbor.
1349 * Don't set contiguous if the combined extent would be too large.
1350 */
1351 if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1352 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &LEFT);
1353 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(LEFT.br_startblock));
1354 }
1355 STATE_SET(LEFT_CONTIG,
1356 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
1357 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1358 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1359 LEFT.br_state == newext &&
1360 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1361 /*
1362 * Check and set flags if this segment has a right neighbor.
1363 * Don't set contiguous if the combined extent would be too large.
1364 * Also check for all-three-contiguous being too large.
1365 */
1366 if (STATE_SET_TEST(RIGHT_VALID,
1367 idx <
1368 ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1)) {
1369 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx + 1), &RIGHT);
1370 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(RIGHT.br_startblock));
1371 }
1372 STATE_SET(RIGHT_CONTIG,
1373 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
1374 new_endoff == RIGHT.br_startoff &&
1375 new->br_startblock + new->br_blockcount ==
1376 RIGHT.br_startblock &&
1377 newext == RIGHT.br_state &&
1378 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1379 ((state & MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING)) !=
1380 MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING) ||
1381 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1382 <= MAXEXTLEN));
1383 /*
1384 * Switch out based on the FILLING and CONTIG state bits.
1385 */
1386 switch (SWITCH_STATE) {
1387
1388 case MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1389 /*
1390 * Setting all of a previous oldext extent to newext.
1391 * The left and right neighbors are both contiguous with new.
1392 */
1393 xfs_bmap_trace_pre_update(fname, "LF|RF|LC|RC", ip, idx - 1,
1394 XFS_DATA_FORK);
1395 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1396 LEFT.br_blockcount + PREV.br_blockcount +
1397 RIGHT.br_blockcount);
1398 xfs_bmap_trace_post_update(fname, "LF|RF|LC|RC", ip, idx - 1,
1399 XFS_DATA_FORK);
1400 xfs_bmap_trace_delete(fname, "LF|RF|LC|RC", ip, idx, 2,
1401 XFS_DATA_FORK);
1402 xfs_iext_remove(ifp, idx, 2);
1403 ip->i_df.if_lastex = idx - 1;
1404 ip->i_d.di_nextents -= 2;
1405 if (cur == NULL)
1406 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1407 else {
1408 rval = XFS_ILOG_CORE;
1409 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1410 RIGHT.br_startblock,
1411 RIGHT.br_blockcount, &i)))
1412 goto done;
1413 ASSERT(i == 1);
1414 if ((error = xfs_bmbt_delete(cur, &i)))
1415 goto done;
1416 ASSERT(i == 1);
1417 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1418 goto done;
1419 ASSERT(i == 1);
1420 if ((error = xfs_bmbt_delete(cur, &i)))
1421 goto done;
1422 ASSERT(i == 1);
1423 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1424 goto done;
1425 ASSERT(i == 1);
1426 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1427 LEFT.br_startblock,
1428 LEFT.br_blockcount + PREV.br_blockcount +
1429 RIGHT.br_blockcount, LEFT.br_state)))
1430 goto done;
1431 }
1432 /* DELTA: Three in-core extents are replaced by one. */
1433 temp = LEFT.br_startoff;
1434 temp2 = LEFT.br_blockcount +
1435 PREV.br_blockcount +
1436 RIGHT.br_blockcount;
1437 break;
1438
1439 case MASK3(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG):
1440 /*
1441 * Setting all of a previous oldext extent to newext.
1442 * The left neighbor is contiguous, the right is not.
1443 */
1444 xfs_bmap_trace_pre_update(fname, "LF|RF|LC", ip, idx - 1,
1445 XFS_DATA_FORK);
1446 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1447 LEFT.br_blockcount + PREV.br_blockcount);
1448 xfs_bmap_trace_post_update(fname, "LF|RF|LC", ip, idx - 1,
1449 XFS_DATA_FORK);
1450 ip->i_df.if_lastex = idx - 1;
1451 xfs_bmap_trace_delete(fname, "LF|RF|LC", ip, idx, 1,
1452 XFS_DATA_FORK);
1453 xfs_iext_remove(ifp, idx, 1);
1454 ip->i_d.di_nextents--;
1455 if (cur == NULL)
1456 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1457 else {
1458 rval = XFS_ILOG_CORE;
1459 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1460 PREV.br_startblock, PREV.br_blockcount,
1461 &i)))
1462 goto done;
1463 ASSERT(i == 1);
1464 if ((error = xfs_bmbt_delete(cur, &i)))
1465 goto done;
1466 ASSERT(i == 1);
1467 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1468 goto done;
1469 ASSERT(i == 1);
1470 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1471 LEFT.br_startblock,
1472 LEFT.br_blockcount + PREV.br_blockcount,
1473 LEFT.br_state)))
1474 goto done;
1475 }
1476 /* DELTA: Two in-core extents are replaced by one. */
1477 temp = LEFT.br_startoff;
1478 temp2 = LEFT.br_blockcount +
1479 PREV.br_blockcount;
1480 break;
1481
1482 case MASK3(LEFT_FILLING, RIGHT_FILLING, RIGHT_CONTIG):
1483 /*
1484 * Setting all of a previous oldext extent to newext.
1485 * The right neighbor is contiguous, the left is not.
1486 */
1487 xfs_bmap_trace_pre_update(fname, "LF|RF|RC", ip, idx,
1488 XFS_DATA_FORK);
1489 xfs_bmbt_set_blockcount(ep,
1490 PREV.br_blockcount + RIGHT.br_blockcount);
1491 xfs_bmbt_set_state(ep, newext);
1492 xfs_bmap_trace_post_update(fname, "LF|RF|RC", ip, idx,
1493 XFS_DATA_FORK);
1494 ip->i_df.if_lastex = idx;
1495 xfs_bmap_trace_delete(fname, "LF|RF|RC", ip, idx + 1, 1,
1496 XFS_DATA_FORK);
1497 xfs_iext_remove(ifp, idx + 1, 1);
1498 ip->i_d.di_nextents--;
1499 if (cur == NULL)
1500 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1501 else {
1502 rval = XFS_ILOG_CORE;
1503 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1504 RIGHT.br_startblock,
1505 RIGHT.br_blockcount, &i)))
1506 goto done;
1507 ASSERT(i == 1);
1508 if ((error = xfs_bmbt_delete(cur, &i)))
1509 goto done;
1510 ASSERT(i == 1);
1511 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1512 goto done;
1513 ASSERT(i == 1);
1514 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1515 new->br_startblock,
1516 new->br_blockcount + RIGHT.br_blockcount,
1517 newext)))
1518 goto done;
1519 }
1520 /* DELTA: Two in-core extents are replaced by one. */
1521 temp = PREV.br_startoff;
1522 temp2 = PREV.br_blockcount +
1523 RIGHT.br_blockcount;
1524 break;
1525
1526 case MASK2(LEFT_FILLING, RIGHT_FILLING):
1527 /*
1528 * Setting all of a previous oldext extent to newext.
1529 * Neither the left nor right neighbors are contiguous with
1530 * the new one.
1531 */
1532 xfs_bmap_trace_pre_update(fname, "LF|RF", ip, idx,
1533 XFS_DATA_FORK);
1534 xfs_bmbt_set_state(ep, newext);
1535 xfs_bmap_trace_post_update(fname, "LF|RF", ip, idx,
1536 XFS_DATA_FORK);
1537 ip->i_df.if_lastex = idx;
1538 if (cur == NULL)
1539 rval = XFS_ILOG_DEXT;
1540 else {
1541 rval = 0;
1542 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1543 new->br_startblock, new->br_blockcount,
1544 &i)))
1545 goto done;
1546 ASSERT(i == 1);
1547 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1548 new->br_startblock, new->br_blockcount,
1549 newext)))
1550 goto done;
1551 }
1552 /* DELTA: The in-core extent described by new changed type. */
1553 temp = new->br_startoff;
1554 temp2 = new->br_blockcount;
1555 break;
1556
1557 case MASK2(LEFT_FILLING, LEFT_CONTIG):
1558 /*
1559 * Setting the first part of a previous oldext extent to newext.
1560 * The left neighbor is contiguous.
1561 */
1562 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx - 1,
1563 XFS_DATA_FORK);
1564 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1565 LEFT.br_blockcount + new->br_blockcount);
1566 xfs_bmbt_set_startoff(ep,
1567 PREV.br_startoff + new->br_blockcount);
1568 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx - 1,
1569 XFS_DATA_FORK);
1570 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx,
1571 XFS_DATA_FORK);
1572 xfs_bmbt_set_startblock(ep,
1573 new->br_startblock + new->br_blockcount);
1574 xfs_bmbt_set_blockcount(ep,
1575 PREV.br_blockcount - new->br_blockcount);
1576 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx,
1577 XFS_DATA_FORK);
1578 ip->i_df.if_lastex = idx - 1;
1579 if (cur == NULL)
1580 rval = XFS_ILOG_DEXT;
1581 else {
1582 rval = 0;
1583 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1584 PREV.br_startblock, PREV.br_blockcount,
1585 &i)))
1586 goto done;
1587 ASSERT(i == 1);
1588 if ((error = xfs_bmbt_update(cur,
1589 PREV.br_startoff + new->br_blockcount,
1590 PREV.br_startblock + new->br_blockcount,
1591 PREV.br_blockcount - new->br_blockcount,
1592 oldext)))
1593 goto done;
1594 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1595 goto done;
1596 if (xfs_bmbt_update(cur, LEFT.br_startoff,
1597 LEFT.br_startblock,
1598 LEFT.br_blockcount + new->br_blockcount,
1599 LEFT.br_state))
1600 goto done;
1601 }
1602 /* DELTA: The boundary between two in-core extents moved. */
1603 temp = LEFT.br_startoff;
1604 temp2 = LEFT.br_blockcount +
1605 PREV.br_blockcount;
1606 break;
1607
1608 case MASK(LEFT_FILLING):
1609 /*
1610 * Setting the first part of a previous oldext extent to newext.
1611 * The left neighbor is not contiguous.
1612 */
1613 xfs_bmap_trace_pre_update(fname, "LF", ip, idx, XFS_DATA_FORK);
1614 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1615 xfs_bmbt_set_startoff(ep, new_endoff);
1616 xfs_bmbt_set_blockcount(ep,
1617 PREV.br_blockcount - new->br_blockcount);
1618 xfs_bmbt_set_startblock(ep,
1619 new->br_startblock + new->br_blockcount);
1620 xfs_bmap_trace_post_update(fname, "LF", ip, idx, XFS_DATA_FORK);
1621 xfs_bmap_trace_insert(fname, "LF", ip, idx, 1, new, NULL,
1622 XFS_DATA_FORK);
1623 xfs_iext_insert(ifp, idx, 1, new);
1624 ip->i_df.if_lastex = idx;
1625 ip->i_d.di_nextents++;
1626 if (cur == NULL)
1627 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1628 else {
1629 rval = XFS_ILOG_CORE;
1630 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1631 PREV.br_startblock, PREV.br_blockcount,
1632 &i)))
1633 goto done;
1634 ASSERT(i == 1);
1635 if ((error = xfs_bmbt_update(cur,
1636 PREV.br_startoff + new->br_blockcount,
1637 PREV.br_startblock + new->br_blockcount,
1638 PREV.br_blockcount - new->br_blockcount,
1639 oldext)))
1640 goto done;
1641 cur->bc_rec.b = *new;
1642 if ((error = xfs_bmbt_insert(cur, &i)))
1643 goto done;
1644 ASSERT(i == 1);
1645 }
1646 /* DELTA: One in-core extent is split in two. */
1647 temp = PREV.br_startoff;
1648 temp2 = PREV.br_blockcount;
1649 break;
1650
1651 case MASK2(RIGHT_FILLING, RIGHT_CONTIG):
1652 /*
1653 * Setting the last part of a previous oldext extent to newext.
1654 * The right neighbor is contiguous with the new allocation.
1655 */
1656 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx,
1657 XFS_DATA_FORK);
1658 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx + 1,
1659 XFS_DATA_FORK);
1660 xfs_bmbt_set_blockcount(ep,
1661 PREV.br_blockcount - new->br_blockcount);
1662 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx,
1663 XFS_DATA_FORK);
1664 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, idx + 1),
1665 new->br_startoff, new->br_startblock,
1666 new->br_blockcount + RIGHT.br_blockcount, newext);
1667 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx + 1,
1668 XFS_DATA_FORK);
1669 ip->i_df.if_lastex = idx + 1;
1670 if (cur == NULL)
1671 rval = XFS_ILOG_DEXT;
1672 else {
1673 rval = 0;
1674 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1675 PREV.br_startblock,
1676 PREV.br_blockcount, &i)))
1677 goto done;
1678 ASSERT(i == 1);
1679 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1680 PREV.br_startblock,
1681 PREV.br_blockcount - new->br_blockcount,
1682 oldext)))
1683 goto done;
1684 if ((error = xfs_bmbt_increment(cur, 0, &i)))
1685 goto done;
1686 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1687 new->br_startblock,
1688 new->br_blockcount + RIGHT.br_blockcount,
1689 newext)))
1690 goto done;
1691 }
1692 /* DELTA: The boundary between two in-core extents moved. */
1693 temp = PREV.br_startoff;
1694 temp2 = PREV.br_blockcount +
1695 RIGHT.br_blockcount;
1696 break;
1697
1698 case MASK(RIGHT_FILLING):
1699 /*
1700 * Setting the last part of a previous oldext extent to newext.
1701 * The right neighbor is not contiguous.
1702 */
1703 xfs_bmap_trace_pre_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1704 xfs_bmbt_set_blockcount(ep,
1705 PREV.br_blockcount - new->br_blockcount);
1706 xfs_bmap_trace_post_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1707 xfs_bmap_trace_insert(fname, "RF", ip, idx + 1, 1,
1708 new, NULL, XFS_DATA_FORK);
1709 xfs_iext_insert(ifp, idx + 1, 1, new);
1710 ip->i_df.if_lastex = idx + 1;
1711 ip->i_d.di_nextents++;
1712 if (cur == NULL)
1713 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1714 else {
1715 rval = XFS_ILOG_CORE;
1716 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1717 PREV.br_startblock, PREV.br_blockcount,
1718 &i)))
1719 goto done;
1720 ASSERT(i == 1);
1721 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1722 PREV.br_startblock,
1723 PREV.br_blockcount - new->br_blockcount,
1724 oldext)))
1725 goto done;
1726 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1727 new->br_startblock, new->br_blockcount,
1728 &i)))
1729 goto done;
1730 ASSERT(i == 0);
1731 cur->bc_rec.b.br_state = XFS_EXT_NORM;
1732 if ((error = xfs_bmbt_insert(cur, &i)))
1733 goto done;
1734 ASSERT(i == 1);
1735 }
1736 /* DELTA: One in-core extent is split in two. */
1737 temp = PREV.br_startoff;
1738 temp2 = PREV.br_blockcount;
1739 break;
1740
1741 case 0:
1742 /*
1743 * Setting the middle part of a previous oldext extent to
1744 * newext. Contiguity is impossible here.
1745 * One extent becomes three extents.
1746 */
1747 xfs_bmap_trace_pre_update(fname, "0", ip, idx, XFS_DATA_FORK);
1748 xfs_bmbt_set_blockcount(ep,
1749 new->br_startoff - PREV.br_startoff);
1750 xfs_bmap_trace_post_update(fname, "0", ip, idx, XFS_DATA_FORK);
1751 r[0] = *new;
1752 r[1].br_startoff = new_endoff;
1753 r[1].br_blockcount =
1754 PREV.br_startoff + PREV.br_blockcount - new_endoff;
1755 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1756 r[1].br_state = oldext;
1757 xfs_bmap_trace_insert(fname, "0", ip, idx + 1, 2, &r[0], &r[1],
1758 XFS_DATA_FORK);
1759 xfs_iext_insert(ifp, idx + 1, 2, &r[0]);
1760 ip->i_df.if_lastex = idx + 1;
1761 ip->i_d.di_nextents += 2;
1762 if (cur == NULL)
1763 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1764 else {
1765 rval = XFS_ILOG_CORE;
1766 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1767 PREV.br_startblock, PREV.br_blockcount,
1768 &i)))
1769 goto done;
1770 ASSERT(i == 1);
1771 /* new right extent - oldext */
1772 if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1773 r[1].br_startblock, r[1].br_blockcount,
1774 r[1].br_state)))
1775 goto done;
1776 /* new left extent - oldext */
1777 PREV.br_blockcount =
1778 new->br_startoff - PREV.br_startoff;
1779 cur->bc_rec.b = PREV;
1780 if ((error = xfs_bmbt_insert(cur, &i)))
1781 goto done;
1782 ASSERT(i == 1);
1783 if ((error = xfs_bmbt_increment(cur, 0, &i)))
1784 goto done;
1785 ASSERT(i == 1);
1786 /* new middle extent - newext */
1787 cur->bc_rec.b = *new;
1788 if ((error = xfs_bmbt_insert(cur, &i)))
1789 goto done;
1790 ASSERT(i == 1);
1791 }
1792 /* DELTA: One in-core extent is split in three. */
1793 temp = PREV.br_startoff;
1794 temp2 = PREV.br_blockcount;
1795 break;
1796
1797 case MASK3(LEFT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1798 case MASK3(RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1799 case MASK2(LEFT_FILLING, RIGHT_CONTIG):
1800 case MASK2(RIGHT_FILLING, LEFT_CONTIG):
1801 case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1802 case MASK(LEFT_CONTIG):
1803 case MASK(RIGHT_CONTIG):
1804 /*
1805 * These cases are all impossible.
1806 */
1807 ASSERT(0);
1808 }
1809 *curp = cur;
1810 if (delta) {
1811 temp2 += temp;
1812 if (delta->xed_startoff > temp)
1813 delta->xed_startoff = temp;
1814 if (delta->xed_blockcount < temp2)
1815 delta->xed_blockcount = temp2;
1816 }
1817 done:
1818 *logflagsp = rval;
1819 return error;
1820 #undef LEFT
1821 #undef RIGHT
1822 #undef PREV
1823 #undef MASK
1824 #undef MASK2
1825 #undef MASK3
1826 #undef MASK4
1827 #undef STATE_SET
1828 #undef STATE_TEST
1829 #undef STATE_SET_TEST
1830 #undef SWITCH_STATE
1831 }
1832
1833 /*
1834 * Called by xfs_bmap_add_extent to handle cases converting a hole
1835 * to a delayed allocation.
1836 */
1837 /*ARGSUSED*/
1838 STATIC int /* error */
1839 xfs_bmap_add_extent_hole_delay(
1840 xfs_inode_t *ip, /* incore inode pointer */
1841 xfs_extnum_t idx, /* extent number to update/insert */
1842 xfs_bmbt_irec_t *new, /* new data to add to file extents */
1843 int *logflagsp, /* inode logging flags */
1844 xfs_extdelta_t *delta, /* Change made to incore extents */
1845 int rsvd) /* OK to allocate reserved blocks */
1846 {
1847 xfs_bmbt_rec_t *ep; /* extent record for idx */
1848 #ifdef XFS_BMAP_TRACE
1849 static char fname[] = "xfs_bmap_add_extent_hole_delay";
1850 #endif
1851 xfs_ifork_t *ifp; /* inode fork pointer */
1852 xfs_bmbt_irec_t left; /* left neighbor extent entry */
1853 xfs_filblks_t newlen=0; /* new indirect size */
1854 xfs_filblks_t oldlen=0; /* old indirect size */
1855 xfs_bmbt_irec_t right; /* right neighbor extent entry */
1856 int state; /* state bits, accessed thru macros */
1857 xfs_filblks_t temp=0; /* temp for indirect calculations */
1858 xfs_filblks_t temp2=0;
1859 enum { /* bit number definitions for state */
1860 LEFT_CONTIG, RIGHT_CONTIG,
1861 LEFT_DELAY, RIGHT_DELAY,
1862 LEFT_VALID, RIGHT_VALID
1863 };
1864
1865 #define MASK(b) (1 << (b))
1866 #define MASK2(a,b) (MASK(a) | MASK(b))
1867 #define STATE_SET(b,v) ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1868 #define STATE_TEST(b) (state & MASK(b))
1869 #define STATE_SET_TEST(b,v) ((v) ? ((state |= MASK(b)), 1) : \
1870 ((state &= ~MASK(b)), 0))
1871 #define SWITCH_STATE (state & MASK2(LEFT_CONTIG, RIGHT_CONTIG))
1872
1873 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1874 ep = xfs_iext_get_ext(ifp, idx);
1875 state = 0;
1876 ASSERT(ISNULLSTARTBLOCK(new->br_startblock));
1877 /*
1878 * Check and set flags if this segment has a left neighbor
1879 */
1880 if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1881 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &left);
1882 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(left.br_startblock));
1883 }
1884 /*
1885 * Check and set flags if the current (right) segment exists.
1886 * If it doesn't exist, we're converting the hole at end-of-file.
1887 */
1888 if (STATE_SET_TEST(RIGHT_VALID,
1889 idx <
1890 ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1891 xfs_bmbt_get_all(ep, &right);
1892 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(right.br_startblock));
1893 }
1894 /*
1895 * Set contiguity flags on the left and right neighbors.
1896 * Don't let extents get too large, even if the pieces are contiguous.
1897 */
1898 STATE_SET(LEFT_CONTIG,
1899 STATE_TEST(LEFT_VALID) && STATE_TEST(LEFT_DELAY) &&
1900 left.br_startoff + left.br_blockcount == new->br_startoff &&
1901 left.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1902 STATE_SET(RIGHT_CONTIG,
1903 STATE_TEST(RIGHT_VALID) && STATE_TEST(RIGHT_DELAY) &&
1904 new->br_startoff + new->br_blockcount == right.br_startoff &&
1905 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1906 (!STATE_TEST(LEFT_CONTIG) ||
1907 (left.br_blockcount + new->br_blockcount +
1908 right.br_blockcount <= MAXEXTLEN)));
1909 /*
1910 * Switch out based on the contiguity flags.
1911 */
1912 switch (SWITCH_STATE) {
1913
1914 case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1915 /*
1916 * New allocation is contiguous with delayed allocations
1917 * on the left and on the right.
1918 * Merge all three into a single extent record.
1919 */
1920 temp = left.br_blockcount + new->br_blockcount +
1921 right.br_blockcount;
1922 xfs_bmap_trace_pre_update(fname, "LC|RC", ip, idx - 1,
1923 XFS_DATA_FORK);
1924 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1), temp);
1925 oldlen = STARTBLOCKVAL(left.br_startblock) +
1926 STARTBLOCKVAL(new->br_startblock) +
1927 STARTBLOCKVAL(right.br_startblock);
1928 newlen = xfs_bmap_worst_indlen(ip, temp);
1929 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx - 1),
1930 NULLSTARTBLOCK((int)newlen));
1931 xfs_bmap_trace_post_update(fname, "LC|RC", ip, idx - 1,
1932 XFS_DATA_FORK);
1933 xfs_bmap_trace_delete(fname, "LC|RC", ip, idx, 1,
1934 XFS_DATA_FORK);
1935 xfs_iext_remove(ifp, idx, 1);
1936 ip->i_df.if_lastex = idx - 1;
1937 /* DELTA: Two in-core extents were replaced by one. */
1938 temp2 = temp;
1939 temp = left.br_startoff;
1940 break;
1941
1942 case MASK(LEFT_CONTIG):
1943 /*
1944 * New allocation is contiguous with a delayed allocation
1945 * on the left.
1946 * Merge the new allocation with the left neighbor.
1947 */
1948 temp = left.br_blockcount + new->br_blockcount;
1949 xfs_bmap_trace_pre_update(fname, "LC", ip, idx - 1,
1950 XFS_DATA_FORK);
1951 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1), temp);
1952 oldlen = STARTBLOCKVAL(left.br_startblock) +
1953 STARTBLOCKVAL(new->br_startblock);
1954 newlen = xfs_bmap_worst_indlen(ip, temp);
1955 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx - 1),
1956 NULLSTARTBLOCK((int)newlen));
1957 xfs_bmap_trace_post_update(fname, "LC", ip, idx - 1,
1958 XFS_DATA_FORK);
1959 ip->i_df.if_lastex = idx - 1;
1960 /* DELTA: One in-core extent grew into a hole. */
1961 temp2 = temp;
1962 temp = left.br_startoff;
1963 break;
1964
1965 case MASK(RIGHT_CONTIG):
1966 /*
1967 * New allocation is contiguous with a delayed allocation
1968 * on the right.
1969 * Merge the new allocation with the right neighbor.
1970 */
1971 xfs_bmap_trace_pre_update(fname, "RC", ip, idx, XFS_DATA_FORK);
1972 temp = new->br_blockcount + right.br_blockcount;
1973 oldlen = STARTBLOCKVAL(new->br_startblock) +
1974 STARTBLOCKVAL(right.br_startblock);
1975 newlen = xfs_bmap_worst_indlen(ip, temp);
1976 xfs_bmbt_set_allf(ep, new->br_startoff,
1977 NULLSTARTBLOCK((int)newlen), temp, right.br_state);
1978 xfs_bmap_trace_post_update(fname, "RC", ip, idx, XFS_DATA_FORK);
1979 ip->i_df.if_lastex = idx;
1980 /* DELTA: One in-core extent grew into a hole. */
1981 temp2 = temp;
1982 temp = new->br_startoff;
1983 break;
1984
1985 case 0:
1986 /*
1987 * New allocation is not contiguous with another
1988 * delayed allocation.
1989 * Insert a new entry.
1990 */
1991 oldlen = newlen = 0;
1992 xfs_bmap_trace_insert(fname, "0", ip, idx, 1, new, NULL,
1993 XFS_DATA_FORK);
1994 xfs_iext_insert(ifp, idx, 1, new);
1995 ip->i_df.if_lastex = idx;
1996 /* DELTA: A new in-core extent was added in a hole. */
1997 temp2 = new->br_blockcount;
1998 temp = new->br_startoff;
1999 break;
2000 }
2001 if (oldlen != newlen) {
2002 ASSERT(oldlen > newlen);
2003 xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
2004 (int64_t)(oldlen - newlen), rsvd);
2005 /*
2006 * Nothing to do for disk quota accounting here.
2007 */
2008 }
2009 if (delta) {
2010 temp2 += temp;
2011 if (delta->xed_startoff > temp)
2012 delta->xed_startoff = temp;
2013 if (delta->xed_blockcount < temp2)
2014 delta->xed_blockcount = temp2;
2015 }
2016 *logflagsp = 0;
2017 return 0;
2018 #undef MASK
2019 #undef MASK2
2020 #undef STATE_SET
2021 #undef STATE_TEST
2022 #undef STATE_SET_TEST
2023 #undef SWITCH_STATE
2024 }
2025
2026 /*
2027 * Called by xfs_bmap_add_extent to handle cases converting a hole
2028 * to a real allocation.
2029 */
2030 STATIC int /* error */
2031 xfs_bmap_add_extent_hole_real(
2032 xfs_inode_t *ip, /* incore inode pointer */
2033 xfs_extnum_t idx, /* extent number to update/insert */
2034 xfs_btree_cur_t *cur, /* if null, not a btree */
2035 xfs_bmbt_irec_t *new, /* new data to add to file extents */
2036 int *logflagsp, /* inode logging flags */
2037 xfs_extdelta_t *delta, /* Change made to incore extents */
2038 int whichfork) /* data or attr fork */
2039 {
2040 xfs_bmbt_rec_t *ep; /* pointer to extent entry ins. point */
2041 int error; /* error return value */
2042 #ifdef XFS_BMAP_TRACE
2043 static char fname[] = "xfs_bmap_add_extent_hole_real";
2044 #endif
2045 int i; /* temp state */
2046 xfs_ifork_t *ifp; /* inode fork pointer */
2047 xfs_bmbt_irec_t left; /* left neighbor extent entry */
2048 xfs_bmbt_irec_t right; /* right neighbor extent entry */
2049 int rval=0; /* return value (logging flags) */
2050 int state; /* state bits, accessed thru macros */
2051 xfs_filblks_t temp=0;
2052 xfs_filblks_t temp2=0;
2053 enum { /* bit number definitions for state */
2054 LEFT_CONTIG, RIGHT_CONTIG,
2055 LEFT_DELAY, RIGHT_DELAY,
2056 LEFT_VALID, RIGHT_VALID
2057 };
2058
2059 #define MASK(b) (1 << (b))
2060 #define MASK2(a,b) (MASK(a) | MASK(b))
2061 #define STATE_SET(b,v) ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
2062 #define STATE_TEST(b) (state & MASK(b))
2063 #define STATE_SET_TEST(b,v) ((v) ? ((state |= MASK(b)), 1) : \
2064 ((state &= ~MASK(b)), 0))
2065 #define SWITCH_STATE (state & MASK2(LEFT_CONTIG, RIGHT_CONTIG))
2066
2067 ifp = XFS_IFORK_PTR(ip, whichfork);
2068 ASSERT(idx <= ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
2069 ep = xfs_iext_get_ext(ifp, idx);
2070 state = 0;
2071 /*
2072 * Check and set flags if this segment has a left neighbor.
2073 */
2074 if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
2075 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &left);
2076 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(left.br_startblock));
2077 }
2078 /*
2079 * Check and set flags if this segment has a current value.
2080 * Not true if we're inserting into the "hole" at eof.
2081 */
2082 if (STATE_SET_TEST(RIGHT_VALID,
2083 idx <
2084 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
2085 xfs_bmbt_get_all(ep, &right);
2086 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(right.br_startblock));
2087 }
2088 /*
2089 * We're inserting a real allocation between "left" and "right".
2090 * Set the contiguity flags. Don't let extents get too large.
2091 */
2092 STATE_SET(LEFT_CONTIG,
2093 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
2094 left.br_startoff + left.br_blockcount == new->br_startoff &&
2095 left.br_startblock + left.br_blockcount == new->br_startblock &&
2096 left.br_state == new->br_state &&
2097 left.br_blockcount + new->br_blockcount <= MAXEXTLEN);
2098 STATE_SET(RIGHT_CONTIG,
2099 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
2100 new->br_startoff + new->br_blockcount == right.br_startoff &&
2101 new->br_startblock + new->br_blockcount ==
2102 right.br_startblock &&
2103 new->br_state == right.br_state &&
2104 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2105 (!STATE_TEST(LEFT_CONTIG) ||
2106 left.br_blockcount + new->br_blockcount +
2107 right.br_blockcount <= MAXEXTLEN));
2108
2109 error = 0;
2110 /*
2111 * Select which case we're in here, and implement it.
2112 */
2113 switch (SWITCH_STATE) {
2114
2115 case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
2116 /*
2117 * New allocation is contiguous with real allocations on the
2118 * left and on the right.
2119 * Merge all three into a single extent record.
2120 */
2121 xfs_bmap_trace_pre_update(fname, "LC|RC", ip, idx - 1,
2122 whichfork);
2123 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
2124 left.br_blockcount + new->br_blockcount +
2125 right.br_blockcount);
2126 xfs_bmap_trace_post_update(fname, "LC|RC", ip, idx - 1,
2127 whichfork);
2128 xfs_bmap_trace_delete(fname, "LC|RC", ip,
2129 idx, 1, whichfork);
2130 xfs_iext_remove(ifp, idx, 1);
2131 ifp->if_lastex = idx - 1;
2132 XFS_IFORK_NEXT_SET(ip, whichfork,
2133 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2134 if (cur == NULL) {
2135 rval = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2136 } else {
2137 rval = XFS_ILOG_CORE;
2138 if ((error = xfs_bmbt_lookup_eq(cur,
2139 right.br_startoff,
2140 right.br_startblock,
2141 right.br_blockcount, &i)))
2142 goto done;
2143 ASSERT(i == 1);
2144 if ((error = xfs_bmbt_delete(cur, &i)))
2145 goto done;
2146 ASSERT(i == 1);
2147 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
2148 goto done;
2149 ASSERT(i == 1);
2150 if ((error = xfs_bmbt_update(cur, left.br_startoff,
2151 left.br_startblock,
2152 left.br_blockcount +
2153 new->br_blockcount +
2154 right.br_blockcount,
2155 left.br_state)))
2156 goto done;
2157 }
2158 /* DELTA: Two in-core extents were replaced by one. */
2159 temp = left.br_startoff;
2160 temp2 = left.br_blockcount +
2161 new->br_blockcount +
2162 right.br_blockcount;
2163 break;
2164
2165 case MASK(LEFT_CONTIG):
2166 /*
2167 * New allocation is contiguous with a real allocation
2168 * on the left.
2169 * Merge the new allocation with the left neighbor.
2170 */
2171 xfs_bmap_trace_pre_update(fname, "LC", ip, idx - 1, whichfork);
2172 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
2173 left.br_blockcount + new->br_blockcount);
2174 xfs_bmap_trace_post_update(fname, "LC", ip, idx - 1, whichfork);
2175 ifp->if_lastex = idx - 1;
2176 if (cur == NULL) {
2177 rval = XFS_ILOG_FEXT(whichfork);
2178 } else {
2179 rval = 0;
2180 if ((error = xfs_bmbt_lookup_eq(cur,
2181 left.br_startoff,
2182 left.br_startblock,
2183 left.br_blockcount, &i)))
2184 goto done;
2185 ASSERT(i == 1);
2186 if ((error = xfs_bmbt_update(cur, left.br_startoff,
2187 left.br_startblock,
2188 left.br_blockcount +
2189 new->br_blockcount,
2190 left.br_state)))
2191 goto done;
2192 }
2193 /* DELTA: One in-core extent grew. */
2194 temp = left.br_startoff;
2195 temp2 = left.br_blockcount +
2196 new->br_blockcount;
2197 break;
2198
2199 case MASK(RIGHT_CONTIG):
2200 /*
2201 * New allocation is contiguous with a real allocation
2202 * on the right.
2203 * Merge the new allocation with the right neighbor.
2204 */
2205 xfs_bmap_trace_pre_update(fname, "RC", ip, idx, whichfork);
2206 xfs_bmbt_set_allf(ep, new->br_startoff, new->br_startblock,
2207 new->br_blockcount + right.br_blockcount,
2208 right.br_state);
2209 xfs_bmap_trace_post_update(fname, "RC", ip, idx, whichfork);
2210 ifp->if_lastex = idx;
2211 if (cur == NULL) {
2212 rval = XFS_ILOG_FEXT(whichfork);
2213 } else {
2214 rval = 0;
2215 if ((error = xfs_bmbt_lookup_eq(cur,
2216 right.br_startoff,
2217 right.br_startblock,
2218 right.br_blockcount, &i)))
2219 goto done;
2220 ASSERT(i == 1);
2221 if ((error = xfs_bmbt_update(cur, new->br_startoff,
2222 new->br_startblock,
2223 new->br_blockcount +
2224 right.br_blockcount,
2225 right.br_state)))
2226 goto done;
2227 }
2228 /* DELTA: One in-core extent grew. */
2229 temp = new->br_startoff;
2230 temp2 = new->br_blockcount +
2231 right.br_blockcount;
2232 break;
2233
2234 case 0:
2235 /*
2236 * New allocation is not contiguous with another
2237 * real allocation.
2238 * Insert a new entry.
2239 */
2240 xfs_bmap_trace_insert(fname, "0", ip, idx, 1, new, NULL,
2241 whichfork);
2242 xfs_iext_insert(ifp, idx, 1, new);
2243 ifp->if_lastex = idx;
2244 XFS_IFORK_NEXT_SET(ip, whichfork,
2245 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2246 if (cur == NULL) {
2247 rval = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2248 } else {
2249 rval = XFS_ILOG_CORE;
2250 if ((error = xfs_bmbt_lookup_eq(cur,
2251 new->br_startoff,
2252 new->br_startblock,
2253 new->br_blockcount, &i)))
2254 goto done;
2255 ASSERT(i == 0);
2256 cur->bc_rec.b.br_state = new->br_state;
2257 if ((error = xfs_bmbt_insert(cur, &i)))
2258 goto done;
2259 ASSERT(i == 1);
2260 }
2261 /* DELTA: A new extent was added in a hole. */
2262 temp = new->br_startoff;
2263 temp2 = new->br_blockcount;
2264 break;
2265 }
2266 if (delta) {
2267 temp2 += temp;
2268 if (delta->xed_startoff > temp)
2269 delta->xed_startoff = temp;
2270 if (delta->xed_blockcount < temp2)
2271 delta->xed_blockcount = temp2;
2272 }
2273 done:
2274 *logflagsp = rval;
2275 return error;
2276 #undef MASK
2277 #undef MASK2
2278 #undef STATE_SET
2279 #undef STATE_TEST
2280 #undef STATE_SET_TEST
2281 #undef SWITCH_STATE
2282 }
2283
2284 /*
2285 * Adjust the size of the new extent based on di_extsize and rt extsize.
2286 */
2287 STATIC int
2288 xfs_bmap_extsize_align(
2289 xfs_mount_t *mp,
2290 xfs_bmbt_irec_t *gotp, /* next extent pointer */
2291 xfs_bmbt_irec_t *prevp, /* previous extent pointer */
2292 xfs_extlen_t extsz, /* align to this extent size */
2293 int rt, /* is this a realtime inode? */
2294 int eof, /* is extent at end-of-file? */
2295 int delay, /* creating delalloc extent? */
2296 int convert, /* overwriting unwritten extent? */
2297 xfs_fileoff_t *offp, /* in/out: aligned offset */
2298 xfs_extlen_t *lenp) /* in/out: aligned length */
2299 {
2300 xfs_fileoff_t orig_off; /* original offset */
2301 xfs_extlen_t orig_alen; /* original length */
2302 xfs_fileoff_t orig_end; /* original off+len */
2303 xfs_fileoff_t nexto; /* next file offset */
2304 xfs_fileoff_t prevo; /* previous file offset */
2305 xfs_fileoff_t align_off; /* temp for offset */
2306 xfs_extlen_t align_alen; /* temp for length */
2307 xfs_extlen_t temp; /* temp for calculations */
2308
2309 if (convert)
2310 return 0;
2311
2312 orig_off = align_off = *offp;
2313 orig_alen = align_alen = *lenp;
2314 orig_end = orig_off + orig_alen;
2315
2316 /*
2317 * If this request overlaps an existing extent, then don't
2318 * attempt to perform any additional alignment.
2319 */
2320 if (!delay && !eof &&
2321 (orig_off >= gotp->br_startoff) &&
2322 (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2323 return 0;
2324 }
2325
2326 /*
2327 * If the file offset is unaligned vs. the extent size
2328 * we need to align it. This will be possible unless
2329 * the file was previously written with a kernel that didn't
2330 * perform this alignment, or if a truncate shot us in the
2331 * foot.
2332 */
2333 temp = do_mod(orig_off, extsz);
2334 if (temp) {
2335 align_alen += temp;
2336 align_off -= temp;
2337 }
2338 /*
2339 * Same adjustment for the end of the requested area.
2340 */
2341 if ((temp = (align_alen % extsz))) {
2342 align_alen += extsz - temp;
2343 }
2344 /*
2345 * If the previous block overlaps with this proposed allocation
2346 * then move the start forward without adjusting the length.
2347 */
2348 if (prevp->br_startoff != NULLFILEOFF) {
2349 if (prevp->br_startblock == HOLESTARTBLOCK)
2350 prevo = prevp->br_startoff;
2351 else
2352 prevo = prevp->br_startoff + prevp->br_blockcount;
2353 } else
2354 prevo = 0;
2355 if (align_off != orig_off && align_off < prevo)
2356 align_off = prevo;
2357 /*
2358 * If the next block overlaps with this proposed allocation
2359 * then move the start back without adjusting the length,
2360 * but not before offset 0.
2361 * This may of course make the start overlap previous block,
2362 * and if we hit the offset 0 limit then the next block
2363 * can still overlap too.
2364 */
2365 if (!eof && gotp->br_startoff != NULLFILEOFF) {
2366 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
2367 (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
2368 nexto = gotp->br_startoff + gotp->br_blockcount;
2369 else
2370 nexto = gotp->br_startoff;
2371 } else
2372 nexto = NULLFILEOFF;
2373 if (!eof &&
2374 align_off + align_alen != orig_end &&
2375 align_off + align_alen > nexto)
2376 align_off = nexto > align_alen ? nexto - align_alen : 0;
2377 /*
2378 * If we're now overlapping the next or previous extent that
2379 * means we can't fit an extsz piece in this hole. Just move
2380 * the start forward to the first valid spot and set
2381 * the length so we hit the end.
2382 */
2383 if (align_off != orig_off && align_off < prevo)
2384 align_off = prevo;
2385 if (align_off + align_alen != orig_end &&
2386 align_off + align_alen > nexto &&
2387 nexto != NULLFILEOFF) {
2388 ASSERT(nexto > prevo);
2389 align_alen = nexto - align_off;
2390 }
2391
2392 /*
2393 * If realtime, and the result isn't a multiple of the realtime
2394 * extent size we need to remove blocks until it is.
2395 */
2396 if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
2397 /*
2398 * We're not covering the original request, or
2399 * we won't be able to once we fix the length.
2400 */
2401 if (orig_off < align_off ||
2402 orig_end > align_off + align_alen ||
2403 align_alen - temp < orig_alen)
2404 return XFS_ERROR(EINVAL);
2405 /*
2406 * Try to fix it by moving the start up.
2407 */
2408 if (align_off + temp <= orig_off) {
2409 align_alen -= temp;
2410 align_off += temp;
2411 }
2412 /*
2413 * Try to fix it by moving the end in.
2414 */
2415 else if (align_off + align_alen - temp >= orig_end)
2416 align_alen -= temp;
2417 /*
2418 * Set the start to the minimum then trim the length.
2419 */
2420 else {
2421 align_alen -= orig_off - align_off;
2422 align_off = orig_off;
2423 align_alen -= align_alen % mp->m_sb.sb_rextsize;
2424 }
2425 /*
2426 * Result doesn't cover the request, fail it.
2427 */
2428 if (orig_off < align_off || orig_end > align_off + align_alen)
2429 return XFS_ERROR(EINVAL);
2430 } else {
2431 ASSERT(orig_off >= align_off);
2432 ASSERT(orig_end <= align_off + align_alen);
2433 }
2434
2435 #ifdef DEBUG
2436 if (!eof && gotp->br_startoff != NULLFILEOFF)
2437 ASSERT(align_off + align_alen <= gotp->br_startoff);
2438 if (prevp->br_startoff != NULLFILEOFF)
2439 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
2440 #endif
2441
2442 *lenp = align_alen;
2443 *offp = align_off;
2444 return 0;
2445 }
2446
2447 #define XFS_ALLOC_GAP_UNITS 4
2448
2449 STATIC int
2450 xfs_bmap_adjacent(
2451 xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2452 {
2453 xfs_fsblock_t adjust; /* adjustment to block numbers */
2454 xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
2455 xfs_mount_t *mp; /* mount point structure */
2456 int nullfb; /* true if ap->firstblock isn't set */
2457 int rt; /* true if inode is realtime */
2458
2459 #define ISVALID(x,y) \
2460 (rt ? \
2461 (x) < mp->m_sb.sb_rblocks : \
2462 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2463 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2464 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2465
2466 mp = ap->ip->i_mount;
2467 nullfb = ap->firstblock == NULLFSBLOCK;
2468 rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2469 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2470 /*
2471 * If allocating at eof, and there's a previous real block,
2472 * try to use it's last block as our starting point.
2473 */
2474 if (ap->eof && ap->prevp->br_startoff != NULLFILEOFF &&
2475 !ISNULLSTARTBLOCK(ap->prevp->br_startblock) &&
2476 ISVALID(ap->prevp->br_startblock + ap->prevp->br_blockcount,
2477 ap->prevp->br_startblock)) {
2478 ap->rval = ap->prevp->br_startblock + ap->prevp->br_blockcount;
2479 /*
2480 * Adjust for the gap between prevp and us.
2481 */
2482 adjust = ap->off -
2483 (ap->prevp->br_startoff + ap->prevp->br_blockcount);
2484 if (adjust &&
2485 ISVALID(ap->rval + adjust, ap->prevp->br_startblock))
2486 ap->rval += adjust;
2487 }
2488 /*
2489 * If not at eof, then compare the two neighbor blocks.
2490 * Figure out whether either one gives us a good starting point,
2491 * and pick the better one.
2492 */
2493 else if (!ap->eof) {
2494 xfs_fsblock_t gotbno; /* right side block number */
2495 xfs_fsblock_t gotdiff=0; /* right side difference */
2496 xfs_fsblock_t prevbno; /* left side block number */
2497 xfs_fsblock_t prevdiff=0; /* left side difference */
2498
2499 /*
2500 * If there's a previous (left) block, select a requested
2501 * start block based on it.
2502 */
2503 if (ap->prevp->br_startoff != NULLFILEOFF &&
2504 !ISNULLSTARTBLOCK(ap->prevp->br_startblock) &&
2505 (prevbno = ap->prevp->br_startblock +
2506 ap->prevp->br_blockcount) &&
2507 ISVALID(prevbno, ap->prevp->br_startblock)) {
2508 /*
2509 * Calculate gap to end of previous block.
2510 */
2511 adjust = prevdiff = ap->off -
2512 (ap->prevp->br_startoff +
2513 ap->prevp->br_blockcount);
2514 /*
2515 * Figure the startblock based on the previous block's
2516 * end and the gap size.
2517 * Heuristic!
2518 * If the gap is large relative to the piece we're
2519 * allocating, or using it gives us an invalid block
2520 * number, then just use the end of the previous block.
2521 */
2522 if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2523 ISVALID(prevbno + prevdiff,
2524 ap->prevp->br_startblock))
2525 prevbno += adjust;
2526 else
2527 prevdiff += adjust;
2528 /*
2529 * If the firstblock forbids it, can't use it,
2530 * must use default.
2531 */
2532 if (!rt && !nullfb &&
2533 XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2534 prevbno = NULLFSBLOCK;
2535 }
2536 /*
2537 * No previous block or can't follow it, just default.
2538 */
2539 else
2540 prevbno = NULLFSBLOCK;
2541 /*
2542 * If there's a following (right) block, select a requested
2543 * start block based on it.
2544 */
2545 if (!ISNULLSTARTBLOCK(ap->gotp->br_startblock)) {
2546 /*
2547 * Calculate gap to start of next block.
2548 */
2549 adjust = gotdiff = ap->gotp->br_startoff - ap->off;
2550 /*
2551 * Figure the startblock based on the next block's
2552 * start and the gap size.
2553 */
2554 gotbno = ap->gotp->br_startblock;
2555 /*
2556 * Heuristic!
2557 * If the gap is large relative to the piece we're
2558 * allocating, or using it gives us an invalid block
2559 * number, then just use the start of the next block
2560 * offset by our length.
2561 */
2562 if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2563 ISVALID(gotbno - gotdiff, gotbno))
2564 gotbno -= adjust;
2565 else if (ISVALID(gotbno - ap->alen, gotbno)) {
2566 gotbno -= ap->alen;
2567 gotdiff += adjust - ap->alen;
2568 } else
2569 gotdiff += adjust;
2570 /*
2571 * If the firstblock forbids it, can't use it,
2572 * must use default.
2573 */
2574 if (!rt && !nullfb &&
2575 XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2576 gotbno = NULLFSBLOCK;
2577 }
2578 /*
2579 * No next block, just default.
2580 */
2581 else
2582 gotbno = NULLFSBLOCK;
2583 /*
2584 * If both valid, pick the better one, else the only good
2585 * one, else ap->rval is already set (to 0 or the inode block).
2586 */
2587 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2588 ap->rval = prevdiff <= gotdiff ? prevbno : gotbno;
2589 else if (prevbno != NULLFSBLOCK)
2590 ap->rval = prevbno;
2591 else if (gotbno != NULLFSBLOCK)
2592 ap->rval = gotbno;
2593 }
2594 #undef ISVALID
2595 return 0;
2596 }
2597
2598 STATIC int
2599 xfs_bmap_rtalloc(
2600 xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2601 {
2602 xfs_alloctype_t atype = 0; /* type for allocation routines */
2603 int error; /* error return value */
2604 xfs_mount_t *mp; /* mount point structure */
2605 xfs_extlen_t prod = 0; /* product factor for allocators */
2606 xfs_extlen_t ralen = 0; /* realtime allocation length */
2607 xfs_extlen_t align; /* minimum allocation alignment */
2608 xfs_rtblock_t rtx; /* realtime extent number */
2609 xfs_rtblock_t rtb;
2610
2611 mp = ap->ip->i_mount;
2612 align = ap->ip->i_d.di_extsize ?
2613 ap->ip->i_d.di_extsize : mp->m_sb.sb_rextsize;
2614 prod = align / mp->m_sb.sb_rextsize;
2615 error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2616 align, 1, ap->eof, 0,
2617 ap->conv, &ap->off, &ap->alen);
2618 if (error)
2619 return error;
2620 ASSERT(ap->alen);
2621 ASSERT(ap->alen % mp->m_sb.sb_rextsize == 0);
2622
2623 /*
2624 * If the offset & length are not perfectly aligned
2625 * then kill prod, it will just get us in trouble.
2626 */
2627 if (do_mod(ap->off, align) || ap->alen % align)
2628 prod = 1;
2629 /*
2630 * Set ralen to be the actual requested length in rtextents.
2631 */
2632 ralen = ap->alen / mp->m_sb.sb_rextsize;
2633 /*
2634 * If the old value was close enough to MAXEXTLEN that
2635 * we rounded up to it, cut it back so it's valid again.
2636 * Note that if it's a really large request (bigger than
2637 * MAXEXTLEN), we don't hear about that number, and can't
2638 * adjust the starting point to match it.
2639 */
2640 if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2641 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2642 /*
2643 * If it's an allocation to an empty file at offset 0,
2644 * pick an extent that will space things out in the rt area.
2645 */
2646 if (ap->eof && ap->off == 0) {
2647 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2648 if (error)
2649 return error;
2650 ap->rval = rtx * mp->m_sb.sb_rextsize;
2651 } else {
2652 ap->rval = 0;
2653 }
2654
2655 xfs_bmap_adjacent(ap);
2656
2657 /*
2658 * Realtime allocation, done through xfs_rtallocate_extent.
2659 */
2660 atype = ap->rval == 0 ? XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2661 do_div(ap->rval, mp->m_sb.sb_rextsize);
2662 rtb = ap->rval;
2663 ap->alen = ralen;
2664 if ((error = xfs_rtallocate_extent(ap->tp, ap->rval, 1, ap->alen,
2665 &ralen, atype, ap->wasdel, prod, &rtb)))
2666 return error;
2667 if (rtb == NULLFSBLOCK && prod > 1 &&
2668 (error = xfs_rtallocate_extent(ap->tp, ap->rval, 1,
2669 ap->alen, &ralen, atype,
2670 ap->wasdel, 1, &rtb)))
2671 return error;
2672 ap->rval = rtb;
2673 if (ap->rval != NULLFSBLOCK) {
2674 ap->rval *= mp->m_sb.sb_rextsize;
2675 ralen *= mp->m_sb.sb_rextsize;
2676 ap->alen = ralen;
2677 ap->ip->i_d.di_nblocks += ralen;
2678 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2679 if (ap->wasdel)
2680 ap->ip->i_delayed_blks -= ralen;
2681 /*
2682 * Adjust the disk quota also. This was reserved
2683 * earlier.
2684 */
2685 XFS_TRANS_MOD_DQUOT_BYINO(mp, ap->tp, ap->ip,
2686 ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2687 XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2688 } else {
2689 ap->alen = 0;
2690 }
2691 return 0;
2692 }
2693
2694 STATIC int
2695 xfs_bmap_btalloc(
2696 xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2697 {
2698 xfs_mount_t *mp; /* mount point structure */
2699 xfs_alloctype_t atype = 0; /* type for allocation routines */
2700 xfs_extlen_t align; /* minimum allocation alignment */
2701 xfs_agnumber_t ag;
2702 xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
2703 xfs_agnumber_t startag;
2704 xfs_alloc_arg_t args;
2705 xfs_extlen_t blen;
2706 xfs_extlen_t delta;
2707 xfs_extlen_t longest;
2708 xfs_extlen_t need;
2709 xfs_extlen_t nextminlen = 0;
2710 xfs_perag_t *pag;
2711 int nullfb; /* true if ap->firstblock isn't set */
2712 int isaligned;
2713 int notinit;
2714 int tryagain;
2715 int error;
2716
2717 mp = ap->ip->i_mount;
2718 align = (ap->userdata && ap->ip->i_d.di_extsize &&
2719 (ap->ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE)) ?
2720 ap->ip->i_d.di_extsize : 0;
2721 if (unlikely(align)) {
2722 error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2723 align, 0, ap->eof, 0, ap->conv,
2724 &ap->off, &ap->alen);
2725 ASSERT(!error);
2726 ASSERT(ap->alen);
2727 }
2728 nullfb = ap->firstblock == NULLFSBLOCK;
2729 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2730 if (nullfb)
2731 ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2732 else
2733 ap->rval = ap->firstblock;
2734
2735 xfs_bmap_adjacent(ap);
2736
2737 /*
2738 * If allowed, use ap->rval; otherwise must use firstblock since
2739 * it's in the right allocation group.
2740 */
2741 if (nullfb || XFS_FSB_TO_AGNO(mp, ap->rval) == fb_agno)
2742 ;
2743 else
2744 ap->rval = ap->firstblock;
2745 /*
2746 * Normal allocation, done through xfs_alloc_vextent.
2747 */
2748 tryagain = isaligned = 0;
2749 args.tp = ap->tp;
2750 args.mp = mp;
2751 args.fsbno = ap->rval;
2752 args.maxlen = MIN(ap->alen, mp->m_sb.sb_agblocks);
2753 args.firstblock = ap->firstblock;
2754 blen = 0;
2755 if (nullfb) {
2756 args.type = XFS_ALLOCTYPE_START_BNO;
2757 args.total = ap->total;
2758 /*
2759 * Find the longest available space.
2760 * We're going to try for the whole allocation at once.
2761 */
2762 startag = ag = XFS_FSB_TO_AGNO(mp, args.fsbno);
2763 notinit = 0;
2764 down_read(&mp->m_peraglock);
2765 while (blen < ap->alen) {
2766 pag = &mp->m_perag[ag];
2767 if (!pag->pagf_init &&
2768 (error = xfs_alloc_pagf_init(mp, args.tp,
2769 ag, XFS_ALLOC_FLAG_TRYLOCK))) {
2770 up_read(&mp->m_peraglock);
2771 return error;
2772 }
2773 /*
2774 * See xfs_alloc_fix_freelist...
2775 */
2776 if (pag->pagf_init) {
2777 need = XFS_MIN_FREELIST_PAG(pag, mp);
2778 delta = need > pag->pagf_flcount ?
2779 need - pag->pagf_flcount : 0;
2780 longest = (pag->pagf_longest > delta) ?
2781 (pag->pagf_longest - delta) :
2782 (pag->pagf_flcount > 0 ||
2783 pag->pagf_longest > 0);
2784 if (blen < longest)
2785 blen = longest;
2786 } else
2787 notinit = 1;
2788 if (++ag == mp->m_sb.sb_agcount)
2789 ag = 0;
2790 if (ag == startag)
2791 break;
2792 }
2793 up_read(&mp->m_peraglock);
2794 /*
2795 * Since the above loop did a BUF_TRYLOCK, it is
2796 * possible that there is space for this request.
2797 */
2798 if (notinit || blen < ap->minlen)
2799 args.minlen = ap->minlen;
2800 /*
2801 * If the best seen length is less than the request
2802 * length, use the best as the minimum.
2803 */
2804 else if (blen < ap->alen)
2805 args.minlen = blen;
2806 /*
2807 * Otherwise we've seen an extent as big as alen,
2808 * use that as the minimum.
2809 */
2810 else
2811 args.minlen = ap->alen;
2812 } else if (ap->low) {
2813 args.type = XFS_ALLOCTYPE_START_BNO;
2814 args.total = args.minlen = ap->minlen;
2815 } else {
2816 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2817 args.total = ap->total;
2818 args.minlen = ap->minlen;
2819 }
2820 if (unlikely(ap->userdata && ap->ip->i_d.di_extsize &&
2821 (ap->ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE))) {
2822 args.prod = ap->ip->i_d.di_extsize;
2823 if ((args.mod = (xfs_extlen_t)do_mod(ap->off, args.prod)))
2824 args.mod = (xfs_extlen_t)(args.prod - args.mod);
2825 } else if (mp->m_sb.sb_blocksize >= NBPP) {
2826 args.prod = 1;
2827 args.mod = 0;
2828 } else {
2829 args.prod = NBPP >> mp->m_sb.sb_blocklog;
2830 if ((args.mod = (xfs_extlen_t)(do_mod(ap->off, args.prod))))
2831 args.mod = (xfs_extlen_t)(args.prod - args.mod);
2832 }
2833 /*
2834 * If we are not low on available data blocks, and the
2835 * underlying logical volume manager is a stripe, and
2836 * the file offset is zero then try to allocate data
2837 * blocks on stripe unit boundary.
2838 * NOTE: ap->aeof is only set if the allocation length
2839 * is >= the stripe unit and the allocation offset is
2840 * at the end of file.
2841 */
2842 if (!ap->low && ap->aeof) {
2843 if (!ap->off) {
2844 args.alignment = mp->m_dalign;
2845 atype = args.type;
2846 isaligned = 1;
2847 /*
2848 * Adjust for alignment
2849 */
2850 if (blen > args.alignment && blen <= ap->alen)
2851 args.minlen = blen - args.alignment;
2852 args.minalignslop = 0;
2853 } else {
2854 /*
2855 * First try an exact bno allocation.
2856 * If it fails then do a near or start bno
2857 * allocation with alignment turned on.
2858 */
2859 atype = args.type;
2860 tryagain = 1;
2861 args.type = XFS_ALLOCTYPE_THIS_BNO;
2862 args.alignment = 1;
2863 /*
2864 * Compute the minlen+alignment for the
2865 * next case. Set slop so that the value
2866 * of minlen+alignment+slop doesn't go up
2867 * between the calls.
2868 */
2869 if (blen > mp->m_dalign && blen <= ap->alen)
2870 nextminlen = blen - mp->m_dalign;
2871 else
2872 nextminlen = args.minlen;
2873 if (nextminlen + mp->m_dalign > args.minlen + 1)
2874 args.minalignslop =
2875 nextminlen + mp->m_dalign -
2876 args.minlen - 1;
2877 else
2878 args.minalignslop = 0;
2879 }
2880 } else {
2881 args.alignment = 1;
2882 args.minalignslop = 0;
2883 }
2884 args.minleft = ap->minleft;
2885 args.wasdel = ap->wasdel;
2886 args.isfl = 0;
2887 args.userdata = ap->userdata;
2888 if ((error = xfs_alloc_vextent(&args)))
2889 return error;
2890 if (tryagain && args.fsbno == NULLFSBLOCK) {
2891 /*
2892 * Exact allocation failed. Now try with alignment
2893 * turned on.
2894 */
2895 args.type = atype;
2896 args.fsbno = ap->rval;
2897 args.alignment = mp->m_dalign;
2898 args.minlen = nextminlen;
2899 args.minalignslop = 0;
2900 isaligned = 1;
2901 if ((error = xfs_alloc_vextent(&args)))
2902 return error;
2903 }
2904 if (isaligned && args.fsbno == NULLFSBLOCK) {
2905 /*
2906 * allocation failed, so turn off alignment and
2907 * try again.
2908 */
2909 args.type = atype;
2910 args.fsbno = ap->rval;
2911 args.alignment = 0;
2912 if ((error = xfs_alloc_vextent(&args)))
2913 return error;
2914 }
2915 if (args.fsbno == NULLFSBLOCK && nullfb &&
2916 args.minlen > ap->minlen) {
2917 args.minlen = ap->minlen;
2918 args.type = XFS_ALLOCTYPE_START_BNO;
2919 args.fsbno = ap->rval;
2920 if ((error = xfs_alloc_vextent(&args)))
2921 return error;
2922 }
2923 if (args.fsbno == NULLFSBLOCK && nullfb) {
2924 args.fsbno = 0;
2925 args.type = XFS_ALLOCTYPE_FIRST_AG;
2926 args.total = ap->minlen;
2927 args.minleft = 0;
2928 if ((error = xfs_alloc_vextent(&args)))
2929 return error;
2930 ap->low = 1;
2931 }
2932 if (args.fsbno != NULLFSBLOCK) {
2933 ap->firstblock = ap->rval = args.fsbno;
2934 ASSERT(nullfb || fb_agno == args.agno ||
2935 (ap->low && fb_agno < args.agno));
2936 ap->alen = args.len;
2937 ap->ip->i_d.di_nblocks += args.len;
2938 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2939 if (ap->wasdel)
2940 ap->ip->i_delayed_blks -= args.len;
2941 /*
2942 * Adjust the disk quota also. This was reserved
2943 * earlier.
2944 */
2945 XFS_TRANS_MOD_DQUOT_BYINO(mp, ap->tp, ap->ip,
2946 ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2947 XFS_TRANS_DQ_BCOUNT,
2948 (long) args.len);
2949 } else {
2950 ap->rval = NULLFSBLOCK;
2951 ap->alen = 0;
2952 }
2953 return 0;
2954 }
2955
2956 /*
2957 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2958 * It figures out where to ask the underlying allocator to put the new extent.
2959 */
2960 STATIC int
2961 xfs_bmap_alloc(
2962 xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2963 {
2964 if ((ap->ip->i_d.di_flags & XFS_DIFLAG_REALTIME) && ap->userdata)
2965 return xfs_bmap_rtalloc(ap);
2966 return xfs_bmap_btalloc(ap);
2967 }
2968
2969 /*
2970 * Transform a btree format file with only one leaf node, where the
2971 * extents list will fit in the inode, into an extents format file.
2972 * Since the file extents are already in-core, all we have to do is
2973 * give up the space for the btree root and pitch the leaf block.
2974 */
2975 STATIC int /* error */
2976 xfs_bmap_btree_to_extents(
2977 xfs_trans_t *tp, /* transaction pointer */
2978 xfs_inode_t *ip, /* incore inode pointer */
2979 xfs_btree_cur_t *cur, /* btree cursor */
2980 int *logflagsp, /* inode logging flags */
2981 int whichfork) /* data or attr fork */
2982 {
2983 /* REFERENCED */
2984 xfs_bmbt_block_t *cblock;/* child btree block */
2985 xfs_fsblock_t cbno; /* child block number */
2986 xfs_buf_t *cbp; /* child block's buffer */
2987 int error; /* error return value */
2988 xfs_ifork_t *ifp; /* inode fork data */
2989 xfs_mount_t *mp; /* mount point structure */
2990 __be64 *pp; /* ptr to block address */
2991 xfs_bmbt_block_t *rblock;/* root btree block */
2992
2993 ifp = XFS_IFORK_PTR(ip, whichfork);
2994 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2995 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
2996 rblock = ifp->if_broot;
2997 ASSERT(be16_to_cpu(rblock->bb_level) == 1);
2998 ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
2999 ASSERT(XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes) == 1);
3000 mp = ip->i_mount;
3001 pp = XFS_BMAP_BROOT_PTR_ADDR(rblock, 1, ifp->if_broot_bytes);
3002 cbno = be64_to_cpu(*pp);
3003 *logflagsp = 0;
3004 #ifdef DEBUG
3005 if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
3006 return error;
3007 #endif
3008 if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
3009 XFS_BMAP_BTREE_REF)))
3010 return error;
3011 cblock = XFS_BUF_TO_BMBT_BLOCK(cbp);
3012 if ((error = xfs_btree_check_lblock(cur, cblock, 0, cbp)))
3013 return error;
3014 xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
3015 ip->i_d.di_nblocks--;
3016 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
3017 xfs_trans_binval(tp, cbp);
3018 if (cur->bc_bufs[0] == cbp)
3019 cur->bc_bufs[0] = NULL;
3020 xfs_iroot_realloc(ip, -1, whichfork);
3021 ASSERT(ifp->if_broot == NULL);
3022 ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
3023 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3024 *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
3025 return 0;
3026 }
3027
3028 /*
3029 * Called by xfs_bmapi to update file extent records and the btree
3030 * after removing space (or undoing a delayed allocation).
3031 */
3032 STATIC int /* error */
3033 xfs_bmap_del_extent(
3034 xfs_inode_t *ip, /* incore inode pointer */
3035 xfs_trans_t *tp, /* current transaction pointer */
3036 xfs_extnum_t idx, /* extent number to update/delete */
3037 xfs_bmap_free_t *flist, /* list of extents to be freed */
3038 xfs_btree_cur_t *cur, /* if null, not a btree */
3039 xfs_bmbt_irec_t *del, /* data to remove from extents */
3040 int *logflagsp, /* inode logging flags */
3041 xfs_extdelta_t *delta, /* Change made to incore extents */
3042 int whichfork, /* data or attr fork */
3043 int rsvd) /* OK to allocate reserved blocks */
3044 {
3045 xfs_filblks_t da_new; /* new delay-alloc indirect blocks */
3046 xfs_filblks_t da_old; /* old delay-alloc indirect blocks */
3047 xfs_fsblock_t del_endblock=0; /* first block past del */
3048 xfs_fileoff_t del_endoff; /* first offset past del */
3049 int delay; /* current block is delayed allocated */
3050 int do_fx; /* free extent at end of routine */
3051 xfs_bmbt_rec_t *ep; /* current extent entry pointer */
3052 int error; /* error return value */
3053 int flags; /* inode logging flags */
3054 #ifdef XFS_BMAP_TRACE
3055 static char fname[] = "xfs_bmap_del_extent";
3056 #endif
3057 xfs_bmbt_irec_t got; /* current extent entry */
3058 xfs_fileoff_t got_endoff; /* first offset past got */
3059 int i; /* temp state */
3060 xfs_ifork_t *ifp; /* inode fork pointer */
3061 xfs_mount_t *mp; /* mount structure */
3062 xfs_filblks_t nblks; /* quota/sb block count */
3063 xfs_bmbt_irec_t new; /* new record to be inserted */
3064 /* REFERENCED */
3065 uint qfield; /* quota field to update */
3066 xfs_filblks_t temp; /* for indirect length calculations */
3067 xfs_filblks_t temp2; /* for indirect length calculations */
3068
3069 XFS_STATS_INC(xs_del_exlist);
3070 mp = ip->i_mount;
3071 ifp = XFS_IFORK_PTR(ip, whichfork);
3072 ASSERT((idx >= 0) && (idx < ifp->if_bytes /
3073 (uint)sizeof(xfs_bmbt_rec_t)));
3074 ASSERT(del->br_blockcount > 0);
3075 ep = xfs_iext_get_ext(ifp, idx);
3076 xfs_bmbt_get_all(ep, &got);
3077 ASSERT(got.br_startoff <= del->br_startoff);
3078 del_endoff = del->br_startoff + del->br_blockcount;
3079 got_endoff = got.br_startoff + got.br_blockcount;
3080 ASSERT(got_endoff >= del_endoff);
3081 delay = ISNULLSTARTBLOCK(got.br_startblock);
3082 ASSERT(ISNULLSTARTBLOCK(del->br_startblock) == delay);
3083 flags = 0;
3084 qfield = 0;
3085 error = 0;
3086 /*
3087 * If deleting a real allocation, must free up the disk space.
3088 */
3089 if (!delay) {
3090 flags = XFS_ILOG_CORE;
3091 /*
3092 * Realtime allocation. Free it and record di_nblocks update.
3093 */
3094 if (whichfork == XFS_DATA_FORK &&
3095 (ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
3096 xfs_fsblock_t bno;
3097 xfs_filblks_t len;
3098
3099 ASSERT(do_mod(del->br_blockcount,
3100 mp->m_sb.sb_rextsize) == 0);
3101 ASSERT(do_mod(del->br_startblock,
3102 mp->m_sb.sb_rextsize) == 0);
3103 bno = del->br_startblock;
3104 len = del->br_blockcount;
3105 do_div(bno, mp->m_sb.sb_rextsize);
3106 do_div(len, mp->m_sb.sb_rextsize);
3107 if ((error = xfs_rtfree_extent(ip->i_transp, bno,
3108 (xfs_extlen_t)len)))
3109 goto done;
3110 do_fx = 0;
3111 nblks = len * mp->m_sb.sb_rextsize;
3112 qfield = XFS_TRANS_DQ_RTBCOUNT;
3113 }
3114 /*
3115 * Ordinary allocation.
3116 */
3117 else {
3118 do_fx = 1;
3119 nblks = del->br_blockcount;
3120 qfield = XFS_TRANS_DQ_BCOUNT;
3121 }
3122 /*
3123 * Set up del_endblock and cur for later.
3124 */
3125 del_endblock = del->br_startblock + del->br_blockcount;
3126 if (cur) {
3127 if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
3128 got.br_startblock, got.br_blockcount,
3129 &i)))
3130 goto done;
3131 ASSERT(i == 1);
3132 }
3133 da_old = da_new = 0;
3134 } else {
3135 da_old = STARTBLOCKVAL(got.br_startblock);
3136 da_new = 0;
3137 nblks = 0;
3138 do_fx = 0;
3139 }
3140 /*
3141 * Set flag value to use in switch statement.
3142 * Left-contig is 2, right-contig is 1.
3143 */
3144 switch (((got.br_startoff == del->br_startoff) << 1) |
3145 (got_endoff == del_endoff)) {
3146 case 3:
3147 /*
3148 * Matches the whole extent. Delete the entry.
3149 */
3150 xfs_bmap_trace_delete(fname, "3", ip, idx, 1, whichfork);
3151 xfs_iext_remove(ifp, idx, 1);
3152 ifp->if_lastex = idx;
3153 if (delay)
3154 break;
3155 XFS_IFORK_NEXT_SET(ip, whichfork,
3156 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
3157 flags |= XFS_ILOG_CORE;
3158 if (!cur) {
3159 flags |= XFS_ILOG_FEXT(whichfork);
3160 break;
3161 }
3162 if ((error = xfs_bmbt_delete(cur, &i)))
3163 goto done;
3164 ASSERT(i == 1);
3165 break;
3166
3167 case 2:
3168 /*
3169 * Deleting the first part of the extent.
3170 */
3171 xfs_bmap_trace_pre_update(fname, "2", ip, idx, whichfork);
3172 xfs_bmbt_set_startoff(ep, del_endoff);
3173 temp = got.br_blockcount - del->br_blockcount;
3174 xfs_bmbt_set_blockcount(ep, temp);
3175 ifp->if_lastex = idx;
3176 if (delay) {
3177 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
3178 da_old);
3179 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3180 xfs_bmap_trace_post_update(fname, "2", ip, idx,
3181 whichfork);
3182 da_new = temp;
3183 break;
3184 }
3185 xfs_bmbt_set_startblock(ep, del_endblock);
3186 xfs_bmap_trace_post_update(fname, "2", ip, idx, whichfork);
3187 if (!cur) {
3188 flags |= XFS_ILOG_FEXT(whichfork);
3189 break;
3190 }
3191 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
3192 got.br_blockcount - del->br_blockcount,
3193 got.br_state)))
3194 goto done;
3195 break;
3196
3197 case 1:
3198 /*
3199 * Deleting the last part of the extent.
3200 */
3201 temp = got.br_blockcount - del->br_blockcount;
3202 xfs_bmap_trace_pre_update(fname, "1", ip, idx, whichfork);
3203 xfs_bmbt_set_blockcount(ep, temp);
3204 ifp->if_lastex = idx;
3205 if (delay) {
3206 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
3207 da_old);
3208 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3209 xfs_bmap_trace_post_update(fname, "1", ip, idx,
3210 whichfork);
3211 da_new = temp;
3212 break;
3213 }
3214 xfs_bmap_trace_post_update(fname, "1", ip, idx, whichfork);
3215 if (!cur) {
3216 flags |= XFS_ILOG_FEXT(whichfork);
3217 break;
3218 }
3219 if ((error = xfs_bmbt_update(cur, got.br_startoff,
3220 got.br_startblock,
3221 got.br_blockcount - del->br_blockcount,
3222 got.br_state)))
3223 goto done;
3224 break;
3225
3226 case 0:
3227 /*
3228 * Deleting the middle of the extent.
3229 */
3230 temp = del->br_startoff - got.br_startoff;
3231 xfs_bmap_trace_pre_update(fname, "0", ip, idx, whichfork);
3232 xfs_bmbt_set_blockcount(ep, temp);
3233 new.br_startoff = del_endoff;
3234 temp2 = got_endoff - del_endoff;
3235 new.br_blockcount = temp2;
3236 new.br_state = got.br_state;
3237 if (!delay) {
3238 new.br_startblock = del_endblock;
3239 flags |= XFS_ILOG_CORE;
3240 if (cur) {
3241 if ((error = xfs_bmbt_update(cur,
3242 got.br_startoff,
3243 got.br_startblock, temp,
3244 got.br_state)))
3245 goto done;
3246 if ((error = xfs_bmbt_increment(cur, 0, &i)))
3247 goto done;
3248 cur->bc_rec.b = new;
3249 error = xfs_bmbt_insert(cur, &i);
3250 if (error && error != ENOSPC)
3251 goto done;
3252 /*
3253 * If get no-space back from btree insert,
3254 * it tried a split, and we have a zero
3255 * block reservation.
3256 * Fix up our state and return the error.
3257 */
3258 if (error == ENOSPC) {
3259 /*
3260 * Reset the cursor, don't trust
3261 * it after any insert operation.
3262 */
3263 if ((error = xfs_bmbt_lookup_eq(cur,
3264 got.br_startoff,
3265 got.br_startblock,
3266 temp, &i)))
3267 goto done;
3268 ASSERT(i == 1);
3269 /*
3270 * Update the btree record back
3271 * to the original value.
3272 */
3273 if ((error = xfs_bmbt_update(cur,
3274 got.br_startoff,
3275 got.br_startblock,
3276 got.br_blockcount,
3277 got.br_state)))
3278 goto done;
3279 /*
3280 * Reset the extent record back
3281 * to the original value.
3282 */
3283 xfs_bmbt_set_blockcount(ep,
3284 got.br_blockcount);
3285 flags = 0;
3286 error = XFS_ERROR(ENOSPC);
3287 goto done;
3288 }
3289 ASSERT(i == 1);
3290 } else
3291 flags |= XFS_ILOG_FEXT(whichfork);
3292 XFS_IFORK_NEXT_SET(ip, whichfork,
3293 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
3294 } else {
3295 ASSERT(whichfork == XFS_DATA_FORK);
3296 temp = xfs_bmap_worst_indlen(ip, temp);
3297 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3298 temp2 = xfs_bmap_worst_indlen(ip, temp2);
3299 new.br_startblock = NULLSTARTBLOCK((int)temp2);
3300 da_new = temp + temp2;
3301 while (da_new > da_old) {
3302 if (temp) {
3303 temp--;
3304 da_new--;
3305 xfs_bmbt_set_startblock(ep,
3306 NULLSTARTBLOCK((int)temp));
3307 }
3308 if (da_new == da_old)
3309 break;
3310 if (temp2) {
3311 temp2--;
3312 da_new--;
3313 new.br_startblock =
3314 NULLSTARTBLOCK((int)temp2);
3315 }
3316 }
3317 }
3318 xfs_bmap_trace_post_update(fname, "0", ip, idx, whichfork);
3319 xfs_bmap_trace_insert(fname, "0", ip, idx + 1, 1, &new, NULL,
3320 whichfork);
3321 xfs_iext_insert(ifp, idx + 1, 1, &new);
3322 ifp->if_lastex = idx + 1;
3323 break;
3324 }
3325 /*
3326 * If we need to, add to list of extents to delete.
3327 */
3328 if (do_fx)
3329 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
3330 mp);
3331 /*
3332 * Adjust inode # blocks in the file.
3333 */
3334 if (nblks)
3335 ip->i_d.di_nblocks -= nblks;
3336 /*
3337 * Adjust quota data.
3338 */
3339 if (qfield)
3340 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, qfield, (long)-nblks);
3341
3342 /*
3343 * Account for change in delayed indirect blocks.
3344 * Nothing to do for disk quota accounting here.
3345 */
3346 ASSERT(da_old >= da_new);
3347 if (da_old > da_new)
3348 xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS, (int64_t)(da_old - da_new),
3349 rsvd);
3350 if (delta) {
3351 /* DELTA: report the original extent. */
3352 if (delta->xed_startoff > got.br_startoff)
3353 delta->xed_startoff = got.br_startoff;
3354 if (delta->xed_blockcount < got.br_startoff+got.br_blockcount)
3355 delta->xed_blockcount = got.br_startoff +
3356 got.br_blockcount;
3357 }
3358 done:
3359 *logflagsp = flags;
3360 return error;
3361 }
3362
3363 /*
3364 * Remove the entry "free" from the free item list. Prev points to the
3365 * previous entry, unless "free" is the head of the list.
3366 */
3367 STATIC void
3368 xfs_bmap_del_free(
3369 xfs_bmap_free_t *flist, /* free item list header */
3370 xfs_bmap_free_item_t *prev, /* previous item on list, if any */
3371 xfs_bmap_free_item_t *free) /* list item to be freed */
3372 {
3373 if (prev)
3374 prev->xbfi_next = free->xbfi_next;
3375 else
3376 flist->xbf_first = free->xbfi_next;
3377 flist->xbf_count--;
3378 kmem_zone_free(xfs_bmap_free_item_zone, free);
3379 }
3380
3381 /*
3382 * Convert an extents-format file into a btree-format file.
3383 * The new file will have a root block (in the inode) and a single child block.
3384 */
3385 STATIC int /* error */
3386 xfs_bmap_extents_to_btree(
3387 xfs_trans_t *tp, /* transaction pointer */
3388 xfs_inode_t *ip, /* incore inode pointer */
3389 xfs_fsblock_t *firstblock, /* first-block-allocated */
3390 xfs_bmap_free_t *flist, /* blocks freed in xaction */
3391 xfs_btree_cur_t **curp, /* cursor returned to caller */
3392 int wasdel, /* converting a delayed alloc */
3393 int *logflagsp, /* inode logging flags */
3394 int whichfork) /* data or attr fork */
3395 {
3396 xfs_bmbt_block_t *ablock; /* allocated (child) bt block */
3397 xfs_buf_t *abp; /* buffer for ablock */
3398 xfs_alloc_arg_t args; /* allocation arguments */
3399 xfs_bmbt_rec_t *arp; /* child record pointer */
3400 xfs_bmbt_block_t *block; /* btree root block */
3401 xfs_btree_cur_t *cur; /* bmap btree cursor */
3402 xfs_bmbt_rec_t *ep; /* extent record pointer */
3403 int error; /* error return value */
3404 xfs_extnum_t i, cnt; /* extent record index */
3405 xfs_ifork_t *ifp; /* inode fork pointer */
3406 xfs_bmbt_key_t *kp; /* root block key pointer */
3407 xfs_mount_t *mp; /* mount structure */
3408 xfs_extnum_t nextents; /* number of file extents */
3409 xfs_bmbt_ptr_t *pp; /* root block address pointer */
3410
3411 ifp = XFS_IFORK_PTR(ip, whichfork);
3412 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3413 ASSERT(ifp->if_ext_max ==
3414 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
3415 /*
3416 * Make space in the inode incore.
3417 */
3418 xfs_iroot_realloc(ip, 1, whichfork);
3419 ifp->if_flags |= XFS_IFBROOT;
3420 /*
3421 * Fill in the root.
3422 */
3423 block = ifp->if_broot;
3424 block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3425 block->bb_level = cpu_to_be16(1);
3426 block->bb_numrecs = cpu_to_be16(1);
3427 block->bb_leftsib = cpu_to_be64(NULLDFSBNO);
3428 block->bb_rightsib = cpu_to_be64(NULLDFSBNO);
3429 /*
3430 * Need a cursor. Can't allocate until bb_level is filled in.
3431 */
3432 mp = ip->i_mount;
3433 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
3434 whichfork);
3435 cur->bc_private.b.firstblock = *firstblock;
3436 cur->bc_private.b.flist = flist;
3437 cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3438 /*
3439 * Convert to a btree with two levels, one record in root.
3440 */
3441 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3442 args.tp = tp;
3443 args.mp = mp;
3444 args.firstblock = *firstblock;
3445 if (*firstblock == NULLFSBLOCK) {
3446 args.type = XFS_ALLOCTYPE_START_BNO;
3447 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3448 } else if (flist->xbf_low) {
3449 args.type = XFS_ALLOCTYPE_START_BNO;
3450 args.fsbno = *firstblock;
3451 } else {
3452 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3453 args.fsbno = *firstblock;
3454 }
3455 args.minlen = args.maxlen = args.prod = 1;
3456 args.total = args.minleft = args.alignment = args.mod = args.isfl =
3457 args.minalignslop = 0;
3458 args.wasdel = wasdel;
3459 *logflagsp = 0;
3460 if ((error = xfs_alloc_vextent(&args))) {
3461 xfs_iroot_realloc(ip, -1, whichfork);
3462 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3463 return error;
3464 }
3465 /*
3466 * Allocation can't fail, the space was reserved.
3467 */
3468 ASSERT(args.fsbno != NULLFSBLOCK);
3469 ASSERT(*firstblock == NULLFSBLOCK ||
3470 args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3471 (flist->xbf_low &&
3472 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3473 *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3474 cur->bc_private.b.allocated++;
3475 ip->i_d.di_nblocks++;
3476 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3477 abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3478 /*
3479 * Fill in the child block.
3480 */
3481 ablock = XFS_BUF_TO_BMBT_BLOCK(abp);
3482 ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3483 ablock->bb_level = 0;
3484 ablock->bb_leftsib = cpu_to_be64(NULLDFSBNO);
3485 ablock->bb_rightsib = cpu_to_be64(NULLDFSBNO);
3486 arp = XFS_BMAP_REC_IADDR(ablock, 1, cur);
3487 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3488 for (cnt = i = 0; i < nextents; i++) {
3489 ep = xfs_iext_get_ext(ifp, i);
3490 if (!ISNULLSTARTBLOCK(xfs_bmbt_get_startblock(ep))) {
3491 arp->l0 = INT_GET(ep->l0, ARCH_CONVERT);
3492 arp->l1 = INT_GET(ep->l1, ARCH_CONVERT);
3493 arp++; cnt++;
3494 }
3495 }
3496 ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3497 ablock->bb_numrecs = cpu_to_be16(cnt);
3498 /*
3499 * Fill in the root key and pointer.
3500 */
3501 kp = XFS_BMAP_KEY_IADDR(block, 1, cur);
3502 arp = XFS_BMAP_REC_IADDR(ablock, 1, cur);
3503 kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
3504 pp = XFS_BMAP_PTR_IADDR(block, 1, cur);
3505 *pp = cpu_to_be64(args.fsbno);
3506 /*
3507 * Do all this logging at the end so that
3508 * the root is at the right level.
3509 */
3510 xfs_bmbt_log_block(cur, abp, XFS_BB_ALL_BITS);
3511 xfs_bmbt_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3512 ASSERT(*curp == NULL);
3513 *curp = cur;
3514 *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FBROOT(whichfork);
3515 return 0;
3516 }
3517
3518 /*
3519 * Helper routine to reset inode di_forkoff field when switching
3520 * attribute fork from local to extent format - we reset it where
3521 * possible to make space available for inline data fork extents.
3522 */
3523 STATIC void
3524 xfs_bmap_forkoff_reset(
3525 xfs_mount_t *mp,
3526 xfs_inode_t *ip,
3527 int whichfork)
3528 {
3529 if (whichfork == XFS_ATTR_FORK &&
3530 (ip->i_d.di_format != XFS_DINODE_FMT_DEV) &&
3531 (ip->i_d.di_format != XFS_DINODE_FMT_UUID) &&
3532 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3533 ((mp->m_attroffset >> 3) > ip->i_d.di_forkoff)) {
3534 ip->i_d.di_forkoff = mp->m_attroffset >> 3;
3535 ip->i_df.if_ext_max = XFS_IFORK_DSIZE(ip) /
3536 (uint)sizeof(xfs_bmbt_rec_t);
3537 ip->i_afp->if_ext_max = XFS_IFORK_ASIZE(ip) /
3538 (uint)sizeof(xfs_bmbt_rec_t);
3539 }
3540 }
3541
3542 /*
3543 * Convert a local file to an extents file.
3544 * This code is out of bounds for data forks of regular files,
3545 * since the file data needs to get logged so things will stay consistent.
3546 * (The bmap-level manipulations are ok, though).
3547 */
3548 STATIC int /* error */
3549 xfs_bmap_local_to_extents(
3550 xfs_trans_t *tp, /* transaction pointer */
3551 xfs_inode_t *ip, /* incore inode pointer */
3552 xfs_fsblock_t *firstblock, /* first block allocated in xaction */
3553 xfs_extlen_t total, /* total blocks needed by transaction */
3554 int *logflagsp, /* inode logging flags */
3555 int whichfork) /* data or attr fork */
3556 {
3557 int error; /* error return value */
3558 int flags; /* logging flags returned */
3559 #ifdef XFS_BMAP_TRACE
3560 static char fname[] = "xfs_bmap_local_to_extents";
3561 #endif
3562 xfs_ifork_t *ifp; /* inode fork pointer */
3563
3564 /*
3565 * We don't want to deal with the case of keeping inode data inline yet.
3566 * So sending the data fork of a regular inode is invalid.
3567 */
3568 ASSERT(!((ip->i_d.di_mode & S_IFMT) == S_IFREG &&
3569 whichfork == XFS_DATA_FORK));
3570 ifp = XFS_IFORK_PTR(ip, whichfork);
3571 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3572 flags = 0;
3573 error = 0;
3574 if (ifp->if_bytes) {
3575 xfs_alloc_arg_t args; /* allocation arguments */
3576 xfs_buf_t *bp; /* buffer for extent block */
3577 xfs_bmbt_rec_t *ep; /* extent record pointer */
3578
3579 args.tp = tp;
3580 args.mp = ip->i_mount;
3581 args.firstblock = *firstblock;
3582 ASSERT((ifp->if_flags &
3583 (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == XFS_IFINLINE);
3584 /*
3585 * Allocate a block. We know we need only one, since the
3586 * file currently fits in an inode.
3587 */
3588 if (*firstblock == NULLFSBLOCK) {
3589 args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3590 args.type = XFS_ALLOCTYPE_START_BNO;
3591 } else {
3592 args.fsbno = *firstblock;
3593 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3594 }
3595 args.total = total;
3596 args.mod = args.minleft = args.alignment = args.wasdel =
3597 args.isfl = args.minalignslop = 0;
3598 args.minlen = args.maxlen = args.prod = 1;
3599 if ((error = xfs_alloc_vextent(&args)))
3600 goto done;
3601 /*
3602 * Can't fail, the space was reserved.
3603 */
3604 ASSERT(args.fsbno != NULLFSBLOCK);
3605 ASSERT(args.len == 1);
3606 *firstblock = args.fsbno;
3607 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3608 memcpy((char *)XFS_BUF_PTR(bp), ifp->if_u1.if_data,
3609 ifp->if_bytes);
3610 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3611 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3612 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3613 xfs_iext_add(ifp, 0, 1);
3614 ep = xfs_iext_get_ext(ifp, 0);
3615 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3616 xfs_bmap_trace_post_update(fname, "new", ip, 0, whichfork);
3617 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3618 ip->i_d.di_nblocks = 1;
3619 XFS_TRANS_MOD_DQUOT_BYINO(args.mp, tp, ip,
3620 XFS_TRANS_DQ_BCOUNT, 1L);
3621 flags |= XFS_ILOG_FEXT(whichfork);
3622 } else {
3623 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3624 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3625 }
3626 ifp->if_flags &= ~XFS_IFINLINE;
3627 ifp->if_flags |= XFS_IFEXTENTS;
3628 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3629 flags |= XFS_ILOG_CORE;
3630 done:
3631 *logflagsp = flags;
3632 return error;
3633 }
3634
3635 /*
3636 * Search the extent records for the entry containing block bno.
3637 * If bno lies in a hole, point to the next entry. If bno lies
3638 * past eof, *eofp will be set, and *prevp will contain the last
3639 * entry (null if none). Else, *lastxp will be set to the index
3640 * of the found entry; *gotp will contain the entry.
3641 */
3642 xfs_bmbt_rec_t * /* pointer to found extent entry */
3643 xfs_bmap_search_multi_extents(
3644 xfs_ifork_t *ifp, /* inode fork pointer */
3645 xfs_fileoff_t bno, /* block number searched for */
3646 int *eofp, /* out: end of file found */
3647 xfs_extnum_t *lastxp, /* out: last extent index */
3648 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
3649 xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
3650 {
3651 xfs_bmbt_rec_t *ep; /* extent record pointer */
3652 xfs_extnum_t lastx; /* last extent index */
3653
3654 /*
3655 * Initialize the extent entry structure to catch access to
3656 * uninitialized br_startblock field.
3657 */
3658 gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
3659 gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3660 gotp->br_state = XFS_EXT_INVALID;
3661 #if XFS_BIG_BLKNOS
3662 gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
3663 #else
3664 gotp->br_startblock = 0xffffa5a5;
3665 #endif
3666 prevp->br_startoff = NULLFILEOFF;
3667
3668 ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
3669 if (lastx > 0) {
3670 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
3671 }
3672 if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
3673 xfs_bmbt_get_all(ep, gotp);
3674 *eofp = 0;
3675 } else {
3676 if (lastx > 0) {
3677 *gotp = *prevp;
3678 }
3679 *eofp = 1;
3680 ep = NULL;
3681 }
3682 *lastxp = lastx;
3683 return ep;
3684 }
3685
3686 /*
3687 * Search the extents list for the inode, for the extent containing bno.
3688 * If bno lies in a hole, point to the next entry. If bno lies past eof,
3689 * *eofp will be set, and *prevp will contain the last entry (null if none).
3690 * Else, *lastxp will be set to the index of the found
3691 * entry; *gotp will contain the entry.
3692 */
3693 STATIC xfs_bmbt_rec_t * /* pointer to found extent entry */
3694 xfs_bmap_search_extents(
3695 xfs_inode_t *ip, /* incore inode pointer */
3696 xfs_fileoff_t bno, /* block number searched for */
3697 int fork, /* data or attr fork */
3698 int *eofp, /* out: end of file found */
3699 xfs_extnum_t *lastxp, /* out: last extent index */
3700 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
3701 xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
3702 {
3703 xfs_ifork_t *ifp; /* inode fork pointer */
3704 xfs_bmbt_rec_t *ep; /* extent record pointer */
3705
3706 XFS_STATS_INC(xs_look_exlist);
3707 ifp = XFS_IFORK_PTR(ip, fork);
3708
3709 ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
3710
3711 if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
3712 !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
3713 xfs_cmn_err(XFS_PTAG_FSBLOCK_ZERO, CE_ALERT, ip->i_mount,
3714 "Access to block zero in inode %llu "
3715 "start_block: %llx start_off: %llx "
3716 "blkcnt: %llx extent-state: %x lastx: %x\n",
3717 (unsigned long long)ip->i_ino,
3718 (unsigned long long)gotp->br_startblock,
3719 (unsigned long long)gotp->br_startoff,
3720 (unsigned long long)gotp->br_blockcount,
3721 gotp->br_state, *lastxp);
3722 *lastxp = NULLEXTNUM;
3723 *eofp = 1;
3724 return NULL;
3725 }
3726 return ep;
3727 }
3728
3729
3730 #ifdef XFS_BMAP_TRACE
3731 ktrace_t *xfs_bmap_trace_buf;
3732
3733 /*
3734 * Add a bmap trace buffer entry. Base routine for the others.
3735 */
3736 STATIC void
3737 xfs_bmap_trace_addentry(
3738 int opcode, /* operation */
3739 char *fname, /* function name */
3740 char *desc, /* operation description */
3741 xfs_inode_t *ip, /* incore inode pointer */
3742 xfs_extnum_t idx, /* index of entry(ies) */
3743 xfs_extnum_t cnt, /* count of entries, 1 or 2 */
3744 xfs_bmbt_rec_t *r1, /* first record */
3745 xfs_bmbt_rec_t *r2, /* second record or null */
3746 int whichfork) /* data or attr fork */
3747 {
3748 xfs_bmbt_rec_t tr2;
3749
3750 ASSERT(cnt == 1 || cnt == 2);
3751 ASSERT(r1 != NULL);
3752 if (cnt == 1) {
3753 ASSERT(r2 == NULL);
3754 r2 = &tr2;
3755 memset(&tr2, 0, sizeof(tr2));
3756 } else
3757 ASSERT(r2 != NULL);
3758 ktrace_enter(xfs_bmap_trace_buf,
3759 (void *)(__psint_t)(opcode | (whichfork << 16)),
3760 (void *)fname, (void *)desc, (void *)ip,
3761 (void *)(__psint_t)idx,
3762 (void *)(__psint_t)cnt,
3763 (void *)(__psunsigned_t)(ip->i_ino >> 32),
3764 (void *)(__psunsigned_t)(unsigned)ip->i_ino,
3765 (void *)(__psunsigned_t)(r1->l0 >> 32),
3766 (void *)(__psunsigned_t)(unsigned)(r1->l0),
3767 (void *)(__psunsigned_t)(r1->l1 >> 32),
3768 (void *)(__psunsigned_t)(unsigned)(r1->l1),
3769 (void *)(__psunsigned_t)(r2->l0 >> 32),
3770 (void *)(__psunsigned_t)(unsigned)(r2->l0),
3771 (void *)(__psunsigned_t)(r2->l1 >> 32),
3772 (void *)(__psunsigned_t)(unsigned)(r2->l1)
3773 );
3774 ASSERT(ip->i_xtrace);
3775 ktrace_enter(ip->i_xtrace,
3776 (void *)(__psint_t)(opcode | (whichfork << 16)),
3777 (void *)fname, (void *)desc, (void *)ip,
3778 (void *)(__psint_t)idx,
3779 (void *)(__psint_t)cnt,
3780 (void *)(__psunsigned_t)(ip->i_ino >> 32),
3781 (void *)(__psunsigned_t)(unsigned)ip->i_ino,
3782 (void *)(__psunsigned_t)(r1->l0 >> 32),
3783 (void *)(__psunsigned_t)(unsigned)(r1->l0),
3784 (void *)(__psunsigned_t)(r1->l1 >> 32),
3785 (void *)(__psunsigned_t)(unsigned)(r1->l1),
3786 (void *)(__psunsigned_t)(r2->l0 >> 32),
3787 (void *)(__psunsigned_t)(unsigned)(r2->l0),
3788 (void *)(__psunsigned_t)(r2->l1 >> 32),
3789 (void *)(__psunsigned_t)(unsigned)(r2->l1)
3790 );
3791 }
3792
3793 /*
3794 * Add bmap trace entry prior to a call to xfs_iext_remove.
3795 */
3796 STATIC void
3797 xfs_bmap_trace_delete(
3798 char *fname, /* function name */
3799 char *desc, /* operation description */
3800 xfs_inode_t *ip, /* incore inode pointer */
3801 xfs_extnum_t idx, /* index of entry(entries) deleted */
3802 xfs_extnum_t cnt, /* count of entries deleted, 1 or 2 */
3803 int whichfork) /* data or attr fork */
3804 {
3805 xfs_ifork_t *ifp; /* inode fork pointer */
3806
3807 ifp = XFS_IFORK_PTR(ip, whichfork);
3808 xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_DELETE, fname, desc, ip, idx,
3809 cnt, xfs_iext_get_ext(ifp, idx),
3810 cnt == 2 ? xfs_iext_get_ext(ifp, idx + 1) : NULL,
3811 whichfork);
3812 }
3813
3814 /*
3815 * Add bmap trace entry prior to a call to xfs_iext_insert, or
3816 * reading in the extents list from the disk (in the btree).
3817 */
3818 STATIC void
3819 xfs_bmap_trace_insert(
3820 char *fname, /* function name */
3821 char *desc, /* operation description */
3822 xfs_inode_t *ip, /* incore inode pointer */
3823 xfs_extnum_t idx, /* index of entry(entries) inserted */
3824 xfs_extnum_t cnt, /* count of entries inserted, 1 or 2 */
3825 xfs_bmbt_irec_t *r1, /* inserted record 1 */
3826 xfs_bmbt_irec_t *r2, /* inserted record 2 or null */
3827 int whichfork) /* data or attr fork */
3828 {
3829 xfs_bmbt_rec_t tr1; /* compressed record 1 */
3830 xfs_bmbt_rec_t tr2; /* compressed record 2 if needed */
3831
3832 xfs_bmbt_set_all(&tr1, r1);
3833 if (cnt == 2) {
3834 ASSERT(r2 != NULL);
3835 xfs_bmbt_set_all(&tr2, r2);
3836 } else {
3837 ASSERT(cnt == 1);
3838 ASSERT(r2 == NULL);
3839 }
3840 xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_INSERT, fname, desc, ip, idx,
3841 cnt, &tr1, cnt == 2 ? &tr2 : NULL, whichfork);
3842 }
3843
3844 /*
3845 * Add bmap trace entry after updating an extent record in place.
3846 */
3847 STATIC void
3848 xfs_bmap_trace_post_update(
3849 char *fname, /* function name */
3850 char *desc, /* operation description */
3851 xfs_inode_t *ip, /* incore inode pointer */
3852 xfs_extnum_t idx, /* index of entry updated */
3853 int whichfork) /* data or attr fork */
3854 {
3855 xfs_ifork_t *ifp; /* inode fork pointer */
3856
3857 ifp = XFS_IFORK_PTR(ip, whichfork);
3858 xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_POST_UP, fname, desc, ip, idx,
3859 1, xfs_iext_get_ext(ifp, idx), NULL, whichfork);
3860 }
3861
3862 /*
3863 * Add bmap trace entry prior to updating an extent record in place.
3864 */
3865 STATIC void
3866 xfs_bmap_trace_pre_update(
3867 char *fname, /* function name */
3868 char *desc, /* operation description */
3869 xfs_inode_t *ip, /* incore inode pointer */
3870 xfs_extnum_t idx, /* index of entry to be updated */
3871 int whichfork) /* data or attr fork */
3872 {
3873 xfs_ifork_t *ifp; /* inode fork pointer */
3874
3875 ifp = XFS_IFORK_PTR(ip, whichfork);
3876 xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_PRE_UP, fname, desc, ip, idx, 1,
3877 xfs_iext_get_ext(ifp, idx), NULL, whichfork);
3878 }
3879 #endif /* XFS_BMAP_TRACE */
3880
3881 /*
3882 * Compute the worst-case number of indirect blocks that will be used
3883 * for ip's delayed extent of length "len".
3884 */
3885 STATIC xfs_filblks_t
3886 xfs_bmap_worst_indlen(
3887 xfs_inode_t *ip, /* incore inode pointer */
3888 xfs_filblks_t len) /* delayed extent length */
3889 {
3890 int level; /* btree level number */
3891 int maxrecs; /* maximum record count at this level */
3892 xfs_mount_t *mp; /* mount structure */
3893 xfs_filblks_t rval; /* return value */
3894
3895 mp = ip->i_mount;
3896 maxrecs = mp->m_bmap_dmxr[0];
3897 for (level = 0, rval = 0;
3898 level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3899 level++) {
3900 len += maxrecs - 1;
3901 do_div(len, maxrecs);
3902 rval += len;
3903 if (len == 1)
3904 return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3905 level - 1;
3906 if (level == 0)
3907 maxrecs = mp->m_bmap_dmxr[1];
3908 }
3909 return rval;
3910 }
3911
3912 #if defined(XFS_RW_TRACE)
3913 STATIC void
3914 xfs_bunmap_trace(
3915 xfs_inode_t *ip,
3916 xfs_fileoff_t bno,
3917 xfs_filblks_t len,
3918 int flags,
3919 inst_t *ra)
3920 {
3921 if (ip->i_rwtrace == NULL)
3922 return;
3923 ktrace_enter(ip->i_rwtrace,
3924 (void *)(__psint_t)XFS_BUNMAP,
3925 (void *)ip,
3926 (void *)(__psint_t)((ip->i_d.di_size >> 32) & 0xffffffff),
3927 (void *)(__psint_t)(ip->i_d.di_size & 0xffffffff),
3928 (void *)(__psint_t)(((xfs_dfiloff_t)bno >> 32) & 0xffffffff),
3929 (void *)(__psint_t)((xfs_dfiloff_t)bno & 0xffffffff),
3930 (void *)(__psint_t)len,
3931 (void *)(__psint_t)flags,
3932 (void *)(unsigned long)current_cpu(),
3933 (void *)ra,
3934 (void *)0,
3935 (void *)0,
3936 (void *)0,
3937 (void *)0,
3938 (void *)0,
3939 (void *)0);
3940 }
3941 #endif
3942
3943 /*
3944 * Convert inode from non-attributed to attributed.
3945 * Must not be in a transaction, ip must not be locked.
3946 */
3947 int /* error code */
3948 xfs_bmap_add_attrfork(
3949 xfs_inode_t *ip, /* incore inode pointer */
3950 int size, /* space new attribute needs */
3951 int rsvd) /* xact may use reserved blks */
3952 {
3953 xfs_fsblock_t firstblock; /* 1st block/ag allocated */
3954 xfs_bmap_free_t flist; /* freed extent records */
3955 xfs_mount_t *mp; /* mount structure */
3956 xfs_trans_t *tp; /* transaction pointer */
3957 unsigned long s; /* spinlock spl value */
3958 int blks; /* space reservation */
3959 int version = 1; /* superblock attr version */
3960 int committed; /* xaction was committed */
3961 int logflags; /* logging flags */
3962 int error; /* error return value */
3963
3964 ASSERT(XFS_IFORK_Q(ip) == 0);
3965 ASSERT(ip->i_df.if_ext_max ==
3966 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3967
3968 mp = ip->i_mount;
3969 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3970 tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3971 blks = XFS_ADDAFORK_SPACE_RES(mp);
3972 if (rsvd)
3973 tp->t_flags |= XFS_TRANS_RESERVE;
3974 if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3975 XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3976 goto error0;
3977 xfs_ilock(ip, XFS_ILOCK_EXCL);
3978 error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip, blks, 0, rsvd ?
3979 XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3980 XFS_QMOPT_RES_REGBLKS);
3981 if (error) {
3982 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3983 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3984 return error;
3985 }
3986 if (XFS_IFORK_Q(ip))
3987 goto error1;
3988 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3989 /*
3990 * For inodes coming from pre-6.2 filesystems.
3991 */
3992 ASSERT(ip->i_d.di_aformat == 0);
3993 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
3994 }
3995 ASSERT(ip->i_d.di_anextents == 0);
3996 VN_HOLD(XFS_ITOV(ip));
3997 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3998 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3999 switch (ip->i_d.di_format) {
4000 case XFS_DINODE_FMT_DEV:
4001 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
4002 break;
4003 case XFS_DINODE_FMT_UUID:
4004 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
4005 break;
4006 case XFS_DINODE_FMT_LOCAL:
4007 case XFS_DINODE_FMT_EXTENTS:
4008 case XFS_DINODE_FMT_BTREE:
4009 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
4010 if (!ip->i_d.di_forkoff)
4011 ip->i_d.di_forkoff = mp->m_attroffset >> 3;
4012 else if (mp->m_flags & XFS_MOUNT_ATTR2)
4013 version = 2;
4014 break;
4015 default:
4016 ASSERT(0);
4017 error = XFS_ERROR(EINVAL);
4018 goto error1;
4019 }
4020 ip->i_df.if_ext_max =
4021 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
4022 ASSERT(ip->i_afp == NULL);
4023 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
4024 ip->i_afp->if_ext_max =
4025 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
4026 ip->i_afp->if_flags = XFS_IFEXTENTS;
4027 logflags = 0;
4028 XFS_BMAP_INIT(&flist, &firstblock);
4029 switch (ip->i_d.di_format) {
4030 case XFS_DINODE_FMT_LOCAL:
4031 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
4032 &logflags);
4033 break;
4034 case XFS_DINODE_FMT_EXTENTS:
4035 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
4036 &flist, &logflags);
4037 break;
4038 case XFS_DINODE_FMT_BTREE:
4039 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
4040 &logflags);
4041 break;
4042 default:
4043 error = 0;
4044 break;
4045 }
4046 if (logflags)
4047 xfs_trans_log_inode(tp, ip, logflags);
4048 if (error)
4049 goto error2;
4050 if (!XFS_SB_VERSION_HASATTR(&mp->m_sb) ||
4051 (!XFS_SB_VERSION_HASATTR2(&mp->m_sb) && version == 2)) {
4052 __int64_t sbfields = 0;
4053
4054 s = XFS_SB_LOCK(mp);
4055 if (!XFS_SB_VERSION_HASATTR(&mp->m_sb)) {
4056 XFS_SB_VERSION_ADDATTR(&mp->m_sb);
4057 sbfields |= XFS_SB_VERSIONNUM;
4058 }
4059 if (!XFS_SB_VERSION_HASATTR2(&mp->m_sb) && version == 2) {
4060 XFS_SB_VERSION_ADDATTR2(&mp->m_sb);
4061 sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
4062 }
4063 if (sbfields) {
4064 XFS_SB_UNLOCK(mp, s);
4065 xfs_mod_sb(tp, sbfields);
4066 } else
4067 XFS_SB_UNLOCK(mp, s);
4068 }
4069 if ((error = xfs_bmap_finish(&tp, &flist, &committed)))
4070 goto error2;
4071 error = xfs_trans_commit(tp, XFS_TRANS_PERM_LOG_RES);
4072 ASSERT(ip->i_df.if_ext_max ==
4073 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
4074 return error;
4075 error2:
4076 xfs_bmap_cancel(&flist);
4077 error1:
4078 ASSERT(ismrlocked(&ip->i_lock,MR_UPDATE));
4079 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4080 error0:
4081 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
4082 ASSERT(ip->i_df.if_ext_max ==
4083 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
4084 return error;
4085 }
4086
4087 /*
4088 * Add the extent to the list of extents to be free at transaction end.
4089 * The list is maintained sorted (by block number).
4090 */
4091 /* ARGSUSED */
4092 void
4093 xfs_bmap_add_free(
4094 xfs_fsblock_t bno, /* fs block number of extent */
4095 xfs_filblks_t len, /* length of extent */
4096 xfs_bmap_free_t *flist, /* list of extents */
4097 xfs_mount_t *mp) /* mount point structure */
4098 {
4099 xfs_bmap_free_item_t *cur; /* current (next) element */
4100 xfs_bmap_free_item_t *new; /* new element */
4101 xfs_bmap_free_item_t *prev; /* previous element */
4102 #ifdef DEBUG
4103 xfs_agnumber_t agno;
4104 xfs_agblock_t agbno;
4105
4106 ASSERT(bno != NULLFSBLOCK);
4107 ASSERT(len > 0);
4108 ASSERT(len <= MAXEXTLEN);
4109 ASSERT(!ISNULLSTARTBLOCK(bno));
4110 agno = XFS_FSB_TO_AGNO(mp, bno);
4111 agbno = XFS_FSB_TO_AGBNO(mp, bno);
4112 ASSERT(agno < mp->m_sb.sb_agcount);
4113 ASSERT(agbno < mp->m_sb.sb_agblocks);
4114 ASSERT(len < mp->m_sb.sb_agblocks);
4115 ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
4116 #endif
4117 ASSERT(xfs_bmap_free_item_zone != NULL);
4118 new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
4119 new->xbfi_startblock = bno;
4120 new->xbfi_blockcount = (xfs_extlen_t)len;
4121 for (prev = NULL, cur = flist->xbf_first;
4122 cur != NULL;
4123 prev = cur, cur = cur->xbfi_next) {
4124 if (cur->xbfi_startblock >= bno)
4125 break;
4126 }
4127 if (prev)
4128 prev->xbfi_next = new;
4129 else
4130 flist->xbf_first = new;
4131 new->xbfi_next = cur;
4132 flist->xbf_count++;
4133 }
4134
4135 /*
4136 * Compute and fill in the value of the maximum depth of a bmap btree
4137 * in this filesystem. Done once, during mount.
4138 */
4139 void
4140 xfs_bmap_compute_maxlevels(
4141 xfs_mount_t *mp, /* file system mount structure */
4142 int whichfork) /* data or attr fork */
4143 {
4144 int level; /* btree level */
4145 uint maxblocks; /* max blocks at this level */
4146 uint maxleafents; /* max leaf entries possible */
4147 int maxrootrecs; /* max records in root block */
4148 int minleafrecs; /* min records in leaf block */
4149 int minnoderecs; /* min records in node block */
4150 int sz; /* root block size */
4151
4152 /*
4153 * The maximum number of extents in a file, hence the maximum
4154 * number of leaf entries, is controlled by the type of di_nextents
4155 * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
4156 * (a signed 16-bit number, xfs_aextnum_t).
4157 */
4158 if (whichfork == XFS_DATA_FORK) {
4159 maxleafents = MAXEXTNUM;
4160 sz = (mp->m_flags & XFS_MOUNT_ATTR2) ?
4161 XFS_BMDR_SPACE_CALC(MINDBTPTRS) : mp->m_attroffset;
4162 } else {
4163 maxleafents = MAXAEXTNUM;
4164 sz = (mp->m_flags & XFS_MOUNT_ATTR2) ?
4165 XFS_BMDR_SPACE_CALC(MINABTPTRS) :
4166 mp->m_sb.sb_inodesize - mp->m_attroffset;
4167 }
4168 maxrootrecs = (int)XFS_BTREE_BLOCK_MAXRECS(sz, xfs_bmdr, 0);
4169 minleafrecs = mp->m_bmap_dmnr[0];
4170 minnoderecs = mp->m_bmap_dmnr[1];
4171 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
4172 for (level = 1; maxblocks > 1; level++) {
4173 if (maxblocks <= maxrootrecs)
4174 maxblocks = 1;
4175 else
4176 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
4177 }
4178 mp->m_bm_maxlevels[whichfork] = level;
4179 }
4180
4181 /*
4182 * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
4183 * caller. Frees all the extents that need freeing, which must be done
4184 * last due to locking considerations. We never free any extents in
4185 * the first transaction. This is to allow the caller to make the first
4186 * transaction a synchronous one so that the pointers to the data being
4187 * broken in this transaction will be permanent before the data is actually
4188 * freed. This is necessary to prevent blocks from being reallocated
4189 * and written to before the free and reallocation are actually permanent.
4190 * We do not just make the first transaction synchronous here, because
4191 * there are more efficient ways to gain the same protection in some cases
4192 * (see the file truncation code).
4193 *
4194 * Return 1 if the given transaction was committed and a new one
4195 * started, and 0 otherwise in the committed parameter.
4196 */
4197 /*ARGSUSED*/
4198 int /* error */
4199 xfs_bmap_finish(
4200 xfs_trans_t **tp, /* transaction pointer addr */
4201 xfs_bmap_free_t *flist, /* i/o: list extents to free */
4202 int *committed) /* xact committed or not */
4203 {
4204 xfs_efd_log_item_t *efd; /* extent free data */
4205 xfs_efi_log_item_t *efi; /* extent free intention */
4206 int error; /* error return value */
4207 xfs_bmap_free_item_t *free; /* free extent item */
4208 unsigned int logres; /* new log reservation */
4209 unsigned int logcount; /* new log count */
4210 xfs_mount_t *mp; /* filesystem mount structure */
4211 xfs_bmap_free_item_t *next; /* next item on free list */
4212 xfs_trans_t *ntp; /* new transaction pointer */
4213
4214 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
4215 if (flist->xbf_count == 0) {
4216 *committed = 0;
4217 return 0;
4218 }
4219 ntp = *tp;
4220 efi = xfs_trans_get_efi(ntp, flist->xbf_count);
4221 for (free = flist->xbf_first; free; free = free->xbfi_next)
4222 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
4223 free->xbfi_blockcount);
4224 logres = ntp->t_log_res;
4225 logcount = ntp->t_log_count;
4226 ntp = xfs_trans_dup(*tp);
4227 error = xfs_trans_commit(*tp, 0);
4228 *tp = ntp;
4229 *committed = 1;
4230 /*
4231 * We have a new transaction, so we should return committed=1,
4232 * even though we're returning an error.
4233 */
4234 if (error) {
4235 return error;
4236 }
4237 if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
4238 logcount)))
4239 return error;
4240 efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
4241 for (free = flist->xbf_first; free != NULL; free = next) {
4242 next = free->xbfi_next;
4243 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
4244 free->xbfi_blockcount))) {
4245 /*
4246 * The bmap free list will be cleaned up at a
4247 * higher level. The EFI will be canceled when
4248 * this transaction is aborted.
4249 * Need to force shutdown here to make sure it
4250 * happens, since this transaction may not be
4251 * dirty yet.
4252 */
4253 mp = ntp->t_mountp;
4254 if (!XFS_FORCED_SHUTDOWN(mp))
4255 xfs_force_shutdown(mp,
4256 (error == EFSCORRUPTED) ?
4257 SHUTDOWN_CORRUPT_INCORE :
4258 SHUTDOWN_META_IO_ERROR);
4259 return error;
4260 }
4261 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
4262 free->xbfi_blockcount);
4263 xfs_bmap_del_free(flist, NULL, free);
4264 }
4265 return 0;
4266 }
4267
4268 /*
4269 * Free up any items left in the list.
4270 */
4271 void
4272 xfs_bmap_cancel(
4273 xfs_bmap_free_t *flist) /* list of bmap_free_items */
4274 {
4275 xfs_bmap_free_item_t *free; /* free list item */
4276 xfs_bmap_free_item_t *next;
4277
4278 if (flist->xbf_count == 0)
4279 return;
4280 ASSERT(flist->xbf_first != NULL);
4281 for (free = flist->xbf_first; free; free = next) {
4282 next = free->xbfi_next;
4283 xfs_bmap_del_free(flist, NULL, free);
4284 }
4285 ASSERT(flist->xbf_count == 0);
4286 }
4287
4288 /*
4289 * Returns the file-relative block number of the first unused block(s)
4290 * in the file with at least "len" logically contiguous blocks free.
4291 * This is the lowest-address hole if the file has holes, else the first block
4292 * past the end of file.
4293 * Return 0 if the file is currently local (in-inode).
4294 */
4295 int /* error */
4296 xfs_bmap_first_unused(
4297 xfs_trans_t *tp, /* transaction pointer */
4298 xfs_inode_t *ip, /* incore inode */
4299 xfs_extlen_t len, /* size of hole to find */
4300 xfs_fileoff_t *first_unused, /* unused block */
4301 int whichfork) /* data or attr fork */
4302 {
4303 xfs_bmbt_rec_t *ep; /* pointer to an extent entry */
4304 int error; /* error return value */
4305 int idx; /* extent record index */
4306 xfs_ifork_t *ifp; /* inode fork pointer */
4307 xfs_fileoff_t lastaddr; /* last block number seen */
4308 xfs_fileoff_t lowest; /* lowest useful block */
4309 xfs_fileoff_t max; /* starting useful block */
4310 xfs_fileoff_t off; /* offset for this block */
4311 xfs_extnum_t nextents; /* number of extent entries */
4312
4313 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
4314 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
4315 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
4316 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4317 *first_unused = 0;
4318 return 0;
4319 }
4320 ifp = XFS_IFORK_PTR(ip, whichfork);
4321 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4322 (error = xfs_iread_extents(tp, ip, whichfork)))
4323 return error;
4324 lowest = *first_unused;
4325 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4326 for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
4327 ep = xfs_iext_get_ext(ifp, idx);
4328 off = xfs_bmbt_get_startoff(ep);
4329 /*
4330 * See if the hole before this extent will work.
4331 */
4332 if (off >= lowest + len && off - max >= len) {
4333 *first_unused = max;
4334 return 0;
4335 }
4336 lastaddr = off + xfs_bmbt_get_blockcount(ep);
4337 max = XFS_FILEOFF_MAX(lastaddr, lowest);
4338 }
4339 *first_unused = max;
4340 return 0;
4341 }
4342
4343 /*
4344 * Returns the file-relative block number of the last block + 1 before
4345 * last_block (input value) in the file.
4346 * This is not based on i_size, it is based on the extent records.
4347 * Returns 0 for local files, as they do not have extent records.
4348 */
4349 int /* error */
4350 xfs_bmap_last_before(
4351 xfs_trans_t *tp, /* transaction pointer */
4352 xfs_inode_t *ip, /* incore inode */
4353 xfs_fileoff_t *last_block, /* last block */
4354 int whichfork) /* data or attr fork */
4355 {
4356 xfs_fileoff_t bno; /* input file offset */
4357 int eof; /* hit end of file */
4358 xfs_bmbt_rec_t *ep; /* pointer to last extent */
4359 int error; /* error return value */
4360 xfs_bmbt_irec_t got; /* current extent value */
4361 xfs_ifork_t *ifp; /* inode fork pointer */
4362 xfs_extnum_t lastx; /* last extent used */
4363 xfs_bmbt_irec_t prev; /* previous extent value */
4364
4365 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4366 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4367 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4368 return XFS_ERROR(EIO);
4369 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4370 *last_block = 0;
4371 return 0;
4372 }
4373 ifp = XFS_IFORK_PTR(ip, whichfork);
4374 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4375 (error = xfs_iread_extents(tp, ip, whichfork)))
4376 return error;
4377 bno = *last_block - 1;
4378 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4379 &prev);
4380 if (eof || xfs_bmbt_get_startoff(ep) > bno) {
4381 if (prev.br_startoff == NULLFILEOFF)
4382 *last_block = 0;
4383 else
4384 *last_block = prev.br_startoff + prev.br_blockcount;
4385 }
4386 /*
4387 * Otherwise *last_block is already the right answer.
4388 */
4389 return 0;
4390 }
4391
4392 /*
4393 * Returns the file-relative block number of the first block past eof in
4394 * the file. This is not based on i_size, it is based on the extent records.
4395 * Returns 0 for local files, as they do not have extent records.
4396 */
4397 int /* error */
4398 xfs_bmap_last_offset(
4399 xfs_trans_t *tp, /* transaction pointer */
4400 xfs_inode_t *ip, /* incore inode */
4401 xfs_fileoff_t *last_block, /* last block */
4402 int whichfork) /* data or attr fork */
4403 {
4404 xfs_bmbt_rec_t *ep; /* pointer to last extent */
4405 int error; /* error return value */
4406 xfs_ifork_t *ifp; /* inode fork pointer */
4407 xfs_extnum_t nextents; /* number of extent entries */
4408
4409 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4410 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4411 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4412 return XFS_ERROR(EIO);
4413 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4414 *last_block = 0;
4415 return 0;
4416 }
4417 ifp = XFS_IFORK_PTR(ip, whichfork);
4418 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4419 (error = xfs_iread_extents(tp, ip, whichfork)))
4420 return error;
4421 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4422 if (!nextents) {
4423 *last_block = 0;
4424 return 0;
4425 }
4426 ep = xfs_iext_get_ext(ifp, nextents - 1);
4427 *last_block = xfs_bmbt_get_startoff(ep) + xfs_bmbt_get_blockcount(ep);
4428 return 0;
4429 }
4430
4431 /*
4432 * Returns whether the selected fork of the inode has exactly one
4433 * block or not. For the data fork we check this matches di_size,
4434 * implying the file's range is 0..bsize-1.
4435 */
4436 int /* 1=>1 block, 0=>otherwise */
4437 xfs_bmap_one_block(
4438 xfs_inode_t *ip, /* incore inode */
4439 int whichfork) /* data or attr fork */
4440 {
4441 xfs_bmbt_rec_t *ep; /* ptr to fork's extent */
4442 xfs_ifork_t *ifp; /* inode fork pointer */
4443 int rval; /* return value */
4444 xfs_bmbt_irec_t s; /* internal version of extent */
4445
4446 #ifndef DEBUG
4447 if (whichfork == XFS_DATA_FORK) {
4448 return ((ip->i_d.di_mode & S_IFMT) == S_IFREG) ?
4449 (ip->i_size == ip->i_mount->m_sb.sb_blocksize) :
4450 (ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize);
4451 }
4452 #endif /* !DEBUG */
4453 if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4454 return 0;
4455 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4456 return 0;
4457 ifp = XFS_IFORK_PTR(ip, whichfork);
4458 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4459 ep = xfs_iext_get_ext(ifp, 0);
4460 xfs_bmbt_get_all(ep, &s);
4461 rval = s.br_startoff == 0 && s.br_blockcount == 1;
4462 if (rval && whichfork == XFS_DATA_FORK)
4463 ASSERT(ip->i_size == ip->i_mount->m_sb.sb_blocksize);
4464 return rval;
4465 }
4466
4467 /*
4468 * Read in the extents to if_extents.
4469 * All inode fields are set up by caller, we just traverse the btree
4470 * and copy the records in. If the file system cannot contain unwritten
4471 * extents, the records are checked for no "state" flags.
4472 */
4473 int /* error */
4474 xfs_bmap_read_extents(
4475 xfs_trans_t *tp, /* transaction pointer */
4476 xfs_inode_t *ip, /* incore inode */
4477 int whichfork) /* data or attr fork */
4478 {
4479 xfs_bmbt_block_t *block; /* current btree block */
4480 xfs_fsblock_t bno; /* block # of "block" */
4481 xfs_buf_t *bp; /* buffer for "block" */
4482 int error; /* error return value */
4483 xfs_exntfmt_t exntf; /* XFS_EXTFMT_NOSTATE, if checking */
4484 #ifdef XFS_BMAP_TRACE
4485 static char fname[] = "xfs_bmap_read_extents";
4486 #endif
4487 xfs_extnum_t i, j; /* index into the extents list */
4488 xfs_ifork_t *ifp; /* fork structure */
4489 int level; /* btree level, for checking */
4490 xfs_mount_t *mp; /* file system mount structure */
4491 __be64 *pp; /* pointer to block address */
4492 /* REFERENCED */
4493 xfs_extnum_t room; /* number of entries there's room for */
4494
4495 bno = NULLFSBLOCK;
4496 mp = ip->i_mount;
4497 ifp = XFS_IFORK_PTR(ip, whichfork);
4498 exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4499 XFS_EXTFMT_INODE(ip);
4500 block = ifp->if_broot;
4501 /*
4502 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4503 */
4504 level = be16_to_cpu(block->bb_level);
4505 ASSERT(level > 0);
4506 pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
4507 bno = be64_to_cpu(*pp);
4508 ASSERT(bno != NULLDFSBNO);
4509 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
4510 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
4511 /*
4512 * Go down the tree until leaf level is reached, following the first
4513 * pointer (leftmost) at each level.
4514 */
4515 while (level-- > 0) {
4516 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4517 XFS_BMAP_BTREE_REF)))
4518 return error;
4519 block = XFS_BUF_TO_BMBT_BLOCK(bp);
4520 XFS_WANT_CORRUPTED_GOTO(
4521 XFS_BMAP_SANITY_CHECK(mp, block, level),
4522 error0);
4523 if (level == 0)
4524 break;
4525 pp = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, 1, mp->m_bmap_dmxr[1]);
4526 bno = be64_to_cpu(*pp);
4527 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
4528 xfs_trans_brelse(tp, bp);
4529 }
4530 /*
4531 * Here with bp and block set to the leftmost leaf node in the tree.
4532 */
4533 room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4534 i = 0;
4535 /*
4536 * Loop over all leaf nodes. Copy information to the extent records.
4537 */
4538 for (;;) {
4539 xfs_bmbt_rec_t *frp, *trp;
4540 xfs_fsblock_t nextbno;
4541 xfs_extnum_t num_recs;
4542 xfs_extnum_t start;
4543
4544
4545 num_recs = be16_to_cpu(block->bb_numrecs);
4546 if (unlikely(i + num_recs > room)) {
4547 ASSERT(i + num_recs <= room);
4548 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
4549 "corrupt dinode %Lu, (btree extents).",
4550 (unsigned long long) ip->i_ino);
4551 XFS_ERROR_REPORT("xfs_bmap_read_extents(1)",
4552 XFS_ERRLEVEL_LOW,
4553 ip->i_mount);
4554 goto error0;
4555 }
4556 XFS_WANT_CORRUPTED_GOTO(
4557 XFS_BMAP_SANITY_CHECK(mp, block, 0),
4558 error0);
4559 /*
4560 * Read-ahead the next leaf block, if any.
4561 */
4562 nextbno = be64_to_cpu(block->bb_rightsib);
4563 if (nextbno != NULLFSBLOCK)
4564 xfs_btree_reada_bufl(mp, nextbno, 1);
4565 /*
4566 * Copy records into the extent records.
4567 */
4568 frp = XFS_BTREE_REC_ADDR(xfs_bmbt, block, 1);
4569 start = i;
4570 for (j = 0; j < num_recs; j++, i++, frp++) {
4571 trp = xfs_iext_get_ext(ifp, i);
4572 trp->l0 = INT_GET(frp->l0, ARCH_CONVERT);
4573 trp->l1 = INT_GET(frp->l1, ARCH_CONVERT);
4574 }
4575 if (exntf == XFS_EXTFMT_NOSTATE) {
4576 /*
4577 * Check all attribute bmap btree records and
4578 * any "older" data bmap btree records for a
4579 * set bit in the "extent flag" position.
4580 */
4581 if (unlikely(xfs_check_nostate_extents(ifp,
4582 start, num_recs))) {
4583 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4584 XFS_ERRLEVEL_LOW,
4585 ip->i_mount);
4586 goto error0;
4587 }
4588 }
4589 xfs_trans_brelse(tp, bp);
4590 bno = nextbno;
4591 /*
4592 * If we've reached the end, stop.
4593 */
4594 if (bno == NULLFSBLOCK)
4595 break;
4596 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4597 XFS_BMAP_BTREE_REF)))
4598 return error;
4599 block = XFS_BUF_TO_BMBT_BLOCK(bp);
4600 }
4601 ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4602 ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4603 xfs_bmap_trace_exlist(fname, ip, i, whichfork);
4604 return 0;
4605 error0:
4606 xfs_trans_brelse(tp, bp);
4607 return XFS_ERROR(EFSCORRUPTED);
4608 }
4609
4610 #ifdef XFS_BMAP_TRACE
4611 /*
4612 * Add bmap trace insert entries for all the contents of the extent records.
4613 */
4614 void
4615 xfs_bmap_trace_exlist(
4616 char *fname, /* function name */
4617 xfs_inode_t *ip, /* incore inode pointer */
4618 xfs_extnum_t cnt, /* count of entries in the list */
4619 int whichfork) /* data or attr fork */
4620 {
4621 xfs_bmbt_rec_t *ep; /* current extent record */
4622 xfs_extnum_t idx; /* extent record index */
4623 xfs_ifork_t *ifp; /* inode fork pointer */
4624 xfs_bmbt_irec_t s; /* file extent record */
4625
4626 ifp = XFS_IFORK_PTR(ip, whichfork);
4627 ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4628 for (idx = 0; idx < cnt; idx++) {
4629 ep = xfs_iext_get_ext(ifp, idx);
4630 xfs_bmbt_get_all(ep, &s);
4631 xfs_bmap_trace_insert(fname, "exlist", ip, idx, 1, &s, NULL,
4632 whichfork);
4633 }
4634 }
4635 #endif
4636
4637 #ifdef DEBUG
4638 /*
4639 * Validate that the bmbt_irecs being returned from bmapi are valid
4640 * given the callers original parameters. Specifically check the
4641 * ranges of the returned irecs to ensure that they only extent beyond
4642 * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4643 */
4644 STATIC void
4645 xfs_bmap_validate_ret(
4646 xfs_fileoff_t bno,
4647 xfs_filblks_t len,
4648 int flags,
4649 xfs_bmbt_irec_t *mval,
4650 int nmap,
4651 int ret_nmap)
4652 {
4653 int i; /* index to map values */
4654
4655 ASSERT(ret_nmap <= nmap);
4656
4657 for (i = 0; i < ret_nmap; i++) {
4658 ASSERT(mval[i].br_blockcount > 0);
4659 if (!(flags & XFS_BMAPI_ENTIRE)) {
4660 ASSERT(mval[i].br_startoff >= bno);
4661 ASSERT(mval[i].br_blockcount <= len);
4662 ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4663 bno + len);
4664 } else {
4665 ASSERT(mval[i].br_startoff < bno + len);
4666 ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4667 bno);
4668 }
4669 ASSERT(i == 0 ||
4670 mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4671 mval[i].br_startoff);
4672 if ((flags & XFS_BMAPI_WRITE) && !(flags & XFS_BMAPI_DELAY))
4673 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4674 mval[i].br_startblock != HOLESTARTBLOCK);
4675 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4676 mval[i].br_state == XFS_EXT_UNWRITTEN);
4677 }
4678 }
4679 #endif /* DEBUG */
4680
4681
4682 /*
4683 * Map file blocks to filesystem blocks.
4684 * File range is given by the bno/len pair.
4685 * Adds blocks to file if a write ("flags & XFS_BMAPI_WRITE" set)
4686 * into a hole or past eof.
4687 * Only allocates blocks from a single allocation group,
4688 * to avoid locking problems.
4689 * The returned value in "firstblock" from the first call in a transaction
4690 * must be remembered and presented to subsequent calls in "firstblock".
4691 * An upper bound for the number of blocks to be allocated is supplied to
4692 * the first call in "total"; if no allocation group has that many free
4693 * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4694 */
4695 int /* error */
4696 xfs_bmapi(
4697 xfs_trans_t *tp, /* transaction pointer */
4698 xfs_inode_t *ip, /* incore inode */
4699 xfs_fileoff_t bno, /* starting file offs. mapped */
4700 xfs_filblks_t len, /* length to map in file */
4701 int flags, /* XFS_BMAPI_... */
4702 xfs_fsblock_t *firstblock, /* first allocated block
4703 controls a.g. for allocs */
4704 xfs_extlen_t total, /* total blocks needed */
4705 xfs_bmbt_irec_t *mval, /* output: map values */
4706 int *nmap, /* i/o: mval size/count */
4707 xfs_bmap_free_t *flist, /* i/o: list extents to free */
4708 xfs_extdelta_t *delta) /* o: change made to incore extents */
4709 {
4710 xfs_fsblock_t abno; /* allocated block number */
4711 xfs_extlen_t alen; /* allocated extent length */
4712 xfs_fileoff_t aoff; /* allocated file offset */
4713 xfs_bmalloca_t bma; /* args for xfs_bmap_alloc */
4714 xfs_btree_cur_t *cur; /* bmap btree cursor */
4715 xfs_fileoff_t end; /* end of mapped file region */
4716 int eof; /* we've hit the end of extents */
4717 xfs_bmbt_rec_t *ep; /* extent record pointer */
4718 int error; /* error return */
4719 xfs_bmbt_irec_t got; /* current file extent record */
4720 xfs_ifork_t *ifp; /* inode fork pointer */
4721 xfs_extlen_t indlen; /* indirect blocks length */
4722 xfs_extnum_t lastx; /* last useful extent number */
4723 int logflags; /* flags for transaction logging */
4724 xfs_extlen_t minleft; /* min blocks left after allocation */
4725 xfs_extlen_t minlen; /* min allocation size */
4726 xfs_mount_t *mp; /* xfs mount structure */
4727 int n; /* current extent index */
4728 int nallocs; /* number of extents alloc\'d */
4729 xfs_extnum_t nextents; /* number of extents in file */
4730 xfs_fileoff_t obno; /* old block number (offset) */
4731 xfs_bmbt_irec_t prev; /* previous file extent record */
4732 int tmp_logflags; /* temp flags holder */
4733 int whichfork; /* data or attr fork */
4734 char inhole; /* current location is hole in file */
4735 char wasdelay; /* old extent was delayed */
4736 char wr; /* this is a write request */
4737 char rt; /* this is a realtime file */
4738 #ifdef DEBUG
4739 xfs_fileoff_t orig_bno; /* original block number value */
4740 int orig_flags; /* original flags arg value */
4741 xfs_filblks_t orig_len; /* original value of len arg */
4742 xfs_bmbt_irec_t *orig_mval; /* original value of mval */
4743 int orig_nmap; /* original value of *nmap */
4744
4745 orig_bno = bno;
4746 orig_len = len;
4747 orig_flags = flags;
4748 orig_mval = mval;
4749 orig_nmap = *nmap;
4750 #endif
4751 ASSERT(*nmap >= 1);
4752 ASSERT(*nmap <= XFS_BMAP_MAX_NMAP || !(flags & XFS_BMAPI_WRITE));
4753 whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4754 XFS_ATTR_FORK : XFS_DATA_FORK;
4755 mp = ip->i_mount;
4756 if (unlikely(XFS_TEST_ERROR(
4757 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4758 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4759 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4760 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4761 XFS_ERROR_REPORT("xfs_bmapi", XFS_ERRLEVEL_LOW, mp);
4762 return XFS_ERROR(EFSCORRUPTED);
4763 }
4764 if (XFS_FORCED_SHUTDOWN(mp))
4765 return XFS_ERROR(EIO);
4766 rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4767 ifp = XFS_IFORK_PTR(ip, whichfork);
4768 ASSERT(ifp->if_ext_max ==
4769 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4770 if ((wr = (flags & XFS_BMAPI_WRITE)) != 0)
4771 XFS_STATS_INC(xs_blk_mapw);
4772 else
4773 XFS_STATS_INC(xs_blk_mapr);
4774 /*
4775 * IGSTATE flag is used to combine extents which
4776 * differ only due to the state of the extents.
4777 * This technique is used from xfs_getbmap()
4778 * when the caller does not wish to see the
4779 * separation (which is the default).
4780 *
4781 * This technique is also used when writing a
4782 * buffer which has been partially written,
4783 * (usually by being flushed during a chunkread),
4784 * to ensure one write takes place. This also
4785 * prevents a change in the xfs inode extents at
4786 * this time, intentionally. This change occurs
4787 * on completion of the write operation, in
4788 * xfs_strat_comp(), where the xfs_bmapi() call
4789 * is transactioned, and the extents combined.
4790 */
4791 if ((flags & XFS_BMAPI_IGSTATE) && wr) /* if writing unwritten space */
4792 wr = 0; /* no allocations are allowed */
4793 ASSERT(wr || !(flags & XFS_BMAPI_DELAY));
4794 logflags = 0;
4795 nallocs = 0;
4796 cur = NULL;
4797 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4798 ASSERT(wr && tp);
4799 if ((error = xfs_bmap_local_to_extents(tp, ip,
4800 firstblock, total, &logflags, whichfork)))
4801 goto error0;
4802 }
4803 if (wr && *firstblock == NULLFSBLOCK) {
4804 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4805 minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4806 else
4807 minleft = 1;
4808 } else
4809 minleft = 0;
4810 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4811 (error = xfs_iread_extents(tp, ip, whichfork)))
4812 goto error0;
4813 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4814 &prev);
4815 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4816 n = 0;
4817 end = bno + len;
4818 obno = bno;
4819 bma.ip = NULL;
4820 if (delta) {
4821 delta->xed_startoff = NULLFILEOFF;
4822 delta->xed_blockcount = 0;
4823 }
4824 while (bno < end && n < *nmap) {
4825 /*
4826 * Reading past eof, act as though there's a hole
4827 * up to end.
4828 */
4829 if (eof && !wr)
4830 got.br_startoff = end;
4831 inhole = eof || got.br_startoff > bno;
4832 wasdelay = wr && !inhole && !(flags & XFS_BMAPI_DELAY) &&
4833 ISNULLSTARTBLOCK(got.br_startblock);
4834 /*
4835 * First, deal with the hole before the allocated space
4836 * that we found, if any.
4837 */
4838 if (wr && (inhole || wasdelay)) {
4839 /*
4840 * For the wasdelay case, we could also just
4841 * allocate the stuff asked for in this bmap call
4842 * but that wouldn't be as good.
4843 */
4844 if (wasdelay && !(flags & XFS_BMAPI_EXACT)) {
4845 alen = (xfs_extlen_t)got.br_blockcount;
4846 aoff = got.br_startoff;
4847 if (lastx != NULLEXTNUM && lastx) {
4848 ep = xfs_iext_get_ext(ifp, lastx - 1);
4849 xfs_bmbt_get_all(ep, &prev);
4850 }
4851 } else if (wasdelay) {
4852 alen = (xfs_extlen_t)
4853 XFS_FILBLKS_MIN(len,
4854 (got.br_startoff +
4855 got.br_blockcount) - bno);
4856 aoff = bno;
4857 } else {
4858 alen = (xfs_extlen_t)
4859 XFS_FILBLKS_MIN(len, MAXEXTLEN);
4860 if (!eof)
4861 alen = (xfs_extlen_t)
4862 XFS_FILBLKS_MIN(alen,
4863 got.br_startoff - bno);
4864 aoff = bno;
4865 }
4866 minlen = (flags & XFS_BMAPI_CONTIG) ? alen : 1;
4867 if (flags & XFS_BMAPI_DELAY) {
4868 xfs_extlen_t extsz;
4869
4870 /* Figure out the extent size, adjust alen */
4871 if (rt) {
4872 if (!(extsz = ip->i_d.di_extsize))
4873 extsz = mp->m_sb.sb_rextsize;
4874 } else {
4875 extsz = ip->i_d.di_extsize;
4876 }
4877 if (extsz) {
4878 error = xfs_bmap_extsize_align(mp,
4879 &got, &prev, extsz,
4880 rt, eof,
4881 flags&XFS_BMAPI_DELAY,
4882 flags&XFS_BMAPI_CONVERT,
4883 &aoff, &alen);
4884 ASSERT(!error);
4885 }
4886
4887 if (rt)
4888 extsz = alen / mp->m_sb.sb_rextsize;
4889
4890 /*
4891 * Make a transaction-less quota reservation for
4892 * delayed allocation blocks. This number gets
4893 * adjusted later. We return if we haven't
4894 * allocated blocks already inside this loop.
4895 */
4896 if ((error = XFS_TRANS_RESERVE_QUOTA_NBLKS(
4897 mp, NULL, ip, (long)alen, 0,
4898 rt ? XFS_QMOPT_RES_RTBLKS :
4899 XFS_QMOPT_RES_REGBLKS))) {
4900 if (n == 0) {
4901 *nmap = 0;
4902 ASSERT(cur == NULL);
4903 return error;
4904 }
4905 break;
4906 }
4907
4908 /*
4909 * Split changing sb for alen and indlen since
4910 * they could be coming from different places.
4911 */
4912 indlen = (xfs_extlen_t)
4913 xfs_bmap_worst_indlen(ip, alen);
4914 ASSERT(indlen > 0);
4915
4916 if (rt) {
4917 error = xfs_mod_incore_sb(mp,
4918 XFS_SBS_FREXTENTS,
4919 -((int64_t)extsz), (flags &
4920 XFS_BMAPI_RSVBLOCKS));
4921 } else {
4922 error = xfs_mod_incore_sb(mp,
4923 XFS_SBS_FDBLOCKS,
4924 -((int64_t)alen), (flags &
4925 XFS_BMAPI_RSVBLOCKS));
4926 }
4927 if (!error) {
4928 error = xfs_mod_incore_sb(mp,
4929 XFS_SBS_FDBLOCKS,
4930 -((int64_t)indlen), (flags &
4931 XFS_BMAPI_RSVBLOCKS));
4932 if (error && rt)
4933 xfs_mod_incore_sb(mp,
4934 XFS_SBS_FREXTENTS,
4935 (int64_t)extsz, (flags &
4936 XFS_BMAPI_RSVBLOCKS));
4937 else if (error)
4938 xfs_mod_incore_sb(mp,
4939 XFS_SBS_FDBLOCKS,
4940 (int64_t)alen, (flags &
4941 XFS_BMAPI_RSVBLOCKS));
4942 }
4943
4944 if (error) {
4945 if (XFS_IS_QUOTA_ON(mp))
4946 /* unreserve the blocks now */
4947 (void)
4948 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(
4949 mp, NULL, ip,
4950 (long)alen, 0, rt ?
4951 XFS_QMOPT_RES_RTBLKS :
4952 XFS_QMOPT_RES_REGBLKS);
4953 break;
4954 }
4955
4956 ip->i_delayed_blks += alen;
4957 abno = NULLSTARTBLOCK(indlen);
4958 } else {
4959 /*
4960 * If first time, allocate and fill in
4961 * once-only bma fields.
4962 */
4963 if (bma.ip == NULL) {
4964 bma.tp = tp;
4965 bma.ip = ip;
4966 bma.prevp = &prev;
4967 bma.gotp = &got;
4968 bma.total = total;
4969 bma.userdata = 0;
4970 }
4971 /* Indicate if this is the first user data
4972 * in the file, or just any user data.
4973 */
4974 if (!(flags & XFS_BMAPI_METADATA)) {
4975 bma.userdata = (aoff == 0) ?
4976 XFS_ALLOC_INITIAL_USER_DATA :
4977 XFS_ALLOC_USERDATA;
4978 }
4979 /*
4980 * Fill in changeable bma fields.
4981 */
4982 bma.eof = eof;
4983 bma.firstblock = *firstblock;
4984 bma.alen = alen;
4985 bma.off = aoff;
4986 bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4987 bma.wasdel = wasdelay;
4988 bma.minlen = minlen;
4989 bma.low = flist->xbf_low;
4990 bma.minleft = minleft;
4991 /*
4992 * Only want to do the alignment at the
4993 * eof if it is userdata and allocation length
4994 * is larger than a stripe unit.
4995 */
4996 if (mp->m_dalign && alen >= mp->m_dalign &&
4997 (!(flags & XFS_BMAPI_METADATA)) &&
4998 (whichfork == XFS_DATA_FORK)) {
4999 if ((error = xfs_bmap_isaeof(ip, aoff,
5000 whichfork, &bma.aeof)))
5001 goto error0;
5002 } else
5003 bma.aeof = 0;
5004 /*
5005 * Call allocator.
5006 */
5007 if ((error = xfs_bmap_alloc(&bma)))
5008 goto error0;
5009 /*
5010 * Copy out result fields.
5011 */
5012 abno = bma.rval;
5013 if ((flist->xbf_low = bma.low))
5014 minleft = 0;
5015 alen = bma.alen;
5016 aoff = bma.off;
5017 ASSERT(*firstblock == NULLFSBLOCK ||
5018 XFS_FSB_TO_AGNO(mp, *firstblock) ==
5019 XFS_FSB_TO_AGNO(mp, bma.firstblock) ||
5020 (flist->xbf_low &&
5021 XFS_FSB_TO_AGNO(mp, *firstblock) <
5022 XFS_FSB_TO_AGNO(mp, bma.firstblock)));
5023 *firstblock = bma.firstblock;
5024 if (cur)
5025 cur->bc_private.b.firstblock =
5026 *firstblock;
5027 if (abno == NULLFSBLOCK)
5028 break;
5029 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
5030 cur = xfs_btree_init_cursor(mp,
5031 tp, NULL, 0, XFS_BTNUM_BMAP,
5032 ip, whichfork);
5033 cur->bc_private.b.firstblock =
5034 *firstblock;
5035 cur->bc_private.b.flist = flist;
5036 }
5037 /*
5038 * Bump the number of extents we've allocated
5039 * in this call.
5040 */
5041 nallocs++;
5042 }
5043 if (cur)
5044 cur->bc_private.b.flags =
5045 wasdelay ? XFS_BTCUR_BPRV_WASDEL : 0;
5046 got.br_startoff = aoff;
5047 got.br_startblock = abno;
5048 got.br_blockcount = alen;
5049 got.br_state = XFS_EXT_NORM; /* assume normal */
5050 /*
5051 * Determine state of extent, and the filesystem.
5052 * A wasdelay extent has been initialized, so
5053 * shouldn't be flagged as unwritten.
5054 */
5055 if (wr && XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
5056 if (!wasdelay && (flags & XFS_BMAPI_PREALLOC))
5057 got.br_state = XFS_EXT_UNWRITTEN;
5058 }
5059 error = xfs_bmap_add_extent(ip, lastx, &cur, &got,
5060 firstblock, flist, &tmp_logflags, delta,
5061 whichfork, (flags & XFS_BMAPI_RSVBLOCKS));
5062 logflags |= tmp_logflags;
5063 if (error)
5064 goto error0;
5065 lastx = ifp->if_lastex;
5066 ep = xfs_iext_get_ext(ifp, lastx);
5067 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5068 xfs_bmbt_get_all(ep, &got);
5069 ASSERT(got.br_startoff <= aoff);
5070 ASSERT(got.br_startoff + got.br_blockcount >=
5071 aoff + alen);
5072 #ifdef DEBUG
5073 if (flags & XFS_BMAPI_DELAY) {
5074 ASSERT(ISNULLSTARTBLOCK(got.br_startblock));
5075 ASSERT(STARTBLOCKVAL(got.br_startblock) > 0);
5076 }
5077 ASSERT(got.br_state == XFS_EXT_NORM ||
5078 got.br_state == XFS_EXT_UNWRITTEN);
5079 #endif
5080 /*
5081 * Fall down into the found allocated space case.
5082 */
5083 } else if (inhole) {
5084 /*
5085 * Reading in a hole.
5086 */
5087 mval->br_startoff = bno;
5088 mval->br_startblock = HOLESTARTBLOCK;
5089 mval->br_blockcount =
5090 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
5091 mval->br_state = XFS_EXT_NORM;
5092 bno += mval->br_blockcount;
5093 len -= mval->br_blockcount;
5094 mval++;
5095 n++;
5096 continue;
5097 }
5098 /*
5099 * Then deal with the allocated space we found.
5100 */
5101 ASSERT(ep != NULL);
5102 if (!(flags & XFS_BMAPI_ENTIRE) &&
5103 (got.br_startoff + got.br_blockcount > obno)) {
5104 if (obno > bno)
5105 bno = obno;
5106 ASSERT((bno >= obno) || (n == 0));
5107 ASSERT(bno < end);
5108 mval->br_startoff = bno;
5109 if (ISNULLSTARTBLOCK(got.br_startblock)) {
5110 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
5111 mval->br_startblock = DELAYSTARTBLOCK;
5112 } else
5113 mval->br_startblock =
5114 got.br_startblock +
5115 (bno - got.br_startoff);
5116 /*
5117 * Return the minimum of what we got and what we
5118 * asked for for the length. We can use the len
5119 * variable here because it is modified below
5120 * and we could have been there before coming
5121 * here if the first part of the allocation
5122 * didn't overlap what was asked for.
5123 */
5124 mval->br_blockcount =
5125 XFS_FILBLKS_MIN(end - bno, got.br_blockcount -
5126 (bno - got.br_startoff));
5127 mval->br_state = got.br_state;
5128 ASSERT(mval->br_blockcount <= len);
5129 } else {
5130 *mval = got;
5131 if (ISNULLSTARTBLOCK(mval->br_startblock)) {
5132 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
5133 mval->br_startblock = DELAYSTARTBLOCK;
5134 }
5135 }
5136
5137 /*
5138 * Check if writing previously allocated but
5139 * unwritten extents.
5140 */
5141 if (wr && mval->br_state == XFS_EXT_UNWRITTEN &&
5142 ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_DELAY)) == 0)) {
5143 /*
5144 * Modify (by adding) the state flag, if writing.
5145 */
5146 ASSERT(mval->br_blockcount <= len);
5147 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
5148 cur = xfs_btree_init_cursor(mp,
5149 tp, NULL, 0, XFS_BTNUM_BMAP,
5150 ip, whichfork);
5151 cur->bc_private.b.firstblock =
5152 *firstblock;
5153 cur->bc_private.b.flist = flist;
5154 }
5155 mval->br_state = XFS_EXT_NORM;
5156 error = xfs_bmap_add_extent(ip, lastx, &cur, mval,
5157 firstblock, flist, &tmp_logflags, delta,
5158 whichfork, (flags & XFS_BMAPI_RSVBLOCKS));
5159 logflags |= tmp_logflags;
5160 if (error)
5161 goto error0;
5162 lastx = ifp->if_lastex;
5163 ep = xfs_iext_get_ext(ifp, lastx);
5164 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5165 xfs_bmbt_get_all(ep, &got);
5166 /*
5167 * We may have combined previously unwritten
5168 * space with written space, so generate
5169 * another request.
5170 */
5171 if (mval->br_blockcount < len)
5172 continue;
5173 }
5174
5175 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
5176 ((mval->br_startoff + mval->br_blockcount) <= end));
5177 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
5178 (mval->br_blockcount <= len) ||
5179 (mval->br_startoff < obno));
5180 bno = mval->br_startoff + mval->br_blockcount;
5181 len = end - bno;
5182 if (n > 0 && mval->br_startoff == mval[-1].br_startoff) {
5183 ASSERT(mval->br_startblock == mval[-1].br_startblock);
5184 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
5185 ASSERT(mval->br_state == mval[-1].br_state);
5186 mval[-1].br_blockcount = mval->br_blockcount;
5187 mval[-1].br_state = mval->br_state;
5188 } else if (n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
5189 mval[-1].br_startblock != DELAYSTARTBLOCK &&
5190 mval[-1].br_startblock != HOLESTARTBLOCK &&
5191 mval->br_startblock ==
5192 mval[-1].br_startblock + mval[-1].br_blockcount &&
5193 ((flags & XFS_BMAPI_IGSTATE) ||
5194 mval[-1].br_state == mval->br_state)) {
5195 ASSERT(mval->br_startoff ==
5196 mval[-1].br_startoff + mval[-1].br_blockcount);
5197 mval[-1].br_blockcount += mval->br_blockcount;
5198 } else if (n > 0 &&
5199 mval->br_startblock == DELAYSTARTBLOCK &&
5200 mval[-1].br_startblock == DELAYSTARTBLOCK &&
5201 mval->br_startoff ==
5202 mval[-1].br_startoff + mval[-1].br_blockcount) {
5203 mval[-1].br_blockcount += mval->br_blockcount;
5204 mval[-1].br_state = mval->br_state;
5205 } else if (!((n == 0) &&
5206 ((mval->br_startoff + mval->br_blockcount) <=
5207 obno))) {
5208 mval++;
5209 n++;
5210 }
5211 /*
5212 * If we're done, stop now. Stop when we've allocated
5213 * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise
5214 * the transaction may get too big.
5215 */
5216 if (bno >= end || n >= *nmap || nallocs >= *nmap)
5217 break;
5218 /*
5219 * Else go on to the next record.
5220 */
5221 ep = xfs_iext_get_ext(ifp, ++lastx);
5222 if (lastx >= nextents) {
5223 eof = 1;
5224 prev = got;
5225 } else
5226 xfs_bmbt_get_all(ep, &got);
5227 }
5228 ifp->if_lastex = lastx;
5229 *nmap = n;
5230 /*
5231 * Transform from btree to extents, give it cur.
5232 */
5233 if (tp && XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5234 XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5235 ASSERT(wr && cur);
5236 error = xfs_bmap_btree_to_extents(tp, ip, cur,
5237 &tmp_logflags, whichfork);
5238 logflags |= tmp_logflags;
5239 if (error)
5240 goto error0;
5241 }
5242 ASSERT(ifp->if_ext_max ==
5243 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5244 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
5245 XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max);
5246 error = 0;
5247 if (delta && delta->xed_startoff != NULLFILEOFF) {
5248 /* A change was actually made.
5249 * Note that delta->xed_blockount is an offset at this
5250 * point and needs to be converted to a block count.
5251 */
5252 ASSERT(delta->xed_blockcount > delta->xed_startoff);
5253 delta->xed_blockcount -= delta->xed_startoff;
5254 }
5255 error0:
5256 /*
5257 * Log everything. Do this after conversion, there's no point in
5258 * logging the extent records if we've converted to btree format.
5259 */
5260 if ((logflags & XFS_ILOG_FEXT(whichfork)) &&
5261 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5262 logflags &= ~XFS_ILOG_FEXT(whichfork);
5263 else if ((logflags & XFS_ILOG_FBROOT(whichfork)) &&
5264 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5265 logflags &= ~XFS_ILOG_FBROOT(whichfork);
5266 /*
5267 * Log whatever the flags say, even if error. Otherwise we might miss
5268 * detecting a case where the data is changed, there's an error,
5269 * and it's not logged so we don't shutdown when we should.
5270 */
5271 if (logflags) {
5272 ASSERT(tp && wr);
5273 xfs_trans_log_inode(tp, ip, logflags);
5274 }
5275 if (cur) {
5276 if (!error) {
5277 ASSERT(*firstblock == NULLFSBLOCK ||
5278 XFS_FSB_TO_AGNO(mp, *firstblock) ==
5279 XFS_FSB_TO_AGNO(mp,
5280 cur->bc_private.b.firstblock) ||
5281 (flist->xbf_low &&
5282 XFS_FSB_TO_AGNO(mp, *firstblock) <
5283 XFS_FSB_TO_AGNO(mp,
5284 cur->bc_private.b.firstblock)));
5285 *firstblock = cur->bc_private.b.firstblock;
5286 }
5287 xfs_btree_del_cursor(cur,
5288 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5289 }
5290 if (!error)
5291 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
5292 orig_nmap, *nmap);
5293 return error;
5294 }
5295
5296 /*
5297 * Map file blocks to filesystem blocks, simple version.
5298 * One block (extent) only, read-only.
5299 * For flags, only the XFS_BMAPI_ATTRFORK flag is examined.
5300 * For the other flag values, the effect is as if XFS_BMAPI_METADATA
5301 * was set and all the others were clear.
5302 */
5303 int /* error */
5304 xfs_bmapi_single(
5305 xfs_trans_t *tp, /* transaction pointer */
5306 xfs_inode_t *ip, /* incore inode */
5307 int whichfork, /* data or attr fork */
5308 xfs_fsblock_t *fsb, /* output: mapped block */
5309 xfs_fileoff_t bno) /* starting file offs. mapped */
5310 {
5311 int eof; /* we've hit the end of extents */
5312 int error; /* error return */
5313 xfs_bmbt_irec_t got; /* current file extent record */
5314 xfs_ifork_t *ifp; /* inode fork pointer */
5315 xfs_extnum_t lastx; /* last useful extent number */
5316 xfs_bmbt_irec_t prev; /* previous file extent record */
5317
5318 ifp = XFS_IFORK_PTR(ip, whichfork);
5319 if (unlikely(
5320 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
5321 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)) {
5322 XFS_ERROR_REPORT("xfs_bmapi_single", XFS_ERRLEVEL_LOW,
5323 ip->i_mount);
5324 return XFS_ERROR(EFSCORRUPTED);
5325 }
5326 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
5327 return XFS_ERROR(EIO);
5328 XFS_STATS_INC(xs_blk_mapr);
5329 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5330 (error = xfs_iread_extents(tp, ip, whichfork)))
5331 return error;
5332 (void)xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5333 &prev);
5334 /*
5335 * Reading past eof, act as though there's a hole
5336 * up to end.
5337 */
5338 if (eof || got.br_startoff > bno) {
5339 *fsb = NULLFSBLOCK;
5340 return 0;
5341 }
5342 ASSERT(!ISNULLSTARTBLOCK(got.br_startblock));
5343 ASSERT(bno < got.br_startoff + got.br_blockcount);
5344 *fsb = got.br_startblock + (bno - got.br_startoff);
5345 ifp->if_lastex = lastx;
5346 return 0;
5347 }
5348
5349 /*
5350 * Unmap (remove) blocks from a file.
5351 * If nexts is nonzero then the number of extents to remove is limited to
5352 * that value. If not all extents in the block range can be removed then
5353 * *done is set.
5354 */
5355 int /* error */
5356 xfs_bunmapi(
5357 xfs_trans_t *tp, /* transaction pointer */
5358 struct xfs_inode *ip, /* incore inode */
5359 xfs_fileoff_t bno, /* starting offset to unmap */
5360 xfs_filblks_t len, /* length to unmap in file */
5361 int flags, /* misc flags */
5362 xfs_extnum_t nexts, /* number of extents max */
5363 xfs_fsblock_t *firstblock, /* first allocated block
5364 controls a.g. for allocs */
5365 xfs_bmap_free_t *flist, /* i/o: list extents to free */
5366 xfs_extdelta_t *delta, /* o: change made to incore
5367 extents */
5368 int *done) /* set if not done yet */
5369 {
5370 xfs_btree_cur_t *cur; /* bmap btree cursor */
5371 xfs_bmbt_irec_t del; /* extent being deleted */
5372 int eof; /* is deleting at eof */
5373 xfs_bmbt_rec_t *ep; /* extent record pointer */
5374 int error; /* error return value */
5375 xfs_extnum_t extno; /* extent number in list */
5376 xfs_bmbt_irec_t got; /* current extent record */
5377 xfs_ifork_t *ifp; /* inode fork pointer */
5378 int isrt; /* freeing in rt area */
5379 xfs_extnum_t lastx; /* last extent index used */
5380 int logflags; /* transaction logging flags */
5381 xfs_extlen_t mod; /* rt extent offset */
5382 xfs_mount_t *mp; /* mount structure */
5383 xfs_extnum_t nextents; /* number of file extents */
5384 xfs_bmbt_irec_t prev; /* previous extent record */
5385 xfs_fileoff_t start; /* first file offset deleted */
5386 int tmp_logflags; /* partial logging flags */
5387 int wasdel; /* was a delayed alloc extent */
5388 int whichfork; /* data or attribute fork */
5389 int rsvd; /* OK to allocate reserved blocks */
5390 xfs_fsblock_t sum;
5391
5392 xfs_bunmap_trace(ip, bno, len, flags, (inst_t *)__return_address);
5393 whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5394 XFS_ATTR_FORK : XFS_DATA_FORK;
5395 ifp = XFS_IFORK_PTR(ip, whichfork);
5396 if (unlikely(
5397 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5398 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5399 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5400 ip->i_mount);
5401 return XFS_ERROR(EFSCORRUPTED);
5402 }
5403 mp = ip->i_mount;
5404 if (XFS_FORCED_SHUTDOWN(mp))
5405 return XFS_ERROR(EIO);
5406 rsvd = (flags & XFS_BMAPI_RSVBLOCKS) != 0;
5407 ASSERT(len > 0);
5408 ASSERT(nexts >= 0);
5409 ASSERT(ifp->if_ext_max ==
5410 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5411 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5412 (error = xfs_iread_extents(tp, ip, whichfork)))
5413 return error;
5414 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5415 if (nextents == 0) {
5416 *done = 1;
5417 return 0;
5418 }
5419 XFS_STATS_INC(xs_blk_unmap);
5420 isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5421 start = bno;
5422 bno = start + len - 1;
5423 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5424 &prev);
5425 if (delta) {
5426 delta->xed_startoff = NULLFILEOFF;
5427 delta->xed_blockcount = 0;
5428 }
5429 /*
5430 * Check to see if the given block number is past the end of the
5431 * file, back up to the last block if so...
5432 */
5433 if (eof) {
5434 ep = xfs_iext_get_ext(ifp, --lastx);
5435 xfs_bmbt_get_all(ep, &got);
5436 bno = got.br_startoff + got.br_blockcount - 1;
5437 }
5438 logflags = 0;
5439 if (ifp->if_flags & XFS_IFBROOT) {
5440 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5441 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
5442 whichfork);
5443 cur->bc_private.b.firstblock = *firstblock;
5444 cur->bc_private.b.flist = flist;
5445 cur->bc_private.b.flags = 0;
5446 } else
5447 cur = NULL;
5448 extno = 0;
5449 while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5450 (nexts == 0 || extno < nexts)) {
5451 /*
5452 * Is the found extent after a hole in which bno lives?
5453 * Just back up to the previous extent, if so.
5454 */
5455 if (got.br_startoff > bno) {
5456 if (--lastx < 0)
5457 break;
5458 ep = xfs_iext_get_ext(ifp, lastx);
5459 xfs_bmbt_get_all(ep, &got);
5460 }
5461 /*
5462 * Is the last block of this extent before the range
5463 * we're supposed to delete? If so, we're done.
5464 */
5465 bno = XFS_FILEOFF_MIN(bno,
5466 got.br_startoff + got.br_blockcount - 1);
5467 if (bno < start)
5468 break;
5469 /*
5470 * Then deal with the (possibly delayed) allocated space
5471 * we found.
5472 */
5473 ASSERT(ep != NULL);
5474 del = got;
5475 wasdel = ISNULLSTARTBLOCK(del.br_startblock);
5476 if (got.br_startoff < start) {
5477 del.br_startoff = start;
5478 del.br_blockcount -= start - got.br_startoff;
5479 if (!wasdel)
5480 del.br_startblock += start - got.br_startoff;
5481 }
5482 if (del.br_startoff + del.br_blockcount > bno + 1)
5483 del.br_blockcount = bno + 1 - del.br_startoff;
5484 sum = del.br_startblock + del.br_blockcount;
5485 if (isrt &&
5486 (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5487 /*
5488 * Realtime extent not lined up at the end.
5489 * The extent could have been split into written
5490 * and unwritten pieces, or we could just be
5491 * unmapping part of it. But we can't really
5492 * get rid of part of a realtime extent.
5493 */
5494 if (del.br_state == XFS_EXT_UNWRITTEN ||
5495 !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
5496 /*
5497 * This piece is unwritten, or we're not
5498 * using unwritten extents. Skip over it.
5499 */
5500 ASSERT(bno >= mod);
5501 bno -= mod > del.br_blockcount ?
5502 del.br_blockcount : mod;
5503 if (bno < got.br_startoff) {
5504 if (--lastx >= 0)
5505 xfs_bmbt_get_all(xfs_iext_get_ext(
5506 ifp, lastx), &got);
5507 }
5508 continue;
5509 }
5510 /*
5511 * It's written, turn it unwritten.
5512 * This is better than zeroing it.
5513 */
5514 ASSERT(del.br_state == XFS_EXT_NORM);
5515 ASSERT(xfs_trans_get_block_res(tp) > 0);
5516 /*
5517 * If this spans a realtime extent boundary,
5518 * chop it back to the start of the one we end at.
5519 */
5520 if (del.br_blockcount > mod) {
5521 del.br_startoff += del.br_blockcount - mod;
5522 del.br_startblock += del.br_blockcount - mod;
5523 del.br_blockcount = mod;
5524 }
5525 del.br_state = XFS_EXT_UNWRITTEN;
5526 error = xfs_bmap_add_extent(ip, lastx, &cur, &del,
5527 firstblock, flist, &logflags, delta,
5528 XFS_DATA_FORK, 0);
5529 if (error)
5530 goto error0;
5531 goto nodelete;
5532 }
5533 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5534 /*
5535 * Realtime extent is lined up at the end but not
5536 * at the front. We'll get rid of full extents if
5537 * we can.
5538 */
5539 mod = mp->m_sb.sb_rextsize - mod;
5540 if (del.br_blockcount > mod) {
5541 del.br_blockcount -= mod;
5542 del.br_startoff += mod;
5543 del.br_startblock += mod;
5544 } else if ((del.br_startoff == start &&
5545 (del.br_state == XFS_EXT_UNWRITTEN ||
5546 xfs_trans_get_block_res(tp) == 0)) ||
5547 !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
5548 /*
5549 * Can't make it unwritten. There isn't
5550 * a full extent here so just skip it.
5551 */
5552 ASSERT(bno >= del.br_blockcount);
5553 bno -= del.br_blockcount;
5554 if (bno < got.br_startoff) {
5555 if (--lastx >= 0)
5556 xfs_bmbt_get_all(--ep, &got);
5557 }
5558 continue;
5559 } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5560 /*
5561 * This one is already unwritten.
5562 * It must have a written left neighbor.
5563 * Unwrite the killed part of that one and
5564 * try again.
5565 */
5566 ASSERT(lastx > 0);
5567 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5568 lastx - 1), &prev);
5569 ASSERT(prev.br_state == XFS_EXT_NORM);
5570 ASSERT(!ISNULLSTARTBLOCK(prev.br_startblock));
5571 ASSERT(del.br_startblock ==
5572 prev.br_startblock + prev.br_blockcount);
5573 if (prev.br_startoff < start) {
5574 mod = start - prev.br_startoff;
5575 prev.br_blockcount -= mod;
5576 prev.br_startblock += mod;
5577 prev.br_startoff = start;
5578 }
5579 prev.br_state = XFS_EXT_UNWRITTEN;
5580 error = xfs_bmap_add_extent(ip, lastx - 1, &cur,
5581 &prev, firstblock, flist, &logflags,
5582 delta, XFS_DATA_FORK, 0);
5583 if (error)
5584 goto error0;
5585 goto nodelete;
5586 } else {
5587 ASSERT(del.br_state == XFS_EXT_NORM);
5588 del.br_state = XFS_EXT_UNWRITTEN;
5589 error = xfs_bmap_add_extent(ip, lastx, &cur,
5590 &del, firstblock, flist, &logflags,
5591 delta, XFS_DATA_FORK, 0);
5592 if (error)
5593 goto error0;
5594 goto nodelete;
5595 }
5596 }
5597 if (wasdel) {
5598 ASSERT(STARTBLOCKVAL(del.br_startblock) > 0);
5599 /* Update realtime/data freespace, unreserve quota */
5600 if (isrt) {
5601 xfs_filblks_t rtexts;
5602
5603 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5604 do_div(rtexts, mp->m_sb.sb_rextsize);
5605 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5606 (int64_t)rtexts, rsvd);
5607 (void)XFS_TRANS_RESERVE_QUOTA_NBLKS(mp,
5608 NULL, ip, -((long)del.br_blockcount), 0,
5609 XFS_QMOPT_RES_RTBLKS);
5610 } else {
5611 xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS,
5612 (int64_t)del.br_blockcount, rsvd);
5613 (void)XFS_TRANS_RESERVE_QUOTA_NBLKS(mp,
5614 NULL, ip, -((long)del.br_blockcount), 0,
5615 XFS_QMOPT_RES_REGBLKS);
5616 }
5617 ip->i_delayed_blks -= del.br_blockcount;
5618 if (cur)
5619 cur->bc_private.b.flags |=
5620 XFS_BTCUR_BPRV_WASDEL;
5621 } else if (cur)
5622 cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5623 /*
5624 * If it's the case where the directory code is running
5625 * with no block reservation, and the deleted block is in
5626 * the middle of its extent, and the resulting insert
5627 * of an extent would cause transformation to btree format,
5628 * then reject it. The calling code will then swap
5629 * blocks around instead.
5630 * We have to do this now, rather than waiting for the
5631 * conversion to btree format, since the transaction
5632 * will be dirty.
5633 */
5634 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5635 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5636 XFS_IFORK_NEXTENTS(ip, whichfork) >= ifp->if_ext_max &&
5637 del.br_startoff > got.br_startoff &&
5638 del.br_startoff + del.br_blockcount <
5639 got.br_startoff + got.br_blockcount) {
5640 error = XFS_ERROR(ENOSPC);
5641 goto error0;
5642 }
5643 error = xfs_bmap_del_extent(ip, tp, lastx, flist, cur, &del,
5644 &tmp_logflags, delta, whichfork, rsvd);
5645 logflags |= tmp_logflags;
5646 if (error)
5647 goto error0;
5648 bno = del.br_startoff - 1;
5649 nodelete:
5650 lastx = ifp->if_lastex;
5651 /*
5652 * If not done go on to the next (previous) record.
5653 * Reset ep in case the extents array was re-alloced.
5654 */
5655 ep = xfs_iext_get_ext(ifp, lastx);
5656 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5657 if (lastx >= XFS_IFORK_NEXTENTS(ip, whichfork) ||
5658 xfs_bmbt_get_startoff(ep) > bno) {
5659 if (--lastx >= 0)
5660 ep = xfs_iext_get_ext(ifp, lastx);
5661 }
5662 if (lastx >= 0)
5663 xfs_bmbt_get_all(ep, &got);
5664 extno++;
5665 }
5666 }
5667 ifp->if_lastex = lastx;
5668 *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5669 ASSERT(ifp->if_ext_max ==
5670 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5671 /*
5672 * Convert to a btree if necessary.
5673 */
5674 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5675 XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
5676 ASSERT(cur == NULL);
5677 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5678 &cur, 0, &tmp_logflags, whichfork);
5679 logflags |= tmp_logflags;
5680 if (error)
5681 goto error0;
5682 }
5683 /*
5684 * transform from btree to extents, give it cur
5685 */
5686 else if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5687 XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5688 ASSERT(cur != NULL);
5689 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5690 whichfork);
5691 logflags |= tmp_logflags;
5692 if (error)
5693 goto error0;
5694 }
5695 /*
5696 * transform from extents to local?
5697 */
5698 ASSERT(ifp->if_ext_max ==
5699 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5700 error = 0;
5701 if (delta && delta->xed_startoff != NULLFILEOFF) {
5702 /* A change was actually made.
5703 * Note that delta->xed_blockount is an offset at this
5704 * point and needs to be converted to a block count.
5705 */
5706 ASSERT(delta->xed_blockcount > delta->xed_startoff);
5707 delta->xed_blockcount -= delta->xed_startoff;
5708 }
5709 error0:
5710 /*
5711 * Log everything. Do this after conversion, there's no point in
5712 * logging the extent records if we've converted to btree format.
5713 */
5714 if ((logflags & XFS_ILOG_FEXT(whichfork)) &&
5715 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5716 logflags &= ~XFS_ILOG_FEXT(whichfork);
5717 else if ((logflags & XFS_ILOG_FBROOT(whichfork)) &&
5718 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5719 logflags &= ~XFS_ILOG_FBROOT(whichfork);
5720 /*
5721 * Log inode even in the error case, if the transaction
5722 * is dirty we'll need to shut down the filesystem.
5723 */
5724 if (logflags)
5725 xfs_trans_log_inode(tp, ip, logflags);
5726 if (cur) {
5727 if (!error) {
5728 *firstblock = cur->bc_private.b.firstblock;
5729 cur->bc_private.b.allocated = 0;
5730 }
5731 xfs_btree_del_cursor(cur,
5732 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5733 }
5734 return error;
5735 }
5736
5737 /*
5738 * Fcntl interface to xfs_bmapi.
5739 */
5740 int /* error code */
5741 xfs_getbmap(
5742 bhv_desc_t *bdp, /* XFS behavior descriptor*/
5743 struct getbmap *bmv, /* user bmap structure */
5744 void __user *ap, /* pointer to user's array */
5745 int interface) /* interface flags */
5746 {
5747 __int64_t bmvend; /* last block requested */
5748 int error; /* return value */
5749 __int64_t fixlen; /* length for -1 case */
5750 int i; /* extent number */
5751 xfs_inode_t *ip; /* xfs incore inode pointer */
5752 bhv_vnode_t *vp; /* corresponding vnode */
5753 int lock; /* lock state */
5754 xfs_bmbt_irec_t *map; /* buffer for user's data */
5755 xfs_mount_t *mp; /* file system mount point */
5756 int nex; /* # of user extents can do */
5757 int nexleft; /* # of user extents left */
5758 int subnex; /* # of bmapi's can do */
5759 int nmap; /* number of map entries */
5760 struct getbmap out; /* output structure */
5761 int whichfork; /* data or attr fork */
5762 int prealloced; /* this is a file with
5763 * preallocated data space */
5764 int sh_unwritten; /* true, if unwritten */
5765 /* extents listed separately */
5766 int bmapi_flags; /* flags for xfs_bmapi */
5767 __int32_t oflags; /* getbmapx bmv_oflags field */
5768
5769 vp = BHV_TO_VNODE(bdp);
5770 ip = XFS_BHVTOI(bdp);
5771 mp = ip->i_mount;
5772
5773 whichfork = interface & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5774 sh_unwritten = (interface & BMV_IF_PREALLOC) != 0;
5775
5776 /* If the BMV_IF_NO_DMAPI_READ interface bit specified, do not
5777 * generate a DMAPI read event. Otherwise, if the DM_EVENT_READ
5778 * bit is set for the file, generate a read event in order
5779 * that the DMAPI application may do its thing before we return
5780 * the extents. Usually this means restoring user file data to
5781 * regions of the file that look like holes.
5782 *
5783 * The "old behavior" (from XFS_IOC_GETBMAP) is to not specify
5784 * BMV_IF_NO_DMAPI_READ so that read events are generated.
5785 * If this were not true, callers of ioctl( XFS_IOC_GETBMAP )
5786 * could misinterpret holes in a DMAPI file as true holes,
5787 * when in fact they may represent offline user data.
5788 */
5789 if ( (interface & BMV_IF_NO_DMAPI_READ) == 0
5790 && DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_READ)
5791 && whichfork == XFS_DATA_FORK) {
5792
5793 error = XFS_SEND_DATA(mp, DM_EVENT_READ, vp, 0, 0, 0, NULL);
5794 if (error)
5795 return XFS_ERROR(error);
5796 }
5797
5798 if (whichfork == XFS_ATTR_FORK) {
5799 if (XFS_IFORK_Q(ip)) {
5800 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5801 ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5802 ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5803 return XFS_ERROR(EINVAL);
5804 } else if (unlikely(
5805 ip->i_d.di_aformat != 0 &&
5806 ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5807 XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5808 ip->i_mount);
5809 return XFS_ERROR(EFSCORRUPTED);
5810 }
5811 } else if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5812 ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5813 ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5814 return XFS_ERROR(EINVAL);
5815 if (whichfork == XFS_DATA_FORK) {
5816 if ((ip->i_d.di_extsize && (ip->i_d.di_flags &
5817 (XFS_DIFLAG_REALTIME|XFS_DIFLAG_EXTSIZE))) ||
5818 ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5819 prealloced = 1;
5820 fixlen = XFS_MAXIOFFSET(mp);
5821 } else {
5822 prealloced = 0;
5823 fixlen = ip->i_size;
5824 }
5825 } else {
5826 prealloced = 0;
5827 fixlen = 1LL << 32;
5828 }
5829
5830 if (bmv->bmv_length == -1) {
5831 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5832 bmv->bmv_length = MAX( (__int64_t)(fixlen - bmv->bmv_offset),
5833 (__int64_t)0);
5834 } else if (bmv->bmv_length < 0)
5835 return XFS_ERROR(EINVAL);
5836 if (bmv->bmv_length == 0) {
5837 bmv->bmv_entries = 0;
5838 return 0;
5839 }
5840 nex = bmv->bmv_count - 1;
5841 if (nex <= 0)
5842 return XFS_ERROR(EINVAL);
5843 bmvend = bmv->bmv_offset + bmv->bmv_length;
5844
5845 xfs_ilock(ip, XFS_IOLOCK_SHARED);
5846
5847 if (whichfork == XFS_DATA_FORK &&
5848 (ip->i_delayed_blks || ip->i_size > ip->i_d.di_size)) {
5849 /* xfs_fsize_t last_byte = xfs_file_last_byte(ip); */
5850 error = bhv_vop_flush_pages(vp, (xfs_off_t)0, -1, 0, FI_REMAPF);
5851 }
5852
5853 ASSERT(whichfork == XFS_ATTR_FORK || ip->i_delayed_blks == 0);
5854
5855 lock = xfs_ilock_map_shared(ip);
5856
5857 /*
5858 * Don't let nex be bigger than the number of extents
5859 * we can have assuming alternating holes and real extents.
5860 */
5861 if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5862 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5863
5864 bmapi_flags = XFS_BMAPI_AFLAG(whichfork) |
5865 ((sh_unwritten) ? 0 : XFS_BMAPI_IGSTATE);
5866
5867 /*
5868 * Allocate enough space to handle "subnex" maps at a time.
5869 */
5870 subnex = 16;
5871 map = kmem_alloc(subnex * sizeof(*map), KM_SLEEP);
5872
5873 bmv->bmv_entries = 0;
5874
5875 if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0) {
5876 error = 0;
5877 goto unlock_and_return;
5878 }
5879
5880 nexleft = nex;
5881
5882 do {
5883 nmap = (nexleft > subnex) ? subnex : nexleft;
5884 error = xfs_bmapi(NULL, ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5885 XFS_BB_TO_FSB(mp, bmv->bmv_length),
5886 bmapi_flags, NULL, 0, map, &nmap,
5887 NULL, NULL);
5888 if (error)
5889 goto unlock_and_return;
5890 ASSERT(nmap <= subnex);
5891
5892 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5893 nexleft--;
5894 oflags = (map[i].br_state == XFS_EXT_UNWRITTEN) ?
5895 BMV_OF_PREALLOC : 0;
5896 out.bmv_offset = XFS_FSB_TO_BB(mp, map[i].br_startoff);
5897 out.bmv_length = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5898 ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
5899 if (map[i].br_startblock == HOLESTARTBLOCK &&
5900 ((prealloced && out.bmv_offset + out.bmv_length == bmvend) ||
5901 whichfork == XFS_ATTR_FORK )) {
5902 /*
5903 * came to hole at end of file or the end of
5904 attribute fork
5905 */
5906 goto unlock_and_return;
5907 } else {
5908 out.bmv_block =
5909 (map[i].br_startblock == HOLESTARTBLOCK) ?
5910 -1 :
5911 XFS_FSB_TO_DB(ip, map[i].br_startblock);
5912
5913 /* return either getbmap/getbmapx structure. */
5914 if (interface & BMV_IF_EXTENDED) {
5915 struct getbmapx outx;
5916
5917 GETBMAP_CONVERT(out,outx);
5918 outx.bmv_oflags = oflags;
5919 outx.bmv_unused1 = outx.bmv_unused2 = 0;
5920 if (copy_to_user(ap, &outx,
5921 sizeof(outx))) {
5922 error = XFS_ERROR(EFAULT);
5923 goto unlock_and_return;
5924 }
5925 } else {
5926 if (copy_to_user(ap, &out,
5927 sizeof(out))) {
5928 error = XFS_ERROR(EFAULT);
5929 goto unlock_and_return;
5930 }
5931 }
5932 bmv->bmv_offset =
5933 out.bmv_offset + out.bmv_length;
5934 bmv->bmv_length = MAX((__int64_t)0,
5935 (__int64_t)(bmvend - bmv->bmv_offset));
5936 bmv->bmv_entries++;
5937 ap = (interface & BMV_IF_EXTENDED) ?
5938 (void __user *)
5939 ((struct getbmapx __user *)ap + 1) :
5940 (void __user *)
5941 ((struct getbmap __user *)ap + 1);
5942 }
5943 }
5944 } while (nmap && nexleft && bmv->bmv_length);
5945
5946 unlock_and_return:
5947 xfs_iunlock_map_shared(ip, lock);
5948 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5949
5950 kmem_free(map, subnex * sizeof(*map));
5951
5952 return error;
5953 }
5954
5955 /*
5956 * Check the last inode extent to determine whether this allocation will result
5957 * in blocks being allocated at the end of the file. When we allocate new data
5958 * blocks at the end of the file which do not start at the previous data block,
5959 * we will try to align the new blocks at stripe unit boundaries.
5960 */
5961 STATIC int /* error */
5962 xfs_bmap_isaeof(
5963 xfs_inode_t *ip, /* incore inode pointer */
5964 xfs_fileoff_t off, /* file offset in fsblocks */
5965 int whichfork, /* data or attribute fork */
5966 char *aeof) /* return value */
5967 {
5968 int error; /* error return value */
5969 xfs_ifork_t *ifp; /* inode fork pointer */
5970 xfs_bmbt_rec_t *lastrec; /* extent record pointer */
5971 xfs_extnum_t nextents; /* number of file extents */
5972 xfs_bmbt_irec_t s; /* expanded extent record */
5973
5974 ASSERT(whichfork == XFS_DATA_FORK);
5975 ifp = XFS_IFORK_PTR(ip, whichfork);
5976 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5977 (error = xfs_iread_extents(NULL, ip, whichfork)))
5978 return error;
5979 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5980 if (nextents == 0) {
5981 *aeof = 1;
5982 return 0;
5983 }
5984 /*
5985 * Go to the last extent
5986 */
5987 lastrec = xfs_iext_get_ext(ifp, nextents - 1);
5988 xfs_bmbt_get_all(lastrec, &s);
5989 /*
5990 * Check we are allocating in the last extent (for delayed allocations)
5991 * or past the last extent for non-delayed allocations.
5992 */
5993 *aeof = (off >= s.br_startoff &&
5994 off < s.br_startoff + s.br_blockcount &&
5995 ISNULLSTARTBLOCK(s.br_startblock)) ||
5996 off >= s.br_startoff + s.br_blockcount;
5997 return 0;
5998 }
5999
6000 /*
6001 * Check if the endoff is outside the last extent. If so the caller will grow
6002 * the allocation to a stripe unit boundary.
6003 */
6004 int /* error */
6005 xfs_bmap_eof(
6006 xfs_inode_t *ip, /* incore inode pointer */
6007 xfs_fileoff_t endoff, /* file offset in fsblocks */
6008 int whichfork, /* data or attribute fork */
6009 int *eof) /* result value */
6010 {
6011 xfs_fsblock_t blockcount; /* extent block count */
6012 int error; /* error return value */
6013 xfs_ifork_t *ifp; /* inode fork pointer */
6014 xfs_bmbt_rec_t *lastrec; /* extent record pointer */
6015 xfs_extnum_t nextents; /* number of file extents */
6016 xfs_fileoff_t startoff; /* extent starting file offset */
6017
6018 ASSERT(whichfork == XFS_DATA_FORK);
6019 ifp = XFS_IFORK_PTR(ip, whichfork);
6020 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
6021 (error = xfs_iread_extents(NULL, ip, whichfork)))
6022 return error;
6023 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
6024 if (nextents == 0) {
6025 *eof = 1;
6026 return 0;
6027 }
6028 /*
6029 * Go to the last extent
6030 */
6031 lastrec = xfs_iext_get_ext(ifp, nextents - 1);
6032 startoff = xfs_bmbt_get_startoff(lastrec);
6033 blockcount = xfs_bmbt_get_blockcount(lastrec);
6034 *eof = endoff >= startoff + blockcount;
6035 return 0;
6036 }
6037
6038 #ifdef DEBUG
6039 STATIC
6040 xfs_buf_t *
6041 xfs_bmap_get_bp(
6042 xfs_btree_cur_t *cur,
6043 xfs_fsblock_t bno)
6044 {
6045 int i;
6046 xfs_buf_t *bp;
6047
6048 if (!cur)
6049 return(NULL);
6050
6051 bp = NULL;
6052 for(i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
6053 bp = cur->bc_bufs[i];
6054 if (!bp) break;
6055 if (XFS_BUF_ADDR(bp) == bno)
6056 break; /* Found it */
6057 }
6058 if (i == XFS_BTREE_MAXLEVELS)
6059 bp = NULL;
6060
6061 if (!bp) { /* Chase down all the log items to see if the bp is there */
6062 xfs_log_item_chunk_t *licp;
6063 xfs_trans_t *tp;
6064
6065 tp = cur->bc_tp;
6066 licp = &tp->t_items;
6067 while (!bp && licp != NULL) {
6068 if (XFS_LIC_ARE_ALL_FREE(licp)) {
6069 licp = licp->lic_next;
6070 continue;
6071 }
6072 for (i = 0; i < licp->lic_unused; i++) {
6073 xfs_log_item_desc_t *lidp;
6074 xfs_log_item_t *lip;
6075 xfs_buf_log_item_t *bip;
6076 xfs_buf_t *lbp;
6077
6078 if (XFS_LIC_ISFREE(licp, i)) {
6079 continue;
6080 }
6081
6082 lidp = XFS_LIC_SLOT(licp, i);
6083 lip = lidp->lid_item;
6084 if (lip->li_type != XFS_LI_BUF)
6085 continue;
6086
6087 bip = (xfs_buf_log_item_t *)lip;
6088 lbp = bip->bli_buf;
6089
6090 if (XFS_BUF_ADDR(lbp) == bno) {
6091 bp = lbp;
6092 break; /* Found it */
6093 }
6094 }
6095 licp = licp->lic_next;
6096 }
6097 }
6098 return(bp);
6099 }
6100
6101 void
6102 xfs_check_block(
6103 xfs_bmbt_block_t *block,
6104 xfs_mount_t *mp,
6105 int root,
6106 short sz)
6107 {
6108 int i, j, dmxr;
6109 __be64 *pp, *thispa; /* pointer to block address */
6110 xfs_bmbt_key_t *prevp, *keyp;
6111
6112 ASSERT(be16_to_cpu(block->bb_level) > 0);
6113
6114 prevp = NULL;
6115 for( i = 1; i <= be16_to_cpu(block->bb_numrecs); i++) {
6116 dmxr = mp->m_bmap_dmxr[0];
6117
6118 if (root) {
6119 keyp = XFS_BMAP_BROOT_KEY_ADDR(block, i, sz);
6120 } else {
6121 keyp = XFS_BTREE_KEY_ADDR(xfs_bmbt, block, i);
6122 }
6123
6124 if (prevp) {
6125 xfs_btree_check_key(XFS_BTNUM_BMAP, prevp, keyp);
6126 }
6127 prevp = keyp;
6128
6129 /*
6130 * Compare the block numbers to see if there are dups.
6131 */
6132
6133 if (root) {
6134 pp = XFS_BMAP_BROOT_PTR_ADDR(block, i, sz);
6135 } else {
6136 pp = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, i, dmxr);
6137 }
6138 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
6139 if (root) {
6140 thispa = XFS_BMAP_BROOT_PTR_ADDR(block, j, sz);
6141 } else {
6142 thispa = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, j,
6143 dmxr);
6144 }
6145 if (*thispa == *pp) {
6146 cmn_err(CE_WARN, "%s: thispa(%d) == pp(%d) %Ld",
6147 __FUNCTION__, j, i,
6148 (unsigned long long)be64_to_cpu(*thispa));
6149 panic("%s: ptrs are equal in node\n",
6150 __FUNCTION__);
6151 }
6152 }
6153 }
6154 }
6155
6156 /*
6157 * Check that the extents for the inode ip are in the right order in all
6158 * btree leaves.
6159 */
6160
6161 STATIC void
6162 xfs_bmap_check_leaf_extents(
6163 xfs_btree_cur_t *cur, /* btree cursor or null */
6164 xfs_inode_t *ip, /* incore inode pointer */
6165 int whichfork) /* data or attr fork */
6166 {
6167 xfs_bmbt_block_t *block; /* current btree block */
6168 xfs_fsblock_t bno; /* block # of "block" */
6169 xfs_buf_t *bp; /* buffer for "block" */
6170 int error; /* error return value */
6171 xfs_extnum_t i=0, j; /* index into the extents list */
6172 xfs_ifork_t *ifp; /* fork structure */
6173 int level; /* btree level, for checking */
6174 xfs_mount_t *mp; /* file system mount structure */
6175 __be64 *pp; /* pointer to block address */
6176 xfs_bmbt_rec_t *ep; /* pointer to current extent */
6177 xfs_bmbt_rec_t *lastp; /* pointer to previous extent */
6178 xfs_bmbt_rec_t *nextp; /* pointer to next extent */
6179 int bp_release = 0;
6180
6181 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
6182 return;
6183 }
6184
6185 bno = NULLFSBLOCK;
6186 mp = ip->i_mount;
6187 ifp = XFS_IFORK_PTR(ip, whichfork);
6188 block = ifp->if_broot;
6189 /*
6190 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6191 */
6192 level = be16_to_cpu(block->bb_level);
6193 ASSERT(level > 0);
6194 xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
6195 pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
6196 bno = be64_to_cpu(*pp);
6197
6198 ASSERT(bno != NULLDFSBNO);
6199 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
6200 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
6201
6202 /*
6203 * Go down the tree until leaf level is reached, following the first
6204 * pointer (leftmost) at each level.
6205 */
6206 while (level-- > 0) {
6207 /* See if buf is in cur first */
6208 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
6209 if (bp) {
6210 bp_release = 0;
6211 } else {
6212 bp_release = 1;
6213 }
6214 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
6215 XFS_BMAP_BTREE_REF)))
6216 goto error_norelse;
6217 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6218 XFS_WANT_CORRUPTED_GOTO(
6219 XFS_BMAP_SANITY_CHECK(mp, block, level),
6220 error0);
6221 if (level == 0)
6222 break;
6223
6224 /*
6225 * Check this block for basic sanity (increasing keys and
6226 * no duplicate blocks).
6227 */
6228
6229 xfs_check_block(block, mp, 0, 0);
6230 pp = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, 1, mp->m_bmap_dmxr[1]);
6231 bno = be64_to_cpu(*pp);
6232 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
6233 if (bp_release) {
6234 bp_release = 0;
6235 xfs_trans_brelse(NULL, bp);
6236 }
6237 }
6238
6239 /*
6240 * Here with bp and block set to the leftmost leaf node in the tree.
6241 */
6242 i = 0;
6243
6244 /*
6245 * Loop over all leaf nodes checking that all extents are in the right order.
6246 */
6247 lastp = NULL;
6248 for (;;) {
6249 xfs_fsblock_t nextbno;
6250 xfs_extnum_t num_recs;
6251
6252
6253 num_recs = be16_to_cpu(block->bb_numrecs);
6254
6255 /*
6256 * Read-ahead the next leaf block, if any.
6257 */
6258
6259 nextbno = be64_to_cpu(block->bb_rightsib);
6260
6261 /*
6262 * Check all the extents to make sure they are OK.
6263 * If we had a previous block, the last entry should
6264 * conform with the first entry in this one.
6265 */
6266
6267 ep = XFS_BTREE_REC_ADDR(xfs_bmbt, block, 1);
6268 for (j = 1; j < num_recs; j++) {
6269 nextp = XFS_BTREE_REC_ADDR(xfs_bmbt, block, j + 1);
6270 if (lastp) {
6271 xfs_btree_check_rec(XFS_BTNUM_BMAP,
6272 (void *)lastp, (void *)ep);
6273 }
6274 xfs_btree_check_rec(XFS_BTNUM_BMAP, (void *)ep,
6275 (void *)(nextp));
6276 lastp = ep;
6277 ep = nextp;
6278 }
6279
6280 i += num_recs;
6281 if (bp_release) {
6282 bp_release = 0;
6283 xfs_trans_brelse(NULL, bp);
6284 }
6285 bno = nextbno;
6286 /*
6287 * If we've reached the end, stop.
6288 */
6289 if (bno == NULLFSBLOCK)
6290 break;
6291
6292 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
6293 if (bp) {
6294 bp_release = 0;
6295 } else {
6296 bp_release = 1;
6297 }
6298 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
6299 XFS_BMAP_BTREE_REF)))
6300 goto error_norelse;
6301 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6302 }
6303 if (bp_release) {
6304 bp_release = 0;
6305 xfs_trans_brelse(NULL, bp);
6306 }
6307 return;
6308
6309 error0:
6310 cmn_err(CE_WARN, "%s: at error0", __FUNCTION__);
6311 if (bp_release)
6312 xfs_trans_brelse(NULL, bp);
6313 error_norelse:
6314 cmn_err(CE_WARN, "%s: BAD after btree leaves for %d extents",
6315 __FUNCTION__, i);
6316 panic("%s: CORRUPTED BTREE OR SOMETHING", __FUNCTION__);
6317 return;
6318 }
6319 #endif
6320
6321 /*
6322 * Count fsblocks of the given fork.
6323 */
6324 int /* error */
6325 xfs_bmap_count_blocks(
6326 xfs_trans_t *tp, /* transaction pointer */
6327 xfs_inode_t *ip, /* incore inode */
6328 int whichfork, /* data or attr fork */
6329 int *count) /* out: count of blocks */
6330 {
6331 xfs_bmbt_block_t *block; /* current btree block */
6332 xfs_fsblock_t bno; /* block # of "block" */
6333 xfs_ifork_t *ifp; /* fork structure */
6334 int level; /* btree level, for checking */
6335 xfs_mount_t *mp; /* file system mount structure */
6336 __be64 *pp; /* pointer to block address */
6337
6338 bno = NULLFSBLOCK;
6339 mp = ip->i_mount;
6340 ifp = XFS_IFORK_PTR(ip, whichfork);
6341 if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
6342 if (unlikely(xfs_bmap_count_leaves(ifp, 0,
6343 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
6344 count) < 0)) {
6345 XFS_ERROR_REPORT("xfs_bmap_count_blocks(1)",
6346 XFS_ERRLEVEL_LOW, mp);
6347 return XFS_ERROR(EFSCORRUPTED);
6348 }
6349 return 0;
6350 }
6351
6352 /*
6353 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6354 */
6355 block = ifp->if_broot;
6356 level = be16_to_cpu(block->bb_level);
6357 ASSERT(level > 0);
6358 pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
6359 bno = be64_to_cpu(*pp);
6360 ASSERT(bno != NULLDFSBNO);
6361 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
6362 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
6363
6364 if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
6365 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
6366 mp);
6367 return XFS_ERROR(EFSCORRUPTED);
6368 }
6369
6370 return 0;
6371 }
6372
6373 /*
6374 * Recursively walks each level of a btree
6375 * to count total fsblocks is use.
6376 */
6377 int /* error */
6378 xfs_bmap_count_tree(
6379 xfs_mount_t *mp, /* file system mount point */
6380 xfs_trans_t *tp, /* transaction pointer */
6381 xfs_ifork_t *ifp, /* inode fork pointer */
6382 xfs_fsblock_t blockno, /* file system block number */
6383 int levelin, /* level in btree */
6384 int *count) /* Count of blocks */
6385 {
6386 int error;
6387 xfs_buf_t *bp, *nbp;
6388 int level = levelin;
6389 __be64 *pp;
6390 xfs_fsblock_t bno = blockno;
6391 xfs_fsblock_t nextbno;
6392 xfs_bmbt_block_t *block, *nextblock;
6393 int numrecs;
6394
6395 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
6396 return error;
6397 *count += 1;
6398 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6399
6400 if (--level) {
6401 /* Not at node above leafs, count this level of nodes */
6402 nextbno = be64_to_cpu(block->bb_rightsib);
6403 while (nextbno != NULLFSBLOCK) {
6404 if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
6405 0, &nbp, XFS_BMAP_BTREE_REF)))
6406 return error;
6407 *count += 1;
6408 nextblock = XFS_BUF_TO_BMBT_BLOCK(nbp);
6409 nextbno = be64_to_cpu(nextblock->bb_rightsib);
6410 xfs_trans_brelse(tp, nbp);
6411 }
6412
6413 /* Dive to the next level */
6414 pp = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, 1, mp->m_bmap_dmxr[1]);
6415 bno = be64_to_cpu(*pp);
6416 if (unlikely((error =
6417 xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
6418 xfs_trans_brelse(tp, bp);
6419 XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6420 XFS_ERRLEVEL_LOW, mp);
6421 return XFS_ERROR(EFSCORRUPTED);
6422 }
6423 xfs_trans_brelse(tp, bp);
6424 } else {
6425 /* count all level 1 nodes and their leaves */
6426 for (;;) {
6427 nextbno = be64_to_cpu(block->bb_rightsib);
6428 numrecs = be16_to_cpu(block->bb_numrecs);
6429 if (unlikely(xfs_bmap_disk_count_leaves(0,
6430 block, numrecs, count) < 0)) {
6431 xfs_trans_brelse(tp, bp);
6432 XFS_ERROR_REPORT("xfs_bmap_count_tree(2)",
6433 XFS_ERRLEVEL_LOW, mp);
6434 return XFS_ERROR(EFSCORRUPTED);
6435 }
6436 xfs_trans_brelse(tp, bp);
6437 if (nextbno == NULLFSBLOCK)
6438 break;
6439 bno = nextbno;
6440 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6441 XFS_BMAP_BTREE_REF)))
6442 return error;
6443 *count += 1;
6444 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6445 }
6446 }
6447 return 0;
6448 }
6449
6450 /*
6451 * Count leaf blocks given a range of extent records.
6452 */
6453 int
6454 xfs_bmap_count_leaves(
6455 xfs_ifork_t *ifp,
6456 xfs_extnum_t idx,
6457 int numrecs,
6458 int *count)
6459 {
6460 int b;
6461 xfs_bmbt_rec_t *frp;
6462
6463 for (b = 0; b < numrecs; b++) {
6464 frp = xfs_iext_get_ext(ifp, idx + b);
6465 *count += xfs_bmbt_get_blockcount(frp);
6466 }
6467 return 0;
6468 }
6469
6470 /*
6471 * Count leaf blocks given a range of extent records originally
6472 * in btree format.
6473 */
6474 int
6475 xfs_bmap_disk_count_leaves(
6476 xfs_extnum_t idx,
6477 xfs_bmbt_block_t *block,
6478 int numrecs,
6479 int *count)
6480 {
6481 int b;
6482 xfs_bmbt_rec_t *frp;
6483
6484 for (b = 1; b <= numrecs; b++) {
6485 frp = XFS_BTREE_REC_ADDR(xfs_bmbt, block, idx + b);
6486 *count += xfs_bmbt_disk_get_blockcount(frp);
6487 }
6488 return 0;
6489 }
Verdex Pro support