aboutsummaryrefslogtreecommitdiffstats
path: root/fs/xfs/libxfs/xfs_bmap.c
diff options
context:
space:
mode:
Diffstat (limited to 'fs/xfs/libxfs/xfs_bmap.c')
-rw-r--r--fs/xfs/libxfs/xfs_bmap.c5606
1 files changed, 5606 insertions, 0 deletions
diff --git a/fs/xfs/libxfs/xfs_bmap.c b/fs/xfs/libxfs/xfs_bmap.c
new file mode 100644
index 000000000000..94ac88306fa6
--- /dev/null
+++ b/fs/xfs/libxfs/xfs_bmap.c
@@ -0,0 +1,5606 @@
1/*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18#include "xfs.h"
19#include "xfs_fs.h"
20#include "xfs_shared.h"
21#include "xfs_format.h"
22#include "xfs_log_format.h"
23#include "xfs_trans_resv.h"
24#include "xfs_bit.h"
25#include "xfs_inum.h"
26#include "xfs_sb.h"
27#include "xfs_ag.h"
28#include "xfs_mount.h"
29#include "xfs_da_format.h"
30#include "xfs_da_btree.h"
31#include "xfs_dir2.h"
32#include "xfs_inode.h"
33#include "xfs_btree.h"
34#include "xfs_trans.h"
35#include "xfs_inode_item.h"
36#include "xfs_extfree_item.h"
37#include "xfs_alloc.h"
38#include "xfs_bmap.h"
39#include "xfs_bmap_util.h"
40#include "xfs_bmap_btree.h"
41#include "xfs_rtalloc.h"
42#include "xfs_error.h"
43#include "xfs_quota.h"
44#include "xfs_trans_space.h"
45#include "xfs_buf_item.h"
46#include "xfs_trace.h"
47#include "xfs_symlink.h"
48#include "xfs_attr_leaf.h"
49#include "xfs_dinode.h"
50#include "xfs_filestream.h"
51
52
53kmem_zone_t *xfs_bmap_free_item_zone;
54
55/*
56 * Miscellaneous helper functions
57 */
58
59/*
60 * Compute and fill in the value of the maximum depth of a bmap btree
61 * in this filesystem. Done once, during mount.
62 */
63void
64xfs_bmap_compute_maxlevels(
65 xfs_mount_t *mp, /* file system mount structure */
66 int whichfork) /* data or attr fork */
67{
68 int level; /* btree level */
69 uint maxblocks; /* max blocks at this level */
70 uint maxleafents; /* max leaf entries possible */
71 int maxrootrecs; /* max records in root block */
72 int minleafrecs; /* min records in leaf block */
73 int minnoderecs; /* min records in node block */
74 int sz; /* root block size */
75
76 /*
77 * The maximum number of extents in a file, hence the maximum
78 * number of leaf entries, is controlled by the type of di_nextents
79 * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
80 * (a signed 16-bit number, xfs_aextnum_t).
81 *
82 * Note that we can no longer assume that if we are in ATTR1 that
83 * the fork offset of all the inodes will be
84 * (xfs_default_attroffset(ip) >> 3) because we could have mounted
85 * with ATTR2 and then mounted back with ATTR1, keeping the
86 * di_forkoff's fixed but probably at various positions. Therefore,
87 * for both ATTR1 and ATTR2 we have to assume the worst case scenario
88 * of a minimum size available.
89 */
90 if (whichfork == XFS_DATA_FORK) {
91 maxleafents = MAXEXTNUM;
92 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
93 } else {
94 maxleafents = MAXAEXTNUM;
95 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
96 }
97 maxrootrecs = xfs_bmdr_maxrecs(sz, 0);
98 minleafrecs = mp->m_bmap_dmnr[0];
99 minnoderecs = mp->m_bmap_dmnr[1];
100 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
101 for (level = 1; maxblocks > 1; level++) {
102 if (maxblocks <= maxrootrecs)
103 maxblocks = 1;
104 else
105 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
106 }
107 mp->m_bm_maxlevels[whichfork] = level;
108}
109
110STATIC int /* error */
111xfs_bmbt_lookup_eq(
112 struct xfs_btree_cur *cur,
113 xfs_fileoff_t off,
114 xfs_fsblock_t bno,
115 xfs_filblks_t len,
116 int *stat) /* success/failure */
117{
118 cur->bc_rec.b.br_startoff = off;
119 cur->bc_rec.b.br_startblock = bno;
120 cur->bc_rec.b.br_blockcount = len;
121 return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
122}
123
124STATIC int /* error */
125xfs_bmbt_lookup_ge(
126 struct xfs_btree_cur *cur,
127 xfs_fileoff_t off,
128 xfs_fsblock_t bno,
129 xfs_filblks_t len,
130 int *stat) /* success/failure */
131{
132 cur->bc_rec.b.br_startoff = off;
133 cur->bc_rec.b.br_startblock = bno;
134 cur->bc_rec.b.br_blockcount = len;
135 return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
136}
137
138/*
139 * Check if the inode needs to be converted to btree format.
140 */
141static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
142{
143 return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
144 XFS_IFORK_NEXTENTS(ip, whichfork) >
145 XFS_IFORK_MAXEXT(ip, whichfork);
146}
147
148/*
149 * Check if the inode should be converted to extent format.
150 */
151static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
152{
153 return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
154 XFS_IFORK_NEXTENTS(ip, whichfork) <=
155 XFS_IFORK_MAXEXT(ip, whichfork);
156}
157
158/*
159 * Update the record referred to by cur to the value given
160 * by [off, bno, len, state].
161 * This either works (return 0) or gets an EFSCORRUPTED error.
162 */
163STATIC int
164xfs_bmbt_update(
165 struct xfs_btree_cur *cur,
166 xfs_fileoff_t off,
167 xfs_fsblock_t bno,
168 xfs_filblks_t len,
169 xfs_exntst_t state)
170{
171 union xfs_btree_rec rec;
172
173 xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
174 return xfs_btree_update(cur, &rec);
175}
176
177/*
178 * Compute the worst-case number of indirect blocks that will be used
179 * for ip's delayed extent of length "len".
180 */
181STATIC xfs_filblks_t
182xfs_bmap_worst_indlen(
183 xfs_inode_t *ip, /* incore inode pointer */
184 xfs_filblks_t len) /* delayed extent length */
185{
186 int level; /* btree level number */
187 int maxrecs; /* maximum record count at this level */
188 xfs_mount_t *mp; /* mount structure */
189 xfs_filblks_t rval; /* return value */
190
191 mp = ip->i_mount;
192 maxrecs = mp->m_bmap_dmxr[0];
193 for (level = 0, rval = 0;
194 level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
195 level++) {
196 len += maxrecs - 1;
197 do_div(len, maxrecs);
198 rval += len;
199 if (len == 1)
200 return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
201 level - 1;
202 if (level == 0)
203 maxrecs = mp->m_bmap_dmxr[1];
204 }
205 return rval;
206}
207
208/*
209 * Calculate the default attribute fork offset for newly created inodes.
210 */
211uint
212xfs_default_attroffset(
213 struct xfs_inode *ip)
214{
215 struct xfs_mount *mp = ip->i_mount;
216 uint offset;
217
218 if (mp->m_sb.sb_inodesize == 256) {
219 offset = XFS_LITINO(mp, ip->i_d.di_version) -
220 XFS_BMDR_SPACE_CALC(MINABTPTRS);
221 } else {
222 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
223 }
224
225 ASSERT(offset < XFS_LITINO(mp, ip->i_d.di_version));
226 return offset;
227}
228
229/*
230 * Helper routine to reset inode di_forkoff field when switching
231 * attribute fork from local to extent format - we reset it where
232 * possible to make space available for inline data fork extents.
233 */
234STATIC void
235xfs_bmap_forkoff_reset(
236 xfs_inode_t *ip,
237 int whichfork)
238{
239 if (whichfork == XFS_ATTR_FORK &&
240 ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
241 ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
242 ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
243 uint dfl_forkoff = xfs_default_attroffset(ip) >> 3;
244
245 if (dfl_forkoff > ip->i_d.di_forkoff)
246 ip->i_d.di_forkoff = dfl_forkoff;
247 }
248}
249
250/*
251 * Debug/sanity checking code
252 */
253
254STATIC int
255xfs_bmap_sanity_check(
256 struct xfs_mount *mp,
257 struct xfs_buf *bp,
258 int level)
259{
260 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
261
262 if (block->bb_magic != cpu_to_be32(XFS_BMAP_CRC_MAGIC) &&
263 block->bb_magic != cpu_to_be32(XFS_BMAP_MAGIC))
264 return 0;
265
266 if (be16_to_cpu(block->bb_level) != level ||
267 be16_to_cpu(block->bb_numrecs) == 0 ||
268 be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
269 return 0;
270
271 return 1;
272}
273
274#ifdef DEBUG
275STATIC struct xfs_buf *
276xfs_bmap_get_bp(
277 struct xfs_btree_cur *cur,
278 xfs_fsblock_t bno)
279{
280 struct xfs_log_item_desc *lidp;
281 int i;
282
283 if (!cur)
284 return NULL;
285
286 for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
287 if (!cur->bc_bufs[i])
288 break;
289 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
290 return cur->bc_bufs[i];
291 }
292
293 /* Chase down all the log items to see if the bp is there */
294 list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
295 struct xfs_buf_log_item *bip;
296 bip = (struct xfs_buf_log_item *)lidp->lid_item;
297 if (bip->bli_item.li_type == XFS_LI_BUF &&
298 XFS_BUF_ADDR(bip->bli_buf) == bno)
299 return bip->bli_buf;
300 }
301
302 return NULL;
303}
304
305STATIC void
306xfs_check_block(
307 struct xfs_btree_block *block,
308 xfs_mount_t *mp,
309 int root,
310 short sz)
311{
312 int i, j, dmxr;
313 __be64 *pp, *thispa; /* pointer to block address */
314 xfs_bmbt_key_t *prevp, *keyp;
315
316 ASSERT(be16_to_cpu(block->bb_level) > 0);
317
318 prevp = NULL;
319 for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
320 dmxr = mp->m_bmap_dmxr[0];
321 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
322
323 if (prevp) {
324 ASSERT(be64_to_cpu(prevp->br_startoff) <
325 be64_to_cpu(keyp->br_startoff));
326 }
327 prevp = keyp;
328
329 /*
330 * Compare the block numbers to see if there are dups.
331 */
332 if (root)
333 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
334 else
335 pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
336
337 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
338 if (root)
339 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
340 else
341 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
342 if (*thispa == *pp) {
343 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
344 __func__, j, i,
345 (unsigned long long)be64_to_cpu(*thispa));
346 panic("%s: ptrs are equal in node\n",
347 __func__);
348 }
349 }
350 }
351}
352
353/*
354 * Check that the extents for the inode ip are in the right order in all
355 * btree leaves.
356 */
357
358STATIC void
359xfs_bmap_check_leaf_extents(
360 xfs_btree_cur_t *cur, /* btree cursor or null */
361 xfs_inode_t *ip, /* incore inode pointer */
362 int whichfork) /* data or attr fork */
363{
364 struct xfs_btree_block *block; /* current btree block */
365 xfs_fsblock_t bno; /* block # of "block" */
366 xfs_buf_t *bp; /* buffer for "block" */
367 int error; /* error return value */
368 xfs_extnum_t i=0, j; /* index into the extents list */
369 xfs_ifork_t *ifp; /* fork structure */
370 int level; /* btree level, for checking */
371 xfs_mount_t *mp; /* file system mount structure */
372 __be64 *pp; /* pointer to block address */
373 xfs_bmbt_rec_t *ep; /* pointer to current extent */
374 xfs_bmbt_rec_t last = {0, 0}; /* last extent in prev block */
375 xfs_bmbt_rec_t *nextp; /* pointer to next extent */
376 int bp_release = 0;
377
378 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
379 return;
380 }
381
382 bno = NULLFSBLOCK;
383 mp = ip->i_mount;
384 ifp = XFS_IFORK_PTR(ip, whichfork);
385 block = ifp->if_broot;
386 /*
387 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
388 */
389 level = be16_to_cpu(block->bb_level);
390 ASSERT(level > 0);
391 xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
392 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
393 bno = be64_to_cpu(*pp);
394
395 ASSERT(bno != NULLDFSBNO);
396 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
397 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
398
399 /*
400 * Go down the tree until leaf level is reached, following the first
401 * pointer (leftmost) at each level.
402 */
403 while (level-- > 0) {
404 /* See if buf is in cur first */
405 bp_release = 0;
406 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
407 if (!bp) {
408 bp_release = 1;
409 error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
410 XFS_BMAP_BTREE_REF,
411 &xfs_bmbt_buf_ops);
412 if (error)
413 goto error_norelse;
414 }
415 block = XFS_BUF_TO_BLOCK(bp);
416 XFS_WANT_CORRUPTED_GOTO(
417 xfs_bmap_sanity_check(mp, bp, level),
418 error0);
419 if (level == 0)
420 break;
421
422 /*
423 * Check this block for basic sanity (increasing keys and
424 * no duplicate blocks).
425 */
426
427 xfs_check_block(block, mp, 0, 0);
428 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
429 bno = be64_to_cpu(*pp);
430 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
431 if (bp_release) {
432 bp_release = 0;
433 xfs_trans_brelse(NULL, bp);
434 }
435 }
436
437 /*
438 * Here with bp and block set to the leftmost leaf node in the tree.
439 */
440 i = 0;
441
442 /*
443 * Loop over all leaf nodes checking that all extents are in the right order.
444 */
445 for (;;) {
446 xfs_fsblock_t nextbno;
447 xfs_extnum_t num_recs;
448
449
450 num_recs = xfs_btree_get_numrecs(block);
451
452 /*
453 * Read-ahead the next leaf block, if any.
454 */
455
456 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
457
458 /*
459 * Check all the extents to make sure they are OK.
460 * If we had a previous block, the last entry should
461 * conform with the first entry in this one.
462 */
463
464 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
465 if (i) {
466 ASSERT(xfs_bmbt_disk_get_startoff(&last) +
467 xfs_bmbt_disk_get_blockcount(&last) <=
468 xfs_bmbt_disk_get_startoff(ep));
469 }
470 for (j = 1; j < num_recs; j++) {
471 nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
472 ASSERT(xfs_bmbt_disk_get_startoff(ep) +
473 xfs_bmbt_disk_get_blockcount(ep) <=
474 xfs_bmbt_disk_get_startoff(nextp));
475 ep = nextp;
476 }
477
478 last = *ep;
479 i += num_recs;
480 if (bp_release) {
481 bp_release = 0;
482 xfs_trans_brelse(NULL, bp);
483 }
484 bno = nextbno;
485 /*
486 * If we've reached the end, stop.
487 */
488 if (bno == NULLFSBLOCK)
489 break;
490
491 bp_release = 0;
492 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
493 if (!bp) {
494 bp_release = 1;
495 error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
496 XFS_BMAP_BTREE_REF,
497 &xfs_bmbt_buf_ops);
498 if (error)
499 goto error_norelse;
500 }
501 block = XFS_BUF_TO_BLOCK(bp);
502 }
503 if (bp_release) {
504 bp_release = 0;
505 xfs_trans_brelse(NULL, bp);
506 }
507 return;
508
509error0:
510 xfs_warn(mp, "%s: at error0", __func__);
511 if (bp_release)
512 xfs_trans_brelse(NULL, bp);
513error_norelse:
514 xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
515 __func__, i);
516 panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
517 return;
518}
519
520/*
521 * Add bmap trace insert entries for all the contents of the extent records.
522 */
523void
524xfs_bmap_trace_exlist(
525 xfs_inode_t *ip, /* incore inode pointer */
526 xfs_extnum_t cnt, /* count of entries in the list */
527 int whichfork, /* data or attr fork */
528 unsigned long caller_ip)
529{
530 xfs_extnum_t idx; /* extent record index */
531 xfs_ifork_t *ifp; /* inode fork pointer */
532 int state = 0;
533
534 if (whichfork == XFS_ATTR_FORK)
535 state |= BMAP_ATTRFORK;
536
537 ifp = XFS_IFORK_PTR(ip, whichfork);
538 ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
539 for (idx = 0; idx < cnt; idx++)
540 trace_xfs_extlist(ip, idx, whichfork, caller_ip);
541}
542
543/*
544 * Validate that the bmbt_irecs being returned from bmapi are valid
545 * given the caller's original parameters. Specifically check the
546 * ranges of the returned irecs to ensure that they only extend beyond
547 * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
548 */
549STATIC void
550xfs_bmap_validate_ret(
551 xfs_fileoff_t bno,
552 xfs_filblks_t len,
553 int flags,
554 xfs_bmbt_irec_t *mval,
555 int nmap,
556 int ret_nmap)
557{
558 int i; /* index to map values */
559
560 ASSERT(ret_nmap <= nmap);
561
562 for (i = 0; i < ret_nmap; i++) {
563 ASSERT(mval[i].br_blockcount > 0);
564 if (!(flags & XFS_BMAPI_ENTIRE)) {
565 ASSERT(mval[i].br_startoff >= bno);
566 ASSERT(mval[i].br_blockcount <= len);
567 ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
568 bno + len);
569 } else {
570 ASSERT(mval[i].br_startoff < bno + len);
571 ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
572 bno);
573 }
574 ASSERT(i == 0 ||
575 mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
576 mval[i].br_startoff);
577 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
578 mval[i].br_startblock != HOLESTARTBLOCK);
579 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
580 mval[i].br_state == XFS_EXT_UNWRITTEN);
581 }
582}
583
584#else
585#define xfs_bmap_check_leaf_extents(cur, ip, whichfork) do { } while (0)
586#define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
587#endif /* DEBUG */
588
589/*
590 * bmap free list manipulation functions
591 */
592
593/*
594 * Add the extent to the list of extents to be free at transaction end.
595 * The list is maintained sorted (by block number).
596 */
597void
598xfs_bmap_add_free(
599 xfs_fsblock_t bno, /* fs block number of extent */
600 xfs_filblks_t len, /* length of extent */
601 xfs_bmap_free_t *flist, /* list of extents */
602 xfs_mount_t *mp) /* mount point structure */
603{
604 xfs_bmap_free_item_t *cur; /* current (next) element */
605 xfs_bmap_free_item_t *new; /* new element */
606 xfs_bmap_free_item_t *prev; /* previous element */
607#ifdef DEBUG
608 xfs_agnumber_t agno;
609 xfs_agblock_t agbno;
610
611 ASSERT(bno != NULLFSBLOCK);
612 ASSERT(len > 0);
613 ASSERT(len <= MAXEXTLEN);
614 ASSERT(!isnullstartblock(bno));
615 agno = XFS_FSB_TO_AGNO(mp, bno);
616 agbno = XFS_FSB_TO_AGBNO(mp, bno);
617 ASSERT(agno < mp->m_sb.sb_agcount);
618 ASSERT(agbno < mp->m_sb.sb_agblocks);
619 ASSERT(len < mp->m_sb.sb_agblocks);
620 ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
621#endif
622 ASSERT(xfs_bmap_free_item_zone != NULL);
623 new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
624 new->xbfi_startblock = bno;
625 new->xbfi_blockcount = (xfs_extlen_t)len;
626 for (prev = NULL, cur = flist->xbf_first;
627 cur != NULL;
628 prev = cur, cur = cur->xbfi_next) {
629 if (cur->xbfi_startblock >= bno)
630 break;
631 }
632 if (prev)
633 prev->xbfi_next = new;
634 else
635 flist->xbf_first = new;
636 new->xbfi_next = cur;
637 flist->xbf_count++;
638}
639
640/*
641 * Remove the entry "free" from the free item list. Prev points to the
642 * previous entry, unless "free" is the head of the list.
643 */
644void
645xfs_bmap_del_free(
646 xfs_bmap_free_t *flist, /* free item list header */
647 xfs_bmap_free_item_t *prev, /* previous item on list, if any */
648 xfs_bmap_free_item_t *free) /* list item to be freed */
649{
650 if (prev)
651 prev->xbfi_next = free->xbfi_next;
652 else
653 flist->xbf_first = free->xbfi_next;
654 flist->xbf_count--;
655 kmem_zone_free(xfs_bmap_free_item_zone, free);
656}
657
658/*
659 * Free up any items left in the list.
660 */
661void
662xfs_bmap_cancel(
663 xfs_bmap_free_t *flist) /* list of bmap_free_items */
664{
665 xfs_bmap_free_item_t *free; /* free list item */
666 xfs_bmap_free_item_t *next;
667
668 if (flist->xbf_count == 0)
669 return;
670 ASSERT(flist->xbf_first != NULL);
671 for (free = flist->xbf_first; free; free = next) {
672 next = free->xbfi_next;
673 xfs_bmap_del_free(flist, NULL, free);
674 }
675 ASSERT(flist->xbf_count == 0);
676}
677
678/*
679 * Inode fork format manipulation functions
680 */
681
682/*
683 * Transform a btree format file with only one leaf node, where the
684 * extents list will fit in the inode, into an extents format file.
685 * Since the file extents are already in-core, all we have to do is
686 * give up the space for the btree root and pitch the leaf block.
687 */
688STATIC int /* error */
689xfs_bmap_btree_to_extents(
690 xfs_trans_t *tp, /* transaction pointer */
691 xfs_inode_t *ip, /* incore inode pointer */
692 xfs_btree_cur_t *cur, /* btree cursor */
693 int *logflagsp, /* inode logging flags */
694 int whichfork) /* data or attr fork */
695{
696 /* REFERENCED */
697 struct xfs_btree_block *cblock;/* child btree block */
698 xfs_fsblock_t cbno; /* child block number */
699 xfs_buf_t *cbp; /* child block's buffer */
700 int error; /* error return value */
701 xfs_ifork_t *ifp; /* inode fork data */
702 xfs_mount_t *mp; /* mount point structure */
703 __be64 *pp; /* ptr to block address */
704 struct xfs_btree_block *rblock;/* root btree block */
705
706 mp = ip->i_mount;
707 ifp = XFS_IFORK_PTR(ip, whichfork);
708 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
709 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
710 rblock = ifp->if_broot;
711 ASSERT(be16_to_cpu(rblock->bb_level) == 1);
712 ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
713 ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
714 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
715 cbno = be64_to_cpu(*pp);
716 *logflagsp = 0;
717#ifdef DEBUG
718 if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
719 return error;
720#endif
721 error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF,
722 &xfs_bmbt_buf_ops);
723 if (error)
724 return error;
725 cblock = XFS_BUF_TO_BLOCK(cbp);
726 if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
727 return error;
728 xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
729 ip->i_d.di_nblocks--;
730 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
731 xfs_trans_binval(tp, cbp);
732 if (cur->bc_bufs[0] == cbp)
733 cur->bc_bufs[0] = NULL;
734 xfs_iroot_realloc(ip, -1, whichfork);
735 ASSERT(ifp->if_broot == NULL);
736 ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
737 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
738 *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
739 return 0;
740}
741
742/*
743 * Convert an extents-format file into a btree-format file.
744 * The new file will have a root block (in the inode) and a single child block.
745 */
746STATIC int /* error */
747xfs_bmap_extents_to_btree(
748 xfs_trans_t *tp, /* transaction pointer */
749 xfs_inode_t *ip, /* incore inode pointer */
750 xfs_fsblock_t *firstblock, /* first-block-allocated */
751 xfs_bmap_free_t *flist, /* blocks freed in xaction */
752 xfs_btree_cur_t **curp, /* cursor returned to caller */
753 int wasdel, /* converting a delayed alloc */
754 int *logflagsp, /* inode logging flags */
755 int whichfork) /* data or attr fork */
756{
757 struct xfs_btree_block *ablock; /* allocated (child) bt block */
758 xfs_buf_t *abp; /* buffer for ablock */
759 xfs_alloc_arg_t args; /* allocation arguments */
760 xfs_bmbt_rec_t *arp; /* child record pointer */
761 struct xfs_btree_block *block; /* btree root block */
762 xfs_btree_cur_t *cur; /* bmap btree cursor */
763 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
764 int error; /* error return value */
765 xfs_extnum_t i, cnt; /* extent record index */
766 xfs_ifork_t *ifp; /* inode fork pointer */
767 xfs_bmbt_key_t *kp; /* root block key pointer */
768 xfs_mount_t *mp; /* mount structure */
769 xfs_extnum_t nextents; /* number of file extents */
770 xfs_bmbt_ptr_t *pp; /* root block address pointer */
771
772 mp = ip->i_mount;
773 ifp = XFS_IFORK_PTR(ip, whichfork);
774 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
775
776 /*
777 * Make space in the inode incore.
778 */
779 xfs_iroot_realloc(ip, 1, whichfork);
780 ifp->if_flags |= XFS_IFBROOT;
781
782 /*
783 * Fill in the root.
784 */
785 block = ifp->if_broot;
786 if (xfs_sb_version_hascrc(&mp->m_sb))
787 xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
788 XFS_BMAP_CRC_MAGIC, 1, 1, ip->i_ino,
789 XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
790 else
791 xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
792 XFS_BMAP_MAGIC, 1, 1, ip->i_ino,
793 XFS_BTREE_LONG_PTRS);
794
795 /*
796 * Need a cursor. Can't allocate until bb_level is filled in.
797 */
798 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
799 cur->bc_private.b.firstblock = *firstblock;
800 cur->bc_private.b.flist = flist;
801 cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
802 /*
803 * Convert to a btree with two levels, one record in root.
804 */
805 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
806 memset(&args, 0, sizeof(args));
807 args.tp = tp;
808 args.mp = mp;
809 args.firstblock = *firstblock;
810 if (*firstblock == NULLFSBLOCK) {
811 args.type = XFS_ALLOCTYPE_START_BNO;
812 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
813 } else if (flist->xbf_low) {
814 args.type = XFS_ALLOCTYPE_START_BNO;
815 args.fsbno = *firstblock;
816 } else {
817 args.type = XFS_ALLOCTYPE_NEAR_BNO;
818 args.fsbno = *firstblock;
819 }
820 args.minlen = args.maxlen = args.prod = 1;
821 args.wasdel = wasdel;
822 *logflagsp = 0;
823 if ((error = xfs_alloc_vextent(&args))) {
824 xfs_iroot_realloc(ip, -1, whichfork);
825 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
826 return error;
827 }
828 /*
829 * Allocation can't fail, the space was reserved.
830 */
831 ASSERT(args.fsbno != NULLFSBLOCK);
832 ASSERT(*firstblock == NULLFSBLOCK ||
833 args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
834 (flist->xbf_low &&
835 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
836 *firstblock = cur->bc_private.b.firstblock = args.fsbno;
837 cur->bc_private.b.allocated++;
838 ip->i_d.di_nblocks++;
839 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
840 abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
841 /*
842 * Fill in the child block.
843 */
844 abp->b_ops = &xfs_bmbt_buf_ops;
845 ablock = XFS_BUF_TO_BLOCK(abp);
846 if (xfs_sb_version_hascrc(&mp->m_sb))
847 xfs_btree_init_block_int(mp, ablock, abp->b_bn,
848 XFS_BMAP_CRC_MAGIC, 0, 0, ip->i_ino,
849 XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
850 else
851 xfs_btree_init_block_int(mp, ablock, abp->b_bn,
852 XFS_BMAP_MAGIC, 0, 0, ip->i_ino,
853 XFS_BTREE_LONG_PTRS);
854
855 arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
856 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
857 for (cnt = i = 0; i < nextents; i++) {
858 ep = xfs_iext_get_ext(ifp, i);
859 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
860 arp->l0 = cpu_to_be64(ep->l0);
861 arp->l1 = cpu_to_be64(ep->l1);
862 arp++; cnt++;
863 }
864 }
865 ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
866 xfs_btree_set_numrecs(ablock, cnt);
867
868 /*
869 * Fill in the root key and pointer.
870 */
871 kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
872 arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
873 kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
874 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
875 be16_to_cpu(block->bb_level)));
876 *pp = cpu_to_be64(args.fsbno);
877
878 /*
879 * Do all this logging at the end so that
880 * the root is at the right level.
881 */
882 xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
883 xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
884 ASSERT(*curp == NULL);
885 *curp = cur;
886 *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
887 return 0;
888}
889
890/*
891 * Convert a local file to an extents file.
892 * This code is out of bounds for data forks of regular files,
893 * since the file data needs to get logged so things will stay consistent.
894 * (The bmap-level manipulations are ok, though).
895 */
896void
897xfs_bmap_local_to_extents_empty(
898 struct xfs_inode *ip,
899 int whichfork)
900{
901 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
902
903 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
904 ASSERT(ifp->if_bytes == 0);
905 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
906
907 xfs_bmap_forkoff_reset(ip, whichfork);
908 ifp->if_flags &= ~XFS_IFINLINE;
909 ifp->if_flags |= XFS_IFEXTENTS;
910 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
911}
912
913
914STATIC int /* error */
915xfs_bmap_local_to_extents(
916 xfs_trans_t *tp, /* transaction pointer */
917 xfs_inode_t *ip, /* incore inode pointer */
918 xfs_fsblock_t *firstblock, /* first block allocated in xaction */
919 xfs_extlen_t total, /* total blocks needed by transaction */
920 int *logflagsp, /* inode logging flags */
921 int whichfork,
922 void (*init_fn)(struct xfs_trans *tp,
923 struct xfs_buf *bp,
924 struct xfs_inode *ip,
925 struct xfs_ifork *ifp))
926{
927 int error = 0;
928 int flags; /* logging flags returned */
929 xfs_ifork_t *ifp; /* inode fork pointer */
930 xfs_alloc_arg_t args; /* allocation arguments */
931 xfs_buf_t *bp; /* buffer for extent block */
932 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
933
934 /*
935 * We don't want to deal with the case of keeping inode data inline yet.
936 * So sending the data fork of a regular inode is invalid.
937 */
938 ASSERT(!(S_ISREG(ip->i_d.di_mode) && whichfork == XFS_DATA_FORK));
939 ifp = XFS_IFORK_PTR(ip, whichfork);
940 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
941
942 if (!ifp->if_bytes) {
943 xfs_bmap_local_to_extents_empty(ip, whichfork);
944 flags = XFS_ILOG_CORE;
945 goto done;
946 }
947
948 flags = 0;
949 error = 0;
950 ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) ==
951 XFS_IFINLINE);
952 memset(&args, 0, sizeof(args));
953 args.tp = tp;
954 args.mp = ip->i_mount;
955 args.firstblock = *firstblock;
956 /*
957 * Allocate a block. We know we need only one, since the
958 * file currently fits in an inode.
959 */
960 if (*firstblock == NULLFSBLOCK) {
961 args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
962 args.type = XFS_ALLOCTYPE_START_BNO;
963 } else {
964 args.fsbno = *firstblock;
965 args.type = XFS_ALLOCTYPE_NEAR_BNO;
966 }
967 args.total = total;
968 args.minlen = args.maxlen = args.prod = 1;
969 error = xfs_alloc_vextent(&args);
970 if (error)
971 goto done;
972
973 /* Can't fail, the space was reserved. */
974 ASSERT(args.fsbno != NULLFSBLOCK);
975 ASSERT(args.len == 1);
976 *firstblock = args.fsbno;
977 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
978
979 /* initialise the block and copy the data */
980 init_fn(tp, bp, ip, ifp);
981
982 /* account for the change in fork size and log everything */
983 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
984 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
985 xfs_bmap_local_to_extents_empty(ip, whichfork);
986 flags |= XFS_ILOG_CORE;
987
988 xfs_iext_add(ifp, 0, 1);
989 ep = xfs_iext_get_ext(ifp, 0);
990 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
991 trace_xfs_bmap_post_update(ip, 0,
992 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
993 _THIS_IP_);
994 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
995 ip->i_d.di_nblocks = 1;
996 xfs_trans_mod_dquot_byino(tp, ip,
997 XFS_TRANS_DQ_BCOUNT, 1L);
998 flags |= xfs_ilog_fext(whichfork);
999
1000done:
1001 *logflagsp = flags;
1002 return error;
1003}
1004
1005/*
1006 * Called from xfs_bmap_add_attrfork to handle btree format files.
1007 */
1008STATIC int /* error */
1009xfs_bmap_add_attrfork_btree(
1010 xfs_trans_t *tp, /* transaction pointer */
1011 xfs_inode_t *ip, /* incore inode pointer */
1012 xfs_fsblock_t *firstblock, /* first block allocated */
1013 xfs_bmap_free_t *flist, /* blocks to free at commit */
1014 int *flags) /* inode logging flags */
1015{
1016 xfs_btree_cur_t *cur; /* btree cursor */
1017 int error; /* error return value */
1018 xfs_mount_t *mp; /* file system mount struct */
1019 int stat; /* newroot status */
1020
1021 mp = ip->i_mount;
1022 if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
1023 *flags |= XFS_ILOG_DBROOT;
1024 else {
1025 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
1026 cur->bc_private.b.flist = flist;
1027 cur->bc_private.b.firstblock = *firstblock;
1028 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
1029 goto error0;
1030 /* must be at least one entry */
1031 XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
1032 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
1033 goto error0;
1034 if (stat == 0) {
1035 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1036 return -ENOSPC;
1037 }
1038 *firstblock = cur->bc_private.b.firstblock;
1039 cur->bc_private.b.allocated = 0;
1040 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1041 }
1042 return 0;
1043error0:
1044 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1045 return error;
1046}
1047
1048/*
1049 * Called from xfs_bmap_add_attrfork to handle extents format files.
1050 */
1051STATIC int /* error */
1052xfs_bmap_add_attrfork_extents(
1053 xfs_trans_t *tp, /* transaction pointer */
1054 xfs_inode_t *ip, /* incore inode pointer */
1055 xfs_fsblock_t *firstblock, /* first block allocated */
1056 xfs_bmap_free_t *flist, /* blocks to free at commit */
1057 int *flags) /* inode logging flags */
1058{
1059 xfs_btree_cur_t *cur; /* bmap btree cursor */
1060 int error; /* error return value */
1061
1062 if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
1063 return 0;
1064 cur = NULL;
1065 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
1066 flags, XFS_DATA_FORK);
1067 if (cur) {
1068 cur->bc_private.b.allocated = 0;
1069 xfs_btree_del_cursor(cur,
1070 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
1071 }
1072 return error;
1073}
1074
1075/*
1076 * Called from xfs_bmap_add_attrfork to handle local format files. Each
1077 * different data fork content type needs a different callout to do the
1078 * conversion. Some are basic and only require special block initialisation
1079 * callouts for the data formating, others (directories) are so specialised they
1080 * handle everything themselves.
1081 *
1082 * XXX (dgc): investigate whether directory conversion can use the generic
1083 * formatting callout. It should be possible - it's just a very complex
1084 * formatter.
1085 */
1086STATIC int /* error */
1087xfs_bmap_add_attrfork_local(
1088 xfs_trans_t *tp, /* transaction pointer */
1089 xfs_inode_t *ip, /* incore inode pointer */
1090 xfs_fsblock_t *firstblock, /* first block allocated */
1091 xfs_bmap_free_t *flist, /* blocks to free at commit */
1092 int *flags) /* inode logging flags */
1093{
1094 xfs_da_args_t dargs; /* args for dir/attr code */
1095
1096 if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
1097 return 0;
1098
1099 if (S_ISDIR(ip->i_d.di_mode)) {
1100 memset(&dargs, 0, sizeof(dargs));
1101 dargs.geo = ip->i_mount->m_dir_geo;
1102 dargs.dp = ip;
1103 dargs.firstblock = firstblock;
1104 dargs.flist = flist;
1105 dargs.total = dargs.geo->fsbcount;
1106 dargs.whichfork = XFS_DATA_FORK;
1107 dargs.trans = tp;
1108 return xfs_dir2_sf_to_block(&dargs);
1109 }
1110
1111 if (S_ISLNK(ip->i_d.di_mode))
1112 return xfs_bmap_local_to_extents(tp, ip, firstblock, 1,
1113 flags, XFS_DATA_FORK,
1114 xfs_symlink_local_to_remote);
1115
1116 /* should only be called for types that support local format data */
1117 ASSERT(0);
1118 return -EFSCORRUPTED;
1119}
1120
1121/*
1122 * Convert inode from non-attributed to attributed.
1123 * Must not be in a transaction, ip must not be locked.
1124 */
1125int /* error code */
1126xfs_bmap_add_attrfork(
1127 xfs_inode_t *ip, /* incore inode pointer */
1128 int size, /* space new attribute needs */
1129 int rsvd) /* xact may use reserved blks */
1130{
1131 xfs_fsblock_t firstblock; /* 1st block/ag allocated */
1132 xfs_bmap_free_t flist; /* freed extent records */
1133 xfs_mount_t *mp; /* mount structure */
1134 xfs_trans_t *tp; /* transaction pointer */
1135 int blks; /* space reservation */
1136 int version = 1; /* superblock attr version */
1137 int committed; /* xaction was committed */
1138 int logflags; /* logging flags */
1139 int error; /* error return value */
1140 int cancel_flags = 0;
1141
1142 ASSERT(XFS_IFORK_Q(ip) == 0);
1143
1144 mp = ip->i_mount;
1145 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1146 tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
1147 blks = XFS_ADDAFORK_SPACE_RES(mp);
1148 if (rsvd)
1149 tp->t_flags |= XFS_TRANS_RESERVE;
1150 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_addafork, blks, 0);
1151 if (error) {
1152 xfs_trans_cancel(tp, 0);
1153 return error;
1154 }
1155 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1156 xfs_ilock(ip, XFS_ILOCK_EXCL);
1157 error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
1158 XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
1159 XFS_QMOPT_RES_REGBLKS);
1160 if (error)
1161 goto trans_cancel;
1162 cancel_flags |= XFS_TRANS_ABORT;
1163 if (XFS_IFORK_Q(ip))
1164 goto trans_cancel;
1165 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
1166 /*
1167 * For inodes coming from pre-6.2 filesystems.
1168 */
1169 ASSERT(ip->i_d.di_aformat == 0);
1170 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1171 }
1172 ASSERT(ip->i_d.di_anextents == 0);
1173
1174 xfs_trans_ijoin(tp, ip, 0);
1175 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1176
1177 switch (ip->i_d.di_format) {
1178 case XFS_DINODE_FMT_DEV:
1179 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1180 break;
1181 case XFS_DINODE_FMT_UUID:
1182 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
1183 break;
1184 case XFS_DINODE_FMT_LOCAL:
1185 case XFS_DINODE_FMT_EXTENTS:
1186 case XFS_DINODE_FMT_BTREE:
1187 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1188 if (!ip->i_d.di_forkoff)
1189 ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1190 else if (mp->m_flags & XFS_MOUNT_ATTR2)
1191 version = 2;
1192 break;
1193 default:
1194 ASSERT(0);
1195 error = -EINVAL;
1196 goto trans_cancel;
1197 }
1198
1199 ASSERT(ip->i_afp == NULL);
1200 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
1201 ip->i_afp->if_flags = XFS_IFEXTENTS;
1202 logflags = 0;
1203 xfs_bmap_init(&flist, &firstblock);
1204 switch (ip->i_d.di_format) {
1205 case XFS_DINODE_FMT_LOCAL:
1206 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
1207 &logflags);
1208 break;
1209 case XFS_DINODE_FMT_EXTENTS:
1210 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
1211 &flist, &logflags);
1212 break;
1213 case XFS_DINODE_FMT_BTREE:
1214 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
1215 &logflags);
1216 break;
1217 default:
1218 error = 0;
1219 break;
1220 }
1221 if (logflags)
1222 xfs_trans_log_inode(tp, ip, logflags);
1223 if (error)
1224 goto bmap_cancel;
1225 if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1226 (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
1227 __int64_t sbfields = 0;
1228
1229 spin_lock(&mp->m_sb_lock);
1230 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1231 xfs_sb_version_addattr(&mp->m_sb);
1232 sbfields |= XFS_SB_VERSIONNUM;
1233 }
1234 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1235 xfs_sb_version_addattr2(&mp->m_sb);
1236 sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
1237 }
1238 if (sbfields) {
1239 spin_unlock(&mp->m_sb_lock);
1240 xfs_mod_sb(tp, sbfields);
1241 } else
1242 spin_unlock(&mp->m_sb_lock);
1243 }
1244
1245 error = xfs_bmap_finish(&tp, &flist, &committed);
1246 if (error)
1247 goto bmap_cancel;
1248 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1249 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1250 return error;
1251
1252bmap_cancel:
1253 xfs_bmap_cancel(&flist);
1254trans_cancel:
1255 xfs_trans_cancel(tp, cancel_flags);
1256 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1257 return error;
1258}
1259
1260/*
1261 * Internal and external extent tree search functions.
1262 */
1263
1264/*
1265 * Read in the extents to if_extents.
1266 * All inode fields are set up by caller, we just traverse the btree
1267 * and copy the records in. If the file system cannot contain unwritten
1268 * extents, the records are checked for no "state" flags.
1269 */
1270int /* error */
1271xfs_bmap_read_extents(
1272 xfs_trans_t *tp, /* transaction pointer */
1273 xfs_inode_t *ip, /* incore inode */
1274 int whichfork) /* data or attr fork */
1275{
1276 struct xfs_btree_block *block; /* current btree block */
1277 xfs_fsblock_t bno; /* block # of "block" */
1278 xfs_buf_t *bp; /* buffer for "block" */
1279 int error; /* error return value */
1280 xfs_exntfmt_t exntf; /* XFS_EXTFMT_NOSTATE, if checking */
1281 xfs_extnum_t i, j; /* index into the extents list */
1282 xfs_ifork_t *ifp; /* fork structure */
1283 int level; /* btree level, for checking */
1284 xfs_mount_t *mp; /* file system mount structure */
1285 __be64 *pp; /* pointer to block address */
1286 /* REFERENCED */
1287 xfs_extnum_t room; /* number of entries there's room for */
1288
1289 bno = NULLFSBLOCK;
1290 mp = ip->i_mount;
1291 ifp = XFS_IFORK_PTR(ip, whichfork);
1292 exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
1293 XFS_EXTFMT_INODE(ip);
1294 block = ifp->if_broot;
1295 /*
1296 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1297 */
1298 level = be16_to_cpu(block->bb_level);
1299 ASSERT(level > 0);
1300 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
1301 bno = be64_to_cpu(*pp);
1302 ASSERT(bno != NULLDFSBNO);
1303 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
1304 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
1305 /*
1306 * Go down the tree until leaf level is reached, following the first
1307 * pointer (leftmost) at each level.
1308 */
1309 while (level-- > 0) {
1310 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1311 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1312 if (error)
1313 return error;
1314 block = XFS_BUF_TO_BLOCK(bp);
1315 XFS_WANT_CORRUPTED_GOTO(
1316 xfs_bmap_sanity_check(mp, bp, level),
1317 error0);
1318 if (level == 0)
1319 break;
1320 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
1321 bno = be64_to_cpu(*pp);
1322 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
1323 xfs_trans_brelse(tp, bp);
1324 }
1325 /*
1326 * Here with bp and block set to the leftmost leaf node in the tree.
1327 */
1328 room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1329 i = 0;
1330 /*
1331 * Loop over all leaf nodes. Copy information to the extent records.
1332 */
1333 for (;;) {
1334 xfs_bmbt_rec_t *frp;
1335 xfs_fsblock_t nextbno;
1336 xfs_extnum_t num_recs;
1337 xfs_extnum_t start;
1338
1339 num_recs = xfs_btree_get_numrecs(block);
1340 if (unlikely(i + num_recs > room)) {
1341 ASSERT(i + num_recs <= room);
1342 xfs_warn(ip->i_mount,
1343 "corrupt dinode %Lu, (btree extents).",
1344 (unsigned long long) ip->i_ino);
1345 XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
1346 XFS_ERRLEVEL_LOW, ip->i_mount, block);
1347 goto error0;
1348 }
1349 XFS_WANT_CORRUPTED_GOTO(
1350 xfs_bmap_sanity_check(mp, bp, 0),
1351 error0);
1352 /*
1353 * Read-ahead the next leaf block, if any.
1354 */
1355 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
1356 if (nextbno != NULLFSBLOCK)
1357 xfs_btree_reada_bufl(mp, nextbno, 1,
1358 &xfs_bmbt_buf_ops);
1359 /*
1360 * Copy records into the extent records.
1361 */
1362 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1363 start = i;
1364 for (j = 0; j < num_recs; j++, i++, frp++) {
1365 xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
1366 trp->l0 = be64_to_cpu(frp->l0);
1367 trp->l1 = be64_to_cpu(frp->l1);
1368 }
1369 if (exntf == XFS_EXTFMT_NOSTATE) {
1370 /*
1371 * Check all attribute bmap btree records and
1372 * any "older" data bmap btree records for a
1373 * set bit in the "extent flag" position.
1374 */
1375 if (unlikely(xfs_check_nostate_extents(ifp,
1376 start, num_recs))) {
1377 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
1378 XFS_ERRLEVEL_LOW,
1379 ip->i_mount);
1380 goto error0;
1381 }
1382 }
1383 xfs_trans_brelse(tp, bp);
1384 bno = nextbno;
1385 /*
1386 * If we've reached the end, stop.
1387 */
1388 if (bno == NULLFSBLOCK)
1389 break;
1390 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1391 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1392 if (error)
1393 return error;
1394 block = XFS_BUF_TO_BLOCK(bp);
1395 }
1396 ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
1397 ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
1398 XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
1399 return 0;
1400error0:
1401 xfs_trans_brelse(tp, bp);
1402 return -EFSCORRUPTED;
1403}
1404
1405
1406/*
1407 * Search the extent records for the entry containing block bno.
1408 * If bno lies in a hole, point to the next entry. If bno lies
1409 * past eof, *eofp will be set, and *prevp will contain the last
1410 * entry (null if none). Else, *lastxp will be set to the index
1411 * of the found entry; *gotp will contain the entry.
1412 */
1413STATIC xfs_bmbt_rec_host_t * /* pointer to found extent entry */
1414xfs_bmap_search_multi_extents(
1415 xfs_ifork_t *ifp, /* inode fork pointer */
1416 xfs_fileoff_t bno, /* block number searched for */
1417 int *eofp, /* out: end of file found */
1418 xfs_extnum_t *lastxp, /* out: last extent index */
1419 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
1420 xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
1421{
1422 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
1423 xfs_extnum_t lastx; /* last extent index */
1424
1425 /*
1426 * Initialize the extent entry structure to catch access to
1427 * uninitialized br_startblock field.
1428 */
1429 gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
1430 gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
1431 gotp->br_state = XFS_EXT_INVALID;
1432#if XFS_BIG_BLKNOS
1433 gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
1434#else
1435 gotp->br_startblock = 0xffffa5a5;
1436#endif
1437 prevp->br_startoff = NULLFILEOFF;
1438
1439 ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
1440 if (lastx > 0) {
1441 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
1442 }
1443 if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1444 xfs_bmbt_get_all(ep, gotp);
1445 *eofp = 0;
1446 } else {
1447 if (lastx > 0) {
1448 *gotp = *prevp;
1449 }
1450 *eofp = 1;
1451 ep = NULL;
1452 }
1453 *lastxp = lastx;
1454 return ep;
1455}
1456
1457/*
1458 * Search the extents list for the inode, for the extent containing bno.
1459 * If bno lies in a hole, point to the next entry. If bno lies past eof,
1460 * *eofp will be set, and *prevp will contain the last entry (null if none).
1461 * Else, *lastxp will be set to the index of the found
1462 * entry; *gotp will contain the entry.
1463 */
1464STATIC xfs_bmbt_rec_host_t * /* pointer to found extent entry */
1465xfs_bmap_search_extents(
1466 xfs_inode_t *ip, /* incore inode pointer */
1467 xfs_fileoff_t bno, /* block number searched for */
1468 int fork, /* data or attr fork */
1469 int *eofp, /* out: end of file found */
1470 xfs_extnum_t *lastxp, /* out: last extent index */
1471 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
1472 xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
1473{
1474 xfs_ifork_t *ifp; /* inode fork pointer */
1475 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
1476
1477 XFS_STATS_INC(xs_look_exlist);
1478 ifp = XFS_IFORK_PTR(ip, fork);
1479
1480 ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
1481
1482 if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
1483 !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
1484 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
1485 "Access to block zero in inode %llu "
1486 "start_block: %llx start_off: %llx "
1487 "blkcnt: %llx extent-state: %x lastx: %x",
1488 (unsigned long long)ip->i_ino,
1489 (unsigned long long)gotp->br_startblock,
1490 (unsigned long long)gotp->br_startoff,
1491 (unsigned long long)gotp->br_blockcount,
1492 gotp->br_state, *lastxp);
1493 *lastxp = NULLEXTNUM;
1494 *eofp = 1;
1495 return NULL;
1496 }
1497 return ep;
1498}
1499
1500/*
1501 * Returns the file-relative block number of the first unused block(s)
1502 * in the file with at least "len" logically contiguous blocks free.
1503 * This is the lowest-address hole if the file has holes, else the first block
1504 * past the end of file.
1505 * Return 0 if the file is currently local (in-inode).
1506 */
1507int /* error */
1508xfs_bmap_first_unused(
1509 xfs_trans_t *tp, /* transaction pointer */
1510 xfs_inode_t *ip, /* incore inode */
1511 xfs_extlen_t len, /* size of hole to find */
1512 xfs_fileoff_t *first_unused, /* unused block */
1513 int whichfork) /* data or attr fork */
1514{
1515 int error; /* error return value */
1516 int idx; /* extent record index */
1517 xfs_ifork_t *ifp; /* inode fork pointer */
1518 xfs_fileoff_t lastaddr; /* last block number seen */
1519 xfs_fileoff_t lowest; /* lowest useful block */
1520 xfs_fileoff_t max; /* starting useful block */
1521 xfs_fileoff_t off; /* offset for this block */
1522 xfs_extnum_t nextents; /* number of extent entries */
1523
1524 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
1525 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
1526 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1527 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1528 *first_unused = 0;
1529 return 0;
1530 }
1531 ifp = XFS_IFORK_PTR(ip, whichfork);
1532 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1533 (error = xfs_iread_extents(tp, ip, whichfork)))
1534 return error;
1535 lowest = *first_unused;
1536 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1537 for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
1538 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
1539 off = xfs_bmbt_get_startoff(ep);
1540 /*
1541 * See if the hole before this extent will work.
1542 */
1543 if (off >= lowest + len && off - max >= len) {
1544 *first_unused = max;
1545 return 0;
1546 }
1547 lastaddr = off + xfs_bmbt_get_blockcount(ep);
1548 max = XFS_FILEOFF_MAX(lastaddr, lowest);
1549 }
1550 *first_unused = max;
1551 return 0;
1552}
1553
1554/*
1555 * Returns the file-relative block number of the last block - 1 before
1556 * last_block (input value) in the file.
1557 * This is not based on i_size, it is based on the extent records.
1558 * Returns 0 for local files, as they do not have extent records.
1559 */
1560int /* error */
1561xfs_bmap_last_before(
1562 xfs_trans_t *tp, /* transaction pointer */
1563 xfs_inode_t *ip, /* incore inode */
1564 xfs_fileoff_t *last_block, /* last block */
1565 int whichfork) /* data or attr fork */
1566{
1567 xfs_fileoff_t bno; /* input file offset */
1568 int eof; /* hit end of file */
1569 xfs_bmbt_rec_host_t *ep; /* pointer to last extent */
1570 int error; /* error return value */
1571 xfs_bmbt_irec_t got; /* current extent value */
1572 xfs_ifork_t *ifp; /* inode fork pointer */
1573 xfs_extnum_t lastx; /* last extent used */
1574 xfs_bmbt_irec_t prev; /* previous extent value */
1575
1576 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1577 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
1578 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
1579 return -EIO;
1580 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1581 *last_block = 0;
1582 return 0;
1583 }
1584 ifp = XFS_IFORK_PTR(ip, whichfork);
1585 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1586 (error = xfs_iread_extents(tp, ip, whichfork)))
1587 return error;
1588 bno = *last_block - 1;
1589 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
1590 &prev);
1591 if (eof || xfs_bmbt_get_startoff(ep) > bno) {
1592 if (prev.br_startoff == NULLFILEOFF)
1593 *last_block = 0;
1594 else
1595 *last_block = prev.br_startoff + prev.br_blockcount;
1596 }
1597 /*
1598 * Otherwise *last_block is already the right answer.
1599 */
1600 return 0;
1601}
1602
1603int
1604xfs_bmap_last_extent(
1605 struct xfs_trans *tp,
1606 struct xfs_inode *ip,
1607 int whichfork,
1608 struct xfs_bmbt_irec *rec,
1609 int *is_empty)
1610{
1611 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
1612 int error;
1613 int nextents;
1614
1615 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1616 error = xfs_iread_extents(tp, ip, whichfork);
1617 if (error)
1618 return error;
1619 }
1620
1621 nextents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
1622 if (nextents == 0) {
1623 *is_empty = 1;
1624 return 0;
1625 }
1626
1627 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
1628 *is_empty = 0;
1629 return 0;
1630}
1631
1632/*
1633 * Check the last inode extent to determine whether this allocation will result
1634 * in blocks being allocated at the end of the file. When we allocate new data
1635 * blocks at the end of the file which do not start at the previous data block,
1636 * we will try to align the new blocks at stripe unit boundaries.
1637 *
1638 * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1639 * at, or past the EOF.
1640 */
1641STATIC int
1642xfs_bmap_isaeof(
1643 struct xfs_bmalloca *bma,
1644 int whichfork)
1645{
1646 struct xfs_bmbt_irec rec;
1647 int is_empty;
1648 int error;
1649
1650 bma->aeof = 0;
1651 error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1652 &is_empty);
1653 if (error)
1654 return error;
1655
1656 if (is_empty) {
1657 bma->aeof = 1;
1658 return 0;
1659 }
1660
1661 /*
1662 * Check if we are allocation or past the last extent, or at least into
1663 * the last delayed allocated extent.
1664 */
1665 bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1666 (bma->offset >= rec.br_startoff &&
1667 isnullstartblock(rec.br_startblock));
1668 return 0;
1669}
1670
1671/*
1672 * Returns the file-relative block number of the first block past eof in
1673 * the file. This is not based on i_size, it is based on the extent records.
1674 * Returns 0 for local files, as they do not have extent records.
1675 */
1676int
1677xfs_bmap_last_offset(
1678 struct xfs_inode *ip,
1679 xfs_fileoff_t *last_block,
1680 int whichfork)
1681{
1682 struct xfs_bmbt_irec rec;
1683 int is_empty;
1684 int error;
1685
1686 *last_block = 0;
1687
1688 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
1689 return 0;
1690
1691 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1692 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1693 return -EIO;
1694
1695 error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1696 if (error || is_empty)
1697 return error;
1698
1699 *last_block = rec.br_startoff + rec.br_blockcount;
1700 return 0;
1701}
1702
1703/*
1704 * Returns whether the selected fork of the inode has exactly one
1705 * block or not. For the data fork we check this matches di_size,
1706 * implying the file's range is 0..bsize-1.
1707 */
1708int /* 1=>1 block, 0=>otherwise */
1709xfs_bmap_one_block(
1710 xfs_inode_t *ip, /* incore inode */
1711 int whichfork) /* data or attr fork */
1712{
1713 xfs_bmbt_rec_host_t *ep; /* ptr to fork's extent */
1714 xfs_ifork_t *ifp; /* inode fork pointer */
1715 int rval; /* return value */
1716 xfs_bmbt_irec_t s; /* internal version of extent */
1717
1718#ifndef DEBUG
1719 if (whichfork == XFS_DATA_FORK)
1720 return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
1721#endif /* !DEBUG */
1722 if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
1723 return 0;
1724 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1725 return 0;
1726 ifp = XFS_IFORK_PTR(ip, whichfork);
1727 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1728 ep = xfs_iext_get_ext(ifp, 0);
1729 xfs_bmbt_get_all(ep, &s);
1730 rval = s.br_startoff == 0 && s.br_blockcount == 1;
1731 if (rval && whichfork == XFS_DATA_FORK)
1732 ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
1733 return rval;
1734}
1735
1736/*
1737 * Extent tree manipulation functions used during allocation.
1738 */
1739
1740/*
1741 * Convert a delayed allocation to a real allocation.
1742 */
1743STATIC int /* error */
1744xfs_bmap_add_extent_delay_real(
1745 struct xfs_bmalloca *bma)
1746{
1747 struct xfs_bmbt_irec *new = &bma->got;
1748 int diff; /* temp value */
1749 xfs_bmbt_rec_host_t *ep; /* extent entry for idx */
1750 int error; /* error return value */
1751 int i; /* temp state */
1752 xfs_ifork_t *ifp; /* inode fork pointer */
1753 xfs_fileoff_t new_endoff; /* end offset of new entry */
1754 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
1755 /* left is 0, right is 1, prev is 2 */
1756 int rval=0; /* return value (logging flags) */
1757 int state = 0;/* state bits, accessed thru macros */
1758 xfs_filblks_t da_new; /* new count del alloc blocks used */
1759 xfs_filblks_t da_old; /* old count del alloc blocks used */
1760 xfs_filblks_t temp=0; /* value for da_new calculations */
1761 xfs_filblks_t temp2=0;/* value for da_new calculations */
1762 int tmp_rval; /* partial logging flags */
1763
1764 ifp = XFS_IFORK_PTR(bma->ip, XFS_DATA_FORK);
1765
1766 ASSERT(bma->idx >= 0);
1767 ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
1768 ASSERT(!isnullstartblock(new->br_startblock));
1769 ASSERT(!bma->cur ||
1770 (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1771
1772 XFS_STATS_INC(xs_add_exlist);
1773
1774#define LEFT r[0]
1775#define RIGHT r[1]
1776#define PREV r[2]
1777
1778 /*
1779 * Set up a bunch of variables to make the tests simpler.
1780 */
1781 ep = xfs_iext_get_ext(ifp, bma->idx);
1782 xfs_bmbt_get_all(ep, &PREV);
1783 new_endoff = new->br_startoff + new->br_blockcount;
1784 ASSERT(PREV.br_startoff <= new->br_startoff);
1785 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1786
1787 da_old = startblockval(PREV.br_startblock);
1788 da_new = 0;
1789
1790 /*
1791 * Set flags determining what part of the previous delayed allocation
1792 * extent is being replaced by a real allocation.
1793 */
1794 if (PREV.br_startoff == new->br_startoff)
1795 state |= BMAP_LEFT_FILLING;
1796 if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1797 state |= BMAP_RIGHT_FILLING;
1798
1799 /*
1800 * Check and set flags if this segment has a left neighbor.
1801 * Don't set contiguous if the combined extent would be too large.
1802 */
1803 if (bma->idx > 0) {
1804 state |= BMAP_LEFT_VALID;
1805 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
1806
1807 if (isnullstartblock(LEFT.br_startblock))
1808 state |= BMAP_LEFT_DELAY;
1809 }
1810
1811 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1812 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1813 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1814 LEFT.br_state == new->br_state &&
1815 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1816 state |= BMAP_LEFT_CONTIG;
1817
1818 /*
1819 * Check and set flags if this segment has a right neighbor.
1820 * Don't set contiguous if the combined extent would be too large.
1821 * Also check for all-three-contiguous being too large.
1822 */
1823 if (bma->idx < bma->ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1824 state |= BMAP_RIGHT_VALID;
1825 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
1826
1827 if (isnullstartblock(RIGHT.br_startblock))
1828 state |= BMAP_RIGHT_DELAY;
1829 }
1830
1831 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1832 new_endoff == RIGHT.br_startoff &&
1833 new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1834 new->br_state == RIGHT.br_state &&
1835 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1836 ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1837 BMAP_RIGHT_FILLING)) !=
1838 (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1839 BMAP_RIGHT_FILLING) ||
1840 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1841 <= MAXEXTLEN))
1842 state |= BMAP_RIGHT_CONTIG;
1843
1844 error = 0;
1845 /*
1846 * Switch out based on the FILLING and CONTIG state bits.
1847 */
1848 switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1849 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1850 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1851 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1852 /*
1853 * Filling in all of a previously delayed allocation extent.
1854 * The left and right neighbors are both contiguous with new.
1855 */
1856 bma->idx--;
1857 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1858 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1859 LEFT.br_blockcount + PREV.br_blockcount +
1860 RIGHT.br_blockcount);
1861 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1862
1863 xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
1864 bma->ip->i_d.di_nextents--;
1865 if (bma->cur == NULL)
1866 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1867 else {
1868 rval = XFS_ILOG_CORE;
1869 error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1870 RIGHT.br_startblock,
1871 RIGHT.br_blockcount, &i);
1872 if (error)
1873 goto done;
1874 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1875 error = xfs_btree_delete(bma->cur, &i);
1876 if (error)
1877 goto done;
1878 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1879 error = xfs_btree_decrement(bma->cur, 0, &i);
1880 if (error)
1881 goto done;
1882 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1883 error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1884 LEFT.br_startblock,
1885 LEFT.br_blockcount +
1886 PREV.br_blockcount +
1887 RIGHT.br_blockcount, LEFT.br_state);
1888 if (error)
1889 goto done;
1890 }
1891 break;
1892
1893 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1894 /*
1895 * Filling in all of a previously delayed allocation extent.
1896 * The left neighbor is contiguous, the right is not.
1897 */
1898 bma->idx--;
1899
1900 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1901 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1902 LEFT.br_blockcount + PREV.br_blockcount);
1903 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1904
1905 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1906 if (bma->cur == NULL)
1907 rval = XFS_ILOG_DEXT;
1908 else {
1909 rval = 0;
1910 error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1911 LEFT.br_startblock, LEFT.br_blockcount,
1912 &i);
1913 if (error)
1914 goto done;
1915 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1916 error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1917 LEFT.br_startblock,
1918 LEFT.br_blockcount +
1919 PREV.br_blockcount, LEFT.br_state);
1920 if (error)
1921 goto done;
1922 }
1923 break;
1924
1925 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1926 /*
1927 * Filling in all of a previously delayed allocation extent.
1928 * The right neighbor is contiguous, the left is not.
1929 */
1930 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1931 xfs_bmbt_set_startblock(ep, new->br_startblock);
1932 xfs_bmbt_set_blockcount(ep,
1933 PREV.br_blockcount + RIGHT.br_blockcount);
1934 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1935
1936 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1937 if (bma->cur == NULL)
1938 rval = XFS_ILOG_DEXT;
1939 else {
1940 rval = 0;
1941 error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1942 RIGHT.br_startblock,
1943 RIGHT.br_blockcount, &i);
1944 if (error)
1945 goto done;
1946 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1947 error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
1948 new->br_startblock,
1949 PREV.br_blockcount +
1950 RIGHT.br_blockcount, PREV.br_state);
1951 if (error)
1952 goto done;
1953 }
1954 break;
1955
1956 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1957 /*
1958 * Filling in all of a previously delayed allocation extent.
1959 * Neither the left nor right neighbors are contiguous with
1960 * the new one.
1961 */
1962 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1963 xfs_bmbt_set_startblock(ep, new->br_startblock);
1964 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1965
1966 bma->ip->i_d.di_nextents++;
1967 if (bma->cur == NULL)
1968 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1969 else {
1970 rval = XFS_ILOG_CORE;
1971 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1972 new->br_startblock, new->br_blockcount,
1973 &i);
1974 if (error)
1975 goto done;
1976 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1977 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1978 error = xfs_btree_insert(bma->cur, &i);
1979 if (error)
1980 goto done;
1981 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1982 }
1983 break;
1984
1985 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1986 /*
1987 * Filling in the first part of a previous delayed allocation.
1988 * The left neighbor is contiguous.
1989 */
1990 trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1991 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
1992 LEFT.br_blockcount + new->br_blockcount);
1993 xfs_bmbt_set_startoff(ep,
1994 PREV.br_startoff + new->br_blockcount);
1995 trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1996
1997 temp = PREV.br_blockcount - new->br_blockcount;
1998 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1999 xfs_bmbt_set_blockcount(ep, temp);
2000 if (bma->cur == NULL)
2001 rval = XFS_ILOG_DEXT;
2002 else {
2003 rval = 0;
2004 error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
2005 LEFT.br_startblock, LEFT.br_blockcount,
2006 &i);
2007 if (error)
2008 goto done;
2009 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2010 error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
2011 LEFT.br_startblock,
2012 LEFT.br_blockcount +
2013 new->br_blockcount,
2014 LEFT.br_state);
2015 if (error)
2016 goto done;
2017 }
2018 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2019 startblockval(PREV.br_startblock));
2020 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2021 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2022
2023 bma->idx--;
2024 break;
2025
2026 case BMAP_LEFT_FILLING:
2027 /*
2028 * Filling in the first part of a previous delayed allocation.
2029 * The left neighbor is not contiguous.
2030 */
2031 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2032 xfs_bmbt_set_startoff(ep, new_endoff);
2033 temp = PREV.br_blockcount - new->br_blockcount;
2034 xfs_bmbt_set_blockcount(ep, temp);
2035 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
2036 bma->ip->i_d.di_nextents++;
2037 if (bma->cur == NULL)
2038 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2039 else {
2040 rval = XFS_ILOG_CORE;
2041 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2042 new->br_startblock, new->br_blockcount,
2043 &i);
2044 if (error)
2045 goto done;
2046 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2047 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2048 error = xfs_btree_insert(bma->cur, &i);
2049 if (error)
2050 goto done;
2051 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2052 }
2053
2054 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2055 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2056 bma->firstblock, bma->flist,
2057 &bma->cur, 1, &tmp_rval, XFS_DATA_FORK);
2058 rval |= tmp_rval;
2059 if (error)
2060 goto done;
2061 }
2062 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2063 startblockval(PREV.br_startblock) -
2064 (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2065 ep = xfs_iext_get_ext(ifp, bma->idx + 1);
2066 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2067 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2068 break;
2069
2070 case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2071 /*
2072 * Filling in the last part of a previous delayed allocation.
2073 * The right neighbor is contiguous with the new allocation.
2074 */
2075 temp = PREV.br_blockcount - new->br_blockcount;
2076 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2077 xfs_bmbt_set_blockcount(ep, temp);
2078 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
2079 new->br_startoff, new->br_startblock,
2080 new->br_blockcount + RIGHT.br_blockcount,
2081 RIGHT.br_state);
2082 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2083 if (bma->cur == NULL)
2084 rval = XFS_ILOG_DEXT;
2085 else {
2086 rval = 0;
2087 error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
2088 RIGHT.br_startblock,
2089 RIGHT.br_blockcount, &i);
2090 if (error)
2091 goto done;
2092 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2093 error = xfs_bmbt_update(bma->cur, new->br_startoff,
2094 new->br_startblock,
2095 new->br_blockcount +
2096 RIGHT.br_blockcount,
2097 RIGHT.br_state);
2098 if (error)
2099 goto done;
2100 }
2101
2102 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2103 startblockval(PREV.br_startblock));
2104 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2105 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2106 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2107
2108 bma->idx++;
2109 break;
2110
2111 case BMAP_RIGHT_FILLING:
2112 /*
2113 * Filling in the last part of a previous delayed allocation.
2114 * The right neighbor is not contiguous.
2115 */
2116 temp = PREV.br_blockcount - new->br_blockcount;
2117 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2118 xfs_bmbt_set_blockcount(ep, temp);
2119 xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
2120 bma->ip->i_d.di_nextents++;
2121 if (bma->cur == NULL)
2122 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2123 else {
2124 rval = XFS_ILOG_CORE;
2125 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2126 new->br_startblock, new->br_blockcount,
2127 &i);
2128 if (error)
2129 goto done;
2130 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2131 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2132 error = xfs_btree_insert(bma->cur, &i);
2133 if (error)
2134 goto done;
2135 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2136 }
2137
2138 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2139 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2140 bma->firstblock, bma->flist, &bma->cur, 1,
2141 &tmp_rval, XFS_DATA_FORK);
2142 rval |= tmp_rval;
2143 if (error)
2144 goto done;
2145 }
2146 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2147 startblockval(PREV.br_startblock) -
2148 (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2149 ep = xfs_iext_get_ext(ifp, bma->idx);
2150 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2151 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2152
2153 bma->idx++;
2154 break;
2155
2156 case 0:
2157 /*
2158 * Filling in the middle part of a previous delayed allocation.
2159 * Contiguity is impossible here.
2160 * This case is avoided almost all the time.
2161 *
2162 * We start with a delayed allocation:
2163 *
2164 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
2165 * PREV @ idx
2166 *
2167 * and we are allocating:
2168 * +rrrrrrrrrrrrrrrrr+
2169 * new
2170 *
2171 * and we set it up for insertion as:
2172 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
2173 * new
2174 * PREV @ idx LEFT RIGHT
2175 * inserted at idx + 1
2176 */
2177 temp = new->br_startoff - PREV.br_startoff;
2178 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
2179 trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
2180 xfs_bmbt_set_blockcount(ep, temp); /* truncate PREV */
2181 LEFT = *new;
2182 RIGHT.br_state = PREV.br_state;
2183 RIGHT.br_startblock = nullstartblock(
2184 (int)xfs_bmap_worst_indlen(bma->ip, temp2));
2185 RIGHT.br_startoff = new_endoff;
2186 RIGHT.br_blockcount = temp2;
2187 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
2188 xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
2189 bma->ip->i_d.di_nextents++;
2190 if (bma->cur == NULL)
2191 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2192 else {
2193 rval = XFS_ILOG_CORE;
2194 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2195 new->br_startblock, new->br_blockcount,
2196 &i);
2197 if (error)
2198 goto done;
2199 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2200 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2201 error = xfs_btree_insert(bma->cur, &i);
2202 if (error)
2203 goto done;
2204 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2205 }
2206
2207 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2208 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2209 bma->firstblock, bma->flist, &bma->cur,
2210 1, &tmp_rval, XFS_DATA_FORK);
2211 rval |= tmp_rval;
2212 if (error)
2213 goto done;
2214 }
2215 temp = xfs_bmap_worst_indlen(bma->ip, temp);
2216 temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
2217 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
2218 (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2219 if (diff > 0) {
2220 error = xfs_icsb_modify_counters(bma->ip->i_mount,
2221 XFS_SBS_FDBLOCKS,
2222 -((int64_t)diff), 0);
2223 ASSERT(!error);
2224 if (error)
2225 goto done;
2226 }
2227
2228 ep = xfs_iext_get_ext(ifp, bma->idx);
2229 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2230 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2231 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2232 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
2233 nullstartblock((int)temp2));
2234 trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2235
2236 bma->idx++;
2237 da_new = temp + temp2;
2238 break;
2239
2240 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2241 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2242 case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2243 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2244 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2245 case BMAP_LEFT_CONTIG:
2246 case BMAP_RIGHT_CONTIG:
2247 /*
2248 * These cases are all impossible.
2249 */
2250 ASSERT(0);
2251 }
2252
2253 /* convert to a btree if necessary */
2254 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2255 int tmp_logflags; /* partial log flag return val */
2256
2257 ASSERT(bma->cur == NULL);
2258 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2259 bma->firstblock, bma->flist, &bma->cur,
2260 da_old > 0, &tmp_logflags, XFS_DATA_FORK);
2261 bma->logflags |= tmp_logflags;
2262 if (error)
2263 goto done;
2264 }
2265
2266 /* adjust for changes in reserved delayed indirect blocks */
2267 if (da_old || da_new) {
2268 temp = da_new;
2269 if (bma->cur)
2270 temp += bma->cur->bc_private.b.allocated;
2271 ASSERT(temp <= da_old);
2272 if (temp < da_old)
2273 xfs_icsb_modify_counters(bma->ip->i_mount,
2274 XFS_SBS_FDBLOCKS,
2275 (int64_t)(da_old - temp), 0);
2276 }
2277
2278 /* clear out the allocated field, done with it now in any case. */
2279 if (bma->cur)
2280 bma->cur->bc_private.b.allocated = 0;
2281
2282 xfs_bmap_check_leaf_extents(bma->cur, bma->ip, XFS_DATA_FORK);
2283done:
2284 bma->logflags |= rval;
2285 return error;
2286#undef LEFT
2287#undef RIGHT
2288#undef PREV
2289}
2290
2291/*
2292 * Convert an unwritten allocation to a real allocation or vice versa.
2293 */
2294STATIC int /* error */
2295xfs_bmap_add_extent_unwritten_real(
2296 struct xfs_trans *tp,
2297 xfs_inode_t *ip, /* incore inode pointer */
2298 xfs_extnum_t *idx, /* extent number to update/insert */
2299 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
2300 xfs_bmbt_irec_t *new, /* new data to add to file extents */
2301 xfs_fsblock_t *first, /* pointer to firstblock variable */
2302 xfs_bmap_free_t *flist, /* list of extents to be freed */
2303 int *logflagsp) /* inode logging flags */
2304{
2305 xfs_btree_cur_t *cur; /* btree cursor */
2306 xfs_bmbt_rec_host_t *ep; /* extent entry for idx */
2307 int error; /* error return value */
2308 int i; /* temp state */
2309 xfs_ifork_t *ifp; /* inode fork pointer */
2310 xfs_fileoff_t new_endoff; /* end offset of new entry */
2311 xfs_exntst_t newext; /* new extent state */
2312 xfs_exntst_t oldext; /* old extent state */
2313 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
2314 /* left is 0, right is 1, prev is 2 */
2315 int rval=0; /* return value (logging flags) */
2316 int state = 0;/* state bits, accessed thru macros */
2317
2318 *logflagsp = 0;
2319
2320 cur = *curp;
2321 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2322
2323 ASSERT(*idx >= 0);
2324 ASSERT(*idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2325 ASSERT(!isnullstartblock(new->br_startblock));
2326
2327 XFS_STATS_INC(xs_add_exlist);
2328
2329#define LEFT r[0]
2330#define RIGHT r[1]
2331#define PREV r[2]
2332
2333 /*
2334 * Set up a bunch of variables to make the tests simpler.
2335 */
2336 error = 0;
2337 ep = xfs_iext_get_ext(ifp, *idx);
2338 xfs_bmbt_get_all(ep, &PREV);
2339 newext = new->br_state;
2340 oldext = (newext == XFS_EXT_UNWRITTEN) ?
2341 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
2342 ASSERT(PREV.br_state == oldext);
2343 new_endoff = new->br_startoff + new->br_blockcount;
2344 ASSERT(PREV.br_startoff <= new->br_startoff);
2345 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2346
2347 /*
2348 * Set flags determining what part of the previous oldext allocation
2349 * extent is being replaced by a newext allocation.
2350 */
2351 if (PREV.br_startoff == new->br_startoff)
2352 state |= BMAP_LEFT_FILLING;
2353 if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2354 state |= BMAP_RIGHT_FILLING;
2355
2356 /*
2357 * Check and set flags if this segment has a left neighbor.
2358 * Don't set contiguous if the combined extent would be too large.
2359 */
2360 if (*idx > 0) {
2361 state |= BMAP_LEFT_VALID;
2362 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
2363
2364 if (isnullstartblock(LEFT.br_startblock))
2365 state |= BMAP_LEFT_DELAY;
2366 }
2367
2368 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2369 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2370 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2371 LEFT.br_state == newext &&
2372 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2373 state |= BMAP_LEFT_CONTIG;
2374
2375 /*
2376 * Check and set flags if this segment has a right neighbor.
2377 * Don't set contiguous if the combined extent would be too large.
2378 * Also check for all-three-contiguous being too large.
2379 */
2380 if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
2381 state |= BMAP_RIGHT_VALID;
2382 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
2383 if (isnullstartblock(RIGHT.br_startblock))
2384 state |= BMAP_RIGHT_DELAY;
2385 }
2386
2387 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2388 new_endoff == RIGHT.br_startoff &&
2389 new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2390 newext == RIGHT.br_state &&
2391 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2392 ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2393 BMAP_RIGHT_FILLING)) !=
2394 (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2395 BMAP_RIGHT_FILLING) ||
2396 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2397 <= MAXEXTLEN))
2398 state |= BMAP_RIGHT_CONTIG;
2399
2400 /*
2401 * Switch out based on the FILLING and CONTIG state bits.
2402 */
2403 switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2404 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2405 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2406 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2407 /*
2408 * Setting all of a previous oldext extent to newext.
2409 * The left and right neighbors are both contiguous with new.
2410 */
2411 --*idx;
2412
2413 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2414 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2415 LEFT.br_blockcount + PREV.br_blockcount +
2416 RIGHT.br_blockcount);
2417 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2418
2419 xfs_iext_remove(ip, *idx + 1, 2, state);
2420 ip->i_d.di_nextents -= 2;
2421 if (cur == NULL)
2422 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2423 else {
2424 rval = XFS_ILOG_CORE;
2425 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2426 RIGHT.br_startblock,
2427 RIGHT.br_blockcount, &i)))
2428 goto done;
2429 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2430 if ((error = xfs_btree_delete(cur, &i)))
2431 goto done;
2432 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2433 if ((error = xfs_btree_decrement(cur, 0, &i)))
2434 goto done;
2435 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2436 if ((error = xfs_btree_delete(cur, &i)))
2437 goto done;
2438 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2439 if ((error = xfs_btree_decrement(cur, 0, &i)))
2440 goto done;
2441 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2442 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2443 LEFT.br_startblock,
2444 LEFT.br_blockcount + PREV.br_blockcount +
2445 RIGHT.br_blockcount, LEFT.br_state)))
2446 goto done;
2447 }
2448 break;
2449
2450 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2451 /*
2452 * Setting all of a previous oldext extent to newext.
2453 * The left neighbor is contiguous, the right is not.
2454 */
2455 --*idx;
2456
2457 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2458 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2459 LEFT.br_blockcount + PREV.br_blockcount);
2460 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2461
2462 xfs_iext_remove(ip, *idx + 1, 1, state);
2463 ip->i_d.di_nextents--;
2464 if (cur == NULL)
2465 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2466 else {
2467 rval = XFS_ILOG_CORE;
2468 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2469 PREV.br_startblock, PREV.br_blockcount,
2470 &i)))
2471 goto done;
2472 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2473 if ((error = xfs_btree_delete(cur, &i)))
2474 goto done;
2475 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2476 if ((error = xfs_btree_decrement(cur, 0, &i)))
2477 goto done;
2478 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2479 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2480 LEFT.br_startblock,
2481 LEFT.br_blockcount + PREV.br_blockcount,
2482 LEFT.br_state)))
2483 goto done;
2484 }
2485 break;
2486
2487 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2488 /*
2489 * Setting all of a previous oldext extent to newext.
2490 * The right neighbor is contiguous, the left is not.
2491 */
2492 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2493 xfs_bmbt_set_blockcount(ep,
2494 PREV.br_blockcount + RIGHT.br_blockcount);
2495 xfs_bmbt_set_state(ep, newext);
2496 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2497 xfs_iext_remove(ip, *idx + 1, 1, state);
2498 ip->i_d.di_nextents--;
2499 if (cur == NULL)
2500 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2501 else {
2502 rval = XFS_ILOG_CORE;
2503 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2504 RIGHT.br_startblock,
2505 RIGHT.br_blockcount, &i)))
2506 goto done;
2507 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2508 if ((error = xfs_btree_delete(cur, &i)))
2509 goto done;
2510 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2511 if ((error = xfs_btree_decrement(cur, 0, &i)))
2512 goto done;
2513 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2514 if ((error = xfs_bmbt_update(cur, new->br_startoff,
2515 new->br_startblock,
2516 new->br_blockcount + RIGHT.br_blockcount,
2517 newext)))
2518 goto done;
2519 }
2520 break;
2521
2522 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2523 /*
2524 * Setting all of a previous oldext extent to newext.
2525 * Neither the left nor right neighbors are contiguous with
2526 * the new one.
2527 */
2528 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2529 xfs_bmbt_set_state(ep, newext);
2530 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2531
2532 if (cur == NULL)
2533 rval = XFS_ILOG_DEXT;
2534 else {
2535 rval = 0;
2536 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2537 new->br_startblock, new->br_blockcount,
2538 &i)))
2539 goto done;
2540 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2541 if ((error = xfs_bmbt_update(cur, new->br_startoff,
2542 new->br_startblock, new->br_blockcount,
2543 newext)))
2544 goto done;
2545 }
2546 break;
2547
2548 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2549 /*
2550 * Setting the first part of a previous oldext extent to newext.
2551 * The left neighbor is contiguous.
2552 */
2553 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
2554 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
2555 LEFT.br_blockcount + new->br_blockcount);
2556 xfs_bmbt_set_startoff(ep,
2557 PREV.br_startoff + new->br_blockcount);
2558 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
2559
2560 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2561 xfs_bmbt_set_startblock(ep,
2562 new->br_startblock + new->br_blockcount);
2563 xfs_bmbt_set_blockcount(ep,
2564 PREV.br_blockcount - new->br_blockcount);
2565 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2566
2567 --*idx;
2568
2569 if (cur == NULL)
2570 rval = XFS_ILOG_DEXT;
2571 else {
2572 rval = 0;
2573 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2574 PREV.br_startblock, PREV.br_blockcount,
2575 &i)))
2576 goto done;
2577 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2578 if ((error = xfs_bmbt_update(cur,
2579 PREV.br_startoff + new->br_blockcount,
2580 PREV.br_startblock + new->br_blockcount,
2581 PREV.br_blockcount - new->br_blockcount,
2582 oldext)))
2583 goto done;
2584 if ((error = xfs_btree_decrement(cur, 0, &i)))
2585 goto done;
2586 error = xfs_bmbt_update(cur, LEFT.br_startoff,
2587 LEFT.br_startblock,
2588 LEFT.br_blockcount + new->br_blockcount,
2589 LEFT.br_state);
2590 if (error)
2591 goto done;
2592 }
2593 break;
2594
2595 case BMAP_LEFT_FILLING:
2596 /*
2597 * Setting the first part of a previous oldext extent to newext.
2598 * The left neighbor is not contiguous.
2599 */
2600 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2601 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
2602 xfs_bmbt_set_startoff(ep, new_endoff);
2603 xfs_bmbt_set_blockcount(ep,
2604 PREV.br_blockcount - new->br_blockcount);
2605 xfs_bmbt_set_startblock(ep,
2606 new->br_startblock + new->br_blockcount);
2607 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2608
2609 xfs_iext_insert(ip, *idx, 1, new, state);
2610 ip->i_d.di_nextents++;
2611 if (cur == NULL)
2612 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2613 else {
2614 rval = XFS_ILOG_CORE;
2615 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2616 PREV.br_startblock, PREV.br_blockcount,
2617 &i)))
2618 goto done;
2619 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2620 if ((error = xfs_bmbt_update(cur,
2621 PREV.br_startoff + new->br_blockcount,
2622 PREV.br_startblock + new->br_blockcount,
2623 PREV.br_blockcount - new->br_blockcount,
2624 oldext)))
2625 goto done;
2626 cur->bc_rec.b = *new;
2627 if ((error = xfs_btree_insert(cur, &i)))
2628 goto done;
2629 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2630 }
2631 break;
2632
2633 case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2634 /*
2635 * Setting the last part of a previous oldext extent to newext.
2636 * The right neighbor is contiguous with the new allocation.
2637 */
2638 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2639 xfs_bmbt_set_blockcount(ep,
2640 PREV.br_blockcount - new->br_blockcount);
2641 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2642
2643 ++*idx;
2644
2645 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2646 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2647 new->br_startoff, new->br_startblock,
2648 new->br_blockcount + RIGHT.br_blockcount, newext);
2649 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2650
2651 if (cur == NULL)
2652 rval = XFS_ILOG_DEXT;
2653 else {
2654 rval = 0;
2655 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2656 PREV.br_startblock,
2657 PREV.br_blockcount, &i)))
2658 goto done;
2659 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2660 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2661 PREV.br_startblock,
2662 PREV.br_blockcount - new->br_blockcount,
2663 oldext)))
2664 goto done;
2665 if ((error = xfs_btree_increment(cur, 0, &i)))
2666 goto done;
2667 if ((error = xfs_bmbt_update(cur, new->br_startoff,
2668 new->br_startblock,
2669 new->br_blockcount + RIGHT.br_blockcount,
2670 newext)))
2671 goto done;
2672 }
2673 break;
2674
2675 case BMAP_RIGHT_FILLING:
2676 /*
2677 * Setting the last part of a previous oldext extent to newext.
2678 * The right neighbor is not contiguous.
2679 */
2680 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2681 xfs_bmbt_set_blockcount(ep,
2682 PREV.br_blockcount - new->br_blockcount);
2683 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2684
2685 ++*idx;
2686 xfs_iext_insert(ip, *idx, 1, new, state);
2687
2688 ip->i_d.di_nextents++;
2689 if (cur == NULL)
2690 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2691 else {
2692 rval = XFS_ILOG_CORE;
2693 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2694 PREV.br_startblock, PREV.br_blockcount,
2695 &i)))
2696 goto done;
2697 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2698 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2699 PREV.br_startblock,
2700 PREV.br_blockcount - new->br_blockcount,
2701 oldext)))
2702 goto done;
2703 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2704 new->br_startblock, new->br_blockcount,
2705 &i)))
2706 goto done;
2707 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2708 cur->bc_rec.b.br_state = XFS_EXT_NORM;
2709 if ((error = xfs_btree_insert(cur, &i)))
2710 goto done;
2711 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2712 }
2713 break;
2714
2715 case 0:
2716 /*
2717 * Setting the middle part of a previous oldext extent to
2718 * newext. Contiguity is impossible here.
2719 * One extent becomes three extents.
2720 */
2721 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2722 xfs_bmbt_set_blockcount(ep,
2723 new->br_startoff - PREV.br_startoff);
2724 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2725
2726 r[0] = *new;
2727 r[1].br_startoff = new_endoff;
2728 r[1].br_blockcount =
2729 PREV.br_startoff + PREV.br_blockcount - new_endoff;
2730 r[1].br_startblock = new->br_startblock + new->br_blockcount;
2731 r[1].br_state = oldext;
2732
2733 ++*idx;
2734 xfs_iext_insert(ip, *idx, 2, &r[0], state);
2735
2736 ip->i_d.di_nextents += 2;
2737 if (cur == NULL)
2738 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2739 else {
2740 rval = XFS_ILOG_CORE;
2741 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2742 PREV.br_startblock, PREV.br_blockcount,
2743 &i)))
2744 goto done;
2745 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2746 /* new right extent - oldext */
2747 if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
2748 r[1].br_startblock, r[1].br_blockcount,
2749 r[1].br_state)))
2750 goto done;
2751 /* new left extent - oldext */
2752 cur->bc_rec.b = PREV;
2753 cur->bc_rec.b.br_blockcount =
2754 new->br_startoff - PREV.br_startoff;
2755 if ((error = xfs_btree_insert(cur, &i)))
2756 goto done;
2757 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2758 /*
2759 * Reset the cursor to the position of the new extent
2760 * we are about to insert as we can't trust it after
2761 * the previous insert.
2762 */
2763 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2764 new->br_startblock, new->br_blockcount,
2765 &i)))
2766 goto done;
2767 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2768 /* new middle extent - newext */
2769 cur->bc_rec.b.br_state = new->br_state;
2770 if ((error = xfs_btree_insert(cur, &i)))
2771 goto done;
2772 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2773 }
2774 break;
2775
2776 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2777 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2778 case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2779 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2780 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2781 case BMAP_LEFT_CONTIG:
2782 case BMAP_RIGHT_CONTIG:
2783 /*
2784 * These cases are all impossible.
2785 */
2786 ASSERT(0);
2787 }
2788
2789 /* convert to a btree if necessary */
2790 if (xfs_bmap_needs_btree(ip, XFS_DATA_FORK)) {
2791 int tmp_logflags; /* partial log flag return val */
2792
2793 ASSERT(cur == NULL);
2794 error = xfs_bmap_extents_to_btree(tp, ip, first, flist, &cur,
2795 0, &tmp_logflags, XFS_DATA_FORK);
2796 *logflagsp |= tmp_logflags;
2797 if (error)
2798 goto done;
2799 }
2800
2801 /* clear out the allocated field, done with it now in any case. */
2802 if (cur) {
2803 cur->bc_private.b.allocated = 0;
2804 *curp = cur;
2805 }
2806
2807 xfs_bmap_check_leaf_extents(*curp, ip, XFS_DATA_FORK);
2808done:
2809 *logflagsp |= rval;
2810 return error;
2811#undef LEFT
2812#undef RIGHT
2813#undef PREV
2814}
2815
2816/*
2817 * Convert a hole to a delayed allocation.
2818 */
2819STATIC void
2820xfs_bmap_add_extent_hole_delay(
2821 xfs_inode_t *ip, /* incore inode pointer */
2822 xfs_extnum_t *idx, /* extent number to update/insert */
2823 xfs_bmbt_irec_t *new) /* new data to add to file extents */
2824{
2825 xfs_ifork_t *ifp; /* inode fork pointer */
2826 xfs_bmbt_irec_t left; /* left neighbor extent entry */
2827 xfs_filblks_t newlen=0; /* new indirect size */
2828 xfs_filblks_t oldlen=0; /* old indirect size */
2829 xfs_bmbt_irec_t right; /* right neighbor extent entry */
2830 int state; /* state bits, accessed thru macros */
2831 xfs_filblks_t temp=0; /* temp for indirect calculations */
2832
2833 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2834 state = 0;
2835 ASSERT(isnullstartblock(new->br_startblock));
2836
2837 /*
2838 * Check and set flags if this segment has a left neighbor
2839 */
2840 if (*idx > 0) {
2841 state |= BMAP_LEFT_VALID;
2842 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
2843
2844 if (isnullstartblock(left.br_startblock))
2845 state |= BMAP_LEFT_DELAY;
2846 }
2847
2848 /*
2849 * Check and set flags if the current (right) segment exists.
2850 * If it doesn't exist, we're converting the hole at end-of-file.
2851 */
2852 if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
2853 state |= BMAP_RIGHT_VALID;
2854 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
2855
2856 if (isnullstartblock(right.br_startblock))
2857 state |= BMAP_RIGHT_DELAY;
2858 }
2859
2860 /*
2861 * Set contiguity flags on the left and right neighbors.
2862 * Don't let extents get too large, even if the pieces are contiguous.
2863 */
2864 if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2865 left.br_startoff + left.br_blockcount == new->br_startoff &&
2866 left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2867 state |= BMAP_LEFT_CONTIG;
2868
2869 if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2870 new->br_startoff + new->br_blockcount == right.br_startoff &&
2871 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2872 (!(state & BMAP_LEFT_CONTIG) ||
2873 (left.br_blockcount + new->br_blockcount +
2874 right.br_blockcount <= MAXEXTLEN)))
2875 state |= BMAP_RIGHT_CONTIG;
2876
2877 /*
2878 * Switch out based on the contiguity flags.
2879 */
2880 switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2881 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2882 /*
2883 * New allocation is contiguous with delayed allocations
2884 * on the left and on the right.
2885 * Merge all three into a single extent record.
2886 */
2887 --*idx;
2888 temp = left.br_blockcount + new->br_blockcount +
2889 right.br_blockcount;
2890
2891 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2892 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2893 oldlen = startblockval(left.br_startblock) +
2894 startblockval(new->br_startblock) +
2895 startblockval(right.br_startblock);
2896 newlen = xfs_bmap_worst_indlen(ip, temp);
2897 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2898 nullstartblock((int)newlen));
2899 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2900
2901 xfs_iext_remove(ip, *idx + 1, 1, state);
2902 break;
2903
2904 case BMAP_LEFT_CONTIG:
2905 /*
2906 * New allocation is contiguous with a delayed allocation
2907 * on the left.
2908 * Merge the new allocation with the left neighbor.
2909 */
2910 --*idx;
2911 temp = left.br_blockcount + new->br_blockcount;
2912
2913 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2914 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2915 oldlen = startblockval(left.br_startblock) +
2916 startblockval(new->br_startblock);
2917 newlen = xfs_bmap_worst_indlen(ip, temp);
2918 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2919 nullstartblock((int)newlen));
2920 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2921 break;
2922
2923 case BMAP_RIGHT_CONTIG:
2924 /*
2925 * New allocation is contiguous with a delayed allocation
2926 * on the right.
2927 * Merge the new allocation with the right neighbor.
2928 */
2929 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2930 temp = new->br_blockcount + right.br_blockcount;
2931 oldlen = startblockval(new->br_startblock) +
2932 startblockval(right.br_startblock);
2933 newlen = xfs_bmap_worst_indlen(ip, temp);
2934 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2935 new->br_startoff,
2936 nullstartblock((int)newlen), temp, right.br_state);
2937 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2938 break;
2939
2940 case 0:
2941 /*
2942 * New allocation is not contiguous with another
2943 * delayed allocation.
2944 * Insert a new entry.
2945 */
2946 oldlen = newlen = 0;
2947 xfs_iext_insert(ip, *idx, 1, new, state);
2948 break;
2949 }
2950 if (oldlen != newlen) {
2951 ASSERT(oldlen > newlen);
2952 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
2953 (int64_t)(oldlen - newlen), 0);
2954 /*
2955 * Nothing to do for disk quota accounting here.
2956 */
2957 }
2958}
2959
2960/*
2961 * Convert a hole to a real allocation.
2962 */
2963STATIC int /* error */
2964xfs_bmap_add_extent_hole_real(
2965 struct xfs_bmalloca *bma,
2966 int whichfork)
2967{
2968 struct xfs_bmbt_irec *new = &bma->got;
2969 int error; /* error return value */
2970 int i; /* temp state */
2971 xfs_ifork_t *ifp; /* inode fork pointer */
2972 xfs_bmbt_irec_t left; /* left neighbor extent entry */
2973 xfs_bmbt_irec_t right; /* right neighbor extent entry */
2974 int rval=0; /* return value (logging flags) */
2975 int state; /* state bits, accessed thru macros */
2976
2977 ifp = XFS_IFORK_PTR(bma->ip, whichfork);
2978
2979 ASSERT(bma->idx >= 0);
2980 ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2981 ASSERT(!isnullstartblock(new->br_startblock));
2982 ASSERT(!bma->cur ||
2983 !(bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
2984
2985 XFS_STATS_INC(xs_add_exlist);
2986
2987 state = 0;
2988 if (whichfork == XFS_ATTR_FORK)
2989 state |= BMAP_ATTRFORK;
2990
2991 /*
2992 * Check and set flags if this segment has a left neighbor.
2993 */
2994 if (bma->idx > 0) {
2995 state |= BMAP_LEFT_VALID;
2996 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &left);
2997 if (isnullstartblock(left.br_startblock))
2998 state |= BMAP_LEFT_DELAY;
2999 }
3000
3001 /*
3002 * Check and set flags if this segment has a current value.
3003 * Not true if we're inserting into the "hole" at eof.
3004 */
3005 if (bma->idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
3006 state |= BMAP_RIGHT_VALID;
3007 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &right);
3008 if (isnullstartblock(right.br_startblock))
3009 state |= BMAP_RIGHT_DELAY;
3010 }
3011
3012 /*
3013 * We're inserting a real allocation between "left" and "right".
3014 * Set the contiguity flags. Don't let extents get too large.
3015 */
3016 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
3017 left.br_startoff + left.br_blockcount == new->br_startoff &&
3018 left.br_startblock + left.br_blockcount == new->br_startblock &&
3019 left.br_state == new->br_state &&
3020 left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
3021 state |= BMAP_LEFT_CONTIG;
3022
3023 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
3024 new->br_startoff + new->br_blockcount == right.br_startoff &&
3025 new->br_startblock + new->br_blockcount == right.br_startblock &&
3026 new->br_state == right.br_state &&
3027 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
3028 (!(state & BMAP_LEFT_CONTIG) ||
3029 left.br_blockcount + new->br_blockcount +
3030 right.br_blockcount <= MAXEXTLEN))
3031 state |= BMAP_RIGHT_CONTIG;
3032
3033 error = 0;
3034 /*
3035 * Select which case we're in here, and implement it.
3036 */
3037 switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
3038 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
3039 /*
3040 * New allocation is contiguous with real allocations on the
3041 * left and on the right.
3042 * Merge all three into a single extent record.
3043 */
3044 --bma->idx;
3045 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3046 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3047 left.br_blockcount + new->br_blockcount +
3048 right.br_blockcount);
3049 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3050
3051 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
3052
3053 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3054 XFS_IFORK_NEXTENTS(bma->ip, whichfork) - 1);
3055 if (bma->cur == NULL) {
3056 rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3057 } else {
3058 rval = XFS_ILOG_CORE;
3059 error = xfs_bmbt_lookup_eq(bma->cur, right.br_startoff,
3060 right.br_startblock, right.br_blockcount,
3061 &i);
3062 if (error)
3063 goto done;
3064 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3065 error = xfs_btree_delete(bma->cur, &i);
3066 if (error)
3067 goto done;
3068 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3069 error = xfs_btree_decrement(bma->cur, 0, &i);
3070 if (error)
3071 goto done;
3072 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3073 error = xfs_bmbt_update(bma->cur, left.br_startoff,
3074 left.br_startblock,
3075 left.br_blockcount +
3076 new->br_blockcount +
3077 right.br_blockcount,
3078 left.br_state);
3079 if (error)
3080 goto done;
3081 }
3082 break;
3083
3084 case BMAP_LEFT_CONTIG:
3085 /*
3086 * New allocation is contiguous with a real allocation
3087 * on the left.
3088 * Merge the new allocation with the left neighbor.
3089 */
3090 --bma->idx;
3091 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3092 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3093 left.br_blockcount + new->br_blockcount);
3094 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3095
3096 if (bma->cur == NULL) {
3097 rval = xfs_ilog_fext(whichfork);
3098 } else {
3099 rval = 0;
3100 error = xfs_bmbt_lookup_eq(bma->cur, left.br_startoff,
3101 left.br_startblock, left.br_blockcount,
3102 &i);
3103 if (error)
3104 goto done;
3105 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3106 error = xfs_bmbt_update(bma->cur, left.br_startoff,
3107 left.br_startblock,
3108 left.br_blockcount +
3109 new->br_blockcount,
3110 left.br_state);
3111 if (error)
3112 goto done;
3113 }
3114 break;
3115
3116 case BMAP_RIGHT_CONTIG:
3117 /*
3118 * New allocation is contiguous with a real allocation
3119 * on the right.
3120 * Merge the new allocation with the right neighbor.
3121 */
3122 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3123 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx),
3124 new->br_startoff, new->br_startblock,
3125 new->br_blockcount + right.br_blockcount,
3126 right.br_state);
3127 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3128
3129 if (bma->cur == NULL) {
3130 rval = xfs_ilog_fext(whichfork);
3131 } else {
3132 rval = 0;
3133 error = xfs_bmbt_lookup_eq(bma->cur,
3134 right.br_startoff,
3135 right.br_startblock,
3136 right.br_blockcount, &i);
3137 if (error)
3138 goto done;
3139 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3140 error = xfs_bmbt_update(bma->cur, new->br_startoff,
3141 new->br_startblock,
3142 new->br_blockcount +
3143 right.br_blockcount,
3144 right.br_state);
3145 if (error)
3146 goto done;
3147 }
3148 break;
3149
3150 case 0:
3151 /*
3152 * New allocation is not contiguous with another
3153 * real allocation.
3154 * Insert a new entry.
3155 */
3156 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
3157 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3158 XFS_IFORK_NEXTENTS(bma->ip, whichfork) + 1);
3159 if (bma->cur == NULL) {
3160 rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3161 } else {
3162 rval = XFS_ILOG_CORE;
3163 error = xfs_bmbt_lookup_eq(bma->cur,
3164 new->br_startoff,
3165 new->br_startblock,
3166 new->br_blockcount, &i);
3167 if (error)
3168 goto done;
3169 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
3170 bma->cur->bc_rec.b.br_state = new->br_state;
3171 error = xfs_btree_insert(bma->cur, &i);
3172 if (error)
3173 goto done;
3174 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3175 }
3176 break;
3177 }
3178
3179 /* convert to a btree if necessary */
3180 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
3181 int tmp_logflags; /* partial log flag return val */
3182
3183 ASSERT(bma->cur == NULL);
3184 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
3185 bma->firstblock, bma->flist, &bma->cur,
3186 0, &tmp_logflags, whichfork);
3187 bma->logflags |= tmp_logflags;
3188 if (error)
3189 goto done;
3190 }
3191
3192 /* clear out the allocated field, done with it now in any case. */
3193 if (bma->cur)
3194 bma->cur->bc_private.b.allocated = 0;
3195
3196 xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
3197done:
3198 bma->logflags |= rval;
3199 return error;
3200}
3201
3202/*
3203 * Functions used in the extent read, allocate and remove paths
3204 */
3205
3206/*
3207 * Adjust the size of the new extent based on di_extsize and rt extsize.
3208 */
3209int
3210xfs_bmap_extsize_align(
3211 xfs_mount_t *mp,
3212 xfs_bmbt_irec_t *gotp, /* next extent pointer */
3213 xfs_bmbt_irec_t *prevp, /* previous extent pointer */
3214 xfs_extlen_t extsz, /* align to this extent size */
3215 int rt, /* is this a realtime inode? */
3216 int eof, /* is extent at end-of-file? */
3217 int delay, /* creating delalloc extent? */
3218 int convert, /* overwriting unwritten extent? */
3219 xfs_fileoff_t *offp, /* in/out: aligned offset */
3220 xfs_extlen_t *lenp) /* in/out: aligned length */
3221{
3222 xfs_fileoff_t orig_off; /* original offset */
3223 xfs_extlen_t orig_alen; /* original length */
3224 xfs_fileoff_t orig_end; /* original off+len */
3225 xfs_fileoff_t nexto; /* next file offset */
3226 xfs_fileoff_t prevo; /* previous file offset */
3227 xfs_fileoff_t align_off; /* temp for offset */
3228 xfs_extlen_t align_alen; /* temp for length */
3229 xfs_extlen_t temp; /* temp for calculations */
3230
3231 if (convert)
3232 return 0;
3233
3234 orig_off = align_off = *offp;
3235 orig_alen = align_alen = *lenp;
3236 orig_end = orig_off + orig_alen;
3237
3238 /*
3239 * If this request overlaps an existing extent, then don't
3240 * attempt to perform any additional alignment.
3241 */
3242 if (!delay && !eof &&
3243 (orig_off >= gotp->br_startoff) &&
3244 (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
3245 return 0;
3246 }
3247
3248 /*
3249 * If the file offset is unaligned vs. the extent size
3250 * we need to align it. This will be possible unless
3251 * the file was previously written with a kernel that didn't
3252 * perform this alignment, or if a truncate shot us in the
3253 * foot.
3254 */
3255 temp = do_mod(orig_off, extsz);
3256 if (temp) {
3257 align_alen += temp;
3258 align_off -= temp;
3259 }
3260 /*
3261 * Same adjustment for the end of the requested area.
3262 */
3263 if ((temp = (align_alen % extsz))) {
3264 align_alen += extsz - temp;
3265 }
3266 /*
3267 * If the previous block overlaps with this proposed allocation
3268 * then move the start forward without adjusting the length.
3269 */
3270 if (prevp->br_startoff != NULLFILEOFF) {
3271 if (prevp->br_startblock == HOLESTARTBLOCK)
3272 prevo = prevp->br_startoff;
3273 else
3274 prevo = prevp->br_startoff + prevp->br_blockcount;
3275 } else
3276 prevo = 0;
3277 if (align_off != orig_off && align_off < prevo)
3278 align_off = prevo;
3279 /*
3280 * If the next block overlaps with this proposed allocation
3281 * then move the start back without adjusting the length,
3282 * but not before offset 0.
3283 * This may of course make the start overlap previous block,
3284 * and if we hit the offset 0 limit then the next block
3285 * can still overlap too.
3286 */
3287 if (!eof && gotp->br_startoff != NULLFILEOFF) {
3288 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
3289 (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
3290 nexto = gotp->br_startoff + gotp->br_blockcount;
3291 else
3292 nexto = gotp->br_startoff;
3293 } else
3294 nexto = NULLFILEOFF;
3295 if (!eof &&
3296 align_off + align_alen != orig_end &&
3297 align_off + align_alen > nexto)
3298 align_off = nexto > align_alen ? nexto - align_alen : 0;
3299 /*
3300 * If we're now overlapping the next or previous extent that
3301 * means we can't fit an extsz piece in this hole. Just move
3302 * the start forward to the first valid spot and set
3303 * the length so we hit the end.
3304 */
3305 if (align_off != orig_off && align_off < prevo)
3306 align_off = prevo;
3307 if (align_off + align_alen != orig_end &&
3308 align_off + align_alen > nexto &&
3309 nexto != NULLFILEOFF) {
3310 ASSERT(nexto > prevo);
3311 align_alen = nexto - align_off;
3312 }
3313
3314 /*
3315 * If realtime, and the result isn't a multiple of the realtime
3316 * extent size we need to remove blocks until it is.
3317 */
3318 if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
3319 /*
3320 * We're not covering the original request, or
3321 * we won't be able to once we fix the length.
3322 */
3323 if (orig_off < align_off ||
3324 orig_end > align_off + align_alen ||
3325 align_alen - temp < orig_alen)
3326 return -EINVAL;
3327 /*
3328 * Try to fix it by moving the start up.
3329 */
3330 if (align_off + temp <= orig_off) {
3331 align_alen -= temp;
3332 align_off += temp;
3333 }
3334 /*
3335 * Try to fix it by moving the end in.
3336 */
3337 else if (align_off + align_alen - temp >= orig_end)
3338 align_alen -= temp;
3339 /*
3340 * Set the start to the minimum then trim the length.
3341 */
3342 else {
3343 align_alen -= orig_off - align_off;
3344 align_off = orig_off;
3345 align_alen -= align_alen % mp->m_sb.sb_rextsize;
3346 }
3347 /*
3348 * Result doesn't cover the request, fail it.
3349 */
3350 if (orig_off < align_off || orig_end > align_off + align_alen)
3351 return -EINVAL;
3352 } else {
3353 ASSERT(orig_off >= align_off);
3354 ASSERT(orig_end <= align_off + align_alen);
3355 }
3356
3357#ifdef DEBUG
3358 if (!eof && gotp->br_startoff != NULLFILEOFF)
3359 ASSERT(align_off + align_alen <= gotp->br_startoff);
3360 if (prevp->br_startoff != NULLFILEOFF)
3361 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3362#endif
3363
3364 *lenp = align_alen;
3365 *offp = align_off;
3366 return 0;
3367}
3368
3369#define XFS_ALLOC_GAP_UNITS 4
3370
3371void
3372xfs_bmap_adjacent(
3373 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
3374{
3375 xfs_fsblock_t adjust; /* adjustment to block numbers */
3376 xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
3377 xfs_mount_t *mp; /* mount point structure */
3378 int nullfb; /* true if ap->firstblock isn't set */
3379 int rt; /* true if inode is realtime */
3380
3381#define ISVALID(x,y) \
3382 (rt ? \
3383 (x) < mp->m_sb.sb_rblocks : \
3384 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3385 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3386 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3387
3388 mp = ap->ip->i_mount;
3389 nullfb = *ap->firstblock == NULLFSBLOCK;
3390 rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
3391 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3392 /*
3393 * If allocating at eof, and there's a previous real block,
3394 * try to use its last block as our starting point.
3395 */
3396 if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3397 !isnullstartblock(ap->prev.br_startblock) &&
3398 ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3399 ap->prev.br_startblock)) {
3400 ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3401 /*
3402 * Adjust for the gap between prevp and us.
3403 */
3404 adjust = ap->offset -
3405 (ap->prev.br_startoff + ap->prev.br_blockcount);
3406 if (adjust &&
3407 ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3408 ap->blkno += adjust;
3409 }
3410 /*
3411 * If not at eof, then compare the two neighbor blocks.
3412 * Figure out whether either one gives us a good starting point,
3413 * and pick the better one.
3414 */
3415 else if (!ap->eof) {
3416 xfs_fsblock_t gotbno; /* right side block number */
3417 xfs_fsblock_t gotdiff=0; /* right side difference */
3418 xfs_fsblock_t prevbno; /* left side block number */
3419 xfs_fsblock_t prevdiff=0; /* left side difference */
3420
3421 /*
3422 * If there's a previous (left) block, select a requested
3423 * start block based on it.
3424 */
3425 if (ap->prev.br_startoff != NULLFILEOFF &&
3426 !isnullstartblock(ap->prev.br_startblock) &&
3427 (prevbno = ap->prev.br_startblock +
3428 ap->prev.br_blockcount) &&
3429 ISVALID(prevbno, ap->prev.br_startblock)) {
3430 /*
3431 * Calculate gap to end of previous block.
3432 */
3433 adjust = prevdiff = ap->offset -
3434 (ap->prev.br_startoff +
3435 ap->prev.br_blockcount);
3436 /*
3437 * Figure the startblock based on the previous block's
3438 * end and the gap size.
3439 * Heuristic!
3440 * If the gap is large relative to the piece we're
3441 * allocating, or using it gives us an invalid block
3442 * number, then just use the end of the previous block.
3443 */
3444 if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3445 ISVALID(prevbno + prevdiff,
3446 ap->prev.br_startblock))
3447 prevbno += adjust;
3448 else
3449 prevdiff += adjust;
3450 /*
3451 * If the firstblock forbids it, can't use it,
3452 * must use default.
3453 */
3454 if (!rt && !nullfb &&
3455 XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3456 prevbno = NULLFSBLOCK;
3457 }
3458 /*
3459 * No previous block or can't follow it, just default.
3460 */
3461 else
3462 prevbno = NULLFSBLOCK;
3463 /*
3464 * If there's a following (right) block, select a requested
3465 * start block based on it.
3466 */
3467 if (!isnullstartblock(ap->got.br_startblock)) {
3468 /*
3469 * Calculate gap to start of next block.
3470 */
3471 adjust = gotdiff = ap->got.br_startoff - ap->offset;
3472 /*
3473 * Figure the startblock based on the next block's
3474 * start and the gap size.
3475 */
3476 gotbno = ap->got.br_startblock;
3477 /*
3478 * Heuristic!
3479 * If the gap is large relative to the piece we're
3480 * allocating, or using it gives us an invalid block
3481 * number, then just use the start of the next block
3482 * offset by our length.
3483 */
3484 if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3485 ISVALID(gotbno - gotdiff, gotbno))
3486 gotbno -= adjust;
3487 else if (ISVALID(gotbno - ap->length, gotbno)) {
3488 gotbno -= ap->length;
3489 gotdiff += adjust - ap->length;
3490 } else
3491 gotdiff += adjust;
3492 /*
3493 * If the firstblock forbids it, can't use it,
3494 * must use default.
3495 */
3496 if (!rt && !nullfb &&
3497 XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3498 gotbno = NULLFSBLOCK;
3499 }
3500 /*
3501 * No next block, just default.
3502 */
3503 else
3504 gotbno = NULLFSBLOCK;
3505 /*
3506 * If both valid, pick the better one, else the only good
3507 * one, else ap->blkno is already set (to 0 or the inode block).
3508 */
3509 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3510 ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3511 else if (prevbno != NULLFSBLOCK)
3512 ap->blkno = prevbno;
3513 else if (gotbno != NULLFSBLOCK)
3514 ap->blkno = gotbno;
3515 }
3516#undef ISVALID
3517}
3518
3519static int
3520xfs_bmap_longest_free_extent(
3521 struct xfs_trans *tp,
3522 xfs_agnumber_t ag,
3523 xfs_extlen_t *blen,
3524 int *notinit)
3525{
3526 struct xfs_mount *mp = tp->t_mountp;
3527 struct xfs_perag *pag;
3528 xfs_extlen_t longest;
3529 int error = 0;
3530
3531 pag = xfs_perag_get(mp, ag);
3532 if (!pag->pagf_init) {
3533 error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK);
3534 if (error)
3535 goto out;
3536
3537 if (!pag->pagf_init) {
3538 *notinit = 1;
3539 goto out;
3540 }
3541 }
3542
3543 longest = xfs_alloc_longest_free_extent(mp, pag);
3544 if (*blen < longest)
3545 *blen = longest;
3546
3547out:
3548 xfs_perag_put(pag);
3549 return error;
3550}
3551
3552static void
3553xfs_bmap_select_minlen(
3554 struct xfs_bmalloca *ap,
3555 struct xfs_alloc_arg *args,
3556 xfs_extlen_t *blen,
3557 int notinit)
3558{
3559 if (notinit || *blen < ap->minlen) {
3560 /*
3561 * Since we did a BUF_TRYLOCK above, it is possible that
3562 * there is space for this request.
3563 */
3564 args->minlen = ap->minlen;
3565 } else if (*blen < args->maxlen) {
3566 /*
3567 * If the best seen length is less than the request length,
3568 * use the best as the minimum.
3569 */
3570 args->minlen = *blen;
3571 } else {
3572 /*
3573 * Otherwise we've seen an extent as big as maxlen, use that
3574 * as the minimum.
3575 */
3576 args->minlen = args->maxlen;
3577 }
3578}
3579
3580STATIC int
3581xfs_bmap_btalloc_nullfb(
3582 struct xfs_bmalloca *ap,
3583 struct xfs_alloc_arg *args,
3584 xfs_extlen_t *blen)
3585{
3586 struct xfs_mount *mp = ap->ip->i_mount;
3587 xfs_agnumber_t ag, startag;
3588 int notinit = 0;
3589 int error;
3590
3591 args->type = XFS_ALLOCTYPE_START_BNO;
3592 args->total = ap->total;
3593
3594 startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3595 if (startag == NULLAGNUMBER)
3596 startag = ag = 0;
3597
3598 while (*blen < args->maxlen) {
3599 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3600 &notinit);
3601 if (error)
3602 return error;
3603
3604 if (++ag == mp->m_sb.sb_agcount)
3605 ag = 0;
3606 if (ag == startag)
3607 break;
3608 }
3609
3610 xfs_bmap_select_minlen(ap, args, blen, notinit);
3611 return 0;
3612}
3613
3614STATIC int
3615xfs_bmap_btalloc_filestreams(
3616 struct xfs_bmalloca *ap,
3617 struct xfs_alloc_arg *args,
3618 xfs_extlen_t *blen)
3619{
3620 struct xfs_mount *mp = ap->ip->i_mount;
3621 xfs_agnumber_t ag;
3622 int notinit = 0;
3623 int error;
3624
3625 args->type = XFS_ALLOCTYPE_NEAR_BNO;
3626 args->total = ap->total;
3627
3628 ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3629 if (ag == NULLAGNUMBER)
3630 ag = 0;
3631
3632 error = xfs_bmap_longest_free_extent(args->tp, ag, blen, &notinit);
3633 if (error)
3634 return error;
3635
3636 if (*blen < args->maxlen) {
3637 error = xfs_filestream_new_ag(ap, &ag);
3638 if (error)
3639 return error;
3640
3641 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3642 &notinit);
3643 if (error)
3644 return error;
3645
3646 }
3647
3648 xfs_bmap_select_minlen(ap, args, blen, notinit);
3649
3650 /*
3651 * Set the failure fallback case to look in the selected AG as stream
3652 * may have moved.
3653 */
3654 ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
3655 return 0;
3656}
3657
3658STATIC int
3659xfs_bmap_btalloc(
3660 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
3661{
3662 xfs_mount_t *mp; /* mount point structure */
3663 xfs_alloctype_t atype = 0; /* type for allocation routines */
3664 xfs_extlen_t align; /* minimum allocation alignment */
3665 xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
3666 xfs_agnumber_t ag;
3667 xfs_alloc_arg_t args;
3668 xfs_extlen_t blen;
3669 xfs_extlen_t nextminlen = 0;
3670 int nullfb; /* true if ap->firstblock isn't set */
3671 int isaligned;
3672 int tryagain;
3673 int error;
3674 int stripe_align;
3675
3676 ASSERT(ap->length);
3677
3678 mp = ap->ip->i_mount;
3679
3680 /* stripe alignment for allocation is determined by mount parameters */
3681 stripe_align = 0;
3682 if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
3683 stripe_align = mp->m_swidth;
3684 else if (mp->m_dalign)
3685 stripe_align = mp->m_dalign;
3686
3687 align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
3688 if (unlikely(align)) {
3689 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
3690 align, 0, ap->eof, 0, ap->conv,
3691 &ap->offset, &ap->length);
3692 ASSERT(!error);
3693 ASSERT(ap->length);
3694 }
3695
3696
3697 nullfb = *ap->firstblock == NULLFSBLOCK;
3698 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3699 if (nullfb) {
3700 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
3701 ag = xfs_filestream_lookup_ag(ap->ip);
3702 ag = (ag != NULLAGNUMBER) ? ag : 0;
3703 ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3704 } else {
3705 ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3706 }
3707 } else
3708 ap->blkno = *ap->firstblock;
3709
3710 xfs_bmap_adjacent(ap);
3711
3712 /*
3713 * If allowed, use ap->blkno; otherwise must use firstblock since
3714 * it's in the right allocation group.
3715 */
3716 if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3717 ;
3718 else
3719 ap->blkno = *ap->firstblock;
3720 /*
3721 * Normal allocation, done through xfs_alloc_vextent.
3722 */
3723 tryagain = isaligned = 0;
3724 memset(&args, 0, sizeof(args));
3725 args.tp = ap->tp;
3726 args.mp = mp;
3727 args.fsbno = ap->blkno;
3728
3729 /* Trim the allocation back to the maximum an AG can fit. */
3730 args.maxlen = MIN(ap->length, XFS_ALLOC_AG_MAX_USABLE(mp));
3731 args.firstblock = *ap->firstblock;
3732 blen = 0;
3733 if (nullfb) {
3734 /*
3735 * Search for an allocation group with a single extent large
3736 * enough for the request. If one isn't found, then adjust
3737 * the minimum allocation size to the largest space found.
3738 */
3739 if (ap->userdata && xfs_inode_is_filestream(ap->ip))
3740 error = xfs_bmap_btalloc_filestreams(ap, &args, &blen);
3741 else
3742 error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
3743 if (error)
3744 return error;
3745 } else if (ap->flist->xbf_low) {
3746 if (xfs_inode_is_filestream(ap->ip))
3747 args.type = XFS_ALLOCTYPE_FIRST_AG;
3748 else
3749 args.type = XFS_ALLOCTYPE_START_BNO;
3750 args.total = args.minlen = ap->minlen;
3751 } else {
3752 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3753 args.total = ap->total;
3754 args.minlen = ap->minlen;
3755 }
3756 /* apply extent size hints if obtained earlier */
3757 if (unlikely(align)) {
3758 args.prod = align;
3759 if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
3760 args.mod = (xfs_extlen_t)(args.prod - args.mod);
3761 } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
3762 args.prod = 1;
3763 args.mod = 0;
3764 } else {
3765 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
3766 if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
3767 args.mod = (xfs_extlen_t)(args.prod - args.mod);
3768 }
3769 /*
3770 * If we are not low on available data blocks, and the
3771 * underlying logical volume manager is a stripe, and
3772 * the file offset is zero then try to allocate data
3773 * blocks on stripe unit boundary.
3774 * NOTE: ap->aeof is only set if the allocation length
3775 * is >= the stripe unit and the allocation offset is
3776 * at the end of file.
3777 */
3778 if (!ap->flist->xbf_low && ap->aeof) {
3779 if (!ap->offset) {
3780 args.alignment = stripe_align;
3781 atype = args.type;
3782 isaligned = 1;
3783 /*
3784 * Adjust for alignment
3785 */
3786 if (blen > args.alignment && blen <= args.maxlen)
3787 args.minlen = blen - args.alignment;
3788 args.minalignslop = 0;
3789 } else {
3790 /*
3791 * First try an exact bno allocation.
3792 * If it fails then do a near or start bno
3793 * allocation with alignment turned on.
3794 */
3795 atype = args.type;
3796 tryagain = 1;
3797 args.type = XFS_ALLOCTYPE_THIS_BNO;
3798 args.alignment = 1;
3799 /*
3800 * Compute the minlen+alignment for the
3801 * next case. Set slop so that the value
3802 * of minlen+alignment+slop doesn't go up
3803 * between the calls.
3804 */
3805 if (blen > stripe_align && blen <= args.maxlen)
3806 nextminlen = blen - stripe_align;
3807 else
3808 nextminlen = args.minlen;
3809 if (nextminlen + stripe_align > args.minlen + 1)
3810 args.minalignslop =
3811 nextminlen + stripe_align -
3812 args.minlen - 1;
3813 else
3814 args.minalignslop = 0;
3815 }
3816 } else {
3817 args.alignment = 1;
3818 args.minalignslop = 0;
3819 }
3820 args.minleft = ap->minleft;
3821 args.wasdel = ap->wasdel;
3822 args.isfl = 0;
3823 args.userdata = ap->userdata;
3824 if ((error = xfs_alloc_vextent(&args)))
3825 return error;
3826 if (tryagain && args.fsbno == NULLFSBLOCK) {
3827 /*
3828 * Exact allocation failed. Now try with alignment
3829 * turned on.
3830 */
3831 args.type = atype;
3832 args.fsbno = ap->blkno;
3833 args.alignment = stripe_align;
3834 args.minlen = nextminlen;
3835 args.minalignslop = 0;
3836 isaligned = 1;
3837 if ((error = xfs_alloc_vextent(&args)))
3838 return error;
3839 }
3840 if (isaligned && args.fsbno == NULLFSBLOCK) {
3841 /*
3842 * allocation failed, so turn off alignment and
3843 * try again.
3844 */
3845 args.type = atype;
3846 args.fsbno = ap->blkno;
3847 args.alignment = 0;
3848 if ((error = xfs_alloc_vextent(&args)))
3849 return error;
3850 }
3851 if (args.fsbno == NULLFSBLOCK && nullfb &&
3852 args.minlen > ap->minlen) {
3853 args.minlen = ap->minlen;
3854 args.type = XFS_ALLOCTYPE_START_BNO;
3855 args.fsbno = ap->blkno;
3856 if ((error = xfs_alloc_vextent(&args)))
3857 return error;
3858 }
3859 if (args.fsbno == NULLFSBLOCK && nullfb) {
3860 args.fsbno = 0;
3861 args.type = XFS_ALLOCTYPE_FIRST_AG;
3862 args.total = ap->minlen;
3863 args.minleft = 0;
3864 if ((error = xfs_alloc_vextent(&args)))
3865 return error;
3866 ap->flist->xbf_low = 1;
3867 }
3868 if (args.fsbno != NULLFSBLOCK) {
3869 /*
3870 * check the allocation happened at the same or higher AG than
3871 * the first block that was allocated.
3872 */
3873 ASSERT(*ap->firstblock == NULLFSBLOCK ||
3874 XFS_FSB_TO_AGNO(mp, *ap->firstblock) ==
3875 XFS_FSB_TO_AGNO(mp, args.fsbno) ||
3876 (ap->flist->xbf_low &&
3877 XFS_FSB_TO_AGNO(mp, *ap->firstblock) <
3878 XFS_FSB_TO_AGNO(mp, args.fsbno)));
3879
3880 ap->blkno = args.fsbno;
3881 if (*ap->firstblock == NULLFSBLOCK)
3882 *ap->firstblock = args.fsbno;
3883 ASSERT(nullfb || fb_agno == args.agno ||
3884 (ap->flist->xbf_low && fb_agno < args.agno));
3885 ap->length = args.len;
3886 ap->ip->i_d.di_nblocks += args.len;
3887 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3888 if (ap->wasdel)
3889 ap->ip->i_delayed_blks -= args.len;
3890 /*
3891 * Adjust the disk quota also. This was reserved
3892 * earlier.
3893 */
3894 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3895 ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
3896 XFS_TRANS_DQ_BCOUNT,
3897 (long) args.len);
3898 } else {
3899 ap->blkno = NULLFSBLOCK;
3900 ap->length = 0;
3901 }
3902 return 0;
3903}
3904
3905/*
3906 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
3907 * It figures out where to ask the underlying allocator to put the new extent.
3908 */
3909STATIC int
3910xfs_bmap_alloc(
3911 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
3912{
3913 if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
3914 return xfs_bmap_rtalloc(ap);
3915 return xfs_bmap_btalloc(ap);
3916}
3917
3918/*
3919 * Trim the returned map to the required bounds
3920 */
3921STATIC void
3922xfs_bmapi_trim_map(
3923 struct xfs_bmbt_irec *mval,
3924 struct xfs_bmbt_irec *got,
3925 xfs_fileoff_t *bno,
3926 xfs_filblks_t len,
3927 xfs_fileoff_t obno,
3928 xfs_fileoff_t end,
3929 int n,
3930 int flags)
3931{
3932 if ((flags & XFS_BMAPI_ENTIRE) ||
3933 got->br_startoff + got->br_blockcount <= obno) {
3934 *mval = *got;
3935 if (isnullstartblock(got->br_startblock))
3936 mval->br_startblock = DELAYSTARTBLOCK;
3937 return;
3938 }
3939
3940 if (obno > *bno)
3941 *bno = obno;
3942 ASSERT((*bno >= obno) || (n == 0));
3943 ASSERT(*bno < end);
3944 mval->br_startoff = *bno;
3945 if (isnullstartblock(got->br_startblock))
3946 mval->br_startblock = DELAYSTARTBLOCK;
3947 else
3948 mval->br_startblock = got->br_startblock +
3949 (*bno - got->br_startoff);
3950 /*
3951 * Return the minimum of what we got and what we asked for for
3952 * the length. We can use the len variable here because it is
3953 * modified below and we could have been there before coming
3954 * here if the first part of the allocation didn't overlap what
3955 * was asked for.
3956 */
3957 mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3958 got->br_blockcount - (*bno - got->br_startoff));
3959 mval->br_state = got->br_state;
3960 ASSERT(mval->br_blockcount <= len);
3961 return;
3962}
3963
3964/*
3965 * Update and validate the extent map to return
3966 */
3967STATIC void
3968xfs_bmapi_update_map(
3969 struct xfs_bmbt_irec **map,
3970 xfs_fileoff_t *bno,
3971 xfs_filblks_t *len,
3972 xfs_fileoff_t obno,
3973 xfs_fileoff_t end,
3974 int *n,
3975 int flags)
3976{
3977 xfs_bmbt_irec_t *mval = *map;
3978
3979 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3980 ((mval->br_startoff + mval->br_blockcount) <= end));
3981 ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3982 (mval->br_startoff < obno));
3983
3984 *bno = mval->br_startoff + mval->br_blockcount;
3985 *len = end - *bno;
3986 if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3987 /* update previous map with new information */
3988 ASSERT(mval->br_startblock == mval[-1].br_startblock);
3989 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3990 ASSERT(mval->br_state == mval[-1].br_state);
3991 mval[-1].br_blockcount = mval->br_blockcount;
3992 mval[-1].br_state = mval->br_state;
3993 } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3994 mval[-1].br_startblock != DELAYSTARTBLOCK &&
3995 mval[-1].br_startblock != HOLESTARTBLOCK &&
3996 mval->br_startblock == mval[-1].br_startblock +
3997 mval[-1].br_blockcount &&
3998 ((flags & XFS_BMAPI_IGSTATE) ||
3999 mval[-1].br_state == mval->br_state)) {
4000 ASSERT(mval->br_startoff ==
4001 mval[-1].br_startoff + mval[-1].br_blockcount);
4002 mval[-1].br_blockcount += mval->br_blockcount;
4003 } else if (*n > 0 &&
4004 mval->br_startblock == DELAYSTARTBLOCK &&
4005 mval[-1].br_startblock == DELAYSTARTBLOCK &&
4006 mval->br_startoff ==
4007 mval[-1].br_startoff + mval[-1].br_blockcount) {
4008 mval[-1].br_blockcount += mval->br_blockcount;
4009 mval[-1].br_state = mval->br_state;
4010 } else if (!((*n == 0) &&
4011 ((mval->br_startoff + mval->br_blockcount) <=
4012 obno))) {
4013 mval++;
4014 (*n)++;
4015 }
4016 *map = mval;
4017}
4018
4019/*
4020 * Map file blocks to filesystem blocks without allocation.
4021 */
4022int
4023xfs_bmapi_read(
4024 struct xfs_inode *ip,
4025 xfs_fileoff_t bno,
4026 xfs_filblks_t len,
4027 struct xfs_bmbt_irec *mval,
4028 int *nmap,
4029 int flags)
4030{
4031 struct xfs_mount *mp = ip->i_mount;
4032 struct xfs_ifork *ifp;
4033 struct xfs_bmbt_irec got;
4034 struct xfs_bmbt_irec prev;
4035 xfs_fileoff_t obno;
4036 xfs_fileoff_t end;
4037 xfs_extnum_t lastx;
4038 int error;
4039 int eof;
4040 int n = 0;
4041 int whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4042 XFS_ATTR_FORK : XFS_DATA_FORK;
4043
4044 ASSERT(*nmap >= 1);
4045 ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
4046 XFS_BMAPI_IGSTATE)));
4047 ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL));
4048
4049 if (unlikely(XFS_TEST_ERROR(
4050 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4051 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4052 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4053 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
4054 return -EFSCORRUPTED;
4055 }
4056
4057 if (XFS_FORCED_SHUTDOWN(mp))
4058 return -EIO;
4059
4060 XFS_STATS_INC(xs_blk_mapr);
4061
4062 ifp = XFS_IFORK_PTR(ip, whichfork);
4063
4064 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4065 error = xfs_iread_extents(NULL, ip, whichfork);
4066 if (error)
4067 return error;
4068 }
4069
4070 xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got, &prev);
4071 end = bno + len;
4072 obno = bno;
4073
4074 while (bno < end && n < *nmap) {
4075 /* Reading past eof, act as though there's a hole up to end. */
4076 if (eof)
4077 got.br_startoff = end;
4078 if (got.br_startoff > bno) {
4079 /* Reading in a hole. */
4080 mval->br_startoff = bno;
4081 mval->br_startblock = HOLESTARTBLOCK;
4082 mval->br_blockcount =
4083 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4084 mval->br_state = XFS_EXT_NORM;
4085 bno += mval->br_blockcount;
4086 len -= mval->br_blockcount;
4087 mval++;
4088 n++;
4089 continue;
4090 }
4091
4092 /* set up the extent map to return. */
4093 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4094 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4095
4096 /* If we're done, stop now. */
4097 if (bno >= end || n >= *nmap)
4098 break;
4099
4100 /* Else go on to the next record. */
4101 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4102 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4103 else
4104 eof = 1;
4105 }
4106 *nmap = n;
4107 return 0;
4108}
4109
4110STATIC int
4111xfs_bmapi_reserve_delalloc(
4112 struct xfs_inode *ip,
4113 xfs_fileoff_t aoff,
4114 xfs_filblks_t len,
4115 struct xfs_bmbt_irec *got,
4116 struct xfs_bmbt_irec *prev,
4117 xfs_extnum_t *lastx,
4118 int eof)
4119{
4120 struct xfs_mount *mp = ip->i_mount;
4121 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4122 xfs_extlen_t alen;
4123 xfs_extlen_t indlen;
4124 char rt = XFS_IS_REALTIME_INODE(ip);
4125 xfs_extlen_t extsz;
4126 int error;
4127
4128 alen = XFS_FILBLKS_MIN(len, MAXEXTLEN);
4129 if (!eof)
4130 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4131
4132 /* Figure out the extent size, adjust alen */
4133 extsz = xfs_get_extsz_hint(ip);
4134 if (extsz) {
4135 /*
4136 * Make sure we don't exceed a single extent length when we
4137 * align the extent by reducing length we are going to
4138 * allocate by the maximum amount extent size aligment may
4139 * require.
4140 */
4141 alen = XFS_FILBLKS_MIN(len, MAXEXTLEN - (2 * extsz - 1));
4142 error = xfs_bmap_extsize_align(mp, got, prev, extsz, rt, eof,
4143 1, 0, &aoff, &alen);
4144 ASSERT(!error);
4145 }
4146
4147 if (rt)
4148 extsz = alen / mp->m_sb.sb_rextsize;
4149
4150 /*
4151 * Make a transaction-less quota reservation for delayed allocation
4152 * blocks. This number gets adjusted later. We return if we haven't
4153 * allocated blocks already inside this loop.
4154 */
4155 error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4156 rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4157 if (error)
4158 return error;
4159
4160 /*
4161 * Split changing sb for alen and indlen since they could be coming
4162 * from different places.
4163 */
4164 indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4165 ASSERT(indlen > 0);
4166
4167 if (rt) {
4168 error = xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
4169 -((int64_t)extsz), 0);
4170 } else {
4171 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4172 -((int64_t)alen), 0);
4173 }
4174
4175 if (error)
4176 goto out_unreserve_quota;
4177
4178 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4179 -((int64_t)indlen), 0);
4180 if (error)
4181 goto out_unreserve_blocks;
4182
4183
4184 ip->i_delayed_blks += alen;
4185
4186 got->br_startoff = aoff;
4187 got->br_startblock = nullstartblock(indlen);
4188 got->br_blockcount = alen;
4189 got->br_state = XFS_EXT_NORM;
4190 xfs_bmap_add_extent_hole_delay(ip, lastx, got);
4191
4192 /*
4193 * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4194 * might have merged it into one of the neighbouring ones.
4195 */
4196 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), got);
4197
4198 ASSERT(got->br_startoff <= aoff);
4199 ASSERT(got->br_startoff + got->br_blockcount >= aoff + alen);
4200 ASSERT(isnullstartblock(got->br_startblock));
4201 ASSERT(got->br_state == XFS_EXT_NORM);
4202 return 0;
4203
4204out_unreserve_blocks:
4205 if (rt)
4206 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS, extsz, 0);
4207 else
4208 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, alen, 0);
4209out_unreserve_quota:
4210 if (XFS_IS_QUOTA_ON(mp))
4211 xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ?
4212 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4213 return error;
4214}
4215
4216/*
4217 * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4218 */
4219int
4220xfs_bmapi_delay(
4221 struct xfs_inode *ip, /* incore inode */
4222 xfs_fileoff_t bno, /* starting file offs. mapped */
4223 xfs_filblks_t len, /* length to map in file */
4224 struct xfs_bmbt_irec *mval, /* output: map values */
4225 int *nmap, /* i/o: mval size/count */
4226 int flags) /* XFS_BMAPI_... */
4227{
4228 struct xfs_mount *mp = ip->i_mount;
4229 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4230 struct xfs_bmbt_irec got; /* current file extent record */
4231 struct xfs_bmbt_irec prev; /* previous file extent record */
4232 xfs_fileoff_t obno; /* old block number (offset) */
4233 xfs_fileoff_t end; /* end of mapped file region */
4234 xfs_extnum_t lastx; /* last useful extent number */
4235 int eof; /* we've hit the end of extents */
4236 int n = 0; /* current extent index */
4237 int error = 0;
4238
4239 ASSERT(*nmap >= 1);
4240 ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4241 ASSERT(!(flags & ~XFS_BMAPI_ENTIRE));
4242 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4243
4244 if (unlikely(XFS_TEST_ERROR(
4245 (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4246 XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4247 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4248 XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW, mp);
4249 return -EFSCORRUPTED;
4250 }
4251
4252 if (XFS_FORCED_SHUTDOWN(mp))
4253 return -EIO;
4254
4255 XFS_STATS_INC(xs_blk_mapw);
4256
4257 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4258 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4259 if (error)
4260 return error;
4261 }
4262
4263 xfs_bmap_search_extents(ip, bno, XFS_DATA_FORK, &eof, &lastx, &got, &prev);
4264 end = bno + len;
4265 obno = bno;
4266
4267 while (bno < end && n < *nmap) {
4268 if (eof || got.br_startoff > bno) {
4269 error = xfs_bmapi_reserve_delalloc(ip, bno, len, &got,
4270 &prev, &lastx, eof);
4271 if (error) {
4272 if (n == 0) {
4273 *nmap = 0;
4274 return error;
4275 }
4276 break;
4277 }
4278 }
4279
4280 /* set up the extent map to return. */
4281 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4282 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4283
4284 /* If we're done, stop now. */
4285 if (bno >= end || n >= *nmap)
4286 break;
4287
4288 /* Else go on to the next record. */
4289 prev = got;
4290 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4291 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4292 else
4293 eof = 1;
4294 }
4295
4296 *nmap = n;
4297 return 0;
4298}
4299
4300
4301static int
4302xfs_bmapi_allocate(
4303 struct xfs_bmalloca *bma)
4304{
4305 struct xfs_mount *mp = bma->ip->i_mount;
4306 int whichfork = (bma->flags & XFS_BMAPI_ATTRFORK) ?
4307 XFS_ATTR_FORK : XFS_DATA_FORK;
4308 struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4309 int tmp_logflags = 0;
4310 int error;
4311
4312 ASSERT(bma->length > 0);
4313
4314 /*
4315 * For the wasdelay case, we could also just allocate the stuff asked
4316 * for in this bmap call but that wouldn't be as good.
4317 */
4318 if (bma->wasdel) {
4319 bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4320 bma->offset = bma->got.br_startoff;
4321 if (bma->idx != NULLEXTNUM && bma->idx) {
4322 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4323 &bma->prev);
4324 }
4325 } else {
4326 bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4327 if (!bma->eof)
4328 bma->length = XFS_FILBLKS_MIN(bma->length,
4329 bma->got.br_startoff - bma->offset);
4330 }
4331
4332 /*
4333 * Indicate if this is the first user data in the file, or just any
4334 * user data.
4335 */
4336 if (!(bma->flags & XFS_BMAPI_METADATA)) {
4337 bma->userdata = (bma->offset == 0) ?
4338 XFS_ALLOC_INITIAL_USER_DATA : XFS_ALLOC_USERDATA;
4339 }
4340
4341 bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4342
4343 /*
4344 * Only want to do the alignment at the eof if it is userdata and
4345 * allocation length is larger than a stripe unit.
4346 */
4347 if (mp->m_dalign && bma->length >= mp->m_dalign &&
4348 !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4349 error = xfs_bmap_isaeof(bma, whichfork);
4350 if (error)
4351 return error;
4352 }
4353
4354 error = xfs_bmap_alloc(bma);
4355 if (error)
4356 return error;
4357
4358 if (bma->flist->xbf_low)
4359 bma->minleft = 0;
4360 if (bma->cur)
4361 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4362 if (bma->blkno == NULLFSBLOCK)
4363 return 0;
4364 if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4365 bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4366 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4367 bma->cur->bc_private.b.flist = bma->flist;
4368 }
4369 /*
4370 * Bump the number of extents we've allocated
4371 * in this call.
4372 */
4373 bma->nallocs++;
4374
4375 if (bma->cur)
4376 bma->cur->bc_private.b.flags =
4377 bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4378
4379 bma->got.br_startoff = bma->offset;
4380 bma->got.br_startblock = bma->blkno;
4381 bma->got.br_blockcount = bma->length;
4382 bma->got.br_state = XFS_EXT_NORM;
4383
4384 /*
4385 * A wasdelay extent has been initialized, so shouldn't be flagged
4386 * as unwritten.
4387 */
4388 if (!bma->wasdel && (bma->flags & XFS_BMAPI_PREALLOC) &&
4389 xfs_sb_version_hasextflgbit(&mp->m_sb))
4390 bma->got.br_state = XFS_EXT_UNWRITTEN;
4391
4392 if (bma->wasdel)
4393 error = xfs_bmap_add_extent_delay_real(bma);
4394 else
4395 error = xfs_bmap_add_extent_hole_real(bma, whichfork);
4396
4397 bma->logflags |= tmp_logflags;
4398 if (error)
4399 return error;
4400
4401 /*
4402 * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4403 * or xfs_bmap_add_extent_hole_real might have merged it into one of
4404 * the neighbouring ones.
4405 */
4406 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4407
4408 ASSERT(bma->got.br_startoff <= bma->offset);
4409 ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4410 bma->offset + bma->length);
4411 ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4412 bma->got.br_state == XFS_EXT_UNWRITTEN);
4413 return 0;
4414}
4415
4416STATIC int
4417xfs_bmapi_convert_unwritten(
4418 struct xfs_bmalloca *bma,
4419 struct xfs_bmbt_irec *mval,
4420 xfs_filblks_t len,
4421 int flags)
4422{
4423 int whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4424 XFS_ATTR_FORK : XFS_DATA_FORK;
4425 struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4426 int tmp_logflags = 0;
4427 int error;
4428
4429 /* check if we need to do unwritten->real conversion */
4430 if (mval->br_state == XFS_EXT_UNWRITTEN &&
4431 (flags & XFS_BMAPI_PREALLOC))
4432 return 0;
4433
4434 /* check if we need to do real->unwritten conversion */
4435 if (mval->br_state == XFS_EXT_NORM &&
4436 (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4437 (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4438 return 0;
4439
4440 /*
4441 * Modify (by adding) the state flag, if writing.
4442 */
4443 ASSERT(mval->br_blockcount <= len);
4444 if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4445 bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4446 bma->ip, whichfork);
4447 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4448 bma->cur->bc_private.b.flist = bma->flist;
4449 }
4450 mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4451 ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4452
4453 error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, &bma->idx,
4454 &bma->cur, mval, bma->firstblock, bma->flist,
4455 &tmp_logflags);
4456 bma->logflags |= tmp_logflags;
4457 if (error)
4458 return error;
4459
4460 /*
4461 * Update our extent pointer, given that
4462 * xfs_bmap_add_extent_unwritten_real might have merged it into one
4463 * of the neighbouring ones.
4464 */
4465 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4466
4467 /*
4468 * We may have combined previously unwritten space with written space,
4469 * so generate another request.
4470 */
4471 if (mval->br_blockcount < len)
4472 return -EAGAIN;
4473 return 0;
4474}
4475
4476/*
4477 * Map file blocks to filesystem blocks, and allocate blocks or convert the
4478 * extent state if necessary. Details behaviour is controlled by the flags
4479 * parameter. Only allocates blocks from a single allocation group, to avoid
4480 * locking problems.
4481 *
4482 * The returned value in "firstblock" from the first call in a transaction
4483 * must be remembered and presented to subsequent calls in "firstblock".
4484 * An upper bound for the number of blocks to be allocated is supplied to
4485 * the first call in "total"; if no allocation group has that many free
4486 * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4487 */
4488int
4489xfs_bmapi_write(
4490 struct xfs_trans *tp, /* transaction pointer */
4491 struct xfs_inode *ip, /* incore inode */
4492 xfs_fileoff_t bno, /* starting file offs. mapped */
4493 xfs_filblks_t len, /* length to map in file */
4494 int flags, /* XFS_BMAPI_... */
4495 xfs_fsblock_t *firstblock, /* first allocated block
4496 controls a.g. for allocs */
4497 xfs_extlen_t total, /* total blocks needed */
4498 struct xfs_bmbt_irec *mval, /* output: map values */
4499 int *nmap, /* i/o: mval size/count */
4500 struct xfs_bmap_free *flist) /* i/o: list extents to free */
4501{
4502 struct xfs_mount *mp = ip->i_mount;
4503 struct xfs_ifork *ifp;
4504 struct xfs_bmalloca bma = { NULL }; /* args for xfs_bmap_alloc */
4505 xfs_fileoff_t end; /* end of mapped file region */
4506 int eof; /* after the end of extents */
4507 int error; /* error return */
4508 int n; /* current extent index */
4509 xfs_fileoff_t obno; /* old block number (offset) */
4510 int whichfork; /* data or attr fork */
4511 char inhole; /* current location is hole in file */
4512 char wasdelay; /* old extent was delayed */
4513
4514#ifdef DEBUG
4515 xfs_fileoff_t orig_bno; /* original block number value */
4516 int orig_flags; /* original flags arg value */
4517 xfs_filblks_t orig_len; /* original value of len arg */
4518 struct xfs_bmbt_irec *orig_mval; /* original value of mval */
4519 int orig_nmap; /* original value of *nmap */
4520
4521 orig_bno = bno;
4522 orig_len = len;
4523 orig_flags = flags;
4524 orig_mval = mval;
4525 orig_nmap = *nmap;
4526#endif
4527 whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4528 XFS_ATTR_FORK : XFS_DATA_FORK;
4529
4530 ASSERT(*nmap >= 1);
4531 ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4532 ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4533 ASSERT(tp != NULL);
4534 ASSERT(len > 0);
4535 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL);
4536 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4537
4538 if (unlikely(XFS_TEST_ERROR(
4539 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4540 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4541 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4542 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4543 return -EFSCORRUPTED;
4544 }
4545
4546 if (XFS_FORCED_SHUTDOWN(mp))
4547 return -EIO;
4548
4549 ifp = XFS_IFORK_PTR(ip, whichfork);
4550
4551 XFS_STATS_INC(xs_blk_mapw);
4552
4553 if (*firstblock == NULLFSBLOCK) {
4554 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4555 bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4556 else
4557 bma.minleft = 1;
4558 } else {
4559 bma.minleft = 0;
4560 }
4561
4562 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4563 error = xfs_iread_extents(tp, ip, whichfork);
4564 if (error)
4565 goto error0;
4566 }
4567
4568 xfs_bmap_search_extents(ip, bno, whichfork, &eof, &bma.idx, &bma.got,
4569 &bma.prev);
4570 n = 0;
4571 end = bno + len;
4572 obno = bno;
4573
4574 bma.tp = tp;
4575 bma.ip = ip;
4576 bma.total = total;
4577 bma.userdata = 0;
4578 bma.flist = flist;
4579 bma.firstblock = firstblock;
4580
4581 while (bno < end && n < *nmap) {
4582 inhole = eof || bma.got.br_startoff > bno;
4583 wasdelay = !inhole && isnullstartblock(bma.got.br_startblock);
4584
4585 /*
4586 * First, deal with the hole before the allocated space
4587 * that we found, if any.
4588 */
4589 if (inhole || wasdelay) {
4590 bma.eof = eof;
4591 bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4592 bma.wasdel = wasdelay;
4593 bma.offset = bno;
4594 bma.flags = flags;
4595
4596 /*
4597 * There's a 32/64 bit type mismatch between the
4598 * allocation length request (which can be 64 bits in
4599 * length) and the bma length request, which is
4600 * xfs_extlen_t and therefore 32 bits. Hence we have to
4601 * check for 32-bit overflows and handle them here.
4602 */
4603 if (len > (xfs_filblks_t)MAXEXTLEN)
4604 bma.length = MAXEXTLEN;
4605 else
4606 bma.length = len;
4607
4608 ASSERT(len > 0);
4609 ASSERT(bma.length > 0);
4610 error = xfs_bmapi_allocate(&bma);
4611 if (error)
4612 goto error0;
4613 if (bma.blkno == NULLFSBLOCK)
4614 break;
4615 }
4616
4617 /* Deal with the allocated space we found. */
4618 xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4619 end, n, flags);
4620
4621 /* Execute unwritten extent conversion if necessary */
4622 error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4623 if (error == -EAGAIN)
4624 continue;
4625 if (error)
4626 goto error0;
4627
4628 /* update the extent map to return */
4629 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4630
4631 /*
4632 * If we're done, stop now. Stop when we've allocated
4633 * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise
4634 * the transaction may get too big.
4635 */
4636 if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4637 break;
4638
4639 /* Else go on to the next record. */
4640 bma.prev = bma.got;
4641 if (++bma.idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t)) {
4642 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma.idx),
4643 &bma.got);
4644 } else
4645 eof = 1;
4646 }
4647 *nmap = n;
4648
4649 /*
4650 * Transform from btree to extents, give it cur.
4651 */
4652 if (xfs_bmap_wants_extents(ip, whichfork)) {
4653 int tmp_logflags = 0;
4654
4655 ASSERT(bma.cur);
4656 error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
4657 &tmp_logflags, whichfork);
4658 bma.logflags |= tmp_logflags;
4659 if (error)
4660 goto error0;
4661 }
4662
4663 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4664 XFS_IFORK_NEXTENTS(ip, whichfork) >
4665 XFS_IFORK_MAXEXT(ip, whichfork));
4666 error = 0;
4667error0:
4668 /*
4669 * Log everything. Do this after conversion, there's no point in
4670 * logging the extent records if we've converted to btree format.
4671 */
4672 if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
4673 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4674 bma.logflags &= ~xfs_ilog_fext(whichfork);
4675 else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
4676 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4677 bma.logflags &= ~xfs_ilog_fbroot(whichfork);
4678 /*
4679 * Log whatever the flags say, even if error. Otherwise we might miss
4680 * detecting a case where the data is changed, there's an error,
4681 * and it's not logged so we don't shutdown when we should.
4682 */
4683 if (bma.logflags)
4684 xfs_trans_log_inode(tp, ip, bma.logflags);
4685
4686 if (bma.cur) {
4687 if (!error) {
4688 ASSERT(*firstblock == NULLFSBLOCK ||
4689 XFS_FSB_TO_AGNO(mp, *firstblock) ==
4690 XFS_FSB_TO_AGNO(mp,
4691 bma.cur->bc_private.b.firstblock) ||
4692 (flist->xbf_low &&
4693 XFS_FSB_TO_AGNO(mp, *firstblock) <
4694 XFS_FSB_TO_AGNO(mp,
4695 bma.cur->bc_private.b.firstblock)));
4696 *firstblock = bma.cur->bc_private.b.firstblock;
4697 }
4698 xfs_btree_del_cursor(bma.cur,
4699 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4700 }
4701 if (!error)
4702 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4703 orig_nmap, *nmap);
4704 return error;
4705}
4706
4707/*
4708 * Called by xfs_bmapi to update file extent records and the btree
4709 * after removing space (or undoing a delayed allocation).
4710 */
4711STATIC int /* error */
4712xfs_bmap_del_extent(
4713 xfs_inode_t *ip, /* incore inode pointer */
4714 xfs_trans_t *tp, /* current transaction pointer */
4715 xfs_extnum_t *idx, /* extent number to update/delete */
4716 xfs_bmap_free_t *flist, /* list of extents to be freed */
4717 xfs_btree_cur_t *cur, /* if null, not a btree */
4718 xfs_bmbt_irec_t *del, /* data to remove from extents */
4719 int *logflagsp, /* inode logging flags */
4720 int whichfork) /* data or attr fork */
4721{
4722 xfs_filblks_t da_new; /* new delay-alloc indirect blocks */
4723 xfs_filblks_t da_old; /* old delay-alloc indirect blocks */
4724 xfs_fsblock_t del_endblock=0; /* first block past del */
4725 xfs_fileoff_t del_endoff; /* first offset past del */
4726 int delay; /* current block is delayed allocated */
4727 int do_fx; /* free extent at end of routine */
4728 xfs_bmbt_rec_host_t *ep; /* current extent entry pointer */
4729 int error; /* error return value */
4730 int flags; /* inode logging flags */
4731 xfs_bmbt_irec_t got; /* current extent entry */
4732 xfs_fileoff_t got_endoff; /* first offset past got */
4733 int i; /* temp state */
4734 xfs_ifork_t *ifp; /* inode fork pointer */
4735 xfs_mount_t *mp; /* mount structure */
4736 xfs_filblks_t nblks; /* quota/sb block count */
4737 xfs_bmbt_irec_t new; /* new record to be inserted */
4738 /* REFERENCED */
4739 uint qfield; /* quota field to update */
4740 xfs_filblks_t temp; /* for indirect length calculations */
4741 xfs_filblks_t temp2; /* for indirect length calculations */
4742 int state = 0;
4743
4744 XFS_STATS_INC(xs_del_exlist);
4745
4746 if (whichfork == XFS_ATTR_FORK)
4747 state |= BMAP_ATTRFORK;
4748
4749 mp = ip->i_mount;
4750 ifp = XFS_IFORK_PTR(ip, whichfork);
4751 ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
4752 (uint)sizeof(xfs_bmbt_rec_t)));
4753 ASSERT(del->br_blockcount > 0);
4754 ep = xfs_iext_get_ext(ifp, *idx);
4755 xfs_bmbt_get_all(ep, &got);
4756 ASSERT(got.br_startoff <= del->br_startoff);
4757 del_endoff = del->br_startoff + del->br_blockcount;
4758 got_endoff = got.br_startoff + got.br_blockcount;
4759 ASSERT(got_endoff >= del_endoff);
4760 delay = isnullstartblock(got.br_startblock);
4761 ASSERT(isnullstartblock(del->br_startblock) == delay);
4762 flags = 0;
4763 qfield = 0;
4764 error = 0;
4765 /*
4766 * If deleting a real allocation, must free up the disk space.
4767 */
4768 if (!delay) {
4769 flags = XFS_ILOG_CORE;
4770 /*
4771 * Realtime allocation. Free it and record di_nblocks update.
4772 */
4773 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
4774 xfs_fsblock_t bno;
4775 xfs_filblks_t len;
4776
4777 ASSERT(do_mod(del->br_blockcount,
4778 mp->m_sb.sb_rextsize) == 0);
4779 ASSERT(do_mod(del->br_startblock,
4780 mp->m_sb.sb_rextsize) == 0);
4781 bno = del->br_startblock;
4782 len = del->br_blockcount;
4783 do_div(bno, mp->m_sb.sb_rextsize);
4784 do_div(len, mp->m_sb.sb_rextsize);
4785 error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
4786 if (error)
4787 goto done;
4788 do_fx = 0;
4789 nblks = len * mp->m_sb.sb_rextsize;
4790 qfield = XFS_TRANS_DQ_RTBCOUNT;
4791 }
4792 /*
4793 * Ordinary allocation.
4794 */
4795 else {
4796 do_fx = 1;
4797 nblks = del->br_blockcount;
4798 qfield = XFS_TRANS_DQ_BCOUNT;
4799 }
4800 /*
4801 * Set up del_endblock and cur for later.
4802 */
4803 del_endblock = del->br_startblock + del->br_blockcount;
4804 if (cur) {
4805 if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
4806 got.br_startblock, got.br_blockcount,
4807 &i)))
4808 goto done;
4809 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4810 }
4811 da_old = da_new = 0;
4812 } else {
4813 da_old = startblockval(got.br_startblock);
4814 da_new = 0;
4815 nblks = 0;
4816 do_fx = 0;
4817 }
4818 /*
4819 * Set flag value to use in switch statement.
4820 * Left-contig is 2, right-contig is 1.
4821 */
4822 switch (((got.br_startoff == del->br_startoff) << 1) |
4823 (got_endoff == del_endoff)) {
4824 case 3:
4825 /*
4826 * Matches the whole extent. Delete the entry.
4827 */
4828 xfs_iext_remove(ip, *idx, 1,
4829 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
4830 --*idx;
4831 if (delay)
4832 break;
4833
4834 XFS_IFORK_NEXT_SET(ip, whichfork,
4835 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
4836 flags |= XFS_ILOG_CORE;
4837 if (!cur) {
4838 flags |= xfs_ilog_fext(whichfork);
4839 break;
4840 }
4841 if ((error = xfs_btree_delete(cur, &i)))
4842 goto done;
4843 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4844 break;
4845
4846 case 2:
4847 /*
4848 * Deleting the first part of the extent.
4849 */
4850 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4851 xfs_bmbt_set_startoff(ep, del_endoff);
4852 temp = got.br_blockcount - del->br_blockcount;
4853 xfs_bmbt_set_blockcount(ep, temp);
4854 if (delay) {
4855 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4856 da_old);
4857 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4858 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4859 da_new = temp;
4860 break;
4861 }
4862 xfs_bmbt_set_startblock(ep, del_endblock);
4863 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4864 if (!cur) {
4865 flags |= xfs_ilog_fext(whichfork);
4866 break;
4867 }
4868 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
4869 got.br_blockcount - del->br_blockcount,
4870 got.br_state)))
4871 goto done;
4872 break;
4873
4874 case 1:
4875 /*
4876 * Deleting the last part of the extent.
4877 */
4878 temp = got.br_blockcount - del->br_blockcount;
4879 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4880 xfs_bmbt_set_blockcount(ep, temp);
4881 if (delay) {
4882 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4883 da_old);
4884 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4885 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4886 da_new = temp;
4887 break;
4888 }
4889 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4890 if (!cur) {
4891 flags |= xfs_ilog_fext(whichfork);
4892 break;
4893 }
4894 if ((error = xfs_bmbt_update(cur, got.br_startoff,
4895 got.br_startblock,
4896 got.br_blockcount - del->br_blockcount,
4897 got.br_state)))
4898 goto done;
4899 break;
4900
4901 case 0:
4902 /*
4903 * Deleting the middle of the extent.
4904 */
4905 temp = del->br_startoff - got.br_startoff;
4906 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4907 xfs_bmbt_set_blockcount(ep, temp);
4908 new.br_startoff = del_endoff;
4909 temp2 = got_endoff - del_endoff;
4910 new.br_blockcount = temp2;
4911 new.br_state = got.br_state;
4912 if (!delay) {
4913 new.br_startblock = del_endblock;
4914 flags |= XFS_ILOG_CORE;
4915 if (cur) {
4916 if ((error = xfs_bmbt_update(cur,
4917 got.br_startoff,
4918 got.br_startblock, temp,
4919 got.br_state)))
4920 goto done;
4921 if ((error = xfs_btree_increment(cur, 0, &i)))
4922 goto done;
4923 cur->bc_rec.b = new;
4924 error = xfs_btree_insert(cur, &i);
4925 if (error && error != -ENOSPC)
4926 goto done;
4927 /*
4928 * If get no-space back from btree insert,
4929 * it tried a split, and we have a zero
4930 * block reservation.
4931 * Fix up our state and return the error.
4932 */
4933 if (error == -ENOSPC) {
4934 /*
4935 * Reset the cursor, don't trust
4936 * it after any insert operation.
4937 */
4938 if ((error = xfs_bmbt_lookup_eq(cur,
4939 got.br_startoff,
4940 got.br_startblock,
4941 temp, &i)))
4942 goto done;
4943 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4944 /*
4945 * Update the btree record back
4946 * to the original value.
4947 */
4948 if ((error = xfs_bmbt_update(cur,
4949 got.br_startoff,
4950 got.br_startblock,
4951 got.br_blockcount,
4952 got.br_state)))
4953 goto done;
4954 /*
4955 * Reset the extent record back
4956 * to the original value.
4957 */
4958 xfs_bmbt_set_blockcount(ep,
4959 got.br_blockcount);
4960 flags = 0;
4961 error = -ENOSPC;
4962 goto done;
4963 }
4964 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4965 } else
4966 flags |= xfs_ilog_fext(whichfork);
4967 XFS_IFORK_NEXT_SET(ip, whichfork,
4968 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
4969 } else {
4970 ASSERT(whichfork == XFS_DATA_FORK);
4971 temp = xfs_bmap_worst_indlen(ip, temp);
4972 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4973 temp2 = xfs_bmap_worst_indlen(ip, temp2);
4974 new.br_startblock = nullstartblock((int)temp2);
4975 da_new = temp + temp2;
4976 while (da_new > da_old) {
4977 if (temp) {
4978 temp--;
4979 da_new--;
4980 xfs_bmbt_set_startblock(ep,
4981 nullstartblock((int)temp));
4982 }
4983 if (da_new == da_old)
4984 break;
4985 if (temp2) {
4986 temp2--;
4987 da_new--;
4988 new.br_startblock =
4989 nullstartblock((int)temp2);
4990 }
4991 }
4992 }
4993 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4994 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
4995 ++*idx;
4996 break;
4997 }
4998 /*
4999 * If we need to, add to list of extents to delete.
5000 */
5001 if (do_fx)
5002 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
5003 mp);
5004 /*
5005 * Adjust inode # blocks in the file.
5006 */
5007 if (nblks)
5008 ip->i_d.di_nblocks -= nblks;
5009 /*
5010 * Adjust quota data.
5011 */
5012 if (qfield)
5013 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5014
5015 /*
5016 * Account for change in delayed indirect blocks.
5017 * Nothing to do for disk quota accounting here.
5018 */
5019 ASSERT(da_old >= da_new);
5020 if (da_old > da_new) {
5021 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5022 (int64_t)(da_old - da_new), 0);
5023 }
5024done:
5025 *logflagsp = flags;
5026 return error;
5027}
5028
5029/*
5030 * Unmap (remove) blocks from a file.
5031 * If nexts is nonzero then the number of extents to remove is limited to
5032 * that value. If not all extents in the block range can be removed then
5033 * *done is set.
5034 */
5035int /* error */
5036xfs_bunmapi(
5037 xfs_trans_t *tp, /* transaction pointer */
5038 struct xfs_inode *ip, /* incore inode */
5039 xfs_fileoff_t bno, /* starting offset to unmap */
5040 xfs_filblks_t len, /* length to unmap in file */
5041 int flags, /* misc flags */
5042 xfs_extnum_t nexts, /* number of extents max */
5043 xfs_fsblock_t *firstblock, /* first allocated block
5044 controls a.g. for allocs */
5045 xfs_bmap_free_t *flist, /* i/o: list extents to free */
5046 int *done) /* set if not done yet */
5047{
5048 xfs_btree_cur_t *cur; /* bmap btree cursor */
5049 xfs_bmbt_irec_t del; /* extent being deleted */
5050 int eof; /* is deleting at eof */
5051 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
5052 int error; /* error return value */
5053 xfs_extnum_t extno; /* extent number in list */
5054 xfs_bmbt_irec_t got; /* current extent record */
5055 xfs_ifork_t *ifp; /* inode fork pointer */
5056 int isrt; /* freeing in rt area */
5057 xfs_extnum_t lastx; /* last extent index used */
5058 int logflags; /* transaction logging flags */
5059 xfs_extlen_t mod; /* rt extent offset */
5060 xfs_mount_t *mp; /* mount structure */
5061 xfs_extnum_t nextents; /* number of file extents */
5062 xfs_bmbt_irec_t prev; /* previous extent record */
5063 xfs_fileoff_t start; /* first file offset deleted */
5064 int tmp_logflags; /* partial logging flags */
5065 int wasdel; /* was a delayed alloc extent */
5066 int whichfork; /* data or attribute fork */
5067 xfs_fsblock_t sum;
5068
5069 trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5070
5071 whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5072 XFS_ATTR_FORK : XFS_DATA_FORK;
5073 ifp = XFS_IFORK_PTR(ip, whichfork);
5074 if (unlikely(
5075 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5076 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5077 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5078 ip->i_mount);
5079 return -EFSCORRUPTED;
5080 }
5081 mp = ip->i_mount;
5082 if (XFS_FORCED_SHUTDOWN(mp))
5083 return -EIO;
5084
5085 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5086 ASSERT(len > 0);
5087 ASSERT(nexts >= 0);
5088
5089 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5090 (error = xfs_iread_extents(tp, ip, whichfork)))
5091 return error;
5092 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5093 if (nextents == 0) {
5094 *done = 1;
5095 return 0;
5096 }
5097 XFS_STATS_INC(xs_blk_unmap);
5098 isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5099 start = bno;
5100 bno = start + len - 1;
5101 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5102 &prev);
5103
5104 /*
5105 * Check to see if the given block number is past the end of the
5106 * file, back up to the last block if so...
5107 */
5108 if (eof) {
5109 ep = xfs_iext_get_ext(ifp, --lastx);
5110 xfs_bmbt_get_all(ep, &got);
5111 bno = got.br_startoff + got.br_blockcount - 1;
5112 }
5113 logflags = 0;
5114 if (ifp->if_flags & XFS_IFBROOT) {
5115 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5116 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5117 cur->bc_private.b.firstblock = *firstblock;
5118 cur->bc_private.b.flist = flist;
5119 cur->bc_private.b.flags = 0;
5120 } else
5121 cur = NULL;
5122
5123 if (isrt) {
5124 /*
5125 * Synchronize by locking the bitmap inode.
5126 */
5127 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
5128 xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5129 }
5130
5131 extno = 0;
5132 while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5133 (nexts == 0 || extno < nexts)) {
5134 /*
5135 * Is the found extent after a hole in which bno lives?
5136 * Just back up to the previous extent, if so.
5137 */
5138 if (got.br_startoff > bno) {
5139 if (--lastx < 0)
5140 break;
5141 ep = xfs_iext_get_ext(ifp, lastx);
5142 xfs_bmbt_get_all(ep, &got);
5143 }
5144 /*
5145 * Is the last block of this extent before the range
5146 * we're supposed to delete? If so, we're done.
5147 */
5148 bno = XFS_FILEOFF_MIN(bno,
5149 got.br_startoff + got.br_blockcount - 1);
5150 if (bno < start)
5151 break;
5152 /*
5153 * Then deal with the (possibly delayed) allocated space
5154 * we found.
5155 */
5156 ASSERT(ep != NULL);
5157 del = got;
5158 wasdel = isnullstartblock(del.br_startblock);
5159 if (got.br_startoff < start) {
5160 del.br_startoff = start;
5161 del.br_blockcount -= start - got.br_startoff;
5162 if (!wasdel)
5163 del.br_startblock += start - got.br_startoff;
5164 }
5165 if (del.br_startoff + del.br_blockcount > bno + 1)
5166 del.br_blockcount = bno + 1 - del.br_startoff;
5167 sum = del.br_startblock + del.br_blockcount;
5168 if (isrt &&
5169 (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5170 /*
5171 * Realtime extent not lined up at the end.
5172 * The extent could have been split into written
5173 * and unwritten pieces, or we could just be
5174 * unmapping part of it. But we can't really
5175 * get rid of part of a realtime extent.
5176 */
5177 if (del.br_state == XFS_EXT_UNWRITTEN ||
5178 !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5179 /*
5180 * This piece is unwritten, or we're not
5181 * using unwritten extents. Skip over it.
5182 */
5183 ASSERT(bno >= mod);
5184 bno -= mod > del.br_blockcount ?
5185 del.br_blockcount : mod;
5186 if (bno < got.br_startoff) {
5187 if (--lastx >= 0)
5188 xfs_bmbt_get_all(xfs_iext_get_ext(
5189 ifp, lastx), &got);
5190 }
5191 continue;
5192 }
5193 /*
5194 * It's written, turn it unwritten.
5195 * This is better than zeroing it.
5196 */
5197 ASSERT(del.br_state == XFS_EXT_NORM);
5198 ASSERT(xfs_trans_get_block_res(tp) > 0);
5199 /*
5200 * If this spans a realtime extent boundary,
5201 * chop it back to the start of the one we end at.
5202 */
5203 if (del.br_blockcount > mod) {
5204 del.br_startoff += del.br_blockcount - mod;
5205 del.br_startblock += del.br_blockcount - mod;
5206 del.br_blockcount = mod;
5207 }
5208 del.br_state = XFS_EXT_UNWRITTEN;
5209 error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5210 &lastx, &cur, &del, firstblock, flist,
5211 &logflags);
5212 if (error)
5213 goto error0;
5214 goto nodelete;
5215 }
5216 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5217 /*
5218 * Realtime extent is lined up at the end but not
5219 * at the front. We'll get rid of full extents if
5220 * we can.
5221 */
5222 mod = mp->m_sb.sb_rextsize - mod;
5223 if (del.br_blockcount > mod) {
5224 del.br_blockcount -= mod;
5225 del.br_startoff += mod;
5226 del.br_startblock += mod;
5227 } else if ((del.br_startoff == start &&
5228 (del.br_state == XFS_EXT_UNWRITTEN ||
5229 xfs_trans_get_block_res(tp) == 0)) ||
5230 !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5231 /*
5232 * Can't make it unwritten. There isn't
5233 * a full extent here so just skip it.
5234 */
5235 ASSERT(bno >= del.br_blockcount);
5236 bno -= del.br_blockcount;
5237 if (got.br_startoff > bno) {
5238 if (--lastx >= 0) {
5239 ep = xfs_iext_get_ext(ifp,
5240 lastx);
5241 xfs_bmbt_get_all(ep, &got);
5242 }
5243 }
5244 continue;
5245 } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5246 /*
5247 * This one is already unwritten.
5248 * It must have a written left neighbor.
5249 * Unwrite the killed part of that one and
5250 * try again.
5251 */
5252 ASSERT(lastx > 0);
5253 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5254 lastx - 1), &prev);
5255 ASSERT(prev.br_state == XFS_EXT_NORM);
5256 ASSERT(!isnullstartblock(prev.br_startblock));
5257 ASSERT(del.br_startblock ==
5258 prev.br_startblock + prev.br_blockcount);
5259 if (prev.br_startoff < start) {
5260 mod = start - prev.br_startoff;
5261 prev.br_blockcount -= mod;
5262 prev.br_startblock += mod;
5263 prev.br_startoff = start;
5264 }
5265 prev.br_state = XFS_EXT_UNWRITTEN;
5266 lastx--;
5267 error = xfs_bmap_add_extent_unwritten_real(tp,
5268 ip, &lastx, &cur, &prev,
5269 firstblock, flist, &logflags);
5270 if (error)
5271 goto error0;
5272 goto nodelete;
5273 } else {
5274 ASSERT(del.br_state == XFS_EXT_NORM);
5275 del.br_state = XFS_EXT_UNWRITTEN;
5276 error = xfs_bmap_add_extent_unwritten_real(tp,
5277 ip, &lastx, &cur, &del,
5278 firstblock, flist, &logflags);
5279 if (error)
5280 goto error0;
5281 goto nodelete;
5282 }
5283 }
5284 if (wasdel) {
5285 ASSERT(startblockval(del.br_startblock) > 0);
5286 /* Update realtime/data freespace, unreserve quota */
5287 if (isrt) {
5288 xfs_filblks_t rtexts;
5289
5290 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5291 do_div(rtexts, mp->m_sb.sb_rextsize);
5292 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5293 (int64_t)rtexts, 0);
5294 (void)xfs_trans_reserve_quota_nblks(NULL,
5295 ip, -((long)del.br_blockcount), 0,
5296 XFS_QMOPT_RES_RTBLKS);
5297 } else {
5298 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5299 (int64_t)del.br_blockcount, 0);
5300 (void)xfs_trans_reserve_quota_nblks(NULL,
5301 ip, -((long)del.br_blockcount), 0,
5302 XFS_QMOPT_RES_REGBLKS);
5303 }
5304 ip->i_delayed_blks -= del.br_blockcount;
5305 if (cur)
5306 cur->bc_private.b.flags |=
5307 XFS_BTCUR_BPRV_WASDEL;
5308 } else if (cur)
5309 cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5310 /*
5311 * If it's the case where the directory code is running
5312 * with no block reservation, and the deleted block is in
5313 * the middle of its extent, and the resulting insert
5314 * of an extent would cause transformation to btree format,
5315 * then reject it. The calling code will then swap
5316 * blocks around instead.
5317 * We have to do this now, rather than waiting for the
5318 * conversion to btree format, since the transaction
5319 * will be dirty.
5320 */
5321 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5322 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5323 XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
5324 XFS_IFORK_MAXEXT(ip, whichfork) &&
5325 del.br_startoff > got.br_startoff &&
5326 del.br_startoff + del.br_blockcount <
5327 got.br_startoff + got.br_blockcount) {
5328 error = -ENOSPC;
5329 goto error0;
5330 }
5331 error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5332 &tmp_logflags, whichfork);
5333 logflags |= tmp_logflags;
5334 if (error)
5335 goto error0;
5336 bno = del.br_startoff - 1;
5337nodelete:
5338 /*
5339 * If not done go on to the next (previous) record.
5340 */
5341 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5342 if (lastx >= 0) {
5343 ep = xfs_iext_get_ext(ifp, lastx);
5344 if (xfs_bmbt_get_startoff(ep) > bno) {
5345 if (--lastx >= 0)
5346 ep = xfs_iext_get_ext(ifp,
5347 lastx);
5348 }
5349 xfs_bmbt_get_all(ep, &got);
5350 }
5351 extno++;
5352 }
5353 }
5354 *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5355
5356 /*
5357 * Convert to a btree if necessary.
5358 */
5359 if (xfs_bmap_needs_btree(ip, whichfork)) {
5360 ASSERT(cur == NULL);
5361 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5362 &cur, 0, &tmp_logflags, whichfork);
5363 logflags |= tmp_logflags;
5364 if (error)
5365 goto error0;
5366 }
5367 /*
5368 * transform from btree to extents, give it cur
5369 */
5370 else if (xfs_bmap_wants_extents(ip, whichfork)) {
5371 ASSERT(cur != NULL);
5372 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5373 whichfork);
5374 logflags |= tmp_logflags;
5375 if (error)
5376 goto error0;
5377 }
5378 /*
5379 * transform from extents to local?
5380 */
5381 error = 0;
5382error0:
5383 /*
5384 * Log everything. Do this after conversion, there's no point in
5385 * logging the extent records if we've converted to btree format.
5386 */
5387 if ((logflags & xfs_ilog_fext(whichfork)) &&
5388 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5389 logflags &= ~xfs_ilog_fext(whichfork);
5390 else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5391 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5392 logflags &= ~xfs_ilog_fbroot(whichfork);
5393 /*
5394 * Log inode even in the error case, if the transaction
5395 * is dirty we'll need to shut down the filesystem.
5396 */
5397 if (logflags)
5398 xfs_trans_log_inode(tp, ip, logflags);
5399 if (cur) {
5400 if (!error) {
5401 *firstblock = cur->bc_private.b.firstblock;
5402 cur->bc_private.b.allocated = 0;
5403 }
5404 xfs_btree_del_cursor(cur,
5405 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5406 }
5407 return error;
5408}
5409
5410/*
5411 * Shift extent records to the left to cover a hole.
5412 *
5413 * The maximum number of extents to be shifted in a single operation
5414 * is @num_exts, and @current_ext keeps track of the current extent
5415 * index we have shifted. @offset_shift_fsb is the length by which each
5416 * extent is shifted. If there is no hole to shift the extents
5417 * into, this will be considered invalid operation and we abort immediately.
5418 */
5419int
5420xfs_bmap_shift_extents(
5421 struct xfs_trans *tp,
5422 struct xfs_inode *ip,
5423 int *done,
5424 xfs_fileoff_t start_fsb,
5425 xfs_fileoff_t offset_shift_fsb,
5426 xfs_extnum_t *current_ext,
5427 xfs_fsblock_t *firstblock,
5428 struct xfs_bmap_free *flist,
5429 int num_exts)
5430{
5431 struct xfs_btree_cur *cur;
5432 struct xfs_bmbt_rec_host *gotp;
5433 struct xfs_bmbt_irec got;
5434 struct xfs_bmbt_irec left;
5435 struct xfs_mount *mp = ip->i_mount;
5436 struct xfs_ifork *ifp;
5437 xfs_extnum_t nexts = 0;
5438 xfs_fileoff_t startoff;
5439 int error = 0;
5440 int i;
5441 int whichfork = XFS_DATA_FORK;
5442 int logflags;
5443 xfs_filblks_t blockcount = 0;
5444 int total_extents;
5445
5446 if (unlikely(XFS_TEST_ERROR(
5447 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5448 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
5449 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
5450 XFS_ERROR_REPORT("xfs_bmap_shift_extents",
5451 XFS_ERRLEVEL_LOW, mp);
5452 return -EFSCORRUPTED;
5453 }
5454
5455 if (XFS_FORCED_SHUTDOWN(mp))
5456 return -EIO;
5457
5458 ASSERT(current_ext != NULL);
5459
5460 ifp = XFS_IFORK_PTR(ip, whichfork);
5461 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5462 /* Read in all the extents */
5463 error = xfs_iread_extents(tp, ip, whichfork);
5464 if (error)
5465 return error;
5466 }
5467
5468 /*
5469 * If *current_ext is 0, we would need to lookup the extent
5470 * from where we would start shifting and store it in gotp.
5471 */
5472 if (!*current_ext) {
5473 gotp = xfs_iext_bno_to_ext(ifp, start_fsb, current_ext);
5474 /*
5475 * gotp can be null in 2 cases: 1) if there are no extents
5476 * or 2) start_fsb lies in a hole beyond which there are
5477 * no extents. Either way, we are done.
5478 */
5479 if (!gotp) {
5480 *done = 1;
5481 return 0;
5482 }
5483 }
5484
5485 /* We are going to change core inode */
5486 logflags = XFS_ILOG_CORE;
5487 if (ifp->if_flags & XFS_IFBROOT) {
5488 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5489 cur->bc_private.b.firstblock = *firstblock;
5490 cur->bc_private.b.flist = flist;
5491 cur->bc_private.b.flags = 0;
5492 } else {
5493 cur = NULL;
5494 logflags |= XFS_ILOG_DEXT;
5495 }
5496
5497 /*
5498 * There may be delalloc extents in the data fork before the range we
5499 * are collapsing out, so we cannot
5500 * use the count of real extents here. Instead we have to calculate it
5501 * from the incore fork.
5502 */
5503 total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
5504 while (nexts++ < num_exts && *current_ext < total_extents) {
5505
5506 gotp = xfs_iext_get_ext(ifp, *current_ext);
5507 xfs_bmbt_get_all(gotp, &got);
5508 startoff = got.br_startoff - offset_shift_fsb;
5509
5510 /*
5511 * Before shifting extent into hole, make sure that the hole
5512 * is large enough to accomodate the shift.
5513 */
5514 if (*current_ext) {
5515 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5516 *current_ext - 1), &left);
5517
5518 if (startoff < left.br_startoff + left.br_blockcount)
5519 error = -EINVAL;
5520 } else if (offset_shift_fsb > got.br_startoff) {
5521 /*
5522 * When first extent is shifted, offset_shift_fsb
5523 * should be less than the stating offset of
5524 * the first extent.
5525 */
5526 error = -EINVAL;
5527 }
5528
5529 if (error)
5530 goto del_cursor;
5531
5532 if (cur) {
5533 error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
5534 got.br_startblock,
5535 got.br_blockcount,
5536 &i);
5537 if (error)
5538 goto del_cursor;
5539 XFS_WANT_CORRUPTED_GOTO(i == 1, del_cursor);
5540 }
5541
5542 /* Check if we can merge 2 adjacent extents */
5543 if (*current_ext &&
5544 left.br_startoff + left.br_blockcount == startoff &&
5545 left.br_startblock + left.br_blockcount ==
5546 got.br_startblock &&
5547 left.br_state == got.br_state &&
5548 left.br_blockcount + got.br_blockcount <= MAXEXTLEN) {
5549 blockcount = left.br_blockcount +
5550 got.br_blockcount;
5551 xfs_iext_remove(ip, *current_ext, 1, 0);
5552 if (cur) {
5553 error = xfs_btree_delete(cur, &i);
5554 if (error)
5555 goto del_cursor;
5556 XFS_WANT_CORRUPTED_GOTO(i == 1, del_cursor);
5557 }
5558 XFS_IFORK_NEXT_SET(ip, whichfork,
5559 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5560 gotp = xfs_iext_get_ext(ifp, --*current_ext);
5561 xfs_bmbt_get_all(gotp, &got);
5562
5563 /* Make cursor point to the extent we will update */
5564 if (cur) {
5565 error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
5566 got.br_startblock,
5567 got.br_blockcount,
5568 &i);
5569 if (error)
5570 goto del_cursor;
5571 XFS_WANT_CORRUPTED_GOTO(i == 1, del_cursor);
5572 }
5573
5574 xfs_bmbt_set_blockcount(gotp, blockcount);
5575 got.br_blockcount = blockcount;
5576 } else {
5577 /* We have to update the startoff */
5578 xfs_bmbt_set_startoff(gotp, startoff);
5579 got.br_startoff = startoff;
5580 }
5581
5582 if (cur) {
5583 error = xfs_bmbt_update(cur, got.br_startoff,
5584 got.br_startblock,
5585 got.br_blockcount,
5586 got.br_state);
5587 if (error)
5588 goto del_cursor;
5589 }
5590
5591 (*current_ext)++;
5592 total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
5593 }
5594
5595 /* Check if we are done */
5596 if (*current_ext == total_extents)
5597 *done = 1;
5598
5599del_cursor:
5600 if (cur)
5601 xfs_btree_del_cursor(cur,
5602 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5603
5604 xfs_trans_log_inode(tp, ip, logflags);
5605 return error;
5606}
generated by cgit v1.2.2 (git 2.25.0) at 2025-01-16 09:53:49 -0500