7b46b78ba3f011b1ed32ab7c2ee09c959650c361
[pandora-kernel.git] / fs / xfs / xfs_bmap.c
1 /*
2  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write the Free Software Foundation,
16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17  */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_inum.h"
24 #include "xfs_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_dir2.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_alloc_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_dir2_sf.h"
33 #include "xfs_dinode.h"
34 #include "xfs_inode.h"
35 #include "xfs_btree.h"
36 #include "xfs_mount.h"
37 #include "xfs_itable.h"
38 #include "xfs_dir2_data.h"
39 #include "xfs_dir2_leaf.h"
40 #include "xfs_dir2_block.h"
41 #include "xfs_inode_item.h"
42 #include "xfs_extfree_item.h"
43 #include "xfs_alloc.h"
44 #include "xfs_bmap.h"
45 #include "xfs_rtalloc.h"
46 #include "xfs_error.h"
47 #include "xfs_attr_leaf.h"
48 #include "xfs_rw.h"
49 #include "xfs_quota.h"
50 #include "xfs_trans_space.h"
51 #include "xfs_buf_item.h"
52 #include "xfs_filestream.h"
53 #include "xfs_vnodeops.h"
54 #include "xfs_trace.h"
55
56
57 #ifdef DEBUG
58 STATIC void
59 xfs_bmap_check_leaf_extents(xfs_btree_cur_t *cur, xfs_inode_t *ip, int whichfork);
60 #endif
61
62 kmem_zone_t             *xfs_bmap_free_item_zone;
63
64 /*
65  * Prototypes for internal bmap routines.
66  */
67
68
69 /*
70  * Called from xfs_bmap_add_attrfork to handle extents format files.
71  */
72 STATIC int                                      /* error */
73 xfs_bmap_add_attrfork_extents(
74         xfs_trans_t             *tp,            /* transaction pointer */
75         xfs_inode_t             *ip,            /* incore inode pointer */
76         xfs_fsblock_t           *firstblock,    /* first block allocated */
77         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
78         int                     *flags);        /* inode logging flags */
79
80 /*
81  * Called from xfs_bmap_add_attrfork to handle local format files.
82  */
83 STATIC int                                      /* error */
84 xfs_bmap_add_attrfork_local(
85         xfs_trans_t             *tp,            /* transaction pointer */
86         xfs_inode_t             *ip,            /* incore inode pointer */
87         xfs_fsblock_t           *firstblock,    /* first block allocated */
88         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
89         int                     *flags);        /* inode logging flags */
90
91 /*
92  * Called by xfs_bmap_add_extent to handle cases converting a delayed
93  * allocation to a real allocation.
94  */
95 STATIC int                              /* error */
96 xfs_bmap_add_extent_delay_real(
97         struct xfs_trans        *tp,    /* transaction pointer */
98         xfs_inode_t             *ip,    /* incore inode pointer */
99         xfs_extnum_t            *idx,   /* extent number to update/insert */
100         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
101         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
102         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
103         xfs_fsblock_t           *first, /* pointer to firstblock variable */
104         xfs_bmap_free_t         *flist, /* list of extents to be freed */
105         int                     *logflagsp); /* inode logging flags */
106
107 /*
108  * Called by xfs_bmap_add_extent to handle cases converting a hole
109  * to a delayed allocation.
110  */
111 STATIC int                              /* error */
112 xfs_bmap_add_extent_hole_delay(
113         xfs_inode_t             *ip,    /* incore inode pointer */
114         xfs_extnum_t            *idx,   /* extent number to update/insert */
115         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
116         int                     *logflagsp); /* inode logging flags */
117
118 /*
119  * Called by xfs_bmap_add_extent to handle cases converting a hole
120  * to a real allocation.
121  */
122 STATIC int                              /* error */
123 xfs_bmap_add_extent_hole_real(
124         xfs_inode_t             *ip,    /* incore inode pointer */
125         xfs_extnum_t            *idx,   /* extent number to update/insert */
126         xfs_btree_cur_t         *cur,   /* if null, not a btree */
127         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
128         int                     *logflagsp, /* inode logging flags */
129         int                     whichfork); /* data or attr fork */
130
131 /*
132  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
133  * allocation to a real allocation or vice versa.
134  */
135 STATIC int                              /* error */
136 xfs_bmap_add_extent_unwritten_real(
137         xfs_inode_t             *ip,    /* incore inode pointer */
138         xfs_extnum_t            *idx,   /* extent number to update/insert */
139         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
140         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
141         int                     *logflagsp); /* inode logging flags */
142
143 /*
144  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
145  * It figures out where to ask the underlying allocator to put the new extent.
146  */
147 STATIC int                              /* error */
148 xfs_bmap_alloc(
149         xfs_bmalloca_t          *ap);   /* bmap alloc argument struct */
150
151 /*
152  * Transform a btree format file with only one leaf node, where the
153  * extents list will fit in the inode, into an extents format file.
154  * Since the file extents are already in-core, all we have to do is
155  * give up the space for the btree root and pitch the leaf block.
156  */
157 STATIC int                              /* error */
158 xfs_bmap_btree_to_extents(
159         xfs_trans_t             *tp,    /* transaction pointer */
160         xfs_inode_t             *ip,    /* incore inode pointer */
161         xfs_btree_cur_t         *cur,   /* btree cursor */
162         int                     *logflagsp, /* inode logging flags */
163         int                     whichfork); /* data or attr fork */
164
165 /*
166  * Remove the entry "free" from the free item list.  Prev points to the
167  * previous entry, unless "free" is the head of the list.
168  */
169 STATIC void
170 xfs_bmap_del_free(
171         xfs_bmap_free_t         *flist, /* free item list header */
172         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
173         xfs_bmap_free_item_t    *free); /* list item to be freed */
174
175 /*
176  * Convert an extents-format file into a btree-format file.
177  * The new file will have a root block (in the inode) and a single child block.
178  */
179 STATIC int                                      /* error */
180 xfs_bmap_extents_to_btree(
181         xfs_trans_t             *tp,            /* transaction pointer */
182         xfs_inode_t             *ip,            /* incore inode pointer */
183         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
184         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
185         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
186         int                     wasdel,         /* converting a delayed alloc */
187         int                     *logflagsp,     /* inode logging flags */
188         int                     whichfork);     /* data or attr fork */
189
190 /*
191  * Convert a local file to an extents file.
192  * This code is sort of bogus, since the file data needs to get
193  * logged so it won't be lost.  The bmap-level manipulations are ok, though.
194  */
195 STATIC int                              /* error */
196 xfs_bmap_local_to_extents(
197         xfs_trans_t     *tp,            /* transaction pointer */
198         xfs_inode_t     *ip,            /* incore inode pointer */
199         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
200         xfs_extlen_t    total,          /* total blocks needed by transaction */
201         int             *logflagsp,     /* inode logging flags */
202         int             whichfork);     /* data or attr fork */
203
204 /*
205  * Search the extents list for the inode, for the extent containing bno.
206  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
207  * *eofp will be set, and *prevp will contain the last entry (null if none).
208  * Else, *lastxp will be set to the index of the found
209  * entry; *gotp will contain the entry.
210  */
211 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
212 xfs_bmap_search_extents(
213         xfs_inode_t     *ip,            /* incore inode pointer */
214         xfs_fileoff_t   bno,            /* block number searched for */
215         int             whichfork,      /* data or attr fork */
216         int             *eofp,          /* out: end of file found */
217         xfs_extnum_t    *lastxp,        /* out: last extent index */
218         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
219         xfs_bmbt_irec_t *prevp);        /* out: previous extent entry found */
220
221 /*
222  * Check the last inode extent to determine whether this allocation will result
223  * in blocks being allocated at the end of the file. When we allocate new data
224  * blocks at the end of the file which do not start at the previous data block,
225  * we will try to align the new blocks at stripe unit boundaries.
226  */
227 STATIC int                              /* error */
228 xfs_bmap_isaeof(
229         xfs_inode_t     *ip,            /* incore inode pointer */
230         xfs_fileoff_t   off,            /* file offset in fsblocks */
231         int             whichfork,      /* data or attribute fork */
232         char            *aeof);         /* return value */
233
234 /*
235  * Compute the worst-case number of indirect blocks that will be used
236  * for ip's delayed extent of length "len".
237  */
238 STATIC xfs_filblks_t
239 xfs_bmap_worst_indlen(
240         xfs_inode_t             *ip,    /* incore inode pointer */
241         xfs_filblks_t           len);   /* delayed extent length */
242
243 #ifdef DEBUG
244 /*
245  * Perform various validation checks on the values being returned
246  * from xfs_bmapi().
247  */
248 STATIC void
249 xfs_bmap_validate_ret(
250         xfs_fileoff_t           bno,
251         xfs_filblks_t           len,
252         int                     flags,
253         xfs_bmbt_irec_t         *mval,
254         int                     nmap,
255         int                     ret_nmap);
256 #else
257 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
258 #endif /* DEBUG */
259
260 STATIC int
261 xfs_bmap_count_tree(
262         xfs_mount_t     *mp,
263         xfs_trans_t     *tp,
264         xfs_ifork_t     *ifp,
265         xfs_fsblock_t   blockno,
266         int             levelin,
267         int             *count);
268
269 STATIC void
270 xfs_bmap_count_leaves(
271         xfs_ifork_t             *ifp,
272         xfs_extnum_t            idx,
273         int                     numrecs,
274         int                     *count);
275
276 STATIC void
277 xfs_bmap_disk_count_leaves(
278         struct xfs_mount        *mp,
279         struct xfs_btree_block  *block,
280         int                     numrecs,
281         int                     *count);
282
283 /*
284  * Bmap internal routines.
285  */
286
287 STATIC int                              /* error */
288 xfs_bmbt_lookup_eq(
289         struct xfs_btree_cur    *cur,
290         xfs_fileoff_t           off,
291         xfs_fsblock_t           bno,
292         xfs_filblks_t           len,
293         int                     *stat)  /* success/failure */
294 {
295         cur->bc_rec.b.br_startoff = off;
296         cur->bc_rec.b.br_startblock = bno;
297         cur->bc_rec.b.br_blockcount = len;
298         return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
299 }
300
301 STATIC int                              /* error */
302 xfs_bmbt_lookup_ge(
303         struct xfs_btree_cur    *cur,
304         xfs_fileoff_t           off,
305         xfs_fsblock_t           bno,
306         xfs_filblks_t           len,
307         int                     *stat)  /* success/failure */
308 {
309         cur->bc_rec.b.br_startoff = off;
310         cur->bc_rec.b.br_startblock = bno;
311         cur->bc_rec.b.br_blockcount = len;
312         return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
313 }
314
315 /*
316 * Update the record referred to by cur to the value given
317  * by [off, bno, len, state].
318  * This either works (return 0) or gets an EFSCORRUPTED error.
319  */
320 STATIC int
321 xfs_bmbt_update(
322         struct xfs_btree_cur    *cur,
323         xfs_fileoff_t           off,
324         xfs_fsblock_t           bno,
325         xfs_filblks_t           len,
326         xfs_exntst_t            state)
327 {
328         union xfs_btree_rec     rec;
329
330         xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
331         return xfs_btree_update(cur, &rec);
332 }
333
334 /*
335  * Called from xfs_bmap_add_attrfork to handle btree format files.
336  */
337 STATIC int                                      /* error */
338 xfs_bmap_add_attrfork_btree(
339         xfs_trans_t             *tp,            /* transaction pointer */
340         xfs_inode_t             *ip,            /* incore inode pointer */
341         xfs_fsblock_t           *firstblock,    /* first block allocated */
342         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
343         int                     *flags)         /* inode logging flags */
344 {
345         xfs_btree_cur_t         *cur;           /* btree cursor */
346         int                     error;          /* error return value */
347         xfs_mount_t             *mp;            /* file system mount struct */
348         int                     stat;           /* newroot status */
349
350         mp = ip->i_mount;
351         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
352                 *flags |= XFS_ILOG_DBROOT;
353         else {
354                 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
355                 cur->bc_private.b.flist = flist;
356                 cur->bc_private.b.firstblock = *firstblock;
357                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
358                         goto error0;
359                 /* must be at least one entry */
360                 XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
361                 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
362                         goto error0;
363                 if (stat == 0) {
364                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
365                         return XFS_ERROR(ENOSPC);
366                 }
367                 *firstblock = cur->bc_private.b.firstblock;
368                 cur->bc_private.b.allocated = 0;
369                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
370         }
371         return 0;
372 error0:
373         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
374         return error;
375 }
376
377 /*
378  * Called from xfs_bmap_add_attrfork to handle extents format files.
379  */
380 STATIC int                                      /* error */
381 xfs_bmap_add_attrfork_extents(
382         xfs_trans_t             *tp,            /* transaction pointer */
383         xfs_inode_t             *ip,            /* incore inode pointer */
384         xfs_fsblock_t           *firstblock,    /* first block allocated */
385         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
386         int                     *flags)         /* inode logging flags */
387 {
388         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
389         int                     error;          /* error return value */
390
391         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
392                 return 0;
393         cur = NULL;
394         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
395                 flags, XFS_DATA_FORK);
396         if (cur) {
397                 cur->bc_private.b.allocated = 0;
398                 xfs_btree_del_cursor(cur,
399                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
400         }
401         return error;
402 }
403
404 /*
405  * Called from xfs_bmap_add_attrfork to handle local format files.
406  */
407 STATIC int                                      /* error */
408 xfs_bmap_add_attrfork_local(
409         xfs_trans_t             *tp,            /* transaction pointer */
410         xfs_inode_t             *ip,            /* incore inode pointer */
411         xfs_fsblock_t           *firstblock,    /* first block allocated */
412         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
413         int                     *flags)         /* inode logging flags */
414 {
415         xfs_da_args_t           dargs;          /* args for dir/attr code */
416         int                     error;          /* error return value */
417         xfs_mount_t             *mp;            /* mount structure pointer */
418
419         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
420                 return 0;
421         if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
422                 mp = ip->i_mount;
423                 memset(&dargs, 0, sizeof(dargs));
424                 dargs.dp = ip;
425                 dargs.firstblock = firstblock;
426                 dargs.flist = flist;
427                 dargs.total = mp->m_dirblkfsbs;
428                 dargs.whichfork = XFS_DATA_FORK;
429                 dargs.trans = tp;
430                 error = xfs_dir2_sf_to_block(&dargs);
431         } else
432                 error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
433                         XFS_DATA_FORK);
434         return error;
435 }
436
437 /*
438  * Called by xfs_bmapi to update file extent records and the btree
439  * after allocating space (or doing a delayed allocation).
440  */
441 STATIC int                              /* error */
442 xfs_bmap_add_extent(
443         struct xfs_trans        *tp,    /* transaction pointer */
444         xfs_inode_t             *ip,    /* incore inode pointer */
445         xfs_extnum_t            *idx,   /* extent number to update/insert */
446         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
447         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
448         xfs_fsblock_t           *first, /* pointer to firstblock variable */
449         xfs_bmap_free_t         *flist, /* list of extents to be freed */
450         int                     *logflagsp, /* inode logging flags */
451         int                     whichfork) /* data or attr fork */
452 {
453         xfs_btree_cur_t         *cur;   /* btree cursor or null */
454         xfs_filblks_t           da_new; /* new count del alloc blocks used */
455         xfs_filblks_t           da_old; /* old count del alloc blocks used */
456         int                     error;  /* error return value */
457         xfs_ifork_t             *ifp;   /* inode fork ptr */
458         int                     logflags; /* returned value */
459         xfs_extnum_t            nextents; /* number of extents in file now */
460
461         XFS_STATS_INC(xs_add_exlist);
462
463         cur = *curp;
464         ifp = XFS_IFORK_PTR(ip, whichfork);
465         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
466         da_old = da_new = 0;
467         error = 0;
468
469         ASSERT(*idx >= 0);
470         ASSERT(*idx <= nextents);
471
472         /*
473          * This is the first extent added to a new/empty file.
474          * Special case this one, so other routines get to assume there are
475          * already extents in the list.
476          */
477         if (nextents == 0) {
478                 xfs_iext_insert(ip, *idx, 1, new,
479                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
480
481                 ASSERT(cur == NULL);
482
483                 if (!isnullstartblock(new->br_startblock)) {
484                         XFS_IFORK_NEXT_SET(ip, whichfork, 1);
485                         logflags = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
486                 } else
487                         logflags = 0;
488         }
489         /*
490          * Any kind of new delayed allocation goes here.
491          */
492         else if (isnullstartblock(new->br_startblock)) {
493                 if (cur)
494                         ASSERT((cur->bc_private.b.flags &
495                                 XFS_BTCUR_BPRV_WASDEL) == 0);
496                 error = xfs_bmap_add_extent_hole_delay(ip, idx, new,
497                                                        &logflags);
498         }
499         /*
500          * Real allocation off the end of the file.
501          */
502         else if (*idx == nextents) {
503                 if (cur)
504                         ASSERT((cur->bc_private.b.flags &
505                                 XFS_BTCUR_BPRV_WASDEL) == 0);
506                 error = xfs_bmap_add_extent_hole_real(ip, idx, cur, new,
507                                 &logflags, whichfork);
508         } else {
509                 xfs_bmbt_irec_t prev;   /* old extent at offset idx */
510
511                 /*
512                  * Get the record referred to by idx.
513                  */
514                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &prev);
515                 /*
516                  * If it's a real allocation record, and the new allocation ends
517                  * after the start of the referred to record, then we're filling
518                  * in a delayed or unwritten allocation with a real one, or
519                  * converting real back to unwritten.
520                  */
521                 if (!isnullstartblock(new->br_startblock) &&
522                     new->br_startoff + new->br_blockcount > prev.br_startoff) {
523                         if (prev.br_state != XFS_EXT_UNWRITTEN &&
524                             isnullstartblock(prev.br_startblock)) {
525                                 da_old = startblockval(prev.br_startblock);
526                                 if (cur)
527                                         ASSERT(cur->bc_private.b.flags &
528                                                 XFS_BTCUR_BPRV_WASDEL);
529                                 error = xfs_bmap_add_extent_delay_real(tp, ip,
530                                                 idx, &cur, new, &da_new,
531                                                 first, flist, &logflags);
532                         } else {
533                                 ASSERT(new->br_state == XFS_EXT_NORM ||
534                                        new->br_state == XFS_EXT_UNWRITTEN);
535
536                                 error = xfs_bmap_add_extent_unwritten_real(ip,
537                                                 idx, &cur, new, &logflags);
538                                 if (error)
539                                         goto done;
540                         }
541                 }
542                 /*
543                  * Otherwise we're filling in a hole with an allocation.
544                  */
545                 else {
546                         if (cur)
547                                 ASSERT((cur->bc_private.b.flags &
548                                         XFS_BTCUR_BPRV_WASDEL) == 0);
549                         error = xfs_bmap_add_extent_hole_real(ip, idx, cur,
550                                         new, &logflags, whichfork);
551                 }
552         }
553
554         if (error)
555                 goto done;
556         ASSERT(*curp == cur || *curp == NULL);
557
558         /*
559          * Convert to a btree if necessary.
560          */
561         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
562             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
563                 int     tmp_logflags;   /* partial log flag return val */
564
565                 ASSERT(cur == NULL);
566                 error = xfs_bmap_extents_to_btree(tp, ip, first,
567                         flist, &cur, da_old > 0, &tmp_logflags, whichfork);
568                 logflags |= tmp_logflags;
569                 if (error)
570                         goto done;
571         }
572         /*
573          * Adjust for changes in reserved delayed indirect blocks.
574          * Nothing to do for disk quotas here.
575          */
576         if (da_old || da_new) {
577                 xfs_filblks_t   nblks;
578
579                 nblks = da_new;
580                 if (cur)
581                         nblks += cur->bc_private.b.allocated;
582                 ASSERT(nblks <= da_old);
583                 if (nblks < da_old)
584                         xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
585                                 (int64_t)(da_old - nblks), 0);
586         }
587         /*
588          * Clear out the allocated field, done with it now in any case.
589          */
590         if (cur) {
591                 cur->bc_private.b.allocated = 0;
592                 *curp = cur;
593         }
594 done:
595 #ifdef DEBUG
596         if (!error)
597                 xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
598 #endif
599         *logflagsp = logflags;
600         return error;
601 }
602
603 /*
604  * Called by xfs_bmap_add_extent to handle cases converting a delayed
605  * allocation to a real allocation.
606  */
607 STATIC int                              /* error */
608 xfs_bmap_add_extent_delay_real(
609         struct xfs_trans        *tp,    /* transaction pointer */
610         xfs_inode_t             *ip,    /* incore inode pointer */
611         xfs_extnum_t            *idx,   /* extent number to update/insert */
612         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
613         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
614         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
615         xfs_fsblock_t           *first, /* pointer to firstblock variable */
616         xfs_bmap_free_t         *flist, /* list of extents to be freed */
617         int                     *logflagsp) /* inode logging flags */
618 {
619         xfs_btree_cur_t         *cur;   /* btree cursor */
620         int                     diff;   /* temp value */
621         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
622         int                     error;  /* error return value */
623         int                     i;      /* temp state */
624         xfs_ifork_t             *ifp;   /* inode fork pointer */
625         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
626         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
627                                         /* left is 0, right is 1, prev is 2 */
628         int                     rval=0; /* return value (logging flags) */
629         int                     state = 0;/* state bits, accessed thru macros */
630         xfs_filblks_t           temp=0; /* value for dnew calculations */
631         xfs_filblks_t           temp2=0;/* value for dnew calculations */
632         int                     tmp_rval;       /* partial logging flags */
633
634 #define LEFT            r[0]
635 #define RIGHT           r[1]
636 #define PREV            r[2]
637
638         /*
639          * Set up a bunch of variables to make the tests simpler.
640          */
641         cur = *curp;
642         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
643         ep = xfs_iext_get_ext(ifp, *idx);
644         xfs_bmbt_get_all(ep, &PREV);
645         new_endoff = new->br_startoff + new->br_blockcount;
646         ASSERT(PREV.br_startoff <= new->br_startoff);
647         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
648
649         /*
650          * Set flags determining what part of the previous delayed allocation
651          * extent is being replaced by a real allocation.
652          */
653         if (PREV.br_startoff == new->br_startoff)
654                 state |= BMAP_LEFT_FILLING;
655         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
656                 state |= BMAP_RIGHT_FILLING;
657
658         /*
659          * Check and set flags if this segment has a left neighbor.
660          * Don't set contiguous if the combined extent would be too large.
661          */
662         if (*idx > 0) {
663                 state |= BMAP_LEFT_VALID;
664                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
665
666                 if (isnullstartblock(LEFT.br_startblock))
667                         state |= BMAP_LEFT_DELAY;
668         }
669
670         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
671             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
672             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
673             LEFT.br_state == new->br_state &&
674             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
675                 state |= BMAP_LEFT_CONTIG;
676
677         /*
678          * Check and set flags if this segment has a right neighbor.
679          * Don't set contiguous if the combined extent would be too large.
680          * Also check for all-three-contiguous being too large.
681          */
682         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
683                 state |= BMAP_RIGHT_VALID;
684                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
685
686                 if (isnullstartblock(RIGHT.br_startblock))
687                         state |= BMAP_RIGHT_DELAY;
688         }
689
690         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
691             new_endoff == RIGHT.br_startoff &&
692             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
693             new->br_state == RIGHT.br_state &&
694             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
695             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
696                        BMAP_RIGHT_FILLING)) !=
697                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
698                        BMAP_RIGHT_FILLING) ||
699              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
700                         <= MAXEXTLEN))
701                 state |= BMAP_RIGHT_CONTIG;
702
703         error = 0;
704         /*
705          * Switch out based on the FILLING and CONTIG state bits.
706          */
707         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
708                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
709         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
710              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
711                 /*
712                  * Filling in all of a previously delayed allocation extent.
713                  * The left and right neighbors are both contiguous with new.
714                  */
715                 --*idx;
716                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
717                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
718                         LEFT.br_blockcount + PREV.br_blockcount +
719                         RIGHT.br_blockcount);
720                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
721
722                 xfs_iext_remove(ip, *idx + 1, 2, state);
723                 ip->i_d.di_nextents--;
724                 if (cur == NULL)
725                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
726                 else {
727                         rval = XFS_ILOG_CORE;
728                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
729                                         RIGHT.br_startblock,
730                                         RIGHT.br_blockcount, &i)))
731                                 goto done;
732                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
733                         if ((error = xfs_btree_delete(cur, &i)))
734                                 goto done;
735                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
736                         if ((error = xfs_btree_decrement(cur, 0, &i)))
737                                 goto done;
738                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
739                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
740                                         LEFT.br_startblock,
741                                         LEFT.br_blockcount +
742                                         PREV.br_blockcount +
743                                         RIGHT.br_blockcount, LEFT.br_state)))
744                                 goto done;
745                 }
746                 *dnew = 0;
747                 break;
748
749         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
750                 /*
751                  * Filling in all of a previously delayed allocation extent.
752                  * The left neighbor is contiguous, the right is not.
753                  */
754                 --*idx;
755
756                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
757                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
758                         LEFT.br_blockcount + PREV.br_blockcount);
759                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
760
761                 xfs_iext_remove(ip, *idx + 1, 1, state);
762                 if (cur == NULL)
763                         rval = XFS_ILOG_DEXT;
764                 else {
765                         rval = 0;
766                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
767                                         LEFT.br_startblock, LEFT.br_blockcount,
768                                         &i)))
769                                 goto done;
770                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
771                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
772                                         LEFT.br_startblock,
773                                         LEFT.br_blockcount +
774                                         PREV.br_blockcount, LEFT.br_state)))
775                                 goto done;
776                 }
777                 *dnew = 0;
778                 break;
779
780         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
781                 /*
782                  * Filling in all of a previously delayed allocation extent.
783                  * The right neighbor is contiguous, the left is not.
784                  */
785                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
786                 xfs_bmbt_set_startblock(ep, new->br_startblock);
787                 xfs_bmbt_set_blockcount(ep,
788                         PREV.br_blockcount + RIGHT.br_blockcount);
789                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
790
791                 xfs_iext_remove(ip, *idx + 1, 1, state);
792                 if (cur == NULL)
793                         rval = XFS_ILOG_DEXT;
794                 else {
795                         rval = 0;
796                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
797                                         RIGHT.br_startblock,
798                                         RIGHT.br_blockcount, &i)))
799                                 goto done;
800                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
801                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
802                                         new->br_startblock,
803                                         PREV.br_blockcount +
804                                         RIGHT.br_blockcount, PREV.br_state)))
805                                 goto done;
806                 }
807
808                 *dnew = 0;
809                 break;
810
811         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
812                 /*
813                  * Filling in all of a previously delayed allocation extent.
814                  * Neither the left nor right neighbors are contiguous with
815                  * the new one.
816                  */
817                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
818                 xfs_bmbt_set_startblock(ep, new->br_startblock);
819                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
820
821                 ip->i_d.di_nextents++;
822                 if (cur == NULL)
823                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
824                 else {
825                         rval = XFS_ILOG_CORE;
826                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
827                                         new->br_startblock, new->br_blockcount,
828                                         &i)))
829                                 goto done;
830                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
831                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
832                         if ((error = xfs_btree_insert(cur, &i)))
833                                 goto done;
834                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
835                 }
836
837                 *dnew = 0;
838                 break;
839
840         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
841                 /*
842                  * Filling in the first part of a previous delayed allocation.
843                  * The left neighbor is contiguous.
844                  */
845                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
846                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
847                         LEFT.br_blockcount + new->br_blockcount);
848                 xfs_bmbt_set_startoff(ep,
849                         PREV.br_startoff + new->br_blockcount);
850                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
851
852                 temp = PREV.br_blockcount - new->br_blockcount;
853                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
854                 xfs_bmbt_set_blockcount(ep, temp);
855                 if (cur == NULL)
856                         rval = XFS_ILOG_DEXT;
857                 else {
858                         rval = 0;
859                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
860                                         LEFT.br_startblock, LEFT.br_blockcount,
861                                         &i)))
862                                 goto done;
863                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
864                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
865                                         LEFT.br_startblock,
866                                         LEFT.br_blockcount +
867                                         new->br_blockcount,
868                                         LEFT.br_state)))
869                                 goto done;
870                 }
871                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
872                         startblockval(PREV.br_startblock));
873                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
874                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
875
876                 --*idx;
877                 *dnew = temp;
878                 break;
879
880         case BMAP_LEFT_FILLING:
881                 /*
882                  * Filling in the first part of a previous delayed allocation.
883                  * The left neighbor is not contiguous.
884                  */
885                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
886                 xfs_bmbt_set_startoff(ep, new_endoff);
887                 temp = PREV.br_blockcount - new->br_blockcount;
888                 xfs_bmbt_set_blockcount(ep, temp);
889                 xfs_iext_insert(ip, *idx, 1, new, state);
890                 ip->i_d.di_nextents++;
891                 if (cur == NULL)
892                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
893                 else {
894                         rval = XFS_ILOG_CORE;
895                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
896                                         new->br_startblock, new->br_blockcount,
897                                         &i)))
898                                 goto done;
899                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
900                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
901                         if ((error = xfs_btree_insert(cur, &i)))
902                                 goto done;
903                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
904                 }
905                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
906                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
907                         error = xfs_bmap_extents_to_btree(tp, ip,
908                                         first, flist, &cur, 1, &tmp_rval,
909                                         XFS_DATA_FORK);
910                         rval |= tmp_rval;
911                         if (error)
912                                 goto done;
913                 }
914                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
915                         startblockval(PREV.br_startblock) -
916                         (cur ? cur->bc_private.b.allocated : 0));
917                 ep = xfs_iext_get_ext(ifp, *idx + 1);
918                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
919                 trace_xfs_bmap_post_update(ip, *idx + 1, state, _THIS_IP_);
920
921                 *dnew = temp;
922                 break;
923
924         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
925                 /*
926                  * Filling in the last part of a previous delayed allocation.
927                  * The right neighbor is contiguous with the new allocation.
928                  */
929                 temp = PREV.br_blockcount - new->br_blockcount;
930                 trace_xfs_bmap_pre_update(ip, *idx + 1, state, _THIS_IP_);
931                 xfs_bmbt_set_blockcount(ep, temp);
932                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx + 1),
933                         new->br_startoff, new->br_startblock,
934                         new->br_blockcount + RIGHT.br_blockcount,
935                         RIGHT.br_state);
936                 trace_xfs_bmap_post_update(ip, *idx + 1, state, _THIS_IP_);
937                 if (cur == NULL)
938                         rval = XFS_ILOG_DEXT;
939                 else {
940                         rval = 0;
941                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
942                                         RIGHT.br_startblock,
943                                         RIGHT.br_blockcount, &i)))
944                                 goto done;
945                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
946                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
947                                         new->br_startblock,
948                                         new->br_blockcount +
949                                         RIGHT.br_blockcount,
950                                         RIGHT.br_state)))
951                                 goto done;
952                 }
953
954                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
955                         startblockval(PREV.br_startblock));
956                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
957                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
958                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
959
960                 ++*idx;
961                 *dnew = temp;
962                 break;
963
964         case BMAP_RIGHT_FILLING:
965                 /*
966                  * Filling in the last part of a previous delayed allocation.
967                  * The right neighbor is not contiguous.
968                  */
969                 temp = PREV.br_blockcount - new->br_blockcount;
970                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
971                 xfs_bmbt_set_blockcount(ep, temp);
972                 xfs_iext_insert(ip, *idx + 1, 1, new, state);
973                 ip->i_d.di_nextents++;
974                 if (cur == NULL)
975                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
976                 else {
977                         rval = XFS_ILOG_CORE;
978                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
979                                         new->br_startblock, new->br_blockcount,
980                                         &i)))
981                                 goto done;
982                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
983                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
984                         if ((error = xfs_btree_insert(cur, &i)))
985                                 goto done;
986                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
987                 }
988                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
989                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
990                         error = xfs_bmap_extents_to_btree(tp, ip,
991                                 first, flist, &cur, 1, &tmp_rval,
992                                 XFS_DATA_FORK);
993                         rval |= tmp_rval;
994                         if (error)
995                                 goto done;
996                 }
997                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
998                         startblockval(PREV.br_startblock) -
999                         (cur ? cur->bc_private.b.allocated : 0));
1000                 ep = xfs_iext_get_ext(ifp, *idx);
1001                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
1002                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1003
1004                 ++*idx;
1005                 *dnew = temp;
1006                 break;
1007
1008         case 0:
1009                 /*
1010                  * Filling in the middle part of a previous delayed allocation.
1011                  * Contiguity is impossible here.
1012                  * This case is avoided almost all the time.
1013                  *
1014                  * We start with a delayed allocation:
1015                  *
1016                  * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
1017                  *  PREV @ idx
1018                  *
1019                  * and we are allocating:
1020                  *                     +rrrrrrrrrrrrrrrrr+
1021                  *                            new
1022                  *
1023                  * and we set it up for insertion as:
1024                  * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
1025                  *                            new
1026                  *  PREV @ idx          LEFT              RIGHT
1027                  *                      inserted at idx + 1
1028                  */
1029                 temp = new->br_startoff - PREV.br_startoff;
1030                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
1031                 trace_xfs_bmap_pre_update(ip, *idx, 0, _THIS_IP_);
1032                 xfs_bmbt_set_blockcount(ep, temp);      /* truncate PREV */
1033                 LEFT = *new;
1034                 RIGHT.br_state = PREV.br_state;
1035                 RIGHT.br_startblock = nullstartblock(
1036                                 (int)xfs_bmap_worst_indlen(ip, temp2));
1037                 RIGHT.br_startoff = new_endoff;
1038                 RIGHT.br_blockcount = temp2;
1039                 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
1040                 xfs_iext_insert(ip, *idx + 1, 2, &LEFT, state);
1041                 ip->i_d.di_nextents++;
1042                 if (cur == NULL)
1043                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1044                 else {
1045                         rval = XFS_ILOG_CORE;
1046                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1047                                         new->br_startblock, new->br_blockcount,
1048                                         &i)))
1049                                 goto done;
1050                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1051                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1052                         if ((error = xfs_btree_insert(cur, &i)))
1053                                 goto done;
1054                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1055                 }
1056                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1057                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1058                         error = xfs_bmap_extents_to_btree(tp, ip,
1059                                         first, flist, &cur, 1, &tmp_rval,
1060                                         XFS_DATA_FORK);
1061                         rval |= tmp_rval;
1062                         if (error)
1063                                 goto done;
1064                 }
1065                 temp = xfs_bmap_worst_indlen(ip, temp);
1066                 temp2 = xfs_bmap_worst_indlen(ip, temp2);
1067                 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
1068                         (cur ? cur->bc_private.b.allocated : 0));
1069                 if (diff > 0 &&
1070                     xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
1071                                              -((int64_t)diff), 0)) {
1072                         /*
1073                          * Ick gross gag me with a spoon.
1074                          */
1075                         ASSERT(0);      /* want to see if this ever happens! */
1076                         while (diff > 0) {
1077                                 if (temp) {
1078                                         temp--;
1079                                         diff--;
1080                                         if (!diff ||
1081                                             !xfs_icsb_modify_counters(ip->i_mount,
1082                                                     XFS_SBS_FDBLOCKS,
1083                                                     -((int64_t)diff), 0))
1084                                                 break;
1085                                 }
1086                                 if (temp2) {
1087                                         temp2--;
1088                                         diff--;
1089                                         if (!diff ||
1090                                             !xfs_icsb_modify_counters(ip->i_mount,
1091                                                     XFS_SBS_FDBLOCKS,
1092                                                     -((int64_t)diff), 0))
1093                                                 break;
1094                                 }
1095                         }
1096                 }
1097                 ep = xfs_iext_get_ext(ifp, *idx);
1098                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
1099                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1100                 trace_xfs_bmap_pre_update(ip, *idx + 2, state, _THIS_IP_);
1101                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx + 2),
1102                         nullstartblock((int)temp2));
1103                 trace_xfs_bmap_post_update(ip, *idx + 2, state, _THIS_IP_);
1104
1105                 ++*idx;
1106                 *dnew = temp + temp2;
1107                 break;
1108
1109         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1110         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1111         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1112         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1113         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1114         case BMAP_LEFT_CONTIG:
1115         case BMAP_RIGHT_CONTIG:
1116                 /*
1117                  * These cases are all impossible.
1118                  */
1119                 ASSERT(0);
1120         }
1121         *curp = cur;
1122 done:
1123         *logflagsp = rval;
1124         return error;
1125 #undef  LEFT
1126 #undef  RIGHT
1127 #undef  PREV
1128 }
1129
1130 /*
1131  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
1132  * allocation to a real allocation or vice versa.
1133  */
1134 STATIC int                              /* error */
1135 xfs_bmap_add_extent_unwritten_real(
1136         xfs_inode_t             *ip,    /* incore inode pointer */
1137         xfs_extnum_t            *idx,   /* extent number to update/insert */
1138         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
1139         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1140         int                     *logflagsp) /* inode logging flags */
1141 {
1142         xfs_btree_cur_t         *cur;   /* btree cursor */
1143         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
1144         int                     error;  /* error return value */
1145         int                     i;      /* temp state */
1146         xfs_ifork_t             *ifp;   /* inode fork pointer */
1147         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1148         xfs_exntst_t            newext; /* new extent state */
1149         xfs_exntst_t            oldext; /* old extent state */
1150         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1151                                         /* left is 0, right is 1, prev is 2 */
1152         int                     rval=0; /* return value (logging flags) */
1153         int                     state = 0;/* state bits, accessed thru macros */
1154
1155 #define LEFT            r[0]
1156 #define RIGHT           r[1]
1157 #define PREV            r[2]
1158         /*
1159          * Set up a bunch of variables to make the tests simpler.
1160          */
1161         error = 0;
1162         cur = *curp;
1163         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1164         ep = xfs_iext_get_ext(ifp, *idx);
1165         xfs_bmbt_get_all(ep, &PREV);
1166         newext = new->br_state;
1167         oldext = (newext == XFS_EXT_UNWRITTEN) ?
1168                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
1169         ASSERT(PREV.br_state == oldext);
1170         new_endoff = new->br_startoff + new->br_blockcount;
1171         ASSERT(PREV.br_startoff <= new->br_startoff);
1172         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1173
1174         /*
1175          * Set flags determining what part of the previous oldext allocation
1176          * extent is being replaced by a newext allocation.
1177          */
1178         if (PREV.br_startoff == new->br_startoff)
1179                 state |= BMAP_LEFT_FILLING;
1180         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1181                 state |= BMAP_RIGHT_FILLING;
1182
1183         /*
1184          * Check and set flags if this segment has a left neighbor.
1185          * Don't set contiguous if the combined extent would be too large.
1186          */
1187         if (*idx > 0) {
1188                 state |= BMAP_LEFT_VALID;
1189                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
1190
1191                 if (isnullstartblock(LEFT.br_startblock))
1192                         state |= BMAP_LEFT_DELAY;
1193         }
1194
1195         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1196             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1197             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1198             LEFT.br_state == newext &&
1199             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1200                 state |= BMAP_LEFT_CONTIG;
1201
1202         /*
1203          * Check and set flags if this segment has a right neighbor.
1204          * Don't set contiguous if the combined extent would be too large.
1205          * Also check for all-three-contiguous being too large.
1206          */
1207         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1208                 state |= BMAP_RIGHT_VALID;
1209                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
1210                 if (isnullstartblock(RIGHT.br_startblock))
1211                         state |= BMAP_RIGHT_DELAY;
1212         }
1213
1214         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1215             new_endoff == RIGHT.br_startoff &&
1216             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1217             newext == RIGHT.br_state &&
1218             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1219             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1220                        BMAP_RIGHT_FILLING)) !=
1221                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1222                        BMAP_RIGHT_FILLING) ||
1223              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1224                         <= MAXEXTLEN))
1225                 state |= BMAP_RIGHT_CONTIG;
1226
1227         /*
1228          * Switch out based on the FILLING and CONTIG state bits.
1229          */
1230         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1231                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1232         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1233              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1234                 /*
1235                  * Setting all of a previous oldext extent to newext.
1236                  * The left and right neighbors are both contiguous with new.
1237                  */
1238                 --*idx;
1239
1240                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1241                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1242                         LEFT.br_blockcount + PREV.br_blockcount +
1243                         RIGHT.br_blockcount);
1244                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1245
1246                 xfs_iext_remove(ip, *idx + 1, 2, state);
1247                 ip->i_d.di_nextents -= 2;
1248                 if (cur == NULL)
1249                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1250                 else {
1251                         rval = XFS_ILOG_CORE;
1252                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1253                                         RIGHT.br_startblock,
1254                                         RIGHT.br_blockcount, &i)))
1255                                 goto done;
1256                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1257                         if ((error = xfs_btree_delete(cur, &i)))
1258                                 goto done;
1259                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1260                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1261                                 goto done;
1262                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1263                         if ((error = xfs_btree_delete(cur, &i)))
1264                                 goto done;
1265                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1266                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1267                                 goto done;
1268                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1269                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1270                                 LEFT.br_startblock,
1271                                 LEFT.br_blockcount + PREV.br_blockcount +
1272                                 RIGHT.br_blockcount, LEFT.br_state)))
1273                                 goto done;
1274                 }
1275                 break;
1276
1277         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1278                 /*
1279                  * Setting all of a previous oldext extent to newext.
1280                  * The left neighbor is contiguous, the right is not.
1281                  */
1282                 --*idx;
1283
1284                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1285                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1286                         LEFT.br_blockcount + PREV.br_blockcount);
1287                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1288
1289                 xfs_iext_remove(ip, *idx + 1, 1, state);
1290                 ip->i_d.di_nextents--;
1291                 if (cur == NULL)
1292                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1293                 else {
1294                         rval = XFS_ILOG_CORE;
1295                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1296                                         PREV.br_startblock, PREV.br_blockcount,
1297                                         &i)))
1298                                 goto done;
1299                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1300                         if ((error = xfs_btree_delete(cur, &i)))
1301                                 goto done;
1302                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1303                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1304                                 goto done;
1305                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1306                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1307                                 LEFT.br_startblock,
1308                                 LEFT.br_blockcount + PREV.br_blockcount,
1309                                 LEFT.br_state)))
1310                                 goto done;
1311                 }
1312                 break;
1313
1314         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1315                 /*
1316                  * Setting all of a previous oldext extent to newext.
1317                  * The right neighbor is contiguous, the left is not.
1318                  */
1319                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1320                 xfs_bmbt_set_blockcount(ep,
1321                         PREV.br_blockcount + RIGHT.br_blockcount);
1322                 xfs_bmbt_set_state(ep, newext);
1323                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1324                 xfs_iext_remove(ip, *idx + 1, 1, state);
1325                 ip->i_d.di_nextents--;
1326                 if (cur == NULL)
1327                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1328                 else {
1329                         rval = XFS_ILOG_CORE;
1330                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1331                                         RIGHT.br_startblock,
1332                                         RIGHT.br_blockcount, &i)))
1333                                 goto done;
1334                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1335                         if ((error = xfs_btree_delete(cur, &i)))
1336                                 goto done;
1337                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1338                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1339                                 goto done;
1340                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1341                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1342                                 new->br_startblock,
1343                                 new->br_blockcount + RIGHT.br_blockcount,
1344                                 newext)))
1345                                 goto done;
1346                 }
1347                 break;
1348
1349         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1350                 /*
1351                  * Setting all of a previous oldext extent to newext.
1352                  * Neither the left nor right neighbors are contiguous with
1353                  * the new one.
1354                  */
1355                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1356                 xfs_bmbt_set_state(ep, newext);
1357                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1358
1359                 if (cur == NULL)
1360                         rval = XFS_ILOG_DEXT;
1361                 else {
1362                         rval = 0;
1363                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1364                                         new->br_startblock, new->br_blockcount,
1365                                         &i)))
1366                                 goto done;
1367                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1368                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1369                                 new->br_startblock, new->br_blockcount,
1370                                 newext)))
1371                                 goto done;
1372                 }
1373                 break;
1374
1375         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1376                 /*
1377                  * Setting the first part of a previous oldext extent to newext.
1378                  * The left neighbor is contiguous.
1379                  */
1380                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
1381                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
1382                         LEFT.br_blockcount + new->br_blockcount);
1383                 xfs_bmbt_set_startoff(ep,
1384                         PREV.br_startoff + new->br_blockcount);
1385                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
1386
1387                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1388                 xfs_bmbt_set_startblock(ep,
1389                         new->br_startblock + new->br_blockcount);
1390                 xfs_bmbt_set_blockcount(ep,
1391                         PREV.br_blockcount - new->br_blockcount);
1392                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1393
1394                 --*idx;
1395
1396                 if (cur == NULL)
1397                         rval = XFS_ILOG_DEXT;
1398                 else {
1399                         rval = 0;
1400                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1401                                         PREV.br_startblock, PREV.br_blockcount,
1402                                         &i)))
1403                                 goto done;
1404                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1405                         if ((error = xfs_bmbt_update(cur,
1406                                 PREV.br_startoff + new->br_blockcount,
1407                                 PREV.br_startblock + new->br_blockcount,
1408                                 PREV.br_blockcount - new->br_blockcount,
1409                                 oldext)))
1410                                 goto done;
1411                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1412                                 goto done;
1413                         if (xfs_bmbt_update(cur, LEFT.br_startoff,
1414                                 LEFT.br_startblock,
1415                                 LEFT.br_blockcount + new->br_blockcount,
1416                                 LEFT.br_state))
1417                                 goto done;
1418                 }
1419                 break;
1420
1421         case BMAP_LEFT_FILLING:
1422                 /*
1423                  * Setting the first part of a previous oldext extent to newext.
1424                  * The left neighbor is not contiguous.
1425                  */
1426                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1427                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1428                 xfs_bmbt_set_startoff(ep, new_endoff);
1429                 xfs_bmbt_set_blockcount(ep,
1430                         PREV.br_blockcount - new->br_blockcount);
1431                 xfs_bmbt_set_startblock(ep,
1432                         new->br_startblock + new->br_blockcount);
1433                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1434
1435                 xfs_iext_insert(ip, *idx, 1, new, state);
1436                 ip->i_d.di_nextents++;
1437                 if (cur == NULL)
1438                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1439                 else {
1440                         rval = XFS_ILOG_CORE;
1441                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1442                                         PREV.br_startblock, PREV.br_blockcount,
1443                                         &i)))
1444                                 goto done;
1445                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1446                         if ((error = xfs_bmbt_update(cur,
1447                                 PREV.br_startoff + new->br_blockcount,
1448                                 PREV.br_startblock + new->br_blockcount,
1449                                 PREV.br_blockcount - new->br_blockcount,
1450                                 oldext)))
1451                                 goto done;
1452                         cur->bc_rec.b = *new;
1453                         if ((error = xfs_btree_insert(cur, &i)))
1454                                 goto done;
1455                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1456                 }
1457                 break;
1458
1459         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1460                 /*
1461                  * Setting the last part of a previous oldext extent to newext.
1462                  * The right neighbor is contiguous with the new allocation.
1463                  */
1464                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1465                 xfs_bmbt_set_blockcount(ep,
1466                         PREV.br_blockcount - new->br_blockcount);
1467                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1468
1469                 ++*idx;
1470
1471                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1472                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1473                         new->br_startoff, new->br_startblock,
1474                         new->br_blockcount + RIGHT.br_blockcount, newext);
1475                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1476
1477                 if (cur == NULL)
1478                         rval = XFS_ILOG_DEXT;
1479                 else {
1480                         rval = 0;
1481                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1482                                         PREV.br_startblock,
1483                                         PREV.br_blockcount, &i)))
1484                                 goto done;
1485                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1486                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1487                                 PREV.br_startblock,
1488                                 PREV.br_blockcount - new->br_blockcount,
1489                                 oldext)))
1490                                 goto done;
1491                         if ((error = xfs_btree_increment(cur, 0, &i)))
1492                                 goto done;
1493                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1494                                 new->br_startblock,
1495                                 new->br_blockcount + RIGHT.br_blockcount,
1496                                 newext)))
1497                                 goto done;
1498                 }
1499                 break;
1500
1501         case BMAP_RIGHT_FILLING:
1502                 /*
1503                  * Setting the last part of a previous oldext extent to newext.
1504                  * The right neighbor is not contiguous.
1505                  */
1506                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1507                 xfs_bmbt_set_blockcount(ep,
1508                         PREV.br_blockcount - new->br_blockcount);
1509                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1510
1511                 ++*idx;
1512                 xfs_iext_insert(ip, *idx, 1, new, state);
1513
1514                 ip->i_d.di_nextents++;
1515                 if (cur == NULL)
1516                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1517                 else {
1518                         rval = XFS_ILOG_CORE;
1519                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1520                                         PREV.br_startblock, PREV.br_blockcount,
1521                                         &i)))
1522                                 goto done;
1523                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1524                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1525                                 PREV.br_startblock,
1526                                 PREV.br_blockcount - new->br_blockcount,
1527                                 oldext)))
1528                                 goto done;
1529                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1530                                         new->br_startblock, new->br_blockcount,
1531                                         &i)))
1532                                 goto done;
1533                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1534                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1535                         if ((error = xfs_btree_insert(cur, &i)))
1536                                 goto done;
1537                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1538                 }
1539                 break;
1540
1541         case 0:
1542                 /*
1543                  * Setting the middle part of a previous oldext extent to
1544                  * newext.  Contiguity is impossible here.
1545                  * One extent becomes three extents.
1546                  */
1547                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1548                 xfs_bmbt_set_blockcount(ep,
1549                         new->br_startoff - PREV.br_startoff);
1550                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1551
1552                 r[0] = *new;
1553                 r[1].br_startoff = new_endoff;
1554                 r[1].br_blockcount =
1555                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
1556                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1557                 r[1].br_state = oldext;
1558
1559                 ++*idx;
1560                 xfs_iext_insert(ip, *idx, 2, &r[0], state);
1561
1562                 ip->i_d.di_nextents += 2;
1563                 if (cur == NULL)
1564                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1565                 else {
1566                         rval = XFS_ILOG_CORE;
1567                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1568                                         PREV.br_startblock, PREV.br_blockcount,
1569                                         &i)))
1570                                 goto done;
1571                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1572                         /* new right extent - oldext */
1573                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1574                                 r[1].br_startblock, r[1].br_blockcount,
1575                                 r[1].br_state)))
1576                                 goto done;
1577                         /* new left extent - oldext */
1578                         cur->bc_rec.b = PREV;
1579                         cur->bc_rec.b.br_blockcount =
1580                                 new->br_startoff - PREV.br_startoff;
1581                         if ((error = xfs_btree_insert(cur, &i)))
1582                                 goto done;
1583                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1584                         /*
1585                          * Reset the cursor to the position of the new extent
1586                          * we are about to insert as we can't trust it after
1587                          * the previous insert.
1588                          */
1589                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1590                                         new->br_startblock, new->br_blockcount,
1591                                         &i)))
1592                                 goto done;
1593                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1594                         /* new middle extent - newext */
1595                         cur->bc_rec.b.br_state = new->br_state;
1596                         if ((error = xfs_btree_insert(cur, &i)))
1597                                 goto done;
1598                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1599                 }
1600                 break;
1601
1602         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1603         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1604         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1605         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1606         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1607         case BMAP_LEFT_CONTIG:
1608         case BMAP_RIGHT_CONTIG:
1609                 /*
1610                  * These cases are all impossible.
1611                  */
1612                 ASSERT(0);
1613         }
1614         *curp = cur;
1615 done:
1616         *logflagsp = rval;
1617         return error;
1618 #undef  LEFT
1619 #undef  RIGHT
1620 #undef  PREV
1621 }
1622
1623 /*
1624  * Called by xfs_bmap_add_extent to handle cases converting a hole
1625  * to a delayed allocation.
1626  */
1627 /*ARGSUSED*/
1628 STATIC int                              /* error */
1629 xfs_bmap_add_extent_hole_delay(
1630         xfs_inode_t             *ip,    /* incore inode pointer */
1631         xfs_extnum_t            *idx,   /* extent number to update/insert */
1632         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1633         int                     *logflagsp) /* inode logging flags */
1634 {
1635         xfs_ifork_t             *ifp;   /* inode fork pointer */
1636         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1637         xfs_filblks_t           newlen=0;       /* new indirect size */
1638         xfs_filblks_t           oldlen=0;       /* old indirect size */
1639         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1640         int                     state;  /* state bits, accessed thru macros */
1641         xfs_filblks_t           temp=0; /* temp for indirect calculations */
1642
1643         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1644         state = 0;
1645         ASSERT(isnullstartblock(new->br_startblock));
1646
1647         /*
1648          * Check and set flags if this segment has a left neighbor
1649          */
1650         if (*idx > 0) {
1651                 state |= BMAP_LEFT_VALID;
1652                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
1653
1654                 if (isnullstartblock(left.br_startblock))
1655                         state |= BMAP_LEFT_DELAY;
1656         }
1657
1658         /*
1659          * Check and set flags if the current (right) segment exists.
1660          * If it doesn't exist, we're converting the hole at end-of-file.
1661          */
1662         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1663                 state |= BMAP_RIGHT_VALID;
1664                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
1665
1666                 if (isnullstartblock(right.br_startblock))
1667                         state |= BMAP_RIGHT_DELAY;
1668         }
1669
1670         /*
1671          * Set contiguity flags on the left and right neighbors.
1672          * Don't let extents get too large, even if the pieces are contiguous.
1673          */
1674         if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
1675             left.br_startoff + left.br_blockcount == new->br_startoff &&
1676             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1677                 state |= BMAP_LEFT_CONTIG;
1678
1679         if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
1680             new->br_startoff + new->br_blockcount == right.br_startoff &&
1681             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1682             (!(state & BMAP_LEFT_CONTIG) ||
1683              (left.br_blockcount + new->br_blockcount +
1684               right.br_blockcount <= MAXEXTLEN)))
1685                 state |= BMAP_RIGHT_CONTIG;
1686
1687         /*
1688          * Switch out based on the contiguity flags.
1689          */
1690         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1691         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1692                 /*
1693                  * New allocation is contiguous with delayed allocations
1694                  * on the left and on the right.
1695                  * Merge all three into a single extent record.
1696                  */
1697                 --*idx;
1698                 temp = left.br_blockcount + new->br_blockcount +
1699                         right.br_blockcount;
1700
1701                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1702                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1703                 oldlen = startblockval(left.br_startblock) +
1704                         startblockval(new->br_startblock) +
1705                         startblockval(right.br_startblock);
1706                 newlen = xfs_bmap_worst_indlen(ip, temp);
1707                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1708                         nullstartblock((int)newlen));
1709                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1710
1711                 xfs_iext_remove(ip, *idx + 1, 1, state);
1712                 break;
1713
1714         case BMAP_LEFT_CONTIG:
1715                 /*
1716                  * New allocation is contiguous with a delayed allocation
1717                  * on the left.
1718                  * Merge the new allocation with the left neighbor.
1719                  */
1720                 --*idx;
1721                 temp = left.br_blockcount + new->br_blockcount;
1722
1723                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1724                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1725                 oldlen = startblockval(left.br_startblock) +
1726                         startblockval(new->br_startblock);
1727                 newlen = xfs_bmap_worst_indlen(ip, temp);
1728                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1729                         nullstartblock((int)newlen));
1730                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1731                 break;
1732
1733         case BMAP_RIGHT_CONTIG:
1734                 /*
1735                  * New allocation is contiguous with a delayed allocation
1736                  * on the right.
1737                  * Merge the new allocation with the right neighbor.
1738                  */
1739                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1740                 temp = new->br_blockcount + right.br_blockcount;
1741                 oldlen = startblockval(new->br_startblock) +
1742                         startblockval(right.br_startblock);
1743                 newlen = xfs_bmap_worst_indlen(ip, temp);
1744                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1745                         new->br_startoff,
1746                         nullstartblock((int)newlen), temp, right.br_state);
1747                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1748                 break;
1749
1750         case 0:
1751                 /*
1752                  * New allocation is not contiguous with another
1753                  * delayed allocation.
1754                  * Insert a new entry.
1755                  */
1756                 oldlen = newlen = 0;
1757                 xfs_iext_insert(ip, *idx, 1, new, state);
1758                 break;
1759         }
1760         if (oldlen != newlen) {
1761                 ASSERT(oldlen > newlen);
1762                 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
1763                         (int64_t)(oldlen - newlen), 0);
1764                 /*
1765                  * Nothing to do for disk quota accounting here.
1766                  */
1767         }
1768         *logflagsp = 0;
1769         return 0;
1770 }
1771
1772 /*
1773  * Called by xfs_bmap_add_extent to handle cases converting a hole
1774  * to a real allocation.
1775  */
1776 STATIC int                              /* error */
1777 xfs_bmap_add_extent_hole_real(
1778         xfs_inode_t             *ip,    /* incore inode pointer */
1779         xfs_extnum_t            *idx,   /* extent number to update/insert */
1780         xfs_btree_cur_t         *cur,   /* if null, not a btree */
1781         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1782         int                     *logflagsp, /* inode logging flags */
1783         int                     whichfork) /* data or attr fork */
1784 {
1785         int                     error;  /* error return value */
1786         int                     i;      /* temp state */
1787         xfs_ifork_t             *ifp;   /* inode fork pointer */
1788         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1789         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1790         int                     rval=0; /* return value (logging flags) */
1791         int                     state;  /* state bits, accessed thru macros */
1792
1793         ifp = XFS_IFORK_PTR(ip, whichfork);
1794         ASSERT(*idx <= ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
1795         state = 0;
1796
1797         if (whichfork == XFS_ATTR_FORK)
1798                 state |= BMAP_ATTRFORK;
1799
1800         /*
1801          * Check and set flags if this segment has a left neighbor.
1802          */
1803         if (*idx > 0) {
1804                 state |= BMAP_LEFT_VALID;
1805                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
1806                 if (isnullstartblock(left.br_startblock))
1807                         state |= BMAP_LEFT_DELAY;
1808         }
1809
1810         /*
1811          * Check and set flags if this segment has a current value.
1812          * Not true if we're inserting into the "hole" at eof.
1813          */
1814         if (*idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1815                 state |= BMAP_RIGHT_VALID;
1816                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
1817                 if (isnullstartblock(right.br_startblock))
1818                         state |= BMAP_RIGHT_DELAY;
1819         }
1820
1821         /*
1822          * We're inserting a real allocation between "left" and "right".
1823          * Set the contiguity flags.  Don't let extents get too large.
1824          */
1825         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1826             left.br_startoff + left.br_blockcount == new->br_startoff &&
1827             left.br_startblock + left.br_blockcount == new->br_startblock &&
1828             left.br_state == new->br_state &&
1829             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1830                 state |= BMAP_LEFT_CONTIG;
1831
1832         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1833             new->br_startoff + new->br_blockcount == right.br_startoff &&
1834             new->br_startblock + new->br_blockcount == right.br_startblock &&
1835             new->br_state == right.br_state &&
1836             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1837             (!(state & BMAP_LEFT_CONTIG) ||
1838              left.br_blockcount + new->br_blockcount +
1839              right.br_blockcount <= MAXEXTLEN))
1840                 state |= BMAP_RIGHT_CONTIG;
1841
1842         error = 0;
1843         /*
1844          * Select which case we're in here, and implement it.
1845          */
1846         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1847         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1848                 /*
1849                  * New allocation is contiguous with real allocations on the
1850                  * left and on the right.
1851                  * Merge all three into a single extent record.
1852                  */
1853                 --*idx;
1854                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1855                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1856                         left.br_blockcount + new->br_blockcount +
1857                         right.br_blockcount);
1858                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1859
1860                 xfs_iext_remove(ip, *idx + 1, 1, state);
1861
1862                 XFS_IFORK_NEXT_SET(ip, whichfork,
1863                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
1864                 if (cur == NULL) {
1865                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1866                 } else {
1867                         rval = XFS_ILOG_CORE;
1868                         if ((error = xfs_bmbt_lookup_eq(cur,
1869                                         right.br_startoff,
1870                                         right.br_startblock,
1871                                         right.br_blockcount, &i)))
1872                                 goto done;
1873                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1874                         if ((error = xfs_btree_delete(cur, &i)))
1875                                 goto done;
1876                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1877                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1878                                 goto done;
1879                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1880                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
1881                                         left.br_startblock,
1882                                         left.br_blockcount +
1883                                                 new->br_blockcount +
1884                                                 right.br_blockcount,
1885                                         left.br_state)))
1886                                 goto done;
1887                 }
1888                 break;
1889
1890         case BMAP_LEFT_CONTIG:
1891                 /*
1892                  * New allocation is contiguous with a real allocation
1893                  * on the left.
1894                  * Merge the new allocation with the left neighbor.
1895                  */
1896                 --*idx;
1897                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1898                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1899                         left.br_blockcount + new->br_blockcount);
1900                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1901
1902                 if (cur == NULL) {
1903                         rval = xfs_ilog_fext(whichfork);
1904                 } else {
1905                         rval = 0;
1906                         if ((error = xfs_bmbt_lookup_eq(cur,
1907                                         left.br_startoff,
1908                                         left.br_startblock,
1909                                         left.br_blockcount, &i)))
1910                                 goto done;
1911                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1912                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
1913                                         left.br_startblock,
1914                                         left.br_blockcount +
1915                                                 new->br_blockcount,
1916                                         left.br_state)))
1917                                 goto done;
1918                 }
1919                 break;
1920
1921         case BMAP_RIGHT_CONTIG:
1922                 /*
1923                  * New allocation is contiguous with a real allocation
1924                  * on the right.
1925                  * Merge the new allocation with the right neighbor.
1926                  */
1927                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1928                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1929                         new->br_startoff, new->br_startblock,
1930                         new->br_blockcount + right.br_blockcount,
1931                         right.br_state);
1932                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1933
1934                 if (cur == NULL) {
1935                         rval = xfs_ilog_fext(whichfork);
1936                 } else {
1937                         rval = 0;
1938                         if ((error = xfs_bmbt_lookup_eq(cur,
1939                                         right.br_startoff,
1940                                         right.br_startblock,
1941                                         right.br_blockcount, &i)))
1942                                 goto done;
1943                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1944                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1945                                         new->br_startblock,
1946                                         new->br_blockcount +
1947                                                 right.br_blockcount,
1948                                         right.br_state)))
1949                                 goto done;
1950                 }
1951                 break;
1952
1953         case 0:
1954                 /*
1955                  * New allocation is not contiguous with another
1956                  * real allocation.
1957                  * Insert a new entry.
1958                  */
1959                 xfs_iext_insert(ip, *idx, 1, new, state);
1960                 XFS_IFORK_NEXT_SET(ip, whichfork,
1961                         XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
1962                 if (cur == NULL) {
1963                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1964                 } else {
1965                         rval = XFS_ILOG_CORE;
1966                         if ((error = xfs_bmbt_lookup_eq(cur,
1967                                         new->br_startoff,
1968                                         new->br_startblock,
1969                                         new->br_blockcount, &i)))
1970                                 goto done;
1971                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1972                         cur->bc_rec.b.br_state = new->br_state;
1973                         if ((error = xfs_btree_insert(cur, &i)))
1974                                 goto done;
1975                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1976                 }
1977                 break;
1978         }
1979 done:
1980         *logflagsp = rval;
1981         return error;
1982 }
1983
1984 /*
1985  * Adjust the size of the new extent based on di_extsize and rt extsize.
1986  */
1987 STATIC int
1988 xfs_bmap_extsize_align(
1989         xfs_mount_t     *mp,
1990         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
1991         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
1992         xfs_extlen_t    extsz,          /* align to this extent size */
1993         int             rt,             /* is this a realtime inode? */
1994         int             eof,            /* is extent at end-of-file? */
1995         int             delay,          /* creating delalloc extent? */
1996         int             convert,        /* overwriting unwritten extent? */
1997         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
1998         xfs_extlen_t    *lenp)          /* in/out: aligned length */
1999 {
2000         xfs_fileoff_t   orig_off;       /* original offset */
2001         xfs_extlen_t    orig_alen;      /* original length */
2002         xfs_fileoff_t   orig_end;       /* original off+len */
2003         xfs_fileoff_t   nexto;          /* next file offset */
2004         xfs_fileoff_t   prevo;          /* previous file offset */
2005         xfs_fileoff_t   align_off;      /* temp for offset */
2006         xfs_extlen_t    align_alen;     /* temp for length */
2007         xfs_extlen_t    temp;           /* temp for calculations */
2008
2009         if (convert)
2010                 return 0;
2011
2012         orig_off = align_off = *offp;
2013         orig_alen = align_alen = *lenp;
2014         orig_end = orig_off + orig_alen;
2015
2016         /*
2017          * If this request overlaps an existing extent, then don't
2018          * attempt to perform any additional alignment.
2019          */
2020         if (!delay && !eof &&
2021             (orig_off >= gotp->br_startoff) &&
2022             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2023                 return 0;
2024         }
2025
2026         /*
2027          * If the file offset is unaligned vs. the extent size
2028          * we need to align it.  This will be possible unless
2029          * the file was previously written with a kernel that didn't
2030          * perform this alignment, or if a truncate shot us in the
2031          * foot.
2032          */
2033         temp = do_mod(orig_off, extsz);
2034         if (temp) {
2035                 align_alen += temp;
2036                 align_off -= temp;
2037         }
2038         /*
2039          * Same adjustment for the end of the requested area.
2040          */
2041         if ((temp = (align_alen % extsz))) {
2042                 align_alen += extsz - temp;
2043         }
2044         /*
2045          * If the previous block overlaps with this proposed allocation
2046          * then move the start forward without adjusting the length.
2047          */
2048         if (prevp->br_startoff != NULLFILEOFF) {
2049                 if (prevp->br_startblock == HOLESTARTBLOCK)
2050                         prevo = prevp->br_startoff;
2051                 else
2052                         prevo = prevp->br_startoff + prevp->br_blockcount;
2053         } else
2054                 prevo = 0;
2055         if (align_off != orig_off && align_off < prevo)
2056                 align_off = prevo;
2057         /*
2058          * If the next block overlaps with this proposed allocation
2059          * then move the start back without adjusting the length,
2060          * but not before offset 0.
2061          * This may of course make the start overlap previous block,
2062          * and if we hit the offset 0 limit then the next block
2063          * can still overlap too.
2064          */
2065         if (!eof && gotp->br_startoff != NULLFILEOFF) {
2066                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
2067                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
2068                         nexto = gotp->br_startoff + gotp->br_blockcount;
2069                 else
2070                         nexto = gotp->br_startoff;
2071         } else
2072                 nexto = NULLFILEOFF;
2073         if (!eof &&
2074             align_off + align_alen != orig_end &&
2075             align_off + align_alen > nexto)
2076                 align_off = nexto > align_alen ? nexto - align_alen : 0;
2077         /*
2078          * If we're now overlapping the next or previous extent that
2079          * means we can't fit an extsz piece in this hole.  Just move
2080          * the start forward to the first valid spot and set
2081          * the length so we hit the end.
2082          */
2083         if (align_off != orig_off && align_off < prevo)
2084                 align_off = prevo;
2085         if (align_off + align_alen != orig_end &&
2086             align_off + align_alen > nexto &&
2087             nexto != NULLFILEOFF) {
2088                 ASSERT(nexto > prevo);
2089                 align_alen = nexto - align_off;
2090         }
2091
2092         /*
2093          * If realtime, and the result isn't a multiple of the realtime
2094          * extent size we need to remove blocks until it is.
2095          */
2096         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
2097                 /*
2098                  * We're not covering the original request, or
2099                  * we won't be able to once we fix the length.
2100                  */
2101                 if (orig_off < align_off ||
2102                     orig_end > align_off + align_alen ||
2103                     align_alen - temp < orig_alen)
2104                         return XFS_ERROR(EINVAL);
2105                 /*
2106                  * Try to fix it by moving the start up.
2107                  */
2108                 if (align_off + temp <= orig_off) {
2109                         align_alen -= temp;
2110                         align_off += temp;
2111                 }
2112                 /*
2113                  * Try to fix it by moving the end in.
2114                  */
2115                 else if (align_off + align_alen - temp >= orig_end)
2116                         align_alen -= temp;
2117                 /*
2118                  * Set the start to the minimum then trim the length.
2119                  */
2120                 else {
2121                         align_alen -= orig_off - align_off;
2122                         align_off = orig_off;
2123                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
2124                 }
2125                 /*
2126                  * Result doesn't cover the request, fail it.
2127                  */
2128                 if (orig_off < align_off || orig_end > align_off + align_alen)
2129                         return XFS_ERROR(EINVAL);
2130         } else {
2131                 ASSERT(orig_off >= align_off);
2132                 ASSERT(orig_end <= align_off + align_alen);
2133         }
2134
2135 #ifdef DEBUG
2136         if (!eof && gotp->br_startoff != NULLFILEOFF)
2137                 ASSERT(align_off + align_alen <= gotp->br_startoff);
2138         if (prevp->br_startoff != NULLFILEOFF)
2139                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
2140 #endif
2141
2142         *lenp = align_alen;
2143         *offp = align_off;
2144         return 0;
2145 }
2146
2147 #define XFS_ALLOC_GAP_UNITS     4
2148
2149 STATIC void
2150 xfs_bmap_adjacent(
2151         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2152 {
2153         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
2154         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2155         xfs_mount_t     *mp;            /* mount point structure */
2156         int             nullfb;         /* true if ap->firstblock isn't set */
2157         int             rt;             /* true if inode is realtime */
2158
2159 #define ISVALID(x,y)    \
2160         (rt ? \
2161                 (x) < mp->m_sb.sb_rblocks : \
2162                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2163                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2164                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2165
2166         mp = ap->ip->i_mount;
2167         nullfb = ap->firstblock == NULLFSBLOCK;
2168         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2169         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2170         /*
2171          * If allocating at eof, and there's a previous real block,
2172          * try to use its last block as our starting point.
2173          */
2174         if (ap->eof && ap->prevp->br_startoff != NULLFILEOFF &&
2175             !isnullstartblock(ap->prevp->br_startblock) &&
2176             ISVALID(ap->prevp->br_startblock + ap->prevp->br_blockcount,
2177                     ap->prevp->br_startblock)) {
2178                 ap->rval = ap->prevp->br_startblock + ap->prevp->br_blockcount;
2179                 /*
2180                  * Adjust for the gap between prevp and us.
2181                  */
2182                 adjust = ap->off -
2183                         (ap->prevp->br_startoff + ap->prevp->br_blockcount);
2184                 if (adjust &&
2185                     ISVALID(ap->rval + adjust, ap->prevp->br_startblock))
2186                         ap->rval += adjust;
2187         }
2188         /*
2189          * If not at eof, then compare the two neighbor blocks.
2190          * Figure out whether either one gives us a good starting point,
2191          * and pick the better one.
2192          */
2193         else if (!ap->eof) {
2194                 xfs_fsblock_t   gotbno;         /* right side block number */
2195                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
2196                 xfs_fsblock_t   prevbno;        /* left side block number */
2197                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
2198
2199                 /*
2200                  * If there's a previous (left) block, select a requested
2201                  * start block based on it.
2202                  */
2203                 if (ap->prevp->br_startoff != NULLFILEOFF &&
2204                     !isnullstartblock(ap->prevp->br_startblock) &&
2205                     (prevbno = ap->prevp->br_startblock +
2206                                ap->prevp->br_blockcount) &&
2207                     ISVALID(prevbno, ap->prevp->br_startblock)) {
2208                         /*
2209                          * Calculate gap to end of previous block.
2210                          */
2211                         adjust = prevdiff = ap->off -
2212                                 (ap->prevp->br_startoff +
2213                                  ap->prevp->br_blockcount);
2214                         /*
2215                          * Figure the startblock based on the previous block's
2216                          * end and the gap size.
2217                          * Heuristic!
2218                          * If the gap is large relative to the piece we're
2219                          * allocating, or using it gives us an invalid block
2220                          * number, then just use the end of the previous block.
2221                          */
2222                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2223                             ISVALID(prevbno + prevdiff,
2224                                     ap->prevp->br_startblock))
2225                                 prevbno += adjust;
2226                         else
2227                                 prevdiff += adjust;
2228                         /*
2229                          * If the firstblock forbids it, can't use it,
2230                          * must use default.
2231                          */
2232                         if (!rt && !nullfb &&
2233                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2234                                 prevbno = NULLFSBLOCK;
2235                 }
2236                 /*
2237                  * No previous block or can't follow it, just default.
2238                  */
2239                 else
2240                         prevbno = NULLFSBLOCK;
2241                 /*
2242                  * If there's a following (right) block, select a requested
2243                  * start block based on it.
2244                  */
2245                 if (!isnullstartblock(ap->gotp->br_startblock)) {
2246                         /*
2247                          * Calculate gap to start of next block.
2248                          */
2249                         adjust = gotdiff = ap->gotp->br_startoff - ap->off;
2250                         /*
2251                          * Figure the startblock based on the next block's
2252                          * start and the gap size.
2253                          */
2254                         gotbno = ap->gotp->br_startblock;
2255                         /*
2256                          * Heuristic!
2257                          * If the gap is large relative to the piece we're
2258                          * allocating, or using it gives us an invalid block
2259                          * number, then just use the start of the next block
2260                          * offset by our length.
2261                          */
2262                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2263                             ISVALID(gotbno - gotdiff, gotbno))
2264                                 gotbno -= adjust;
2265                         else if (ISVALID(gotbno - ap->alen, gotbno)) {
2266                                 gotbno -= ap->alen;
2267                                 gotdiff += adjust - ap->alen;
2268                         } else
2269                                 gotdiff += adjust;
2270                         /*
2271                          * If the firstblock forbids it, can't use it,
2272                          * must use default.
2273                          */
2274                         if (!rt && !nullfb &&
2275                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2276                                 gotbno = NULLFSBLOCK;
2277                 }
2278                 /*
2279                  * No next block, just default.
2280                  */
2281                 else
2282                         gotbno = NULLFSBLOCK;
2283                 /*
2284                  * If both valid, pick the better one, else the only good
2285                  * one, else ap->rval is already set (to 0 or the inode block).
2286                  */
2287                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2288                         ap->rval = prevdiff <= gotdiff ? prevbno : gotbno;
2289                 else if (prevbno != NULLFSBLOCK)
2290                         ap->rval = prevbno;
2291                 else if (gotbno != NULLFSBLOCK)
2292                         ap->rval = gotbno;
2293         }
2294 #undef ISVALID
2295 }
2296
2297 STATIC int
2298 xfs_bmap_rtalloc(
2299         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2300 {
2301         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2302         int             error;          /* error return value */
2303         xfs_mount_t     *mp;            /* mount point structure */
2304         xfs_extlen_t    prod = 0;       /* product factor for allocators */
2305         xfs_extlen_t    ralen = 0;      /* realtime allocation length */
2306         xfs_extlen_t    align;          /* minimum allocation alignment */
2307         xfs_rtblock_t   rtb;
2308
2309         mp = ap->ip->i_mount;
2310         align = xfs_get_extsz_hint(ap->ip);
2311         prod = align / mp->m_sb.sb_rextsize;
2312         error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2313                                         align, 1, ap->eof, 0,
2314                                         ap->conv, &ap->off, &ap->alen);
2315         if (error)
2316                 return error;
2317         ASSERT(ap->alen);
2318         ASSERT(ap->alen % mp->m_sb.sb_rextsize == 0);
2319
2320         /*
2321          * If the offset & length are not perfectly aligned
2322          * then kill prod, it will just get us in trouble.
2323          */
2324         if (do_mod(ap->off, align) || ap->alen % align)
2325                 prod = 1;
2326         /*
2327          * Set ralen to be the actual requested length in rtextents.
2328          */
2329         ralen = ap->alen / mp->m_sb.sb_rextsize;
2330         /*
2331          * If the old value was close enough to MAXEXTLEN that
2332          * we rounded up to it, cut it back so it's valid again.
2333          * Note that if it's a really large request (bigger than
2334          * MAXEXTLEN), we don't hear about that number, and can't
2335          * adjust the starting point to match it.
2336          */
2337         if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2338                 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2339
2340         /*
2341          * Lock out other modifications to the RT bitmap inode.
2342          */
2343         xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
2344         xfs_trans_ijoin_ref(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
2345
2346         /*
2347          * If it's an allocation to an empty file at offset 0,
2348          * pick an extent that will space things out in the rt area.
2349          */
2350         if (ap->eof && ap->off == 0) {
2351                 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
2352
2353                 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2354                 if (error)
2355                         return error;
2356                 ap->rval = rtx * mp->m_sb.sb_rextsize;
2357         } else {
2358                 ap->rval = 0;
2359         }
2360
2361         xfs_bmap_adjacent(ap);
2362
2363         /*
2364          * Realtime allocation, done through xfs_rtallocate_extent.
2365          */
2366         atype = ap->rval == 0 ?  XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2367         do_div(ap->rval, mp->m_sb.sb_rextsize);
2368         rtb = ap->rval;
2369         ap->alen = ralen;
2370         if ((error = xfs_rtallocate_extent(ap->tp, ap->rval, 1, ap->alen,
2371                                 &ralen, atype, ap->wasdel, prod, &rtb)))
2372                 return error;
2373         if (rtb == NULLFSBLOCK && prod > 1 &&
2374             (error = xfs_rtallocate_extent(ap->tp, ap->rval, 1,
2375                                            ap->alen, &ralen, atype,
2376                                            ap->wasdel, 1, &rtb)))
2377                 return error;
2378         ap->rval = rtb;
2379         if (ap->rval != NULLFSBLOCK) {
2380                 ap->rval *= mp->m_sb.sb_rextsize;
2381                 ralen *= mp->m_sb.sb_rextsize;
2382                 ap->alen = ralen;
2383                 ap->ip->i_d.di_nblocks += ralen;
2384                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2385                 if (ap->wasdel)
2386                         ap->ip->i_delayed_blks -= ralen;
2387                 /*
2388                  * Adjust the disk quota also. This was reserved
2389                  * earlier.
2390                  */
2391                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2392                         ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2393                                         XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2394         } else {
2395                 ap->alen = 0;
2396         }
2397         return 0;
2398 }
2399
2400 STATIC int
2401 xfs_bmap_btalloc_nullfb(
2402         struct xfs_bmalloca     *ap,
2403         struct xfs_alloc_arg    *args,
2404         xfs_extlen_t            *blen)
2405 {
2406         struct xfs_mount        *mp = ap->ip->i_mount;
2407         struct xfs_perag        *pag;
2408         xfs_agnumber_t          ag, startag;
2409         int                     notinit = 0;
2410         int                     error;
2411
2412         if (ap->userdata && xfs_inode_is_filestream(ap->ip))
2413                 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2414         else
2415                 args->type = XFS_ALLOCTYPE_START_BNO;
2416         args->total = ap->total;
2417
2418         /*
2419          * Search for an allocation group with a single extent large enough
2420          * for the request.  If one isn't found, then adjust the minimum
2421          * allocation size to the largest space found.
2422          */
2423         startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
2424         if (startag == NULLAGNUMBER)
2425                 startag = ag = 0;
2426
2427         pag = xfs_perag_get(mp, ag);
2428         while (*blen < args->maxlen) {
2429                 if (!pag->pagf_init) {
2430                         error = xfs_alloc_pagf_init(mp, args->tp, ag,
2431                                                     XFS_ALLOC_FLAG_TRYLOCK);
2432                         if (error) {
2433                                 xfs_perag_put(pag);
2434                                 return error;
2435                         }
2436                 }
2437
2438                 /*
2439                  * See xfs_alloc_fix_freelist...
2440                  */
2441                 if (pag->pagf_init) {
2442                         xfs_extlen_t    longest;
2443                         longest = xfs_alloc_longest_free_extent(mp, pag);
2444                         if (*blen < longest)
2445                                 *blen = longest;
2446                 } else
2447                         notinit = 1;
2448
2449                 if (xfs_inode_is_filestream(ap->ip)) {
2450                         if (*blen >= args->maxlen)
2451                                 break;
2452
2453                         if (ap->userdata) {
2454                                 /*
2455                                  * If startag is an invalid AG, we've
2456                                  * come here once before and
2457                                  * xfs_filestream_new_ag picked the
2458                                  * best currently available.
2459                                  *
2460                                  * Don't continue looping, since we
2461                                  * could loop forever.
2462                                  */
2463                                 if (startag == NULLAGNUMBER)
2464                                         break;
2465
2466                                 error = xfs_filestream_new_ag(ap, &ag);
2467                                 xfs_perag_put(pag);
2468                                 if (error)
2469                                         return error;
2470
2471                                 /* loop again to set 'blen'*/
2472                                 startag = NULLAGNUMBER;
2473                                 pag = xfs_perag_get(mp, ag);
2474                                 continue;
2475                         }
2476                 }
2477                 if (++ag == mp->m_sb.sb_agcount)
2478                         ag = 0;
2479                 if (ag == startag)
2480                         break;
2481                 xfs_perag_put(pag);
2482                 pag = xfs_perag_get(mp, ag);
2483         }
2484         xfs_perag_put(pag);
2485
2486         /*
2487          * Since the above loop did a BUF_TRYLOCK, it is
2488          * possible that there is space for this request.
2489          */
2490         if (notinit || *blen < ap->minlen)
2491                 args->minlen = ap->minlen;
2492         /*
2493          * If the best seen length is less than the request
2494          * length, use the best as the minimum.
2495          */
2496         else if (*blen < args->maxlen)
2497                 args->minlen = *blen;
2498         /*
2499          * Otherwise we've seen an extent as big as maxlen,
2500          * use that as the minimum.
2501          */
2502         else
2503                 args->minlen = args->maxlen;
2504
2505         /*
2506          * set the failure fallback case to look in the selected
2507          * AG as the stream may have moved.
2508          */
2509         if (xfs_inode_is_filestream(ap->ip))
2510                 ap->rval = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
2511
2512         return 0;
2513 }
2514
2515 STATIC int
2516 xfs_bmap_btalloc(
2517         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2518 {
2519         xfs_mount_t     *mp;            /* mount point structure */
2520         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2521         xfs_extlen_t    align;          /* minimum allocation alignment */
2522         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2523         xfs_agnumber_t  ag;
2524         xfs_alloc_arg_t args;
2525         xfs_extlen_t    blen;
2526         xfs_extlen_t    nextminlen = 0;
2527         int             nullfb;         /* true if ap->firstblock isn't set */
2528         int             isaligned;
2529         int             tryagain;
2530         int             error;
2531
2532         mp = ap->ip->i_mount;
2533         align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
2534         if (unlikely(align)) {
2535                 error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2536                                                 align, 0, ap->eof, 0, ap->conv,
2537                                                 &ap->off, &ap->alen);
2538                 ASSERT(!error);
2539                 ASSERT(ap->alen);
2540         }
2541         nullfb = ap->firstblock == NULLFSBLOCK;
2542         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2543         if (nullfb) {
2544                 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
2545                         ag = xfs_filestream_lookup_ag(ap->ip);
2546                         ag = (ag != NULLAGNUMBER) ? ag : 0;
2547                         ap->rval = XFS_AGB_TO_FSB(mp, ag, 0);
2548                 } else {
2549                         ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2550                 }
2551         } else
2552                 ap->rval = ap->firstblock;
2553
2554         xfs_bmap_adjacent(ap);
2555
2556         /*
2557          * If allowed, use ap->rval; otherwise must use firstblock since
2558          * it's in the right allocation group.
2559          */
2560         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->rval) == fb_agno)
2561                 ;
2562         else
2563                 ap->rval = ap->firstblock;
2564         /*
2565          * Normal allocation, done through xfs_alloc_vextent.
2566          */
2567         tryagain = isaligned = 0;
2568         args.tp = ap->tp;
2569         args.mp = mp;
2570         args.fsbno = ap->rval;
2571
2572         /* Trim the allocation back to the maximum an AG can fit. */
2573         args.maxlen = MIN(ap->alen, XFS_ALLOC_AG_MAX_USABLE(mp));
2574         args.firstblock = ap->firstblock;
2575         blen = 0;
2576         if (nullfb) {
2577                 error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
2578                 if (error)
2579                         return error;
2580         } else if (ap->low) {
2581                 if (xfs_inode_is_filestream(ap->ip))
2582                         args.type = XFS_ALLOCTYPE_FIRST_AG;
2583                 else
2584                         args.type = XFS_ALLOCTYPE_START_BNO;
2585                 args.total = args.minlen = ap->minlen;
2586         } else {
2587                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2588                 args.total = ap->total;
2589                 args.minlen = ap->minlen;
2590         }
2591         /* apply extent size hints if obtained earlier */
2592         if (unlikely(align)) {
2593                 args.prod = align;
2594                 if ((args.mod = (xfs_extlen_t)do_mod(ap->off, args.prod)))
2595                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2596         } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
2597                 args.prod = 1;
2598                 args.mod = 0;
2599         } else {
2600                 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
2601                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->off, args.prod))))
2602                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2603         }
2604         /*
2605          * If we are not low on available data blocks, and the
2606          * underlying logical volume manager is a stripe, and
2607          * the file offset is zero then try to allocate data
2608          * blocks on stripe unit boundary.
2609          * NOTE: ap->aeof is only set if the allocation length
2610          * is >= the stripe unit and the allocation offset is
2611          * at the end of file.
2612          */
2613         if (!ap->low && ap->aeof) {
2614                 if (!ap->off) {
2615                         args.alignment = mp->m_dalign;
2616                         atype = args.type;
2617                         isaligned = 1;
2618                         /*
2619                          * Adjust for alignment
2620                          */
2621                         if (blen > args.alignment && blen <= args.maxlen)
2622                                 args.minlen = blen - args.alignment;
2623                         args.minalignslop = 0;
2624                 } else {
2625                         /*
2626                          * First try an exact bno allocation.
2627                          * If it fails then do a near or start bno
2628                          * allocation with alignment turned on.
2629                          */
2630                         atype = args.type;
2631                         tryagain = 1;
2632                         args.type = XFS_ALLOCTYPE_THIS_BNO;
2633                         args.alignment = 1;
2634                         /*
2635                          * Compute the minlen+alignment for the
2636                          * next case.  Set slop so that the value
2637                          * of minlen+alignment+slop doesn't go up
2638                          * between the calls.
2639                          */
2640                         if (blen > mp->m_dalign && blen <= args.maxlen)
2641                                 nextminlen = blen - mp->m_dalign;
2642                         else
2643                                 nextminlen = args.minlen;
2644                         if (nextminlen + mp->m_dalign > args.minlen + 1)
2645                                 args.minalignslop =
2646                                         nextminlen + mp->m_dalign -
2647                                         args.minlen - 1;
2648                         else
2649                                 args.minalignslop = 0;
2650                 }
2651         } else {
2652                 args.alignment = 1;
2653                 args.minalignslop = 0;
2654         }
2655         args.minleft = ap->minleft;
2656         args.wasdel = ap->wasdel;
2657         args.isfl = 0;
2658         args.userdata = ap->userdata;
2659         if ((error = xfs_alloc_vextent(&args)))
2660                 return error;
2661         if (tryagain && args.fsbno == NULLFSBLOCK) {
2662                 /*
2663                  * Exact allocation failed. Now try with alignment
2664                  * turned on.
2665                  */
2666                 args.type = atype;
2667                 args.fsbno = ap->rval;
2668                 args.alignment = mp->m_dalign;
2669                 args.minlen = nextminlen;
2670                 args.minalignslop = 0;
2671                 isaligned = 1;
2672                 if ((error = xfs_alloc_vextent(&args)))
2673                         return error;
2674         }
2675         if (isaligned && args.fsbno == NULLFSBLOCK) {
2676                 /*
2677                  * allocation failed, so turn off alignment and
2678                  * try again.
2679                  */
2680                 args.type = atype;
2681                 args.fsbno = ap->rval;
2682                 args.alignment = 0;
2683                 if ((error = xfs_alloc_vextent(&args)))
2684                         return error;
2685         }
2686         if (args.fsbno == NULLFSBLOCK && nullfb &&
2687             args.minlen > ap->minlen) {
2688                 args.minlen = ap->minlen;
2689                 args.type = XFS_ALLOCTYPE_START_BNO;
2690                 args.fsbno = ap->rval;
2691                 if ((error = xfs_alloc_vextent(&args)))
2692                         return error;
2693         }
2694         if (args.fsbno == NULLFSBLOCK && nullfb) {
2695                 args.fsbno = 0;
2696                 args.type = XFS_ALLOCTYPE_FIRST_AG;
2697                 args.total = ap->minlen;
2698                 args.minleft = 0;
2699                 if ((error = xfs_alloc_vextent(&args)))
2700                         return error;
2701                 ap->low = 1;
2702         }
2703         if (args.fsbno != NULLFSBLOCK) {
2704                 ap->firstblock = ap->rval = args.fsbno;
2705                 ASSERT(nullfb || fb_agno == args.agno ||
2706                        (ap->low && fb_agno < args.agno));
2707                 ap->alen = args.len;
2708                 ap->ip->i_d.di_nblocks += args.len;
2709                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2710                 if (ap->wasdel)
2711                         ap->ip->i_delayed_blks -= args.len;
2712                 /*
2713                  * Adjust the disk quota also. This was reserved
2714                  * earlier.
2715                  */
2716                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2717                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2718                                         XFS_TRANS_DQ_BCOUNT,
2719                         (long) args.len);
2720         } else {
2721                 ap->rval = NULLFSBLOCK;
2722                 ap->alen = 0;
2723         }
2724         return 0;
2725 }
2726
2727 /*
2728  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2729  * It figures out where to ask the underlying allocator to put the new extent.
2730  */
2731 STATIC int
2732 xfs_bmap_alloc(
2733         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2734 {
2735         if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
2736                 return xfs_bmap_rtalloc(ap);
2737         return xfs_bmap_btalloc(ap);
2738 }
2739
2740 /*
2741  * Transform a btree format file with only one leaf node, where the
2742  * extents list will fit in the inode, into an extents format file.
2743  * Since the file extents are already in-core, all we have to do is
2744  * give up the space for the btree root and pitch the leaf block.
2745  */
2746 STATIC int                              /* error */
2747 xfs_bmap_btree_to_extents(
2748         xfs_trans_t             *tp,    /* transaction pointer */
2749         xfs_inode_t             *ip,    /* incore inode pointer */
2750         xfs_btree_cur_t         *cur,   /* btree cursor */
2751         int                     *logflagsp, /* inode logging flags */
2752         int                     whichfork)  /* data or attr fork */
2753 {
2754         /* REFERENCED */
2755         struct xfs_btree_block  *cblock;/* child btree block */
2756         xfs_fsblock_t           cbno;   /* child block number */
2757         xfs_buf_t               *cbp;   /* child block's buffer */
2758         int                     error;  /* error return value */
2759         xfs_ifork_t             *ifp;   /* inode fork data */
2760         xfs_mount_t             *mp;    /* mount point structure */
2761         __be64                  *pp;    /* ptr to block address */
2762         struct xfs_btree_block  *rblock;/* root btree block */
2763
2764         mp = ip->i_mount;
2765         ifp = XFS_IFORK_PTR(ip, whichfork);
2766         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2767         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
2768         rblock = ifp->if_broot;
2769         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
2770         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
2771         ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
2772         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
2773         cbno = be64_to_cpu(*pp);
2774         *logflagsp = 0;
2775 #ifdef DEBUG
2776         if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
2777                 return error;
2778 #endif
2779         if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
2780                         XFS_BMAP_BTREE_REF)))
2781                 return error;
2782         cblock = XFS_BUF_TO_BLOCK(cbp);
2783         if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
2784                 return error;
2785         xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
2786         ip->i_d.di_nblocks--;
2787         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
2788         xfs_trans_binval(tp, cbp);
2789         if (cur->bc_bufs[0] == cbp)
2790                 cur->bc_bufs[0] = NULL;
2791         xfs_iroot_realloc(ip, -1, whichfork);
2792         ASSERT(ifp->if_broot == NULL);
2793         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
2794         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
2795         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2796         return 0;
2797 }
2798
2799 /*
2800  * Called by xfs_bmapi to update file extent records and the btree
2801  * after removing space (or undoing a delayed allocation).
2802  */
2803 STATIC int                              /* error */
2804 xfs_bmap_del_extent(
2805         xfs_inode_t             *ip,    /* incore inode pointer */
2806         xfs_trans_t             *tp,    /* current transaction pointer */
2807         xfs_extnum_t            *idx,   /* extent number to update/delete */
2808         xfs_bmap_free_t         *flist, /* list of extents to be freed */
2809         xfs_btree_cur_t         *cur,   /* if null, not a btree */
2810         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
2811         int                     *logflagsp, /* inode logging flags */
2812         int                     whichfork) /* data or attr fork */
2813 {
2814         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
2815         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
2816         xfs_fsblock_t           del_endblock=0; /* first block past del */
2817         xfs_fileoff_t           del_endoff;     /* first offset past del */
2818         int                     delay;  /* current block is delayed allocated */
2819         int                     do_fx;  /* free extent at end of routine */
2820         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
2821         int                     error;  /* error return value */
2822         int                     flags;  /* inode logging flags */
2823         xfs_bmbt_irec_t         got;    /* current extent entry */
2824         xfs_fileoff_t           got_endoff;     /* first offset past got */
2825         int                     i;      /* temp state */
2826         xfs_ifork_t             *ifp;   /* inode fork pointer */
2827         xfs_mount_t             *mp;    /* mount structure */
2828         xfs_filblks_t           nblks;  /* quota/sb block count */
2829         xfs_bmbt_irec_t         new;    /* new record to be inserted */
2830         /* REFERENCED */
2831         uint                    qfield; /* quota field to update */
2832         xfs_filblks_t           temp;   /* for indirect length calculations */
2833         xfs_filblks_t           temp2;  /* for indirect length calculations */
2834         int                     state = 0;
2835
2836         XFS_STATS_INC(xs_del_exlist);
2837
2838         if (whichfork == XFS_ATTR_FORK)
2839                 state |= BMAP_ATTRFORK;
2840
2841         mp = ip->i_mount;
2842         ifp = XFS_IFORK_PTR(ip, whichfork);
2843         ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
2844                 (uint)sizeof(xfs_bmbt_rec_t)));
2845         ASSERT(del->br_blockcount > 0);
2846         ep = xfs_iext_get_ext(ifp, *idx);
2847         xfs_bmbt_get_all(ep, &got);
2848         ASSERT(got.br_startoff <= del->br_startoff);
2849         del_endoff = del->br_startoff + del->br_blockcount;
2850         got_endoff = got.br_startoff + got.br_blockcount;
2851         ASSERT(got_endoff >= del_endoff);
2852         delay = isnullstartblock(got.br_startblock);
2853         ASSERT(isnullstartblock(del->br_startblock) == delay);
2854         flags = 0;
2855         qfield = 0;
2856         error = 0;
2857         /*
2858          * If deleting a real allocation, must free up the disk space.
2859          */
2860         if (!delay) {
2861                 flags = XFS_ILOG_CORE;
2862                 /*
2863                  * Realtime allocation.  Free it and record di_nblocks update.
2864                  */
2865                 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
2866                         xfs_fsblock_t   bno;
2867                         xfs_filblks_t   len;
2868
2869                         ASSERT(do_mod(del->br_blockcount,
2870                                       mp->m_sb.sb_rextsize) == 0);
2871                         ASSERT(do_mod(del->br_startblock,
2872                                       mp->m_sb.sb_rextsize) == 0);
2873                         bno = del->br_startblock;
2874                         len = del->br_blockcount;
2875                         do_div(bno, mp->m_sb.sb_rextsize);
2876                         do_div(len, mp->m_sb.sb_rextsize);
2877                         error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
2878                         if (error)
2879                                 goto done;
2880                         do_fx = 0;
2881                         nblks = len * mp->m_sb.sb_rextsize;
2882                         qfield = XFS_TRANS_DQ_RTBCOUNT;
2883                 }
2884                 /*
2885                  * Ordinary allocation.
2886                  */
2887                 else {
2888                         do_fx = 1;
2889                         nblks = del->br_blockcount;
2890                         qfield = XFS_TRANS_DQ_BCOUNT;
2891                 }
2892                 /*
2893                  * Set up del_endblock and cur for later.
2894                  */
2895                 del_endblock = del->br_startblock + del->br_blockcount;
2896                 if (cur) {
2897                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
2898                                         got.br_startblock, got.br_blockcount,
2899                                         &i)))
2900                                 goto done;
2901                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2902                 }
2903                 da_old = da_new = 0;
2904         } else {
2905                 da_old = startblockval(got.br_startblock);
2906                 da_new = 0;
2907                 nblks = 0;
2908                 do_fx = 0;
2909         }
2910         /*
2911          * Set flag value to use in switch statement.
2912          * Left-contig is 2, right-contig is 1.
2913          */
2914         switch (((got.br_startoff == del->br_startoff) << 1) |
2915                 (got_endoff == del_endoff)) {
2916         case 3:
2917                 /*
2918                  * Matches the whole extent.  Delete the entry.
2919                  */
2920                 xfs_iext_remove(ip, *idx, 1,
2921                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
2922                 --*idx;
2923                 if (delay)
2924                         break;
2925
2926                 XFS_IFORK_NEXT_SET(ip, whichfork,
2927                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2928                 flags |= XFS_ILOG_CORE;
2929                 if (!cur) {
2930                         flags |= xfs_ilog_fext(whichfork);
2931                         break;
2932                 }
2933                 if ((error = xfs_btree_delete(cur, &i)))
2934                         goto done;
2935                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2936                 break;
2937
2938         case 2:
2939                 /*
2940                  * Deleting the first part of the extent.
2941                  */
2942                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2943                 xfs_bmbt_set_startoff(ep, del_endoff);
2944                 temp = got.br_blockcount - del->br_blockcount;
2945                 xfs_bmbt_set_blockcount(ep, temp);
2946                 if (delay) {
2947                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2948                                 da_old);
2949                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2950                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2951                         da_new = temp;
2952                         break;
2953                 }
2954                 xfs_bmbt_set_startblock(ep, del_endblock);
2955                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2956                 if (!cur) {
2957                         flags |= xfs_ilog_fext(whichfork);
2958                         break;
2959                 }
2960                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
2961                                 got.br_blockcount - del->br_blockcount,
2962                                 got.br_state)))
2963                         goto done;
2964                 break;
2965
2966         case 1:
2967                 /*
2968                  * Deleting the last part of the extent.
2969                  */
2970                 temp = got.br_blockcount - del->br_blockcount;
2971                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2972                 xfs_bmbt_set_blockcount(ep, temp);
2973                 if (delay) {
2974                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2975                                 da_old);
2976                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2977                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2978                         da_new = temp;
2979                         break;
2980                 }
2981                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2982                 if (!cur) {
2983                         flags |= xfs_ilog_fext(whichfork);
2984                         break;
2985                 }
2986                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
2987                                 got.br_startblock,
2988                                 got.br_blockcount - del->br_blockcount,
2989                                 got.br_state)))
2990                         goto done;
2991                 break;
2992
2993         case 0:
2994                 /*
2995                  * Deleting the middle of the extent.
2996                  */
2997                 temp = del->br_startoff - got.br_startoff;
2998                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2999                 xfs_bmbt_set_blockcount(ep, temp);
3000                 new.br_startoff = del_endoff;
3001                 temp2 = got_endoff - del_endoff;
3002                 new.br_blockcount = temp2;
3003                 new.br_state = got.br_state;
3004                 if (!delay) {
3005                         new.br_startblock = del_endblock;
3006                         flags |= XFS_ILOG_CORE;
3007                         if (cur) {
3008                                 if ((error = xfs_bmbt_update(cur,
3009                                                 got.br_startoff,
3010                                                 got.br_startblock, temp,
3011                                                 got.br_state)))
3012                                         goto done;
3013                                 if ((error = xfs_btree_increment(cur, 0, &i)))
3014                                         goto done;
3015                                 cur->bc_rec.b = new;
3016                                 error = xfs_btree_insert(cur, &i);
3017                                 if (error && error != ENOSPC)
3018                                         goto done;
3019                                 /*
3020                                  * If get no-space back from btree insert,
3021                                  * it tried a split, and we have a zero
3022                                  * block reservation.
3023                                  * Fix up our state and return the error.
3024                                  */
3025                                 if (error == ENOSPC) {
3026                                         /*
3027                                          * Reset the cursor, don't trust
3028                                          * it after any insert operation.
3029                                          */
3030                                         if ((error = xfs_bmbt_lookup_eq(cur,
3031                                                         got.br_startoff,
3032                                                         got.br_startblock,
3033                                                         temp, &i)))
3034                                                 goto done;
3035                                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3036                                         /*
3037                                          * Update the btree record back
3038                                          * to the original value.
3039                                          */
3040                                         if ((error = xfs_bmbt_update(cur,
3041                                                         got.br_startoff,
3042                                                         got.br_startblock,
3043                                                         got.br_blockcount,
3044                                                         got.br_state)))
3045                                                 goto done;
3046                                         /*
3047                                          * Reset the extent record back
3048                                          * to the original value.
3049                                          */
3050                                         xfs_bmbt_set_blockcount(ep,
3051                                                 got.br_blockcount);
3052                                         flags = 0;
3053                                         error = XFS_ERROR(ENOSPC);
3054                                         goto done;
3055                                 }
3056                                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3057                         } else
3058                                 flags |= xfs_ilog_fext(whichfork);
3059                         XFS_IFORK_NEXT_SET(ip, whichfork,
3060                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
3061                 } else {
3062                         ASSERT(whichfork == XFS_DATA_FORK);
3063                         temp = xfs_bmap_worst_indlen(ip, temp);
3064                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
3065                         temp2 = xfs_bmap_worst_indlen(ip, temp2);
3066                         new.br_startblock = nullstartblock((int)temp2);
3067                         da_new = temp + temp2;
3068                         while (da_new > da_old) {
3069                                 if (temp) {
3070                                         temp--;
3071                                         da_new--;
3072                                         xfs_bmbt_set_startblock(ep,
3073                                                 nullstartblock((int)temp));
3074                                 }
3075                                 if (da_new == da_old)
3076                                         break;
3077                                 if (temp2) {
3078                                         temp2--;
3079                                         da_new--;
3080                                         new.br_startblock =
3081                                                 nullstartblock((int)temp2);
3082                                 }
3083                         }
3084                 }
3085                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
3086                 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
3087                 ++*idx;
3088                 break;
3089         }
3090         /*
3091          * If we need to, add to list of extents to delete.
3092          */
3093         if (do_fx)
3094                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
3095                         mp);
3096         /*
3097          * Adjust inode # blocks in the file.
3098          */
3099         if (nblks)
3100                 ip->i_d.di_nblocks -= nblks;
3101         /*
3102          * Adjust quota data.
3103          */
3104         if (qfield)
3105                 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
3106
3107         /*
3108          * Account for change in delayed indirect blocks.
3109          * Nothing to do for disk quota accounting here.
3110          */
3111         ASSERT(da_old >= da_new);
3112         if (da_old > da_new) {
3113                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
3114                         (int64_t)(da_old - da_new), 0);
3115         }
3116 done:
3117         *logflagsp = flags;
3118         return error;
3119 }
3120
3121 /*
3122  * Remove the entry "free" from the free item list.  Prev points to the
3123  * previous entry, unless "free" is the head of the list.
3124  */
3125 STATIC void
3126 xfs_bmap_del_free(
3127         xfs_bmap_free_t         *flist, /* free item list header */
3128         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
3129         xfs_bmap_free_item_t    *free)  /* list item to be freed */
3130 {
3131         if (prev)
3132                 prev->xbfi_next = free->xbfi_next;
3133         else
3134                 flist->xbf_first = free->xbfi_next;
3135         flist->xbf_count--;
3136         kmem_zone_free(xfs_bmap_free_item_zone, free);
3137 }
3138
3139 /*
3140  * Convert an extents-format file into a btree-format file.
3141  * The new file will have a root block (in the inode) and a single child block.
3142  */
3143 STATIC int                                      /* error */
3144 xfs_bmap_extents_to_btree(
3145         xfs_trans_t             *tp,            /* transaction pointer */
3146         xfs_inode_t             *ip,            /* incore inode pointer */
3147         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
3148         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
3149         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
3150         int                     wasdel,         /* converting a delayed alloc */
3151         int                     *logflagsp,     /* inode logging flags */
3152         int                     whichfork)      /* data or attr fork */
3153 {
3154         struct xfs_btree_block  *ablock;        /* allocated (child) bt block */
3155         xfs_buf_t               *abp;           /* buffer for ablock */
3156         xfs_alloc_arg_t         args;           /* allocation arguments */
3157         xfs_bmbt_rec_t          *arp;           /* child record pointer */
3158         struct xfs_btree_block  *block;         /* btree root block */
3159         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
3160         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
3161         int                     error;          /* error return value */
3162         xfs_extnum_t            i, cnt;         /* extent record index */
3163         xfs_ifork_t             *ifp;           /* inode fork pointer */
3164         xfs_bmbt_key_t          *kp;            /* root block key pointer */
3165         xfs_mount_t             *mp;            /* mount structure */
3166         xfs_extnum_t            nextents;       /* number of file extents */
3167         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
3168
3169         ifp = XFS_IFORK_PTR(ip, whichfork);
3170         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3171         ASSERT(ifp->if_ext_max ==
3172                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
3173         /*
3174          * Make space in the inode incore.
3175          */
3176         xfs_iroot_realloc(ip, 1, whichfork);
3177         ifp->if_flags |= XFS_IFBROOT;
3178
3179         /*
3180          * Fill in the root.
3181          */
3182         block = ifp->if_broot;
3183         block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3184         block->bb_level = cpu_to_be16(1);
3185         block->bb_numrecs = cpu_to_be16(1);
3186         block->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3187         block->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3188
3189         /*
3190          * Need a cursor.  Can't allocate until bb_level is filled in.
3191          */
3192         mp = ip->i_mount;
3193         cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
3194         cur->bc_private.b.firstblock = *firstblock;
3195         cur->bc_private.b.flist = flist;
3196         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3197         /*
3198          * Convert to a btree with two levels, one record in root.
3199          */
3200         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3201         args.tp = tp;
3202         args.mp = mp;
3203         args.firstblock = *firstblock;
3204         if (*firstblock == NULLFSBLOCK) {
3205                 args.type = XFS_ALLOCTYPE_START_BNO;
3206                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3207         } else if (flist->xbf_low) {
3208                 args.type = XFS_ALLOCTYPE_START_BNO;
3209                 args.fsbno = *firstblock;
3210         } else {
3211                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3212                 args.fsbno = *firstblock;
3213         }
3214         args.minlen = args.maxlen = args.prod = 1;
3215         args.total = args.minleft = args.alignment = args.mod = args.isfl =
3216                 args.minalignslop = 0;
3217         args.wasdel = wasdel;
3218         *logflagsp = 0;
3219         if ((error = xfs_alloc_vextent(&args))) {
3220                 xfs_iroot_realloc(ip, -1, whichfork);
3221                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3222                 return error;
3223         }
3224         /*
3225          * Allocation can't fail, the space was reserved.
3226          */
3227         ASSERT(args.fsbno != NULLFSBLOCK);
3228         ASSERT(*firstblock == NULLFSBLOCK ||
3229                args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3230                (flist->xbf_low &&
3231                 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3232         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3233         cur->bc_private.b.allocated++;
3234         ip->i_d.di_nblocks++;
3235         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3236         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3237         /*
3238          * Fill in the child block.
3239          */
3240         ablock = XFS_BUF_TO_BLOCK(abp);
3241         ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3242         ablock->bb_level = 0;
3243         ablock->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3244         ablock->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3245         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3246         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3247         for (cnt = i = 0; i < nextents; i++) {
3248                 ep = xfs_iext_get_ext(ifp, i);
3249                 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
3250                         arp->l0 = cpu_to_be64(ep->l0);
3251                         arp->l1 = cpu_to_be64(ep->l1);
3252                         arp++; cnt++;
3253                 }
3254         }
3255         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3256         xfs_btree_set_numrecs(ablock, cnt);
3257
3258         /*
3259          * Fill in the root key and pointer.
3260          */
3261         kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
3262         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3263         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
3264         pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
3265                                                 be16_to_cpu(block->bb_level)));
3266         *pp = cpu_to_be64(args.fsbno);
3267
3268         /*
3269          * Do all this logging at the end so that
3270          * the root is at the right level.
3271          */
3272         xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
3273         xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3274         ASSERT(*curp == NULL);
3275         *curp = cur;
3276         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
3277         return 0;
3278 }
3279
3280 /*
3281  * Calculate the default attribute fork offset for newly created inodes.
3282  */
3283 uint
3284 xfs_default_attroffset(
3285         struct xfs_inode        *ip)
3286 {
3287         struct xfs_mount        *mp = ip->i_mount;
3288         uint                    offset;
3289
3290         if (mp->m_sb.sb_inodesize == 256) {
3291                 offset = XFS_LITINO(mp) -
3292                                 XFS_BMDR_SPACE_CALC(MINABTPTRS);
3293         } else {
3294                 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
3295         }
3296
3297         ASSERT(offset < XFS_LITINO(mp));
3298         return offset;
3299 }
3300
3301 /*
3302  * Helper routine to reset inode di_forkoff field when switching
3303  * attribute fork from local to extent format - we reset it where
3304  * possible to make space available for inline data fork extents.
3305  */
3306 STATIC void
3307 xfs_bmap_forkoff_reset(
3308         xfs_mount_t     *mp,
3309         xfs_inode_t     *ip,
3310         int             whichfork)
3311 {
3312         if (whichfork == XFS_ATTR_FORK &&
3313             ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
3314             ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
3315             ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
3316                 uint    dfl_forkoff = xfs_default_attroffset(ip) >> 3;
3317
3318                 if (dfl_forkoff > ip->i_d.di_forkoff) {
3319                         ip->i_d.di_forkoff = dfl_forkoff;
3320                         ip->i_df.if_ext_max =
3321                                 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t);
3322                         ip->i_afp->if_ext_max =
3323                                 XFS_IFORK_ASIZE(ip) / sizeof(xfs_bmbt_rec_t);
3324                 }
3325         }
3326 }
3327
3328 /*
3329  * Convert a local file to an extents file.
3330  * This code is out of bounds for data forks of regular files,
3331  * since the file data needs to get logged so things will stay consistent.
3332  * (The bmap-level manipulations are ok, though).
3333  */
3334 STATIC int                              /* error */
3335 xfs_bmap_local_to_extents(
3336         xfs_trans_t     *tp,            /* transaction pointer */
3337         xfs_inode_t     *ip,            /* incore inode pointer */
3338         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
3339         xfs_extlen_t    total,          /* total blocks needed by transaction */
3340         int             *logflagsp,     /* inode logging flags */
3341         int             whichfork)      /* data or attr fork */
3342 {
3343         int             error;          /* error return value */
3344         int             flags;          /* logging flags returned */
3345         xfs_ifork_t     *ifp;           /* inode fork pointer */
3346
3347         /*
3348          * We don't want to deal with the case of keeping inode data inline yet.
3349          * So sending the data fork of a regular inode is invalid.
3350          */
3351         ASSERT(!((ip->i_d.di_mode & S_IFMT) == S_IFREG &&
3352                  whichfork == XFS_DATA_FORK));
3353         ifp = XFS_IFORK_PTR(ip, whichfork);
3354         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3355         flags = 0;
3356         error = 0;
3357         if (ifp->if_bytes) {
3358                 xfs_alloc_arg_t args;   /* allocation arguments */
3359                 xfs_buf_t       *bp;    /* buffer for extent block */
3360                 xfs_bmbt_rec_host_t *ep;/* extent record pointer */
3361
3362                 args.tp = tp;
3363                 args.mp = ip->i_mount;
3364                 args.firstblock = *firstblock;
3365                 ASSERT((ifp->if_flags &
3366                         (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == XFS_IFINLINE);
3367                 /*
3368                  * Allocate a block.  We know we need only one, since the
3369                  * file currently fits in an inode.
3370                  */
3371                 if (*firstblock == NULLFSBLOCK) {
3372                         args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3373                         args.type = XFS_ALLOCTYPE_START_BNO;
3374                 } else {
3375                         args.fsbno = *firstblock;
3376                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
3377                 }
3378                 args.total = total;
3379                 args.mod = args.minleft = args.alignment = args.wasdel =
3380                         args.isfl = args.minalignslop = 0;
3381                 args.minlen = args.maxlen = args.prod = 1;
3382                 if ((error = xfs_alloc_vextent(&args)))
3383                         goto done;
3384                 /*
3385                  * Can't fail, the space was reserved.
3386                  */
3387                 ASSERT(args.fsbno != NULLFSBLOCK);
3388                 ASSERT(args.len == 1);
3389                 *firstblock = args.fsbno;
3390                 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3391                 memcpy((char *)XFS_BUF_PTR(bp), ifp->if_u1.if_data,
3392                         ifp->if_bytes);
3393                 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3394                 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3395                 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3396                 xfs_iext_add(ifp, 0, 1);
3397                 ep = xfs_iext_get_ext(ifp, 0);
3398                 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3399                 trace_xfs_bmap_post_update(ip, 0,
3400                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
3401                                 _THIS_IP_);
3402                 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3403                 ip->i_d.di_nblocks = 1;
3404                 xfs_trans_mod_dquot_byino(tp, ip,
3405                         XFS_TRANS_DQ_BCOUNT, 1L);
3406                 flags |= xfs_ilog_fext(whichfork);
3407         } else {
3408                 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3409                 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3410         }
3411         ifp->if_flags &= ~XFS_IFINLINE;
3412         ifp->if_flags |= XFS_IFEXTENTS;
3413         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3414         flags |= XFS_ILOG_CORE;
3415 done:
3416         *logflagsp = flags;
3417         return error;
3418 }
3419
3420 /*
3421  * Search the extent records for the entry containing block bno.
3422  * If bno lies in a hole, point to the next entry.  If bno lies
3423  * past eof, *eofp will be set, and *prevp will contain the last
3424  * entry (null if none).  Else, *lastxp will be set to the index
3425  * of the found entry; *gotp will contain the entry.
3426  */
3427 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
3428 xfs_bmap_search_multi_extents(
3429         xfs_ifork_t     *ifp,           /* inode fork pointer */
3430         xfs_fileoff_t   bno,            /* block number searched for */
3431         int             *eofp,          /* out: end of file found */
3432         xfs_extnum_t    *lastxp,        /* out: last extent index */
3433         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3434         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3435 {
3436         xfs_bmbt_rec_host_t *ep;                /* extent record pointer */
3437         xfs_extnum_t    lastx;          /* last extent index */
3438
3439         /*
3440          * Initialize the extent entry structure to catch access to
3441          * uninitialized br_startblock field.
3442          */
3443         gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
3444         gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3445         gotp->br_state = XFS_EXT_INVALID;
3446 #if XFS_BIG_BLKNOS
3447         gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
3448 #else
3449         gotp->br_startblock = 0xffffa5a5;
3450 #endif
3451         prevp->br_startoff = NULLFILEOFF;
3452
3453         ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
3454         if (lastx > 0) {
3455                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
3456         }
3457         if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
3458                 xfs_bmbt_get_all(ep, gotp);
3459                 *eofp = 0;
3460         } else {
3461                 if (lastx > 0) {
3462                         *gotp = *prevp;
3463                 }
3464                 *eofp = 1;
3465                 ep = NULL;
3466         }
3467         *lastxp = lastx;
3468         return ep;
3469 }
3470
3471 /*
3472  * Search the extents list for the inode, for the extent containing bno.
3473  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
3474  * *eofp will be set, and *prevp will contain the last entry (null if none).
3475  * Else, *lastxp will be set to the index of the found
3476  * entry; *gotp will contain the entry.
3477  */
3478 STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
3479 xfs_bmap_search_extents(
3480         xfs_inode_t     *ip,            /* incore inode pointer */
3481         xfs_fileoff_t   bno,            /* block number searched for */
3482         int             fork,           /* data or attr fork */
3483         int             *eofp,          /* out: end of file found */
3484         xfs_extnum_t    *lastxp,        /* out: last extent index */
3485         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3486         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3487 {
3488         xfs_ifork_t     *ifp;           /* inode fork pointer */
3489         xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
3490
3491         XFS_STATS_INC(xs_look_exlist);
3492         ifp = XFS_IFORK_PTR(ip, fork);
3493
3494         ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
3495
3496         if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
3497                      !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
3498                 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
3499                                 "Access to block zero in inode %llu "
3500                                 "start_block: %llx start_off: %llx "
3501                                 "blkcnt: %llx extent-state: %x lastx: %x\n",
3502                         (unsigned long long)ip->i_ino,
3503                         (unsigned long long)gotp->br_startblock,
3504                         (unsigned long long)gotp->br_startoff,
3505                         (unsigned long long)gotp->br_blockcount,
3506                         gotp->br_state, *lastxp);
3507                 *lastxp = NULLEXTNUM;
3508                 *eofp = 1;
3509                 return NULL;
3510         }
3511         return ep;
3512 }
3513
3514 /*
3515  * Compute the worst-case number of indirect blocks that will be used
3516  * for ip's delayed extent of length "len".
3517  */
3518 STATIC xfs_filblks_t
3519 xfs_bmap_worst_indlen(
3520         xfs_inode_t     *ip,            /* incore inode pointer */
3521         xfs_filblks_t   len)            /* delayed extent length */
3522 {
3523         int             level;          /* btree level number */
3524         int             maxrecs;        /* maximum record count at this level */
3525         xfs_mount_t     *mp;            /* mount structure */
3526         xfs_filblks_t   rval;           /* return value */
3527
3528         mp = ip->i_mount;
3529         maxrecs = mp->m_bmap_dmxr[0];
3530         for (level = 0, rval = 0;
3531              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3532              level++) {
3533                 len += maxrecs - 1;
3534                 do_div(len, maxrecs);
3535                 rval += len;
3536                 if (len == 1)
3537                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3538                                 level - 1;
3539                 if (level == 0)
3540                         maxrecs = mp->m_bmap_dmxr[1];
3541         }
3542         return rval;
3543 }
3544
3545 /*
3546  * Convert inode from non-attributed to attributed.
3547  * Must not be in a transaction, ip must not be locked.
3548  */
3549 int                                             /* error code */
3550 xfs_bmap_add_attrfork(
3551         xfs_inode_t             *ip,            /* incore inode pointer */
3552         int                     size,           /* space new attribute needs */
3553         int                     rsvd)           /* xact may use reserved blks */
3554 {
3555         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
3556         xfs_bmap_free_t         flist;          /* freed extent records */
3557         xfs_mount_t             *mp;            /* mount structure */
3558         xfs_trans_t             *tp;            /* transaction pointer */
3559         int                     blks;           /* space reservation */
3560         int                     version = 1;    /* superblock attr version */
3561         int                     committed;      /* xaction was committed */
3562         int                     logflags;       /* logging flags */
3563         int                     error;          /* error return value */
3564
3565         ASSERT(XFS_IFORK_Q(ip) == 0);
3566         ASSERT(ip->i_df.if_ext_max ==
3567                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3568
3569         mp = ip->i_mount;
3570         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3571         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3572         blks = XFS_ADDAFORK_SPACE_RES(mp);
3573         if (rsvd)
3574                 tp->t_flags |= XFS_TRANS_RESERVE;
3575         if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3576                         XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3577                 goto error0;
3578         xfs_ilock(ip, XFS_ILOCK_EXCL);
3579         error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
3580                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3581                         XFS_QMOPT_RES_REGBLKS);
3582         if (error) {
3583                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3584                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3585                 return error;
3586         }
3587         if (XFS_IFORK_Q(ip))
3588                 goto error1;
3589         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3590                 /*
3591                  * For inodes coming from pre-6.2 filesystems.
3592                  */
3593                 ASSERT(ip->i_d.di_aformat == 0);
3594                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
3595         }
3596         ASSERT(ip->i_d.di_anextents == 0);
3597
3598         xfs_trans_ijoin_ref(tp, ip, XFS_ILOCK_EXCL);
3599         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3600
3601         switch (ip->i_d.di_format) {
3602         case XFS_DINODE_FMT_DEV:
3603                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
3604                 break;
3605         case XFS_DINODE_FMT_UUID:
3606                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
3607                 break;
3608         case XFS_DINODE_FMT_LOCAL:
3609         case XFS_DINODE_FMT_EXTENTS:
3610         case XFS_DINODE_FMT_BTREE:
3611                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
3612                 if (!ip->i_d.di_forkoff)
3613                         ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
3614                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
3615                         version = 2;
3616                 break;
3617         default:
3618                 ASSERT(0);
3619                 error = XFS_ERROR(EINVAL);
3620                 goto error1;
3621         }
3622         ip->i_df.if_ext_max =
3623                 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3624         ASSERT(ip->i_afp == NULL);
3625         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
3626         ip->i_afp->if_ext_max =
3627                 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3628         ip->i_afp->if_flags = XFS_IFEXTENTS;
3629         logflags = 0;
3630         xfs_bmap_init(&flist, &firstblock);
3631         switch (ip->i_d.di_format) {
3632         case XFS_DINODE_FMT_LOCAL:
3633                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
3634                         &logflags);
3635                 break;
3636         case XFS_DINODE_FMT_EXTENTS:
3637                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
3638                         &flist, &logflags);
3639                 break;
3640         case XFS_DINODE_FMT_BTREE:
3641                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
3642                         &logflags);
3643                 break;
3644         default:
3645                 error = 0;
3646                 break;
3647         }
3648         if (logflags)
3649                 xfs_trans_log_inode(tp, ip, logflags);
3650         if (error)
3651                 goto error2;
3652         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
3653            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
3654                 __int64_t sbfields = 0;
3655
3656                 spin_lock(&mp->m_sb_lock);
3657                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
3658                         xfs_sb_version_addattr(&mp->m_sb);
3659                         sbfields |= XFS_SB_VERSIONNUM;
3660                 }
3661                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
3662                         xfs_sb_version_addattr2(&mp->m_sb);
3663                         sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
3664                 }
3665                 if (sbfields) {
3666                         spin_unlock(&mp->m_sb_lock);
3667                         xfs_mod_sb(tp, sbfields);
3668                 } else
3669                         spin_unlock(&mp->m_sb_lock);
3670         }
3671         if ((error = xfs_bmap_finish(&tp, &flist, &committed)))
3672                 goto error2;
3673         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3674         ASSERT(ip->i_df.if_ext_max ==
3675                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3676         return error;
3677 error2:
3678         xfs_bmap_cancel(&flist);
3679 error1:
3680         xfs_iunlock(ip, XFS_ILOCK_EXCL);
3681 error0:
3682         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
3683         ASSERT(ip->i_df.if_ext_max ==
3684                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3685         return error;
3686 }
3687
3688 /*
3689  * Add the extent to the list of extents to be free at transaction end.
3690  * The list is maintained sorted (by block number).
3691  */
3692 /* ARGSUSED */
3693 void
3694 xfs_bmap_add_free(
3695         xfs_fsblock_t           bno,            /* fs block number of extent */
3696         xfs_filblks_t           len,            /* length of extent */
3697         xfs_bmap_free_t         *flist,         /* list of extents */
3698         xfs_mount_t             *mp)            /* mount point structure */
3699 {
3700         xfs_bmap_free_item_t    *cur;           /* current (next) element */
3701         xfs_bmap_free_item_t    *new;           /* new element */
3702         xfs_bmap_free_item_t    *prev;          /* previous element */
3703 #ifdef DEBUG
3704         xfs_agnumber_t          agno;
3705         xfs_agblock_t           agbno;
3706
3707         ASSERT(bno != NULLFSBLOCK);
3708         ASSERT(len > 0);
3709         ASSERT(len <= MAXEXTLEN);
3710         ASSERT(!isnullstartblock(bno));
3711         agno = XFS_FSB_TO_AGNO(mp, bno);
3712         agbno = XFS_FSB_TO_AGBNO(mp, bno);
3713         ASSERT(agno < mp->m_sb.sb_agcount);
3714         ASSERT(agbno < mp->m_sb.sb_agblocks);
3715         ASSERT(len < mp->m_sb.sb_agblocks);
3716         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
3717 #endif
3718         ASSERT(xfs_bmap_free_item_zone != NULL);
3719         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
3720         new->xbfi_startblock = bno;
3721         new->xbfi_blockcount = (xfs_extlen_t)len;
3722         for (prev = NULL, cur = flist->xbf_first;
3723              cur != NULL;
3724              prev = cur, cur = cur->xbfi_next) {
3725                 if (cur->xbfi_startblock >= bno)
3726                         break;
3727         }
3728         if (prev)
3729                 prev->xbfi_next = new;
3730         else
3731                 flist->xbf_first = new;
3732         new->xbfi_next = cur;
3733         flist->xbf_count++;
3734 }
3735
3736 /*
3737  * Compute and fill in the value of the maximum depth of a bmap btree
3738  * in this filesystem.  Done once, during mount.
3739  */
3740 void
3741 xfs_bmap_compute_maxlevels(
3742         xfs_mount_t     *mp,            /* file system mount structure */
3743         int             whichfork)      /* data or attr fork */
3744 {
3745         int             level;          /* btree level */
3746         uint            maxblocks;      /* max blocks at this level */
3747         uint            maxleafents;    /* max leaf entries possible */
3748         int             maxrootrecs;    /* max records in root block */
3749         int             minleafrecs;    /* min records in leaf block */
3750         int             minnoderecs;    /* min records in node block */
3751         int             sz;             /* root block size */
3752
3753         /*
3754          * The maximum number of extents in a file, hence the maximum
3755          * number of leaf entries, is controlled by the type of di_nextents
3756          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
3757          * (a signed 16-bit number, xfs_aextnum_t).
3758          *
3759          * Note that we can no longer assume that if we are in ATTR1 that
3760          * the fork offset of all the inodes will be
3761          * (xfs_default_attroffset(ip) >> 3) because we could have mounted
3762          * with ATTR2 and then mounted back with ATTR1, keeping the
3763          * di_forkoff's fixed but probably at various positions. Therefore,
3764          * for both ATTR1 and ATTR2 we have to assume the worst case scenario
3765          * of a minimum size available.
3766          */
3767         if (whichfork == XFS_DATA_FORK) {
3768                 maxleafents = MAXEXTNUM;
3769                 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
3770         } else {
3771                 maxleafents = MAXAEXTNUM;
3772                 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
3773         }
3774         maxrootrecs = xfs_bmdr_maxrecs(mp, sz, 0);
3775         minleafrecs = mp->m_bmap_dmnr[0];
3776         minnoderecs = mp->m_bmap_dmnr[1];
3777         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
3778         for (level = 1; maxblocks > 1; level++) {
3779                 if (maxblocks <= maxrootrecs)
3780                         maxblocks = 1;
3781                 else
3782                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
3783         }
3784         mp->m_bm_maxlevels[whichfork] = level;
3785 }
3786
3787 /*
3788  * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
3789  * caller.  Frees all the extents that need freeing, which must be done
3790  * last due to locking considerations.  We never free any extents in
3791  * the first transaction.  This is to allow the caller to make the first
3792  * transaction a synchronous one so that the pointers to the data being
3793  * broken in this transaction will be permanent before the data is actually
3794  * freed.  This is necessary to prevent blocks from being reallocated
3795  * and written to before the free and reallocation are actually permanent.
3796  * We do not just make the first transaction synchronous here, because
3797  * there are more efficient ways to gain the same protection in some cases
3798  * (see the file truncation code).
3799  *
3800  * Return 1 if the given transaction was committed and a new one
3801  * started, and 0 otherwise in the committed parameter.
3802  */
3803 /*ARGSUSED*/
3804 int                                             /* error */
3805 xfs_bmap_finish(
3806         xfs_trans_t             **tp,           /* transaction pointer addr */
3807         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
3808         int                     *committed)     /* xact committed or not */
3809 {
3810         xfs_efd_log_item_t      *efd;           /* extent free data */
3811         xfs_efi_log_item_t      *efi;           /* extent free intention */
3812         int                     error;          /* error return value */
3813         xfs_bmap_free_item_t    *free;          /* free extent item */
3814         unsigned int            logres;         /* new log reservation */
3815         unsigned int            logcount;       /* new log count */
3816         xfs_mount_t             *mp;            /* filesystem mount structure */
3817         xfs_bmap_free_item_t    *next;          /* next item on free list */
3818         xfs_trans_t             *ntp;           /* new transaction pointer */
3819
3820         ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
3821         if (flist->xbf_count == 0) {
3822                 *committed = 0;
3823                 return 0;
3824         }
3825         ntp = *tp;
3826         efi = xfs_trans_get_efi(ntp, flist->xbf_count);
3827         for (free = flist->xbf_first; free; free = free->xbfi_next)
3828                 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
3829                         free->xbfi_blockcount);
3830         logres = ntp->t_log_res;
3831         logcount = ntp->t_log_count;
3832         ntp = xfs_trans_dup(*tp);
3833         error = xfs_trans_commit(*tp, 0);
3834         *tp = ntp;
3835         *committed = 1;
3836         /*
3837          * We have a new transaction, so we should return committed=1,
3838          * even though we're returning an error.
3839          */
3840         if (error)
3841                 return error;
3842
3843         /*
3844          * transaction commit worked ok so we can drop the extra ticket
3845          * reference that we gained in xfs_trans_dup()
3846          */
3847         xfs_log_ticket_put(ntp->t_ticket);
3848
3849         if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
3850                         logcount)))
3851                 return error;
3852         efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
3853         for (free = flist->xbf_first; free != NULL; free = next) {
3854                 next = free->xbfi_next;
3855                 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
3856                                 free->xbfi_blockcount))) {
3857                         /*
3858                          * The bmap free list will be cleaned up at a
3859                          * higher level.  The EFI will be canceled when
3860                          * this transaction is aborted.
3861                          * Need to force shutdown here to make sure it
3862                          * happens, since this transaction may not be
3863                          * dirty yet.
3864                          */
3865                         mp = ntp->t_mountp;
3866                         if (!XFS_FORCED_SHUTDOWN(mp))
3867                                 xfs_force_shutdown(mp,
3868                                                    (error == EFSCORRUPTED) ?
3869                                                    SHUTDOWN_CORRUPT_INCORE :
3870                                                    SHUTDOWN_META_IO_ERROR);
3871                         return error;
3872                 }
3873                 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
3874                         free->xbfi_blockcount);
3875                 xfs_bmap_del_free(flist, NULL, free);
3876         }
3877         return 0;
3878 }
3879
3880 /*
3881  * Free up any items left in the list.
3882  */
3883 void
3884 xfs_bmap_cancel(
3885         xfs_bmap_free_t         *flist) /* list of bmap_free_items */
3886 {
3887         xfs_bmap_free_item_t    *free;  /* free list item */
3888         xfs_bmap_free_item_t    *next;
3889
3890         if (flist->xbf_count == 0)
3891                 return;
3892         ASSERT(flist->xbf_first != NULL);
3893         for (free = flist->xbf_first; free; free = next) {
3894                 next = free->xbfi_next;
3895                 xfs_bmap_del_free(flist, NULL, free);
3896         }
3897         ASSERT(flist->xbf_count == 0);
3898 }
3899
3900 /*
3901  * Returns the file-relative block number of the first unused block(s)
3902  * in the file with at least "len" logically contiguous blocks free.
3903  * This is the lowest-address hole if the file has holes, else the first block
3904  * past the end of file.
3905  * Return 0 if the file is currently local (in-inode).
3906  */
3907 int                                             /* error */
3908 xfs_bmap_first_unused(
3909         xfs_trans_t     *tp,                    /* transaction pointer */
3910         xfs_inode_t     *ip,                    /* incore inode */
3911         xfs_extlen_t    len,                    /* size of hole to find */
3912         xfs_fileoff_t   *first_unused,          /* unused block */
3913         int             whichfork)              /* data or attr fork */
3914 {
3915         int             error;                  /* error return value */
3916         int             idx;                    /* extent record index */
3917         xfs_ifork_t     *ifp;                   /* inode fork pointer */
3918         xfs_fileoff_t   lastaddr;               /* last block number seen */
3919         xfs_fileoff_t   lowest;                 /* lowest useful block */
3920         xfs_fileoff_t   max;                    /* starting useful block */
3921         xfs_fileoff_t   off;                    /* offset for this block */
3922         xfs_extnum_t    nextents;               /* number of extent entries */
3923
3924         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
3925                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
3926                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3927         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3928                 *first_unused = 0;
3929                 return 0;
3930         }
3931         ifp = XFS_IFORK_PTR(ip, whichfork);
3932         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3933             (error = xfs_iread_extents(tp, ip, whichfork)))
3934                 return error;
3935         lowest = *first_unused;
3936         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3937         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
3938                 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
3939                 off = xfs_bmbt_get_startoff(ep);
3940                 /*
3941                  * See if the hole before this extent will work.
3942                  */
3943                 if (off >= lowest + len && off - max >= len) {
3944                         *first_unused = max;
3945                         return 0;
3946                 }
3947                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
3948                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
3949         }
3950         *first_unused = max;
3951         return 0;
3952 }
3953
3954 /*
3955  * Returns the file-relative block number of the last block + 1 before
3956  * last_block (input value) in the file.
3957  * This is not based on i_size, it is based on the extent records.
3958  * Returns 0 for local files, as they do not have extent records.
3959  */
3960 int                                             /* error */
3961 xfs_bmap_last_before(
3962         xfs_trans_t     *tp,                    /* transaction pointer */
3963         xfs_inode_t     *ip,                    /* incore inode */
3964         xfs_fileoff_t   *last_block,            /* last block */
3965         int             whichfork)              /* data or attr fork */
3966 {
3967         xfs_fileoff_t   bno;                    /* input file offset */
3968         int             eof;                    /* hit end of file */
3969         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
3970         int             error;                  /* error return value */
3971         xfs_bmbt_irec_t got;                    /* current extent value */
3972         xfs_ifork_t     *ifp;                   /* inode fork pointer */
3973         xfs_extnum_t    lastx;                  /* last extent used */
3974         xfs_bmbt_irec_t prev;                   /* previous extent value */
3975
3976         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
3977             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
3978             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
3979                return XFS_ERROR(EIO);
3980         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3981                 *last_block = 0;
3982                 return 0;
3983         }
3984         ifp = XFS_IFORK_PTR(ip, whichfork);
3985         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3986             (error = xfs_iread_extents(tp, ip, whichfork)))
3987                 return error;
3988         bno = *last_block - 1;
3989         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
3990                 &prev);
3991         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
3992                 if (prev.br_startoff == NULLFILEOFF)
3993                         *last_block = 0;
3994                 else
3995                         *last_block = prev.br_startoff + prev.br_blockcount;
3996         }
3997         /*
3998          * Otherwise *last_block is already the right answer.
3999          */
4000         return 0;
4001 }
4002
4003 /*
4004  * Returns the file-relative block number of the first block past eof in
4005  * the file.  This is not based on i_size, it is based on the extent records.
4006  * Returns 0 for local files, as they do not have extent records.
4007  */
4008 int                                             /* error */
4009 xfs_bmap_last_offset(
4010         xfs_trans_t     *tp,                    /* transaction pointer */
4011         xfs_inode_t     *ip,                    /* incore inode */
4012         xfs_fileoff_t   *last_block,            /* last block */
4013         int             whichfork)              /* data or attr fork */
4014 {
4015         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
4016         int             error;                  /* error return value */
4017         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4018         xfs_extnum_t    nextents;               /* number of extent entries */
4019
4020         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4021             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4022             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4023                return XFS_ERROR(EIO);
4024         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4025                 *last_block = 0;
4026                 return 0;
4027         }
4028         ifp = XFS_IFORK_PTR(ip, whichfork);
4029         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4030             (error = xfs_iread_extents(tp, ip, whichfork)))
4031                 return error;
4032         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4033         if (!nextents) {
4034                 *last_block = 0;
4035                 return 0;
4036         }
4037         ep = xfs_iext_get_ext(ifp, nextents - 1);
4038         *last_block = xfs_bmbt_get_startoff(ep) + xfs_bmbt_get_blockcount(ep);
4039         return 0;
4040 }
4041
4042 /*
4043  * Returns whether the selected fork of the inode has exactly one
4044  * block or not.  For the data fork we check this matches di_size,
4045  * implying the file's range is 0..bsize-1.
4046  */
4047 int                                     /* 1=>1 block, 0=>otherwise */
4048 xfs_bmap_one_block(
4049         xfs_inode_t     *ip,            /* incore inode */
4050         int             whichfork)      /* data or attr fork */
4051 {
4052         xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
4053         xfs_ifork_t     *ifp;           /* inode fork pointer */
4054         int             rval;           /* return value */
4055         xfs_bmbt_irec_t s;              /* internal version of extent */
4056
4057 #ifndef DEBUG
4058         if (whichfork == XFS_DATA_FORK) {
4059                 return ((ip->i_d.di_mode & S_IFMT) == S_IFREG) ?
4060                         (ip->i_size == ip->i_mount->m_sb.sb_blocksize) :
4061                         (ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize);
4062         }
4063 #endif  /* !DEBUG */
4064         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4065                 return 0;
4066         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4067                 return 0;
4068         ifp = XFS_IFORK_PTR(ip, whichfork);
4069         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4070         ep = xfs_iext_get_ext(ifp, 0);
4071         xfs_bmbt_get_all(ep, &s);
4072         rval = s.br_startoff == 0 && s.br_blockcount == 1;
4073         if (rval && whichfork == XFS_DATA_FORK)
4074                 ASSERT(ip->i_size == ip->i_mount->m_sb.sb_blocksize);
4075         return rval;
4076 }
4077
4078 STATIC int
4079 xfs_bmap_sanity_check(
4080         struct xfs_mount        *mp,
4081         struct xfs_buf          *bp,
4082         int                     level)
4083 {
4084         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4085
4086         if (be32_to_cpu(block->bb_magic) != XFS_BMAP_MAGIC ||
4087             be16_to_cpu(block->bb_level) != level ||
4088             be16_to_cpu(block->bb_numrecs) == 0 ||
4089             be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
4090                 return 0;
4091         return 1;
4092 }
4093
4094 /*
4095  * Read in the extents to if_extents.
4096  * All inode fields are set up by caller, we just traverse the btree
4097  * and copy the records in. If the file system cannot contain unwritten
4098  * extents, the records are checked for no "state" flags.
4099  */
4100 int                                     /* error */
4101 xfs_bmap_read_extents(
4102         xfs_trans_t             *tp,    /* transaction pointer */
4103         xfs_inode_t             *ip,    /* incore inode */
4104         int                     whichfork) /* data or attr fork */
4105 {
4106         struct xfs_btree_block  *block; /* current btree block */
4107         xfs_fsblock_t           bno;    /* block # of "block" */
4108         xfs_buf_t               *bp;    /* buffer for "block" */
4109         int                     error;  /* error return value */
4110         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
4111         xfs_extnum_t            i, j;   /* index into the extents list */
4112         xfs_ifork_t             *ifp;   /* fork structure */
4113         int                     level;  /* btree level, for checking */
4114         xfs_mount_t             *mp;    /* file system mount structure */
4115         __be64                  *pp;    /* pointer to block address */
4116         /* REFERENCED */
4117         xfs_extnum_t            room;   /* number of entries there's room for */
4118
4119         bno = NULLFSBLOCK;
4120         mp = ip->i_mount;
4121         ifp = XFS_IFORK_PTR(ip, whichfork);
4122         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4123                                         XFS_EXTFMT_INODE(ip);
4124         block = ifp->if_broot;
4125         /*
4126          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4127          */
4128         level = be16_to_cpu(block->bb_level);
4129         ASSERT(level > 0);
4130         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
4131         bno = be64_to_cpu(*pp);
4132         ASSERT(bno != NULLDFSBNO);
4133         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
4134         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
4135         /*
4136          * Go down the tree until leaf level is reached, following the first
4137          * pointer (leftmost) at each level.
4138          */
4139         while (level-- > 0) {
4140                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4141                                 XFS_BMAP_BTREE_REF)))
4142                         return error;
4143                 block = XFS_BUF_TO_BLOCK(bp);
4144                 XFS_WANT_CORRUPTED_GOTO(
4145                         xfs_bmap_sanity_check(mp, bp, level),
4146                         error0);
4147                 if (level == 0)
4148                         break;
4149                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
4150                 bno = be64_to_cpu(*pp);
4151                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
4152                 xfs_trans_brelse(tp, bp);
4153         }
4154         /*
4155          * Here with bp and block set to the leftmost leaf node in the tree.
4156          */
4157         room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4158         i = 0;
4159         /*
4160          * Loop over all leaf nodes.  Copy information to the extent records.
4161          */
4162         for (;;) {
4163                 xfs_bmbt_rec_t  *frp;
4164                 xfs_fsblock_t   nextbno;
4165                 xfs_extnum_t    num_recs;
4166                 xfs_extnum_t    start;
4167
4168
4169                 num_recs = xfs_btree_get_numrecs(block);
4170                 if (unlikely(i + num_recs > room)) {
4171                         ASSERT(i + num_recs <= room);
4172                         xfs_warn(ip->i_mount,
4173                                 "corrupt dinode %Lu, (btree extents).",
4174                                 (unsigned long long) ip->i_ino);
4175                         XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
4176                                 XFS_ERRLEVEL_LOW, ip->i_mount, block);
4177                         goto error0;
4178                 }
4179                 XFS_WANT_CORRUPTED_GOTO(
4180                         xfs_bmap_sanity_check(mp, bp, 0),
4181                         error0);
4182                 /*
4183                  * Read-ahead the next leaf block, if any.
4184                  */
4185                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
4186                 if (nextbno != NULLFSBLOCK)
4187                         xfs_btree_reada_bufl(mp, nextbno, 1);
4188                 /*
4189                  * Copy records into the extent records.
4190                  */
4191                 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
4192                 start = i;
4193                 for (j = 0; j < num_recs; j++, i++, frp++) {
4194                         xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
4195                         trp->l0 = be64_to_cpu(frp->l0);
4196                         trp->l1 = be64_to_cpu(frp->l1);
4197                 }
4198                 if (exntf == XFS_EXTFMT_NOSTATE) {
4199                         /*
4200                          * Check all attribute bmap btree records and
4201                          * any "older" data bmap btree records for a
4202                          * set bit in the "extent flag" position.
4203                          */
4204                         if (unlikely(xfs_check_nostate_extents(ifp,
4205                                         start, num_recs))) {
4206                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4207                                                  XFS_ERRLEVEL_LOW,
4208                                                  ip->i_mount);
4209                                 goto error0;
4210                         }
4211                 }
4212                 xfs_trans_brelse(tp, bp);
4213                 bno = nextbno;
4214                 /*
4215                  * If we've reached the end, stop.
4216                  */
4217                 if (bno == NULLFSBLOCK)
4218                         break;
4219                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4220                                 XFS_BMAP_BTREE_REF)))
4221                         return error;
4222                 block = XFS_BUF_TO_BLOCK(bp);
4223         }
4224         ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4225         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4226         XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
4227         return 0;
4228 error0:
4229         xfs_trans_brelse(tp, bp);
4230         return XFS_ERROR(EFSCORRUPTED);
4231 }
4232
4233 #ifdef DEBUG
4234 /*
4235  * Add bmap trace insert entries for all the contents of the extent records.
4236  */
4237 void
4238 xfs_bmap_trace_exlist(
4239         xfs_inode_t     *ip,            /* incore inode pointer */
4240         xfs_extnum_t    cnt,            /* count of entries in the list */
4241         int             whichfork,      /* data or attr fork */
4242         unsigned long   caller_ip)
4243 {
4244         xfs_extnum_t    idx;            /* extent record index */
4245         xfs_ifork_t     *ifp;           /* inode fork pointer */
4246         int             state = 0;
4247
4248         if (whichfork == XFS_ATTR_FORK)
4249                 state |= BMAP_ATTRFORK;
4250
4251         ifp = XFS_IFORK_PTR(ip, whichfork);
4252         ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4253         for (idx = 0; idx < cnt; idx++)
4254                 trace_xfs_extlist(ip, idx, whichfork, caller_ip);
4255 }
4256
4257 /*
4258  * Validate that the bmbt_irecs being returned from bmapi are valid
4259  * given the callers original parameters.  Specifically check the
4260  * ranges of the returned irecs to ensure that they only extent beyond
4261  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4262  */
4263 STATIC void
4264 xfs_bmap_validate_ret(
4265         xfs_fileoff_t           bno,
4266         xfs_filblks_t           len,
4267         int                     flags,
4268         xfs_bmbt_irec_t         *mval,
4269         int                     nmap,
4270         int                     ret_nmap)
4271 {
4272         int                     i;              /* index to map values */
4273
4274         ASSERT(ret_nmap <= nmap);
4275
4276         for (i = 0; i < ret_nmap; i++) {
4277                 ASSERT(mval[i].br_blockcount > 0);
4278                 if (!(flags & XFS_BMAPI_ENTIRE)) {
4279                         ASSERT(mval[i].br_startoff >= bno);
4280                         ASSERT(mval[i].br_blockcount <= len);
4281                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4282                                bno + len);
4283                 } else {
4284                         ASSERT(mval[i].br_startoff < bno + len);
4285                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4286                                bno);
4287                 }
4288                 ASSERT(i == 0 ||
4289                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4290                        mval[i].br_startoff);
4291                 if ((flags & XFS_BMAPI_WRITE) && !(flags & XFS_BMAPI_DELAY))
4292                         ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4293                                mval[i].br_startblock != HOLESTARTBLOCK);
4294                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4295                        mval[i].br_state == XFS_EXT_UNWRITTEN);
4296         }
4297 }
4298 #endif /* DEBUG */
4299
4300
4301 /*
4302  * Map file blocks to filesystem blocks.
4303  * File range is given by the bno/len pair.
4304  * Adds blocks to file if a write ("flags & XFS_BMAPI_WRITE" set)
4305  * into a hole or past eof.
4306  * Only allocates blocks from a single allocation group,
4307  * to avoid locking problems.
4308  * The returned value in "firstblock" from the first call in a transaction
4309  * must be remembered and presented to subsequent calls in "firstblock".
4310  * An upper bound for the number of blocks to be allocated is supplied to
4311  * the first call in "total"; if no allocation group has that many free
4312  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4313  */
4314 int                                     /* error */
4315 xfs_bmapi(
4316         xfs_trans_t     *tp,            /* transaction pointer */
4317         xfs_inode_t     *ip,            /* incore inode */
4318         xfs_fileoff_t   bno,            /* starting file offs. mapped */
4319         xfs_filblks_t   len,            /* length to map in file */
4320         int             flags,          /* XFS_BMAPI_... */
4321         xfs_fsblock_t   *firstblock,    /* first allocated block
4322                                            controls a.g. for allocs */
4323         xfs_extlen_t    total,          /* total blocks needed */
4324         xfs_bmbt_irec_t *mval,          /* output: map values */
4325         int             *nmap,          /* i/o: mval size/count */
4326         xfs_bmap_free_t *flist)         /* i/o: list extents to free */
4327 {
4328         xfs_fsblock_t   abno;           /* allocated block number */
4329         xfs_extlen_t    alen;           /* allocated extent length */
4330         xfs_fileoff_t   aoff;           /* allocated file offset */
4331         xfs_bmalloca_t  bma = { 0 };    /* args for xfs_bmap_alloc */
4332         xfs_btree_cur_t *cur;           /* bmap btree cursor */
4333         xfs_fileoff_t   end;            /* end of mapped file region */
4334         int             eof;            /* we've hit the end of extents */
4335         xfs_bmbt_rec_host_t *ep;        /* extent record pointer */
4336         int             error;          /* error return */
4337         xfs_bmbt_irec_t got;            /* current file extent record */
4338         xfs_ifork_t     *ifp;           /* inode fork pointer */
4339         xfs_extlen_t    indlen;         /* indirect blocks length */
4340         xfs_extnum_t    lastx;          /* last useful extent number */
4341         int             logflags;       /* flags for transaction logging */
4342         xfs_extlen_t    minleft;        /* min blocks left after allocation */
4343         xfs_extlen_t    minlen;         /* min allocation size */
4344         xfs_mount_t     *mp;            /* xfs mount structure */
4345         int             n;              /* current extent index */
4346         int             nallocs;        /* number of extents alloc'd */
4347         xfs_extnum_t    nextents;       /* number of extents in file */
4348         xfs_fileoff_t   obno;           /* old block number (offset) */
4349         xfs_bmbt_irec_t prev;           /* previous file extent record */
4350         int             tmp_logflags;   /* temp flags holder */
4351         int             whichfork;      /* data or attr fork */
4352         char            inhole;         /* current location is hole in file */
4353         char            wasdelay;       /* old extent was delayed */
4354         char            wr;             /* this is a write request */
4355         char            rt;             /* this is a realtime file */
4356 #ifdef DEBUG
4357         xfs_fileoff_t   orig_bno;       /* original block number value */
4358         int             orig_flags;     /* original flags arg value */
4359         xfs_filblks_t   orig_len;       /* original value of len arg */
4360         xfs_bmbt_irec_t *orig_mval;     /* original value of mval */
4361         int             orig_nmap;      /* original value of *nmap */
4362
4363         orig_bno = bno;
4364         orig_len = len;
4365         orig_flags = flags;
4366         orig_mval = mval;
4367         orig_nmap = *nmap;
4368 #endif
4369         ASSERT(*nmap >= 1);
4370         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP || !(flags & XFS_BMAPI_WRITE));
4371         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4372                 XFS_ATTR_FORK : XFS_DATA_FORK;
4373         mp = ip->i_mount;
4374         if (unlikely(XFS_TEST_ERROR(
4375             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4376              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4377              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4378              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4379                 XFS_ERROR_REPORT("xfs_bmapi", XFS_ERRLEVEL_LOW, mp);
4380                 return XFS_ERROR(EFSCORRUPTED);
4381         }
4382         if (XFS_FORCED_SHUTDOWN(mp))
4383                 return XFS_ERROR(EIO);
4384         rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4385         ifp = XFS_IFORK_PTR(ip, whichfork);
4386         ASSERT(ifp->if_ext_max ==
4387                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4388         if ((wr = (flags & XFS_BMAPI_WRITE)) != 0)
4389                 XFS_STATS_INC(xs_blk_mapw);
4390         else
4391                 XFS_STATS_INC(xs_blk_mapr);
4392         /*
4393          * IGSTATE flag is used to combine extents which
4394          * differ only due to the state of the extents.
4395          * This technique is used from xfs_getbmap()
4396          * when the caller does not wish to see the
4397          * separation (which is the default).
4398          *
4399          * This technique is also used when writing a
4400          * buffer which has been partially written,
4401          * (usually by being flushed during a chunkread),
4402          * to ensure one write takes place. This also
4403          * prevents a change in the xfs inode extents at
4404          * this time, intentionally. This change occurs
4405          * on completion of the write operation, in
4406          * xfs_strat_comp(), where the xfs_bmapi() call
4407          * is transactioned, and the extents combined.
4408          */
4409         if ((flags & XFS_BMAPI_IGSTATE) && wr)  /* if writing unwritten space */
4410                 wr = 0;                         /* no allocations are allowed */
4411         ASSERT(wr || !(flags & XFS_BMAPI_DELAY));
4412         logflags = 0;
4413         nallocs = 0;
4414         cur = NULL;
4415         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4416                 ASSERT(wr && tp);
4417                 if ((error = xfs_bmap_local_to_extents(tp, ip,
4418                                 firstblock, total, &logflags, whichfork)))
4419                         goto error0;
4420         }
4421         if (wr && *firstblock == NULLFSBLOCK) {
4422                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4423                         minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4424                 else
4425                         minleft = 1;
4426         } else
4427                 minleft = 0;
4428         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4429             (error = xfs_iread_extents(tp, ip, whichfork)))
4430                 goto error0;
4431         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4432                 &prev);
4433         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4434         n = 0;
4435         end = bno + len;
4436         obno = bno;
4437         bma.ip = NULL;
4438
4439         while (bno < end && n < *nmap) {
4440                 /*
4441                  * Reading past eof, act as though there's a hole
4442                  * up to end.
4443                  */
4444                 if (eof && !wr)
4445                         got.br_startoff = end;
4446                 inhole = eof || got.br_startoff > bno;
4447                 wasdelay = wr && !inhole && !(flags & XFS_BMAPI_DELAY) &&
4448                         isnullstartblock(got.br_startblock);
4449                 /*
4450                  * First, deal with the hole before the allocated space
4451                  * that we found, if any.
4452                  */
4453                 if (wr && (inhole || wasdelay)) {
4454                         /*
4455                          * For the wasdelay case, we could also just
4456                          * allocate the stuff asked for in this bmap call
4457                          * but that wouldn't be as good.
4458                          */
4459                         if (wasdelay) {
4460                                 alen = (xfs_extlen_t)got.br_blockcount;
4461                                 aoff = got.br_startoff;
4462                                 if (lastx != NULLEXTNUM && lastx) {
4463                                         ep = xfs_iext_get_ext(ifp, lastx - 1);
4464                                         xfs_bmbt_get_all(ep, &prev);
4465                                 }
4466                         } else {
4467                                 alen = (xfs_extlen_t)
4468                                         XFS_FILBLKS_MIN(len, MAXEXTLEN);
4469                                 if (!eof)
4470                                         alen = (xfs_extlen_t)
4471                                                 XFS_FILBLKS_MIN(alen,
4472                                                         got.br_startoff - bno);
4473                                 aoff = bno;
4474                         }
4475                         minlen = (flags & XFS_BMAPI_CONTIG) ? alen : 1;
4476                         if (flags & XFS_BMAPI_DELAY) {
4477                                 xfs_extlen_t    extsz;
4478
4479                                 /* Figure out the extent size, adjust alen */
4480                                 extsz = xfs_get_extsz_hint(ip);
4481                                 if (extsz) {
4482                                         /*
4483                                          * make sure we don't exceed a single
4484                                          * extent length when we align the
4485                                          * extent by reducing length we are
4486                                          * going to allocate by the maximum
4487                                          * amount extent size aligment may
4488                                          * require.
4489                                          */
4490                                         alen = XFS_FILBLKS_MIN(len,
4491                                                    MAXEXTLEN - (2 * extsz - 1));
4492                                         error = xfs_bmap_extsize_align(mp,
4493                                                         &got, &prev, extsz,
4494                                                         rt, eof,
4495                                                         flags&XFS_BMAPI_DELAY,
4496                                                         flags&XFS_BMAPI_CONVERT,
4497                                                         &aoff, &alen);
4498                                         ASSERT(!error);
4499                                 }
4500
4501                                 if (rt)
4502                                         extsz = alen / mp->m_sb.sb_rextsize;
4503
4504                                 /*
4505                                  * Make a transaction-less quota reservation for
4506                                  * delayed allocation blocks. This number gets
4507                                  * adjusted later.  We return if we haven't
4508                                  * allocated blocks already inside this loop.
4509                                  */
4510                                 error = xfs_trans_reserve_quota_nblks(
4511                                                 NULL, ip, (long)alen, 0,
4512                                                 rt ? XFS_QMOPT_RES_RTBLKS :
4513                                                      XFS_QMOPT_RES_REGBLKS);
4514                                 if (error) {
4515                                         if (n == 0) {
4516                                                 *nmap = 0;
4517                                                 ASSERT(cur == NULL);
4518                                                 return error;
4519                                         }
4520                                         break;
4521                                 }
4522
4523                                 /*
4524                                  * Split changing sb for alen and indlen since
4525                                  * they could be coming from different places.
4526                                  */
4527                                 indlen = (xfs_extlen_t)
4528                                         xfs_bmap_worst_indlen(ip, alen);
4529                                 ASSERT(indlen > 0);
4530
4531                                 if (rt) {
4532                                         error = xfs_mod_incore_sb(mp,
4533                                                         XFS_SBS_FREXTENTS,
4534                                                         -((int64_t)extsz), 0);
4535                                 } else {
4536                                         error = xfs_icsb_modify_counters(mp,
4537                                                         XFS_SBS_FDBLOCKS,
4538                                                         -((int64_t)alen), 0);
4539                                 }
4540                                 if (!error) {
4541                                         error = xfs_icsb_modify_counters(mp,
4542                                                         XFS_SBS_FDBLOCKS,
4543                                                         -((int64_t)indlen), 0);
4544                                         if (error && rt)
4545                                                 xfs_mod_incore_sb(mp,
4546                                                         XFS_SBS_FREXTENTS,
4547                                                         (int64_t)extsz, 0);
4548                                         else if (error)
4549                                                 xfs_icsb_modify_counters(mp,
4550                                                         XFS_SBS_FDBLOCKS,
4551                                                         (int64_t)alen, 0);
4552                                 }
4553
4554                                 if (error) {
4555                                         if (XFS_IS_QUOTA_ON(mp))
4556                                                 /* unreserve the blocks now */
4557                                                 (void)
4558                                                 xfs_trans_unreserve_quota_nblks(
4559                                                         NULL, ip,
4560                                                         (long)alen, 0, rt ?
4561                                                         XFS_QMOPT_RES_RTBLKS :
4562                                                         XFS_QMOPT_RES_REGBLKS);
4563                                         break;
4564                                 }
4565
4566                                 ip->i_delayed_blks += alen;
4567                                 abno = nullstartblock(indlen);
4568                         } else {
4569                                 /*
4570                                  * If first time, allocate and fill in
4571                                  * once-only bma fields.
4572                                  */
4573                                 if (bma.ip == NULL) {
4574                                         bma.tp = tp;
4575                                         bma.ip = ip;
4576                                         bma.prevp = &prev;
4577                                         bma.gotp = &got;
4578                                         bma.total = total;
4579                                         bma.userdata = 0;
4580                                 }
4581                                 /* Indicate if this is the first user data
4582                                  * in the file, or just any user data.
4583                                  */
4584                                 if (!(flags & XFS_BMAPI_METADATA)) {
4585                                         bma.userdata = (aoff == 0) ?
4586                                                 XFS_ALLOC_INITIAL_USER_DATA :
4587                                                 XFS_ALLOC_USERDATA;
4588                                 }
4589                                 /*
4590                                  * Fill in changeable bma fields.
4591                                  */
4592                                 bma.eof = eof;
4593                                 bma.firstblock = *firstblock;
4594                                 bma.alen = alen;
4595                                 bma.off = aoff;
4596                                 bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4597                                 bma.wasdel = wasdelay;
4598                                 bma.minlen = minlen;
4599                                 bma.low = flist->xbf_low;
4600                                 bma.minleft = minleft;
4601                                 /*
4602                                  * Only want to do the alignment at the
4603                                  * eof if it is userdata and allocation length
4604                                  * is larger than a stripe unit.
4605                                  */
4606                                 if (mp->m_dalign && alen >= mp->m_dalign &&
4607                                     (!(flags & XFS_BMAPI_METADATA)) &&
4608                                     (whichfork == XFS_DATA_FORK)) {
4609                                         if ((error = xfs_bmap_isaeof(ip, aoff,
4610                                                         whichfork, &bma.aeof)))
4611                                                 goto error0;
4612                                 } else
4613                                         bma.aeof = 0;
4614                                 /*
4615                                  * Call allocator.
4616                                  */
4617                                 if ((error = xfs_bmap_alloc(&bma)))
4618                                         goto error0;
4619                                 /*
4620                                  * Copy out result fields.
4621                                  */
4622                                 abno = bma.rval;
4623                                 if ((flist->xbf_low = bma.low))
4624                                         minleft = 0;
4625                                 alen = bma.alen;
4626                                 aoff = bma.off;
4627                                 ASSERT(*firstblock == NULLFSBLOCK ||
4628                                        XFS_FSB_TO_AGNO(mp, *firstblock) ==
4629                                        XFS_FSB_TO_AGNO(mp, bma.firstblock) ||
4630                                        (flist->xbf_low &&
4631                                         XFS_FSB_TO_AGNO(mp, *firstblock) <
4632                                         XFS_FSB_TO_AGNO(mp, bma.firstblock)));
4633                                 *firstblock = bma.firstblock;
4634                                 if (cur)
4635                                         cur->bc_private.b.firstblock =
4636                                                 *firstblock;
4637                                 if (abno == NULLFSBLOCK)
4638                                         break;
4639                                 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
4640                                         cur = xfs_bmbt_init_cursor(mp, tp,
4641                                                 ip, whichfork);
4642                                         cur->bc_private.b.firstblock =
4643                                                 *firstblock;
4644                                         cur->bc_private.b.flist = flist;
4645                                 }
4646                                 /*
4647                                  * Bump the number of extents we've allocated
4648                                  * in this call.
4649                                  */
4650                                 nallocs++;
4651                         }
4652                         if (cur)
4653                                 cur->bc_private.b.flags =
4654                                         wasdelay ? XFS_BTCUR_BPRV_WASDEL : 0;
4655                         got.br_startoff = aoff;
4656                         got.br_startblock = abno;
4657                         got.br_blockcount = alen;
4658                         got.br_state = XFS_EXT_NORM;    /* assume normal */
4659                         /*
4660                          * Determine state of extent, and the filesystem.
4661                          * A wasdelay extent has been initialized, so
4662                          * shouldn't be flagged as unwritten.
4663                          */
4664                         if (wr && xfs_sb_version_hasextflgbit(&mp->m_sb)) {
4665                                 if (!wasdelay && (flags & XFS_BMAPI_PREALLOC))
4666                                         got.br_state = XFS_EXT_UNWRITTEN;
4667                         }
4668                         error = xfs_bmap_add_extent(tp, ip, &lastx, &cur, &got,
4669                                 firstblock, flist, &tmp_logflags,
4670                                 whichfork);
4671                         logflags |= tmp_logflags;
4672                         if (error)
4673                                 goto error0;
4674                         ep = xfs_iext_get_ext(ifp, lastx);
4675                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4676                         xfs_bmbt_get_all(ep, &got);
4677                         ASSERT(got.br_startoff <= aoff);
4678                         ASSERT(got.br_startoff + got.br_blockcount >=
4679                                 aoff + alen);
4680 #ifdef DEBUG
4681                         if (flags & XFS_BMAPI_DELAY) {
4682                                 ASSERT(isnullstartblock(got.br_startblock));
4683                                 ASSERT(startblockval(got.br_startblock) > 0);
4684                         }
4685                         ASSERT(got.br_state == XFS_EXT_NORM ||
4686                                got.br_state == XFS_EXT_UNWRITTEN);
4687 #endif
4688                         /*
4689                          * Fall down into the found allocated space case.
4690                          */
4691                 } else if (inhole) {
4692                         /*
4693                          * Reading in a hole.
4694                          */
4695                         mval->br_startoff = bno;
4696                         mval->br_startblock = HOLESTARTBLOCK;
4697                         mval->br_blockcount =
4698                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4699                         mval->br_state = XFS_EXT_NORM;
4700                         bno += mval->br_blockcount;
4701                         len -= mval->br_blockcount;
4702                         mval++;
4703                         n++;
4704                         continue;
4705                 }
4706                 /*
4707                  * Then deal with the allocated space we found.
4708                  */
4709                 ASSERT(ep != NULL);
4710                 if (!(flags & XFS_BMAPI_ENTIRE) &&
4711                     (got.br_startoff + got.br_blockcount > obno)) {
4712                         if (obno > bno)
4713                                 bno = obno;
4714                         ASSERT((bno >= obno) || (n == 0));
4715                         ASSERT(bno < end);
4716                         mval->br_startoff = bno;
4717                         if (isnullstartblock(got.br_startblock)) {
4718                                 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
4719                                 mval->br_startblock = DELAYSTARTBLOCK;
4720                         } else
4721                                 mval->br_startblock =
4722                                         got.br_startblock +
4723                                         (bno - got.br_startoff);
4724                         /*
4725                          * Return the minimum of what we got and what we
4726                          * asked for for the length.  We can use the len
4727                          * variable here because it is modified below
4728                          * and we could have been there before coming
4729                          * here if the first part of the allocation
4730                          * didn't overlap what was asked for.
4731                          */
4732                         mval->br_blockcount =
4733                                 XFS_FILBLKS_MIN(end - bno, got.br_blockcount -
4734                                         (bno - got.br_startoff));
4735                         mval->br_state = got.br_state;
4736                         ASSERT(mval->br_blockcount <= len);
4737                 } else {
4738                         *mval = got;
4739                         if (isnullstartblock(mval->br_startblock)) {
4740                                 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
4741                                 mval->br_startblock = DELAYSTARTBLOCK;
4742                         }
4743                 }
4744
4745                 /*
4746                  * Check if writing previously allocated but
4747                  * unwritten extents.
4748                  */
4749                 if (wr &&
4750                     ((mval->br_state == XFS_EXT_UNWRITTEN &&
4751                       ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_DELAY)) == 0)) ||
4752                      (mval->br_state == XFS_EXT_NORM &&
4753                       ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_CONVERT)) ==
4754                                 (XFS_BMAPI_PREALLOC|XFS_BMAPI_CONVERT))))) {
4755                         /*
4756                          * Modify (by adding) the state flag, if writing.
4757                          */
4758                         ASSERT(mval->br_blockcount <= len);
4759                         if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
4760                                 cur = xfs_bmbt_init_cursor(mp,
4761                                         tp, ip, whichfork);
4762                                 cur->bc_private.b.firstblock =
4763                                         *firstblock;
4764                                 cur->bc_private.b.flist = flist;
4765                         }
4766                         mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4767                                                 ? XFS_EXT_NORM
4768                                                 : XFS_EXT_UNWRITTEN;
4769                         error = xfs_bmap_add_extent(tp, ip, &lastx, &cur, mval,
4770                                 firstblock, flist, &tmp_logflags,
4771                                 whichfork);
4772                         logflags |= tmp_logflags;
4773                         if (error)
4774                                 goto error0;
4775                         ep = xfs_iext_get_ext(ifp, lastx);
4776                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4777                         xfs_bmbt_get_all(ep, &got);
4778                         /*
4779                          * We may have combined previously unwritten
4780                          * space with written space, so generate
4781                          * another request.
4782                          */
4783                         if (mval->br_blockcount < len)
4784                                 continue;
4785                 }
4786
4787                 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4788                        ((mval->br_startoff + mval->br_blockcount) <= end));
4789                 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4790                        (mval->br_blockcount <= len) ||
4791                        (mval->br_startoff < obno));
4792                 bno = mval->br_startoff + mval->br_blockcount;
4793                 len = end - bno;
4794                 if (n > 0 && mval->br_startoff == mval[-1].br_startoff) {
4795                         ASSERT(mval->br_startblock == mval[-1].br_startblock);
4796                         ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
4797                         ASSERT(mval->br_state == mval[-1].br_state);
4798                         mval[-1].br_blockcount = mval->br_blockcount;
4799                         mval[-1].br_state = mval->br_state;
4800                 } else if (n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
4801                            mval[-1].br_startblock != DELAYSTARTBLOCK &&
4802                            mval[-1].br_startblock != HOLESTARTBLOCK &&
4803                            mval->br_startblock ==
4804                            mval[-1].br_startblock + mval[-1].br_blockcount &&
4805                            ((flags & XFS_BMAPI_IGSTATE) ||
4806                                 mval[-1].br_state == mval->br_state)) {
4807                         ASSERT(mval->br_startoff ==
4808                                mval[-1].br_startoff + mval[-1].br_blockcount);
4809                         mval[-1].br_blockcount += mval->br_blockcount;
4810                 } else if (n > 0 &&
4811                            mval->br_startblock == DELAYSTARTBLOCK &&
4812                            mval[-1].br_startblock == DELAYSTARTBLOCK &&
4813                            mval->br_startoff ==
4814                            mval[-1].br_startoff + mval[-1].br_blockcount) {
4815                         mval[-1].br_blockcount += mval->br_blockcount;
4816                         mval[-1].br_state = mval->br_state;
4817                 } else if (!((n == 0) &&
4818                              ((mval->br_startoff + mval->br_blockcount) <=
4819                               obno))) {
4820                         mval++;
4821                         n++;
4822                 }
4823                 /*
4824                  * If we're done, stop now.  Stop when we've allocated
4825                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4826                  * the transaction may get too big.
4827                  */
4828                 if (bno >= end || n >= *nmap || nallocs >= *nmap)
4829                         break;
4830                 /*
4831                  * Else go on to the next record.
4832                  */
4833                 prev = got;
4834                 if (++lastx < nextents) {
4835                         ep = xfs_iext_get_ext(ifp, lastx);
4836                         xfs_bmbt_get_all(ep, &got);
4837                 } else {
4838                         eof = 1;
4839                 }
4840         }
4841         *nmap = n;
4842         /*
4843          * Transform from btree to extents, give it cur.
4844          */
4845         if (tp && XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
4846             XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
4847                 ASSERT(wr && cur);
4848                 error = xfs_bmap_btree_to_extents(tp, ip, cur,
4849                         &tmp_logflags, whichfork);
4850                 logflags |= tmp_logflags;
4851                 if (error)
4852                         goto error0;
4853         }
4854         ASSERT(ifp->if_ext_max ==
4855                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4856         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4857                XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max);
4858         error = 0;
4859 error0:
4860         /*
4861          * Log everything.  Do this after conversion, there's no point in
4862          * logging the extent records if we've converted to btree format.
4863          */
4864         if ((logflags & xfs_ilog_fext(whichfork)) &&
4865             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4866                 logflags &= ~xfs_ilog_fext(whichfork);
4867         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
4868                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4869                 logflags &= ~xfs_ilog_fbroot(whichfork);
4870         /*
4871          * Log whatever the flags say, even if error.  Otherwise we might miss
4872          * detecting a case where the data is changed, there's an error,
4873          * and it's not logged so we don't shutdown when we should.
4874          */
4875         if (logflags) {
4876                 ASSERT(tp && wr);
4877                 xfs_trans_log_inode(tp, ip, logflags);
4878         }
4879         if (cur) {
4880                 if (!error) {
4881                         ASSERT(*firstblock == NULLFSBLOCK ||
4882                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
4883                                XFS_FSB_TO_AGNO(mp,
4884                                        cur->bc_private.b.firstblock) ||
4885                                (flist->xbf_low &&
4886                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
4887                                 XFS_FSB_TO_AGNO(mp,
4888                                         cur->bc_private.b.firstblock)));
4889                         *firstblock = cur->bc_private.b.firstblock;
4890                 }
4891                 xfs_btree_del_cursor(cur,
4892                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4893         }
4894         if (!error)
4895                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4896                         orig_nmap, *nmap);
4897         return error;
4898 }
4899
4900 /*
4901  * Map file blocks to filesystem blocks, simple version.
4902  * One block (extent) only, read-only.
4903  * For flags, only the XFS_BMAPI_ATTRFORK flag is examined.
4904  * For the other flag values, the effect is as if XFS_BMAPI_METADATA
4905  * was set and all the others were clear.
4906  */
4907 int                                             /* error */
4908 xfs_bmapi_single(
4909         xfs_trans_t     *tp,            /* transaction pointer */
4910         xfs_inode_t     *ip,            /* incore inode */
4911         int             whichfork,      /* data or attr fork */
4912         xfs_fsblock_t   *fsb,           /* output: mapped block */
4913         xfs_fileoff_t   bno)            /* starting file offs. mapped */
4914 {
4915         int             eof;            /* we've hit the end of extents */
4916         int             error;          /* error return */
4917         xfs_bmbt_irec_t got;            /* current file extent record */
4918         xfs_ifork_t     *ifp;           /* inode fork pointer */
4919         xfs_extnum_t    lastx;          /* last useful extent number */
4920         xfs_bmbt_irec_t prev;           /* previous file extent record */
4921
4922         ifp = XFS_IFORK_PTR(ip, whichfork);
4923         if (unlikely(
4924             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4925             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)) {
4926                XFS_ERROR_REPORT("xfs_bmapi_single", XFS_ERRLEVEL_LOW,
4927                                 ip->i_mount);
4928                return XFS_ERROR(EFSCORRUPTED);
4929         }
4930         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
4931                 return XFS_ERROR(EIO);
4932         XFS_STATS_INC(xs_blk_mapr);
4933         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4934             (error = xfs_iread_extents(tp, ip, whichfork)))
4935                 return error;
4936         (void)xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4937                 &prev);
4938         /*
4939          * Reading past eof, act as though there's a hole
4940          * up to end.
4941          */
4942         if (eof || got.br_startoff > bno) {
4943                 *fsb = NULLFSBLOCK;
4944                 return 0;
4945         }
4946         ASSERT(!isnullstartblock(got.br_startblock));
4947         ASSERT(bno < got.br_startoff + got.br_blockcount);
4948         *fsb = got.br_startblock + (bno - got.br_startoff);
4949         return 0;
4950 }
4951
4952 /*
4953  * Unmap (remove) blocks from a file.
4954  * If nexts is nonzero then the number of extents to remove is limited to
4955  * that value.  If not all extents in the block range can be removed then
4956  * *done is set.
4957  */
4958 int                                             /* error */
4959 xfs_bunmapi(
4960         xfs_trans_t             *tp,            /* transaction pointer */
4961         struct xfs_inode        *ip,            /* incore inode */
4962         xfs_fileoff_t           bno,            /* starting offset to unmap */
4963         xfs_filblks_t           len,            /* length to unmap in file */
4964         int                     flags,          /* misc flags */
4965         xfs_extnum_t            nexts,          /* number of extents max */
4966         xfs_fsblock_t           *firstblock,    /* first allocated block
4967                                                    controls a.g. for allocs */
4968         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
4969         int                     *done)          /* set if not done yet */
4970 {
4971         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
4972         xfs_bmbt_irec_t         del;            /* extent being deleted */
4973         int                     eof;            /* is deleting at eof */
4974         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
4975         int                     error;          /* error return value */
4976         xfs_extnum_t            extno;          /* extent number in list */
4977         xfs_bmbt_irec_t         got;            /* current extent record */
4978         xfs_ifork_t             *ifp;           /* inode fork pointer */
4979         int                     isrt;           /* freeing in rt area */
4980         xfs_extnum_t            lastx;          /* last extent index used */
4981         int                     logflags;       /* transaction logging flags */
4982         xfs_extlen_t            mod;            /* rt extent offset */
4983         xfs_mount_t             *mp;            /* mount structure */
4984         xfs_extnum_t            nextents;       /* number of file extents */
4985         xfs_bmbt_irec_t         prev;           /* previous extent record */
4986         xfs_fileoff_t           start;          /* first file offset deleted */
4987         int                     tmp_logflags;   /* partial logging flags */
4988         int                     wasdel;         /* was a delayed alloc extent */
4989         int                     whichfork;      /* data or attribute fork */
4990         xfs_fsblock_t           sum;
4991
4992         trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
4993
4994         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4995                 XFS_ATTR_FORK : XFS_DATA_FORK;
4996         ifp = XFS_IFORK_PTR(ip, whichfork);
4997         if (unlikely(
4998             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4999             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5000                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5001                                  ip->i_mount);
5002                 return XFS_ERROR(EFSCORRUPTED);
5003         }
5004         mp = ip->i_mount;
5005         if (XFS_FORCED_SHUTDOWN(mp))
5006                 return XFS_ERROR(EIO);
5007
5008         ASSERT(len > 0);
5009         ASSERT(nexts >= 0);
5010         ASSERT(ifp->if_ext_max ==
5011                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5012         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5013             (error = xfs_iread_extents(tp, ip, whichfork)))
5014                 return error;
5015         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5016         if (nextents == 0) {
5017                 *done = 1;
5018                 return 0;
5019         }
5020         XFS_STATS_INC(xs_blk_unmap);
5021         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5022         start = bno;
5023         bno = start + len - 1;
5024         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5025                 &prev);
5026
5027         /*
5028          * Check to see if the given block number is past the end of the
5029          * file, back up to the last block if so...
5030          */
5031         if (eof) {
5032                 ep = xfs_iext_get_ext(ifp, --lastx);
5033                 xfs_bmbt_get_all(ep, &got);
5034                 bno = got.br_startoff + got.br_blockcount - 1;
5035         }
5036         logflags = 0;
5037         if (ifp->if_flags & XFS_IFBROOT) {
5038                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5039                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5040                 cur->bc_private.b.firstblock = *firstblock;
5041                 cur->bc_private.b.flist = flist;
5042                 cur->bc_private.b.flags = 0;
5043         } else
5044                 cur = NULL;
5045         extno = 0;
5046         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5047                (nexts == 0 || extno < nexts)) {
5048                 /*
5049                  * Is the found extent after a hole in which bno lives?
5050                  * Just back up to the previous extent, if so.
5051                  */
5052                 if (got.br_startoff > bno) {
5053                         if (--lastx < 0)
5054                                 break;
5055                         ep = xfs_iext_get_ext(ifp, lastx);
5056                         xfs_bmbt_get_all(ep, &got);
5057                 }
5058                 /*
5059                  * Is the last block of this extent before the range
5060                  * we're supposed to delete?  If so, we're done.
5061                  */
5062                 bno = XFS_FILEOFF_MIN(bno,
5063                         got.br_startoff + got.br_blockcount - 1);
5064                 if (bno < start)
5065                         break;
5066                 /*
5067                  * Then deal with the (possibly delayed) allocated space
5068                  * we found.
5069                  */
5070                 ASSERT(ep != NULL);
5071                 del = got;
5072                 wasdel = isnullstartblock(del.br_startblock);
5073                 if (got.br_startoff < start) {
5074                         del.br_startoff = start;
5075                         del.br_blockcount -= start - got.br_startoff;
5076                         if (!wasdel)
5077                                 del.br_startblock += start - got.br_startoff;
5078                 }
5079                 if (del.br_startoff + del.br_blockcount > bno + 1)
5080                         del.br_blockcount = bno + 1 - del.br_startoff;
5081                 sum = del.br_startblock + del.br_blockcount;
5082                 if (isrt &&
5083                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5084                         /*
5085                          * Realtime extent not lined up at the end.
5086                          * The extent could have been split into written
5087                          * and unwritten pieces, or we could just be
5088                          * unmapping part of it.  But we can't really
5089                          * get rid of part of a realtime extent.
5090                          */
5091                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5092                             !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5093                                 /*
5094                                  * This piece is unwritten, or we're not
5095                                  * using unwritten extents.  Skip over it.
5096                                  */
5097                                 ASSERT(bno >= mod);
5098                                 bno -= mod > del.br_blockcount ?
5099                                         del.br_blockcount : mod;
5100                                 if (bno < got.br_startoff) {
5101                                         if (--lastx >= 0)
5102                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5103                                                         ifp, lastx), &got);
5104                                 }
5105                                 continue;
5106                         }
5107                         /*
5108                          * It's written, turn it unwritten.
5109                          * This is better than zeroing it.
5110                          */
5111                         ASSERT(del.br_state == XFS_EXT_NORM);
5112                         ASSERT(xfs_trans_get_block_res(tp) > 0);
5113                         /*
5114                          * If this spans a realtime extent boundary,
5115                          * chop it back to the start of the one we end at.
5116                          */
5117                         if (del.br_blockcount > mod) {
5118                                 del.br_startoff += del.br_blockcount - mod;
5119                                 del.br_startblock += del.br_blockcount - mod;
5120                                 del.br_blockcount = mod;
5121                         }
5122                         del.br_state = XFS_EXT_UNWRITTEN;
5123                         error = xfs_bmap_add_extent(tp, ip, &lastx, &cur, &del,
5124                                 firstblock, flist, &logflags,
5125                                 XFS_DATA_FORK);
5126                         if (error)
5127                                 goto error0;
5128                         goto nodelete;
5129                 }
5130                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5131                         /*
5132                          * Realtime extent is lined up at the end but not
5133                          * at the front.  We'll get rid of full extents if
5134                          * we can.
5135                          */
5136                         mod = mp->m_sb.sb_rextsize - mod;
5137                         if (del.br_blockcount > mod) {
5138                                 del.br_blockcount -= mod;
5139                                 del.br_startoff += mod;
5140                                 del.br_startblock += mod;
5141                         } else if ((del.br_startoff == start &&
5142                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5143                                      xfs_trans_get_block_res(tp) == 0)) ||
5144                                    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5145                                 /*
5146                                  * Can't make it unwritten.  There isn't
5147                                  * a full extent here so just skip it.
5148                                  */
5149                                 ASSERT(bno >= del.br_blockcount);
5150                                 bno -= del.br_blockcount;
5151                                 if (got.br_startoff > bno) {
5152                                         if (--lastx >= 0) {
5153                                                 ep = xfs_iext_get_ext(ifp,
5154                                                                       lastx);
5155                                                 xfs_bmbt_get_all(ep, &got);
5156                                         }
5157                                 }
5158                                 continue;
5159                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5160                                 /*
5161                                  * This one is already unwritten.
5162                                  * It must have a written left neighbor.
5163                                  * Unwrite the killed part of that one and
5164                                  * try again.
5165                                  */
5166                                 ASSERT(lastx > 0);
5167                                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5168                                                 lastx - 1), &prev);
5169                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5170                                 ASSERT(!isnullstartblock(prev.br_startblock));
5171                                 ASSERT(del.br_startblock ==
5172                                        prev.br_startblock + prev.br_blockcount);
5173                                 if (prev.br_startoff < start) {
5174                                         mod = start - prev.br_startoff;
5175                                         prev.br_blockcount -= mod;
5176                                         prev.br_startblock += mod;
5177                                         prev.br_startoff = start;
5178                                 }
5179                                 prev.br_state = XFS_EXT_UNWRITTEN;
5180                                 lastx--;
5181                                 error = xfs_bmap_add_extent(tp, ip, &lastx,
5182                                                 &cur, &prev, firstblock, flist,
5183                                                 &logflags, XFS_DATA_FORK);
5184                                 if (error)
5185                                         goto error0;
5186                                 goto nodelete;
5187                         } else {
5188                                 ASSERT(del.br_state == XFS_EXT_NORM);
5189                                 del.br_state = XFS_EXT_UNWRITTEN;
5190                                 error = xfs_bmap_add_extent(tp, ip, &lastx,
5191                                                 &cur, &del, firstblock, flist,
5192                                                 &logflags, XFS_DATA_FORK);
5193                                 if (error)
5194                                         goto error0;
5195                                 goto nodelete;
5196                         }
5197                 }
5198                 if (wasdel) {
5199                         ASSERT(startblockval(del.br_startblock) > 0);
5200                         /* Update realtime/data freespace, unreserve quota */
5201                         if (isrt) {
5202                                 xfs_filblks_t rtexts;
5203
5204                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5205                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5206                                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5207                                                 (int64_t)rtexts, 0);
5208                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5209                                         ip, -((long)del.br_blockcount), 0,
5210                                         XFS_QMOPT_RES_RTBLKS);
5211                         } else {
5212                                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5213                                                 (int64_t)del.br_blockcount, 0);
5214                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5215                                         ip, -((long)del.br_blockcount), 0,
5216                                         XFS_QMOPT_RES_REGBLKS);
5217                         }
5218                         ip->i_delayed_blks -= del.br_blockcount;
5219                         if (cur)
5220                                 cur->bc_private.b.flags |=
5221                                         XFS_BTCUR_BPRV_WASDEL;
5222                 } else if (cur)
5223                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5224                 /*
5225                  * If it's the case where the directory code is running
5226                  * with no block reservation, and the deleted block is in
5227                  * the middle of its extent, and the resulting insert
5228                  * of an extent would cause transformation to btree format,
5229                  * then reject it.  The calling code will then swap
5230                  * blocks around instead.
5231                  * We have to do this now, rather than waiting for the
5232                  * conversion to btree format, since the transaction
5233                  * will be dirty.
5234                  */
5235                 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5236                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5237                     XFS_IFORK_NEXTENTS(ip, whichfork) >= ifp->if_ext_max &&
5238                     del.br_startoff > got.br_startoff &&
5239                     del.br_startoff + del.br_blockcount <
5240                     got.br_startoff + got.br_blockcount) {
5241                         error = XFS_ERROR(ENOSPC);
5242                         goto error0;
5243                 }
5244                 error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5245                                 &tmp_logflags, whichfork);
5246                 logflags |= tmp_logflags;
5247                 if (error)
5248                         goto error0;
5249                 bno = del.br_startoff - 1;
5250 nodelete:
5251                 /*
5252                  * If not done go on to the next (previous) record.
5253                  */
5254                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5255                         if (lastx >= 0) {
5256                                 ep = xfs_iext_get_ext(ifp, lastx);
5257                                 if (xfs_bmbt_get_startoff(ep) > bno) {
5258                                         if (--lastx >= 0)
5259                                                 ep = xfs_iext_get_ext(ifp,
5260                                                                       lastx);
5261                                 }
5262                                 xfs_bmbt_get_all(ep, &got);
5263                         }
5264                         extno++;
5265                 }
5266         }
5267         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5268         ASSERT(ifp->if_ext_max ==
5269                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5270         /*
5271          * Convert to a btree if necessary.
5272          */
5273         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5274             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
5275                 ASSERT(cur == NULL);
5276                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5277                         &cur, 0, &tmp_logflags, whichfork);
5278                 logflags |= tmp_logflags;
5279                 if (error)
5280                         goto error0;
5281         }
5282         /*
5283          * transform from btree to extents, give it cur
5284          */
5285         else if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5286                  XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5287                 ASSERT(cur != NULL);
5288                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5289                         whichfork);
5290                 logflags |= tmp_logflags;
5291                 if (error)
5292                         goto error0;
5293         }
5294         /*
5295          * transform from extents to local?
5296          */
5297         ASSERT(ifp->if_ext_max ==
5298                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5299         error = 0;
5300 error0:
5301         /*
5302          * Log everything.  Do this after conversion, there's no point in
5303          * logging the extent records if we've converted to btree format.
5304          */
5305         if ((logflags & xfs_ilog_fext(whichfork)) &&
5306             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5307                 logflags &= ~xfs_ilog_fext(whichfork);
5308         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5309                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5310                 logflags &= ~xfs_ilog_fbroot(whichfork);
5311         /*
5312          * Log inode even in the error case, if the transaction
5313          * is dirty we'll need to shut down the filesystem.
5314          */
5315         if (logflags)
5316                 xfs_trans_log_inode(tp, ip, logflags);
5317         if (cur) {
5318                 if (!error) {
5319                         *firstblock = cur->bc_private.b.firstblock;
5320                         cur->bc_private.b.allocated = 0;
5321                 }
5322                 xfs_btree_del_cursor(cur,
5323                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5324         }
5325         return error;
5326 }
5327
5328 /*
5329  * returns 1 for success, 0 if we failed to map the extent.
5330  */
5331 STATIC int
5332 xfs_getbmapx_fix_eof_hole(
5333         xfs_inode_t             *ip,            /* xfs incore inode pointer */
5334         struct getbmapx         *out,           /* output structure */
5335         int                     prealloced,     /* this is a file with
5336                                                  * preallocated data space */
5337         __int64_t               end,            /* last block requested */
5338         xfs_fsblock_t           startblock)
5339 {
5340         __int64_t               fixlen;
5341         xfs_mount_t             *mp;            /* file system mount point */
5342         xfs_ifork_t             *ifp;           /* inode fork pointer */
5343         xfs_extnum_t            lastx;          /* last extent pointer */
5344         xfs_fileoff_t           fileblock;
5345
5346         if (startblock == HOLESTARTBLOCK) {
5347                 mp = ip->i_mount;
5348                 out->bmv_block = -1;
5349                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, ip->i_size));
5350                 fixlen -= out->bmv_offset;
5351                 if (prealloced && out->bmv_offset + out->bmv_length == end) {
5352                         /* Came to hole at EOF. Trim it. */
5353                         if (fixlen <= 0)
5354                                 return 0;
5355                         out->bmv_length = fixlen;
5356                 }
5357         } else {
5358                 if (startblock == DELAYSTARTBLOCK)
5359                         out->bmv_block = -2;
5360                 else
5361                         out->bmv_block = xfs_fsb_to_db(ip, startblock);
5362                 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
5363                 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
5364                 if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
5365                    (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
5366                         out->bmv_oflags |= BMV_OF_LAST;
5367         }
5368
5369         return 1;
5370 }
5371
5372 /*
5373  * Get inode's extents as described in bmv, and format for output.
5374  * Calls formatter to fill the user's buffer until all extents
5375  * are mapped, until the passed-in bmv->bmv_count slots have
5376  * been filled, or until the formatter short-circuits the loop,
5377  * if it is tracking filled-in extents on its own.
5378  */
5379 int                                             /* error code */
5380 xfs_getbmap(
5381         xfs_inode_t             *ip,
5382         struct getbmapx         *bmv,           /* user bmap structure */
5383         xfs_bmap_format_t       formatter,      /* format to user */
5384         void                    *arg)           /* formatter arg */
5385 {
5386         __int64_t               bmvend;         /* last block requested */
5387         int                     error = 0;      /* return value */
5388         __int64_t               fixlen;         /* length for -1 case */
5389         int                     i;              /* extent number */
5390         int                     lock;           /* lock state */
5391         xfs_bmbt_irec_t         *map;           /* buffer for user's data */
5392         xfs_mount_t             *mp;            /* file system mount point */
5393         int                     nex;            /* # of user extents can do */
5394         int                     nexleft;        /* # of user extents left */
5395         int                     subnex;         /* # of bmapi's can do */
5396         int                     nmap;           /* number of map entries */
5397         struct getbmapx         *out;           /* output structure */
5398         int                     whichfork;      /* data or attr fork */
5399         int                     prealloced;     /* this is a file with
5400                                                  * preallocated data space */
5401         int                     iflags;         /* interface flags */
5402         int                     bmapi_flags;    /* flags for xfs_bmapi */
5403         int                     cur_ext = 0;
5404
5405         mp = ip->i_mount;
5406         iflags = bmv->bmv_iflags;
5407         whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5408
5409         if (whichfork == XFS_ATTR_FORK) {
5410                 if (XFS_IFORK_Q(ip)) {
5411                         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5412                             ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5413                             ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5414                                 return XFS_ERROR(EINVAL);
5415                 } else if (unlikely(
5416                            ip->i_d.di_aformat != 0 &&
5417                            ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5418                         XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5419                                          ip->i_mount);
5420                         return XFS_ERROR(EFSCORRUPTED);
5421                 }
5422
5423                 prealloced = 0;
5424                 fixlen = 1LL << 32;
5425         } else {
5426                 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5427                     ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5428                     ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5429                         return XFS_ERROR(EINVAL);
5430
5431                 if (xfs_get_extsz_hint(ip) ||
5432                     ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5433                         prealloced = 1;
5434                         fixlen = XFS_MAXIOFFSET(mp);
5435                 } else {
5436                         prealloced = 0;
5437                         fixlen = ip->i_size;
5438                 }
5439         }
5440
5441         if (bmv->bmv_length == -1) {
5442                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5443                 bmv->bmv_length =
5444                         max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
5445         } else if (bmv->bmv_length == 0) {
5446                 bmv->bmv_entries = 0;
5447                 return 0;
5448         } else if (bmv->bmv_length < 0) {
5449                 return XFS_ERROR(EINVAL);
5450         }
5451
5452         nex = bmv->bmv_count - 1;
5453         if (nex <= 0)
5454                 return XFS_ERROR(EINVAL);
5455         bmvend = bmv->bmv_offset + bmv->bmv_length;
5456
5457
5458         if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
5459                 return XFS_ERROR(ENOMEM);
5460         out = kmem_zalloc(bmv->bmv_count * sizeof(struct getbmapx), KM_MAYFAIL);
5461         if (!out)
5462                 return XFS_ERROR(ENOMEM);
5463
5464         xfs_ilock(ip, XFS_IOLOCK_SHARED);
5465         if (whichfork == XFS_DATA_FORK && !(iflags & BMV_IF_DELALLOC)) {
5466                 if (ip->i_delayed_blks || ip->i_size > ip->i_d.di_size) {
5467                         error = xfs_flush_pages(ip, 0, -1, 0, FI_REMAPF);
5468                         if (error)
5469                                 goto out_unlock_iolock;
5470                 }
5471                 /*
5472                  * even after flushing the inode, there can still be delalloc
5473                  * blocks on the inode beyond EOF due to speculative
5474                  * preallocation. These are not removed until the release
5475                  * function is called or the inode is inactivated. Hence we
5476                  * cannot assert here that ip->i_delayed_blks == 0.
5477                  */
5478         }
5479
5480         lock = xfs_ilock_map_shared(ip);
5481
5482         /*
5483          * Don't let nex be bigger than the number of extents
5484          * we can have assuming alternating holes and real extents.
5485          */
5486         if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5487                 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5488
5489         bmapi_flags = xfs_bmapi_aflag(whichfork);
5490         if (!(iflags & BMV_IF_PREALLOC))
5491                 bmapi_flags |= XFS_BMAPI_IGSTATE;
5492
5493         /*
5494          * Allocate enough space to handle "subnex" maps at a time.
5495          */
5496         error = ENOMEM;
5497         subnex = 16;
5498         map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
5499         if (!map)
5500                 goto out_unlock_ilock;
5501
5502         bmv->bmv_entries = 0;
5503
5504         if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
5505             (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
5506                 error = 0;
5507                 goto out_free_map;
5508         }
5509
5510         nexleft = nex;
5511
5512         do {
5513                 nmap = (nexleft > subnex) ? subnex : nexleft;
5514                 error = xfs_bmapi(NULL, ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5515                                   XFS_BB_TO_FSB(mp, bmv->bmv_length),
5516                                   bmapi_flags, NULL, 0, map, &nmap,
5517                                   NULL);
5518                 if (error)
5519                         goto out_free_map;
5520                 ASSERT(nmap <= subnex);
5521
5522                 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5523                         out[cur_ext].bmv_oflags = 0;
5524                         if (map[i].br_state == XFS_EXT_UNWRITTEN)
5525                                 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
5526                         else if (map[i].br_startblock == DELAYSTARTBLOCK)
5527                                 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
5528                         out[cur_ext].bmv_offset =
5529                                 XFS_FSB_TO_BB(mp, map[i].br_startoff);
5530                         out[cur_ext].bmv_length =
5531                                 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5532                         out[cur_ext].bmv_unused1 = 0;
5533                         out[cur_ext].bmv_unused2 = 0;
5534                         ASSERT(((iflags & BMV_IF_DELALLOC) != 0) ||
5535                               (map[i].br_startblock != DELAYSTARTBLOCK));
5536                         if (map[i].br_startblock == HOLESTARTBLOCK &&
5537                             whichfork == XFS_ATTR_FORK) {
5538                                 /* came to the end of attribute fork */
5539                                 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
5540                                 goto out_free_map;
5541                         }
5542
5543                         if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
5544                                         prealloced, bmvend,
5545                                         map[i].br_startblock))
5546                                 goto out_free_map;
5547
5548                         bmv->bmv_offset =
5549                                 out[cur_ext].bmv_offset +
5550                                 out[cur_ext].bmv_length;
5551                         bmv->bmv_length =
5552                                 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
5553
5554                         /*
5555                          * In case we don't want to return the hole,
5556                          * don't increase cur_ext so that we can reuse
5557                          * it in the next loop.
5558                          */
5559                         if ((iflags & BMV_IF_NO_HOLES) &&
5560                             map[i].br_startblock == HOLESTARTBLOCK) {
5561                                 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
5562                                 continue;
5563                         }
5564
5565                         nexleft--;
5566                         bmv->bmv_entries++;
5567                         cur_ext++;
5568                 }
5569         } while (nmap && nexleft && bmv->bmv_length);
5570
5571  out_free_map:
5572         kmem_free(map);
5573  out_unlock_ilock:
5574         xfs_iunlock_map_shared(ip, lock);
5575  out_unlock_iolock:
5576         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5577
5578         for (i = 0; i < cur_ext; i++) {
5579                 int full = 0;   /* user array is full */
5580
5581                 /* format results & advance arg */
5582                 error = formatter(&arg, &out[i], &full);
5583                 if (error || full)
5584                         break;
5585         }
5586
5587         kmem_free(out);
5588         return error;
5589 }
5590
5591 /*
5592  * Check the last inode extent to determine whether this allocation will result
5593  * in blocks being allocated at the end of the file. When we allocate new data
5594  * blocks at the end of the file which do not start at the previous data block,
5595  * we will try to align the new blocks at stripe unit boundaries.
5596  */
5597 STATIC int                              /* error */
5598 xfs_bmap_isaeof(
5599         xfs_inode_t     *ip,            /* incore inode pointer */
5600         xfs_fileoff_t   off,            /* file offset in fsblocks */
5601         int             whichfork,      /* data or attribute fork */
5602         char            *aeof)          /* return value */
5603 {
5604         int             error;          /* error return value */
5605         xfs_ifork_t     *ifp;           /* inode fork pointer */
5606         xfs_bmbt_rec_host_t *lastrec;   /* extent record pointer */
5607         xfs_extnum_t    nextents;       /* number of file extents */
5608         xfs_bmbt_irec_t s;              /* expanded extent record */
5609
5610         ASSERT(whichfork == XFS_DATA_FORK);
5611         ifp = XFS_IFORK_PTR(ip, whichfork);
5612         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5613             (error = xfs_iread_extents(NULL, ip, whichfork)))
5614                 return error;
5615         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5616         if (nextents == 0) {
5617                 *aeof = 1;
5618                 return 0;
5619         }
5620         /*
5621          * Go to the last extent
5622          */
5623         lastrec = xfs_iext_get_ext(ifp, nextents - 1);
5624         xfs_bmbt_get_all(lastrec, &s);
5625         /*
5626          * Check we are allocating in the last extent (for delayed allocations)
5627          * or past the last extent for non-delayed allocations.
5628          */
5629         *aeof = (off >= s.br_startoff &&
5630                  off < s.br_startoff + s.br_blockcount &&
5631                  isnullstartblock(s.br_startblock)) ||
5632                 off >= s.br_startoff + s.br_blockcount;
5633         return 0;
5634 }
5635
5636 /*
5637  * Check if the endoff is outside the last extent. If so the caller will grow
5638  * the allocation to a stripe unit boundary.
5639  */
5640 int                                     /* error */
5641 xfs_bmap_eof(
5642         xfs_inode_t     *ip,            /* incore inode pointer */
5643         xfs_fileoff_t   endoff,         /* file offset in fsblocks */
5644         int             whichfork,      /* data or attribute fork */
5645         int             *eof)           /* result value */
5646 {
5647         xfs_fsblock_t   blockcount;     /* extent block count */
5648         int             error;          /* error return value */
5649         xfs_ifork_t     *ifp;           /* inode fork pointer */
5650         xfs_bmbt_rec_host_t *lastrec;   /* extent record pointer */
5651         xfs_extnum_t    nextents;       /* number of file extents */
5652         xfs_fileoff_t   startoff;       /* extent starting file offset */
5653
5654         ASSERT(whichfork == XFS_DATA_FORK);
5655         ifp = XFS_IFORK_PTR(ip, whichfork);
5656         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5657             (error = xfs_iread_extents(NULL, ip, whichfork)))
5658                 return error;
5659         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5660         if (nextents == 0) {
5661                 *eof = 1;
5662                 return 0;
5663         }
5664         /*
5665          * Go to the last extent
5666          */
5667         lastrec = xfs_iext_get_ext(ifp, nextents - 1);
5668         startoff = xfs_bmbt_get_startoff(lastrec);
5669         blockcount = xfs_bmbt_get_blockcount(lastrec);
5670         *eof = endoff >= startoff + blockcount;
5671         return 0;
5672 }
5673
5674 #ifdef DEBUG
5675 STATIC struct xfs_buf *
5676 xfs_bmap_get_bp(
5677         struct xfs_btree_cur    *cur,
5678         xfs_fsblock_t           bno)
5679 {
5680         struct xfs_log_item_desc *lidp;
5681         int                     i;
5682
5683         if (!cur)
5684                 return NULL;
5685
5686         for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
5687                 if (!cur->bc_bufs[i])
5688                         break;
5689                 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
5690                         return cur->bc_bufs[i];
5691         }
5692
5693         /* Chase down all the log items to see if the bp is there */
5694         list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
5695                 struct xfs_buf_log_item *bip;
5696                 bip = (struct xfs_buf_log_item *)lidp->lid_item;
5697                 if (bip->bli_item.li_type == XFS_LI_BUF &&
5698                     XFS_BUF_ADDR(bip->bli_buf) == bno)
5699                         return bip->bli_buf;
5700         }
5701
5702         return NULL;
5703 }
5704
5705 STATIC void
5706 xfs_check_block(
5707         struct xfs_btree_block  *block,
5708         xfs_mount_t             *mp,
5709         int                     root,
5710         short                   sz)
5711 {
5712         int                     i, j, dmxr;
5713         __be64                  *pp, *thispa;   /* pointer to block address */
5714         xfs_bmbt_key_t          *prevp, *keyp;
5715
5716         ASSERT(be16_to_cpu(block->bb_level) > 0);
5717
5718         prevp = NULL;
5719         for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
5720                 dmxr = mp->m_bmap_dmxr[0];
5721                 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
5722
5723                 if (prevp) {
5724                         ASSERT(be64_to_cpu(prevp->br_startoff) <
5725                                be64_to_cpu(keyp->br_startoff));
5726                 }
5727                 prevp = keyp;
5728
5729                 /*
5730                  * Compare the block numbers to see if there are dups.
5731                  */
5732                 if (root)
5733                         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
5734                 else
5735                         pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
5736
5737                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
5738                         if (root)
5739                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
5740                         else
5741                                 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
5742                         if (*thispa == *pp) {
5743                                 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
5744                                         __func__, j, i,
5745                                         (unsigned long long)be64_to_cpu(*thispa));
5746                                 panic("%s: ptrs are equal in node\n",
5747                                         __func__);
5748                         }
5749                 }
5750         }
5751 }
5752
5753 /*
5754  * Check that the extents for the inode ip are in the right order in all
5755  * btree leaves.
5756  */
5757
5758 STATIC void
5759 xfs_bmap_check_leaf_extents(
5760         xfs_btree_cur_t         *cur,   /* btree cursor or null */
5761         xfs_inode_t             *ip,            /* incore inode pointer */
5762         int                     whichfork)      /* data or attr fork */
5763 {
5764         struct xfs_btree_block  *block; /* current btree block */
5765         xfs_fsblock_t           bno;    /* block # of "block" */
5766         xfs_buf_t               *bp;    /* buffer for "block" */
5767         int                     error;  /* error return value */
5768         xfs_extnum_t            i=0, j; /* index into the extents list */
5769         xfs_ifork_t             *ifp;   /* fork structure */
5770         int                     level;  /* btree level, for checking */
5771         xfs_mount_t             *mp;    /* file system mount structure */
5772         __be64                  *pp;    /* pointer to block address */
5773         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
5774         xfs_bmbt_rec_t          last = {0, 0}; /* last extent in prev block */
5775         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
5776         int                     bp_release = 0;
5777
5778         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
5779                 return;
5780         }
5781
5782         bno = NULLFSBLOCK;
5783         mp = ip->i_mount;
5784         ifp = XFS_IFORK_PTR(ip, whichfork);
5785         block = ifp->if_broot;
5786         /*
5787          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5788          */
5789         level = be16_to_cpu(block->bb_level);
5790         ASSERT(level > 0);
5791         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
5792         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5793         bno = be64_to_cpu(*pp);
5794
5795         ASSERT(bno != NULLDFSBNO);
5796         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5797         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5798
5799         /*
5800          * Go down the tree until leaf level is reached, following the first
5801          * pointer (leftmost) at each level.
5802          */
5803         while (level-- > 0) {
5804                 /* See if buf is in cur first */
5805                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5806                 if (bp) {
5807                         bp_release = 0;
5808                 } else {
5809                         bp_release = 1;
5810                 }
5811                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5812                                 XFS_BMAP_BTREE_REF)))
5813                         goto error_norelse;
5814                 block = XFS_BUF_TO_BLOCK(bp);
5815                 XFS_WANT_CORRUPTED_GOTO(
5816                         xfs_bmap_sanity_check(mp, bp, level),
5817                         error0);
5818                 if (level == 0)
5819                         break;
5820
5821                 /*
5822                  * Check this block for basic sanity (increasing keys and
5823                  * no duplicate blocks).
5824                  */
5825
5826                 xfs_check_block(block, mp, 0, 0);
5827                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
5828                 bno = be64_to_cpu(*pp);
5829                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
5830                 if (bp_release) {
5831                         bp_release = 0;
5832                         xfs_trans_brelse(NULL, bp);
5833                 }
5834         }
5835
5836         /*
5837          * Here with bp and block set to the leftmost leaf node in the tree.
5838          */
5839         i = 0;
5840
5841         /*
5842          * Loop over all leaf nodes checking that all extents are in the right order.
5843          */
5844         for (;;) {
5845                 xfs_fsblock_t   nextbno;
5846                 xfs_extnum_t    num_recs;
5847
5848
5849                 num_recs = xfs_btree_get_numrecs(block);
5850
5851                 /*
5852                  * Read-ahead the next leaf block, if any.
5853                  */
5854
5855                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
5856
5857                 /*
5858                  * Check all the extents to make sure they are OK.
5859                  * If we had a previous block, the last entry should
5860                  * conform with the first entry in this one.
5861                  */
5862
5863                 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
5864                 if (i) {
5865                         ASSERT(xfs_bmbt_disk_get_startoff(&last) +
5866                                xfs_bmbt_disk_get_blockcount(&last) <=
5867                                xfs_bmbt_disk_get_startoff(ep));
5868                 }
5869                 for (j = 1; j < num_recs; j++) {
5870                         nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
5871                         ASSERT(xfs_bmbt_disk_get_startoff(ep) +
5872                                xfs_bmbt_disk_get_blockcount(ep) <=
5873                                xfs_bmbt_disk_get_startoff(nextp));
5874                         ep = nextp;
5875                 }
5876
5877                 last = *ep;
5878                 i += num_recs;
5879                 if (bp_release) {
5880                         bp_release = 0;
5881                         xfs_trans_brelse(NULL, bp);
5882                 }
5883                 bno = nextbno;
5884                 /*
5885                  * If we've reached the end, stop.
5886                  */
5887                 if (bno == NULLFSBLOCK)
5888                         break;
5889
5890                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5891                 if (bp) {
5892                         bp_release = 0;
5893                 } else {
5894                         bp_release = 1;
5895                 }
5896                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5897                                 XFS_BMAP_BTREE_REF)))
5898                         goto error_norelse;
5899                 block = XFS_BUF_TO_BLOCK(bp);
5900         }
5901         if (bp_release) {
5902                 bp_release = 0;
5903                 xfs_trans_brelse(NULL, bp);
5904         }
5905         return;
5906
5907 error0:
5908         xfs_warn(mp, "%s: at error0", __func__);
5909         if (bp_release)
5910                 xfs_trans_brelse(NULL, bp);
5911 error_norelse:
5912         xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
5913                 __func__, i);
5914         panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
5915         return;
5916 }
5917 #endif
5918
5919 /*
5920  * Count fsblocks of the given fork.
5921  */
5922 int                                             /* error */
5923 xfs_bmap_count_blocks(
5924         xfs_trans_t             *tp,            /* transaction pointer */
5925         xfs_inode_t             *ip,            /* incore inode */
5926         int                     whichfork,      /* data or attr fork */
5927         int                     *count)         /* out: count of blocks */
5928 {
5929         struct xfs_btree_block  *block; /* current btree block */
5930         xfs_fsblock_t           bno;    /* block # of "block" */
5931         xfs_ifork_t             *ifp;   /* fork structure */
5932         int                     level;  /* btree level, for checking */
5933         xfs_mount_t             *mp;    /* file system mount structure */
5934         __be64                  *pp;    /* pointer to block address */
5935
5936         bno = NULLFSBLOCK;
5937         mp = ip->i_mount;
5938         ifp = XFS_IFORK_PTR(ip, whichfork);
5939         if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
5940                 xfs_bmap_count_leaves(ifp, 0,
5941                         ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
5942                         count);
5943                 return 0;
5944         }
5945
5946         /*
5947          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5948          */
5949         block = ifp->if_broot;
5950         level = be16_to_cpu(block->bb_level);
5951         ASSERT(level > 0);
5952         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5953         bno = be64_to_cpu(*pp);
5954         ASSERT(bno != NULLDFSBNO);
5955         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5956         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5957
5958         if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
5959                 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
5960                                  mp);
5961                 return XFS_ERROR(EFSCORRUPTED);
5962         }
5963
5964         return 0;
5965 }
5966
5967 /*
5968  * Recursively walks each level of a btree
5969  * to count total fsblocks is use.
5970  */
5971 STATIC int                                     /* error */
5972 xfs_bmap_count_tree(
5973         xfs_mount_t     *mp,            /* file system mount point */
5974         xfs_trans_t     *tp,            /* transaction pointer */
5975         xfs_ifork_t     *ifp,           /* inode fork pointer */
5976         xfs_fsblock_t   blockno,        /* file system block number */
5977         int             levelin,        /* level in btree */
5978         int             *count)         /* Count of blocks */
5979 {
5980         int                     error;
5981         xfs_buf_t               *bp, *nbp;
5982         int                     level = levelin;
5983         __be64                  *pp;
5984         xfs_fsblock_t           bno = blockno;
5985         xfs_fsblock_t           nextbno;
5986         struct xfs_btree_block  *block, *nextblock;
5987         int                     numrecs;
5988
5989         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
5990                 return error;
5991         *count += 1;
5992         block = XFS_BUF_TO_BLOCK(bp);
5993
5994         if (--level) {
5995                 /* Not at node above leaves, count this level of nodes */
5996                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
5997                 while (nextbno != NULLFSBLOCK) {
5998                         if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
5999                                 0, &nbp, XFS_BMAP_BTREE_REF)))
6000                                 return error;
6001                         *count += 1;
6002                         nextblock = XFS_BUF_TO_BLOCK(nbp);
6003                         nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
6004                         xfs_trans_brelse(tp, nbp);
6005                 }
6006
6007                 /* Dive to the next level */
6008                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
6009                 bno = be64_to_cpu(*pp);
6010                 if (unlikely((error =
6011                      xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
6012                         xfs_trans_brelse(tp, bp);
6013                         XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6014                                          XFS_ERRLEVEL_LOW, mp);
6015                         return XFS_ERROR(EFSCORRUPTED);
6016                 }
6017                 xfs_trans_brelse(tp, bp);
6018         } else {
6019                 /* count all level 1 nodes and their leaves */
6020                 for (;;) {
6021                         nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
6022                         numrecs = be16_to_cpu(block->bb_numrecs);
6023                         xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
6024                         xfs_trans_brelse(tp, bp);
6025                         if (nextbno == NULLFSBLOCK)
6026                                 break;
6027                         bno = nextbno;
6028                         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6029                                 XFS_BMAP_BTREE_REF)))
6030                                 return error;
6031                         *count += 1;
6032                         block = XFS_BUF_TO_BLOCK(bp);
6033                 }
6034         }
6035         return 0;
6036 }
6037
6038 /*
6039  * Count leaf blocks given a range of extent records.
6040  */
6041 STATIC void
6042 xfs_bmap_count_leaves(
6043         xfs_ifork_t             *ifp,
6044         xfs_extnum_t            idx,
6045         int                     numrecs,
6046         int                     *count)
6047 {
6048         int             b;
6049
6050         for (b = 0; b < numrecs; b++) {
6051                 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
6052                 *count += xfs_bmbt_get_blockcount(frp);
6053         }
6054 }
6055
6056 /*
6057  * Count leaf blocks given a range of extent records originally
6058  * in btree format.
6059  */
6060 STATIC void
6061 xfs_bmap_disk_count_leaves(
6062         struct xfs_mount        *mp,
6063         struct xfs_btree_block  *block,
6064         int                     numrecs,
6065         int                     *count)
6066 {
6067         int             b;
6068         xfs_bmbt_rec_t  *frp;
6069
6070         for (b = 1; b <= numrecs; b++) {
6071                 frp = XFS_BMBT_REC_ADDR(mp, block, b);
6072                 *count += xfs_bmbt_disk_get_blockcount(frp);
6073         }
6074 }
6075
6076 /*
6077  * dead simple method of punching delalyed allocation blocks from a range in
6078  * the inode. Walks a block at a time so will be slow, but is only executed in
6079  * rare error cases so the overhead is not critical. This will alays punch out
6080  * both the start and end blocks, even if the ranges only partially overlap
6081  * them, so it is up to the caller to ensure that partial blocks are not
6082  * passed in.
6083  */
6084 int
6085 xfs_bmap_punch_delalloc_range(
6086         struct xfs_inode        *ip,
6087         xfs_fileoff_t           start_fsb,
6088         xfs_fileoff_t           length)
6089 {
6090         xfs_fileoff_t           remaining = length;
6091         int                     error = 0;
6092
6093         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
6094
6095         do {
6096                 int             done;
6097                 xfs_bmbt_irec_t imap;
6098                 int             nimaps = 1;
6099                 xfs_fsblock_t   firstblock;
6100                 xfs_bmap_free_t flist;
6101
6102                 /*
6103                  * Map the range first and check that it is a delalloc extent
6104                  * before trying to unmap the range. Otherwise we will be
6105                  * trying to remove a real extent (which requires a
6106                  * transaction) or a hole, which is probably a bad idea...
6107                  */
6108                 error = xfs_bmapi(NULL, ip, start_fsb, 1,
6109                                 XFS_BMAPI_ENTIRE,  NULL, 0, &imap,
6110                                 &nimaps, NULL);
6111
6112                 if (error) {
6113                         /* something screwed, just bail */
6114                         if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
6115                                 xfs_alert(ip->i_mount,
6116                         "Failed delalloc mapping lookup ino %lld fsb %lld.",
6117                                                 ip->i_ino, start_fsb);
6118                         }
6119                         break;
6120                 }
6121                 if (!nimaps) {
6122                         /* nothing there */
6123                         goto next_block;
6124                 }
6125                 if (imap.br_startblock != DELAYSTARTBLOCK) {
6126                         /* been converted, ignore */
6127                         goto next_block;
6128                 }
6129                 WARN_ON(imap.br_blockcount == 0);
6130
6131                 /*
6132                  * Note: while we initialise the firstblock/flist pair, they
6133                  * should never be used because blocks should never be
6134                  * allocated or freed for a delalloc extent and hence we need
6135                  * don't cancel or finish them after the xfs_bunmapi() call.
6136                  */
6137                 xfs_bmap_init(&flist, &firstblock);
6138                 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
6139                                         &flist, &done);
6140                 if (error)
6141                         break;
6142
6143                 ASSERT(!flist.xbf_count && !flist.xbf_first);
6144 next_block:
6145                 start_fsb++;
6146                 remaining--;
6147         } while(remaining > 0);
6148
6149         return error;
6150 }