xfs: remove if_lastex
[linux-2.6-microblaze.git] / fs / xfs / xfs_bmap.c
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_dir2_sf.h"
33 #include "xfs_dinode.h"
34 #include "xfs_inode.h"
35 #include "xfs_btree.h"
36 #include "xfs_mount.h"
37 #include "xfs_itable.h"
38 #include "xfs_dir2_data.h"
39 #include "xfs_dir2_leaf.h"
40 #include "xfs_dir2_block.h"
41 #include "xfs_inode_item.h"
42 #include "xfs_extfree_item.h"
43 #include "xfs_alloc.h"
44 #include "xfs_bmap.h"
45 #include "xfs_rtalloc.h"
46 #include "xfs_error.h"
47 #include "xfs_attr_leaf.h"
48 #include "xfs_rw.h"
49 #include "xfs_quota.h"
50 #include "xfs_trans_space.h"
51 #include "xfs_buf_item.h"
52 #include "xfs_filestream.h"
53 #include "xfs_vnodeops.h"
54 #include "xfs_trace.h"
55
56
57 #ifdef DEBUG
58 STATIC void
59 xfs_bmap_check_leaf_extents(xfs_btree_cur_t *cur, xfs_inode_t *ip, int whichfork);
60 #endif
61
62 kmem_zone_t             *xfs_bmap_free_item_zone;
63
64 /*
65  * Prototypes for internal bmap routines.
66  */
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  * Called by xfs_bmap_add_extent to handle cases converting a delayed
93  * allocation to a real allocation.
94  */
95 STATIC int                              /* error */
96 xfs_bmap_add_extent_delay_real(
97         xfs_inode_t             *ip,    /* incore inode pointer */
98         xfs_extnum_t            *idx,   /* extent number to update/insert */
99         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
100         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
101         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
102         xfs_fsblock_t           *first, /* pointer to firstblock variable */
103         xfs_bmap_free_t         *flist, /* list of extents to be freed */
104         int                     *logflagsp); /* inode logging flags */
105
106 /*
107  * Called by xfs_bmap_add_extent to handle cases converting a hole
108  * to a delayed allocation.
109  */
110 STATIC int                              /* error */
111 xfs_bmap_add_extent_hole_delay(
112         xfs_inode_t             *ip,    /* incore inode pointer */
113         xfs_extnum_t            *idx,   /* extent number to update/insert */
114         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
115         int                     *logflagsp); /* inode logging flags */
116
117 /*
118  * Called by xfs_bmap_add_extent to handle cases converting a hole
119  * to a real allocation.
120  */
121 STATIC int                              /* error */
122 xfs_bmap_add_extent_hole_real(
123         xfs_inode_t             *ip,    /* incore inode pointer */
124         xfs_extnum_t            *idx,   /* extent number to update/insert */
125         xfs_btree_cur_t         *cur,   /* if null, not a btree */
126         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
127         int                     *logflagsp, /* inode logging flags */
128         int                     whichfork); /* data or attr fork */
129
130 /*
131  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
132  * allocation to a real allocation or vice versa.
133  */
134 STATIC int                              /* error */
135 xfs_bmap_add_extent_unwritten_real(
136         xfs_inode_t             *ip,    /* incore inode pointer */
137         xfs_extnum_t            *idx,   /* extent number to update/insert */
138         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
139         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
140         int                     *logflagsp); /* inode logging flags */
141
142 /*
143  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
144  * It figures out where to ask the underlying allocator to put the new extent.
145  */
146 STATIC int                              /* error */
147 xfs_bmap_alloc(
148         xfs_bmalloca_t          *ap);   /* bmap alloc argument struct */
149
150 /*
151  * Transform a btree format file with only one leaf node, where the
152  * extents list will fit in the inode, into an extents format file.
153  * Since the file extents are already in-core, all we have to do is
154  * give up the space for the btree root and pitch the leaf block.
155  */
156 STATIC int                              /* error */
157 xfs_bmap_btree_to_extents(
158         xfs_trans_t             *tp,    /* transaction pointer */
159         xfs_inode_t             *ip,    /* incore inode pointer */
160         xfs_btree_cur_t         *cur,   /* btree cursor */
161         int                     *logflagsp, /* inode logging flags */
162         int                     whichfork); /* data or attr fork */
163
164 /*
165  * Remove the entry "free" from the free item list.  Prev points to the
166  * previous entry, unless "free" is the head of the list.
167  */
168 STATIC void
169 xfs_bmap_del_free(
170         xfs_bmap_free_t         *flist, /* free item list header */
171         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
172         xfs_bmap_free_item_t    *free); /* list item to be freed */
173
174 /*
175  * Convert an extents-format file into a btree-format file.
176  * The new file will have a root block (in the inode) and a single child block.
177  */
178 STATIC int                                      /* error */
179 xfs_bmap_extents_to_btree(
180         xfs_trans_t             *tp,            /* transaction pointer */
181         xfs_inode_t             *ip,            /* incore inode pointer */
182         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
183         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
184         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
185         int                     wasdel,         /* converting a delayed alloc */
186         int                     *logflagsp,     /* inode logging flags */
187         int                     whichfork);     /* data or attr fork */
188
189 /*
190  * Convert a local file to an extents file.
191  * This code is sort of bogus, since the file data needs to get
192  * logged so it won't be lost.  The bmap-level manipulations are ok, though.
193  */
194 STATIC int                              /* error */
195 xfs_bmap_local_to_extents(
196         xfs_trans_t     *tp,            /* transaction pointer */
197         xfs_inode_t     *ip,            /* incore inode pointer */
198         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
199         xfs_extlen_t    total,          /* total blocks needed by transaction */
200         int             *logflagsp,     /* inode logging flags */
201         int             whichfork);     /* data or attr fork */
202
203 /*
204  * Search the extents list for the inode, for the extent containing bno.
205  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
206  * *eofp will be set, and *prevp will contain the last entry (null if none).
207  * Else, *lastxp will be set to the index of the found
208  * entry; *gotp will contain the entry.
209  */
210 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
211 xfs_bmap_search_extents(
212         xfs_inode_t     *ip,            /* incore inode pointer */
213         xfs_fileoff_t   bno,            /* block number searched for */
214         int             whichfork,      /* data or attr fork */
215         int             *eofp,          /* out: end of file found */
216         xfs_extnum_t    *lastxp,        /* out: last extent index */
217         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
218         xfs_bmbt_irec_t *prevp);        /* out: previous extent entry found */
219
220 /*
221  * Check the last inode extent to determine whether this allocation will result
222  * in blocks being allocated at the end of the file. When we allocate new data
223  * blocks at the end of the file which do not start at the previous data block,
224  * we will try to align the new blocks at stripe unit boundaries.
225  */
226 STATIC int                              /* error */
227 xfs_bmap_isaeof(
228         xfs_inode_t     *ip,            /* incore inode pointer */
229         xfs_fileoff_t   off,            /* file offset in fsblocks */
230         int             whichfork,      /* data or attribute fork */
231         char            *aeof);         /* return value */
232
233 /*
234  * Compute the worst-case number of indirect blocks that will be used
235  * for ip's delayed extent of length "len".
236  */
237 STATIC xfs_filblks_t
238 xfs_bmap_worst_indlen(
239         xfs_inode_t             *ip,    /* incore inode pointer */
240         xfs_filblks_t           len);   /* delayed extent length */
241
242 #ifdef DEBUG
243 /*
244  * Perform various validation checks on the values being returned
245  * from xfs_bmapi().
246  */
247 STATIC void
248 xfs_bmap_validate_ret(
249         xfs_fileoff_t           bno,
250         xfs_filblks_t           len,
251         int                     flags,
252         xfs_bmbt_irec_t         *mval,
253         int                     nmap,
254         int                     ret_nmap);
255 #else
256 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
257 #endif /* DEBUG */
258
259 STATIC int
260 xfs_bmap_count_tree(
261         xfs_mount_t     *mp,
262         xfs_trans_t     *tp,
263         xfs_ifork_t     *ifp,
264         xfs_fsblock_t   blockno,
265         int             levelin,
266         int             *count);
267
268 STATIC void
269 xfs_bmap_count_leaves(
270         xfs_ifork_t             *ifp,
271         xfs_extnum_t            idx,
272         int                     numrecs,
273         int                     *count);
274
275 STATIC void
276 xfs_bmap_disk_count_leaves(
277         struct xfs_mount        *mp,
278         struct xfs_btree_block  *block,
279         int                     numrecs,
280         int                     *count);
281
282 /*
283  * Bmap internal routines.
284  */
285
286 STATIC int                              /* error */
287 xfs_bmbt_lookup_eq(
288         struct xfs_btree_cur    *cur,
289         xfs_fileoff_t           off,
290         xfs_fsblock_t           bno,
291         xfs_filblks_t           len,
292         int                     *stat)  /* success/failure */
293 {
294         cur->bc_rec.b.br_startoff = off;
295         cur->bc_rec.b.br_startblock = bno;
296         cur->bc_rec.b.br_blockcount = len;
297         return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
298 }
299
300 STATIC int                              /* error */
301 xfs_bmbt_lookup_ge(
302         struct xfs_btree_cur    *cur,
303         xfs_fileoff_t           off,
304         xfs_fsblock_t           bno,
305         xfs_filblks_t           len,
306         int                     *stat)  /* success/failure */
307 {
308         cur->bc_rec.b.br_startoff = off;
309         cur->bc_rec.b.br_startblock = bno;
310         cur->bc_rec.b.br_blockcount = len;
311         return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
312 }
313
314 /*
315 * Update the record referred to by cur to the value given
316  * by [off, bno, len, state].
317  * This either works (return 0) or gets an EFSCORRUPTED error.
318  */
319 STATIC int
320 xfs_bmbt_update(
321         struct xfs_btree_cur    *cur,
322         xfs_fileoff_t           off,
323         xfs_fsblock_t           bno,
324         xfs_filblks_t           len,
325         xfs_exntst_t            state)
326 {
327         union xfs_btree_rec     rec;
328
329         xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
330         return xfs_btree_update(cur, &rec);
331 }
332
333 /*
334  * Called from xfs_bmap_add_attrfork to handle btree format files.
335  */
336 STATIC int                                      /* error */
337 xfs_bmap_add_attrfork_btree(
338         xfs_trans_t             *tp,            /* transaction pointer */
339         xfs_inode_t             *ip,            /* incore inode pointer */
340         xfs_fsblock_t           *firstblock,    /* first block allocated */
341         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
342         int                     *flags)         /* inode logging flags */
343 {
344         xfs_btree_cur_t         *cur;           /* btree cursor */
345         int                     error;          /* error return value */
346         xfs_mount_t             *mp;            /* file system mount struct */
347         int                     stat;           /* newroot status */
348
349         mp = ip->i_mount;
350         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
351                 *flags |= XFS_ILOG_DBROOT;
352         else {
353                 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
354                 cur->bc_private.b.flist = flist;
355                 cur->bc_private.b.firstblock = *firstblock;
356                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
357                         goto error0;
358                 /* must be at least one entry */
359                 XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
360                 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
361                         goto error0;
362                 if (stat == 0) {
363                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
364                         return XFS_ERROR(ENOSPC);
365                 }
366                 *firstblock = cur->bc_private.b.firstblock;
367                 cur->bc_private.b.allocated = 0;
368                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
369         }
370         return 0;
371 error0:
372         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
373         return error;
374 }
375
376 /*
377  * Called from xfs_bmap_add_attrfork to handle extents format files.
378  */
379 STATIC int                                      /* error */
380 xfs_bmap_add_attrfork_extents(
381         xfs_trans_t             *tp,            /* transaction pointer */
382         xfs_inode_t             *ip,            /* incore inode pointer */
383         xfs_fsblock_t           *firstblock,    /* first block allocated */
384         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
385         int                     *flags)         /* inode logging flags */
386 {
387         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
388         int                     error;          /* error return value */
389
390         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
391                 return 0;
392         cur = NULL;
393         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
394                 flags, XFS_DATA_FORK);
395         if (cur) {
396                 cur->bc_private.b.allocated = 0;
397                 xfs_btree_del_cursor(cur,
398                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
399         }
400         return error;
401 }
402
403 /*
404  * Called from xfs_bmap_add_attrfork to handle local format files.
405  */
406 STATIC int                                      /* error */
407 xfs_bmap_add_attrfork_local(
408         xfs_trans_t             *tp,            /* transaction pointer */
409         xfs_inode_t             *ip,            /* incore inode pointer */
410         xfs_fsblock_t           *firstblock,    /* first block allocated */
411         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
412         int                     *flags)         /* inode logging flags */
413 {
414         xfs_da_args_t           dargs;          /* args for dir/attr code */
415         int                     error;          /* error return value */
416         xfs_mount_t             *mp;            /* mount structure pointer */
417
418         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
419                 return 0;
420         if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
421                 mp = ip->i_mount;
422                 memset(&dargs, 0, sizeof(dargs));
423                 dargs.dp = ip;
424                 dargs.firstblock = firstblock;
425                 dargs.flist = flist;
426                 dargs.total = mp->m_dirblkfsbs;
427                 dargs.whichfork = XFS_DATA_FORK;
428                 dargs.trans = tp;
429                 error = xfs_dir2_sf_to_block(&dargs);
430         } else
431                 error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
432                         XFS_DATA_FORK);
433         return error;
434 }
435
436 /*
437  * Called by xfs_bmapi to update file extent records and the btree
438  * after allocating space (or doing a delayed allocation).
439  */
440 STATIC int                              /* error */
441 xfs_bmap_add_extent(
442         xfs_inode_t             *ip,    /* incore inode pointer */
443         xfs_extnum_t            *idx,   /* extent number to update/insert */
444         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
445         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
446         xfs_fsblock_t           *first, /* pointer to firstblock variable */
447         xfs_bmap_free_t         *flist, /* list of extents to be freed */
448         int                     *logflagsp, /* inode logging flags */
449         int                     whichfork) /* data or attr fork */
450 {
451         xfs_btree_cur_t         *cur;   /* btree cursor or null */
452         xfs_filblks_t           da_new; /* new count del alloc blocks used */
453         xfs_filblks_t           da_old; /* old count del alloc blocks used */
454         int                     error;  /* error return value */
455         xfs_ifork_t             *ifp;   /* inode fork ptr */
456         int                     logflags; /* returned value */
457         xfs_extnum_t            nextents; /* number of extents in file now */
458
459         XFS_STATS_INC(xs_add_exlist);
460
461         cur = *curp;
462         ifp = XFS_IFORK_PTR(ip, whichfork);
463         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
464         da_old = da_new = 0;
465         error = 0;
466
467         ASSERT(*idx >= 0);
468         ASSERT(*idx <= nextents);
469
470         /*
471          * This is the first extent added to a new/empty file.
472          * Special case this one, so other routines get to assume there are
473          * already extents in the list.
474          */
475         if (nextents == 0) {
476                 xfs_iext_insert(ip, *idx, 1, new,
477                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
478
479                 ASSERT(cur == NULL);
480
481                 if (!isnullstartblock(new->br_startblock)) {
482                         XFS_IFORK_NEXT_SET(ip, whichfork, 1);
483                         logflags = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
484                 } else
485                         logflags = 0;
486         }
487         /*
488          * Any kind of new delayed allocation goes here.
489          */
490         else if (isnullstartblock(new->br_startblock)) {
491                 if (cur)
492                         ASSERT((cur->bc_private.b.flags &
493                                 XFS_BTCUR_BPRV_WASDEL) == 0);
494                 error = xfs_bmap_add_extent_hole_delay(ip, idx, new,
495                                                        &logflags);
496         }
497         /*
498          * Real allocation off the end of the file.
499          */
500         else if (*idx == nextents) {
501                 if (cur)
502                         ASSERT((cur->bc_private.b.flags &
503                                 XFS_BTCUR_BPRV_WASDEL) == 0);
504                 error = xfs_bmap_add_extent_hole_real(ip, idx, cur, new,
505                                 &logflags, whichfork);
506         } else {
507                 xfs_bmbt_irec_t prev;   /* old extent at offset idx */
508
509                 /*
510                  * Get the record referred to by idx.
511                  */
512                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &prev);
513                 /*
514                  * If it's a real allocation record, and the new allocation ends
515                  * after the start of the referred to record, then we're filling
516                  * in a delayed or unwritten allocation with a real one, or
517                  * converting real back to unwritten.
518                  */
519                 if (!isnullstartblock(new->br_startblock) &&
520                     new->br_startoff + new->br_blockcount > prev.br_startoff) {
521                         if (prev.br_state != XFS_EXT_UNWRITTEN &&
522                             isnullstartblock(prev.br_startblock)) {
523                                 da_old = startblockval(prev.br_startblock);
524                                 if (cur)
525                                         ASSERT(cur->bc_private.b.flags &
526                                                 XFS_BTCUR_BPRV_WASDEL);
527                                 error = xfs_bmap_add_extent_delay_real(ip,
528                                                 idx, &cur, new, &da_new,
529                                                 first, flist, &logflags);
530                         } else {
531                                 ASSERT(new->br_state == XFS_EXT_NORM ||
532                                        new->br_state == XFS_EXT_UNWRITTEN);
533
534                                 error = xfs_bmap_add_extent_unwritten_real(ip,
535                                                 idx, &cur, new, &logflags);
536                                 if (error)
537                                         goto done;
538                         }
539                 }
540                 /*
541                  * Otherwise we're filling in a hole with an allocation.
542                  */
543                 else {
544                         if (cur)
545                                 ASSERT((cur->bc_private.b.flags &
546                                         XFS_BTCUR_BPRV_WASDEL) == 0);
547                         error = xfs_bmap_add_extent_hole_real(ip, idx, cur,
548                                         new, &logflags, whichfork);
549                 }
550         }
551
552         if (error)
553                 goto done;
554         ASSERT(*curp == cur || *curp == NULL);
555
556         /*
557          * Convert to a btree if necessary.
558          */
559         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
560             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
561                 int     tmp_logflags;   /* partial log flag return val */
562
563                 ASSERT(cur == NULL);
564                 error = xfs_bmap_extents_to_btree(ip->i_transp, ip, first,
565                         flist, &cur, da_old > 0, &tmp_logflags, whichfork);
566                 logflags |= tmp_logflags;
567                 if (error)
568                         goto done;
569         }
570         /*
571          * Adjust for changes in reserved delayed indirect blocks.
572          * Nothing to do for disk quotas here.
573          */
574         if (da_old || da_new) {
575                 xfs_filblks_t   nblks;
576
577                 nblks = da_new;
578                 if (cur)
579                         nblks += cur->bc_private.b.allocated;
580                 ASSERT(nblks <= da_old);
581                 if (nblks < da_old)
582                         xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
583                                 (int64_t)(da_old - nblks), 0);
584         }
585         /*
586          * Clear out the allocated field, done with it now in any case.
587          */
588         if (cur) {
589                 cur->bc_private.b.allocated = 0;
590                 *curp = cur;
591         }
592 done:
593 #ifdef DEBUG
594         if (!error)
595                 xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
596 #endif
597         *logflagsp = logflags;
598         return error;
599 }
600
601 /*
602  * Called by xfs_bmap_add_extent to handle cases converting a delayed
603  * allocation to a real allocation.
604  */
605 STATIC int                              /* error */
606 xfs_bmap_add_extent_delay_real(
607         xfs_inode_t             *ip,    /* incore inode pointer */
608         xfs_extnum_t            *idx,   /* extent number to update/insert */
609         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
610         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
611         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
612         xfs_fsblock_t           *first, /* pointer to firstblock variable */
613         xfs_bmap_free_t         *flist, /* list of extents to be freed */
614         int                     *logflagsp) /* inode logging flags */
615 {
616         xfs_btree_cur_t         *cur;   /* btree cursor */
617         int                     diff;   /* temp value */
618         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
619         int                     error;  /* error return value */
620         int                     i;      /* temp state */
621         xfs_ifork_t             *ifp;   /* inode fork pointer */
622         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
623         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
624                                         /* left is 0, right is 1, prev is 2 */
625         int                     rval=0; /* return value (logging flags) */
626         int                     state = 0;/* state bits, accessed thru macros */
627         xfs_filblks_t           temp=0; /* value for dnew calculations */
628         xfs_filblks_t           temp2=0;/* value for dnew calculations */
629         int                     tmp_rval;       /* partial logging flags */
630
631 #define LEFT            r[0]
632 #define RIGHT           r[1]
633 #define PREV            r[2]
634
635         /*
636          * Set up a bunch of variables to make the tests simpler.
637          */
638         cur = *curp;
639         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
640         ep = xfs_iext_get_ext(ifp, *idx);
641         xfs_bmbt_get_all(ep, &PREV);
642         new_endoff = new->br_startoff + new->br_blockcount;
643         ASSERT(PREV.br_startoff <= new->br_startoff);
644         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
645
646         /*
647          * Set flags determining what part of the previous delayed allocation
648          * extent is being replaced by a real allocation.
649          */
650         if (PREV.br_startoff == new->br_startoff)
651                 state |= BMAP_LEFT_FILLING;
652         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
653                 state |= BMAP_RIGHT_FILLING;
654
655         /*
656          * Check and set flags if this segment has a left neighbor.
657          * Don't set contiguous if the combined extent would be too large.
658          */
659         if (*idx > 0) {
660                 state |= BMAP_LEFT_VALID;
661                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
662
663                 if (isnullstartblock(LEFT.br_startblock))
664                         state |= BMAP_LEFT_DELAY;
665         }
666
667         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
668             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
669             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
670             LEFT.br_state == new->br_state &&
671             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
672                 state |= BMAP_LEFT_CONTIG;
673
674         /*
675          * Check and set flags if this segment has a right neighbor.
676          * Don't set contiguous if the combined extent would be too large.
677          * Also check for all-three-contiguous being too large.
678          */
679         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
680                 state |= BMAP_RIGHT_VALID;
681                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
682
683                 if (isnullstartblock(RIGHT.br_startblock))
684                         state |= BMAP_RIGHT_DELAY;
685         }
686
687         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
688             new_endoff == RIGHT.br_startoff &&
689             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
690             new->br_state == RIGHT.br_state &&
691             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
692             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
693                        BMAP_RIGHT_FILLING)) !=
694                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
695                        BMAP_RIGHT_FILLING) ||
696              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
697                         <= MAXEXTLEN))
698                 state |= BMAP_RIGHT_CONTIG;
699
700         error = 0;
701         /*
702          * Switch out based on the FILLING and CONTIG state bits.
703          */
704         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
705                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
706         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
707              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
708                 /*
709                  * Filling in all of a previously delayed allocation extent.
710                  * The left and right neighbors are both contiguous with new.
711                  */
712                 --*idx;
713                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
714                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
715                         LEFT.br_blockcount + PREV.br_blockcount +
716                         RIGHT.br_blockcount);
717                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
718
719                 xfs_iext_remove(ip, *idx + 1, 2, state);
720                 ip->i_d.di_nextents--;
721                 if (cur == NULL)
722                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
723                 else {
724                         rval = XFS_ILOG_CORE;
725                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
726                                         RIGHT.br_startblock,
727                                         RIGHT.br_blockcount, &i)))
728                                 goto done;
729                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
730                         if ((error = xfs_btree_delete(cur, &i)))
731                                 goto done;
732                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
733                         if ((error = xfs_btree_decrement(cur, 0, &i)))
734                                 goto done;
735                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
736                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
737                                         LEFT.br_startblock,
738                                         LEFT.br_blockcount +
739                                         PREV.br_blockcount +
740                                         RIGHT.br_blockcount, LEFT.br_state)))
741                                 goto done;
742                 }
743                 *dnew = 0;
744                 break;
745
746         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
747                 /*
748                  * Filling in all of a previously delayed allocation extent.
749                  * The left neighbor is contiguous, the right is not.
750                  */
751                 --*idx;
752
753                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
754                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
755                         LEFT.br_blockcount + PREV.br_blockcount);
756                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
757
758                 xfs_iext_remove(ip, *idx + 1, 1, state);
759                 if (cur == NULL)
760                         rval = XFS_ILOG_DEXT;
761                 else {
762                         rval = 0;
763                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
764                                         LEFT.br_startblock, LEFT.br_blockcount,
765                                         &i)))
766                                 goto done;
767                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
768                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
769                                         LEFT.br_startblock,
770                                         LEFT.br_blockcount +
771                                         PREV.br_blockcount, LEFT.br_state)))
772                                 goto done;
773                 }
774                 *dnew = 0;
775                 break;
776
777         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
778                 /*
779                  * Filling in all of a previously delayed allocation extent.
780                  * The right neighbor is contiguous, the left is not.
781                  */
782                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
783                 xfs_bmbt_set_startblock(ep, new->br_startblock);
784                 xfs_bmbt_set_blockcount(ep,
785                         PREV.br_blockcount + RIGHT.br_blockcount);
786                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
787
788                 xfs_iext_remove(ip, *idx + 1, 1, state);
789                 if (cur == NULL)
790                         rval = XFS_ILOG_DEXT;
791                 else {
792                         rval = 0;
793                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
794                                         RIGHT.br_startblock,
795                                         RIGHT.br_blockcount, &i)))
796                                 goto done;
797                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
798                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
799                                         new->br_startblock,
800                                         PREV.br_blockcount +
801                                         RIGHT.br_blockcount, PREV.br_state)))
802                                 goto done;
803                 }
804
805                 *dnew = 0;
806                 break;
807
808         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
809                 /*
810                  * Filling in all of a previously delayed allocation extent.
811                  * Neither the left nor right neighbors are contiguous with
812                  * the new one.
813                  */
814                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
815                 xfs_bmbt_set_startblock(ep, new->br_startblock);
816                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
817
818                 ip->i_d.di_nextents++;
819                 if (cur == NULL)
820                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
821                 else {
822                         rval = XFS_ILOG_CORE;
823                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
824                                         new->br_startblock, new->br_blockcount,
825                                         &i)))
826                                 goto done;
827                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
828                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
829                         if ((error = xfs_btree_insert(cur, &i)))
830                                 goto done;
831                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
832                 }
833
834                 *dnew = 0;
835                 break;
836
837         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
838                 /*
839                  * Filling in the first part of a previous delayed allocation.
840                  * The left neighbor is contiguous.
841                  */
842                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
843                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
844                         LEFT.br_blockcount + new->br_blockcount);
845                 xfs_bmbt_set_startoff(ep,
846                         PREV.br_startoff + new->br_blockcount);
847                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
848
849                 temp = PREV.br_blockcount - new->br_blockcount;
850                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
851                 xfs_bmbt_set_blockcount(ep, temp);
852                 if (cur == NULL)
853                         rval = XFS_ILOG_DEXT;
854                 else {
855                         rval = 0;
856                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
857                                         LEFT.br_startblock, LEFT.br_blockcount,
858                                         &i)))
859                                 goto done;
860                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
861                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
862                                         LEFT.br_startblock,
863                                         LEFT.br_blockcount +
864                                         new->br_blockcount,
865                                         LEFT.br_state)))
866                                 goto done;
867                 }
868                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
869                         startblockval(PREV.br_startblock));
870                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
871                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
872
873                 --*idx;
874                 *dnew = temp;
875                 break;
876
877         case BMAP_LEFT_FILLING:
878                 /*
879                  * Filling in the first part of a previous delayed allocation.
880                  * The left neighbor is not contiguous.
881                  */
882                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
883                 xfs_bmbt_set_startoff(ep, new_endoff);
884                 temp = PREV.br_blockcount - new->br_blockcount;
885                 xfs_bmbt_set_blockcount(ep, temp);
886                 xfs_iext_insert(ip, *idx, 1, new, state);
887                 ip->i_d.di_nextents++;
888                 if (cur == NULL)
889                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
890                 else {
891                         rval = XFS_ILOG_CORE;
892                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
893                                         new->br_startblock, new->br_blockcount,
894                                         &i)))
895                                 goto done;
896                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
897                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
898                         if ((error = xfs_btree_insert(cur, &i)))
899                                 goto done;
900                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
901                 }
902                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
903                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
904                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
905                                         first, flist, &cur, 1, &tmp_rval,
906                                         XFS_DATA_FORK);
907                         rval |= tmp_rval;
908                         if (error)
909                                 goto done;
910                 }
911                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
912                         startblockval(PREV.br_startblock) -
913                         (cur ? cur->bc_private.b.allocated : 0));
914                 ep = xfs_iext_get_ext(ifp, *idx + 1);
915                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
916                 trace_xfs_bmap_post_update(ip, *idx + 1, state, _THIS_IP_);
917
918                 *dnew = temp;
919                 break;
920
921         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
922                 /*
923                  * Filling in the last part of a previous delayed allocation.
924                  * The right neighbor is contiguous with the new allocation.
925                  */
926                 temp = PREV.br_blockcount - new->br_blockcount;
927                 trace_xfs_bmap_pre_update(ip, *idx + 1, state, _THIS_IP_);
928                 xfs_bmbt_set_blockcount(ep, temp);
929                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx + 1),
930                         new->br_startoff, new->br_startblock,
931                         new->br_blockcount + RIGHT.br_blockcount,
932                         RIGHT.br_state);
933                 trace_xfs_bmap_post_update(ip, *idx + 1, state, _THIS_IP_);
934                 if (cur == NULL)
935                         rval = XFS_ILOG_DEXT;
936                 else {
937                         rval = 0;
938                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
939                                         RIGHT.br_startblock,
940                                         RIGHT.br_blockcount, &i)))
941                                 goto done;
942                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
943                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
944                                         new->br_startblock,
945                                         new->br_blockcount +
946                                         RIGHT.br_blockcount,
947                                         RIGHT.br_state)))
948                                 goto done;
949                 }
950
951                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
952                         startblockval(PREV.br_startblock));
953                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
954                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
955                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
956
957                 ++*idx;
958                 *dnew = temp;
959                 break;
960
961         case BMAP_RIGHT_FILLING:
962                 /*
963                  * Filling in the last part of a previous delayed allocation.
964                  * The right neighbor is not contiguous.
965                  */
966                 temp = PREV.br_blockcount - new->br_blockcount;
967                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
968                 xfs_bmbt_set_blockcount(ep, temp);
969                 xfs_iext_insert(ip, *idx + 1, 1, new, state);
970                 ip->i_d.di_nextents++;
971                 if (cur == NULL)
972                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
973                 else {
974                         rval = XFS_ILOG_CORE;
975                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
976                                         new->br_startblock, new->br_blockcount,
977                                         &i)))
978                                 goto done;
979                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
980                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
981                         if ((error = xfs_btree_insert(cur, &i)))
982                                 goto done;
983                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
984                 }
985                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
986                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
987                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
988                                 first, flist, &cur, 1, &tmp_rval,
989                                 XFS_DATA_FORK);
990                         rval |= tmp_rval;
991                         if (error)
992                                 goto done;
993                 }
994                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
995                         startblockval(PREV.br_startblock) -
996                         (cur ? cur->bc_private.b.allocated : 0));
997                 ep = xfs_iext_get_ext(ifp, *idx);
998                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
999                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1000
1001                 ++*idx;
1002                 *dnew = temp;
1003                 break;
1004
1005         case 0:
1006                 /*
1007                  * Filling in the middle part of a previous delayed allocation.
1008                  * Contiguity is impossible here.
1009                  * This case is avoided almost all the time.
1010                  *
1011                  * We start with a delayed allocation:
1012                  *
1013                  * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
1014                  *  PREV @ idx
1015                  *
1016                  * and we are allocating:
1017                  *                     +rrrrrrrrrrrrrrrrr+
1018                  *                            new
1019                  *
1020                  * and we set it up for insertion as:
1021                  * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
1022                  *                            new
1023                  *  PREV @ idx          LEFT              RIGHT
1024                  *                      inserted at idx + 1
1025                  */
1026                 temp = new->br_startoff - PREV.br_startoff;
1027                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
1028                 trace_xfs_bmap_pre_update(ip, *idx, 0, _THIS_IP_);
1029                 xfs_bmbt_set_blockcount(ep, temp);      /* truncate PREV */
1030                 LEFT = *new;
1031                 RIGHT.br_state = PREV.br_state;
1032                 RIGHT.br_startblock = nullstartblock(
1033                                 (int)xfs_bmap_worst_indlen(ip, temp2));
1034                 RIGHT.br_startoff = new_endoff;
1035                 RIGHT.br_blockcount = temp2;
1036                 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
1037                 xfs_iext_insert(ip, *idx + 1, 2, &LEFT, state);
1038                 ip->i_d.di_nextents++;
1039                 if (cur == NULL)
1040                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1041                 else {
1042                         rval = XFS_ILOG_CORE;
1043                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1044                                         new->br_startblock, new->br_blockcount,
1045                                         &i)))
1046                                 goto done;
1047                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1048                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1049                         if ((error = xfs_btree_insert(cur, &i)))
1050                                 goto done;
1051                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1052                 }
1053                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1054                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1055                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1056                                         first, flist, &cur, 1, &tmp_rval,
1057                                         XFS_DATA_FORK);
1058                         rval |= tmp_rval;
1059                         if (error)
1060                                 goto done;
1061                 }
1062                 temp = xfs_bmap_worst_indlen(ip, temp);
1063                 temp2 = xfs_bmap_worst_indlen(ip, temp2);
1064                 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
1065                         (cur ? cur->bc_private.b.allocated : 0));
1066                 if (diff > 0 &&
1067                     xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
1068                                              -((int64_t)diff), 0)) {
1069                         /*
1070                          * Ick gross gag me with a spoon.
1071                          */
1072                         ASSERT(0);      /* want to see if this ever happens! */
1073                         while (diff > 0) {
1074                                 if (temp) {
1075                                         temp--;
1076                                         diff--;
1077                                         if (!diff ||
1078                                             !xfs_icsb_modify_counters(ip->i_mount,
1079                                                     XFS_SBS_FDBLOCKS,
1080                                                     -((int64_t)diff), 0))
1081                                                 break;
1082                                 }
1083                                 if (temp2) {
1084                                         temp2--;
1085                                         diff--;
1086                                         if (!diff ||
1087                                             !xfs_icsb_modify_counters(ip->i_mount,
1088                                                     XFS_SBS_FDBLOCKS,
1089                                                     -((int64_t)diff), 0))
1090                                                 break;
1091                                 }
1092                         }
1093                 }
1094                 ep = xfs_iext_get_ext(ifp, *idx);
1095                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
1096                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1097                 trace_xfs_bmap_pre_update(ip, *idx + 2, state, _THIS_IP_);
1098                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx + 2),
1099                         nullstartblock((int)temp2));
1100                 trace_xfs_bmap_post_update(ip, *idx + 2, state, _THIS_IP_);
1101
1102                 ++*idx;
1103                 *dnew = temp + temp2;
1104                 break;
1105
1106         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1107         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1108         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1109         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1110         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1111         case BMAP_LEFT_CONTIG:
1112         case BMAP_RIGHT_CONTIG:
1113                 /*
1114                  * These cases are all impossible.
1115                  */
1116                 ASSERT(0);
1117         }
1118         *curp = cur;
1119 done:
1120         *logflagsp = rval;
1121         return error;
1122 #undef  LEFT
1123 #undef  RIGHT
1124 #undef  PREV
1125 }
1126
1127 /*
1128  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
1129  * allocation to a real allocation or vice versa.
1130  */
1131 STATIC int                              /* error */
1132 xfs_bmap_add_extent_unwritten_real(
1133         xfs_inode_t             *ip,    /* incore inode pointer */
1134         xfs_extnum_t            *idx,   /* extent number to update/insert */
1135         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
1136         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1137         int                     *logflagsp) /* inode logging flags */
1138 {
1139         xfs_btree_cur_t         *cur;   /* btree cursor */
1140         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
1141         int                     error;  /* error return value */
1142         int                     i;      /* temp state */
1143         xfs_ifork_t             *ifp;   /* inode fork pointer */
1144         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1145         xfs_exntst_t            newext; /* new extent state */
1146         xfs_exntst_t            oldext; /* old extent state */
1147         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1148                                         /* left is 0, right is 1, prev is 2 */
1149         int                     rval=0; /* return value (logging flags) */
1150         int                     state = 0;/* state bits, accessed thru macros */
1151
1152 #define LEFT            r[0]
1153 #define RIGHT           r[1]
1154 #define PREV            r[2]
1155         /*
1156          * Set up a bunch of variables to make the tests simpler.
1157          */
1158         error = 0;
1159         cur = *curp;
1160         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1161         ep = xfs_iext_get_ext(ifp, *idx);
1162         xfs_bmbt_get_all(ep, &PREV);
1163         newext = new->br_state;
1164         oldext = (newext == XFS_EXT_UNWRITTEN) ?
1165                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
1166         ASSERT(PREV.br_state == oldext);
1167         new_endoff = new->br_startoff + new->br_blockcount;
1168         ASSERT(PREV.br_startoff <= new->br_startoff);
1169         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1170
1171         /*
1172          * Set flags determining what part of the previous oldext allocation
1173          * extent is being replaced by a newext allocation.
1174          */
1175         if (PREV.br_startoff == new->br_startoff)
1176                 state |= BMAP_LEFT_FILLING;
1177         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1178                 state |= BMAP_RIGHT_FILLING;
1179
1180         /*
1181          * Check and set flags if this segment has a left neighbor.
1182          * Don't set contiguous if the combined extent would be too large.
1183          */
1184         if (*idx > 0) {
1185                 state |= BMAP_LEFT_VALID;
1186                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
1187
1188                 if (isnullstartblock(LEFT.br_startblock))
1189                         state |= BMAP_LEFT_DELAY;
1190         }
1191
1192         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1193             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1194             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1195             LEFT.br_state == newext &&
1196             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1197                 state |= BMAP_LEFT_CONTIG;
1198
1199         /*
1200          * Check and set flags if this segment has a right neighbor.
1201          * Don't set contiguous if the combined extent would be too large.
1202          * Also check for all-three-contiguous being too large.
1203          */
1204         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1205                 state |= BMAP_RIGHT_VALID;
1206                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
1207                 if (isnullstartblock(RIGHT.br_startblock))
1208                         state |= BMAP_RIGHT_DELAY;
1209         }
1210
1211         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1212             new_endoff == RIGHT.br_startoff &&
1213             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1214             newext == RIGHT.br_state &&
1215             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1216             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1217                        BMAP_RIGHT_FILLING)) !=
1218                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1219                        BMAP_RIGHT_FILLING) ||
1220              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1221                         <= MAXEXTLEN))
1222                 state |= BMAP_RIGHT_CONTIG;
1223
1224         /*
1225          * Switch out based on the FILLING and CONTIG state bits.
1226          */
1227         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1228                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1229         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1230              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1231                 /*
1232                  * Setting all of a previous oldext extent to newext.
1233                  * The left and right neighbors are both contiguous with new.
1234                  */
1235                 --*idx;
1236
1237                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1238                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1239                         LEFT.br_blockcount + PREV.br_blockcount +
1240                         RIGHT.br_blockcount);
1241                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1242
1243                 xfs_iext_remove(ip, *idx + 1, 2, state);
1244                 ip->i_d.di_nextents -= 2;
1245                 if (cur == NULL)
1246                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1247                 else {
1248                         rval = XFS_ILOG_CORE;
1249                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1250                                         RIGHT.br_startblock,
1251                                         RIGHT.br_blockcount, &i)))
1252                                 goto done;
1253                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1254                         if ((error = xfs_btree_delete(cur, &i)))
1255                                 goto done;
1256                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1257                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1258                                 goto done;
1259                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1260                         if ((error = xfs_btree_delete(cur, &i)))
1261                                 goto done;
1262                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1263                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1264                                 goto done;
1265                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1266                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1267                                 LEFT.br_startblock,
1268                                 LEFT.br_blockcount + PREV.br_blockcount +
1269                                 RIGHT.br_blockcount, LEFT.br_state)))
1270                                 goto done;
1271                 }
1272                 break;
1273
1274         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1275                 /*
1276                  * Setting all of a previous oldext extent to newext.
1277                  * The left neighbor is contiguous, the right is not.
1278                  */
1279                 --*idx;
1280
1281                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1282                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1283                         LEFT.br_blockcount + PREV.br_blockcount);
1284                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1285
1286                 xfs_iext_remove(ip, *idx + 1, 1, state);
1287                 ip->i_d.di_nextents--;
1288                 if (cur == NULL)
1289                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1290                 else {
1291                         rval = XFS_ILOG_CORE;
1292                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1293                                         PREV.br_startblock, PREV.br_blockcount,
1294                                         &i)))
1295                                 goto done;
1296                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1297                         if ((error = xfs_btree_delete(cur, &i)))
1298                                 goto done;
1299                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1300                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1301                                 goto done;
1302                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1303                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1304                                 LEFT.br_startblock,
1305                                 LEFT.br_blockcount + PREV.br_blockcount,
1306                                 LEFT.br_state)))
1307                                 goto done;
1308                 }
1309                 break;
1310
1311         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1312                 /*
1313                  * Setting all of a previous oldext extent to newext.
1314                  * The right neighbor is contiguous, the left is not.
1315                  */
1316                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1317                 xfs_bmbt_set_blockcount(ep,
1318                         PREV.br_blockcount + RIGHT.br_blockcount);
1319                 xfs_bmbt_set_state(ep, newext);
1320                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1321                 xfs_iext_remove(ip, *idx + 1, 1, state);
1322                 ip->i_d.di_nextents--;
1323                 if (cur == NULL)
1324                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1325                 else {
1326                         rval = XFS_ILOG_CORE;
1327                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1328                                         RIGHT.br_startblock,
1329                                         RIGHT.br_blockcount, &i)))
1330                                 goto done;
1331                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1332                         if ((error = xfs_btree_delete(cur, &i)))
1333                                 goto done;
1334                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1335                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1336                                 goto done;
1337                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1338                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1339                                 new->br_startblock,
1340                                 new->br_blockcount + RIGHT.br_blockcount,
1341                                 newext)))
1342                                 goto done;
1343                 }
1344                 break;
1345
1346         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1347                 /*
1348                  * Setting all of a previous oldext extent to newext.
1349                  * Neither the left nor right neighbors are contiguous with
1350                  * the new one.
1351                  */
1352                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1353                 xfs_bmbt_set_state(ep, newext);
1354                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1355
1356                 if (cur == NULL)
1357                         rval = XFS_ILOG_DEXT;
1358                 else {
1359                         rval = 0;
1360                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1361                                         new->br_startblock, new->br_blockcount,
1362                                         &i)))
1363                                 goto done;
1364                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1365                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1366                                 new->br_startblock, new->br_blockcount,
1367                                 newext)))
1368                                 goto done;
1369                 }
1370                 break;
1371
1372         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1373                 /*
1374                  * Setting the first part of a previous oldext extent to newext.
1375                  * The left neighbor is contiguous.
1376                  */
1377                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
1378                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
1379                         LEFT.br_blockcount + new->br_blockcount);
1380                 xfs_bmbt_set_startoff(ep,
1381                         PREV.br_startoff + new->br_blockcount);
1382                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
1383
1384                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1385                 xfs_bmbt_set_startblock(ep,
1386                         new->br_startblock + new->br_blockcount);
1387                 xfs_bmbt_set_blockcount(ep,
1388                         PREV.br_blockcount - new->br_blockcount);
1389                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1390
1391                 --*idx;
1392
1393                 if (cur == NULL)
1394                         rval = XFS_ILOG_DEXT;
1395                 else {
1396                         rval = 0;
1397                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1398                                         PREV.br_startblock, PREV.br_blockcount,
1399                                         &i)))
1400                                 goto done;
1401                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1402                         if ((error = xfs_bmbt_update(cur,
1403                                 PREV.br_startoff + new->br_blockcount,
1404                                 PREV.br_startblock + new->br_blockcount,
1405                                 PREV.br_blockcount - new->br_blockcount,
1406                                 oldext)))
1407                                 goto done;
1408                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1409                                 goto done;
1410                         if (xfs_bmbt_update(cur, LEFT.br_startoff,
1411                                 LEFT.br_startblock,
1412                                 LEFT.br_blockcount + new->br_blockcount,
1413                                 LEFT.br_state))
1414                                 goto done;
1415                 }
1416                 break;
1417
1418         case BMAP_LEFT_FILLING:
1419                 /*
1420                  * Setting the first part of a previous oldext extent to newext.
1421                  * The left neighbor is not contiguous.
1422                  */
1423                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1424                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1425                 xfs_bmbt_set_startoff(ep, new_endoff);
1426                 xfs_bmbt_set_blockcount(ep,
1427                         PREV.br_blockcount - new->br_blockcount);
1428                 xfs_bmbt_set_startblock(ep,
1429                         new->br_startblock + new->br_blockcount);
1430                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1431
1432                 xfs_iext_insert(ip, *idx, 1, new, state);
1433                 ip->i_d.di_nextents++;
1434                 if (cur == NULL)
1435                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1436                 else {
1437                         rval = XFS_ILOG_CORE;
1438                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1439                                         PREV.br_startblock, PREV.br_blockcount,
1440                                         &i)))
1441                                 goto done;
1442                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1443                         if ((error = xfs_bmbt_update(cur,
1444                                 PREV.br_startoff + new->br_blockcount,
1445                                 PREV.br_startblock + new->br_blockcount,
1446                                 PREV.br_blockcount - new->br_blockcount,
1447                                 oldext)))
1448                                 goto done;
1449                         cur->bc_rec.b = *new;
1450                         if ((error = xfs_btree_insert(cur, &i)))
1451                                 goto done;
1452                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1453                 }
1454                 break;
1455
1456         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1457                 /*
1458                  * Setting the last part of a previous oldext extent to newext.
1459                  * The right neighbor is contiguous with the new allocation.
1460                  */
1461                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1462                 xfs_bmbt_set_blockcount(ep,
1463                         PREV.br_blockcount - new->br_blockcount);
1464                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1465
1466                 ++*idx;
1467
1468                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1469                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1470                         new->br_startoff, new->br_startblock,
1471                         new->br_blockcount + RIGHT.br_blockcount, newext);
1472                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1473
1474                 if (cur == NULL)
1475                         rval = XFS_ILOG_DEXT;
1476                 else {
1477                         rval = 0;
1478                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1479                                         PREV.br_startblock,
1480                                         PREV.br_blockcount, &i)))
1481                                 goto done;
1482                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1483                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1484                                 PREV.br_startblock,
1485                                 PREV.br_blockcount - new->br_blockcount,
1486                                 oldext)))
1487                                 goto done;
1488                         if ((error = xfs_btree_increment(cur, 0, &i)))
1489                                 goto done;
1490                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1491                                 new->br_startblock,
1492                                 new->br_blockcount + RIGHT.br_blockcount,
1493                                 newext)))
1494                                 goto done;
1495                 }
1496                 break;
1497
1498         case BMAP_RIGHT_FILLING:
1499                 /*
1500                  * Setting the last part of a previous oldext extent to newext.
1501                  * The right neighbor is not contiguous.
1502                  */
1503                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1504                 xfs_bmbt_set_blockcount(ep,
1505                         PREV.br_blockcount - new->br_blockcount);
1506                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1507
1508                 ++*idx;
1509                 xfs_iext_insert(ip, *idx, 1, new, state);
1510
1511                 ip->i_d.di_nextents++;
1512                 if (cur == NULL)
1513                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1514                 else {
1515                         rval = XFS_ILOG_CORE;
1516                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1517                                         PREV.br_startblock, PREV.br_blockcount,
1518                                         &i)))
1519                                 goto done;
1520                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1521                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1522                                 PREV.br_startblock,
1523                                 PREV.br_blockcount - new->br_blockcount,
1524                                 oldext)))
1525                                 goto done;
1526                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1527                                         new->br_startblock, new->br_blockcount,
1528                                         &i)))
1529                                 goto done;
1530                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1531                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1532                         if ((error = xfs_btree_insert(cur, &i)))
1533                                 goto done;
1534                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1535                 }
1536                 break;
1537
1538         case 0:
1539                 /*
1540                  * Setting the middle part of a previous oldext extent to
1541                  * newext.  Contiguity is impossible here.
1542                  * One extent becomes three extents.
1543                  */
1544                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1545                 xfs_bmbt_set_blockcount(ep,
1546                         new->br_startoff - PREV.br_startoff);
1547                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1548
1549                 r[0] = *new;
1550                 r[1].br_startoff = new_endoff;
1551                 r[1].br_blockcount =
1552                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
1553                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1554                 r[1].br_state = oldext;
1555
1556                 ++*idx;
1557                 xfs_iext_insert(ip, *idx, 2, &r[0], state);
1558
1559                 ip->i_d.di_nextents += 2;
1560                 if (cur == NULL)
1561                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1562                 else {
1563                         rval = XFS_ILOG_CORE;
1564                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1565                                         PREV.br_startblock, PREV.br_blockcount,
1566                                         &i)))
1567                                 goto done;
1568                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1569                         /* new right extent - oldext */
1570                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1571                                 r[1].br_startblock, r[1].br_blockcount,
1572                                 r[1].br_state)))
1573                                 goto done;
1574                         /* new left extent - oldext */
1575                         cur->bc_rec.b = PREV;
1576                         cur->bc_rec.b.br_blockcount =
1577                                 new->br_startoff - PREV.br_startoff;
1578                         if ((error = xfs_btree_insert(cur, &i)))
1579                                 goto done;
1580                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1581                         /*
1582                          * Reset the cursor to the position of the new extent
1583                          * we are about to insert as we can't trust it after
1584                          * the previous insert.
1585                          */
1586                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1587                                         new->br_startblock, new->br_blockcount,
1588                                         &i)))
1589                                 goto done;
1590                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1591                         /* new middle extent - newext */
1592                         cur->bc_rec.b.br_state = new->br_state;
1593                         if ((error = xfs_btree_insert(cur, &i)))
1594                                 goto done;
1595                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1596                 }
1597                 break;
1598
1599         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1600         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1601         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1602         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1603         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1604         case BMAP_LEFT_CONTIG:
1605         case BMAP_RIGHT_CONTIG:
1606                 /*
1607                  * These cases are all impossible.
1608                  */
1609                 ASSERT(0);
1610         }
1611         *curp = cur;
1612 done:
1613         *logflagsp = rval;
1614         return error;
1615 #undef  LEFT
1616 #undef  RIGHT
1617 #undef  PREV
1618 }
1619
1620 /*
1621  * Called by xfs_bmap_add_extent to handle cases converting a hole
1622  * to a delayed allocation.
1623  */
1624 /*ARGSUSED*/
1625 STATIC int                              /* error */
1626 xfs_bmap_add_extent_hole_delay(
1627         xfs_inode_t             *ip,    /* incore inode pointer */
1628         xfs_extnum_t            *idx,   /* extent number to update/insert */
1629         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1630         int                     *logflagsp) /* inode logging flags */
1631 {
1632         xfs_bmbt_rec_host_t     *ep;    /* extent record for idx */
1633         xfs_ifork_t             *ifp;   /* inode fork pointer */
1634         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1635         xfs_filblks_t           newlen=0;       /* new indirect size */
1636         xfs_filblks_t           oldlen=0;       /* old indirect size */
1637         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1638         int                     state;  /* state bits, accessed thru macros */
1639         xfs_filblks_t           temp=0; /* temp for indirect calculations */
1640
1641         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1642         ep = xfs_iext_get_ext(ifp, *idx);
1643         state = 0;
1644         ASSERT(isnullstartblock(new->br_startblock));
1645
1646         /*
1647          * Check and set flags if this segment has a left neighbor
1648          */
1649         if (*idx > 0) {
1650                 state |= BMAP_LEFT_VALID;
1651                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
1652
1653                 if (isnullstartblock(left.br_startblock))
1654                         state |= BMAP_LEFT_DELAY;
1655         }
1656
1657         /*
1658          * Check and set flags if the current (right) segment exists.
1659          * If it doesn't exist, we're converting the hole at end-of-file.
1660          */
1661         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1662                 state |= BMAP_RIGHT_VALID;
1663                 xfs_bmbt_get_all(ep, &right);
1664
1665                 if (isnullstartblock(right.br_startblock))
1666                         state |= BMAP_RIGHT_DELAY;
1667         }
1668
1669         /*
1670          * Set contiguity flags on the left and right neighbors.
1671          * Don't let extents get too large, even if the pieces are contiguous.
1672          */
1673         if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
1674             left.br_startoff + left.br_blockcount == new->br_startoff &&
1675             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1676                 state |= BMAP_LEFT_CONTIG;
1677
1678         if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
1679             new->br_startoff + new->br_blockcount == right.br_startoff &&
1680             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1681             (!(state & BMAP_LEFT_CONTIG) ||
1682              (left.br_blockcount + new->br_blockcount +
1683               right.br_blockcount <= MAXEXTLEN)))
1684                 state |= BMAP_RIGHT_CONTIG;
1685
1686         /*
1687          * Switch out based on the contiguity flags.
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 delayed allocations
1693                  * on the left and on the right.
1694                  * Merge all three into a single extent record.
1695                  */
1696                 --*idx;
1697                 temp = left.br_blockcount + new->br_blockcount +
1698                         right.br_blockcount;
1699
1700                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1701                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1702                 oldlen = startblockval(left.br_startblock) +
1703                         startblockval(new->br_startblock) +
1704                         startblockval(right.br_startblock);
1705                 newlen = xfs_bmap_worst_indlen(ip, temp);
1706                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1707                         nullstartblock((int)newlen));
1708                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1709
1710                 xfs_iext_remove(ip, *idx + 1, 1, state);
1711                 break;
1712
1713         case BMAP_LEFT_CONTIG:
1714                 /*
1715                  * New allocation is contiguous with a delayed allocation
1716                  * on the left.
1717                  * Merge the new allocation with the left neighbor.
1718                  */
1719                 --*idx;
1720                 temp = left.br_blockcount + new->br_blockcount;
1721
1722                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1723                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1724                 oldlen = startblockval(left.br_startblock) +
1725                         startblockval(new->br_startblock);
1726                 newlen = xfs_bmap_worst_indlen(ip, temp);
1727                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1728                         nullstartblock((int)newlen));
1729                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1730                 break;
1731
1732         case BMAP_RIGHT_CONTIG:
1733                 /*
1734                  * New allocation is contiguous with a delayed allocation
1735                  * on the right.
1736                  * Merge the new allocation with the right neighbor.
1737                  */
1738                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1739                 temp = new->br_blockcount + right.br_blockcount;
1740                 oldlen = startblockval(new->br_startblock) +
1741                         startblockval(right.br_startblock);
1742                 newlen = xfs_bmap_worst_indlen(ip, temp);
1743                 xfs_bmbt_set_allf(ep, new->br_startoff,
1744                         nullstartblock((int)newlen), temp, right.br_state);
1745                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1746                 break;
1747
1748         case 0:
1749                 /*
1750                  * New allocation is not contiguous with another
1751                  * delayed allocation.
1752                  * Insert a new entry.
1753                  */
1754                 oldlen = newlen = 0;
1755                 xfs_iext_insert(ip, *idx, 1, new, state);
1756                 break;
1757         }
1758         if (oldlen != newlen) {
1759                 ASSERT(oldlen > newlen);
1760                 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
1761                         (int64_t)(oldlen - newlen), 0);
1762                 /*
1763                  * Nothing to do for disk quota accounting here.
1764                  */
1765         }
1766         *logflagsp = 0;
1767         return 0;
1768 }
1769
1770 /*
1771  * Called by xfs_bmap_add_extent to handle cases converting a hole
1772  * to a real allocation.
1773  */
1774 STATIC int                              /* error */
1775 xfs_bmap_add_extent_hole_real(
1776         xfs_inode_t             *ip,    /* incore inode pointer */
1777         xfs_extnum_t            *idx,   /* extent number to update/insert */
1778         xfs_btree_cur_t         *cur,   /* if null, not a btree */
1779         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1780         int                     *logflagsp, /* inode logging flags */
1781         int                     whichfork) /* data or attr fork */
1782 {
1783         xfs_bmbt_rec_host_t     *ep;    /* pointer to extent entry ins. point */
1784         int                     error;  /* error return value */
1785         int                     i;      /* temp state */
1786         xfs_ifork_t             *ifp;   /* inode fork pointer */
1787         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1788         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1789         int                     rval=0; /* return value (logging flags) */
1790         int                     state;  /* state bits, accessed thru macros */
1791
1792         ifp = XFS_IFORK_PTR(ip, whichfork);
1793         ASSERT(*idx <= ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
1794         ep = xfs_iext_get_ext(ifp, *idx);
1795         state = 0;
1796
1797         if (whichfork == XFS_ATTR_FORK)
1798                 state |= BMAP_ATTRFORK;
1799
1800         /*
1801          * Check and set flags if this segment has a left neighbor.
1802          */
1803         if (*idx > 0) {
1804                 state |= BMAP_LEFT_VALID;
1805                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
1806                 if (isnullstartblock(left.br_startblock))
1807                         state |= BMAP_LEFT_DELAY;
1808         }
1809
1810         /*
1811          * Check and set flags if this segment has a current value.
1812          * Not true if we're inserting into the "hole" at eof.
1813          */
1814         if (*idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1815                 state |= BMAP_RIGHT_VALID;
1816                 xfs_bmbt_get_all(ep, &right);
1817                 if (isnullstartblock(right.br_startblock))
1818                         state |= BMAP_RIGHT_DELAY;
1819         }
1820
1821         /*
1822          * We're inserting a real allocation between "left" and "right".
1823          * Set the contiguity flags.  Don't let extents get too large.
1824          */
1825         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1826             left.br_startoff + left.br_blockcount == new->br_startoff &&
1827             left.br_startblock + left.br_blockcount == new->br_startblock &&
1828             left.br_state == new->br_state &&
1829             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1830                 state |= BMAP_LEFT_CONTIG;
1831
1832         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1833             new->br_startoff + new->br_blockcount == right.br_startoff &&
1834             new->br_startblock + new->br_blockcount == right.br_startblock &&
1835             new->br_state == right.br_state &&
1836             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1837             (!(state & BMAP_LEFT_CONTIG) ||
1838              left.br_blockcount + new->br_blockcount +
1839              right.br_blockcount <= MAXEXTLEN))
1840                 state |= BMAP_RIGHT_CONTIG;
1841
1842         error = 0;
1843         /*
1844          * Select which case we're in here, and implement it.
1845          */
1846         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1847         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1848                 /*
1849                  * New allocation is contiguous with real allocations on the
1850                  * left and on the right.
1851                  * Merge all three into a single extent record.
1852                  */
1853                 --*idx;
1854                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1855                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1856                         left.br_blockcount + new->br_blockcount +
1857                         right.br_blockcount);
1858                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1859
1860                 xfs_iext_remove(ip, *idx + 1, 1, state);
1861
1862                 XFS_IFORK_NEXT_SET(ip, whichfork,
1863                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
1864                 if (cur == NULL) {
1865                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1866                 } else {
1867                         rval = XFS_ILOG_CORE;
1868                         if ((error = xfs_bmbt_lookup_eq(cur,
1869                                         right.br_startoff,
1870                                         right.br_startblock,
1871                                         right.br_blockcount, &i)))
1872                                 goto done;
1873                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1874                         if ((error = xfs_btree_delete(cur, &i)))
1875                                 goto done;
1876                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1877                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1878                                 goto done;
1879                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1880                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
1881                                         left.br_startblock,
1882                                         left.br_blockcount +
1883                                                 new->br_blockcount +
1884                                                 right.br_blockcount,
1885                                         left.br_state)))
1886                                 goto done;
1887                 }
1888                 break;
1889
1890         case BMAP_LEFT_CONTIG:
1891                 /*
1892                  * New allocation is contiguous with a real allocation
1893                  * on the left.
1894                  * Merge the new allocation with the left neighbor.
1895                  */
1896                 --*idx;
1897                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1898                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1899                         left.br_blockcount + new->br_blockcount);
1900                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1901
1902                 if (cur == NULL) {
1903                         rval = xfs_ilog_fext(whichfork);
1904                 } else {
1905                         rval = 0;
1906                         if ((error = xfs_bmbt_lookup_eq(cur,
1907                                         left.br_startoff,
1908                                         left.br_startblock,
1909                                         left.br_blockcount, &i)))
1910                                 goto done;
1911                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1912                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
1913                                         left.br_startblock,
1914                                         left.br_blockcount +
1915                                                 new->br_blockcount,
1916                                         left.br_state)))
1917                                 goto done;
1918                 }
1919                 break;
1920
1921         case BMAP_RIGHT_CONTIG:
1922                 /*
1923                  * New allocation is contiguous with a real allocation
1924                  * on the right.
1925                  * Merge the new allocation with the right neighbor.
1926                  */
1927                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1928                 xfs_bmbt_set_allf(ep, new->br_startoff, new->br_startblock,
1929                         new->br_blockcount + right.br_blockcount,
1930                         right.br_state);
1931                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1932
1933                 if (cur == NULL) {
1934                         rval = xfs_ilog_fext(whichfork);
1935                 } else {
1936                         rval = 0;
1937                         if ((error = xfs_bmbt_lookup_eq(cur,
1938                                         right.br_startoff,
1939                                         right.br_startblock,
1940                                         right.br_blockcount, &i)))
1941                                 goto done;
1942                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1943                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1944                                         new->br_startblock,
1945                                         new->br_blockcount +
1946                                                 right.br_blockcount,
1947                                         right.br_state)))
1948                                 goto done;
1949                 }
1950                 break;
1951
1952         case 0:
1953                 /*
1954                  * New allocation is not contiguous with another
1955                  * real allocation.
1956                  * Insert a new entry.
1957                  */
1958                 xfs_iext_insert(ip, *idx, 1, new, state);
1959                 XFS_IFORK_NEXT_SET(ip, whichfork,
1960                         XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
1961                 if (cur == NULL) {
1962                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1963                 } else {
1964                         rval = XFS_ILOG_CORE;
1965                         if ((error = xfs_bmbt_lookup_eq(cur,
1966                                         new->br_startoff,
1967                                         new->br_startblock,
1968                                         new->br_blockcount, &i)))
1969                                 goto done;
1970                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1971                         cur->bc_rec.b.br_state = new->br_state;
1972                         if ((error = xfs_btree_insert(cur, &i)))
1973                                 goto done;
1974                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1975                 }
1976                 break;
1977         }
1978 done:
1979         *logflagsp = rval;
1980         return error;
1981 }
1982
1983 /*
1984  * Adjust the size of the new extent based on di_extsize and rt extsize.
1985  */
1986 STATIC int
1987 xfs_bmap_extsize_align(
1988         xfs_mount_t     *mp,
1989         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
1990         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
1991         xfs_extlen_t    extsz,          /* align to this extent size */
1992         int             rt,             /* is this a realtime inode? */
1993         int             eof,            /* is extent at end-of-file? */
1994         int             delay,          /* creating delalloc extent? */
1995         int             convert,        /* overwriting unwritten extent? */
1996         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
1997         xfs_extlen_t    *lenp)          /* in/out: aligned length */
1998 {
1999         xfs_fileoff_t   orig_off;       /* original offset */
2000         xfs_extlen_t    orig_alen;      /* original length */
2001         xfs_fileoff_t   orig_end;       /* original off+len */
2002         xfs_fileoff_t   nexto;          /* next file offset */
2003         xfs_fileoff_t   prevo;          /* previous file offset */
2004         xfs_fileoff_t   align_off;      /* temp for offset */
2005         xfs_extlen_t    align_alen;     /* temp for length */
2006         xfs_extlen_t    temp;           /* temp for calculations */
2007
2008         if (convert)
2009                 return 0;
2010
2011         orig_off = align_off = *offp;
2012         orig_alen = align_alen = *lenp;
2013         orig_end = orig_off + orig_alen;
2014
2015         /*
2016          * If this request overlaps an existing extent, then don't
2017          * attempt to perform any additional alignment.
2018          */
2019         if (!delay && !eof &&
2020             (orig_off >= gotp->br_startoff) &&
2021             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2022                 return 0;
2023         }
2024
2025         /*
2026          * If the file offset is unaligned vs. the extent size
2027          * we need to align it.  This will be possible unless
2028          * the file was previously written with a kernel that didn't
2029          * perform this alignment, or if a truncate shot us in the
2030          * foot.
2031          */
2032         temp = do_mod(orig_off, extsz);
2033         if (temp) {
2034                 align_alen += temp;
2035                 align_off -= temp;
2036         }
2037         /*
2038          * Same adjustment for the end of the requested area.
2039          */
2040         if ((temp = (align_alen % extsz))) {
2041                 align_alen += extsz - temp;
2042         }
2043         /*
2044          * If the previous block overlaps with this proposed allocation
2045          * then move the start forward without adjusting the length.
2046          */
2047         if (prevp->br_startoff != NULLFILEOFF) {
2048                 if (prevp->br_startblock == HOLESTARTBLOCK)
2049                         prevo = prevp->br_startoff;
2050                 else
2051                         prevo = prevp->br_startoff + prevp->br_blockcount;
2052         } else
2053                 prevo = 0;
2054         if (align_off != orig_off && align_off < prevo)
2055                 align_off = prevo;
2056         /*
2057          * If the next block overlaps with this proposed allocation
2058          * then move the start back without adjusting the length,
2059          * but not before offset 0.
2060          * This may of course make the start overlap previous block,
2061          * and if we hit the offset 0 limit then the next block
2062          * can still overlap too.
2063          */
2064         if (!eof && gotp->br_startoff != NULLFILEOFF) {
2065                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
2066                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
2067                         nexto = gotp->br_startoff + gotp->br_blockcount;
2068                 else
2069                         nexto = gotp->br_startoff;
2070         } else
2071                 nexto = NULLFILEOFF;
2072         if (!eof &&
2073             align_off + align_alen != orig_end &&
2074             align_off + align_alen > nexto)
2075                 align_off = nexto > align_alen ? nexto - align_alen : 0;
2076         /*
2077          * If we're now overlapping the next or previous extent that
2078          * means we can't fit an extsz piece in this hole.  Just move
2079          * the start forward to the first valid spot and set
2080          * the length so we hit the end.
2081          */
2082         if (align_off != orig_off && align_off < prevo)
2083                 align_off = prevo;
2084         if (align_off + align_alen != orig_end &&
2085             align_off + align_alen > nexto &&
2086             nexto != NULLFILEOFF) {
2087                 ASSERT(nexto > prevo);
2088                 align_alen = nexto - align_off;
2089         }
2090
2091         /*
2092          * If realtime, and the result isn't a multiple of the realtime
2093          * extent size we need to remove blocks until it is.
2094          */
2095         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
2096                 /*
2097                  * We're not covering the original request, or
2098                  * we won't be able to once we fix the length.
2099                  */
2100                 if (orig_off < align_off ||
2101                     orig_end > align_off + align_alen ||
2102                     align_alen - temp < orig_alen)
2103                         return XFS_ERROR(EINVAL);
2104                 /*
2105                  * Try to fix it by moving the start up.
2106                  */
2107                 if (align_off + temp <= orig_off) {
2108                         align_alen -= temp;
2109                         align_off += temp;
2110                 }
2111                 /*
2112                  * Try to fix it by moving the end in.
2113                  */
2114                 else if (align_off + align_alen - temp >= orig_end)
2115                         align_alen -= temp;
2116                 /*
2117                  * Set the start to the minimum then trim the length.
2118                  */
2119                 else {
2120                         align_alen -= orig_off - align_off;
2121                         align_off = orig_off;
2122                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
2123                 }
2124                 /*
2125                  * Result doesn't cover the request, fail it.
2126                  */
2127                 if (orig_off < align_off || orig_end > align_off + align_alen)
2128                         return XFS_ERROR(EINVAL);
2129         } else {
2130                 ASSERT(orig_off >= align_off);
2131                 ASSERT(orig_end <= align_off + align_alen);
2132         }
2133
2134 #ifdef DEBUG
2135         if (!eof && gotp->br_startoff != NULLFILEOFF)
2136                 ASSERT(align_off + align_alen <= gotp->br_startoff);
2137         if (prevp->br_startoff != NULLFILEOFF)
2138                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
2139 #endif
2140
2141         *lenp = align_alen;
2142         *offp = align_off;
2143         return 0;
2144 }
2145
2146 #define XFS_ALLOC_GAP_UNITS     4
2147
2148 STATIC void
2149 xfs_bmap_adjacent(
2150         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2151 {
2152         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
2153         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2154         xfs_mount_t     *mp;            /* mount point structure */
2155         int             nullfb;         /* true if ap->firstblock isn't set */
2156         int             rt;             /* true if inode is realtime */
2157
2158 #define ISVALID(x,y)    \
2159         (rt ? \
2160                 (x) < mp->m_sb.sb_rblocks : \
2161                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2162                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2163                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2164
2165         mp = ap->ip->i_mount;
2166         nullfb = ap->firstblock == NULLFSBLOCK;
2167         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2168         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2169         /*
2170          * If allocating at eof, and there's a previous real block,
2171          * try to use its last block as our starting point.
2172          */
2173         if (ap->eof && ap->prevp->br_startoff != NULLFILEOFF &&
2174             !isnullstartblock(ap->prevp->br_startblock) &&
2175             ISVALID(ap->prevp->br_startblock + ap->prevp->br_blockcount,
2176                     ap->prevp->br_startblock)) {
2177                 ap->rval = ap->prevp->br_startblock + ap->prevp->br_blockcount;
2178                 /*
2179                  * Adjust for the gap between prevp and us.
2180                  */
2181                 adjust = ap->off -
2182                         (ap->prevp->br_startoff + ap->prevp->br_blockcount);
2183                 if (adjust &&
2184                     ISVALID(ap->rval + adjust, ap->prevp->br_startblock))
2185                         ap->rval += adjust;
2186         }
2187         /*
2188          * If not at eof, then compare the two neighbor blocks.
2189          * Figure out whether either one gives us a good starting point,
2190          * and pick the better one.
2191          */
2192         else if (!ap->eof) {
2193                 xfs_fsblock_t   gotbno;         /* right side block number */
2194                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
2195                 xfs_fsblock_t   prevbno;        /* left side block number */
2196                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
2197
2198                 /*
2199                  * If there's a previous (left) block, select a requested
2200                  * start block based on it.
2201                  */
2202                 if (ap->prevp->br_startoff != NULLFILEOFF &&
2203                     !isnullstartblock(ap->prevp->br_startblock) &&
2204                     (prevbno = ap->prevp->br_startblock +
2205                                ap->prevp->br_blockcount) &&
2206                     ISVALID(prevbno, ap->prevp->br_startblock)) {
2207                         /*
2208                          * Calculate gap to end of previous block.
2209                          */
2210                         adjust = prevdiff = ap->off -
2211                                 (ap->prevp->br_startoff +
2212                                  ap->prevp->br_blockcount);
2213                         /*
2214                          * Figure the startblock based on the previous block's
2215                          * end and the gap size.
2216                          * Heuristic!
2217                          * If the gap is large relative to the piece we're
2218                          * allocating, or using it gives us an invalid block
2219                          * number, then just use the end of the previous block.
2220                          */
2221                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2222                             ISVALID(prevbno + prevdiff,
2223                                     ap->prevp->br_startblock))
2224                                 prevbno += adjust;
2225                         else
2226                                 prevdiff += adjust;
2227                         /*
2228                          * If the firstblock forbids it, can't use it,
2229                          * must use default.
2230                          */
2231                         if (!rt && !nullfb &&
2232                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2233                                 prevbno = NULLFSBLOCK;
2234                 }
2235                 /*
2236                  * No previous block or can't follow it, just default.
2237                  */
2238                 else
2239                         prevbno = NULLFSBLOCK;
2240                 /*
2241                  * If there's a following (right) block, select a requested
2242                  * start block based on it.
2243                  */
2244                 if (!isnullstartblock(ap->gotp->br_startblock)) {
2245                         /*
2246                          * Calculate gap to start of next block.
2247                          */
2248                         adjust = gotdiff = ap->gotp->br_startoff - ap->off;
2249                         /*
2250                          * Figure the startblock based on the next block's
2251                          * start and the gap size.
2252                          */
2253                         gotbno = ap->gotp->br_startblock;
2254                         /*
2255                          * Heuristic!
2256                          * If the gap is large relative to the piece we're
2257                          * allocating, or using it gives us an invalid block
2258                          * number, then just use the start of the next block
2259                          * offset by our length.
2260                          */
2261                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2262                             ISVALID(gotbno - gotdiff, gotbno))
2263                                 gotbno -= adjust;
2264                         else if (ISVALID(gotbno - ap->alen, gotbno)) {
2265                                 gotbno -= ap->alen;
2266                                 gotdiff += adjust - ap->alen;
2267                         } else
2268                                 gotdiff += adjust;
2269                         /*
2270                          * If the firstblock forbids it, can't use it,
2271                          * must use default.
2272                          */
2273                         if (!rt && !nullfb &&
2274                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2275                                 gotbno = NULLFSBLOCK;
2276                 }
2277                 /*
2278                  * No next block, just default.
2279                  */
2280                 else
2281                         gotbno = NULLFSBLOCK;
2282                 /*
2283                  * If both valid, pick the better one, else the only good
2284                  * one, else ap->rval is already set (to 0 or the inode block).
2285                  */
2286                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2287                         ap->rval = prevdiff <= gotdiff ? prevbno : gotbno;
2288                 else if (prevbno != NULLFSBLOCK)
2289                         ap->rval = prevbno;
2290                 else if (gotbno != NULLFSBLOCK)
2291                         ap->rval = gotbno;
2292         }
2293 #undef ISVALID
2294 }
2295
2296 STATIC int
2297 xfs_bmap_rtalloc(
2298         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2299 {
2300         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2301         int             error;          /* error return value */
2302         xfs_mount_t     *mp;            /* mount point structure */
2303         xfs_extlen_t    prod = 0;       /* product factor for allocators */
2304         xfs_extlen_t    ralen = 0;      /* realtime allocation length */
2305         xfs_extlen_t    align;          /* minimum allocation alignment */
2306         xfs_rtblock_t   rtb;
2307
2308         mp = ap->ip->i_mount;
2309         align = xfs_get_extsz_hint(ap->ip);
2310         prod = align / mp->m_sb.sb_rextsize;
2311         error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2312                                         align, 1, ap->eof, 0,
2313                                         ap->conv, &ap->off, &ap->alen);
2314         if (error)
2315                 return error;
2316         ASSERT(ap->alen);
2317         ASSERT(ap->alen % mp->m_sb.sb_rextsize == 0);
2318
2319         /*
2320          * If the offset & length are not perfectly aligned
2321          * then kill prod, it will just get us in trouble.
2322          */
2323         if (do_mod(ap->off, align) || ap->alen % align)
2324                 prod = 1;
2325         /*
2326          * Set ralen to be the actual requested length in rtextents.
2327          */
2328         ralen = ap->alen / mp->m_sb.sb_rextsize;
2329         /*
2330          * If the old value was close enough to MAXEXTLEN that
2331          * we rounded up to it, cut it back so it's valid again.
2332          * Note that if it's a really large request (bigger than
2333          * MAXEXTLEN), we don't hear about that number, and can't
2334          * adjust the starting point to match it.
2335          */
2336         if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2337                 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2338
2339         /*
2340          * Lock out other modifications to the RT bitmap inode.
2341          */
2342         xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
2343         xfs_trans_ijoin_ref(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
2344
2345         /*
2346          * If it's an allocation to an empty file at offset 0,
2347          * pick an extent that will space things out in the rt area.
2348          */
2349         if (ap->eof && ap->off == 0) {
2350                 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
2351
2352                 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2353                 if (error)
2354                         return error;
2355                 ap->rval = rtx * mp->m_sb.sb_rextsize;
2356         } else {
2357                 ap->rval = 0;
2358         }
2359
2360         xfs_bmap_adjacent(ap);
2361
2362         /*
2363          * Realtime allocation, done through xfs_rtallocate_extent.
2364          */
2365         atype = ap->rval == 0 ?  XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2366         do_div(ap->rval, mp->m_sb.sb_rextsize);
2367         rtb = ap->rval;
2368         ap->alen = ralen;
2369         if ((error = xfs_rtallocate_extent(ap->tp, ap->rval, 1, ap->alen,
2370                                 &ralen, atype, ap->wasdel, prod, &rtb)))
2371                 return error;
2372         if (rtb == NULLFSBLOCK && prod > 1 &&
2373             (error = xfs_rtallocate_extent(ap->tp, ap->rval, 1,
2374                                            ap->alen, &ralen, atype,
2375                                            ap->wasdel, 1, &rtb)))
2376                 return error;
2377         ap->rval = rtb;
2378         if (ap->rval != NULLFSBLOCK) {
2379                 ap->rval *= mp->m_sb.sb_rextsize;
2380                 ralen *= mp->m_sb.sb_rextsize;
2381                 ap->alen = ralen;
2382                 ap->ip->i_d.di_nblocks += ralen;
2383                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2384                 if (ap->wasdel)
2385                         ap->ip->i_delayed_blks -= ralen;
2386                 /*
2387                  * Adjust the disk quota also. This was reserved
2388                  * earlier.
2389                  */
2390                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2391                         ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2392                                         XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2393         } else {
2394                 ap->alen = 0;
2395         }
2396         return 0;
2397 }
2398
2399 STATIC int
2400 xfs_bmap_btalloc_nullfb(
2401         struct xfs_bmalloca     *ap,
2402         struct xfs_alloc_arg    *args,
2403         xfs_extlen_t            *blen)
2404 {
2405         struct xfs_mount        *mp = ap->ip->i_mount;
2406         struct xfs_perag        *pag;
2407         xfs_agnumber_t          ag, startag;
2408         int                     notinit = 0;
2409         int                     error;
2410
2411         if (ap->userdata && xfs_inode_is_filestream(ap->ip))
2412                 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2413         else
2414                 args->type = XFS_ALLOCTYPE_START_BNO;
2415         args->total = ap->total;
2416
2417         /*
2418          * Search for an allocation group with a single extent large enough
2419          * for the request.  If one isn't found, then adjust the minimum
2420          * allocation size to the largest space found.
2421          */
2422         startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
2423         if (startag == NULLAGNUMBER)
2424                 startag = ag = 0;
2425
2426         pag = xfs_perag_get(mp, ag);
2427         while (*blen < args->maxlen) {
2428                 if (!pag->pagf_init) {
2429                         error = xfs_alloc_pagf_init(mp, args->tp, ag,
2430                                                     XFS_ALLOC_FLAG_TRYLOCK);
2431                         if (error) {
2432                                 xfs_perag_put(pag);
2433                                 return error;
2434                         }
2435                 }
2436
2437                 /*
2438                  * See xfs_alloc_fix_freelist...
2439                  */
2440                 if (pag->pagf_init) {
2441                         xfs_extlen_t    longest;
2442                         longest = xfs_alloc_longest_free_extent(mp, pag);
2443                         if (*blen < longest)
2444                                 *blen = longest;
2445                 } else
2446                         notinit = 1;
2447
2448                 if (xfs_inode_is_filestream(ap->ip)) {
2449                         if (*blen >= args->maxlen)
2450                                 break;
2451
2452                         if (ap->userdata) {
2453                                 /*
2454                                  * If startag is an invalid AG, we've
2455                                  * come here once before and
2456                                  * xfs_filestream_new_ag picked the
2457                                  * best currently available.
2458                                  *
2459                                  * Don't continue looping, since we
2460                                  * could loop forever.
2461                                  */
2462                                 if (startag == NULLAGNUMBER)
2463                                         break;
2464
2465                                 error = xfs_filestream_new_ag(ap, &ag);
2466                                 xfs_perag_put(pag);
2467                                 if (error)
2468                                         return error;
2469
2470                                 /* loop again to set 'blen'*/
2471                                 startag = NULLAGNUMBER;
2472                                 pag = xfs_perag_get(mp, ag);
2473                                 continue;
2474                         }
2475                 }
2476                 if (++ag == mp->m_sb.sb_agcount)
2477                         ag = 0;
2478                 if (ag == startag)
2479                         break;
2480                 xfs_perag_put(pag);
2481                 pag = xfs_perag_get(mp, ag);
2482         }
2483         xfs_perag_put(pag);
2484
2485         /*
2486          * Since the above loop did a BUF_TRYLOCK, it is
2487          * possible that there is space for this request.
2488          */
2489         if (notinit || *blen < ap->minlen)
2490                 args->minlen = ap->minlen;
2491         /*
2492          * If the best seen length is less than the request
2493          * length, use the best as the minimum.
2494          */
2495         else if (*blen < args->maxlen)
2496                 args->minlen = *blen;
2497         /*
2498          * Otherwise we've seen an extent as big as maxlen,
2499          * use that as the minimum.
2500          */
2501         else
2502                 args->minlen = args->maxlen;
2503
2504         /*
2505          * set the failure fallback case to look in the selected
2506          * AG as the stream may have moved.
2507          */
2508         if (xfs_inode_is_filestream(ap->ip))
2509                 ap->rval = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
2510
2511         return 0;
2512 }
2513
2514 STATIC int
2515 xfs_bmap_btalloc(
2516         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2517 {
2518         xfs_mount_t     *mp;            /* mount point structure */
2519         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2520         xfs_extlen_t    align;          /* minimum allocation alignment */
2521         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2522         xfs_agnumber_t  ag;
2523         xfs_alloc_arg_t args;
2524         xfs_extlen_t    blen;
2525         xfs_extlen_t    nextminlen = 0;
2526         int             nullfb;         /* true if ap->firstblock isn't set */
2527         int             isaligned;
2528         int             tryagain;
2529         int             error;
2530
2531         mp = ap->ip->i_mount;
2532         align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
2533         if (unlikely(align)) {
2534                 error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2535                                                 align, 0, ap->eof, 0, ap->conv,
2536                                                 &ap->off, &ap->alen);
2537                 ASSERT(!error);
2538                 ASSERT(ap->alen);
2539         }
2540         nullfb = ap->firstblock == NULLFSBLOCK;
2541         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2542         if (nullfb) {
2543                 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
2544                         ag = xfs_filestream_lookup_ag(ap->ip);
2545                         ag = (ag != NULLAGNUMBER) ? ag : 0;
2546                         ap->rval = XFS_AGB_TO_FSB(mp, ag, 0);
2547                 } else {
2548                         ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2549                 }
2550         } else
2551                 ap->rval = ap->firstblock;
2552
2553         xfs_bmap_adjacent(ap);
2554
2555         /*
2556          * If allowed, use ap->rval; otherwise must use firstblock since
2557          * it's in the right allocation group.
2558          */
2559         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->rval) == fb_agno)
2560                 ;
2561         else
2562                 ap->rval = ap->firstblock;
2563         /*
2564          * Normal allocation, done through xfs_alloc_vextent.
2565          */
2566         tryagain = isaligned = 0;
2567         args.tp = ap->tp;
2568         args.mp = mp;
2569         args.fsbno = ap->rval;
2570
2571         /* Trim the allocation back to the maximum an AG can fit. */
2572         args.maxlen = MIN(ap->alen, XFS_ALLOC_AG_MAX_USABLE(mp));
2573         args.firstblock = ap->firstblock;
2574         blen = 0;
2575         if (nullfb) {
2576                 error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
2577                 if (error)
2578                         return error;
2579         } else if (ap->low) {
2580                 if (xfs_inode_is_filestream(ap->ip))
2581                         args.type = XFS_ALLOCTYPE_FIRST_AG;
2582                 else
2583                         args.type = XFS_ALLOCTYPE_START_BNO;
2584                 args.total = args.minlen = ap->minlen;
2585         } else {
2586                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2587                 args.total = ap->total;
2588                 args.minlen = ap->minlen;
2589         }
2590         /* apply extent size hints if obtained earlier */
2591         if (unlikely(align)) {
2592                 args.prod = align;
2593                 if ((args.mod = (xfs_extlen_t)do_mod(ap->off, args.prod)))
2594                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2595         } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
2596                 args.prod = 1;
2597                 args.mod = 0;
2598         } else {
2599                 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
2600                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->off, args.prod))))
2601                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2602         }
2603         /*
2604          * If we are not low on available data blocks, and the
2605          * underlying logical volume manager is a stripe, and
2606          * the file offset is zero then try to allocate data
2607          * blocks on stripe unit boundary.
2608          * NOTE: ap->aeof is only set if the allocation length
2609          * is >= the stripe unit and the allocation offset is
2610          * at the end of file.
2611          */
2612         if (!ap->low && ap->aeof) {
2613                 if (!ap->off) {
2614                         args.alignment = mp->m_dalign;
2615                         atype = args.type;
2616                         isaligned = 1;
2617                         /*
2618                          * Adjust for alignment
2619                          */
2620                         if (blen > args.alignment && blen <= args.maxlen)
2621                                 args.minlen = blen - args.alignment;
2622                         args.minalignslop = 0;
2623                 } else {
2624                         /*
2625                          * First try an exact bno allocation.
2626                          * If it fails then do a near or start bno
2627                          * allocation with alignment turned on.
2628                          */
2629                         atype = args.type;
2630                         tryagain = 1;
2631                         args.type = XFS_ALLOCTYPE_THIS_BNO;
2632                         args.alignment = 1;
2633                         /*
2634                          * Compute the minlen+alignment for the
2635                          * next case.  Set slop so that the value
2636                          * of minlen+alignment+slop doesn't go up
2637                          * between the calls.
2638                          */
2639                         if (blen > mp->m_dalign && blen <= args.maxlen)
2640                                 nextminlen = blen - mp->m_dalign;
2641                         else
2642                                 nextminlen = args.minlen;
2643                         if (nextminlen + mp->m_dalign > args.minlen + 1)
2644                                 args.minalignslop =
2645                                         nextminlen + mp->m_dalign -
2646                                         args.minlen - 1;
2647                         else
2648                                 args.minalignslop = 0;
2649                 }
2650         } else {
2651                 args.alignment = 1;
2652                 args.minalignslop = 0;
2653         }
2654         args.minleft = ap->minleft;
2655         args.wasdel = ap->wasdel;
2656         args.isfl = 0;
2657         args.userdata = ap->userdata;
2658         if ((error = xfs_alloc_vextent(&args)))
2659                 return error;
2660         if (tryagain && args.fsbno == NULLFSBLOCK) {
2661                 /*
2662                  * Exact allocation failed. Now try with alignment
2663                  * turned on.
2664                  */
2665                 args.type = atype;
2666                 args.fsbno = ap->rval;
2667                 args.alignment = mp->m_dalign;
2668                 args.minlen = nextminlen;
2669                 args.minalignslop = 0;
2670                 isaligned = 1;
2671                 if ((error = xfs_alloc_vextent(&args)))
2672                         return error;
2673         }
2674         if (isaligned && args.fsbno == NULLFSBLOCK) {
2675                 /*
2676                  * allocation failed, so turn off alignment and
2677                  * try again.
2678                  */
2679                 args.type = atype;
2680                 args.fsbno = ap->rval;
2681                 args.alignment = 0;
2682                 if ((error = xfs_alloc_vextent(&args)))
2683                         return error;
2684         }
2685         if (args.fsbno == NULLFSBLOCK && nullfb &&
2686             args.minlen > ap->minlen) {
2687                 args.minlen = ap->minlen;
2688                 args.type = XFS_ALLOCTYPE_START_BNO;
2689                 args.fsbno = ap->rval;
2690                 if ((error = xfs_alloc_vextent(&args)))
2691                         return error;
2692         }
2693         if (args.fsbno == NULLFSBLOCK && nullfb) {
2694                 args.fsbno = 0;
2695                 args.type = XFS_ALLOCTYPE_FIRST_AG;
2696                 args.total = ap->minlen;
2697                 args.minleft = 0;
2698                 if ((error = xfs_alloc_vextent(&args)))
2699                         return error;
2700                 ap->low = 1;
2701         }
2702         if (args.fsbno != NULLFSBLOCK) {
2703                 ap->firstblock = ap->rval = args.fsbno;
2704                 ASSERT(nullfb || fb_agno == args.agno ||
2705                        (ap->low && fb_agno < args.agno));
2706                 ap->alen = args.len;
2707                 ap->ip->i_d.di_nblocks += args.len;
2708                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2709                 if (ap->wasdel)
2710                         ap->ip->i_delayed_blks -= args.len;
2711                 /*
2712                  * Adjust the disk quota also. This was reserved
2713                  * earlier.
2714                  */
2715                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2716                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2717                                         XFS_TRANS_DQ_BCOUNT,
2718                         (long) args.len);
2719         } else {
2720                 ap->rval = NULLFSBLOCK;
2721                 ap->alen = 0;
2722         }
2723         return 0;
2724 }
2725
2726 /*
2727  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2728  * It figures out where to ask the underlying allocator to put the new extent.
2729  */
2730 STATIC int
2731 xfs_bmap_alloc(
2732         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2733 {
2734         if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
2735                 return xfs_bmap_rtalloc(ap);
2736         return xfs_bmap_btalloc(ap);
2737 }
2738
2739 /*
2740  * Transform a btree format file with only one leaf node, where the
2741  * extents list will fit in the inode, into an extents format file.
2742  * Since the file extents are already in-core, all we have to do is
2743  * give up the space for the btree root and pitch the leaf block.
2744  */
2745 STATIC int                              /* error */
2746 xfs_bmap_btree_to_extents(
2747         xfs_trans_t             *tp,    /* transaction pointer */
2748         xfs_inode_t             *ip,    /* incore inode pointer */
2749         xfs_btree_cur_t         *cur,   /* btree cursor */
2750         int                     *logflagsp, /* inode logging flags */
2751         int                     whichfork)  /* data or attr fork */
2752 {
2753         /* REFERENCED */
2754         struct xfs_btree_block  *cblock;/* child btree block */
2755         xfs_fsblock_t           cbno;   /* child block number */
2756         xfs_buf_t               *cbp;   /* child block's buffer */
2757         int                     error;  /* error return value */
2758         xfs_ifork_t             *ifp;   /* inode fork data */
2759         xfs_mount_t             *mp;    /* mount point structure */
2760         __be64                  *pp;    /* ptr to block address */
2761         struct xfs_btree_block  *rblock;/* root btree block */
2762
2763         mp = ip->i_mount;
2764         ifp = XFS_IFORK_PTR(ip, whichfork);
2765         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2766         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
2767         rblock = ifp->if_broot;
2768         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
2769         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
2770         ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
2771         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
2772         cbno = be64_to_cpu(*pp);
2773         *logflagsp = 0;
2774 #ifdef DEBUG
2775         if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
2776                 return error;
2777 #endif
2778         if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
2779                         XFS_BMAP_BTREE_REF)))
2780                 return error;
2781         cblock = XFS_BUF_TO_BLOCK(cbp);
2782         if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
2783                 return error;
2784         xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
2785         ip->i_d.di_nblocks--;
2786         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
2787         xfs_trans_binval(tp, cbp);
2788         if (cur->bc_bufs[0] == cbp)
2789                 cur->bc_bufs[0] = NULL;
2790         xfs_iroot_realloc(ip, -1, whichfork);
2791         ASSERT(ifp->if_broot == NULL);
2792         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
2793         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
2794         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2795         return 0;
2796 }
2797
2798 /*
2799  * Called by xfs_bmapi to update file extent records and the btree
2800  * after removing space (or undoing a delayed allocation).
2801  */
2802 STATIC int                              /* error */
2803 xfs_bmap_del_extent(
2804         xfs_inode_t             *ip,    /* incore inode pointer */
2805         xfs_trans_t             *tp,    /* current transaction pointer */
2806         xfs_extnum_t            *idx,   /* extent number to update/delete */
2807         xfs_bmap_free_t         *flist, /* list of extents to be freed */
2808         xfs_btree_cur_t         *cur,   /* if null, not a btree */
2809         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
2810         int                     *logflagsp, /* inode logging flags */
2811         int                     whichfork) /* data or attr fork */
2812 {
2813         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
2814         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
2815         xfs_fsblock_t           del_endblock=0; /* first block past del */
2816         xfs_fileoff_t           del_endoff;     /* first offset past del */
2817         int                     delay;  /* current block is delayed allocated */
2818         int                     do_fx;  /* free extent at end of routine */
2819         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
2820         int                     error;  /* error return value */
2821         int                     flags;  /* inode logging flags */
2822         xfs_bmbt_irec_t         got;    /* current extent entry */
2823         xfs_fileoff_t           got_endoff;     /* first offset past got */
2824         int                     i;      /* temp state */
2825         xfs_ifork_t             *ifp;   /* inode fork pointer */
2826         xfs_mount_t             *mp;    /* mount structure */
2827         xfs_filblks_t           nblks;  /* quota/sb block count */
2828         xfs_bmbt_irec_t         new;    /* new record to be inserted */
2829         /* REFERENCED */
2830         uint                    qfield; /* quota field to update */
2831         xfs_filblks_t           temp;   /* for indirect length calculations */
2832         xfs_filblks_t           temp2;  /* for indirect length calculations */
2833         int                     state = 0;
2834
2835         XFS_STATS_INC(xs_del_exlist);
2836
2837         if (whichfork == XFS_ATTR_FORK)
2838                 state |= BMAP_ATTRFORK;
2839
2840         mp = ip->i_mount;
2841         ifp = XFS_IFORK_PTR(ip, whichfork);
2842         ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
2843                 (uint)sizeof(xfs_bmbt_rec_t)));
2844         ASSERT(del->br_blockcount > 0);
2845         ep = xfs_iext_get_ext(ifp, *idx);
2846         xfs_bmbt_get_all(ep, &got);
2847         ASSERT(got.br_startoff <= del->br_startoff);
2848         del_endoff = del->br_startoff + del->br_blockcount;
2849         got_endoff = got.br_startoff + got.br_blockcount;
2850         ASSERT(got_endoff >= del_endoff);
2851         delay = isnullstartblock(got.br_startblock);
2852         ASSERT(isnullstartblock(del->br_startblock) == delay);
2853         flags = 0;
2854         qfield = 0;
2855         error = 0;
2856         /*
2857          * If deleting a real allocation, must free up the disk space.
2858          */
2859         if (!delay) {
2860                 flags = XFS_ILOG_CORE;
2861                 /*
2862                  * Realtime allocation.  Free it and record di_nblocks update.
2863                  */
2864                 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
2865                         xfs_fsblock_t   bno;
2866                         xfs_filblks_t   len;
2867
2868                         ASSERT(do_mod(del->br_blockcount,
2869                                       mp->m_sb.sb_rextsize) == 0);
2870                         ASSERT(do_mod(del->br_startblock,
2871                                       mp->m_sb.sb_rextsize) == 0);
2872                         bno = del->br_startblock;
2873                         len = del->br_blockcount;
2874                         do_div(bno, mp->m_sb.sb_rextsize);
2875                         do_div(len, mp->m_sb.sb_rextsize);
2876                         if ((error = xfs_rtfree_extent(ip->i_transp, bno,
2877                                         (xfs_extlen_t)len)))
2878                                 goto done;
2879                         do_fx = 0;
2880                         nblks = len * mp->m_sb.sb_rextsize;
2881                         qfield = XFS_TRANS_DQ_RTBCOUNT;
2882                 }
2883                 /*
2884                  * Ordinary allocation.
2885                  */
2886                 else {
2887                         do_fx = 1;
2888                         nblks = del->br_blockcount;
2889                         qfield = XFS_TRANS_DQ_BCOUNT;
2890                 }
2891                 /*
2892                  * Set up del_endblock and cur for later.
2893                  */
2894                 del_endblock = del->br_startblock + del->br_blockcount;
2895                 if (cur) {
2896                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
2897                                         got.br_startblock, got.br_blockcount,
2898                                         &i)))
2899                                 goto done;
2900                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2901                 }
2902                 da_old = da_new = 0;
2903         } else {
2904                 da_old = startblockval(got.br_startblock);
2905                 da_new = 0;
2906                 nblks = 0;
2907                 do_fx = 0;
2908         }
2909         /*
2910          * Set flag value to use in switch statement.
2911          * Left-contig is 2, right-contig is 1.
2912          */
2913         switch (((got.br_startoff == del->br_startoff) << 1) |
2914                 (got_endoff == del_endoff)) {
2915         case 3:
2916                 /*
2917                  * Matches the whole extent.  Delete the entry.
2918                  */
2919                 xfs_iext_remove(ip, *idx, 1,
2920                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
2921                 if (delay)
2922                         break;
2923                 XFS_IFORK_NEXT_SET(ip, whichfork,
2924                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2925                 flags |= XFS_ILOG_CORE;
2926                 if (!cur) {
2927                         flags |= xfs_ilog_fext(whichfork);
2928                         break;
2929                 }
2930                 if ((error = xfs_btree_delete(cur, &i)))
2931                         goto done;
2932                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2933                 break;
2934
2935         case 2:
2936                 /*
2937                  * Deleting the first part of the extent.
2938                  */
2939                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2940                 xfs_bmbt_set_startoff(ep, del_endoff);
2941                 temp = got.br_blockcount - del->br_blockcount;
2942                 xfs_bmbt_set_blockcount(ep, temp);
2943                 if (delay) {
2944                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2945                                 da_old);
2946                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2947                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2948                         da_new = temp;
2949                         break;
2950                 }
2951                 xfs_bmbt_set_startblock(ep, del_endblock);
2952                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2953                 if (!cur) {
2954                         flags |= xfs_ilog_fext(whichfork);
2955                         break;
2956                 }
2957                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
2958                                 got.br_blockcount - del->br_blockcount,
2959                                 got.br_state)))
2960                         goto done;
2961                 break;
2962
2963         case 1:
2964                 /*
2965                  * Deleting the last part of the extent.
2966                  */
2967                 temp = got.br_blockcount - del->br_blockcount;
2968                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2969                 xfs_bmbt_set_blockcount(ep, temp);
2970                 if (delay) {
2971                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2972                                 da_old);
2973                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2974                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2975                         da_new = temp;
2976                         break;
2977                 }
2978                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2979                 if (!cur) {
2980                         flags |= xfs_ilog_fext(whichfork);
2981                         break;
2982                 }
2983                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
2984                                 got.br_startblock,
2985                                 got.br_blockcount - del->br_blockcount,
2986                                 got.br_state)))
2987                         goto done;
2988                 break;
2989
2990         case 0:
2991                 /*
2992                  * Deleting the middle of the extent.
2993                  */
2994                 temp = del->br_startoff - got.br_startoff;
2995                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2996                 xfs_bmbt_set_blockcount(ep, temp);
2997                 new.br_startoff = del_endoff;
2998                 temp2 = got_endoff - del_endoff;
2999                 new.br_blockcount = temp2;
3000                 new.br_state = got.br_state;
3001                 if (!delay) {
3002                         new.br_startblock = del_endblock;
3003                         flags |= XFS_ILOG_CORE;
3004                         if (cur) {
3005                                 if ((error = xfs_bmbt_update(cur,
3006                                                 got.br_startoff,
3007                                                 got.br_startblock, temp,
3008                                                 got.br_state)))
3009                                         goto done;
3010                                 if ((error = xfs_btree_increment(cur, 0, &i)))
3011                                         goto done;
3012                                 cur->bc_rec.b = new;
3013                                 error = xfs_btree_insert(cur, &i);
3014                                 if (error && error != ENOSPC)
3015                                         goto done;
3016                                 /*
3017                                  * If get no-space back from btree insert,
3018                                  * it tried a split, and we have a zero
3019                                  * block reservation.
3020                                  * Fix up our state and return the error.
3021                                  */
3022                                 if (error == ENOSPC) {
3023                                         /*
3024                                          * Reset the cursor, don't trust
3025                                          * it after any insert operation.
3026                                          */
3027                                         if ((error = xfs_bmbt_lookup_eq(cur,
3028                                                         got.br_startoff,
3029                                                         got.br_startblock,
3030                                                         temp, &i)))
3031                                                 goto done;
3032                                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3033                                         /*
3034                                          * Update the btree record back
3035                                          * to the original value.
3036                                          */
3037                                         if ((error = xfs_bmbt_update(cur,
3038                                                         got.br_startoff,
3039                                                         got.br_startblock,
3040                                                         got.br_blockcount,
3041                                                         got.br_state)))
3042                                                 goto done;
3043                                         /*
3044                                          * Reset the extent record back
3045                                          * to the original value.
3046                                          */
3047                                         xfs_bmbt_set_blockcount(ep,
3048                                                 got.br_blockcount);
3049                                         flags = 0;
3050                                         error = XFS_ERROR(ENOSPC);
3051                                         goto done;
3052                                 }
3053                                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3054                         } else
3055                                 flags |= xfs_ilog_fext(whichfork);
3056                         XFS_IFORK_NEXT_SET(ip, whichfork,
3057                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
3058                 } else {
3059                         ASSERT(whichfork == XFS_DATA_FORK);
3060                         temp = xfs_bmap_worst_indlen(ip, temp);
3061                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
3062                         temp2 = xfs_bmap_worst_indlen(ip, temp2);
3063                         new.br_startblock = nullstartblock((int)temp2);
3064                         da_new = temp + temp2;
3065                         while (da_new > da_old) {
3066                                 if (temp) {
3067                                         temp--;
3068                                         da_new--;
3069                                         xfs_bmbt_set_startblock(ep,
3070                                                 nullstartblock((int)temp));
3071                                 }
3072                                 if (da_new == da_old)
3073                                         break;
3074                                 if (temp2) {
3075                                         temp2--;
3076                                         da_new--;
3077                                         new.br_startblock =
3078                                                 nullstartblock((int)temp2);
3079                                 }
3080                         }
3081                 }
3082                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
3083                 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
3084                 ++*idx;
3085                 break;
3086         }
3087         /*
3088          * If we need to, add to list of extents to delete.
3089          */
3090         if (do_fx)
3091                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
3092                         mp);
3093         /*
3094          * Adjust inode # blocks in the file.
3095          */
3096         if (nblks)
3097                 ip->i_d.di_nblocks -= nblks;
3098         /*
3099          * Adjust quota data.
3100          */
3101         if (qfield)
3102                 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
3103
3104         /*
3105          * Account for change in delayed indirect blocks.
3106          * Nothing to do for disk quota accounting here.
3107          */
3108         ASSERT(da_old >= da_new);
3109         if (da_old > da_new) {
3110                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
3111                         (int64_t)(da_old - da_new), 0);
3112         }
3113 done:
3114         *logflagsp = flags;
3115         return error;
3116 }
3117
3118 /*
3119  * Remove the entry "free" from the free item list.  Prev points to the
3120  * previous entry, unless "free" is the head of the list.
3121  */
3122 STATIC void
3123 xfs_bmap_del_free(
3124         xfs_bmap_free_t         *flist, /* free item list header */
3125         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
3126         xfs_bmap_free_item_t    *free)  /* list item to be freed */
3127 {
3128         if (prev)
3129                 prev->xbfi_next = free->xbfi_next;
3130         else
3131                 flist->xbf_first = free->xbfi_next;
3132         flist->xbf_count--;
3133         kmem_zone_free(xfs_bmap_free_item_zone, free);
3134 }
3135
3136 /*
3137  * Convert an extents-format file into a btree-format file.
3138  * The new file will have a root block (in the inode) and a single child block.
3139  */
3140 STATIC int                                      /* error */
3141 xfs_bmap_extents_to_btree(
3142         xfs_trans_t             *tp,            /* transaction pointer */
3143         xfs_inode_t             *ip,            /* incore inode pointer */
3144         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
3145         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
3146         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
3147         int                     wasdel,         /* converting a delayed alloc */
3148         int                     *logflagsp,     /* inode logging flags */
3149         int                     whichfork)      /* data or attr fork */
3150 {
3151         struct xfs_btree_block  *ablock;        /* allocated (child) bt block */
3152         xfs_buf_t               *abp;           /* buffer for ablock */
3153         xfs_alloc_arg_t         args;           /* allocation arguments */
3154         xfs_bmbt_rec_t          *arp;           /* child record pointer */
3155         struct xfs_btree_block  *block;         /* btree root block */
3156         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
3157         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
3158         int                     error;          /* error return value */
3159         xfs_extnum_t            i, cnt;         /* extent record index */
3160         xfs_ifork_t             *ifp;           /* inode fork pointer */
3161         xfs_bmbt_key_t          *kp;            /* root block key pointer */
3162         xfs_mount_t             *mp;            /* mount structure */
3163         xfs_extnum_t            nextents;       /* number of file extents */
3164         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
3165
3166         ifp = XFS_IFORK_PTR(ip, whichfork);
3167         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3168         ASSERT(ifp->if_ext_max ==
3169                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
3170         /*
3171          * Make space in the inode incore.
3172          */
3173         xfs_iroot_realloc(ip, 1, whichfork);
3174         ifp->if_flags |= XFS_IFBROOT;
3175
3176         /*
3177          * Fill in the root.
3178          */
3179         block = ifp->if_broot;
3180         block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3181         block->bb_level = cpu_to_be16(1);
3182         block->bb_numrecs = cpu_to_be16(1);
3183         block->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3184         block->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3185
3186         /*
3187          * Need a cursor.  Can't allocate until bb_level is filled in.
3188          */
3189         mp = ip->i_mount;
3190         cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
3191         cur->bc_private.b.firstblock = *firstblock;
3192         cur->bc_private.b.flist = flist;
3193         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3194         /*
3195          * Convert to a btree with two levels, one record in root.
3196          */
3197         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3198         args.tp = tp;
3199         args.mp = mp;
3200         args.firstblock = *firstblock;
3201         if (*firstblock == NULLFSBLOCK) {
3202                 args.type = XFS_ALLOCTYPE_START_BNO;
3203                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3204         } else if (flist->xbf_low) {
3205                 args.type = XFS_ALLOCTYPE_START_BNO;
3206                 args.fsbno = *firstblock;
3207         } else {
3208                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3209                 args.fsbno = *firstblock;
3210         }
3211         args.minlen = args.maxlen = args.prod = 1;
3212         args.total = args.minleft = args.alignment = args.mod = args.isfl =
3213                 args.minalignslop = 0;
3214         args.wasdel = wasdel;
3215         *logflagsp = 0;
3216         if ((error = xfs_alloc_vextent(&args))) {
3217                 xfs_iroot_realloc(ip, -1, whichfork);
3218                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3219                 return error;
3220         }
3221         /*
3222          * Allocation can't fail, the space was reserved.
3223          */
3224         ASSERT(args.fsbno != NULLFSBLOCK);
3225         ASSERT(*firstblock == NULLFSBLOCK ||
3226                args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3227                (flist->xbf_low &&
3228                 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3229         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3230         cur->bc_private.b.allocated++;
3231         ip->i_d.di_nblocks++;
3232         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3233         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3234         /*
3235          * Fill in the child block.
3236          */
3237         ablock = XFS_BUF_TO_BLOCK(abp);
3238         ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3239         ablock->bb_level = 0;
3240         ablock->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3241         ablock->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3242         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3243         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3244         for (cnt = i = 0; i < nextents; i++) {
3245                 ep = xfs_iext_get_ext(ifp, i);
3246                 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
3247                         arp->l0 = cpu_to_be64(ep->l0);
3248                         arp->l1 = cpu_to_be64(ep->l1);
3249                         arp++; cnt++;
3250                 }
3251         }
3252         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3253         xfs_btree_set_numrecs(ablock, cnt);
3254
3255         /*
3256          * Fill in the root key and pointer.
3257          */
3258         kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
3259         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3260         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
3261         pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
3262                                                 be16_to_cpu(block->bb_level)));
3263         *pp = cpu_to_be64(args.fsbno);
3264
3265         /*
3266          * Do all this logging at the end so that
3267          * the root is at the right level.
3268          */
3269         xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
3270         xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3271         ASSERT(*curp == NULL);
3272         *curp = cur;
3273         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
3274         return 0;
3275 }
3276
3277 /*
3278  * Calculate the default attribute fork offset for newly created inodes.
3279  */
3280 uint
3281 xfs_default_attroffset(
3282         struct xfs_inode        *ip)
3283 {
3284         struct xfs_mount        *mp = ip->i_mount;
3285         uint                    offset;
3286
3287         if (mp->m_sb.sb_inodesize == 256) {
3288                 offset = XFS_LITINO(mp) -
3289                                 XFS_BMDR_SPACE_CALC(MINABTPTRS);
3290         } else {
3291                 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
3292         }
3293
3294         ASSERT(offset < XFS_LITINO(mp));
3295         return offset;
3296 }
3297
3298 /*
3299  * Helper routine to reset inode di_forkoff field when switching
3300  * attribute fork from local to extent format - we reset it where
3301  * possible to make space available for inline data fork extents.
3302  */
3303 STATIC void
3304 xfs_bmap_forkoff_reset(
3305         xfs_mount_t     *mp,
3306         xfs_inode_t     *ip,
3307         int             whichfork)
3308 {
3309         if (whichfork == XFS_ATTR_FORK &&
3310             ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
3311             ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
3312             ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
3313                 uint    dfl_forkoff = xfs_default_attroffset(ip) >> 3;
3314
3315                 if (dfl_forkoff > ip->i_d.di_forkoff) {
3316                         ip->i_d.di_forkoff = dfl_forkoff;
3317                         ip->i_df.if_ext_max =
3318                                 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t);
3319                         ip->i_afp->if_ext_max =
3320                                 XFS_IFORK_ASIZE(ip) / sizeof(xfs_bmbt_rec_t);
3321                 }
3322         }
3323 }
3324
3325 /*
3326  * Convert a local file to an extents file.
3327  * This code is out of bounds for data forks of regular files,
3328  * since the file data needs to get logged so things will stay consistent.
3329  * (The bmap-level manipulations are ok, though).
3330  */
3331 STATIC int                              /* error */
3332 xfs_bmap_local_to_extents(
3333         xfs_trans_t     *tp,            /* transaction pointer */
3334         xfs_inode_t     *ip,            /* incore inode pointer */
3335         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
3336         xfs_extlen_t    total,          /* total blocks needed by transaction */
3337         int             *logflagsp,     /* inode logging flags */
3338         int             whichfork)      /* data or attr fork */
3339 {
3340         int             error;          /* error return value */
3341         int             flags;          /* logging flags returned */
3342         xfs_ifork_t     *ifp;           /* inode fork pointer */
3343
3344         /*
3345          * We don't want to deal with the case of keeping inode data inline yet.
3346          * So sending the data fork of a regular inode is invalid.
3347          */
3348         ASSERT(!((ip->i_d.di_mode & S_IFMT) == S_IFREG &&
3349                  whichfork == XFS_DATA_FORK));
3350         ifp = XFS_IFORK_PTR(ip, whichfork);
3351         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3352         flags = 0;
3353         error = 0;
3354         if (ifp->if_bytes) {
3355                 xfs_alloc_arg_t args;   /* allocation arguments */
3356                 xfs_buf_t       *bp;    /* buffer for extent block */
3357                 xfs_bmbt_rec_host_t *ep;/* extent record pointer */
3358
3359                 args.tp = tp;
3360                 args.mp = ip->i_mount;
3361                 args.firstblock = *firstblock;
3362                 ASSERT((ifp->if_flags &
3363                         (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == XFS_IFINLINE);
3364                 /*
3365                  * Allocate a block.  We know we need only one, since the
3366                  * file currently fits in an inode.
3367                  */
3368                 if (*firstblock == NULLFSBLOCK) {
3369                         args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3370                         args.type = XFS_ALLOCTYPE_START_BNO;
3371                 } else {
3372                         args.fsbno = *firstblock;
3373                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
3374                 }
3375                 args.total = total;
3376                 args.mod = args.minleft = args.alignment = args.wasdel =
3377                         args.isfl = args.minalignslop = 0;
3378                 args.minlen = args.maxlen = args.prod = 1;
3379                 if ((error = xfs_alloc_vextent(&args)))
3380                         goto done;
3381                 /*
3382                  * Can't fail, the space was reserved.
3383                  */
3384                 ASSERT(args.fsbno != NULLFSBLOCK);
3385                 ASSERT(args.len == 1);
3386                 *firstblock = args.fsbno;
3387                 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3388                 memcpy((char *)XFS_BUF_PTR(bp), ifp->if_u1.if_data,
3389                         ifp->if_bytes);
3390                 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3391                 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3392                 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3393                 xfs_iext_add(ifp, 0, 1);
3394                 ep = xfs_iext_get_ext(ifp, 0);
3395                 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3396                 trace_xfs_bmap_post_update(ip, 0,
3397                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
3398                                 _THIS_IP_);
3399                 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3400                 ip->i_d.di_nblocks = 1;
3401                 xfs_trans_mod_dquot_byino(tp, ip,
3402                         XFS_TRANS_DQ_BCOUNT, 1L);
3403                 flags |= xfs_ilog_fext(whichfork);
3404         } else {
3405                 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3406                 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3407         }
3408         ifp->if_flags &= ~XFS_IFINLINE;
3409         ifp->if_flags |= XFS_IFEXTENTS;
3410         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3411         flags |= XFS_ILOG_CORE;
3412 done:
3413         *logflagsp = flags;
3414         return error;
3415 }
3416
3417 /*
3418  * Search the extent records for the entry containing block bno.
3419  * If bno lies in a hole, point to the next entry.  If bno lies
3420  * past eof, *eofp will be set, and *prevp will contain the last
3421  * entry (null if none).  Else, *lastxp will be set to the index
3422  * of the found entry; *gotp will contain the entry.
3423  */
3424 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
3425 xfs_bmap_search_multi_extents(
3426         xfs_ifork_t     *ifp,           /* inode fork pointer */
3427         xfs_fileoff_t   bno,            /* block number searched for */
3428         int             *eofp,          /* out: end of file found */
3429         xfs_extnum_t    *lastxp,        /* out: last extent index */
3430         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3431         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3432 {
3433         xfs_bmbt_rec_host_t *ep;                /* extent record pointer */
3434         xfs_extnum_t    lastx;          /* last extent index */
3435
3436         /*
3437          * Initialize the extent entry structure to catch access to
3438          * uninitialized br_startblock field.
3439          */
3440         gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
3441         gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3442         gotp->br_state = XFS_EXT_INVALID;
3443 #if XFS_BIG_BLKNOS
3444         gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
3445 #else
3446         gotp->br_startblock = 0xffffa5a5;
3447 #endif
3448         prevp->br_startoff = NULLFILEOFF;
3449
3450         ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
3451         if (lastx > 0) {
3452                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
3453         }
3454         if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
3455                 xfs_bmbt_get_all(ep, gotp);
3456                 *eofp = 0;
3457         } else {
3458                 if (lastx > 0) {
3459                         *gotp = *prevp;
3460                 }
3461                 *eofp = 1;
3462                 ep = NULL;
3463         }
3464         *lastxp = lastx;
3465         return ep;
3466 }
3467
3468 /*
3469  * Search the extents list for the inode, for the extent containing bno.
3470  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
3471  * *eofp will be set, and *prevp will contain the last entry (null if none).
3472  * Else, *lastxp will be set to the index of the found
3473  * entry; *gotp will contain the entry.
3474  */
3475 STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
3476 xfs_bmap_search_extents(
3477         xfs_inode_t     *ip,            /* incore inode pointer */
3478         xfs_fileoff_t   bno,            /* block number searched for */
3479         int             fork,           /* data or attr fork */
3480         int             *eofp,          /* out: end of file found */
3481         xfs_extnum_t    *lastxp,        /* out: last extent index */
3482         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3483         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3484 {
3485         xfs_ifork_t     *ifp;           /* inode fork pointer */
3486         xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
3487
3488         XFS_STATS_INC(xs_look_exlist);
3489         ifp = XFS_IFORK_PTR(ip, fork);
3490
3491         ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
3492
3493         if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
3494                      !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
3495                 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
3496                                 "Access to block zero in inode %llu "
3497                                 "start_block: %llx start_off: %llx "
3498                                 "blkcnt: %llx extent-state: %x lastx: %x\n",
3499                         (unsigned long long)ip->i_ino,
3500                         (unsigned long long)gotp->br_startblock,
3501                         (unsigned long long)gotp->br_startoff,
3502                         (unsigned long long)gotp->br_blockcount,
3503                         gotp->br_state, *lastxp);
3504                 *lastxp = NULLEXTNUM;
3505                 *eofp = 1;
3506                 return NULL;
3507         }
3508         return ep;
3509 }
3510
3511 /*
3512  * Compute the worst-case number of indirect blocks that will be used
3513  * for ip's delayed extent of length "len".
3514  */
3515 STATIC xfs_filblks_t
3516 xfs_bmap_worst_indlen(
3517         xfs_inode_t     *ip,            /* incore inode pointer */
3518         xfs_filblks_t   len)            /* delayed extent length */
3519 {
3520         int             level;          /* btree level number */
3521         int             maxrecs;        /* maximum record count at this level */
3522         xfs_mount_t     *mp;            /* mount structure */
3523         xfs_filblks_t   rval;           /* return value */
3524
3525         mp = ip->i_mount;
3526         maxrecs = mp->m_bmap_dmxr[0];
3527         for (level = 0, rval = 0;
3528              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3529              level++) {
3530                 len += maxrecs - 1;
3531                 do_div(len, maxrecs);
3532                 rval += len;
3533                 if (len == 1)
3534                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3535                                 level - 1;
3536                 if (level == 0)
3537                         maxrecs = mp->m_bmap_dmxr[1];
3538         }
3539         return rval;
3540 }
3541
3542 /*
3543  * Convert inode from non-attributed to attributed.
3544  * Must not be in a transaction, ip must not be locked.
3545  */
3546 int                                             /* error code */
3547 xfs_bmap_add_attrfork(
3548         xfs_inode_t             *ip,            /* incore inode pointer */
3549         int                     size,           /* space new attribute needs */
3550         int                     rsvd)           /* xact may use reserved blks */
3551 {
3552         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
3553         xfs_bmap_free_t         flist;          /* freed extent records */
3554         xfs_mount_t             *mp;            /* mount structure */
3555         xfs_trans_t             *tp;            /* transaction pointer */
3556         int                     blks;           /* space reservation */
3557         int                     version = 1;    /* superblock attr version */
3558         int                     committed;      /* xaction was committed */
3559         int                     logflags;       /* logging flags */
3560         int                     error;          /* error return value */
3561
3562         ASSERT(XFS_IFORK_Q(ip) == 0);
3563         ASSERT(ip->i_df.if_ext_max ==
3564                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3565
3566         mp = ip->i_mount;
3567         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3568         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3569         blks = XFS_ADDAFORK_SPACE_RES(mp);
3570         if (rsvd)
3571                 tp->t_flags |= XFS_TRANS_RESERVE;
3572         if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3573                         XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3574                 goto error0;
3575         xfs_ilock(ip, XFS_ILOCK_EXCL);
3576         error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
3577                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3578                         XFS_QMOPT_RES_REGBLKS);
3579         if (error) {
3580                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3581                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3582                 return error;
3583         }
3584         if (XFS_IFORK_Q(ip))
3585                 goto error1;
3586         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3587                 /*
3588                  * For inodes coming from pre-6.2 filesystems.
3589                  */
3590                 ASSERT(ip->i_d.di_aformat == 0);
3591                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
3592         }
3593         ASSERT(ip->i_d.di_anextents == 0);
3594
3595         xfs_trans_ijoin_ref(tp, ip, XFS_ILOCK_EXCL);
3596         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3597
3598         switch (ip->i_d.di_format) {
3599         case XFS_DINODE_FMT_DEV:
3600                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
3601                 break;
3602         case XFS_DINODE_FMT_UUID:
3603                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
3604                 break;
3605         case XFS_DINODE_FMT_LOCAL:
3606         case XFS_DINODE_FMT_EXTENTS:
3607         case XFS_DINODE_FMT_BTREE:
3608                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
3609                 if (!ip->i_d.di_forkoff)
3610                         ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
3611                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
3612                         version = 2;
3613                 break;
3614         default:
3615                 ASSERT(0);
3616                 error = XFS_ERROR(EINVAL);
3617                 goto error1;
3618         }
3619         ip->i_df.if_ext_max =
3620                 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3621         ASSERT(ip->i_afp == NULL);
3622         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
3623         ip->i_afp->if_ext_max =
3624                 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3625         ip->i_afp->if_flags = XFS_IFEXTENTS;
3626         logflags = 0;
3627         xfs_bmap_init(&flist, &firstblock);
3628         switch (ip->i_d.di_format) {
3629         case XFS_DINODE_FMT_LOCAL:
3630                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
3631                         &logflags);
3632                 break;
3633         case XFS_DINODE_FMT_EXTENTS:
3634                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
3635                         &flist, &logflags);
3636                 break;
3637         case XFS_DINODE_FMT_BTREE:
3638                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
3639                         &logflags);
3640                 break;
3641         default:
3642                 error = 0;
3643                 break;
3644         }
3645         if (logflags)
3646                 xfs_trans_log_inode(tp, ip, logflags);
3647         if (error)
3648                 goto error2;
3649         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
3650            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
3651                 __int64_t sbfields = 0;
3652
3653                 spin_lock(&mp->m_sb_lock);
3654                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
3655                         xfs_sb_version_addattr(&mp->m_sb);
3656                         sbfields |= XFS_SB_VERSIONNUM;
3657                 }
3658                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
3659                         xfs_sb_version_addattr2(&mp->m_sb);
3660                         sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
3661                 }
3662                 if (sbfields) {
3663                         spin_unlock(&mp->m_sb_lock);
3664                         xfs_mod_sb(tp, sbfields);
3665                 } else
3666                         spin_unlock(&mp->m_sb_lock);
3667         }
3668         if ((error = xfs_bmap_finish(&tp, &flist, &committed)))
3669                 goto error2;
3670         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3671         ASSERT(ip->i_df.if_ext_max ==
3672                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3673         return error;
3674 error2:
3675         xfs_bmap_cancel(&flist);
3676 error1:
3677         xfs_iunlock(ip, XFS_ILOCK_EXCL);
3678 error0:
3679         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
3680         ASSERT(ip->i_df.if_ext_max ==
3681                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3682         return error;
3683 }
3684
3685 /*
3686  * Add the extent to the list of extents to be free at transaction end.
3687  * The list is maintained sorted (by block number).
3688  */
3689 /* ARGSUSED */
3690 void
3691 xfs_bmap_add_free(
3692         xfs_fsblock_t           bno,            /* fs block number of extent */
3693         xfs_filblks_t           len,            /* length of extent */
3694         xfs_bmap_free_t         *flist,         /* list of extents */
3695         xfs_mount_t             *mp)            /* mount point structure */
3696 {
3697         xfs_bmap_free_item_t    *cur;           /* current (next) element */
3698         xfs_bmap_free_item_t    *new;           /* new element */
3699         xfs_bmap_free_item_t    *prev;          /* previous element */
3700 #ifdef DEBUG
3701         xfs_agnumber_t          agno;
3702         xfs_agblock_t           agbno;
3703
3704         ASSERT(bno != NULLFSBLOCK);
3705         ASSERT(len > 0);
3706         ASSERT(len <= MAXEXTLEN);
3707         ASSERT(!isnullstartblock(bno));
3708         agno = XFS_FSB_TO_AGNO(mp, bno);
3709         agbno = XFS_FSB_TO_AGBNO(mp, bno);
3710         ASSERT(agno < mp->m_sb.sb_agcount);
3711         ASSERT(agbno < mp->m_sb.sb_agblocks);
3712         ASSERT(len < mp->m_sb.sb_agblocks);
3713         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
3714 #endif
3715         ASSERT(xfs_bmap_free_item_zone != NULL);
3716         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
3717         new->xbfi_startblock = bno;
3718         new->xbfi_blockcount = (xfs_extlen_t)len;
3719         for (prev = NULL, cur = flist->xbf_first;
3720              cur != NULL;
3721              prev = cur, cur = cur->xbfi_next) {
3722                 if (cur->xbfi_startblock >= bno)
3723                         break;
3724         }
3725         if (prev)
3726                 prev->xbfi_next = new;
3727         else
3728                 flist->xbf_first = new;
3729         new->xbfi_next = cur;
3730         flist->xbf_count++;
3731 }
3732
3733 /*
3734  * Compute and fill in the value of the maximum depth of a bmap btree
3735  * in this filesystem.  Done once, during mount.
3736  */
3737 void
3738 xfs_bmap_compute_maxlevels(
3739         xfs_mount_t     *mp,            /* file system mount structure */
3740         int             whichfork)      /* data or attr fork */
3741 {
3742         int             level;          /* btree level */
3743         uint            maxblocks;      /* max blocks at this level */
3744         uint            maxleafents;    /* max leaf entries possible */
3745         int             maxrootrecs;    /* max records in root block */
3746         int             minleafrecs;    /* min records in leaf block */
3747         int             minnoderecs;    /* min records in node block */
3748         int             sz;             /* root block size */
3749
3750         /*
3751          * The maximum number of extents in a file, hence the maximum
3752          * number of leaf entries, is controlled by the type of di_nextents
3753          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
3754          * (a signed 16-bit number, xfs_aextnum_t).
3755          *
3756          * Note that we can no longer assume that if we are in ATTR1 that
3757          * the fork offset of all the inodes will be
3758          * (xfs_default_attroffset(ip) >> 3) because we could have mounted
3759          * with ATTR2 and then mounted back with ATTR1, keeping the
3760          * di_forkoff's fixed but probably at various positions. Therefore,
3761          * for both ATTR1 and ATTR2 we have to assume the worst case scenario
3762          * of a minimum size available.
3763          */
3764         if (whichfork == XFS_DATA_FORK) {
3765                 maxleafents = MAXEXTNUM;
3766                 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
3767         } else {
3768                 maxleafents = MAXAEXTNUM;
3769                 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
3770         }
3771         maxrootrecs = xfs_bmdr_maxrecs(mp, sz, 0);
3772         minleafrecs = mp->m_bmap_dmnr[0];
3773         minnoderecs = mp->m_bmap_dmnr[1];
3774         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
3775         for (level = 1; maxblocks > 1; level++) {
3776                 if (maxblocks <= maxrootrecs)
3777                         maxblocks = 1;
3778                 else
3779                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
3780         }
3781         mp->m_bm_maxlevels[whichfork] = level;
3782 }
3783
3784 /*
3785  * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
3786  * caller.  Frees all the extents that need freeing, which must be done
3787  * last due to locking considerations.  We never free any extents in
3788  * the first transaction.  This is to allow the caller to make the first
3789  * transaction a synchronous one so that the pointers to the data being
3790  * broken in this transaction will be permanent before the data is actually
3791  * freed.  This is necessary to prevent blocks from being reallocated
3792  * and written to before the free and reallocation are actually permanent.
3793  * We do not just make the first transaction synchronous here, because
3794  * there are more efficient ways to gain the same protection in some cases
3795  * (see the file truncation code).
3796  *
3797  * Return 1 if the given transaction was committed and a new one
3798  * started, and 0 otherwise in the committed parameter.
3799  */
3800 /*ARGSUSED*/
3801 int                                             /* error */
3802 xfs_bmap_finish(
3803         xfs_trans_t             **tp,           /* transaction pointer addr */
3804         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
3805         int                     *committed)     /* xact committed or not */
3806 {
3807         xfs_efd_log_item_t      *efd;           /* extent free data */
3808         xfs_efi_log_item_t      *efi;           /* extent free intention */
3809         int                     error;          /* error return value */
3810         xfs_bmap_free_item_t    *free;          /* free extent item */
3811         unsigned int            logres;         /* new log reservation */
3812         unsigned int            logcount;       /* new log count */
3813         xfs_mount_t             *mp;            /* filesystem mount structure */
3814         xfs_bmap_free_item_t    *next;          /* next item on free list */
3815         xfs_trans_t             *ntp;           /* new transaction pointer */
3816
3817         ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
3818         if (flist->xbf_count == 0) {
3819                 *committed = 0;
3820                 return 0;
3821         }
3822         ntp = *tp;
3823         efi = xfs_trans_get_efi(ntp, flist->xbf_count);
3824         for (free = flist->xbf_first; free; free = free->xbfi_next)
3825                 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
3826                         free->xbfi_blockcount);
3827         logres = ntp->t_log_res;
3828         logcount = ntp->t_log_count;
3829         ntp = xfs_trans_dup(*tp);
3830         error = xfs_trans_commit(*tp, 0);
3831         *tp = ntp;
3832         *committed = 1;
3833         /*
3834          * We have a new transaction, so we should return committed=1,
3835          * even though we're returning an error.
3836          */
3837         if (error)
3838                 return error;
3839
3840         /*
3841          * transaction commit worked ok so we can drop the extra ticket
3842          * reference that we gained in xfs_trans_dup()
3843          */
3844         xfs_log_ticket_put(ntp->t_ticket);
3845
3846         if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
3847                         logcount)))
3848                 return error;
3849         efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
3850         for (free = flist->xbf_first; free != NULL; free = next) {
3851                 next = free->xbfi_next;
3852                 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
3853                                 free->xbfi_blockcount))) {
3854                         /*
3855                          * The bmap free list will be cleaned up at a
3856                          * higher level.  The EFI will be canceled when
3857                          * this transaction is aborted.
3858                          * Need to force shutdown here to make sure it
3859                          * happens, since this transaction may not be
3860                          * dirty yet.
3861                          */
3862                         mp = ntp->t_mountp;
3863                         if (!XFS_FORCED_SHUTDOWN(mp))
3864                                 xfs_force_shutdown(mp,
3865                                                    (error == EFSCORRUPTED) ?
3866                                                    SHUTDOWN_CORRUPT_INCORE :
3867                                                    SHUTDOWN_META_IO_ERROR);
3868                         return error;
3869                 }
3870                 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
3871                         free->xbfi_blockcount);
3872                 xfs_bmap_del_free(flist, NULL, free);
3873         }
3874         return 0;
3875 }
3876
3877 /*
3878  * Free up any items left in the list.
3879  */
3880 void
3881 xfs_bmap_cancel(
3882         xfs_bmap_free_t         *flist) /* list of bmap_free_items */
3883 {
3884         xfs_bmap_free_item_t    *free;  /* free list item */
3885         xfs_bmap_free_item_t    *next;
3886
3887         if (flist->xbf_count == 0)
3888                 return;
3889         ASSERT(flist->xbf_first != NULL);
3890         for (free = flist->xbf_first; free; free = next) {
3891                 next = free->xbfi_next;
3892                 xfs_bmap_del_free(flist, NULL, free);
3893         }
3894         ASSERT(flist->xbf_count == 0);
3895 }
3896
3897 /*
3898  * Returns the file-relative block number of the first unused block(s)
3899  * in the file with at least "len" logically contiguous blocks free.
3900  * This is the lowest-address hole if the file has holes, else the first block
3901  * past the end of file.
3902  * Return 0 if the file is currently local (in-inode).
3903  */
3904 int                                             /* error */
3905 xfs_bmap_first_unused(
3906         xfs_trans_t     *tp,                    /* transaction pointer */
3907         xfs_inode_t     *ip,                    /* incore inode */
3908         xfs_extlen_t    len,                    /* size of hole to find */
3909         xfs_fileoff_t   *first_unused,          /* unused block */
3910         int             whichfork)              /* data or attr fork */
3911 {
3912         int             error;                  /* error return value */
3913         int             idx;                    /* extent record index */
3914         xfs_ifork_t     *ifp;                   /* inode fork pointer */
3915         xfs_fileoff_t   lastaddr;               /* last block number seen */
3916         xfs_fileoff_t   lowest;                 /* lowest useful block */
3917         xfs_fileoff_t   max;                    /* starting useful block */
3918         xfs_fileoff_t   off;                    /* offset for this block */
3919         xfs_extnum_t    nextents;               /* number of extent entries */
3920
3921         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
3922                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
3923                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3924         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3925                 *first_unused = 0;
3926                 return 0;
3927         }
3928         ifp = XFS_IFORK_PTR(ip, whichfork);
3929         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3930             (error = xfs_iread_extents(tp, ip, whichfork)))
3931                 return error;
3932         lowest = *first_unused;
3933         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3934         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
3935                 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
3936                 off = xfs_bmbt_get_startoff(ep);
3937                 /*
3938                  * See if the hole before this extent will work.
3939                  */
3940                 if (off >= lowest + len && off - max >= len) {
3941                         *first_unused = max;
3942                         return 0;
3943                 }
3944                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
3945                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
3946         }
3947         *first_unused = max;
3948         return 0;
3949 }
3950
3951 /*
3952  * Returns the file-relative block number of the last block + 1 before
3953  * last_block (input value) in the file.
3954  * This is not based on i_size, it is based on the extent records.
3955  * Returns 0 for local files, as they do not have extent records.
3956  */
3957 int                                             /* error */
3958 xfs_bmap_last_before(
3959         xfs_trans_t     *tp,                    /* transaction pointer */
3960         xfs_inode_t     *ip,                    /* incore inode */
3961         xfs_fileoff_t   *last_block,            /* last block */
3962         int             whichfork)              /* data or attr fork */
3963 {
3964         xfs_fileoff_t   bno;                    /* input file offset */
3965         int             eof;                    /* hit end of file */
3966         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
3967         int             error;                  /* error return value */
3968         xfs_bmbt_irec_t got;                    /* current extent value */
3969         xfs_ifork_t     *ifp;                   /* inode fork pointer */
3970         xfs_extnum_t    lastx;                  /* last extent used */
3971         xfs_bmbt_irec_t prev;                   /* previous extent value */
3972
3973         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
3974             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
3975             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
3976                return XFS_ERROR(EIO);
3977         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3978                 *last_block = 0;
3979                 return 0;
3980         }
3981         ifp = XFS_IFORK_PTR(ip, whichfork);
3982         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3983             (error = xfs_iread_extents(tp, ip, whichfork)))
3984                 return error;
3985         bno = *last_block - 1;
3986         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
3987                 &prev);
3988         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
3989                 if (prev.br_startoff == NULLFILEOFF)
3990                         *last_block = 0;
3991                 else
3992                         *last_block = prev.br_startoff + prev.br_blockcount;
3993         }
3994         /*
3995          * Otherwise *last_block is already the right answer.
3996          */
3997         return 0;
3998 }
3999
4000 /*
4001  * Returns the file-relative block number of the first block past eof in
4002  * the file.  This is not based on i_size, it is based on the extent records.
4003  * Returns 0 for local files, as they do not have extent records.
4004  */
4005 int                                             /* error */
4006 xfs_bmap_last_offset(
4007         xfs_trans_t     *tp,                    /* transaction pointer */
4008         xfs_inode_t     *ip,                    /* incore inode */
4009         xfs_fileoff_t   *last_block,            /* last block */
4010         int             whichfork)              /* data or attr fork */
4011 {
4012         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
4013         int             error;                  /* error return value */
4014         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4015         xfs_extnum_t    nextents;               /* number of extent entries */
4016
4017         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4018             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4019             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4020                return XFS_ERROR(EIO);
4021         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4022                 *last_block = 0;
4023                 return 0;
4024         }
4025         ifp = XFS_IFORK_PTR(ip, whichfork);
4026         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4027             (error = xfs_iread_extents(tp, ip, whichfork)))
4028                 return error;
4029         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4030         if (!nextents) {
4031                 *last_block = 0;
4032                 return 0;
4033         }
4034         ep = xfs_iext_get_ext(ifp, nextents - 1);
4035         *last_block = xfs_bmbt_get_startoff(ep) + xfs_bmbt_get_blockcount(ep);
4036         return 0;
4037 }
4038
4039 /*
4040  * Returns whether the selected fork of the inode has exactly one
4041  * block or not.  For the data fork we check this matches di_size,
4042  * implying the file's range is 0..bsize-1.
4043  */
4044 int                                     /* 1=>1 block, 0=>otherwise */
4045 xfs_bmap_one_block(
4046         xfs_inode_t     *ip,            /* incore inode */
4047         int             whichfork)      /* data or attr fork */
4048 {
4049         xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
4050         xfs_ifork_t     *ifp;           /* inode fork pointer */
4051         int             rval;           /* return value */
4052         xfs_bmbt_irec_t s;              /* internal version of extent */
4053
4054 #ifndef DEBUG
4055         if (whichfork == XFS_DATA_FORK) {
4056                 return ((ip->i_d.di_mode & S_IFMT) == S_IFREG) ?
4057                         (ip->i_size == ip->i_mount->m_sb.sb_blocksize) :
4058                         (ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize);
4059         }
4060 #endif  /* !DEBUG */
4061         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4062                 return 0;
4063         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4064                 return 0;
4065         ifp = XFS_IFORK_PTR(ip, whichfork);
4066         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4067         ep = xfs_iext_get_ext(ifp, 0);
4068         xfs_bmbt_get_all(ep, &s);
4069         rval = s.br_startoff == 0 && s.br_blockcount == 1;
4070         if (rval && whichfork == XFS_DATA_FORK)
4071                 ASSERT(ip->i_size == ip->i_mount->m_sb.sb_blocksize);
4072         return rval;
4073 }
4074
4075 STATIC int
4076 xfs_bmap_sanity_check(
4077         struct xfs_mount        *mp,
4078         struct xfs_buf          *bp,
4079         int                     level)
4080 {
4081         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4082
4083         if (be32_to_cpu(block->bb_magic) != XFS_BMAP_MAGIC ||
4084             be16_to_cpu(block->bb_level) != level ||
4085             be16_to_cpu(block->bb_numrecs) == 0 ||
4086             be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
4087                 return 0;
4088         return 1;
4089 }
4090
4091 /*
4092  * Read in the extents to if_extents.
4093  * All inode fields are set up by caller, we just traverse the btree
4094  * and copy the records in. If the file system cannot contain unwritten
4095  * extents, the records are checked for no "state" flags.
4096  */
4097 int                                     /* error */
4098 xfs_bmap_read_extents(
4099         xfs_trans_t             *tp,    /* transaction pointer */
4100         xfs_inode_t             *ip,    /* incore inode */
4101         int                     whichfork) /* data or attr fork */
4102 {
4103         struct xfs_btree_block  *block; /* current btree block */
4104         xfs_fsblock_t           bno;    /* block # of "block" */
4105         xfs_buf_t               *bp;    /* buffer for "block" */
4106         int                     error;  /* error return value */
4107         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
4108         xfs_extnum_t            i, j;   /* index into the extents list */
4109         xfs_ifork_t             *ifp;   /* fork structure */
4110         int                     level;  /* btree level, for checking */
4111         xfs_mount_t             *mp;    /* file system mount structure */
4112         __be64                  *pp;    /* pointer to block address */
4113         /* REFERENCED */
4114         xfs_extnum_t            room;   /* number of entries there's room for */
4115
4116         bno = NULLFSBLOCK;
4117         mp = ip->i_mount;
4118         ifp = XFS_IFORK_PTR(ip, whichfork);
4119         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4120                                         XFS_EXTFMT_INODE(ip);
4121         block = ifp->if_broot;
4122         /*
4123          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4124          */
4125         level = be16_to_cpu(block->bb_level);
4126         ASSERT(level > 0);
4127         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
4128         bno = be64_to_cpu(*pp);
4129         ASSERT(bno != NULLDFSBNO);
4130         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
4131         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
4132         /*
4133          * Go down the tree until leaf level is reached, following the first
4134          * pointer (leftmost) at each level.
4135          */
4136         while (level-- > 0) {
4137                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4138                                 XFS_BMAP_BTREE_REF)))
4139                         return error;
4140                 block = XFS_BUF_TO_BLOCK(bp);
4141                 XFS_WANT_CORRUPTED_GOTO(
4142                         xfs_bmap_sanity_check(mp, bp, level),
4143                         error0);
4144                 if (level == 0)
4145                         break;
4146                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
4147                 bno = be64_to_cpu(*pp);
4148                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
4149                 xfs_trans_brelse(tp, bp);
4150         }
4151         /*
4152          * Here with bp and block set to the leftmost leaf node in the tree.
4153          */
4154         room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4155         i = 0;
4156         /*
4157          * Loop over all leaf nodes.  Copy information to the extent records.
4158          */
4159         for (;;) {
4160                 xfs_bmbt_rec_t  *frp;
4161                 xfs_fsblock_t   nextbno;
4162                 xfs_extnum_t    num_recs;
4163                 xfs_extnum_t    start;
4164
4165
4166                 num_recs = xfs_btree_get_numrecs(block);
4167                 if (unlikely(i + num_recs > room)) {
4168                         ASSERT(i + num_recs <= room);
4169                         xfs_warn(ip->i_mount,
4170                                 "corrupt dinode %Lu, (btree extents).",
4171                                 (unsigned long long) ip->i_ino);
4172                         XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
4173                                 XFS_ERRLEVEL_LOW, ip->i_mount, block);
4174                         goto error0;
4175                 }
4176                 XFS_WANT_CORRUPTED_GOTO(
4177                         xfs_bmap_sanity_check(mp, bp, 0),
4178                         error0);
4179                 /*
4180                  * Read-ahead the next leaf block, if any.
4181                  */
4182                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
4183                 if (nextbno != NULLFSBLOCK)
4184                         xfs_btree_reada_bufl(mp, nextbno, 1);
4185                 /*
4186                  * Copy records into the extent records.
4187                  */
4188                 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
4189                 start = i;
4190                 for (j = 0; j < num_recs; j++, i++, frp++) {
4191                         xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
4192                         trp->l0 = be64_to_cpu(frp->l0);
4193                         trp->l1 = be64_to_cpu(frp->l1);
4194                 }
4195                 if (exntf == XFS_EXTFMT_NOSTATE) {
4196                         /*
4197                          * Check all attribute bmap btree records and
4198                          * any "older" data bmap btree records for a
4199                          * set bit in the "extent flag" position.
4200                          */
4201                         if (unlikely(xfs_check_nostate_extents(ifp,
4202                                         start, num_recs))) {
4203                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4204                                                  XFS_ERRLEVEL_LOW,
4205                                                  ip->i_mount);
4206                                 goto error0;
4207                         }
4208                 }
4209                 xfs_trans_brelse(tp, bp);
4210                 bno = nextbno;
4211                 /*
4212                  * If we've reached the end, stop.
4213                  */
4214                 if (bno == NULLFSBLOCK)
4215                         break;
4216                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4217                                 XFS_BMAP_BTREE_REF)))
4218                         return error;
4219                 block = XFS_BUF_TO_BLOCK(bp);
4220         }
4221         ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4222         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4223         XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
4224         return 0;
4225 error0:
4226         xfs_trans_brelse(tp, bp);
4227         return XFS_ERROR(EFSCORRUPTED);
4228 }
4229
4230 #ifdef DEBUG
4231 /*
4232  * Add bmap trace insert entries for all the contents of the extent records.
4233  */
4234 void
4235 xfs_bmap_trace_exlist(
4236         xfs_inode_t     *ip,            /* incore inode pointer */
4237         xfs_extnum_t    cnt,            /* count of entries in the list */
4238         int             whichfork,      /* data or attr fork */
4239         unsigned long   caller_ip)
4240 {
4241         xfs_extnum_t    idx;            /* extent record index */
4242         xfs_ifork_t     *ifp;           /* inode fork pointer */
4243         int             state = 0;
4244
4245         if (whichfork == XFS_ATTR_FORK)
4246                 state |= BMAP_ATTRFORK;
4247
4248         ifp = XFS_IFORK_PTR(ip, whichfork);
4249         ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4250         for (idx = 0; idx < cnt; idx++)
4251                 trace_xfs_extlist(ip, idx, whichfork, caller_ip);
4252 }
4253
4254 /*
4255  * Validate that the bmbt_irecs being returned from bmapi are valid
4256  * given the callers original parameters.  Specifically check the
4257  * ranges of the returned irecs to ensure that they only extent beyond
4258  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4259  */
4260 STATIC void
4261 xfs_bmap_validate_ret(
4262         xfs_fileoff_t           bno,
4263         xfs_filblks_t           len,
4264         int                     flags,
4265         xfs_bmbt_irec_t         *mval,
4266         int                     nmap,
4267         int                     ret_nmap)
4268 {
4269         int                     i;              /* index to map values */
4270
4271         ASSERT(ret_nmap <= nmap);
4272
4273         for (i = 0; i < ret_nmap; i++) {
4274                 ASSERT(mval[i].br_blockcount > 0);
4275                 if (!(flags & XFS_BMAPI_ENTIRE)) {
4276                         ASSERT(mval[i].br_startoff >= bno);
4277                         ASSERT(mval[i].br_blockcount <= len);
4278                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4279                                bno + len);
4280                 } else {
4281                         ASSERT(mval[i].br_startoff < bno + len);
4282                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4283                                bno);
4284                 }
4285                 ASSERT(i == 0 ||
4286                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4287                        mval[i].br_startoff);
4288                 if ((flags & XFS_BMAPI_WRITE) && !(flags & XFS_BMAPI_DELAY))
4289                         ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4290                                mval[i].br_startblock != HOLESTARTBLOCK);
4291                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4292                        mval[i].br_state == XFS_EXT_UNWRITTEN);
4293         }
4294 }
4295 #endif /* DEBUG */
4296
4297
4298 /*
4299  * Map file blocks to filesystem blocks.
4300  * File range is given by the bno/len pair.
4301  * Adds blocks to file if a write ("flags & XFS_BMAPI_WRITE" set)
4302  * into a hole or past eof.
4303  * Only allocates blocks from a single allocation group,
4304  * to avoid locking problems.
4305  * The returned value in "firstblock" from the first call in a transaction
4306  * must be remembered and presented to subsequent calls in "firstblock".
4307  * An upper bound for the number of blocks to be allocated is supplied to
4308  * the first call in "total"; if no allocation group has that many free
4309  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4310  */
4311 int                                     /* error */
4312 xfs_bmapi(
4313         xfs_trans_t     *tp,            /* transaction pointer */
4314         xfs_inode_t     *ip,            /* incore inode */
4315         xfs_fileoff_t   bno,            /* starting file offs. mapped */
4316         xfs_filblks_t   len,            /* length to map in file */
4317         int             flags,          /* XFS_BMAPI_... */
4318         xfs_fsblock_t   *firstblock,    /* first allocated block
4319                                            controls a.g. for allocs */
4320         xfs_extlen_t    total,          /* total blocks needed */
4321         xfs_bmbt_irec_t *mval,          /* output: map values */
4322         int             *nmap,          /* i/o: mval size/count */
4323         xfs_bmap_free_t *flist)         /* i/o: list extents to free */
4324 {
4325         xfs_fsblock_t   abno;           /* allocated block number */
4326         xfs_extlen_t    alen;           /* allocated extent length */
4327         xfs_fileoff_t   aoff;           /* allocated file offset */
4328         xfs_bmalloca_t  bma = { 0 };    /* args for xfs_bmap_alloc */
4329         xfs_btree_cur_t *cur;           /* bmap btree cursor */
4330         xfs_fileoff_t   end;            /* end of mapped file region */
4331         int             eof;            /* we've hit the end of extents */
4332         xfs_bmbt_rec_host_t *ep;        /* extent record pointer */
4333         int             error;          /* error return */
4334         xfs_bmbt_irec_t got;            /* current file extent record */
4335         xfs_ifork_t     *ifp;           /* inode fork pointer */
4336         xfs_extlen_t    indlen;         /* indirect blocks length */
4337         xfs_extnum_t    lastx;          /* last useful extent number */
4338         int             logflags;       /* flags for transaction logging */
4339         xfs_extlen_t    minleft;        /* min blocks left after allocation */
4340         xfs_extlen_t    minlen;         /* min allocation size */
4341         xfs_mount_t     *mp;            /* xfs mount structure */
4342         int             n;              /* current extent index */
4343         int             nallocs;        /* number of extents alloc'd */
4344         xfs_extnum_t    nextents;       /* number of extents in file */
4345         xfs_fileoff_t   obno;           /* old block number (offset) */
4346         xfs_bmbt_irec_t prev;           /* previous file extent record */
4347         int             tmp_logflags;   /* temp flags holder */
4348         int             whichfork;      /* data or attr fork */
4349         char            inhole;         /* current location is hole in file */
4350         char            wasdelay;       /* old extent was delayed */
4351         char            wr;             /* this is a write request */
4352         char            rt;             /* this is a realtime file */
4353 #ifdef DEBUG
4354         xfs_fileoff_t   orig_bno;       /* original block number value */
4355         int             orig_flags;     /* original flags arg value */
4356         xfs_filblks_t   orig_len;       /* original value of len arg */
4357         xfs_bmbt_irec_t *orig_mval;     /* original value of mval */
4358         int             orig_nmap;      /* original value of *nmap */
4359
4360         orig_bno = bno;
4361         orig_len = len;
4362         orig_flags = flags;
4363         orig_mval = mval;
4364         orig_nmap = *nmap;
4365 #endif
4366         ASSERT(*nmap >= 1);
4367         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP || !(flags & XFS_BMAPI_WRITE));
4368         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4369                 XFS_ATTR_FORK : XFS_DATA_FORK;
4370         mp = ip->i_mount;
4371         if (unlikely(XFS_TEST_ERROR(
4372             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4373              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4374              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4375              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4376                 XFS_ERROR_REPORT("xfs_bmapi", XFS_ERRLEVEL_LOW, mp);
4377                 return XFS_ERROR(EFSCORRUPTED);
4378         }
4379         if (XFS_FORCED_SHUTDOWN(mp))
4380                 return XFS_ERROR(EIO);
4381         rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4382         ifp = XFS_IFORK_PTR(ip, whichfork);
4383         ASSERT(ifp->if_ext_max ==
4384                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4385         if ((wr = (flags & XFS_BMAPI_WRITE)) != 0)
4386                 XFS_STATS_INC(xs_blk_mapw);
4387         else
4388                 XFS_STATS_INC(xs_blk_mapr);
4389         /*
4390          * IGSTATE flag is used to combine extents which
4391          * differ only due to the state of the extents.
4392          * This technique is used from xfs_getbmap()
4393          * when the caller does not wish to see the
4394          * separation (which is the default).
4395          *
4396          * This technique is also used when writing a
4397          * buffer which has been partially written,
4398          * (usually by being flushed during a chunkread),
4399          * to ensure one write takes place. This also
4400          * prevents a change in the xfs inode extents at
4401          * this time, intentionally. This change occurs
4402          * on completion of the write operation, in
4403          * xfs_strat_comp(), where the xfs_bmapi() call
4404          * is transactioned, and the extents combined.
4405          */
4406         if ((flags & XFS_BMAPI_IGSTATE) && wr)  /* if writing unwritten space */
4407                 wr = 0;                         /* no allocations are allowed */
4408         ASSERT(wr || !(flags & XFS_BMAPI_DELAY));
4409         logflags = 0;
4410         nallocs = 0;
4411         cur = NULL;
4412         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4413                 ASSERT(wr && tp);
4414                 if ((error = xfs_bmap_local_to_extents(tp, ip,
4415                                 firstblock, total, &logflags, whichfork)))
4416                         goto error0;
4417         }
4418         if (wr && *firstblock == NULLFSBLOCK) {
4419                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4420                         minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4421                 else
4422                         minleft = 1;
4423         } else
4424                 minleft = 0;
4425         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4426             (error = xfs_iread_extents(tp, ip, whichfork)))
4427                 goto error0;
4428         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4429                 &prev);
4430         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4431         n = 0;
4432         end = bno + len;
4433         obno = bno;
4434         bma.ip = NULL;
4435
4436         while (bno < end && n < *nmap) {
4437                 /*
4438                  * Reading past eof, act as though there's a hole
4439                  * up to end.
4440                  */
4441                 if (eof && !wr)
4442                         got.br_startoff = end;
4443                 inhole = eof || got.br_startoff > bno;
4444                 wasdelay = wr && !inhole && !(flags & XFS_BMAPI_DELAY) &&
4445                         isnullstartblock(got.br_startblock);
4446                 /*
4447                  * First, deal with the hole before the allocated space
4448                  * that we found, if any.
4449                  */
4450                 if (wr && (inhole || wasdelay)) {
4451                         /*
4452                          * For the wasdelay case, we could also just
4453                          * allocate the stuff asked for in this bmap call
4454                          * but that wouldn't be as good.
4455                          */
4456                         if (wasdelay) {
4457                                 alen = (xfs_extlen_t)got.br_blockcount;
4458                                 aoff = got.br_startoff;
4459                                 if (lastx != NULLEXTNUM && lastx) {
4460                                         ep = xfs_iext_get_ext(ifp, lastx - 1);
4461                                         xfs_bmbt_get_all(ep, &prev);
4462                                 }
4463                         } else {
4464                                 alen = (xfs_extlen_t)
4465                                         XFS_FILBLKS_MIN(len, MAXEXTLEN);
4466                                 if (!eof)
4467                                         alen = (xfs_extlen_t)
4468                                                 XFS_FILBLKS_MIN(alen,
4469                                                         got.br_startoff - bno);
4470                                 aoff = bno;
4471                         }
4472                         minlen = (flags & XFS_BMAPI_CONTIG) ? alen : 1;
4473                         if (flags & XFS_BMAPI_DELAY) {
4474                                 xfs_extlen_t    extsz;
4475
4476                                 /* Figure out the extent size, adjust alen */
4477                                 extsz = xfs_get_extsz_hint(ip);
4478                                 if (extsz) {
4479                                         /*
4480                                          * make sure we don't exceed a single
4481                                          * extent length when we align the
4482                                          * extent by reducing length we are
4483                                          * going to allocate by the maximum
4484                                          * amount extent size aligment may
4485                                          * require.
4486                                          */
4487                                         alen = XFS_FILBLKS_MIN(len,
4488                                                    MAXEXTLEN - (2 * extsz - 1));
4489                                         error = xfs_bmap_extsize_align(mp,
4490                                                         &got, &prev, extsz,
4491                                                         rt, eof,
4492                                                         flags&XFS_BMAPI_DELAY,
4493                                                         flags&XFS_BMAPI_CONVERT,
4494                                                         &aoff, &alen);
4495                                         ASSERT(!error);
4496                                 }
4497
4498                                 if (rt)
4499                                         extsz = alen / mp->m_sb.sb_rextsize;
4500
4501                                 /*
4502                                  * Make a transaction-less quota reservation for
4503                                  * delayed allocation blocks. This number gets
4504                                  * adjusted later.  We return if we haven't
4505                                  * allocated blocks already inside this loop.
4506                                  */
4507                                 error = xfs_trans_reserve_quota_nblks(
4508                                                 NULL, ip, (long)alen, 0,
4509                                                 rt ? XFS_QMOPT_RES_RTBLKS :
4510                                                      XFS_QMOPT_RES_REGBLKS);
4511                                 if (error) {
4512                                         if (n == 0) {
4513                                                 *nmap = 0;
4514                                                 ASSERT(cur == NULL);
4515                                                 return error;
4516                                         }
4517                                         break;
4518                                 }
4519
4520                                 /*
4521                                  * Split changing sb for alen and indlen since
4522                                  * they could be coming from different places.
4523                                  */
4524                                 indlen = (xfs_extlen_t)
4525                                         xfs_bmap_worst_indlen(ip, alen);
4526                                 ASSERT(indlen > 0);
4527
4528                                 if (rt) {
4529                                         error = xfs_mod_incore_sb(mp,
4530                                                         XFS_SBS_FREXTENTS,
4531                                                         -((int64_t)extsz), 0);
4532                                 } else {
4533                                         error = xfs_icsb_modify_counters(mp,
4534                                                         XFS_SBS_FDBLOCKS,
4535                                                         -((int64_t)alen), 0);
4536                                 }
4537                                 if (!error) {
4538                                         error = xfs_icsb_modify_counters(mp,
4539                                                         XFS_SBS_FDBLOCKS,
4540                                                         -((int64_t)indlen), 0);
4541                                         if (error && rt)
4542                                                 xfs_mod_incore_sb(mp,
4543                                                         XFS_SBS_FREXTENTS,
4544                                                         (int64_t)extsz, 0);
4545                                         else if (error)
4546                                                 xfs_icsb_modify_counters(mp,
4547                                                         XFS_SBS_FDBLOCKS,
4548                                                         (int64_t)alen, 0);
4549                                 }
4550
4551                                 if (error) {
4552                                         if (XFS_IS_QUOTA_ON(mp))
4553                                                 /* unreserve the blocks now */
4554                                                 (void)
4555                                                 xfs_trans_unreserve_quota_nblks(
4556                                                         NULL, ip,
4557                                                         (long)alen, 0, rt ?
4558                                                         XFS_QMOPT_RES_RTBLKS :
4559                                                         XFS_QMOPT_RES_REGBLKS);
4560                                         break;
4561                                 }
4562
4563                                 ip->i_delayed_blks += alen;
4564                                 abno = nullstartblock(indlen);
4565                         } else {
4566                                 /*
4567                                  * If first time, allocate and fill in
4568                                  * once-only bma fields.
4569                                  */
4570                                 if (bma.ip == NULL) {
4571                                         bma.tp = tp;
4572                                         bma.ip = ip;
4573                                         bma.prevp = &prev;
4574                                         bma.gotp = &got;
4575                                         bma.total = total;
4576                                         bma.userdata = 0;
4577                                 }
4578                                 /* Indicate if this is the first user data
4579                                  * in the file, or just any user data.
4580                                  */
4581                                 if (!(flags & XFS_BMAPI_METADATA)) {
4582                                         bma.userdata = (aoff == 0) ?
4583                                                 XFS_ALLOC_INITIAL_USER_DATA :
4584                                                 XFS_ALLOC_USERDATA;
4585                                 }
4586                                 /*
4587                                  * Fill in changeable bma fields.
4588                                  */
4589                                 bma.eof = eof;
4590                                 bma.firstblock = *firstblock;
4591                                 bma.alen = alen;
4592                                 bma.off = aoff;
4593                                 bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4594                                 bma.wasdel = wasdelay;
4595                                 bma.minlen = minlen;
4596                                 bma.low = flist->xbf_low;
4597                                 bma.minleft = minleft;
4598                                 /*
4599                                  * Only want to do the alignment at the
4600                                  * eof if it is userdata and allocation length
4601                                  * is larger than a stripe unit.
4602                                  */
4603                                 if (mp->m_dalign && alen >= mp->m_dalign &&
4604                                     (!(flags & XFS_BMAPI_METADATA)) &&
4605                                     (whichfork == XFS_DATA_FORK)) {
4606                                         if ((error = xfs_bmap_isaeof(ip, aoff,
4607                                                         whichfork, &bma.aeof)))
4608                                                 goto error0;
4609                                 } else
4610                                         bma.aeof = 0;
4611                                 /*
4612                                  * Call allocator.
4613                                  */
4614                                 if ((error = xfs_bmap_alloc(&bma)))
4615                                         goto error0;
4616                                 /*
4617                                  * Copy out result fields.
4618                                  */
4619                                 abno = bma.rval;
4620                                 if ((flist->xbf_low = bma.low))
4621                                         minleft = 0;
4622                                 alen = bma.alen;
4623                                 aoff = bma.off;
4624                                 ASSERT(*firstblock == NULLFSBLOCK ||
4625                                        XFS_FSB_TO_AGNO(mp, *firstblock) ==
4626                                        XFS_FSB_TO_AGNO(mp, bma.firstblock) ||
4627                                        (flist->xbf_low &&
4628                                         XFS_FSB_TO_AGNO(mp, *firstblock) <
4629                                         XFS_FSB_TO_AGNO(mp, bma.firstblock)));
4630                                 *firstblock = bma.firstblock;
4631                                 if (cur)
4632                                         cur->bc_private.b.firstblock =
4633                                                 *firstblock;
4634                                 if (abno == NULLFSBLOCK)
4635                                         break;
4636                                 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
4637                                         cur = xfs_bmbt_init_cursor(mp, tp,
4638                                                 ip, whichfork);
4639                                         cur->bc_private.b.firstblock =
4640                                                 *firstblock;
4641                                         cur->bc_private.b.flist = flist;
4642                                 }
4643                                 /*
4644                                  * Bump the number of extents we've allocated
4645                                  * in this call.
4646                                  */
4647                                 nallocs++;
4648                         }
4649                         if (cur)
4650                                 cur->bc_private.b.flags =
4651                                         wasdelay ? XFS_BTCUR_BPRV_WASDEL : 0;
4652                         got.br_startoff = aoff;
4653                         got.br_startblock = abno;
4654                         got.br_blockcount = alen;
4655                         got.br_state = XFS_EXT_NORM;    /* assume normal */
4656                         /*
4657                          * Determine state of extent, and the filesystem.
4658                          * A wasdelay extent has been initialized, so
4659                          * shouldn't be flagged as unwritten.
4660                          */
4661                         if (wr && xfs_sb_version_hasextflgbit(&mp->m_sb)) {
4662                                 if (!wasdelay && (flags & XFS_BMAPI_PREALLOC))
4663                                         got.br_state = XFS_EXT_UNWRITTEN;
4664                         }
4665                         error = xfs_bmap_add_extent(ip, &lastx, &cur, &got,
4666                                 firstblock, flist, &tmp_logflags,
4667                                 whichfork);
4668                         logflags |= tmp_logflags;
4669                         if (error)
4670                                 goto error0;
4671                         ep = xfs_iext_get_ext(ifp, lastx);
4672                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4673                         xfs_bmbt_get_all(ep, &got);
4674                         ASSERT(got.br_startoff <= aoff);
4675                         ASSERT(got.br_startoff + got.br_blockcount >=
4676                                 aoff + alen);
4677 #ifdef DEBUG
4678                         if (flags & XFS_BMAPI_DELAY) {
4679                                 ASSERT(isnullstartblock(got.br_startblock));
4680                                 ASSERT(startblockval(got.br_startblock) > 0);
4681                         }
4682                         ASSERT(got.br_state == XFS_EXT_NORM ||
4683                                got.br_state == XFS_EXT_UNWRITTEN);
4684 #endif
4685                         /*
4686                          * Fall down into the found allocated space case.
4687                          */
4688                 } else if (inhole) {
4689                         /*
4690                          * Reading in a hole.
4691                          */
4692                         mval->br_startoff = bno;
4693                         mval->br_startblock = HOLESTARTBLOCK;
4694                         mval->br_blockcount =
4695                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4696                         mval->br_state = XFS_EXT_NORM;
4697                         bno += mval->br_blockcount;
4698                         len -= mval->br_blockcount;
4699                         mval++;
4700                         n++;
4701                         continue;
4702                 }
4703                 /*
4704                  * Then deal with the allocated space we found.
4705                  */
4706                 ASSERT(ep != NULL);
4707                 if (!(flags & XFS_BMAPI_ENTIRE) &&
4708                     (got.br_startoff + got.br_blockcount > obno)) {
4709                         if (obno > bno)
4710                                 bno = obno;
4711                         ASSERT((bno >= obno) || (n == 0));
4712                         ASSERT(bno < end);
4713                         mval->br_startoff = bno;
4714                         if (isnullstartblock(got.br_startblock)) {
4715                                 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
4716                                 mval->br_startblock = DELAYSTARTBLOCK;
4717                         } else
4718                                 mval->br_startblock =
4719                                         got.br_startblock +
4720                                         (bno - got.br_startoff);
4721                         /*
4722                          * Return the minimum of what we got and what we
4723                          * asked for for the length.  We can use the len
4724                          * variable here because it is modified below
4725                          * and we could have been there before coming
4726                          * here if the first part of the allocation
4727                          * didn't overlap what was asked for.
4728                          */
4729                         mval->br_blockcount =
4730                                 XFS_FILBLKS_MIN(end - bno, got.br_blockcount -
4731                                         (bno - got.br_startoff));
4732                         mval->br_state = got.br_state;
4733                         ASSERT(mval->br_blockcount <= len);
4734                 } else {
4735                         *mval = got;
4736                         if (isnullstartblock(mval->br_startblock)) {
4737                                 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
4738                                 mval->br_startblock = DELAYSTARTBLOCK;
4739                         }
4740                 }
4741
4742                 /*
4743                  * Check if writing previously allocated but
4744                  * unwritten extents.
4745                  */
4746                 if (wr &&
4747                     ((mval->br_state == XFS_EXT_UNWRITTEN &&
4748                       ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_DELAY)) == 0)) ||
4749                      (mval->br_state == XFS_EXT_NORM &&
4750                       ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_CONVERT)) ==
4751                                 (XFS_BMAPI_PREALLOC|XFS_BMAPI_CONVERT))))) {
4752                         /*
4753                          * Modify (by adding) the state flag, if writing.
4754                          */
4755                         ASSERT(mval->br_blockcount <= len);
4756                         if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
4757                                 cur = xfs_bmbt_init_cursor(mp,
4758                                         tp, ip, whichfork);
4759                                 cur->bc_private.b.firstblock =
4760                                         *firstblock;
4761                                 cur->bc_private.b.flist = flist;
4762                         }
4763                         mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4764                                                 ? XFS_EXT_NORM
4765                                                 : XFS_EXT_UNWRITTEN;
4766                         error = xfs_bmap_add_extent(ip, &lastx, &cur, mval,
4767                                 firstblock, flist, &tmp_logflags,
4768                                 whichfork);
4769                         logflags |= tmp_logflags;
4770                         if (error)
4771                                 goto error0;
4772                         ep = xfs_iext_get_ext(ifp, lastx);
4773                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4774                         xfs_bmbt_get_all(ep, &got);
4775                         /*
4776                          * We may have combined previously unwritten
4777                          * space with written space, so generate
4778                          * another request.
4779                          */
4780                         if (mval->br_blockcount < len)
4781                                 continue;
4782                 }
4783
4784                 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4785                        ((mval->br_startoff + mval->br_blockcount) <= end));
4786                 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4787                        (mval->br_blockcount <= len) ||
4788                        (mval->br_startoff < obno));
4789                 bno = mval->br_startoff + mval->br_blockcount;
4790                 len = end - bno;
4791                 if (n > 0 && mval->br_startoff == mval[-1].br_startoff) {
4792                         ASSERT(mval->br_startblock == mval[-1].br_startblock);
4793                         ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
4794                         ASSERT(mval->br_state == mval[-1].br_state);
4795                         mval[-1].br_blockcount = mval->br_blockcount;
4796                         mval[-1].br_state = mval->br_state;
4797                 } else if (n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
4798                            mval[-1].br_startblock != DELAYSTARTBLOCK &&
4799                            mval[-1].br_startblock != HOLESTARTBLOCK &&
4800                            mval->br_startblock ==
4801                            mval[-1].br_startblock + mval[-1].br_blockcount &&
4802                            ((flags & XFS_BMAPI_IGSTATE) ||
4803                                 mval[-1].br_state == mval->br_state)) {
4804                         ASSERT(mval->br_startoff ==
4805                                mval[-1].br_startoff + mval[-1].br_blockcount);
4806                         mval[-1].br_blockcount += mval->br_blockcount;
4807                 } else if (n > 0 &&
4808                            mval->br_startblock == DELAYSTARTBLOCK &&
4809                            mval[-1].br_startblock == DELAYSTARTBLOCK &&
4810                            mval->br_startoff ==
4811                            mval[-1].br_startoff + mval[-1].br_blockcount) {
4812                         mval[-1].br_blockcount += mval->br_blockcount;
4813                         mval[-1].br_state = mval->br_state;
4814                 } else if (!((n == 0) &&
4815                              ((mval->br_startoff + mval->br_blockcount) <=
4816                               obno))) {
4817                         mval++;
4818                         n++;
4819                 }
4820                 /*
4821                  * If we're done, stop now.  Stop when we've allocated
4822                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4823                  * the transaction may get too big.
4824                  */
4825                 if (bno >= end || n >= *nmap || nallocs >= *nmap)
4826                         break;
4827                 /*
4828                  * Else go on to the next record.
4829                  */
4830                 ep = xfs_iext_get_ext(ifp, ++lastx);
4831                 prev = got;
4832                 if (lastx >= nextents)
4833                         eof = 1;
4834                 else
4835                         xfs_bmbt_get_all(ep, &got);
4836         }
4837         *nmap = n;
4838         /*
4839          * Transform from btree to extents, give it cur.
4840          */
4841         if (tp && XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
4842             XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
4843                 ASSERT(wr && cur);
4844                 error = xfs_bmap_btree_to_extents(tp, ip, cur,
4845                         &tmp_logflags, whichfork);
4846                 logflags |= tmp_logflags;
4847                 if (error)
4848                         goto error0;
4849         }
4850         ASSERT(ifp->if_ext_max ==
4851                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4852         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4853                XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max);
4854         error = 0;
4855 error0:
4856         /*
4857          * Log everything.  Do this after conversion, there's no point in
4858          * logging the extent records if we've converted to btree format.
4859          */
4860         if ((logflags & xfs_ilog_fext(whichfork)) &&
4861             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4862                 logflags &= ~xfs_ilog_fext(whichfork);
4863         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
4864                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4865                 logflags &= ~xfs_ilog_fbroot(whichfork);
4866         /*
4867          * Log whatever the flags say, even if error.  Otherwise we might miss
4868          * detecting a case where the data is changed, there's an error,
4869          * and it's not logged so we don't shutdown when we should.
4870          */
4871         if (logflags) {
4872                 ASSERT(tp && wr);
4873                 xfs_trans_log_inode(tp, ip, logflags);
4874         }
4875         if (cur) {
4876                 if (!error) {
4877                         ASSERT(*firstblock == NULLFSBLOCK ||
4878                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
4879                                XFS_FSB_TO_AGNO(mp,
4880                                        cur->bc_private.b.firstblock) ||
4881                                (flist->xbf_low &&
4882                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
4883                                 XFS_FSB_TO_AGNO(mp,
4884                                         cur->bc_private.b.firstblock)));
4885                         *firstblock = cur->bc_private.b.firstblock;
4886                 }
4887                 xfs_btree_del_cursor(cur,
4888                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4889         }
4890         if (!error)
4891                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4892                         orig_nmap, *nmap);
4893         return error;
4894 }
4895
4896 /*
4897  * Map file blocks to filesystem blocks, simple version.
4898  * One block (extent) only, read-only.
4899  * For flags, only the XFS_BMAPI_ATTRFORK flag is examined.
4900  * For the other flag values, the effect is as if XFS_BMAPI_METADATA
4901  * was set and all the others were clear.
4902  */
4903 int                                             /* error */
4904 xfs_bmapi_single(
4905         xfs_trans_t     *tp,            /* transaction pointer */
4906         xfs_inode_t     *ip,            /* incore inode */
4907         int             whichfork,      /* data or attr fork */
4908         xfs_fsblock_t   *fsb,           /* output: mapped block */
4909         xfs_fileoff_t   bno)            /* starting file offs. mapped */
4910 {
4911         int             eof;            /* we've hit the end of extents */
4912         int             error;          /* error return */
4913         xfs_bmbt_irec_t got;            /* current file extent record */
4914         xfs_ifork_t     *ifp;           /* inode fork pointer */
4915         xfs_extnum_t    lastx;          /* last useful extent number */
4916         xfs_bmbt_irec_t prev;           /* previous file extent record */
4917
4918         ifp = XFS_IFORK_PTR(ip, whichfork);
4919         if (unlikely(
4920             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4921             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)) {
4922                XFS_ERROR_REPORT("xfs_bmapi_single", XFS_ERRLEVEL_LOW,
4923                                 ip->i_mount);
4924                return XFS_ERROR(EFSCORRUPTED);
4925         }
4926         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
4927                 return XFS_ERROR(EIO);
4928         XFS_STATS_INC(xs_blk_mapr);
4929         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4930             (error = xfs_iread_extents(tp, ip, whichfork)))
4931                 return error;
4932         (void)xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4933                 &prev);
4934         /*
4935          * Reading past eof, act as though there's a hole
4936          * up to end.
4937          */
4938         if (eof || got.br_startoff > bno) {
4939                 *fsb = NULLFSBLOCK;
4940                 return 0;
4941         }
4942         ASSERT(!isnullstartblock(got.br_startblock));
4943         ASSERT(bno < got.br_startoff + got.br_blockcount);
4944         *fsb = got.br_startblock + (bno - got.br_startoff);
4945         return 0;
4946 }
4947
4948 /*
4949  * Unmap (remove) blocks from a file.
4950  * If nexts is nonzero then the number of extents to remove is limited to
4951  * that value.  If not all extents in the block range can be removed then
4952  * *done is set.
4953  */
4954 int                                             /* error */
4955 xfs_bunmapi(
4956         xfs_trans_t             *tp,            /* transaction pointer */
4957         struct xfs_inode        *ip,            /* incore inode */
4958         xfs_fileoff_t           bno,            /* starting offset to unmap */
4959         xfs_filblks_t           len,            /* length to unmap in file */
4960         int                     flags,          /* misc flags */
4961         xfs_extnum_t            nexts,          /* number of extents max */
4962         xfs_fsblock_t           *firstblock,    /* first allocated block
4963                                                    controls a.g. for allocs */
4964         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
4965         int                     *done)          /* set if not done yet */
4966 {
4967         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
4968         xfs_bmbt_irec_t         del;            /* extent being deleted */
4969         int                     eof;            /* is deleting at eof */
4970         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
4971         int                     error;          /* error return value */
4972         xfs_extnum_t            extno;          /* extent number in list */
4973         xfs_bmbt_irec_t         got;            /* current extent record */
4974         xfs_ifork_t             *ifp;           /* inode fork pointer */
4975         int                     isrt;           /* freeing in rt area */
4976         xfs_extnum_t            lastx;          /* last extent index used */
4977         int                     logflags;       /* transaction logging flags */
4978         xfs_extlen_t            mod;            /* rt extent offset */
4979         xfs_mount_t             *mp;            /* mount structure */
4980         xfs_extnum_t            nextents;       /* number of file extents */
4981         xfs_bmbt_irec_t         prev;           /* previous extent record */
4982         xfs_fileoff_t           start;          /* first file offset deleted */
4983         int                     tmp_logflags;   /* partial logging flags */
4984         int                     wasdel;         /* was a delayed alloc extent */
4985         int                     whichfork;      /* data or attribute fork */
4986         xfs_fsblock_t           sum;
4987
4988         trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
4989
4990         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4991                 XFS_ATTR_FORK : XFS_DATA_FORK;
4992         ifp = XFS_IFORK_PTR(ip, whichfork);
4993         if (unlikely(
4994             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4995             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
4996                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
4997                                  ip->i_mount);
4998                 return XFS_ERROR(EFSCORRUPTED);
4999         }
5000         mp = ip->i_mount;
5001         if (XFS_FORCED_SHUTDOWN(mp))
5002                 return XFS_ERROR(EIO);
5003
5004         ASSERT(len > 0);
5005         ASSERT(nexts >= 0);
5006         ASSERT(ifp->if_ext_max ==
5007                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5008         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5009             (error = xfs_iread_extents(tp, ip, whichfork)))
5010                 return error;
5011         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5012         if (nextents == 0) {
5013                 *done = 1;
5014                 return 0;
5015         }
5016         XFS_STATS_INC(xs_blk_unmap);
5017         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5018         start = bno;
5019         bno = start + len - 1;
5020         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5021                 &prev);
5022
5023         /*
5024          * Check to see if the given block number is past the end of the
5025          * file, back up to the last block if so...
5026          */
5027         if (eof) {
5028                 ep = xfs_iext_get_ext(ifp, --lastx);
5029                 xfs_bmbt_get_all(ep, &got);
5030                 bno = got.br_startoff + got.br_blockcount - 1;
5031         }
5032         logflags = 0;
5033         if (ifp->if_flags & XFS_IFBROOT) {
5034                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5035                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5036                 cur->bc_private.b.firstblock = *firstblock;
5037                 cur->bc_private.b.flist = flist;
5038                 cur->bc_private.b.flags = 0;
5039         } else
5040                 cur = NULL;
5041         extno = 0;
5042         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5043                (nexts == 0 || extno < nexts)) {
5044                 /*
5045                  * Is the found extent after a hole in which bno lives?
5046                  * Just back up to the previous extent, if so.
5047                  */
5048                 if (got.br_startoff > bno) {
5049                         if (--lastx < 0)
5050                                 break;
5051                         ep = xfs_iext_get_ext(ifp, lastx);
5052                         xfs_bmbt_get_all(ep, &got);
5053                 }
5054                 /*
5055                  * Is the last block of this extent before the range
5056                  * we're supposed to delete?  If so, we're done.
5057                  */
5058                 bno = XFS_FILEOFF_MIN(bno,
5059                         got.br_startoff + got.br_blockcount - 1);
5060                 if (bno < start)
5061                         break;
5062                 /*
5063                  * Then deal with the (possibly delayed) allocated space
5064                  * we found.
5065                  */
5066                 ASSERT(ep != NULL);
5067                 del = got;
5068                 wasdel = isnullstartblock(del.br_startblock);
5069                 if (got.br_startoff < start) {
5070                         del.br_startoff = start;
5071                         del.br_blockcount -= start - got.br_startoff;
5072                         if (!wasdel)
5073                                 del.br_startblock += start - got.br_startoff;
5074                 }
5075                 if (del.br_startoff + del.br_blockcount > bno + 1)
5076                         del.br_blockcount = bno + 1 - del.br_startoff;
5077                 sum = del.br_startblock + del.br_blockcount;
5078                 if (isrt &&
5079                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5080                         /*
5081                          * Realtime extent not lined up at the end.
5082                          * The extent could have been split into written
5083                          * and unwritten pieces, or we could just be
5084                          * unmapping part of it.  But we can't really
5085                          * get rid of part of a realtime extent.
5086                          */
5087                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5088                             !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5089                                 /*
5090                                  * This piece is unwritten, or we're not
5091                                  * using unwritten extents.  Skip over it.
5092                                  */
5093                                 ASSERT(bno >= mod);
5094                                 bno -= mod > del.br_blockcount ?
5095                                         del.br_blockcount : mod;
5096                                 if (bno < got.br_startoff) {
5097                                         if (--lastx >= 0)
5098                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5099                                                         ifp, lastx), &got);
5100                                 }
5101                                 continue;
5102                         }
5103                         /*
5104                          * It's written, turn it unwritten.
5105                          * This is better than zeroing it.
5106                          */
5107                         ASSERT(del.br_state == XFS_EXT_NORM);
5108                         ASSERT(xfs_trans_get_block_res(tp) > 0);
5109                         /*
5110                          * If this spans a realtime extent boundary,
5111                          * chop it back to the start of the one we end at.
5112                          */
5113                         if (del.br_blockcount > mod) {
5114                                 del.br_startoff += del.br_blockcount - mod;
5115                                 del.br_startblock += del.br_blockcount - mod;
5116                                 del.br_blockcount = mod;
5117                         }
5118                         del.br_state = XFS_EXT_UNWRITTEN;
5119                         error = xfs_bmap_add_extent(ip, &lastx, &cur, &del,
5120                                 firstblock, flist, &logflags,
5121                                 XFS_DATA_FORK);
5122                         if (error)
5123                                 goto error0;
5124                         goto nodelete;
5125                 }
5126                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5127                         /*
5128                          * Realtime extent is lined up at the end but not
5129                          * at the front.  We'll get rid of full extents if
5130                          * we can.
5131                          */
5132                         mod = mp->m_sb.sb_rextsize - mod;
5133                         if (del.br_blockcount > mod) {
5134                                 del.br_blockcount -= mod;
5135                                 del.br_startoff += mod;
5136                                 del.br_startblock += mod;
5137                         } else if ((del.br_startoff == start &&
5138                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5139                                      xfs_trans_get_block_res(tp) == 0)) ||
5140                                    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5141                                 /*
5142                                  * Can't make it unwritten.  There isn't
5143                                  * a full extent here so just skip it.
5144                                  */
5145                                 ASSERT(bno >= del.br_blockcount);
5146                                 bno -= del.br_blockcount;
5147                                 if (bno < got.br_startoff) {
5148                                         if (--lastx >= 0)
5149                                                 xfs_bmbt_get_all(--ep, &got);
5150                                 }
5151                                 continue;
5152                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5153                                 /*
5154                                  * This one is already unwritten.
5155                                  * It must have a written left neighbor.
5156                                  * Unwrite the killed part of that one and
5157                                  * try again.
5158                                  */
5159                                 ASSERT(lastx > 0);
5160                                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5161                                                 lastx - 1), &prev);
5162                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5163                                 ASSERT(!isnullstartblock(prev.br_startblock));
5164                                 ASSERT(del.br_startblock ==
5165                                        prev.br_startblock + prev.br_blockcount);
5166                                 if (prev.br_startoff < start) {
5167                                         mod = start - prev.br_startoff;
5168                                         prev.br_blockcount -= mod;
5169                                         prev.br_startblock += mod;
5170                                         prev.br_startoff = start;
5171                                 }
5172                                 prev.br_state = XFS_EXT_UNWRITTEN;
5173                                 lastx--;
5174                                 error = xfs_bmap_add_extent(ip, &lastx, &cur,
5175                                         &prev, firstblock, flist, &logflags,
5176                                         XFS_DATA_FORK);
5177                                 if (error)
5178                                         goto error0;
5179                                 goto nodelete;
5180                         } else {
5181                                 ASSERT(del.br_state == XFS_EXT_NORM);
5182                                 del.br_state = XFS_EXT_UNWRITTEN;
5183                                 error = xfs_bmap_add_extent(ip, &lastx, &cur,
5184                                         &del, firstblock, flist, &logflags,
5185                                         XFS_DATA_FORK);
5186                                 if (error)
5187                                         goto error0;
5188                                 goto nodelete;
5189                         }
5190                 }
5191                 if (wasdel) {
5192                         ASSERT(startblockval(del.br_startblock) > 0);
5193                         /* Update realtime/data freespace, unreserve quota */
5194                         if (isrt) {
5195                                 xfs_filblks_t rtexts;
5196
5197                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5198                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5199                                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5200                                                 (int64_t)rtexts, 0);
5201                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5202                                         ip, -((long)del.br_blockcount), 0,
5203                                         XFS_QMOPT_RES_RTBLKS);
5204                         } else {
5205                                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5206                                                 (int64_t)del.br_blockcount, 0);
5207                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5208                                         ip, -((long)del.br_blockcount), 0,
5209                                         XFS_QMOPT_RES_REGBLKS);
5210                         }
5211                         ip->i_delayed_blks -= del.br_blockcount;
5212                         if (cur)
5213                                 cur->bc_private.b.flags |=
5214                                         XFS_BTCUR_BPRV_WASDEL;
5215                 } else if (cur)
5216                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5217                 /*
5218                  * If it's the case where the directory code is running
5219                  * with no block reservation, and the deleted block is in
5220                  * the middle of its extent, and the resulting insert
5221                  * of an extent would cause transformation to btree format,
5222                  * then reject it.  The calling code will then swap
5223                  * blocks around instead.
5224                  * We have to do this now, rather than waiting for the
5225                  * conversion to btree format, since the transaction
5226                  * will be dirty.
5227                  */
5228                 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5229                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5230                     XFS_IFORK_NEXTENTS(ip, whichfork) >= ifp->if_ext_max &&
5231                     del.br_startoff > got.br_startoff &&
5232                     del.br_startoff + del.br_blockcount <
5233                     got.br_startoff + got.br_blockcount) {
5234                         error = XFS_ERROR(ENOSPC);
5235                         goto error0;
5236                 }
5237                 error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5238                                 &tmp_logflags, whichfork);
5239                 logflags |= tmp_logflags;
5240                 if (error)
5241                         goto error0;
5242                 bno = del.br_startoff - 1;
5243 nodelete:
5244                 /*
5245                  * If not done go on to the next (previous) record.
5246                  * Reset ep in case the extents array was re-alloced.
5247                  */
5248                 ep = xfs_iext_get_ext(ifp, lastx);
5249                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5250                         if (lastx >= XFS_IFORK_NEXTENTS(ip, whichfork) ||
5251                             xfs_bmbt_get_startoff(ep) > bno) {
5252                                 if (--lastx >= 0)
5253                                         ep = xfs_iext_get_ext(ifp, lastx);
5254                         }
5255                         if (lastx >= 0)
5256                                 xfs_bmbt_get_all(ep, &got);
5257                         extno++;
5258                 }
5259         }
5260         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5261         ASSERT(ifp->if_ext_max ==
5262                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5263         /*
5264          * Convert to a btree if necessary.
5265          */
5266         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5267             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
5268                 ASSERT(cur == NULL);
5269                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5270                         &cur, 0, &tmp_logflags, whichfork);
5271                 logflags |= tmp_logflags;
5272                 if (error)
5273                         goto error0;
5274         }
5275         /*
5276          * transform from btree to extents, give it cur
5277          */
5278         else if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5279                  XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5280                 ASSERT(cur != NULL);
5281                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5282                         whichfork);
5283                 logflags |= tmp_logflags;
5284                 if (error)
5285                         goto error0;
5286         }
5287         /*
5288          * transform from extents to local?
5289          */
5290         ASSERT(ifp->if_ext_max ==
5291                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5292         error = 0;
5293 error0:
5294         /*
5295          * Log everything.  Do this after conversion, there's no point in
5296          * logging the extent records if we've converted to btree format.
5297          */
5298         if ((logflags & xfs_ilog_fext(whichfork)) &&
5299             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5300                 logflags &= ~xfs_ilog_fext(whichfork);
5301         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5302                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5303                 logflags &= ~xfs_ilog_fbroot(whichfork);
5304         /*
5305          * Log inode even in the error case, if the transaction
5306          * is dirty we'll need to shut down the filesystem.
5307          */
5308         if (logflags)
5309                 xfs_trans_log_inode(tp, ip, logflags);
5310         if (cur) {
5311                 if (!error) {
5312                         *firstblock = cur->bc_private.b.firstblock;
5313                         cur->bc_private.b.allocated = 0;
5314                 }
5315                 xfs_btree_del_cursor(cur,
5316                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5317         }
5318         return error;
5319 }
5320
5321 /*
5322  * returns 1 for success, 0 if we failed to map the extent.
5323  */
5324 STATIC int
5325 xfs_getbmapx_fix_eof_hole(
5326         xfs_inode_t             *ip,            /* xfs incore inode pointer */
5327         struct getbmapx         *out,           /* output structure */
5328         int                     prealloced,     /* this is a file with
5329                                                  * preallocated data space */
5330         __int64_t               end,            /* last block requested */
5331         xfs_fsblock_t           startblock)
5332 {
5333         __int64_t               fixlen;
5334         xfs_mount_t             *mp;            /* file system mount point */
5335         xfs_ifork_t             *ifp;           /* inode fork pointer */
5336         xfs_extnum_t            lastx;          /* last extent pointer */
5337         xfs_fileoff_t           fileblock;
5338
5339         if (startblock == HOLESTARTBLOCK) {
5340                 mp = ip->i_mount;
5341                 out->bmv_block = -1;
5342                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, ip->i_size));
5343                 fixlen -= out->bmv_offset;
5344                 if (prealloced && out->bmv_offset + out->bmv_length == end) {
5345                         /* Came to hole at EOF. Trim it. */
5346                         if (fixlen <= 0)
5347                                 return 0;
5348                         out->bmv_length = fixlen;
5349                 }
5350         } else {
5351                 if (startblock == DELAYSTARTBLOCK)
5352                         out->bmv_block = -2;
5353                 else
5354                         out->bmv_block = xfs_fsb_to_db(ip, startblock);
5355                 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
5356                 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
5357                 if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
5358                    (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
5359                         out->bmv_oflags |= BMV_OF_LAST;
5360         }
5361
5362         return 1;
5363 }
5364
5365 /*
5366  * Get inode's extents as described in bmv, and format for output.
5367  * Calls formatter to fill the user's buffer until all extents
5368  * are mapped, until the passed-in bmv->bmv_count slots have
5369  * been filled, or until the formatter short-circuits the loop,
5370  * if it is tracking filled-in extents on its own.
5371  */
5372 int                                             /* error code */
5373 xfs_getbmap(
5374         xfs_inode_t             *ip,
5375         struct getbmapx         *bmv,           /* user bmap structure */
5376         xfs_bmap_format_t       formatter,      /* format to user */
5377         void                    *arg)           /* formatter arg */
5378 {
5379         __int64_t               bmvend;         /* last block requested */
5380         int                     error = 0;      /* return value */
5381         __int64_t               fixlen;         /* length for -1 case */
5382         int                     i;              /* extent number */
5383         int                     lock;           /* lock state */
5384         xfs_bmbt_irec_t         *map;           /* buffer for user's data */
5385         xfs_mount_t             *mp;            /* file system mount point */
5386         int                     nex;            /* # of user extents can do */
5387         int                     nexleft;        /* # of user extents left */
5388         int                     subnex;         /* # of bmapi's can do */
5389         int                     nmap;           /* number of map entries */
5390         struct getbmapx         *out;           /* output structure */
5391         int                     whichfork;      /* data or attr fork */
5392         int                     prealloced;     /* this is a file with
5393                                                  * preallocated data space */
5394         int                     iflags;         /* interface flags */
5395         int                     bmapi_flags;    /* flags for xfs_bmapi */
5396         int                     cur_ext = 0;
5397
5398         mp = ip->i_mount;
5399         iflags = bmv->bmv_iflags;
5400         whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5401
5402         if (whichfork == XFS_ATTR_FORK) {
5403                 if (XFS_IFORK_Q(ip)) {
5404                         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5405                             ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5406                             ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5407                                 return XFS_ERROR(EINVAL);
5408                 } else if (unlikely(
5409                            ip->i_d.di_aformat != 0 &&
5410                            ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5411                         XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5412                                          ip->i_mount);
5413                         return XFS_ERROR(EFSCORRUPTED);
5414                 }
5415
5416                 prealloced = 0;
5417                 fixlen = 1LL << 32;
5418         } else {
5419                 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5420                     ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5421                     ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5422                         return XFS_ERROR(EINVAL);
5423
5424                 if (xfs_get_extsz_hint(ip) ||
5425                     ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5426                         prealloced = 1;
5427                         fixlen = XFS_MAXIOFFSET(mp);
5428                 } else {
5429                         prealloced = 0;
5430                         fixlen = ip->i_size;
5431                 }
5432         }
5433
5434         if (bmv->bmv_length == -1) {
5435                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5436                 bmv->bmv_length =
5437                         max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
5438         } else if (bmv->bmv_length == 0) {
5439                 bmv->bmv_entries = 0;
5440                 return 0;
5441         } else if (bmv->bmv_length < 0) {
5442                 return XFS_ERROR(EINVAL);
5443         }
5444
5445         nex = bmv->bmv_count - 1;
5446         if (nex <= 0)
5447                 return XFS_ERROR(EINVAL);
5448         bmvend = bmv->bmv_offset + bmv->bmv_length;
5449
5450
5451         if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
5452                 return XFS_ERROR(ENOMEM);
5453         out = kmem_zalloc(bmv->bmv_count * sizeof(struct getbmapx), KM_MAYFAIL);
5454         if (!out)
5455                 return XFS_ERROR(ENOMEM);
5456
5457         xfs_ilock(ip, XFS_IOLOCK_SHARED);
5458         if (whichfork == XFS_DATA_FORK && !(iflags & BMV_IF_DELALLOC)) {
5459                 if (ip->i_delayed_blks || ip->i_size > ip->i_d.di_size) {
5460                         error = xfs_flush_pages(ip, 0, -1, 0, FI_REMAPF);
5461                         if (error)
5462                                 goto out_unlock_iolock;
5463                 }
5464                 /*
5465                  * even after flushing the inode, there can still be delalloc
5466                  * blocks on the inode beyond EOF due to speculative
5467                  * preallocation. These are not removed until the release
5468                  * function is called or the inode is inactivated. Hence we
5469                  * cannot assert here that ip->i_delayed_blks == 0.
5470                  */
5471         }
5472
5473         lock = xfs_ilock_map_shared(ip);
5474
5475         /*
5476          * Don't let nex be bigger than the number of extents
5477          * we can have assuming alternating holes and real extents.
5478          */
5479         if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5480                 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5481
5482         bmapi_flags = xfs_bmapi_aflag(whichfork);
5483         if (!(iflags & BMV_IF_PREALLOC))
5484                 bmapi_flags |= XFS_BMAPI_IGSTATE;
5485
5486         /*
5487          * Allocate enough space to handle "subnex" maps at a time.
5488          */
5489         error = ENOMEM;
5490         subnex = 16;
5491         map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
5492         if (!map)
5493                 goto out_unlock_ilock;
5494
5495         bmv->bmv_entries = 0;
5496
5497         if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
5498             (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
5499                 error = 0;
5500                 goto out_free_map;
5501         }
5502
5503         nexleft = nex;
5504
5505         do {
5506                 nmap = (nexleft > subnex) ? subnex : nexleft;
5507                 error = xfs_bmapi(NULL, ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5508                                   XFS_BB_TO_FSB(mp, bmv->bmv_length),
5509                                   bmapi_flags, NULL, 0, map, &nmap,
5510                                   NULL);
5511                 if (error)
5512                         goto out_free_map;
5513                 ASSERT(nmap <= subnex);
5514
5515                 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5516                         out[cur_ext].bmv_oflags = 0;
5517                         if (map[i].br_state == XFS_EXT_UNWRITTEN)
5518                                 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
5519                         else if (map[i].br_startblock == DELAYSTARTBLOCK)
5520                                 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
5521                         out[cur_ext].bmv_offset =
5522                                 XFS_FSB_TO_BB(mp, map[i].br_startoff);
5523                         out[cur_ext].bmv_length =
5524                                 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5525                         out[cur_ext].bmv_unused1 = 0;
5526                         out[cur_ext].bmv_unused2 = 0;
5527                         ASSERT(((iflags & BMV_IF_DELALLOC) != 0) ||
5528                               (map[i].br_startblock != DELAYSTARTBLOCK));
5529                         if (map[i].br_startblock == HOLESTARTBLOCK &&
5530                             whichfork == XFS_ATTR_FORK) {
5531                                 /* came to the end of attribute fork */
5532                                 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
5533                                 goto out_free_map;
5534                         }
5535
5536                         if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
5537                                         prealloced, bmvend,
5538                                         map[i].br_startblock))
5539                                 goto out_free_map;
5540
5541                         bmv->bmv_offset =
5542                                 out[cur_ext].bmv_offset +
5543                                 out[cur_ext].bmv_length;
5544                         bmv->bmv_length =
5545                                 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
5546
5547                         /*
5548                          * In case we don't want to return the hole,
5549                          * don't increase cur_ext so that we can reuse
5550                          * it in the next loop.
5551                          */
5552                         if ((iflags & BMV_IF_NO_HOLES) &&
5553                             map[i].br_startblock == HOLESTARTBLOCK) {
5554                                 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
5555                                 continue;
5556                         }
5557
5558                         nexleft--;
5559                         bmv->bmv_entries++;
5560                         cur_ext++;
5561                 }
5562         } while (nmap && nexleft && bmv->bmv_length);
5563
5564  out_free_map:
5565         kmem_free(map);
5566  out_unlock_ilock:
5567         xfs_iunlock_map_shared(ip, lock);
5568  out_unlock_iolock:
5569         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5570
5571         for (i = 0; i < cur_ext; i++) {
5572                 int full = 0;   /* user array is full */
5573
5574                 /* format results & advance arg */
5575                 error = formatter(&arg, &out[i], &full);
5576                 if (error || full)
5577                         break;
5578         }
5579
5580         kmem_free(out);
5581         return error;
5582 }
5583
5584 /*
5585  * Check the last inode extent to determine whether this allocation will result
5586  * in blocks being allocated at the end of the file. When we allocate new data
5587  * blocks at the end of the file which do not start at the previous data block,
5588  * we will try to align the new blocks at stripe unit boundaries.
5589  */
5590 STATIC int                              /* error */
5591 xfs_bmap_isaeof(
5592         xfs_inode_t     *ip,            /* incore inode pointer */
5593         xfs_fileoff_t   off,            /* file offset in fsblocks */
5594         int             whichfork,      /* data or attribute fork */
5595         char            *aeof)          /* return value */
5596 {
5597         int             error;          /* error return value */
5598         xfs_ifork_t     *ifp;           /* inode fork pointer */
5599         xfs_bmbt_rec_host_t *lastrec;   /* extent record pointer */
5600         xfs_extnum_t    nextents;       /* number of file extents */
5601         xfs_bmbt_irec_t s;              /* expanded extent record */
5602
5603         ASSERT(whichfork == XFS_DATA_FORK);
5604         ifp = XFS_IFORK_PTR(ip, whichfork);
5605         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5606             (error = xfs_iread_extents(NULL, ip, whichfork)))
5607                 return error;
5608         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5609         if (nextents == 0) {
5610                 *aeof = 1;
5611                 return 0;
5612         }
5613         /*
5614          * Go to the last extent
5615          */
5616         lastrec = xfs_iext_get_ext(ifp, nextents - 1);
5617         xfs_bmbt_get_all(lastrec, &s);
5618         /*
5619          * Check we are allocating in the last extent (for delayed allocations)
5620          * or past the last extent for non-delayed allocations.
5621          */
5622         *aeof = (off >= s.br_startoff &&
5623                  off < s.br_startoff + s.br_blockcount &&
5624                  isnullstartblock(s.br_startblock)) ||
5625                 off >= s.br_startoff + s.br_blockcount;
5626         return 0;
5627 }
5628
5629 /*
5630  * Check if the endoff is outside the last extent. If so the caller will grow
5631  * the allocation to a stripe unit boundary.
5632  */
5633 int                                     /* error */
5634 xfs_bmap_eof(
5635         xfs_inode_t     *ip,            /* incore inode pointer */
5636         xfs_fileoff_t   endoff,         /* file offset in fsblocks */
5637         int             whichfork,      /* data or attribute fork */
5638         int             *eof)           /* result value */
5639 {
5640         xfs_fsblock_t   blockcount;     /* extent block count */
5641         int             error;          /* error return value */
5642         xfs_ifork_t     *ifp;           /* inode fork pointer */
5643         xfs_bmbt_rec_host_t *lastrec;   /* extent record pointer */
5644         xfs_extnum_t    nextents;       /* number of file extents */
5645         xfs_fileoff_t   startoff;       /* extent starting file offset */
5646
5647         ASSERT(whichfork == XFS_DATA_FORK);
5648         ifp = XFS_IFORK_PTR(ip, whichfork);
5649         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5650             (error = xfs_iread_extents(NULL, ip, whichfork)))
5651                 return error;
5652         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5653         if (nextents == 0) {
5654                 *eof = 1;
5655                 return 0;
5656         }
5657         /*
5658          * Go to the last extent
5659          */
5660         lastrec = xfs_iext_get_ext(ifp, nextents - 1);
5661         startoff = xfs_bmbt_get_startoff(lastrec);
5662         blockcount = xfs_bmbt_get_blockcount(lastrec);
5663         *eof = endoff >= startoff + blockcount;
5664         return 0;
5665 }
5666
5667 #ifdef DEBUG
5668 STATIC struct xfs_buf *
5669 xfs_bmap_get_bp(
5670         struct xfs_btree_cur    *cur,
5671         xfs_fsblock_t           bno)
5672 {
5673         struct xfs_log_item_desc *lidp;
5674         int                     i;
5675
5676         if (!cur)
5677                 return NULL;
5678
5679         for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
5680                 if (!cur->bc_bufs[i])
5681                         break;
5682                 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
5683                         return cur->bc_bufs[i];
5684         }
5685
5686         /* Chase down all the log items to see if the bp is there */
5687         list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
5688                 struct xfs_buf_log_item *bip;
5689                 bip = (struct xfs_buf_log_item *)lidp->lid_item;
5690                 if (bip->bli_item.li_type == XFS_LI_BUF &&
5691                     XFS_BUF_ADDR(bip->bli_buf) == bno)
5692                         return bip->bli_buf;
5693         }
5694
5695         return NULL;
5696 }
5697
5698 STATIC void
5699 xfs_check_block(
5700         struct xfs_btree_block  *block,
5701         xfs_mount_t             *mp,
5702         int                     root,
5703         short                   sz)
5704 {
5705         int                     i, j, dmxr;
5706         __be64                  *pp, *thispa;   /* pointer to block address */
5707         xfs_bmbt_key_t          *prevp, *keyp;
5708
5709         ASSERT(be16_to_cpu(block->bb_level) > 0);
5710
5711         prevp = NULL;
5712         for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
5713                 dmxr = mp->m_bmap_dmxr[0];
5714                 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
5715
5716                 if (prevp) {
5717                         ASSERT(be64_to_cpu(prevp->br_startoff) <
5718                                be64_to_cpu(keyp->br_startoff));
5719                 }
5720                 prevp = keyp;
5721
5722                 /*
5723                  * Compare the block numbers to see if there are dups.
5724                  */
5725                 if (root)
5726                         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
5727                 else
5728                         pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
5729
5730                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
5731                         if (root)
5732                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
5733                         else
5734                                 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
5735                         if (*thispa == *pp) {
5736                                 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
5737                                         __func__, j, i,
5738                                         (unsigned long long)be64_to_cpu(*thispa));
5739                                 panic("%s: ptrs are equal in node\n",
5740                                         __func__);
5741                         }
5742                 }
5743         }
5744 }
5745
5746 /*
5747  * Check that the extents for the inode ip are in the right order in all
5748  * btree leaves.
5749  */
5750
5751 STATIC void
5752 xfs_bmap_check_leaf_extents(
5753         xfs_btree_cur_t         *cur,   /* btree cursor or null */
5754         xfs_inode_t             *ip,            /* incore inode pointer */
5755         int                     whichfork)      /* data or attr fork */
5756 {
5757         struct xfs_btree_block  *block; /* current btree block */
5758         xfs_fsblock_t           bno;    /* block # of "block" */
5759         xfs_buf_t               *bp;    /* buffer for "block" */
5760         int                     error;  /* error return value */
5761         xfs_extnum_t            i=0, j; /* index into the extents list */
5762         xfs_ifork_t             *ifp;   /* fork structure */
5763         int                     level;  /* btree level, for checking */
5764         xfs_mount_t             *mp;    /* file system mount structure */
5765         __be64                  *pp;    /* pointer to block address */
5766         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
5767         xfs_bmbt_rec_t          last = {0, 0}; /* last extent in prev block */
5768         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
5769         int                     bp_release = 0;
5770
5771         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
5772                 return;
5773         }
5774
5775         bno = NULLFSBLOCK;
5776         mp = ip->i_mount;
5777         ifp = XFS_IFORK_PTR(ip, whichfork);
5778         block = ifp->if_broot;
5779         /*
5780          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5781          */
5782         level = be16_to_cpu(block->bb_level);
5783         ASSERT(level > 0);
5784         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
5785         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5786         bno = be64_to_cpu(*pp);
5787
5788         ASSERT(bno != NULLDFSBNO);
5789         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5790         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5791
5792         /*
5793          * Go down the tree until leaf level is reached, following the first
5794          * pointer (leftmost) at each level.
5795          */
5796         while (level-- > 0) {
5797                 /* See if buf is in cur first */
5798                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5799                 if (bp) {
5800                         bp_release = 0;
5801                 } else {
5802                         bp_release = 1;
5803                 }
5804                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5805                                 XFS_BMAP_BTREE_REF)))
5806                         goto error_norelse;
5807                 block = XFS_BUF_TO_BLOCK(bp);
5808                 XFS_WANT_CORRUPTED_GOTO(
5809                         xfs_bmap_sanity_check(mp, bp, level),
5810                         error0);
5811                 if (level == 0)
5812                         break;
5813
5814                 /*
5815                  * Check this block for basic sanity (increasing keys and
5816                  * no duplicate blocks).
5817                  */
5818
5819                 xfs_check_block(block, mp, 0, 0);
5820                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
5821                 bno = be64_to_cpu(*pp);
5822                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
5823                 if (bp_release) {
5824                         bp_release = 0;
5825                         xfs_trans_brelse(NULL, bp);
5826                 }
5827         }
5828
5829         /*
5830          * Here with bp and block set to the leftmost leaf node in the tree.
5831          */
5832         i = 0;
5833
5834         /*
5835          * Loop over all leaf nodes checking that all extents are in the right order.
5836          */
5837         for (;;) {
5838                 xfs_fsblock_t   nextbno;
5839                 xfs_extnum_t    num_recs;
5840
5841
5842                 num_recs = xfs_btree_get_numrecs(block);
5843
5844                 /*
5845                  * Read-ahead the next leaf block, if any.
5846                  */
5847
5848                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
5849
5850                 /*
5851                  * Check all the extents to make sure they are OK.
5852                  * If we had a previous block, the last entry should
5853                  * conform with the first entry in this one.
5854                  */
5855
5856                 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
5857                 if (i) {
5858                         ASSERT(xfs_bmbt_disk_get_startoff(&last) +
5859                                xfs_bmbt_disk_get_blockcount(&last) <=
5860                                xfs_bmbt_disk_get_startoff(ep));
5861                 }
5862                 for (j = 1; j < num_recs; j++) {
5863                         nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
5864                         ASSERT(xfs_bmbt_disk_get_startoff(ep) +
5865                                xfs_bmbt_disk_get_blockcount(ep) <=
5866                                xfs_bmbt_disk_get_startoff(nextp));
5867                         ep = nextp;
5868                 }
5869
5870                 last = *ep;
5871                 i += num_recs;
5872                 if (bp_release) {
5873                         bp_release = 0;
5874                         xfs_trans_brelse(NULL, bp);
5875                 }
5876                 bno = nextbno;
5877                 /*
5878                  * If we've reached the end, stop.
5879                  */
5880                 if (bno == NULLFSBLOCK)
5881                         break;
5882
5883                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5884                 if (bp) {
5885                         bp_release = 0;
5886                 } else {
5887                         bp_release = 1;
5888                 }
5889                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5890                                 XFS_BMAP_BTREE_REF)))
5891                         goto error_norelse;
5892                 block = XFS_BUF_TO_BLOCK(bp);
5893         }
5894         if (bp_release) {
5895                 bp_release = 0;
5896                 xfs_trans_brelse(NULL, bp);
5897         }
5898         return;
5899
5900 error0:
5901         xfs_warn(mp, "%s: at error0", __func__);
5902         if (bp_release)
5903                 xfs_trans_brelse(NULL, bp);
5904 error_norelse:
5905         xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
5906                 __func__, i);
5907         panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
5908         return;
5909 }
5910 #endif
5911
5912 /*
5913  * Count fsblocks of the given fork.
5914  */
5915 int                                             /* error */
5916 xfs_bmap_count_blocks(
5917         xfs_trans_t             *tp,            /* transaction pointer */
5918         xfs_inode_t             *ip,            /* incore inode */
5919         int                     whichfork,      /* data or attr fork */
5920         int                     *count)         /* out: count of blocks */
5921 {
5922         struct xfs_btree_block  *block; /* current btree block */
5923         xfs_fsblock_t           bno;    /* block # of "block" */
5924         xfs_ifork_t             *ifp;   /* fork structure */
5925         int                     level;  /* btree level, for checking */
5926         xfs_mount_t             *mp;    /* file system mount structure */
5927         __be64                  *pp;    /* pointer to block address */
5928
5929         bno = NULLFSBLOCK;
5930         mp = ip->i_mount;
5931         ifp = XFS_IFORK_PTR(ip, whichfork);
5932         if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
5933                 xfs_bmap_count_leaves(ifp, 0,
5934                         ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
5935                         count);
5936                 return 0;
5937         }
5938
5939         /*
5940          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5941          */
5942         block = ifp->if_broot;
5943         level = be16_to_cpu(block->bb_level);
5944         ASSERT(level > 0);
5945         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5946         bno = be64_to_cpu(*pp);
5947         ASSERT(bno != NULLDFSBNO);
5948         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5949         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5950
5951         if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
5952                 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
5953                                  mp);
5954                 return XFS_ERROR(EFSCORRUPTED);
5955         }
5956
5957         return 0;
5958 }
5959
5960 /*
5961  * Recursively walks each level of a btree
5962  * to count total fsblocks is use.
5963  */
5964 STATIC int                                     /* error */
5965 xfs_bmap_count_tree(
5966         xfs_mount_t     *mp,            /* file system mount point */
5967         xfs_trans_t     *tp,            /* transaction pointer */
5968         xfs_ifork_t     *ifp,           /* inode fork pointer */
5969         xfs_fsblock_t   blockno,        /* file system block number */
5970         int             levelin,        /* level in btree */
5971         int             *count)         /* Count of blocks */
5972 {
5973         int                     error;
5974         xfs_buf_t               *bp, *nbp;
5975         int                     level = levelin;
5976         __be64                  *pp;
5977         xfs_fsblock_t           bno = blockno;
5978         xfs_fsblock_t           nextbno;
5979         struct xfs_btree_block  *block, *nextblock;
5980         int                     numrecs;
5981
5982         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
5983                 return error;
5984         *count += 1;
5985         block = XFS_BUF_TO_BLOCK(bp);
5986
5987         if (--level) {
5988                 /* Not at node above leaves, count this level of nodes */
5989                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
5990                 while (nextbno != NULLFSBLOCK) {
5991                         if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
5992                                 0, &nbp, XFS_BMAP_BTREE_REF)))
5993                                 return error;
5994                         *count += 1;
5995                         nextblock = XFS_BUF_TO_BLOCK(nbp);
5996                         nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
5997                         xfs_trans_brelse(tp, nbp);
5998                 }
5999
6000                 /* Dive to the next level */
6001                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
6002                 bno = be64_to_cpu(*pp);
6003                 if (unlikely((error =
6004                      xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
6005                         xfs_trans_brelse(tp, bp);
6006                         XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6007                                          XFS_ERRLEVEL_LOW, mp);
6008                         return XFS_ERROR(EFSCORRUPTED);
6009                 }
6010                 xfs_trans_brelse(tp, bp);
6011         } else {
6012                 /* count all level 1 nodes and their leaves */
6013                 for (;;) {
6014                         nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
6015                         numrecs = be16_to_cpu(block->bb_numrecs);
6016                         xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
6017                         xfs_trans_brelse(tp, bp);
6018                         if (nextbno == NULLFSBLOCK)
6019                                 break;
6020                         bno = nextbno;
6021                         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6022                                 XFS_BMAP_BTREE_REF)))
6023                                 return error;
6024                         *count += 1;
6025                         block = XFS_BUF_TO_BLOCK(bp);
6026                 }
6027         }
6028         return 0;
6029 }
6030
6031 /*
6032  * Count leaf blocks given a range of extent records.
6033  */
6034 STATIC void
6035 xfs_bmap_count_leaves(
6036         xfs_ifork_t             *ifp,
6037         xfs_extnum_t            idx,
6038         int                     numrecs,
6039         int                     *count)
6040 {
6041         int             b;
6042
6043         for (b = 0; b < numrecs; b++) {
6044                 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
6045                 *count += xfs_bmbt_get_blockcount(frp);
6046         }
6047 }
6048
6049 /*
6050  * Count leaf blocks given a range of extent records originally
6051  * in btree format.
6052  */
6053 STATIC void
6054 xfs_bmap_disk_count_leaves(
6055         struct xfs_mount        *mp,
6056         struct xfs_btree_block  *block,
6057         int                     numrecs,
6058         int                     *count)
6059 {
6060         int             b;
6061         xfs_bmbt_rec_t  *frp;
6062
6063         for (b = 1; b <= numrecs; b++) {
6064                 frp = XFS_BMBT_REC_ADDR(mp, block, b);
6065                 *count += xfs_bmbt_disk_get_blockcount(frp);
6066         }
6067 }
6068
6069 /*
6070  * dead simple method of punching delalyed allocation blocks from a range in
6071  * the inode. Walks a block at a time so will be slow, but is only executed in
6072  * rare error cases so the overhead is not critical. This will alays punch out
6073  * both the start and end blocks, even if the ranges only partially overlap
6074  * them, so it is up to the caller to ensure that partial blocks are not
6075  * passed in.
6076  */
6077 int
6078 xfs_bmap_punch_delalloc_range(
6079         struct xfs_inode        *ip,
6080         xfs_fileoff_t           start_fsb,
6081         xfs_fileoff_t           length)
6082 {
6083         xfs_fileoff_t           remaining = length;
6084         int                     error = 0;
6085
6086         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
6087
6088         do {
6089                 int             done;
6090                 xfs_bmbt_irec_t imap;
6091                 int             nimaps = 1;
6092                 xfs_fsblock_t   firstblock;
6093                 xfs_bmap_free_t flist;
6094
6095                 /*
6096                  * Map the range first and check that it is a delalloc extent
6097                  * before trying to unmap the range. Otherwise we will be
6098                  * trying to remove a real extent (which requires a
6099                  * transaction) or a hole, which is probably a bad idea...
6100                  */
6101                 error = xfs_bmapi(NULL, ip, start_fsb, 1,
6102                                 XFS_BMAPI_ENTIRE,  NULL, 0, &imap,
6103                                 &nimaps, NULL);
6104
6105                 if (error) {
6106                         /* something screwed, just bail */
6107                         if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
6108                                 xfs_alert(ip->i_mount,
6109                         "Failed delalloc mapping lookup ino %lld fsb %lld.",
6110                                                 ip->i_ino, start_fsb);
6111                         }
6112                         break;
6113                 }
6114                 if (!nimaps) {
6115                         /* nothing there */
6116                         goto next_block;
6117                 }
6118                 if (imap.br_startblock != DELAYSTARTBLOCK) {
6119                         /* been converted, ignore */
6120                         goto next_block;
6121                 }
6122                 WARN_ON(imap.br_blockcount == 0);
6123
6124                 /*
6125                  * Note: while we initialise the firstblock/flist pair, they
6126                  * should never be used because blocks should never be
6127                  * allocated or freed for a delalloc extent and hence we need
6128                  * don't cancel or finish them after the xfs_bunmapi() call.
6129                  */
6130                 xfs_bmap_init(&flist, &firstblock);
6131                 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
6132                                         &flist, &done);
6133                 if (error)
6134                         break;
6135
6136                 ASSERT(!flist.xbf_count && !flist.xbf_first);
6137 next_block:
6138                 start_fsb++;
6139                 remaining--;
6140         } while(remaining > 0);
6141
6142         return error;
6143 }