Line data Source code
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_rtalloc.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_symlink_remote.h"
41 :
42 : struct kmem_cache *xfs_bmap_intent_cache;
43 :
44 : /*
45 : * Miscellaneous helper functions
46 : */
47 :
48 : /*
49 : * Compute and fill in the value of the maximum depth of a bmap btree
50 : * in this filesystem. Done once, during mount.
51 : */
52 : void
53 48238 : xfs_bmap_compute_maxlevels(
54 : xfs_mount_t *mp, /* file system mount structure */
55 : int whichfork) /* data or attr fork */
56 : {
57 48238 : uint64_t maxblocks; /* max blocks at this level */
58 48238 : xfs_extnum_t maxleafents; /* max leaf entries possible */
59 48238 : int level; /* btree level */
60 48238 : int maxrootrecs; /* max records in root block */
61 48238 : int minleafrecs; /* min records in leaf block */
62 48238 : int minnoderecs; /* min records in node block */
63 48238 : int sz; /* root block size */
64 :
65 : /*
66 : * The maximum number of extents in a fork, hence the maximum number of
67 : * leaf entries, is controlled by the size of the on-disk extent count.
68 : *
69 : * Note that we can no longer assume that if we are in ATTR1 that the
70 : * fork offset of all the inodes will be
71 : * (xfs_default_attroffset(ip) >> 3) because we could have mounted with
72 : * ATTR2 and then mounted back with ATTR1, keeping the i_forkoff's fixed
73 : * but probably at various positions. Therefore, for both ATTR1 and
74 : * ATTR2 we have to assume the worst case scenario of a minimum size
75 : * available.
76 : */
77 48238 : maxleafents = xfs_iext_max_nextents(xfs_has_large_extent_counts(mp),
78 : whichfork);
79 48238 : if (whichfork == XFS_DATA_FORK)
80 : sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
81 : else
82 24119 : sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
83 :
84 48238 : maxrootrecs = xfs_bmdr_maxrecs(sz, 0);
85 48238 : minleafrecs = mp->m_bmap_dmnr[0];
86 48238 : minnoderecs = mp->m_bmap_dmnr[1];
87 48238 : maxblocks = howmany_64(maxleafents, minleafrecs);
88 288990 : for (level = 1; maxblocks > 1; level++) {
89 240752 : if (maxblocks <= maxrootrecs)
90 : maxblocks = 1;
91 : else
92 231252 : maxblocks = howmany_64(maxblocks, minnoderecs);
93 : }
94 48238 : mp->m_bm_maxlevels[whichfork] = level;
95 48238 : ASSERT(mp->m_bm_maxlevels[whichfork] <= xfs_bmbt_maxlevels_ondisk());
96 48238 : }
97 :
98 : unsigned int
99 24119 : xfs_bmap_compute_attr_offset(
100 : struct xfs_mount *mp)
101 : {
102 24119 : if (mp->m_sb.sb_inodesize == 256)
103 80 : return XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS);
104 : return XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
105 : }
106 :
107 : STATIC int /* error */
108 134524845 : xfs_bmbt_lookup_eq(
109 : struct xfs_btree_cur *cur,
110 : struct xfs_bmbt_irec *irec,
111 : int *stat) /* success/failure */
112 : {
113 134524845 : cur->bc_rec.b = *irec;
114 134524845 : return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
115 : }
116 :
117 : STATIC int /* error */
118 0 : xfs_bmbt_lookup_first(
119 : struct xfs_btree_cur *cur,
120 : int *stat) /* success/failure */
121 : {
122 0 : cur->bc_rec.b.br_startoff = 0;
123 0 : cur->bc_rec.b.br_startblock = 0;
124 0 : cur->bc_rec.b.br_blockcount = 0;
125 0 : return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
126 : }
127 :
128 : /*
129 : * Check if the inode needs to be converted to btree format.
130 : */
131 222041515 : static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
132 : {
133 222041515 : struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
134 :
135 442084487 : return whichfork != XFS_COW_FORK &&
136 222040679 : ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
137 124347752 : ifp->if_nextents > XFS_IFORK_MAXEXT(ip, whichfork);
138 : }
139 :
140 : /*
141 : * Check if the inode should be converted to extent format.
142 : */
143 222751944 : static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
144 : {
145 222751944 : struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
146 :
147 445168935 : return whichfork != XFS_COW_FORK &&
148 222752812 : ifp->if_format == XFS_DINODE_FMT_BTREE &&
149 96900427 : ifp->if_nextents <= XFS_IFORK_MAXEXT(ip, whichfork);
150 : }
151 :
152 : /*
153 : * Update the record referred to by cur to the value given by irec
154 : * This either works (return 0) or gets an EFSCORRUPTED error.
155 : */
156 : STATIC int
157 33315799 : xfs_bmbt_update(
158 : struct xfs_btree_cur *cur,
159 : struct xfs_bmbt_irec *irec)
160 : {
161 33315799 : union xfs_btree_rec rec;
162 :
163 33315799 : xfs_bmbt_disk_set_all(&rec.bmbt, irec);
164 33315765 : return xfs_btree_update(cur, &rec);
165 : }
166 :
167 : /*
168 : * Compute the worst-case number of indirect blocks that will be used
169 : * for ip's delayed extent of length "len".
170 : */
171 : STATIC xfs_filblks_t
172 21581822 : xfs_bmap_worst_indlen(
173 : xfs_inode_t *ip, /* incore inode pointer */
174 : xfs_filblks_t len) /* delayed extent length */
175 : {
176 21581822 : int level; /* btree level number */
177 21581822 : int maxrecs; /* maximum record count at this level */
178 21581822 : xfs_mount_t *mp; /* mount structure */
179 21581822 : xfs_filblks_t rval; /* return value */
180 :
181 21581822 : mp = ip->i_mount;
182 21581822 : maxrecs = mp->m_bmap_dmxr[0];
183 21581822 : for (level = 0, rval = 0;
184 22953556 : level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
185 1371734 : level++) {
186 22953662 : len += maxrecs - 1;
187 22953662 : do_div(len, maxrecs);
188 22953662 : rval += len;
189 22953662 : if (len == 1)
190 21581928 : return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
191 21581928 : level - 1;
192 1371734 : if (level == 0)
193 1155942 : maxrecs = mp->m_bmap_dmxr[1];
194 : }
195 : return rval;
196 : }
197 :
198 : /*
199 : * Calculate the default attribute fork offset for newly created inodes.
200 : */
201 : uint
202 61494997 : xfs_default_attroffset(
203 : struct xfs_inode *ip)
204 : {
205 61494997 : if (ip->i_df.if_format == XFS_DINODE_FMT_DEV)
206 : return roundup(sizeof(xfs_dev_t), 8);
207 53973115 : return M_IGEO(ip->i_mount)->attr_fork_offset;
208 : }
209 :
210 : /*
211 : * Helper routine to reset inode i_forkoff field when switching attribute fork
212 : * from local to extent format - we reset it where possible to make space
213 : * available for inline data fork extents.
214 : */
215 : STATIC void
216 4575688 : xfs_bmap_forkoff_reset(
217 : xfs_inode_t *ip,
218 : int whichfork)
219 : {
220 4575688 : if (whichfork == XFS_ATTR_FORK &&
221 4263445 : ip->i_df.if_format != XFS_DINODE_FMT_DEV &&
222 : ip->i_df.if_format != XFS_DINODE_FMT_BTREE) {
223 4194251 : uint dfl_forkoff = xfs_default_attroffset(ip) >> 3;
224 :
225 4194251 : if (dfl_forkoff > ip->i_forkoff)
226 4177122 : ip->i_forkoff = dfl_forkoff;
227 : }
228 4575688 : }
229 :
230 : #ifdef DEBUG
231 : STATIC struct xfs_buf *
232 451357795 : xfs_bmap_get_bp(
233 : struct xfs_btree_cur *cur,
234 : xfs_fsblock_t bno)
235 : {
236 451357795 : struct xfs_log_item *lip;
237 451357795 : int i;
238 :
239 451357795 : if (!cur)
240 : return NULL;
241 :
242 1230384791 : for (i = 0; i < cur->bc_maxlevels; i++) {
243 1230384600 : if (!cur->bc_levels[i].bp)
244 : break;
245 828250135 : if (xfs_buf_daddr(cur->bc_levels[i].bp) == bno)
246 49223139 : return cur->bc_levels[i].bp;
247 : }
248 :
249 : /* Chase down all the log items to see if the bp is there */
250 2235621287 : list_for_each_entry(lip, &cur->bc_tp->t_items, li_trans) {
251 1845045178 : struct xfs_buf_log_item *bip = (struct xfs_buf_log_item *)lip;
252 :
253 1845045178 : if (bip->bli_item.li_type == XFS_LI_BUF &&
254 1067027401 : xfs_buf_daddr(bip->bli_buf) == bno)
255 11558547 : return bip->bli_buf;
256 : }
257 :
258 : return NULL;
259 : }
260 :
261 : STATIC void
262 50024793 : xfs_check_block(
263 : struct xfs_btree_block *block,
264 : xfs_mount_t *mp,
265 : int root,
266 : short sz)
267 : {
268 50024793 : int i, j, dmxr;
269 50024793 : __be64 *pp, *thispa; /* pointer to block address */
270 50024793 : xfs_bmbt_key_t *prevp, *keyp;
271 :
272 50024793 : ASSERT(be16_to_cpu(block->bb_level) > 0);
273 :
274 : prevp = NULL;
275 1002763738 : for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
276 451357130 : dmxr = mp->m_bmap_dmxr[0];
277 451357130 : keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
278 :
279 451357130 : if (prevp) {
280 401333052 : ASSERT(be64_to_cpu(prevp->br_startoff) <
281 : be64_to_cpu(keyp->br_startoff));
282 : }
283 451357130 : prevp = keyp;
284 :
285 : /*
286 : * Compare the block numbers to see if there are dups.
287 : */
288 451357130 : if (root)
289 73635398 : pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
290 : else
291 377721732 : pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
292 :
293 5772904120 : for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
294 5321547044 : if (root)
295 162001672 : thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
296 : else
297 5159545372 : thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
298 5321547167 : if (*thispa == *pp) {
299 0 : xfs_warn(mp, "%s: thispa(%d) == pp(%d) %lld",
300 : __func__, j, i,
301 : (unsigned long long)be64_to_cpu(*thispa));
302 0 : xfs_err(mp, "%s: ptrs are equal in node\n",
303 : __func__);
304 0 : xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
305 : }
306 : }
307 : }
308 50024739 : }
309 :
310 : /*
311 : * Check that the extents for the inode ip are in the right order in all
312 : * btree leaves. THis becomes prohibitively expensive for large extent count
313 : * files, so don't bother with inodes that have more than 10,000 extents in
314 : * them. The btree record ordering checks will still be done, so for such large
315 : * bmapbt constructs that is going to catch most corruptions.
316 : */
317 : STATIC void
318 144090399 : xfs_bmap_check_leaf_extents(
319 : struct xfs_btree_cur *cur, /* btree cursor or null */
320 : xfs_inode_t *ip, /* incore inode pointer */
321 : int whichfork) /* data or attr fork */
322 : {
323 144090399 : struct xfs_mount *mp = ip->i_mount;
324 144090399 : struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
325 144088127 : struct xfs_btree_block *block; /* current btree block */
326 144088127 : xfs_fsblock_t bno; /* block # of "block" */
327 144088127 : struct xfs_buf *bp; /* buffer for "block" */
328 144088127 : int error; /* error return value */
329 144088127 : xfs_extnum_t i=0, j; /* index into the extents list */
330 144088127 : int level; /* btree level, for checking */
331 144088127 : __be64 *pp; /* pointer to block address */
332 144088127 : xfs_bmbt_rec_t *ep; /* pointer to current extent */
333 144088127 : xfs_bmbt_rec_t last = {0, 0}; /* last extent in prev block */
334 144088127 : xfs_bmbt_rec_t *nextp; /* pointer to next extent */
335 144088127 : int bp_release = 0;
336 :
337 144088127 : if (ifp->if_format != XFS_DINODE_FMT_BTREE)
338 : return;
339 :
340 : /* skip large extent count inodes */
341 66364219 : if (ip->i_df.if_nextents > 10000)
342 : return;
343 :
344 35192750 : bno = NULLFSBLOCK;
345 35192750 : block = ifp->if_broot;
346 : /*
347 : * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
348 : */
349 35192750 : level = be16_to_cpu(block->bb_level);
350 35192750 : ASSERT(level > 0);
351 35192750 : xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
352 35192697 : pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
353 35192606 : bno = be64_to_cpu(*pp);
354 :
355 35192606 : ASSERT(bno != NULLFSBLOCK);
356 35192606 : ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
357 35192606 : ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
358 :
359 : /*
360 : * Go down the tree until leaf level is reached, following the first
361 : * pointer (leftmost) at each level.
362 : */
363 50024360 : while (level-- > 0) {
364 : /* See if buf is in cur first */
365 50024220 : bp_release = 0;
366 50024220 : bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
367 50024693 : if (!bp) {
368 21344150 : bp_release = 1;
369 21344150 : error = xfs_btree_read_bufl(mp, NULL, bno, &bp,
370 : XFS_BMAP_BTREE_REF,
371 : &xfs_bmbt_buf_ops);
372 21344158 : if (xfs_metadata_is_sick(error))
373 0 : xfs_btree_mark_sick(cur);
374 21344158 : if (error)
375 5 : goto error_norelse;
376 : }
377 50024696 : block = XFS_BUF_TO_BLOCK(bp);
378 50024696 : if (level == 0)
379 : break;
380 :
381 : /*
382 : * Check this block for basic sanity (increasing keys and
383 : * no duplicate blocks).
384 : */
385 :
386 14831822 : xfs_check_block(block, mp, 0, 0);
387 14831817 : pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
388 14831817 : bno = be64_to_cpu(*pp);
389 14831817 : if (XFS_IS_CORRUPT(mp, !xfs_verify_fsbno(mp, bno))) {
390 0 : xfs_btree_mark_sick(cur);
391 0 : error = -EFSCORRUPTED;
392 0 : goto error0;
393 : }
394 14831754 : if (bp_release) {
395 0 : bp_release = 0;
396 0 : xfs_trans_brelse(NULL, bp);
397 : }
398 : }
399 :
400 : /*
401 : * Here with bp and block set to the leftmost leaf node in the tree.
402 : */
403 : i = 0;
404 :
405 : /*
406 : * Loop over all leaf nodes checking that all extents are in the right order.
407 : */
408 837856540 : for (;;) {
409 436524777 : xfs_fsblock_t nextbno;
410 436524777 : xfs_extnum_t num_recs;
411 :
412 :
413 436524777 : num_recs = xfs_btree_get_numrecs(block);
414 :
415 : /*
416 : * Read-ahead the next leaf block, if any.
417 : */
418 :
419 436524777 : nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
420 :
421 : /*
422 : * Check all the extents to make sure they are OK.
423 : * If we had a previous block, the last entry should
424 : * conform with the first entry in this one.
425 : */
426 :
427 436524777 : ep = XFS_BMBT_REC_ADDR(mp, block, 1);
428 436524777 : if (i) {
429 401332413 : ASSERT(xfs_bmbt_disk_get_startoff(&last) +
430 : xfs_bmbt_disk_get_blockcount(&last) <=
431 : xfs_bmbt_disk_get_startoff(ep));
432 : }
433 >10349*10^7 : for (j = 1; j < num_recs; j++) {
434 >10306*10^7 : nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
435 >10306*10^7 : ASSERT(xfs_bmbt_disk_get_startoff(ep) +
436 : xfs_bmbt_disk_get_blockcount(ep) <=
437 : xfs_bmbt_disk_get_startoff(nextp));
438 >10306*10^7 : ep = nextp;
439 : }
440 :
441 436525213 : last = *ep;
442 436525213 : i += num_recs;
443 436525213 : if (bp_release) {
444 390576155 : bp_release = 0;
445 390576155 : xfs_trans_brelse(NULL, bp);
446 : }
447 436526161 : bno = nextbno;
448 : /*
449 : * If we've reached the end, stop.
450 : */
451 436526161 : if (bno == NULLFSBLOCK)
452 : break;
453 :
454 401333168 : bp_release = 0;
455 401333168 : bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
456 401331354 : if (!bp) {
457 369232422 : bp_release = 1;
458 369232422 : error = xfs_btree_read_bufl(mp, NULL, bno, &bp,
459 : XFS_BMAP_BTREE_REF,
460 : &xfs_bmbt_buf_ops);
461 369232886 : if (xfs_metadata_is_sick(error))
462 0 : xfs_btree_mark_sick(cur);
463 369232886 : if (error)
464 42 : goto error_norelse;
465 : }
466 401331776 : block = XFS_BUF_TO_BLOCK(bp);
467 : }
468 :
469 : return;
470 :
471 : error0:
472 0 : xfs_warn(mp, "%s: at error0", __func__);
473 0 : if (bp_release)
474 0 : xfs_trans_brelse(NULL, bp);
475 0 : error_norelse:
476 47 : xfs_warn(mp, "%s: BAD after btree leaves for %llu extents",
477 : __func__, i);
478 47 : xfs_err(mp, "%s: CORRUPTED BTREE OR SOMETHING", __func__);
479 47 : xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
480 47 : return;
481 : }
482 :
483 : /*
484 : * Validate that the bmbt_irecs being returned from bmapi are valid
485 : * given the caller's original parameters. Specifically check the
486 : * ranges of the returned irecs to ensure that they only extend beyond
487 : * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
488 : */
489 : STATIC void
490 76199331 : xfs_bmap_validate_ret(
491 : xfs_fileoff_t bno,
492 : xfs_filblks_t len,
493 : uint32_t flags,
494 : xfs_bmbt_irec_t *mval,
495 : int nmap,
496 : int ret_nmap)
497 : {
498 76199331 : int i; /* index to map values */
499 :
500 76199331 : ASSERT(ret_nmap <= nmap);
501 :
502 152398202 : for (i = 0; i < ret_nmap; i++) {
503 76198860 : ASSERT(mval[i].br_blockcount > 0);
504 76198860 : if (!(flags & XFS_BMAPI_ENTIRE)) {
505 76198860 : ASSERT(mval[i].br_startoff >= bno);
506 76198860 : ASSERT(mval[i].br_blockcount <= len);
507 76198860 : ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
508 : bno + len);
509 : } else {
510 0 : ASSERT(mval[i].br_startoff < bno + len);
511 0 : ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
512 : bno);
513 : }
514 76198860 : ASSERT(i == 0 ||
515 : mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
516 : mval[i].br_startoff);
517 76198860 : ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
518 : mval[i].br_startblock != HOLESTARTBLOCK);
519 76198860 : ASSERT(mval[i].br_state == XFS_EXT_NORM ||
520 : mval[i].br_state == XFS_EXT_UNWRITTEN);
521 : }
522 76199342 : }
523 :
524 : #else
525 : #define xfs_bmap_check_leaf_extents(cur, ip, whichfork) do { } while (0)
526 : #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap) do { } while (0)
527 : #endif /* DEBUG */
528 :
529 : /*
530 : * Inode fork format manipulation functions
531 : */
532 :
533 : /*
534 : * Convert the inode format to extent format if it currently is in btree format,
535 : * but the extent list is small enough that it fits into the extent format.
536 : *
537 : * Since the extents are already in-core, all we have to do is give up the space
538 : * for the btree root and pitch the leaf block.
539 : */
540 : STATIC int /* error */
541 222739390 : xfs_bmap_btree_to_extents(
542 : struct xfs_trans *tp, /* transaction pointer */
543 : struct xfs_inode *ip, /* incore inode pointer */
544 : struct xfs_btree_cur *cur, /* btree cursor */
545 : int *logflagsp, /* inode logging flags */
546 : int whichfork) /* data or attr fork */
547 : {
548 222739390 : struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
549 222755499 : struct xfs_mount *mp = ip->i_mount;
550 222755499 : struct xfs_btree_block *rblock = ifp->if_broot;
551 222755499 : struct xfs_btree_block *cblock;/* child btree block */
552 222755499 : xfs_fsblock_t cbno; /* child block number */
553 222755499 : struct xfs_buf *cbp; /* child block's buffer */
554 222755499 : int error; /* error return value */
555 222755499 : __be64 *pp; /* ptr to block address */
556 222755499 : struct xfs_owner_info oinfo;
557 :
558 : /* check if we actually need the extent format first: */
559 222755499 : if (!xfs_bmap_wants_extents(ip, whichfork))
560 : return 0;
561 :
562 446626 : ASSERT(cur);
563 446626 : ASSERT(whichfork != XFS_COW_FORK);
564 446626 : ASSERT(ifp->if_format == XFS_DINODE_FMT_BTREE);
565 446626 : ASSERT(be16_to_cpu(rblock->bb_level) == 1);
566 446626 : ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
567 446626 : ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
568 :
569 446626 : pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
570 446626 : cbno = be64_to_cpu(*pp);
571 : #ifdef DEBUG
572 446626 : if (XFS_IS_CORRUPT(cur->bc_mp, !xfs_btree_check_lptr(cur, cbno, 1))) {
573 0 : xfs_btree_mark_sick(cur);
574 0 : return -EFSCORRUPTED;
575 : }
576 : #endif
577 446626 : error = xfs_btree_read_bufl(mp, tp, cbno, &cbp, XFS_BMAP_BTREE_REF,
578 : &xfs_bmbt_buf_ops);
579 446626 : if (xfs_metadata_is_sick(error))
580 0 : xfs_btree_mark_sick(cur);
581 446626 : if (error)
582 : return error;
583 446626 : cblock = XFS_BUF_TO_BLOCK(cbp);
584 446626 : if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
585 : return error;
586 :
587 446626 : xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork);
588 446626 : error = xfs_free_extent_later(cur->bc_tp, cbno, 1, &oinfo,
589 : XFS_AG_RESV_NONE);
590 446626 : if (error)
591 : return error;
592 :
593 446626 : ip->i_nblocks--;
594 446626 : xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
595 446626 : xfs_trans_binval(tp, cbp);
596 446626 : if (cur->bc_levels[0].bp == cbp)
597 446626 : cur->bc_levels[0].bp = NULL;
598 446626 : xfs_iroot_realloc(ip, -1, whichfork);
599 446626 : ASSERT(ifp->if_broot == NULL);
600 446626 : ifp->if_format = XFS_DINODE_FMT_EXTENTS;
601 446626 : *logflagsp |= XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
602 446626 : return 0;
603 : }
604 :
605 : /*
606 : * Convert an extents-format file into a btree-format file.
607 : * The new file will have a root block (in the inode) and a single child block.
608 : */
609 : STATIC int /* error */
610 858297 : xfs_bmap_extents_to_btree(
611 : struct xfs_trans *tp, /* transaction pointer */
612 : struct xfs_inode *ip, /* incore inode pointer */
613 : struct xfs_btree_cur **curp, /* cursor returned to caller */
614 : int wasdel, /* converting a delayed alloc */
615 : int *logflagsp, /* inode logging flags */
616 : int whichfork) /* data or attr fork */
617 : {
618 858297 : struct xfs_btree_block *ablock; /* allocated (child) bt block */
619 858297 : struct xfs_buf *abp; /* buffer for ablock */
620 858297 : struct xfs_alloc_arg args; /* allocation arguments */
621 858297 : struct xfs_bmbt_rec *arp; /* child record pointer */
622 858297 : struct xfs_btree_block *block; /* btree root block */
623 858297 : struct xfs_btree_cur *cur; /* bmap btree cursor */
624 858297 : int error; /* error return value */
625 858297 : struct xfs_ifork *ifp; /* inode fork pointer */
626 858297 : struct xfs_bmbt_key *kp; /* root block key pointer */
627 858297 : struct xfs_mount *mp; /* mount structure */
628 858297 : xfs_bmbt_ptr_t *pp; /* root block address pointer */
629 858297 : struct xfs_iext_cursor icur;
630 858297 : struct xfs_bmbt_irec rec;
631 858297 : xfs_extnum_t cnt = 0;
632 :
633 858297 : mp = ip->i_mount;
634 858297 : ASSERT(whichfork != XFS_COW_FORK);
635 858297 : ifp = xfs_ifork_ptr(ip, whichfork);
636 858297 : ASSERT(ifp->if_format == XFS_DINODE_FMT_EXTENTS);
637 :
638 : /*
639 : * Make space in the inode incore. This needs to be undone if we fail
640 : * to expand the root.
641 : */
642 858297 : xfs_iroot_realloc(ip, 1, whichfork);
643 :
644 : /*
645 : * Fill in the root.
646 : */
647 858298 : block = ifp->if_broot;
648 858298 : xfs_btree_init_block(mp, block, &xfs_bmbt_ops, 1, 1, ip->i_ino);
649 : /*
650 : * Need a cursor. Can't allocate until bb_level is filled in.
651 : */
652 858296 : cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
653 858297 : cur->bc_ino.flags = wasdel ? XFS_BTCUR_BMBT_WASDEL : 0;
654 : /*
655 : * Convert to a btree with two levels, one record in root.
656 : */
657 858297 : ifp->if_format = XFS_DINODE_FMT_BTREE;
658 858297 : memset(&args, 0, sizeof(args));
659 858297 : args.tp = tp;
660 858297 : args.mp = mp;
661 858297 : xfs_rmap_ino_bmbt_owner(&args.oinfo, ip->i_ino, whichfork);
662 :
663 858297 : args.minlen = args.maxlen = args.prod = 1;
664 858297 : args.wasdel = wasdel;
665 858297 : *logflagsp = 0;
666 858297 : error = xfs_alloc_vextent_start_ag(&args,
667 858297 : XFS_INO_TO_FSB(mp, ip->i_ino));
668 858299 : if (error)
669 53 : goto out_root_realloc;
670 :
671 : /*
672 : * Allocation can't fail, the space was reserved.
673 : */
674 858246 : if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) {
675 0 : error = -ENOSPC;
676 0 : goto out_root_realloc;
677 : }
678 :
679 858246 : cur->bc_ino.allocated++;
680 858246 : ip->i_nblocks++;
681 858246 : xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
682 858246 : error = xfs_trans_get_buf(tp, mp->m_ddev_targp,
683 858246 : XFS_FSB_TO_DADDR(mp, args.fsbno),
684 : mp->m_bsize, 0, &abp);
685 858246 : if (error)
686 0 : goto out_unreserve_dquot;
687 :
688 : /*
689 : * Fill in the child block.
690 : */
691 858246 : ablock = XFS_BUF_TO_BLOCK(abp);
692 858246 : xfs_btree_init_buf(mp, abp, &xfs_bmbt_ops, 0, 0, ip->i_ino);
693 :
694 12114812 : for_each_xfs_iext(ifp, &icur, &rec) {
695 11256566 : if (isnullstartblock(rec.br_startblock))
696 206132 : continue;
697 11050434 : arp = XFS_BMBT_REC_ADDR(mp, ablock, 1 + cnt);
698 11050434 : xfs_bmbt_disk_set_all(arp, &rec);
699 11050434 : cnt++;
700 : }
701 858246 : ASSERT(cnt == ifp->if_nextents);
702 858246 : xfs_btree_set_numrecs(ablock, cnt);
703 :
704 : /*
705 : * Fill in the root key and pointer.
706 : */
707 858246 : kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
708 858246 : arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
709 858246 : kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
710 858246 : pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
711 : be16_to_cpu(block->bb_level)));
712 858246 : *pp = cpu_to_be64(args.fsbno);
713 :
714 : /*
715 : * Do all this logging at the end so that
716 : * the root is at the right level.
717 : */
718 858246 : xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
719 858246 : xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
720 858246 : ASSERT(*curp == NULL);
721 858246 : *curp = cur;
722 858246 : *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
723 858246 : return 0;
724 :
725 : out_unreserve_dquot:
726 0 : xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
727 53 : out_root_realloc:
728 53 : xfs_iroot_realloc(ip, -1, whichfork);
729 53 : ifp->if_format = XFS_DINODE_FMT_EXTENTS;
730 53 : ASSERT(ifp->if_broot == NULL);
731 53 : xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
732 :
733 53 : return error;
734 : }
735 :
736 : /*
737 : * Convert a local file to an extents file.
738 : * This code is out of bounds for data forks of regular files,
739 : * since the file data needs to get logged so things will stay consistent.
740 : * (The bmap-level manipulations are ok, though).
741 : */
742 : void
743 4575666 : xfs_bmap_local_to_extents_empty(
744 : struct xfs_trans *tp,
745 : struct xfs_inode *ip,
746 : int whichfork)
747 : {
748 4575666 : struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
749 :
750 4575710 : ASSERT(whichfork != XFS_COW_FORK);
751 4575710 : ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
752 4575710 : ASSERT(ifp->if_bytes == 0);
753 4575710 : ASSERT(ifp->if_nextents == 0);
754 :
755 4575710 : xfs_bmap_forkoff_reset(ip, whichfork);
756 4575662 : ifp->if_u1.if_root = NULL;
757 4575662 : ifp->if_height = 0;
758 4575662 : ifp->if_format = XFS_DINODE_FMT_EXTENTS;
759 4575662 : xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
760 4575687 : }
761 :
762 :
763 : int /* error */
764 559 : xfs_bmap_local_to_extents(
765 : xfs_trans_t *tp, /* transaction pointer */
766 : xfs_inode_t *ip, /* incore inode pointer */
767 : xfs_extlen_t total, /* total blocks needed by transaction */
768 : int *logflagsp, /* inode logging flags */
769 : int whichfork,
770 : void (*init_fn)(struct xfs_trans *tp,
771 : struct xfs_buf *bp,
772 : struct xfs_inode *ip,
773 : struct xfs_ifork *ifp, void *priv),
774 : void *priv)
775 : {
776 559 : int error = 0;
777 559 : int flags; /* logging flags returned */
778 559 : struct xfs_ifork *ifp; /* inode fork pointer */
779 559 : xfs_alloc_arg_t args; /* allocation arguments */
780 559 : struct xfs_buf *bp; /* buffer for extent block */
781 559 : struct xfs_bmbt_irec rec;
782 559 : struct xfs_iext_cursor icur;
783 :
784 : /*
785 : * We don't want to deal with the case of keeping inode data inline yet.
786 : * So sending the data fork of a regular inode is invalid.
787 : */
788 559 : ASSERT(!(S_ISREG(VFS_I(ip)->i_mode) && whichfork == XFS_DATA_FORK));
789 559 : ifp = xfs_ifork_ptr(ip, whichfork);
790 559 : ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
791 :
792 559 : if (!ifp->if_bytes) {
793 0 : xfs_bmap_local_to_extents_empty(tp, ip, whichfork);
794 0 : flags = XFS_ILOG_CORE;
795 0 : goto done;
796 : }
797 :
798 559 : flags = 0;
799 559 : error = 0;
800 559 : memset(&args, 0, sizeof(args));
801 559 : args.tp = tp;
802 559 : args.mp = ip->i_mount;
803 559 : args.total = total;
804 559 : args.minlen = args.maxlen = args.prod = 1;
805 559 : xfs_rmap_ino_owner(&args.oinfo, ip->i_ino, whichfork, 0);
806 :
807 : /*
808 : * Allocate a block. We know we need only one, since the
809 : * file currently fits in an inode.
810 : */
811 559 : args.total = total;
812 559 : args.minlen = args.maxlen = args.prod = 1;
813 559 : error = xfs_alloc_vextent_start_ag(&args,
814 559 : XFS_INO_TO_FSB(args.mp, ip->i_ino));
815 559 : if (error)
816 0 : goto done;
817 :
818 : /* Can't fail, the space was reserved. */
819 559 : ASSERT(args.fsbno != NULLFSBLOCK);
820 559 : ASSERT(args.len == 1);
821 559 : error = xfs_trans_get_buf(tp, args.mp->m_ddev_targp,
822 559 : XFS_FSB_TO_DADDR(args.mp, args.fsbno),
823 : args.mp->m_bsize, 0, &bp);
824 559 : if (error)
825 0 : goto done;
826 :
827 : /*
828 : * Initialize the block, copy the data and log the remote buffer.
829 : *
830 : * The callout is responsible for logging because the remote format
831 : * might differ from the local format and thus we don't know how much to
832 : * log here. Note that init_fn must also set the buffer log item type
833 : * correctly.
834 : */
835 559 : init_fn(tp, bp, ip, ifp, priv);
836 :
837 : /* account for the change in fork size */
838 559 : xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
839 559 : xfs_bmap_local_to_extents_empty(tp, ip, whichfork);
840 559 : flags |= XFS_ILOG_CORE;
841 :
842 559 : ifp->if_u1.if_root = NULL;
843 559 : ifp->if_height = 0;
844 :
845 559 : rec.br_startoff = 0;
846 559 : rec.br_startblock = args.fsbno;
847 559 : rec.br_blockcount = 1;
848 559 : rec.br_state = XFS_EXT_NORM;
849 559 : xfs_iext_first(ifp, &icur);
850 559 : xfs_iext_insert(ip, &icur, &rec, 0);
851 :
852 559 : ifp->if_nextents = 1;
853 559 : ip->i_nblocks = 1;
854 559 : xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
855 559 : flags |= xfs_ilog_fext(whichfork);
856 :
857 559 : done:
858 559 : *logflagsp = flags;
859 559 : return error;
860 : }
861 :
862 : /*
863 : * Called from xfs_bmap_add_attrfork to handle btree format files.
864 : */
865 : STATIC int /* error */
866 11 : xfs_bmap_add_attrfork_btree(
867 : xfs_trans_t *tp, /* transaction pointer */
868 : xfs_inode_t *ip, /* incore inode pointer */
869 : int *flags) /* inode logging flags */
870 : {
871 11 : struct xfs_btree_block *block = ip->i_df.if_broot;
872 11 : struct xfs_btree_cur *cur; /* btree cursor */
873 11 : int error; /* error return value */
874 11 : xfs_mount_t *mp; /* file system mount struct */
875 11 : int stat; /* newroot status */
876 :
877 11 : mp = ip->i_mount;
878 :
879 11 : if (XFS_BMAP_BMDR_SPACE(block) <= xfs_inode_data_fork_size(ip))
880 11 : *flags |= XFS_ILOG_DBROOT;
881 : else {
882 0 : cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
883 0 : error = xfs_bmbt_lookup_first(cur, &stat);
884 0 : if (error)
885 0 : goto error0;
886 : /* must be at least one entry */
887 0 : if (XFS_IS_CORRUPT(mp, stat != 1)) {
888 0 : xfs_btree_mark_sick(cur);
889 0 : error = -EFSCORRUPTED;
890 0 : goto error0;
891 : }
892 0 : if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
893 0 : goto error0;
894 0 : if (stat == 0) {
895 0 : xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
896 0 : return -ENOSPC;
897 : }
898 0 : cur->bc_ino.allocated = 0;
899 0 : xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
900 : }
901 : return 0;
902 0 : error0:
903 0 : xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
904 0 : return error;
905 : }
906 :
907 : /*
908 : * Called from xfs_bmap_add_attrfork to handle extents format files.
909 : */
910 : STATIC int /* error */
911 197692 : xfs_bmap_add_attrfork_extents(
912 : struct xfs_trans *tp, /* transaction pointer */
913 : struct xfs_inode *ip, /* incore inode pointer */
914 : int *flags) /* inode logging flags */
915 : {
916 197692 : struct xfs_btree_cur *cur; /* bmap btree cursor */
917 197692 : int error; /* error return value */
918 :
919 197692 : if (ip->i_df.if_nextents * sizeof(struct xfs_bmbt_rec) <=
920 197692 : xfs_inode_data_fork_size(ip))
921 : return 0;
922 9 : cur = NULL;
923 9 : error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0, flags,
924 : XFS_DATA_FORK);
925 9 : if (cur) {
926 9 : cur->bc_ino.allocated = 0;
927 9 : xfs_btree_del_cursor(cur, error);
928 : }
929 : return error;
930 : }
931 :
932 : /*
933 : * Called from xfs_bmap_add_attrfork to handle local format files. Each
934 : * different data fork content type needs a different callout to do the
935 : * conversion. Some are basic and only require special block initialisation
936 : * callouts for the data formating, others (directories) are so specialised they
937 : * handle everything themselves.
938 : *
939 : * XXX (dgc): investigate whether directory conversion can use the generic
940 : * formatting callout. It should be possible - it's just a very complex
941 : * formatter.
942 : */
943 : STATIC int /* error */
944 367 : xfs_bmap_add_attrfork_local(
945 : struct xfs_trans *tp, /* transaction pointer */
946 : struct xfs_inode *ip, /* incore inode pointer */
947 : int *flags) /* inode logging flags */
948 : {
949 367 : struct xfs_da_args dargs; /* args for dir/attr code */
950 :
951 367 : if (ip->i_df.if_bytes <= xfs_inode_data_fork_size(ip))
952 : return 0;
953 :
954 0 : if (S_ISDIR(VFS_I(ip)->i_mode)) {
955 0 : memset(&dargs, 0, sizeof(dargs));
956 0 : dargs.geo = ip->i_mount->m_dir_geo;
957 0 : dargs.dp = ip;
958 0 : dargs.total = dargs.geo->fsbcount;
959 0 : dargs.whichfork = XFS_DATA_FORK;
960 0 : dargs.trans = tp;
961 0 : dargs.owner = ip->i_ino;
962 0 : return xfs_dir2_sf_to_block(&dargs);
963 : }
964 :
965 0 : if (S_ISLNK(VFS_I(ip)->i_mode))
966 0 : return xfs_bmap_local_to_extents(tp, ip, 1, flags,
967 : XFS_DATA_FORK, xfs_symlink_local_to_remote,
968 : NULL);
969 :
970 : /* should only be called for types that support local format data */
971 0 : ASSERT(0);
972 0 : xfs_bmap_mark_sick(ip, XFS_ATTR_FORK);
973 0 : return -EFSCORRUPTED;
974 : }
975 :
976 : /*
977 : * Set an inode attr fork offset based on the format of the data fork.
978 : */
979 : static int
980 198073 : xfs_bmap_set_attrforkoff(
981 : struct xfs_inode *ip,
982 : int size,
983 : int *version)
984 : {
985 198073 : int default_size = xfs_default_attroffset(ip) >> 3;
986 :
987 198073 : switch (ip->i_df.if_format) {
988 0 : case XFS_DINODE_FMT_DEV:
989 0 : ip->i_forkoff = default_size;
990 0 : break;
991 198073 : case XFS_DINODE_FMT_LOCAL:
992 : case XFS_DINODE_FMT_EXTENTS:
993 : case XFS_DINODE_FMT_BTREE:
994 198073 : ip->i_forkoff = xfs_attr_shortform_bytesfit(ip, size);
995 198063 : if (!ip->i_forkoff)
996 0 : ip->i_forkoff = default_size;
997 198063 : else if (xfs_has_attr2(ip->i_mount) && version)
998 198063 : *version = 2;
999 : break;
1000 0 : default:
1001 0 : ASSERT(0);
1002 0 : return -EINVAL;
1003 : }
1004 :
1005 : return 0;
1006 : }
1007 :
1008 : /*
1009 : * Convert inode from non-attributed to attributed.
1010 : * Must not be in a transaction, ip must not be locked.
1011 : */
1012 : int /* error code */
1013 352372 : xfs_bmap_add_attrfork(
1014 : xfs_inode_t *ip, /* incore inode pointer */
1015 : int size, /* space new attribute needs */
1016 : int rsvd) /* xact may use reserved blks */
1017 : {
1018 352372 : xfs_mount_t *mp; /* mount structure */
1019 352372 : xfs_trans_t *tp; /* transaction pointer */
1020 352372 : int blks; /* space reservation */
1021 352372 : int version = 1; /* superblock attr version */
1022 352372 : int logflags; /* logging flags */
1023 352372 : int error; /* error return value */
1024 :
1025 352372 : mp = ip->i_mount;
1026 352372 : ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1027 :
1028 352372 : blks = XFS_ADDAFORK_SPACE_RES(mp);
1029 :
1030 352372 : error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_addafork, blks, 0,
1031 : rsvd, &tp);
1032 352375 : if (error)
1033 : return error;
1034 352375 : if (xfs_inode_has_attr_fork(ip))
1035 154302 : goto trans_cancel;
1036 :
1037 198073 : xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1038 198073 : error = xfs_bmap_set_attrforkoff(ip, size, &version);
1039 198063 : if (error)
1040 0 : goto trans_cancel;
1041 :
1042 198063 : xfs_ifork_init_attr(ip, XFS_DINODE_FMT_EXTENTS, 0);
1043 198064 : logflags = 0;
1044 198064 : switch (ip->i_df.if_format) {
1045 367 : case XFS_DINODE_FMT_LOCAL:
1046 367 : error = xfs_bmap_add_attrfork_local(tp, ip, &logflags);
1047 367 : break;
1048 197686 : case XFS_DINODE_FMT_EXTENTS:
1049 197686 : error = xfs_bmap_add_attrfork_extents(tp, ip, &logflags);
1050 197686 : break;
1051 11 : case XFS_DINODE_FMT_BTREE:
1052 11 : error = xfs_bmap_add_attrfork_btree(tp, ip, &logflags);
1053 11 : break;
1054 : default:
1055 : error = 0;
1056 : break;
1057 : }
1058 198064 : if (logflags)
1059 20 : xfs_trans_log_inode(tp, ip, logflags);
1060 198063 : if (error)
1061 0 : goto trans_cancel;
1062 198063 : if (!xfs_has_attr(mp) ||
1063 0 : (!xfs_has_attr2(mp) && version == 2)) {
1064 0 : bool log_sb = false;
1065 :
1066 0 : spin_lock(&mp->m_sb_lock);
1067 0 : if (!xfs_has_attr(mp)) {
1068 0 : xfs_add_attr(mp);
1069 0 : log_sb = true;
1070 : }
1071 0 : if (!xfs_has_attr2(mp) && version == 2) {
1072 0 : xfs_add_attr2(mp);
1073 0 : log_sb = true;
1074 : }
1075 0 : spin_unlock(&mp->m_sb_lock);
1076 0 : if (log_sb)
1077 0 : xfs_log_sb(tp);
1078 : }
1079 :
1080 198063 : error = xfs_trans_commit(tp);
1081 198073 : xfs_iunlock(ip, XFS_ILOCK_EXCL);
1082 198073 : return error;
1083 :
1084 154302 : trans_cancel:
1085 154302 : xfs_trans_cancel(tp);
1086 154302 : xfs_iunlock(ip, XFS_ILOCK_EXCL);
1087 154302 : return error;
1088 : }
1089 :
1090 : /*
1091 : * Internal and external extent tree search functions.
1092 : */
1093 :
1094 : struct xfs_iread_state {
1095 : struct xfs_iext_cursor icur;
1096 : xfs_extnum_t loaded;
1097 : };
1098 :
1099 : int
1100 0 : xfs_bmap_complain_bad_rec(
1101 : struct xfs_inode *ip,
1102 : int whichfork,
1103 : xfs_failaddr_t fa,
1104 : const struct xfs_bmbt_irec *irec)
1105 : {
1106 0 : struct xfs_mount *mp = ip->i_mount;
1107 0 : const char *forkname;
1108 :
1109 0 : switch (whichfork) {
1110 : case XFS_DATA_FORK: forkname = "data"; break;
1111 : case XFS_ATTR_FORK: forkname = "attr"; break;
1112 : case XFS_COW_FORK: forkname = "CoW"; break;
1113 : default: forkname = "???"; break;
1114 : }
1115 :
1116 0 : xfs_warn(mp,
1117 : "Bmap BTree record corruption in inode 0x%llx %s fork detected at %pS!",
1118 : ip->i_ino, forkname, fa);
1119 0 : xfs_warn(mp,
1120 : "Offset 0x%llx, start block 0x%llx, block count 0x%llx state 0x%x",
1121 : irec->br_startoff, irec->br_startblock, irec->br_blockcount,
1122 : irec->br_state);
1123 :
1124 0 : return -EFSCORRUPTED;
1125 : }
1126 :
1127 : /* Stuff every bmbt record from this block into the incore extent map. */
1128 : static int
1129 5773545 : xfs_iread_bmbt_block(
1130 : struct xfs_btree_cur *cur,
1131 : int level,
1132 : void *priv)
1133 : {
1134 5773545 : struct xfs_iread_state *ir = priv;
1135 5773545 : struct xfs_mount *mp = cur->bc_mp;
1136 5773545 : struct xfs_inode *ip = cur->bc_ino.ip;
1137 5773545 : struct xfs_btree_block *block;
1138 5773545 : struct xfs_buf *bp;
1139 5773545 : struct xfs_bmbt_rec *frp;
1140 5773545 : xfs_extnum_t num_recs;
1141 5773545 : xfs_extnum_t j;
1142 5773545 : int whichfork = cur->bc_ino.whichfork;
1143 5773545 : struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
1144 :
1145 5773541 : block = xfs_btree_get_block(cur, level, &bp);
1146 :
1147 : /* Abort if we find more records than nextents. */
1148 5773537 : num_recs = xfs_btree_get_numrecs(block);
1149 5773537 : if (unlikely(ir->loaded + num_recs > ifp->if_nextents)) {
1150 0 : xfs_warn(ip->i_mount, "corrupt dinode %llu, (btree extents).",
1151 : (unsigned long long)ip->i_ino);
1152 0 : xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__, block,
1153 0 : sizeof(*block), __this_address);
1154 0 : xfs_bmap_mark_sick(ip, whichfork);
1155 0 : return -EFSCORRUPTED;
1156 : }
1157 :
1158 : /* Copy records into the incore cache. */
1159 5773537 : frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1160 972772572 : for (j = 0; j < num_recs; j++, frp++, ir->loaded++) {
1161 966999032 : struct xfs_bmbt_irec new;
1162 966999032 : xfs_failaddr_t fa;
1163 :
1164 966999032 : xfs_bmbt_disk_get_all(frp, &new);
1165 966998143 : fa = xfs_bmap_validate_extent(ip, whichfork, &new);
1166 966999258 : if (fa) {
1167 0 : xfs_inode_verifier_error(ip, -EFSCORRUPTED,
1168 : "xfs_iread_extents(2)", frp,
1169 : sizeof(*frp), fa);
1170 0 : xfs_bmap_mark_sick(ip, whichfork);
1171 0 : return xfs_bmap_complain_bad_rec(ip, whichfork, fa,
1172 : &new);
1173 : }
1174 1933993546 : xfs_iext_insert(ip, &ir->icur, &new,
1175 : xfs_bmap_fork_to_state(whichfork));
1176 1933991094 : trace_xfs_read_extent(ip, &ir->icur,
1177 966998032 : xfs_bmap_fork_to_state(whichfork), _THIS_IP_);
1178 966998566 : xfs_iext_next(ifp, &ir->icur);
1179 : }
1180 :
1181 : return 0;
1182 : }
1183 :
1184 : /*
1185 : * Read in extents from a btree-format inode.
1186 : */
1187 : int
1188 4451562865 : xfs_iread_extents(
1189 : struct xfs_trans *tp,
1190 : struct xfs_inode *ip,
1191 : int whichfork)
1192 : {
1193 4451562865 : struct xfs_iread_state ir;
1194 4451562865 : struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
1195 4453142112 : struct xfs_mount *mp = ip->i_mount;
1196 4453142112 : struct xfs_btree_cur *cur;
1197 4453142112 : int error;
1198 :
1199 4453142112 : if (!xfs_need_iread_extents(ifp))
1200 : return 0;
1201 :
1202 1252869 : ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1203 :
1204 1252869 : ir.loaded = 0;
1205 1252869 : xfs_iext_first(ifp, &ir.icur);
1206 1252869 : cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
1207 1252867 : error = xfs_btree_visit_blocks(cur, xfs_iread_bmbt_block,
1208 : XFS_BTREE_VISIT_RECORDS, &ir);
1209 1252855 : xfs_btree_del_cursor(cur, error);
1210 1252867 : if (error)
1211 198 : goto out;
1212 :
1213 1252669 : if (XFS_IS_CORRUPT(mp, ir.loaded != ifp->if_nextents)) {
1214 0 : xfs_bmap_mark_sick(ip, whichfork);
1215 0 : error = -EFSCORRUPTED;
1216 0 : goto out;
1217 : }
1218 1252669 : ASSERT(ir.loaded == xfs_iext_count(ifp));
1219 : /*
1220 : * Use release semantics so that we can use acquire semantics in
1221 : * xfs_need_iread_extents and be guaranteed to see a valid mapping tree
1222 : * after that load.
1223 : */
1224 1252669 : smp_store_release(&ifp->if_needextents, 0);
1225 1252669 : return 0;
1226 198 : out:
1227 198 : if (xfs_metadata_is_sick(error))
1228 2 : xfs_bmap_mark_sick(ip, whichfork);
1229 198 : xfs_iext_destroy(ifp);
1230 198 : return error;
1231 : }
1232 :
1233 : /*
1234 : * Returns the relative block number of the first unused block(s) in the given
1235 : * fork with at least "len" logically contiguous blocks free. This is the
1236 : * lowest-address hole if the fork has holes, else the first block past the end
1237 : * of fork. Return 0 if the fork is currently local (in-inode).
1238 : */
1239 : int /* error */
1240 4750991 : xfs_bmap_first_unused(
1241 : struct xfs_trans *tp, /* transaction pointer */
1242 : struct xfs_inode *ip, /* incore inode */
1243 : xfs_extlen_t len, /* size of hole to find */
1244 : xfs_fileoff_t *first_unused, /* unused block */
1245 : int whichfork) /* data or attr fork */
1246 : {
1247 4750991 : struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
1248 4751042 : struct xfs_bmbt_irec got;
1249 4751042 : struct xfs_iext_cursor icur;
1250 4751042 : xfs_fileoff_t lastaddr = 0;
1251 4751042 : xfs_fileoff_t lowest, max;
1252 4751042 : int error;
1253 :
1254 4751042 : if (ifp->if_format == XFS_DINODE_FMT_LOCAL) {
1255 0 : *first_unused = 0;
1256 0 : return 0;
1257 : }
1258 :
1259 4751042 : ASSERT(xfs_ifork_has_extents(ifp));
1260 :
1261 4751042 : error = xfs_iread_extents(tp, ip, whichfork);
1262 4751010 : if (error)
1263 : return error;
1264 :
1265 4751102 : lowest = max = *first_unused;
1266 16779225 : for_each_xfs_iext(ifp, &icur, &got) {
1267 : /*
1268 : * See if the hole before this extent will work.
1269 : */
1270 12098478 : if (got.br_startoff >= lowest + len &&
1271 8763621 : got.br_startoff - max >= len)
1272 : break;
1273 12028283 : lastaddr = got.br_startoff + got.br_blockcount;
1274 12028283 : max = XFS_FILEOFF_MAX(lastaddr, lowest);
1275 : }
1276 :
1277 4751003 : *first_unused = max;
1278 4751003 : return 0;
1279 : }
1280 :
1281 : /*
1282 : * Returns the file-relative block number of the last block - 1 before
1283 : * last_block (input value) in the file.
1284 : * This is not based on i_size, it is based on the extent records.
1285 : * Returns 0 for local files, as they do not have extent records.
1286 : */
1287 : int /* error */
1288 239655 : xfs_bmap_last_before(
1289 : struct xfs_trans *tp, /* transaction pointer */
1290 : struct xfs_inode *ip, /* incore inode */
1291 : xfs_fileoff_t *last_block, /* last block */
1292 : int whichfork) /* data or attr fork */
1293 : {
1294 239655 : struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
1295 239655 : struct xfs_bmbt_irec got;
1296 239655 : struct xfs_iext_cursor icur;
1297 239655 : int error;
1298 :
1299 239655 : switch (ifp->if_format) {
1300 0 : case XFS_DINODE_FMT_LOCAL:
1301 0 : *last_block = 0;
1302 0 : return 0;
1303 : case XFS_DINODE_FMT_BTREE:
1304 : case XFS_DINODE_FMT_EXTENTS:
1305 239655 : break;
1306 0 : default:
1307 0 : ASSERT(0);
1308 0 : xfs_bmap_mark_sick(ip, whichfork);
1309 0 : return -EFSCORRUPTED;
1310 : }
1311 :
1312 239655 : error = xfs_iread_extents(tp, ip, whichfork);
1313 239655 : if (error)
1314 : return error;
1315 :
1316 239655 : if (!xfs_iext_lookup_extent_before(ip, ifp, last_block, &icur, &got))
1317 217278 : *last_block = 0;
1318 : return 0;
1319 : }
1320 :
1321 : int
1322 475199889 : xfs_bmap_last_extent(
1323 : struct xfs_trans *tp,
1324 : struct xfs_inode *ip,
1325 : int whichfork,
1326 : struct xfs_bmbt_irec *rec,
1327 : int *is_empty)
1328 : {
1329 475199889 : struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
1330 475233301 : struct xfs_iext_cursor icur;
1331 475233301 : int error;
1332 :
1333 475233301 : error = xfs_iread_extents(tp, ip, whichfork);
1334 475322702 : if (error)
1335 : return error;
1336 :
1337 475305742 : xfs_iext_last(ifp, &icur);
1338 475314394 : if (!xfs_iext_get_extent(ifp, &icur, rec))
1339 18061586 : *is_empty = 1;
1340 : else
1341 457260922 : *is_empty = 0;
1342 : return 0;
1343 : }
1344 :
1345 : /*
1346 : * Check the last inode extent to determine whether this allocation will result
1347 : * in blocks being allocated at the end of the file. When we allocate new data
1348 : * blocks at the end of the file which do not start at the previous data block,
1349 : * we will try to align the new blocks at stripe unit boundaries.
1350 : *
1351 : * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1352 : * at, or past the EOF.
1353 : */
1354 : STATIC int
1355 6443 : xfs_bmap_isaeof(
1356 : struct xfs_bmalloca *bma,
1357 : int whichfork)
1358 : {
1359 6443 : struct xfs_bmbt_irec rec;
1360 6443 : int is_empty;
1361 6443 : int error;
1362 :
1363 6443 : bma->aeof = false;
1364 6443 : error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1365 : &is_empty);
1366 6444 : if (error)
1367 : return error;
1368 :
1369 6444 : if (is_empty) {
1370 1231 : bma->aeof = true;
1371 1231 : return 0;
1372 : }
1373 :
1374 : /*
1375 : * Check if we are allocation or past the last extent, or at least into
1376 : * the last delayed allocated extent.
1377 : */
1378 5213 : bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1379 1274 : (bma->offset >= rec.br_startoff &&
1380 1274 : isnullstartblock(rec.br_startblock));
1381 5213 : return 0;
1382 : }
1383 :
1384 : /*
1385 : * Returns the file-relative block number of the first block past eof in
1386 : * the file. This is not based on i_size, it is based on the extent records.
1387 : * Returns 0 for local files, as they do not have extent records.
1388 : */
1389 : int
1390 474798102 : xfs_bmap_last_offset(
1391 : struct xfs_inode *ip,
1392 : xfs_fileoff_t *last_block,
1393 : int whichfork)
1394 : {
1395 474798102 : struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
1396 474839686 : struct xfs_bmbt_irec rec;
1397 474839686 : int is_empty;
1398 474839686 : int error;
1399 :
1400 474839686 : *last_block = 0;
1401 :
1402 474839686 : if (ifp->if_format == XFS_DINODE_FMT_LOCAL)
1403 : return 0;
1404 :
1405 474839686 : if (XFS_IS_CORRUPT(ip->i_mount, !xfs_ifork_has_extents(ifp))) {
1406 0 : xfs_bmap_mark_sick(ip, whichfork);
1407 0 : return -EFSCORRUPTED;
1408 : }
1409 :
1410 474839686 : error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1411 474964766 : if (error || is_empty)
1412 : return error;
1413 :
1414 456912650 : *last_block = rec.br_startoff + rec.br_blockcount;
1415 456912650 : return 0;
1416 : }
1417 :
1418 : /*
1419 : * Extent tree manipulation functions used during allocation.
1420 : */
1421 :
1422 : /*
1423 : * Convert a delayed allocation to a real allocation.
1424 : */
1425 : STATIC int /* error */
1426 15729650 : xfs_bmap_add_extent_delay_real(
1427 : struct xfs_bmalloca *bma,
1428 : int whichfork)
1429 : {
1430 15729650 : struct xfs_mount *mp = bma->ip->i_mount;
1431 15729650 : struct xfs_ifork *ifp = xfs_ifork_ptr(bma->ip, whichfork);
1432 15729656 : struct xfs_bmbt_irec *new = &bma->got;
1433 15729656 : int error; /* error return value */
1434 15729656 : int i; /* temp state */
1435 15729656 : xfs_fileoff_t new_endoff; /* end offset of new entry */
1436 15729656 : xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
1437 : /* left is 0, right is 1, prev is 2 */
1438 15729656 : int rval=0; /* return value (logging flags) */
1439 15729656 : uint32_t state = xfs_bmap_fork_to_state(whichfork);
1440 15729656 : xfs_filblks_t da_new; /* new count del alloc blocks used */
1441 15729656 : xfs_filblks_t da_old; /* old count del alloc blocks used */
1442 15729656 : xfs_filblks_t temp=0; /* value for da_new calculations */
1443 15729656 : int tmp_rval; /* partial logging flags */
1444 15729656 : struct xfs_bmbt_irec old;
1445 :
1446 15729656 : ASSERT(whichfork != XFS_ATTR_FORK);
1447 15729656 : ASSERT(!isnullstartblock(new->br_startblock));
1448 15729656 : ASSERT(!bma->cur ||
1449 : (bma->cur->bc_ino.flags & XFS_BTCUR_BMBT_WASDEL));
1450 :
1451 15729656 : XFS_STATS_INC(mp, xs_add_exlist);
1452 :
1453 : #define LEFT r[0]
1454 : #define RIGHT r[1]
1455 : #define PREV r[2]
1456 :
1457 : /*
1458 : * Set up a bunch of variables to make the tests simpler.
1459 : */
1460 15729656 : xfs_iext_get_extent(ifp, &bma->icur, &PREV);
1461 15729657 : new_endoff = new->br_startoff + new->br_blockcount;
1462 15729657 : ASSERT(isnullstartblock(PREV.br_startblock));
1463 15729657 : ASSERT(PREV.br_startoff <= new->br_startoff);
1464 15729657 : ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1465 :
1466 15729657 : da_old = startblockval(PREV.br_startblock);
1467 15729657 : da_new = 0;
1468 :
1469 : /*
1470 : * Set flags determining what part of the previous delayed allocation
1471 : * extent is being replaced by a real allocation.
1472 : */
1473 15729657 : if (PREV.br_startoff == new->br_startoff)
1474 15729664 : state |= BMAP_LEFT_FILLING;
1475 15729657 : if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1476 15497896 : state |= BMAP_RIGHT_FILLING;
1477 :
1478 : /*
1479 : * Check and set flags if this segment has a left neighbor.
1480 : * Don't set contiguous if the combined extent would be too large.
1481 : */
1482 15729657 : if (xfs_iext_peek_prev_extent(ifp, &bma->icur, &LEFT)) {
1483 10557601 : state |= BMAP_LEFT_VALID;
1484 10557601 : if (isnullstartblock(LEFT.br_startblock))
1485 44518 : state |= BMAP_LEFT_DELAY;
1486 : }
1487 :
1488 15729637 : if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1489 10513075 : LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1490 872128 : LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1491 196792 : LEFT.br_state == new->br_state &&
1492 93771 : LEFT.br_blockcount + new->br_blockcount <= XFS_MAX_BMBT_EXTLEN)
1493 89543 : state |= BMAP_LEFT_CONTIG;
1494 :
1495 : /*
1496 : * Check and set flags if this segment has a right neighbor.
1497 : * Don't set contiguous if the combined extent would be too large.
1498 : * Also check for all-three-contiguous being too large.
1499 : */
1500 15729637 : if (xfs_iext_peek_next_extent(ifp, &bma->icur, &RIGHT)) {
1501 7599912 : state |= BMAP_RIGHT_VALID;
1502 7599912 : if (isnullstartblock(RIGHT.br_startblock))
1503 787161 : state |= BMAP_RIGHT_DELAY;
1504 : }
1505 :
1506 15729628 : if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1507 6812752 : new_endoff == RIGHT.br_startoff &&
1508 582262 : new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1509 54026 : new->br_state == RIGHT.br_state &&
1510 25410 : new->br_blockcount + RIGHT.br_blockcount <= XFS_MAX_BMBT_EXTLEN &&
1511 25410 : ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1512 : BMAP_RIGHT_FILLING)) !=
1513 : (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1514 11517 : BMAP_RIGHT_FILLING) ||
1515 11517 : LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1516 : <= XFS_MAX_BMBT_EXTLEN))
1517 25410 : state |= BMAP_RIGHT_CONTIG;
1518 :
1519 15729628 : error = 0;
1520 : /*
1521 : * Switch out based on the FILLING and CONTIG state bits.
1522 : */
1523 15729628 : switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1524 : BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1525 11517 : case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1526 : BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1527 : /*
1528 : * Filling in all of a previously delayed allocation extent.
1529 : * The left and right neighbors are both contiguous with new.
1530 : */
1531 11517 : LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount;
1532 :
1533 11517 : xfs_iext_remove(bma->ip, &bma->icur, state);
1534 11517 : xfs_iext_remove(bma->ip, &bma->icur, state);
1535 11517 : xfs_iext_prev(ifp, &bma->icur);
1536 11517 : xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1537 11517 : ifp->if_nextents--;
1538 :
1539 11517 : if (bma->cur == NULL)
1540 : rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1541 : else {
1542 1837 : rval = XFS_ILOG_CORE;
1543 1837 : error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i);
1544 1837 : if (error)
1545 0 : goto done;
1546 1837 : if (XFS_IS_CORRUPT(mp, i != 1)) {
1547 0 : xfs_btree_mark_sick(bma->cur);
1548 0 : error = -EFSCORRUPTED;
1549 0 : goto done;
1550 : }
1551 1837 : error = xfs_btree_delete(bma->cur, &i);
1552 1837 : if (error)
1553 0 : goto done;
1554 1837 : if (XFS_IS_CORRUPT(mp, i != 1)) {
1555 0 : xfs_btree_mark_sick(bma->cur);
1556 0 : error = -EFSCORRUPTED;
1557 0 : goto done;
1558 : }
1559 1837 : error = xfs_btree_decrement(bma->cur, 0, &i);
1560 1837 : if (error)
1561 0 : goto done;
1562 1837 : if (XFS_IS_CORRUPT(mp, i != 1)) {
1563 0 : xfs_btree_mark_sick(bma->cur);
1564 0 : error = -EFSCORRUPTED;
1565 0 : goto done;
1566 : }
1567 1837 : error = xfs_bmbt_update(bma->cur, &LEFT);
1568 1837 : if (error)
1569 0 : goto done;
1570 : }
1571 : break;
1572 :
1573 63754 : case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1574 : /*
1575 : * Filling in all of a previously delayed allocation extent.
1576 : * The left neighbor is contiguous, the right is not.
1577 : */
1578 63754 : old = LEFT;
1579 63754 : LEFT.br_blockcount += PREV.br_blockcount;
1580 :
1581 63754 : xfs_iext_remove(bma->ip, &bma->icur, state);
1582 63754 : xfs_iext_prev(ifp, &bma->icur);
1583 63754 : xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1584 :
1585 63754 : if (bma->cur == NULL)
1586 : rval = XFS_ILOG_DEXT;
1587 : else {
1588 7787 : rval = 0;
1589 7787 : error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1590 7787 : if (error)
1591 0 : goto done;
1592 7787 : if (XFS_IS_CORRUPT(mp, i != 1)) {
1593 0 : xfs_btree_mark_sick(bma->cur);
1594 0 : error = -EFSCORRUPTED;
1595 0 : goto done;
1596 : }
1597 7787 : error = xfs_bmbt_update(bma->cur, &LEFT);
1598 7787 : if (error)
1599 0 : goto done;
1600 : }
1601 : break;
1602 :
1603 13893 : case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1604 : /*
1605 : * Filling in all of a previously delayed allocation extent.
1606 : * The right neighbor is contiguous, the left is not. Take care
1607 : * with delay -> unwritten extent allocation here because the
1608 : * delalloc record we are overwriting is always written.
1609 : */
1610 13893 : PREV.br_startblock = new->br_startblock;
1611 13893 : PREV.br_blockcount += RIGHT.br_blockcount;
1612 13893 : PREV.br_state = new->br_state;
1613 :
1614 13893 : xfs_iext_next(ifp, &bma->icur);
1615 13893 : xfs_iext_remove(bma->ip, &bma->icur, state);
1616 13893 : xfs_iext_prev(ifp, &bma->icur);
1617 13893 : xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1618 :
1619 13893 : if (bma->cur == NULL)
1620 : rval = XFS_ILOG_DEXT;
1621 : else {
1622 1769 : rval = 0;
1623 1769 : error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i);
1624 1769 : if (error)
1625 0 : goto done;
1626 1769 : if (XFS_IS_CORRUPT(mp, i != 1)) {
1627 0 : xfs_btree_mark_sick(bma->cur);
1628 0 : error = -EFSCORRUPTED;
1629 0 : goto done;
1630 : }
1631 1769 : error = xfs_bmbt_update(bma->cur, &PREV);
1632 1769 : if (error)
1633 0 : goto done;
1634 : }
1635 : break;
1636 :
1637 15408700 : case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1638 : /*
1639 : * Filling in all of a previously delayed allocation extent.
1640 : * Neither the left nor right neighbors are contiguous with
1641 : * the new one.
1642 : */
1643 15408700 : PREV.br_startblock = new->br_startblock;
1644 15408700 : PREV.br_state = new->br_state;
1645 15408700 : xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1646 15408702 : ifp->if_nextents++;
1647 :
1648 15408702 : if (bma->cur == NULL)
1649 : rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1650 : else {
1651 2279366 : rval = XFS_ILOG_CORE;
1652 2279366 : error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1653 2279366 : if (error)
1654 1 : goto done;
1655 2279365 : if (XFS_IS_CORRUPT(mp, i != 0)) {
1656 0 : xfs_btree_mark_sick(bma->cur);
1657 0 : error = -EFSCORRUPTED;
1658 0 : goto done;
1659 : }
1660 2279365 : error = xfs_btree_insert(bma->cur, &i);
1661 2279365 : if (error)
1662 0 : goto done;
1663 2279365 : if (XFS_IS_CORRUPT(mp, i != 1)) {
1664 0 : xfs_btree_mark_sick(bma->cur);
1665 0 : error = -EFSCORRUPTED;
1666 0 : goto done;
1667 : }
1668 : }
1669 : break;
1670 :
1671 14272 : case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1672 : /*
1673 : * Filling in the first part of a previous delayed allocation.
1674 : * The left neighbor is contiguous.
1675 : */
1676 14272 : old = LEFT;
1677 14272 : temp = PREV.br_blockcount - new->br_blockcount;
1678 14272 : da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1679 : startblockval(PREV.br_startblock));
1680 :
1681 14272 : LEFT.br_blockcount += new->br_blockcount;
1682 :
1683 14272 : PREV.br_blockcount = temp;
1684 14272 : PREV.br_startoff += new->br_blockcount;
1685 14272 : PREV.br_startblock = nullstartblock(da_new);
1686 :
1687 14272 : xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1688 14272 : xfs_iext_prev(ifp, &bma->icur);
1689 14272 : xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1690 :
1691 14272 : if (bma->cur == NULL)
1692 : rval = XFS_ILOG_DEXT;
1693 : else {
1694 11873 : rval = 0;
1695 11873 : error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1696 11873 : if (error)
1697 0 : goto done;
1698 11873 : if (XFS_IS_CORRUPT(mp, i != 1)) {
1699 0 : xfs_btree_mark_sick(bma->cur);
1700 0 : error = -EFSCORRUPTED;
1701 0 : goto done;
1702 : }
1703 11873 : error = xfs_bmbt_update(bma->cur, &LEFT);
1704 11873 : if (error)
1705 0 : goto done;
1706 : }
1707 : break;
1708 :
1709 217492 : case BMAP_LEFT_FILLING:
1710 : /*
1711 : * Filling in the first part of a previous delayed allocation.
1712 : * The left neighbor is not contiguous.
1713 : */
1714 217492 : xfs_iext_update_extent(bma->ip, state, &bma->icur, new);
1715 217495 : ifp->if_nextents++;
1716 :
1717 217495 : if (bma->cur == NULL)
1718 : rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1719 : else {
1720 183567 : rval = XFS_ILOG_CORE;
1721 183567 : error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1722 183565 : if (error)
1723 0 : goto done;
1724 183565 : if (XFS_IS_CORRUPT(mp, i != 0)) {
1725 0 : xfs_btree_mark_sick(bma->cur);
1726 0 : error = -EFSCORRUPTED;
1727 0 : goto done;
1728 : }
1729 183565 : error = xfs_btree_insert(bma->cur, &i);
1730 183566 : if (error)
1731 0 : goto done;
1732 183566 : if (XFS_IS_CORRUPT(mp, i != 1)) {
1733 0 : xfs_btree_mark_sick(bma->cur);
1734 0 : error = -EFSCORRUPTED;
1735 0 : goto done;
1736 : }
1737 : }
1738 :
1739 217494 : if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1740 2057 : error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1741 : &bma->cur, 1, &tmp_rval, whichfork);
1742 2057 : rval |= tmp_rval;
1743 2057 : if (error)
1744 0 : goto done;
1745 : }
1746 :
1747 217495 : temp = PREV.br_blockcount - new->br_blockcount;
1748 217495 : da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1749 : startblockval(PREV.br_startblock) -
1750 : (bma->cur ? bma->cur->bc_ino.allocated : 0));
1751 :
1752 217496 : PREV.br_startoff = new_endoff;
1753 217496 : PREV.br_blockcount = temp;
1754 217496 : PREV.br_startblock = nullstartblock(da_new);
1755 217496 : xfs_iext_next(ifp, &bma->icur);
1756 217495 : xfs_iext_insert(bma->ip, &bma->icur, &PREV, state);
1757 217496 : xfs_iext_prev(ifp, &bma->icur);
1758 217496 : break;
1759 :
1760 0 : case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1761 : /*
1762 : * Filling in the last part of a previous delayed allocation.
1763 : * The right neighbor is contiguous with the new allocation.
1764 : */
1765 0 : old = RIGHT;
1766 0 : RIGHT.br_startoff = new->br_startoff;
1767 0 : RIGHT.br_startblock = new->br_startblock;
1768 0 : RIGHT.br_blockcount += new->br_blockcount;
1769 :
1770 0 : if (bma->cur == NULL)
1771 : rval = XFS_ILOG_DEXT;
1772 : else {
1773 0 : rval = 0;
1774 0 : error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1775 0 : if (error)
1776 0 : goto done;
1777 0 : if (XFS_IS_CORRUPT(mp, i != 1)) {
1778 0 : xfs_btree_mark_sick(bma->cur);
1779 0 : error = -EFSCORRUPTED;
1780 0 : goto done;
1781 : }
1782 0 : error = xfs_bmbt_update(bma->cur, &RIGHT);
1783 0 : if (error)
1784 0 : goto done;
1785 : }
1786 :
1787 0 : temp = PREV.br_blockcount - new->br_blockcount;
1788 0 : da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1789 : startblockval(PREV.br_startblock));
1790 :
1791 0 : PREV.br_blockcount = temp;
1792 0 : PREV.br_startblock = nullstartblock(da_new);
1793 :
1794 0 : xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1795 0 : xfs_iext_next(ifp, &bma->icur);
1796 0 : xfs_iext_update_extent(bma->ip, state, &bma->icur, &RIGHT);
1797 0 : break;
1798 :
1799 0 : case BMAP_RIGHT_FILLING:
1800 : /*
1801 : * Filling in the last part of a previous delayed allocation.
1802 : * The right neighbor is not contiguous.
1803 : */
1804 0 : xfs_iext_update_extent(bma->ip, state, &bma->icur, new);
1805 0 : ifp->if_nextents++;
1806 :
1807 0 : if (bma->cur == NULL)
1808 : rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1809 : else {
1810 0 : rval = XFS_ILOG_CORE;
1811 0 : error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1812 0 : if (error)
1813 0 : goto done;
1814 0 : if (XFS_IS_CORRUPT(mp, i != 0)) {
1815 0 : xfs_btree_mark_sick(bma->cur);
1816 0 : error = -EFSCORRUPTED;
1817 0 : goto done;
1818 : }
1819 0 : error = xfs_btree_insert(bma->cur, &i);
1820 0 : if (error)
1821 0 : goto done;
1822 0 : if (XFS_IS_CORRUPT(mp, i != 1)) {
1823 0 : xfs_btree_mark_sick(bma->cur);
1824 0 : error = -EFSCORRUPTED;
1825 0 : goto done;
1826 : }
1827 : }
1828 :
1829 0 : if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1830 0 : error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1831 : &bma->cur, 1, &tmp_rval, whichfork);
1832 0 : rval |= tmp_rval;
1833 0 : if (error)
1834 0 : goto done;
1835 : }
1836 :
1837 0 : temp = PREV.br_blockcount - new->br_blockcount;
1838 0 : da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1839 : startblockval(PREV.br_startblock) -
1840 : (bma->cur ? bma->cur->bc_ino.allocated : 0));
1841 :
1842 0 : PREV.br_startblock = nullstartblock(da_new);
1843 0 : PREV.br_blockcount = temp;
1844 0 : xfs_iext_insert(bma->ip, &bma->icur, &PREV, state);
1845 0 : xfs_iext_next(ifp, &bma->icur);
1846 0 : break;
1847 :
1848 0 : case 0:
1849 : /*
1850 : * Filling in the middle part of a previous delayed allocation.
1851 : * Contiguity is impossible here.
1852 : * This case is avoided almost all the time.
1853 : *
1854 : * We start with a delayed allocation:
1855 : *
1856 : * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
1857 : * PREV @ idx
1858 : *
1859 : * and we are allocating:
1860 : * +rrrrrrrrrrrrrrrrr+
1861 : * new
1862 : *
1863 : * and we set it up for insertion as:
1864 : * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
1865 : * new
1866 : * PREV @ idx LEFT RIGHT
1867 : * inserted at idx + 1
1868 : */
1869 0 : old = PREV;
1870 :
1871 : /* LEFT is the new middle */
1872 0 : LEFT = *new;
1873 :
1874 : /* RIGHT is the new right */
1875 0 : RIGHT.br_state = PREV.br_state;
1876 0 : RIGHT.br_startoff = new_endoff;
1877 0 : RIGHT.br_blockcount =
1878 0 : PREV.br_startoff + PREV.br_blockcount - new_endoff;
1879 0 : RIGHT.br_startblock =
1880 0 : nullstartblock(xfs_bmap_worst_indlen(bma->ip,
1881 : RIGHT.br_blockcount));
1882 :
1883 : /* truncate PREV */
1884 0 : PREV.br_blockcount = new->br_startoff - PREV.br_startoff;
1885 0 : PREV.br_startblock =
1886 0 : nullstartblock(xfs_bmap_worst_indlen(bma->ip,
1887 : PREV.br_blockcount));
1888 0 : xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1889 :
1890 0 : xfs_iext_next(ifp, &bma->icur);
1891 0 : xfs_iext_insert(bma->ip, &bma->icur, &RIGHT, state);
1892 0 : xfs_iext_insert(bma->ip, &bma->icur, &LEFT, state);
1893 0 : ifp->if_nextents++;
1894 :
1895 0 : if (bma->cur == NULL)
1896 : rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1897 : else {
1898 0 : rval = XFS_ILOG_CORE;
1899 0 : error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1900 0 : if (error)
1901 0 : goto done;
1902 0 : if (XFS_IS_CORRUPT(mp, i != 0)) {
1903 0 : xfs_btree_mark_sick(bma->cur);
1904 0 : error = -EFSCORRUPTED;
1905 0 : goto done;
1906 : }
1907 0 : error = xfs_btree_insert(bma->cur, &i);
1908 0 : if (error)
1909 0 : goto done;
1910 0 : if (XFS_IS_CORRUPT(mp, i != 1)) {
1911 0 : xfs_btree_mark_sick(bma->cur);
1912 0 : error = -EFSCORRUPTED;
1913 0 : goto done;
1914 : }
1915 : }
1916 :
1917 0 : if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1918 0 : error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1919 : &bma->cur, 1, &tmp_rval, whichfork);
1920 0 : rval |= tmp_rval;
1921 0 : if (error)
1922 0 : goto done;
1923 : }
1924 :
1925 0 : da_new = startblockval(PREV.br_startblock) +
1926 0 : startblockval(RIGHT.br_startblock);
1927 0 : break;
1928 :
1929 0 : case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1930 : case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1931 : case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1932 : case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1933 : case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1934 : case BMAP_LEFT_CONTIG:
1935 : case BMAP_RIGHT_CONTIG:
1936 : /*
1937 : * These cases are all impossible.
1938 : */
1939 0 : ASSERT(0);
1940 : }
1941 :
1942 : /* add reverse mapping unless caller opted out */
1943 15729633 : if (!(bma->flags & XFS_BMAPI_NORMAP))
1944 15729618 : xfs_rmap_map_extent(bma->tp, bma->ip, whichfork, new);
1945 :
1946 : /* convert to a btree if necessary */
1947 15729636 : if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1948 315975 : int tmp_logflags; /* partial log flag return val */
1949 :
1950 315975 : ASSERT(bma->cur == NULL);
1951 315975 : error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1952 : &bma->cur, da_old > 0, &tmp_logflags,
1953 : whichfork);
1954 315975 : bma->logflags |= tmp_logflags;
1955 315975 : if (error)
1956 39 : goto done;
1957 : }
1958 :
1959 15729603 : if (da_new != da_old)
1960 15500538 : xfs_mod_delalloc(mp, (int64_t)da_new - da_old);
1961 :
1962 15729556 : if (bma->cur) {
1963 2804161 : da_new += bma->cur->bc_ino.allocated;
1964 2804161 : bma->cur->bc_ino.allocated = 0;
1965 : }
1966 :
1967 : /* adjust for changes in reserved delayed indirect blocks */
1968 15729556 : if (da_new != da_old) {
1969 15498176 : ASSERT(state == 0 || da_new < da_old);
1970 15498176 : error = xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new),
1971 : false);
1972 : }
1973 :
1974 15729589 : xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
1975 15729600 : done:
1976 15729600 : if (whichfork != XFS_COW_FORK)
1977 15612263 : bma->logflags |= rval;
1978 15729600 : return error;
1979 : #undef LEFT
1980 : #undef RIGHT
1981 : #undef PREV
1982 : }
1983 :
1984 : /*
1985 : * Convert an unwritten allocation to a real allocation or vice versa.
1986 : */
1987 : int /* error */
1988 26537967 : xfs_bmap_add_extent_unwritten_real(
1989 : struct xfs_trans *tp,
1990 : xfs_inode_t *ip, /* incore inode pointer */
1991 : int whichfork,
1992 : struct xfs_iext_cursor *icur,
1993 : struct xfs_btree_cur **curp, /* if *curp is null, not a btree */
1994 : xfs_bmbt_irec_t *new, /* new data to add to file extents */
1995 : int *logflagsp) /* inode logging flags */
1996 : {
1997 26537967 : struct xfs_btree_cur *cur; /* btree cursor */
1998 26537967 : int error; /* error return value */
1999 26537967 : int i; /* temp state */
2000 26537967 : struct xfs_ifork *ifp; /* inode fork pointer */
2001 26537967 : xfs_fileoff_t new_endoff; /* end offset of new entry */
2002 26537967 : xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
2003 : /* left is 0, right is 1, prev is 2 */
2004 26537967 : int rval=0; /* return value (logging flags) */
2005 26537967 : uint32_t state = xfs_bmap_fork_to_state(whichfork);
2006 26537967 : struct xfs_mount *mp = ip->i_mount;
2007 26537967 : struct xfs_bmbt_irec old;
2008 :
2009 26537967 : *logflagsp = 0;
2010 :
2011 26537967 : cur = *curp;
2012 26537967 : ifp = xfs_ifork_ptr(ip, whichfork);
2013 :
2014 26537814 : ASSERT(!isnullstartblock(new->br_startblock));
2015 :
2016 26537814 : XFS_STATS_INC(mp, xs_add_exlist);
2017 :
2018 : #define LEFT r[0]
2019 : #define RIGHT r[1]
2020 : #define PREV r[2]
2021 :
2022 : /*
2023 : * Set up a bunch of variables to make the tests simpler.
2024 : */
2025 26537814 : error = 0;
2026 26537814 : xfs_iext_get_extent(ifp, icur, &PREV);
2027 26537949 : ASSERT(new->br_state != PREV.br_state);
2028 26537949 : new_endoff = new->br_startoff + new->br_blockcount;
2029 26537949 : ASSERT(PREV.br_startoff <= new->br_startoff);
2030 26537949 : ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2031 :
2032 : /*
2033 : * Set flags determining what part of the previous oldext allocation
2034 : * extent is being replaced by a newext allocation.
2035 : */
2036 26537949 : if (PREV.br_startoff == new->br_startoff)
2037 24918245 : state |= BMAP_LEFT_FILLING;
2038 26537949 : if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2039 22336822 : state |= BMAP_RIGHT_FILLING;
2040 :
2041 : /*
2042 : * Check and set flags if this segment has a left neighbor.
2043 : * Don't set contiguous if the combined extent would be too large.
2044 : */
2045 26537949 : if (xfs_iext_peek_prev_extent(ifp, icur, &LEFT)) {
2046 19172722 : state |= BMAP_LEFT_VALID;
2047 19172722 : if (isnullstartblock(LEFT.br_startblock))
2048 115786 : state |= BMAP_LEFT_DELAY;
2049 : }
2050 :
2051 26536204 : if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2052 19057245 : LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2053 3634284 : LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2054 953095 : LEFT.br_state == new->br_state &&
2055 948867 : LEFT.br_blockcount + new->br_blockcount <= XFS_MAX_BMBT_EXTLEN)
2056 948867 : state |= BMAP_LEFT_CONTIG;
2057 :
2058 : /*
2059 : * Check and set flags if this segment has a right neighbor.
2060 : * Don't set contiguous if the combined extent would be too large.
2061 : * Also check for all-three-contiguous being too large.
2062 : */
2063 26536204 : if (xfs_iext_peek_next_extent(ifp, icur, &RIGHT)) {
2064 14564357 : state |= BMAP_RIGHT_VALID;
2065 14564357 : if (isnullstartblock(RIGHT.br_startblock))
2066 143701 : state |= BMAP_RIGHT_DELAY;
2067 : }
2068 :
2069 26537231 : if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2070 14420602 : new_endoff == RIGHT.br_startoff &&
2071 3036595 : new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2072 325374 : new->br_state == RIGHT.br_state &&
2073 325373 : new->br_blockcount + RIGHT.br_blockcount <= XFS_MAX_BMBT_EXTLEN &&
2074 325373 : ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2075 : BMAP_RIGHT_FILLING)) !=
2076 : (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2077 97624 : BMAP_RIGHT_FILLING) ||
2078 97624 : LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2079 : <= XFS_MAX_BMBT_EXTLEN))
2080 325372 : state |= BMAP_RIGHT_CONTIG;
2081 :
2082 : /*
2083 : * Switch out based on the FILLING and CONTIG state bits.
2084 : */
2085 26537231 : switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2086 : BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2087 97623 : case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2088 : BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2089 : /*
2090 : * Setting all of a previous oldext extent to newext.
2091 : * The left and right neighbors are both contiguous with new.
2092 : */
2093 97623 : LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount;
2094 :
2095 97623 : xfs_iext_remove(ip, icur, state);
2096 97623 : xfs_iext_remove(ip, icur, state);
2097 97623 : xfs_iext_prev(ifp, icur);
2098 97623 : xfs_iext_update_extent(ip, state, icur, &LEFT);
2099 97623 : ifp->if_nextents -= 2;
2100 97623 : if (cur == NULL)
2101 : rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2102 : else {
2103 45980 : rval = XFS_ILOG_CORE;
2104 45980 : error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i);
2105 45980 : if (error)
2106 0 : goto done;
2107 45980 : if (XFS_IS_CORRUPT(mp, i != 1)) {
2108 0 : xfs_btree_mark_sick(cur);
2109 0 : error = -EFSCORRUPTED;
2110 0 : goto done;
2111 : }
2112 45980 : if ((error = xfs_btree_delete(cur, &i)))
2113 0 : goto done;
2114 45980 : if (XFS_IS_CORRUPT(mp, i != 1)) {
2115 0 : xfs_btree_mark_sick(cur);
2116 0 : error = -EFSCORRUPTED;
2117 0 : goto done;
2118 : }
2119 45980 : if ((error = xfs_btree_decrement(cur, 0, &i)))
2120 0 : goto done;
2121 45980 : if (XFS_IS_CORRUPT(mp, i != 1)) {
2122 0 : xfs_btree_mark_sick(cur);
2123 0 : error = -EFSCORRUPTED;
2124 0 : goto done;
2125 : }
2126 45980 : if ((error = xfs_btree_delete(cur, &i)))
2127 0 : goto done;
2128 45980 : if (XFS_IS_CORRUPT(mp, i != 1)) {
2129 0 : xfs_btree_mark_sick(cur);
2130 0 : error = -EFSCORRUPTED;
2131 0 : goto done;
2132 : }
2133 45980 : if ((error = xfs_btree_decrement(cur, 0, &i)))
2134 0 : goto done;
2135 45980 : if (XFS_IS_CORRUPT(mp, i != 1)) {
2136 0 : xfs_btree_mark_sick(cur);
2137 0 : error = -EFSCORRUPTED;
2138 0 : goto done;
2139 : }
2140 45980 : error = xfs_bmbt_update(cur, &LEFT);
2141 45980 : if (error)
2142 0 : goto done;
2143 : }
2144 : break;
2145 :
2146 531586 : case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2147 : /*
2148 : * Setting all of a previous oldext extent to newext.
2149 : * The left neighbor is contiguous, the right is not.
2150 : */
2151 531586 : LEFT.br_blockcount += PREV.br_blockcount;
2152 :
2153 531586 : xfs_iext_remove(ip, icur, state);
2154 531586 : xfs_iext_prev(ifp, icur);
2155 531586 : xfs_iext_update_extent(ip, state, icur, &LEFT);
2156 531586 : ifp->if_nextents--;
2157 531586 : if (cur == NULL)
2158 : rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2159 : else {
2160 165921 : rval = XFS_ILOG_CORE;
2161 165921 : error = xfs_bmbt_lookup_eq(cur, &PREV, &i);
2162 165921 : if (error)
2163 0 : goto done;
2164 165921 : if (XFS_IS_CORRUPT(mp, i != 1)) {
2165 0 : xfs_btree_mark_sick(cur);
2166 0 : error = -EFSCORRUPTED;
2167 0 : goto done;
2168 : }
2169 165921 : if ((error = xfs_btree_delete(cur, &i)))
2170 0 : goto done;
2171 165921 : if (XFS_IS_CORRUPT(mp, i != 1)) {
2172 0 : xfs_btree_mark_sick(cur);
2173 0 : error = -EFSCORRUPTED;
2174 0 : goto done;
2175 : }
2176 165921 : if ((error = xfs_btree_decrement(cur, 0, &i)))
2177 0 : goto done;
2178 165921 : if (XFS_IS_CORRUPT(mp, i != 1)) {
2179 0 : xfs_btree_mark_sick(cur);
2180 0 : error = -EFSCORRUPTED;
2181 0 : goto done;
2182 : }
2183 165921 : error = xfs_bmbt_update(cur, &LEFT);
2184 165921 : if (error)
2185 0 : goto done;
2186 : }
2187 : break;
2188 :
2189 162514 : case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2190 : /*
2191 : * Setting all of a previous oldext extent to newext.
2192 : * The right neighbor is contiguous, the left is not.
2193 : */
2194 162514 : PREV.br_blockcount += RIGHT.br_blockcount;
2195 162514 : PREV.br_state = new->br_state;
2196 :
2197 162514 : xfs_iext_next(ifp, icur);
2198 162514 : xfs_iext_remove(ip, icur, state);
2199 162514 : xfs_iext_prev(ifp, icur);
2200 162514 : xfs_iext_update_extent(ip, state, icur, &PREV);
2201 162514 : ifp->if_nextents--;
2202 :
2203 162514 : if (cur == NULL)
2204 : rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2205 : else {
2206 32596 : rval = XFS_ILOG_CORE;
2207 32596 : error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i);
2208 32596 : if (error)
2209 0 : goto done;
2210 32596 : if (XFS_IS_CORRUPT(mp, i != 1)) {
2211 0 : xfs_btree_mark_sick(cur);
2212 0 : error = -EFSCORRUPTED;
2213 0 : goto done;
2214 : }
2215 32596 : if ((error = xfs_btree_delete(cur, &i)))
2216 0 : goto done;
2217 32596 : if (XFS_IS_CORRUPT(mp, i != 1)) {
2218 0 : xfs_btree_mark_sick(cur);
2219 0 : error = -EFSCORRUPTED;
2220 0 : goto done;
2221 : }
2222 32596 : if ((error = xfs_btree_decrement(cur, 0, &i)))
2223 0 : goto done;
2224 32596 : if (XFS_IS_CORRUPT(mp, i != 1)) {
2225 0 : xfs_btree_mark_sick(cur);
2226 0 : error = -EFSCORRUPTED;
2227 0 : goto done;
2228 : }
2229 32596 : error = xfs_bmbt_update(cur, &PREV);
2230 32596 : if (error)
2231 0 : goto done;
2232 : }
2233 : break;
2234 :
2235 20958502 : case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2236 : /*
2237 : * Setting all of a previous oldext extent to newext.
2238 : * Neither the left nor right neighbors are contiguous with
2239 : * the new one.
2240 : */
2241 20958502 : PREV.br_state = new->br_state;
2242 20958502 : xfs_iext_update_extent(ip, state, icur, &PREV);
2243 :
2244 20957970 : if (cur == NULL)
2245 : rval = XFS_ILOG_DEXT;
2246 : else {
2247 4923217 : rval = 0;
2248 4923217 : error = xfs_bmbt_lookup_eq(cur, new, &i);
2249 4923240 : if (error)
2250 0 : goto done;
2251 4923240 : if (XFS_IS_CORRUPT(mp, i != 1)) {
2252 0 : xfs_btree_mark_sick(cur);
2253 0 : error = -EFSCORRUPTED;
2254 0 : goto done;
2255 : }
2256 4923240 : error = xfs_bmbt_update(cur, &PREV);
2257 4923166 : if (error)
2258 0 : goto done;
2259 : }
2260 : break;
2261 :
2262 319658 : case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2263 : /*
2264 : * Setting the first part of a previous oldext extent to newext.
2265 : * The left neighbor is contiguous.
2266 : */
2267 319658 : LEFT.br_blockcount += new->br_blockcount;
2268 :
2269 319658 : old = PREV;
2270 319658 : PREV.br_startoff += new->br_blockcount;
2271 319658 : PREV.br_startblock += new->br_blockcount;
2272 319658 : PREV.br_blockcount -= new->br_blockcount;
2273 :
2274 319658 : xfs_iext_update_extent(ip, state, icur, &PREV);
2275 319658 : xfs_iext_prev(ifp, icur);
2276 319658 : xfs_iext_update_extent(ip, state, icur, &LEFT);
2277 :
2278 319658 : if (cur == NULL)
2279 : rval = XFS_ILOG_DEXT;
2280 : else {
2281 107168 : rval = 0;
2282 107168 : error = xfs_bmbt_lookup_eq(cur, &old, &i);
2283 107168 : if (error)
2284 0 : goto done;
2285 107168 : if (XFS_IS_CORRUPT(mp, i != 1)) {
2286 0 : xfs_btree_mark_sick(cur);
2287 0 : error = -EFSCORRUPTED;
2288 0 : goto done;
2289 : }
2290 107168 : error = xfs_bmbt_update(cur, &PREV);
2291 107168 : if (error)
2292 0 : goto done;
2293 107168 : error = xfs_btree_decrement(cur, 0, &i);
2294 107168 : if (error)
2295 0 : goto done;
2296 107168 : error = xfs_bmbt_update(cur, &LEFT);
2297 107168 : if (error)
2298 0 : goto done;
2299 : }
2300 : break;
2301 :
2302 2847647 : case BMAP_LEFT_FILLING:
2303 : /*
2304 : * Setting the first part of a previous oldext extent to newext.
2305 : * The left neighbor is not contiguous.
2306 : */
2307 2847647 : old = PREV;
2308 2847647 : PREV.br_startoff += new->br_blockcount;
2309 2847647 : PREV.br_startblock += new->br_blockcount;
2310 2847647 : PREV.br_blockcount -= new->br_blockcount;
2311 :
2312 2847647 : xfs_iext_update_extent(ip, state, icur, &PREV);
2313 2847631 : xfs_iext_insert(ip, icur, new, state);
2314 2847631 : ifp->if_nextents++;
2315 :
2316 2847631 : if (cur == NULL)
2317 : rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2318 : else {
2319 364572 : rval = XFS_ILOG_CORE;
2320 364572 : error = xfs_bmbt_lookup_eq(cur, &old, &i);
2321 364572 : if (error)
2322 0 : goto done;
2323 364572 : if (XFS_IS_CORRUPT(mp, i != 1)) {
2324 0 : xfs_btree_mark_sick(cur);
2325 0 : error = -EFSCORRUPTED;
2326 0 : goto done;
2327 : }
2328 364572 : error = xfs_bmbt_update(cur, &PREV);
2329 364572 : if (error)
2330 0 : goto done;
2331 364572 : cur->bc_rec.b = *new;
2332 364572 : if ((error = xfs_btree_insert(cur, &i)))
2333 0 : goto done;
2334 364572 : if (XFS_IS_CORRUPT(mp, i != 1)) {
2335 0 : xfs_btree_mark_sick(cur);
2336 0 : error = -EFSCORRUPTED;
2337 0 : goto done;
2338 : }
2339 : }
2340 : break;
2341 :
2342 65235 : case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2343 : /*
2344 : * Setting the last part of a previous oldext extent to newext.
2345 : * The right neighbor is contiguous with the new allocation.
2346 : */
2347 65235 : old = PREV;
2348 65235 : PREV.br_blockcount -= new->br_blockcount;
2349 :
2350 65235 : RIGHT.br_startoff = new->br_startoff;
2351 65235 : RIGHT.br_startblock = new->br_startblock;
2352 65235 : RIGHT.br_blockcount += new->br_blockcount;
2353 :
2354 65235 : xfs_iext_update_extent(ip, state, icur, &PREV);
2355 65235 : xfs_iext_next(ifp, icur);
2356 65235 : xfs_iext_update_extent(ip, state, icur, &RIGHT);
2357 :
2358 65235 : if (cur == NULL)
2359 : rval = XFS_ILOG_DEXT;
2360 : else {
2361 35222 : rval = 0;
2362 35222 : error = xfs_bmbt_lookup_eq(cur, &old, &i);
2363 35222 : if (error)
2364 0 : goto done;
2365 35222 : if (XFS_IS_CORRUPT(mp, i != 1)) {
2366 0 : xfs_btree_mark_sick(cur);
2367 0 : error = -EFSCORRUPTED;
2368 0 : goto done;
2369 : }
2370 35222 : error = xfs_bmbt_update(cur, &PREV);
2371 35222 : if (error)
2372 0 : goto done;
2373 35222 : error = xfs_btree_increment(cur, 0, &i);
2374 35222 : if (error)
2375 0 : goto done;
2376 35222 : error = xfs_bmbt_update(cur, &RIGHT);
2377 35222 : if (error)
2378 0 : goto done;
2379 : }
2380 : break;
2381 :
2382 520623 : case BMAP_RIGHT_FILLING:
2383 : /*
2384 : * Setting the last part of a previous oldext extent to newext.
2385 : * The right neighbor is not contiguous.
2386 : */
2387 520623 : old = PREV;
2388 520623 : PREV.br_blockcount -= new->br_blockcount;
2389 :
2390 520623 : xfs_iext_update_extent(ip, state, icur, &PREV);
2391 520622 : xfs_iext_next(ifp, icur);
2392 520623 : xfs_iext_insert(ip, icur, new, state);
2393 520622 : ifp->if_nextents++;
2394 :
2395 520622 : if (cur == NULL)
2396 : rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2397 : else {
2398 89708 : rval = XFS_ILOG_CORE;
2399 89708 : error = xfs_bmbt_lookup_eq(cur, &old, &i);
2400 89708 : if (error)
2401 0 : goto done;
2402 89708 : if (XFS_IS_CORRUPT(mp, i != 1)) {
2403 0 : xfs_btree_mark_sick(cur);
2404 0 : error = -EFSCORRUPTED;
2405 0 : goto done;
2406 : }
2407 89708 : error = xfs_bmbt_update(cur, &PREV);
2408 89708 : if (error)
2409 0 : goto done;
2410 89708 : error = xfs_bmbt_lookup_eq(cur, new, &i);
2411 89708 : if (error)
2412 0 : goto done;
2413 89708 : if (XFS_IS_CORRUPT(mp, i != 0)) {
2414 0 : xfs_btree_mark_sick(cur);
2415 0 : error = -EFSCORRUPTED;
2416 0 : goto done;
2417 : }
2418 89708 : if ((error = xfs_btree_insert(cur, &i)))
2419 0 : goto done;
2420 89708 : if (XFS_IS_CORRUPT(mp, i != 1)) {
2421 0 : xfs_btree_mark_sick(cur);
2422 0 : error = -EFSCORRUPTED;
2423 0 : goto done;
2424 : }
2425 : }
2426 : break;
2427 :
2428 1033843 : case 0:
2429 : /*
2430 : * Setting the middle part of a previous oldext extent to
2431 : * newext. Contiguity is impossible here.
2432 : * One extent becomes three extents.
2433 : */
2434 1033843 : old = PREV;
2435 1033843 : PREV.br_blockcount = new->br_startoff - PREV.br_startoff;
2436 :
2437 1033843 : r[0] = *new;
2438 1033843 : r[1].br_startoff = new_endoff;
2439 1033843 : r[1].br_blockcount =
2440 1033843 : old.br_startoff + old.br_blockcount - new_endoff;
2441 1033843 : r[1].br_startblock = new->br_startblock + new->br_blockcount;
2442 1033843 : r[1].br_state = PREV.br_state;
2443 :
2444 1033843 : xfs_iext_update_extent(ip, state, icur, &PREV);
2445 1033808 : xfs_iext_next(ifp, icur);
2446 1033545 : xfs_iext_insert(ip, icur, &r[1], state);
2447 1033903 : xfs_iext_insert(ip, icur, &r[0], state);
2448 1033372 : ifp->if_nextents += 2;
2449 :
2450 1033372 : if (cur == NULL)
2451 : rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2452 : else {
2453 185135 : rval = XFS_ILOG_CORE;
2454 185135 : error = xfs_bmbt_lookup_eq(cur, &old, &i);
2455 185135 : if (error)
2456 0 : goto done;
2457 185135 : if (XFS_IS_CORRUPT(mp, i != 1)) {
2458 0 : xfs_btree_mark_sick(cur);
2459 0 : error = -EFSCORRUPTED;
2460 0 : goto done;
2461 : }
2462 : /* new right extent - oldext */
2463 185135 : error = xfs_bmbt_update(cur, &r[1]);
2464 185135 : if (error)
2465 0 : goto done;
2466 : /* new left extent - oldext */
2467 185135 : cur->bc_rec.b = PREV;
2468 185135 : if ((error = xfs_btree_insert(cur, &i)))
2469 0 : goto done;
2470 185135 : if (XFS_IS_CORRUPT(mp, i != 1)) {
2471 0 : xfs_btree_mark_sick(cur);
2472 0 : error = -EFSCORRUPTED;
2473 0 : goto done;
2474 : }
2475 : /*
2476 : * Reset the cursor to the position of the new extent
2477 : * we are about to insert as we can't trust it after
2478 : * the previous insert.
2479 : */
2480 185135 : error = xfs_bmbt_lookup_eq(cur, new, &i);
2481 185135 : if (error)
2482 0 : goto done;
2483 185135 : if (XFS_IS_CORRUPT(mp, i != 0)) {
2484 0 : xfs_btree_mark_sick(cur);
2485 0 : error = -EFSCORRUPTED;
2486 0 : goto done;
2487 : }
2488 : /* new middle extent - newext */
2489 185135 : if ((error = xfs_btree_insert(cur, &i)))
2490 0 : goto done;
2491 185135 : if (XFS_IS_CORRUPT(mp, i != 1)) {
2492 0 : xfs_btree_mark_sick(cur);
2493 0 : error = -EFSCORRUPTED;
2494 0 : goto done;
2495 : }
2496 : }
2497 : break;
2498 :
2499 0 : case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2500 : case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2501 : case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2502 : case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2503 : case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2504 : case BMAP_LEFT_CONTIG:
2505 : case BMAP_RIGHT_CONTIG:
2506 : /*
2507 : * These cases are all impossible.
2508 : */
2509 0 : ASSERT(0);
2510 : }
2511 :
2512 : /* update reverse mappings */
2513 26536160 : xfs_rmap_convert_extent(mp, tp, ip, whichfork, new);
2514 :
2515 : /* convert to a btree if necessary */
2516 26536154 : if (xfs_bmap_needs_btree(ip, whichfork)) {
2517 101820 : int tmp_logflags; /* partial log flag return val */
2518 :
2519 101820 : ASSERT(cur == NULL);
2520 101820 : error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
2521 : &tmp_logflags, whichfork);
2522 101820 : *logflagsp |= tmp_logflags;
2523 101820 : if (error)
2524 7 : goto done;
2525 : }
2526 :
2527 : /* clear out the allocated field, done with it now in any case. */
2528 26535843 : if (cur) {
2529 6051252 : cur->bc_ino.allocated = 0;
2530 6051252 : *curp = cur;
2531 : }
2532 :
2533 26535843 : xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
2534 26535600 : done:
2535 26535600 : *logflagsp |= rval;
2536 26535600 : return error;
2537 : #undef LEFT
2538 : #undef RIGHT
2539 : #undef PREV
2540 : }
2541 :
2542 : /*
2543 : * Convert a hole to a delayed allocation.
2544 : */
2545 : STATIC void
2546 18061755 : xfs_bmap_add_extent_hole_delay(
2547 : xfs_inode_t *ip, /* incore inode pointer */
2548 : int whichfork,
2549 : struct xfs_iext_cursor *icur,
2550 : xfs_bmbt_irec_t *new) /* new data to add to file extents */
2551 : {
2552 18061755 : struct xfs_ifork *ifp; /* inode fork pointer */
2553 18061755 : xfs_bmbt_irec_t left; /* left neighbor extent entry */
2554 18061755 : xfs_filblks_t newlen=0; /* new indirect size */
2555 18061755 : xfs_filblks_t oldlen=0; /* old indirect size */
2556 18061755 : xfs_bmbt_irec_t right; /* right neighbor extent entry */
2557 18061755 : uint32_t state = xfs_bmap_fork_to_state(whichfork);
2558 18061755 : xfs_filblks_t temp; /* temp for indirect calculations */
2559 :
2560 18061755 : ifp = xfs_ifork_ptr(ip, whichfork);
2561 18076645 : ASSERT(isnullstartblock(new->br_startblock));
2562 :
2563 : /*
2564 : * Check and set flags if this segment has a left neighbor
2565 : */
2566 18076645 : if (xfs_iext_peek_prev_extent(ifp, icur, &left)) {
2567 12502527 : state |= BMAP_LEFT_VALID;
2568 12502527 : if (isnullstartblock(left.br_startblock))
2569 2465336 : state |= BMAP_LEFT_DELAY;
2570 : }
2571 :
2572 : /*
2573 : * Check and set flags if the current (right) segment exists.
2574 : * If it doesn't exist, we're converting the hole at end-of-file.
2575 : */
2576 18074613 : if (xfs_iext_get_extent(ifp, icur, &right)) {
2577 8475871 : state |= BMAP_RIGHT_VALID;
2578 8475871 : if (isnullstartblock(right.br_startblock))
2579 1383623 : state |= BMAP_RIGHT_DELAY;
2580 : }
2581 :
2582 : /*
2583 : * Set contiguity flags on the left and right neighbors.
2584 : * Don't let extents get too large, even if the pieces are contiguous.
2585 : */
2586 18074765 : if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2587 2465325 : left.br_startoff + left.br_blockcount == new->br_startoff &&
2588 1847404 : left.br_blockcount + new->br_blockcount <= XFS_MAX_BMBT_EXTLEN)
2589 1847403 : state |= BMAP_LEFT_CONTIG;
2590 :
2591 18074765 : if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2592 1383623 : new->br_startoff + new->br_blockcount == right.br_startoff &&
2593 89421 : new->br_blockcount + right.br_blockcount <= XFS_MAX_BMBT_EXTLEN &&
2594 85125 : (!(state & BMAP_LEFT_CONTIG) ||
2595 19661 : (left.br_blockcount + new->br_blockcount +
2596 : right.br_blockcount <= XFS_MAX_BMBT_EXTLEN)))
2597 85125 : state |= BMAP_RIGHT_CONTIG;
2598 :
2599 : /*
2600 : * Switch out based on the contiguity flags.
2601 : */
2602 18074765 : switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2603 19661 : case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2604 : /*
2605 : * New allocation is contiguous with delayed allocations
2606 : * on the left and on the right.
2607 : * Merge all three into a single extent record.
2608 : */
2609 19661 : temp = left.br_blockcount + new->br_blockcount +
2610 19661 : right.br_blockcount;
2611 :
2612 19661 : oldlen = startblockval(left.br_startblock) +
2613 19661 : startblockval(new->br_startblock) +
2614 19661 : startblockval(right.br_startblock);
2615 19661 : newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2616 : oldlen);
2617 19661 : left.br_startblock = nullstartblock(newlen);
2618 19661 : left.br_blockcount = temp;
2619 :
2620 19661 : xfs_iext_remove(ip, icur, state);
2621 19661 : xfs_iext_prev(ifp, icur);
2622 19661 : xfs_iext_update_extent(ip, state, icur, &left);
2623 19661 : break;
2624 :
2625 1827726 : case BMAP_LEFT_CONTIG:
2626 : /*
2627 : * New allocation is contiguous with a delayed allocation
2628 : * on the left.
2629 : * Merge the new allocation with the left neighbor.
2630 : */
2631 1827726 : temp = left.br_blockcount + new->br_blockcount;
2632 :
2633 1827726 : oldlen = startblockval(left.br_startblock) +
2634 1827726 : startblockval(new->br_startblock);
2635 1827726 : newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2636 : oldlen);
2637 1827741 : left.br_blockcount = temp;
2638 1827741 : left.br_startblock = nullstartblock(newlen);
2639 :
2640 1827741 : xfs_iext_prev(ifp, icur);
2641 1827703 : xfs_iext_update_extent(ip, state, icur, &left);
2642 1827703 : break;
2643 :
2644 65464 : case BMAP_RIGHT_CONTIG:
2645 : /*
2646 : * New allocation is contiguous with a delayed allocation
2647 : * on the right.
2648 : * Merge the new allocation with the right neighbor.
2649 : */
2650 65464 : temp = new->br_blockcount + right.br_blockcount;
2651 65464 : oldlen = startblockval(new->br_startblock) +
2652 65464 : startblockval(right.br_startblock);
2653 65464 : newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2654 : oldlen);
2655 65464 : right.br_startoff = new->br_startoff;
2656 65464 : right.br_startblock = nullstartblock(newlen);
2657 65464 : right.br_blockcount = temp;
2658 65464 : xfs_iext_update_extent(ip, state, icur, &right);
2659 65464 : break;
2660 :
2661 16161914 : case 0:
2662 : /*
2663 : * New allocation is not contiguous with another
2664 : * delayed allocation.
2665 : * Insert a new entry.
2666 : */
2667 16161914 : oldlen = newlen = 0;
2668 16161914 : xfs_iext_insert(ip, icur, new, state);
2669 16161914 : break;
2670 : }
2671 18074807 : if (oldlen != newlen) {
2672 1912811 : ASSERT(oldlen > newlen);
2673 1912811 : xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen),
2674 : false);
2675 : /*
2676 : * Nothing to do for disk quota accounting here.
2677 : */
2678 1912812 : xfs_mod_delalloc(ip->i_mount, (int64_t)newlen - oldlen);
2679 : }
2680 18055938 : }
2681 :
2682 : /*
2683 : * Convert a hole to a real allocation.
2684 : */
2685 : STATIC int /* error */
2686 101831260 : xfs_bmap_add_extent_hole_real(
2687 : struct xfs_trans *tp,
2688 : struct xfs_inode *ip,
2689 : int whichfork,
2690 : struct xfs_iext_cursor *icur,
2691 : struct xfs_btree_cur **curp,
2692 : struct xfs_bmbt_irec *new,
2693 : int *logflagsp,
2694 : uint32_t flags)
2695 : {
2696 101831260 : struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
2697 101831611 : struct xfs_mount *mp = ip->i_mount;
2698 101831611 : struct xfs_btree_cur *cur = *curp;
2699 101831611 : int error; /* error return value */
2700 101831611 : int i; /* temp state */
2701 101831611 : xfs_bmbt_irec_t left; /* left neighbor extent entry */
2702 101831611 : xfs_bmbt_irec_t right; /* right neighbor extent entry */
2703 101831611 : int rval=0; /* return value (logging flags) */
2704 101831611 : uint32_t state = xfs_bmap_fork_to_state(whichfork);
2705 101831611 : struct xfs_bmbt_irec old;
2706 :
2707 101831611 : ASSERT(!isnullstartblock(new->br_startblock));
2708 101831611 : ASSERT(!cur || !(cur->bc_ino.flags & XFS_BTCUR_BMBT_WASDEL));
2709 :
2710 101831611 : XFS_STATS_INC(mp, xs_add_exlist);
2711 :
2712 : /*
2713 : * Check and set flags if this segment has a left neighbor.
2714 : */
2715 101831611 : if (xfs_iext_peek_prev_extent(ifp, icur, &left)) {
2716 68850869 : state |= BMAP_LEFT_VALID;
2717 68850869 : if (isnullstartblock(left.br_startblock))
2718 138357 : state |= BMAP_LEFT_DELAY;
2719 : }
2720 :
2721 : /*
2722 : * Check and set flags if this segment has a current value.
2723 : * Not true if we're inserting into the "hole" at eof.
2724 : */
2725 101832260 : if (xfs_iext_get_extent(ifp, icur, &right)) {
2726 12996301 : state |= BMAP_RIGHT_VALID;
2727 12996301 : if (isnullstartblock(right.br_startblock))
2728 205596 : state |= BMAP_RIGHT_DELAY;
2729 : }
2730 :
2731 : /*
2732 : * We're inserting a real allocation between "left" and "right".
2733 : * Set the contiguity flags. Don't let extents get too large.
2734 : */
2735 101830671 : if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2736 68712466 : left.br_startoff + left.br_blockcount == new->br_startoff &&
2737 50480494 : left.br_startblock + left.br_blockcount == new->br_startblock &&
2738 5322312 : left.br_state == new->br_state &&
2739 4856345 : left.br_blockcount + new->br_blockcount <= XFS_MAX_BMBT_EXTLEN)
2740 4856332 : state |= BMAP_LEFT_CONTIG;
2741 :
2742 101830671 : if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2743 12790647 : new->br_startoff + new->br_blockcount == right.br_startoff &&
2744 4368886 : new->br_startblock + new->br_blockcount == right.br_startblock &&
2745 653036 : new->br_state == right.br_state &&
2746 554557 : new->br_blockcount + right.br_blockcount <= XFS_MAX_BMBT_EXTLEN &&
2747 554557 : (!(state & BMAP_LEFT_CONTIG) ||
2748 344063 : left.br_blockcount + new->br_blockcount +
2749 : right.br_blockcount <= XFS_MAX_BMBT_EXTLEN))
2750 554557 : state |= BMAP_RIGHT_CONTIG;
2751 :
2752 101830671 : error = 0;
2753 : /*
2754 : * Select which case we're in here, and implement it.
2755 : */
2756 101830671 : switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2757 344063 : case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2758 : /*
2759 : * New allocation is contiguous with real allocations on the
2760 : * left and on the right.
2761 : * Merge all three into a single extent record.
2762 : */
2763 344063 : left.br_blockcount += new->br_blockcount + right.br_blockcount;
2764 :
2765 344063 : xfs_iext_remove(ip, icur, state);
2766 344063 : xfs_iext_prev(ifp, icur);
2767 344063 : xfs_iext_update_extent(ip, state, icur, &left);
2768 344063 : ifp->if_nextents--;
2769 :
2770 344063 : if (cur == NULL) {
2771 60244 : rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2772 : } else {
2773 283849 : rval = XFS_ILOG_CORE;
2774 283849 : error = xfs_bmbt_lookup_eq(cur, &right, &i);
2775 283849 : if (error)
2776 0 : goto done;
2777 283849 : if (XFS_IS_CORRUPT(mp, i != 1)) {
2778 0 : xfs_btree_mark_sick(cur);
2779 0 : error = -EFSCORRUPTED;
2780 0 : goto done;
2781 : }
2782 283849 : error = xfs_btree_delete(cur, &i);
2783 283849 : if (error)
2784 0 : goto done;
2785 283849 : if (XFS_IS_CORRUPT(mp, i != 1)) {
2786 0 : xfs_btree_mark_sick(cur);
2787 0 : error = -EFSCORRUPTED;
2788 0 : goto done;
2789 : }
2790 283849 : error = xfs_btree_decrement(cur, 0, &i);
2791 283849 : if (error)
2792 0 : goto done;
2793 283849 : if (XFS_IS_CORRUPT(mp, i != 1)) {
2794 0 : xfs_btree_mark_sick(cur);
2795 0 : error = -EFSCORRUPTED;
2796 0 : goto done;
2797 : }
2798 283849 : error = xfs_bmbt_update(cur, &left);
2799 283849 : if (error)
2800 0 : goto done;
2801 : }
2802 : break;
2803 :
2804 4512269 : case BMAP_LEFT_CONTIG:
2805 : /*
2806 : * New allocation is contiguous with a real allocation
2807 : * on the left.
2808 : * Merge the new allocation with the left neighbor.
2809 : */
2810 4512269 : old = left;
2811 4512269 : left.br_blockcount += new->br_blockcount;
2812 :
2813 4512269 : xfs_iext_prev(ifp, icur);
2814 4512269 : xfs_iext_update_extent(ip, state, icur, &left);
2815 :
2816 4512269 : if (cur == NULL) {
2817 3676273 : rval = xfs_ilog_fext(whichfork);
2818 : } else {
2819 835996 : rval = 0;
2820 835996 : error = xfs_bmbt_lookup_eq(cur, &old, &i);
2821 835996 : if (error)
2822 0 : goto done;
2823 835996 : if (XFS_IS_CORRUPT(mp, i != 1)) {
2824 0 : xfs_btree_mark_sick(cur);
2825 0 : error = -EFSCORRUPTED;
2826 0 : goto done;
2827 : }
2828 835996 : error = xfs_bmbt_update(cur, &left);
2829 835996 : if (error)
2830 0 : goto done;
2831 : }
2832 : break;
2833 :
2834 210494 : case BMAP_RIGHT_CONTIG:
2835 : /*
2836 : * New allocation is contiguous with a real allocation
2837 : * on the right.
2838 : * Merge the new allocation with the right neighbor.
2839 : */
2840 210494 : old = right;
2841 :
2842 210494 : right.br_startoff = new->br_startoff;
2843 210494 : right.br_startblock = new->br_startblock;
2844 210494 : right.br_blockcount += new->br_blockcount;
2845 210494 : xfs_iext_update_extent(ip, state, icur, &right);
2846 :
2847 210494 : if (cur == NULL) {
2848 67201 : rval = xfs_ilog_fext(whichfork);
2849 : } else {
2850 143293 : rval = 0;
2851 143293 : error = xfs_bmbt_lookup_eq(cur, &old, &i);
2852 143293 : if (error)
2853 0 : goto done;
2854 143293 : if (XFS_IS_CORRUPT(mp, i != 1)) {
2855 0 : xfs_btree_mark_sick(cur);
2856 0 : error = -EFSCORRUPTED;
2857 0 : goto done;
2858 : }
2859 143293 : error = xfs_bmbt_update(cur, &right);
2860 143293 : if (error)
2861 0 : goto done;
2862 : }
2863 : break;
2864 :
2865 96763845 : case 0:
2866 : /*
2867 : * New allocation is not contiguous with another
2868 : * real allocation.
2869 : * Insert a new entry.
2870 : */
2871 96763845 : xfs_iext_insert(ip, icur, new, state);
2872 96761327 : ifp->if_nextents++;
2873 :
2874 96761327 : if (cur == NULL) {
2875 45532413 : rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2876 : } else {
2877 55867746 : rval = XFS_ILOG_CORE;
2878 55867746 : error = xfs_bmbt_lookup_eq(cur, new, &i);
2879 55867805 : if (error)
2880 994 : goto done;
2881 55866811 : if (XFS_IS_CORRUPT(mp, i != 0)) {
2882 0 : xfs_btree_mark_sick(cur);
2883 0 : error = -EFSCORRUPTED;
2884 0 : goto done;
2885 : }
2886 55866811 : error = xfs_btree_insert(cur, &i);
2887 55866818 : if (error)
2888 118 : goto done;
2889 55866700 : if (XFS_IS_CORRUPT(mp, i != 1)) {
2890 0 : xfs_btree_mark_sick(cur);
2891 0 : error = -EFSCORRUPTED;
2892 0 : goto done;
2893 : }
2894 : }
2895 : break;
2896 : }
2897 :
2898 : /* add reverse mapping unless caller opted out */
2899 101827107 : if (!(flags & XFS_BMAPI_NORMAP))
2900 101825925 : xfs_rmap_map_extent(tp, ip, whichfork, new);
2901 :
2902 : /* convert to a btree if necessary */
2903 101829137 : if (xfs_bmap_needs_btree(ip, whichfork)) {
2904 379283 : int tmp_logflags; /* partial log flag return val */
2905 :
2906 379283 : ASSERT(cur == NULL);
2907 379283 : error = xfs_bmap_extents_to_btree(tp, ip, curp, 0,
2908 : &tmp_logflags, whichfork);
2909 379283 : *logflagsp |= tmp_logflags;
2910 379283 : cur = *curp;
2911 379283 : if (error)
2912 7 : goto done;
2913 : }
2914 :
2915 : /* clear out the allocated field, done with it now in any case. */
2916 101827507 : if (cur)
2917 57509063 : cur->bc_ino.allocated = 0;
2918 :
2919 101827507 : xfs_bmap_check_leaf_extents(cur, ip, whichfork);
2920 101826942 : done:
2921 101826942 : *logflagsp |= rval;
2922 101826942 : return error;
2923 : }
2924 :
2925 : /*
2926 : * Functions used in the extent read, allocate and remove paths
2927 : */
2928 :
2929 : /*
2930 : * Adjust the size of the new extent based on i_extsize and rt extsize.
2931 : */
2932 : int
2933 11738938 : xfs_bmap_extsize_align(
2934 : xfs_mount_t *mp,
2935 : xfs_bmbt_irec_t *gotp, /* next extent pointer */
2936 : xfs_bmbt_irec_t *prevp, /* previous extent pointer */
2937 : xfs_extlen_t extsz, /* align to this extent size */
2938 : int rt, /* is this a realtime inode? */
2939 : int eof, /* is extent at end-of-file? */
2940 : int delay, /* creating delalloc extent? */
2941 : int convert, /* overwriting unwritten extent? */
2942 : xfs_fileoff_t *offp, /* in/out: aligned offset */
2943 : xfs_extlen_t *lenp) /* in/out: aligned length */
2944 : {
2945 11738938 : xfs_fileoff_t orig_off; /* original offset */
2946 11738938 : xfs_extlen_t orig_alen; /* original length */
2947 11738938 : xfs_fileoff_t orig_end; /* original off+len */
2948 11738938 : xfs_fileoff_t nexto; /* next file offset */
2949 11738938 : xfs_fileoff_t prevo; /* previous file offset */
2950 11738938 : xfs_fileoff_t align_off; /* temp for offset */
2951 11738938 : xfs_extlen_t align_alen; /* temp for length */
2952 11738938 : xfs_extlen_t temp; /* temp for calculations */
2953 :
2954 11738938 : if (convert)
2955 : return 0;
2956 :
2957 11738938 : orig_off = align_off = *offp;
2958 11738938 : orig_alen = align_alen = *lenp;
2959 11738938 : orig_end = orig_off + orig_alen;
2960 :
2961 : /*
2962 : * If this request overlaps an existing extent, then don't
2963 : * attempt to perform any additional alignment.
2964 : */
2965 11738938 : if (!delay && !eof &&
2966 4693790 : (orig_off >= gotp->br_startoff) &&
2967 117376 : (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2968 : return 0;
2969 : }
2970 :
2971 : /*
2972 : * If the file offset is unaligned vs. the extent size
2973 : * we need to align it. This will be possible unless
2974 : * the file was previously written with a kernel that didn't
2975 : * perform this alignment, or if a truncate shot us in the
2976 : * foot.
2977 : */
2978 11621562 : div_u64_rem(orig_off, extsz, &temp);
2979 11621562 : if (temp) {
2980 300065 : align_alen += temp;
2981 300065 : align_off -= temp;
2982 : }
2983 :
2984 : /* Same adjustment for the end of the requested area. */
2985 11621562 : temp = (align_alen % extsz);
2986 11621562 : if (temp)
2987 337908 : align_alen += extsz - temp;
2988 :
2989 : /*
2990 : * For large extent hint sizes, the aligned extent might be larger than
2991 : * XFS_BMBT_MAX_EXTLEN. In that case, reduce the size by an extsz so
2992 : * that it pulls the length back under XFS_BMBT_MAX_EXTLEN. The outer
2993 : * allocation loops handle short allocation just fine, so it is safe to
2994 : * do this. We only want to do it when we are forced to, though, because
2995 : * it means more allocation operations are required.
2996 : */
2997 11621565 : while (align_alen > XFS_MAX_BMBT_EXTLEN)
2998 3 : align_alen -= extsz;
2999 11621562 : ASSERT(align_alen <= XFS_MAX_BMBT_EXTLEN);
3000 :
3001 : /*
3002 : * If the previous block overlaps with this proposed allocation
3003 : * then move the start forward without adjusting the length.
3004 : */
3005 11621562 : if (prevp->br_startoff != NULLFILEOFF) {
3006 9990106 : if (prevp->br_startblock == HOLESTARTBLOCK)
3007 : prevo = prevp->br_startoff;
3008 : else
3009 9990106 : prevo = prevp->br_startoff + prevp->br_blockcount;
3010 : } else
3011 : prevo = 0;
3012 11621562 : if (align_off != orig_off && align_off < prevo)
3013 5110 : align_off = prevo;
3014 : /*
3015 : * If the next block overlaps with this proposed allocation
3016 : * then move the start back without adjusting the length,
3017 : * but not before offset 0.
3018 : * This may of course make the start overlap previous block,
3019 : * and if we hit the offset 0 limit then the next block
3020 : * can still overlap too.
3021 : */
3022 11621562 : if (!eof && gotp->br_startoff != NULLFILEOFF) {
3023 4631094 : if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
3024 4576472 : (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
3025 0 : nexto = gotp->br_startoff + gotp->br_blockcount;
3026 : else
3027 : nexto = gotp->br_startoff;
3028 : } else
3029 : nexto = NULLFILEOFF;
3030 11621562 : if (!eof &&
3031 4631065 : align_off + align_alen != orig_end &&
3032 : align_off + align_alen > nexto)
3033 8331 : align_off = nexto > align_alen ? nexto - align_alen : 0;
3034 : /*
3035 : * If we're now overlapping the next or previous extent that
3036 : * means we can't fit an extsz piece in this hole. Just move
3037 : * the start forward to the first valid spot and set
3038 : * the length so we hit the end.
3039 : */
3040 11621562 : if (align_off != orig_off && align_off < prevo)
3041 2718 : align_off = prevo;
3042 11621562 : if (align_off + align_alen != orig_end &&
3043 : align_off + align_alen > nexto &&
3044 : nexto != NULLFILEOFF) {
3045 3524 : ASSERT(nexto > prevo);
3046 3524 : align_alen = nexto - align_off;
3047 : }
3048 :
3049 : /*
3050 : * If realtime, and the result isn't a multiple of the realtime
3051 : * extent size we need to remove blocks until it is.
3052 : */
3053 11621562 : if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
3054 : /*
3055 : * We're not covering the original request, or
3056 : * we won't be able to once we fix the length.
3057 : */
3058 0 : if (orig_off < align_off ||
3059 0 : orig_end > align_off + align_alen ||
3060 0 : align_alen - temp < orig_alen)
3061 : return -EINVAL;
3062 : /*
3063 : * Try to fix it by moving the start up.
3064 : */
3065 0 : if (align_off + temp <= orig_off) {
3066 : align_alen -= temp;
3067 : align_off += temp;
3068 : }
3069 : /*
3070 : * Try to fix it by moving the end in.
3071 : */
3072 0 : else if (align_off + align_alen - temp >= orig_end)
3073 : align_alen -= temp;
3074 : /*
3075 : * Set the start to the minimum then trim the length.
3076 : */
3077 : else {
3078 0 : align_alen -= orig_off - align_off;
3079 0 : align_off = orig_off;
3080 0 : align_alen -= align_alen % mp->m_sb.sb_rextsize;
3081 : }
3082 : /*
3083 : * Result doesn't cover the request, fail it.
3084 : */
3085 0 : if (orig_off < align_off || orig_end > align_off + align_alen)
3086 : return -EINVAL;
3087 : } else {
3088 11621562 : ASSERT(orig_off >= align_off);
3089 : /* see XFS_BMBT_MAX_EXTLEN handling above */
3090 11621562 : ASSERT(orig_end <= align_off + align_alen ||
3091 : align_alen + extsz > XFS_MAX_BMBT_EXTLEN);
3092 : }
3093 :
3094 : #ifdef DEBUG
3095 11621562 : if (!eof && gotp->br_startoff != NULLFILEOFF)
3096 4631152 : ASSERT(align_off + align_alen <= gotp->br_startoff);
3097 11621562 : if (prevp->br_startoff != NULLFILEOFF)
3098 9990509 : ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3099 : #endif
3100 :
3101 11621562 : *lenp = align_alen;
3102 11621562 : *offp = align_off;
3103 11621562 : return 0;
3104 : }
3105 :
3106 : #define XFS_ALLOC_GAP_UNITS 4
3107 :
3108 : void
3109 63895157 : xfs_bmap_adjacent(
3110 : struct xfs_bmalloca *ap) /* bmap alloc argument struct */
3111 : {
3112 63895157 : xfs_fsblock_t adjust; /* adjustment to block numbers */
3113 63895157 : xfs_mount_t *mp; /* mount point structure */
3114 63895157 : int rt; /* true if inode is realtime */
3115 :
3116 : #define ISVALID(x,y) \
3117 : (rt ? \
3118 : (x) < mp->m_sb.sb_rblocks : \
3119 : XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3120 : XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3121 : XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3122 :
3123 63895157 : mp = ap->ip->i_mount;
3124 63895157 : rt = XFS_IS_REALTIME_INODE(ap->ip) &&
3125 9457760 : (ap->datatype & XFS_ALLOC_USERDATA);
3126 : /*
3127 : * If allocating at eof, and there's a previous real block,
3128 : * try to use its last block as our starting point.
3129 : */
3130 63895157 : if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3131 11968636 : !isnullstartblock(ap->prev.br_startblock) &&
3132 7381243 : ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3133 : ap->prev.br_startblock)) {
3134 7381052 : ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3135 : /*
3136 : * Adjust for the gap between prevp and us.
3137 : */
3138 7381052 : adjust = ap->offset -
3139 7381052 : (ap->prev.br_startoff + ap->prev.br_blockcount);
3140 10132785 : if (adjust &&
3141 2751733 : ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3142 2742208 : ap->blkno += adjust;
3143 : }
3144 : /*
3145 : * If not at eof, then compare the two neighbor blocks.
3146 : * Figure out whether either one gives us a good starting point,
3147 : * and pick the better one.
3148 : */
3149 56514105 : else if (!ap->eof) {
3150 24853765 : xfs_fsblock_t gotbno; /* right side block number */
3151 24853765 : xfs_fsblock_t gotdiff=0; /* right side difference */
3152 24853765 : xfs_fsblock_t prevbno; /* left side block number */
3153 24853765 : xfs_fsblock_t prevdiff=0; /* left side difference */
3154 :
3155 : /*
3156 : * If there's a previous (left) block, select a requested
3157 : * start block based on it.
3158 : */
3159 43403685 : if (ap->prev.br_startoff != NULLFILEOFF &&
3160 18658346 : !isnullstartblock(ap->prev.br_startblock) &&
3161 18549983 : (prevbno = ap->prev.br_startblock +
3162 33324106 : ap->prev.br_blockcount) &&
3163 18549920 : ISVALID(prevbno, ap->prev.br_startblock)) {
3164 : /*
3165 : * Calculate gap to end of previous block.
3166 : */
3167 18549813 : adjust = prevdiff = ap->offset -
3168 18549813 : (ap->prev.br_startoff +
3169 : ap->prev.br_blockcount);
3170 : /*
3171 : * Figure the startblock based on the previous block's
3172 : * end and the gap size.
3173 : * Heuristic!
3174 : * If the gap is large relative to the piece we're
3175 : * allocating, or using it gives us an invalid block
3176 : * number, then just use the end of the previous block.
3177 : */
3178 21882668 : if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3179 6016578 : ISVALID(prevbno + prevdiff,
3180 : ap->prev.br_startblock))
3181 6016529 : prevbno += adjust;
3182 : else
3183 12533284 : prevdiff += adjust;
3184 : }
3185 : /*
3186 : * No previous block or can't follow it, just default.
3187 : */
3188 : else
3189 : prevbno = NULLFSBLOCK;
3190 : /*
3191 : * If there's a following (right) block, select a requested
3192 : * start block based on it.
3193 : */
3194 24853765 : if (!isnullstartblock(ap->got.br_startblock)) {
3195 : /*
3196 : * Calculate gap to start of next block.
3197 : */
3198 9191715 : adjust = gotdiff = ap->got.br_startoff - ap->offset;
3199 : /*
3200 : * Figure the startblock based on the next block's
3201 : * start and the gap size.
3202 : */
3203 9191715 : gotbno = ap->got.br_startblock;
3204 : /*
3205 : * Heuristic!
3206 : * If the gap is large relative to the piece we're
3207 : * allocating, or using it gives us an invalid block
3208 : * number, then just use the start of the next block
3209 : * offset by our length.
3210 : */
3211 10662533 : if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3212 3856996 : ISVALID(gotbno - gotdiff, gotbno))
3213 3831752 : gotbno -= adjust;
3214 5359963 : else if (ISVALID(gotbno - ap->length, gotbno)) {
3215 5330891 : gotbno -= ap->length;
3216 5330891 : gotdiff += adjust - ap->length;
3217 : } else
3218 29072 : gotdiff += adjust;
3219 : }
3220 : /*
3221 : * No next block, just default.
3222 : */
3223 : else
3224 : gotbno = NULLFSBLOCK;
3225 : /*
3226 : * If both valid, pick the better one, else the only good
3227 : * one, else ap->blkno is already set (to 0 or the inode block).
3228 : */
3229 24853765 : if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3230 10812691 : ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3231 16696549 : else if (prevbno != NULLFSBLOCK)
3232 10392744 : ap->blkno = prevbno;
3233 6303805 : else if (gotbno != NULLFSBLOCK)
3234 1034461 : ap->blkno = gotbno;
3235 : }
3236 : #undef ISVALID
3237 63895157 : }
3238 :
3239 : int
3240 56087809 : xfs_bmap_longest_free_extent(
3241 : struct xfs_perag *pag,
3242 : struct xfs_trans *tp,
3243 : xfs_extlen_t *blen)
3244 : {
3245 56087809 : xfs_extlen_t longest;
3246 56087809 : int error = 0;
3247 :
3248 112175618 : if (!xfs_perag_initialised_agf(pag)) {
3249 1621 : error = xfs_alloc_read_agf(pag, tp, XFS_ALLOC_FLAG_TRYLOCK,
3250 : NULL);
3251 1621 : if (error)
3252 : return error;
3253 : }
3254 :
3255 56086204 : longest = xfs_alloc_longest_free_extent(pag,
3256 : xfs_alloc_min_freelist(pag->pag_mount, pag),
3257 : xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE));
3258 56086770 : if (*blen < longest)
3259 55944750 : *blen = longest;
3260 :
3261 : return 0;
3262 : }
3263 :
3264 : static xfs_extlen_t
3265 : xfs_bmap_select_minlen(
3266 : struct xfs_bmalloca *ap,
3267 : struct xfs_alloc_arg *args,
3268 : xfs_extlen_t blen)
3269 : {
3270 :
3271 : /*
3272 : * Since we used XFS_ALLOC_FLAG_TRYLOCK in _longest_free_extent(), it is
3273 : * possible that there is enough contiguous free space for this request.
3274 : */
3275 55751365 : if (blen < ap->minlen)
3276 : return ap->minlen;
3277 :
3278 : /*
3279 : * If the best seen length is less than the request length,
3280 : * use the best as the minimum, otherwise we've got the maxlen we
3281 : * were asked for.
3282 : */
3283 55751327 : if (blen < args->maxlen)
3284 : return blen;
3285 : return args->maxlen;
3286 : }
3287 :
3288 : static int
3289 55740402 : xfs_bmap_btalloc_select_lengths(
3290 : struct xfs_bmalloca *ap,
3291 : struct xfs_alloc_arg *args,
3292 : xfs_extlen_t *blen)
3293 : {
3294 55740402 : struct xfs_mount *mp = args->mp;
3295 55740402 : struct xfs_perag *pag;
3296 55740402 : xfs_agnumber_t agno, startag;
3297 55740402 : int error = 0;
3298 :
3299 55740402 : if (ap->tp->t_flags & XFS_TRANS_LOWMODE) {
3300 122 : args->total = ap->minlen;
3301 122 : args->minlen = ap->minlen;
3302 122 : return 0;
3303 : }
3304 :
3305 55740280 : args->total = ap->total;
3306 55740280 : startag = XFS_FSB_TO_AGNO(mp, ap->blkno);
3307 55740280 : if (startag == NULLAGNUMBER)
3308 0 : startag = 0;
3309 :
3310 55740280 : *blen = 0;
3311 56070841 : for_each_perag_wrap(mp, startag, agno, pag) {
3312 56001284 : error = xfs_bmap_longest_free_extent(pag, args->tp, blen);
3313 56000348 : if (error && error != -EAGAIN)
3314 : break;
3315 56000348 : error = 0;
3316 56000348 : if (*blen >= args->maxlen)
3317 : break;
3318 : }
3319 55738813 : if (pag)
3320 55671608 : xfs_perag_rele(pag);
3321 :
3322 55740140 : args->minlen = xfs_bmap_select_minlen(ap, args, *blen);
3323 55740140 : return error;
3324 : }
3325 :
3326 : /* Update all inode and quota accounting for the allocation we just did. */
3327 : static void
3328 56120309 : xfs_bmap_btalloc_accounting(
3329 : struct xfs_bmalloca *ap,
3330 : struct xfs_alloc_arg *args)
3331 : {
3332 56120309 : if (ap->flags & XFS_BMAPI_COWFORK) {
3333 : /*
3334 : * COW fork blocks are in-core only and thus are treated as
3335 : * in-core quota reservation (like delalloc blocks) even when
3336 : * converted to real blocks. The quota reservation is not
3337 : * accounted to disk until blocks are remapped to the data
3338 : * fork. So if these blocks were previously delalloc, we
3339 : * already have quota reservation and there's nothing to do
3340 : * yet.
3341 : */
3342 336674 : if (ap->wasdel) {
3343 117343 : xfs_mod_delalloc(ap->ip->i_mount, -(int64_t)args->len);
3344 117343 : return;
3345 : }
3346 :
3347 : /*
3348 : * Otherwise, we've allocated blocks in a hole. The transaction
3349 : * has acquired in-core quota reservation for this extent.
3350 : * Rather than account these as real blocks, however, we reduce
3351 : * the transaction quota reservation based on the allocation.
3352 : * This essentially transfers the transaction quota reservation
3353 : * to that of a delalloc extent.
3354 : */
3355 219331 : ap->ip->i_delayed_blks += args->len;
3356 219331 : xfs_trans_mod_dquot_byino(ap->tp, ap->ip, XFS_TRANS_DQ_RES_BLKS,
3357 219331 : -(long)args->len);
3358 219331 : return;
3359 : }
3360 :
3361 : /* data/attr fork only */
3362 55783635 : ap->ip->i_nblocks += args->len;
3363 55783635 : xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3364 55785616 : if (ap->wasdel) {
3365 15612300 : ap->ip->i_delayed_blks -= args->len;
3366 15612300 : xfs_mod_delalloc(ap->ip->i_mount, -(int64_t)args->len);
3367 : }
3368 111571226 : xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3369 55785613 : ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT : XFS_TRANS_DQ_BCOUNT,
3370 55785613 : args->len);
3371 : }
3372 :
3373 : static int
3374 56119309 : xfs_bmap_compute_alignments(
3375 : struct xfs_bmalloca *ap,
3376 : struct xfs_alloc_arg *args)
3377 : {
3378 56119309 : struct xfs_mount *mp = args->mp;
3379 56119309 : xfs_extlen_t align = 0; /* minimum allocation alignment */
3380 56119309 : int stripe_align = 0;
3381 :
3382 : /* stripe alignment for allocation is determined by mount parameters */
3383 56119309 : if (mp->m_swidth && xfs_has_swalloc(mp))
3384 : stripe_align = mp->m_swidth;
3385 56119311 : else if (mp->m_dalign)
3386 40541 : stripe_align = mp->m_dalign;
3387 :
3388 56119309 : if (ap->flags & XFS_BMAPI_COWFORK)
3389 336690 : align = xfs_get_cowextsz_hint(ap->ip);
3390 55782619 : else if (ap->datatype & XFS_ALLOC_USERDATA)
3391 23837319 : align = xfs_get_extsz_hint(ap->ip);
3392 24174203 : if (align) {
3393 3433906 : if (xfs_bmap_extsize_align(mp, &ap->got, &ap->prev, align, 0,
3394 3433903 : ap->eof, 0, ap->conv, &ap->offset,
3395 : &ap->length))
3396 0 : ASSERT(0);
3397 3433906 : ASSERT(ap->length);
3398 : }
3399 :
3400 : /* apply extent size hints if obtained earlier */
3401 56119506 : if (align) {
3402 3433911 : args->prod = align;
3403 3433911 : div_u64_rem(ap->offset, args->prod, &args->mod);
3404 3433911 : if (args->mod)
3405 6385 : args->mod = args->prod - args->mod;
3406 52685595 : } else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) {
3407 358 : args->prod = 1;
3408 358 : args->mod = 0;
3409 : } else {
3410 52685237 : args->prod = PAGE_SIZE >> mp->m_sb.sb_blocklog;
3411 52685237 : div_u64_rem(ap->offset, args->prod, &args->mod);
3412 52685237 : if (args->mod)
3413 18052161 : args->mod = args->prod - args->mod;
3414 : }
3415 :
3416 56119506 : return stripe_align;
3417 : }
3418 :
3419 : static void
3420 56120373 : xfs_bmap_process_allocated_extent(
3421 : struct xfs_bmalloca *ap,
3422 : struct xfs_alloc_arg *args,
3423 : xfs_fileoff_t orig_offset,
3424 : xfs_extlen_t orig_length)
3425 : {
3426 56120373 : ap->blkno = args->fsbno;
3427 56120373 : ap->length = args->len;
3428 : /*
3429 : * If the extent size hint is active, we tried to round the
3430 : * caller's allocation request offset down to extsz and the
3431 : * length up to another extsz boundary. If we found a free
3432 : * extent we mapped it in starting at this new offset. If the
3433 : * newly mapped space isn't long enough to cover any of the
3434 : * range of offsets that was originally requested, move the
3435 : * mapping up so that we can fill as much of the caller's
3436 : * original request as possible. Free space is apparently
3437 : * very fragmented so we're unlikely to be able to satisfy the
3438 : * hints anyway.
3439 : */
3440 56120373 : if (ap->length <= orig_length)
3441 55885056 : ap->offset = orig_offset;
3442 235317 : else if (ap->offset + ap->length < orig_offset + orig_length)
3443 0 : ap->offset = orig_offset + orig_length - ap->length;
3444 56120373 : xfs_bmap_btalloc_accounting(ap, args);
3445 56121704 : }
3446 :
3447 : #ifdef DEBUG
3448 : static int
3449 370107 : xfs_bmap_exact_minlen_extent_alloc(
3450 : struct xfs_bmalloca *ap)
3451 : {
3452 370107 : struct xfs_mount *mp = ap->ip->i_mount;
3453 370107 : struct xfs_alloc_arg args = { .tp = ap->tp, .mp = mp };
3454 370107 : xfs_fileoff_t orig_offset;
3455 370107 : xfs_extlen_t orig_length;
3456 370107 : int error;
3457 :
3458 370107 : ASSERT(ap->length);
3459 :
3460 370107 : if (ap->minlen != 1) {
3461 0 : ap->blkno = NULLFSBLOCK;
3462 0 : ap->length = 0;
3463 0 : return 0;
3464 : }
3465 :
3466 370107 : orig_offset = ap->offset;
3467 370107 : orig_length = ap->length;
3468 :
3469 370107 : args.alloc_minlen_only = 1;
3470 :
3471 370107 : xfs_bmap_compute_alignments(ap, &args);
3472 :
3473 : /*
3474 : * Unlike the longest extent available in an AG, we don't track
3475 : * the length of an AG's shortest extent.
3476 : * XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT is a debug only knob and
3477 : * hence we can afford to start traversing from the 0th AG since
3478 : * we need not be concerned about a drop in performance in
3479 : * "debug only" code paths.
3480 : */
3481 370109 : ap->blkno = XFS_AGB_TO_FSB(mp, 0, 0);
3482 :
3483 370109 : args.oinfo = XFS_RMAP_OINFO_SKIP_UPDATE;
3484 370109 : args.minlen = args.maxlen = ap->minlen;
3485 370109 : args.total = ap->total;
3486 :
3487 370109 : args.alignment = 1;
3488 370109 : args.minalignslop = 0;
3489 :
3490 370109 : args.minleft = ap->minleft;
3491 370109 : args.wasdel = ap->wasdel;
3492 370109 : args.resv = XFS_AG_RESV_NONE;
3493 370109 : args.datatype = ap->datatype;
3494 :
3495 370109 : error = xfs_alloc_vextent_first_ag(&args, ap->blkno);
3496 370112 : if (error)
3497 : return error;
3498 :
3499 370112 : if (args.fsbno != NULLFSBLOCK) {
3500 370112 : xfs_bmap_process_allocated_extent(ap, &args, orig_offset,
3501 : orig_length);
3502 : } else {
3503 0 : ap->blkno = NULLFSBLOCK;
3504 0 : ap->length = 0;
3505 : }
3506 :
3507 : return 0;
3508 : }
3509 : #else
3510 :
3511 : #define xfs_bmap_exact_minlen_extent_alloc(bma) (-EFSCORRUPTED)
3512 :
3513 : #endif
3514 :
3515 : /*
3516 : * If we are not low on available data blocks and we are allocating at
3517 : * EOF, optimise allocation for contiguous file extension and/or stripe
3518 : * alignment of the new extent.
3519 : *
3520 : * NOTE: ap->aeof is only set if the allocation length is >= the
3521 : * stripe unit and the allocation offset is at the end of file.
3522 : */
3523 : static int
3524 3829 : xfs_bmap_btalloc_at_eof(
3525 : struct xfs_bmalloca *ap,
3526 : struct xfs_alloc_arg *args,
3527 : xfs_extlen_t blen,
3528 : int stripe_align,
3529 : bool ag_only)
3530 : {
3531 3829 : struct xfs_mount *mp = args->mp;
3532 3829 : struct xfs_perag *caller_pag = args->pag;
3533 3829 : int error;
3534 :
3535 : /*
3536 : * If there are already extents in the file, try an exact EOF block
3537 : * allocation to extend the file as a contiguous extent. If that fails,
3538 : * or it's the first allocation in a file, just try for a stripe aligned
3539 : * allocation.
3540 : */
3541 3829 : if (ap->offset) {
3542 3378 : xfs_extlen_t nextminlen = 0;
3543 :
3544 : /*
3545 : * Compute the minlen+alignment for the next case. Set slop so
3546 : * that the value of minlen+alignment+slop doesn't go up between
3547 : * the calls.
3548 : */
3549 3378 : args->alignment = 1;
3550 3378 : if (blen > stripe_align && blen <= args->maxlen)
3551 0 : nextminlen = blen - stripe_align;
3552 : else
3553 3378 : nextminlen = args->minlen;
3554 3378 : if (nextminlen + stripe_align > args->minlen + 1)
3555 3378 : args->minalignslop = nextminlen + stripe_align -
3556 3378 : args->minlen - 1;
3557 : else
3558 0 : args->minalignslop = 0;
3559 :
3560 3378 : if (!caller_pag)
3561 3379 : args->pag = xfs_perag_get(mp, XFS_FSB_TO_AGNO(mp, ap->blkno));
3562 3377 : error = xfs_alloc_vextent_exact_bno(args, ap->blkno);
3563 3379 : if (!caller_pag) {
3564 3379 : xfs_perag_put(args->pag);
3565 3379 : args->pag = NULL;
3566 : }
3567 3379 : if (error)
3568 : return error;
3569 :
3570 3379 : if (args->fsbno != NULLFSBLOCK)
3571 : return 0;
3572 : /*
3573 : * Exact allocation failed. Reset to try an aligned allocation
3574 : * according to the original allocation specification.
3575 : */
3576 2925 : args->alignment = stripe_align;
3577 2925 : args->minlen = nextminlen;
3578 2925 : args->minalignslop = 0;
3579 : } else {
3580 : /*
3581 : * Adjust minlen to try and preserve alignment if we
3582 : * can't guarantee an aligned maxlen extent.
3583 : */
3584 451 : args->alignment = stripe_align;
3585 451 : if (blen > args->alignment &&
3586 451 : blen <= args->maxlen + args->alignment)
3587 0 : args->minlen = blen - args->alignment;
3588 451 : args->minalignslop = 0;
3589 : }
3590 :
3591 3376 : if (ag_only) {
3592 0 : error = xfs_alloc_vextent_near_bno(args, ap->blkno);
3593 : } else {
3594 3376 : args->pag = NULL;
3595 3376 : error = xfs_alloc_vextent_start_ag(args, ap->blkno);
3596 3376 : ASSERT(args->pag == NULL);
3597 3376 : args->pag = caller_pag;
3598 : }
3599 3376 : if (error)
3600 : return error;
3601 :
3602 3376 : if (args->fsbno != NULLFSBLOCK)
3603 : return 0;
3604 :
3605 : /*
3606 : * Allocation failed, so turn return the allocation args to their
3607 : * original non-aligned state so the caller can proceed on allocation
3608 : * failure as if this function was never called.
3609 : */
3610 0 : args->alignment = 1;
3611 0 : return 0;
3612 : }
3613 :
3614 : /*
3615 : * We have failed multiple allocation attempts so now are in a low space
3616 : * allocation situation. Try a locality first full filesystem minimum length
3617 : * allocation whilst still maintaining necessary total block reservation
3618 : * requirements.
3619 : *
3620 : * If that fails, we are now critically low on space, so perform a last resort
3621 : * allocation attempt: no reserve, no locality, blocking, minimum length, full
3622 : * filesystem free space scan. We also indicate to future allocations in this
3623 : * transaction that we are critically low on space so they don't waste time on
3624 : * allocation modes that are unlikely to succeed.
3625 : */
3626 : int
3627 31224 : xfs_bmap_btalloc_low_space(
3628 : struct xfs_bmalloca *ap,
3629 : struct xfs_alloc_arg *args)
3630 : {
3631 31224 : int error;
3632 :
3633 31224 : if (args->minlen > ap->minlen) {
3634 31184 : args->minlen = ap->minlen;
3635 31184 : error = xfs_alloc_vextent_start_ag(args, ap->blkno);
3636 31184 : if (error || args->fsbno != NULLFSBLOCK)
3637 : return error;
3638 : }
3639 :
3640 : /* Last ditch attempt before failure is declared. */
3641 40 : args->total = ap->minlen;
3642 40 : error = xfs_alloc_vextent_first_ag(args, 0);
3643 40 : if (error)
3644 : return error;
3645 40 : ap->tp->t_flags |= XFS_TRANS_LOWMODE;
3646 40 : return 0;
3647 : }
3648 :
3649 : static int
3650 11222 : xfs_bmap_btalloc_filestreams(
3651 : struct xfs_bmalloca *ap,
3652 : struct xfs_alloc_arg *args,
3653 : int stripe_align)
3654 : {
3655 11222 : xfs_extlen_t blen = 0;
3656 11222 : int error = 0;
3657 :
3658 :
3659 11222 : error = xfs_filestream_select_ag(ap, args, &blen);
3660 11225 : if (error)
3661 : return error;
3662 11225 : ASSERT(args->pag);
3663 :
3664 : /*
3665 : * If we are in low space mode, then optimal allocation will fail so
3666 : * prepare for minimal allocation and jump to the low space algorithm
3667 : * immediately.
3668 : */
3669 11225 : if (ap->tp->t_flags & XFS_TRANS_LOWMODE) {
3670 0 : args->minlen = ap->minlen;
3671 0 : ASSERT(args->fsbno == NULLFSBLOCK);
3672 0 : goto out_low_space;
3673 : }
3674 :
3675 11225 : args->minlen = xfs_bmap_select_minlen(ap, args, blen);
3676 11225 : if (ap->aeof)
3677 0 : error = xfs_bmap_btalloc_at_eof(ap, args, blen, stripe_align,
3678 : true);
3679 :
3680 11225 : if (!error && args->fsbno == NULLFSBLOCK)
3681 11225 : error = xfs_alloc_vextent_near_bno(args, ap->blkno);
3682 :
3683 0 : out_low_space:
3684 : /*
3685 : * We are now done with the perag reference for the filestreams
3686 : * association provided by xfs_filestream_select_ag(). Release it now as
3687 : * we've either succeeded, had a fatal error or we are out of space and
3688 : * need to do a full filesystem scan for free space which will take it's
3689 : * own references.
3690 : */
3691 11222 : xfs_perag_rele(args->pag);
3692 11223 : args->pag = NULL;
3693 11223 : if (error || args->fsbno != NULLFSBLOCK)
3694 : return error;
3695 :
3696 1 : return xfs_bmap_btalloc_low_space(ap, args);
3697 : }
3698 :
3699 : static int
3700 55738303 : xfs_bmap_btalloc_best_length(
3701 : struct xfs_bmalloca *ap,
3702 : struct xfs_alloc_arg *args,
3703 : int stripe_align)
3704 : {
3705 55738303 : xfs_extlen_t blen = 0;
3706 55738303 : int error;
3707 :
3708 55738303 : ap->blkno = XFS_INO_TO_FSB(args->mp, ap->ip->i_ino);
3709 55738303 : xfs_bmap_adjacent(ap);
3710 :
3711 : /*
3712 : * Search for an allocation group with a single extent large enough for
3713 : * the request. If one isn't found, then adjust the minimum allocation
3714 : * size to the largest space found.
3715 : */
3716 55738116 : error = xfs_bmap_btalloc_select_lengths(ap, args, &blen);
3717 55740014 : if (error)
3718 : return error;
3719 :
3720 : /*
3721 : * Don't attempt optimal EOF allocation if previous allocations barely
3722 : * succeeded due to being near ENOSPC. It is highly unlikely we'll get
3723 : * optimal or even aligned allocations in this case, so don't waste time
3724 : * trying.
3725 : */
3726 55740014 : if (ap->aeof && !(ap->tp->t_flags & XFS_TRANS_LOWMODE)) {
3727 3829 : error = xfs_bmap_btalloc_at_eof(ap, args, blen, stripe_align,
3728 : false);
3729 3830 : if (error || args->fsbno != NULLFSBLOCK)
3730 : return error;
3731 : }
3732 :
3733 55736185 : error = xfs_alloc_vextent_start_ag(args, ap->blkno);
3734 55736778 : if (error || args->fsbno != NULLFSBLOCK)
3735 : return error;
3736 :
3737 31223 : return xfs_bmap_btalloc_low_space(ap, args);
3738 : }
3739 :
3740 : static int
3741 55751722 : xfs_bmap_btalloc(
3742 : struct xfs_bmalloca *ap)
3743 : {
3744 55751722 : struct xfs_mount *mp = ap->ip->i_mount;
3745 55751722 : struct xfs_alloc_arg args = {
3746 55751722 : .tp = ap->tp,
3747 : .mp = mp,
3748 : .fsbno = NULLFSBLOCK,
3749 : .oinfo = XFS_RMAP_OINFO_SKIP_UPDATE,
3750 55751722 : .minleft = ap->minleft,
3751 55751722 : .wasdel = ap->wasdel,
3752 : .resv = XFS_AG_RESV_NONE,
3753 55751722 : .datatype = ap->datatype,
3754 : .alignment = 1,
3755 : .minalignslop = 0,
3756 : };
3757 55751722 : xfs_fileoff_t orig_offset;
3758 55751722 : xfs_extlen_t orig_length;
3759 55751722 : int error;
3760 55751722 : int stripe_align;
3761 :
3762 55751722 : ASSERT(ap->length);
3763 55751722 : orig_offset = ap->offset;
3764 55751722 : orig_length = ap->length;
3765 :
3766 55751722 : stripe_align = xfs_bmap_compute_alignments(ap, &args);
3767 :
3768 : /* Trim the allocation back to the maximum an AG can fit. */
3769 55751004 : args.maxlen = min(ap->length, mp->m_ag_max_usable);
3770 :
3771 55751004 : if ((ap->datatype & XFS_ALLOC_USERDATA) &&
3772 23809289 : xfs_inode_is_filestream(ap->ip))
3773 11318 : error = xfs_bmap_btalloc_filestreams(ap, &args, stripe_align);
3774 : else
3775 55739694 : error = xfs_bmap_btalloc_best_length(ap, &args, stripe_align);
3776 55752274 : if (error)
3777 : return error;
3778 :
3779 55750783 : if (args.fsbno != NULLFSBLOCK) {
3780 55750743 : xfs_bmap_process_allocated_extent(ap, &args, orig_offset,
3781 : orig_length);
3782 : } else {
3783 40 : ap->blkno = NULLFSBLOCK;
3784 40 : ap->length = 0;
3785 : }
3786 : return 0;
3787 : }
3788 :
3789 : /* Trim extent to fit a logical block range. */
3790 : void
3791 278259771 : xfs_trim_extent(
3792 : struct xfs_bmbt_irec *irec,
3793 : xfs_fileoff_t bno,
3794 : xfs_filblks_t len)
3795 : {
3796 278259771 : xfs_fileoff_t distance;
3797 278259771 : xfs_fileoff_t end = bno + len;
3798 :
3799 278259771 : if (irec->br_startoff + irec->br_blockcount <= bno ||
3800 : irec->br_startoff >= end) {
3801 93866037 : irec->br_blockcount = 0;
3802 93866037 : return;
3803 : }
3804 :
3805 184393734 : if (irec->br_startoff < bno) {
3806 56018655 : distance = bno - irec->br_startoff;
3807 56018655 : if (isnullstartblock(irec->br_startblock))
3808 274418 : irec->br_startblock = DELAYSTARTBLOCK;
3809 56018655 : if (irec->br_startblock != DELAYSTARTBLOCK &&
3810 : irec->br_startblock != HOLESTARTBLOCK)
3811 55745135 : irec->br_startblock += distance;
3812 56018655 : irec->br_startoff += distance;
3813 56018655 : irec->br_blockcount -= distance;
3814 : }
3815 :
3816 184393734 : if (end < irec->br_startoff + irec->br_blockcount) {
3817 3248367 : distance = irec->br_startoff + irec->br_blockcount - end;
3818 3248367 : irec->br_blockcount -= distance;
3819 : }
3820 : }
3821 :
3822 : /*
3823 : * Trim the returned map to the required bounds
3824 : */
3825 : STATIC void
3826 2912203957 : xfs_bmapi_trim_map(
3827 : struct xfs_bmbt_irec *mval,
3828 : struct xfs_bmbt_irec *got,
3829 : xfs_fileoff_t *bno,
3830 : xfs_filblks_t len,
3831 : xfs_fileoff_t obno,
3832 : xfs_fileoff_t end,
3833 : int n,
3834 : uint32_t flags)
3835 : {
3836 2912203957 : if ((flags & XFS_BMAPI_ENTIRE) ||
3837 2912203957 : got->br_startoff + got->br_blockcount <= obno) {
3838 8 : *mval = *got;
3839 8 : if (isnullstartblock(got->br_startblock))
3840 0 : mval->br_startblock = DELAYSTARTBLOCK;
3841 8 : return;
3842 : }
3843 :
3844 2912203949 : if (obno > *bno)
3845 0 : *bno = obno;
3846 2912203949 : ASSERT((*bno >= obno) || (n == 0));
3847 2912203949 : ASSERT(*bno < end);
3848 2912203949 : mval->br_startoff = *bno;
3849 2912203949 : if (isnullstartblock(got->br_startblock))
3850 3598080 : mval->br_startblock = DELAYSTARTBLOCK;
3851 : else
3852 2908605869 : mval->br_startblock = got->br_startblock +
3853 2908605869 : (*bno - got->br_startoff);
3854 : /*
3855 : * Return the minimum of what we got and what we asked for for
3856 : * the length. We can use the len variable here because it is
3857 : * modified below and we could have been there before coming
3858 : * here if the first part of the allocation didn't overlap what
3859 : * was asked for.
3860 : */
3861 2912203949 : mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3862 : got->br_blockcount - (*bno - got->br_startoff));
3863 2912203949 : mval->br_state = got->br_state;
3864 2912203949 : ASSERT(mval->br_blockcount <= len);
3865 : return;
3866 : }
3867 :
3868 : /*
3869 : * Update and validate the extent map to return
3870 : */
3871 : STATIC void
3872 2912493873 : xfs_bmapi_update_map(
3873 : struct xfs_bmbt_irec **map,
3874 : xfs_fileoff_t *bno,
3875 : xfs_filblks_t *len,
3876 : xfs_fileoff_t obno,
3877 : xfs_fileoff_t end,
3878 : int *n,
3879 : uint32_t flags)
3880 : {
3881 2912493873 : xfs_bmbt_irec_t *mval = *map;
3882 :
3883 2912493873 : ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3884 : ((mval->br_startoff + mval->br_blockcount) <= end));
3885 2912493873 : ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3886 : (mval->br_startoff < obno));
3887 :
3888 2912493873 : *bno = mval->br_startoff + mval->br_blockcount;
3889 2912493873 : *len = end - *bno;
3890 2912493873 : if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3891 : /* update previous map with new information */
3892 0 : ASSERT(mval->br_startblock == mval[-1].br_startblock);
3893 0 : ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3894 0 : ASSERT(mval->br_state == mval[-1].br_state);
3895 0 : mval[-1].br_blockcount = mval->br_blockcount;
3896 0 : mval[-1].br_state = mval->br_state;
3897 2912493873 : } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3898 138052 : mval[-1].br_startblock != DELAYSTARTBLOCK &&
3899 137586 : mval[-1].br_startblock != HOLESTARTBLOCK &&
3900 137586 : mval->br_startblock == mval[-1].br_startblock +
3901 137586 : mval[-1].br_blockcount &&
3902 0 : mval[-1].br_state == mval->br_state) {
3903 0 : ASSERT(mval->br_startoff ==
3904 : mval[-1].br_startoff + mval[-1].br_blockcount);
3905 0 : mval[-1].br_blockcount += mval->br_blockcount;
3906 2912493873 : } else if (*n > 0 &&
3907 138052 : mval->br_startblock == DELAYSTARTBLOCK &&
3908 0 : mval[-1].br_startblock == DELAYSTARTBLOCK &&
3909 0 : mval->br_startoff ==
3910 0 : mval[-1].br_startoff + mval[-1].br_blockcount) {
3911 0 : mval[-1].br_blockcount += mval->br_blockcount;
3912 0 : mval[-1].br_state = mval->br_state;
3913 2912493873 : } else if (!((*n == 0) &&
3914 2912291794 : ((mval->br_startoff + mval->br_blockcount) <=
3915 : obno))) {
3916 2912512938 : mval++;
3917 2912512938 : (*n)++;
3918 : }
3919 2912493873 : *map = mval;
3920 2912493873 : }
3921 :
3922 : /*
3923 : * Map file blocks to filesystem blocks without allocation.
3924 : */
3925 : int
3926 3390642069 : xfs_bmapi_read(
3927 : struct xfs_inode *ip,
3928 : xfs_fileoff_t bno,
3929 : xfs_filblks_t len,
3930 : struct xfs_bmbt_irec *mval,
3931 : int *nmap,
3932 : uint32_t flags)
3933 : {
3934 3390642069 : struct xfs_mount *mp = ip->i_mount;
3935 3390642069 : int whichfork = xfs_bmapi_whichfork(flags);
3936 3390642069 : struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
3937 3390932569 : struct xfs_bmbt_irec got;
3938 3390932569 : xfs_fileoff_t obno;
3939 3390932569 : xfs_fileoff_t end;
3940 3390932569 : struct xfs_iext_cursor icur;
3941 3390932569 : int error;
3942 3390932569 : bool eof = false;
3943 3390932569 : int n = 0;
3944 :
3945 3390932569 : ASSERT(*nmap >= 1);
3946 3390932569 : ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK | XFS_BMAPI_ENTIRE)));
3947 3390932569 : ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL));
3948 :
3949 3390836089 : if (WARN_ON_ONCE(!ifp)) {
3950 0 : xfs_bmap_mark_sick(ip, whichfork);
3951 0 : return -EFSCORRUPTED;
3952 : }
3953 :
3954 6782092232 : if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
3955 3390836089 : XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
3956 0 : xfs_bmap_mark_sick(ip, whichfork);
3957 0 : return -EFSCORRUPTED;
3958 : }
3959 :
3960 6782512286 : if (xfs_is_shutdown(mp))
3961 : return -EIO;
3962 :
3963 3391251372 : XFS_STATS_INC(mp, xs_blk_mapr);
3964 :
3965 3391251372 : error = xfs_iread_extents(NULL, ip, whichfork);
3966 3390953796 : if (error)
3967 : return error;
3968 :
3969 3391435131 : if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got))
3970 266907710 : eof = true;
3971 3391409664 : end = bno + len;
3972 3391409664 : obno = bno;
3973 :
3974 3995814842 : while (bno < end && n < *nmap) {
3975 : /* Reading past eof, act as though there's a hole up to end. */
3976 3415735857 : if (eof)
3977 291091626 : got.br_startoff = end;
3978 3415735857 : if (got.br_startoff > bno) {
3979 : /* Reading in a hole. */
3980 580084362 : mval->br_startoff = bno;
3981 580084362 : mval->br_startblock = HOLESTARTBLOCK;
3982 580084362 : mval->br_blockcount =
3983 580084362 : XFS_FILBLKS_MIN(len, got.br_startoff - bno);
3984 580084362 : mval->br_state = XFS_EXT_NORM;
3985 580084362 : bno += mval->br_blockcount;
3986 580084362 : len -= mval->br_blockcount;
3987 580084362 : mval++;
3988 580084362 : n++;
3989 580084362 : continue;
3990 : }
3991 :
3992 : /* set up the extent map to return. */
3993 2835651495 : xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
3994 2836266908 : xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
3995 :
3996 : /* If we're done, stop now. */
3997 2836479794 : if (bno >= end || n >= *nmap)
3998 : break;
3999 :
4000 : /* Else go on to the next record. */
4001 24320801 : if (!xfs_iext_next_extent(ifp, &icur, &got))
4002 24182752 : eof = true;
4003 : }
4004 3392237978 : *nmap = n;
4005 3392237978 : return 0;
4006 : }
4007 :
4008 : /*
4009 : * Add a delayed allocation extent to an inode. Blocks are reserved from the
4010 : * global pool and the extent inserted into the inode in-core extent tree.
4011 : *
4012 : * On entry, got refers to the first extent beyond the offset of the extent to
4013 : * allocate or eof is specified if no such extent exists. On return, got refers
4014 : * to the extent record that was inserted to the inode fork.
4015 : *
4016 : * Note that the allocated extent may have been merged with contiguous extents
4017 : * during insertion into the inode fork. Thus, got does not reflect the current
4018 : * state of the inode fork on return. If necessary, the caller can use lastx to
4019 : * look up the updated record in the inode fork.
4020 : */
4021 : int
4022 18808087 : xfs_bmapi_reserve_delalloc(
4023 : struct xfs_inode *ip,
4024 : int whichfork,
4025 : xfs_fileoff_t off,
4026 : xfs_filblks_t len,
4027 : xfs_filblks_t prealloc,
4028 : struct xfs_bmbt_irec *got,
4029 : struct xfs_iext_cursor *icur,
4030 : int eof)
4031 : {
4032 18808087 : struct xfs_mount *mp = ip->i_mount;
4033 18808087 : struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
4034 18808157 : xfs_extlen_t alen;
4035 18808157 : xfs_extlen_t indlen;
4036 18808157 : int error;
4037 18808157 : xfs_fileoff_t aoff = off;
4038 :
4039 : /*
4040 : * Cap the alloc length. Keep track of prealloc so we know whether to
4041 : * tag the inode before we return.
4042 : */
4043 18808157 : alen = XFS_FILBLKS_MIN(len + prealloc, XFS_MAX_BMBT_EXTLEN);
4044 18808157 : if (!eof)
4045 8798718 : alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4046 18808157 : if (prealloc && alen >= len)
4047 3150396 : prealloc = alen - len;
4048 :
4049 : /* Figure out the extent size, adjust alen */
4050 18808157 : if (whichfork == XFS_COW_FORK) {
4051 160892 : struct xfs_bmbt_irec prev;
4052 160892 : xfs_extlen_t extsz = xfs_get_cowextsz_hint(ip);
4053 :
4054 160892 : if (!xfs_iext_peek_prev_extent(ifp, icur, &prev))
4055 100800 : prev.br_startoff = NULLFILEOFF;
4056 :
4057 160892 : error = xfs_bmap_extsize_align(mp, got, &prev, extsz, 0, eof,
4058 : 1, 0, &aoff, &alen);
4059 160892 : ASSERT(!error);
4060 : }
4061 :
4062 : /*
4063 : * Make a transaction-less quota reservation for delayed allocation
4064 : * blocks. This number gets adjusted later. We return if we haven't
4065 : * allocated blocks already inside this loop.
4066 : */
4067 18808157 : error = xfs_quota_reserve_blkres(ip, alen);
4068 18811971 : if (error)
4069 : return error;
4070 :
4071 : /*
4072 : * Split changing sb for alen and indlen since they could be coming
4073 : * from different places.
4074 : */
4075 18811249 : indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4076 18810709 : ASSERT(indlen > 0);
4077 :
4078 18810709 : error = xfs_mod_fdblocks(mp, -((int64_t)alen), false);
4079 18810030 : if (error)
4080 567211 : goto out_unreserve_quota;
4081 :
4082 18242819 : error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false);
4083 18243488 : if (error)
4084 166643 : goto out_unreserve_blocks;
4085 :
4086 :
4087 18076845 : ip->i_delayed_blks += alen;
4088 18076845 : xfs_mod_delalloc(ip->i_mount, alen + indlen);
4089 :
4090 18076268 : got->br_startoff = aoff;
4091 18076268 : got->br_startblock = nullstartblock(indlen);
4092 18076207 : got->br_blockcount = alen;
4093 18076207 : got->br_state = XFS_EXT_NORM;
4094 :
4095 18076207 : xfs_bmap_add_extent_hole_delay(ip, whichfork, icur, got);
4096 :
4097 : /*
4098 : * Tag the inode if blocks were preallocated. Note that COW fork
4099 : * preallocation can occur at the start or end of the extent, even when
4100 : * prealloc == 0, so we must also check the aligned offset and length.
4101 : */
4102 18071680 : if (whichfork == XFS_DATA_FORK && prealloc)
4103 3016123 : xfs_inode_set_eofblocks_tag(ip);
4104 18071715 : if (whichfork == XFS_COW_FORK && (prealloc || aoff < off || alen > len))
4105 151463 : xfs_inode_set_cowblocks_tag(ip);
4106 :
4107 : return 0;
4108 :
4109 : out_unreserve_blocks:
4110 166643 : xfs_mod_fdblocks(mp, alen, false);
4111 733855 : out_unreserve_quota:
4112 733855 : if (XFS_IS_QUOTA_ON(mp))
4113 733828 : xfs_quota_unreserve_blkres(ip, alen);
4114 : return error;
4115 : }
4116 :
4117 : static int
4118 32318501 : xfs_bmap_alloc_userdata(
4119 : struct xfs_bmalloca *bma)
4120 : {
4121 32318501 : struct xfs_mount *mp = bma->ip->i_mount;
4122 32318501 : int whichfork = xfs_bmapi_whichfork(bma->flags);
4123 32318501 : int error;
4124 :
4125 : /*
4126 : * Set the data type being allocated. For the data fork, the first data
4127 : * in the file is treated differently to all other allocations. For the
4128 : * attribute fork, we only need to ensure the allocated range is not on
4129 : * the busy list.
4130 : */
4131 32318501 : bma->datatype = XFS_ALLOC_NOBUSY;
4132 32318501 : if (whichfork == XFS_DATA_FORK || whichfork == XFS_COW_FORK) {
4133 32317974 : bma->datatype |= XFS_ALLOC_USERDATA;
4134 32317974 : if (bma->offset == 0)
4135 2528199 : bma->datatype |= XFS_ALLOC_INITIAL_USER_DATA;
4136 :
4137 32317974 : if (mp->m_dalign && bma->length >= mp->m_dalign) {
4138 6443 : error = xfs_bmap_isaeof(bma, whichfork);
4139 6443 : if (error)
4140 : return error;
4141 : }
4142 :
4143 32317974 : if (XFS_IS_REALTIME_INODE(bma->ip))
4144 8143832 : return xfs_bmap_rtalloc(bma);
4145 : }
4146 :
4147 24174669 : if (unlikely(XFS_TEST_ERROR(false, mp,
4148 : XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT)))
4149 364807 : return xfs_bmap_exact_minlen_extent_alloc(bma);
4150 :
4151 23809789 : return xfs_bmap_btalloc(bma);
4152 : }
4153 :
4154 : static int
4155 64264378 : xfs_bmapi_allocate(
4156 : struct xfs_bmalloca *bma)
4157 : {
4158 64264378 : struct xfs_mount *mp = bma->ip->i_mount;
4159 64264378 : int whichfork = xfs_bmapi_whichfork(bma->flags);
4160 64264378 : struct xfs_ifork *ifp = xfs_ifork_ptr(bma->ip, whichfork);
4161 64266181 : int tmp_logflags = 0;
4162 64266181 : int error;
4163 :
4164 64266181 : ASSERT(bma->length > 0);
4165 :
4166 : /*
4167 : * For the wasdelay case, we could also just allocate the stuff asked
4168 : * for in this bmap call but that wouldn't be as good.
4169 : */
4170 64266181 : if (bma->wasdel) {
4171 15730356 : bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4172 15730356 : bma->offset = bma->got.br_startoff;
4173 15730356 : if (!xfs_iext_peek_prev_extent(ifp, &bma->icur, &bma->prev))
4174 5172692 : bma->prev.br_startoff = NULLFILEOFF;
4175 : } else {
4176 48535825 : bma->length = XFS_FILBLKS_MIN(bma->length, XFS_MAX_BMBT_EXTLEN);
4177 48535825 : if (!bma->eof)
4178 9416511 : bma->length = XFS_FILBLKS_MIN(bma->length,
4179 : bma->got.br_startoff - bma->offset);
4180 : }
4181 :
4182 64266178 : if (bma->flags & XFS_BMAPI_CONTIG)
4183 4750469 : bma->minlen = bma->length;
4184 : else
4185 59515709 : bma->minlen = 1;
4186 :
4187 64266178 : if (bma->flags & XFS_BMAPI_METADATA) {
4188 31946643 : if (unlikely(XFS_TEST_ERROR(false, mp,
4189 : XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT)))
4190 5307 : error = xfs_bmap_exact_minlen_extent_alloc(bma);
4191 : else
4192 31940362 : error = xfs_bmap_btalloc(bma);
4193 : } else {
4194 32319535 : error = xfs_bmap_alloc_userdata(bma);
4195 : }
4196 64265136 : if (error || bma->blkno == NULLFSBLOCK)
4197 : return error;
4198 :
4199 64263605 : if (bma->flags & XFS_BMAPI_ZERO) {
4200 0 : error = xfs_zero_extent(bma->ip, bma->blkno, bma->length);
4201 0 : if (error)
4202 : return error;
4203 : }
4204 :
4205 64263605 : if (ifp->if_format == XFS_DINODE_FMT_BTREE && !bma->cur)
4206 8118254 : bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4207 : /*
4208 : * Bump the number of extents we've allocated
4209 : * in this call.
4210 : */
4211 64263532 : bma->nallocs++;
4212 :
4213 64263532 : if (bma->cur)
4214 8116167 : bma->cur->bc_ino.flags =
4215 8116167 : bma->wasdel ? XFS_BTCUR_BMBT_WASDEL : 0;
4216 :
4217 64263532 : bma->got.br_startoff = bma->offset;
4218 64263532 : bma->got.br_startblock = bma->blkno;
4219 64263532 : bma->got.br_blockcount = bma->length;
4220 64263532 : bma->got.br_state = XFS_EXT_NORM;
4221 :
4222 64263532 : if (bma->flags & XFS_BMAPI_PREALLOC)
4223 32317478 : bma->got.br_state = XFS_EXT_UNWRITTEN;
4224 :
4225 64263532 : if (bma->wasdel)
4226 15729665 : error = xfs_bmap_add_extent_delay_real(bma, whichfork);
4227 : else
4228 48533867 : 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 :
4232 64263502 : bma->logflags |= tmp_logflags;
4233 64263502 : if (error)
4234 : return error;
4235 :
4236 : /*
4237 : * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4238 : * or xfs_bmap_add_extent_hole_real might have merged it into one of
4239 : * the neighbouring ones.
4240 : */
4241 64263202 : xfs_iext_get_extent(ifp, &bma->icur, &bma->got);
4242 :
4243 64257967 : ASSERT(bma->got.br_startoff <= bma->offset);
4244 64257967 : ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4245 : bma->offset + bma->length);
4246 64257967 : ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4247 : bma->got.br_state == XFS_EXT_UNWRITTEN);
4248 : return 0;
4249 : }
4250 :
4251 : STATIC int
4252 76589841 : xfs_bmapi_convert_unwritten(
4253 : struct xfs_bmalloca *bma,
4254 : struct xfs_bmbt_irec *mval,
4255 : xfs_filblks_t len,
4256 : uint32_t flags)
4257 : {
4258 76589841 : int whichfork = xfs_bmapi_whichfork(flags);
4259 76589841 : struct xfs_ifork *ifp = xfs_ifork_ptr(bma->ip, whichfork);
4260 76586223 : int tmp_logflags = 0;
4261 76586223 : int error;
4262 :
4263 : /* check if we need to do unwritten->real conversion */
4264 76586223 : if (mval->br_state == XFS_EXT_UNWRITTEN &&
4265 42221176 : (flags & XFS_BMAPI_PREALLOC))
4266 : return 0;
4267 :
4268 : /* check if we need to do real->unwritten conversion */
4269 59240682 : if (mval->br_state == XFS_EXT_NORM &&
4270 34364237 : (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4271 : (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4272 : return 0;
4273 :
4274 : /*
4275 : * Modify (by adding) the state flag, if writing.
4276 : */
4277 24876445 : ASSERT(mval->br_blockcount <= len);
4278 24876445 : if (ifp->if_format == XFS_DINODE_FMT_BTREE && !bma->cur) {
4279 5949565 : bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4280 : bma->ip, whichfork);
4281 : }
4282 24876409 : mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4283 24876409 : ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4284 :
4285 : /*
4286 : * Before insertion into the bmbt, zero the range being converted
4287 : * if required.
4288 : */
4289 24876409 : if (flags & XFS_BMAPI_ZERO) {
4290 0 : error = xfs_zero_extent(bma->ip, mval->br_startblock,
4291 0 : mval->br_blockcount);
4292 0 : if (error)
4293 : return error;
4294 : }
4295 :
4296 24876409 : error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, whichfork,
4297 : &bma->icur, &bma->cur, mval, &tmp_logflags);
4298 : /*
4299 : * Log the inode core unconditionally in the unwritten extent conversion
4300 : * path because the conversion might not have done so (e.g., if the
4301 : * extent count hasn't changed). We need to make sure the inode is dirty
4302 : * in the transaction for the sake of fsync(), even if nothing has
4303 : * changed, because fsync() will not force the log for this transaction
4304 : * unless it sees the inode pinned.
4305 : *
4306 : * Note: If we're only converting cow fork extents, there aren't
4307 : * any on-disk updates to make, so we don't need to log anything.
4308 : */
4309 24877612 : if (whichfork != XFS_COW_FORK)
4310 24877612 : bma->logflags |= tmp_logflags | XFS_ILOG_CORE;
4311 24877612 : if (error)
4312 : return error;
4313 :
4314 : /*
4315 : * Update our extent pointer, given that
4316 : * xfs_bmap_add_extent_unwritten_real might have merged it into one
4317 : * of the neighbouring ones.
4318 : */
4319 24877629 : xfs_iext_get_extent(ifp, &bma->icur, &bma->got);
4320 :
4321 : /*
4322 : * We may have combined previously unwritten space with written space,
4323 : * so generate another request.
4324 : */
4325 24876725 : if (mval->br_blockcount < len)
4326 389748 : return -EAGAIN;
4327 : return 0;
4328 : }
4329 :
4330 : xfs_extlen_t
4331 92090940 : xfs_bmapi_minleft(
4332 : struct xfs_trans *tp,
4333 : struct xfs_inode *ip,
4334 : int fork)
4335 : {
4336 92090940 : struct xfs_ifork *ifp = xfs_ifork_ptr(ip, fork);
4337 :
4338 92090335 : if (tp && tp->t_highest_agno != NULLAGNUMBER)
4339 : return 0;
4340 92052565 : if (ifp->if_format != XFS_DINODE_FMT_BTREE)
4341 : return 1;
4342 14881724 : return be16_to_cpu(ifp->if_broot->bb_level) + 1;
4343 : }
4344 :
4345 : /*
4346 : * Log whatever the flags say, even if error. Otherwise we might miss detecting
4347 : * a case where the data is changed, there's an error, and it's not logged so we
4348 : * don't shutdown when we should. Don't bother logging extents/btree changes if
4349 : * we converted to the other format.
4350 : */
4351 : static void
4352 91879198 : xfs_bmapi_finish(
4353 : struct xfs_bmalloca *bma,
4354 : int whichfork,
4355 : int error)
4356 : {
4357 91879198 : struct xfs_ifork *ifp = xfs_ifork_ptr(bma->ip, whichfork);
4358 :
4359 96578631 : if ((bma->logflags & xfs_ilog_fext(whichfork)) &&
4360 74984114 : ifp->if_format != XFS_DINODE_FMT_EXTENTS)
4361 725448 : bma->logflags &= ~xfs_ilog_fext(whichfork);
4362 95853214 : else if ((bma->logflags & xfs_ilog_fbroot(whichfork)) &&
4363 0 : ifp->if_format != XFS_DINODE_FMT_BTREE)
4364 0 : bma->logflags &= ~xfs_ilog_fbroot(whichfork);
4365 :
4366 91880004 : if (bma->logflags)
4367 88847591 : xfs_trans_log_inode(bma->tp, bma->ip, bma->logflags);
4368 91880457 : if (bma->cur)
4369 14793079 : xfs_btree_del_cursor(bma->cur, error);
4370 91880387 : }
4371 :
4372 : /*
4373 : * Map file blocks to filesystem blocks, and allocate blocks or convert the
4374 : * extent state if necessary. Details behaviour is controlled by the flags
4375 : * parameter. Only allocates blocks from a single allocation group, to avoid
4376 : * locking problems.
4377 : */
4378 : int
4379 76205124 : xfs_bmapi_write(
4380 : struct xfs_trans *tp, /* transaction pointer */
4381 : struct xfs_inode *ip, /* incore inode */
4382 : xfs_fileoff_t bno, /* starting file offs. mapped */
4383 : xfs_filblks_t len, /* length to map in file */
4384 : uint32_t flags, /* XFS_BMAPI_... */
4385 : xfs_extlen_t total, /* total blocks needed */
4386 : struct xfs_bmbt_irec *mval, /* output: map values */
4387 : int *nmap) /* i/o: mval size/count */
4388 : {
4389 76205124 : struct xfs_bmalloca bma = {
4390 : .tp = tp,
4391 : .ip = ip,
4392 : .total = total,
4393 : };
4394 76205124 : struct xfs_mount *mp = ip->i_mount;
4395 76205124 : int whichfork = xfs_bmapi_whichfork(flags);
4396 76205124 : struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
4397 76205746 : xfs_fileoff_t end; /* end of mapped file region */
4398 76205746 : bool eof = false; /* after the end of extents */
4399 76205746 : int error; /* error return */
4400 76205746 : int n; /* current extent index */
4401 76205746 : xfs_fileoff_t obno; /* old block number (offset) */
4402 :
4403 : #ifdef DEBUG
4404 76205746 : xfs_fileoff_t orig_bno; /* original block number value */
4405 76205746 : int orig_flags; /* original flags arg value */
4406 76205746 : xfs_filblks_t orig_len; /* original value of len arg */
4407 76205746 : struct xfs_bmbt_irec *orig_mval; /* original value of mval */
4408 76205746 : int orig_nmap; /* original value of *nmap */
4409 :
4410 76205746 : orig_bno = bno;
4411 76205746 : orig_len = len;
4412 76205746 : orig_flags = flags;
4413 76205746 : orig_mval = mval;
4414 76205746 : orig_nmap = *nmap;
4415 : #endif
4416 :
4417 76205746 : ASSERT(*nmap >= 1);
4418 76205746 : ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4419 76205746 : ASSERT(tp != NULL);
4420 76205746 : ASSERT(len > 0);
4421 76205746 : ASSERT(ifp->if_format != XFS_DINODE_FMT_LOCAL);
4422 76205746 : ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4423 76203841 : ASSERT(!(flags & XFS_BMAPI_REMAP));
4424 :
4425 : /* zeroing is for currently only for data extents, not metadata */
4426 76203841 : ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) !=
4427 : (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO));
4428 : /*
4429 : * we can allocate unwritten extents or pre-zero allocated blocks,
4430 : * but it makes no sense to do both at once. This would result in
4431 : * zeroing the unwritten extent twice, but it still being an
4432 : * unwritten extent....
4433 : */
4434 76203841 : ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) !=
4435 : (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO));
4436 :
4437 152407322 : if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
4438 76203841 : XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
4439 0 : xfs_bmap_mark_sick(ip, whichfork);
4440 0 : return -EFSCORRUPTED;
4441 : }
4442 :
4443 152406962 : if (xfs_is_shutdown(mp))
4444 : return -EIO;
4445 :
4446 76203423 : XFS_STATS_INC(mp, xs_blk_mapw);
4447 :
4448 76203423 : error = xfs_iread_extents(tp, ip, whichfork);
4449 76202474 : if (error)
4450 0 : goto error0;
4451 :
4452 76202474 : if (!xfs_iext_lookup_extent(ip, ifp, bno, &bma.icur, &bma.got))
4453 39120366 : eof = true;
4454 76202792 : if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev))
4455 40145091 : bma.prev.br_startoff = NULLFILEOFF;
4456 76201496 : bma.minleft = xfs_bmapi_minleft(tp, ip, whichfork);
4457 :
4458 76199999 : n = 0;
4459 76199999 : end = bno + len;
4460 76199999 : obno = bno;
4461 76591187 : while (bno < end && n < *nmap) {
4462 76591187 : bool need_alloc = false, wasdelay = false;
4463 :
4464 : /* in hole or beyond EOF? */
4465 76591187 : if (eof || bma.got.br_startoff > bno) {
4466 : /*
4467 : * CoW fork conversions should /never/ hit EOF or
4468 : * holes. There should always be something for us
4469 : * to work on.
4470 : */
4471 48535539 : ASSERT(!((flags & XFS_BMAPI_CONVERT) &&
4472 : (flags & XFS_BMAPI_COWFORK)));
4473 :
4474 : need_alloc = true;
4475 28055648 : } else if (isnullstartblock(bma.got.br_startblock)) {
4476 45355 : wasdelay = true;
4477 : }
4478 :
4479 : /*
4480 : * First, deal with the hole before the allocated space
4481 : * that we found, if any.
4482 : */
4483 76591187 : if (need_alloc || wasdelay) {
4484 48580797 : bma.eof = eof;
4485 48580797 : bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4486 48580797 : bma.wasdel = wasdelay;
4487 48580797 : bma.offset = bno;
4488 48580797 : bma.flags = flags;
4489 :
4490 : /*
4491 : * There's a 32/64 bit type mismatch between the
4492 : * allocation length request (which can be 64 bits in
4493 : * length) and the bma length request, which is
4494 : * xfs_extlen_t and therefore 32 bits. Hence we have to
4495 : * check for 32-bit overflows and handle them here.
4496 : */
4497 48580797 : if (len > (xfs_filblks_t)XFS_MAX_BMBT_EXTLEN)
4498 6772 : bma.length = XFS_MAX_BMBT_EXTLEN;
4499 : else
4500 48574025 : bma.length = len;
4501 :
4502 48580797 : ASSERT(len > 0);
4503 48580797 : ASSERT(bma.length > 0);
4504 48580797 : error = xfs_bmapi_allocate(&bma);
4505 48579182 : if (error)
4506 1906 : goto error0;
4507 48577276 : if (bma.blkno == NULLFSBLOCK)
4508 : break;
4509 :
4510 : /*
4511 : * If this is a CoW allocation, record the data in
4512 : * the refcount btree for orphan recovery.
4513 : */
4514 48577236 : if (whichfork == XFS_COW_FORK)
4515 219820 : xfs_refcount_alloc_cow_extent(tp, bma.blkno,
4516 : bma.length);
4517 : }
4518 :
4519 : /* Deal with the allocated space we found. */
4520 76587626 : xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4521 : end, n, flags);
4522 :
4523 : /* Execute unwritten extent conversion if necessary */
4524 76586850 : error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4525 76587214 : if (error == -EAGAIN)
4526 389739 : continue;
4527 76197475 : if (error)
4528 7 : goto error0;
4529 :
4530 : /* update the extent map to return */
4531 76197468 : xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4532 :
4533 : /*
4534 : * If we're done, stop now. Stop when we've allocated
4535 : * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise
4536 : * the transaction may get too big.
4537 : */
4538 76198124 : if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4539 : break;
4540 :
4541 : /* Else go on to the next record. */
4542 80 : bma.prev = bma.got;
4543 80 : if (!xfs_iext_next_extent(ifp, &bma.icur, &bma.got))
4544 24 : eof = true;
4545 : }
4546 76198084 : *nmap = n;
4547 :
4548 76198084 : error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, &bma.logflags,
4549 : whichfork);
4550 76193993 : if (error)
4551 0 : goto error0;
4552 :
4553 76193993 : ASSERT(ifp->if_format != XFS_DINODE_FMT_BTREE ||
4554 : ifp->if_nextents > XFS_IFORK_MAXEXT(ip, whichfork));
4555 76193993 : xfs_bmapi_finish(&bma, whichfork, 0);
4556 76200119 : xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4557 : orig_nmap, *nmap);
4558 76200119 : return 0;
4559 1913 : error0:
4560 1913 : xfs_bmapi_finish(&bma, whichfork, error);
4561 1913 : return error;
4562 : }
4563 :
4564 : /*
4565 : * Convert an existing delalloc extent to real blocks based on file offset. This
4566 : * attempts to allocate the entire delalloc extent and may require multiple
4567 : * invocations to allocate the target offset if a large enough physical extent
4568 : * is not available.
4569 : */
4570 : int
4571 15818941 : xfs_bmapi_convert_delalloc(
4572 : struct xfs_inode *ip,
4573 : int whichfork,
4574 : xfs_off_t offset,
4575 : struct iomap *iomap,
4576 : unsigned int *seq)
4577 : {
4578 15818941 : struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
4579 15818960 : struct xfs_mount *mp = ip->i_mount;
4580 15818960 : xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
4581 15818960 : struct xfs_bmalloca bma = { NULL };
4582 15818960 : uint16_t flags = 0;
4583 15818960 : struct xfs_trans *tp;
4584 15818960 : int error;
4585 :
4586 15818960 : if (whichfork == XFS_COW_FORK)
4587 250431 : flags |= IOMAP_F_SHARED;
4588 :
4589 : /*
4590 : * Space for the extent and indirect blocks was reserved when the
4591 : * delalloc extent was created so there's no need to do so here.
4592 : */
4593 15818960 : error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0,
4594 : XFS_TRANS_RESERVE, &tp);
4595 15819051 : if (error)
4596 : return error;
4597 :
4598 15819040 : xfs_ilock(ip, XFS_ILOCK_EXCL);
4599 15819049 : xfs_trans_ijoin(tp, ip, 0);
4600 :
4601 15818637 : error = xfs_iext_count_may_overflow(ip, whichfork,
4602 : XFS_IEXT_ADD_NOSPLIT_CNT);
4603 15819108 : if (error == -EFBIG)
4604 1 : error = xfs_iext_count_upgrade(tp, ip,
4605 : XFS_IEXT_ADD_NOSPLIT_CNT);
4606 15819108 : if (error)
4607 1 : goto out_trans_cancel;
4608 :
4609 15819107 : if (!xfs_iext_lookup_extent(ip, ifp, offset_fsb, &bma.icur, &bma.got) ||
4610 15818957 : bma.got.br_startoff > offset_fsb) {
4611 : /*
4612 : * No extent found in the range we are trying to convert. This
4613 : * should only happen for the COW fork, where another thread
4614 : * might have moved the extent to the data fork in the meantime.
4615 : */
4616 0 : WARN_ON_ONCE(whichfork != XFS_COW_FORK);
4617 0 : error = -EAGAIN;
4618 0 : goto out_trans_cancel;
4619 : }
4620 :
4621 : /*
4622 : * If we find a real extent here we raced with another thread converting
4623 : * the extent. Just return the real extent at this offset.
4624 : */
4625 15818957 : if (!isnullstartblock(bma.got.br_startblock)) {
4626 133934 : xfs_bmbt_to_iomap(ip, iomap, &bma.got, 0, flags,
4627 : xfs_iomap_inode_sequence(ip, flags));
4628 133595 : *seq = READ_ONCE(ifp->if_seq);
4629 133595 : goto out_trans_cancel;
4630 : }
4631 :
4632 15685023 : bma.tp = tp;
4633 15685023 : bma.ip = ip;
4634 15685023 : bma.wasdel = true;
4635 15685023 : bma.offset = bma.got.br_startoff;
4636 15685023 : bma.length = max_t(xfs_filblks_t, bma.got.br_blockcount,
4637 : XFS_MAX_BMBT_EXTLEN);
4638 15685023 : bma.minleft = xfs_bmapi_minleft(tp, ip, whichfork);
4639 :
4640 : /*
4641 : * When we're converting the delalloc reservations backing dirty pages
4642 : * in the page cache, we must be careful about how we create the new
4643 : * extents:
4644 : *
4645 : * New CoW fork extents are created unwritten, turned into real extents
4646 : * when we're about to write the data to disk, and mapped into the data
4647 : * fork after the write finishes. End of story.
4648 : *
4649 : * New data fork extents must be mapped in as unwritten and converted
4650 : * to real extents after the write succeeds to avoid exposing stale
4651 : * disk contents if we crash.
4652 : */
4653 15684984 : bma.flags = XFS_BMAPI_PREALLOC;
4654 15684984 : if (whichfork == XFS_COW_FORK)
4655 116869 : bma.flags |= XFS_BMAPI_COWFORK;
4656 :
4657 15684984 : if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev))
4658 5170861 : bma.prev.br_startoff = NULLFILEOFF;
4659 :
4660 15684984 : error = xfs_bmapi_allocate(&bma);
4661 15684945 : if (error)
4662 743 : goto out_finish;
4663 :
4664 15684202 : error = -ENOSPC;
4665 15684202 : if (WARN_ON_ONCE(bma.blkno == NULLFSBLOCK))
4666 0 : goto out_finish;
4667 31368404 : if (WARN_ON_ONCE(!xfs_valid_startblock(ip, bma.got.br_startblock))) {
4668 0 : xfs_bmap_mark_sick(ip, whichfork);
4669 0 : error = -EFSCORRUPTED;
4670 0 : goto out_finish;
4671 : }
4672 :
4673 15684202 : XFS_STATS_ADD(mp, xs_xstrat_bytes, XFS_FSB_TO_B(mp, bma.length));
4674 15684202 : XFS_STATS_INC(mp, xs_xstrat_quick);
4675 :
4676 15684202 : ASSERT(!isnullstartblock(bma.got.br_startblock));
4677 15684202 : xfs_bmbt_to_iomap(ip, iomap, &bma.got, 0, flags,
4678 : xfs_iomap_inode_sequence(ip, flags));
4679 15684198 : *seq = READ_ONCE(ifp->if_seq);
4680 :
4681 15684198 : if (whichfork == XFS_COW_FORK)
4682 116849 : xfs_refcount_alloc_cow_extent(tp, bma.blkno, bma.length);
4683 :
4684 15684217 : error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, &bma.logflags,
4685 : whichfork);
4686 15684210 : if (error)
4687 0 : goto out_finish;
4688 :
4689 15684210 : xfs_bmapi_finish(&bma, whichfork, 0);
4690 15684232 : error = xfs_trans_commit(tp);
4691 15684283 : xfs_iunlock(ip, XFS_ILOCK_EXCL);
4692 15684283 : return error;
4693 :
4694 743 : out_finish:
4695 743 : xfs_bmapi_finish(&bma, whichfork, error);
4696 134339 : out_trans_cancel:
4697 134339 : xfs_trans_cancel(tp);
4698 134854 : xfs_iunlock(ip, XFS_ILOCK_EXCL);
4699 134854 : return error;
4700 : }
4701 :
4702 : int
4703 53297250 : xfs_bmapi_remap(
4704 : struct xfs_trans *tp,
4705 : struct xfs_inode *ip,
4706 : xfs_fileoff_t bno,
4707 : xfs_filblks_t len,
4708 : xfs_fsblock_t startblock,
4709 : uint32_t flags)
4710 : {
4711 53297250 : struct xfs_mount *mp = ip->i_mount;
4712 53297250 : struct xfs_ifork *ifp;
4713 53297250 : struct xfs_btree_cur *cur = NULL;
4714 53297250 : struct xfs_bmbt_irec got;
4715 53297250 : struct xfs_iext_cursor icur;
4716 53297250 : int whichfork = xfs_bmapi_whichfork(flags);
4717 53297250 : int logflags = 0, error;
4718 :
4719 53297250 : ifp = xfs_ifork_ptr(ip, whichfork);
4720 53297250 : ASSERT(len > 0);
4721 53297250 : ASSERT(len <= (xfs_filblks_t)XFS_MAX_BMBT_EXTLEN);
4722 53297250 : ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4723 53297250 : ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC |
4724 : XFS_BMAPI_NORMAP)));
4725 53297250 : ASSERT((flags & (XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC)) !=
4726 : (XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC));
4727 :
4728 106594500 : if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
4729 53297250 : XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
4730 0 : xfs_bmap_mark_sick(ip, whichfork);
4731 0 : return -EFSCORRUPTED;
4732 : }
4733 :
4734 106594500 : if (xfs_is_shutdown(mp))
4735 : return -EIO;
4736 :
4737 53297247 : error = xfs_iread_extents(tp, ip, whichfork);
4738 53297247 : if (error)
4739 : return error;
4740 :
4741 53297247 : if (xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) {
4742 : /* make sure we only reflink into a hole. */
4743 3580328 : ASSERT(got.br_startoff > bno);
4744 3580328 : ASSERT(got.br_startoff - bno >= len);
4745 : }
4746 :
4747 53297247 : ip->i_nblocks += len;
4748 53297247 : xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4749 :
4750 53297247 : if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
4751 51498850 : cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
4752 51498850 : cur->bc_ino.flags = 0;
4753 : }
4754 :
4755 53297247 : got.br_startoff = bno;
4756 53297247 : got.br_startblock = startblock;
4757 53297247 : got.br_blockcount = len;
4758 53297247 : if (flags & XFS_BMAPI_PREALLOC)
4759 264542 : got.br_state = XFS_EXT_UNWRITTEN;
4760 : else
4761 53032705 : got.br_state = XFS_EXT_NORM;
4762 :
4763 53297247 : error = xfs_bmap_add_extent_hole_real(tp, ip, whichfork, &icur,
4764 : &cur, &got, &logflags, flags);
4765 53297246 : if (error)
4766 1 : goto error0;
4767 :
4768 53297245 : error = xfs_bmap_btree_to_extents(tp, ip, cur, &logflags, whichfork);
4769 :
4770 53297247 : error0:
4771 53297247 : if (ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS)
4772 51568925 : logflags &= ~XFS_ILOG_DEXT;
4773 1728322 : else if (ip->i_df.if_format != XFS_DINODE_FMT_BTREE)
4774 1728322 : logflags &= ~XFS_ILOG_DBROOT;
4775 :
4776 53297247 : if (logflags)
4777 52470813 : xfs_trans_log_inode(tp, ip, logflags);
4778 53297246 : if (cur)
4779 51572484 : xfs_btree_del_cursor(cur, error);
4780 : return error;
4781 : }
4782 :
4783 : /*
4784 : * When a delalloc extent is split (e.g., due to a hole punch), the original
4785 : * indlen reservation must be shared across the two new extents that are left
4786 : * behind.
4787 : *
4788 : * Given the original reservation and the worst case indlen for the two new
4789 : * extents (as calculated by xfs_bmap_worst_indlen()), split the original
4790 : * reservation fairly across the two new extents. If necessary, steal available
4791 : * blocks from a deleted extent to make up a reservation deficiency (e.g., if
4792 : * ores == 1). The number of stolen blocks is returned. The availability and
4793 : * subsequent accounting of stolen blocks is the responsibility of the caller.
4794 : */
4795 : static xfs_filblks_t
4796 585 : xfs_bmap_split_indlen(
4797 : xfs_filblks_t ores, /* original res. */
4798 : xfs_filblks_t *indlen1, /* ext1 worst indlen */
4799 : xfs_filblks_t *indlen2, /* ext2 worst indlen */
4800 : xfs_filblks_t avail) /* stealable blocks */
4801 : {
4802 585 : xfs_filblks_t len1 = *indlen1;
4803 585 : xfs_filblks_t len2 = *indlen2;
4804 585 : xfs_filblks_t nres = len1 + len2; /* new total res. */
4805 585 : xfs_filblks_t stolen = 0;
4806 585 : xfs_filblks_t resfactor;
4807 :
4808 : /*
4809 : * Steal as many blocks as we can to try and satisfy the worst case
4810 : * indlen for both new extents.
4811 : */
4812 585 : if (ores < nres && avail)
4813 585 : stolen = XFS_FILBLKS_MIN(nres - ores, avail);
4814 585 : ores += stolen;
4815 :
4816 : /* nothing else to do if we've satisfied the new reservation */
4817 585 : if (ores >= nres)
4818 : return stolen;
4819 :
4820 : /*
4821 : * We can't meet the total required reservation for the two extents.
4822 : * Calculate the percent of the overall shortage between both extents
4823 : * and apply this percentage to each of the requested indlen values.
4824 : * This distributes the shortage fairly and reduces the chances that one
4825 : * of the two extents is left with nothing when extents are repeatedly
4826 : * split.
4827 : */
4828 96 : resfactor = (ores * 100);
4829 96 : do_div(resfactor, nres);
4830 96 : len1 *= resfactor;
4831 96 : do_div(len1, 100);
4832 96 : len2 *= resfactor;
4833 96 : do_div(len2, 100);
4834 96 : ASSERT(len1 + len2 <= ores);
4835 96 : ASSERT(len1 < *indlen1 && len2 < *indlen2);
4836 :
4837 : /*
4838 : * Hand out the remainder to each extent. If one of the two reservations
4839 : * is zero, we want to make sure that one gets a block first. The loop
4840 : * below starts with len1, so hand len2 a block right off the bat if it
4841 : * is zero.
4842 : */
4843 96 : ores -= (len1 + len2);
4844 96 : ASSERT((*indlen1 - len1) + (*indlen2 - len2) >= ores);
4845 96 : if (ores && !len2 && *indlen2) {
4846 0 : len2++;
4847 0 : ores--;
4848 : }
4849 158 : while (ores) {
4850 96 : if (len1 < *indlen1) {
4851 96 : len1++;
4852 96 : ores--;
4853 : }
4854 96 : if (!ores)
4855 : break;
4856 62 : if (len2 < *indlen2) {
4857 62 : len2++;
4858 62 : ores--;
4859 : }
4860 : }
4861 :
4862 96 : *indlen1 = len1;
4863 96 : *indlen2 = len2;
4864 :
4865 96 : return stolen;
4866 : }
4867 :
4868 : int
4869 1187498 : xfs_bmap_del_extent_delay(
4870 : struct xfs_inode *ip,
4871 : int whichfork,
4872 : struct xfs_iext_cursor *icur,
4873 : struct xfs_bmbt_irec *got,
4874 : struct xfs_bmbt_irec *del)
4875 : {
4876 1187498 : struct xfs_mount *mp = ip->i_mount;
4877 1187498 : struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
4878 1187970 : struct xfs_bmbt_irec new;
4879 1187970 : int64_t da_old, da_new, da_diff = 0;
4880 1187970 : xfs_fileoff_t del_endoff, got_endoff;
4881 1187970 : xfs_filblks_t got_indlen, new_indlen, stolen;
4882 1187970 : uint32_t state = xfs_bmap_fork_to_state(whichfork);
4883 1187970 : int error = 0;
4884 1187970 : bool isrt;
4885 :
4886 1187970 : XFS_STATS_INC(mp, xs_del_exlist);
4887 :
4888 1187970 : isrt = xfs_ifork_is_realtime(ip, whichfork);
4889 1187943 : del_endoff = del->br_startoff + del->br_blockcount;
4890 1187943 : got_endoff = got->br_startoff + got->br_blockcount;
4891 1187943 : da_old = startblockval(got->br_startblock);
4892 1187943 : da_new = 0;
4893 :
4894 1187943 : ASSERT(del->br_blockcount > 0);
4895 1187943 : ASSERT(got->br_startoff <= del->br_startoff);
4896 1187943 : ASSERT(got_endoff >= del_endoff);
4897 :
4898 1187943 : if (isrt) {
4899 0 : uint64_t rtexts = XFS_FSB_TO_B(mp, del->br_blockcount);
4900 :
4901 0 : do_div(rtexts, mp->m_sb.sb_rextsize);
4902 0 : xfs_mod_frextents(mp, rtexts);
4903 : }
4904 :
4905 : /*
4906 : * Update the inode delalloc counter now and wait to update the
4907 : * sb counters as we might have to borrow some blocks for the
4908 : * indirect block accounting.
4909 : */
4910 1187943 : ASSERT(!isrt);
4911 1187943 : error = xfs_quota_unreserve_blkres(ip, del->br_blockcount);
4912 1187914 : if (error)
4913 : return error;
4914 1187914 : ip->i_delayed_blks -= del->br_blockcount;
4915 :
4916 1187914 : if (got->br_startoff == del->br_startoff)
4917 563109 : state |= BMAP_LEFT_FILLING;
4918 1187914 : if (got_endoff == del_endoff)
4919 1186159 : state |= BMAP_RIGHT_FILLING;
4920 :
4921 1187914 : switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
4922 561709 : case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
4923 : /*
4924 : * Matches the whole extent. Delete the entry.
4925 : */
4926 561709 : xfs_iext_remove(ip, icur, state);
4927 561630 : xfs_iext_prev(ifp, icur);
4928 561630 : break;
4929 1287 : case BMAP_LEFT_FILLING:
4930 : /*
4931 : * Deleting the first part of the extent.
4932 : */
4933 1287 : got->br_startoff = del_endoff;
4934 1287 : got->br_blockcount -= del->br_blockcount;
4935 1287 : da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4936 : got->br_blockcount), da_old);
4937 1287 : got->br_startblock = nullstartblock((int)da_new);
4938 1287 : xfs_iext_update_extent(ip, state, icur, got);
4939 1287 : break;
4940 624333 : case BMAP_RIGHT_FILLING:
4941 : /*
4942 : * Deleting the last part of the extent.
4943 : */
4944 624333 : got->br_blockcount = got->br_blockcount - del->br_blockcount;
4945 624333 : da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4946 : got->br_blockcount), da_old);
4947 624341 : got->br_startblock = nullstartblock((int)da_new);
4948 624340 : xfs_iext_update_extent(ip, state, icur, got);
4949 624340 : break;
4950 585 : case 0:
4951 : /*
4952 : * Deleting the middle of the extent.
4953 : *
4954 : * Distribute the original indlen reservation across the two new
4955 : * extents. Steal blocks from the deleted extent if necessary.
4956 : * Stealing blocks simply fudges the fdblocks accounting below.
4957 : * Warn if either of the new indlen reservations is zero as this
4958 : * can lead to delalloc problems.
4959 : */
4960 585 : got->br_blockcount = del->br_startoff - got->br_startoff;
4961 585 : got_indlen = xfs_bmap_worst_indlen(ip, got->br_blockcount);
4962 :
4963 585 : new.br_blockcount = got_endoff - del_endoff;
4964 585 : new_indlen = xfs_bmap_worst_indlen(ip, new.br_blockcount);
4965 :
4966 585 : WARN_ON_ONCE(!got_indlen || !new_indlen);
4967 585 : stolen = xfs_bmap_split_indlen(da_old, &got_indlen, &new_indlen,
4968 : del->br_blockcount);
4969 :
4970 585 : got->br_startblock = nullstartblock((int)got_indlen);
4971 :
4972 585 : new.br_startoff = del_endoff;
4973 585 : new.br_state = got->br_state;
4974 585 : new.br_startblock = nullstartblock((int)new_indlen);
4975 :
4976 585 : xfs_iext_update_extent(ip, state, icur, got);
4977 585 : xfs_iext_next(ifp, icur);
4978 585 : xfs_iext_insert(ip, icur, &new, state);
4979 :
4980 585 : da_new = got_indlen + new_indlen - stolen;
4981 585 : del->br_blockcount -= stolen;
4982 585 : break;
4983 : }
4984 :
4985 1187760 : ASSERT(da_old >= da_new);
4986 1187760 : da_diff = da_old - da_new;
4987 1187760 : if (!isrt)
4988 1187911 : da_diff += del->br_blockcount;
4989 1187760 : if (da_diff) {
4990 1187670 : xfs_mod_fdblocks(mp, da_diff, false);
4991 1187850 : xfs_mod_delalloc(mp, -da_diff);
4992 : }
4993 : return error;
4994 : }
4995 :
4996 : void
4997 1960109 : xfs_bmap_del_extent_cow(
4998 : struct xfs_inode *ip,
4999 : struct xfs_iext_cursor *icur,
5000 : struct xfs_bmbt_irec *got,
5001 : struct xfs_bmbt_irec *del)
5002 : {
5003 1960109 : struct xfs_mount *mp = ip->i_mount;
5004 1960109 : struct xfs_ifork *ifp = xfs_ifork_ptr(ip, XFS_COW_FORK);
5005 1960109 : struct xfs_bmbt_irec new;
5006 1960109 : xfs_fileoff_t del_endoff, got_endoff;
5007 1960109 : uint32_t state = BMAP_COWFORK;
5008 :
5009 1960109 : XFS_STATS_INC(mp, xs_del_exlist);
5010 :
5011 1960109 : del_endoff = del->br_startoff + del->br_blockcount;
5012 1960109 : got_endoff = got->br_startoff + got->br_blockcount;
5013 :
5014 1960109 : ASSERT(del->br_blockcount > 0);
5015 1960109 : ASSERT(got->br_startoff <= del->br_startoff);
5016 1960109 : ASSERT(got_endoff >= del_endoff);
5017 1960109 : ASSERT(!isnullstartblock(got->br_startblock));
5018 :
5019 1960109 : if (got->br_startoff == del->br_startoff)
5020 1946977 : state |= BMAP_LEFT_FILLING;
5021 1960109 : if (got_endoff == del_endoff)
5022 1667137 : state |= BMAP_RIGHT_FILLING;
5023 :
5024 1960109 : switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
5025 1654469 : case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
5026 : /*
5027 : * Matches the whole extent. Delete the entry.
5028 : */
5029 1654469 : xfs_iext_remove(ip, icur, state);
5030 1654469 : xfs_iext_prev(ifp, icur);
5031 1654469 : break;
5032 292508 : case BMAP_LEFT_FILLING:
5033 : /*
5034 : * Deleting the first part of the extent.
5035 : */
5036 292508 : got->br_startoff = del_endoff;
5037 292508 : got->br_blockcount -= del->br_blockcount;
5038 292508 : got->br_startblock = del->br_startblock + del->br_blockcount;
5039 292508 : xfs_iext_update_extent(ip, state, icur, got);
5040 292508 : break;
5041 12668 : case BMAP_RIGHT_FILLING:
5042 : /*
5043 : * Deleting the last part of the extent.
5044 : */
5045 12668 : got->br_blockcount -= del->br_blockcount;
5046 12668 : xfs_iext_update_extent(ip, state, icur, got);
5047 12668 : break;
5048 464 : case 0:
5049 : /*
5050 : * Deleting the middle of the extent.
5051 : */
5052 464 : got->br_blockcount = del->br_startoff - got->br_startoff;
5053 :
5054 464 : new.br_startoff = del_endoff;
5055 464 : new.br_blockcount = got_endoff - del_endoff;
5056 464 : new.br_state = got->br_state;
5057 464 : new.br_startblock = del->br_startblock + del->br_blockcount;
5058 :
5059 464 : xfs_iext_update_extent(ip, state, icur, got);
5060 464 : xfs_iext_next(ifp, icur);
5061 464 : xfs_iext_insert(ip, icur, &new, state);
5062 464 : break;
5063 : }
5064 1960109 : ip->i_delayed_blks -= del->br_blockcount;
5065 1960109 : }
5066 :
5067 : /*
5068 : * Called by xfs_bmapi to update file extent records and the btree
5069 : * after removing space.
5070 : */
5071 : STATIC int /* error */
5072 93811630 : xfs_bmap_del_extent_real(
5073 : xfs_inode_t *ip, /* incore inode pointer */
5074 : xfs_trans_t *tp, /* current transaction pointer */
5075 : struct xfs_iext_cursor *icur,
5076 : struct xfs_btree_cur *cur, /* if null, not a btree */
5077 : xfs_bmbt_irec_t *del, /* data to remove from extents */
5078 : int *logflagsp, /* inode logging flags */
5079 : int whichfork, /* data or attr fork */
5080 : uint32_t bflags) /* bmapi flags */
5081 : {
5082 93811630 : xfs_fsblock_t del_endblock=0; /* first block past del */
5083 93811630 : xfs_fileoff_t del_endoff; /* first offset past del */
5084 93811630 : int do_fx; /* free extent at end of routine */
5085 93811630 : int error; /* error return value */
5086 93811630 : int flags = 0;/* inode logging flags */
5087 93811630 : struct xfs_bmbt_irec got; /* current extent entry */
5088 93811630 : xfs_fileoff_t got_endoff; /* first offset past got */
5089 93811630 : int i; /* temp state */
5090 93811630 : struct xfs_ifork *ifp; /* inode fork pointer */
5091 93811630 : xfs_mount_t *mp; /* mount structure */
5092 93811630 : xfs_filblks_t nblks; /* quota/sb block count */
5093 93811630 : xfs_bmbt_irec_t new; /* new record to be inserted */
5094 : /* REFERENCED */
5095 93811630 : uint qfield; /* quota field to update */
5096 93811630 : uint32_t state = xfs_bmap_fork_to_state(whichfork);
5097 93811630 : struct xfs_bmbt_irec old;
5098 :
5099 93811630 : mp = ip->i_mount;
5100 93811630 : XFS_STATS_INC(mp, xs_del_exlist);
5101 :
5102 93811630 : ifp = xfs_ifork_ptr(ip, whichfork);
5103 93806513 : ASSERT(del->br_blockcount > 0);
5104 93806513 : xfs_iext_get_extent(ifp, icur, &got);
5105 93807576 : ASSERT(got.br_startoff <= del->br_startoff);
5106 93807576 : del_endoff = del->br_startoff + del->br_blockcount;
5107 93807576 : got_endoff = got.br_startoff + got.br_blockcount;
5108 93807576 : ASSERT(got_endoff >= del_endoff);
5109 93807576 : ASSERT(!isnullstartblock(got.br_startblock));
5110 93807576 : qfield = 0;
5111 93807576 : error = 0;
5112 :
5113 : /*
5114 : * If it's the case where the directory code is running with no block
5115 : * reservation, and the deleted block is in the middle of its extent,
5116 : * and the resulting insert of an extent would cause transformation to
5117 : * btree format, then reject it. The calling code will then swap blocks
5118 : * around instead. We have to do this now, rather than waiting for the
5119 : * conversion to btree format, since the transaction will be dirty then.
5120 : */
5121 93807576 : if (tp->t_blk_res == 0 &&
5122 77084535 : ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
5123 35894741 : ifp->if_nextents >= XFS_IFORK_MAXEXT(ip, whichfork) &&
5124 208614 : del->br_startoff > got.br_startoff && del_endoff < got_endoff)
5125 : return -ENOSPC;
5126 :
5127 93807576 : flags = XFS_ILOG_CORE;
5128 93807576 : if (xfs_ifork_is_realtime(ip, whichfork)) {
5129 3204583 : if (!(bflags & XFS_BMAPI_REMAP)) {
5130 3204583 : error = xfs_rtfree_blocks(tp, del->br_startblock,
5131 : del->br_blockcount);
5132 3204583 : if (error)
5133 0 : goto done;
5134 : }
5135 : do_fx = 0;
5136 : qfield = XFS_TRANS_DQ_RTBCOUNT;
5137 : } else {
5138 : do_fx = 1;
5139 : qfield = XFS_TRANS_DQ_BCOUNT;
5140 : }
5141 93806115 : nblks = del->br_blockcount;
5142 :
5143 93806115 : del_endblock = del->br_startblock + del->br_blockcount;
5144 93806115 : if (cur) {
5145 50116788 : error = xfs_bmbt_lookup_eq(cur, &got, &i);
5146 50116803 : if (error)
5147 0 : goto done;
5148 50116803 : if (XFS_IS_CORRUPT(mp, i != 1)) {
5149 0 : xfs_btree_mark_sick(cur);
5150 0 : error = -EFSCORRUPTED;
5151 0 : goto done;
5152 : }
5153 : }
5154 :
5155 93806130 : if (got.br_startoff == del->br_startoff)
5156 84703596 : state |= BMAP_LEFT_FILLING;
5157 93806130 : if (got_endoff == del_endoff)
5158 84139304 : state |= BMAP_RIGHT_FILLING;
5159 :
5160 93806130 : switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
5161 82617888 : case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
5162 : /*
5163 : * Matches the whole extent. Delete the entry.
5164 : */
5165 82617888 : xfs_iext_remove(ip, icur, state);
5166 82620942 : xfs_iext_prev(ifp, icur);
5167 82622195 : ifp->if_nextents--;
5168 :
5169 82622195 : flags |= XFS_ILOG_CORE;
5170 82622195 : if (!cur) {
5171 39885954 : flags |= xfs_ilog_fext(whichfork);
5172 39885954 : break;
5173 : }
5174 42736241 : if ((error = xfs_btree_delete(cur, &i)))
5175 0 : goto done;
5176 42736241 : if (XFS_IS_CORRUPT(mp, i != 1)) {
5177 0 : xfs_btree_mark_sick(cur);
5178 0 : error = -EFSCORRUPTED;
5179 0 : goto done;
5180 : }
5181 : break;
5182 2081027 : case BMAP_LEFT_FILLING:
5183 : /*
5184 : * Deleting the first part of the extent.
5185 : */
5186 2081027 : got.br_startoff = del_endoff;
5187 2081027 : got.br_startblock = del_endblock;
5188 2081027 : got.br_blockcount -= del->br_blockcount;
5189 2081027 : xfs_iext_update_extent(ip, state, icur, &got);
5190 2081032 : if (!cur) {
5191 1238614 : flags |= xfs_ilog_fext(whichfork);
5192 1238614 : break;
5193 : }
5194 842418 : error = xfs_bmbt_update(cur, &got);
5195 842386 : if (error)
5196 0 : goto done;
5197 : break;
5198 1517479 : case BMAP_RIGHT_FILLING:
5199 : /*
5200 : * Deleting the last part of the extent.
5201 : */
5202 1517479 : got.br_blockcount -= del->br_blockcount;
5203 1517479 : xfs_iext_update_extent(ip, state, icur, &got);
5204 1517481 : if (!cur) {
5205 1141263 : flags |= xfs_ilog_fext(whichfork);
5206 1141263 : break;
5207 : }
5208 376218 : error = xfs_bmbt_update(cur, &got);
5209 376218 : if (error)
5210 0 : goto done;
5211 : break;
5212 7589736 : case 0:
5213 : /*
5214 : * Deleting the middle of the extent.
5215 : */
5216 :
5217 7589736 : old = got;
5218 :
5219 7589736 : got.br_blockcount = del->br_startoff - got.br_startoff;
5220 7589736 : xfs_iext_update_extent(ip, state, icur, &got);
5221 :
5222 7589738 : new.br_startoff = del_endoff;
5223 7589738 : new.br_blockcount = got_endoff - del_endoff;
5224 7589738 : new.br_state = got.br_state;
5225 7589738 : new.br_startblock = del_endblock;
5226 :
5227 7589738 : flags |= XFS_ILOG_CORE;
5228 7589738 : if (cur) {
5229 6161926 : error = xfs_bmbt_update(cur, &got);
5230 6161926 : if (error)
5231 0 : goto done;
5232 6161926 : error = xfs_btree_increment(cur, 0, &i);
5233 6161926 : if (error)
5234 0 : goto done;
5235 6161926 : cur->bc_rec.b = new;
5236 6161926 : error = xfs_btree_insert(cur, &i);
5237 6161925 : if (error && error != -ENOSPC)
5238 0 : goto done;
5239 : /*
5240 : * If get no-space back from btree insert, it tried a
5241 : * split, and we have a zero block reservation. Fix up
5242 : * our state and return the error.
5243 : */
5244 6161925 : if (error == -ENOSPC) {
5245 : /*
5246 : * Reset the cursor, don't trust it after any
5247 : * insert operation.
5248 : */
5249 0 : error = xfs_bmbt_lookup_eq(cur, &got, &i);
5250 0 : if (error)
5251 0 : goto done;
5252 0 : if (XFS_IS_CORRUPT(mp, i != 1)) {
5253 0 : xfs_btree_mark_sick(cur);
5254 0 : error = -EFSCORRUPTED;
5255 0 : goto done;
5256 : }
5257 : /*
5258 : * Update the btree record back
5259 : * to the original value.
5260 : */
5261 0 : error = xfs_bmbt_update(cur, &old);
5262 0 : if (error)
5263 0 : goto done;
5264 : /*
5265 : * Reset the extent record back
5266 : * to the original value.
5267 : */
5268 0 : xfs_iext_update_extent(ip, state, icur, &old);
5269 0 : flags = 0;
5270 0 : error = -ENOSPC;
5271 0 : goto done;
5272 : }
5273 6161925 : if (XFS_IS_CORRUPT(mp, i != 1)) {
5274 0 : xfs_btree_mark_sick(cur);
5275 0 : error = -EFSCORRUPTED;
5276 0 : goto done;
5277 : }
5278 : } else
5279 1427829 : flags |= xfs_ilog_fext(whichfork);
5280 :
5281 7589737 : ifp->if_nextents++;
5282 7589737 : xfs_iext_next(ifp, icur);
5283 7589737 : xfs_iext_insert(ip, icur, &new, state);
5284 7589737 : break;
5285 : }
5286 :
5287 : /* remove reverse mapping */
5288 93810414 : xfs_rmap_unmap_extent(tp, ip, whichfork, del);
5289 :
5290 : /*
5291 : * If we need to, add to list of extents to delete.
5292 : */
5293 93812383 : if (do_fx && !(bflags & XFS_BMAPI_REMAP)) {
5294 84650032 : if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) {
5295 50195050 : xfs_refcount_decrease_extent(tp, del);
5296 : } else {
5297 34454982 : error = __xfs_free_extent_later(tp, del->br_startblock,
5298 : del->br_blockcount, NULL,
5299 : XFS_AG_RESV_NONE,
5300 34454982 : ((bflags & XFS_BMAPI_NODISCARD) ||
5301 34439378 : del->br_state == XFS_EXT_UNWRITTEN));
5302 34451902 : if (error)
5303 0 : goto done;
5304 : }
5305 : }
5306 :
5307 : /*
5308 : * Adjust inode # blocks in the file.
5309 : */
5310 93809415 : if (nblks)
5311 93809026 : ip->i_nblocks -= nblks;
5312 : /*
5313 : * Adjust quota data.
5314 : */
5315 93809415 : if (qfield && !(bflags & XFS_BMAPI_REMAP))
5316 87852664 : xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5317 :
5318 5956751 : done:
5319 93808035 : *logflagsp = flags;
5320 93808035 : return error;
5321 : }
5322 :
5323 : /*
5324 : * Unmap (remove) blocks from a file.
5325 : * If nexts is nonzero then the number of extents to remove is limited to
5326 : * that value. If not all extents in the block range can be removed then
5327 : * *done is set.
5328 : */
5329 : int /* error */
5330 78608057 : __xfs_bunmapi(
5331 : struct xfs_trans *tp, /* transaction pointer */
5332 : struct xfs_inode *ip, /* incore inode */
5333 : xfs_fileoff_t start, /* first file offset deleted */
5334 : xfs_filblks_t *rlen, /* i/o: amount remaining */
5335 : uint32_t flags, /* misc flags */
5336 : xfs_extnum_t nexts) /* number of extents max */
5337 : {
5338 78608057 : struct xfs_btree_cur *cur; /* bmap btree cursor */
5339 78608057 : struct xfs_bmbt_irec del; /* extent being deleted */
5340 78608057 : int error; /* error return value */
5341 78608057 : xfs_extnum_t extno; /* extent number in list */
5342 78608057 : struct xfs_bmbt_irec got; /* current extent record */
5343 78608057 : struct xfs_ifork *ifp; /* inode fork pointer */
5344 78608057 : int isrt; /* freeing in rt area */
5345 78608057 : int logflags; /* transaction logging flags */
5346 78608057 : xfs_extlen_t mod; /* rt extent offset */
5347 78608057 : struct xfs_mount *mp = ip->i_mount;
5348 78608057 : int tmp_logflags; /* partial logging flags */
5349 78608057 : int wasdel; /* was a delayed alloc extent */
5350 78608057 : int whichfork; /* data or attribute fork */
5351 78608057 : xfs_fsblock_t sum;
5352 78608057 : xfs_filblks_t len = *rlen; /* length to unmap in file */
5353 78608057 : xfs_fileoff_t end;
5354 78608057 : struct xfs_iext_cursor icur;
5355 78608057 : bool done = false;
5356 :
5357 78608057 : trace_xfs_bunmap(ip, start, len, flags, _RET_IP_);
5358 :
5359 78606833 : whichfork = xfs_bmapi_whichfork(flags);
5360 78606833 : ASSERT(whichfork != XFS_COW_FORK);
5361 78606833 : ifp = xfs_ifork_ptr(ip, whichfork);
5362 78607235 : if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp))) {
5363 0 : xfs_bmap_mark_sick(ip, whichfork);
5364 0 : return -EFSCORRUPTED;
5365 : }
5366 157214470 : if (xfs_is_shutdown(mp))
5367 : return -EIO;
5368 :
5369 78606906 : ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5370 78606391 : ASSERT(len > 0);
5371 78606391 : ASSERT(nexts >= 0);
5372 :
5373 78606391 : error = xfs_iread_extents(tp, ip, whichfork);
5374 78609425 : if (error)
5375 : return error;
5376 :
5377 78609559 : if (xfs_iext_count(ifp) == 0) {
5378 727011 : *rlen = 0;
5379 727011 : return 0;
5380 : }
5381 77878413 : XFS_STATS_INC(mp, xs_blk_unmap);
5382 77878413 : isrt = xfs_ifork_is_realtime(ip, whichfork);
5383 77875174 : end = start + len;
5384 :
5385 77875174 : if (!xfs_iext_lookup_extent_before(ip, ifp, &end, &icur, &got)) {
5386 219002 : *rlen = 0;
5387 219002 : return 0;
5388 : }
5389 77656033 : end--;
5390 :
5391 77656033 : logflags = 0;
5392 77656033 : if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
5393 29939284 : ASSERT(ifp->if_format == XFS_DINODE_FMT_BTREE);
5394 29939284 : cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5395 29939287 : cur->bc_ino.flags = 0;
5396 : } else
5397 47716749 : cur = NULL;
5398 :
5399 77656036 : if (isrt) {
5400 : /*
5401 : * Synchronize by locking the bitmap inode.
5402 : */
5403 2506874 : xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
5404 2506884 : xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5405 2506884 : xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
5406 2506884 : xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL);
5407 : }
5408 :
5409 : extno = 0;
5410 171510510 : while (end != (xfs_fileoff_t)-1 && end >= start &&
5411 : (nexts == 0 || extno < nexts)) {
5412 : /*
5413 : * Is the found extent after a hole in which end lives?
5414 : * Just back up to the previous extent, if so.
5415 : */
5416 95832380 : if (got.br_startoff > end &&
5417 0 : !xfs_iext_prev_extent(ifp, &icur, &got)) {
5418 : done = true;
5419 : break;
5420 : }
5421 : /*
5422 : * Is the last block of this extent before the range
5423 : * we're supposed to delete? If so, we're done.
5424 : */
5425 95832380 : end = XFS_FILEOFF_MIN(end,
5426 : got.br_startoff + got.br_blockcount - 1);
5427 95832380 : if (end < start)
5428 : break;
5429 : /*
5430 : * Then deal with the (possibly delayed) allocated space
5431 : * we found.
5432 : */
5433 94972396 : del = got;
5434 94972396 : wasdel = isnullstartblock(del.br_startblock);
5435 :
5436 94972396 : if (got.br_startoff < start) {
5437 9720162 : del.br_startoff = start;
5438 9720162 : del.br_blockcount -= start - got.br_startoff;
5439 9720162 : if (!wasdel)
5440 9107214 : del.br_startblock += start - got.br_startoff;
5441 : }
5442 94972396 : if (del.br_startoff + del.br_blockcount > end + 1)
5443 9671365 : del.br_blockcount = end + 1 - del.br_startoff;
5444 :
5445 94972396 : if (!isrt || (flags & XFS_BMAPI_REMAP))
5446 91767811 : goto delete;
5447 :
5448 3204585 : sum = del.br_startblock + del.br_blockcount;
5449 3204585 : div_u64_rem(sum, mp->m_sb.sb_rextsize, &mod);
5450 3204585 : if (mod) {
5451 : /*
5452 : * Realtime extent not lined up at the end.
5453 : * The extent could have been split into written
5454 : * and unwritten pieces, or we could just be
5455 : * unmapping part of it. But we can't really
5456 : * get rid of part of a realtime extent.
5457 : */
5458 0 : if (del.br_state == XFS_EXT_UNWRITTEN) {
5459 : /*
5460 : * This piece is unwritten, or we're not
5461 : * using unwritten extents. Skip over it.
5462 : */
5463 0 : ASSERT((flags & XFS_BMAPI_REMAP) || end >= mod);
5464 0 : end -= mod > del.br_blockcount ?
5465 0 : del.br_blockcount : mod;
5466 0 : if (end < got.br_startoff &&
5467 0 : !xfs_iext_prev_extent(ifp, &icur, &got)) {
5468 : done = true;
5469 : break;
5470 : }
5471 0 : continue;
5472 : }
5473 : /*
5474 : * It's written, turn it unwritten.
5475 : * This is better than zeroing it.
5476 : */
5477 0 : ASSERT(del.br_state == XFS_EXT_NORM);
5478 0 : ASSERT(tp->t_blk_res > 0);
5479 : /*
5480 : * If this spans a realtime extent boundary,
5481 : * chop it back to the start of the one we end at.
5482 : */
5483 0 : if (del.br_blockcount > mod) {
5484 0 : del.br_startoff += del.br_blockcount - mod;
5485 0 : del.br_startblock += del.br_blockcount - mod;
5486 0 : del.br_blockcount = mod;
5487 : }
5488 0 : del.br_state = XFS_EXT_UNWRITTEN;
5489 0 : error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5490 : whichfork, &icur, &cur, &del,
5491 : &logflags);
5492 0 : if (error)
5493 0 : goto error0;
5494 0 : goto nodelete;
5495 : }
5496 3204585 : div_u64_rem(del.br_startblock, mp->m_sb.sb_rextsize, &mod);
5497 3204585 : if (mod) {
5498 2 : xfs_extlen_t off = mp->m_sb.sb_rextsize - mod;
5499 :
5500 : /*
5501 : * Realtime extent is lined up at the end but not
5502 : * at the front. We'll get rid of full extents if
5503 : * we can.
5504 : */
5505 2 : if (del.br_blockcount > off) {
5506 0 : del.br_blockcount -= off;
5507 0 : del.br_startoff += off;
5508 0 : del.br_startblock += off;
5509 2 : } else if (del.br_startoff == start &&
5510 0 : (del.br_state == XFS_EXT_UNWRITTEN ||
5511 0 : tp->t_blk_res == 0)) {
5512 : /*
5513 : * Can't make it unwritten. There isn't
5514 : * a full extent here so just skip it.
5515 : */
5516 0 : ASSERT(end >= del.br_blockcount);
5517 0 : end -= del.br_blockcount;
5518 0 : if (got.br_startoff > end &&
5519 0 : !xfs_iext_prev_extent(ifp, &icur, &got)) {
5520 : done = true;
5521 : break;
5522 : }
5523 0 : continue;
5524 2 : } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5525 2 : struct xfs_bmbt_irec prev;
5526 2 : xfs_fileoff_t unwrite_start;
5527 :
5528 : /*
5529 : * This one is already unwritten.
5530 : * It must have a written left neighbor.
5531 : * Unwrite the killed part of that one and
5532 : * try again.
5533 : */
5534 2 : if (!xfs_iext_prev_extent(ifp, &icur, &prev))
5535 0 : ASSERT(0);
5536 2 : ASSERT(prev.br_state == XFS_EXT_NORM);
5537 2 : ASSERT(!isnullstartblock(prev.br_startblock));
5538 2 : ASSERT(del.br_startblock ==
5539 : prev.br_startblock + prev.br_blockcount);
5540 2 : unwrite_start = max3(start,
5541 : del.br_startoff - mod,
5542 : prev.br_startoff);
5543 2 : mod = unwrite_start - prev.br_startoff;
5544 2 : prev.br_startoff = unwrite_start;
5545 2 : prev.br_startblock += mod;
5546 2 : prev.br_blockcount -= mod;
5547 2 : prev.br_state = XFS_EXT_UNWRITTEN;
5548 2 : error = xfs_bmap_add_extent_unwritten_real(tp,
5549 : ip, whichfork, &icur, &cur,
5550 : &prev, &logflags);
5551 2 : if (error)
5552 0 : goto error0;
5553 2 : goto nodelete;
5554 : } else {
5555 0 : ASSERT(del.br_state == XFS_EXT_NORM);
5556 0 : del.br_state = XFS_EXT_UNWRITTEN;
5557 0 : error = xfs_bmap_add_extent_unwritten_real(tp,
5558 : ip, whichfork, &icur, &cur,
5559 : &del, &logflags);
5560 0 : if (error)
5561 0 : goto error0;
5562 0 : goto nodelete;
5563 : }
5564 : }
5565 :
5566 3204583 : delete:
5567 94972394 : if (wasdel) {
5568 1163997 : error = xfs_bmap_del_extent_delay(ip, whichfork, &icur,
5569 : &got, &del);
5570 : } else {
5571 93808397 : error = xfs_bmap_del_extent_real(ip, tp, &icur, cur,
5572 : &del, &tmp_logflags, whichfork,
5573 : flags);
5574 93809405 : logflags |= tmp_logflags;
5575 : }
5576 :
5577 94976967 : if (error)
5578 0 : goto error0;
5579 :
5580 94976967 : end = del.br_startoff - 1;
5581 94976969 : nodelete:
5582 : /*
5583 : * If not done go on to the next (previous) record.
5584 : */
5585 94976969 : if (end != (xfs_fileoff_t)-1 && end >= start) {
5586 51219274 : if (!xfs_iext_get_extent(ifp, &icur, &got) ||
5587 50616315 : (got.br_startoff > end &&
5588 486989 : !xfs_iext_prev_extent(ifp, &icur, &got))) {
5589 : done = true;
5590 : break;
5591 : }
5592 50096769 : extno++;
5593 : }
5594 : }
5595 76538114 : if (done || end == (xfs_fileoff_t)-1 || end < start)
5596 53513692 : *rlen = 0;
5597 : else
5598 24146920 : *rlen = end - start + 1;
5599 :
5600 : /*
5601 : * Convert to a btree if necessary.
5602 : */
5603 77660612 : if (xfs_bmap_needs_btree(ip, whichfork)) {
5604 56127 : ASSERT(cur == NULL);
5605 56127 : error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
5606 : &tmp_logflags, whichfork);
5607 56127 : logflags |= tmp_logflags;
5608 : } else {
5609 77600898 : error = xfs_bmap_btree_to_extents(tp, ip, cur, &logflags,
5610 : whichfork);
5611 : }
5612 :
5613 77652438 : error0:
5614 : /*
5615 : * Log everything. Do this after conversion, there's no point in
5616 : * logging the extent records if we've converted to btree format.
5617 : */
5618 81021969 : if ((logflags & xfs_ilog_fext(whichfork)) &&
5619 39972447 : ifp->if_format != XFS_DINODE_FMT_EXTENTS)
5620 56127 : logflags &= ~xfs_ilog_fext(whichfork);
5621 80968290 : else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5622 0 : ifp->if_format != XFS_DINODE_FMT_BTREE)
5623 0 : logflags &= ~xfs_ilog_fbroot(whichfork);
5624 : /*
5625 : * Log inode even in the error case, if the transaction
5626 : * is dirty we'll need to shut down the filesystem.
5627 : */
5628 77652438 : if (logflags)
5629 68925875 : xfs_trans_log_inode(tp, ip, logflags);
5630 77653254 : if (cur) {
5631 29991202 : if (!error)
5632 29995413 : cur->bc_ino.allocated = 0;
5633 29991202 : xfs_btree_del_cursor(cur, error);
5634 : }
5635 : return error;
5636 : }
5637 :
5638 : /* Unmap a range of a file. */
5639 : int
5640 36996823 : xfs_bunmapi(
5641 : xfs_trans_t *tp,
5642 : struct xfs_inode *ip,
5643 : xfs_fileoff_t bno,
5644 : xfs_filblks_t len,
5645 : uint32_t flags,
5646 : xfs_extnum_t nexts,
5647 : int *done)
5648 : {
5649 36996823 : int error;
5650 :
5651 36996823 : error = __xfs_bunmapi(tp, ip, bno, &len, flags, nexts);
5652 36995581 : *done = (len == 0);
5653 36995581 : return error;
5654 : }
5655 :
5656 : /*
5657 : * Determine whether an extent shift can be accomplished by a merge with the
5658 : * extent that precedes the target hole of the shift.
5659 : */
5660 : STATIC bool
5661 20047464 : xfs_bmse_can_merge(
5662 : struct xfs_bmbt_irec *left, /* preceding extent */
5663 : struct xfs_bmbt_irec *got, /* current extent to shift */
5664 : xfs_fileoff_t shift) /* shift fsb */
5665 : {
5666 20047464 : xfs_fileoff_t startoff;
5667 :
5668 20047464 : startoff = got->br_startoff - shift;
5669 :
5670 : /*
5671 : * The extent, once shifted, must be adjacent in-file and on-disk with
5672 : * the preceding extent.
5673 : */
5674 20047464 : if ((left->br_startoff + left->br_blockcount != startoff) ||
5675 18846154 : (left->br_startblock + left->br_blockcount != got->br_startblock) ||
5676 226700 : (left->br_state != got->br_state) ||
5677 5190 : (left->br_blockcount + got->br_blockcount > XFS_MAX_BMBT_EXTLEN))
5678 20042274 : return false;
5679 :
5680 : return true;
5681 : }
5682 :
5683 : /*
5684 : * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5685 : * hole in the file. If an extent shift would result in the extent being fully
5686 : * adjacent to the extent that currently precedes the hole, we can merge with
5687 : * the preceding extent rather than do the shift.
5688 : *
5689 : * This function assumes the caller has verified a shift-by-merge is possible
5690 : * with the provided extents via xfs_bmse_can_merge().
5691 : */
5692 : STATIC int
5693 2595 : xfs_bmse_merge(
5694 : struct xfs_trans *tp,
5695 : struct xfs_inode *ip,
5696 : int whichfork,
5697 : xfs_fileoff_t shift, /* shift fsb */
5698 : struct xfs_iext_cursor *icur,
5699 : struct xfs_bmbt_irec *got, /* extent to shift */
5700 : struct xfs_bmbt_irec *left, /* preceding extent */
5701 : struct xfs_btree_cur *cur,
5702 : int *logflags) /* output */
5703 : {
5704 2595 : struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
5705 2595 : struct xfs_bmbt_irec new;
5706 2595 : xfs_filblks_t blockcount;
5707 2595 : int error, i;
5708 2595 : struct xfs_mount *mp = ip->i_mount;
5709 :
5710 2595 : blockcount = left->br_blockcount + got->br_blockcount;
5711 :
5712 2595 : ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5713 2595 : ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5714 2595 : ASSERT(xfs_bmse_can_merge(left, got, shift));
5715 :
5716 2595 : new = *left;
5717 2595 : new.br_blockcount = blockcount;
5718 :
5719 : /*
5720 : * Update the on-disk extent count, the btree if necessary and log the
5721 : * inode.
5722 : */
5723 2595 : ifp->if_nextents--;
5724 2595 : *logflags |= XFS_ILOG_CORE;
5725 2595 : if (!cur) {
5726 2239 : *logflags |= XFS_ILOG_DEXT;
5727 2239 : goto done;
5728 : }
5729 :
5730 : /* lookup and remove the extent to merge */
5731 356 : error = xfs_bmbt_lookup_eq(cur, got, &i);
5732 356 : if (error)
5733 : return error;
5734 356 : if (XFS_IS_CORRUPT(mp, i != 1)) {
5735 0 : xfs_btree_mark_sick(cur);
5736 0 : return -EFSCORRUPTED;
5737 : }
5738 :
5739 356 : error = xfs_btree_delete(cur, &i);
5740 356 : if (error)
5741 : return error;
5742 356 : if (XFS_IS_CORRUPT(mp, i != 1)) {
5743 0 : xfs_btree_mark_sick(cur);
5744 0 : return -EFSCORRUPTED;
5745 : }
5746 :
5747 : /* lookup and update size of the previous extent */
5748 356 : error = xfs_bmbt_lookup_eq(cur, left, &i);
5749 356 : if (error)
5750 : return error;
5751 356 : if (XFS_IS_CORRUPT(mp, i != 1)) {
5752 0 : xfs_btree_mark_sick(cur);
5753 0 : return -EFSCORRUPTED;
5754 : }
5755 :
5756 356 : error = xfs_bmbt_update(cur, &new);
5757 356 : if (error)
5758 : return error;
5759 :
5760 : /* change to extent format if required after extent removal */
5761 356 : error = xfs_bmap_btree_to_extents(tp, ip, cur, logflags, whichfork);
5762 356 : if (error)
5763 : return error;
5764 :
5765 356 : done:
5766 2595 : xfs_iext_remove(ip, icur, 0);
5767 2595 : xfs_iext_prev(ifp, icur);
5768 5190 : xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur,
5769 : &new);
5770 :
5771 : /* update reverse mapping. rmap functions merge the rmaps for us */
5772 2595 : xfs_rmap_unmap_extent(tp, ip, whichfork, got);
5773 2595 : memcpy(&new, got, sizeof(new));
5774 2595 : new.br_startoff = left->br_startoff + left->br_blockcount;
5775 2595 : xfs_rmap_map_extent(tp, ip, whichfork, &new);
5776 2595 : return 0;
5777 : }
5778 :
5779 : static int
5780 20444338 : xfs_bmap_shift_update_extent(
5781 : struct xfs_trans *tp,
5782 : struct xfs_inode *ip,
5783 : int whichfork,
5784 : struct xfs_iext_cursor *icur,
5785 : struct xfs_bmbt_irec *got,
5786 : struct xfs_btree_cur *cur,
5787 : int *logflags,
5788 : xfs_fileoff_t startoff)
5789 : {
5790 20444338 : struct xfs_mount *mp = ip->i_mount;
5791 20444338 : struct xfs_bmbt_irec prev = *got;
5792 20444338 : int error, i;
5793 :
5794 20444338 : *logflags |= XFS_ILOG_CORE;
5795 :
5796 20444338 : got->br_startoff = startoff;
5797 :
5798 20444338 : if (cur) {
5799 18546973 : error = xfs_bmbt_lookup_eq(cur, &prev, &i);
5800 18546973 : if (error)
5801 : return error;
5802 18546973 : if (XFS_IS_CORRUPT(mp, i != 1)) {
5803 0 : xfs_btree_mark_sick(cur);
5804 0 : return -EFSCORRUPTED;
5805 : }
5806 :
5807 18546973 : error = xfs_bmbt_update(cur, got);
5808 18546973 : if (error)
5809 : return error;
5810 : } else {
5811 1897365 : *logflags |= XFS_ILOG_DEXT;
5812 : }
5813 :
5814 40888676 : xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur,
5815 : got);
5816 :
5817 : /* update reverse mapping */
5818 20444337 : xfs_rmap_unmap_extent(tp, ip, whichfork, &prev);
5819 20444338 : xfs_rmap_map_extent(tp, ip, whichfork, got);
5820 20444338 : return 0;
5821 : }
5822 :
5823 : int
5824 18963645 : xfs_bmap_collapse_extents(
5825 : struct xfs_trans *tp,
5826 : struct xfs_inode *ip,
5827 : xfs_fileoff_t *next_fsb,
5828 : xfs_fileoff_t offset_shift_fsb,
5829 : bool *done)
5830 : {
5831 18963645 : int whichfork = XFS_DATA_FORK;
5832 18963645 : struct xfs_mount *mp = ip->i_mount;
5833 18963645 : struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
5834 18963645 : struct xfs_btree_cur *cur = NULL;
5835 18963645 : struct xfs_bmbt_irec got, prev;
5836 18963645 : struct xfs_iext_cursor icur;
5837 18963645 : xfs_fileoff_t new_startoff;
5838 18963645 : int error = 0;
5839 18963645 : int logflags = 0;
5840 :
5841 37927290 : if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
5842 18963645 : XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
5843 0 : xfs_bmap_mark_sick(ip, whichfork);
5844 0 : return -EFSCORRUPTED;
5845 : }
5846 :
5847 37927290 : if (xfs_is_shutdown(mp))
5848 : return -EIO;
5849 :
5850 18963643 : ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL));
5851 :
5852 18963643 : error = xfs_iread_extents(tp, ip, whichfork);
5853 18963643 : if (error)
5854 : return error;
5855 :
5856 18963643 : if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
5857 17896167 : cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5858 17896167 : cur->bc_ino.flags = 0;
5859 : }
5860 :
5861 18963643 : if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) {
5862 45676 : *done = true;
5863 45676 : goto del_cursor;
5864 : }
5865 18917967 : if (XFS_IS_CORRUPT(mp, isnullstartblock(got.br_startblock))) {
5866 0 : xfs_bmap_mark_sick(ip, whichfork);
5867 0 : error = -EFSCORRUPTED;
5868 0 : goto del_cursor;
5869 : }
5870 :
5871 18917967 : new_startoff = got.br_startoff - offset_shift_fsb;
5872 18917967 : if (xfs_iext_peek_prev_extent(ifp, &icur, &prev)) {
5873 18854442 : if (new_startoff < prev.br_startoff + prev.br_blockcount) {
5874 0 : error = -EINVAL;
5875 0 : goto del_cursor;
5876 : }
5877 :
5878 18854442 : if (xfs_bmse_can_merge(&prev, &got, offset_shift_fsb)) {
5879 2595 : error = xfs_bmse_merge(tp, ip, whichfork,
5880 : offset_shift_fsb, &icur, &got, &prev,
5881 : cur, &logflags);
5882 2595 : if (error)
5883 0 : goto del_cursor;
5884 2595 : goto done;
5885 : }
5886 : } else {
5887 63525 : if (got.br_startoff < offset_shift_fsb) {
5888 0 : error = -EINVAL;
5889 0 : goto del_cursor;
5890 : }
5891 : }
5892 :
5893 18915372 : error = xfs_bmap_shift_update_extent(tp, ip, whichfork, &icur, &got,
5894 : cur, &logflags, new_startoff);
5895 18915372 : if (error)
5896 0 : goto del_cursor;
5897 :
5898 18915372 : done:
5899 18917967 : if (!xfs_iext_next_extent(ifp, &icur, &got)) {
5900 411771 : *done = true;
5901 411771 : goto del_cursor;
5902 : }
5903 :
5904 18506196 : *next_fsb = got.br_startoff;
5905 18963643 : del_cursor:
5906 18963643 : if (cur)
5907 17896167 : xfs_btree_del_cursor(cur, error);
5908 18963643 : if (logflags)
5909 18917967 : xfs_trans_log_inode(tp, ip, logflags);
5910 : return error;
5911 : }
5912 :
5913 : /* Make sure we won't be right-shifting an extent past the maximum bound. */
5914 : int
5915 379186 : xfs_bmap_can_insert_extents(
5916 : struct xfs_inode *ip,
5917 : xfs_fileoff_t off,
5918 : xfs_fileoff_t shift)
5919 : {
5920 379186 : struct xfs_bmbt_irec got;
5921 379186 : int is_empty;
5922 379186 : int error = 0;
5923 :
5924 379186 : ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5925 :
5926 758372 : if (xfs_is_shutdown(ip->i_mount))
5927 : return -EIO;
5928 :
5929 379184 : xfs_ilock(ip, XFS_ILOCK_EXCL);
5930 379184 : error = xfs_bmap_last_extent(NULL, ip, XFS_DATA_FORK, &got, &is_empty);
5931 379184 : if (!error && !is_empty && got.br_startoff >= off &&
5932 321839 : ((got.br_startoff + shift) & BMBT_STARTOFF_MASK) < got.br_startoff)
5933 0 : error = -EINVAL;
5934 379184 : xfs_iunlock(ip, XFS_ILOCK_EXCL);
5935 :
5936 379184 : return error;
5937 : }
5938 :
5939 : int
5940 1568460 : xfs_bmap_insert_extents(
5941 : struct xfs_trans *tp,
5942 : struct xfs_inode *ip,
5943 : xfs_fileoff_t *next_fsb,
5944 : xfs_fileoff_t offset_shift_fsb,
5945 : bool *done,
5946 : xfs_fileoff_t stop_fsb)
5947 : {
5948 1568460 : int whichfork = XFS_DATA_FORK;
5949 1568460 : struct xfs_mount *mp = ip->i_mount;
5950 1568460 : struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
5951 1568461 : struct xfs_btree_cur *cur = NULL;
5952 1568461 : struct xfs_bmbt_irec got, next;
5953 1568461 : struct xfs_iext_cursor icur;
5954 1568461 : xfs_fileoff_t new_startoff;
5955 1568461 : int error = 0;
5956 1568461 : int logflags = 0;
5957 :
5958 3136922 : if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
5959 1568461 : XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
5960 0 : xfs_bmap_mark_sick(ip, whichfork);
5961 0 : return -EFSCORRUPTED;
5962 : }
5963 :
5964 3136922 : if (xfs_is_shutdown(mp))
5965 : return -EIO;
5966 :
5967 1568461 : ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL));
5968 :
5969 1568461 : error = xfs_iread_extents(tp, ip, whichfork);
5970 1568460 : if (error)
5971 : return error;
5972 :
5973 1568460 : if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
5974 656267 : cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5975 656267 : cur->bc_ino.flags = 0;
5976 : }
5977 :
5978 1568460 : if (*next_fsb == NULLFSBLOCK) {
5979 378033 : xfs_iext_last(ifp, &icur);
5980 378034 : if (!xfs_iext_get_extent(ifp, &icur, &got) ||
5981 369874 : stop_fsb > got.br_startoff) {
5982 39495 : *done = true;
5983 39495 : goto del_cursor;
5984 : }
5985 : } else {
5986 1190427 : if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) {
5987 0 : *done = true;
5988 0 : goto del_cursor;
5989 : }
5990 : }
5991 1528966 : if (XFS_IS_CORRUPT(mp, isnullstartblock(got.br_startblock))) {
5992 0 : xfs_bmap_mark_sick(ip, whichfork);
5993 0 : error = -EFSCORRUPTED;
5994 0 : goto del_cursor;
5995 : }
5996 :
5997 1528966 : if (XFS_IS_CORRUPT(mp, stop_fsb > got.br_startoff)) {
5998 0 : xfs_bmap_mark_sick(ip, whichfork);
5999 0 : error = -EFSCORRUPTED;
6000 0 : goto del_cursor;
6001 : }
6002 :
6003 1528966 : new_startoff = got.br_startoff + offset_shift_fsb;
6004 1528966 : if (xfs_iext_peek_next_extent(ifp, &icur, &next)) {
6005 1190427 : if (new_startoff + got.br_blockcount > next.br_startoff) {
6006 0 : error = -EINVAL;
6007 0 : goto del_cursor;
6008 : }
6009 :
6010 : /*
6011 : * Unlike a left shift (which involves a hole punch), a right
6012 : * shift does not modify extent neighbors in any way. We should
6013 : * never find mergeable extents in this scenario. Check anyways
6014 : * and warn if we encounter two extents that could be one.
6015 : */
6016 1190427 : if (xfs_bmse_can_merge(&got, &next, offset_shift_fsb))
6017 0 : WARN_ON_ONCE(1);
6018 : }
6019 :
6020 1528966 : error = xfs_bmap_shift_update_extent(tp, ip, whichfork, &icur, &got,
6021 : cur, &logflags, new_startoff);
6022 1528966 : if (error)
6023 0 : goto del_cursor;
6024 :
6025 1528966 : if (!xfs_iext_prev_extent(ifp, &icur, &got) ||
6026 1465829 : stop_fsb >= got.br_startoff + got.br_blockcount) {
6027 338539 : *done = true;
6028 338539 : goto del_cursor;
6029 : }
6030 :
6031 1190427 : *next_fsb = got.br_startoff;
6032 1568461 : del_cursor:
6033 1568461 : if (cur)
6034 656268 : xfs_btree_del_cursor(cur, error);
6035 1568461 : if (logflags)
6036 1528965 : xfs_trans_log_inode(tp, ip, logflags);
6037 : return error;
6038 : }
6039 :
6040 : /*
6041 : * Splits an extent into two extents at split_fsb block such that it is the
6042 : * first block of the current_ext. @ext is a target extent to be split.
6043 : * @split_fsb is a block where the extents is split. If split_fsb lies in a
6044 : * hole or the first block of extents, just return 0.
6045 : */
6046 : int
6047 378034 : xfs_bmap_split_extent(
6048 : struct xfs_trans *tp,
6049 : struct xfs_inode *ip,
6050 : xfs_fileoff_t split_fsb)
6051 : {
6052 378034 : int whichfork = XFS_DATA_FORK;
6053 378034 : struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
6054 378034 : struct xfs_btree_cur *cur = NULL;
6055 378034 : struct xfs_bmbt_irec got;
6056 378034 : struct xfs_bmbt_irec new; /* split extent */
6057 378034 : struct xfs_mount *mp = ip->i_mount;
6058 378034 : xfs_fsblock_t gotblkcnt; /* new block count for got */
6059 378034 : struct xfs_iext_cursor icur;
6060 378034 : int error = 0;
6061 378034 : int logflags = 0;
6062 378034 : int i = 0;
6063 :
6064 756067 : if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
6065 378034 : XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
6066 0 : xfs_bmap_mark_sick(ip, whichfork);
6067 0 : return -EFSCORRUPTED;
6068 : }
6069 :
6070 756066 : if (xfs_is_shutdown(mp))
6071 : return -EIO;
6072 :
6073 : /* Read in all the extents */
6074 378033 : error = xfs_iread_extents(tp, ip, whichfork);
6075 378033 : if (error)
6076 : return error;
6077 :
6078 : /*
6079 : * If there are not extents, or split_fsb lies in a hole we are done.
6080 : */
6081 378033 : if (!xfs_iext_lookup_extent(ip, ifp, split_fsb, &icur, &got) ||
6082 338538 : got.br_startoff >= split_fsb)
6083 : return 0;
6084 :
6085 80670 : gotblkcnt = split_fsb - got.br_startoff;
6086 80670 : new.br_startoff = split_fsb;
6087 80670 : new.br_startblock = got.br_startblock + gotblkcnt;
6088 80670 : new.br_blockcount = got.br_blockcount - gotblkcnt;
6089 80670 : new.br_state = got.br_state;
6090 :
6091 80670 : if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
6092 9594 : cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
6093 9594 : cur->bc_ino.flags = 0;
6094 9594 : error = xfs_bmbt_lookup_eq(cur, &got, &i);
6095 9594 : if (error)
6096 0 : goto del_cursor;
6097 9594 : if (XFS_IS_CORRUPT(mp, i != 1)) {
6098 0 : xfs_btree_mark_sick(cur);
6099 0 : error = -EFSCORRUPTED;
6100 0 : goto del_cursor;
6101 : }
6102 : }
6103 :
6104 80670 : got.br_blockcount = gotblkcnt;
6105 80670 : xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), &icur,
6106 : &got);
6107 :
6108 80670 : logflags = XFS_ILOG_CORE;
6109 80670 : if (cur) {
6110 9594 : error = xfs_bmbt_update(cur, &got);
6111 9594 : if (error)
6112 0 : goto del_cursor;
6113 : } else
6114 : logflags |= XFS_ILOG_DEXT;
6115 :
6116 : /* Add new extent */
6117 80670 : xfs_iext_next(ifp, &icur);
6118 80670 : xfs_iext_insert(ip, &icur, &new, 0);
6119 80670 : ifp->if_nextents++;
6120 :
6121 80670 : if (cur) {
6122 9594 : error = xfs_bmbt_lookup_eq(cur, &new, &i);
6123 9594 : if (error)
6124 0 : goto del_cursor;
6125 9594 : if (XFS_IS_CORRUPT(mp, i != 0)) {
6126 0 : xfs_btree_mark_sick(cur);
6127 0 : error = -EFSCORRUPTED;
6128 0 : goto del_cursor;
6129 : }
6130 9594 : error = xfs_btree_insert(cur, &i);
6131 9594 : if (error)
6132 0 : goto del_cursor;
6133 9594 : if (XFS_IS_CORRUPT(mp, i != 1)) {
6134 0 : xfs_btree_mark_sick(cur);
6135 0 : error = -EFSCORRUPTED;
6136 0 : goto del_cursor;
6137 : }
6138 : }
6139 :
6140 : /*
6141 : * Convert to a btree if necessary.
6142 : */
6143 80670 : if (xfs_bmap_needs_btree(ip, whichfork)) {
6144 3028 : int tmp_logflags; /* partial log flag return val */
6145 :
6146 3028 : ASSERT(cur == NULL);
6147 3028 : error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
6148 : &tmp_logflags, whichfork);
6149 3028 : logflags |= tmp_logflags;
6150 : }
6151 :
6152 77642 : del_cursor:
6153 80670 : if (cur) {
6154 12622 : cur->bc_ino.allocated = 0;
6155 12622 : xfs_btree_del_cursor(cur, error);
6156 : }
6157 :
6158 80670 : if (logflags)
6159 80670 : xfs_trans_log_inode(tp, ip, logflags);
6160 : return error;
6161 : }
6162 :
6163 : /* Record a bmap intent. */
6164 : static void
6165 60004635 : __xfs_bmap_add(
6166 : struct xfs_trans *tp,
6167 : enum xfs_bmap_intent_type type,
6168 : struct xfs_inode *ip,
6169 : int whichfork,
6170 : struct xfs_bmbt_irec *bmap)
6171 : {
6172 60004635 : struct xfs_bmap_intent *bi;
6173 :
6174 60004635 : if ((whichfork != XFS_DATA_FORK && whichfork != XFS_ATTR_FORK) ||
6175 60004635 : bmap->br_startblock == HOLESTARTBLOCK ||
6176 : bmap->br_startblock == DELAYSTARTBLOCK)
6177 : return;
6178 :
6179 59253869 : bi = kmem_cache_alloc(xfs_bmap_intent_cache, GFP_NOFS | __GFP_NOFAIL);
6180 59253894 : INIT_LIST_HEAD(&bi->bi_list);
6181 59253894 : bi->bi_type = type;
6182 59253894 : bi->bi_owner = ip;
6183 59253894 : bi->bi_whichfork = whichfork;
6184 59253894 : bi->bi_bmap = *bmap;
6185 :
6186 59253894 : trace_xfs_bmap_defer(bi);
6187 :
6188 59253898 : xfs_bmap_update_get_group(tp->t_mountp, bi);
6189 59253870 : xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_BMAP, &bi->bi_list);
6190 : }
6191 :
6192 : /* Map an extent into a file. */
6193 : void
6194 53672559 : xfs_bmap_map_extent(
6195 : struct xfs_trans *tp,
6196 : struct xfs_inode *ip,
6197 : int whichfork,
6198 : struct xfs_bmbt_irec *PREV)
6199 : {
6200 53672559 : __xfs_bmap_add(tp, XFS_BMAP_MAP, ip, whichfork, PREV);
6201 53672565 : }
6202 :
6203 : /* Unmap an extent out of a file. */
6204 : void
6205 6332095 : xfs_bmap_unmap_extent(
6206 : struct xfs_trans *tp,
6207 : struct xfs_inode *ip,
6208 : int whichfork,
6209 : struct xfs_bmbt_irec *PREV)
6210 : {
6211 6332095 : __xfs_bmap_add(tp, XFS_BMAP_UNMAP, ip, whichfork, PREV);
6212 6332099 : }
6213 :
6214 : /*
6215 : * Process one of the deferred bmap operations. We pass back the
6216 : * btree cursor to maintain our lock on the bmapbt between calls.
6217 : */
6218 : int
6219 59254016 : xfs_bmap_finish_one(
6220 : struct xfs_trans *tp,
6221 : struct xfs_bmap_intent *bi)
6222 : {
6223 59254016 : struct xfs_bmbt_irec *bmap = &bi->bi_bmap;
6224 59254016 : int error = 0;
6225 59254016 : int flags = 0;
6226 :
6227 59254016 : if (bi->bi_whichfork == XFS_ATTR_FORK)
6228 218466 : flags |= XFS_BMAPI_ATTRFORK;
6229 :
6230 59254016 : ASSERT(tp->t_highest_agno == NULLAGNUMBER);
6231 :
6232 59254016 : trace_xfs_bmap_deferred(bi);
6233 :
6234 59254015 : if (XFS_TEST_ERROR(false, tp->t_mountp, XFS_ERRTAG_BMAP_FINISH_ONE))
6235 : return -EIO;
6236 :
6237 59253999 : switch (bi->bi_type) {
6238 53297250 : case XFS_BMAP_MAP:
6239 53297250 : if (bi->bi_bmap.br_state == XFS_EXT_UNWRITTEN)
6240 264542 : flags |= XFS_BMAPI_PREALLOC;
6241 53297250 : error = xfs_bmapi_remap(tp, bi->bi_owner, bmap->br_startoff,
6242 : bmap->br_blockcount, bmap->br_startblock,
6243 : flags);
6244 53297250 : bmap->br_blockcount = 0;
6245 53297250 : break;
6246 5956749 : case XFS_BMAP_UNMAP:
6247 5956749 : error = __xfs_bunmapi(tp, bi->bi_owner, bmap->br_startoff,
6248 : &bmap->br_blockcount, flags | XFS_BMAPI_REMAP,
6249 : 1);
6250 5956749 : break;
6251 0 : default:
6252 0 : ASSERT(0);
6253 0 : xfs_bmap_mark_sick(bi->bi_owner, bi->bi_whichfork);
6254 0 : error = -EFSCORRUPTED;
6255 : }
6256 :
6257 : return error;
6258 : }
6259 :
6260 : /* Check that an extent does not have invalid flags or bad ranges. */
6261 : xfs_failaddr_t
6262 2460853677 : xfs_bmap_validate_extent_raw(
6263 : struct xfs_mount *mp,
6264 : bool rtfile,
6265 : int whichfork,
6266 : struct xfs_bmbt_irec *irec)
6267 : {
6268 2460853677 : if (!xfs_verify_fileext(mp, irec->br_startoff, irec->br_blockcount))
6269 0 : return __this_address;
6270 :
6271 2460884022 : if (rtfile && whichfork == XFS_DATA_FORK) {
6272 144989847 : if (!xfs_verify_rtext(mp, irec->br_startblock,
6273 : irec->br_blockcount))
6274 0 : return __this_address;
6275 : } else {
6276 2315894175 : if (!xfs_verify_fsbext(mp, irec->br_startblock,
6277 : irec->br_blockcount))
6278 0 : return __this_address;
6279 : }
6280 2460930736 : if (irec->br_state != XFS_EXT_NORM && whichfork != XFS_DATA_FORK)
6281 0 : return __this_address;
6282 : return NULL;
6283 : }
6284 :
6285 : int __init
6286 12 : xfs_bmap_intent_init_cache(void)
6287 : {
6288 12 : xfs_bmap_intent_cache = kmem_cache_create("xfs_bmap_intent",
6289 : sizeof(struct xfs_bmap_intent),
6290 : 0, 0, NULL);
6291 :
6292 12 : return xfs_bmap_intent_cache != NULL ? 0 : -ENOMEM;
6293 : }
6294 :
6295 : void
6296 12 : xfs_bmap_intent_destroy_cache(void)
6297 : {
6298 12 : kmem_cache_destroy(xfs_bmap_intent_cache);
6299 12 : xfs_bmap_intent_cache = NULL;
6300 12 : }
6301 :
6302 : /* Check that an inode's extent does not have invalid flags or bad ranges. */
6303 : xfs_failaddr_t
6304 2460832587 : xfs_bmap_validate_extent(
6305 : struct xfs_inode *ip,
6306 : int whichfork,
6307 : struct xfs_bmbt_irec *irec)
6308 : {
6309 2460832587 : return xfs_bmap_validate_extent_raw(ip->i_mount,
6310 2460832587 : XFS_IS_REALTIME_INODE(ip), whichfork, irec);
6311 : }
6312 :
6313 : /*
6314 : * Used in xfs_itruncate_extents(). This is the maximum number of extents
6315 : * freed from a file in a single transaction.
6316 : */
6317 : #define XFS_ITRUNC_MAX_EXTENTS 2
6318 :
6319 : /*
6320 : * Unmap every extent in part of an inode's fork. We don't do any higher level
6321 : * invalidation work at all.
6322 : */
6323 : int
6324 11946806 : xfs_bunmapi_range(
6325 : struct xfs_trans **tpp,
6326 : struct xfs_inode *ip,
6327 : uint32_t flags,
6328 : xfs_fileoff_t startoff,
6329 : xfs_fileoff_t endoff)
6330 : {
6331 11946806 : xfs_filblks_t unmap_len = endoff - startoff + 1;
6332 11946806 : int error = 0;
6333 :
6334 11946806 : ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
6335 :
6336 47595452 : while (unmap_len > 0) {
6337 35653158 : ASSERT((*tpp)->t_highest_agno == NULLAGNUMBER);
6338 35653158 : error = __xfs_bunmapi(*tpp, ip, startoff, &unmap_len, flags,
6339 : XFS_ITRUNC_MAX_EXTENTS);
6340 35652492 : if (error)
6341 523 : goto out;
6342 :
6343 : /* free the just unmapped extents */
6344 35651969 : error = xfs_defer_finish(tpp);
6345 35653173 : if (error)
6346 1284 : goto out;
6347 : }
6348 11942294 : out:
6349 11944101 : return error;
6350 : }
6351 :
6352 : struct xfs_bmap_query_range {
6353 : xfs_bmap_query_range_fn fn;
6354 : void *priv;
6355 : };
6356 :
6357 : /* Format btree record and pass to our callback. */
6358 : STATIC int
6359 687274 : xfs_bmap_query_range_helper(
6360 : struct xfs_btree_cur *cur,
6361 : const union xfs_btree_rec *rec,
6362 : void *priv)
6363 : {
6364 687274 : struct xfs_bmap_query_range *query = priv;
6365 687274 : struct xfs_bmbt_irec irec;
6366 687274 : xfs_failaddr_t fa;
6367 :
6368 687274 : xfs_bmbt_disk_get_all(&rec->bmbt, &irec);
6369 687274 : fa = xfs_bmap_validate_extent(cur->bc_ino.ip, cur->bc_ino.whichfork,
6370 : &irec);
6371 687274 : if (fa) {
6372 0 : xfs_btree_mark_sick(cur);
6373 0 : return xfs_bmap_complain_bad_rec(cur->bc_ino.ip,
6374 0 : cur->bc_ino.whichfork, fa, &irec);
6375 : }
6376 :
6377 687274 : return query->fn(cur, &irec, query->priv);
6378 : }
6379 :
6380 : /* Find all bmaps. */
6381 : int
6382 37458 : xfs_bmap_query_all(
6383 : struct xfs_btree_cur *cur,
6384 : xfs_bmap_query_range_fn fn,
6385 : void *priv)
6386 : {
6387 37458 : struct xfs_bmap_query_range query = {
6388 : .priv = priv,
6389 : .fn = fn,
6390 : };
6391 :
6392 37458 : return xfs_btree_query_all(cur, xfs_bmap_query_range_helper, &query);
6393 : }
|
