]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - fs/xfs/libxfs/xfs_bmap.c
projects / mirror_ubuntu-artful-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
xfs: remove the flags argument to xfs_trans_cancel
[mirror_ubuntu-artful-kernel.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#ifdef DEBUG
248STATIC struct xfs_buf *
249xfs_bmap_get_bp(
250 struct xfs_btree_cur *cur,
251 xfs_fsblock_t bno)
252{
253 struct xfs_log_item_desc *lidp;
254 int i;
7574aa92 255
9e5987a7
DC		 256 if (!cur)
257 return NULL;
258
259 for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
260 if (!cur->bc_bufs[i])
261 break;
262 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
263 return cur->bc_bufs[i];
1da177e4 264 }
7574aa92 265
9e5987a7
DC		 266 /* Chase down all the log items to see if the bp is there */
267 list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
268 struct xfs_buf_log_item *bip;
269 bip = (struct xfs_buf_log_item *)lidp->lid_item;
270 if (bip->bli_item.li_type == XFS_LI_BUF &&
271 XFS_BUF_ADDR(bip->bli_buf) == bno)
272 return bip->bli_buf;
273 }
7574aa92 274
9e5987a7
DC		 275 return NULL;
276}
0b1b213f 277
9e5987a7
DC		 278STATIC void
279xfs_check_block(
280 struct xfs_btree_block *block,
281 xfs_mount_t *mp,
282 int root,
283 short sz)
284{
285 int i, j, dmxr;
286 __be64 *pp, *thispa; /* pointer to block address */
287 xfs_bmbt_key_t *prevp, *keyp;
1da177e4 288
9e5987a7 289 ASSERT(be16_to_cpu(block->bb_level) > 0);
ec90c556 290
9e5987a7
DC		 291 prevp = NULL;
292 for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
293 dmxr = mp->m_bmap_dmxr[0];
294 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
0b1b213f 295
9e5987a7
DC		 296 if (prevp) {
297 ASSERT(be64_to_cpu(prevp->br_startoff) <
298 be64_to_cpu(keyp->br_startoff));
1da177e4 299 }
9e5987a7 300 prevp = keyp;
1da177e4 301
1da177e4 302 /*
9e5987a7 303 * Compare the block numbers to see if there are dups.
1da177e4 304 */
9e5987a7
DC		 305 if (root)
306 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
307 else
308 pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
0b1b213f 309
9e5987a7
DC		 310 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
311 if (root)
312 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
313 else
314 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
315 if (*thispa == *pp) {
316 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
317 __func__, j, i,
318 (unsigned long long)be64_to_cpu(*thispa));
319 panic("%s: ptrs are equal in node\n",
320 __func__);
321 }
1da177e4 322 }
9e5987a7
DC		 323 }
324}
1da177e4 325
9e5987a7
DC		 326/*
327 * Check that the extents for the inode ip are in the right order in all
328 * btree leaves.
329 */
0b1b213f 330
9e5987a7
DC		 331STATIC void
332xfs_bmap_check_leaf_extents(
333 xfs_btree_cur_t *cur, /* btree cursor or null */
334 xfs_inode_t *ip, /* incore inode pointer */
335 int whichfork) /* data or attr fork */
336{
337 struct xfs_btree_block *block; /* current btree block */
338 xfs_fsblock_t bno; /* block # of "block" */
339 xfs_buf_t *bp; /* buffer for "block" */
340 int error; /* error return value */
341 xfs_extnum_t i=0, j; /* index into the extents list */
342 xfs_ifork_t *ifp; /* fork structure */
343 int level; /* btree level, for checking */
344 xfs_mount_t *mp; /* file system mount structure */
345 __be64 *pp; /* pointer to block address */
346 xfs_bmbt_rec_t *ep; /* pointer to current extent */
347 xfs_bmbt_rec_t last = {0, 0}; /* last extent in prev block */
348 xfs_bmbt_rec_t *nextp; /* pointer to next extent */
349 int bp_release = 0;
350
351 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
352 return;
353 }
354
355 bno = NULLFSBLOCK;
356 mp = ip->i_mount;
357 ifp = XFS_IFORK_PTR(ip, whichfork);
358 block = ifp->if_broot;
359 /*
360 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
361 */
362 level = be16_to_cpu(block->bb_level);
363 ASSERT(level > 0);
364 xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
365 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
366 bno = be64_to_cpu(*pp);
367
d5cf09ba 368 ASSERT(bno != NULLFSBLOCK);
9e5987a7
DC		 369 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
370 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
371
372 /*
373 * Go down the tree until leaf level is reached, following the first
374 * pointer (leftmost) at each level.
375 */
376 while (level-- > 0) {
377 /* See if buf is in cur first */
378 bp_release = 0;
379 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
380 if (!bp) {
381 bp_release = 1;
382 error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
383 XFS_BMAP_BTREE_REF,
384 &xfs_bmbt_buf_ops);
572a4cf0 385 if (error)
9e5987a7 386 goto error_norelse;
1da177e4 387 }
9e5987a7 388 block = XFS_BUF_TO_BLOCK(bp);
9e5987a7
DC		 389 if (level == 0)
390 break;
1da177e4 391
1da177e4 392 /*
9e5987a7
DC		 393 * Check this block for basic sanity (increasing keys and
394 * no duplicate blocks).
1da177e4 395 */
0b1b213f 396
9e5987a7
DC		 397 xfs_check_block(block, mp, 0, 0);
398 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
399 bno = be64_to_cpu(*pp);
c29aad41
ES		 400 XFS_WANT_CORRUPTED_GOTO(mp,
401 XFS_FSB_SANITY_CHECK(mp, bno), error0);
9e5987a7
DC		 402 if (bp_release) {
403 bp_release = 0;
404 xfs_trans_brelse(NULL, bp);
1da177e4 405 }
9e5987a7 406 }
ec90c556 407
9e5987a7
DC		 408 /*
409 * Here with bp and block set to the leftmost leaf node in the tree.
410 */
411 i = 0;
412
413 /*
414 * Loop over all leaf nodes checking that all extents are in the right order.
415 */
416 for (;;) {
417 xfs_fsblock_t nextbno;
418 xfs_extnum_t num_recs;
419
420
421 num_recs = xfs_btree_get_numrecs(block);
1da177e4 422
1da177e4 423 /*
9e5987a7 424 * Read-ahead the next leaf block, if any.
1da177e4 425 */
8096b1eb 426
9e5987a7 427 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
1da177e4 428
1da177e4 429 /*
9e5987a7
DC		 430 * Check all the extents to make sure they are OK.
431 * If we had a previous block, the last entry should
432 * conform with the first entry in this one.
1da177e4 433 */
ec90c556 434
9e5987a7
DC		 435 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
436 if (i) {
437 ASSERT(xfs_bmbt_disk_get_startoff(&last) +
438 xfs_bmbt_disk_get_blockcount(&last) <=
439 xfs_bmbt_disk_get_startoff(ep));
440 }
441 for (j = 1; j < num_recs; j++) {
442 nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
443 ASSERT(xfs_bmbt_disk_get_startoff(ep) +
444 xfs_bmbt_disk_get_blockcount(ep) <=
445 xfs_bmbt_disk_get_startoff(nextp));
446 ep = nextp;
447 }
1da177e4 448
9e5987a7
DC		 449 last = *ep;
450 i += num_recs;
451 if (bp_release) {
452 bp_release = 0;
453 xfs_trans_brelse(NULL, bp);
454 }
455 bno = nextbno;
1da177e4 456 /*
9e5987a7 457 * If we've reached the end, stop.
1da177e4 458 */
9e5987a7
DC		 459 if (bno == NULLFSBLOCK)
460 break;
8096b1eb 461
9e5987a7
DC		 462 bp_release = 0;
463 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
464 if (!bp) {
465 bp_release = 1;
466 error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
467 XFS_BMAP_BTREE_REF,
468 &xfs_bmbt_buf_ops);
b9b984d7 469 if (error)
9e5987a7 470 goto error_norelse;
1da177e4 471 }
9e5987a7 472 block = XFS_BUF_TO_BLOCK(bp);
a5bd606b 473 }
9e5987a7
DC		 474 if (bp_release) {
475 bp_release = 0;
476 xfs_trans_brelse(NULL, bp);
a5bd606b 477 }
9e5987a7 478 return;
a5bd606b 479
9e5987a7
DC		 480error0:
481 xfs_warn(mp, "%s: at error0", __func__);
482 if (bp_release)
483 xfs_trans_brelse(NULL, bp);
484error_norelse:
485 xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
486 __func__, i);
487 panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
488 return;
1da177e4
LT		 489}
490
491/*
9e5987a7 492 * Add bmap trace insert entries for all the contents of the extent records.
1da177e4 493 */
9e5987a7
DC		 494void
495xfs_bmap_trace_exlist(
496 xfs_inode_t *ip, /* incore inode pointer */
497 xfs_extnum_t cnt, /* count of entries in the list */
498 int whichfork, /* data or attr fork */
499 unsigned long caller_ip)
1da177e4 500{
9e5987a7
DC		 501 xfs_extnum_t idx; /* extent record index */
502 xfs_ifork_t *ifp; /* inode fork pointer */
503 int state = 0;
a5bd606b 504
9e5987a7
DC		 505 if (whichfork == XFS_ATTR_FORK)
506 state |= BMAP_ATTRFORK;
a5bd606b 507
9e5987a7
DC		 508 ifp = XFS_IFORK_PTR(ip, whichfork);
509 ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
510 for (idx = 0; idx < cnt; idx++)
511 trace_xfs_extlist(ip, idx, whichfork, caller_ip);
512}
a5bd606b 513
9e5987a7
DC		 514/*
515 * Validate that the bmbt_irecs being returned from bmapi are valid
a97f4df7
ZYW		 516 * given the caller's original parameters. Specifically check the
517 * ranges of the returned irecs to ensure that they only extend beyond
9e5987a7
DC		 518 * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
519 */
520STATIC void
521xfs_bmap_validate_ret(
522 xfs_fileoff_t bno,
523 xfs_filblks_t len,
524 int flags,
525 xfs_bmbt_irec_t *mval,
526 int nmap,
527 int ret_nmap)
528{
529 int i; /* index to map values */
a5bd606b 530
9e5987a7 531 ASSERT(ret_nmap <= nmap);
a5bd606b 532
9e5987a7
DC		 533 for (i = 0; i < ret_nmap; i++) {
534 ASSERT(mval[i].br_blockcount > 0);
535 if (!(flags & XFS_BMAPI_ENTIRE)) {
536 ASSERT(mval[i].br_startoff >= bno);
537 ASSERT(mval[i].br_blockcount <= len);
538 ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
539 bno + len);
540 } else {
541 ASSERT(mval[i].br_startoff < bno + len);
542 ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
543 bno);
544 }
545 ASSERT(i == 0 ||
546 mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
547 mval[i].br_startoff);
548 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
549 mval[i].br_startblock != HOLESTARTBLOCK);
550 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
551 mval[i].br_state == XFS_EXT_UNWRITTEN);
552 }
553}
7574aa92 554
9e5987a7
DC		 555#else
556#define xfs_bmap_check_leaf_extents(cur, ip, whichfork) do { } while (0)
557#define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
558#endif /* DEBUG */
7574aa92 559
9e5987a7
DC		 560/*
561 * bmap free list manipulation functions
562 */
7574aa92 563
9e5987a7
DC		 564/*
565 * Add the extent to the list of extents to be free at transaction end.
566 * The list is maintained sorted (by block number).
567 */
568void
569xfs_bmap_add_free(
570 xfs_fsblock_t bno, /* fs block number of extent */
571 xfs_filblks_t len, /* length of extent */
572 xfs_bmap_free_t *flist, /* list of extents */
573 xfs_mount_t *mp) /* mount point structure */
574{
575 xfs_bmap_free_item_t *cur; /* current (next) element */
576 xfs_bmap_free_item_t *new; /* new element */
577 xfs_bmap_free_item_t *prev; /* previous element */
578#ifdef DEBUG
579 xfs_agnumber_t agno;
580 xfs_agblock_t agbno;
581
582 ASSERT(bno != NULLFSBLOCK);
583 ASSERT(len > 0);
584 ASSERT(len <= MAXEXTLEN);
585 ASSERT(!isnullstartblock(bno));
586 agno = XFS_FSB_TO_AGNO(mp, bno);
587 agbno = XFS_FSB_TO_AGBNO(mp, bno);
588 ASSERT(agno < mp->m_sb.sb_agcount);
589 ASSERT(agbno < mp->m_sb.sb_agblocks);
590 ASSERT(len < mp->m_sb.sb_agblocks);
591 ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
592#endif
593 ASSERT(xfs_bmap_free_item_zone != NULL);
594 new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
595 new->xbfi_startblock = bno;
596 new->xbfi_blockcount = (xfs_extlen_t)len;
597 for (prev = NULL, cur = flist->xbf_first;
598 cur != NULL;
599 prev = cur, cur = cur->xbfi_next) {
600 if (cur->xbfi_startblock >= bno)
601 break;
1da177e4 602 }
9e5987a7
DC		 603 if (prev)
604 prev->xbfi_next = new;
605 else
606 flist->xbf_first = new;
607 new->xbfi_next = cur;
608 flist->xbf_count++;
609}
7574aa92 610
9e5987a7
DC		 611/*
612 * Remove the entry "free" from the free item list. Prev points to the
613 * previous entry, unless "free" is the head of the list.
614 */
68988114 615void
9e5987a7
DC		 616xfs_bmap_del_free(
617 xfs_bmap_free_t *flist, /* free item list header */
618 xfs_bmap_free_item_t *prev, /* previous item on list, if any */
619 xfs_bmap_free_item_t *free) /* list item to be freed */
620{
621 if (prev)
622 prev->xbfi_next = free->xbfi_next;
623 else
624 flist->xbf_first = free->xbfi_next;
625 flist->xbf_count--;
626 kmem_zone_free(xfs_bmap_free_item_zone, free);
627}
628
9e5987a7
DC		 629/*
630 * Free up any items left in the list.
631 */
632void
633xfs_bmap_cancel(
634 xfs_bmap_free_t *flist) /* list of bmap_free_items */
635{
636 xfs_bmap_free_item_t *free; /* free list item */
637 xfs_bmap_free_item_t *next;
ec90c556 638
9e5987a7
DC		 639 if (flist->xbf_count == 0)
640 return;
641 ASSERT(flist->xbf_first != NULL);
642 for (free = flist->xbf_first; free; free = next) {
643 next = free->xbfi_next;
644 xfs_bmap_del_free(flist, NULL, free);
645 }
646 ASSERT(flist->xbf_count == 0);
647}
0b1b213f 648
9e5987a7
DC		 649/*
650 * Inode fork format manipulation functions
651 */
1da177e4 652
9e5987a7
DC		 653/*
654 * Transform a btree format file with only one leaf node, where the
655 * extents list will fit in the inode, into an extents format file.
656 * Since the file extents are already in-core, all we have to do is
657 * give up the space for the btree root and pitch the leaf block.
658 */
659STATIC int /* error */
660xfs_bmap_btree_to_extents(
661 xfs_trans_t *tp, /* transaction pointer */
662 xfs_inode_t *ip, /* incore inode pointer */
663 xfs_btree_cur_t *cur, /* btree cursor */
664 int *logflagsp, /* inode logging flags */
665 int whichfork) /* data or attr fork */
666{
667 /* REFERENCED */
668 struct xfs_btree_block *cblock;/* child btree block */
669 xfs_fsblock_t cbno; /* child block number */
670 xfs_buf_t *cbp; /* child block's buffer */
671 int error; /* error return value */
672 xfs_ifork_t *ifp; /* inode fork data */
673 xfs_mount_t *mp; /* mount point structure */
674 __be64 *pp; /* ptr to block address */
675 struct xfs_btree_block *rblock;/* root btree block */
1da177e4 676
9e5987a7
DC		 677 mp = ip->i_mount;
678 ifp = XFS_IFORK_PTR(ip, whichfork);
679 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
680 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
681 rblock = ifp->if_broot;
682 ASSERT(be16_to_cpu(rblock->bb_level) == 1);
683 ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
684 ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
685 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
686 cbno = be64_to_cpu(*pp);
687 *logflagsp = 0;
688#ifdef DEBUG
689 if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
690 return error;
691#endif
692 error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF,
693 &xfs_bmbt_buf_ops);
694 if (error)
695 return error;
696 cblock = XFS_BUF_TO_BLOCK(cbp);
697 if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
698 return error;
699 xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
700 ip->i_d.di_nblocks--;
701 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
702 xfs_trans_binval(tp, cbp);
703 if (cur->bc_bufs[0] == cbp)
704 cur->bc_bufs[0] = NULL;
705 xfs_iroot_realloc(ip, -1, whichfork);
706 ASSERT(ifp->if_broot == NULL);
707 ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
708 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
709 *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
710 return 0;
711}
0b1b213f 712
9e5987a7
DC		 713/*
714 * Convert an extents-format file into a btree-format file.
715 * The new file will have a root block (in the inode) and a single child block.
716 */
717STATIC int /* error */
718xfs_bmap_extents_to_btree(
719 xfs_trans_t *tp, /* transaction pointer */
720 xfs_inode_t *ip, /* incore inode pointer */
721 xfs_fsblock_t *firstblock, /* first-block-allocated */
722 xfs_bmap_free_t *flist, /* blocks freed in xaction */
723 xfs_btree_cur_t **curp, /* cursor returned to caller */
724 int wasdel, /* converting a delayed alloc */
725 int *logflagsp, /* inode logging flags */
726 int whichfork) /* data or attr fork */
727{
728 struct xfs_btree_block *ablock; /* allocated (child) bt block */
729 xfs_buf_t *abp; /* buffer for ablock */
730 xfs_alloc_arg_t args; /* allocation arguments */
731 xfs_bmbt_rec_t *arp; /* child record pointer */
732 struct xfs_btree_block *block; /* btree root block */
733 xfs_btree_cur_t *cur; /* bmap btree cursor */
734 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
735 int error; /* error return value */
736 xfs_extnum_t i, cnt; /* extent record index */
737 xfs_ifork_t *ifp; /* inode fork pointer */
738 xfs_bmbt_key_t *kp; /* root block key pointer */
739 xfs_mount_t *mp; /* mount structure */
740 xfs_extnum_t nextents; /* number of file extents */
741 xfs_bmbt_ptr_t *pp; /* root block address pointer */
1da177e4 742
ee1a47ab 743 mp = ip->i_mount;
9e5987a7
DC		 744 ifp = XFS_IFORK_PTR(ip, whichfork);
745 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
0b1b213f 746
9e5987a7
DC		 747 /*
748 * Make space in the inode incore.
749 */
750 xfs_iroot_realloc(ip, 1, whichfork);
751 ifp->if_flags |= XFS_IFBROOT;
ec90c556 752
9e5987a7
DC		 753 /*
754 * Fill in the root.
755 */
756 block = ifp->if_broot;
ee1a47ab
CH		 757 if (xfs_sb_version_hascrc(&mp->m_sb))
758 xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
759 XFS_BMAP_CRC_MAGIC, 1, 1, ip->i_ino,
760 XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
761 else
762 xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
763 XFS_BMAP_MAGIC, 1, 1, ip->i_ino,
764 XFS_BTREE_LONG_PTRS);
0b1b213f 765
9e5987a7
DC		 766 /*
767 * Need a cursor. Can't allocate until bb_level is filled in.
768 */
9e5987a7
DC		 769 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
770 cur->bc_private.b.firstblock = *firstblock;
771 cur->bc_private.b.flist = flist;
772 cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
773 /*
774 * Convert to a btree with two levels, one record in root.
775 */
776 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
777 memset(&args, 0, sizeof(args));
778 args.tp = tp;
779 args.mp = mp;
780 args.firstblock = *firstblock;
781 if (*firstblock == NULLFSBLOCK) {
782 args.type = XFS_ALLOCTYPE_START_BNO;
783 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
784 } else if (flist->xbf_low) {
785 args.type = XFS_ALLOCTYPE_START_BNO;
786 args.fsbno = *firstblock;
787 } else {
788 args.type = XFS_ALLOCTYPE_NEAR_BNO;
789 args.fsbno = *firstblock;
790 }
791 args.minlen = args.maxlen = args.prod = 1;
792 args.wasdel = wasdel;
793 *logflagsp = 0;
794 if ((error = xfs_alloc_vextent(&args))) {
795 xfs_iroot_realloc(ip, -1, whichfork);
796 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
797 return error;
798 }
799 /*
800 * Allocation can't fail, the space was reserved.
801 */
802 ASSERT(args.fsbno != NULLFSBLOCK);
803 ASSERT(*firstblock == NULLFSBLOCK ||
804 args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
805 (flist->xbf_low &&
806 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
807 *firstblock = cur->bc_private.b.firstblock = args.fsbno;
808 cur->bc_private.b.allocated++;
809 ip->i_d.di_nblocks++;
810 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
811 abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
812 /*
813 * Fill in the child block.
814 */
815 abp->b_ops = &xfs_bmbt_buf_ops;
816 ablock = XFS_BUF_TO_BLOCK(abp);
ee1a47ab
CH		 817 if (xfs_sb_version_hascrc(&mp->m_sb))
818 xfs_btree_init_block_int(mp, ablock, abp->b_bn,
819 XFS_BMAP_CRC_MAGIC, 0, 0, ip->i_ino,
820 XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
821 else
822 xfs_btree_init_block_int(mp, ablock, abp->b_bn,
823 XFS_BMAP_MAGIC, 0, 0, ip->i_ino,
824 XFS_BTREE_LONG_PTRS);
825
9e5987a7
DC		 826 arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
827 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
828 for (cnt = i = 0; i < nextents; i++) {
829 ep = xfs_iext_get_ext(ifp, i);
830 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
831 arp->l0 = cpu_to_be64(ep->l0);
832 arp->l1 = cpu_to_be64(ep->l1);
833 arp++; cnt++;
1da177e4 834 }
9e5987a7
DC		 835 }
836 ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
837 xfs_btree_set_numrecs(ablock, cnt);
1da177e4 838
9e5987a7
DC		 839 /*
840 * Fill in the root key and pointer.
841 */
842 kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
843 arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
844 kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
845 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
846 be16_to_cpu(block->bb_level)));
847 *pp = cpu_to_be64(args.fsbno);
ec90c556 848
9e5987a7
DC		 849 /*
850 * Do all this logging at the end so that
851 * the root is at the right level.
852 */
853 xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
854 xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
855 ASSERT(*curp == NULL);
856 *curp = cur;
857 *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
858 return 0;
859}
ec90c556 860
9e5987a7
DC		 861/*
862 * Convert a local file to an extents file.
863 * This code is out of bounds for data forks of regular files,
864 * since the file data needs to get logged so things will stay consistent.
865 * (The bmap-level manipulations are ok, though).
866 */
f3508bcd
DC		 867void
868xfs_bmap_local_to_extents_empty(
869 struct xfs_inode *ip,
870 int whichfork)
871{
872 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
873
874 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
875 ASSERT(ifp->if_bytes == 0);
876 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
877
6a9edd3d 878 xfs_bmap_forkoff_reset(ip, whichfork);
f3508bcd
DC		 879 ifp->if_flags &= ~XFS_IFINLINE;
880 ifp->if_flags |= XFS_IFEXTENTS;
881 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
882}
883
884
9e5987a7
DC		 885STATIC int /* error */
886xfs_bmap_local_to_extents(
887 xfs_trans_t *tp, /* transaction pointer */
888 xfs_inode_t *ip, /* incore inode pointer */
889 xfs_fsblock_t *firstblock, /* first block allocated in xaction */
890 xfs_extlen_t total, /* total blocks needed by transaction */
891 int *logflagsp, /* inode logging flags */
892 int whichfork,
ee1a47ab
CH		 893 void (*init_fn)(struct xfs_trans *tp,
894 struct xfs_buf *bp,
9e5987a7
DC		 895 struct xfs_inode *ip,
896 struct xfs_ifork *ifp))
897{
f3508bcd 898 int error = 0;
9e5987a7
DC		 899 int flags; /* logging flags returned */
900 xfs_ifork_t *ifp; /* inode fork pointer */
f3508bcd
DC		 901 xfs_alloc_arg_t args; /* allocation arguments */
902 xfs_buf_t *bp; /* buffer for extent block */
903 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
0b1b213f 904
9e5987a7
DC		 905 /*
906 * We don't want to deal with the case of keeping inode data inline yet.
907 * So sending the data fork of a regular inode is invalid.
908 */
909 ASSERT(!(S_ISREG(ip->i_d.di_mode) && whichfork == XFS_DATA_FORK));
910 ifp = XFS_IFORK_PTR(ip, whichfork);
911 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
f3508bcd
DC		 912
913 if (!ifp->if_bytes) {
914 xfs_bmap_local_to_extents_empty(ip, whichfork);
915 flags = XFS_ILOG_CORE;
916 goto done;
917 }
918
9e5987a7
DC		 919 flags = 0;
920 error = 0;
f3508bcd
DC		 921 ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) ==
922 XFS_IFINLINE);
923 memset(&args, 0, sizeof(args));
924 args.tp = tp;
925 args.mp = ip->i_mount;
926 args.firstblock = *firstblock;
927 /*
928 * Allocate a block. We know we need only one, since the
929 * file currently fits in an inode.
930 */
931 if (*firstblock == NULLFSBLOCK) {
932 args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
933 args.type = XFS_ALLOCTYPE_START_BNO;
9e5987a7 934 } else {
f3508bcd
DC		 935 args.fsbno = *firstblock;
936 args.type = XFS_ALLOCTYPE_NEAR_BNO;
9e5987a7 937 }
f3508bcd
DC		 938 args.total = total;
939 args.minlen = args.maxlen = args.prod = 1;
940 error = xfs_alloc_vextent(&args);
941 if (error)
942 goto done;
943
944 /* Can't fail, the space was reserved. */
945 ASSERT(args.fsbno != NULLFSBLOCK);
946 ASSERT(args.len == 1);
947 *firstblock = args.fsbno;
948 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
949
fe22d552
DC		 950 /*
951 * Initialise the block and copy the data
952 *
953 * Note: init_fn must set the buffer log item type correctly!
954 */
f3508bcd
DC		 955 init_fn(tp, bp, ip, ifp);
956
957 /* account for the change in fork size and log everything */
958 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
959 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
960 xfs_bmap_local_to_extents_empty(ip, whichfork);
9e5987a7 961 flags |= XFS_ILOG_CORE;
f3508bcd
DC		 962
963 xfs_iext_add(ifp, 0, 1);
964 ep = xfs_iext_get_ext(ifp, 0);
965 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
966 trace_xfs_bmap_post_update(ip, 0,
967 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
968 _THIS_IP_);
969 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
970 ip->i_d.di_nblocks = 1;
971 xfs_trans_mod_dquot_byino(tp, ip,
972 XFS_TRANS_DQ_BCOUNT, 1L);
973 flags |= xfs_ilog_fext(whichfork);
974
9e5987a7
DC		 975done:
976 *logflagsp = flags;
977 return error;
978}
ec90c556 979
9e5987a7
DC		 980/*
981 * Called from xfs_bmap_add_attrfork to handle btree format files.
982 */
983STATIC int /* error */
984xfs_bmap_add_attrfork_btree(
985 xfs_trans_t *tp, /* transaction pointer */
986 xfs_inode_t *ip, /* incore inode pointer */
987 xfs_fsblock_t *firstblock, /* first block allocated */
988 xfs_bmap_free_t *flist, /* blocks to free at commit */
989 int *flags) /* inode logging flags */
990{
991 xfs_btree_cur_t *cur; /* btree cursor */
992 int error; /* error return value */
993 xfs_mount_t *mp; /* file system mount struct */
994 int stat; /* newroot status */
ec90c556 995
9e5987a7
DC		 996 mp = ip->i_mount;
997 if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
998 *flags |= XFS_ILOG_DBROOT;
999 else {
1000 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
1001 cur->bc_private.b.flist = flist;
1002 cur->bc_private.b.firstblock = *firstblock;
1003 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
1004 goto error0;
1005 /* must be at least one entry */
c29aad41 1006 XFS_WANT_CORRUPTED_GOTO(mp, stat == 1, error0);
9e5987a7
DC		 1007 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
1008 goto error0;
1009 if (stat == 0) {
1010 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
2451337d 1011 return -ENOSPC;
1da177e4 1012 }
9e5987a7
DC		 1013 *firstblock = cur->bc_private.b.firstblock;
1014 cur->bc_private.b.allocated = 0;
1015 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1da177e4 1016 }
9e5987a7
DC		 1017 return 0;
1018error0:
1019 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1020 return error;
1021}
a5bd606b 1022
9e5987a7
DC		 1023/*
1024 * Called from xfs_bmap_add_attrfork to handle extents format files.
1025 */
1026STATIC int /* error */
1027xfs_bmap_add_attrfork_extents(
1028 xfs_trans_t *tp, /* transaction pointer */
1029 xfs_inode_t *ip, /* incore inode pointer */
1030 xfs_fsblock_t *firstblock, /* first block allocated */
1031 xfs_bmap_free_t *flist, /* blocks to free at commit */
1032 int *flags) /* inode logging flags */
1033{
1034 xfs_btree_cur_t *cur; /* bmap btree cursor */
1035 int error; /* error return value */
a5bd606b 1036
9e5987a7
DC		 1037 if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
1038 return 0;
1039 cur = NULL;
1040 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
1041 flags, XFS_DATA_FORK);
a5bd606b
CH		 1042 if (cur) {
1043 cur->bc_private.b.allocated = 0;
9e5987a7
DC		 1044 xfs_btree_del_cursor(cur,
1045 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
a5bd606b 1046 }
1da177e4 1047 return error;
1da177e4
LT		 1048}
1049
9e5987a7
DC		 1050/*
1051 * Called from xfs_bmap_add_attrfork to handle local format files. Each
1052 * different data fork content type needs a different callout to do the
1053 * conversion. Some are basic and only require special block initialisation
1054 * callouts for the data formating, others (directories) are so specialised they
1055 * handle everything themselves.
1056 *
1057 * XXX (dgc): investigate whether directory conversion can use the generic
1058 * formatting callout. It should be possible - it's just a very complex
ee1a47ab 1059 * formatter.
9e5987a7
DC		 1060 */
1061STATIC int /* error */
1062xfs_bmap_add_attrfork_local(
1063 xfs_trans_t *tp, /* transaction pointer */
1064 xfs_inode_t *ip, /* incore inode pointer */
1065 xfs_fsblock_t *firstblock, /* first block allocated */
1066 xfs_bmap_free_t *flist, /* blocks to free at commit */
1067 int *flags) /* inode logging flags */
1068{
1069 xfs_da_args_t dargs; /* args for dir/attr code */
7574aa92 1070
9e5987a7
DC		 1071 if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
1072 return 0;
7574aa92 1073
9e5987a7
DC		 1074 if (S_ISDIR(ip->i_d.di_mode)) {
1075 memset(&dargs, 0, sizeof(dargs));
d6cf1305 1076 dargs.geo = ip->i_mount->m_dir_geo;
9e5987a7
DC		 1077 dargs.dp = ip;
1078 dargs.firstblock = firstblock;
1079 dargs.flist = flist;
d6cf1305 1080 dargs.total = dargs.geo->fsbcount;
9e5987a7
DC		 1081 dargs.whichfork = XFS_DATA_FORK;
1082 dargs.trans = tp;
1083 return xfs_dir2_sf_to_block(&dargs);
1da177e4 1084 }
7574aa92 1085
9e5987a7
DC		 1086 if (S_ISLNK(ip->i_d.di_mode))
1087 return xfs_bmap_local_to_extents(tp, ip, firstblock, 1,
1088 flags, XFS_DATA_FORK,
1089 xfs_symlink_local_to_remote);
7574aa92 1090
f3508bcd
DC		 1091 /* should only be called for types that support local format data */
1092 ASSERT(0);
2451337d 1093 return -EFSCORRUPTED;
9e5987a7 1094}
0b1b213f 1095
9e5987a7
DC		 1096/*
1097 * Convert inode from non-attributed to attributed.
1098 * Must not be in a transaction, ip must not be locked.
1099 */
1100int /* error code */
1101xfs_bmap_add_attrfork(
1102 xfs_inode_t *ip, /* incore inode pointer */
1103 int size, /* space new attribute needs */
1104 int rsvd) /* xact may use reserved blks */
1105{
1106 xfs_fsblock_t firstblock; /* 1st block/ag allocated */
1107 xfs_bmap_free_t flist; /* freed extent records */
1108 xfs_mount_t *mp; /* mount structure */
1109 xfs_trans_t *tp; /* transaction pointer */
1110 int blks; /* space reservation */
1111 int version = 1; /* superblock attr version */
1112 int committed; /* xaction was committed */
1113 int logflags; /* logging flags */
1114 int error; /* error return value */
0b1b213f 1115
9e5987a7 1116 ASSERT(XFS_IFORK_Q(ip) == 0);
1da177e4 1117
9e5987a7
DC		 1118 mp = ip->i_mount;
1119 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1120 tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
1121 blks = XFS_ADDAFORK_SPACE_RES(mp);
1122 if (rsvd)
1123 tp->t_flags |= XFS_TRANS_RESERVE;
3d3c8b52 1124 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_addafork, blks, 0);
9e3908e3 1125 if (error) {
4906e215 1126 xfs_trans_cancel(tp);
9e3908e3
MT		 1127 return error;
1128 }
9e5987a7
DC		 1129 xfs_ilock(ip, XFS_ILOCK_EXCL);
1130 error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
1131 XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
1132 XFS_QMOPT_RES_REGBLKS);
9e3908e3
MT		 1133 if (error)
1134 goto trans_cancel;
9e5987a7 1135 if (XFS_IFORK_Q(ip))
9e3908e3 1136 goto trans_cancel;
9e5987a7 1137 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
1da177e4 1138 /*
9e5987a7 1139 * For inodes coming from pre-6.2 filesystems.
1da177e4 1140 */
9e5987a7
DC		 1141 ASSERT(ip->i_d.di_aformat == 0);
1142 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1143 }
1144 ASSERT(ip->i_d.di_anextents == 0);
ec90c556 1145
9e3908e3 1146 xfs_trans_ijoin(tp, ip, 0);
9e5987a7 1147 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1da177e4 1148
9e5987a7
DC		 1149 switch (ip->i_d.di_format) {
1150 case XFS_DINODE_FMT_DEV:
1151 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1152 break;
1153 case XFS_DINODE_FMT_UUID:
1154 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
1155 break;
1156 case XFS_DINODE_FMT_LOCAL:
1157 case XFS_DINODE_FMT_EXTENTS:
1158 case XFS_DINODE_FMT_BTREE:
1159 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1160 if (!ip->i_d.di_forkoff)
1161 ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1162 else if (mp->m_flags & XFS_MOUNT_ATTR2)
1163 version = 2;
1da177e4 1164 break;
9e5987a7
DC		 1165 default:
1166 ASSERT(0);
2451337d 1167 error = -EINVAL;
9e3908e3 1168 goto trans_cancel;
9e5987a7 1169 }
1da177e4 1170
9e5987a7
DC		 1171 ASSERT(ip->i_afp == NULL);
1172 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
1173 ip->i_afp->if_flags = XFS_IFEXTENTS;
1174 logflags = 0;
1175 xfs_bmap_init(&flist, &firstblock);
1176 switch (ip->i_d.di_format) {
1177 case XFS_DINODE_FMT_LOCAL:
1178 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
1179 &logflags);
1180 break;
1181 case XFS_DINODE_FMT_EXTENTS:
1182 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
1183 &flist, &logflags);
1184 break;
1185 case XFS_DINODE_FMT_BTREE:
1186 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
1187 &logflags);
1188 break;
1189 default:
1190 error = 0;
1da177e4
LT		 1191 break;
1192 }
9e5987a7
DC		 1193 if (logflags)
1194 xfs_trans_log_inode(tp, ip, logflags);
1195 if (error)
9e3908e3 1196 goto bmap_cancel;
9e5987a7
DC		 1197 if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1198 (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
61e63ecb 1199 bool log_sb = false;
9e5987a7
DC		 1200
1201 spin_lock(&mp->m_sb_lock);
1202 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1203 xfs_sb_version_addattr(&mp->m_sb);
61e63ecb 1204 log_sb = true;
9e5987a7
DC		 1205 }
1206 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1207 xfs_sb_version_addattr2(&mp->m_sb);
61e63ecb 1208 log_sb = true;
9e5987a7 1209 }
4d11a402 1210 spin_unlock(&mp->m_sb_lock);
61e63ecb
DC		 1211 if (log_sb)
1212 xfs_log_sb(tp);
1da177e4 1213 }
9e5987a7
DC		 1214
1215 error = xfs_bmap_finish(&tp, &flist, &committed);
1216 if (error)
9e3908e3
MT		 1217 goto bmap_cancel;
1218 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1219 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1220 return error;
1221
1222bmap_cancel:
9e5987a7 1223 xfs_bmap_cancel(&flist);
9e3908e3 1224trans_cancel:
4906e215 1225 xfs_trans_cancel(tp);
9e5987a7 1226 xfs_iunlock(ip, XFS_ILOCK_EXCL);
9e5987a7 1227 return error;
1da177e4
LT		 1228}
1229
1230/*
9e5987a7 1231 * Internal and external extent tree search functions.
1da177e4 1232 */
a5bd606b 1233
9e5987a7
DC		 1234/*
1235 * Read in the extents to if_extents.
1236 * All inode fields are set up by caller, we just traverse the btree
1237 * and copy the records in. If the file system cannot contain unwritten
1238 * extents, the records are checked for no "state" flags.
1239 */
1240int /* error */
1241xfs_bmap_read_extents(
1242 xfs_trans_t *tp, /* transaction pointer */
1243 xfs_inode_t *ip, /* incore inode */
1244 int whichfork) /* data or attr fork */
1245{
1246 struct xfs_btree_block *block; /* current btree block */
1247 xfs_fsblock_t bno; /* block # of "block" */
1248 xfs_buf_t *bp; /* buffer for "block" */
1249 int error; /* error return value */
1250 xfs_exntfmt_t exntf; /* XFS_EXTFMT_NOSTATE, if checking */
1251 xfs_extnum_t i, j; /* index into the extents list */
1252 xfs_ifork_t *ifp; /* fork structure */
1253 int level; /* btree level, for checking */
1254 xfs_mount_t *mp; /* file system mount structure */
1255 __be64 *pp; /* pointer to block address */
1256 /* REFERENCED */
1257 xfs_extnum_t room; /* number of entries there's room for */
6ef35544 1258
9e5987a7
DC		 1259 bno = NULLFSBLOCK;
1260 mp = ip->i_mount;
1261 ifp = XFS_IFORK_PTR(ip, whichfork);
1262 exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
1263 XFS_EXTFMT_INODE(ip);
1264 block = ifp->if_broot;
1da177e4 1265 /*
9e5987a7 1266 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1da177e4 1267 */
9e5987a7
DC		 1268 level = be16_to_cpu(block->bb_level);
1269 ASSERT(level > 0);
1270 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
1271 bno = be64_to_cpu(*pp);
d5cf09ba 1272 ASSERT(bno != NULLFSBLOCK);
9e5987a7
DC		 1273 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
1274 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
1da177e4 1275 /*
9e5987a7
DC		 1276 * Go down the tree until leaf level is reached, following the first
1277 * pointer (leftmost) at each level.
1da177e4 1278 */
9e5987a7
DC		 1279 while (level-- > 0) {
1280 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1281 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1282 if (error)
1283 return error;
1284 block = XFS_BUF_TO_BLOCK(bp);
9e5987a7
DC		 1285 if (level == 0)
1286 break;
1287 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
1288 bno = be64_to_cpu(*pp);
c29aad41
ES		 1289 XFS_WANT_CORRUPTED_GOTO(mp,
1290 XFS_FSB_SANITY_CHECK(mp, bno), error0);
9e5987a7 1291 xfs_trans_brelse(tp, bp);
1da177e4
LT		 1292 }
1293 /*
9e5987a7 1294 * Here with bp and block set to the leftmost leaf node in the tree.
1da177e4 1295 */
9e5987a7
DC		 1296 room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1297 i = 0;
1da177e4 1298 /*
9e5987a7 1299 * Loop over all leaf nodes. Copy information to the extent records.
1da177e4 1300 */
9e5987a7
DC		 1301 for (;;) {
1302 xfs_bmbt_rec_t *frp;
1303 xfs_fsblock_t nextbno;
1304 xfs_extnum_t num_recs;
1305 xfs_extnum_t start;
0b1b213f 1306
9e5987a7
DC		 1307 num_recs = xfs_btree_get_numrecs(block);
1308 if (unlikely(i + num_recs > room)) {
1309 ASSERT(i + num_recs <= room);
1310 xfs_warn(ip->i_mount,
1311 "corrupt dinode %Lu, (btree extents).",
1312 (unsigned long long) ip->i_ino);
1313 XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
1314 XFS_ERRLEVEL_LOW, ip->i_mount, block);
1315 goto error0;
1da177e4 1316 }
1da177e4 1317 /*
9e5987a7 1318 * Read-ahead the next leaf block, if any.
1da177e4 1319 */
9e5987a7
DC		 1320 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
1321 if (nextbno != NULLFSBLOCK)
1322 xfs_btree_reada_bufl(mp, nextbno, 1,
1323 &xfs_bmbt_buf_ops);
1da177e4 1324 /*
9e5987a7 1325 * Copy records into the extent records.
1da177e4 1326 */
9e5987a7
DC		 1327 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1328 start = i;
1329 for (j = 0; j < num_recs; j++, i++, frp++) {
1330 xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
1331 trp->l0 = be64_to_cpu(frp->l0);
1332 trp->l1 = be64_to_cpu(frp->l1);
1da177e4 1333 }
9e5987a7
DC		 1334 if (exntf == XFS_EXTFMT_NOSTATE) {
1335 /*
1336 * Check all attribute bmap btree records and
1337 * any "older" data bmap btree records for a
1338 * set bit in the "extent flag" position.
1339 */
1340 if (unlikely(xfs_check_nostate_extents(ifp,
1341 start, num_recs))) {
1342 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
1343 XFS_ERRLEVEL_LOW,
1344 ip->i_mount);
1345 goto error0;
1346 }
1347 }
1348 xfs_trans_brelse(tp, bp);
1349 bno = nextbno;
1da177e4 1350 /*
9e5987a7 1351 * If we've reached the end, stop.
1da177e4 1352 */
9e5987a7
DC		 1353 if (bno == NULLFSBLOCK)
1354 break;
1355 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1356 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1357 if (error)
1358 return error;
1359 block = XFS_BUF_TO_BLOCK(bp);
1da177e4 1360 }
9e5987a7
DC		 1361 ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
1362 ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
1363 XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
1364 return 0;
1365error0:
1366 xfs_trans_brelse(tp, bp);
2451337d 1367 return -EFSCORRUPTED;
9e5987a7 1368}
a5bd606b 1369
a5bd606b 1370
9e5987a7
DC		 1371/*
1372 * Search the extent records for the entry containing block bno.
1373 * If bno lies in a hole, point to the next entry. If bno lies
1374 * past eof, *eofp will be set, and *prevp will contain the last
1375 * entry (null if none). Else, *lastxp will be set to the index
1376 * of the found entry; *gotp will contain the entry.
1377 */
1378STATIC xfs_bmbt_rec_host_t * /* pointer to found extent entry */
1379xfs_bmap_search_multi_extents(
1380 xfs_ifork_t *ifp, /* inode fork pointer */
1381 xfs_fileoff_t bno, /* block number searched for */
1382 int *eofp, /* out: end of file found */
1383 xfs_extnum_t *lastxp, /* out: last extent index */
1384 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
1385 xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
1386{
1387 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
1388 xfs_extnum_t lastx; /* last extent index */
a5bd606b 1389
9e5987a7
DC		 1390 /*
1391 * Initialize the extent entry structure to catch access to
1392 * uninitialized br_startblock field.
1393 */
1394 gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
1395 gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
1396 gotp->br_state = XFS_EXT_INVALID;
9e5987a7 1397 gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
9e5987a7 1398 prevp->br_startoff = NULLFILEOFF;
c6534249 1399
9e5987a7
DC		 1400 ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
1401 if (lastx > 0) {
1402 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
1403 }
1404 if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1405 xfs_bmbt_get_all(ep, gotp);
1406 *eofp = 0;
1407 } else {
1408 if (lastx > 0) {
1409 *gotp = *prevp;
1410 }
1411 *eofp = 1;
1412 ep = NULL;
1413 }
1414 *lastxp = lastx;
1415 return ep;
1da177e4
LT		 1416}
1417
dd9f438e 1418/*
9e5987a7
DC		 1419 * Search the extents list for the inode, for the extent containing bno.
1420 * If bno lies in a hole, point to the next entry. If bno lies past eof,
1421 * *eofp will be set, and *prevp will contain the last entry (null if none).
1422 * Else, *lastxp will be set to the index of the found
1423 * entry; *gotp will contain the entry.
dd9f438e 1424 */
9e5987a7
DC		 1425STATIC xfs_bmbt_rec_host_t * /* pointer to found extent entry */
1426xfs_bmap_search_extents(
1427 xfs_inode_t *ip, /* incore inode pointer */
1428 xfs_fileoff_t bno, /* block number searched for */
1429 int fork, /* data or attr fork */
1430 int *eofp, /* out: end of file found */
1431 xfs_extnum_t *lastxp, /* out: last extent index */
1432 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
1433 xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
dd9f438e 1434{
9e5987a7
DC		 1435 xfs_ifork_t *ifp; /* inode fork pointer */
1436 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
dd9f438e 1437
9e5987a7
DC		 1438 XFS_STATS_INC(xs_look_exlist);
1439 ifp = XFS_IFORK_PTR(ip, fork);
dd9f438e 1440
9e5987a7 1441 ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
dd9f438e 1442
9e5987a7
DC		 1443 if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
1444 !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
1445 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
1446 "Access to block zero in inode %llu "
1447 "start_block: %llx start_off: %llx "
08e96e1a 1448 "blkcnt: %llx extent-state: %x lastx: %x",
9e5987a7
DC		 1449 (unsigned long long)ip->i_ino,
1450 (unsigned long long)gotp->br_startblock,
1451 (unsigned long long)gotp->br_startoff,
1452 (unsigned long long)gotp->br_blockcount,
1453 gotp->br_state, *lastxp);
1454 *lastxp = NULLEXTNUM;
1455 *eofp = 1;
1456 return NULL;
dd9f438e 1457 }
9e5987a7
DC		 1458 return ep;
1459}
dd9f438e 1460
9e5987a7
DC		 1461/*
1462 * Returns the file-relative block number of the first unused block(s)
1463 * in the file with at least "len" logically contiguous blocks free.
1464 * This is the lowest-address hole if the file has holes, else the first block
1465 * past the end of file.
1466 * Return 0 if the file is currently local (in-inode).
1467 */
1468int /* error */
1469xfs_bmap_first_unused(
1470 xfs_trans_t *tp, /* transaction pointer */
1471 xfs_inode_t *ip, /* incore inode */
1472 xfs_extlen_t len, /* size of hole to find */
1473 xfs_fileoff_t *first_unused, /* unused block */
1474 int whichfork) /* data or attr fork */
1475{
1476 int error; /* error return value */
1477 int idx; /* extent record index */
1478 xfs_ifork_t *ifp; /* inode fork pointer */
1479 xfs_fileoff_t lastaddr; /* last block number seen */
1480 xfs_fileoff_t lowest; /* lowest useful block */
1481 xfs_fileoff_t max; /* starting useful block */
1482 xfs_fileoff_t off; /* offset for this block */
1483 xfs_extnum_t nextents; /* number of extent entries */
1484
1485 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
1486 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
1487 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1488 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1489 *first_unused = 0;
1490 return 0;
dd9f438e 1491 }
9e5987a7
DC		 1492 ifp = XFS_IFORK_PTR(ip, whichfork);
1493 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1494 (error = xfs_iread_extents(tp, ip, whichfork)))
1495 return error;
1496 lowest = *first_unused;
1497 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1498 for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
1499 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
1500 off = xfs_bmbt_get_startoff(ep);
1501 /*
1502 * See if the hole before this extent will work.
1503 */
1504 if (off >= lowest + len && off - max >= len) {
1505 *first_unused = max;
1506 return 0;
1507 }
1508 lastaddr = off + xfs_bmbt_get_blockcount(ep);
1509 max = XFS_FILEOFF_MAX(lastaddr, lowest);
dd9f438e 1510 }
9e5987a7
DC		 1511 *first_unused = max;
1512 return 0;
1513}
1514
1515/*
02bb4873 1516 * Returns the file-relative block number of the last block - 1 before
9e5987a7
DC		 1517 * last_block (input value) in the file.
1518 * This is not based on i_size, it is based on the extent records.
1519 * Returns 0 for local files, as they do not have extent records.
1520 */
1521int /* error */
1522xfs_bmap_last_before(
1523 xfs_trans_t *tp, /* transaction pointer */
1524 xfs_inode_t *ip, /* incore inode */
1525 xfs_fileoff_t *last_block, /* last block */
1526 int whichfork) /* data or attr fork */
1527{
1528 xfs_fileoff_t bno; /* input file offset */
1529 int eof; /* hit end of file */
1530 xfs_bmbt_rec_host_t *ep; /* pointer to last extent */
1531 int error; /* error return value */
1532 xfs_bmbt_irec_t got; /* current extent value */
1533 xfs_ifork_t *ifp; /* inode fork pointer */
1534 xfs_extnum_t lastx; /* last extent used */
1535 xfs_bmbt_irec_t prev; /* previous extent value */
1536
1537 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1538 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
1539 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
2451337d 1540 return -EIO;
9e5987a7
DC		 1541 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1542 *last_block = 0;
1543 return 0;
1544 }
1545 ifp = XFS_IFORK_PTR(ip, whichfork);
1546 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1547 (error = xfs_iread_extents(tp, ip, whichfork)))
1548 return error;
1549 bno = *last_block - 1;
1550 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
1551 &prev);
1552 if (eof || xfs_bmbt_get_startoff(ep) > bno) {
1553 if (prev.br_startoff == NULLFILEOFF)
1554 *last_block = 0;
dd9f438e 1555 else
9e5987a7 1556 *last_block = prev.br_startoff + prev.br_blockcount;
dd9f438e 1557 }
dd9f438e 1558 /*
9e5987a7 1559 * Otherwise *last_block is already the right answer.
dd9f438e 1560 */
9e5987a7
DC		 1561 return 0;
1562}
1563
68988114 1564int
9e5987a7
DC		 1565xfs_bmap_last_extent(
1566 struct xfs_trans *tp,
1567 struct xfs_inode *ip,
1568 int whichfork,
1569 struct xfs_bmbt_irec *rec,
1570 int *is_empty)
1571{
1572 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
1573 int error;
1574 int nextents;
1575
1576 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1577 error = xfs_iread_extents(tp, ip, whichfork);
1578 if (error)
1579 return error;
dd9f438e
NS		 1580 }
1581
9e5987a7
DC		 1582 nextents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
1583 if (nextents == 0) {
1584 *is_empty = 1;
1585 return 0;
1586 }
dd9f438e 1587
9e5987a7
DC		 1588 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
1589 *is_empty = 0;
dd9f438e
NS		 1590 return 0;
1591}
1592
9e5987a7
DC		 1593/*
1594 * Check the last inode extent to determine whether this allocation will result
1595 * in blocks being allocated at the end of the file. When we allocate new data
1596 * blocks at the end of the file which do not start at the previous data block,
1597 * we will try to align the new blocks at stripe unit boundaries.
1598 *
6e708bcf 1599 * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
9e5987a7
DC		 1600 * at, or past the EOF.
1601 */
1602STATIC int
1603xfs_bmap_isaeof(
1604 struct xfs_bmalloca *bma,
1605 int whichfork)
1da177e4 1606{
9e5987a7
DC		 1607 struct xfs_bmbt_irec rec;
1608 int is_empty;
1609 int error;
1da177e4 1610
9e5987a7
DC		 1611 bma->aeof = 0;
1612 error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1613 &is_empty);
6e708bcf 1614 if (error)
9e5987a7 1615 return error;
1da177e4 1616
6e708bcf
DC		 1617 if (is_empty) {
1618 bma->aeof = 1;
1619 return 0;
1620 }
1621
1da177e4 1622 /*
9e5987a7
DC		 1623 * Check if we are allocation or past the last extent, or at least into
1624 * the last delayed allocated extent.
1da177e4 1625 */
9e5987a7
DC		 1626 bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1627 (bma->offset >= rec.br_startoff &&
1628 isnullstartblock(rec.br_startblock));
1629 return 0;
1630}
1da177e4 1631
9e5987a7
DC		 1632/*
1633 * Returns the file-relative block number of the first block past eof in
1634 * the file. This is not based on i_size, it is based on the extent records.
1635 * Returns 0 for local files, as they do not have extent records.
1636 */
1637int
1638xfs_bmap_last_offset(
9e5987a7
DC		 1639 struct xfs_inode *ip,
1640 xfs_fileoff_t *last_block,
1641 int whichfork)
1642{
1643 struct xfs_bmbt_irec rec;
1644 int is_empty;
1645 int error;
04e99455 1646
9e5987a7 1647 *last_block = 0;
0892ccd6 1648
9e5987a7
DC		 1649 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
1650 return 0;
a365bdd5 1651
9e5987a7
DC		 1652 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1653 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
2451337d 1654 return -EIO;
a365bdd5 1655
9e5987a7
DC		 1656 error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1657 if (error || is_empty)
a365bdd5 1658 return error;
9e5987a7
DC		 1659
1660 *last_block = rec.br_startoff + rec.br_blockcount;
a365bdd5
NS		 1661 return 0;
1662}
1663
9e5987a7
DC		 1664/*
1665 * Returns whether the selected fork of the inode has exactly one
1666 * block or not. For the data fork we check this matches di_size,
1667 * implying the file's range is 0..bsize-1.
1668 */
1669int /* 1=>1 block, 0=>otherwise */
1670xfs_bmap_one_block(
1671 xfs_inode_t *ip, /* incore inode */
1672 int whichfork) /* data or attr fork */
c467c049 1673{
9e5987a7
DC		 1674 xfs_bmbt_rec_host_t *ep; /* ptr to fork's extent */
1675 xfs_ifork_t *ifp; /* inode fork pointer */
1676 int rval; /* return value */
1677 xfs_bmbt_irec_t s; /* internal version of extent */
c467c049 1678
9e5987a7
DC		 1679#ifndef DEBUG
1680 if (whichfork == XFS_DATA_FORK)
1681 return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
1682#endif /* !DEBUG */
1683 if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
1684 return 0;
1685 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1686 return 0;
1687 ifp = XFS_IFORK_PTR(ip, whichfork);
1688 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1689 ep = xfs_iext_get_ext(ifp, 0);
1690 xfs_bmbt_get_all(ep, &s);
1691 rval = s.br_startoff == 0 && s.br_blockcount == 1;
1692 if (rval && whichfork == XFS_DATA_FORK)
1693 ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
1694 return rval;
1695}
c467c049 1696
9e5987a7
DC		 1697/*
1698 * Extent tree manipulation functions used during allocation.
1699 */
c467c049 1700
9e5987a7
DC		 1701/*
1702 * Convert a delayed allocation to a real allocation.
1703 */
1704STATIC int /* error */
1705xfs_bmap_add_extent_delay_real(
1706 struct xfs_bmalloca *bma)
1707{
1708 struct xfs_bmbt_irec *new = &bma->got;
1709 int diff; /* temp value */
1710 xfs_bmbt_rec_host_t *ep; /* extent entry for idx */
1711 int error; /* error return value */
1712 int i; /* temp state */
1713 xfs_ifork_t *ifp; /* inode fork pointer */
1714 xfs_fileoff_t new_endoff; /* end offset of new entry */
1715 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
1716 /* left is 0, right is 1, prev is 2 */
1717 int rval=0; /* return value (logging flags) */
1718 int state = 0;/* state bits, accessed thru macros */
1719 xfs_filblks_t da_new; /* new count del alloc blocks used */
1720 xfs_filblks_t da_old; /* old count del alloc blocks used */
1721 xfs_filblks_t temp=0; /* value for da_new calculations */
1722 xfs_filblks_t temp2=0;/* value for da_new calculations */
1723 int tmp_rval; /* partial logging flags */
c29aad41 1724 struct xfs_mount *mp;
c467c049 1725
c29aad41 1726 mp = bma->tp ? bma->tp->t_mountp : NULL;
9e5987a7 1727 ifp = XFS_IFORK_PTR(bma->ip, XFS_DATA_FORK);
c467c049 1728
9e5987a7
DC		 1729 ASSERT(bma->idx >= 0);
1730 ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
1731 ASSERT(!isnullstartblock(new->br_startblock));
1732 ASSERT(!bma->cur ||
1733 (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
c467c049 1734
9e5987a7 1735 XFS_STATS_INC(xs_add_exlist);
c467c049 1736
9e5987a7
DC		 1737#define LEFT r[0]
1738#define RIGHT r[1]
1739#define PREV r[2]
c467c049
CH		 1740
1741 /*
9e5987a7 1742 * Set up a bunch of variables to make the tests simpler.
c467c049 1743 */
9e5987a7
DC		 1744 ep = xfs_iext_get_ext(ifp, bma->idx);
1745 xfs_bmbt_get_all(ep, &PREV);
1746 new_endoff = new->br_startoff + new->br_blockcount;
1747 ASSERT(PREV.br_startoff <= new->br_startoff);
1748 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1749
1750 da_old = startblockval(PREV.br_startblock);
1751 da_new = 0;
1752
c467c049 1753 /*
9e5987a7
DC		 1754 * Set flags determining what part of the previous delayed allocation
1755 * extent is being replaced by a real allocation.
c467c049 1756 */
9e5987a7
DC		 1757 if (PREV.br_startoff == new->br_startoff)
1758 state |= BMAP_LEFT_FILLING;
1759 if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1760 state |= BMAP_RIGHT_FILLING;
c467c049
CH		 1761
1762 /*
9e5987a7
DC		 1763 * Check and set flags if this segment has a left neighbor.
1764 * Don't set contiguous if the combined extent would be too large.
c467c049 1765 */
9e5987a7
DC		 1766 if (bma->idx > 0) {
1767 state |= BMAP_LEFT_VALID;
1768 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
a99ebf43 1769
9e5987a7
DC		 1770 if (isnullstartblock(LEFT.br_startblock))
1771 state |= BMAP_LEFT_DELAY;
a365bdd5 1772 }
a365bdd5 1773
9e5987a7
DC		 1774 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1775 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1776 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1777 LEFT.br_state == new->br_state &&
1778 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1779 state |= BMAP_LEFT_CONTIG;
a365bdd5 1780
1da177e4 1781 /*
9e5987a7
DC		 1782 * Check and set flags if this segment has a right neighbor.
1783 * Don't set contiguous if the combined extent would be too large.
1784 * Also check for all-three-contiguous being too large.
1da177e4 1785 */
9e5987a7
DC		 1786 if (bma->idx < bma->ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1787 state |= BMAP_RIGHT_VALID;
1788 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
14b064ce 1789
9e5987a7
DC		 1790 if (isnullstartblock(RIGHT.br_startblock))
1791 state |= BMAP_RIGHT_DELAY;
1da177e4 1792 }
9e5987a7
DC		 1793
1794 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1795 new_endoff == RIGHT.br_startoff &&
1796 new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1797 new->br_state == RIGHT.br_state &&
1798 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1799 ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1800 BMAP_RIGHT_FILLING)) !=
1801 (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1802 BMAP_RIGHT_FILLING) ||
1803 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1804 <= MAXEXTLEN))
1805 state |= BMAP_RIGHT_CONTIG;
1806
1807 error = 0;
1da177e4 1808 /*
9e5987a7 1809 * Switch out based on the FILLING and CONTIG state bits.
1da177e4 1810 */
9e5987a7
DC		 1811 switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1812 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1813 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1814 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1da177e4 1815 /*
9e5987a7
DC		 1816 * Filling in all of a previously delayed allocation extent.
1817 * The left and right neighbors are both contiguous with new.
1da177e4 1818 */
9e5987a7
DC		 1819 bma->idx--;
1820 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1821 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1822 LEFT.br_blockcount + PREV.br_blockcount +
1823 RIGHT.br_blockcount);
1824 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1825
1826 xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
1827 bma->ip->i_d.di_nextents--;
1828 if (bma->cur == NULL)
1829 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1830 else {
1831 rval = XFS_ILOG_CORE;
1832 error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1833 RIGHT.br_startblock,
1834 RIGHT.br_blockcount, &i);
1835 if (error)
1836 goto done;
c29aad41 1837 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 1838 error = xfs_btree_delete(bma->cur, &i);
1839 if (error)
1840 goto done;
c29aad41 1841 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 1842 error = xfs_btree_decrement(bma->cur, 0, &i);
1843 if (error)
1844 goto done;
c29aad41 1845 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 1846 error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1847 LEFT.br_startblock,
1848 LEFT.br_blockcount +
1849 PREV.br_blockcount +
1850 RIGHT.br_blockcount, LEFT.br_state);
1851 if (error)
1852 goto done;
1853 }
1854 break;
1855
1856 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
a365bdd5 1857 /*
9e5987a7
DC		 1858 * Filling in all of a previously delayed allocation extent.
1859 * The left neighbor is contiguous, the right is not.
a365bdd5 1860 */
9e5987a7
DC		 1861 bma->idx--;
1862
1863 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1864 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1865 LEFT.br_blockcount + PREV.br_blockcount);
1866 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1867
1868 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1869 if (bma->cur == NULL)
1870 rval = XFS_ILOG_DEXT;
1871 else {
1872 rval = 0;
1873 error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1874 LEFT.br_startblock, LEFT.br_blockcount,
1875 &i);
1876 if (error)
1877 goto done;
c29aad41 1878 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 1879 error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1880 LEFT.br_startblock,
1881 LEFT.br_blockcount +
1882 PREV.br_blockcount, LEFT.br_state);
1883 if (error)
1884 goto done;
1885 }
1886 break;
1887
1888 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
0937e0fd 1889 /*
9e5987a7
DC		 1890 * Filling in all of a previously delayed allocation extent.
1891 * The right neighbor is contiguous, the left is not.
0937e0fd 1892 */
9e5987a7
DC		 1893 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1894 xfs_bmbt_set_startblock(ep, new->br_startblock);
1895 xfs_bmbt_set_blockcount(ep,
1896 PREV.br_blockcount + RIGHT.br_blockcount);
1897 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
0937e0fd 1898
9e5987a7
DC		 1899 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1900 if (bma->cur == NULL)
1901 rval = XFS_ILOG_DEXT;
1902 else {
1903 rval = 0;
1904 error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1905 RIGHT.br_startblock,
1906 RIGHT.br_blockcount, &i);
1907 if (error)
1908 goto done;
c29aad41 1909 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 1910 error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
1911 new->br_startblock,
1912 PREV.br_blockcount +
1913 RIGHT.br_blockcount, PREV.br_state);
1914 if (error)
1915 goto done;
1916 }
1917 break;
1918
1919 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
a365bdd5 1920 /*
9e5987a7
DC		 1921 * Filling in all of a previously delayed allocation extent.
1922 * Neither the left nor right neighbors are contiguous with
1923 * the new one.
a365bdd5 1924 */
9e5987a7
DC		 1925 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1926 xfs_bmbt_set_startblock(ep, new->br_startblock);
1927 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
a365bdd5 1928
9e5987a7
DC		 1929 bma->ip->i_d.di_nextents++;
1930 if (bma->cur == NULL)
1931 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1932 else {
1933 rval = XFS_ILOG_CORE;
1934 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1935 new->br_startblock, new->br_blockcount,
1936 &i);
1937 if (error)
1938 goto done;
c29aad41 1939 XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
9e5987a7
DC		 1940 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1941 error = xfs_btree_insert(bma->cur, &i);
1942 if (error)
1943 goto done;
c29aad41 1944 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 1945 }
1946 break;
1da177e4 1947
9e5987a7 1948 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1da177e4 1949 /*
9e5987a7
DC		 1950 * Filling in the first part of a previous delayed allocation.
1951 * The left neighbor is contiguous.
1da177e4 1952 */
9e5987a7
DC		 1953 trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1954 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
1955 LEFT.br_blockcount + new->br_blockcount);
1956 xfs_bmbt_set_startoff(ep,
1957 PREV.br_startoff + new->br_blockcount);
1958 trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1da177e4 1959
9e5987a7
DC		 1960 temp = PREV.br_blockcount - new->br_blockcount;
1961 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1962 xfs_bmbt_set_blockcount(ep, temp);
1963 if (bma->cur == NULL)
1964 rval = XFS_ILOG_DEXT;
1965 else {
1966 rval = 0;
1967 error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1968 LEFT.br_startblock, LEFT.br_blockcount,
1969 &i);
1970 if (error)
1971 goto done;
c29aad41 1972 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 1973 error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1974 LEFT.br_startblock,
1975 LEFT.br_blockcount +
1976 new->br_blockcount,
1977 LEFT.br_state);
f3ca8738 1978 if (error)
1da177e4 1979 goto done;
1da177e4 1980 }
9e5987a7
DC		 1981 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1982 startblockval(PREV.br_startblock));
1983 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
1984 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1985
1986 bma->idx--;
1987 break;
1988
1989 case BMAP_LEFT_FILLING:
1da177e4 1990 /*
9e5987a7
DC		 1991 * Filling in the first part of a previous delayed allocation.
1992 * The left neighbor is not contiguous.
1da177e4 1993 */
9e5987a7
DC		 1994 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1995 xfs_bmbt_set_startoff(ep, new_endoff);
1996 temp = PREV.br_blockcount - new->br_blockcount;
1997 xfs_bmbt_set_blockcount(ep, temp);
1998 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
1999 bma->ip->i_d.di_nextents++;
2000 if (bma->cur == NULL)
2001 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1da177e4 2002 else {
9e5987a7
DC		 2003 rval = XFS_ILOG_CORE;
2004 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2005 new->br_startblock, new->br_blockcount,
2006 &i);
2007 if (error)
2008 goto done;
c29aad41 2009 XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
9e5987a7
DC		 2010 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2011 error = xfs_btree_insert(bma->cur, &i);
2012 if (error)
1da177e4 2013 goto done;
c29aad41 2014 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1da177e4 2015 }
233eebb9 2016
9e5987a7
DC		 2017 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2018 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2019 bma->firstblock, bma->flist,
2020 &bma->cur, 1, &tmp_rval, XFS_DATA_FORK);
2021 rval |= tmp_rval;
2022 if (error)
2023 goto done;
1da177e4 2024 }
9e5987a7
DC		 2025 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2026 startblockval(PREV.br_startblock) -
2027 (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2028 ep = xfs_iext_get_ext(ifp, bma->idx + 1);
2029 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2030 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
1da177e4
LT		 2031 break;
2032
9e5987a7 2033 case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1da177e4 2034 /*
9e5987a7
DC		 2035 * Filling in the last part of a previous delayed allocation.
2036 * The right neighbor is contiguous with the new allocation.
1da177e4 2037 */
9e5987a7
DC		 2038 temp = PREV.br_blockcount - new->br_blockcount;
2039 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
1da177e4 2040 xfs_bmbt_set_blockcount(ep, temp);
9e5987a7
DC		 2041 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
2042 new->br_startoff, new->br_startblock,
2043 new->br_blockcount + RIGHT.br_blockcount,
2044 RIGHT.br_state);
2045 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2046 if (bma->cur == NULL)
2047 rval = XFS_ILOG_DEXT;
2048 else {
2049 rval = 0;
2050 error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
2051 RIGHT.br_startblock,
2052 RIGHT.br_blockcount, &i);
2053 if (error)
2054 goto done;
c29aad41 2055 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 2056 error = xfs_bmbt_update(bma->cur, new->br_startoff,
2057 new->br_startblock,
2058 new->br_blockcount +
2059 RIGHT.br_blockcount,
2060 RIGHT.br_state);
2061 if (error)
2062 goto done;
1da177e4 2063 }
9e5987a7
DC		 2064
2065 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2066 startblockval(PREV.br_startblock));
2067 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2068 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2069 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2070
2071 bma->idx++;
1da177e4
LT		 2072 break;
2073
9e5987a7 2074 case BMAP_RIGHT_FILLING:
1da177e4 2075 /*
9e5987a7
DC		 2076 * Filling in the last part of a previous delayed allocation.
2077 * The right neighbor is not contiguous.
1da177e4 2078 */
9e5987a7
DC		 2079 temp = PREV.br_blockcount - new->br_blockcount;
2080 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1da177e4 2081 xfs_bmbt_set_blockcount(ep, temp);
9e5987a7
DC		 2082 xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
2083 bma->ip->i_d.di_nextents++;
2084 if (bma->cur == NULL)
2085 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2086 else {
2087 rval = XFS_ILOG_CORE;
2088 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2089 new->br_startblock, new->br_blockcount,
2090 &i);
2091 if (error)
2092 goto done;
c29aad41 2093 XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
9e5987a7
DC		 2094 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2095 error = xfs_btree_insert(bma->cur, &i);
2096 if (error)
2097 goto done;
c29aad41 2098 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1da177e4 2099 }
9e5987a7
DC		 2100
2101 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2102 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2103 bma->firstblock, bma->flist, &bma->cur, 1,
2104 &tmp_rval, XFS_DATA_FORK);
2105 rval |= tmp_rval;
2106 if (error)
2107 goto done;
1da177e4 2108 }
9e5987a7
DC		 2109 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2110 startblockval(PREV.br_startblock) -
2111 (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2112 ep = xfs_iext_get_ext(ifp, bma->idx);
2113 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2114 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2115
2116 bma->idx++;
1da177e4
LT		 2117 break;
2118
2119 case 0:
2120 /*
9e5987a7
DC		 2121 * Filling in the middle part of a previous delayed allocation.
2122 * Contiguity is impossible here.
2123 * This case is avoided almost all the time.
2124 *
2125 * We start with a delayed allocation:
2126 *
2127 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
2128 * PREV @ idx
2129 *
2130 * and we are allocating:
2131 * +rrrrrrrrrrrrrrrrr+
2132 * new
2133 *
2134 * and we set it up for insertion as:
2135 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
2136 * new
2137 * PREV @ idx LEFT RIGHT
2138 * inserted at idx + 1
1da177e4 2139 */
9e5987a7
DC		 2140 temp = new->br_startoff - PREV.br_startoff;
2141 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
2142 trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
2143 xfs_bmbt_set_blockcount(ep, temp); /* truncate PREV */
2144 LEFT = *new;
2145 RIGHT.br_state = PREV.br_state;
2146 RIGHT.br_startblock = nullstartblock(
2147 (int)xfs_bmap_worst_indlen(bma->ip, temp2));
2148 RIGHT.br_startoff = new_endoff;
2149 RIGHT.br_blockcount = temp2;
2150 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
2151 xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
2152 bma->ip->i_d.di_nextents++;
2153 if (bma->cur == NULL)
2154 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2155 else {
2156 rval = XFS_ILOG_CORE;
2157 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2158 new->br_startblock, new->br_blockcount,
2159 &i);
2160 if (error)
2161 goto done;
c29aad41 2162 XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
9e5987a7
DC		 2163 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2164 error = xfs_btree_insert(bma->cur, &i);
2165 if (error)
2166 goto done;
c29aad41 2167 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1da177e4 2168 }
1da177e4 2169
9e5987a7
DC		 2170 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2171 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2172 bma->firstblock, bma->flist, &bma->cur,
2173 1, &tmp_rval, XFS_DATA_FORK);
2174 rval |= tmp_rval;
2175 if (error)
2176 goto done;
2177 }
2178 temp = xfs_bmap_worst_indlen(bma->ip, temp);
2179 temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
2180 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
2181 (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2182 if (diff > 0) {
0d485ada
DC		 2183 error = xfs_mod_fdblocks(bma->ip->i_mount,
2184 -((int64_t)diff), false);
9e5987a7
DC		 2185 ASSERT(!error);
2186 if (error)
2187 goto done;
2188 }
2189
2190 ep = xfs_iext_get_ext(ifp, bma->idx);
2191 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2192 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2193 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2194 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
2195 nullstartblock((int)temp2));
2196 trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2197
2198 bma->idx++;
2199 da_new = temp + temp2;
2200 break;
2201
2202 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2203 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2204 case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2205 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2206 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2207 case BMAP_LEFT_CONTIG:
2208 case BMAP_RIGHT_CONTIG:
2209 /*
2210 * These cases are all impossible.
2211 */
2212 ASSERT(0);
2213 }
2214
2215 /* convert to a btree if necessary */
2216 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2217 int tmp_logflags; /* partial log flag return val */
2218
2219 ASSERT(bma->cur == NULL);
2220 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2221 bma->firstblock, bma->flist, &bma->cur,
2222 da_old > 0, &tmp_logflags, XFS_DATA_FORK);
2223 bma->logflags |= tmp_logflags;
2224 if (error)
2225 goto done;
2226 }
2227
2228 /* adjust for changes in reserved delayed indirect blocks */
2229 if (da_old || da_new) {
2230 temp = da_new;
2231 if (bma->cur)
2232 temp += bma->cur->bc_private.b.allocated;
2233 ASSERT(temp <= da_old);
2234 if (temp < da_old)
0d485ada
DC		 2235 xfs_mod_fdblocks(bma->ip->i_mount,
2236 (int64_t)(da_old - temp), false);
96540c78 2237 }
9e5987a7
DC		 2238
2239 /* clear out the allocated field, done with it now in any case. */
2240 if (bma->cur)
2241 bma->cur->bc_private.b.allocated = 0;
2242
2243 xfs_bmap_check_leaf_extents(bma->cur, bma->ip, XFS_DATA_FORK);
1da177e4 2244done:
9e5987a7 2245 bma->logflags |= rval;
1da177e4 2246 return error;
9e5987a7
DC		 2247#undef LEFT
2248#undef RIGHT
2249#undef PREV
1da177e4
LT		 2250}
2251
2252/*
9e5987a7 2253 * Convert an unwritten allocation to a real allocation or vice versa.
1da177e4 2254 */
9e5987a7
DC		 2255STATIC int /* error */
2256xfs_bmap_add_extent_unwritten_real(
2257 struct xfs_trans *tp,
2258 xfs_inode_t *ip, /* incore inode pointer */
2259 xfs_extnum_t *idx, /* extent number to update/insert */
2260 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
2261 xfs_bmbt_irec_t *new, /* new data to add to file extents */
2262 xfs_fsblock_t *first, /* pointer to firstblock variable */
2263 xfs_bmap_free_t *flist, /* list of extents to be freed */
2264 int *logflagsp) /* inode logging flags */
1da177e4 2265{
9e5987a7
DC		 2266 xfs_btree_cur_t *cur; /* btree cursor */
2267 xfs_bmbt_rec_host_t *ep; /* extent entry for idx */
2268 int error; /* error return value */
2269 int i; /* temp state */
2270 xfs_ifork_t *ifp; /* inode fork pointer */
2271 xfs_fileoff_t new_endoff; /* end offset of new entry */
2272 xfs_exntst_t newext; /* new extent state */
2273 xfs_exntst_t oldext; /* old extent state */
2274 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
2275 /* left is 0, right is 1, prev is 2 */
2276 int rval=0; /* return value (logging flags) */
2277 int state = 0;/* state bits, accessed thru macros */
c29aad41 2278 struct xfs_mount *mp = tp->t_mountp;
1da177e4 2279
9e5987a7 2280 *logflagsp = 0;
1da177e4 2281
9e5987a7
DC		 2282 cur = *curp;
2283 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
8096b1eb 2284
9e5987a7
DC		 2285 ASSERT(*idx >= 0);
2286 ASSERT(*idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2287 ASSERT(!isnullstartblock(new->br_startblock));
2288
2289 XFS_STATS_INC(xs_add_exlist);
2290
2291#define LEFT r[0]
2292#define RIGHT r[1]
2293#define PREV r[2]
7cc95a82 2294
1da177e4 2295 /*
9e5987a7 2296 * Set up a bunch of variables to make the tests simpler.
1da177e4 2297 */
9e5987a7
DC		 2298 error = 0;
2299 ep = xfs_iext_get_ext(ifp, *idx);
2300 xfs_bmbt_get_all(ep, &PREV);
2301 newext = new->br_state;
2302 oldext = (newext == XFS_EXT_UNWRITTEN) ?
2303 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
2304 ASSERT(PREV.br_state == oldext);
2305 new_endoff = new->br_startoff + new->br_blockcount;
2306 ASSERT(PREV.br_startoff <= new->br_startoff);
2307 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
7cc95a82 2308
1da177e4 2309 /*
9e5987a7
DC		 2310 * Set flags determining what part of the previous oldext allocation
2311 * extent is being replaced by a newext allocation.
1da177e4 2312 */
9e5987a7
DC		 2313 if (PREV.br_startoff == new->br_startoff)
2314 state |= BMAP_LEFT_FILLING;
2315 if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2316 state |= BMAP_RIGHT_FILLING;
2317
1da177e4 2318 /*
9e5987a7
DC		 2319 * Check and set flags if this segment has a left neighbor.
2320 * Don't set contiguous if the combined extent would be too large.
1da177e4 2321 */
9e5987a7
DC		 2322 if (*idx > 0) {
2323 state |= BMAP_LEFT_VALID;
2324 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
2325
2326 if (isnullstartblock(LEFT.br_startblock))
2327 state |= BMAP_LEFT_DELAY;
1da177e4 2328 }
9e5987a7
DC		 2329
2330 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2331 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2332 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2333 LEFT.br_state == newext &&
2334 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2335 state |= BMAP_LEFT_CONTIG;
2336
1da177e4 2337 /*
9e5987a7
DC		 2338 * Check and set flags if this segment has a right neighbor.
2339 * Don't set contiguous if the combined extent would be too large.
2340 * Also check for all-three-contiguous being too large.
1da177e4 2341 */
9e5987a7
DC		 2342 if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
2343 state |= BMAP_RIGHT_VALID;
2344 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
2345 if (isnullstartblock(RIGHT.br_startblock))
2346 state |= BMAP_RIGHT_DELAY;
1da177e4 2347 }
7cc95a82 2348
9e5987a7
DC		 2349 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2350 new_endoff == RIGHT.br_startoff &&
2351 new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2352 newext == RIGHT.br_state &&
2353 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2354 ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2355 BMAP_RIGHT_FILLING)) !=
2356 (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2357 BMAP_RIGHT_FILLING) ||
2358 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2359 <= MAXEXTLEN))
2360 state |= BMAP_RIGHT_CONTIG;
136341b4 2361
1da177e4 2362 /*
9e5987a7 2363 * Switch out based on the FILLING and CONTIG state bits.
1da177e4 2364 */
9e5987a7
DC		 2365 switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2366 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2367 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2368 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2369 /*
2370 * Setting all of a previous oldext extent to newext.
2371 * The left and right neighbors are both contiguous with new.
2372 */
2373 --*idx;
1a5902c5 2374
9e5987a7
DC		 2375 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2376 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2377 LEFT.br_blockcount + PREV.br_blockcount +
2378 RIGHT.br_blockcount);
2379 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1a5902c5 2380
9e5987a7
DC		 2381 xfs_iext_remove(ip, *idx + 1, 2, state);
2382 ip->i_d.di_nextents -= 2;
2383 if (cur == NULL)
2384 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2385 else {
2386 rval = XFS_ILOG_CORE;
2387 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2388 RIGHT.br_startblock,
2389 RIGHT.br_blockcount, &i)))
2390 goto done;
c29aad41 2391 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 2392 if ((error = xfs_btree_delete(cur, &i)))
2393 goto done;
c29aad41 2394 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 2395 if ((error = xfs_btree_decrement(cur, 0, &i)))
2396 goto done;
c29aad41 2397 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 2398 if ((error = xfs_btree_delete(cur, &i)))
2399 goto done;
c29aad41 2400 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 2401 if ((error = xfs_btree_decrement(cur, 0, &i)))
2402 goto done;
c29aad41 2403 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 2404 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2405 LEFT.br_startblock,
2406 LEFT.br_blockcount + PREV.br_blockcount +
2407 RIGHT.br_blockcount, LEFT.br_state)))
2408 goto done;
2409 }
2410 break;
1a5902c5 2411
9e5987a7
DC		 2412 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2413 /*
2414 * Setting all of a previous oldext extent to newext.
2415 * The left neighbor is contiguous, the right is not.
2416 */
2417 --*idx;
d8cc890d 2418
9e5987a7
DC		 2419 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2420 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2421 LEFT.br_blockcount + PREV.br_blockcount);
2422 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1da177e4 2423
9e5987a7
DC		 2424 xfs_iext_remove(ip, *idx + 1, 1, state);
2425 ip->i_d.di_nextents--;
2426 if (cur == NULL)
2427 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2428 else {
2429 rval = XFS_ILOG_CORE;
2430 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2431 PREV.br_startblock, PREV.br_blockcount,
2432 &i)))
2433 goto done;
c29aad41 2434 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 2435 if ((error = xfs_btree_delete(cur, &i)))
2436 goto done;
c29aad41 2437 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 2438 if ((error = xfs_btree_decrement(cur, 0, &i)))
2439 goto done;
c29aad41 2440 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 2441 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2442 LEFT.br_startblock,
2443 LEFT.br_blockcount + PREV.br_blockcount,
2444 LEFT.br_state)))
2445 goto done;
2446 }
2447 break;
1da177e4 2448
9e5987a7 2449 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1da177e4 2450 /*
9e5987a7
DC		 2451 * Setting all of a previous oldext extent to newext.
2452 * The right neighbor is contiguous, the left is not.
1da177e4 2453 */
9e5987a7
DC		 2454 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2455 xfs_bmbt_set_blockcount(ep,
2456 PREV.br_blockcount + RIGHT.br_blockcount);
2457 xfs_bmbt_set_state(ep, newext);
2458 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2459 xfs_iext_remove(ip, *idx + 1, 1, state);
2460 ip->i_d.di_nextents--;
2461 if (cur == NULL)
2462 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2463 else {
2464 rval = XFS_ILOG_CORE;
2465 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2466 RIGHT.br_startblock,
2467 RIGHT.br_blockcount, &i)))
2468 goto done;
c29aad41 2469 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 2470 if ((error = xfs_btree_delete(cur, &i)))
2471 goto done;
c29aad41 2472 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 2473 if ((error = xfs_btree_decrement(cur, 0, &i)))
2474 goto done;
c29aad41 2475 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 2476 if ((error = xfs_bmbt_update(cur, new->br_startoff,
2477 new->br_startblock,
2478 new->br_blockcount + RIGHT.br_blockcount,
2479 newext)))
2480 goto done;
1da177e4 2481 }
9e5987a7 2482 break;
1e82379b 2483
9e5987a7
DC		 2484 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2485 /*
2486 * Setting all of a previous oldext extent to newext.
2487 * Neither the left nor right neighbors are contiguous with
2488 * the new one.
2489 */
2490 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2491 xfs_bmbt_set_state(ep, newext);
2492 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1e82379b 2493
9e5987a7
DC		 2494 if (cur == NULL)
2495 rval = XFS_ILOG_DEXT;
2496 else {
2497 rval = 0;
2498 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2499 new->br_startblock, new->br_blockcount,
2500 &i)))
2501 goto done;
c29aad41 2502 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 2503 if ((error = xfs_bmbt_update(cur, new->br_startoff,
2504 new->br_startblock, new->br_blockcount,
2505 newext)))
2506 goto done;
2507 }
2508 break;
1e82379b 2509
9e5987a7
DC		 2510 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2511 /*
2512 * Setting the first part of a previous oldext extent to newext.
2513 * The left neighbor is contiguous.
2514 */
2515 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
2516 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
2517 LEFT.br_blockcount + new->br_blockcount);
2518 xfs_bmbt_set_startoff(ep,
2519 PREV.br_startoff + new->br_blockcount);
2520 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
1da177e4 2521
9e5987a7
DC		 2522 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2523 xfs_bmbt_set_startblock(ep,
2524 new->br_startblock + new->br_blockcount);
2525 xfs_bmbt_set_blockcount(ep,
2526 PREV.br_blockcount - new->br_blockcount);
2527 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
0293ce3a 2528
9e5987a7 2529 --*idx;
8867bc9b 2530
9e5987a7
DC		 2531 if (cur == NULL)
2532 rval = XFS_ILOG_DEXT;
2533 else {
2534 rval = 0;
2535 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2536 PREV.br_startblock, PREV.br_blockcount,
2537 &i)))
2538 goto done;
c29aad41 2539 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 2540 if ((error = xfs_bmbt_update(cur,
2541 PREV.br_startoff + new->br_blockcount,
2542 PREV.br_startblock + new->br_blockcount,
2543 PREV.br_blockcount - new->br_blockcount,
2544 oldext)))
2545 goto done;
2546 if ((error = xfs_btree_decrement(cur, 0, &i)))
2547 goto done;
2548 error = xfs_bmbt_update(cur, LEFT.br_startoff,
2549 LEFT.br_startblock,
2550 LEFT.br_blockcount + new->br_blockcount,
2551 LEFT.br_state);
2552 if (error)
2553 goto done;
8867bc9b 2554 }
9e5987a7 2555 break;
0293ce3a 2556
9e5987a7
DC		 2557 case BMAP_LEFT_FILLING:
2558 /*
2559 * Setting the first part of a previous oldext extent to newext.
2560 * The left neighbor is not contiguous.
2561 */
2562 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2563 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
2564 xfs_bmbt_set_startoff(ep, new_endoff);
2565 xfs_bmbt_set_blockcount(ep,
2566 PREV.br_blockcount - new->br_blockcount);
2567 xfs_bmbt_set_startblock(ep,
2568 new->br_startblock + new->br_blockcount);
2569 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1da177e4 2570
9e5987a7
DC		 2571 xfs_iext_insert(ip, *idx, 1, new, state);
2572 ip->i_d.di_nextents++;
2573 if (cur == NULL)
2574 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2575 else {
2576 rval = XFS_ILOG_CORE;
2577 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2578 PREV.br_startblock, PREV.br_blockcount,
2579 &i)))
2580 goto done;
c29aad41 2581 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 2582 if ((error = xfs_bmbt_update(cur,
2583 PREV.br_startoff + new->br_blockcount,
2584 PREV.br_startblock + new->br_blockcount,
2585 PREV.br_blockcount - new->br_blockcount,
2586 oldext)))
2587 goto done;
2588 cur->bc_rec.b = *new;
2589 if ((error = xfs_btree_insert(cur, &i)))
2590 goto done;
c29aad41 2591 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 2592 }
2593 break;
1da177e4 2594
9e5987a7
DC		 2595 case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2596 /*
2597 * Setting the last part of a previous oldext extent to newext.
2598 * The right neighbor is contiguous with the new allocation.
2599 */
2600 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2601 xfs_bmbt_set_blockcount(ep,
2602 PREV.br_blockcount - new->br_blockcount);
2603 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
0293ce3a 2604
9e5987a7 2605 ++*idx;
1da177e4 2606
9e5987a7
DC		 2607 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2608 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2609 new->br_startoff, new->br_startblock,
2610 new->br_blockcount + RIGHT.br_blockcount, newext);
2611 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1da177e4 2612
9e5987a7
DC		 2613 if (cur == NULL)
2614 rval = XFS_ILOG_DEXT;
2615 else {
2616 rval = 0;
2617 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2618 PREV.br_startblock,
2619 PREV.br_blockcount, &i)))
2620 goto done;
c29aad41 2621 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 2622 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2623 PREV.br_startblock,
2624 PREV.br_blockcount - new->br_blockcount,
2625 oldext)))
2626 goto done;
2627 if ((error = xfs_btree_increment(cur, 0, &i)))
2628 goto done;
2629 if ((error = xfs_bmbt_update(cur, new->br_startoff,
2630 new->br_startblock,
2631 new->br_blockcount + RIGHT.br_blockcount,
2632 newext)))
2633 goto done;
2634 }
2635 break;
1da177e4 2636
9e5987a7
DC		 2637 case BMAP_RIGHT_FILLING:
2638 /*
2639 * Setting the last part of a previous oldext extent to newext.
2640 * The right neighbor is not contiguous.
2641 */
2642 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2643 xfs_bmbt_set_blockcount(ep,
2644 PREV.br_blockcount - new->br_blockcount);
2645 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1da177e4 2646
9e5987a7
DC		 2647 ++*idx;
2648 xfs_iext_insert(ip, *idx, 1, new, state);
d8cc890d 2649
9e5987a7
DC		 2650 ip->i_d.di_nextents++;
2651 if (cur == NULL)
2652 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2653 else {
2654 rval = XFS_ILOG_CORE;
2655 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2656 PREV.br_startblock, PREV.br_blockcount,
2657 &i)))
2658 goto done;
c29aad41 2659 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 2660 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2661 PREV.br_startblock,
2662 PREV.br_blockcount - new->br_blockcount,
2663 oldext)))
2664 goto done;
2665 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2666 new->br_startblock, new->br_blockcount,
2667 &i)))
2668 goto done;
c29aad41 2669 XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
9e5987a7
DC		 2670 cur->bc_rec.b.br_state = XFS_EXT_NORM;
2671 if ((error = xfs_btree_insert(cur, &i)))
2672 goto done;
c29aad41 2673 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 2674 }
2675 break;
2676
2677 case 0:
1da177e4 2678 /*
9e5987a7
DC		 2679 * Setting the middle part of a previous oldext extent to
2680 * newext. Contiguity is impossible here.
2681 * One extent becomes three extents.
1da177e4 2682 */
9e5987a7
DC		 2683 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2684 xfs_bmbt_set_blockcount(ep,
2685 new->br_startoff - PREV.br_startoff);
2686 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
898621d5 2687
9e5987a7
DC		 2688 r[0] = *new;
2689 r[1].br_startoff = new_endoff;
2690 r[1].br_blockcount =
2691 PREV.br_startoff + PREV.br_blockcount - new_endoff;
2692 r[1].br_startblock = new->br_startblock + new->br_blockcount;
2693 r[1].br_state = oldext;
898621d5 2694
9e5987a7
DC		 2695 ++*idx;
2696 xfs_iext_insert(ip, *idx, 2, &r[0], state);
2697
2698 ip->i_d.di_nextents += 2;
2699 if (cur == NULL)
2700 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2701 else {
2702 rval = XFS_ILOG_CORE;
2703 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2704 PREV.br_startblock, PREV.br_blockcount,
2705 &i)))
2706 goto done;
c29aad41 2707 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 2708 /* new right extent - oldext */
2709 if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
2710 r[1].br_startblock, r[1].br_blockcount,
2711 r[1].br_state)))
2712 goto done;
2713 /* new left extent - oldext */
2714 cur->bc_rec.b = PREV;
2715 cur->bc_rec.b.br_blockcount =
2716 new->br_startoff - PREV.br_startoff;
2717 if ((error = xfs_btree_insert(cur, &i)))
2718 goto done;
c29aad41 2719 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 2720 /*
2721 * Reset the cursor to the position of the new extent
2722 * we are about to insert as we can't trust it after
2723 * the previous insert.
2724 */
2725 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2726 new->br_startblock, new->br_blockcount,
2727 &i)))
2728 goto done;
c29aad41 2729 XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
9e5987a7
DC		 2730 /* new middle extent - newext */
2731 cur->bc_rec.b.br_state = new->br_state;
2732 if ((error = xfs_btree_insert(cur, &i)))
2733 goto done;
c29aad41 2734 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7 2735 }
1da177e4 2736 break;
9e5987a7
DC		 2737
2738 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2739 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2740 case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2741 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2742 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2743 case BMAP_LEFT_CONTIG:
2744 case BMAP_RIGHT_CONTIG:
2745 /*
2746 * These cases are all impossible.
2747 */
1da177e4 2748 ASSERT(0);
1da177e4 2749 }
8096b1eb 2750
9e5987a7
DC		 2751 /* convert to a btree if necessary */
2752 if (xfs_bmap_needs_btree(ip, XFS_DATA_FORK)) {
2753 int tmp_logflags; /* partial log flag return val */
2754
2755 ASSERT(cur == NULL);
2756 error = xfs_bmap_extents_to_btree(tp, ip, first, flist, &cur,
2757 0, &tmp_logflags, XFS_DATA_FORK);
2758 *logflagsp |= tmp_logflags;
2759 if (error)
2760 goto done;
1da177e4 2761 }
da087bad 2762
9e5987a7
DC		 2763 /* clear out the allocated field, done with it now in any case. */
2764 if (cur) {
2765 cur->bc_private.b.allocated = 0;
2766 *curp = cur;
1da177e4 2767 }
8096b1eb 2768
9e5987a7
DC		 2769 xfs_bmap_check_leaf_extents(*curp, ip, XFS_DATA_FORK);
2770done:
2771 *logflagsp |= rval;
1da177e4 2772 return error;
9e5987a7
DC		 2773#undef LEFT
2774#undef RIGHT
2775#undef PREV
1da177e4
LT		 2776}
2777
2778/*
9e5987a7 2779 * Convert a hole to a delayed allocation.
1da177e4 2780 */
9e5987a7
DC		 2781STATIC void
2782xfs_bmap_add_extent_hole_delay(
2783 xfs_inode_t *ip, /* incore inode pointer */
2784 xfs_extnum_t *idx, /* extent number to update/insert */
2785 xfs_bmbt_irec_t *new) /* new data to add to file extents */
1da177e4 2786{
9e5987a7
DC		 2787 xfs_ifork_t *ifp; /* inode fork pointer */
2788 xfs_bmbt_irec_t left; /* left neighbor extent entry */
2789 xfs_filblks_t newlen=0; /* new indirect size */
2790 xfs_filblks_t oldlen=0; /* old indirect size */
2791 xfs_bmbt_irec_t right; /* right neighbor extent entry */
2792 int state; /* state bits, accessed thru macros */
2793 xfs_filblks_t temp=0; /* temp for indirect calculations */
1da177e4 2794
9e5987a7
DC		 2795 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2796 state = 0;
2797 ASSERT(isnullstartblock(new->br_startblock));
1da177e4
LT		 2798
2799 /*
9e5987a7 2800 * Check and set flags if this segment has a left neighbor
1da177e4 2801 */
9e5987a7
DC		 2802 if (*idx > 0) {
2803 state |= BMAP_LEFT_VALID;
2804 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
2805
2806 if (isnullstartblock(left.br_startblock))
2807 state |= BMAP_LEFT_DELAY;
1da177e4 2808 }
1da177e4 2809
9e5987a7
DC		 2810 /*
2811 * Check and set flags if the current (right) segment exists.
2812 * If it doesn't exist, we're converting the hole at end-of-file.
2813 */
2814 if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
2815 state |= BMAP_RIGHT_VALID;
2816 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
1da177e4 2817
9e5987a7
DC		 2818 if (isnullstartblock(right.br_startblock))
2819 state |= BMAP_RIGHT_DELAY;
1da177e4 2820 }
9e5987a7 2821
1da177e4 2822 /*
9e5987a7
DC		 2823 * Set contiguity flags on the left and right neighbors.
2824 * Don't let extents get too large, even if the pieces are contiguous.
1da177e4 2825 */
9e5987a7
DC		 2826 if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2827 left.br_startoff + left.br_blockcount == new->br_startoff &&
2828 left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2829 state |= BMAP_LEFT_CONTIG;
2830
2831 if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2832 new->br_startoff + new->br_blockcount == right.br_startoff &&
2833 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2834 (!(state & BMAP_LEFT_CONTIG) ||
2835 (left.br_blockcount + new->br_blockcount +
2836 right.br_blockcount <= MAXEXTLEN)))
2837 state |= BMAP_RIGHT_CONTIG;
cc09c0dc
DC		 2838
2839 /*
9e5987a7 2840 * Switch out based on the contiguity flags.
cc09c0dc 2841 */
9e5987a7
DC		 2842 switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2843 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2844 /*
2845 * New allocation is contiguous with delayed allocations
2846 * on the left and on the right.
2847 * Merge all three into a single extent record.
2848 */
2849 --*idx;
2850 temp = left.br_blockcount + new->br_blockcount +
2851 right.br_blockcount;
cc09c0dc 2852
9e5987a7
DC		 2853 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2854 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2855 oldlen = startblockval(left.br_startblock) +
2856 startblockval(new->br_startblock) +
2857 startblockval(right.br_startblock);
2858 newlen = xfs_bmap_worst_indlen(ip, temp);
2859 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2860 nullstartblock((int)newlen));
2861 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1da177e4 2862
9e5987a7
DC		 2863 xfs_iext_remove(ip, *idx + 1, 1, state);
2864 break;
1da177e4 2865
9e5987a7
DC		 2866 case BMAP_LEFT_CONTIG:
2867 /*
2868 * New allocation is contiguous with a delayed allocation
2869 * on the left.
2870 * Merge the new allocation with the left neighbor.
2871 */
2872 --*idx;
2873 temp = left.br_blockcount + new->br_blockcount;
1da177e4 2874
9e5987a7
DC		 2875 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2876 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2877 oldlen = startblockval(left.br_startblock) +
2878 startblockval(new->br_startblock);
2879 newlen = xfs_bmap_worst_indlen(ip, temp);
2880 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2881 nullstartblock((int)newlen));
2882 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2883 break;
1da177e4 2884
9e5987a7
DC		 2885 case BMAP_RIGHT_CONTIG:
2886 /*
2887 * New allocation is contiguous with a delayed allocation
2888 * on the right.
2889 * Merge the new allocation with the right neighbor.
2890 */
2891 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2892 temp = new->br_blockcount + right.br_blockcount;
2893 oldlen = startblockval(new->br_startblock) +
2894 startblockval(right.br_startblock);
2895 newlen = xfs_bmap_worst_indlen(ip, temp);
2896 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2897 new->br_startoff,
2898 nullstartblock((int)newlen), temp, right.br_state);
2899 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2900 break;
2901
2902 case 0:
2903 /*
2904 * New allocation is not contiguous with another
2905 * delayed allocation.
2906 * Insert a new entry.
2907 */
2908 oldlen = newlen = 0;
2909 xfs_iext_insert(ip, *idx, 1, new, state);
2910 break;
1da177e4 2911 }
9e5987a7
DC		 2912 if (oldlen != newlen) {
2913 ASSERT(oldlen > newlen);
0d485ada
DC		 2914 xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen),
2915 false);
1da177e4 2916 /*
9e5987a7 2917 * Nothing to do for disk quota accounting here.
1da177e4 2918 */
1da177e4 2919 }
1da177e4
LT		 2920}
2921
2922/*
9e5987a7 2923 * Convert a hole to a real allocation.
1da177e4 2924 */
9e5987a7
DC		 2925STATIC int /* error */
2926xfs_bmap_add_extent_hole_real(
2927 struct xfs_bmalloca *bma,
2928 int whichfork)
27a3f8f2 2929{
9e5987a7
DC		 2930 struct xfs_bmbt_irec *new = &bma->got;
2931 int error; /* error return value */
2932 int i; /* temp state */
2933 xfs_ifork_t *ifp; /* inode fork pointer */
2934 xfs_bmbt_irec_t left; /* left neighbor extent entry */
2935 xfs_bmbt_irec_t right; /* right neighbor extent entry */
2936 int rval=0; /* return value (logging flags) */
2937 int state; /* state bits, accessed thru macros */
c29aad41 2938 struct xfs_mount *mp;
27a3f8f2 2939
c29aad41 2940 mp = bma->tp ? bma->tp->t_mountp : NULL;
9e5987a7 2941 ifp = XFS_IFORK_PTR(bma->ip, whichfork);
27a3f8f2 2942
9e5987a7
DC		 2943 ASSERT(bma->idx >= 0);
2944 ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2945 ASSERT(!isnullstartblock(new->br_startblock));
2946 ASSERT(!bma->cur ||
2947 !(bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
27a3f8f2 2948
9e5987a7 2949 XFS_STATS_INC(xs_add_exlist);
27a3f8f2 2950
9e5987a7
DC		 2951 state = 0;
2952 if (whichfork == XFS_ATTR_FORK)
2953 state |= BMAP_ATTRFORK;
27a3f8f2 2954
9e5987a7
DC		 2955 /*
2956 * Check and set flags if this segment has a left neighbor.
2957 */
2958 if (bma->idx > 0) {
2959 state |= BMAP_LEFT_VALID;
2960 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &left);
2961 if (isnullstartblock(left.br_startblock))
2962 state |= BMAP_LEFT_DELAY;
2963 }
27a3f8f2
CH		 2964
2965 /*
9e5987a7
DC		 2966 * Check and set flags if this segment has a current value.
2967 * Not true if we're inserting into the "hole" at eof.
27a3f8f2 2968 */
9e5987a7
DC		 2969 if (bma->idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
2970 state |= BMAP_RIGHT_VALID;
2971 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &right);
2972 if (isnullstartblock(right.br_startblock))
2973 state |= BMAP_RIGHT_DELAY;
2974 }
27a3f8f2 2975
9e5987a7
DC		 2976 /*
2977 * We're inserting a real allocation between "left" and "right".
2978 * Set the contiguity flags. Don't let extents get too large.
2979 */
2980 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2981 left.br_startoff + left.br_blockcount == new->br_startoff &&
2982 left.br_startblock + left.br_blockcount == new->br_startblock &&
2983 left.br_state == new->br_state &&
2984 left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2985 state |= BMAP_LEFT_CONTIG;
27a3f8f2 2986
9e5987a7
DC		 2987 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2988 new->br_startoff + new->br_blockcount == right.br_startoff &&
2989 new->br_startblock + new->br_blockcount == right.br_startblock &&
2990 new->br_state == right.br_state &&
2991 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2992 (!(state & BMAP_LEFT_CONTIG) ||
2993 left.br_blockcount + new->br_blockcount +
2994 right.br_blockcount <= MAXEXTLEN))
2995 state |= BMAP_RIGHT_CONTIG;
27a3f8f2 2996
9e5987a7
DC		 2997 error = 0;
2998 /*
2999 * Select which case we're in here, and implement it.
3000 */
3001 switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
3002 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
3003 /*
3004 * New allocation is contiguous with real allocations on the
3005 * left and on the right.
3006 * Merge all three into a single extent record.
3007 */
3008 --bma->idx;
3009 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3010 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3011 left.br_blockcount + new->br_blockcount +
3012 right.br_blockcount);
3013 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
27a3f8f2 3014
9e5987a7 3015 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
27a3f8f2 3016
9e5987a7
DC		 3017 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3018 XFS_IFORK_NEXTENTS(bma->ip, whichfork) - 1);
3019 if (bma->cur == NULL) {
3020 rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3021 } else {
3022 rval = XFS_ILOG_CORE;
3023 error = xfs_bmbt_lookup_eq(bma->cur, right.br_startoff,
3024 right.br_startblock, right.br_blockcount,
3025 &i);
3026 if (error)
3027 goto done;
c29aad41 3028 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 3029 error = xfs_btree_delete(bma->cur, &i);
3030 if (error)
3031 goto done;
c29aad41 3032 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 3033 error = xfs_btree_decrement(bma->cur, 0, &i);
3034 if (error)
3035 goto done;
c29aad41 3036 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 3037 error = xfs_bmbt_update(bma->cur, left.br_startoff,
3038 left.br_startblock,
3039 left.br_blockcount +
3040 new->br_blockcount +
3041 right.br_blockcount,
3042 left.br_state);
3043 if (error)
3044 goto done;
3045 }
3046 break;
27a3f8f2 3047
9e5987a7
DC		 3048 case BMAP_LEFT_CONTIG:
3049 /*
3050 * New allocation is contiguous with a real allocation
3051 * on the left.
3052 * Merge the new allocation with the left neighbor.
3053 */
3054 --bma->idx;
3055 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3056 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3057 left.br_blockcount + new->br_blockcount);
3058 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1da177e4 3059
9e5987a7
DC		 3060 if (bma->cur == NULL) {
3061 rval = xfs_ilog_fext(whichfork);
3062 } else {
3063 rval = 0;
3064 error = xfs_bmbt_lookup_eq(bma->cur, left.br_startoff,
3065 left.br_startblock, left.br_blockcount,
3066 &i);
3067 if (error)
3068 goto done;
c29aad41 3069 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 3070 error = xfs_bmbt_update(bma->cur, left.br_startoff,
3071 left.br_startblock,
3072 left.br_blockcount +
3073 new->br_blockcount,
3074 left.br_state);
3075 if (error)
3076 goto done;
3077 }
3078 break;
27a3f8f2 3079
9e5987a7
DC		 3080 case BMAP_RIGHT_CONTIG:
3081 /*
3082 * New allocation is contiguous with a real allocation
3083 * on the right.
3084 * Merge the new allocation with the right neighbor.
3085 */
3086 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3087 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx),
3088 new->br_startoff, new->br_startblock,
3089 new->br_blockcount + right.br_blockcount,
3090 right.br_state);
3091 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
27a3f8f2 3092
9e5987a7
DC		 3093 if (bma->cur == NULL) {
3094 rval = xfs_ilog_fext(whichfork);
3095 } else {
3096 rval = 0;
3097 error = xfs_bmbt_lookup_eq(bma->cur,
3098 right.br_startoff,
3099 right.br_startblock,
3100 right.br_blockcount, &i);
3101 if (error)
3102 goto done;
c29aad41 3103 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 3104 error = xfs_bmbt_update(bma->cur, new->br_startoff,
3105 new->br_startblock,
3106 new->br_blockcount +
3107 right.br_blockcount,
3108 right.br_state);
3109 if (error)
3110 goto done;
3111 }
3112 break;
3113
3114 case 0:
3115 /*
3116 * New allocation is not contiguous with another
3117 * real allocation.
3118 * Insert a new entry.
3119 */
3120 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
3121 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3122 XFS_IFORK_NEXTENTS(bma->ip, whichfork) + 1);
3123 if (bma->cur == NULL) {
3124 rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3125 } else {
3126 rval = XFS_ILOG_CORE;
3127 error = xfs_bmbt_lookup_eq(bma->cur,
3128 new->br_startoff,
3129 new->br_startblock,
3130 new->br_blockcount, &i);
3131 if (error)
3132 goto done;
c29aad41 3133 XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
9e5987a7
DC		 3134 bma->cur->bc_rec.b.br_state = new->br_state;
3135 error = xfs_btree_insert(bma->cur, &i);
3136 if (error)
3137 goto done;
c29aad41 3138 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 3139 }
3140 break;
3141 }
3142
3143 /* convert to a btree if necessary */
3144 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
3145 int tmp_logflags; /* partial log flag return val */
3146
3147 ASSERT(bma->cur == NULL);
3148 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
3149 bma->firstblock, bma->flist, &bma->cur,
3150 0, &tmp_logflags, whichfork);
3151 bma->logflags |= tmp_logflags;
3152 if (error)
3153 goto done;
3154 }
3155
3156 /* clear out the allocated field, done with it now in any case. */
3157 if (bma->cur)
3158 bma->cur->bc_private.b.allocated = 0;
3159
3160 xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
3161done:
3162 bma->logflags |= rval;
3163 return error;
1da177e4
LT		 3164}
3165
3166/*
9e5987a7 3167 * Functions used in the extent read, allocate and remove paths
1da177e4 3168 */
1da177e4 3169
9e5987a7
DC		 3170/*
3171 * Adjust the size of the new extent based on di_extsize and rt extsize.
3172 */
68988114 3173int
9e5987a7
DC		 3174xfs_bmap_extsize_align(
3175 xfs_mount_t *mp,
3176 xfs_bmbt_irec_t *gotp, /* next extent pointer */
3177 xfs_bmbt_irec_t *prevp, /* previous extent pointer */
3178 xfs_extlen_t extsz, /* align to this extent size */
3179 int rt, /* is this a realtime inode? */
3180 int eof, /* is extent at end-of-file? */
3181 int delay, /* creating delalloc extent? */
3182 int convert, /* overwriting unwritten extent? */
3183 xfs_fileoff_t *offp, /* in/out: aligned offset */
3184 xfs_extlen_t *lenp) /* in/out: aligned length */
4e8938fe 3185{
9e5987a7
DC		 3186 xfs_fileoff_t orig_off; /* original offset */
3187 xfs_extlen_t orig_alen; /* original length */
3188 xfs_fileoff_t orig_end; /* original off+len */
3189 xfs_fileoff_t nexto; /* next file offset */
3190 xfs_fileoff_t prevo; /* previous file offset */
3191 xfs_fileoff_t align_off; /* temp for offset */
3192 xfs_extlen_t align_alen; /* temp for length */
3193 xfs_extlen_t temp; /* temp for calculations */
4e8938fe 3194
9e5987a7 3195 if (convert)
4e8938fe 3196 return 0;
4e8938fe 3197
9e5987a7
DC		 3198 orig_off = align_off = *offp;
3199 orig_alen = align_alen = *lenp;
3200 orig_end = orig_off + orig_alen;
1da177e4 3201
1da177e4 3202 /*
9e5987a7
DC		 3203 * If this request overlaps an existing extent, then don't
3204 * attempt to perform any additional alignment.
1da177e4 3205 */
9e5987a7
DC		 3206 if (!delay && !eof &&
3207 (orig_off >= gotp->br_startoff) &&
3208 (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
3209 return 0;
3210 }
3211
1da177e4 3212 /*
9e5987a7
DC		 3213 * If the file offset is unaligned vs. the extent size
3214 * we need to align it. This will be possible unless
3215 * the file was previously written with a kernel that didn't
3216 * perform this alignment, or if a truncate shot us in the
3217 * foot.
1da177e4 3218 */
9e5987a7
DC		 3219 temp = do_mod(orig_off, extsz);
3220 if (temp) {
3221 align_alen += temp;
3222 align_off -= temp;
1da177e4
LT		 3223 }
3224 /*
9e5987a7 3225 * Same adjustment for the end of the requested area.
1da177e4 3226 */
9e5987a7
DC		 3227 if ((temp = (align_alen % extsz))) {
3228 align_alen += extsz - temp;
3229 }
1da177e4 3230 /*
9e5987a7
DC		 3231 * If the previous block overlaps with this proposed allocation
3232 * then move the start forward without adjusting the length.
1da177e4 3233 */
9e5987a7
DC		 3234 if (prevp->br_startoff != NULLFILEOFF) {
3235 if (prevp->br_startblock == HOLESTARTBLOCK)
3236 prevo = prevp->br_startoff;
3237 else
3238 prevo = prevp->br_startoff + prevp->br_blockcount;
3239 } else
3240 prevo = 0;
3241 if (align_off != orig_off && align_off < prevo)
3242 align_off = prevo;
3243 /*
3244 * If the next block overlaps with this proposed allocation
3245 * then move the start back without adjusting the length,
3246 * but not before offset 0.
3247 * This may of course make the start overlap previous block,
3248 * and if we hit the offset 0 limit then the next block
3249 * can still overlap too.
3250 */
3251 if (!eof && gotp->br_startoff != NULLFILEOFF) {
3252 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
3253 (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
3254 nexto = gotp->br_startoff + gotp->br_blockcount;
3255 else
3256 nexto = gotp->br_startoff;
3257 } else
3258 nexto = NULLFILEOFF;
3259 if (!eof &&
3260 align_off + align_alen != orig_end &&
3261 align_off + align_alen > nexto)
3262 align_off = nexto > align_alen ? nexto - align_alen : 0;
3263 /*
3264 * If we're now overlapping the next or previous extent that
3265 * means we can't fit an extsz piece in this hole. Just move
3266 * the start forward to the first valid spot and set
3267 * the length so we hit the end.
3268 */
3269 if (align_off != orig_off && align_off < prevo)
3270 align_off = prevo;
3271 if (align_off + align_alen != orig_end &&
3272 align_off + align_alen > nexto &&
3273 nexto != NULLFILEOFF) {
3274 ASSERT(nexto > prevo);
3275 align_alen = nexto - align_off;
3276 }
1da177e4 3277
9e5987a7
DC		 3278 /*
3279 * If realtime, and the result isn't a multiple of the realtime
3280 * extent size we need to remove blocks until it is.
3281 */
3282 if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
1da177e4 3283 /*
9e5987a7
DC		 3284 * We're not covering the original request, or
3285 * we won't be able to once we fix the length.
1da177e4 3286 */
9e5987a7
DC		 3287 if (orig_off < align_off ||
3288 orig_end > align_off + align_alen ||
3289 align_alen - temp < orig_alen)
2451337d 3290 return -EINVAL;
1da177e4 3291 /*
9e5987a7 3292 * Try to fix it by moving the start up.
1da177e4 3293 */
9e5987a7
DC		 3294 if (align_off + temp <= orig_off) {
3295 align_alen -= temp;
3296 align_off += temp;
1da177e4 3297 }
9e5987a7
DC		 3298 /*
3299 * Try to fix it by moving the end in.
3300 */
3301 else if (align_off + align_alen - temp >= orig_end)
3302 align_alen -= temp;
3303 /*
3304 * Set the start to the minimum then trim the length.
3305 */
3306 else {
3307 align_alen -= orig_off - align_off;
3308 align_off = orig_off;
3309 align_alen -= align_alen % mp->m_sb.sb_rextsize;
1da177e4 3310 }
1da177e4 3311 /*
9e5987a7 3312 * Result doesn't cover the request, fail it.
1da177e4 3313 */
9e5987a7 3314 if (orig_off < align_off || orig_end > align_off + align_alen)
2451337d 3315 return -EINVAL;
9e5987a7
DC		 3316 } else {
3317 ASSERT(orig_off >= align_off);
3318 ASSERT(orig_end <= align_off + align_alen);
1da177e4 3319 }
1da177e4 3320
0b1b213f 3321#ifdef DEBUG
9e5987a7
DC		 3322 if (!eof && gotp->br_startoff != NULLFILEOFF)
3323 ASSERT(align_off + align_alen <= gotp->br_startoff);
3324 if (prevp->br_startoff != NULLFILEOFF)
3325 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3326#endif
1da177e4 3327
9e5987a7
DC		 3328 *lenp = align_alen;
3329 *offp = align_off;
3330 return 0;
1da177e4 3331}
1da177e4 3332
9e5987a7
DC		 3333#define XFS_ALLOC_GAP_UNITS 4
3334
68988114 3335void
9e5987a7 3336xfs_bmap_adjacent(
68988114 3337 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
1da177e4 3338{
9e5987a7
DC		 3339 xfs_fsblock_t adjust; /* adjustment to block numbers */
3340 xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
3341 xfs_mount_t *mp; /* mount point structure */
3342 int nullfb; /* true if ap->firstblock isn't set */
3343 int rt; /* true if inode is realtime */
1da177e4 3344
9e5987a7
DC		 3345#define ISVALID(x,y) \
3346 (rt ? \
3347 (x) < mp->m_sb.sb_rblocks : \
3348 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3349 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3350 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
1da177e4 3351
9e5987a7
DC		 3352 mp = ap->ip->i_mount;
3353 nullfb = *ap->firstblock == NULLFSBLOCK;
3354 rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
3355 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3356 /*
3357 * If allocating at eof, and there's a previous real block,
3358 * try to use its last block as our starting point.
3359 */
3360 if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3361 !isnullstartblock(ap->prev.br_startblock) &&
3362 ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3363 ap->prev.br_startblock)) {
3364 ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3365 /*
3366 * Adjust for the gap between prevp and us.
3367 */
3368 adjust = ap->offset -
3369 (ap->prev.br_startoff + ap->prev.br_blockcount);
3370 if (adjust &&
3371 ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3372 ap->blkno += adjust;
3373 }
3374 /*
3375 * If not at eof, then compare the two neighbor blocks.
3376 * Figure out whether either one gives us a good starting point,
3377 * and pick the better one.
3378 */
3379 else if (!ap->eof) {
3380 xfs_fsblock_t gotbno; /* right side block number */
3381 xfs_fsblock_t gotdiff=0; /* right side difference */
3382 xfs_fsblock_t prevbno; /* left side block number */
3383 xfs_fsblock_t prevdiff=0; /* left side difference */
3384
3385 /*
3386 * If there's a previous (left) block, select a requested
3387 * start block based on it.
3388 */
3389 if (ap->prev.br_startoff != NULLFILEOFF &&
3390 !isnullstartblock(ap->prev.br_startblock) &&
3391 (prevbno = ap->prev.br_startblock +
3392 ap->prev.br_blockcount) &&
3393 ISVALID(prevbno, ap->prev.br_startblock)) {
3394 /*
3395 * Calculate gap to end of previous block.
3396 */
3397 adjust = prevdiff = ap->offset -
3398 (ap->prev.br_startoff +
3399 ap->prev.br_blockcount);
3400 /*
3401 * Figure the startblock based on the previous block's
3402 * end and the gap size.
3403 * Heuristic!
3404 * If the gap is large relative to the piece we're
3405 * allocating, or using it gives us an invalid block
3406 * number, then just use the end of the previous block.
3407 */
3408 if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3409 ISVALID(prevbno + prevdiff,
3410 ap->prev.br_startblock))
3411 prevbno += adjust;
3412 else
3413 prevdiff += adjust;
3414 /*
3415 * If the firstblock forbids it, can't use it,
3416 * must use default.
3417 */
3418 if (!rt && !nullfb &&
3419 XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3420 prevbno = NULLFSBLOCK;
1da177e4 3421 }
9e5987a7
DC		 3422 /*
3423 * No previous block or can't follow it, just default.
3424 */
3425 else
3426 prevbno = NULLFSBLOCK;
3427 /*
3428 * If there's a following (right) block, select a requested
3429 * start block based on it.
3430 */
3431 if (!isnullstartblock(ap->got.br_startblock)) {
3432 /*
3433 * Calculate gap to start of next block.
3434 */
3435 adjust = gotdiff = ap->got.br_startoff - ap->offset;
3436 /*
3437 * Figure the startblock based on the next block's
3438 * start and the gap size.
3439 */
3440 gotbno = ap->got.br_startblock;
3441 /*
3442 * Heuristic!
3443 * If the gap is large relative to the piece we're
3444 * allocating, or using it gives us an invalid block
3445 * number, then just use the start of the next block
3446 * offset by our length.
3447 */
3448 if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3449 ISVALID(gotbno - gotdiff, gotbno))
3450 gotbno -= adjust;
3451 else if (ISVALID(gotbno - ap->length, gotbno)) {
3452 gotbno -= ap->length;
3453 gotdiff += adjust - ap->length;
3454 } else
3455 gotdiff += adjust;
3456 /*
3457 * If the firstblock forbids it, can't use it,
3458 * must use default.
3459 */
3460 if (!rt && !nullfb &&
3461 XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3462 gotbno = NULLFSBLOCK;
3463 }
3464 /*
3465 * No next block, just default.
3466 */
3467 else
3468 gotbno = NULLFSBLOCK;
3469 /*
3470 * If both valid, pick the better one, else the only good
3471 * one, else ap->blkno is already set (to 0 or the inode block).
3472 */
3473 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3474 ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3475 else if (prevbno != NULLFSBLOCK)
3476 ap->blkno = prevbno;
3477 else if (gotbno != NULLFSBLOCK)
3478 ap->blkno = gotbno;
1da177e4 3479 }
9e5987a7 3480#undef ISVALID
1da177e4 3481}
1da177e4 3482
c977eb10
CH		 3483static int
3484xfs_bmap_longest_free_extent(
3485 struct xfs_trans *tp,
3486 xfs_agnumber_t ag,
3487 xfs_extlen_t *blen,
3488 int *notinit)
3489{
3490 struct xfs_mount *mp = tp->t_mountp;
3491 struct xfs_perag *pag;
3492 xfs_extlen_t longest;
3493 int error = 0;
3494
3495 pag = xfs_perag_get(mp, ag);
3496 if (!pag->pagf_init) {
3497 error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK);
3498 if (error)
3499 goto out;
3500
3501 if (!pag->pagf_init) {
3502 *notinit = 1;
3503 goto out;
3504 }
3505 }
3506
3507 longest = xfs_alloc_longest_free_extent(mp, pag);
3508 if (*blen < longest)
3509 *blen = longest;
3510
3511out:
3512 xfs_perag_put(pag);
3513 return error;
3514}
3515
3516static void
3517xfs_bmap_select_minlen(
3518 struct xfs_bmalloca *ap,
3519 struct xfs_alloc_arg *args,
3520 xfs_extlen_t *blen,
3521 int notinit)
3522{
3523 if (notinit || *blen < ap->minlen) {
3524 /*
3525 * Since we did a BUF_TRYLOCK above, it is possible that
3526 * there is space for this request.
3527 */
3528 args->minlen = ap->minlen;
3529 } else if (*blen < args->maxlen) {
3530 /*
3531 * If the best seen length is less than the request length,
3532 * use the best as the minimum.
3533 */
3534 args->minlen = *blen;
3535 } else {
3536 /*
3537 * Otherwise we've seen an extent as big as maxlen, use that
3538 * as the minimum.
3539 */
3540 args->minlen = args->maxlen;
3541 }
3542}
3543
b64dfe4e 3544STATIC int
9e5987a7
DC		 3545xfs_bmap_btalloc_nullfb(
3546 struct xfs_bmalloca *ap,
3547 struct xfs_alloc_arg *args,
3548 xfs_extlen_t *blen)
b64dfe4e 3549{
9e5987a7 3550 struct xfs_mount *mp = ap->ip->i_mount;
9e5987a7
DC		 3551 xfs_agnumber_t ag, startag;
3552 int notinit = 0;
b64dfe4e
CH		 3553 int error;
3554
c977eb10 3555 args->type = XFS_ALLOCTYPE_START_BNO;
9e5987a7 3556 args->total = ap->total;
b64dfe4e 3557
9e5987a7
DC		 3558 startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3559 if (startag == NULLAGNUMBER)
3560 startag = ag = 0;
b64dfe4e 3561
9e5987a7 3562 while (*blen < args->maxlen) {
c977eb10
CH		 3563 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3564 &notinit);
3565 if (error)
3566 return error;
b64dfe4e 3567
9e5987a7
DC		 3568 if (++ag == mp->m_sb.sb_agcount)
3569 ag = 0;
3570 if (ag == startag)
3571 break;
9e5987a7 3572 }
b64dfe4e 3573
c977eb10
CH		 3574 xfs_bmap_select_minlen(ap, args, blen, notinit);
3575 return 0;
3576}
3577
3578STATIC int
3579xfs_bmap_btalloc_filestreams(
3580 struct xfs_bmalloca *ap,
3581 struct xfs_alloc_arg *args,
3582 xfs_extlen_t *blen)
3583{
3584 struct xfs_mount *mp = ap->ip->i_mount;
3585 xfs_agnumber_t ag;
3586 int notinit = 0;
3587 int error;
3588
3589 args->type = XFS_ALLOCTYPE_NEAR_BNO;
3590 args->total = ap->total;
3591
3592 ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3593 if (ag == NULLAGNUMBER)
3594 ag = 0;
3595
3596 error = xfs_bmap_longest_free_extent(args->tp, ag, blen, &notinit);
3597 if (error)
3598 return error;
3599
3600 if (*blen < args->maxlen) {
3601 error = xfs_filestream_new_ag(ap, &ag);
3602 if (error)
3603 return error;
3604
3605 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3606 &notinit);
3607 if (error)
3608 return error;
3609
3610 }
3611
3612 xfs_bmap_select_minlen(ap, args, blen, notinit);
b64dfe4e
CH		 3613
3614 /*
c977eb10
CH		 3615 * Set the failure fallback case to look in the selected AG as stream
3616 * may have moved.
b64dfe4e 3617 */
c977eb10 3618 ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
b64dfe4e 3619 return 0;
b64dfe4e
CH		 3620}
3621
9e5987a7
DC		 3622STATIC int
3623xfs_bmap_btalloc(
68988114 3624 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
4403280a 3625{
9e5987a7
DC		 3626 xfs_mount_t *mp; /* mount point structure */
3627 xfs_alloctype_t atype = 0; /* type for allocation routines */
3628 xfs_extlen_t align; /* minimum allocation alignment */
3629 xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
3630 xfs_agnumber_t ag;
3631 xfs_alloc_arg_t args;
3632 xfs_extlen_t blen;
3633 xfs_extlen_t nextminlen = 0;
3634 int nullfb; /* true if ap->firstblock isn't set */
3635 int isaligned;
3636 int tryagain;
3637 int error;
33177f05 3638 int stripe_align;
4403280a 3639
9e5987a7 3640 ASSERT(ap->length);
4403280a 3641
9e5987a7 3642 mp = ap->ip->i_mount;
33177f05
DC		 3643
3644 /* stripe alignment for allocation is determined by mount parameters */
3645 stripe_align = 0;
3646 if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
3647 stripe_align = mp->m_swidth;
3648 else if (mp->m_dalign)
3649 stripe_align = mp->m_dalign;
3650
9e5987a7
DC		 3651 align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
3652 if (unlikely(align)) {
3653 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
3654 align, 0, ap->eof, 0, ap->conv,
3655 &ap->offset, &ap->length);
3656 ASSERT(!error);
3657 ASSERT(ap->length);
4403280a 3658 }
33177f05
DC		 3659
3660
9e5987a7
DC		 3661 nullfb = *ap->firstblock == NULLFSBLOCK;
3662 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3663 if (nullfb) {
3664 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
3665 ag = xfs_filestream_lookup_ag(ap->ip);
3666 ag = (ag != NULLAGNUMBER) ? ag : 0;
3667 ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3668 } else {
3669 ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3670 }
3671 } else
3672 ap->blkno = *ap->firstblock;
4403280a 3673
9e5987a7 3674 xfs_bmap_adjacent(ap);
4403280a 3675
9e5987a7
DC		 3676 /*
3677 * If allowed, use ap->blkno; otherwise must use firstblock since
3678 * it's in the right allocation group.
3679 */
3680 if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3681 ;
3682 else
3683 ap->blkno = *ap->firstblock;
3684 /*
3685 * Normal allocation, done through xfs_alloc_vextent.
3686 */
3687 tryagain = isaligned = 0;
3688 memset(&args, 0, sizeof(args));
3689 args.tp = ap->tp;
3690 args.mp = mp;
3691 args.fsbno = ap->blkno;
4403280a 3692
9e5987a7
DC		 3693 /* Trim the allocation back to the maximum an AG can fit. */
3694 args.maxlen = MIN(ap->length, XFS_ALLOC_AG_MAX_USABLE(mp));
3695 args.firstblock = *ap->firstblock;
3696 blen = 0;
3697 if (nullfb) {
c977eb10
CH		 3698 /*
3699 * Search for an allocation group with a single extent large
3700 * enough for the request. If one isn't found, then adjust
3701 * the minimum allocation size to the largest space found.
3702 */
3703 if (ap->userdata && xfs_inode_is_filestream(ap->ip))
3704 error = xfs_bmap_btalloc_filestreams(ap, &args, &blen);
3705 else
3706 error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
4403280a
CH		 3707 if (error)
3708 return error;
9e5987a7
DC		 3709 } else if (ap->flist->xbf_low) {
3710 if (xfs_inode_is_filestream(ap->ip))
3711 args.type = XFS_ALLOCTYPE_FIRST_AG;
3712 else
3713 args.type = XFS_ALLOCTYPE_START_BNO;
3714 args.total = args.minlen = ap->minlen;
3715 } else {
3716 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3717 args.total = ap->total;
3718 args.minlen = ap->minlen;
4403280a 3719 }
9e5987a7
DC		 3720 /* apply extent size hints if obtained earlier */
3721 if (unlikely(align)) {
3722 args.prod = align;
3723 if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
3724 args.mod = (xfs_extlen_t)(args.prod - args.mod);
3725 } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
3726 args.prod = 1;
3727 args.mod = 0;
3728 } else {
3729 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
3730 if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
3731 args.mod = (xfs_extlen_t)(args.prod - args.mod);
4403280a 3732 }
7e47a4ef 3733 /*
9e5987a7
DC		 3734 * If we are not low on available data blocks, and the
3735 * underlying logical volume manager is a stripe, and
3736 * the file offset is zero then try to allocate data
3737 * blocks on stripe unit boundary.
3738 * NOTE: ap->aeof is only set if the allocation length
3739 * is >= the stripe unit and the allocation offset is
3740 * at the end of file.
7e47a4ef 3741 */
9e5987a7
DC		 3742 if (!ap->flist->xbf_low && ap->aeof) {
3743 if (!ap->offset) {
33177f05 3744 args.alignment = stripe_align;
9e5987a7
DC		 3745 atype = args.type;
3746 isaligned = 1;
3747 /*
3748 * Adjust for alignment
3749 */
3750 if (blen > args.alignment && blen <= args.maxlen)
3751 args.minlen = blen - args.alignment;
3752 args.minalignslop = 0;
3753 } else {
3754 /*
3755 * First try an exact bno allocation.
3756 * If it fails then do a near or start bno
3757 * allocation with alignment turned on.
3758 */
3759 atype = args.type;
3760 tryagain = 1;
3761 args.type = XFS_ALLOCTYPE_THIS_BNO;
3762 args.alignment = 1;
3763 /*
3764 * Compute the minlen+alignment for the
3765 * next case. Set slop so that the value
3766 * of minlen+alignment+slop doesn't go up
3767 * between the calls.
3768 */
33177f05
DC		 3769 if (blen > stripe_align && blen <= args.maxlen)
3770 nextminlen = blen - stripe_align;
9e5987a7
DC		 3771 else
3772 nextminlen = args.minlen;
33177f05 3773 if (nextminlen + stripe_align > args.minlen + 1)
9e5987a7 3774 args.minalignslop =
33177f05 3775 nextminlen + stripe_align -
9e5987a7
DC		 3776 args.minlen - 1;
3777 else
3778 args.minalignslop = 0;
7e47a4ef
DC		 3779 }
3780 } else {
9e5987a7
DC		 3781 args.alignment = 1;
3782 args.minalignslop = 0;
7e47a4ef 3783 }
9e5987a7
DC		 3784 args.minleft = ap->minleft;
3785 args.wasdel = ap->wasdel;
3786 args.isfl = 0;
3787 args.userdata = ap->userdata;
3788 if ((error = xfs_alloc_vextent(&args)))
3789 return error;
3790 if (tryagain && args.fsbno == NULLFSBLOCK) {
3791 /*
3792 * Exact allocation failed. Now try with alignment
3793 * turned on.
3794 */
3795 args.type = atype;
3796 args.fsbno = ap->blkno;
33177f05 3797 args.alignment = stripe_align;
9e5987a7
DC		 3798 args.minlen = nextminlen;
3799 args.minalignslop = 0;
3800 isaligned = 1;
3801 if ((error = xfs_alloc_vextent(&args)))
3802 return error;
7e47a4ef 3803 }
9e5987a7
DC		 3804 if (isaligned && args.fsbno == NULLFSBLOCK) {
3805 /*
3806 * allocation failed, so turn off alignment and
3807 * try again.
3808 */
3809 args.type = atype;
3810 args.fsbno = ap->blkno;
3811 args.alignment = 0;
3812 if ((error = xfs_alloc_vextent(&args)))
7e47a4ef
DC		 3813 return error;
3814 }
9e5987a7
DC		 3815 if (args.fsbno == NULLFSBLOCK && nullfb &&
3816 args.minlen > ap->minlen) {
3817 args.minlen = ap->minlen;
3818 args.type = XFS_ALLOCTYPE_START_BNO;
3819 args.fsbno = ap->blkno;
3820 if ((error = xfs_alloc_vextent(&args)))
3821 return error;
7e47a4ef 3822 }
9e5987a7
DC		 3823 if (args.fsbno == NULLFSBLOCK && nullfb) {
3824 args.fsbno = 0;
3825 args.type = XFS_ALLOCTYPE_FIRST_AG;
3826 args.total = ap->minlen;
3827 args.minleft = 0;
3828 if ((error = xfs_alloc_vextent(&args)))
3829 return error;
3830 ap->flist->xbf_low = 1;
3831 }
3832 if (args.fsbno != NULLFSBLOCK) {
3833 /*
3834 * check the allocation happened at the same or higher AG than
3835 * the first block that was allocated.
3836 */
3837 ASSERT(*ap->firstblock == NULLFSBLOCK ||
3838 XFS_FSB_TO_AGNO(mp, *ap->firstblock) ==
3839 XFS_FSB_TO_AGNO(mp, args.fsbno) ||
3840 (ap->flist->xbf_low &&
3841 XFS_FSB_TO_AGNO(mp, *ap->firstblock) <
3842 XFS_FSB_TO_AGNO(mp, args.fsbno)));
7e47a4ef 3843
9e5987a7
DC		 3844 ap->blkno = args.fsbno;
3845 if (*ap->firstblock == NULLFSBLOCK)
3846 *ap->firstblock = args.fsbno;
3847 ASSERT(nullfb || fb_agno == args.agno ||
3848 (ap->flist->xbf_low && fb_agno < args.agno));
3849 ap->length = args.len;
3850 ap->ip->i_d.di_nblocks += args.len;
3851 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3852 if (ap->wasdel)
3853 ap->ip->i_delayed_blks -= args.len;
3854 /*
3855 * Adjust the disk quota also. This was reserved
3856 * earlier.
3857 */
3858 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3859 ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
3860 XFS_TRANS_DQ_BCOUNT,
3861 (long) args.len);
3862 } else {
3863 ap->blkno = NULLFSBLOCK;
3864 ap->length = 0;
3865 }
3866 return 0;
3867}
7e47a4ef 3868
9e5987a7
DC		 3869/*
3870 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
3871 * It figures out where to ask the underlying allocator to put the new extent.
3872 */
3873STATIC int
3874xfs_bmap_alloc(
68988114 3875 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
9e5987a7
DC		 3876{
3877 if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
3878 return xfs_bmap_rtalloc(ap);
3879 return xfs_bmap_btalloc(ap);
3880}
a5bd606b 3881
9e5987a7
DC		 3882/*
3883 * Trim the returned map to the required bounds
3884 */
3885STATIC void
3886xfs_bmapi_trim_map(
3887 struct xfs_bmbt_irec *mval,
3888 struct xfs_bmbt_irec *got,
3889 xfs_fileoff_t *bno,
3890 xfs_filblks_t len,
3891 xfs_fileoff_t obno,
3892 xfs_fileoff_t end,
3893 int n,
3894 int flags)
3895{
3896 if ((flags & XFS_BMAPI_ENTIRE) ||
3897 got->br_startoff + got->br_blockcount <= obno) {
3898 *mval = *got;
3899 if (isnullstartblock(got->br_startblock))
3900 mval->br_startblock = DELAYSTARTBLOCK;
3901 return;
3902 }
7e47a4ef 3903
9e5987a7
DC		 3904 if (obno > *bno)
3905 *bno = obno;
3906 ASSERT((*bno >= obno) || (n == 0));
3907 ASSERT(*bno < end);
3908 mval->br_startoff = *bno;
3909 if (isnullstartblock(got->br_startblock))
3910 mval->br_startblock = DELAYSTARTBLOCK;
3911 else
3912 mval->br_startblock = got->br_startblock +
3913 (*bno - got->br_startoff);
7e47a4ef 3914 /*
9e5987a7
DC		 3915 * Return the minimum of what we got and what we asked for for
3916 * the length. We can use the len variable here because it is
3917 * modified below and we could have been there before coming
3918 * here if the first part of the allocation didn't overlap what
3919 * was asked for.
7e47a4ef 3920 */
9e5987a7
DC		 3921 mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3922 got->br_blockcount - (*bno - got->br_startoff));
3923 mval->br_state = got->br_state;
3924 ASSERT(mval->br_blockcount <= len);
3925 return;
7e47a4ef
DC		 3926}
3927
9e5987a7
DC		 3928/*
3929 * Update and validate the extent map to return
3930 */
3931STATIC void
3932xfs_bmapi_update_map(
3933 struct xfs_bmbt_irec **map,
3934 xfs_fileoff_t *bno,
3935 xfs_filblks_t *len,
3936 xfs_fileoff_t obno,
3937 xfs_fileoff_t end,
3938 int *n,
3939 int flags)
e04426b9 3940{
9e5987a7 3941 xfs_bmbt_irec_t *mval = *map;
e04426b9 3942
9e5987a7
DC		 3943 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3944 ((mval->br_startoff + mval->br_blockcount) <= end));
3945 ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3946 (mval->br_startoff < obno));
e04426b9 3947
9e5987a7
DC		 3948 *bno = mval->br_startoff + mval->br_blockcount;
3949 *len = end - *bno;
3950 if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3951 /* update previous map with new information */
3952 ASSERT(mval->br_startblock == mval[-1].br_startblock);
3953 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3954 ASSERT(mval->br_state == mval[-1].br_state);
3955 mval[-1].br_blockcount = mval->br_blockcount;
3956 mval[-1].br_state = mval->br_state;
3957 } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3958 mval[-1].br_startblock != DELAYSTARTBLOCK &&
3959 mval[-1].br_startblock != HOLESTARTBLOCK &&
3960 mval->br_startblock == mval[-1].br_startblock +
3961 mval[-1].br_blockcount &&
3962 ((flags & XFS_BMAPI_IGSTATE) ||
3963 mval[-1].br_state == mval->br_state)) {
3964 ASSERT(mval->br_startoff ==
3965 mval[-1].br_startoff + mval[-1].br_blockcount);
3966 mval[-1].br_blockcount += mval->br_blockcount;
3967 } else if (*n > 0 &&
3968 mval->br_startblock == DELAYSTARTBLOCK &&
3969 mval[-1].br_startblock == DELAYSTARTBLOCK &&
3970 mval->br_startoff ==
3971 mval[-1].br_startoff + mval[-1].br_blockcount) {
3972 mval[-1].br_blockcount += mval->br_blockcount;
3973 mval[-1].br_state = mval->br_state;
3974 } else if (!((*n == 0) &&
3975 ((mval->br_startoff + mval->br_blockcount) <=
3976 obno))) {
3977 mval++;
3978 (*n)++;
3979 }
3980 *map = mval;
e04426b9
DC		 3981}
3982
3983/*
9e5987a7 3984 * Map file blocks to filesystem blocks without allocation.
e04426b9
DC		 3985 */
3986int
9e5987a7
DC		 3987xfs_bmapi_read(
3988 struct xfs_inode *ip,
3989 xfs_fileoff_t bno,
b447fe5a 3990 xfs_filblks_t len,
9e5987a7
DC		 3991 struct xfs_bmbt_irec *mval,
3992 int *nmap,
c315c90b 3993 int flags)
b447fe5a 3994{
9e5987a7
DC		 3995 struct xfs_mount *mp = ip->i_mount;
3996 struct xfs_ifork *ifp;
3997 struct xfs_bmbt_irec got;
3998 struct xfs_bmbt_irec prev;
3999 xfs_fileoff_t obno;
4000 xfs_fileoff_t end;
4001 xfs_extnum_t lastx;
4002 int error;
4003 int eof;
4004 int n = 0;
b447fe5a
DC		 4005 int whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4006 XFS_ATTR_FORK : XFS_DATA_FORK;
b447fe5a 4007
9e5987a7
DC		 4008 ASSERT(*nmap >= 1);
4009 ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
4010 XFS_BMAPI_IGSTATE)));
eef334e5 4011 ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL));
b447fe5a 4012
9e5987a7
DC		 4013 if (unlikely(XFS_TEST_ERROR(
4014 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4015 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4016 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4017 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
2451337d 4018 return -EFSCORRUPTED;
9e5987a7 4019 }
b447fe5a 4020
9e5987a7 4021 if (XFS_FORCED_SHUTDOWN(mp))
2451337d 4022 return -EIO;
9e5987a7
DC		 4023
4024 XFS_STATS_INC(xs_blk_mapr);
4025
4026 ifp = XFS_IFORK_PTR(ip, whichfork);
4027
4028 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4029 error = xfs_iread_extents(NULL, ip, whichfork);
4030 if (error)
4031 return error;
b447fe5a 4032 }
b447fe5a 4033
9e5987a7
DC		 4034 xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got, &prev);
4035 end = bno + len;
4036 obno = bno;
b447fe5a 4037
9e5987a7
DC		 4038 while (bno < end && n < *nmap) {
4039 /* Reading past eof, act as though there's a hole up to end. */
4040 if (eof)
4041 got.br_startoff = end;
4042 if (got.br_startoff > bno) {
4043 /* Reading in a hole. */
4044 mval->br_startoff = bno;
4045 mval->br_startblock = HOLESTARTBLOCK;
4046 mval->br_blockcount =
4047 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4048 mval->br_state = XFS_EXT_NORM;
4049 bno += mval->br_blockcount;
4050 len -= mval->br_blockcount;
4051 mval++;
4052 n++;
4053 continue;
4054 }
b447fe5a 4055
9e5987a7
DC		 4056 /* set up the extent map to return. */
4057 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4058 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4059
4060 /* If we're done, stop now. */
4061 if (bno >= end || n >= *nmap)
4062 break;
4063
4064 /* Else go on to the next record. */
4065 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4066 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4067 else
4068 eof = 1;
4069 }
4070 *nmap = n;
b447fe5a
DC		 4071 return 0;
4072}
4073
9e5987a7
DC		 4074STATIC int
4075xfs_bmapi_reserve_delalloc(
4076 struct xfs_inode *ip,
4077 xfs_fileoff_t aoff,
4078 xfs_filblks_t len,
4079 struct xfs_bmbt_irec *got,
4080 struct xfs_bmbt_irec *prev,
4081 xfs_extnum_t *lastx,
4082 int eof)
1da177e4 4083{
c0dc7828 4084 struct xfs_mount *mp = ip->i_mount;
9e5987a7
DC		 4085 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4086 xfs_extlen_t alen;
4087 xfs_extlen_t indlen;
4088 char rt = XFS_IS_REALTIME_INODE(ip);
4089 xfs_extlen_t extsz;
4090 int error;
c0dc7828 4091
9e5987a7
DC		 4092 alen = XFS_FILBLKS_MIN(len, MAXEXTLEN);
4093 if (!eof)
4094 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
1da177e4 4095
9e5987a7
DC		 4096 /* Figure out the extent size, adjust alen */
4097 extsz = xfs_get_extsz_hint(ip);
4098 if (extsz) {
4099 /*
4100 * Make sure we don't exceed a single extent length when we
4101 * align the extent by reducing length we are going to
4102 * allocate by the maximum amount extent size aligment may
4103 * require.
4104 */
4105 alen = XFS_FILBLKS_MIN(len, MAXEXTLEN - (2 * extsz - 1));
4106 error = xfs_bmap_extsize_align(mp, got, prev, extsz, rt, eof,
4107 1, 0, &aoff, &alen);
4108 ASSERT(!error);
4109 }
4110
4111 if (rt)
4112 extsz = alen / mp->m_sb.sb_rextsize;
4113
4114 /*
4115 * Make a transaction-less quota reservation for delayed allocation
4116 * blocks. This number gets adjusted later. We return if we haven't
4117 * allocated blocks already inside this loop.
4118 */
4119 error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4120 rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4121 if (error)
4122 return error;
4123
4124 /*
4125 * Split changing sb for alen and indlen since they could be coming
4126 * from different places.
4127 */
4128 indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4129 ASSERT(indlen > 0);
4130
4131 if (rt) {
bab98bbe 4132 error = xfs_mod_frextents(mp, -((int64_t)extsz));
9e5987a7 4133 } else {
0d485ada 4134 error = xfs_mod_fdblocks(mp, -((int64_t)alen), false);
9e5987a7
DC		 4135 }
4136
4137 if (error)
4138 goto out_unreserve_quota;
4139
0d485ada 4140 error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false);
9e5987a7
DC		 4141 if (error)
4142 goto out_unreserve_blocks;
4143
4144
4145 ip->i_delayed_blks += alen;
4146
4147 got->br_startoff = aoff;
4148 got->br_startblock = nullstartblock(indlen);
4149 got->br_blockcount = alen;
4150 got->br_state = XFS_EXT_NORM;
4151 xfs_bmap_add_extent_hole_delay(ip, lastx, got);
4152
4153 /*
4154 * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4155 * might have merged it into one of the neighbouring ones.
4156 */
4157 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), got);
4158
4159 ASSERT(got->br_startoff <= aoff);
4160 ASSERT(got->br_startoff + got->br_blockcount >= aoff + alen);
4161 ASSERT(isnullstartblock(got->br_startblock));
4162 ASSERT(got->br_state == XFS_EXT_NORM);
4163 return 0;
4164
4165out_unreserve_blocks:
4166 if (rt)
bab98bbe 4167 xfs_mod_frextents(mp, extsz);
9e5987a7 4168 else
0d485ada 4169 xfs_mod_fdblocks(mp, alen, false);
9e5987a7
DC		 4170out_unreserve_quota:
4171 if (XFS_IS_QUOTA_ON(mp))
4172 xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ?
4173 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4174 return error;
4175}
4176
4177/*
4178 * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4179 */
4180int
4181xfs_bmapi_delay(
4182 struct xfs_inode *ip, /* incore inode */
4183 xfs_fileoff_t bno, /* starting file offs. mapped */
4184 xfs_filblks_t len, /* length to map in file */
4185 struct xfs_bmbt_irec *mval, /* output: map values */
4186 int *nmap, /* i/o: mval size/count */
4187 int flags) /* XFS_BMAPI_... */
4188{
4189 struct xfs_mount *mp = ip->i_mount;
4190 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4191 struct xfs_bmbt_irec got; /* current file extent record */
4192 struct xfs_bmbt_irec prev; /* previous file extent record */
4193 xfs_fileoff_t obno; /* old block number (offset) */
4194 xfs_fileoff_t end; /* end of mapped file region */
4195 xfs_extnum_t lastx; /* last useful extent number */
4196 int eof; /* we've hit the end of extents */
4197 int n = 0; /* current extent index */
4198 int error = 0;
c0dc7828 4199
1da177e4 4200 ASSERT(*nmap >= 1);
c0dc7828 4201 ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
9e5987a7 4202 ASSERT(!(flags & ~XFS_BMAPI_ENTIRE));
eef334e5 4203 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
c0dc7828 4204
1da177e4 4205 if (unlikely(XFS_TEST_ERROR(
9e5987a7
DC		 4206 (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4207 XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
1da177e4 4208 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
9e5987a7 4209 XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW, mp);
2451337d 4210 return -EFSCORRUPTED;
1da177e4 4211 }
c0dc7828 4212
1da177e4 4213 if (XFS_FORCED_SHUTDOWN(mp))
2451337d 4214 return -EIO;
c0dc7828 4215
c0dc7828
DC		 4216 XFS_STATS_INC(xs_blk_mapw);
4217
c0dc7828 4218 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
9e5987a7 4219 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
c0dc7828 4220 if (error)
9e5987a7 4221 return error;
c0dc7828
DC		 4222 }
4223
9e5987a7 4224 xfs_bmap_search_extents(ip, bno, XFS_DATA_FORK, &eof, &lastx, &got, &prev);
1da177e4
LT		 4225 end = bno + len;
4226 obno = bno;
7e47a4ef 4227
1da177e4 4228 while (bno < end && n < *nmap) {
9e5987a7
DC		 4229 if (eof || got.br_startoff > bno) {
4230 error = xfs_bmapi_reserve_delalloc(ip, bno, len, &got,
4231 &prev, &lastx, eof);
4232 if (error) {
4233 if (n == 0) {
4234 *nmap = 0;
4235 return error;
4236 }
4237 break;
4238 }
4239 }
c0dc7828 4240
9e5987a7
DC		 4241 /* set up the extent map to return. */
4242 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
aef9a895
DC		 4243 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4244
9e5987a7
DC		 4245 /* If we're done, stop now. */
4246 if (bno >= end || n >= *nmap)
1da177e4 4247 break;
c0dc7828
DC		 4248
4249 /* Else go on to the next record. */
9e5987a7
DC		 4250 prev = got;
4251 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4252 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4253 else
5690f921 4254 eof = 1;
1da177e4 4255 }
9e5987a7 4256
1da177e4 4257 *nmap = n;
9e5987a7
DC		 4258 return 0;
4259}
c0dc7828 4260
c315c90b 4261
cf11da9c
DC		 4262static int
4263xfs_bmapi_allocate(
9e5987a7
DC		 4264 struct xfs_bmalloca *bma)
4265{
4266 struct xfs_mount *mp = bma->ip->i_mount;
4267 int whichfork = (bma->flags & XFS_BMAPI_ATTRFORK) ?
4268 XFS_ATTR_FORK : XFS_DATA_FORK;
4269 struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4270 int tmp_logflags = 0;
4271 int error;
9e5987a7
DC		 4272
4273 ASSERT(bma->length > 0);
4274
1da177e4 4275 /*
9e5987a7
DC		 4276 * For the wasdelay case, we could also just allocate the stuff asked
4277 * for in this bmap call but that wouldn't be as good.
1da177e4 4278 */
9e5987a7
DC		 4279 if (bma->wasdel) {
4280 bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4281 bma->offset = bma->got.br_startoff;
4282 if (bma->idx != NULLEXTNUM && bma->idx) {
4283 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4284 &bma->prev);
1da177e4 4285 }
9e5987a7
DC		 4286 } else {
4287 bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4288 if (!bma->eof)
4289 bma->length = XFS_FILBLKS_MIN(bma->length,
4290 bma->got.br_startoff - bma->offset);
1da177e4 4291 }
1da177e4 4292
9e5987a7
DC		 4293 /*
4294 * Indicate if this is the first user data in the file, or just any
4295 * user data.
4296 */
4297 if (!(bma->flags & XFS_BMAPI_METADATA)) {
4298 bma->userdata = (bma->offset == 0) ?
4299 XFS_ALLOC_INITIAL_USER_DATA : XFS_ALLOC_USERDATA;
4300 }
1da177e4 4301
9e5987a7 4302 bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
0b1b213f 4303
9e5987a7
DC		 4304 /*
4305 * Only want to do the alignment at the eof if it is userdata and
4306 * allocation length is larger than a stripe unit.
4307 */
4308 if (mp->m_dalign && bma->length >= mp->m_dalign &&
4309 !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4310 error = xfs_bmap_isaeof(bma, whichfork);
4311 if (error)
4312 return error;
1da177e4 4313 }
8096b1eb 4314
9e5987a7
DC		 4315 error = xfs_bmap_alloc(bma);
4316 if (error)
1da177e4 4317 return error;
9e5987a7
DC		 4318
4319 if (bma->flist->xbf_low)
4320 bma->minleft = 0;
4321 if (bma->cur)
4322 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4323 if (bma->blkno == NULLFSBLOCK)
1da177e4 4324 return 0;
9e5987a7
DC		 4325 if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4326 bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4327 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4328 bma->cur->bc_private.b.flist = bma->flist;
1da177e4 4329 }
9e5987a7
DC		 4330 /*
4331 * Bump the number of extents we've allocated
4332 * in this call.
4333 */
4334 bma->nallocs++;
4335
4336 if (bma->cur)
4337 bma->cur->bc_private.b.flags =
4338 bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4339
4340 bma->got.br_startoff = bma->offset;
4341 bma->got.br_startblock = bma->blkno;
4342 bma->got.br_blockcount = bma->length;
4343 bma->got.br_state = XFS_EXT_NORM;
b4e9181e 4344
1da177e4 4345 /*
9e5987a7
DC		 4346 * A wasdelay extent has been initialized, so shouldn't be flagged
4347 * as unwritten.
1da177e4 4348 */
9e5987a7
DC		 4349 if (!bma->wasdel && (bma->flags & XFS_BMAPI_PREALLOC) &&
4350 xfs_sb_version_hasextflgbit(&mp->m_sb))
4351 bma->got.br_state = XFS_EXT_UNWRITTEN;
4352
4353 if (bma->wasdel)
4354 error = xfs_bmap_add_extent_delay_real(bma);
4355 else
4356 error = xfs_bmap_add_extent_hole_real(bma, whichfork);
4357
4358 bma->logflags |= tmp_logflags;
4359 if (error)
4360 return error;
4361
4362 /*
4363 * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4364 * or xfs_bmap_add_extent_hole_real might have merged it into one of
4365 * the neighbouring ones.
4366 */
4367 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4368
4369 ASSERT(bma->got.br_startoff <= bma->offset);
4370 ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4371 bma->offset + bma->length);
4372 ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4373 bma->got.br_state == XFS_EXT_UNWRITTEN);
4374 return 0;
4375}
4376
9e5987a7
DC		 4377STATIC int
4378xfs_bmapi_convert_unwritten(
4379 struct xfs_bmalloca *bma,
4380 struct xfs_bmbt_irec *mval,
4381 xfs_filblks_t len,
4382 int flags)
4383{
4384 int whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4385 XFS_ATTR_FORK : XFS_DATA_FORK;
4386 struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4387 int tmp_logflags = 0;
4388 int error;
4389
4390 /* check if we need to do unwritten->real conversion */
4391 if (mval->br_state == XFS_EXT_UNWRITTEN &&
4392 (flags & XFS_BMAPI_PREALLOC))
4393 return 0;
4394
4395 /* check if we need to do real->unwritten conversion */
4396 if (mval->br_state == XFS_EXT_NORM &&
4397 (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4398 (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4399 return 0;
4400
4401 /*
4402 * Modify (by adding) the state flag, if writing.
4403 */
4404 ASSERT(mval->br_blockcount <= len);
4405 if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4406 bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4407 bma->ip, whichfork);
4408 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4409 bma->cur->bc_private.b.flist = bma->flist;
1da177e4 4410 }
9e5987a7
DC		 4411 mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4412 ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
5575acc7 4413
9e5987a7
DC		 4414 error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, &bma->idx,
4415 &bma->cur, mval, bma->firstblock, bma->flist,
4416 &tmp_logflags);
4417 bma->logflags |= tmp_logflags;
4418 if (error)
4419 return error;
4420
4421 /*
4422 * Update our extent pointer, given that
4423 * xfs_bmap_add_extent_unwritten_real might have merged it into one
4424 * of the neighbouring ones.
4425 */
4426 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4427
4428 /*
4429 * We may have combined previously unwritten space with written space,
4430 * so generate another request.
4431 */
4432 if (mval->br_blockcount < len)
2451337d 4433 return -EAGAIN;
9e5987a7
DC		 4434 return 0;
4435}
4436
4437/*
4438 * Map file blocks to filesystem blocks, and allocate blocks or convert the
4439 * extent state if necessary. Details behaviour is controlled by the flags
4440 * parameter. Only allocates blocks from a single allocation group, to avoid
4441 * locking problems.
4442 *
4443 * The returned value in "firstblock" from the first call in a transaction
4444 * must be remembered and presented to subsequent calls in "firstblock".
4445 * An upper bound for the number of blocks to be allocated is supplied to
4446 * the first call in "total"; if no allocation group has that many free
4447 * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4448 */
4449int
4450xfs_bmapi_write(
4451 struct xfs_trans *tp, /* transaction pointer */
4452 struct xfs_inode *ip, /* incore inode */
4453 xfs_fileoff_t bno, /* starting file offs. mapped */
4454 xfs_filblks_t len, /* length to map in file */
4455 int flags, /* XFS_BMAPI_... */
4456 xfs_fsblock_t *firstblock, /* first allocated block
4457 controls a.g. for allocs */
4458 xfs_extlen_t total, /* total blocks needed */
4459 struct xfs_bmbt_irec *mval, /* output: map values */
4460 int *nmap, /* i/o: mval size/count */
4461 struct xfs_bmap_free *flist) /* i/o: list extents to free */
4462{
4463 struct xfs_mount *mp = ip->i_mount;
4464 struct xfs_ifork *ifp;
a30b0367 4465 struct xfs_bmalloca bma = { NULL }; /* args for xfs_bmap_alloc */
9e5987a7
DC		 4466 xfs_fileoff_t end; /* end of mapped file region */
4467 int eof; /* after the end of extents */
4468 int error; /* error return */
4469 int n; /* current extent index */
4470 xfs_fileoff_t obno; /* old block number (offset) */
4471 int whichfork; /* data or attr fork */
4472 char inhole; /* current location is hole in file */
4473 char wasdelay; /* old extent was delayed */
4474
4475#ifdef DEBUG
4476 xfs_fileoff_t orig_bno; /* original block number value */
4477 int orig_flags; /* original flags arg value */
4478 xfs_filblks_t orig_len; /* original value of len arg */
4479 struct xfs_bmbt_irec *orig_mval; /* original value of mval */
4480 int orig_nmap; /* original value of *nmap */
4481
4482 orig_bno = bno;
4483 orig_len = len;
4484 orig_flags = flags;
4485 orig_mval = mval;
4486 orig_nmap = *nmap;
4487#endif
f3508bcd
DC		 4488 whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4489 XFS_ATTR_FORK : XFS_DATA_FORK;
9e5987a7
DC		 4490
4491 ASSERT(*nmap >= 1);
4492 ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4493 ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4494 ASSERT(tp != NULL);
4495 ASSERT(len > 0);
f3508bcd 4496 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL);
eef334e5 4497 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
9e5987a7
DC		 4498
4499 if (unlikely(XFS_TEST_ERROR(
4500 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
f3508bcd 4501 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
9e5987a7
DC		 4502 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4503 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
2451337d 4504 return -EFSCORRUPTED;
5575acc7
KD		 4505 }
4506
9e5987a7 4507 if (XFS_FORCED_SHUTDOWN(mp))
2451337d 4508 return -EIO;
9e5987a7
DC		 4509
4510 ifp = XFS_IFORK_PTR(ip, whichfork);
4511
4512 XFS_STATS_INC(xs_blk_mapw);
4513
9e5987a7
DC		 4514 if (*firstblock == NULLFSBLOCK) {
4515 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4516 bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4517 else
4518 bma.minleft = 1;
4519 } else {
4520 bma.minleft = 0;
4521 }
4522
4523 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4524 error = xfs_iread_extents(tp, ip, whichfork);
4525 if (error)
4526 goto error0;
4527 }
4528
4529 xfs_bmap_search_extents(ip, bno, whichfork, &eof, &bma.idx, &bma.got,
4530 &bma.prev);
4531 n = 0;
4532 end = bno + len;
4533 obno = bno;
4534
4535 bma.tp = tp;
4536 bma.ip = ip;
4537 bma.total = total;
4538 bma.userdata = 0;
4539 bma.flist = flist;
4540 bma.firstblock = firstblock;
4541
9e5987a7
DC		 4542 while (bno < end && n < *nmap) {
4543 inhole = eof || bma.got.br_startoff > bno;
4544 wasdelay = !inhole && isnullstartblock(bma.got.br_startblock);
4545
1da177e4 4546 /*
9e5987a7
DC		 4547 * First, deal with the hole before the allocated space
4548 * that we found, if any.
1da177e4 4549 */
9e5987a7
DC		 4550 if (inhole || wasdelay) {
4551 bma.eof = eof;
4552 bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4553 bma.wasdel = wasdelay;
4554 bma.offset = bno;
4555 bma.flags = flags;
4556
1da177e4 4557 /*
9e5987a7
DC		 4558 * There's a 32/64 bit type mismatch between the
4559 * allocation length request (which can be 64 bits in
4560 * length) and the bma length request, which is
4561 * xfs_extlen_t and therefore 32 bits. Hence we have to
4562 * check for 32-bit overflows and handle them here.
1da177e4 4563 */
9e5987a7
DC		 4564 if (len > (xfs_filblks_t)MAXEXTLEN)
4565 bma.length = MAXEXTLEN;
4566 else
4567 bma.length = len;
4568
4569 ASSERT(len > 0);
4570 ASSERT(bma.length > 0);
4571 error = xfs_bmapi_allocate(&bma);
1da177e4
LT		 4572 if (error)
4573 goto error0;
9e5987a7
DC		 4574 if (bma.blkno == NULLFSBLOCK)
4575 break;
1da177e4 4576 }
9e5987a7
DC		 4577
4578 /* Deal with the allocated space we found. */
4579 xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4580 end, n, flags);
4581
4582 /* Execute unwritten extent conversion if necessary */
4583 error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
2451337d 4584 if (error == -EAGAIN)
9e5987a7
DC		 4585 continue;
4586 if (error)
4587 goto error0;
4588
4589 /* update the extent map to return */
4590 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4591
4592 /*
4593 * If we're done, stop now. Stop when we've allocated
4594 * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise
4595 * the transaction may get too big.
4596 */
4597 if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4598 break;
4599
4600 /* Else go on to the next record. */
4601 bma.prev = bma.got;
4602 if (++bma.idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t)) {
4603 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma.idx),
4604 &bma.got);
4605 } else
4606 eof = 1;
4607 }
4608 *nmap = n;
4609
4610 /*
4611 * Transform from btree to extents, give it cur.
4612 */
4613 if (xfs_bmap_wants_extents(ip, whichfork)) {
4614 int tmp_logflags = 0;
4615
4616 ASSERT(bma.cur);
4617 error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
4618 &tmp_logflags, whichfork);
4619 bma.logflags |= tmp_logflags;
4620 if (error)
4621 goto error0;
4622 }
4623
4624 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4625 XFS_IFORK_NEXTENTS(ip, whichfork) >
4626 XFS_IFORK_MAXEXT(ip, whichfork));
4627 error = 0;
4628error0:
4629 /*
4630 * Log everything. Do this after conversion, there's no point in
4631 * logging the extent records if we've converted to btree format.
4632 */
4633 if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
4634 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4635 bma.logflags &= ~xfs_ilog_fext(whichfork);
4636 else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
4637 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4638 bma.logflags &= ~xfs_ilog_fbroot(whichfork);
4639 /*
4640 * Log whatever the flags say, even if error. Otherwise we might miss
4641 * detecting a case where the data is changed, there's an error,
4642 * and it's not logged so we don't shutdown when we should.
4643 */
4644 if (bma.logflags)
4645 xfs_trans_log_inode(tp, ip, bma.logflags);
4646
4647 if (bma.cur) {
4648 if (!error) {
4649 ASSERT(*firstblock == NULLFSBLOCK ||
4650 XFS_FSB_TO_AGNO(mp, *firstblock) ==
4651 XFS_FSB_TO_AGNO(mp,
4652 bma.cur->bc_private.b.firstblock) ||
4653 (flist->xbf_low &&
4654 XFS_FSB_TO_AGNO(mp, *firstblock) <
4655 XFS_FSB_TO_AGNO(mp,
4656 bma.cur->bc_private.b.firstblock)));
4657 *firstblock = bma.cur->bc_private.b.firstblock;
1da177e4 4658 }
9e5987a7
DC		 4659 xfs_btree_del_cursor(bma.cur,
4660 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4661 }
4662 if (!error)
4663 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4664 orig_nmap, *nmap);
4665 return error;
4666}
06d10dd9 4667
9e5987a7
DC		 4668/*
4669 * Called by xfs_bmapi to update file extent records and the btree
4670 * after removing space (or undoing a delayed allocation).
4671 */
4672STATIC int /* error */
4673xfs_bmap_del_extent(
4674 xfs_inode_t *ip, /* incore inode pointer */
4675 xfs_trans_t *tp, /* current transaction pointer */
4676 xfs_extnum_t *idx, /* extent number to update/delete */
4677 xfs_bmap_free_t *flist, /* list of extents to be freed */
4678 xfs_btree_cur_t *cur, /* if null, not a btree */
4679 xfs_bmbt_irec_t *del, /* data to remove from extents */
4680 int *logflagsp, /* inode logging flags */
4681 int whichfork) /* data or attr fork */
4682{
4683 xfs_filblks_t da_new; /* new delay-alloc indirect blocks */
4684 xfs_filblks_t da_old; /* old delay-alloc indirect blocks */
4685 xfs_fsblock_t del_endblock=0; /* first block past del */
4686 xfs_fileoff_t del_endoff; /* first offset past del */
4687 int delay; /* current block is delayed allocated */
4688 int do_fx; /* free extent at end of routine */
4689 xfs_bmbt_rec_host_t *ep; /* current extent entry pointer */
4690 int error; /* error return value */
4691 int flags; /* inode logging flags */
4692 xfs_bmbt_irec_t got; /* current extent entry */
4693 xfs_fileoff_t got_endoff; /* first offset past got */
4694 int i; /* temp state */
4695 xfs_ifork_t *ifp; /* inode fork pointer */
4696 xfs_mount_t *mp; /* mount structure */
4697 xfs_filblks_t nblks; /* quota/sb block count */
4698 xfs_bmbt_irec_t new; /* new record to be inserted */
4699 /* REFERENCED */
4700 uint qfield; /* quota field to update */
4701 xfs_filblks_t temp; /* for indirect length calculations */
4702 xfs_filblks_t temp2; /* for indirect length calculations */
4703 int state = 0;
4704
4705 XFS_STATS_INC(xs_del_exlist);
4706
4707 if (whichfork == XFS_ATTR_FORK)
4708 state |= BMAP_ATTRFORK;
4709
4710 mp = ip->i_mount;
4711 ifp = XFS_IFORK_PTR(ip, whichfork);
4712 ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
4713 (uint)sizeof(xfs_bmbt_rec_t)));
4714 ASSERT(del->br_blockcount > 0);
4715 ep = xfs_iext_get_ext(ifp, *idx);
4716 xfs_bmbt_get_all(ep, &got);
4717 ASSERT(got.br_startoff <= del->br_startoff);
4718 del_endoff = del->br_startoff + del->br_blockcount;
4719 got_endoff = got.br_startoff + got.br_blockcount;
4720 ASSERT(got_endoff >= del_endoff);
4721 delay = isnullstartblock(got.br_startblock);
4722 ASSERT(isnullstartblock(del->br_startblock) == delay);
4723 flags = 0;
4724 qfield = 0;
4725 error = 0;
4726 /*
4727 * If deleting a real allocation, must free up the disk space.
4728 */
4729 if (!delay) {
4730 flags = XFS_ILOG_CORE;
4731 /*
4732 * Realtime allocation. Free it and record di_nblocks update.
4733 */
4734 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
4735 xfs_fsblock_t bno;
4736 xfs_filblks_t len;
4737
4738 ASSERT(do_mod(del->br_blockcount,
4739 mp->m_sb.sb_rextsize) == 0);
4740 ASSERT(do_mod(del->br_startblock,
4741 mp->m_sb.sb_rextsize) == 0);
4742 bno = del->br_startblock;
4743 len = del->br_blockcount;
4744 do_div(bno, mp->m_sb.sb_rextsize);
4745 do_div(len, mp->m_sb.sb_rextsize);
4746 error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
4747 if (error)
4748 goto done;
4749 do_fx = 0;
4750 nblks = len * mp->m_sb.sb_rextsize;
4751 qfield = XFS_TRANS_DQ_RTBCOUNT;
4752 }
1da177e4 4753 /*
9e5987a7 4754 * Ordinary allocation.
1da177e4 4755 */
9e5987a7
DC		 4756 else {
4757 do_fx = 1;
4758 nblks = del->br_blockcount;
4759 qfield = XFS_TRANS_DQ_BCOUNT;
1da177e4 4760 }
1da177e4 4761 /*
9e5987a7 4762 * Set up del_endblock and cur for later.
1da177e4 4763 */
9e5987a7
DC		 4764 del_endblock = del->br_startblock + del->br_blockcount;
4765 if (cur) {
4766 if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
4767 got.br_startblock, got.br_blockcount,
4768 &i)))
4769 goto done;
c29aad41 4770 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1da177e4 4771 }
9e5987a7
DC		 4772 da_old = da_new = 0;
4773 } else {
4774 da_old = startblockval(got.br_startblock);
4775 da_new = 0;
4776 nblks = 0;
4777 do_fx = 0;
1da177e4
LT		 4778 }
4779 /*
9e5987a7
DC		 4780 * Set flag value to use in switch statement.
4781 * Left-contig is 2, right-contig is 1.
1da177e4 4782 */
9e5987a7
DC		 4783 switch (((got.br_startoff == del->br_startoff) << 1) |
4784 (got_endoff == del_endoff)) {
4785 case 3:
4786 /*
4787 * Matches the whole extent. Delete the entry.
4788 */
4789 xfs_iext_remove(ip, *idx, 1,
4790 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
4791 --*idx;
4792 if (delay)
4793 break;
4794
4795 XFS_IFORK_NEXT_SET(ip, whichfork,
4796 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
4797 flags |= XFS_ILOG_CORE;
4798 if (!cur) {
4799 flags |= xfs_ilog_fext(whichfork);
4800 break;
4801 }
4802 if ((error = xfs_btree_delete(cur, &i)))
4803 goto done;
c29aad41 4804 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 4805 break;
4806
4807 case 2:
4808 /*
4809 * Deleting the first part of the extent.
4810 */
4811 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4812 xfs_bmbt_set_startoff(ep, del_endoff);
4813 temp = got.br_blockcount - del->br_blockcount;
4814 xfs_bmbt_set_blockcount(ep, temp);
4815 if (delay) {
4816 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4817 da_old);
4818 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4819 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4820 da_new = temp;
4821 break;
4822 }
4823 xfs_bmbt_set_startblock(ep, del_endblock);
4824 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4825 if (!cur) {
4826 flags |= xfs_ilog_fext(whichfork);
4827 break;
4828 }
4829 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
4830 got.br_blockcount - del->br_blockcount,
4831 got.br_state)))
4832 goto done;
4833 break;
4834
4835 case 1:
4836 /*
4837 * Deleting the last part of the extent.
4838 */
4839 temp = got.br_blockcount - del->br_blockcount;
4840 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4841 xfs_bmbt_set_blockcount(ep, temp);
4842 if (delay) {
4843 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4844 da_old);
4845 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4846 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4847 da_new = temp;
4848 break;
4849 }
4850 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4851 if (!cur) {
4852 flags |= xfs_ilog_fext(whichfork);
4853 break;
4854 }
4855 if ((error = xfs_bmbt_update(cur, got.br_startoff,
4856 got.br_startblock,
4857 got.br_blockcount - del->br_blockcount,
4858 got.br_state)))
4859 goto done;
4860 break;
4861
4862 case 0:
4863 /*
4864 * Deleting the middle of the extent.
4865 */
4866 temp = del->br_startoff - got.br_startoff;
4867 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4868 xfs_bmbt_set_blockcount(ep, temp);
4869 new.br_startoff = del_endoff;
4870 temp2 = got_endoff - del_endoff;
4871 new.br_blockcount = temp2;
4872 new.br_state = got.br_state;
4873 if (!delay) {
4874 new.br_startblock = del_endblock;
4875 flags |= XFS_ILOG_CORE;
4876 if (cur) {
4877 if ((error = xfs_bmbt_update(cur,
4878 got.br_startoff,
4879 got.br_startblock, temp,
4880 got.br_state)))
4881 goto done;
4882 if ((error = xfs_btree_increment(cur, 0, &i)))
4883 goto done;
4884 cur->bc_rec.b = new;
4885 error = xfs_btree_insert(cur, &i);
2451337d 4886 if (error && error != -ENOSPC)
9e5987a7
DC		 4887 goto done;
4888 /*
4889 * If get no-space back from btree insert,
4890 * it tried a split, and we have a zero
4891 * block reservation.
4892 * Fix up our state and return the error.
4893 */
2451337d 4894 if (error == -ENOSPC) {
9e5987a7
DC		 4895 /*
4896 * Reset the cursor, don't trust
4897 * it after any insert operation.
4898 */
4899 if ((error = xfs_bmbt_lookup_eq(cur,
4900 got.br_startoff,
4901 got.br_startblock,
4902 temp, &i)))
4903 goto done;
c29aad41
ES		 4904 XFS_WANT_CORRUPTED_GOTO(mp,
4905 i == 1, done);
9e5987a7
DC		 4906 /*
4907 * Update the btree record back
4908 * to the original value.
4909 */
4910 if ((error = xfs_bmbt_update(cur,
4911 got.br_startoff,
4912 got.br_startblock,
4913 got.br_blockcount,
4914 got.br_state)))
4915 goto done;
4916 /*
4917 * Reset the extent record back
4918 * to the original value.
4919 */
4920 xfs_bmbt_set_blockcount(ep,
4921 got.br_blockcount);
4922 flags = 0;
2451337d 4923 error = -ENOSPC;
9e5987a7
DC		 4924 goto done;
4925 }
c29aad41 4926 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
9e5987a7
DC		 4927 } else
4928 flags |= xfs_ilog_fext(whichfork);
4929 XFS_IFORK_NEXT_SET(ip, whichfork,
4930 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
4931 } else {
4932 ASSERT(whichfork == XFS_DATA_FORK);
4933 temp = xfs_bmap_worst_indlen(ip, temp);
4934 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4935 temp2 = xfs_bmap_worst_indlen(ip, temp2);
4936 new.br_startblock = nullstartblock((int)temp2);
4937 da_new = temp + temp2;
4938 while (da_new > da_old) {
4939 if (temp) {
4940 temp--;
4941 da_new--;
4942 xfs_bmbt_set_startblock(ep,
4943 nullstartblock((int)temp));
4944 }
4945 if (da_new == da_old)
4946 break;
4947 if (temp2) {
4948 temp2--;
4949 da_new--;
4950 new.br_startblock =
4951 nullstartblock((int)temp2);
4952 }
4953 }
4954 }
4955 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4956 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
4957 ++*idx;
4958 break;
1da177e4
LT		 4959 }
4960 /*
9e5987a7 4961 * If we need to, add to list of extents to delete.
1da177e4 4962 */
9e5987a7
DC		 4963 if (do_fx)
4964 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
4965 mp);
1da177e4 4966 /*
9e5987a7 4967 * Adjust inode # blocks in the file.
1da177e4 4968 */
9e5987a7
DC		 4969 if (nblks)
4970 ip->i_d.di_nblocks -= nblks;
1da177e4 4971 /*
9e5987a7 4972 * Adjust quota data.
1da177e4 4973 */
9e5987a7
DC		 4974 if (qfield)
4975 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
4976
4977 /*
4978 * Account for change in delayed indirect blocks.
4979 * Nothing to do for disk quota accounting here.
4980 */
4981 ASSERT(da_old >= da_new);
0d485ada
DC		 4982 if (da_old > da_new)
4983 xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new), false);
9e5987a7
DC		 4984done:
4985 *logflagsp = flags;
1da177e4
LT		 4986 return error;
4987}
4988
3bacbcd8 4989/*
9e5987a7
DC		 4990 * Unmap (remove) blocks from a file.
4991 * If nexts is nonzero then the number of extents to remove is limited to
4992 * that value. If not all extents in the block range can be removed then
4993 * *done is set.
3bacbcd8 4994 */
9e5987a7
DC		 4995int /* error */
4996xfs_bunmapi(
4997 xfs_trans_t *tp, /* transaction pointer */
4998 struct xfs_inode *ip, /* incore inode */
4999 xfs_fileoff_t bno, /* starting offset to unmap */
5000 xfs_filblks_t len, /* length to unmap in file */
5001 int flags, /* misc flags */
5002 xfs_extnum_t nexts, /* number of extents max */
5003 xfs_fsblock_t *firstblock, /* first allocated block
5004 controls a.g. for allocs */
5005 xfs_bmap_free_t *flist, /* i/o: list extents to free */
5006 int *done) /* set if not done yet */
3bacbcd8 5007{
9e5987a7
DC		 5008 xfs_btree_cur_t *cur; /* bmap btree cursor */
5009 xfs_bmbt_irec_t del; /* extent being deleted */
5010 int eof; /* is deleting at eof */
5011 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
5012 int error; /* error return value */
5013 xfs_extnum_t extno; /* extent number in list */
5014 xfs_bmbt_irec_t got; /* current extent record */
5af317c9 5015 xfs_ifork_t *ifp; /* inode fork pointer */
9e5987a7
DC		 5016 int isrt; /* freeing in rt area */
5017 xfs_extnum_t lastx; /* last extent index used */
5018 int logflags; /* transaction logging flags */
5019 xfs_extlen_t mod; /* rt extent offset */
5020 xfs_mount_t *mp; /* mount structure */
5021 xfs_extnum_t nextents; /* number of file extents */
5022 xfs_bmbt_irec_t prev; /* previous extent record */
5023 xfs_fileoff_t start; /* first file offset deleted */
5024 int tmp_logflags; /* partial logging flags */
5025 int wasdel; /* was a delayed alloc extent */
5026 int whichfork; /* data or attribute fork */
5027 xfs_fsblock_t sum;
1da177e4 5028
9e5987a7 5029 trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
1da177e4 5030
9e5987a7
DC		 5031 whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5032 XFS_ATTR_FORK : XFS_DATA_FORK;
5033 ifp = XFS_IFORK_PTR(ip, whichfork);
5034 if (unlikely(
5035 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5036 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5037 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5038 ip->i_mount);
2451337d 5039 return -EFSCORRUPTED;
1da177e4 5040 }
9e5987a7
DC		 5041 mp = ip->i_mount;
5042 if (XFS_FORCED_SHUTDOWN(mp))
2451337d 5043 return -EIO;
1da177e4 5044
eef334e5 5045 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
9e5987a7
DC		 5046 ASSERT(len > 0);
5047 ASSERT(nexts >= 0);
1da177e4 5048
9e5987a7
DC		 5049 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5050 (error = xfs_iread_extents(tp, ip, whichfork)))
5051 return error;
5052 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5053 if (nextents == 0) {
5054 *done = 1;
5055 return 0;
5056 }
5057 XFS_STATS_INC(xs_blk_unmap);
5058 isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5059 start = bno;
5060 bno = start + len - 1;
5061 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5062 &prev);
1da177e4 5063
9e5987a7
DC		 5064 /*
5065 * Check to see if the given block number is past the end of the
5066 * file, back up to the last block if so...
5067 */
5068 if (eof) {
5069 ep = xfs_iext_get_ext(ifp, --lastx);
5070 xfs_bmbt_get_all(ep, &got);
5071 bno = got.br_startoff + got.br_blockcount - 1;
5072 }
5073 logflags = 0;
5074 if (ifp->if_flags & XFS_IFBROOT) {
5075 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5076 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5077 cur->bc_private.b.firstblock = *firstblock;
5078 cur->bc_private.b.flist = flist;
5079 cur->bc_private.b.flags = 0;
5080 } else
5081 cur = NULL;
5082
5083 if (isrt) {
5084 /*
5085 * Synchronize by locking the bitmap inode.
5086 */
5087 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
5088 xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5089 }
58e20770 5090
9e5987a7
DC		 5091 extno = 0;
5092 while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5093 (nexts == 0 || extno < nexts)) {
5094 /*
5095 * Is the found extent after a hole in which bno lives?
5096 * Just back up to the previous extent, if so.
5097 */
5098 if (got.br_startoff > bno) {
5099 if (--lastx < 0)
5100 break;
5101 ep = xfs_iext_get_ext(ifp, lastx);
5102 xfs_bmbt_get_all(ep, &got);
5103 }
5104 /*
5105 * Is the last block of this extent before the range
5106 * we're supposed to delete? If so, we're done.
5107 */
5108 bno = XFS_FILEOFF_MIN(bno,
5109 got.br_startoff + got.br_blockcount - 1);
5110 if (bno < start)
5111 break;
5112 /*
5113 * Then deal with the (possibly delayed) allocated space
5114 * we found.
5115 */
5116 ASSERT(ep != NULL);
5117 del = got;
5118 wasdel = isnullstartblock(del.br_startblock);
5119 if (got.br_startoff < start) {
5120 del.br_startoff = start;
5121 del.br_blockcount -= start - got.br_startoff;
5122 if (!wasdel)
5123 del.br_startblock += start - got.br_startoff;
5124 }
5125 if (del.br_startoff + del.br_blockcount > bno + 1)
5126 del.br_blockcount = bno + 1 - del.br_startoff;
5127 sum = del.br_startblock + del.br_blockcount;
5128 if (isrt &&
5129 (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
58e20770 5130 /*
9e5987a7
DC		 5131 * Realtime extent not lined up at the end.
5132 * The extent could have been split into written
5133 * and unwritten pieces, or we could just be
5134 * unmapping part of it. But we can't really
5135 * get rid of part of a realtime extent.
58e20770 5136 */
9e5987a7
DC		 5137 if (del.br_state == XFS_EXT_UNWRITTEN ||
5138 !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5139 /*
5140 * This piece is unwritten, or we're not
5141 * using unwritten extents. Skip over it.
5142 */
5143 ASSERT(bno >= mod);
5144 bno -= mod > del.br_blockcount ?
5145 del.br_blockcount : mod;
5146 if (bno < got.br_startoff) {
5147 if (--lastx >= 0)
5148 xfs_bmbt_get_all(xfs_iext_get_ext(
5149 ifp, lastx), &got);
5150 }
5151 continue;
1da177e4 5152 }
9af25465 5153 /*
9e5987a7
DC		 5154 * It's written, turn it unwritten.
5155 * This is better than zeroing it.
9af25465 5156 */
9e5987a7
DC		 5157 ASSERT(del.br_state == XFS_EXT_NORM);
5158 ASSERT(xfs_trans_get_block_res(tp) > 0);
5159 /*
5160 * If this spans a realtime extent boundary,
5161 * chop it back to the start of the one we end at.
5162 */
5163 if (del.br_blockcount > mod) {
5164 del.br_startoff += del.br_blockcount - mod;
5165 del.br_startblock += del.br_blockcount - mod;
5166 del.br_blockcount = mod;
5167 }
5168 del.br_state = XFS_EXT_UNWRITTEN;
5169 error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5170 &lastx, &cur, &del, firstblock, flist,
5171 &logflags);
5172 if (error)
5173 goto error0;
5174 goto nodelete;
5175 }
5176 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5177 /*
5178 * Realtime extent is lined up at the end but not
5179 * at the front. We'll get rid of full extents if
5180 * we can.
5181 */
5182 mod = mp->m_sb.sb_rextsize - mod;
5183 if (del.br_blockcount > mod) {
5184 del.br_blockcount -= mod;
5185 del.br_startoff += mod;
5186 del.br_startblock += mod;
5187 } else if ((del.br_startoff == start &&
5188 (del.br_state == XFS_EXT_UNWRITTEN ||
5189 xfs_trans_get_block_res(tp) == 0)) ||
5190 !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5191 /*
5192 * Can't make it unwritten. There isn't
5193 * a full extent here so just skip it.
5194 */
5195 ASSERT(bno >= del.br_blockcount);
5196 bno -= del.br_blockcount;
5197 if (got.br_startoff > bno) {
5198 if (--lastx >= 0) {
5199 ep = xfs_iext_get_ext(ifp,
5200 lastx);
5201 xfs_bmbt_get_all(ep, &got);
5202 }
5203 }
9af25465 5204 continue;
9e5987a7
DC		 5205 } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5206 /*
5207 * This one is already unwritten.
5208 * It must have a written left neighbor.
5209 * Unwrite the killed part of that one and
5210 * try again.
5211 */
5212 ASSERT(lastx > 0);
5213 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5214 lastx - 1), &prev);
5215 ASSERT(prev.br_state == XFS_EXT_NORM);
5216 ASSERT(!isnullstartblock(prev.br_startblock));
5217 ASSERT(del.br_startblock ==
5218 prev.br_startblock + prev.br_blockcount);
5219 if (prev.br_startoff < start) {
5220 mod = start - prev.br_startoff;
5221 prev.br_blockcount -= mod;
5222 prev.br_startblock += mod;
5223 prev.br_startoff = start;
5224 }
5225 prev.br_state = XFS_EXT_UNWRITTEN;
5226 lastx--;
5227 error = xfs_bmap_add_extent_unwritten_real(tp,
5228 ip, &lastx, &cur, &prev,
5229 firstblock, flist, &logflags);
5230 if (error)
5231 goto error0;
5232 goto nodelete;
5233 } else {
5234 ASSERT(del.br_state == XFS_EXT_NORM);
5235 del.br_state = XFS_EXT_UNWRITTEN;
5236 error = xfs_bmap_add_extent_unwritten_real(tp,
5237 ip, &lastx, &cur, &del,
5238 firstblock, flist, &logflags);
5239 if (error)
5240 goto error0;
5241 goto nodelete;
9af25465 5242 }
1da177e4 5243 }
9e5987a7
DC		 5244 if (wasdel) {
5245 ASSERT(startblockval(del.br_startblock) > 0);
5246 /* Update realtime/data freespace, unreserve quota */
5247 if (isrt) {
5248 xfs_filblks_t rtexts;
1da177e4 5249
9e5987a7
DC		 5250 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5251 do_div(rtexts, mp->m_sb.sb_rextsize);
bab98bbe 5252 xfs_mod_frextents(mp, (int64_t)rtexts);
9e5987a7
DC		 5253 (void)xfs_trans_reserve_quota_nblks(NULL,
5254 ip, -((long)del.br_blockcount), 0,
5255 XFS_QMOPT_RES_RTBLKS);
5256 } else {
0d485ada
DC		 5257 xfs_mod_fdblocks(mp, (int64_t)del.br_blockcount,
5258 false);
9e5987a7
DC		 5259 (void)xfs_trans_reserve_quota_nblks(NULL,
5260 ip, -((long)del.br_blockcount), 0,
5261 XFS_QMOPT_RES_REGBLKS);
5262 }
5263 ip->i_delayed_blks -= del.br_blockcount;
5264 if (cur)
5265 cur->bc_private.b.flags |=
5266 XFS_BTCUR_BPRV_WASDEL;
5267 } else if (cur)
5268 cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5269 /*
5270 * If it's the case where the directory code is running
5271 * with no block reservation, and the deleted block is in
5272 * the middle of its extent, and the resulting insert
5273 * of an extent would cause transformation to btree format,
5274 * then reject it. The calling code will then swap
5275 * blocks around instead.
5276 * We have to do this now, rather than waiting for the
5277 * conversion to btree format, since the transaction
5278 * will be dirty.
5279 */
5280 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5281 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5282 XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
5283 XFS_IFORK_MAXEXT(ip, whichfork) &&
5284 del.br_startoff > got.br_startoff &&
5285 del.br_startoff + del.br_blockcount <
5286 got.br_startoff + got.br_blockcount) {
2451337d 5287 error = -ENOSPC;
9e5987a7 5288 goto error0;
1da177e4 5289 }
9e5987a7
DC		 5290 error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5291 &tmp_logflags, whichfork);
5292 logflags |= tmp_logflags;
5293 if (error)
5294 goto error0;
5295 bno = del.br_startoff - 1;
5296nodelete:
1da177e4 5297 /*
9e5987a7 5298 * If not done go on to the next (previous) record.
1da177e4 5299 */
9e5987a7
DC		 5300 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5301 if (lastx >= 0) {
5302 ep = xfs_iext_get_ext(ifp, lastx);
5303 if (xfs_bmbt_get_startoff(ep) > bno) {
5304 if (--lastx >= 0)
5305 ep = xfs_iext_get_ext(ifp,
5306 lastx);
5307 }
5308 xfs_bmbt_get_all(ep, &got);
1da177e4 5309 }
9e5987a7 5310 extno++;
1da177e4
LT		 5311 }
5312 }
9e5987a7 5313 *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
576039cf 5314
1da177e4 5315 /*
9e5987a7 5316 * Convert to a btree if necessary.
1da177e4 5317 */
9e5987a7
DC		 5318 if (xfs_bmap_needs_btree(ip, whichfork)) {
5319 ASSERT(cur == NULL);
5320 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5321 &cur, 0, &tmp_logflags, whichfork);
5322 logflags |= tmp_logflags;
5323 if (error)
5324 goto error0;
1da177e4 5325 }
1da177e4 5326 /*
9e5987a7 5327 * transform from btree to extents, give it cur
1da177e4 5328 */
9e5987a7
DC		 5329 else if (xfs_bmap_wants_extents(ip, whichfork)) {
5330 ASSERT(cur != NULL);
5331 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5332 whichfork);
5333 logflags |= tmp_logflags;
5334 if (error)
5335 goto error0;
5336 }
1da177e4 5337 /*
9e5987a7 5338 * transform from extents to local?
1da177e4 5339 */
9e5987a7
DC		 5340 error = 0;
5341error0:
5342 /*
5343 * Log everything. Do this after conversion, there's no point in
5344 * logging the extent records if we've converted to btree format.
5345 */
5346 if ((logflags & xfs_ilog_fext(whichfork)) &&
5347 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5348 logflags &= ~xfs_ilog_fext(whichfork);
5349 else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5350 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5351 logflags &= ~xfs_ilog_fbroot(whichfork);
5352 /*
5353 * Log inode even in the error case, if the transaction
5354 * is dirty we'll need to shut down the filesystem.
5355 */
5356 if (logflags)
5357 xfs_trans_log_inode(tp, ip, logflags);
5358 if (cur) {
5359 if (!error) {
5360 *firstblock = cur->bc_private.b.firstblock;
5361 cur->bc_private.b.allocated = 0;
5362 }
5363 xfs_btree_del_cursor(cur,
5364 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5365 }
5366 return error;
5367}
e1d8fb88 5368
ddb19e31
BF		 5369/*
5370 * Determine whether an extent shift can be accomplished by a merge with the
5371 * extent that precedes the target hole of the shift.
5372 */
5373STATIC bool
5374xfs_bmse_can_merge(
5375 struct xfs_bmbt_irec *left, /* preceding extent */
5376 struct xfs_bmbt_irec *got, /* current extent to shift */
5377 xfs_fileoff_t shift) /* shift fsb */
5378{
5379 xfs_fileoff_t startoff;
5380
5381 startoff = got->br_startoff - shift;
5382
5383 /*
5384 * The extent, once shifted, must be adjacent in-file and on-disk with
5385 * the preceding extent.
5386 */
5387 if ((left->br_startoff + left->br_blockcount != startoff) ||
5388 (left->br_startblock + left->br_blockcount != got->br_startblock) ||
5389 (left->br_state != got->br_state) ||
5390 (left->br_blockcount + got->br_blockcount > MAXEXTLEN))
5391 return false;
5392
5393 return true;
5394}
5395
5396/*
5397 * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5398 * hole in the file. If an extent shift would result in the extent being fully
5399 * adjacent to the extent that currently precedes the hole, we can merge with
5400 * the preceding extent rather than do the shift.
5401 *
5402 * This function assumes the caller has verified a shift-by-merge is possible
5403 * with the provided extents via xfs_bmse_can_merge().
5404 */
5405STATIC int
5406xfs_bmse_merge(
5407 struct xfs_inode *ip,
5408 int whichfork,
5409 xfs_fileoff_t shift, /* shift fsb */
5410 int current_ext, /* idx of gotp */
5411 struct xfs_bmbt_rec_host *gotp, /* extent to shift */
5412 struct xfs_bmbt_rec_host *leftp, /* preceding extent */
5413 struct xfs_btree_cur *cur,
5414 int *logflags) /* output */
5415{
ddb19e31
BF		 5416 struct xfs_bmbt_irec got;
5417 struct xfs_bmbt_irec left;
5418 xfs_filblks_t blockcount;
5419 int error, i;
5fb5aeee 5420 struct xfs_mount *mp = ip->i_mount;
ddb19e31 5421
ddb19e31
BF		 5422 xfs_bmbt_get_all(gotp, &got);
5423 xfs_bmbt_get_all(leftp, &left);
5424 blockcount = left.br_blockcount + got.br_blockcount;
5425
5426 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5427 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5428 ASSERT(xfs_bmse_can_merge(&left, &got, shift));
5429
5430 /*
5431 * Merge the in-core extents. Note that the host record pointers and
5432 * current_ext index are invalid once the extent has been removed via
5433 * xfs_iext_remove().
5434 */
5435 xfs_bmbt_set_blockcount(leftp, blockcount);
5436 xfs_iext_remove(ip, current_ext, 1, 0);
5437
5438 /*
5439 * Update the on-disk extent count, the btree if necessary and log the
5440 * inode.
5441 */
5442 XFS_IFORK_NEXT_SET(ip, whichfork,
5443 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5444 *logflags |= XFS_ILOG_CORE;
5445 if (!cur) {
5446 *logflags |= XFS_ILOG_DEXT;
5447 return 0;
5448 }
5449
5450 /* lookup and remove the extent to merge */
5451 error = xfs_bmbt_lookup_eq(cur, got.br_startoff, got.br_startblock,
5452 got.br_blockcount, &i);
5453 if (error)
4db431f5 5454 return error;
5fb5aeee 5455 XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
ddb19e31
BF		 5456
5457 error = xfs_btree_delete(cur, &i);
5458 if (error)
4db431f5 5459 return error;
5fb5aeee 5460 XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
ddb19e31
BF		 5461
5462 /* lookup and update size of the previous extent */
5463 error = xfs_bmbt_lookup_eq(cur, left.br_startoff, left.br_startblock,
5464 left.br_blockcount, &i);
5465 if (error)
4db431f5 5466 return error;
5fb5aeee 5467 XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
ddb19e31
BF		 5468
5469 left.br_blockcount = blockcount;
5470
4db431f5
DC		 5471 return xfs_bmbt_update(cur, left.br_startoff, left.br_startblock,
5472 left.br_blockcount, left.br_state);
ddb19e31
BF		 5473}
5474
a979bdfe
BF		 5475/*
5476 * Shift a single extent.
5477 */
5478STATIC int
5479xfs_bmse_shift_one(
5480 struct xfs_inode *ip,
5481 int whichfork,
5482 xfs_fileoff_t offset_shift_fsb,
5483 int *current_ext,
5484 struct xfs_bmbt_rec_host *gotp,
5485 struct xfs_btree_cur *cur,
a904b1ca
NJ		 5486 int *logflags,
5487 enum shift_direction direction)
a979bdfe
BF		 5488{
5489 struct xfs_ifork *ifp;
5fb5aeee 5490 struct xfs_mount *mp;
a979bdfe 5491 xfs_fileoff_t startoff;
a904b1ca 5492 struct xfs_bmbt_rec_host *adj_irecp;
a979bdfe 5493 struct xfs_bmbt_irec got;
a904b1ca 5494 struct xfs_bmbt_irec adj_irec;
a979bdfe
BF		 5495 int error;
5496 int i;
a904b1ca 5497 int total_extents;
a979bdfe 5498
5fb5aeee 5499 mp = ip->i_mount;
a979bdfe 5500 ifp = XFS_IFORK_PTR(ip, whichfork);
a904b1ca 5501 total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
a979bdfe
BF		 5502
5503 xfs_bmbt_get_all(gotp, &got);
a979bdfe 5504
f71721d0 5505 /* delalloc extents should be prevented by caller */
5fb5aeee 5506 XFS_WANT_CORRUPTED_RETURN(mp, !isnullstartblock(got.br_startblock));
f71721d0 5507
a904b1ca
NJ		 5508 if (direction == SHIFT_LEFT) {
5509 startoff = got.br_startoff - offset_shift_fsb;
5510
5511 /*
5512 * Check for merge if we've got an extent to the left,
5513 * otherwise make sure there's enough room at the start
5514 * of the file for the shift.
5515 */
5516 if (!*current_ext) {
5517 if (got.br_startoff < offset_shift_fsb)
5518 return -EINVAL;
5519 goto update_current_ext;
5520 }
5521 /*
5522 * grab the left extent and check for a large
5523 * enough hole.
5524 */
5525 adj_irecp = xfs_iext_get_ext(ifp, *current_ext - 1);
5526 xfs_bmbt_get_all(adj_irecp, &adj_irec);
a979bdfe 5527
a904b1ca
NJ		 5528 if (startoff <
5529 adj_irec.br_startoff + adj_irec.br_blockcount)
b11bd671 5530 return -EINVAL;
a979bdfe 5531
b11bd671 5532 /* check whether to merge the extent or shift it down */
a904b1ca
NJ		 5533 if (xfs_bmse_can_merge(&adj_irec, &got,
5534 offset_shift_fsb)) {
b11bd671 5535 return xfs_bmse_merge(ip, whichfork, offset_shift_fsb,
a904b1ca
NJ		 5536 *current_ext, gotp, adj_irecp,
5537 cur, logflags);
b11bd671 5538 }
a904b1ca
NJ		 5539 } else {
5540 startoff = got.br_startoff + offset_shift_fsb;
5541 /* nothing to move if this is the last extent */
5542 if (*current_ext >= (total_extents - 1))
5543 goto update_current_ext;
5544 /*
5545 * If this is not the last extent in the file, make sure there
5546 * is enough room between current extent and next extent for
5547 * accommodating the shift.
5548 */
5549 adj_irecp = xfs_iext_get_ext(ifp, *current_ext + 1);
5550 xfs_bmbt_get_all(adj_irecp, &adj_irec);
5551 if (startoff + got.br_blockcount > adj_irec.br_startoff)
5552 return -EINVAL;
5553 /*
5554 * Unlike a left shift (which involves a hole punch),
5555 * a right shift does not modify extent neighbors
5556 * in any way. We should never find mergeable extents
5557 * in this scenario. Check anyways and warn if we
5558 * encounter two extents that could be one.
5559 */
5560 if (xfs_bmse_can_merge(&got, &adj_irec, offset_shift_fsb))
5561 WARN_ON_ONCE(1);
5562 }
a979bdfe
BF		 5563 /*
5564 * Increment the extent index for the next iteration, update the start
5565 * offset of the in-core extent and update the btree if applicable.
5566 */
a904b1ca
NJ		 5567update_current_ext:
5568 if (direction == SHIFT_LEFT)
5569 (*current_ext)++;
5570 else
5571 (*current_ext)--;
a979bdfe
BF		 5572 xfs_bmbt_set_startoff(gotp, startoff);
5573 *logflags |= XFS_ILOG_CORE;
5574 if (!cur) {
5575 *logflags |= XFS_ILOG_DEXT;
5576 return 0;
5577 }
5578
5579 error = xfs_bmbt_lookup_eq(cur, got.br_startoff, got.br_startblock,
5580 got.br_blockcount, &i);
5581 if (error)
5582 return error;
5fb5aeee 5583 XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
a979bdfe
BF		 5584
5585 got.br_startoff = startoff;
d254aaec 5586 return xfs_bmbt_update(cur, got.br_startoff, got.br_startblock,
a904b1ca 5587 got.br_blockcount, got.br_state);
a979bdfe
BF		 5588}
5589
e1d8fb88 5590/*
a904b1ca 5591 * Shift extent records to the left/right to cover/create a hole.
e1d8fb88 5592 *
2c845f5a 5593 * The maximum number of extents to be shifted in a single operation is
a904b1ca 5594 * @num_exts. @stop_fsb specifies the file offset at which to stop shift and the
2c845f5a
BF		 5595 * file offset where we've left off is returned in @next_fsb. @offset_shift_fsb
5596 * is the length by which each extent is shifted. If there is no hole to shift
5597 * the extents into, this will be considered invalid operation and we abort
5598 * immediately.
e1d8fb88
NJ		 5599 */
5600int
5601xfs_bmap_shift_extents(
5602 struct xfs_trans *tp,
5603 struct xfs_inode *ip,
a904b1ca 5604 xfs_fileoff_t *next_fsb,
e1d8fb88 5605 xfs_fileoff_t offset_shift_fsb,
2c845f5a 5606 int *done,
a904b1ca 5607 xfs_fileoff_t stop_fsb,
e1d8fb88
NJ		 5608 xfs_fsblock_t *firstblock,
5609 struct xfs_bmap_free *flist,
a904b1ca 5610 enum shift_direction direction,
e1d8fb88
NJ		 5611 int num_exts)
5612{
ca446d88 5613 struct xfs_btree_cur *cur = NULL;
e1d8fb88
NJ		 5614 struct xfs_bmbt_rec_host *gotp;
5615 struct xfs_bmbt_irec got;
e1d8fb88
NJ		 5616 struct xfs_mount *mp = ip->i_mount;
5617 struct xfs_ifork *ifp;
5618 xfs_extnum_t nexts = 0;
2c845f5a 5619 xfs_extnum_t current_ext;
a904b1ca
NJ		 5620 xfs_extnum_t total_extents;
5621 xfs_extnum_t stop_extent;
e1d8fb88 5622 int error = 0;
e1d8fb88 5623 int whichfork = XFS_DATA_FORK;
ca446d88 5624 int logflags = 0;
e1d8fb88
NJ		 5625
5626 if (unlikely(XFS_TEST_ERROR(
5627 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5628 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
5629 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
5630 XFS_ERROR_REPORT("xfs_bmap_shift_extents",
5631 XFS_ERRLEVEL_LOW, mp);
2451337d 5632 return -EFSCORRUPTED;
e1d8fb88
NJ		 5633 }
5634
5635 if (XFS_FORCED_SHUTDOWN(mp))
2451337d 5636 return -EIO;
e1d8fb88 5637
2c845f5a
BF		 5638 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5639 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
a904b1ca
NJ		 5640 ASSERT(direction == SHIFT_LEFT || direction == SHIFT_RIGHT);
5641 ASSERT(*next_fsb != NULLFSBLOCK || direction == SHIFT_RIGHT);
e1d8fb88
NJ		 5642
5643 ifp = XFS_IFORK_PTR(ip, whichfork);
e1d8fb88
NJ		 5644 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5645 /* Read in all the extents */
5646 error = xfs_iread_extents(tp, ip, whichfork);
5647 if (error)
5648 return error;
5649 }
5650
ddb19e31
BF		 5651 if (ifp->if_flags & XFS_IFBROOT) {
5652 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5653 cur->bc_private.b.firstblock = *firstblock;
5654 cur->bc_private.b.flist = flist;
5655 cur->bc_private.b.flags = 0;
5656 }
5657
a904b1ca
NJ		 5658 /*
5659 * There may be delalloc extents in the data fork before the range we
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.
5662 */
5663 total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
5664 if (total_extents == 0) {
5665 *done = 1;
5666 goto del_cursor;
5667 }
5668
5669 /*
5670 * In case of first right shift, we need to initialize next_fsb
5671 */
5672 if (*next_fsb == NULLFSBLOCK) {
5673 gotp = xfs_iext_get_ext(ifp, total_extents - 1);
5674 xfs_bmbt_get_all(gotp, &got);
5675 *next_fsb = got.br_startoff;
5676 if (stop_fsb > *next_fsb) {
5677 *done = 1;
5678 goto del_cursor;
5679 }
5680 }
5681
5682 /* Lookup the extent index at which we have to stop */
5683 if (direction == SHIFT_RIGHT) {
5684 gotp = xfs_iext_bno_to_ext(ifp, stop_fsb, &stop_extent);
5685 /* Make stop_extent exclusive of shift range */
5686 stop_extent--;
5687 } else
5688 stop_extent = total_extents;
5689
e1d8fb88 5690 /*
2c845f5a
BF		 5691 * Look up the extent index for the fsb where we start shifting. We can
5692 * henceforth iterate with current_ext as extent list changes are locked
5693 * out via ilock.
5694 *
5695 * gotp can be null in 2 cases: 1) if there are no extents or 2)
a904b1ca 5696 * *next_fsb lies in a hole beyond which there are no extents. Either
2c845f5a 5697 * way, we are done.
e1d8fb88 5698 */
a904b1ca 5699 gotp = xfs_iext_bno_to_ext(ifp, *next_fsb, &current_ext);
2c845f5a
BF		 5700 if (!gotp) {
5701 *done = 1;
ddb19e31 5702 goto del_cursor;
e1d8fb88 5703 }
e1d8fb88 5704
a904b1ca
NJ		 5705 /* some sanity checking before we finally start shifting extents */
5706 if ((direction == SHIFT_LEFT && current_ext >= stop_extent) ||
5707 (direction == SHIFT_RIGHT && current_ext <= stop_extent)) {
5708 error = -EIO;
5709 goto del_cursor;
5710 }
5711
5712 while (nexts++ < num_exts) {
a979bdfe 5713 error = xfs_bmse_shift_one(ip, whichfork, offset_shift_fsb,
a904b1ca
NJ		 5714 &current_ext, gotp, cur, &logflags,
5715 direction);
a979bdfe
BF		 5716 if (error)
5717 goto del_cursor;
a904b1ca
NJ		 5718 /*
5719 * If there was an extent merge during the shift, the extent
5720 * count can change. Update the total and grade the next record.
5721 */
5722 if (direction == SHIFT_LEFT) {
5723 total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
5724 stop_extent = total_extents;
5725 }
e1d8fb88 5726
a904b1ca
NJ		 5727 if (current_ext == stop_extent) {
5728 *done = 1;
5729 *next_fsb = NULLFSBLOCK;
ddb19e31 5730 break;
a904b1ca 5731 }
ddb19e31 5732 gotp = xfs_iext_get_ext(ifp, current_ext);
e1d8fb88
NJ		 5733 }
5734
a904b1ca 5735 if (!*done) {
a979bdfe 5736 xfs_bmbt_get_all(gotp, &got);
2c845f5a 5737 *next_fsb = got.br_startoff;
a979bdfe 5738 }
e1d8fb88
NJ		 5739
5740del_cursor:
5741 if (cur)
5742 xfs_btree_del_cursor(cur,
5743 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5744
ca446d88
BF		 5745 if (logflags)
5746 xfs_trans_log_inode(tp, ip, logflags);
2c845f5a 5747
e1d8fb88
NJ		 5748 return error;
5749}
a904b1ca
NJ		 5750
5751/*
5752 * Splits an extent into two extents at split_fsb block such that it is
5753 * the first block of the current_ext. @current_ext is a target extent
5754 * to be split. @split_fsb is a block where the extents is split.
5755 * If split_fsb lies in a hole or the first block of extents, just return 0.
5756 */
5757STATIC int
5758xfs_bmap_split_extent_at(
5759 struct xfs_trans *tp,
5760 struct xfs_inode *ip,
5761 xfs_fileoff_t split_fsb,
5762 xfs_fsblock_t *firstfsb,
5763 struct xfs_bmap_free *free_list)
5764{
5765 int whichfork = XFS_DATA_FORK;
5766 struct xfs_btree_cur *cur = NULL;
5767 struct xfs_bmbt_rec_host *gotp;
5768 struct xfs_bmbt_irec got;
5769 struct xfs_bmbt_irec new; /* split extent */
5770 struct xfs_mount *mp = ip->i_mount;
5771 struct xfs_ifork *ifp;
5772 xfs_fsblock_t gotblkcnt; /* new block count for got */
5773 xfs_extnum_t current_ext;
5774 int error = 0;
5775 int logflags = 0;
5776 int i = 0;
5777
5778 if (unlikely(XFS_TEST_ERROR(
5779 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5780 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
5781 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
5782 XFS_ERROR_REPORT("xfs_bmap_split_extent_at",
5783 XFS_ERRLEVEL_LOW, mp);
5784 return -EFSCORRUPTED;
5785 }
5786
5787 if (XFS_FORCED_SHUTDOWN(mp))
5788 return -EIO;
5789
5790 ifp = XFS_IFORK_PTR(ip, whichfork);
5791 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5792 /* Read in all the extents */
5793 error = xfs_iread_extents(tp, ip, whichfork);
5794 if (error)
5795 return error;
5796 }
5797
5798 /*
5799 * gotp can be null in 2 cases: 1) if there are no extents
5800 * or 2) split_fsb lies in a hole beyond which there are
5801 * no extents. Either way, we are done.
5802 */
5803 gotp = xfs_iext_bno_to_ext(ifp, split_fsb, &current_ext);
5804 if (!gotp)
5805 return 0;
5806
5807 xfs_bmbt_get_all(gotp, &got);
5808
5809 /*
5810 * Check split_fsb lies in a hole or the start boundary offset
5811 * of the extent.
5812 */
5813 if (got.br_startoff >= split_fsb)
5814 return 0;
5815
5816 gotblkcnt = split_fsb - got.br_startoff;
5817 new.br_startoff = split_fsb;
5818 new.br_startblock = got.br_startblock + gotblkcnt;
5819 new.br_blockcount = got.br_blockcount - gotblkcnt;
5820 new.br_state = got.br_state;
5821
5822 if (ifp->if_flags & XFS_IFBROOT) {
5823 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5824 cur->bc_private.b.firstblock = *firstfsb;
5825 cur->bc_private.b.flist = free_list;
5826 cur->bc_private.b.flags = 0;
5827 error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
5828 got.br_startblock,
5829 got.br_blockcount,
5830 &i);
5831 if (error)
5832 goto del_cursor;
5833 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
5834 }
5835
5836 xfs_bmbt_set_blockcount(gotp, gotblkcnt);
5837 got.br_blockcount = gotblkcnt;
5838
5839 logflags = XFS_ILOG_CORE;
5840 if (cur) {
5841 error = xfs_bmbt_update(cur, got.br_startoff,
5842 got.br_startblock,
5843 got.br_blockcount,
5844 got.br_state);
5845 if (error)
5846 goto del_cursor;
5847 } else
5848 logflags |= XFS_ILOG_DEXT;
5849
5850 /* Add new extent */
5851 current_ext++;
5852 xfs_iext_insert(ip, current_ext, 1, &new, 0);
5853 XFS_IFORK_NEXT_SET(ip, whichfork,
5854 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
5855
5856 if (cur) {
5857 error = xfs_bmbt_lookup_eq(cur, new.br_startoff,
5858 new.br_startblock, new.br_blockcount,
5859 &i);
5860 if (error)
5861 goto del_cursor;
5862 XFS_WANT_CORRUPTED_GOTO(mp, i == 0, del_cursor);
5863 cur->bc_rec.b.br_state = new.br_state;
5864
5865 error = xfs_btree_insert(cur, &i);
5866 if (error)
5867 goto del_cursor;
5868 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
5869 }
5870
5871 /*
5872 * Convert to a btree if necessary.
5873 */
5874 if (xfs_bmap_needs_btree(ip, whichfork)) {
5875 int tmp_logflags; /* partial log flag return val */
5876
5877 ASSERT(cur == NULL);
5878 error = xfs_bmap_extents_to_btree(tp, ip, firstfsb, free_list,
5879 &cur, 0, &tmp_logflags, whichfork);
5880 logflags |= tmp_logflags;
5881 }
5882
5883del_cursor:
5884 if (cur) {
5885 cur->bc_private.b.allocated = 0;
5886 xfs_btree_del_cursor(cur,
5887 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5888 }
5889
5890 if (logflags)
5891 xfs_trans_log_inode(tp, ip, logflags);
5892 return error;
5893}
5894
5895int
5896xfs_bmap_split_extent(
5897 struct xfs_inode *ip,
5898 xfs_fileoff_t split_fsb)
5899{
5900 struct xfs_mount *mp = ip->i_mount;
5901 struct xfs_trans *tp;
5902 struct xfs_bmap_free free_list;
5903 xfs_fsblock_t firstfsb;
5904 int committed;
5905 int error;
5906
5907 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
5908 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
5909 XFS_DIOSTRAT_SPACE_RES(mp, 0), 0);
5910 if (error) {
4906e215 5911 xfs_trans_cancel(tp);
a904b1ca
NJ		 5912 return error;
5913 }
5914
5915 xfs_ilock(ip, XFS_ILOCK_EXCL);
5916 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
5917
5918 xfs_bmap_init(&free_list, &firstfsb);
5919
5920 error = xfs_bmap_split_extent_at(tp, ip, split_fsb,
5921 &firstfsb, &free_list);
5922 if (error)
5923 goto out;
5924
5925 error = xfs_bmap_finish(&tp, &free_list, &committed);
5926 if (error)
5927 goto out;
5928
5929 return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
5930
5931
5932out:
4906e215 5933 xfs_trans_cancel(tp);
a904b1ca
NJ		 5934 return error;
5935}
Ubuntu Artful kernel mirror