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_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_bit.h"
25 #include "xfs_sb.h"
26 #include "xfs_mount.h"
27 #include "xfs_defer.h"
28 #include "xfs_da_format.h"
29 #include "xfs_da_btree.h"
30 #include "xfs_dir2.h"
31 #include "xfs_inode.h"
32 #include "xfs_btree.h"
33 #include "xfs_trans.h"
34 #include "xfs_inode_item.h"
35 #include "xfs_extfree_item.h"
36 #include "xfs_alloc.h"
37 #include "xfs_bmap.h"
38 #include "xfs_bmap_util.h"
39 #include "xfs_bmap_btree.h"
40 #include "xfs_rtalloc.h"
41 #include "xfs_error.h"
42 #include "xfs_quota.h"
43 #include "xfs_trans_space.h"
44 #include "xfs_buf_item.h"
45 #include "xfs_trace.h"
46 #include "xfs_symlink.h"
47 #include "xfs_attr_leaf.h"
48 #include "xfs_filestream.h"
49 #include "xfs_rmap.h"
50 #include "xfs_ag_resv.h"
51 #include "xfs_refcount.h"
52 #include "xfs_icache.h"
53
54
55 kmem_zone_t *xfs_bmap_free_item_zone;
56
57 /*
58 * Miscellaneous helper functions
59 */
60
61 /*
62 * Compute and fill in the value of the maximum depth of a bmap btree
63 * in this filesystem. Done once, during mount.
64 */
65 void
xfs_bmap_compute_maxlevels(xfs_mount_t * mp,int whichfork)66 xfs_bmap_compute_maxlevels(
67 xfs_mount_t *mp, /* file system mount structure */
68 int whichfork) /* data or attr fork */
69 {
70 int level; /* btree level */
71 uint maxblocks; /* max blocks at this level */
72 uint maxleafents; /* max leaf entries possible */
73 int maxrootrecs; /* max records in root block */
74 int minleafrecs; /* min records in leaf block */
75 int minnoderecs; /* min records in node block */
76 int sz; /* root block size */
77
78 /*
79 * The maximum number of extents in a file, hence the maximum
80 * number of leaf entries, is controlled by the type of di_nextents
81 * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
82 * (a signed 16-bit number, xfs_aextnum_t).
83 *
84 * Note that we can no longer assume that if we are in ATTR1 that
85 * the fork offset of all the inodes will be
86 * (xfs_default_attroffset(ip) >> 3) because we could have mounted
87 * with ATTR2 and then mounted back with ATTR1, keeping the
88 * di_forkoff's fixed but probably at various positions. Therefore,
89 * for both ATTR1 and ATTR2 we have to assume the worst case scenario
90 * of a minimum size available.
91 */
92 if (whichfork == XFS_DATA_FORK) {
93 maxleafents = MAXEXTNUM;
94 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
95 } else {
96 maxleafents = MAXAEXTNUM;
97 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
98 }
99 maxrootrecs = xfs_bmdr_maxrecs(sz, 0);
100 minleafrecs = mp->m_bmap_dmnr[0];
101 minnoderecs = mp->m_bmap_dmnr[1];
102 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
103 for (level = 1; maxblocks > 1; level++) {
104 if (maxblocks <= maxrootrecs)
105 maxblocks = 1;
106 else
107 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
108 }
109 mp->m_bm_maxlevels[whichfork] = level;
110 }
111
112 STATIC int /* error */
xfs_bmbt_lookup_eq(struct xfs_btree_cur * cur,xfs_fileoff_t off,xfs_fsblock_t bno,xfs_filblks_t len,int * stat)113 xfs_bmbt_lookup_eq(
114 struct xfs_btree_cur *cur,
115 xfs_fileoff_t off,
116 xfs_fsblock_t bno,
117 xfs_filblks_t len,
118 int *stat) /* success/failure */
119 {
120 cur->bc_rec.b.br_startoff = off;
121 cur->bc_rec.b.br_startblock = bno;
122 cur->bc_rec.b.br_blockcount = len;
123 return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
124 }
125
126 STATIC int /* error */
xfs_bmbt_lookup_ge(struct xfs_btree_cur * cur,xfs_fileoff_t off,xfs_fsblock_t bno,xfs_filblks_t len,int * stat)127 xfs_bmbt_lookup_ge(
128 struct xfs_btree_cur *cur,
129 xfs_fileoff_t off,
130 xfs_fsblock_t bno,
131 xfs_filblks_t len,
132 int *stat) /* success/failure */
133 {
134 cur->bc_rec.b.br_startoff = off;
135 cur->bc_rec.b.br_startblock = bno;
136 cur->bc_rec.b.br_blockcount = len;
137 return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
138 }
139
140 /*
141 * Check if the inode needs to be converted to btree format.
142 */
xfs_bmap_needs_btree(struct xfs_inode * ip,int whichfork)143 static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
144 {
145 return whichfork != XFS_COW_FORK &&
146 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
147 XFS_IFORK_NEXTENTS(ip, whichfork) >
148 XFS_IFORK_MAXEXT(ip, whichfork);
149 }
150
151 /*
152 * Check if the inode should be converted to extent format.
153 */
xfs_bmap_wants_extents(struct xfs_inode * ip,int whichfork)154 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
155 {
156 return whichfork != XFS_COW_FORK &&
157 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
158 XFS_IFORK_NEXTENTS(ip, whichfork) <=
159 XFS_IFORK_MAXEXT(ip, whichfork);
160 }
161
162 /*
163 * Update the record referred to by cur to the value given
164 * by [off, bno, len, state].
165 * This either works (return 0) or gets an EFSCORRUPTED error.
166 */
167 STATIC int
xfs_bmbt_update(struct xfs_btree_cur * cur,xfs_fileoff_t off,xfs_fsblock_t bno,xfs_filblks_t len,xfs_exntst_t state)168 xfs_bmbt_update(
169 struct xfs_btree_cur *cur,
170 xfs_fileoff_t off,
171 xfs_fsblock_t bno,
172 xfs_filblks_t len,
173 xfs_exntst_t state)
174 {
175 union xfs_btree_rec rec;
176
177 xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
178 return xfs_btree_update(cur, &rec);
179 }
180
181 /*
182 * Compute the worst-case number of indirect blocks that will be used
183 * for ip's delayed extent of length "len".
184 */
185 STATIC xfs_filblks_t
xfs_bmap_worst_indlen(xfs_inode_t * ip,xfs_filblks_t len)186 xfs_bmap_worst_indlen(
187 xfs_inode_t *ip, /* incore inode pointer */
188 xfs_filblks_t len) /* delayed extent length */
189 {
190 int level; /* btree level number */
191 int maxrecs; /* maximum record count at this level */
192 xfs_mount_t *mp; /* mount structure */
193 xfs_filblks_t rval; /* return value */
194
195 mp = ip->i_mount;
196 maxrecs = mp->m_bmap_dmxr[0];
197 for (level = 0, rval = 0;
198 level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
199 level++) {
200 len += maxrecs - 1;
201 do_div(len, maxrecs);
202 rval += len;
203 if (len == 1)
204 return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
205 level - 1;
206 if (level == 0)
207 maxrecs = mp->m_bmap_dmxr[1];
208 }
209 return rval;
210 }
211
212 /*
213 * Calculate the default attribute fork offset for newly created inodes.
214 */
215 uint
xfs_default_attroffset(struct xfs_inode * ip)216 xfs_default_attroffset(
217 struct xfs_inode *ip)
218 {
219 struct xfs_mount *mp = ip->i_mount;
220 uint offset;
221
222 if (mp->m_sb.sb_inodesize == 256) {
223 offset = XFS_LITINO(mp, ip->i_d.di_version) -
224 XFS_BMDR_SPACE_CALC(MINABTPTRS);
225 } else {
226 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
227 }
228
229 ASSERT(offset < XFS_LITINO(mp, ip->i_d.di_version));
230 return offset;
231 }
232
233 /*
234 * Helper routine to reset inode di_forkoff field when switching
235 * attribute fork from local to extent format - we reset it where
236 * possible to make space available for inline data fork extents.
237 */
238 STATIC void
xfs_bmap_forkoff_reset(xfs_inode_t * ip,int whichfork)239 xfs_bmap_forkoff_reset(
240 xfs_inode_t *ip,
241 int whichfork)
242 {
243 if (whichfork == XFS_ATTR_FORK &&
244 ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
245 ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
246 ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
247 uint dfl_forkoff = xfs_default_attroffset(ip) >> 3;
248
249 if (dfl_forkoff > ip->i_d.di_forkoff)
250 ip->i_d.di_forkoff = dfl_forkoff;
251 }
252 }
253
254 #ifdef DEBUG
255 STATIC struct xfs_buf *
xfs_bmap_get_bp(struct xfs_btree_cur * cur,xfs_fsblock_t bno)256 xfs_bmap_get_bp(
257 struct xfs_btree_cur *cur,
258 xfs_fsblock_t bno)
259 {
260 struct xfs_log_item_desc *lidp;
261 int i;
262
263 if (!cur)
264 return NULL;
265
266 for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
267 if (!cur->bc_bufs[i])
268 break;
269 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
270 return cur->bc_bufs[i];
271 }
272
273 /* Chase down all the log items to see if the bp is there */
274 list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
275 struct xfs_buf_log_item *bip;
276 bip = (struct xfs_buf_log_item *)lidp->lid_item;
277 if (bip->bli_item.li_type == XFS_LI_BUF &&
278 XFS_BUF_ADDR(bip->bli_buf) == bno)
279 return bip->bli_buf;
280 }
281
282 return NULL;
283 }
284
285 STATIC void
xfs_check_block(struct xfs_btree_block * block,xfs_mount_t * mp,int root,short sz)286 xfs_check_block(
287 struct xfs_btree_block *block,
288 xfs_mount_t *mp,
289 int root,
290 short sz)
291 {
292 int i, j, dmxr;
293 __be64 *pp, *thispa; /* pointer to block address */
294 xfs_bmbt_key_t *prevp, *keyp;
295
296 ASSERT(be16_to_cpu(block->bb_level) > 0);
297
298 prevp = NULL;
299 for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
300 dmxr = mp->m_bmap_dmxr[0];
301 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
302
303 if (prevp) {
304 ASSERT(be64_to_cpu(prevp->br_startoff) <
305 be64_to_cpu(keyp->br_startoff));
306 }
307 prevp = keyp;
308
309 /*
310 * Compare the block numbers to see if there are dups.
311 */
312 if (root)
313 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
314 else
315 pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
316
317 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
318 if (root)
319 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
320 else
321 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
322 if (*thispa == *pp) {
323 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
324 __func__, j, i,
325 (unsigned long long)be64_to_cpu(*thispa));
326 panic("%s: ptrs are equal in node\n",
327 __func__);
328 }
329 }
330 }
331 }
332
333 /*
334 * Check that the extents for the inode ip are in the right order in all
335 * btree leaves. THis becomes prohibitively expensive for large extent count
336 * files, so don't bother with inodes that have more than 10,000 extents in
337 * them. The btree record ordering checks will still be done, so for such large
338 * bmapbt constructs that is going to catch most corruptions.
339 */
340 STATIC void
xfs_bmap_check_leaf_extents(xfs_btree_cur_t * cur,xfs_inode_t * ip,int whichfork)341 xfs_bmap_check_leaf_extents(
342 xfs_btree_cur_t *cur, /* btree cursor or null */
343 xfs_inode_t *ip, /* incore inode pointer */
344 int whichfork) /* data or attr fork */
345 {
346 struct xfs_btree_block *block; /* current btree block */
347 xfs_fsblock_t bno; /* block # of "block" */
348 xfs_buf_t *bp; /* buffer for "block" */
349 int error; /* error return value */
350 xfs_extnum_t i=0, j; /* index into the extents list */
351 xfs_ifork_t *ifp; /* fork structure */
352 int level; /* btree level, for checking */
353 xfs_mount_t *mp; /* file system mount structure */
354 __be64 *pp; /* pointer to block address */
355 xfs_bmbt_rec_t *ep; /* pointer to current extent */
356 xfs_bmbt_rec_t last = {0, 0}; /* last extent in prev block */
357 xfs_bmbt_rec_t *nextp; /* pointer to next extent */
358 int bp_release = 0;
359
360 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
361 return;
362 }
363
364 /* skip large extent count inodes */
365 if (ip->i_d.di_nextents > 10000)
366 return;
367
368 bno = NULLFSBLOCK;
369 mp = ip->i_mount;
370 ifp = XFS_IFORK_PTR(ip, whichfork);
371 block = ifp->if_broot;
372 /*
373 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
374 */
375 level = be16_to_cpu(block->bb_level);
376 ASSERT(level > 0);
377 xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
378 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
379 bno = be64_to_cpu(*pp);
380
381 ASSERT(bno != NULLFSBLOCK);
382 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
383 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
384
385 /*
386 * Go down the tree until leaf level is reached, following the first
387 * pointer (leftmost) at each level.
388 */
389 while (level-- > 0) {
390 /* See if buf is in cur first */
391 bp_release = 0;
392 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
393 if (!bp) {
394 bp_release = 1;
395 error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
396 XFS_BMAP_BTREE_REF,
397 &xfs_bmbt_buf_ops);
398 if (error)
399 goto error_norelse;
400 }
401 block = XFS_BUF_TO_BLOCK(bp);
402 if (level == 0)
403 break;
404
405 /*
406 * Check this block for basic sanity (increasing keys and
407 * no duplicate blocks).
408 */
409
410 xfs_check_block(block, mp, 0, 0);
411 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
412 bno = be64_to_cpu(*pp);
413 XFS_WANT_CORRUPTED_GOTO(mp,
414 XFS_FSB_SANITY_CHECK(mp, bno), error0);
415 if (bp_release) {
416 bp_release = 0;
417 xfs_trans_brelse(NULL, bp);
418 }
419 }
420
421 /*
422 * Here with bp and block set to the leftmost leaf node in the tree.
423 */
424 i = 0;
425
426 /*
427 * Loop over all leaf nodes checking that all extents are in the right order.
428 */
429 for (;;) {
430 xfs_fsblock_t nextbno;
431 xfs_extnum_t num_recs;
432
433
434 num_recs = xfs_btree_get_numrecs(block);
435
436 /*
437 * Read-ahead the next leaf block, if any.
438 */
439
440 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
441
442 /*
443 * Check all the extents to make sure they are OK.
444 * If we had a previous block, the last entry should
445 * conform with the first entry in this one.
446 */
447
448 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
449 if (i) {
450 ASSERT(xfs_bmbt_disk_get_startoff(&last) +
451 xfs_bmbt_disk_get_blockcount(&last) <=
452 xfs_bmbt_disk_get_startoff(ep));
453 }
454 for (j = 1; j < num_recs; j++) {
455 nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
456 ASSERT(xfs_bmbt_disk_get_startoff(ep) +
457 xfs_bmbt_disk_get_blockcount(ep) <=
458 xfs_bmbt_disk_get_startoff(nextp));
459 ep = nextp;
460 }
461
462 last = *ep;
463 i += num_recs;
464 if (bp_release) {
465 bp_release = 0;
466 xfs_trans_brelse(NULL, bp);
467 }
468 bno = nextbno;
469 /*
470 * If we've reached the end, stop.
471 */
472 if (bno == NULLFSBLOCK)
473 break;
474
475 bp_release = 0;
476 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
477 if (!bp) {
478 bp_release = 1;
479 error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
480 XFS_BMAP_BTREE_REF,
481 &xfs_bmbt_buf_ops);
482 if (error)
483 goto error_norelse;
484 }
485 block = XFS_BUF_TO_BLOCK(bp);
486 }
487
488 return;
489
490 error0:
491 xfs_warn(mp, "%s: at error0", __func__);
492 if (bp_release)
493 xfs_trans_brelse(NULL, bp);
494 error_norelse:
495 xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
496 __func__, i);
497 panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
498 return;
499 }
500
501 /*
502 * Add bmap trace insert entries for all the contents of the extent records.
503 */
504 void
xfs_bmap_trace_exlist(xfs_inode_t * ip,xfs_extnum_t cnt,int whichfork,unsigned long caller_ip)505 xfs_bmap_trace_exlist(
506 xfs_inode_t *ip, /* incore inode pointer */
507 xfs_extnum_t cnt, /* count of entries in the list */
508 int whichfork, /* data or attr or cow fork */
509 unsigned long caller_ip)
510 {
511 xfs_extnum_t idx; /* extent record index */
512 xfs_ifork_t *ifp; /* inode fork pointer */
513 int state = 0;
514
515 if (whichfork == XFS_ATTR_FORK)
516 state |= BMAP_ATTRFORK;
517 else if (whichfork == XFS_COW_FORK)
518 state |= BMAP_COWFORK;
519
520 ifp = XFS_IFORK_PTR(ip, whichfork);
521 ASSERT(cnt == xfs_iext_count(ifp));
522 for (idx = 0; idx < cnt; idx++)
523 trace_xfs_extlist(ip, idx, state, caller_ip);
524 }
525
526 /*
527 * Validate that the bmbt_irecs being returned from bmapi are valid
528 * given the caller's original parameters. Specifically check the
529 * ranges of the returned irecs to ensure that they only extend beyond
530 * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
531 */
532 STATIC void
xfs_bmap_validate_ret(xfs_fileoff_t bno,xfs_filblks_t len,int flags,xfs_bmbt_irec_t * mval,int nmap,int ret_nmap)533 xfs_bmap_validate_ret(
534 xfs_fileoff_t bno,
535 xfs_filblks_t len,
536 int flags,
537 xfs_bmbt_irec_t *mval,
538 int nmap,
539 int ret_nmap)
540 {
541 int i; /* index to map values */
542
543 ASSERT(ret_nmap <= nmap);
544
545 for (i = 0; i < ret_nmap; i++) {
546 ASSERT(mval[i].br_blockcount > 0);
547 if (!(flags & XFS_BMAPI_ENTIRE)) {
548 ASSERT(mval[i].br_startoff >= bno);
549 ASSERT(mval[i].br_blockcount <= len);
550 ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
551 bno + len);
552 } else {
553 ASSERT(mval[i].br_startoff < bno + len);
554 ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
555 bno);
556 }
557 ASSERT(i == 0 ||
558 mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
559 mval[i].br_startoff);
560 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
561 mval[i].br_startblock != HOLESTARTBLOCK);
562 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
563 mval[i].br_state == XFS_EXT_UNWRITTEN);
564 }
565 }
566
567 #else
568 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork) do { } while (0)
569 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap) do { } while (0)
570 #endif /* DEBUG */
571
572 /*
573 * bmap free list manipulation functions
574 */
575
576 /*
577 * Add the extent to the list of extents to be free at transaction end.
578 * The list is maintained sorted (by block number).
579 */
580 void
xfs_bmap_add_free(struct xfs_mount * mp,struct xfs_defer_ops * dfops,xfs_fsblock_t bno,xfs_filblks_t len,struct xfs_owner_info * oinfo)581 xfs_bmap_add_free(
582 struct xfs_mount *mp,
583 struct xfs_defer_ops *dfops,
584 xfs_fsblock_t bno,
585 xfs_filblks_t len,
586 struct xfs_owner_info *oinfo)
587 {
588 struct xfs_extent_free_item *new; /* new element */
589 #ifdef DEBUG
590 xfs_agnumber_t agno;
591 xfs_agblock_t agbno;
592
593 ASSERT(bno != NULLFSBLOCK);
594 ASSERT(len > 0);
595 ASSERT(len <= MAXEXTLEN);
596 ASSERT(!isnullstartblock(bno));
597 agno = XFS_FSB_TO_AGNO(mp, bno);
598 agbno = XFS_FSB_TO_AGBNO(mp, bno);
599 ASSERT(agno < mp->m_sb.sb_agcount);
600 ASSERT(agbno < mp->m_sb.sb_agblocks);
601 ASSERT(len < mp->m_sb.sb_agblocks);
602 ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
603 #endif
604 ASSERT(xfs_bmap_free_item_zone != NULL);
605
606 new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
607 new->xefi_startblock = bno;
608 new->xefi_blockcount = (xfs_extlen_t)len;
609 if (oinfo)
610 new->xefi_oinfo = *oinfo;
611 else
612 xfs_rmap_skip_owner_update(&new->xefi_oinfo);
613 trace_xfs_bmap_free_defer(mp, XFS_FSB_TO_AGNO(mp, bno), 0,
614 XFS_FSB_TO_AGBNO(mp, bno), len);
615 xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_FREE, &new->xefi_list);
616 }
617
618 /*
619 * Inode fork format manipulation functions
620 */
621
622 /*
623 * Transform a btree format file with only one leaf node, where the
624 * extents list will fit in the inode, into an extents format file.
625 * Since the file extents are already in-core, all we have to do is
626 * give up the space for the btree root and pitch the leaf block.
627 */
628 STATIC int /* error */
xfs_bmap_btree_to_extents(xfs_trans_t * tp,xfs_inode_t * ip,xfs_btree_cur_t * cur,int * logflagsp,int whichfork)629 xfs_bmap_btree_to_extents(
630 xfs_trans_t *tp, /* transaction pointer */
631 xfs_inode_t *ip, /* incore inode pointer */
632 xfs_btree_cur_t *cur, /* btree cursor */
633 int *logflagsp, /* inode logging flags */
634 int whichfork) /* data or attr fork */
635 {
636 /* REFERENCED */
637 struct xfs_btree_block *cblock;/* child btree block */
638 xfs_fsblock_t cbno; /* child block number */
639 xfs_buf_t *cbp; /* child block's buffer */
640 int error; /* error return value */
641 xfs_ifork_t *ifp; /* inode fork data */
642 xfs_mount_t *mp; /* mount point structure */
643 __be64 *pp; /* ptr to block address */
644 struct xfs_btree_block *rblock;/* root btree block */
645 struct xfs_owner_info oinfo;
646
647 mp = ip->i_mount;
648 ifp = XFS_IFORK_PTR(ip, whichfork);
649 ASSERT(whichfork != XFS_COW_FORK);
650 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
651 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
652 rblock = ifp->if_broot;
653 ASSERT(be16_to_cpu(rblock->bb_level) == 1);
654 ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
655 ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
656 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
657 cbno = be64_to_cpu(*pp);
658 *logflagsp = 0;
659 #ifdef DEBUG
660 if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
661 return error;
662 #endif
663 error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF,
664 &xfs_bmbt_buf_ops);
665 if (error)
666 return error;
667 cblock = XFS_BUF_TO_BLOCK(cbp);
668 if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
669 return error;
670 xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork);
671 xfs_bmap_add_free(mp, cur->bc_private.b.dfops, cbno, 1, &oinfo);
672 ip->i_d.di_nblocks--;
673 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
674 xfs_trans_binval(tp, cbp);
675 if (cur->bc_bufs[0] == cbp)
676 cur->bc_bufs[0] = NULL;
677 xfs_iroot_realloc(ip, -1, whichfork);
678 ASSERT(ifp->if_broot == NULL);
679 ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
680 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
681 *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
682 return 0;
683 }
684
685 /*
686 * Convert an extents-format file into a btree-format file.
687 * The new file will have a root block (in the inode) and a single child block.
688 */
689 STATIC int /* error */
xfs_bmap_extents_to_btree(xfs_trans_t * tp,xfs_inode_t * ip,xfs_fsblock_t * firstblock,struct xfs_defer_ops * dfops,xfs_btree_cur_t ** curp,int wasdel,int * logflagsp,int whichfork)690 xfs_bmap_extents_to_btree(
691 xfs_trans_t *tp, /* transaction pointer */
692 xfs_inode_t *ip, /* incore inode pointer */
693 xfs_fsblock_t *firstblock, /* first-block-allocated */
694 struct xfs_defer_ops *dfops, /* blocks freed in xaction */
695 xfs_btree_cur_t **curp, /* cursor returned to caller */
696 int wasdel, /* converting a delayed alloc */
697 int *logflagsp, /* inode logging flags */
698 int whichfork) /* data or attr fork */
699 {
700 struct xfs_btree_block *ablock; /* allocated (child) bt block */
701 xfs_buf_t *abp; /* buffer for ablock */
702 xfs_alloc_arg_t args; /* allocation arguments */
703 xfs_bmbt_rec_t *arp; /* child record pointer */
704 struct xfs_btree_block *block; /* btree root block */
705 xfs_btree_cur_t *cur; /* bmap btree cursor */
706 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
707 int error; /* error return value */
708 xfs_extnum_t i, cnt; /* extent record index */
709 xfs_ifork_t *ifp; /* inode fork pointer */
710 xfs_bmbt_key_t *kp; /* root block key pointer */
711 xfs_mount_t *mp; /* mount structure */
712 xfs_extnum_t nextents; /* number of file extents */
713 xfs_bmbt_ptr_t *pp; /* root block address pointer */
714
715 mp = ip->i_mount;
716 ASSERT(whichfork != XFS_COW_FORK);
717 ifp = XFS_IFORK_PTR(ip, whichfork);
718 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
719
720 /*
721 * Make space in the inode incore.
722 */
723 xfs_iroot_realloc(ip, 1, whichfork);
724 ifp->if_flags |= XFS_IFBROOT;
725
726 /*
727 * Fill in the root.
728 */
729 block = ifp->if_broot;
730 xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
731 XFS_BTNUM_BMAP, 1, 1, ip->i_ino,
732 XFS_BTREE_LONG_PTRS);
733 /*
734 * Need a cursor. Can't allocate until bb_level is filled in.
735 */
736 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
737 cur->bc_private.b.firstblock = *firstblock;
738 cur->bc_private.b.dfops = dfops;
739 cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
740 /*
741 * Convert to a btree with two levels, one record in root.
742 */
743 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
744 memset(&args, 0, sizeof(args));
745 args.tp = tp;
746 args.mp = mp;
747 xfs_rmap_ino_bmbt_owner(&args.oinfo, ip->i_ino, whichfork);
748 args.firstblock = *firstblock;
749 if (*firstblock == NULLFSBLOCK) {
750 args.type = XFS_ALLOCTYPE_START_BNO;
751 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
752 } else if (dfops->dop_low) {
753 args.type = XFS_ALLOCTYPE_START_BNO;
754 args.fsbno = *firstblock;
755 } else {
756 args.type = XFS_ALLOCTYPE_NEAR_BNO;
757 args.fsbno = *firstblock;
758 }
759 args.minlen = args.maxlen = args.prod = 1;
760 args.wasdel = wasdel;
761 *logflagsp = 0;
762 if ((error = xfs_alloc_vextent(&args))) {
763 xfs_iroot_realloc(ip, -1, whichfork);
764 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
765 return error;
766 }
767
768 if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) {
769 xfs_iroot_realloc(ip, -1, whichfork);
770 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
771 return -ENOSPC;
772 }
773 /*
774 * Allocation can't fail, the space was reserved.
775 */
776 ASSERT(*firstblock == NULLFSBLOCK ||
777 args.agno >= XFS_FSB_TO_AGNO(mp, *firstblock));
778 *firstblock = cur->bc_private.b.firstblock = args.fsbno;
779 cur->bc_private.b.allocated++;
780 ip->i_d.di_nblocks++;
781 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
782 abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
783 /*
784 * Fill in the child block.
785 */
786 abp->b_ops = &xfs_bmbt_buf_ops;
787 ablock = XFS_BUF_TO_BLOCK(abp);
788 xfs_btree_init_block_int(mp, ablock, abp->b_bn,
789 XFS_BTNUM_BMAP, 0, 0, ip->i_ino,
790 XFS_BTREE_LONG_PTRS);
791
792 arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
793 nextents = xfs_iext_count(ifp);
794 for (cnt = i = 0; i < nextents; i++) {
795 ep = xfs_iext_get_ext(ifp, i);
796 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
797 arp->l0 = cpu_to_be64(ep->l0);
798 arp->l1 = cpu_to_be64(ep->l1);
799 arp++; cnt++;
800 }
801 }
802 ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
803 xfs_btree_set_numrecs(ablock, cnt);
804
805 /*
806 * Fill in the root key and pointer.
807 */
808 kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
809 arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
810 kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
811 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
812 be16_to_cpu(block->bb_level)));
813 *pp = cpu_to_be64(args.fsbno);
814
815 /*
816 * Do all this logging at the end so that
817 * the root is at the right level.
818 */
819 xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
820 xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
821 ASSERT(*curp == NULL);
822 *curp = cur;
823 *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
824 return 0;
825 }
826
827 /*
828 * Convert a local file to an extents file.
829 * This code is out of bounds for data forks of regular files,
830 * since the file data needs to get logged so things will stay consistent.
831 * (The bmap-level manipulations are ok, though).
832 */
833 void
xfs_bmap_local_to_extents_empty(struct xfs_inode * ip,int whichfork)834 xfs_bmap_local_to_extents_empty(
835 struct xfs_inode *ip,
836 int whichfork)
837 {
838 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
839
840 ASSERT(whichfork != XFS_COW_FORK);
841 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
842 ASSERT(ifp->if_bytes == 0);
843 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
844
845 xfs_bmap_forkoff_reset(ip, whichfork);
846 ifp->if_flags &= ~XFS_IFINLINE;
847 ifp->if_flags |= XFS_IFEXTENTS;
848 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
849 }
850
851
852 STATIC int /* error */
xfs_bmap_local_to_extents(xfs_trans_t * tp,xfs_inode_t * ip,xfs_fsblock_t * firstblock,xfs_extlen_t total,int * logflagsp,int whichfork,void (* init_fn)(struct xfs_trans * tp,struct xfs_buf * bp,struct xfs_inode * ip,struct xfs_ifork * ifp))853 xfs_bmap_local_to_extents(
854 xfs_trans_t *tp, /* transaction pointer */
855 xfs_inode_t *ip, /* incore inode pointer */
856 xfs_fsblock_t *firstblock, /* first block allocated in xaction */
857 xfs_extlen_t total, /* total blocks needed by transaction */
858 int *logflagsp, /* inode logging flags */
859 int whichfork,
860 void (*init_fn)(struct xfs_trans *tp,
861 struct xfs_buf *bp,
862 struct xfs_inode *ip,
863 struct xfs_ifork *ifp))
864 {
865 int error = 0;
866 int flags; /* logging flags returned */
867 xfs_ifork_t *ifp; /* inode fork pointer */
868 xfs_alloc_arg_t args; /* allocation arguments */
869 xfs_buf_t *bp; /* buffer for extent block */
870 struct xfs_bmbt_irec rec;
871
872 /*
873 * We don't want to deal with the case of keeping inode data inline yet.
874 * So sending the data fork of a regular inode is invalid.
875 */
876 ASSERT(!(S_ISREG(VFS_I(ip)->i_mode) && whichfork == XFS_DATA_FORK));
877 ifp = XFS_IFORK_PTR(ip, whichfork);
878 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
879
880 if (!ifp->if_bytes) {
881 xfs_bmap_local_to_extents_empty(ip, whichfork);
882 flags = XFS_ILOG_CORE;
883 goto done;
884 }
885
886 flags = 0;
887 error = 0;
888 ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) ==
889 XFS_IFINLINE);
890 memset(&args, 0, sizeof(args));
891 args.tp = tp;
892 args.mp = ip->i_mount;
893 xfs_rmap_ino_owner(&args.oinfo, ip->i_ino, whichfork, 0);
894 args.firstblock = *firstblock;
895 /*
896 * Allocate a block. We know we need only one, since the
897 * file currently fits in an inode.
898 */
899 if (*firstblock == NULLFSBLOCK) {
900 args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
901 args.type = XFS_ALLOCTYPE_START_BNO;
902 } else {
903 args.fsbno = *firstblock;
904 args.type = XFS_ALLOCTYPE_NEAR_BNO;
905 }
906 args.total = total;
907 args.minlen = args.maxlen = args.prod = 1;
908 error = xfs_alloc_vextent(&args);
909 if (error)
910 goto done;
911
912 /* Can't fail, the space was reserved. */
913 ASSERT(args.fsbno != NULLFSBLOCK);
914 ASSERT(args.len == 1);
915 *firstblock = args.fsbno;
916 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
917
918 /*
919 * Initialize the block, copy the data and log the remote buffer.
920 *
921 * The callout is responsible for logging because the remote format
922 * might differ from the local format and thus we don't know how much to
923 * log here. Note that init_fn must also set the buffer log item type
924 * correctly.
925 */
926 init_fn(tp, bp, ip, ifp);
927
928 /* account for the change in fork size */
929 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
930 xfs_bmap_local_to_extents_empty(ip, whichfork);
931 flags |= XFS_ILOG_CORE;
932
933 rec.br_startoff = 0;
934 rec.br_startblock = args.fsbno;
935 rec.br_blockcount = 1;
936 rec.br_state = XFS_EXT_NORM;
937 xfs_iext_insert(ip, 0, 1, &rec, 0);
938
939 trace_xfs_bmap_post_update(ip, 0,
940 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
941 _THIS_IP_);
942 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
943 ip->i_d.di_nblocks = 1;
944 xfs_trans_mod_dquot_byino(tp, ip,
945 XFS_TRANS_DQ_BCOUNT, 1L);
946 flags |= xfs_ilog_fext(whichfork);
947
948 done:
949 *logflagsp = flags;
950 return error;
951 }
952
953 /*
954 * Called from xfs_bmap_add_attrfork to handle btree format files.
955 */
956 STATIC int /* error */
xfs_bmap_add_attrfork_btree(xfs_trans_t * tp,xfs_inode_t * ip,xfs_fsblock_t * firstblock,struct xfs_defer_ops * dfops,int * flags)957 xfs_bmap_add_attrfork_btree(
958 xfs_trans_t *tp, /* transaction pointer */
959 xfs_inode_t *ip, /* incore inode pointer */
960 xfs_fsblock_t *firstblock, /* first block allocated */
961 struct xfs_defer_ops *dfops, /* blocks to free at commit */
962 int *flags) /* inode logging flags */
963 {
964 xfs_btree_cur_t *cur; /* btree cursor */
965 int error; /* error return value */
966 xfs_mount_t *mp; /* file system mount struct */
967 int stat; /* newroot status */
968
969 mp = ip->i_mount;
970 if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
971 *flags |= XFS_ILOG_DBROOT;
972 else {
973 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
974 cur->bc_private.b.dfops = dfops;
975 cur->bc_private.b.firstblock = *firstblock;
976 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
977 goto error0;
978 /* must be at least one entry */
979 XFS_WANT_CORRUPTED_GOTO(mp, stat == 1, error0);
980 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
981 goto error0;
982 if (stat == 0) {
983 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
984 return -ENOSPC;
985 }
986 *firstblock = cur->bc_private.b.firstblock;
987 cur->bc_private.b.allocated = 0;
988 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
989 }
990 return 0;
991 error0:
992 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
993 return error;
994 }
995
996 /*
997 * Called from xfs_bmap_add_attrfork to handle extents format files.
998 */
999 STATIC int /* error */
xfs_bmap_add_attrfork_extents(xfs_trans_t * tp,xfs_inode_t * ip,xfs_fsblock_t * firstblock,struct xfs_defer_ops * dfops,int * flags)1000 xfs_bmap_add_attrfork_extents(
1001 xfs_trans_t *tp, /* transaction pointer */
1002 xfs_inode_t *ip, /* incore inode pointer */
1003 xfs_fsblock_t *firstblock, /* first block allocated */
1004 struct xfs_defer_ops *dfops, /* blocks to free at commit */
1005 int *flags) /* inode logging flags */
1006 {
1007 xfs_btree_cur_t *cur; /* bmap btree cursor */
1008 int error; /* error return value */
1009
1010 if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
1011 return 0;
1012 cur = NULL;
1013 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, dfops, &cur, 0,
1014 flags, XFS_DATA_FORK);
1015 if (cur) {
1016 cur->bc_private.b.allocated = 0;
1017 xfs_btree_del_cursor(cur,
1018 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
1019 }
1020 return error;
1021 }
1022
1023 /*
1024 * Called from xfs_bmap_add_attrfork to handle local format files. Each
1025 * different data fork content type needs a different callout to do the
1026 * conversion. Some are basic and only require special block initialisation
1027 * callouts for the data formating, others (directories) are so specialised they
1028 * handle everything themselves.
1029 *
1030 * XXX (dgc): investigate whether directory conversion can use the generic
1031 * formatting callout. It should be possible - it's just a very complex
1032 * formatter.
1033 */
1034 STATIC int /* error */
xfs_bmap_add_attrfork_local(xfs_trans_t * tp,xfs_inode_t * ip,xfs_fsblock_t * firstblock,struct xfs_defer_ops * dfops,int * flags)1035 xfs_bmap_add_attrfork_local(
1036 xfs_trans_t *tp, /* transaction pointer */
1037 xfs_inode_t *ip, /* incore inode pointer */
1038 xfs_fsblock_t *firstblock, /* first block allocated */
1039 struct xfs_defer_ops *dfops, /* blocks to free at commit */
1040 int *flags) /* inode logging flags */
1041 {
1042 xfs_da_args_t dargs; /* args for dir/attr code */
1043
1044 if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
1045 return 0;
1046
1047 if (S_ISDIR(VFS_I(ip)->i_mode)) {
1048 memset(&dargs, 0, sizeof(dargs));
1049 dargs.geo = ip->i_mount->m_dir_geo;
1050 dargs.dp = ip;
1051 dargs.firstblock = firstblock;
1052 dargs.dfops = dfops;
1053 dargs.total = dargs.geo->fsbcount;
1054 dargs.whichfork = XFS_DATA_FORK;
1055 dargs.trans = tp;
1056 return xfs_dir2_sf_to_block(&dargs);
1057 }
1058
1059 if (S_ISLNK(VFS_I(ip)->i_mode))
1060 return xfs_bmap_local_to_extents(tp, ip, firstblock, 1,
1061 flags, XFS_DATA_FORK,
1062 xfs_symlink_local_to_remote);
1063
1064 /* should only be called for types that support local format data */
1065 ASSERT(0);
1066 return -EFSCORRUPTED;
1067 }
1068
1069 /*
1070 * Convert inode from non-attributed to attributed.
1071 * Must not be in a transaction, ip must not be locked.
1072 */
1073 int /* error code */
xfs_bmap_add_attrfork(xfs_inode_t * ip,int size,int rsvd)1074 xfs_bmap_add_attrfork(
1075 xfs_inode_t *ip, /* incore inode pointer */
1076 int size, /* space new attribute needs */
1077 int rsvd) /* xact may use reserved blks */
1078 {
1079 xfs_fsblock_t firstblock; /* 1st block/ag allocated */
1080 struct xfs_defer_ops dfops; /* freed extent records */
1081 xfs_mount_t *mp; /* mount structure */
1082 xfs_trans_t *tp; /* transaction pointer */
1083 int blks; /* space reservation */
1084 int version = 1; /* superblock attr version */
1085 int logflags; /* logging flags */
1086 int error; /* error return value */
1087
1088 ASSERT(XFS_IFORK_Q(ip) == 0);
1089
1090 mp = ip->i_mount;
1091 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1092
1093 blks = XFS_ADDAFORK_SPACE_RES(mp);
1094
1095 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_addafork, blks, 0,
1096 rsvd ? XFS_TRANS_RESERVE : 0, &tp);
1097 if (error)
1098 return error;
1099
1100 xfs_ilock(ip, XFS_ILOCK_EXCL);
1101 error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
1102 XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
1103 XFS_QMOPT_RES_REGBLKS);
1104 if (error)
1105 goto trans_cancel;
1106 if (XFS_IFORK_Q(ip))
1107 goto trans_cancel;
1108 if (ip->i_d.di_anextents != 0) {
1109 error = -EFSCORRUPTED;
1110 goto trans_cancel;
1111 }
1112 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
1113 /*
1114 * For inodes coming from pre-6.2 filesystems.
1115 */
1116 ASSERT(ip->i_d.di_aformat == 0);
1117 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1118 }
1119
1120 xfs_trans_ijoin(tp, ip, 0);
1121 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1122
1123 switch (ip->i_d.di_format) {
1124 case XFS_DINODE_FMT_DEV:
1125 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1126 break;
1127 case XFS_DINODE_FMT_UUID:
1128 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
1129 break;
1130 case XFS_DINODE_FMT_LOCAL:
1131 case XFS_DINODE_FMT_EXTENTS:
1132 case XFS_DINODE_FMT_BTREE:
1133 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1134 if (!ip->i_d.di_forkoff)
1135 ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1136 else if (mp->m_flags & XFS_MOUNT_ATTR2)
1137 version = 2;
1138 break;
1139 default:
1140 ASSERT(0);
1141 error = -EINVAL;
1142 goto trans_cancel;
1143 }
1144
1145 ASSERT(ip->i_afp == NULL);
1146 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
1147 ip->i_afp->if_flags = XFS_IFEXTENTS;
1148 logflags = 0;
1149 xfs_defer_init(&dfops, &firstblock);
1150 switch (ip->i_d.di_format) {
1151 case XFS_DINODE_FMT_LOCAL:
1152 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &dfops,
1153 &logflags);
1154 break;
1155 case XFS_DINODE_FMT_EXTENTS:
1156 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
1157 &dfops, &logflags);
1158 break;
1159 case XFS_DINODE_FMT_BTREE:
1160 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &dfops,
1161 &logflags);
1162 break;
1163 default:
1164 error = 0;
1165 break;
1166 }
1167 if (logflags)
1168 xfs_trans_log_inode(tp, ip, logflags);
1169 if (error)
1170 goto bmap_cancel;
1171 if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1172 (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
1173 bool log_sb = false;
1174
1175 spin_lock(&mp->m_sb_lock);
1176 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1177 xfs_sb_version_addattr(&mp->m_sb);
1178 log_sb = true;
1179 }
1180 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1181 xfs_sb_version_addattr2(&mp->m_sb);
1182 log_sb = true;
1183 }
1184 spin_unlock(&mp->m_sb_lock);
1185 if (log_sb)
1186 xfs_log_sb(tp);
1187 }
1188
1189 error = xfs_defer_finish(&tp, &dfops);
1190 if (error)
1191 goto bmap_cancel;
1192 error = xfs_trans_commit(tp);
1193 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1194 return error;
1195
1196 bmap_cancel:
1197 xfs_defer_cancel(&dfops);
1198 trans_cancel:
1199 xfs_trans_cancel(tp);
1200 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1201 return error;
1202 }
1203
1204 /*
1205 * Internal and external extent tree search functions.
1206 */
1207
1208 /*
1209 * Read in the extents to if_extents.
1210 * All inode fields are set up by caller, we just traverse the btree
1211 * and copy the records in. If the file system cannot contain unwritten
1212 * extents, the records are checked for no "state" flags.
1213 */
1214 int /* error */
xfs_bmap_read_extents(xfs_trans_t * tp,xfs_inode_t * ip,int whichfork)1215 xfs_bmap_read_extents(
1216 xfs_trans_t *tp, /* transaction pointer */
1217 xfs_inode_t *ip, /* incore inode */
1218 int whichfork) /* data or attr fork */
1219 {
1220 struct xfs_btree_block *block; /* current btree block */
1221 xfs_fsblock_t bno; /* block # of "block" */
1222 xfs_buf_t *bp; /* buffer for "block" */
1223 int error; /* error return value */
1224 xfs_extnum_t i, j; /* index into the extents list */
1225 xfs_ifork_t *ifp; /* fork structure */
1226 int level; /* btree level, for checking */
1227 xfs_mount_t *mp; /* file system mount structure */
1228 __be64 *pp; /* pointer to block address */
1229 /* REFERENCED */
1230 xfs_extnum_t room; /* number of entries there's room for */
1231
1232 mp = ip->i_mount;
1233 ifp = XFS_IFORK_PTR(ip, whichfork);
1234 block = ifp->if_broot;
1235 /*
1236 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1237 */
1238 level = be16_to_cpu(block->bb_level);
1239 ASSERT(level > 0);
1240 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
1241 bno = be64_to_cpu(*pp);
1242
1243 /*
1244 * Go down the tree until leaf level is reached, following the first
1245 * pointer (leftmost) at each level.
1246 */
1247 while (level-- > 0) {
1248 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1249 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1250 if (error)
1251 return error;
1252 block = XFS_BUF_TO_BLOCK(bp);
1253 if (level == 0)
1254 break;
1255 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
1256 bno = be64_to_cpu(*pp);
1257 XFS_WANT_CORRUPTED_GOTO(mp,
1258 XFS_FSB_SANITY_CHECK(mp, bno), error0);
1259 xfs_trans_brelse(tp, bp);
1260 }
1261 /*
1262 * Here with bp and block set to the leftmost leaf node in the tree.
1263 */
1264 room = xfs_iext_count(ifp);
1265 i = 0;
1266 /*
1267 * Loop over all leaf nodes. Copy information to the extent records.
1268 */
1269 for (;;) {
1270 xfs_bmbt_rec_t *frp;
1271 xfs_fsblock_t nextbno;
1272 xfs_extnum_t num_recs;
1273
1274 num_recs = xfs_btree_get_numrecs(block);
1275 if (unlikely(i + num_recs > room)) {
1276 ASSERT(i + num_recs <= room);
1277 xfs_warn(ip->i_mount,
1278 "corrupt dinode %Lu, (btree extents).",
1279 (unsigned long long) ip->i_ino);
1280 XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
1281 XFS_ERRLEVEL_LOW, ip->i_mount, block);
1282 goto error0;
1283 }
1284 /*
1285 * Read-ahead the next leaf block, if any.
1286 */
1287 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
1288 if (nextbno != NULLFSBLOCK)
1289 xfs_btree_reada_bufl(mp, nextbno, 1,
1290 &xfs_bmbt_buf_ops);
1291 /*
1292 * Copy records into the extent records.
1293 */
1294 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1295 for (j = 0; j < num_recs; j++, i++, frp++) {
1296 xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
1297 trp->l0 = be64_to_cpu(frp->l0);
1298 trp->l1 = be64_to_cpu(frp->l1);
1299 if (!xfs_bmbt_validate_extent(mp, whichfork, trp)) {
1300 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
1301 XFS_ERRLEVEL_LOW, mp);
1302 goto error0;
1303 }
1304 }
1305 xfs_trans_brelse(tp, bp);
1306 bno = nextbno;
1307 /*
1308 * If we've reached the end, stop.
1309 */
1310 if (bno == NULLFSBLOCK)
1311 break;
1312 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1313 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1314 if (error)
1315 return error;
1316 block = XFS_BUF_TO_BLOCK(bp);
1317 }
1318 if (i != XFS_IFORK_NEXTENTS(ip, whichfork))
1319 return -EFSCORRUPTED;
1320 ASSERT(i == xfs_iext_count(ifp));
1321 XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
1322 return 0;
1323 error0:
1324 xfs_trans_brelse(tp, bp);
1325 return -EFSCORRUPTED;
1326 }
1327
1328 /*
1329 * Returns the file-relative block number of the first unused block(s)
1330 * in the file with at least "len" logically contiguous blocks free.
1331 * This is the lowest-address hole if the file has holes, else the first block
1332 * past the end of file.
1333 * Return 0 if the file is currently local (in-inode).
1334 */
1335 int /* error */
xfs_bmap_first_unused(xfs_trans_t * tp,xfs_inode_t * ip,xfs_extlen_t len,xfs_fileoff_t * first_unused,int whichfork)1336 xfs_bmap_first_unused(
1337 xfs_trans_t *tp, /* transaction pointer */
1338 xfs_inode_t *ip, /* incore inode */
1339 xfs_extlen_t len, /* size of hole to find */
1340 xfs_fileoff_t *first_unused, /* unused block */
1341 int whichfork) /* data or attr fork */
1342 {
1343 int error; /* error return value */
1344 int idx; /* extent record index */
1345 xfs_ifork_t *ifp; /* inode fork pointer */
1346 xfs_fileoff_t lastaddr; /* last block number seen */
1347 xfs_fileoff_t lowest; /* lowest useful block */
1348 xfs_fileoff_t max; /* starting useful block */
1349 xfs_extnum_t nextents; /* number of extent entries */
1350
1351 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
1352 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
1353 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1354 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1355 *first_unused = 0;
1356 return 0;
1357 }
1358 ifp = XFS_IFORK_PTR(ip, whichfork);
1359 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1360 (error = xfs_iread_extents(tp, ip, whichfork)))
1361 return error;
1362 lowest = *first_unused;
1363 nextents = xfs_iext_count(ifp);
1364 for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
1365 struct xfs_bmbt_irec got;
1366
1367 xfs_iext_get_extent(ifp, idx, &got);
1368
1369 /*
1370 * See if the hole before this extent will work.
1371 */
1372 if (got.br_startoff >= lowest + len &&
1373 got.br_startoff - max >= len) {
1374 *first_unused = max;
1375 return 0;
1376 }
1377 lastaddr = got.br_startoff + got.br_blockcount;
1378 max = XFS_FILEOFF_MAX(lastaddr, lowest);
1379 }
1380 *first_unused = max;
1381 return 0;
1382 }
1383
1384 /*
1385 * Returns the file-relative block number of the last block - 1 before
1386 * last_block (input value) in the file.
1387 * This is not based on i_size, it is based on the extent records.
1388 * Returns 0 for local files, as they do not have extent records.
1389 */
1390 int /* error */
xfs_bmap_last_before(struct xfs_trans * tp,struct xfs_inode * ip,xfs_fileoff_t * last_block,int whichfork)1391 xfs_bmap_last_before(
1392 struct xfs_trans *tp, /* transaction pointer */
1393 struct xfs_inode *ip, /* incore inode */
1394 xfs_fileoff_t *last_block, /* last block */
1395 int whichfork) /* data or attr fork */
1396 {
1397 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
1398 struct xfs_bmbt_irec got;
1399 xfs_extnum_t idx;
1400 int error;
1401
1402 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
1403 case XFS_DINODE_FMT_LOCAL:
1404 *last_block = 0;
1405 return 0;
1406 case XFS_DINODE_FMT_BTREE:
1407 case XFS_DINODE_FMT_EXTENTS:
1408 break;
1409 default:
1410 return -EIO;
1411 }
1412
1413 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1414 error = xfs_iread_extents(tp, ip, whichfork);
1415 if (error)
1416 return error;
1417 }
1418
1419 if (xfs_iext_lookup_extent(ip, ifp, *last_block - 1, &idx, &got)) {
1420 if (got.br_startoff <= *last_block - 1)
1421 return 0;
1422 }
1423
1424 if (xfs_iext_get_extent(ifp, idx - 1, &got)) {
1425 *last_block = got.br_startoff + got.br_blockcount;
1426 return 0;
1427 }
1428
1429 *last_block = 0;
1430 return 0;
1431 }
1432
1433 int
xfs_bmap_last_extent(struct xfs_trans * tp,struct xfs_inode * ip,int whichfork,struct xfs_bmbt_irec * rec,int * is_empty)1434 xfs_bmap_last_extent(
1435 struct xfs_trans *tp,
1436 struct xfs_inode *ip,
1437 int whichfork,
1438 struct xfs_bmbt_irec *rec,
1439 int *is_empty)
1440 {
1441 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
1442 int error;
1443 int nextents;
1444
1445 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1446 error = xfs_iread_extents(tp, ip, whichfork);
1447 if (error)
1448 return error;
1449 }
1450
1451 nextents = xfs_iext_count(ifp);
1452 if (nextents == 0) {
1453 *is_empty = 1;
1454 return 0;
1455 }
1456
1457 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
1458 *is_empty = 0;
1459 return 0;
1460 }
1461
1462 /*
1463 * Check the last inode extent to determine whether this allocation will result
1464 * in blocks being allocated at the end of the file. When we allocate new data
1465 * blocks at the end of the file which do not start at the previous data block,
1466 * we will try to align the new blocks at stripe unit boundaries.
1467 *
1468 * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1469 * at, or past the EOF.
1470 */
1471 STATIC int
xfs_bmap_isaeof(struct xfs_bmalloca * bma,int whichfork)1472 xfs_bmap_isaeof(
1473 struct xfs_bmalloca *bma,
1474 int whichfork)
1475 {
1476 struct xfs_bmbt_irec rec;
1477 int is_empty;
1478 int error;
1479
1480 bma->aeof = false;
1481 error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1482 &is_empty);
1483 if (error)
1484 return error;
1485
1486 if (is_empty) {
1487 bma->aeof = true;
1488 return 0;
1489 }
1490
1491 /*
1492 * Check if we are allocation or past the last extent, or at least into
1493 * the last delayed allocated extent.
1494 */
1495 bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1496 (bma->offset >= rec.br_startoff &&
1497 isnullstartblock(rec.br_startblock));
1498 return 0;
1499 }
1500
1501 /*
1502 * Returns the file-relative block number of the first block past eof in
1503 * the file. This is not based on i_size, it is based on the extent records.
1504 * Returns 0 for local files, as they do not have extent records.
1505 */
1506 int
xfs_bmap_last_offset(struct xfs_inode * ip,xfs_fileoff_t * last_block,int whichfork)1507 xfs_bmap_last_offset(
1508 struct xfs_inode *ip,
1509 xfs_fileoff_t *last_block,
1510 int whichfork)
1511 {
1512 struct xfs_bmbt_irec rec;
1513 int is_empty;
1514 int error;
1515
1516 *last_block = 0;
1517
1518 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
1519 return 0;
1520
1521 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1522 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1523 return -EIO;
1524
1525 error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1526 if (error || is_empty)
1527 return error;
1528
1529 *last_block = rec.br_startoff + rec.br_blockcount;
1530 return 0;
1531 }
1532
1533 /*
1534 * Returns whether the selected fork of the inode has exactly one
1535 * block or not. For the data fork we check this matches di_size,
1536 * implying the file's range is 0..bsize-1.
1537 */
1538 int /* 1=>1 block, 0=>otherwise */
xfs_bmap_one_block(xfs_inode_t * ip,int whichfork)1539 xfs_bmap_one_block(
1540 xfs_inode_t *ip, /* incore inode */
1541 int whichfork) /* data or attr fork */
1542 {
1543 xfs_bmbt_rec_host_t *ep; /* ptr to fork's extent */
1544 xfs_ifork_t *ifp; /* inode fork pointer */
1545 int rval; /* return value */
1546 xfs_bmbt_irec_t s; /* internal version of extent */
1547
1548 #ifndef DEBUG
1549 if (whichfork == XFS_DATA_FORK)
1550 return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
1551 #endif /* !DEBUG */
1552 if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
1553 return 0;
1554 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1555 return 0;
1556 ifp = XFS_IFORK_PTR(ip, whichfork);
1557 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1558 ep = xfs_iext_get_ext(ifp, 0);
1559 xfs_bmbt_get_all(ep, &s);
1560 rval = s.br_startoff == 0 && s.br_blockcount == 1;
1561 if (rval && whichfork == XFS_DATA_FORK)
1562 ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
1563 return rval;
1564 }
1565
1566 /*
1567 * Extent tree manipulation functions used during allocation.
1568 */
1569
1570 /*
1571 * Convert a delayed allocation to a real allocation.
1572 */
1573 STATIC int /* error */
xfs_bmap_add_extent_delay_real(struct xfs_bmalloca * bma,int whichfork)1574 xfs_bmap_add_extent_delay_real(
1575 struct xfs_bmalloca *bma,
1576 int whichfork)
1577 {
1578 struct xfs_bmbt_irec *new = &bma->got;
1579 int diff; /* temp value */
1580 xfs_bmbt_rec_host_t *ep; /* extent entry for idx */
1581 int error; /* error return value */
1582 int i; /* temp state */
1583 xfs_ifork_t *ifp; /* inode fork pointer */
1584 xfs_fileoff_t new_endoff; /* end offset of new entry */
1585 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
1586 /* left is 0, right is 1, prev is 2 */
1587 int rval=0; /* return value (logging flags) */
1588 int state = 0;/* state bits, accessed thru macros */
1589 xfs_filblks_t da_new; /* new count del alloc blocks used */
1590 xfs_filblks_t da_old; /* old count del alloc blocks used */
1591 xfs_filblks_t temp=0; /* value for da_new calculations */
1592 xfs_filblks_t temp2=0;/* value for da_new calculations */
1593 int tmp_rval; /* partial logging flags */
1594 struct xfs_mount *mp;
1595 xfs_extnum_t *nextents;
1596
1597 mp = bma->ip->i_mount;
1598 ifp = XFS_IFORK_PTR(bma->ip, whichfork);
1599 ASSERT(whichfork != XFS_ATTR_FORK);
1600 nextents = (whichfork == XFS_COW_FORK ? &bma->ip->i_cnextents :
1601 &bma->ip->i_d.di_nextents);
1602
1603 ASSERT(bma->idx >= 0);
1604 ASSERT(bma->idx <= xfs_iext_count(ifp));
1605 ASSERT(!isnullstartblock(new->br_startblock));
1606 ASSERT(!bma->cur ||
1607 (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1608
1609 XFS_STATS_INC(mp, xs_add_exlist);
1610
1611 #define LEFT r[0]
1612 #define RIGHT r[1]
1613 #define PREV r[2]
1614
1615 if (whichfork == XFS_COW_FORK)
1616 state |= BMAP_COWFORK;
1617
1618 /*
1619 * Set up a bunch of variables to make the tests simpler.
1620 */
1621 ep = xfs_iext_get_ext(ifp, bma->idx);
1622 xfs_bmbt_get_all(ep, &PREV);
1623 new_endoff = new->br_startoff + new->br_blockcount;
1624 ASSERT(PREV.br_startoff <= new->br_startoff);
1625 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1626
1627 da_old = startblockval(PREV.br_startblock);
1628 da_new = 0;
1629
1630 /*
1631 * Set flags determining what part of the previous delayed allocation
1632 * extent is being replaced by a real allocation.
1633 */
1634 if (PREV.br_startoff == new->br_startoff)
1635 state |= BMAP_LEFT_FILLING;
1636 if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1637 state |= BMAP_RIGHT_FILLING;
1638
1639 /*
1640 * Check and set flags if this segment has a left neighbor.
1641 * Don't set contiguous if the combined extent would be too large.
1642 */
1643 if (bma->idx > 0) {
1644 state |= BMAP_LEFT_VALID;
1645 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
1646
1647 if (isnullstartblock(LEFT.br_startblock))
1648 state |= BMAP_LEFT_DELAY;
1649 }
1650
1651 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1652 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1653 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1654 LEFT.br_state == new->br_state &&
1655 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1656 state |= BMAP_LEFT_CONTIG;
1657
1658 /*
1659 * Check and set flags if this segment has a right neighbor.
1660 * Don't set contiguous if the combined extent would be too large.
1661 * Also check for all-three-contiguous being too large.
1662 */
1663 if (bma->idx < xfs_iext_count(ifp) - 1) {
1664 state |= BMAP_RIGHT_VALID;
1665 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
1666
1667 if (isnullstartblock(RIGHT.br_startblock))
1668 state |= BMAP_RIGHT_DELAY;
1669 }
1670
1671 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1672 new_endoff == RIGHT.br_startoff &&
1673 new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1674 new->br_state == RIGHT.br_state &&
1675 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1676 ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1677 BMAP_RIGHT_FILLING)) !=
1678 (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1679 BMAP_RIGHT_FILLING) ||
1680 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1681 <= MAXEXTLEN))
1682 state |= BMAP_RIGHT_CONTIG;
1683
1684 error = 0;
1685 /*
1686 * Switch out based on the FILLING and CONTIG state bits.
1687 */
1688 switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1689 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1690 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1691 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1692 /*
1693 * Filling in all of a previously delayed allocation extent.
1694 * The left and right neighbors are both contiguous with new.
1695 */
1696 bma->idx--;
1697 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1698 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1699 LEFT.br_blockcount + PREV.br_blockcount +
1700 RIGHT.br_blockcount);
1701 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1702
1703 xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
1704 (*nextents)--;
1705 if (bma->cur == NULL)
1706 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1707 else {
1708 rval = XFS_ILOG_CORE;
1709 error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1710 RIGHT.br_startblock,
1711 RIGHT.br_blockcount, &i);
1712 if (error)
1713 goto done;
1714 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1715 error = xfs_btree_delete(bma->cur, &i);
1716 if (error)
1717 goto done;
1718 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1719 error = xfs_btree_decrement(bma->cur, 0, &i);
1720 if (error)
1721 goto done;
1722 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1723 error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1724 LEFT.br_startblock,
1725 LEFT.br_blockcount +
1726 PREV.br_blockcount +
1727 RIGHT.br_blockcount, LEFT.br_state);
1728 if (error)
1729 goto done;
1730 }
1731 break;
1732
1733 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1734 /*
1735 * Filling in all of a previously delayed allocation extent.
1736 * The left neighbor is contiguous, the right is not.
1737 */
1738 bma->idx--;
1739
1740 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1741 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1742 LEFT.br_blockcount + PREV.br_blockcount);
1743 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1744
1745 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1746 if (bma->cur == NULL)
1747 rval = XFS_ILOG_DEXT;
1748 else {
1749 rval = 0;
1750 error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1751 LEFT.br_startblock, LEFT.br_blockcount,
1752 &i);
1753 if (error)
1754 goto done;
1755 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1756 error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1757 LEFT.br_startblock,
1758 LEFT.br_blockcount +
1759 PREV.br_blockcount, LEFT.br_state);
1760 if (error)
1761 goto done;
1762 }
1763 break;
1764
1765 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1766 /*
1767 * Filling in all of a previously delayed allocation extent.
1768 * The right neighbor is contiguous, the left is not.
1769 */
1770 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1771 xfs_bmbt_set_startblock(ep, new->br_startblock);
1772 xfs_bmbt_set_blockcount(ep,
1773 PREV.br_blockcount + RIGHT.br_blockcount);
1774 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1775
1776 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1777 if (bma->cur == NULL)
1778 rval = XFS_ILOG_DEXT;
1779 else {
1780 rval = 0;
1781 error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1782 RIGHT.br_startblock,
1783 RIGHT.br_blockcount, &i);
1784 if (error)
1785 goto done;
1786 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1787 error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
1788 new->br_startblock,
1789 PREV.br_blockcount +
1790 RIGHT.br_blockcount, PREV.br_state);
1791 if (error)
1792 goto done;
1793 }
1794 break;
1795
1796 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1797 /*
1798 * Filling in all of a previously delayed allocation extent.
1799 * Neither the left nor right neighbors are contiguous with
1800 * the new one.
1801 */
1802 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1803 xfs_bmbt_set_startblock(ep, new->br_startblock);
1804 xfs_bmbt_set_state(ep, new->br_state);
1805 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1806
1807 (*nextents)++;
1808 if (bma->cur == NULL)
1809 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1810 else {
1811 rval = XFS_ILOG_CORE;
1812 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1813 new->br_startblock, new->br_blockcount,
1814 &i);
1815 if (error)
1816 goto done;
1817 XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1818 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1819 error = xfs_btree_insert(bma->cur, &i);
1820 if (error)
1821 goto done;
1822 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1823 }
1824 break;
1825
1826 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1827 /*
1828 * Filling in the first part of a previous delayed allocation.
1829 * The left neighbor is contiguous.
1830 */
1831 trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1832 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
1833 LEFT.br_blockcount + new->br_blockcount);
1834 xfs_bmbt_set_startoff(ep,
1835 PREV.br_startoff + new->br_blockcount);
1836 trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1837
1838 temp = PREV.br_blockcount - new->br_blockcount;
1839 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1840 xfs_bmbt_set_blockcount(ep, temp);
1841 if (bma->cur == NULL)
1842 rval = XFS_ILOG_DEXT;
1843 else {
1844 rval = 0;
1845 error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1846 LEFT.br_startblock, LEFT.br_blockcount,
1847 &i);
1848 if (error)
1849 goto done;
1850 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1851 error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1852 LEFT.br_startblock,
1853 LEFT.br_blockcount +
1854 new->br_blockcount,
1855 LEFT.br_state);
1856 if (error)
1857 goto done;
1858 }
1859 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1860 startblockval(PREV.br_startblock));
1861 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
1862 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1863
1864 bma->idx--;
1865 break;
1866
1867 case BMAP_LEFT_FILLING:
1868 /*
1869 * Filling in the first part of a previous delayed allocation.
1870 * The left neighbor is not contiguous.
1871 */
1872 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1873 xfs_bmbt_set_startoff(ep, new_endoff);
1874 temp = PREV.br_blockcount - new->br_blockcount;
1875 xfs_bmbt_set_blockcount(ep, temp);
1876 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
1877 (*nextents)++;
1878 if (bma->cur == NULL)
1879 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1880 else {
1881 rval = XFS_ILOG_CORE;
1882 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1883 new->br_startblock, new->br_blockcount,
1884 &i);
1885 if (error)
1886 goto done;
1887 XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1888 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1889 error = xfs_btree_insert(bma->cur, &i);
1890 if (error)
1891 goto done;
1892 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1893 }
1894
1895 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1896 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1897 bma->firstblock, bma->dfops,
1898 &bma->cur, 1, &tmp_rval, whichfork);
1899 rval |= tmp_rval;
1900 if (error)
1901 goto done;
1902 }
1903 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1904 startblockval(PREV.br_startblock) -
1905 (bma->cur ? bma->cur->bc_private.b.allocated : 0));
1906 ep = xfs_iext_get_ext(ifp, bma->idx + 1);
1907 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
1908 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
1909 break;
1910
1911 case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1912 /*
1913 * Filling in the last part of a previous delayed allocation.
1914 * The right neighbor is contiguous with the new allocation.
1915 */
1916 temp = PREV.br_blockcount - new->br_blockcount;
1917 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
1918 xfs_bmbt_set_blockcount(ep, temp);
1919 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
1920 new->br_startoff, new->br_startblock,
1921 new->br_blockcount + RIGHT.br_blockcount,
1922 RIGHT.br_state);
1923 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
1924 if (bma->cur == NULL)
1925 rval = XFS_ILOG_DEXT;
1926 else {
1927 rval = 0;
1928 error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1929 RIGHT.br_startblock,
1930 RIGHT.br_blockcount, &i);
1931 if (error)
1932 goto done;
1933 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1934 error = xfs_bmbt_update(bma->cur, new->br_startoff,
1935 new->br_startblock,
1936 new->br_blockcount +
1937 RIGHT.br_blockcount,
1938 RIGHT.br_state);
1939 if (error)
1940 goto done;
1941 }
1942
1943 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1944 startblockval(PREV.br_startblock));
1945 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1946 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
1947 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1948
1949 bma->idx++;
1950 break;
1951
1952 case BMAP_RIGHT_FILLING:
1953 /*
1954 * Filling in the last part of a previous delayed allocation.
1955 * The right neighbor is not contiguous.
1956 */
1957 temp = PREV.br_blockcount - new->br_blockcount;
1958 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1959 xfs_bmbt_set_blockcount(ep, temp);
1960 xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
1961 (*nextents)++;
1962 if (bma->cur == NULL)
1963 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1964 else {
1965 rval = XFS_ILOG_CORE;
1966 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1967 new->br_startblock, new->br_blockcount,
1968 &i);
1969 if (error)
1970 goto done;
1971 XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1972 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1973 error = xfs_btree_insert(bma->cur, &i);
1974 if (error)
1975 goto done;
1976 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1977 }
1978
1979 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1980 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1981 bma->firstblock, bma->dfops, &bma->cur, 1,
1982 &tmp_rval, whichfork);
1983 rval |= tmp_rval;
1984 if (error)
1985 goto done;
1986 }
1987 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1988 startblockval(PREV.br_startblock) -
1989 (bma->cur ? bma->cur->bc_private.b.allocated : 0));
1990 ep = xfs_iext_get_ext(ifp, bma->idx);
1991 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
1992 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1993
1994 bma->idx++;
1995 break;
1996
1997 case 0:
1998 /*
1999 * Filling in the middle part of a previous delayed allocation.
2000 * Contiguity is impossible here.
2001 * This case is avoided almost all the time.
2002 *
2003 * We start with a delayed allocation:
2004 *
2005 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
2006 * PREV @ idx
2007 *
2008 * and we are allocating:
2009 * +rrrrrrrrrrrrrrrrr+
2010 * new
2011 *
2012 * and we set it up for insertion as:
2013 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
2014 * new
2015 * PREV @ idx LEFT RIGHT
2016 * inserted at idx + 1
2017 */
2018 temp = new->br_startoff - PREV.br_startoff;
2019 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
2020 trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
2021 xfs_bmbt_set_blockcount(ep, temp); /* truncate PREV */
2022 LEFT = *new;
2023 RIGHT.br_state = PREV.br_state;
2024 RIGHT.br_startblock = nullstartblock(
2025 (int)xfs_bmap_worst_indlen(bma->ip, temp2));
2026 RIGHT.br_startoff = new_endoff;
2027 RIGHT.br_blockcount = temp2;
2028 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
2029 xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
2030 (*nextents)++;
2031 if (bma->cur == NULL)
2032 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2033 else {
2034 rval = XFS_ILOG_CORE;
2035 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2036 new->br_startblock, new->br_blockcount,
2037 &i);
2038 if (error)
2039 goto done;
2040 XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2041 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2042 error = xfs_btree_insert(bma->cur, &i);
2043 if (error)
2044 goto done;
2045 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2046 }
2047
2048 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2049 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2050 bma->firstblock, bma->dfops, &bma->cur,
2051 1, &tmp_rval, whichfork);
2052 rval |= tmp_rval;
2053 if (error)
2054 goto done;
2055 }
2056 temp = xfs_bmap_worst_indlen(bma->ip, temp);
2057 temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
2058 diff = (int)(temp + temp2 -
2059 (startblockval(PREV.br_startblock) -
2060 (bma->cur ?
2061 bma->cur->bc_private.b.allocated : 0)));
2062 if (diff > 0) {
2063 error = xfs_mod_fdblocks(bma->ip->i_mount,
2064 -((int64_t)diff), false);
2065 ASSERT(!error);
2066 if (error)
2067 goto done;
2068 }
2069
2070 ep = xfs_iext_get_ext(ifp, bma->idx);
2071 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2072 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2073 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2074 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
2075 nullstartblock((int)temp2));
2076 trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2077
2078 bma->idx++;
2079 da_new = temp + temp2;
2080 break;
2081
2082 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2083 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2084 case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2085 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2086 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2087 case BMAP_LEFT_CONTIG:
2088 case BMAP_RIGHT_CONTIG:
2089 /*
2090 * These cases are all impossible.
2091 */
2092 ASSERT(0);
2093 }
2094
2095 /* add reverse mapping */
2096 error = xfs_rmap_map_extent(mp, bma->dfops, bma->ip, whichfork, new);
2097 if (error)
2098 goto done;
2099
2100 /* convert to a btree if necessary */
2101 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2102 int tmp_logflags; /* partial log flag return val */
2103
2104 ASSERT(bma->cur == NULL);
2105 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2106 bma->firstblock, bma->dfops, &bma->cur,
2107 da_old > 0, &tmp_logflags, whichfork);
2108 bma->logflags |= tmp_logflags;
2109 if (error)
2110 goto done;
2111 }
2112
2113 /* adjust for changes in reserved delayed indirect blocks */
2114 if (da_old || da_new) {
2115 temp = da_new;
2116 if (bma->cur)
2117 temp += bma->cur->bc_private.b.allocated;
2118 if (temp < da_old)
2119 xfs_mod_fdblocks(bma->ip->i_mount,
2120 (int64_t)(da_old - temp), false);
2121 }
2122
2123 /* clear out the allocated field, done with it now in any case. */
2124 if (bma->cur)
2125 bma->cur->bc_private.b.allocated = 0;
2126
2127 xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
2128 done:
2129 if (whichfork != XFS_COW_FORK)
2130 bma->logflags |= rval;
2131 return error;
2132 #undef LEFT
2133 #undef RIGHT
2134 #undef PREV
2135 }
2136
2137 /*
2138 * Convert an unwritten allocation to a real allocation or vice versa.
2139 */
2140 STATIC int /* error */
xfs_bmap_add_extent_unwritten_real(struct xfs_trans * tp,xfs_inode_t * ip,int whichfork,xfs_extnum_t * idx,xfs_btree_cur_t ** curp,xfs_bmbt_irec_t * new,xfs_fsblock_t * first,struct xfs_defer_ops * dfops,int * logflagsp)2141 xfs_bmap_add_extent_unwritten_real(
2142 struct xfs_trans *tp,
2143 xfs_inode_t *ip, /* incore inode pointer */
2144 int whichfork,
2145 xfs_extnum_t *idx, /* extent number to update/insert */
2146 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
2147 xfs_bmbt_irec_t *new, /* new data to add to file extents */
2148 xfs_fsblock_t *first, /* pointer to firstblock variable */
2149 struct xfs_defer_ops *dfops, /* list of extents to be freed */
2150 int *logflagsp) /* inode logging flags */
2151 {
2152 xfs_btree_cur_t *cur; /* btree cursor */
2153 xfs_bmbt_rec_host_t *ep; /* extent entry for idx */
2154 int error; /* error return value */
2155 int i; /* temp state */
2156 xfs_ifork_t *ifp; /* inode fork pointer */
2157 xfs_fileoff_t new_endoff; /* end offset of new entry */
2158 xfs_exntst_t newext; /* new extent state */
2159 xfs_exntst_t oldext; /* old extent state */
2160 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
2161 /* left is 0, right is 1, prev is 2 */
2162 int rval=0; /* return value (logging flags) */
2163 int state = 0;/* state bits, accessed thru macros */
2164 struct xfs_mount *mp = ip->i_mount;
2165
2166 *logflagsp = 0;
2167
2168 cur = *curp;
2169 ifp = XFS_IFORK_PTR(ip, whichfork);
2170 if (whichfork == XFS_COW_FORK)
2171 state |= BMAP_COWFORK;
2172
2173 ASSERT(*idx >= 0);
2174 ASSERT(*idx <= xfs_iext_count(ifp));
2175 ASSERT(!isnullstartblock(new->br_startblock));
2176
2177 XFS_STATS_INC(mp, xs_add_exlist);
2178
2179 #define LEFT r[0]
2180 #define RIGHT r[1]
2181 #define PREV r[2]
2182
2183 /*
2184 * Set up a bunch of variables to make the tests simpler.
2185 */
2186 error = 0;
2187 ep = xfs_iext_get_ext(ifp, *idx);
2188 xfs_bmbt_get_all(ep, &PREV);
2189 newext = new->br_state;
2190 oldext = (newext == XFS_EXT_UNWRITTEN) ?
2191 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
2192 ASSERT(PREV.br_state == oldext);
2193 new_endoff = new->br_startoff + new->br_blockcount;
2194 ASSERT(PREV.br_startoff <= new->br_startoff);
2195 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2196
2197 /*
2198 * Set flags determining what part of the previous oldext allocation
2199 * extent is being replaced by a newext allocation.
2200 */
2201 if (PREV.br_startoff == new->br_startoff)
2202 state |= BMAP_LEFT_FILLING;
2203 if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2204 state |= BMAP_RIGHT_FILLING;
2205
2206 /*
2207 * Check and set flags if this segment has a left neighbor.
2208 * Don't set contiguous if the combined extent would be too large.
2209 */
2210 if (*idx > 0) {
2211 state |= BMAP_LEFT_VALID;
2212 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
2213
2214 if (isnullstartblock(LEFT.br_startblock))
2215 state |= BMAP_LEFT_DELAY;
2216 }
2217
2218 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2219 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2220 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2221 LEFT.br_state == newext &&
2222 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2223 state |= BMAP_LEFT_CONTIG;
2224
2225 /*
2226 * Check and set flags if this segment has a right neighbor.
2227 * Don't set contiguous if the combined extent would be too large.
2228 * Also check for all-three-contiguous being too large.
2229 */
2230 if (*idx < xfs_iext_count(ifp) - 1) {
2231 state |= BMAP_RIGHT_VALID;
2232 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
2233 if (isnullstartblock(RIGHT.br_startblock))
2234 state |= BMAP_RIGHT_DELAY;
2235 }
2236
2237 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2238 new_endoff == RIGHT.br_startoff &&
2239 new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2240 newext == RIGHT.br_state &&
2241 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2242 ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2243 BMAP_RIGHT_FILLING)) !=
2244 (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2245 BMAP_RIGHT_FILLING) ||
2246 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2247 <= MAXEXTLEN))
2248 state |= BMAP_RIGHT_CONTIG;
2249
2250 /*
2251 * Switch out based on the FILLING and CONTIG state bits.
2252 */
2253 switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2254 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2255 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2256 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2257 /*
2258 * Setting all of a previous oldext extent to newext.
2259 * The left and right neighbors are both contiguous with new.
2260 */
2261 --*idx;
2262
2263 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2264 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2265 LEFT.br_blockcount + PREV.br_blockcount +
2266 RIGHT.br_blockcount);
2267 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2268
2269 xfs_iext_remove(ip, *idx + 1, 2, state);
2270 XFS_IFORK_NEXT_SET(ip, whichfork,
2271 XFS_IFORK_NEXTENTS(ip, whichfork) - 2);
2272 if (cur == NULL)
2273 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2274 else {
2275 rval = XFS_ILOG_CORE;
2276 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2277 RIGHT.br_startblock,
2278 RIGHT.br_blockcount, &i)))
2279 goto done;
2280 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2281 if ((error = xfs_btree_delete(cur, &i)))
2282 goto done;
2283 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2284 if ((error = xfs_btree_decrement(cur, 0, &i)))
2285 goto done;
2286 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2287 if ((error = xfs_btree_delete(cur, &i)))
2288 goto done;
2289 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2290 if ((error = xfs_btree_decrement(cur, 0, &i)))
2291 goto done;
2292 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2293 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2294 LEFT.br_startblock,
2295 LEFT.br_blockcount + PREV.br_blockcount +
2296 RIGHT.br_blockcount, LEFT.br_state)))
2297 goto done;
2298 }
2299 break;
2300
2301 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2302 /*
2303 * Setting all of a previous oldext extent to newext.
2304 * The left neighbor is contiguous, the right is not.
2305 */
2306 --*idx;
2307
2308 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2309 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2310 LEFT.br_blockcount + PREV.br_blockcount);
2311 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2312
2313 xfs_iext_remove(ip, *idx + 1, 1, state);
2314 XFS_IFORK_NEXT_SET(ip, whichfork,
2315 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2316 if (cur == NULL)
2317 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2318 else {
2319 rval = XFS_ILOG_CORE;
2320 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2321 PREV.br_startblock, PREV.br_blockcount,
2322 &i)))
2323 goto done;
2324 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2325 if ((error = xfs_btree_delete(cur, &i)))
2326 goto done;
2327 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2328 if ((error = xfs_btree_decrement(cur, 0, &i)))
2329 goto done;
2330 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2331 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2332 LEFT.br_startblock,
2333 LEFT.br_blockcount + PREV.br_blockcount,
2334 LEFT.br_state)))
2335 goto done;
2336 }
2337 break;
2338
2339 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2340 /*
2341 * Setting all of a previous oldext extent to newext.
2342 * The right neighbor is contiguous, the left is not.
2343 */
2344 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2345 xfs_bmbt_set_blockcount(ep,
2346 PREV.br_blockcount + RIGHT.br_blockcount);
2347 xfs_bmbt_set_state(ep, newext);
2348 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2349 xfs_iext_remove(ip, *idx + 1, 1, state);
2350 XFS_IFORK_NEXT_SET(ip, whichfork,
2351 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2352 if (cur == NULL)
2353 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2354 else {
2355 rval = XFS_ILOG_CORE;
2356 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2357 RIGHT.br_startblock,
2358 RIGHT.br_blockcount, &i)))
2359 goto done;
2360 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2361 if ((error = xfs_btree_delete(cur, &i)))
2362 goto done;
2363 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2364 if ((error = xfs_btree_decrement(cur, 0, &i)))
2365 goto done;
2366 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2367 if ((error = xfs_bmbt_update(cur, new->br_startoff,
2368 new->br_startblock,
2369 new->br_blockcount + RIGHT.br_blockcount,
2370 newext)))
2371 goto done;
2372 }
2373 break;
2374
2375 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2376 /*
2377 * Setting all of a previous oldext extent to newext.
2378 * Neither the left nor right neighbors are contiguous with
2379 * the new one.
2380 */
2381 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2382 xfs_bmbt_set_state(ep, newext);
2383 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2384
2385 if (cur == NULL)
2386 rval = XFS_ILOG_DEXT;
2387 else {
2388 rval = 0;
2389 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2390 new->br_startblock, new->br_blockcount,
2391 &i)))
2392 goto done;
2393 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2394 if ((error = xfs_bmbt_update(cur, new->br_startoff,
2395 new->br_startblock, new->br_blockcount,
2396 newext)))
2397 goto done;
2398 }
2399 break;
2400
2401 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2402 /*
2403 * Setting the first part of a previous oldext extent to newext.
2404 * The left neighbor is contiguous.
2405 */
2406 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
2407 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
2408 LEFT.br_blockcount + new->br_blockcount);
2409 xfs_bmbt_set_startoff(ep,
2410 PREV.br_startoff + new->br_blockcount);
2411 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
2412
2413 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2414 xfs_bmbt_set_startblock(ep,
2415 new->br_startblock + new->br_blockcount);
2416 xfs_bmbt_set_blockcount(ep,
2417 PREV.br_blockcount - new->br_blockcount);
2418 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2419
2420 --*idx;
2421
2422 if (cur == NULL)
2423 rval = XFS_ILOG_DEXT;
2424 else {
2425 rval = 0;
2426 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2427 PREV.br_startblock, PREV.br_blockcount,
2428 &i)))
2429 goto done;
2430 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2431 if ((error = xfs_bmbt_update(cur,
2432 PREV.br_startoff + new->br_blockcount,
2433 PREV.br_startblock + new->br_blockcount,
2434 PREV.br_blockcount - new->br_blockcount,
2435 oldext)))
2436 goto done;
2437 if ((error = xfs_btree_decrement(cur, 0, &i)))
2438 goto done;
2439 error = xfs_bmbt_update(cur, LEFT.br_startoff,
2440 LEFT.br_startblock,
2441 LEFT.br_blockcount + new->br_blockcount,
2442 LEFT.br_state);
2443 if (error)
2444 goto done;
2445 }
2446 break;
2447
2448 case BMAP_LEFT_FILLING:
2449 /*
2450 * Setting the first part of a previous oldext extent to newext.
2451 * The left neighbor is not contiguous.
2452 */
2453 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2454 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
2455 xfs_bmbt_set_startoff(ep, new_endoff);
2456 xfs_bmbt_set_blockcount(ep,
2457 PREV.br_blockcount - new->br_blockcount);
2458 xfs_bmbt_set_startblock(ep,
2459 new->br_startblock + new->br_blockcount);
2460 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2461
2462 xfs_iext_insert(ip, *idx, 1, new, state);
2463 XFS_IFORK_NEXT_SET(ip, whichfork,
2464 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2465 if (cur == NULL)
2466 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2467 else {
2468 rval = XFS_ILOG_CORE;
2469 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2470 PREV.br_startblock, PREV.br_blockcount,
2471 &i)))
2472 goto done;
2473 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2474 if ((error = xfs_bmbt_update(cur,
2475 PREV.br_startoff + new->br_blockcount,
2476 PREV.br_startblock + new->br_blockcount,
2477 PREV.br_blockcount - new->br_blockcount,
2478 oldext)))
2479 goto done;
2480 cur->bc_rec.b = *new;
2481 if ((error = xfs_btree_insert(cur, &i)))
2482 goto done;
2483 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2484 }
2485 break;
2486
2487 case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2488 /*
2489 * Setting the last part of a previous oldext extent to newext.
2490 * The right neighbor is contiguous with the new allocation.
2491 */
2492 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2493 xfs_bmbt_set_blockcount(ep,
2494 PREV.br_blockcount - new->br_blockcount);
2495 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2496
2497 ++*idx;
2498
2499 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2500 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2501 new->br_startoff, new->br_startblock,
2502 new->br_blockcount + RIGHT.br_blockcount, newext);
2503 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2504
2505 if (cur == NULL)
2506 rval = XFS_ILOG_DEXT;
2507 else {
2508 rval = 0;
2509 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2510 PREV.br_startblock,
2511 PREV.br_blockcount, &i)))
2512 goto done;
2513 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2514 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2515 PREV.br_startblock,
2516 PREV.br_blockcount - new->br_blockcount,
2517 oldext)))
2518 goto done;
2519 if ((error = xfs_btree_increment(cur, 0, &i)))
2520 goto done;
2521 if ((error = xfs_bmbt_update(cur, new->br_startoff,
2522 new->br_startblock,
2523 new->br_blockcount + RIGHT.br_blockcount,
2524 newext)))
2525 goto done;
2526 }
2527 break;
2528
2529 case BMAP_RIGHT_FILLING:
2530 /*
2531 * Setting the last part of a previous oldext extent to newext.
2532 * The right neighbor is not contiguous.
2533 */
2534 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2535 xfs_bmbt_set_blockcount(ep,
2536 PREV.br_blockcount - new->br_blockcount);
2537 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2538
2539 ++*idx;
2540 xfs_iext_insert(ip, *idx, 1, new, state);
2541
2542 XFS_IFORK_NEXT_SET(ip, whichfork,
2543 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2544 if (cur == NULL)
2545 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2546 else {
2547 rval = XFS_ILOG_CORE;
2548 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2549 PREV.br_startblock, PREV.br_blockcount,
2550 &i)))
2551 goto done;
2552 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2553 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2554 PREV.br_startblock,
2555 PREV.br_blockcount - new->br_blockcount,
2556 oldext)))
2557 goto done;
2558 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2559 new->br_startblock, new->br_blockcount,
2560 &i)))
2561 goto done;
2562 XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2563 cur->bc_rec.b.br_state = new->br_state;
2564 if ((error = xfs_btree_insert(cur, &i)))
2565 goto done;
2566 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2567 }
2568 break;
2569
2570 case 0:
2571 /*
2572 * Setting the middle part of a previous oldext extent to
2573 * newext. Contiguity is impossible here.
2574 * One extent becomes three extents.
2575 */
2576 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2577 xfs_bmbt_set_blockcount(ep,
2578 new->br_startoff - PREV.br_startoff);
2579 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2580
2581 r[0] = *new;
2582 r[1].br_startoff = new_endoff;
2583 r[1].br_blockcount =
2584 PREV.br_startoff + PREV.br_blockcount - new_endoff;
2585 r[1].br_startblock = new->br_startblock + new->br_blockcount;
2586 r[1].br_state = oldext;
2587
2588 ++*idx;
2589 xfs_iext_insert(ip, *idx, 2, &r[0], state);
2590
2591 XFS_IFORK_NEXT_SET(ip, whichfork,
2592 XFS_IFORK_NEXTENTS(ip, whichfork) + 2);
2593 if (cur == NULL)
2594 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2595 else {
2596 rval = XFS_ILOG_CORE;
2597 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2598 PREV.br_startblock, PREV.br_blockcount,
2599 &i)))
2600 goto done;
2601 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2602 /* new right extent - oldext */
2603 if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
2604 r[1].br_startblock, r[1].br_blockcount,
2605 r[1].br_state)))
2606 goto done;
2607 /* new left extent - oldext */
2608 cur->bc_rec.b = PREV;
2609 cur->bc_rec.b.br_blockcount =
2610 new->br_startoff - PREV.br_startoff;
2611 if ((error = xfs_btree_insert(cur, &i)))
2612 goto done;
2613 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2614 /*
2615 * Reset the cursor to the position of the new extent
2616 * we are about to insert as we can't trust it after
2617 * the previous insert.
2618 */
2619 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2620 new->br_startblock, new->br_blockcount,
2621 &i)))
2622 goto done;
2623 XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2624 /* new middle extent - newext */
2625 cur->bc_rec.b.br_state = new->br_state;
2626 if ((error = xfs_btree_insert(cur, &i)))
2627 goto done;
2628 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2629 }
2630 break;
2631
2632 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2633 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2634 case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2635 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2636 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2637 case BMAP_LEFT_CONTIG:
2638 case BMAP_RIGHT_CONTIG:
2639 /*
2640 * These cases are all impossible.
2641 */
2642 ASSERT(0);
2643 }
2644
2645 /* update reverse mappings */
2646 error = xfs_rmap_convert_extent(mp, dfops, ip, whichfork, new);
2647 if (error)
2648 goto done;
2649
2650 /* convert to a btree if necessary */
2651 if (xfs_bmap_needs_btree(ip, whichfork)) {
2652 int tmp_logflags; /* partial log flag return val */
2653
2654 ASSERT(cur == NULL);
2655 error = xfs_bmap_extents_to_btree(tp, ip, first, dfops, &cur,
2656 0, &tmp_logflags, whichfork);
2657 *logflagsp |= tmp_logflags;
2658 if (error)
2659 goto done;
2660 }
2661
2662 /* clear out the allocated field, done with it now in any case. */
2663 if (cur) {
2664 cur->bc_private.b.allocated = 0;
2665 *curp = cur;
2666 }
2667
2668 xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
2669 done:
2670 *logflagsp |= rval;
2671 return error;
2672 #undef LEFT
2673 #undef RIGHT
2674 #undef PREV
2675 }
2676
2677 /*
2678 * Convert a hole to a delayed allocation.
2679 */
2680 STATIC void
xfs_bmap_add_extent_hole_delay(xfs_inode_t * ip,int whichfork,xfs_extnum_t * idx,xfs_bmbt_irec_t * new)2681 xfs_bmap_add_extent_hole_delay(
2682 xfs_inode_t *ip, /* incore inode pointer */
2683 int whichfork,
2684 xfs_extnum_t *idx, /* extent number to update/insert */
2685 xfs_bmbt_irec_t *new) /* new data to add to file extents */
2686 {
2687 xfs_ifork_t *ifp; /* inode fork pointer */
2688 xfs_bmbt_irec_t left; /* left neighbor extent entry */
2689 xfs_filblks_t newlen=0; /* new indirect size */
2690 xfs_filblks_t oldlen=0; /* old indirect size */
2691 xfs_bmbt_irec_t right; /* right neighbor extent entry */
2692 int state; /* state bits, accessed thru macros */
2693 xfs_filblks_t temp=0; /* temp for indirect calculations */
2694
2695 ifp = XFS_IFORK_PTR(ip, whichfork);
2696 state = 0;
2697 if (whichfork == XFS_COW_FORK)
2698 state |= BMAP_COWFORK;
2699 ASSERT(isnullstartblock(new->br_startblock));
2700
2701 /*
2702 * Check and set flags if this segment has a left neighbor
2703 */
2704 if (*idx > 0) {
2705 state |= BMAP_LEFT_VALID;
2706 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
2707
2708 if (isnullstartblock(left.br_startblock))
2709 state |= BMAP_LEFT_DELAY;
2710 }
2711
2712 /*
2713 * Check and set flags if the current (right) segment exists.
2714 * If it doesn't exist, we're converting the hole at end-of-file.
2715 */
2716 if (*idx < xfs_iext_count(ifp)) {
2717 state |= BMAP_RIGHT_VALID;
2718 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
2719
2720 if (isnullstartblock(right.br_startblock))
2721 state |= BMAP_RIGHT_DELAY;
2722 }
2723
2724 /*
2725 * Set contiguity flags on the left and right neighbors.
2726 * Don't let extents get too large, even if the pieces are contiguous.
2727 */
2728 if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2729 left.br_startoff + left.br_blockcount == new->br_startoff &&
2730 left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2731 state |= BMAP_LEFT_CONTIG;
2732
2733 if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2734 new->br_startoff + new->br_blockcount == right.br_startoff &&
2735 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2736 (!(state & BMAP_LEFT_CONTIG) ||
2737 (left.br_blockcount + new->br_blockcount +
2738 right.br_blockcount <= MAXEXTLEN)))
2739 state |= BMAP_RIGHT_CONTIG;
2740
2741 /*
2742 * Switch out based on the contiguity flags.
2743 */
2744 switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2745 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2746 /*
2747 * New allocation is contiguous with delayed allocations
2748 * on the left and on the right.
2749 * Merge all three into a single extent record.
2750 */
2751 --*idx;
2752 temp = left.br_blockcount + new->br_blockcount +
2753 right.br_blockcount;
2754
2755 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2756 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2757 oldlen = startblockval(left.br_startblock) +
2758 startblockval(new->br_startblock) +
2759 startblockval(right.br_startblock);
2760 newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2761 oldlen);
2762 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2763 nullstartblock((int)newlen));
2764 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2765
2766 xfs_iext_remove(ip, *idx + 1, 1, state);
2767 break;
2768
2769 case BMAP_LEFT_CONTIG:
2770 /*
2771 * New allocation is contiguous with a delayed allocation
2772 * on the left.
2773 * Merge the new allocation with the left neighbor.
2774 */
2775 --*idx;
2776 temp = left.br_blockcount + new->br_blockcount;
2777
2778 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2779 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2780 oldlen = startblockval(left.br_startblock) +
2781 startblockval(new->br_startblock);
2782 newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2783 oldlen);
2784 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2785 nullstartblock((int)newlen));
2786 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2787 break;
2788
2789 case BMAP_RIGHT_CONTIG:
2790 /*
2791 * New allocation is contiguous with a delayed allocation
2792 * on the right.
2793 * Merge the new allocation with the right neighbor.
2794 */
2795 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2796 temp = new->br_blockcount + right.br_blockcount;
2797 oldlen = startblockval(new->br_startblock) +
2798 startblockval(right.br_startblock);
2799 newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2800 oldlen);
2801 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2802 new->br_startoff,
2803 nullstartblock((int)newlen), temp, right.br_state);
2804 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2805 break;
2806
2807 case 0:
2808 /*
2809 * New allocation is not contiguous with another
2810 * delayed allocation.
2811 * Insert a new entry.
2812 */
2813 oldlen = newlen = 0;
2814 xfs_iext_insert(ip, *idx, 1, new, state);
2815 break;
2816 }
2817 if (oldlen != newlen) {
2818 ASSERT(oldlen > newlen);
2819 xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen),
2820 false);
2821 /*
2822 * Nothing to do for disk quota accounting here.
2823 */
2824 }
2825 }
2826
2827 /*
2828 * Convert a hole to a real allocation.
2829 */
2830 STATIC int /* error */
xfs_bmap_add_extent_hole_real(struct xfs_trans * tp,struct xfs_inode * ip,int whichfork,xfs_extnum_t * idx,struct xfs_btree_cur ** curp,struct xfs_bmbt_irec * new,xfs_fsblock_t * first,struct xfs_defer_ops * dfops,int * logflagsp)2831 xfs_bmap_add_extent_hole_real(
2832 struct xfs_trans *tp,
2833 struct xfs_inode *ip,
2834 int whichfork,
2835 xfs_extnum_t *idx,
2836 struct xfs_btree_cur **curp,
2837 struct xfs_bmbt_irec *new,
2838 xfs_fsblock_t *first,
2839 struct xfs_defer_ops *dfops,
2840 int *logflagsp)
2841 {
2842 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
2843 struct xfs_mount *mp = ip->i_mount;
2844 struct xfs_btree_cur *cur = *curp;
2845 int error; /* error return value */
2846 int i; /* temp state */
2847 xfs_bmbt_irec_t left; /* left neighbor extent entry */
2848 xfs_bmbt_irec_t right; /* right neighbor extent entry */
2849 int rval=0; /* return value (logging flags) */
2850 int state; /* state bits, accessed thru macros */
2851
2852 ASSERT(*idx >= 0);
2853 ASSERT(*idx <= xfs_iext_count(ifp));
2854 ASSERT(!isnullstartblock(new->br_startblock));
2855 ASSERT(!cur || !(cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
2856
2857 XFS_STATS_INC(mp, xs_add_exlist);
2858
2859 state = 0;
2860 if (whichfork == XFS_ATTR_FORK)
2861 state |= BMAP_ATTRFORK;
2862 if (whichfork == XFS_COW_FORK)
2863 state |= BMAP_COWFORK;
2864
2865 /*
2866 * Check and set flags if this segment has a left neighbor.
2867 */
2868 if (*idx > 0) {
2869 state |= BMAP_LEFT_VALID;
2870 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
2871 if (isnullstartblock(left.br_startblock))
2872 state |= BMAP_LEFT_DELAY;
2873 }
2874
2875 /*
2876 * Check and set flags if this segment has a current value.
2877 * Not true if we're inserting into the "hole" at eof.
2878 */
2879 if (*idx < xfs_iext_count(ifp)) {
2880 state |= BMAP_RIGHT_VALID;
2881 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
2882 if (isnullstartblock(right.br_startblock))
2883 state |= BMAP_RIGHT_DELAY;
2884 }
2885
2886 /*
2887 * We're inserting a real allocation between "left" and "right".
2888 * Set the contiguity flags. Don't let extents get too large.
2889 */
2890 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2891 left.br_startoff + left.br_blockcount == new->br_startoff &&
2892 left.br_startblock + left.br_blockcount == new->br_startblock &&
2893 left.br_state == new->br_state &&
2894 left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2895 state |= BMAP_LEFT_CONTIG;
2896
2897 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2898 new->br_startoff + new->br_blockcount == right.br_startoff &&
2899 new->br_startblock + new->br_blockcount == right.br_startblock &&
2900 new->br_state == right.br_state &&
2901 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2902 (!(state & BMAP_LEFT_CONTIG) ||
2903 left.br_blockcount + new->br_blockcount +
2904 right.br_blockcount <= MAXEXTLEN))
2905 state |= BMAP_RIGHT_CONTIG;
2906
2907 error = 0;
2908 /*
2909 * Select which case we're in here, and implement it.
2910 */
2911 switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2912 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2913 /*
2914 * New allocation is contiguous with real allocations on the
2915 * left and on the right.
2916 * Merge all three into a single extent record.
2917 */
2918 --*idx;
2919 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2920 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2921 left.br_blockcount + new->br_blockcount +
2922 right.br_blockcount);
2923 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2924
2925 xfs_iext_remove(ip, *idx + 1, 1, state);
2926
2927 XFS_IFORK_NEXT_SET(ip, whichfork,
2928 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2929 if (cur == NULL) {
2930 rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2931 } else {
2932 rval = XFS_ILOG_CORE;
2933 error = xfs_bmbt_lookup_eq(cur, right.br_startoff,
2934 right.br_startblock, right.br_blockcount,
2935 &i);
2936 if (error)
2937 goto done;
2938 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2939 error = xfs_btree_delete(cur, &i);
2940 if (error)
2941 goto done;
2942 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2943 error = xfs_btree_decrement(cur, 0, &i);
2944 if (error)
2945 goto done;
2946 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2947 error = xfs_bmbt_update(cur, left.br_startoff,
2948 left.br_startblock,
2949 left.br_blockcount +
2950 new->br_blockcount +
2951 right.br_blockcount,
2952 left.br_state);
2953 if (error)
2954 goto done;
2955 }
2956 break;
2957
2958 case BMAP_LEFT_CONTIG:
2959 /*
2960 * New allocation is contiguous with a real allocation
2961 * on the left.
2962 * Merge the new allocation with the left neighbor.
2963 */
2964 --*idx;
2965 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2966 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2967 left.br_blockcount + new->br_blockcount);
2968 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2969
2970 if (cur == NULL) {
2971 rval = xfs_ilog_fext(whichfork);
2972 } else {
2973 rval = 0;
2974 error = xfs_bmbt_lookup_eq(cur, left.br_startoff,
2975 left.br_startblock, left.br_blockcount,
2976 &i);
2977 if (error)
2978 goto done;
2979 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2980 error = xfs_bmbt_update(cur, left.br_startoff,
2981 left.br_startblock,
2982 left.br_blockcount +
2983 new->br_blockcount,
2984 left.br_state);
2985 if (error)
2986 goto done;
2987 }
2988 break;
2989
2990 case BMAP_RIGHT_CONTIG:
2991 /*
2992 * New allocation is contiguous with a real allocation
2993 * on the right.
2994 * Merge the new allocation with the right neighbor.
2995 */
2996 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2997 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2998 new->br_startoff, new->br_startblock,
2999 new->br_blockcount + right.br_blockcount,
3000 right.br_state);
3001 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
3002
3003 if (cur == NULL) {
3004 rval = xfs_ilog_fext(whichfork);
3005 } else {
3006 rval = 0;
3007 error = xfs_bmbt_lookup_eq(cur,
3008 right.br_startoff,
3009 right.br_startblock,
3010 right.br_blockcount, &i);
3011 if (error)
3012 goto done;
3013 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3014 error = xfs_bmbt_update(cur, new->br_startoff,
3015 new->br_startblock,
3016 new->br_blockcount +
3017 right.br_blockcount,
3018 right.br_state);
3019 if (error)
3020 goto done;
3021 }
3022 break;
3023
3024 case 0:
3025 /*
3026 * New allocation is not contiguous with another
3027 * real allocation.
3028 * Insert a new entry.
3029 */
3030 xfs_iext_insert(ip, *idx, 1, new, state);
3031 XFS_IFORK_NEXT_SET(ip, whichfork,
3032 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
3033 if (cur == NULL) {
3034 rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3035 } else {
3036 rval = XFS_ILOG_CORE;
3037 error = xfs_bmbt_lookup_eq(cur,
3038 new->br_startoff,
3039 new->br_startblock,
3040 new->br_blockcount, &i);
3041 if (error)
3042 goto done;
3043 XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
3044 cur->bc_rec.b.br_state = new->br_state;
3045 error = xfs_btree_insert(cur, &i);
3046 if (error)
3047 goto done;
3048 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3049 }
3050 break;
3051 }
3052
3053 /* add reverse mapping */
3054 error = xfs_rmap_map_extent(mp, dfops, ip, whichfork, new);
3055 if (error)
3056 goto done;
3057
3058 /* convert to a btree if necessary */
3059 if (xfs_bmap_needs_btree(ip, whichfork)) {
3060 int tmp_logflags; /* partial log flag return val */
3061
3062 ASSERT(cur == NULL);
3063 error = xfs_bmap_extents_to_btree(tp, ip, first, dfops, curp,
3064 0, &tmp_logflags, whichfork);
3065 *logflagsp |= tmp_logflags;
3066 cur = *curp;
3067 if (error)
3068 goto done;
3069 }
3070
3071 /* clear out the allocated field, done with it now in any case. */
3072 if (cur)
3073 cur->bc_private.b.allocated = 0;
3074
3075 xfs_bmap_check_leaf_extents(cur, ip, whichfork);
3076 done:
3077 *logflagsp |= rval;
3078 return error;
3079 }
3080
3081 /*
3082 * Functions used in the extent read, allocate and remove paths
3083 */
3084
3085 /*
3086 * Adjust the size of the new extent based on di_extsize and rt extsize.
3087 */
3088 int
xfs_bmap_extsize_align(xfs_mount_t * mp,xfs_bmbt_irec_t * gotp,xfs_bmbt_irec_t * prevp,xfs_extlen_t extsz,int rt,int eof,int delay,int convert,xfs_fileoff_t * offp,xfs_extlen_t * lenp)3089 xfs_bmap_extsize_align(
3090 xfs_mount_t *mp,
3091 xfs_bmbt_irec_t *gotp, /* next extent pointer */
3092 xfs_bmbt_irec_t *prevp, /* previous extent pointer */
3093 xfs_extlen_t extsz, /* align to this extent size */
3094 int rt, /* is this a realtime inode? */
3095 int eof, /* is extent at end-of-file? */
3096 int delay, /* creating delalloc extent? */
3097 int convert, /* overwriting unwritten extent? */
3098 xfs_fileoff_t *offp, /* in/out: aligned offset */
3099 xfs_extlen_t *lenp) /* in/out: aligned length */
3100 {
3101 xfs_fileoff_t orig_off; /* original offset */
3102 xfs_extlen_t orig_alen; /* original length */
3103 xfs_fileoff_t orig_end; /* original off+len */
3104 xfs_fileoff_t nexto; /* next file offset */
3105 xfs_fileoff_t prevo; /* previous file offset */
3106 xfs_fileoff_t align_off; /* temp for offset */
3107 xfs_extlen_t align_alen; /* temp for length */
3108 xfs_extlen_t temp; /* temp for calculations */
3109
3110 if (convert)
3111 return 0;
3112
3113 orig_off = align_off = *offp;
3114 orig_alen = align_alen = *lenp;
3115 orig_end = orig_off + orig_alen;
3116
3117 /*
3118 * If this request overlaps an existing extent, then don't
3119 * attempt to perform any additional alignment.
3120 */
3121 if (!delay && !eof &&
3122 (orig_off >= gotp->br_startoff) &&
3123 (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
3124 return 0;
3125 }
3126
3127 /*
3128 * If the file offset is unaligned vs. the extent size
3129 * we need to align it. This will be possible unless
3130 * the file was previously written with a kernel that didn't
3131 * perform this alignment, or if a truncate shot us in the
3132 * foot.
3133 */
3134 temp = do_mod(orig_off, extsz);
3135 if (temp) {
3136 align_alen += temp;
3137 align_off -= temp;
3138 }
3139
3140 /* Same adjustment for the end of the requested area. */
3141 temp = (align_alen % extsz);
3142 if (temp)
3143 align_alen += extsz - temp;
3144
3145 /*
3146 * For large extent hint sizes, the aligned extent might be larger than
3147 * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls
3148 * the length back under MAXEXTLEN. The outer allocation loops handle
3149 * short allocation just fine, so it is safe to do this. We only want to
3150 * do it when we are forced to, though, because it means more allocation
3151 * operations are required.
3152 */
3153 while (align_alen > MAXEXTLEN)
3154 align_alen -= extsz;
3155 ASSERT(align_alen <= MAXEXTLEN);
3156
3157 /*
3158 * If the previous block overlaps with this proposed allocation
3159 * then move the start forward without adjusting the length.
3160 */
3161 if (prevp->br_startoff != NULLFILEOFF) {
3162 if (prevp->br_startblock == HOLESTARTBLOCK)
3163 prevo = prevp->br_startoff;
3164 else
3165 prevo = prevp->br_startoff + prevp->br_blockcount;
3166 } else
3167 prevo = 0;
3168 if (align_off != orig_off && align_off < prevo)
3169 align_off = prevo;
3170 /*
3171 * If the next block overlaps with this proposed allocation
3172 * then move the start back without adjusting the length,
3173 * but not before offset 0.
3174 * This may of course make the start overlap previous block,
3175 * and if we hit the offset 0 limit then the next block
3176 * can still overlap too.
3177 */
3178 if (!eof && gotp->br_startoff != NULLFILEOFF) {
3179 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
3180 (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
3181 nexto = gotp->br_startoff + gotp->br_blockcount;
3182 else
3183 nexto = gotp->br_startoff;
3184 } else
3185 nexto = NULLFILEOFF;
3186 if (!eof &&
3187 align_off + align_alen != orig_end &&
3188 align_off + align_alen > nexto)
3189 align_off = nexto > align_alen ? nexto - align_alen : 0;
3190 /*
3191 * If we're now overlapping the next or previous extent that
3192 * means we can't fit an extsz piece in this hole. Just move
3193 * the start forward to the first valid spot and set
3194 * the length so we hit the end.
3195 */
3196 if (align_off != orig_off && align_off < prevo)
3197 align_off = prevo;
3198 if (align_off + align_alen != orig_end &&
3199 align_off + align_alen > nexto &&
3200 nexto != NULLFILEOFF) {
3201 ASSERT(nexto > prevo);
3202 align_alen = nexto - align_off;
3203 }
3204
3205 /*
3206 * If realtime, and the result isn't a multiple of the realtime
3207 * extent size we need to remove blocks until it is.
3208 */
3209 if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
3210 /*
3211 * We're not covering the original request, or
3212 * we won't be able to once we fix the length.
3213 */
3214 if (orig_off < align_off ||
3215 orig_end > align_off + align_alen ||
3216 align_alen - temp < orig_alen)
3217 return -EINVAL;
3218 /*
3219 * Try to fix it by moving the start up.
3220 */
3221 if (align_off + temp <= orig_off) {
3222 align_alen -= temp;
3223 align_off += temp;
3224 }
3225 /*
3226 * Try to fix it by moving the end in.
3227 */
3228 else if (align_off + align_alen - temp >= orig_end)
3229 align_alen -= temp;
3230 /*
3231 * Set the start to the minimum then trim the length.
3232 */
3233 else {
3234 align_alen -= orig_off - align_off;
3235 align_off = orig_off;
3236 align_alen -= align_alen % mp->m_sb.sb_rextsize;
3237 }
3238 /*
3239 * Result doesn't cover the request, fail it.
3240 */
3241 if (orig_off < align_off || orig_end > align_off + align_alen)
3242 return -EINVAL;
3243 } else {
3244 ASSERT(orig_off >= align_off);
3245 /* see MAXEXTLEN handling above */
3246 ASSERT(orig_end <= align_off + align_alen ||
3247 align_alen + extsz > MAXEXTLEN);
3248 }
3249
3250 #ifdef DEBUG
3251 if (!eof && gotp->br_startoff != NULLFILEOFF)
3252 ASSERT(align_off + align_alen <= gotp->br_startoff);
3253 if (prevp->br_startoff != NULLFILEOFF)
3254 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3255 #endif
3256
3257 *lenp = align_alen;
3258 *offp = align_off;
3259 return 0;
3260 }
3261
3262 #define XFS_ALLOC_GAP_UNITS 4
3263
3264 void
xfs_bmap_adjacent(struct xfs_bmalloca * ap)3265 xfs_bmap_adjacent(
3266 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
3267 {
3268 xfs_fsblock_t adjust; /* adjustment to block numbers */
3269 xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
3270 xfs_mount_t *mp; /* mount point structure */
3271 int nullfb; /* true if ap->firstblock isn't set */
3272 int rt; /* true if inode is realtime */
3273
3274 #define ISVALID(x,y) \
3275 (rt ? \
3276 (x) < mp->m_sb.sb_rblocks : \
3277 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3278 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3279 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3280
3281 mp = ap->ip->i_mount;
3282 nullfb = *ap->firstblock == NULLFSBLOCK;
3283 rt = XFS_IS_REALTIME_INODE(ap->ip) &&
3284 xfs_alloc_is_userdata(ap->datatype);
3285 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3286 /*
3287 * If allocating at eof, and there's a previous real block,
3288 * try to use its last block as our starting point.
3289 */
3290 if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3291 !isnullstartblock(ap->prev.br_startblock) &&
3292 ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3293 ap->prev.br_startblock)) {
3294 ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3295 /*
3296 * Adjust for the gap between prevp and us.
3297 */
3298 adjust = ap->offset -
3299 (ap->prev.br_startoff + ap->prev.br_blockcount);
3300 if (adjust &&
3301 ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3302 ap->blkno += adjust;
3303 }
3304 /*
3305 * If not at eof, then compare the two neighbor blocks.
3306 * Figure out whether either one gives us a good starting point,
3307 * and pick the better one.
3308 */
3309 else if (!ap->eof) {
3310 xfs_fsblock_t gotbno; /* right side block number */
3311 xfs_fsblock_t gotdiff=0; /* right side difference */
3312 xfs_fsblock_t prevbno; /* left side block number */
3313 xfs_fsblock_t prevdiff=0; /* left side difference */
3314
3315 /*
3316 * If there's a previous (left) block, select a requested
3317 * start block based on it.
3318 */
3319 if (ap->prev.br_startoff != NULLFILEOFF &&
3320 !isnullstartblock(ap->prev.br_startblock) &&
3321 (prevbno = ap->prev.br_startblock +
3322 ap->prev.br_blockcount) &&
3323 ISVALID(prevbno, ap->prev.br_startblock)) {
3324 /*
3325 * Calculate gap to end of previous block.
3326 */
3327 adjust = prevdiff = ap->offset -
3328 (ap->prev.br_startoff +
3329 ap->prev.br_blockcount);
3330 /*
3331 * Figure the startblock based on the previous block's
3332 * end and the gap size.
3333 * Heuristic!
3334 * If the gap is large relative to the piece we're
3335 * allocating, or using it gives us an invalid block
3336 * number, then just use the end of the previous block.
3337 */
3338 if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3339 ISVALID(prevbno + prevdiff,
3340 ap->prev.br_startblock))
3341 prevbno += adjust;
3342 else
3343 prevdiff += adjust;
3344 /*
3345 * If the firstblock forbids it, can't use it,
3346 * must use default.
3347 */
3348 if (!rt && !nullfb &&
3349 XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3350 prevbno = NULLFSBLOCK;
3351 }
3352 /*
3353 * No previous block or can't follow it, just default.
3354 */
3355 else
3356 prevbno = NULLFSBLOCK;
3357 /*
3358 * If there's a following (right) block, select a requested
3359 * start block based on it.
3360 */
3361 if (!isnullstartblock(ap->got.br_startblock)) {
3362 /*
3363 * Calculate gap to start of next block.
3364 */
3365 adjust = gotdiff = ap->got.br_startoff - ap->offset;
3366 /*
3367 * Figure the startblock based on the next block's
3368 * start and the gap size.
3369 */
3370 gotbno = ap->got.br_startblock;
3371 /*
3372 * Heuristic!
3373 * If the gap is large relative to the piece we're
3374 * allocating, or using it gives us an invalid block
3375 * number, then just use the start of the next block
3376 * offset by our length.
3377 */
3378 if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3379 ISVALID(gotbno - gotdiff, gotbno))
3380 gotbno -= adjust;
3381 else if (ISVALID(gotbno - ap->length, gotbno)) {
3382 gotbno -= ap->length;
3383 gotdiff += adjust - ap->length;
3384 } else
3385 gotdiff += adjust;
3386 /*
3387 * If the firstblock forbids it, can't use it,
3388 * must use default.
3389 */
3390 if (!rt && !nullfb &&
3391 XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3392 gotbno = NULLFSBLOCK;
3393 }
3394 /*
3395 * No next block, just default.
3396 */
3397 else
3398 gotbno = NULLFSBLOCK;
3399 /*
3400 * If both valid, pick the better one, else the only good
3401 * one, else ap->blkno is already set (to 0 or the inode block).
3402 */
3403 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3404 ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3405 else if (prevbno != NULLFSBLOCK)
3406 ap->blkno = prevbno;
3407 else if (gotbno != NULLFSBLOCK)
3408 ap->blkno = gotbno;
3409 }
3410 #undef ISVALID
3411 }
3412
3413 static int
xfs_bmap_longest_free_extent(struct xfs_trans * tp,xfs_agnumber_t ag,xfs_extlen_t * blen,int * notinit)3414 xfs_bmap_longest_free_extent(
3415 struct xfs_trans *tp,
3416 xfs_agnumber_t ag,
3417 xfs_extlen_t *blen,
3418 int *notinit)
3419 {
3420 struct xfs_mount *mp = tp->t_mountp;
3421 struct xfs_perag *pag;
3422 xfs_extlen_t longest;
3423 int error = 0;
3424
3425 pag = xfs_perag_get(mp, ag);
3426 if (!pag->pagf_init) {
3427 error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK);
3428 if (error)
3429 goto out;
3430
3431 if (!pag->pagf_init) {
3432 *notinit = 1;
3433 goto out;
3434 }
3435 }
3436
3437 longest = xfs_alloc_longest_free_extent(mp, pag,
3438 xfs_alloc_min_freelist(mp, pag),
3439 xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE));
3440 if (*blen < longest)
3441 *blen = longest;
3442
3443 out:
3444 xfs_perag_put(pag);
3445 return error;
3446 }
3447
3448 static void
xfs_bmap_select_minlen(struct xfs_bmalloca * ap,struct xfs_alloc_arg * args,xfs_extlen_t * blen,int notinit)3449 xfs_bmap_select_minlen(
3450 struct xfs_bmalloca *ap,
3451 struct xfs_alloc_arg *args,
3452 xfs_extlen_t *blen,
3453 int notinit)
3454 {
3455 if (notinit || *blen < ap->minlen) {
3456 /*
3457 * Since we did a BUF_TRYLOCK above, it is possible that
3458 * there is space for this request.
3459 */
3460 args->minlen = ap->minlen;
3461 } else if (*blen < args->maxlen) {
3462 /*
3463 * If the best seen length is less than the request length,
3464 * use the best as the minimum.
3465 */
3466 args->minlen = *blen;
3467 } else {
3468 /*
3469 * Otherwise we've seen an extent as big as maxlen, use that
3470 * as the minimum.
3471 */
3472 args->minlen = args->maxlen;
3473 }
3474 }
3475
3476 STATIC int
xfs_bmap_btalloc_nullfb(struct xfs_bmalloca * ap,struct xfs_alloc_arg * args,xfs_extlen_t * blen)3477 xfs_bmap_btalloc_nullfb(
3478 struct xfs_bmalloca *ap,
3479 struct xfs_alloc_arg *args,
3480 xfs_extlen_t *blen)
3481 {
3482 struct xfs_mount *mp = ap->ip->i_mount;
3483 xfs_agnumber_t ag, startag;
3484 int notinit = 0;
3485 int error;
3486
3487 args->type = XFS_ALLOCTYPE_START_BNO;
3488 args->total = ap->total;
3489
3490 startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3491 if (startag == NULLAGNUMBER)
3492 startag = ag = 0;
3493
3494 while (*blen < args->maxlen) {
3495 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3496 ¬init);
3497 if (error)
3498 return error;
3499
3500 if (++ag == mp->m_sb.sb_agcount)
3501 ag = 0;
3502 if (ag == startag)
3503 break;
3504 }
3505
3506 xfs_bmap_select_minlen(ap, args, blen, notinit);
3507 return 0;
3508 }
3509
3510 STATIC int
xfs_bmap_btalloc_filestreams(struct xfs_bmalloca * ap,struct xfs_alloc_arg * args,xfs_extlen_t * blen)3511 xfs_bmap_btalloc_filestreams(
3512 struct xfs_bmalloca *ap,
3513 struct xfs_alloc_arg *args,
3514 xfs_extlen_t *blen)
3515 {
3516 struct xfs_mount *mp = ap->ip->i_mount;
3517 xfs_agnumber_t ag;
3518 int notinit = 0;
3519 int error;
3520
3521 args->type = XFS_ALLOCTYPE_NEAR_BNO;
3522 args->total = ap->total;
3523
3524 ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3525 if (ag == NULLAGNUMBER)
3526 ag = 0;
3527
3528 error = xfs_bmap_longest_free_extent(args->tp, ag, blen, ¬init);
3529 if (error)
3530 return error;
3531
3532 if (*blen < args->maxlen) {
3533 error = xfs_filestream_new_ag(ap, &ag);
3534 if (error)
3535 return error;
3536
3537 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3538 ¬init);
3539 if (error)
3540 return error;
3541
3542 }
3543
3544 xfs_bmap_select_minlen(ap, args, blen, notinit);
3545
3546 /*
3547 * Set the failure fallback case to look in the selected AG as stream
3548 * may have moved.
3549 */
3550 ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
3551 return 0;
3552 }
3553
3554 STATIC int
xfs_bmap_btalloc(struct xfs_bmalloca * ap)3555 xfs_bmap_btalloc(
3556 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
3557 {
3558 xfs_mount_t *mp; /* mount point structure */
3559 xfs_alloctype_t atype = 0; /* type for allocation routines */
3560 xfs_extlen_t align = 0; /* minimum allocation alignment */
3561 xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
3562 xfs_agnumber_t ag;
3563 xfs_alloc_arg_t args;
3564 xfs_extlen_t blen;
3565 xfs_extlen_t nextminlen = 0;
3566 int nullfb; /* true if ap->firstblock isn't set */
3567 int isaligned;
3568 int tryagain;
3569 int error;
3570 int stripe_align;
3571
3572 ASSERT(ap->length);
3573
3574 mp = ap->ip->i_mount;
3575
3576 /* stripe alignment for allocation is determined by mount parameters */
3577 stripe_align = 0;
3578 if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
3579 stripe_align = mp->m_swidth;
3580 else if (mp->m_dalign)
3581 stripe_align = mp->m_dalign;
3582
3583 if (ap->flags & XFS_BMAPI_COWFORK)
3584 align = xfs_get_cowextsz_hint(ap->ip);
3585 else if (xfs_alloc_is_userdata(ap->datatype))
3586 align = xfs_get_extsz_hint(ap->ip);
3587 if (align) {
3588 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
3589 align, 0, ap->eof, 0, ap->conv,
3590 &ap->offset, &ap->length);
3591 ASSERT(!error);
3592 ASSERT(ap->length);
3593 }
3594
3595
3596 nullfb = *ap->firstblock == NULLFSBLOCK;
3597 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3598 if (nullfb) {
3599 if (xfs_alloc_is_userdata(ap->datatype) &&
3600 xfs_inode_is_filestream(ap->ip)) {
3601 ag = xfs_filestream_lookup_ag(ap->ip);
3602 ag = (ag != NULLAGNUMBER) ? ag : 0;
3603 ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3604 } else {
3605 ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3606 }
3607 } else
3608 ap->blkno = *ap->firstblock;
3609
3610 xfs_bmap_adjacent(ap);
3611
3612 /*
3613 * If allowed, use ap->blkno; otherwise must use firstblock since
3614 * it's in the right allocation group.
3615 */
3616 if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3617 ;
3618 else
3619 ap->blkno = *ap->firstblock;
3620 /*
3621 * Normal allocation, done through xfs_alloc_vextent.
3622 */
3623 tryagain = isaligned = 0;
3624 memset(&args, 0, sizeof(args));
3625 args.tp = ap->tp;
3626 args.mp = mp;
3627 args.fsbno = ap->blkno;
3628 xfs_rmap_skip_owner_update(&args.oinfo);
3629
3630 /* Trim the allocation back to the maximum an AG can fit. */
3631 args.maxlen = MIN(ap->length, mp->m_ag_max_usable);
3632 args.firstblock = *ap->firstblock;
3633 blen = 0;
3634 if (nullfb) {
3635 /*
3636 * Search for an allocation group with a single extent large
3637 * enough for the request. If one isn't found, then adjust
3638 * the minimum allocation size to the largest space found.
3639 */
3640 if (xfs_alloc_is_userdata(ap->datatype) &&
3641 xfs_inode_is_filestream(ap->ip))
3642 error = xfs_bmap_btalloc_filestreams(ap, &args, &blen);
3643 else
3644 error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
3645 if (error)
3646 return error;
3647 } else if (ap->dfops->dop_low) {
3648 if (xfs_inode_is_filestream(ap->ip))
3649 args.type = XFS_ALLOCTYPE_FIRST_AG;
3650 else
3651 args.type = XFS_ALLOCTYPE_START_BNO;
3652 args.total = args.minlen = ap->minlen;
3653 } else {
3654 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3655 args.total = ap->total;
3656 args.minlen = ap->minlen;
3657 }
3658 /* apply extent size hints if obtained earlier */
3659 if (align) {
3660 args.prod = align;
3661 if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
3662 args.mod = (xfs_extlen_t)(args.prod - args.mod);
3663 } else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) {
3664 args.prod = 1;
3665 args.mod = 0;
3666 } else {
3667 args.prod = PAGE_SIZE >> mp->m_sb.sb_blocklog;
3668 if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
3669 args.mod = (xfs_extlen_t)(args.prod - args.mod);
3670 }
3671 /*
3672 * If we are not low on available data blocks, and the
3673 * underlying logical volume manager is a stripe, and
3674 * the file offset is zero then try to allocate data
3675 * blocks on stripe unit boundary.
3676 * NOTE: ap->aeof is only set if the allocation length
3677 * is >= the stripe unit and the allocation offset is
3678 * at the end of file.
3679 */
3680 if (!ap->dfops->dop_low && ap->aeof) {
3681 if (!ap->offset) {
3682 args.alignment = stripe_align;
3683 atype = args.type;
3684 isaligned = 1;
3685 /*
3686 * Adjust for alignment
3687 */
3688 if (blen > args.alignment && blen <= args.maxlen)
3689 args.minlen = blen - args.alignment;
3690 args.minalignslop = 0;
3691 } else {
3692 /*
3693 * First try an exact bno allocation.
3694 * If it fails then do a near or start bno
3695 * allocation with alignment turned on.
3696 */
3697 atype = args.type;
3698 tryagain = 1;
3699 args.type = XFS_ALLOCTYPE_THIS_BNO;
3700 args.alignment = 1;
3701 /*
3702 * Compute the minlen+alignment for the
3703 * next case. Set slop so that the value
3704 * of minlen+alignment+slop doesn't go up
3705 * between the calls.
3706 */
3707 if (blen > stripe_align && blen <= args.maxlen)
3708 nextminlen = blen - stripe_align;
3709 else
3710 nextminlen = args.minlen;
3711 if (nextminlen + stripe_align > args.minlen + 1)
3712 args.minalignslop =
3713 nextminlen + stripe_align -
3714 args.minlen - 1;
3715 else
3716 args.minalignslop = 0;
3717 }
3718 } else {
3719 args.alignment = 1;
3720 args.minalignslop = 0;
3721 }
3722 args.minleft = ap->minleft;
3723 args.wasdel = ap->wasdel;
3724 args.resv = XFS_AG_RESV_NONE;
3725 args.datatype = ap->datatype;
3726 if (ap->datatype & XFS_ALLOC_USERDATA_ZERO)
3727 args.ip = ap->ip;
3728
3729 error = xfs_alloc_vextent(&args);
3730 if (error)
3731 return error;
3732
3733 if (tryagain && args.fsbno == NULLFSBLOCK) {
3734 /*
3735 * Exact allocation failed. Now try with alignment
3736 * turned on.
3737 */
3738 args.type = atype;
3739 args.fsbno = ap->blkno;
3740 args.alignment = stripe_align;
3741 args.minlen = nextminlen;
3742 args.minalignslop = 0;
3743 isaligned = 1;
3744 if ((error = xfs_alloc_vextent(&args)))
3745 return error;
3746 }
3747 if (isaligned && args.fsbno == NULLFSBLOCK) {
3748 /*
3749 * allocation failed, so turn off alignment and
3750 * try again.
3751 */
3752 args.type = atype;
3753 args.fsbno = ap->blkno;
3754 args.alignment = 0;
3755 if ((error = xfs_alloc_vextent(&args)))
3756 return error;
3757 }
3758 if (args.fsbno == NULLFSBLOCK && nullfb &&
3759 args.minlen > ap->minlen) {
3760 args.minlen = ap->minlen;
3761 args.type = XFS_ALLOCTYPE_START_BNO;
3762 args.fsbno = ap->blkno;
3763 if ((error = xfs_alloc_vextent(&args)))
3764 return error;
3765 }
3766 if (args.fsbno == NULLFSBLOCK && nullfb) {
3767 args.fsbno = 0;
3768 args.type = XFS_ALLOCTYPE_FIRST_AG;
3769 args.total = ap->minlen;
3770 if ((error = xfs_alloc_vextent(&args)))
3771 return error;
3772 ap->dfops->dop_low = true;
3773 }
3774 if (args.fsbno != NULLFSBLOCK) {
3775 /*
3776 * check the allocation happened at the same or higher AG than
3777 * the first block that was allocated.
3778 */
3779 ASSERT(*ap->firstblock == NULLFSBLOCK ||
3780 XFS_FSB_TO_AGNO(mp, *ap->firstblock) <=
3781 XFS_FSB_TO_AGNO(mp, args.fsbno));
3782
3783 ap->blkno = args.fsbno;
3784 if (*ap->firstblock == NULLFSBLOCK)
3785 *ap->firstblock = args.fsbno;
3786 ASSERT(nullfb || fb_agno <= args.agno);
3787 ap->length = args.len;
3788 if (!(ap->flags & XFS_BMAPI_COWFORK))
3789 ap->ip->i_d.di_nblocks += args.len;
3790 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3791 if (ap->wasdel)
3792 ap->ip->i_delayed_blks -= args.len;
3793 /*
3794 * Adjust the disk quota also. This was reserved
3795 * earlier.
3796 */
3797 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3798 ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
3799 XFS_TRANS_DQ_BCOUNT,
3800 (long) args.len);
3801 } else {
3802 ap->blkno = NULLFSBLOCK;
3803 ap->length = 0;
3804 }
3805 return 0;
3806 }
3807
3808 /*
3809 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
3810 * It figures out where to ask the underlying allocator to put the new extent.
3811 */
3812 STATIC int
xfs_bmap_alloc(struct xfs_bmalloca * ap)3813 xfs_bmap_alloc(
3814 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
3815 {
3816 if (XFS_IS_REALTIME_INODE(ap->ip) &&
3817 xfs_alloc_is_userdata(ap->datatype))
3818 return xfs_bmap_rtalloc(ap);
3819 return xfs_bmap_btalloc(ap);
3820 }
3821
3822 /* Trim extent to fit a logical block range. */
3823 void
xfs_trim_extent(struct xfs_bmbt_irec * irec,xfs_fileoff_t bno,xfs_filblks_t len)3824 xfs_trim_extent(
3825 struct xfs_bmbt_irec *irec,
3826 xfs_fileoff_t bno,
3827 xfs_filblks_t len)
3828 {
3829 xfs_fileoff_t distance;
3830 xfs_fileoff_t end = bno + len;
3831
3832 if (irec->br_startoff + irec->br_blockcount <= bno ||
3833 irec->br_startoff >= end) {
3834 irec->br_blockcount = 0;
3835 return;
3836 }
3837
3838 if (irec->br_startoff < bno) {
3839 distance = bno - irec->br_startoff;
3840 if (isnullstartblock(irec->br_startblock))
3841 irec->br_startblock = DELAYSTARTBLOCK;
3842 if (irec->br_startblock != DELAYSTARTBLOCK &&
3843 irec->br_startblock != HOLESTARTBLOCK)
3844 irec->br_startblock += distance;
3845 irec->br_startoff += distance;
3846 irec->br_blockcount -= distance;
3847 }
3848
3849 if (end < irec->br_startoff + irec->br_blockcount) {
3850 distance = irec->br_startoff + irec->br_blockcount - end;
3851 irec->br_blockcount -= distance;
3852 }
3853 }
3854
3855 /* trim extent to within eof */
3856 void
xfs_trim_extent_eof(struct xfs_bmbt_irec * irec,struct xfs_inode * ip)3857 xfs_trim_extent_eof(
3858 struct xfs_bmbt_irec *irec,
3859 struct xfs_inode *ip)
3860
3861 {
3862 xfs_trim_extent(irec, 0, XFS_B_TO_FSB(ip->i_mount,
3863 i_size_read(VFS_I(ip))));
3864 }
3865
3866 /*
3867 * Trim the returned map to the required bounds
3868 */
3869 STATIC void
xfs_bmapi_trim_map(struct xfs_bmbt_irec * mval,struct xfs_bmbt_irec * got,xfs_fileoff_t * bno,xfs_filblks_t len,xfs_fileoff_t obno,xfs_fileoff_t end,int n,int flags)3870 xfs_bmapi_trim_map(
3871 struct xfs_bmbt_irec *mval,
3872 struct xfs_bmbt_irec *got,
3873 xfs_fileoff_t *bno,
3874 xfs_filblks_t len,
3875 xfs_fileoff_t obno,
3876 xfs_fileoff_t end,
3877 int n,
3878 int flags)
3879 {
3880 if ((flags & XFS_BMAPI_ENTIRE) ||
3881 got->br_startoff + got->br_blockcount <= obno) {
3882 *mval = *got;
3883 if (isnullstartblock(got->br_startblock))
3884 mval->br_startblock = DELAYSTARTBLOCK;
3885 return;
3886 }
3887
3888 if (obno > *bno)
3889 *bno = obno;
3890 ASSERT((*bno >= obno) || (n == 0));
3891 ASSERT(*bno < end);
3892 mval->br_startoff = *bno;
3893 if (isnullstartblock(got->br_startblock))
3894 mval->br_startblock = DELAYSTARTBLOCK;
3895 else
3896 mval->br_startblock = got->br_startblock +
3897 (*bno - got->br_startoff);
3898 /*
3899 * Return the minimum of what we got and what we asked for for
3900 * the length. We can use the len variable here because it is
3901 * modified below and we could have been there before coming
3902 * here if the first part of the allocation didn't overlap what
3903 * was asked for.
3904 */
3905 mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3906 got->br_blockcount - (*bno - got->br_startoff));
3907 mval->br_state = got->br_state;
3908 ASSERT(mval->br_blockcount <= len);
3909 return;
3910 }
3911
3912 /*
3913 * Update and validate the extent map to return
3914 */
3915 STATIC void
xfs_bmapi_update_map(struct xfs_bmbt_irec ** map,xfs_fileoff_t * bno,xfs_filblks_t * len,xfs_fileoff_t obno,xfs_fileoff_t end,int * n,int flags)3916 xfs_bmapi_update_map(
3917 struct xfs_bmbt_irec **map,
3918 xfs_fileoff_t *bno,
3919 xfs_filblks_t *len,
3920 xfs_fileoff_t obno,
3921 xfs_fileoff_t end,
3922 int *n,
3923 int flags)
3924 {
3925 xfs_bmbt_irec_t *mval = *map;
3926
3927 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3928 ((mval->br_startoff + mval->br_blockcount) <= end));
3929 ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3930 (mval->br_startoff < obno));
3931
3932 *bno = mval->br_startoff + mval->br_blockcount;
3933 *len = end - *bno;
3934 if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3935 /* update previous map with new information */
3936 ASSERT(mval->br_startblock == mval[-1].br_startblock);
3937 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3938 ASSERT(mval->br_state == mval[-1].br_state);
3939 mval[-1].br_blockcount = mval->br_blockcount;
3940 mval[-1].br_state = mval->br_state;
3941 } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3942 mval[-1].br_startblock != DELAYSTARTBLOCK &&
3943 mval[-1].br_startblock != HOLESTARTBLOCK &&
3944 mval->br_startblock == mval[-1].br_startblock +
3945 mval[-1].br_blockcount &&
3946 ((flags & XFS_BMAPI_IGSTATE) ||
3947 mval[-1].br_state == mval->br_state)) {
3948 ASSERT(mval->br_startoff ==
3949 mval[-1].br_startoff + mval[-1].br_blockcount);
3950 mval[-1].br_blockcount += mval->br_blockcount;
3951 } else if (*n > 0 &&
3952 mval->br_startblock == DELAYSTARTBLOCK &&
3953 mval[-1].br_startblock == DELAYSTARTBLOCK &&
3954 mval->br_startoff ==
3955 mval[-1].br_startoff + mval[-1].br_blockcount) {
3956 mval[-1].br_blockcount += mval->br_blockcount;
3957 mval[-1].br_state = mval->br_state;
3958 } else if (!((*n == 0) &&
3959 ((mval->br_startoff + mval->br_blockcount) <=
3960 obno))) {
3961 mval++;
3962 (*n)++;
3963 }
3964 *map = mval;
3965 }
3966
3967 /*
3968 * Map file blocks to filesystem blocks without allocation.
3969 */
3970 int
xfs_bmapi_read(struct xfs_inode * ip,xfs_fileoff_t bno,xfs_filblks_t len,struct xfs_bmbt_irec * mval,int * nmap,int flags)3971 xfs_bmapi_read(
3972 struct xfs_inode *ip,
3973 xfs_fileoff_t bno,
3974 xfs_filblks_t len,
3975 struct xfs_bmbt_irec *mval,
3976 int *nmap,
3977 int flags)
3978 {
3979 struct xfs_mount *mp = ip->i_mount;
3980 struct xfs_ifork *ifp;
3981 struct xfs_bmbt_irec got;
3982 xfs_fileoff_t obno;
3983 xfs_fileoff_t end;
3984 xfs_extnum_t idx;
3985 int error;
3986 bool eof = false;
3987 int n = 0;
3988 int whichfork = xfs_bmapi_whichfork(flags);
3989
3990 ASSERT(*nmap >= 1);
3991 ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
3992 XFS_BMAPI_IGSTATE|XFS_BMAPI_COWFORK)));
3993 ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL));
3994
3995 if (unlikely(XFS_TEST_ERROR(
3996 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
3997 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
3998 mp, XFS_ERRTAG_BMAPIFORMAT))) {
3999 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
4000 return -EFSCORRUPTED;
4001 }
4002
4003 if (XFS_FORCED_SHUTDOWN(mp))
4004 return -EIO;
4005
4006 XFS_STATS_INC(mp, xs_blk_mapr);
4007
4008 ifp = XFS_IFORK_PTR(ip, whichfork);
4009 if (!ifp) {
4010 /* No CoW fork? Return a hole. */
4011 if (whichfork == XFS_COW_FORK) {
4012 mval->br_startoff = bno;
4013 mval->br_startblock = HOLESTARTBLOCK;
4014 mval->br_blockcount = len;
4015 mval->br_state = XFS_EXT_NORM;
4016 *nmap = 1;
4017 return 0;
4018 }
4019
4020 /*
4021 * A missing attr ifork implies that the inode says we're in
4022 * extents or btree format but failed to pass the inode fork
4023 * verifier while trying to load it. Treat that as a file
4024 * corruption too.
4025 */
4026 #ifdef DEBUG
4027 xfs_alert(mp, "%s: inode %llu missing fork %d",
4028 __func__, ip->i_ino, whichfork);
4029 #endif /* DEBUG */
4030 return -EFSCORRUPTED;
4031 }
4032
4033 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4034 error = xfs_iread_extents(NULL, ip, whichfork);
4035 if (error)
4036 return error;
4037 }
4038
4039 if (!xfs_iext_lookup_extent(ip, ifp, bno, &idx, &got))
4040 eof = true;
4041 end = bno + len;
4042 obno = bno;
4043
4044 while (bno < end && n < *nmap) {
4045 /* Reading past eof, act as though there's a hole up to end. */
4046 if (eof)
4047 got.br_startoff = end;
4048 if (got.br_startoff > bno) {
4049 /* Reading in a hole. */
4050 mval->br_startoff = bno;
4051 mval->br_startblock = HOLESTARTBLOCK;
4052 mval->br_blockcount =
4053 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4054 mval->br_state = XFS_EXT_NORM;
4055 bno += mval->br_blockcount;
4056 len -= mval->br_blockcount;
4057 mval++;
4058 n++;
4059 continue;
4060 }
4061
4062 /* set up the extent map to return. */
4063 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4064 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4065
4066 /* If we're done, stop now. */
4067 if (bno >= end || n >= *nmap)
4068 break;
4069
4070 /* Else go on to the next record. */
4071 if (!xfs_iext_get_extent(ifp, ++idx, &got))
4072 eof = true;
4073 }
4074 *nmap = n;
4075 return 0;
4076 }
4077
4078 /*
4079 * Add a delayed allocation extent to an inode. Blocks are reserved from the
4080 * global pool and the extent inserted into the inode in-core extent tree.
4081 *
4082 * On entry, got refers to the first extent beyond the offset of the extent to
4083 * allocate or eof is specified if no such extent exists. On return, got refers
4084 * to the extent record that was inserted to the inode fork.
4085 *
4086 * Note that the allocated extent may have been merged with contiguous extents
4087 * during insertion into the inode fork. Thus, got does not reflect the current
4088 * state of the inode fork on return. If necessary, the caller can use lastx to
4089 * look up the updated record in the inode fork.
4090 */
4091 int
xfs_bmapi_reserve_delalloc(struct xfs_inode * ip,int whichfork,xfs_fileoff_t off,xfs_filblks_t len,xfs_filblks_t prealloc,struct xfs_bmbt_irec * got,xfs_extnum_t * lastx,int eof)4092 xfs_bmapi_reserve_delalloc(
4093 struct xfs_inode *ip,
4094 int whichfork,
4095 xfs_fileoff_t off,
4096 xfs_filblks_t len,
4097 xfs_filblks_t prealloc,
4098 struct xfs_bmbt_irec *got,
4099 xfs_extnum_t *lastx,
4100 int eof)
4101 {
4102 struct xfs_mount *mp = ip->i_mount;
4103 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
4104 xfs_extlen_t alen;
4105 xfs_extlen_t indlen;
4106 char rt = XFS_IS_REALTIME_INODE(ip);
4107 xfs_extlen_t extsz;
4108 int error;
4109 xfs_fileoff_t aoff = off;
4110
4111 /*
4112 * Cap the alloc length. Keep track of prealloc so we know whether to
4113 * tag the inode before we return.
4114 */
4115 alen = XFS_FILBLKS_MIN(len + prealloc, MAXEXTLEN);
4116 if (!eof)
4117 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4118 if (prealloc && alen >= len)
4119 prealloc = alen - len;
4120
4121 /* Figure out the extent size, adjust alen */
4122 if (whichfork == XFS_COW_FORK)
4123 extsz = xfs_get_cowextsz_hint(ip);
4124 else
4125 extsz = xfs_get_extsz_hint(ip);
4126 if (extsz) {
4127 struct xfs_bmbt_irec prev;
4128
4129 if (!xfs_iext_get_extent(ifp, *lastx - 1, &prev))
4130 prev.br_startoff = NULLFILEOFF;
4131
4132 error = xfs_bmap_extsize_align(mp, got, &prev, extsz, rt, eof,
4133 1, 0, &aoff, &alen);
4134 ASSERT(!error);
4135 }
4136
4137 if (rt)
4138 extsz = alen / mp->m_sb.sb_rextsize;
4139
4140 /*
4141 * Make a transaction-less quota reservation for delayed allocation
4142 * blocks. This number gets adjusted later. We return if we haven't
4143 * allocated blocks already inside this loop.
4144 */
4145 error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4146 rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4147 if (error)
4148 return error;
4149
4150 /*
4151 * Split changing sb for alen and indlen since they could be coming
4152 * from different places.
4153 */
4154 indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4155 ASSERT(indlen > 0);
4156
4157 if (rt) {
4158 error = xfs_mod_frextents(mp, -((int64_t)extsz));
4159 } else {
4160 error = xfs_mod_fdblocks(mp, -((int64_t)alen), false);
4161 }
4162
4163 if (error)
4164 goto out_unreserve_quota;
4165
4166 error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false);
4167 if (error)
4168 goto out_unreserve_blocks;
4169
4170
4171 ip->i_delayed_blks += alen;
4172
4173 got->br_startoff = aoff;
4174 got->br_startblock = nullstartblock(indlen);
4175 got->br_blockcount = alen;
4176 got->br_state = XFS_EXT_NORM;
4177
4178 xfs_bmap_add_extent_hole_delay(ip, whichfork, lastx, got);
4179
4180 /*
4181 * Tag the inode if blocks were preallocated. Note that COW fork
4182 * preallocation can occur at the start or end of the extent, even when
4183 * prealloc == 0, so we must also check the aligned offset and length.
4184 */
4185 if (whichfork == XFS_DATA_FORK && prealloc)
4186 xfs_inode_set_eofblocks_tag(ip);
4187 if (whichfork == XFS_COW_FORK && (prealloc || aoff < off || alen > len))
4188 xfs_inode_set_cowblocks_tag(ip);
4189
4190 return 0;
4191
4192 out_unreserve_blocks:
4193 if (rt)
4194 xfs_mod_frextents(mp, extsz);
4195 else
4196 xfs_mod_fdblocks(mp, alen, false);
4197 out_unreserve_quota:
4198 if (XFS_IS_QUOTA_ON(mp))
4199 xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ?
4200 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4201 return error;
4202 }
4203
4204 static int
xfs_bmapi_allocate(struct xfs_bmalloca * bma)4205 xfs_bmapi_allocate(
4206 struct xfs_bmalloca *bma)
4207 {
4208 struct xfs_mount *mp = bma->ip->i_mount;
4209 int whichfork = xfs_bmapi_whichfork(bma->flags);
4210 struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4211 int tmp_logflags = 0;
4212 int error;
4213
4214 ASSERT(bma->length > 0);
4215
4216 /*
4217 * For the wasdelay case, we could also just allocate the stuff asked
4218 * for in this bmap call but that wouldn't be as good.
4219 */
4220 if (bma->wasdel) {
4221 bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4222 bma->offset = bma->got.br_startoff;
4223 if (bma->idx) {
4224 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4225 &bma->prev);
4226 }
4227 } else {
4228 bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4229 if (!bma->eof)
4230 bma->length = XFS_FILBLKS_MIN(bma->length,
4231 bma->got.br_startoff - bma->offset);
4232 }
4233
4234 /*
4235 * Set the data type being allocated. For the data fork, the first data
4236 * in the file is treated differently to all other allocations. For the
4237 * attribute fork, we only need to ensure the allocated range is not on
4238 * the busy list.
4239 */
4240 if (!(bma->flags & XFS_BMAPI_METADATA)) {
4241 bma->datatype = XFS_ALLOC_NOBUSY;
4242 if (whichfork == XFS_DATA_FORK) {
4243 if (bma->offset == 0)
4244 bma->datatype |= XFS_ALLOC_INITIAL_USER_DATA;
4245 else
4246 bma->datatype |= XFS_ALLOC_USERDATA;
4247 }
4248 if (bma->flags & XFS_BMAPI_ZERO)
4249 bma->datatype |= XFS_ALLOC_USERDATA_ZERO;
4250 }
4251
4252 bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4253
4254 /*
4255 * Only want to do the alignment at the eof if it is userdata and
4256 * allocation length is larger than a stripe unit.
4257 */
4258 if (mp->m_dalign && bma->length >= mp->m_dalign &&
4259 !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4260 error = xfs_bmap_isaeof(bma, whichfork);
4261 if (error)
4262 return error;
4263 }
4264
4265 error = xfs_bmap_alloc(bma);
4266 if (error)
4267 return error;
4268
4269 if (bma->cur)
4270 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4271 if (bma->blkno == NULLFSBLOCK)
4272 return 0;
4273 if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4274 bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4275 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4276 bma->cur->bc_private.b.dfops = bma->dfops;
4277 }
4278 /*
4279 * Bump the number of extents we've allocated
4280 * in this call.
4281 */
4282 bma->nallocs++;
4283
4284 if (bma->cur)
4285 bma->cur->bc_private.b.flags =
4286 bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4287
4288 bma->got.br_startoff = bma->offset;
4289 bma->got.br_startblock = bma->blkno;
4290 bma->got.br_blockcount = bma->length;
4291 bma->got.br_state = XFS_EXT_NORM;
4292
4293 /*
4294 * In the data fork, a wasdelay extent has been initialized, so
4295 * shouldn't be flagged as unwritten.
4296 *
4297 * For the cow fork, however, we convert delalloc reservations
4298 * (extents allocated for speculative preallocation) to
4299 * allocated unwritten extents, and only convert the unwritten
4300 * extents to real extents when we're about to write the data.
4301 */
4302 if ((!bma->wasdel || (bma->flags & XFS_BMAPI_COWFORK)) &&
4303 (bma->flags & XFS_BMAPI_PREALLOC) &&
4304 xfs_sb_version_hasextflgbit(&mp->m_sb))
4305 bma->got.br_state = XFS_EXT_UNWRITTEN;
4306
4307 if (bma->wasdel)
4308 error = xfs_bmap_add_extent_delay_real(bma, whichfork);
4309 else
4310 error = xfs_bmap_add_extent_hole_real(bma->tp, bma->ip,
4311 whichfork, &bma->idx, &bma->cur, &bma->got,
4312 bma->firstblock, bma->dfops, &bma->logflags);
4313
4314 bma->logflags |= tmp_logflags;
4315 if (error)
4316 return error;
4317
4318 /*
4319 * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4320 * or xfs_bmap_add_extent_hole_real might have merged it into one of
4321 * the neighbouring ones.
4322 */
4323 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4324
4325 ASSERT(bma->got.br_startoff <= bma->offset);
4326 ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4327 bma->offset + bma->length);
4328 ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4329 bma->got.br_state == XFS_EXT_UNWRITTEN);
4330 return 0;
4331 }
4332
4333 STATIC int
xfs_bmapi_convert_unwritten(struct xfs_bmalloca * bma,struct xfs_bmbt_irec * mval,xfs_filblks_t len,int flags)4334 xfs_bmapi_convert_unwritten(
4335 struct xfs_bmalloca *bma,
4336 struct xfs_bmbt_irec *mval,
4337 xfs_filblks_t len,
4338 int flags)
4339 {
4340 int whichfork = xfs_bmapi_whichfork(flags);
4341 struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4342 int tmp_logflags = 0;
4343 int error;
4344
4345 /* check if we need to do unwritten->real conversion */
4346 if (mval->br_state == XFS_EXT_UNWRITTEN &&
4347 (flags & XFS_BMAPI_PREALLOC))
4348 return 0;
4349
4350 /* check if we need to do real->unwritten conversion */
4351 if (mval->br_state == XFS_EXT_NORM &&
4352 (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4353 (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4354 return 0;
4355
4356 /*
4357 * Modify (by adding) the state flag, if writing.
4358 */
4359 ASSERT(mval->br_blockcount <= len);
4360 if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4361 bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4362 bma->ip, whichfork);
4363 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4364 bma->cur->bc_private.b.dfops = bma->dfops;
4365 }
4366 mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4367 ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4368
4369 /*
4370 * Before insertion into the bmbt, zero the range being converted
4371 * if required.
4372 */
4373 if (flags & XFS_BMAPI_ZERO) {
4374 error = xfs_zero_extent(bma->ip, mval->br_startblock,
4375 mval->br_blockcount);
4376 if (error)
4377 return error;
4378 }
4379
4380 error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, whichfork,
4381 &bma->idx, &bma->cur, mval, bma->firstblock, bma->dfops,
4382 &tmp_logflags);
4383 /*
4384 * Log the inode core unconditionally in the unwritten extent conversion
4385 * path because the conversion might not have done so (e.g., if the
4386 * extent count hasn't changed). We need to make sure the inode is dirty
4387 * in the transaction for the sake of fsync(), even if nothing has
4388 * changed, because fsync() will not force the log for this transaction
4389 * unless it sees the inode pinned.
4390 *
4391 * Note: If we're only converting cow fork extents, there aren't
4392 * any on-disk updates to make, so we don't need to log anything.
4393 */
4394 if (whichfork != XFS_COW_FORK)
4395 bma->logflags |= tmp_logflags | XFS_ILOG_CORE;
4396 if (error)
4397 return error;
4398
4399 /*
4400 * Update our extent pointer, given that
4401 * xfs_bmap_add_extent_unwritten_real might have merged it into one
4402 * of the neighbouring ones.
4403 */
4404 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4405
4406 /*
4407 * We may have combined previously unwritten space with written space,
4408 * so generate another request.
4409 */
4410 if (mval->br_blockcount < len)
4411 return -EAGAIN;
4412 return 0;
4413 }
4414
4415 /*
4416 * Map file blocks to filesystem blocks, and allocate blocks or convert the
4417 * extent state if necessary. Details behaviour is controlled by the flags
4418 * parameter. Only allocates blocks from a single allocation group, to avoid
4419 * locking problems.
4420 *
4421 * The returned value in "firstblock" from the first call in a transaction
4422 * must be remembered and presented to subsequent calls in "firstblock".
4423 * An upper bound for the number of blocks to be allocated is supplied to
4424 * the first call in "total"; if no allocation group has that many free
4425 * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4426 */
4427 int
xfs_bmapi_write(struct xfs_trans * tp,struct xfs_inode * ip,xfs_fileoff_t bno,xfs_filblks_t len,int flags,xfs_fsblock_t * firstblock,xfs_extlen_t total,struct xfs_bmbt_irec * mval,int * nmap,struct xfs_defer_ops * dfops)4428 xfs_bmapi_write(
4429 struct xfs_trans *tp, /* transaction pointer */
4430 struct xfs_inode *ip, /* incore inode */
4431 xfs_fileoff_t bno, /* starting file offs. mapped */
4432 xfs_filblks_t len, /* length to map in file */
4433 int flags, /* XFS_BMAPI_... */
4434 xfs_fsblock_t *firstblock, /* first allocated block
4435 controls a.g. for allocs */
4436 xfs_extlen_t total, /* total blocks needed */
4437 struct xfs_bmbt_irec *mval, /* output: map values */
4438 int *nmap, /* i/o: mval size/count */
4439 struct xfs_defer_ops *dfops) /* i/o: list extents to free */
4440 {
4441 struct xfs_mount *mp = ip->i_mount;
4442 struct xfs_ifork *ifp;
4443 struct xfs_bmalloca bma = { NULL }; /* args for xfs_bmap_alloc */
4444 xfs_fileoff_t end; /* end of mapped file region */
4445 bool eof = false; /* after the end of extents */
4446 int error; /* error return */
4447 int n; /* current extent index */
4448 xfs_fileoff_t obno; /* old block number (offset) */
4449 int whichfork; /* data or attr fork */
4450
4451 #ifdef DEBUG
4452 xfs_fileoff_t orig_bno; /* original block number value */
4453 int orig_flags; /* original flags arg value */
4454 xfs_filblks_t orig_len; /* original value of len arg */
4455 struct xfs_bmbt_irec *orig_mval; /* original value of mval */
4456 int orig_nmap; /* original value of *nmap */
4457
4458 orig_bno = bno;
4459 orig_len = len;
4460 orig_flags = flags;
4461 orig_mval = mval;
4462 orig_nmap = *nmap;
4463 #endif
4464 whichfork = xfs_bmapi_whichfork(flags);
4465
4466 ASSERT(*nmap >= 1);
4467 ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4468 ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4469 ASSERT(tp != NULL ||
4470 (flags & (XFS_BMAPI_CONVERT | XFS_BMAPI_COWFORK)) ==
4471 (XFS_BMAPI_CONVERT | XFS_BMAPI_COWFORK));
4472 ASSERT(len > 0);
4473 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL);
4474 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4475 ASSERT(!(flags & XFS_BMAPI_REMAP));
4476
4477 /* zeroing is for currently only for data extents, not metadata */
4478 ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) !=
4479 (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO));
4480 /*
4481 * we can allocate unwritten extents or pre-zero allocated blocks,
4482 * but it makes no sense to do both at once. This would result in
4483 * zeroing the unwritten extent twice, but it still being an
4484 * unwritten extent....
4485 */
4486 ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) !=
4487 (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO));
4488
4489 if (unlikely(XFS_TEST_ERROR(
4490 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4491 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4492 mp, XFS_ERRTAG_BMAPIFORMAT))) {
4493 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4494 return -EFSCORRUPTED;
4495 }
4496
4497 if (XFS_FORCED_SHUTDOWN(mp))
4498 return -EIO;
4499
4500 ifp = XFS_IFORK_PTR(ip, whichfork);
4501
4502 XFS_STATS_INC(mp, xs_blk_mapw);
4503
4504 if (*firstblock == NULLFSBLOCK) {
4505 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4506 bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4507 else
4508 bma.minleft = 1;
4509 } else {
4510 bma.minleft = 0;
4511 }
4512
4513 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4514 error = xfs_iread_extents(tp, ip, whichfork);
4515 if (error)
4516 goto error0;
4517 }
4518
4519 n = 0;
4520 end = bno + len;
4521 obno = bno;
4522
4523 if (!xfs_iext_lookup_extent(ip, ifp, bno, &bma.idx, &bma.got))
4524 eof = true;
4525 if (!xfs_iext_get_extent(ifp, bma.idx - 1, &bma.prev))
4526 bma.prev.br_startoff = NULLFILEOFF;
4527 bma.tp = tp;
4528 bma.ip = ip;
4529 bma.total = total;
4530 bma.datatype = 0;
4531 bma.dfops = dfops;
4532 bma.firstblock = firstblock;
4533
4534 while (bno < end && n < *nmap) {
4535 bool need_alloc = false, wasdelay = false;
4536
4537 /* in hole or beyoned EOF? */
4538 if (eof || bma.got.br_startoff > bno) {
4539 if (flags & XFS_BMAPI_DELALLOC) {
4540 /*
4541 * For the COW fork we can reasonably get a
4542 * request for converting an extent that races
4543 * with other threads already having converted
4544 * part of it, as there converting COW to
4545 * regular blocks is not protected using the
4546 * IOLOCK.
4547 */
4548 ASSERT(flags & XFS_BMAPI_COWFORK);
4549 if (!(flags & XFS_BMAPI_COWFORK)) {
4550 error = -EIO;
4551 goto error0;
4552 }
4553
4554 if (eof || bno >= end)
4555 break;
4556 } else {
4557 need_alloc = true;
4558 }
4559 } else if (isnullstartblock(bma.got.br_startblock)) {
4560 wasdelay = true;
4561 }
4562
4563 /*
4564 * First, deal with the hole before the allocated space
4565 * that we found, if any.
4566 */
4567 if (need_alloc || wasdelay) {
4568 bma.eof = eof;
4569 bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4570 bma.wasdel = wasdelay;
4571 bma.offset = bno;
4572 bma.flags = flags;
4573
4574 /*
4575 * There's a 32/64 bit type mismatch between the
4576 * allocation length request (which can be 64 bits in
4577 * length) and the bma length request, which is
4578 * xfs_extlen_t and therefore 32 bits. Hence we have to
4579 * check for 32-bit overflows and handle them here.
4580 */
4581 if (len > (xfs_filblks_t)MAXEXTLEN)
4582 bma.length = MAXEXTLEN;
4583 else
4584 bma.length = len;
4585
4586 ASSERT(len > 0);
4587 ASSERT(bma.length > 0);
4588 error = xfs_bmapi_allocate(&bma);
4589 if (error)
4590 goto error0;
4591 if (bma.blkno == NULLFSBLOCK)
4592 break;
4593
4594 /*
4595 * If this is a CoW allocation, record the data in
4596 * the refcount btree for orphan recovery.
4597 */
4598 if (whichfork == XFS_COW_FORK) {
4599 error = xfs_refcount_alloc_cow_extent(mp, dfops,
4600 bma.blkno, bma.length);
4601 if (error)
4602 goto error0;
4603 }
4604 }
4605
4606 /* Deal with the allocated space we found. */
4607 xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4608 end, n, flags);
4609
4610 /* Execute unwritten extent conversion if necessary */
4611 error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4612 if (error == -EAGAIN)
4613 continue;
4614 if (error)
4615 goto error0;
4616
4617 /* update the extent map to return */
4618 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4619
4620 /*
4621 * If we're done, stop now. Stop when we've allocated
4622 * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise
4623 * the transaction may get too big.
4624 */
4625 if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4626 break;
4627
4628 /* Else go on to the next record. */
4629 bma.prev = bma.got;
4630 if (!xfs_iext_get_extent(ifp, ++bma.idx, &bma.got))
4631 eof = true;
4632 }
4633 *nmap = n;
4634
4635 /*
4636 * Transform from btree to extents, give it cur.
4637 */
4638 if (xfs_bmap_wants_extents(ip, whichfork)) {
4639 int tmp_logflags = 0;
4640
4641 ASSERT(bma.cur);
4642 error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
4643 &tmp_logflags, whichfork);
4644 bma.logflags |= tmp_logflags;
4645 if (error)
4646 goto error0;
4647 }
4648
4649 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4650 XFS_IFORK_NEXTENTS(ip, whichfork) >
4651 XFS_IFORK_MAXEXT(ip, whichfork));
4652 error = 0;
4653 error0:
4654 /*
4655 * Log everything. Do this after conversion, there's no point in
4656 * logging the extent records if we've converted to btree format.
4657 */
4658 if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
4659 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4660 bma.logflags &= ~xfs_ilog_fext(whichfork);
4661 else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
4662 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4663 bma.logflags &= ~xfs_ilog_fbroot(whichfork);
4664 /*
4665 * Log whatever the flags say, even if error. Otherwise we might miss
4666 * detecting a case where the data is changed, there's an error,
4667 * and it's not logged so we don't shutdown when we should.
4668 */
4669 if (bma.logflags)
4670 xfs_trans_log_inode(tp, ip, bma.logflags);
4671
4672 if (bma.cur) {
4673 if (!error) {
4674 ASSERT(*firstblock == NULLFSBLOCK ||
4675 XFS_FSB_TO_AGNO(mp, *firstblock) <=
4676 XFS_FSB_TO_AGNO(mp,
4677 bma.cur->bc_private.b.firstblock));
4678 *firstblock = bma.cur->bc_private.b.firstblock;
4679 }
4680 xfs_btree_del_cursor(bma.cur,
4681 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4682 }
4683 if (!error)
4684 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4685 orig_nmap, *nmap);
4686 return error;
4687 }
4688
4689 static int
xfs_bmapi_remap(struct xfs_trans * tp,struct xfs_inode * ip,xfs_fileoff_t bno,xfs_filblks_t len,xfs_fsblock_t startblock,struct xfs_defer_ops * dfops)4690 xfs_bmapi_remap(
4691 struct xfs_trans *tp,
4692 struct xfs_inode *ip,
4693 xfs_fileoff_t bno,
4694 xfs_filblks_t len,
4695 xfs_fsblock_t startblock,
4696 struct xfs_defer_ops *dfops)
4697 {
4698 struct xfs_mount *mp = ip->i_mount;
4699 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4700 struct xfs_btree_cur *cur = NULL;
4701 xfs_fsblock_t firstblock = NULLFSBLOCK;
4702 struct xfs_bmbt_irec got;
4703 xfs_extnum_t idx;
4704 int logflags = 0, error;
4705
4706 ASSERT(len > 0);
4707 ASSERT(len <= (xfs_filblks_t)MAXEXTLEN);
4708 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4709
4710 if (unlikely(XFS_TEST_ERROR(
4711 (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4712 XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4713 mp, XFS_ERRTAG_BMAPIFORMAT))) {
4714 XFS_ERROR_REPORT("xfs_bmapi_remap", XFS_ERRLEVEL_LOW, mp);
4715 return -EFSCORRUPTED;
4716 }
4717
4718 if (XFS_FORCED_SHUTDOWN(mp))
4719 return -EIO;
4720
4721 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4722 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4723 if (error)
4724 return error;
4725 }
4726
4727 if (xfs_iext_lookup_extent(ip, ifp, bno, &idx, &got)) {
4728 /* make sure we only reflink into a hole. */
4729 ASSERT(got.br_startoff > bno);
4730 ASSERT(got.br_startoff - bno >= len);
4731 }
4732
4733 ip->i_d.di_nblocks += len;
4734 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4735
4736 if (ifp->if_flags & XFS_IFBROOT) {
4737 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
4738 cur->bc_private.b.firstblock = firstblock;
4739 cur->bc_private.b.dfops = dfops;
4740 cur->bc_private.b.flags = 0;
4741 }
4742
4743 got.br_startoff = bno;
4744 got.br_startblock = startblock;
4745 got.br_blockcount = len;
4746 got.br_state = XFS_EXT_NORM;
4747
4748 error = xfs_bmap_add_extent_hole_real(tp, ip, XFS_DATA_FORK, &idx, &cur,
4749 &got, &firstblock, dfops, &logflags);
4750 if (error)
4751 goto error0;
4752
4753 if (xfs_bmap_wants_extents(ip, XFS_DATA_FORK)) {
4754 int tmp_logflags = 0;
4755
4756 error = xfs_bmap_btree_to_extents(tp, ip, cur,
4757 &tmp_logflags, XFS_DATA_FORK);
4758 logflags |= tmp_logflags;
4759 }
4760
4761 error0:
4762 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS)
4763 logflags &= ~XFS_ILOG_DEXT;
4764 else if (ip->i_d.di_format != XFS_DINODE_FMT_BTREE)
4765 logflags &= ~XFS_ILOG_DBROOT;
4766
4767 if (logflags)
4768 xfs_trans_log_inode(tp, ip, logflags);
4769 if (cur) {
4770 xfs_btree_del_cursor(cur,
4771 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4772 }
4773 return error;
4774 }
4775
4776 /*
4777 * When a delalloc extent is split (e.g., due to a hole punch), the original
4778 * indlen reservation must be shared across the two new extents that are left
4779 * behind.
4780 *
4781 * Given the original reservation and the worst case indlen for the two new
4782 * extents (as calculated by xfs_bmap_worst_indlen()), split the original
4783 * reservation fairly across the two new extents. If necessary, steal available
4784 * blocks from a deleted extent to make up a reservation deficiency (e.g., if
4785 * ores == 1). The number of stolen blocks is returned. The availability and
4786 * subsequent accounting of stolen blocks is the responsibility of the caller.
4787 */
4788 static xfs_filblks_t
xfs_bmap_split_indlen(xfs_filblks_t ores,xfs_filblks_t * indlen1,xfs_filblks_t * indlen2,xfs_filblks_t avail)4789 xfs_bmap_split_indlen(
4790 xfs_filblks_t ores, /* original res. */
4791 xfs_filblks_t *indlen1, /* ext1 worst indlen */
4792 xfs_filblks_t *indlen2, /* ext2 worst indlen */
4793 xfs_filblks_t avail) /* stealable blocks */
4794 {
4795 xfs_filblks_t len1 = *indlen1;
4796 xfs_filblks_t len2 = *indlen2;
4797 xfs_filblks_t nres = len1 + len2; /* new total res. */
4798 xfs_filblks_t stolen = 0;
4799 xfs_filblks_t resfactor;
4800
4801 /*
4802 * Steal as many blocks as we can to try and satisfy the worst case
4803 * indlen for both new extents.
4804 */
4805 if (ores < nres && avail)
4806 stolen = XFS_FILBLKS_MIN(nres - ores, avail);
4807 ores += stolen;
4808
4809 /* nothing else to do if we've satisfied the new reservation */
4810 if (ores >= nres)
4811 return stolen;
4812
4813 /*
4814 * We can't meet the total required reservation for the two extents.
4815 * Calculate the percent of the overall shortage between both extents
4816 * and apply this percentage to each of the requested indlen values.
4817 * This distributes the shortage fairly and reduces the chances that one
4818 * of the two extents is left with nothing when extents are repeatedly
4819 * split.
4820 */
4821 resfactor = (ores * 100);
4822 do_div(resfactor, nres);
4823 len1 *= resfactor;
4824 do_div(len1, 100);
4825 len2 *= resfactor;
4826 do_div(len2, 100);
4827 ASSERT(len1 + len2 <= ores);
4828 ASSERT(len1 < *indlen1 && len2 < *indlen2);
4829
4830 /*
4831 * Hand out the remainder to each extent. If one of the two reservations
4832 * is zero, we want to make sure that one gets a block first. The loop
4833 * below starts with len1, so hand len2 a block right off the bat if it
4834 * is zero.
4835 */
4836 ores -= (len1 + len2);
4837 ASSERT((*indlen1 - len1) + (*indlen2 - len2) >= ores);
4838 if (ores && !len2 && *indlen2) {
4839 len2++;
4840 ores--;
4841 }
4842 while (ores) {
4843 if (len1 < *indlen1) {
4844 len1++;
4845 ores--;
4846 }
4847 if (!ores)
4848 break;
4849 if (len2 < *indlen2) {
4850 len2++;
4851 ores--;
4852 }
4853 }
4854
4855 *indlen1 = len1;
4856 *indlen2 = len2;
4857
4858 return stolen;
4859 }
4860
4861 int
xfs_bmap_del_extent_delay(struct xfs_inode * ip,int whichfork,xfs_extnum_t * idx,struct xfs_bmbt_irec * got,struct xfs_bmbt_irec * del)4862 xfs_bmap_del_extent_delay(
4863 struct xfs_inode *ip,
4864 int whichfork,
4865 xfs_extnum_t *idx,
4866 struct xfs_bmbt_irec *got,
4867 struct xfs_bmbt_irec *del)
4868 {
4869 struct xfs_mount *mp = ip->i_mount;
4870 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
4871 struct xfs_bmbt_irec new;
4872 int64_t da_old, da_new, da_diff = 0;
4873 xfs_fileoff_t del_endoff, got_endoff;
4874 xfs_filblks_t got_indlen, new_indlen, stolen;
4875 int error = 0, state = 0;
4876 bool isrt;
4877
4878 XFS_STATS_INC(mp, xs_del_exlist);
4879
4880 isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4881 del_endoff = del->br_startoff + del->br_blockcount;
4882 got_endoff = got->br_startoff + got->br_blockcount;
4883 da_old = startblockval(got->br_startblock);
4884 da_new = 0;
4885
4886 ASSERT(*idx >= 0);
4887 ASSERT(*idx <= xfs_iext_count(ifp));
4888 ASSERT(del->br_blockcount > 0);
4889 ASSERT(got->br_startoff <= del->br_startoff);
4890 ASSERT(got_endoff >= del_endoff);
4891
4892 if (isrt) {
4893 uint64_t rtexts = XFS_FSB_TO_B(mp, del->br_blockcount);
4894
4895 do_div(rtexts, mp->m_sb.sb_rextsize);
4896 xfs_mod_frextents(mp, rtexts);
4897 }
4898
4899 /*
4900 * Update the inode delalloc counter now and wait to update the
4901 * sb counters as we might have to borrow some blocks for the
4902 * indirect block accounting.
4903 */
4904 error = xfs_trans_reserve_quota_nblks(NULL, ip,
4905 -((long)del->br_blockcount), 0,
4906 isrt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4907 if (error)
4908 return error;
4909 ip->i_delayed_blks -= del->br_blockcount;
4910
4911 if (whichfork == XFS_COW_FORK)
4912 state |= BMAP_COWFORK;
4913
4914 if (got->br_startoff == del->br_startoff)
4915 state |= BMAP_LEFT_CONTIG;
4916 if (got_endoff == del_endoff)
4917 state |= BMAP_RIGHT_CONTIG;
4918
4919 switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
4920 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
4921 /*
4922 * Matches the whole extent. Delete the entry.
4923 */
4924 xfs_iext_remove(ip, *idx, 1, state);
4925 --*idx;
4926 break;
4927 case BMAP_LEFT_CONTIG:
4928 /*
4929 * Deleting the first part of the extent.
4930 */
4931 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4932 got->br_startoff = del_endoff;
4933 got->br_blockcount -= del->br_blockcount;
4934 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4935 got->br_blockcount), da_old);
4936 got->br_startblock = nullstartblock((int)da_new);
4937 xfs_iext_update_extent(ifp, *idx, got);
4938 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4939 break;
4940 case BMAP_RIGHT_CONTIG:
4941 /*
4942 * Deleting the last part of the extent.
4943 */
4944 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4945 got->br_blockcount = got->br_blockcount - del->br_blockcount;
4946 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4947 got->br_blockcount), da_old);
4948 got->br_startblock = nullstartblock((int)da_new);
4949 xfs_iext_update_extent(ifp, *idx, got);
4950 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4951 break;
4952 case 0:
4953 /*
4954 * Deleting the middle of the extent.
4955 *
4956 * Distribute the original indlen reservation across the two new
4957 * extents. Steal blocks from the deleted extent if necessary.
4958 * Stealing blocks simply fudges the fdblocks accounting below.
4959 * Warn if either of the new indlen reservations is zero as this
4960 * can lead to delalloc problems.
4961 */
4962 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4963
4964 got->br_blockcount = del->br_startoff - got->br_startoff;
4965 got_indlen = xfs_bmap_worst_indlen(ip, got->br_blockcount);
4966
4967 new.br_blockcount = got_endoff - del_endoff;
4968 new_indlen = xfs_bmap_worst_indlen(ip, new.br_blockcount);
4969
4970 WARN_ON_ONCE(!got_indlen || !new_indlen);
4971 stolen = xfs_bmap_split_indlen(da_old, &got_indlen, &new_indlen,
4972 del->br_blockcount);
4973
4974 got->br_startblock = nullstartblock((int)got_indlen);
4975 xfs_iext_update_extent(ifp, *idx, got);
4976 trace_xfs_bmap_post_update(ip, *idx, 0, _THIS_IP_);
4977
4978 new.br_startoff = del_endoff;
4979 new.br_state = got->br_state;
4980 new.br_startblock = nullstartblock((int)new_indlen);
4981
4982 ++*idx;
4983 xfs_iext_insert(ip, *idx, 1, &new, state);
4984
4985 da_new = got_indlen + new_indlen - stolen;
4986 del->br_blockcount -= stolen;
4987 break;
4988 }
4989
4990 ASSERT(da_old >= da_new);
4991 da_diff = da_old - da_new;
4992 if (!isrt)
4993 da_diff += del->br_blockcount;
4994 if (da_diff)
4995 xfs_mod_fdblocks(mp, da_diff, false);
4996 return error;
4997 }
4998
4999 void
xfs_bmap_del_extent_cow(struct xfs_inode * ip,xfs_extnum_t * idx,struct xfs_bmbt_irec * got,struct xfs_bmbt_irec * del)5000 xfs_bmap_del_extent_cow(
5001 struct xfs_inode *ip,
5002 xfs_extnum_t *idx,
5003 struct xfs_bmbt_irec *got,
5004 struct xfs_bmbt_irec *del)
5005 {
5006 struct xfs_mount *mp = ip->i_mount;
5007 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
5008 struct xfs_bmbt_irec new;
5009 xfs_fileoff_t del_endoff, got_endoff;
5010 int state = BMAP_COWFORK;
5011
5012 XFS_STATS_INC(mp, xs_del_exlist);
5013
5014 del_endoff = del->br_startoff + del->br_blockcount;
5015 got_endoff = got->br_startoff + got->br_blockcount;
5016
5017 ASSERT(*idx >= 0);
5018 ASSERT(*idx <= xfs_iext_count(ifp));
5019 ASSERT(del->br_blockcount > 0);
5020 ASSERT(got->br_startoff <= del->br_startoff);
5021 ASSERT(got_endoff >= del_endoff);
5022 ASSERT(!isnullstartblock(got->br_startblock));
5023
5024 if (got->br_startoff == del->br_startoff)
5025 state |= BMAP_LEFT_CONTIG;
5026 if (got_endoff == del_endoff)
5027 state |= BMAP_RIGHT_CONTIG;
5028
5029 switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
5030 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
5031 /*
5032 * Matches the whole extent. Delete the entry.
5033 */
5034 xfs_iext_remove(ip, *idx, 1, state);
5035 --*idx;
5036 break;
5037 case BMAP_LEFT_CONTIG:
5038 /*
5039 * Deleting the first part of the extent.
5040 */
5041 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5042 got->br_startoff = del_endoff;
5043 got->br_blockcount -= del->br_blockcount;
5044 got->br_startblock = del->br_startblock + del->br_blockcount;
5045 xfs_iext_update_extent(ifp, *idx, got);
5046 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5047 break;
5048 case BMAP_RIGHT_CONTIG:
5049 /*
5050 * Deleting the last part of the extent.
5051 */
5052 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5053 got->br_blockcount -= del->br_blockcount;
5054 xfs_iext_update_extent(ifp, *idx, got);
5055 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5056 break;
5057 case 0:
5058 /*
5059 * Deleting the middle of the extent.
5060 */
5061 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5062 got->br_blockcount = del->br_startoff - got->br_startoff;
5063 xfs_iext_update_extent(ifp, *idx, got);
5064 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5065
5066 new.br_startoff = del_endoff;
5067 new.br_blockcount = got_endoff - del_endoff;
5068 new.br_state = got->br_state;
5069 new.br_startblock = del->br_startblock + del->br_blockcount;
5070
5071 ++*idx;
5072 xfs_iext_insert(ip, *idx, 1, &new, state);
5073 break;
5074 }
5075 }
5076
5077 /*
5078 * Called by xfs_bmapi to update file extent records and the btree
5079 * after removing space (or undoing a delayed allocation).
5080 */
5081 STATIC int /* error */
xfs_bmap_del_extent(xfs_inode_t * ip,xfs_trans_t * tp,xfs_extnum_t * idx,struct xfs_defer_ops * dfops,xfs_btree_cur_t * cur,xfs_bmbt_irec_t * del,int * logflagsp,int whichfork,int bflags)5082 xfs_bmap_del_extent(
5083 xfs_inode_t *ip, /* incore inode pointer */
5084 xfs_trans_t *tp, /* current transaction pointer */
5085 xfs_extnum_t *idx, /* extent number to update/delete */
5086 struct xfs_defer_ops *dfops, /* list of extents to be freed */
5087 xfs_btree_cur_t *cur, /* if null, not a btree */
5088 xfs_bmbt_irec_t *del, /* data to remove from extents */
5089 int *logflagsp, /* inode logging flags */
5090 int whichfork, /* data or attr fork */
5091 int bflags) /* bmapi flags */
5092 {
5093 xfs_filblks_t da_new; /* new delay-alloc indirect blocks */
5094 xfs_filblks_t da_old; /* old delay-alloc indirect blocks */
5095 xfs_fsblock_t del_endblock=0; /* first block past del */
5096 xfs_fileoff_t del_endoff; /* first offset past del */
5097 int delay; /* current block is delayed allocated */
5098 int do_fx; /* free extent at end of routine */
5099 xfs_bmbt_rec_host_t *ep; /* current extent entry pointer */
5100 int error; /* error return value */
5101 int flags; /* inode logging flags */
5102 xfs_bmbt_irec_t got; /* current extent entry */
5103 xfs_fileoff_t got_endoff; /* first offset past got */
5104 int i; /* temp state */
5105 xfs_ifork_t *ifp; /* inode fork pointer */
5106 xfs_mount_t *mp; /* mount structure */
5107 xfs_filblks_t nblks; /* quota/sb block count */
5108 xfs_bmbt_irec_t new; /* new record to be inserted */
5109 /* REFERENCED */
5110 uint qfield; /* quota field to update */
5111 xfs_filblks_t temp; /* for indirect length calculations */
5112 xfs_filblks_t temp2; /* for indirect length calculations */
5113 int state = 0;
5114
5115 mp = ip->i_mount;
5116 XFS_STATS_INC(mp, xs_del_exlist);
5117
5118 if (whichfork == XFS_ATTR_FORK)
5119 state |= BMAP_ATTRFORK;
5120 else if (whichfork == XFS_COW_FORK)
5121 state |= BMAP_COWFORK;
5122
5123 ifp = XFS_IFORK_PTR(ip, whichfork);
5124 ASSERT((*idx >= 0) && (*idx < xfs_iext_count(ifp)));
5125 ASSERT(del->br_blockcount > 0);
5126 ep = xfs_iext_get_ext(ifp, *idx);
5127 xfs_bmbt_get_all(ep, &got);
5128 ASSERT(got.br_startoff <= del->br_startoff);
5129 del_endoff = del->br_startoff + del->br_blockcount;
5130 got_endoff = got.br_startoff + got.br_blockcount;
5131 ASSERT(got_endoff >= del_endoff);
5132 delay = isnullstartblock(got.br_startblock);
5133 ASSERT(isnullstartblock(del->br_startblock) == delay);
5134 flags = 0;
5135 qfield = 0;
5136 error = 0;
5137 /*
5138 * If deleting a real allocation, must free up the disk space.
5139 */
5140 if (!delay) {
5141 flags = XFS_ILOG_CORE;
5142 /*
5143 * Realtime allocation. Free it and record di_nblocks update.
5144 */
5145 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
5146 xfs_fsblock_t bno;
5147 xfs_filblks_t len;
5148
5149 ASSERT(do_mod(del->br_blockcount,
5150 mp->m_sb.sb_rextsize) == 0);
5151 ASSERT(do_mod(del->br_startblock,
5152 mp->m_sb.sb_rextsize) == 0);
5153 bno = del->br_startblock;
5154 len = del->br_blockcount;
5155 do_div(bno, mp->m_sb.sb_rextsize);
5156 do_div(len, mp->m_sb.sb_rextsize);
5157 error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
5158 if (error)
5159 goto done;
5160 do_fx = 0;
5161 nblks = len * mp->m_sb.sb_rextsize;
5162 qfield = XFS_TRANS_DQ_RTBCOUNT;
5163 }
5164 /*
5165 * Ordinary allocation.
5166 */
5167 else {
5168 do_fx = 1;
5169 nblks = del->br_blockcount;
5170 qfield = XFS_TRANS_DQ_BCOUNT;
5171 }
5172 /*
5173 * Set up del_endblock and cur for later.
5174 */
5175 del_endblock = del->br_startblock + del->br_blockcount;
5176 if (cur) {
5177 if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
5178 got.br_startblock, got.br_blockcount,
5179 &i)))
5180 goto done;
5181 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5182 }
5183 da_old = da_new = 0;
5184 } else {
5185 da_old = startblockval(got.br_startblock);
5186 da_new = 0;
5187 nblks = 0;
5188 do_fx = 0;
5189 }
5190
5191 /*
5192 * Set flag value to use in switch statement.
5193 * Left-contig is 2, right-contig is 1.
5194 */
5195 switch (((got.br_startoff == del->br_startoff) << 1) |
5196 (got_endoff == del_endoff)) {
5197 case 3:
5198 /*
5199 * Matches the whole extent. Delete the entry.
5200 */
5201 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5202 xfs_iext_remove(ip, *idx, 1,
5203 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
5204 --*idx;
5205 if (delay)
5206 break;
5207
5208 XFS_IFORK_NEXT_SET(ip, whichfork,
5209 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5210 flags |= XFS_ILOG_CORE;
5211 if (!cur) {
5212 flags |= xfs_ilog_fext(whichfork);
5213 break;
5214 }
5215 if ((error = xfs_btree_delete(cur, &i)))
5216 goto done;
5217 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5218 break;
5219
5220 case 2:
5221 /*
5222 * Deleting the first part of the extent.
5223 */
5224 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5225 xfs_bmbt_set_startoff(ep, del_endoff);
5226 temp = got.br_blockcount - del->br_blockcount;
5227 xfs_bmbt_set_blockcount(ep, temp);
5228 if (delay) {
5229 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
5230 da_old);
5231 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
5232 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5233 da_new = temp;
5234 break;
5235 }
5236 xfs_bmbt_set_startblock(ep, del_endblock);
5237 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5238 if (!cur) {
5239 flags |= xfs_ilog_fext(whichfork);
5240 break;
5241 }
5242 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
5243 got.br_blockcount - del->br_blockcount,
5244 got.br_state)))
5245 goto done;
5246 break;
5247
5248 case 1:
5249 /*
5250 * Deleting the last part of the extent.
5251 */
5252 temp = got.br_blockcount - del->br_blockcount;
5253 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5254 xfs_bmbt_set_blockcount(ep, temp);
5255 if (delay) {
5256 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
5257 da_old);
5258 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
5259 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5260 da_new = temp;
5261 break;
5262 }
5263 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5264 if (!cur) {
5265 flags |= xfs_ilog_fext(whichfork);
5266 break;
5267 }
5268 if ((error = xfs_bmbt_update(cur, got.br_startoff,
5269 got.br_startblock,
5270 got.br_blockcount - del->br_blockcount,
5271 got.br_state)))
5272 goto done;
5273 break;
5274
5275 case 0:
5276 /*
5277 * Deleting the middle of the extent.
5278 */
5279 temp = del->br_startoff - got.br_startoff;
5280 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5281 xfs_bmbt_set_blockcount(ep, temp);
5282 new.br_startoff = del_endoff;
5283 temp2 = got_endoff - del_endoff;
5284 new.br_blockcount = temp2;
5285 new.br_state = got.br_state;
5286 if (!delay) {
5287 new.br_startblock = del_endblock;
5288 flags |= XFS_ILOG_CORE;
5289 if (cur) {
5290 if ((error = xfs_bmbt_update(cur,
5291 got.br_startoff,
5292 got.br_startblock, temp,
5293 got.br_state)))
5294 goto done;
5295 if ((error = xfs_btree_increment(cur, 0, &i)))
5296 goto done;
5297 cur->bc_rec.b = new;
5298 error = xfs_btree_insert(cur, &i);
5299 if (error && error != -ENOSPC)
5300 goto done;
5301 /*
5302 * If get no-space back from btree insert,
5303 * it tried a split, and we have a zero
5304 * block reservation.
5305 * Fix up our state and return the error.
5306 */
5307 if (error == -ENOSPC) {
5308 /*
5309 * Reset the cursor, don't trust
5310 * it after any insert operation.
5311 */
5312 if ((error = xfs_bmbt_lookup_eq(cur,
5313 got.br_startoff,
5314 got.br_startblock,
5315 temp, &i)))
5316 goto done;
5317 XFS_WANT_CORRUPTED_GOTO(mp,
5318 i == 1, done);
5319 /*
5320 * Update the btree record back
5321 * to the original value.
5322 */
5323 if ((error = xfs_bmbt_update(cur,
5324 got.br_startoff,
5325 got.br_startblock,
5326 got.br_blockcount,
5327 got.br_state)))
5328 goto done;
5329 /*
5330 * Reset the extent record back
5331 * to the original value.
5332 */
5333 xfs_bmbt_set_blockcount(ep,
5334 got.br_blockcount);
5335 flags = 0;
5336 error = -ENOSPC;
5337 goto done;
5338 }
5339 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5340 } else
5341 flags |= xfs_ilog_fext(whichfork);
5342 XFS_IFORK_NEXT_SET(ip, whichfork,
5343 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
5344 } else {
5345 xfs_filblks_t stolen;
5346 ASSERT(whichfork == XFS_DATA_FORK);
5347
5348 /*
5349 * Distribute the original indlen reservation across the
5350 * two new extents. Steal blocks from the deleted extent
5351 * if necessary. Stealing blocks simply fudges the
5352 * fdblocks accounting in xfs_bunmapi().
5353 */
5354 temp = xfs_bmap_worst_indlen(ip, got.br_blockcount);
5355 temp2 = xfs_bmap_worst_indlen(ip, new.br_blockcount);
5356 stolen = xfs_bmap_split_indlen(da_old, &temp, &temp2,
5357 del->br_blockcount);
5358 da_new = temp + temp2 - stolen;
5359 del->br_blockcount -= stolen;
5360
5361 /*
5362 * Set the reservation for each extent. Warn if either
5363 * is zero as this can lead to delalloc problems.
5364 */
5365 WARN_ON_ONCE(!temp || !temp2);
5366 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
5367 new.br_startblock = nullstartblock((int)temp2);
5368 }
5369 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5370 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
5371 ++*idx;
5372 break;
5373 }
5374
5375 /* remove reverse mapping */
5376 if (!delay) {
5377 error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, del);
5378 if (error)
5379 goto done;
5380 }
5381
5382 /*
5383 * If we need to, add to list of extents to delete.
5384 */
5385 if (do_fx && !(bflags & XFS_BMAPI_REMAP)) {
5386 if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) {
5387 error = xfs_refcount_decrease_extent(mp, dfops, del);
5388 if (error)
5389 goto done;
5390 } else
5391 xfs_bmap_add_free(mp, dfops, del->br_startblock,
5392 del->br_blockcount, NULL);
5393 }
5394
5395 /*
5396 * Adjust inode # blocks in the file.
5397 */
5398 if (nblks)
5399 ip->i_d.di_nblocks -= nblks;
5400 /*
5401 * Adjust quota data.
5402 */
5403 if (qfield && !(bflags & XFS_BMAPI_REMAP))
5404 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5405
5406 /*
5407 * Account for change in delayed indirect blocks.
5408 * Nothing to do for disk quota accounting here.
5409 */
5410 ASSERT(da_old >= da_new);
5411 if (da_old > da_new)
5412 xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new), false);
5413 done:
5414 *logflagsp = flags;
5415 return error;
5416 }
5417
5418 /*
5419 * Unmap (remove) blocks from a file.
5420 * If nexts is nonzero then the number of extents to remove is limited to
5421 * that value. If not all extents in the block range can be removed then
5422 * *done is set.
5423 */
5424 int /* error */
__xfs_bunmapi(xfs_trans_t * tp,struct xfs_inode * ip,xfs_fileoff_t bno,xfs_filblks_t * rlen,int flags,xfs_extnum_t nexts,xfs_fsblock_t * firstblock,struct xfs_defer_ops * dfops)5425 __xfs_bunmapi(
5426 xfs_trans_t *tp, /* transaction pointer */
5427 struct xfs_inode *ip, /* incore inode */
5428 xfs_fileoff_t bno, /* starting offset to unmap */
5429 xfs_filblks_t *rlen, /* i/o: amount remaining */
5430 int flags, /* misc flags */
5431 xfs_extnum_t nexts, /* number of extents max */
5432 xfs_fsblock_t *firstblock, /* first allocated block
5433 controls a.g. for allocs */
5434 struct xfs_defer_ops *dfops) /* i/o: deferred updates */
5435 {
5436 xfs_btree_cur_t *cur; /* bmap btree cursor */
5437 xfs_bmbt_irec_t del; /* extent being deleted */
5438 int error; /* error return value */
5439 xfs_extnum_t extno; /* extent number in list */
5440 xfs_bmbt_irec_t got; /* current extent record */
5441 xfs_ifork_t *ifp; /* inode fork pointer */
5442 int isrt; /* freeing in rt area */
5443 xfs_extnum_t lastx; /* last extent index used */
5444 int logflags; /* transaction logging flags */
5445 xfs_extlen_t mod; /* rt extent offset */
5446 xfs_mount_t *mp; /* mount structure */
5447 xfs_fileoff_t start; /* first file offset deleted */
5448 int tmp_logflags; /* partial logging flags */
5449 int wasdel; /* was a delayed alloc extent */
5450 int whichfork; /* data or attribute fork */
5451 xfs_fsblock_t sum;
5452 xfs_filblks_t len = *rlen; /* length to unmap in file */
5453 xfs_fileoff_t max_len;
5454 xfs_agnumber_t prev_agno = NULLAGNUMBER, agno;
5455
5456 trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5457
5458 whichfork = xfs_bmapi_whichfork(flags);
5459 ASSERT(whichfork != XFS_COW_FORK);
5460 ifp = XFS_IFORK_PTR(ip, whichfork);
5461 if (unlikely(
5462 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5463 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5464 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5465 ip->i_mount);
5466 return -EFSCORRUPTED;
5467 }
5468 mp = ip->i_mount;
5469 if (XFS_FORCED_SHUTDOWN(mp))
5470 return -EIO;
5471
5472 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5473 ASSERT(len > 0);
5474 ASSERT(nexts >= 0);
5475
5476 /*
5477 * Guesstimate how many blocks we can unmap without running the risk of
5478 * blowing out the transaction with a mix of EFIs and reflink
5479 * adjustments.
5480 */
5481 if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK)
5482 max_len = min(len, xfs_refcount_max_unmap(tp->t_log_res));
5483 else
5484 max_len = len;
5485
5486 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5487 (error = xfs_iread_extents(tp, ip, whichfork)))
5488 return error;
5489 if (xfs_iext_count(ifp) == 0) {
5490 *rlen = 0;
5491 return 0;
5492 }
5493 XFS_STATS_INC(mp, xs_blk_unmap);
5494 isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5495 start = bno;
5496 bno = start + len - 1;
5497
5498 /*
5499 * Check to see if the given block number is past the end of the
5500 * file, back up to the last block if so...
5501 */
5502 if (!xfs_iext_lookup_extent(ip, ifp, bno, &lastx, &got)) {
5503 ASSERT(lastx > 0);
5504 xfs_iext_get_extent(ifp, --lastx, &got);
5505 bno = got.br_startoff + got.br_blockcount - 1;
5506 }
5507
5508 logflags = 0;
5509 if (ifp->if_flags & XFS_IFBROOT) {
5510 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5511 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5512 cur->bc_private.b.firstblock = *firstblock;
5513 cur->bc_private.b.dfops = dfops;
5514 cur->bc_private.b.flags = 0;
5515 } else
5516 cur = NULL;
5517
5518 if (isrt) {
5519 /*
5520 * Synchronize by locking the bitmap inode.
5521 */
5522 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
5523 xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5524 xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
5525 xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL);
5526 }
5527
5528 extno = 0;
5529 while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5530 (nexts == 0 || extno < nexts) && max_len > 0) {
5531 /*
5532 * Is the found extent after a hole in which bno lives?
5533 * Just back up to the previous extent, if so.
5534 */
5535 if (got.br_startoff > bno) {
5536 if (--lastx < 0)
5537 break;
5538 xfs_iext_get_extent(ifp, lastx, &got);
5539 }
5540 /*
5541 * Is the last block of this extent before the range
5542 * we're supposed to delete? If so, we're done.
5543 */
5544 bno = XFS_FILEOFF_MIN(bno,
5545 got.br_startoff + got.br_blockcount - 1);
5546 if (bno < start)
5547 break;
5548 /*
5549 * Then deal with the (possibly delayed) allocated space
5550 * we found.
5551 */
5552 del = got;
5553 wasdel = isnullstartblock(del.br_startblock);
5554
5555 /*
5556 * Make sure we don't touch multiple AGF headers out of order
5557 * in a single transaction, as that could cause AB-BA deadlocks.
5558 */
5559 if (!wasdel && !isrt) {
5560 agno = XFS_FSB_TO_AGNO(mp, del.br_startblock);
5561 if (prev_agno != NULLAGNUMBER && prev_agno > agno)
5562 break;
5563 prev_agno = agno;
5564 }
5565 if (got.br_startoff < start) {
5566 del.br_startoff = start;
5567 del.br_blockcount -= start - got.br_startoff;
5568 if (!wasdel)
5569 del.br_startblock += start - got.br_startoff;
5570 }
5571 if (del.br_startoff + del.br_blockcount > bno + 1)
5572 del.br_blockcount = bno + 1 - del.br_startoff;
5573
5574 /* How much can we safely unmap? */
5575 if (max_len < del.br_blockcount) {
5576 del.br_startoff += del.br_blockcount - max_len;
5577 if (!wasdel)
5578 del.br_startblock += del.br_blockcount - max_len;
5579 del.br_blockcount = max_len;
5580 }
5581
5582 sum = del.br_startblock + del.br_blockcount;
5583 if (isrt &&
5584 (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5585 /*
5586 * Realtime extent not lined up at the end.
5587 * The extent could have been split into written
5588 * and unwritten pieces, or we could just be
5589 * unmapping part of it. But we can't really
5590 * get rid of part of a realtime extent.
5591 */
5592 if (del.br_state == XFS_EXT_UNWRITTEN ||
5593 !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5594 /*
5595 * This piece is unwritten, or we're not
5596 * using unwritten extents. Skip over it.
5597 */
5598 ASSERT(bno >= mod);
5599 bno -= mod > del.br_blockcount ?
5600 del.br_blockcount : mod;
5601 if (bno < got.br_startoff) {
5602 if (--lastx >= 0)
5603 xfs_bmbt_get_all(xfs_iext_get_ext(
5604 ifp, lastx), &got);
5605 }
5606 continue;
5607 }
5608 /*
5609 * It's written, turn it unwritten.
5610 * This is better than zeroing it.
5611 */
5612 ASSERT(del.br_state == XFS_EXT_NORM);
5613 ASSERT(tp->t_blk_res > 0);
5614 /*
5615 * If this spans a realtime extent boundary,
5616 * chop it back to the start of the one we end at.
5617 */
5618 if (del.br_blockcount > mod) {
5619 del.br_startoff += del.br_blockcount - mod;
5620 del.br_startblock += del.br_blockcount - mod;
5621 del.br_blockcount = mod;
5622 }
5623 del.br_state = XFS_EXT_UNWRITTEN;
5624 error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5625 whichfork, &lastx, &cur, &del,
5626 firstblock, dfops, &logflags);
5627 if (error)
5628 goto error0;
5629 goto nodelete;
5630 }
5631 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5632 /*
5633 * Realtime extent is lined up at the end but not
5634 * at the front. We'll get rid of full extents if
5635 * we can.
5636 */
5637 mod = mp->m_sb.sb_rextsize - mod;
5638 if (del.br_blockcount > mod) {
5639 del.br_blockcount -= mod;
5640 del.br_startoff += mod;
5641 del.br_startblock += mod;
5642 } else if ((del.br_startoff == start &&
5643 (del.br_state == XFS_EXT_UNWRITTEN ||
5644 tp->t_blk_res == 0)) ||
5645 !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5646 /*
5647 * Can't make it unwritten. There isn't
5648 * a full extent here so just skip it.
5649 */
5650 ASSERT(bno >= del.br_blockcount);
5651 bno -= del.br_blockcount;
5652 if (got.br_startoff > bno && --lastx >= 0)
5653 xfs_iext_get_extent(ifp, lastx, &got);
5654 continue;
5655 } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5656 struct xfs_bmbt_irec prev;
5657
5658 /*
5659 * This one is already unwritten.
5660 * It must have a written left neighbor.
5661 * Unwrite the killed part of that one and
5662 * try again.
5663 */
5664 ASSERT(lastx > 0);
5665 xfs_iext_get_extent(ifp, lastx - 1, &prev);
5666 ASSERT(prev.br_state == XFS_EXT_NORM);
5667 ASSERT(!isnullstartblock(prev.br_startblock));
5668 ASSERT(del.br_startblock ==
5669 prev.br_startblock + prev.br_blockcount);
5670 if (prev.br_startoff < start) {
5671 mod = start - prev.br_startoff;
5672 prev.br_blockcount -= mod;
5673 prev.br_startblock += mod;
5674 prev.br_startoff = start;
5675 }
5676 prev.br_state = XFS_EXT_UNWRITTEN;
5677 lastx--;
5678 error = xfs_bmap_add_extent_unwritten_real(tp,
5679 ip, whichfork, &lastx, &cur,
5680 &prev, firstblock, dfops,
5681 &logflags);
5682 if (error)
5683 goto error0;
5684 goto nodelete;
5685 } else {
5686 ASSERT(del.br_state == XFS_EXT_NORM);
5687 del.br_state = XFS_EXT_UNWRITTEN;
5688 error = xfs_bmap_add_extent_unwritten_real(tp,
5689 ip, whichfork, &lastx, &cur,
5690 &del, firstblock, dfops,
5691 &logflags);
5692 if (error)
5693 goto error0;
5694 goto nodelete;
5695 }
5696 }
5697
5698 /*
5699 * If it's the case where the directory code is running
5700 * with no block reservation, and the deleted block is in
5701 * the middle of its extent, and the resulting insert
5702 * of an extent would cause transformation to btree format,
5703 * then reject it. The calling code will then swap
5704 * blocks around instead.
5705 * We have to do this now, rather than waiting for the
5706 * conversion to btree format, since the transaction
5707 * will be dirty.
5708 */
5709 if (!wasdel && tp->t_blk_res == 0 &&
5710 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5711 XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
5712 XFS_IFORK_MAXEXT(ip, whichfork) &&
5713 del.br_startoff > got.br_startoff &&
5714 del.br_startoff + del.br_blockcount <
5715 got.br_startoff + got.br_blockcount) {
5716 error = -ENOSPC;
5717 goto error0;
5718 }
5719
5720 /*
5721 * Unreserve quota and update realtime free space, if
5722 * appropriate. If delayed allocation, update the inode delalloc
5723 * counter now and wait to update the sb counters as
5724 * xfs_bmap_del_extent() might need to borrow some blocks.
5725 */
5726 if (wasdel) {
5727 ASSERT(startblockval(del.br_startblock) > 0);
5728 if (isrt) {
5729 xfs_filblks_t rtexts;
5730
5731 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5732 do_div(rtexts, mp->m_sb.sb_rextsize);
5733 xfs_mod_frextents(mp, (int64_t)rtexts);
5734 (void)xfs_trans_reserve_quota_nblks(NULL,
5735 ip, -((long)del.br_blockcount), 0,
5736 XFS_QMOPT_RES_RTBLKS);
5737 } else {
5738 (void)xfs_trans_reserve_quota_nblks(NULL,
5739 ip, -((long)del.br_blockcount), 0,
5740 XFS_QMOPT_RES_REGBLKS);
5741 }
5742 ip->i_delayed_blks -= del.br_blockcount;
5743 if (cur)
5744 cur->bc_private.b.flags |=
5745 XFS_BTCUR_BPRV_WASDEL;
5746 } else if (cur)
5747 cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5748
5749 error = xfs_bmap_del_extent(ip, tp, &lastx, dfops, cur, &del,
5750 &tmp_logflags, whichfork, flags);
5751 logflags |= tmp_logflags;
5752 if (error)
5753 goto error0;
5754
5755 if (!isrt && wasdel)
5756 xfs_mod_fdblocks(mp, (int64_t)del.br_blockcount, false);
5757
5758 max_len -= del.br_blockcount;
5759 bno = del.br_startoff - 1;
5760 nodelete:
5761 /*
5762 * If not done go on to the next (previous) record.
5763 */
5764 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5765 if (lastx >= 0) {
5766 xfs_iext_get_extent(ifp, lastx, &got);
5767 if (got.br_startoff > bno && --lastx >= 0)
5768 xfs_iext_get_extent(ifp, lastx, &got);
5769 }
5770 extno++;
5771 }
5772 }
5773 if (bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0)
5774 *rlen = 0;
5775 else
5776 *rlen = bno - start + 1;
5777
5778 /*
5779 * Convert to a btree if necessary.
5780 */
5781 if (xfs_bmap_needs_btree(ip, whichfork)) {
5782 ASSERT(cur == NULL);
5783 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, dfops,
5784 &cur, 0, &tmp_logflags, whichfork);
5785 logflags |= tmp_logflags;
5786 if (error)
5787 goto error0;
5788 }
5789 /*
5790 * transform from btree to extents, give it cur
5791 */
5792 else if (xfs_bmap_wants_extents(ip, whichfork)) {
5793 ASSERT(cur != NULL);
5794 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5795 whichfork);
5796 logflags |= tmp_logflags;
5797 if (error)
5798 goto error0;
5799 }
5800 /*
5801 * transform from extents to local?
5802 */
5803 error = 0;
5804 error0:
5805 /*
5806 * Log everything. Do this after conversion, there's no point in
5807 * logging the extent records if we've converted to btree format.
5808 */
5809 if ((logflags & xfs_ilog_fext(whichfork)) &&
5810 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5811 logflags &= ~xfs_ilog_fext(whichfork);
5812 else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5813 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5814 logflags &= ~xfs_ilog_fbroot(whichfork);
5815 /*
5816 * Log inode even in the error case, if the transaction
5817 * is dirty we'll need to shut down the filesystem.
5818 */
5819 if (logflags)
5820 xfs_trans_log_inode(tp, ip, logflags);
5821 if (cur) {
5822 if (!error) {
5823 *firstblock = cur->bc_private.b.firstblock;
5824 cur->bc_private.b.allocated = 0;
5825 }
5826 xfs_btree_del_cursor(cur,
5827 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5828 }
5829 return error;
5830 }
5831
5832 /* Unmap a range of a file. */
5833 int
xfs_bunmapi(xfs_trans_t * tp,struct xfs_inode * ip,xfs_fileoff_t bno,xfs_filblks_t len,int flags,xfs_extnum_t nexts,xfs_fsblock_t * firstblock,struct xfs_defer_ops * dfops,int * done)5834 xfs_bunmapi(
5835 xfs_trans_t *tp,
5836 struct xfs_inode *ip,
5837 xfs_fileoff_t bno,
5838 xfs_filblks_t len,
5839 int flags,
5840 xfs_extnum_t nexts,
5841 xfs_fsblock_t *firstblock,
5842 struct xfs_defer_ops *dfops,
5843 int *done)
5844 {
5845 int error;
5846
5847 error = __xfs_bunmapi(tp, ip, bno, &len, flags, nexts, firstblock,
5848 dfops);
5849 *done = (len == 0);
5850 return error;
5851 }
5852
5853 /*
5854 * Determine whether an extent shift can be accomplished by a merge with the
5855 * extent that precedes the target hole of the shift.
5856 */
5857 STATIC bool
xfs_bmse_can_merge(struct xfs_bmbt_irec * left,struct xfs_bmbt_irec * got,xfs_fileoff_t shift)5858 xfs_bmse_can_merge(
5859 struct xfs_bmbt_irec *left, /* preceding extent */
5860 struct xfs_bmbt_irec *got, /* current extent to shift */
5861 xfs_fileoff_t shift) /* shift fsb */
5862 {
5863 xfs_fileoff_t startoff;
5864
5865 startoff = got->br_startoff - shift;
5866
5867 /*
5868 * The extent, once shifted, must be adjacent in-file and on-disk with
5869 * the preceding extent.
5870 */
5871 if ((left->br_startoff + left->br_blockcount != startoff) ||
5872 (left->br_startblock + left->br_blockcount != got->br_startblock) ||
5873 (left->br_state != got->br_state) ||
5874 (left->br_blockcount + got->br_blockcount > MAXEXTLEN))
5875 return false;
5876
5877 return true;
5878 }
5879
5880 /*
5881 * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5882 * hole in the file. If an extent shift would result in the extent being fully
5883 * adjacent to the extent that currently precedes the hole, we can merge with
5884 * the preceding extent rather than do the shift.
5885 *
5886 * This function assumes the caller has verified a shift-by-merge is possible
5887 * with the provided extents via xfs_bmse_can_merge().
5888 */
5889 STATIC int
xfs_bmse_merge(struct xfs_inode * ip,int whichfork,xfs_fileoff_t shift,int current_ext,struct xfs_bmbt_irec * got,struct xfs_bmbt_irec * left,struct xfs_btree_cur * cur,int * logflags,struct xfs_defer_ops * dfops)5890 xfs_bmse_merge(
5891 struct xfs_inode *ip,
5892 int whichfork,
5893 xfs_fileoff_t shift, /* shift fsb */
5894 int current_ext, /* idx of gotp */
5895 struct xfs_bmbt_irec *got, /* extent to shift */
5896 struct xfs_bmbt_irec *left, /* preceding extent */
5897 struct xfs_btree_cur *cur,
5898 int *logflags, /* output */
5899 struct xfs_defer_ops *dfops)
5900 {
5901 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
5902 struct xfs_bmbt_irec new;
5903 xfs_filblks_t blockcount;
5904 int error, i;
5905 struct xfs_mount *mp = ip->i_mount;
5906
5907 blockcount = left->br_blockcount + got->br_blockcount;
5908
5909 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5910 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5911 ASSERT(xfs_bmse_can_merge(left, got, shift));
5912
5913 new = *left;
5914 new.br_blockcount = blockcount;
5915
5916 /*
5917 * Update the on-disk extent count, the btree if necessary and log the
5918 * inode.
5919 */
5920 XFS_IFORK_NEXT_SET(ip, whichfork,
5921 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5922 *logflags |= XFS_ILOG_CORE;
5923 if (!cur) {
5924 *logflags |= XFS_ILOG_DEXT;
5925 goto done;
5926 }
5927
5928 /* lookup and remove the extent to merge */
5929 error = xfs_bmbt_lookup_eq(cur, got->br_startoff, got->br_startblock,
5930 got->br_blockcount, &i);
5931 if (error)
5932 return error;
5933 XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5934
5935 error = xfs_btree_delete(cur, &i);
5936 if (error)
5937 return error;
5938 XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5939
5940 /* lookup and update size of the previous extent */
5941 error = xfs_bmbt_lookup_eq(cur, left->br_startoff, left->br_startblock,
5942 left->br_blockcount, &i);
5943 if (error)
5944 return error;
5945 XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5946
5947 error = xfs_bmbt_update(cur, new.br_startoff, new.br_startblock,
5948 new.br_blockcount, new.br_state);
5949 if (error)
5950 return error;
5951
5952 done:
5953 xfs_iext_update_extent(ifp, current_ext - 1, &new);
5954 xfs_iext_remove(ip, current_ext, 1, 0);
5955
5956 /* update reverse mapping. rmap functions merge the rmaps for us */
5957 error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, got);
5958 if (error)
5959 return error;
5960 memcpy(&new, got, sizeof(new));
5961 new.br_startoff = left->br_startoff + left->br_blockcount;
5962 return xfs_rmap_map_extent(mp, dfops, ip, whichfork, &new);
5963 }
5964
5965 /*
5966 * Shift a single extent.
5967 */
5968 STATIC int
xfs_bmse_shift_one(struct xfs_inode * ip,int whichfork,xfs_fileoff_t offset_shift_fsb,int * current_ext,struct xfs_bmbt_irec * got,struct xfs_btree_cur * cur,int * logflags,enum shift_direction direction,struct xfs_defer_ops * dfops)5969 xfs_bmse_shift_one(
5970 struct xfs_inode *ip,
5971 int whichfork,
5972 xfs_fileoff_t offset_shift_fsb,
5973 int *current_ext,
5974 struct xfs_bmbt_irec *got,
5975 struct xfs_btree_cur *cur,
5976 int *logflags,
5977 enum shift_direction direction,
5978 struct xfs_defer_ops *dfops)
5979 {
5980 struct xfs_ifork *ifp;
5981 struct xfs_mount *mp;
5982 xfs_fileoff_t startoff;
5983 struct xfs_bmbt_irec adj_irec, new;
5984 int error;
5985 int i;
5986 int total_extents;
5987
5988 mp = ip->i_mount;
5989 ifp = XFS_IFORK_PTR(ip, whichfork);
5990 total_extents = xfs_iext_count(ifp);
5991
5992 /* delalloc extents should be prevented by caller */
5993 XFS_WANT_CORRUPTED_RETURN(mp, !isnullstartblock(got->br_startblock));
5994
5995 if (direction == SHIFT_LEFT) {
5996 startoff = got->br_startoff - offset_shift_fsb;
5997
5998 /*
5999 * Check for merge if we've got an extent to the left,
6000 * otherwise make sure there's enough room at the start
6001 * of the file for the shift.
6002 */
6003 if (!*current_ext) {
6004 if (got->br_startoff < offset_shift_fsb)
6005 return -EINVAL;
6006 goto update_current_ext;
6007 }
6008
6009 /*
6010 * grab the left extent and check for a large enough hole.
6011 */
6012 xfs_iext_get_extent(ifp, *current_ext - 1, &adj_irec);
6013 if (startoff < adj_irec.br_startoff + adj_irec.br_blockcount)
6014 return -EINVAL;
6015
6016 /* check whether to merge the extent or shift it down */
6017 if (xfs_bmse_can_merge(&adj_irec, got, offset_shift_fsb)) {
6018 return xfs_bmse_merge(ip, whichfork, offset_shift_fsb,
6019 *current_ext, got, &adj_irec,
6020 cur, logflags, dfops);
6021 }
6022 } else {
6023 startoff = got->br_startoff + offset_shift_fsb;
6024 /* nothing to move if this is the last extent */
6025 if (*current_ext >= (total_extents - 1))
6026 goto update_current_ext;
6027
6028 /*
6029 * If this is not the last extent in the file, make sure there
6030 * is enough room between current extent and next extent for
6031 * accommodating the shift.
6032 */
6033 xfs_iext_get_extent(ifp, *current_ext + 1, &adj_irec);
6034 if (startoff + got->br_blockcount > adj_irec.br_startoff)
6035 return -EINVAL;
6036
6037 /*
6038 * Unlike a left shift (which involves a hole punch),
6039 * a right shift does not modify extent neighbors
6040 * in any way. We should never find mergeable extents
6041 * in this scenario. Check anyways and warn if we
6042 * encounter two extents that could be one.
6043 */
6044 if (xfs_bmse_can_merge(got, &adj_irec, offset_shift_fsb))
6045 WARN_ON_ONCE(1);
6046 }
6047
6048 /*
6049 * Increment the extent index for the next iteration, update the start
6050 * offset of the in-core extent and update the btree if applicable.
6051 */
6052 update_current_ext:
6053 *logflags |= XFS_ILOG_CORE;
6054
6055 new = *got;
6056 new.br_startoff = startoff;
6057
6058 if (cur) {
6059 error = xfs_bmbt_lookup_eq(cur, got->br_startoff,
6060 got->br_startblock, got->br_blockcount, &i);
6061 if (error)
6062 return error;
6063 XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
6064
6065 error = xfs_bmbt_update(cur, new.br_startoff,
6066 new.br_startblock, new.br_blockcount,
6067 new.br_state);
6068 if (error)
6069 return error;
6070 } else {
6071 *logflags |= XFS_ILOG_DEXT;
6072 }
6073
6074 xfs_iext_update_extent(ifp, *current_ext, &new);
6075
6076 if (direction == SHIFT_LEFT)
6077 (*current_ext)++;
6078 else
6079 (*current_ext)--;
6080
6081 /* update reverse mapping */
6082 error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, got);
6083 if (error)
6084 return error;
6085 return xfs_rmap_map_extent(mp, dfops, ip, whichfork, &new);
6086 }
6087
6088 /*
6089 * Shift extent records to the left/right to cover/create a hole.
6090 *
6091 * The maximum number of extents to be shifted in a single operation is
6092 * @num_exts. @stop_fsb specifies the file offset at which to stop shift and the
6093 * file offset where we've left off is returned in @next_fsb. @offset_shift_fsb
6094 * is the length by which each extent is shifted. If there is no hole to shift
6095 * the extents into, this will be considered invalid operation and we abort
6096 * immediately.
6097 */
6098 int
xfs_bmap_shift_extents(struct xfs_trans * tp,struct xfs_inode * ip,xfs_fileoff_t * next_fsb,xfs_fileoff_t offset_shift_fsb,int * done,xfs_fileoff_t stop_fsb,xfs_fsblock_t * firstblock,struct xfs_defer_ops * dfops,enum shift_direction direction,int num_exts)6099 xfs_bmap_shift_extents(
6100 struct xfs_trans *tp,
6101 struct xfs_inode *ip,
6102 xfs_fileoff_t *next_fsb,
6103 xfs_fileoff_t offset_shift_fsb,
6104 int *done,
6105 xfs_fileoff_t stop_fsb,
6106 xfs_fsblock_t *firstblock,
6107 struct xfs_defer_ops *dfops,
6108 enum shift_direction direction,
6109 int num_exts)
6110 {
6111 struct xfs_btree_cur *cur = NULL;
6112 struct xfs_bmbt_irec got;
6113 struct xfs_mount *mp = ip->i_mount;
6114 struct xfs_ifork *ifp;
6115 xfs_extnum_t nexts = 0;
6116 xfs_extnum_t current_ext;
6117 xfs_extnum_t total_extents;
6118 xfs_extnum_t stop_extent;
6119 int error = 0;
6120 int whichfork = XFS_DATA_FORK;
6121 int logflags = 0;
6122
6123 if (unlikely(XFS_TEST_ERROR(
6124 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
6125 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
6126 mp, XFS_ERRTAG_BMAPIFORMAT))) {
6127 XFS_ERROR_REPORT("xfs_bmap_shift_extents",
6128 XFS_ERRLEVEL_LOW, mp);
6129 return -EFSCORRUPTED;
6130 }
6131
6132 if (XFS_FORCED_SHUTDOWN(mp))
6133 return -EIO;
6134
6135 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
6136 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
6137 ASSERT(direction == SHIFT_LEFT || direction == SHIFT_RIGHT);
6138
6139 ifp = XFS_IFORK_PTR(ip, whichfork);
6140 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
6141 /* Read in all the extents */
6142 error = xfs_iread_extents(tp, ip, whichfork);
6143 if (error)
6144 return error;
6145 }
6146
6147 if (ifp->if_flags & XFS_IFBROOT) {
6148 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
6149 cur->bc_private.b.firstblock = *firstblock;
6150 cur->bc_private.b.dfops = dfops;
6151 cur->bc_private.b.flags = 0;
6152 }
6153
6154 /*
6155 * There may be delalloc extents in the data fork before the range we
6156 * are collapsing out, so we cannot use the count of real extents here.
6157 * Instead we have to calculate it from the incore fork.
6158 */
6159 total_extents = xfs_iext_count(ifp);
6160 if (total_extents == 0) {
6161 *done = 1;
6162 goto del_cursor;
6163 }
6164
6165 /*
6166 * In case of first right shift, we need to initialize next_fsb
6167 */
6168 if (*next_fsb == NULLFSBLOCK) {
6169 ASSERT(direction == SHIFT_RIGHT);
6170
6171 current_ext = total_extents - 1;
6172 xfs_iext_get_extent(ifp, current_ext, &got);
6173 if (stop_fsb > got.br_startoff) {
6174 *done = 1;
6175 goto del_cursor;
6176 }
6177 *next_fsb = got.br_startoff;
6178 } else {
6179 /*
6180 * Look up the extent index for the fsb where we start shifting. We can
6181 * henceforth iterate with current_ext as extent list changes are locked
6182 * out via ilock.
6183 *
6184 * If next_fsb lies in a hole beyond which there are no extents we are
6185 * done.
6186 */
6187 if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, ¤t_ext,
6188 &got)) {
6189 *done = 1;
6190 goto del_cursor;
6191 }
6192 }
6193
6194 /* Lookup the extent index at which we have to stop */
6195 if (direction == SHIFT_RIGHT) {
6196 struct xfs_bmbt_irec s;
6197
6198 xfs_iext_lookup_extent(ip, ifp, stop_fsb, &stop_extent, &s);
6199 /* Make stop_extent exclusive of shift range */
6200 stop_extent--;
6201 if (current_ext <= stop_extent) {
6202 error = -EIO;
6203 goto del_cursor;
6204 }
6205 } else {
6206 stop_extent = total_extents;
6207 if (current_ext >= stop_extent) {
6208 error = -EIO;
6209 goto del_cursor;
6210 }
6211 }
6212
6213 while (nexts++ < num_exts) {
6214 error = xfs_bmse_shift_one(ip, whichfork, offset_shift_fsb,
6215 ¤t_ext, &got, cur, &logflags,
6216 direction, dfops);
6217 if (error)
6218 goto del_cursor;
6219 /*
6220 * If there was an extent merge during the shift, the extent
6221 * count can change. Update the total and grade the next record.
6222 */
6223 if (direction == SHIFT_LEFT) {
6224 total_extents = xfs_iext_count(ifp);
6225 stop_extent = total_extents;
6226 }
6227
6228 if (current_ext == stop_extent) {
6229 *done = 1;
6230 *next_fsb = NULLFSBLOCK;
6231 break;
6232 }
6233 xfs_iext_get_extent(ifp, current_ext, &got);
6234 }
6235
6236 if (!*done)
6237 *next_fsb = got.br_startoff;
6238
6239 del_cursor:
6240 if (cur)
6241 xfs_btree_del_cursor(cur,
6242 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
6243
6244 if (logflags)
6245 xfs_trans_log_inode(tp, ip, logflags);
6246
6247 return error;
6248 }
6249
6250 /*
6251 * Splits an extent into two extents at split_fsb block such that it is
6252 * the first block of the current_ext. @current_ext is a target extent
6253 * to be split. @split_fsb is a block where the extents is split.
6254 * If split_fsb lies in a hole or the first block of extents, just return 0.
6255 */
6256 STATIC int
xfs_bmap_split_extent_at(struct xfs_trans * tp,struct xfs_inode * ip,xfs_fileoff_t split_fsb,xfs_fsblock_t * firstfsb,struct xfs_defer_ops * dfops)6257 xfs_bmap_split_extent_at(
6258 struct xfs_trans *tp,
6259 struct xfs_inode *ip,
6260 xfs_fileoff_t split_fsb,
6261 xfs_fsblock_t *firstfsb,
6262 struct xfs_defer_ops *dfops)
6263 {
6264 int whichfork = XFS_DATA_FORK;
6265 struct xfs_btree_cur *cur = NULL;
6266 struct xfs_bmbt_irec got;
6267 struct xfs_bmbt_irec new; /* split extent */
6268 struct xfs_mount *mp = ip->i_mount;
6269 struct xfs_ifork *ifp;
6270 xfs_fsblock_t gotblkcnt; /* new block count for got */
6271 xfs_extnum_t current_ext;
6272 int error = 0;
6273 int logflags = 0;
6274 int i = 0;
6275
6276 if (unlikely(XFS_TEST_ERROR(
6277 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
6278 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
6279 mp, XFS_ERRTAG_BMAPIFORMAT))) {
6280 XFS_ERROR_REPORT("xfs_bmap_split_extent_at",
6281 XFS_ERRLEVEL_LOW, mp);
6282 return -EFSCORRUPTED;
6283 }
6284
6285 if (XFS_FORCED_SHUTDOWN(mp))
6286 return -EIO;
6287
6288 ifp = XFS_IFORK_PTR(ip, whichfork);
6289 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
6290 /* Read in all the extents */
6291 error = xfs_iread_extents(tp, ip, whichfork);
6292 if (error)
6293 return error;
6294 }
6295
6296 /*
6297 * If there are not extents, or split_fsb lies in a hole we are done.
6298 */
6299 if (!xfs_iext_lookup_extent(ip, ifp, split_fsb, ¤t_ext, &got) ||
6300 got.br_startoff >= split_fsb)
6301 return 0;
6302
6303 gotblkcnt = split_fsb - got.br_startoff;
6304 new.br_startoff = split_fsb;
6305 new.br_startblock = got.br_startblock + gotblkcnt;
6306 new.br_blockcount = got.br_blockcount - gotblkcnt;
6307 new.br_state = got.br_state;
6308
6309 if (ifp->if_flags & XFS_IFBROOT) {
6310 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
6311 cur->bc_private.b.firstblock = *firstfsb;
6312 cur->bc_private.b.dfops = dfops;
6313 cur->bc_private.b.flags = 0;
6314 error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
6315 got.br_startblock,
6316 got.br_blockcount,
6317 &i);
6318 if (error)
6319 goto del_cursor;
6320 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
6321 }
6322
6323 got.br_blockcount = gotblkcnt;
6324 xfs_iext_update_extent(ifp, current_ext, &got);
6325
6326 logflags = XFS_ILOG_CORE;
6327 if (cur) {
6328 error = xfs_bmbt_update(cur, got.br_startoff,
6329 got.br_startblock,
6330 got.br_blockcount,
6331 got.br_state);
6332 if (error)
6333 goto del_cursor;
6334 } else
6335 logflags |= XFS_ILOG_DEXT;
6336
6337 /* Add new extent */
6338 current_ext++;
6339 xfs_iext_insert(ip, current_ext, 1, &new, 0);
6340 XFS_IFORK_NEXT_SET(ip, whichfork,
6341 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
6342
6343 if (cur) {
6344 error = xfs_bmbt_lookup_eq(cur, new.br_startoff,
6345 new.br_startblock, new.br_blockcount,
6346 &i);
6347 if (error)
6348 goto del_cursor;
6349 XFS_WANT_CORRUPTED_GOTO(mp, i == 0, del_cursor);
6350 cur->bc_rec.b.br_state = new.br_state;
6351
6352 error = xfs_btree_insert(cur, &i);
6353 if (error)
6354 goto del_cursor;
6355 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
6356 }
6357
6358 /*
6359 * Convert to a btree if necessary.
6360 */
6361 if (xfs_bmap_needs_btree(ip, whichfork)) {
6362 int tmp_logflags; /* partial log flag return val */
6363
6364 ASSERT(cur == NULL);
6365 error = xfs_bmap_extents_to_btree(tp, ip, firstfsb, dfops,
6366 &cur, 0, &tmp_logflags, whichfork);
6367 logflags |= tmp_logflags;
6368 }
6369
6370 del_cursor:
6371 if (cur) {
6372 cur->bc_private.b.allocated = 0;
6373 xfs_btree_del_cursor(cur,
6374 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
6375 }
6376
6377 if (logflags)
6378 xfs_trans_log_inode(tp, ip, logflags);
6379 return error;
6380 }
6381
6382 int
xfs_bmap_split_extent(struct xfs_inode * ip,xfs_fileoff_t split_fsb)6383 xfs_bmap_split_extent(
6384 struct xfs_inode *ip,
6385 xfs_fileoff_t split_fsb)
6386 {
6387 struct xfs_mount *mp = ip->i_mount;
6388 struct xfs_trans *tp;
6389 struct xfs_defer_ops dfops;
6390 xfs_fsblock_t firstfsb;
6391 int error;
6392
6393 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write,
6394 XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp);
6395 if (error)
6396 return error;
6397
6398 xfs_ilock(ip, XFS_ILOCK_EXCL);
6399 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
6400
6401 xfs_defer_init(&dfops, &firstfsb);
6402
6403 error = xfs_bmap_split_extent_at(tp, ip, split_fsb,
6404 &firstfsb, &dfops);
6405 if (error)
6406 goto out;
6407
6408 error = xfs_defer_finish(&tp, &dfops);
6409 if (error)
6410 goto out;
6411
6412 return xfs_trans_commit(tp);
6413
6414 out:
6415 xfs_defer_cancel(&dfops);
6416 xfs_trans_cancel(tp);
6417 return error;
6418 }
6419
6420 /* Deferred mapping is only for real extents in the data fork. */
6421 static bool
xfs_bmap_is_update_needed(struct xfs_bmbt_irec * bmap)6422 xfs_bmap_is_update_needed(
6423 struct xfs_bmbt_irec *bmap)
6424 {
6425 return bmap->br_startblock != HOLESTARTBLOCK &&
6426 bmap->br_startblock != DELAYSTARTBLOCK;
6427 }
6428
6429 /* Record a bmap intent. */
6430 static int
__xfs_bmap_add(struct xfs_mount * mp,struct xfs_defer_ops * dfops,enum xfs_bmap_intent_type type,struct xfs_inode * ip,int whichfork,struct xfs_bmbt_irec * bmap)6431 __xfs_bmap_add(
6432 struct xfs_mount *mp,
6433 struct xfs_defer_ops *dfops,
6434 enum xfs_bmap_intent_type type,
6435 struct xfs_inode *ip,
6436 int whichfork,
6437 struct xfs_bmbt_irec *bmap)
6438 {
6439 int error;
6440 struct xfs_bmap_intent *bi;
6441
6442 trace_xfs_bmap_defer(mp,
6443 XFS_FSB_TO_AGNO(mp, bmap->br_startblock),
6444 type,
6445 XFS_FSB_TO_AGBNO(mp, bmap->br_startblock),
6446 ip->i_ino, whichfork,
6447 bmap->br_startoff,
6448 bmap->br_blockcount,
6449 bmap->br_state);
6450
6451 bi = kmem_alloc(sizeof(struct xfs_bmap_intent), KM_SLEEP | KM_NOFS);
6452 INIT_LIST_HEAD(&bi->bi_list);
6453 bi->bi_type = type;
6454 bi->bi_owner = ip;
6455 bi->bi_whichfork = whichfork;
6456 bi->bi_bmap = *bmap;
6457
6458 error = xfs_defer_ijoin(dfops, bi->bi_owner);
6459 if (error) {
6460 kmem_free(bi);
6461 return error;
6462 }
6463
6464 xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_BMAP, &bi->bi_list);
6465 return 0;
6466 }
6467
6468 /* Map an extent into a file. */
6469 int
xfs_bmap_map_extent(struct xfs_mount * mp,struct xfs_defer_ops * dfops,struct xfs_inode * ip,struct xfs_bmbt_irec * PREV)6470 xfs_bmap_map_extent(
6471 struct xfs_mount *mp,
6472 struct xfs_defer_ops *dfops,
6473 struct xfs_inode *ip,
6474 struct xfs_bmbt_irec *PREV)
6475 {
6476 if (!xfs_bmap_is_update_needed(PREV))
6477 return 0;
6478
6479 return __xfs_bmap_add(mp, dfops, XFS_BMAP_MAP, ip,
6480 XFS_DATA_FORK, PREV);
6481 }
6482
6483 /* Unmap an extent out of a file. */
6484 int
xfs_bmap_unmap_extent(struct xfs_mount * mp,struct xfs_defer_ops * dfops,struct xfs_inode * ip,struct xfs_bmbt_irec * PREV)6485 xfs_bmap_unmap_extent(
6486 struct xfs_mount *mp,
6487 struct xfs_defer_ops *dfops,
6488 struct xfs_inode *ip,
6489 struct xfs_bmbt_irec *PREV)
6490 {
6491 if (!xfs_bmap_is_update_needed(PREV))
6492 return 0;
6493
6494 return __xfs_bmap_add(mp, dfops, XFS_BMAP_UNMAP, ip,
6495 XFS_DATA_FORK, PREV);
6496 }
6497
6498 /*
6499 * Process one of the deferred bmap operations. We pass back the
6500 * btree cursor to maintain our lock on the bmapbt between calls.
6501 */
6502 int
xfs_bmap_finish_one(struct xfs_trans * tp,struct xfs_defer_ops * dfops,struct xfs_inode * ip,enum xfs_bmap_intent_type type,int whichfork,xfs_fileoff_t startoff,xfs_fsblock_t startblock,xfs_filblks_t * blockcount,xfs_exntst_t state)6503 xfs_bmap_finish_one(
6504 struct xfs_trans *tp,
6505 struct xfs_defer_ops *dfops,
6506 struct xfs_inode *ip,
6507 enum xfs_bmap_intent_type type,
6508 int whichfork,
6509 xfs_fileoff_t startoff,
6510 xfs_fsblock_t startblock,
6511 xfs_filblks_t *blockcount,
6512 xfs_exntst_t state)
6513 {
6514 xfs_fsblock_t firstfsb;
6515 int error = 0;
6516
6517 /*
6518 * firstfsb is tied to the transaction lifetime and is used to
6519 * ensure correct AG locking order and schedule work item
6520 * continuations. XFS_BUI_MAX_FAST_EXTENTS (== 1) restricts us
6521 * to only making one bmap call per transaction, so it should
6522 * be safe to have it as a local variable here.
6523 */
6524 firstfsb = NULLFSBLOCK;
6525
6526 trace_xfs_bmap_deferred(tp->t_mountp,
6527 XFS_FSB_TO_AGNO(tp->t_mountp, startblock), type,
6528 XFS_FSB_TO_AGBNO(tp->t_mountp, startblock),
6529 ip->i_ino, whichfork, startoff, *blockcount, state);
6530
6531 if (WARN_ON_ONCE(whichfork != XFS_DATA_FORK))
6532 return -EFSCORRUPTED;
6533
6534 if (XFS_TEST_ERROR(false, tp->t_mountp,
6535 XFS_ERRTAG_BMAP_FINISH_ONE))
6536 return -EIO;
6537
6538 switch (type) {
6539 case XFS_BMAP_MAP:
6540 error = xfs_bmapi_remap(tp, ip, startoff, *blockcount,
6541 startblock, dfops);
6542 *blockcount = 0;
6543 break;
6544 case XFS_BMAP_UNMAP:
6545 error = __xfs_bunmapi(tp, ip, startoff, blockcount,
6546 XFS_BMAPI_REMAP, 1, &firstfsb, dfops);
6547 break;
6548 default:
6549 ASSERT(0);
6550 error = -EFSCORRUPTED;
6551 }
6552
6553 return error;
6554 }
6555