]> Git Repo - linux.git/blame - fs/xfs/libxfs/xfs_bmap.c
projects / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
xfs: consolidate superblock logging functions
[linux.git] / fs / xfs / libxfs / xfs_bmap.c
CommitLineData
1da177e4 1/*
3e57ecf6 2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
7b718769 3 * All Rights Reserved.
1da177e4 4 *
7b718769
NS		 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
1da177e4
LT		 7 * published by the Free Software Foundation.
8 *
7b718769
NS		 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.
1da177e4 13 *
7b718769
NS		 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
1da177e4 17 */
1da177e4 18#include "xfs.h"
a844f451 19#include "xfs_fs.h"
70a9883c 20#include "xfs_shared.h"
239880ef
DC		 21#include "xfs_format.h"
22#include "xfs_log_format.h"
23#include "xfs_trans_resv.h"
a844f451 24#include "xfs_bit.h"
1da177e4 25#include "xfs_sb.h"
f5ea1100 26#include "xfs_mount.h"
57062787 27#include "xfs_da_format.h"
a844f451 28#include "xfs_da_btree.h"
2b9ab5ab 29#include "xfs_dir2.h"
1da177e4 30#include "xfs_inode.h"
a844f451 31#include "xfs_btree.h"
239880ef 32#include "xfs_trans.h"
a844f451 33#include "xfs_inode_item.h"
1da177e4
LT		 34#include "xfs_extfree_item.h"
35#include "xfs_alloc.h"
36#include "xfs_bmap.h"
68988114 37#include "xfs_bmap_util.h"
a4fbe6ab 38#include "xfs_bmap_btree.h"
1da177e4
LT		 39#include "xfs_rtalloc.h"
40#include "xfs_error.h"
1da177e4
LT		 41#include "xfs_quota.h"
42#include "xfs_trans_space.h"
43#include "xfs_buf_item.h"
0b1b213f 44#include "xfs_trace.h"
19de7351 45#include "xfs_symlink.h"
a4fbe6ab 46#include "xfs_attr_leaf.h"
a4fbe6ab 47#include "xfs_filestream.h"
1da177e4
LT		 48
49
1da177e4
LT		 50kmem_zone_t *xfs_bmap_free_item_zone;
51
52/*
9e5987a7 53 * Miscellaneous helper functions
1da177e4
LT		 54 */
55
1da177e4 56/*
9e5987a7
DC		 57 * Compute and fill in the value of the maximum depth of a bmap btree
58 * in this filesystem. Done once, during mount.
1da177e4 59 */
9e5987a7
DC		 60void
61xfs_bmap_compute_maxlevels(
62 xfs_mount_t *mp, /* file system mount structure */
63 int whichfork) /* data or attr fork */
64{
65 int level; /* btree level */
66 uint maxblocks; /* max blocks at this level */
67 uint maxleafents; /* max leaf entries possible */
68 int maxrootrecs; /* max records in root block */
69 int minleafrecs; /* min records in leaf block */
70 int minnoderecs; /* min records in node block */
71 int sz; /* root block size */
1da177e4 72
9e5987a7
DC		 73 /*
74 * The maximum number of extents in a file, hence the maximum
75 * number of leaf entries, is controlled by the type of di_nextents
76 * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
77 * (a signed 16-bit number, xfs_aextnum_t).
78 *
79 * Note that we can no longer assume that if we are in ATTR1 that
80 * the fork offset of all the inodes will be
81 * (xfs_default_attroffset(ip) >> 3) because we could have mounted
82 * with ATTR2 and then mounted back with ATTR1, keeping the
83 * di_forkoff's fixed but probably at various positions. Therefore,
84 * for both ATTR1 and ATTR2 we have to assume the worst case scenario
85 * of a minimum size available.
86 */
87 if (whichfork == XFS_DATA_FORK) {
88 maxleafents = MAXEXTNUM;
89 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
90 } else {
91 maxleafents = MAXAEXTNUM;
92 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
93 }
152d93b7 94 maxrootrecs = xfs_bmdr_maxrecs(sz, 0);
9e5987a7
DC		 95 minleafrecs = mp->m_bmap_dmnr[0];
96 minnoderecs = mp->m_bmap_dmnr[1];
97 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
98 for (level = 1; maxblocks > 1; level++) {
99 if (maxblocks <= maxrootrecs)
100 maxblocks = 1;
101 else
102 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
103 }
104 mp->m_bm_maxlevels[whichfork] = level;
105}
91e11088 106
fe033cc8
CH		 107STATIC int /* error */
108xfs_bmbt_lookup_eq(
109 struct xfs_btree_cur *cur,
110 xfs_fileoff_t off,
111 xfs_fsblock_t bno,
112 xfs_filblks_t len,
113 int *stat) /* success/failure */
114{
115 cur->bc_rec.b.br_startoff = off;
116 cur->bc_rec.b.br_startblock = bno;
117 cur->bc_rec.b.br_blockcount = len;
118 return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
119}
120
121STATIC int /* error */
122xfs_bmbt_lookup_ge(
123 struct xfs_btree_cur *cur,
124 xfs_fileoff_t off,
125 xfs_fsblock_t bno,
126 xfs_filblks_t len,
127 int *stat) /* success/failure */
128{
129 cur->bc_rec.b.br_startoff = off;
130 cur->bc_rec.b.br_startblock = bno;
131 cur->bc_rec.b.br_blockcount = len;
132 return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
133}
134
278d0ca1 135/*
8096b1eb
CH		 136 * Check if the inode needs to be converted to btree format.
137 */
138static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
139{
140 return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
141 XFS_IFORK_NEXTENTS(ip, whichfork) >
142 XFS_IFORK_MAXEXT(ip, whichfork);
143}
144
145/*
146 * Check if the inode should be converted to extent format.
147 */
148static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
149{
150 return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
151 XFS_IFORK_NEXTENTS(ip, whichfork) <=
152 XFS_IFORK_MAXEXT(ip, whichfork);
153}
154
155/*
156 * Update the record referred to by cur to the value given
278d0ca1
CH		 157 * by [off, bno, len, state].
158 * This either works (return 0) or gets an EFSCORRUPTED error.
159 */
160STATIC int
161xfs_bmbt_update(
162 struct xfs_btree_cur *cur,
163 xfs_fileoff_t off,
164 xfs_fsblock_t bno,
165 xfs_filblks_t len,
166 xfs_exntst_t state)
167{
168 union xfs_btree_rec rec;
169
170 xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
171 return xfs_btree_update(cur, &rec);
172}
fe033cc8 173
1da177e4 174/*
9e5987a7
DC		 175 * Compute the worst-case number of indirect blocks that will be used
176 * for ip's delayed extent of length "len".
1da177e4 177 */
9e5987a7
DC		 178STATIC xfs_filblks_t
179xfs_bmap_worst_indlen(
180 xfs_inode_t *ip, /* incore inode pointer */
181 xfs_filblks_t len) /* delayed extent length */
1da177e4 182{
9e5987a7
DC		 183 int level; /* btree level number */
184 int maxrecs; /* maximum record count at this level */
185 xfs_mount_t *mp; /* mount structure */
186 xfs_filblks_t rval; /* return value */
1da177e4
LT		 187
188 mp = ip->i_mount;
9e5987a7
DC		 189 maxrecs = mp->m_bmap_dmxr[0];
190 for (level = 0, rval = 0;
191 level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
192 level++) {
193 len += maxrecs - 1;
194 do_div(len, maxrecs);
195 rval += len;
196 if (len == 1)
197 return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
198 level - 1;
199 if (level == 0)
200 maxrecs = mp->m_bmap_dmxr[1];
1da177e4 201 }
9e5987a7 202 return rval;
1da177e4
LT		 203}
204
205/*
9e5987a7 206 * Calculate the default attribute fork offset for newly created inodes.
1da177e4 207 */
9e5987a7
DC		 208uint
209xfs_default_attroffset(
210 struct xfs_inode *ip)
1da177e4 211{
9e5987a7
DC		 212 struct xfs_mount *mp = ip->i_mount;
213 uint offset;
1da177e4 214
9e5987a7 215 if (mp->m_sb.sb_inodesize == 256) {
56cea2d0 216 offset = XFS_LITINO(mp, ip->i_d.di_version) -
9e5987a7
DC		 217 XFS_BMDR_SPACE_CALC(MINABTPTRS);
218 } else {
219 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
1da177e4 220 }
9e5987a7 221
56cea2d0 222 ASSERT(offset < XFS_LITINO(mp, ip->i_d.di_version));
9e5987a7 223 return offset;
1da177e4
LT		 224}
225
226/*
9e5987a7
DC		 227 * Helper routine to reset inode di_forkoff field when switching
228 * attribute fork from local to extent format - we reset it where
229 * possible to make space available for inline data fork extents.
1e82379b
DC		 230 */
231STATIC void
9e5987a7 232xfs_bmap_forkoff_reset(
9e5987a7
DC		 233 xfs_inode_t *ip,
234 int whichfork)
1e82379b 235{
9e5987a7
DC		 236 if (whichfork == XFS_ATTR_FORK &&
237 ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
238 ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
239 ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
240 uint dfl_forkoff = xfs_default_attroffset(ip) >> 3;
241
242 if (dfl_forkoff > ip->i_d.di_forkoff)
243 ip->i_d.di_forkoff = dfl_forkoff;
244 }
1e82379b
DC		 245}
246
9e5987a7
DC		 247/*
248 * Debug/sanity checking code
249 */
7574aa92 250
9e5987a7
DC		 251STATIC int
252xfs_bmap_sanity_check(
253 struct xfs_mount *mp,
254 struct xfs_buf *bp,
255 int level)
256{
257 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
7574aa92 258
ee1a47ab
CH		 259 if (block->bb_magic != cpu_to_be32(XFS_BMAP_CRC_MAGIC) &&
260 block->bb_magic != cpu_to_be32(XFS_BMAP_MAGIC))
261 return 0;
262
263 if (be16_to_cpu(block->bb_level) != level ||
9e5987a7
DC		 264 be16_to_cpu(block->bb_numrecs) == 0 ||
265 be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
266 return 0;
ee1a47ab 267
9e5987a7
DC		 268 return 1;
269}
7574aa92 270
9e5987a7
DC		 271#ifdef DEBUG
272STATIC struct xfs_buf *
273xfs_bmap_get_bp(
274 struct xfs_btree_cur *cur,
275 xfs_fsblock_t bno)
276{
277 struct xfs_log_item_desc *lidp;
278 int i;
7574aa92 279
9e5987a7
DC		 280 if (!cur)
281 return NULL;
282
283 for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
284 if (!cur->bc_bufs[i])
285 break;
286 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
287 return cur->bc_bufs[i];
1da177e4 288 }
7574aa92 289
9e5987a7
DC		 290 /* Chase down all the log items to see if the bp is there */
291 list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
292 struct xfs_buf_log_item *bip;
293 bip = (struct xfs_buf_log_item *)lidp->lid_item;
294 if (bip->bli_item.li_type == XFS_LI_BUF &&
295 XFS_BUF_ADDR(bip->bli_buf) == bno)
296 return bip->bli_buf;
297 }
7574aa92 298
9e5987a7
DC		 299 return NULL;
300}
0b1b213f 301
9e5987a7
DC		 302STATIC void
303xfs_check_block(
304 struct xfs_btree_block *block,
305 xfs_mount_t *mp,
306 int root,
307 short sz)
308{
309 int i, j, dmxr;
310 __be64 *pp, *thispa; /* pointer to block address */
311 xfs_bmbt_key_t *prevp, *keyp;
1da177e4 312
9e5987a7 313 ASSERT(be16_to_cpu(block->bb_level) > 0);
ec90c556 314
9e5987a7
DC		 315 prevp = NULL;
316 for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
317 dmxr = mp->m_bmap_dmxr[0];
318 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
0b1b213f 319
9e5987a7
DC		 320 if (prevp) {
321 ASSERT(be64_to_cpu(prevp->br_startoff) <
322 be64_to_cpu(keyp->br_startoff));
1da177e4 323 }
9e5987a7 324 prevp = keyp;
1da177e4 325
1da177e4 326 /*
9e5987a7 327 * Compare the block numbers to see if there are dups.
1da177e4 328 */
9e5987a7
DC		 329 if (root)
330 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
331 else
332 pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
0b1b213f 333
9e5987a7
DC		 334 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
335 if (root)
336 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
337 else
338 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
339 if (*thispa == *pp) {
340 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
341 __func__, j, i,
342 (unsigned long long)be64_to_cpu(*thispa));
343 panic("%s: ptrs are equal in node\n",
344 __func__);
345 }
1da177e4 346 }
9e5987a7
DC		 347 }
348}
1da177e4 349
9e5987a7
DC		 350/*
351 * Check that the extents for the inode ip are in the right order in all
352 * btree leaves.
353 */
0b1b213f 354
9e5987a7
DC		 355STATIC void
356xfs_bmap_check_leaf_extents(
357 xfs_btree_cur_t *cur, /* btree cursor or null */
358 xfs_inode_t *ip, /* incore inode pointer */
359 int whichfork) /* data or attr fork */
360{
361 struct xfs_btree_block *block; /* current btree block */
362 xfs_fsblock_t bno; /* block # of "block" */
363 xfs_buf_t *bp; /* buffer for "block" */
364 int error; /* error return value */
365 xfs_extnum_t i=0, j; /* index into the extents list */
366 xfs_ifork_t *ifp; /* fork structure */
367 int level; /* btree level, for checking */
368 xfs_mount_t *mp; /* file system mount structure */
369 __be64 *pp; /* pointer to block address */
370 xfs_bmbt_rec_t *ep; /* pointer to current extent */
371 xfs_bmbt_rec_t last = {0, 0}; /* last extent in prev block */
372 xfs_bmbt_rec_t *nextp; /* pointer to next extent */
373 int bp_release = 0;
374
375 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
376 return;
377 }
378
379 bno = NULLFSBLOCK;
380 mp = ip->i_mount;
381 ifp = XFS_IFORK_PTR(ip, whichfork);
382 block = ifp->if_broot;
383 /*
384 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
385 */
386 level = be16_to_cpu(block->bb_level);
387 ASSERT(level > 0);
388 xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
389 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
390 bno = be64_to_cpu(*pp);
391
d5cf09ba 392 ASSERT(bno != NULLFSBLOCK);
9e5987a7
DC		 393 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
394 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
395
396 /*
397 * Go down the tree until leaf level is reached, following the first
398 * pointer (leftmost) at each level.
399 */
400 while (level-- > 0) {
401 /* See if buf is in cur first */
402 bp_release = 0;
403 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
404 if (!bp) {
405 bp_release = 1;
406 error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
407 XFS_BMAP_BTREE_REF,
408 &xfs_bmbt_buf_ops);
572a4cf0 409 if (error)
9e5987a7 410 goto error_norelse;
1da177e4 411 }
9e5987a7
DC		 412 block = XFS_BUF_TO_BLOCK(bp);
413 XFS_WANT_CORRUPTED_GOTO(
414 xfs_bmap_sanity_check(mp, bp, level),
415 error0);
416 if (level == 0)
417 break;
1da177e4 418
1da177e4 419 /*
9e5987a7
DC		 420 * Check this block for basic sanity (increasing keys and
421 * no duplicate blocks).
1da177e4 422 */
0b1b213f 423
9e5987a7
DC		 424 xfs_check_block(block, mp, 0, 0);
425 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
426 bno = be64_to_cpu(*pp);
427 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
428 if (bp_release) {
429 bp_release = 0;
430 xfs_trans_brelse(NULL, bp);
1da177e4 431 }
9e5987a7 432 }
ec90c556 433
9e5987a7
DC		 434 /*
435 * Here with bp and block set to the leftmost leaf node in the tree.
436 */
437 i = 0;
438
439 /*
440 * Loop over all leaf nodes checking that all extents are in the right order.
441 */
442 for (;;) {
443 xfs_fsblock_t nextbno;
444 xfs_extnum_t num_recs;
445
446
447 num_recs = xfs_btree_get_numrecs(block);
1da177e4 448
1da177e4 449 /*
9e5987a7 450 * Read-ahead the next leaf block, if any.
1da177e4 451 */
8096b1eb 452
9e5987a7 453 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
1da177e4 454
1da177e4 455 /*
9e5987a7
DC		 456 * Check all the extents to make sure they are OK.
457 * If we had a previous block, the last entry should
458 * conform with the first entry in this one.
1da177e4 459 */
ec90c556 460
9e5987a7
DC		 461 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
462 if (i) {
463 ASSERT(xfs_bmbt_disk_get_startoff(&last) +
464 xfs_bmbt_disk_get_blockcount(&last) <=
465 xfs_bmbt_disk_get_startoff(ep));
466 }
467 for (j = 1; j < num_recs; j++) {
468 nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
469 ASSERT(xfs_bmbt_disk_get_startoff(ep) +
470 xfs_bmbt_disk_get_blockcount(ep) <=
471 xfs_bmbt_disk_get_startoff(nextp));
472 ep = nextp;
473 }
1da177e4 474
9e5987a7
DC		 475 last = *ep;
476 i += num_recs;
477 if (bp_release) {
478 bp_release = 0;
479 xfs_trans_brelse(NULL, bp);
480 }
481 bno = nextbno;
1da177e4 482 /*
9e5987a7 483 * If we've reached the end, stop.
1da177e4 484 */
9e5987a7
DC		 485 if (bno == NULLFSBLOCK)
486 break;
8096b1eb 487
9e5987a7
DC		 488 bp_release = 0;
489 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
490 if (!bp) {
491 bp_release = 1;
492 error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
493 XFS_BMAP_BTREE_REF,
494 &xfs_bmbt_buf_ops);
b9b984d7 495 if (error)
9e5987a7 496 goto error_norelse;
1da177e4 497 }
9e5987a7 498 block = XFS_BUF_TO_BLOCK(bp);
a5bd606b 499 }
9e5987a7
DC		 500 if (bp_release) {
501 bp_release = 0;
502 xfs_trans_brelse(NULL, bp);
a5bd606b 503 }
9e5987a7 504 return;
a5bd606b 505
9e5987a7
DC		 506error0:
507 xfs_warn(mp, "%s: at error0", __func__);
508 if (bp_release)
509 xfs_trans_brelse(NULL, bp);
510error_norelse:
511 xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
512 __func__, i);
513 panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
514 return;
1da177e4
LT		 515}
516
517/*
9e5987a7 518 * Add bmap trace insert entries for all the contents of the extent records.
1da177e4 519 */
9e5987a7
DC		 520void
521xfs_bmap_trace_exlist(
522 xfs_inode_t *ip, /* incore inode pointer */
523 xfs_extnum_t cnt, /* count of entries in the list */
524 int whichfork, /* data or attr fork */
525 unsigned long caller_ip)
1da177e4 526{
9e5987a7
DC		 527 xfs_extnum_t idx; /* extent record index */
528 xfs_ifork_t *ifp; /* inode fork pointer */
529 int state = 0;
a5bd606b 530
9e5987a7
DC		 531 if (whichfork == XFS_ATTR_FORK)
532 state |= BMAP_ATTRFORK;
a5bd606b 533
9e5987a7
DC		 534 ifp = XFS_IFORK_PTR(ip, whichfork);
535 ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
536 for (idx = 0; idx < cnt; idx++)
537 trace_xfs_extlist(ip, idx, whichfork, caller_ip);
538}
a5bd606b 539
9e5987a7
DC		 540/*
541 * Validate that the bmbt_irecs being returned from bmapi are valid
a97f4df7
ZYW		 542 * given the caller's original parameters. Specifically check the
543 * ranges of the returned irecs to ensure that they only extend beyond
9e5987a7
DC		 544 * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
545 */
546STATIC void
547xfs_bmap_validate_ret(
548 xfs_fileoff_t bno,
549 xfs_filblks_t len,
550 int flags,
551 xfs_bmbt_irec_t *mval,
552 int nmap,
553 int ret_nmap)
554{
555 int i; /* index to map values */
a5bd606b 556
9e5987a7 557 ASSERT(ret_nmap <= nmap);
a5bd606b 558
9e5987a7
DC		 559 for (i = 0; i < ret_nmap; i++) {
560 ASSERT(mval[i].br_blockcount > 0);
561 if (!(flags & XFS_BMAPI_ENTIRE)) {
562 ASSERT(mval[i].br_startoff >= bno);
563 ASSERT(mval[i].br_blockcount <= len);
564 ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
565 bno + len);
566 } else {
567 ASSERT(mval[i].br_startoff < bno + len);
568 ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
569 bno);
570 }
571 ASSERT(i == 0 ||
572 mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
573 mval[i].br_startoff);
574 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
575 mval[i].br_startblock != HOLESTARTBLOCK);
576 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
577 mval[i].br_state == XFS_EXT_UNWRITTEN);
578 }
579}
7574aa92 580
9e5987a7
DC		 581#else
582#define xfs_bmap_check_leaf_extents(cur, ip, whichfork) do { } while (0)
583#define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
584#endif /* DEBUG */
7574aa92 585
9e5987a7
DC		 586/*
587 * bmap free list manipulation functions
588 */
7574aa92 589
9e5987a7
DC		 590/*
591 * Add the extent to the list of extents to be free at transaction end.
592 * The list is maintained sorted (by block number).
593 */
594void
595xfs_bmap_add_free(
596 xfs_fsblock_t bno, /* fs block number of extent */
597 xfs_filblks_t len, /* length of extent */
598 xfs_bmap_free_t *flist, /* list of extents */
599 xfs_mount_t *mp) /* mount point structure */
600{
601 xfs_bmap_free_item_t *cur; /* current (next) element */
602 xfs_bmap_free_item_t *new; /* new element */
603 xfs_bmap_free_item_t *prev; /* previous element */
604#ifdef DEBUG
605 xfs_agnumber_t agno;
606 xfs_agblock_t agbno;
607
608 ASSERT(bno != NULLFSBLOCK);
609 ASSERT(len > 0);
610 ASSERT(len <= MAXEXTLEN);
611 ASSERT(!isnullstartblock(bno));
612 agno = XFS_FSB_TO_AGNO(mp, bno);
613 agbno = XFS_FSB_TO_AGBNO(mp, bno);
614 ASSERT(agno < mp->m_sb.sb_agcount);
615 ASSERT(agbno < mp->m_sb.sb_agblocks);
616 ASSERT(len < mp->m_sb.sb_agblocks);
617 ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
618#endif
619 ASSERT(xfs_bmap_free_item_zone != NULL);
620 new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
621 new->xbfi_startblock = bno;
622 new->xbfi_blockcount = (xfs_extlen_t)len;
623 for (prev = NULL, cur = flist->xbf_first;
624 cur != NULL;
625 prev = cur, cur = cur->xbfi_next) {
626 if (cur->xbfi_startblock >= bno)
627 break;
1da177e4 628 }
9e5987a7
DC		 629 if (prev)
630 prev->xbfi_next = new;
631 else
632 flist->xbf_first = new;
633 new->xbfi_next = cur;
634 flist->xbf_count++;
635}
7574aa92 636
9e5987a7
DC		 637/*
638 * Remove the entry "free" from the free item list. Prev points to the
639 * previous entry, unless "free" is the head of the list.
640 */
68988114 641void
9e5987a7
DC		 642xfs_bmap_del_free(
643 xfs_bmap_free_t *flist, /* free item list header */
644 xfs_bmap_free_item_t *prev, /* previous item on list, if any */
645 xfs_bmap_free_item_t *free) /* list item to be freed */
646{
647 if (prev)
648 prev->xbfi_next = free->xbfi_next;
649 else
650 flist->xbf_first = free->xbfi_next;
651 flist->xbf_count--;
652 kmem_zone_free(xfs_bmap_free_item_zone, free);
653}
654
9e5987a7
DC		 655/*
656 * Free up any items left in the list.
657 */
658void
659xfs_bmap_cancel(
660 xfs_bmap_free_t *flist) /* list of bmap_free_items */
661{
662 xfs_bmap_free_item_t *free; /* free list item */
663 xfs_bmap_free_item_t *next;
ec90c556 664
9e5987a7
DC		 665 if (flist->xbf_count == 0)
666 return;
667 ASSERT(flist->xbf_first != NULL);
668 for (free = flist->xbf_first; free; free = next) {
669 next = free->xbfi_next;
670 xfs_bmap_del_free(flist, NULL, free);
671 }
672 ASSERT(flist->xbf_count == 0);
673}
0b1b213f 674
9e5987a7
DC		 675/*
676 * Inode fork format manipulation functions
677 */
1da177e4 678
9e5987a7
DC		 679/*
680 * Transform a btree format file with only one leaf node, where the
681 * extents list will fit in the inode, into an extents format file.
682 * Since the file extents are already in-core, all we have to do is
683 * give up the space for the btree root and pitch the leaf block.
684 */
685STATIC int /* error */
686xfs_bmap_btree_to_extents(
687 xfs_trans_t *tp, /* transaction pointer */
688 xfs_inode_t *ip, /* incore inode pointer */
689 xfs_btree_cur_t *cur, /* btree cursor */
690 int *logflagsp, /* inode logging flags */
691 int whichfork) /* data or attr fork */
692{
693 /* REFERENCED */
694 struct xfs_btree_block *cblock;/* child btree block */
695 xfs_fsblock_t cbno; /* child block number */
696 xfs_buf_t *cbp; /* child block's buffer */
697 int error; /* error return value */
698 xfs_ifork_t *ifp; /* inode fork data */
699 xfs_mount_t *mp; /* mount point structure */
700 __be64 *pp; /* ptr to block address */
701 struct xfs_btree_block *rblock;/* root btree block */
1da177e4 702
9e5987a7
DC		 703 mp = ip->i_mount;
704 ifp = XFS_IFORK_PTR(ip, whichfork);
705 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
706 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
707 rblock = ifp->if_broot;
708 ASSERT(be16_to_cpu(rblock->bb_level) == 1);
709 ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
710 ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
711 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
712 cbno = be64_to_cpu(*pp);
713 *logflagsp = 0;
714#ifdef DEBUG
715 if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
716 return error;
717#endif
718 error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF,
719 &xfs_bmbt_buf_ops);
720 if (error)
721 return error;
722 cblock = XFS_BUF_TO_BLOCK(cbp);
723 if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
724 return error;
725 xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
726 ip->i_d.di_nblocks--;
727 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
728 xfs_trans_binval(tp, cbp);
729 if (cur->bc_bufs[0] == cbp)
730 cur->bc_bufs[0] = NULL;
731 xfs_iroot_realloc(ip, -1, whichfork);
732 ASSERT(ifp->if_broot == NULL);
733 ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
734 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
735 *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
736 return 0;
737}
0b1b213f 738
9e5987a7
DC		 739/*
740 * Convert an extents-format file into a btree-format file.
741 * The new file will have a root block (in the inode) and a single child block.
742 */
743STATIC int /* error */
744xfs_bmap_extents_to_btree(
745 xfs_trans_t *tp, /* transaction pointer */
746 xfs_inode_t *ip, /* incore inode pointer */
747 xfs_fsblock_t *firstblock, /* first-block-allocated */
748 xfs_bmap_free_t *flist, /* blocks freed in xaction */
749 xfs_btree_cur_t **curp, /* cursor returned to caller */
750 int wasdel, /* converting a delayed alloc */
751 int *logflagsp, /* inode logging flags */
752 int whichfork) /* data or attr fork */
753{
754 struct xfs_btree_block *ablock; /* allocated (child) bt block */
755 xfs_buf_t *abp; /* buffer for ablock */
756 xfs_alloc_arg_t args; /* allocation arguments */
757 xfs_bmbt_rec_t *arp; /* child record pointer */
758 struct xfs_btree_block *block; /* btree root block */
759 xfs_btree_cur_t *cur; /* bmap btree cursor */
760 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
761 int error; /* error return value */
762 xfs_extnum_t i, cnt; /* extent record index */
763 xfs_ifork_t *ifp; /* inode fork pointer */
764 xfs_bmbt_key_t *kp; /* root block key pointer */
765 xfs_mount_t *mp; /* mount structure */
766 xfs_extnum_t nextents; /* number of file extents */
767 xfs_bmbt_ptr_t *pp; /* root block address pointer */
1da177e4 768
ee1a47ab 769 mp = ip->i_mount;
9e5987a7
DC		 770 ifp = XFS_IFORK_PTR(ip, whichfork);
771 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
0b1b213f 772
9e5987a7
DC		 773 /*
774 * Make space in the inode incore.
775 */
776 xfs_iroot_realloc(ip, 1, whichfork);
777 ifp->if_flags |= XFS_IFBROOT;
ec90c556 778
9e5987a7
DC		 779 /*
780 * Fill in the root.
781 */
782 block = ifp->if_broot;
ee1a47ab
CH		 783 if (xfs_sb_version_hascrc(&mp->m_sb))
784 xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
785 XFS_BMAP_CRC_MAGIC, 1, 1, ip->i_ino,
786 XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
787 else
788 xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
789 XFS_BMAP_MAGIC, 1, 1, ip->i_ino,
790 XFS_BTREE_LONG_PTRS);
0b1b213f 791
9e5987a7
DC		 792 /*
793 * Need a cursor. Can't allocate until bb_level is filled in.
794 */
9e5987a7
DC		 795 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
796 cur->bc_private.b.firstblock = *firstblock;
797 cur->bc_private.b.flist = flist;
798 cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
799 /*
800 * Convert to a btree with two levels, one record in root.
801 */
802 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
803 memset(&args, 0, sizeof(args));
804 args.tp = tp;
805 args.mp = mp;
806 args.firstblock = *firstblock;
807 if (*firstblock == NULLFSBLOCK) {
808 args.type = XFS_ALLOCTYPE_START_BNO;
809 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
810 } else if (flist->xbf_low) {
811 args.type = XFS_ALLOCTYPE_START_BNO;
812 args.fsbno = *firstblock;
813 } else {
814 args.type = XFS_ALLOCTYPE_NEAR_BNO;
815 args.fsbno = *firstblock;
816 }
817 args.minlen = args.maxlen = args.prod = 1;
818 args.wasdel = wasdel;
819 *logflagsp = 0;
820 if ((error = xfs_alloc_vextent(&args))) {
821 xfs_iroot_realloc(ip, -1, whichfork);
822 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
823 return error;
824 }
825 /*
826 * Allocation can't fail, the space was reserved.
827 */
828 ASSERT(args.fsbno != NULLFSBLOCK);
829 ASSERT(*firstblock == NULLFSBLOCK ||
830 args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
831 (flist->xbf_low &&
832 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
833 *firstblock = cur->bc_private.b.firstblock = args.fsbno;
834 cur->bc_private.b.allocated++;
835 ip->i_d.di_nblocks++;
836 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
837 abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
838 /*
839 * Fill in the child block.
840 */
841 abp->b_ops = &xfs_bmbt_buf_ops;
842 ablock = XFS_BUF_TO_BLOCK(abp);
ee1a47ab
CH		 843 if (xfs_sb_version_hascrc(&mp->m_sb))
844 xfs_btree_init_block_int(mp, ablock, abp->b_bn,
845 XFS_BMAP_CRC_MAGIC, 0, 0, ip->i_ino,
846 XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
847 else
848 xfs_btree_init_block_int(mp, ablock, abp->b_bn,
849 XFS_BMAP_MAGIC, 0, 0, ip->i_ino,
850 XFS_BTREE_LONG_PTRS);
851
9e5987a7
DC		 852 arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
853 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
854 for (cnt = i = 0; i < nextents; i++) {
855 ep = xfs_iext_get_ext(ifp, i);
856 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
857 arp->l0 = cpu_to_be64(ep->l0);
858 arp->l1 = cpu_to_be64(ep->l1);
859 arp++; cnt++;
1da177e4 860 }
9e5987a7
DC		 861 }
862 ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
863 xfs_btree_set_numrecs(ablock, cnt);
1da177e4 864
9e5987a7
DC		 865 /*
866 * Fill in the root key and pointer.
867 */
868 kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
869 arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
870 kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
871 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
872 be16_to_cpu(block->bb_level)));
873 *pp = cpu_to_be64(args.fsbno);
ec90c556 874
9e5987a7
DC		 875 /*
876 * Do all this logging at the end so that
877 * the root is at the right level.
878 */
879 xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
880 xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
881 ASSERT(*curp == NULL);
882 *curp = cur;
883 *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
884 return 0;
885}
ec90c556 886
9e5987a7
DC		 887/*
888 * Convert a local file to an extents file.
889 * This code is out of bounds for data forks of regular files,
890 * since the file data needs to get logged so things will stay consistent.
891 * (The bmap-level manipulations are ok, though).
892 */
f3508bcd
DC		 893void
894xfs_bmap_local_to_extents_empty(
895 struct xfs_inode *ip,
896 int whichfork)
897{
898 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
899
900 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
901 ASSERT(ifp->if_bytes == 0);
902 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
903
6a9edd3d 904 xfs_bmap_forkoff_reset(ip, whichfork);
f3508bcd
DC		 905 ifp->if_flags &= ~XFS_IFINLINE;
906 ifp->if_flags |= XFS_IFEXTENTS;
907 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
908}
909
910
9e5987a7
DC		 911STATIC int /* error */
912xfs_bmap_local_to_extents(
913 xfs_trans_t *tp, /* transaction pointer */
914 xfs_inode_t *ip, /* incore inode pointer */
915 xfs_fsblock_t *firstblock, /* first block allocated in xaction */
916 xfs_extlen_t total, /* total blocks needed by transaction */
917 int *logflagsp, /* inode logging flags */
918 int whichfork,
ee1a47ab
CH		 919 void (*init_fn)(struct xfs_trans *tp,
920 struct xfs_buf *bp,
9e5987a7
DC		 921 struct xfs_inode *ip,
922 struct xfs_ifork *ifp))
923{
f3508bcd 924 int error = 0;
9e5987a7
DC		 925 int flags; /* logging flags returned */
926 xfs_ifork_t *ifp; /* inode fork pointer */
f3508bcd
DC		 927 xfs_alloc_arg_t args; /* allocation arguments */
928 xfs_buf_t *bp; /* buffer for extent block */
929 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
0b1b213f 930
9e5987a7
DC		 931 /*
932 * We don't want to deal with the case of keeping inode data inline yet.
933 * So sending the data fork of a regular inode is invalid.
934 */
935 ASSERT(!(S_ISREG(ip->i_d.di_mode) && whichfork == XFS_DATA_FORK));
936 ifp = XFS_IFORK_PTR(ip, whichfork);
937 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
f3508bcd
DC		 938
939 if (!ifp->if_bytes) {
940 xfs_bmap_local_to_extents_empty(ip, whichfork);
941 flags = XFS_ILOG_CORE;
942 goto done;
943 }
944
9e5987a7
DC		 945 flags = 0;
946 error = 0;
f3508bcd
DC		 947 ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) ==
948 XFS_IFINLINE);
949 memset(&args, 0, sizeof(args));
950 args.tp = tp;
951 args.mp = ip->i_mount;
952 args.firstblock = *firstblock;
953 /*
954 * Allocate a block. We know we need only one, since the
955 * file currently fits in an inode.
956 */
957 if (*firstblock == NULLFSBLOCK) {
958 args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
959 args.type = XFS_ALLOCTYPE_START_BNO;
9e5987a7 960 } else {
f3508bcd
DC		 961 args.fsbno = *firstblock;
962 args.type = XFS_ALLOCTYPE_NEAR_BNO;
9e5987a7 963 }
f3508bcd
DC		 964 args.total = total;
965 args.minlen = args.maxlen = args.prod = 1;
966 error = xfs_alloc_vextent(&args);
967 if (error)
968 goto done;
969
970 /* Can't fail, the space was reserved. */
971 ASSERT(args.fsbno != NULLFSBLOCK);
972 ASSERT(args.len == 1);
973 *firstblock = args.fsbno;
974 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
975
976 /* initialise the block and copy the data */
977 init_fn(tp, bp, ip, ifp);
978
979 /* account for the change in fork size and log everything */
980 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
981 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
982 xfs_bmap_local_to_extents_empty(ip, whichfork);
9e5987a7 983 flags |= XFS_ILOG_CORE;
f3508bcd
DC		 984
985 xfs_iext_add(ifp, 0, 1);
986 ep = xfs_iext_get_ext(ifp, 0);
987 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
988 trace_xfs_bmap_post_update(ip, 0,
989 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
990 _THIS_IP_);
991 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
992 ip->i_d.di_nblocks = 1;
993 xfs_trans_mod_dquot_byino(tp, ip,
994 XFS_TRANS_DQ_BCOUNT, 1L);
995 flags |= xfs_ilog_fext(whichfork);
996
9e5987a7
DC		 997done:
998 *logflagsp = flags;
999 return error;
1000}
ec90c556 1001
9e5987a7
DC		 1002/*
1003 * Called from xfs_bmap_add_attrfork to handle btree format files.
1004 */
1005STATIC int /* error */
1006xfs_bmap_add_attrfork_btree(
1007 xfs_trans_t *tp, /* transaction pointer */
1008 xfs_inode_t *ip, /* incore inode pointer */
1009 xfs_fsblock_t *firstblock, /* first block allocated */
1010 xfs_bmap_free_t *flist, /* blocks to free at commit */
1011 int *flags) /* inode logging flags */
1012{
1013 xfs_btree_cur_t *cur; /* btree cursor */
1014 int error; /* error return value */
1015 xfs_mount_t *mp; /* file system mount struct */
1016 int stat; /* newroot status */
ec90c556 1017
9e5987a7
DC		 1018 mp = ip->i_mount;
1019 if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
1020 *flags |= XFS_ILOG_DBROOT;
1021 else {
1022 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
1023 cur->bc_private.b.flist = flist;
1024 cur->bc_private.b.firstblock = *firstblock;
1025 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
1026 goto error0;
1027 /* must be at least one entry */
1028 XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
1029 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
1030 goto error0;
1031 if (stat == 0) {
1032 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
2451337d 1033 return -ENOSPC;
1da177e4 1034 }
9e5987a7
DC		 1035 *firstblock = cur->bc_private.b.firstblock;
1036 cur->bc_private.b.allocated = 0;
1037 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1da177e4 1038 }
9e5987a7
DC		 1039 return 0;
1040error0:
1041 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1042 return error;
1043}
a5bd606b 1044
9e5987a7
DC		 1045/*
1046 * Called from xfs_bmap_add_attrfork to handle extents format files.
1047 */
1048STATIC int /* error */
1049xfs_bmap_add_attrfork_extents(
1050 xfs_trans_t *tp, /* transaction pointer */
1051 xfs_inode_t *ip, /* incore inode pointer */
1052 xfs_fsblock_t *firstblock, /* first block allocated */
1053 xfs_bmap_free_t *flist, /* blocks to free at commit */
1054 int *flags) /* inode logging flags */
1055{
1056 xfs_btree_cur_t *cur; /* bmap btree cursor */
1057 int error; /* error return value */
a5bd606b 1058
9e5987a7
DC		 1059 if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
1060 return 0;
1061 cur = NULL;
1062 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
1063 flags, XFS_DATA_FORK);
a5bd606b
CH		 1064 if (cur) {
1065 cur->bc_private.b.allocated = 0;
9e5987a7
DC		 1066 xfs_btree_del_cursor(cur,
1067 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
a5bd606b 1068 }
1da177e4 1069 return error;
1da177e4
LT		 1070}
1071
9e5987a7
DC		 1072/*
1073 * Called from xfs_bmap_add_attrfork to handle local format files. Each
1074 * different data fork content type needs a different callout to do the
1075 * conversion. Some are basic and only require special block initialisation
1076 * callouts for the data formating, others (directories) are so specialised they
1077 * handle everything themselves.
1078 *
1079 * XXX (dgc): investigate whether directory conversion can use the generic
1080 * formatting callout. It should be possible - it's just a very complex
ee1a47ab 1081 * formatter.
9e5987a7
DC		 1082 */
1083STATIC int /* error */
1084xfs_bmap_add_attrfork_local(
1085 xfs_trans_t *tp, /* transaction pointer */
1086 xfs_inode_t *ip, /* incore inode pointer */
1087 xfs_fsblock_t *firstblock, /* first block allocated */
1088 xfs_bmap_free_t *flist, /* blocks to free at commit */
1089 int *flags) /* inode logging flags */
1090{
1091 xfs_da_args_t dargs; /* args for dir/attr code */
7574aa92 1092
9e5987a7
DC		 1093 if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
1094 return 0;
7574aa92 1095
9e5987a7
DC		 1096 if (S_ISDIR(ip->i_d.di_mode)) {
1097 memset(&dargs, 0, sizeof(dargs));
d6cf1305 1098 dargs.geo = ip->i_mount->m_dir_geo;
9e5987a7
DC		 1099 dargs.dp = ip;
1100 dargs.firstblock = firstblock;
1101 dargs.flist = flist;
d6cf1305 1102 dargs.total = dargs.geo->fsbcount;
9e5987a7
DC		 1103 dargs.whichfork = XFS_DATA_FORK;
1104 dargs.trans = tp;
1105 return xfs_dir2_sf_to_block(&dargs);
1da177e4 1106 }
7574aa92 1107
9e5987a7
DC		 1108 if (S_ISLNK(ip->i_d.di_mode))
1109 return xfs_bmap_local_to_extents(tp, ip, firstblock, 1,
1110 flags, XFS_DATA_FORK,
1111 xfs_symlink_local_to_remote);
7574aa92 1112
f3508bcd
DC		 1113 /* should only be called for types that support local format data */
1114 ASSERT(0);
2451337d 1115 return -EFSCORRUPTED;
9e5987a7 1116}
0b1b213f 1117
9e5987a7
DC		 1118/*
1119 * Convert inode from non-attributed to attributed.
1120 * Must not be in a transaction, ip must not be locked.
1121 */
1122int /* error code */
1123xfs_bmap_add_attrfork(
1124 xfs_inode_t *ip, /* incore inode pointer */
1125 int size, /* space new attribute needs */
1126 int rsvd) /* xact may use reserved blks */
1127{
1128 xfs_fsblock_t firstblock; /* 1st block/ag allocated */
1129 xfs_bmap_free_t flist; /* freed extent records */
1130 xfs_mount_t *mp; /* mount structure */
1131 xfs_trans_t *tp; /* transaction pointer */
1132 int blks; /* space reservation */
1133 int version = 1; /* superblock attr version */
1134 int committed; /* xaction was committed */
1135 int logflags; /* logging flags */
1136 int error; /* error return value */
9e3908e3 1137 int cancel_flags = 0;
0b1b213f 1138
9e5987a7 1139 ASSERT(XFS_IFORK_Q(ip) == 0);
1da177e4 1140
9e5987a7
DC		 1141 mp = ip->i_mount;
1142 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1143 tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
1144 blks = XFS_ADDAFORK_SPACE_RES(mp);
1145 if (rsvd)
1146 tp->t_flags |= XFS_TRANS_RESERVE;
3d3c8b52 1147 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_addafork, blks, 0);
9e3908e3
MT		 1148 if (error) {
1149 xfs_trans_cancel(tp, 0);
1150 return error;
1151 }
1152 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
9e5987a7
DC		 1153 xfs_ilock(ip, XFS_ILOCK_EXCL);
1154 error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
1155 XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
1156 XFS_QMOPT_RES_REGBLKS);
9e3908e3
MT		 1157 if (error)
1158 goto trans_cancel;
1159 cancel_flags |= XFS_TRANS_ABORT;
9e5987a7 1160 if (XFS_IFORK_Q(ip))
9e3908e3 1161 goto trans_cancel;
9e5987a7 1162 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
1da177e4 1163 /*
9e5987a7 1164 * For inodes coming from pre-6.2 filesystems.
1da177e4 1165 */
9e5987a7
DC		 1166 ASSERT(ip->i_d.di_aformat == 0);
1167 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1168 }
1169 ASSERT(ip->i_d.di_anextents == 0);
ec90c556 1170
9e3908e3 1171 xfs_trans_ijoin(tp, ip, 0);
9e5987a7 1172 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1da177e4 1173
9e5987a7
DC		 1174 switch (ip->i_d.di_format) {
1175 case XFS_DINODE_FMT_DEV:
1176 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1177 break;
1178 case XFS_DINODE_FMT_UUID:
1179 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
1180 break;
1181 case XFS_DINODE_FMT_LOCAL:
1182 case XFS_DINODE_FMT_EXTENTS:
1183 case XFS_DINODE_FMT_BTREE:
1184 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1185 if (!ip->i_d.di_forkoff)
1186 ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1187 else if (mp->m_flags & XFS_MOUNT_ATTR2)
1188 version = 2;
1da177e4 1189 break;
9e5987a7
DC		 1190 default:
1191 ASSERT(0);
2451337d 1192 error = -EINVAL;
9e3908e3 1193 goto trans_cancel;
9e5987a7 1194 }
1da177e4 1195
9e5987a7
DC		 1196 ASSERT(ip->i_afp == NULL);
1197 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
1198 ip->i_afp->if_flags = XFS_IFEXTENTS;
1199 logflags = 0;
1200 xfs_bmap_init(&flist, &firstblock);
1201 switch (ip->i_d.di_format) {
1202 case XFS_DINODE_FMT_LOCAL:
1203 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
1204 &logflags);
1205 break;
1206 case XFS_DINODE_FMT_EXTENTS:
1207 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
1208 &flist, &logflags);
1209 break;
1210 case XFS_DINODE_FMT_BTREE:
1211 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
1212 &logflags);
1213 break;
1214 default:
1215 error = 0;
1da177e4
LT		 1216 break;
1217 }
9e5987a7
DC		 1218 if (logflags)
1219 xfs_trans_log_inode(tp, ip, logflags);
1220 if (error)
9e3908e3 1221 goto bmap_cancel;
9e5987a7
DC		 1222 if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1223 (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
61e63ecb 1224 bool log_sb = false;
9e5987a7
DC		 1225
1226 spin_lock(&mp->m_sb_lock);
1227 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1228 xfs_sb_version_addattr(&mp->m_sb);
61e63ecb 1229 log_sb = true;
9e5987a7
DC		 1230 }
1231 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1232 xfs_sb_version_addattr2(&mp->m_sb);
61e63ecb 1233 log_sb = true;
9e5987a7 1234 }
4d11a402 1235 spin_unlock(&mp->m_sb_lock);
61e63ecb
DC		 1236 if (log_sb)
1237 xfs_log_sb(tp);
1da177e4 1238 }
9e5987a7
DC		 1239
1240 error = xfs_bmap_finish(&tp, &flist, &committed);
1241 if (error)
9e3908e3
MT		 1242 goto bmap_cancel;
1243 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1244 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1245 return error;
1246
1247bmap_cancel:
9e5987a7 1248 xfs_bmap_cancel(&flist);
9e3908e3
MT		 1249trans_cancel:
1250 xfs_trans_cancel(tp, cancel_flags);
9e5987a7 1251 xfs_iunlock(ip, XFS_ILOCK_EXCL);
9e5987a7 1252 return error;
1da177e4
LT		 1253}
1254
1255/*
9e5987a7 1256 * Internal and external extent tree search functions.
1da177e4 1257 */
a5bd606b 1258
9e5987a7
DC		 1259/*
1260 * Read in the extents to if_extents.
1261 * All inode fields are set up by caller, we just traverse the btree
1262 * and copy the records in. If the file system cannot contain unwritten
1263 * extents, the records are checked for no "state" flags.
1264 */
1265int /* error */
1266xfs_bmap_read_extents(
1267 xfs_trans_t *tp, /* transaction pointer */
1268 xfs_inode_t *ip, /* incore inode */
1269 int whichfork) /* data or attr fork */
1270{
1271 struct xfs_btree_block *block; /* current btree block */
1272 xfs_fsblock_t bno; /* block # of "block" */
1273 xfs_buf_t *bp; /* buffer for "block" */
1274 int error; /* error return value */
1275 xfs_exntfmt_t exntf; /* XFS_EXTFMT_NOSTATE, if checking */
1276 xfs_extnum_t i, j; /* index into the extents list */
1277 xfs_ifork_t *ifp; /* fork structure */
1278 int level; /* btree level, for checking */
1279 xfs_mount_t *mp; /* file system mount structure */
1280 __be64 *pp; /* pointer to block address */
1281 /* REFERENCED */
1282 xfs_extnum_t room; /* number of entries there's room for */
6ef35544 1283
9e5987a7
DC		 1284 bno = NULLFSBLOCK;
1285 mp = ip->i_mount;
1286 ifp = XFS_IFORK_PTR(ip, whichfork);
1287 exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
1288 XFS_EXTFMT_INODE(ip);
1289 block = ifp->if_broot;
1da177e4 1290 /*
9e5987a7 1291 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1da177e4 1292 */
9e5987a7
DC		 1293 level = be16_to_cpu(block->bb_level);
1294 ASSERT(level > 0);
1295 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
1296 bno = be64_to_cpu(*pp);
d5cf09ba 1297 ASSERT(bno != NULLFSBLOCK);
9e5987a7
DC		 1298 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
1299 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
1da177e4 1300 /*
9e5987a7
DC		 1301 * Go down the tree until leaf level is reached, following the first
1302 * pointer (leftmost) at each level.
1da177e4 1303 */
9e5987a7
DC		 1304 while (level-- > 0) {
1305 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1306 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1307 if (error)
1308 return error;
1309 block = XFS_BUF_TO_BLOCK(bp);
1310 XFS_WANT_CORRUPTED_GOTO(
1311 xfs_bmap_sanity_check(mp, bp, level),
1312 error0);
1313 if (level == 0)
1314 break;
1315 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
1316 bno = be64_to_cpu(*pp);
1317 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
1318 xfs_trans_brelse(tp, bp);
1da177e4
LT		 1319 }
1320 /*
9e5987a7 1321 * Here with bp and block set to the leftmost leaf node in the tree.
1da177e4 1322 */
9e5987a7
DC		 1323 room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1324 i = 0;
1da177e4 1325 /*
9e5987a7 1326 * Loop over all leaf nodes. Copy information to the extent records.
1da177e4 1327 */
9e5987a7
DC		 1328 for (;;) {
1329 xfs_bmbt_rec_t *frp;
1330 xfs_fsblock_t nextbno;
1331 xfs_extnum_t num_recs;
1332 xfs_extnum_t start;
0b1b213f 1333
9e5987a7
DC		 1334 num_recs = xfs_btree_get_numrecs(block);
1335 if (unlikely(i + num_recs > room)) {
1336 ASSERT(i + num_recs <= room);
1337 xfs_warn(ip->i_mount,
1338 "corrupt dinode %Lu, (btree extents).",
1339 (unsigned long long) ip->i_ino);
1340 XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
1341 XFS_ERRLEVEL_LOW, ip->i_mount, block);
1342 goto error0;
1da177e4 1343 }
9e5987a7
DC		 1344 XFS_WANT_CORRUPTED_GOTO(
1345 xfs_bmap_sanity_check(mp, bp, 0),
1346 error0);
1da177e4 1347 /*
9e5987a7 1348 * Read-ahead the next leaf block, if any.
1da177e4 1349 */
9e5987a7
DC		 1350 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
1351 if (nextbno != NULLFSBLOCK)
1352 xfs_btree_reada_bufl(mp, nextbno, 1,
1353 &xfs_bmbt_buf_ops);
1da177e4 1354 /*
9e5987a7 1355 * Copy records into the extent records.
1da177e4 1356 */
9e5987a7
DC		 1357 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1358 start = i;
1359 for (j = 0; j < num_recs; j++, i++, frp++) {
1360 xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
1361 trp->l0 = be64_to_cpu(frp->l0);
1362 trp->l1 = be64_to_cpu(frp->l1);
1da177e4 1363 }
9e5987a7
DC		 1364 if (exntf == XFS_EXTFMT_NOSTATE) {
1365 /*
1366 * Check all attribute bmap btree records and
1367 * any "older" data bmap btree records for a
1368 * set bit in the "extent flag" position.
1369 */
1370 if (unlikely(xfs_check_nostate_extents(ifp,
1371 start, num_recs))) {
1372 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
1373 XFS_ERRLEVEL_LOW,
1374 ip->i_mount);
1375 goto error0;
1376 }
1377 }
1378 xfs_trans_brelse(tp, bp);
1379 bno = nextbno;
1da177e4 1380 /*
9e5987a7 1381 * If we've reached the end, stop.
1da177e4 1382 */
9e5987a7
DC		 1383 if (bno == NULLFSBLOCK)
1384 break;
1385 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1386 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1387 if (error)
1388 return error;
1389 block = XFS_BUF_TO_BLOCK(bp);
1da177e4 1390 }
9e5987a7
DC		 1391 ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
1392 ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
1393 XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
1394 return 0;
1395error0:
1396 xfs_trans_brelse(tp, bp);
2451337d 1397 return -EFSCORRUPTED;
9e5987a7 1398}
a5bd606b 1399
a5bd606b 1400
9e5987a7
DC		 1401/*
1402 * Search the extent records for the entry containing block bno.
1403 * If bno lies in a hole, point to the next entry. If bno lies
1404 * past eof, *eofp will be set, and *prevp will contain the last
1405 * entry (null if none). Else, *lastxp will be set to the index
1406 * of the found entry; *gotp will contain the entry.
1407 */
1408STATIC xfs_bmbt_rec_host_t * /* pointer to found extent entry */
1409xfs_bmap_search_multi_extents(
1410 xfs_ifork_t *ifp, /* inode fork pointer */
1411 xfs_fileoff_t bno, /* block number searched for */
1412 int *eofp, /* out: end of file found */
1413 xfs_extnum_t *lastxp, /* out: last extent index */
1414 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
1415 xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
1416{
1417 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
1418 xfs_extnum_t lastx; /* last extent index */
a5bd606b 1419
9e5987a7
DC		 1420 /*
1421 * Initialize the extent entry structure to catch access to
1422 * uninitialized br_startblock field.
1423 */
1424 gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
1425 gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
1426 gotp->br_state = XFS_EXT_INVALID;
9e5987a7 1427 gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
9e5987a7 1428 prevp->br_startoff = NULLFILEOFF;
c6534249 1429
9e5987a7
DC		 1430 ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
1431 if (lastx > 0) {
1432 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
1433 }
1434 if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1435 xfs_bmbt_get_all(ep, gotp);
1436 *eofp = 0;
1437 } else {
1438 if (lastx > 0) {
1439 *gotp = *prevp;
1440 }
1441 *eofp = 1;
1442 ep = NULL;
1443 }
1444 *lastxp = lastx;
1445 return ep;
1da177e4
LT		 1446}
1447
dd9f438e 1448/*
9e5987a7
DC		 1449 * Search the extents list for the inode, for the extent containing bno.
1450 * If bno lies in a hole, point to the next entry. If bno lies past eof,
1451 * *eofp will be set, and *prevp will contain the last entry (null if none).
1452 * Else, *lastxp will be set to the index of the found
1453 * entry; *gotp will contain the entry.
dd9f438e 1454 */
9e5987a7
DC		 1455STATIC xfs_bmbt_rec_host_t * /* pointer to found extent entry */
1456xfs_bmap_search_extents(
1457 xfs_inode_t *ip, /* incore inode pointer */
1458 xfs_fileoff_t bno, /* block number searched for */
1459 int fork, /* data or attr fork */
1460 int *eofp, /* out: end of file found */
1461 xfs_extnum_t *lastxp, /* out: last extent index */
1462 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
1463 xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
dd9f438e 1464{
9e5987a7
DC		 1465 xfs_ifork_t *ifp; /* inode fork pointer */
1466 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
dd9f438e 1467
9e5987a7
DC		 1468 XFS_STATS_INC(xs_look_exlist);
1469 ifp = XFS_IFORK_PTR(ip, fork);
dd9f438e 1470
9e5987a7 1471 ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
dd9f438e 1472
9e5987a7
DC		 1473 if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
1474 !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
1475 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
1476 "Access to block zero in inode %llu "
1477 "start_block: %llx start_off: %llx "
08e96e1a 1478 "blkcnt: %llx extent-state: %x lastx: %x",
9e5987a7
DC		 1479 (unsigned long long)ip->i_ino,
1480 (unsigned long long)gotp->br_startblock,
1481 (unsigned long long)gotp->br_startoff,
1482 (unsigned long long)gotp->br_blockcount,
1483 gotp->br_state, *lastxp);
1484 *lastxp = NULLEXTNUM;
1485 *eofp = 1;
1486 return NULL;
dd9f438e 1487 }
9e5987a7
DC		 1488 return ep;
1489}
dd9f438e 1490
9e5987a7
DC		 1491/*
1492 * Returns the file-relative block number of the first unused block(s)
1493 * in the file with at least "len" logically contiguous blocks free.
1494 * This is the lowest-address hole if the file has holes, else the first block
1495 * past the end of file.
1496 * Return 0 if the file is currently local (in-inode).
1497 */
1498int /* error */
1499xfs_bmap_first_unused(
1500 xfs_trans_t *tp, /* transaction pointer */
1501 xfs_inode_t *ip, /* incore inode */
1502 xfs_extlen_t len, /* size of hole to find */
1503 xfs_fileoff_t *first_unused, /* unused block */
1504 int whichfork) /* data or attr fork */
1505{
1506 int error; /* error return value */
1507 int idx; /* extent record index */
1508 xfs_ifork_t *ifp; /* inode fork pointer */
1509 xfs_fileoff_t lastaddr; /* last block number seen */
1510 xfs_fileoff_t lowest; /* lowest useful block */
1511 xfs_fileoff_t max; /* starting useful block */
1512 xfs_fileoff_t off; /* offset for this block */
1513 xfs_extnum_t nextents; /* number of extent entries */
1514
1515 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
1516 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
1517 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1518 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1519 *first_unused = 0;
1520 return 0;
dd9f438e 1521 }
9e5987a7
DC		 1522 ifp = XFS_IFORK_PTR(ip, whichfork);
1523 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1524 (error = xfs_iread_extents(tp, ip, whichfork)))
1525 return error;
1526 lowest = *first_unused;
1527 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1528 for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
1529 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
1530 off = xfs_bmbt_get_startoff(ep);
1531 /*
1532 * See if the hole before this extent will work.
1533 */
1534 if (off >= lowest + len && off - max >= len) {
1535 *first_unused = max;
1536 return 0;
1537 }
1538 lastaddr = off + xfs_bmbt_get_blockcount(ep);
1539 max = XFS_FILEOFF_MAX(lastaddr, lowest);
dd9f438e 1540 }
9e5987a7
DC		 1541 *first_unused = max;
1542 return 0;
1543}
1544
1545/*
02bb4873 1546 * Returns the file-relative block number of the last block - 1 before
9e5987a7
DC		 1547 * last_block (input value) in the file.
1548 * This is not based on i_size, it is based on the extent records.
1549 * Returns 0 for local files, as they do not have extent records.
1550 */
1551int /* error */
1552xfs_bmap_last_before(
1553 xfs_trans_t *tp, /* transaction pointer */
1554 xfs_inode_t *ip, /* incore inode */
1555 xfs_fileoff_t *last_block, /* last block */
1556 int whichfork) /* data or attr fork */
1557{
1558 xfs_fileoff_t bno; /* input file offset */
1559 int eof; /* hit end of file */
1560 xfs_bmbt_rec_host_t *ep; /* pointer to last extent */
1561 int error; /* error return value */
1562 xfs_bmbt_irec_t got; /* current extent value */
1563 xfs_ifork_t *ifp; /* inode fork pointer */
1564 xfs_extnum_t lastx; /* last extent used */
1565 xfs_bmbt_irec_t prev; /* previous extent value */
1566
1567 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1568 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
1569 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
2451337d 1570 return -EIO;
9e5987a7
DC		 1571 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1572 *last_block = 0;
1573 return 0;
1574 }
1575 ifp = XFS_IFORK_PTR(ip, whichfork);
1576 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1577 (error = xfs_iread_extents(tp, ip, whichfork)))
1578 return error;
1579 bno = *last_block - 1;
1580 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
1581 &prev);
1582 if (eof || xfs_bmbt_get_startoff(ep) > bno) {
1583 if (prev.br_startoff == NULLFILEOFF)
1584 *last_block = 0;
dd9f438e 1585 else
9e5987a7 1586 *last_block = prev.br_startoff + prev.br_blockcount;
dd9f438e 1587 }
dd9f438e 1588 /*
9e5987a7 1589 * Otherwise *last_block is already the right answer.
dd9f438e 1590 */
9e5987a7
DC		 1591 return 0;
1592}
1593
68988114 1594int
9e5987a7
DC		 1595xfs_bmap_last_extent(
1596 struct xfs_trans *tp,
1597 struct xfs_inode *ip,
1598 int whichfork,
1599 struct xfs_bmbt_irec *rec,
1600 int *is_empty)
1601{
1602 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
1603 int error;
1604 int nextents;
1605
1606 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1607 error = xfs_iread_extents(tp, ip, whichfork);
1608 if (error)
1609 return error;
dd9f438e
NS		 1610 }
1611
9e5987a7
DC		 1612 nextents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
1613 if (nextents == 0) {
1614 *is_empty = 1;
1615 return 0;
1616 }
dd9f438e 1617
9e5987a7
DC		 1618 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
1619 *is_empty = 0;
dd9f438e
NS		 1620 return 0;
1621}
1622
9e5987a7
DC		 1623/*
1624 * Check the last inode extent to determine whether this allocation will result
1625 * in blocks being allocated at the end of the file. When we allocate new data
1626 * blocks at the end of the file which do not start at the previous data block,
1627 * we will try to align the new blocks at stripe unit boundaries.
1628 *
6e708bcf 1629 * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
9e5987a7
DC		 1630 * at, or past the EOF.
1631 */
1632STATIC int
1633xfs_bmap_isaeof(
1634 struct xfs_bmalloca *bma,
1635 int whichfork)
1da177e4 1636{
9e5987a7
DC		 1637 struct xfs_bmbt_irec rec;
1638 int is_empty;
1639 int error;
1da177e4 1640
9e5987a7
DC		 1641 bma->aeof = 0;
1642 error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1643 &is_empty);
6e708bcf 1644 if (error)
9e5987a7 1645 return error;
1da177e4 1646
6e708bcf
DC		 1647 if (is_empty) {
1648 bma->aeof = 1;
1649 return 0;
1650 }
1651
1da177e4 1652 /*
9e5987a7
DC		 1653 * Check if we are allocation or past the last extent, or at least into
1654 * the last delayed allocated extent.
1da177e4 1655 */
9e5987a7
DC		 1656 bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1657 (bma->offset >= rec.br_startoff &&
1658 isnullstartblock(rec.br_startblock));
1659 return 0;
1660}
1da177e4 1661
9e5987a7
DC		 1662/*
1663 * Returns the file-relative block number of the first block past eof in
1664 * the file. This is not based on i_size, it is based on the extent records.
1665 * Returns 0 for local files, as they do not have extent records.
1666 */
1667int
1668xfs_bmap_last_offset(
9e5987a7
DC		 1669 struct xfs_inode *ip,
1670 xfs_fileoff_t *last_block,
1671 int whichfork)
1672{
1673 struct xfs_bmbt_irec rec;
1674 int is_empty;
1675 int error;
04e99455 1676
9e5987a7 1677 *last_block = 0;
0892ccd6 1678
9e5987a7
DC		 1679 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
1680 return 0;
a365bdd5 1681
9e5987a7
DC		 1682 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1683 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
2451337d 1684 return -EIO;
a365bdd5 1685
9e5987a7
DC		 1686 error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1687 if (error || is_empty)
a365bdd5 1688 return error;
9e5987a7
DC		 1689
1690 *last_block = rec.br_startoff + rec.br_blockcount;
a365bdd5
NS		 1691 return 0;
1692}
1693
9e5987a7
DC		 1694/*
1695 * Returns whether the selected fork of the inode has exactly one
1696 * block or not. For the data fork we check this matches di_size,
1697 * implying the file's range is 0..bsize-1.
1698 */
1699int /* 1=>1 block, 0=>otherwise */
1700xfs_bmap_one_block(
1701 xfs_inode_t *ip, /* incore inode */
1702 int whichfork) /* data or attr fork */
c467c049 1703{
9e5987a7
DC		 1704 xfs_bmbt_rec_host_t *ep; /* ptr to fork's extent */
1705 xfs_ifork_t *ifp; /* inode fork pointer */
1706 int rval; /* return value */
1707 xfs_bmbt_irec_t s; /* internal version of extent */
c467c049 1708
9e5987a7
DC		 1709#ifndef DEBUG
1710 if (whichfork == XFS_DATA_FORK)
1711 return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
1712#endif /* !DEBUG */
1713 if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
1714 return 0;
1715 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1716 return 0;
1717 ifp = XFS_IFORK_PTR(ip, whichfork);
1718 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1719 ep = xfs_iext_get_ext(ifp, 0);
1720 xfs_bmbt_get_all(ep, &s);
1721 rval = s.br_startoff == 0 && s.br_blockcount == 1;
1722 if (rval && whichfork == XFS_DATA_FORK)
1723 ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
1724 return rval;
1725}
c467c049 1726
9e5987a7
DC		 1727/*
1728 * Extent tree manipulation functions used during allocation.
1729 */
c467c049 1730
9e5987a7
DC		 1731/*
1732 * Convert a delayed allocation to a real allocation.
1733 */
1734STATIC int /* error */
1735xfs_bmap_add_extent_delay_real(
1736 struct xfs_bmalloca *bma)
1737{
1738 struct xfs_bmbt_irec *new = &bma->got;
1739 int diff; /* temp value */
1740 xfs_bmbt_rec_host_t *ep; /* extent entry for idx */
1741 int error; /* error return value */
1742 int i; /* temp state */
1743 xfs_ifork_t *ifp; /* inode fork pointer */
1744 xfs_fileoff_t new_endoff; /* end offset of new entry */
1745 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
1746 /* left is 0, right is 1, prev is 2 */
1747 int rval=0; /* return value (logging flags) */
1748 int state = 0;/* state bits, accessed thru macros */
1749 xfs_filblks_t da_new; /* new count del alloc blocks used */
1750 xfs_filblks_t da_old; /* old count del alloc blocks used */
1751 xfs_filblks_t temp=0; /* value for da_new calculations */
1752 xfs_filblks_t temp2=0;/* value for da_new calculations */
1753 int tmp_rval; /* partial logging flags */
c467c049 1754
9e5987a7 1755 ifp = XFS_IFORK_PTR(bma->ip, XFS_DATA_FORK);
c467c049 1756
9e5987a7
DC		 1757 ASSERT(bma->idx >= 0);
1758 ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
1759 ASSERT(!isnullstartblock(new->br_startblock));
1760 ASSERT(!bma->cur ||
1761 (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
c467c049 1762
9e5987a7 1763 XFS_STATS_INC(xs_add_exlist);
c467c049 1764
9e5987a7
DC		 1765#define LEFT r[0]
1766#define RIGHT r[1]
1767#define PREV r[2]
c467c049
CH		 1768
1769 /*
9e5987a7 1770 * Set up a bunch of variables to make the tests simpler.
c467c049 1771 */
9e5987a7
DC		 1772 ep = xfs_iext_get_ext(ifp, bma->idx);
1773 xfs_bmbt_get_all(ep, &PREV);
1774 new_endoff = new->br_startoff + new->br_blockcount;
1775 ASSERT(PREV.br_startoff <= new->br_startoff);
1776 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1777
1778 da_old = startblockval(PREV.br_startblock);
1779 da_new = 0;
1780
c467c049 1781 /*
9e5987a7
DC		 1782 * Set flags determining what part of the previous delayed allocation
1783 * extent is being replaced by a real allocation.
c467c049 1784 */
9e5987a7
DC		 1785 if (PREV.br_startoff == new->br_startoff)
1786 state |= BMAP_LEFT_FILLING;
1787 if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1788 state |= BMAP_RIGHT_FILLING;
c467c049
CH		 1789
1790 /*
9e5987a7
DC		 1791 * Check and set flags if this segment has a left neighbor.
1792 * Don't set contiguous if the combined extent would be too large.
c467c049 1793 */
9e5987a7
DC		 1794 if (bma->idx > 0) {
1795 state |= BMAP_LEFT_VALID;
1796 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
a99ebf43 1797
9e5987a7
DC		 1798 if (isnullstartblock(LEFT.br_startblock))
1799 state |= BMAP_LEFT_DELAY;
a365bdd5 1800 }
a365bdd5 1801
9e5987a7
DC		 1802 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1803 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1804 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1805 LEFT.br_state == new->br_state &&
1806 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1807 state |= BMAP_LEFT_CONTIG;
a365bdd5 1808
1da177e4 1809 /*
9e5987a7
DC		 1810 * Check and set flags if this segment has a right neighbor.
1811 * Don't set contiguous if the combined extent would be too large.
1812 * Also check for all-three-contiguous being too large.
1da177e4 1813 */
9e5987a7
DC		 1814 if (bma->idx < bma->ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1815 state |= BMAP_RIGHT_VALID;
1816 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
14b064ce 1817
9e5987a7
DC		 1818 if (isnullstartblock(RIGHT.br_startblock))
1819 state |= BMAP_RIGHT_DELAY;
1da177e4 1820 }
9e5987a7
DC		 1821
1822 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1823 new_endoff == RIGHT.br_startoff &&
1824 new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1825 new->br_state == RIGHT.br_state &&
1826 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1827 ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1828 BMAP_RIGHT_FILLING)) !=
1829 (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1830 BMAP_RIGHT_FILLING) ||
1831 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1832 <= MAXEXTLEN))
1833 state |= BMAP_RIGHT_CONTIG;
1834
1835 error = 0;
1da177e4 1836 /*
9e5987a7 1837 * Switch out based on the FILLING and CONTIG state bits.
1da177e4 1838 */
9e5987a7
DC		 1839 switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1840 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1841 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1842 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1da177e4 1843 /*
9e5987a7
DC		 1844 * Filling in all of a previously delayed allocation extent.
1845 * The left and right neighbors are both contiguous with new.
1da177e4 1846 */
9e5987a7
DC		 1847 bma->idx--;
1848 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1849 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1850 LEFT.br_blockcount + PREV.br_blockcount +
1851 RIGHT.br_blockcount);
1852 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1853
1854 xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
1855 bma->ip->i_d.di_nextents--;
1856 if (bma->cur == NULL)
1857 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1858 else {
1859 rval = XFS_ILOG_CORE;
1860 error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1861 RIGHT.br_startblock,
1862 RIGHT.br_blockcount, &i);
1863 if (error)
1864 goto done;
1865 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1866 error = xfs_btree_delete(bma->cur, &i);
1867 if (error)
1868 goto done;
1869 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1870 error = xfs_btree_decrement(bma->cur, 0, &i);
1871 if (error)
1872 goto done;
1873 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1874 error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1875 LEFT.br_startblock,
1876 LEFT.br_blockcount +
1877 PREV.br_blockcount +
1878 RIGHT.br_blockcount, LEFT.br_state);
1879 if (error)
1880 goto done;
1881 }
1882 break;
1883
1884 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
a365bdd5 1885 /*
9e5987a7
DC		 1886 * Filling in all of a previously delayed allocation extent.
1887 * The left neighbor is contiguous, the right is not.
a365bdd5 1888 */
9e5987a7
DC		 1889 bma->idx--;
1890
1891 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1892 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1893 LEFT.br_blockcount + PREV.br_blockcount);
1894 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1895
1896 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1897 if (bma->cur == NULL)
1898 rval = XFS_ILOG_DEXT;
1899 else {
1900 rval = 0;
1901 error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1902 LEFT.br_startblock, LEFT.br_blockcount,
1903 &i);
1904 if (error)
1905 goto done;
1906 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1907 error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1908 LEFT.br_startblock,
1909 LEFT.br_blockcount +
1910 PREV.br_blockcount, LEFT.br_state);
1911 if (error)
1912 goto done;
1913 }
1914 break;
1915
1916 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
0937e0fd 1917 /*
9e5987a7
DC		 1918 * Filling in all of a previously delayed allocation extent.
1919 * The right neighbor is contiguous, the left is not.
0937e0fd 1920 */
9e5987a7
DC		 1921 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1922 xfs_bmbt_set_startblock(ep, new->br_startblock);
1923 xfs_bmbt_set_blockcount(ep,
1924 PREV.br_blockcount + RIGHT.br_blockcount);
1925 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
0937e0fd 1926
9e5987a7
DC		 1927 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1928 if (bma->cur == NULL)
1929 rval = XFS_ILOG_DEXT;
1930 else {
1931 rval = 0;
1932 error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1933 RIGHT.br_startblock,
1934 RIGHT.br_blockcount, &i);
1935 if (error)
1936 goto done;
1937 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1938 error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
1939 new->br_startblock,
1940 PREV.br_blockcount +
1941 RIGHT.br_blockcount, PREV.br_state);
1942 if (error)
1943 goto done;
1944 }
1945 break;
1946
1947 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
a365bdd5 1948 /*
9e5987a7
DC		 1949 * Filling in all of a previously delayed allocation extent.
1950 * Neither the left nor right neighbors are contiguous with
1951 * the new one.
a365bdd5 1952 */
9e5987a7
DC		 1953 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1954 xfs_bmbt_set_startblock(ep, new->br_startblock);
1955 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
a365bdd5 1956
9e5987a7
DC		 1957 bma->ip->i_d.di_nextents++;
1958 if (bma->cur == NULL)
1959 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1960 else {
1961 rval = XFS_ILOG_CORE;
1962 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1963 new->br_startblock, new->br_blockcount,
1964 &i);
1965 if (error)
1966 goto done;
1967 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1968 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1969 error = xfs_btree_insert(bma->cur, &i);
1970 if (error)
1971 goto done;
1972 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1973 }
1974 break;
1da177e4 1975
9e5987a7 1976 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1da177e4 1977 /*
9e5987a7
DC		 1978 * Filling in the first part of a previous delayed allocation.
1979 * The left neighbor is contiguous.
1da177e4 1980 */
9e5987a7
DC		 1981 trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1982 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
1983 LEFT.br_blockcount + new->br_blockcount);
1984 xfs_bmbt_set_startoff(ep,
1985 PREV.br_startoff + new->br_blockcount);
1986 trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1da177e4 1987
9e5987a7
DC		 1988 temp = PREV.br_blockcount - new->br_blockcount;
1989 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1990 xfs_bmbt_set_blockcount(ep, temp);
1991 if (bma->cur == NULL)
1992 rval = XFS_ILOG_DEXT;
1993 else {
1994 rval = 0;
1995 error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1996 LEFT.br_startblock, LEFT.br_blockcount,
1997 &i);
1998 if (error)
1999 goto done;
2000 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2001 error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
2002 LEFT.br_startblock,
2003 LEFT.br_blockcount +
2004 new->br_blockcount,
2005 LEFT.br_state);
f3ca8738 2006 if (error)
1da177e4 2007 goto done;
1da177e4 2008 }
9e5987a7
DC		 2009 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2010 startblockval(PREV.br_startblock));
2011 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2012 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2013
2014 bma->idx--;
2015 break;
2016
2017 case BMAP_LEFT_FILLING:
1da177e4 2018 /*
9e5987a7
DC		 2019 * Filling in the first part of a previous delayed allocation.
2020 * The left neighbor is not contiguous.
1da177e4 2021 */
9e5987a7
DC		 2022 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2023 xfs_bmbt_set_startoff(ep, new_endoff);
2024 temp = PREV.br_blockcount - new->br_blockcount;
2025 xfs_bmbt_set_blockcount(ep, temp);
2026 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
2027 bma->ip->i_d.di_nextents++;
2028 if (bma->cur == NULL)
2029 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1da177e4 2030 else {
9e5987a7
DC		 2031 rval = XFS_ILOG_CORE;
2032 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2033 new->br_startblock, new->br_blockcount,
2034 &i);
2035 if (error)
2036 goto done;
2037 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2038 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2039 error = xfs_btree_insert(bma->cur, &i);
2040 if (error)
1da177e4 2041 goto done;
6bd8fc8a 2042 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1da177e4 2043 }
233eebb9 2044
9e5987a7
DC		 2045 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2046 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2047 bma->firstblock, bma->flist,
2048 &bma->cur, 1, &tmp_rval, XFS_DATA_FORK);
2049 rval |= tmp_rval;
2050 if (error)
2051 goto done;
1da177e4 2052 }
9e5987a7
DC		 2053 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2054 startblockval(PREV.br_startblock) -
2055 (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2056 ep = xfs_iext_get_ext(ifp, bma->idx + 1);
2057 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2058 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
1da177e4
LT		 2059 break;
2060
9e5987a7 2061 case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1da177e4 2062 /*
9e5987a7
DC		 2063 * Filling in the last part of a previous delayed allocation.
2064 * The right neighbor is contiguous with the new allocation.
1da177e4 2065 */
9e5987a7
DC		 2066 temp = PREV.br_blockcount - new->br_blockcount;
2067 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
1da177e4 2068 xfs_bmbt_set_blockcount(ep, temp);
9e5987a7
DC		 2069 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
2070 new->br_startoff, new->br_startblock,
2071 new->br_blockcount + RIGHT.br_blockcount,
2072 RIGHT.br_state);
2073 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2074 if (bma->cur == NULL)
2075 rval = XFS_ILOG_DEXT;
2076 else {
2077 rval = 0;
2078 error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
2079 RIGHT.br_startblock,
2080 RIGHT.br_blockcount, &i);
2081 if (error)
2082 goto done;
2083 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2084 error = xfs_bmbt_update(bma->cur, new->br_startoff,
2085 new->br_startblock,
2086 new->br_blockcount +
2087 RIGHT.br_blockcount,
2088 RIGHT.br_state);
2089 if (error)
2090 goto done;
1da177e4 2091 }
9e5987a7
DC		 2092
2093 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2094 startblockval(PREV.br_startblock));
2095 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2096 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2097 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2098
2099 bma->idx++;
1da177e4
LT		 2100 break;
2101
9e5987a7 2102 case BMAP_RIGHT_FILLING:
1da177e4 2103 /*
9e5987a7
DC		 2104 * Filling in the last part of a previous delayed allocation.
2105 * The right neighbor is not contiguous.
1da177e4 2106 */
9e5987a7
DC		 2107 temp = PREV.br_blockcount - new->br_blockcount;
2108 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1da177e4 2109 xfs_bmbt_set_blockcount(ep, temp);
9e5987a7
DC		 2110 xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
2111 bma->ip->i_d.di_nextents++;
2112 if (bma->cur == NULL)
2113 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2114 else {
2115 rval = XFS_ILOG_CORE;
2116 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2117 new->br_startblock, new->br_blockcount,
2118 &i);
2119 if (error)
2120 goto done;
2121 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2122 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2123 error = xfs_btree_insert(bma->cur, &i);
2124 if (error)
2125 goto done;
2126 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1da177e4 2127 }
9e5987a7
DC		 2128
2129 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2130 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2131 bma->firstblock, bma->flist, &bma->cur, 1,
2132 &tmp_rval, XFS_DATA_FORK);
2133 rval |= tmp_rval;
2134 if (error)
2135 goto done;
1da177e4 2136 }
9e5987a7
DC		 2137 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2138 startblockval(PREV.br_startblock) -
2139 (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2140 ep = xfs_iext_get_ext(ifp, bma->idx);
2141 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2142 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2143
2144 bma->idx++;
1da177e4
LT		 2145 break;
2146
2147 case 0:
2148 /*
9e5987a7
DC		 2149 * Filling in the middle part of a previous delayed allocation.
2150 * Contiguity is impossible here.
2151 * This case is avoided almost all the time.
2152 *
2153 * We start with a delayed allocation:
2154 *
2155 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
2156 * PREV @ idx
2157 *
2158 * and we are allocating:
2159 * +rrrrrrrrrrrrrrrrr+
2160 * new
2161 *
2162 * and we set it up for insertion as:
2163 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
2164 * new
2165 * PREV @ idx LEFT RIGHT
2166 * inserted at idx + 1
1da177e4 2167 */
9e5987a7
DC		 2168 temp = new->br_startoff - PREV.br_startoff;
2169 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
2170 trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
2171 xfs_bmbt_set_blockcount(ep, temp); /* truncate PREV */
2172 LEFT = *new;
2173 RIGHT.br_state = PREV.br_state;
2174 RIGHT.br_startblock = nullstartblock(
2175 (int)xfs_bmap_worst_indlen(bma->ip, temp2));
2176 RIGHT.br_startoff = new_endoff;
2177 RIGHT.br_blockcount = temp2;
2178 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
2179 xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
2180 bma->ip->i_d.di_nextents++;
2181 if (bma->cur == NULL)
2182 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2183 else {
2184 rval = XFS_ILOG_CORE;
2185 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2186 new->br_startblock, new->br_blockcount,
2187 &i);
2188 if (error)
2189 goto done;
2190 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2191 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2192 error = xfs_btree_insert(bma->cur, &i);
2193 if (error)
2194 goto done;
2195 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1da177e4 2196 }
1da177e4 2197
9e5987a7
DC		 2198 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2199 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2200 bma->firstblock, bma->flist, &bma->cur,
2201 1, &tmp_rval, XFS_DATA_FORK);
2202 rval |= tmp_rval;
2203 if (error)
2204 goto done;
2205 }
2206 temp = xfs_bmap_worst_indlen(bma->ip, temp);
2207 temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
2208 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
2209 (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2210 if (diff > 0) {
2211 error = xfs_icsb_modify_counters(bma->ip->i_mount,
2212 XFS_SBS_FDBLOCKS,
2213 -((int64_t)diff), 0);
2214 ASSERT(!error);
2215 if (error)
2216 goto done;
2217 }
2218
2219 ep = xfs_iext_get_ext(ifp, bma->idx);
2220 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2221 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2222 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2223 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
2224 nullstartblock((int)temp2));
2225 trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2226
2227 bma->idx++;
2228 da_new = temp + temp2;
2229 break;
2230
2231 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2232 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2233 case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2234 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2235 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2236 case BMAP_LEFT_CONTIG:
2237 case BMAP_RIGHT_CONTIG:
2238 /*
2239 * These cases are all impossible.
2240 */
2241 ASSERT(0);
2242 }
2243
2244 /* convert to a btree if necessary */
2245 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2246 int tmp_logflags; /* partial log flag return val */
2247
2248 ASSERT(bma->cur == NULL);
2249 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2250 bma->firstblock, bma->flist, &bma->cur,
2251 da_old > 0, &tmp_logflags, XFS_DATA_FORK);
2252 bma->logflags |= tmp_logflags;
2253 if (error)
2254 goto done;
2255 }
2256
2257 /* adjust for changes in reserved delayed indirect blocks */
2258 if (da_old || da_new) {
2259 temp = da_new;
2260 if (bma->cur)
2261 temp += bma->cur->bc_private.b.allocated;
2262 ASSERT(temp <= da_old);
2263 if (temp < da_old)
2264 xfs_icsb_modify_counters(bma->ip->i_mount,
2265 XFS_SBS_FDBLOCKS,
2266 (int64_t)(da_old - temp), 0);
96540c78 2267 }
9e5987a7
DC		 2268
2269 /* clear out the allocated field, done with it now in any case. */
2270 if (bma->cur)
2271 bma->cur->bc_private.b.allocated = 0;
2272
2273 xfs_bmap_check_leaf_extents(bma->cur, bma->ip, XFS_DATA_FORK);
1da177e4 2274done:
9e5987a7 2275 bma->logflags |= rval;
1da177e4 2276 return error;
9e5987a7
DC		 2277#undef LEFT
2278#undef RIGHT
2279#undef PREV
1da177e4
LT		 2280}
2281
2282/*
9e5987a7 2283 * Convert an unwritten allocation to a real allocation or vice versa.
1da177e4 2284 */
9e5987a7
DC		 2285STATIC int /* error */
2286xfs_bmap_add_extent_unwritten_real(
2287 struct xfs_trans *tp,
2288 xfs_inode_t *ip, /* incore inode pointer */
2289 xfs_extnum_t *idx, /* extent number to update/insert */
2290 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
2291 xfs_bmbt_irec_t *new, /* new data to add to file extents */
2292 xfs_fsblock_t *first, /* pointer to firstblock variable */
2293 xfs_bmap_free_t *flist, /* list of extents to be freed */
2294 int *logflagsp) /* inode logging flags */
1da177e4 2295{
9e5987a7
DC		 2296 xfs_btree_cur_t *cur; /* btree cursor */
2297 xfs_bmbt_rec_host_t *ep; /* extent entry for idx */
2298 int error; /* error return value */
2299 int i; /* temp state */
2300 xfs_ifork_t *ifp; /* inode fork pointer */
2301 xfs_fileoff_t new_endoff; /* end offset of new entry */
2302 xfs_exntst_t newext; /* new extent state */
2303 xfs_exntst_t oldext; /* old extent state */
2304 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
2305 /* left is 0, right is 1, prev is 2 */
2306 int rval=0; /* return value (logging flags) */
2307 int state = 0;/* state bits, accessed thru macros */
1da177e4 2308
9e5987a7 2309 *logflagsp = 0;
1da177e4 2310
9e5987a7
DC		 2311 cur = *curp;
2312 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
8096b1eb 2313
9e5987a7
DC		 2314 ASSERT(*idx >= 0);
2315 ASSERT(*idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2316 ASSERT(!isnullstartblock(new->br_startblock));
2317
2318 XFS_STATS_INC(xs_add_exlist);
2319
2320#define LEFT r[0]
2321#define RIGHT r[1]
2322#define PREV r[2]
7cc95a82 2323
1da177e4 2324 /*
9e5987a7 2325 * Set up a bunch of variables to make the tests simpler.
1da177e4 2326 */
9e5987a7
DC		 2327 error = 0;
2328 ep = xfs_iext_get_ext(ifp, *idx);
2329 xfs_bmbt_get_all(ep, &PREV);
2330 newext = new->br_state;
2331 oldext = (newext == XFS_EXT_UNWRITTEN) ?
2332 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
2333 ASSERT(PREV.br_state == oldext);
2334 new_endoff = new->br_startoff + new->br_blockcount;
2335 ASSERT(PREV.br_startoff <= new->br_startoff);
2336 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
7cc95a82 2337
1da177e4 2338 /*
9e5987a7
DC		 2339 * Set flags determining what part of the previous oldext allocation
2340 * extent is being replaced by a newext allocation.
1da177e4 2341 */
9e5987a7
DC		 2342 if (PREV.br_startoff == new->br_startoff)
2343 state |= BMAP_LEFT_FILLING;
2344 if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2345 state |= BMAP_RIGHT_FILLING;
2346
1da177e4 2347 /*
9e5987a7
DC		 2348 * Check and set flags if this segment has a left neighbor.
2349 * Don't set contiguous if the combined extent would be too large.
1da177e4 2350 */
9e5987a7
DC		 2351 if (*idx > 0) {
2352 state |= BMAP_LEFT_VALID;
2353 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
2354
2355 if (isnullstartblock(LEFT.br_startblock))
2356 state |= BMAP_LEFT_DELAY;
1da177e4 2357 }
9e5987a7
DC		 2358
2359 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2360 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2361 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2362 LEFT.br_state == newext &&
2363 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2364 state |= BMAP_LEFT_CONTIG;
2365
1da177e4 2366 /*
9e5987a7
DC		 2367 * Check and set flags if this segment has a right neighbor.
2368 * Don't set contiguous if the combined extent would be too large.
2369 * Also check for all-three-contiguous being too large.
1da177e4 2370 */
9e5987a7
DC		 2371 if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
2372 state |= BMAP_RIGHT_VALID;
2373 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
2374 if (isnullstartblock(RIGHT.br_startblock))
2375 state |= BMAP_RIGHT_DELAY;
1da177e4 2376 }
7cc95a82 2377
9e5987a7
DC		 2378 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2379 new_endoff == RIGHT.br_startoff &&
2380 new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2381 newext == RIGHT.br_state &&
2382 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2383 ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2384 BMAP_RIGHT_FILLING)) !=
2385 (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2386 BMAP_RIGHT_FILLING) ||
2387 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2388 <= MAXEXTLEN))
2389 state |= BMAP_RIGHT_CONTIG;
136341b4 2390
1da177e4 2391 /*
9e5987a7 2392 * Switch out based on the FILLING and CONTIG state bits.
1da177e4 2393 */
9e5987a7
DC		 2394 switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2395 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2396 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2397 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2398 /*
2399 * Setting all of a previous oldext extent to newext.
2400 * The left and right neighbors are both contiguous with new.
2401 */
2402 --*idx;
1a5902c5 2403
9e5987a7
DC		 2404 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2405 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2406 LEFT.br_blockcount + PREV.br_blockcount +
2407 RIGHT.br_blockcount);
2408 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1a5902c5 2409
9e5987a7
DC		 2410 xfs_iext_remove(ip, *idx + 1, 2, state);
2411 ip->i_d.di_nextents -= 2;
2412 if (cur == NULL)
2413 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2414 else {
2415 rval = XFS_ILOG_CORE;
2416 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2417 RIGHT.br_startblock,
2418 RIGHT.br_blockcount, &i)))
2419 goto done;
2420 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2421 if ((error = xfs_btree_delete(cur, &i)))
2422 goto done;
2423 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2424 if ((error = xfs_btree_decrement(cur, 0, &i)))
2425 goto done;
2426 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2427 if ((error = xfs_btree_delete(cur, &i)))
2428 goto done;
2429 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2430 if ((error = xfs_btree_decrement(cur, 0, &i)))
2431 goto done;
2432 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2433 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2434 LEFT.br_startblock,
2435 LEFT.br_blockcount + PREV.br_blockcount +
2436 RIGHT.br_blockcount, LEFT.br_state)))
2437 goto done;
2438 }
2439 break;
1a5902c5 2440
9e5987a7
DC		 2441 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2442 /*
2443 * Setting all of a previous oldext extent to newext.
2444 * The left neighbor is contiguous, the right is not.
2445 */
2446 --*idx;
d8cc890d 2447
9e5987a7
DC		 2448 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2449 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2450 LEFT.br_blockcount + PREV.br_blockcount);
2451 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1da177e4 2452
9e5987a7
DC		 2453 xfs_iext_remove(ip, *idx + 1, 1, state);
2454 ip->i_d.di_nextents--;
2455 if (cur == NULL)
2456 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2457 else {
2458 rval = XFS_ILOG_CORE;
2459 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2460 PREV.br_startblock, PREV.br_blockcount,
2461 &i)))
2462 goto done;
2463 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2464 if ((error = xfs_btree_delete(cur, &i)))
2465 goto done;
2466 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2467 if ((error = xfs_btree_decrement(cur, 0, &i)))
2468 goto done;
2469 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2470 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2471 LEFT.br_startblock,
2472 LEFT.br_blockcount + PREV.br_blockcount,
2473 LEFT.br_state)))
2474 goto done;
2475 }
2476 break;
1da177e4 2477
9e5987a7 2478 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1da177e4 2479 /*
9e5987a7
DC		 2480 * Setting all of a previous oldext extent to newext.
2481 * The right neighbor is contiguous, the left is not.
1da177e4 2482 */
9e5987a7
DC		 2483 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2484 xfs_bmbt_set_blockcount(ep,
2485 PREV.br_blockcount + RIGHT.br_blockcount);
2486 xfs_bmbt_set_state(ep, newext);
2487 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2488 xfs_iext_remove(ip, *idx + 1, 1, state);
2489 ip->i_d.di_nextents--;
2490 if (cur == NULL)
2491 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2492 else {
2493 rval = XFS_ILOG_CORE;
2494 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2495 RIGHT.br_startblock,
2496 RIGHT.br_blockcount, &i)))
2497 goto done;
2498 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2499 if ((error = xfs_btree_delete(cur, &i)))
2500 goto done;
2501 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2502 if ((error = xfs_btree_decrement(cur, 0, &i)))
2503 goto done;
2504 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2505 if ((error = xfs_bmbt_update(cur, new->br_startoff,
2506 new->br_startblock,
2507 new->br_blockcount + RIGHT.br_blockcount,
2508 newext)))
2509 goto done;
1da177e4 2510 }
9e5987a7 2511 break;
1e82379b 2512
9e5987a7
DC		 2513 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2514 /*
2515 * Setting all of a previous oldext extent to newext.
2516 * Neither the left nor right neighbors are contiguous with
2517 * the new one.
2518 */
2519 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2520 xfs_bmbt_set_state(ep, newext);
2521 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1e82379b 2522
9e5987a7
DC		 2523 if (cur == NULL)
2524 rval = XFS_ILOG_DEXT;
2525 else {
2526 rval = 0;
2527 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2528 new->br_startblock, new->br_blockcount,
2529 &i)))
2530 goto done;
2531 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2532 if ((error = xfs_bmbt_update(cur, new->br_startoff,
2533 new->br_startblock, new->br_blockcount,
2534 newext)))
2535 goto done;
2536 }
2537 break;
1e82379b 2538
9e5987a7
DC		 2539 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2540 /*
2541 * Setting the first part of a previous oldext extent to newext.
2542 * The left neighbor is contiguous.
2543 */
2544 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
2545 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
2546 LEFT.br_blockcount + new->br_blockcount);
2547 xfs_bmbt_set_startoff(ep,
2548 PREV.br_startoff + new->br_blockcount);
2549 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
1da177e4 2550
9e5987a7
DC		 2551 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2552 xfs_bmbt_set_startblock(ep,
2553 new->br_startblock + new->br_blockcount);
2554 xfs_bmbt_set_blockcount(ep,
2555 PREV.br_blockcount - new->br_blockcount);
2556 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
0293ce3a 2557
9e5987a7 2558 --*idx;
8867bc9b 2559
9e5987a7
DC		 2560 if (cur == NULL)
2561 rval = XFS_ILOG_DEXT;
2562 else {
2563 rval = 0;
2564 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2565 PREV.br_startblock, PREV.br_blockcount,
2566 &i)))
2567 goto done;
2568 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2569 if ((error = xfs_bmbt_update(cur,
2570 PREV.br_startoff + new->br_blockcount,
2571 PREV.br_startblock + new->br_blockcount,
2572 PREV.br_blockcount - new->br_blockcount,
2573 oldext)))
2574 goto done;
2575 if ((error = xfs_btree_decrement(cur, 0, &i)))
2576 goto done;
2577 error = xfs_bmbt_update(cur, LEFT.br_startoff,
2578 LEFT.br_startblock,
2579 LEFT.br_blockcount + new->br_blockcount,
2580 LEFT.br_state);
2581 if (error)
2582 goto done;
8867bc9b 2583 }
9e5987a7 2584 break;
0293ce3a 2585
9e5987a7
DC		 2586 case BMAP_LEFT_FILLING:
2587 /*
2588 * Setting the first part of a previous oldext extent to newext.
2589 * The left neighbor is not contiguous.
2590 */
2591 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2592 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
2593 xfs_bmbt_set_startoff(ep, new_endoff);
2594 xfs_bmbt_set_blockcount(ep,
2595 PREV.br_blockcount - new->br_blockcount);
2596 xfs_bmbt_set_startblock(ep,
2597 new->br_startblock + new->br_blockcount);
2598 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1da177e4 2599
9e5987a7
DC		 2600 xfs_iext_insert(ip, *idx, 1, new, state);
2601 ip->i_d.di_nextents++;
2602 if (cur == NULL)
2603 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2604 else {
2605 rval = XFS_ILOG_CORE;
2606 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2607 PREV.br_startblock, PREV.br_blockcount,
2608 &i)))
2609 goto done;
2610 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2611 if ((error = xfs_bmbt_update(cur,
2612 PREV.br_startoff + new->br_blockcount,
2613 PREV.br_startblock + new->br_blockcount,
2614 PREV.br_blockcount - new->br_blockcount,
2615 oldext)))
2616 goto done;
2617 cur->bc_rec.b = *new;
2618 if ((error = xfs_btree_insert(cur, &i)))
2619 goto done;
2620 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2621 }
2622 break;
1da177e4 2623
9e5987a7
DC		 2624 case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2625 /*
2626 * Setting the last part of a previous oldext extent to newext.
2627 * The right neighbor is contiguous with the new allocation.
2628 */
2629 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2630 xfs_bmbt_set_blockcount(ep,
2631 PREV.br_blockcount - new->br_blockcount);
2632 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
0293ce3a 2633
9e5987a7 2634 ++*idx;
1da177e4 2635
9e5987a7
DC		 2636 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2637 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2638 new->br_startoff, new->br_startblock,
2639 new->br_blockcount + RIGHT.br_blockcount, newext);
2640 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1da177e4 2641
9e5987a7
DC		 2642 if (cur == NULL)
2643 rval = XFS_ILOG_DEXT;
2644 else {
2645 rval = 0;
2646 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2647 PREV.br_startblock,
2648 PREV.br_blockcount, &i)))
2649 goto done;
2650 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2651 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2652 PREV.br_startblock,
2653 PREV.br_blockcount - new->br_blockcount,
2654 oldext)))
2655 goto done;
2656 if ((error = xfs_btree_increment(cur, 0, &i)))
2657 goto done;
2658 if ((error = xfs_bmbt_update(cur, new->br_startoff,
2659 new->br_startblock,
2660 new->br_blockcount + RIGHT.br_blockcount,
2661 newext)))
2662 goto done;
2663 }
2664 break;
1da177e4 2665
9e5987a7
DC		 2666 case BMAP_RIGHT_FILLING:
2667 /*
2668 * Setting the last part of a previous oldext extent to newext.
2669 * The right neighbor is not contiguous.
2670 */
2671 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2672 xfs_bmbt_set_blockcount(ep,
2673 PREV.br_blockcount - new->br_blockcount);
2674 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1da177e4 2675
9e5987a7
DC		 2676 ++*idx;
2677 xfs_iext_insert(ip, *idx, 1, new, state);
d8cc890d 2678
9e5987a7
DC		 2679 ip->i_d.di_nextents++;
2680 if (cur == NULL)
2681 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2682 else {
2683 rval = XFS_ILOG_CORE;
2684 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2685 PREV.br_startblock, PREV.br_blockcount,
2686 &i)))
2687 goto done;
2688 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2689 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2690 PREV.br_startblock,
2691 PREV.br_blockcount - new->br_blockcount,
2692 oldext)))
2693 goto done;
2694 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2695 new->br_startblock, new->br_blockcount,
2696 &i)))
2697 goto done;
2698 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2699 cur->bc_rec.b.br_state = XFS_EXT_NORM;
2700 if ((error = xfs_btree_insert(cur, &i)))
2701 goto done;
2702 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2703 }
2704 break;
2705
2706 case 0:
1da177e4 2707 /*
9e5987a7
DC		 2708 * Setting the middle part of a previous oldext extent to
2709 * newext. Contiguity is impossible here.
2710 * One extent becomes three extents.
1da177e4 2711 */
9e5987a7
DC		 2712 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2713 xfs_bmbt_set_blockcount(ep,
2714 new->br_startoff - PREV.br_startoff);
2715 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
898621d5 2716
9e5987a7
DC		 2717 r[0] = *new;
2718 r[1].br_startoff = new_endoff;
2719 r[1].br_blockcount =
2720 PREV.br_startoff + PREV.br_blockcount - new_endoff;
2721 r[1].br_startblock = new->br_startblock + new->br_blockcount;
2722 r[1].br_state = oldext;
898621d5 2723
9e5987a7
DC		 2724 ++*idx;
2725 xfs_iext_insert(ip, *idx, 2, &r[0], state);
2726
2727 ip->i_d.di_nextents += 2;
2728 if (cur == NULL)
2729 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2730 else {
2731 rval = XFS_ILOG_CORE;
2732 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2733 PREV.br_startblock, PREV.br_blockcount,
2734 &i)))
2735 goto done;
2736 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2737 /* new right extent - oldext */
2738 if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
2739 r[1].br_startblock, r[1].br_blockcount,
2740 r[1].br_state)))
2741 goto done;
2742 /* new left extent - oldext */
2743 cur->bc_rec.b = PREV;
2744 cur->bc_rec.b.br_blockcount =
2745 new->br_startoff - PREV.br_startoff;
2746 if ((error = xfs_btree_insert(cur, &i)))
2747 goto done;
2748 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2749 /*
2750 * Reset the cursor to the position of the new extent
2751 * we are about to insert as we can't trust it after
2752 * the previous insert.
2753 */
2754 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2755 new->br_startblock, new->br_blockcount,
2756 &i)))
2757 goto done;
2758 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2759 /* new middle extent - newext */
2760 cur->bc_rec.b.br_state = new->br_state;
2761 if ((error = xfs_btree_insert(cur, &i)))
2762 goto done;
2763 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2764 }
1da177e4 2765 break;
9e5987a7
DC		 2766
2767 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2768 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2769 case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2770 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2771 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2772 case BMAP_LEFT_CONTIG:
2773 case BMAP_RIGHT_CONTIG:
2774 /*
2775 * These cases are all impossible.
2776 */
1da177e4 2777 ASSERT(0);
1da177e4 2778 }
8096b1eb 2779
9e5987a7
DC		 2780 /* convert to a btree if necessary */
2781 if (xfs_bmap_needs_btree(ip, XFS_DATA_FORK)) {
2782 int tmp_logflags; /* partial log flag return val */
2783
2784 ASSERT(cur == NULL);
2785 error = xfs_bmap_extents_to_btree(tp, ip, first, flist, &cur,
2786 0, &tmp_logflags, XFS_DATA_FORK);
2787 *logflagsp |= tmp_logflags;
2788 if (error)
2789 goto done;
1da177e4 2790 }
da087bad 2791
9e5987a7
DC		 2792 /* clear out the allocated field, done with it now in any case. */
2793 if (cur) {
2794 cur->bc_private.b.allocated = 0;
2795 *curp = cur;
1da177e4 2796 }
8096b1eb 2797
9e5987a7
DC		 2798 xfs_bmap_check_leaf_extents(*curp, ip, XFS_DATA_FORK);
2799done:
2800 *logflagsp |= rval;
1da177e4 2801 return error;
9e5987a7
DC		 2802#undef LEFT
2803#undef RIGHT
2804#undef PREV
1da177e4
LT		 2805}
2806
2807/*
9e5987a7 2808 * Convert a hole to a delayed allocation.
1da177e4 2809 */
9e5987a7
DC		 2810STATIC void
2811xfs_bmap_add_extent_hole_delay(
2812 xfs_inode_t *ip, /* incore inode pointer */
2813 xfs_extnum_t *idx, /* extent number to update/insert */
2814 xfs_bmbt_irec_t *new) /* new data to add to file extents */
1da177e4 2815{
9e5987a7
DC		 2816 xfs_ifork_t *ifp; /* inode fork pointer */
2817 xfs_bmbt_irec_t left; /* left neighbor extent entry */
2818 xfs_filblks_t newlen=0; /* new indirect size */
2819 xfs_filblks_t oldlen=0; /* old indirect size */
2820 xfs_bmbt_irec_t right; /* right neighbor extent entry */
2821 int state; /* state bits, accessed thru macros */
2822 xfs_filblks_t temp=0; /* temp for indirect calculations */
1da177e4 2823
9e5987a7
DC		 2824 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2825 state = 0;
2826 ASSERT(isnullstartblock(new->br_startblock));
1da177e4
LT		 2827
2828 /*
9e5987a7 2829 * Check and set flags if this segment has a left neighbor
1da177e4 2830 */
9e5987a7
DC		 2831 if (*idx > 0) {
2832 state |= BMAP_LEFT_VALID;
2833 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
2834
2835 if (isnullstartblock(left.br_startblock))
2836 state |= BMAP_LEFT_DELAY;
1da177e4 2837 }
1da177e4 2838
9e5987a7
DC		 2839 /*
2840 * Check and set flags if the current (right) segment exists.
2841 * If it doesn't exist, we're converting the hole at end-of-file.
2842 */
2843 if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
2844 state |= BMAP_RIGHT_VALID;
2845 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
1da177e4 2846
9e5987a7
DC		 2847 if (isnullstartblock(right.br_startblock))
2848 state |= BMAP_RIGHT_DELAY;
1da177e4 2849 }
9e5987a7 2850
1da177e4 2851 /*
9e5987a7
DC		 2852 * Set contiguity flags on the left and right neighbors.
2853 * Don't let extents get too large, even if the pieces are contiguous.
1da177e4 2854 */
9e5987a7
DC		 2855 if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2856 left.br_startoff + left.br_blockcount == new->br_startoff &&
2857 left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2858 state |= BMAP_LEFT_CONTIG;
2859
2860 if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2861 new->br_startoff + new->br_blockcount == right.br_startoff &&
2862 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2863 (!(state & BMAP_LEFT_CONTIG) ||
2864 (left.br_blockcount + new->br_blockcount +
2865 right.br_blockcount <= MAXEXTLEN)))
2866 state |= BMAP_RIGHT_CONTIG;
cc09c0dc
DC		 2867
2868 /*
9e5987a7 2869 * Switch out based on the contiguity flags.
cc09c0dc 2870 */
9e5987a7
DC		 2871 switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2872 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2873 /*
2874 * New allocation is contiguous with delayed allocations
2875 * on the left and on the right.
2876 * Merge all three into a single extent record.
2877 */
2878 --*idx;
2879 temp = left.br_blockcount + new->br_blockcount +
2880 right.br_blockcount;
cc09c0dc 2881
9e5987a7
DC		 2882 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2883 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2884 oldlen = startblockval(left.br_startblock) +
2885 startblockval(new->br_startblock) +
2886 startblockval(right.br_startblock);
2887 newlen = xfs_bmap_worst_indlen(ip, temp);
2888 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2889 nullstartblock((int)newlen));
2890 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1da177e4 2891
9e5987a7
DC		 2892 xfs_iext_remove(ip, *idx + 1, 1, state);
2893 break;
1da177e4 2894
9e5987a7
DC		 2895 case BMAP_LEFT_CONTIG:
2896 /*
2897 * New allocation is contiguous with a delayed allocation
2898 * on the left.
2899 * Merge the new allocation with the left neighbor.
2900 */
2901 --*idx;
2902 temp = left.br_blockcount + new->br_blockcount;
1da177e4 2903
9e5987a7
DC		 2904 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2905 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2906 oldlen = startblockval(left.br_startblock) +
2907 startblockval(new->br_startblock);
2908 newlen = xfs_bmap_worst_indlen(ip, temp);
2909 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2910 nullstartblock((int)newlen));
2911 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2912 break;
1da177e4 2913
9e5987a7
DC		 2914 case BMAP_RIGHT_CONTIG:
2915 /*
2916 * New allocation is contiguous with a delayed allocation
2917 * on the right.
2918 * Merge the new allocation with the right neighbor.
2919 */
2920 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2921 temp = new->br_blockcount + right.br_blockcount;
2922 oldlen = startblockval(new->br_startblock) +
2923 startblockval(right.br_startblock);
2924 newlen = xfs_bmap_worst_indlen(ip, temp);
2925 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2926 new->br_startoff,
2927 nullstartblock((int)newlen), temp, right.br_state);
2928 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2929 break;
2930
2931 case 0:
2932 /*
2933 * New allocation is not contiguous with another
2934 * delayed allocation.
2935 * Insert a new entry.
2936 */
2937 oldlen = newlen = 0;
2938 xfs_iext_insert(ip, *idx, 1, new, state);
2939 break;
1da177e4 2940 }
9e5987a7
DC		 2941 if (oldlen != newlen) {
2942 ASSERT(oldlen > newlen);
2943 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
2944 (int64_t)(oldlen - newlen), 0);
1da177e4 2945 /*
9e5987a7 2946 * Nothing to do for disk quota accounting here.
1da177e4 2947 */
1da177e4 2948 }
1da177e4
LT		 2949}
2950
2951/*
9e5987a7 2952 * Convert a hole to a real allocation.
1da177e4 2953 */
9e5987a7
DC		 2954STATIC int /* error */
2955xfs_bmap_add_extent_hole_real(
2956 struct xfs_bmalloca *bma,
2957 int whichfork)
27a3f8f2 2958{
9e5987a7
DC		 2959 struct xfs_bmbt_irec *new = &bma->got;
2960 int error; /* error return value */
2961 int i; /* temp state */
2962 xfs_ifork_t *ifp; /* inode fork pointer */
2963 xfs_bmbt_irec_t left; /* left neighbor extent entry */
2964 xfs_bmbt_irec_t right; /* right neighbor extent entry */
2965 int rval=0; /* return value (logging flags) */
2966 int state; /* state bits, accessed thru macros */
27a3f8f2 2967
9e5987a7 2968 ifp = XFS_IFORK_PTR(bma->ip, whichfork);
27a3f8f2 2969
9e5987a7
DC		 2970 ASSERT(bma->idx >= 0);
2971 ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2972 ASSERT(!isnullstartblock(new->br_startblock));
2973 ASSERT(!bma->cur ||
2974 !(bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
27a3f8f2 2975
9e5987a7 2976 XFS_STATS_INC(xs_add_exlist);
27a3f8f2 2977
9e5987a7
DC		 2978 state = 0;
2979 if (whichfork == XFS_ATTR_FORK)
2980 state |= BMAP_ATTRFORK;
27a3f8f2 2981
9e5987a7
DC		 2982 /*
2983 * Check and set flags if this segment has a left neighbor.
2984 */
2985 if (bma->idx > 0) {
2986 state |= BMAP_LEFT_VALID;
2987 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &left);
2988 if (isnullstartblock(left.br_startblock))
2989 state |= BMAP_LEFT_DELAY;
2990 }
27a3f8f2
CH		 2991
2992 /*
9e5987a7
DC		 2993 * Check and set flags if this segment has a current value.
2994 * Not true if we're inserting into the "hole" at eof.
27a3f8f2 2995 */
9e5987a7
DC		 2996 if (bma->idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
2997 state |= BMAP_RIGHT_VALID;
2998 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &right);
2999 if (isnullstartblock(right.br_startblock))
3000 state |= BMAP_RIGHT_DELAY;
3001 }
27a3f8f2 3002
9e5987a7
DC		 3003 /*
3004 * We're inserting a real allocation between "left" and "right".
3005 * Set the contiguity flags. Don't let extents get too large.
3006 */
3007 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
3008 left.br_startoff + left.br_blockcount == new->br_startoff &&
3009 left.br_startblock + left.br_blockcount == new->br_startblock &&
3010 left.br_state == new->br_state &&
3011 left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
3012 state |= BMAP_LEFT_CONTIG;
27a3f8f2 3013
9e5987a7
DC		 3014 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
3015 new->br_startoff + new->br_blockcount == right.br_startoff &&
3016 new->br_startblock + new->br_blockcount == right.br_startblock &&
3017 new->br_state == right.br_state &&
3018 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
3019 (!(state & BMAP_LEFT_CONTIG) ||
3020 left.br_blockcount + new->br_blockcount +
3021 right.br_blockcount <= MAXEXTLEN))
3022 state |= BMAP_RIGHT_CONTIG;
27a3f8f2 3023
9e5987a7
DC		 3024 error = 0;
3025 /*
3026 * Select which case we're in here, and implement it.
3027 */
3028 switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
3029 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
3030 /*
3031 * New allocation is contiguous with real allocations on the
3032 * left and on the right.
3033 * Merge all three into a single extent record.
3034 */
3035 --bma->idx;
3036 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3037 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3038 left.br_blockcount + new->br_blockcount +
3039 right.br_blockcount);
3040 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
27a3f8f2 3041
9e5987a7 3042 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
27a3f8f2 3043
9e5987a7
DC		 3044 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3045 XFS_IFORK_NEXTENTS(bma->ip, whichfork) - 1);
3046 if (bma->cur == NULL) {
3047 rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3048 } else {
3049 rval = XFS_ILOG_CORE;
3050 error = xfs_bmbt_lookup_eq(bma->cur, right.br_startoff,
3051 right.br_startblock, right.br_blockcount,
3052 &i);
3053 if (error)
3054 goto done;
3055 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3056 error = xfs_btree_delete(bma->cur, &i);
3057 if (error)
3058 goto done;
3059 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3060 error = xfs_btree_decrement(bma->cur, 0, &i);
3061 if (error)
3062 goto done;
3063 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3064 error = xfs_bmbt_update(bma->cur, left.br_startoff,
3065 left.br_startblock,
3066 left.br_blockcount +
3067 new->br_blockcount +
3068 right.br_blockcount,
3069 left.br_state);
3070 if (error)
3071 goto done;
3072 }
3073 break;
27a3f8f2 3074
9e5987a7
DC		 3075 case BMAP_LEFT_CONTIG:
3076 /*
3077 * New allocation is contiguous with a real allocation
3078 * on the left.
3079 * Merge the new allocation with the left neighbor.
3080 */
3081 --bma->idx;
3082 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3083 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3084 left.br_blockcount + new->br_blockcount);
3085 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1da177e4 3086
9e5987a7
DC		 3087 if (bma->cur == NULL) {
3088 rval = xfs_ilog_fext(whichfork);
3089 } else {
3090 rval = 0;
3091 error = xfs_bmbt_lookup_eq(bma->cur, left.br_startoff,
3092 left.br_startblock, left.br_blockcount,
3093 &i);
3094 if (error)
3095 goto done;
3096 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3097 error = xfs_bmbt_update(bma->cur, left.br_startoff,
3098 left.br_startblock,
3099 left.br_blockcount +
3100 new->br_blockcount,
3101 left.br_state);
3102 if (error)
3103 goto done;
3104 }
3105 break;
27a3f8f2 3106
9e5987a7
DC		 3107 case BMAP_RIGHT_CONTIG:
3108 /*
3109 * New allocation is contiguous with a real allocation
3110 * on the right.
3111 * Merge the new allocation with the right neighbor.
3112 */
3113 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3114 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx),
3115 new->br_startoff, new->br_startblock,
3116 new->br_blockcount + right.br_blockcount,
3117 right.br_state);
3118 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
27a3f8f2 3119
9e5987a7
DC		 3120 if (bma->cur == NULL) {
3121 rval = xfs_ilog_fext(whichfork);
3122 } else {
3123 rval = 0;
3124 error = xfs_bmbt_lookup_eq(bma->cur,
3125 right.br_startoff,
3126 right.br_startblock,
3127 right.br_blockcount, &i);
3128 if (error)
3129 goto done;
3130 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3131 error = xfs_bmbt_update(bma->cur, new->br_startoff,
3132 new->br_startblock,
3133 new->br_blockcount +
3134 right.br_blockcount,
3135 right.br_state);
3136 if (error)
3137 goto done;
3138 }
3139 break;
3140
3141 case 0:
3142 /*
3143 * New allocation is not contiguous with another
3144 * real allocation.
3145 * Insert a new entry.
3146 */
3147 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
3148 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3149 XFS_IFORK_NEXTENTS(bma->ip, whichfork) + 1);
3150 if (bma->cur == NULL) {
3151 rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3152 } else {
3153 rval = XFS_ILOG_CORE;
3154 error = xfs_bmbt_lookup_eq(bma->cur,
3155 new->br_startoff,
3156 new->br_startblock,
3157 new->br_blockcount, &i);
3158 if (error)
3159 goto done;
3160 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
3161 bma->cur->bc_rec.b.br_state = new->br_state;
3162 error = xfs_btree_insert(bma->cur, &i);
3163 if (error)
3164 goto done;
3165 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3166 }
3167 break;
3168 }
3169
3170 /* convert to a btree if necessary */
3171 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
3172 int tmp_logflags; /* partial log flag return val */
3173
3174 ASSERT(bma->cur == NULL);
3175 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
3176 bma->firstblock, bma->flist, &bma->cur,
3177 0, &tmp_logflags, whichfork);
3178 bma->logflags |= tmp_logflags;
3179 if (error)
3180 goto done;
3181 }
3182
3183 /* clear out the allocated field, done with it now in any case. */
3184 if (bma->cur)
3185 bma->cur->bc_private.b.allocated = 0;
3186
3187 xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
3188done:
3189 bma->logflags |= rval;
3190 return error;
1da177e4
LT		 3191}
3192
3193/*
9e5987a7 3194 * Functions used in the extent read, allocate and remove paths
1da177e4 3195 */
1da177e4 3196
9e5987a7
DC		 3197/*
3198 * Adjust the size of the new extent based on di_extsize and rt extsize.
3199 */
68988114 3200int
9e5987a7
DC		 3201xfs_bmap_extsize_align(
3202 xfs_mount_t *mp,
3203 xfs_bmbt_irec_t *gotp, /* next extent pointer */
3204 xfs_bmbt_irec_t *prevp, /* previous extent pointer */
3205 xfs_extlen_t extsz, /* align to this extent size */
3206 int rt, /* is this a realtime inode? */
3207 int eof, /* is extent at end-of-file? */
3208 int delay, /* creating delalloc extent? */
3209 int convert, /* overwriting unwritten extent? */
3210 xfs_fileoff_t *offp, /* in/out: aligned offset */
3211 xfs_extlen_t *lenp) /* in/out: aligned length */
4e8938fe 3212{
9e5987a7
DC		 3213 xfs_fileoff_t orig_off; /* original offset */
3214 xfs_extlen_t orig_alen; /* original length */
3215 xfs_fileoff_t orig_end; /* original off+len */
3216 xfs_fileoff_t nexto; /* next file offset */
3217 xfs_fileoff_t prevo; /* previous file offset */
3218 xfs_fileoff_t align_off; /* temp for offset */
3219 xfs_extlen_t align_alen; /* temp for length */
3220 xfs_extlen_t temp; /* temp for calculations */
4e8938fe 3221
9e5987a7 3222 if (convert)
4e8938fe 3223 return 0;
4e8938fe 3224
9e5987a7
DC		 3225 orig_off = align_off = *offp;
3226 orig_alen = align_alen = *lenp;
3227 orig_end = orig_off + orig_alen;
1da177e4 3228
1da177e4 3229 /*
9e5987a7
DC		 3230 * If this request overlaps an existing extent, then don't
3231 * attempt to perform any additional alignment.
1da177e4 3232 */
9e5987a7
DC		 3233 if (!delay && !eof &&
3234 (orig_off >= gotp->br_startoff) &&
3235 (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
3236 return 0;
3237 }
3238
1da177e4 3239 /*
9e5987a7
DC		 3240 * If the file offset is unaligned vs. the extent size
3241 * we need to align it. This will be possible unless
3242 * the file was previously written with a kernel that didn't
3243 * perform this alignment, or if a truncate shot us in the
3244 * foot.
1da177e4 3245 */
9e5987a7
DC		 3246 temp = do_mod(orig_off, extsz);
3247 if (temp) {
3248 align_alen += temp;
3249 align_off -= temp;
1da177e4
LT		 3250 }
3251 /*
9e5987a7 3252 * Same adjustment for the end of the requested area.
1da177e4 3253 */
9e5987a7
DC		 3254 if ((temp = (align_alen % extsz))) {
3255 align_alen += extsz - temp;
3256 }
1da177e4 3257 /*
9e5987a7
DC		 3258 * If the previous block overlaps with this proposed allocation
3259 * then move the start forward without adjusting the length.
1da177e4 3260 */
9e5987a7
DC		 3261 if (prevp->br_startoff != NULLFILEOFF) {
3262 if (prevp->br_startblock == HOLESTARTBLOCK)
3263 prevo = prevp->br_startoff;
3264 else
3265 prevo = prevp->br_startoff + prevp->br_blockcount;
3266 } else
3267 prevo = 0;
3268 if (align_off != orig_off && align_off < prevo)
3269 align_off = prevo;
3270 /*
3271 * If the next block overlaps with this proposed allocation
3272 * then move the start back without adjusting the length,
3273 * but not before offset 0.
3274 * This may of course make the start overlap previous block,
3275 * and if we hit the offset 0 limit then the next block
3276 * can still overlap too.
3277 */
3278 if (!eof && gotp->br_startoff != NULLFILEOFF) {
3279 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
3280 (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
3281 nexto = gotp->br_startoff + gotp->br_blockcount;
3282 else
3283 nexto = gotp->br_startoff;
3284 } else
3285 nexto = NULLFILEOFF;
3286 if (!eof &&
3287 align_off + align_alen != orig_end &&
3288 align_off + align_alen > nexto)
3289 align_off = nexto > align_alen ? nexto - align_alen : 0;
3290 /*
3291 * If we're now overlapping the next or previous extent that
3292 * means we can't fit an extsz piece in this hole. Just move
3293 * the start forward to the first valid spot and set
3294 * the length so we hit the end.
3295 */
3296 if (align_off != orig_off && align_off < prevo)
3297 align_off = prevo;
3298 if (align_off + align_alen != orig_end &&
3299 align_off + align_alen > nexto &&
3300 nexto != NULLFILEOFF) {
3301 ASSERT(nexto > prevo);
3302 align_alen = nexto - align_off;
3303 }
1da177e4 3304
9e5987a7
DC		 3305 /*
3306 * If realtime, and the result isn't a multiple of the realtime
3307 * extent size we need to remove blocks until it is.
3308 */
3309 if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
1da177e4 3310 /*
9e5987a7
DC		 3311 * We're not covering the original request, or
3312 * we won't be able to once we fix the length.
1da177e4 3313 */
9e5987a7
DC		 3314 if (orig_off < align_off ||
3315 orig_end > align_off + align_alen ||
3316 align_alen - temp < orig_alen)
2451337d 3317 return -EINVAL;
1da177e4 3318 /*
9e5987a7 3319 * Try to fix it by moving the start up.
1da177e4 3320 */
9e5987a7
DC		 3321 if (align_off + temp <= orig_off) {
3322 align_alen -= temp;
3323 align_off += temp;
1da177e4 3324 }
9e5987a7
DC		 3325 /*
3326 * Try to fix it by moving the end in.
3327 */
3328 else if (align_off + align_alen - temp >= orig_end)
3329 align_alen -= temp;
3330 /*
3331 * Set the start to the minimum then trim the length.
3332 */
3333 else {
3334 align_alen -= orig_off - align_off;
3335 align_off = orig_off;
3336 align_alen -= align_alen % mp->m_sb.sb_rextsize;
1da177e4 3337 }
1da177e4 3338 /*
9e5987a7 3339 * Result doesn't cover the request, fail it.
1da177e4 3340 */
9e5987a7 3341 if (orig_off < align_off || orig_end > align_off + align_alen)
2451337d 3342 return -EINVAL;
9e5987a7
DC		 3343 } else {
3344 ASSERT(orig_off >= align_off);
3345 ASSERT(orig_end <= align_off + align_alen);
1da177e4 3346 }
1da177e4 3347
0b1b213f 3348#ifdef DEBUG
9e5987a7
DC		 3349 if (!eof && gotp->br_startoff != NULLFILEOFF)
3350 ASSERT(align_off + align_alen <= gotp->br_startoff);
3351 if (prevp->br_startoff != NULLFILEOFF)
3352 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3353#endif
1da177e4 3354
9e5987a7
DC		 3355 *lenp = align_alen;
3356 *offp = align_off;
3357 return 0;
1da177e4 3358}
1da177e4 3359
9e5987a7
DC		 3360#define XFS_ALLOC_GAP_UNITS 4
3361
68988114 3362void
9e5987a7 3363xfs_bmap_adjacent(
68988114 3364 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
1da177e4 3365{
9e5987a7
DC		 3366 xfs_fsblock_t adjust; /* adjustment to block numbers */
3367 xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
3368 xfs_mount_t *mp; /* mount point structure */
3369 int nullfb; /* true if ap->firstblock isn't set */
3370 int rt; /* true if inode is realtime */
1da177e4 3371
9e5987a7
DC		 3372#define ISVALID(x,y) \
3373 (rt ? \
3374 (x) < mp->m_sb.sb_rblocks : \
3375 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3376 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3377 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
1da177e4 3378
9e5987a7
DC		 3379 mp = ap->ip->i_mount;
3380 nullfb = *ap->firstblock == NULLFSBLOCK;
3381 rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
3382 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3383 /*
3384 * If allocating at eof, and there's a previous real block,
3385 * try to use its last block as our starting point.
3386 */
3387 if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3388 !isnullstartblock(ap->prev.br_startblock) &&
3389 ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3390 ap->prev.br_startblock)) {
3391 ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3392 /*
3393 * Adjust for the gap between prevp and us.
3394 */
3395 adjust = ap->offset -
3396 (ap->prev.br_startoff + ap->prev.br_blockcount);
3397 if (adjust &&
3398 ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3399 ap->blkno += adjust;
3400 }
3401 /*
3402 * If not at eof, then compare the two neighbor blocks.
3403 * Figure out whether either one gives us a good starting point,
3404 * and pick the better one.
3405 */
3406 else if (!ap->eof) {
3407 xfs_fsblock_t gotbno; /* right side block number */
3408 xfs_fsblock_t gotdiff=0; /* right side difference */
3409 xfs_fsblock_t prevbno; /* left side block number */
3410 xfs_fsblock_t prevdiff=0; /* left side difference */
3411
3412 /*
3413 * If there's a previous (left) block, select a requested
3414 * start block based on it.
3415 */
3416 if (ap->prev.br_startoff != NULLFILEOFF &&
3417 !isnullstartblock(ap->prev.br_startblock) &&
3418 (prevbno = ap->prev.br_startblock +
3419 ap->prev.br_blockcount) &&
3420 ISVALID(prevbno, ap->prev.br_startblock)) {
3421 /*
3422 * Calculate gap to end of previous block.
3423 */
3424 adjust = prevdiff = ap->offset -
3425 (ap->prev.br_startoff +
3426 ap->prev.br_blockcount);
3427 /*
3428 * Figure the startblock based on the previous block's
3429 * end and the gap size.
3430 * Heuristic!
3431 * If the gap is large relative to the piece we're
3432 * allocating, or using it gives us an invalid block
3433 * number, then just use the end of the previous block.
3434 */
3435 if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3436 ISVALID(prevbno + prevdiff,
3437 ap->prev.br_startblock))
3438 prevbno += adjust;
3439 else
3440 prevdiff += adjust;
3441 /*
3442 * If the firstblock forbids it, can't use it,
3443 * must use default.
3444 */
3445 if (!rt && !nullfb &&
3446 XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3447 prevbno = NULLFSBLOCK;
1da177e4 3448 }
9e5987a7
DC		 3449 /*
3450 * No previous block or can't follow it, just default.
3451 */
3452 else
3453 prevbno = NULLFSBLOCK;
3454 /*
3455 * If there's a following (right) block, select a requested
3456 * start block based on it.
3457 */
3458 if (!isnullstartblock(ap->got.br_startblock)) {
3459 /*
3460 * Calculate gap to start of next block.
3461 */
3462 adjust = gotdiff = ap->got.br_startoff - ap->offset;
3463 /*
3464 * Figure the startblock based on the next block's
3465 * start and the gap size.
3466 */
3467 gotbno = ap->got.br_startblock;
3468 /*
3469 * Heuristic!
3470 * If the gap is large relative to the piece we're
3471 * allocating, or using it gives us an invalid block
3472 * number, then just use the start of the next block
3473 * offset by our length.
3474 */
3475 if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3476 ISVALID(gotbno - gotdiff, gotbno))
3477 gotbno -= adjust;
3478 else if (ISVALID(gotbno - ap->length, gotbno)) {
3479 gotbno -= ap->length;
3480 gotdiff += adjust - ap->length;
3481 } else
3482 gotdiff += adjust;
3483 /*
3484 * If the firstblock forbids it, can't use it,
3485 * must use default.
3486 */
3487 if (!rt && !nullfb &&
3488 XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3489 gotbno = NULLFSBLOCK;
3490 }
3491 /*
3492 * No next block, just default.
3493 */
3494 else
3495 gotbno = NULLFSBLOCK;
3496 /*
3497 * If both valid, pick the better one, else the only good
3498 * one, else ap->blkno is already set (to 0 or the inode block).
3499 */
3500 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3501 ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3502 else if (prevbno != NULLFSBLOCK)
3503 ap->blkno = prevbno;
3504 else if (gotbno != NULLFSBLOCK)
3505 ap->blkno = gotbno;
1da177e4 3506 }
9e5987a7 3507#undef ISVALID
1da177e4 3508}
1da177e4 3509
c977eb10
CH		 3510static int
3511xfs_bmap_longest_free_extent(
3512 struct xfs_trans *tp,
3513 xfs_agnumber_t ag,
3514 xfs_extlen_t *blen,
3515 int *notinit)
3516{
3517 struct xfs_mount *mp = tp->t_mountp;
3518 struct xfs_perag *pag;
3519 xfs_extlen_t longest;
3520 int error = 0;
3521
3522 pag = xfs_perag_get(mp, ag);
3523 if (!pag->pagf_init) {
3524 error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK);
3525 if (error)
3526 goto out;
3527
3528 if (!pag->pagf_init) {
3529 *notinit = 1;
3530 goto out;
3531 }
3532 }
3533
3534 longest = xfs_alloc_longest_free_extent(mp, pag);
3535 if (*blen < longest)
3536 *blen = longest;
3537
3538out:
3539 xfs_perag_put(pag);
3540 return error;
3541}
3542
3543static void
3544xfs_bmap_select_minlen(
3545 struct xfs_bmalloca *ap,
3546 struct xfs_alloc_arg *args,
3547 xfs_extlen_t *blen,
3548 int notinit)
3549{
3550 if (notinit || *blen < ap->minlen) {
3551 /*
3552 * Since we did a BUF_TRYLOCK above, it is possible that
3553 * there is space for this request.
3554 */
3555 args->minlen = ap->minlen;
3556 } else if (*blen < args->maxlen) {
3557 /*
3558 * If the best seen length is less than the request length,
3559 * use the best as the minimum.
3560 */
3561 args->minlen = *blen;
3562 } else {
3563 /*
3564 * Otherwise we've seen an extent as big as maxlen, use that
3565 * as the minimum.
3566 */
3567 args->minlen = args->maxlen;
3568 }
3569}
3570
b64dfe4e 3571STATIC int
9e5987a7
DC		 3572xfs_bmap_btalloc_nullfb(
3573 struct xfs_bmalloca *ap,
3574 struct xfs_alloc_arg *args,
3575 xfs_extlen_t *blen)
b64dfe4e 3576{
9e5987a7 3577 struct xfs_mount *mp = ap->ip->i_mount;
9e5987a7
DC		 3578 xfs_agnumber_t ag, startag;
3579 int notinit = 0;
b64dfe4e
CH		 3580 int error;
3581
c977eb10 3582 args->type = XFS_ALLOCTYPE_START_BNO;
9e5987a7 3583 args->total = ap->total;
b64dfe4e 3584
9e5987a7
DC		 3585 startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3586 if (startag == NULLAGNUMBER)
3587 startag = ag = 0;
b64dfe4e 3588
9e5987a7 3589 while (*blen < args->maxlen) {
c977eb10
CH		 3590 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3591 &notinit);
3592 if (error)
3593 return error;
b64dfe4e 3594
9e5987a7
DC		 3595 if (++ag == mp->m_sb.sb_agcount)
3596 ag = 0;
3597 if (ag == startag)
3598 break;
9e5987a7 3599 }
b64dfe4e 3600
c977eb10
CH		 3601 xfs_bmap_select_minlen(ap, args, blen, notinit);
3602 return 0;
3603}
3604
3605STATIC int
3606xfs_bmap_btalloc_filestreams(
3607 struct xfs_bmalloca *ap,
3608 struct xfs_alloc_arg *args,
3609 xfs_extlen_t *blen)
3610{
3611 struct xfs_mount *mp = ap->ip->i_mount;
3612 xfs_agnumber_t ag;
3613 int notinit = 0;
3614 int error;
3615
3616 args->type = XFS_ALLOCTYPE_NEAR_BNO;
3617 args->total = ap->total;
3618
3619 ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3620 if (ag == NULLAGNUMBER)
3621 ag = 0;
3622
3623 error = xfs_bmap_longest_free_extent(args->tp, ag, blen, &notinit);
3624 if (error)
3625 return error;
3626
3627 if (*blen < args->maxlen) {
3628 error = xfs_filestream_new_ag(ap, &ag);
3629 if (error)
3630 return error;
3631
3632 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3633 &notinit);
3634 if (error)
3635 return error;
3636
3637 }
3638
3639 xfs_bmap_select_minlen(ap, args, blen, notinit);
b64dfe4e
CH		 3640
3641 /*
c977eb10
CH		 3642 * Set the failure fallback case to look in the selected AG as stream
3643 * may have moved.
b64dfe4e 3644 */
c977eb10 3645 ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
b64dfe4e 3646 return 0;
b64dfe4e
CH		 3647}
3648
9e5987a7
DC		 3649STATIC int
3650xfs_bmap_btalloc(
68988114 3651 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
4403280a 3652{
9e5987a7
DC		 3653 xfs_mount_t *mp; /* mount point structure */
3654 xfs_alloctype_t atype = 0; /* type for allocation routines */
3655 xfs_extlen_t align; /* minimum allocation alignment */
3656 xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
3657 xfs_agnumber_t ag;
3658 xfs_alloc_arg_t args;
3659 xfs_extlen_t blen;
3660 xfs_extlen_t nextminlen = 0;
3661 int nullfb; /* true if ap->firstblock isn't set */
3662 int isaligned;
3663 int tryagain;
3664 int error;
33177f05 3665 int stripe_align;
4403280a 3666
9e5987a7 3667 ASSERT(ap->length);
4403280a 3668
9e5987a7 3669 mp = ap->ip->i_mount;
33177f05
DC		 3670
3671 /* stripe alignment for allocation is determined by mount parameters */
3672 stripe_align = 0;
3673 if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
3674 stripe_align = mp->m_swidth;
3675 else if (mp->m_dalign)
3676 stripe_align = mp->m_dalign;
3677
9e5987a7
DC		 3678 align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
3679 if (unlikely(align)) {
3680 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
3681 align, 0, ap->eof, 0, ap->conv,
3682 &ap->offset, &ap->length);
3683 ASSERT(!error);
3684 ASSERT(ap->length);
4403280a 3685 }
33177f05
DC		 3686
3687
9e5987a7
DC		 3688 nullfb = *ap->firstblock == NULLFSBLOCK;
3689 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3690 if (nullfb) {
3691 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
3692 ag = xfs_filestream_lookup_ag(ap->ip);
3693 ag = (ag != NULLAGNUMBER) ? ag : 0;
3694 ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3695 } else {
3696 ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3697 }
3698 } else
3699 ap->blkno = *ap->firstblock;
4403280a 3700
9e5987a7 3701 xfs_bmap_adjacent(ap);
4403280a 3702
9e5987a7
DC		 3703 /*
3704 * If allowed, use ap->blkno; otherwise must use firstblock since
3705 * it's in the right allocation group.
3706 */
3707 if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3708 ;
3709 else
3710 ap->blkno = *ap->firstblock;
3711 /*
3712 * Normal allocation, done through xfs_alloc_vextent.
3713 */
3714 tryagain = isaligned = 0;
3715 memset(&args, 0, sizeof(args));
3716 args.tp = ap->tp;
3717 args.mp = mp;
3718 args.fsbno = ap->blkno;
4403280a 3719
9e5987a7
DC		 3720 /* Trim the allocation back to the maximum an AG can fit. */
3721 args.maxlen = MIN(ap->length, XFS_ALLOC_AG_MAX_USABLE(mp));
3722 args.firstblock = *ap->firstblock;
3723 blen = 0;
3724 if (nullfb) {
c977eb10
CH		 3725 /*
3726 * Search for an allocation group with a single extent large
3727 * enough for the request. If one isn't found, then adjust
3728 * the minimum allocation size to the largest space found.
3729 */
3730 if (ap->userdata && xfs_inode_is_filestream(ap->ip))
3731 error = xfs_bmap_btalloc_filestreams(ap, &args, &blen);
3732 else
3733 error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
4403280a
CH		 3734 if (error)
3735 return error;
9e5987a7
DC		 3736 } else if (ap->flist->xbf_low) {
3737 if (xfs_inode_is_filestream(ap->ip))
3738 args.type = XFS_ALLOCTYPE_FIRST_AG;
3739 else
3740 args.type = XFS_ALLOCTYPE_START_BNO;
3741 args.total = args.minlen = ap->minlen;
3742 } else {
3743 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3744 args.total = ap->total;
3745 args.minlen = ap->minlen;
4403280a 3746 }
9e5987a7
DC		 3747 /* apply extent size hints if obtained earlier */
3748 if (unlikely(align)) {
3749 args.prod = align;
3750 if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
3751 args.mod = (xfs_extlen_t)(args.prod - args.mod);
3752 } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
3753 args.prod = 1;
3754 args.mod = 0;
3755 } else {
3756 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
3757 if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
3758 args.mod = (xfs_extlen_t)(args.prod - args.mod);
4403280a 3759 }
7e47a4ef 3760 /*
9e5987a7
DC		 3761 * If we are not low on available data blocks, and the
3762 * underlying logical volume manager is a stripe, and
3763 * the file offset is zero then try to allocate data
3764 * blocks on stripe unit boundary.
3765 * NOTE: ap->aeof is only set if the allocation length
3766 * is >= the stripe unit and the allocation offset is
3767 * at the end of file.
7e47a4ef 3768 */
9e5987a7
DC		 3769 if (!ap->flist->xbf_low && ap->aeof) {
3770 if (!ap->offset) {
33177f05 3771 args.alignment = stripe_align;
9e5987a7
DC		 3772 atype = args.type;
3773 isaligned = 1;
3774 /*
3775 * Adjust for alignment
3776 */
3777 if (blen > args.alignment && blen <= args.maxlen)
3778 args.minlen = blen - args.alignment;
3779 args.minalignslop = 0;
3780 } else {
3781 /*
3782 * First try an exact bno allocation.
3783 * If it fails then do a near or start bno
3784 * allocation with alignment turned on.
3785 */
3786 atype = args.type;
3787 tryagain = 1;
3788 args.type = XFS_ALLOCTYPE_THIS_BNO;
3789 args.alignment = 1;
3790 /*
3791 * Compute the minlen+alignment for the
3792 * next case. Set slop so that the value
3793 * of minlen+alignment+slop doesn't go up
3794 * between the calls.
3795 */
33177f05
DC		 3796 if (blen > stripe_align && blen <= args.maxlen)
3797 nextminlen = blen - stripe_align;
9e5987a7
DC		 3798 else
3799 nextminlen = args.minlen;
33177f05 3800 if (nextminlen + stripe_align > args.minlen + 1)
9e5987a7 3801 args.minalignslop =
33177f05 3802 nextminlen + stripe_align -
9e5987a7
DC		 3803 args.minlen - 1;
3804 else
3805 args.minalignslop = 0;
7e47a4ef
DC		 3806 }
3807 } else {
9e5987a7
DC		 3808 args.alignment = 1;
3809 args.minalignslop = 0;
7e47a4ef 3810 }
9e5987a7
DC		 3811 args.minleft = ap->minleft;
3812 args.wasdel = ap->wasdel;
3813 args.isfl = 0;
3814 args.userdata = ap->userdata;
3815 if ((error = xfs_alloc_vextent(&args)))
3816 return error;
3817 if (tryagain && args.fsbno == NULLFSBLOCK) {
3818 /*
3819 * Exact allocation failed. Now try with alignment
3820 * turned on.
3821 */
3822 args.type = atype;
3823 args.fsbno = ap->blkno;
33177f05 3824 args.alignment = stripe_align;
9e5987a7
DC		 3825 args.minlen = nextminlen;
3826 args.minalignslop = 0;
3827 isaligned = 1;
3828 if ((error = xfs_alloc_vextent(&args)))
3829 return error;
7e47a4ef 3830 }
9e5987a7
DC		 3831 if (isaligned && args.fsbno == NULLFSBLOCK) {
3832 /*
3833 * allocation failed, so turn off alignment and
3834 * try again.
3835 */
3836 args.type = atype;
3837 args.fsbno = ap->blkno;
3838 args.alignment = 0;
3839 if ((error = xfs_alloc_vextent(&args)))
7e47a4ef
DC		 3840 return error;
3841 }
9e5987a7
DC		 3842 if (args.fsbno == NULLFSBLOCK && nullfb &&
3843 args.minlen > ap->minlen) {
3844 args.minlen = ap->minlen;
3845 args.type = XFS_ALLOCTYPE_START_BNO;
3846 args.fsbno = ap->blkno;
3847 if ((error = xfs_alloc_vextent(&args)))
3848 return error;
7e47a4ef 3849 }
9e5987a7
DC		 3850 if (args.fsbno == NULLFSBLOCK && nullfb) {
3851 args.fsbno = 0;
3852 args.type = XFS_ALLOCTYPE_FIRST_AG;
3853 args.total = ap->minlen;
3854 args.minleft = 0;
3855 if ((error = xfs_alloc_vextent(&args)))
3856 return error;
3857 ap->flist->xbf_low = 1;
3858 }
3859 if (args.fsbno != NULLFSBLOCK) {
3860 /*
3861 * check the allocation happened at the same or higher AG than
3862 * the first block that was allocated.
3863 */
3864 ASSERT(*ap->firstblock == NULLFSBLOCK ||
3865 XFS_FSB_TO_AGNO(mp, *ap->firstblock) ==
3866 XFS_FSB_TO_AGNO(mp, args.fsbno) ||
3867 (ap->flist->xbf_low &&
3868 XFS_FSB_TO_AGNO(mp, *ap->firstblock) <
3869 XFS_FSB_TO_AGNO(mp, args.fsbno)));
7e47a4ef 3870
9e5987a7
DC		 3871 ap->blkno = args.fsbno;
3872 if (*ap->firstblock == NULLFSBLOCK)
3873 *ap->firstblock = args.fsbno;
3874 ASSERT(nullfb || fb_agno == args.agno ||
3875 (ap->flist->xbf_low && fb_agno < args.agno));
3876 ap->length = args.len;
3877 ap->ip->i_d.di_nblocks += args.len;
3878 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3879 if (ap->wasdel)
3880 ap->ip->i_delayed_blks -= args.len;
3881 /*
3882 * Adjust the disk quota also. This was reserved
3883 * earlier.
3884 */
3885 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3886 ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
3887 XFS_TRANS_DQ_BCOUNT,
3888 (long) args.len);
3889 } else {
3890 ap->blkno = NULLFSBLOCK;
3891 ap->length = 0;
3892 }
3893 return 0;
3894}
7e47a4ef 3895
9e5987a7
DC		 3896/*
3897 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
3898 * It figures out where to ask the underlying allocator to put the new extent.
3899 */
3900STATIC int
3901xfs_bmap_alloc(
68988114 3902 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
9e5987a7
DC		 3903{
3904 if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
3905 return xfs_bmap_rtalloc(ap);
3906 return xfs_bmap_btalloc(ap);
3907}
a5bd606b 3908
9e5987a7
DC		 3909/*
3910 * Trim the returned map to the required bounds
3911 */
3912STATIC void
3913xfs_bmapi_trim_map(
3914 struct xfs_bmbt_irec *mval,
3915 struct xfs_bmbt_irec *got,
3916 xfs_fileoff_t *bno,
3917 xfs_filblks_t len,
3918 xfs_fileoff_t obno,
3919 xfs_fileoff_t end,
3920 int n,
3921 int flags)
3922{
3923 if ((flags & XFS_BMAPI_ENTIRE) ||
3924 got->br_startoff + got->br_blockcount <= obno) {
3925 *mval = *got;
3926 if (isnullstartblock(got->br_startblock))
3927 mval->br_startblock = DELAYSTARTBLOCK;
3928 return;
3929 }
7e47a4ef 3930
9e5987a7
DC		 3931 if (obno > *bno)
3932 *bno = obno;
3933 ASSERT((*bno >= obno) || (n == 0));
3934 ASSERT(*bno < end);
3935 mval->br_startoff = *bno;
3936 if (isnullstartblock(got->br_startblock))
3937 mval->br_startblock = DELAYSTARTBLOCK;
3938 else
3939 mval->br_startblock = got->br_startblock +
3940 (*bno - got->br_startoff);
7e47a4ef 3941 /*
9e5987a7
DC		 3942 * Return the minimum of what we got and what we asked for for
3943 * the length. We can use the len variable here because it is
3944 * modified below and we could have been there before coming
3945 * here if the first part of the allocation didn't overlap what
3946 * was asked for.
7e47a4ef 3947 */
9e5987a7
DC		 3948 mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3949 got->br_blockcount - (*bno - got->br_startoff));
3950 mval->br_state = got->br_state;
3951 ASSERT(mval->br_blockcount <= len);
3952 return;
7e47a4ef
DC		 3953}
3954
9e5987a7
DC		 3955/*
3956 * Update and validate the extent map to return
3957 */
3958STATIC void
3959xfs_bmapi_update_map(
3960 struct xfs_bmbt_irec **map,
3961 xfs_fileoff_t *bno,
3962 xfs_filblks_t *len,
3963 xfs_fileoff_t obno,
3964 xfs_fileoff_t end,
3965 int *n,
3966 int flags)
e04426b9 3967{
9e5987a7 3968 xfs_bmbt_irec_t *mval = *map;
e04426b9 3969
9e5987a7
DC		 3970 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3971 ((mval->br_startoff + mval->br_blockcount) <= end));
3972 ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3973 (mval->br_startoff < obno));
e04426b9 3974
9e5987a7
DC		 3975 *bno = mval->br_startoff + mval->br_blockcount;
3976 *len = end - *bno;
3977 if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3978 /* update previous map with new information */
3979 ASSERT(mval->br_startblock == mval[-1].br_startblock);
3980 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3981 ASSERT(mval->br_state == mval[-1].br_state);
3982 mval[-1].br_blockcount = mval->br_blockcount;
3983 mval[-1].br_state = mval->br_state;
3984 } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3985 mval[-1].br_startblock != DELAYSTARTBLOCK &&
3986 mval[-1].br_startblock != HOLESTARTBLOCK &&
3987 mval->br_startblock == mval[-1].br_startblock +
3988 mval[-1].br_blockcount &&
3989 ((flags & XFS_BMAPI_IGSTATE) ||
3990 mval[-1].br_state == mval->br_state)) {
3991 ASSERT(mval->br_startoff ==
3992 mval[-1].br_startoff + mval[-1].br_blockcount);
3993 mval[-1].br_blockcount += mval->br_blockcount;
3994 } else if (*n > 0 &&
3995 mval->br_startblock == DELAYSTARTBLOCK &&
3996 mval[-1].br_startblock == DELAYSTARTBLOCK &&
3997 mval->br_startoff ==
3998 mval[-1].br_startoff + mval[-1].br_blockcount) {
3999 mval[-1].br_blockcount += mval->br_blockcount;
4000 mval[-1].br_state = mval->br_state;
4001 } else if (!((*n == 0) &&
4002 ((mval->br_startoff + mval->br_blockcount) <=
4003 obno))) {
4004 mval++;
4005 (*n)++;
4006 }
4007 *map = mval;
e04426b9
DC		 4008}
4009
4010/*
9e5987a7 4011 * Map file blocks to filesystem blocks without allocation.
e04426b9
DC		 4012 */
4013int
9e5987a7
DC		 4014xfs_bmapi_read(
4015 struct xfs_inode *ip,
4016 xfs_fileoff_t bno,
b447fe5a 4017 xfs_filblks_t len,
9e5987a7
DC		 4018 struct xfs_bmbt_irec *mval,
4019 int *nmap,
c315c90b 4020 int flags)
b447fe5a 4021{
9e5987a7
DC		 4022 struct xfs_mount *mp = ip->i_mount;
4023 struct xfs_ifork *ifp;
4024 struct xfs_bmbt_irec got;
4025 struct xfs_bmbt_irec prev;
4026 xfs_fileoff_t obno;
4027 xfs_fileoff_t end;
4028 xfs_extnum_t lastx;
4029 int error;
4030 int eof;
4031 int n = 0;
b447fe5a
DC		 4032 int whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4033 XFS_ATTR_FORK : XFS_DATA_FORK;
b447fe5a 4034
9e5987a7
DC		 4035 ASSERT(*nmap >= 1);
4036 ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
4037 XFS_BMAPI_IGSTATE)));
eef334e5 4038 ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL));
b447fe5a 4039
9e5987a7
DC		 4040 if (unlikely(XFS_TEST_ERROR(
4041 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4042 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4043 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4044 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
2451337d 4045 return -EFSCORRUPTED;
9e5987a7 4046 }
b447fe5a 4047
9e5987a7 4048 if (XFS_FORCED_SHUTDOWN(mp))
2451337d 4049 return -EIO;
9e5987a7
DC		 4050
4051 XFS_STATS_INC(xs_blk_mapr);
4052
4053 ifp = XFS_IFORK_PTR(ip, whichfork);
4054
4055 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4056 error = xfs_iread_extents(NULL, ip, whichfork);
4057 if (error)
4058 return error;
b447fe5a 4059 }
b447fe5a 4060
9e5987a7
DC		 4061 xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got, &prev);
4062 end = bno + len;
4063 obno = bno;
b447fe5a 4064
9e5987a7
DC		 4065 while (bno < end && n < *nmap) {
4066 /* Reading past eof, act as though there's a hole up to end. */
4067 if (eof)
4068 got.br_startoff = end;
4069 if (got.br_startoff > bno) {
4070 /* Reading in a hole. */
4071 mval->br_startoff = bno;
4072 mval->br_startblock = HOLESTARTBLOCK;
4073 mval->br_blockcount =
4074 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4075 mval->br_state = XFS_EXT_NORM;
4076 bno += mval->br_blockcount;
4077 len -= mval->br_blockcount;
4078 mval++;
4079 n++;
4080 continue;
4081 }
b447fe5a 4082
9e5987a7
DC		 4083 /* set up the extent map to return. */
4084 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4085 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4086
4087 /* If we're done, stop now. */
4088 if (bno >= end || n >= *nmap)
4089 break;
4090
4091 /* Else go on to the next record. */
4092 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4093 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4094 else
4095 eof = 1;
4096 }
4097 *nmap = n;
b447fe5a
DC		 4098 return 0;
4099}
4100
9e5987a7
DC		 4101STATIC int
4102xfs_bmapi_reserve_delalloc(
4103 struct xfs_inode *ip,
4104 xfs_fileoff_t aoff,
4105 xfs_filblks_t len,
4106 struct xfs_bmbt_irec *got,
4107 struct xfs_bmbt_irec *prev,
4108 xfs_extnum_t *lastx,
4109 int eof)
1da177e4 4110{
c0dc7828 4111 struct xfs_mount *mp = ip->i_mount;
9e5987a7
DC		 4112 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4113 xfs_extlen_t alen;
4114 xfs_extlen_t indlen;
4115 char rt = XFS_IS_REALTIME_INODE(ip);
4116 xfs_extlen_t extsz;
4117 int error;
c0dc7828 4118
9e5987a7
DC		 4119 alen = XFS_FILBLKS_MIN(len, MAXEXTLEN);
4120 if (!eof)
4121 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
1da177e4 4122
9e5987a7
DC		 4123 /* Figure out the extent size, adjust alen */
4124 extsz = xfs_get_extsz_hint(ip);
4125 if (extsz) {
4126 /*
4127 * Make sure we don't exceed a single extent length when we
4128 * align the extent by reducing length we are going to
4129 * allocate by the maximum amount extent size aligment may
4130 * require.
4131 */
4132 alen = XFS_FILBLKS_MIN(len, MAXEXTLEN - (2 * extsz - 1));
4133 error = xfs_bmap_extsize_align(mp, got, prev, extsz, rt, eof,
4134 1, 0, &aoff, &alen);
4135 ASSERT(!error);
4136 }
4137
4138 if (rt)
4139 extsz = alen / mp->m_sb.sb_rextsize;
4140
4141 /*
4142 * Make a transaction-less quota reservation for delayed allocation
4143 * blocks. This number gets adjusted later. We return if we haven't
4144 * allocated blocks already inside this loop.
4145 */
4146 error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4147 rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4148 if (error)
4149 return error;
4150
4151 /*
4152 * Split changing sb for alen and indlen since they could be coming
4153 * from different places.
4154 */
4155 indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4156 ASSERT(indlen > 0);
4157
4158 if (rt) {
4159 error = xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
4160 -((int64_t)extsz), 0);
4161 } else {
4162 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4163 -((int64_t)alen), 0);
4164 }
4165
4166 if (error)
4167 goto out_unreserve_quota;
4168
4169 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4170 -((int64_t)indlen), 0);
4171 if (error)
4172 goto out_unreserve_blocks;
4173
4174
4175 ip->i_delayed_blks += alen;
4176
4177 got->br_startoff = aoff;
4178 got->br_startblock = nullstartblock(indlen);
4179 got->br_blockcount = alen;
4180 got->br_state = XFS_EXT_NORM;
4181 xfs_bmap_add_extent_hole_delay(ip, lastx, got);
4182
4183 /*
4184 * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4185 * might have merged it into one of the neighbouring ones.
4186 */
4187 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), got);
4188
4189 ASSERT(got->br_startoff <= aoff);
4190 ASSERT(got->br_startoff + got->br_blockcount >= aoff + alen);
4191 ASSERT(isnullstartblock(got->br_startblock));
4192 ASSERT(got->br_state == XFS_EXT_NORM);
4193 return 0;
4194
4195out_unreserve_blocks:
4196 if (rt)
4197 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS, extsz, 0);
4198 else
4199 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, alen, 0);
4200out_unreserve_quota:
4201 if (XFS_IS_QUOTA_ON(mp))
4202 xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ?
4203 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4204 return error;
4205}
4206
4207/*
4208 * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4209 */
4210int
4211xfs_bmapi_delay(
4212 struct xfs_inode *ip, /* incore inode */
4213 xfs_fileoff_t bno, /* starting file offs. mapped */
4214 xfs_filblks_t len, /* length to map in file */
4215 struct xfs_bmbt_irec *mval, /* output: map values */
4216 int *nmap, /* i/o: mval size/count */
4217 int flags) /* XFS_BMAPI_... */
4218{
4219 struct xfs_mount *mp = ip->i_mount;
4220 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4221 struct xfs_bmbt_irec got; /* current file extent record */
4222 struct xfs_bmbt_irec prev; /* previous file extent record */
4223 xfs_fileoff_t obno; /* old block number (offset) */
4224 xfs_fileoff_t end; /* end of mapped file region */
4225 xfs_extnum_t lastx; /* last useful extent number */
4226 int eof; /* we've hit the end of extents */
4227 int n = 0; /* current extent index */
4228 int error = 0;
c0dc7828 4229
1da177e4 4230 ASSERT(*nmap >= 1);
c0dc7828 4231 ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
9e5987a7 4232 ASSERT(!(flags & ~XFS_BMAPI_ENTIRE));
eef334e5 4233 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
c0dc7828 4234
1da177e4 4235 if (unlikely(XFS_TEST_ERROR(
9e5987a7
DC		 4236 (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4237 XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
1da177e4 4238 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
9e5987a7 4239 XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW, mp);
2451337d 4240 return -EFSCORRUPTED;
1da177e4 4241 }
c0dc7828 4242
1da177e4 4243 if (XFS_FORCED_SHUTDOWN(mp))
2451337d 4244 return -EIO;
c0dc7828 4245
c0dc7828
DC		 4246 XFS_STATS_INC(xs_blk_mapw);
4247
c0dc7828 4248 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
9e5987a7 4249 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
c0dc7828 4250 if (error)
9e5987a7 4251 return error;
c0dc7828
DC		 4252 }
4253
9e5987a7 4254 xfs_bmap_search_extents(ip, bno, XFS_DATA_FORK, &eof, &lastx, &got, &prev);
1da177e4
LT		 4255 end = bno + len;
4256 obno = bno;
7e47a4ef 4257
1da177e4 4258 while (bno < end && n < *nmap) {
9e5987a7
DC		 4259 if (eof || got.br_startoff > bno) {
4260 error = xfs_bmapi_reserve_delalloc(ip, bno, len, &got,
4261 &prev, &lastx, eof);
4262 if (error) {
4263 if (n == 0) {
4264 *nmap = 0;
4265 return error;
4266 }
4267 break;
4268 }
4269 }
c0dc7828 4270
9e5987a7
DC		 4271 /* set up the extent map to return. */
4272 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
aef9a895
DC		 4273 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4274
9e5987a7
DC		 4275 /* If we're done, stop now. */
4276 if (bno >= end || n >= *nmap)
1da177e4 4277 break;
c0dc7828
DC		 4278
4279 /* Else go on to the next record. */
9e5987a7
DC		 4280 prev = got;
4281 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4282 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4283 else
5690f921 4284 eof = 1;
1da177e4 4285 }
9e5987a7 4286
1da177e4 4287 *nmap = n;
9e5987a7
DC		 4288 return 0;
4289}
c0dc7828 4290
c315c90b 4291
cf11da9c
DC		 4292static int
4293xfs_bmapi_allocate(
9e5987a7
DC		 4294 struct xfs_bmalloca *bma)
4295{
4296 struct xfs_mount *mp = bma->ip->i_mount;
4297 int whichfork = (bma->flags & XFS_BMAPI_ATTRFORK) ?
4298 XFS_ATTR_FORK : XFS_DATA_FORK;
4299 struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4300 int tmp_logflags = 0;
4301 int error;
9e5987a7
DC		 4302
4303 ASSERT(bma->length > 0);
4304
1da177e4 4305 /*
9e5987a7
DC		 4306 * For the wasdelay case, we could also just allocate the stuff asked
4307 * for in this bmap call but that wouldn't be as good.
1da177e4 4308 */
9e5987a7
DC		 4309 if (bma->wasdel) {
4310 bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4311 bma->offset = bma->got.br_startoff;
4312 if (bma->idx != NULLEXTNUM && bma->idx) {
4313 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4314 &bma->prev);
1da177e4 4315 }
9e5987a7
DC		 4316 } else {
4317 bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4318 if (!bma->eof)
4319 bma->length = XFS_FILBLKS_MIN(bma->length,
4320 bma->got.br_startoff - bma->offset);
1da177e4 4321 }
1da177e4 4322
9e5987a7
DC		 4323 /*
4324 * Indicate if this is the first user data in the file, or just any
4325 * user data.
4326 */
4327 if (!(bma->flags & XFS_BMAPI_METADATA)) {
4328 bma->userdata = (bma->offset == 0) ?
4329 XFS_ALLOC_INITIAL_USER_DATA : XFS_ALLOC_USERDATA;
4330 }
1da177e4 4331
9e5987a7 4332 bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
0b1b213f 4333
9e5987a7
DC		 4334 /*
4335 * Only want to do the alignment at the eof if it is userdata and
4336 * allocation length is larger than a stripe unit.
4337 */
4338 if (mp->m_dalign && bma->length >= mp->m_dalign &&
4339 !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4340 error = xfs_bmap_isaeof(bma, whichfork);
4341 if (error)
4342 return error;
1da177e4 4343 }
8096b1eb 4344
9e5987a7
DC		 4345 error = xfs_bmap_alloc(bma);
4346 if (error)
1da177e4 4347 return error;
9e5987a7
DC		 4348
4349 if (bma->flist->xbf_low)
4350 bma->minleft = 0;
4351 if (bma->cur)
4352 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4353 if (bma->blkno == NULLFSBLOCK)
1da177e4 4354 return 0;
9e5987a7
DC		 4355 if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4356 bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4357 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4358 bma->cur->bc_private.b.flist = bma->flist;
1da177e4 4359 }
9e5987a7
DC		 4360 /*
4361 * Bump the number of extents we've allocated
4362 * in this call.
4363 */
4364 bma->nallocs++;
4365
4366 if (bma->cur)
4367 bma->cur->bc_private.b.flags =
4368 bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4369
4370 bma->got.br_startoff = bma->offset;
4371 bma->got.br_startblock = bma->blkno;
4372 bma->got.br_blockcount = bma->length;
4373 bma->got.br_state = XFS_EXT_NORM;
b4e9181e 4374
1da177e4 4375 /*
9e5987a7
DC		 4376 * A wasdelay extent has been initialized, so shouldn't be flagged
4377 * as unwritten.
1da177e4 4378 */
9e5987a7
DC		 4379 if (!bma->wasdel && (bma->flags & XFS_BMAPI_PREALLOC) &&
4380 xfs_sb_version_hasextflgbit(&mp->m_sb))
4381 bma->got.br_state = XFS_EXT_UNWRITTEN;
4382
4383 if (bma->wasdel)
4384 error = xfs_bmap_add_extent_delay_real(bma);
4385 else
4386 error = xfs_bmap_add_extent_hole_real(bma, whichfork);
4387
4388 bma->logflags |= tmp_logflags;
4389 if (error)
4390 return error;
4391
4392 /*
4393 * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4394 * or xfs_bmap_add_extent_hole_real might have merged it into one of
4395 * the neighbouring ones.
4396 */
4397 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4398
4399 ASSERT(bma->got.br_startoff <= bma->offset);
4400 ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4401 bma->offset + bma->length);
4402 ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4403 bma->got.br_state == XFS_EXT_UNWRITTEN);
4404 return 0;
4405}
4406
9e5987a7
DC		 4407STATIC int
4408xfs_bmapi_convert_unwritten(
4409 struct xfs_bmalloca *bma,
4410 struct xfs_bmbt_irec *mval,
4411 xfs_filblks_t len,
4412 int flags)
4413{
4414 int whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4415 XFS_ATTR_FORK : XFS_DATA_FORK;
4416 struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4417 int tmp_logflags = 0;
4418 int error;
4419
4420 /* check if we need to do unwritten->real conversion */
4421 if (mval->br_state == XFS_EXT_UNWRITTEN &&
4422 (flags & XFS_BMAPI_PREALLOC))
4423 return 0;
4424
4425 /* check if we need to do real->unwritten conversion */
4426 if (mval->br_state == XFS_EXT_NORM &&
4427 (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4428 (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4429 return 0;
4430
4431 /*
4432 * Modify (by adding) the state flag, if writing.
4433 */
4434 ASSERT(mval->br_blockcount <= len);
4435 if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4436 bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4437 bma->ip, whichfork);
4438 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4439 bma->cur->bc_private.b.flist = bma->flist;
1da177e4 4440 }
9e5987a7
DC		 4441 mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4442 ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
5575acc7 4443
9e5987a7
DC		 4444 error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, &bma->idx,
4445 &bma->cur, mval, bma->firstblock, bma->flist,
4446 &tmp_logflags);
4447 bma->logflags |= tmp_logflags;
4448 if (error)
4449 return error;
4450
4451 /*
4452 * Update our extent pointer, given that
4453 * xfs_bmap_add_extent_unwritten_real might have merged it into one
4454 * of the neighbouring ones.
4455 */
4456 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4457
4458 /*
4459 * We may have combined previously unwritten space with written space,
4460 * so generate another request.
4461 */
4462 if (mval->br_blockcount < len)
2451337d 4463 return -EAGAIN;
9e5987a7
DC		 4464 return 0;
4465}
4466
4467/*
4468 * Map file blocks to filesystem blocks, and allocate blocks or convert the
4469 * extent state if necessary. Details behaviour is controlled by the flags
4470 * parameter. Only allocates blocks from a single allocation group, to avoid
4471 * locking problems.
4472 *
4473 * The returned value in "firstblock" from the first call in a transaction
4474 * must be remembered and presented to subsequent calls in "firstblock".
4475 * An upper bound for the number of blocks to be allocated is supplied to
4476 * the first call in "total"; if no allocation group has that many free
4477 * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4478 */
4479int
4480xfs_bmapi_write(
4481 struct xfs_trans *tp, /* transaction pointer */
4482 struct xfs_inode *ip, /* incore inode */
4483 xfs_fileoff_t bno, /* starting file offs. mapped */
4484 xfs_filblks_t len, /* length to map in file */
4485 int flags, /* XFS_BMAPI_... */
4486 xfs_fsblock_t *firstblock, /* first allocated block
4487 controls a.g. for allocs */
4488 xfs_extlen_t total, /* total blocks needed */
4489 struct xfs_bmbt_irec *mval, /* output: map values */
4490 int *nmap, /* i/o: mval size/count */
4491 struct xfs_bmap_free *flist) /* i/o: list extents to free */
4492{
4493 struct xfs_mount *mp = ip->i_mount;
4494 struct xfs_ifork *ifp;
a30b0367 4495 struct xfs_bmalloca bma = { NULL }; /* args for xfs_bmap_alloc */
9e5987a7
DC		 4496 xfs_fileoff_t end; /* end of mapped file region */
4497 int eof; /* after the end of extents */
4498 int error; /* error return */
4499 int n; /* current extent index */
4500 xfs_fileoff_t obno; /* old block number (offset) */
4501 int whichfork; /* data or attr fork */
4502 char inhole; /* current location is hole in file */
4503 char wasdelay; /* old extent was delayed */
4504
4505#ifdef DEBUG
4506 xfs_fileoff_t orig_bno; /* original block number value */
4507 int orig_flags; /* original flags arg value */
4508 xfs_filblks_t orig_len; /* original value of len arg */
4509 struct xfs_bmbt_irec *orig_mval; /* original value of mval */
4510 int orig_nmap; /* original value of *nmap */
4511
4512 orig_bno = bno;
4513 orig_len = len;
4514 orig_flags = flags;
4515 orig_mval = mval;
4516 orig_nmap = *nmap;
4517#endif
f3508bcd
DC		 4518 whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4519 XFS_ATTR_FORK : XFS_DATA_FORK;
9e5987a7
DC		 4520
4521 ASSERT(*nmap >= 1);
4522 ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4523 ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4524 ASSERT(tp != NULL);
4525 ASSERT(len > 0);
f3508bcd 4526 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL);
eef334e5 4527 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
9e5987a7
DC		 4528
4529 if (unlikely(XFS_TEST_ERROR(
4530 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
f3508bcd 4531 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
9e5987a7
DC		 4532 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4533 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
2451337d 4534 return -EFSCORRUPTED;
5575acc7
KD		 4535 }
4536
9e5987a7 4537 if (XFS_FORCED_SHUTDOWN(mp))
2451337d 4538 return -EIO;
9e5987a7
DC		 4539
4540 ifp = XFS_IFORK_PTR(ip, whichfork);
4541
4542 XFS_STATS_INC(xs_blk_mapw);
4543
9e5987a7
DC		 4544 if (*firstblock == NULLFSBLOCK) {
4545 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4546 bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4547 else
4548 bma.minleft = 1;
4549 } else {
4550 bma.minleft = 0;
4551 }
4552
4553 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4554 error = xfs_iread_extents(tp, ip, whichfork);
4555 if (error)
4556 goto error0;
4557 }
4558
4559 xfs_bmap_search_extents(ip, bno, whichfork, &eof, &bma.idx, &bma.got,
4560 &bma.prev);
4561 n = 0;
4562 end = bno + len;
4563 obno = bno;
4564
4565 bma.tp = tp;
4566 bma.ip = ip;
4567 bma.total = total;
4568 bma.userdata = 0;
4569 bma.flist = flist;
4570 bma.firstblock = firstblock;
4571
9e5987a7
DC		 4572 while (bno < end && n < *nmap) {
4573 inhole = eof || bma.got.br_startoff > bno;
4574 wasdelay = !inhole && isnullstartblock(bma.got.br_startblock);
4575
1da177e4 4576 /*
9e5987a7
DC		 4577 * First, deal with the hole before the allocated space
4578 * that we found, if any.
1da177e4 4579 */
9e5987a7
DC		 4580 if (inhole || wasdelay) {
4581 bma.eof = eof;
4582 bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4583 bma.wasdel = wasdelay;
4584 bma.offset = bno;
4585 bma.flags = flags;
4586
1da177e4 4587 /*
9e5987a7
DC		 4588 * There's a 32/64 bit type mismatch between the
4589 * allocation length request (which can be 64 bits in
4590 * length) and the bma length request, which is
4591 * xfs_extlen_t and therefore 32 bits. Hence we have to
4592 * check for 32-bit overflows and handle them here.
1da177e4 4593 */
9e5987a7
DC		 4594 if (len > (xfs_filblks_t)MAXEXTLEN)
4595 bma.length = MAXEXTLEN;
4596 else
4597 bma.length = len;
4598
4599 ASSERT(len > 0);
4600 ASSERT(bma.length > 0);
4601 error = xfs_bmapi_allocate(&bma);
1da177e4
LT		 4602 if (error)
4603 goto error0;
9e5987a7
DC		 4604 if (bma.blkno == NULLFSBLOCK)
4605 break;
1da177e4 4606 }
9e5987a7
DC		 4607
4608 /* Deal with the allocated space we found. */
4609 xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4610 end, n, flags);
4611
4612 /* Execute unwritten extent conversion if necessary */
4613 error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
2451337d 4614 if (error == -EAGAIN)
9e5987a7
DC		 4615 continue;
4616 if (error)
4617 goto error0;
4618
4619 /* update the extent map to return */
4620 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4621
4622 /*
4623 * If we're done, stop now. Stop when we've allocated
4624 * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise
4625 * the transaction may get too big.
4626 */
4627 if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4628 break;
4629
4630 /* Else go on to the next record. */
4631 bma.prev = bma.got;
4632 if (++bma.idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t)) {
4633 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma.idx),
4634 &bma.got);
4635 } else
4636 eof = 1;
4637 }
4638 *nmap = n;
4639
4640 /*
4641 * Transform from btree to extents, give it cur.
4642 */
4643 if (xfs_bmap_wants_extents(ip, whichfork)) {
4644 int tmp_logflags = 0;
4645
4646 ASSERT(bma.cur);
4647 error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
4648 &tmp_logflags, whichfork);
4649 bma.logflags |= tmp_logflags;
4650 if (error)
4651 goto error0;
4652 }
4653
4654 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4655 XFS_IFORK_NEXTENTS(ip, whichfork) >
4656 XFS_IFORK_MAXEXT(ip, whichfork));
4657 error = 0;
4658error0:
4659 /*
4660 * Log everything. Do this after conversion, there's no point in
4661 * logging the extent records if we've converted to btree format.
4662 */
4663 if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
4664 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4665 bma.logflags &= ~xfs_ilog_fext(whichfork);
4666 else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
4667 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4668 bma.logflags &= ~xfs_ilog_fbroot(whichfork);
4669 /*
4670 * Log whatever the flags say, even if error. Otherwise we might miss
4671 * detecting a case where the data is changed, there's an error,
4672 * and it's not logged so we don't shutdown when we should.
4673 */
4674 if (bma.logflags)
4675 xfs_trans_log_inode(tp, ip, bma.logflags);
4676
4677 if (bma.cur) {
4678 if (!error) {
4679 ASSERT(*firstblock == NULLFSBLOCK ||
4680 XFS_FSB_TO_AGNO(mp, *firstblock) ==
4681 XFS_FSB_TO_AGNO(mp,
4682 bma.cur->bc_private.b.firstblock) ||
4683 (flist->xbf_low &&
4684 XFS_FSB_TO_AGNO(mp, *firstblock) <
4685 XFS_FSB_TO_AGNO(mp,
4686 bma.cur->bc_private.b.firstblock)));
4687 *firstblock = bma.cur->bc_private.b.firstblock;
1da177e4 4688 }
9e5987a7
DC		 4689 xfs_btree_del_cursor(bma.cur,
4690 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4691 }
4692 if (!error)
4693 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4694 orig_nmap, *nmap);
4695 return error;
4696}
06d10dd9 4697
9e5987a7
DC		 4698/*
4699 * Called by xfs_bmapi to update file extent records and the btree
4700 * after removing space (or undoing a delayed allocation).
4701 */
4702STATIC int /* error */
4703xfs_bmap_del_extent(
4704 xfs_inode_t *ip, /* incore inode pointer */
4705 xfs_trans_t *tp, /* current transaction pointer */
4706 xfs_extnum_t *idx, /* extent number to update/delete */
4707 xfs_bmap_free_t *flist, /* list of extents to be freed */
4708 xfs_btree_cur_t *cur, /* if null, not a btree */
4709 xfs_bmbt_irec_t *del, /* data to remove from extents */
4710 int *logflagsp, /* inode logging flags */
4711 int whichfork) /* data or attr fork */
4712{
4713 xfs_filblks_t da_new; /* new delay-alloc indirect blocks */
4714 xfs_filblks_t da_old; /* old delay-alloc indirect blocks */
4715 xfs_fsblock_t del_endblock=0; /* first block past del */
4716 xfs_fileoff_t del_endoff; /* first offset past del */
4717 int delay; /* current block is delayed allocated */
4718 int do_fx; /* free extent at end of routine */
4719 xfs_bmbt_rec_host_t *ep; /* current extent entry pointer */
4720 int error; /* error return value */
4721 int flags; /* inode logging flags */
4722 xfs_bmbt_irec_t got; /* current extent entry */
4723 xfs_fileoff_t got_endoff; /* first offset past got */
4724 int i; /* temp state */
4725 xfs_ifork_t *ifp; /* inode fork pointer */
4726 xfs_mount_t *mp; /* mount structure */
4727 xfs_filblks_t nblks; /* quota/sb block count */
4728 xfs_bmbt_irec_t new; /* new record to be inserted */
4729 /* REFERENCED */
4730 uint qfield; /* quota field to update */
4731 xfs_filblks_t temp; /* for indirect length calculations */
4732 xfs_filblks_t temp2; /* for indirect length calculations */
4733 int state = 0;
4734
4735 XFS_STATS_INC(xs_del_exlist);
4736
4737 if (whichfork == XFS_ATTR_FORK)
4738 state |= BMAP_ATTRFORK;
4739
4740 mp = ip->i_mount;
4741 ifp = XFS_IFORK_PTR(ip, whichfork);
4742 ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
4743 (uint)sizeof(xfs_bmbt_rec_t)));
4744 ASSERT(del->br_blockcount > 0);
4745 ep = xfs_iext_get_ext(ifp, *idx);
4746 xfs_bmbt_get_all(ep, &got);
4747 ASSERT(got.br_startoff <= del->br_startoff);
4748 del_endoff = del->br_startoff + del->br_blockcount;
4749 got_endoff = got.br_startoff + got.br_blockcount;
4750 ASSERT(got_endoff >= del_endoff);
4751 delay = isnullstartblock(got.br_startblock);
4752 ASSERT(isnullstartblock(del->br_startblock) == delay);
4753 flags = 0;
4754 qfield = 0;
4755 error = 0;
4756 /*
4757 * If deleting a real allocation, must free up the disk space.
4758 */
4759 if (!delay) {
4760 flags = XFS_ILOG_CORE;
4761 /*
4762 * Realtime allocation. Free it and record di_nblocks update.
4763 */
4764 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
4765 xfs_fsblock_t bno;
4766 xfs_filblks_t len;
4767
4768 ASSERT(do_mod(del->br_blockcount,
4769 mp->m_sb.sb_rextsize) == 0);
4770 ASSERT(do_mod(del->br_startblock,
4771 mp->m_sb.sb_rextsize) == 0);
4772 bno = del->br_startblock;
4773 len = del->br_blockcount;
4774 do_div(bno, mp->m_sb.sb_rextsize);
4775 do_div(len, mp->m_sb.sb_rextsize);
4776 error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
4777 if (error)
4778 goto done;
4779 do_fx = 0;
4780 nblks = len * mp->m_sb.sb_rextsize;
4781 qfield = XFS_TRANS_DQ_RTBCOUNT;
4782 }
1da177e4 4783 /*
9e5987a7 4784 * Ordinary allocation.
1da177e4 4785 */
9e5987a7
DC		 4786 else {
4787 do_fx = 1;
4788 nblks = del->br_blockcount;
4789 qfield = XFS_TRANS_DQ_BCOUNT;
1da177e4 4790 }
1da177e4 4791 /*
9e5987a7 4792 * Set up del_endblock and cur for later.
1da177e4 4793 */
9e5987a7
DC		 4794 del_endblock = del->br_startblock + del->br_blockcount;
4795 if (cur) {
4796 if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
4797 got.br_startblock, got.br_blockcount,
4798 &i)))
4799 goto done;
4800 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1da177e4 4801 }
9e5987a7
DC		 4802 da_old = da_new = 0;
4803 } else {
4804 da_old = startblockval(got.br_startblock);
4805 da_new = 0;
4806 nblks = 0;
4807 do_fx = 0;
1da177e4
LT		 4808 }
4809 /*
9e5987a7
DC		 4810 * Set flag value to use in switch statement.
4811 * Left-contig is 2, right-contig is 1.
1da177e4 4812 */
9e5987a7
DC		 4813 switch (((got.br_startoff == del->br_startoff) << 1) |
4814 (got_endoff == del_endoff)) {
4815 case 3:
4816 /*
4817 * Matches the whole extent. Delete the entry.
4818 */
4819 xfs_iext_remove(ip, *idx, 1,
4820 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
4821 --*idx;
4822 if (delay)
4823 break;
4824
4825 XFS_IFORK_NEXT_SET(ip, whichfork,
4826 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
4827 flags |= XFS_ILOG_CORE;
4828 if (!cur) {
4829 flags |= xfs_ilog_fext(whichfork);
4830 break;
4831 }
4832 if ((error = xfs_btree_delete(cur, &i)))
4833 goto done;
4834 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4835 break;
4836
4837 case 2:
4838 /*
4839 * Deleting the first part of the extent.
4840 */
4841 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4842 xfs_bmbt_set_startoff(ep, del_endoff);
4843 temp = got.br_blockcount - del->br_blockcount;
4844 xfs_bmbt_set_blockcount(ep, temp);
4845 if (delay) {
4846 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4847 da_old);
4848 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4849 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4850 da_new = temp;
4851 break;
4852 }
4853 xfs_bmbt_set_startblock(ep, del_endblock);
4854 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4855 if (!cur) {
4856 flags |= xfs_ilog_fext(whichfork);
4857 break;
4858 }
4859 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
4860 got.br_blockcount - del->br_blockcount,
4861 got.br_state)))
4862 goto done;
4863 break;
4864
4865 case 1:
4866 /*
4867 * Deleting the last part of the extent.
4868 */
4869 temp = got.br_blockcount - del->br_blockcount;
4870 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4871 xfs_bmbt_set_blockcount(ep, temp);
4872 if (delay) {
4873 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4874 da_old);
4875 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4876 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4877 da_new = temp;
4878 break;
4879 }
4880 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4881 if (!cur) {
4882 flags |= xfs_ilog_fext(whichfork);
4883 break;
4884 }
4885 if ((error = xfs_bmbt_update(cur, got.br_startoff,
4886 got.br_startblock,
4887 got.br_blockcount - del->br_blockcount,
4888 got.br_state)))
4889 goto done;
4890 break;
4891
4892 case 0:
4893 /*
4894 * Deleting the middle of the extent.
4895 */
4896 temp = del->br_startoff - got.br_startoff;
4897 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4898 xfs_bmbt_set_blockcount(ep, temp);
4899 new.br_startoff = del_endoff;
4900 temp2 = got_endoff - del_endoff;
4901 new.br_blockcount = temp2;
4902 new.br_state = got.br_state;
4903 if (!delay) {
4904 new.br_startblock = del_endblock;
4905 flags |= XFS_ILOG_CORE;
4906 if (cur) {
4907 if ((error = xfs_bmbt_update(cur,
4908 got.br_startoff,
4909 got.br_startblock, temp,
4910 got.br_state)))
4911 goto done;
4912 if ((error = xfs_btree_increment(cur, 0, &i)))
4913 goto done;
4914 cur->bc_rec.b = new;
4915 error = xfs_btree_insert(cur, &i);
2451337d 4916 if (error && error != -ENOSPC)
9e5987a7
DC		 4917 goto done;
4918 /*
4919 * If get no-space back from btree insert,
4920 * it tried a split, and we have a zero
4921 * block reservation.
4922 * Fix up our state and return the error.
4923 */
2451337d 4924 if (error == -ENOSPC) {
9e5987a7
DC		 4925 /*
4926 * Reset the cursor, don't trust
4927 * it after any insert operation.
4928 */
4929 if ((error = xfs_bmbt_lookup_eq(cur,
4930 got.br_startoff,
4931 got.br_startblock,
4932 temp, &i)))
4933 goto done;
4934 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4935 /*
4936 * Update the btree record back
4937 * to the original value.
4938 */
4939 if ((error = xfs_bmbt_update(cur,
4940 got.br_startoff,
4941 got.br_startblock,
4942 got.br_blockcount,
4943 got.br_state)))
4944 goto done;
4945 /*
4946 * Reset the extent record back
4947 * to the original value.
4948 */
4949 xfs_bmbt_set_blockcount(ep,
4950 got.br_blockcount);
4951 flags = 0;
2451337d 4952 error = -ENOSPC;
9e5987a7
DC		 4953 goto done;
4954 }
4955 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4956 } else
4957 flags |= xfs_ilog_fext(whichfork);
4958 XFS_IFORK_NEXT_SET(ip, whichfork,
4959 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
4960 } else {
4961 ASSERT(whichfork == XFS_DATA_FORK);
4962 temp = xfs_bmap_worst_indlen(ip, temp);
4963 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4964 temp2 = xfs_bmap_worst_indlen(ip, temp2);
4965 new.br_startblock = nullstartblock((int)temp2);
4966 da_new = temp + temp2;
4967 while (da_new > da_old) {
4968 if (temp) {
4969 temp--;
4970 da_new--;
4971 xfs_bmbt_set_startblock(ep,
4972 nullstartblock((int)temp));
4973 }
4974 if (da_new == da_old)
4975 break;
4976 if (temp2) {
4977 temp2--;
4978 da_new--;
4979 new.br_startblock =
4980 nullstartblock((int)temp2);
4981 }
4982 }
4983 }
4984 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4985 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
4986 ++*idx;
4987 break;
1da177e4
LT		 4988 }
4989 /*
9e5987a7 4990 * If we need to, add to list of extents to delete.
1da177e4 4991 */
9e5987a7
DC		 4992 if (do_fx)
4993 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
4994 mp);
1da177e4 4995 /*
9e5987a7 4996 * Adjust inode # blocks in the file.
1da177e4 4997 */
9e5987a7
DC		 4998 if (nblks)
4999 ip->i_d.di_nblocks -= nblks;
1da177e4 5000 /*
9e5987a7 5001 * Adjust quota data.
1da177e4 5002 */
9e5987a7
DC		 5003 if (qfield)
5004 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5005
5006 /*
5007 * Account for change in delayed indirect blocks.
5008 * Nothing to do for disk quota accounting here.
5009 */
5010 ASSERT(da_old >= da_new);
5011 if (da_old > da_new) {
5012 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5013 (int64_t)(da_old - da_new), 0);
1da177e4 5014 }
9e5987a7
DC		 5015done:
5016 *logflagsp = flags;
1da177e4
LT		 5017 return error;
5018}
5019
3bacbcd8 5020/*
9e5987a7
DC		 5021 * Unmap (remove) blocks from a file.
5022 * If nexts is nonzero then the number of extents to remove is limited to
5023 * that value. If not all extents in the block range can be removed then
5024 * *done is set.
3bacbcd8 5025 */
9e5987a7
DC		 5026int /* error */
5027xfs_bunmapi(
5028 xfs_trans_t *tp, /* transaction pointer */
5029 struct xfs_inode *ip, /* incore inode */
5030 xfs_fileoff_t bno, /* starting offset to unmap */
5031 xfs_filblks_t len, /* length to unmap in file */
5032 int flags, /* misc flags */
5033 xfs_extnum_t nexts, /* number of extents max */
5034 xfs_fsblock_t *firstblock, /* first allocated block
5035 controls a.g. for allocs */
5036 xfs_bmap_free_t *flist, /* i/o: list extents to free */
5037 int *done) /* set if not done yet */
3bacbcd8 5038{
9e5987a7
DC		 5039 xfs_btree_cur_t *cur; /* bmap btree cursor */
5040 xfs_bmbt_irec_t del; /* extent being deleted */
5041 int eof; /* is deleting at eof */
5042 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
5043 int error; /* error return value */
5044 xfs_extnum_t extno; /* extent number in list */
5045 xfs_bmbt_irec_t got; /* current extent record */
5af317c9 5046 xfs_ifork_t *ifp; /* inode fork pointer */
9e5987a7
DC		 5047 int isrt; /* freeing in rt area */
5048 xfs_extnum_t lastx; /* last extent index used */
5049 int logflags; /* transaction logging flags */
5050 xfs_extlen_t mod; /* rt extent offset */
5051 xfs_mount_t *mp; /* mount structure */
5052 xfs_extnum_t nextents; /* number of file extents */
5053 xfs_bmbt_irec_t prev; /* previous extent record */
5054 xfs_fileoff_t start; /* first file offset deleted */
5055 int tmp_logflags; /* partial logging flags */
5056 int wasdel; /* was a delayed alloc extent */
5057 int whichfork; /* data or attribute fork */
5058 xfs_fsblock_t sum;
1da177e4 5059
9e5987a7 5060 trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
1da177e4 5061
9e5987a7
DC		 5062 whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5063 XFS_ATTR_FORK : XFS_DATA_FORK;
5064 ifp = XFS_IFORK_PTR(ip, whichfork);
5065 if (unlikely(
5066 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5067 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5068 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5069 ip->i_mount);
2451337d 5070 return -EFSCORRUPTED;
1da177e4 5071 }
9e5987a7
DC		 5072 mp = ip->i_mount;
5073 if (XFS_FORCED_SHUTDOWN(mp))
2451337d 5074 return -EIO;
1da177e4 5075
eef334e5 5076 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
9e5987a7
DC		 5077 ASSERT(len > 0);
5078 ASSERT(nexts >= 0);
1da177e4 5079
9e5987a7
DC		 5080 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5081 (error = xfs_iread_extents(tp, ip, whichfork)))
5082 return error;
5083 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5084 if (nextents == 0) {
5085 *done = 1;
5086 return 0;
5087 }
5088 XFS_STATS_INC(xs_blk_unmap);
5089 isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5090 start = bno;
5091 bno = start + len - 1;
5092 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5093 &prev);
1da177e4 5094
9e5987a7
DC		 5095 /*
5096 * Check to see if the given block number is past the end of the
5097 * file, back up to the last block if so...
5098 */
5099 if (eof) {
5100 ep = xfs_iext_get_ext(ifp, --lastx);
5101 xfs_bmbt_get_all(ep, &got);
5102 bno = got.br_startoff + got.br_blockcount - 1;
5103 }
5104 logflags = 0;
5105 if (ifp->if_flags & XFS_IFBROOT) {
5106 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5107 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5108 cur->bc_private.b.firstblock = *firstblock;
5109 cur->bc_private.b.flist = flist;
5110 cur->bc_private.b.flags = 0;
5111 } else
5112 cur = NULL;
5113
5114 if (isrt) {
5115 /*
5116 * Synchronize by locking the bitmap inode.
5117 */
5118 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
5119 xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5120 }
58e20770 5121
9e5987a7
DC		 5122 extno = 0;
5123 while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5124 (nexts == 0 || extno < nexts)) {
5125 /*
5126 * Is the found extent after a hole in which bno lives?
5127 * Just back up to the previous extent, if so.
5128 */
5129 if (got.br_startoff > bno) {
5130 if (--lastx < 0)
5131 break;
5132 ep = xfs_iext_get_ext(ifp, lastx);
5133 xfs_bmbt_get_all(ep, &got);
5134 }
5135 /*
5136 * Is the last block of this extent before the range
5137 * we're supposed to delete? If so, we're done.
5138 */
5139 bno = XFS_FILEOFF_MIN(bno,
5140 got.br_startoff + got.br_blockcount - 1);
5141 if (bno < start)
5142 break;
5143 /*
5144 * Then deal with the (possibly delayed) allocated space
5145 * we found.
5146 */
5147 ASSERT(ep != NULL);
5148 del = got;
5149 wasdel = isnullstartblock(del.br_startblock);
5150 if (got.br_startoff < start) {
5151 del.br_startoff = start;
5152 del.br_blockcount -= start - got.br_startoff;
5153 if (!wasdel)
5154 del.br_startblock += start - got.br_startoff;
5155 }
5156 if (del.br_startoff + del.br_blockcount > bno + 1)
5157 del.br_blockcount = bno + 1 - del.br_startoff;
5158 sum = del.br_startblock + del.br_blockcount;
5159 if (isrt &&
5160 (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
58e20770 5161 /*
9e5987a7
DC		 5162 * Realtime extent not lined up at the end.
5163 * The extent could have been split into written
5164 * and unwritten pieces, or we could just be
5165 * unmapping part of it. But we can't really
5166 * get rid of part of a realtime extent.
58e20770 5167 */
9e5987a7
DC		 5168 if (del.br_state == XFS_EXT_UNWRITTEN ||
5169 !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5170 /*
5171 * This piece is unwritten, or we're not
5172 * using unwritten extents. Skip over it.
5173 */
5174 ASSERT(bno >= mod);
5175 bno -= mod > del.br_blockcount ?
5176 del.br_blockcount : mod;
5177 if (bno < got.br_startoff) {
5178 if (--lastx >= 0)
5179 xfs_bmbt_get_all(xfs_iext_get_ext(
5180 ifp, lastx), &got);
5181 }
5182 continue;
1da177e4 5183 }
9af25465 5184 /*
9e5987a7
DC		 5185 * It's written, turn it unwritten.
5186 * This is better than zeroing it.
9af25465 5187 */
9e5987a7
DC		 5188 ASSERT(del.br_state == XFS_EXT_NORM);
5189 ASSERT(xfs_trans_get_block_res(tp) > 0);
5190 /*
5191 * If this spans a realtime extent boundary,
5192 * chop it back to the start of the one we end at.
5193 */
5194 if (del.br_blockcount > mod) {
5195 del.br_startoff += del.br_blockcount - mod;
5196 del.br_startblock += del.br_blockcount - mod;
5197 del.br_blockcount = mod;
5198 }
5199 del.br_state = XFS_EXT_UNWRITTEN;
5200 error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5201 &lastx, &cur, &del, firstblock, flist,
5202 &logflags);
5203 if (error)
5204 goto error0;
5205 goto nodelete;
5206 }
5207 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5208 /*
5209 * Realtime extent is lined up at the end but not
5210 * at the front. We'll get rid of full extents if
5211 * we can.
5212 */
5213 mod = mp->m_sb.sb_rextsize - mod;
5214 if (del.br_blockcount > mod) {
5215 del.br_blockcount -= mod;
5216 del.br_startoff += mod;
5217 del.br_startblock += mod;
5218 } else if ((del.br_startoff == start &&
5219 (del.br_state == XFS_EXT_UNWRITTEN ||
5220 xfs_trans_get_block_res(tp) == 0)) ||
5221 !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5222 /*
5223 * Can't make it unwritten. There isn't
5224 * a full extent here so just skip it.
5225 */
5226 ASSERT(bno >= del.br_blockcount);
5227 bno -= del.br_blockcount;
5228 if (got.br_startoff > bno) {
5229 if (--lastx >= 0) {
5230 ep = xfs_iext_get_ext(ifp,
5231 lastx);
5232 xfs_bmbt_get_all(ep, &got);
5233 }
5234 }
9af25465 5235 continue;
9e5987a7
DC		 5236 } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5237 /*
5238 * This one is already unwritten.
5239 * It must have a written left neighbor.
5240 * Unwrite the killed part of that one and
5241 * try again.
5242 */
5243 ASSERT(lastx > 0);
5244 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5245 lastx - 1), &prev);
5246 ASSERT(prev.br_state == XFS_EXT_NORM);
5247 ASSERT(!isnullstartblock(prev.br_startblock));
5248 ASSERT(del.br_startblock ==
5249 prev.br_startblock + prev.br_blockcount);
5250 if (prev.br_startoff < start) {
5251 mod = start - prev.br_startoff;
5252 prev.br_blockcount -= mod;
5253 prev.br_startblock += mod;
5254 prev.br_startoff = start;
5255 }
5256 prev.br_state = XFS_EXT_UNWRITTEN;
5257 lastx--;
5258 error = xfs_bmap_add_extent_unwritten_real(tp,
5259 ip, &lastx, &cur, &prev,
5260 firstblock, flist, &logflags);
5261 if (error)
5262 goto error0;
5263 goto nodelete;
5264 } else {
5265 ASSERT(del.br_state == XFS_EXT_NORM);
5266 del.br_state = XFS_EXT_UNWRITTEN;
5267 error = xfs_bmap_add_extent_unwritten_real(tp,
5268 ip, &lastx, &cur, &del,
5269 firstblock, flist, &logflags);
5270 if (error)
5271 goto error0;
5272 goto nodelete;
9af25465 5273 }
1da177e4 5274 }
9e5987a7
DC		 5275 if (wasdel) {
5276 ASSERT(startblockval(del.br_startblock) > 0);
5277 /* Update realtime/data freespace, unreserve quota */
5278 if (isrt) {
5279 xfs_filblks_t rtexts;
1da177e4 5280
9e5987a7
DC		 5281 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5282 do_div(rtexts, mp->m_sb.sb_rextsize);
5283 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5284 (int64_t)rtexts, 0);
5285 (void)xfs_trans_reserve_quota_nblks(NULL,
5286 ip, -((long)del.br_blockcount), 0,
5287 XFS_QMOPT_RES_RTBLKS);
5288 } else {
5289 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5290 (int64_t)del.br_blockcount, 0);
5291 (void)xfs_trans_reserve_quota_nblks(NULL,
5292 ip, -((long)del.br_blockcount), 0,
5293 XFS_QMOPT_RES_REGBLKS);
5294 }
5295 ip->i_delayed_blks -= del.br_blockcount;
5296 if (cur)
5297 cur->bc_private.b.flags |=
5298 XFS_BTCUR_BPRV_WASDEL;
5299 } else if (cur)
5300 cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5301 /*
5302 * If it's the case where the directory code is running
5303 * with no block reservation, and the deleted block is in
5304 * the middle of its extent, and the resulting insert
5305 * of an extent would cause transformation to btree format,
5306 * then reject it. The calling code will then swap
5307 * blocks around instead.
5308 * We have to do this now, rather than waiting for the
5309 * conversion to btree format, since the transaction
5310 * will be dirty.
5311 */
5312 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5313 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5314 XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
5315 XFS_IFORK_MAXEXT(ip, whichfork) &&
5316 del.br_startoff > got.br_startoff &&
5317 del.br_startoff + del.br_blockcount <
5318 got.br_startoff + got.br_blockcount) {
2451337d 5319 error = -ENOSPC;
9e5987a7 5320 goto error0;
1da177e4 5321 }
9e5987a7
DC		 5322 error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5323 &tmp_logflags, whichfork);
5324 logflags |= tmp_logflags;
5325 if (error)
5326 goto error0;
5327 bno = del.br_startoff - 1;
5328nodelete:
1da177e4 5329 /*
9e5987a7 5330 * If not done go on to the next (previous) record.
1da177e4 5331 */
9e5987a7
DC		 5332 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5333 if (lastx >= 0) {
5334 ep = xfs_iext_get_ext(ifp, lastx);
5335 if (xfs_bmbt_get_startoff(ep) > bno) {
5336 if (--lastx >= 0)
5337 ep = xfs_iext_get_ext(ifp,
5338 lastx);
5339 }
5340 xfs_bmbt_get_all(ep, &got);
1da177e4 5341 }
9e5987a7 5342 extno++;
1da177e4
LT		 5343 }
5344 }
9e5987a7 5345 *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
576039cf 5346
1da177e4 5347 /*
9e5987a7 5348 * Convert to a btree if necessary.
1da177e4 5349 */
9e5987a7
DC		 5350 if (xfs_bmap_needs_btree(ip, whichfork)) {
5351 ASSERT(cur == NULL);
5352 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5353 &cur, 0, &tmp_logflags, whichfork);
5354 logflags |= tmp_logflags;
5355 if (error)
5356 goto error0;
1da177e4 5357 }
1da177e4 5358 /*
9e5987a7 5359 * transform from btree to extents, give it cur
1da177e4 5360 */
9e5987a7
DC		 5361 else if (xfs_bmap_wants_extents(ip, whichfork)) {
5362 ASSERT(cur != NULL);
5363 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5364 whichfork);
5365 logflags |= tmp_logflags;
5366 if (error)
5367 goto error0;
5368 }
1da177e4 5369 /*
9e5987a7 5370 * transform from extents to local?
1da177e4 5371 */
9e5987a7
DC		 5372 error = 0;
5373error0:
5374 /*
5375 * Log everything. Do this after conversion, there's no point in
5376 * logging the extent records if we've converted to btree format.
5377 */
5378 if ((logflags & xfs_ilog_fext(whichfork)) &&
5379 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5380 logflags &= ~xfs_ilog_fext(whichfork);
5381 else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5382 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5383 logflags &= ~xfs_ilog_fbroot(whichfork);
5384 /*
5385 * Log inode even in the error case, if the transaction
5386 * is dirty we'll need to shut down the filesystem.
5387 */
5388 if (logflags)
5389 xfs_trans_log_inode(tp, ip, logflags);
5390 if (cur) {
5391 if (!error) {
5392 *firstblock = cur->bc_private.b.firstblock;
5393 cur->bc_private.b.allocated = 0;
5394 }
5395 xfs_btree_del_cursor(cur,
5396 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5397 }
5398 return error;
5399}
e1d8fb88 5400
ddb19e31
BF		 5401/*
5402 * Determine whether an extent shift can be accomplished by a merge with the
5403 * extent that precedes the target hole of the shift.
5404 */
5405STATIC bool
5406xfs_bmse_can_merge(
5407 struct xfs_bmbt_irec *left, /* preceding extent */
5408 struct xfs_bmbt_irec *got, /* current extent to shift */
5409 xfs_fileoff_t shift) /* shift fsb */
5410{
5411 xfs_fileoff_t startoff;
5412
5413 startoff = got->br_startoff - shift;
5414
5415 /*
5416 * The extent, once shifted, must be adjacent in-file and on-disk with
5417 * the preceding extent.
5418 */
5419 if ((left->br_startoff + left->br_blockcount != startoff) ||
5420 (left->br_startblock + left->br_blockcount != got->br_startblock) ||
5421 (left->br_state != got->br_state) ||
5422 (left->br_blockcount + got->br_blockcount > MAXEXTLEN))
5423 return false;
5424
5425 return true;
5426}
5427
5428/*
5429 * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5430 * hole in the file. If an extent shift would result in the extent being fully
5431 * adjacent to the extent that currently precedes the hole, we can merge with
5432 * the preceding extent rather than do the shift.
5433 *
5434 * This function assumes the caller has verified a shift-by-merge is possible
5435 * with the provided extents via xfs_bmse_can_merge().
5436 */
5437STATIC int
5438xfs_bmse_merge(
5439 struct xfs_inode *ip,
5440 int whichfork,
5441 xfs_fileoff_t shift, /* shift fsb */
5442 int current_ext, /* idx of gotp */
5443 struct xfs_bmbt_rec_host *gotp, /* extent to shift */
5444 struct xfs_bmbt_rec_host *leftp, /* preceding extent */
5445 struct xfs_btree_cur *cur,
5446 int *logflags) /* output */
5447{
ddb19e31
BF		 5448 struct xfs_bmbt_irec got;
5449 struct xfs_bmbt_irec left;
5450 xfs_filblks_t blockcount;
5451 int error, i;
5452
ddb19e31
BF		 5453 xfs_bmbt_get_all(gotp, &got);
5454 xfs_bmbt_get_all(leftp, &left);
5455 blockcount = left.br_blockcount + got.br_blockcount;
5456
5457 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5458 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5459 ASSERT(xfs_bmse_can_merge(&left, &got, shift));
5460
5461 /*
5462 * Merge the in-core extents. Note that the host record pointers and
5463 * current_ext index are invalid once the extent has been removed via
5464 * xfs_iext_remove().
5465 */
5466 xfs_bmbt_set_blockcount(leftp, blockcount);
5467 xfs_iext_remove(ip, current_ext, 1, 0);
5468
5469 /*
5470 * Update the on-disk extent count, the btree if necessary and log the
5471 * inode.
5472 */
5473 XFS_IFORK_NEXT_SET(ip, whichfork,
5474 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5475 *logflags |= XFS_ILOG_CORE;
5476 if (!cur) {
5477 *logflags |= XFS_ILOG_DEXT;
5478 return 0;
5479 }
5480
5481 /* lookup and remove the extent to merge */
5482 error = xfs_bmbt_lookup_eq(cur, got.br_startoff, got.br_startblock,
5483 got.br_blockcount, &i);
5484 if (error)
4db431f5
DC		 5485 return error;
5486 XFS_WANT_CORRUPTED_RETURN(i == 1);
ddb19e31
BF		 5487
5488 error = xfs_btree_delete(cur, &i);
5489 if (error)
4db431f5
DC		 5490 return error;
5491 XFS_WANT_CORRUPTED_RETURN(i == 1);
ddb19e31
BF		 5492
5493 /* lookup and update size of the previous extent */
5494 error = xfs_bmbt_lookup_eq(cur, left.br_startoff, left.br_startblock,
5495 left.br_blockcount, &i);
5496 if (error)
4db431f5
DC		 5497 return error;
5498 XFS_WANT_CORRUPTED_RETURN(i == 1);
ddb19e31
BF		 5499
5500 left.br_blockcount = blockcount;
5501
4db431f5
DC		 5502 return xfs_bmbt_update(cur, left.br_startoff, left.br_startblock,
5503 left.br_blockcount, left.br_state);
ddb19e31
BF		 5504}
5505
a979bdfe
BF		 5506/*
5507 * Shift a single extent.
5508 */
5509STATIC int
5510xfs_bmse_shift_one(
5511 struct xfs_inode *ip,
5512 int whichfork,
5513 xfs_fileoff_t offset_shift_fsb,
5514 int *current_ext,
5515 struct xfs_bmbt_rec_host *gotp,
5516 struct xfs_btree_cur *cur,
5517 int *logflags)
5518{
5519 struct xfs_ifork *ifp;
5520 xfs_fileoff_t startoff;
5521 struct xfs_bmbt_rec_host *leftp;
5522 struct xfs_bmbt_irec got;
5523 struct xfs_bmbt_irec left;
5524 int error;
5525 int i;
5526
5527 ifp = XFS_IFORK_PTR(ip, whichfork);
5528
5529 xfs_bmbt_get_all(gotp, &got);
5530 startoff = got.br_startoff - offset_shift_fsb;
5531
f71721d0 5532 /* delalloc extents should be prevented by caller */
b11bd671 5533 XFS_WANT_CORRUPTED_RETURN(!isnullstartblock(got.br_startblock));
f71721d0 5534
a979bdfe 5535 /*
b11bd671
DC		 5536 * Check for merge if we've got an extent to the left, otherwise make
5537 * sure there's enough room at the start of the file for the shift.
a979bdfe 5538 */
b11bd671
DC		 5539 if (*current_ext) {
5540 /* grab the left extent and check for a large enough hole */
5541 leftp = xfs_iext_get_ext(ifp, *current_ext - 1);
5542 xfs_bmbt_get_all(leftp, &left);
a979bdfe 5543
b11bd671
DC		 5544 if (startoff < left.br_startoff + left.br_blockcount)
5545 return -EINVAL;
a979bdfe 5546
b11bd671
DC		 5547 /* check whether to merge the extent or shift it down */
5548 if (xfs_bmse_can_merge(&left, &got, offset_shift_fsb)) {
5549 return xfs_bmse_merge(ip, whichfork, offset_shift_fsb,
5550 *current_ext, gotp, leftp, cur,
5551 logflags);
5552 }
5553 } else if (got.br_startoff < offset_shift_fsb)
a979bdfe
BF		 5554 return -EINVAL;
5555
a979bdfe
BF		 5556 /*
5557 * Increment the extent index for the next iteration, update the start
5558 * offset of the in-core extent and update the btree if applicable.
5559 */
5560 (*current_ext)++;
5561 xfs_bmbt_set_startoff(gotp, startoff);
5562 *logflags |= XFS_ILOG_CORE;
5563 if (!cur) {
5564 *logflags |= XFS_ILOG_DEXT;
5565 return 0;
5566 }
5567
5568 error = xfs_bmbt_lookup_eq(cur, got.br_startoff, got.br_startblock,
5569 got.br_blockcount, &i);
5570 if (error)
5571 return error;
b11bd671 5572 XFS_WANT_CORRUPTED_RETURN(i == 1);
a979bdfe
BF		 5573
5574 got.br_startoff = startoff;
d254aaec 5575 return xfs_bmbt_update(cur, got.br_startoff, got.br_startblock,
a979bdfe 5576 got.br_blockcount, got.br_state);
a979bdfe
BF		 5577}
5578
e1d8fb88
NJ		 5579/*
5580 * Shift extent records to the left to cover a hole.
5581 *
2c845f5a
BF		 5582 * The maximum number of extents to be shifted in a single operation is
5583 * @num_exts. @start_fsb specifies the file offset to start the shift and the
5584 * file offset where we've left off is returned in @next_fsb. @offset_shift_fsb
5585 * is the length by which each extent is shifted. If there is no hole to shift
5586 * the extents into, this will be considered invalid operation and we abort
5587 * immediately.
e1d8fb88
NJ		 5588 */
5589int
5590xfs_bmap_shift_extents(
5591 struct xfs_trans *tp,
5592 struct xfs_inode *ip,
e1d8fb88
NJ		 5593 xfs_fileoff_t start_fsb,
5594 xfs_fileoff_t offset_shift_fsb,
2c845f5a
BF		 5595 int *done,
5596 xfs_fileoff_t *next_fsb,
e1d8fb88
NJ		 5597 xfs_fsblock_t *firstblock,
5598 struct xfs_bmap_free *flist,
5599 int num_exts)
5600{
ca446d88 5601 struct xfs_btree_cur *cur = NULL;
e1d8fb88
NJ		 5602 struct xfs_bmbt_rec_host *gotp;
5603 struct xfs_bmbt_irec got;
e1d8fb88
NJ		 5604 struct xfs_mount *mp = ip->i_mount;
5605 struct xfs_ifork *ifp;
5606 xfs_extnum_t nexts = 0;
2c845f5a 5607 xfs_extnum_t current_ext;
e1d8fb88 5608 int error = 0;
e1d8fb88 5609 int whichfork = XFS_DATA_FORK;
ca446d88 5610 int logflags = 0;
d39a2ced 5611 int total_extents;
e1d8fb88
NJ		 5612
5613 if (unlikely(XFS_TEST_ERROR(
5614 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5615 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
5616 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
5617 XFS_ERROR_REPORT("xfs_bmap_shift_extents",
5618 XFS_ERRLEVEL_LOW, mp);
2451337d 5619 return -EFSCORRUPTED;
e1d8fb88
NJ		 5620 }
5621
5622 if (XFS_FORCED_SHUTDOWN(mp))
2451337d 5623 return -EIO;
e1d8fb88 5624
2c845f5a
BF		 5625 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5626 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
e1d8fb88
NJ		 5627
5628 ifp = XFS_IFORK_PTR(ip, whichfork);
e1d8fb88
NJ		 5629 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5630 /* Read in all the extents */
5631 error = xfs_iread_extents(tp, ip, whichfork);
5632 if (error)
5633 return error;
5634 }
5635
ddb19e31
BF		 5636 if (ifp->if_flags & XFS_IFBROOT) {
5637 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5638 cur->bc_private.b.firstblock = *firstblock;
5639 cur->bc_private.b.flist = flist;
5640 cur->bc_private.b.flags = 0;
5641 }
5642
e1d8fb88 5643 /*
2c845f5a
BF		 5644 * Look up the extent index for the fsb where we start shifting. We can
5645 * henceforth iterate with current_ext as extent list changes are locked
5646 * out via ilock.
5647 *
5648 * gotp can be null in 2 cases: 1) if there are no extents or 2)
5649 * start_fsb lies in a hole beyond which there are no extents. Either
5650 * way, we are done.
e1d8fb88 5651 */
2c845f5a
BF		 5652 gotp = xfs_iext_bno_to_ext(ifp, start_fsb, &current_ext);
5653 if (!gotp) {
5654 *done = 1;
ddb19e31 5655 goto del_cursor;
e1d8fb88 5656 }
e1d8fb88 5657
d39a2ced
DC		 5658 /*
5659 * There may be delalloc extents in the data fork before the range we
2c845f5a
BF		 5660 * are collapsing out, so we cannot use the count of real extents here.
5661 * Instead we have to calculate it from the incore fork.
d39a2ced
DC		 5662 */
5663 total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
2c845f5a 5664 while (nexts++ < num_exts && current_ext < total_extents) {
a979bdfe
BF		 5665 error = xfs_bmse_shift_one(ip, whichfork, offset_shift_fsb,
5666 &current_ext, gotp, cur, &logflags);
5667 if (error)
5668 goto del_cursor;
e1d8fb88 5669
ddb19e31 5670 /* update total extent count and grab the next record */
d39a2ced 5671 total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
a979bdfe 5672 if (current_ext >= total_extents)
ddb19e31
BF		 5673 break;
5674 gotp = xfs_iext_get_ext(ifp, current_ext);
e1d8fb88
NJ		 5675 }
5676
5677 /* Check if we are done */
a979bdfe 5678 if (current_ext == total_extents) {
e1d8fb88 5679 *done = 1;
a979bdfe
BF		 5680 } else if (next_fsb) {
5681 xfs_bmbt_get_all(gotp, &got);
2c845f5a 5682 *next_fsb = got.br_startoff;
a979bdfe 5683 }
e1d8fb88
NJ		 5684
5685del_cursor:
5686 if (cur)
5687 xfs_btree_del_cursor(cur,
5688 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5689
ca446d88
BF		 5690 if (logflags)
5691 xfs_trans_log_inode(tp, ip, logflags);
2c845f5a 5692
e1d8fb88
NJ		 5693 return error;
5694}
Empty description
This page took 1.683844 seconds and 4 git commands to generate.