1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6 #include "xfs.h"
7 #include "xfs_fs.h"
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_bit.h"
13 #include "xfs_sb.h"
14 #include "xfs_mount.h"
15 #include "xfs_defer.h"
16 #include "xfs_dir2.h"
17 #include "xfs_inode.h"
18 #include "xfs_btree.h"
19 #include "xfs_trans.h"
20 #include "xfs_alloc.h"
21 #include "xfs_bmap.h"
22 #include "xfs_bmap_util.h"
23 #include "xfs_bmap_btree.h"
24 #include "xfs_rtbitmap.h"
25 #include "xfs_errortag.h"
26 #include "xfs_error.h"
27 #include "xfs_quota.h"
28 #include "xfs_trans_space.h"
29 #include "xfs_buf_item.h"
30 #include "xfs_trace.h"
31 #include "xfs_attr_leaf.h"
32 #include "xfs_filestream.h"
33 #include "xfs_rmap.h"
34 #include "xfs_ag.h"
35 #include "xfs_ag_resv.h"
36 #include "xfs_refcount.h"
37 #include "xfs_icache.h"
38 #include "xfs_iomap.h"
39 #include "xfs_health.h"
40 #include "xfs_bmap_item.h"
41 #include "xfs_symlink_remote.h"
42 #include "xfs_inode_util.h"
43
44 struct kmem_cache *xfs_bmap_intent_cache;
45
46 /*
47 * Miscellaneous helper functions
48 */
49
50 /*
51 * Compute and fill in the value of the maximum depth of a bmap btree
52 * in this filesystem. Done once, during mount.
53 */
54 void
xfs_bmap_compute_maxlevels(xfs_mount_t * mp,int whichfork)55 xfs_bmap_compute_maxlevels(
56 xfs_mount_t *mp, /* file system mount structure */
57 int whichfork) /* data or attr fork */
58 {
59 uint64_t maxblocks; /* max blocks at this level */
60 xfs_extnum_t maxleafents; /* max leaf entries possible */
61 int level; /* btree level */
62 int maxrootrecs; /* max records in root block */
63 int minleafrecs; /* min records in leaf block */
64 int minnoderecs; /* min records in node block */
65 int sz; /* root block size */
66
67 /*
68 * The maximum number of extents in a fork, hence the maximum number of
69 * leaf entries, is controlled by the size of the on-disk extent count.
70 *
71 * Note that we can no longer assume that if we are in ATTR1 that the
72 * fork offset of all the inodes will be
73 * (xfs_default_attroffset(ip) >> 3) because we could have mounted with
74 * ATTR2 and then mounted back with ATTR1, keeping the i_forkoff's fixed
75 * but probably at various positions. Therefore, for both ATTR1 and
76 * ATTR2 we have to assume the worst case scenario of a minimum size
77 * available.
78 */
79 maxleafents = xfs_iext_max_nextents(xfs_has_large_extent_counts(mp),
80 whichfork);
81 if (whichfork == XFS_DATA_FORK)
82 sz = xfs_bmdr_space_calc(MINDBTPTRS);
83 else
84 sz = xfs_bmdr_space_calc(MINABTPTRS);
85
86 maxrootrecs = xfs_bmdr_maxrecs(sz, 0);
87 minleafrecs = mp->m_bmap_dmnr[0];
88 minnoderecs = mp->m_bmap_dmnr[1];
89 maxblocks = howmany_64(maxleafents, minleafrecs);
90 for (level = 1; maxblocks > 1; level++) {
91 if (maxblocks <= maxrootrecs)
92 maxblocks = 1;
93 else
94 maxblocks = howmany_64(maxblocks, minnoderecs);
95 }
96 mp->m_bm_maxlevels[whichfork] = level;
97 ASSERT(mp->m_bm_maxlevels[whichfork] <= xfs_bmbt_maxlevels_ondisk());
98 }
99
100 unsigned int
xfs_bmap_compute_attr_offset(struct xfs_mount * mp)101 xfs_bmap_compute_attr_offset(
102 struct xfs_mount *mp)
103 {
104 if (mp->m_sb.sb_inodesize == 256)
105 return XFS_LITINO(mp) - xfs_bmdr_space_calc(MINABTPTRS);
106 return xfs_bmdr_space_calc(6 * MINABTPTRS);
107 }
108
109 STATIC int /* error */
xfs_bmbt_lookup_eq(struct xfs_btree_cur * cur,struct xfs_bmbt_irec * irec,int * stat)110 xfs_bmbt_lookup_eq(
111 struct xfs_btree_cur *cur,
112 struct xfs_bmbt_irec *irec,
113 int *stat) /* success/failure */
114 {
115 cur->bc_rec.b = *irec;
116 return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
117 }
118
119 STATIC int /* error */
xfs_bmbt_lookup_first(struct xfs_btree_cur * cur,int * stat)120 xfs_bmbt_lookup_first(
121 struct xfs_btree_cur *cur,
122 int *stat) /* success/failure */
123 {
124 cur->bc_rec.b.br_startoff = 0;
125 cur->bc_rec.b.br_startblock = 0;
126 cur->bc_rec.b.br_blockcount = 0;
127 return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
128 }
129
130 /*
131 * Check if the inode needs to be converted to btree format.
132 */
xfs_bmap_needs_btree(struct xfs_inode * ip,int whichfork)133 static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
134 {
135 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
136
137 return whichfork != XFS_COW_FORK &&
138 ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
139 ifp->if_nextents > XFS_IFORK_MAXEXT(ip, whichfork);
140 }
141
142 /*
143 * Check if the inode should be converted to extent format.
144 */
xfs_bmap_wants_extents(struct xfs_inode * ip,int whichfork)145 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
146 {
147 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
148
149 return whichfork != XFS_COW_FORK &&
150 ifp->if_format == XFS_DINODE_FMT_BTREE &&
151 ifp->if_nextents <= XFS_IFORK_MAXEXT(ip, whichfork);
152 }
153
154 /*
155 * Update the record referred to by cur to the value given by irec
156 * This either works (return 0) or gets an EFSCORRUPTED error.
157 */
158 STATIC int
xfs_bmbt_update(struct xfs_btree_cur * cur,struct xfs_bmbt_irec * irec)159 xfs_bmbt_update(
160 struct xfs_btree_cur *cur,
161 struct xfs_bmbt_irec *irec)
162 {
163 union xfs_btree_rec rec;
164
165 xfs_bmbt_disk_set_all(&rec.bmbt, irec);
166 return xfs_btree_update(cur, &rec);
167 }
168
169 /*
170 * Compute the worst-case number of indirect blocks that will be used
171 * for ip's delayed extent of length "len".
172 */
173 STATIC xfs_filblks_t
xfs_bmap_worst_indlen(xfs_inode_t * ip,xfs_filblks_t len)174 xfs_bmap_worst_indlen(
175 xfs_inode_t *ip, /* incore inode pointer */
176 xfs_filblks_t len) /* delayed extent length */
177 {
178 int level; /* btree level number */
179 int maxrecs; /* maximum record count at this level */
180 xfs_mount_t *mp; /* mount structure */
181 xfs_filblks_t rval; /* return value */
182
183 mp = ip->i_mount;
184 maxrecs = mp->m_bmap_dmxr[0];
185 for (level = 0, rval = 0;
186 level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
187 level++) {
188 len += maxrecs - 1;
189 do_div(len, maxrecs);
190 rval += len;
191 if (len == 1)
192 return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
193 level - 1;
194 if (level == 0)
195 maxrecs = mp->m_bmap_dmxr[1];
196 }
197 return rval;
198 }
199
200 /*
201 * Calculate the default attribute fork offset for newly created inodes.
202 */
203 uint
xfs_default_attroffset(struct xfs_inode * ip)204 xfs_default_attroffset(
205 struct xfs_inode *ip)
206 {
207 if (ip->i_df.if_format == XFS_DINODE_FMT_DEV)
208 return roundup(sizeof(xfs_dev_t), 8);
209 return M_IGEO(ip->i_mount)->attr_fork_offset;
210 }
211
212 /*
213 * Helper routine to reset inode i_forkoff field when switching attribute fork
214 * from local to extent format - we reset it where possible to make space
215 * available for inline data fork extents.
216 */
217 STATIC void
xfs_bmap_forkoff_reset(xfs_inode_t * ip,int whichfork)218 xfs_bmap_forkoff_reset(
219 xfs_inode_t *ip,
220 int whichfork)
221 {
222 if (whichfork == XFS_ATTR_FORK &&
223 ip->i_df.if_format != XFS_DINODE_FMT_DEV &&
224 ip->i_df.if_format != XFS_DINODE_FMT_BTREE) {
225 uint dfl_forkoff = xfs_default_attroffset(ip) >> 3;
226
227 if (dfl_forkoff > ip->i_forkoff)
228 ip->i_forkoff = dfl_forkoff;
229 }
230 }
231
232 static int
xfs_bmap_read_buf(struct xfs_mount * mp,struct xfs_trans * tp,xfs_fsblock_t fsbno,struct xfs_buf ** bpp)233 xfs_bmap_read_buf(
234 struct xfs_mount *mp, /* file system mount point */
235 struct xfs_trans *tp, /* transaction pointer */
236 xfs_fsblock_t fsbno, /* file system block number */
237 struct xfs_buf **bpp) /* buffer for fsbno */
238 {
239 struct xfs_buf *bp; /* return value */
240 int error;
241
242 if (!xfs_verify_fsbno(mp, fsbno))
243 return -EFSCORRUPTED;
244 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
245 XFS_FSB_TO_DADDR(mp, fsbno), mp->m_bsize, 0, &bp,
246 &xfs_bmbt_buf_ops);
247 if (!error) {
248 xfs_buf_set_ref(bp, XFS_BMAP_BTREE_REF);
249 *bpp = bp;
250 }
251 return error;
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 *lip;
261 int i;
262
263 if (!cur)
264 return NULL;
265
266 for (i = 0; i < cur->bc_maxlevels; i++) {
267 if (!cur->bc_levels[i].bp)
268 break;
269 if (xfs_buf_daddr(cur->bc_levels[i].bp) == bno)
270 return cur->bc_levels[i].bp;
271 }
272
273 /* Chase down all the log items to see if the bp is there */
274 list_for_each_entry(lip, &cur->bc_tp->t_items, li_trans) {
275 struct xfs_buf_log_item *bip = (struct xfs_buf_log_item *)lip;
276
277 if (bip->bli_item.li_type == XFS_LI_BUF &&
278 xfs_buf_daddr(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) %lld",
324 __func__, j, i,
325 (unsigned long long)be64_to_cpu(*thispa));
326 xfs_err(mp, "%s: ptrs are equal in node\n",
327 __func__);
328 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
329 }
330 }
331 }
332 }
333
334 /*
335 * Check that the extents for the inode ip are in the right order in all
336 * btree leaves. THis becomes prohibitively expensive for large extent count
337 * files, so don't bother with inodes that have more than 10,000 extents in
338 * them. The btree record ordering checks will still be done, so for such large
339 * bmapbt constructs that is going to catch most corruptions.
340 */
341 STATIC void
xfs_bmap_check_leaf_extents(struct xfs_btree_cur * cur,xfs_inode_t * ip,int whichfork)342 xfs_bmap_check_leaf_extents(
343 struct xfs_btree_cur *cur, /* btree cursor or null */
344 xfs_inode_t *ip, /* incore inode pointer */
345 int whichfork) /* data or attr fork */
346 {
347 struct xfs_mount *mp = ip->i_mount;
348 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
349 struct xfs_btree_block *block; /* current btree block */
350 xfs_fsblock_t bno; /* block # of "block" */
351 struct xfs_buf *bp; /* buffer for "block" */
352 int error; /* error return value */
353 xfs_extnum_t i=0, j; /* index into the extents list */
354 int level; /* btree level, for checking */
355 __be64 *pp; /* pointer to block address */
356 xfs_bmbt_rec_t *ep; /* pointer to current extent */
357 xfs_bmbt_rec_t last = {0, 0}; /* last extent in prev block */
358 xfs_bmbt_rec_t *nextp; /* pointer to next extent */
359 int bp_release = 0;
360
361 if (ifp->if_format != XFS_DINODE_FMT_BTREE)
362 return;
363
364 /* skip large extent count inodes */
365 if (ip->i_df.if_nextents > 10000)
366 return;
367
368 bno = NULLFSBLOCK;
369 block = ifp->if_broot;
370 /*
371 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
372 */
373 level = be16_to_cpu(block->bb_level);
374 ASSERT(level > 0);
375 xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
376 pp = xfs_bmap_broot_ptr_addr(mp, block, 1, ifp->if_broot_bytes);
377 bno = be64_to_cpu(*pp);
378
379 ASSERT(bno != NULLFSBLOCK);
380 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
381 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
382
383 /*
384 * Go down the tree until leaf level is reached, following the first
385 * pointer (leftmost) at each level.
386 */
387 while (level-- > 0) {
388 /* See if buf is in cur first */
389 bp_release = 0;
390 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
391 if (!bp) {
392 bp_release = 1;
393 error = xfs_bmap_read_buf(mp, NULL, bno, &bp);
394 if (xfs_metadata_is_sick(error))
395 xfs_btree_mark_sick(cur);
396 if (error)
397 goto error_norelse;
398 }
399 block = XFS_BUF_TO_BLOCK(bp);
400 if (level == 0)
401 break;
402
403 /*
404 * Check this block for basic sanity (increasing keys and
405 * no duplicate blocks).
406 */
407
408 xfs_check_block(block, mp, 0, 0);
409 pp = xfs_bmbt_ptr_addr(mp, block, 1, mp->m_bmap_dmxr[1]);
410 bno = be64_to_cpu(*pp);
411 if (XFS_IS_CORRUPT(mp, !xfs_verify_fsbno(mp, bno))) {
412 xfs_btree_mark_sick(cur);
413 error = -EFSCORRUPTED;
414 goto error0;
415 }
416 if (bp_release) {
417 bp_release = 0;
418 xfs_trans_brelse(NULL, bp);
419 }
420 }
421
422 /*
423 * Here with bp and block set to the leftmost leaf node in the tree.
424 */
425 i = 0;
426
427 /*
428 * Loop over all leaf nodes checking that all extents are in the right order.
429 */
430 for (;;) {
431 xfs_fsblock_t nextbno;
432 xfs_extnum_t num_recs;
433
434
435 num_recs = xfs_btree_get_numrecs(block);
436
437 /*
438 * Read-ahead the next leaf block, if any.
439 */
440
441 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
442
443 /*
444 * Check all the extents to make sure they are OK.
445 * If we had a previous block, the last entry should
446 * conform with the first entry in this one.
447 */
448
449 ep = xfs_bmbt_rec_addr(mp, block, 1);
450 if (i) {
451 ASSERT(xfs_bmbt_disk_get_startoff(&last) +
452 xfs_bmbt_disk_get_blockcount(&last) <=
453 xfs_bmbt_disk_get_startoff(ep));
454 }
455 for (j = 1; j < num_recs; j++) {
456 nextp = xfs_bmbt_rec_addr(mp, block, j + 1);
457 ASSERT(xfs_bmbt_disk_get_startoff(ep) +
458 xfs_bmbt_disk_get_blockcount(ep) <=
459 xfs_bmbt_disk_get_startoff(nextp));
460 ep = nextp;
461 }
462
463 last = *ep;
464 i += num_recs;
465 if (bp_release) {
466 bp_release = 0;
467 xfs_trans_brelse(NULL, bp);
468 }
469 bno = nextbno;
470 /*
471 * If we've reached the end, stop.
472 */
473 if (bno == NULLFSBLOCK)
474 break;
475
476 bp_release = 0;
477 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
478 if (!bp) {
479 bp_release = 1;
480 error = xfs_bmap_read_buf(mp, NULL, bno, &bp);
481 if (xfs_metadata_is_sick(error))
482 xfs_btree_mark_sick(cur);
483 if (error)
484 goto error_norelse;
485 }
486 block = XFS_BUF_TO_BLOCK(bp);
487 }
488
489 return;
490
491 error0:
492 xfs_warn(mp, "%s: at error0", __func__);
493 if (bp_release)
494 xfs_trans_brelse(NULL, bp);
495 error_norelse:
496 xfs_warn(mp, "%s: BAD after btree leaves for %llu extents",
497 __func__, i);
498 xfs_err(mp, "%s: CORRUPTED BTREE OR SOMETHING", __func__);
499 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
500 return;
501 }
502
503 /*
504 * Validate that the bmbt_irecs being returned from bmapi are valid
505 * given the caller's original parameters. Specifically check the
506 * ranges of the returned irecs to ensure that they only extend beyond
507 * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
508 */
509 STATIC void
xfs_bmap_validate_ret(xfs_fileoff_t bno,xfs_filblks_t len,uint32_t flags,xfs_bmbt_irec_t * mval,int nmap,int ret_nmap)510 xfs_bmap_validate_ret(
511 xfs_fileoff_t bno,
512 xfs_filblks_t len,
513 uint32_t flags,
514 xfs_bmbt_irec_t *mval,
515 int nmap,
516 int ret_nmap)
517 {
518 int i; /* index to map values */
519
520 ASSERT(ret_nmap <= nmap);
521
522 for (i = 0; i < ret_nmap; i++) {
523 ASSERT(mval[i].br_blockcount > 0);
524 if (!(flags & XFS_BMAPI_ENTIRE)) {
525 ASSERT(mval[i].br_startoff >= bno);
526 ASSERT(mval[i].br_blockcount <= len);
527 ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
528 bno + len);
529 } else {
530 ASSERT(mval[i].br_startoff < bno + len);
531 ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
532 bno);
533 }
534 ASSERT(i == 0 ||
535 mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
536 mval[i].br_startoff);
537 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
538 mval[i].br_startblock != HOLESTARTBLOCK);
539 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
540 mval[i].br_state == XFS_EXT_UNWRITTEN);
541 }
542 }
543
544 #else
545 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork) do { } while (0)
546 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap) do { } while (0)
547 #endif /* DEBUG */
548
549 /*
550 * Inode fork format manipulation functions
551 */
552
553 /*
554 * Convert the inode format to extent format if it currently is in btree format,
555 * but the extent list is small enough that it fits into the extent format.
556 *
557 * Since the extents are already in-core, all we have to do is give up the space
558 * for the btree root and pitch the leaf block.
559 */
560 STATIC int /* error */
xfs_bmap_btree_to_extents(struct xfs_trans * tp,struct xfs_inode * ip,struct xfs_btree_cur * cur,int * logflagsp,int whichfork)561 xfs_bmap_btree_to_extents(
562 struct xfs_trans *tp, /* transaction pointer */
563 struct xfs_inode *ip, /* incore inode pointer */
564 struct xfs_btree_cur *cur, /* btree cursor */
565 int *logflagsp, /* inode logging flags */
566 int whichfork) /* data or attr fork */
567 {
568 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
569 struct xfs_mount *mp = ip->i_mount;
570 struct xfs_btree_block *rblock = ifp->if_broot;
571 struct xfs_btree_block *cblock;/* child btree block */
572 xfs_fsblock_t cbno; /* child block number */
573 struct xfs_buf *cbp; /* child block's buffer */
574 int error; /* error return value */
575 __be64 *pp; /* ptr to block address */
576 struct xfs_owner_info oinfo;
577
578 /* check if we actually need the extent format first: */
579 if (!xfs_bmap_wants_extents(ip, whichfork))
580 return 0;
581
582 ASSERT(cur);
583 ASSERT(whichfork != XFS_COW_FORK);
584 ASSERT(ifp->if_format == XFS_DINODE_FMT_BTREE);
585 ASSERT(be16_to_cpu(rblock->bb_level) == 1);
586 ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
587 ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, false) == 1);
588
589 pp = xfs_bmap_broot_ptr_addr(mp, rblock, 1, ifp->if_broot_bytes);
590 cbno = be64_to_cpu(*pp);
591 #ifdef DEBUG
592 if (XFS_IS_CORRUPT(cur->bc_mp, !xfs_verify_fsbno(mp, cbno))) {
593 xfs_btree_mark_sick(cur);
594 return -EFSCORRUPTED;
595 }
596 #endif
597 error = xfs_bmap_read_buf(mp, tp, cbno, &cbp);
598 if (xfs_metadata_is_sick(error))
599 xfs_btree_mark_sick(cur);
600 if (error)
601 return error;
602 cblock = XFS_BUF_TO_BLOCK(cbp);
603 if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
604 return error;
605
606 xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork);
607 error = xfs_free_extent_later(cur->bc_tp, cbno, 1, &oinfo,
608 XFS_AG_RESV_NONE, 0);
609 if (error)
610 return error;
611
612 ip->i_nblocks--;
613 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
614 xfs_trans_binval(tp, cbp);
615 if (cur->bc_levels[0].bp == cbp)
616 cur->bc_levels[0].bp = NULL;
617 xfs_iroot_realloc(ip, -1, whichfork);
618 ASSERT(ifp->if_broot == NULL);
619 ifp->if_format = XFS_DINODE_FMT_EXTENTS;
620 *logflagsp |= XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
621 return 0;
622 }
623
624 /*
625 * Convert an extents-format file into a btree-format file.
626 * The new file will have a root block (in the inode) and a single child block.
627 */
628 STATIC int /* error */
xfs_bmap_extents_to_btree(struct xfs_trans * tp,struct xfs_inode * ip,struct xfs_btree_cur ** curp,int wasdel,int * logflagsp,int whichfork)629 xfs_bmap_extents_to_btree(
630 struct xfs_trans *tp, /* transaction pointer */
631 struct xfs_inode *ip, /* incore inode pointer */
632 struct xfs_btree_cur **curp, /* cursor returned to caller */
633 int wasdel, /* converting a delayed alloc */
634 int *logflagsp, /* inode logging flags */
635 int whichfork) /* data or attr fork */
636 {
637 struct xfs_btree_block *ablock; /* allocated (child) bt block */
638 struct xfs_buf *abp; /* buffer for ablock */
639 struct xfs_alloc_arg args; /* allocation arguments */
640 struct xfs_bmbt_rec *arp; /* child record pointer */
641 struct xfs_btree_block *block; /* btree root block */
642 struct xfs_btree_cur *cur; /* bmap btree cursor */
643 int error; /* error return value */
644 struct xfs_ifork *ifp; /* inode fork pointer */
645 struct xfs_bmbt_key *kp; /* root block key pointer */
646 struct xfs_mount *mp; /* mount structure */
647 xfs_bmbt_ptr_t *pp; /* root block address pointer */
648 struct xfs_iext_cursor icur;
649 struct xfs_bmbt_irec rec;
650 xfs_extnum_t cnt = 0;
651
652 mp = ip->i_mount;
653 ASSERT(whichfork != XFS_COW_FORK);
654 ifp = xfs_ifork_ptr(ip, whichfork);
655 ASSERT(ifp->if_format == XFS_DINODE_FMT_EXTENTS);
656
657 /*
658 * Make space in the inode incore. This needs to be undone if we fail
659 * to expand the root.
660 */
661 xfs_iroot_realloc(ip, 1, whichfork);
662
663 /*
664 * Fill in the root.
665 */
666 block = ifp->if_broot;
667 xfs_bmbt_init_block(ip, block, NULL, 1, 1);
668 /*
669 * Need a cursor. Can't allocate until bb_level is filled in.
670 */
671 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
672 if (wasdel)
673 cur->bc_flags |= XFS_BTREE_BMBT_WASDEL;
674 /*
675 * Convert to a btree with two levels, one record in root.
676 */
677 ifp->if_format = XFS_DINODE_FMT_BTREE;
678 memset(&args, 0, sizeof(args));
679 args.tp = tp;
680 args.mp = mp;
681 xfs_rmap_ino_bmbt_owner(&args.oinfo, ip->i_ino, whichfork);
682
683 args.minlen = args.maxlen = args.prod = 1;
684 args.wasdel = wasdel;
685 *logflagsp = 0;
686 error = xfs_alloc_vextent_start_ag(&args,
687 XFS_INO_TO_FSB(mp, ip->i_ino));
688 if (error)
689 goto out_root_realloc;
690
691 /*
692 * Allocation can't fail, the space was reserved.
693 */
694 if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) {
695 error = -ENOSPC;
696 goto out_root_realloc;
697 }
698
699 cur->bc_bmap.allocated++;
700 ip->i_nblocks++;
701 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
702 error = xfs_trans_get_buf(tp, mp->m_ddev_targp,
703 XFS_FSB_TO_DADDR(mp, args.fsbno),
704 mp->m_bsize, 0, &abp);
705 if (error)
706 goto out_unreserve_dquot;
707
708 /*
709 * Fill in the child block.
710 */
711 ablock = XFS_BUF_TO_BLOCK(abp);
712 xfs_bmbt_init_block(ip, ablock, abp, 0, 0);
713
714 for_each_xfs_iext(ifp, &icur, &rec) {
715 if (isnullstartblock(rec.br_startblock))
716 continue;
717 arp = xfs_bmbt_rec_addr(mp, ablock, 1 + cnt);
718 xfs_bmbt_disk_set_all(arp, &rec);
719 cnt++;
720 }
721 ASSERT(cnt == ifp->if_nextents);
722 xfs_btree_set_numrecs(ablock, cnt);
723
724 /*
725 * Fill in the root key and pointer.
726 */
727 kp = xfs_bmbt_key_addr(mp, block, 1);
728 arp = xfs_bmbt_rec_addr(mp, ablock, 1);
729 kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
730 pp = xfs_bmbt_ptr_addr(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
731 be16_to_cpu(block->bb_level)));
732 *pp = cpu_to_be64(args.fsbno);
733
734 /*
735 * Do all this logging at the end so that
736 * the root is at the right level.
737 */
738 xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
739 xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
740 ASSERT(*curp == NULL);
741 *curp = cur;
742 *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
743 return 0;
744
745 out_unreserve_dquot:
746 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
747 out_root_realloc:
748 xfs_iroot_realloc(ip, -1, whichfork);
749 ifp->if_format = XFS_DINODE_FMT_EXTENTS;
750 ASSERT(ifp->if_broot == NULL);
751 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
752
753 return error;
754 }
755
756 /*
757 * Convert a local file to an extents file.
758 * This code is out of bounds for data forks of regular files,
759 * since the file data needs to get logged so things will stay consistent.
760 * (The bmap-level manipulations are ok, though).
761 */
762 void
xfs_bmap_local_to_extents_empty(struct xfs_trans * tp,struct xfs_inode * ip,int whichfork)763 xfs_bmap_local_to_extents_empty(
764 struct xfs_trans *tp,
765 struct xfs_inode *ip,
766 int whichfork)
767 {
768 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
769
770 ASSERT(whichfork != XFS_COW_FORK);
771 ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
772 ASSERT(ifp->if_bytes == 0);
773 ASSERT(ifp->if_nextents == 0);
774
775 xfs_bmap_forkoff_reset(ip, whichfork);
776 ifp->if_data = NULL;
777 ifp->if_height = 0;
778 ifp->if_format = XFS_DINODE_FMT_EXTENTS;
779 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
780 }
781
782
783 int /* error */
xfs_bmap_local_to_extents(xfs_trans_t * tp,xfs_inode_t * ip,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,void * priv),void * priv)784 xfs_bmap_local_to_extents(
785 xfs_trans_t *tp, /* transaction pointer */
786 xfs_inode_t *ip, /* incore inode pointer */
787 xfs_extlen_t total, /* total blocks needed by transaction */
788 int *logflagsp, /* inode logging flags */
789 int whichfork,
790 void (*init_fn)(struct xfs_trans *tp,
791 struct xfs_buf *bp,
792 struct xfs_inode *ip,
793 struct xfs_ifork *ifp, void *priv),
794 void *priv)
795 {
796 int error = 0;
797 int flags; /* logging flags returned */
798 struct xfs_ifork *ifp; /* inode fork pointer */
799 xfs_alloc_arg_t args; /* allocation arguments */
800 struct xfs_buf *bp; /* buffer for extent block */
801 struct xfs_bmbt_irec rec;
802 struct xfs_iext_cursor icur;
803
804 /*
805 * We don't want to deal with the case of keeping inode data inline yet.
806 * So sending the data fork of a regular inode is invalid.
807 */
808 ASSERT(!(S_ISREG(VFS_I(ip)->i_mode) && whichfork == XFS_DATA_FORK));
809 ifp = xfs_ifork_ptr(ip, whichfork);
810 ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
811
812 if (!ifp->if_bytes) {
813 xfs_bmap_local_to_extents_empty(tp, ip, whichfork);
814 flags = XFS_ILOG_CORE;
815 goto done;
816 }
817
818 flags = 0;
819 error = 0;
820 memset(&args, 0, sizeof(args));
821 args.tp = tp;
822 args.mp = ip->i_mount;
823 args.total = total;
824 args.minlen = args.maxlen = args.prod = 1;
825 xfs_rmap_ino_owner(&args.oinfo, ip->i_ino, whichfork, 0);
826
827 /*
828 * Allocate a block. We know we need only one, since the
829 * file currently fits in an inode.
830 */
831 args.total = total;
832 args.minlen = args.maxlen = args.prod = 1;
833 error = xfs_alloc_vextent_start_ag(&args,
834 XFS_INO_TO_FSB(args.mp, ip->i_ino));
835 if (error)
836 goto done;
837
838 /* Can't fail, the space was reserved. */
839 ASSERT(args.fsbno != NULLFSBLOCK);
840 ASSERT(args.len == 1);
841 error = xfs_trans_get_buf(tp, args.mp->m_ddev_targp,
842 XFS_FSB_TO_DADDR(args.mp, args.fsbno),
843 args.mp->m_bsize, 0, &bp);
844 if (error)
845 goto done;
846
847 /*
848 * Initialize the block, copy the data and log the remote buffer.
849 *
850 * The callout is responsible for logging because the remote format
851 * might differ from the local format and thus we don't know how much to
852 * log here. Note that init_fn must also set the buffer log item type
853 * correctly.
854 */
855 init_fn(tp, bp, ip, ifp, priv);
856
857 /* account for the change in fork size */
858 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
859 xfs_bmap_local_to_extents_empty(tp, ip, whichfork);
860 flags |= XFS_ILOG_CORE;
861
862 ifp->if_data = NULL;
863 ifp->if_height = 0;
864
865 rec.br_startoff = 0;
866 rec.br_startblock = args.fsbno;
867 rec.br_blockcount = 1;
868 rec.br_state = XFS_EXT_NORM;
869 xfs_iext_first(ifp, &icur);
870 xfs_iext_insert(ip, &icur, &rec, 0);
871
872 ifp->if_nextents = 1;
873 ip->i_nblocks = 1;
874 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
875 flags |= xfs_ilog_fext(whichfork);
876
877 done:
878 *logflagsp = flags;
879 return error;
880 }
881
882 /*
883 * Called from xfs_bmap_add_attrfork to handle btree format files.
884 */
885 STATIC int /* error */
xfs_bmap_add_attrfork_btree(xfs_trans_t * tp,xfs_inode_t * ip,int * flags)886 xfs_bmap_add_attrfork_btree(
887 xfs_trans_t *tp, /* transaction pointer */
888 xfs_inode_t *ip, /* incore inode pointer */
889 int *flags) /* inode logging flags */
890 {
891 struct xfs_btree_block *block = ip->i_df.if_broot;
892 struct xfs_btree_cur *cur; /* btree cursor */
893 int error; /* error return value */
894 xfs_mount_t *mp; /* file system mount struct */
895 int stat; /* newroot status */
896
897 mp = ip->i_mount;
898
899 if (xfs_bmap_bmdr_space(block) <= xfs_inode_data_fork_size(ip))
900 *flags |= XFS_ILOG_DBROOT;
901 else {
902 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
903 error = xfs_bmbt_lookup_first(cur, &stat);
904 if (error)
905 goto error0;
906 /* must be at least one entry */
907 if (XFS_IS_CORRUPT(mp, stat != 1)) {
908 xfs_btree_mark_sick(cur);
909 error = -EFSCORRUPTED;
910 goto error0;
911 }
912 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
913 goto error0;
914 if (stat == 0) {
915 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
916 return -ENOSPC;
917 }
918 cur->bc_bmap.allocated = 0;
919 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
920 }
921 return 0;
922 error0:
923 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
924 return error;
925 }
926
927 /*
928 * Called from xfs_bmap_add_attrfork to handle extents format files.
929 */
930 STATIC int /* error */
xfs_bmap_add_attrfork_extents(struct xfs_trans * tp,struct xfs_inode * ip,int * flags)931 xfs_bmap_add_attrfork_extents(
932 struct xfs_trans *tp, /* transaction pointer */
933 struct xfs_inode *ip, /* incore inode pointer */
934 int *flags) /* inode logging flags */
935 {
936 struct xfs_btree_cur *cur; /* bmap btree cursor */
937 int error; /* error return value */
938
939 if (ip->i_df.if_nextents * sizeof(struct xfs_bmbt_rec) <=
940 xfs_inode_data_fork_size(ip))
941 return 0;
942 cur = NULL;
943 error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0, flags,
944 XFS_DATA_FORK);
945 if (cur) {
946 cur->bc_bmap.allocated = 0;
947 xfs_btree_del_cursor(cur, error);
948 }
949 return error;
950 }
951
952 /*
953 * Called from xfs_bmap_add_attrfork to handle local format files. Each
954 * different data fork content type needs a different callout to do the
955 * conversion. Some are basic and only require special block initialisation
956 * callouts for the data formating, others (directories) are so specialised they
957 * handle everything themselves.
958 *
959 * XXX (dgc): investigate whether directory conversion can use the generic
960 * formatting callout. It should be possible - it's just a very complex
961 * formatter.
962 */
963 STATIC int /* error */
xfs_bmap_add_attrfork_local(struct xfs_trans * tp,struct xfs_inode * ip,int * flags)964 xfs_bmap_add_attrfork_local(
965 struct xfs_trans *tp, /* transaction pointer */
966 struct xfs_inode *ip, /* incore inode pointer */
967 int *flags) /* inode logging flags */
968 {
969 struct xfs_da_args dargs; /* args for dir/attr code */
970
971 if (ip->i_df.if_bytes <= xfs_inode_data_fork_size(ip))
972 return 0;
973
974 if (S_ISDIR(VFS_I(ip)->i_mode)) {
975 memset(&dargs, 0, sizeof(dargs));
976 dargs.geo = ip->i_mount->m_dir_geo;
977 dargs.dp = ip;
978 dargs.total = dargs.geo->fsbcount;
979 dargs.whichfork = XFS_DATA_FORK;
980 dargs.trans = tp;
981 dargs.owner = ip->i_ino;
982 return xfs_dir2_sf_to_block(&dargs);
983 }
984
985 if (S_ISLNK(VFS_I(ip)->i_mode))
986 return xfs_bmap_local_to_extents(tp, ip, 1, flags,
987 XFS_DATA_FORK, xfs_symlink_local_to_remote,
988 NULL);
989
990 /* should only be called for types that support local format data */
991 ASSERT(0);
992 xfs_bmap_mark_sick(ip, XFS_ATTR_FORK);
993 return -EFSCORRUPTED;
994 }
995
996 /*
997 * Set an inode attr fork offset based on the format of the data fork.
998 */
999 static int
xfs_bmap_set_attrforkoff(struct xfs_inode * ip,int size,int * version)1000 xfs_bmap_set_attrforkoff(
1001 struct xfs_inode *ip,
1002 int size,
1003 int *version)
1004 {
1005 int default_size = xfs_default_attroffset(ip) >> 3;
1006
1007 switch (ip->i_df.if_format) {
1008 case XFS_DINODE_FMT_DEV:
1009 ip->i_forkoff = default_size;
1010 break;
1011 case XFS_DINODE_FMT_LOCAL:
1012 case XFS_DINODE_FMT_EXTENTS:
1013 case XFS_DINODE_FMT_BTREE:
1014 ip->i_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1015 if (!ip->i_forkoff)
1016 ip->i_forkoff = default_size;
1017 else if (xfs_has_attr2(ip->i_mount) && version)
1018 *version = 2;
1019 break;
1020 default:
1021 ASSERT(0);
1022 return -EINVAL;
1023 }
1024
1025 return 0;
1026 }
1027
1028 /*
1029 * Convert inode from non-attributed to attributed. Caller must hold the
1030 * ILOCK_EXCL and the file cannot have an attr fork.
1031 */
1032 int /* error code */
xfs_bmap_add_attrfork(struct xfs_trans * tp,struct xfs_inode * ip,int size,int rsvd)1033 xfs_bmap_add_attrfork(
1034 struct xfs_trans *tp,
1035 struct xfs_inode *ip, /* incore inode pointer */
1036 int size, /* space new attribute needs */
1037 int rsvd) /* xact may use reserved blks */
1038 {
1039 struct xfs_mount *mp = tp->t_mountp;
1040 int version = 1; /* superblock attr version */
1041 int logflags; /* logging flags */
1042 int error; /* error return value */
1043
1044 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
1045 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1046 ASSERT(!xfs_inode_has_attr_fork(ip));
1047
1048 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1049 error = xfs_bmap_set_attrforkoff(ip, size, &version);
1050 if (error)
1051 return error;
1052
1053 xfs_ifork_init_attr(ip, XFS_DINODE_FMT_EXTENTS, 0);
1054 logflags = 0;
1055 switch (ip->i_df.if_format) {
1056 case XFS_DINODE_FMT_LOCAL:
1057 error = xfs_bmap_add_attrfork_local(tp, ip, &logflags);
1058 break;
1059 case XFS_DINODE_FMT_EXTENTS:
1060 error = xfs_bmap_add_attrfork_extents(tp, ip, &logflags);
1061 break;
1062 case XFS_DINODE_FMT_BTREE:
1063 error = xfs_bmap_add_attrfork_btree(tp, ip, &logflags);
1064 break;
1065 default:
1066 error = 0;
1067 break;
1068 }
1069 if (logflags)
1070 xfs_trans_log_inode(tp, ip, logflags);
1071 if (error)
1072 return error;
1073 if (!xfs_has_attr(mp) ||
1074 (!xfs_has_attr2(mp) && version == 2)) {
1075 bool log_sb = false;
1076
1077 spin_lock(&mp->m_sb_lock);
1078 if (!xfs_has_attr(mp)) {
1079 xfs_add_attr(mp);
1080 log_sb = true;
1081 }
1082 if (!xfs_has_attr2(mp) && version == 2) {
1083 xfs_add_attr2(mp);
1084 log_sb = true;
1085 }
1086 spin_unlock(&mp->m_sb_lock);
1087 if (log_sb)
1088 xfs_log_sb(tp);
1089 }
1090
1091 return 0;
1092 }
1093
1094 /*
1095 * Internal and external extent tree search functions.
1096 */
1097
1098 struct xfs_iread_state {
1099 struct xfs_iext_cursor icur;
1100 xfs_extnum_t loaded;
1101 };
1102
1103 int
xfs_bmap_complain_bad_rec(struct xfs_inode * ip,int whichfork,xfs_failaddr_t fa,const struct xfs_bmbt_irec * irec)1104 xfs_bmap_complain_bad_rec(
1105 struct xfs_inode *ip,
1106 int whichfork,
1107 xfs_failaddr_t fa,
1108 const struct xfs_bmbt_irec *irec)
1109 {
1110 struct xfs_mount *mp = ip->i_mount;
1111 const char *forkname;
1112
1113 switch (whichfork) {
1114 case XFS_DATA_FORK: forkname = "data"; break;
1115 case XFS_ATTR_FORK: forkname = "attr"; break;
1116 case XFS_COW_FORK: forkname = "CoW"; break;
1117 default: forkname = "???"; break;
1118 }
1119
1120 xfs_warn(mp,
1121 "Bmap BTree record corruption in inode 0x%llx %s fork detected at %pS!",
1122 ip->i_ino, forkname, fa);
1123 xfs_warn(mp,
1124 "Offset 0x%llx, start block 0x%llx, block count 0x%llx state 0x%x",
1125 irec->br_startoff, irec->br_startblock, irec->br_blockcount,
1126 irec->br_state);
1127
1128 return -EFSCORRUPTED;
1129 }
1130
1131 /* Stuff every bmbt record from this block into the incore extent map. */
1132 static int
xfs_iread_bmbt_block(struct xfs_btree_cur * cur,int level,void * priv)1133 xfs_iread_bmbt_block(
1134 struct xfs_btree_cur *cur,
1135 int level,
1136 void *priv)
1137 {
1138 struct xfs_iread_state *ir = priv;
1139 struct xfs_mount *mp = cur->bc_mp;
1140 struct xfs_inode *ip = cur->bc_ino.ip;
1141 struct xfs_btree_block *block;
1142 struct xfs_buf *bp;
1143 struct xfs_bmbt_rec *frp;
1144 xfs_extnum_t num_recs;
1145 xfs_extnum_t j;
1146 int whichfork = cur->bc_ino.whichfork;
1147 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
1148
1149 block = xfs_btree_get_block(cur, level, &bp);
1150
1151 /* Abort if we find more records than nextents. */
1152 num_recs = xfs_btree_get_numrecs(block);
1153 if (unlikely(ir->loaded + num_recs > ifp->if_nextents)) {
1154 xfs_warn(ip->i_mount, "corrupt dinode %llu, (btree extents).",
1155 (unsigned long long)ip->i_ino);
1156 xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__, block,
1157 sizeof(*block), __this_address);
1158 xfs_bmap_mark_sick(ip, whichfork);
1159 return -EFSCORRUPTED;
1160 }
1161
1162 /* Copy records into the incore cache. */
1163 frp = xfs_bmbt_rec_addr(mp, block, 1);
1164 for (j = 0; j < num_recs; j++, frp++, ir->loaded++) {
1165 struct xfs_bmbt_irec new;
1166 xfs_failaddr_t fa;
1167
1168 xfs_bmbt_disk_get_all(frp, &new);
1169 fa = xfs_bmap_validate_extent(ip, whichfork, &new);
1170 if (fa) {
1171 xfs_inode_verifier_error(ip, -EFSCORRUPTED,
1172 "xfs_iread_extents(2)", frp,
1173 sizeof(*frp), fa);
1174 xfs_bmap_mark_sick(ip, whichfork);
1175 return xfs_bmap_complain_bad_rec(ip, whichfork, fa,
1176 &new);
1177 }
1178 xfs_iext_insert(ip, &ir->icur, &new,
1179 xfs_bmap_fork_to_state(whichfork));
1180 trace_xfs_read_extent(ip, &ir->icur,
1181 xfs_bmap_fork_to_state(whichfork), _THIS_IP_);
1182 xfs_iext_next(ifp, &ir->icur);
1183 }
1184
1185 return 0;
1186 }
1187
1188 /*
1189 * Read in extents from a btree-format inode.
1190 */
1191 int
xfs_iread_extents(struct xfs_trans * tp,struct xfs_inode * ip,int whichfork)1192 xfs_iread_extents(
1193 struct xfs_trans *tp,
1194 struct xfs_inode *ip,
1195 int whichfork)
1196 {
1197 struct xfs_iread_state ir;
1198 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
1199 struct xfs_mount *mp = ip->i_mount;
1200 struct xfs_btree_cur *cur;
1201 int error;
1202
1203 if (!xfs_need_iread_extents(ifp))
1204 return 0;
1205
1206 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
1207
1208 ir.loaded = 0;
1209 xfs_iext_first(ifp, &ir.icur);
1210 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
1211 error = xfs_btree_visit_blocks(cur, xfs_iread_bmbt_block,
1212 XFS_BTREE_VISIT_RECORDS, &ir);
1213 xfs_btree_del_cursor(cur, error);
1214 if (error)
1215 goto out;
1216
1217 if (XFS_IS_CORRUPT(mp, ir.loaded != ifp->if_nextents)) {
1218 xfs_bmap_mark_sick(ip, whichfork);
1219 error = -EFSCORRUPTED;
1220 goto out;
1221 }
1222 ASSERT(ir.loaded == xfs_iext_count(ifp));
1223 /*
1224 * Use release semantics so that we can use acquire semantics in
1225 * xfs_need_iread_extents and be guaranteed to see a valid mapping tree
1226 * after that load.
1227 */
1228 smp_store_release(&ifp->if_needextents, 0);
1229 return 0;
1230 out:
1231 if (xfs_metadata_is_sick(error))
1232 xfs_bmap_mark_sick(ip, whichfork);
1233 xfs_iext_destroy(ifp);
1234 return error;
1235 }
1236
1237 /*
1238 * Returns the relative block number of the first unused block(s) in the given
1239 * fork with at least "len" logically contiguous blocks free. This is the
1240 * lowest-address hole if the fork has holes, else the first block past the end
1241 * of fork. Return 0 if the fork is currently local (in-inode).
1242 */
1243 int /* error */
xfs_bmap_first_unused(struct xfs_trans * tp,struct xfs_inode * ip,xfs_extlen_t len,xfs_fileoff_t * first_unused,int whichfork)1244 xfs_bmap_first_unused(
1245 struct xfs_trans *tp, /* transaction pointer */
1246 struct xfs_inode *ip, /* incore inode */
1247 xfs_extlen_t len, /* size of hole to find */
1248 xfs_fileoff_t *first_unused, /* unused block */
1249 int whichfork) /* data or attr fork */
1250 {
1251 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
1252 struct xfs_bmbt_irec got;
1253 struct xfs_iext_cursor icur;
1254 xfs_fileoff_t lastaddr = 0;
1255 xfs_fileoff_t lowest, max;
1256 int error;
1257
1258 if (ifp->if_format == XFS_DINODE_FMT_LOCAL) {
1259 *first_unused = 0;
1260 return 0;
1261 }
1262
1263 ASSERT(xfs_ifork_has_extents(ifp));
1264
1265 error = xfs_iread_extents(tp, ip, whichfork);
1266 if (error)
1267 return error;
1268
1269 lowest = max = *first_unused;
1270 for_each_xfs_iext(ifp, &icur, &got) {
1271 /*
1272 * See if the hole before this extent will work.
1273 */
1274 if (got.br_startoff >= lowest + len &&
1275 got.br_startoff - max >= len)
1276 break;
1277 lastaddr = got.br_startoff + got.br_blockcount;
1278 max = XFS_FILEOFF_MAX(lastaddr, lowest);
1279 }
1280
1281 *first_unused = max;
1282 return 0;
1283 }
1284
1285 /*
1286 * Returns the file-relative block number of the last block - 1 before
1287 * last_block (input value) in the file.
1288 * This is not based on i_size, it is based on the extent records.
1289 * Returns 0 for local files, as they do not have extent records.
1290 */
1291 int /* error */
xfs_bmap_last_before(struct xfs_trans * tp,struct xfs_inode * ip,xfs_fileoff_t * last_block,int whichfork)1292 xfs_bmap_last_before(
1293 struct xfs_trans *tp, /* transaction pointer */
1294 struct xfs_inode *ip, /* incore inode */
1295 xfs_fileoff_t *last_block, /* last block */
1296 int whichfork) /* data or attr fork */
1297 {
1298 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
1299 struct xfs_bmbt_irec got;
1300 struct xfs_iext_cursor icur;
1301 int error;
1302
1303 switch (ifp->if_format) {
1304 case XFS_DINODE_FMT_LOCAL:
1305 *last_block = 0;
1306 return 0;
1307 case XFS_DINODE_FMT_BTREE:
1308 case XFS_DINODE_FMT_EXTENTS:
1309 break;
1310 default:
1311 ASSERT(0);
1312 xfs_bmap_mark_sick(ip, whichfork);
1313 return -EFSCORRUPTED;
1314 }
1315
1316 error = xfs_iread_extents(tp, ip, whichfork);
1317 if (error)
1318 return error;
1319
1320 if (!xfs_iext_lookup_extent_before(ip, ifp, last_block, &icur, &got))
1321 *last_block = 0;
1322 return 0;
1323 }
1324
1325 int
xfs_bmap_last_extent(struct xfs_trans * tp,struct xfs_inode * ip,int whichfork,struct xfs_bmbt_irec * rec,int * is_empty)1326 xfs_bmap_last_extent(
1327 struct xfs_trans *tp,
1328 struct xfs_inode *ip,
1329 int whichfork,
1330 struct xfs_bmbt_irec *rec,
1331 int *is_empty)
1332 {
1333 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
1334 struct xfs_iext_cursor icur;
1335 int error;
1336
1337 error = xfs_iread_extents(tp, ip, whichfork);
1338 if (error)
1339 return error;
1340
1341 xfs_iext_last(ifp, &icur);
1342 if (!xfs_iext_get_extent(ifp, &icur, rec))
1343 *is_empty = 1;
1344 else
1345 *is_empty = 0;
1346 return 0;
1347 }
1348
1349 /*
1350 * Check the last inode extent to determine whether this allocation will result
1351 * in blocks being allocated at the end of the file. When we allocate new data
1352 * blocks at the end of the file which do not start at the previous data block,
1353 * we will try to align the new blocks at stripe unit boundaries.
1354 *
1355 * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1356 * at, or past the EOF.
1357 */
1358 STATIC int
xfs_bmap_isaeof(struct xfs_bmalloca * bma,int whichfork)1359 xfs_bmap_isaeof(
1360 struct xfs_bmalloca *bma,
1361 int whichfork)
1362 {
1363 struct xfs_bmbt_irec rec;
1364 int is_empty;
1365 int error;
1366
1367 bma->aeof = false;
1368 error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1369 &is_empty);
1370 if (error)
1371 return error;
1372
1373 if (is_empty) {
1374 bma->aeof = true;
1375 return 0;
1376 }
1377
1378 /*
1379 * Check if we are allocation or past the last extent, or at least into
1380 * the last delayed allocated extent.
1381 */
1382 bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1383 (bma->offset >= rec.br_startoff &&
1384 isnullstartblock(rec.br_startblock));
1385 return 0;
1386 }
1387
1388 /*
1389 * Returns the file-relative block number of the first block past eof in
1390 * the file. This is not based on i_size, it is based on the extent records.
1391 * Returns 0 for local files, as they do not have extent records.
1392 */
1393 int
xfs_bmap_last_offset(struct xfs_inode * ip,xfs_fileoff_t * last_block,int whichfork)1394 xfs_bmap_last_offset(
1395 struct xfs_inode *ip,
1396 xfs_fileoff_t *last_block,
1397 int whichfork)
1398 {
1399 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
1400 struct xfs_bmbt_irec rec;
1401 int is_empty;
1402 int error;
1403
1404 *last_block = 0;
1405
1406 if (ifp->if_format == XFS_DINODE_FMT_LOCAL)
1407 return 0;
1408
1409 if (XFS_IS_CORRUPT(ip->i_mount, !xfs_ifork_has_extents(ifp))) {
1410 xfs_bmap_mark_sick(ip, whichfork);
1411 return -EFSCORRUPTED;
1412 }
1413
1414 error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1415 if (error || is_empty)
1416 return error;
1417
1418 *last_block = rec.br_startoff + rec.br_blockcount;
1419 return 0;
1420 }
1421
1422 /*
1423 * Extent tree manipulation functions used during allocation.
1424 */
1425
1426 /*
1427 * Convert a delayed allocation to a real allocation.
1428 */
1429 STATIC int /* error */
xfs_bmap_add_extent_delay_real(struct xfs_bmalloca * bma,int whichfork)1430 xfs_bmap_add_extent_delay_real(
1431 struct xfs_bmalloca *bma,
1432 int whichfork)
1433 {
1434 struct xfs_mount *mp = bma->ip->i_mount;
1435 struct xfs_ifork *ifp = xfs_ifork_ptr(bma->ip, whichfork);
1436 struct xfs_bmbt_irec *new = &bma->got;
1437 int error; /* error return value */
1438 int i; /* temp state */
1439 xfs_fileoff_t new_endoff; /* end offset of new entry */
1440 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
1441 /* left is 0, right is 1, prev is 2 */
1442 int rval=0; /* return value (logging flags) */
1443 uint32_t state = xfs_bmap_fork_to_state(whichfork);
1444 xfs_filblks_t da_new; /* new count del alloc blocks used */
1445 xfs_filblks_t da_old; /* old count del alloc blocks used */
1446 xfs_filblks_t temp=0; /* value for da_new calculations */
1447 int tmp_rval; /* partial logging flags */
1448 struct xfs_bmbt_irec old;
1449
1450 ASSERT(whichfork != XFS_ATTR_FORK);
1451 ASSERT(!isnullstartblock(new->br_startblock));
1452 ASSERT(!bma->cur || (bma->cur->bc_flags & XFS_BTREE_BMBT_WASDEL));
1453
1454 XFS_STATS_INC(mp, xs_add_exlist);
1455
1456 #define LEFT r[0]
1457 #define RIGHT r[1]
1458 #define PREV r[2]
1459
1460 /*
1461 * Set up a bunch of variables to make the tests simpler.
1462 */
1463 xfs_iext_get_extent(ifp, &bma->icur, &PREV);
1464 new_endoff = new->br_startoff + new->br_blockcount;
1465 ASSERT(isnullstartblock(PREV.br_startblock));
1466 ASSERT(PREV.br_startoff <= new->br_startoff);
1467 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1468
1469 da_old = startblockval(PREV.br_startblock);
1470 da_new = 0;
1471
1472 /*
1473 * Set flags determining what part of the previous delayed allocation
1474 * extent is being replaced by a real allocation.
1475 */
1476 if (PREV.br_startoff == new->br_startoff)
1477 state |= BMAP_LEFT_FILLING;
1478 if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1479 state |= BMAP_RIGHT_FILLING;
1480
1481 /*
1482 * Check and set flags if this segment has a left neighbor.
1483 * Don't set contiguous if the combined extent would be too large.
1484 */
1485 if (xfs_iext_peek_prev_extent(ifp, &bma->icur, &LEFT)) {
1486 state |= BMAP_LEFT_VALID;
1487 if (isnullstartblock(LEFT.br_startblock))
1488 state |= BMAP_LEFT_DELAY;
1489 }
1490
1491 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1492 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1493 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1494 LEFT.br_state == new->br_state &&
1495 LEFT.br_blockcount + new->br_blockcount <= XFS_MAX_BMBT_EXTLEN)
1496 state |= BMAP_LEFT_CONTIG;
1497
1498 /*
1499 * Check and set flags if this segment has a right neighbor.
1500 * Don't set contiguous if the combined extent would be too large.
1501 * Also check for all-three-contiguous being too large.
1502 */
1503 if (xfs_iext_peek_next_extent(ifp, &bma->icur, &RIGHT)) {
1504 state |= BMAP_RIGHT_VALID;
1505 if (isnullstartblock(RIGHT.br_startblock))
1506 state |= BMAP_RIGHT_DELAY;
1507 }
1508
1509 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1510 new_endoff == RIGHT.br_startoff &&
1511 new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1512 new->br_state == RIGHT.br_state &&
1513 new->br_blockcount + RIGHT.br_blockcount <= XFS_MAX_BMBT_EXTLEN &&
1514 ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1515 BMAP_RIGHT_FILLING)) !=
1516 (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1517 BMAP_RIGHT_FILLING) ||
1518 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1519 <= XFS_MAX_BMBT_EXTLEN))
1520 state |= BMAP_RIGHT_CONTIG;
1521
1522 error = 0;
1523 /*
1524 * Switch out based on the FILLING and CONTIG state bits.
1525 */
1526 switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1527 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1528 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1529 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1530 /*
1531 * Filling in all of a previously delayed allocation extent.
1532 * The left and right neighbors are both contiguous with new.
1533 */
1534 LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount;
1535
1536 xfs_iext_remove(bma->ip, &bma->icur, state);
1537 xfs_iext_remove(bma->ip, &bma->icur, state);
1538 xfs_iext_prev(ifp, &bma->icur);
1539 xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1540 ifp->if_nextents--;
1541
1542 if (bma->cur == NULL)
1543 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1544 else {
1545 rval = XFS_ILOG_CORE;
1546 error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i);
1547 if (error)
1548 goto done;
1549 if (XFS_IS_CORRUPT(mp, i != 1)) {
1550 xfs_btree_mark_sick(bma->cur);
1551 error = -EFSCORRUPTED;
1552 goto done;
1553 }
1554 error = xfs_btree_delete(bma->cur, &i);
1555 if (error)
1556 goto done;
1557 if (XFS_IS_CORRUPT(mp, i != 1)) {
1558 xfs_btree_mark_sick(bma->cur);
1559 error = -EFSCORRUPTED;
1560 goto done;
1561 }
1562 error = xfs_btree_decrement(bma->cur, 0, &i);
1563 if (error)
1564 goto done;
1565 if (XFS_IS_CORRUPT(mp, i != 1)) {
1566 xfs_btree_mark_sick(bma->cur);
1567 error = -EFSCORRUPTED;
1568 goto done;
1569 }
1570 error = xfs_bmbt_update(bma->cur, &LEFT);
1571 if (error)
1572 goto done;
1573 }
1574 ASSERT(da_new <= da_old);
1575 break;
1576
1577 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1578 /*
1579 * Filling in all of a previously delayed allocation extent.
1580 * The left neighbor is contiguous, the right is not.
1581 */
1582 old = LEFT;
1583 LEFT.br_blockcount += PREV.br_blockcount;
1584
1585 xfs_iext_remove(bma->ip, &bma->icur, state);
1586 xfs_iext_prev(ifp, &bma->icur);
1587 xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1588
1589 if (bma->cur == NULL)
1590 rval = XFS_ILOG_DEXT;
1591 else {
1592 rval = 0;
1593 error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1594 if (error)
1595 goto done;
1596 if (XFS_IS_CORRUPT(mp, i != 1)) {
1597 xfs_btree_mark_sick(bma->cur);
1598 error = -EFSCORRUPTED;
1599 goto done;
1600 }
1601 error = xfs_bmbt_update(bma->cur, &LEFT);
1602 if (error)
1603 goto done;
1604 }
1605 ASSERT(da_new <= da_old);
1606 break;
1607
1608 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1609 /*
1610 * Filling in all of a previously delayed allocation extent.
1611 * The right neighbor is contiguous, the left is not. Take care
1612 * with delay -> unwritten extent allocation here because the
1613 * delalloc record we are overwriting is always written.
1614 */
1615 PREV.br_startblock = new->br_startblock;
1616 PREV.br_blockcount += RIGHT.br_blockcount;
1617 PREV.br_state = new->br_state;
1618
1619 xfs_iext_next(ifp, &bma->icur);
1620 xfs_iext_remove(bma->ip, &bma->icur, state);
1621 xfs_iext_prev(ifp, &bma->icur);
1622 xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1623
1624 if (bma->cur == NULL)
1625 rval = XFS_ILOG_DEXT;
1626 else {
1627 rval = 0;
1628 error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i);
1629 if (error)
1630 goto done;
1631 if (XFS_IS_CORRUPT(mp, i != 1)) {
1632 xfs_btree_mark_sick(bma->cur);
1633 error = -EFSCORRUPTED;
1634 goto done;
1635 }
1636 error = xfs_bmbt_update(bma->cur, &PREV);
1637 if (error)
1638 goto done;
1639 }
1640 ASSERT(da_new <= da_old);
1641 break;
1642
1643 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1644 /*
1645 * Filling in all of a previously delayed allocation extent.
1646 * Neither the left nor right neighbors are contiguous with
1647 * the new one.
1648 */
1649 PREV.br_startblock = new->br_startblock;
1650 PREV.br_state = new->br_state;
1651 xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1652 ifp->if_nextents++;
1653
1654 if (bma->cur == NULL)
1655 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1656 else {
1657 rval = XFS_ILOG_CORE;
1658 error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1659 if (error)
1660 goto done;
1661 if (XFS_IS_CORRUPT(mp, i != 0)) {
1662 xfs_btree_mark_sick(bma->cur);
1663 error = -EFSCORRUPTED;
1664 goto done;
1665 }
1666 error = xfs_btree_insert(bma->cur, &i);
1667 if (error)
1668 goto done;
1669 if (XFS_IS_CORRUPT(mp, i != 1)) {
1670 xfs_btree_mark_sick(bma->cur);
1671 error = -EFSCORRUPTED;
1672 goto done;
1673 }
1674 }
1675 ASSERT(da_new <= da_old);
1676 break;
1677
1678 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1679 /*
1680 * Filling in the first part of a previous delayed allocation.
1681 * The left neighbor is contiguous.
1682 */
1683 old = LEFT;
1684 temp = PREV.br_blockcount - new->br_blockcount;
1685 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1686 startblockval(PREV.br_startblock));
1687
1688 LEFT.br_blockcount += new->br_blockcount;
1689
1690 PREV.br_blockcount = temp;
1691 PREV.br_startoff += new->br_blockcount;
1692 PREV.br_startblock = nullstartblock(da_new);
1693
1694 xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1695 xfs_iext_prev(ifp, &bma->icur);
1696 xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1697
1698 if (bma->cur == NULL)
1699 rval = XFS_ILOG_DEXT;
1700 else {
1701 rval = 0;
1702 error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1703 if (error)
1704 goto done;
1705 if (XFS_IS_CORRUPT(mp, i != 1)) {
1706 xfs_btree_mark_sick(bma->cur);
1707 error = -EFSCORRUPTED;
1708 goto done;
1709 }
1710 error = xfs_bmbt_update(bma->cur, &LEFT);
1711 if (error)
1712 goto done;
1713 }
1714 ASSERT(da_new <= da_old);
1715 break;
1716
1717 case BMAP_LEFT_FILLING:
1718 /*
1719 * Filling in the first part of a previous delayed allocation.
1720 * The left neighbor is not contiguous.
1721 */
1722 xfs_iext_update_extent(bma->ip, state, &bma->icur, new);
1723 ifp->if_nextents++;
1724
1725 if (bma->cur == NULL)
1726 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1727 else {
1728 rval = XFS_ILOG_CORE;
1729 error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1730 if (error)
1731 goto done;
1732 if (XFS_IS_CORRUPT(mp, i != 0)) {
1733 xfs_btree_mark_sick(bma->cur);
1734 error = -EFSCORRUPTED;
1735 goto done;
1736 }
1737 error = xfs_btree_insert(bma->cur, &i);
1738 if (error)
1739 goto done;
1740 if (XFS_IS_CORRUPT(mp, i != 1)) {
1741 xfs_btree_mark_sick(bma->cur);
1742 error = -EFSCORRUPTED;
1743 goto done;
1744 }
1745 }
1746
1747 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1748 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1749 &bma->cur, 1, &tmp_rval, whichfork);
1750 rval |= tmp_rval;
1751 if (error)
1752 goto done;
1753 }
1754
1755 temp = PREV.br_blockcount - new->br_blockcount;
1756 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1757 startblockval(PREV.br_startblock) -
1758 (bma->cur ? bma->cur->bc_bmap.allocated : 0));
1759
1760 PREV.br_startoff = new_endoff;
1761 PREV.br_blockcount = temp;
1762 PREV.br_startblock = nullstartblock(da_new);
1763 xfs_iext_next(ifp, &bma->icur);
1764 xfs_iext_insert(bma->ip, &bma->icur, &PREV, state);
1765 xfs_iext_prev(ifp, &bma->icur);
1766 break;
1767
1768 case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1769 /*
1770 * Filling in the last part of a previous delayed allocation.
1771 * The right neighbor is contiguous with the new allocation.
1772 */
1773 old = RIGHT;
1774 RIGHT.br_startoff = new->br_startoff;
1775 RIGHT.br_startblock = new->br_startblock;
1776 RIGHT.br_blockcount += new->br_blockcount;
1777
1778 if (bma->cur == NULL)
1779 rval = XFS_ILOG_DEXT;
1780 else {
1781 rval = 0;
1782 error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1783 if (error)
1784 goto done;
1785 if (XFS_IS_CORRUPT(mp, i != 1)) {
1786 xfs_btree_mark_sick(bma->cur);
1787 error = -EFSCORRUPTED;
1788 goto done;
1789 }
1790 error = xfs_bmbt_update(bma->cur, &RIGHT);
1791 if (error)
1792 goto done;
1793 }
1794
1795 temp = PREV.br_blockcount - new->br_blockcount;
1796 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1797 startblockval(PREV.br_startblock));
1798
1799 PREV.br_blockcount = temp;
1800 PREV.br_startblock = nullstartblock(da_new);
1801
1802 xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1803 xfs_iext_next(ifp, &bma->icur);
1804 xfs_iext_update_extent(bma->ip, state, &bma->icur, &RIGHT);
1805 ASSERT(da_new <= da_old);
1806 break;
1807
1808 case BMAP_RIGHT_FILLING:
1809 /*
1810 * Filling in the last part of a previous delayed allocation.
1811 * The right neighbor is not contiguous.
1812 */
1813 xfs_iext_update_extent(bma->ip, state, &bma->icur, new);
1814 ifp->if_nextents++;
1815
1816 if (bma->cur == NULL)
1817 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1818 else {
1819 rval = XFS_ILOG_CORE;
1820 error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1821 if (error)
1822 goto done;
1823 if (XFS_IS_CORRUPT(mp, i != 0)) {
1824 xfs_btree_mark_sick(bma->cur);
1825 error = -EFSCORRUPTED;
1826 goto done;
1827 }
1828 error = xfs_btree_insert(bma->cur, &i);
1829 if (error)
1830 goto done;
1831 if (XFS_IS_CORRUPT(mp, i != 1)) {
1832 xfs_btree_mark_sick(bma->cur);
1833 error = -EFSCORRUPTED;
1834 goto done;
1835 }
1836 }
1837
1838 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1839 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1840 &bma->cur, 1, &tmp_rval, whichfork);
1841 rval |= tmp_rval;
1842 if (error)
1843 goto done;
1844 }
1845
1846 temp = PREV.br_blockcount - new->br_blockcount;
1847 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1848 startblockval(PREV.br_startblock) -
1849 (bma->cur ? bma->cur->bc_bmap.allocated : 0));
1850
1851 PREV.br_startblock = nullstartblock(da_new);
1852 PREV.br_blockcount = temp;
1853 xfs_iext_insert(bma->ip, &bma->icur, &PREV, state);
1854 xfs_iext_next(ifp, &bma->icur);
1855 ASSERT(da_new <= da_old);
1856 break;
1857
1858 case 0:
1859 /*
1860 * Filling in the middle part of a previous delayed allocation.
1861 * Contiguity is impossible here.
1862 * This case is avoided almost all the time.
1863 *
1864 * We start with a delayed allocation:
1865 *
1866 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
1867 * PREV @ idx
1868 *
1869 * and we are allocating:
1870 * +rrrrrrrrrrrrrrrrr+
1871 * new
1872 *
1873 * and we set it up for insertion as:
1874 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
1875 * new
1876 * PREV @ idx LEFT RIGHT
1877 * inserted at idx + 1
1878 */
1879 old = PREV;
1880
1881 /* LEFT is the new middle */
1882 LEFT = *new;
1883
1884 /* RIGHT is the new right */
1885 RIGHT.br_state = PREV.br_state;
1886 RIGHT.br_startoff = new_endoff;
1887 RIGHT.br_blockcount =
1888 PREV.br_startoff + PREV.br_blockcount - new_endoff;
1889 RIGHT.br_startblock =
1890 nullstartblock(xfs_bmap_worst_indlen(bma->ip,
1891 RIGHT.br_blockcount));
1892
1893 /* truncate PREV */
1894 PREV.br_blockcount = new->br_startoff - PREV.br_startoff;
1895 PREV.br_startblock =
1896 nullstartblock(xfs_bmap_worst_indlen(bma->ip,
1897 PREV.br_blockcount));
1898 xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1899
1900 xfs_iext_next(ifp, &bma->icur);
1901 xfs_iext_insert(bma->ip, &bma->icur, &RIGHT, state);
1902 xfs_iext_insert(bma->ip, &bma->icur, &LEFT, state);
1903 ifp->if_nextents++;
1904
1905 if (bma->cur == NULL)
1906 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1907 else {
1908 rval = XFS_ILOG_CORE;
1909 error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1910 if (error)
1911 goto done;
1912 if (XFS_IS_CORRUPT(mp, i != 0)) {
1913 xfs_btree_mark_sick(bma->cur);
1914 error = -EFSCORRUPTED;
1915 goto done;
1916 }
1917 error = xfs_btree_insert(bma->cur, &i);
1918 if (error)
1919 goto done;
1920 if (XFS_IS_CORRUPT(mp, i != 1)) {
1921 xfs_btree_mark_sick(bma->cur);
1922 error = -EFSCORRUPTED;
1923 goto done;
1924 }
1925 }
1926
1927 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1928 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1929 &bma->cur, 1, &tmp_rval, whichfork);
1930 rval |= tmp_rval;
1931 if (error)
1932 goto done;
1933 }
1934
1935 da_new = startblockval(PREV.br_startblock) +
1936 startblockval(RIGHT.br_startblock);
1937 break;
1938
1939 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1940 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1941 case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1942 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1943 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1944 case BMAP_LEFT_CONTIG:
1945 case BMAP_RIGHT_CONTIG:
1946 /*
1947 * These cases are all impossible.
1948 */
1949 ASSERT(0);
1950 }
1951
1952 /* add reverse mapping unless caller opted out */
1953 if (!(bma->flags & XFS_BMAPI_NORMAP))
1954 xfs_rmap_map_extent(bma->tp, bma->ip, whichfork, new);
1955
1956 /* convert to a btree if necessary */
1957 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1958 int tmp_logflags; /* partial log flag return val */
1959
1960 ASSERT(bma->cur == NULL);
1961 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1962 &bma->cur, da_old > 0, &tmp_logflags,
1963 whichfork);
1964 bma->logflags |= tmp_logflags;
1965 if (error)
1966 goto done;
1967 }
1968
1969 if (da_new != da_old)
1970 xfs_mod_delalloc(bma->ip, 0, (int64_t)da_new - da_old);
1971
1972 if (bma->cur) {
1973 da_new += bma->cur->bc_bmap.allocated;
1974 bma->cur->bc_bmap.allocated = 0;
1975 }
1976
1977 /* adjust for changes in reserved delayed indirect blocks */
1978 if (da_new < da_old)
1979 xfs_add_fdblocks(mp, da_old - da_new);
1980 else if (da_new > da_old)
1981 error = xfs_dec_fdblocks(mp, da_new - da_old, true);
1982
1983 xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
1984 done:
1985 if (whichfork != XFS_COW_FORK)
1986 bma->logflags |= rval;
1987 return error;
1988 #undef LEFT
1989 #undef RIGHT
1990 #undef PREV
1991 }
1992
1993 /*
1994 * Convert an unwritten allocation to a real allocation or vice versa.
1995 */
1996 int /* error */
xfs_bmap_add_extent_unwritten_real(struct xfs_trans * tp,xfs_inode_t * ip,int whichfork,struct xfs_iext_cursor * icur,struct xfs_btree_cur ** curp,xfs_bmbt_irec_t * new,int * logflagsp)1997 xfs_bmap_add_extent_unwritten_real(
1998 struct xfs_trans *tp,
1999 xfs_inode_t *ip, /* incore inode pointer */
2000 int whichfork,
2001 struct xfs_iext_cursor *icur,
2002 struct xfs_btree_cur **curp, /* if *curp is null, not a btree */
2003 xfs_bmbt_irec_t *new, /* new data to add to file extents */
2004 int *logflagsp) /* inode logging flags */
2005 {
2006 struct xfs_btree_cur *cur; /* btree cursor */
2007 int error; /* error return value */
2008 int i; /* temp state */
2009 struct xfs_ifork *ifp; /* inode fork pointer */
2010 xfs_fileoff_t new_endoff; /* end offset of new entry */
2011 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
2012 /* left is 0, right is 1, prev is 2 */
2013 int rval=0; /* return value (logging flags) */
2014 uint32_t state = xfs_bmap_fork_to_state(whichfork);
2015 struct xfs_mount *mp = ip->i_mount;
2016 struct xfs_bmbt_irec old;
2017
2018 *logflagsp = 0;
2019
2020 cur = *curp;
2021 ifp = xfs_ifork_ptr(ip, whichfork);
2022
2023 ASSERT(!isnullstartblock(new->br_startblock));
2024
2025 XFS_STATS_INC(mp, xs_add_exlist);
2026
2027 #define LEFT r[0]
2028 #define RIGHT r[1]
2029 #define PREV r[2]
2030
2031 /*
2032 * Set up a bunch of variables to make the tests simpler.
2033 */
2034 error = 0;
2035 xfs_iext_get_extent(ifp, icur, &PREV);
2036 ASSERT(new->br_state != PREV.br_state);
2037 new_endoff = new->br_startoff + new->br_blockcount;
2038 ASSERT(PREV.br_startoff <= new->br_startoff);
2039 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2040
2041 /*
2042 * Set flags determining what part of the previous oldext allocation
2043 * extent is being replaced by a newext allocation.
2044 */
2045 if (PREV.br_startoff == new->br_startoff)
2046 state |= BMAP_LEFT_FILLING;
2047 if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2048 state |= BMAP_RIGHT_FILLING;
2049
2050 /*
2051 * Check and set flags if this segment has a left neighbor.
2052 * Don't set contiguous if the combined extent would be too large.
2053 */
2054 if (xfs_iext_peek_prev_extent(ifp, icur, &LEFT)) {
2055 state |= BMAP_LEFT_VALID;
2056 if (isnullstartblock(LEFT.br_startblock))
2057 state |= BMAP_LEFT_DELAY;
2058 }
2059
2060 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2061 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2062 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2063 LEFT.br_state == new->br_state &&
2064 LEFT.br_blockcount + new->br_blockcount <= XFS_MAX_BMBT_EXTLEN)
2065 state |= BMAP_LEFT_CONTIG;
2066
2067 /*
2068 * Check and set flags if this segment has a right neighbor.
2069 * Don't set contiguous if the combined extent would be too large.
2070 * Also check for all-three-contiguous being too large.
2071 */
2072 if (xfs_iext_peek_next_extent(ifp, icur, &RIGHT)) {
2073 state |= BMAP_RIGHT_VALID;
2074 if (isnullstartblock(RIGHT.br_startblock))
2075 state |= BMAP_RIGHT_DELAY;
2076 }
2077
2078 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2079 new_endoff == RIGHT.br_startoff &&
2080 new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2081 new->br_state == RIGHT.br_state &&
2082 new->br_blockcount + RIGHT.br_blockcount <= XFS_MAX_BMBT_EXTLEN &&
2083 ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2084 BMAP_RIGHT_FILLING)) !=
2085 (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2086 BMAP_RIGHT_FILLING) ||
2087 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2088 <= XFS_MAX_BMBT_EXTLEN))
2089 state |= BMAP_RIGHT_CONTIG;
2090
2091 /*
2092 * Switch out based on the FILLING and CONTIG state bits.
2093 */
2094 switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2095 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2096 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2097 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2098 /*
2099 * Setting all of a previous oldext extent to newext.
2100 * The left and right neighbors are both contiguous with new.
2101 */
2102 LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount;
2103
2104 xfs_iext_remove(ip, icur, state);
2105 xfs_iext_remove(ip, icur, state);
2106 xfs_iext_prev(ifp, icur);
2107 xfs_iext_update_extent(ip, state, icur, &LEFT);
2108 ifp->if_nextents -= 2;
2109 if (cur == NULL)
2110 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2111 else {
2112 rval = XFS_ILOG_CORE;
2113 error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i);
2114 if (error)
2115 goto done;
2116 if (XFS_IS_CORRUPT(mp, i != 1)) {
2117 xfs_btree_mark_sick(cur);
2118 error = -EFSCORRUPTED;
2119 goto done;
2120 }
2121 if ((error = xfs_btree_delete(cur, &i)))
2122 goto done;
2123 if (XFS_IS_CORRUPT(mp, i != 1)) {
2124 xfs_btree_mark_sick(cur);
2125 error = -EFSCORRUPTED;
2126 goto done;
2127 }
2128 if ((error = xfs_btree_decrement(cur, 0, &i)))
2129 goto done;
2130 if (XFS_IS_CORRUPT(mp, i != 1)) {
2131 xfs_btree_mark_sick(cur);
2132 error = -EFSCORRUPTED;
2133 goto done;
2134 }
2135 if ((error = xfs_btree_delete(cur, &i)))
2136 goto done;
2137 if (XFS_IS_CORRUPT(mp, i != 1)) {
2138 xfs_btree_mark_sick(cur);
2139 error = -EFSCORRUPTED;
2140 goto done;
2141 }
2142 if ((error = xfs_btree_decrement(cur, 0, &i)))
2143 goto done;
2144 if (XFS_IS_CORRUPT(mp, i != 1)) {
2145 xfs_btree_mark_sick(cur);
2146 error = -EFSCORRUPTED;
2147 goto done;
2148 }
2149 error = xfs_bmbt_update(cur, &LEFT);
2150 if (error)
2151 goto done;
2152 }
2153 break;
2154
2155 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2156 /*
2157 * Setting all of a previous oldext extent to newext.
2158 * The left neighbor is contiguous, the right is not.
2159 */
2160 LEFT.br_blockcount += PREV.br_blockcount;
2161
2162 xfs_iext_remove(ip, icur, state);
2163 xfs_iext_prev(ifp, icur);
2164 xfs_iext_update_extent(ip, state, icur, &LEFT);
2165 ifp->if_nextents--;
2166 if (cur == NULL)
2167 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2168 else {
2169 rval = XFS_ILOG_CORE;
2170 error = xfs_bmbt_lookup_eq(cur, &PREV, &i);
2171 if (error)
2172 goto done;
2173 if (XFS_IS_CORRUPT(mp, i != 1)) {
2174 xfs_btree_mark_sick(cur);
2175 error = -EFSCORRUPTED;
2176 goto done;
2177 }
2178 if ((error = xfs_btree_delete(cur, &i)))
2179 goto done;
2180 if (XFS_IS_CORRUPT(mp, i != 1)) {
2181 xfs_btree_mark_sick(cur);
2182 error = -EFSCORRUPTED;
2183 goto done;
2184 }
2185 if ((error = xfs_btree_decrement(cur, 0, &i)))
2186 goto done;
2187 if (XFS_IS_CORRUPT(mp, i != 1)) {
2188 xfs_btree_mark_sick(cur);
2189 error = -EFSCORRUPTED;
2190 goto done;
2191 }
2192 error = xfs_bmbt_update(cur, &LEFT);
2193 if (error)
2194 goto done;
2195 }
2196 break;
2197
2198 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2199 /*
2200 * Setting all of a previous oldext extent to newext.
2201 * The right neighbor is contiguous, the left is not.
2202 */
2203 PREV.br_blockcount += RIGHT.br_blockcount;
2204 PREV.br_state = new->br_state;
2205
2206 xfs_iext_next(ifp, icur);
2207 xfs_iext_remove(ip, icur, state);
2208 xfs_iext_prev(ifp, icur);
2209 xfs_iext_update_extent(ip, state, icur, &PREV);
2210 ifp->if_nextents--;
2211
2212 if (cur == NULL)
2213 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2214 else {
2215 rval = XFS_ILOG_CORE;
2216 error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i);
2217 if (error)
2218 goto done;
2219 if (XFS_IS_CORRUPT(mp, i != 1)) {
2220 xfs_btree_mark_sick(cur);
2221 error = -EFSCORRUPTED;
2222 goto done;
2223 }
2224 if ((error = xfs_btree_delete(cur, &i)))
2225 goto done;
2226 if (XFS_IS_CORRUPT(mp, i != 1)) {
2227 xfs_btree_mark_sick(cur);
2228 error = -EFSCORRUPTED;
2229 goto done;
2230 }
2231 if ((error = xfs_btree_decrement(cur, 0, &i)))
2232 goto done;
2233 if (XFS_IS_CORRUPT(mp, i != 1)) {
2234 xfs_btree_mark_sick(cur);
2235 error = -EFSCORRUPTED;
2236 goto done;
2237 }
2238 error = xfs_bmbt_update(cur, &PREV);
2239 if (error)
2240 goto done;
2241 }
2242 break;
2243
2244 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2245 /*
2246 * Setting all of a previous oldext extent to newext.
2247 * Neither the left nor right neighbors are contiguous with
2248 * the new one.
2249 */
2250 PREV.br_state = new->br_state;
2251 xfs_iext_update_extent(ip, state, icur, &PREV);
2252
2253 if (cur == NULL)
2254 rval = XFS_ILOG_DEXT;
2255 else {
2256 rval = 0;
2257 error = xfs_bmbt_lookup_eq(cur, new, &i);
2258 if (error)
2259 goto done;
2260 if (XFS_IS_CORRUPT(mp, i != 1)) {
2261 xfs_btree_mark_sick(cur);
2262 error = -EFSCORRUPTED;
2263 goto done;
2264 }
2265 error = xfs_bmbt_update(cur, &PREV);
2266 if (error)
2267 goto done;
2268 }
2269 break;
2270
2271 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2272 /*
2273 * Setting the first part of a previous oldext extent to newext.
2274 * The left neighbor is contiguous.
2275 */
2276 LEFT.br_blockcount += new->br_blockcount;
2277
2278 old = PREV;
2279 PREV.br_startoff += new->br_blockcount;
2280 PREV.br_startblock += new->br_blockcount;
2281 PREV.br_blockcount -= new->br_blockcount;
2282
2283 xfs_iext_update_extent(ip, state, icur, &PREV);
2284 xfs_iext_prev(ifp, icur);
2285 xfs_iext_update_extent(ip, state, icur, &LEFT);
2286
2287 if (cur == NULL)
2288 rval = XFS_ILOG_DEXT;
2289 else {
2290 rval = 0;
2291 error = xfs_bmbt_lookup_eq(cur, &old, &i);
2292 if (error)
2293 goto done;
2294 if (XFS_IS_CORRUPT(mp, i != 1)) {
2295 xfs_btree_mark_sick(cur);
2296 error = -EFSCORRUPTED;
2297 goto done;
2298 }
2299 error = xfs_bmbt_update(cur, &PREV);
2300 if (error)
2301 goto done;
2302 error = xfs_btree_decrement(cur, 0, &i);
2303 if (error)
2304 goto done;
2305 error = xfs_bmbt_update(cur, &LEFT);
2306 if (error)
2307 goto done;
2308 }
2309 break;
2310
2311 case BMAP_LEFT_FILLING:
2312 /*
2313 * Setting the first part of a previous oldext extent to newext.
2314 * The left neighbor is not contiguous.
2315 */
2316 old = PREV;
2317 PREV.br_startoff += new->br_blockcount;
2318 PREV.br_startblock += new->br_blockcount;
2319 PREV.br_blockcount -= new->br_blockcount;
2320
2321 xfs_iext_update_extent(ip, state, icur, &PREV);
2322 xfs_iext_insert(ip, icur, new, state);
2323 ifp->if_nextents++;
2324
2325 if (cur == NULL)
2326 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2327 else {
2328 rval = XFS_ILOG_CORE;
2329 error = xfs_bmbt_lookup_eq(cur, &old, &i);
2330 if (error)
2331 goto done;
2332 if (XFS_IS_CORRUPT(mp, i != 1)) {
2333 xfs_btree_mark_sick(cur);
2334 error = -EFSCORRUPTED;
2335 goto done;
2336 }
2337 error = xfs_bmbt_update(cur, &PREV);
2338 if (error)
2339 goto done;
2340 cur->bc_rec.b = *new;
2341 if ((error = xfs_btree_insert(cur, &i)))
2342 goto done;
2343 if (XFS_IS_CORRUPT(mp, i != 1)) {
2344 xfs_btree_mark_sick(cur);
2345 error = -EFSCORRUPTED;
2346 goto done;
2347 }
2348 }
2349 break;
2350
2351 case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2352 /*
2353 * Setting the last part of a previous oldext extent to newext.
2354 * The right neighbor is contiguous with the new allocation.
2355 */
2356 old = PREV;
2357 PREV.br_blockcount -= new->br_blockcount;
2358
2359 RIGHT.br_startoff = new->br_startoff;
2360 RIGHT.br_startblock = new->br_startblock;
2361 RIGHT.br_blockcount += new->br_blockcount;
2362
2363 xfs_iext_update_extent(ip, state, icur, &PREV);
2364 xfs_iext_next(ifp, icur);
2365 xfs_iext_update_extent(ip, state, icur, &RIGHT);
2366
2367 if (cur == NULL)
2368 rval = XFS_ILOG_DEXT;
2369 else {
2370 rval = 0;
2371 error = xfs_bmbt_lookup_eq(cur, &old, &i);
2372 if (error)
2373 goto done;
2374 if (XFS_IS_CORRUPT(mp, i != 1)) {
2375 xfs_btree_mark_sick(cur);
2376 error = -EFSCORRUPTED;
2377 goto done;
2378 }
2379 error = xfs_bmbt_update(cur, &PREV);
2380 if (error)
2381 goto done;
2382 error = xfs_btree_increment(cur, 0, &i);
2383 if (error)
2384 goto done;
2385 error = xfs_bmbt_update(cur, &RIGHT);
2386 if (error)
2387 goto done;
2388 }
2389 break;
2390
2391 case BMAP_RIGHT_FILLING:
2392 /*
2393 * Setting the last part of a previous oldext extent to newext.
2394 * The right neighbor is not contiguous.
2395 */
2396 old = PREV;
2397 PREV.br_blockcount -= new->br_blockcount;
2398
2399 xfs_iext_update_extent(ip, state, icur, &PREV);
2400 xfs_iext_next(ifp, icur);
2401 xfs_iext_insert(ip, icur, new, state);
2402 ifp->if_nextents++;
2403
2404 if (cur == NULL)
2405 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2406 else {
2407 rval = XFS_ILOG_CORE;
2408 error = xfs_bmbt_lookup_eq(cur, &old, &i);
2409 if (error)
2410 goto done;
2411 if (XFS_IS_CORRUPT(mp, i != 1)) {
2412 xfs_btree_mark_sick(cur);
2413 error = -EFSCORRUPTED;
2414 goto done;
2415 }
2416 error = xfs_bmbt_update(cur, &PREV);
2417 if (error)
2418 goto done;
2419 error = xfs_bmbt_lookup_eq(cur, new, &i);
2420 if (error)
2421 goto done;
2422 if (XFS_IS_CORRUPT(mp, i != 0)) {
2423 xfs_btree_mark_sick(cur);
2424 error = -EFSCORRUPTED;
2425 goto done;
2426 }
2427 if ((error = xfs_btree_insert(cur, &i)))
2428 goto done;
2429 if (XFS_IS_CORRUPT(mp, i != 1)) {
2430 xfs_btree_mark_sick(cur);
2431 error = -EFSCORRUPTED;
2432 goto done;
2433 }
2434 }
2435 break;
2436
2437 case 0:
2438 /*
2439 * Setting the middle part of a previous oldext extent to
2440 * newext. Contiguity is impossible here.
2441 * One extent becomes three extents.
2442 */
2443 old = PREV;
2444 PREV.br_blockcount = new->br_startoff - PREV.br_startoff;
2445
2446 r[0] = *new;
2447 r[1].br_startoff = new_endoff;
2448 r[1].br_blockcount =
2449 old.br_startoff + old.br_blockcount - new_endoff;
2450 r[1].br_startblock = new->br_startblock + new->br_blockcount;
2451 r[1].br_state = PREV.br_state;
2452
2453 xfs_iext_update_extent(ip, state, icur, &PREV);
2454 xfs_iext_next(ifp, icur);
2455 xfs_iext_insert(ip, icur, &r[1], state);
2456 xfs_iext_insert(ip, icur, &r[0], state);
2457 ifp->if_nextents += 2;
2458
2459 if (cur == NULL)
2460 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2461 else {
2462 rval = XFS_ILOG_CORE;
2463 error = xfs_bmbt_lookup_eq(cur, &old, &i);
2464 if (error)
2465 goto done;
2466 if (XFS_IS_CORRUPT(mp, i != 1)) {
2467 xfs_btree_mark_sick(cur);
2468 error = -EFSCORRUPTED;
2469 goto done;
2470 }
2471 /* new right extent - oldext */
2472 error = xfs_bmbt_update(cur, &r[1]);
2473 if (error)
2474 goto done;
2475 /* new left extent - oldext */
2476 cur->bc_rec.b = PREV;
2477 if ((error = xfs_btree_insert(cur, &i)))
2478 goto done;
2479 if (XFS_IS_CORRUPT(mp, i != 1)) {
2480 xfs_btree_mark_sick(cur);
2481 error = -EFSCORRUPTED;
2482 goto done;
2483 }
2484 /*
2485 * Reset the cursor to the position of the new extent
2486 * we are about to insert as we can't trust it after
2487 * the previous insert.
2488 */
2489 error = xfs_bmbt_lookup_eq(cur, new, &i);
2490 if (error)
2491 goto done;
2492 if (XFS_IS_CORRUPT(mp, i != 0)) {
2493 xfs_btree_mark_sick(cur);
2494 error = -EFSCORRUPTED;
2495 goto done;
2496 }
2497 /* new middle extent - newext */
2498 if ((error = xfs_btree_insert(cur, &i)))
2499 goto done;
2500 if (XFS_IS_CORRUPT(mp, i != 1)) {
2501 xfs_btree_mark_sick(cur);
2502 error = -EFSCORRUPTED;
2503 goto done;
2504 }
2505 }
2506 break;
2507
2508 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2509 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2510 case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2511 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2512 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2513 case BMAP_LEFT_CONTIG:
2514 case BMAP_RIGHT_CONTIG:
2515 /*
2516 * These cases are all impossible.
2517 */
2518 ASSERT(0);
2519 }
2520
2521 /* update reverse mappings */
2522 xfs_rmap_convert_extent(mp, tp, ip, whichfork, new);
2523
2524 /* convert to a btree if necessary */
2525 if (xfs_bmap_needs_btree(ip, whichfork)) {
2526 int tmp_logflags; /* partial log flag return val */
2527
2528 ASSERT(cur == NULL);
2529 error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
2530 &tmp_logflags, whichfork);
2531 *logflagsp |= tmp_logflags;
2532 if (error)
2533 goto done;
2534 }
2535
2536 /* clear out the allocated field, done with it now in any case. */
2537 if (cur) {
2538 cur->bc_bmap.allocated = 0;
2539 *curp = cur;
2540 }
2541
2542 xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
2543 done:
2544 *logflagsp |= rval;
2545 return error;
2546 #undef LEFT
2547 #undef RIGHT
2548 #undef PREV
2549 }
2550
2551 /*
2552 * Convert a hole to a delayed allocation.
2553 */
2554 STATIC void
xfs_bmap_add_extent_hole_delay(xfs_inode_t * ip,int whichfork,struct xfs_iext_cursor * icur,xfs_bmbt_irec_t * new)2555 xfs_bmap_add_extent_hole_delay(
2556 xfs_inode_t *ip, /* incore inode pointer */
2557 int whichfork,
2558 struct xfs_iext_cursor *icur,
2559 xfs_bmbt_irec_t *new) /* new data to add to file extents */
2560 {
2561 struct xfs_ifork *ifp; /* inode fork pointer */
2562 xfs_bmbt_irec_t left; /* left neighbor extent entry */
2563 xfs_filblks_t newlen=0; /* new indirect size */
2564 xfs_filblks_t oldlen=0; /* old indirect size */
2565 xfs_bmbt_irec_t right; /* right neighbor extent entry */
2566 uint32_t state = xfs_bmap_fork_to_state(whichfork);
2567 xfs_filblks_t temp; /* temp for indirect calculations */
2568
2569 ifp = xfs_ifork_ptr(ip, whichfork);
2570 ASSERT(isnullstartblock(new->br_startblock));
2571
2572 /*
2573 * Check and set flags if this segment has a left neighbor
2574 */
2575 if (xfs_iext_peek_prev_extent(ifp, icur, &left)) {
2576 state |= BMAP_LEFT_VALID;
2577 if (isnullstartblock(left.br_startblock))
2578 state |= BMAP_LEFT_DELAY;
2579 }
2580
2581 /*
2582 * Check and set flags if the current (right) segment exists.
2583 * If it doesn't exist, we're converting the hole at end-of-file.
2584 */
2585 if (xfs_iext_get_extent(ifp, icur, &right)) {
2586 state |= BMAP_RIGHT_VALID;
2587 if (isnullstartblock(right.br_startblock))
2588 state |= BMAP_RIGHT_DELAY;
2589 }
2590
2591 /*
2592 * Set contiguity flags on the left and right neighbors.
2593 * Don't let extents get too large, even if the pieces are contiguous.
2594 */
2595 if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2596 left.br_startoff + left.br_blockcount == new->br_startoff &&
2597 left.br_blockcount + new->br_blockcount <= XFS_MAX_BMBT_EXTLEN)
2598 state |= BMAP_LEFT_CONTIG;
2599
2600 if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2601 new->br_startoff + new->br_blockcount == right.br_startoff &&
2602 new->br_blockcount + right.br_blockcount <= XFS_MAX_BMBT_EXTLEN &&
2603 (!(state & BMAP_LEFT_CONTIG) ||
2604 (left.br_blockcount + new->br_blockcount +
2605 right.br_blockcount <= XFS_MAX_BMBT_EXTLEN)))
2606 state |= BMAP_RIGHT_CONTIG;
2607
2608 /*
2609 * Switch out based on the contiguity flags.
2610 */
2611 switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2612 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2613 /*
2614 * New allocation is contiguous with delayed allocations
2615 * on the left and on the right.
2616 * Merge all three into a single extent record.
2617 */
2618 temp = left.br_blockcount + new->br_blockcount +
2619 right.br_blockcount;
2620
2621 oldlen = startblockval(left.br_startblock) +
2622 startblockval(new->br_startblock) +
2623 startblockval(right.br_startblock);
2624 newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2625 oldlen);
2626 left.br_startblock = nullstartblock(newlen);
2627 left.br_blockcount = temp;
2628
2629 xfs_iext_remove(ip, icur, state);
2630 xfs_iext_prev(ifp, icur);
2631 xfs_iext_update_extent(ip, state, icur, &left);
2632 break;
2633
2634 case BMAP_LEFT_CONTIG:
2635 /*
2636 * New allocation is contiguous with a delayed allocation
2637 * on the left.
2638 * Merge the new allocation with the left neighbor.
2639 */
2640 temp = left.br_blockcount + new->br_blockcount;
2641
2642 oldlen = startblockval(left.br_startblock) +
2643 startblockval(new->br_startblock);
2644 newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2645 oldlen);
2646 left.br_blockcount = temp;
2647 left.br_startblock = nullstartblock(newlen);
2648
2649 xfs_iext_prev(ifp, icur);
2650 xfs_iext_update_extent(ip, state, icur, &left);
2651 break;
2652
2653 case BMAP_RIGHT_CONTIG:
2654 /*
2655 * New allocation is contiguous with a delayed allocation
2656 * on the right.
2657 * Merge the new allocation with the right neighbor.
2658 */
2659 temp = new->br_blockcount + right.br_blockcount;
2660 oldlen = startblockval(new->br_startblock) +
2661 startblockval(right.br_startblock);
2662 newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2663 oldlen);
2664 right.br_startoff = new->br_startoff;
2665 right.br_startblock = nullstartblock(newlen);
2666 right.br_blockcount = temp;
2667 xfs_iext_update_extent(ip, state, icur, &right);
2668 break;
2669
2670 case 0:
2671 /*
2672 * New allocation is not contiguous with another
2673 * delayed allocation.
2674 * Insert a new entry.
2675 */
2676 oldlen = newlen = 0;
2677 xfs_iext_insert(ip, icur, new, state);
2678 break;
2679 }
2680 if (oldlen != newlen) {
2681 ASSERT(oldlen > newlen);
2682 xfs_add_fdblocks(ip->i_mount, oldlen - newlen);
2683
2684 /*
2685 * Nothing to do for disk quota accounting here.
2686 */
2687 xfs_mod_delalloc(ip, 0, (int64_t)newlen - oldlen);
2688 }
2689 }
2690
2691 /*
2692 * Convert a hole to a real allocation.
2693 */
2694 STATIC int /* error */
xfs_bmap_add_extent_hole_real(struct xfs_trans * tp,struct xfs_inode * ip,int whichfork,struct xfs_iext_cursor * icur,struct xfs_btree_cur ** curp,struct xfs_bmbt_irec * new,int * logflagsp,uint32_t flags)2695 xfs_bmap_add_extent_hole_real(
2696 struct xfs_trans *tp,
2697 struct xfs_inode *ip,
2698 int whichfork,
2699 struct xfs_iext_cursor *icur,
2700 struct xfs_btree_cur **curp,
2701 struct xfs_bmbt_irec *new,
2702 int *logflagsp,
2703 uint32_t flags)
2704 {
2705 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
2706 struct xfs_mount *mp = ip->i_mount;
2707 struct xfs_btree_cur *cur = *curp;
2708 int error; /* error return value */
2709 int i; /* temp state */
2710 xfs_bmbt_irec_t left; /* left neighbor extent entry */
2711 xfs_bmbt_irec_t right; /* right neighbor extent entry */
2712 int rval=0; /* return value (logging flags) */
2713 uint32_t state = xfs_bmap_fork_to_state(whichfork);
2714 struct xfs_bmbt_irec old;
2715
2716 ASSERT(!isnullstartblock(new->br_startblock));
2717 ASSERT(!cur || !(cur->bc_flags & XFS_BTREE_BMBT_WASDEL));
2718
2719 XFS_STATS_INC(mp, xs_add_exlist);
2720
2721 /*
2722 * Check and set flags if this segment has a left neighbor.
2723 */
2724 if (xfs_iext_peek_prev_extent(ifp, icur, &left)) {
2725 state |= BMAP_LEFT_VALID;
2726 if (isnullstartblock(left.br_startblock))
2727 state |= BMAP_LEFT_DELAY;
2728 }
2729
2730 /*
2731 * Check and set flags if this segment has a current value.
2732 * Not true if we're inserting into the "hole" at eof.
2733 */
2734 if (xfs_iext_get_extent(ifp, icur, &right)) {
2735 state |= BMAP_RIGHT_VALID;
2736 if (isnullstartblock(right.br_startblock))
2737 state |= BMAP_RIGHT_DELAY;
2738 }
2739
2740 /*
2741 * We're inserting a real allocation between "left" and "right".
2742 * Set the contiguity flags. Don't let extents get too large.
2743 */
2744 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2745 left.br_startoff + left.br_blockcount == new->br_startoff &&
2746 left.br_startblock + left.br_blockcount == new->br_startblock &&
2747 left.br_state == new->br_state &&
2748 left.br_blockcount + new->br_blockcount <= XFS_MAX_BMBT_EXTLEN)
2749 state |= BMAP_LEFT_CONTIG;
2750
2751 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2752 new->br_startoff + new->br_blockcount == right.br_startoff &&
2753 new->br_startblock + new->br_blockcount == right.br_startblock &&
2754 new->br_state == right.br_state &&
2755 new->br_blockcount + right.br_blockcount <= XFS_MAX_BMBT_EXTLEN &&
2756 (!(state & BMAP_LEFT_CONTIG) ||
2757 left.br_blockcount + new->br_blockcount +
2758 right.br_blockcount <= XFS_MAX_BMBT_EXTLEN))
2759 state |= BMAP_RIGHT_CONTIG;
2760
2761 error = 0;
2762 /*
2763 * Select which case we're in here, and implement it.
2764 */
2765 switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2766 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2767 /*
2768 * New allocation is contiguous with real allocations on the
2769 * left and on the right.
2770 * Merge all three into a single extent record.
2771 */
2772 left.br_blockcount += new->br_blockcount + right.br_blockcount;
2773
2774 xfs_iext_remove(ip, icur, state);
2775 xfs_iext_prev(ifp, icur);
2776 xfs_iext_update_extent(ip, state, icur, &left);
2777 ifp->if_nextents--;
2778
2779 if (cur == NULL) {
2780 rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2781 } else {
2782 rval = XFS_ILOG_CORE;
2783 error = xfs_bmbt_lookup_eq(cur, &right, &i);
2784 if (error)
2785 goto done;
2786 if (XFS_IS_CORRUPT(mp, i != 1)) {
2787 xfs_btree_mark_sick(cur);
2788 error = -EFSCORRUPTED;
2789 goto done;
2790 }
2791 error = xfs_btree_delete(cur, &i);
2792 if (error)
2793 goto done;
2794 if (XFS_IS_CORRUPT(mp, i != 1)) {
2795 xfs_btree_mark_sick(cur);
2796 error = -EFSCORRUPTED;
2797 goto done;
2798 }
2799 error = xfs_btree_decrement(cur, 0, &i);
2800 if (error)
2801 goto done;
2802 if (XFS_IS_CORRUPT(mp, i != 1)) {
2803 xfs_btree_mark_sick(cur);
2804 error = -EFSCORRUPTED;
2805 goto done;
2806 }
2807 error = xfs_bmbt_update(cur, &left);
2808 if (error)
2809 goto done;
2810 }
2811 break;
2812
2813 case BMAP_LEFT_CONTIG:
2814 /*
2815 * New allocation is contiguous with a real allocation
2816 * on the left.
2817 * Merge the new allocation with the left neighbor.
2818 */
2819 old = left;
2820 left.br_blockcount += new->br_blockcount;
2821
2822 xfs_iext_prev(ifp, icur);
2823 xfs_iext_update_extent(ip, state, icur, &left);
2824
2825 if (cur == NULL) {
2826 rval = xfs_ilog_fext(whichfork);
2827 } else {
2828 rval = 0;
2829 error = xfs_bmbt_lookup_eq(cur, &old, &i);
2830 if (error)
2831 goto done;
2832 if (XFS_IS_CORRUPT(mp, i != 1)) {
2833 xfs_btree_mark_sick(cur);
2834 error = -EFSCORRUPTED;
2835 goto done;
2836 }
2837 error = xfs_bmbt_update(cur, &left);
2838 if (error)
2839 goto done;
2840 }
2841 break;
2842
2843 case BMAP_RIGHT_CONTIG:
2844 /*
2845 * New allocation is contiguous with a real allocation
2846 * on the right.
2847 * Merge the new allocation with the right neighbor.
2848 */
2849 old = right;
2850
2851 right.br_startoff = new->br_startoff;
2852 right.br_startblock = new->br_startblock;
2853 right.br_blockcount += new->br_blockcount;
2854 xfs_iext_update_extent(ip, state, icur, &right);
2855
2856 if (cur == NULL) {
2857 rval = xfs_ilog_fext(whichfork);
2858 } else {
2859 rval = 0;
2860 error = xfs_bmbt_lookup_eq(cur, &old, &i);
2861 if (error)
2862 goto done;
2863 if (XFS_IS_CORRUPT(mp, i != 1)) {
2864 xfs_btree_mark_sick(cur);
2865 error = -EFSCORRUPTED;
2866 goto done;
2867 }
2868 error = xfs_bmbt_update(cur, &right);
2869 if (error)
2870 goto done;
2871 }
2872 break;
2873
2874 case 0:
2875 /*
2876 * New allocation is not contiguous with another
2877 * real allocation.
2878 * Insert a new entry.
2879 */
2880 xfs_iext_insert(ip, icur, new, state);
2881 ifp->if_nextents++;
2882
2883 if (cur == NULL) {
2884 rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2885 } else {
2886 rval = XFS_ILOG_CORE;
2887 error = xfs_bmbt_lookup_eq(cur, new, &i);
2888 if (error)
2889 goto done;
2890 if (XFS_IS_CORRUPT(mp, i != 0)) {
2891 xfs_btree_mark_sick(cur);
2892 error = -EFSCORRUPTED;
2893 goto done;
2894 }
2895 error = xfs_btree_insert(cur, &i);
2896 if (error)
2897 goto done;
2898 if (XFS_IS_CORRUPT(mp, i != 1)) {
2899 xfs_btree_mark_sick(cur);
2900 error = -EFSCORRUPTED;
2901 goto done;
2902 }
2903 }
2904 break;
2905 }
2906
2907 /* add reverse mapping unless caller opted out */
2908 if (!(flags & XFS_BMAPI_NORMAP))
2909 xfs_rmap_map_extent(tp, ip, whichfork, new);
2910
2911 /* convert to a btree if necessary */
2912 if (xfs_bmap_needs_btree(ip, whichfork)) {
2913 int tmp_logflags; /* partial log flag return val */
2914
2915 ASSERT(cur == NULL);
2916 error = xfs_bmap_extents_to_btree(tp, ip, curp, 0,
2917 &tmp_logflags, whichfork);
2918 *logflagsp |= tmp_logflags;
2919 cur = *curp;
2920 if (error)
2921 goto done;
2922 }
2923
2924 /* clear out the allocated field, done with it now in any case. */
2925 if (cur)
2926 cur->bc_bmap.allocated = 0;
2927
2928 xfs_bmap_check_leaf_extents(cur, ip, whichfork);
2929 done:
2930 *logflagsp |= rval;
2931 return error;
2932 }
2933
2934 /*
2935 * Functions used in the extent read, allocate and remove paths
2936 */
2937
2938 /*
2939 * Adjust the size of the new extent based on i_extsize and rt extsize.
2940 */
2941 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)2942 xfs_bmap_extsize_align(
2943 xfs_mount_t *mp,
2944 xfs_bmbt_irec_t *gotp, /* next extent pointer */
2945 xfs_bmbt_irec_t *prevp, /* previous extent pointer */
2946 xfs_extlen_t extsz, /* align to this extent size */
2947 int rt, /* is this a realtime inode? */
2948 int eof, /* is extent at end-of-file? */
2949 int delay, /* creating delalloc extent? */
2950 int convert, /* overwriting unwritten extent? */
2951 xfs_fileoff_t *offp, /* in/out: aligned offset */
2952 xfs_extlen_t *lenp) /* in/out: aligned length */
2953 {
2954 xfs_fileoff_t orig_off; /* original offset */
2955 xfs_extlen_t orig_alen; /* original length */
2956 xfs_fileoff_t orig_end; /* original off+len */
2957 xfs_fileoff_t nexto; /* next file offset */
2958 xfs_fileoff_t prevo; /* previous file offset */
2959 xfs_fileoff_t align_off; /* temp for offset */
2960 xfs_extlen_t align_alen; /* temp for length */
2961 xfs_extlen_t temp; /* temp for calculations */
2962
2963 if (convert)
2964 return 0;
2965
2966 orig_off = align_off = *offp;
2967 orig_alen = align_alen = *lenp;
2968 orig_end = orig_off + orig_alen;
2969
2970 /*
2971 * If this request overlaps an existing extent, then don't
2972 * attempt to perform any additional alignment.
2973 */
2974 if (!delay && !eof &&
2975 (orig_off >= gotp->br_startoff) &&
2976 (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2977 return 0;
2978 }
2979
2980 /*
2981 * If the file offset is unaligned vs. the extent size
2982 * we need to align it. This will be possible unless
2983 * the file was previously written with a kernel that didn't
2984 * perform this alignment, or if a truncate shot us in the
2985 * foot.
2986 */
2987 div_u64_rem(orig_off, extsz, &temp);
2988 if (temp) {
2989 align_alen += temp;
2990 align_off -= temp;
2991 }
2992
2993 /* Same adjustment for the end of the requested area. */
2994 temp = (align_alen % extsz);
2995 if (temp)
2996 align_alen += extsz - temp;
2997
2998 /*
2999 * For large extent hint sizes, the aligned extent might be larger than
3000 * XFS_BMBT_MAX_EXTLEN. In that case, reduce the size by an extsz so
3001 * that it pulls the length back under XFS_BMBT_MAX_EXTLEN. The outer
3002 * allocation loops handle short allocation just fine, so it is safe to
3003 * do this. We only want to do it when we are forced to, though, because
3004 * it means more allocation operations are required.
3005 */
3006 while (align_alen > XFS_MAX_BMBT_EXTLEN)
3007 align_alen -= extsz;
3008 ASSERT(align_alen <= XFS_MAX_BMBT_EXTLEN);
3009
3010 /*
3011 * If the previous block overlaps with this proposed allocation
3012 * then move the start forward without adjusting the length.
3013 */
3014 if (prevp->br_startoff != NULLFILEOFF) {
3015 if (prevp->br_startblock == HOLESTARTBLOCK)
3016 prevo = prevp->br_startoff;
3017 else
3018 prevo = prevp->br_startoff + prevp->br_blockcount;
3019 } else
3020 prevo = 0;
3021 if (align_off != orig_off && align_off < prevo)
3022 align_off = prevo;
3023 /*
3024 * If the next block overlaps with this proposed allocation
3025 * then move the start back without adjusting the length,
3026 * but not before offset 0.
3027 * This may of course make the start overlap previous block,
3028 * and if we hit the offset 0 limit then the next block
3029 * can still overlap too.
3030 */
3031 if (!eof && gotp->br_startoff != NULLFILEOFF) {
3032 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
3033 (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
3034 nexto = gotp->br_startoff + gotp->br_blockcount;
3035 else
3036 nexto = gotp->br_startoff;
3037 } else
3038 nexto = NULLFILEOFF;
3039 if (!eof &&
3040 align_off + align_alen != orig_end &&
3041 align_off + align_alen > nexto)
3042 align_off = nexto > align_alen ? nexto - align_alen : 0;
3043 /*
3044 * If we're now overlapping the next or previous extent that
3045 * means we can't fit an extsz piece in this hole. Just move
3046 * the start forward to the first valid spot and set
3047 * the length so we hit the end.
3048 */
3049 if (align_off != orig_off && align_off < prevo)
3050 align_off = prevo;
3051 if (align_off + align_alen != orig_end &&
3052 align_off + align_alen > nexto &&
3053 nexto != NULLFILEOFF) {
3054 ASSERT(nexto > prevo);
3055 align_alen = nexto - align_off;
3056 }
3057
3058 /*
3059 * If realtime, and the result isn't a multiple of the realtime
3060 * extent size we need to remove blocks until it is.
3061 */
3062 if (rt && (temp = xfs_extlen_to_rtxmod(mp, align_alen))) {
3063 /*
3064 * We're not covering the original request, or
3065 * we won't be able to once we fix the length.
3066 */
3067 if (orig_off < align_off ||
3068 orig_end > align_off + align_alen ||
3069 align_alen - temp < orig_alen)
3070 return -EINVAL;
3071 /*
3072 * Try to fix it by moving the start up.
3073 */
3074 if (align_off + temp <= orig_off) {
3075 align_alen -= temp;
3076 align_off += temp;
3077 }
3078 /*
3079 * Try to fix it by moving the end in.
3080 */
3081 else if (align_off + align_alen - temp >= orig_end)
3082 align_alen -= temp;
3083 /*
3084 * Set the start to the minimum then trim the length.
3085 */
3086 else {
3087 align_alen -= orig_off - align_off;
3088 align_off = orig_off;
3089 align_alen -= xfs_extlen_to_rtxmod(mp, align_alen);
3090 }
3091 /*
3092 * Result doesn't cover the request, fail it.
3093 */
3094 if (orig_off < align_off || orig_end > align_off + align_alen)
3095 return -EINVAL;
3096 } else {
3097 ASSERT(orig_off >= align_off);
3098 /* see XFS_BMBT_MAX_EXTLEN handling above */
3099 ASSERT(orig_end <= align_off + align_alen ||
3100 align_alen + extsz > XFS_MAX_BMBT_EXTLEN);
3101 }
3102
3103 #ifdef DEBUG
3104 if (!eof && gotp->br_startoff != NULLFILEOFF)
3105 ASSERT(align_off + align_alen <= gotp->br_startoff);
3106 if (prevp->br_startoff != NULLFILEOFF)
3107 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3108 #endif
3109
3110 *lenp = align_alen;
3111 *offp = align_off;
3112 return 0;
3113 }
3114
3115 static inline bool
xfs_bmap_adjacent_valid(struct xfs_bmalloca * ap,xfs_fsblock_t x,xfs_fsblock_t y)3116 xfs_bmap_adjacent_valid(
3117 struct xfs_bmalloca *ap,
3118 xfs_fsblock_t x,
3119 xfs_fsblock_t y)
3120 {
3121 struct xfs_mount *mp = ap->ip->i_mount;
3122
3123 if (XFS_IS_REALTIME_INODE(ap->ip) &&
3124 (ap->datatype & XFS_ALLOC_USERDATA))
3125 return x < mp->m_sb.sb_rblocks;
3126
3127 return XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) &&
3128 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount &&
3129 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks;
3130 }
3131
3132 #define XFS_ALLOC_GAP_UNITS 4
3133
3134 /* returns true if ap->blkno was modified */
3135 bool
xfs_bmap_adjacent(struct xfs_bmalloca * ap)3136 xfs_bmap_adjacent(
3137 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
3138 {
3139 xfs_fsblock_t adjust; /* adjustment to block numbers */
3140
3141 /*
3142 * If allocating at eof, and there's a previous real block,
3143 * try to use its last block as our starting point.
3144 */
3145 if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3146 !isnullstartblock(ap->prev.br_startblock) &&
3147 xfs_bmap_adjacent_valid(ap,
3148 ap->prev.br_startblock + ap->prev.br_blockcount,
3149 ap->prev.br_startblock)) {
3150 ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3151 /*
3152 * Adjust for the gap between prevp and us.
3153 */
3154 adjust = ap->offset -
3155 (ap->prev.br_startoff + ap->prev.br_blockcount);
3156 if (adjust && xfs_bmap_adjacent_valid(ap, ap->blkno + adjust,
3157 ap->prev.br_startblock))
3158 ap->blkno += adjust;
3159 return true;
3160 }
3161 /*
3162 * If not at eof, then compare the two neighbor blocks.
3163 * Figure out whether either one gives us a good starting point,
3164 * and pick the better one.
3165 */
3166 if (!ap->eof) {
3167 xfs_fsblock_t gotbno; /* right side block number */
3168 xfs_fsblock_t gotdiff=0; /* right side difference */
3169 xfs_fsblock_t prevbno; /* left side block number */
3170 xfs_fsblock_t prevdiff=0; /* left side difference */
3171
3172 /*
3173 * If there's a previous (left) block, select a requested
3174 * start block based on it.
3175 */
3176 if (ap->prev.br_startoff != NULLFILEOFF &&
3177 !isnullstartblock(ap->prev.br_startblock) &&
3178 (prevbno = ap->prev.br_startblock +
3179 ap->prev.br_blockcount) &&
3180 xfs_bmap_adjacent_valid(ap, prevbno,
3181 ap->prev.br_startblock)) {
3182 /*
3183 * Calculate gap to end of previous block.
3184 */
3185 adjust = prevdiff = ap->offset -
3186 (ap->prev.br_startoff +
3187 ap->prev.br_blockcount);
3188 /*
3189 * Figure the startblock based on the previous block's
3190 * end and the gap size.
3191 * Heuristic!
3192 * If the gap is large relative to the piece we're
3193 * allocating, or using it gives us an invalid block
3194 * number, then just use the end of the previous block.
3195 */
3196 if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3197 xfs_bmap_adjacent_valid(ap, prevbno + prevdiff,
3198 ap->prev.br_startblock))
3199 prevbno += adjust;
3200 else
3201 prevdiff += adjust;
3202 }
3203 /*
3204 * No previous block or can't follow it, just default.
3205 */
3206 else
3207 prevbno = NULLFSBLOCK;
3208 /*
3209 * If there's a following (right) block, select a requested
3210 * start block based on it.
3211 */
3212 if (!isnullstartblock(ap->got.br_startblock)) {
3213 /*
3214 * Calculate gap to start of next block.
3215 */
3216 adjust = gotdiff = ap->got.br_startoff - ap->offset;
3217 /*
3218 * Figure the startblock based on the next block's
3219 * start and the gap size.
3220 */
3221 gotbno = ap->got.br_startblock;
3222 /*
3223 * Heuristic!
3224 * If the gap is large relative to the piece we're
3225 * allocating, or using it gives us an invalid block
3226 * number, then just use the start of the next block
3227 * offset by our length.
3228 */
3229 if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3230 xfs_bmap_adjacent_valid(ap, gotbno - gotdiff,
3231 gotbno))
3232 gotbno -= adjust;
3233 else if (xfs_bmap_adjacent_valid(ap, gotbno - ap->length,
3234 gotbno)) {
3235 gotbno -= ap->length;
3236 gotdiff += adjust - ap->length;
3237 } else
3238 gotdiff += adjust;
3239 }
3240 /*
3241 * No next block, just default.
3242 */
3243 else
3244 gotbno = NULLFSBLOCK;
3245 /*
3246 * If both valid, pick the better one, else the only good
3247 * one, else ap->blkno is already set (to 0 or the inode block).
3248 */
3249 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK) {
3250 ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3251 return true;
3252 }
3253 if (prevbno != NULLFSBLOCK) {
3254 ap->blkno = prevbno;
3255 return true;
3256 }
3257 if (gotbno != NULLFSBLOCK) {
3258 ap->blkno = gotbno;
3259 return true;
3260 }
3261 }
3262
3263 return false;
3264 }
3265
3266 int
xfs_bmap_longest_free_extent(struct xfs_perag * pag,struct xfs_trans * tp,xfs_extlen_t * blen)3267 xfs_bmap_longest_free_extent(
3268 struct xfs_perag *pag,
3269 struct xfs_trans *tp,
3270 xfs_extlen_t *blen)
3271 {
3272 xfs_extlen_t longest;
3273 int error = 0;
3274
3275 if (!xfs_perag_initialised_agf(pag)) {
3276 error = xfs_alloc_read_agf(pag, tp, XFS_ALLOC_FLAG_TRYLOCK,
3277 NULL);
3278 if (error)
3279 return error;
3280 }
3281
3282 longest = xfs_alloc_longest_free_extent(pag,
3283 xfs_alloc_min_freelist(pag->pag_mount, pag),
3284 xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE));
3285 if (*blen < longest)
3286 *blen = longest;
3287
3288 return 0;
3289 }
3290
3291 static xfs_extlen_t
xfs_bmap_select_minlen(struct xfs_bmalloca * ap,struct xfs_alloc_arg * args,xfs_extlen_t blen)3292 xfs_bmap_select_minlen(
3293 struct xfs_bmalloca *ap,
3294 struct xfs_alloc_arg *args,
3295 xfs_extlen_t blen)
3296 {
3297
3298 /*
3299 * Since we used XFS_ALLOC_FLAG_TRYLOCK in _longest_free_extent(), it is
3300 * possible that there is enough contiguous free space for this request.
3301 */
3302 if (blen < ap->minlen)
3303 return ap->minlen;
3304
3305 /*
3306 * If the best seen length is less than the request length,
3307 * use the best as the minimum, otherwise we've got the maxlen we
3308 * were asked for.
3309 */
3310 if (blen < args->maxlen)
3311 return blen;
3312 return args->maxlen;
3313 }
3314
3315 static int
xfs_bmap_btalloc_select_lengths(struct xfs_bmalloca * ap,struct xfs_alloc_arg * args,xfs_extlen_t * blen)3316 xfs_bmap_btalloc_select_lengths(
3317 struct xfs_bmalloca *ap,
3318 struct xfs_alloc_arg *args,
3319 xfs_extlen_t *blen)
3320 {
3321 struct xfs_mount *mp = args->mp;
3322 struct xfs_perag *pag;
3323 xfs_agnumber_t agno, startag;
3324 int error = 0;
3325
3326 if (ap->tp->t_flags & XFS_TRANS_LOWMODE) {
3327 args->total = ap->minlen;
3328 args->minlen = ap->minlen;
3329 return 0;
3330 }
3331
3332 args->total = ap->total;
3333 startag = XFS_FSB_TO_AGNO(mp, ap->blkno);
3334 if (startag == NULLAGNUMBER)
3335 startag = 0;
3336
3337 *blen = 0;
3338 for_each_perag_wrap(mp, startag, agno, pag) {
3339 error = xfs_bmap_longest_free_extent(pag, args->tp, blen);
3340 if (error && error != -EAGAIN)
3341 break;
3342 error = 0;
3343 if (*blen >= args->maxlen)
3344 break;
3345 }
3346 if (pag)
3347 xfs_perag_rele(pag);
3348
3349 args->minlen = xfs_bmap_select_minlen(ap, args, *blen);
3350 return error;
3351 }
3352
3353 /* Update all inode and quota accounting for the allocation we just did. */
3354 void
xfs_bmap_alloc_account(struct xfs_bmalloca * ap)3355 xfs_bmap_alloc_account(
3356 struct xfs_bmalloca *ap)
3357 {
3358 bool isrt = XFS_IS_REALTIME_INODE(ap->ip) &&
3359 !(ap->flags & XFS_BMAPI_ATTRFORK);
3360 uint fld;
3361
3362 if (ap->flags & XFS_BMAPI_COWFORK) {
3363 /*
3364 * COW fork blocks are in-core only and thus are treated as
3365 * in-core quota reservation (like delalloc blocks) even when
3366 * converted to real blocks. The quota reservation is not
3367 * accounted to disk until blocks are remapped to the data
3368 * fork. So if these blocks were previously delalloc, we
3369 * already have quota reservation and there's nothing to do
3370 * yet.
3371 */
3372 if (ap->wasdel) {
3373 xfs_mod_delalloc(ap->ip, -(int64_t)ap->length, 0);
3374 return;
3375 }
3376
3377 /*
3378 * Otherwise, we've allocated blocks in a hole. The transaction
3379 * has acquired in-core quota reservation for this extent.
3380 * Rather than account these as real blocks, however, we reduce
3381 * the transaction quota reservation based on the allocation.
3382 * This essentially transfers the transaction quota reservation
3383 * to that of a delalloc extent.
3384 */
3385 ap->ip->i_delayed_blks += ap->length;
3386 xfs_trans_mod_dquot_byino(ap->tp, ap->ip, isrt ?
3387 XFS_TRANS_DQ_RES_RTBLKS : XFS_TRANS_DQ_RES_BLKS,
3388 -(long)ap->length);
3389 return;
3390 }
3391
3392 /* data/attr fork only */
3393 ap->ip->i_nblocks += ap->length;
3394 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3395 if (ap->wasdel) {
3396 ap->ip->i_delayed_blks -= ap->length;
3397 xfs_mod_delalloc(ap->ip, -(int64_t)ap->length, 0);
3398 fld = isrt ? XFS_TRANS_DQ_DELRTBCOUNT : XFS_TRANS_DQ_DELBCOUNT;
3399 } else {
3400 fld = isrt ? XFS_TRANS_DQ_RTBCOUNT : XFS_TRANS_DQ_BCOUNT;
3401 }
3402
3403 xfs_trans_mod_dquot_byino(ap->tp, ap->ip, fld, ap->length);
3404 }
3405
3406 static int
xfs_bmap_compute_alignments(struct xfs_bmalloca * ap,struct xfs_alloc_arg * args)3407 xfs_bmap_compute_alignments(
3408 struct xfs_bmalloca *ap,
3409 struct xfs_alloc_arg *args)
3410 {
3411 struct xfs_mount *mp = args->mp;
3412 xfs_extlen_t align = 0; /* minimum allocation alignment */
3413 int stripe_align = 0;
3414
3415 /* stripe alignment for allocation is determined by mount parameters */
3416 if (mp->m_swidth && xfs_has_swalloc(mp))
3417 stripe_align = mp->m_swidth;
3418 else if (mp->m_dalign)
3419 stripe_align = mp->m_dalign;
3420
3421 if (ap->flags & XFS_BMAPI_COWFORK)
3422 align = xfs_get_cowextsz_hint(ap->ip);
3423 else if (ap->datatype & XFS_ALLOC_USERDATA)
3424 align = xfs_get_extsz_hint(ap->ip);
3425 if (align) {
3426 if (xfs_bmap_extsize_align(mp, &ap->got, &ap->prev, align, 0,
3427 ap->eof, 0, ap->conv, &ap->offset,
3428 &ap->length))
3429 ASSERT(0);
3430 ASSERT(ap->length);
3431 }
3432
3433 /* apply extent size hints if obtained earlier */
3434 if (align) {
3435 args->prod = align;
3436 div_u64_rem(ap->offset, args->prod, &args->mod);
3437 if (args->mod)
3438 args->mod = args->prod - args->mod;
3439 } else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) {
3440 args->prod = 1;
3441 args->mod = 0;
3442 } else {
3443 args->prod = PAGE_SIZE >> mp->m_sb.sb_blocklog;
3444 div_u64_rem(ap->offset, args->prod, &args->mod);
3445 if (args->mod)
3446 args->mod = args->prod - args->mod;
3447 }
3448
3449 return stripe_align;
3450 }
3451
3452 static void
xfs_bmap_process_allocated_extent(struct xfs_bmalloca * ap,struct xfs_alloc_arg * args,xfs_fileoff_t orig_offset,xfs_extlen_t orig_length)3453 xfs_bmap_process_allocated_extent(
3454 struct xfs_bmalloca *ap,
3455 struct xfs_alloc_arg *args,
3456 xfs_fileoff_t orig_offset,
3457 xfs_extlen_t orig_length)
3458 {
3459 ap->blkno = args->fsbno;
3460 ap->length = args->len;
3461 /*
3462 * If the extent size hint is active, we tried to round the
3463 * caller's allocation request offset down to extsz and the
3464 * length up to another extsz boundary. If we found a free
3465 * extent we mapped it in starting at this new offset. If the
3466 * newly mapped space isn't long enough to cover any of the
3467 * range of offsets that was originally requested, move the
3468 * mapping up so that we can fill as much of the caller's
3469 * original request as possible. Free space is apparently
3470 * very fragmented so we're unlikely to be able to satisfy the
3471 * hints anyway.
3472 */
3473 if (ap->length <= orig_length)
3474 ap->offset = orig_offset;
3475 else if (ap->offset + ap->length < orig_offset + orig_length)
3476 ap->offset = orig_offset + orig_length - ap->length;
3477 xfs_bmap_alloc_account(ap);
3478 }
3479
3480 static int
xfs_bmap_exact_minlen_extent_alloc(struct xfs_bmalloca * ap,struct xfs_alloc_arg * args)3481 xfs_bmap_exact_minlen_extent_alloc(
3482 struct xfs_bmalloca *ap,
3483 struct xfs_alloc_arg *args)
3484 {
3485 if (ap->minlen != 1) {
3486 args->fsbno = NULLFSBLOCK;
3487 return 0;
3488 }
3489
3490 args->alloc_minlen_only = 1;
3491 args->minlen = args->maxlen = ap->minlen;
3492 args->total = ap->total;
3493
3494 /*
3495 * Unlike the longest extent available in an AG, we don't track
3496 * the length of an AG's shortest extent.
3497 * XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT is a debug only knob and
3498 * hence we can afford to start traversing from the 0th AG since
3499 * we need not be concerned about a drop in performance in
3500 * "debug only" code paths.
3501 */
3502 ap->blkno = XFS_AGB_TO_FSB(ap->ip->i_mount, 0, 0);
3503
3504 /*
3505 * Call xfs_bmap_btalloc_low_space here as it first does a "normal" AG
3506 * iteration and then drops args->total to args->minlen, which might be
3507 * required to find an allocation for the transaction reservation when
3508 * the file system is very full.
3509 */
3510 return xfs_bmap_btalloc_low_space(ap, args);
3511 }
3512
3513 /*
3514 * If we are not low on available data blocks and we are allocating at
3515 * EOF, optimise allocation for contiguous file extension and/or stripe
3516 * alignment of the new extent.
3517 *
3518 * NOTE: ap->aeof is only set if the allocation length is >= the
3519 * stripe unit and the allocation offset is at the end of file.
3520 */
3521 static int
xfs_bmap_btalloc_at_eof(struct xfs_bmalloca * ap,struct xfs_alloc_arg * args,xfs_extlen_t blen,int stripe_align,bool ag_only)3522 xfs_bmap_btalloc_at_eof(
3523 struct xfs_bmalloca *ap,
3524 struct xfs_alloc_arg *args,
3525 xfs_extlen_t blen,
3526 int stripe_align,
3527 bool ag_only)
3528 {
3529 struct xfs_mount *mp = args->mp;
3530 struct xfs_perag *caller_pag = args->pag;
3531 int error;
3532
3533 /*
3534 * If there are already extents in the file, try an exact EOF block
3535 * allocation to extend the file as a contiguous extent. If that fails,
3536 * or it's the first allocation in a file, just try for a stripe aligned
3537 * allocation.
3538 */
3539 if (ap->offset) {
3540 xfs_extlen_t nextminlen = 0;
3541
3542 /*
3543 * Compute the minlen+alignment for the next case. Set slop so
3544 * that the value of minlen+alignment+slop doesn't go up between
3545 * the calls.
3546 */
3547 args->alignment = 1;
3548 if (blen > stripe_align && blen <= args->maxlen)
3549 nextminlen = blen - stripe_align;
3550 else
3551 nextminlen = args->minlen;
3552 if (nextminlen + stripe_align > args->minlen + 1)
3553 args->minalignslop = nextminlen + stripe_align -
3554 args->minlen - 1;
3555 else
3556 args->minalignslop = 0;
3557
3558 if (!caller_pag)
3559 args->pag = xfs_perag_get(mp, XFS_FSB_TO_AGNO(mp, ap->blkno));
3560 error = xfs_alloc_vextent_exact_bno(args, ap->blkno);
3561 if (!caller_pag) {
3562 xfs_perag_put(args->pag);
3563 args->pag = NULL;
3564 }
3565 if (error)
3566 return error;
3567
3568 if (args->fsbno != NULLFSBLOCK)
3569 return 0;
3570 /*
3571 * Exact allocation failed. Reset to try an aligned allocation
3572 * according to the original allocation specification.
3573 */
3574 args->alignment = stripe_align;
3575 args->minlen = nextminlen;
3576 args->minalignslop = 0;
3577 } else {
3578 /*
3579 * Adjust minlen to try and preserve alignment if we
3580 * can't guarantee an aligned maxlen extent.
3581 */
3582 args->alignment = stripe_align;
3583 if (blen > args->alignment &&
3584 blen <= args->maxlen + args->alignment)
3585 args->minlen = blen - args->alignment;
3586 args->minalignslop = 0;
3587 }
3588
3589 if (ag_only) {
3590 error = xfs_alloc_vextent_near_bno(args, ap->blkno);
3591 } else {
3592 args->pag = NULL;
3593 error = xfs_alloc_vextent_start_ag(args, ap->blkno);
3594 ASSERT(args->pag == NULL);
3595 args->pag = caller_pag;
3596 }
3597 if (error)
3598 return error;
3599
3600 if (args->fsbno != NULLFSBLOCK)
3601 return 0;
3602
3603 /*
3604 * Allocation failed, so turn return the allocation args to their
3605 * original non-aligned state so the caller can proceed on allocation
3606 * failure as if this function was never called.
3607 */
3608 args->alignment = 1;
3609 return 0;
3610 }
3611
3612 /*
3613 * We have failed multiple allocation attempts so now are in a low space
3614 * allocation situation. Try a locality first full filesystem minimum length
3615 * allocation whilst still maintaining necessary total block reservation
3616 * requirements.
3617 *
3618 * If that fails, we are now critically low on space, so perform a last resort
3619 * allocation attempt: no reserve, no locality, blocking, minimum length, full
3620 * filesystem free space scan. We also indicate to future allocations in this
3621 * transaction that we are critically low on space so they don't waste time on
3622 * allocation modes that are unlikely to succeed.
3623 */
3624 int
xfs_bmap_btalloc_low_space(struct xfs_bmalloca * ap,struct xfs_alloc_arg * args)3625 xfs_bmap_btalloc_low_space(
3626 struct xfs_bmalloca *ap,
3627 struct xfs_alloc_arg *args)
3628 {
3629 int error;
3630
3631 if (args->minlen > ap->minlen) {
3632 args->minlen = ap->minlen;
3633 error = xfs_alloc_vextent_start_ag(args, ap->blkno);
3634 if (error || args->fsbno != NULLFSBLOCK)
3635 return error;
3636 }
3637
3638 /* Last ditch attempt before failure is declared. */
3639 args->total = ap->minlen;
3640 error = xfs_alloc_vextent_first_ag(args, 0);
3641 if (error)
3642 return error;
3643 ap->tp->t_flags |= XFS_TRANS_LOWMODE;
3644 return 0;
3645 }
3646
3647 static int
xfs_bmap_btalloc_filestreams(struct xfs_bmalloca * ap,struct xfs_alloc_arg * args,int stripe_align)3648 xfs_bmap_btalloc_filestreams(
3649 struct xfs_bmalloca *ap,
3650 struct xfs_alloc_arg *args,
3651 int stripe_align)
3652 {
3653 xfs_extlen_t blen = 0;
3654 int error = 0;
3655
3656
3657 error = xfs_filestream_select_ag(ap, args, &blen);
3658 if (error)
3659 return error;
3660 ASSERT(args->pag);
3661
3662 /*
3663 * If we are in low space mode, then optimal allocation will fail so
3664 * prepare for minimal allocation and jump to the low space algorithm
3665 * immediately.
3666 */
3667 if (ap->tp->t_flags & XFS_TRANS_LOWMODE) {
3668 args->minlen = ap->minlen;
3669 ASSERT(args->fsbno == NULLFSBLOCK);
3670 goto out_low_space;
3671 }
3672
3673 args->minlen = xfs_bmap_select_minlen(ap, args, blen);
3674 if (ap->aeof)
3675 error = xfs_bmap_btalloc_at_eof(ap, args, blen, stripe_align,
3676 true);
3677
3678 if (!error && args->fsbno == NULLFSBLOCK)
3679 error = xfs_alloc_vextent_near_bno(args, ap->blkno);
3680
3681 out_low_space:
3682 /*
3683 * We are now done with the perag reference for the filestreams
3684 * association provided by xfs_filestream_select_ag(). Release it now as
3685 * we've either succeeded, had a fatal error or we are out of space and
3686 * need to do a full filesystem scan for free space which will take it's
3687 * own references.
3688 */
3689 xfs_perag_rele(args->pag);
3690 args->pag = NULL;
3691 if (error || args->fsbno != NULLFSBLOCK)
3692 return error;
3693
3694 return xfs_bmap_btalloc_low_space(ap, args);
3695 }
3696
3697 static int
xfs_bmap_btalloc_best_length(struct xfs_bmalloca * ap,struct xfs_alloc_arg * args,int stripe_align)3698 xfs_bmap_btalloc_best_length(
3699 struct xfs_bmalloca *ap,
3700 struct xfs_alloc_arg *args,
3701 int stripe_align)
3702 {
3703 xfs_extlen_t blen = 0;
3704 int error;
3705
3706 ap->blkno = XFS_INO_TO_FSB(args->mp, ap->ip->i_ino);
3707 xfs_bmap_adjacent(ap);
3708
3709 /*
3710 * Search for an allocation group with a single extent large enough for
3711 * the request. If one isn't found, then adjust the minimum allocation
3712 * size to the largest space found.
3713 */
3714 error = xfs_bmap_btalloc_select_lengths(ap, args, &blen);
3715 if (error)
3716 return error;
3717
3718 /*
3719 * Don't attempt optimal EOF allocation if previous allocations barely
3720 * succeeded due to being near ENOSPC. It is highly unlikely we'll get
3721 * optimal or even aligned allocations in this case, so don't waste time
3722 * trying.
3723 */
3724 if (ap->aeof && !(ap->tp->t_flags & XFS_TRANS_LOWMODE)) {
3725 error = xfs_bmap_btalloc_at_eof(ap, args, blen, stripe_align,
3726 false);
3727 if (error || args->fsbno != NULLFSBLOCK)
3728 return error;
3729 }
3730
3731 error = xfs_alloc_vextent_start_ag(args, ap->blkno);
3732 if (error || args->fsbno != NULLFSBLOCK)
3733 return error;
3734
3735 return xfs_bmap_btalloc_low_space(ap, args);
3736 }
3737
3738 static int
xfs_bmap_btalloc(struct xfs_bmalloca * ap)3739 xfs_bmap_btalloc(
3740 struct xfs_bmalloca *ap)
3741 {
3742 struct xfs_mount *mp = ap->ip->i_mount;
3743 struct xfs_alloc_arg args = {
3744 .tp = ap->tp,
3745 .mp = mp,
3746 .fsbno = NULLFSBLOCK,
3747 .oinfo = XFS_RMAP_OINFO_SKIP_UPDATE,
3748 .minleft = ap->minleft,
3749 .wasdel = ap->wasdel,
3750 .resv = XFS_AG_RESV_NONE,
3751 .datatype = ap->datatype,
3752 .alignment = 1,
3753 .minalignslop = 0,
3754 };
3755 xfs_fileoff_t orig_offset;
3756 xfs_extlen_t orig_length;
3757 int error;
3758 int stripe_align;
3759
3760 ASSERT(ap->length);
3761 orig_offset = ap->offset;
3762 orig_length = ap->length;
3763
3764 stripe_align = xfs_bmap_compute_alignments(ap, &args);
3765
3766 /* Trim the allocation back to the maximum an AG can fit. */
3767 args.maxlen = min(ap->length, mp->m_ag_max_usable);
3768
3769 if (unlikely(XFS_TEST_ERROR(false, mp,
3770 XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT)))
3771 error = xfs_bmap_exact_minlen_extent_alloc(ap, &args);
3772 else if ((ap->datatype & XFS_ALLOC_USERDATA) &&
3773 xfs_inode_is_filestream(ap->ip))
3774 error = xfs_bmap_btalloc_filestreams(ap, &args, stripe_align);
3775 else
3776 error = xfs_bmap_btalloc_best_length(ap, &args, stripe_align);
3777 if (error)
3778 return error;
3779
3780 if (args.fsbno != NULLFSBLOCK) {
3781 xfs_bmap_process_allocated_extent(ap, &args, orig_offset,
3782 orig_length);
3783 } else {
3784 ap->blkno = NULLFSBLOCK;
3785 ap->length = 0;
3786 }
3787 return 0;
3788 }
3789
3790 /* Trim extent to fit a logical block range. */
3791 void
xfs_trim_extent(struct xfs_bmbt_irec * irec,xfs_fileoff_t bno,xfs_filblks_t len)3792 xfs_trim_extent(
3793 struct xfs_bmbt_irec *irec,
3794 xfs_fileoff_t bno,
3795 xfs_filblks_t len)
3796 {
3797 xfs_fileoff_t distance;
3798 xfs_fileoff_t end = bno + len;
3799
3800 if (irec->br_startoff + irec->br_blockcount <= bno ||
3801 irec->br_startoff >= end) {
3802 irec->br_blockcount = 0;
3803 return;
3804 }
3805
3806 if (irec->br_startoff < bno) {
3807 distance = bno - irec->br_startoff;
3808 if (isnullstartblock(irec->br_startblock))
3809 irec->br_startblock = DELAYSTARTBLOCK;
3810 if (irec->br_startblock != DELAYSTARTBLOCK &&
3811 irec->br_startblock != HOLESTARTBLOCK)
3812 irec->br_startblock += distance;
3813 irec->br_startoff += distance;
3814 irec->br_blockcount -= distance;
3815 }
3816
3817 if (end < irec->br_startoff + irec->br_blockcount) {
3818 distance = irec->br_startoff + irec->br_blockcount - end;
3819 irec->br_blockcount -= distance;
3820 }
3821 }
3822
3823 /*
3824 * Trim the returned map to the required bounds
3825 */
3826 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,uint32_t flags)3827 xfs_bmapi_trim_map(
3828 struct xfs_bmbt_irec *mval,
3829 struct xfs_bmbt_irec *got,
3830 xfs_fileoff_t *bno,
3831 xfs_filblks_t len,
3832 xfs_fileoff_t obno,
3833 xfs_fileoff_t end,
3834 int n,
3835 uint32_t flags)
3836 {
3837 if ((flags & XFS_BMAPI_ENTIRE) ||
3838 got->br_startoff + got->br_blockcount <= obno) {
3839 *mval = *got;
3840 if (isnullstartblock(got->br_startblock))
3841 mval->br_startblock = DELAYSTARTBLOCK;
3842 return;
3843 }
3844
3845 if (obno > *bno)
3846 *bno = obno;
3847 ASSERT((*bno >= obno) || (n == 0));
3848 ASSERT(*bno < end);
3849 mval->br_startoff = *bno;
3850 if (isnullstartblock(got->br_startblock))
3851 mval->br_startblock = DELAYSTARTBLOCK;
3852 else
3853 mval->br_startblock = got->br_startblock +
3854 (*bno - got->br_startoff);
3855 /*
3856 * Return the minimum of what we got and what we asked for for
3857 * the length. We can use the len variable here because it is
3858 * modified below and we could have been there before coming
3859 * here if the first part of the allocation didn't overlap what
3860 * was asked for.
3861 */
3862 mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3863 got->br_blockcount - (*bno - got->br_startoff));
3864 mval->br_state = got->br_state;
3865 ASSERT(mval->br_blockcount <= len);
3866 return;
3867 }
3868
3869 /*
3870 * Update and validate the extent map to return
3871 */
3872 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,uint32_t flags)3873 xfs_bmapi_update_map(
3874 struct xfs_bmbt_irec **map,
3875 xfs_fileoff_t *bno,
3876 xfs_filblks_t *len,
3877 xfs_fileoff_t obno,
3878 xfs_fileoff_t end,
3879 int *n,
3880 uint32_t flags)
3881 {
3882 xfs_bmbt_irec_t *mval = *map;
3883
3884 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3885 ((mval->br_startoff + mval->br_blockcount) <= end));
3886 ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3887 (mval->br_startoff < obno));
3888
3889 *bno = mval->br_startoff + mval->br_blockcount;
3890 *len = end - *bno;
3891 if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3892 /* update previous map with new information */
3893 ASSERT(mval->br_startblock == mval[-1].br_startblock);
3894 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3895 ASSERT(mval->br_state == mval[-1].br_state);
3896 mval[-1].br_blockcount = mval->br_blockcount;
3897 mval[-1].br_state = mval->br_state;
3898 } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3899 mval[-1].br_startblock != DELAYSTARTBLOCK &&
3900 mval[-1].br_startblock != HOLESTARTBLOCK &&
3901 mval->br_startblock == mval[-1].br_startblock +
3902 mval[-1].br_blockcount &&
3903 mval[-1].br_state == mval->br_state) {
3904 ASSERT(mval->br_startoff ==
3905 mval[-1].br_startoff + mval[-1].br_blockcount);
3906 mval[-1].br_blockcount += mval->br_blockcount;
3907 } else if (*n > 0 &&
3908 mval->br_startblock == DELAYSTARTBLOCK &&
3909 mval[-1].br_startblock == DELAYSTARTBLOCK &&
3910 mval->br_startoff ==
3911 mval[-1].br_startoff + mval[-1].br_blockcount) {
3912 mval[-1].br_blockcount += mval->br_blockcount;
3913 mval[-1].br_state = mval->br_state;
3914 } else if (!((*n == 0) &&
3915 ((mval->br_startoff + mval->br_blockcount) <=
3916 obno))) {
3917 mval++;
3918 (*n)++;
3919 }
3920 *map = mval;
3921 }
3922
3923 /*
3924 * Map file blocks to filesystem blocks without allocation.
3925 */
3926 int
xfs_bmapi_read(struct xfs_inode * ip,xfs_fileoff_t bno,xfs_filblks_t len,struct xfs_bmbt_irec * mval,int * nmap,uint32_t flags)3927 xfs_bmapi_read(
3928 struct xfs_inode *ip,
3929 xfs_fileoff_t bno,
3930 xfs_filblks_t len,
3931 struct xfs_bmbt_irec *mval,
3932 int *nmap,
3933 uint32_t flags)
3934 {
3935 struct xfs_mount *mp = ip->i_mount;
3936 int whichfork = xfs_bmapi_whichfork(flags);
3937 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
3938 struct xfs_bmbt_irec got;
3939 xfs_fileoff_t obno;
3940 xfs_fileoff_t end;
3941 struct xfs_iext_cursor icur;
3942 int error;
3943 bool eof = false;
3944 int n = 0;
3945
3946 ASSERT(*nmap >= 1);
3947 ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK | XFS_BMAPI_ENTIRE)));
3948 xfs_assert_ilocked(ip, XFS_ILOCK_SHARED | XFS_ILOCK_EXCL);
3949
3950 if (WARN_ON_ONCE(!ifp)) {
3951 xfs_bmap_mark_sick(ip, whichfork);
3952 return -EFSCORRUPTED;
3953 }
3954
3955 if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
3956 XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
3957 xfs_bmap_mark_sick(ip, whichfork);
3958 return -EFSCORRUPTED;
3959 }
3960
3961 if (xfs_is_shutdown(mp))
3962 return -EIO;
3963
3964 XFS_STATS_INC(mp, xs_blk_mapr);
3965
3966 error = xfs_iread_extents(NULL, ip, whichfork);
3967 if (error)
3968 return error;
3969
3970 if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got))
3971 eof = true;
3972 end = bno + len;
3973 obno = bno;
3974
3975 while (bno < end && n < *nmap) {
3976 /* Reading past eof, act as though there's a hole up to end. */
3977 if (eof)
3978 got.br_startoff = end;
3979 if (got.br_startoff > bno) {
3980 /* Reading in a hole. */
3981 mval->br_startoff = bno;
3982 mval->br_startblock = HOLESTARTBLOCK;
3983 mval->br_blockcount =
3984 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
3985 mval->br_state = XFS_EXT_NORM;
3986 bno += mval->br_blockcount;
3987 len -= mval->br_blockcount;
3988 mval++;
3989 n++;
3990 continue;
3991 }
3992
3993 /* set up the extent map to return. */
3994 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
3995 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
3996
3997 /* If we're done, stop now. */
3998 if (bno >= end || n >= *nmap)
3999 break;
4000
4001 /* Else go on to the next record. */
4002 if (!xfs_iext_next_extent(ifp, &icur, &got))
4003 eof = true;
4004 }
4005 *nmap = n;
4006 return 0;
4007 }
4008
4009 /*
4010 * Add a delayed allocation extent to an inode. Blocks are reserved from the
4011 * global pool and the extent inserted into the inode in-core extent tree.
4012 *
4013 * On entry, got refers to the first extent beyond the offset of the extent to
4014 * allocate or eof is specified if no such extent exists. On return, got refers
4015 * to the extent record that was inserted to the inode fork.
4016 *
4017 * Note that the allocated extent may have been merged with contiguous extents
4018 * during insertion into the inode fork. Thus, got does not reflect the current
4019 * state of the inode fork on return. If necessary, the caller can use lastx to
4020 * look up the updated record in the inode fork.
4021 */
4022 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,struct xfs_iext_cursor * icur,int eof)4023 xfs_bmapi_reserve_delalloc(
4024 struct xfs_inode *ip,
4025 int whichfork,
4026 xfs_fileoff_t off,
4027 xfs_filblks_t len,
4028 xfs_filblks_t prealloc,
4029 struct xfs_bmbt_irec *got,
4030 struct xfs_iext_cursor *icur,
4031 int eof)
4032 {
4033 struct xfs_mount *mp = ip->i_mount;
4034 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
4035 xfs_extlen_t alen;
4036 xfs_extlen_t indlen;
4037 uint64_t fdblocks;
4038 int error;
4039 xfs_fileoff_t aoff;
4040 bool use_cowextszhint =
4041 whichfork == XFS_COW_FORK && !prealloc;
4042
4043 retry:
4044 /*
4045 * Cap the alloc length. Keep track of prealloc so we know whether to
4046 * tag the inode before we return.
4047 */
4048 aoff = off;
4049 alen = XFS_FILBLKS_MIN(len + prealloc, XFS_MAX_BMBT_EXTLEN);
4050 if (!eof)
4051 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4052 if (prealloc && alen >= len)
4053 prealloc = alen - len;
4054
4055 /*
4056 * If we're targetting the COW fork but aren't creating a speculative
4057 * posteof preallocation, try to expand the reservation to align with
4058 * the COW extent size hint if there's sufficient free space.
4059 *
4060 * Unlike the data fork, the CoW cancellation functions will free all
4061 * the reservations at inactivation, so we don't require that every
4062 * delalloc reservation have a dirty pagecache.
4063 */
4064 if (use_cowextszhint) {
4065 struct xfs_bmbt_irec prev;
4066 xfs_extlen_t extsz = xfs_get_cowextsz_hint(ip);
4067
4068 if (!xfs_iext_peek_prev_extent(ifp, icur, &prev))
4069 prev.br_startoff = NULLFILEOFF;
4070
4071 error = xfs_bmap_extsize_align(mp, got, &prev, extsz, 0, eof,
4072 1, 0, &aoff, &alen);
4073 ASSERT(!error);
4074 }
4075
4076 /*
4077 * Make a transaction-less quota reservation for delayed allocation
4078 * blocks. This number gets adjusted later. We return if we haven't
4079 * allocated blocks already inside this loop.
4080 */
4081 error = xfs_quota_reserve_blkres(ip, alen);
4082 if (error)
4083 goto out;
4084
4085 /*
4086 * Split changing sb for alen and indlen since they could be coming
4087 * from different places.
4088 */
4089 indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4090 ASSERT(indlen > 0);
4091
4092 fdblocks = indlen;
4093 if (XFS_IS_REALTIME_INODE(ip)) {
4094 error = xfs_dec_frextents(mp, xfs_rtb_to_rtx(mp, alen));
4095 if (error)
4096 goto out_unreserve_quota;
4097 } else {
4098 fdblocks += alen;
4099 }
4100
4101 error = xfs_dec_fdblocks(mp, fdblocks, false);
4102 if (error)
4103 goto out_unreserve_frextents;
4104
4105 ip->i_delayed_blks += alen;
4106 xfs_mod_delalloc(ip, alen, indlen);
4107
4108 got->br_startoff = aoff;
4109 got->br_startblock = nullstartblock(indlen);
4110 got->br_blockcount = alen;
4111 got->br_state = XFS_EXT_NORM;
4112
4113 xfs_bmap_add_extent_hole_delay(ip, whichfork, icur, got);
4114
4115 /*
4116 * Tag the inode if blocks were preallocated. Note that COW fork
4117 * preallocation can occur at the start or end of the extent, even when
4118 * prealloc == 0, so we must also check the aligned offset and length.
4119 */
4120 if (whichfork == XFS_DATA_FORK && prealloc)
4121 xfs_inode_set_eofblocks_tag(ip);
4122 if (whichfork == XFS_COW_FORK && (prealloc || aoff < off || alen > len))
4123 xfs_inode_set_cowblocks_tag(ip);
4124
4125 return 0;
4126
4127 out_unreserve_frextents:
4128 if (XFS_IS_REALTIME_INODE(ip))
4129 xfs_add_frextents(mp, xfs_rtb_to_rtx(mp, alen));
4130 out_unreserve_quota:
4131 if (XFS_IS_QUOTA_ON(mp))
4132 xfs_quota_unreserve_blkres(ip, alen);
4133 out:
4134 if (error == -ENOSPC || error == -EDQUOT) {
4135 trace_xfs_delalloc_enospc(ip, off, len);
4136
4137 if (prealloc || use_cowextszhint) {
4138 /* retry without any preallocation */
4139 use_cowextszhint = false;
4140 prealloc = 0;
4141 goto retry;
4142 }
4143 }
4144 return error;
4145 }
4146
4147 static int
xfs_bmapi_allocate(struct xfs_bmalloca * bma)4148 xfs_bmapi_allocate(
4149 struct xfs_bmalloca *bma)
4150 {
4151 struct xfs_mount *mp = bma->ip->i_mount;
4152 int whichfork = xfs_bmapi_whichfork(bma->flags);
4153 struct xfs_ifork *ifp = xfs_ifork_ptr(bma->ip, whichfork);
4154 int error;
4155
4156 ASSERT(bma->length > 0);
4157 ASSERT(bma->length <= XFS_MAX_BMBT_EXTLEN);
4158
4159 if (bma->flags & XFS_BMAPI_CONTIG)
4160 bma->minlen = bma->length;
4161 else
4162 bma->minlen = 1;
4163
4164 if (!(bma->flags & XFS_BMAPI_METADATA)) {
4165 /*
4166 * For the data and COW fork, the first data in the file is
4167 * treated differently to all other allocations. For the
4168 * attribute fork, we only need to ensure the allocated range
4169 * is not on the busy list.
4170 */
4171 bma->datatype = XFS_ALLOC_NOBUSY;
4172 if (whichfork == XFS_DATA_FORK || whichfork == XFS_COW_FORK) {
4173 bma->datatype |= XFS_ALLOC_USERDATA;
4174 if (bma->offset == 0)
4175 bma->datatype |= XFS_ALLOC_INITIAL_USER_DATA;
4176
4177 if (mp->m_dalign && bma->length >= mp->m_dalign) {
4178 error = xfs_bmap_isaeof(bma, whichfork);
4179 if (error)
4180 return error;
4181 }
4182 }
4183 }
4184
4185 if ((bma->datatype & XFS_ALLOC_USERDATA) &&
4186 XFS_IS_REALTIME_INODE(bma->ip))
4187 error = xfs_bmap_rtalloc(bma);
4188 else
4189 error = xfs_bmap_btalloc(bma);
4190 if (error)
4191 return error;
4192 if (bma->blkno == NULLFSBLOCK)
4193 return -ENOSPC;
4194
4195 if (WARN_ON_ONCE(!xfs_valid_startblock(bma->ip, bma->blkno))) {
4196 xfs_bmap_mark_sick(bma->ip, whichfork);
4197 return -EFSCORRUPTED;
4198 }
4199
4200 if (bma->flags & XFS_BMAPI_ZERO) {
4201 error = xfs_zero_extent(bma->ip, bma->blkno, bma->length);
4202 if (error)
4203 return error;
4204 }
4205
4206 if (ifp->if_format == XFS_DINODE_FMT_BTREE && !bma->cur)
4207 bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4208 /*
4209 * Bump the number of extents we've allocated
4210 * in this call.
4211 */
4212 bma->nallocs++;
4213
4214 if (bma->cur && bma->wasdel)
4215 bma->cur->bc_flags |= XFS_BTREE_BMBT_WASDEL;
4216
4217 bma->got.br_startoff = bma->offset;
4218 bma->got.br_startblock = bma->blkno;
4219 bma->got.br_blockcount = bma->length;
4220 bma->got.br_state = XFS_EXT_NORM;
4221
4222 if (bma->flags & XFS_BMAPI_PREALLOC)
4223 bma->got.br_state = XFS_EXT_UNWRITTEN;
4224
4225 if (bma->wasdel)
4226 error = xfs_bmap_add_extent_delay_real(bma, whichfork);
4227 else
4228 error = xfs_bmap_add_extent_hole_real(bma->tp, bma->ip,
4229 whichfork, &bma->icur, &bma->cur, &bma->got,
4230 &bma->logflags, bma->flags);
4231 if (error)
4232 return error;
4233
4234 /*
4235 * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4236 * or xfs_bmap_add_extent_hole_real might have merged it into one of
4237 * the neighbouring ones.
4238 */
4239 xfs_iext_get_extent(ifp, &bma->icur, &bma->got);
4240
4241 ASSERT(bma->got.br_startoff <= bma->offset);
4242 ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4243 bma->offset + bma->length);
4244 ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4245 bma->got.br_state == XFS_EXT_UNWRITTEN);
4246 return 0;
4247 }
4248
4249 STATIC int
xfs_bmapi_convert_unwritten(struct xfs_bmalloca * bma,struct xfs_bmbt_irec * mval,xfs_filblks_t len,uint32_t flags)4250 xfs_bmapi_convert_unwritten(
4251 struct xfs_bmalloca *bma,
4252 struct xfs_bmbt_irec *mval,
4253 xfs_filblks_t len,
4254 uint32_t flags)
4255 {
4256 int whichfork = xfs_bmapi_whichfork(flags);
4257 struct xfs_ifork *ifp = xfs_ifork_ptr(bma->ip, whichfork);
4258 int tmp_logflags = 0;
4259 int error;
4260
4261 /* check if we need to do unwritten->real conversion */
4262 if (mval->br_state == XFS_EXT_UNWRITTEN &&
4263 (flags & XFS_BMAPI_PREALLOC))
4264 return 0;
4265
4266 /* check if we need to do real->unwritten conversion */
4267 if (mval->br_state == XFS_EXT_NORM &&
4268 (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4269 (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4270 return 0;
4271
4272 /*
4273 * Modify (by adding) the state flag, if writing.
4274 */
4275 ASSERT(mval->br_blockcount <= len);
4276 if (ifp->if_format == XFS_DINODE_FMT_BTREE && !bma->cur) {
4277 bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4278 bma->ip, whichfork);
4279 }
4280 mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4281 ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4282
4283 /*
4284 * Before insertion into the bmbt, zero the range being converted
4285 * if required.
4286 */
4287 if (flags & XFS_BMAPI_ZERO) {
4288 error = xfs_zero_extent(bma->ip, mval->br_startblock,
4289 mval->br_blockcount);
4290 if (error)
4291 return error;
4292 }
4293
4294 error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, whichfork,
4295 &bma->icur, &bma->cur, mval, &tmp_logflags);
4296 /*
4297 * Log the inode core unconditionally in the unwritten extent conversion
4298 * path because the conversion might not have done so (e.g., if the
4299 * extent count hasn't changed). We need to make sure the inode is dirty
4300 * in the transaction for the sake of fsync(), even if nothing has
4301 * changed, because fsync() will not force the log for this transaction
4302 * unless it sees the inode pinned.
4303 *
4304 * Note: If we're only converting cow fork extents, there aren't
4305 * any on-disk updates to make, so we don't need to log anything.
4306 */
4307 if (whichfork != XFS_COW_FORK)
4308 bma->logflags |= tmp_logflags | XFS_ILOG_CORE;
4309 if (error)
4310 return error;
4311
4312 /*
4313 * Update our extent pointer, given that
4314 * xfs_bmap_add_extent_unwritten_real might have merged it into one
4315 * of the neighbouring ones.
4316 */
4317 xfs_iext_get_extent(ifp, &bma->icur, &bma->got);
4318
4319 /*
4320 * We may have combined previously unwritten space with written space,
4321 * so generate another request.
4322 */
4323 if (mval->br_blockcount < len)
4324 return -EAGAIN;
4325 return 0;
4326 }
4327
4328 xfs_extlen_t
xfs_bmapi_minleft(struct xfs_trans * tp,struct xfs_inode * ip,int fork)4329 xfs_bmapi_minleft(
4330 struct xfs_trans *tp,
4331 struct xfs_inode *ip,
4332 int fork)
4333 {
4334 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, fork);
4335
4336 if (tp && tp->t_highest_agno != NULLAGNUMBER)
4337 return 0;
4338 if (ifp->if_format != XFS_DINODE_FMT_BTREE)
4339 return 1;
4340 return be16_to_cpu(ifp->if_broot->bb_level) + 1;
4341 }
4342
4343 /*
4344 * Log whatever the flags say, even if error. Otherwise we might miss detecting
4345 * a case where the data is changed, there's an error, and it's not logged so we
4346 * don't shutdown when we should. Don't bother logging extents/btree changes if
4347 * we converted to the other format.
4348 */
4349 static void
xfs_bmapi_finish(struct xfs_bmalloca * bma,int whichfork,int error)4350 xfs_bmapi_finish(
4351 struct xfs_bmalloca *bma,
4352 int whichfork,
4353 int error)
4354 {
4355 struct xfs_ifork *ifp = xfs_ifork_ptr(bma->ip, whichfork);
4356
4357 if ((bma->logflags & xfs_ilog_fext(whichfork)) &&
4358 ifp->if_format != XFS_DINODE_FMT_EXTENTS)
4359 bma->logflags &= ~xfs_ilog_fext(whichfork);
4360 else if ((bma->logflags & xfs_ilog_fbroot(whichfork)) &&
4361 ifp->if_format != XFS_DINODE_FMT_BTREE)
4362 bma->logflags &= ~xfs_ilog_fbroot(whichfork);
4363
4364 if (bma->logflags)
4365 xfs_trans_log_inode(bma->tp, bma->ip, bma->logflags);
4366 if (bma->cur)
4367 xfs_btree_del_cursor(bma->cur, error);
4368 }
4369
4370 /*
4371 * Map file blocks to filesystem blocks, and allocate blocks or convert the
4372 * extent state if necessary. Details behaviour is controlled by the flags
4373 * parameter. Only allocates blocks from a single allocation group, to avoid
4374 * locking problems.
4375 *
4376 * Returns 0 on success and places the extent mappings in mval. nmaps is used
4377 * as an input/output parameter where the caller specifies the maximum number
4378 * of mappings that may be returned and xfs_bmapi_write passes back the number
4379 * of mappings (including existing mappings) it found.
4380 *
4381 * Returns a negative error code on failure, including -ENOSPC when it could not
4382 * allocate any blocks and -ENOSR when it did allocate blocks to convert a
4383 * delalloc range, but those blocks were before the passed in range.
4384 */
4385 int
xfs_bmapi_write(struct xfs_trans * tp,struct xfs_inode * ip,xfs_fileoff_t bno,xfs_filblks_t len,uint32_t flags,xfs_extlen_t total,struct xfs_bmbt_irec * mval,int * nmap)4386 xfs_bmapi_write(
4387 struct xfs_trans *tp, /* transaction pointer */
4388 struct xfs_inode *ip, /* incore inode */
4389 xfs_fileoff_t bno, /* starting file offs. mapped */
4390 xfs_filblks_t len, /* length to map in file */
4391 uint32_t flags, /* XFS_BMAPI_... */
4392 xfs_extlen_t total, /* total blocks needed */
4393 struct xfs_bmbt_irec *mval, /* output: map values */
4394 int *nmap) /* i/o: mval size/count */
4395 {
4396 struct xfs_bmalloca bma = {
4397 .tp = tp,
4398 .ip = ip,
4399 .total = total,
4400 };
4401 struct xfs_mount *mp = ip->i_mount;
4402 int whichfork = xfs_bmapi_whichfork(flags);
4403 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
4404 xfs_fileoff_t end; /* end of mapped file region */
4405 bool eof = false; /* after the end of extents */
4406 int error; /* error return */
4407 int n; /* current extent index */
4408 xfs_fileoff_t obno; /* old block number (offset) */
4409
4410 #ifdef DEBUG
4411 xfs_fileoff_t orig_bno; /* original block number value */
4412 int orig_flags; /* original flags arg value */
4413 xfs_filblks_t orig_len; /* original value of len arg */
4414 struct xfs_bmbt_irec *orig_mval; /* original value of mval */
4415 int orig_nmap; /* original value of *nmap */
4416
4417 orig_bno = bno;
4418 orig_len = len;
4419 orig_flags = flags;
4420 orig_mval = mval;
4421 orig_nmap = *nmap;
4422 #endif
4423
4424 ASSERT(*nmap >= 1);
4425 ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4426 ASSERT(tp != NULL);
4427 ASSERT(len > 0);
4428 ASSERT(ifp->if_format != XFS_DINODE_FMT_LOCAL);
4429 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
4430 ASSERT(!(flags & XFS_BMAPI_REMAP));
4431
4432 /* zeroing is for currently only for data extents, not metadata */
4433 ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) !=
4434 (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO));
4435 /*
4436 * we can allocate unwritten extents or pre-zero allocated blocks,
4437 * but it makes no sense to do both at once. This would result in
4438 * zeroing the unwritten extent twice, but it still being an
4439 * unwritten extent....
4440 */
4441 ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) !=
4442 (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO));
4443
4444 if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
4445 XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
4446 xfs_bmap_mark_sick(ip, whichfork);
4447 return -EFSCORRUPTED;
4448 }
4449
4450 if (xfs_is_shutdown(mp))
4451 return -EIO;
4452
4453 XFS_STATS_INC(mp, xs_blk_mapw);
4454
4455 error = xfs_iread_extents(tp, ip, whichfork);
4456 if (error)
4457 goto error0;
4458
4459 if (!xfs_iext_lookup_extent(ip, ifp, bno, &bma.icur, &bma.got))
4460 eof = true;
4461 if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev))
4462 bma.prev.br_startoff = NULLFILEOFF;
4463 bma.minleft = xfs_bmapi_minleft(tp, ip, whichfork);
4464
4465 n = 0;
4466 end = bno + len;
4467 obno = bno;
4468 while (bno < end && n < *nmap) {
4469 bool need_alloc = false, wasdelay = false;
4470
4471 /* in hole or beyond EOF? */
4472 if (eof || bma.got.br_startoff > bno) {
4473 /*
4474 * CoW fork conversions should /never/ hit EOF or
4475 * holes. There should always be something for us
4476 * to work on.
4477 */
4478 ASSERT(!((flags & XFS_BMAPI_CONVERT) &&
4479 (flags & XFS_BMAPI_COWFORK)));
4480
4481 need_alloc = true;
4482 } else if (isnullstartblock(bma.got.br_startblock)) {
4483 wasdelay = true;
4484 }
4485
4486 /*
4487 * First, deal with the hole before the allocated space
4488 * that we found, if any.
4489 */
4490 if (need_alloc || wasdelay) {
4491 bma.eof = eof;
4492 bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4493 bma.wasdel = wasdelay;
4494 bma.offset = bno;
4495 bma.flags = flags;
4496
4497 /*
4498 * There's a 32/64 bit type mismatch between the
4499 * allocation length request (which can be 64 bits in
4500 * length) and the bma length request, which is
4501 * xfs_extlen_t and therefore 32 bits. Hence we have to
4502 * be careful and do the min() using the larger type to
4503 * avoid overflows.
4504 */
4505 bma.length = XFS_FILBLKS_MIN(len, XFS_MAX_BMBT_EXTLEN);
4506
4507 if (wasdelay) {
4508 bma.length = XFS_FILBLKS_MIN(bma.length,
4509 bma.got.br_blockcount -
4510 (bno - bma.got.br_startoff));
4511 } else {
4512 if (!eof)
4513 bma.length = XFS_FILBLKS_MIN(bma.length,
4514 bma.got.br_startoff - bno);
4515 }
4516
4517 ASSERT(bma.length > 0);
4518 error = xfs_bmapi_allocate(&bma);
4519 if (error) {
4520 /*
4521 * If we already allocated space in a previous
4522 * iteration return what we go so far when
4523 * running out of space.
4524 */
4525 if (error == -ENOSPC && bma.nallocs)
4526 break;
4527 goto error0;
4528 }
4529
4530 /*
4531 * If this is a CoW allocation, record the data in
4532 * the refcount btree for orphan recovery.
4533 */
4534 if (whichfork == XFS_COW_FORK)
4535 xfs_refcount_alloc_cow_extent(tp, bma.blkno,
4536 bma.length);
4537 }
4538
4539 /* Deal with the allocated space we found. */
4540 xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4541 end, n, flags);
4542
4543 /* Execute unwritten extent conversion if necessary */
4544 error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4545 if (error == -EAGAIN)
4546 continue;
4547 if (error)
4548 goto error0;
4549
4550 /* update the extent map to return */
4551 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4552
4553 /*
4554 * If we're done, stop now. Stop when we've allocated
4555 * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise
4556 * the transaction may get too big.
4557 */
4558 if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4559 break;
4560
4561 /* Else go on to the next record. */
4562 bma.prev = bma.got;
4563 if (!xfs_iext_next_extent(ifp, &bma.icur, &bma.got))
4564 eof = true;
4565 }
4566
4567 error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, &bma.logflags,
4568 whichfork);
4569 if (error)
4570 goto error0;
4571
4572 ASSERT(ifp->if_format != XFS_DINODE_FMT_BTREE ||
4573 ifp->if_nextents > XFS_IFORK_MAXEXT(ip, whichfork));
4574 xfs_bmapi_finish(&bma, whichfork, 0);
4575 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4576 orig_nmap, n);
4577
4578 /*
4579 * When converting delayed allocations, xfs_bmapi_allocate ignores
4580 * the passed in bno and always converts from the start of the found
4581 * delalloc extent.
4582 *
4583 * To avoid a successful return with *nmap set to 0, return the magic
4584 * -ENOSR error code for this particular case so that the caller can
4585 * handle it.
4586 */
4587 if (!n) {
4588 ASSERT(bma.nallocs >= *nmap);
4589 return -ENOSR;
4590 }
4591 *nmap = n;
4592 return 0;
4593 error0:
4594 xfs_bmapi_finish(&bma, whichfork, error);
4595 return error;
4596 }
4597
4598 /*
4599 * Convert an existing delalloc extent to real blocks based on file offset. This
4600 * attempts to allocate the entire delalloc extent and may require multiple
4601 * invocations to allocate the target offset if a large enough physical extent
4602 * is not available.
4603 */
4604 static int
xfs_bmapi_convert_one_delalloc(struct xfs_inode * ip,int whichfork,xfs_off_t offset,struct iomap * iomap,unsigned int * seq)4605 xfs_bmapi_convert_one_delalloc(
4606 struct xfs_inode *ip,
4607 int whichfork,
4608 xfs_off_t offset,
4609 struct iomap *iomap,
4610 unsigned int *seq)
4611 {
4612 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
4613 struct xfs_mount *mp = ip->i_mount;
4614 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
4615 struct xfs_bmalloca bma = { NULL };
4616 uint16_t flags = 0;
4617 struct xfs_trans *tp;
4618 int error;
4619
4620 if (whichfork == XFS_COW_FORK)
4621 flags |= IOMAP_F_SHARED;
4622
4623 /*
4624 * Space for the extent and indirect blocks was reserved when the
4625 * delalloc extent was created so there's no need to do so here.
4626 */
4627 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0,
4628 XFS_TRANS_RESERVE, &tp);
4629 if (error)
4630 return error;
4631
4632 xfs_ilock(ip, XFS_ILOCK_EXCL);
4633 xfs_trans_ijoin(tp, ip, 0);
4634
4635 error = xfs_iext_count_extend(tp, ip, whichfork,
4636 XFS_IEXT_ADD_NOSPLIT_CNT);
4637 if (error)
4638 goto out_trans_cancel;
4639
4640 if (!xfs_iext_lookup_extent(ip, ifp, offset_fsb, &bma.icur, &bma.got) ||
4641 bma.got.br_startoff > offset_fsb) {
4642 /*
4643 * No extent found in the range we are trying to convert. This
4644 * should only happen for the COW fork, where another thread
4645 * might have moved the extent to the data fork in the meantime.
4646 */
4647 WARN_ON_ONCE(whichfork != XFS_COW_FORK);
4648 error = -EAGAIN;
4649 goto out_trans_cancel;
4650 }
4651
4652 /*
4653 * If we find a real extent here we raced with another thread converting
4654 * the extent. Just return the real extent at this offset.
4655 */
4656 if (!isnullstartblock(bma.got.br_startblock)) {
4657 xfs_bmbt_to_iomap(ip, iomap, &bma.got, 0, flags,
4658 xfs_iomap_inode_sequence(ip, flags));
4659 if (seq)
4660 *seq = READ_ONCE(ifp->if_seq);
4661 goto out_trans_cancel;
4662 }
4663
4664 bma.tp = tp;
4665 bma.ip = ip;
4666 bma.wasdel = true;
4667 bma.minleft = xfs_bmapi_minleft(tp, ip, whichfork);
4668
4669 /*
4670 * Always allocate convert from the start of the delalloc extent even if
4671 * that is outside the passed in range to create large contiguous
4672 * extents on disk.
4673 */
4674 bma.offset = bma.got.br_startoff;
4675 bma.length = bma.got.br_blockcount;
4676
4677 /*
4678 * When we're converting the delalloc reservations backing dirty pages
4679 * in the page cache, we must be careful about how we create the new
4680 * extents:
4681 *
4682 * New CoW fork extents are created unwritten, turned into real extents
4683 * when we're about to write the data to disk, and mapped into the data
4684 * fork after the write finishes. End of story.
4685 *
4686 * New data fork extents must be mapped in as unwritten and converted
4687 * to real extents after the write succeeds to avoid exposing stale
4688 * disk contents if we crash.
4689 */
4690 bma.flags = XFS_BMAPI_PREALLOC;
4691 if (whichfork == XFS_COW_FORK)
4692 bma.flags |= XFS_BMAPI_COWFORK;
4693
4694 if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev))
4695 bma.prev.br_startoff = NULLFILEOFF;
4696
4697 error = xfs_bmapi_allocate(&bma);
4698 if (error)
4699 goto out_finish;
4700
4701 XFS_STATS_ADD(mp, xs_xstrat_bytes, XFS_FSB_TO_B(mp, bma.length));
4702 XFS_STATS_INC(mp, xs_xstrat_quick);
4703
4704 ASSERT(!isnullstartblock(bma.got.br_startblock));
4705 xfs_bmbt_to_iomap(ip, iomap, &bma.got, 0, flags,
4706 xfs_iomap_inode_sequence(ip, flags));
4707 if (seq)
4708 *seq = READ_ONCE(ifp->if_seq);
4709
4710 if (whichfork == XFS_COW_FORK)
4711 xfs_refcount_alloc_cow_extent(tp, bma.blkno, bma.length);
4712
4713 error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, &bma.logflags,
4714 whichfork);
4715 if (error)
4716 goto out_finish;
4717
4718 xfs_bmapi_finish(&bma, whichfork, 0);
4719 error = xfs_trans_commit(tp);
4720 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4721 return error;
4722
4723 out_finish:
4724 xfs_bmapi_finish(&bma, whichfork, error);
4725 out_trans_cancel:
4726 xfs_trans_cancel(tp);
4727 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4728 return error;
4729 }
4730
4731 /*
4732 * Pass in a dellalloc extent and convert it to real extents, return the real
4733 * extent that maps offset_fsb in iomap.
4734 */
4735 int
xfs_bmapi_convert_delalloc(struct xfs_inode * ip,int whichfork,loff_t offset,struct iomap * iomap,unsigned int * seq)4736 xfs_bmapi_convert_delalloc(
4737 struct xfs_inode *ip,
4738 int whichfork,
4739 loff_t offset,
4740 struct iomap *iomap,
4741 unsigned int *seq)
4742 {
4743 int error;
4744
4745 /*
4746 * Attempt to allocate whatever delalloc extent currently backs offset
4747 * and put the result into iomap. Allocate in a loop because it may
4748 * take several attempts to allocate real blocks for a contiguous
4749 * delalloc extent if free space is sufficiently fragmented.
4750 */
4751 do {
4752 error = xfs_bmapi_convert_one_delalloc(ip, whichfork, offset,
4753 iomap, seq);
4754 if (error)
4755 return error;
4756 } while (iomap->offset + iomap->length <= offset);
4757
4758 return 0;
4759 }
4760
4761 int
xfs_bmapi_remap(struct xfs_trans * tp,struct xfs_inode * ip,xfs_fileoff_t bno,xfs_filblks_t len,xfs_fsblock_t startblock,uint32_t flags)4762 xfs_bmapi_remap(
4763 struct xfs_trans *tp,
4764 struct xfs_inode *ip,
4765 xfs_fileoff_t bno,
4766 xfs_filblks_t len,
4767 xfs_fsblock_t startblock,
4768 uint32_t flags)
4769 {
4770 struct xfs_mount *mp = ip->i_mount;
4771 struct xfs_ifork *ifp;
4772 struct xfs_btree_cur *cur = NULL;
4773 struct xfs_bmbt_irec got;
4774 struct xfs_iext_cursor icur;
4775 int whichfork = xfs_bmapi_whichfork(flags);
4776 int logflags = 0, error;
4777
4778 ifp = xfs_ifork_ptr(ip, whichfork);
4779 ASSERT(len > 0);
4780 ASSERT(len <= (xfs_filblks_t)XFS_MAX_BMBT_EXTLEN);
4781 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
4782 ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC |
4783 XFS_BMAPI_NORMAP)));
4784 ASSERT((flags & (XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC)) !=
4785 (XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC));
4786
4787 if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
4788 XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
4789 xfs_bmap_mark_sick(ip, whichfork);
4790 return -EFSCORRUPTED;
4791 }
4792
4793 if (xfs_is_shutdown(mp))
4794 return -EIO;
4795
4796 error = xfs_iread_extents(tp, ip, whichfork);
4797 if (error)
4798 return error;
4799
4800 if (xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) {
4801 /* make sure we only reflink into a hole. */
4802 ASSERT(got.br_startoff > bno);
4803 ASSERT(got.br_startoff - bno >= len);
4804 }
4805
4806 ip->i_nblocks += len;
4807 ip->i_delayed_blks -= len; /* see xfs_bmap_defer_add */
4808 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4809
4810 if (ifp->if_format == XFS_DINODE_FMT_BTREE)
4811 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
4812
4813 got.br_startoff = bno;
4814 got.br_startblock = startblock;
4815 got.br_blockcount = len;
4816 if (flags & XFS_BMAPI_PREALLOC)
4817 got.br_state = XFS_EXT_UNWRITTEN;
4818 else
4819 got.br_state = XFS_EXT_NORM;
4820
4821 error = xfs_bmap_add_extent_hole_real(tp, ip, whichfork, &icur,
4822 &cur, &got, &logflags, flags);
4823 if (error)
4824 goto error0;
4825
4826 error = xfs_bmap_btree_to_extents(tp, ip, cur, &logflags, whichfork);
4827
4828 error0:
4829 if (ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS)
4830 logflags &= ~XFS_ILOG_DEXT;
4831 else if (ip->i_df.if_format != XFS_DINODE_FMT_BTREE)
4832 logflags &= ~XFS_ILOG_DBROOT;
4833
4834 if (logflags)
4835 xfs_trans_log_inode(tp, ip, logflags);
4836 if (cur)
4837 xfs_btree_del_cursor(cur, error);
4838 return error;
4839 }
4840
4841 /*
4842 * When a delalloc extent is split (e.g., due to a hole punch), the original
4843 * indlen reservation must be shared across the two new extents that are left
4844 * behind.
4845 *
4846 * Given the original reservation and the worst case indlen for the two new
4847 * extents (as calculated by xfs_bmap_worst_indlen()), split the original
4848 * reservation fairly across the two new extents. If necessary, steal available
4849 * blocks from a deleted extent to make up a reservation deficiency (e.g., if
4850 * ores == 1). The number of stolen blocks is returned. The availability and
4851 * subsequent accounting of stolen blocks is the responsibility of the caller.
4852 */
4853 static void
xfs_bmap_split_indlen(xfs_filblks_t ores,xfs_filblks_t * indlen1,xfs_filblks_t * indlen2)4854 xfs_bmap_split_indlen(
4855 xfs_filblks_t ores, /* original res. */
4856 xfs_filblks_t *indlen1, /* ext1 worst indlen */
4857 xfs_filblks_t *indlen2) /* ext2 worst indlen */
4858 {
4859 xfs_filblks_t len1 = *indlen1;
4860 xfs_filblks_t len2 = *indlen2;
4861 xfs_filblks_t nres = len1 + len2; /* new total res. */
4862 xfs_filblks_t resfactor;
4863
4864 /*
4865 * We can't meet the total required reservation for the two extents.
4866 * Calculate the percent of the overall shortage between both extents
4867 * and apply this percentage to each of the requested indlen values.
4868 * This distributes the shortage fairly and reduces the chances that one
4869 * of the two extents is left with nothing when extents are repeatedly
4870 * split.
4871 */
4872 resfactor = (ores * 100);
4873 do_div(resfactor, nres);
4874 len1 *= resfactor;
4875 do_div(len1, 100);
4876 len2 *= resfactor;
4877 do_div(len2, 100);
4878 ASSERT(len1 + len2 <= ores);
4879 ASSERT(len1 < *indlen1 && len2 < *indlen2);
4880
4881 /*
4882 * Hand out the remainder to each extent. If one of the two reservations
4883 * is zero, we want to make sure that one gets a block first. The loop
4884 * below starts with len1, so hand len2 a block right off the bat if it
4885 * is zero.
4886 */
4887 ores -= (len1 + len2);
4888 ASSERT((*indlen1 - len1) + (*indlen2 - len2) >= ores);
4889 if (ores && !len2 && *indlen2) {
4890 len2++;
4891 ores--;
4892 }
4893 while (ores) {
4894 if (len1 < *indlen1) {
4895 len1++;
4896 ores--;
4897 }
4898 if (!ores)
4899 break;
4900 if (len2 < *indlen2) {
4901 len2++;
4902 ores--;
4903 }
4904 }
4905
4906 *indlen1 = len1;
4907 *indlen2 = len2;
4908 }
4909
4910 void
xfs_bmap_del_extent_delay(struct xfs_inode * ip,int whichfork,struct xfs_iext_cursor * icur,struct xfs_bmbt_irec * got,struct xfs_bmbt_irec * del)4911 xfs_bmap_del_extent_delay(
4912 struct xfs_inode *ip,
4913 int whichfork,
4914 struct xfs_iext_cursor *icur,
4915 struct xfs_bmbt_irec *got,
4916 struct xfs_bmbt_irec *del)
4917 {
4918 struct xfs_mount *mp = ip->i_mount;
4919 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
4920 struct xfs_bmbt_irec new;
4921 int64_t da_old, da_new, da_diff = 0;
4922 xfs_fileoff_t del_endoff, got_endoff;
4923 xfs_filblks_t got_indlen, new_indlen, stolen = 0;
4924 uint32_t state = xfs_bmap_fork_to_state(whichfork);
4925 uint64_t fdblocks;
4926 bool isrt;
4927
4928 XFS_STATS_INC(mp, xs_del_exlist);
4929
4930 isrt = xfs_ifork_is_realtime(ip, whichfork);
4931 del_endoff = del->br_startoff + del->br_blockcount;
4932 got_endoff = got->br_startoff + got->br_blockcount;
4933 da_old = startblockval(got->br_startblock);
4934 da_new = 0;
4935
4936 ASSERT(del->br_blockcount > 0);
4937 ASSERT(got->br_startoff <= del->br_startoff);
4938 ASSERT(got_endoff >= del_endoff);
4939
4940 /*
4941 * Update the inode delalloc counter now and wait to update the
4942 * sb counters as we might have to borrow some blocks for the
4943 * indirect block accounting.
4944 */
4945 xfs_quota_unreserve_blkres(ip, del->br_blockcount);
4946 ip->i_delayed_blks -= del->br_blockcount;
4947
4948 if (got->br_startoff == del->br_startoff)
4949 state |= BMAP_LEFT_FILLING;
4950 if (got_endoff == del_endoff)
4951 state |= BMAP_RIGHT_FILLING;
4952
4953 switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
4954 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
4955 /*
4956 * Matches the whole extent. Delete the entry.
4957 */
4958 xfs_iext_remove(ip, icur, state);
4959 xfs_iext_prev(ifp, icur);
4960 break;
4961 case BMAP_LEFT_FILLING:
4962 /*
4963 * Deleting the first part of the extent.
4964 */
4965 got->br_startoff = del_endoff;
4966 got->br_blockcount -= del->br_blockcount;
4967 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4968 got->br_blockcount), da_old);
4969 got->br_startblock = nullstartblock((int)da_new);
4970 xfs_iext_update_extent(ip, state, icur, got);
4971 break;
4972 case BMAP_RIGHT_FILLING:
4973 /*
4974 * Deleting the last part of the extent.
4975 */
4976 got->br_blockcount = got->br_blockcount - del->br_blockcount;
4977 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4978 got->br_blockcount), da_old);
4979 got->br_startblock = nullstartblock((int)da_new);
4980 xfs_iext_update_extent(ip, state, icur, got);
4981 break;
4982 case 0:
4983 /*
4984 * Deleting the middle of the extent.
4985 *
4986 * Distribute the original indlen reservation across the two new
4987 * extents. Steal blocks from the deleted extent if necessary.
4988 * Stealing blocks simply fudges the fdblocks accounting below.
4989 * Warn if either of the new indlen reservations is zero as this
4990 * can lead to delalloc problems.
4991 */
4992 got->br_blockcount = del->br_startoff - got->br_startoff;
4993 got_indlen = xfs_bmap_worst_indlen(ip, got->br_blockcount);
4994
4995 new.br_blockcount = got_endoff - del_endoff;
4996 new_indlen = xfs_bmap_worst_indlen(ip, new.br_blockcount);
4997
4998 WARN_ON_ONCE(!got_indlen || !new_indlen);
4999 /*
5000 * Steal as many blocks as we can to try and satisfy the worst
5001 * case indlen for both new extents.
5002 *
5003 * However, we can't just steal reservations from the data
5004 * blocks if this is an RT inodes as the data and metadata
5005 * blocks come from different pools. We'll have to live with
5006 * under-filled indirect reservation in this case.
5007 */
5008 da_new = got_indlen + new_indlen;
5009 if (da_new > da_old && !isrt) {
5010 stolen = XFS_FILBLKS_MIN(da_new - da_old,
5011 del->br_blockcount);
5012 da_old += stolen;
5013 }
5014 if (da_new > da_old)
5015 xfs_bmap_split_indlen(da_old, &got_indlen, &new_indlen);
5016 da_new = got_indlen + new_indlen;
5017
5018 got->br_startblock = nullstartblock((int)got_indlen);
5019
5020 new.br_startoff = del_endoff;
5021 new.br_state = got->br_state;
5022 new.br_startblock = nullstartblock((int)new_indlen);
5023
5024 xfs_iext_update_extent(ip, state, icur, got);
5025 xfs_iext_next(ifp, icur);
5026 xfs_iext_insert(ip, icur, &new, state);
5027
5028 del->br_blockcount -= stolen;
5029 break;
5030 }
5031
5032 ASSERT(da_old >= da_new);
5033 da_diff = da_old - da_new;
5034 fdblocks = da_diff;
5035
5036 if (isrt)
5037 xfs_add_frextents(mp, xfs_rtb_to_rtx(mp, del->br_blockcount));
5038 else
5039 fdblocks += del->br_blockcount;
5040
5041 xfs_add_fdblocks(mp, fdblocks);
5042 xfs_mod_delalloc(ip, -(int64_t)del->br_blockcount, -da_diff);
5043 }
5044
5045 void
xfs_bmap_del_extent_cow(struct xfs_inode * ip,struct xfs_iext_cursor * icur,struct xfs_bmbt_irec * got,struct xfs_bmbt_irec * del)5046 xfs_bmap_del_extent_cow(
5047 struct xfs_inode *ip,
5048 struct xfs_iext_cursor *icur,
5049 struct xfs_bmbt_irec *got,
5050 struct xfs_bmbt_irec *del)
5051 {
5052 struct xfs_mount *mp = ip->i_mount;
5053 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, XFS_COW_FORK);
5054 struct xfs_bmbt_irec new;
5055 xfs_fileoff_t del_endoff, got_endoff;
5056 uint32_t state = BMAP_COWFORK;
5057
5058 XFS_STATS_INC(mp, xs_del_exlist);
5059
5060 del_endoff = del->br_startoff + del->br_blockcount;
5061 got_endoff = got->br_startoff + got->br_blockcount;
5062
5063 ASSERT(del->br_blockcount > 0);
5064 ASSERT(got->br_startoff <= del->br_startoff);
5065 ASSERT(got_endoff >= del_endoff);
5066 ASSERT(!isnullstartblock(got->br_startblock));
5067
5068 if (got->br_startoff == del->br_startoff)
5069 state |= BMAP_LEFT_FILLING;
5070 if (got_endoff == del_endoff)
5071 state |= BMAP_RIGHT_FILLING;
5072
5073 switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
5074 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
5075 /*
5076 * Matches the whole extent. Delete the entry.
5077 */
5078 xfs_iext_remove(ip, icur, state);
5079 xfs_iext_prev(ifp, icur);
5080 break;
5081 case BMAP_LEFT_FILLING:
5082 /*
5083 * Deleting the first part of the extent.
5084 */
5085 got->br_startoff = del_endoff;
5086 got->br_blockcount -= del->br_blockcount;
5087 got->br_startblock = del->br_startblock + del->br_blockcount;
5088 xfs_iext_update_extent(ip, state, icur, got);
5089 break;
5090 case BMAP_RIGHT_FILLING:
5091 /*
5092 * Deleting the last part of the extent.
5093 */
5094 got->br_blockcount -= del->br_blockcount;
5095 xfs_iext_update_extent(ip, state, icur, got);
5096 break;
5097 case 0:
5098 /*
5099 * Deleting the middle of the extent.
5100 */
5101 got->br_blockcount = del->br_startoff - got->br_startoff;
5102
5103 new.br_startoff = del_endoff;
5104 new.br_blockcount = got_endoff - del_endoff;
5105 new.br_state = got->br_state;
5106 new.br_startblock = del->br_startblock + del->br_blockcount;
5107
5108 xfs_iext_update_extent(ip, state, icur, got);
5109 xfs_iext_next(ifp, icur);
5110 xfs_iext_insert(ip, icur, &new, state);
5111 break;
5112 }
5113 ip->i_delayed_blks -= del->br_blockcount;
5114 }
5115
5116 /*
5117 * Called by xfs_bmapi to update file extent records and the btree
5118 * after removing space.
5119 */
5120 STATIC int /* error */
xfs_bmap_del_extent_real(xfs_inode_t * ip,xfs_trans_t * tp,struct xfs_iext_cursor * icur,struct xfs_btree_cur * cur,xfs_bmbt_irec_t * del,int * logflagsp,int whichfork,uint32_t bflags)5121 xfs_bmap_del_extent_real(
5122 xfs_inode_t *ip, /* incore inode pointer */
5123 xfs_trans_t *tp, /* current transaction pointer */
5124 struct xfs_iext_cursor *icur,
5125 struct xfs_btree_cur *cur, /* if null, not a btree */
5126 xfs_bmbt_irec_t *del, /* data to remove from extents */
5127 int *logflagsp, /* inode logging flags */
5128 int whichfork, /* data or attr fork */
5129 uint32_t bflags) /* bmapi flags */
5130 {
5131 xfs_fsblock_t del_endblock=0; /* first block past del */
5132 xfs_fileoff_t del_endoff; /* first offset past del */
5133 int error = 0; /* error return value */
5134 struct xfs_bmbt_irec got; /* current extent entry */
5135 xfs_fileoff_t got_endoff; /* first offset past got */
5136 int i; /* temp state */
5137 struct xfs_ifork *ifp; /* inode fork pointer */
5138 xfs_mount_t *mp; /* mount structure */
5139 xfs_filblks_t nblks; /* quota/sb block count */
5140 xfs_bmbt_irec_t new; /* new record to be inserted */
5141 /* REFERENCED */
5142 uint qfield; /* quota field to update */
5143 uint32_t state = xfs_bmap_fork_to_state(whichfork);
5144 struct xfs_bmbt_irec old;
5145
5146 *logflagsp = 0;
5147
5148 mp = ip->i_mount;
5149 XFS_STATS_INC(mp, xs_del_exlist);
5150
5151 ifp = xfs_ifork_ptr(ip, whichfork);
5152 ASSERT(del->br_blockcount > 0);
5153 xfs_iext_get_extent(ifp, icur, &got);
5154 ASSERT(got.br_startoff <= del->br_startoff);
5155 del_endoff = del->br_startoff + del->br_blockcount;
5156 got_endoff = got.br_startoff + got.br_blockcount;
5157 ASSERT(got_endoff >= del_endoff);
5158 ASSERT(!isnullstartblock(got.br_startblock));
5159 qfield = 0;
5160
5161 /*
5162 * If it's the case where the directory code is running with no block
5163 * reservation, and the deleted block is in the middle of its extent,
5164 * and the resulting insert of an extent would cause transformation to
5165 * btree format, then reject it. The calling code will then swap blocks
5166 * around instead. We have to do this now, rather than waiting for the
5167 * conversion to btree format, since the transaction will be dirty then.
5168 */
5169 if (tp->t_blk_res == 0 &&
5170 ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
5171 ifp->if_nextents >= XFS_IFORK_MAXEXT(ip, whichfork) &&
5172 del->br_startoff > got.br_startoff && del_endoff < got_endoff)
5173 return -ENOSPC;
5174
5175 *logflagsp = XFS_ILOG_CORE;
5176 if (xfs_ifork_is_realtime(ip, whichfork))
5177 qfield = XFS_TRANS_DQ_RTBCOUNT;
5178 else
5179 qfield = XFS_TRANS_DQ_BCOUNT;
5180 nblks = del->br_blockcount;
5181
5182 del_endblock = del->br_startblock + del->br_blockcount;
5183 if (cur) {
5184 error = xfs_bmbt_lookup_eq(cur, &got, &i);
5185 if (error)
5186 return error;
5187 if (XFS_IS_CORRUPT(mp, i != 1)) {
5188 xfs_btree_mark_sick(cur);
5189 return -EFSCORRUPTED;
5190 }
5191 }
5192
5193 if (got.br_startoff == del->br_startoff)
5194 state |= BMAP_LEFT_FILLING;
5195 if (got_endoff == del_endoff)
5196 state |= BMAP_RIGHT_FILLING;
5197
5198 switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
5199 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
5200 /*
5201 * Matches the whole extent. Delete the entry.
5202 */
5203 xfs_iext_remove(ip, icur, state);
5204 xfs_iext_prev(ifp, icur);
5205 ifp->if_nextents--;
5206
5207 *logflagsp |= XFS_ILOG_CORE;
5208 if (!cur) {
5209 *logflagsp |= xfs_ilog_fext(whichfork);
5210 break;
5211 }
5212 if ((error = xfs_btree_delete(cur, &i)))
5213 return error;
5214 if (XFS_IS_CORRUPT(mp, i != 1)) {
5215 xfs_btree_mark_sick(cur);
5216 return -EFSCORRUPTED;
5217 }
5218 break;
5219 case BMAP_LEFT_FILLING:
5220 /*
5221 * Deleting the first part of the extent.
5222 */
5223 got.br_startoff = del_endoff;
5224 got.br_startblock = del_endblock;
5225 got.br_blockcount -= del->br_blockcount;
5226 xfs_iext_update_extent(ip, state, icur, &got);
5227 if (!cur) {
5228 *logflagsp |= xfs_ilog_fext(whichfork);
5229 break;
5230 }
5231 error = xfs_bmbt_update(cur, &got);
5232 if (error)
5233 return error;
5234 break;
5235 case BMAP_RIGHT_FILLING:
5236 /*
5237 * Deleting the last part of the extent.
5238 */
5239 got.br_blockcount -= del->br_blockcount;
5240 xfs_iext_update_extent(ip, state, icur, &got);
5241 if (!cur) {
5242 *logflagsp |= xfs_ilog_fext(whichfork);
5243 break;
5244 }
5245 error = xfs_bmbt_update(cur, &got);
5246 if (error)
5247 return error;
5248 break;
5249 case 0:
5250 /*
5251 * Deleting the middle of the extent.
5252 */
5253
5254 old = got;
5255
5256 got.br_blockcount = del->br_startoff - got.br_startoff;
5257 xfs_iext_update_extent(ip, state, icur, &got);
5258
5259 new.br_startoff = del_endoff;
5260 new.br_blockcount = got_endoff - del_endoff;
5261 new.br_state = got.br_state;
5262 new.br_startblock = del_endblock;
5263
5264 *logflagsp |= XFS_ILOG_CORE;
5265 if (cur) {
5266 error = xfs_bmbt_update(cur, &got);
5267 if (error)
5268 return error;
5269 error = xfs_btree_increment(cur, 0, &i);
5270 if (error)
5271 return error;
5272 cur->bc_rec.b = new;
5273 error = xfs_btree_insert(cur, &i);
5274 if (error && error != -ENOSPC)
5275 return error;
5276 /*
5277 * If get no-space back from btree insert, it tried a
5278 * split, and we have a zero block reservation. Fix up
5279 * our state and return the error.
5280 */
5281 if (error == -ENOSPC) {
5282 /*
5283 * Reset the cursor, don't trust it after any
5284 * insert operation.
5285 */
5286 error = xfs_bmbt_lookup_eq(cur, &got, &i);
5287 if (error)
5288 return error;
5289 if (XFS_IS_CORRUPT(mp, i != 1)) {
5290 xfs_btree_mark_sick(cur);
5291 return -EFSCORRUPTED;
5292 }
5293 /*
5294 * Update the btree record back
5295 * to the original value.
5296 */
5297 error = xfs_bmbt_update(cur, &old);
5298 if (error)
5299 return error;
5300 /*
5301 * Reset the extent record back
5302 * to the original value.
5303 */
5304 xfs_iext_update_extent(ip, state, icur, &old);
5305 *logflagsp = 0;
5306 return -ENOSPC;
5307 }
5308 if (XFS_IS_CORRUPT(mp, i != 1)) {
5309 xfs_btree_mark_sick(cur);
5310 return -EFSCORRUPTED;
5311 }
5312 } else
5313 *logflagsp |= xfs_ilog_fext(whichfork);
5314
5315 ifp->if_nextents++;
5316 xfs_iext_next(ifp, icur);
5317 xfs_iext_insert(ip, icur, &new, state);
5318 break;
5319 }
5320
5321 /* remove reverse mapping */
5322 xfs_rmap_unmap_extent(tp, ip, whichfork, del);
5323
5324 /*
5325 * If we need to, add to list of extents to delete.
5326 */
5327 if (!(bflags & XFS_BMAPI_REMAP)) {
5328 if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) {
5329 xfs_refcount_decrease_extent(tp, del);
5330 } else if (xfs_ifork_is_realtime(ip, whichfork)) {
5331 /*
5332 * Ensure the bitmap and summary inodes are locked
5333 * and joined to the transaction before modifying them.
5334 */
5335 if (!(tp->t_flags & XFS_TRANS_RTBITMAP_LOCKED)) {
5336 tp->t_flags |= XFS_TRANS_RTBITMAP_LOCKED;
5337 xfs_rtbitmap_lock(mp);
5338 xfs_rtbitmap_trans_join(tp);
5339 }
5340 error = xfs_rtfree_blocks(tp, del->br_startblock,
5341 del->br_blockcount);
5342 } else {
5343 unsigned int efi_flags = 0;
5344
5345 if ((bflags & XFS_BMAPI_NODISCARD) ||
5346 del->br_state == XFS_EXT_UNWRITTEN)
5347 efi_flags |= XFS_FREE_EXTENT_SKIP_DISCARD;
5348
5349 error = xfs_free_extent_later(tp, del->br_startblock,
5350 del->br_blockcount, NULL,
5351 XFS_AG_RESV_NONE, efi_flags);
5352 }
5353 if (error)
5354 return error;
5355 }
5356
5357 /*
5358 * Adjust inode # blocks in the file.
5359 */
5360 if (nblks)
5361 ip->i_nblocks -= nblks;
5362 /*
5363 * Adjust quota data.
5364 */
5365 if (qfield && !(bflags & XFS_BMAPI_REMAP))
5366 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5367
5368 return 0;
5369 }
5370
5371 /*
5372 * Unmap (remove) blocks from a file.
5373 * If nexts is nonzero then the number of extents to remove is limited to
5374 * that value. If not all extents in the block range can be removed then
5375 * *done is set.
5376 */
5377 static int
__xfs_bunmapi(struct xfs_trans * tp,struct xfs_inode * ip,xfs_fileoff_t start,xfs_filblks_t * rlen,uint32_t flags,xfs_extnum_t nexts)5378 __xfs_bunmapi(
5379 struct xfs_trans *tp, /* transaction pointer */
5380 struct xfs_inode *ip, /* incore inode */
5381 xfs_fileoff_t start, /* first file offset deleted */
5382 xfs_filblks_t *rlen, /* i/o: amount remaining */
5383 uint32_t flags, /* misc flags */
5384 xfs_extnum_t nexts) /* number of extents max */
5385 {
5386 struct xfs_btree_cur *cur; /* bmap btree cursor */
5387 struct xfs_bmbt_irec del; /* extent being deleted */
5388 int error; /* error return value */
5389 xfs_extnum_t extno; /* extent number in list */
5390 struct xfs_bmbt_irec got; /* current extent record */
5391 struct xfs_ifork *ifp; /* inode fork pointer */
5392 int isrt; /* freeing in rt area */
5393 int logflags; /* transaction logging flags */
5394 xfs_extlen_t mod; /* rt extent offset */
5395 struct xfs_mount *mp = ip->i_mount;
5396 int tmp_logflags; /* partial logging flags */
5397 int wasdel; /* was a delayed alloc extent */
5398 int whichfork; /* data or attribute fork */
5399 xfs_filblks_t len = *rlen; /* length to unmap in file */
5400 xfs_fileoff_t end;
5401 struct xfs_iext_cursor icur;
5402 bool done = false;
5403
5404 trace_xfs_bunmap(ip, start, len, flags, _RET_IP_);
5405
5406 whichfork = xfs_bmapi_whichfork(flags);
5407 ASSERT(whichfork != XFS_COW_FORK);
5408 ifp = xfs_ifork_ptr(ip, whichfork);
5409 if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp))) {
5410 xfs_bmap_mark_sick(ip, whichfork);
5411 return -EFSCORRUPTED;
5412 }
5413 if (xfs_is_shutdown(mp))
5414 return -EIO;
5415
5416 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
5417 ASSERT(len > 0);
5418 ASSERT(nexts >= 0);
5419
5420 error = xfs_iread_extents(tp, ip, whichfork);
5421 if (error)
5422 return error;
5423
5424 if (xfs_iext_count(ifp) == 0) {
5425 *rlen = 0;
5426 return 0;
5427 }
5428 XFS_STATS_INC(mp, xs_blk_unmap);
5429 isrt = xfs_ifork_is_realtime(ip, whichfork);
5430 end = start + len;
5431
5432 if (!xfs_iext_lookup_extent_before(ip, ifp, &end, &icur, &got)) {
5433 *rlen = 0;
5434 return 0;
5435 }
5436 end--;
5437
5438 logflags = 0;
5439 if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
5440 ASSERT(ifp->if_format == XFS_DINODE_FMT_BTREE);
5441 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5442 } else
5443 cur = NULL;
5444
5445 extno = 0;
5446 while (end != (xfs_fileoff_t)-1 && end >= start &&
5447 (nexts == 0 || extno < nexts)) {
5448 /*
5449 * Is the found extent after a hole in which end lives?
5450 * Just back up to the previous extent, if so.
5451 */
5452 if (got.br_startoff > end &&
5453 !xfs_iext_prev_extent(ifp, &icur, &got)) {
5454 done = true;
5455 break;
5456 }
5457 /*
5458 * Is the last block of this extent before the range
5459 * we're supposed to delete? If so, we're done.
5460 */
5461 end = XFS_FILEOFF_MIN(end,
5462 got.br_startoff + got.br_blockcount - 1);
5463 if (end < start)
5464 break;
5465 /*
5466 * Then deal with the (possibly delayed) allocated space
5467 * we found.
5468 */
5469 del = got;
5470 wasdel = isnullstartblock(del.br_startblock);
5471
5472 if (got.br_startoff < start) {
5473 del.br_startoff = start;
5474 del.br_blockcount -= start - got.br_startoff;
5475 if (!wasdel)
5476 del.br_startblock += start - got.br_startoff;
5477 }
5478 if (del.br_startoff + del.br_blockcount > end + 1)
5479 del.br_blockcount = end + 1 - del.br_startoff;
5480
5481 if (!isrt || (flags & XFS_BMAPI_REMAP))
5482 goto delete;
5483
5484 mod = xfs_rtb_to_rtxoff(mp,
5485 del.br_startblock + del.br_blockcount);
5486 if (mod) {
5487 /*
5488 * Realtime extent not lined up at the end.
5489 * The extent could have been split into written
5490 * and unwritten pieces, or we could just be
5491 * unmapping part of it. But we can't really
5492 * get rid of part of a realtime extent.
5493 */
5494 if (del.br_state == XFS_EXT_UNWRITTEN) {
5495 /*
5496 * This piece is unwritten, or we're not
5497 * using unwritten extents. Skip over it.
5498 */
5499 ASSERT((flags & XFS_BMAPI_REMAP) || end >= mod);
5500 end -= mod > del.br_blockcount ?
5501 del.br_blockcount : mod;
5502 if (end < got.br_startoff &&
5503 !xfs_iext_prev_extent(ifp, &icur, &got)) {
5504 done = true;
5505 break;
5506 }
5507 continue;
5508 }
5509 /*
5510 * It's written, turn it unwritten.
5511 * This is better than zeroing it.
5512 */
5513 ASSERT(del.br_state == XFS_EXT_NORM);
5514 ASSERT(tp->t_blk_res > 0);
5515 /*
5516 * If this spans a realtime extent boundary,
5517 * chop it back to the start of the one we end at.
5518 */
5519 if (del.br_blockcount > mod) {
5520 del.br_startoff += del.br_blockcount - mod;
5521 del.br_startblock += del.br_blockcount - mod;
5522 del.br_blockcount = mod;
5523 }
5524 del.br_state = XFS_EXT_UNWRITTEN;
5525 error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5526 whichfork, &icur, &cur, &del,
5527 &logflags);
5528 if (error)
5529 goto error0;
5530 goto nodelete;
5531 }
5532
5533 mod = xfs_rtb_to_rtxoff(mp, del.br_startblock);
5534 if (mod) {
5535 xfs_extlen_t off = mp->m_sb.sb_rextsize - mod;
5536
5537 /*
5538 * Realtime extent is lined up at the end but not
5539 * at the front. We'll get rid of full extents if
5540 * we can.
5541 */
5542 if (del.br_blockcount > off) {
5543 del.br_blockcount -= off;
5544 del.br_startoff += off;
5545 del.br_startblock += off;
5546 } else if (del.br_startoff == start &&
5547 (del.br_state == XFS_EXT_UNWRITTEN ||
5548 tp->t_blk_res == 0)) {
5549 /*
5550 * Can't make it unwritten. There isn't
5551 * a full extent here so just skip it.
5552 */
5553 ASSERT(end >= del.br_blockcount);
5554 end -= del.br_blockcount;
5555 if (got.br_startoff > end &&
5556 !xfs_iext_prev_extent(ifp, &icur, &got)) {
5557 done = true;
5558 break;
5559 }
5560 continue;
5561 } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5562 struct xfs_bmbt_irec prev;
5563 xfs_fileoff_t unwrite_start;
5564
5565 /*
5566 * This one is already unwritten.
5567 * It must have a written left neighbor.
5568 * Unwrite the killed part of that one and
5569 * try again.
5570 */
5571 if (!xfs_iext_prev_extent(ifp, &icur, &prev))
5572 ASSERT(0);
5573 ASSERT(prev.br_state == XFS_EXT_NORM);
5574 ASSERT(!isnullstartblock(prev.br_startblock));
5575 ASSERT(del.br_startblock ==
5576 prev.br_startblock + prev.br_blockcount);
5577 unwrite_start = max3(start,
5578 del.br_startoff - mod,
5579 prev.br_startoff);
5580 mod = unwrite_start - prev.br_startoff;
5581 prev.br_startoff = unwrite_start;
5582 prev.br_startblock += mod;
5583 prev.br_blockcount -= mod;
5584 prev.br_state = XFS_EXT_UNWRITTEN;
5585 error = xfs_bmap_add_extent_unwritten_real(tp,
5586 ip, whichfork, &icur, &cur,
5587 &prev, &logflags);
5588 if (error)
5589 goto error0;
5590 goto nodelete;
5591 } else {
5592 ASSERT(del.br_state == XFS_EXT_NORM);
5593 del.br_state = XFS_EXT_UNWRITTEN;
5594 error = xfs_bmap_add_extent_unwritten_real(tp,
5595 ip, whichfork, &icur, &cur,
5596 &del, &logflags);
5597 if (error)
5598 goto error0;
5599 goto nodelete;
5600 }
5601 }
5602
5603 delete:
5604 if (wasdel) {
5605 xfs_bmap_del_extent_delay(ip, whichfork, &icur, &got, &del);
5606 } else {
5607 error = xfs_bmap_del_extent_real(ip, tp, &icur, cur,
5608 &del, &tmp_logflags, whichfork,
5609 flags);
5610 logflags |= tmp_logflags;
5611 if (error)
5612 goto error0;
5613 }
5614
5615 end = del.br_startoff - 1;
5616 nodelete:
5617 /*
5618 * If not done go on to the next (previous) record.
5619 */
5620 if (end != (xfs_fileoff_t)-1 && end >= start) {
5621 if (!xfs_iext_get_extent(ifp, &icur, &got) ||
5622 (got.br_startoff > end &&
5623 !xfs_iext_prev_extent(ifp, &icur, &got))) {
5624 done = true;
5625 break;
5626 }
5627 extno++;
5628 }
5629 }
5630 if (done || end == (xfs_fileoff_t)-1 || end < start)
5631 *rlen = 0;
5632 else
5633 *rlen = end - start + 1;
5634
5635 /*
5636 * Convert to a btree if necessary.
5637 */
5638 if (xfs_bmap_needs_btree(ip, whichfork)) {
5639 ASSERT(cur == NULL);
5640 error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
5641 &tmp_logflags, whichfork);
5642 logflags |= tmp_logflags;
5643 } else {
5644 error = xfs_bmap_btree_to_extents(tp, ip, cur, &logflags,
5645 whichfork);
5646 }
5647
5648 error0:
5649 /*
5650 * Log everything. Do this after conversion, there's no point in
5651 * logging the extent records if we've converted to btree format.
5652 */
5653 if ((logflags & xfs_ilog_fext(whichfork)) &&
5654 ifp->if_format != XFS_DINODE_FMT_EXTENTS)
5655 logflags &= ~xfs_ilog_fext(whichfork);
5656 else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5657 ifp->if_format != XFS_DINODE_FMT_BTREE)
5658 logflags &= ~xfs_ilog_fbroot(whichfork);
5659 /*
5660 * Log inode even in the error case, if the transaction
5661 * is dirty we'll need to shut down the filesystem.
5662 */
5663 if (logflags)
5664 xfs_trans_log_inode(tp, ip, logflags);
5665 if (cur) {
5666 if (!error)
5667 cur->bc_bmap.allocated = 0;
5668 xfs_btree_del_cursor(cur, error);
5669 }
5670 return error;
5671 }
5672
5673 /* Unmap a range of a file. */
5674 int
xfs_bunmapi(xfs_trans_t * tp,struct xfs_inode * ip,xfs_fileoff_t bno,xfs_filblks_t len,uint32_t flags,xfs_extnum_t nexts,int * done)5675 xfs_bunmapi(
5676 xfs_trans_t *tp,
5677 struct xfs_inode *ip,
5678 xfs_fileoff_t bno,
5679 xfs_filblks_t len,
5680 uint32_t flags,
5681 xfs_extnum_t nexts,
5682 int *done)
5683 {
5684 int error;
5685
5686 error = __xfs_bunmapi(tp, ip, bno, &len, flags, nexts);
5687 *done = (len == 0);
5688 return error;
5689 }
5690
5691 /*
5692 * Determine whether an extent shift can be accomplished by a merge with the
5693 * extent that precedes the target hole of the shift.
5694 */
5695 STATIC bool
xfs_bmse_can_merge(struct xfs_bmbt_irec * left,struct xfs_bmbt_irec * got,xfs_fileoff_t shift)5696 xfs_bmse_can_merge(
5697 struct xfs_bmbt_irec *left, /* preceding extent */
5698 struct xfs_bmbt_irec *got, /* current extent to shift */
5699 xfs_fileoff_t shift) /* shift fsb */
5700 {
5701 xfs_fileoff_t startoff;
5702
5703 startoff = got->br_startoff - shift;
5704
5705 /*
5706 * The extent, once shifted, must be adjacent in-file and on-disk with
5707 * the preceding extent.
5708 */
5709 if ((left->br_startoff + left->br_blockcount != startoff) ||
5710 (left->br_startblock + left->br_blockcount != got->br_startblock) ||
5711 (left->br_state != got->br_state) ||
5712 (left->br_blockcount + got->br_blockcount > XFS_MAX_BMBT_EXTLEN))
5713 return false;
5714
5715 return true;
5716 }
5717
5718 /*
5719 * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5720 * hole in the file. If an extent shift would result in the extent being fully
5721 * adjacent to the extent that currently precedes the hole, we can merge with
5722 * the preceding extent rather than do the shift.
5723 *
5724 * This function assumes the caller has verified a shift-by-merge is possible
5725 * with the provided extents via xfs_bmse_can_merge().
5726 */
5727 STATIC int
xfs_bmse_merge(struct xfs_trans * tp,struct xfs_inode * ip,int whichfork,xfs_fileoff_t shift,struct xfs_iext_cursor * icur,struct xfs_bmbt_irec * got,struct xfs_bmbt_irec * left,struct xfs_btree_cur * cur,int * logflags)5728 xfs_bmse_merge(
5729 struct xfs_trans *tp,
5730 struct xfs_inode *ip,
5731 int whichfork,
5732 xfs_fileoff_t shift, /* shift fsb */
5733 struct xfs_iext_cursor *icur,
5734 struct xfs_bmbt_irec *got, /* extent to shift */
5735 struct xfs_bmbt_irec *left, /* preceding extent */
5736 struct xfs_btree_cur *cur,
5737 int *logflags) /* output */
5738 {
5739 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
5740 struct xfs_bmbt_irec new;
5741 xfs_filblks_t blockcount;
5742 int error, i;
5743 struct xfs_mount *mp = ip->i_mount;
5744
5745 blockcount = left->br_blockcount + got->br_blockcount;
5746
5747 xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
5748 ASSERT(xfs_bmse_can_merge(left, got, shift));
5749
5750 new = *left;
5751 new.br_blockcount = blockcount;
5752
5753 /*
5754 * Update the on-disk extent count, the btree if necessary and log the
5755 * inode.
5756 */
5757 ifp->if_nextents--;
5758 *logflags |= XFS_ILOG_CORE;
5759 if (!cur) {
5760 *logflags |= XFS_ILOG_DEXT;
5761 goto done;
5762 }
5763
5764 /* lookup and remove the extent to merge */
5765 error = xfs_bmbt_lookup_eq(cur, got, &i);
5766 if (error)
5767 return error;
5768 if (XFS_IS_CORRUPT(mp, i != 1)) {
5769 xfs_btree_mark_sick(cur);
5770 return -EFSCORRUPTED;
5771 }
5772
5773 error = xfs_btree_delete(cur, &i);
5774 if (error)
5775 return error;
5776 if (XFS_IS_CORRUPT(mp, i != 1)) {
5777 xfs_btree_mark_sick(cur);
5778 return -EFSCORRUPTED;
5779 }
5780
5781 /* lookup and update size of the previous extent */
5782 error = xfs_bmbt_lookup_eq(cur, left, &i);
5783 if (error)
5784 return error;
5785 if (XFS_IS_CORRUPT(mp, i != 1)) {
5786 xfs_btree_mark_sick(cur);
5787 return -EFSCORRUPTED;
5788 }
5789
5790 error = xfs_bmbt_update(cur, &new);
5791 if (error)
5792 return error;
5793
5794 /* change to extent format if required after extent removal */
5795 error = xfs_bmap_btree_to_extents(tp, ip, cur, logflags, whichfork);
5796 if (error)
5797 return error;
5798
5799 done:
5800 xfs_iext_remove(ip, icur, 0);
5801 xfs_iext_prev(ifp, icur);
5802 xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur,
5803 &new);
5804
5805 /* update reverse mapping. rmap functions merge the rmaps for us */
5806 xfs_rmap_unmap_extent(tp, ip, whichfork, got);
5807 memcpy(&new, got, sizeof(new));
5808 new.br_startoff = left->br_startoff + left->br_blockcount;
5809 xfs_rmap_map_extent(tp, ip, whichfork, &new);
5810 return 0;
5811 }
5812
5813 static int
xfs_bmap_shift_update_extent(struct xfs_trans * tp,struct xfs_inode * ip,int whichfork,struct xfs_iext_cursor * icur,struct xfs_bmbt_irec * got,struct xfs_btree_cur * cur,int * logflags,xfs_fileoff_t startoff)5814 xfs_bmap_shift_update_extent(
5815 struct xfs_trans *tp,
5816 struct xfs_inode *ip,
5817 int whichfork,
5818 struct xfs_iext_cursor *icur,
5819 struct xfs_bmbt_irec *got,
5820 struct xfs_btree_cur *cur,
5821 int *logflags,
5822 xfs_fileoff_t startoff)
5823 {
5824 struct xfs_mount *mp = ip->i_mount;
5825 struct xfs_bmbt_irec prev = *got;
5826 int error, i;
5827
5828 *logflags |= XFS_ILOG_CORE;
5829
5830 got->br_startoff = startoff;
5831
5832 if (cur) {
5833 error = xfs_bmbt_lookup_eq(cur, &prev, &i);
5834 if (error)
5835 return error;
5836 if (XFS_IS_CORRUPT(mp, i != 1)) {
5837 xfs_btree_mark_sick(cur);
5838 return -EFSCORRUPTED;
5839 }
5840
5841 error = xfs_bmbt_update(cur, got);
5842 if (error)
5843 return error;
5844 } else {
5845 *logflags |= XFS_ILOG_DEXT;
5846 }
5847
5848 xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur,
5849 got);
5850
5851 /* update reverse mapping */
5852 xfs_rmap_unmap_extent(tp, ip, whichfork, &prev);
5853 xfs_rmap_map_extent(tp, ip, whichfork, got);
5854 return 0;
5855 }
5856
5857 int
xfs_bmap_collapse_extents(struct xfs_trans * tp,struct xfs_inode * ip,xfs_fileoff_t * next_fsb,xfs_fileoff_t offset_shift_fsb,bool * done)5858 xfs_bmap_collapse_extents(
5859 struct xfs_trans *tp,
5860 struct xfs_inode *ip,
5861 xfs_fileoff_t *next_fsb,
5862 xfs_fileoff_t offset_shift_fsb,
5863 bool *done)
5864 {
5865 int whichfork = XFS_DATA_FORK;
5866 struct xfs_mount *mp = ip->i_mount;
5867 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
5868 struct xfs_btree_cur *cur = NULL;
5869 struct xfs_bmbt_irec got, prev;
5870 struct xfs_iext_cursor icur;
5871 xfs_fileoff_t new_startoff;
5872 int error = 0;
5873 int logflags = 0;
5874
5875 if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
5876 XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
5877 xfs_bmap_mark_sick(ip, whichfork);
5878 return -EFSCORRUPTED;
5879 }
5880
5881 if (xfs_is_shutdown(mp))
5882 return -EIO;
5883
5884 xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
5885
5886 error = xfs_iread_extents(tp, ip, whichfork);
5887 if (error)
5888 return error;
5889
5890 if (ifp->if_format == XFS_DINODE_FMT_BTREE)
5891 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5892
5893 if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) {
5894 *done = true;
5895 goto del_cursor;
5896 }
5897 if (XFS_IS_CORRUPT(mp, isnullstartblock(got.br_startblock))) {
5898 xfs_bmap_mark_sick(ip, whichfork);
5899 error = -EFSCORRUPTED;
5900 goto del_cursor;
5901 }
5902
5903 new_startoff = got.br_startoff - offset_shift_fsb;
5904 if (xfs_iext_peek_prev_extent(ifp, &icur, &prev)) {
5905 if (new_startoff < prev.br_startoff + prev.br_blockcount) {
5906 error = -EINVAL;
5907 goto del_cursor;
5908 }
5909
5910 if (xfs_bmse_can_merge(&prev, &got, offset_shift_fsb)) {
5911 error = xfs_bmse_merge(tp, ip, whichfork,
5912 offset_shift_fsb, &icur, &got, &prev,
5913 cur, &logflags);
5914 if (error)
5915 goto del_cursor;
5916 goto done;
5917 }
5918 } else {
5919 if (got.br_startoff < offset_shift_fsb) {
5920 error = -EINVAL;
5921 goto del_cursor;
5922 }
5923 }
5924
5925 error = xfs_bmap_shift_update_extent(tp, ip, whichfork, &icur, &got,
5926 cur, &logflags, new_startoff);
5927 if (error)
5928 goto del_cursor;
5929
5930 done:
5931 if (!xfs_iext_next_extent(ifp, &icur, &got)) {
5932 *done = true;
5933 goto del_cursor;
5934 }
5935
5936 *next_fsb = got.br_startoff;
5937 del_cursor:
5938 if (cur)
5939 xfs_btree_del_cursor(cur, error);
5940 if (logflags)
5941 xfs_trans_log_inode(tp, ip, logflags);
5942 return error;
5943 }
5944
5945 /* Make sure we won't be right-shifting an extent past the maximum bound. */
5946 int
xfs_bmap_can_insert_extents(struct xfs_inode * ip,xfs_fileoff_t off,xfs_fileoff_t shift)5947 xfs_bmap_can_insert_extents(
5948 struct xfs_inode *ip,
5949 xfs_fileoff_t off,
5950 xfs_fileoff_t shift)
5951 {
5952 struct xfs_bmbt_irec got;
5953 int is_empty;
5954 int error = 0;
5955
5956 xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL);
5957
5958 if (xfs_is_shutdown(ip->i_mount))
5959 return -EIO;
5960
5961 xfs_ilock(ip, XFS_ILOCK_EXCL);
5962 error = xfs_bmap_last_extent(NULL, ip, XFS_DATA_FORK, &got, &is_empty);
5963 if (!error && !is_empty && got.br_startoff >= off &&
5964 ((got.br_startoff + shift) & BMBT_STARTOFF_MASK) < got.br_startoff)
5965 error = -EINVAL;
5966 xfs_iunlock(ip, XFS_ILOCK_EXCL);
5967
5968 return error;
5969 }
5970
5971 int
xfs_bmap_insert_extents(struct xfs_trans * tp,struct xfs_inode * ip,xfs_fileoff_t * next_fsb,xfs_fileoff_t offset_shift_fsb,bool * done,xfs_fileoff_t stop_fsb)5972 xfs_bmap_insert_extents(
5973 struct xfs_trans *tp,
5974 struct xfs_inode *ip,
5975 xfs_fileoff_t *next_fsb,
5976 xfs_fileoff_t offset_shift_fsb,
5977 bool *done,
5978 xfs_fileoff_t stop_fsb)
5979 {
5980 int whichfork = XFS_DATA_FORK;
5981 struct xfs_mount *mp = ip->i_mount;
5982 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
5983 struct xfs_btree_cur *cur = NULL;
5984 struct xfs_bmbt_irec got, next;
5985 struct xfs_iext_cursor icur;
5986 xfs_fileoff_t new_startoff;
5987 int error = 0;
5988 int logflags = 0;
5989
5990 if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
5991 XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
5992 xfs_bmap_mark_sick(ip, whichfork);
5993 return -EFSCORRUPTED;
5994 }
5995
5996 if (xfs_is_shutdown(mp))
5997 return -EIO;
5998
5999 xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
6000
6001 error = xfs_iread_extents(tp, ip, whichfork);
6002 if (error)
6003 return error;
6004
6005 if (ifp->if_format == XFS_DINODE_FMT_BTREE)
6006 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
6007
6008 if (*next_fsb == NULLFSBLOCK) {
6009 xfs_iext_last(ifp, &icur);
6010 if (!xfs_iext_get_extent(ifp, &icur, &got) ||
6011 stop_fsb > got.br_startoff) {
6012 *done = true;
6013 goto del_cursor;
6014 }
6015 } else {
6016 if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) {
6017 *done = true;
6018 goto del_cursor;
6019 }
6020 }
6021 if (XFS_IS_CORRUPT(mp, isnullstartblock(got.br_startblock))) {
6022 xfs_bmap_mark_sick(ip, whichfork);
6023 error = -EFSCORRUPTED;
6024 goto del_cursor;
6025 }
6026
6027 if (XFS_IS_CORRUPT(mp, stop_fsb > got.br_startoff)) {
6028 xfs_bmap_mark_sick(ip, whichfork);
6029 error = -EFSCORRUPTED;
6030 goto del_cursor;
6031 }
6032
6033 new_startoff = got.br_startoff + offset_shift_fsb;
6034 if (xfs_iext_peek_next_extent(ifp, &icur, &next)) {
6035 if (new_startoff + got.br_blockcount > next.br_startoff) {
6036 error = -EINVAL;
6037 goto del_cursor;
6038 }
6039
6040 /*
6041 * Unlike a left shift (which involves a hole punch), a right
6042 * shift does not modify extent neighbors in any way. We should
6043 * never find mergeable extents in this scenario. Check anyways
6044 * and warn if we encounter two extents that could be one.
6045 */
6046 if (xfs_bmse_can_merge(&got, &next, offset_shift_fsb))
6047 WARN_ON_ONCE(1);
6048 }
6049
6050 error = xfs_bmap_shift_update_extent(tp, ip, whichfork, &icur, &got,
6051 cur, &logflags, new_startoff);
6052 if (error)
6053 goto del_cursor;
6054
6055 if (!xfs_iext_prev_extent(ifp, &icur, &got) ||
6056 stop_fsb >= got.br_startoff + got.br_blockcount) {
6057 *done = true;
6058 goto del_cursor;
6059 }
6060
6061 *next_fsb = got.br_startoff;
6062 del_cursor:
6063 if (cur)
6064 xfs_btree_del_cursor(cur, error);
6065 if (logflags)
6066 xfs_trans_log_inode(tp, ip, logflags);
6067 return error;
6068 }
6069
6070 /*
6071 * Splits an extent into two extents at split_fsb block such that it is the
6072 * first block of the current_ext. @ext is a target extent to be split.
6073 * @split_fsb is a block where the extents is split. If split_fsb lies in a
6074 * hole or the first block of extents, just return 0.
6075 */
6076 int
xfs_bmap_split_extent(struct xfs_trans * tp,struct xfs_inode * ip,xfs_fileoff_t split_fsb)6077 xfs_bmap_split_extent(
6078 struct xfs_trans *tp,
6079 struct xfs_inode *ip,
6080 xfs_fileoff_t split_fsb)
6081 {
6082 int whichfork = XFS_DATA_FORK;
6083 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
6084 struct xfs_btree_cur *cur = NULL;
6085 struct xfs_bmbt_irec got;
6086 struct xfs_bmbt_irec new; /* split extent */
6087 struct xfs_mount *mp = ip->i_mount;
6088 xfs_fsblock_t gotblkcnt; /* new block count for got */
6089 struct xfs_iext_cursor icur;
6090 int error = 0;
6091 int logflags = 0;
6092 int i = 0;
6093
6094 if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
6095 XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
6096 xfs_bmap_mark_sick(ip, whichfork);
6097 return -EFSCORRUPTED;
6098 }
6099
6100 if (xfs_is_shutdown(mp))
6101 return -EIO;
6102
6103 /* Read in all the extents */
6104 error = xfs_iread_extents(tp, ip, whichfork);
6105 if (error)
6106 return error;
6107
6108 /*
6109 * If there are not extents, or split_fsb lies in a hole we are done.
6110 */
6111 if (!xfs_iext_lookup_extent(ip, ifp, split_fsb, &icur, &got) ||
6112 got.br_startoff >= split_fsb)
6113 return 0;
6114
6115 gotblkcnt = split_fsb - got.br_startoff;
6116 new.br_startoff = split_fsb;
6117 new.br_startblock = got.br_startblock + gotblkcnt;
6118 new.br_blockcount = got.br_blockcount - gotblkcnt;
6119 new.br_state = got.br_state;
6120
6121 if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
6122 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
6123 error = xfs_bmbt_lookup_eq(cur, &got, &i);
6124 if (error)
6125 goto del_cursor;
6126 if (XFS_IS_CORRUPT(mp, i != 1)) {
6127 xfs_btree_mark_sick(cur);
6128 error = -EFSCORRUPTED;
6129 goto del_cursor;
6130 }
6131 }
6132
6133 got.br_blockcount = gotblkcnt;
6134 xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), &icur,
6135 &got);
6136
6137 logflags = XFS_ILOG_CORE;
6138 if (cur) {
6139 error = xfs_bmbt_update(cur, &got);
6140 if (error)
6141 goto del_cursor;
6142 } else
6143 logflags |= XFS_ILOG_DEXT;
6144
6145 /* Add new extent */
6146 xfs_iext_next(ifp, &icur);
6147 xfs_iext_insert(ip, &icur, &new, 0);
6148 ifp->if_nextents++;
6149
6150 if (cur) {
6151 error = xfs_bmbt_lookup_eq(cur, &new, &i);
6152 if (error)
6153 goto del_cursor;
6154 if (XFS_IS_CORRUPT(mp, i != 0)) {
6155 xfs_btree_mark_sick(cur);
6156 error = -EFSCORRUPTED;
6157 goto del_cursor;
6158 }
6159 error = xfs_btree_insert(cur, &i);
6160 if (error)
6161 goto del_cursor;
6162 if (XFS_IS_CORRUPT(mp, i != 1)) {
6163 xfs_btree_mark_sick(cur);
6164 error = -EFSCORRUPTED;
6165 goto del_cursor;
6166 }
6167 }
6168
6169 /*
6170 * Convert to a btree if necessary.
6171 */
6172 if (xfs_bmap_needs_btree(ip, whichfork)) {
6173 int tmp_logflags; /* partial log flag return val */
6174
6175 ASSERT(cur == NULL);
6176 error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
6177 &tmp_logflags, whichfork);
6178 logflags |= tmp_logflags;
6179 }
6180
6181 del_cursor:
6182 if (cur) {
6183 cur->bc_bmap.allocated = 0;
6184 xfs_btree_del_cursor(cur, error);
6185 }
6186
6187 if (logflags)
6188 xfs_trans_log_inode(tp, ip, logflags);
6189 return error;
6190 }
6191
6192 /* Record a bmap intent. */
6193 static inline void
__xfs_bmap_add(struct xfs_trans * tp,enum xfs_bmap_intent_type type,struct xfs_inode * ip,int whichfork,struct xfs_bmbt_irec * bmap)6194 __xfs_bmap_add(
6195 struct xfs_trans *tp,
6196 enum xfs_bmap_intent_type type,
6197 struct xfs_inode *ip,
6198 int whichfork,
6199 struct xfs_bmbt_irec *bmap)
6200 {
6201 struct xfs_bmap_intent *bi;
6202
6203 if ((whichfork != XFS_DATA_FORK && whichfork != XFS_ATTR_FORK) ||
6204 bmap->br_startblock == HOLESTARTBLOCK ||
6205 bmap->br_startblock == DELAYSTARTBLOCK)
6206 return;
6207
6208 bi = kmem_cache_alloc(xfs_bmap_intent_cache, GFP_KERNEL | __GFP_NOFAIL);
6209 INIT_LIST_HEAD(&bi->bi_list);
6210 bi->bi_type = type;
6211 bi->bi_owner = ip;
6212 bi->bi_whichfork = whichfork;
6213 bi->bi_bmap = *bmap;
6214
6215 xfs_bmap_defer_add(tp, bi);
6216 }
6217
6218 /* Map an extent into a file. */
6219 void
xfs_bmap_map_extent(struct xfs_trans * tp,struct xfs_inode * ip,int whichfork,struct xfs_bmbt_irec * PREV)6220 xfs_bmap_map_extent(
6221 struct xfs_trans *tp,
6222 struct xfs_inode *ip,
6223 int whichfork,
6224 struct xfs_bmbt_irec *PREV)
6225 {
6226 __xfs_bmap_add(tp, XFS_BMAP_MAP, ip, whichfork, PREV);
6227 }
6228
6229 /* Unmap an extent out of a file. */
6230 void
xfs_bmap_unmap_extent(struct xfs_trans * tp,struct xfs_inode * ip,int whichfork,struct xfs_bmbt_irec * PREV)6231 xfs_bmap_unmap_extent(
6232 struct xfs_trans *tp,
6233 struct xfs_inode *ip,
6234 int whichfork,
6235 struct xfs_bmbt_irec *PREV)
6236 {
6237 __xfs_bmap_add(tp, XFS_BMAP_UNMAP, ip, whichfork, PREV);
6238 }
6239
6240 /*
6241 * Process one of the deferred bmap operations. We pass back the
6242 * btree cursor to maintain our lock on the bmapbt between calls.
6243 */
6244 int
xfs_bmap_finish_one(struct xfs_trans * tp,struct xfs_bmap_intent * bi)6245 xfs_bmap_finish_one(
6246 struct xfs_trans *tp,
6247 struct xfs_bmap_intent *bi)
6248 {
6249 struct xfs_bmbt_irec *bmap = &bi->bi_bmap;
6250 int error = 0;
6251 int flags = 0;
6252
6253 if (bi->bi_whichfork == XFS_ATTR_FORK)
6254 flags |= XFS_BMAPI_ATTRFORK;
6255
6256 ASSERT(tp->t_highest_agno == NULLAGNUMBER);
6257
6258 trace_xfs_bmap_deferred(bi);
6259
6260 if (XFS_TEST_ERROR(false, tp->t_mountp, XFS_ERRTAG_BMAP_FINISH_ONE))
6261 return -EIO;
6262
6263 switch (bi->bi_type) {
6264 case XFS_BMAP_MAP:
6265 if (bi->bi_bmap.br_state == XFS_EXT_UNWRITTEN)
6266 flags |= XFS_BMAPI_PREALLOC;
6267 error = xfs_bmapi_remap(tp, bi->bi_owner, bmap->br_startoff,
6268 bmap->br_blockcount, bmap->br_startblock,
6269 flags);
6270 bmap->br_blockcount = 0;
6271 break;
6272 case XFS_BMAP_UNMAP:
6273 error = __xfs_bunmapi(tp, bi->bi_owner, bmap->br_startoff,
6274 &bmap->br_blockcount, flags | XFS_BMAPI_REMAP,
6275 1);
6276 break;
6277 default:
6278 ASSERT(0);
6279 xfs_bmap_mark_sick(bi->bi_owner, bi->bi_whichfork);
6280 error = -EFSCORRUPTED;
6281 }
6282
6283 return error;
6284 }
6285
6286 /* Check that an extent does not have invalid flags or bad ranges. */
6287 xfs_failaddr_t
xfs_bmap_validate_extent_raw(struct xfs_mount * mp,bool rtfile,int whichfork,struct xfs_bmbt_irec * irec)6288 xfs_bmap_validate_extent_raw(
6289 struct xfs_mount *mp,
6290 bool rtfile,
6291 int whichfork,
6292 struct xfs_bmbt_irec *irec)
6293 {
6294 if (!xfs_verify_fileext(mp, irec->br_startoff, irec->br_blockcount))
6295 return __this_address;
6296
6297 if (rtfile && whichfork == XFS_DATA_FORK) {
6298 if (!xfs_verify_rtbext(mp, irec->br_startblock,
6299 irec->br_blockcount))
6300 return __this_address;
6301 } else {
6302 if (!xfs_verify_fsbext(mp, irec->br_startblock,
6303 irec->br_blockcount))
6304 return __this_address;
6305 }
6306 if (irec->br_state != XFS_EXT_NORM && whichfork != XFS_DATA_FORK)
6307 return __this_address;
6308 return NULL;
6309 }
6310
6311 int __init
xfs_bmap_intent_init_cache(void)6312 xfs_bmap_intent_init_cache(void)
6313 {
6314 xfs_bmap_intent_cache = kmem_cache_create("xfs_bmap_intent",
6315 sizeof(struct xfs_bmap_intent),
6316 0, 0, NULL);
6317
6318 return xfs_bmap_intent_cache != NULL ? 0 : -ENOMEM;
6319 }
6320
6321 void
xfs_bmap_intent_destroy_cache(void)6322 xfs_bmap_intent_destroy_cache(void)
6323 {
6324 kmem_cache_destroy(xfs_bmap_intent_cache);
6325 xfs_bmap_intent_cache = NULL;
6326 }
6327
6328 /* Check that an inode's extent does not have invalid flags or bad ranges. */
6329 xfs_failaddr_t
xfs_bmap_validate_extent(struct xfs_inode * ip,int whichfork,struct xfs_bmbt_irec * irec)6330 xfs_bmap_validate_extent(
6331 struct xfs_inode *ip,
6332 int whichfork,
6333 struct xfs_bmbt_irec *irec)
6334 {
6335 return xfs_bmap_validate_extent_raw(ip->i_mount,
6336 XFS_IS_REALTIME_INODE(ip), whichfork, irec);
6337 }
6338
6339 /*
6340 * Used in xfs_itruncate_extents(). This is the maximum number of extents
6341 * freed from a file in a single transaction.
6342 */
6343 #define XFS_ITRUNC_MAX_EXTENTS 2
6344
6345 /*
6346 * Unmap every extent in part of an inode's fork. We don't do any higher level
6347 * invalidation work at all.
6348 */
6349 int
xfs_bunmapi_range(struct xfs_trans ** tpp,struct xfs_inode * ip,uint32_t flags,xfs_fileoff_t startoff,xfs_fileoff_t endoff)6350 xfs_bunmapi_range(
6351 struct xfs_trans **tpp,
6352 struct xfs_inode *ip,
6353 uint32_t flags,
6354 xfs_fileoff_t startoff,
6355 xfs_fileoff_t endoff)
6356 {
6357 xfs_filblks_t unmap_len = endoff - startoff + 1;
6358 int error = 0;
6359
6360 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
6361
6362 while (unmap_len > 0) {
6363 ASSERT((*tpp)->t_highest_agno == NULLAGNUMBER);
6364 error = __xfs_bunmapi(*tpp, ip, startoff, &unmap_len, flags,
6365 XFS_ITRUNC_MAX_EXTENTS);
6366 if (error)
6367 goto out;
6368
6369 /* free the just unmapped extents */
6370 error = xfs_defer_finish(tpp);
6371 if (error)
6372 goto out;
6373 cond_resched();
6374 }
6375 out:
6376 return error;
6377 }
6378
6379 struct xfs_bmap_query_range {
6380 xfs_bmap_query_range_fn fn;
6381 void *priv;
6382 };
6383
6384 /* Format btree record and pass to our callback. */
6385 STATIC int
xfs_bmap_query_range_helper(struct xfs_btree_cur * cur,const union xfs_btree_rec * rec,void * priv)6386 xfs_bmap_query_range_helper(
6387 struct xfs_btree_cur *cur,
6388 const union xfs_btree_rec *rec,
6389 void *priv)
6390 {
6391 struct xfs_bmap_query_range *query = priv;
6392 struct xfs_bmbt_irec irec;
6393 xfs_failaddr_t fa;
6394
6395 xfs_bmbt_disk_get_all(&rec->bmbt, &irec);
6396 fa = xfs_bmap_validate_extent(cur->bc_ino.ip, cur->bc_ino.whichfork,
6397 &irec);
6398 if (fa) {
6399 xfs_btree_mark_sick(cur);
6400 return xfs_bmap_complain_bad_rec(cur->bc_ino.ip,
6401 cur->bc_ino.whichfork, fa, &irec);
6402 }
6403
6404 return query->fn(cur, &irec, query->priv);
6405 }
6406
6407 /* Find all bmaps. */
6408 int
xfs_bmap_query_all(struct xfs_btree_cur * cur,xfs_bmap_query_range_fn fn,void * priv)6409 xfs_bmap_query_all(
6410 struct xfs_btree_cur *cur,
6411 xfs_bmap_query_range_fn fn,
6412 void *priv)
6413 {
6414 struct xfs_bmap_query_range query = {
6415 .priv = priv,
6416 .fn = fn,
6417 };
6418
6419 return xfs_btree_query_all(cur, xfs_bmap_query_range_helper, &query);
6420 }
6421
6422 /* Helper function to extract extent size hint from inode */
6423 xfs_extlen_t
xfs_get_extsz_hint(struct xfs_inode * ip)6424 xfs_get_extsz_hint(
6425 struct xfs_inode *ip)
6426 {
6427 /*
6428 * No point in aligning allocations if we need to COW to actually
6429 * write to them.
6430 */
6431 if (xfs_is_always_cow_inode(ip))
6432 return 0;
6433 if ((ip->i_diflags & XFS_DIFLAG_EXTSIZE) && ip->i_extsize)
6434 return ip->i_extsize;
6435 if (XFS_IS_REALTIME_INODE(ip) &&
6436 ip->i_mount->m_sb.sb_rextsize > 1)
6437 return ip->i_mount->m_sb.sb_rextsize;
6438 return 0;
6439 }
6440
6441 /*
6442 * Helper function to extract CoW extent size hint from inode.
6443 * Between the extent size hint and the CoW extent size hint, we
6444 * return the greater of the two. If the value is zero (automatic),
6445 * use the default size.
6446 */
6447 xfs_extlen_t
xfs_get_cowextsz_hint(struct xfs_inode * ip)6448 xfs_get_cowextsz_hint(
6449 struct xfs_inode *ip)
6450 {
6451 xfs_extlen_t a, b;
6452
6453 a = 0;
6454 if (ip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE)
6455 a = ip->i_cowextsize;
6456 b = xfs_get_extsz_hint(ip);
6457
6458 a = max(a, b);
6459 if (a == 0)
6460 return XFS_DEFAULT_COWEXTSZ_HINT;
6461 return a;
6462 }
6463