Merge branch 'for-linus' of master.kernel.org:/pub/scm/linux/kernel/git/roland/infiniband
[linux-2.6] / fs / xfs / xfs_bmap.c
1 /*
2  * Copyright (c) 2000-2005 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_dir.h"
28 #include "xfs_dir2.h"
29 #include "xfs_da_btree.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_alloc_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_dir_sf.h"
34 #include "xfs_dir2_sf.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
38 #include "xfs_btree.h"
39 #include "xfs_dmapi.h"
40 #include "xfs_mount.h"
41 #include "xfs_ialloc.h"
42 #include "xfs_itable.h"
43 #include "xfs_inode_item.h"
44 #include "xfs_extfree_item.h"
45 #include "xfs_alloc.h"
46 #include "xfs_bmap.h"
47 #include "xfs_rtalloc.h"
48 #include "xfs_error.h"
49 #include "xfs_dir_leaf.h"
50 #include "xfs_attr_leaf.h"
51 #include "xfs_rw.h"
52 #include "xfs_quota.h"
53 #include "xfs_trans_space.h"
54 #include "xfs_buf_item.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_bmapi to update file extent records and the btree
93  * after allocating space (or doing a delayed allocation).
94  */
95 STATIC int                              /* error */
96 xfs_bmap_add_extent(
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_fsblock_t           *first, /* pointer to firstblock variable */
102         xfs_bmap_free_t         *flist, /* list of extents to be freed */
103         int                     *logflagsp, /* inode logging flags */
104         int                     whichfork, /* data or attr fork */
105         int                     rsvd);  /* OK to allocate reserved blocks */
106
107 /*
108  * Called by xfs_bmap_add_extent to handle cases converting a delayed
109  * allocation to a real allocation.
110  */
111 STATIC int                              /* error */
112 xfs_bmap_add_extent_delay_real(
113         xfs_inode_t             *ip,    /* incore inode pointer */
114         xfs_extnum_t            idx,    /* extent number to update/insert */
115         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
116         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
117         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
118         xfs_fsblock_t           *first, /* pointer to firstblock variable */
119         xfs_bmap_free_t         *flist, /* list of extents to be freed */
120         int                     *logflagsp, /* inode logging flags */
121         int                     rsvd);  /* OK to allocate reserved blocks */
122
123 /*
124  * Called by xfs_bmap_add_extent to handle cases converting a hole
125  * to a delayed allocation.
126  */
127 STATIC int                              /* error */
128 xfs_bmap_add_extent_hole_delay(
129         xfs_inode_t             *ip,    /* incore inode pointer */
130         xfs_extnum_t            idx,    /* extent number to update/insert */
131         xfs_btree_cur_t         *cur,   /* if null, not a btree */
132         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
133         int                     *logflagsp,/* inode logging flags */
134         int                     rsvd);  /* OK to allocate reserved blocks */
135
136 /*
137  * Called by xfs_bmap_add_extent to handle cases converting a hole
138  * to a real allocation.
139  */
140 STATIC int                              /* error */
141 xfs_bmap_add_extent_hole_real(
142         xfs_inode_t             *ip,    /* incore inode pointer */
143         xfs_extnum_t            idx,    /* extent number to update/insert */
144         xfs_btree_cur_t         *cur,   /* if null, not a btree */
145         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
146         int                     *logflagsp, /* inode logging flags */
147         int                     whichfork); /* data or attr fork */
148
149 /*
150  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
151  * allocation to a real allocation or vice versa.
152  */
153 STATIC int                              /* error */
154 xfs_bmap_add_extent_unwritten_real(
155         xfs_inode_t             *ip,    /* incore inode pointer */
156         xfs_extnum_t            idx,    /* extent number to update/insert */
157         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
158         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
159         int                     *logflagsp); /* inode logging flags */
160
161 /*
162  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
163  * It figures out where to ask the underlying allocator to put the new extent.
164  */
165 STATIC int                              /* error */
166 xfs_bmap_alloc(
167         xfs_bmalloca_t          *ap);   /* bmap alloc argument struct */
168
169 /*
170  * Transform a btree format file with only one leaf node, where the
171  * extents list will fit in the inode, into an extents format file.
172  * Since the file extents are already in-core, all we have to do is
173  * give up the space for the btree root and pitch the leaf block.
174  */
175 STATIC int                              /* error */
176 xfs_bmap_btree_to_extents(
177         xfs_trans_t             *tp,    /* transaction pointer */
178         xfs_inode_t             *ip,    /* incore inode pointer */
179         xfs_btree_cur_t         *cur,   /* btree cursor */
180         int                     *logflagsp, /* inode logging flags */
181         int                     whichfork); /* data or attr fork */
182
183 #ifdef DEBUG
184 /*
185  * Check that the extents list for the inode ip is in the right order.
186  */
187 STATIC void
188 xfs_bmap_check_extents(
189         xfs_inode_t             *ip,            /* incore inode pointer */
190         int                     whichfork);     /* data or attr fork */
191 #endif
192
193 /*
194  * Called by xfs_bmapi to update file extent records and the btree
195  * after removing space (or undoing a delayed allocation).
196  */
197 STATIC int                              /* error */
198 xfs_bmap_del_extent(
199         xfs_inode_t             *ip,    /* incore inode pointer */
200         xfs_trans_t             *tp,    /* current trans pointer */
201         xfs_extnum_t            idx,    /* extent number to update/insert */
202         xfs_bmap_free_t         *flist, /* list of extents to be freed */
203         xfs_btree_cur_t         *cur,   /* if null, not a btree */
204         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
205         int                     *logflagsp,/* inode logging flags */
206         int                     whichfork, /* data or attr fork */
207         int                     rsvd);   /* OK to allocate reserved blocks */
208
209 /*
210  * Remove the entry "free" from the free item list.  Prev points to the
211  * previous entry, unless "free" is the head of the list.
212  */
213 STATIC void
214 xfs_bmap_del_free(
215         xfs_bmap_free_t         *flist, /* free item list header */
216         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
217         xfs_bmap_free_item_t    *free); /* list item to be freed */
218
219 /*
220  * Convert an extents-format file into a btree-format file.
221  * The new file will have a root block (in the inode) and a single child block.
222  */
223 STATIC int                                      /* error */
224 xfs_bmap_extents_to_btree(
225         xfs_trans_t             *tp,            /* transaction pointer */
226         xfs_inode_t             *ip,            /* incore inode pointer */
227         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
228         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
229         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
230         int                     wasdel,         /* converting a delayed alloc */
231         int                     *logflagsp,     /* inode logging flags */
232         int                     whichfork);     /* data or attr fork */
233
234 /*
235  * Convert a local file to an extents file.
236  * This code is sort of bogus, since the file data needs to get
237  * logged so it won't be lost.  The bmap-level manipulations are ok, though.
238  */
239 STATIC int                              /* error */
240 xfs_bmap_local_to_extents(
241         xfs_trans_t     *tp,            /* transaction pointer */
242         xfs_inode_t     *ip,            /* incore inode pointer */
243         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
244         xfs_extlen_t    total,          /* total blocks needed by transaction */
245         int             *logflagsp,     /* inode logging flags */
246         int             whichfork);     /* data or attr fork */
247
248 /*
249  * Search the extents list for the inode, for the extent containing bno.
250  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
251  * *eofp will be set, and *prevp will contain the last entry (null if none).
252  * Else, *lastxp will be set to the index of the found
253  * entry; *gotp will contain the entry.
254  */
255 STATIC xfs_bmbt_rec_t *                 /* pointer to found extent entry */
256 xfs_bmap_search_extents(
257         xfs_inode_t     *ip,            /* incore inode pointer */
258         xfs_fileoff_t   bno,            /* block number searched for */
259         int             whichfork,      /* data or attr fork */
260         int             *eofp,          /* out: end of file found */
261         xfs_extnum_t    *lastxp,        /* out: last extent index */
262         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
263         xfs_bmbt_irec_t *prevp);        /* out: previous extent entry found */
264
265 /*
266  * Check the last inode extent to determine whether this allocation will result
267  * in blocks being allocated at the end of the file. When we allocate new data
268  * blocks at the end of the file which do not start at the previous data block,
269  * we will try to align the new blocks at stripe unit boundaries.
270  */
271 STATIC int                              /* error */
272 xfs_bmap_isaeof(
273         xfs_inode_t     *ip,            /* incore inode pointer */
274         xfs_fileoff_t   off,            /* file offset in fsblocks */
275         int             whichfork,      /* data or attribute fork */
276         char            *aeof);         /* return value */
277
278 #ifdef XFS_BMAP_TRACE
279 /*
280  * Add a bmap trace buffer entry.  Base routine for the others.
281  */
282 STATIC void
283 xfs_bmap_trace_addentry(
284         int             opcode,         /* operation */
285         char            *fname,         /* function name */
286         char            *desc,          /* operation description */
287         xfs_inode_t     *ip,            /* incore inode pointer */
288         xfs_extnum_t    idx,            /* index of entry(ies) */
289         xfs_extnum_t    cnt,            /* count of entries, 1 or 2 */
290         xfs_bmbt_rec_t  *r1,            /* first record */
291         xfs_bmbt_rec_t  *r2,            /* second record or null */
292         int             whichfork);     /* data or attr fork */
293
294 /*
295  * Add bmap trace entry prior to a call to xfs_iext_remove.
296  */
297 STATIC void
298 xfs_bmap_trace_delete(
299         char            *fname,         /* function name */
300         char            *desc,          /* operation description */
301         xfs_inode_t     *ip,            /* incore inode pointer */
302         xfs_extnum_t    idx,            /* index of entry(entries) deleted */
303         xfs_extnum_t    cnt,            /* count of entries deleted, 1 or 2 */
304         int             whichfork);     /* data or attr fork */
305
306 /*
307  * Add bmap trace entry prior to a call to xfs_iext_insert, or
308  * reading in the extents list from the disk (in the btree).
309  */
310 STATIC void
311 xfs_bmap_trace_insert(
312         char            *fname,         /* function name */
313         char            *desc,          /* operation description */
314         xfs_inode_t     *ip,            /* incore inode pointer */
315         xfs_extnum_t    idx,            /* index of entry(entries) inserted */
316         xfs_extnum_t    cnt,            /* count of entries inserted, 1 or 2 */
317         xfs_bmbt_irec_t *r1,            /* inserted record 1 */
318         xfs_bmbt_irec_t *r2,            /* inserted record 2 or null */
319         int             whichfork);     /* data or attr fork */
320
321 /*
322  * Add bmap trace entry after updating an extent record in place.
323  */
324 STATIC void
325 xfs_bmap_trace_post_update(
326         char            *fname,         /* function name */
327         char            *desc,          /* operation description */
328         xfs_inode_t     *ip,            /* incore inode pointer */
329         xfs_extnum_t    idx,            /* index of entry updated */
330         int             whichfork);     /* data or attr fork */
331
332 /*
333  * Add bmap trace entry prior to updating an extent record in place.
334  */
335 STATIC void
336 xfs_bmap_trace_pre_update(
337         char            *fname,         /* function name */
338         char            *desc,          /* operation description */
339         xfs_inode_t     *ip,            /* incore inode pointer */
340         xfs_extnum_t    idx,            /* index of entry to be updated */
341         int             whichfork);     /* data or attr fork */
342
343 #else
344 #define xfs_bmap_trace_delete(f,d,ip,i,c,w)
345 #define xfs_bmap_trace_insert(f,d,ip,i,c,r1,r2,w)
346 #define xfs_bmap_trace_post_update(f,d,ip,i,w)
347 #define xfs_bmap_trace_pre_update(f,d,ip,i,w)
348 #endif  /* XFS_BMAP_TRACE */
349
350 /*
351  * Compute the worst-case number of indirect blocks that will be used
352  * for ip's delayed extent of length "len".
353  */
354 STATIC xfs_filblks_t
355 xfs_bmap_worst_indlen(
356         xfs_inode_t             *ip,    /* incore inode pointer */
357         xfs_filblks_t           len);   /* delayed extent length */
358
359 #ifdef DEBUG
360 /*
361  * Perform various validation checks on the values being returned
362  * from xfs_bmapi().
363  */
364 STATIC void
365 xfs_bmap_validate_ret(
366         xfs_fileoff_t           bno,
367         xfs_filblks_t           len,
368         int                     flags,
369         xfs_bmbt_irec_t         *mval,
370         int                     nmap,
371         int                     ret_nmap);
372 #else
373 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
374 #endif /* DEBUG */
375
376 #if defined(XFS_RW_TRACE)
377 STATIC void
378 xfs_bunmap_trace(
379         xfs_inode_t             *ip,
380         xfs_fileoff_t           bno,
381         xfs_filblks_t           len,
382         int                     flags,
383         inst_t                  *ra);
384 #else
385 #define xfs_bunmap_trace(ip, bno, len, flags, ra)
386 #endif  /* XFS_RW_TRACE */
387
388 STATIC int
389 xfs_bmap_count_tree(
390         xfs_mount_t     *mp,
391         xfs_trans_t     *tp,
392         xfs_ifork_t     *ifp,
393         xfs_fsblock_t   blockno,
394         int             levelin,
395         int             *count);
396
397 STATIC int
398 xfs_bmap_count_leaves(
399         xfs_ifork_t             *ifp,
400         xfs_extnum_t            idx,
401         int                     numrecs,
402         int                     *count);
403
404 STATIC int
405 xfs_bmap_disk_count_leaves(
406         xfs_ifork_t             *ifp,
407         xfs_mount_t             *mp,
408         xfs_extnum_t            idx,
409         xfs_bmbt_block_t        *block,
410         int                     numrecs,
411         int                     *count);
412
413 /*
414  * Bmap internal routines.
415  */
416
417 /*
418  * Called from xfs_bmap_add_attrfork to handle btree format files.
419  */
420 STATIC int                                      /* error */
421 xfs_bmap_add_attrfork_btree(
422         xfs_trans_t             *tp,            /* transaction pointer */
423         xfs_inode_t             *ip,            /* incore inode pointer */
424         xfs_fsblock_t           *firstblock,    /* first block allocated */
425         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
426         int                     *flags)         /* inode logging flags */
427 {
428         xfs_btree_cur_t         *cur;           /* btree cursor */
429         int                     error;          /* error return value */
430         xfs_mount_t             *mp;            /* file system mount struct */
431         int                     stat;           /* newroot status */
432
433         mp = ip->i_mount;
434         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
435                 *flags |= XFS_ILOG_DBROOT;
436         else {
437                 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
438                         XFS_DATA_FORK);
439                 cur->bc_private.b.flist = flist;
440                 cur->bc_private.b.firstblock = *firstblock;
441                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
442                         goto error0;
443                 ASSERT(stat == 1);      /* must be at least one entry */
444                 if ((error = xfs_bmbt_newroot(cur, flags, &stat)))
445                         goto error0;
446                 if (stat == 0) {
447                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
448                         return XFS_ERROR(ENOSPC);
449                 }
450                 *firstblock = cur->bc_private.b.firstblock;
451                 cur->bc_private.b.allocated = 0;
452                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
453         }
454         return 0;
455 error0:
456         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
457         return error;
458 }
459
460 /*
461  * Called from xfs_bmap_add_attrfork to handle extents format files.
462  */
463 STATIC int                                      /* error */
464 xfs_bmap_add_attrfork_extents(
465         xfs_trans_t             *tp,            /* transaction pointer */
466         xfs_inode_t             *ip,            /* incore inode pointer */
467         xfs_fsblock_t           *firstblock,    /* first block allocated */
468         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
469         int                     *flags)         /* inode logging flags */
470 {
471         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
472         int                     error;          /* error return value */
473
474         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
475                 return 0;
476         cur = NULL;
477         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
478                 flags, XFS_DATA_FORK);
479         if (cur) {
480                 cur->bc_private.b.allocated = 0;
481                 xfs_btree_del_cursor(cur,
482                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
483         }
484         return error;
485 }
486
487 /*
488  * Called from xfs_bmap_add_attrfork to handle local format files.
489  */
490 STATIC int                                      /* error */
491 xfs_bmap_add_attrfork_local(
492         xfs_trans_t             *tp,            /* transaction pointer */
493         xfs_inode_t             *ip,            /* incore inode pointer */
494         xfs_fsblock_t           *firstblock,    /* first block allocated */
495         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
496         int                     *flags)         /* inode logging flags */
497 {
498         xfs_da_args_t           dargs;          /* args for dir/attr code */
499         int                     error;          /* error return value */
500         xfs_mount_t             *mp;            /* mount structure pointer */
501
502         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
503                 return 0;
504         if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
505                 mp = ip->i_mount;
506                 memset(&dargs, 0, sizeof(dargs));
507                 dargs.dp = ip;
508                 dargs.firstblock = firstblock;
509                 dargs.flist = flist;
510                 dargs.total = mp->m_dirblkfsbs;
511                 dargs.whichfork = XFS_DATA_FORK;
512                 dargs.trans = tp;
513                 error = XFS_DIR_SHORTFORM_TO_SINGLE(mp, &dargs);
514         } else
515                 error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
516                         XFS_DATA_FORK);
517         return error;
518 }
519
520 /*
521  * Called by xfs_bmapi to update file extent records and the btree
522  * after allocating space (or doing a delayed allocation).
523  */
524 STATIC int                              /* error */
525 xfs_bmap_add_extent(
526         xfs_inode_t             *ip,    /* incore inode pointer */
527         xfs_extnum_t            idx,    /* extent number to update/insert */
528         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
529         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
530         xfs_fsblock_t           *first, /* pointer to firstblock variable */
531         xfs_bmap_free_t         *flist, /* list of extents to be freed */
532         int                     *logflagsp, /* inode logging flags */
533         int                     whichfork, /* data or attr fork */
534         int                     rsvd)   /* OK to use reserved data blocks */
535 {
536         xfs_btree_cur_t         *cur;   /* btree cursor or null */
537         xfs_filblks_t           da_new; /* new count del alloc blocks used */
538         xfs_filblks_t           da_old; /* old count del alloc blocks used */
539         int                     error;  /* error return value */
540 #ifdef XFS_BMAP_TRACE
541         static char             fname[] = "xfs_bmap_add_extent";
542 #endif
543         xfs_ifork_t             *ifp;   /* inode fork ptr */
544         int                     logflags; /* returned value */
545         xfs_extnum_t            nextents; /* number of extents in file now */
546
547         XFS_STATS_INC(xs_add_exlist);
548         cur = *curp;
549         ifp = XFS_IFORK_PTR(ip, whichfork);
550         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
551         ASSERT(idx <= nextents);
552         da_old = da_new = 0;
553         error = 0;
554         /*
555          * This is the first extent added to a new/empty file.
556          * Special case this one, so other routines get to assume there are
557          * already extents in the list.
558          */
559         if (nextents == 0) {
560                 xfs_bmap_trace_insert(fname, "insert empty", ip, 0, 1, new,
561                         NULL, whichfork);
562                 xfs_iext_insert(ifp, 0, 1, new);
563                 ASSERT(cur == NULL);
564                 ifp->if_lastex = 0;
565                 if (!ISNULLSTARTBLOCK(new->br_startblock)) {
566                         XFS_IFORK_NEXT_SET(ip, whichfork, 1);
567                         logflags = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
568                 } else
569                         logflags = 0;
570         }
571         /*
572          * Any kind of new delayed allocation goes here.
573          */
574         else if (ISNULLSTARTBLOCK(new->br_startblock)) {
575                 if (cur)
576                         ASSERT((cur->bc_private.b.flags &
577                                 XFS_BTCUR_BPRV_WASDEL) == 0);
578                 if ((error = xfs_bmap_add_extent_hole_delay(ip, idx, cur, new,
579                                 &logflags, rsvd)))
580                         goto done;
581         }
582         /*
583          * Real allocation off the end of the file.
584          */
585         else if (idx == nextents) {
586                 if (cur)
587                         ASSERT((cur->bc_private.b.flags &
588                                 XFS_BTCUR_BPRV_WASDEL) == 0);
589                 if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur, new,
590                                 &logflags, whichfork)))
591                         goto done;
592         } else {
593                 xfs_bmbt_irec_t prev;   /* old extent at offset idx */
594
595                 /*
596                  * Get the record referred to by idx.
597                  */
598                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx), &prev);
599                 /*
600                  * If it's a real allocation record, and the new allocation ends
601                  * after the start of the referred to record, then we're filling
602                  * in a delayed or unwritten allocation with a real one, or
603                  * converting real back to unwritten.
604                  */
605                 if (!ISNULLSTARTBLOCK(new->br_startblock) &&
606                     new->br_startoff + new->br_blockcount > prev.br_startoff) {
607                         if (prev.br_state != XFS_EXT_UNWRITTEN &&
608                             ISNULLSTARTBLOCK(prev.br_startblock)) {
609                                 da_old = STARTBLOCKVAL(prev.br_startblock);
610                                 if (cur)
611                                         ASSERT(cur->bc_private.b.flags &
612                                                 XFS_BTCUR_BPRV_WASDEL);
613                                 if ((error = xfs_bmap_add_extent_delay_real(ip,
614                                         idx, &cur, new, &da_new, first, flist,
615                                         &logflags, rsvd)))
616                                         goto done;
617                         } else if (new->br_state == XFS_EXT_NORM) {
618                                 ASSERT(new->br_state == XFS_EXT_NORM);
619                                 if ((error = xfs_bmap_add_extent_unwritten_real(
620                                         ip, idx, &cur, new, &logflags)))
621                                         goto done;
622                         } else {
623                                 ASSERT(new->br_state == XFS_EXT_UNWRITTEN);
624                                 if ((error = xfs_bmap_add_extent_unwritten_real(
625                                         ip, idx, &cur, new, &logflags)))
626                                         goto done;
627                         }
628                         ASSERT(*curp == cur || *curp == NULL);
629                 }
630                 /*
631                  * Otherwise we're filling in a hole with an allocation.
632                  */
633                 else {
634                         if (cur)
635                                 ASSERT((cur->bc_private.b.flags &
636                                         XFS_BTCUR_BPRV_WASDEL) == 0);
637                         if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur,
638                                         new, &logflags, whichfork)))
639                                 goto done;
640                 }
641         }
642
643         ASSERT(*curp == cur || *curp == NULL);
644         /*
645          * Convert to a btree if necessary.
646          */
647         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
648             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
649                 int     tmp_logflags;   /* partial log flag return val */
650
651                 ASSERT(cur == NULL);
652                 error = xfs_bmap_extents_to_btree(ip->i_transp, ip, first,
653                         flist, &cur, da_old > 0, &tmp_logflags, whichfork);
654                 logflags |= tmp_logflags;
655                 if (error)
656                         goto done;
657         }
658         /*
659          * Adjust for changes in reserved delayed indirect blocks.
660          * Nothing to do for disk quotas here.
661          */
662         if (da_old || da_new) {
663                 xfs_filblks_t   nblks;
664
665                 nblks = da_new;
666                 if (cur)
667                         nblks += cur->bc_private.b.allocated;
668                 ASSERT(nblks <= da_old);
669                 if (nblks < da_old)
670                         xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
671                                 (int)(da_old - nblks), rsvd);
672         }
673         /*
674          * Clear out the allocated field, done with it now in any case.
675          */
676         if (cur) {
677                 cur->bc_private.b.allocated = 0;
678                 *curp = cur;
679         }
680 done:
681 #ifdef DEBUG
682         if (!error)
683                 xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
684 #endif
685         *logflagsp = logflags;
686         return error;
687 }
688
689 /*
690  * Called by xfs_bmap_add_extent to handle cases converting a delayed
691  * allocation to a real allocation.
692  */
693 STATIC int                              /* error */
694 xfs_bmap_add_extent_delay_real(
695         xfs_inode_t             *ip,    /* incore inode pointer */
696         xfs_extnum_t            idx,    /* extent number to update/insert */
697         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
698         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
699         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
700         xfs_fsblock_t           *first, /* pointer to firstblock variable */
701         xfs_bmap_free_t         *flist, /* list of extents to be freed */
702         int                     *logflagsp, /* inode logging flags */
703         int                     rsvd)   /* OK to use reserved data block allocation */
704 {
705         xfs_btree_cur_t         *cur;   /* btree cursor */
706         int                     diff;   /* temp value */
707         xfs_bmbt_rec_t          *ep;    /* extent entry for idx */
708         int                     error;  /* error return value */
709 #ifdef XFS_BMAP_TRACE
710         static char             fname[] = "xfs_bmap_add_extent_delay_real";
711 #endif
712         int                     i;      /* temp state */
713         xfs_ifork_t             *ifp;   /* inode fork pointer */
714         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
715         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
716                                         /* left is 0, right is 1, prev is 2 */
717         int                     rval=0; /* return value (logging flags) */
718         int                     state = 0;/* state bits, accessed thru macros */
719         xfs_filblks_t           temp;   /* value for dnew calculations */
720         xfs_filblks_t           temp2;  /* value for dnew calculations */
721         int                     tmp_rval;       /* partial logging flags */
722         enum {                          /* bit number definitions for state */
723                 LEFT_CONTIG,    RIGHT_CONTIG,
724                 LEFT_FILLING,   RIGHT_FILLING,
725                 LEFT_DELAY,     RIGHT_DELAY,
726                 LEFT_VALID,     RIGHT_VALID
727         };
728
729 #define LEFT            r[0]
730 #define RIGHT           r[1]
731 #define PREV            r[2]
732 #define MASK(b)         (1 << (b))
733 #define MASK2(a,b)      (MASK(a) | MASK(b))
734 #define MASK3(a,b,c)    (MASK2(a,b) | MASK(c))
735 #define MASK4(a,b,c,d)  (MASK3(a,b,c) | MASK(d))
736 #define STATE_SET(b,v)  ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
737 #define STATE_TEST(b)   (state & MASK(b))
738 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
739                                        ((state &= ~MASK(b)), 0))
740 #define SWITCH_STATE            \
741         (state & MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG))
742
743         /*
744          * Set up a bunch of variables to make the tests simpler.
745          */
746         cur = *curp;
747         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
748         ep = xfs_iext_get_ext(ifp, idx);
749         xfs_bmbt_get_all(ep, &PREV);
750         new_endoff = new->br_startoff + new->br_blockcount;
751         ASSERT(PREV.br_startoff <= new->br_startoff);
752         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
753         /*
754          * Set flags determining what part of the previous delayed allocation
755          * extent is being replaced by a real allocation.
756          */
757         STATE_SET(LEFT_FILLING, PREV.br_startoff == new->br_startoff);
758         STATE_SET(RIGHT_FILLING,
759                 PREV.br_startoff + PREV.br_blockcount == new_endoff);
760         /*
761          * Check and set flags if this segment has a left neighbor.
762          * Don't set contiguous if the combined extent would be too large.
763          */
764         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
765                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &LEFT);
766                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(LEFT.br_startblock));
767         }
768         STATE_SET(LEFT_CONTIG,
769                 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
770                 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
771                 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
772                 LEFT.br_state == new->br_state &&
773                 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN);
774         /*
775          * Check and set flags if this segment has a right neighbor.
776          * Don't set contiguous if the combined extent would be too large.
777          * Also check for all-three-contiguous being too large.
778          */
779         if (STATE_SET_TEST(RIGHT_VALID,
780                         idx <
781                         ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1)) {
782                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx + 1), &RIGHT);
783                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(RIGHT.br_startblock));
784         }
785         STATE_SET(RIGHT_CONTIG,
786                 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
787                 new_endoff == RIGHT.br_startoff &&
788                 new->br_startblock + new->br_blockcount ==
789                     RIGHT.br_startblock &&
790                 new->br_state == RIGHT.br_state &&
791                 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
792                 ((state & MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING)) !=
793                   MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING) ||
794                  LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
795                      <= MAXEXTLEN));
796         error = 0;
797         /*
798          * Switch out based on the FILLING and CONTIG state bits.
799          */
800         switch (SWITCH_STATE) {
801
802         case MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
803                 /*
804                  * Filling in all of a previously delayed allocation extent.
805                  * The left and right neighbors are both contiguous with new.
806                  */
807                 xfs_bmap_trace_pre_update(fname, "LF|RF|LC|RC", ip, idx - 1,
808                         XFS_DATA_FORK);
809                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
810                         LEFT.br_blockcount + PREV.br_blockcount +
811                         RIGHT.br_blockcount);
812                 xfs_bmap_trace_post_update(fname, "LF|RF|LC|RC", ip, idx - 1,
813                         XFS_DATA_FORK);
814                 xfs_bmap_trace_delete(fname, "LF|RF|LC|RC", ip, idx, 2,
815                         XFS_DATA_FORK);
816                 xfs_iext_remove(ifp, idx, 2);
817                 ip->i_df.if_lastex = idx - 1;
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, RIGHT.br_startoff,
824                                         RIGHT.br_startblock,
825                                         RIGHT.br_blockcount, &i)))
826                                 goto done;
827                         ASSERT(i == 1);
828                         if ((error = xfs_bmbt_delete(cur, &i)))
829                                 goto done;
830                         ASSERT(i == 1);
831                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
832                                 goto done;
833                         ASSERT(i == 1);
834                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
835                                         LEFT.br_startblock,
836                                         LEFT.br_blockcount +
837                                         PREV.br_blockcount +
838                                         RIGHT.br_blockcount, LEFT.br_state)))
839                                 goto done;
840                 }
841                 *dnew = 0;
842                 break;
843
844         case MASK3(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG):
845                 /*
846                  * Filling in all of a previously delayed allocation extent.
847                  * The left neighbor is contiguous, the right is not.
848                  */
849                 xfs_bmap_trace_pre_update(fname, "LF|RF|LC", ip, idx - 1,
850                         XFS_DATA_FORK);
851                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
852                         LEFT.br_blockcount + PREV.br_blockcount);
853                 xfs_bmap_trace_post_update(fname, "LF|RF|LC", ip, idx - 1,
854                         XFS_DATA_FORK);
855                 ip->i_df.if_lastex = idx - 1;
856                 xfs_bmap_trace_delete(fname, "LF|RF|LC", ip, idx, 1,
857                         XFS_DATA_FORK);
858                 xfs_iext_remove(ifp, idx, 1);
859                 if (cur == NULL)
860                         rval = XFS_ILOG_DEXT;
861                 else {
862                         rval = 0;
863                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
864                                         LEFT.br_startblock, LEFT.br_blockcount,
865                                         &i)))
866                                 goto done;
867                         ASSERT(i == 1);
868                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
869                                         LEFT.br_startblock,
870                                         LEFT.br_blockcount +
871                                         PREV.br_blockcount, LEFT.br_state)))
872                                 goto done;
873                 }
874                 *dnew = 0;
875                 break;
876
877         case MASK3(LEFT_FILLING, RIGHT_FILLING, RIGHT_CONTIG):
878                 /*
879                  * Filling in all of a previously delayed allocation extent.
880                  * The right neighbor is contiguous, the left is not.
881                  */
882                 xfs_bmap_trace_pre_update(fname, "LF|RF|RC", ip, idx,
883                         XFS_DATA_FORK);
884                 xfs_bmbt_set_startblock(ep, new->br_startblock);
885                 xfs_bmbt_set_blockcount(ep,
886                         PREV.br_blockcount + RIGHT.br_blockcount);
887                 xfs_bmap_trace_post_update(fname, "LF|RF|RC", ip, idx,
888                         XFS_DATA_FORK);
889                 ip->i_df.if_lastex = idx;
890                 xfs_bmap_trace_delete(fname, "LF|RF|RC", ip, idx + 1, 1,
891                         XFS_DATA_FORK);
892                 xfs_iext_remove(ifp, idx + 1, 1);
893                 if (cur == NULL)
894                         rval = XFS_ILOG_DEXT;
895                 else {
896                         rval = 0;
897                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
898                                         RIGHT.br_startblock,
899                                         RIGHT.br_blockcount, &i)))
900                                 goto done;
901                         ASSERT(i == 1);
902                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
903                                         new->br_startblock,
904                                         PREV.br_blockcount +
905                                         RIGHT.br_blockcount, PREV.br_state)))
906                                 goto done;
907                 }
908                 *dnew = 0;
909                 break;
910
911         case MASK2(LEFT_FILLING, RIGHT_FILLING):
912                 /*
913                  * Filling in all of a previously delayed allocation extent.
914                  * Neither the left nor right neighbors are contiguous with
915                  * the new one.
916                  */
917                 xfs_bmap_trace_pre_update(fname, "LF|RF", ip, idx,
918                         XFS_DATA_FORK);
919                 xfs_bmbt_set_startblock(ep, new->br_startblock);
920                 xfs_bmap_trace_post_update(fname, "LF|RF", ip, idx,
921                         XFS_DATA_FORK);
922                 ip->i_df.if_lastex = idx;
923                 ip->i_d.di_nextents++;
924                 if (cur == NULL)
925                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
926                 else {
927                         rval = XFS_ILOG_CORE;
928                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
929                                         new->br_startblock, new->br_blockcount,
930                                         &i)))
931                                 goto done;
932                         ASSERT(i == 0);
933                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
934                         if ((error = xfs_bmbt_insert(cur, &i)))
935                                 goto done;
936                         ASSERT(i == 1);
937                 }
938                 *dnew = 0;
939                 break;
940
941         case MASK2(LEFT_FILLING, LEFT_CONTIG):
942                 /*
943                  * Filling in the first part of a previous delayed allocation.
944                  * The left neighbor is contiguous.
945                  */
946                 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx - 1,
947                         XFS_DATA_FORK);
948                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
949                         LEFT.br_blockcount + new->br_blockcount);
950                 xfs_bmbt_set_startoff(ep,
951                         PREV.br_startoff + new->br_blockcount);
952                 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx - 1,
953                         XFS_DATA_FORK);
954                 temp = PREV.br_blockcount - new->br_blockcount;
955                 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx,
956                         XFS_DATA_FORK);
957                 xfs_bmbt_set_blockcount(ep, temp);
958                 ip->i_df.if_lastex = idx - 1;
959                 if (cur == NULL)
960                         rval = XFS_ILOG_DEXT;
961                 else {
962                         rval = 0;
963                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
964                                         LEFT.br_startblock, LEFT.br_blockcount,
965                                         &i)))
966                                 goto done;
967                         ASSERT(i == 1);
968                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
969                                         LEFT.br_startblock,
970                                         LEFT.br_blockcount +
971                                         new->br_blockcount,
972                                         LEFT.br_state)))
973                                 goto done;
974                 }
975                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
976                         STARTBLOCKVAL(PREV.br_startblock));
977                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
978                 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx,
979                         XFS_DATA_FORK);
980                 *dnew = temp;
981                 break;
982
983         case MASK(LEFT_FILLING):
984                 /*
985                  * Filling in the first part of a previous delayed allocation.
986                  * The left neighbor is not contiguous.
987                  */
988                 xfs_bmap_trace_pre_update(fname, "LF", ip, idx, XFS_DATA_FORK);
989                 xfs_bmbt_set_startoff(ep, new_endoff);
990                 temp = PREV.br_blockcount - new->br_blockcount;
991                 xfs_bmbt_set_blockcount(ep, temp);
992                 xfs_bmap_trace_insert(fname, "LF", ip, idx, 1, new, NULL,
993                         XFS_DATA_FORK);
994                 xfs_iext_insert(ifp, idx, 1, new);
995                 ip->i_df.if_lastex = idx;
996                 ip->i_d.di_nextents++;
997                 if (cur == NULL)
998                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
999                 else {
1000                         rval = XFS_ILOG_CORE;
1001                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1002                                         new->br_startblock, new->br_blockcount,
1003                                         &i)))
1004                                 goto done;
1005                         ASSERT(i == 0);
1006                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1007                         if ((error = xfs_bmbt_insert(cur, &i)))
1008                                 goto done;
1009                         ASSERT(i == 1);
1010                 }
1011                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1012                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1013                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1014                                         first, flist, &cur, 1, &tmp_rval,
1015                                         XFS_DATA_FORK);
1016                         rval |= tmp_rval;
1017                         if (error)
1018                                 goto done;
1019                 }
1020                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1021                         STARTBLOCKVAL(PREV.br_startblock) -
1022                         (cur ? cur->bc_private.b.allocated : 0));
1023                 ep = xfs_iext_get_ext(ifp, idx + 1);
1024                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1025                 xfs_bmap_trace_post_update(fname, "LF", ip, idx + 1,
1026                         XFS_DATA_FORK);
1027                 *dnew = temp;
1028                 break;
1029
1030         case MASK2(RIGHT_FILLING, RIGHT_CONTIG):
1031                 /*
1032                  * Filling in the last part of a previous delayed allocation.
1033                  * The right neighbor is contiguous with the new allocation.
1034                  */
1035                 temp = PREV.br_blockcount - new->br_blockcount;
1036                 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx,
1037                         XFS_DATA_FORK);
1038                 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx + 1,
1039                         XFS_DATA_FORK);
1040                 xfs_bmbt_set_blockcount(ep, temp);
1041                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, idx + 1),
1042                         new->br_startoff, new->br_startblock,
1043                         new->br_blockcount + RIGHT.br_blockcount,
1044                         RIGHT.br_state);
1045                 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx + 1,
1046                         XFS_DATA_FORK);
1047                 ip->i_df.if_lastex = idx + 1;
1048                 if (cur == NULL)
1049                         rval = XFS_ILOG_DEXT;
1050                 else {
1051                         rval = 0;
1052                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1053                                         RIGHT.br_startblock,
1054                                         RIGHT.br_blockcount, &i)))
1055                                 goto done;
1056                         ASSERT(i == 1);
1057                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1058                                         new->br_startblock,
1059                                         new->br_blockcount +
1060                                         RIGHT.br_blockcount,
1061                                         RIGHT.br_state)))
1062                                 goto done;
1063                 }
1064                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1065                         STARTBLOCKVAL(PREV.br_startblock));
1066                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1067                 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx,
1068                         XFS_DATA_FORK);
1069                 *dnew = temp;
1070                 break;
1071
1072         case MASK(RIGHT_FILLING):
1073                 /*
1074                  * Filling in the last part of a previous delayed allocation.
1075                  * The right neighbor is not contiguous.
1076                  */
1077                 temp = PREV.br_blockcount - new->br_blockcount;
1078                 xfs_bmap_trace_pre_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1079                 xfs_bmbt_set_blockcount(ep, temp);
1080                 xfs_bmap_trace_insert(fname, "RF", ip, idx + 1, 1,
1081                         new, NULL, XFS_DATA_FORK);
1082                 xfs_iext_insert(ifp, idx + 1, 1, new);
1083                 ip->i_df.if_lastex = idx + 1;
1084                 ip->i_d.di_nextents++;
1085                 if (cur == NULL)
1086                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1087                 else {
1088                         rval = XFS_ILOG_CORE;
1089                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1090                                         new->br_startblock, new->br_blockcount,
1091                                         &i)))
1092                                 goto done;
1093                         ASSERT(i == 0);
1094                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1095                         if ((error = xfs_bmbt_insert(cur, &i)))
1096                                 goto done;
1097                         ASSERT(i == 1);
1098                 }
1099                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1100                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1101                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1102                                 first, flist, &cur, 1, &tmp_rval,
1103                                 XFS_DATA_FORK);
1104                         rval |= tmp_rval;
1105                         if (error)
1106                                 goto done;
1107                 }
1108                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1109                         STARTBLOCKVAL(PREV.br_startblock) -
1110                         (cur ? cur->bc_private.b.allocated : 0));
1111                 ep = xfs_iext_get_ext(ifp, idx);
1112                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1113                 xfs_bmap_trace_post_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1114                 *dnew = temp;
1115                 break;
1116
1117         case 0:
1118                 /*
1119                  * Filling in the middle part of a previous delayed allocation.
1120                  * Contiguity is impossible here.
1121                  * This case is avoided almost all the time.
1122                  */
1123                 temp = new->br_startoff - PREV.br_startoff;
1124                 xfs_bmap_trace_pre_update(fname, "0", ip, idx, XFS_DATA_FORK);
1125                 xfs_bmbt_set_blockcount(ep, temp);
1126                 r[0] = *new;
1127                 r[1].br_startoff = new_endoff;
1128                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
1129                 r[1].br_blockcount = temp2;
1130                 xfs_bmap_trace_insert(fname, "0", ip, idx + 1, 2, &r[0], &r[1],
1131                         XFS_DATA_FORK);
1132                 xfs_iext_insert(ifp, idx + 1, 2, &r[0]);
1133                 ip->i_df.if_lastex = idx + 1;
1134                 ip->i_d.di_nextents++;
1135                 if (cur == NULL)
1136                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1137                 else {
1138                         rval = XFS_ILOG_CORE;
1139                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1140                                         new->br_startblock, new->br_blockcount,
1141                                         &i)))
1142                                 goto done;
1143                         ASSERT(i == 0);
1144                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1145                         if ((error = xfs_bmbt_insert(cur, &i)))
1146                                 goto done;
1147                         ASSERT(i == 1);
1148                 }
1149                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1150                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1151                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1152                                         first, flist, &cur, 1, &tmp_rval,
1153                                         XFS_DATA_FORK);
1154                         rval |= tmp_rval;
1155                         if (error)
1156                                 goto done;
1157                 }
1158                 temp = xfs_bmap_worst_indlen(ip, temp);
1159                 temp2 = xfs_bmap_worst_indlen(ip, temp2);
1160                 diff = (int)(temp + temp2 - STARTBLOCKVAL(PREV.br_startblock) -
1161                         (cur ? cur->bc_private.b.allocated : 0));
1162                 if (diff > 0 &&
1163                     xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS, -diff, rsvd)) {
1164                         /*
1165                          * Ick gross gag me with a spoon.
1166                          */
1167                         ASSERT(0);      /* want to see if this ever happens! */
1168                         while (diff > 0) {
1169                                 if (temp) {
1170                                         temp--;
1171                                         diff--;
1172                                         if (!diff ||
1173                                             !xfs_mod_incore_sb(ip->i_mount,
1174                                                     XFS_SBS_FDBLOCKS, -diff, rsvd))
1175                                                 break;
1176                                 }
1177                                 if (temp2) {
1178                                         temp2--;
1179                                         diff--;
1180                                         if (!diff ||
1181                                             !xfs_mod_incore_sb(ip->i_mount,
1182                                                     XFS_SBS_FDBLOCKS, -diff, rsvd))
1183                                                 break;
1184                                 }
1185                         }
1186                 }
1187                 ep = xfs_iext_get_ext(ifp, idx);
1188                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1189                 xfs_bmap_trace_post_update(fname, "0", ip, idx, XFS_DATA_FORK);
1190                 xfs_bmap_trace_pre_update(fname, "0", ip, idx + 2,
1191                         XFS_DATA_FORK);
1192                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx + 2),
1193                         NULLSTARTBLOCK((int)temp2));
1194                 xfs_bmap_trace_post_update(fname, "0", ip, idx + 2,
1195                         XFS_DATA_FORK);
1196                 *dnew = temp + temp2;
1197                 break;
1198
1199         case MASK3(LEFT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1200         case MASK3(RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1201         case MASK2(LEFT_FILLING, RIGHT_CONTIG):
1202         case MASK2(RIGHT_FILLING, LEFT_CONTIG):
1203         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1204         case MASK(LEFT_CONTIG):
1205         case MASK(RIGHT_CONTIG):
1206                 /*
1207                  * These cases are all impossible.
1208                  */
1209                 ASSERT(0);
1210         }
1211         *curp = cur;
1212 done:
1213         *logflagsp = rval;
1214         return error;
1215 #undef  LEFT
1216 #undef  RIGHT
1217 #undef  PREV
1218 #undef  MASK
1219 #undef  MASK2
1220 #undef  MASK3
1221 #undef  MASK4
1222 #undef  STATE_SET
1223 #undef  STATE_TEST
1224 #undef  STATE_SET_TEST
1225 #undef  SWITCH_STATE
1226 }
1227
1228 /*
1229  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
1230  * allocation to a real allocation or vice versa.
1231  */
1232 STATIC int                              /* error */
1233 xfs_bmap_add_extent_unwritten_real(
1234         xfs_inode_t             *ip,    /* incore inode pointer */
1235         xfs_extnum_t            idx,    /* extent number to update/insert */
1236         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
1237         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1238         int                     *logflagsp) /* inode logging flags */
1239 {
1240         xfs_btree_cur_t         *cur;   /* btree cursor */
1241         xfs_bmbt_rec_t          *ep;    /* extent entry for idx */
1242         int                     error;  /* error return value */
1243 #ifdef XFS_BMAP_TRACE
1244         static char             fname[] = "xfs_bmap_add_extent_unwritten_real";
1245 #endif
1246         int                     i;      /* temp state */
1247         xfs_ifork_t             *ifp;   /* inode fork pointer */
1248         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1249         xfs_exntst_t            newext; /* new extent state */
1250         xfs_exntst_t            oldext; /* old extent state */
1251         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1252                                         /* left is 0, right is 1, prev is 2 */
1253         int                     rval=0; /* return value (logging flags) */
1254         int                     state = 0;/* state bits, accessed thru macros */
1255         enum {                          /* bit number definitions for state */
1256                 LEFT_CONTIG,    RIGHT_CONTIG,
1257                 LEFT_FILLING,   RIGHT_FILLING,
1258                 LEFT_DELAY,     RIGHT_DELAY,
1259                 LEFT_VALID,     RIGHT_VALID
1260         };
1261
1262 #define LEFT            r[0]
1263 #define RIGHT           r[1]
1264 #define PREV            r[2]
1265 #define MASK(b)         (1 << (b))
1266 #define MASK2(a,b)      (MASK(a) | MASK(b))
1267 #define MASK3(a,b,c)    (MASK2(a,b) | MASK(c))
1268 #define MASK4(a,b,c,d)  (MASK3(a,b,c) | MASK(d))
1269 #define STATE_SET(b,v)  ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1270 #define STATE_TEST(b)   (state & MASK(b))
1271 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
1272                                        ((state &= ~MASK(b)), 0))
1273 #define SWITCH_STATE            \
1274         (state & MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG))
1275
1276         /*
1277          * Set up a bunch of variables to make the tests simpler.
1278          */
1279         error = 0;
1280         cur = *curp;
1281         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1282         ep = xfs_iext_get_ext(ifp, idx);
1283         xfs_bmbt_get_all(ep, &PREV);
1284         newext = new->br_state;
1285         oldext = (newext == XFS_EXT_UNWRITTEN) ?
1286                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
1287         ASSERT(PREV.br_state == oldext);
1288         new_endoff = new->br_startoff + new->br_blockcount;
1289         ASSERT(PREV.br_startoff <= new->br_startoff);
1290         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1291         /*
1292          * Set flags determining what part of the previous oldext allocation
1293          * extent is being replaced by a newext allocation.
1294          */
1295         STATE_SET(LEFT_FILLING, PREV.br_startoff == new->br_startoff);
1296         STATE_SET(RIGHT_FILLING,
1297                 PREV.br_startoff + PREV.br_blockcount == new_endoff);
1298         /*
1299          * Check and set flags if this segment has a left neighbor.
1300          * Don't set contiguous if the combined extent would be too large.
1301          */
1302         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1303                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &LEFT);
1304                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(LEFT.br_startblock));
1305         }
1306         STATE_SET(LEFT_CONTIG,
1307                 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
1308                 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1309                 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1310                 LEFT.br_state == newext &&
1311                 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1312         /*
1313          * Check and set flags if this segment has a right neighbor.
1314          * Don't set contiguous if the combined extent would be too large.
1315          * Also check for all-three-contiguous being too large.
1316          */
1317         if (STATE_SET_TEST(RIGHT_VALID,
1318                         idx <
1319                         ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1)) {
1320                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx + 1), &RIGHT);
1321                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(RIGHT.br_startblock));
1322         }
1323         STATE_SET(RIGHT_CONTIG,
1324                 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
1325                 new_endoff == RIGHT.br_startoff &&
1326                 new->br_startblock + new->br_blockcount ==
1327                     RIGHT.br_startblock &&
1328                 newext == RIGHT.br_state &&
1329                 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1330                 ((state & MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING)) !=
1331                   MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING) ||
1332                  LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1333                      <= MAXEXTLEN));
1334         /*
1335          * Switch out based on the FILLING and CONTIG state bits.
1336          */
1337         switch (SWITCH_STATE) {
1338
1339         case MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1340                 /*
1341                  * Setting all of a previous oldext extent to newext.
1342                  * The left and right neighbors are both contiguous with new.
1343                  */
1344                 xfs_bmap_trace_pre_update(fname, "LF|RF|LC|RC", ip, idx - 1,
1345                         XFS_DATA_FORK);
1346                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1347                         LEFT.br_blockcount + PREV.br_blockcount +
1348                         RIGHT.br_blockcount);
1349                 xfs_bmap_trace_post_update(fname, "LF|RF|LC|RC", ip, idx - 1,
1350                         XFS_DATA_FORK);
1351                 xfs_bmap_trace_delete(fname, "LF|RF|LC|RC", ip, idx, 2,
1352                         XFS_DATA_FORK);
1353                 xfs_iext_remove(ifp, idx, 2);
1354                 ip->i_df.if_lastex = idx - 1;
1355                 ip->i_d.di_nextents -= 2;
1356                 if (cur == NULL)
1357                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1358                 else {
1359                         rval = XFS_ILOG_CORE;
1360                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1361                                         RIGHT.br_startblock,
1362                                         RIGHT.br_blockcount, &i)))
1363                                 goto done;
1364                         ASSERT(i == 1);
1365                         if ((error = xfs_bmbt_delete(cur, &i)))
1366                                 goto done;
1367                         ASSERT(i == 1);
1368                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1369                                 goto done;
1370                         ASSERT(i == 1);
1371                         if ((error = xfs_bmbt_delete(cur, &i)))
1372                                 goto done;
1373                         ASSERT(i == 1);
1374                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1375                                 goto done;
1376                         ASSERT(i == 1);
1377                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1378                                 LEFT.br_startblock,
1379                                 LEFT.br_blockcount + PREV.br_blockcount +
1380                                 RIGHT.br_blockcount, LEFT.br_state)))
1381                                 goto done;
1382                 }
1383                 break;
1384
1385         case MASK3(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG):
1386                 /*
1387                  * Setting all of a previous oldext extent to newext.
1388                  * The left neighbor is contiguous, the right is not.
1389                  */
1390                 xfs_bmap_trace_pre_update(fname, "LF|RF|LC", ip, idx - 1,
1391                         XFS_DATA_FORK);
1392                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1393                         LEFT.br_blockcount + PREV.br_blockcount);
1394                 xfs_bmap_trace_post_update(fname, "LF|RF|LC", ip, idx - 1,
1395                         XFS_DATA_FORK);
1396                 ip->i_df.if_lastex = idx - 1;
1397                 xfs_bmap_trace_delete(fname, "LF|RF|LC", ip, idx, 1,
1398                         XFS_DATA_FORK);
1399                 xfs_iext_remove(ifp, idx, 1);
1400                 ip->i_d.di_nextents--;
1401                 if (cur == NULL)
1402                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1403                 else {
1404                         rval = XFS_ILOG_CORE;
1405                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1406                                         PREV.br_startblock, PREV.br_blockcount,
1407                                         &i)))
1408                                 goto done;
1409                         ASSERT(i == 1);
1410                         if ((error = xfs_bmbt_delete(cur, &i)))
1411                                 goto done;
1412                         ASSERT(i == 1);
1413                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1414                                 goto done;
1415                         ASSERT(i == 1);
1416                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1417                                 LEFT.br_startblock,
1418                                 LEFT.br_blockcount + PREV.br_blockcount,
1419                                 LEFT.br_state)))
1420                                 goto done;
1421                 }
1422                 break;
1423
1424         case MASK3(LEFT_FILLING, RIGHT_FILLING, RIGHT_CONTIG):
1425                 /*
1426                  * Setting all of a previous oldext extent to newext.
1427                  * The right neighbor is contiguous, the left is not.
1428                  */
1429                 xfs_bmap_trace_pre_update(fname, "LF|RF|RC", ip, idx,
1430                         XFS_DATA_FORK);
1431                 xfs_bmbt_set_blockcount(ep,
1432                         PREV.br_blockcount + RIGHT.br_blockcount);
1433                 xfs_bmbt_set_state(ep, newext);
1434                 xfs_bmap_trace_post_update(fname, "LF|RF|RC", ip, idx,
1435                         XFS_DATA_FORK);
1436                 ip->i_df.if_lastex = idx;
1437                 xfs_bmap_trace_delete(fname, "LF|RF|RC", ip, idx + 1, 1,
1438                         XFS_DATA_FORK);
1439                 xfs_iext_remove(ifp, idx + 1, 1);
1440                 ip->i_d.di_nextents--;
1441                 if (cur == NULL)
1442                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1443                 else {
1444                         rval = XFS_ILOG_CORE;
1445                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1446                                         RIGHT.br_startblock,
1447                                         RIGHT.br_blockcount, &i)))
1448                                 goto done;
1449                         ASSERT(i == 1);
1450                         if ((error = xfs_bmbt_delete(cur, &i)))
1451                                 goto done;
1452                         ASSERT(i == 1);
1453                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1454                                 goto done;
1455                         ASSERT(i == 1);
1456                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1457                                 new->br_startblock,
1458                                 new->br_blockcount + RIGHT.br_blockcount,
1459                                 newext)))
1460                                 goto done;
1461                 }
1462                 break;
1463
1464         case MASK2(LEFT_FILLING, RIGHT_FILLING):
1465                 /*
1466                  * Setting all of a previous oldext extent to newext.
1467                  * Neither the left nor right neighbors are contiguous with
1468                  * the new one.
1469                  */
1470                 xfs_bmap_trace_pre_update(fname, "LF|RF", ip, idx,
1471                         XFS_DATA_FORK);
1472                 xfs_bmbt_set_state(ep, newext);
1473                 xfs_bmap_trace_post_update(fname, "LF|RF", ip, idx,
1474                         XFS_DATA_FORK);
1475                 ip->i_df.if_lastex = idx;
1476                 if (cur == NULL)
1477                         rval = XFS_ILOG_DEXT;
1478                 else {
1479                         rval = 0;
1480                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1481                                         new->br_startblock, new->br_blockcount,
1482                                         &i)))
1483                                 goto done;
1484                         ASSERT(i == 1);
1485                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1486                                 new->br_startblock, new->br_blockcount,
1487                                 newext)))
1488                                 goto done;
1489                 }
1490                 break;
1491
1492         case MASK2(LEFT_FILLING, LEFT_CONTIG):
1493                 /*
1494                  * Setting the first part of a previous oldext extent to newext.
1495                  * The left neighbor is contiguous.
1496                  */
1497                 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx - 1,
1498                         XFS_DATA_FORK);
1499                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1500                         LEFT.br_blockcount + new->br_blockcount);
1501                 xfs_bmbt_set_startoff(ep,
1502                         PREV.br_startoff + new->br_blockcount);
1503                 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx - 1,
1504                         XFS_DATA_FORK);
1505                 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx,
1506                         XFS_DATA_FORK);
1507                 xfs_bmbt_set_startblock(ep,
1508                         new->br_startblock + new->br_blockcount);
1509                 xfs_bmbt_set_blockcount(ep,
1510                         PREV.br_blockcount - new->br_blockcount);
1511                 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx,
1512                         XFS_DATA_FORK);
1513                 ip->i_df.if_lastex = idx - 1;
1514                 if (cur == NULL)
1515                         rval = XFS_ILOG_DEXT;
1516                 else {
1517                         rval = 0;
1518                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1519                                         PREV.br_startblock, PREV.br_blockcount,
1520                                         &i)))
1521                                 goto done;
1522                         ASSERT(i == 1);
1523                         if ((error = xfs_bmbt_update(cur,
1524                                 PREV.br_startoff + new->br_blockcount,
1525                                 PREV.br_startblock + new->br_blockcount,
1526                                 PREV.br_blockcount - new->br_blockcount,
1527                                 oldext)))
1528                                 goto done;
1529                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1530                                 goto done;
1531                         if (xfs_bmbt_update(cur, LEFT.br_startoff,
1532                                 LEFT.br_startblock,
1533                                 LEFT.br_blockcount + new->br_blockcount,
1534                                 LEFT.br_state))
1535                                 goto done;
1536                 }
1537                 break;
1538
1539         case MASK(LEFT_FILLING):
1540                 /*
1541                  * Setting the first part of a previous oldext extent to newext.
1542                  * The left neighbor is not contiguous.
1543                  */
1544                 xfs_bmap_trace_pre_update(fname, "LF", ip, idx, XFS_DATA_FORK);
1545                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1546                 xfs_bmbt_set_startoff(ep, new_endoff);
1547                 xfs_bmbt_set_blockcount(ep,
1548                         PREV.br_blockcount - new->br_blockcount);
1549                 xfs_bmbt_set_startblock(ep,
1550                         new->br_startblock + new->br_blockcount);
1551                 xfs_bmap_trace_post_update(fname, "LF", ip, idx, XFS_DATA_FORK);
1552                 xfs_bmap_trace_insert(fname, "LF", ip, idx, 1, new, NULL,
1553                         XFS_DATA_FORK);
1554                 xfs_iext_insert(ifp, idx, 1, new);
1555                 ip->i_df.if_lastex = idx;
1556                 ip->i_d.di_nextents++;
1557                 if (cur == NULL)
1558                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1559                 else {
1560                         rval = XFS_ILOG_CORE;
1561                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1562                                         PREV.br_startblock, PREV.br_blockcount,
1563                                         &i)))
1564                                 goto done;
1565                         ASSERT(i == 1);
1566                         if ((error = xfs_bmbt_update(cur,
1567                                 PREV.br_startoff + new->br_blockcount,
1568                                 PREV.br_startblock + new->br_blockcount,
1569                                 PREV.br_blockcount - new->br_blockcount,
1570                                 oldext)))
1571                                 goto done;
1572                         cur->bc_rec.b = *new;
1573                         if ((error = xfs_bmbt_insert(cur, &i)))
1574                                 goto done;
1575                         ASSERT(i == 1);
1576                 }
1577                 break;
1578
1579         case MASK2(RIGHT_FILLING, RIGHT_CONTIG):
1580                 /*
1581                  * Setting the last part of a previous oldext extent to newext.
1582                  * The right neighbor is contiguous with the new allocation.
1583                  */
1584                 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx,
1585                         XFS_DATA_FORK);
1586                 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx + 1,
1587                         XFS_DATA_FORK);
1588                 xfs_bmbt_set_blockcount(ep,
1589                         PREV.br_blockcount - new->br_blockcount);
1590                 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx,
1591                         XFS_DATA_FORK);
1592                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, idx + 1),
1593                         new->br_startoff, new->br_startblock,
1594                         new->br_blockcount + RIGHT.br_blockcount, newext);
1595                 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx + 1,
1596                         XFS_DATA_FORK);
1597                 ip->i_df.if_lastex = idx + 1;
1598                 if (cur == NULL)
1599                         rval = XFS_ILOG_DEXT;
1600                 else {
1601                         rval = 0;
1602                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1603                                         PREV.br_startblock,
1604                                         PREV.br_blockcount, &i)))
1605                                 goto done;
1606                         ASSERT(i == 1);
1607                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1608                                 PREV.br_startblock,
1609                                 PREV.br_blockcount - new->br_blockcount,
1610                                 oldext)))
1611                                 goto done;
1612                         if ((error = xfs_bmbt_increment(cur, 0, &i)))
1613                                 goto done;
1614                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1615                                 new->br_startblock,
1616                                 new->br_blockcount + RIGHT.br_blockcount,
1617                                 newext)))
1618                                 goto done;
1619                 }
1620                 break;
1621
1622         case MASK(RIGHT_FILLING):
1623                 /*
1624                  * Setting the last part of a previous oldext extent to newext.
1625                  * The right neighbor is not contiguous.
1626                  */
1627                 xfs_bmap_trace_pre_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1628                 xfs_bmbt_set_blockcount(ep,
1629                         PREV.br_blockcount - new->br_blockcount);
1630                 xfs_bmap_trace_post_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1631                 xfs_bmap_trace_insert(fname, "RF", ip, idx + 1, 1,
1632                         new, NULL, XFS_DATA_FORK);
1633                 xfs_iext_insert(ifp, idx + 1, 1, new);
1634                 ip->i_df.if_lastex = idx + 1;
1635                 ip->i_d.di_nextents++;
1636                 if (cur == NULL)
1637                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1638                 else {
1639                         rval = XFS_ILOG_CORE;
1640                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1641                                         PREV.br_startblock, PREV.br_blockcount,
1642                                         &i)))
1643                                 goto done;
1644                         ASSERT(i == 1);
1645                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1646                                 PREV.br_startblock,
1647                                 PREV.br_blockcount - new->br_blockcount,
1648                                 oldext)))
1649                                 goto done;
1650                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1651                                         new->br_startblock, new->br_blockcount,
1652                                         &i)))
1653                                 goto done;
1654                         ASSERT(i == 0);
1655                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1656                         if ((error = xfs_bmbt_insert(cur, &i)))
1657                                 goto done;
1658                         ASSERT(i == 1);
1659                 }
1660                 break;
1661
1662         case 0:
1663                 /*
1664                  * Setting the middle part of a previous oldext extent to
1665                  * newext.  Contiguity is impossible here.
1666                  * One extent becomes three extents.
1667                  */
1668                 xfs_bmap_trace_pre_update(fname, "0", ip, idx, XFS_DATA_FORK);
1669                 xfs_bmbt_set_blockcount(ep,
1670                         new->br_startoff - PREV.br_startoff);
1671                 xfs_bmap_trace_post_update(fname, "0", ip, idx, XFS_DATA_FORK);
1672                 r[0] = *new;
1673                 r[1].br_startoff = new_endoff;
1674                 r[1].br_blockcount =
1675                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
1676                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1677                 r[1].br_state = oldext;
1678                 xfs_bmap_trace_insert(fname, "0", ip, idx + 1, 2, &r[0], &r[1],
1679                         XFS_DATA_FORK);
1680                 xfs_iext_insert(ifp, idx + 1, 2, &r[0]);
1681                 ip->i_df.if_lastex = idx + 1;
1682                 ip->i_d.di_nextents += 2;
1683                 if (cur == NULL)
1684                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1685                 else {
1686                         rval = XFS_ILOG_CORE;
1687                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1688                                         PREV.br_startblock, PREV.br_blockcount,
1689                                         &i)))
1690                                 goto done;
1691                         ASSERT(i == 1);
1692                         /* new right extent - oldext */
1693                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1694                                 r[1].br_startblock, r[1].br_blockcount,
1695                                 r[1].br_state)))
1696                                 goto done;
1697                         /* new left extent - oldext */
1698                         PREV.br_blockcount =
1699                                 new->br_startoff - PREV.br_startoff;
1700                         cur->bc_rec.b = PREV;
1701                         if ((error = xfs_bmbt_insert(cur, &i)))
1702                                 goto done;
1703                         ASSERT(i == 1);
1704                         if ((error = xfs_bmbt_increment(cur, 0, &i)))
1705                                 goto done;
1706                         ASSERT(i == 1);
1707                         /* new middle extent - newext */
1708                         cur->bc_rec.b = *new;
1709                         if ((error = xfs_bmbt_insert(cur, &i)))
1710                                 goto done;
1711                         ASSERT(i == 1);
1712                 }
1713                 break;
1714
1715         case MASK3(LEFT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1716         case MASK3(RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1717         case MASK2(LEFT_FILLING, RIGHT_CONTIG):
1718         case MASK2(RIGHT_FILLING, LEFT_CONTIG):
1719         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1720         case MASK(LEFT_CONTIG):
1721         case MASK(RIGHT_CONTIG):
1722                 /*
1723                  * These cases are all impossible.
1724                  */
1725                 ASSERT(0);
1726         }
1727         *curp = cur;
1728 done:
1729         *logflagsp = rval;
1730         return error;
1731 #undef  LEFT
1732 #undef  RIGHT
1733 #undef  PREV
1734 #undef  MASK
1735 #undef  MASK2
1736 #undef  MASK3
1737 #undef  MASK4
1738 #undef  STATE_SET
1739 #undef  STATE_TEST
1740 #undef  STATE_SET_TEST
1741 #undef  SWITCH_STATE
1742 }
1743
1744 /*
1745  * Called by xfs_bmap_add_extent to handle cases converting a hole
1746  * to a delayed allocation.
1747  */
1748 /*ARGSUSED*/
1749 STATIC int                              /* error */
1750 xfs_bmap_add_extent_hole_delay(
1751         xfs_inode_t             *ip,    /* incore inode pointer */
1752         xfs_extnum_t            idx,    /* extent number to update/insert */
1753         xfs_btree_cur_t         *cur,   /* if null, not a btree */
1754         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1755         int                     *logflagsp, /* inode logging flags */
1756         int                     rsvd)           /* OK to allocate reserved blocks */
1757 {
1758         xfs_bmbt_rec_t          *ep;    /* extent record for idx */
1759 #ifdef XFS_BMAP_TRACE
1760         static char             fname[] = "xfs_bmap_add_extent_hole_delay";
1761 #endif
1762         xfs_ifork_t             *ifp;   /* inode fork pointer */
1763         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1764         xfs_filblks_t           newlen=0;       /* new indirect size */
1765         xfs_filblks_t           oldlen=0;       /* old indirect size */
1766         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1767         int                     state;  /* state bits, accessed thru macros */
1768         xfs_filblks_t           temp;   /* temp for indirect calculations */
1769         enum {                          /* bit number definitions for state */
1770                 LEFT_CONTIG,    RIGHT_CONTIG,
1771                 LEFT_DELAY,     RIGHT_DELAY,
1772                 LEFT_VALID,     RIGHT_VALID
1773         };
1774
1775 #define MASK(b)                 (1 << (b))
1776 #define MASK2(a,b)              (MASK(a) | MASK(b))
1777 #define STATE_SET(b,v)          ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1778 #define STATE_TEST(b)           (state & MASK(b))
1779 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
1780                                        ((state &= ~MASK(b)), 0))
1781 #define SWITCH_STATE            (state & MASK2(LEFT_CONTIG, RIGHT_CONTIG))
1782
1783         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1784         ep = xfs_iext_get_ext(ifp, idx);
1785         state = 0;
1786         ASSERT(ISNULLSTARTBLOCK(new->br_startblock));
1787         /*
1788          * Check and set flags if this segment has a left neighbor
1789          */
1790         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1791                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &left);
1792                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(left.br_startblock));
1793         }
1794         /*
1795          * Check and set flags if the current (right) segment exists.
1796          * If it doesn't exist, we're converting the hole at end-of-file.
1797          */
1798         if (STATE_SET_TEST(RIGHT_VALID,
1799                            idx <
1800                            ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1801                 xfs_bmbt_get_all(ep, &right);
1802                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(right.br_startblock));
1803         }
1804         /*
1805          * Set contiguity flags on the left and right neighbors.
1806          * Don't let extents get too large, even if the pieces are contiguous.
1807          */
1808         STATE_SET(LEFT_CONTIG,
1809                 STATE_TEST(LEFT_VALID) && STATE_TEST(LEFT_DELAY) &&
1810                 left.br_startoff + left.br_blockcount == new->br_startoff &&
1811                 left.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1812         STATE_SET(RIGHT_CONTIG,
1813                 STATE_TEST(RIGHT_VALID) && STATE_TEST(RIGHT_DELAY) &&
1814                 new->br_startoff + new->br_blockcount == right.br_startoff &&
1815                 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1816                 (!STATE_TEST(LEFT_CONTIG) ||
1817                  (left.br_blockcount + new->br_blockcount +
1818                      right.br_blockcount <= MAXEXTLEN)));
1819         /*
1820          * Switch out based on the contiguity flags.
1821          */
1822         switch (SWITCH_STATE) {
1823
1824         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1825                 /*
1826                  * New allocation is contiguous with delayed allocations
1827                  * on the left and on the right.
1828                  * Merge all three into a single extent record.
1829                  */
1830                 temp = left.br_blockcount + new->br_blockcount +
1831                         right.br_blockcount;
1832                 xfs_bmap_trace_pre_update(fname, "LC|RC", ip, idx - 1,
1833                         XFS_DATA_FORK);
1834                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1), temp);
1835                 oldlen = STARTBLOCKVAL(left.br_startblock) +
1836                         STARTBLOCKVAL(new->br_startblock) +
1837                         STARTBLOCKVAL(right.br_startblock);
1838                 newlen = xfs_bmap_worst_indlen(ip, temp);
1839                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx - 1),
1840                         NULLSTARTBLOCK((int)newlen));
1841                 xfs_bmap_trace_post_update(fname, "LC|RC", ip, idx - 1,
1842                         XFS_DATA_FORK);
1843                 xfs_bmap_trace_delete(fname, "LC|RC", ip, idx, 1,
1844                         XFS_DATA_FORK);
1845                 xfs_iext_remove(ifp, idx, 1);
1846                 ip->i_df.if_lastex = idx - 1;
1847                 break;
1848
1849         case MASK(LEFT_CONTIG):
1850                 /*
1851                  * New allocation is contiguous with a delayed allocation
1852                  * on the left.
1853                  * Merge the new allocation with the left neighbor.
1854                  */
1855                 temp = left.br_blockcount + new->br_blockcount;
1856                 xfs_bmap_trace_pre_update(fname, "LC", ip, idx - 1,
1857                         XFS_DATA_FORK);
1858                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1), temp);
1859                 oldlen = STARTBLOCKVAL(left.br_startblock) +
1860                         STARTBLOCKVAL(new->br_startblock);
1861                 newlen = xfs_bmap_worst_indlen(ip, temp);
1862                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx - 1),
1863                         NULLSTARTBLOCK((int)newlen));
1864                 xfs_bmap_trace_post_update(fname, "LC", ip, idx - 1,
1865                         XFS_DATA_FORK);
1866                 ip->i_df.if_lastex = idx - 1;
1867                 break;
1868
1869         case MASK(RIGHT_CONTIG):
1870                 /*
1871                  * New allocation is contiguous with a delayed allocation
1872                  * on the right.
1873                  * Merge the new allocation with the right neighbor.
1874                  */
1875                 xfs_bmap_trace_pre_update(fname, "RC", ip, idx, XFS_DATA_FORK);
1876                 temp = new->br_blockcount + right.br_blockcount;
1877                 oldlen = STARTBLOCKVAL(new->br_startblock) +
1878                         STARTBLOCKVAL(right.br_startblock);
1879                 newlen = xfs_bmap_worst_indlen(ip, temp);
1880                 xfs_bmbt_set_allf(ep, new->br_startoff,
1881                         NULLSTARTBLOCK((int)newlen), temp, right.br_state);
1882                 xfs_bmap_trace_post_update(fname, "RC", ip, idx, XFS_DATA_FORK);
1883                 ip->i_df.if_lastex = idx;
1884                 break;
1885
1886         case 0:
1887                 /*
1888                  * New allocation is not contiguous with another
1889                  * delayed allocation.
1890                  * Insert a new entry.
1891                  */
1892                 oldlen = newlen = 0;
1893                 xfs_bmap_trace_insert(fname, "0", ip, idx, 1, new, NULL,
1894                         XFS_DATA_FORK);
1895                 xfs_iext_insert(ifp, idx, 1, new);
1896                 ip->i_df.if_lastex = idx;
1897                 break;
1898         }
1899         if (oldlen != newlen) {
1900                 ASSERT(oldlen > newlen);
1901                 xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
1902                         (int)(oldlen - newlen), rsvd);
1903                 /*
1904                  * Nothing to do for disk quota accounting here.
1905                  */
1906         }
1907         *logflagsp = 0;
1908         return 0;
1909 #undef  MASK
1910 #undef  MASK2
1911 #undef  STATE_SET
1912 #undef  STATE_TEST
1913 #undef  STATE_SET_TEST
1914 #undef  SWITCH_STATE
1915 }
1916
1917 /*
1918  * Called by xfs_bmap_add_extent to handle cases converting a hole
1919  * to a real allocation.
1920  */
1921 STATIC int                              /* error */
1922 xfs_bmap_add_extent_hole_real(
1923         xfs_inode_t             *ip,    /* incore inode pointer */
1924         xfs_extnum_t            idx,    /* extent number to update/insert */
1925         xfs_btree_cur_t         *cur,   /* if null, not a btree */
1926         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1927         int                     *logflagsp, /* inode logging flags */
1928         int                     whichfork) /* data or attr fork */
1929 {
1930         xfs_bmbt_rec_t          *ep;    /* pointer to extent entry ins. point */
1931         int                     error;  /* error return value */
1932 #ifdef XFS_BMAP_TRACE
1933         static char             fname[] = "xfs_bmap_add_extent_hole_real";
1934 #endif
1935         int                     i;      /* temp state */
1936         xfs_ifork_t             *ifp;   /* inode fork pointer */
1937         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1938         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1939         int                     state;  /* state bits, accessed thru macros */
1940         enum {                          /* bit number definitions for state */
1941                 LEFT_CONTIG,    RIGHT_CONTIG,
1942                 LEFT_DELAY,     RIGHT_DELAY,
1943                 LEFT_VALID,     RIGHT_VALID
1944         };
1945
1946 #define MASK(b)                 (1 << (b))
1947 #define MASK2(a,b)              (MASK(a) | MASK(b))
1948 #define STATE_SET(b,v)          ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1949 #define STATE_TEST(b)           (state & MASK(b))
1950 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
1951                                        ((state &= ~MASK(b)), 0))
1952 #define SWITCH_STATE            (state & MASK2(LEFT_CONTIG, RIGHT_CONTIG))
1953
1954         ifp = XFS_IFORK_PTR(ip, whichfork);
1955         ASSERT(idx <= ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
1956         ep = xfs_iext_get_ext(ifp, idx);
1957         state = 0;
1958         /*
1959          * Check and set flags if this segment has a left neighbor.
1960          */
1961         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1962                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &left);
1963                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(left.br_startblock));
1964         }
1965         /*
1966          * Check and set flags if this segment has a current value.
1967          * Not true if we're inserting into the "hole" at eof.
1968          */
1969         if (STATE_SET_TEST(RIGHT_VALID,
1970                            idx <
1971                            ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1972                 xfs_bmbt_get_all(ep, &right);
1973                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(right.br_startblock));
1974         }
1975         /*
1976          * We're inserting a real allocation between "left" and "right".
1977          * Set the contiguity flags.  Don't let extents get too large.
1978          */
1979         STATE_SET(LEFT_CONTIG,
1980                 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
1981                 left.br_startoff + left.br_blockcount == new->br_startoff &&
1982                 left.br_startblock + left.br_blockcount == new->br_startblock &&
1983                 left.br_state == new->br_state &&
1984                 left.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1985         STATE_SET(RIGHT_CONTIG,
1986                 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
1987                 new->br_startoff + new->br_blockcount == right.br_startoff &&
1988                 new->br_startblock + new->br_blockcount ==
1989                     right.br_startblock &&
1990                 new->br_state == right.br_state &&
1991                 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1992                 (!STATE_TEST(LEFT_CONTIG) ||
1993                  left.br_blockcount + new->br_blockcount +
1994                      right.br_blockcount <= MAXEXTLEN));
1995
1996         /*
1997          * Select which case we're in here, and implement it.
1998          */
1999         switch (SWITCH_STATE) {
2000
2001         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
2002                 /*
2003                  * New allocation is contiguous with real allocations on the
2004                  * left and on the right.
2005                  * Merge all three into a single extent record.
2006                  */
2007                 xfs_bmap_trace_pre_update(fname, "LC|RC", ip, idx - 1,
2008                         whichfork);
2009                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
2010                         left.br_blockcount + new->br_blockcount +
2011                         right.br_blockcount);
2012                 xfs_bmap_trace_post_update(fname, "LC|RC", ip, idx - 1,
2013                         whichfork);
2014                 xfs_bmap_trace_delete(fname, "LC|RC", ip,
2015                         idx, 1, whichfork);
2016                 xfs_iext_remove(ifp, idx, 1);
2017                 ifp->if_lastex = idx - 1;
2018                 XFS_IFORK_NEXT_SET(ip, whichfork,
2019                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2020                 if (cur == NULL) {
2021                         *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2022                         return 0;
2023                 }
2024                 *logflagsp = XFS_ILOG_CORE;
2025                 if ((error = xfs_bmbt_lookup_eq(cur, right.br_startoff,
2026                                 right.br_startblock, right.br_blockcount, &i)))
2027                         return error;
2028                 ASSERT(i == 1);
2029                 if ((error = xfs_bmbt_delete(cur, &i)))
2030                         return error;
2031                 ASSERT(i == 1);
2032                 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
2033                         return error;
2034                 ASSERT(i == 1);
2035                 error = xfs_bmbt_update(cur, left.br_startoff,
2036                                 left.br_startblock,
2037                                 left.br_blockcount + new->br_blockcount +
2038                                 right.br_blockcount, left.br_state);
2039                 return error;
2040
2041         case MASK(LEFT_CONTIG):
2042                 /*
2043                  * New allocation is contiguous with a real allocation
2044                  * on the left.
2045                  * Merge the new allocation with the left neighbor.
2046                  */
2047                 xfs_bmap_trace_pre_update(fname, "LC", ip, idx - 1, whichfork);
2048                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
2049                         left.br_blockcount + new->br_blockcount);
2050                 xfs_bmap_trace_post_update(fname, "LC", ip, idx - 1, whichfork);
2051                 ifp->if_lastex = idx - 1;
2052                 if (cur == NULL) {
2053                         *logflagsp = XFS_ILOG_FEXT(whichfork);
2054                         return 0;
2055                 }
2056                 *logflagsp = 0;
2057                 if ((error = xfs_bmbt_lookup_eq(cur, left.br_startoff,
2058                                 left.br_startblock, left.br_blockcount, &i)))
2059                         return error;
2060                 ASSERT(i == 1);
2061                 error = xfs_bmbt_update(cur, left.br_startoff,
2062                                 left.br_startblock,
2063                                 left.br_blockcount + new->br_blockcount,
2064                                 left.br_state);
2065                 return error;
2066
2067         case MASK(RIGHT_CONTIG):
2068                 /*
2069                  * New allocation is contiguous with a real allocation
2070                  * on the right.
2071                  * Merge the new allocation with the right neighbor.
2072                  */
2073                 xfs_bmap_trace_pre_update(fname, "RC", ip, idx, whichfork);
2074                 xfs_bmbt_set_allf(ep, new->br_startoff, new->br_startblock,
2075                         new->br_blockcount + right.br_blockcount,
2076                         right.br_state);
2077                 xfs_bmap_trace_post_update(fname, "RC", ip, idx, whichfork);
2078                 ifp->if_lastex = idx;
2079                 if (cur == NULL) {
2080                         *logflagsp = XFS_ILOG_FEXT(whichfork);
2081                         return 0;
2082                 }
2083                 *logflagsp = 0;
2084                 if ((error = xfs_bmbt_lookup_eq(cur, right.br_startoff,
2085                                 right.br_startblock, right.br_blockcount, &i)))
2086                         return error;
2087                 ASSERT(i == 1);
2088                 error = xfs_bmbt_update(cur, new->br_startoff,
2089                                 new->br_startblock,
2090                                 new->br_blockcount + right.br_blockcount,
2091                                 right.br_state);
2092                 return error;
2093
2094         case 0:
2095                 /*
2096                  * New allocation is not contiguous with another
2097                  * real allocation.
2098                  * Insert a new entry.
2099                  */
2100                 xfs_bmap_trace_insert(fname, "0", ip, idx, 1, new, NULL,
2101                         whichfork);
2102                 xfs_iext_insert(ifp, idx, 1, new);
2103                 ifp->if_lastex = idx;
2104                 XFS_IFORK_NEXT_SET(ip, whichfork,
2105                         XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2106                 if (cur == NULL) {
2107                         *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2108                         return 0;
2109                 }
2110                 *logflagsp = XFS_ILOG_CORE;
2111                 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2112                                 new->br_startblock, new->br_blockcount, &i)))
2113                         return error;
2114                 ASSERT(i == 0);
2115                 cur->bc_rec.b.br_state = new->br_state;
2116                 if ((error = xfs_bmbt_insert(cur, &i)))
2117                         return error;
2118                 ASSERT(i == 1);
2119                 return 0;
2120         }
2121 #undef  MASK
2122 #undef  MASK2
2123 #undef  STATE_SET
2124 #undef  STATE_TEST
2125 #undef  STATE_SET_TEST
2126 #undef  SWITCH_STATE
2127         /* NOTREACHED */
2128         ASSERT(0);
2129         return 0; /* keep gcc quite */
2130 }
2131
2132 /*
2133  * Adjust the size of the new extent based on di_extsize and rt extsize.
2134  */
2135 STATIC int
2136 xfs_bmap_extsize_align(
2137         xfs_mount_t     *mp,
2138         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
2139         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
2140         xfs_extlen_t    extsz,          /* align to this extent size */
2141         int             rt,             /* is this a realtime inode? */
2142         int             eof,            /* is extent at end-of-file? */
2143         int             delay,          /* creating delalloc extent? */
2144         int             convert,        /* overwriting unwritten extent? */
2145         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
2146         xfs_extlen_t    *lenp)          /* in/out: aligned length */
2147 {
2148         xfs_fileoff_t   orig_off;       /* original offset */
2149         xfs_extlen_t    orig_alen;      /* original length */
2150         xfs_fileoff_t   orig_end;       /* original off+len */
2151         xfs_fileoff_t   nexto;          /* next file offset */
2152         xfs_fileoff_t   prevo;          /* previous file offset */
2153         xfs_fileoff_t   align_off;      /* temp for offset */
2154         xfs_extlen_t    align_alen;     /* temp for length */
2155         xfs_extlen_t    temp;           /* temp for calculations */
2156
2157         if (convert)
2158                 return 0;
2159
2160         orig_off = align_off = *offp;
2161         orig_alen = align_alen = *lenp;
2162         orig_end = orig_off + orig_alen;
2163
2164         /*
2165          * If this request overlaps an existing extent, then don't
2166          * attempt to perform any additional alignment.
2167          */
2168         if (!delay && !eof &&
2169             (orig_off >= gotp->br_startoff) &&
2170             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2171                 return 0;
2172         }
2173
2174         /*
2175          * If the file offset is unaligned vs. the extent size
2176          * we need to align it.  This will be possible unless
2177          * the file was previously written with a kernel that didn't
2178          * perform this alignment, or if a truncate shot us in the
2179          * foot.
2180          */
2181         temp = do_mod(orig_off, extsz);
2182         if (temp) {
2183                 align_alen += temp;
2184                 align_off -= temp;
2185         }
2186         /*
2187          * Same adjustment for the end of the requested area.
2188          */
2189         if ((temp = (align_alen % extsz))) {
2190                 align_alen += extsz - temp;
2191         }
2192         /*
2193          * If the previous block overlaps with this proposed allocation
2194          * then move the start forward without adjusting the length.
2195          */
2196         if (prevp->br_startoff != NULLFILEOFF) {
2197                 if (prevp->br_startblock == HOLESTARTBLOCK)
2198                         prevo = prevp->br_startoff;
2199                 else
2200                         prevo = prevp->br_startoff + prevp->br_blockcount;
2201         } else
2202                 prevo = 0;
2203         if (align_off != orig_off && align_off < prevo)
2204                 align_off = prevo;
2205         /*
2206          * If the next block overlaps with this proposed allocation
2207          * then move the start back without adjusting the length,
2208          * but not before offset 0.
2209          * This may of course make the start overlap previous block,
2210          * and if we hit the offset 0 limit then the next block
2211          * can still overlap too.
2212          */
2213         if (!eof && gotp->br_startoff != NULLFILEOFF) {
2214                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
2215                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
2216                         nexto = gotp->br_startoff + gotp->br_blockcount;
2217                 else
2218                         nexto = gotp->br_startoff;
2219         } else
2220                 nexto = NULLFILEOFF;
2221         if (!eof &&
2222             align_off + align_alen != orig_end &&
2223             align_off + align_alen > nexto)
2224                 align_off = nexto > align_alen ? nexto - align_alen : 0;
2225         /*
2226          * If we're now overlapping the next or previous extent that
2227          * means we can't fit an extsz piece in this hole.  Just move
2228          * the start forward to the first valid spot and set
2229          * the length so we hit the end.
2230          */
2231         if (align_off != orig_off && align_off < prevo)
2232                 align_off = prevo;
2233         if (align_off + align_alen != orig_end &&
2234             align_off + align_alen > nexto &&
2235             nexto != NULLFILEOFF) {
2236                 ASSERT(nexto > prevo);
2237                 align_alen = nexto - align_off;
2238         }
2239
2240         /*
2241          * If realtime, and the result isn't a multiple of the realtime
2242          * extent size we need to remove blocks until it is.
2243          */
2244         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
2245                 /*
2246                  * We're not covering the original request, or
2247                  * we won't be able to once we fix the length.
2248                  */
2249                 if (orig_off < align_off ||
2250                     orig_end > align_off + align_alen ||
2251                     align_alen - temp < orig_alen)
2252                         return XFS_ERROR(EINVAL);
2253                 /*
2254                  * Try to fix it by moving the start up.
2255                  */
2256                 if (align_off + temp <= orig_off) {
2257                         align_alen -= temp;
2258                         align_off += temp;
2259                 }
2260                 /*
2261                  * Try to fix it by moving the end in.
2262                  */
2263                 else if (align_off + align_alen - temp >= orig_end)
2264                         align_alen -= temp;
2265                 /*
2266                  * Set the start to the minimum then trim the length.
2267                  */
2268                 else {
2269                         align_alen -= orig_off - align_off;
2270                         align_off = orig_off;
2271                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
2272                 }
2273                 /*
2274                  * Result doesn't cover the request, fail it.
2275                  */
2276                 if (orig_off < align_off || orig_end > align_off + align_alen)
2277                         return XFS_ERROR(EINVAL);
2278         } else {
2279                 ASSERT(orig_off >= align_off);
2280                 ASSERT(orig_end <= align_off + align_alen);
2281         }
2282
2283 #ifdef DEBUG
2284         if (!eof && gotp->br_startoff != NULLFILEOFF)
2285                 ASSERT(align_off + align_alen <= gotp->br_startoff);
2286         if (prevp->br_startoff != NULLFILEOFF)
2287                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
2288 #endif
2289
2290         *lenp = align_alen;
2291         *offp = align_off;
2292         return 0;
2293 }
2294
2295 #define XFS_ALLOC_GAP_UNITS     4
2296
2297 STATIC int
2298 xfs_bmap_adjacent(
2299         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2300 {
2301         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
2302         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2303         xfs_mount_t     *mp;            /* mount point structure */
2304         int             nullfb;         /* true if ap->firstblock isn't set */
2305         int             rt;             /* true if inode is realtime */
2306
2307 #define ISVALID(x,y)    \
2308         (rt ? \
2309                 (x) < mp->m_sb.sb_rblocks : \
2310                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2311                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2312                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2313
2314         mp = ap->ip->i_mount;
2315         nullfb = ap->firstblock == NULLFSBLOCK;
2316         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2317         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2318         /*
2319          * If allocating at eof, and there's a previous real block,
2320          * try to use it's last block as our starting point.
2321          */
2322         if (ap->eof && ap->prevp->br_startoff != NULLFILEOFF &&
2323             !ISNULLSTARTBLOCK(ap->prevp->br_startblock) &&
2324             ISVALID(ap->prevp->br_startblock + ap->prevp->br_blockcount,
2325                     ap->prevp->br_startblock)) {
2326                 ap->rval = ap->prevp->br_startblock + ap->prevp->br_blockcount;
2327                 /*
2328                  * Adjust for the gap between prevp and us.
2329                  */
2330                 adjust = ap->off -
2331                         (ap->prevp->br_startoff + ap->prevp->br_blockcount);
2332                 if (adjust &&
2333                     ISVALID(ap->rval + adjust, ap->prevp->br_startblock))
2334                         ap->rval += adjust;
2335         }
2336         /*
2337          * If not at eof, then compare the two neighbor blocks.
2338          * Figure out whether either one gives us a good starting point,
2339          * and pick the better one.
2340          */
2341         else if (!ap->eof) {
2342                 xfs_fsblock_t   gotbno;         /* right side block number */
2343                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
2344                 xfs_fsblock_t   prevbno;        /* left side block number */
2345                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
2346
2347                 /*
2348                  * If there's a previous (left) block, select a requested
2349                  * start block based on it.
2350                  */
2351                 if (ap->prevp->br_startoff != NULLFILEOFF &&
2352                     !ISNULLSTARTBLOCK(ap->prevp->br_startblock) &&
2353                     (prevbno = ap->prevp->br_startblock +
2354                                ap->prevp->br_blockcount) &&
2355                     ISVALID(prevbno, ap->prevp->br_startblock)) {
2356                         /*
2357                          * Calculate gap to end of previous block.
2358                          */
2359                         adjust = prevdiff = ap->off -
2360                                 (ap->prevp->br_startoff +
2361                                  ap->prevp->br_blockcount);
2362                         /*
2363                          * Figure the startblock based on the previous block's
2364                          * end and the gap size.
2365                          * Heuristic!
2366                          * If the gap is large relative to the piece we're
2367                          * allocating, or using it gives us an invalid block
2368                          * number, then just use the end of the previous block.
2369                          */
2370                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2371                             ISVALID(prevbno + prevdiff,
2372                                     ap->prevp->br_startblock))
2373                                 prevbno += adjust;
2374                         else
2375                                 prevdiff += adjust;
2376                         /*
2377                          * If the firstblock forbids it, can't use it,
2378                          * must use default.
2379                          */
2380                         if (!rt && !nullfb &&
2381                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2382                                 prevbno = NULLFSBLOCK;
2383                 }
2384                 /*
2385                  * No previous block or can't follow it, just default.
2386                  */
2387                 else
2388                         prevbno = NULLFSBLOCK;
2389                 /*
2390                  * If there's a following (right) block, select a requested
2391                  * start block based on it.
2392                  */
2393                 if (!ISNULLSTARTBLOCK(ap->gotp->br_startblock)) {
2394                         /*
2395                          * Calculate gap to start of next block.
2396                          */
2397                         adjust = gotdiff = ap->gotp->br_startoff - ap->off;
2398                         /*
2399                          * Figure the startblock based on the next block's
2400                          * start and the gap size.
2401                          */
2402                         gotbno = ap->gotp->br_startblock;
2403                         /*
2404                          * Heuristic!
2405                          * If the gap is large relative to the piece we're
2406                          * allocating, or using it gives us an invalid block
2407                          * number, then just use the start of the next block
2408                          * offset by our length.
2409                          */
2410                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2411                             ISVALID(gotbno - gotdiff, gotbno))
2412                                 gotbno -= adjust;
2413                         else if (ISVALID(gotbno - ap->alen, gotbno)) {
2414                                 gotbno -= ap->alen;
2415                                 gotdiff += adjust - ap->alen;
2416                         } else
2417                                 gotdiff += adjust;
2418                         /*
2419                          * If the firstblock forbids it, can't use it,
2420                          * must use default.
2421                          */
2422                         if (!rt && !nullfb &&
2423                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2424                                 gotbno = NULLFSBLOCK;
2425                 }
2426                 /*
2427                  * No next block, just default.
2428                  */
2429                 else
2430                         gotbno = NULLFSBLOCK;
2431                 /*
2432                  * If both valid, pick the better one, else the only good
2433                  * one, else ap->rval is already set (to 0 or the inode block).
2434                  */
2435                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2436                         ap->rval = prevdiff <= gotdiff ? prevbno : gotbno;
2437                 else if (prevbno != NULLFSBLOCK)
2438                         ap->rval = prevbno;
2439                 else if (gotbno != NULLFSBLOCK)
2440                         ap->rval = gotbno;
2441         }
2442 #undef ISVALID
2443         return 0;
2444 }
2445
2446 STATIC int
2447 xfs_bmap_rtalloc(
2448         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2449 {
2450         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2451         int             error;          /* error return value */
2452         xfs_mount_t     *mp;            /* mount point structure */
2453         xfs_extlen_t    prod = 0;       /* product factor for allocators */
2454         xfs_extlen_t    ralen = 0;      /* realtime allocation length */
2455         xfs_extlen_t    align;          /* minimum allocation alignment */
2456         xfs_rtblock_t   rtx;            /* realtime extent number */
2457         xfs_rtblock_t   rtb;
2458
2459         mp = ap->ip->i_mount;
2460         align = ap->ip->i_d.di_extsize ?
2461                 ap->ip->i_d.di_extsize : mp->m_sb.sb_rextsize;
2462         prod = align / mp->m_sb.sb_rextsize;
2463         error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2464                                         align, 1, ap->eof, 0,
2465                                         ap->conv, &ap->off, &ap->alen);
2466         if (error)
2467                 return error;
2468         ASSERT(ap->alen);
2469         ASSERT(ap->alen % mp->m_sb.sb_rextsize == 0);
2470
2471         /*
2472          * If the offset & length are not perfectly aligned
2473          * then kill prod, it will just get us in trouble.
2474          */
2475         if (do_mod(ap->off, align) || ap->alen % align)
2476                 prod = 1;
2477         /*
2478          * Set ralen to be the actual requested length in rtextents.
2479          */
2480         ralen = ap->alen / mp->m_sb.sb_rextsize;
2481         /*
2482          * If the old value was close enough to MAXEXTLEN that
2483          * we rounded up to it, cut it back so it's valid again.
2484          * Note that if it's a really large request (bigger than
2485          * MAXEXTLEN), we don't hear about that number, and can't
2486          * adjust the starting point to match it.
2487          */
2488         if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2489                 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2490         /*
2491          * If it's an allocation to an empty file at offset 0,
2492          * pick an extent that will space things out in the rt area.
2493          */
2494         if (ap->eof && ap->off == 0) {
2495                 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2496                 if (error)
2497                         return error;
2498                 ap->rval = rtx * mp->m_sb.sb_rextsize;
2499         } else {
2500                 ap->rval = 0;
2501         }
2502
2503         xfs_bmap_adjacent(ap);
2504
2505         /*
2506          * Realtime allocation, done through xfs_rtallocate_extent.
2507          */
2508         atype = ap->rval == 0 ?  XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2509         do_div(ap->rval, mp->m_sb.sb_rextsize);
2510         rtb = ap->rval;
2511         ap->alen = ralen;
2512         if ((error = xfs_rtallocate_extent(ap->tp, ap->rval, 1, ap->alen,
2513                                 &ralen, atype, ap->wasdel, prod, &rtb)))
2514                 return error;
2515         if (rtb == NULLFSBLOCK && prod > 1 &&
2516             (error = xfs_rtallocate_extent(ap->tp, ap->rval, 1,
2517                                            ap->alen, &ralen, atype,
2518                                            ap->wasdel, 1, &rtb)))
2519                 return error;
2520         ap->rval = rtb;
2521         if (ap->rval != NULLFSBLOCK) {
2522                 ap->rval *= mp->m_sb.sb_rextsize;
2523                 ralen *= mp->m_sb.sb_rextsize;
2524                 ap->alen = ralen;
2525                 ap->ip->i_d.di_nblocks += ralen;
2526                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2527                 if (ap->wasdel)
2528                         ap->ip->i_delayed_blks -= ralen;
2529                 /*
2530                  * Adjust the disk quota also. This was reserved
2531                  * earlier.
2532                  */
2533                 XFS_TRANS_MOD_DQUOT_BYINO(mp, ap->tp, ap->ip,
2534                         ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2535                                         XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2536         } else {
2537                 ap->alen = 0;
2538         }
2539         return 0;
2540 }
2541
2542 STATIC int
2543 xfs_bmap_btalloc(
2544         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2545 {
2546         xfs_mount_t     *mp;            /* mount point structure */
2547         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2548         xfs_extlen_t    align;          /* minimum allocation alignment */
2549         xfs_agnumber_t  ag;
2550         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2551         xfs_agnumber_t  startag;
2552         xfs_alloc_arg_t args;
2553         xfs_extlen_t    blen;
2554         xfs_extlen_t    delta;
2555         xfs_extlen_t    longest;
2556         xfs_extlen_t    need;
2557         xfs_extlen_t    nextminlen = 0;
2558         xfs_perag_t     *pag;
2559         int             nullfb;         /* true if ap->firstblock isn't set */
2560         int             isaligned;
2561         int             notinit;
2562         int             tryagain;
2563         int             error;
2564
2565         mp = ap->ip->i_mount;
2566         align = (ap->userdata && ap->ip->i_d.di_extsize &&
2567                 (ap->ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE)) ?
2568                 ap->ip->i_d.di_extsize : 0;
2569         if (unlikely(align)) {
2570                 error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2571                                                 align, 0, ap->eof, 0, ap->conv,
2572                                                 &ap->off, &ap->alen);
2573                 ASSERT(!error);
2574                 ASSERT(ap->alen);
2575         }
2576         nullfb = ap->firstblock == NULLFSBLOCK;
2577         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2578         if (nullfb)
2579                 ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2580         else
2581                 ap->rval = ap->firstblock;
2582
2583         xfs_bmap_adjacent(ap);
2584
2585         /*
2586          * If allowed, use ap->rval; otherwise must use firstblock since
2587          * it's in the right allocation group.
2588          */
2589         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->rval) == fb_agno)
2590                 ;
2591         else
2592                 ap->rval = ap->firstblock;
2593         /*
2594          * Normal allocation, done through xfs_alloc_vextent.
2595          */
2596         tryagain = isaligned = 0;
2597         args.tp = ap->tp;
2598         args.mp = mp;
2599         args.fsbno = ap->rval;
2600         args.maxlen = MIN(ap->alen, mp->m_sb.sb_agblocks);
2601         blen = 0;
2602         if (nullfb) {
2603                 args.type = XFS_ALLOCTYPE_START_BNO;
2604                 args.total = ap->total;
2605                 /*
2606                  * Find the longest available space.
2607                  * We're going to try for the whole allocation at once.
2608                  */
2609                 startag = ag = XFS_FSB_TO_AGNO(mp, args.fsbno);
2610                 notinit = 0;
2611                 down_read(&mp->m_peraglock);
2612                 while (blen < ap->alen) {
2613                         pag = &mp->m_perag[ag];
2614                         if (!pag->pagf_init &&
2615                             (error = xfs_alloc_pagf_init(mp, args.tp,
2616                                     ag, XFS_ALLOC_FLAG_TRYLOCK))) {
2617                                 up_read(&mp->m_peraglock);
2618                                 return error;
2619                         }
2620                         /*
2621                          * See xfs_alloc_fix_freelist...
2622                          */
2623                         if (pag->pagf_init) {
2624                                 need = XFS_MIN_FREELIST_PAG(pag, mp);
2625                                 delta = need > pag->pagf_flcount ?
2626                                         need - pag->pagf_flcount : 0;
2627                                 longest = (pag->pagf_longest > delta) ?
2628                                         (pag->pagf_longest - delta) :
2629                                         (pag->pagf_flcount > 0 ||
2630                                          pag->pagf_longest > 0);
2631                                 if (blen < longest)
2632                                         blen = longest;
2633                         } else
2634                                 notinit = 1;
2635                         if (++ag == mp->m_sb.sb_agcount)
2636                                 ag = 0;
2637                         if (ag == startag)
2638                                 break;
2639                 }
2640                 up_read(&mp->m_peraglock);
2641                 /*
2642                  * Since the above loop did a BUF_TRYLOCK, it is
2643                  * possible that there is space for this request.
2644                  */
2645                 if (notinit || blen < ap->minlen)
2646                         args.minlen = ap->minlen;
2647                 /*
2648                  * If the best seen length is less than the request
2649                  * length, use the best as the minimum.
2650                  */
2651                 else if (blen < ap->alen)
2652                         args.minlen = blen;
2653                 /*
2654                  * Otherwise we've seen an extent as big as alen,
2655                  * use that as the minimum.
2656                  */
2657                 else
2658                         args.minlen = ap->alen;
2659         } else if (ap->low) {
2660                 args.type = XFS_ALLOCTYPE_FIRST_AG;
2661                 args.total = args.minlen = ap->minlen;
2662         } else {
2663                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2664                 args.total = ap->total;
2665                 args.minlen = ap->minlen;
2666         }
2667         if (unlikely(ap->userdata && ap->ip->i_d.di_extsize &&
2668                     (ap->ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE))) {
2669                 args.prod = ap->ip->i_d.di_extsize;
2670                 if ((args.mod = (xfs_extlen_t)do_mod(ap->off, args.prod)))
2671                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2672         } else if (unlikely(mp->m_sb.sb_blocksize >= NBPP)) {
2673                 args.prod = 1;
2674                 args.mod = 0;
2675         } else {
2676                 args.prod = NBPP >> mp->m_sb.sb_blocklog;
2677                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->off, args.prod))))
2678                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2679         }
2680         /*
2681          * If we are not low on available data blocks, and the
2682          * underlying logical volume manager is a stripe, and
2683          * the file offset is zero then try to allocate data
2684          * blocks on stripe unit boundary.
2685          * NOTE: ap->aeof is only set if the allocation length
2686          * is >= the stripe unit and the allocation offset is
2687          * at the end of file.
2688          */
2689         if (!ap->low && ap->aeof) {
2690                 if (!ap->off) {
2691                         args.alignment = mp->m_dalign;
2692                         atype = args.type;
2693                         isaligned = 1;
2694                         /*
2695                          * Adjust for alignment
2696                          */
2697                         if (blen > args.alignment && blen <= ap->alen)
2698                                 args.minlen = blen - args.alignment;
2699                         args.minalignslop = 0;
2700                 } else {
2701                         /*
2702                          * First try an exact bno allocation.
2703                          * If it fails then do a near or start bno
2704                          * allocation with alignment turned on.
2705                          */
2706                         atype = args.type;
2707                         tryagain = 1;
2708                         args.type = XFS_ALLOCTYPE_THIS_BNO;
2709                         args.alignment = 1;
2710                         /*
2711                          * Compute the minlen+alignment for the
2712                          * next case.  Set slop so that the value
2713                          * of minlen+alignment+slop doesn't go up
2714                          * between the calls.
2715                          */
2716                         if (blen > mp->m_dalign && blen <= ap->alen)
2717                                 nextminlen = blen - mp->m_dalign;
2718                         else
2719                                 nextminlen = args.minlen;
2720                         if (nextminlen + mp->m_dalign > args.minlen + 1)
2721                                 args.minalignslop =
2722                                         nextminlen + mp->m_dalign -
2723                                         args.minlen - 1;
2724                         else
2725                                 args.minalignslop = 0;
2726                 }
2727         } else {
2728                 args.alignment = 1;
2729                 args.minalignslop = 0;
2730         }
2731         args.minleft = ap->minleft;
2732         args.wasdel = ap->wasdel;
2733         args.isfl = 0;
2734         args.userdata = ap->userdata;
2735         if ((error = xfs_alloc_vextent(&args)))
2736                 return error;
2737         if (tryagain && args.fsbno == NULLFSBLOCK) {
2738                 /*
2739                  * Exact allocation failed. Now try with alignment
2740                  * turned on.
2741                  */
2742                 args.type = atype;
2743                 args.fsbno = ap->rval;
2744                 args.alignment = mp->m_dalign;
2745                 args.minlen = nextminlen;
2746                 args.minalignslop = 0;
2747                 isaligned = 1;
2748                 if ((error = xfs_alloc_vextent(&args)))
2749                         return error;
2750         }
2751         if (isaligned && args.fsbno == NULLFSBLOCK) {
2752                 /*
2753                  * allocation failed, so turn off alignment and
2754                  * try again.
2755                  */
2756                 args.type = atype;
2757                 args.fsbno = ap->rval;
2758                 args.alignment = 0;
2759                 if ((error = xfs_alloc_vextent(&args)))
2760                         return error;
2761         }
2762         if (args.fsbno == NULLFSBLOCK && nullfb &&
2763             args.minlen > ap->minlen) {
2764                 args.minlen = ap->minlen;
2765                 args.type = XFS_ALLOCTYPE_START_BNO;
2766                 args.fsbno = ap->rval;
2767                 if ((error = xfs_alloc_vextent(&args)))
2768                         return error;
2769         }
2770         if (args.fsbno == NULLFSBLOCK && nullfb) {
2771                 args.fsbno = 0;
2772                 args.type = XFS_ALLOCTYPE_FIRST_AG;
2773                 args.total = ap->minlen;
2774                 args.minleft = 0;
2775                 if ((error = xfs_alloc_vextent(&args)))
2776                         return error;
2777                 ap->low = 1;
2778         }
2779         if (args.fsbno != NULLFSBLOCK) {
2780                 ap->firstblock = ap->rval = args.fsbno;
2781                 ASSERT(nullfb || fb_agno == args.agno ||
2782                        (ap->low && fb_agno < args.agno));
2783                 ap->alen = args.len;
2784                 ap->ip->i_d.di_nblocks += args.len;
2785                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2786                 if (ap->wasdel)
2787                         ap->ip->i_delayed_blks -= args.len;
2788                 /*
2789                  * Adjust the disk quota also. This was reserved
2790                  * earlier.
2791                  */
2792                 XFS_TRANS_MOD_DQUOT_BYINO(mp, ap->tp, ap->ip,
2793                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2794                                         XFS_TRANS_DQ_BCOUNT,
2795                         (long) args.len);
2796         } else {
2797                 ap->rval = NULLFSBLOCK;
2798                 ap->alen = 0;
2799         }
2800         return 0;
2801 }
2802
2803 /*
2804  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2805  * It figures out where to ask the underlying allocator to put the new extent.
2806  */
2807 STATIC int
2808 xfs_bmap_alloc(
2809         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2810 {
2811         if ((ap->ip->i_d.di_flags & XFS_DIFLAG_REALTIME) && ap->userdata)
2812                 return xfs_bmap_rtalloc(ap);
2813         return xfs_bmap_btalloc(ap);
2814 }
2815
2816 /*
2817  * Transform a btree format file with only one leaf node, where the
2818  * extents list will fit in the inode, into an extents format file.
2819  * Since the file extents are already in-core, all we have to do is
2820  * give up the space for the btree root and pitch the leaf block.
2821  */
2822 STATIC int                              /* error */
2823 xfs_bmap_btree_to_extents(
2824         xfs_trans_t             *tp,    /* transaction pointer */
2825         xfs_inode_t             *ip,    /* incore inode pointer */
2826         xfs_btree_cur_t         *cur,   /* btree cursor */
2827         int                     *logflagsp, /* inode logging flags */
2828         int                     whichfork)  /* data or attr fork */
2829 {
2830         /* REFERENCED */
2831         xfs_bmbt_block_t        *cblock;/* child btree block */
2832         xfs_fsblock_t           cbno;   /* child block number */
2833         xfs_buf_t               *cbp;   /* child block's buffer */
2834         int                     error;  /* error return value */
2835         xfs_ifork_t             *ifp;   /* inode fork data */
2836         xfs_mount_t             *mp;    /* mount point structure */
2837         xfs_bmbt_ptr_t          *pp;    /* ptr to block address */
2838         xfs_bmbt_block_t        *rblock;/* root btree block */
2839
2840         ifp = XFS_IFORK_PTR(ip, whichfork);
2841         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2842         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
2843         rblock = ifp->if_broot;
2844         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
2845         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
2846         ASSERT(XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes) == 1);
2847         mp = ip->i_mount;
2848         pp = XFS_BMAP_BROOT_PTR_ADDR(rblock, 1, ifp->if_broot_bytes);
2849         *logflagsp = 0;
2850 #ifdef DEBUG
2851         if ((error = xfs_btree_check_lptr(cur, INT_GET(*pp, ARCH_CONVERT), 1)))
2852                 return error;
2853 #endif
2854         cbno = INT_GET(*pp, ARCH_CONVERT);
2855         if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
2856                         XFS_BMAP_BTREE_REF)))
2857                 return error;
2858         cblock = XFS_BUF_TO_BMBT_BLOCK(cbp);
2859         if ((error = xfs_btree_check_lblock(cur, cblock, 0, cbp)))
2860                 return error;
2861         xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
2862         ip->i_d.di_nblocks--;
2863         XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
2864         xfs_trans_binval(tp, cbp);
2865         if (cur->bc_bufs[0] == cbp)
2866                 cur->bc_bufs[0] = NULL;
2867         xfs_iroot_realloc(ip, -1, whichfork);
2868         ASSERT(ifp->if_broot == NULL);
2869         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
2870         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
2871         *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2872         return 0;
2873 }
2874
2875 /*
2876  * Called by xfs_bmapi to update file extent records and the btree
2877  * after removing space (or undoing a delayed allocation).
2878  */
2879 STATIC int                              /* error */
2880 xfs_bmap_del_extent(
2881         xfs_inode_t             *ip,    /* incore inode pointer */
2882         xfs_trans_t             *tp,    /* current transaction pointer */
2883         xfs_extnum_t            idx,    /* extent number to update/delete */
2884         xfs_bmap_free_t         *flist, /* list of extents to be freed */
2885         xfs_btree_cur_t         *cur,   /* if null, not a btree */
2886         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
2887         int                     *logflagsp, /* inode logging flags */
2888         int                     whichfork, /* data or attr fork */
2889         int                     rsvd)   /* OK to allocate reserved blocks */
2890 {
2891         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
2892         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
2893         xfs_fsblock_t           del_endblock=0; /* first block past del */
2894         xfs_fileoff_t           del_endoff;     /* first offset past del */
2895         int                     delay;  /* current block is delayed allocated */
2896         int                     do_fx;  /* free extent at end of routine */
2897         xfs_bmbt_rec_t          *ep;    /* current extent entry pointer */
2898         int                     error;  /* error return value */
2899         int                     flags;  /* inode logging flags */
2900 #ifdef XFS_BMAP_TRACE
2901         static char             fname[] = "xfs_bmap_del_extent";
2902 #endif
2903         xfs_bmbt_irec_t         got;    /* current extent entry */
2904         xfs_fileoff_t           got_endoff;     /* first offset past got */
2905         int                     i;      /* temp state */
2906         xfs_ifork_t             *ifp;   /* inode fork pointer */
2907         xfs_mount_t             *mp;    /* mount structure */
2908         xfs_filblks_t           nblks;  /* quota/sb block count */
2909         xfs_bmbt_irec_t         new;    /* new record to be inserted */
2910         /* REFERENCED */
2911         uint                    qfield; /* quota field to update */
2912         xfs_filblks_t           temp;   /* for indirect length calculations */
2913         xfs_filblks_t           temp2;  /* for indirect length calculations */
2914
2915         XFS_STATS_INC(xs_del_exlist);
2916         mp = ip->i_mount;
2917         ifp = XFS_IFORK_PTR(ip, whichfork);
2918         ASSERT((idx >= 0) && (idx < ifp->if_bytes /
2919                 (uint)sizeof(xfs_bmbt_rec_t)));
2920         ASSERT(del->br_blockcount > 0);
2921         ep = xfs_iext_get_ext(ifp, idx);
2922         xfs_bmbt_get_all(ep, &got);
2923         ASSERT(got.br_startoff <= del->br_startoff);
2924         del_endoff = del->br_startoff + del->br_blockcount;
2925         got_endoff = got.br_startoff + got.br_blockcount;
2926         ASSERT(got_endoff >= del_endoff);
2927         delay = ISNULLSTARTBLOCK(got.br_startblock);
2928         ASSERT(ISNULLSTARTBLOCK(del->br_startblock) == delay);
2929         flags = 0;
2930         qfield = 0;
2931         error = 0;
2932         /*
2933          * If deleting a real allocation, must free up the disk space.
2934          */
2935         if (!delay) {
2936                 flags = XFS_ILOG_CORE;
2937                 /*
2938                  * Realtime allocation.  Free it and record di_nblocks update.
2939                  */
2940                 if (whichfork == XFS_DATA_FORK &&
2941                     (ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
2942                         xfs_fsblock_t   bno;
2943                         xfs_filblks_t   len;
2944
2945                         ASSERT(do_mod(del->br_blockcount,
2946                                       mp->m_sb.sb_rextsize) == 0);
2947                         ASSERT(do_mod(del->br_startblock,
2948                                       mp->m_sb.sb_rextsize) == 0);
2949                         bno = del->br_startblock;
2950                         len = del->br_blockcount;
2951                         do_div(bno, mp->m_sb.sb_rextsize);
2952                         do_div(len, mp->m_sb.sb_rextsize);
2953                         if ((error = xfs_rtfree_extent(ip->i_transp, bno,
2954                                         (xfs_extlen_t)len)))
2955                                 goto done;
2956                         do_fx = 0;
2957                         nblks = len * mp->m_sb.sb_rextsize;
2958                         qfield = XFS_TRANS_DQ_RTBCOUNT;
2959                 }
2960                 /*
2961                  * Ordinary allocation.
2962                  */
2963                 else {
2964                         do_fx = 1;
2965                         nblks = del->br_blockcount;
2966                         qfield = XFS_TRANS_DQ_BCOUNT;
2967                 }
2968                 /*
2969                  * Set up del_endblock and cur for later.
2970                  */
2971                 del_endblock = del->br_startblock + del->br_blockcount;
2972                 if (cur) {
2973                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
2974                                         got.br_startblock, got.br_blockcount,
2975                                         &i)))
2976                                 goto done;
2977                         ASSERT(i == 1);
2978                 }
2979                 da_old = da_new = 0;
2980         } else {
2981                 da_old = STARTBLOCKVAL(got.br_startblock);
2982                 da_new = 0;
2983                 nblks = 0;
2984                 do_fx = 0;
2985         }
2986         /*
2987          * Set flag value to use in switch statement.
2988          * Left-contig is 2, right-contig is 1.
2989          */
2990         switch (((got.br_startoff == del->br_startoff) << 1) |
2991                 (got_endoff == del_endoff)) {
2992         case 3:
2993                 /*
2994                  * Matches the whole extent.  Delete the entry.
2995                  */
2996                 xfs_bmap_trace_delete(fname, "3", ip, idx, 1, whichfork);
2997                 xfs_iext_remove(ifp, idx, 1);
2998                 ifp->if_lastex = idx;
2999                 if (delay)
3000                         break;
3001                 XFS_IFORK_NEXT_SET(ip, whichfork,
3002                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
3003                 flags |= XFS_ILOG_CORE;
3004                 if (!cur) {
3005                         flags |= XFS_ILOG_FEXT(whichfork);
3006                         break;
3007                 }
3008                 if ((error = xfs_bmbt_delete(cur, &i)))
3009                         goto done;
3010                 ASSERT(i == 1);
3011                 break;
3012
3013         case 2:
3014                 /*
3015                  * Deleting the first part of the extent.
3016                  */
3017                 xfs_bmap_trace_pre_update(fname, "2", ip, idx, whichfork);
3018                 xfs_bmbt_set_startoff(ep, del_endoff);
3019                 temp = got.br_blockcount - del->br_blockcount;
3020                 xfs_bmbt_set_blockcount(ep, temp);
3021                 ifp->if_lastex = idx;
3022                 if (delay) {
3023                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
3024                                 da_old);
3025                         xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3026                         xfs_bmap_trace_post_update(fname, "2", ip, idx,
3027                                 whichfork);
3028                         da_new = temp;
3029                         break;
3030                 }
3031                 xfs_bmbt_set_startblock(ep, del_endblock);
3032                 xfs_bmap_trace_post_update(fname, "2", ip, idx, whichfork);
3033                 if (!cur) {
3034                         flags |= XFS_ILOG_FEXT(whichfork);
3035                         break;
3036                 }
3037                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
3038                                 got.br_blockcount - del->br_blockcount,
3039                                 got.br_state)))
3040                         goto done;
3041                 break;
3042
3043         case 1:
3044                 /*
3045                  * Deleting the last part of the extent.
3046                  */
3047                 temp = got.br_blockcount - del->br_blockcount;
3048                 xfs_bmap_trace_pre_update(fname, "1", ip, idx, whichfork);
3049                 xfs_bmbt_set_blockcount(ep, temp);
3050                 ifp->if_lastex = idx;
3051                 if (delay) {
3052                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
3053                                 da_old);
3054                         xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3055                         xfs_bmap_trace_post_update(fname, "1", ip, idx,
3056                                 whichfork);
3057                         da_new = temp;
3058                         break;
3059                 }
3060                 xfs_bmap_trace_post_update(fname, "1", ip, idx, whichfork);
3061                 if (!cur) {
3062                         flags |= XFS_ILOG_FEXT(whichfork);
3063                         break;
3064                 }
3065                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
3066                                 got.br_startblock,
3067                                 got.br_blockcount - del->br_blockcount,
3068                                 got.br_state)))
3069                         goto done;
3070                 break;
3071
3072         case 0:
3073                 /*
3074                  * Deleting the middle of the extent.
3075                  */
3076                 temp = del->br_startoff - got.br_startoff;
3077                 xfs_bmap_trace_pre_update(fname, "0", ip, idx, whichfork);
3078                 xfs_bmbt_set_blockcount(ep, temp);
3079                 new.br_startoff = del_endoff;
3080                 temp2 = got_endoff - del_endoff;
3081                 new.br_blockcount = temp2;
3082                 new.br_state = got.br_state;
3083                 if (!delay) {
3084                         new.br_startblock = del_endblock;
3085                         flags |= XFS_ILOG_CORE;
3086                         if (cur) {
3087                                 if ((error = xfs_bmbt_update(cur,
3088                                                 got.br_startoff,
3089                                                 got.br_startblock, temp,
3090                                                 got.br_state)))
3091                                         goto done;
3092                                 if ((error = xfs_bmbt_increment(cur, 0, &i)))
3093                                         goto done;
3094                                 cur->bc_rec.b = new;
3095                                 error = xfs_bmbt_insert(cur, &i);
3096                                 if (error && error != ENOSPC)
3097                                         goto done;
3098                                 /*
3099                                  * If get no-space back from btree insert,
3100                                  * it tried a split, and we have a zero
3101                                  * block reservation.
3102                                  * Fix up our state and return the error.
3103                                  */
3104                                 if (error == ENOSPC) {
3105                                         /*
3106                                          * Reset the cursor, don't trust
3107                                          * it after any insert operation.
3108                                          */
3109                                         if ((error = xfs_bmbt_lookup_eq(cur,
3110                                                         got.br_startoff,
3111                                                         got.br_startblock,
3112                                                         temp, &i)))
3113                                                 goto done;
3114                                         ASSERT(i == 1);
3115                                         /*
3116                                          * Update the btree record back
3117                                          * to the original value.
3118                                          */
3119                                         if ((error = xfs_bmbt_update(cur,
3120                                                         got.br_startoff,
3121                                                         got.br_startblock,
3122                                                         got.br_blockcount,
3123                                                         got.br_state)))
3124                                                 goto done;
3125                                         /*
3126                                          * Reset the extent record back
3127                                          * to the original value.
3128                                          */
3129                                         xfs_bmbt_set_blockcount(ep,
3130                                                 got.br_blockcount);
3131                                         flags = 0;
3132                                         error = XFS_ERROR(ENOSPC);
3133                                         goto done;
3134                                 }
3135                                 ASSERT(i == 1);
3136                         } else
3137                                 flags |= XFS_ILOG_FEXT(whichfork);
3138                         XFS_IFORK_NEXT_SET(ip, whichfork,
3139                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
3140                 } else {
3141                         ASSERT(whichfork == XFS_DATA_FORK);
3142                         temp = xfs_bmap_worst_indlen(ip, temp);
3143                         xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3144                         temp2 = xfs_bmap_worst_indlen(ip, temp2);
3145                         new.br_startblock = NULLSTARTBLOCK((int)temp2);
3146                         da_new = temp + temp2;
3147                         while (da_new > da_old) {
3148                                 if (temp) {
3149                                         temp--;
3150                                         da_new--;
3151                                         xfs_bmbt_set_startblock(ep,
3152                                                 NULLSTARTBLOCK((int)temp));
3153                                 }
3154                                 if (da_new == da_old)
3155                                         break;
3156                                 if (temp2) {
3157                                         temp2--;
3158                                         da_new--;
3159                                         new.br_startblock =
3160                                                 NULLSTARTBLOCK((int)temp2);
3161                                 }
3162                         }
3163                 }
3164                 xfs_bmap_trace_post_update(fname, "0", ip, idx, whichfork);
3165                 xfs_bmap_trace_insert(fname, "0", ip, idx + 1, 1, &new, NULL,
3166                         whichfork);
3167                 xfs_iext_insert(ifp, idx + 1, 1, &new);
3168                 ifp->if_lastex = idx + 1;
3169                 break;
3170         }
3171         /*
3172          * If we need to, add to list of extents to delete.
3173          */
3174         if (do_fx)
3175                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
3176                         mp);
3177         /*
3178          * Adjust inode # blocks in the file.
3179          */
3180         if (nblks)
3181                 ip->i_d.di_nblocks -= nblks;
3182         /*
3183          * Adjust quota data.
3184          */
3185         if (qfield)
3186                 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, qfield, (long)-nblks);
3187
3188         /*
3189          * Account for change in delayed indirect blocks.
3190          * Nothing to do for disk quota accounting here.
3191          */
3192         ASSERT(da_old >= da_new);
3193         if (da_old > da_new)
3194                 xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS, (int)(da_old - da_new),
3195                         rsvd);
3196 done:
3197         *logflagsp = flags;
3198         return error;
3199 }
3200
3201 /*
3202  * Remove the entry "free" from the free item list.  Prev points to the
3203  * previous entry, unless "free" is the head of the list.
3204  */
3205 STATIC void
3206 xfs_bmap_del_free(
3207         xfs_bmap_free_t         *flist, /* free item list header */
3208         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
3209         xfs_bmap_free_item_t    *free)  /* list item to be freed */
3210 {
3211         if (prev)
3212                 prev->xbfi_next = free->xbfi_next;
3213         else
3214                 flist->xbf_first = free->xbfi_next;
3215         flist->xbf_count--;
3216         kmem_zone_free(xfs_bmap_free_item_zone, free);
3217 }
3218
3219 /*
3220  * Convert an extents-format file into a btree-format file.
3221  * The new file will have a root block (in the inode) and a single child block.
3222  */
3223 STATIC int                                      /* error */
3224 xfs_bmap_extents_to_btree(
3225         xfs_trans_t             *tp,            /* transaction pointer */
3226         xfs_inode_t             *ip,            /* incore inode pointer */
3227         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
3228         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
3229         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
3230         int                     wasdel,         /* converting a delayed alloc */
3231         int                     *logflagsp,     /* inode logging flags */
3232         int                     whichfork)      /* data or attr fork */
3233 {
3234         xfs_bmbt_block_t        *ablock;        /* allocated (child) bt block */
3235         xfs_buf_t               *abp;           /* buffer for ablock */
3236         xfs_alloc_arg_t         args;           /* allocation arguments */
3237         xfs_bmbt_rec_t          *arp;           /* child record pointer */
3238         xfs_bmbt_block_t        *block;         /* btree root block */
3239         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
3240         xfs_bmbt_rec_t          *ep;            /* extent record pointer */
3241         int                     error;          /* error return value */
3242         xfs_extnum_t            i, cnt;         /* extent record index */
3243         xfs_ifork_t             *ifp;           /* inode fork pointer */
3244         xfs_bmbt_key_t          *kp;            /* root block key pointer */
3245         xfs_mount_t             *mp;            /* mount structure */
3246         xfs_extnum_t            nextents;       /* number of file extents */
3247         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
3248
3249         ifp = XFS_IFORK_PTR(ip, whichfork);
3250         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3251         ASSERT(ifp->if_ext_max ==
3252                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
3253         /*
3254          * Make space in the inode incore.
3255          */
3256         xfs_iroot_realloc(ip, 1, whichfork);
3257         ifp->if_flags |= XFS_IFBROOT;
3258         /*
3259          * Fill in the root.
3260          */
3261         block = ifp->if_broot;
3262         block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3263         block->bb_level = cpu_to_be16(1);
3264         block->bb_numrecs = cpu_to_be16(1);
3265         block->bb_leftsib = cpu_to_be64(NULLDFSBNO);
3266         block->bb_rightsib = cpu_to_be64(NULLDFSBNO);
3267         /*
3268          * Need a cursor.  Can't allocate until bb_level is filled in.
3269          */
3270         mp = ip->i_mount;
3271         cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
3272                 whichfork);
3273         cur->bc_private.b.firstblock = *firstblock;
3274         cur->bc_private.b.flist = flist;
3275         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3276         /*
3277          * Convert to a btree with two levels, one record in root.
3278          */
3279         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3280         args.tp = tp;
3281         args.mp = mp;
3282         if (*firstblock == NULLFSBLOCK) {
3283                 args.type = XFS_ALLOCTYPE_START_BNO;
3284                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3285         } else if (flist->xbf_low) {
3286                 args.type = XFS_ALLOCTYPE_START_BNO;
3287                 args.fsbno = *firstblock;
3288         } else {
3289                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3290                 args.fsbno = *firstblock;
3291         }
3292         args.minlen = args.maxlen = args.prod = 1;
3293         args.total = args.minleft = args.alignment = args.mod = args.isfl =
3294                 args.minalignslop = 0;
3295         args.wasdel = wasdel;
3296         *logflagsp = 0;
3297         if ((error = xfs_alloc_vextent(&args))) {
3298                 xfs_iroot_realloc(ip, -1, whichfork);
3299                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3300                 return error;
3301         }
3302         /*
3303          * Allocation can't fail, the space was reserved.
3304          */
3305         ASSERT(args.fsbno != NULLFSBLOCK);
3306         ASSERT(*firstblock == NULLFSBLOCK ||
3307                args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3308                (flist->xbf_low &&
3309                 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3310         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3311         cur->bc_private.b.allocated++;
3312         ip->i_d.di_nblocks++;
3313         XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3314         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3315         /*
3316          * Fill in the child block.
3317          */
3318         ablock = XFS_BUF_TO_BMBT_BLOCK(abp);
3319         ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3320         ablock->bb_level = 0;
3321         ablock->bb_leftsib = cpu_to_be64(NULLDFSBNO);
3322         ablock->bb_rightsib = cpu_to_be64(NULLDFSBNO);
3323         arp = XFS_BMAP_REC_IADDR(ablock, 1, cur);
3324         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3325         for (cnt = i = 0; i < nextents; i++) {
3326                 ep = xfs_iext_get_ext(ifp, i);
3327                 if (!ISNULLSTARTBLOCK(xfs_bmbt_get_startblock(ep))) {
3328                         arp->l0 = INT_GET(ep->l0, ARCH_CONVERT);
3329                         arp->l1 = INT_GET(ep->l1, ARCH_CONVERT);
3330                         arp++; cnt++;
3331                 }
3332         }
3333         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3334         ablock->bb_numrecs = cpu_to_be16(cnt);
3335         /*
3336          * Fill in the root key and pointer.
3337          */
3338         kp = XFS_BMAP_KEY_IADDR(block, 1, cur);
3339         arp = XFS_BMAP_REC_IADDR(ablock, 1, cur);
3340         INT_SET(kp->br_startoff, ARCH_CONVERT, xfs_bmbt_disk_get_startoff(arp));
3341         pp = XFS_BMAP_PTR_IADDR(block, 1, cur);
3342         INT_SET(*pp, ARCH_CONVERT, args.fsbno);
3343         /*
3344          * Do all this logging at the end so that
3345          * the root is at the right level.
3346          */
3347         xfs_bmbt_log_block(cur, abp, XFS_BB_ALL_BITS);
3348         xfs_bmbt_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3349         ASSERT(*curp == NULL);
3350         *curp = cur;
3351         *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FBROOT(whichfork);
3352         return 0;
3353 }
3354
3355 /*
3356  * Helper routine to reset inode di_forkoff field when switching
3357  * attribute fork from local to extent format - we reset it where
3358  * possible to make space available for inline data fork extents.
3359  */
3360 STATIC void
3361 xfs_bmap_forkoff_reset(
3362         xfs_mount_t     *mp,
3363         xfs_inode_t     *ip,
3364         int             whichfork)
3365 {
3366         if (whichfork == XFS_ATTR_FORK &&
3367             (ip->i_d.di_format != XFS_DINODE_FMT_DEV) &&
3368             (ip->i_d.di_format != XFS_DINODE_FMT_UUID) &&
3369             ((mp->m_attroffset >> 3) > ip->i_d.di_forkoff)) {
3370                 ip->i_d.di_forkoff = mp->m_attroffset >> 3;
3371                 ip->i_df.if_ext_max = XFS_IFORK_DSIZE(ip) /
3372                                         (uint)sizeof(xfs_bmbt_rec_t);
3373                 ip->i_afp->if_ext_max = XFS_IFORK_ASIZE(ip) /
3374                                         (uint)sizeof(xfs_bmbt_rec_t);
3375         }
3376 }
3377
3378 /*
3379  * Convert a local file to an extents file.
3380  * This code is out of bounds for data forks of regular files,
3381  * since the file data needs to get logged so things will stay consistent.
3382  * (The bmap-level manipulations are ok, though).
3383  */
3384 STATIC int                              /* error */
3385 xfs_bmap_local_to_extents(
3386         xfs_trans_t     *tp,            /* transaction pointer */
3387         xfs_inode_t     *ip,            /* incore inode pointer */
3388         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
3389         xfs_extlen_t    total,          /* total blocks needed by transaction */
3390         int             *logflagsp,     /* inode logging flags */
3391         int             whichfork)      /* data or attr fork */
3392 {
3393         int             error;          /* error return value */
3394         int             flags;          /* logging flags returned */
3395 #ifdef XFS_BMAP_TRACE
3396         static char     fname[] = "xfs_bmap_local_to_extents";
3397 #endif
3398         xfs_ifork_t     *ifp;           /* inode fork pointer */
3399
3400         /*
3401          * We don't want to deal with the case of keeping inode data inline yet.
3402          * So sending the data fork of a regular inode is invalid.
3403          */
3404         ASSERT(!((ip->i_d.di_mode & S_IFMT) == S_IFREG &&
3405                  whichfork == XFS_DATA_FORK));
3406         ifp = XFS_IFORK_PTR(ip, whichfork);
3407         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3408         flags = 0;
3409         error = 0;
3410         if (ifp->if_bytes) {
3411                 xfs_alloc_arg_t args;   /* allocation arguments */
3412                 xfs_buf_t       *bp;    /* buffer for extent block */
3413                 xfs_bmbt_rec_t  *ep;    /* extent record pointer */
3414
3415                 args.tp = tp;
3416                 args.mp = ip->i_mount;
3417                 ASSERT((ifp->if_flags &
3418                         (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == XFS_IFINLINE);
3419                 /*
3420                  * Allocate a block.  We know we need only one, since the
3421                  * file currently fits in an inode.
3422                  */
3423                 if (*firstblock == NULLFSBLOCK) {
3424                         args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3425                         args.type = XFS_ALLOCTYPE_START_BNO;
3426                 } else {
3427                         args.fsbno = *firstblock;
3428                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
3429                 }
3430                 args.total = total;
3431                 args.mod = args.minleft = args.alignment = args.wasdel =
3432                         args.isfl = args.minalignslop = 0;
3433                 args.minlen = args.maxlen = args.prod = 1;
3434                 if ((error = xfs_alloc_vextent(&args)))
3435                         goto done;
3436                 /*
3437                  * Can't fail, the space was reserved.
3438                  */
3439                 ASSERT(args.fsbno != NULLFSBLOCK);
3440                 ASSERT(args.len == 1);
3441                 *firstblock = args.fsbno;
3442                 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3443                 memcpy((char *)XFS_BUF_PTR(bp), ifp->if_u1.if_data,
3444                         ifp->if_bytes);
3445                 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3446                 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3447                 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3448                 xfs_iext_add(ifp, 0, 1);
3449                 ep = xfs_iext_get_ext(ifp, 0);
3450                 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3451                 xfs_bmap_trace_post_update(fname, "new", ip, 0, whichfork);
3452                 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3453                 ip->i_d.di_nblocks = 1;
3454                 XFS_TRANS_MOD_DQUOT_BYINO(args.mp, tp, ip,
3455                         XFS_TRANS_DQ_BCOUNT, 1L);
3456                 flags |= XFS_ILOG_FEXT(whichfork);
3457         } else {
3458                 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3459                 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3460         }
3461         ifp->if_flags &= ~XFS_IFINLINE;
3462         ifp->if_flags |= XFS_IFEXTENTS;
3463         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3464         flags |= XFS_ILOG_CORE;
3465 done:
3466         *logflagsp = flags;
3467         return error;
3468 }
3469
3470 /*
3471  * Search the extent records for the entry containing block bno.
3472  * If bno lies in a hole, point to the next entry.  If bno lies
3473  * past eof, *eofp will be set, and *prevp will contain the last
3474  * entry (null if none).  Else, *lastxp will be set to the index
3475  * of the found entry; *gotp will contain the entry.
3476  */
3477 xfs_bmbt_rec_t *                        /* pointer to found extent entry */
3478 xfs_bmap_search_multi_extents(
3479         xfs_ifork_t     *ifp,           /* inode fork pointer */
3480         xfs_fileoff_t   bno,            /* block number searched for */
3481         int             *eofp,          /* out: end of file found */
3482         xfs_extnum_t    *lastxp,        /* out: last extent index */
3483         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3484         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3485 {
3486         xfs_bmbt_rec_t  *ep;            /* extent record pointer */
3487         xfs_extnum_t    lastx;          /* last extent index */
3488
3489         /*
3490          * Initialize the extent entry structure to catch access to
3491          * uninitialized br_startblock field.
3492          */
3493         gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
3494         gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3495         gotp->br_state = XFS_EXT_INVALID;
3496 #if XFS_BIG_BLKNOS
3497         gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
3498 #else
3499         gotp->br_startblock = 0xffffa5a5;
3500 #endif
3501         prevp->br_startoff = NULLFILEOFF;
3502
3503         ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
3504         if (lastx > 0) {
3505                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
3506         }
3507         if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
3508                 xfs_bmbt_get_all(ep, gotp);
3509                 *eofp = 0;
3510         } else {
3511                 if (lastx > 0) {
3512                         *gotp = *prevp;
3513                 }
3514                 *eofp = 1;
3515                 ep = NULL;
3516         }
3517         *lastxp = lastx;
3518         return ep;
3519 }
3520
3521 /*
3522  * Search the extents list for the inode, for the extent containing bno.
3523  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
3524  * *eofp will be set, and *prevp will contain the last entry (null if none).
3525  * Else, *lastxp will be set to the index of the found
3526  * entry; *gotp will contain the entry.
3527  */
3528 STATIC xfs_bmbt_rec_t *                 /* pointer to found extent entry */
3529 xfs_bmap_search_extents(
3530         xfs_inode_t     *ip,            /* incore inode pointer */
3531         xfs_fileoff_t   bno,            /* block number searched for */
3532         int             whichfork,      /* data or attr fork */
3533         int             *eofp,          /* out: end of file found */
3534         xfs_extnum_t    *lastxp,        /* out: last extent index */
3535         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3536         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3537 {
3538         xfs_ifork_t     *ifp;           /* inode fork pointer */
3539         xfs_bmbt_rec_t  *ep;            /* extent record pointer */
3540         int             rt;             /* realtime flag    */
3541
3542         XFS_STATS_INC(xs_look_exlist);
3543         ifp = XFS_IFORK_PTR(ip, whichfork);
3544
3545         ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
3546
3547         rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
3548         if (unlikely(!rt && !gotp->br_startblock && (*lastxp != NULLEXTNUM))) {
3549                 cmn_err(CE_PANIC,"Access to block zero: fs: <%s> inode: %lld "
3550                         "start_block : %llx start_off : %llx blkcnt : %llx "
3551                         "extent-state : %x \n",
3552                         (ip->i_mount)->m_fsname, (long long)ip->i_ino,
3553                         (unsigned long long)gotp->br_startblock,
3554                         (unsigned long long)gotp->br_startoff,
3555                         (unsigned long long)gotp->br_blockcount,
3556                         gotp->br_state);
3557         }
3558         return ep;
3559 }
3560
3561
3562 #ifdef XFS_BMAP_TRACE
3563 ktrace_t        *xfs_bmap_trace_buf;
3564
3565 /*
3566  * Add a bmap trace buffer entry.  Base routine for the others.
3567  */
3568 STATIC void
3569 xfs_bmap_trace_addentry(
3570         int             opcode,         /* operation */
3571         char            *fname,         /* function name */
3572         char            *desc,          /* operation description */
3573         xfs_inode_t     *ip,            /* incore inode pointer */
3574         xfs_extnum_t    idx,            /* index of entry(ies) */
3575         xfs_extnum_t    cnt,            /* count of entries, 1 or 2 */
3576         xfs_bmbt_rec_t  *r1,            /* first record */
3577         xfs_bmbt_rec_t  *r2,            /* second record or null */
3578         int             whichfork)      /* data or attr fork */
3579 {
3580         xfs_bmbt_rec_t  tr2;
3581
3582         ASSERT(cnt == 1 || cnt == 2);
3583         ASSERT(r1 != NULL);
3584         if (cnt == 1) {
3585                 ASSERT(r2 == NULL);
3586                 r2 = &tr2;
3587                 memset(&tr2, 0, sizeof(tr2));
3588         } else
3589                 ASSERT(r2 != NULL);
3590         ktrace_enter(xfs_bmap_trace_buf,
3591                 (void *)(__psint_t)(opcode | (whichfork << 16)),
3592                 (void *)fname, (void *)desc, (void *)ip,
3593                 (void *)(__psint_t)idx,
3594                 (void *)(__psint_t)cnt,
3595                 (void *)(__psunsigned_t)(ip->i_ino >> 32),
3596                 (void *)(__psunsigned_t)(unsigned)ip->i_ino,
3597                 (void *)(__psunsigned_t)(r1->l0 >> 32),
3598                 (void *)(__psunsigned_t)(unsigned)(r1->l0),
3599                 (void *)(__psunsigned_t)(r1->l1 >> 32),
3600                 (void *)(__psunsigned_t)(unsigned)(r1->l1),
3601                 (void *)(__psunsigned_t)(r2->l0 >> 32),
3602                 (void *)(__psunsigned_t)(unsigned)(r2->l0),
3603                 (void *)(__psunsigned_t)(r2->l1 >> 32),
3604                 (void *)(__psunsigned_t)(unsigned)(r2->l1)
3605                 );
3606         ASSERT(ip->i_xtrace);
3607         ktrace_enter(ip->i_xtrace,
3608                 (void *)(__psint_t)(opcode | (whichfork << 16)),
3609                 (void *)fname, (void *)desc, (void *)ip,
3610                 (void *)(__psint_t)idx,
3611                 (void *)(__psint_t)cnt,
3612                 (void *)(__psunsigned_t)(ip->i_ino >> 32),
3613                 (void *)(__psunsigned_t)(unsigned)ip->i_ino,
3614                 (void *)(__psunsigned_t)(r1->l0 >> 32),
3615                 (void *)(__psunsigned_t)(unsigned)(r1->l0),
3616                 (void *)(__psunsigned_t)(r1->l1 >> 32),
3617                 (void *)(__psunsigned_t)(unsigned)(r1->l1),
3618                 (void *)(__psunsigned_t)(r2->l0 >> 32),
3619                 (void *)(__psunsigned_t)(unsigned)(r2->l0),
3620                 (void *)(__psunsigned_t)(r2->l1 >> 32),
3621                 (void *)(__psunsigned_t)(unsigned)(r2->l1)
3622                 );
3623 }
3624
3625 /*
3626  * Add bmap trace entry prior to a call to xfs_iext_remove.
3627  */
3628 STATIC void
3629 xfs_bmap_trace_delete(
3630         char            *fname,         /* function name */
3631         char            *desc,          /* operation description */
3632         xfs_inode_t     *ip,            /* incore inode pointer */
3633         xfs_extnum_t    idx,            /* index of entry(entries) deleted */
3634         xfs_extnum_t    cnt,            /* count of entries deleted, 1 or 2 */
3635         int             whichfork)      /* data or attr fork */
3636 {
3637         xfs_ifork_t     *ifp;           /* inode fork pointer */
3638
3639         ifp = XFS_IFORK_PTR(ip, whichfork);
3640         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_DELETE, fname, desc, ip, idx,
3641                 cnt, xfs_iext_get_ext(ifp, idx),
3642                 cnt == 2 ? xfs_iext_get_ext(ifp, idx + 1) : NULL,
3643                 whichfork);
3644 }
3645
3646 /*
3647  * Add bmap trace entry prior to a call to xfs_iext_insert, or
3648  * reading in the extents list from the disk (in the btree).
3649  */
3650 STATIC void
3651 xfs_bmap_trace_insert(
3652         char            *fname,         /* function name */
3653         char            *desc,          /* operation description */
3654         xfs_inode_t     *ip,            /* incore inode pointer */
3655         xfs_extnum_t    idx,            /* index of entry(entries) inserted */
3656         xfs_extnum_t    cnt,            /* count of entries inserted, 1 or 2 */
3657         xfs_bmbt_irec_t *r1,            /* inserted record 1 */
3658         xfs_bmbt_irec_t *r2,            /* inserted record 2 or null */
3659         int             whichfork)      /* data or attr fork */
3660 {
3661         xfs_bmbt_rec_t  tr1;            /* compressed record 1 */
3662         xfs_bmbt_rec_t  tr2;            /* compressed record 2 if needed */
3663
3664         xfs_bmbt_set_all(&tr1, r1);
3665         if (cnt == 2) {
3666                 ASSERT(r2 != NULL);
3667                 xfs_bmbt_set_all(&tr2, r2);
3668         } else {
3669                 ASSERT(cnt == 1);
3670                 ASSERT(r2 == NULL);
3671         }
3672         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_INSERT, fname, desc, ip, idx,
3673                 cnt, &tr1, cnt == 2 ? &tr2 : NULL, whichfork);
3674 }
3675
3676 /*
3677  * Add bmap trace entry after updating an extent record in place.
3678  */
3679 STATIC void
3680 xfs_bmap_trace_post_update(
3681         char            *fname,         /* function name */
3682         char            *desc,          /* operation description */
3683         xfs_inode_t     *ip,            /* incore inode pointer */
3684         xfs_extnum_t    idx,            /* index of entry updated */
3685         int             whichfork)      /* data or attr fork */
3686 {
3687         xfs_ifork_t     *ifp;           /* inode fork pointer */
3688
3689         ifp = XFS_IFORK_PTR(ip, whichfork);
3690         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_POST_UP, fname, desc, ip, idx,
3691                 1, xfs_iext_get_ext(ifp, idx), NULL, whichfork);
3692 }
3693
3694 /*
3695  * Add bmap trace entry prior to updating an extent record in place.
3696  */
3697 STATIC void
3698 xfs_bmap_trace_pre_update(
3699         char            *fname,         /* function name */
3700         char            *desc,          /* operation description */
3701         xfs_inode_t     *ip,            /* incore inode pointer */
3702         xfs_extnum_t    idx,            /* index of entry to be updated */
3703         int             whichfork)      /* data or attr fork */
3704 {
3705         xfs_ifork_t     *ifp;           /* inode fork pointer */
3706
3707         ifp = XFS_IFORK_PTR(ip, whichfork);
3708         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_PRE_UP, fname, desc, ip, idx, 1,
3709                 xfs_iext_get_ext(ifp, idx), NULL, whichfork);
3710 }
3711 #endif  /* XFS_BMAP_TRACE */
3712
3713 /*
3714  * Compute the worst-case number of indirect blocks that will be used
3715  * for ip's delayed extent of length "len".
3716  */
3717 STATIC xfs_filblks_t
3718 xfs_bmap_worst_indlen(
3719         xfs_inode_t     *ip,            /* incore inode pointer */
3720         xfs_filblks_t   len)            /* delayed extent length */
3721 {
3722         int             level;          /* btree level number */
3723         int             maxrecs;        /* maximum record count at this level */
3724         xfs_mount_t     *mp;            /* mount structure */
3725         xfs_filblks_t   rval;           /* return value */
3726
3727         mp = ip->i_mount;
3728         maxrecs = mp->m_bmap_dmxr[0];
3729         for (level = 0, rval = 0;
3730              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3731              level++) {
3732                 len += maxrecs - 1;
3733                 do_div(len, maxrecs);
3734                 rval += len;
3735                 if (len == 1)
3736                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3737                                 level - 1;
3738                 if (level == 0)
3739                         maxrecs = mp->m_bmap_dmxr[1];
3740         }
3741         return rval;
3742 }
3743
3744 #if defined(XFS_RW_TRACE)
3745 STATIC void
3746 xfs_bunmap_trace(
3747         xfs_inode_t             *ip,
3748         xfs_fileoff_t           bno,
3749         xfs_filblks_t           len,
3750         int                     flags,
3751         inst_t                  *ra)
3752 {
3753         if (ip->i_rwtrace == NULL)
3754                 return;
3755         ktrace_enter(ip->i_rwtrace,
3756                 (void *)(__psint_t)XFS_BUNMAPI,
3757                 (void *)ip,
3758                 (void *)(__psint_t)((ip->i_d.di_size >> 32) & 0xffffffff),
3759                 (void *)(__psint_t)(ip->i_d.di_size & 0xffffffff),
3760                 (void *)(__psint_t)(((xfs_dfiloff_t)bno >> 32) & 0xffffffff),
3761                 (void *)(__psint_t)((xfs_dfiloff_t)bno & 0xffffffff),
3762                 (void *)(__psint_t)len,
3763                 (void *)(__psint_t)flags,
3764                 (void *)(unsigned long)current_cpu(),
3765                 (void *)ra,
3766                 (void *)0,
3767                 (void *)0,
3768                 (void *)0,
3769                 (void *)0,
3770                 (void *)0,
3771                 (void *)0);
3772 }
3773 #endif
3774
3775 /*
3776  * Convert inode from non-attributed to attributed.
3777  * Must not be in a transaction, ip must not be locked.
3778  */
3779 int                                             /* error code */
3780 xfs_bmap_add_attrfork(
3781         xfs_inode_t             *ip,            /* incore inode pointer */
3782         int                     size,           /* space new attribute needs */
3783         int                     rsvd)           /* xact may use reserved blks */
3784 {
3785         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
3786         xfs_bmap_free_t         flist;          /* freed extent records */
3787         xfs_mount_t             *mp;            /* mount structure */
3788         xfs_trans_t             *tp;            /* transaction pointer */
3789         unsigned long           s;              /* spinlock spl value */
3790         int                     blks;           /* space reservation */
3791         int                     version = 1;    /* superblock attr version */
3792         int                     committed;      /* xaction was committed */
3793         int                     logflags;       /* logging flags */
3794         int                     error;          /* error return value */
3795
3796         ASSERT(XFS_IFORK_Q(ip) == 0);
3797         ASSERT(ip->i_df.if_ext_max ==
3798                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3799
3800         mp = ip->i_mount;
3801         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3802         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3803         blks = XFS_ADDAFORK_SPACE_RES(mp);
3804         if (rsvd)
3805                 tp->t_flags |= XFS_TRANS_RESERVE;
3806         if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3807                         XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3808                 goto error0;
3809         xfs_ilock(ip, XFS_ILOCK_EXCL);
3810         error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip, blks, 0, rsvd ?
3811                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3812                         XFS_QMOPT_RES_REGBLKS);
3813         if (error) {
3814                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3815                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3816                 return error;
3817         }
3818         if (XFS_IFORK_Q(ip))
3819                 goto error1;
3820         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3821                 /*
3822                  * For inodes coming from pre-6.2 filesystems.
3823                  */
3824                 ASSERT(ip->i_d.di_aformat == 0);
3825                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
3826         }
3827         ASSERT(ip->i_d.di_anextents == 0);
3828         VN_HOLD(XFS_ITOV(ip));
3829         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3830         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3831         switch (ip->i_d.di_format) {
3832         case XFS_DINODE_FMT_DEV:
3833                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
3834                 break;
3835         case XFS_DINODE_FMT_UUID:
3836                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
3837                 break;
3838         case XFS_DINODE_FMT_LOCAL:
3839         case XFS_DINODE_FMT_EXTENTS:
3840         case XFS_DINODE_FMT_BTREE:
3841                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
3842                 if (!ip->i_d.di_forkoff)
3843                         ip->i_d.di_forkoff = mp->m_attroffset >> 3;
3844                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
3845                         version = 2;
3846                 break;
3847         default:
3848                 ASSERT(0);
3849                 error = XFS_ERROR(EINVAL);
3850                 goto error1;
3851         }
3852         ip->i_df.if_ext_max =
3853                 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3854         ASSERT(ip->i_afp == NULL);
3855         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
3856         ip->i_afp->if_ext_max =
3857                 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3858         ip->i_afp->if_flags = XFS_IFEXTENTS;
3859         logflags = 0;
3860         XFS_BMAP_INIT(&flist, &firstblock);
3861         switch (ip->i_d.di_format) {
3862         case XFS_DINODE_FMT_LOCAL:
3863                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
3864                         &logflags);
3865                 break;
3866         case XFS_DINODE_FMT_EXTENTS:
3867                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
3868                         &flist, &logflags);
3869                 break;
3870         case XFS_DINODE_FMT_BTREE:
3871                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
3872                         &logflags);
3873                 break;
3874         default:
3875                 error = 0;
3876                 break;
3877         }
3878         if (logflags)
3879                 xfs_trans_log_inode(tp, ip, logflags);
3880         if (error)
3881                 goto error2;
3882         if (!XFS_SB_VERSION_HASATTR(&mp->m_sb) ||
3883            (!XFS_SB_VERSION_HASATTR2(&mp->m_sb) && version == 2)) {
3884                 __int64_t sbfields = 0;
3885
3886                 s = XFS_SB_LOCK(mp);
3887                 if (!XFS_SB_VERSION_HASATTR(&mp->m_sb)) {
3888                         XFS_SB_VERSION_ADDATTR(&mp->m_sb);
3889                         sbfields |= XFS_SB_VERSIONNUM;
3890                 }
3891                 if (!XFS_SB_VERSION_HASATTR2(&mp->m_sb) && version == 2) {
3892                         XFS_SB_VERSION_ADDATTR2(&mp->m_sb);
3893                         sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
3894                 }
3895                 if (sbfields) {
3896                         XFS_SB_UNLOCK(mp, s);
3897                         xfs_mod_sb(tp, sbfields);
3898                 } else
3899                         XFS_SB_UNLOCK(mp, s);
3900         }
3901         if ((error = xfs_bmap_finish(&tp, &flist, firstblock, &committed)))
3902                 goto error2;
3903         error = xfs_trans_commit(tp, XFS_TRANS_PERM_LOG_RES, NULL);
3904         ASSERT(ip->i_df.if_ext_max ==
3905                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3906         return error;
3907 error2:
3908         xfs_bmap_cancel(&flist);
3909 error1:
3910         ASSERT(ismrlocked(&ip->i_lock,MR_UPDATE));
3911         xfs_iunlock(ip, XFS_ILOCK_EXCL);
3912 error0:
3913         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
3914         ASSERT(ip->i_df.if_ext_max ==
3915                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3916         return error;
3917 }
3918
3919 /*
3920  * Add the extent to the list of extents to be free at transaction end.
3921  * The list is maintained sorted (by block number).
3922  */
3923 /* ARGSUSED */
3924 void
3925 xfs_bmap_add_free(
3926         xfs_fsblock_t           bno,            /* fs block number of extent */
3927         xfs_filblks_t           len,            /* length of extent */
3928         xfs_bmap_free_t         *flist,         /* list of extents */
3929         xfs_mount_t             *mp)            /* mount point structure */
3930 {
3931         xfs_bmap_free_item_t    *cur;           /* current (next) element */
3932         xfs_bmap_free_item_t    *new;           /* new element */
3933         xfs_bmap_free_item_t    *prev;          /* previous element */
3934 #ifdef DEBUG
3935         xfs_agnumber_t          agno;
3936         xfs_agblock_t           agbno;
3937
3938         ASSERT(bno != NULLFSBLOCK);
3939         ASSERT(len > 0);
3940         ASSERT(len <= MAXEXTLEN);
3941         ASSERT(!ISNULLSTARTBLOCK(bno));
3942         agno = XFS_FSB_TO_AGNO(mp, bno);
3943         agbno = XFS_FSB_TO_AGBNO(mp, bno);
3944         ASSERT(agno < mp->m_sb.sb_agcount);
3945         ASSERT(agbno < mp->m_sb.sb_agblocks);
3946         ASSERT(len < mp->m_sb.sb_agblocks);
3947         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
3948 #endif
3949         ASSERT(xfs_bmap_free_item_zone != NULL);
3950         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
3951         new->xbfi_startblock = bno;
3952         new->xbfi_blockcount = (xfs_extlen_t)len;
3953         for (prev = NULL, cur = flist->xbf_first;
3954              cur != NULL;
3955              prev = cur, cur = cur->xbfi_next) {
3956                 if (cur->xbfi_startblock >= bno)
3957                         break;
3958         }
3959         if (prev)
3960                 prev->xbfi_next = new;
3961         else
3962                 flist->xbf_first = new;
3963         new->xbfi_next = cur;
3964         flist->xbf_count++;
3965 }
3966
3967 /*
3968  * Compute and fill in the value of the maximum depth of a bmap btree
3969  * in this filesystem.  Done once, during mount.
3970  */
3971 void
3972 xfs_bmap_compute_maxlevels(
3973         xfs_mount_t     *mp,            /* file system mount structure */
3974         int             whichfork)      /* data or attr fork */
3975 {
3976         int             level;          /* btree level */
3977         uint            maxblocks;      /* max blocks at this level */
3978         uint            maxleafents;    /* max leaf entries possible */
3979         int             maxrootrecs;    /* max records in root block */
3980         int             minleafrecs;    /* min records in leaf block */
3981         int             minnoderecs;    /* min records in node block */
3982         int             sz;             /* root block size */
3983
3984         /*
3985          * The maximum number of extents in a file, hence the maximum
3986          * number of leaf entries, is controlled by the type of di_nextents
3987          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
3988          * (a signed 16-bit number, xfs_aextnum_t).
3989          */
3990         if (whichfork == XFS_DATA_FORK) {
3991                 maxleafents = MAXEXTNUM;
3992                 sz = (mp->m_flags & XFS_MOUNT_ATTR2) ?
3993                         XFS_BMDR_SPACE_CALC(MINDBTPTRS) : mp->m_attroffset;
3994         } else {
3995                 maxleafents = MAXAEXTNUM;
3996                 sz = (mp->m_flags & XFS_MOUNT_ATTR2) ?
3997                         XFS_BMDR_SPACE_CALC(MINABTPTRS) :
3998                         mp->m_sb.sb_inodesize - mp->m_attroffset;
3999         }
4000         maxrootrecs = (int)XFS_BTREE_BLOCK_MAXRECS(sz, xfs_bmdr, 0);
4001         minleafrecs = mp->m_bmap_dmnr[0];
4002         minnoderecs = mp->m_bmap_dmnr[1];
4003         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
4004         for (level = 1; maxblocks > 1; level++) {
4005                 if (maxblocks <= maxrootrecs)
4006                         maxblocks = 1;
4007                 else
4008                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
4009         }
4010         mp->m_bm_maxlevels[whichfork] = level;
4011 }
4012
4013 /*
4014  * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
4015  * caller.  Frees all the extents that need freeing, which must be done
4016  * last due to locking considerations.  We never free any extents in
4017  * the first transaction.  This is to allow the caller to make the first
4018  * transaction a synchronous one so that the pointers to the data being
4019  * broken in this transaction will be permanent before the data is actually
4020  * freed.  This is necessary to prevent blocks from being reallocated
4021  * and written to before the free and reallocation are actually permanent.
4022  * We do not just make the first transaction synchronous here, because
4023  * there are more efficient ways to gain the same protection in some cases
4024  * (see the file truncation code).
4025  *
4026  * Return 1 if the given transaction was committed and a new one
4027  * started, and 0 otherwise in the committed parameter.
4028  */
4029 /*ARGSUSED*/
4030 int                                             /* error */
4031 xfs_bmap_finish(
4032         xfs_trans_t             **tp,           /* transaction pointer addr */
4033         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
4034         xfs_fsblock_t           firstblock,     /* controlled ag for allocs */
4035         int                     *committed)     /* xact committed or not */
4036 {
4037         xfs_efd_log_item_t      *efd;           /* extent free data */
4038         xfs_efi_log_item_t      *efi;           /* extent free intention */
4039         int                     error;          /* error return value */
4040         xfs_bmap_free_item_t    *free;          /* free extent item */
4041         unsigned int            logres;         /* new log reservation */
4042         unsigned int            logcount;       /* new log count */
4043         xfs_mount_t             *mp;            /* filesystem mount structure */
4044         xfs_bmap_free_item_t    *next;          /* next item on free list */
4045         xfs_trans_t             *ntp;           /* new transaction pointer */
4046
4047         ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
4048         if (flist->xbf_count == 0) {
4049                 *committed = 0;
4050                 return 0;
4051         }
4052         ntp = *tp;
4053         efi = xfs_trans_get_efi(ntp, flist->xbf_count);
4054         for (free = flist->xbf_first; free; free = free->xbfi_next)
4055                 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
4056                         free->xbfi_blockcount);
4057         logres = ntp->t_log_res;
4058         logcount = ntp->t_log_count;
4059         ntp = xfs_trans_dup(*tp);
4060         error = xfs_trans_commit(*tp, 0, NULL);
4061         *tp = ntp;
4062         *committed = 1;
4063         /*
4064          * We have a new transaction, so we should return committed=1,
4065          * even though we're returning an error.
4066          */
4067         if (error) {
4068                 return error;
4069         }
4070         if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
4071                         logcount)))
4072                 return error;
4073         efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
4074         for (free = flist->xbf_first; free != NULL; free = next) {
4075                 next = free->xbfi_next;
4076                 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
4077                                 free->xbfi_blockcount))) {
4078                         /*
4079                          * The bmap free list will be cleaned up at a
4080                          * higher level.  The EFI will be canceled when
4081                          * this transaction is aborted.
4082                          * Need to force shutdown here to make sure it
4083                          * happens, since this transaction may not be
4084                          * dirty yet.
4085                          */
4086                         mp = ntp->t_mountp;
4087                         if (!XFS_FORCED_SHUTDOWN(mp))
4088                                 xfs_force_shutdown(mp,
4089                                                    (error == EFSCORRUPTED) ?
4090                                                    XFS_CORRUPT_INCORE :
4091                                                    XFS_METADATA_IO_ERROR);
4092                         return error;
4093                 }
4094                 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
4095                         free->xbfi_blockcount);
4096                 xfs_bmap_del_free(flist, NULL, free);
4097         }
4098         return 0;
4099 }
4100
4101 /*
4102  * Free up any items left in the list.
4103  */
4104 void
4105 xfs_bmap_cancel(
4106         xfs_bmap_free_t         *flist) /* list of bmap_free_items */
4107 {
4108         xfs_bmap_free_item_t    *free;  /* free list item */
4109         xfs_bmap_free_item_t    *next;
4110
4111         if (flist->xbf_count == 0)
4112                 return;
4113         ASSERT(flist->xbf_first != NULL);
4114         for (free = flist->xbf_first; free; free = next) {
4115                 next = free->xbfi_next;
4116                 xfs_bmap_del_free(flist, NULL, free);
4117         }
4118         ASSERT(flist->xbf_count == 0);
4119 }
4120
4121 /*
4122  * Returns the file-relative block number of the first unused block(s)
4123  * in the file with at least "len" logically contiguous blocks free.
4124  * This is the lowest-address hole if the file has holes, else the first block
4125  * past the end of file.
4126  * Return 0 if the file is currently local (in-inode).
4127  */
4128 int                                             /* error */
4129 xfs_bmap_first_unused(
4130         xfs_trans_t     *tp,                    /* transaction pointer */
4131         xfs_inode_t     *ip,                    /* incore inode */
4132         xfs_extlen_t    len,                    /* size of hole to find */
4133         xfs_fileoff_t   *first_unused,          /* unused block */
4134         int             whichfork)              /* data or attr fork */
4135 {
4136         xfs_bmbt_rec_t  *ep;                    /* pointer to an extent entry */
4137         int             error;                  /* error return value */
4138         int             idx;                    /* extent record index */
4139         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4140         xfs_fileoff_t   lastaddr;               /* last block number seen */
4141         xfs_fileoff_t   lowest;                 /* lowest useful block */
4142         xfs_fileoff_t   max;                    /* starting useful block */
4143         xfs_fileoff_t   off;                    /* offset for this block */
4144         xfs_extnum_t    nextents;               /* number of extent entries */
4145
4146         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
4147                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
4148                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
4149         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4150                 *first_unused = 0;
4151                 return 0;
4152         }
4153         ifp = XFS_IFORK_PTR(ip, whichfork);
4154         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4155             (error = xfs_iread_extents(tp, ip, whichfork)))
4156                 return error;
4157         lowest = *first_unused;
4158         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4159         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
4160                 ep = xfs_iext_get_ext(ifp, idx);
4161                 off = xfs_bmbt_get_startoff(ep);
4162                 /*
4163                  * See if the hole before this extent will work.
4164                  */
4165                 if (off >= lowest + len && off - max >= len) {
4166                         *first_unused = max;
4167                         return 0;
4168                 }
4169                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
4170                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
4171         }
4172         *first_unused = max;
4173         return 0;
4174 }
4175
4176 /*
4177  * Returns the file-relative block number of the last block + 1 before
4178  * last_block (input value) in the file.
4179  * This is not based on i_size, it is based on the extent records.
4180  * Returns 0 for local files, as they do not have extent records.
4181  */
4182 int                                             /* error */
4183 xfs_bmap_last_before(
4184         xfs_trans_t     *tp,                    /* transaction pointer */
4185         xfs_inode_t     *ip,                    /* incore inode */
4186         xfs_fileoff_t   *last_block,            /* last block */
4187         int             whichfork)              /* data or attr fork */
4188 {
4189         xfs_fileoff_t   bno;                    /* input file offset */
4190         int             eof;                    /* hit end of file */
4191         xfs_bmbt_rec_t  *ep;                    /* pointer to last extent */
4192         int             error;                  /* error return value */
4193         xfs_bmbt_irec_t got;                    /* current extent value */
4194         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4195         xfs_extnum_t    lastx;                  /* last extent used */
4196         xfs_bmbt_irec_t prev;                   /* previous extent value */
4197
4198         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4199             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4200             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4201                return XFS_ERROR(EIO);
4202         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4203                 *last_block = 0;
4204                 return 0;
4205         }
4206         ifp = XFS_IFORK_PTR(ip, whichfork);
4207         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4208             (error = xfs_iread_extents(tp, ip, whichfork)))
4209                 return error;
4210         bno = *last_block - 1;
4211         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4212                 &prev);
4213         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
4214                 if (prev.br_startoff == NULLFILEOFF)
4215                         *last_block = 0;
4216                 else
4217                         *last_block = prev.br_startoff + prev.br_blockcount;
4218         }
4219         /*
4220          * Otherwise *last_block is already the right answer.
4221          */
4222         return 0;
4223 }
4224
4225 /*
4226  * Returns the file-relative block number of the first block past eof in
4227  * the file.  This is not based on i_size, it is based on the extent records.
4228  * Returns 0 for local files, as they do not have extent records.
4229  */
4230 int                                             /* error */
4231 xfs_bmap_last_offset(
4232         xfs_trans_t     *tp,                    /* transaction pointer */
4233         xfs_inode_t     *ip,                    /* incore inode */
4234         xfs_fileoff_t   *last_block,            /* last block */
4235         int             whichfork)              /* data or attr fork */
4236 {
4237         xfs_bmbt_rec_t  *ep;                    /* pointer to last extent */
4238         int             error;                  /* error return value */
4239         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4240         xfs_extnum_t    nextents;               /* number of extent entries */
4241
4242         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4243             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4244             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4245                return XFS_ERROR(EIO);
4246         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4247                 *last_block = 0;
4248                 return 0;
4249         }
4250         ifp = XFS_IFORK_PTR(ip, whichfork);
4251         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4252             (error = xfs_iread_extents(tp, ip, whichfork)))
4253                 return error;
4254         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4255         if (!nextents) {
4256                 *last_block = 0;
4257                 return 0;
4258         }
4259         ep = xfs_iext_get_ext(ifp, nextents - 1);
4260         *last_block = xfs_bmbt_get_startoff(ep) + xfs_bmbt_get_blockcount(ep);
4261         return 0;
4262 }
4263
4264 /*
4265  * Returns whether the selected fork of the inode has exactly one
4266  * block or not.  For the data fork we check this matches di_size,
4267  * implying the file's range is 0..bsize-1.
4268  */
4269 int                                     /* 1=>1 block, 0=>otherwise */
4270 xfs_bmap_one_block(
4271         xfs_inode_t     *ip,            /* incore inode */
4272         int             whichfork)      /* data or attr fork */
4273 {
4274         xfs_bmbt_rec_t  *ep;            /* ptr to fork's extent */
4275         xfs_ifork_t     *ifp;           /* inode fork pointer */
4276         int             rval;           /* return value */
4277         xfs_bmbt_irec_t s;              /* internal version of extent */
4278
4279 #ifndef DEBUG
4280         if (whichfork == XFS_DATA_FORK)
4281                 return ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize;
4282 #endif  /* !DEBUG */
4283         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4284                 return 0;
4285         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4286                 return 0;
4287         ifp = XFS_IFORK_PTR(ip, whichfork);
4288         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4289         ep = xfs_iext_get_ext(ifp, 0);
4290         xfs_bmbt_get_all(ep, &s);
4291         rval = s.br_startoff == 0 && s.br_blockcount == 1;
4292         if (rval && whichfork == XFS_DATA_FORK)
4293                 ASSERT(ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize);
4294         return rval;
4295 }
4296
4297 /*
4298  * Read in the extents to if_extents.
4299  * All inode fields are set up by caller, we just traverse the btree
4300  * and copy the records in. If the file system cannot contain unwritten
4301  * extents, the records are checked for no "state" flags.
4302  */
4303 int                                     /* error */
4304 xfs_bmap_read_extents(
4305         xfs_trans_t             *tp,    /* transaction pointer */
4306         xfs_inode_t             *ip,    /* incore inode */
4307         int                     whichfork) /* data or attr fork */
4308 {
4309         xfs_bmbt_block_t        *block; /* current btree block */
4310         xfs_fsblock_t           bno;    /* block # of "block" */
4311         xfs_buf_t               *bp;    /* buffer for "block" */
4312         int                     error;  /* error return value */
4313         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
4314 #ifdef XFS_BMAP_TRACE
4315         static char             fname[] = "xfs_bmap_read_extents";
4316 #endif
4317         xfs_extnum_t            i, j;   /* index into the extents list */
4318         xfs_ifork_t             *ifp;   /* fork structure */
4319         int                     level;  /* btree level, for checking */
4320         xfs_mount_t             *mp;    /* file system mount structure */
4321         xfs_bmbt_ptr_t          *pp;    /* pointer to block address */
4322         /* REFERENCED */
4323         xfs_extnum_t            room;   /* number of entries there's room for */
4324
4325         bno = NULLFSBLOCK;
4326         mp = ip->i_mount;
4327         ifp = XFS_IFORK_PTR(ip, whichfork);
4328         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4329                                         XFS_EXTFMT_INODE(ip);
4330         block = ifp->if_broot;
4331         /*
4332          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4333          */
4334         level = be16_to_cpu(block->bb_level);
4335         ASSERT(level > 0);
4336         pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
4337         ASSERT(INT_GET(*pp, ARCH_CONVERT) != NULLDFSBNO);
4338         ASSERT(XFS_FSB_TO_AGNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agcount);
4339         ASSERT(XFS_FSB_TO_AGBNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agblocks);
4340         bno = INT_GET(*pp, ARCH_CONVERT);
4341         /*
4342          * Go down the tree until leaf level is reached, following the first
4343          * pointer (leftmost) at each level.
4344          */
4345         while (level-- > 0) {
4346                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4347                                 XFS_BMAP_BTREE_REF)))
4348                         return error;
4349                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
4350                 XFS_WANT_CORRUPTED_GOTO(
4351                         XFS_BMAP_SANITY_CHECK(mp, block, level),
4352                         error0);
4353                 if (level == 0)
4354                         break;
4355                 pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt, block,
4356                         1, mp->m_bmap_dmxr[1]);
4357                 XFS_WANT_CORRUPTED_GOTO(
4358                         XFS_FSB_SANITY_CHECK(mp, INT_GET(*pp, ARCH_CONVERT)),
4359                         error0);
4360                 bno = INT_GET(*pp, ARCH_CONVERT);
4361                 xfs_trans_brelse(tp, bp);
4362         }
4363         /*
4364          * Here with bp and block set to the leftmost leaf node in the tree.
4365          */
4366         room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4367         i = 0;
4368         /*
4369          * Loop over all leaf nodes.  Copy information to the extent records.
4370          */
4371         for (;;) {
4372                 xfs_bmbt_rec_t  *frp, *trp;
4373                 xfs_fsblock_t   nextbno;
4374                 xfs_extnum_t    num_recs;
4375                 xfs_extnum_t    start;
4376
4377
4378                 num_recs = be16_to_cpu(block->bb_numrecs);
4379                 if (unlikely(i + num_recs > room)) {
4380                         ASSERT(i + num_recs <= room);
4381                         xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
4382                                 "corrupt dinode %Lu, (btree extents).",
4383                                 (unsigned long long) ip->i_ino);
4384                         XFS_ERROR_REPORT("xfs_bmap_read_extents(1)",
4385                                          XFS_ERRLEVEL_LOW,
4386                                         ip->i_mount);
4387                         goto error0;
4388                 }
4389                 XFS_WANT_CORRUPTED_GOTO(
4390                         XFS_BMAP_SANITY_CHECK(mp, block, 0),
4391                         error0);
4392                 /*
4393                  * Read-ahead the next leaf block, if any.
4394                  */
4395                 nextbno = be64_to_cpu(block->bb_rightsib);
4396                 if (nextbno != NULLFSBLOCK)
4397                         xfs_btree_reada_bufl(mp, nextbno, 1);
4398                 /*
4399                  * Copy records into the extent records.
4400                  */
4401                 frp = XFS_BTREE_REC_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt,
4402                         block, 1, mp->m_bmap_dmxr[0]);
4403                 start = i;
4404                 for (j = 0; j < num_recs; j++, i++, frp++) {
4405                         trp = xfs_iext_get_ext(ifp, i);
4406                         trp->l0 = INT_GET(frp->l0, ARCH_CONVERT);
4407                         trp->l1 = INT_GET(frp->l1, ARCH_CONVERT);
4408                 }
4409                 if (exntf == XFS_EXTFMT_NOSTATE) {
4410                         /*
4411                          * Check all attribute bmap btree records and
4412                          * any "older" data bmap btree records for a
4413                          * set bit in the "extent flag" position.
4414                          */
4415                         if (unlikely(xfs_check_nostate_extents(ifp,
4416                                         start, num_recs))) {
4417                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4418                                                  XFS_ERRLEVEL_LOW,
4419                                                  ip->i_mount);
4420                                 goto error0;
4421                         }
4422                 }
4423                 xfs_trans_brelse(tp, bp);
4424                 bno = nextbno;
4425                 /*
4426                  * If we've reached the end, stop.
4427                  */
4428                 if (bno == NULLFSBLOCK)
4429                         break;
4430                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4431                                 XFS_BMAP_BTREE_REF)))
4432                         return error;
4433                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
4434         }
4435         ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4436         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4437         xfs_bmap_trace_exlist(fname, ip, i, whichfork);
4438         return 0;
4439 error0:
4440         xfs_trans_brelse(tp, bp);
4441         return XFS_ERROR(EFSCORRUPTED);
4442 }
4443
4444 #ifdef XFS_BMAP_TRACE
4445 /*
4446  * Add bmap trace insert entries for all the contents of the extent records.
4447  */
4448 void
4449 xfs_bmap_trace_exlist(
4450         char            *fname,         /* function name */
4451         xfs_inode_t     *ip,            /* incore inode pointer */
4452         xfs_extnum_t    cnt,            /* count of entries in the list */
4453         int             whichfork)      /* data or attr fork */
4454 {
4455         xfs_bmbt_rec_t  *ep;            /* current extent record */
4456         xfs_extnum_t    idx;            /* extent record index */
4457         xfs_ifork_t     *ifp;           /* inode fork pointer */
4458         xfs_bmbt_irec_t s;              /* file extent record */
4459
4460         ifp = XFS_IFORK_PTR(ip, whichfork);
4461         ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4462         for (idx = 0; idx < cnt; idx++) {
4463                 ep = xfs_iext_get_ext(ifp, idx);
4464                 xfs_bmbt_get_all(ep, &s);
4465                 xfs_bmap_trace_insert(fname, "exlist", ip, idx, 1, &s, NULL,
4466                         whichfork);
4467         }
4468 }
4469 #endif
4470
4471 #ifdef DEBUG
4472 /*
4473  * Validate that the bmbt_irecs being returned from bmapi are valid
4474  * given the callers original parameters.  Specifically check the
4475  * ranges of the returned irecs to ensure that they only extent beyond
4476  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4477  */
4478 STATIC void
4479 xfs_bmap_validate_ret(
4480         xfs_fileoff_t           bno,
4481         xfs_filblks_t           len,
4482         int                     flags,
4483         xfs_bmbt_irec_t         *mval,
4484         int                     nmap,
4485         int                     ret_nmap)
4486 {
4487         int                     i;              /* index to map values */
4488
4489         ASSERT(ret_nmap <= nmap);
4490
4491         for (i = 0; i < ret_nmap; i++) {
4492                 ASSERT(mval[i].br_blockcount > 0);
4493                 if (!(flags & XFS_BMAPI_ENTIRE)) {
4494                         ASSERT(mval[i].br_startoff >= bno);
4495                         ASSERT(mval[i].br_blockcount <= len);
4496                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4497                                bno + len);
4498                 } else {
4499                         ASSERT(mval[i].br_startoff < bno + len);
4500                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4501                                bno);
4502                 }
4503                 ASSERT(i == 0 ||
4504                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4505                        mval[i].br_startoff);
4506                 if ((flags & XFS_BMAPI_WRITE) && !(flags & XFS_BMAPI_DELAY))
4507                         ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4508                                mval[i].br_startblock != HOLESTARTBLOCK);
4509                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4510                        mval[i].br_state == XFS_EXT_UNWRITTEN);
4511         }
4512 }
4513 #endif /* DEBUG */
4514
4515
4516 /*
4517  * Map file blocks to filesystem blocks.
4518  * File range is given by the bno/len pair.
4519  * Adds blocks to file if a write ("flags & XFS_BMAPI_WRITE" set)
4520  * into a hole or past eof.
4521  * Only allocates blocks from a single allocation group,
4522  * to avoid locking problems.
4523  * The returned value in "firstblock" from the first call in a transaction
4524  * must be remembered and presented to subsequent calls in "firstblock".
4525  * An upper bound for the number of blocks to be allocated is supplied to
4526  * the first call in "total"; if no allocation group has that many free
4527  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4528  */
4529 int                                     /* error */
4530 xfs_bmapi(
4531         xfs_trans_t     *tp,            /* transaction pointer */
4532         xfs_inode_t     *ip,            /* incore inode */
4533         xfs_fileoff_t   bno,            /* starting file offs. mapped */
4534         xfs_filblks_t   len,            /* length to map in file */
4535         int             flags,          /* XFS_BMAPI_... */
4536         xfs_fsblock_t   *firstblock,    /* first allocated block
4537                                            controls a.g. for allocs */
4538         xfs_extlen_t    total,          /* total blocks needed */
4539         xfs_bmbt_irec_t *mval,          /* output: map values */
4540         int             *nmap,          /* i/o: mval size/count */
4541         xfs_bmap_free_t *flist)         /* i/o: list extents to free */
4542 {
4543         xfs_fsblock_t   abno;           /* allocated block number */
4544         xfs_extlen_t    alen;           /* allocated extent length */
4545         xfs_fileoff_t   aoff;           /* allocated file offset */
4546         xfs_bmalloca_t  bma;            /* args for xfs_bmap_alloc */
4547         xfs_btree_cur_t *cur;           /* bmap btree cursor */
4548         xfs_fileoff_t   end;            /* end of mapped file region */
4549         int             eof;            /* we've hit the end of extents */
4550         xfs_bmbt_rec_t  *ep;            /* extent record pointer */
4551         int             error;          /* error return */
4552         xfs_bmbt_irec_t got;            /* current file extent record */
4553         xfs_ifork_t     *ifp;           /* inode fork pointer */
4554         xfs_extlen_t    indlen;         /* indirect blocks length */
4555         xfs_extnum_t    lastx;          /* last useful extent number */
4556         int             logflags;       /* flags for transaction logging */
4557         xfs_extlen_t    minleft;        /* min blocks left after allocation */
4558         xfs_extlen_t    minlen;         /* min allocation size */
4559         xfs_mount_t     *mp;            /* xfs mount structure */
4560         int             n;              /* current extent index */
4561         int             nallocs;        /* number of extents alloc\'d */
4562         xfs_extnum_t    nextents;       /* number of extents in file */
4563         xfs_fileoff_t   obno;           /* old block number (offset) */
4564         xfs_bmbt_irec_t prev;           /* previous file extent record */
4565         int             tmp_logflags;   /* temp flags holder */
4566         int             whichfork;      /* data or attr fork */
4567         char            inhole;         /* current location is hole in file */
4568         char            wasdelay;       /* old extent was delayed */
4569         char            wr;             /* this is a write request */
4570         char            rt;             /* this is a realtime file */
4571 #ifdef DEBUG
4572         xfs_fileoff_t   orig_bno;       /* original block number value */
4573         int             orig_flags;     /* original flags arg value */
4574         xfs_filblks_t   orig_len;       /* original value of len arg */
4575         xfs_bmbt_irec_t *orig_mval;     /* original value of mval */
4576         int             orig_nmap;      /* original value of *nmap */
4577
4578         orig_bno = bno;
4579         orig_len = len;
4580         orig_flags = flags;
4581         orig_mval = mval;
4582         orig_nmap = *nmap;
4583 #endif
4584         ASSERT(*nmap >= 1);
4585         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP || !(flags & XFS_BMAPI_WRITE));
4586         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4587                 XFS_ATTR_FORK : XFS_DATA_FORK;
4588         mp = ip->i_mount;
4589         if (unlikely(XFS_TEST_ERROR(
4590             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4591              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4592              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4593              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4594                 XFS_ERROR_REPORT("xfs_bmapi", XFS_ERRLEVEL_LOW, mp);
4595                 return XFS_ERROR(EFSCORRUPTED);
4596         }
4597         if (XFS_FORCED_SHUTDOWN(mp))
4598                 return XFS_ERROR(EIO);
4599         rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4600         ifp = XFS_IFORK_PTR(ip, whichfork);
4601         ASSERT(ifp->if_ext_max ==
4602                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4603         if ((wr = (flags & XFS_BMAPI_WRITE)) != 0)
4604                 XFS_STATS_INC(xs_blk_mapw);
4605         else
4606                 XFS_STATS_INC(xs_blk_mapr);
4607         /*
4608          * IGSTATE flag is used to combine extents which
4609          * differ only due to the state of the extents.
4610          * This technique is used from xfs_getbmap()
4611          * when the caller does not wish to see the
4612          * separation (which is the default).
4613          *
4614          * This technique is also used when writing a
4615          * buffer which has been partially written,
4616          * (usually by being flushed during a chunkread),
4617          * to ensure one write takes place. This also
4618          * prevents a change in the xfs inode extents at
4619          * this time, intentionally. This change occurs
4620          * on completion of the write operation, in
4621          * xfs_strat_comp(), where the xfs_bmapi() call
4622          * is transactioned, and the extents combined.
4623          */
4624         if ((flags & XFS_BMAPI_IGSTATE) && wr)  /* if writing unwritten space */
4625                 wr = 0;                         /* no allocations are allowed */
4626         ASSERT(wr || !(flags & XFS_BMAPI_DELAY));
4627         logflags = 0;
4628         nallocs = 0;
4629         cur = NULL;
4630         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4631                 ASSERT(wr && tp);
4632                 if ((error = xfs_bmap_local_to_extents(tp, ip,
4633                                 firstblock, total, &logflags, whichfork)))
4634                         goto error0;
4635         }
4636         if (wr && *firstblock == NULLFSBLOCK) {
4637                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4638                         minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4639                 else
4640                         minleft = 1;
4641         } else
4642                 minleft = 0;
4643         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4644             (error = xfs_iread_extents(tp, ip, whichfork)))
4645                 goto error0;
4646         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4647                 &prev);
4648         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4649         n = 0;
4650         end = bno + len;
4651         obno = bno;
4652         bma.ip = NULL;
4653         while (bno < end && n < *nmap) {
4654                 /*
4655                  * Reading past eof, act as though there's a hole
4656                  * up to end.
4657                  */
4658                 if (eof && !wr)
4659                         got.br_startoff = end;
4660                 inhole = eof || got.br_startoff > bno;
4661                 wasdelay = wr && !inhole && !(flags & XFS_BMAPI_DELAY) &&
4662                         ISNULLSTARTBLOCK(got.br_startblock);
4663                 /*
4664                  * First, deal with the hole before the allocated space
4665                  * that we found, if any.
4666                  */
4667                 if (wr && (inhole || wasdelay)) {
4668                         /*
4669                          * For the wasdelay case, we could also just
4670                          * allocate the stuff asked for in this bmap call
4671                          * but that wouldn't be as good.
4672                          */
4673                         if (wasdelay && !(flags & XFS_BMAPI_EXACT)) {
4674                                 alen = (xfs_extlen_t)got.br_blockcount;
4675                                 aoff = got.br_startoff;
4676                                 if (lastx != NULLEXTNUM && lastx) {
4677                                         ep = xfs_iext_get_ext(ifp, lastx - 1);
4678                                         xfs_bmbt_get_all(ep, &prev);
4679                                 }
4680                         } else if (wasdelay) {
4681                                 alen = (xfs_extlen_t)
4682                                         XFS_FILBLKS_MIN(len,
4683                                                 (got.br_startoff +
4684                                                  got.br_blockcount) - bno);
4685                                 aoff = bno;
4686                         } else {
4687                                 alen = (xfs_extlen_t)
4688                                         XFS_FILBLKS_MIN(len, MAXEXTLEN);
4689                                 if (!eof)
4690                                         alen = (xfs_extlen_t)
4691                                                 XFS_FILBLKS_MIN(alen,
4692                                                         got.br_startoff - bno);
4693                                 aoff = bno;
4694                         }
4695                         minlen = (flags & XFS_BMAPI_CONTIG) ? alen : 1;
4696                         if (flags & XFS_BMAPI_DELAY) {
4697                                 xfs_extlen_t    extsz;
4698
4699                                 /* Figure out the extent size, adjust alen */
4700                                 if (rt) {
4701                                         if (!(extsz = ip->i_d.di_extsize))
4702                                                 extsz = mp->m_sb.sb_rextsize;
4703                                 } else {
4704                                         extsz = ip->i_d.di_extsize;
4705                                 }
4706                                 if (extsz) {
4707                                         error = xfs_bmap_extsize_align(mp,
4708                                                         &got, &prev, extsz,
4709                                                         rt, eof,
4710                                                         flags&XFS_BMAPI_DELAY,
4711                                                         flags&XFS_BMAPI_CONVERT,
4712                                                         &aoff, &alen);
4713                                         ASSERT(!error);
4714                                 }
4715
4716                                 if (rt)
4717                                         extsz = alen / mp->m_sb.sb_rextsize;
4718
4719                                 /*
4720                                  * Make a transaction-less quota reservation for
4721                                  * delayed allocation blocks. This number gets
4722                                  * adjusted later.  We return if we haven't
4723                                  * allocated blocks already inside this loop.
4724                                  */
4725                                 if ((error = XFS_TRANS_RESERVE_QUOTA_NBLKS(
4726                                                 mp, NULL, ip, (long)alen, 0,
4727                                                 rt ? XFS_QMOPT_RES_RTBLKS :
4728                                                      XFS_QMOPT_RES_REGBLKS))) {
4729                                         if (n == 0) {
4730                                                 *nmap = 0;
4731                                                 ASSERT(cur == NULL);
4732                                                 return error;
4733                                         }
4734                                         break;
4735                                 }
4736
4737                                 /*
4738                                  * Split changing sb for alen and indlen since
4739                                  * they could be coming from different places.
4740                                  */
4741                                 indlen = (xfs_extlen_t)
4742                                         xfs_bmap_worst_indlen(ip, alen);
4743                                 ASSERT(indlen > 0);
4744
4745                                 if (rt) {
4746                                         error = xfs_mod_incore_sb(mp,
4747                                                         XFS_SBS_FREXTENTS,
4748                                                         -(extsz), (flags &
4749                                                         XFS_BMAPI_RSVBLOCKS));
4750                                 } else {
4751                                         error = xfs_mod_incore_sb(mp,
4752                                                         XFS_SBS_FDBLOCKS,
4753                                                         -(alen), (flags &
4754                                                         XFS_BMAPI_RSVBLOCKS));
4755                                 }
4756                                 if (!error) {
4757                                         error = xfs_mod_incore_sb(mp,
4758                                                         XFS_SBS_FDBLOCKS,
4759                                                         -(indlen), (flags &
4760                                                         XFS_BMAPI_RSVBLOCKS));
4761                                         if (error && rt)
4762                                                 xfs_mod_incore_sb(mp,
4763                                                         XFS_SBS_FREXTENTS,
4764                                                         extsz, (flags &
4765                                                         XFS_BMAPI_RSVBLOCKS));
4766                                         else if (error)
4767                                                 xfs_mod_incore_sb(mp,
4768                                                         XFS_SBS_FDBLOCKS,
4769                                                         alen, (flags &
4770                                                         XFS_BMAPI_RSVBLOCKS));
4771                                 }
4772
4773                                 if (error) {
4774                                         if (XFS_IS_QUOTA_ON(mp))
4775                                                 /* unreserve the blocks now */
4776                                                 (void)
4777                                                 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(
4778                                                         mp, NULL, ip,
4779                                                         (long)alen, 0, rt ?
4780                                                         XFS_QMOPT_RES_RTBLKS :
4781                                                         XFS_QMOPT_RES_REGBLKS);
4782                                         break;
4783                                 }
4784
4785                                 ip->i_delayed_blks += alen;
4786                                 abno = NULLSTARTBLOCK(indlen);
4787                         } else {
4788                                 /*
4789                                  * If first time, allocate and fill in
4790                                  * once-only bma fields.
4791                                  */
4792                                 if (bma.ip == NULL) {
4793                                         bma.tp = tp;
4794                                         bma.ip = ip;
4795                                         bma.prevp = &prev;
4796                                         bma.gotp = &got;
4797                                         bma.total = total;
4798                                         bma.userdata = 0;
4799                                 }
4800                                 /* Indicate if this is the first user data
4801                                  * in the file, or just any user data.
4802                                  */
4803                                 if (!(flags & XFS_BMAPI_METADATA)) {
4804                                         bma.userdata = (aoff == 0) ?
4805                                                 XFS_ALLOC_INITIAL_USER_DATA :
4806                                                 XFS_ALLOC_USERDATA;
4807                                 }
4808                                 /*
4809                                  * Fill in changeable bma fields.
4810                                  */
4811                                 bma.eof = eof;
4812                                 bma.firstblock = *firstblock;
4813                                 bma.alen = alen;
4814                                 bma.off = aoff;
4815                                 bma.conv = (flags & XFS_BMAPI_CONVERT);
4816                                 bma.wasdel = wasdelay;
4817                                 bma.minlen = minlen;
4818                                 bma.low = flist->xbf_low;
4819                                 bma.minleft = minleft;
4820                                 /*
4821                                  * Only want to do the alignment at the
4822                                  * eof if it is userdata and allocation length
4823                                  * is larger than a stripe unit.
4824                                  */
4825                                 if (mp->m_dalign && alen >= mp->m_dalign &&
4826                                     (!(flags & XFS_BMAPI_METADATA)) &&
4827                                     (whichfork == XFS_DATA_FORK)) {
4828                                         if ((error = xfs_bmap_isaeof(ip, aoff,
4829                                                         whichfork, &bma.aeof)))
4830                                                 goto error0;
4831                                 } else
4832                                         bma.aeof = 0;
4833                                 /*
4834                                  * Call allocator.
4835                                  */
4836                                 if ((error = xfs_bmap_alloc(&bma)))
4837                                         goto error0;
4838                                 /*
4839                                  * Copy out result fields.
4840                                  */
4841                                 abno = bma.rval;
4842                                 if ((flist->xbf_low = bma.low))
4843                                         minleft = 0;
4844                                 alen = bma.alen;
4845                                 aoff = bma.off;
4846                                 ASSERT(*firstblock == NULLFSBLOCK ||
4847                                        XFS_FSB_TO_AGNO(mp, *firstblock) ==
4848                                        XFS_FSB_TO_AGNO(mp, bma.firstblock) ||
4849                                        (flist->xbf_low &&
4850                                         XFS_FSB_TO_AGNO(mp, *firstblock) <
4851                                         XFS_FSB_TO_AGNO(mp, bma.firstblock)));
4852                                 *firstblock = bma.firstblock;
4853                                 if (cur)
4854                                         cur->bc_private.b.firstblock =
4855                                                 *firstblock;
4856                                 if (abno == NULLFSBLOCK)
4857                                         break;
4858                                 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
4859                                         cur = xfs_btree_init_cursor(mp,
4860                                                 tp, NULL, 0, XFS_BTNUM_BMAP,
4861                                                 ip, whichfork);
4862                                         cur->bc_private.b.firstblock =
4863                                                 *firstblock;
4864                                         cur->bc_private.b.flist = flist;
4865                                 }
4866                                 /*
4867                                  * Bump the number of extents we've allocated
4868                                  * in this call.
4869                                  */
4870                                 nallocs++;
4871                         }
4872                         if (cur)
4873                                 cur->bc_private.b.flags =
4874                                         wasdelay ? XFS_BTCUR_BPRV_WASDEL : 0;
4875                         got.br_startoff = aoff;
4876                         got.br_startblock = abno;
4877                         got.br_blockcount = alen;
4878                         got.br_state = XFS_EXT_NORM;    /* assume normal */
4879                         /*
4880                          * Determine state of extent, and the filesystem.
4881                          * A wasdelay extent has been initialized, so
4882                          * shouldn't be flagged as unwritten.
4883                          */
4884                         if (wr && XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4885                                 if (!wasdelay && (flags & XFS_BMAPI_PREALLOC))
4886                                         got.br_state = XFS_EXT_UNWRITTEN;
4887                         }
4888                         error = xfs_bmap_add_extent(ip, lastx, &cur, &got,
4889                                 firstblock, flist, &tmp_logflags, whichfork,
4890                                 (flags & XFS_BMAPI_RSVBLOCKS));
4891                         logflags |= tmp_logflags;
4892                         if (error)
4893                                 goto error0;
4894                         lastx = ifp->if_lastex;
4895                         ep = xfs_iext_get_ext(ifp, lastx);
4896                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4897                         xfs_bmbt_get_all(ep, &got);
4898                         ASSERT(got.br_startoff <= aoff);
4899                         ASSERT(got.br_startoff + got.br_blockcount >=
4900                                 aoff + alen);
4901 #ifdef DEBUG
4902                         if (flags & XFS_BMAPI_DELAY) {
4903                                 ASSERT(ISNULLSTARTBLOCK(got.br_startblock));
4904                                 ASSERT(STARTBLOCKVAL(got.br_startblock) > 0);
4905                         }
4906                         ASSERT(got.br_state == XFS_EXT_NORM ||
4907                                got.br_state == XFS_EXT_UNWRITTEN);
4908 #endif
4909                         /*
4910                          * Fall down into the found allocated space case.
4911                          */
4912                 } else if (inhole) {
4913                         /*
4914                          * Reading in a hole.
4915                          */
4916                         mval->br_startoff = bno;
4917                         mval->br_startblock = HOLESTARTBLOCK;
4918                         mval->br_blockcount =
4919                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4920                         mval->br_state = XFS_EXT_NORM;
4921                         bno += mval->br_blockcount;
4922                         len -= mval->br_blockcount;
4923                         mval++;
4924                         n++;
4925                         continue;
4926                 }
4927                 /*
4928                  * Then deal with the allocated space we found.
4929                  */
4930                 ASSERT(ep != NULL);
4931                 if (!(flags & XFS_BMAPI_ENTIRE) &&
4932                     (got.br_startoff + got.br_blockcount > obno)) {
4933                         if (obno > bno)
4934                                 bno = obno;
4935                         ASSERT((bno >= obno) || (n == 0));
4936                         ASSERT(bno < end);
4937                         mval->br_startoff = bno;
4938                         if (ISNULLSTARTBLOCK(got.br_startblock)) {
4939                                 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
4940                                 mval->br_startblock = DELAYSTARTBLOCK;
4941                         } else
4942                                 mval->br_startblock =
4943                                         got.br_startblock +
4944                                         (bno - got.br_startoff);
4945                         /*
4946                          * Return the minimum of what we got and what we
4947                          * asked for for the length.  We can use the len
4948                          * variable here because it is modified below
4949                          * and we could have been there before coming
4950                          * here if the first part of the allocation
4951                          * didn't overlap what was asked for.
4952                          */
4953                         mval->br_blockcount =
4954                                 XFS_FILBLKS_MIN(end - bno, got.br_blockcount -
4955                                         (bno - got.br_startoff));
4956                         mval->br_state = got.br_state;
4957                         ASSERT(mval->br_blockcount <= len);
4958                 } else {
4959                         *mval = got;
4960                         if (ISNULLSTARTBLOCK(mval->br_startblock)) {
4961                                 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
4962                                 mval->br_startblock = DELAYSTARTBLOCK;
4963                         }
4964                 }
4965
4966                 /*
4967                  * Check if writing previously allocated but
4968                  * unwritten extents.
4969                  */
4970                 if (wr && mval->br_state == XFS_EXT_UNWRITTEN &&
4971                     ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_DELAY)) == 0)) {
4972                         /*
4973                          * Modify (by adding) the state flag, if writing.
4974                          */
4975                         ASSERT(mval->br_blockcount <= len);
4976                         if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
4977                                 cur = xfs_btree_init_cursor(mp,
4978                                         tp, NULL, 0, XFS_BTNUM_BMAP,
4979                                         ip, whichfork);
4980                                 cur->bc_private.b.firstblock =
4981                                         *firstblock;
4982                                 cur->bc_private.b.flist = flist;
4983                         }
4984                         mval->br_state = XFS_EXT_NORM;
4985                         error = xfs_bmap_add_extent(ip, lastx, &cur, mval,
4986                                 firstblock, flist, &tmp_logflags, whichfork,
4987                                 (flags & XFS_BMAPI_RSVBLOCKS));
4988                         logflags |= tmp_logflags;
4989                         if (error)
4990                                 goto error0;
4991                         lastx = ifp->if_lastex;
4992                         ep = xfs_iext_get_ext(ifp, lastx);
4993                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4994                         xfs_bmbt_get_all(ep, &got);
4995                         /*
4996                          * We may have combined previously unwritten
4997                          * space with written space, so generate
4998                          * another request.
4999                          */
5000                         if (mval->br_blockcount < len)
5001                                 continue;
5002                 }
5003
5004                 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
5005                        ((mval->br_startoff + mval->br_blockcount) <= end));
5006                 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
5007                        (mval->br_blockcount <= len) ||
5008                        (mval->br_startoff < obno));
5009                 bno = mval->br_startoff + mval->br_blockcount;
5010                 len = end - bno;
5011                 if (n > 0 && mval->br_startoff == mval[-1].br_startoff) {
5012                         ASSERT(mval->br_startblock == mval[-1].br_startblock);
5013                         ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
5014                         ASSERT(mval->br_state == mval[-1].br_state);
5015                         mval[-1].br_blockcount = mval->br_blockcount;
5016                         mval[-1].br_state = mval->br_state;
5017                 } else if (n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
5018                            mval[-1].br_startblock != DELAYSTARTBLOCK &&
5019                            mval[-1].br_startblock != HOLESTARTBLOCK &&
5020                            mval->br_startblock ==
5021                            mval[-1].br_startblock + mval[-1].br_blockcount &&
5022                            ((flags & XFS_BMAPI_IGSTATE) ||
5023                                 mval[-1].br_state == mval->br_state)) {
5024                         ASSERT(mval->br_startoff ==
5025                                mval[-1].br_startoff + mval[-1].br_blockcount);
5026                         mval[-1].br_blockcount += mval->br_blockcount;
5027                 } else if (n > 0 &&
5028                            mval->br_startblock == DELAYSTARTBLOCK &&
5029                            mval[-1].br_startblock == DELAYSTARTBLOCK &&
5030                            mval->br_startoff ==
5031                            mval[-1].br_startoff + mval[-1].br_blockcount) {
5032                         mval[-1].br_blockcount += mval->br_blockcount;
5033                         mval[-1].br_state = mval->br_state;
5034                 } else if (!((n == 0) &&
5035                              ((mval->br_startoff + mval->br_blockcount) <=
5036                               obno))) {
5037                         mval++;
5038                         n++;
5039                 }
5040                 /*
5041                  * If we're done, stop now.  Stop when we've allocated
5042                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
5043                  * the transaction may get too big.
5044                  */
5045                 if (bno >= end || n >= *nmap || nallocs >= *nmap)
5046                         break;
5047                 /*
5048                  * Else go on to the next record.
5049                  */
5050                 ep = xfs_iext_get_ext(ifp, ++lastx);
5051                 if (lastx >= nextents) {
5052                         eof = 1;
5053                         prev = got;
5054                 } else
5055                         xfs_bmbt_get_all(ep, &got);
5056         }
5057         ifp->if_lastex = lastx;
5058         *nmap = n;
5059         /*
5060          * Transform from btree to extents, give it cur.
5061          */
5062         if (tp && XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5063             XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5064                 ASSERT(wr && cur);
5065                 error = xfs_bmap_btree_to_extents(tp, ip, cur,
5066                         &tmp_logflags, whichfork);
5067                 logflags |= tmp_logflags;
5068                 if (error)
5069                         goto error0;
5070         }
5071         ASSERT(ifp->if_ext_max ==
5072                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5073         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
5074                XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max);
5075         error = 0;
5076
5077 error0:
5078         /*
5079          * Log everything.  Do this after conversion, there's no point in
5080          * logging the extent records if we've converted to btree format.
5081          */
5082         if ((logflags & XFS_ILOG_FEXT(whichfork)) &&
5083             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5084                 logflags &= ~XFS_ILOG_FEXT(whichfork);
5085         else if ((logflags & XFS_ILOG_FBROOT(whichfork)) &&
5086                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5087                 logflags &= ~XFS_ILOG_FBROOT(whichfork);
5088         /*
5089          * Log whatever the flags say, even if error.  Otherwise we might miss
5090          * detecting a case where the data is changed, there's an error,
5091          * and it's not logged so we don't shutdown when we should.
5092          */
5093         if (logflags) {
5094                 ASSERT(tp && wr);
5095                 xfs_trans_log_inode(tp, ip, logflags);
5096         }
5097         if (cur) {
5098                 if (!error) {
5099                         ASSERT(*firstblock == NULLFSBLOCK ||
5100                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
5101                                XFS_FSB_TO_AGNO(mp,
5102                                        cur->bc_private.b.firstblock) ||
5103                                (flist->xbf_low &&
5104                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
5105                                 XFS_FSB_TO_AGNO(mp,
5106                                         cur->bc_private.b.firstblock)));
5107                         *firstblock = cur->bc_private.b.firstblock;
5108                 }
5109                 xfs_btree_del_cursor(cur,
5110                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5111         }
5112         if (!error)
5113                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
5114                         orig_nmap, *nmap);
5115         return error;
5116 }
5117
5118 /*
5119  * Map file blocks to filesystem blocks, simple version.
5120  * One block (extent) only, read-only.
5121  * For flags, only the XFS_BMAPI_ATTRFORK flag is examined.
5122  * For the other flag values, the effect is as if XFS_BMAPI_METADATA
5123  * was set and all the others were clear.
5124  */
5125 int                                             /* error */
5126 xfs_bmapi_single(
5127         xfs_trans_t     *tp,            /* transaction pointer */
5128         xfs_inode_t     *ip,            /* incore inode */
5129         int             whichfork,      /* data or attr fork */
5130         xfs_fsblock_t   *fsb,           /* output: mapped block */
5131         xfs_fileoff_t   bno)            /* starting file offs. mapped */
5132 {
5133         int             eof;            /* we've hit the end of extents */
5134         int             error;          /* error return */
5135         xfs_bmbt_irec_t got;            /* current file extent record */
5136         xfs_ifork_t     *ifp;           /* inode fork pointer */
5137         xfs_extnum_t    lastx;          /* last useful extent number */
5138         xfs_bmbt_irec_t prev;           /* previous file extent record */
5139
5140         ifp = XFS_IFORK_PTR(ip, whichfork);
5141         if (unlikely(
5142             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
5143             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)) {
5144                XFS_ERROR_REPORT("xfs_bmapi_single", XFS_ERRLEVEL_LOW,
5145                                 ip->i_mount);
5146                return XFS_ERROR(EFSCORRUPTED);
5147         }
5148         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
5149                 return XFS_ERROR(EIO);
5150         XFS_STATS_INC(xs_blk_mapr);
5151         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5152             (error = xfs_iread_extents(tp, ip, whichfork)))
5153                 return error;
5154         (void)xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5155                 &prev);
5156         /*
5157          * Reading past eof, act as though there's a hole
5158          * up to end.
5159          */
5160         if (eof || got.br_startoff > bno) {
5161                 *fsb = NULLFSBLOCK;
5162                 return 0;
5163         }
5164         ASSERT(!ISNULLSTARTBLOCK(got.br_startblock));
5165         ASSERT(bno < got.br_startoff + got.br_blockcount);
5166         *fsb = got.br_startblock + (bno - got.br_startoff);
5167         ifp->if_lastex = lastx;
5168         return 0;
5169 }
5170
5171 /*
5172  * Unmap (remove) blocks from a file.
5173  * If nexts is nonzero then the number of extents to remove is limited to
5174  * that value.  If not all extents in the block range can be removed then
5175  * *done is set.
5176  */
5177 int                                             /* error */
5178 xfs_bunmapi(
5179         xfs_trans_t             *tp,            /* transaction pointer */
5180         struct xfs_inode        *ip,            /* incore inode */
5181         xfs_fileoff_t           bno,            /* starting offset to unmap */
5182         xfs_filblks_t           len,            /* length to unmap in file */
5183         int                     flags,          /* misc flags */
5184         xfs_extnum_t            nexts,          /* number of extents max */
5185         xfs_fsblock_t           *firstblock,    /* first allocated block
5186                                                    controls a.g. for allocs */
5187         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
5188         int                     *done)          /* set if not done yet */
5189 {
5190         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
5191         xfs_bmbt_irec_t         del;            /* extent being deleted */
5192         int                     eof;            /* is deleting at eof */
5193         xfs_bmbt_rec_t          *ep;            /* extent record pointer */
5194         int                     error;          /* error return value */
5195         xfs_extnum_t            extno;          /* extent number in list */
5196         xfs_bmbt_irec_t         got;            /* current extent record */
5197         xfs_ifork_t             *ifp;           /* inode fork pointer */
5198         int                     isrt;           /* freeing in rt area */
5199         xfs_extnum_t            lastx;          /* last extent index used */
5200         int                     logflags;       /* transaction logging flags */
5201         xfs_extlen_t            mod;            /* rt extent offset */
5202         xfs_mount_t             *mp;            /* mount structure */
5203         xfs_extnum_t            nextents;       /* number of file extents */
5204         xfs_bmbt_irec_t         prev;           /* previous extent record */
5205         xfs_fileoff_t           start;          /* first file offset deleted */
5206         int                     tmp_logflags;   /* partial logging flags */
5207         int                     wasdel;         /* was a delayed alloc extent */
5208         int                     whichfork;      /* data or attribute fork */
5209         int                     rsvd;           /* OK to allocate reserved blocks */
5210         xfs_fsblock_t           sum;
5211
5212         xfs_bunmap_trace(ip, bno, len, flags, (inst_t *)__return_address);
5213         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5214                 XFS_ATTR_FORK : XFS_DATA_FORK;
5215         ifp = XFS_IFORK_PTR(ip, whichfork);
5216         if (unlikely(
5217             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5218             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5219                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5220                                  ip->i_mount);
5221                 return XFS_ERROR(EFSCORRUPTED);
5222         }
5223         mp = ip->i_mount;
5224         if (XFS_FORCED_SHUTDOWN(mp))
5225                 return XFS_ERROR(EIO);
5226         rsvd = (flags & XFS_BMAPI_RSVBLOCKS) != 0;
5227         ASSERT(len > 0);
5228         ASSERT(nexts >= 0);
5229         ASSERT(ifp->if_ext_max ==
5230                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5231         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5232             (error = xfs_iread_extents(tp, ip, whichfork)))
5233                 return error;
5234         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5235         if (nextents == 0) {
5236                 *done = 1;
5237                 return 0;
5238         }
5239         XFS_STATS_INC(xs_blk_unmap);
5240         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5241         start = bno;
5242         bno = start + len - 1;
5243         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5244                 &prev);
5245         /*
5246          * Check to see if the given block number is past the end of the
5247          * file, back up to the last block if so...
5248          */
5249         if (eof) {
5250                 ep = xfs_iext_get_ext(ifp, --lastx);
5251                 xfs_bmbt_get_all(ep, &got);
5252                 bno = got.br_startoff + got.br_blockcount - 1;
5253         }
5254         logflags = 0;
5255         if (ifp->if_flags & XFS_IFBROOT) {
5256                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5257                 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
5258                         whichfork);
5259                 cur->bc_private.b.firstblock = *firstblock;
5260                 cur->bc_private.b.flist = flist;
5261                 cur->bc_private.b.flags = 0;
5262         } else
5263                 cur = NULL;
5264         extno = 0;
5265         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5266                (nexts == 0 || extno < nexts)) {
5267                 /*
5268                  * Is the found extent after a hole in which bno lives?
5269                  * Just back up to the previous extent, if so.
5270                  */
5271                 if (got.br_startoff > bno) {
5272                         if (--lastx < 0)
5273                                 break;
5274                         ep = xfs_iext_get_ext(ifp, lastx);
5275                         xfs_bmbt_get_all(ep, &got);
5276                 }
5277                 /*
5278                  * Is the last block of this extent before the range
5279                  * we're supposed to delete?  If so, we're done.
5280                  */
5281                 bno = XFS_FILEOFF_MIN(bno,
5282                         got.br_startoff + got.br_blockcount - 1);
5283                 if (bno < start)
5284                         break;
5285                 /*
5286                  * Then deal with the (possibly delayed) allocated space
5287                  * we found.
5288                  */
5289                 ASSERT(ep != NULL);
5290                 del = got;
5291                 wasdel = ISNULLSTARTBLOCK(del.br_startblock);
5292                 if (got.br_startoff < start) {
5293                         del.br_startoff = start;
5294                         del.br_blockcount -= start - got.br_startoff;
5295                         if (!wasdel)
5296                                 del.br_startblock += start - got.br_startoff;
5297                 }
5298                 if (del.br_startoff + del.br_blockcount > bno + 1)
5299                         del.br_blockcount = bno + 1 - del.br_startoff;
5300                 sum = del.br_startblock + del.br_blockcount;
5301                 if (isrt &&
5302                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5303                         /*
5304                          * Realtime extent not lined up at the end.
5305                          * The extent could have been split into written
5306                          * and unwritten pieces, or we could just be
5307                          * unmapping part of it.  But we can't really
5308                          * get rid of part of a realtime extent.
5309                          */
5310                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5311                             !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
5312                                 /*
5313                                  * This piece is unwritten, or we're not
5314                                  * using unwritten extents.  Skip over it.
5315                                  */
5316                                 ASSERT(bno >= mod);
5317                                 bno -= mod > del.br_blockcount ?
5318                                         del.br_blockcount : mod;
5319                                 if (bno < got.br_startoff) {
5320                                         if (--lastx >= 0)
5321                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5322                                                         ifp, lastx), &got);
5323                                 }
5324                                 continue;
5325                         }
5326                         /*
5327                          * It's written, turn it unwritten.
5328                          * This is better than zeroing it.
5329                          */
5330                         ASSERT(del.br_state == XFS_EXT_NORM);
5331                         ASSERT(xfs_trans_get_block_res(tp) > 0);
5332                         /*
5333                          * If this spans a realtime extent boundary,
5334                          * chop it back to the start of the one we end at.
5335                          */
5336                         if (del.br_blockcount > mod) {
5337                                 del.br_startoff += del.br_blockcount - mod;
5338                                 del.br_startblock += del.br_blockcount - mod;
5339                                 del.br_blockcount = mod;
5340                         }
5341                         del.br_state = XFS_EXT_UNWRITTEN;
5342                         error = xfs_bmap_add_extent(ip, lastx, &cur, &del,
5343                                 firstblock, flist, &logflags, XFS_DATA_FORK, 0);
5344                         if (error)
5345                                 goto error0;
5346                         goto nodelete;
5347                 }
5348                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5349                         /*
5350                          * Realtime extent is lined up at the end but not
5351                          * at the front.  We'll get rid of full extents if
5352                          * we can.
5353                          */
5354                         mod = mp->m_sb.sb_rextsize - mod;
5355                         if (del.br_blockcount > mod) {
5356                                 del.br_blockcount -= mod;
5357                                 del.br_startoff += mod;
5358                                 del.br_startblock += mod;
5359                         } else if ((del.br_startoff == start &&
5360                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5361                                      xfs_trans_get_block_res(tp) == 0)) ||
5362                                    !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
5363                                 /*
5364                                  * Can't make it unwritten.  There isn't
5365                                  * a full extent here so just skip it.
5366                                  */
5367                                 ASSERT(bno >= del.br_blockcount);
5368                                 bno -= del.br_blockcount;
5369                                 if (bno < got.br_startoff) {
5370                                         if (--lastx >= 0)
5371                                                 xfs_bmbt_get_all(--ep, &got);
5372                                 }
5373                                 continue;
5374                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5375                                 /*
5376                                  * This one is already unwritten.
5377                                  * It must have a written left neighbor.
5378                                  * Unwrite the killed part of that one and
5379                                  * try again.
5380                                  */
5381                                 ASSERT(lastx > 0);
5382                                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5383                                                 lastx - 1), &prev);
5384                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5385                                 ASSERT(!ISNULLSTARTBLOCK(prev.br_startblock));
5386                                 ASSERT(del.br_startblock ==
5387                                        prev.br_startblock + prev.br_blockcount);
5388                                 if (prev.br_startoff < start) {
5389                                         mod = start - prev.br_startoff;
5390                                         prev.br_blockcount -= mod;
5391                                         prev.br_startblock += mod;
5392                                         prev.br_startoff = start;
5393                                 }
5394                                 prev.br_state = XFS_EXT_UNWRITTEN;
5395                                 error = xfs_bmap_add_extent(ip, lastx - 1, &cur,
5396                                         &prev, firstblock, flist, &logflags,
5397                                         XFS_DATA_FORK, 0);
5398                                 if (error)
5399                                         goto error0;
5400                                 goto nodelete;
5401                         } else {
5402                                 ASSERT(del.br_state == XFS_EXT_NORM);
5403                                 del.br_state = XFS_EXT_UNWRITTEN;
5404                                 error = xfs_bmap_add_extent(ip, lastx, &cur,
5405                                         &del, firstblock, flist, &logflags,
5406                                         XFS_DATA_FORK, 0);
5407                                 if (error)
5408                                         goto error0;
5409                                 goto nodelete;
5410                         }
5411                 }
5412                 if (wasdel) {
5413                         ASSERT(STARTBLOCKVAL(del.br_startblock) > 0);
5414                         /* Update realtime/data freespace, unreserve quota */
5415                         if (isrt) {
5416                                 xfs_filblks_t rtexts;
5417
5418                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5419                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5420                                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5421                                                 (int)rtexts, rsvd);
5422                                 (void)XFS_TRANS_RESERVE_QUOTA_NBLKS(mp,
5423                                         NULL, ip, -((long)del.br_blockcount), 0,
5424                                         XFS_QMOPT_RES_RTBLKS);
5425                         } else {
5426                                 xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS,
5427                                                 (int)del.br_blockcount, rsvd);
5428                                 (void)XFS_TRANS_RESERVE_QUOTA_NBLKS(mp,
5429                                         NULL, ip, -((long)del.br_blockcount), 0,
5430                                         XFS_QMOPT_RES_REGBLKS);
5431                         }
5432                         ip->i_delayed_blks -= del.br_blockcount;
5433                         if (cur)
5434                                 cur->bc_private.b.flags |=
5435                                         XFS_BTCUR_BPRV_WASDEL;
5436                 } else if (cur)
5437                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5438                 /*
5439                  * If it's the case where the directory code is running
5440                  * with no block reservation, and the deleted block is in
5441                  * the middle of its extent, and the resulting insert
5442                  * of an extent would cause transformation to btree format,
5443                  * then reject it.  The calling code will then swap
5444                  * blocks around instead.
5445                  * We have to do this now, rather than waiting for the
5446                  * conversion to btree format, since the transaction
5447                  * will be dirty.
5448                  */
5449                 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5450                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5451                     XFS_IFORK_NEXTENTS(ip, whichfork) >= ifp->if_ext_max &&
5452                     del.br_startoff > got.br_startoff &&
5453                     del.br_startoff + del.br_blockcount <
5454                     got.br_startoff + got.br_blockcount) {
5455                         error = XFS_ERROR(ENOSPC);
5456                         goto error0;
5457                 }
5458                 error = xfs_bmap_del_extent(ip, tp, lastx, flist, cur, &del,
5459                         &tmp_logflags, whichfork, rsvd);
5460                 logflags |= tmp_logflags;
5461                 if (error)
5462                         goto error0;
5463                 bno = del.br_startoff - 1;
5464 nodelete:
5465                 lastx = ifp->if_lastex;
5466                 /*
5467                  * If not done go on to the next (previous) record.
5468                  * Reset ep in case the extents array was re-alloced.
5469                  */
5470                 ep = xfs_iext_get_ext(ifp, lastx);
5471                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5472                         if (lastx >= XFS_IFORK_NEXTENTS(ip, whichfork) ||
5473                             xfs_bmbt_get_startoff(ep) > bno) {
5474                                 if (--lastx >= 0)
5475                                         ep = xfs_iext_get_ext(ifp, lastx);
5476                         }
5477                         if (lastx >= 0)
5478                                 xfs_bmbt_get_all(ep, &got);
5479                         extno++;
5480                 }
5481         }
5482         ifp->if_lastex = lastx;
5483         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5484         ASSERT(ifp->if_ext_max ==
5485                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5486         /*
5487          * Convert to a btree if necessary.
5488          */
5489         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5490             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
5491                 ASSERT(cur == NULL);
5492                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5493                         &cur, 0, &tmp_logflags, whichfork);
5494                 logflags |= tmp_logflags;
5495                 if (error)
5496                         goto error0;
5497         }
5498         /*
5499          * transform from btree to extents, give it cur
5500          */
5501         else if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5502                  XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5503                 ASSERT(cur != NULL);
5504                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5505                         whichfork);
5506                 logflags |= tmp_logflags;
5507                 if (error)
5508                         goto error0;
5509         }
5510         /*
5511          * transform from extents to local?
5512          */
5513         ASSERT(ifp->if_ext_max ==
5514                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5515         error = 0;
5516 error0:
5517         /*
5518          * Log everything.  Do this after conversion, there's no point in
5519          * logging the extent records if we've converted to btree format.
5520          */
5521         if ((logflags & XFS_ILOG_FEXT(whichfork)) &&
5522             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5523                 logflags &= ~XFS_ILOG_FEXT(whichfork);
5524         else if ((logflags & XFS_ILOG_FBROOT(whichfork)) &&
5525                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5526                 logflags &= ~XFS_ILOG_FBROOT(whichfork);
5527         /*
5528          * Log inode even in the error case, if the transaction
5529          * is dirty we'll need to shut down the filesystem.
5530          */
5531         if (logflags)
5532                 xfs_trans_log_inode(tp, ip, logflags);
5533         if (cur) {
5534                 if (!error) {
5535                         *firstblock = cur->bc_private.b.firstblock;
5536                         cur->bc_private.b.allocated = 0;
5537                 }
5538                 xfs_btree_del_cursor(cur,
5539                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5540         }
5541         return error;
5542 }
5543
5544 /*
5545  * Fcntl interface to xfs_bmapi.
5546  */
5547 int                                             /* error code */
5548 xfs_getbmap(
5549         bhv_desc_t              *bdp,           /* XFS behavior descriptor*/
5550         struct getbmap          *bmv,           /* user bmap structure */
5551         void                    __user *ap,     /* pointer to user's array */
5552         int                     interface)      /* interface flags */
5553 {
5554         __int64_t               bmvend;         /* last block requested */
5555         int                     error;          /* return value */
5556         __int64_t               fixlen;         /* length for -1 case */
5557         int                     i;              /* extent number */
5558         xfs_inode_t             *ip;            /* xfs incore inode pointer */
5559         vnode_t                 *vp;            /* corresponding vnode */
5560         int                     lock;           /* lock state */
5561         xfs_bmbt_irec_t         *map;           /* buffer for user's data */
5562         xfs_mount_t             *mp;            /* file system mount point */
5563         int                     nex;            /* # of user extents can do */
5564         int                     nexleft;        /* # of user extents left */
5565         int                     subnex;         /* # of bmapi's can do */
5566         int                     nmap;           /* number of map entries */
5567         struct getbmap          out;            /* output structure */
5568         int                     whichfork;      /* data or attr fork */
5569         int                     prealloced;     /* this is a file with
5570                                                  * preallocated data space */
5571         int                     sh_unwritten;   /* true, if unwritten */
5572                                                 /* extents listed separately */
5573         int                     bmapi_flags;    /* flags for xfs_bmapi */
5574         __int32_t               oflags;         /* getbmapx bmv_oflags field */
5575
5576         vp = BHV_TO_VNODE(bdp);
5577         ip = XFS_BHVTOI(bdp);
5578         mp = ip->i_mount;
5579
5580         whichfork = interface & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5581         sh_unwritten = (interface & BMV_IF_PREALLOC) != 0;
5582
5583         /*      If the BMV_IF_NO_DMAPI_READ interface bit specified, do not
5584          *      generate a DMAPI read event.  Otherwise, if the DM_EVENT_READ
5585          *      bit is set for the file, generate a read event in order
5586          *      that the DMAPI application may do its thing before we return
5587          *      the extents.  Usually this means restoring user file data to
5588          *      regions of the file that look like holes.
5589          *
5590          *      The "old behavior" (from XFS_IOC_GETBMAP) is to not specify
5591          *      BMV_IF_NO_DMAPI_READ so that read events are generated.
5592          *      If this were not true, callers of ioctl( XFS_IOC_GETBMAP )
5593          *      could misinterpret holes in a DMAPI file as true holes,
5594          *      when in fact they may represent offline user data.
5595          */
5596         if (   (interface & BMV_IF_NO_DMAPI_READ) == 0
5597             && DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_READ)
5598             && whichfork == XFS_DATA_FORK) {
5599
5600                 error = XFS_SEND_DATA(mp, DM_EVENT_READ, vp, 0, 0, 0, NULL);
5601                 if (error)
5602                         return XFS_ERROR(error);
5603         }
5604
5605         if (whichfork == XFS_ATTR_FORK) {
5606                 if (XFS_IFORK_Q(ip)) {
5607                         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5608                             ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5609                             ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5610                                 return XFS_ERROR(EINVAL);
5611                 } else if (unlikely(
5612                            ip->i_d.di_aformat != 0 &&
5613                            ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5614                         XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5615                                          ip->i_mount);
5616                         return XFS_ERROR(EFSCORRUPTED);
5617                 }
5618         } else if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5619                    ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5620                    ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5621                 return XFS_ERROR(EINVAL);
5622         if (whichfork == XFS_DATA_FORK) {
5623                 if ((ip->i_d.di_extsize && (ip->i_d.di_flags &
5624                                 (XFS_DIFLAG_REALTIME|XFS_DIFLAG_EXTSIZE))) ||
5625                     ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5626                         prealloced = 1;
5627                         fixlen = XFS_MAXIOFFSET(mp);
5628                 } else {
5629                         prealloced = 0;
5630                         fixlen = ip->i_d.di_size;
5631                 }
5632         } else {
5633                 prealloced = 0;
5634                 fixlen = 1LL << 32;
5635         }
5636
5637         if (bmv->bmv_length == -1) {
5638                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5639                 bmv->bmv_length = MAX( (__int64_t)(fixlen - bmv->bmv_offset),
5640                                         (__int64_t)0);
5641         } else if (bmv->bmv_length < 0)
5642                 return XFS_ERROR(EINVAL);
5643         if (bmv->bmv_length == 0) {
5644                 bmv->bmv_entries = 0;
5645                 return 0;
5646         }
5647         nex = bmv->bmv_count - 1;
5648         if (nex <= 0)
5649                 return XFS_ERROR(EINVAL);
5650         bmvend = bmv->bmv_offset + bmv->bmv_length;
5651
5652         xfs_ilock(ip, XFS_IOLOCK_SHARED);
5653
5654         if (whichfork == XFS_DATA_FORK && ip->i_delayed_blks) {
5655                 /* xfs_fsize_t last_byte = xfs_file_last_byte(ip); */
5656                 VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1, 0, FI_REMAPF, error);
5657         }
5658
5659         ASSERT(whichfork == XFS_ATTR_FORK || ip->i_delayed_blks == 0);
5660
5661         lock = xfs_ilock_map_shared(ip);
5662
5663         /*
5664          * Don't let nex be bigger than the number of extents
5665          * we can have assuming alternating holes and real extents.
5666          */
5667         if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5668                 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5669
5670         bmapi_flags = XFS_BMAPI_AFLAG(whichfork) |
5671                         ((sh_unwritten) ? 0 : XFS_BMAPI_IGSTATE);
5672
5673         /*
5674          * Allocate enough space to handle "subnex" maps at a time.
5675          */
5676         subnex = 16;
5677         map = kmem_alloc(subnex * sizeof(*map), KM_SLEEP);
5678
5679         bmv->bmv_entries = 0;
5680
5681         if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0) {
5682                 error = 0;
5683                 goto unlock_and_return;
5684         }
5685
5686         nexleft = nex;
5687
5688         do {
5689                 nmap = (nexleft > subnex) ? subnex : nexleft;
5690                 error = xfs_bmapi(NULL, ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5691                                   XFS_BB_TO_FSB(mp, bmv->bmv_length),
5692                                   bmapi_flags, NULL, 0, map, &nmap, NULL);
5693                 if (error)
5694                         goto unlock_and_return;
5695                 ASSERT(nmap <= subnex);
5696
5697                 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5698                         nexleft--;
5699                         oflags = (map[i].br_state == XFS_EXT_UNWRITTEN) ?
5700                                         BMV_OF_PREALLOC : 0;
5701                         out.bmv_offset = XFS_FSB_TO_BB(mp, map[i].br_startoff);
5702                         out.bmv_length = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5703                         ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
5704                         if (map[i].br_startblock == HOLESTARTBLOCK &&
5705                            ((prealloced && out.bmv_offset + out.bmv_length == bmvend) ||
5706                              whichfork == XFS_ATTR_FORK )) {
5707                                 /*
5708                                  * came to hole at end of file or the end of
5709                                    attribute fork
5710                                  */
5711                                 goto unlock_and_return;
5712                         } else {
5713                                 out.bmv_block =
5714                                     (map[i].br_startblock == HOLESTARTBLOCK) ?
5715                                         -1 :
5716                                         XFS_FSB_TO_DB(ip, map[i].br_startblock);
5717
5718                                 /* return either getbmap/getbmapx structure. */
5719                                 if (interface & BMV_IF_EXTENDED) {
5720                                         struct  getbmapx        outx;
5721
5722                                         GETBMAP_CONVERT(out,outx);
5723                                         outx.bmv_oflags = oflags;
5724                                         outx.bmv_unused1 = outx.bmv_unused2 = 0;
5725                                         if (copy_to_user(ap, &outx,
5726                                                         sizeof(outx))) {
5727                                                 error = XFS_ERROR(EFAULT);
5728                                                 goto unlock_and_return;
5729                                         }
5730                                 } else {
5731                                         if (copy_to_user(ap, &out,
5732                                                         sizeof(out))) {
5733                                                 error = XFS_ERROR(EFAULT);
5734                                                 goto unlock_and_return;
5735                                         }
5736                                 }
5737                                 bmv->bmv_offset =
5738                                         out.bmv_offset + out.bmv_length;
5739                                 bmv->bmv_length = MAX((__int64_t)0,
5740                                         (__int64_t)(bmvend - bmv->bmv_offset));
5741                                 bmv->bmv_entries++;
5742                                 ap = (interface & BMV_IF_EXTENDED) ?
5743                                                 (void __user *)
5744                                         ((struct getbmapx __user *)ap + 1) :
5745                                                 (void __user *)
5746                                         ((struct getbmap __user *)ap + 1);
5747                         }
5748                 }
5749         } while (nmap && nexleft && bmv->bmv_length);
5750
5751 unlock_and_return:
5752         xfs_iunlock_map_shared(ip, lock);
5753         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5754
5755         kmem_free(map, subnex * sizeof(*map));
5756
5757         return error;
5758 }
5759
5760 /*
5761  * Check the last inode extent to determine whether this allocation will result
5762  * in blocks being allocated at the end of the file. When we allocate new data
5763  * blocks at the end of the file which do not start at the previous data block,
5764  * we will try to align the new blocks at stripe unit boundaries.
5765  */
5766 STATIC int                              /* error */
5767 xfs_bmap_isaeof(
5768         xfs_inode_t     *ip,            /* incore inode pointer */
5769         xfs_fileoff_t   off,            /* file offset in fsblocks */
5770         int             whichfork,      /* data or attribute fork */
5771         char            *aeof)          /* return value */
5772 {
5773         int             error;          /* error return value */
5774         xfs_ifork_t     *ifp;           /* inode fork pointer */
5775         xfs_bmbt_rec_t  *lastrec;       /* extent record pointer */
5776         xfs_extnum_t    nextents;       /* number of file extents */
5777         xfs_bmbt_irec_t s;              /* expanded extent record */
5778
5779         ASSERT(whichfork == XFS_DATA_FORK);
5780         ifp = XFS_IFORK_PTR(ip, whichfork);
5781         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5782             (error = xfs_iread_extents(NULL, ip, whichfork)))
5783                 return error;
5784         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5785         if (nextents == 0) {
5786                 *aeof = 1;
5787                 return 0;
5788         }
5789         /*
5790          * Go to the last extent
5791          */
5792         lastrec = xfs_iext_get_ext(ifp, nextents - 1);
5793         xfs_bmbt_get_all(lastrec, &s);
5794         /*
5795          * Check we are allocating in the last extent (for delayed allocations)
5796          * or past the last extent for non-delayed allocations.
5797          */
5798         *aeof = (off >= s.br_startoff &&
5799                  off < s.br_startoff + s.br_blockcount &&
5800                  ISNULLSTARTBLOCK(s.br_startblock)) ||
5801                 off >= s.br_startoff + s.br_blockcount;
5802         return 0;
5803 }
5804
5805 /*
5806  * Check if the endoff is outside the last extent. If so the caller will grow
5807  * the allocation to a stripe unit boundary.
5808  */
5809 int                                     /* error */
5810 xfs_bmap_eof(
5811         xfs_inode_t     *ip,            /* incore inode pointer */
5812         xfs_fileoff_t   endoff,         /* file offset in fsblocks */
5813         int             whichfork,      /* data or attribute fork */
5814         int             *eof)           /* result value */
5815 {
5816         xfs_fsblock_t   blockcount;     /* extent block count */
5817         int             error;          /* error return value */
5818         xfs_ifork_t     *ifp;           /* inode fork pointer */
5819         xfs_bmbt_rec_t  *lastrec;       /* extent record pointer */
5820         xfs_extnum_t    nextents;       /* number of file extents */
5821         xfs_fileoff_t   startoff;       /* extent starting file offset */
5822
5823         ASSERT(whichfork == XFS_DATA_FORK);
5824         ifp = XFS_IFORK_PTR(ip, whichfork);
5825         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5826             (error = xfs_iread_extents(NULL, ip, whichfork)))
5827                 return error;
5828         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5829         if (nextents == 0) {
5830                 *eof = 1;
5831                 return 0;
5832         }
5833         /*
5834          * Go to the last extent
5835          */
5836         lastrec = xfs_iext_get_ext(ifp, nextents - 1);
5837         startoff = xfs_bmbt_get_startoff(lastrec);
5838         blockcount = xfs_bmbt_get_blockcount(lastrec);
5839         *eof = endoff >= startoff + blockcount;
5840         return 0;
5841 }
5842
5843 #ifdef DEBUG
5844 /*
5845  * Check that the extents list for the inode ip is in the right order.
5846  */
5847 STATIC void
5848 xfs_bmap_check_extents(
5849         xfs_inode_t             *ip,            /* incore inode pointer */
5850         int                     whichfork)      /* data or attr fork */
5851 {
5852         xfs_bmbt_rec_t          *ep;            /* current extent entry */
5853         xfs_extnum_t            idx;            /* extent record index */
5854         xfs_ifork_t             *ifp;           /* inode fork pointer */
5855         xfs_extnum_t            nextents;       /* number of extents in list */
5856         xfs_bmbt_rec_t          *nextp;         /* next extent entry */
5857
5858         ifp = XFS_IFORK_PTR(ip, whichfork);
5859         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
5860         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5861         ep = xfs_iext_get_ext(ifp, 0);
5862         for (idx = 0; idx < nextents - 1; idx++) {
5863                 nextp = xfs_iext_get_ext(ifp, idx + 1);
5864                 xfs_btree_check_rec(XFS_BTNUM_BMAP, (void *)ep,
5865                         (void *)(nextp));
5866                 ep = nextp;
5867         }
5868 }
5869
5870 STATIC
5871 xfs_buf_t *
5872 xfs_bmap_get_bp(
5873         xfs_btree_cur_t         *cur,
5874         xfs_fsblock_t           bno)
5875 {
5876         int i;
5877         xfs_buf_t *bp;
5878
5879         if (!cur)
5880                 return(NULL);
5881
5882         bp = NULL;
5883         for(i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
5884                 bp = cur->bc_bufs[i];
5885                 if (!bp) break;
5886                 if (XFS_BUF_ADDR(bp) == bno)
5887                         break;  /* Found it */
5888         }
5889         if (i == XFS_BTREE_MAXLEVELS)
5890                 bp = NULL;
5891
5892         if (!bp) { /* Chase down all the log items to see if the bp is there */
5893                 xfs_log_item_chunk_t    *licp;
5894                 xfs_trans_t             *tp;
5895
5896                 tp = cur->bc_tp;
5897                 licp = &tp->t_items;
5898                 while (!bp && licp != NULL) {
5899                         if (XFS_LIC_ARE_ALL_FREE(licp)) {
5900                                 licp = licp->lic_next;
5901                                 continue;
5902                         }
5903                         for (i = 0; i < licp->lic_unused; i++) {
5904                                 xfs_log_item_desc_t     *lidp;
5905                                 xfs_log_item_t          *lip;
5906                                 xfs_buf_log_item_t      *bip;
5907                                 xfs_buf_t               *lbp;
5908
5909                                 if (XFS_LIC_ISFREE(licp, i)) {
5910                                         continue;
5911                                 }
5912
5913                                 lidp = XFS_LIC_SLOT(licp, i);
5914                                 lip = lidp->lid_item;
5915                                 if (lip->li_type != XFS_LI_BUF)
5916                                         continue;
5917
5918                                 bip = (xfs_buf_log_item_t *)lip;
5919                                 lbp = bip->bli_buf;
5920
5921                                 if (XFS_BUF_ADDR(lbp) == bno) {
5922                                         bp = lbp;
5923                                         break; /* Found it */
5924                                 }
5925                         }
5926                         licp = licp->lic_next;
5927                 }
5928         }
5929         return(bp);
5930 }
5931
5932 void
5933 xfs_check_block(
5934         xfs_bmbt_block_t        *block,
5935         xfs_mount_t             *mp,
5936         int                     root,
5937         short                   sz)
5938 {
5939         int                     i, j, dmxr;
5940         xfs_bmbt_ptr_t          *pp, *thispa;   /* pointer to block address */
5941         xfs_bmbt_key_t          *prevp, *keyp;
5942
5943         ASSERT(be16_to_cpu(block->bb_level) > 0);
5944
5945         prevp = NULL;
5946         for( i = 1; i <= be16_to_cpu(block->bb_numrecs); i++) {
5947                 dmxr = mp->m_bmap_dmxr[0];
5948
5949                 if (root) {
5950                         keyp = XFS_BMAP_BROOT_KEY_ADDR(block, i, sz);
5951                 } else {
5952                         keyp = XFS_BTREE_KEY_ADDR(mp->m_sb.sb_blocksize,
5953                                 xfs_bmbt, block, i, dmxr);
5954                 }
5955
5956                 if (prevp) {
5957                         xfs_btree_check_key(XFS_BTNUM_BMAP, prevp, keyp);
5958                 }
5959                 prevp = keyp;
5960
5961                 /*
5962                  * Compare the block numbers to see if there are dups.
5963                  */
5964
5965                 if (root) {
5966                         pp = XFS_BMAP_BROOT_PTR_ADDR(block, i, sz);
5967                 } else {
5968                         pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize,
5969                                 xfs_bmbt, block, i, dmxr);
5970                 }
5971                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
5972                         if (root) {
5973                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(block, j, sz);
5974                         } else {
5975                                 thispa = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize,
5976                                         xfs_bmbt, block, j, dmxr);
5977                         }
5978                         if (INT_GET(*thispa, ARCH_CONVERT) ==
5979                             INT_GET(*pp, ARCH_CONVERT)) {
5980                                 cmn_err(CE_WARN, "%s: thispa(%d) == pp(%d) %Ld",
5981                                         __FUNCTION__, j, i,
5982                                         INT_GET(*thispa, ARCH_CONVERT));
5983                                 panic("%s: ptrs are equal in node\n",
5984                                         __FUNCTION__);
5985                         }
5986                 }
5987         }
5988 }
5989
5990 /*
5991  * Check that the extents for the inode ip are in the right order in all
5992  * btree leaves.
5993  */
5994
5995 STATIC void
5996 xfs_bmap_check_leaf_extents(
5997         xfs_btree_cur_t         *cur,   /* btree cursor or null */
5998         xfs_inode_t             *ip,            /* incore inode pointer */
5999         int                     whichfork)      /* data or attr fork */
6000 {
6001         xfs_bmbt_block_t        *block; /* current btree block */
6002         xfs_fsblock_t           bno;    /* block # of "block" */
6003         xfs_buf_t               *bp;    /* buffer for "block" */
6004         int                     error;  /* error return value */
6005         xfs_extnum_t            i=0, j; /* index into the extents list */
6006         xfs_ifork_t             *ifp;   /* fork structure */
6007         int                     level;  /* btree level, for checking */
6008         xfs_mount_t             *mp;    /* file system mount structure */
6009         xfs_bmbt_ptr_t          *pp;    /* pointer to block address */
6010         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
6011         xfs_bmbt_rec_t          *lastp; /* pointer to previous extent */
6012         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
6013         int                     bp_release = 0;
6014
6015         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
6016                 return;
6017         }
6018
6019         bno = NULLFSBLOCK;
6020         mp = ip->i_mount;
6021         ifp = XFS_IFORK_PTR(ip, whichfork);
6022         block = ifp->if_broot;
6023         /*
6024          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6025          */
6026         level = be16_to_cpu(block->bb_level);
6027         ASSERT(level > 0);
6028         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
6029         pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
6030         ASSERT(INT_GET(*pp, ARCH_CONVERT) != NULLDFSBNO);
6031         ASSERT(XFS_FSB_TO_AGNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agcount);
6032         ASSERT(XFS_FSB_TO_AGBNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agblocks);
6033         bno = INT_GET(*pp, ARCH_CONVERT);
6034         /*
6035          * Go down the tree until leaf level is reached, following the first
6036          * pointer (leftmost) at each level.
6037          */
6038         while (level-- > 0) {
6039                 /* See if buf is in cur first */
6040                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
6041                 if (bp) {
6042                         bp_release = 0;
6043                 } else {
6044                         bp_release = 1;
6045                 }
6046                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
6047                                 XFS_BMAP_BTREE_REF)))
6048                         goto error_norelse;
6049                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6050                 XFS_WANT_CORRUPTED_GOTO(
6051                         XFS_BMAP_SANITY_CHECK(mp, block, level),
6052                         error0);
6053                 if (level == 0)
6054                         break;
6055
6056                 /*
6057                  * Check this block for basic sanity (increasing keys and
6058                  * no duplicate blocks).
6059                  */
6060
6061                 xfs_check_block(block, mp, 0, 0);
6062                 pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt, block,
6063                         1, mp->m_bmap_dmxr[1]);
6064                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, INT_GET(*pp, ARCH_CONVERT)), error0);
6065                 bno = INT_GET(*pp, ARCH_CONVERT);
6066                 if (bp_release) {
6067                         bp_release = 0;
6068                         xfs_trans_brelse(NULL, bp);
6069                 }
6070         }
6071
6072         /*
6073          * Here with bp and block set to the leftmost leaf node in the tree.
6074          */
6075         i = 0;
6076
6077         /*
6078          * Loop over all leaf nodes checking that all extents are in the right order.
6079          */
6080         lastp = NULL;
6081         for (;;) {
6082                 xfs_fsblock_t   nextbno;
6083                 xfs_extnum_t    num_recs;
6084
6085
6086                 num_recs = be16_to_cpu(block->bb_numrecs);
6087
6088                 /*
6089                  * Read-ahead the next leaf block, if any.
6090                  */
6091
6092                 nextbno = be64_to_cpu(block->bb_rightsib);
6093
6094                 /*
6095                  * Check all the extents to make sure they are OK.
6096                  * If we had a previous block, the last entry should
6097                  * conform with the first entry in this one.
6098                  */
6099
6100                 ep = XFS_BTREE_REC_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt,
6101                         block, 1, mp->m_bmap_dmxr[0]);
6102                 for (j = 1; j < num_recs; j++) {
6103                         nextp = XFS_BTREE_REC_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt,
6104                                 block, j + 1, mp->m_bmap_dmxr[0]);
6105                         if (lastp) {
6106                                 xfs_btree_check_rec(XFS_BTNUM_BMAP,
6107                                         (void *)lastp, (void *)ep);
6108                         }
6109                         xfs_btree_check_rec(XFS_BTNUM_BMAP, (void *)ep,
6110                                 (void *)(nextp));
6111                         lastp = ep;
6112                         ep = nextp;
6113                 }
6114
6115                 i += num_recs;
6116                 if (bp_release) {
6117                         bp_release = 0;
6118                         xfs_trans_brelse(NULL, bp);
6119                 }
6120                 bno = nextbno;
6121                 /*
6122                  * If we've reached the end, stop.
6123                  */
6124                 if (bno == NULLFSBLOCK)
6125                         break;
6126
6127                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
6128                 if (bp) {
6129                         bp_release = 0;
6130                 } else {
6131                         bp_release = 1;
6132                 }
6133                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
6134                                 XFS_BMAP_BTREE_REF)))
6135                         goto error_norelse;
6136                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6137         }
6138         if (bp_release) {
6139                 bp_release = 0;
6140                 xfs_trans_brelse(NULL, bp);
6141         }
6142         return;
6143
6144 error0:
6145         cmn_err(CE_WARN, "%s: at error0", __FUNCTION__);
6146         if (bp_release)
6147                 xfs_trans_brelse(NULL, bp);
6148 error_norelse:
6149         cmn_err(CE_WARN, "%s: BAD after btree leaves for %d extents",
6150                 __FUNCTION__, i);
6151         panic("%s: CORRUPTED BTREE OR SOMETHING", __FUNCTION__);
6152         return;
6153 }
6154 #endif
6155
6156 /*
6157  * Count fsblocks of the given fork.
6158  */
6159 int                                             /* error */
6160 xfs_bmap_count_blocks(
6161         xfs_trans_t             *tp,            /* transaction pointer */
6162         xfs_inode_t             *ip,            /* incore inode */
6163         int                     whichfork,      /* data or attr fork */
6164         int                     *count)         /* out: count of blocks */
6165 {
6166         xfs_bmbt_block_t        *block; /* current btree block */
6167         xfs_fsblock_t           bno;    /* block # of "block" */
6168         xfs_ifork_t             *ifp;   /* fork structure */
6169         int                     level;  /* btree level, for checking */
6170         xfs_mount_t             *mp;    /* file system mount structure */
6171         xfs_bmbt_ptr_t          *pp;    /* pointer to block address */
6172
6173         bno = NULLFSBLOCK;
6174         mp = ip->i_mount;
6175         ifp = XFS_IFORK_PTR(ip, whichfork);
6176         if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
6177                 if (unlikely(xfs_bmap_count_leaves(ifp, 0,
6178                         ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
6179                         count) < 0)) {
6180                         XFS_ERROR_REPORT("xfs_bmap_count_blocks(1)",
6181                                          XFS_ERRLEVEL_LOW, mp);
6182                         return XFS_ERROR(EFSCORRUPTED);
6183                 }
6184                 return 0;
6185         }
6186
6187         /*
6188          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6189          */
6190         block = ifp->if_broot;
6191         level = be16_to_cpu(block->bb_level);
6192         ASSERT(level > 0);
6193         pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
6194         ASSERT(INT_GET(*pp, ARCH_CONVERT) != NULLDFSBNO);
6195         ASSERT(XFS_FSB_TO_AGNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agcount);
6196         ASSERT(XFS_FSB_TO_AGBNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agblocks);
6197         bno = INT_GET(*pp, ARCH_CONVERT);
6198
6199         if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
6200                 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
6201                                  mp);
6202                 return XFS_ERROR(EFSCORRUPTED);
6203         }
6204
6205         return 0;
6206 }
6207
6208 /*
6209  * Recursively walks each level of a btree
6210  * to count total fsblocks is use.
6211  */
6212 int                                     /* error */
6213 xfs_bmap_count_tree(
6214         xfs_mount_t     *mp,            /* file system mount point */
6215         xfs_trans_t     *tp,            /* transaction pointer */
6216         xfs_ifork_t     *ifp,           /* inode fork pointer */
6217         xfs_fsblock_t   blockno,        /* file system block number */
6218         int             levelin,        /* level in btree */
6219         int             *count)         /* Count of blocks */
6220 {
6221         int                     error;
6222         xfs_buf_t               *bp, *nbp;
6223         int                     level = levelin;
6224         xfs_bmbt_ptr_t          *pp;
6225         xfs_fsblock_t           bno = blockno;
6226         xfs_fsblock_t           nextbno;
6227         xfs_bmbt_block_t        *block, *nextblock;
6228         int                     numrecs;
6229
6230         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
6231                 return error;
6232         *count += 1;
6233         block = XFS_BUF_TO_BMBT_BLOCK(bp);
6234
6235         if (--level) {
6236                 /* Not at node above leafs, count this level of nodes */
6237                 nextbno = be64_to_cpu(block->bb_rightsib);
6238                 while (nextbno != NULLFSBLOCK) {
6239                         if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
6240                                 0, &nbp, XFS_BMAP_BTREE_REF)))
6241                                 return error;
6242                         *count += 1;
6243                         nextblock = XFS_BUF_TO_BMBT_BLOCK(nbp);
6244                         nextbno = be64_to_cpu(nextblock->bb_rightsib);
6245                         xfs_trans_brelse(tp, nbp);
6246                 }
6247
6248                 /* Dive to the next level */
6249                 pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize,
6250                         xfs_bmbt, block, 1, mp->m_bmap_dmxr[1]);
6251                 bno = INT_GET(*pp, ARCH_CONVERT);
6252                 if (unlikely((error =
6253                      xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
6254                         xfs_trans_brelse(tp, bp);
6255                         XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6256                                          XFS_ERRLEVEL_LOW, mp);
6257                         return XFS_ERROR(EFSCORRUPTED);
6258                 }
6259                 xfs_trans_brelse(tp, bp);
6260         } else {
6261                 /* count all level 1 nodes and their leaves */
6262                 for (;;) {
6263                         nextbno = be64_to_cpu(block->bb_rightsib);
6264                         numrecs = be16_to_cpu(block->bb_numrecs);
6265                         if (unlikely(xfs_bmap_disk_count_leaves(ifp, mp,
6266                                         0, block, numrecs, count) < 0)) {
6267                                 xfs_trans_brelse(tp, bp);
6268                                 XFS_ERROR_REPORT("xfs_bmap_count_tree(2)",
6269                                                  XFS_ERRLEVEL_LOW, mp);
6270                                 return XFS_ERROR(EFSCORRUPTED);
6271                         }
6272                         xfs_trans_brelse(tp, bp);
6273                         if (nextbno == NULLFSBLOCK)
6274                                 break;
6275                         bno = nextbno;
6276                         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6277                                 XFS_BMAP_BTREE_REF)))
6278                                 return error;
6279                         *count += 1;
6280                         block = XFS_BUF_TO_BMBT_BLOCK(bp);
6281                 }
6282         }
6283         return 0;
6284 }
6285
6286 /*
6287  * Count leaf blocks given a range of extent records.
6288  */
6289 int
6290 xfs_bmap_count_leaves(
6291         xfs_ifork_t             *ifp,
6292         xfs_extnum_t            idx,
6293         int                     numrecs,
6294         int                     *count)
6295 {
6296         int             b;
6297         xfs_bmbt_rec_t  *frp;
6298
6299         for (b = 0; b < numrecs; b++) {
6300                 frp = xfs_iext_get_ext(ifp, idx + b);
6301                 *count += xfs_bmbt_get_blockcount(frp);
6302         }
6303         return 0;
6304 }
6305
6306 /*
6307  * Count leaf blocks given a range of extent records originally
6308  * in btree format.
6309  */
6310 int
6311 xfs_bmap_disk_count_leaves(
6312         xfs_ifork_t             *ifp,
6313         xfs_mount_t             *mp,
6314         xfs_extnum_t            idx,
6315         xfs_bmbt_block_t        *block,
6316         int                     numrecs,
6317         int                     *count)
6318 {
6319         int             b;
6320         xfs_bmbt_rec_t  *frp;
6321
6322         for (b = 1; b <= numrecs; b++) {
6323                 frp = XFS_BTREE_REC_ADDR(mp->m_sb.sb_blocksize,
6324                         xfs_bmbt, block, idx + b, mp->m_bmap_dmxr[0]);
6325                 *count += xfs_bmbt_disk_get_blockcount(frp);
6326         }
6327         return 0;
6328 }