Linux Kernel  3.7.1
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
xfs_bmap.c
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write the Free Software Foundation,
16  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17  */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_inum.h"
24 #include "xfs_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_dir2.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_alloc_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_dinode.h"
33 #include "xfs_inode.h"
34 #include "xfs_btree.h"
35 #include "xfs_mount.h"
36 #include "xfs_itable.h"
37 #include "xfs_inode_item.h"
38 #include "xfs_extfree_item.h"
39 #include "xfs_alloc.h"
40 #include "xfs_bmap.h"
41 #include "xfs_rtalloc.h"
42 #include "xfs_error.h"
43 #include "xfs_attr_leaf.h"
44 #include "xfs_quota.h"
45 #include "xfs_trans_space.h"
46 #include "xfs_buf_item.h"
47 #include "xfs_filestream.h"
48 #include "xfs_vnodeops.h"
49 #include "xfs_trace.h"
50 
51 
52 kmem_zone_t *xfs_bmap_free_item_zone;
53 
54 /*
55  * Prototypes for internal bmap routines.
56  */
57 
58 #ifdef DEBUG
59 STATIC void
60 xfs_bmap_check_leaf_extents(
61  struct xfs_btree_cur *cur,
62  struct xfs_inode *ip,
63  int whichfork);
64 #else
65 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork) do { } while (0)
66 #endif
67 
68 
69 /*
70  * Called from xfs_bmap_add_attrfork to handle extents format files.
71  */
72 STATIC int /* error */
73 xfs_bmap_add_attrfork_extents(
74  xfs_trans_t *tp, /* transaction pointer */
75  xfs_inode_t *ip, /* incore inode pointer */
76  xfs_fsblock_t *firstblock, /* first block allocated */
77  xfs_bmap_free_t *flist, /* blocks to free at commit */
78  int *flags); /* inode logging flags */
79 
80 /*
81  * Called from xfs_bmap_add_attrfork to handle local format files.
82  */
83 STATIC int /* error */
84 xfs_bmap_add_attrfork_local(
85  xfs_trans_t *tp, /* transaction pointer */
86  xfs_inode_t *ip, /* incore inode pointer */
87  xfs_fsblock_t *firstblock, /* first block allocated */
88  xfs_bmap_free_t *flist, /* blocks to free at commit */
89  int *flags); /* inode logging flags */
90 
91 /*
92  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
93  * It figures out where to ask the underlying allocator to put the new extent.
94  */
95 STATIC int /* error */
96 xfs_bmap_alloc(
97  xfs_bmalloca_t *ap); /* bmap alloc argument struct */
98 
99 /*
100  * Transform a btree format file with only one leaf node, where the
101  * extents list will fit in the inode, into an extents format file.
102  * Since the file extents are already in-core, all we have to do is
103  * give up the space for the btree root and pitch the leaf block.
104  */
105 STATIC int /* error */
106 xfs_bmap_btree_to_extents(
107  xfs_trans_t *tp, /* transaction pointer */
108  xfs_inode_t *ip, /* incore inode pointer */
109  xfs_btree_cur_t *cur, /* btree cursor */
110  int *logflagsp, /* inode logging flags */
111  int whichfork); /* data or attr fork */
112 
113 /*
114  * Remove the entry "free" from the free item list. Prev points to the
115  * previous entry, unless "free" is the head of the list.
116  */
117 STATIC void
118 xfs_bmap_del_free(
119  xfs_bmap_free_t *flist, /* free item list header */
120  xfs_bmap_free_item_t *prev, /* previous item on list, if any */
121  xfs_bmap_free_item_t *free); /* list item to be freed */
122 
123 /*
124  * Convert an extents-format file into a btree-format file.
125  * The new file will have a root block (in the inode) and a single child block.
126  */
127 STATIC int /* error */
128 xfs_bmap_extents_to_btree(
129  xfs_trans_t *tp, /* transaction pointer */
130  xfs_inode_t *ip, /* incore inode pointer */
131  xfs_fsblock_t *firstblock, /* first-block-allocated */
132  xfs_bmap_free_t *flist, /* blocks freed in xaction */
133  xfs_btree_cur_t **curp, /* cursor returned to caller */
134  int wasdel, /* converting a delayed alloc */
135  int *logflagsp, /* inode logging flags */
136  int whichfork); /* data or attr fork */
137 
138 /*
139  * Convert a local file to an extents file.
140  * This code is sort of bogus, since the file data needs to get
141  * logged so it won't be lost. The bmap-level manipulations are ok, though.
142  */
143 STATIC int /* error */
144 xfs_bmap_local_to_extents(
145  xfs_trans_t *tp, /* transaction pointer */
146  xfs_inode_t *ip, /* incore inode pointer */
147  xfs_fsblock_t *firstblock, /* first block allocated in xaction */
148  xfs_extlen_t total, /* total blocks needed by transaction */
149  int *logflagsp, /* inode logging flags */
150  int whichfork); /* data or attr fork */
151 
152 /*
153  * Search the extents list for the inode, for the extent containing bno.
154  * If bno lies in a hole, point to the next entry. If bno lies past eof,
155  * *eofp will be set, and *prevp will contain the last entry (null if none).
156  * Else, *lastxp will be set to the index of the found
157  * entry; *gotp will contain the entry.
158  */
159 STATIC xfs_bmbt_rec_host_t * /* pointer to found extent entry */
160 xfs_bmap_search_extents(
161  xfs_inode_t *ip, /* incore inode pointer */
162  xfs_fileoff_t bno, /* block number searched for */
163  int whichfork, /* data or attr fork */
164  int *eofp, /* out: end of file found */
165  xfs_extnum_t *lastxp, /* out: last extent index */
166  xfs_bmbt_irec_t *gotp, /* out: extent entry found */
167  xfs_bmbt_irec_t *prevp); /* out: previous extent entry found */
168 
169 /*
170  * Compute the worst-case number of indirect blocks that will be used
171  * for ip's delayed extent of length "len".
172  */
173 STATIC xfs_filblks_t
174 xfs_bmap_worst_indlen(
175  xfs_inode_t *ip, /* incore inode pointer */
176  xfs_filblks_t len); /* delayed extent length */
177 
178 #ifdef DEBUG
179 /*
180  * Perform various validation checks on the values being returned
181  * from xfs_bmapi().
182  */
183 STATIC void
184 xfs_bmap_validate_ret(
185  xfs_fileoff_t bno,
186  xfs_filblks_t len,
187  int flags,
188  xfs_bmbt_irec_t *mval,
189  int nmap,
190  int ret_nmap);
191 #else
192 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
193 #endif /* DEBUG */
194 
195 STATIC int
196 xfs_bmap_count_tree(
197  xfs_mount_t *mp,
198  xfs_trans_t *tp,
199  xfs_ifork_t *ifp,
200  xfs_fsblock_t blockno,
201  int levelin,
202  int *count);
203 
204 STATIC void
205 xfs_bmap_count_leaves(
206  xfs_ifork_t *ifp,
207  xfs_extnum_t idx,
208  int numrecs,
209  int *count);
210 
211 STATIC void
212 xfs_bmap_disk_count_leaves(
213  struct xfs_mount *mp,
214  struct xfs_btree_block *block,
215  int numrecs,
216  int *count);
217 
218 /*
219  * Bmap internal routines.
220  */
221 
222 STATIC int /* error */
223 xfs_bmbt_lookup_eq(
224  struct xfs_btree_cur *cur,
225  xfs_fileoff_t off,
226  xfs_fsblock_t bno,
227  xfs_filblks_t len,
228  int *stat) /* success/failure */
229 {
230  cur->bc_rec.b.br_startoff = off;
231  cur->bc_rec.b.br_startblock = bno;
232  cur->bc_rec.b.br_blockcount = len;
233  return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
234 }
235 
236 STATIC int /* error */
237 xfs_bmbt_lookup_ge(
238  struct xfs_btree_cur *cur,
239  xfs_fileoff_t off,
240  xfs_fsblock_t bno,
241  xfs_filblks_t len,
242  int *stat) /* success/failure */
243 {
244  cur->bc_rec.b.br_startoff = off;
245  cur->bc_rec.b.br_startblock = bno;
246  cur->bc_rec.b.br_blockcount = len;
247  return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
248 }
249 
250 /*
251  * Check if the inode needs to be converted to btree format.
252  */
253 static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
254 {
255  return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
256  XFS_IFORK_NEXTENTS(ip, whichfork) >
257  XFS_IFORK_MAXEXT(ip, whichfork);
258 }
259 
260 /*
261  * Check if the inode should be converted to extent format.
262  */
263 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
264 {
265  return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
266  XFS_IFORK_NEXTENTS(ip, whichfork) <=
267  XFS_IFORK_MAXEXT(ip, whichfork);
268 }
269 
270 /*
271  * Update the record referred to by cur to the value given
272  * by [off, bno, len, state].
273  * This either works (return 0) or gets an EFSCORRUPTED error.
274  */
275 STATIC int
276 xfs_bmbt_update(
277  struct xfs_btree_cur *cur,
278  xfs_fileoff_t off,
279  xfs_fsblock_t bno,
280  xfs_filblks_t len,
281  xfs_exntst_t state)
282 {
283  union xfs_btree_rec rec;
284 
285  xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
286  return xfs_btree_update(cur, &rec);
287 }
288 
289 /*
290  * Called from xfs_bmap_add_attrfork to handle btree format files.
291  */
292 STATIC int /* error */
293 xfs_bmap_add_attrfork_btree(
294  xfs_trans_t *tp, /* transaction pointer */
295  xfs_inode_t *ip, /* incore inode pointer */
296  xfs_fsblock_t *firstblock, /* first block allocated */
297  xfs_bmap_free_t *flist, /* blocks to free at commit */
298  int *flags) /* inode logging flags */
299 {
300  xfs_btree_cur_t *cur; /* btree cursor */
301  int error; /* error return value */
302  xfs_mount_t *mp; /* file system mount struct */
303  int stat; /* newroot status */
304 
305  mp = ip->i_mount;
306  if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
307  *flags |= XFS_ILOG_DBROOT;
308  else {
309  cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
310  cur->bc_private.b.flist = flist;
311  cur->bc_private.b.firstblock = *firstblock;
312  if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
313  goto error0;
314  /* must be at least one entry */
315  XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
316  if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
317  goto error0;
318  if (stat == 0) {
319  xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
320  return XFS_ERROR(ENOSPC);
321  }
322  *firstblock = cur->bc_private.b.firstblock;
323  cur->bc_private.b.allocated = 0;
324  xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
325  }
326  return 0;
327 error0:
328  xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
329  return error;
330 }
331 
332 /*
333  * Called from xfs_bmap_add_attrfork to handle extents format files.
334  */
335 STATIC int /* error */
336 xfs_bmap_add_attrfork_extents(
337  xfs_trans_t *tp, /* transaction pointer */
338  xfs_inode_t *ip, /* incore inode pointer */
339  xfs_fsblock_t *firstblock, /* first block allocated */
340  xfs_bmap_free_t *flist, /* blocks to free at commit */
341  int *flags) /* inode logging flags */
342 {
343  xfs_btree_cur_t *cur; /* bmap btree cursor */
344  int error; /* error return value */
345 
346  if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
347  return 0;
348  cur = NULL;
349  error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
350  flags, XFS_DATA_FORK);
351  if (cur) {
352  cur->bc_private.b.allocated = 0;
353  xfs_btree_del_cursor(cur,
354  error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
355  }
356  return error;
357 }
358 
359 /*
360  * Called from xfs_bmap_add_attrfork to handle local format files.
361  */
362 STATIC int /* error */
363 xfs_bmap_add_attrfork_local(
364  xfs_trans_t *tp, /* transaction pointer */
365  xfs_inode_t *ip, /* incore inode pointer */
366  xfs_fsblock_t *firstblock, /* first block allocated */
367  xfs_bmap_free_t *flist, /* blocks to free at commit */
368  int *flags) /* inode logging flags */
369 {
370  xfs_da_args_t dargs; /* args for dir/attr code */
371  int error; /* error return value */
372  xfs_mount_t *mp; /* mount structure pointer */
373 
374  if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
375  return 0;
376  if (S_ISDIR(ip->i_d.di_mode)) {
377  mp = ip->i_mount;
378  memset(&dargs, 0, sizeof(dargs));
379  dargs.dp = ip;
380  dargs.firstblock = firstblock;
381  dargs.flist = flist;
382  dargs.total = mp->m_dirblkfsbs;
383  dargs.whichfork = XFS_DATA_FORK;
384  dargs.trans = tp;
385  error = xfs_dir2_sf_to_block(&dargs);
386  } else
387  error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
388  XFS_DATA_FORK);
389  return error;
390 }
391 
392 /*
393  * Convert a delayed allocation to a real allocation.
394  */
395 STATIC int /* error */
396 xfs_bmap_add_extent_delay_real(
397  struct xfs_bmalloca *bma)
398 {
399  struct xfs_bmbt_irec *new = &bma->got;
400  int diff; /* temp value */
401  xfs_bmbt_rec_host_t *ep; /* extent entry for idx */
402  int error; /* error return value */
403  int i; /* temp state */
404  xfs_ifork_t *ifp; /* inode fork pointer */
405  xfs_fileoff_t new_endoff; /* end offset of new entry */
406  xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
407  /* left is 0, right is 1, prev is 2 */
408  int rval=0; /* return value (logging flags) */
409  int state = 0;/* state bits, accessed thru macros */
410  xfs_filblks_t da_new; /* new count del alloc blocks used */
411  xfs_filblks_t da_old; /* old count del alloc blocks used */
412  xfs_filblks_t temp=0; /* value for da_new calculations */
413  xfs_filblks_t temp2=0;/* value for da_new calculations */
414  int tmp_rval; /* partial logging flags */
415 
416  ifp = XFS_IFORK_PTR(bma->ip, XFS_DATA_FORK);
417 
418  ASSERT(bma->idx >= 0);
419  ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
420  ASSERT(!isnullstartblock(new->br_startblock));
421  ASSERT(!bma->cur ||
422  (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
423 
424  XFS_STATS_INC(xs_add_exlist);
425 
426 #define LEFT r[0]
427 #define RIGHT r[1]
428 #define PREV r[2]
429 
430  /*
431  * Set up a bunch of variables to make the tests simpler.
432  */
433  ep = xfs_iext_get_ext(ifp, bma->idx);
434  xfs_bmbt_get_all(ep, &PREV);
435  new_endoff = new->br_startoff + new->br_blockcount;
436  ASSERT(PREV.br_startoff <= new->br_startoff);
437  ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
438 
439  da_old = startblockval(PREV.br_startblock);
440  da_new = 0;
441 
442  /*
443  * Set flags determining what part of the previous delayed allocation
444  * extent is being replaced by a real allocation.
445  */
446  if (PREV.br_startoff == new->br_startoff)
447  state |= BMAP_LEFT_FILLING;
448  if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
449  state |= BMAP_RIGHT_FILLING;
450 
451  /*
452  * Check and set flags if this segment has a left neighbor.
453  * Don't set contiguous if the combined extent would be too large.
454  */
455  if (bma->idx > 0) {
456  state |= BMAP_LEFT_VALID;
457  xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
458 
459  if (isnullstartblock(LEFT.br_startblock))
460  state |= BMAP_LEFT_DELAY;
461  }
462 
463  if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
464  LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
465  LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
466  LEFT.br_state == new->br_state &&
467  LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
468  state |= BMAP_LEFT_CONTIG;
469 
470  /*
471  * Check and set flags if this segment has a right neighbor.
472  * Don't set contiguous if the combined extent would be too large.
473  * Also check for all-three-contiguous being too large.
474  */
475  if (bma->idx < bma->ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
476  state |= BMAP_RIGHT_VALID;
477  xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
478 
479  if (isnullstartblock(RIGHT.br_startblock))
480  state |= BMAP_RIGHT_DELAY;
481  }
482 
483  if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
484  new_endoff == RIGHT.br_startoff &&
485  new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
486  new->br_state == RIGHT.br_state &&
487  new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
488  ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
489  BMAP_RIGHT_FILLING)) !=
490  (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
491  BMAP_RIGHT_FILLING) ||
492  LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
493  <= MAXEXTLEN))
494  state |= BMAP_RIGHT_CONTIG;
495 
496  error = 0;
497  /*
498  * Switch out based on the FILLING and CONTIG state bits.
499  */
500  switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
501  BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
502  case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
503  BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
504  /*
505  * Filling in all of a previously delayed allocation extent.
506  * The left and right neighbors are both contiguous with new.
507  */
508  bma->idx--;
509  trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
510  xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
511  LEFT.br_blockcount + PREV.br_blockcount +
512  RIGHT.br_blockcount);
513  trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
514 
515  xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
516  bma->ip->i_d.di_nextents--;
517  if (bma->cur == NULL)
518  rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
519  else {
520  rval = XFS_ILOG_CORE;
521  error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
522  RIGHT.br_startblock,
523  RIGHT.br_blockcount, &i);
524  if (error)
525  goto done;
526  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
527  error = xfs_btree_delete(bma->cur, &i);
528  if (error)
529  goto done;
530  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
531  error = xfs_btree_decrement(bma->cur, 0, &i);
532  if (error)
533  goto done;
534  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
535  error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
536  LEFT.br_startblock,
537  LEFT.br_blockcount +
538  PREV.br_blockcount +
539  RIGHT.br_blockcount, LEFT.br_state);
540  if (error)
541  goto done;
542  }
543  break;
544 
545  case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
546  /*
547  * Filling in all of a previously delayed allocation extent.
548  * The left neighbor is contiguous, the right is not.
549  */
550  bma->idx--;
551 
552  trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
553  xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
554  LEFT.br_blockcount + PREV.br_blockcount);
555  trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
556 
557  xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
558  if (bma->cur == NULL)
559  rval = XFS_ILOG_DEXT;
560  else {
561  rval = 0;
562  error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
563  LEFT.br_startblock, LEFT.br_blockcount,
564  &i);
565  if (error)
566  goto done;
567  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
568  error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
569  LEFT.br_startblock,
570  LEFT.br_blockcount +
571  PREV.br_blockcount, LEFT.br_state);
572  if (error)
573  goto done;
574  }
575  break;
576 
577  case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
578  /*
579  * Filling in all of a previously delayed allocation extent.
580  * The right neighbor is contiguous, the left is not.
581  */
582  trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
583  xfs_bmbt_set_startblock(ep, new->br_startblock);
584  xfs_bmbt_set_blockcount(ep,
585  PREV.br_blockcount + RIGHT.br_blockcount);
586  trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
587 
588  xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
589  if (bma->cur == NULL)
590  rval = XFS_ILOG_DEXT;
591  else {
592  rval = 0;
593  error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
594  RIGHT.br_startblock,
595  RIGHT.br_blockcount, &i);
596  if (error)
597  goto done;
598  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
599  error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
600  new->br_startblock,
601  PREV.br_blockcount +
602  RIGHT.br_blockcount, PREV.br_state);
603  if (error)
604  goto done;
605  }
606  break;
607 
608  case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
609  /*
610  * Filling in all of a previously delayed allocation extent.
611  * Neither the left nor right neighbors are contiguous with
612  * the new one.
613  */
614  trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
615  xfs_bmbt_set_startblock(ep, new->br_startblock);
616  trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
617 
618  bma->ip->i_d.di_nextents++;
619  if (bma->cur == NULL)
620  rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
621  else {
622  rval = XFS_ILOG_CORE;
623  error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
624  new->br_startblock, new->br_blockcount,
625  &i);
626  if (error)
627  goto done;
628  XFS_WANT_CORRUPTED_GOTO(i == 0, done);
629  bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
630  error = xfs_btree_insert(bma->cur, &i);
631  if (error)
632  goto done;
633  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
634  }
635  break;
636 
637  case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
638  /*
639  * Filling in the first part of a previous delayed allocation.
640  * The left neighbor is contiguous.
641  */
642  trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
643  xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
644  LEFT.br_blockcount + new->br_blockcount);
645  xfs_bmbt_set_startoff(ep,
646  PREV.br_startoff + new->br_blockcount);
647  trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
648 
649  temp = PREV.br_blockcount - new->br_blockcount;
650  trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
651  xfs_bmbt_set_blockcount(ep, temp);
652  if (bma->cur == NULL)
653  rval = XFS_ILOG_DEXT;
654  else {
655  rval = 0;
656  error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
657  LEFT.br_startblock, LEFT.br_blockcount,
658  &i);
659  if (error)
660  goto done;
661  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
662  error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
663  LEFT.br_startblock,
664  LEFT.br_blockcount +
665  new->br_blockcount,
666  LEFT.br_state);
667  if (error)
668  goto done;
669  }
670  da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
671  startblockval(PREV.br_startblock));
672  xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
673  trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
674 
675  bma->idx--;
676  break;
677 
678  case BMAP_LEFT_FILLING:
679  /*
680  * Filling in the first part of a previous delayed allocation.
681  * The left neighbor is not contiguous.
682  */
683  trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
684  xfs_bmbt_set_startoff(ep, new_endoff);
685  temp = PREV.br_blockcount - new->br_blockcount;
686  xfs_bmbt_set_blockcount(ep, temp);
687  xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
688  bma->ip->i_d.di_nextents++;
689  if (bma->cur == NULL)
690  rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
691  else {
692  rval = XFS_ILOG_CORE;
693  error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
694  new->br_startblock, new->br_blockcount,
695  &i);
696  if (error)
697  goto done;
698  XFS_WANT_CORRUPTED_GOTO(i == 0, done);
699  bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
700  error = xfs_btree_insert(bma->cur, &i);
701  if (error)
702  goto done;
703  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
704  }
705 
706  if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
707  error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
708  bma->firstblock, bma->flist,
709  &bma->cur, 1, &tmp_rval, XFS_DATA_FORK);
710  rval |= tmp_rval;
711  if (error)
712  goto done;
713  }
714  da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
715  startblockval(PREV.br_startblock) -
716  (bma->cur ? bma->cur->bc_private.b.allocated : 0));
717  ep = xfs_iext_get_ext(ifp, bma->idx + 1);
718  xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
719  trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
720  break;
721 
722  case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
723  /*
724  * Filling in the last part of a previous delayed allocation.
725  * The right neighbor is contiguous with the new allocation.
726  */
727  temp = PREV.br_blockcount - new->br_blockcount;
728  trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
729  xfs_bmbt_set_blockcount(ep, temp);
730  xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
731  new->br_startoff, new->br_startblock,
732  new->br_blockcount + RIGHT.br_blockcount,
733  RIGHT.br_state);
734  trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
735  if (bma->cur == NULL)
736  rval = XFS_ILOG_DEXT;
737  else {
738  rval = 0;
739  error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
740  RIGHT.br_startblock,
741  RIGHT.br_blockcount, &i);
742  if (error)
743  goto done;
744  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
745  error = xfs_bmbt_update(bma->cur, new->br_startoff,
746  new->br_startblock,
747  new->br_blockcount +
748  RIGHT.br_blockcount,
749  RIGHT.br_state);
750  if (error)
751  goto done;
752  }
753 
754  da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
755  startblockval(PREV.br_startblock));
756  trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
757  xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
758  trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
759 
760  bma->idx++;
761  break;
762 
763  case BMAP_RIGHT_FILLING:
764  /*
765  * Filling in the last part of a previous delayed allocation.
766  * The right neighbor is not contiguous.
767  */
768  temp = PREV.br_blockcount - new->br_blockcount;
769  trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
770  xfs_bmbt_set_blockcount(ep, temp);
771  xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
772  bma->ip->i_d.di_nextents++;
773  if (bma->cur == NULL)
774  rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
775  else {
776  rval = XFS_ILOG_CORE;
777  error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
778  new->br_startblock, new->br_blockcount,
779  &i);
780  if (error)
781  goto done;
782  XFS_WANT_CORRUPTED_GOTO(i == 0, done);
783  bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
784  error = xfs_btree_insert(bma->cur, &i);
785  if (error)
786  goto done;
787  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
788  }
789 
790  if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
791  error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
792  bma->firstblock, bma->flist, &bma->cur, 1,
793  &tmp_rval, XFS_DATA_FORK);
794  rval |= tmp_rval;
795  if (error)
796  goto done;
797  }
798  da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
799  startblockval(PREV.br_startblock) -
800  (bma->cur ? bma->cur->bc_private.b.allocated : 0));
801  ep = xfs_iext_get_ext(ifp, bma->idx);
802  xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
803  trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
804 
805  bma->idx++;
806  break;
807 
808  case 0:
809  /*
810  * Filling in the middle part of a previous delayed allocation.
811  * Contiguity is impossible here.
812  * This case is avoided almost all the time.
813  *
814  * We start with a delayed allocation:
815  *
816  * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
817  * PREV @ idx
818  *
819  * and we are allocating:
820  * +rrrrrrrrrrrrrrrrr+
821  * new
822  *
823  * and we set it up for insertion as:
824  * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
825  * new
826  * PREV @ idx LEFT RIGHT
827  * inserted at idx + 1
828  */
829  temp = new->br_startoff - PREV.br_startoff;
830  temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
831  trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
832  xfs_bmbt_set_blockcount(ep, temp); /* truncate PREV */
833  LEFT = *new;
834  RIGHT.br_state = PREV.br_state;
835  RIGHT.br_startblock = nullstartblock(
836  (int)xfs_bmap_worst_indlen(bma->ip, temp2));
837  RIGHT.br_startoff = new_endoff;
838  RIGHT.br_blockcount = temp2;
839  /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
840  xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
841  bma->ip->i_d.di_nextents++;
842  if (bma->cur == NULL)
843  rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
844  else {
845  rval = XFS_ILOG_CORE;
846  error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
847  new->br_startblock, new->br_blockcount,
848  &i);
849  if (error)
850  goto done;
851  XFS_WANT_CORRUPTED_GOTO(i == 0, done);
852  bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
853  error = xfs_btree_insert(bma->cur, &i);
854  if (error)
855  goto done;
856  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
857  }
858 
859  if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
860  error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
861  bma->firstblock, bma->flist, &bma->cur,
862  1, &tmp_rval, XFS_DATA_FORK);
863  rval |= tmp_rval;
864  if (error)
865  goto done;
866  }
867  temp = xfs_bmap_worst_indlen(bma->ip, temp);
868  temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
869  diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
870  (bma->cur ? bma->cur->bc_private.b.allocated : 0));
871  if (diff > 0) {
872  error = xfs_icsb_modify_counters(bma->ip->i_mount,
873  XFS_SBS_FDBLOCKS,
874  -((int64_t)diff), 0);
875  ASSERT(!error);
876  if (error)
877  goto done;
878  }
879 
880  ep = xfs_iext_get_ext(ifp, bma->idx);
881  xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
882  trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
883  trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
884  xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
885  nullstartblock((int)temp2));
886  trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
887 
888  bma->idx++;
889  da_new = temp + temp2;
890  break;
891 
892  case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
893  case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
894  case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
895  case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
896  case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
897  case BMAP_LEFT_CONTIG:
898  case BMAP_RIGHT_CONTIG:
899  /*
900  * These cases are all impossible.
901  */
902  ASSERT(0);
903  }
904 
905  /* convert to a btree if necessary */
906  if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
907  int tmp_logflags; /* partial log flag return val */
908 
909  ASSERT(bma->cur == NULL);
910  error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
911  bma->firstblock, bma->flist, &bma->cur,
912  da_old > 0, &tmp_logflags, XFS_DATA_FORK);
913  bma->logflags |= tmp_logflags;
914  if (error)
915  goto done;
916  }
917 
918  /* adjust for changes in reserved delayed indirect blocks */
919  if (da_old || da_new) {
920  temp = da_new;
921  if (bma->cur)
922  temp += bma->cur->bc_private.b.allocated;
923  ASSERT(temp <= da_old);
924  if (temp < da_old)
925  xfs_icsb_modify_counters(bma->ip->i_mount,
926  XFS_SBS_FDBLOCKS,
927  (int64_t)(da_old - temp), 0);
928  }
929 
930  /* clear out the allocated field, done with it now in any case. */
931  if (bma->cur)
932  bma->cur->bc_private.b.allocated = 0;
933 
934  xfs_bmap_check_leaf_extents(bma->cur, bma->ip, XFS_DATA_FORK);
935 done:
936  bma->logflags |= rval;
937  return error;
938 #undef LEFT
939 #undef RIGHT
940 #undef PREV
941 }
942 
943 /*
944  * Convert an unwritten allocation to a real allocation or vice versa.
945  */
946 STATIC int /* error */
947 xfs_bmap_add_extent_unwritten_real(
948  struct xfs_trans *tp,
949  xfs_inode_t *ip, /* incore inode pointer */
950  xfs_extnum_t *idx, /* extent number to update/insert */
951  xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
952  xfs_bmbt_irec_t *new, /* new data to add to file extents */
953  xfs_fsblock_t *first, /* pointer to firstblock variable */
954  xfs_bmap_free_t *flist, /* list of extents to be freed */
955  int *logflagsp) /* inode logging flags */
956 {
957  xfs_btree_cur_t *cur; /* btree cursor */
958  xfs_bmbt_rec_host_t *ep; /* extent entry for idx */
959  int error; /* error return value */
960  int i; /* temp state */
961  xfs_ifork_t *ifp; /* inode fork pointer */
962  xfs_fileoff_t new_endoff; /* end offset of new entry */
963  xfs_exntst_t newext; /* new extent state */
964  xfs_exntst_t oldext; /* old extent state */
965  xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
966  /* left is 0, right is 1, prev is 2 */
967  int rval=0; /* return value (logging flags) */
968  int state = 0;/* state bits, accessed thru macros */
969 
970  *logflagsp = 0;
971 
972  cur = *curp;
973  ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
974 
975  ASSERT(*idx >= 0);
976  ASSERT(*idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
977  ASSERT(!isnullstartblock(new->br_startblock));
978 
979  XFS_STATS_INC(xs_add_exlist);
980 
981 #define LEFT r[0]
982 #define RIGHT r[1]
983 #define PREV r[2]
984 
985  /*
986  * Set up a bunch of variables to make the tests simpler.
987  */
988  error = 0;
989  ep = xfs_iext_get_ext(ifp, *idx);
990  xfs_bmbt_get_all(ep, &PREV);
991  newext = new->br_state;
992  oldext = (newext == XFS_EXT_UNWRITTEN) ?
993  XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
994  ASSERT(PREV.br_state == oldext);
995  new_endoff = new->br_startoff + new->br_blockcount;
996  ASSERT(PREV.br_startoff <= new->br_startoff);
997  ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
998 
999  /*
1000  * Set flags determining what part of the previous oldext allocation
1001  * extent is being replaced by a newext allocation.
1002  */
1003  if (PREV.br_startoff == new->br_startoff)
1004  state |= BMAP_LEFT_FILLING;
1005  if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1006  state |= BMAP_RIGHT_FILLING;
1007 
1008  /*
1009  * Check and set flags if this segment has a left neighbor.
1010  * Don't set contiguous if the combined extent would be too large.
1011  */
1012  if (*idx > 0) {
1013  state |= BMAP_LEFT_VALID;
1014  xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
1015 
1016  if (isnullstartblock(LEFT.br_startblock))
1017  state |= BMAP_LEFT_DELAY;
1018  }
1019 
1020  if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1021  LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1022  LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1023  LEFT.br_state == newext &&
1024  LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1025  state |= BMAP_LEFT_CONTIG;
1026 
1027  /*
1028  * Check and set flags if this segment has a right neighbor.
1029  * Don't set contiguous if the combined extent would be too large.
1030  * Also check for all-three-contiguous being too large.
1031  */
1032  if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1033  state |= BMAP_RIGHT_VALID;
1034  xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
1035  if (isnullstartblock(RIGHT.br_startblock))
1036  state |= BMAP_RIGHT_DELAY;
1037  }
1038 
1039  if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1040  new_endoff == RIGHT.br_startoff &&
1041  new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1042  newext == RIGHT.br_state &&
1043  new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1044  ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1045  BMAP_RIGHT_FILLING)) !=
1046  (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1047  BMAP_RIGHT_FILLING) ||
1048  LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1049  <= MAXEXTLEN))
1050  state |= BMAP_RIGHT_CONTIG;
1051 
1052  /*
1053  * Switch out based on the FILLING and CONTIG state bits.
1054  */
1055  switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1056  BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1057  case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1058  BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1059  /*
1060  * Setting all of a previous oldext extent to newext.
1061  * The left and right neighbors are both contiguous with new.
1062  */
1063  --*idx;
1064 
1065  trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1066  xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1067  LEFT.br_blockcount + PREV.br_blockcount +
1068  RIGHT.br_blockcount);
1069  trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1070 
1071  xfs_iext_remove(ip, *idx + 1, 2, state);
1072  ip->i_d.di_nextents -= 2;
1073  if (cur == NULL)
1074  rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1075  else {
1076  rval = XFS_ILOG_CORE;
1077  if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1078  RIGHT.br_startblock,
1079  RIGHT.br_blockcount, &i)))
1080  goto done;
1081  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1082  if ((error = xfs_btree_delete(cur, &i)))
1083  goto done;
1084  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1085  if ((error = xfs_btree_decrement(cur, 0, &i)))
1086  goto done;
1087  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1088  if ((error = xfs_btree_delete(cur, &i)))
1089  goto done;
1090  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1091  if ((error = xfs_btree_decrement(cur, 0, &i)))
1092  goto done;
1093  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1094  if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1095  LEFT.br_startblock,
1096  LEFT.br_blockcount + PREV.br_blockcount +
1097  RIGHT.br_blockcount, LEFT.br_state)))
1098  goto done;
1099  }
1100  break;
1101 
1102  case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1103  /*
1104  * Setting all of a previous oldext extent to newext.
1105  * The left neighbor is contiguous, the right is not.
1106  */
1107  --*idx;
1108 
1109  trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1110  xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1111  LEFT.br_blockcount + PREV.br_blockcount);
1112  trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1113 
1114  xfs_iext_remove(ip, *idx + 1, 1, state);
1115  ip->i_d.di_nextents--;
1116  if (cur == NULL)
1117  rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1118  else {
1119  rval = XFS_ILOG_CORE;
1120  if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1121  PREV.br_startblock, PREV.br_blockcount,
1122  &i)))
1123  goto done;
1124  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1125  if ((error = xfs_btree_delete(cur, &i)))
1126  goto done;
1127  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1128  if ((error = xfs_btree_decrement(cur, 0, &i)))
1129  goto done;
1130  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1131  if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1132  LEFT.br_startblock,
1133  LEFT.br_blockcount + PREV.br_blockcount,
1134  LEFT.br_state)))
1135  goto done;
1136  }
1137  break;
1138 
1139  case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1140  /*
1141  * Setting all of a previous oldext extent to newext.
1142  * The right neighbor is contiguous, the left is not.
1143  */
1144  trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1145  xfs_bmbt_set_blockcount(ep,
1146  PREV.br_blockcount + RIGHT.br_blockcount);
1147  xfs_bmbt_set_state(ep, newext);
1148  trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1149  xfs_iext_remove(ip, *idx + 1, 1, state);
1150  ip->i_d.di_nextents--;
1151  if (cur == NULL)
1152  rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1153  else {
1154  rval = XFS_ILOG_CORE;
1155  if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1156  RIGHT.br_startblock,
1157  RIGHT.br_blockcount, &i)))
1158  goto done;
1159  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1160  if ((error = xfs_btree_delete(cur, &i)))
1161  goto done;
1162  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1163  if ((error = xfs_btree_decrement(cur, 0, &i)))
1164  goto done;
1165  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1166  if ((error = xfs_bmbt_update(cur, new->br_startoff,
1167  new->br_startblock,
1168  new->br_blockcount + RIGHT.br_blockcount,
1169  newext)))
1170  goto done;
1171  }
1172  break;
1173 
1174  case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1175  /*
1176  * Setting all of a previous oldext extent to newext.
1177  * Neither the left nor right neighbors are contiguous with
1178  * the new one.
1179  */
1180  trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1181  xfs_bmbt_set_state(ep, newext);
1182  trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1183 
1184  if (cur == NULL)
1185  rval = XFS_ILOG_DEXT;
1186  else {
1187  rval = 0;
1188  if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1189  new->br_startblock, new->br_blockcount,
1190  &i)))
1191  goto done;
1192  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1193  if ((error = xfs_bmbt_update(cur, new->br_startoff,
1194  new->br_startblock, new->br_blockcount,
1195  newext)))
1196  goto done;
1197  }
1198  break;
1199 
1200  case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1201  /*
1202  * Setting the first part of a previous oldext extent to newext.
1203  * The left neighbor is contiguous.
1204  */
1205  trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
1206  xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
1207  LEFT.br_blockcount + new->br_blockcount);
1208  xfs_bmbt_set_startoff(ep,
1209  PREV.br_startoff + new->br_blockcount);
1210  trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
1211 
1212  trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1213  xfs_bmbt_set_startblock(ep,
1214  new->br_startblock + new->br_blockcount);
1215  xfs_bmbt_set_blockcount(ep,
1216  PREV.br_blockcount - new->br_blockcount);
1217  trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1218 
1219  --*idx;
1220 
1221  if (cur == NULL)
1222  rval = XFS_ILOG_DEXT;
1223  else {
1224  rval = 0;
1225  if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1226  PREV.br_startblock, PREV.br_blockcount,
1227  &i)))
1228  goto done;
1229  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1230  if ((error = xfs_bmbt_update(cur,
1231  PREV.br_startoff + new->br_blockcount,
1232  PREV.br_startblock + new->br_blockcount,
1233  PREV.br_blockcount - new->br_blockcount,
1234  oldext)))
1235  goto done;
1236  if ((error = xfs_btree_decrement(cur, 0, &i)))
1237  goto done;
1238  error = xfs_bmbt_update(cur, LEFT.br_startoff,
1239  LEFT.br_startblock,
1240  LEFT.br_blockcount + new->br_blockcount,
1241  LEFT.br_state);
1242  if (error)
1243  goto done;
1244  }
1245  break;
1246 
1247  case BMAP_LEFT_FILLING:
1248  /*
1249  * Setting the first part of a previous oldext extent to newext.
1250  * The left neighbor is not contiguous.
1251  */
1252  trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1253  ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1254  xfs_bmbt_set_startoff(ep, new_endoff);
1255  xfs_bmbt_set_blockcount(ep,
1256  PREV.br_blockcount - new->br_blockcount);
1257  xfs_bmbt_set_startblock(ep,
1258  new->br_startblock + new->br_blockcount);
1259  trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1260 
1261  xfs_iext_insert(ip, *idx, 1, new, state);
1262  ip->i_d.di_nextents++;
1263  if (cur == NULL)
1264  rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1265  else {
1266  rval = XFS_ILOG_CORE;
1267  if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1268  PREV.br_startblock, PREV.br_blockcount,
1269  &i)))
1270  goto done;
1271  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1272  if ((error = xfs_bmbt_update(cur,
1273  PREV.br_startoff + new->br_blockcount,
1274  PREV.br_startblock + new->br_blockcount,
1275  PREV.br_blockcount - new->br_blockcount,
1276  oldext)))
1277  goto done;
1278  cur->bc_rec.b = *new;
1279  if ((error = xfs_btree_insert(cur, &i)))
1280  goto done;
1281  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1282  }
1283  break;
1284 
1285  case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1286  /*
1287  * Setting the last part of a previous oldext extent to newext.
1288  * The right neighbor is contiguous with the new allocation.
1289  */
1290  trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1291  xfs_bmbt_set_blockcount(ep,
1292  PREV.br_blockcount - new->br_blockcount);
1293  trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1294 
1295  ++*idx;
1296 
1297  trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1298  xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1299  new->br_startoff, new->br_startblock,
1300  new->br_blockcount + RIGHT.br_blockcount, newext);
1301  trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1302 
1303  if (cur == NULL)
1304  rval = XFS_ILOG_DEXT;
1305  else {
1306  rval = 0;
1307  if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1308  PREV.br_startblock,
1309  PREV.br_blockcount, &i)))
1310  goto done;
1311  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1312  if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1313  PREV.br_startblock,
1314  PREV.br_blockcount - new->br_blockcount,
1315  oldext)))
1316  goto done;
1317  if ((error = xfs_btree_increment(cur, 0, &i)))
1318  goto done;
1319  if ((error = xfs_bmbt_update(cur, new->br_startoff,
1320  new->br_startblock,
1321  new->br_blockcount + RIGHT.br_blockcount,
1322  newext)))
1323  goto done;
1324  }
1325  break;
1326 
1327  case BMAP_RIGHT_FILLING:
1328  /*
1329  * Setting the last part of a previous oldext extent to newext.
1330  * The right neighbor is not contiguous.
1331  */
1332  trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1333  xfs_bmbt_set_blockcount(ep,
1334  PREV.br_blockcount - new->br_blockcount);
1335  trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1336 
1337  ++*idx;
1338  xfs_iext_insert(ip, *idx, 1, new, state);
1339 
1340  ip->i_d.di_nextents++;
1341  if (cur == NULL)
1342  rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1343  else {
1344  rval = XFS_ILOG_CORE;
1345  if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1346  PREV.br_startblock, PREV.br_blockcount,
1347  &i)))
1348  goto done;
1349  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1350  if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1351  PREV.br_startblock,
1352  PREV.br_blockcount - new->br_blockcount,
1353  oldext)))
1354  goto done;
1355  if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1356  new->br_startblock, new->br_blockcount,
1357  &i)))
1358  goto done;
1359  XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1360  cur->bc_rec.b.br_state = XFS_EXT_NORM;
1361  if ((error = xfs_btree_insert(cur, &i)))
1362  goto done;
1363  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1364  }
1365  break;
1366 
1367  case 0:
1368  /*
1369  * Setting the middle part of a previous oldext extent to
1370  * newext. Contiguity is impossible here.
1371  * One extent becomes three extents.
1372  */
1373  trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1374  xfs_bmbt_set_blockcount(ep,
1375  new->br_startoff - PREV.br_startoff);
1376  trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1377 
1378  r[0] = *new;
1379  r[1].br_startoff = new_endoff;
1380  r[1].br_blockcount =
1381  PREV.br_startoff + PREV.br_blockcount - new_endoff;
1382  r[1].br_startblock = new->br_startblock + new->br_blockcount;
1383  r[1].br_state = oldext;
1384 
1385  ++*idx;
1386  xfs_iext_insert(ip, *idx, 2, &r[0], state);
1387 
1388  ip->i_d.di_nextents += 2;
1389  if (cur == NULL)
1390  rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1391  else {
1392  rval = XFS_ILOG_CORE;
1393  if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1394  PREV.br_startblock, PREV.br_blockcount,
1395  &i)))
1396  goto done;
1397  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1398  /* new right extent - oldext */
1399  if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1400  r[1].br_startblock, r[1].br_blockcount,
1401  r[1].br_state)))
1402  goto done;
1403  /* new left extent - oldext */
1404  cur->bc_rec.b = PREV;
1405  cur->bc_rec.b.br_blockcount =
1406  new->br_startoff - PREV.br_startoff;
1407  if ((error = xfs_btree_insert(cur, &i)))
1408  goto done;
1409  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1410  /*
1411  * Reset the cursor to the position of the new extent
1412  * we are about to insert as we can't trust it after
1413  * the previous insert.
1414  */
1415  if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1416  new->br_startblock, new->br_blockcount,
1417  &i)))
1418  goto done;
1419  XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1420  /* new middle extent - newext */
1421  cur->bc_rec.b.br_state = new->br_state;
1422  if ((error = xfs_btree_insert(cur, &i)))
1423  goto done;
1424  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1425  }
1426  break;
1427 
1428  case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1429  case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1430  case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1431  case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1432  case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1433  case BMAP_LEFT_CONTIG:
1434  case BMAP_RIGHT_CONTIG:
1435  /*
1436  * These cases are all impossible.
1437  */
1438  ASSERT(0);
1439  }
1440 
1441  /* convert to a btree if necessary */
1442  if (xfs_bmap_needs_btree(ip, XFS_DATA_FORK)) {
1443  int tmp_logflags; /* partial log flag return val */
1444 
1445  ASSERT(cur == NULL);
1446  error = xfs_bmap_extents_to_btree(tp, ip, first, flist, &cur,
1447  0, &tmp_logflags, XFS_DATA_FORK);
1448  *logflagsp |= tmp_logflags;
1449  if (error)
1450  goto done;
1451  }
1452 
1453  /* clear out the allocated field, done with it now in any case. */
1454  if (cur) {
1455  cur->bc_private.b.allocated = 0;
1456  *curp = cur;
1457  }
1458 
1459  xfs_bmap_check_leaf_extents(*curp, ip, XFS_DATA_FORK);
1460 done:
1461  *logflagsp |= rval;
1462  return error;
1463 #undef LEFT
1464 #undef RIGHT
1465 #undef PREV
1466 }
1467 
1468 /*
1469  * Convert a hole to a delayed allocation.
1470  */
1471 STATIC void
1472 xfs_bmap_add_extent_hole_delay(
1473  xfs_inode_t *ip, /* incore inode pointer */
1474  xfs_extnum_t *idx, /* extent number to update/insert */
1475  xfs_bmbt_irec_t *new) /* new data to add to file extents */
1476 {
1477  xfs_ifork_t *ifp; /* inode fork pointer */
1478  xfs_bmbt_irec_t left; /* left neighbor extent entry */
1479  xfs_filblks_t newlen=0; /* new indirect size */
1480  xfs_filblks_t oldlen=0; /* old indirect size */
1481  xfs_bmbt_irec_t right; /* right neighbor extent entry */
1482  int state; /* state bits, accessed thru macros */
1483  xfs_filblks_t temp=0; /* temp for indirect calculations */
1484 
1485  ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1486  state = 0;
1487  ASSERT(isnullstartblock(new->br_startblock));
1488 
1489  /*
1490  * Check and set flags if this segment has a left neighbor
1491  */
1492  if (*idx > 0) {
1493  state |= BMAP_LEFT_VALID;
1494  xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
1495 
1496  if (isnullstartblock(left.br_startblock))
1497  state |= BMAP_LEFT_DELAY;
1498  }
1499 
1500  /*
1501  * Check and set flags if the current (right) segment exists.
1502  * If it doesn't exist, we're converting the hole at end-of-file.
1503  */
1504  if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1505  state |= BMAP_RIGHT_VALID;
1506  xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
1507 
1508  if (isnullstartblock(right.br_startblock))
1509  state |= BMAP_RIGHT_DELAY;
1510  }
1511 
1512  /*
1513  * Set contiguity flags on the left and right neighbors.
1514  * Don't let extents get too large, even if the pieces are contiguous.
1515  */
1516  if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
1517  left.br_startoff + left.br_blockcount == new->br_startoff &&
1518  left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1519  state |= BMAP_LEFT_CONTIG;
1520 
1521  if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
1522  new->br_startoff + new->br_blockcount == right.br_startoff &&
1523  new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1524  (!(state & BMAP_LEFT_CONTIG) ||
1525  (left.br_blockcount + new->br_blockcount +
1526  right.br_blockcount <= MAXEXTLEN)))
1527  state |= BMAP_RIGHT_CONTIG;
1528 
1529  /*
1530  * Switch out based on the contiguity flags.
1531  */
1532  switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1533  case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1534  /*
1535  * New allocation is contiguous with delayed allocations
1536  * on the left and on the right.
1537  * Merge all three into a single extent record.
1538  */
1539  --*idx;
1540  temp = left.br_blockcount + new->br_blockcount +
1541  right.br_blockcount;
1542 
1543  trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1544  xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1545  oldlen = startblockval(left.br_startblock) +
1546  startblockval(new->br_startblock) +
1547  startblockval(right.br_startblock);
1548  newlen = xfs_bmap_worst_indlen(ip, temp);
1549  xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1550  nullstartblock((int)newlen));
1551  trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1552 
1553  xfs_iext_remove(ip, *idx + 1, 1, state);
1554  break;
1555 
1556  case BMAP_LEFT_CONTIG:
1557  /*
1558  * New allocation is contiguous with a delayed allocation
1559  * on the left.
1560  * Merge the new allocation with the left neighbor.
1561  */
1562  --*idx;
1563  temp = left.br_blockcount + new->br_blockcount;
1564 
1565  trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1566  xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1567  oldlen = startblockval(left.br_startblock) +
1568  startblockval(new->br_startblock);
1569  newlen = xfs_bmap_worst_indlen(ip, temp);
1570  xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1571  nullstartblock((int)newlen));
1572  trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1573  break;
1574 
1575  case BMAP_RIGHT_CONTIG:
1576  /*
1577  * New allocation is contiguous with a delayed allocation
1578  * on the right.
1579  * Merge the new allocation with the right neighbor.
1580  */
1581  trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1582  temp = new->br_blockcount + right.br_blockcount;
1583  oldlen = startblockval(new->br_startblock) +
1584  startblockval(right.br_startblock);
1585  newlen = xfs_bmap_worst_indlen(ip, temp);
1586  xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1587  new->br_startoff,
1588  nullstartblock((int)newlen), temp, right.br_state);
1589  trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1590  break;
1591 
1592  case 0:
1593  /*
1594  * New allocation is not contiguous with another
1595  * delayed allocation.
1596  * Insert a new entry.
1597  */
1598  oldlen = newlen = 0;
1599  xfs_iext_insert(ip, *idx, 1, new, state);
1600  break;
1601  }
1602  if (oldlen != newlen) {
1603  ASSERT(oldlen > newlen);
1604  xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
1605  (int64_t)(oldlen - newlen), 0);
1606  /*
1607  * Nothing to do for disk quota accounting here.
1608  */
1609  }
1610 }
1611 
1612 /*
1613  * Convert a hole to a real allocation.
1614  */
1615 STATIC int /* error */
1616 xfs_bmap_add_extent_hole_real(
1617  struct xfs_bmalloca *bma,
1618  int whichfork)
1619 {
1620  struct xfs_bmbt_irec *new = &bma->got;
1621  int error; /* error return value */
1622  int i; /* temp state */
1623  xfs_ifork_t *ifp; /* inode fork pointer */
1624  xfs_bmbt_irec_t left; /* left neighbor extent entry */
1625  xfs_bmbt_irec_t right; /* right neighbor extent entry */
1626  int rval=0; /* return value (logging flags) */
1627  int state; /* state bits, accessed thru macros */
1628 
1629  ifp = XFS_IFORK_PTR(bma->ip, whichfork);
1630 
1631  ASSERT(bma->idx >= 0);
1632  ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
1633  ASSERT(!isnullstartblock(new->br_startblock));
1634  ASSERT(!bma->cur ||
1635  !(bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1636 
1637  XFS_STATS_INC(xs_add_exlist);
1638 
1639  state = 0;
1640  if (whichfork == XFS_ATTR_FORK)
1641  state |= BMAP_ATTRFORK;
1642 
1643  /*
1644  * Check and set flags if this segment has a left neighbor.
1645  */
1646  if (bma->idx > 0) {
1647  state |= BMAP_LEFT_VALID;
1648  xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &left);
1649  if (isnullstartblock(left.br_startblock))
1650  state |= BMAP_LEFT_DELAY;
1651  }
1652 
1653  /*
1654  * Check and set flags if this segment has a current value.
1655  * Not true if we're inserting into the "hole" at eof.
1656  */
1657  if (bma->idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1658  state |= BMAP_RIGHT_VALID;
1659  xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &right);
1660  if (isnullstartblock(right.br_startblock))
1661  state |= BMAP_RIGHT_DELAY;
1662  }
1663 
1664  /*
1665  * We're inserting a real allocation between "left" and "right".
1666  * Set the contiguity flags. Don't let extents get too large.
1667  */
1668  if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1669  left.br_startoff + left.br_blockcount == new->br_startoff &&
1670  left.br_startblock + left.br_blockcount == new->br_startblock &&
1671  left.br_state == new->br_state &&
1672  left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1673  state |= BMAP_LEFT_CONTIG;
1674 
1675  if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1676  new->br_startoff + new->br_blockcount == right.br_startoff &&
1677  new->br_startblock + new->br_blockcount == right.br_startblock &&
1678  new->br_state == right.br_state &&
1679  new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1680  (!(state & BMAP_LEFT_CONTIG) ||
1681  left.br_blockcount + new->br_blockcount +
1682  right.br_blockcount <= MAXEXTLEN))
1683  state |= BMAP_RIGHT_CONTIG;
1684 
1685  error = 0;
1686  /*
1687  * Select which case we're in here, and implement it.
1688  */
1689  switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1690  case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1691  /*
1692  * New allocation is contiguous with real allocations on the
1693  * left and on the right.
1694  * Merge all three into a single extent record.
1695  */
1696  --bma->idx;
1697  trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1698  xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1699  left.br_blockcount + new->br_blockcount +
1700  right.br_blockcount);
1701  trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1702 
1703  xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1704 
1705  XFS_IFORK_NEXT_SET(bma->ip, whichfork,
1706  XFS_IFORK_NEXTENTS(bma->ip, whichfork) - 1);
1707  if (bma->cur == NULL) {
1708  rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1709  } else {
1710  rval = XFS_ILOG_CORE;
1711  error = xfs_bmbt_lookup_eq(bma->cur, right.br_startoff,
1712  right.br_startblock, right.br_blockcount,
1713  &i);
1714  if (error)
1715  goto done;
1716  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1717  error = xfs_btree_delete(bma->cur, &i);
1718  if (error)
1719  goto done;
1720  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1721  error = xfs_btree_decrement(bma->cur, 0, &i);
1722  if (error)
1723  goto done;
1724  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1725  error = xfs_bmbt_update(bma->cur, left.br_startoff,
1726  left.br_startblock,
1727  left.br_blockcount +
1728  new->br_blockcount +
1729  right.br_blockcount,
1730  left.br_state);
1731  if (error)
1732  goto done;
1733  }
1734  break;
1735 
1736  case BMAP_LEFT_CONTIG:
1737  /*
1738  * New allocation is contiguous with a real allocation
1739  * on the left.
1740  * Merge the new allocation with the left neighbor.
1741  */
1742  --bma->idx;
1743  trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1744  xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1745  left.br_blockcount + new->br_blockcount);
1746  trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1747 
1748  if (bma->cur == NULL) {
1749  rval = xfs_ilog_fext(whichfork);
1750  } else {
1751  rval = 0;
1752  error = xfs_bmbt_lookup_eq(bma->cur, left.br_startoff,
1753  left.br_startblock, left.br_blockcount,
1754  &i);
1755  if (error)
1756  goto done;
1757  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1758  error = xfs_bmbt_update(bma->cur, left.br_startoff,
1759  left.br_startblock,
1760  left.br_blockcount +
1761  new->br_blockcount,
1762  left.br_state);
1763  if (error)
1764  goto done;
1765  }
1766  break;
1767 
1768  case BMAP_RIGHT_CONTIG:
1769  /*
1770  * New allocation is contiguous with a real allocation
1771  * on the right.
1772  * Merge the new allocation with the right neighbor.
1773  */
1774  trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1775  xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx),
1776  new->br_startoff, new->br_startblock,
1777  new->br_blockcount + right.br_blockcount,
1778  right.br_state);
1779  trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1780 
1781  if (bma->cur == NULL) {
1782  rval = xfs_ilog_fext(whichfork);
1783  } else {
1784  rval = 0;
1785  error = xfs_bmbt_lookup_eq(bma->cur,
1786  right.br_startoff,
1787  right.br_startblock,
1788  right.br_blockcount, &i);
1789  if (error)
1790  goto done;
1791  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1792  error = xfs_bmbt_update(bma->cur, new->br_startoff,
1793  new->br_startblock,
1794  new->br_blockcount +
1795  right.br_blockcount,
1796  right.br_state);
1797  if (error)
1798  goto done;
1799  }
1800  break;
1801 
1802  case 0:
1803  /*
1804  * New allocation is not contiguous with another
1805  * real allocation.
1806  * Insert a new entry.
1807  */
1808  xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
1809  XFS_IFORK_NEXT_SET(bma->ip, whichfork,
1810  XFS_IFORK_NEXTENTS(bma->ip, whichfork) + 1);
1811  if (bma->cur == NULL) {
1812  rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1813  } else {
1814  rval = XFS_ILOG_CORE;
1815  error = xfs_bmbt_lookup_eq(bma->cur,
1816  new->br_startoff,
1817  new->br_startblock,
1818  new->br_blockcount, &i);
1819  if (error)
1820  goto done;
1821  XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1822  bma->cur->bc_rec.b.br_state = new->br_state;
1823  error = xfs_btree_insert(bma->cur, &i);
1824  if (error)
1825  goto done;
1826  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1827  }
1828  break;
1829  }
1830 
1831  /* convert to a btree if necessary */
1832  if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1833  int tmp_logflags; /* partial log flag return val */
1834 
1835  ASSERT(bma->cur == NULL);
1836  error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1837  bma->firstblock, bma->flist, &bma->cur,
1838  0, &tmp_logflags, whichfork);
1839  bma->logflags |= tmp_logflags;
1840  if (error)
1841  goto done;
1842  }
1843 
1844  /* clear out the allocated field, done with it now in any case. */
1845  if (bma->cur)
1846  bma->cur->bc_private.b.allocated = 0;
1847 
1848  xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
1849 done:
1850  bma->logflags |= rval;
1851  return error;
1852 }
1853 
1854 /*
1855  * Adjust the size of the new extent based on di_extsize and rt extsize.
1856  */
1857 STATIC int
1858 xfs_bmap_extsize_align(
1859  xfs_mount_t *mp,
1860  xfs_bmbt_irec_t *gotp, /* next extent pointer */
1861  xfs_bmbt_irec_t *prevp, /* previous extent pointer */
1862  xfs_extlen_t extsz, /* align to this extent size */
1863  int rt, /* is this a realtime inode? */
1864  int eof, /* is extent at end-of-file? */
1865  int delay, /* creating delalloc extent? */
1866  int convert, /* overwriting unwritten extent? */
1867  xfs_fileoff_t *offp, /* in/out: aligned offset */
1868  xfs_extlen_t *lenp) /* in/out: aligned length */
1869 {
1870  xfs_fileoff_t orig_off; /* original offset */
1871  xfs_extlen_t orig_alen; /* original length */
1872  xfs_fileoff_t orig_end; /* original off+len */
1873  xfs_fileoff_t nexto; /* next file offset */
1874  xfs_fileoff_t prevo; /* previous file offset */
1875  xfs_fileoff_t align_off; /* temp for offset */
1876  xfs_extlen_t align_alen; /* temp for length */
1877  xfs_extlen_t temp; /* temp for calculations */
1878 
1879  if (convert)
1880  return 0;
1881 
1882  orig_off = align_off = *offp;
1883  orig_alen = align_alen = *lenp;
1884  orig_end = orig_off + orig_alen;
1885 
1886  /*
1887  * If this request overlaps an existing extent, then don't
1888  * attempt to perform any additional alignment.
1889  */
1890  if (!delay && !eof &&
1891  (orig_off >= gotp->br_startoff) &&
1892  (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
1893  return 0;
1894  }
1895 
1896  /*
1897  * If the file offset is unaligned vs. the extent size
1898  * we need to align it. This will be possible unless
1899  * the file was previously written with a kernel that didn't
1900  * perform this alignment, or if a truncate shot us in the
1901  * foot.
1902  */
1903  temp = do_mod(orig_off, extsz);
1904  if (temp) {
1905  align_alen += temp;
1906  align_off -= temp;
1907  }
1908  /*
1909  * Same adjustment for the end of the requested area.
1910  */
1911  if ((temp = (align_alen % extsz))) {
1912  align_alen += extsz - temp;
1913  }
1914  /*
1915  * If the previous block overlaps with this proposed allocation
1916  * then move the start forward without adjusting the length.
1917  */
1918  if (prevp->br_startoff != NULLFILEOFF) {
1919  if (prevp->br_startblock == HOLESTARTBLOCK)
1920  prevo = prevp->br_startoff;
1921  else
1922  prevo = prevp->br_startoff + prevp->br_blockcount;
1923  } else
1924  prevo = 0;
1925  if (align_off != orig_off && align_off < prevo)
1926  align_off = prevo;
1927  /*
1928  * If the next block overlaps with this proposed allocation
1929  * then move the start back without adjusting the length,
1930  * but not before offset 0.
1931  * This may of course make the start overlap previous block,
1932  * and if we hit the offset 0 limit then the next block
1933  * can still overlap too.
1934  */
1935  if (!eof && gotp->br_startoff != NULLFILEOFF) {
1936  if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
1937  (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
1938  nexto = gotp->br_startoff + gotp->br_blockcount;
1939  else
1940  nexto = gotp->br_startoff;
1941  } else
1942  nexto = NULLFILEOFF;
1943  if (!eof &&
1944  align_off + align_alen != orig_end &&
1945  align_off + align_alen > nexto)
1946  align_off = nexto > align_alen ? nexto - align_alen : 0;
1947  /*
1948  * If we're now overlapping the next or previous extent that
1949  * means we can't fit an extsz piece in this hole. Just move
1950  * the start forward to the first valid spot and set
1951  * the length so we hit the end.
1952  */
1953  if (align_off != orig_off && align_off < prevo)
1954  align_off = prevo;
1955  if (align_off + align_alen != orig_end &&
1956  align_off + align_alen > nexto &&
1957  nexto != NULLFILEOFF) {
1958  ASSERT(nexto > prevo);
1959  align_alen = nexto - align_off;
1960  }
1961 
1962  /*
1963  * If realtime, and the result isn't a multiple of the realtime
1964  * extent size we need to remove blocks until it is.
1965  */
1966  if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
1967  /*
1968  * We're not covering the original request, or
1969  * we won't be able to once we fix the length.
1970  */
1971  if (orig_off < align_off ||
1972  orig_end > align_off + align_alen ||
1973  align_alen - temp < orig_alen)
1974  return XFS_ERROR(EINVAL);
1975  /*
1976  * Try to fix it by moving the start up.
1977  */
1978  if (align_off + temp <= orig_off) {
1979  align_alen -= temp;
1980  align_off += temp;
1981  }
1982  /*
1983  * Try to fix it by moving the end in.
1984  */
1985  else if (align_off + align_alen - temp >= orig_end)
1986  align_alen -= temp;
1987  /*
1988  * Set the start to the minimum then trim the length.
1989  */
1990  else {
1991  align_alen -= orig_off - align_off;
1992  align_off = orig_off;
1993  align_alen -= align_alen % mp->m_sb.sb_rextsize;
1994  }
1995  /*
1996  * Result doesn't cover the request, fail it.
1997  */
1998  if (orig_off < align_off || orig_end > align_off + align_alen)
1999  return XFS_ERROR(EINVAL);
2000  } else {
2001  ASSERT(orig_off >= align_off);
2002  ASSERT(orig_end <= align_off + align_alen);
2003  }
2004 
2005 #ifdef DEBUG
2006  if (!eof && gotp->br_startoff != NULLFILEOFF)
2007  ASSERT(align_off + align_alen <= gotp->br_startoff);
2008  if (prevp->br_startoff != NULLFILEOFF)
2009  ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
2010 #endif
2011 
2012  *lenp = align_alen;
2013  *offp = align_off;
2014  return 0;
2015 }
2016 
2017 #define XFS_ALLOC_GAP_UNITS 4
2018 
2019 STATIC void
2020 xfs_bmap_adjacent(
2021  xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2022 {
2023  xfs_fsblock_t adjust; /* adjustment to block numbers */
2024  xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
2025  xfs_mount_t *mp; /* mount point structure */
2026  int nullfb; /* true if ap->firstblock isn't set */
2027  int rt; /* true if inode is realtime */
2028 
2029 #define ISVALID(x,y) \
2030  (rt ? \
2031  (x) < mp->m_sb.sb_rblocks : \
2032  XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2033  XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2034  XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2035 
2036  mp = ap->ip->i_mount;
2037  nullfb = *ap->firstblock == NULLFSBLOCK;
2038  rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2039  fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
2040  /*
2041  * If allocating at eof, and there's a previous real block,
2042  * try to use its last block as our starting point.
2043  */
2044  if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
2045  !isnullstartblock(ap->prev.br_startblock) &&
2046  ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
2047  ap->prev.br_startblock)) {
2048  ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
2049  /*
2050  * Adjust for the gap between prevp and us.
2051  */
2052  adjust = ap->offset -
2053  (ap->prev.br_startoff + ap->prev.br_blockcount);
2054  if (adjust &&
2055  ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
2056  ap->blkno += adjust;
2057  }
2058  /*
2059  * If not at eof, then compare the two neighbor blocks.
2060  * Figure out whether either one gives us a good starting point,
2061  * and pick the better one.
2062  */
2063  else if (!ap->eof) {
2064  xfs_fsblock_t gotbno; /* right side block number */
2065  xfs_fsblock_t gotdiff=0; /* right side difference */
2066  xfs_fsblock_t prevbno; /* left side block number */
2067  xfs_fsblock_t prevdiff=0; /* left side difference */
2068 
2069  /*
2070  * If there's a previous (left) block, select a requested
2071  * start block based on it.
2072  */
2073  if (ap->prev.br_startoff != NULLFILEOFF &&
2074  !isnullstartblock(ap->prev.br_startblock) &&
2075  (prevbno = ap->prev.br_startblock +
2076  ap->prev.br_blockcount) &&
2077  ISVALID(prevbno, ap->prev.br_startblock)) {
2078  /*
2079  * Calculate gap to end of previous block.
2080  */
2081  adjust = prevdiff = ap->offset -
2082  (ap->prev.br_startoff +
2083  ap->prev.br_blockcount);
2084  /*
2085  * Figure the startblock based on the previous block's
2086  * end and the gap size.
2087  * Heuristic!
2088  * If the gap is large relative to the piece we're
2089  * allocating, or using it gives us an invalid block
2090  * number, then just use the end of the previous block.
2091  */
2092  if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
2093  ISVALID(prevbno + prevdiff,
2094  ap->prev.br_startblock))
2095  prevbno += adjust;
2096  else
2097  prevdiff += adjust;
2098  /*
2099  * If the firstblock forbids it, can't use it,
2100  * must use default.
2101  */
2102  if (!rt && !nullfb &&
2103  XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2104  prevbno = NULLFSBLOCK;
2105  }
2106  /*
2107  * No previous block or can't follow it, just default.
2108  */
2109  else
2110  prevbno = NULLFSBLOCK;
2111  /*
2112  * If there's a following (right) block, select a requested
2113  * start block based on it.
2114  */
2115  if (!isnullstartblock(ap->got.br_startblock)) {
2116  /*
2117  * Calculate gap to start of next block.
2118  */
2119  adjust = gotdiff = ap->got.br_startoff - ap->offset;
2120  /*
2121  * Figure the startblock based on the next block's
2122  * start and the gap size.
2123  */
2124  gotbno = ap->got.br_startblock;
2125  /*
2126  * Heuristic!
2127  * If the gap is large relative to the piece we're
2128  * allocating, or using it gives us an invalid block
2129  * number, then just use the start of the next block
2130  * offset by our length.
2131  */
2132  if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
2133  ISVALID(gotbno - gotdiff, gotbno))
2134  gotbno -= adjust;
2135  else if (ISVALID(gotbno - ap->length, gotbno)) {
2136  gotbno -= ap->length;
2137  gotdiff += adjust - ap->length;
2138  } else
2139  gotdiff += adjust;
2140  /*
2141  * If the firstblock forbids it, can't use it,
2142  * must use default.
2143  */
2144  if (!rt && !nullfb &&
2145  XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2146  gotbno = NULLFSBLOCK;
2147  }
2148  /*
2149  * No next block, just default.
2150  */
2151  else
2152  gotbno = NULLFSBLOCK;
2153  /*
2154  * If both valid, pick the better one, else the only good
2155  * one, else ap->blkno is already set (to 0 or the inode block).
2156  */
2157  if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2158  ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
2159  else if (prevbno != NULLFSBLOCK)
2160  ap->blkno = prevbno;
2161  else if (gotbno != NULLFSBLOCK)
2162  ap->blkno = gotbno;
2163  }
2164 #undef ISVALID
2165 }
2166 
2167 STATIC int
2168 xfs_bmap_rtalloc(
2169  xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2170 {
2171  xfs_alloctype_t atype = 0; /* type for allocation routines */
2172  int error; /* error return value */
2173  xfs_mount_t *mp; /* mount point structure */
2174  xfs_extlen_t prod = 0; /* product factor for allocators */
2175  xfs_extlen_t ralen = 0; /* realtime allocation length */
2176  xfs_extlen_t align; /* minimum allocation alignment */
2177  xfs_rtblock_t rtb;
2178 
2179  mp = ap->ip->i_mount;
2180  align = xfs_get_extsz_hint(ap->ip);
2181  prod = align / mp->m_sb.sb_rextsize;
2182  error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
2183  align, 1, ap->eof, 0,
2184  ap->conv, &ap->offset, &ap->length);
2185  if (error)
2186  return error;
2187  ASSERT(ap->length);
2188  ASSERT(ap->length % mp->m_sb.sb_rextsize == 0);
2189 
2190  /*
2191  * If the offset & length are not perfectly aligned
2192  * then kill prod, it will just get us in trouble.
2193  */
2194  if (do_mod(ap->offset, align) || ap->length % align)
2195  prod = 1;
2196  /*
2197  * Set ralen to be the actual requested length in rtextents.
2198  */
2199  ralen = ap->length / mp->m_sb.sb_rextsize;
2200  /*
2201  * If the old value was close enough to MAXEXTLEN that
2202  * we rounded up to it, cut it back so it's valid again.
2203  * Note that if it's a really large request (bigger than
2204  * MAXEXTLEN), we don't hear about that number, and can't
2205  * adjust the starting point to match it.
2206  */
2207  if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2208  ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2209 
2210  /*
2211  * Lock out other modifications to the RT bitmap inode.
2212  */
2213  xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
2214  xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
2215 
2216  /*
2217  * If it's an allocation to an empty file at offset 0,
2218  * pick an extent that will space things out in the rt area.
2219  */
2220  if (ap->eof && ap->offset == 0) {
2221  xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
2222 
2223  error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2224  if (error)
2225  return error;
2226  ap->blkno = rtx * mp->m_sb.sb_rextsize;
2227  } else {
2228  ap->blkno = 0;
2229  }
2230 
2231  xfs_bmap_adjacent(ap);
2232 
2233  /*
2234  * Realtime allocation, done through xfs_rtallocate_extent.
2235  */
2236  atype = ap->blkno == 0 ? XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2237  do_div(ap->blkno, mp->m_sb.sb_rextsize);
2238  rtb = ap->blkno;
2239  ap->length = ralen;
2240  if ((error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length,
2241  &ralen, atype, ap->wasdel, prod, &rtb)))
2242  return error;
2243  if (rtb == NULLFSBLOCK && prod > 1 &&
2244  (error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1,
2245  ap->length, &ralen, atype,
2246  ap->wasdel, 1, &rtb)))
2247  return error;
2248  ap->blkno = rtb;
2249  if (ap->blkno != NULLFSBLOCK) {
2250  ap->blkno *= mp->m_sb.sb_rextsize;
2251  ralen *= mp->m_sb.sb_rextsize;
2252  ap->length = ralen;
2253  ap->ip->i_d.di_nblocks += ralen;
2254  xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2255  if (ap->wasdel)
2256  ap->ip->i_delayed_blks -= ralen;
2257  /*
2258  * Adjust the disk quota also. This was reserved
2259  * earlier.
2260  */
2261  xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2262  ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2263  XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2264  } else {
2265  ap->length = 0;
2266  }
2267  return 0;
2268 }
2269 
2270 STATIC int
2271 xfs_bmap_btalloc_nullfb(
2272  struct xfs_bmalloca *ap,
2273  struct xfs_alloc_arg *args,
2274  xfs_extlen_t *blen)
2275 {
2276  struct xfs_mount *mp = ap->ip->i_mount;
2277  struct xfs_perag *pag;
2278  xfs_agnumber_t ag, startag;
2279  int notinit = 0;
2280  int error;
2281 
2282  if (ap->userdata && xfs_inode_is_filestream(ap->ip))
2283  args->type = XFS_ALLOCTYPE_NEAR_BNO;
2284  else
2285  args->type = XFS_ALLOCTYPE_START_BNO;
2286  args->total = ap->total;
2287 
2288  /*
2289  * Search for an allocation group with a single extent large enough
2290  * for the request. If one isn't found, then adjust the minimum
2291  * allocation size to the largest space found.
2292  */
2293  startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
2294  if (startag == NULLAGNUMBER)
2295  startag = ag = 0;
2296 
2297  pag = xfs_perag_get(mp, ag);
2298  while (*blen < args->maxlen) {
2299  if (!pag->pagf_init) {
2300  error = xfs_alloc_pagf_init(mp, args->tp, ag,
2301  XFS_ALLOC_FLAG_TRYLOCK);
2302  if (error) {
2303  xfs_perag_put(pag);
2304  return error;
2305  }
2306  }
2307 
2308  /*
2309  * See xfs_alloc_fix_freelist...
2310  */
2311  if (pag->pagf_init) {
2312  xfs_extlen_t longest;
2313  longest = xfs_alloc_longest_free_extent(mp, pag);
2314  if (*blen < longest)
2315  *blen = longest;
2316  } else
2317  notinit = 1;
2318 
2319  if (xfs_inode_is_filestream(ap->ip)) {
2320  if (*blen >= args->maxlen)
2321  break;
2322 
2323  if (ap->userdata) {
2324  /*
2325  * If startag is an invalid AG, we've
2326  * come here once before and
2327  * xfs_filestream_new_ag picked the
2328  * best currently available.
2329  *
2330  * Don't continue looping, since we
2331  * could loop forever.
2332  */
2333  if (startag == NULLAGNUMBER)
2334  break;
2335 
2336  error = xfs_filestream_new_ag(ap, &ag);
2337  xfs_perag_put(pag);
2338  if (error)
2339  return error;
2340 
2341  /* loop again to set 'blen'*/
2342  startag = NULLAGNUMBER;
2343  pag = xfs_perag_get(mp, ag);
2344  continue;
2345  }
2346  }
2347  if (++ag == mp->m_sb.sb_agcount)
2348  ag = 0;
2349  if (ag == startag)
2350  break;
2351  xfs_perag_put(pag);
2352  pag = xfs_perag_get(mp, ag);
2353  }
2354  xfs_perag_put(pag);
2355 
2356  /*
2357  * Since the above loop did a BUF_TRYLOCK, it is
2358  * possible that there is space for this request.
2359  */
2360  if (notinit || *blen < ap->minlen)
2361  args->minlen = ap->minlen;
2362  /*
2363  * If the best seen length is less than the request
2364  * length, use the best as the minimum.
2365  */
2366  else if (*blen < args->maxlen)
2367  args->minlen = *blen;
2368  /*
2369  * Otherwise we've seen an extent as big as maxlen,
2370  * use that as the minimum.
2371  */
2372  else
2373  args->minlen = args->maxlen;
2374 
2375  /*
2376  * set the failure fallback case to look in the selected
2377  * AG as the stream may have moved.
2378  */
2379  if (xfs_inode_is_filestream(ap->ip))
2380  ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
2381 
2382  return 0;
2383 }
2384 
2385 STATIC int
2386 xfs_bmap_btalloc(
2387  xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2388 {
2389  xfs_mount_t *mp; /* mount point structure */
2390  xfs_alloctype_t atype = 0; /* type for allocation routines */
2391  xfs_extlen_t align; /* minimum allocation alignment */
2392  xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
2393  xfs_agnumber_t ag;
2394  xfs_alloc_arg_t args;
2395  xfs_extlen_t blen;
2396  xfs_extlen_t nextminlen = 0;
2397  int nullfb; /* true if ap->firstblock isn't set */
2398  int isaligned;
2399  int tryagain;
2400  int error;
2401 
2402  ASSERT(ap->length);
2403 
2404  mp = ap->ip->i_mount;
2405  align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
2406  if (unlikely(align)) {
2407  error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
2408  align, 0, ap->eof, 0, ap->conv,
2409  &ap->offset, &ap->length);
2410  ASSERT(!error);
2411  ASSERT(ap->length);
2412  }
2413  nullfb = *ap->firstblock == NULLFSBLOCK;
2414  fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
2415  if (nullfb) {
2416  if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
2417  ag = xfs_filestream_lookup_ag(ap->ip);
2418  ag = (ag != NULLAGNUMBER) ? ag : 0;
2419  ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
2420  } else {
2421  ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2422  }
2423  } else
2424  ap->blkno = *ap->firstblock;
2425 
2426  xfs_bmap_adjacent(ap);
2427 
2428  /*
2429  * If allowed, use ap->blkno; otherwise must use firstblock since
2430  * it's in the right allocation group.
2431  */
2432  if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
2433  ;
2434  else
2435  ap->blkno = *ap->firstblock;
2436  /*
2437  * Normal allocation, done through xfs_alloc_vextent.
2438  */
2439  tryagain = isaligned = 0;
2440  memset(&args, 0, sizeof(args));
2441  args.tp = ap->tp;
2442  args.mp = mp;
2443  args.fsbno = ap->blkno;
2444 
2445  /* Trim the allocation back to the maximum an AG can fit. */
2446  args.maxlen = MIN(ap->length, XFS_ALLOC_AG_MAX_USABLE(mp));
2447  args.firstblock = *ap->firstblock;
2448  blen = 0;
2449  if (nullfb) {
2450  error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
2451  if (error)
2452  return error;
2453  } else if (ap->flist->xbf_low) {
2454  if (xfs_inode_is_filestream(ap->ip))
2455  args.type = XFS_ALLOCTYPE_FIRST_AG;
2456  else
2457  args.type = XFS_ALLOCTYPE_START_BNO;
2458  args.total = args.minlen = ap->minlen;
2459  } else {
2460  args.type = XFS_ALLOCTYPE_NEAR_BNO;
2461  args.total = ap->total;
2462  args.minlen = ap->minlen;
2463  }
2464  /* apply extent size hints if obtained earlier */
2465  if (unlikely(align)) {
2466  args.prod = align;
2467  if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
2468  args.mod = (xfs_extlen_t)(args.prod - args.mod);
2469  } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
2470  args.prod = 1;
2471  args.mod = 0;
2472  } else {
2473  args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
2474  if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
2475  args.mod = (xfs_extlen_t)(args.prod - args.mod);
2476  }
2477  /*
2478  * If we are not low on available data blocks, and the
2479  * underlying logical volume manager is a stripe, and
2480  * the file offset is zero then try to allocate data
2481  * blocks on stripe unit boundary.
2482  * NOTE: ap->aeof is only set if the allocation length
2483  * is >= the stripe unit and the allocation offset is
2484  * at the end of file.
2485  */
2486  if (!ap->flist->xbf_low && ap->aeof) {
2487  if (!ap->offset) {
2488  args.alignment = mp->m_dalign;
2489  atype = args.type;
2490  isaligned = 1;
2491  /*
2492  * Adjust for alignment
2493  */
2494  if (blen > args.alignment && blen <= args.maxlen)
2495  args.minlen = blen - args.alignment;
2496  args.minalignslop = 0;
2497  } else {
2498  /*
2499  * First try an exact bno allocation.
2500  * If it fails then do a near or start bno
2501  * allocation with alignment turned on.
2502  */
2503  atype = args.type;
2504  tryagain = 1;
2505  args.type = XFS_ALLOCTYPE_THIS_BNO;
2506  args.alignment = 1;
2507  /*
2508  * Compute the minlen+alignment for the
2509  * next case. Set slop so that the value
2510  * of minlen+alignment+slop doesn't go up
2511  * between the calls.
2512  */
2513  if (blen > mp->m_dalign && blen <= args.maxlen)
2514  nextminlen = blen - mp->m_dalign;
2515  else
2516  nextminlen = args.minlen;
2517  if (nextminlen + mp->m_dalign > args.minlen + 1)
2518  args.minalignslop =
2519  nextminlen + mp->m_dalign -
2520  args.minlen - 1;
2521  else
2522  args.minalignslop = 0;
2523  }
2524  } else {
2525  args.alignment = 1;
2526  args.minalignslop = 0;
2527  }
2528  args.minleft = ap->minleft;
2529  args.wasdel = ap->wasdel;
2530  args.isfl = 0;
2531  args.userdata = ap->userdata;
2532  if ((error = xfs_alloc_vextent(&args)))
2533  return error;
2534  if (tryagain && args.fsbno == NULLFSBLOCK) {
2535  /*
2536  * Exact allocation failed. Now try with alignment
2537  * turned on.
2538  */
2539  args.type = atype;
2540  args.fsbno = ap->blkno;
2541  args.alignment = mp->m_dalign;
2542  args.minlen = nextminlen;
2543  args.minalignslop = 0;
2544  isaligned = 1;
2545  if ((error = xfs_alloc_vextent(&args)))
2546  return error;
2547  }
2548  if (isaligned && args.fsbno == NULLFSBLOCK) {
2549  /*
2550  * allocation failed, so turn off alignment and
2551  * try again.
2552  */
2553  args.type = atype;
2554  args.fsbno = ap->blkno;
2555  args.alignment = 0;
2556  if ((error = xfs_alloc_vextent(&args)))
2557  return error;
2558  }
2559  if (args.fsbno == NULLFSBLOCK && nullfb &&
2560  args.minlen > ap->minlen) {
2561  args.minlen = ap->minlen;
2562  args.type = XFS_ALLOCTYPE_START_BNO;
2563  args.fsbno = ap->blkno;
2564  if ((error = xfs_alloc_vextent(&args)))
2565  return error;
2566  }
2567  if (args.fsbno == NULLFSBLOCK && nullfb) {
2568  args.fsbno = 0;
2569  args.type = XFS_ALLOCTYPE_FIRST_AG;
2570  args.total = ap->minlen;
2571  args.minleft = 0;
2572  if ((error = xfs_alloc_vextent(&args)))
2573  return error;
2574  ap->flist->xbf_low = 1;
2575  }
2576  if (args.fsbno != NULLFSBLOCK) {
2577  /*
2578  * check the allocation happened at the same or higher AG than
2579  * the first block that was allocated.
2580  */
2581  ASSERT(*ap->firstblock == NULLFSBLOCK ||
2582  XFS_FSB_TO_AGNO(mp, *ap->firstblock) ==
2583  XFS_FSB_TO_AGNO(mp, args.fsbno) ||
2584  (ap->flist->xbf_low &&
2585  XFS_FSB_TO_AGNO(mp, *ap->firstblock) <
2586  XFS_FSB_TO_AGNO(mp, args.fsbno)));
2587 
2588  ap->blkno = args.fsbno;
2589  if (*ap->firstblock == NULLFSBLOCK)
2590  *ap->firstblock = args.fsbno;
2591  ASSERT(nullfb || fb_agno == args.agno ||
2592  (ap->flist->xbf_low && fb_agno < args.agno));
2593  ap->length = args.len;
2594  ap->ip->i_d.di_nblocks += args.len;
2595  xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2596  if (ap->wasdel)
2597  ap->ip->i_delayed_blks -= args.len;
2598  /*
2599  * Adjust the disk quota also. This was reserved
2600  * earlier.
2601  */
2602  xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2603  ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2604  XFS_TRANS_DQ_BCOUNT,
2605  (long) args.len);
2606  } else {
2607  ap->blkno = NULLFSBLOCK;
2608  ap->length = 0;
2609  }
2610  return 0;
2611 }
2612 
2613 /*
2614  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2615  * It figures out where to ask the underlying allocator to put the new extent.
2616  */
2617 STATIC int
2618 xfs_bmap_alloc(
2619  xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2620 {
2621  if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
2622  return xfs_bmap_rtalloc(ap);
2623  return xfs_bmap_btalloc(ap);
2624 }
2625 
2626 /*
2627  * Transform a btree format file with only one leaf node, where the
2628  * extents list will fit in the inode, into an extents format file.
2629  * Since the file extents are already in-core, all we have to do is
2630  * give up the space for the btree root and pitch the leaf block.
2631  */
2632 STATIC int /* error */
2633 xfs_bmap_btree_to_extents(
2634  xfs_trans_t *tp, /* transaction pointer */
2635  xfs_inode_t *ip, /* incore inode pointer */
2636  xfs_btree_cur_t *cur, /* btree cursor */
2637  int *logflagsp, /* inode logging flags */
2638  int whichfork) /* data or attr fork */
2639 {
2640  /* REFERENCED */
2641  struct xfs_btree_block *cblock;/* child btree block */
2642  xfs_fsblock_t cbno; /* child block number */
2643  xfs_buf_t *cbp; /* child block's buffer */
2644  int error; /* error return value */
2645  xfs_ifork_t *ifp; /* inode fork data */
2646  xfs_mount_t *mp; /* mount point structure */
2647  __be64 *pp; /* ptr to block address */
2648  struct xfs_btree_block *rblock;/* root btree block */
2649 
2650  mp = ip->i_mount;
2651  ifp = XFS_IFORK_PTR(ip, whichfork);
2652  ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2653  ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
2654  rblock = ifp->if_broot;
2655  ASSERT(be16_to_cpu(rblock->bb_level) == 1);
2656  ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
2657  ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
2658  pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
2659  cbno = be64_to_cpu(*pp);
2660  *logflagsp = 0;
2661 #ifdef DEBUG
2662  if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
2663  return error;
2664 #endif
2665  if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
2666  XFS_BMAP_BTREE_REF)))
2667  return error;
2668  cblock = XFS_BUF_TO_BLOCK(cbp);
2669  if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
2670  return error;
2671  xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
2672  ip->i_d.di_nblocks--;
2673  xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
2674  xfs_trans_binval(tp, cbp);
2675  if (cur->bc_bufs[0] == cbp)
2676  cur->bc_bufs[0] = NULL;
2677  xfs_iroot_realloc(ip, -1, whichfork);
2678  ASSERT(ifp->if_broot == NULL);
2679  ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
2680  XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
2681  *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2682  return 0;
2683 }
2684 
2685 /*
2686  * Called by xfs_bmapi to update file extent records and the btree
2687  * after removing space (or undoing a delayed allocation).
2688  */
2689 STATIC int /* error */
2690 xfs_bmap_del_extent(
2691  xfs_inode_t *ip, /* incore inode pointer */
2692  xfs_trans_t *tp, /* current transaction pointer */
2693  xfs_extnum_t *idx, /* extent number to update/delete */
2694  xfs_bmap_free_t *flist, /* list of extents to be freed */
2695  xfs_btree_cur_t *cur, /* if null, not a btree */
2696  xfs_bmbt_irec_t *del, /* data to remove from extents */
2697  int *logflagsp, /* inode logging flags */
2698  int whichfork) /* data or attr fork */
2699 {
2700  xfs_filblks_t da_new; /* new delay-alloc indirect blocks */
2701  xfs_filblks_t da_old; /* old delay-alloc indirect blocks */
2702  xfs_fsblock_t del_endblock=0; /* first block past del */
2703  xfs_fileoff_t del_endoff; /* first offset past del */
2704  int delay; /* current block is delayed allocated */
2705  int do_fx; /* free extent at end of routine */
2706  xfs_bmbt_rec_host_t *ep; /* current extent entry pointer */
2707  int error; /* error return value */
2708  int flags; /* inode logging flags */
2709  xfs_bmbt_irec_t got; /* current extent entry */
2710  xfs_fileoff_t got_endoff; /* first offset past got */
2711  int i; /* temp state */
2712  xfs_ifork_t *ifp; /* inode fork pointer */
2713  xfs_mount_t *mp; /* mount structure */
2714  xfs_filblks_t nblks; /* quota/sb block count */
2715  xfs_bmbt_irec_t new; /* new record to be inserted */
2716  /* REFERENCED */
2717  uint qfield; /* quota field to update */
2718  xfs_filblks_t temp; /* for indirect length calculations */
2719  xfs_filblks_t temp2; /* for indirect length calculations */
2720  int state = 0;
2721 
2722  XFS_STATS_INC(xs_del_exlist);
2723 
2724  if (whichfork == XFS_ATTR_FORK)
2725  state |= BMAP_ATTRFORK;
2726 
2727  mp = ip->i_mount;
2728  ifp = XFS_IFORK_PTR(ip, whichfork);
2729  ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
2730  (uint)sizeof(xfs_bmbt_rec_t)));
2731  ASSERT(del->br_blockcount > 0);
2732  ep = xfs_iext_get_ext(ifp, *idx);
2733  xfs_bmbt_get_all(ep, &got);
2734  ASSERT(got.br_startoff <= del->br_startoff);
2735  del_endoff = del->br_startoff + del->br_blockcount;
2736  got_endoff = got.br_startoff + got.br_blockcount;
2737  ASSERT(got_endoff >= del_endoff);
2738  delay = isnullstartblock(got.br_startblock);
2739  ASSERT(isnullstartblock(del->br_startblock) == delay);
2740  flags = 0;
2741  qfield = 0;
2742  error = 0;
2743  /*
2744  * If deleting a real allocation, must free up the disk space.
2745  */
2746  if (!delay) {
2747  flags = XFS_ILOG_CORE;
2748  /*
2749  * Realtime allocation. Free it and record di_nblocks update.
2750  */
2751  if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
2752  xfs_fsblock_t bno;
2753  xfs_filblks_t len;
2754 
2755  ASSERT(do_mod(del->br_blockcount,
2756  mp->m_sb.sb_rextsize) == 0);
2757  ASSERT(do_mod(del->br_startblock,
2758  mp->m_sb.sb_rextsize) == 0);
2759  bno = del->br_startblock;
2760  len = del->br_blockcount;
2761  do_div(bno, mp->m_sb.sb_rextsize);
2762  do_div(len, mp->m_sb.sb_rextsize);
2763  error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
2764  if (error)
2765  goto done;
2766  do_fx = 0;
2767  nblks = len * mp->m_sb.sb_rextsize;
2768  qfield = XFS_TRANS_DQ_RTBCOUNT;
2769  }
2770  /*
2771  * Ordinary allocation.
2772  */
2773  else {
2774  do_fx = 1;
2775  nblks = del->br_blockcount;
2776  qfield = XFS_TRANS_DQ_BCOUNT;
2777  }
2778  /*
2779  * Set up del_endblock and cur for later.
2780  */
2781  del_endblock = del->br_startblock + del->br_blockcount;
2782  if (cur) {
2783  if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
2784  got.br_startblock, got.br_blockcount,
2785  &i)))
2786  goto done;
2787  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2788  }
2789  da_old = da_new = 0;
2790  } else {
2791  da_old = startblockval(got.br_startblock);
2792  da_new = 0;
2793  nblks = 0;
2794  do_fx = 0;
2795  }
2796  /*
2797  * Set flag value to use in switch statement.
2798  * Left-contig is 2, right-contig is 1.
2799  */
2800  switch (((got.br_startoff == del->br_startoff) << 1) |
2801  (got_endoff == del_endoff)) {
2802  case 3:
2803  /*
2804  * Matches the whole extent. Delete the entry.
2805  */
2806  xfs_iext_remove(ip, *idx, 1,
2807  whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
2808  --*idx;
2809  if (delay)
2810  break;
2811 
2812  XFS_IFORK_NEXT_SET(ip, whichfork,
2813  XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2814  flags |= XFS_ILOG_CORE;
2815  if (!cur) {
2816  flags |= xfs_ilog_fext(whichfork);
2817  break;
2818  }
2819  if ((error = xfs_btree_delete(cur, &i)))
2820  goto done;
2821  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2822  break;
2823 
2824  case 2:
2825  /*
2826  * Deleting the first part of the extent.
2827  */
2828  trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2829  xfs_bmbt_set_startoff(ep, del_endoff);
2830  temp = got.br_blockcount - del->br_blockcount;
2831  xfs_bmbt_set_blockcount(ep, temp);
2832  if (delay) {
2833  temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2834  da_old);
2835  xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2836  trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2837  da_new = temp;
2838  break;
2839  }
2840  xfs_bmbt_set_startblock(ep, del_endblock);
2841  trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2842  if (!cur) {
2843  flags |= xfs_ilog_fext(whichfork);
2844  break;
2845  }
2846  if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
2847  got.br_blockcount - del->br_blockcount,
2848  got.br_state)))
2849  goto done;
2850  break;
2851 
2852  case 1:
2853  /*
2854  * Deleting the last part of the extent.
2855  */
2856  temp = got.br_blockcount - del->br_blockcount;
2857  trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2858  xfs_bmbt_set_blockcount(ep, temp);
2859  if (delay) {
2860  temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2861  da_old);
2862  xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2863  trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2864  da_new = temp;
2865  break;
2866  }
2867  trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2868  if (!cur) {
2869  flags |= xfs_ilog_fext(whichfork);
2870  break;
2871  }
2872  if ((error = xfs_bmbt_update(cur, got.br_startoff,
2873  got.br_startblock,
2874  got.br_blockcount - del->br_blockcount,
2875  got.br_state)))
2876  goto done;
2877  break;
2878 
2879  case 0:
2880  /*
2881  * Deleting the middle of the extent.
2882  */
2883  temp = del->br_startoff - got.br_startoff;
2884  trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2885  xfs_bmbt_set_blockcount(ep, temp);
2886  new.br_startoff = del_endoff;
2887  temp2 = got_endoff - del_endoff;
2888  new.br_blockcount = temp2;
2889  new.br_state = got.br_state;
2890  if (!delay) {
2891  new.br_startblock = del_endblock;
2892  flags |= XFS_ILOG_CORE;
2893  if (cur) {
2894  if ((error = xfs_bmbt_update(cur,
2895  got.br_startoff,
2896  got.br_startblock, temp,
2897  got.br_state)))
2898  goto done;
2899  if ((error = xfs_btree_increment(cur, 0, &i)))
2900  goto done;
2901  cur->bc_rec.b = new;
2902  error = xfs_btree_insert(cur, &i);
2903  if (error && error != ENOSPC)
2904  goto done;
2905  /*
2906  * If get no-space back from btree insert,
2907  * it tried a split, and we have a zero
2908  * block reservation.
2909  * Fix up our state and return the error.
2910  */
2911  if (error == ENOSPC) {
2912  /*
2913  * Reset the cursor, don't trust
2914  * it after any insert operation.
2915  */
2916  if ((error = xfs_bmbt_lookup_eq(cur,
2917  got.br_startoff,
2918  got.br_startblock,
2919  temp, &i)))
2920  goto done;
2921  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2922  /*
2923  * Update the btree record back
2924  * to the original value.
2925  */
2926  if ((error = xfs_bmbt_update(cur,
2927  got.br_startoff,
2928  got.br_startblock,
2929  got.br_blockcount,
2930  got.br_state)))
2931  goto done;
2932  /*
2933  * Reset the extent record back
2934  * to the original value.
2935  */
2936  xfs_bmbt_set_blockcount(ep,
2937  got.br_blockcount);
2938  flags = 0;
2939  error = XFS_ERROR(ENOSPC);
2940  goto done;
2941  }
2942  XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2943  } else
2944  flags |= xfs_ilog_fext(whichfork);
2945  XFS_IFORK_NEXT_SET(ip, whichfork,
2946  XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2947  } else {
2948  ASSERT(whichfork == XFS_DATA_FORK);
2949  temp = xfs_bmap_worst_indlen(ip, temp);
2950  xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2951  temp2 = xfs_bmap_worst_indlen(ip, temp2);
2952  new.br_startblock = nullstartblock((int)temp2);
2953  da_new = temp + temp2;
2954  while (da_new > da_old) {
2955  if (temp) {
2956  temp--;
2957  da_new--;
2958  xfs_bmbt_set_startblock(ep,
2959  nullstartblock((int)temp));
2960  }
2961  if (da_new == da_old)
2962  break;
2963  if (temp2) {
2964  temp2--;
2965  da_new--;
2966  new.br_startblock =
2967  nullstartblock((int)temp2);
2968  }
2969  }
2970  }
2971  trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2972  xfs_iext_insert(ip, *idx + 1, 1, &new, state);
2973  ++*idx;
2974  break;
2975  }
2976  /*
2977  * If we need to, add to list of extents to delete.
2978  */
2979  if (do_fx)
2980  xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
2981  mp);
2982  /*
2983  * Adjust inode # blocks in the file.
2984  */
2985  if (nblks)
2986  ip->i_d.di_nblocks -= nblks;
2987  /*
2988  * Adjust quota data.
2989  */
2990  if (qfield)
2991  xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
2992 
2993  /*
2994  * Account for change in delayed indirect blocks.
2995  * Nothing to do for disk quota accounting here.
2996  */
2997  ASSERT(da_old >= da_new);
2998  if (da_old > da_new) {
2999  xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
3000  (int64_t)(da_old - da_new), 0);
3001  }
3002 done:
3003  *logflagsp = flags;
3004  return error;
3005 }
3006 
3007 /*
3008  * Remove the entry "free" from the free item list. Prev points to the
3009  * previous entry, unless "free" is the head of the list.
3010  */
3011 STATIC void
3012 xfs_bmap_del_free(
3013  xfs_bmap_free_t *flist, /* free item list header */
3014  xfs_bmap_free_item_t *prev, /* previous item on list, if any */
3015  xfs_bmap_free_item_t *free) /* list item to be freed */
3016 {
3017  if (prev)
3018  prev->xbfi_next = free->xbfi_next;
3019  else
3020  flist->xbf_first = free->xbfi_next;
3021  flist->xbf_count--;
3022  kmem_zone_free(xfs_bmap_free_item_zone, free);
3023 }
3024 
3025 /*
3026  * Convert an extents-format file into a btree-format file.
3027  * The new file will have a root block (in the inode) and a single child block.
3028  */
3029 STATIC int /* error */
3030 xfs_bmap_extents_to_btree(
3031  xfs_trans_t *tp, /* transaction pointer */
3032  xfs_inode_t *ip, /* incore inode pointer */
3033  xfs_fsblock_t *firstblock, /* first-block-allocated */
3034  xfs_bmap_free_t *flist, /* blocks freed in xaction */
3035  xfs_btree_cur_t **curp, /* cursor returned to caller */
3036  int wasdel, /* converting a delayed alloc */
3037  int *logflagsp, /* inode logging flags */
3038  int whichfork) /* data or attr fork */
3039 {
3040  struct xfs_btree_block *ablock; /* allocated (child) bt block */
3041  xfs_buf_t *abp; /* buffer for ablock */
3042  xfs_alloc_arg_t args; /* allocation arguments */
3043  xfs_bmbt_rec_t *arp; /* child record pointer */
3044  struct xfs_btree_block *block; /* btree root block */
3045  xfs_btree_cur_t *cur; /* bmap btree cursor */
3046  xfs_bmbt_rec_host_t *ep; /* extent record pointer */
3047  int error; /* error return value */
3048  xfs_extnum_t i, cnt; /* extent record index */
3049  xfs_ifork_t *ifp; /* inode fork pointer */
3050  xfs_bmbt_key_t *kp; /* root block key pointer */
3051  xfs_mount_t *mp; /* mount structure */
3052  xfs_extnum_t nextents; /* number of file extents */
3053  xfs_bmbt_ptr_t *pp; /* root block address pointer */
3054 
3055  ifp = XFS_IFORK_PTR(ip, whichfork);
3056  ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3057 
3058  /*
3059  * Make space in the inode incore.
3060  */
3061  xfs_iroot_realloc(ip, 1, whichfork);
3062  ifp->if_flags |= XFS_IFBROOT;
3063 
3064  /*
3065  * Fill in the root.
3066  */
3067  block = ifp->if_broot;
3068  block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3069  block->bb_level = cpu_to_be16(1);
3070  block->bb_numrecs = cpu_to_be16(1);
3071  block->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3072  block->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3073 
3074  /*
3075  * Need a cursor. Can't allocate until bb_level is filled in.
3076  */
3077  mp = ip->i_mount;
3078  cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
3079  cur->bc_private.b.firstblock = *firstblock;
3080  cur->bc_private.b.flist = flist;
3081  cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3082  /*
3083  * Convert to a btree with two levels, one record in root.
3084  */
3085  XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3086  memset(&args, 0, sizeof(args));
3087  args.tp = tp;
3088  args.mp = mp;
3089  args.firstblock = *firstblock;
3090  if (*firstblock == NULLFSBLOCK) {
3091  args.type = XFS_ALLOCTYPE_START_BNO;
3092  args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3093  } else if (flist->xbf_low) {
3094  args.type = XFS_ALLOCTYPE_START_BNO;
3095  args.fsbno = *firstblock;
3096  } else {
3097  args.type = XFS_ALLOCTYPE_NEAR_BNO;
3098  args.fsbno = *firstblock;
3099  }
3100  args.minlen = args.maxlen = args.prod = 1;
3101  args.total = args.minleft = args.alignment = args.mod = args.isfl =
3102  args.minalignslop = 0;
3103  args.wasdel = wasdel;
3104  *logflagsp = 0;
3105  if ((error = xfs_alloc_vextent(&args))) {
3106  xfs_iroot_realloc(ip, -1, whichfork);
3107  xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3108  return error;
3109  }
3110  /*
3111  * Allocation can't fail, the space was reserved.
3112  */
3113  ASSERT(args.fsbno != NULLFSBLOCK);
3114  ASSERT(*firstblock == NULLFSBLOCK ||
3115  args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3116  (flist->xbf_low &&
3117  args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3118  *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3119  cur->bc_private.b.allocated++;
3120  ip->i_d.di_nblocks++;
3121  xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3122  abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3123  /*
3124  * Fill in the child block.
3125  */
3126  ablock = XFS_BUF_TO_BLOCK(abp);
3127  ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3128  ablock->bb_level = 0;
3129  ablock->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3130  ablock->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3131  arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3132  nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3133  for (cnt = i = 0; i < nextents; i++) {
3134  ep = xfs_iext_get_ext(ifp, i);
3135  if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
3136  arp->l0 = cpu_to_be64(ep->l0);
3137  arp->l1 = cpu_to_be64(ep->l1);
3138  arp++; cnt++;
3139  }
3140  }
3141  ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3142  xfs_btree_set_numrecs(ablock, cnt);
3143 
3144  /*
3145  * Fill in the root key and pointer.
3146  */
3147  kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
3148  arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3149  kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
3150  pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
3151  be16_to_cpu(block->bb_level)));
3152  *pp = cpu_to_be64(args.fsbno);
3153 
3154  /*
3155  * Do all this logging at the end so that
3156  * the root is at the right level.
3157  */
3158  xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
3159  xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3160  ASSERT(*curp == NULL);
3161  *curp = cur;
3162  *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
3163  return 0;
3164 }
3165 
3166 /*
3167  * Calculate the default attribute fork offset for newly created inodes.
3168  */
3169 uint
3170 xfs_default_attroffset(
3171  struct xfs_inode *ip)
3172 {
3173  struct xfs_mount *mp = ip->i_mount;
3174  uint offset;
3175 
3176  if (mp->m_sb.sb_inodesize == 256) {
3177  offset = XFS_LITINO(mp) -
3178  XFS_BMDR_SPACE_CALC(MINABTPTRS);
3179  } else {
3180  offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
3181  }
3182 
3183  ASSERT(offset < XFS_LITINO(mp));
3184  return offset;
3185 }
3186 
3187 /*
3188  * Helper routine to reset inode di_forkoff field when switching
3189  * attribute fork from local to extent format - we reset it where
3190  * possible to make space available for inline data fork extents.
3191  */
3192 STATIC void
3193 xfs_bmap_forkoff_reset(
3194  xfs_mount_t *mp,
3195  xfs_inode_t *ip,
3196  int whichfork)
3197 {
3198  if (whichfork == XFS_ATTR_FORK &&
3199  ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
3200  ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
3201  ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
3202  uint dfl_forkoff = xfs_default_attroffset(ip) >> 3;
3203 
3204  if (dfl_forkoff > ip->i_d.di_forkoff)
3205  ip->i_d.di_forkoff = dfl_forkoff;
3206  }
3207 }
3208 
3209 /*
3210  * Convert a local file to an extents file.
3211  * This code is out of bounds for data forks of regular files,
3212  * since the file data needs to get logged so things will stay consistent.
3213  * (The bmap-level manipulations are ok, though).
3214  */
3215 STATIC int /* error */
3216 xfs_bmap_local_to_extents(
3217  xfs_trans_t *tp, /* transaction pointer */
3218  xfs_inode_t *ip, /* incore inode pointer */
3219  xfs_fsblock_t *firstblock, /* first block allocated in xaction */
3220  xfs_extlen_t total, /* total blocks needed by transaction */
3221  int *logflagsp, /* inode logging flags */
3222  int whichfork) /* data or attr fork */
3223 {
3224  int error; /* error return value */
3225  int flags; /* logging flags returned */
3226  xfs_ifork_t *ifp; /* inode fork pointer */
3227 
3228  /*
3229  * We don't want to deal with the case of keeping inode data inline yet.
3230  * So sending the data fork of a regular inode is invalid.
3231  */
3232  ASSERT(!(S_ISREG(ip->i_d.di_mode) && whichfork == XFS_DATA_FORK));
3233  ifp = XFS_IFORK_PTR(ip, whichfork);
3234  ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3235  flags = 0;
3236  error = 0;
3237  if (ifp->if_bytes) {
3238  xfs_alloc_arg_t args; /* allocation arguments */
3239  xfs_buf_t *bp; /* buffer for extent block */
3240  xfs_bmbt_rec_host_t *ep;/* extent record pointer */
3241 
3242  memset(&args, 0, sizeof(args));
3243  args.tp = tp;
3244  args.mp = ip->i_mount;
3245  args.firstblock = *firstblock;
3246  ASSERT((ifp->if_flags &
3247  (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == XFS_IFINLINE);
3248  /*
3249  * Allocate a block. We know we need only one, since the
3250  * file currently fits in an inode.
3251  */
3252  if (*firstblock == NULLFSBLOCK) {
3253  args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3254  args.type = XFS_ALLOCTYPE_START_BNO;
3255  } else {
3256  args.fsbno = *firstblock;
3257  args.type = XFS_ALLOCTYPE_NEAR_BNO;
3258  }
3259  args.total = total;
3260  args.mod = args.minleft = args.alignment = args.wasdel =
3261  args.isfl = args.minalignslop = 0;
3262  args.minlen = args.maxlen = args.prod = 1;
3263  if ((error = xfs_alloc_vextent(&args)))
3264  goto done;
3265  /*
3266  * Can't fail, the space was reserved.
3267  */
3268  ASSERT(args.fsbno != NULLFSBLOCK);
3269  ASSERT(args.len == 1);
3270  *firstblock = args.fsbno;
3271  bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3272  memcpy(bp->b_addr, ifp->if_u1.if_data, ifp->if_bytes);
3273  xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3274  xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3275  xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3276  xfs_iext_add(ifp, 0, 1);
3277  ep = xfs_iext_get_ext(ifp, 0);
3278  xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3279  trace_xfs_bmap_post_update(ip, 0,
3280  whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
3281  _THIS_IP_);
3282  XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3283  ip->i_d.di_nblocks = 1;
3284  xfs_trans_mod_dquot_byino(tp, ip,
3285  XFS_TRANS_DQ_BCOUNT, 1L);
3286  flags |= xfs_ilog_fext(whichfork);
3287  } else {
3288  ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3289  xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3290  }
3291  ifp->if_flags &= ~XFS_IFINLINE;
3292  ifp->if_flags |= XFS_IFEXTENTS;
3293  XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3294  flags |= XFS_ILOG_CORE;
3295 done:
3296  *logflagsp = flags;
3297  return error;
3298 }
3299 
3300 /*
3301  * Search the extent records for the entry containing block bno.
3302  * If bno lies in a hole, point to the next entry. If bno lies
3303  * past eof, *eofp will be set, and *prevp will contain the last
3304  * entry (null if none). Else, *lastxp will be set to the index
3305  * of the found entry; *gotp will contain the entry.
3306  */
3307 STATIC xfs_bmbt_rec_host_t * /* pointer to found extent entry */
3308 xfs_bmap_search_multi_extents(
3309  xfs_ifork_t *ifp, /* inode fork pointer */
3310  xfs_fileoff_t bno, /* block number searched for */
3311  int *eofp, /* out: end of file found */
3312  xfs_extnum_t *lastxp, /* out: last extent index */
3313  xfs_bmbt_irec_t *gotp, /* out: extent entry found */
3314  xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
3315 {
3316  xfs_bmbt_rec_host_t *ep; /* extent record pointer */
3317  xfs_extnum_t lastx; /* last extent index */
3318 
3319  /*
3320  * Initialize the extent entry structure to catch access to
3321  * uninitialized br_startblock field.
3322  */
3323  gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
3324  gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3325  gotp->br_state = XFS_EXT_INVALID;
3326 #if XFS_BIG_BLKNOS
3327  gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
3328 #else
3329  gotp->br_startblock = 0xffffa5a5;
3330 #endif
3331  prevp->br_startoff = NULLFILEOFF;
3332 
3333  ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
3334  if (lastx > 0) {
3335  xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
3336  }
3337  if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
3338  xfs_bmbt_get_all(ep, gotp);
3339  *eofp = 0;
3340  } else {
3341  if (lastx > 0) {
3342  *gotp = *prevp;
3343  }
3344  *eofp = 1;
3345  ep = NULL;
3346  }
3347  *lastxp = lastx;
3348  return ep;
3349 }
3350 
3351 /*
3352  * Search the extents list for the inode, for the extent containing bno.
3353  * If bno lies in a hole, point to the next entry. If bno lies past eof,
3354  * *eofp will be set, and *prevp will contain the last entry (null if none).
3355  * Else, *lastxp will be set to the index of the found
3356  * entry; *gotp will contain the entry.
3357  */
3358 STATIC xfs_bmbt_rec_host_t * /* pointer to found extent entry */
3359 xfs_bmap_search_extents(
3360  xfs_inode_t *ip, /* incore inode pointer */
3361  xfs_fileoff_t bno, /* block number searched for */
3362  int fork, /* data or attr fork */
3363  int *eofp, /* out: end of file found */
3364  xfs_extnum_t *lastxp, /* out: last extent index */
3365  xfs_bmbt_irec_t *gotp, /* out: extent entry found */
3366  xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
3367 {
3368  xfs_ifork_t *ifp; /* inode fork pointer */
3369  xfs_bmbt_rec_host_t *ep; /* extent record pointer */
3370 
3371  XFS_STATS_INC(xs_look_exlist);
3372  ifp = XFS_IFORK_PTR(ip, fork);
3373 
3374  ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
3375 
3376  if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
3377  !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
3378  xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
3379  "Access to block zero in inode %llu "
3380  "start_block: %llx start_off: %llx "
3381  "blkcnt: %llx extent-state: %x lastx: %x\n",
3382  (unsigned long long)ip->i_ino,
3383  (unsigned long long)gotp->br_startblock,
3384  (unsigned long long)gotp->br_startoff,
3385  (unsigned long long)gotp->br_blockcount,
3386  gotp->br_state, *lastxp);
3387  *lastxp = NULLEXTNUM;
3388  *eofp = 1;
3389  return NULL;
3390  }
3391  return ep;
3392 }
3393 
3394 /*
3395  * Compute the worst-case number of indirect blocks that will be used
3396  * for ip's delayed extent of length "len".
3397  */
3398 STATIC xfs_filblks_t
3399 xfs_bmap_worst_indlen(
3400  xfs_inode_t *ip, /* incore inode pointer */
3401  xfs_filblks_t len) /* delayed extent length */
3402 {
3403  int level; /* btree level number */
3404  int maxrecs; /* maximum record count at this level */
3405  xfs_mount_t *mp; /* mount structure */
3406  xfs_filblks_t rval; /* return value */
3407 
3408  mp = ip->i_mount;
3409  maxrecs = mp->m_bmap_dmxr[0];
3410  for (level = 0, rval = 0;
3411  level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3412  level++) {
3413  len += maxrecs - 1;
3414  do_div(len, maxrecs);
3415  rval += len;
3416  if (len == 1)
3417  return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3418  level - 1;
3419  if (level == 0)
3420  maxrecs = mp->m_bmap_dmxr[1];
3421  }
3422  return rval;
3423 }
3424 
3425 /*
3426  * Convert inode from non-attributed to attributed.
3427  * Must not be in a transaction, ip must not be locked.
3428  */
3429 int /* error code */
3430 xfs_bmap_add_attrfork(
3431  xfs_inode_t *ip, /* incore inode pointer */
3432  int size, /* space new attribute needs */
3433  int rsvd) /* xact may use reserved blks */
3434 {
3435  xfs_fsblock_t firstblock; /* 1st block/ag allocated */
3436  xfs_bmap_free_t flist; /* freed extent records */
3437  xfs_mount_t *mp; /* mount structure */
3438  xfs_trans_t *tp; /* transaction pointer */
3439  int blks; /* space reservation */
3440  int version = 1; /* superblock attr version */
3441  int committed; /* xaction was committed */
3442  int logflags; /* logging flags */
3443  int error; /* error return value */
3444 
3445  ASSERT(XFS_IFORK_Q(ip) == 0);
3446 
3447  mp = ip->i_mount;
3448  ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3449  tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3450  blks = XFS_ADDAFORK_SPACE_RES(mp);
3451  if (rsvd)
3452  tp->t_flags |= XFS_TRANS_RESERVE;
3453  if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3454  XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3455  goto error0;
3456  xfs_ilock(ip, XFS_ILOCK_EXCL);
3457  error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
3458  XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3459  XFS_QMOPT_RES_REGBLKS);
3460  if (error) {
3461  xfs_iunlock(ip, XFS_ILOCK_EXCL);
3462  xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3463  return error;
3464  }
3465  if (XFS_IFORK_Q(ip))
3466  goto error1;
3467  if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3468  /*
3469  * For inodes coming from pre-6.2 filesystems.
3470  */
3471  ASSERT(ip->i_d.di_aformat == 0);
3472  ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
3473  }
3474  ASSERT(ip->i_d.di_anextents == 0);
3475 
3476  xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3477  xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3478 
3479  switch (ip->i_d.di_format) {
3480  case XFS_DINODE_FMT_DEV:
3481  ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
3482  break;
3483  case XFS_DINODE_FMT_UUID:
3484  ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
3485  break;
3486  case XFS_DINODE_FMT_LOCAL:
3487  case XFS_DINODE_FMT_EXTENTS:
3488  case XFS_DINODE_FMT_BTREE:
3489  ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
3490  if (!ip->i_d.di_forkoff)
3491  ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
3492  else if (mp->m_flags & XFS_MOUNT_ATTR2)
3493  version = 2;
3494  break;
3495  default:
3496  ASSERT(0);
3497  error = XFS_ERROR(EINVAL);
3498  goto error1;
3499  }
3500 
3501  ASSERT(ip->i_afp == NULL);
3502  ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
3503  ip->i_afp->if_flags = XFS_IFEXTENTS;
3504  logflags = 0;
3505  xfs_bmap_init(&flist, &firstblock);
3506  switch (ip->i_d.di_format) {
3507  case XFS_DINODE_FMT_LOCAL:
3508  error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
3509  &logflags);
3510  break;
3511  case XFS_DINODE_FMT_EXTENTS:
3512  error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
3513  &flist, &logflags);
3514  break;
3515  case XFS_DINODE_FMT_BTREE:
3516  error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
3517  &logflags);
3518  break;
3519  default:
3520  error = 0;
3521  break;
3522  }
3523  if (logflags)
3524  xfs_trans_log_inode(tp, ip, logflags);
3525  if (error)
3526  goto error2;
3527  if (!xfs_sb_version_hasattr(&mp->m_sb) ||
3528  (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
3529  __int64_t sbfields = 0;
3530 
3531  spin_lock(&mp->m_sb_lock);
3532  if (!xfs_sb_version_hasattr(&mp->m_sb)) {
3533  xfs_sb_version_addattr(&mp->m_sb);
3534  sbfields |= XFS_SB_VERSIONNUM;
3535  }
3536  if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
3537  xfs_sb_version_addattr2(&mp->m_sb);
3538  sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
3539  }
3540  if (sbfields) {
3541  spin_unlock(&mp->m_sb_lock);
3542  xfs_mod_sb(tp, sbfields);
3543  } else
3544  spin_unlock(&mp->m_sb_lock);
3545  }
3546 
3547  error = xfs_bmap_finish(&tp, &flist, &committed);
3548  if (error)
3549  goto error2;
3550  return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3551 error2:
3552  xfs_bmap_cancel(&flist);
3553 error1:
3554  xfs_iunlock(ip, XFS_ILOCK_EXCL);
3555 error0:
3556  xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
3557  return error;
3558 }
3559 
3560 /*
3561  * Add the extent to the list of extents to be free at transaction end.
3562  * The list is maintained sorted (by block number).
3563  */
3564 /* ARGSUSED */
3565 void
3566 xfs_bmap_add_free(
3567  xfs_fsblock_t bno, /* fs block number of extent */
3568  xfs_filblks_t len, /* length of extent */
3569  xfs_bmap_free_t *flist, /* list of extents */
3570  xfs_mount_t *mp) /* mount point structure */
3571 {
3572  xfs_bmap_free_item_t *cur; /* current (next) element */
3573  xfs_bmap_free_item_t *new; /* new element */
3574  xfs_bmap_free_item_t *prev; /* previous element */
3575 #ifdef DEBUG
3576  xfs_agnumber_t agno;
3577  xfs_agblock_t agbno;
3578 
3579  ASSERT(bno != NULLFSBLOCK);
3580  ASSERT(len > 0);
3581  ASSERT(len <= MAXEXTLEN);
3582  ASSERT(!isnullstartblock(bno));
3583  agno = XFS_FSB_TO_AGNO(mp, bno);
3584  agbno = XFS_FSB_TO_AGBNO(mp, bno);
3585  ASSERT(agno < mp->m_sb.sb_agcount);
3586  ASSERT(agbno < mp->m_sb.sb_agblocks);
3587  ASSERT(len < mp->m_sb.sb_agblocks);
3588  ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
3589 #endif
3590  ASSERT(xfs_bmap_free_item_zone != NULL);
3591  new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
3592  new->xbfi_startblock = bno;
3593  new->xbfi_blockcount = (xfs_extlen_t)len;
3594  for (prev = NULL, cur = flist->xbf_first;
3595  cur != NULL;
3596  prev = cur, cur = cur->xbfi_next) {
3597  if (cur->xbfi_startblock >= bno)
3598  break;
3599  }
3600  if (prev)
3601  prev->xbfi_next = new;
3602  else
3603  flist->xbf_first = new;
3604  new->xbfi_next = cur;
3605  flist->xbf_count++;
3606 }
3607 
3608 /*
3609  * Compute and fill in the value of the maximum depth of a bmap btree
3610  * in this filesystem. Done once, during mount.
3611  */
3612 void
3613 xfs_bmap_compute_maxlevels(
3614  xfs_mount_t *mp, /* file system mount structure */
3615  int whichfork) /* data or attr fork */
3616 {
3617  int level; /* btree level */
3618  uint maxblocks; /* max blocks at this level */
3619  uint maxleafents; /* max leaf entries possible */
3620  int maxrootrecs; /* max records in root block */
3621  int minleafrecs; /* min records in leaf block */
3622  int minnoderecs; /* min records in node block */
3623  int sz; /* root block size */
3624 
3625  /*
3626  * The maximum number of extents in a file, hence the maximum
3627  * number of leaf entries, is controlled by the type of di_nextents
3628  * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
3629  * (a signed 16-bit number, xfs_aextnum_t).
3630  *
3631  * Note that we can no longer assume that if we are in ATTR1 that
3632  * the fork offset of all the inodes will be
3633  * (xfs_default_attroffset(ip) >> 3) because we could have mounted
3634  * with ATTR2 and then mounted back with ATTR1, keeping the
3635  * di_forkoff's fixed but probably at various positions. Therefore,
3636  * for both ATTR1 and ATTR2 we have to assume the worst case scenario
3637  * of a minimum size available.
3638  */
3639  if (whichfork == XFS_DATA_FORK) {
3640  maxleafents = MAXEXTNUM;
3641  sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
3642  } else {
3643  maxleafents = MAXAEXTNUM;
3644  sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
3645  }
3646  maxrootrecs = xfs_bmdr_maxrecs(mp, sz, 0);
3647  minleafrecs = mp->m_bmap_dmnr[0];
3648  minnoderecs = mp->m_bmap_dmnr[1];
3649  maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
3650  for (level = 1; maxblocks > 1; level++) {
3651  if (maxblocks <= maxrootrecs)
3652  maxblocks = 1;
3653  else
3654  maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
3655  }
3656  mp->m_bm_maxlevels[whichfork] = level;
3657 }
3658 
3659 /*
3660  * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
3661  * caller. Frees all the extents that need freeing, which must be done
3662  * last due to locking considerations. We never free any extents in
3663  * the first transaction.
3664  *
3665  * Return 1 if the given transaction was committed and a new one
3666  * started, and 0 otherwise in the committed parameter.
3667  */
3668 int /* error */
3669 xfs_bmap_finish(
3670  xfs_trans_t **tp, /* transaction pointer addr */
3671  xfs_bmap_free_t *flist, /* i/o: list extents to free */
3672  int *committed) /* xact committed or not */
3673 {
3674  xfs_efd_log_item_t *efd; /* extent free data */
3675  xfs_efi_log_item_t *efi; /* extent free intention */
3676  int error; /* error return value */
3677  xfs_bmap_free_item_t *free; /* free extent item */
3678  unsigned int logres; /* new log reservation */
3679  unsigned int logcount; /* new log count */
3680  xfs_mount_t *mp; /* filesystem mount structure */
3681  xfs_bmap_free_item_t *next; /* next item on free list */
3682  xfs_trans_t *ntp; /* new transaction pointer */
3683 
3684  ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
3685  if (flist->xbf_count == 0) {
3686  *committed = 0;
3687  return 0;
3688  }
3689  ntp = *tp;
3690  efi = xfs_trans_get_efi(ntp, flist->xbf_count);
3691  for (free = flist->xbf_first; free; free = free->xbfi_next)
3692  xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
3693  free->xbfi_blockcount);
3694  logres = ntp->t_log_res;
3695  logcount = ntp->t_log_count;
3696  ntp = xfs_trans_dup(*tp);
3697  error = xfs_trans_commit(*tp, 0);
3698  *tp = ntp;
3699  *committed = 1;
3700  /*
3701  * We have a new transaction, so we should return committed=1,
3702  * even though we're returning an error.
3703  */
3704  if (error)
3705  return error;
3706 
3707  /*
3708  * transaction commit worked ok so we can drop the extra ticket
3709  * reference that we gained in xfs_trans_dup()
3710  */
3711  xfs_log_ticket_put(ntp->t_ticket);
3712 
3713  if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
3714  logcount)))
3715  return error;
3716  efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
3717  for (free = flist->xbf_first; free != NULL; free = next) {
3718  next = free->xbfi_next;
3719  if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
3720  free->xbfi_blockcount))) {
3721  /*
3722  * The bmap free list will be cleaned up at a
3723  * higher level. The EFI will be canceled when
3724  * this transaction is aborted.
3725  * Need to force shutdown here to make sure it
3726  * happens, since this transaction may not be
3727  * dirty yet.
3728  */
3729  mp = ntp->t_mountp;
3730  if (!XFS_FORCED_SHUTDOWN(mp))
3731  xfs_force_shutdown(mp,
3732  (error == EFSCORRUPTED) ?
3733  SHUTDOWN_CORRUPT_INCORE :
3734  SHUTDOWN_META_IO_ERROR);
3735  return error;
3736  }
3737  xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
3738  free->xbfi_blockcount);
3739  xfs_bmap_del_free(flist, NULL, free);
3740  }
3741  return 0;
3742 }
3743 
3744 /*
3745  * Free up any items left in the list.
3746  */
3747 void
3748 xfs_bmap_cancel(
3749  xfs_bmap_free_t *flist) /* list of bmap_free_items */
3750 {
3751  xfs_bmap_free_item_t *free; /* free list item */
3752  xfs_bmap_free_item_t *next;
3753 
3754  if (flist->xbf_count == 0)
3755  return;
3756  ASSERT(flist->xbf_first != NULL);
3757  for (free = flist->xbf_first; free; free = next) {
3758  next = free->xbfi_next;
3759  xfs_bmap_del_free(flist, NULL, free);
3760  }
3761  ASSERT(flist->xbf_count == 0);
3762 }
3763 
3764 /*
3765  * Returns the file-relative block number of the first unused block(s)
3766  * in the file with at least "len" logically contiguous blocks free.
3767  * This is the lowest-address hole if the file has holes, else the first block
3768  * past the end of file.
3769  * Return 0 if the file is currently local (in-inode).
3770  */
3771 int /* error */
3772 xfs_bmap_first_unused(
3773  xfs_trans_t *tp, /* transaction pointer */
3774  xfs_inode_t *ip, /* incore inode */
3775  xfs_extlen_t len, /* size of hole to find */
3776  xfs_fileoff_t *first_unused, /* unused block */
3777  int whichfork) /* data or attr fork */
3778 {
3779  int error; /* error return value */
3780  int idx; /* extent record index */
3781  xfs_ifork_t *ifp; /* inode fork pointer */
3782  xfs_fileoff_t lastaddr; /* last block number seen */
3783  xfs_fileoff_t lowest; /* lowest useful block */
3784  xfs_fileoff_t max; /* starting useful block */
3785  xfs_fileoff_t off; /* offset for this block */
3786  xfs_extnum_t nextents; /* number of extent entries */
3787 
3788  ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
3789  XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
3790  XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3791  if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3792  *first_unused = 0;
3793  return 0;
3794  }
3795  ifp = XFS_IFORK_PTR(ip, whichfork);
3796  if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3797  (error = xfs_iread_extents(tp, ip, whichfork)))
3798  return error;
3799  lowest = *first_unused;
3800  nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3801  for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
3802  xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
3803  off = xfs_bmbt_get_startoff(ep);
3804  /*
3805  * See if the hole before this extent will work.
3806  */
3807  if (off >= lowest + len && off - max >= len) {
3808  *first_unused = max;
3809  return 0;
3810  }
3811  lastaddr = off + xfs_bmbt_get_blockcount(ep);
3812  max = XFS_FILEOFF_MAX(lastaddr, lowest);
3813  }
3814  *first_unused = max;
3815  return 0;
3816 }
3817 
3818 /*
3819  * Returns the file-relative block number of the last block + 1 before
3820  * last_block (input value) in the file.
3821  * This is not based on i_size, it is based on the extent records.
3822  * Returns 0 for local files, as they do not have extent records.
3823  */
3824 int /* error */
3825 xfs_bmap_last_before(
3826  xfs_trans_t *tp, /* transaction pointer */
3827  xfs_inode_t *ip, /* incore inode */
3828  xfs_fileoff_t *last_block, /* last block */
3829  int whichfork) /* data or attr fork */
3830 {
3831  xfs_fileoff_t bno; /* input file offset */
3832  int eof; /* hit end of file */
3833  xfs_bmbt_rec_host_t *ep; /* pointer to last extent */
3834  int error; /* error return value */
3835  xfs_bmbt_irec_t got; /* current extent value */
3836  xfs_ifork_t *ifp; /* inode fork pointer */
3837  xfs_extnum_t lastx; /* last extent used */
3838  xfs_bmbt_irec_t prev; /* previous extent value */
3839 
3840  if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
3841  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
3842  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
3843  return XFS_ERROR(EIO);
3844  if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3845  *last_block = 0;
3846  return 0;
3847  }
3848  ifp = XFS_IFORK_PTR(ip, whichfork);
3849  if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3850  (error = xfs_iread_extents(tp, ip, whichfork)))
3851  return error;
3852  bno = *last_block - 1;
3853  ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
3854  &prev);
3855  if (eof || xfs_bmbt_get_startoff(ep) > bno) {
3856  if (prev.br_startoff == NULLFILEOFF)
3857  *last_block = 0;
3858  else
3859  *last_block = prev.br_startoff + prev.br_blockcount;
3860  }
3861  /*
3862  * Otherwise *last_block is already the right answer.
3863  */
3864  return 0;
3865 }
3866 
3867 STATIC int
3868 xfs_bmap_last_extent(
3869  struct xfs_trans *tp,
3870  struct xfs_inode *ip,
3871  int whichfork,
3872  struct xfs_bmbt_irec *rec,
3873  int *is_empty)
3874 {
3875  struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
3876  int error;
3877  int nextents;
3878 
3879  if (!(ifp->if_flags & XFS_IFEXTENTS)) {
3880  error = xfs_iread_extents(tp, ip, whichfork);
3881  if (error)
3882  return error;
3883  }
3884 
3885  nextents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
3886  if (nextents == 0) {
3887  *is_empty = 1;
3888  return 0;
3889  }
3890 
3891  xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
3892  *is_empty = 0;
3893  return 0;
3894 }
3895 
3896 /*
3897  * Check the last inode extent to determine whether this allocation will result
3898  * in blocks being allocated at the end of the file. When we allocate new data
3899  * blocks at the end of the file which do not start at the previous data block,
3900  * we will try to align the new blocks at stripe unit boundaries.
3901  *
3902  * Returns 0 in bma->aeof if the file (fork) is empty as any new write will be
3903  * at, or past the EOF.
3904  */
3905 STATIC int
3906 xfs_bmap_isaeof(
3907  struct xfs_bmalloca *bma,
3908  int whichfork)
3909 {
3910  struct xfs_bmbt_irec rec;
3911  int is_empty;
3912  int error;
3913 
3914  bma->aeof = 0;
3915  error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
3916  &is_empty);
3917  if (error || is_empty)
3918  return error;
3919 
3920  /*
3921  * Check if we are allocation or past the last extent, or at least into
3922  * the last delayed allocated extent.
3923  */
3924  bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
3925  (bma->offset >= rec.br_startoff &&
3926  isnullstartblock(rec.br_startblock));
3927  return 0;
3928 }
3929 
3930 /*
3931  * Check if the endoff is outside the last extent. If so the caller will grow
3932  * the allocation to a stripe unit boundary. All offsets are considered outside
3933  * the end of file for an empty fork, so 1 is returned in *eof in that case.
3934  */
3935 int
3936 xfs_bmap_eof(
3937  struct xfs_inode *ip,
3938  xfs_fileoff_t endoff,
3939  int whichfork,
3940  int *eof)
3941 {
3942  struct xfs_bmbt_irec rec;
3943  int error;
3944 
3945  error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof);
3946  if (error || *eof)
3947  return error;
3948 
3949  *eof = endoff >= rec.br_startoff + rec.br_blockcount;
3950  return 0;
3951 }
3952 
3953 /*
3954  * Returns the file-relative block number of the first block past eof in
3955  * the file. This is not based on i_size, it is based on the extent records.
3956  * Returns 0 for local files, as they do not have extent records.
3957  */
3958 int
3959 xfs_bmap_last_offset(
3960  struct xfs_trans *tp,
3961  struct xfs_inode *ip,
3962  xfs_fileoff_t *last_block,
3963  int whichfork)
3964 {
3965  struct xfs_bmbt_irec rec;
3966  int is_empty;
3967  int error;
3968 
3969  *last_block = 0;
3970 
3971  if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
3972  return 0;
3973 
3974  if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
3975  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
3976  return XFS_ERROR(EIO);
3977 
3978  error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
3979  if (error || is_empty)
3980  return error;
3981 
3982  *last_block = rec.br_startoff + rec.br_blockcount;
3983  return 0;
3984 }
3985 
3986 /*
3987  * Returns whether the selected fork of the inode has exactly one
3988  * block or not. For the data fork we check this matches di_size,
3989  * implying the file's range is 0..bsize-1.
3990  */
3991 int /* 1=>1 block, 0=>otherwise */
3992 xfs_bmap_one_block(
3993  xfs_inode_t *ip, /* incore inode */
3994  int whichfork) /* data or attr fork */
3995 {
3996  xfs_bmbt_rec_host_t *ep; /* ptr to fork's extent */
3997  xfs_ifork_t *ifp; /* inode fork pointer */
3998  int rval; /* return value */
3999  xfs_bmbt_irec_t s; /* internal version of extent */
4000 
4001 #ifndef DEBUG
4002  if (whichfork == XFS_DATA_FORK)
4003  return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
4004 #endif /* !DEBUG */
4005  if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4006  return 0;
4007  if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4008  return 0;
4009  ifp = XFS_IFORK_PTR(ip, whichfork);
4010  ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4011  ep = xfs_iext_get_ext(ifp, 0);
4012  xfs_bmbt_get_all(ep, &s);
4013  rval = s.br_startoff == 0 && s.br_blockcount == 1;
4014  if (rval && whichfork == XFS_DATA_FORK)
4015  ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
4016  return rval;
4017 }
4018 
4019 STATIC int
4020 xfs_bmap_sanity_check(
4021  struct xfs_mount *mp,
4022  struct xfs_buf *bp,
4023  int level)
4024 {
4025  struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
4026 
4027  if (block->bb_magic != cpu_to_be32(XFS_BMAP_MAGIC) ||
4028  be16_to_cpu(block->bb_level) != level ||
4029  be16_to_cpu(block->bb_numrecs) == 0 ||
4030  be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
4031  return 0;
4032  return 1;
4033 }
4034 
4035 /*
4036  * Read in the extents to if_extents.
4037  * All inode fields are set up by caller, we just traverse the btree
4038  * and copy the records in. If the file system cannot contain unwritten
4039  * extents, the records are checked for no "state" flags.
4040  */
4041 int /* error */
4042 xfs_bmap_read_extents(
4043  xfs_trans_t *tp, /* transaction pointer */
4044  xfs_inode_t *ip, /* incore inode */
4045  int whichfork) /* data or attr fork */
4046 {
4047  struct xfs_btree_block *block; /* current btree block */
4048  xfs_fsblock_t bno; /* block # of "block" */
4049  xfs_buf_t *bp; /* buffer for "block" */
4050  int error; /* error return value */
4051  xfs_exntfmt_t exntf; /* XFS_EXTFMT_NOSTATE, if checking */
4052  xfs_extnum_t i, j; /* index into the extents list */
4053  xfs_ifork_t *ifp; /* fork structure */
4054  int level; /* btree level, for checking */
4055  xfs_mount_t *mp; /* file system mount structure */
4056  __be64 *pp; /* pointer to block address */
4057  /* REFERENCED */
4058  xfs_extnum_t room; /* number of entries there's room for */
4059 
4060  bno = NULLFSBLOCK;
4061  mp = ip->i_mount;
4062  ifp = XFS_IFORK_PTR(ip, whichfork);
4063  exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4064  XFS_EXTFMT_INODE(ip);
4065  block = ifp->if_broot;
4066  /*
4067  * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4068  */
4069  level = be16_to_cpu(block->bb_level);
4070  ASSERT(level > 0);
4071  pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
4072  bno = be64_to_cpu(*pp);
4073  ASSERT(bno != NULLDFSBNO);
4074  ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
4075  ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
4076  /*
4077  * Go down the tree until leaf level is reached, following the first
4078  * pointer (leftmost) at each level.
4079  */
4080  while (level-- > 0) {
4081  if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4082  XFS_BMAP_BTREE_REF)))
4083  return error;
4084  block = XFS_BUF_TO_BLOCK(bp);
4085  XFS_WANT_CORRUPTED_GOTO(
4086  xfs_bmap_sanity_check(mp, bp, level),
4087  error0);
4088  if (level == 0)
4089  break;
4090  pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
4091  bno = be64_to_cpu(*pp);
4092  XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
4093  xfs_trans_brelse(tp, bp);
4094  }
4095  /*
4096  * Here with bp and block set to the leftmost leaf node in the tree.
4097  */
4098  room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4099  i = 0;
4100  /*
4101  * Loop over all leaf nodes. Copy information to the extent records.
4102  */
4103  for (;;) {
4104  xfs_bmbt_rec_t *frp;
4105  xfs_fsblock_t nextbno;
4106  xfs_extnum_t num_recs;
4107  xfs_extnum_t start;
4108 
4109  num_recs = xfs_btree_get_numrecs(block);
4110  if (unlikely(i + num_recs > room)) {
4111  ASSERT(i + num_recs <= room);
4112  xfs_warn(ip->i_mount,
4113  "corrupt dinode %Lu, (btree extents).",
4114  (unsigned long long) ip->i_ino);
4115  XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
4116  XFS_ERRLEVEL_LOW, ip->i_mount, block);
4117  goto error0;
4118  }
4119  XFS_WANT_CORRUPTED_GOTO(
4120  xfs_bmap_sanity_check(mp, bp, 0),
4121  error0);
4122  /*
4123  * Read-ahead the next leaf block, if any.
4124  */
4125  nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
4126  if (nextbno != NULLFSBLOCK)
4127  xfs_btree_reada_bufl(mp, nextbno, 1);
4128  /*
4129  * Copy records into the extent records.
4130  */
4131  frp = XFS_BMBT_REC_ADDR(mp, block, 1);
4132  start = i;
4133  for (j = 0; j < num_recs; j++, i++, frp++) {
4134  xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
4135  trp->l0 = be64_to_cpu(frp->l0);
4136  trp->l1 = be64_to_cpu(frp->l1);
4137  }
4138  if (exntf == XFS_EXTFMT_NOSTATE) {
4139  /*
4140  * Check all attribute bmap btree records and
4141  * any "older" data bmap btree records for a
4142  * set bit in the "extent flag" position.
4143  */
4144  if (unlikely(xfs_check_nostate_extents(ifp,
4145  start, num_recs))) {
4146  XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4147  XFS_ERRLEVEL_LOW,
4148  ip->i_mount);
4149  goto error0;
4150  }
4151  }
4152  xfs_trans_brelse(tp, bp);
4153  bno = nextbno;
4154  /*
4155  * If we've reached the end, stop.
4156  */
4157  if (bno == NULLFSBLOCK)
4158  break;
4159  if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4160  XFS_BMAP_BTREE_REF)))
4161  return error;
4162  block = XFS_BUF_TO_BLOCK(bp);
4163  }
4164  ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4165  ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4166  XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
4167  return 0;
4168 error0:
4169  xfs_trans_brelse(tp, bp);
4170  return XFS_ERROR(EFSCORRUPTED);
4171 }
4172 
4173 #ifdef DEBUG
4174 /*
4175  * Add bmap trace insert entries for all the contents of the extent records.
4176  */
4177 void
4178 xfs_bmap_trace_exlist(
4179  xfs_inode_t *ip, /* incore inode pointer */
4180  xfs_extnum_t cnt, /* count of entries in the list */
4181  int whichfork, /* data or attr fork */
4182  unsigned long caller_ip)
4183 {
4184  xfs_extnum_t idx; /* extent record index */
4185  xfs_ifork_t *ifp; /* inode fork pointer */
4186  int state = 0;
4187 
4188  if (whichfork == XFS_ATTR_FORK)
4189  state |= BMAP_ATTRFORK;
4190 
4191  ifp = XFS_IFORK_PTR(ip, whichfork);
4192  ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4193  for (idx = 0; idx < cnt; idx++)
4194  trace_xfs_extlist(ip, idx, whichfork, caller_ip);
4195 }
4196 
4197 /*
4198  * Validate that the bmbt_irecs being returned from bmapi are valid
4199  * given the callers original parameters. Specifically check the
4200  * ranges of the returned irecs to ensure that they only extent beyond
4201  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4202  */
4203 STATIC void
4204 xfs_bmap_validate_ret(
4205  xfs_fileoff_t bno,
4206  xfs_filblks_t len,
4207  int flags,
4208  xfs_bmbt_irec_t *mval,
4209  int nmap,
4210  int ret_nmap)
4211 {
4212  int i; /* index to map values */
4213 
4214  ASSERT(ret_nmap <= nmap);
4215 
4216  for (i = 0; i < ret_nmap; i++) {
4217  ASSERT(mval[i].br_blockcount > 0);
4218  if (!(flags & XFS_BMAPI_ENTIRE)) {
4219  ASSERT(mval[i].br_startoff >= bno);
4220  ASSERT(mval[i].br_blockcount <= len);
4221  ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4222  bno + len);
4223  } else {
4224  ASSERT(mval[i].br_startoff < bno + len);
4225  ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4226  bno);
4227  }
4228  ASSERT(i == 0 ||
4229  mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4230  mval[i].br_startoff);
4231  ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4232  mval[i].br_startblock != HOLESTARTBLOCK);
4233  ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4234  mval[i].br_state == XFS_EXT_UNWRITTEN);
4235  }
4236 }
4237 #endif /* DEBUG */
4238 
4239 
4240 /*
4241  * Trim the returned map to the required bounds
4242  */
4243 STATIC void
4244 xfs_bmapi_trim_map(
4245  struct xfs_bmbt_irec *mval,
4246  struct xfs_bmbt_irec *got,
4247  xfs_fileoff_t *bno,
4248  xfs_filblks_t len,
4249  xfs_fileoff_t obno,
4250  xfs_fileoff_t end,
4251  int n,
4252  int flags)
4253 {
4254  if ((flags & XFS_BMAPI_ENTIRE) ||
4255  got->br_startoff + got->br_blockcount <= obno) {
4256  *mval = *got;
4257  if (isnullstartblock(got->br_startblock))
4258  mval->br_startblock = DELAYSTARTBLOCK;
4259  return;
4260  }
4261 
4262  if (obno > *bno)
4263  *bno = obno;
4264  ASSERT((*bno >= obno) || (n == 0));
4265  ASSERT(*bno < end);
4266  mval->br_startoff = *bno;
4267  if (isnullstartblock(got->br_startblock))
4268  mval->br_startblock = DELAYSTARTBLOCK;
4269  else
4270  mval->br_startblock = got->br_startblock +
4271  (*bno - got->br_startoff);
4272  /*
4273  * Return the minimum of what we got and what we asked for for
4274  * the length. We can use the len variable here because it is
4275  * modified below and we could have been there before coming
4276  * here if the first part of the allocation didn't overlap what
4277  * was asked for.
4278  */
4279  mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
4280  got->br_blockcount - (*bno - got->br_startoff));
4281  mval->br_state = got->br_state;
4282  ASSERT(mval->br_blockcount <= len);
4283  return;
4284 }
4285 
4286 /*
4287  * Update and validate the extent map to return
4288  */
4289 STATIC void
4290 xfs_bmapi_update_map(
4291  struct xfs_bmbt_irec **map,
4292  xfs_fileoff_t *bno,
4293  xfs_filblks_t *len,
4294  xfs_fileoff_t obno,
4295  xfs_fileoff_t end,
4296  int *n,
4297  int flags)
4298 {
4299  xfs_bmbt_irec_t *mval = *map;
4300 
4301  ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4302  ((mval->br_startoff + mval->br_blockcount) <= end));
4303  ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
4304  (mval->br_startoff < obno));
4305 
4306  *bno = mval->br_startoff + mval->br_blockcount;
4307  *len = end - *bno;
4308  if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
4309  /* update previous map with new information */
4310  ASSERT(mval->br_startblock == mval[-1].br_startblock);
4311  ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
4312  ASSERT(mval->br_state == mval[-1].br_state);
4313  mval[-1].br_blockcount = mval->br_blockcount;
4314  mval[-1].br_state = mval->br_state;
4315  } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
4316  mval[-1].br_startblock != DELAYSTARTBLOCK &&
4317  mval[-1].br_startblock != HOLESTARTBLOCK &&
4318  mval->br_startblock == mval[-1].br_startblock +
4319  mval[-1].br_blockcount &&
4320  ((flags & XFS_BMAPI_IGSTATE) ||
4321  mval[-1].br_state == mval->br_state)) {
4322  ASSERT(mval->br_startoff ==
4323  mval[-1].br_startoff + mval[-1].br_blockcount);
4324  mval[-1].br_blockcount += mval->br_blockcount;
4325  } else if (*n > 0 &&
4326  mval->br_startblock == DELAYSTARTBLOCK &&
4327  mval[-1].br_startblock == DELAYSTARTBLOCK &&
4328  mval->br_startoff ==
4329  mval[-1].br_startoff + mval[-1].br_blockcount) {
4330  mval[-1].br_blockcount += mval->br_blockcount;
4331  mval[-1].br_state = mval->br_state;
4332  } else if (!((*n == 0) &&
4333  ((mval->br_startoff + mval->br_blockcount) <=
4334  obno))) {
4335  mval++;
4336  (*n)++;
4337  }
4338  *map = mval;
4339 }
4340 
4341 /*
4342  * Map file blocks to filesystem blocks without allocation.
4343  */
4344 int
4345 xfs_bmapi_read(
4346  struct xfs_inode *ip,
4347  xfs_fileoff_t bno,
4348  xfs_filblks_t len,
4349  struct xfs_bmbt_irec *mval,
4350  int *nmap,
4351  int flags)
4352 {
4353  struct xfs_mount *mp = ip->i_mount;
4354  struct xfs_ifork *ifp;
4355  struct xfs_bmbt_irec got;
4356  struct xfs_bmbt_irec prev;
4357  xfs_fileoff_t obno;
4358  xfs_fileoff_t end;
4359  xfs_extnum_t lastx;
4360  int error;
4361  int eof;
4362  int n = 0;
4363  int whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4364  XFS_ATTR_FORK : XFS_DATA_FORK;
4365 
4366  ASSERT(*nmap >= 1);
4367  ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
4368  XFS_BMAPI_IGSTATE)));
4369 
4370  if (unlikely(XFS_TEST_ERROR(
4371  (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4372  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4373  mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4374  XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
4375  return XFS_ERROR(EFSCORRUPTED);
4376  }
4377 
4378  if (XFS_FORCED_SHUTDOWN(mp))
4379  return XFS_ERROR(EIO);
4380 
4381  XFS_STATS_INC(xs_blk_mapr);
4382 
4383  ifp = XFS_IFORK_PTR(ip, whichfork);
4384 
4385  if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4386  error = xfs_iread_extents(NULL, ip, whichfork);
4387  if (error)
4388  return error;
4389  }
4390 
4391  xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got, &prev);
4392  end = bno + len;
4393  obno = bno;
4394 
4395  while (bno < end && n < *nmap) {
4396  /* Reading past eof, act as though there's a hole up to end. */
4397  if (eof)
4398  got.br_startoff = end;
4399  if (got.br_startoff > bno) {
4400  /* Reading in a hole. */
4401  mval->br_startoff = bno;
4402  mval->br_startblock = HOLESTARTBLOCK;
4403  mval->br_blockcount =
4404  XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4405  mval->br_state = XFS_EXT_NORM;
4406  bno += mval->br_blockcount;
4407  len -= mval->br_blockcount;
4408  mval++;
4409  n++;
4410  continue;
4411  }
4412 
4413  /* set up the extent map to return. */
4414  xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4415  xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4416 
4417  /* If we're done, stop now. */
4418  if (bno >= end || n >= *nmap)
4419  break;
4420 
4421  /* Else go on to the next record. */
4422  if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4423  xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4424  else
4425  eof = 1;
4426  }
4427  *nmap = n;
4428  return 0;
4429 }
4430 
4431 STATIC int
4432 xfs_bmapi_reserve_delalloc(
4433  struct xfs_inode *ip,
4434  xfs_fileoff_t aoff,
4435  xfs_filblks_t len,
4436  struct xfs_bmbt_irec *got,
4437  struct xfs_bmbt_irec *prev,
4438  xfs_extnum_t *lastx,
4439  int eof)
4440 {
4441  struct xfs_mount *mp = ip->i_mount;
4442  struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4443  xfs_extlen_t alen;
4444  xfs_extlen_t indlen;
4445  char rt = XFS_IS_REALTIME_INODE(ip);
4446  xfs_extlen_t extsz;
4447  int error;
4448 
4449  alen = XFS_FILBLKS_MIN(len, MAXEXTLEN);
4450  if (!eof)
4451  alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4452 
4453  /* Figure out the extent size, adjust alen */
4454  extsz = xfs_get_extsz_hint(ip);
4455  if (extsz) {
4456  /*
4457  * Make sure we don't exceed a single extent length when we
4458  * align the extent by reducing length we are going to
4459  * allocate by the maximum amount extent size aligment may
4460  * require.
4461  */
4462  alen = XFS_FILBLKS_MIN(len, MAXEXTLEN - (2 * extsz - 1));
4463  error = xfs_bmap_extsize_align(mp, got, prev, extsz, rt, eof,
4464  1, 0, &aoff, &alen);
4465  ASSERT(!error);
4466  }
4467 
4468  if (rt)
4469  extsz = alen / mp->m_sb.sb_rextsize;
4470 
4471  /*
4472  * Make a transaction-less quota reservation for delayed allocation
4473  * blocks. This number gets adjusted later. We return if we haven't
4474  * allocated blocks already inside this loop.
4475  */
4476  error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4477  rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4478  if (error)
4479  return error;
4480 
4481  /*
4482  * Split changing sb for alen and indlen since they could be coming
4483  * from different places.
4484  */
4485  indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4486  ASSERT(indlen > 0);
4487 
4488  if (rt) {
4489  error = xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
4490  -((int64_t)extsz), 0);
4491  } else {
4492  error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4493  -((int64_t)alen), 0);
4494  }
4495 
4496  if (error)
4497  goto out_unreserve_quota;
4498 
4499  error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4500  -((int64_t)indlen), 0);
4501  if (error)
4502  goto out_unreserve_blocks;
4503 
4504 
4505  ip->i_delayed_blks += alen;
4506 
4507  got->br_startoff = aoff;
4508  got->br_startblock = nullstartblock(indlen);
4509  got->br_blockcount = alen;
4510  got->br_state = XFS_EXT_NORM;
4511  xfs_bmap_add_extent_hole_delay(ip, lastx, got);
4512 
4513  /*
4514  * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4515  * might have merged it into one of the neighbouring ones.
4516  */
4517  xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), got);
4518 
4519  ASSERT(got->br_startoff <= aoff);
4520  ASSERT(got->br_startoff + got->br_blockcount >= aoff + alen);
4521  ASSERT(isnullstartblock(got->br_startblock));
4522  ASSERT(got->br_state == XFS_EXT_NORM);
4523  return 0;
4524 
4525 out_unreserve_blocks:
4526  if (rt)
4527  xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS, extsz, 0);
4528  else
4529  xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, alen, 0);
4530 out_unreserve_quota:
4531  if (XFS_IS_QUOTA_ON(mp))
4532  xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ?
4533  XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4534  return error;
4535 }
4536 
4537 /*
4538  * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4539  */
4540 int
4541 xfs_bmapi_delay(
4542  struct xfs_inode *ip, /* incore inode */
4543  xfs_fileoff_t bno, /* starting file offs. mapped */
4544  xfs_filblks_t len, /* length to map in file */
4545  struct xfs_bmbt_irec *mval, /* output: map values */
4546  int *nmap, /* i/o: mval size/count */
4547  int flags) /* XFS_BMAPI_... */
4548 {
4549  struct xfs_mount *mp = ip->i_mount;
4550  struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4551  struct xfs_bmbt_irec got; /* current file extent record */
4552  struct xfs_bmbt_irec prev; /* previous file extent record */
4553  xfs_fileoff_t obno; /* old block number (offset) */
4554  xfs_fileoff_t end; /* end of mapped file region */
4555  xfs_extnum_t lastx; /* last useful extent number */
4556  int eof; /* we've hit the end of extents */
4557  int n = 0; /* current extent index */
4558  int error = 0;
4559 
4560  ASSERT(*nmap >= 1);
4561  ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4562  ASSERT(!(flags & ~XFS_BMAPI_ENTIRE));
4563 
4564  if (unlikely(XFS_TEST_ERROR(
4565  (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4566  XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4567  mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4568  XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW, mp);
4569  return XFS_ERROR(EFSCORRUPTED);
4570  }
4571 
4572  if (XFS_FORCED_SHUTDOWN(mp))
4573  return XFS_ERROR(EIO);
4574 
4575  XFS_STATS_INC(xs_blk_mapw);
4576 
4577  if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4578  error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4579  if (error)
4580  return error;
4581  }
4582 
4583  xfs_bmap_search_extents(ip, bno, XFS_DATA_FORK, &eof, &lastx, &got, &prev);
4584  end = bno + len;
4585  obno = bno;
4586 
4587  while (bno < end && n < *nmap) {
4588  if (eof || got.br_startoff > bno) {
4589  error = xfs_bmapi_reserve_delalloc(ip, bno, len, &got,
4590  &prev, &lastx, eof);
4591  if (error) {
4592  if (n == 0) {
4593  *nmap = 0;
4594  return error;
4595  }
4596  break;
4597  }
4598  }
4599 
4600  /* set up the extent map to return. */
4601  xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4602  xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4603 
4604  /* If we're done, stop now. */
4605  if (bno >= end || n >= *nmap)
4606  break;
4607 
4608  /* Else go on to the next record. */
4609  prev = got;
4610  if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4611  xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4612  else
4613  eof = 1;
4614  }
4615 
4616  *nmap = n;
4617  return 0;
4618 }
4619 
4620 
4621 STATIC int
4622 __xfs_bmapi_allocate(
4623  struct xfs_bmalloca *bma)
4624 {
4625  struct xfs_mount *mp = bma->ip->i_mount;
4626  int whichfork = (bma->flags & XFS_BMAPI_ATTRFORK) ?
4627  XFS_ATTR_FORK : XFS_DATA_FORK;
4628  struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4629  int tmp_logflags = 0;
4630  int error;
4631  int rt;
4632 
4633  ASSERT(bma->length > 0);
4634 
4635  rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(bma->ip);
4636 
4637  /*
4638  * For the wasdelay case, we could also just allocate the stuff asked
4639  * for in this bmap call but that wouldn't be as good.
4640  */
4641  if (bma->wasdel) {
4642  bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4643  bma->offset = bma->got.br_startoff;
4644  if (bma->idx != NULLEXTNUM && bma->idx) {
4645  xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4646  &bma->prev);
4647  }
4648  } else {
4649  bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4650  if (!bma->eof)
4651  bma->length = XFS_FILBLKS_MIN(bma->length,
4652  bma->got.br_startoff - bma->offset);
4653  }
4654 
4655  /*
4656  * Indicate if this is the first user data in the file, or just any
4657  * user data.
4658  */
4659  if (!(bma->flags & XFS_BMAPI_METADATA)) {
4660  bma->userdata = (bma->offset == 0) ?
4661  XFS_ALLOC_INITIAL_USER_DATA : XFS_ALLOC_USERDATA;
4662  }
4663 
4664  bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4665 
4666  /*
4667  * Only want to do the alignment at the eof if it is userdata and
4668  * allocation length is larger than a stripe unit.
4669  */
4670  if (mp->m_dalign && bma->length >= mp->m_dalign &&
4671  !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4672  error = xfs_bmap_isaeof(bma, whichfork);
4673  if (error)
4674  return error;
4675  }
4676 
4677  if (bma->flags & XFS_BMAPI_STACK_SWITCH)
4678  bma->stack_switch = 1;
4679 
4680  error = xfs_bmap_alloc(bma);
4681  if (error)
4682  return error;
4683 
4684  if (bma->flist->xbf_low)
4685  bma->minleft = 0;
4686  if (bma->cur)
4687  bma->cur->bc_private.b.firstblock = *bma->firstblock;
4688  if (bma->blkno == NULLFSBLOCK)
4689  return 0;
4690  if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4691  bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4692  bma->cur->bc_private.b.firstblock = *bma->firstblock;
4693  bma->cur->bc_private.b.flist = bma->flist;
4694  }
4695  /*
4696  * Bump the number of extents we've allocated
4697  * in this call.
4698  */
4699  bma->nallocs++;
4700 
4701  if (bma->cur)
4702  bma->cur->bc_private.b.flags =
4703  bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4704 
4705  bma->got.br_startoff = bma->offset;
4706  bma->got.br_startblock = bma->blkno;
4707  bma->got.br_blockcount = bma->length;
4708  bma->got.br_state = XFS_EXT_NORM;
4709 
4710  /*
4711  * A wasdelay extent has been initialized, so shouldn't be flagged
4712  * as unwritten.
4713  */
4714  if (!bma->wasdel && (bma->flags & XFS_BMAPI_PREALLOC) &&
4715  xfs_sb_version_hasextflgbit(&mp->m_sb))
4716  bma->got.br_state = XFS_EXT_UNWRITTEN;
4717 
4718  if (bma->wasdel)
4719  error = xfs_bmap_add_extent_delay_real(bma);
4720  else
4721  error = xfs_bmap_add_extent_hole_real(bma, whichfork);
4722 
4723  bma->logflags |= tmp_logflags;
4724  if (error)
4725  return error;
4726 
4727  /*
4728  * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4729  * or xfs_bmap_add_extent_hole_real might have merged it into one of
4730  * the neighbouring ones.
4731  */
4732  xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4733 
4734  ASSERT(bma->got.br_startoff <= bma->offset);
4735  ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4736  bma->offset + bma->length);
4737  ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4738  bma->got.br_state == XFS_EXT_UNWRITTEN);
4739  return 0;
4740 }
4741 
4742 static void
4743 xfs_bmapi_allocate_worker(
4744  struct work_struct *work)
4745 {
4746  struct xfs_bmalloca *args = container_of(work,
4747  struct xfs_bmalloca, work);
4748  unsigned long pflags;
4749 
4750  /* we are in a transaction context here */
4751  current_set_flags_nested(&pflags, PF_FSTRANS);
4752 
4753  args->result = __xfs_bmapi_allocate(args);
4754  complete(args->done);
4755 
4756  current_restore_flags_nested(&pflags, PF_FSTRANS);
4757 }
4758 
4759 /*
4760  * Some allocation requests often come in with little stack to work on. Push
4761  * them off to a worker thread so there is lots of stack to use. Otherwise just
4762  * call directly to avoid the context switch overhead here.
4763  */
4764 int
4765 xfs_bmapi_allocate(
4766  struct xfs_bmalloca *args)
4767 {
4768  DECLARE_COMPLETION_ONSTACK(done);
4769 
4770  if (!args->stack_switch)
4771  return __xfs_bmapi_allocate(args);
4772 
4773 
4774  args->done = &done;
4775  INIT_WORK_ONSTACK(&args->work, xfs_bmapi_allocate_worker);
4776  queue_work(xfs_alloc_wq, &args->work);
4777  wait_for_completion(&done);
4778  return args->result;
4779 }
4780 
4781 STATIC int
4782 xfs_bmapi_convert_unwritten(
4783  struct xfs_bmalloca *bma,
4784  struct xfs_bmbt_irec *mval,
4785  xfs_filblks_t len,
4786  int flags)
4787 {
4788  int whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4789  XFS_ATTR_FORK : XFS_DATA_FORK;
4790  struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4791  int tmp_logflags = 0;
4792  int error;
4793 
4794  /* check if we need to do unwritten->real conversion */
4795  if (mval->br_state == XFS_EXT_UNWRITTEN &&
4796  (flags & XFS_BMAPI_PREALLOC))
4797  return 0;
4798 
4799  /* check if we need to do real->unwritten conversion */
4800  if (mval->br_state == XFS_EXT_NORM &&
4801  (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4802  (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4803  return 0;
4804 
4805  /*
4806  * Modify (by adding) the state flag, if writing.
4807  */
4808  ASSERT(mval->br_blockcount <= len);
4809  if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4810  bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4811  bma->ip, whichfork);
4812  bma->cur->bc_private.b.firstblock = *bma->firstblock;
4813  bma->cur->bc_private.b.flist = bma->flist;
4814  }
4815  mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4816  ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4817 
4818  error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, &bma->idx,
4819  &bma->cur, mval, bma->firstblock, bma->flist,
4820  &tmp_logflags);
4821  bma->logflags |= tmp_logflags;
4822  if (error)
4823  return error;
4824 
4825  /*
4826  * Update our extent pointer, given that
4827  * xfs_bmap_add_extent_unwritten_real might have merged it into one
4828  * of the neighbouring ones.
4829  */
4830  xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4831 
4832  /*
4833  * We may have combined previously unwritten space with written space,
4834  * so generate another request.
4835  */
4836  if (mval->br_blockcount < len)
4837  return EAGAIN;
4838  return 0;
4839 }
4840 
4841 /*
4842  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4843  * extent state if necessary. Details behaviour is controlled by the flags
4844  * parameter. Only allocates blocks from a single allocation group, to avoid
4845  * locking problems.
4846  *
4847  * The returned value in "firstblock" from the first call in a transaction
4848  * must be remembered and presented to subsequent calls in "firstblock".
4849  * An upper bound for the number of blocks to be allocated is supplied to
4850  * the first call in "total"; if no allocation group has that many free
4851  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4852  */
4853 int
4854 xfs_bmapi_write(
4855  struct xfs_trans *tp, /* transaction pointer */
4856  struct xfs_inode *ip, /* incore inode */
4857  xfs_fileoff_t bno, /* starting file offs. mapped */
4858  xfs_filblks_t len, /* length to map in file */
4859  int flags, /* XFS_BMAPI_... */
4860  xfs_fsblock_t *firstblock, /* first allocated block
4861  controls a.g. for allocs */
4862  xfs_extlen_t total, /* total blocks needed */
4863  struct xfs_bmbt_irec *mval, /* output: map values */
4864  int *nmap, /* i/o: mval size/count */
4865  struct xfs_bmap_free *flist) /* i/o: list extents to free */
4866 {
4867  struct xfs_mount *mp = ip->i_mount;
4868  struct xfs_ifork *ifp;
4869  struct xfs_bmalloca bma = { 0 }; /* args for xfs_bmap_alloc */
4870  xfs_fileoff_t end; /* end of mapped file region */
4871  int eof; /* after the end of extents */
4872  int error; /* error return */
4873  int n; /* current extent index */
4874  xfs_fileoff_t obno; /* old block number (offset) */
4875  int whichfork; /* data or attr fork */
4876  char inhole; /* current location is hole in file */
4877  char wasdelay; /* old extent was delayed */
4878 
4879 #ifdef DEBUG
4880  xfs_fileoff_t orig_bno; /* original block number value */
4881  int orig_flags; /* original flags arg value */
4882  xfs_filblks_t orig_len; /* original value of len arg */
4883  struct xfs_bmbt_irec *orig_mval; /* original value of mval */
4884  int orig_nmap; /* original value of *nmap */
4885 
4886  orig_bno = bno;
4887  orig_len = len;
4888  orig_flags = flags;
4889  orig_mval = mval;
4890  orig_nmap = *nmap;
4891 #endif
4892 
4893  ASSERT(*nmap >= 1);
4894  ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4895  ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4896  ASSERT(tp != NULL);
4897  ASSERT(len > 0);
4898 
4899  whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4900  XFS_ATTR_FORK : XFS_DATA_FORK;
4901 
4902  if (unlikely(XFS_TEST_ERROR(
4903  (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4904  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4905  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4906  mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4907  XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4908  return XFS_ERROR(EFSCORRUPTED);
4909  }
4910 
4911  if (XFS_FORCED_SHUTDOWN(mp))
4912  return XFS_ERROR(EIO);
4913 
4914  ifp = XFS_IFORK_PTR(ip, whichfork);
4915 
4916  XFS_STATS_INC(xs_blk_mapw);
4917 
4918  if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4919  error = xfs_bmap_local_to_extents(tp, ip, firstblock, total,
4920  &bma.logflags, whichfork);
4921  if (error)
4922  goto error0;
4923  }
4924 
4925  if (*firstblock == NULLFSBLOCK) {
4926  if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4927  bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4928  else
4929  bma.minleft = 1;
4930  } else {
4931  bma.minleft = 0;
4932  }
4933 
4934  if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4935  error = xfs_iread_extents(tp, ip, whichfork);
4936  if (error)
4937  goto error0;
4938  }
4939 
4940  xfs_bmap_search_extents(ip, bno, whichfork, &eof, &bma.idx, &bma.got,
4941  &bma.prev);
4942  n = 0;
4943  end = bno + len;
4944  obno = bno;
4945 
4946  bma.tp = tp;
4947  bma.ip = ip;
4948  bma.total = total;
4949  bma.userdata = 0;
4950  bma.flist = flist;
4951  bma.firstblock = firstblock;
4952 
4953  while (bno < end && n < *nmap) {
4954  inhole = eof || bma.got.br_startoff > bno;
4955  wasdelay = !inhole && isnullstartblock(bma.got.br_startblock);
4956 
4957  /*
4958  * First, deal with the hole before the allocated space
4959  * that we found, if any.
4960  */
4961  if (inhole || wasdelay) {
4962  bma.eof = eof;
4963  bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4964  bma.wasdel = wasdelay;
4965  bma.offset = bno;
4966  bma.flags = flags;
4967 
4968  /*
4969  * There's a 32/64 bit type mismatch between the
4970  * allocation length request (which can be 64 bits in
4971  * length) and the bma length request, which is
4972  * xfs_extlen_t and therefore 32 bits. Hence we have to
4973  * check for 32-bit overflows and handle them here.
4974  */
4975  if (len > (xfs_filblks_t)MAXEXTLEN)
4976  bma.length = MAXEXTLEN;
4977  else
4978  bma.length = len;
4979 
4980  ASSERT(len > 0);
4981  ASSERT(bma.length > 0);
4982  error = xfs_bmapi_allocate(&bma);
4983  if (error)
4984  goto error0;
4985  if (bma.blkno == NULLFSBLOCK)
4986  break;
4987  }
4988 
4989  /* Deal with the allocated space we found. */
4990  xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4991  end, n, flags);
4992 
4993  /* Execute unwritten extent conversion if necessary */
4994  error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4995  if (error == EAGAIN)
4996  continue;
4997  if (error)
4998  goto error0;
4999 
5000  /* update the extent map to return */
5001  xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
5002 
5003  /*
5004  * If we're done, stop now. Stop when we've allocated
5005  * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise
5006  * the transaction may get too big.
5007  */
5008  if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
5009  break;
5010 
5011  /* Else go on to the next record. */
5012  bma.prev = bma.got;
5013  if (++bma.idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t)) {
5014  xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma.idx),
5015  &bma.got);
5016  } else
5017  eof = 1;
5018  }
5019  *nmap = n;
5020 
5021  /*
5022  * Transform from btree to extents, give it cur.
5023  */
5024  if (xfs_bmap_wants_extents(ip, whichfork)) {
5025  int tmp_logflags = 0;
5026 
5027  ASSERT(bma.cur);
5028  error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
5029  &tmp_logflags, whichfork);
5030  bma.logflags |= tmp_logflags;
5031  if (error)
5032  goto error0;
5033  }
5034 
5035  ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
5036  XFS_IFORK_NEXTENTS(ip, whichfork) >
5037  XFS_IFORK_MAXEXT(ip, whichfork));
5038  error = 0;
5039 error0:
5040  /*
5041  * Log everything. Do this after conversion, there's no point in
5042  * logging the extent records if we've converted to btree format.
5043  */
5044  if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
5045  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5046  bma.logflags &= ~xfs_ilog_fext(whichfork);
5047  else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
5048  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5049  bma.logflags &= ~xfs_ilog_fbroot(whichfork);
5050  /*
5051  * Log whatever the flags say, even if error. Otherwise we might miss
5052  * detecting a case where the data is changed, there's an error,
5053  * and it's not logged so we don't shutdown when we should.
5054  */
5055  if (bma.logflags)
5056  xfs_trans_log_inode(tp, ip, bma.logflags);
5057 
5058  if (bma.cur) {
5059  if (!error) {
5060  ASSERT(*firstblock == NULLFSBLOCK ||
5061  XFS_FSB_TO_AGNO(mp, *firstblock) ==
5062  XFS_FSB_TO_AGNO(mp,
5063  bma.cur->bc_private.b.firstblock) ||
5064  (flist->xbf_low &&
5065  XFS_FSB_TO_AGNO(mp, *firstblock) <
5066  XFS_FSB_TO_AGNO(mp,
5067  bma.cur->bc_private.b.firstblock)));
5068  *firstblock = bma.cur->bc_private.b.firstblock;
5069  }
5070  xfs_btree_del_cursor(bma.cur,
5071  error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5072  }
5073  if (!error)
5074  xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
5075  orig_nmap, *nmap);
5076  return error;
5077 }
5078 
5079 /*
5080  * Unmap (remove) blocks from a file.
5081  * If nexts is nonzero then the number of extents to remove is limited to
5082  * that value. If not all extents in the block range can be removed then
5083  * *done is set.
5084  */
5085 int /* error */
5086 xfs_bunmapi(
5087  xfs_trans_t *tp, /* transaction pointer */
5088  struct xfs_inode *ip, /* incore inode */
5089  xfs_fileoff_t bno, /* starting offset to unmap */
5090  xfs_filblks_t len, /* length to unmap in file */
5091  int flags, /* misc flags */
5092  xfs_extnum_t nexts, /* number of extents max */
5093  xfs_fsblock_t *firstblock, /* first allocated block
5094  controls a.g. for allocs */
5095  xfs_bmap_free_t *flist, /* i/o: list extents to free */
5096  int *done) /* set if not done yet */
5097 {
5098  xfs_btree_cur_t *cur; /* bmap btree cursor */
5099  xfs_bmbt_irec_t del; /* extent being deleted */
5100  int eof; /* is deleting at eof */
5101  xfs_bmbt_rec_host_t *ep; /* extent record pointer */
5102  int error; /* error return value */
5103  xfs_extnum_t extno; /* extent number in list */
5104  xfs_bmbt_irec_t got; /* current extent record */
5105  xfs_ifork_t *ifp; /* inode fork pointer */
5106  int isrt; /* freeing in rt area */
5107  xfs_extnum_t lastx; /* last extent index used */
5108  int logflags; /* transaction logging flags */
5109  xfs_extlen_t mod; /* rt extent offset */
5110  xfs_mount_t *mp; /* mount structure */
5111  xfs_extnum_t nextents; /* number of file extents */
5112  xfs_bmbt_irec_t prev; /* previous extent record */
5113  xfs_fileoff_t start; /* first file offset deleted */
5114  int tmp_logflags; /* partial logging flags */
5115  int wasdel; /* was a delayed alloc extent */
5116  int whichfork; /* data or attribute fork */
5117  xfs_fsblock_t sum;
5118 
5119  trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5120 
5121  whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5122  XFS_ATTR_FORK : XFS_DATA_FORK;
5123  ifp = XFS_IFORK_PTR(ip, whichfork);
5124  if (unlikely(
5125  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5126  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5127  XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5128  ip->i_mount);
5129  return XFS_ERROR(EFSCORRUPTED);
5130  }
5131  mp = ip->i_mount;
5132  if (XFS_FORCED_SHUTDOWN(mp))
5133  return XFS_ERROR(EIO);
5134 
5135  ASSERT(len > 0);
5136  ASSERT(nexts >= 0);
5137 
5138  if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5139  (error = xfs_iread_extents(tp, ip, whichfork)))
5140  return error;
5141  nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5142  if (nextents == 0) {
5143  *done = 1;
5144  return 0;
5145  }
5146  XFS_STATS_INC(xs_blk_unmap);
5147  isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5148  start = bno;
5149  bno = start + len - 1;
5150  ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5151  &prev);
5152 
5153  /*
5154  * Check to see if the given block number is past the end of the
5155  * file, back up to the last block if so...
5156  */
5157  if (eof) {
5158  ep = xfs_iext_get_ext(ifp, --lastx);
5159  xfs_bmbt_get_all(ep, &got);
5160  bno = got.br_startoff + got.br_blockcount - 1;
5161  }
5162  logflags = 0;
5163  if (ifp->if_flags & XFS_IFBROOT) {
5164  ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5165  cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5166  cur->bc_private.b.firstblock = *firstblock;
5167  cur->bc_private.b.flist = flist;
5168  cur->bc_private.b.flags = 0;
5169  } else
5170  cur = NULL;
5171 
5172  if (isrt) {
5173  /*
5174  * Synchronize by locking the bitmap inode.
5175  */
5176  xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
5177  xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5178  }
5179 
5180  extno = 0;
5181  while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5182  (nexts == 0 || extno < nexts)) {
5183  /*
5184  * Is the found extent after a hole in which bno lives?
5185  * Just back up to the previous extent, if so.
5186  */
5187  if (got.br_startoff > bno) {
5188  if (--lastx < 0)
5189  break;
5190  ep = xfs_iext_get_ext(ifp, lastx);
5191  xfs_bmbt_get_all(ep, &got);
5192  }
5193  /*
5194  * Is the last block of this extent before the range
5195  * we're supposed to delete? If so, we're done.
5196  */
5197  bno = XFS_FILEOFF_MIN(bno,
5198  got.br_startoff + got.br_blockcount - 1);
5199  if (bno < start)
5200  break;
5201  /*
5202  * Then deal with the (possibly delayed) allocated space
5203  * we found.
5204  */
5205  ASSERT(ep != NULL);
5206  del = got;
5207  wasdel = isnullstartblock(del.br_startblock);
5208  if (got.br_startoff < start) {
5209  del.br_startoff = start;
5210  del.br_blockcount -= start - got.br_startoff;
5211  if (!wasdel)
5212  del.br_startblock += start - got.br_startoff;
5213  }
5214  if (del.br_startoff + del.br_blockcount > bno + 1)
5215  del.br_blockcount = bno + 1 - del.br_startoff;
5216  sum = del.br_startblock + del.br_blockcount;
5217  if (isrt &&
5218  (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5219  /*
5220  * Realtime extent not lined up at the end.
5221  * The extent could have been split into written
5222  * and unwritten pieces, or we could just be
5223  * unmapping part of it. But we can't really
5224  * get rid of part of a realtime extent.
5225  */
5226  if (del.br_state == XFS_EXT_UNWRITTEN ||
5227  !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5228  /*
5229  * This piece is unwritten, or we're not
5230  * using unwritten extents. Skip over it.
5231  */
5232  ASSERT(bno >= mod);
5233  bno -= mod > del.br_blockcount ?
5234  del.br_blockcount : mod;
5235  if (bno < got.br_startoff) {
5236  if (--lastx >= 0)
5237  xfs_bmbt_get_all(xfs_iext_get_ext(
5238  ifp, lastx), &got);
5239  }
5240  continue;
5241  }
5242  /*
5243  * It's written, turn it unwritten.
5244  * This is better than zeroing it.
5245  */
5246  ASSERT(del.br_state == XFS_EXT_NORM);
5247  ASSERT(xfs_trans_get_block_res(tp) > 0);
5248  /*
5249  * If this spans a realtime extent boundary,
5250  * chop it back to the start of the one we end at.
5251  */
5252  if (del.br_blockcount > mod) {
5253  del.br_startoff += del.br_blockcount - mod;
5254  del.br_startblock += del.br_blockcount - mod;
5255  del.br_blockcount = mod;
5256  }
5257  del.br_state = XFS_EXT_UNWRITTEN;
5258  error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5259  &lastx, &cur, &del, firstblock, flist,
5260  &logflags);
5261  if (error)
5262  goto error0;
5263  goto nodelete;
5264  }
5265  if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5266  /*
5267  * Realtime extent is lined up at the end but not
5268  * at the front. We'll get rid of full extents if
5269  * we can.
5270  */
5271  mod = mp->m_sb.sb_rextsize - mod;
5272  if (del.br_blockcount > mod) {
5273  del.br_blockcount -= mod;
5274  del.br_startoff += mod;
5275  del.br_startblock += mod;
5276  } else if ((del.br_startoff == start &&
5277  (del.br_state == XFS_EXT_UNWRITTEN ||
5278  xfs_trans_get_block_res(tp) == 0)) ||
5279  !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5280  /*
5281  * Can't make it unwritten. There isn't
5282  * a full extent here so just skip it.
5283  */
5284  ASSERT(bno >= del.br_blockcount);
5285  bno -= del.br_blockcount;
5286  if (got.br_startoff > bno) {
5287  if (--lastx >= 0) {
5288  ep = xfs_iext_get_ext(ifp,
5289  lastx);
5290  xfs_bmbt_get_all(ep, &got);
5291  }
5292  }
5293  continue;
5294  } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5295  /*
5296  * This one is already unwritten.
5297  * It must have a written left neighbor.
5298  * Unwrite the killed part of that one and
5299  * try again.
5300  */
5301  ASSERT(lastx > 0);
5302  xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5303  lastx - 1), &prev);
5304  ASSERT(prev.br_state == XFS_EXT_NORM);
5305  ASSERT(!isnullstartblock(prev.br_startblock));
5306  ASSERT(del.br_startblock ==
5307  prev.br_startblock + prev.br_blockcount);
5308  if (prev.br_startoff < start) {
5309  mod = start - prev.br_startoff;
5310  prev.br_blockcount -= mod;
5311  prev.br_startblock += mod;
5312  prev.br_startoff = start;
5313  }
5314  prev.br_state = XFS_EXT_UNWRITTEN;
5315  lastx--;
5316  error = xfs_bmap_add_extent_unwritten_real(tp,
5317  ip, &lastx, &cur, &prev,
5318  firstblock, flist, &logflags);
5319  if (error)
5320  goto error0;
5321  goto nodelete;
5322  } else {
5323  ASSERT(del.br_state == XFS_EXT_NORM);
5324  del.br_state = XFS_EXT_UNWRITTEN;
5325  error = xfs_bmap_add_extent_unwritten_real(tp,
5326  ip, &lastx, &cur, &del,
5327  firstblock, flist, &logflags);
5328  if (error)
5329  goto error0;
5330  goto nodelete;
5331  }
5332  }
5333  if (wasdel) {
5334  ASSERT(startblockval(del.br_startblock) > 0);
5335  /* Update realtime/data freespace, unreserve quota */
5336  if (isrt) {
5337  xfs_filblks_t rtexts;
5338 
5339  rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5340  do_div(rtexts, mp->m_sb.sb_rextsize);
5341  xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5342  (int64_t)rtexts, 0);
5343  (void)xfs_trans_reserve_quota_nblks(NULL,
5344  ip, -((long)del.br_blockcount), 0,
5345  XFS_QMOPT_RES_RTBLKS);
5346  } else {
5347  xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5348  (int64_t)del.br_blockcount, 0);
5349  (void)xfs_trans_reserve_quota_nblks(NULL,
5350  ip, -((long)del.br_blockcount), 0,
5351  XFS_QMOPT_RES_REGBLKS);
5352  }
5353  ip->i_delayed_blks -= del.br_blockcount;
5354  if (cur)
5355  cur->bc_private.b.flags |=
5356  XFS_BTCUR_BPRV_WASDEL;
5357  } else if (cur)
5358  cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5359  /*
5360  * If it's the case where the directory code is running
5361  * with no block reservation, and the deleted block is in
5362  * the middle of its extent, and the resulting insert
5363  * of an extent would cause transformation to btree format,
5364  * then reject it. The calling code will then swap
5365  * blocks around instead.
5366  * We have to do this now, rather than waiting for the
5367  * conversion to btree format, since the transaction
5368  * will be dirty.
5369  */
5370  if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5371  XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5372  XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
5373  XFS_IFORK_MAXEXT(ip, whichfork) &&
5374  del.br_startoff > got.br_startoff &&
5375  del.br_startoff + del.br_blockcount <
5376  got.br_startoff + got.br_blockcount) {
5377  error = XFS_ERROR(ENOSPC);
5378  goto error0;
5379  }
5380  error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5381  &tmp_logflags, whichfork);
5382  logflags |= tmp_logflags;
5383  if (error)
5384  goto error0;
5385  bno = del.br_startoff - 1;
5386 nodelete:
5387  /*
5388  * If not done go on to the next (previous) record.
5389  */
5390  if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5391  if (lastx >= 0) {
5392  ep = xfs_iext_get_ext(ifp, lastx);
5393  if (xfs_bmbt_get_startoff(ep) > bno) {
5394  if (--lastx >= 0)
5395  ep = xfs_iext_get_ext(ifp,
5396  lastx);
5397  }
5398  xfs_bmbt_get_all(ep, &got);
5399  }
5400  extno++;
5401  }
5402  }
5403  *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5404 
5405  /*
5406  * Convert to a btree if necessary.
5407  */
5408  if (xfs_bmap_needs_btree(ip, whichfork)) {
5409  ASSERT(cur == NULL);
5410  error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5411  &cur, 0, &tmp_logflags, whichfork);
5412  logflags |= tmp_logflags;
5413  if (error)
5414  goto error0;
5415  }
5416  /*
5417  * transform from btree to extents, give it cur
5418  */
5419  else if (xfs_bmap_wants_extents(ip, whichfork)) {
5420  ASSERT(cur != NULL);
5421  error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5422  whichfork);
5423  logflags |= tmp_logflags;
5424  if (error)
5425  goto error0;
5426  }
5427  /*
5428  * transform from extents to local?
5429  */
5430  error = 0;
5431 error0:
5432  /*
5433  * Log everything. Do this after conversion, there's no point in
5434  * logging the extent records if we've converted to btree format.
5435  */
5436  if ((logflags & xfs_ilog_fext(whichfork)) &&
5437  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5438  logflags &= ~xfs_ilog_fext(whichfork);
5439  else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5440  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5441  logflags &= ~xfs_ilog_fbroot(whichfork);
5442  /*
5443  * Log inode even in the error case, if the transaction
5444  * is dirty we'll need to shut down the filesystem.
5445  */
5446  if (logflags)
5447  xfs_trans_log_inode(tp, ip, logflags);
5448  if (cur) {
5449  if (!error) {
5450  *firstblock = cur->bc_private.b.firstblock;
5451  cur->bc_private.b.allocated = 0;
5452  }
5453  xfs_btree_del_cursor(cur,
5454  error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5455  }
5456  return error;
5457 }
5458 
5459 /*
5460  * returns 1 for success, 0 if we failed to map the extent.
5461  */
5462 STATIC int
5463 xfs_getbmapx_fix_eof_hole(
5464  xfs_inode_t *ip, /* xfs incore inode pointer */
5465  struct getbmapx *out, /* output structure */
5466  int prealloced, /* this is a file with
5467  * preallocated data space */
5468  __int64_t end, /* last block requested */
5469  xfs_fsblock_t startblock)
5470 {
5471  __int64_t fixlen;
5472  xfs_mount_t *mp; /* file system mount point */
5473  xfs_ifork_t *ifp; /* inode fork pointer */
5474  xfs_extnum_t lastx; /* last extent pointer */
5475  xfs_fileoff_t fileblock;
5476 
5477  if (startblock == HOLESTARTBLOCK) {
5478  mp = ip->i_mount;
5479  out->bmv_block = -1;
5480  fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
5481  fixlen -= out->bmv_offset;
5482  if (prealloced && out->bmv_offset + out->bmv_length == end) {
5483  /* Came to hole at EOF. Trim it. */
5484  if (fixlen <= 0)
5485  return 0;
5486  out->bmv_length = fixlen;
5487  }
5488  } else {
5489  if (startblock == DELAYSTARTBLOCK)
5490  out->bmv_block = -2;
5491  else
5492  out->bmv_block = xfs_fsb_to_db(ip, startblock);
5493  fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
5494  ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
5495  if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
5496  (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
5497  out->bmv_oflags |= BMV_OF_LAST;
5498  }
5499 
5500  return 1;
5501 }
5502 
5503 /*
5504  * Get inode's extents as described in bmv, and format for output.
5505  * Calls formatter to fill the user's buffer until all extents
5506  * are mapped, until the passed-in bmv->bmv_count slots have
5507  * been filled, or until the formatter short-circuits the loop,
5508  * if it is tracking filled-in extents on its own.
5509  */
5510 int /* error code */
5511 xfs_getbmap(
5512  xfs_inode_t *ip,
5513  struct getbmapx *bmv, /* user bmap structure */
5514  xfs_bmap_format_t formatter, /* format to user */
5515  void *arg) /* formatter arg */
5516 {
5517  __int64_t bmvend; /* last block requested */
5518  int error = 0; /* return value */
5519  __int64_t fixlen; /* length for -1 case */
5520  int i; /* extent number */
5521  int lock; /* lock state */
5522  xfs_bmbt_irec_t *map; /* buffer for user's data */
5523  xfs_mount_t *mp; /* file system mount point */
5524  int nex; /* # of user extents can do */
5525  int nexleft; /* # of user extents left */
5526  int subnex; /* # of bmapi's can do */
5527  int nmap; /* number of map entries */
5528  struct getbmapx *out; /* output structure */
5529  int whichfork; /* data or attr fork */
5530  int prealloced; /* this is a file with
5531  * preallocated data space */
5532  int iflags; /* interface flags */
5533  int bmapi_flags; /* flags for xfs_bmapi */
5534  int cur_ext = 0;
5535 
5536  mp = ip->i_mount;
5537  iflags = bmv->bmv_iflags;
5538  whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5539 
5540  if (whichfork == XFS_ATTR_FORK) {
5541  if (XFS_IFORK_Q(ip)) {
5542  if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5543  ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5544  ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5545  return XFS_ERROR(EINVAL);
5546  } else if (unlikely(
5547  ip->i_d.di_aformat != 0 &&
5548  ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5549  XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5550  ip->i_mount);
5551  return XFS_ERROR(EFSCORRUPTED);
5552  }
5553 
5554  prealloced = 0;
5555  fixlen = 1LL << 32;
5556  } else {
5557  if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5558  ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5559  ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5560  return XFS_ERROR(EINVAL);
5561 
5562  if (xfs_get_extsz_hint(ip) ||
5563  ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5564  prealloced = 1;
5565  fixlen = mp->m_super->s_maxbytes;
5566  } else {
5567  prealloced = 0;
5568  fixlen = XFS_ISIZE(ip);
5569  }
5570  }
5571 
5572  if (bmv->bmv_length == -1) {
5573  fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5574  bmv->bmv_length =
5575  max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
5576  } else if (bmv->bmv_length == 0) {
5577  bmv->bmv_entries = 0;
5578  return 0;
5579  } else if (bmv->bmv_length < 0) {
5580  return XFS_ERROR(EINVAL);
5581  }
5582 
5583  nex = bmv->bmv_count - 1;
5584  if (nex <= 0)
5585  return XFS_ERROR(EINVAL);
5586  bmvend = bmv->bmv_offset + bmv->bmv_length;
5587 
5588 
5589  if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
5590  return XFS_ERROR(ENOMEM);
5591  out = kmem_zalloc(bmv->bmv_count * sizeof(struct getbmapx), KM_MAYFAIL);
5592  if (!out) {
5593  out = kmem_zalloc_large(bmv->bmv_count *
5594  sizeof(struct getbmapx));
5595  if (!out)
5596  return XFS_ERROR(ENOMEM);
5597  }
5598 
5599  xfs_ilock(ip, XFS_IOLOCK_SHARED);
5600  if (whichfork == XFS_DATA_FORK && !(iflags & BMV_IF_DELALLOC)) {
5601  if (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size) {
5602  error = xfs_flush_pages(ip, 0, -1, 0, FI_REMAPF);
5603  if (error)
5604  goto out_unlock_iolock;
5605  }
5606  /*
5607  * even after flushing the inode, there can still be delalloc
5608  * blocks on the inode beyond EOF due to speculative
5609  * preallocation. These are not removed until the release
5610  * function is called or the inode is inactivated. Hence we
5611  * cannot assert here that ip->i_delayed_blks == 0.
5612  */
5613  }
5614 
5615  lock = xfs_ilock_map_shared(ip);
5616 
5617  /*
5618  * Don't let nex be bigger than the number of extents
5619  * we can have assuming alternating holes and real extents.
5620  */
5621  if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5622  nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5623 
5624  bmapi_flags = xfs_bmapi_aflag(whichfork);
5625  if (!(iflags & BMV_IF_PREALLOC))
5626  bmapi_flags |= XFS_BMAPI_IGSTATE;
5627 
5628  /*
5629  * Allocate enough space to handle "subnex" maps at a time.
5630  */
5631  error = ENOMEM;
5632  subnex = 16;
5633  map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
5634  if (!map)
5635  goto out_unlock_ilock;
5636 
5637  bmv->bmv_entries = 0;
5638 
5639  if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
5640  (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
5641  error = 0;
5642  goto out_free_map;
5643  }
5644 
5645  nexleft = nex;
5646 
5647  do {
5648  nmap = (nexleft > subnex) ? subnex : nexleft;
5649  error = xfs_bmapi_read(ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5650  XFS_BB_TO_FSB(mp, bmv->bmv_length),
5651  map, &nmap, bmapi_flags);
5652  if (error)
5653  goto out_free_map;
5654  ASSERT(nmap <= subnex);
5655 
5656  for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5657  out[cur_ext].bmv_oflags = 0;
5658  if (map[i].br_state == XFS_EXT_UNWRITTEN)
5659  out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
5660  else if (map[i].br_startblock == DELAYSTARTBLOCK)
5661  out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
5662  out[cur_ext].bmv_offset =
5663  XFS_FSB_TO_BB(mp, map[i].br_startoff);
5664  out[cur_ext].bmv_length =
5665  XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5666  out[cur_ext].bmv_unused1 = 0;
5667  out[cur_ext].bmv_unused2 = 0;
5668 
5669  /*
5670  * delayed allocation extents that start beyond EOF can
5671  * occur due to speculative EOF allocation when the
5672  * delalloc extent is larger than the largest freespace
5673  * extent at conversion time. These extents cannot be
5674  * converted by data writeback, so can exist here even
5675  * if we are not supposed to be finding delalloc
5676  * extents.
5677  */
5678  if (map[i].br_startblock == DELAYSTARTBLOCK &&
5679  map[i].br_startoff <= XFS_B_TO_FSB(mp, XFS_ISIZE(ip)))
5680  ASSERT((iflags & BMV_IF_DELALLOC) != 0);
5681 
5682  if (map[i].br_startblock == HOLESTARTBLOCK &&
5683  whichfork == XFS_ATTR_FORK) {
5684  /* came to the end of attribute fork */
5685  out[cur_ext].bmv_oflags |= BMV_OF_LAST;
5686  goto out_free_map;
5687  }
5688 
5689  if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
5690  prealloced, bmvend,
5691  map[i].br_startblock))
5692  goto out_free_map;
5693 
5694  bmv->bmv_offset =
5695  out[cur_ext].bmv_offset +
5696  out[cur_ext].bmv_length;
5697  bmv->bmv_length =
5698  max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
5699 
5700  /*
5701  * In case we don't want to return the hole,
5702  * don't increase cur_ext so that we can reuse
5703  * it in the next loop.
5704  */
5705  if ((iflags & BMV_IF_NO_HOLES) &&
5706  map[i].br_startblock == HOLESTARTBLOCK) {
5707  memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
5708  continue;
5709  }
5710 
5711  nexleft--;
5712  bmv->bmv_entries++;
5713  cur_ext++;
5714  }
5715  } while (nmap && nexleft && bmv->bmv_length);
5716 
5717  out_free_map:
5718  kmem_free(map);
5719  out_unlock_ilock:
5720  xfs_iunlock_map_shared(ip, lock);
5721  out_unlock_iolock:
5722  xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5723 
5724  for (i = 0; i < cur_ext; i++) {
5725  int full = 0; /* user array is full */
5726 
5727  /* format results & advance arg */
5728  error = formatter(&arg, &out[i], &full);
5729  if (error || full)
5730  break;
5731  }
5732 
5733  if (is_vmalloc_addr(out))
5734  kmem_free_large(out);
5735  else
5736  kmem_free(out);
5737  return error;
5738 }
5739 
5740 #ifdef DEBUG
5741 STATIC struct xfs_buf *
5742 xfs_bmap_get_bp(
5743  struct xfs_btree_cur *cur,
5744  xfs_fsblock_t bno)
5745 {
5746  struct xfs_log_item_desc *lidp;
5747  int i;
5748 
5749  if (!cur)
5750  return NULL;
5751 
5752  for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
5753  if (!cur->bc_bufs[i])
5754  break;
5755  if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
5756  return cur->bc_bufs[i];
5757  }
5758 
5759  /* Chase down all the log items to see if the bp is there */
5760  list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
5761  struct xfs_buf_log_item *bip;
5762  bip = (struct xfs_buf_log_item *)lidp->lid_item;
5763  if (bip->bli_item.li_type == XFS_LI_BUF &&
5764  XFS_BUF_ADDR(bip->bli_buf) == bno)
5765  return bip->bli_buf;
5766  }
5767 
5768  return NULL;
5769 }
5770 
5771 STATIC void
5772 xfs_check_block(
5773  struct xfs_btree_block *block,
5774  xfs_mount_t *mp,
5775  int root,
5776  short sz)
5777 {
5778  int i, j, dmxr;
5779  __be64 *pp, *thispa; /* pointer to block address */
5780  xfs_bmbt_key_t *prevp, *keyp;
5781 
5782  ASSERT(be16_to_cpu(block->bb_level) > 0);
5783 
5784  prevp = NULL;
5785  for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
5786  dmxr = mp->m_bmap_dmxr[0];
5787  keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
5788 
5789  if (prevp) {
5790  ASSERT(be64_to_cpu(prevp->br_startoff) <
5791  be64_to_cpu(keyp->br_startoff));
5792  }
5793  prevp = keyp;
5794 
5795  /*
5796  * Compare the block numbers to see if there are dups.
5797  */
5798  if (root)
5799  pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
5800  else
5801  pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
5802 
5803  for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
5804  if (root)
5805  thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
5806  else
5807  thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
5808  if (*thispa == *pp) {
5809  xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
5810  __func__, j, i,
5811  (unsigned long long)be64_to_cpu(*thispa));
5812  panic("%s: ptrs are equal in node\n",
5813  __func__);
5814  }
5815  }
5816  }
5817 }
5818 
5819 /*
5820  * Check that the extents for the inode ip are in the right order in all
5821  * btree leaves.
5822  */
5823 
5824 STATIC void
5825 xfs_bmap_check_leaf_extents(
5826  xfs_btree_cur_t *cur, /* btree cursor or null */
5827  xfs_inode_t *ip, /* incore inode pointer */
5828  int whichfork) /* data or attr fork */
5829 {
5830  struct xfs_btree_block *block; /* current btree block */
5831  xfs_fsblock_t bno; /* block # of "block" */
5832  xfs_buf_t *bp; /* buffer for "block" */
5833  int error; /* error return value */
5834  xfs_extnum_t i=0, j; /* index into the extents list */
5835  xfs_ifork_t *ifp; /* fork structure */
5836  int level; /* btree level, for checking */
5837  xfs_mount_t *mp; /* file system mount structure */
5838  __be64 *pp; /* pointer to block address */
5839  xfs_bmbt_rec_t *ep; /* pointer to current extent */
5840  xfs_bmbt_rec_t last = {0, 0}; /* last extent in prev block */
5841  xfs_bmbt_rec_t *nextp; /* pointer to next extent */
5842  int bp_release = 0;
5843 
5844  if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
5845  return;
5846  }
5847 
5848  bno = NULLFSBLOCK;
5849  mp = ip->i_mount;
5850  ifp = XFS_IFORK_PTR(ip, whichfork);
5851  block = ifp->if_broot;
5852  /*
5853  * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5854  */
5855  level = be16_to_cpu(block->bb_level);
5856  ASSERT(level > 0);
5857  xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
5858  pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5859  bno = be64_to_cpu(*pp);
5860 
5861  ASSERT(bno != NULLDFSBNO);
5862  ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5863  ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5864 
5865  /*
5866  * Go down the tree until leaf level is reached, following the first
5867  * pointer (leftmost) at each level.
5868  */
5869  while (level-- > 0) {
5870  /* See if buf is in cur first */
5871  bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5872  if (bp) {
5873  bp_release = 0;
5874  } else {
5875  bp_release = 1;
5876  }
5877  if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5878  XFS_BMAP_BTREE_REF)))
5879  goto error_norelse;
5880  block = XFS_BUF_TO_BLOCK(bp);
5881  XFS_WANT_CORRUPTED_GOTO(
5882  xfs_bmap_sanity_check(mp, bp, level),
5883  error0);
5884  if (level == 0)
5885  break;
5886 
5887  /*
5888  * Check this block for basic sanity (increasing keys and
5889  * no duplicate blocks).
5890  */
5891 
5892  xfs_check_block(block, mp, 0, 0);
5893  pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
5894  bno = be64_to_cpu(*pp);
5895  XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
5896  if (bp_release) {
5897  bp_release = 0;
5898  xfs_trans_brelse(NULL, bp);
5899  }
5900  }
5901 
5902  /*
5903  * Here with bp and block set to the leftmost leaf node in the tree.
5904  */
5905  i = 0;
5906 
5907  /*
5908  * Loop over all leaf nodes checking that all extents are in the right order.
5909  */
5910  for (;;) {
5911  xfs_fsblock_t nextbno;
5912  xfs_extnum_t num_recs;
5913 
5914 
5915  num_recs = xfs_btree_get_numrecs(block);
5916 
5917  /*
5918  * Read-ahead the next leaf block, if any.
5919  */
5920 
5921  nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
5922 
5923  /*
5924  * Check all the extents to make sure they are OK.
5925  * If we had a previous block, the last entry should
5926  * conform with the first entry in this one.
5927  */
5928 
5929  ep = XFS_BMBT_REC_ADDR(mp, block, 1);
5930  if (i) {
5931  ASSERT(xfs_bmbt_disk_get_startoff(&last) +
5932  xfs_bmbt_disk_get_blockcount(&last) <=
5933  xfs_bmbt_disk_get_startoff(ep));
5934  }
5935  for (j = 1; j < num_recs; j++) {
5936  nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
5937  ASSERT(xfs_bmbt_disk_get_startoff(ep) +
5938  xfs_bmbt_disk_get_blockcount(ep) <=
5939  xfs_bmbt_disk_get_startoff(nextp));
5940  ep = nextp;
5941  }
5942 
5943  last = *ep;
5944  i += num_recs;
5945  if (bp_release) {
5946  bp_release = 0;
5947  xfs_trans_brelse(NULL, bp);
5948  }
5949  bno = nextbno;
5950  /*
5951  * If we've reached the end, stop.
5952  */
5953  if (bno == NULLFSBLOCK)
5954  break;
5955 
5956  bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5957  if (bp) {
5958  bp_release = 0;
5959  } else {
5960  bp_release = 1;
5961  }
5962  if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5963  XFS_BMAP_BTREE_REF)))
5964  goto error_norelse;
5965  block = XFS_BUF_TO_BLOCK(bp);
5966  }
5967  if (bp_release) {
5968  bp_release = 0;
5969  xfs_trans_brelse(NULL, bp);
5970  }
5971  return;
5972 
5973 error0:
5974  xfs_warn(mp, "%s: at error0", __func__);
5975  if (bp_release)
5976  xfs_trans_brelse(NULL, bp);
5977 error_norelse:
5978  xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
5979  __func__, i);
5980  panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
5981  return;
5982 }
5983 #endif
5984 
5985 /*
5986  * Count fsblocks of the given fork.
5987  */
5988 int /* error */
5989 xfs_bmap_count_blocks(
5990  xfs_trans_t *tp, /* transaction pointer */
5991  xfs_inode_t *ip, /* incore inode */
5992  int whichfork, /* data or attr fork */
5993  int *count) /* out: count of blocks */
5994 {
5995  struct xfs_btree_block *block; /* current btree block */
5996  xfs_fsblock_t bno; /* block # of "block" */
5997  xfs_ifork_t *ifp; /* fork structure */
5998  int level; /* btree level, for checking */
5999  xfs_mount_t *mp; /* file system mount structure */
6000  __be64 *pp; /* pointer to block address */
6001 
6002  bno = NULLFSBLOCK;
6003  mp = ip->i_mount;
6004  ifp = XFS_IFORK_PTR(ip, whichfork);
6005  if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
6006  xfs_bmap_count_leaves(ifp, 0,
6007  ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
6008  count);
6009  return 0;
6010  }
6011 
6012  /*
6013  * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6014  */
6015  block = ifp->if_broot;
6016  level = be16_to_cpu(block->bb_level);
6017  ASSERT(level > 0);
6018  pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
6019  bno = be64_to_cpu(*pp);
6020  ASSERT(bno != NULLDFSBNO);
6021  ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
6022  ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
6023 
6024  if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
6025  XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
6026  mp);
6027  return XFS_ERROR(EFSCORRUPTED);
6028  }
6029 
6030  return 0;
6031 }
6032 
6033 /*
6034  * Recursively walks each level of a btree
6035  * to count total fsblocks is use.
6036  */
6037 STATIC int /* error */
6038 xfs_bmap_count_tree(
6039  xfs_mount_t *mp, /* file system mount point */
6040  xfs_trans_t *tp, /* transaction pointer */
6041  xfs_ifork_t *ifp, /* inode fork pointer */
6042  xfs_fsblock_t blockno, /* file system block number */
6043  int levelin, /* level in btree */
6044  int *count) /* Count of blocks */
6045 {
6046  int error;
6047  xfs_buf_t *bp, *nbp;
6048  int level = levelin;
6049  __be64 *pp;
6050  xfs_fsblock_t bno = blockno;
6051  xfs_fsblock_t nextbno;
6052  struct xfs_btree_block *block, *nextblock;
6053  int numrecs;
6054 
6055  if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
6056  return error;
6057  *count += 1;
6058  block = XFS_BUF_TO_BLOCK(bp);
6059 
6060  if (--level) {
6061  /* Not at node above leaves, count this level of nodes */
6062  nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
6063  while (nextbno != NULLFSBLOCK) {
6064  if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
6065  0, &nbp, XFS_BMAP_BTREE_REF)))
6066  return error;
6067  *count += 1;
6068  nextblock = XFS_BUF_TO_BLOCK(nbp);
6069  nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
6070  xfs_trans_brelse(tp, nbp);
6071  }
6072 
6073  /* Dive to the next level */
6074  pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
6075  bno = be64_to_cpu(*pp);
6076  if (unlikely((error =
6077  xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
6078  xfs_trans_brelse(tp, bp);
6079  XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6080  XFS_ERRLEVEL_LOW, mp);
6081  return XFS_ERROR(EFSCORRUPTED);
6082  }
6083  xfs_trans_brelse(tp, bp);
6084  } else {
6085  /* count all level 1 nodes and their leaves */
6086  for (;;) {
6087  nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
6088  numrecs = be16_to_cpu(block->bb_numrecs);
6089  xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
6090  xfs_trans_brelse(tp, bp);
6091  if (nextbno == NULLFSBLOCK)
6092  break;
6093  bno = nextbno;
6094  if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6095  XFS_BMAP_BTREE_REF)))
6096  return error;
6097  *count += 1;
6098  block = XFS_BUF_TO_BLOCK(bp);
6099  }
6100  }
6101  return 0;
6102 }
6103 
6104 /*
6105  * Count leaf blocks given a range of extent records.
6106  */
6107 STATIC void
6108 xfs_bmap_count_leaves(
6109  xfs_ifork_t *ifp,
6110  xfs_extnum_t idx,
6111  int numrecs,
6112  int *count)
6113 {
6114  int b;
6115 
6116  for (b = 0; b < numrecs; b++) {
6117  xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
6118  *count += xfs_bmbt_get_blockcount(frp);
6119  }
6120 }
6121 
6122 /*
6123  * Count leaf blocks given a range of extent records originally
6124  * in btree format.
6125  */
6126 STATIC void
6127 xfs_bmap_disk_count_leaves(
6128  struct xfs_mount *mp,
6129  struct xfs_btree_block *block,
6130  int numrecs,
6131  int *count)
6132 {
6133  int b;
6134  xfs_bmbt_rec_t *frp;
6135 
6136  for (b = 1; b <= numrecs; b++) {
6137  frp = XFS_BMBT_REC_ADDR(mp, block, b);
6138  *count += xfs_bmbt_disk_get_blockcount(frp);
6139  }
6140 }
6141 
6142 /*
6143  * dead simple method of punching delalyed allocation blocks from a range in
6144  * the inode. Walks a block at a time so will be slow, but is only executed in
6145  * rare error cases so the overhead is not critical. This will alays punch out
6146  * both the start and end blocks, even if the ranges only partially overlap
6147  * them, so it is up to the caller to ensure that partial blocks are not
6148  * passed in.
6149  */
6150 int
6151 xfs_bmap_punch_delalloc_range(
6152  struct xfs_inode *ip,
6153  xfs_fileoff_t start_fsb,
6154  xfs_fileoff_t length)
6155 {
6156  xfs_fileoff_t remaining = length;
6157  int error = 0;
6158 
6159  ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
6160 
6161  do {
6162  int done;
6163  xfs_bmbt_irec_t imap;
6164  int nimaps = 1;
6165  xfs_fsblock_t firstblock;
6166  xfs_bmap_free_t flist;
6167 
6168  /*
6169  * Map the range first and check that it is a delalloc extent
6170  * before trying to unmap the range. Otherwise we will be
6171  * trying to remove a real extent (which requires a
6172  * transaction) or a hole, which is probably a bad idea...
6173  */
6174  error = xfs_bmapi_read(ip, start_fsb, 1, &imap, &nimaps,
6175  XFS_BMAPI_ENTIRE);
6176 
6177  if (error) {
6178  /* something screwed, just bail */
6179  if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
6180  xfs_alert(ip->i_mount,
6181  "Failed delalloc mapping lookup ino %lld fsb %lld.",
6182  ip->i_ino, start_fsb);
6183  }
6184  break;
6185  }
6186  if (!nimaps) {
6187  /* nothing there */
6188  goto next_block;
6189  }
6190  if (imap.br_startblock != DELAYSTARTBLOCK) {
6191  /* been converted, ignore */
6192  goto next_block;
6193  }
6194  WARN_ON(imap.br_blockcount == 0);
6195 
6196  /*
6197  * Note: while we initialise the firstblock/flist pair, they
6198  * should never be used because blocks should never be
6199  * allocated or freed for a delalloc extent and hence we need
6200  * don't cancel or finish them after the xfs_bunmapi() call.
6201  */
6202  xfs_bmap_init(&flist, &firstblock);
6203  error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
6204  &flist, &done);
6205  if (error)
6206  break;
6207 
6208  ASSERT(!flist.xbf_count && !flist.xbf_first);
6209 next_block:
6210  start_fsb++;
6211  remaining--;
6212  } while(remaining > 0);
6213 
6214  return error;
6215 }
6216 
6217 /*
6218  * Convert the given file system block to a disk block. We have to treat it
6219  * differently based on whether the file is a real time file or not, because the
6220  * bmap code does.
6221  */
6222 xfs_daddr_t
6223 xfs_fsb_to_db(struct xfs_inode *ip, xfs_fsblock_t fsb)
6224 {
6225  return (XFS_IS_REALTIME_INODE(ip) ? \
6226  (xfs_daddr_t)XFS_FSB_TO_BB((ip)->i_mount, (fsb)) : \
6227  XFS_FSB_TO_DADDR((ip)->i_mount, (fsb)));
6228 }
Generated on Thu Jan 10 2013 14:50:19 for Linux Kernel by   doxygen 1.8.2