Merge branch 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ieee13...
[linux-2.6] / 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_attr_sf.h"
34 #include "xfs_dinode.h"
35 #include "xfs_inode.h"
36 #include "xfs_btree.h"
37 #include "xfs_dmapi.h"
38 #include "xfs_mount.h"
39 #include "xfs_ialloc.h"
40 #include "xfs_itable.h"
41 #include "xfs_dir2_data.h"
42 #include "xfs_dir2_leaf.h"
43 #include "xfs_dir2_block.h"
44 #include "xfs_inode_item.h"
45 #include "xfs_extfree_item.h"
46 #include "xfs_alloc.h"
47 #include "xfs_bmap.h"
48 #include "xfs_rtalloc.h"
49 #include "xfs_error.h"
50 #include "xfs_attr_leaf.h"
51 #include "xfs_rw.h"
52 #include "xfs_quota.h"
53 #include "xfs_trans_space.h"
54 #include "xfs_buf_item.h"
55
56
57 #ifdef DEBUG
58 STATIC void
59 xfs_bmap_check_leaf_extents(xfs_btree_cur_t *cur, xfs_inode_t *ip, int whichfork);
60 #endif
61
62 kmem_zone_t             *xfs_bmap_free_item_zone;
63
64 /*
65  * Prototypes for internal bmap routines.
66  */
67
68
69 /*
70  * Called from xfs_bmap_add_attrfork to handle extents format files.
71  */
72 STATIC int                                      /* error */
73 xfs_bmap_add_attrfork_extents(
74         xfs_trans_t             *tp,            /* transaction pointer */
75         xfs_inode_t             *ip,            /* incore inode pointer */
76         xfs_fsblock_t           *firstblock,    /* first block allocated */
77         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
78         int                     *flags);        /* inode logging flags */
79
80 /*
81  * Called from xfs_bmap_add_attrfork to handle local format files.
82  */
83 STATIC int                                      /* error */
84 xfs_bmap_add_attrfork_local(
85         xfs_trans_t             *tp,            /* transaction pointer */
86         xfs_inode_t             *ip,            /* incore inode pointer */
87         xfs_fsblock_t           *firstblock,    /* first block allocated */
88         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
89         int                     *flags);        /* inode logging flags */
90
91 /*
92  * Called by xfs_bmapi to update file extent records and the btree
93  * after allocating space (or doing a delayed allocation).
94  */
95 STATIC int                              /* error */
96 xfs_bmap_add_extent(
97         xfs_inode_t             *ip,    /* incore inode pointer */
98         xfs_extnum_t            idx,    /* extent number to update/insert */
99         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
100         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
101         xfs_fsblock_t           *first, /* pointer to firstblock variable */
102         xfs_bmap_free_t         *flist, /* list of extents to be freed */
103         int                     *logflagsp, /* inode logging flags */
104         xfs_extdelta_t          *delta, /* Change made to incore extents */
105         int                     whichfork, /* data or attr fork */
106         int                     rsvd);  /* OK to allocate reserved blocks */
107
108 /*
109  * Called by xfs_bmap_add_extent to handle cases converting a delayed
110  * allocation to a real allocation.
111  */
112 STATIC int                              /* error */
113 xfs_bmap_add_extent_delay_real(
114         xfs_inode_t             *ip,    /* incore inode pointer */
115         xfs_extnum_t            idx,    /* extent number to update/insert */
116         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
117         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
118         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
119         xfs_fsblock_t           *first, /* pointer to firstblock variable */
120         xfs_bmap_free_t         *flist, /* list of extents to be freed */
121         int                     *logflagsp, /* inode logging flags */
122         xfs_extdelta_t          *delta, /* Change made to incore extents */
123         int                     rsvd);  /* OK to allocate reserved blocks */
124
125 /*
126  * Called by xfs_bmap_add_extent to handle cases converting a hole
127  * to a delayed allocation.
128  */
129 STATIC int                              /* error */
130 xfs_bmap_add_extent_hole_delay(
131         xfs_inode_t             *ip,    /* incore inode pointer */
132         xfs_extnum_t            idx,    /* extent number to update/insert */
133         xfs_btree_cur_t         *cur,   /* if null, not a btree */
134         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
135         int                     *logflagsp,/* inode logging flags */
136         xfs_extdelta_t          *delta, /* Change made to incore extents */
137         int                     rsvd);  /* OK to allocate reserved blocks */
138
139 /*
140  * Called by xfs_bmap_add_extent to handle cases converting a hole
141  * to a real allocation.
142  */
143 STATIC int                              /* error */
144 xfs_bmap_add_extent_hole_real(
145         xfs_inode_t             *ip,    /* incore inode pointer */
146         xfs_extnum_t            idx,    /* extent number to update/insert */
147         xfs_btree_cur_t         *cur,   /* if null, not a btree */
148         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
149         int                     *logflagsp, /* inode logging flags */
150         xfs_extdelta_t          *delta, /* Change made to incore extents */
151         int                     whichfork); /* data or attr fork */
152
153 /*
154  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
155  * allocation to a real allocation or vice versa.
156  */
157 STATIC int                              /* error */
158 xfs_bmap_add_extent_unwritten_real(
159         xfs_inode_t             *ip,    /* incore inode pointer */
160         xfs_extnum_t            idx,    /* extent number to update/insert */
161         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
162         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
163         int                     *logflagsp, /* inode logging flags */
164         xfs_extdelta_t          *delta); /* Change made to incore extents */
165
166 /*
167  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
168  * It figures out where to ask the underlying allocator to put the new extent.
169  */
170 STATIC int                              /* error */
171 xfs_bmap_alloc(
172         xfs_bmalloca_t          *ap);   /* bmap alloc argument struct */
173
174 /*
175  * Transform a btree format file with only one leaf node, where the
176  * extents list will fit in the inode, into an extents format file.
177  * Since the file extents are already in-core, all we have to do is
178  * give up the space for the btree root and pitch the leaf block.
179  */
180 STATIC int                              /* error */
181 xfs_bmap_btree_to_extents(
182         xfs_trans_t             *tp,    /* transaction pointer */
183         xfs_inode_t             *ip,    /* incore inode pointer */
184         xfs_btree_cur_t         *cur,   /* btree cursor */
185         int                     *logflagsp, /* inode logging flags */
186         int                     whichfork); /* data or attr fork */
187
188 #ifdef DEBUG
189 /*
190  * Check that the extents list for the inode ip is in the right order.
191  */
192 STATIC void
193 xfs_bmap_check_extents(
194         xfs_inode_t             *ip,            /* incore inode pointer */
195         int                     whichfork);     /* data or attr fork */
196 #endif
197
198 /*
199  * Called by xfs_bmapi to update file extent records and the btree
200  * after removing space (or undoing a delayed allocation).
201  */
202 STATIC int                              /* error */
203 xfs_bmap_del_extent(
204         xfs_inode_t             *ip,    /* incore inode pointer */
205         xfs_trans_t             *tp,    /* current trans pointer */
206         xfs_extnum_t            idx,    /* extent number to update/insert */
207         xfs_bmap_free_t         *flist, /* list of extents to be freed */
208         xfs_btree_cur_t         *cur,   /* if null, not a btree */
209         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
210         int                     *logflagsp,/* inode logging flags */
211         xfs_extdelta_t          *delta, /* Change made to incore extents */
212         int                     whichfork, /* data or attr fork */
213         int                     rsvd);   /* OK to allocate reserved blocks */
214
215 /*
216  * Remove the entry "free" from the free item list.  Prev points to the
217  * previous entry, unless "free" is the head of the list.
218  */
219 STATIC void
220 xfs_bmap_del_free(
221         xfs_bmap_free_t         *flist, /* free item list header */
222         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
223         xfs_bmap_free_item_t    *free); /* list item to be freed */
224
225 /*
226  * Convert an extents-format file into a btree-format file.
227  * The new file will have a root block (in the inode) and a single child block.
228  */
229 STATIC int                                      /* error */
230 xfs_bmap_extents_to_btree(
231         xfs_trans_t             *tp,            /* transaction pointer */
232         xfs_inode_t             *ip,            /* incore inode pointer */
233         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
234         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
235         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
236         int                     wasdel,         /* converting a delayed alloc */
237         int                     *logflagsp,     /* inode logging flags */
238         int                     whichfork);     /* data or attr fork */
239
240 /*
241  * Convert a local file to an extents file.
242  * This code is sort of bogus, since the file data needs to get
243  * logged so it won't be lost.  The bmap-level manipulations are ok, though.
244  */
245 STATIC int                              /* error */
246 xfs_bmap_local_to_extents(
247         xfs_trans_t     *tp,            /* transaction pointer */
248         xfs_inode_t     *ip,            /* incore inode pointer */
249         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
250         xfs_extlen_t    total,          /* total blocks needed by transaction */
251         int             *logflagsp,     /* inode logging flags */
252         int             whichfork);     /* data or attr fork */
253
254 /*
255  * Search the extents list for the inode, for the extent containing bno.
256  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
257  * *eofp will be set, and *prevp will contain the last entry (null if none).
258  * Else, *lastxp will be set to the index of the found
259  * entry; *gotp will contain the entry.
260  */
261 STATIC xfs_bmbt_rec_t *                 /* pointer to found extent entry */
262 xfs_bmap_search_extents(
263         xfs_inode_t     *ip,            /* incore inode pointer */
264         xfs_fileoff_t   bno,            /* block number searched for */
265         int             whichfork,      /* data or attr fork */
266         int             *eofp,          /* out: end of file found */
267         xfs_extnum_t    *lastxp,        /* out: last extent index */
268         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
269         xfs_bmbt_irec_t *prevp);        /* out: previous extent entry found */
270
271 /*
272  * Check the last inode extent to determine whether this allocation will result
273  * in blocks being allocated at the end of the file. When we allocate new data
274  * blocks at the end of the file which do not start at the previous data block,
275  * we will try to align the new blocks at stripe unit boundaries.
276  */
277 STATIC int                              /* error */
278 xfs_bmap_isaeof(
279         xfs_inode_t     *ip,            /* incore inode pointer */
280         xfs_fileoff_t   off,            /* file offset in fsblocks */
281         int             whichfork,      /* data or attribute fork */
282         char            *aeof);         /* return value */
283
284 #ifdef XFS_BMAP_TRACE
285 /*
286  * Add a bmap trace buffer entry.  Base routine for the others.
287  */
288 STATIC void
289 xfs_bmap_trace_addentry(
290         int             opcode,         /* operation */
291         char            *fname,         /* function name */
292         char            *desc,          /* operation description */
293         xfs_inode_t     *ip,            /* incore inode pointer */
294         xfs_extnum_t    idx,            /* index of entry(ies) */
295         xfs_extnum_t    cnt,            /* count of entries, 1 or 2 */
296         xfs_bmbt_rec_t  *r1,            /* first record */
297         xfs_bmbt_rec_t  *r2,            /* second record or null */
298         int             whichfork);     /* data or attr fork */
299
300 /*
301  * Add bmap trace entry prior to a call to xfs_iext_remove.
302  */
303 STATIC void
304 xfs_bmap_trace_delete(
305         char            *fname,         /* function name */
306         char            *desc,          /* operation description */
307         xfs_inode_t     *ip,            /* incore inode pointer */
308         xfs_extnum_t    idx,            /* index of entry(entries) deleted */
309         xfs_extnum_t    cnt,            /* count of entries deleted, 1 or 2 */
310         int             whichfork);     /* data or attr fork */
311
312 /*
313  * Add bmap trace entry prior to a call to xfs_iext_insert, or
314  * reading in the extents list from the disk (in the btree).
315  */
316 STATIC void
317 xfs_bmap_trace_insert(
318         char            *fname,         /* function name */
319         char            *desc,          /* operation description */
320         xfs_inode_t     *ip,            /* incore inode pointer */
321         xfs_extnum_t    idx,            /* index of entry(entries) inserted */
322         xfs_extnum_t    cnt,            /* count of entries inserted, 1 or 2 */
323         xfs_bmbt_irec_t *r1,            /* inserted record 1 */
324         xfs_bmbt_irec_t *r2,            /* inserted record 2 or null */
325         int             whichfork);     /* data or attr fork */
326
327 /*
328  * Add bmap trace entry after updating an extent record in place.
329  */
330 STATIC void
331 xfs_bmap_trace_post_update(
332         char            *fname,         /* function name */
333         char            *desc,          /* operation description */
334         xfs_inode_t     *ip,            /* incore inode pointer */
335         xfs_extnum_t    idx,            /* index of entry updated */
336         int             whichfork);     /* data or attr fork */
337
338 /*
339  * Add bmap trace entry prior to updating an extent record in place.
340  */
341 STATIC void
342 xfs_bmap_trace_pre_update(
343         char            *fname,         /* function name */
344         char            *desc,          /* operation description */
345         xfs_inode_t     *ip,            /* incore inode pointer */
346         xfs_extnum_t    idx,            /* index of entry to be updated */
347         int             whichfork);     /* data or attr fork */
348
349 #else
350 #define xfs_bmap_trace_delete(f,d,ip,i,c,w)
351 #define xfs_bmap_trace_insert(f,d,ip,i,c,r1,r2,w)
352 #define xfs_bmap_trace_post_update(f,d,ip,i,w)
353 #define xfs_bmap_trace_pre_update(f,d,ip,i,w)
354 #endif  /* XFS_BMAP_TRACE */
355
356 /*
357  * Compute the worst-case number of indirect blocks that will be used
358  * for ip's delayed extent of length "len".
359  */
360 STATIC xfs_filblks_t
361 xfs_bmap_worst_indlen(
362         xfs_inode_t             *ip,    /* incore inode pointer */
363         xfs_filblks_t           len);   /* delayed extent length */
364
365 #ifdef DEBUG
366 /*
367  * Perform various validation checks on the values being returned
368  * from xfs_bmapi().
369  */
370 STATIC void
371 xfs_bmap_validate_ret(
372         xfs_fileoff_t           bno,
373         xfs_filblks_t           len,
374         int                     flags,
375         xfs_bmbt_irec_t         *mval,
376         int                     nmap,
377         int                     ret_nmap);
378 #else
379 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
380 #endif /* DEBUG */
381
382 #if defined(XFS_RW_TRACE)
383 STATIC void
384 xfs_bunmap_trace(
385         xfs_inode_t             *ip,
386         xfs_fileoff_t           bno,
387         xfs_filblks_t           len,
388         int                     flags,
389         inst_t                  *ra);
390 #else
391 #define xfs_bunmap_trace(ip, bno, len, flags, ra)
392 #endif  /* XFS_RW_TRACE */
393
394 STATIC int
395 xfs_bmap_count_tree(
396         xfs_mount_t     *mp,
397         xfs_trans_t     *tp,
398         xfs_ifork_t     *ifp,
399         xfs_fsblock_t   blockno,
400         int             levelin,
401         int             *count);
402
403 STATIC int
404 xfs_bmap_count_leaves(
405         xfs_ifork_t             *ifp,
406         xfs_extnum_t            idx,
407         int                     numrecs,
408         int                     *count);
409
410 STATIC int
411 xfs_bmap_disk_count_leaves(
412         xfs_ifork_t             *ifp,
413         xfs_mount_t             *mp,
414         xfs_extnum_t            idx,
415         xfs_bmbt_block_t        *block,
416         int                     numrecs,
417         int                     *count);
418
419 /*
420  * Bmap internal routines.
421  */
422
423 /*
424  * Called from xfs_bmap_add_attrfork to handle btree format files.
425  */
426 STATIC int                                      /* error */
427 xfs_bmap_add_attrfork_btree(
428         xfs_trans_t             *tp,            /* transaction pointer */
429         xfs_inode_t             *ip,            /* incore inode pointer */
430         xfs_fsblock_t           *firstblock,    /* first block allocated */
431         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
432         int                     *flags)         /* inode logging flags */
433 {
434         xfs_btree_cur_t         *cur;           /* btree cursor */
435         int                     error;          /* error return value */
436         xfs_mount_t             *mp;            /* file system mount struct */
437         int                     stat;           /* newroot status */
438
439         mp = ip->i_mount;
440         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
441                 *flags |= XFS_ILOG_DBROOT;
442         else {
443                 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
444                         XFS_DATA_FORK);
445                 cur->bc_private.b.flist = flist;
446                 cur->bc_private.b.firstblock = *firstblock;
447                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
448                         goto error0;
449                 ASSERT(stat == 1);      /* must be at least one entry */
450                 if ((error = xfs_bmbt_newroot(cur, flags, &stat)))
451                         goto error0;
452                 if (stat == 0) {
453                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
454                         return XFS_ERROR(ENOSPC);
455                 }
456                 *firstblock = cur->bc_private.b.firstblock;
457                 cur->bc_private.b.allocated = 0;
458                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
459         }
460         return 0;
461 error0:
462         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
463         return error;
464 }
465
466 /*
467  * Called from xfs_bmap_add_attrfork to handle extents format files.
468  */
469 STATIC int                                      /* error */
470 xfs_bmap_add_attrfork_extents(
471         xfs_trans_t             *tp,            /* transaction pointer */
472         xfs_inode_t             *ip,            /* incore inode pointer */
473         xfs_fsblock_t           *firstblock,    /* first block allocated */
474         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
475         int                     *flags)         /* inode logging flags */
476 {
477         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
478         int                     error;          /* error return value */
479
480         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
481                 return 0;
482         cur = NULL;
483         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
484                 flags, XFS_DATA_FORK);
485         if (cur) {
486                 cur->bc_private.b.allocated = 0;
487                 xfs_btree_del_cursor(cur,
488                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
489         }
490         return error;
491 }
492
493 /*
494  * Called from xfs_bmap_add_attrfork to handle local format files.
495  */
496 STATIC int                                      /* error */
497 xfs_bmap_add_attrfork_local(
498         xfs_trans_t             *tp,            /* transaction pointer */
499         xfs_inode_t             *ip,            /* incore inode pointer */
500         xfs_fsblock_t           *firstblock,    /* first block allocated */
501         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
502         int                     *flags)         /* inode logging flags */
503 {
504         xfs_da_args_t           dargs;          /* args for dir/attr code */
505         int                     error;          /* error return value */
506         xfs_mount_t             *mp;            /* mount structure pointer */
507
508         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
509                 return 0;
510         if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
511                 mp = ip->i_mount;
512                 memset(&dargs, 0, sizeof(dargs));
513                 dargs.dp = ip;
514                 dargs.firstblock = firstblock;
515                 dargs.flist = flist;
516                 dargs.total = mp->m_dirblkfsbs;
517                 dargs.whichfork = XFS_DATA_FORK;
518                 dargs.trans = tp;
519                 error = xfs_dir2_sf_to_block(&dargs);
520         } else
521                 error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
522                         XFS_DATA_FORK);
523         return error;
524 }
525
526 /*
527  * Called by xfs_bmapi to update file extent records and the btree
528  * after allocating space (or doing a delayed allocation).
529  */
530 STATIC int                              /* error */
531 xfs_bmap_add_extent(
532         xfs_inode_t             *ip,    /* incore inode pointer */
533         xfs_extnum_t            idx,    /* extent number to update/insert */
534         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
535         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
536         xfs_fsblock_t           *first, /* pointer to firstblock variable */
537         xfs_bmap_free_t         *flist, /* list of extents to be freed */
538         int                     *logflagsp, /* inode logging flags */
539         xfs_extdelta_t          *delta, /* Change made to incore extents */
540         int                     whichfork, /* data or attr fork */
541         int                     rsvd)   /* OK to use reserved data blocks */
542 {
543         xfs_btree_cur_t         *cur;   /* btree cursor or null */
544         xfs_filblks_t           da_new; /* new count del alloc blocks used */
545         xfs_filblks_t           da_old; /* old count del alloc blocks used */
546         int                     error;  /* error return value */
547 #ifdef XFS_BMAP_TRACE
548         static char             fname[] = "xfs_bmap_add_extent";
549 #endif
550         xfs_ifork_t             *ifp;   /* inode fork ptr */
551         int                     logflags; /* returned value */
552         xfs_extnum_t            nextents; /* number of extents in file now */
553
554         XFS_STATS_INC(xs_add_exlist);
555         cur = *curp;
556         ifp = XFS_IFORK_PTR(ip, whichfork);
557         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
558         ASSERT(idx <= nextents);
559         da_old = da_new = 0;
560         error = 0;
561         /*
562          * This is the first extent added to a new/empty file.
563          * Special case this one, so other routines get to assume there are
564          * already extents in the list.
565          */
566         if (nextents == 0) {
567                 xfs_bmap_trace_insert(fname, "insert empty", ip, 0, 1, new,
568                         NULL, whichfork);
569                 xfs_iext_insert(ifp, 0, 1, new);
570                 ASSERT(cur == NULL);
571                 ifp->if_lastex = 0;
572                 if (!ISNULLSTARTBLOCK(new->br_startblock)) {
573                         XFS_IFORK_NEXT_SET(ip, whichfork, 1);
574                         logflags = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
575                 } else
576                         logflags = 0;
577                 /* DELTA: single new extent */
578                 if (delta) {
579                         if (delta->xed_startoff > new->br_startoff)
580                                 delta->xed_startoff = new->br_startoff;
581                         if (delta->xed_blockcount <
582                                         new->br_startoff + new->br_blockcount)
583                                 delta->xed_blockcount = new->br_startoff +
584                                                 new->br_blockcount;
585                 }
586         }
587         /*
588          * Any kind of new delayed allocation goes here.
589          */
590         else if (ISNULLSTARTBLOCK(new->br_startblock)) {
591                 if (cur)
592                         ASSERT((cur->bc_private.b.flags &
593                                 XFS_BTCUR_BPRV_WASDEL) == 0);
594                 if ((error = xfs_bmap_add_extent_hole_delay(ip, idx, cur, new,
595                                 &logflags, delta, rsvd)))
596                         goto done;
597         }
598         /*
599          * Real allocation off the end of the file.
600          */
601         else if (idx == nextents) {
602                 if (cur)
603                         ASSERT((cur->bc_private.b.flags &
604                                 XFS_BTCUR_BPRV_WASDEL) == 0);
605                 if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur, new,
606                                 &logflags, delta, whichfork)))
607                         goto done;
608         } else {
609                 xfs_bmbt_irec_t prev;   /* old extent at offset idx */
610
611                 /*
612                  * Get the record referred to by idx.
613                  */
614                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx), &prev);
615                 /*
616                  * If it's a real allocation record, and the new allocation ends
617                  * after the start of the referred to record, then we're filling
618                  * in a delayed or unwritten allocation with a real one, or
619                  * converting real back to unwritten.
620                  */
621                 if (!ISNULLSTARTBLOCK(new->br_startblock) &&
622                     new->br_startoff + new->br_blockcount > prev.br_startoff) {
623                         if (prev.br_state != XFS_EXT_UNWRITTEN &&
624                             ISNULLSTARTBLOCK(prev.br_startblock)) {
625                                 da_old = STARTBLOCKVAL(prev.br_startblock);
626                                 if (cur)
627                                         ASSERT(cur->bc_private.b.flags &
628                                                 XFS_BTCUR_BPRV_WASDEL);
629                                 if ((error = xfs_bmap_add_extent_delay_real(ip,
630                                         idx, &cur, new, &da_new, first, flist,
631                                         &logflags, delta, rsvd)))
632                                         goto done;
633                         } else if (new->br_state == XFS_EXT_NORM) {
634                                 ASSERT(new->br_state == XFS_EXT_NORM);
635                                 if ((error = xfs_bmap_add_extent_unwritten_real(
636                                         ip, idx, &cur, new, &logflags, delta)))
637                                         goto done;
638                         } else {
639                                 ASSERT(new->br_state == XFS_EXT_UNWRITTEN);
640                                 if ((error = xfs_bmap_add_extent_unwritten_real(
641                                         ip, idx, &cur, new, &logflags, delta)))
642                                         goto done;
643                         }
644                         ASSERT(*curp == cur || *curp == NULL);
645                 }
646                 /*
647                  * Otherwise we're filling in a hole with an allocation.
648                  */
649                 else {
650                         if (cur)
651                                 ASSERT((cur->bc_private.b.flags &
652                                         XFS_BTCUR_BPRV_WASDEL) == 0);
653                         if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur,
654                                         new, &logflags, delta, whichfork)))
655                                 goto done;
656                 }
657         }
658
659         ASSERT(*curp == cur || *curp == NULL);
660         /*
661          * Convert to a btree if necessary.
662          */
663         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
664             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
665                 int     tmp_logflags;   /* partial log flag return val */
666
667                 ASSERT(cur == NULL);
668                 error = xfs_bmap_extents_to_btree(ip->i_transp, ip, first,
669                         flist, &cur, da_old > 0, &tmp_logflags, whichfork);
670                 logflags |= tmp_logflags;
671                 if (error)
672                         goto done;
673         }
674         /*
675          * Adjust for changes in reserved delayed indirect blocks.
676          * Nothing to do for disk quotas here.
677          */
678         if (da_old || da_new) {
679                 xfs_filblks_t   nblks;
680
681                 nblks = da_new;
682                 if (cur)
683                         nblks += cur->bc_private.b.allocated;
684                 ASSERT(nblks <= da_old);
685                 if (nblks < da_old)
686                         xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
687                                 (int)(da_old - nblks), rsvd);
688         }
689         /*
690          * Clear out the allocated field, done with it now in any case.
691          */
692         if (cur) {
693                 cur->bc_private.b.allocated = 0;
694                 *curp = cur;
695         }
696 done:
697 #ifdef DEBUG
698         if (!error)
699                 xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
700 #endif
701         *logflagsp = logflags;
702         return error;
703 }
704
705 /*
706  * Called by xfs_bmap_add_extent to handle cases converting a delayed
707  * allocation to a real allocation.
708  */
709 STATIC int                              /* error */
710 xfs_bmap_add_extent_delay_real(
711         xfs_inode_t             *ip,    /* incore inode pointer */
712         xfs_extnum_t            idx,    /* extent number to update/insert */
713         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
714         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
715         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
716         xfs_fsblock_t           *first, /* pointer to firstblock variable */
717         xfs_bmap_free_t         *flist, /* list of extents to be freed */
718         int                     *logflagsp, /* inode logging flags */
719         xfs_extdelta_t          *delta, /* Change made to incore extents */
720         int                     rsvd)   /* OK to use reserved data block allocation */
721 {
722         xfs_btree_cur_t         *cur;   /* btree cursor */
723         int                     diff;   /* temp value */
724         xfs_bmbt_rec_t          *ep;    /* extent entry for idx */
725         int                     error;  /* error return value */
726 #ifdef XFS_BMAP_TRACE
727         static char             fname[] = "xfs_bmap_add_extent_delay_real";
728 #endif
729         int                     i;      /* temp state */
730         xfs_ifork_t             *ifp;   /* inode fork pointer */
731         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
732         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
733                                         /* left is 0, right is 1, prev is 2 */
734         int                     rval=0; /* return value (logging flags) */
735         int                     state = 0;/* state bits, accessed thru macros */
736         xfs_filblks_t           temp=0; /* value for dnew calculations */
737         xfs_filblks_t           temp2=0;/* value for dnew calculations */
738         int                     tmp_rval;       /* partial logging flags */
739         enum {                          /* bit number definitions for state */
740                 LEFT_CONTIG,    RIGHT_CONTIG,
741                 LEFT_FILLING,   RIGHT_FILLING,
742                 LEFT_DELAY,     RIGHT_DELAY,
743                 LEFT_VALID,     RIGHT_VALID
744         };
745
746 #define LEFT            r[0]
747 #define RIGHT           r[1]
748 #define PREV            r[2]
749 #define MASK(b)         (1 << (b))
750 #define MASK2(a,b)      (MASK(a) | MASK(b))
751 #define MASK3(a,b,c)    (MASK2(a,b) | MASK(c))
752 #define MASK4(a,b,c,d)  (MASK3(a,b,c) | MASK(d))
753 #define STATE_SET(b,v)  ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
754 #define STATE_TEST(b)   (state & MASK(b))
755 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
756                                        ((state &= ~MASK(b)), 0))
757 #define SWITCH_STATE            \
758         (state & MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG))
759
760         /*
761          * Set up a bunch of variables to make the tests simpler.
762          */
763         cur = *curp;
764         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
765         ep = xfs_iext_get_ext(ifp, idx);
766         xfs_bmbt_get_all(ep, &PREV);
767         new_endoff = new->br_startoff + new->br_blockcount;
768         ASSERT(PREV.br_startoff <= new->br_startoff);
769         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
770         /*
771          * Set flags determining what part of the previous delayed allocation
772          * extent is being replaced by a real allocation.
773          */
774         STATE_SET(LEFT_FILLING, PREV.br_startoff == new->br_startoff);
775         STATE_SET(RIGHT_FILLING,
776                 PREV.br_startoff + PREV.br_blockcount == new_endoff);
777         /*
778          * Check and set flags if this segment has a left neighbor.
779          * Don't set contiguous if the combined extent would be too large.
780          */
781         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
782                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &LEFT);
783                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(LEFT.br_startblock));
784         }
785         STATE_SET(LEFT_CONTIG,
786                 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
787                 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
788                 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
789                 LEFT.br_state == new->br_state &&
790                 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN);
791         /*
792          * Check and set flags if this segment has a right neighbor.
793          * Don't set contiguous if the combined extent would be too large.
794          * Also check for all-three-contiguous being too large.
795          */
796         if (STATE_SET_TEST(RIGHT_VALID,
797                         idx <
798                         ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1)) {
799                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx + 1), &RIGHT);
800                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(RIGHT.br_startblock));
801         }
802         STATE_SET(RIGHT_CONTIG,
803                 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
804                 new_endoff == RIGHT.br_startoff &&
805                 new->br_startblock + new->br_blockcount ==
806                     RIGHT.br_startblock &&
807                 new->br_state == RIGHT.br_state &&
808                 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
809                 ((state & MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING)) !=
810                   MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING) ||
811                  LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
812                      <= MAXEXTLEN));
813         error = 0;
814         /*
815          * Switch out based on the FILLING and CONTIG state bits.
816          */
817         switch (SWITCH_STATE) {
818
819         case MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
820                 /*
821                  * Filling in all of a previously delayed allocation extent.
822                  * The left and right neighbors are both contiguous with new.
823                  */
824                 xfs_bmap_trace_pre_update(fname, "LF|RF|LC|RC", ip, idx - 1,
825                         XFS_DATA_FORK);
826                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
827                         LEFT.br_blockcount + PREV.br_blockcount +
828                         RIGHT.br_blockcount);
829                 xfs_bmap_trace_post_update(fname, "LF|RF|LC|RC", ip, idx - 1,
830                         XFS_DATA_FORK);
831                 xfs_bmap_trace_delete(fname, "LF|RF|LC|RC", ip, idx, 2,
832                         XFS_DATA_FORK);
833                 xfs_iext_remove(ifp, idx, 2);
834                 ip->i_df.if_lastex = idx - 1;
835                 ip->i_d.di_nextents--;
836                 if (cur == NULL)
837                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
838                 else {
839                         rval = XFS_ILOG_CORE;
840                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
841                                         RIGHT.br_startblock,
842                                         RIGHT.br_blockcount, &i)))
843                                 goto done;
844                         ASSERT(i == 1);
845                         if ((error = xfs_bmbt_delete(cur, &i)))
846                                 goto done;
847                         ASSERT(i == 1);
848                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
849                                 goto done;
850                         ASSERT(i == 1);
851                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
852                                         LEFT.br_startblock,
853                                         LEFT.br_blockcount +
854                                         PREV.br_blockcount +
855                                         RIGHT.br_blockcount, LEFT.br_state)))
856                                 goto done;
857                 }
858                 *dnew = 0;
859                 /* DELTA: Three in-core extents are replaced by one. */
860                 temp = LEFT.br_startoff;
861                 temp2 = LEFT.br_blockcount +
862                         PREV.br_blockcount +
863                         RIGHT.br_blockcount;
864                 break;
865
866         case MASK3(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG):
867                 /*
868                  * Filling in all of a previously delayed allocation extent.
869                  * The left neighbor is contiguous, the right is not.
870                  */
871                 xfs_bmap_trace_pre_update(fname, "LF|RF|LC", ip, idx - 1,
872                         XFS_DATA_FORK);
873                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
874                         LEFT.br_blockcount + PREV.br_blockcount);
875                 xfs_bmap_trace_post_update(fname, "LF|RF|LC", ip, idx - 1,
876                         XFS_DATA_FORK);
877                 ip->i_df.if_lastex = idx - 1;
878                 xfs_bmap_trace_delete(fname, "LF|RF|LC", ip, idx, 1,
879                         XFS_DATA_FORK);
880                 xfs_iext_remove(ifp, idx, 1);
881                 if (cur == NULL)
882                         rval = XFS_ILOG_DEXT;
883                 else {
884                         rval = 0;
885                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
886                                         LEFT.br_startblock, LEFT.br_blockcount,
887                                         &i)))
888                                 goto done;
889                         ASSERT(i == 1);
890                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
891                                         LEFT.br_startblock,
892                                         LEFT.br_blockcount +
893                                         PREV.br_blockcount, LEFT.br_state)))
894                                 goto done;
895                 }
896                 *dnew = 0;
897                 /* DELTA: Two in-core extents are replaced by one. */
898                 temp = LEFT.br_startoff;
899                 temp2 = LEFT.br_blockcount +
900                         PREV.br_blockcount;
901                 break;
902
903         case MASK3(LEFT_FILLING, RIGHT_FILLING, RIGHT_CONTIG):
904                 /*
905                  * Filling in all of a previously delayed allocation extent.
906                  * The right neighbor is contiguous, the left is not.
907                  */
908                 xfs_bmap_trace_pre_update(fname, "LF|RF|RC", ip, idx,
909                         XFS_DATA_FORK);
910                 xfs_bmbt_set_startblock(ep, new->br_startblock);
911                 xfs_bmbt_set_blockcount(ep,
912                         PREV.br_blockcount + RIGHT.br_blockcount);
913                 xfs_bmap_trace_post_update(fname, "LF|RF|RC", ip, idx,
914                         XFS_DATA_FORK);
915                 ip->i_df.if_lastex = idx;
916                 xfs_bmap_trace_delete(fname, "LF|RF|RC", ip, idx + 1, 1,
917                         XFS_DATA_FORK);
918                 xfs_iext_remove(ifp, idx + 1, 1);
919                 if (cur == NULL)
920                         rval = XFS_ILOG_DEXT;
921                 else {
922                         rval = 0;
923                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
924                                         RIGHT.br_startblock,
925                                         RIGHT.br_blockcount, &i)))
926                                 goto done;
927                         ASSERT(i == 1);
928                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
929                                         new->br_startblock,
930                                         PREV.br_blockcount +
931                                         RIGHT.br_blockcount, PREV.br_state)))
932                                 goto done;
933                 }
934                 *dnew = 0;
935                 /* DELTA: Two in-core extents are replaced by one. */
936                 temp = PREV.br_startoff;
937                 temp2 = PREV.br_blockcount +
938                         RIGHT.br_blockcount;
939                 break;
940
941         case MASK2(LEFT_FILLING, RIGHT_FILLING):
942                 /*
943                  * Filling in all of a previously delayed allocation extent.
944                  * Neither the left nor right neighbors are contiguous with
945                  * the new one.
946                  */
947                 xfs_bmap_trace_pre_update(fname, "LF|RF", ip, idx,
948                         XFS_DATA_FORK);
949                 xfs_bmbt_set_startblock(ep, new->br_startblock);
950                 xfs_bmap_trace_post_update(fname, "LF|RF", ip, idx,
951                         XFS_DATA_FORK);
952                 ip->i_df.if_lastex = idx;
953                 ip->i_d.di_nextents++;
954                 if (cur == NULL)
955                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
956                 else {
957                         rval = XFS_ILOG_CORE;
958                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
959                                         new->br_startblock, new->br_blockcount,
960                                         &i)))
961                                 goto done;
962                         ASSERT(i == 0);
963                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
964                         if ((error = xfs_bmbt_insert(cur, &i)))
965                                 goto done;
966                         ASSERT(i == 1);
967                 }
968                 *dnew = 0;
969                 /* DELTA: The in-core extent described by new changed type. */
970                 temp = new->br_startoff;
971                 temp2 = new->br_blockcount;
972                 break;
973
974         case MASK2(LEFT_FILLING, LEFT_CONTIG):
975                 /*
976                  * Filling in the first part of a previous delayed allocation.
977                  * The left neighbor is contiguous.
978                  */
979                 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx - 1,
980                         XFS_DATA_FORK);
981                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
982                         LEFT.br_blockcount + new->br_blockcount);
983                 xfs_bmbt_set_startoff(ep,
984                         PREV.br_startoff + new->br_blockcount);
985                 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx - 1,
986                         XFS_DATA_FORK);
987                 temp = PREV.br_blockcount - new->br_blockcount;
988                 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx,
989                         XFS_DATA_FORK);
990                 xfs_bmbt_set_blockcount(ep, temp);
991                 ip->i_df.if_lastex = idx - 1;
992                 if (cur == NULL)
993                         rval = XFS_ILOG_DEXT;
994                 else {
995                         rval = 0;
996                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
997                                         LEFT.br_startblock, LEFT.br_blockcount,
998                                         &i)))
999                                 goto done;
1000                         ASSERT(i == 1);
1001                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1002                                         LEFT.br_startblock,
1003                                         LEFT.br_blockcount +
1004                                         new->br_blockcount,
1005                                         LEFT.br_state)))
1006                                 goto done;
1007                 }
1008                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1009                         STARTBLOCKVAL(PREV.br_startblock));
1010                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1011                 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx,
1012                         XFS_DATA_FORK);
1013                 *dnew = temp;
1014                 /* DELTA: The boundary between two in-core extents moved. */
1015                 temp = LEFT.br_startoff;
1016                 temp2 = LEFT.br_blockcount +
1017                         PREV.br_blockcount;
1018                 break;
1019
1020         case MASK(LEFT_FILLING):
1021                 /*
1022                  * Filling in the first part of a previous delayed allocation.
1023                  * The left neighbor is not contiguous.
1024                  */
1025                 xfs_bmap_trace_pre_update(fname, "LF", ip, idx, XFS_DATA_FORK);
1026                 xfs_bmbt_set_startoff(ep, new_endoff);
1027                 temp = PREV.br_blockcount - new->br_blockcount;
1028                 xfs_bmbt_set_blockcount(ep, temp);
1029                 xfs_bmap_trace_insert(fname, "LF", ip, idx, 1, new, NULL,
1030                         XFS_DATA_FORK);
1031                 xfs_iext_insert(ifp, idx, 1, new);
1032                 ip->i_df.if_lastex = idx;
1033                 ip->i_d.di_nextents++;
1034                 if (cur == NULL)
1035                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1036                 else {
1037                         rval = XFS_ILOG_CORE;
1038                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1039                                         new->br_startblock, new->br_blockcount,
1040                                         &i)))
1041                                 goto done;
1042                         ASSERT(i == 0);
1043                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1044                         if ((error = xfs_bmbt_insert(cur, &i)))
1045                                 goto done;
1046                         ASSERT(i == 1);
1047                 }
1048                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1049                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1050                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1051                                         first, flist, &cur, 1, &tmp_rval,
1052                                         XFS_DATA_FORK);
1053                         rval |= tmp_rval;
1054                         if (error)
1055                                 goto done;
1056                 }
1057                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1058                         STARTBLOCKVAL(PREV.br_startblock) -
1059                         (cur ? cur->bc_private.b.allocated : 0));
1060                 ep = xfs_iext_get_ext(ifp, idx + 1);
1061                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1062                 xfs_bmap_trace_post_update(fname, "LF", ip, idx + 1,
1063                         XFS_DATA_FORK);
1064                 *dnew = temp;
1065                 /* DELTA: One in-core extent is split in two. */
1066                 temp = PREV.br_startoff;
1067                 temp2 = PREV.br_blockcount;
1068                 break;
1069
1070         case MASK2(RIGHT_FILLING, RIGHT_CONTIG):
1071                 /*
1072                  * Filling in the last part of a previous delayed allocation.
1073                  * The right neighbor is contiguous with the new allocation.
1074                  */
1075                 temp = PREV.br_blockcount - new->br_blockcount;
1076                 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx,
1077                         XFS_DATA_FORK);
1078                 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx + 1,
1079                         XFS_DATA_FORK);
1080                 xfs_bmbt_set_blockcount(ep, temp);
1081                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, idx + 1),
1082                         new->br_startoff, new->br_startblock,
1083                         new->br_blockcount + RIGHT.br_blockcount,
1084                         RIGHT.br_state);
1085                 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx + 1,
1086                         XFS_DATA_FORK);
1087                 ip->i_df.if_lastex = idx + 1;
1088                 if (cur == NULL)
1089                         rval = XFS_ILOG_DEXT;
1090                 else {
1091                         rval = 0;
1092                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1093                                         RIGHT.br_startblock,
1094                                         RIGHT.br_blockcount, &i)))
1095                                 goto done;
1096                         ASSERT(i == 1);
1097                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1098                                         new->br_startblock,
1099                                         new->br_blockcount +
1100                                         RIGHT.br_blockcount,
1101                                         RIGHT.br_state)))
1102                                 goto done;
1103                 }
1104                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1105                         STARTBLOCKVAL(PREV.br_startblock));
1106                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1107                 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx,
1108                         XFS_DATA_FORK);
1109                 *dnew = temp;
1110                 /* DELTA: The boundary between two in-core extents moved. */
1111                 temp = PREV.br_startoff;
1112                 temp2 = PREV.br_blockcount +
1113                         RIGHT.br_blockcount;
1114                 break;
1115
1116         case MASK(RIGHT_FILLING):
1117                 /*
1118                  * Filling in the last part of a previous delayed allocation.
1119                  * The right neighbor is not contiguous.
1120                  */
1121                 temp = PREV.br_blockcount - new->br_blockcount;
1122                 xfs_bmap_trace_pre_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1123                 xfs_bmbt_set_blockcount(ep, temp);
1124                 xfs_bmap_trace_insert(fname, "RF", ip, idx + 1, 1,
1125                         new, NULL, XFS_DATA_FORK);
1126                 xfs_iext_insert(ifp, idx + 1, 1, new);
1127                 ip->i_df.if_lastex = idx + 1;
1128                 ip->i_d.di_nextents++;
1129                 if (cur == NULL)
1130                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1131                 else {
1132                         rval = XFS_ILOG_CORE;
1133                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1134                                         new->br_startblock, new->br_blockcount,
1135                                         &i)))
1136                                 goto done;
1137                         ASSERT(i == 0);
1138                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1139                         if ((error = xfs_bmbt_insert(cur, &i)))
1140                                 goto done;
1141                         ASSERT(i == 1);
1142                 }
1143                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1144                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1145                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1146                                 first, flist, &cur, 1, &tmp_rval,
1147                                 XFS_DATA_FORK);
1148                         rval |= tmp_rval;
1149                         if (error)
1150                                 goto done;
1151                 }
1152                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1153                         STARTBLOCKVAL(PREV.br_startblock) -
1154                         (cur ? cur->bc_private.b.allocated : 0));
1155                 ep = xfs_iext_get_ext(ifp, idx);
1156                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1157                 xfs_bmap_trace_post_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1158                 *dnew = temp;
1159                 /* DELTA: One in-core extent is split in two. */
1160                 temp = PREV.br_startoff;
1161                 temp2 = PREV.br_blockcount;
1162                 break;
1163
1164         case 0:
1165                 /*
1166                  * Filling in the middle part of a previous delayed allocation.
1167                  * Contiguity is impossible here.
1168                  * This case is avoided almost all the time.
1169                  */
1170                 temp = new->br_startoff - PREV.br_startoff;
1171                 xfs_bmap_trace_pre_update(fname, "0", ip, idx, XFS_DATA_FORK);
1172                 xfs_bmbt_set_blockcount(ep, temp);
1173                 r[0] = *new;
1174                 r[1].br_startoff = new_endoff;
1175                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
1176                 r[1].br_blockcount = temp2;
1177                 xfs_bmap_trace_insert(fname, "0", ip, idx + 1, 2, &r[0], &r[1],
1178                         XFS_DATA_FORK);
1179                 xfs_iext_insert(ifp, idx + 1, 2, &r[0]);
1180                 ip->i_df.if_lastex = idx + 1;
1181                 ip->i_d.di_nextents++;
1182                 if (cur == NULL)
1183                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1184                 else {
1185                         rval = XFS_ILOG_CORE;
1186                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1187                                         new->br_startblock, new->br_blockcount,
1188                                         &i)))
1189                                 goto done;
1190                         ASSERT(i == 0);
1191                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1192                         if ((error = xfs_bmbt_insert(cur, &i)))
1193                                 goto done;
1194                         ASSERT(i == 1);
1195                 }
1196                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1197                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1198                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1199                                         first, flist, &cur, 1, &tmp_rval,
1200                                         XFS_DATA_FORK);
1201                         rval |= tmp_rval;
1202                         if (error)
1203                                 goto done;
1204                 }
1205                 temp = xfs_bmap_worst_indlen(ip, temp);
1206                 temp2 = xfs_bmap_worst_indlen(ip, temp2);
1207                 diff = (int)(temp + temp2 - STARTBLOCKVAL(PREV.br_startblock) -
1208                         (cur ? cur->bc_private.b.allocated : 0));
1209                 if (diff > 0 &&
1210                     xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS, -diff, rsvd)) {
1211                         /*
1212                          * Ick gross gag me with a spoon.
1213                          */
1214                         ASSERT(0);      /* want to see if this ever happens! */
1215                         while (diff > 0) {
1216                                 if (temp) {
1217                                         temp--;
1218                                         diff--;
1219                                         if (!diff ||
1220                                             !xfs_mod_incore_sb(ip->i_mount,
1221                                                     XFS_SBS_FDBLOCKS, -diff, rsvd))
1222                                                 break;
1223                                 }
1224                                 if (temp2) {
1225                                         temp2--;
1226                                         diff--;
1227                                         if (!diff ||
1228                                             !xfs_mod_incore_sb(ip->i_mount,
1229                                                     XFS_SBS_FDBLOCKS, -diff, rsvd))
1230                                                 break;
1231                                 }
1232                         }
1233                 }
1234                 ep = xfs_iext_get_ext(ifp, idx);
1235                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1236                 xfs_bmap_trace_post_update(fname, "0", ip, idx, XFS_DATA_FORK);
1237                 xfs_bmap_trace_pre_update(fname, "0", ip, idx + 2,
1238                         XFS_DATA_FORK);
1239                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx + 2),
1240                         NULLSTARTBLOCK((int)temp2));
1241                 xfs_bmap_trace_post_update(fname, "0", ip, idx + 2,
1242                         XFS_DATA_FORK);
1243                 *dnew = temp + temp2;
1244                 /* DELTA: One in-core extent is split in three. */
1245                 temp = PREV.br_startoff;
1246                 temp2 = PREV.br_blockcount;
1247                 break;
1248
1249         case MASK3(LEFT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1250         case MASK3(RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1251         case MASK2(LEFT_FILLING, RIGHT_CONTIG):
1252         case MASK2(RIGHT_FILLING, LEFT_CONTIG):
1253         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1254         case MASK(LEFT_CONTIG):
1255         case MASK(RIGHT_CONTIG):
1256                 /*
1257                  * These cases are all impossible.
1258                  */
1259                 ASSERT(0);
1260         }
1261         *curp = cur;
1262         if (delta) {
1263                 temp2 += temp;
1264                 if (delta->xed_startoff > temp)
1265                         delta->xed_startoff = temp;
1266                 if (delta->xed_blockcount < temp2)
1267                         delta->xed_blockcount = temp2;
1268         }
1269 done:
1270         *logflagsp = rval;
1271         return error;
1272 #undef  LEFT
1273 #undef  RIGHT
1274 #undef  PREV
1275 #undef  MASK
1276 #undef  MASK2
1277 #undef  MASK3
1278 #undef  MASK4
1279 #undef  STATE_SET
1280 #undef  STATE_TEST
1281 #undef  STATE_SET_TEST
1282 #undef  SWITCH_STATE
1283 }
1284
1285 /*
1286  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
1287  * allocation to a real allocation or vice versa.
1288  */
1289 STATIC int                              /* error */
1290 xfs_bmap_add_extent_unwritten_real(
1291         xfs_inode_t             *ip,    /* incore inode pointer */
1292         xfs_extnum_t            idx,    /* extent number to update/insert */
1293         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
1294         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1295         int                     *logflagsp, /* inode logging flags */
1296         xfs_extdelta_t          *delta) /* Change made to incore extents */
1297 {
1298         xfs_btree_cur_t         *cur;   /* btree cursor */
1299         xfs_bmbt_rec_t          *ep;    /* extent entry for idx */
1300         int                     error;  /* error return value */
1301 #ifdef XFS_BMAP_TRACE
1302         static char             fname[] = "xfs_bmap_add_extent_unwritten_real";
1303 #endif
1304         int                     i;      /* temp state */
1305         xfs_ifork_t             *ifp;   /* inode fork pointer */
1306         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1307         xfs_exntst_t            newext; /* new extent state */
1308         xfs_exntst_t            oldext; /* old extent state */
1309         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1310                                         /* left is 0, right is 1, prev is 2 */
1311         int                     rval=0; /* return value (logging flags) */
1312         int                     state = 0;/* state bits, accessed thru macros */
1313         xfs_filblks_t           temp=0;
1314         xfs_filblks_t           temp2=0;
1315         enum {                          /* bit number definitions for state */
1316                 LEFT_CONTIG,    RIGHT_CONTIG,
1317                 LEFT_FILLING,   RIGHT_FILLING,
1318                 LEFT_DELAY,     RIGHT_DELAY,
1319                 LEFT_VALID,     RIGHT_VALID
1320         };
1321
1322 #define LEFT            r[0]
1323 #define RIGHT           r[1]
1324 #define PREV            r[2]
1325 #define MASK(b)         (1 << (b))
1326 #define MASK2(a,b)      (MASK(a) | MASK(b))
1327 #define MASK3(a,b,c)    (MASK2(a,b) | MASK(c))
1328 #define MASK4(a,b,c,d)  (MASK3(a,b,c) | MASK(d))
1329 #define STATE_SET(b,v)  ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1330 #define STATE_TEST(b)   (state & MASK(b))
1331 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
1332                                        ((state &= ~MASK(b)), 0))
1333 #define SWITCH_STATE            \
1334         (state & MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG))
1335
1336         /*
1337          * Set up a bunch of variables to make the tests simpler.
1338          */
1339         error = 0;
1340         cur = *curp;
1341         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1342         ep = xfs_iext_get_ext(ifp, idx);
1343         xfs_bmbt_get_all(ep, &PREV);
1344         newext = new->br_state;
1345         oldext = (newext == XFS_EXT_UNWRITTEN) ?
1346                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
1347         ASSERT(PREV.br_state == oldext);
1348         new_endoff = new->br_startoff + new->br_blockcount;
1349         ASSERT(PREV.br_startoff <= new->br_startoff);
1350         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1351         /*
1352          * Set flags determining what part of the previous oldext allocation
1353          * extent is being replaced by a newext allocation.
1354          */
1355         STATE_SET(LEFT_FILLING, PREV.br_startoff == new->br_startoff);
1356         STATE_SET(RIGHT_FILLING,
1357                 PREV.br_startoff + PREV.br_blockcount == new_endoff);
1358         /*
1359          * Check and set flags if this segment has a left neighbor.
1360          * Don't set contiguous if the combined extent would be too large.
1361          */
1362         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1363                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &LEFT);
1364                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(LEFT.br_startblock));
1365         }
1366         STATE_SET(LEFT_CONTIG,
1367                 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
1368                 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1369                 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1370                 LEFT.br_state == newext &&
1371                 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1372         /*
1373          * Check and set flags if this segment has a right neighbor.
1374          * Don't set contiguous if the combined extent would be too large.
1375          * Also check for all-three-contiguous being too large.
1376          */
1377         if (STATE_SET_TEST(RIGHT_VALID,
1378                         idx <
1379                         ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1)) {
1380                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx + 1), &RIGHT);
1381                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(RIGHT.br_startblock));
1382         }
1383         STATE_SET(RIGHT_CONTIG,
1384                 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
1385                 new_endoff == RIGHT.br_startoff &&
1386                 new->br_startblock + new->br_blockcount ==
1387                     RIGHT.br_startblock &&
1388                 newext == RIGHT.br_state &&
1389                 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1390                 ((state & MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING)) !=
1391                   MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING) ||
1392                  LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1393                      <= MAXEXTLEN));
1394         /*
1395          * Switch out based on the FILLING and CONTIG state bits.
1396          */
1397         switch (SWITCH_STATE) {
1398
1399         case MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1400                 /*
1401                  * Setting all of a previous oldext extent to newext.
1402                  * The left and right neighbors are both contiguous with new.
1403                  */
1404                 xfs_bmap_trace_pre_update(fname, "LF|RF|LC|RC", ip, idx - 1,
1405                         XFS_DATA_FORK);
1406                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1407                         LEFT.br_blockcount + PREV.br_blockcount +
1408                         RIGHT.br_blockcount);
1409                 xfs_bmap_trace_post_update(fname, "LF|RF|LC|RC", ip, idx - 1,
1410                         XFS_DATA_FORK);
1411                 xfs_bmap_trace_delete(fname, "LF|RF|LC|RC", ip, idx, 2,
1412                         XFS_DATA_FORK);
1413                 xfs_iext_remove(ifp, idx, 2);
1414                 ip->i_df.if_lastex = idx - 1;
1415                 ip->i_d.di_nextents -= 2;
1416                 if (cur == NULL)
1417                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1418                 else {
1419                         rval = XFS_ILOG_CORE;
1420                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1421                                         RIGHT.br_startblock,
1422                                         RIGHT.br_blockcount, &i)))
1423                                 goto done;
1424                         ASSERT(i == 1);
1425                         if ((error = xfs_bmbt_delete(cur, &i)))
1426                                 goto done;
1427                         ASSERT(i == 1);
1428                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1429                                 goto done;
1430                         ASSERT(i == 1);
1431                         if ((error = xfs_bmbt_delete(cur, &i)))
1432                                 goto done;
1433                         ASSERT(i == 1);
1434                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1435                                 goto done;
1436                         ASSERT(i == 1);
1437                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1438                                 LEFT.br_startblock,
1439                                 LEFT.br_blockcount + PREV.br_blockcount +
1440                                 RIGHT.br_blockcount, LEFT.br_state)))
1441                                 goto done;
1442                 }
1443                 /* DELTA: Three in-core extents are replaced by one. */
1444                 temp = LEFT.br_startoff;
1445                 temp2 = LEFT.br_blockcount +
1446                         PREV.br_blockcount +
1447                         RIGHT.br_blockcount;
1448                 break;
1449
1450         case MASK3(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG):
1451                 /*
1452                  * Setting all of a previous oldext extent to newext.
1453                  * The left neighbor is contiguous, the right is not.
1454                  */
1455                 xfs_bmap_trace_pre_update(fname, "LF|RF|LC", ip, idx - 1,
1456                         XFS_DATA_FORK);
1457                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1458                         LEFT.br_blockcount + PREV.br_blockcount);
1459                 xfs_bmap_trace_post_update(fname, "LF|RF|LC", ip, idx - 1,
1460                         XFS_DATA_FORK);
1461                 ip->i_df.if_lastex = idx - 1;
1462                 xfs_bmap_trace_delete(fname, "LF|RF|LC", ip, idx, 1,
1463                         XFS_DATA_FORK);
1464                 xfs_iext_remove(ifp, idx, 1);
1465                 ip->i_d.di_nextents--;
1466                 if (cur == NULL)
1467                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1468                 else {
1469                         rval = XFS_ILOG_CORE;
1470                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1471                                         PREV.br_startblock, PREV.br_blockcount,
1472                                         &i)))
1473                                 goto done;
1474                         ASSERT(i == 1);
1475                         if ((error = xfs_bmbt_delete(cur, &i)))
1476                                 goto done;
1477                         ASSERT(i == 1);
1478                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1479                                 goto done;
1480                         ASSERT(i == 1);
1481                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1482                                 LEFT.br_startblock,
1483                                 LEFT.br_blockcount + PREV.br_blockcount,
1484                                 LEFT.br_state)))
1485                                 goto done;
1486                 }
1487                 /* DELTA: Two in-core extents are replaced by one. */
1488                 temp = LEFT.br_startoff;
1489                 temp2 = LEFT.br_blockcount +
1490                         PREV.br_blockcount;
1491                 break;
1492
1493         case MASK3(LEFT_FILLING, RIGHT_FILLING, RIGHT_CONTIG):
1494                 /*
1495                  * Setting all of a previous oldext extent to newext.
1496                  * The right neighbor is contiguous, the left is not.
1497                  */
1498                 xfs_bmap_trace_pre_update(fname, "LF|RF|RC", ip, idx,
1499                         XFS_DATA_FORK);
1500                 xfs_bmbt_set_blockcount(ep,
1501                         PREV.br_blockcount + RIGHT.br_blockcount);
1502                 xfs_bmbt_set_state(ep, newext);
1503                 xfs_bmap_trace_post_update(fname, "LF|RF|RC", ip, idx,
1504                         XFS_DATA_FORK);
1505                 ip->i_df.if_lastex = idx;
1506                 xfs_bmap_trace_delete(fname, "LF|RF|RC", ip, idx + 1, 1,
1507                         XFS_DATA_FORK);
1508                 xfs_iext_remove(ifp, idx + 1, 1);
1509                 ip->i_d.di_nextents--;
1510                 if (cur == NULL)
1511                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1512                 else {
1513                         rval = XFS_ILOG_CORE;
1514                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1515                                         RIGHT.br_startblock,
1516                                         RIGHT.br_blockcount, &i)))
1517                                 goto done;
1518                         ASSERT(i == 1);
1519                         if ((error = xfs_bmbt_delete(cur, &i)))
1520                                 goto done;
1521                         ASSERT(i == 1);
1522                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1523                                 goto done;
1524                         ASSERT(i == 1);
1525                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1526                                 new->br_startblock,
1527                                 new->br_blockcount + RIGHT.br_blockcount,
1528                                 newext)))
1529                                 goto done;
1530                 }
1531                 /* DELTA: Two in-core extents are replaced by one. */
1532                 temp = PREV.br_startoff;
1533                 temp2 = PREV.br_blockcount +
1534                         RIGHT.br_blockcount;
1535                 break;
1536
1537         case MASK2(LEFT_FILLING, RIGHT_FILLING):
1538                 /*
1539                  * Setting all of a previous oldext extent to newext.
1540                  * Neither the left nor right neighbors are contiguous with
1541                  * the new one.
1542                  */
1543                 xfs_bmap_trace_pre_update(fname, "LF|RF", ip, idx,
1544                         XFS_DATA_FORK);
1545                 xfs_bmbt_set_state(ep, newext);
1546                 xfs_bmap_trace_post_update(fname, "LF|RF", ip, idx,
1547                         XFS_DATA_FORK);
1548                 ip->i_df.if_lastex = idx;
1549                 if (cur == NULL)
1550                         rval = XFS_ILOG_DEXT;
1551                 else {
1552                         rval = 0;
1553                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1554                                         new->br_startblock, new->br_blockcount,
1555                                         &i)))
1556                                 goto done;
1557                         ASSERT(i == 1);
1558                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1559                                 new->br_startblock, new->br_blockcount,
1560                                 newext)))
1561                                 goto done;
1562                 }
1563                 /* DELTA: The in-core extent described by new changed type. */
1564                 temp = new->br_startoff;
1565                 temp2 = new->br_blockcount;
1566                 break;
1567
1568         case MASK2(LEFT_FILLING, LEFT_CONTIG):
1569                 /*
1570                  * Setting the first part of a previous oldext extent to newext.
1571                  * The left neighbor is contiguous.
1572                  */
1573                 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx - 1,
1574                         XFS_DATA_FORK);
1575                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1576                         LEFT.br_blockcount + new->br_blockcount);
1577                 xfs_bmbt_set_startoff(ep,
1578                         PREV.br_startoff + new->br_blockcount);
1579                 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx - 1,
1580                         XFS_DATA_FORK);
1581                 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx,
1582                         XFS_DATA_FORK);
1583                 xfs_bmbt_set_startblock(ep,
1584                         new->br_startblock + new->br_blockcount);
1585                 xfs_bmbt_set_blockcount(ep,
1586                         PREV.br_blockcount - new->br_blockcount);
1587                 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx,
1588                         XFS_DATA_FORK);
1589                 ip->i_df.if_lastex = idx - 1;
1590                 if (cur == NULL)
1591                         rval = XFS_ILOG_DEXT;
1592                 else {
1593                         rval = 0;
1594                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1595                                         PREV.br_startblock, PREV.br_blockcount,
1596                                         &i)))
1597                                 goto done;
1598                         ASSERT(i == 1);
1599                         if ((error = xfs_bmbt_update(cur,
1600                                 PREV.br_startoff + new->br_blockcount,
1601                                 PREV.br_startblock + new->br_blockcount,
1602                                 PREV.br_blockcount - new->br_blockcount,
1603                                 oldext)))
1604                                 goto done;
1605                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1606                                 goto done;
1607                         if (xfs_bmbt_update(cur, LEFT.br_startoff,
1608                                 LEFT.br_startblock,
1609                                 LEFT.br_blockcount + new->br_blockcount,
1610                                 LEFT.br_state))
1611                                 goto done;
1612                 }
1613                 /* DELTA: The boundary between two in-core extents moved. */
1614                 temp = LEFT.br_startoff;
1615                 temp2 = LEFT.br_blockcount +
1616                         PREV.br_blockcount;
1617                 break;
1618
1619         case MASK(LEFT_FILLING):
1620                 /*
1621                  * Setting the first part of a previous oldext extent to newext.
1622                  * The left neighbor is not contiguous.
1623                  */
1624                 xfs_bmap_trace_pre_update(fname, "LF", ip, idx, XFS_DATA_FORK);
1625                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1626                 xfs_bmbt_set_startoff(ep, new_endoff);
1627                 xfs_bmbt_set_blockcount(ep,
1628                         PREV.br_blockcount - new->br_blockcount);
1629                 xfs_bmbt_set_startblock(ep,
1630                         new->br_startblock + new->br_blockcount);
1631                 xfs_bmap_trace_post_update(fname, "LF", ip, idx, XFS_DATA_FORK);
1632                 xfs_bmap_trace_insert(fname, "LF", ip, idx, 1, new, NULL,
1633                         XFS_DATA_FORK);
1634                 xfs_iext_insert(ifp, idx, 1, new);
1635                 ip->i_df.if_lastex = idx;
1636                 ip->i_d.di_nextents++;
1637                 if (cur == NULL)
1638                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1639                 else {
1640                         rval = XFS_ILOG_CORE;
1641                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1642                                         PREV.br_startblock, PREV.br_blockcount,
1643                                         &i)))
1644                                 goto done;
1645                         ASSERT(i == 1);
1646                         if ((error = xfs_bmbt_update(cur,
1647                                 PREV.br_startoff + new->br_blockcount,
1648                                 PREV.br_startblock + new->br_blockcount,
1649                                 PREV.br_blockcount - new->br_blockcount,
1650                                 oldext)))
1651                                 goto done;
1652                         cur->bc_rec.b = *new;
1653                         if ((error = xfs_bmbt_insert(cur, &i)))
1654                                 goto done;
1655                         ASSERT(i == 1);
1656                 }
1657                 /* DELTA: One in-core extent is split in two. */
1658                 temp = PREV.br_startoff;
1659                 temp2 = PREV.br_blockcount;
1660                 break;
1661
1662         case MASK2(RIGHT_FILLING, RIGHT_CONTIG):
1663                 /*
1664                  * Setting the last part of a previous oldext extent to newext.
1665                  * The right neighbor is contiguous with the new allocation.
1666                  */
1667                 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx,
1668                         XFS_DATA_FORK);
1669                 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx + 1,
1670                         XFS_DATA_FORK);
1671                 xfs_bmbt_set_blockcount(ep,
1672                         PREV.br_blockcount - new->br_blockcount);
1673                 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx,
1674                         XFS_DATA_FORK);
1675                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, idx + 1),
1676                         new->br_startoff, new->br_startblock,
1677                         new->br_blockcount + RIGHT.br_blockcount, newext);
1678                 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx + 1,
1679                         XFS_DATA_FORK);
1680                 ip->i_df.if_lastex = idx + 1;
1681                 if (cur == NULL)
1682                         rval = XFS_ILOG_DEXT;
1683                 else {
1684                         rval = 0;
1685                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1686                                         PREV.br_startblock,
1687                                         PREV.br_blockcount, &i)))
1688                                 goto done;
1689                         ASSERT(i == 1);
1690                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1691                                 PREV.br_startblock,
1692                                 PREV.br_blockcount - new->br_blockcount,
1693                                 oldext)))
1694                                 goto done;
1695                         if ((error = xfs_bmbt_increment(cur, 0, &i)))
1696                                 goto done;
1697                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1698                                 new->br_startblock,
1699                                 new->br_blockcount + RIGHT.br_blockcount,
1700                                 newext)))
1701                                 goto done;
1702                 }
1703                 /* DELTA: The boundary between two in-core extents moved. */
1704                 temp = PREV.br_startoff;
1705                 temp2 = PREV.br_blockcount +
1706                         RIGHT.br_blockcount;
1707                 break;
1708
1709         case MASK(RIGHT_FILLING):
1710                 /*
1711                  * Setting the last part of a previous oldext extent to newext.
1712                  * The right neighbor is not contiguous.
1713                  */
1714                 xfs_bmap_trace_pre_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1715                 xfs_bmbt_set_blockcount(ep,
1716                         PREV.br_blockcount - new->br_blockcount);
1717                 xfs_bmap_trace_post_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1718                 xfs_bmap_trace_insert(fname, "RF", ip, idx + 1, 1,
1719                         new, NULL, XFS_DATA_FORK);
1720                 xfs_iext_insert(ifp, idx + 1, 1, new);
1721                 ip->i_df.if_lastex = idx + 1;
1722                 ip->i_d.di_nextents++;
1723                 if (cur == NULL)
1724                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1725                 else {
1726                         rval = XFS_ILOG_CORE;
1727                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1728                                         PREV.br_startblock, PREV.br_blockcount,
1729                                         &i)))
1730                                 goto done;
1731                         ASSERT(i == 1);
1732                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1733                                 PREV.br_startblock,
1734                                 PREV.br_blockcount - new->br_blockcount,
1735                                 oldext)))
1736                                 goto done;
1737                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1738                                         new->br_startblock, new->br_blockcount,
1739                                         &i)))
1740                                 goto done;
1741                         ASSERT(i == 0);
1742                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1743                         if ((error = xfs_bmbt_insert(cur, &i)))
1744                                 goto done;
1745                         ASSERT(i == 1);
1746                 }
1747                 /* DELTA: One in-core extent is split in two. */
1748                 temp = PREV.br_startoff;
1749                 temp2 = PREV.br_blockcount;
1750                 break;
1751
1752         case 0:
1753                 /*
1754                  * Setting the middle part of a previous oldext extent to
1755                  * newext.  Contiguity is impossible here.
1756                  * One extent becomes three extents.
1757                  */
1758                 xfs_bmap_trace_pre_update(fname, "0", ip, idx, XFS_DATA_FORK);
1759                 xfs_bmbt_set_blockcount(ep,
1760                         new->br_startoff - PREV.br_startoff);
1761                 xfs_bmap_trace_post_update(fname, "0", ip, idx, XFS_DATA_FORK);
1762                 r[0] = *new;
1763                 r[1].br_startoff = new_endoff;
1764                 r[1].br_blockcount =
1765                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
1766                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1767                 r[1].br_state = oldext;
1768                 xfs_bmap_trace_insert(fname, "0", ip, idx + 1, 2, &r[0], &r[1],
1769                         XFS_DATA_FORK);
1770                 xfs_iext_insert(ifp, idx + 1, 2, &r[0]);
1771                 ip->i_df.if_lastex = idx + 1;
1772                 ip->i_d.di_nextents += 2;
1773                 if (cur == NULL)
1774                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1775                 else {
1776                         rval = XFS_ILOG_CORE;
1777                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1778                                         PREV.br_startblock, PREV.br_blockcount,
1779                                         &i)))
1780                                 goto done;
1781                         ASSERT(i == 1);
1782                         /* new right extent - oldext */
1783                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1784                                 r[1].br_startblock, r[1].br_blockcount,
1785                                 r[1].br_state)))
1786                                 goto done;
1787                         /* new left extent - oldext */
1788                         PREV.br_blockcount =
1789                                 new->br_startoff - PREV.br_startoff;
1790                         cur->bc_rec.b = PREV;
1791                         if ((error = xfs_bmbt_insert(cur, &i)))
1792                                 goto done;
1793                         ASSERT(i == 1);
1794                         if ((error = xfs_bmbt_increment(cur, 0, &i)))
1795                                 goto done;
1796                         ASSERT(i == 1);
1797                         /* new middle extent - newext */
1798                         cur->bc_rec.b = *new;
1799                         if ((error = xfs_bmbt_insert(cur, &i)))
1800                                 goto done;
1801                         ASSERT(i == 1);
1802                 }
1803                 /* DELTA: One in-core extent is split in three. */
1804                 temp = PREV.br_startoff;
1805                 temp2 = PREV.br_blockcount;
1806                 break;
1807
1808         case MASK3(LEFT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1809         case MASK3(RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1810         case MASK2(LEFT_FILLING, RIGHT_CONTIG):
1811         case MASK2(RIGHT_FILLING, LEFT_CONTIG):
1812         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1813         case MASK(LEFT_CONTIG):
1814         case MASK(RIGHT_CONTIG):
1815                 /*
1816                  * These cases are all impossible.
1817                  */
1818                 ASSERT(0);
1819         }
1820         *curp = cur;
1821         if (delta) {
1822                 temp2 += temp;
1823                 if (delta->xed_startoff > temp)
1824                         delta->xed_startoff = temp;
1825                 if (delta->xed_blockcount < temp2)
1826                         delta->xed_blockcount = temp2;
1827         }
1828 done:
1829         *logflagsp = rval;
1830         return error;
1831 #undef  LEFT
1832 #undef  RIGHT
1833 #undef  PREV
1834 #undef  MASK
1835 #undef  MASK2
1836 #undef  MASK3
1837 #undef  MASK4
1838 #undef  STATE_SET
1839 #undef  STATE_TEST
1840 #undef  STATE_SET_TEST
1841 #undef  SWITCH_STATE
1842 }
1843
1844 /*
1845  * Called by xfs_bmap_add_extent to handle cases converting a hole
1846  * to a delayed allocation.
1847  */
1848 /*ARGSUSED*/
1849 STATIC int                              /* error */
1850 xfs_bmap_add_extent_hole_delay(
1851         xfs_inode_t             *ip,    /* incore inode pointer */
1852         xfs_extnum_t            idx,    /* extent number to update/insert */
1853         xfs_btree_cur_t         *cur,   /* if null, not a btree */
1854         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1855         int                     *logflagsp, /* inode logging flags */
1856         xfs_extdelta_t          *delta, /* Change made to incore extents */
1857         int                     rsvd)           /* OK to allocate reserved blocks */
1858 {
1859         xfs_bmbt_rec_t          *ep;    /* extent record for idx */
1860 #ifdef XFS_BMAP_TRACE
1861         static char             fname[] = "xfs_bmap_add_extent_hole_delay";
1862 #endif
1863         xfs_ifork_t             *ifp;   /* inode fork pointer */
1864         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1865         xfs_filblks_t           newlen=0;       /* new indirect size */
1866         xfs_filblks_t           oldlen=0;       /* old indirect size */
1867         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1868         int                     state;  /* state bits, accessed thru macros */
1869         xfs_filblks_t           temp=0; /* temp for indirect calculations */
1870         xfs_filblks_t           temp2=0;
1871         enum {                          /* bit number definitions for state */
1872                 LEFT_CONTIG,    RIGHT_CONTIG,
1873                 LEFT_DELAY,     RIGHT_DELAY,
1874                 LEFT_VALID,     RIGHT_VALID
1875         };
1876
1877 #define MASK(b)                 (1 << (b))
1878 #define MASK2(a,b)              (MASK(a) | MASK(b))
1879 #define STATE_SET(b,v)          ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1880 #define STATE_TEST(b)           (state & MASK(b))
1881 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
1882                                        ((state &= ~MASK(b)), 0))
1883 #define SWITCH_STATE            (state & MASK2(LEFT_CONTIG, RIGHT_CONTIG))
1884
1885         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1886         ep = xfs_iext_get_ext(ifp, idx);
1887         state = 0;
1888         ASSERT(ISNULLSTARTBLOCK(new->br_startblock));
1889         /*
1890          * Check and set flags if this segment has a left neighbor
1891          */
1892         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1893                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &left);
1894                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(left.br_startblock));
1895         }
1896         /*
1897          * Check and set flags if the current (right) segment exists.
1898          * If it doesn't exist, we're converting the hole at end-of-file.
1899          */
1900         if (STATE_SET_TEST(RIGHT_VALID,
1901                            idx <
1902                            ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1903                 xfs_bmbt_get_all(ep, &right);
1904                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(right.br_startblock));
1905         }
1906         /*
1907          * Set contiguity flags on the left and right neighbors.
1908          * Don't let extents get too large, even if the pieces are contiguous.
1909          */
1910         STATE_SET(LEFT_CONTIG,
1911                 STATE_TEST(LEFT_VALID) && STATE_TEST(LEFT_DELAY) &&
1912                 left.br_startoff + left.br_blockcount == new->br_startoff &&
1913                 left.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1914         STATE_SET(RIGHT_CONTIG,
1915                 STATE_TEST(RIGHT_VALID) && STATE_TEST(RIGHT_DELAY) &&
1916                 new->br_startoff + new->br_blockcount == right.br_startoff &&
1917                 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1918                 (!STATE_TEST(LEFT_CONTIG) ||
1919                  (left.br_blockcount + new->br_blockcount +
1920                      right.br_blockcount <= MAXEXTLEN)));
1921         /*
1922          * Switch out based on the contiguity flags.
1923          */
1924         switch (SWITCH_STATE) {
1925
1926         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1927                 /*
1928                  * New allocation is contiguous with delayed allocations
1929                  * on the left and on the right.
1930                  * Merge all three into a single extent record.
1931                  */
1932                 temp = left.br_blockcount + new->br_blockcount +
1933                         right.br_blockcount;
1934                 xfs_bmap_trace_pre_update(fname, "LC|RC", ip, idx - 1,
1935                         XFS_DATA_FORK);
1936                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1), temp);
1937                 oldlen = STARTBLOCKVAL(left.br_startblock) +
1938                         STARTBLOCKVAL(new->br_startblock) +
1939                         STARTBLOCKVAL(right.br_startblock);
1940                 newlen = xfs_bmap_worst_indlen(ip, temp);
1941                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx - 1),
1942                         NULLSTARTBLOCK((int)newlen));
1943                 xfs_bmap_trace_post_update(fname, "LC|RC", ip, idx - 1,
1944                         XFS_DATA_FORK);
1945                 xfs_bmap_trace_delete(fname, "LC|RC", ip, idx, 1,
1946                         XFS_DATA_FORK);
1947                 xfs_iext_remove(ifp, idx, 1);
1948                 ip->i_df.if_lastex = idx - 1;
1949                 /* DELTA: Two in-core extents were replaced by one. */
1950                 temp2 = temp;
1951                 temp = left.br_startoff;
1952                 break;
1953
1954         case MASK(LEFT_CONTIG):
1955                 /*
1956                  * New allocation is contiguous with a delayed allocation
1957                  * on the left.
1958                  * Merge the new allocation with the left neighbor.
1959                  */
1960                 temp = left.br_blockcount + new->br_blockcount;
1961                 xfs_bmap_trace_pre_update(fname, "LC", ip, idx - 1,
1962                         XFS_DATA_FORK);
1963                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1), temp);
1964                 oldlen = STARTBLOCKVAL(left.br_startblock) +
1965                         STARTBLOCKVAL(new->br_startblock);
1966                 newlen = xfs_bmap_worst_indlen(ip, temp);
1967                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx - 1),
1968                         NULLSTARTBLOCK((int)newlen));
1969                 xfs_bmap_trace_post_update(fname, "LC", ip, idx - 1,
1970                         XFS_DATA_FORK);
1971                 ip->i_df.if_lastex = idx - 1;
1972                 /* DELTA: One in-core extent grew into a hole. */
1973                 temp2 = temp;
1974                 temp = left.br_startoff;
1975                 break;
1976
1977         case MASK(RIGHT_CONTIG):
1978                 /*
1979                  * New allocation is contiguous with a delayed allocation
1980                  * on the right.
1981                  * Merge the new allocation with the right neighbor.
1982                  */
1983                 xfs_bmap_trace_pre_update(fname, "RC", ip, idx, XFS_DATA_FORK);
1984                 temp = new->br_blockcount + right.br_blockcount;
1985                 oldlen = STARTBLOCKVAL(new->br_startblock) +
1986                         STARTBLOCKVAL(right.br_startblock);
1987                 newlen = xfs_bmap_worst_indlen(ip, temp);
1988                 xfs_bmbt_set_allf(ep, new->br_startoff,
1989                         NULLSTARTBLOCK((int)newlen), temp, right.br_state);
1990                 xfs_bmap_trace_post_update(fname, "RC", ip, idx, XFS_DATA_FORK);
1991                 ip->i_df.if_lastex = idx;
1992                 /* DELTA: One in-core extent grew into a hole. */
1993                 temp2 = temp;
1994                 temp = new->br_startoff;
1995                 break;
1996
1997         case 0:
1998                 /*
1999                  * New allocation is not contiguous with another
2000                  * delayed allocation.
2001                  * Insert a new entry.
2002                  */
2003                 oldlen = newlen = 0;
2004                 xfs_bmap_trace_insert(fname, "0", ip, idx, 1, new, NULL,
2005                         XFS_DATA_FORK);
2006                 xfs_iext_insert(ifp, idx, 1, new);
2007                 ip->i_df.if_lastex = idx;
2008                 /* DELTA: A new in-core extent was added in a hole. */
2009                 temp2 = new->br_blockcount;
2010                 temp = new->br_startoff;
2011                 break;
2012         }
2013         if (oldlen != newlen) {
2014                 ASSERT(oldlen > newlen);
2015                 xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
2016                         (int)(oldlen - newlen), rsvd);
2017                 /*
2018                  * Nothing to do for disk quota accounting here.
2019                  */
2020         }
2021         if (delta) {
2022                 temp2 += temp;
2023                 if (delta->xed_startoff > temp)
2024                         delta->xed_startoff = temp;
2025                 if (delta->xed_blockcount < temp2)
2026                         delta->xed_blockcount = temp2;
2027         }
2028         *logflagsp = 0;
2029         return 0;
2030 #undef  MASK
2031 #undef  MASK2
2032 #undef  STATE_SET
2033 #undef  STATE_TEST
2034 #undef  STATE_SET_TEST
2035 #undef  SWITCH_STATE
2036 }
2037
2038 /*
2039  * Called by xfs_bmap_add_extent to handle cases converting a hole
2040  * to a real allocation.
2041  */
2042 STATIC int                              /* error */
2043 xfs_bmap_add_extent_hole_real(
2044         xfs_inode_t             *ip,    /* incore inode pointer */
2045         xfs_extnum_t            idx,    /* extent number to update/insert */
2046         xfs_btree_cur_t         *cur,   /* if null, not a btree */
2047         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
2048         int                     *logflagsp, /* inode logging flags */
2049         xfs_extdelta_t          *delta, /* Change made to incore extents */
2050         int                     whichfork) /* data or attr fork */
2051 {
2052         xfs_bmbt_rec_t          *ep;    /* pointer to extent entry ins. point */
2053         int                     error;  /* error return value */
2054 #ifdef XFS_BMAP_TRACE
2055         static char             fname[] = "xfs_bmap_add_extent_hole_real";
2056 #endif
2057         int                     i;      /* temp state */
2058         xfs_ifork_t             *ifp;   /* inode fork pointer */
2059         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2060         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2061         int                     rval=0; /* return value (logging flags) */
2062         int                     state;  /* state bits, accessed thru macros */
2063         xfs_filblks_t           temp=0;
2064         xfs_filblks_t           temp2=0;
2065         enum {                          /* bit number definitions for state */
2066                 LEFT_CONTIG,    RIGHT_CONTIG,
2067                 LEFT_DELAY,     RIGHT_DELAY,
2068                 LEFT_VALID,     RIGHT_VALID
2069         };
2070
2071 #define MASK(b)                 (1 << (b))
2072 #define MASK2(a,b)              (MASK(a) | MASK(b))
2073 #define STATE_SET(b,v)          ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
2074 #define STATE_TEST(b)           (state & MASK(b))
2075 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
2076                                        ((state &= ~MASK(b)), 0))
2077 #define SWITCH_STATE            (state & MASK2(LEFT_CONTIG, RIGHT_CONTIG))
2078
2079         ifp = XFS_IFORK_PTR(ip, whichfork);
2080         ASSERT(idx <= ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
2081         ep = xfs_iext_get_ext(ifp, idx);
2082         state = 0;
2083         /*
2084          * Check and set flags if this segment has a left neighbor.
2085          */
2086         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
2087                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &left);
2088                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(left.br_startblock));
2089         }
2090         /*
2091          * Check and set flags if this segment has a current value.
2092          * Not true if we're inserting into the "hole" at eof.
2093          */
2094         if (STATE_SET_TEST(RIGHT_VALID,
2095                            idx <
2096                            ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
2097                 xfs_bmbt_get_all(ep, &right);
2098                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(right.br_startblock));
2099         }
2100         /*
2101          * We're inserting a real allocation between "left" and "right".
2102          * Set the contiguity flags.  Don't let extents get too large.
2103          */
2104         STATE_SET(LEFT_CONTIG,
2105                 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
2106                 left.br_startoff + left.br_blockcount == new->br_startoff &&
2107                 left.br_startblock + left.br_blockcount == new->br_startblock &&
2108                 left.br_state == new->br_state &&
2109                 left.br_blockcount + new->br_blockcount <= MAXEXTLEN);
2110         STATE_SET(RIGHT_CONTIG,
2111                 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
2112                 new->br_startoff + new->br_blockcount == right.br_startoff &&
2113                 new->br_startblock + new->br_blockcount ==
2114                     right.br_startblock &&
2115                 new->br_state == right.br_state &&
2116                 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2117                 (!STATE_TEST(LEFT_CONTIG) ||
2118                  left.br_blockcount + new->br_blockcount +
2119                      right.br_blockcount <= MAXEXTLEN));
2120
2121         error = 0;
2122         /*
2123          * Select which case we're in here, and implement it.
2124          */
2125         switch (SWITCH_STATE) {
2126
2127         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
2128                 /*
2129                  * New allocation is contiguous with real allocations on the
2130                  * left and on the right.
2131                  * Merge all three into a single extent record.
2132                  */
2133                 xfs_bmap_trace_pre_update(fname, "LC|RC", ip, idx - 1,
2134                         whichfork);
2135                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
2136                         left.br_blockcount + new->br_blockcount +
2137                         right.br_blockcount);
2138                 xfs_bmap_trace_post_update(fname, "LC|RC", ip, idx - 1,
2139                         whichfork);
2140                 xfs_bmap_trace_delete(fname, "LC|RC", ip,
2141                         idx, 1, whichfork);
2142                 xfs_iext_remove(ifp, idx, 1);
2143                 ifp->if_lastex = idx - 1;
2144                 XFS_IFORK_NEXT_SET(ip, whichfork,
2145                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2146                 if (cur == NULL) {
2147                         rval = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2148                 } else {
2149                         rval = XFS_ILOG_CORE;
2150                         if ((error = xfs_bmbt_lookup_eq(cur,
2151                                         right.br_startoff,
2152                                         right.br_startblock,
2153                                         right.br_blockcount, &i)))
2154                                 goto done;
2155                         ASSERT(i == 1);
2156                         if ((error = xfs_bmbt_delete(cur, &i)))
2157                                 goto done;
2158                         ASSERT(i == 1);
2159                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
2160                                 goto done;
2161                         ASSERT(i == 1);
2162                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
2163                                         left.br_startblock,
2164                                         left.br_blockcount +
2165                                                 new->br_blockcount +
2166                                                 right.br_blockcount,
2167                                         left.br_state)))
2168                                 goto done;
2169                 }
2170                 /* DELTA: Two in-core extents were replaced by one. */
2171                 temp = left.br_startoff;
2172                 temp2 = left.br_blockcount +
2173                         new->br_blockcount +
2174                         right.br_blockcount;
2175                 break;
2176
2177         case MASK(LEFT_CONTIG):
2178                 /*
2179                  * New allocation is contiguous with a real allocation
2180                  * on the left.
2181                  * Merge the new allocation with the left neighbor.
2182                  */
2183                 xfs_bmap_trace_pre_update(fname, "LC", ip, idx - 1, whichfork);
2184                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
2185                         left.br_blockcount + new->br_blockcount);
2186                 xfs_bmap_trace_post_update(fname, "LC", ip, idx - 1, whichfork);
2187                 ifp->if_lastex = idx - 1;
2188                 if (cur == NULL) {
2189                         rval = XFS_ILOG_FEXT(whichfork);
2190                 } else {
2191                         rval = 0;
2192                         if ((error = xfs_bmbt_lookup_eq(cur,
2193                                         left.br_startoff,
2194                                         left.br_startblock,
2195                                         left.br_blockcount, &i)))
2196                                 goto done;
2197                         ASSERT(i == 1);
2198                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
2199                                         left.br_startblock,
2200                                         left.br_blockcount +
2201                                                 new->br_blockcount,
2202                                         left.br_state)))
2203                                 goto done;
2204                 }
2205                 /* DELTA: One in-core extent grew. */
2206                 temp = left.br_startoff;
2207                 temp2 = left.br_blockcount +
2208                         new->br_blockcount;
2209                 break;
2210
2211         case MASK(RIGHT_CONTIG):
2212                 /*
2213                  * New allocation is contiguous with a real allocation
2214                  * on the right.
2215                  * Merge the new allocation with the right neighbor.
2216                  */
2217                 xfs_bmap_trace_pre_update(fname, "RC", ip, idx, whichfork);
2218                 xfs_bmbt_set_allf(ep, new->br_startoff, new->br_startblock,
2219                         new->br_blockcount + right.br_blockcount,
2220                         right.br_state);
2221                 xfs_bmap_trace_post_update(fname, "RC", ip, idx, whichfork);
2222                 ifp->if_lastex = idx;
2223                 if (cur == NULL) {
2224                         rval = XFS_ILOG_FEXT(whichfork);
2225                 } else {
2226                         rval = 0;
2227                         if ((error = xfs_bmbt_lookup_eq(cur,
2228                                         right.br_startoff,
2229                                         right.br_startblock,
2230                                         right.br_blockcount, &i)))
2231                                 goto done;
2232                         ASSERT(i == 1);
2233                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2234                                         new->br_startblock,
2235                                         new->br_blockcount +
2236                                                 right.br_blockcount,
2237                                         right.br_state)))
2238                                 goto done;
2239                 }
2240                 /* DELTA: One in-core extent grew. */
2241                 temp = new->br_startoff;
2242                 temp2 = new->br_blockcount +
2243                         right.br_blockcount;
2244                 break;
2245
2246         case 0:
2247                 /*
2248                  * New allocation is not contiguous with another
2249                  * real allocation.
2250                  * Insert a new entry.
2251                  */
2252                 xfs_bmap_trace_insert(fname, "0", ip, idx, 1, new, NULL,
2253                         whichfork);
2254                 xfs_iext_insert(ifp, idx, 1, new);
2255                 ifp->if_lastex = idx;
2256                 XFS_IFORK_NEXT_SET(ip, whichfork,
2257                         XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2258                 if (cur == NULL) {
2259                         rval = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2260                 } else {
2261                         rval = XFS_ILOG_CORE;
2262                         if ((error = xfs_bmbt_lookup_eq(cur,
2263                                         new->br_startoff,
2264                                         new->br_startblock,
2265                                         new->br_blockcount, &i)))
2266                                 goto done;
2267                         ASSERT(i == 0);
2268                         cur->bc_rec.b.br_state = new->br_state;
2269                         if ((error = xfs_bmbt_insert(cur, &i)))
2270                                 goto done;
2271                         ASSERT(i == 1);
2272                 }
2273                 /* DELTA: A new extent was added in a hole. */
2274                 temp = new->br_startoff;
2275                 temp2 = new->br_blockcount;
2276                 break;
2277         }
2278         if (delta) {
2279                 temp2 += temp;
2280                 if (delta->xed_startoff > temp)
2281                         delta->xed_startoff = temp;
2282                 if (delta->xed_blockcount < temp2)
2283                         delta->xed_blockcount = temp2;
2284         }
2285 done:
2286         *logflagsp = rval;
2287         return error;
2288 #undef  MASK
2289 #undef  MASK2
2290 #undef  STATE_SET
2291 #undef  STATE_TEST
2292 #undef  STATE_SET_TEST
2293 #undef  SWITCH_STATE
2294 }
2295
2296 /*
2297  * Adjust the size of the new extent based on di_extsize and rt extsize.
2298  */
2299 STATIC int
2300 xfs_bmap_extsize_align(
2301         xfs_mount_t     *mp,
2302         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
2303         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
2304         xfs_extlen_t    extsz,          /* align to this extent size */
2305         int             rt,             /* is this a realtime inode? */
2306         int             eof,            /* is extent at end-of-file? */
2307         int             delay,          /* creating delalloc extent? */
2308         int             convert,        /* overwriting unwritten extent? */
2309         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
2310         xfs_extlen_t    *lenp)          /* in/out: aligned length */
2311 {
2312         xfs_fileoff_t   orig_off;       /* original offset */
2313         xfs_extlen_t    orig_alen;      /* original length */
2314         xfs_fileoff_t   orig_end;       /* original off+len */
2315         xfs_fileoff_t   nexto;          /* next file offset */
2316         xfs_fileoff_t   prevo;          /* previous file offset */
2317         xfs_fileoff_t   align_off;      /* temp for offset */
2318         xfs_extlen_t    align_alen;     /* temp for length */
2319         xfs_extlen_t    temp;           /* temp for calculations */
2320
2321         if (convert)
2322                 return 0;
2323
2324         orig_off = align_off = *offp;
2325         orig_alen = align_alen = *lenp;
2326         orig_end = orig_off + orig_alen;
2327
2328         /*
2329          * If this request overlaps an existing extent, then don't
2330          * attempt to perform any additional alignment.
2331          */
2332         if (!delay && !eof &&
2333             (orig_off >= gotp->br_startoff) &&
2334             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2335                 return 0;
2336         }
2337
2338         /*
2339          * If the file offset is unaligned vs. the extent size
2340          * we need to align it.  This will be possible unless
2341          * the file was previously written with a kernel that didn't
2342          * perform this alignment, or if a truncate shot us in the
2343          * foot.
2344          */
2345         temp = do_mod(orig_off, extsz);
2346         if (temp) {
2347                 align_alen += temp;
2348                 align_off -= temp;
2349         }
2350         /*
2351          * Same adjustment for the end of the requested area.
2352          */
2353         if ((temp = (align_alen % extsz))) {
2354                 align_alen += extsz - temp;
2355         }
2356         /*
2357          * If the previous block overlaps with this proposed allocation
2358          * then move the start forward without adjusting the length.
2359          */
2360         if (prevp->br_startoff != NULLFILEOFF) {
2361                 if (prevp->br_startblock == HOLESTARTBLOCK)
2362                         prevo = prevp->br_startoff;
2363                 else
2364                         prevo = prevp->br_startoff + prevp->br_blockcount;
2365         } else
2366                 prevo = 0;
2367         if (align_off != orig_off && align_off < prevo)
2368                 align_off = prevo;
2369         /*
2370          * If the next block overlaps with this proposed allocation
2371          * then move the start back without adjusting the length,
2372          * but not before offset 0.
2373          * This may of course make the start overlap previous block,
2374          * and if we hit the offset 0 limit then the next block
2375          * can still overlap too.
2376          */
2377         if (!eof && gotp->br_startoff != NULLFILEOFF) {
2378                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
2379                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
2380                         nexto = gotp->br_startoff + gotp->br_blockcount;
2381                 else
2382                         nexto = gotp->br_startoff;
2383         } else
2384                 nexto = NULLFILEOFF;
2385         if (!eof &&
2386             align_off + align_alen != orig_end &&
2387             align_off + align_alen > nexto)
2388                 align_off = nexto > align_alen ? nexto - align_alen : 0;
2389         /*
2390          * If we're now overlapping the next or previous extent that
2391          * means we can't fit an extsz piece in this hole.  Just move
2392          * the start forward to the first valid spot and set
2393          * the length so we hit the end.
2394          */
2395         if (align_off != orig_off && align_off < prevo)
2396                 align_off = prevo;
2397         if (align_off + align_alen != orig_end &&
2398             align_off + align_alen > nexto &&
2399             nexto != NULLFILEOFF) {
2400                 ASSERT(nexto > prevo);
2401                 align_alen = nexto - align_off;
2402         }
2403
2404         /*
2405          * If realtime, and the result isn't a multiple of the realtime
2406          * extent size we need to remove blocks until it is.
2407          */
2408         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
2409                 /*
2410                  * We're not covering the original request, or
2411                  * we won't be able to once we fix the length.
2412                  */
2413                 if (orig_off < align_off ||
2414                     orig_end > align_off + align_alen ||
2415                     align_alen - temp < orig_alen)
2416                         return XFS_ERROR(EINVAL);
2417                 /*
2418                  * Try to fix it by moving the start up.
2419                  */
2420                 if (align_off + temp <= orig_off) {
2421                         align_alen -= temp;
2422                         align_off += temp;
2423                 }
2424                 /*
2425                  * Try to fix it by moving the end in.
2426                  */
2427                 else if (align_off + align_alen - temp >= orig_end)
2428                         align_alen -= temp;
2429                 /*
2430                  * Set the start to the minimum then trim the length.
2431                  */
2432                 else {
2433                         align_alen -= orig_off - align_off;
2434                         align_off = orig_off;
2435                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
2436                 }
2437                 /*
2438                  * Result doesn't cover the request, fail it.
2439                  */
2440                 if (orig_off < align_off || orig_end > align_off + align_alen)
2441                         return XFS_ERROR(EINVAL);
2442         } else {
2443                 ASSERT(orig_off >= align_off);
2444                 ASSERT(orig_end <= align_off + align_alen);
2445         }
2446
2447 #ifdef DEBUG
2448         if (!eof && gotp->br_startoff != NULLFILEOFF)
2449                 ASSERT(align_off + align_alen <= gotp->br_startoff);
2450         if (prevp->br_startoff != NULLFILEOFF)
2451                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
2452 #endif
2453
2454         *lenp = align_alen;
2455         *offp = align_off;
2456         return 0;
2457 }
2458
2459 #define XFS_ALLOC_GAP_UNITS     4
2460
2461 STATIC int
2462 xfs_bmap_adjacent(
2463         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2464 {
2465         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
2466         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2467         xfs_mount_t     *mp;            /* mount point structure */
2468         int             nullfb;         /* true if ap->firstblock isn't set */
2469         int             rt;             /* true if inode is realtime */
2470
2471 #define ISVALID(x,y)    \
2472         (rt ? \
2473                 (x) < mp->m_sb.sb_rblocks : \
2474                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2475                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2476                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2477
2478         mp = ap->ip->i_mount;
2479         nullfb = ap->firstblock == NULLFSBLOCK;
2480         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2481         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2482         /*
2483          * If allocating at eof, and there's a previous real block,
2484          * try to use it's last block as our starting point.
2485          */
2486         if (ap->eof && ap->prevp->br_startoff != NULLFILEOFF &&
2487             !ISNULLSTARTBLOCK(ap->prevp->br_startblock) &&
2488             ISVALID(ap->prevp->br_startblock + ap->prevp->br_blockcount,
2489                     ap->prevp->br_startblock)) {
2490                 ap->rval = ap->prevp->br_startblock + ap->prevp->br_blockcount;
2491                 /*
2492                  * Adjust for the gap between prevp and us.
2493                  */
2494                 adjust = ap->off -
2495                         (ap->prevp->br_startoff + ap->prevp->br_blockcount);
2496                 if (adjust &&
2497                     ISVALID(ap->rval + adjust, ap->prevp->br_startblock))
2498                         ap->rval += adjust;
2499         }
2500         /*
2501          * If not at eof, then compare the two neighbor blocks.
2502          * Figure out whether either one gives us a good starting point,
2503          * and pick the better one.
2504          */
2505         else if (!ap->eof) {
2506                 xfs_fsblock_t   gotbno;         /* right side block number */
2507                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
2508                 xfs_fsblock_t   prevbno;        /* left side block number */
2509                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
2510
2511                 /*
2512                  * If there's a previous (left) block, select a requested
2513                  * start block based on it.
2514                  */
2515                 if (ap->prevp->br_startoff != NULLFILEOFF &&
2516                     !ISNULLSTARTBLOCK(ap->prevp->br_startblock) &&
2517                     (prevbno = ap->prevp->br_startblock +
2518                                ap->prevp->br_blockcount) &&
2519                     ISVALID(prevbno, ap->prevp->br_startblock)) {
2520                         /*
2521                          * Calculate gap to end of previous block.
2522                          */
2523                         adjust = prevdiff = ap->off -
2524                                 (ap->prevp->br_startoff +
2525                                  ap->prevp->br_blockcount);
2526                         /*
2527                          * Figure the startblock based on the previous block's
2528                          * end and the gap size.
2529                          * Heuristic!
2530                          * If the gap is large relative to the piece we're
2531                          * allocating, or using it gives us an invalid block
2532                          * number, then just use the end of the previous block.
2533                          */
2534                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2535                             ISVALID(prevbno + prevdiff,
2536                                     ap->prevp->br_startblock))
2537                                 prevbno += adjust;
2538                         else
2539                                 prevdiff += adjust;
2540                         /*
2541                          * If the firstblock forbids it, can't use it,
2542                          * must use default.
2543                          */
2544                         if (!rt && !nullfb &&
2545                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2546                                 prevbno = NULLFSBLOCK;
2547                 }
2548                 /*
2549                  * No previous block or can't follow it, just default.
2550                  */
2551                 else
2552                         prevbno = NULLFSBLOCK;
2553                 /*
2554                  * If there's a following (right) block, select a requested
2555                  * start block based on it.
2556                  */
2557                 if (!ISNULLSTARTBLOCK(ap->gotp->br_startblock)) {
2558                         /*
2559                          * Calculate gap to start of next block.
2560                          */
2561                         adjust = gotdiff = ap->gotp->br_startoff - ap->off;
2562                         /*
2563                          * Figure the startblock based on the next block's
2564                          * start and the gap size.
2565                          */
2566                         gotbno = ap->gotp->br_startblock;
2567                         /*
2568                          * Heuristic!
2569                          * If the gap is large relative to the piece we're
2570                          * allocating, or using it gives us an invalid block
2571                          * number, then just use the start of the next block
2572                          * offset by our length.
2573                          */
2574                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2575                             ISVALID(gotbno - gotdiff, gotbno))
2576                                 gotbno -= adjust;
2577                         else if (ISVALID(gotbno - ap->alen, gotbno)) {
2578                                 gotbno -= ap->alen;
2579                                 gotdiff += adjust - ap->alen;
2580                         } else
2581                                 gotdiff += adjust;
2582                         /*
2583                          * If the firstblock forbids it, can't use it,
2584                          * must use default.
2585                          */
2586                         if (!rt && !nullfb &&
2587                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2588                                 gotbno = NULLFSBLOCK;
2589                 }
2590                 /*
2591                  * No next block, just default.
2592                  */
2593                 else
2594                         gotbno = NULLFSBLOCK;
2595                 /*
2596                  * If both valid, pick the better one, else the only good
2597                  * one, else ap->rval is already set (to 0 or the inode block).
2598                  */
2599                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2600                         ap->rval = prevdiff <= gotdiff ? prevbno : gotbno;
2601                 else if (prevbno != NULLFSBLOCK)
2602                         ap->rval = prevbno;
2603                 else if (gotbno != NULLFSBLOCK)
2604                         ap->rval = gotbno;
2605         }
2606 #undef ISVALID
2607         return 0;
2608 }
2609
2610 STATIC int
2611 xfs_bmap_rtalloc(
2612         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2613 {
2614         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2615         int             error;          /* error return value */
2616         xfs_mount_t     *mp;            /* mount point structure */
2617         xfs_extlen_t    prod = 0;       /* product factor for allocators */
2618         xfs_extlen_t    ralen = 0;      /* realtime allocation length */
2619         xfs_extlen_t    align;          /* minimum allocation alignment */
2620         xfs_rtblock_t   rtx;            /* realtime extent number */
2621         xfs_rtblock_t   rtb;
2622
2623         mp = ap->ip->i_mount;
2624         align = ap->ip->i_d.di_extsize ?
2625                 ap->ip->i_d.di_extsize : mp->m_sb.sb_rextsize;
2626         prod = align / mp->m_sb.sb_rextsize;
2627         error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2628                                         align, 1, ap->eof, 0,
2629                                         ap->conv, &ap->off, &ap->alen);
2630         if (error)
2631                 return error;
2632         ASSERT(ap->alen);
2633         ASSERT(ap->alen % mp->m_sb.sb_rextsize == 0);
2634
2635         /*
2636          * If the offset & length are not perfectly aligned
2637          * then kill prod, it will just get us in trouble.
2638          */
2639         if (do_mod(ap->off, align) || ap->alen % align)
2640                 prod = 1;
2641         /*
2642          * Set ralen to be the actual requested length in rtextents.
2643          */
2644         ralen = ap->alen / mp->m_sb.sb_rextsize;
2645         /*
2646          * If the old value was close enough to MAXEXTLEN that
2647          * we rounded up to it, cut it back so it's valid again.
2648          * Note that if it's a really large request (bigger than
2649          * MAXEXTLEN), we don't hear about that number, and can't
2650          * adjust the starting point to match it.
2651          */
2652         if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2653                 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2654         /*
2655          * If it's an allocation to an empty file at offset 0,
2656          * pick an extent that will space things out in the rt area.
2657          */
2658         if (ap->eof && ap->off == 0) {
2659                 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2660                 if (error)
2661                         return error;
2662                 ap->rval = rtx * mp->m_sb.sb_rextsize;
2663         } else {
2664                 ap->rval = 0;
2665         }
2666
2667         xfs_bmap_adjacent(ap);
2668
2669         /*
2670          * Realtime allocation, done through xfs_rtallocate_extent.
2671          */
2672         atype = ap->rval == 0 ?  XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2673         do_div(ap->rval, mp->m_sb.sb_rextsize);
2674         rtb = ap->rval;
2675         ap->alen = ralen;
2676         if ((error = xfs_rtallocate_extent(ap->tp, ap->rval, 1, ap->alen,
2677                                 &ralen, atype, ap->wasdel, prod, &rtb)))
2678                 return error;
2679         if (rtb == NULLFSBLOCK && prod > 1 &&
2680             (error = xfs_rtallocate_extent(ap->tp, ap->rval, 1,
2681                                            ap->alen, &ralen, atype,
2682                                            ap->wasdel, 1, &rtb)))
2683                 return error;
2684         ap->rval = rtb;
2685         if (ap->rval != NULLFSBLOCK) {
2686                 ap->rval *= mp->m_sb.sb_rextsize;
2687                 ralen *= mp->m_sb.sb_rextsize;
2688                 ap->alen = ralen;
2689                 ap->ip->i_d.di_nblocks += ralen;
2690                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2691                 if (ap->wasdel)
2692                         ap->ip->i_delayed_blks -= ralen;
2693                 /*
2694                  * Adjust the disk quota also. This was reserved
2695                  * earlier.
2696                  */
2697                 XFS_TRANS_MOD_DQUOT_BYINO(mp, ap->tp, ap->ip,
2698                         ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2699                                         XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2700         } else {
2701                 ap->alen = 0;
2702         }
2703         return 0;
2704 }
2705
2706 STATIC int
2707 xfs_bmap_btalloc(
2708         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2709 {
2710         xfs_mount_t     *mp;            /* mount point structure */
2711         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2712         xfs_extlen_t    align;          /* minimum allocation alignment */
2713         xfs_agnumber_t  ag;
2714         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2715         xfs_agnumber_t  startag;
2716         xfs_alloc_arg_t args;
2717         xfs_extlen_t    blen;
2718         xfs_extlen_t    delta;
2719         xfs_extlen_t    longest;
2720         xfs_extlen_t    need;
2721         xfs_extlen_t    nextminlen = 0;
2722         xfs_perag_t     *pag;
2723         int             nullfb;         /* true if ap->firstblock isn't set */
2724         int             isaligned;
2725         int             notinit;
2726         int             tryagain;
2727         int             error;
2728
2729         mp = ap->ip->i_mount;
2730         align = (ap->userdata && ap->ip->i_d.di_extsize &&
2731                 (ap->ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE)) ?
2732                 ap->ip->i_d.di_extsize : 0;
2733         if (unlikely(align)) {
2734                 error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2735                                                 align, 0, ap->eof, 0, ap->conv,
2736                                                 &ap->off, &ap->alen);
2737                 ASSERT(!error);
2738                 ASSERT(ap->alen);
2739         }
2740         nullfb = ap->firstblock == NULLFSBLOCK;
2741         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2742         if (nullfb)
2743                 ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2744         else
2745                 ap->rval = ap->firstblock;
2746
2747         xfs_bmap_adjacent(ap);
2748
2749         /*
2750          * If allowed, use ap->rval; otherwise must use firstblock since
2751          * it's in the right allocation group.
2752          */
2753         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->rval) == fb_agno)
2754                 ;
2755         else
2756                 ap->rval = ap->firstblock;
2757         /*
2758          * Normal allocation, done through xfs_alloc_vextent.
2759          */
2760         tryagain = isaligned = 0;
2761         args.tp = ap->tp;
2762         args.mp = mp;
2763         args.fsbno = ap->rval;
2764         args.maxlen = MIN(ap->alen, mp->m_sb.sb_agblocks);
2765         args.firstblock = ap->firstblock;
2766         blen = 0;
2767         if (nullfb) {
2768                 args.type = XFS_ALLOCTYPE_START_BNO;
2769                 args.total = ap->total;
2770                 /*
2771                  * Find the longest available space.
2772                  * We're going to try for the whole allocation at once.
2773                  */
2774                 startag = ag = XFS_FSB_TO_AGNO(mp, args.fsbno);
2775                 notinit = 0;
2776                 down_read(&mp->m_peraglock);
2777                 while (blen < ap->alen) {
2778                         pag = &mp->m_perag[ag];
2779                         if (!pag->pagf_init &&
2780                             (error = xfs_alloc_pagf_init(mp, args.tp,
2781                                     ag, XFS_ALLOC_FLAG_TRYLOCK))) {
2782                                 up_read(&mp->m_peraglock);
2783                                 return error;
2784                         }
2785                         /*
2786                          * See xfs_alloc_fix_freelist...
2787                          */
2788                         if (pag->pagf_init) {
2789                                 need = XFS_MIN_FREELIST_PAG(pag, mp);
2790                                 delta = need > pag->pagf_flcount ?
2791                                         need - pag->pagf_flcount : 0;
2792                                 longest = (pag->pagf_longest > delta) ?
2793                                         (pag->pagf_longest - delta) :
2794                                         (pag->pagf_flcount > 0 ||
2795                                          pag->pagf_longest > 0);
2796                                 if (blen < longest)
2797                                         blen = longest;
2798                         } else
2799                                 notinit = 1;
2800                         if (++ag == mp->m_sb.sb_agcount)
2801                                 ag = 0;
2802                         if (ag == startag)
2803                                 break;
2804                 }
2805                 up_read(&mp->m_peraglock);
2806                 /*
2807                  * Since the above loop did a BUF_TRYLOCK, it is
2808                  * possible that there is space for this request.
2809                  */
2810                 if (notinit || blen < ap->minlen)
2811                         args.minlen = ap->minlen;
2812                 /*
2813                  * If the best seen length is less than the request
2814                  * length, use the best as the minimum.
2815                  */
2816                 else if (blen < ap->alen)
2817                         args.minlen = blen;
2818                 /*
2819                  * Otherwise we've seen an extent as big as alen,
2820                  * use that as the minimum.
2821                  */
2822                 else
2823                         args.minlen = ap->alen;
2824         } else if (ap->low) {
2825                 args.type = XFS_ALLOCTYPE_START_BNO;
2826                 args.total = args.minlen = ap->minlen;
2827         } else {
2828                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2829                 args.total = ap->total;
2830                 args.minlen = ap->minlen;
2831         }
2832         if (unlikely(ap->userdata && ap->ip->i_d.di_extsize &&
2833                     (ap->ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE))) {
2834                 args.prod = ap->ip->i_d.di_extsize;
2835                 if ((args.mod = (xfs_extlen_t)do_mod(ap->off, args.prod)))
2836                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2837         } else if (mp->m_sb.sb_blocksize >= NBPP) {
2838                 args.prod = 1;
2839                 args.mod = 0;
2840         } else {
2841                 args.prod = NBPP >> mp->m_sb.sb_blocklog;
2842                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->off, args.prod))))
2843                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2844         }
2845         /*
2846          * If we are not low on available data blocks, and the
2847          * underlying logical volume manager is a stripe, and
2848          * the file offset is zero then try to allocate data
2849          * blocks on stripe unit boundary.
2850          * NOTE: ap->aeof is only set if the allocation length
2851          * is >= the stripe unit and the allocation offset is
2852          * at the end of file.
2853          */
2854         if (!ap->low && ap->aeof) {
2855                 if (!ap->off) {
2856                         args.alignment = mp->m_dalign;
2857                         atype = args.type;
2858                         isaligned = 1;
2859                         /*
2860                          * Adjust for alignment
2861                          */
2862                         if (blen > args.alignment && blen <= ap->alen)
2863                                 args.minlen = blen - args.alignment;
2864                         args.minalignslop = 0;
2865                 } else {
2866                         /*
2867                          * First try an exact bno allocation.
2868                          * If it fails then do a near or start bno
2869                          * allocation with alignment turned on.
2870                          */
2871                         atype = args.type;
2872                         tryagain = 1;
2873                         args.type = XFS_ALLOCTYPE_THIS_BNO;
2874                         args.alignment = 1;
2875                         /*
2876                          * Compute the minlen+alignment for the
2877                          * next case.  Set slop so that the value
2878                          * of minlen+alignment+slop doesn't go up
2879                          * between the calls.
2880                          */
2881                         if (blen > mp->m_dalign && blen <= ap->alen)
2882                                 nextminlen = blen - mp->m_dalign;
2883                         else
2884                                 nextminlen = args.minlen;
2885                         if (nextminlen + mp->m_dalign > args.minlen + 1)
2886                                 args.minalignslop =
2887                                         nextminlen + mp->m_dalign -
2888                                         args.minlen - 1;
2889                         else
2890                                 args.minalignslop = 0;
2891                 }
2892         } else {
2893                 args.alignment = 1;
2894                 args.minalignslop = 0;
2895         }
2896         args.minleft = ap->minleft;
2897         args.wasdel = ap->wasdel;
2898         args.isfl = 0;
2899         args.userdata = ap->userdata;
2900         if ((error = xfs_alloc_vextent(&args)))
2901                 return error;
2902         if (tryagain && args.fsbno == NULLFSBLOCK) {
2903                 /*
2904                  * Exact allocation failed. Now try with alignment
2905                  * turned on.
2906                  */
2907                 args.type = atype;
2908                 args.fsbno = ap->rval;
2909                 args.alignment = mp->m_dalign;
2910                 args.minlen = nextminlen;
2911                 args.minalignslop = 0;
2912                 isaligned = 1;
2913                 if ((error = xfs_alloc_vextent(&args)))
2914                         return error;
2915         }
2916         if (isaligned && args.fsbno == NULLFSBLOCK) {
2917                 /*
2918                  * allocation failed, so turn off alignment and
2919                  * try again.
2920                  */
2921                 args.type = atype;
2922                 args.fsbno = ap->rval;
2923                 args.alignment = 0;
2924                 if ((error = xfs_alloc_vextent(&args)))
2925                         return error;
2926         }
2927         if (args.fsbno == NULLFSBLOCK && nullfb &&
2928             args.minlen > ap->minlen) {
2929                 args.minlen = ap->minlen;
2930                 args.type = XFS_ALLOCTYPE_START_BNO;
2931                 args.fsbno = ap->rval;
2932                 if ((error = xfs_alloc_vextent(&args)))
2933                         return error;
2934         }
2935         if (args.fsbno == NULLFSBLOCK && nullfb) {
2936                 args.fsbno = 0;
2937                 args.type = XFS_ALLOCTYPE_FIRST_AG;
2938                 args.total = ap->minlen;
2939                 args.minleft = 0;
2940                 if ((error = xfs_alloc_vextent(&args)))
2941                         return error;
2942                 ap->low = 1;
2943         }
2944         if (args.fsbno != NULLFSBLOCK) {
2945                 ap->firstblock = ap->rval = args.fsbno;
2946                 ASSERT(nullfb || fb_agno == args.agno ||
2947                        (ap->low && fb_agno < args.agno));
2948                 ap->alen = args.len;
2949                 ap->ip->i_d.di_nblocks += args.len;
2950                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2951                 if (ap->wasdel)
2952                         ap->ip->i_delayed_blks -= args.len;
2953                 /*
2954                  * Adjust the disk quota also. This was reserved
2955                  * earlier.
2956                  */
2957                 XFS_TRANS_MOD_DQUOT_BYINO(mp, ap->tp, ap->ip,
2958                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2959                                         XFS_TRANS_DQ_BCOUNT,
2960                         (long) args.len);
2961         } else {
2962                 ap->rval = NULLFSBLOCK;
2963                 ap->alen = 0;
2964         }
2965         return 0;
2966 }
2967
2968 /*
2969  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2970  * It figures out where to ask the underlying allocator to put the new extent.
2971  */
2972 STATIC int
2973 xfs_bmap_alloc(
2974         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2975 {
2976         if ((ap->ip->i_d.di_flags & XFS_DIFLAG_REALTIME) && ap->userdata)
2977                 return xfs_bmap_rtalloc(ap);
2978         return xfs_bmap_btalloc(ap);
2979 }
2980
2981 /*
2982  * Transform a btree format file with only one leaf node, where the
2983  * extents list will fit in the inode, into an extents format file.
2984  * Since the file extents are already in-core, all we have to do is
2985  * give up the space for the btree root and pitch the leaf block.
2986  */
2987 STATIC int                              /* error */
2988 xfs_bmap_btree_to_extents(
2989         xfs_trans_t             *tp,    /* transaction pointer */
2990         xfs_inode_t             *ip,    /* incore inode pointer */
2991         xfs_btree_cur_t         *cur,   /* btree cursor */
2992         int                     *logflagsp, /* inode logging flags */
2993         int                     whichfork)  /* data or attr fork */
2994 {
2995         /* REFERENCED */
2996         xfs_bmbt_block_t        *cblock;/* child btree block */
2997         xfs_fsblock_t           cbno;   /* child block number */
2998         xfs_buf_t               *cbp;   /* child block's buffer */
2999         int                     error;  /* error return value */
3000         xfs_ifork_t             *ifp;   /* inode fork data */
3001         xfs_mount_t             *mp;    /* mount point structure */
3002         __be64                  *pp;    /* ptr to block address */
3003         xfs_bmbt_block_t        *rblock;/* root btree block */
3004
3005         ifp = XFS_IFORK_PTR(ip, whichfork);
3006         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3007         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
3008         rblock = ifp->if_broot;
3009         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
3010         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
3011         ASSERT(XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes) == 1);
3012         mp = ip->i_mount;
3013         pp = XFS_BMAP_BROOT_PTR_ADDR(rblock, 1, ifp->if_broot_bytes);
3014         cbno = be64_to_cpu(*pp);
3015         *logflagsp = 0;
3016 #ifdef DEBUG
3017         if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
3018                 return error;
3019 #endif
3020         if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
3021                         XFS_BMAP_BTREE_REF)))
3022                 return error;
3023         cblock = XFS_BUF_TO_BMBT_BLOCK(cbp);
3024         if ((error = xfs_btree_check_lblock(cur, cblock, 0, cbp)))
3025                 return error;
3026         xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
3027         ip->i_d.di_nblocks--;
3028         XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
3029         xfs_trans_binval(tp, cbp);
3030         if (cur->bc_bufs[0] == cbp)
3031                 cur->bc_bufs[0] = NULL;
3032         xfs_iroot_realloc(ip, -1, whichfork);
3033         ASSERT(ifp->if_broot == NULL);
3034         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
3035         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3036         *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
3037         return 0;
3038 }
3039
3040 /*
3041  * Called by xfs_bmapi to update file extent records and the btree
3042  * after removing space (or undoing a delayed allocation).
3043  */
3044 STATIC int                              /* error */
3045 xfs_bmap_del_extent(
3046         xfs_inode_t             *ip,    /* incore inode pointer */
3047         xfs_trans_t             *tp,    /* current transaction pointer */
3048         xfs_extnum_t            idx,    /* extent number to update/delete */
3049         xfs_bmap_free_t         *flist, /* list of extents to be freed */
3050         xfs_btree_cur_t         *cur,   /* if null, not a btree */
3051         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
3052         int                     *logflagsp, /* inode logging flags */
3053         xfs_extdelta_t          *delta, /* Change made to incore extents */
3054         int                     whichfork, /* data or attr fork */
3055         int                     rsvd)   /* OK to allocate reserved blocks */
3056 {
3057         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
3058         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
3059         xfs_fsblock_t           del_endblock=0; /* first block past del */
3060         xfs_fileoff_t           del_endoff;     /* first offset past del */
3061         int                     delay;  /* current block is delayed allocated */
3062         int                     do_fx;  /* free extent at end of routine */
3063         xfs_bmbt_rec_t          *ep;    /* current extent entry pointer */
3064         int                     error;  /* error return value */
3065         int                     flags;  /* inode logging flags */
3066 #ifdef XFS_BMAP_TRACE
3067         static char             fname[] = "xfs_bmap_del_extent";
3068 #endif
3069         xfs_bmbt_irec_t         got;    /* current extent entry */
3070         xfs_fileoff_t           got_endoff;     /* first offset past got */
3071         int                     i;      /* temp state */
3072         xfs_ifork_t             *ifp;   /* inode fork pointer */
3073         xfs_mount_t             *mp;    /* mount structure */
3074         xfs_filblks_t           nblks;  /* quota/sb block count */
3075         xfs_bmbt_irec_t         new;    /* new record to be inserted */
3076         /* REFERENCED */
3077         uint                    qfield; /* quota field to update */
3078         xfs_filblks_t           temp;   /* for indirect length calculations */
3079         xfs_filblks_t           temp2;  /* for indirect length calculations */
3080
3081         XFS_STATS_INC(xs_del_exlist);
3082         mp = ip->i_mount;
3083         ifp = XFS_IFORK_PTR(ip, whichfork);
3084         ASSERT((idx >= 0) && (idx < ifp->if_bytes /
3085                 (uint)sizeof(xfs_bmbt_rec_t)));
3086         ASSERT(del->br_blockcount > 0);
3087         ep = xfs_iext_get_ext(ifp, idx);
3088         xfs_bmbt_get_all(ep, &got);
3089         ASSERT(got.br_startoff <= del->br_startoff);
3090         del_endoff = del->br_startoff + del->br_blockcount;
3091         got_endoff = got.br_startoff + got.br_blockcount;
3092         ASSERT(got_endoff >= del_endoff);
3093         delay = ISNULLSTARTBLOCK(got.br_startblock);
3094         ASSERT(ISNULLSTARTBLOCK(del->br_startblock) == delay);
3095         flags = 0;
3096         qfield = 0;
3097         error = 0;
3098         /*
3099          * If deleting a real allocation, must free up the disk space.
3100          */
3101         if (!delay) {
3102                 flags = XFS_ILOG_CORE;
3103                 /*
3104                  * Realtime allocation.  Free it and record di_nblocks update.
3105                  */
3106                 if (whichfork == XFS_DATA_FORK &&
3107                     (ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
3108                         xfs_fsblock_t   bno;
3109                         xfs_filblks_t   len;
3110
3111                         ASSERT(do_mod(del->br_blockcount,
3112                                       mp->m_sb.sb_rextsize) == 0);
3113                         ASSERT(do_mod(del->br_startblock,
3114                                       mp->m_sb.sb_rextsize) == 0);
3115                         bno = del->br_startblock;
3116                         len = del->br_blockcount;
3117                         do_div(bno, mp->m_sb.sb_rextsize);
3118                         do_div(len, mp->m_sb.sb_rextsize);
3119                         if ((error = xfs_rtfree_extent(ip->i_transp, bno,
3120                                         (xfs_extlen_t)len)))
3121                                 goto done;
3122                         do_fx = 0;
3123                         nblks = len * mp->m_sb.sb_rextsize;
3124                         qfield = XFS_TRANS_DQ_RTBCOUNT;
3125                 }
3126                 /*
3127                  * Ordinary allocation.
3128                  */
3129                 else {
3130                         do_fx = 1;
3131                         nblks = del->br_blockcount;
3132                         qfield = XFS_TRANS_DQ_BCOUNT;
3133                 }
3134                 /*
3135                  * Set up del_endblock and cur for later.
3136                  */
3137                 del_endblock = del->br_startblock + del->br_blockcount;
3138                 if (cur) {
3139                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
3140                                         got.br_startblock, got.br_blockcount,
3141                                         &i)))
3142                                 goto done;
3143                         ASSERT(i == 1);
3144                 }
3145                 da_old = da_new = 0;
3146         } else {
3147                 da_old = STARTBLOCKVAL(got.br_startblock);
3148                 da_new = 0;
3149                 nblks = 0;
3150                 do_fx = 0;
3151         }
3152         /*
3153          * Set flag value to use in switch statement.
3154          * Left-contig is 2, right-contig is 1.
3155          */
3156         switch (((got.br_startoff == del->br_startoff) << 1) |
3157                 (got_endoff == del_endoff)) {
3158         case 3:
3159                 /*
3160                  * Matches the whole extent.  Delete the entry.
3161                  */
3162                 xfs_bmap_trace_delete(fname, "3", ip, idx, 1, whichfork);
3163                 xfs_iext_remove(ifp, idx, 1);
3164                 ifp->if_lastex = idx;
3165                 if (delay)
3166                         break;
3167                 XFS_IFORK_NEXT_SET(ip, whichfork,
3168                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
3169                 flags |= XFS_ILOG_CORE;
3170                 if (!cur) {
3171                         flags |= XFS_ILOG_FEXT(whichfork);
3172                         break;
3173                 }
3174                 if ((error = xfs_bmbt_delete(cur, &i)))
3175                         goto done;
3176                 ASSERT(i == 1);
3177                 break;
3178
3179         case 2:
3180                 /*
3181                  * Deleting the first part of the extent.
3182                  */
3183                 xfs_bmap_trace_pre_update(fname, "2", ip, idx, whichfork);
3184                 xfs_bmbt_set_startoff(ep, del_endoff);
3185                 temp = got.br_blockcount - del->br_blockcount;
3186                 xfs_bmbt_set_blockcount(ep, temp);
3187                 ifp->if_lastex = idx;
3188                 if (delay) {
3189                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
3190                                 da_old);
3191                         xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3192                         xfs_bmap_trace_post_update(fname, "2", ip, idx,
3193                                 whichfork);
3194                         da_new = temp;
3195                         break;
3196                 }
3197                 xfs_bmbt_set_startblock(ep, del_endblock);
3198                 xfs_bmap_trace_post_update(fname, "2", ip, idx, whichfork);
3199                 if (!cur) {
3200                         flags |= XFS_ILOG_FEXT(whichfork);
3201                         break;
3202                 }
3203                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
3204                                 got.br_blockcount - del->br_blockcount,
3205                                 got.br_state)))
3206                         goto done;
3207                 break;
3208
3209         case 1:
3210                 /*
3211                  * Deleting the last part of the extent.
3212                  */
3213                 temp = got.br_blockcount - del->br_blockcount;
3214                 xfs_bmap_trace_pre_update(fname, "1", ip, idx, whichfork);
3215                 xfs_bmbt_set_blockcount(ep, temp);
3216                 ifp->if_lastex = idx;
3217                 if (delay) {
3218                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
3219                                 da_old);
3220                         xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3221                         xfs_bmap_trace_post_update(fname, "1", ip, idx,
3222                                 whichfork);
3223                         da_new = temp;
3224                         break;
3225                 }
3226                 xfs_bmap_trace_post_update(fname, "1", ip, idx, whichfork);
3227                 if (!cur) {
3228                         flags |= XFS_ILOG_FEXT(whichfork);
3229                         break;
3230                 }
3231                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
3232                                 got.br_startblock,
3233                                 got.br_blockcount - del->br_blockcount,
3234                                 got.br_state)))
3235                         goto done;
3236                 break;
3237
3238         case 0:
3239                 /*
3240                  * Deleting the middle of the extent.
3241                  */
3242                 temp = del->br_startoff - got.br_startoff;
3243                 xfs_bmap_trace_pre_update(fname, "0", ip, idx, whichfork);
3244                 xfs_bmbt_set_blockcount(ep, temp);
3245                 new.br_startoff = del_endoff;
3246                 temp2 = got_endoff - del_endoff;
3247                 new.br_blockcount = temp2;
3248                 new.br_state = got.br_state;
3249                 if (!delay) {
3250                         new.br_startblock = del_endblock;
3251                         flags |= XFS_ILOG_CORE;
3252                         if (cur) {
3253                                 if ((error = xfs_bmbt_update(cur,
3254                                                 got.br_startoff,
3255                                                 got.br_startblock, temp,
3256                                                 got.br_state)))
3257                                         goto done;
3258                                 if ((error = xfs_bmbt_increment(cur, 0, &i)))
3259                                         goto done;
3260                                 cur->bc_rec.b = new;
3261                                 error = xfs_bmbt_insert(cur, &i);
3262                                 if (error && error != ENOSPC)
3263                                         goto done;
3264                                 /*
3265                                  * If get no-space back from btree insert,
3266                                  * it tried a split, and we have a zero
3267                                  * block reservation.
3268                                  * Fix up our state and return the error.
3269                                  */
3270                                 if (error == ENOSPC) {
3271                                         /*
3272                                          * Reset the cursor, don't trust
3273                                          * it after any insert operation.
3274                                          */
3275                                         if ((error = xfs_bmbt_lookup_eq(cur,
3276                                                         got.br_startoff,
3277                                                         got.br_startblock,
3278                                                         temp, &i)))
3279                                                 goto done;
3280                                         ASSERT(i == 1);
3281                                         /*
3282                                          * Update the btree record back
3283                                          * to the original value.
3284                                          */
3285                                         if ((error = xfs_bmbt_update(cur,
3286                                                         got.br_startoff,
3287                                                         got.br_startblock,
3288                                                         got.br_blockcount,
3289                                                         got.br_state)))
3290                                                 goto done;
3291                                         /*
3292                                          * Reset the extent record back
3293                                          * to the original value.
3294                                          */
3295                                         xfs_bmbt_set_blockcount(ep,
3296                                                 got.br_blockcount);
3297                                         flags = 0;
3298                                         error = XFS_ERROR(ENOSPC);
3299                                         goto done;
3300                                 }
3301                                 ASSERT(i == 1);
3302                         } else
3303                                 flags |= XFS_ILOG_FEXT(whichfork);
3304                         XFS_IFORK_NEXT_SET(ip, whichfork,
3305                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
3306                 } else {
3307                         ASSERT(whichfork == XFS_DATA_FORK);
3308                         temp = xfs_bmap_worst_indlen(ip, temp);
3309                         xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3310                         temp2 = xfs_bmap_worst_indlen(ip, temp2);
3311                         new.br_startblock = NULLSTARTBLOCK((int)temp2);
3312                         da_new = temp + temp2;
3313                         while (da_new > da_old) {
3314                                 if (temp) {
3315                                         temp--;
3316                                         da_new--;
3317                                         xfs_bmbt_set_startblock(ep,
3318                                                 NULLSTARTBLOCK((int)temp));
3319                                 }
3320                                 if (da_new == da_old)
3321                                         break;
3322                                 if (temp2) {
3323                                         temp2--;
3324                                         da_new--;
3325                                         new.br_startblock =
3326                                                 NULLSTARTBLOCK((int)temp2);
3327                                 }
3328                         }
3329                 }
3330                 xfs_bmap_trace_post_update(fname, "0", ip, idx, whichfork);
3331                 xfs_bmap_trace_insert(fname, "0", ip, idx + 1, 1, &new, NULL,
3332                         whichfork);
3333                 xfs_iext_insert(ifp, idx + 1, 1, &new);
3334                 ifp->if_lastex = idx + 1;
3335                 break;
3336         }
3337         /*
3338          * If we need to, add to list of extents to delete.
3339          */
3340         if (do_fx)
3341                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
3342                         mp);
3343         /*
3344          * Adjust inode # blocks in the file.
3345          */
3346         if (nblks)
3347                 ip->i_d.di_nblocks -= nblks;
3348         /*
3349          * Adjust quota data.
3350          */
3351         if (qfield)
3352                 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, qfield, (long)-nblks);
3353
3354         /*
3355          * Account for change in delayed indirect blocks.
3356          * Nothing to do for disk quota accounting here.
3357          */
3358         ASSERT(da_old >= da_new);
3359         if (da_old > da_new)
3360                 xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS, (int)(da_old - da_new),
3361                         rsvd);
3362         if (delta) {
3363                 /* DELTA: report the original extent. */
3364                 if (delta->xed_startoff > got.br_startoff)
3365                         delta->xed_startoff = got.br_startoff;
3366                 if (delta->xed_blockcount < got.br_startoff+got.br_blockcount)
3367                         delta->xed_blockcount = got.br_startoff +
3368                                                         got.br_blockcount;
3369         }
3370 done:
3371         *logflagsp = flags;
3372         return error;
3373 }
3374
3375 /*
3376  * Remove the entry "free" from the free item list.  Prev points to the
3377  * previous entry, unless "free" is the head of the list.
3378  */
3379 STATIC void
3380 xfs_bmap_del_free(
3381         xfs_bmap_free_t         *flist, /* free item list header */
3382         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
3383         xfs_bmap_free_item_t    *free)  /* list item to be freed */
3384 {
3385         if (prev)
3386                 prev->xbfi_next = free->xbfi_next;
3387         else
3388                 flist->xbf_first = free->xbfi_next;
3389         flist->xbf_count--;
3390         kmem_zone_free(xfs_bmap_free_item_zone, free);
3391 }
3392
3393 /*
3394  * Convert an extents-format file into a btree-format file.
3395  * The new file will have a root block (in the inode) and a single child block.
3396  */
3397 STATIC int                                      /* error */
3398 xfs_bmap_extents_to_btree(
3399         xfs_trans_t             *tp,            /* transaction pointer */
3400         xfs_inode_t             *ip,            /* incore inode pointer */
3401         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
3402         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
3403         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
3404         int                     wasdel,         /* converting a delayed alloc */
3405         int                     *logflagsp,     /* inode logging flags */
3406         int                     whichfork)      /* data or attr fork */
3407 {
3408         xfs_bmbt_block_t        *ablock;        /* allocated (child) bt block */
3409         xfs_buf_t               *abp;           /* buffer for ablock */
3410         xfs_alloc_arg_t         args;           /* allocation arguments */
3411         xfs_bmbt_rec_t          *arp;           /* child record pointer */
3412         xfs_bmbt_block_t        *block;         /* btree root block */
3413         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
3414         xfs_bmbt_rec_t          *ep;            /* extent record pointer */
3415         int                     error;          /* error return value */
3416         xfs_extnum_t            i, cnt;         /* extent record index */
3417         xfs_ifork_t             *ifp;           /* inode fork pointer */
3418         xfs_bmbt_key_t          *kp;            /* root block key pointer */
3419         xfs_mount_t             *mp;            /* mount structure */
3420         xfs_extnum_t            nextents;       /* number of file extents */
3421         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
3422
3423         ifp = XFS_IFORK_PTR(ip, whichfork);
3424         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3425         ASSERT(ifp->if_ext_max ==
3426                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
3427         /*
3428          * Make space in the inode incore.
3429          */
3430         xfs_iroot_realloc(ip, 1, whichfork);
3431         ifp->if_flags |= XFS_IFBROOT;
3432         /*
3433          * Fill in the root.
3434          */
3435         block = ifp->if_broot;
3436         block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3437         block->bb_level = cpu_to_be16(1);
3438         block->bb_numrecs = cpu_to_be16(1);
3439         block->bb_leftsib = cpu_to_be64(NULLDFSBNO);
3440         block->bb_rightsib = cpu_to_be64(NULLDFSBNO);
3441         /*
3442          * Need a cursor.  Can't allocate until bb_level is filled in.
3443          */
3444         mp = ip->i_mount;
3445         cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
3446                 whichfork);
3447         cur->bc_private.b.firstblock = *firstblock;
3448         cur->bc_private.b.flist = flist;
3449         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3450         /*
3451          * Convert to a btree with two levels, one record in root.
3452          */
3453         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3454         args.tp = tp;
3455         args.mp = mp;
3456         args.firstblock = *firstblock;
3457         if (*firstblock == NULLFSBLOCK) {
3458                 args.type = XFS_ALLOCTYPE_START_BNO;
3459                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3460         } else if (flist->xbf_low) {
3461                 args.type = XFS_ALLOCTYPE_START_BNO;
3462                 args.fsbno = *firstblock;
3463         } else {
3464                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3465                 args.fsbno = *firstblock;
3466         }
3467         args.minlen = args.maxlen = args.prod = 1;
3468         args.total = args.minleft = args.alignment = args.mod = args.isfl =
3469                 args.minalignslop = 0;
3470         args.wasdel = wasdel;
3471         *logflagsp = 0;
3472         if ((error = xfs_alloc_vextent(&args))) {
3473                 xfs_iroot_realloc(ip, -1, whichfork);
3474                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3475                 return error;
3476         }
3477         /*
3478          * Allocation can't fail, the space was reserved.
3479          */
3480         ASSERT(args.fsbno != NULLFSBLOCK);
3481         ASSERT(*firstblock == NULLFSBLOCK ||
3482                args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3483                (flist->xbf_low &&
3484                 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3485         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3486         cur->bc_private.b.allocated++;
3487         ip->i_d.di_nblocks++;
3488         XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3489         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3490         /*
3491          * Fill in the child block.
3492          */
3493         ablock = XFS_BUF_TO_BMBT_BLOCK(abp);
3494         ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3495         ablock->bb_level = 0;
3496         ablock->bb_leftsib = cpu_to_be64(NULLDFSBNO);
3497         ablock->bb_rightsib = cpu_to_be64(NULLDFSBNO);
3498         arp = XFS_BMAP_REC_IADDR(ablock, 1, cur);
3499         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3500         for (cnt = i = 0; i < nextents; i++) {
3501                 ep = xfs_iext_get_ext(ifp, i);
3502                 if (!ISNULLSTARTBLOCK(xfs_bmbt_get_startblock(ep))) {
3503                         arp->l0 = INT_GET(ep->l0, ARCH_CONVERT);
3504                         arp->l1 = INT_GET(ep->l1, ARCH_CONVERT);
3505                         arp++; cnt++;
3506                 }
3507         }
3508         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3509         ablock->bb_numrecs = cpu_to_be16(cnt);
3510         /*
3511          * Fill in the root key and pointer.
3512          */
3513         kp = XFS_BMAP_KEY_IADDR(block, 1, cur);
3514         arp = XFS_BMAP_REC_IADDR(ablock, 1, cur);
3515         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
3516         pp = XFS_BMAP_PTR_IADDR(block, 1, cur);
3517         *pp = cpu_to_be64(args.fsbno);
3518         /*
3519          * Do all this logging at the end so that
3520          * the root is at the right level.
3521          */
3522         xfs_bmbt_log_block(cur, abp, XFS_BB_ALL_BITS);
3523         xfs_bmbt_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3524         ASSERT(*curp == NULL);
3525         *curp = cur;
3526         *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FBROOT(whichfork);
3527         return 0;
3528 }
3529
3530 /*
3531  * Helper routine to reset inode di_forkoff field when switching
3532  * attribute fork from local to extent format - we reset it where
3533  * possible to make space available for inline data fork extents.
3534  */
3535 STATIC void
3536 xfs_bmap_forkoff_reset(
3537         xfs_mount_t     *mp,
3538         xfs_inode_t     *ip,
3539         int             whichfork)
3540 {
3541         if (whichfork == XFS_ATTR_FORK &&
3542             (ip->i_d.di_format != XFS_DINODE_FMT_DEV) &&
3543             (ip->i_d.di_format != XFS_DINODE_FMT_UUID) &&
3544             ((mp->m_attroffset >> 3) > ip->i_d.di_forkoff)) {
3545                 ip->i_d.di_forkoff = mp->m_attroffset >> 3;
3546                 ip->i_df.if_ext_max = XFS_IFORK_DSIZE(ip) /
3547                                         (uint)sizeof(xfs_bmbt_rec_t);
3548                 ip->i_afp->if_ext_max = XFS_IFORK_ASIZE(ip) /
3549                                         (uint)sizeof(xfs_bmbt_rec_t);
3550         }
3551 }
3552
3553 /*
3554  * Convert a local file to an extents file.
3555  * This code is out of bounds for data forks of regular files,
3556  * since the file data needs to get logged so things will stay consistent.
3557  * (The bmap-level manipulations are ok, though).
3558  */
3559 STATIC int                              /* error */
3560 xfs_bmap_local_to_extents(
3561         xfs_trans_t     *tp,            /* transaction pointer */
3562         xfs_inode_t     *ip,            /* incore inode pointer */
3563         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
3564         xfs_extlen_t    total,          /* total blocks needed by transaction */
3565         int             *logflagsp,     /* inode logging flags */
3566         int             whichfork)      /* data or attr fork */
3567 {
3568         int             error;          /* error return value */
3569         int             flags;          /* logging flags returned */
3570 #ifdef XFS_BMAP_TRACE
3571         static char     fname[] = "xfs_bmap_local_to_extents";
3572 #endif
3573         xfs_ifork_t     *ifp;           /* inode fork pointer */
3574
3575         /*
3576          * We don't want to deal with the case of keeping inode data inline yet.
3577          * So sending the data fork of a regular inode is invalid.
3578          */
3579         ASSERT(!((ip->i_d.di_mode & S_IFMT) == S_IFREG &&
3580                  whichfork == XFS_DATA_FORK));
3581         ifp = XFS_IFORK_PTR(ip, whichfork);
3582         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3583         flags = 0;
3584         error = 0;
3585         if (ifp->if_bytes) {
3586                 xfs_alloc_arg_t args;   /* allocation arguments */
3587                 xfs_buf_t       *bp;    /* buffer for extent block */
3588                 xfs_bmbt_rec_t  *ep;    /* extent record pointer */
3589
3590                 args.tp = tp;
3591                 args.mp = ip->i_mount;
3592                 args.firstblock = *firstblock;
3593                 ASSERT((ifp->if_flags &
3594                         (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == XFS_IFINLINE);
3595                 /*
3596                  * Allocate a block.  We know we need only one, since the
3597                  * file currently fits in an inode.
3598                  */
3599                 if (*firstblock == NULLFSBLOCK) {
3600                         args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3601                         args.type = XFS_ALLOCTYPE_START_BNO;
3602                 } else {
3603                         args.fsbno = *firstblock;
3604                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
3605                 }
3606                 args.total = total;
3607                 args.mod = args.minleft = args.alignment = args.wasdel =
3608                         args.isfl = args.minalignslop = 0;
3609                 args.minlen = args.maxlen = args.prod = 1;
3610                 if ((error = xfs_alloc_vextent(&args)))
3611                         goto done;
3612                 /*
3613                  * Can't fail, the space was reserved.
3614                  */
3615                 ASSERT(args.fsbno != NULLFSBLOCK);
3616                 ASSERT(args.len == 1);
3617                 *firstblock = args.fsbno;
3618                 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3619                 memcpy((char *)XFS_BUF_PTR(bp), ifp->if_u1.if_data,
3620                         ifp->if_bytes);
3621                 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3622                 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3623                 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3624                 xfs_iext_add(ifp, 0, 1);
3625                 ep = xfs_iext_get_ext(ifp, 0);
3626                 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3627                 xfs_bmap_trace_post_update(fname, "new", ip, 0, whichfork);
3628                 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3629                 ip->i_d.di_nblocks = 1;
3630                 XFS_TRANS_MOD_DQUOT_BYINO(args.mp, tp, ip,
3631                         XFS_TRANS_DQ_BCOUNT, 1L);
3632                 flags |= XFS_ILOG_FEXT(whichfork);
3633         } else {
3634                 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3635                 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3636         }
3637         ifp->if_flags &= ~XFS_IFINLINE;
3638         ifp->if_flags |= XFS_IFEXTENTS;
3639         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3640         flags |= XFS_ILOG_CORE;
3641 done:
3642         *logflagsp = flags;
3643         return error;
3644 }
3645
3646 /*
3647  * Search the extent records for the entry containing block bno.
3648  * If bno lies in a hole, point to the next entry.  If bno lies
3649  * past eof, *eofp will be set, and *prevp will contain the last
3650  * entry (null if none).  Else, *lastxp will be set to the index
3651  * of the found entry; *gotp will contain the entry.
3652  */
3653 xfs_bmbt_rec_t *                        /* pointer to found extent entry */
3654 xfs_bmap_search_multi_extents(
3655         xfs_ifork_t     *ifp,           /* inode fork pointer */
3656         xfs_fileoff_t   bno,            /* block number searched for */
3657         int             *eofp,          /* out: end of file found */
3658         xfs_extnum_t    *lastxp,        /* out: last extent index */
3659         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3660         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3661 {
3662         xfs_bmbt_rec_t  *ep;            /* extent record pointer */
3663         xfs_extnum_t    lastx;          /* last extent index */
3664
3665         /*
3666          * Initialize the extent entry structure to catch access to
3667          * uninitialized br_startblock field.
3668          */
3669         gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
3670         gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3671         gotp->br_state = XFS_EXT_INVALID;
3672 #if XFS_BIG_BLKNOS
3673         gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
3674 #else
3675         gotp->br_startblock = 0xffffa5a5;
3676 #endif
3677         prevp->br_startoff = NULLFILEOFF;
3678
3679         ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
3680         if (lastx > 0) {
3681                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
3682         }
3683         if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
3684                 xfs_bmbt_get_all(ep, gotp);
3685                 *eofp = 0;
3686         } else {
3687                 if (lastx > 0) {
3688                         *gotp = *prevp;
3689                 }
3690                 *eofp = 1;
3691                 ep = NULL;
3692         }
3693         *lastxp = lastx;
3694         return ep;
3695 }
3696
3697 /*
3698  * Search the extents list for the inode, for the extent containing bno.
3699  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
3700  * *eofp will be set, and *prevp will contain the last entry (null if none).
3701  * Else, *lastxp will be set to the index of the found
3702  * entry; *gotp will contain the entry.
3703  */
3704 STATIC xfs_bmbt_rec_t *                 /* pointer to found extent entry */
3705 xfs_bmap_search_extents(
3706         xfs_inode_t     *ip,            /* incore inode pointer */
3707         xfs_fileoff_t   bno,            /* block number searched for */
3708         int             fork,           /* data or attr fork */
3709         int             *eofp,          /* out: end of file found */
3710         xfs_extnum_t    *lastxp,        /* out: last extent index */
3711         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3712         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3713 {
3714         xfs_ifork_t     *ifp;           /* inode fork pointer */
3715         xfs_bmbt_rec_t  *ep;            /* extent record pointer */
3716
3717         XFS_STATS_INC(xs_look_exlist);
3718         ifp = XFS_IFORK_PTR(ip, fork);
3719
3720         ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
3721
3722         if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
3723                      !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
3724                 xfs_cmn_err(XFS_PTAG_FSBLOCK_ZERO, CE_ALERT, ip->i_mount,
3725                                 "Access to block zero in inode %llu "
3726                                 "start_block: %llx start_off: %llx "
3727                                 "blkcnt: %llx extent-state: %x lastx: %x\n",
3728                         (unsigned long long)ip->i_ino,
3729                         (unsigned long long)gotp->br_startblock,
3730                         (unsigned long long)gotp->br_startoff,
3731                         (unsigned long long)gotp->br_blockcount,
3732                         gotp->br_state, *lastxp);
3733                 *lastxp = NULLEXTNUM;
3734                 *eofp = 1;
3735                 return NULL;
3736         }
3737         return ep;
3738 }
3739
3740
3741 #ifdef XFS_BMAP_TRACE
3742 ktrace_t        *xfs_bmap_trace_buf;
3743
3744 /*
3745  * Add a bmap trace buffer entry.  Base routine for the others.
3746  */
3747 STATIC void
3748 xfs_bmap_trace_addentry(
3749         int             opcode,         /* operation */
3750         char            *fname,         /* function name */
3751         char            *desc,          /* operation description */
3752         xfs_inode_t     *ip,            /* incore inode pointer */
3753         xfs_extnum_t    idx,            /* index of entry(ies) */
3754         xfs_extnum_t    cnt,            /* count of entries, 1 or 2 */
3755         xfs_bmbt_rec_t  *r1,            /* first record */
3756         xfs_bmbt_rec_t  *r2,            /* second record or null */
3757         int             whichfork)      /* data or attr fork */
3758 {
3759         xfs_bmbt_rec_t  tr2;
3760
3761         ASSERT(cnt == 1 || cnt == 2);
3762         ASSERT(r1 != NULL);
3763         if (cnt == 1) {
3764                 ASSERT(r2 == NULL);
3765                 r2 = &tr2;
3766                 memset(&tr2, 0, sizeof(tr2));
3767         } else
3768                 ASSERT(r2 != NULL);
3769         ktrace_enter(xfs_bmap_trace_buf,
3770                 (void *)(__psint_t)(opcode | (whichfork << 16)),
3771                 (void *)fname, (void *)desc, (void *)ip,
3772                 (void *)(__psint_t)idx,
3773                 (void *)(__psint_t)cnt,
3774                 (void *)(__psunsigned_t)(ip->i_ino >> 32),
3775                 (void *)(__psunsigned_t)(unsigned)ip->i_ino,
3776                 (void *)(__psunsigned_t)(r1->l0 >> 32),
3777                 (void *)(__psunsigned_t)(unsigned)(r1->l0),
3778                 (void *)(__psunsigned_t)(r1->l1 >> 32),
3779                 (void *)(__psunsigned_t)(unsigned)(r1->l1),
3780                 (void *)(__psunsigned_t)(r2->l0 >> 32),
3781                 (void *)(__psunsigned_t)(unsigned)(r2->l0),
3782                 (void *)(__psunsigned_t)(r2->l1 >> 32),
3783                 (void *)(__psunsigned_t)(unsigned)(r2->l1)
3784                 );
3785         ASSERT(ip->i_xtrace);
3786         ktrace_enter(ip->i_xtrace,
3787                 (void *)(__psint_t)(opcode | (whichfork << 16)),
3788                 (void *)fname, (void *)desc, (void *)ip,
3789                 (void *)(__psint_t)idx,
3790                 (void *)(__psint_t)cnt,
3791                 (void *)(__psunsigned_t)(ip->i_ino >> 32),
3792                 (void *)(__psunsigned_t)(unsigned)ip->i_ino,
3793                 (void *)(__psunsigned_t)(r1->l0 >> 32),
3794                 (void *)(__psunsigned_t)(unsigned)(r1->l0),
3795                 (void *)(__psunsigned_t)(r1->l1 >> 32),
3796                 (void *)(__psunsigned_t)(unsigned)(r1->l1),
3797                 (void *)(__psunsigned_t)(r2->l0 >> 32),
3798                 (void *)(__psunsigned_t)(unsigned)(r2->l0),
3799                 (void *)(__psunsigned_t)(r2->l1 >> 32),
3800                 (void *)(__psunsigned_t)(unsigned)(r2->l1)
3801                 );
3802 }
3803
3804 /*
3805  * Add bmap trace entry prior to a call to xfs_iext_remove.
3806  */
3807 STATIC void
3808 xfs_bmap_trace_delete(
3809         char            *fname,         /* function name */
3810         char            *desc,          /* operation description */
3811         xfs_inode_t     *ip,            /* incore inode pointer */
3812         xfs_extnum_t    idx,            /* index of entry(entries) deleted */
3813         xfs_extnum_t    cnt,            /* count of entries deleted, 1 or 2 */
3814         int             whichfork)      /* data or attr fork */
3815 {
3816         xfs_ifork_t     *ifp;           /* inode fork pointer */
3817
3818         ifp = XFS_IFORK_PTR(ip, whichfork);
3819         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_DELETE, fname, desc, ip, idx,
3820                 cnt, xfs_iext_get_ext(ifp, idx),
3821                 cnt == 2 ? xfs_iext_get_ext(ifp, idx + 1) : NULL,
3822                 whichfork);
3823 }
3824
3825 /*
3826  * Add bmap trace entry prior to a call to xfs_iext_insert, or
3827  * reading in the extents list from the disk (in the btree).
3828  */
3829 STATIC void
3830 xfs_bmap_trace_insert(
3831         char            *fname,         /* function name */
3832         char            *desc,          /* operation description */
3833         xfs_inode_t     *ip,            /* incore inode pointer */
3834         xfs_extnum_t    idx,            /* index of entry(entries) inserted */
3835         xfs_extnum_t    cnt,            /* count of entries inserted, 1 or 2 */
3836         xfs_bmbt_irec_t *r1,            /* inserted record 1 */
3837         xfs_bmbt_irec_t *r2,            /* inserted record 2 or null */
3838         int             whichfork)      /* data or attr fork */
3839 {
3840         xfs_bmbt_rec_t  tr1;            /* compressed record 1 */
3841         xfs_bmbt_rec_t  tr2;            /* compressed record 2 if needed */
3842
3843         xfs_bmbt_set_all(&tr1, r1);
3844         if (cnt == 2) {
3845                 ASSERT(r2 != NULL);
3846                 xfs_bmbt_set_all(&tr2, r2);
3847         } else {
3848                 ASSERT(cnt == 1);
3849                 ASSERT(r2 == NULL);
3850         }
3851         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_INSERT, fname, desc, ip, idx,
3852                 cnt, &tr1, cnt == 2 ? &tr2 : NULL, whichfork);
3853 }
3854
3855 /*
3856  * Add bmap trace entry after updating an extent record in place.
3857  */
3858 STATIC void
3859 xfs_bmap_trace_post_update(
3860         char            *fname,         /* function name */
3861         char            *desc,          /* operation description */
3862         xfs_inode_t     *ip,            /* incore inode pointer */
3863         xfs_extnum_t    idx,            /* index of entry updated */
3864         int             whichfork)      /* data or attr fork */
3865 {
3866         xfs_ifork_t     *ifp;           /* inode fork pointer */
3867
3868         ifp = XFS_IFORK_PTR(ip, whichfork);
3869         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_POST_UP, fname, desc, ip, idx,
3870                 1, xfs_iext_get_ext(ifp, idx), NULL, whichfork);
3871 }
3872
3873 /*
3874  * Add bmap trace entry prior to updating an extent record in place.
3875  */
3876 STATIC void
3877 xfs_bmap_trace_pre_update(
3878         char            *fname,         /* function name */
3879         char            *desc,          /* operation description */
3880         xfs_inode_t     *ip,            /* incore inode pointer */
3881         xfs_extnum_t    idx,            /* index of entry to be updated */
3882         int             whichfork)      /* data or attr fork */
3883 {
3884         xfs_ifork_t     *ifp;           /* inode fork pointer */
3885
3886         ifp = XFS_IFORK_PTR(ip, whichfork);
3887         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_PRE_UP, fname, desc, ip, idx, 1,
3888                 xfs_iext_get_ext(ifp, idx), NULL, whichfork);
3889 }
3890 #endif  /* XFS_BMAP_TRACE */
3891
3892 /*
3893  * Compute the worst-case number of indirect blocks that will be used
3894  * for ip's delayed extent of length "len".
3895  */
3896 STATIC xfs_filblks_t
3897 xfs_bmap_worst_indlen(
3898         xfs_inode_t     *ip,            /* incore inode pointer */
3899         xfs_filblks_t   len)            /* delayed extent length */
3900 {
3901         int             level;          /* btree level number */
3902         int             maxrecs;        /* maximum record count at this level */
3903         xfs_mount_t     *mp;            /* mount structure */
3904         xfs_filblks_t   rval;           /* return value */
3905
3906         mp = ip->i_mount;
3907         maxrecs = mp->m_bmap_dmxr[0];
3908         for (level = 0, rval = 0;
3909              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3910              level++) {
3911                 len += maxrecs - 1;
3912                 do_div(len, maxrecs);
3913                 rval += len;
3914                 if (len == 1)
3915                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3916                                 level - 1;
3917                 if (level == 0)
3918                         maxrecs = mp->m_bmap_dmxr[1];
3919         }
3920         return rval;
3921 }
3922
3923 #if defined(XFS_RW_TRACE)
3924 STATIC void
3925 xfs_bunmap_trace(
3926         xfs_inode_t             *ip,
3927         xfs_fileoff_t           bno,
3928         xfs_filblks_t           len,
3929         int                     flags,
3930         inst_t                  *ra)
3931 {
3932         if (ip->i_rwtrace == NULL)
3933                 return;
3934         ktrace_enter(ip->i_rwtrace,
3935                 (void *)(__psint_t)XFS_BUNMAP,
3936                 (void *)ip,
3937                 (void *)(__psint_t)((ip->i_d.di_size >> 32) & 0xffffffff),
3938                 (void *)(__psint_t)(ip->i_d.di_size & 0xffffffff),
3939                 (void *)(__psint_t)(((xfs_dfiloff_t)bno >> 32) & 0xffffffff),
3940                 (void *)(__psint_t)((xfs_dfiloff_t)bno & 0xffffffff),
3941                 (void *)(__psint_t)len,
3942                 (void *)(__psint_t)flags,
3943                 (void *)(unsigned long)current_cpu(),
3944                 (void *)ra,
3945                 (void *)0,
3946                 (void *)0,
3947                 (void *)0,
3948                 (void *)0,
3949                 (void *)0,
3950                 (void *)0);
3951 }
3952 #endif
3953
3954 /*
3955  * Convert inode from non-attributed to attributed.
3956  * Must not be in a transaction, ip must not be locked.
3957  */
3958 int                                             /* error code */
3959 xfs_bmap_add_attrfork(
3960         xfs_inode_t             *ip,            /* incore inode pointer */
3961         int                     size,           /* space new attribute needs */
3962         int                     rsvd)           /* xact may use reserved blks */
3963 {
3964         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
3965         xfs_bmap_free_t         flist;          /* freed extent records */
3966         xfs_mount_t             *mp;            /* mount structure */
3967         xfs_trans_t             *tp;            /* transaction pointer */
3968         unsigned long           s;              /* spinlock spl value */
3969         int                     blks;           /* space reservation */
3970         int                     version = 1;    /* superblock attr version */
3971         int                     committed;      /* xaction was committed */
3972         int                     logflags;       /* logging flags */
3973         int                     error;          /* error return value */
3974
3975         ASSERT(XFS_IFORK_Q(ip) == 0);
3976         ASSERT(ip->i_df.if_ext_max ==
3977                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3978
3979         mp = ip->i_mount;
3980         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3981         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3982         blks = XFS_ADDAFORK_SPACE_RES(mp);
3983         if (rsvd)
3984                 tp->t_flags |= XFS_TRANS_RESERVE;
3985         if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3986                         XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3987                 goto error0;
3988         xfs_ilock(ip, XFS_ILOCK_EXCL);
3989         error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip, blks, 0, rsvd ?
3990                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3991                         XFS_QMOPT_RES_REGBLKS);
3992         if (error) {
3993                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3994                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3995                 return error;
3996         }
3997         if (XFS_IFORK_Q(ip))
3998                 goto error1;
3999         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
4000                 /*
4001                  * For inodes coming from pre-6.2 filesystems.
4002                  */
4003                 ASSERT(ip->i_d.di_aformat == 0);
4004                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
4005         }
4006         ASSERT(ip->i_d.di_anextents == 0);
4007         VN_HOLD(XFS_ITOV(ip));
4008         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4009         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4010         switch (ip->i_d.di_format) {
4011         case XFS_DINODE_FMT_DEV:
4012                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
4013                 break;
4014         case XFS_DINODE_FMT_UUID:
4015                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
4016                 break;
4017         case XFS_DINODE_FMT_LOCAL:
4018         case XFS_DINODE_FMT_EXTENTS:
4019         case XFS_DINODE_FMT_BTREE:
4020                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
4021                 if (!ip->i_d.di_forkoff)
4022                         ip->i_d.di_forkoff = mp->m_attroffset >> 3;
4023                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
4024                         version = 2;
4025                 break;
4026         default:
4027                 ASSERT(0);
4028                 error = XFS_ERROR(EINVAL);
4029                 goto error1;
4030         }
4031         ip->i_df.if_ext_max =
4032                 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
4033         ASSERT(ip->i_afp == NULL);
4034         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
4035         ip->i_afp->if_ext_max =
4036                 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
4037         ip->i_afp->if_flags = XFS_IFEXTENTS;
4038         logflags = 0;
4039         XFS_BMAP_INIT(&flist, &firstblock);
4040         switch (ip->i_d.di_format) {
4041         case XFS_DINODE_FMT_LOCAL:
4042                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
4043                         &logflags);
4044                 break;
4045         case XFS_DINODE_FMT_EXTENTS:
4046                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
4047                         &flist, &logflags);
4048                 break;
4049         case XFS_DINODE_FMT_BTREE:
4050                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
4051                         &logflags);
4052                 break;
4053         default:
4054                 error = 0;
4055                 break;
4056         }
4057         if (logflags)
4058                 xfs_trans_log_inode(tp, ip, logflags);
4059         if (error)
4060                 goto error2;
4061         if (!XFS_SB_VERSION_HASATTR(&mp->m_sb) ||
4062            (!XFS_SB_VERSION_HASATTR2(&mp->m_sb) && version == 2)) {
4063                 __int64_t sbfields = 0;
4064
4065                 s = XFS_SB_LOCK(mp);
4066                 if (!XFS_SB_VERSION_HASATTR(&mp->m_sb)) {
4067                         XFS_SB_VERSION_ADDATTR(&mp->m_sb);
4068                         sbfields |= XFS_SB_VERSIONNUM;
4069                 }
4070                 if (!XFS_SB_VERSION_HASATTR2(&mp->m_sb) && version == 2) {
4071                         XFS_SB_VERSION_ADDATTR2(&mp->m_sb);
4072                         sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
4073                 }
4074                 if (sbfields) {
4075                         XFS_SB_UNLOCK(mp, s);
4076                         xfs_mod_sb(tp, sbfields);
4077                 } else
4078                         XFS_SB_UNLOCK(mp, s);
4079         }
4080         if ((error = xfs_bmap_finish(&tp, &flist, firstblock, &committed)))
4081                 goto error2;
4082         error = xfs_trans_commit(tp, XFS_TRANS_PERM_LOG_RES, NULL);
4083         ASSERT(ip->i_df.if_ext_max ==
4084                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
4085         return error;
4086 error2:
4087         xfs_bmap_cancel(&flist);
4088 error1:
4089         ASSERT(ismrlocked(&ip->i_lock,MR_UPDATE));
4090         xfs_iunlock(ip, XFS_ILOCK_EXCL);
4091 error0:
4092         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
4093         ASSERT(ip->i_df.if_ext_max ==
4094                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
4095         return error;
4096 }
4097
4098 /*
4099  * Add the extent to the list of extents to be free at transaction end.
4100  * The list is maintained sorted (by block number).
4101  */
4102 /* ARGSUSED */
4103 void
4104 xfs_bmap_add_free(
4105         xfs_fsblock_t           bno,            /* fs block number of extent */
4106         xfs_filblks_t           len,            /* length of extent */
4107         xfs_bmap_free_t         *flist,         /* list of extents */
4108         xfs_mount_t             *mp)            /* mount point structure */
4109 {
4110         xfs_bmap_free_item_t    *cur;           /* current (next) element */
4111         xfs_bmap_free_item_t    *new;           /* new element */
4112         xfs_bmap_free_item_t    *prev;          /* previous element */
4113 #ifdef DEBUG
4114         xfs_agnumber_t          agno;
4115         xfs_agblock_t           agbno;
4116
4117         ASSERT(bno != NULLFSBLOCK);
4118         ASSERT(len > 0);
4119         ASSERT(len <= MAXEXTLEN);
4120         ASSERT(!ISNULLSTARTBLOCK(bno));
4121         agno = XFS_FSB_TO_AGNO(mp, bno);
4122         agbno = XFS_FSB_TO_AGBNO(mp, bno);
4123         ASSERT(agno < mp->m_sb.sb_agcount);
4124         ASSERT(agbno < mp->m_sb.sb_agblocks);
4125         ASSERT(len < mp->m_sb.sb_agblocks);
4126         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
4127 #endif
4128         ASSERT(xfs_bmap_free_item_zone != NULL);
4129         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
4130         new->xbfi_startblock = bno;
4131         new->xbfi_blockcount = (xfs_extlen_t)len;
4132         for (prev = NULL, cur = flist->xbf_first;
4133              cur != NULL;
4134              prev = cur, cur = cur->xbfi_next) {
4135                 if (cur->xbfi_startblock >= bno)
4136                         break;
4137         }
4138         if (prev)
4139                 prev->xbfi_next = new;
4140         else
4141                 flist->xbf_first = new;
4142         new->xbfi_next = cur;
4143         flist->xbf_count++;
4144 }
4145
4146 /*
4147  * Compute and fill in the value of the maximum depth of a bmap btree
4148  * in this filesystem.  Done once, during mount.
4149  */
4150 void
4151 xfs_bmap_compute_maxlevels(
4152         xfs_mount_t     *mp,            /* file system mount structure */
4153         int             whichfork)      /* data or attr fork */
4154 {
4155         int             level;          /* btree level */
4156         uint            maxblocks;      /* max blocks at this level */
4157         uint            maxleafents;    /* max leaf entries possible */
4158         int             maxrootrecs;    /* max records in root block */
4159         int             minleafrecs;    /* min records in leaf block */
4160         int             minnoderecs;    /* min records in node block */
4161         int             sz;             /* root block size */
4162
4163         /*
4164          * The maximum number of extents in a file, hence the maximum
4165          * number of leaf entries, is controlled by the type of di_nextents
4166          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
4167          * (a signed 16-bit number, xfs_aextnum_t).
4168          */
4169         if (whichfork == XFS_DATA_FORK) {
4170                 maxleafents = MAXEXTNUM;
4171                 sz = (mp->m_flags & XFS_MOUNT_ATTR2) ?
4172                         XFS_BMDR_SPACE_CALC(MINDBTPTRS) : mp->m_attroffset;
4173         } else {
4174                 maxleafents = MAXAEXTNUM;
4175                 sz = (mp->m_flags & XFS_MOUNT_ATTR2) ?
4176                         XFS_BMDR_SPACE_CALC(MINABTPTRS) :
4177                         mp->m_sb.sb_inodesize - mp->m_attroffset;
4178         }
4179         maxrootrecs = (int)XFS_BTREE_BLOCK_MAXRECS(sz, xfs_bmdr, 0);
4180         minleafrecs = mp->m_bmap_dmnr[0];
4181         minnoderecs = mp->m_bmap_dmnr[1];
4182         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
4183         for (level = 1; maxblocks > 1; level++) {
4184                 if (maxblocks <= maxrootrecs)
4185                         maxblocks = 1;
4186                 else
4187                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
4188         }
4189         mp->m_bm_maxlevels[whichfork] = level;
4190 }
4191
4192 /*
4193  * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
4194  * caller.  Frees all the extents that need freeing, which must be done
4195  * last due to locking considerations.  We never free any extents in
4196  * the first transaction.  This is to allow the caller to make the first
4197  * transaction a synchronous one so that the pointers to the data being
4198  * broken in this transaction will be permanent before the data is actually
4199  * freed.  This is necessary to prevent blocks from being reallocated
4200  * and written to before the free and reallocation are actually permanent.
4201  * We do not just make the first transaction synchronous here, because
4202  * there are more efficient ways to gain the same protection in some cases
4203  * (see the file truncation code).
4204  *
4205  * Return 1 if the given transaction was committed and a new one
4206  * started, and 0 otherwise in the committed parameter.
4207  */
4208 /*ARGSUSED*/
4209 int                                             /* error */
4210 xfs_bmap_finish(
4211         xfs_trans_t             **tp,           /* transaction pointer addr */
4212         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
4213         xfs_fsblock_t           firstblock,     /* controlled ag for allocs */
4214         int                     *committed)     /* xact committed or not */
4215 {
4216         xfs_efd_log_item_t      *efd;           /* extent free data */
4217         xfs_efi_log_item_t      *efi;           /* extent free intention */
4218         int                     error;          /* error return value */
4219         xfs_bmap_free_item_t    *free;          /* free extent item */
4220         unsigned int            logres;         /* new log reservation */
4221         unsigned int            logcount;       /* new log count */
4222         xfs_mount_t             *mp;            /* filesystem mount structure */
4223         xfs_bmap_free_item_t    *next;          /* next item on free list */
4224         xfs_trans_t             *ntp;           /* new transaction pointer */
4225
4226         ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
4227         if (flist->xbf_count == 0) {
4228                 *committed = 0;
4229                 return 0;
4230         }
4231         ntp = *tp;
4232         efi = xfs_trans_get_efi(ntp, flist->xbf_count);
4233         for (free = flist->xbf_first; free; free = free->xbfi_next)
4234                 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
4235                         free->xbfi_blockcount);
4236         logres = ntp->t_log_res;
4237         logcount = ntp->t_log_count;
4238         ntp = xfs_trans_dup(*tp);
4239         error = xfs_trans_commit(*tp, 0, NULL);
4240         *tp = ntp;
4241         *committed = 1;
4242         /*
4243          * We have a new transaction, so we should return committed=1,
4244          * even though we're returning an error.
4245          */
4246         if (error) {
4247                 return error;
4248         }
4249         if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
4250                         logcount)))
4251                 return error;
4252         efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
4253         for (free = flist->xbf_first; free != NULL; free = next) {
4254                 next = free->xbfi_next;
4255                 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
4256                                 free->xbfi_blockcount))) {
4257                         /*
4258                          * The bmap free list will be cleaned up at a
4259                          * higher level.  The EFI will be canceled when
4260                          * this transaction is aborted.
4261                          * Need to force shutdown here to make sure it
4262                          * happens, since this transaction may not be
4263                          * dirty yet.
4264                          */
4265                         mp = ntp->t_mountp;
4266                         if (!XFS_FORCED_SHUTDOWN(mp))
4267                                 xfs_force_shutdown(mp,
4268                                                    (error == EFSCORRUPTED) ?
4269                                                    SHUTDOWN_CORRUPT_INCORE :
4270                                                    SHUTDOWN_META_IO_ERROR);
4271                         return error;
4272                 }
4273                 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
4274                         free->xbfi_blockcount);
4275                 xfs_bmap_del_free(flist, NULL, free);
4276         }
4277         return 0;
4278 }
4279
4280 /*
4281  * Free up any items left in the list.
4282  */
4283 void
4284 xfs_bmap_cancel(
4285         xfs_bmap_free_t         *flist) /* list of bmap_free_items */
4286 {
4287         xfs_bmap_free_item_t    *free;  /* free list item */
4288         xfs_bmap_free_item_t    *next;
4289
4290         if (flist->xbf_count == 0)
4291                 return;
4292         ASSERT(flist->xbf_first != NULL);
4293         for (free = flist->xbf_first; free; free = next) {
4294                 next = free->xbfi_next;
4295                 xfs_bmap_del_free(flist, NULL, free);
4296         }
4297         ASSERT(flist->xbf_count == 0);
4298 }
4299
4300 /*
4301  * Returns the file-relative block number of the first unused block(s)
4302  * in the file with at least "len" logically contiguous blocks free.
4303  * This is the lowest-address hole if the file has holes, else the first block
4304  * past the end of file.
4305  * Return 0 if the file is currently local (in-inode).
4306  */
4307 int                                             /* error */
4308 xfs_bmap_first_unused(
4309         xfs_trans_t     *tp,                    /* transaction pointer */
4310         xfs_inode_t     *ip,                    /* incore inode */
4311         xfs_extlen_t    len,                    /* size of hole to find */
4312         xfs_fileoff_t   *first_unused,          /* unused block */
4313         int             whichfork)              /* data or attr fork */
4314 {
4315         xfs_bmbt_rec_t  *ep;                    /* pointer to an extent entry */
4316         int             error;                  /* error return value */
4317         int             idx;                    /* extent record index */
4318         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4319         xfs_fileoff_t   lastaddr;               /* last block number seen */
4320         xfs_fileoff_t   lowest;                 /* lowest useful block */
4321         xfs_fileoff_t   max;                    /* starting useful block */
4322         xfs_fileoff_t   off;                    /* offset for this block */
4323         xfs_extnum_t    nextents;               /* number of extent entries */
4324
4325         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
4326                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
4327                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
4328         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4329                 *first_unused = 0;
4330                 return 0;
4331         }
4332         ifp = XFS_IFORK_PTR(ip, whichfork);
4333         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4334             (error = xfs_iread_extents(tp, ip, whichfork)))
4335                 return error;
4336         lowest = *first_unused;
4337         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4338         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
4339                 ep = xfs_iext_get_ext(ifp, idx);
4340                 off = xfs_bmbt_get_startoff(ep);
4341                 /*
4342                  * See if the hole before this extent will work.
4343                  */
4344                 if (off >= lowest + len && off - max >= len) {
4345                         *first_unused = max;
4346                         return 0;
4347                 }
4348                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
4349                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
4350         }
4351         *first_unused = max;
4352         return 0;
4353 }
4354
4355 /*
4356  * Returns the file-relative block number of the last block + 1 before
4357  * last_block (input value) in the file.
4358  * This is not based on i_size, it is based on the extent records.
4359  * Returns 0 for local files, as they do not have extent records.
4360  */
4361 int                                             /* error */
4362 xfs_bmap_last_before(
4363         xfs_trans_t     *tp,                    /* transaction pointer */
4364         xfs_inode_t     *ip,                    /* incore inode */
4365         xfs_fileoff_t   *last_block,            /* last block */
4366         int             whichfork)              /* data or attr fork */
4367 {
4368         xfs_fileoff_t   bno;                    /* input file offset */
4369         int             eof;                    /* hit end of file */
4370         xfs_bmbt_rec_t  *ep;                    /* pointer to last extent */
4371         int             error;                  /* error return value */
4372         xfs_bmbt_irec_t got;                    /* current extent value */
4373         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4374         xfs_extnum_t    lastx;                  /* last extent used */
4375         xfs_bmbt_irec_t prev;                   /* previous extent value */
4376
4377         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4378             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4379             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4380                return XFS_ERROR(EIO);
4381         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4382                 *last_block = 0;
4383                 return 0;
4384         }
4385         ifp = XFS_IFORK_PTR(ip, whichfork);
4386         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4387             (error = xfs_iread_extents(tp, ip, whichfork)))
4388                 return error;
4389         bno = *last_block - 1;
4390         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4391                 &prev);
4392         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
4393                 if (prev.br_startoff == NULLFILEOFF)
4394                         *last_block = 0;
4395                 else
4396                         *last_block = prev.br_startoff + prev.br_blockcount;
4397         }
4398         /*
4399          * Otherwise *last_block is already the right answer.
4400          */
4401         return 0;
4402 }
4403
4404 /*
4405  * Returns the file-relative block number of the first block past eof in
4406  * the file.  This is not based on i_size, it is based on the extent records.
4407  * Returns 0 for local files, as they do not have extent records.
4408  */
4409 int                                             /* error */
4410 xfs_bmap_last_offset(
4411         xfs_trans_t     *tp,                    /* transaction pointer */
4412         xfs_inode_t     *ip,                    /* incore inode */
4413         xfs_fileoff_t   *last_block,            /* last block */
4414         int             whichfork)              /* data or attr fork */
4415 {
4416         xfs_bmbt_rec_t  *ep;                    /* pointer to last extent */
4417         int             error;                  /* error return value */
4418         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4419         xfs_extnum_t    nextents;               /* number of extent entries */
4420
4421         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4422             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4423             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4424                return XFS_ERROR(EIO);
4425         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4426                 *last_block = 0;
4427                 return 0;
4428         }
4429         ifp = XFS_IFORK_PTR(ip, whichfork);
4430         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4431             (error = xfs_iread_extents(tp, ip, whichfork)))
4432                 return error;
4433         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4434         if (!nextents) {
4435                 *last_block = 0;
4436                 return 0;
4437         }
4438         ep = xfs_iext_get_ext(ifp, nextents - 1);
4439         *last_block = xfs_bmbt_get_startoff(ep) + xfs_bmbt_get_blockcount(ep);
4440         return 0;
4441 }
4442
4443 /*
4444  * Returns whether the selected fork of the inode has exactly one
4445  * block or not.  For the data fork we check this matches di_size,
4446  * implying the file's range is 0..bsize-1.
4447  */
4448 int                                     /* 1=>1 block, 0=>otherwise */
4449 xfs_bmap_one_block(
4450         xfs_inode_t     *ip,            /* incore inode */
4451         int             whichfork)      /* data or attr fork */
4452 {
4453         xfs_bmbt_rec_t  *ep;            /* ptr to fork's extent */
4454         xfs_ifork_t     *ifp;           /* inode fork pointer */
4455         int             rval;           /* return value */
4456         xfs_bmbt_irec_t s;              /* internal version of extent */
4457
4458 #ifndef DEBUG
4459         if (whichfork == XFS_DATA_FORK)
4460                 return ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize;
4461 #endif  /* !DEBUG */
4462         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4463                 return 0;
4464         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4465                 return 0;
4466         ifp = XFS_IFORK_PTR(ip, whichfork);
4467         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4468         ep = xfs_iext_get_ext(ifp, 0);
4469         xfs_bmbt_get_all(ep, &s);
4470         rval = s.br_startoff == 0 && s.br_blockcount == 1;
4471         if (rval && whichfork == XFS_DATA_FORK)
4472                 ASSERT(ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize);
4473         return rval;
4474 }
4475
4476 /*
4477  * Read in the extents to if_extents.
4478  * All inode fields are set up by caller, we just traverse the btree
4479  * and copy the records in. If the file system cannot contain unwritten
4480  * extents, the records are checked for no "state" flags.
4481  */
4482 int                                     /* error */
4483 xfs_bmap_read_extents(
4484         xfs_trans_t             *tp,    /* transaction pointer */
4485         xfs_inode_t             *ip,    /* incore inode */
4486         int                     whichfork) /* data or attr fork */
4487 {
4488         xfs_bmbt_block_t        *block; /* current btree block */
4489         xfs_fsblock_t           bno;    /* block # of "block" */
4490         xfs_buf_t               *bp;    /* buffer for "block" */
4491         int                     error;  /* error return value */
4492         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
4493 #ifdef XFS_BMAP_TRACE
4494         static char             fname[] = "xfs_bmap_read_extents";
4495 #endif
4496         xfs_extnum_t            i, j;   /* index into the extents list */
4497         xfs_ifork_t             *ifp;   /* fork structure */
4498         int                     level;  /* btree level, for checking */
4499         xfs_mount_t             *mp;    /* file system mount structure */
4500         __be64                  *pp;    /* pointer to block address */
4501         /* REFERENCED */
4502         xfs_extnum_t            room;   /* number of entries there's room for */
4503
4504         bno = NULLFSBLOCK;
4505         mp = ip->i_mount;
4506         ifp = XFS_IFORK_PTR(ip, whichfork);
4507         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4508                                         XFS_EXTFMT_INODE(ip);
4509         block = ifp->if_broot;
4510         /*
4511          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4512          */
4513         level = be16_to_cpu(block->bb_level);
4514         ASSERT(level > 0);
4515         pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
4516         bno = be64_to_cpu(*pp);
4517         ASSERT(bno != NULLDFSBNO);
4518         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
4519         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
4520         /*
4521          * Go down the tree until leaf level is reached, following the first
4522          * pointer (leftmost) at each level.
4523          */
4524         while (level-- > 0) {
4525                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4526                                 XFS_BMAP_BTREE_REF)))
4527                         return error;
4528                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
4529                 XFS_WANT_CORRUPTED_GOTO(
4530                         XFS_BMAP_SANITY_CHECK(mp, block, level),
4531                         error0);
4532                 if (level == 0)
4533                         break;
4534                 pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt, block,
4535                         1, mp->m_bmap_dmxr[1]);
4536                 bno = be64_to_cpu(*pp);
4537                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
4538                 xfs_trans_brelse(tp, bp);
4539         }
4540         /*
4541          * Here with bp and block set to the leftmost leaf node in the tree.
4542          */
4543         room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4544         i = 0;
4545         /*
4546          * Loop over all leaf nodes.  Copy information to the extent records.
4547          */
4548         for (;;) {
4549                 xfs_bmbt_rec_t  *frp, *trp;
4550                 xfs_fsblock_t   nextbno;
4551                 xfs_extnum_t    num_recs;
4552                 xfs_extnum_t    start;
4553
4554
4555                 num_recs = be16_to_cpu(block->bb_numrecs);
4556                 if (unlikely(i + num_recs > room)) {
4557                         ASSERT(i + num_recs <= room);
4558                         xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
4559                                 "corrupt dinode %Lu, (btree extents).",
4560                                 (unsigned long long) ip->i_ino);
4561                         XFS_ERROR_REPORT("xfs_bmap_read_extents(1)",
4562                                          XFS_ERRLEVEL_LOW,
4563                                         ip->i_mount);
4564                         goto error0;
4565                 }
4566                 XFS_WANT_CORRUPTED_GOTO(
4567                         XFS_BMAP_SANITY_CHECK(mp, block, 0),
4568                         error0);
4569                 /*
4570                  * Read-ahead the next leaf block, if any.
4571                  */
4572                 nextbno = be64_to_cpu(block->bb_rightsib);
4573                 if (nextbno != NULLFSBLOCK)
4574                         xfs_btree_reada_bufl(mp, nextbno, 1);
4575                 /*
4576                  * Copy records into the extent records.
4577                  */
4578                 frp = XFS_BTREE_REC_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt,
4579                         block, 1, mp->m_bmap_dmxr[0]);
4580                 start = i;
4581                 for (j = 0; j < num_recs; j++, i++, frp++) {
4582                         trp = xfs_iext_get_ext(ifp, i);
4583                         trp->l0 = INT_GET(frp->l0, ARCH_CONVERT);
4584                         trp->l1 = INT_GET(frp->l1, ARCH_CONVERT);
4585                 }
4586                 if (exntf == XFS_EXTFMT_NOSTATE) {
4587                         /*
4588                          * Check all attribute bmap btree records and
4589                          * any "older" data bmap btree records for a
4590                          * set bit in the "extent flag" position.
4591                          */
4592                         if (unlikely(xfs_check_nostate_extents(ifp,
4593                                         start, num_recs))) {
4594                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4595                                                  XFS_ERRLEVEL_LOW,
4596                                                  ip->i_mount);
4597                                 goto error0;
4598                         }
4599                 }
4600                 xfs_trans_brelse(tp, bp);
4601                 bno = nextbno;
4602                 /*
4603                  * If we've reached the end, stop.
4604                  */
4605                 if (bno == NULLFSBLOCK)
4606                         break;
4607                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4608                                 XFS_BMAP_BTREE_REF)))
4609                         return error;
4610                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
4611         }
4612         ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4613         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4614         xfs_bmap_trace_exlist(fname, ip, i, whichfork);
4615         return 0;
4616 error0:
4617         xfs_trans_brelse(tp, bp);
4618         return XFS_ERROR(EFSCORRUPTED);
4619 }
4620
4621 #ifdef XFS_BMAP_TRACE
4622 /*
4623  * Add bmap trace insert entries for all the contents of the extent records.
4624  */
4625 void
4626 xfs_bmap_trace_exlist(
4627         char            *fname,         /* function name */
4628         xfs_inode_t     *ip,            /* incore inode pointer */
4629         xfs_extnum_t    cnt,            /* count of entries in the list */
4630         int             whichfork)      /* data or attr fork */
4631 {
4632         xfs_bmbt_rec_t  *ep;            /* current extent record */
4633         xfs_extnum_t    idx;            /* extent record index */
4634         xfs_ifork_t     *ifp;           /* inode fork pointer */
4635         xfs_bmbt_irec_t s;              /* file extent record */
4636
4637         ifp = XFS_IFORK_PTR(ip, whichfork);
4638         ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4639         for (idx = 0; idx < cnt; idx++) {
4640                 ep = xfs_iext_get_ext(ifp, idx);
4641                 xfs_bmbt_get_all(ep, &s);
4642                 xfs_bmap_trace_insert(fname, "exlist", ip, idx, 1, &s, NULL,
4643                         whichfork);
4644         }
4645 }
4646 #endif
4647
4648 #ifdef DEBUG
4649 /*
4650  * Validate that the bmbt_irecs being returned from bmapi are valid
4651  * given the callers original parameters.  Specifically check the
4652  * ranges of the returned irecs to ensure that they only extent beyond
4653  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4654  */
4655 STATIC void
4656 xfs_bmap_validate_ret(
4657         xfs_fileoff_t           bno,
4658         xfs_filblks_t           len,
4659         int                     flags,
4660         xfs_bmbt_irec_t         *mval,
4661         int                     nmap,
4662         int                     ret_nmap)
4663 {
4664         int                     i;              /* index to map values */
4665
4666         ASSERT(ret_nmap <= nmap);
4667
4668         for (i = 0; i < ret_nmap; i++) {
4669                 ASSERT(mval[i].br_blockcount > 0);
4670                 if (!(flags & XFS_BMAPI_ENTIRE)) {
4671                         ASSERT(mval[i].br_startoff >= bno);
4672                         ASSERT(mval[i].br_blockcount <= len);
4673                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4674                                bno + len);
4675                 } else {
4676                         ASSERT(mval[i].br_startoff < bno + len);
4677                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4678                                bno);
4679                 }
4680                 ASSERT(i == 0 ||
4681                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4682                        mval[i].br_startoff);
4683                 if ((flags & XFS_BMAPI_WRITE) && !(flags & XFS_BMAPI_DELAY))
4684                         ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4685                                mval[i].br_startblock != HOLESTARTBLOCK);
4686                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4687                        mval[i].br_state == XFS_EXT_UNWRITTEN);
4688         }
4689 }
4690 #endif /* DEBUG */
4691
4692
4693 /*
4694  * Map file blocks to filesystem blocks.
4695  * File range is given by the bno/len pair.
4696  * Adds blocks to file if a write ("flags & XFS_BMAPI_WRITE" set)
4697  * into a hole or past eof.
4698  * Only allocates blocks from a single allocation group,
4699  * to avoid locking problems.
4700  * The returned value in "firstblock" from the first call in a transaction
4701  * must be remembered and presented to subsequent calls in "firstblock".
4702  * An upper bound for the number of blocks to be allocated is supplied to
4703  * the first call in "total"; if no allocation group has that many free
4704  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4705  */
4706 int                                     /* error */
4707 xfs_bmapi(
4708         xfs_trans_t     *tp,            /* transaction pointer */
4709         xfs_inode_t     *ip,            /* incore inode */
4710         xfs_fileoff_t   bno,            /* starting file offs. mapped */
4711         xfs_filblks_t   len,            /* length to map in file */
4712         int             flags,          /* XFS_BMAPI_... */
4713         xfs_fsblock_t   *firstblock,    /* first allocated block
4714                                            controls a.g. for allocs */
4715         xfs_extlen_t    total,          /* total blocks needed */
4716         xfs_bmbt_irec_t *mval,          /* output: map values */
4717         int             *nmap,          /* i/o: mval size/count */
4718         xfs_bmap_free_t *flist,         /* i/o: list extents to free */
4719         xfs_extdelta_t  *delta)         /* o: change made to incore extents */
4720 {
4721         xfs_fsblock_t   abno;           /* allocated block number */
4722         xfs_extlen_t    alen;           /* allocated extent length */
4723         xfs_fileoff_t   aoff;           /* allocated file offset */
4724         xfs_bmalloca_t  bma;            /* args for xfs_bmap_alloc */
4725         xfs_btree_cur_t *cur;           /* bmap btree cursor */
4726         xfs_fileoff_t   end;            /* end of mapped file region */
4727         int             eof;            /* we've hit the end of extents */
4728         xfs_bmbt_rec_t  *ep;            /* extent record pointer */
4729         int             error;          /* error return */
4730         xfs_bmbt_irec_t got;            /* current file extent record */
4731         xfs_ifork_t     *ifp;           /* inode fork pointer */
4732         xfs_extlen_t    indlen;         /* indirect blocks length */
4733         xfs_extnum_t    lastx;          /* last useful extent number */
4734         int             logflags;       /* flags for transaction logging */
4735         xfs_extlen_t    minleft;        /* min blocks left after allocation */
4736         xfs_extlen_t    minlen;         /* min allocation size */
4737         xfs_mount_t     *mp;            /* xfs mount structure */
4738         int             n;              /* current extent index */
4739         int             nallocs;        /* number of extents alloc\'d */
4740         xfs_extnum_t    nextents;       /* number of extents in file */
4741         xfs_fileoff_t   obno;           /* old block number (offset) */
4742         xfs_bmbt_irec_t prev;           /* previous file extent record */
4743         int             tmp_logflags;   /* temp flags holder */
4744         int             whichfork;      /* data or attr fork */
4745         char            inhole;         /* current location is hole in file */
4746         char            wasdelay;       /* old extent was delayed */
4747         char            wr;             /* this is a write request */
4748         char            rt;             /* this is a realtime file */
4749 #ifdef DEBUG
4750         xfs_fileoff_t   orig_bno;       /* original block number value */
4751         int             orig_flags;     /* original flags arg value */
4752         xfs_filblks_t   orig_len;       /* original value of len arg */
4753         xfs_bmbt_irec_t *orig_mval;     /* original value of mval */
4754         int             orig_nmap;      /* original value of *nmap */
4755
4756         orig_bno = bno;
4757         orig_len = len;
4758         orig_flags = flags;
4759         orig_mval = mval;
4760         orig_nmap = *nmap;
4761 #endif
4762         ASSERT(*nmap >= 1);
4763         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP || !(flags & XFS_BMAPI_WRITE));
4764         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4765                 XFS_ATTR_FORK : XFS_DATA_FORK;
4766         mp = ip->i_mount;
4767         if (unlikely(XFS_TEST_ERROR(
4768             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4769              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4770              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4771              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4772                 XFS_ERROR_REPORT("xfs_bmapi", XFS_ERRLEVEL_LOW, mp);
4773                 return XFS_ERROR(EFSCORRUPTED);
4774         }
4775         if (XFS_FORCED_SHUTDOWN(mp))
4776                 return XFS_ERROR(EIO);
4777         rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4778         ifp = XFS_IFORK_PTR(ip, whichfork);
4779         ASSERT(ifp->if_ext_max ==
4780                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4781         if ((wr = (flags & XFS_BMAPI_WRITE)) != 0)
4782                 XFS_STATS_INC(xs_blk_mapw);
4783         else
4784                 XFS_STATS_INC(xs_blk_mapr);
4785         /*
4786          * IGSTATE flag is used to combine extents which
4787          * differ only due to the state of the extents.
4788          * This technique is used from xfs_getbmap()
4789          * when the caller does not wish to see the
4790          * separation (which is the default).
4791          *
4792          * This technique is also used when writing a
4793          * buffer which has been partially written,
4794          * (usually by being flushed during a chunkread),
4795          * to ensure one write takes place. This also
4796          * prevents a change in the xfs inode extents at
4797          * this time, intentionally. This change occurs
4798          * on completion of the write operation, in
4799          * xfs_strat_comp(), where the xfs_bmapi() call
4800          * is transactioned, and the extents combined.
4801          */
4802         if ((flags & XFS_BMAPI_IGSTATE) && wr)  /* if writing unwritten space */
4803                 wr = 0;                         /* no allocations are allowed */
4804         ASSERT(wr || !(flags & XFS_BMAPI_DELAY));
4805         logflags = 0;
4806         nallocs = 0;
4807         cur = NULL;
4808         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4809                 ASSERT(wr && tp);
4810                 if ((error = xfs_bmap_local_to_extents(tp, ip,
4811                                 firstblock, total, &logflags, whichfork)))
4812                         goto error0;
4813         }
4814         if (wr && *firstblock == NULLFSBLOCK) {
4815                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4816                         minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4817                 else
4818                         minleft = 1;
4819         } else
4820                 minleft = 0;
4821         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4822             (error = xfs_iread_extents(tp, ip, whichfork)))
4823                 goto error0;
4824         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4825                 &prev);
4826         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4827         n = 0;
4828         end = bno + len;
4829         obno = bno;
4830         bma.ip = NULL;
4831         if (delta) {
4832                 delta->xed_startoff = NULLFILEOFF;
4833                 delta->xed_blockcount = 0;
4834         }
4835         while (bno < end && n < *nmap) {
4836                 /*
4837                  * Reading past eof, act as though there's a hole
4838                  * up to end.
4839                  */
4840                 if (eof && !wr)
4841                         got.br_startoff = end;
4842                 inhole = eof || got.br_startoff > bno;
4843                 wasdelay = wr && !inhole && !(flags & XFS_BMAPI_DELAY) &&
4844                         ISNULLSTARTBLOCK(got.br_startblock);
4845                 /*
4846                  * First, deal with the hole before the allocated space
4847                  * that we found, if any.
4848                  */
4849                 if (wr && (inhole || wasdelay)) {
4850                         /*
4851                          * For the wasdelay case, we could also just
4852                          * allocate the stuff asked for in this bmap call
4853                          * but that wouldn't be as good.
4854                          */
4855                         if (wasdelay && !(flags & XFS_BMAPI_EXACT)) {
4856                                 alen = (xfs_extlen_t)got.br_blockcount;
4857                                 aoff = got.br_startoff;
4858                                 if (lastx != NULLEXTNUM && lastx) {
4859                                         ep = xfs_iext_get_ext(ifp, lastx - 1);
4860                                         xfs_bmbt_get_all(ep, &prev);
4861                                 }
4862                         } else if (wasdelay) {
4863                                 alen = (xfs_extlen_t)
4864                                         XFS_FILBLKS_MIN(len,
4865                                                 (got.br_startoff +
4866                                                  got.br_blockcount) - bno);
4867                                 aoff = bno;
4868                         } else {
4869                                 alen = (xfs_extlen_t)
4870                                         XFS_FILBLKS_MIN(len, MAXEXTLEN);
4871                                 if (!eof)
4872                                         alen = (xfs_extlen_t)
4873                                                 XFS_FILBLKS_MIN(alen,
4874                                                         got.br_startoff - bno);
4875                                 aoff = bno;
4876                         }
4877                         minlen = (flags & XFS_BMAPI_CONTIG) ? alen : 1;
4878                         if (flags & XFS_BMAPI_DELAY) {
4879                                 xfs_extlen_t    extsz;
4880
4881                                 /* Figure out the extent size, adjust alen */
4882                                 if (rt) {
4883                                         if (!(extsz = ip->i_d.di_extsize))
4884                                                 extsz = mp->m_sb.sb_rextsize;
4885                                 } else {
4886                                         extsz = ip->i_d.di_extsize;
4887                                 }
4888                                 if (extsz) {
4889                                         error = xfs_bmap_extsize_align(mp,
4890                                                         &got, &prev, extsz,
4891                                                         rt, eof,
4892                                                         flags&XFS_BMAPI_DELAY,
4893                                                         flags&XFS_BMAPI_CONVERT,
4894                                                         &aoff, &alen);
4895                                         ASSERT(!error);
4896                                 }
4897
4898                                 if (rt)
4899                                         extsz = alen / mp->m_sb.sb_rextsize;
4900
4901                                 /*
4902                                  * Make a transaction-less quota reservation for
4903                                  * delayed allocation blocks. This number gets
4904                                  * adjusted later.  We return if we haven't
4905                                  * allocated blocks already inside this loop.
4906                                  */
4907                                 if ((error = XFS_TRANS_RESERVE_QUOTA_NBLKS(
4908                                                 mp, NULL, ip, (long)alen, 0,
4909                                                 rt ? XFS_QMOPT_RES_RTBLKS :
4910                                                      XFS_QMOPT_RES_REGBLKS))) {
4911                                         if (n == 0) {
4912                                                 *nmap = 0;
4913                                                 ASSERT(cur == NULL);
4914                                                 return error;
4915                                         }
4916                                         break;
4917                                 }
4918
4919                                 /*
4920                                  * Split changing sb for alen and indlen since
4921                                  * they could be coming from different places.
4922                                  */
4923                                 indlen = (xfs_extlen_t)
4924                                         xfs_bmap_worst_indlen(ip, alen);
4925                                 ASSERT(indlen > 0);
4926
4927                                 if (rt) {
4928                                         error = xfs_mod_incore_sb(mp,
4929                                                         XFS_SBS_FREXTENTS,
4930                                                         -(extsz), (flags &
4931                                                         XFS_BMAPI_RSVBLOCKS));
4932                                 } else {
4933                                         error = xfs_mod_incore_sb(mp,
4934                                                         XFS_SBS_FDBLOCKS,
4935                                                         -(alen), (flags &
4936                                                         XFS_BMAPI_RSVBLOCKS));
4937                                 }
4938                                 if (!error) {
4939                                         error = xfs_mod_incore_sb(mp,
4940                                                         XFS_SBS_FDBLOCKS,
4941                                                         -(indlen), (flags &
4942                                                         XFS_BMAPI_RSVBLOCKS));
4943                                         if (error && rt)
4944                                                 xfs_mod_incore_sb(mp,
4945                                                         XFS_SBS_FREXTENTS,
4946                                                         extsz, (flags &
4947                                                         XFS_BMAPI_RSVBLOCKS));
4948                                         else if (error)
4949                                                 xfs_mod_incore_sb(mp,
4950                                                         XFS_SBS_FDBLOCKS,
4951                                                         alen, (flags &
4952                                                         XFS_BMAPI_RSVBLOCKS));
4953                                 }
4954
4955                                 if (error) {
4956                                         if (XFS_IS_QUOTA_ON(mp))
4957                                                 /* unreserve the blocks now */
4958                                                 (void)
4959                                                 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(
4960                                                         mp, NULL, ip,
4961                                                         (long)alen, 0, rt ?
4962                                                         XFS_QMOPT_RES_RTBLKS :
4963                                                         XFS_QMOPT_RES_REGBLKS);
4964                                         break;
4965                                 }
4966
4967                                 ip->i_delayed_blks += alen;
4968                                 abno = NULLSTARTBLOCK(indlen);
4969                         } else {
4970                                 /*
4971                                  * If first time, allocate and fill in
4972                                  * once-only bma fields.
4973                                  */
4974                                 if (bma.ip == NULL) {
4975                                         bma.tp = tp;
4976                                         bma.ip = ip;
4977                                         bma.prevp = &prev;
4978                                         bma.gotp = &got;
4979                                         bma.total = total;
4980                                         bma.userdata = 0;
4981                                 }
4982                                 /* Indicate if this is the first user data
4983                                  * in the file, or just any user data.
4984                                  */
4985                                 if (!(flags & XFS_BMAPI_METADATA)) {
4986                                         bma.userdata = (aoff == 0) ?
4987                                                 XFS_ALLOC_INITIAL_USER_DATA :
4988                                                 XFS_ALLOC_USERDATA;
4989                                 }
4990                                 /*
4991                                  * Fill in changeable bma fields.
4992                                  */
4993                                 bma.eof = eof;
4994                                 bma.firstblock = *firstblock;
4995                                 bma.alen = alen;
4996                                 bma.off = aoff;
4997                                 bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4998                                 bma.wasdel = wasdelay;
4999                                 bma.minlen = minlen;
5000                                 bma.low = flist->xbf_low;
5001                                 bma.minleft = minleft;
5002                                 /*
5003                                  * Only want to do the alignment at the
5004                                  * eof if it is userdata and allocation length
5005                                  * is larger than a stripe unit.
5006                                  */
5007                                 if (mp->m_dalign && alen >= mp->m_dalign &&
5008                                     (!(flags & XFS_BMAPI_METADATA)) &&
5009                                     (whichfork == XFS_DATA_FORK)) {
5010                                         if ((error = xfs_bmap_isaeof(ip, aoff,
5011                                                         whichfork, &bma.aeof)))
5012                                                 goto error0;
5013                                 } else
5014                                         bma.aeof = 0;
5015                                 /*
5016                                  * Call allocator.
5017                                  */
5018                                 if ((error = xfs_bmap_alloc(&bma)))
5019                                         goto error0;
5020                                 /*
5021                                  * Copy out result fields.
5022                                  */
5023                                 abno = bma.rval;
5024                                 if ((flist->xbf_low = bma.low))
5025                                         minleft = 0;
5026                                 alen = bma.alen;
5027                                 aoff = bma.off;
5028                                 ASSERT(*firstblock == NULLFSBLOCK ||
5029                                        XFS_FSB_TO_AGNO(mp, *firstblock) ==
5030                                        XFS_FSB_TO_AGNO(mp, bma.firstblock) ||
5031                                        (flist->xbf_low &&
5032                                         XFS_FSB_TO_AGNO(mp, *firstblock) <
5033                                         XFS_FSB_TO_AGNO(mp, bma.firstblock)));
5034                                 *firstblock = bma.firstblock;
5035                                 if (cur)
5036                                         cur->bc_private.b.firstblock =
5037                                                 *firstblock;
5038                                 if (abno == NULLFSBLOCK)
5039                                         break;
5040                                 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
5041                                         cur = xfs_btree_init_cursor(mp,
5042                                                 tp, NULL, 0, XFS_BTNUM_BMAP,
5043                                                 ip, whichfork);
5044                                         cur->bc_private.b.firstblock =
5045                                                 *firstblock;
5046                                         cur->bc_private.b.flist = flist;
5047                                 }
5048                                 /*
5049                                  * Bump the number of extents we've allocated
5050                                  * in this call.
5051                                  */
5052                                 nallocs++;
5053                         }
5054                         if (cur)
5055                                 cur->bc_private.b.flags =
5056                                         wasdelay ? XFS_BTCUR_BPRV_WASDEL : 0;
5057                         got.br_startoff = aoff;
5058                         got.br_startblock = abno;
5059                         got.br_blockcount = alen;
5060                         got.br_state = XFS_EXT_NORM;    /* assume normal */
5061                         /*
5062                          * Determine state of extent, and the filesystem.
5063                          * A wasdelay extent has been initialized, so
5064                          * shouldn't be flagged as unwritten.
5065                          */
5066                         if (wr && XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
5067                                 if (!wasdelay && (flags & XFS_BMAPI_PREALLOC))
5068                                         got.br_state = XFS_EXT_UNWRITTEN;
5069                         }
5070                         error = xfs_bmap_add_extent(ip, lastx, &cur, &got,
5071                                 firstblock, flist, &tmp_logflags, delta,
5072                                 whichfork, (flags & XFS_BMAPI_RSVBLOCKS));
5073                         logflags |= tmp_logflags;
5074                         if (error)
5075                                 goto error0;
5076                         lastx = ifp->if_lastex;
5077                         ep = xfs_iext_get_ext(ifp, lastx);
5078                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5079                         xfs_bmbt_get_all(ep, &got);
5080                         ASSERT(got.br_startoff <= aoff);
5081                         ASSERT(got.br_startoff + got.br_blockcount >=
5082                                 aoff + alen);
5083 #ifdef DEBUG
5084                         if (flags & XFS_BMAPI_DELAY) {
5085                                 ASSERT(ISNULLSTARTBLOCK(got.br_startblock));
5086                                 ASSERT(STARTBLOCKVAL(got.br_startblock) > 0);
5087                         }
5088                         ASSERT(got.br_state == XFS_EXT_NORM ||
5089                                got.br_state == XFS_EXT_UNWRITTEN);
5090 #endif
5091                         /*
5092                          * Fall down into the found allocated space case.
5093                          */
5094                 } else if (inhole) {
5095                         /*
5096                          * Reading in a hole.
5097                          */
5098                         mval->br_startoff = bno;
5099                         mval->br_startblock = HOLESTARTBLOCK;
5100                         mval->br_blockcount =
5101                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
5102                         mval->br_state = XFS_EXT_NORM;
5103                         bno += mval->br_blockcount;
5104                         len -= mval->br_blockcount;
5105                         mval++;
5106                         n++;
5107                         continue;
5108                 }
5109                 /*
5110                  * Then deal with the allocated space we found.
5111                  */
5112                 ASSERT(ep != NULL);
5113                 if (!(flags & XFS_BMAPI_ENTIRE) &&
5114                     (got.br_startoff + got.br_blockcount > obno)) {
5115                         if (obno > bno)
5116                                 bno = obno;
5117                         ASSERT((bno >= obno) || (n == 0));
5118                         ASSERT(bno < end);
5119                         mval->br_startoff = bno;
5120                         if (ISNULLSTARTBLOCK(got.br_startblock)) {
5121                                 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
5122                                 mval->br_startblock = DELAYSTARTBLOCK;
5123                         } else
5124                                 mval->br_startblock =
5125                                         got.br_startblock +
5126                                         (bno - got.br_startoff);
5127                         /*
5128                          * Return the minimum of what we got and what we
5129                          * asked for for the length.  We can use the len
5130                          * variable here because it is modified below
5131                          * and we could have been there before coming
5132                          * here if the first part of the allocation
5133                          * didn't overlap what was asked for.
5134                          */
5135                         mval->br_blockcount =
5136                                 XFS_FILBLKS_MIN(end - bno, got.br_blockcount -
5137                                         (bno - got.br_startoff));
5138                         mval->br_state = got.br_state;
5139                         ASSERT(mval->br_blockcount <= len);
5140                 } else {
5141                         *mval = got;
5142                         if (ISNULLSTARTBLOCK(mval->br_startblock)) {
5143                                 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
5144                                 mval->br_startblock = DELAYSTARTBLOCK;
5145                         }
5146                 }
5147
5148                 /*
5149                  * Check if writing previously allocated but
5150                  * unwritten extents.
5151                  */
5152                 if (wr && mval->br_state == XFS_EXT_UNWRITTEN &&
5153                     ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_DELAY)) == 0)) {
5154                         /*
5155                          * Modify (by adding) the state flag, if writing.
5156                          */
5157                         ASSERT(mval->br_blockcount <= len);
5158                         if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
5159                                 cur = xfs_btree_init_cursor(mp,
5160                                         tp, NULL, 0, XFS_BTNUM_BMAP,
5161                                         ip, whichfork);
5162                                 cur->bc_private.b.firstblock =
5163                                         *firstblock;
5164                                 cur->bc_private.b.flist = flist;
5165                         }
5166                         mval->br_state = XFS_EXT_NORM;
5167                         error = xfs_bmap_add_extent(ip, lastx, &cur, mval,
5168                                 firstblock, flist, &tmp_logflags, delta,
5169                                 whichfork, (flags & XFS_BMAPI_RSVBLOCKS));
5170                         logflags |= tmp_logflags;
5171                         if (error)
5172                                 goto error0;
5173                         lastx = ifp->if_lastex;
5174                         ep = xfs_iext_get_ext(ifp, lastx);
5175                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5176                         xfs_bmbt_get_all(ep, &got);
5177                         /*
5178                          * We may have combined previously unwritten
5179                          * space with written space, so generate
5180                          * another request.
5181                          */
5182                         if (mval->br_blockcount < len)
5183                                 continue;
5184                 }
5185
5186                 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
5187                        ((mval->br_startoff + mval->br_blockcount) <= end));
5188                 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
5189                        (mval->br_blockcount <= len) ||
5190                        (mval->br_startoff < obno));
5191                 bno = mval->br_startoff + mval->br_blockcount;
5192                 len = end - bno;
5193                 if (n > 0 && mval->br_startoff == mval[-1].br_startoff) {
5194                         ASSERT(mval->br_startblock == mval[-1].br_startblock);
5195                         ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
5196                         ASSERT(mval->br_state == mval[-1].br_state);
5197                         mval[-1].br_blockcount = mval->br_blockcount;
5198                         mval[-1].br_state = mval->br_state;
5199                 } else if (n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
5200                            mval[-1].br_startblock != DELAYSTARTBLOCK &&
5201                            mval[-1].br_startblock != HOLESTARTBLOCK &&
5202                            mval->br_startblock ==
5203                            mval[-1].br_startblock + mval[-1].br_blockcount &&
5204                            ((flags & XFS_BMAPI_IGSTATE) ||
5205                                 mval[-1].br_state == mval->br_state)) {
5206                         ASSERT(mval->br_startoff ==
5207                                mval[-1].br_startoff + mval[-1].br_blockcount);
5208                         mval[-1].br_blockcount += mval->br_blockcount;
5209                 } else if (n > 0 &&
5210                            mval->br_startblock == DELAYSTARTBLOCK &&
5211                            mval[-1].br_startblock == DELAYSTARTBLOCK &&
5212                            mval->br_startoff ==
5213                            mval[-1].br_startoff + mval[-1].br_blockcount) {
5214                         mval[-1].br_blockcount += mval->br_blockcount;
5215                         mval[-1].br_state = mval->br_state;
5216                 } else if (!((n == 0) &&
5217                              ((mval->br_startoff + mval->br_blockcount) <=
5218                               obno))) {
5219                         mval++;
5220                         n++;
5221                 }
5222                 /*
5223                  * If we're done, stop now.  Stop when we've allocated
5224                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
5225                  * the transaction may get too big.
5226                  */
5227                 if (bno >= end || n >= *nmap || nallocs >= *nmap)
5228                         break;
5229                 /*
5230                  * Else go on to the next record.
5231                  */
5232                 ep = xfs_iext_get_ext(ifp, ++lastx);
5233                 if (lastx >= nextents) {
5234                         eof = 1;
5235                         prev = got;
5236                 } else
5237                         xfs_bmbt_get_all(ep, &got);
5238         }
5239         ifp->if_lastex = lastx;
5240         *nmap = n;
5241         /*
5242          * Transform from btree to extents, give it cur.
5243          */
5244         if (tp && XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5245             XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5246                 ASSERT(wr && cur);
5247                 error = xfs_bmap_btree_to_extents(tp, ip, cur,
5248                         &tmp_logflags, whichfork);
5249                 logflags |= tmp_logflags;
5250                 if (error)
5251                         goto error0;
5252         }
5253         ASSERT(ifp->if_ext_max ==
5254                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5255         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
5256                XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max);
5257         error = 0;
5258         if (delta && delta->xed_startoff != NULLFILEOFF) {
5259                 /* A change was actually made.
5260                  * Note that delta->xed_blockount is an offset at this
5261                  * point and needs to be converted to a block count.
5262                  */
5263                 ASSERT(delta->xed_blockcount > delta->xed_startoff);
5264                 delta->xed_blockcount -= delta->xed_startoff;
5265         }
5266 error0:
5267         /*
5268          * Log everything.  Do this after conversion, there's no point in
5269          * logging the extent records if we've converted to btree format.
5270          */
5271         if ((logflags & XFS_ILOG_FEXT(whichfork)) &&
5272             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5273                 logflags &= ~XFS_ILOG_FEXT(whichfork);
5274         else if ((logflags & XFS_ILOG_FBROOT(whichfork)) &&
5275                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5276                 logflags &= ~XFS_ILOG_FBROOT(whichfork);
5277         /*
5278          * Log whatever the flags say, even if error.  Otherwise we might miss
5279          * detecting a case where the data is changed, there's an error,
5280          * and it's not logged so we don't shutdown when we should.
5281          */
5282         if (logflags) {
5283                 ASSERT(tp && wr);
5284                 xfs_trans_log_inode(tp, ip, logflags);
5285         }
5286         if (cur) {
5287                 if (!error) {
5288                         ASSERT(*firstblock == NULLFSBLOCK ||
5289                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
5290                                XFS_FSB_TO_AGNO(mp,
5291                                        cur->bc_private.b.firstblock) ||
5292                                (flist->xbf_low &&
5293                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
5294                                 XFS_FSB_TO_AGNO(mp,
5295                                         cur->bc_private.b.firstblock)));
5296                         *firstblock = cur->bc_private.b.firstblock;
5297                 }
5298                 xfs_btree_del_cursor(cur,
5299                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5300         }
5301         if (!error)
5302                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
5303                         orig_nmap, *nmap);
5304         return error;
5305 }
5306
5307 /*
5308  * Map file blocks to filesystem blocks, simple version.
5309  * One block (extent) only, read-only.
5310  * For flags, only the XFS_BMAPI_ATTRFORK flag is examined.
5311  * For the other flag values, the effect is as if XFS_BMAPI_METADATA
5312  * was set and all the others were clear.
5313  */
5314 int                                             /* error */
5315 xfs_bmapi_single(
5316         xfs_trans_t     *tp,            /* transaction pointer */
5317         xfs_inode_t     *ip,            /* incore inode */
5318         int             whichfork,      /* data or attr fork */
5319         xfs_fsblock_t   *fsb,           /* output: mapped block */
5320         xfs_fileoff_t   bno)            /* starting file offs. mapped */
5321 {
5322         int             eof;            /* we've hit the end of extents */
5323         int             error;          /* error return */
5324         xfs_bmbt_irec_t got;            /* current file extent record */
5325         xfs_ifork_t     *ifp;           /* inode fork pointer */
5326         xfs_extnum_t    lastx;          /* last useful extent number */
5327         xfs_bmbt_irec_t prev;           /* previous file extent record */
5328
5329         ifp = XFS_IFORK_PTR(ip, whichfork);
5330         if (unlikely(
5331             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
5332             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)) {
5333                XFS_ERROR_REPORT("xfs_bmapi_single", XFS_ERRLEVEL_LOW,
5334                                 ip->i_mount);
5335                return XFS_ERROR(EFSCORRUPTED);
5336         }
5337         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
5338                 return XFS_ERROR(EIO);
5339         XFS_STATS_INC(xs_blk_mapr);
5340         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5341             (error = xfs_iread_extents(tp, ip, whichfork)))
5342                 return error;
5343         (void)xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5344                 &prev);
5345         /*
5346          * Reading past eof, act as though there's a hole
5347          * up to end.
5348          */
5349         if (eof || got.br_startoff > bno) {
5350                 *fsb = NULLFSBLOCK;
5351                 return 0;
5352         }
5353         ASSERT(!ISNULLSTARTBLOCK(got.br_startblock));
5354         ASSERT(bno < got.br_startoff + got.br_blockcount);
5355         *fsb = got.br_startblock + (bno - got.br_startoff);
5356         ifp->if_lastex = lastx;
5357         return 0;
5358 }
5359
5360 /*
5361  * Unmap (remove) blocks from a file.
5362  * If nexts is nonzero then the number of extents to remove is limited to
5363  * that value.  If not all extents in the block range can be removed then
5364  * *done is set.
5365  */
5366 int                                             /* error */
5367 xfs_bunmapi(
5368         xfs_trans_t             *tp,            /* transaction pointer */
5369         struct xfs_inode        *ip,            /* incore inode */
5370         xfs_fileoff_t           bno,            /* starting offset to unmap */
5371         xfs_filblks_t           len,            /* length to unmap in file */
5372         int                     flags,          /* misc flags */
5373         xfs_extnum_t            nexts,          /* number of extents max */
5374         xfs_fsblock_t           *firstblock,    /* first allocated block
5375                                                    controls a.g. for allocs */
5376         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
5377         xfs_extdelta_t          *delta,         /* o: change made to incore
5378                                                    extents */
5379         int                     *done)          /* set if not done yet */
5380 {
5381         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
5382         xfs_bmbt_irec_t         del;            /* extent being deleted */
5383         int                     eof;            /* is deleting at eof */
5384         xfs_bmbt_rec_t          *ep;            /* extent record pointer */
5385         int                     error;          /* error return value */
5386         xfs_extnum_t            extno;          /* extent number in list */
5387         xfs_bmbt_irec_t         got;            /* current extent record */
5388         xfs_ifork_t             *ifp;           /* inode fork pointer */
5389         int                     isrt;           /* freeing in rt area */
5390         xfs_extnum_t            lastx;          /* last extent index used */
5391         int                     logflags;       /* transaction logging flags */
5392         xfs_extlen_t            mod;            /* rt extent offset */
5393         xfs_mount_t             *mp;            /* mount structure */
5394         xfs_extnum_t            nextents;       /* number of file extents */
5395         xfs_bmbt_irec_t         prev;           /* previous extent record */
5396         xfs_fileoff_t           start;          /* first file offset deleted */
5397         int                     tmp_logflags;   /* partial logging flags */
5398         int                     wasdel;         /* was a delayed alloc extent */
5399         int                     whichfork;      /* data or attribute fork */
5400         int                     rsvd;           /* OK to allocate reserved blocks */
5401         xfs_fsblock_t           sum;
5402
5403         xfs_bunmap_trace(ip, bno, len, flags, (inst_t *)__return_address);
5404         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5405                 XFS_ATTR_FORK : XFS_DATA_FORK;
5406         ifp = XFS_IFORK_PTR(ip, whichfork);
5407         if (unlikely(
5408             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5409             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5410                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5411                                  ip->i_mount);
5412                 return XFS_ERROR(EFSCORRUPTED);
5413         }
5414         mp = ip->i_mount;
5415         if (XFS_FORCED_SHUTDOWN(mp))
5416                 return XFS_ERROR(EIO);
5417         rsvd = (flags & XFS_BMAPI_RSVBLOCKS) != 0;
5418         ASSERT(len > 0);
5419         ASSERT(nexts >= 0);
5420         ASSERT(ifp->if_ext_max ==
5421                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5422         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5423             (error = xfs_iread_extents(tp, ip, whichfork)))
5424                 return error;
5425         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5426         if (nextents == 0) {
5427                 *done = 1;
5428                 return 0;
5429         }
5430         XFS_STATS_INC(xs_blk_unmap);
5431         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5432         start = bno;
5433         bno = start + len - 1;
5434         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5435                 &prev);
5436         if (delta) {
5437                 delta->xed_startoff = NULLFILEOFF;
5438                 delta->xed_blockcount = 0;
5439         }
5440         /*
5441          * Check to see if the given block number is past the end of the
5442          * file, back up to the last block if so...
5443          */
5444         if (eof) {
5445                 ep = xfs_iext_get_ext(ifp, --lastx);
5446                 xfs_bmbt_get_all(ep, &got);
5447                 bno = got.br_startoff + got.br_blockcount - 1;
5448         }
5449         logflags = 0;
5450         if (ifp->if_flags & XFS_IFBROOT) {
5451                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5452                 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
5453                         whichfork);
5454                 cur->bc_private.b.firstblock = *firstblock;
5455                 cur->bc_private.b.flist = flist;
5456                 cur->bc_private.b.flags = 0;
5457         } else
5458                 cur = NULL;
5459         extno = 0;
5460         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5461                (nexts == 0 || extno < nexts)) {
5462                 /*
5463                  * Is the found extent after a hole in which bno lives?
5464                  * Just back up to the previous extent, if so.
5465                  */
5466                 if (got.br_startoff > bno) {
5467                         if (--lastx < 0)
5468                                 break;
5469                         ep = xfs_iext_get_ext(ifp, lastx);
5470                         xfs_bmbt_get_all(ep, &got);
5471                 }
5472                 /*
5473                  * Is the last block of this extent before the range
5474                  * we're supposed to delete?  If so, we're done.
5475                  */
5476                 bno = XFS_FILEOFF_MIN(bno,
5477                         got.br_startoff + got.br_blockcount - 1);
5478                 if (bno < start)
5479                         break;
5480                 /*
5481                  * Then deal with the (possibly delayed) allocated space
5482                  * we found.
5483                  */
5484                 ASSERT(ep != NULL);
5485                 del = got;
5486                 wasdel = ISNULLSTARTBLOCK(del.br_startblock);
5487                 if (got.br_startoff < start) {
5488                         del.br_startoff = start;
5489                         del.br_blockcount -= start - got.br_startoff;
5490                         if (!wasdel)
5491                                 del.br_startblock += start - got.br_startoff;
5492                 }
5493                 if (del.br_startoff + del.br_blockcount > bno + 1)
5494                         del.br_blockcount = bno + 1 - del.br_startoff;
5495                 sum = del.br_startblock + del.br_blockcount;
5496                 if (isrt &&
5497                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5498                         /*
5499                          * Realtime extent not lined up at the end.
5500                          * The extent could have been split into written
5501                          * and unwritten pieces, or we could just be
5502                          * unmapping part of it.  But we can't really
5503                          * get rid of part of a realtime extent.
5504                          */
5505                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5506                             !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
5507                                 /*
5508                                  * This piece is unwritten, or we're not
5509                                  * using unwritten extents.  Skip over it.
5510                                  */
5511                                 ASSERT(bno >= mod);
5512                                 bno -= mod > del.br_blockcount ?
5513                                         del.br_blockcount : mod;
5514                                 if (bno < got.br_startoff) {
5515                                         if (--lastx >= 0)
5516                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5517                                                         ifp, lastx), &got);
5518                                 }
5519                                 continue;
5520                         }
5521                         /*
5522                          * It's written, turn it unwritten.
5523                          * This is better than zeroing it.
5524                          */
5525                         ASSERT(del.br_state == XFS_EXT_NORM);
5526                         ASSERT(xfs_trans_get_block_res(tp) > 0);
5527                         /*
5528                          * If this spans a realtime extent boundary,
5529                          * chop it back to the start of the one we end at.
5530                          */
5531                         if (del.br_blockcount > mod) {
5532                                 del.br_startoff += del.br_blockcount - mod;
5533                                 del.br_startblock += del.br_blockcount - mod;
5534                                 del.br_blockcount = mod;
5535                         }
5536                         del.br_state = XFS_EXT_UNWRITTEN;
5537                         error = xfs_bmap_add_extent(ip, lastx, &cur, &del,
5538                                 firstblock, flist, &logflags, delta,
5539                                 XFS_DATA_FORK, 0);
5540                         if (error)
5541                                 goto error0;
5542                         goto nodelete;
5543                 }
5544                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5545                         /*
5546                          * Realtime extent is lined up at the end but not
5547                          * at the front.  We'll get rid of full extents if
5548                          * we can.
5549                          */
5550                         mod = mp->m_sb.sb_rextsize - mod;
5551                         if (del.br_blockcount > mod) {
5552                                 del.br_blockcount -= mod;
5553                                 del.br_startoff += mod;
5554                                 del.br_startblock += mod;
5555                         } else if ((del.br_startoff == start &&
5556                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5557                                      xfs_trans_get_block_res(tp) == 0)) ||
5558                                    !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
5559                                 /*
5560                                  * Can't make it unwritten.  There isn't
5561                                  * a full extent here so just skip it.
5562                                  */
5563                                 ASSERT(bno >= del.br_blockcount);
5564                                 bno -= del.br_blockcount;
5565                                 if (bno < got.br_startoff) {
5566                                         if (--lastx >= 0)
5567                                                 xfs_bmbt_get_all(--ep, &got);
5568                                 }
5569                                 continue;
5570                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5571                                 /*
5572                                  * This one is already unwritten.
5573                                  * It must have a written left neighbor.
5574                                  * Unwrite the killed part of that one and
5575                                  * try again.
5576                                  */
5577                                 ASSERT(lastx > 0);
5578                                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5579                                                 lastx - 1), &prev);
5580                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5581                                 ASSERT(!ISNULLSTARTBLOCK(prev.br_startblock));
5582                                 ASSERT(del.br_startblock ==
5583                                        prev.br_startblock + prev.br_blockcount);
5584                                 if (prev.br_startoff < start) {
5585                                         mod = start - prev.br_startoff;
5586                                         prev.br_blockcount -= mod;
5587                                         prev.br_startblock += mod;
5588                                         prev.br_startoff = start;
5589                                 }
5590                                 prev.br_state = XFS_EXT_UNWRITTEN;
5591                                 error = xfs_bmap_add_extent(ip, lastx - 1, &cur,
5592                                         &prev, firstblock, flist, &logflags,
5593                                         delta, XFS_DATA_FORK, 0);
5594                                 if (error)
5595                                         goto error0;
5596                                 goto nodelete;
5597                         } else {
5598                                 ASSERT(del.br_state == XFS_EXT_NORM);
5599                                 del.br_state = XFS_EXT_UNWRITTEN;
5600                                 error = xfs_bmap_add_extent(ip, lastx, &cur,
5601                                         &del, firstblock, flist, &logflags,
5602                                         delta, XFS_DATA_FORK, 0);
5603                                 if (error)
5604                                         goto error0;
5605                                 goto nodelete;
5606                         }
5607                 }
5608                 if (wasdel) {
5609                         ASSERT(STARTBLOCKVAL(del.br_startblock) > 0);
5610                         /* Update realtime/data freespace, unreserve quota */
5611                         if (isrt) {
5612                                 xfs_filblks_t rtexts;
5613
5614                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5615                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5616                                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5617                                                 (int)rtexts, rsvd);
5618                                 (void)XFS_TRANS_RESERVE_QUOTA_NBLKS(mp,
5619                                         NULL, ip, -((long)del.br_blockcount), 0,
5620                                         XFS_QMOPT_RES_RTBLKS);
5621                         } else {
5622                                 xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS,
5623                                                 (int)del.br_blockcount, rsvd);
5624                                 (void)XFS_TRANS_RESERVE_QUOTA_NBLKS(mp,
5625                                         NULL, ip, -((long)del.br_blockcount), 0,
5626                                         XFS_QMOPT_RES_REGBLKS);
5627                         }
5628                         ip->i_delayed_blks -= del.br_blockcount;
5629                         if (cur)
5630                                 cur->bc_private.b.flags |=
5631                                         XFS_BTCUR_BPRV_WASDEL;
5632                 } else if (cur)
5633                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5634                 /*
5635                  * If it's the case where the directory code is running
5636                  * with no block reservation, and the deleted block is in
5637                  * the middle of its extent, and the resulting insert
5638                  * of an extent would cause transformation to btree format,
5639                  * then reject it.  The calling code will then swap
5640                  * blocks around instead.
5641                  * We have to do this now, rather than waiting for the
5642                  * conversion to btree format, since the transaction
5643                  * will be dirty.
5644                  */
5645                 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5646                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5647                     XFS_IFORK_NEXTENTS(ip, whichfork) >= ifp->if_ext_max &&
5648                     del.br_startoff > got.br_startoff &&
5649                     del.br_startoff + del.br_blockcount <
5650                     got.br_startoff + got.br_blockcount) {
5651                         error = XFS_ERROR(ENOSPC);
5652                         goto error0;
5653                 }
5654                 error = xfs_bmap_del_extent(ip, tp, lastx, flist, cur, &del,
5655                                 &tmp_logflags, delta, whichfork, rsvd);
5656                 logflags |= tmp_logflags;
5657                 if (error)
5658                         goto error0;
5659                 bno = del.br_startoff - 1;
5660 nodelete:
5661                 lastx = ifp->if_lastex;
5662                 /*
5663                  * If not done go on to the next (previous) record.
5664                  * Reset ep in case the extents array was re-alloced.
5665                  */
5666                 ep = xfs_iext_get_ext(ifp, lastx);
5667                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5668                         if (lastx >= XFS_IFORK_NEXTENTS(ip, whichfork) ||
5669                             xfs_bmbt_get_startoff(ep) > bno) {
5670                                 if (--lastx >= 0)
5671                                         ep = xfs_iext_get_ext(ifp, lastx);
5672                         }
5673                         if (lastx >= 0)
5674                                 xfs_bmbt_get_all(ep, &got);
5675                         extno++;
5676                 }
5677         }
5678         ifp->if_lastex = lastx;
5679         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5680         ASSERT(ifp->if_ext_max ==
5681                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5682         /*
5683          * Convert to a btree if necessary.
5684          */
5685         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5686             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
5687                 ASSERT(cur == NULL);
5688                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5689                         &cur, 0, &tmp_logflags, whichfork);
5690                 logflags |= tmp_logflags;
5691                 if (error)
5692                         goto error0;
5693         }
5694         /*
5695          * transform from btree to extents, give it cur
5696          */
5697         else if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5698                  XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5699                 ASSERT(cur != NULL);
5700                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5701                         whichfork);
5702                 logflags |= tmp_logflags;
5703                 if (error)
5704                         goto error0;
5705         }
5706         /*
5707          * transform from extents to local?
5708          */
5709         ASSERT(ifp->if_ext_max ==
5710                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5711         error = 0;
5712         if (delta && delta->xed_startoff != NULLFILEOFF) {
5713                 /* A change was actually made.
5714                  * Note that delta->xed_blockount is an offset at this
5715                  * point and needs to be converted to a block count.
5716                  */
5717                 ASSERT(delta->xed_blockcount > delta->xed_startoff);
5718                 delta->xed_blockcount -= delta->xed_startoff;
5719         }
5720 error0:
5721         /*
5722          * Log everything.  Do this after conversion, there's no point in
5723          * logging the extent records if we've converted to btree format.
5724          */
5725         if ((logflags & XFS_ILOG_FEXT(whichfork)) &&
5726             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5727                 logflags &= ~XFS_ILOG_FEXT(whichfork);
5728         else if ((logflags & XFS_ILOG_FBROOT(whichfork)) &&
5729                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5730                 logflags &= ~XFS_ILOG_FBROOT(whichfork);
5731         /*
5732          * Log inode even in the error case, if the transaction
5733          * is dirty we'll need to shut down the filesystem.
5734          */
5735         if (logflags)
5736                 xfs_trans_log_inode(tp, ip, logflags);
5737         if (cur) {
5738                 if (!error) {
5739                         *firstblock = cur->bc_private.b.firstblock;
5740                         cur->bc_private.b.allocated = 0;
5741                 }
5742                 xfs_btree_del_cursor(cur,
5743                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5744         }
5745         return error;
5746 }
5747
5748 /*
5749  * Fcntl interface to xfs_bmapi.
5750  */
5751 int                                             /* error code */
5752 xfs_getbmap(
5753         bhv_desc_t              *bdp,           /* XFS behavior descriptor*/
5754         struct getbmap          *bmv,           /* user bmap structure */
5755         void                    __user *ap,     /* pointer to user's array */
5756         int                     interface)      /* interface flags */
5757 {
5758         __int64_t               bmvend;         /* last block requested */
5759         int                     error;          /* return value */
5760         __int64_t               fixlen;         /* length for -1 case */
5761         int                     i;              /* extent number */
5762         xfs_inode_t             *ip;            /* xfs incore inode pointer */
5763         bhv_vnode_t             *vp;            /* corresponding vnode */
5764         int                     lock;           /* lock state */
5765         xfs_bmbt_irec_t         *map;           /* buffer for user's data */
5766         xfs_mount_t             *mp;            /* file system mount point */
5767         int                     nex;            /* # of user extents can do */
5768         int                     nexleft;        /* # of user extents left */
5769         int                     subnex;         /* # of bmapi's can do */
5770         int                     nmap;           /* number of map entries */
5771         struct getbmap          out;            /* output structure */
5772         int                     whichfork;      /* data or attr fork */
5773         int                     prealloced;     /* this is a file with
5774                                                  * preallocated data space */
5775         int                     sh_unwritten;   /* true, if unwritten */
5776                                                 /* extents listed separately */
5777         int                     bmapi_flags;    /* flags for xfs_bmapi */
5778         __int32_t               oflags;         /* getbmapx bmv_oflags field */
5779
5780         vp = BHV_TO_VNODE(bdp);
5781         ip = XFS_BHVTOI(bdp);
5782         mp = ip->i_mount;
5783
5784         whichfork = interface & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5785         sh_unwritten = (interface & BMV_IF_PREALLOC) != 0;
5786
5787         /*      If the BMV_IF_NO_DMAPI_READ interface bit specified, do not
5788          *      generate a DMAPI read event.  Otherwise, if the DM_EVENT_READ
5789          *      bit is set for the file, generate a read event in order
5790          *      that the DMAPI application may do its thing before we return
5791          *      the extents.  Usually this means restoring user file data to
5792          *      regions of the file that look like holes.
5793          *
5794          *      The "old behavior" (from XFS_IOC_GETBMAP) is to not specify
5795          *      BMV_IF_NO_DMAPI_READ so that read events are generated.
5796          *      If this were not true, callers of ioctl( XFS_IOC_GETBMAP )
5797          *      could misinterpret holes in a DMAPI file as true holes,
5798          *      when in fact they may represent offline user data.
5799          */
5800         if (   (interface & BMV_IF_NO_DMAPI_READ) == 0
5801             && DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_READ)
5802             && whichfork == XFS_DATA_FORK) {
5803
5804                 error = XFS_SEND_DATA(mp, DM_EVENT_READ, vp, 0, 0, 0, NULL);
5805                 if (error)
5806                         return XFS_ERROR(error);
5807         }
5808
5809         if (whichfork == XFS_ATTR_FORK) {
5810                 if (XFS_IFORK_Q(ip)) {
5811                         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5812                             ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5813                             ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5814                                 return XFS_ERROR(EINVAL);
5815                 } else if (unlikely(
5816                            ip->i_d.di_aformat != 0 &&
5817                            ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5818                         XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5819                                          ip->i_mount);
5820                         return XFS_ERROR(EFSCORRUPTED);
5821                 }
5822         } else if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5823                    ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5824                    ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5825                 return XFS_ERROR(EINVAL);
5826         if (whichfork == XFS_DATA_FORK) {
5827                 if ((ip->i_d.di_extsize && (ip->i_d.di_flags &
5828                                 (XFS_DIFLAG_REALTIME|XFS_DIFLAG_EXTSIZE))) ||
5829                     ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5830                         prealloced = 1;
5831                         fixlen = XFS_MAXIOFFSET(mp);
5832                 } else {
5833                         prealloced = 0;
5834                         fixlen = ip->i_d.di_size;
5835                 }
5836         } else {
5837                 prealloced = 0;
5838                 fixlen = 1LL << 32;
5839         }
5840
5841         if (bmv->bmv_length == -1) {
5842                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5843                 bmv->bmv_length = MAX( (__int64_t)(fixlen - bmv->bmv_offset),
5844                                         (__int64_t)0);
5845         } else if (bmv->bmv_length < 0)
5846                 return XFS_ERROR(EINVAL);
5847         if (bmv->bmv_length == 0) {
5848                 bmv->bmv_entries = 0;
5849                 return 0;
5850         }
5851         nex = bmv->bmv_count - 1;
5852         if (nex <= 0)
5853                 return XFS_ERROR(EINVAL);
5854         bmvend = bmv->bmv_offset + bmv->bmv_length;
5855
5856         xfs_ilock(ip, XFS_IOLOCK_SHARED);
5857
5858         if (whichfork == XFS_DATA_FORK && ip->i_delayed_blks) {
5859                 /* xfs_fsize_t last_byte = xfs_file_last_byte(ip); */
5860                 error = bhv_vop_flush_pages(vp, (xfs_off_t)0, -1, 0, FI_REMAPF);
5861         }
5862
5863         ASSERT(whichfork == XFS_ATTR_FORK || ip->i_delayed_blks == 0);
5864
5865         lock = xfs_ilock_map_shared(ip);
5866
5867         /*
5868          * Don't let nex be bigger than the number of extents
5869          * we can have assuming alternating holes and real extents.
5870          */
5871         if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5872                 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5873
5874         bmapi_flags = XFS_BMAPI_AFLAG(whichfork) |
5875                         ((sh_unwritten) ? 0 : XFS_BMAPI_IGSTATE);
5876
5877         /*
5878          * Allocate enough space to handle "subnex" maps at a time.
5879          */
5880         subnex = 16;
5881         map = kmem_alloc(subnex * sizeof(*map), KM_SLEEP);
5882
5883         bmv->bmv_entries = 0;
5884
5885         if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0) {
5886                 error = 0;
5887                 goto unlock_and_return;
5888         }
5889
5890         nexleft = nex;
5891
5892         do {
5893                 nmap = (nexleft > subnex) ? subnex : nexleft;
5894                 error = xfs_bmapi(NULL, ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5895                                   XFS_BB_TO_FSB(mp, bmv->bmv_length),
5896                                   bmapi_flags, NULL, 0, map, &nmap,
5897                                   NULL, NULL);
5898                 if (error)
5899                         goto unlock_and_return;
5900                 ASSERT(nmap <= subnex);
5901
5902                 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5903                         nexleft--;
5904                         oflags = (map[i].br_state == XFS_EXT_UNWRITTEN) ?
5905                                         BMV_OF_PREALLOC : 0;
5906                         out.bmv_offset = XFS_FSB_TO_BB(mp, map[i].br_startoff);
5907                         out.bmv_length = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5908                         ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
5909                         if (map[i].br_startblock == HOLESTARTBLOCK &&
5910                            ((prealloced && out.bmv_offset + out.bmv_length == bmvend) ||
5911                              whichfork == XFS_ATTR_FORK )) {
5912                                 /*
5913                                  * came to hole at end of file or the end of
5914                                    attribute fork
5915                                  */
5916                                 goto unlock_and_return;
5917                         } else {
5918                                 out.bmv_block =
5919                                     (map[i].br_startblock == HOLESTARTBLOCK) ?
5920                                         -1 :
5921                                         XFS_FSB_TO_DB(ip, map[i].br_startblock);
5922
5923                                 /* return either getbmap/getbmapx structure. */
5924                                 if (interface & BMV_IF_EXTENDED) {
5925                                         struct  getbmapx        outx;
5926
5927                                         GETBMAP_CONVERT(out,outx);
5928                                         outx.bmv_oflags = oflags;
5929                                         outx.bmv_unused1 = outx.bmv_unused2 = 0;
5930                                         if (copy_to_user(ap, &outx,
5931                                                         sizeof(outx))) {
5932                                                 error = XFS_ERROR(EFAULT);
5933                                                 goto unlock_and_return;
5934                                         }
5935                                 } else {
5936                                         if (copy_to_user(ap, &out,
5937                                                         sizeof(out))) {
5938                                                 error = XFS_ERROR(EFAULT);
5939                                                 goto unlock_and_return;
5940                                         }
5941                                 }
5942                                 bmv->bmv_offset =
5943                                         out.bmv_offset + out.bmv_length;
5944                                 bmv->bmv_length = MAX((__int64_t)0,
5945                                         (__int64_t)(bmvend - bmv->bmv_offset));
5946                                 bmv->bmv_entries++;
5947                                 ap = (interface & BMV_IF_EXTENDED) ?
5948                                                 (void __user *)
5949                                         ((struct getbmapx __user *)ap + 1) :
5950                                                 (void __user *)
5951                                         ((struct getbmap __user *)ap + 1);
5952                         }
5953                 }
5954         } while (nmap && nexleft && bmv->bmv_length);
5955
5956 unlock_and_return:
5957         xfs_iunlock_map_shared(ip, lock);
5958         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5959
5960         kmem_free(map, subnex * sizeof(*map));
5961
5962         return error;
5963 }
5964
5965 /*
5966  * Check the last inode extent to determine whether this allocation will result
5967  * in blocks being allocated at the end of the file. When we allocate new data
5968  * blocks at the end of the file which do not start at the previous data block,
5969  * we will try to align the new blocks at stripe unit boundaries.
5970  */
5971 STATIC int                              /* error */
5972 xfs_bmap_isaeof(
5973         xfs_inode_t     *ip,            /* incore inode pointer */
5974         xfs_fileoff_t   off,            /* file offset in fsblocks */
5975         int             whichfork,      /* data or attribute fork */
5976         char            *aeof)          /* return value */
5977 {
5978         int             error;          /* error return value */
5979         xfs_ifork_t     *ifp;           /* inode fork pointer */
5980         xfs_bmbt_rec_t  *lastrec;       /* extent record pointer */
5981         xfs_extnum_t    nextents;       /* number of file extents */
5982         xfs_bmbt_irec_t s;              /* expanded extent record */
5983
5984         ASSERT(whichfork == XFS_DATA_FORK);
5985         ifp = XFS_IFORK_PTR(ip, whichfork);
5986         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5987             (error = xfs_iread_extents(NULL, ip, whichfork)))
5988                 return error;
5989         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5990         if (nextents == 0) {
5991                 *aeof = 1;
5992                 return 0;
5993         }
5994         /*
5995          * Go to the last extent
5996          */
5997         lastrec = xfs_iext_get_ext(ifp, nextents - 1);
5998         xfs_bmbt_get_all(lastrec, &s);
5999         /*
6000          * Check we are allocating in the last extent (for delayed allocations)
6001          * or past the last extent for non-delayed allocations.
6002          */
6003         *aeof = (off >= s.br_startoff &&
6004                  off < s.br_startoff + s.br_blockcount &&
6005                  ISNULLSTARTBLOCK(s.br_startblock)) ||
6006                 off >= s.br_startoff + s.br_blockcount;
6007         return 0;
6008 }
6009
6010 /*
6011  * Check if the endoff is outside the last extent. If so the caller will grow
6012  * the allocation to a stripe unit boundary.
6013  */
6014 int                                     /* error */
6015 xfs_bmap_eof(
6016         xfs_inode_t     *ip,            /* incore inode pointer */
6017         xfs_fileoff_t   endoff,         /* file offset in fsblocks */
6018         int             whichfork,      /* data or attribute fork */
6019         int             *eof)           /* result value */
6020 {
6021         xfs_fsblock_t   blockcount;     /* extent block count */
6022         int             error;          /* error return value */
6023         xfs_ifork_t     *ifp;           /* inode fork pointer */
6024         xfs_bmbt_rec_t  *lastrec;       /* extent record pointer */
6025         xfs_extnum_t    nextents;       /* number of file extents */
6026         xfs_fileoff_t   startoff;       /* extent starting file offset */
6027
6028         ASSERT(whichfork == XFS_DATA_FORK);
6029         ifp = XFS_IFORK_PTR(ip, whichfork);
6030         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
6031             (error = xfs_iread_extents(NULL, ip, whichfork)))
6032                 return error;
6033         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
6034         if (nextents == 0) {
6035                 *eof = 1;
6036                 return 0;
6037         }
6038         /*
6039          * Go to the last extent
6040          */
6041         lastrec = xfs_iext_get_ext(ifp, nextents - 1);
6042         startoff = xfs_bmbt_get_startoff(lastrec);
6043         blockcount = xfs_bmbt_get_blockcount(lastrec);
6044         *eof = endoff >= startoff + blockcount;
6045         return 0;
6046 }
6047
6048 #ifdef DEBUG
6049 /*
6050  * Check that the extents list for the inode ip is in the right order.
6051  */
6052 STATIC void
6053 xfs_bmap_check_extents(
6054         xfs_inode_t             *ip,            /* incore inode pointer */
6055         int                     whichfork)      /* data or attr fork */
6056 {
6057         xfs_bmbt_rec_t          *ep;            /* current extent entry */
6058         xfs_extnum_t            idx;            /* extent record index */
6059         xfs_ifork_t             *ifp;           /* inode fork pointer */
6060         xfs_extnum_t            nextents;       /* number of extents in list */
6061         xfs_bmbt_rec_t          *nextp;         /* next extent entry */
6062
6063         ifp = XFS_IFORK_PTR(ip, whichfork);
6064         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
6065         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
6066         ep = xfs_iext_get_ext(ifp, 0);
6067         for (idx = 0; idx < nextents - 1; idx++) {
6068                 nextp = xfs_iext_get_ext(ifp, idx + 1);
6069                 xfs_btree_check_rec(XFS_BTNUM_BMAP, (void *)ep,
6070                         (void *)(nextp));
6071                 ep = nextp;
6072         }
6073 }
6074
6075 STATIC
6076 xfs_buf_t *
6077 xfs_bmap_get_bp(
6078         xfs_btree_cur_t         *cur,
6079         xfs_fsblock_t           bno)
6080 {
6081         int i;
6082         xfs_buf_t *bp;
6083
6084         if (!cur)
6085                 return(NULL);
6086
6087         bp = NULL;
6088         for(i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
6089                 bp = cur->bc_bufs[i];
6090                 if (!bp) break;
6091                 if (XFS_BUF_ADDR(bp) == bno)
6092                         break;  /* Found it */
6093         }
6094         if (i == XFS_BTREE_MAXLEVELS)
6095                 bp = NULL;
6096
6097         if (!bp) { /* Chase down all the log items to see if the bp is there */
6098                 xfs_log_item_chunk_t    *licp;
6099                 xfs_trans_t             *tp;
6100
6101                 tp = cur->bc_tp;
6102                 licp = &tp->t_items;
6103                 while (!bp && licp != NULL) {
6104                         if (XFS_LIC_ARE_ALL_FREE(licp)) {
6105                                 licp = licp->lic_next;
6106                                 continue;
6107                         }
6108                         for (i = 0; i < licp->lic_unused; i++) {
6109                                 xfs_log_item_desc_t     *lidp;
6110                                 xfs_log_item_t          *lip;
6111                                 xfs_buf_log_item_t      *bip;
6112                                 xfs_buf_t               *lbp;
6113
6114                                 if (XFS_LIC_ISFREE(licp, i)) {
6115                                         continue;
6116                                 }
6117
6118                                 lidp = XFS_LIC_SLOT(licp, i);
6119                                 lip = lidp->lid_item;
6120                                 if (lip->li_type != XFS_LI_BUF)
6121                                         continue;
6122
6123                                 bip = (xfs_buf_log_item_t *)lip;
6124                                 lbp = bip->bli_buf;
6125
6126                                 if (XFS_BUF_ADDR(lbp) == bno) {
6127                                         bp = lbp;
6128                                         break; /* Found it */
6129                                 }
6130                         }
6131                         licp = licp->lic_next;
6132                 }
6133         }
6134         return(bp);
6135 }
6136
6137 void
6138 xfs_check_block(
6139         xfs_bmbt_block_t        *block,
6140         xfs_mount_t             *mp,
6141         int                     root,
6142         short                   sz)
6143 {
6144         int                     i, j, dmxr;
6145         __be64                  *pp, *thispa;   /* pointer to block address */
6146         xfs_bmbt_key_t          *prevp, *keyp;
6147
6148         ASSERT(be16_to_cpu(block->bb_level) > 0);
6149
6150         prevp = NULL;
6151         for( i = 1; i <= be16_to_cpu(block->bb_numrecs); i++) {
6152                 dmxr = mp->m_bmap_dmxr[0];
6153
6154                 if (root) {
6155                         keyp = XFS_BMAP_BROOT_KEY_ADDR(block, i, sz);
6156                 } else {
6157                         keyp = XFS_BTREE_KEY_ADDR(mp->m_sb.sb_blocksize,
6158                                 xfs_bmbt, block, i, dmxr);
6159                 }
6160
6161                 if (prevp) {
6162                         xfs_btree_check_key(XFS_BTNUM_BMAP, prevp, keyp);
6163                 }
6164                 prevp = keyp;
6165
6166                 /*
6167                  * Compare the block numbers to see if there are dups.
6168                  */
6169
6170                 if (root) {
6171                         pp = XFS_BMAP_BROOT_PTR_ADDR(block, i, sz);
6172                 } else {
6173                         pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize,
6174                                 xfs_bmbt, block, i, dmxr);
6175                 }
6176                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
6177                         if (root) {
6178                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(block, j, sz);
6179                         } else {
6180                                 thispa = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize,
6181                                         xfs_bmbt, block, j, dmxr);
6182                         }
6183                         if (*thispa == *pp) {
6184                                 cmn_err(CE_WARN, "%s: thispa(%d) == pp(%d) %Ld",
6185                                         __FUNCTION__, j, i,
6186                                         (unsigned long long)be64_to_cpu(*thispa));
6187                                 panic("%s: ptrs are equal in node\n",
6188                                         __FUNCTION__);
6189                         }
6190                 }
6191         }
6192 }
6193
6194 /*
6195  * Check that the extents for the inode ip are in the right order in all
6196  * btree leaves.
6197  */
6198
6199 STATIC void
6200 xfs_bmap_check_leaf_extents(
6201         xfs_btree_cur_t         *cur,   /* btree cursor or null */
6202         xfs_inode_t             *ip,            /* incore inode pointer */
6203         int                     whichfork)      /* data or attr fork */
6204 {
6205         xfs_bmbt_block_t        *block; /* current btree block */
6206         xfs_fsblock_t           bno;    /* block # of "block" */
6207         xfs_buf_t               *bp;    /* buffer for "block" */
6208         int                     error;  /* error return value */
6209         xfs_extnum_t            i=0, j; /* index into the extents list */
6210         xfs_ifork_t             *ifp;   /* fork structure */
6211         int                     level;  /* btree level, for checking */
6212         xfs_mount_t             *mp;    /* file system mount structure */
6213         __be64                  *pp;    /* pointer to block address */
6214         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
6215         xfs_bmbt_rec_t          *lastp; /* pointer to previous extent */
6216         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
6217         int                     bp_release = 0;
6218
6219         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
6220                 return;
6221         }
6222
6223         bno = NULLFSBLOCK;
6224         mp = ip->i_mount;
6225         ifp = XFS_IFORK_PTR(ip, whichfork);
6226         block = ifp->if_broot;
6227         /*
6228          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6229          */
6230         level = be16_to_cpu(block->bb_level);
6231         ASSERT(level > 0);
6232         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
6233         pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
6234         bno = be64_to_cpu(*pp);
6235
6236         ASSERT(bno != NULLDFSBNO);
6237         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
6238         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
6239
6240         /*
6241          * Go down the tree until leaf level is reached, following the first
6242          * pointer (leftmost) at each level.
6243          */
6244         while (level-- > 0) {
6245                 /* See if buf is in cur first */
6246                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
6247                 if (bp) {
6248                         bp_release = 0;
6249                 } else {
6250                         bp_release = 1;
6251                 }
6252                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
6253                                 XFS_BMAP_BTREE_REF)))
6254                         goto error_norelse;
6255                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6256                 XFS_WANT_CORRUPTED_GOTO(
6257                         XFS_BMAP_SANITY_CHECK(mp, block, level),
6258                         error0);
6259                 if (level == 0)
6260                         break;
6261
6262                 /*
6263                  * Check this block for basic sanity (increasing keys and
6264                  * no duplicate blocks).
6265                  */
6266
6267                 xfs_check_block(block, mp, 0, 0);
6268                 pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt, block,
6269                         1, mp->m_bmap_dmxr[1]);
6270                 bno = be64_to_cpu(*pp);
6271                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
6272                 if (bp_release) {
6273                         bp_release = 0;
6274                         xfs_trans_brelse(NULL, bp);
6275                 }
6276         }
6277
6278         /*
6279          * Here with bp and block set to the leftmost leaf node in the tree.
6280          */
6281         i = 0;
6282
6283         /*
6284          * Loop over all leaf nodes checking that all extents are in the right order.
6285          */
6286         lastp = NULL;
6287         for (;;) {
6288                 xfs_fsblock_t   nextbno;
6289                 xfs_extnum_t    num_recs;
6290
6291
6292                 num_recs = be16_to_cpu(block->bb_numrecs);
6293
6294                 /*
6295                  * Read-ahead the next leaf block, if any.
6296                  */
6297
6298                 nextbno = be64_to_cpu(block->bb_rightsib);
6299
6300                 /*
6301                  * Check all the extents to make sure they are OK.
6302                  * If we had a previous block, the last entry should
6303                  * conform with the first entry in this one.
6304                  */
6305
6306                 ep = XFS_BTREE_REC_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt,
6307                         block, 1, mp->m_bmap_dmxr[0]);
6308                 for (j = 1; j < num_recs; j++) {
6309                         nextp = XFS_BTREE_REC_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt,
6310                                 block, j + 1, mp->m_bmap_dmxr[0]);
6311                         if (lastp) {
6312                                 xfs_btree_check_rec(XFS_BTNUM_BMAP,
6313                                         (void *)lastp, (void *)ep);
6314                         }
6315                         xfs_btree_check_rec(XFS_BTNUM_BMAP, (void *)ep,
6316                                 (void *)(nextp));
6317                         lastp = ep;
6318                         ep = nextp;
6319                 }
6320
6321                 i += num_recs;
6322                 if (bp_release) {
6323                         bp_release = 0;
6324                         xfs_trans_brelse(NULL, bp);
6325                 }
6326                 bno = nextbno;
6327                 /*
6328                  * If we've reached the end, stop.
6329                  */
6330                 if (bno == NULLFSBLOCK)
6331                         break;
6332
6333                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
6334                 if (bp) {
6335                         bp_release = 0;
6336                 } else {
6337                         bp_release = 1;
6338                 }
6339                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
6340                                 XFS_BMAP_BTREE_REF)))
6341                         goto error_norelse;
6342                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6343         }
6344         if (bp_release) {
6345                 bp_release = 0;
6346                 xfs_trans_brelse(NULL, bp);
6347         }
6348         return;
6349
6350 error0:
6351         cmn_err(CE_WARN, "%s: at error0", __FUNCTION__);
6352         if (bp_release)
6353                 xfs_trans_brelse(NULL, bp);
6354 error_norelse:
6355         cmn_err(CE_WARN, "%s: BAD after btree leaves for %d extents",
6356                 __FUNCTION__, i);
6357         panic("%s: CORRUPTED BTREE OR SOMETHING", __FUNCTION__);
6358         return;
6359 }
6360 #endif
6361
6362 /*
6363  * Count fsblocks of the given fork.
6364  */
6365 int                                             /* error */
6366 xfs_bmap_count_blocks(
6367         xfs_trans_t             *tp,            /* transaction pointer */
6368         xfs_inode_t             *ip,            /* incore inode */
6369         int                     whichfork,      /* data or attr fork */
6370         int                     *count)         /* out: count of blocks */
6371 {
6372         xfs_bmbt_block_t        *block; /* current btree block */
6373         xfs_fsblock_t           bno;    /* block # of "block" */
6374         xfs_ifork_t             *ifp;   /* fork structure */
6375         int                     level;  /* btree level, for checking */
6376         xfs_mount_t             *mp;    /* file system mount structure */
6377         __be64                  *pp;    /* pointer to block address */
6378
6379         bno = NULLFSBLOCK;
6380         mp = ip->i_mount;
6381         ifp = XFS_IFORK_PTR(ip, whichfork);
6382         if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
6383                 if (unlikely(xfs_bmap_count_leaves(ifp, 0,
6384                         ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
6385                         count) < 0)) {
6386                         XFS_ERROR_REPORT("xfs_bmap_count_blocks(1)",
6387                                          XFS_ERRLEVEL_LOW, mp);
6388                         return XFS_ERROR(EFSCORRUPTED);
6389                 }
6390                 return 0;
6391         }
6392
6393         /*
6394          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6395          */
6396         block = ifp->if_broot;
6397         level = be16_to_cpu(block->bb_level);
6398         ASSERT(level > 0);
6399         pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
6400         bno = be64_to_cpu(*pp);
6401         ASSERT(bno != NULLDFSBNO);
6402         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
6403         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
6404
6405         if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
6406                 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
6407                                  mp);
6408                 return XFS_ERROR(EFSCORRUPTED);
6409         }
6410
6411         return 0;
6412 }
6413
6414 /*
6415  * Recursively walks each level of a btree
6416  * to count total fsblocks is use.
6417  */
6418 int                                     /* error */
6419 xfs_bmap_count_tree(
6420         xfs_mount_t     *mp,            /* file system mount point */
6421         xfs_trans_t     *tp,            /* transaction pointer */
6422         xfs_ifork_t     *ifp,           /* inode fork pointer */
6423         xfs_fsblock_t   blockno,        /* file system block number */
6424         int             levelin,        /* level in btree */
6425         int             *count)         /* Count of blocks */
6426 {
6427         int                     error;
6428         xfs_buf_t               *bp, *nbp;
6429         int                     level = levelin;
6430         __be64                  *pp;
6431         xfs_fsblock_t           bno = blockno;
6432         xfs_fsblock_t           nextbno;
6433         xfs_bmbt_block_t        *block, *nextblock;
6434         int                     numrecs;
6435
6436         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
6437                 return error;
6438         *count += 1;
6439         block = XFS_BUF_TO_BMBT_BLOCK(bp);
6440
6441         if (--level) {
6442                 /* Not at node above leafs, count this level of nodes */
6443                 nextbno = be64_to_cpu(block->bb_rightsib);
6444                 while (nextbno != NULLFSBLOCK) {
6445                         if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
6446                                 0, &nbp, XFS_BMAP_BTREE_REF)))
6447                                 return error;
6448                         *count += 1;
6449                         nextblock = XFS_BUF_TO_BMBT_BLOCK(nbp);
6450                         nextbno = be64_to_cpu(nextblock->bb_rightsib);
6451                         xfs_trans_brelse(tp, nbp);
6452                 }
6453
6454                 /* Dive to the next level */
6455                 pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize,
6456                         xfs_bmbt, block, 1, mp->m_bmap_dmxr[1]);
6457                 bno = be64_to_cpu(*pp);
6458                 if (unlikely((error =
6459                      xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
6460                         xfs_trans_brelse(tp, bp);
6461                         XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6462                                          XFS_ERRLEVEL_LOW, mp);
6463                         return XFS_ERROR(EFSCORRUPTED);
6464                 }
6465                 xfs_trans_brelse(tp, bp);
6466         } else {
6467                 /* count all level 1 nodes and their leaves */
6468                 for (;;) {
6469                         nextbno = be64_to_cpu(block->bb_rightsib);
6470                         numrecs = be16_to_cpu(block->bb_numrecs);
6471                         if (unlikely(xfs_bmap_disk_count_leaves(ifp, mp,
6472                                         0, block, numrecs, count) < 0)) {
6473                                 xfs_trans_brelse(tp, bp);
6474                                 XFS_ERROR_REPORT("xfs_bmap_count_tree(2)",
6475                                                  XFS_ERRLEVEL_LOW, mp);
6476                                 return XFS_ERROR(EFSCORRUPTED);
6477                         }
6478                         xfs_trans_brelse(tp, bp);
6479                         if (nextbno == NULLFSBLOCK)
6480                                 break;
6481                         bno = nextbno;
6482                         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6483                                 XFS_BMAP_BTREE_REF)))
6484                                 return error;
6485                         *count += 1;
6486                         block = XFS_BUF_TO_BMBT_BLOCK(bp);
6487                 }
6488         }
6489         return 0;
6490 }
6491
6492 /*
6493  * Count leaf blocks given a range of extent records.
6494  */
6495 int
6496 xfs_bmap_count_leaves(
6497         xfs_ifork_t             *ifp,
6498         xfs_extnum_t            idx,
6499         int                     numrecs,
6500         int                     *count)
6501 {
6502         int             b;
6503         xfs_bmbt_rec_t  *frp;
6504
6505         for (b = 0; b < numrecs; b++) {
6506                 frp = xfs_iext_get_ext(ifp, idx + b);
6507                 *count += xfs_bmbt_get_blockcount(frp);
6508         }
6509         return 0;
6510 }
6511
6512 /*
6513  * Count leaf blocks given a range of extent records originally
6514  * in btree format.
6515  */
6516 int
6517 xfs_bmap_disk_count_leaves(
6518         xfs_ifork_t             *ifp,
6519         xfs_mount_t             *mp,
6520         xfs_extnum_t            idx,
6521         xfs_bmbt_block_t        *block,
6522         int                     numrecs,
6523         int                     *count)
6524 {
6525         int             b;
6526         xfs_bmbt_rec_t  *frp;
6527
6528         for (b = 1; b <= numrecs; b++) {
6529                 frp = XFS_BTREE_REC_ADDR(mp->m_sb.sb_blocksize,
6530                         xfs_bmbt, block, idx + b, mp->m_bmap_dmxr[0]);
6531                 *count += xfs_bmbt_disk_get_blockcount(frp);
6532         }
6533         return 0;
6534 }