Merge /pub/scm/linux/kernel/git/torvalds/linux-2.6
[linux-2.6] / fs / xfs / xfs_bmap.c
1 /*
2  * Copyright (c) 2000-2004 Silicon Graphics, Inc.  All Rights Reserved.
3  *
4  * This program is free software; you can redistribute it and/or modify it
5  * under the terms of version 2 of the GNU General Public License as
6  * published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it would be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
11  *
12  * Further, this software is distributed without any warranty that it is
13  * free of the rightful claim of any third person regarding infringement
14  * or the like.  Any license provided herein, whether implied or
15  * otherwise, applies only to this software file.  Patent licenses, if
16  * any, provided herein do not apply to combinations of this program with
17  * other software, or any other product whatsoever.
18  *
19  * You should have received a copy of the GNU General Public License along
20  * with this program; if not, write the Free Software Foundation, Inc., 59
21  * Temple Place - Suite 330, Boston MA 02111-1307, USA.
22  *
23  * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24  * Mountain View, CA  94043, or:
25  *
26  * http://www.sgi.com
27  *
28  * For further information regarding this notice, see:
29  *
30  * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
31  */
32
33 #include "xfs.h"
34
35 #include "xfs_macros.h"
36 #include "xfs_types.h"
37 #include "xfs_inum.h"
38 #include "xfs_log.h"
39 #include "xfs_trans.h"
40 #include "xfs_sb.h"
41 #include "xfs_ag.h"
42 #include "xfs_dir.h"
43 #include "xfs_dir2.h"
44 #include "xfs_dmapi.h"
45 #include "xfs_mount.h"
46 #include "xfs_alloc_btree.h"
47 #include "xfs_bmap_btree.h"
48 #include "xfs_ialloc_btree.h"
49 #include "xfs_btree.h"
50 #include "xfs_ialloc.h"
51 #include "xfs_attr_sf.h"
52 #include "xfs_dir_sf.h"
53 #include "xfs_dir2_sf.h"
54 #include "xfs_dinode.h"
55 #include "xfs_inode_item.h"
56 #include "xfs_inode.h"
57 #include "xfs_itable.h"
58 #include "xfs_extfree_item.h"
59 #include "xfs_alloc.h"
60 #include "xfs_bmap.h"
61 #include "xfs_rtalloc.h"
62 #include "xfs_error.h"
63 #include "xfs_da_btree.h"
64 #include "xfs_dir_leaf.h"
65 #include "xfs_bit.h"
66 #include "xfs_rw.h"
67 #include "xfs_quota.h"
68 #include "xfs_trans_space.h"
69 #include "xfs_buf_item.h"
70
71
72 #ifdef DEBUG
73 STATIC void
74 xfs_bmap_check_leaf_extents(xfs_btree_cur_t *cur, xfs_inode_t *ip, int whichfork);
75 #endif
76
77 kmem_zone_t             *xfs_bmap_free_item_zone;
78
79 /*
80  * Prototypes for internal bmap routines.
81  */
82
83
84 /*
85  * Called from xfs_bmap_add_attrfork to handle extents format files.
86  */
87 STATIC int                                      /* error */
88 xfs_bmap_add_attrfork_extents(
89         xfs_trans_t             *tp,            /* transaction pointer */
90         xfs_inode_t             *ip,            /* incore inode pointer */
91         xfs_fsblock_t           *firstblock,    /* first block allocated */
92         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
93         int                     *flags);        /* inode logging flags */
94
95 /*
96  * Called from xfs_bmap_add_attrfork to handle local format files.
97  */
98 STATIC int                                      /* error */
99 xfs_bmap_add_attrfork_local(
100         xfs_trans_t             *tp,            /* transaction pointer */
101         xfs_inode_t             *ip,            /* incore inode pointer */
102         xfs_fsblock_t           *firstblock,    /* first block allocated */
103         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
104         int                     *flags);        /* inode logging flags */
105
106 /*
107  * Called by xfs_bmapi to update extent list structure and the btree
108  * after allocating space (or doing a delayed allocation).
109  */
110 STATIC int                              /* error */
111 xfs_bmap_add_extent(
112         xfs_inode_t             *ip,    /* incore inode pointer */
113         xfs_extnum_t            idx,    /* extent number to update/insert */
114         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
115         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
116         xfs_fsblock_t           *first, /* pointer to firstblock variable */
117         xfs_bmap_free_t         *flist, /* list of extents to be freed */
118         int                     *logflagsp, /* inode logging flags */
119         int                     whichfork, /* data or attr fork */
120         int                     rsvd);  /* OK to allocate reserved blocks */
121
122 /*
123  * Called by xfs_bmap_add_extent to handle cases converting a delayed
124  * allocation to a real allocation.
125  */
126 STATIC int                              /* error */
127 xfs_bmap_add_extent_delay_real(
128         xfs_inode_t             *ip,    /* incore inode pointer */
129         xfs_extnum_t            idx,    /* extent number to update/insert */
130         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
131         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
132         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
133         xfs_fsblock_t           *first, /* pointer to firstblock variable */
134         xfs_bmap_free_t         *flist, /* list of extents to be freed */
135         int                     *logflagsp, /* inode logging flags */
136         int                     rsvd);  /* OK to allocate reserved blocks */
137
138 /*
139  * Called by xfs_bmap_add_extent to handle cases converting a hole
140  * to a delayed allocation.
141  */
142 STATIC int                              /* error */
143 xfs_bmap_add_extent_hole_delay(
144         xfs_inode_t             *ip,    /* incore inode pointer */
145         xfs_extnum_t            idx,    /* extent number to update/insert */
146         xfs_btree_cur_t         *cur,   /* if null, not a btree */
147         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
148         int                     *logflagsp,/* inode logging flags */
149         int                     rsvd);  /* OK to allocate reserved blocks */
150
151 /*
152  * Called by xfs_bmap_add_extent to handle cases converting a hole
153  * to a real allocation.
154  */
155 STATIC int                              /* error */
156 xfs_bmap_add_extent_hole_real(
157         xfs_inode_t             *ip,    /* incore inode pointer */
158         xfs_extnum_t            idx,    /* extent number to update/insert */
159         xfs_btree_cur_t         *cur,   /* if null, not a btree */
160         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
161         int                     *logflagsp, /* inode logging flags */
162         int                     whichfork); /* data or attr fork */
163
164 /*
165  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
166  * allocation to a real allocation or vice versa.
167  */
168 STATIC int                              /* error */
169 xfs_bmap_add_extent_unwritten_real(
170         xfs_inode_t             *ip,    /* incore inode pointer */
171         xfs_extnum_t            idx,    /* extent number to update/insert */
172         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
173         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
174         int                     *logflagsp); /* inode logging flags */
175
176 /*
177  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
178  * It figures out where to ask the underlying allocator to put the new extent.
179  */
180 STATIC int                              /* error */
181 xfs_bmap_alloc(
182         xfs_bmalloca_t          *ap);   /* bmap alloc argument struct */
183
184 /*
185  * Transform a btree format file with only one leaf node, where the
186  * extents list will fit in the inode, into an extents format file.
187  * Since the extent list is already in-core, all we have to do is
188  * give up the space for the btree root and pitch the leaf block.
189  */
190 STATIC int                              /* error */
191 xfs_bmap_btree_to_extents(
192         xfs_trans_t             *tp,    /* transaction pointer */
193         xfs_inode_t             *ip,    /* incore inode pointer */
194         xfs_btree_cur_t         *cur,   /* btree cursor */
195         int                     *logflagsp, /* inode logging flags */
196         int                     whichfork); /* data or attr fork */
197
198 #ifdef DEBUG
199 /*
200  * Check that the extents list for the inode ip is in the right order.
201  */
202 STATIC void
203 xfs_bmap_check_extents(
204         xfs_inode_t             *ip,            /* incore inode pointer */
205         int                     whichfork);     /* data or attr fork */
206 #endif
207
208 /*
209  * Called by xfs_bmapi to update extent list structure and the btree
210  * after removing space (or undoing a delayed allocation).
211  */
212 STATIC int                              /* error */
213 xfs_bmap_del_extent(
214         xfs_inode_t             *ip,    /* incore inode pointer */
215         xfs_trans_t             *tp,    /* current trans pointer */
216         xfs_extnum_t            idx,    /* extent number to update/insert */
217         xfs_bmap_free_t         *flist, /* list of extents to be freed */
218         xfs_btree_cur_t         *cur,   /* if null, not a btree */
219         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
220         int                     *logflagsp,/* inode logging flags */
221         int                     whichfork, /* data or attr fork */
222         int                     rsvd);   /* OK to allocate reserved blocks */
223
224 /*
225  * Remove the entry "free" from the free item list.  Prev points to the
226  * previous entry, unless "free" is the head of the list.
227  */
228 STATIC void
229 xfs_bmap_del_free(
230         xfs_bmap_free_t         *flist, /* free item list header */
231         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
232         xfs_bmap_free_item_t    *free); /* list item to be freed */
233
234 /*
235  * Remove count entries from the extents array for inode "ip", starting
236  * at index "idx".  Copies the remaining items down over the deleted ones,
237  * and gives back the excess memory.
238  */
239 STATIC void
240 xfs_bmap_delete_exlist(
241         xfs_inode_t     *ip,            /* incode inode pointer */
242         xfs_extnum_t    idx,            /* starting delete index */
243         xfs_extnum_t    count,          /* count of items to delete */
244         int             whichfork);     /* data or attr fork */
245
246 /*
247  * Convert an extents-format file into a btree-format file.
248  * The new file will have a root block (in the inode) and a single child block.
249  */
250 STATIC int                                      /* error */
251 xfs_bmap_extents_to_btree(
252         xfs_trans_t             *tp,            /* transaction pointer */
253         xfs_inode_t             *ip,            /* incore inode pointer */
254         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
255         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
256         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
257         int                     wasdel,         /* converting a delayed alloc */
258         int                     *logflagsp,     /* inode logging flags */
259         int                     whichfork);     /* data or attr fork */
260
261 /*
262  * Insert new item(s) in the extent list for inode "ip".
263  * Count new items are inserted at offset idx.
264  */
265 STATIC void
266 xfs_bmap_insert_exlist(
267         xfs_inode_t     *ip,            /* incore inode pointer */
268         xfs_extnum_t    idx,            /* starting index of new items */
269         xfs_extnum_t    count,          /* number of inserted items */
270         xfs_bmbt_irec_t *new,           /* items to insert */
271         int             whichfork);     /* data or attr fork */
272
273 /*
274  * Convert a local file to an extents file.
275  * This code is sort of bogus, since the file data needs to get
276  * logged so it won't be lost.  The bmap-level manipulations are ok, though.
277  */
278 STATIC int                              /* error */
279 xfs_bmap_local_to_extents(
280         xfs_trans_t     *tp,            /* transaction pointer */
281         xfs_inode_t     *ip,            /* incore inode pointer */
282         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
283         xfs_extlen_t    total,          /* total blocks needed by transaction */
284         int             *logflagsp,     /* inode logging flags */
285         int             whichfork);     /* data or attr fork */
286
287 /*
288  * Search the extents list for the inode, for the extent containing bno.
289  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
290  * *eofp will be set, and *prevp will contain the last entry (null if none).
291  * Else, *lastxp will be set to the index of the found
292  * entry; *gotp will contain the entry.
293  */
294 STATIC xfs_bmbt_rec_t *                 /* pointer to found extent entry */
295 xfs_bmap_search_extents(
296         xfs_inode_t     *ip,            /* incore inode pointer */
297         xfs_fileoff_t   bno,            /* block number searched for */
298         int             whichfork,      /* data or attr fork */
299         int             *eofp,          /* out: end of file found */
300         xfs_extnum_t    *lastxp,        /* out: last extent index */
301         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
302         xfs_bmbt_irec_t *prevp);        /* out: previous extent entry found */
303
304 /*
305  * Check the last inode extent to determine whether this allocation will result
306  * in blocks being allocated at the end of the file. When we allocate new data
307  * blocks at the end of the file which do not start at the previous data block,
308  * we will try to align the new blocks at stripe unit boundaries.
309  */
310 STATIC int                              /* error */
311 xfs_bmap_isaeof(
312         xfs_inode_t     *ip,            /* incore inode pointer */
313         xfs_fileoff_t   off,            /* file offset in fsblocks */
314         int             whichfork,      /* data or attribute fork */
315         char            *aeof);         /* return value */
316
317 #ifdef XFS_BMAP_TRACE
318 /*
319  * Add a bmap trace buffer entry.  Base routine for the others.
320  */
321 STATIC void
322 xfs_bmap_trace_addentry(
323         int             opcode,         /* operation */
324         char            *fname,         /* function name */
325         char            *desc,          /* operation description */
326         xfs_inode_t     *ip,            /* incore inode pointer */
327         xfs_extnum_t    idx,            /* index of entry(ies) */
328         xfs_extnum_t    cnt,            /* count of entries, 1 or 2 */
329         xfs_bmbt_rec_t  *r1,            /* first record */
330         xfs_bmbt_rec_t  *r2,            /* second record or null */
331         int             whichfork);     /* data or attr fork */
332
333 /*
334  * Add bmap trace entry prior to a call to xfs_bmap_delete_exlist.
335  */
336 STATIC void
337 xfs_bmap_trace_delete(
338         char            *fname,         /* function name */
339         char            *desc,          /* operation description */
340         xfs_inode_t     *ip,            /* incore inode pointer */
341         xfs_extnum_t    idx,            /* index of entry(entries) deleted */
342         xfs_extnum_t    cnt,            /* count of entries deleted, 1 or 2 */
343         int             whichfork);     /* data or attr fork */
344
345 /*
346  * Add bmap trace entry prior to a call to xfs_bmap_insert_exlist, or
347  * reading in the extents list from the disk (in the btree).
348  */
349 STATIC void
350 xfs_bmap_trace_insert(
351         char            *fname,         /* function name */
352         char            *desc,          /* operation description */
353         xfs_inode_t     *ip,            /* incore inode pointer */
354         xfs_extnum_t    idx,            /* index of entry(entries) inserted */
355         xfs_extnum_t    cnt,            /* count of entries inserted, 1 or 2 */
356         xfs_bmbt_irec_t *r1,            /* inserted record 1 */
357         xfs_bmbt_irec_t *r2,            /* inserted record 2 or null */
358         int             whichfork);     /* data or attr fork */
359
360 /*
361  * Add bmap trace entry after updating an extent list entry in place.
362  */
363 STATIC void
364 xfs_bmap_trace_post_update(
365         char            *fname,         /* function name */
366         char            *desc,          /* operation description */
367         xfs_inode_t     *ip,            /* incore inode pointer */
368         xfs_extnum_t    idx,            /* index of entry updated */
369         int             whichfork);     /* data or attr fork */
370
371 /*
372  * Add bmap trace entry prior to updating an extent list entry in place.
373  */
374 STATIC void
375 xfs_bmap_trace_pre_update(
376         char            *fname,         /* function name */
377         char            *desc,          /* operation description */
378         xfs_inode_t     *ip,            /* incore inode pointer */
379         xfs_extnum_t    idx,            /* index of entry to be updated */
380         int             whichfork);     /* data or attr fork */
381
382 #else
383 #define xfs_bmap_trace_delete(f,d,ip,i,c,w)
384 #define xfs_bmap_trace_insert(f,d,ip,i,c,r1,r2,w)
385 #define xfs_bmap_trace_post_update(f,d,ip,i,w)
386 #define xfs_bmap_trace_pre_update(f,d,ip,i,w)
387 #endif  /* XFS_BMAP_TRACE */
388
389 /*
390  * Compute the worst-case number of indirect blocks that will be used
391  * for ip's delayed extent of length "len".
392  */
393 STATIC xfs_filblks_t
394 xfs_bmap_worst_indlen(
395         xfs_inode_t             *ip,    /* incore inode pointer */
396         xfs_filblks_t           len);   /* delayed extent length */
397
398 #ifdef DEBUG
399 /*
400  * Perform various validation checks on the values being returned
401  * from xfs_bmapi().
402  */
403 STATIC void
404 xfs_bmap_validate_ret(
405         xfs_fileoff_t           bno,
406         xfs_filblks_t           len,
407         int                     flags,
408         xfs_bmbt_irec_t         *mval,
409         int                     nmap,
410         int                     ret_nmap);
411 #else
412 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
413 #endif /* DEBUG */
414
415 #if defined(XFS_RW_TRACE)
416 STATIC void
417 xfs_bunmap_trace(
418         xfs_inode_t             *ip,
419         xfs_fileoff_t           bno,
420         xfs_filblks_t           len,
421         int                     flags,
422         inst_t                  *ra);
423 #else
424 #define xfs_bunmap_trace(ip, bno, len, flags, ra)
425 #endif  /* XFS_RW_TRACE */
426
427 STATIC int
428 xfs_bmap_count_tree(
429         xfs_mount_t     *mp,
430         xfs_trans_t     *tp,
431         xfs_fsblock_t   blockno,
432         int             levelin,
433         int             *count);
434
435 STATIC int
436 xfs_bmap_count_leaves(
437         xfs_bmbt_rec_t          *frp,
438         int                     numrecs,
439         int                     *count);
440
441 /*
442  * Bmap internal routines.
443  */
444
445 /*
446  * Called from xfs_bmap_add_attrfork to handle btree format files.
447  */
448 STATIC int                                      /* error */
449 xfs_bmap_add_attrfork_btree(
450         xfs_trans_t             *tp,            /* transaction pointer */
451         xfs_inode_t             *ip,            /* incore inode pointer */
452         xfs_fsblock_t           *firstblock,    /* first block allocated */
453         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
454         int                     *flags)         /* inode logging flags */
455 {
456         xfs_btree_cur_t         *cur;           /* btree cursor */
457         int                     error;          /* error return value */
458         xfs_mount_t             *mp;            /* file system mount struct */
459         int                     stat;           /* newroot status */
460
461         mp = ip->i_mount;
462         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
463                 *flags |= XFS_ILOG_DBROOT;
464         else {
465                 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
466                         XFS_DATA_FORK);
467                 cur->bc_private.b.flist = flist;
468                 cur->bc_private.b.firstblock = *firstblock;
469                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
470                         goto error0;
471                 ASSERT(stat == 1);      /* must be at least one entry */
472                 if ((error = xfs_bmbt_newroot(cur, flags, &stat)))
473                         goto error0;
474                 if (stat == 0) {
475                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
476                         return XFS_ERROR(ENOSPC);
477                 }
478                 *firstblock = cur->bc_private.b.firstblock;
479                 cur->bc_private.b.allocated = 0;
480                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
481         }
482         return 0;
483 error0:
484         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
485         return error;
486 }
487
488 /*
489  * Called from xfs_bmap_add_attrfork to handle extents format files.
490  */
491 STATIC int                                      /* error */
492 xfs_bmap_add_attrfork_extents(
493         xfs_trans_t             *tp,            /* transaction pointer */
494         xfs_inode_t             *ip,            /* incore inode pointer */
495         xfs_fsblock_t           *firstblock,    /* first block allocated */
496         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
497         int                     *flags)         /* inode logging flags */
498 {
499         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
500         int                     error;          /* error return value */
501
502         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
503                 return 0;
504         cur = NULL;
505         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
506                 flags, XFS_DATA_FORK);
507         if (cur) {
508                 cur->bc_private.b.allocated = 0;
509                 xfs_btree_del_cursor(cur,
510                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
511         }
512         return error;
513 }
514
515 /*
516  * Called from xfs_bmap_add_attrfork to handle local format files.
517  */
518 STATIC int                                      /* error */
519 xfs_bmap_add_attrfork_local(
520         xfs_trans_t             *tp,            /* transaction pointer */
521         xfs_inode_t             *ip,            /* incore inode pointer */
522         xfs_fsblock_t           *firstblock,    /* first block allocated */
523         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
524         int                     *flags)         /* inode logging flags */
525 {
526         xfs_da_args_t           dargs;          /* args for dir/attr code */
527         int                     error;          /* error return value */
528         xfs_mount_t             *mp;            /* mount structure pointer */
529
530         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
531                 return 0;
532         if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
533                 mp = ip->i_mount;
534                 memset(&dargs, 0, sizeof(dargs));
535                 dargs.dp = ip;
536                 dargs.firstblock = firstblock;
537                 dargs.flist = flist;
538                 dargs.total = mp->m_dirblkfsbs;
539                 dargs.whichfork = XFS_DATA_FORK;
540                 dargs.trans = tp;
541                 error = XFS_DIR_SHORTFORM_TO_SINGLE(mp, &dargs);
542         } else
543                 error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
544                         XFS_DATA_FORK);
545         return error;
546 }
547
548 /*
549  * Called by xfs_bmapi to update extent list structure and the btree
550  * after allocating space (or doing a delayed allocation).
551  */
552 STATIC int                              /* error */
553 xfs_bmap_add_extent(
554         xfs_inode_t             *ip,    /* incore inode pointer */
555         xfs_extnum_t            idx,    /* extent number to update/insert */
556         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
557         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
558         xfs_fsblock_t           *first, /* pointer to firstblock variable */
559         xfs_bmap_free_t         *flist, /* list of extents to be freed */
560         int                     *logflagsp, /* inode logging flags */
561         int                     whichfork, /* data or attr fork */
562         int                     rsvd)   /* OK to use reserved data blocks */
563 {
564         xfs_btree_cur_t         *cur;   /* btree cursor or null */
565         xfs_filblks_t           da_new; /* new count del alloc blocks used */
566         xfs_filblks_t           da_old; /* old count del alloc blocks used */
567         int                     error;  /* error return value */
568 #ifdef XFS_BMAP_TRACE
569         static char             fname[] = "xfs_bmap_add_extent";
570 #endif
571         xfs_ifork_t             *ifp;   /* inode fork ptr */
572         int                     logflags; /* returned value */
573         xfs_extnum_t            nextents; /* number of extents in file now */
574
575         XFS_STATS_INC(xs_add_exlist);
576         cur = *curp;
577         ifp = XFS_IFORK_PTR(ip, whichfork);
578         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
579         ASSERT(idx <= nextents);
580         da_old = da_new = 0;
581         error = 0;
582         /*
583          * This is the first extent added to a new/empty file.
584          * Special case this one, so other routines get to assume there are
585          * already extents in the list.
586          */
587         if (nextents == 0) {
588                 xfs_bmap_trace_insert(fname, "insert empty", ip, 0, 1, new,
589                         NULL, whichfork);
590                 xfs_bmap_insert_exlist(ip, 0, 1, new, whichfork);
591                 ASSERT(cur == NULL);
592                 ifp->if_lastex = 0;
593                 if (!ISNULLSTARTBLOCK(new->br_startblock)) {
594                         XFS_IFORK_NEXT_SET(ip, whichfork, 1);
595                         logflags = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
596                 } else
597                         logflags = 0;
598         }
599         /*
600          * Any kind of new delayed allocation goes here.
601          */
602         else if (ISNULLSTARTBLOCK(new->br_startblock)) {
603                 if (cur)
604                         ASSERT((cur->bc_private.b.flags &
605                                 XFS_BTCUR_BPRV_WASDEL) == 0);
606                 if ((error = xfs_bmap_add_extent_hole_delay(ip, idx, cur, new,
607                                 &logflags, rsvd)))
608                         goto done;
609         }
610         /*
611          * Real allocation off the end of the file.
612          */
613         else if (idx == nextents) {
614                 if (cur)
615                         ASSERT((cur->bc_private.b.flags &
616                                 XFS_BTCUR_BPRV_WASDEL) == 0);
617                 if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur, new,
618                                 &logflags, whichfork)))
619                         goto done;
620         } else {
621                 xfs_bmbt_irec_t prev;   /* old extent at offset idx */
622
623                 /*
624                  * Get the record referred to by idx.
625                  */
626                 xfs_bmbt_get_all(&ifp->if_u1.if_extents[idx], &prev);
627                 /*
628                  * If it's a real allocation record, and the new allocation ends
629                  * after the start of the referred to record, then we're filling
630                  * in a delayed or unwritten allocation with a real one, or
631                  * converting real back to unwritten.
632                  */
633                 if (!ISNULLSTARTBLOCK(new->br_startblock) &&
634                     new->br_startoff + new->br_blockcount > prev.br_startoff) {
635                         if (prev.br_state != XFS_EXT_UNWRITTEN &&
636                             ISNULLSTARTBLOCK(prev.br_startblock)) {
637                                 da_old = STARTBLOCKVAL(prev.br_startblock);
638                                 if (cur)
639                                         ASSERT(cur->bc_private.b.flags &
640                                                 XFS_BTCUR_BPRV_WASDEL);
641                                 if ((error = xfs_bmap_add_extent_delay_real(ip,
642                                         idx, &cur, new, &da_new, first, flist,
643                                         &logflags, rsvd)))
644                                         goto done;
645                         } else if (new->br_state == XFS_EXT_NORM) {
646                                 ASSERT(new->br_state == XFS_EXT_NORM);
647                                 if ((error = xfs_bmap_add_extent_unwritten_real(
648                                         ip, idx, &cur, new, &logflags)))
649                                         goto done;
650                         } else {
651                                 ASSERT(new->br_state == XFS_EXT_UNWRITTEN);
652                                 if ((error = xfs_bmap_add_extent_unwritten_real(
653                                         ip, idx, &cur, new, &logflags)))
654                                         goto done;
655                         }
656                         ASSERT(*curp == cur || *curp == NULL);
657                 }
658                 /*
659                  * Otherwise we're filling in a hole with an allocation.
660                  */
661                 else {
662                         if (cur)
663                                 ASSERT((cur->bc_private.b.flags &
664                                         XFS_BTCUR_BPRV_WASDEL) == 0);
665                         if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur,
666                                         new, &logflags, whichfork)))
667                                 goto done;
668                 }
669         }
670
671         ASSERT(*curp == cur || *curp == NULL);
672         /*
673          * Convert to a btree if necessary.
674          */
675         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
676             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
677                 int     tmp_logflags;   /* partial log flag return val */
678
679                 ASSERT(cur == NULL);
680                 error = xfs_bmap_extents_to_btree(ip->i_transp, ip, first,
681                         flist, &cur, da_old > 0, &tmp_logflags, whichfork);
682                 logflags |= tmp_logflags;
683                 if (error)
684                         goto done;
685         }
686         /*
687          * Adjust for changes in reserved delayed indirect blocks.
688          * Nothing to do for disk quotas here.
689          */
690         if (da_old || da_new) {
691                 xfs_filblks_t   nblks;
692
693                 nblks = da_new;
694                 if (cur)
695                         nblks += cur->bc_private.b.allocated;
696                 ASSERT(nblks <= da_old);
697                 if (nblks < da_old)
698                         xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
699                                 (int)(da_old - nblks), rsvd);
700         }
701         /*
702          * Clear out the allocated field, done with it now in any case.
703          */
704         if (cur) {
705                 cur->bc_private.b.allocated = 0;
706                 *curp = cur;
707         }
708 done:
709 #ifdef DEBUG
710         if (!error)
711                 xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
712 #endif
713         *logflagsp = logflags;
714         return error;
715 }
716
717 /*
718  * Called by xfs_bmap_add_extent to handle cases converting a delayed
719  * allocation to a real allocation.
720  */
721 STATIC int                              /* error */
722 xfs_bmap_add_extent_delay_real(
723         xfs_inode_t             *ip,    /* incore inode pointer */
724         xfs_extnum_t            idx,    /* extent number to update/insert */
725         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
726         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
727         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
728         xfs_fsblock_t           *first, /* pointer to firstblock variable */
729         xfs_bmap_free_t         *flist, /* list of extents to be freed */
730         int                     *logflagsp, /* inode logging flags */
731         int                     rsvd)   /* OK to use reserved data block allocation */
732 {
733         xfs_bmbt_rec_t          *base;  /* base of extent entry list */
734         xfs_btree_cur_t         *cur;   /* btree cursor */
735         int                     diff;   /* temp value */
736         xfs_bmbt_rec_t          *ep;    /* extent entry for idx */
737         int                     error;  /* error return value */
738 #ifdef XFS_BMAP_TRACE
739         static char             fname[] = "xfs_bmap_add_extent_delay_real";
740 #endif
741         int                     i;      /* temp state */
742         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
743         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
744                                         /* left is 0, right is 1, prev is 2 */
745         int                     rval=0; /* return value (logging flags) */
746         int                     state = 0;/* state bits, accessed thru macros */
747         xfs_filblks_t           temp;   /* value for dnew calculations */
748         xfs_filblks_t           temp2;  /* value for dnew calculations */
749         int                     tmp_rval;       /* partial logging flags */
750         enum {                          /* bit number definitions for state */
751                 LEFT_CONTIG,    RIGHT_CONTIG,
752                 LEFT_FILLING,   RIGHT_FILLING,
753                 LEFT_DELAY,     RIGHT_DELAY,
754                 LEFT_VALID,     RIGHT_VALID
755         };
756
757 #define LEFT            r[0]
758 #define RIGHT           r[1]
759 #define PREV            r[2]
760 #define MASK(b)         (1 << (b))
761 #define MASK2(a,b)      (MASK(a) | MASK(b))
762 #define MASK3(a,b,c)    (MASK2(a,b) | MASK(c))
763 #define MASK4(a,b,c,d)  (MASK3(a,b,c) | MASK(d))
764 #define STATE_SET(b,v)  ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
765 #define STATE_TEST(b)   (state & MASK(b))
766 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
767                                        ((state &= ~MASK(b)), 0))
768 #define SWITCH_STATE            \
769         (state & MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG))
770
771         /*
772          * Set up a bunch of variables to make the tests simpler.
773          */
774         cur = *curp;
775         base = ip->i_df.if_u1.if_extents;
776         ep = &base[idx];
777         xfs_bmbt_get_all(ep, &PREV);
778         new_endoff = new->br_startoff + new->br_blockcount;
779         ASSERT(PREV.br_startoff <= new->br_startoff);
780         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
781         /*
782          * Set flags determining what part of the previous delayed allocation
783          * extent is being replaced by a real allocation.
784          */
785         STATE_SET(LEFT_FILLING, PREV.br_startoff == new->br_startoff);
786         STATE_SET(RIGHT_FILLING,
787                 PREV.br_startoff + PREV.br_blockcount == new_endoff);
788         /*
789          * Check and set flags if this segment has a left neighbor.
790          * Don't set contiguous if the combined extent would be too large.
791          */
792         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
793                 xfs_bmbt_get_all(ep - 1, &LEFT);
794                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(LEFT.br_startblock));
795         }
796         STATE_SET(LEFT_CONTIG,
797                 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
798                 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
799                 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
800                 LEFT.br_state == new->br_state &&
801                 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN);
802         /*
803          * Check and set flags if this segment has a right neighbor.
804          * Don't set contiguous if the combined extent would be too large.
805          * Also check for all-three-contiguous being too large.
806          */
807         if (STATE_SET_TEST(RIGHT_VALID,
808                         idx <
809                         ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1)) {
810                 xfs_bmbt_get_all(ep + 1, &RIGHT);
811                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(RIGHT.br_startblock));
812         }
813         STATE_SET(RIGHT_CONTIG,
814                 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
815                 new_endoff == RIGHT.br_startoff &&
816                 new->br_startblock + new->br_blockcount ==
817                     RIGHT.br_startblock &&
818                 new->br_state == RIGHT.br_state &&
819                 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
820                 ((state & MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING)) !=
821                   MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING) ||
822                  LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
823                      <= MAXEXTLEN));
824         error = 0;
825         /*
826          * Switch out based on the FILLING and CONTIG state bits.
827          */
828         switch (SWITCH_STATE) {
829
830         case MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
831                 /*
832                  * Filling in all of a previously delayed allocation extent.
833                  * The left and right neighbors are both contiguous with new.
834                  */
835                 xfs_bmap_trace_pre_update(fname, "LF|RF|LC|RC", ip, idx - 1,
836                         XFS_DATA_FORK);
837                 xfs_bmbt_set_blockcount(ep - 1,
838                         LEFT.br_blockcount + PREV.br_blockcount +
839                         RIGHT.br_blockcount);
840                 xfs_bmap_trace_post_update(fname, "LF|RF|LC|RC", ip, idx - 1,
841                         XFS_DATA_FORK);
842                 xfs_bmap_trace_delete(fname, "LF|RF|LC|RC", ip, idx, 2,
843                         XFS_DATA_FORK);
844                 xfs_bmap_delete_exlist(ip, idx, 2, XFS_DATA_FORK);
845                 ip->i_df.if_lastex = idx - 1;
846                 ip->i_d.di_nextents--;
847                 if (cur == NULL)
848                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
849                 else {
850                         rval = XFS_ILOG_CORE;
851                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
852                                         RIGHT.br_startblock,
853                                         RIGHT.br_blockcount, &i)))
854                                 goto done;
855                         ASSERT(i == 1);
856                         if ((error = xfs_bmbt_delete(cur, &i)))
857                                 goto done;
858                         ASSERT(i == 1);
859                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
860                                 goto done;
861                         ASSERT(i == 1);
862                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
863                                         LEFT.br_startblock,
864                                         LEFT.br_blockcount +
865                                         PREV.br_blockcount +
866                                         RIGHT.br_blockcount, LEFT.br_state)))
867                                 goto done;
868                 }
869                 *dnew = 0;
870                 break;
871
872         case MASK3(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG):
873                 /*
874                  * Filling in all of a previously delayed allocation extent.
875                  * The left neighbor is contiguous, the right is not.
876                  */
877                 xfs_bmap_trace_pre_update(fname, "LF|RF|LC", ip, idx - 1,
878                         XFS_DATA_FORK);
879                 xfs_bmbt_set_blockcount(ep - 1,
880                         LEFT.br_blockcount + PREV.br_blockcount);
881                 xfs_bmap_trace_post_update(fname, "LF|RF|LC", ip, idx - 1,
882                         XFS_DATA_FORK);
883                 ip->i_df.if_lastex = idx - 1;
884                 xfs_bmap_trace_delete(fname, "LF|RF|LC", ip, idx, 1,
885                         XFS_DATA_FORK);
886                 xfs_bmap_delete_exlist(ip, idx, 1, XFS_DATA_FORK);
887                 if (cur == NULL)
888                         rval = XFS_ILOG_DEXT;
889                 else {
890                         rval = 0;
891                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
892                                         LEFT.br_startblock, LEFT.br_blockcount,
893                                         &i)))
894                                 goto done;
895                         ASSERT(i == 1);
896                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
897                                         LEFT.br_startblock,
898                                         LEFT.br_blockcount +
899                                         PREV.br_blockcount, LEFT.br_state)))
900                                 goto done;
901                 }
902                 *dnew = 0;
903                 break;
904
905         case MASK3(LEFT_FILLING, RIGHT_FILLING, RIGHT_CONTIG):
906                 /*
907                  * Filling in all of a previously delayed allocation extent.
908                  * The right neighbor is contiguous, the left is not.
909                  */
910                 xfs_bmap_trace_pre_update(fname, "LF|RF|RC", ip, idx,
911                         XFS_DATA_FORK);
912                 xfs_bmbt_set_startblock(ep, new->br_startblock);
913                 xfs_bmbt_set_blockcount(ep,
914                         PREV.br_blockcount + RIGHT.br_blockcount);
915                 xfs_bmap_trace_post_update(fname, "LF|RF|RC", ip, idx,
916                         XFS_DATA_FORK);
917                 ip->i_df.if_lastex = idx;
918                 xfs_bmap_trace_delete(fname, "LF|RF|RC", ip, idx + 1, 1,
919                         XFS_DATA_FORK);
920                 xfs_bmap_delete_exlist(ip, idx + 1, 1, XFS_DATA_FORK);
921                 if (cur == NULL)
922                         rval = XFS_ILOG_DEXT;
923                 else {
924                         rval = 0;
925                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
926                                         RIGHT.br_startblock,
927                                         RIGHT.br_blockcount, &i)))
928                                 goto done;
929                         ASSERT(i == 1);
930                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
931                                         new->br_startblock,
932                                         PREV.br_blockcount +
933                                         RIGHT.br_blockcount, PREV.br_state)))
934                                 goto done;
935                 }
936                 *dnew = 0;
937                 break;
938
939         case MASK2(LEFT_FILLING, RIGHT_FILLING):
940                 /*
941                  * Filling in all of a previously delayed allocation extent.
942                  * Neither the left nor right neighbors are contiguous with
943                  * the new one.
944                  */
945                 xfs_bmap_trace_pre_update(fname, "LF|RF", ip, idx,
946                         XFS_DATA_FORK);
947                 xfs_bmbt_set_startblock(ep, new->br_startblock);
948                 xfs_bmap_trace_post_update(fname, "LF|RF", ip, idx,
949                         XFS_DATA_FORK);
950                 ip->i_df.if_lastex = idx;
951                 ip->i_d.di_nextents++;
952                 if (cur == NULL)
953                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
954                 else {
955                         rval = XFS_ILOG_CORE;
956                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
957                                         new->br_startblock, new->br_blockcount,
958                                         &i)))
959                                 goto done;
960                         ASSERT(i == 0);
961                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
962                         if ((error = xfs_bmbt_insert(cur, &i)))
963                                 goto done;
964                         ASSERT(i == 1);
965                 }
966                 *dnew = 0;
967                 break;
968
969         case MASK2(LEFT_FILLING, LEFT_CONTIG):
970                 /*
971                  * Filling in the first part of a previous delayed allocation.
972                  * The left neighbor is contiguous.
973                  */
974                 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx - 1,
975                         XFS_DATA_FORK);
976                 xfs_bmbt_set_blockcount(ep - 1,
977                         LEFT.br_blockcount + new->br_blockcount);
978                 xfs_bmbt_set_startoff(ep,
979                         PREV.br_startoff + new->br_blockcount);
980                 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx - 1,
981                         XFS_DATA_FORK);
982                 temp = PREV.br_blockcount - new->br_blockcount;
983                 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx,
984                         XFS_DATA_FORK);
985                 xfs_bmbt_set_blockcount(ep, temp);
986                 ip->i_df.if_lastex = idx - 1;
987                 if (cur == NULL)
988                         rval = XFS_ILOG_DEXT;
989                 else {
990                         rval = 0;
991                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
992                                         LEFT.br_startblock, LEFT.br_blockcount,
993                                         &i)))
994                                 goto done;
995                         ASSERT(i == 1);
996                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
997                                         LEFT.br_startblock,
998                                         LEFT.br_blockcount +
999                                         new->br_blockcount,
1000                                         LEFT.br_state)))
1001                                 goto done;
1002                 }
1003                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1004                         STARTBLOCKVAL(PREV.br_startblock));
1005                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1006                 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx,
1007                         XFS_DATA_FORK);
1008                 *dnew = temp;
1009                 break;
1010
1011         case MASK(LEFT_FILLING):
1012                 /*
1013                  * Filling in the first part of a previous delayed allocation.
1014                  * The left neighbor is not contiguous.
1015                  */
1016                 xfs_bmap_trace_pre_update(fname, "LF", ip, idx, XFS_DATA_FORK);
1017                 xfs_bmbt_set_startoff(ep, new_endoff);
1018                 temp = PREV.br_blockcount - new->br_blockcount;
1019                 xfs_bmbt_set_blockcount(ep, temp);
1020                 xfs_bmap_trace_insert(fname, "LF", ip, idx, 1, new, NULL,
1021                         XFS_DATA_FORK);
1022                 xfs_bmap_insert_exlist(ip, idx, 1, new, XFS_DATA_FORK);
1023                 ip->i_df.if_lastex = idx;
1024                 ip->i_d.di_nextents++;
1025                 if (cur == NULL)
1026                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1027                 else {
1028                         rval = XFS_ILOG_CORE;
1029                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1030                                         new->br_startblock, new->br_blockcount,
1031                                         &i)))
1032                                 goto done;
1033                         ASSERT(i == 0);
1034                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1035                         if ((error = xfs_bmbt_insert(cur, &i)))
1036                                 goto done;
1037                         ASSERT(i == 1);
1038                 }
1039                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1040                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1041                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1042                                         first, flist, &cur, 1, &tmp_rval,
1043                                         XFS_DATA_FORK);
1044                         rval |= tmp_rval;
1045                         if (error)
1046                                 goto done;
1047                 }
1048                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1049                         STARTBLOCKVAL(PREV.br_startblock) -
1050                         (cur ? cur->bc_private.b.allocated : 0));
1051                 base = ip->i_df.if_u1.if_extents;
1052                 ep = &base[idx + 1];
1053                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1054                 xfs_bmap_trace_post_update(fname, "LF", ip, idx + 1,
1055                         XFS_DATA_FORK);
1056                 *dnew = temp;
1057                 break;
1058
1059         case MASK2(RIGHT_FILLING, RIGHT_CONTIG):
1060                 /*
1061                  * Filling in the last part of a previous delayed allocation.
1062                  * The right neighbor is contiguous with the new allocation.
1063                  */
1064                 temp = PREV.br_blockcount - new->br_blockcount;
1065                 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx,
1066                         XFS_DATA_FORK);
1067                 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx + 1,
1068                         XFS_DATA_FORK);
1069                 xfs_bmbt_set_blockcount(ep, temp);
1070                 xfs_bmbt_set_allf(ep + 1, new->br_startoff, new->br_startblock,
1071                         new->br_blockcount + RIGHT.br_blockcount,
1072                         RIGHT.br_state);
1073                 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx + 1,
1074                         XFS_DATA_FORK);
1075                 ip->i_df.if_lastex = idx + 1;
1076                 if (cur == NULL)
1077                         rval = XFS_ILOG_DEXT;
1078                 else {
1079                         rval = 0;
1080                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1081                                         RIGHT.br_startblock,
1082                                         RIGHT.br_blockcount, &i)))
1083                                 goto done;
1084                         ASSERT(i == 1);
1085                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1086                                         new->br_startblock,
1087                                         new->br_blockcount +
1088                                         RIGHT.br_blockcount,
1089                                         RIGHT.br_state)))
1090                                 goto done;
1091                 }
1092                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1093                         STARTBLOCKVAL(PREV.br_startblock));
1094                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1095                 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx,
1096                         XFS_DATA_FORK);
1097                 *dnew = temp;
1098                 break;
1099
1100         case MASK(RIGHT_FILLING):
1101                 /*
1102                  * Filling in the last part of a previous delayed allocation.
1103                  * The right neighbor is not contiguous.
1104                  */
1105                 temp = PREV.br_blockcount - new->br_blockcount;
1106                 xfs_bmap_trace_pre_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1107                 xfs_bmbt_set_blockcount(ep, temp);
1108                 xfs_bmap_trace_insert(fname, "RF", ip, idx + 1, 1,
1109                         new, NULL, XFS_DATA_FORK);
1110                 xfs_bmap_insert_exlist(ip, idx + 1, 1, new, XFS_DATA_FORK);
1111                 ip->i_df.if_lastex = idx + 1;
1112                 ip->i_d.di_nextents++;
1113                 if (cur == NULL)
1114                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1115                 else {
1116                         rval = XFS_ILOG_CORE;
1117                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1118                                         new->br_startblock, new->br_blockcount,
1119                                         &i)))
1120                                 goto done;
1121                         ASSERT(i == 0);
1122                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1123                         if ((error = xfs_bmbt_insert(cur, &i)))
1124                                 goto done;
1125                         ASSERT(i == 1);
1126                 }
1127                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1128                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1129                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1130                                 first, flist, &cur, 1, &tmp_rval,
1131                                 XFS_DATA_FORK);
1132                         rval |= tmp_rval;
1133                         if (error)
1134                                 goto done;
1135                 }
1136                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1137                         STARTBLOCKVAL(PREV.br_startblock) -
1138                         (cur ? cur->bc_private.b.allocated : 0));
1139                 base = ip->i_df.if_u1.if_extents;
1140                 ep = &base[idx];
1141                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1142                 xfs_bmap_trace_post_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1143                 *dnew = temp;
1144                 break;
1145
1146         case 0:
1147                 /*
1148                  * Filling in the middle part of a previous delayed allocation.
1149                  * Contiguity is impossible here.
1150                  * This case is avoided almost all the time.
1151                  */
1152                 temp = new->br_startoff - PREV.br_startoff;
1153                 xfs_bmap_trace_pre_update(fname, "0", ip, idx, XFS_DATA_FORK);
1154                 xfs_bmbt_set_blockcount(ep, temp);
1155                 r[0] = *new;
1156                 r[1].br_startoff = new_endoff;
1157                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
1158                 r[1].br_blockcount = temp2;
1159                 xfs_bmap_trace_insert(fname, "0", ip, idx + 1, 2, &r[0], &r[1],
1160                         XFS_DATA_FORK);
1161                 xfs_bmap_insert_exlist(ip, idx + 1, 2, &r[0], XFS_DATA_FORK);
1162                 ip->i_df.if_lastex = idx + 1;
1163                 ip->i_d.di_nextents++;
1164                 if (cur == NULL)
1165                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1166                 else {
1167                         rval = XFS_ILOG_CORE;
1168                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1169                                         new->br_startblock, new->br_blockcount,
1170                                         &i)))
1171                                 goto done;
1172                         ASSERT(i == 0);
1173                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1174                         if ((error = xfs_bmbt_insert(cur, &i)))
1175                                 goto done;
1176                         ASSERT(i == 1);
1177                 }
1178                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1179                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1180                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1181                                         first, flist, &cur, 1, &tmp_rval,
1182                                         XFS_DATA_FORK);
1183                         rval |= tmp_rval;
1184                         if (error)
1185                                 goto done;
1186                 }
1187                 temp = xfs_bmap_worst_indlen(ip, temp);
1188                 temp2 = xfs_bmap_worst_indlen(ip, temp2);
1189                 diff = (int)(temp + temp2 - STARTBLOCKVAL(PREV.br_startblock) -
1190                         (cur ? cur->bc_private.b.allocated : 0));
1191                 if (diff > 0 &&
1192                     xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS, -diff, rsvd)) {
1193                         /*
1194                          * Ick gross gag me with a spoon.
1195                          */
1196                         ASSERT(0);      /* want to see if this ever happens! */
1197                         while (diff > 0) {
1198                                 if (temp) {
1199                                         temp--;
1200                                         diff--;
1201                                         if (!diff ||
1202                                             !xfs_mod_incore_sb(ip->i_mount,
1203                                                     XFS_SBS_FDBLOCKS, -diff, rsvd))
1204                                                 break;
1205                                 }
1206                                 if (temp2) {
1207                                         temp2--;
1208                                         diff--;
1209                                         if (!diff ||
1210                                             !xfs_mod_incore_sb(ip->i_mount,
1211                                                     XFS_SBS_FDBLOCKS, -diff, rsvd))
1212                                                 break;
1213                                 }
1214                         }
1215                 }
1216                 base = ip->i_df.if_u1.if_extents;
1217                 ep = &base[idx];
1218                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1219                 xfs_bmap_trace_post_update(fname, "0", ip, idx, XFS_DATA_FORK);
1220                 xfs_bmap_trace_pre_update(fname, "0", ip, idx + 2,
1221                         XFS_DATA_FORK);
1222                 xfs_bmbt_set_startblock(ep + 2, NULLSTARTBLOCK((int)temp2));
1223                 xfs_bmap_trace_post_update(fname, "0", ip, idx + 2,
1224                         XFS_DATA_FORK);
1225                 *dnew = temp + temp2;
1226                 break;
1227
1228         case MASK3(LEFT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1229         case MASK3(RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1230         case MASK2(LEFT_FILLING, RIGHT_CONTIG):
1231         case MASK2(RIGHT_FILLING, LEFT_CONTIG):
1232         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1233         case MASK(LEFT_CONTIG):
1234         case MASK(RIGHT_CONTIG):
1235                 /*
1236                  * These cases are all impossible.
1237                  */
1238                 ASSERT(0);
1239         }
1240         *curp = cur;
1241 done:
1242         *logflagsp = rval;
1243         return error;
1244 #undef  LEFT
1245 #undef  RIGHT
1246 #undef  PREV
1247 #undef  MASK
1248 #undef  MASK2
1249 #undef  MASK3
1250 #undef  MASK4
1251 #undef  STATE_SET
1252 #undef  STATE_TEST
1253 #undef  STATE_SET_TEST
1254 #undef  SWITCH_STATE
1255 }
1256
1257 /*
1258  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
1259  * allocation to a real allocation or vice versa.
1260  */
1261 STATIC int                              /* error */
1262 xfs_bmap_add_extent_unwritten_real(
1263         xfs_inode_t             *ip,    /* incore inode pointer */
1264         xfs_extnum_t            idx,    /* extent number to update/insert */
1265         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
1266         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
1267         int                     *logflagsp) /* inode logging flags */
1268 {
1269         xfs_bmbt_rec_t          *base;  /* base of extent entry list */
1270         xfs_btree_cur_t         *cur;   /* btree cursor */
1271         xfs_bmbt_rec_t          *ep;    /* extent entry for idx */
1272         int                     error;  /* error return value */
1273 #ifdef XFS_BMAP_TRACE
1274         static char             fname[] = "xfs_bmap_add_extent_unwritten_real";
1275 #endif
1276         int                     i;      /* temp state */
1277         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1278         xfs_exntst_t            newext; /* new extent state */
1279         xfs_exntst_t            oldext; /* old extent state */
1280         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1281                                         /* left is 0, right is 1, prev is 2 */
1282         int                     rval=0; /* return value (logging flags) */
1283         int                     state = 0;/* state bits, accessed thru macros */
1284         enum {                          /* bit number definitions for state */
1285                 LEFT_CONTIG,    RIGHT_CONTIG,
1286                 LEFT_FILLING,   RIGHT_FILLING,
1287                 LEFT_DELAY,     RIGHT_DELAY,
1288                 LEFT_VALID,     RIGHT_VALID
1289         };
1290
1291 #define LEFT            r[0]
1292 #define RIGHT           r[1]
1293 #define PREV            r[2]
1294 #define MASK(b)         (1 << (b))
1295 #define MASK2(a,b)      (MASK(a) | MASK(b))
1296 #define MASK3(a,b,c)    (MASK2(a,b) | MASK(c))
1297 #define MASK4(a,b,c,d)  (MASK3(a,b,c) | MASK(d))
1298 #define STATE_SET(b,v)  ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1299 #define STATE_TEST(b)   (state & MASK(b))
1300 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
1301                                        ((state &= ~MASK(b)), 0))
1302 #define SWITCH_STATE            \
1303         (state & MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG))
1304
1305         /*
1306          * Set up a bunch of variables to make the tests simpler.
1307          */
1308         error = 0;
1309         cur = *curp;
1310         base = ip->i_df.if_u1.if_extents;
1311         ep = &base[idx];
1312         xfs_bmbt_get_all(ep, &PREV);
1313         newext = new->br_state;
1314         oldext = (newext == XFS_EXT_UNWRITTEN) ?
1315                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
1316         ASSERT(PREV.br_state == oldext);
1317         new_endoff = new->br_startoff + new->br_blockcount;
1318         ASSERT(PREV.br_startoff <= new->br_startoff);
1319         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1320         /*
1321          * Set flags determining what part of the previous oldext allocation
1322          * extent is being replaced by a newext allocation.
1323          */
1324         STATE_SET(LEFT_FILLING, PREV.br_startoff == new->br_startoff);
1325         STATE_SET(RIGHT_FILLING,
1326                 PREV.br_startoff + PREV.br_blockcount == new_endoff);
1327         /*
1328          * Check and set flags if this segment has a left neighbor.
1329          * Don't set contiguous if the combined extent would be too large.
1330          */
1331         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1332                 xfs_bmbt_get_all(ep - 1, &LEFT);
1333                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(LEFT.br_startblock));
1334         }
1335         STATE_SET(LEFT_CONTIG,
1336                 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
1337                 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1338                 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1339                 LEFT.br_state == newext &&
1340                 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1341         /*
1342          * Check and set flags if this segment has a right neighbor.
1343          * Don't set contiguous if the combined extent would be too large.
1344          * Also check for all-three-contiguous being too large.
1345          */
1346         if (STATE_SET_TEST(RIGHT_VALID,
1347                         idx <
1348                         ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1)) {
1349                 xfs_bmbt_get_all(ep + 1, &RIGHT);
1350                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(RIGHT.br_startblock));
1351         }
1352         STATE_SET(RIGHT_CONTIG,
1353                 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
1354                 new_endoff == RIGHT.br_startoff &&
1355                 new->br_startblock + new->br_blockcount ==
1356                     RIGHT.br_startblock &&
1357                 newext == RIGHT.br_state &&
1358                 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1359                 ((state & MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING)) !=
1360                   MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING) ||
1361                  LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1362                      <= MAXEXTLEN));
1363         /*
1364          * Switch out based on the FILLING and CONTIG state bits.
1365          */
1366         switch (SWITCH_STATE) {
1367
1368         case MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1369                 /*
1370                  * Setting all of a previous oldext extent to newext.
1371                  * The left and right neighbors are both contiguous with new.
1372                  */
1373                 xfs_bmap_trace_pre_update(fname, "LF|RF|LC|RC", ip, idx - 1,
1374                         XFS_DATA_FORK);
1375                 xfs_bmbt_set_blockcount(ep - 1,
1376                         LEFT.br_blockcount + PREV.br_blockcount +
1377                         RIGHT.br_blockcount);
1378                 xfs_bmap_trace_post_update(fname, "LF|RF|LC|RC", ip, idx - 1,
1379                         XFS_DATA_FORK);
1380                 xfs_bmap_trace_delete(fname, "LF|RF|LC|RC", ip, idx, 2,
1381                         XFS_DATA_FORK);
1382                 xfs_bmap_delete_exlist(ip, idx, 2, XFS_DATA_FORK);
1383                 ip->i_df.if_lastex = idx - 1;
1384                 ip->i_d.di_nextents -= 2;
1385                 if (cur == NULL)
1386                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1387                 else {
1388                         rval = XFS_ILOG_CORE;
1389                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1390                                         RIGHT.br_startblock,
1391                                         RIGHT.br_blockcount, &i)))
1392                                 goto done;
1393                         ASSERT(i == 1);
1394                         if ((error = xfs_bmbt_delete(cur, &i)))
1395                                 goto done;
1396                         ASSERT(i == 1);
1397                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1398                                 goto done;
1399                         ASSERT(i == 1);
1400                         if ((error = xfs_bmbt_delete(cur, &i)))
1401                                 goto done;
1402                         ASSERT(i == 1);
1403                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1404                                 goto done;
1405                         ASSERT(i == 1);
1406                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1407                                 LEFT.br_startblock,
1408                                 LEFT.br_blockcount + PREV.br_blockcount +
1409                                 RIGHT.br_blockcount, LEFT.br_state)))
1410                                 goto done;
1411                 }
1412                 break;
1413
1414         case MASK3(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG):
1415                 /*
1416                  * Setting all of a previous oldext extent to newext.
1417                  * The left neighbor is contiguous, the right is not.
1418                  */
1419                 xfs_bmap_trace_pre_update(fname, "LF|RF|LC", ip, idx - 1,
1420                         XFS_DATA_FORK);
1421                 xfs_bmbt_set_blockcount(ep - 1,
1422                         LEFT.br_blockcount + PREV.br_blockcount);
1423                 xfs_bmap_trace_post_update(fname, "LF|RF|LC", ip, idx - 1,
1424                         XFS_DATA_FORK);
1425                 ip->i_df.if_lastex = idx - 1;
1426                 xfs_bmap_trace_delete(fname, "LF|RF|LC", ip, idx, 1,
1427                         XFS_DATA_FORK);
1428                 xfs_bmap_delete_exlist(ip, idx, 1, XFS_DATA_FORK);
1429                 ip->i_d.di_nextents--;
1430                 if (cur == NULL)
1431                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1432                 else {
1433                         rval = XFS_ILOG_CORE;
1434                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1435                                         PREV.br_startblock, PREV.br_blockcount,
1436                                         &i)))
1437                                 goto done;
1438                         ASSERT(i == 1);
1439                         if ((error = xfs_bmbt_delete(cur, &i)))
1440                                 goto done;
1441                         ASSERT(i == 1);
1442                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1443                                 goto done;
1444                         ASSERT(i == 1);
1445                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1446                                 LEFT.br_startblock,
1447                                 LEFT.br_blockcount + PREV.br_blockcount,
1448                                 LEFT.br_state)))
1449                                 goto done;
1450                 }
1451                 break;
1452
1453         case MASK3(LEFT_FILLING, RIGHT_FILLING, RIGHT_CONTIG):
1454                 /*
1455                  * Setting all of a previous oldext extent to newext.
1456                  * The right neighbor is contiguous, the left is not.
1457                  */
1458                 xfs_bmap_trace_pre_update(fname, "LF|RF|RC", ip, idx,
1459                         XFS_DATA_FORK);
1460                 xfs_bmbt_set_blockcount(ep,
1461                         PREV.br_blockcount + RIGHT.br_blockcount);
1462                 xfs_bmbt_set_state(ep, newext);
1463                 xfs_bmap_trace_post_update(fname, "LF|RF|RC", ip, idx,
1464                         XFS_DATA_FORK);
1465                 ip->i_df.if_lastex = idx;
1466                 xfs_bmap_trace_delete(fname, "LF|RF|RC", ip, idx + 1, 1,
1467                         XFS_DATA_FORK);
1468                 xfs_bmap_delete_exlist(ip, idx + 1, 1, XFS_DATA_FORK);
1469                 ip->i_d.di_nextents--;
1470                 if (cur == NULL)
1471                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1472                 else {
1473                         rval = XFS_ILOG_CORE;
1474                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1475                                         RIGHT.br_startblock,
1476                                         RIGHT.br_blockcount, &i)))
1477                                 goto done;
1478                         ASSERT(i == 1);
1479                         if ((error = xfs_bmbt_delete(cur, &i)))
1480                                 goto done;
1481                         ASSERT(i == 1);
1482                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1483                                 goto done;
1484                         ASSERT(i == 1);
1485                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1486                                 new->br_startblock,
1487                                 new->br_blockcount + RIGHT.br_blockcount,
1488                                 newext)))
1489                                 goto done;
1490                 }
1491                 break;
1492
1493         case MASK2(LEFT_FILLING, RIGHT_FILLING):
1494                 /*
1495                  * Setting all of a previous oldext extent to newext.
1496                  * Neither the left nor right neighbors are contiguous with
1497                  * the new one.
1498                  */
1499                 xfs_bmap_trace_pre_update(fname, "LF|RF", ip, idx,
1500                         XFS_DATA_FORK);
1501                 xfs_bmbt_set_state(ep, newext);
1502                 xfs_bmap_trace_post_update(fname, "LF|RF", ip, idx,
1503                         XFS_DATA_FORK);
1504                 ip->i_df.if_lastex = idx;
1505                 if (cur == NULL)
1506                         rval = XFS_ILOG_DEXT;
1507                 else {
1508                         rval = 0;
1509                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1510                                         new->br_startblock, new->br_blockcount,
1511                                         &i)))
1512                                 goto done;
1513                         ASSERT(i == 1);
1514                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1515                                 new->br_startblock, new->br_blockcount,
1516                                 newext)))
1517                                 goto done;
1518                 }
1519                 break;
1520
1521         case MASK2(LEFT_FILLING, LEFT_CONTIG):
1522                 /*
1523                  * Setting the first part of a previous oldext extent to newext.
1524                  * The left neighbor is contiguous.
1525                  */
1526                 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx - 1,
1527                         XFS_DATA_FORK);
1528                 xfs_bmbt_set_blockcount(ep - 1,
1529                         LEFT.br_blockcount + new->br_blockcount);
1530                 xfs_bmbt_set_startoff(ep,
1531                         PREV.br_startoff + new->br_blockcount);
1532                 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx - 1,
1533                         XFS_DATA_FORK);
1534                 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx,
1535                         XFS_DATA_FORK);
1536                 xfs_bmbt_set_startblock(ep,
1537                         new->br_startblock + new->br_blockcount);
1538                 xfs_bmbt_set_blockcount(ep,
1539                         PREV.br_blockcount - new->br_blockcount);
1540                 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx,
1541                         XFS_DATA_FORK);
1542                 ip->i_df.if_lastex = idx - 1;
1543                 if (cur == NULL)
1544                         rval = XFS_ILOG_DEXT;
1545                 else {
1546                         rval = 0;
1547                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1548                                         PREV.br_startblock, PREV.br_blockcount,
1549                                         &i)))
1550                                 goto done;
1551                         ASSERT(i == 1);
1552                         if ((error = xfs_bmbt_update(cur,
1553                                 PREV.br_startoff + new->br_blockcount,
1554                                 PREV.br_startblock + new->br_blockcount,
1555                                 PREV.br_blockcount - new->br_blockcount,
1556                                 oldext)))
1557                                 goto done;
1558                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1559                                 goto done;
1560                         if (xfs_bmbt_update(cur, LEFT.br_startoff,
1561                                 LEFT.br_startblock,
1562                                 LEFT.br_blockcount + new->br_blockcount,
1563                                 LEFT.br_state))
1564                                 goto done;
1565                 }
1566                 break;
1567
1568         case MASK(LEFT_FILLING):
1569                 /*
1570                  * Setting the first part of a previous oldext extent to newext.
1571                  * The left neighbor is not contiguous.
1572                  */
1573                 xfs_bmap_trace_pre_update(fname, "LF", ip, idx, XFS_DATA_FORK);
1574                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1575                 xfs_bmbt_set_startoff(ep, new_endoff);
1576                 xfs_bmbt_set_blockcount(ep,
1577                         PREV.br_blockcount - new->br_blockcount);
1578                 xfs_bmbt_set_startblock(ep,
1579                         new->br_startblock + new->br_blockcount);
1580                 xfs_bmap_trace_post_update(fname, "LF", ip, idx, XFS_DATA_FORK);
1581                 xfs_bmap_trace_insert(fname, "LF", ip, idx, 1, new, NULL,
1582                         XFS_DATA_FORK);
1583                 xfs_bmap_insert_exlist(ip, idx, 1, new, XFS_DATA_FORK);
1584                 ip->i_df.if_lastex = idx;
1585                 ip->i_d.di_nextents++;
1586                 if (cur == NULL)
1587                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1588                 else {
1589                         rval = XFS_ILOG_CORE;
1590                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1591                                         PREV.br_startblock, PREV.br_blockcount,
1592                                         &i)))
1593                                 goto done;
1594                         ASSERT(i == 1);
1595                         if ((error = xfs_bmbt_update(cur,
1596                                 PREV.br_startoff + new->br_blockcount,
1597                                 PREV.br_startblock + new->br_blockcount,
1598                                 PREV.br_blockcount - new->br_blockcount,
1599                                 oldext)))
1600                                 goto done;
1601                         cur->bc_rec.b = *new;
1602                         if ((error = xfs_bmbt_insert(cur, &i)))
1603                                 goto done;
1604                         ASSERT(i == 1);
1605                 }
1606                 break;
1607
1608         case MASK2(RIGHT_FILLING, RIGHT_CONTIG):
1609                 /*
1610                  * Setting the last part of a previous oldext extent to newext.
1611                  * The right neighbor is contiguous with the new allocation.
1612                  */
1613                 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx,
1614                         XFS_DATA_FORK);
1615                 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx + 1,
1616                         XFS_DATA_FORK);
1617                 xfs_bmbt_set_blockcount(ep,
1618                         PREV.br_blockcount - new->br_blockcount);
1619                 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx,
1620                         XFS_DATA_FORK);
1621                 xfs_bmbt_set_allf(ep + 1, new->br_startoff, new->br_startblock,
1622                         new->br_blockcount + RIGHT.br_blockcount, newext);
1623                 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx + 1,
1624                         XFS_DATA_FORK);
1625                 ip->i_df.if_lastex = idx + 1;
1626                 if (cur == NULL)
1627                         rval = XFS_ILOG_DEXT;
1628                 else {
1629                         rval = 0;
1630                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1631                                         PREV.br_startblock,
1632                                         PREV.br_blockcount, &i)))
1633                                 goto done;
1634                         ASSERT(i == 1);
1635                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1636                                 PREV.br_startblock,
1637                                 PREV.br_blockcount - new->br_blockcount,
1638                                 oldext)))
1639                                 goto done;
1640                         if ((error = xfs_bmbt_increment(cur, 0, &i)))
1641                                 goto done;
1642                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1643                                 new->br_startblock,
1644                                 new->br_blockcount + RIGHT.br_blockcount,
1645                                 newext)))
1646                                 goto done;
1647                 }
1648                 break;
1649
1650         case MASK(RIGHT_FILLING):
1651                 /*
1652                  * Setting the last part of a previous oldext extent to newext.
1653                  * The right neighbor is not contiguous.
1654                  */
1655                 xfs_bmap_trace_pre_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1656                 xfs_bmbt_set_blockcount(ep,
1657                         PREV.br_blockcount - new->br_blockcount);
1658                 xfs_bmap_trace_post_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1659                 xfs_bmap_trace_insert(fname, "RF", ip, idx + 1, 1,
1660                         new, NULL, XFS_DATA_FORK);
1661                 xfs_bmap_insert_exlist(ip, idx + 1, 1, new, XFS_DATA_FORK);
1662                 ip->i_df.if_lastex = idx + 1;
1663                 ip->i_d.di_nextents++;
1664                 if (cur == NULL)
1665                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1666                 else {
1667                         rval = XFS_ILOG_CORE;
1668                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1669                                         PREV.br_startblock, PREV.br_blockcount,
1670                                         &i)))
1671                                 goto done;
1672                         ASSERT(i == 1);
1673                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1674                                 PREV.br_startblock,
1675                                 PREV.br_blockcount - new->br_blockcount,
1676                                 oldext)))
1677                                 goto done;
1678                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1679                                         new->br_startblock, new->br_blockcount,
1680                                         &i)))
1681                                 goto done;
1682                         ASSERT(i == 0);
1683                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1684                         if ((error = xfs_bmbt_insert(cur, &i)))
1685                                 goto done;
1686                         ASSERT(i == 1);
1687                 }
1688                 break;
1689
1690         case 0:
1691                 /*
1692                  * Setting the middle part of a previous oldext extent to
1693                  * newext.  Contiguity is impossible here.
1694                  * One extent becomes three extents.
1695                  */
1696                 xfs_bmap_trace_pre_update(fname, "0", ip, idx, XFS_DATA_FORK);
1697                 xfs_bmbt_set_blockcount(ep,
1698                         new->br_startoff - PREV.br_startoff);
1699                 xfs_bmap_trace_post_update(fname, "0", ip, idx, XFS_DATA_FORK);
1700                 r[0] = *new;
1701                 r[1].br_startoff = new_endoff;
1702                 r[1].br_blockcount =
1703                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
1704                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1705                 r[1].br_state = oldext;
1706                 xfs_bmap_trace_insert(fname, "0", ip, idx + 1, 2, &r[0], &r[1],
1707                         XFS_DATA_FORK);
1708                 xfs_bmap_insert_exlist(ip, idx + 1, 2, &r[0], XFS_DATA_FORK);
1709                 ip->i_df.if_lastex = idx + 1;
1710                 ip->i_d.di_nextents += 2;
1711                 if (cur == NULL)
1712                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1713                 else {
1714                         rval = XFS_ILOG_CORE;
1715                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1716                                         PREV.br_startblock, PREV.br_blockcount,
1717                                         &i)))
1718                                 goto done;
1719                         ASSERT(i == 1);
1720                         /* new right extent - oldext */
1721                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1722                                 r[1].br_startblock, r[1].br_blockcount,
1723                                 r[1].br_state)))
1724                                 goto done;
1725                         /* new left extent - oldext */
1726                         PREV.br_blockcount =
1727                                 new->br_startoff - PREV.br_startoff;
1728                         cur->bc_rec.b = PREV;
1729                         if ((error = xfs_bmbt_insert(cur, &i)))
1730                                 goto done;
1731                         ASSERT(i == 1);
1732                         if ((error = xfs_bmbt_increment(cur, 0, &i)))
1733                                 goto done;
1734                         ASSERT(i == 1);
1735                         /* new middle extent - newext */
1736                         cur->bc_rec.b = *new;
1737                         if ((error = xfs_bmbt_insert(cur, &i)))
1738                                 goto done;
1739                         ASSERT(i == 1);
1740                 }
1741                 break;
1742
1743         case MASK3(LEFT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1744         case MASK3(RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1745         case MASK2(LEFT_FILLING, RIGHT_CONTIG):
1746         case MASK2(RIGHT_FILLING, LEFT_CONTIG):
1747         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1748         case MASK(LEFT_CONTIG):
1749         case MASK(RIGHT_CONTIG):
1750                 /*
1751                  * These cases are all impossible.
1752                  */
1753                 ASSERT(0);
1754         }
1755         *curp = cur;
1756 done:
1757         *logflagsp = rval;
1758         return error;
1759 #undef  LEFT
1760 #undef  RIGHT
1761 #undef  PREV
1762 #undef  MASK
1763 #undef  MASK2
1764 #undef  MASK3
1765 #undef  MASK4
1766 #undef  STATE_SET
1767 #undef  STATE_TEST
1768 #undef  STATE_SET_TEST
1769 #undef  SWITCH_STATE
1770 }
1771
1772 /*
1773  * Called by xfs_bmap_add_extent to handle cases converting a hole
1774  * to a delayed allocation.
1775  */
1776 /*ARGSUSED*/
1777 STATIC int                              /* error */
1778 xfs_bmap_add_extent_hole_delay(
1779         xfs_inode_t             *ip,    /* incore inode pointer */
1780         xfs_extnum_t            idx,    /* extent number to update/insert */
1781         xfs_btree_cur_t         *cur,   /* if null, not a btree */
1782         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
1783         int                     *logflagsp, /* inode logging flags */
1784         int                     rsvd)           /* OK to allocate reserved blocks */
1785 {
1786         xfs_bmbt_rec_t          *base;  /* base of extent entry list */
1787         xfs_bmbt_rec_t          *ep;    /* extent list entry for idx */
1788 #ifdef XFS_BMAP_TRACE
1789         static char             fname[] = "xfs_bmap_add_extent_hole_delay";
1790 #endif
1791         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1792         xfs_filblks_t           newlen=0;       /* new indirect size */
1793         xfs_filblks_t           oldlen=0;       /* old indirect size */
1794         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1795         int                     state;  /* state bits, accessed thru macros */
1796         xfs_filblks_t           temp;   /* temp for indirect calculations */
1797         enum {                          /* bit number definitions for state */
1798                 LEFT_CONTIG,    RIGHT_CONTIG,
1799                 LEFT_DELAY,     RIGHT_DELAY,
1800                 LEFT_VALID,     RIGHT_VALID
1801         };
1802
1803 #define MASK(b)                 (1 << (b))
1804 #define MASK2(a,b)              (MASK(a) | MASK(b))
1805 #define STATE_SET(b,v)          ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1806 #define STATE_TEST(b)           (state & MASK(b))
1807 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
1808                                        ((state &= ~MASK(b)), 0))
1809 #define SWITCH_STATE            (state & MASK2(LEFT_CONTIG, RIGHT_CONTIG))
1810
1811         base = ip->i_df.if_u1.if_extents;
1812         ep = &base[idx];
1813         state = 0;
1814         ASSERT(ISNULLSTARTBLOCK(new->br_startblock));
1815         /*
1816          * Check and set flags if this segment has a left neighbor
1817          */
1818         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1819                 xfs_bmbt_get_all(ep - 1, &left);
1820                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(left.br_startblock));
1821         }
1822         /*
1823          * Check and set flags if the current (right) segment exists.
1824          * If it doesn't exist, we're converting the hole at end-of-file.
1825          */
1826         if (STATE_SET_TEST(RIGHT_VALID,
1827                            idx <
1828                            ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1829                 xfs_bmbt_get_all(ep, &right);
1830                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(right.br_startblock));
1831         }
1832         /*
1833          * Set contiguity flags on the left and right neighbors.
1834          * Don't let extents get too large, even if the pieces are contiguous.
1835          */
1836         STATE_SET(LEFT_CONTIG,
1837                 STATE_TEST(LEFT_VALID) && STATE_TEST(LEFT_DELAY) &&
1838                 left.br_startoff + left.br_blockcount == new->br_startoff &&
1839                 left.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1840         STATE_SET(RIGHT_CONTIG,
1841                 STATE_TEST(RIGHT_VALID) && STATE_TEST(RIGHT_DELAY) &&
1842                 new->br_startoff + new->br_blockcount == right.br_startoff &&
1843                 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1844                 (!STATE_TEST(LEFT_CONTIG) ||
1845                  (left.br_blockcount + new->br_blockcount +
1846                      right.br_blockcount <= MAXEXTLEN)));
1847         /*
1848          * Switch out based on the contiguity flags.
1849          */
1850         switch (SWITCH_STATE) {
1851
1852         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1853                 /*
1854                  * New allocation is contiguous with delayed allocations
1855                  * on the left and on the right.
1856                  * Merge all three into a single extent list entry.
1857                  */
1858                 temp = left.br_blockcount + new->br_blockcount +
1859                         right.br_blockcount;
1860                 xfs_bmap_trace_pre_update(fname, "LC|RC", ip, idx - 1,
1861                         XFS_DATA_FORK);
1862                 xfs_bmbt_set_blockcount(ep - 1, temp);
1863                 oldlen = STARTBLOCKVAL(left.br_startblock) +
1864                         STARTBLOCKVAL(new->br_startblock) +
1865                         STARTBLOCKVAL(right.br_startblock);
1866                 newlen = xfs_bmap_worst_indlen(ip, temp);
1867                 xfs_bmbt_set_startblock(ep - 1, NULLSTARTBLOCK((int)newlen));
1868                 xfs_bmap_trace_post_update(fname, "LC|RC", ip, idx - 1,
1869                         XFS_DATA_FORK);
1870                 xfs_bmap_trace_delete(fname, "LC|RC", ip, idx, 1,
1871                         XFS_DATA_FORK);
1872                 xfs_bmap_delete_exlist(ip, idx, 1, XFS_DATA_FORK);
1873                 ip->i_df.if_lastex = idx - 1;
1874                 break;
1875
1876         case MASK(LEFT_CONTIG):
1877                 /*
1878                  * New allocation is contiguous with a delayed allocation
1879                  * on the left.
1880                  * Merge the new allocation with the left neighbor.
1881                  */
1882                 temp = left.br_blockcount + new->br_blockcount;
1883                 xfs_bmap_trace_pre_update(fname, "LC", ip, idx - 1,
1884                         XFS_DATA_FORK);
1885                 xfs_bmbt_set_blockcount(ep - 1, temp);
1886                 oldlen = STARTBLOCKVAL(left.br_startblock) +
1887                         STARTBLOCKVAL(new->br_startblock);
1888                 newlen = xfs_bmap_worst_indlen(ip, temp);
1889                 xfs_bmbt_set_startblock(ep - 1, NULLSTARTBLOCK((int)newlen));
1890                 xfs_bmap_trace_post_update(fname, "LC", ip, idx - 1,
1891                         XFS_DATA_FORK);
1892                 ip->i_df.if_lastex = idx - 1;
1893                 break;
1894
1895         case MASK(RIGHT_CONTIG):
1896                 /*
1897                  * New allocation is contiguous with a delayed allocation
1898                  * on the right.
1899                  * Merge the new allocation with the right neighbor.
1900                  */
1901                 xfs_bmap_trace_pre_update(fname, "RC", ip, idx, XFS_DATA_FORK);
1902                 temp = new->br_blockcount + right.br_blockcount;
1903                 oldlen = STARTBLOCKVAL(new->br_startblock) +
1904                         STARTBLOCKVAL(right.br_startblock);
1905                 newlen = xfs_bmap_worst_indlen(ip, temp);
1906                 xfs_bmbt_set_allf(ep, new->br_startoff,
1907                         NULLSTARTBLOCK((int)newlen), temp, right.br_state);
1908                 xfs_bmap_trace_post_update(fname, "RC", ip, idx, XFS_DATA_FORK);
1909                 ip->i_df.if_lastex = idx;
1910                 break;
1911
1912         case 0:
1913                 /*
1914                  * New allocation is not contiguous with another
1915                  * delayed allocation.
1916                  * Insert a new entry.
1917                  */
1918                 oldlen = newlen = 0;
1919                 xfs_bmap_trace_insert(fname, "0", ip, idx, 1, new, NULL,
1920                         XFS_DATA_FORK);
1921                 xfs_bmap_insert_exlist(ip, idx, 1, new, XFS_DATA_FORK);
1922                 ip->i_df.if_lastex = idx;
1923                 break;
1924         }
1925         if (oldlen != newlen) {
1926                 ASSERT(oldlen > newlen);
1927                 xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
1928                         (int)(oldlen - newlen), rsvd);
1929                 /*
1930                  * Nothing to do for disk quota accounting here.
1931                  */
1932         }
1933         *logflagsp = 0;
1934         return 0;
1935 #undef  MASK
1936 #undef  MASK2
1937 #undef  STATE_SET
1938 #undef  STATE_TEST
1939 #undef  STATE_SET_TEST
1940 #undef  SWITCH_STATE
1941 }
1942
1943 /*
1944  * Called by xfs_bmap_add_extent to handle cases converting a hole
1945  * to a real allocation.
1946  */
1947 STATIC int                              /* error */
1948 xfs_bmap_add_extent_hole_real(
1949         xfs_inode_t             *ip,    /* incore inode pointer */
1950         xfs_extnum_t            idx,    /* extent number to update/insert */
1951         xfs_btree_cur_t         *cur,   /* if null, not a btree */
1952         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
1953         int                     *logflagsp, /* inode logging flags */
1954         int                     whichfork) /* data or attr fork */
1955 {
1956         xfs_bmbt_rec_t          *ep;    /* pointer to extent entry ins. point */
1957         int                     error;  /* error return value */
1958 #ifdef XFS_BMAP_TRACE
1959         static char             fname[] = "xfs_bmap_add_extent_hole_real";
1960 #endif
1961         int                     i;      /* temp state */
1962         xfs_ifork_t             *ifp;   /* inode fork pointer */
1963         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1964         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1965         int                     state;  /* state bits, accessed thru macros */
1966         enum {                          /* bit number definitions for state */
1967                 LEFT_CONTIG,    RIGHT_CONTIG,
1968                 LEFT_DELAY,     RIGHT_DELAY,
1969                 LEFT_VALID,     RIGHT_VALID
1970         };
1971
1972 #define MASK(b)                 (1 << (b))
1973 #define MASK2(a,b)              (MASK(a) | MASK(b))
1974 #define STATE_SET(b,v)          ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1975 #define STATE_TEST(b)           (state & MASK(b))
1976 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
1977                                        ((state &= ~MASK(b)), 0))
1978 #define SWITCH_STATE            (state & MASK2(LEFT_CONTIG, RIGHT_CONTIG))
1979
1980         ifp = XFS_IFORK_PTR(ip, whichfork);
1981         ASSERT(idx <= ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
1982         ep = &ifp->if_u1.if_extents[idx];
1983         state = 0;
1984         /*
1985          * Check and set flags if this segment has a left neighbor.
1986          */
1987         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1988                 xfs_bmbt_get_all(ep - 1, &left);
1989                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(left.br_startblock));
1990         }
1991         /*
1992          * Check and set flags if this segment has a current value.
1993          * Not true if we're inserting into the "hole" at eof.
1994          */
1995         if (STATE_SET_TEST(RIGHT_VALID,
1996                            idx <
1997                            ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1998                 xfs_bmbt_get_all(ep, &right);
1999                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(right.br_startblock));
2000         }
2001         /*
2002          * We're inserting a real allocation between "left" and "right".
2003          * Set the contiguity flags.  Don't let extents get too large.
2004          */
2005         STATE_SET(LEFT_CONTIG,
2006                 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
2007                 left.br_startoff + left.br_blockcount == new->br_startoff &&
2008                 left.br_startblock + left.br_blockcount == new->br_startblock &&
2009                 left.br_state == new->br_state &&
2010                 left.br_blockcount + new->br_blockcount <= MAXEXTLEN);
2011         STATE_SET(RIGHT_CONTIG,
2012                 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
2013                 new->br_startoff + new->br_blockcount == right.br_startoff &&
2014                 new->br_startblock + new->br_blockcount ==
2015                     right.br_startblock &&
2016                 new->br_state == right.br_state &&
2017                 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2018                 (!STATE_TEST(LEFT_CONTIG) ||
2019                  left.br_blockcount + new->br_blockcount +
2020                      right.br_blockcount <= MAXEXTLEN));
2021
2022         /*
2023          * Select which case we're in here, and implement it.
2024          */
2025         switch (SWITCH_STATE) {
2026
2027         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
2028                 /*
2029                  * New allocation is contiguous with real allocations on the
2030                  * left and on the right.
2031                  * Merge all three into a single extent list entry.
2032                  */
2033                 xfs_bmap_trace_pre_update(fname, "LC|RC", ip, idx - 1,
2034                         whichfork);
2035                 xfs_bmbt_set_blockcount(ep - 1,
2036                         left.br_blockcount + new->br_blockcount +
2037                         right.br_blockcount);
2038                 xfs_bmap_trace_post_update(fname, "LC|RC", ip, idx - 1,
2039                         whichfork);
2040                 xfs_bmap_trace_delete(fname, "LC|RC", ip,
2041                         idx, 1, whichfork);
2042                 xfs_bmap_delete_exlist(ip, idx, 1, whichfork);
2043                 ifp->if_lastex = idx - 1;
2044                 XFS_IFORK_NEXT_SET(ip, whichfork,
2045                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2046                 if (cur == NULL) {
2047                         *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2048                         return 0;
2049                 }
2050                 *logflagsp = XFS_ILOG_CORE;
2051                 if ((error = xfs_bmbt_lookup_eq(cur, right.br_startoff,
2052                                 right.br_startblock, right.br_blockcount, &i)))
2053                         return error;
2054                 ASSERT(i == 1);
2055                 if ((error = xfs_bmbt_delete(cur, &i)))
2056                         return error;
2057                 ASSERT(i == 1);
2058                 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
2059                         return error;
2060                 ASSERT(i == 1);
2061                 error = xfs_bmbt_update(cur, left.br_startoff,
2062                                 left.br_startblock,
2063                                 left.br_blockcount + new->br_blockcount +
2064                                 right.br_blockcount, left.br_state);
2065                 return error;
2066
2067         case MASK(LEFT_CONTIG):
2068                 /*
2069                  * New allocation is contiguous with a real allocation
2070                  * on the left.
2071                  * Merge the new allocation with the left neighbor.
2072                  */
2073                 xfs_bmap_trace_pre_update(fname, "LC", ip, idx - 1, whichfork);
2074                 xfs_bmbt_set_blockcount(ep - 1,
2075                         left.br_blockcount + new->br_blockcount);
2076                 xfs_bmap_trace_post_update(fname, "LC", ip, idx - 1, whichfork);
2077                 ifp->if_lastex = idx - 1;
2078                 if (cur == NULL) {
2079                         *logflagsp = XFS_ILOG_FEXT(whichfork);
2080                         return 0;
2081                 }
2082                 *logflagsp = 0;
2083                 if ((error = xfs_bmbt_lookup_eq(cur, left.br_startoff,
2084                                 left.br_startblock, left.br_blockcount, &i)))
2085                         return error;
2086                 ASSERT(i == 1);
2087                 error = xfs_bmbt_update(cur, left.br_startoff,
2088                                 left.br_startblock,
2089                                 left.br_blockcount + new->br_blockcount,
2090                                 left.br_state);
2091                 return error;
2092
2093         case MASK(RIGHT_CONTIG):
2094                 /*
2095                  * New allocation is contiguous with a real allocation
2096                  * on the right.
2097                  * Merge the new allocation with the right neighbor.
2098                  */
2099                 xfs_bmap_trace_pre_update(fname, "RC", ip, idx, whichfork);
2100                 xfs_bmbt_set_allf(ep, new->br_startoff, new->br_startblock,
2101                         new->br_blockcount + right.br_blockcount,
2102                         right.br_state);
2103                 xfs_bmap_trace_post_update(fname, "RC", ip, idx, whichfork);
2104                 ifp->if_lastex = idx;
2105                 if (cur == NULL) {
2106                         *logflagsp = XFS_ILOG_FEXT(whichfork);
2107                         return 0;
2108                 }
2109                 *logflagsp = 0;
2110                 if ((error = xfs_bmbt_lookup_eq(cur, right.br_startoff,
2111                                 right.br_startblock, right.br_blockcount, &i)))
2112                         return error;
2113                 ASSERT(i == 1);
2114                 error = xfs_bmbt_update(cur, new->br_startoff,
2115                                 new->br_startblock,
2116                                 new->br_blockcount + right.br_blockcount,
2117                                 right.br_state);
2118                 return error;
2119
2120         case 0:
2121                 /*
2122                  * New allocation is not contiguous with another
2123                  * real allocation.
2124                  * Insert a new entry.
2125                  */
2126                 xfs_bmap_trace_insert(fname, "0", ip, idx, 1, new, NULL,
2127                         whichfork);
2128                 xfs_bmap_insert_exlist(ip, idx, 1, new, whichfork);
2129                 ifp->if_lastex = idx;
2130                 XFS_IFORK_NEXT_SET(ip, whichfork,
2131                         XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2132                 if (cur == NULL) {
2133                         *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2134                         return 0;
2135                 }
2136                 *logflagsp = XFS_ILOG_CORE;
2137                 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2138                                 new->br_startblock, new->br_blockcount, &i)))
2139                         return error;
2140                 ASSERT(i == 0);
2141                 cur->bc_rec.b.br_state = new->br_state;
2142                 if ((error = xfs_bmbt_insert(cur, &i)))
2143                         return error;
2144                 ASSERT(i == 1);
2145                 return 0;
2146         }
2147 #undef  MASK
2148 #undef  MASK2
2149 #undef  STATE_SET
2150 #undef  STATE_TEST
2151 #undef  STATE_SET_TEST
2152 #undef  SWITCH_STATE
2153         /* NOTREACHED */
2154         ASSERT(0);
2155         return 0; /* keep gcc quite */
2156 }
2157
2158 #define XFS_ALLOC_GAP_UNITS     4
2159
2160 /*
2161  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2162  * It figures out where to ask the underlying allocator to put the new extent.
2163  */
2164 STATIC int                              /* error */
2165 xfs_bmap_alloc(
2166         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2167 {
2168         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
2169         xfs_alloctype_t atype=0;        /* type for allocation routines */
2170         int             error;          /* error return value */
2171         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2172         xfs_mount_t     *mp;            /* mount point structure */
2173         int             nullfb;         /* true if ap->firstblock isn't set */
2174         int             rt;             /* true if inode is realtime */
2175 #ifdef __KERNEL__
2176         xfs_extlen_t    prod=0;         /* product factor for allocators */
2177         xfs_extlen_t    ralen=0;        /* realtime allocation length */
2178 #endif
2179
2180 #define ISVALID(x,y)    \
2181         (rt ? \
2182                 (x) < mp->m_sb.sb_rblocks : \
2183                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2184                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2185                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2186
2187         /*
2188          * Set up variables.
2189          */
2190         mp = ap->ip->i_mount;
2191         nullfb = ap->firstblock == NULLFSBLOCK;
2192         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2193         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2194 #ifdef __KERNEL__
2195         if (rt) {
2196                 xfs_extlen_t    extsz;          /* file extent size for rt */
2197                 xfs_fileoff_t   nexto;          /* next file offset */
2198                 xfs_extlen_t    orig_alen;      /* original ap->alen */
2199                 xfs_fileoff_t   orig_end;       /* original off+len */
2200                 xfs_fileoff_t   orig_off;       /* original ap->off */
2201                 xfs_extlen_t    mod_off;        /* modulus calculations */
2202                 xfs_fileoff_t   prevo;          /* previous file offset */
2203                 xfs_rtblock_t   rtx;            /* realtime extent number */
2204                 xfs_extlen_t    temp;           /* temp for rt calculations */
2205
2206                 /*
2207                  * Set prod to match the realtime extent size.
2208                  */
2209                 if (!(extsz = ap->ip->i_d.di_extsize))
2210                         extsz = mp->m_sb.sb_rextsize;
2211                 prod = extsz / mp->m_sb.sb_rextsize;
2212                 orig_off = ap->off;
2213                 orig_alen = ap->alen;
2214                 orig_end = orig_off + orig_alen;
2215                 /*
2216                  * If the file offset is unaligned vs. the extent size
2217                  * we need to align it.  This will be possible unless
2218                  * the file was previously written with a kernel that didn't
2219                  * perform this alignment.
2220                  */
2221                 mod_off = do_mod(orig_off, extsz);
2222                 if (mod_off) {
2223                         ap->alen += mod_off;
2224                         ap->off -= mod_off;
2225                 }
2226                 /*
2227                  * Same adjustment for the end of the requested area.
2228                  */
2229                 if ((temp = (ap->alen % extsz)))
2230                         ap->alen += extsz - temp;
2231                 /*
2232                  * If the previous block overlaps with this proposed allocation
2233                  * then move the start forward without adjusting the length.
2234                  */
2235                 prevo =
2236                         ap->prevp->br_startoff == NULLFILEOFF ?
2237                                 0 :
2238                                 (ap->prevp->br_startoff +
2239                                  ap->prevp->br_blockcount);
2240                 if (ap->off != orig_off && ap->off < prevo)
2241                         ap->off = prevo;
2242                 /*
2243                  * If the next block overlaps with this proposed allocation
2244                  * then move the start back without adjusting the length,
2245                  * but not before offset 0.
2246                  * This may of course make the start overlap previous block,
2247                  * and if we hit the offset 0 limit then the next block
2248                  * can still overlap too.
2249                  */
2250                 nexto = (ap->eof || ap->gotp->br_startoff == NULLFILEOFF) ?
2251                         NULLFILEOFF : ap->gotp->br_startoff;
2252                 if (!ap->eof &&
2253                     ap->off + ap->alen != orig_end &&
2254                     ap->off + ap->alen > nexto)
2255                         ap->off = nexto > ap->alen ? nexto - ap->alen : 0;
2256                 /*
2257                  * If we're now overlapping the next or previous extent that
2258                  * means we can't fit an extsz piece in this hole.  Just move
2259                  * the start forward to the first valid spot and set
2260                  * the length so we hit the end.
2261                  */
2262                 if ((ap->off != orig_off && ap->off < prevo) ||
2263                     (ap->off + ap->alen != orig_end &&
2264                      ap->off + ap->alen > nexto)) {
2265                         ap->off = prevo;
2266                         ap->alen = nexto - prevo;
2267                 }
2268                 /*
2269                  * If the result isn't a multiple of rtextents we need to
2270                  * remove blocks until it is.
2271                  */
2272                 if ((temp = (ap->alen % mp->m_sb.sb_rextsize))) {
2273                         /*
2274                          * We're not covering the original request, or
2275                          * we won't be able to once we fix the length.
2276                          */
2277                         if (orig_off < ap->off ||
2278                             orig_end > ap->off + ap->alen ||
2279                             ap->alen - temp < orig_alen)
2280                                 return XFS_ERROR(EINVAL);
2281                         /*
2282                          * Try to fix it by moving the start up.
2283                          */
2284                         if (ap->off + temp <= orig_off) {
2285                                 ap->alen -= temp;
2286                                 ap->off += temp;
2287                         }
2288                         /*
2289                          * Try to fix it by moving the end in.
2290                          */
2291                         else if (ap->off + ap->alen - temp >= orig_end)
2292                                 ap->alen -= temp;
2293                         /*
2294                          * Set the start to the minimum then trim the length.
2295                          */
2296                         else {
2297                                 ap->alen -= orig_off - ap->off;
2298                                 ap->off = orig_off;
2299                                 ap->alen -= ap->alen % mp->m_sb.sb_rextsize;
2300                         }
2301                         /*
2302                          * Result doesn't cover the request, fail it.
2303                          */
2304                         if (orig_off < ap->off || orig_end > ap->off + ap->alen)
2305                                 return XFS_ERROR(EINVAL);
2306                 }
2307                 ASSERT(ap->alen % mp->m_sb.sb_rextsize == 0);
2308                 /*
2309                  * If the offset & length are not perfectly aligned
2310                  * then kill prod, it will just get us in trouble.
2311                  */
2312                 if (do_mod(ap->off, extsz) || ap->alen % extsz)
2313                         prod = 1;
2314                 /*
2315                  * Set ralen to be the actual requested length in rtextents.
2316                  */
2317                 ralen = ap->alen / mp->m_sb.sb_rextsize;
2318                 /*
2319                  * If the old value was close enough to MAXEXTLEN that
2320                  * we rounded up to it, cut it back so it's valid again.
2321                  * Note that if it's a really large request (bigger than
2322                  * MAXEXTLEN), we don't hear about that number, and can't
2323                  * adjust the starting point to match it.
2324                  */
2325                 if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2326                         ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2327                 /*
2328                  * If it's an allocation to an empty file at offset 0,
2329                  * pick an extent that will space things out in the rt area.
2330                  */
2331                 if (ap->eof && ap->off == 0) {
2332                         error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2333                         if (error)
2334                                 return error;
2335                         ap->rval = rtx * mp->m_sb.sb_rextsize;
2336                 } else
2337                         ap->rval = 0;
2338         }
2339 #else
2340         if (rt)
2341                 ap->rval = 0;
2342 #endif  /* __KERNEL__ */
2343         else if (nullfb)
2344                 ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2345         else
2346                 ap->rval = ap->firstblock;
2347         /*
2348          * If allocating at eof, and there's a previous real block,
2349          * try to use it's last block as our starting point.
2350          */
2351         if (ap->eof && ap->prevp->br_startoff != NULLFILEOFF &&
2352             !ISNULLSTARTBLOCK(ap->prevp->br_startblock) &&
2353             ISVALID(ap->prevp->br_startblock + ap->prevp->br_blockcount,
2354                     ap->prevp->br_startblock)) {
2355                 ap->rval = ap->prevp->br_startblock + ap->prevp->br_blockcount;
2356                 /*
2357                  * Adjust for the gap between prevp and us.
2358                  */
2359                 adjust = ap->off -
2360                         (ap->prevp->br_startoff + ap->prevp->br_blockcount);
2361                 if (adjust &&
2362                     ISVALID(ap->rval + adjust, ap->prevp->br_startblock))
2363                         ap->rval += adjust;
2364         }
2365         /*
2366          * If not at eof, then compare the two neighbor blocks.
2367          * Figure out whether either one gives us a good starting point,
2368          * and pick the better one.
2369          */
2370         else if (!ap->eof) {
2371                 xfs_fsblock_t   gotbno;         /* right side block number */
2372                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
2373                 xfs_fsblock_t   prevbno;        /* left side block number */
2374                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
2375
2376                 /*
2377                  * If there's a previous (left) block, select a requested
2378                  * start block based on it.
2379                  */
2380                 if (ap->prevp->br_startoff != NULLFILEOFF &&
2381                     !ISNULLSTARTBLOCK(ap->prevp->br_startblock) &&
2382                     (prevbno = ap->prevp->br_startblock +
2383                                ap->prevp->br_blockcount) &&
2384                     ISVALID(prevbno, ap->prevp->br_startblock)) {
2385                         /*
2386                          * Calculate gap to end of previous block.
2387                          */
2388                         adjust = prevdiff = ap->off -
2389                                 (ap->prevp->br_startoff +
2390                                  ap->prevp->br_blockcount);
2391                         /*
2392                          * Figure the startblock based on the previous block's
2393                          * end and the gap size.
2394                          * Heuristic!
2395                          * If the gap is large relative to the piece we're
2396                          * allocating, or using it gives us an invalid block
2397                          * number, then just use the end of the previous block.
2398                          */
2399                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2400                             ISVALID(prevbno + prevdiff,
2401                                     ap->prevp->br_startblock))
2402                                 prevbno += adjust;
2403                         else
2404                                 prevdiff += adjust;
2405                         /*
2406                          * If the firstblock forbids it, can't use it,
2407                          * must use default.
2408                          */
2409                         if (!rt && !nullfb &&
2410                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2411                                 prevbno = NULLFSBLOCK;
2412                 }
2413                 /*
2414                  * No previous block or can't follow it, just default.
2415                  */
2416                 else
2417                         prevbno = NULLFSBLOCK;
2418                 /*
2419                  * If there's a following (right) block, select a requested
2420                  * start block based on it.
2421                  */
2422                 if (!ISNULLSTARTBLOCK(ap->gotp->br_startblock)) {
2423                         /*
2424                          * Calculate gap to start of next block.
2425                          */
2426                         adjust = gotdiff = ap->gotp->br_startoff - ap->off;
2427                         /*
2428                          * Figure the startblock based on the next block's
2429                          * start and the gap size.
2430                          */
2431                         gotbno = ap->gotp->br_startblock;
2432                         /*
2433                          * Heuristic!
2434                          * If the gap is large relative to the piece we're
2435                          * allocating, or using it gives us an invalid block
2436                          * number, then just use the start of the next block
2437                          * offset by our length.
2438                          */
2439                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2440                             ISVALID(gotbno - gotdiff, gotbno))
2441                                 gotbno -= adjust;
2442                         else if (ISVALID(gotbno - ap->alen, gotbno)) {
2443                                 gotbno -= ap->alen;
2444                                 gotdiff += adjust - ap->alen;
2445                         } else
2446                                 gotdiff += adjust;
2447                         /*
2448                          * If the firstblock forbids it, can't use it,
2449                          * must use default.
2450                          */
2451                         if (!rt && !nullfb &&
2452                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2453                                 gotbno = NULLFSBLOCK;
2454                 }
2455                 /*
2456                  * No next block, just default.
2457                  */
2458                 else
2459                         gotbno = NULLFSBLOCK;
2460                 /*
2461                  * If both valid, pick the better one, else the only good
2462                  * one, else ap->rval is already set (to 0 or the inode block).
2463                  */
2464                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2465                         ap->rval = prevdiff <= gotdiff ? prevbno : gotbno;
2466                 else if (prevbno != NULLFSBLOCK)
2467                         ap->rval = prevbno;
2468                 else if (gotbno != NULLFSBLOCK)
2469                         ap->rval = gotbno;
2470         }
2471         /*
2472          * If allowed, use ap->rval; otherwise must use firstblock since
2473          * it's in the right allocation group.
2474          */
2475         if (nullfb || rt || XFS_FSB_TO_AGNO(mp, ap->rval) == fb_agno)
2476                 ;
2477         else
2478                 ap->rval = ap->firstblock;
2479         /*
2480          * Realtime allocation, done through xfs_rtallocate_extent.
2481          */
2482         if (rt) {
2483 #ifndef __KERNEL__
2484                 ASSERT(0);
2485 #else
2486                 xfs_rtblock_t   rtb;
2487
2488                 atype = ap->rval == 0 ?
2489                         XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2490                 do_div(ap->rval, mp->m_sb.sb_rextsize);
2491                 rtb = ap->rval;
2492                 ap->alen = ralen;
2493                 if ((error = xfs_rtallocate_extent(ap->tp, ap->rval, 1, ap->alen,
2494                                 &ralen, atype, ap->wasdel, prod, &rtb)))
2495                         return error;
2496                 if (rtb == NULLFSBLOCK && prod > 1 &&
2497                     (error = xfs_rtallocate_extent(ap->tp, ap->rval, 1,
2498                                                    ap->alen, &ralen, atype,
2499                                                    ap->wasdel, 1, &rtb)))
2500                         return error;
2501                 ap->rval = rtb;
2502                 if (ap->rval != NULLFSBLOCK) {
2503                         ap->rval *= mp->m_sb.sb_rextsize;
2504                         ralen *= mp->m_sb.sb_rextsize;
2505                         ap->alen = ralen;
2506                         ap->ip->i_d.di_nblocks += ralen;
2507                         xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2508                         if (ap->wasdel)
2509                                 ap->ip->i_delayed_blks -= ralen;
2510                         /*
2511                          * Adjust the disk quota also. This was reserved
2512                          * earlier.
2513                          */
2514                         XFS_TRANS_MOD_DQUOT_BYINO(mp, ap->tp, ap->ip,
2515                                 ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2516                                                 XFS_TRANS_DQ_RTBCOUNT,
2517                                 (long) ralen);
2518                 } else
2519                         ap->alen = 0;
2520 #endif  /* __KERNEL__ */
2521         }
2522         /*
2523          * Normal allocation, done through xfs_alloc_vextent.
2524          */
2525         else {
2526                 xfs_agnumber_t  ag;
2527                 xfs_alloc_arg_t args;
2528                 xfs_extlen_t    blen;
2529                 xfs_extlen_t    delta;
2530                 int             isaligned;
2531                 xfs_extlen_t    longest;
2532                 xfs_extlen_t    need;
2533                 xfs_extlen_t    nextminlen=0;
2534                 int             notinit;
2535                 xfs_perag_t     *pag;
2536                 xfs_agnumber_t  startag;
2537                 int             tryagain;
2538
2539                 tryagain = isaligned = 0;
2540                 args.tp = ap->tp;
2541                 args.mp = mp;
2542                 args.fsbno = ap->rval;
2543                 args.maxlen = MIN(ap->alen, mp->m_sb.sb_agblocks);
2544                 blen = 0;
2545                 if (nullfb) {
2546                         args.type = XFS_ALLOCTYPE_START_BNO;
2547                         args.total = ap->total;
2548                         /*
2549                          * Find the longest available space.
2550                          * We're going to try for the whole allocation at once.
2551                          */
2552                         startag = ag = XFS_FSB_TO_AGNO(mp, args.fsbno);
2553                         notinit = 0;
2554                         down_read(&mp->m_peraglock);
2555                         while (blen < ap->alen) {
2556                                 pag = &mp->m_perag[ag];
2557                                 if (!pag->pagf_init &&
2558                                     (error = xfs_alloc_pagf_init(mp, args.tp,
2559                                             ag, XFS_ALLOC_FLAG_TRYLOCK))) {
2560                                         up_read(&mp->m_peraglock);
2561                                         return error;
2562                                 }
2563                                 /*
2564                                  * See xfs_alloc_fix_freelist...
2565                                  */
2566                                 if (pag->pagf_init) {
2567                                         need = XFS_MIN_FREELIST_PAG(pag, mp);
2568                                         delta = need > pag->pagf_flcount ?
2569                                                 need - pag->pagf_flcount : 0;
2570                                         longest = (pag->pagf_longest > delta) ?
2571                                                 (pag->pagf_longest - delta) :
2572                                                 (pag->pagf_flcount > 0 ||
2573                                                  pag->pagf_longest > 0);
2574                                         if (blen < longest)
2575                                                 blen = longest;
2576                                 } else
2577                                         notinit = 1;
2578                                 if (++ag == mp->m_sb.sb_agcount)
2579                                         ag = 0;
2580                                 if (ag == startag)
2581                                         break;
2582                         }
2583                         up_read(&mp->m_peraglock);
2584                         /*
2585                          * Since the above loop did a BUF_TRYLOCK, it is
2586                          * possible that there is space for this request.
2587                          */
2588                         if (notinit || blen < ap->minlen)
2589                                 args.minlen = ap->minlen;
2590                         /*
2591                          * If the best seen length is less than the request
2592                          * length, use the best as the minimum.
2593                          */
2594                         else if (blen < ap->alen)
2595                                 args.minlen = blen;
2596                         /*
2597                          * Otherwise we've seen an extent as big as alen,
2598                          * use that as the minimum.
2599                          */
2600                         else
2601                                 args.minlen = ap->alen;
2602                 } else if (ap->low) {
2603                         args.type = XFS_ALLOCTYPE_FIRST_AG;
2604                         args.total = args.minlen = ap->minlen;
2605                 } else {
2606                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
2607                         args.total = ap->total;
2608                         args.minlen = ap->minlen;
2609                 }
2610                 if (ap->ip->i_d.di_extsize) {
2611                         args.prod = ap->ip->i_d.di_extsize;
2612                         if ((args.mod = (xfs_extlen_t)do_mod(ap->off, args.prod)))
2613                                 args.mod = (xfs_extlen_t)(args.prod - args.mod);
2614                 } else if (mp->m_sb.sb_blocksize >= NBPP) {
2615                         args.prod = 1;
2616                         args.mod = 0;
2617                 } else {
2618                         args.prod = NBPP >> mp->m_sb.sb_blocklog;
2619                         if ((args.mod = (xfs_extlen_t)(do_mod(ap->off, args.prod))))
2620                                 args.mod = (xfs_extlen_t)(args.prod - args.mod);
2621                 }
2622                 /*
2623                  * If we are not low on available data blocks, and the
2624                  * underlying logical volume manager is a stripe, and
2625                  * the file offset is zero then try to allocate data
2626                  * blocks on stripe unit boundary.
2627                  * NOTE: ap->aeof is only set if the allocation length
2628                  * is >= the stripe unit and the allocation offset is
2629                  * at the end of file.
2630                  */
2631                 if (!ap->low && ap->aeof) {
2632                         if (!ap->off) {
2633                                 args.alignment = mp->m_dalign;
2634                                 atype = args.type;
2635                                 isaligned = 1;
2636                                 /*
2637                                  * Adjust for alignment
2638                                  */
2639                                 if (blen > args.alignment && blen <= ap->alen)
2640                                         args.minlen = blen - args.alignment;
2641                                 args.minalignslop = 0;
2642                         } else {
2643                                 /*
2644                                  * First try an exact bno allocation.
2645                                  * If it fails then do a near or start bno
2646                                  * allocation with alignment turned on.
2647                                  */
2648                                 atype = args.type;
2649                                 tryagain = 1;
2650                                 args.type = XFS_ALLOCTYPE_THIS_BNO;
2651                                 args.alignment = 1;
2652                                 /*
2653                                  * Compute the minlen+alignment for the
2654                                  * next case.  Set slop so that the value
2655                                  * of minlen+alignment+slop doesn't go up
2656                                  * between the calls.
2657                                  */
2658                                 if (blen > mp->m_dalign && blen <= ap->alen)
2659                                         nextminlen = blen - mp->m_dalign;
2660                                 else
2661                                         nextminlen = args.minlen;
2662                                 if (nextminlen + mp->m_dalign > args.minlen + 1)
2663                                         args.minalignslop =
2664                                                 nextminlen + mp->m_dalign -
2665                                                 args.minlen - 1;
2666                                 else
2667                                         args.minalignslop = 0;
2668                         }
2669                 } else {
2670                         args.alignment = 1;
2671                         args.minalignslop = 0;
2672                 }
2673                 args.minleft = ap->minleft;
2674                 args.wasdel = ap->wasdel;
2675                 args.isfl = 0;
2676                 args.userdata = ap->userdata;
2677                 if ((error = xfs_alloc_vextent(&args)))
2678                         return error;
2679                 if (tryagain && args.fsbno == NULLFSBLOCK) {
2680                         /*
2681                          * Exact allocation failed. Now try with alignment
2682                          * turned on.
2683                          */
2684                         args.type = atype;
2685                         args.fsbno = ap->rval;
2686                         args.alignment = mp->m_dalign;
2687                         args.minlen = nextminlen;
2688                         args.minalignslop = 0;
2689                         isaligned = 1;
2690                         if ((error = xfs_alloc_vextent(&args)))
2691                                 return error;
2692                 }
2693                 if (isaligned && args.fsbno == NULLFSBLOCK) {
2694                         /*
2695                          * allocation failed, so turn off alignment and
2696                          * try again.
2697                          */
2698                         args.type = atype;
2699                         args.fsbno = ap->rval;
2700                         args.alignment = 0;
2701                         if ((error = xfs_alloc_vextent(&args)))
2702                                 return error;
2703                 }
2704                 if (args.fsbno == NULLFSBLOCK && nullfb &&
2705                     args.minlen > ap->minlen) {
2706                         args.minlen = ap->minlen;
2707                         args.type = XFS_ALLOCTYPE_START_BNO;
2708                         args.fsbno = ap->rval;
2709                         if ((error = xfs_alloc_vextent(&args)))
2710                                 return error;
2711                 }
2712                 if (args.fsbno == NULLFSBLOCK && nullfb) {
2713                         args.fsbno = 0;
2714                         args.type = XFS_ALLOCTYPE_FIRST_AG;
2715                         args.total = ap->minlen;
2716                         args.minleft = 0;
2717                         if ((error = xfs_alloc_vextent(&args)))
2718                                 return error;
2719                         ap->low = 1;
2720                 }
2721                 if (args.fsbno != NULLFSBLOCK) {
2722                         ap->firstblock = ap->rval = args.fsbno;
2723                         ASSERT(nullfb || fb_agno == args.agno ||
2724                                (ap->low && fb_agno < args.agno));
2725                         ap->alen = args.len;
2726                         ap->ip->i_d.di_nblocks += args.len;
2727                         xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2728                         if (ap->wasdel)
2729                                 ap->ip->i_delayed_blks -= args.len;
2730                         /*
2731                          * Adjust the disk quota also. This was reserved
2732                          * earlier.
2733                          */
2734                         XFS_TRANS_MOD_DQUOT_BYINO(mp, ap->tp, ap->ip,
2735                                 ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2736                                                 XFS_TRANS_DQ_BCOUNT,
2737                                 (long) args.len);
2738                 } else {
2739                         ap->rval = NULLFSBLOCK;
2740                         ap->alen = 0;
2741                 }
2742         }
2743         return 0;
2744 #undef  ISVALID
2745 }
2746
2747 /*
2748  * Transform a btree format file with only one leaf node, where the
2749  * extents list will fit in the inode, into an extents format file.
2750  * Since the extent list is already in-core, all we have to do is
2751  * give up the space for the btree root and pitch the leaf block.
2752  */
2753 STATIC int                              /* error */
2754 xfs_bmap_btree_to_extents(
2755         xfs_trans_t             *tp,    /* transaction pointer */
2756         xfs_inode_t             *ip,    /* incore inode pointer */
2757         xfs_btree_cur_t         *cur,   /* btree cursor */
2758         int                     *logflagsp, /* inode logging flags */
2759         int                     whichfork)  /* data or attr fork */
2760 {
2761         /* REFERENCED */
2762         xfs_bmbt_block_t        *cblock;/* child btree block */
2763         xfs_fsblock_t           cbno;   /* child block number */
2764         xfs_buf_t               *cbp;   /* child block's buffer */
2765         int                     error;  /* error return value */
2766         xfs_ifork_t             *ifp;   /* inode fork data */
2767         xfs_mount_t             *mp;    /* mount point structure */
2768         xfs_bmbt_ptr_t          *pp;    /* ptr to block address */
2769         xfs_bmbt_block_t        *rblock;/* root btree block */
2770
2771         ifp = XFS_IFORK_PTR(ip, whichfork);
2772         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2773         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
2774         rblock = ifp->if_broot;
2775         ASSERT(INT_GET(rblock->bb_level, ARCH_CONVERT) == 1);
2776         ASSERT(INT_GET(rblock->bb_numrecs, ARCH_CONVERT) == 1);
2777         ASSERT(XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes) == 1);
2778         mp = ip->i_mount;
2779         pp = XFS_BMAP_BROOT_PTR_ADDR(rblock, 1, ifp->if_broot_bytes);
2780         *logflagsp = 0;
2781 #ifdef DEBUG
2782         if ((error = xfs_btree_check_lptr(cur, INT_GET(*pp, ARCH_CONVERT), 1)))
2783                 return error;
2784 #endif
2785         cbno = INT_GET(*pp, ARCH_CONVERT);
2786         if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
2787                         XFS_BMAP_BTREE_REF)))
2788                 return error;
2789         cblock = XFS_BUF_TO_BMBT_BLOCK(cbp);
2790         if ((error = xfs_btree_check_lblock(cur, cblock, 0, cbp)))
2791                 return error;
2792         xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
2793         ip->i_d.di_nblocks--;
2794         XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
2795         xfs_trans_binval(tp, cbp);
2796         if (cur->bc_bufs[0] == cbp)
2797                 cur->bc_bufs[0] = NULL;
2798         xfs_iroot_realloc(ip, -1, whichfork);
2799         ASSERT(ifp->if_broot == NULL);
2800         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
2801         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
2802         *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2803         return 0;
2804 }
2805
2806 /*
2807  * Called by xfs_bmapi to update extent list structure and the btree
2808  * after removing space (or undoing a delayed allocation).
2809  */
2810 STATIC int                              /* error */
2811 xfs_bmap_del_extent(
2812         xfs_inode_t             *ip,    /* incore inode pointer */
2813         xfs_trans_t             *tp,    /* current transaction pointer */
2814         xfs_extnum_t            idx,    /* extent number to update/delete */
2815         xfs_bmap_free_t         *flist, /* list of extents to be freed */
2816         xfs_btree_cur_t         *cur,   /* if null, not a btree */
2817         xfs_bmbt_irec_t         *del,   /* data to remove from extent list */
2818         int                     *logflagsp, /* inode logging flags */
2819         int                     whichfork, /* data or attr fork */
2820         int                     rsvd)   /* OK to allocate reserved blocks */
2821 {
2822         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
2823         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
2824         xfs_fsblock_t           del_endblock=0; /* first block past del */
2825         xfs_fileoff_t           del_endoff;     /* first offset past del */
2826         int                     delay;  /* current block is delayed allocated */
2827         int                     do_fx;  /* free extent at end of routine */
2828         xfs_bmbt_rec_t          *ep;    /* current extent entry pointer */
2829         int                     error;  /* error return value */
2830         int                     flags;  /* inode logging flags */
2831 #ifdef XFS_BMAP_TRACE
2832         static char             fname[] = "xfs_bmap_del_extent";
2833 #endif
2834         xfs_bmbt_irec_t         got;    /* current extent entry */
2835         xfs_fileoff_t           got_endoff;     /* first offset past got */
2836         int                     i;      /* temp state */
2837         xfs_ifork_t             *ifp;   /* inode fork pointer */
2838         xfs_mount_t             *mp;    /* mount structure */
2839         xfs_filblks_t           nblks;  /* quota/sb block count */
2840         xfs_bmbt_irec_t         new;    /* new record to be inserted */
2841         /* REFERENCED */
2842         xfs_extnum_t            nextents;       /* number of extents in list */
2843         uint                    qfield; /* quota field to update */
2844         xfs_filblks_t           temp;   /* for indirect length calculations */
2845         xfs_filblks_t           temp2;  /* for indirect length calculations */
2846
2847         XFS_STATS_INC(xs_del_exlist);
2848         mp = ip->i_mount;
2849         ifp = XFS_IFORK_PTR(ip, whichfork);
2850         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2851         ASSERT(idx >= 0 && idx < nextents);
2852         ASSERT(del->br_blockcount > 0);
2853         ep = &ifp->if_u1.if_extents[idx];
2854         xfs_bmbt_get_all(ep, &got);
2855         ASSERT(got.br_startoff <= del->br_startoff);
2856         del_endoff = del->br_startoff + del->br_blockcount;
2857         got_endoff = got.br_startoff + got.br_blockcount;
2858         ASSERT(got_endoff >= del_endoff);
2859         delay = ISNULLSTARTBLOCK(got.br_startblock);
2860         ASSERT(ISNULLSTARTBLOCK(del->br_startblock) == delay);
2861         flags = 0;
2862         qfield = 0;
2863         error = 0;
2864         /*
2865          * If deleting a real allocation, must free up the disk space.
2866          */
2867         if (!delay) {
2868                 flags = XFS_ILOG_CORE;
2869                 /*
2870                  * Realtime allocation.  Free it and record di_nblocks update.
2871                  */
2872                 if (whichfork == XFS_DATA_FORK &&
2873                     (ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
2874                         xfs_fsblock_t   bno;
2875                         xfs_filblks_t   len;
2876
2877                         ASSERT(do_mod(del->br_blockcount,
2878                                       mp->m_sb.sb_rextsize) == 0);
2879                         ASSERT(do_mod(del->br_startblock,
2880                                       mp->m_sb.sb_rextsize) == 0);
2881                         bno = del->br_startblock;
2882                         len = del->br_blockcount;
2883                         do_div(bno, mp->m_sb.sb_rextsize);
2884                         do_div(len, mp->m_sb.sb_rextsize);
2885                         if ((error = xfs_rtfree_extent(ip->i_transp, bno,
2886                                         (xfs_extlen_t)len)))
2887                                 goto done;
2888                         do_fx = 0;
2889                         nblks = len * mp->m_sb.sb_rextsize;
2890                         qfield = XFS_TRANS_DQ_RTBCOUNT;
2891                 }
2892                 /*
2893                  * Ordinary allocation.
2894                  */
2895                 else {
2896                         do_fx = 1;
2897                         nblks = del->br_blockcount;
2898                         qfield = XFS_TRANS_DQ_BCOUNT;
2899                 }
2900                 /*
2901                  * Set up del_endblock and cur for later.
2902                  */
2903                 del_endblock = del->br_startblock + del->br_blockcount;
2904                 if (cur) {
2905                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
2906                                         got.br_startblock, got.br_blockcount,
2907                                         &i)))
2908                                 goto done;
2909                         ASSERT(i == 1);
2910                 }
2911                 da_old = da_new = 0;
2912         } else {
2913                 da_old = STARTBLOCKVAL(got.br_startblock);
2914                 da_new = 0;
2915                 nblks = 0;
2916                 do_fx = 0;
2917         }
2918         /*
2919          * Set flag value to use in switch statement.
2920          * Left-contig is 2, right-contig is 1.
2921          */
2922         switch (((got.br_startoff == del->br_startoff) << 1) |
2923                 (got_endoff == del_endoff)) {
2924         case 3:
2925                 /*
2926                  * Matches the whole extent.  Delete the entry.
2927                  */
2928                 xfs_bmap_trace_delete(fname, "3", ip, idx, 1, whichfork);
2929                 xfs_bmap_delete_exlist(ip, idx, 1, whichfork);
2930                 ifp->if_lastex = idx;
2931                 if (delay)
2932                         break;
2933                 XFS_IFORK_NEXT_SET(ip, whichfork,
2934                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2935                 flags |= XFS_ILOG_CORE;
2936                 if (!cur) {
2937                         flags |= XFS_ILOG_FEXT(whichfork);
2938                         break;
2939                 }
2940                 if ((error = xfs_bmbt_delete(cur, &i)))
2941                         goto done;
2942                 ASSERT(i == 1);
2943                 break;
2944
2945         case 2:
2946                 /*
2947                  * Deleting the first part of the extent.
2948                  */
2949                 xfs_bmap_trace_pre_update(fname, "2", ip, idx, whichfork);
2950                 xfs_bmbt_set_startoff(ep, del_endoff);
2951                 temp = got.br_blockcount - del->br_blockcount;
2952                 xfs_bmbt_set_blockcount(ep, temp);
2953                 ifp->if_lastex = idx;
2954                 if (delay) {
2955                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2956                                 da_old);
2957                         xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
2958                         xfs_bmap_trace_post_update(fname, "2", ip, idx,
2959                                 whichfork);
2960                         da_new = temp;
2961                         break;
2962                 }
2963                 xfs_bmbt_set_startblock(ep, del_endblock);
2964                 xfs_bmap_trace_post_update(fname, "2", ip, idx, whichfork);
2965                 if (!cur) {
2966                         flags |= XFS_ILOG_FEXT(whichfork);
2967                         break;
2968                 }
2969                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
2970                                 got.br_blockcount - del->br_blockcount,
2971                                 got.br_state)))
2972                         goto done;
2973                 break;
2974
2975         case 1:
2976                 /*
2977                  * Deleting the last part of the extent.
2978                  */
2979                 temp = got.br_blockcount - del->br_blockcount;
2980                 xfs_bmap_trace_pre_update(fname, "1", ip, idx, whichfork);
2981                 xfs_bmbt_set_blockcount(ep, temp);
2982                 ifp->if_lastex = idx;
2983                 if (delay) {
2984                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2985                                 da_old);
2986                         xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
2987                         xfs_bmap_trace_post_update(fname, "1", ip, idx,
2988                                 whichfork);
2989                         da_new = temp;
2990                         break;
2991                 }
2992                 xfs_bmap_trace_post_update(fname, "1", ip, idx, whichfork);
2993                 if (!cur) {
2994                         flags |= XFS_ILOG_FEXT(whichfork);
2995                         break;
2996                 }
2997                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
2998                                 got.br_startblock,
2999                                 got.br_blockcount - del->br_blockcount,
3000                                 got.br_state)))
3001                         goto done;
3002                 break;
3003
3004         case 0:
3005                 /*
3006                  * Deleting the middle of the extent.
3007                  */
3008                 temp = del->br_startoff - got.br_startoff;
3009                 xfs_bmap_trace_pre_update(fname, "0", ip, idx, whichfork);
3010                 xfs_bmbt_set_blockcount(ep, temp);
3011                 new.br_startoff = del_endoff;
3012                 temp2 = got_endoff - del_endoff;
3013                 new.br_blockcount = temp2;
3014                 new.br_state = got.br_state;
3015                 if (!delay) {
3016                         new.br_startblock = del_endblock;
3017                         flags |= XFS_ILOG_CORE;
3018                         if (cur) {
3019                                 if ((error = xfs_bmbt_update(cur,
3020                                                 got.br_startoff,
3021                                                 got.br_startblock, temp,
3022                                                 got.br_state)))
3023                                         goto done;
3024                                 if ((error = xfs_bmbt_increment(cur, 0, &i)))
3025                                         goto done;
3026                                 cur->bc_rec.b = new;
3027                                 error = xfs_bmbt_insert(cur, &i);
3028                                 if (error && error != ENOSPC)
3029                                         goto done;
3030                                 /*
3031                                  * If get no-space back from btree insert,
3032                                  * it tried a split, and we have a zero
3033                                  * block reservation.
3034                                  * Fix up our state and return the error.
3035                                  */
3036                                 if (error == ENOSPC) {
3037                                         /*
3038                                          * Reset the cursor, don't trust
3039                                          * it after any insert operation.
3040                                          */
3041                                         if ((error = xfs_bmbt_lookup_eq(cur,
3042                                                         got.br_startoff,
3043                                                         got.br_startblock,
3044                                                         temp, &i)))
3045                                                 goto done;
3046                                         ASSERT(i == 1);
3047                                         /*
3048                                          * Update the btree record back
3049                                          * to the original value.
3050                                          */
3051                                         if ((error = xfs_bmbt_update(cur,
3052                                                         got.br_startoff,
3053                                                         got.br_startblock,
3054                                                         got.br_blockcount,
3055                                                         got.br_state)))
3056                                                 goto done;
3057                                         /*
3058                                          * Reset the extent record back
3059                                          * to the original value.
3060                                          */
3061                                         xfs_bmbt_set_blockcount(ep,
3062                                                 got.br_blockcount);
3063                                         flags = 0;
3064                                         error = XFS_ERROR(ENOSPC);
3065                                         goto done;
3066                                 }
3067                                 ASSERT(i == 1);
3068                         } else
3069                                 flags |= XFS_ILOG_FEXT(whichfork);
3070                         XFS_IFORK_NEXT_SET(ip, whichfork,
3071                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
3072                 } else {
3073                         ASSERT(whichfork == XFS_DATA_FORK);
3074                         temp = xfs_bmap_worst_indlen(ip, temp);
3075                         xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3076                         temp2 = xfs_bmap_worst_indlen(ip, temp2);
3077                         new.br_startblock = NULLSTARTBLOCK((int)temp2);
3078                         da_new = temp + temp2;
3079                         while (da_new > da_old) {
3080                                 if (temp) {
3081                                         temp--;
3082                                         da_new--;
3083                                         xfs_bmbt_set_startblock(ep,
3084                                                 NULLSTARTBLOCK((int)temp));
3085                                 }
3086                                 if (da_new == da_old)
3087                                         break;
3088                                 if (temp2) {
3089                                         temp2--;
3090                                         da_new--;
3091                                         new.br_startblock =
3092                                                 NULLSTARTBLOCK((int)temp2);
3093                                 }
3094                         }
3095                 }
3096                 xfs_bmap_trace_post_update(fname, "0", ip, idx, whichfork);
3097                 xfs_bmap_trace_insert(fname, "0", ip, idx + 1, 1, &new, NULL,
3098                         whichfork);
3099                 xfs_bmap_insert_exlist(ip, idx + 1, 1, &new, whichfork);
3100                 ifp->if_lastex = idx + 1;
3101                 break;
3102         }
3103         /*
3104          * If we need to, add to list of extents to delete.
3105          */
3106         if (do_fx)
3107                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
3108                         mp);
3109         /*
3110          * Adjust inode # blocks in the file.
3111          */
3112         if (nblks)
3113                 ip->i_d.di_nblocks -= nblks;
3114         /*
3115          * Adjust quota data.
3116          */
3117         if (qfield)
3118                 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, qfield, (long)-nblks);
3119
3120         /*
3121          * Account for change in delayed indirect blocks.
3122          * Nothing to do for disk quota accounting here.
3123          */
3124         ASSERT(da_old >= da_new);
3125         if (da_old > da_new)
3126                 xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS, (int)(da_old - da_new),
3127                         rsvd);
3128 done:
3129         *logflagsp = flags;
3130         return error;
3131 }
3132
3133 /*
3134  * Remove the entry "free" from the free item list.  Prev points to the
3135  * previous entry, unless "free" is the head of the list.
3136  */
3137 STATIC void
3138 xfs_bmap_del_free(
3139         xfs_bmap_free_t         *flist, /* free item list header */
3140         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
3141         xfs_bmap_free_item_t    *free)  /* list item to be freed */
3142 {
3143         if (prev)
3144                 prev->xbfi_next = free->xbfi_next;
3145         else
3146                 flist->xbf_first = free->xbfi_next;
3147         flist->xbf_count--;
3148         kmem_zone_free(xfs_bmap_free_item_zone, free);
3149 }
3150
3151 /*
3152  * Remove count entries from the extents array for inode "ip", starting
3153  * at index "idx".  Copies the remaining items down over the deleted ones,
3154  * and gives back the excess memory.
3155  */
3156 STATIC void
3157 xfs_bmap_delete_exlist(
3158         xfs_inode_t     *ip,            /* incore inode pointer */
3159         xfs_extnum_t    idx,            /* starting delete index */
3160         xfs_extnum_t    count,          /* count of items to delete */
3161         int             whichfork)      /* data or attr fork */
3162 {
3163         xfs_bmbt_rec_t  *base;          /* base of extent list */
3164         xfs_ifork_t     *ifp;           /* inode fork pointer */
3165         xfs_extnum_t    nextents;       /* number of extents in list after */
3166
3167         ifp = XFS_IFORK_PTR(ip, whichfork);
3168         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3169         base = ifp->if_u1.if_extents;
3170         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - count;
3171         memmove(&base[idx], &base[idx + count],
3172                 (nextents - idx) * sizeof(*base));
3173         xfs_iext_realloc(ip, -count, whichfork);
3174 }
3175
3176 /*
3177  * Convert an extents-format file into a btree-format file.
3178  * The new file will have a root block (in the inode) and a single child block.
3179  */
3180 STATIC int                                      /* error */
3181 xfs_bmap_extents_to_btree(
3182         xfs_trans_t             *tp,            /* transaction pointer */
3183         xfs_inode_t             *ip,            /* incore inode pointer */
3184         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
3185         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
3186         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
3187         int                     wasdel,         /* converting a delayed alloc */
3188         int                     *logflagsp,     /* inode logging flags */
3189         int                     whichfork)      /* data or attr fork */
3190 {
3191         xfs_bmbt_block_t        *ablock;        /* allocated (child) bt block */
3192         xfs_buf_t               *abp;           /* buffer for ablock */
3193         xfs_alloc_arg_t         args;           /* allocation arguments */
3194         xfs_bmbt_rec_t          *arp;           /* child record pointer */
3195         xfs_bmbt_block_t        *block;         /* btree root block */
3196         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
3197         xfs_bmbt_rec_t          *ep;            /* extent list pointer */
3198         int                     error;          /* error return value */
3199         xfs_extnum_t            i, cnt;         /* extent list index */
3200         xfs_ifork_t             *ifp;           /* inode fork pointer */
3201         xfs_bmbt_key_t          *kp;            /* root block key pointer */
3202         xfs_mount_t             *mp;            /* mount structure */
3203         xfs_extnum_t            nextents;       /* extent list size */
3204         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
3205
3206         ifp = XFS_IFORK_PTR(ip, whichfork);
3207         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3208         ASSERT(ifp->if_ext_max ==
3209                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
3210         /*
3211          * Make space in the inode incore.
3212          */
3213         xfs_iroot_realloc(ip, 1, whichfork);
3214         ifp->if_flags |= XFS_IFBROOT;
3215         /*
3216          * Fill in the root.
3217          */
3218         block = ifp->if_broot;
3219         INT_SET(block->bb_magic, ARCH_CONVERT, XFS_BMAP_MAGIC);
3220         INT_SET(block->bb_level, ARCH_CONVERT, 1);
3221         INT_SET(block->bb_numrecs, ARCH_CONVERT, 1);
3222         INT_SET(block->bb_leftsib, ARCH_CONVERT, NULLDFSBNO);
3223         INT_SET(block->bb_rightsib, ARCH_CONVERT, NULLDFSBNO);
3224         /*
3225          * Need a cursor.  Can't allocate until bb_level is filled in.
3226          */
3227         mp = ip->i_mount;
3228         cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
3229                 whichfork);
3230         cur->bc_private.b.firstblock = *firstblock;
3231         cur->bc_private.b.flist = flist;
3232         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3233         /*
3234          * Convert to a btree with two levels, one record in root.
3235          */
3236         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3237         args.tp = tp;
3238         args.mp = mp;
3239         if (*firstblock == NULLFSBLOCK) {
3240                 args.type = XFS_ALLOCTYPE_START_BNO;
3241                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3242         } else if (flist->xbf_low) {
3243                 args.type = XFS_ALLOCTYPE_START_BNO;
3244                 args.fsbno = *firstblock;
3245         } else {
3246                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3247                 args.fsbno = *firstblock;
3248         }
3249         args.minlen = args.maxlen = args.prod = 1;
3250         args.total = args.minleft = args.alignment = args.mod = args.isfl =
3251                 args.minalignslop = 0;
3252         args.wasdel = wasdel;
3253         *logflagsp = 0;
3254         if ((error = xfs_alloc_vextent(&args))) {
3255                 xfs_iroot_realloc(ip, -1, whichfork);
3256                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3257                 return error;
3258         }
3259         /*
3260          * Allocation can't fail, the space was reserved.
3261          */
3262         ASSERT(args.fsbno != NULLFSBLOCK);
3263         ASSERT(*firstblock == NULLFSBLOCK ||
3264                args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3265                (flist->xbf_low &&
3266                 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3267         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3268         cur->bc_private.b.allocated++;
3269         ip->i_d.di_nblocks++;
3270         XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3271         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3272         /*
3273          * Fill in the child block.
3274          */
3275         ablock = XFS_BUF_TO_BMBT_BLOCK(abp);
3276         INT_SET(ablock->bb_magic, ARCH_CONVERT, XFS_BMAP_MAGIC);
3277         ablock->bb_level = 0;
3278         INT_SET(ablock->bb_leftsib, ARCH_CONVERT, NULLDFSBNO);
3279         INT_SET(ablock->bb_rightsib, ARCH_CONVERT, NULLDFSBNO);
3280         arp = XFS_BMAP_REC_IADDR(ablock, 1, cur);
3281         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3282         for (ep = ifp->if_u1.if_extents, cnt = i = 0; i < nextents; i++, ep++) {
3283                 if (!ISNULLSTARTBLOCK(xfs_bmbt_get_startblock(ep))) {
3284                         arp->l0 = INT_GET(ep->l0, ARCH_CONVERT);
3285                         arp->l1 = INT_GET(ep->l1, ARCH_CONVERT);
3286                         arp++; cnt++;
3287                 }
3288         }
3289         INT_SET(ablock->bb_numrecs, ARCH_CONVERT, cnt);
3290         ASSERT(INT_GET(ablock->bb_numrecs, ARCH_CONVERT) == XFS_IFORK_NEXTENTS(ip, whichfork));
3291         /*
3292          * Fill in the root key and pointer.
3293          */
3294         kp = XFS_BMAP_KEY_IADDR(block, 1, cur);
3295         arp = XFS_BMAP_REC_IADDR(ablock, 1, cur);
3296         INT_SET(kp->br_startoff, ARCH_CONVERT, xfs_bmbt_disk_get_startoff(arp));
3297         pp = XFS_BMAP_PTR_IADDR(block, 1, cur);
3298         INT_SET(*pp, ARCH_CONVERT, args.fsbno);
3299         /*
3300          * Do all this logging at the end so that
3301          * the root is at the right level.
3302          */
3303         xfs_bmbt_log_block(cur, abp, XFS_BB_ALL_BITS);
3304         xfs_bmbt_log_recs(cur, abp, 1, INT_GET(ablock->bb_numrecs, ARCH_CONVERT));
3305         ASSERT(*curp == NULL);
3306         *curp = cur;
3307         *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FBROOT(whichfork);
3308         return 0;
3309 }
3310
3311 /*
3312  * Insert new item(s) in the extent list for inode "ip".
3313  * Count new items are inserted at offset idx.
3314  */
3315 STATIC void
3316 xfs_bmap_insert_exlist(
3317         xfs_inode_t     *ip,            /* incore inode pointer */
3318         xfs_extnum_t    idx,            /* starting index of new items */
3319         xfs_extnum_t    count,          /* number of inserted items */
3320         xfs_bmbt_irec_t *new,           /* items to insert */
3321         int             whichfork)      /* data or attr fork */
3322 {
3323         xfs_bmbt_rec_t  *base;          /* extent list base */
3324         xfs_ifork_t     *ifp;           /* inode fork pointer */
3325         xfs_extnum_t    nextents;       /* extent list size */
3326         xfs_extnum_t    to;             /* extent list index */
3327
3328         ifp = XFS_IFORK_PTR(ip, whichfork);
3329         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3330         xfs_iext_realloc(ip, count, whichfork);
3331         base = ifp->if_u1.if_extents;
3332         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3333         memmove(&base[idx + count], &base[idx],
3334                 (nextents - (idx + count)) * sizeof(*base));
3335         for (to = idx; to < idx + count; to++, new++)
3336                 xfs_bmbt_set_all(&base[to], new);
3337 }
3338
3339 /*
3340  * Convert a local file to an extents file.
3341  * This code is out of bounds for data forks of regular files,
3342  * since the file data needs to get logged so things will stay consistent.
3343  * (The bmap-level manipulations are ok, though).
3344  */
3345 STATIC int                              /* error */
3346 xfs_bmap_local_to_extents(
3347         xfs_trans_t     *tp,            /* transaction pointer */
3348         xfs_inode_t     *ip,            /* incore inode pointer */
3349         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
3350         xfs_extlen_t    total,          /* total blocks needed by transaction */
3351         int             *logflagsp,     /* inode logging flags */
3352         int             whichfork)      /* data or attr fork */
3353 {
3354         int             error;          /* error return value */
3355         int             flags;          /* logging flags returned */
3356 #ifdef XFS_BMAP_TRACE
3357         static char     fname[] = "xfs_bmap_local_to_extents";
3358 #endif
3359         xfs_ifork_t     *ifp;           /* inode fork pointer */
3360
3361         /*
3362          * We don't want to deal with the case of keeping inode data inline yet.
3363          * So sending the data fork of a regular inode is invalid.
3364          */
3365         ASSERT(!((ip->i_d.di_mode & S_IFMT) == S_IFREG &&
3366                  whichfork == XFS_DATA_FORK));
3367         ifp = XFS_IFORK_PTR(ip, whichfork);
3368         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3369         flags = 0;
3370         error = 0;
3371         if (ifp->if_bytes) {
3372                 xfs_alloc_arg_t args;   /* allocation arguments */
3373                 xfs_buf_t       *bp;    /* buffer for extent list block */
3374                 xfs_bmbt_rec_t  *ep;    /* extent list pointer */
3375
3376                 args.tp = tp;
3377                 args.mp = ip->i_mount;
3378                 ASSERT(ifp->if_flags & XFS_IFINLINE);
3379                 /*
3380                  * Allocate a block.  We know we need only one, since the
3381                  * file currently fits in an inode.
3382                  */
3383                 if (*firstblock == NULLFSBLOCK) {
3384                         args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3385                         args.type = XFS_ALLOCTYPE_START_BNO;
3386                 } else {
3387                         args.fsbno = *firstblock;
3388                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
3389                 }
3390                 args.total = total;
3391                 args.mod = args.minleft = args.alignment = args.wasdel =
3392                         args.isfl = args.minalignslop = 0;
3393                 args.minlen = args.maxlen = args.prod = 1;
3394                 if ((error = xfs_alloc_vextent(&args)))
3395                         goto done;
3396                 /*
3397                  * Can't fail, the space was reserved.
3398                  */
3399                 ASSERT(args.fsbno != NULLFSBLOCK);
3400                 ASSERT(args.len == 1);
3401                 *firstblock = args.fsbno;
3402                 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3403                 memcpy((char *)XFS_BUF_PTR(bp), ifp->if_u1.if_data,
3404                         ifp->if_bytes);
3405                 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3406                 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3407                 xfs_iext_realloc(ip, 1, whichfork);
3408                 ep = ifp->if_u1.if_extents;
3409                 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3410                 xfs_bmap_trace_post_update(fname, "new", ip, 0, whichfork);
3411                 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3412                 ip->i_d.di_nblocks = 1;
3413                 XFS_TRANS_MOD_DQUOT_BYINO(args.mp, tp, ip,
3414                         XFS_TRANS_DQ_BCOUNT, 1L);
3415                 flags |= XFS_ILOG_FEXT(whichfork);
3416         } else
3417                 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3418         ifp->if_flags &= ~XFS_IFINLINE;
3419         ifp->if_flags |= XFS_IFEXTENTS;
3420         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3421         flags |= XFS_ILOG_CORE;
3422 done:
3423         *logflagsp = flags;
3424         return error;
3425 }
3426
3427 xfs_bmbt_rec_t *                        /* pointer to found extent entry */
3428 xfs_bmap_do_search_extents(
3429         xfs_bmbt_rec_t  *base,          /* base of extent list */
3430         xfs_extnum_t    lastx,          /* last extent index used */
3431         xfs_extnum_t    nextents,       /* extent list size */
3432         xfs_fileoff_t   bno,            /* block number searched for */
3433         int             *eofp,          /* out: end of file found */
3434         xfs_extnum_t    *lastxp,        /* out: last extent index */
3435         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3436         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3437 {
3438         xfs_bmbt_rec_t  *ep;            /* extent list entry pointer */
3439         xfs_bmbt_irec_t got;            /* extent list entry, decoded */
3440         int             high;           /* high index of binary search */
3441         int             low;            /* low index of binary search */
3442
3443         /*
3444          * Initialize the extent entry structure to catch access to
3445          * uninitialized br_startblock field.
3446          */
3447         got.br_startoff = 0xffa5a5a5a5a5a5a5LL;
3448         got.br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3449         got.br_state = XFS_EXT_INVALID;
3450
3451 #if XFS_BIG_BLKNOS
3452         got.br_startblock = 0xffffa5a5a5a5a5a5LL;
3453 #else
3454         got.br_startblock = 0xffffa5a5;
3455 #endif
3456
3457         if (lastx != NULLEXTNUM && lastx < nextents)
3458                 ep = base + lastx;
3459         else
3460                 ep = NULL;
3461         prevp->br_startoff = NULLFILEOFF;
3462         if (ep && bno >= (got.br_startoff = xfs_bmbt_get_startoff(ep)) &&
3463             bno < got.br_startoff +
3464                   (got.br_blockcount = xfs_bmbt_get_blockcount(ep)))
3465                 *eofp = 0;
3466         else if (ep && lastx < nextents - 1 &&
3467                  bno >= (got.br_startoff = xfs_bmbt_get_startoff(ep + 1)) &&
3468                  bno < got.br_startoff +
3469                        (got.br_blockcount = xfs_bmbt_get_blockcount(ep + 1))) {
3470                 lastx++;
3471                 ep++;
3472                 *eofp = 0;
3473         } else if (nextents == 0)
3474                 *eofp = 1;
3475         else if (bno == 0 &&
3476                  (got.br_startoff = xfs_bmbt_get_startoff(base)) == 0) {
3477                 ep = base;
3478                 lastx = 0;
3479                 got.br_blockcount = xfs_bmbt_get_blockcount(ep);
3480                 *eofp = 0;
3481         } else {
3482                 /* binary search the extents array */
3483                 low = 0;
3484                 high = nextents - 1;
3485                 while (low <= high) {
3486                         XFS_STATS_INC(xs_cmp_exlist);
3487                         lastx = (low + high) >> 1;
3488                         ep = base + lastx;
3489                         got.br_startoff = xfs_bmbt_get_startoff(ep);
3490                         got.br_blockcount = xfs_bmbt_get_blockcount(ep);
3491                         if (bno < got.br_startoff)
3492                                 high = lastx - 1;
3493                         else if (bno >= got.br_startoff + got.br_blockcount)
3494                                 low = lastx + 1;
3495                         else {
3496                                 got.br_startblock = xfs_bmbt_get_startblock(ep);
3497                                 got.br_state = xfs_bmbt_get_state(ep);
3498                                 *eofp = 0;
3499                                 *lastxp = lastx;
3500                                 *gotp = got;
3501                                 return ep;
3502                         }
3503                 }
3504                 if (bno >= got.br_startoff + got.br_blockcount) {
3505                         lastx++;
3506                         if (lastx == nextents) {
3507                                 *eofp = 1;
3508                                 got.br_startblock = xfs_bmbt_get_startblock(ep);
3509                                 got.br_state = xfs_bmbt_get_state(ep);
3510                                 *prevp = got;
3511                                 ep = NULL;
3512                         } else {
3513                                 *eofp = 0;
3514                                 xfs_bmbt_get_all(ep, prevp);
3515                                 ep++;
3516                                 got.br_startoff = xfs_bmbt_get_startoff(ep);
3517                                 got.br_blockcount = xfs_bmbt_get_blockcount(ep);
3518                         }
3519                 } else {
3520                         *eofp = 0;
3521                         if (ep > base)
3522                                 xfs_bmbt_get_all(ep - 1, prevp);
3523                 }
3524         }
3525         if (ep) {
3526                 got.br_startblock = xfs_bmbt_get_startblock(ep);
3527                 got.br_state = xfs_bmbt_get_state(ep);
3528         }
3529         *lastxp = lastx;
3530         *gotp = got;
3531         return ep;
3532 }
3533
3534 /*
3535  * Search the extents list for the inode, for the extent containing bno.
3536  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
3537  * *eofp will be set, and *prevp will contain the last entry (null if none).
3538  * Else, *lastxp will be set to the index of the found
3539  * entry; *gotp will contain the entry.
3540  */
3541 STATIC xfs_bmbt_rec_t *                 /* pointer to found extent entry */
3542 xfs_bmap_search_extents(
3543         xfs_inode_t     *ip,            /* incore inode pointer */
3544         xfs_fileoff_t   bno,            /* block number searched for */
3545         int             whichfork,      /* data or attr fork */
3546         int             *eofp,          /* out: end of file found */
3547         xfs_extnum_t    *lastxp,        /* out: last extent index */
3548         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3549         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3550 {
3551         xfs_ifork_t     *ifp;           /* inode fork pointer */
3552         xfs_bmbt_rec_t  *base;          /* base of extent list */
3553         xfs_extnum_t    lastx;          /* last extent index used */
3554         xfs_extnum_t    nextents;       /* extent list size */
3555         xfs_bmbt_rec_t  *ep;            /* extent list entry pointer */
3556         int             rt;             /* realtime flag    */
3557
3558         XFS_STATS_INC(xs_look_exlist);
3559         ifp = XFS_IFORK_PTR(ip, whichfork);
3560         lastx = ifp->if_lastex;
3561         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3562         base = &ifp->if_u1.if_extents[0];
3563
3564         ep = xfs_bmap_do_search_extents(base, lastx, nextents, bno, eofp,
3565                                           lastxp, gotp, prevp);
3566         rt = ip->i_d.di_flags & XFS_DIFLAG_REALTIME;
3567         if(!rt && !gotp->br_startblock && (*lastxp != NULLEXTNUM)) {
3568                 cmn_err(CE_PANIC,"Access to block zero: fs: <%s> inode: %lld "
3569                         "start_block : %llx start_off : %llx blkcnt : %llx "
3570                         "extent-state : %x \n",
3571                         (ip->i_mount)->m_fsname,(long long)ip->i_ino,
3572                         gotp->br_startblock, gotp->br_startoff,
3573                         gotp->br_blockcount,gotp->br_state);
3574         }
3575         return ep;
3576 }
3577
3578
3579 #ifdef XFS_BMAP_TRACE
3580 ktrace_t        *xfs_bmap_trace_buf;
3581
3582 /*
3583  * Add a bmap trace buffer entry.  Base routine for the others.
3584  */
3585 STATIC void
3586 xfs_bmap_trace_addentry(
3587         int             opcode,         /* operation */
3588         char            *fname,         /* function name */
3589         char            *desc,          /* operation description */
3590         xfs_inode_t     *ip,            /* incore inode pointer */
3591         xfs_extnum_t    idx,            /* index of entry(ies) */
3592         xfs_extnum_t    cnt,            /* count of entries, 1 or 2 */
3593         xfs_bmbt_rec_t  *r1,            /* first record */
3594         xfs_bmbt_rec_t  *r2,            /* second record or null */
3595         int             whichfork)      /* data or attr fork */
3596 {
3597         xfs_bmbt_rec_t  tr2;
3598
3599         ASSERT(cnt == 1 || cnt == 2);
3600         ASSERT(r1 != NULL);
3601         if (cnt == 1) {
3602                 ASSERT(r2 == NULL);
3603                 r2 = &tr2;
3604                 memset(&tr2, 0, sizeof(tr2));
3605         } else
3606                 ASSERT(r2 != NULL);
3607         ktrace_enter(xfs_bmap_trace_buf,
3608                 (void *)(__psint_t)(opcode | (whichfork << 16)),
3609                 (void *)fname, (void *)desc, (void *)ip,
3610                 (void *)(__psint_t)idx,
3611                 (void *)(__psint_t)cnt,
3612                 (void *)(__psunsigned_t)(ip->i_ino >> 32),
3613                 (void *)(__psunsigned_t)(unsigned)ip->i_ino,
3614                 (void *)(__psunsigned_t)(r1->l0 >> 32),
3615                 (void *)(__psunsigned_t)(unsigned)(r1->l0),
3616                 (void *)(__psunsigned_t)(r1->l1 >> 32),
3617                 (void *)(__psunsigned_t)(unsigned)(r1->l1),
3618                 (void *)(__psunsigned_t)(r2->l0 >> 32),
3619                 (void *)(__psunsigned_t)(unsigned)(r2->l0),
3620                 (void *)(__psunsigned_t)(r2->l1 >> 32),
3621                 (void *)(__psunsigned_t)(unsigned)(r2->l1)
3622                 );
3623         ASSERT(ip->i_xtrace);
3624         ktrace_enter(ip->i_xtrace,
3625                 (void *)(__psint_t)(opcode | (whichfork << 16)),
3626                 (void *)fname, (void *)desc, (void *)ip,
3627                 (void *)(__psint_t)idx,
3628                 (void *)(__psint_t)cnt,
3629                 (void *)(__psunsigned_t)(ip->i_ino >> 32),
3630                 (void *)(__psunsigned_t)(unsigned)ip->i_ino,
3631                 (void *)(__psunsigned_t)(r1->l0 >> 32),
3632                 (void *)(__psunsigned_t)(unsigned)(r1->l0),
3633                 (void *)(__psunsigned_t)(r1->l1 >> 32),
3634                 (void *)(__psunsigned_t)(unsigned)(r1->l1),
3635                 (void *)(__psunsigned_t)(r2->l0 >> 32),
3636                 (void *)(__psunsigned_t)(unsigned)(r2->l0),
3637                 (void *)(__psunsigned_t)(r2->l1 >> 32),
3638                 (void *)(__psunsigned_t)(unsigned)(r2->l1)
3639                 );
3640 }
3641
3642 /*
3643  * Add bmap trace entry prior to a call to xfs_bmap_delete_exlist.
3644  */
3645 STATIC void
3646 xfs_bmap_trace_delete(
3647         char            *fname,         /* function name */
3648         char            *desc,          /* operation description */
3649         xfs_inode_t     *ip,            /* incore inode pointer */
3650         xfs_extnum_t    idx,            /* index of entry(entries) deleted */
3651         xfs_extnum_t    cnt,            /* count of entries deleted, 1 or 2 */
3652         int             whichfork)      /* data or attr fork */
3653 {
3654         xfs_ifork_t     *ifp;           /* inode fork pointer */
3655
3656         ifp = XFS_IFORK_PTR(ip, whichfork);
3657         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_DELETE, fname, desc, ip, idx,
3658                 cnt, &ifp->if_u1.if_extents[idx],
3659                 cnt == 2 ? &ifp->if_u1.if_extents[idx + 1] : NULL,
3660                 whichfork);
3661 }
3662
3663 /*
3664  * Add bmap trace entry prior to a call to xfs_bmap_insert_exlist, or
3665  * reading in the extents list from the disk (in the btree).
3666  */
3667 STATIC void
3668 xfs_bmap_trace_insert(
3669         char            *fname,         /* function name */
3670         char            *desc,          /* operation description */
3671         xfs_inode_t     *ip,            /* incore inode pointer */
3672         xfs_extnum_t    idx,            /* index of entry(entries) inserted */
3673         xfs_extnum_t    cnt,            /* count of entries inserted, 1 or 2 */
3674         xfs_bmbt_irec_t *r1,            /* inserted record 1 */
3675         xfs_bmbt_irec_t *r2,            /* inserted record 2 or null */
3676         int             whichfork)      /* data or attr fork */
3677 {
3678         xfs_bmbt_rec_t  tr1;            /* compressed record 1 */
3679         xfs_bmbt_rec_t  tr2;            /* compressed record 2 if needed */
3680
3681         xfs_bmbt_set_all(&tr1, r1);
3682         if (cnt == 2) {
3683                 ASSERT(r2 != NULL);
3684                 xfs_bmbt_set_all(&tr2, r2);
3685         } else {
3686                 ASSERT(cnt == 1);
3687                 ASSERT(r2 == NULL);
3688         }
3689         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_INSERT, fname, desc, ip, idx,
3690                 cnt, &tr1, cnt == 2 ? &tr2 : NULL, whichfork);
3691 }
3692
3693 /*
3694  * Add bmap trace entry after updating an extent list entry in place.
3695  */
3696 STATIC void
3697 xfs_bmap_trace_post_update(
3698         char            *fname,         /* function name */
3699         char            *desc,          /* operation description */
3700         xfs_inode_t     *ip,            /* incore inode pointer */
3701         xfs_extnum_t    idx,            /* index of entry updated */
3702         int             whichfork)      /* data or attr fork */
3703 {
3704         xfs_ifork_t     *ifp;           /* inode fork pointer */
3705
3706         ifp = XFS_IFORK_PTR(ip, whichfork);
3707         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_POST_UP, fname, desc, ip, idx,
3708                 1, &ifp->if_u1.if_extents[idx], NULL, whichfork);
3709 }
3710
3711 /*
3712  * Add bmap trace entry prior to updating an extent list entry in place.
3713  */
3714 STATIC void
3715 xfs_bmap_trace_pre_update(
3716         char            *fname,         /* function name */
3717         char            *desc,          /* operation description */
3718         xfs_inode_t     *ip,            /* incore inode pointer */
3719         xfs_extnum_t    idx,            /* index of entry to be updated */
3720         int             whichfork)      /* data or attr fork */
3721 {
3722         xfs_ifork_t     *ifp;           /* inode fork pointer */
3723
3724         ifp = XFS_IFORK_PTR(ip, whichfork);
3725         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_PRE_UP, fname, desc, ip, idx, 1,
3726                 &ifp->if_u1.if_extents[idx], NULL, whichfork);
3727 }
3728 #endif  /* XFS_BMAP_TRACE */
3729
3730 /*
3731  * Compute the worst-case number of indirect blocks that will be used
3732  * for ip's delayed extent of length "len".
3733  */
3734 STATIC xfs_filblks_t
3735 xfs_bmap_worst_indlen(
3736         xfs_inode_t     *ip,            /* incore inode pointer */
3737         xfs_filblks_t   len)            /* delayed extent length */
3738 {
3739         int             level;          /* btree level number */
3740         int             maxrecs;        /* maximum record count at this level */
3741         xfs_mount_t     *mp;            /* mount structure */
3742         xfs_filblks_t   rval;           /* return value */
3743
3744         mp = ip->i_mount;
3745         maxrecs = mp->m_bmap_dmxr[0];
3746         for (level = 0, rval = 0;
3747              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3748              level++) {
3749                 len += maxrecs - 1;
3750                 do_div(len, maxrecs);
3751                 rval += len;
3752                 if (len == 1)
3753                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3754                                 level - 1;
3755                 if (level == 0)
3756                         maxrecs = mp->m_bmap_dmxr[1];
3757         }
3758         return rval;
3759 }
3760
3761 #if defined(XFS_RW_TRACE)
3762 STATIC void
3763 xfs_bunmap_trace(
3764         xfs_inode_t             *ip,
3765         xfs_fileoff_t           bno,
3766         xfs_filblks_t           len,
3767         int                     flags,
3768         inst_t                  *ra)
3769 {
3770         if (ip->i_rwtrace == NULL)
3771                 return;
3772         ktrace_enter(ip->i_rwtrace,
3773                 (void *)(__psint_t)XFS_BUNMAPI,
3774                 (void *)ip,
3775                 (void *)(__psint_t)((ip->i_d.di_size >> 32) & 0xffffffff),
3776                 (void *)(__psint_t)(ip->i_d.di_size & 0xffffffff),
3777                 (void *)(__psint_t)(((xfs_dfiloff_t)bno >> 32) & 0xffffffff),
3778                 (void *)(__psint_t)((xfs_dfiloff_t)bno & 0xffffffff),
3779                 (void *)(__psint_t)len,
3780                 (void *)(__psint_t)flags,
3781                 (void *)(unsigned long)current_cpu(),
3782                 (void *)ra,
3783                 (void *)0,
3784                 (void *)0,
3785                 (void *)0,
3786                 (void *)0,
3787                 (void *)0,
3788                 (void *)0);
3789 }
3790 #endif
3791
3792 /*
3793  * Convert inode from non-attributed to attributed.
3794  * Must not be in a transaction, ip must not be locked.
3795  */
3796 int                                             /* error code */
3797 xfs_bmap_add_attrfork(
3798         xfs_inode_t             *ip,            /* incore inode pointer */
3799         int                     rsvd)           /* OK to allocated reserved blocks in trans */
3800 {
3801         int                     blks;           /* space reservation */
3802         int                     committed;      /* xaction was committed */
3803         int                     error;          /* error return value */
3804         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
3805         xfs_bmap_free_t         flist;          /* freed extent list */
3806         int                     logflags;       /* logging flags */
3807         xfs_mount_t             *mp;            /* mount structure */
3808         unsigned long           s;              /* spinlock spl value */
3809         xfs_trans_t             *tp;            /* transaction pointer */
3810
3811         ASSERT(ip->i_df.if_ext_max ==
3812                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3813         if (XFS_IFORK_Q(ip))
3814                 return 0;
3815         mp = ip->i_mount;
3816         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3817         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3818         blks = XFS_ADDAFORK_SPACE_RES(mp);
3819         if (rsvd)
3820                 tp->t_flags |= XFS_TRANS_RESERVE;
3821         if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3822                         XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3823                 goto error0;
3824         xfs_ilock(ip, XFS_ILOCK_EXCL);
3825         error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip, blks, 0, rsvd ?
3826                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3827                         XFS_QMOPT_RES_REGBLKS);
3828         if (error) {
3829                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3830                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3831                 return error;
3832         }
3833         if (XFS_IFORK_Q(ip))
3834                 goto error1;
3835         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3836                 /*
3837                  * For inodes coming from pre-6.2 filesystems.
3838                  */
3839                 ASSERT(ip->i_d.di_aformat == 0);
3840                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
3841         }
3842         ASSERT(ip->i_d.di_anextents == 0);
3843         VN_HOLD(XFS_ITOV(ip));
3844         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3845         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3846         switch (ip->i_d.di_format) {
3847         case XFS_DINODE_FMT_DEV:
3848                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
3849                 break;
3850         case XFS_DINODE_FMT_UUID:
3851                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
3852                 break;
3853         case XFS_DINODE_FMT_LOCAL:
3854         case XFS_DINODE_FMT_EXTENTS:
3855         case XFS_DINODE_FMT_BTREE:
3856                 ip->i_d.di_forkoff = mp->m_attroffset >> 3;
3857                 break;
3858         default:
3859                 ASSERT(0);
3860                 error = XFS_ERROR(EINVAL);
3861                 goto error1;
3862         }
3863         ip->i_df.if_ext_max =
3864                 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3865         ASSERT(ip->i_afp == NULL);
3866         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
3867         ip->i_afp->if_ext_max =
3868                 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3869         ip->i_afp->if_flags = XFS_IFEXTENTS;
3870         logflags = 0;
3871         XFS_BMAP_INIT(&flist, &firstblock);
3872         switch (ip->i_d.di_format) {
3873         case XFS_DINODE_FMT_LOCAL:
3874                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
3875                         &logflags);
3876                 break;
3877         case XFS_DINODE_FMT_EXTENTS:
3878                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
3879                         &flist, &logflags);
3880                 break;
3881         case XFS_DINODE_FMT_BTREE:
3882                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
3883                         &logflags);
3884                 break;
3885         default:
3886                 error = 0;
3887                 break;
3888         }
3889         if (logflags)
3890                 xfs_trans_log_inode(tp, ip, logflags);
3891         if (error)
3892                 goto error2;
3893         if (!XFS_SB_VERSION_HASATTR(&mp->m_sb)) {
3894                 s = XFS_SB_LOCK(mp);
3895                 if (!XFS_SB_VERSION_HASATTR(&mp->m_sb)) {
3896                         XFS_SB_VERSION_ADDATTR(&mp->m_sb);
3897                         XFS_SB_UNLOCK(mp, s);
3898                         xfs_mod_sb(tp, XFS_SB_VERSIONNUM);
3899                 } else
3900                         XFS_SB_UNLOCK(mp, s);
3901         }
3902         if ((error = xfs_bmap_finish(&tp, &flist, firstblock, &committed)))
3903                 goto error2;
3904         error = xfs_trans_commit(tp, XFS_TRANS_PERM_LOG_RES, NULL);
3905         ASSERT(ip->i_df.if_ext_max ==
3906                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3907         return error;
3908 error2:
3909         xfs_bmap_cancel(&flist);
3910 error1:
3911         ASSERT(ismrlocked(&ip->i_lock,MR_UPDATE));
3912         xfs_iunlock(ip, XFS_ILOCK_EXCL);
3913 error0:
3914         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
3915         ASSERT(ip->i_df.if_ext_max ==
3916                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3917         return error;
3918 }
3919
3920 /*
3921  * Add the extent to the list of extents to be free at transaction end.
3922  * The list is maintained sorted (by block number).
3923  */
3924 /* ARGSUSED */
3925 void
3926 xfs_bmap_add_free(
3927         xfs_fsblock_t           bno,            /* fs block number of extent */
3928         xfs_filblks_t           len,            /* length of extent */
3929         xfs_bmap_free_t         *flist,         /* list of extents */
3930         xfs_mount_t             *mp)            /* mount point structure */
3931 {
3932         xfs_bmap_free_item_t    *cur;           /* current (next) element */
3933         xfs_bmap_free_item_t    *new;           /* new element */
3934         xfs_bmap_free_item_t    *prev;          /* previous element */
3935 #ifdef DEBUG
3936         xfs_agnumber_t          agno;
3937         xfs_agblock_t           agbno;
3938
3939         ASSERT(bno != NULLFSBLOCK);
3940         ASSERT(len > 0);
3941         ASSERT(len <= MAXEXTLEN);
3942         ASSERT(!ISNULLSTARTBLOCK(bno));
3943         agno = XFS_FSB_TO_AGNO(mp, bno);
3944         agbno = XFS_FSB_TO_AGBNO(mp, bno);
3945         ASSERT(agno < mp->m_sb.sb_agcount);
3946         ASSERT(agbno < mp->m_sb.sb_agblocks);
3947         ASSERT(len < mp->m_sb.sb_agblocks);
3948         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
3949 #endif
3950         ASSERT(xfs_bmap_free_item_zone != NULL);
3951         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
3952         new->xbfi_startblock = bno;
3953         new->xbfi_blockcount = (xfs_extlen_t)len;
3954         for (prev = NULL, cur = flist->xbf_first;
3955              cur != NULL;
3956              prev = cur, cur = cur->xbfi_next) {
3957                 if (cur->xbfi_startblock >= bno)
3958                         break;
3959         }
3960         if (prev)
3961                 prev->xbfi_next = new;
3962         else
3963                 flist->xbf_first = new;
3964         new->xbfi_next = cur;
3965         flist->xbf_count++;
3966 }
3967
3968 /*
3969  * Compute and fill in the value of the maximum depth of a bmap btree
3970  * in this filesystem.  Done once, during mount.
3971  */
3972 void
3973 xfs_bmap_compute_maxlevels(
3974         xfs_mount_t     *mp,            /* file system mount structure */
3975         int             whichfork)      /* data or attr fork */
3976 {
3977         int             level;          /* btree level */
3978         uint            maxblocks;      /* max blocks at this level */
3979         uint            maxleafents;    /* max leaf entries possible */
3980         int             maxrootrecs;    /* max records in root block */
3981         int             minleafrecs;    /* min records in leaf block */
3982         int             minnoderecs;    /* min records in node block */
3983         int             sz;             /* root block size */
3984
3985         /*
3986          * The maximum number of extents in a file, hence the maximum
3987          * number of leaf entries, is controlled by the type of di_nextents
3988          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
3989          * (a signed 16-bit number, xfs_aextnum_t).
3990          */
3991         maxleafents = (whichfork == XFS_DATA_FORK) ? MAXEXTNUM : MAXAEXTNUM;
3992         minleafrecs = mp->m_bmap_dmnr[0];
3993         minnoderecs = mp->m_bmap_dmnr[1];
3994         sz = (whichfork == XFS_DATA_FORK) ?
3995                 mp->m_attroffset :
3996                 mp->m_sb.sb_inodesize - mp->m_attroffset;
3997         maxrootrecs = (int)XFS_BTREE_BLOCK_MAXRECS(sz, xfs_bmdr, 0);
3998         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
3999         for (level = 1; maxblocks > 1; level++) {
4000                 if (maxblocks <= maxrootrecs)
4001                         maxblocks = 1;
4002                 else
4003                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
4004         }
4005         mp->m_bm_maxlevels[whichfork] = level;
4006 }
4007
4008 /*
4009  * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
4010  * caller.  Frees all the extents that need freeing, which must be done
4011  * last due to locking considerations.  We never free any extents in
4012  * the first transaction.  This is to allow the caller to make the first
4013  * transaction a synchronous one so that the pointers to the data being
4014  * broken in this transaction will be permanent before the data is actually
4015  * freed.  This is necessary to prevent blocks from being reallocated
4016  * and written to before the free and reallocation are actually permanent.
4017  * We do not just make the first transaction synchronous here, because
4018  * there are more efficient ways to gain the same protection in some cases
4019  * (see the file truncation code).
4020  *
4021  * Return 1 if the given transaction was committed and a new one
4022  * started, and 0 otherwise in the committed parameter.
4023  */
4024 /*ARGSUSED*/
4025 int                                             /* error */
4026 xfs_bmap_finish(
4027         xfs_trans_t             **tp,           /* transaction pointer addr */
4028         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
4029         xfs_fsblock_t           firstblock,     /* controlled ag for allocs */
4030         int                     *committed)     /* xact committed or not */
4031 {
4032         xfs_efd_log_item_t      *efd;           /* extent free data */
4033         xfs_efi_log_item_t      *efi;           /* extent free intention */
4034         int                     error;          /* error return value */
4035         xfs_bmap_free_item_t    *free;          /* free extent list item */
4036         unsigned int            logres;         /* new log reservation */
4037         unsigned int            logcount;       /* new log count */
4038         xfs_mount_t             *mp;            /* filesystem mount structure */
4039         xfs_bmap_free_item_t    *next;          /* next item on free list */
4040         xfs_trans_t             *ntp;           /* new transaction pointer */
4041
4042         ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
4043         if (flist->xbf_count == 0) {
4044                 *committed = 0;
4045                 return 0;
4046         }
4047         ntp = *tp;
4048         efi = xfs_trans_get_efi(ntp, flist->xbf_count);
4049         for (free = flist->xbf_first; free; free = free->xbfi_next)
4050                 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
4051                         free->xbfi_blockcount);
4052         logres = ntp->t_log_res;
4053         logcount = ntp->t_log_count;
4054         ntp = xfs_trans_dup(*tp);
4055         error = xfs_trans_commit(*tp, 0, NULL);
4056         *tp = ntp;
4057         *committed = 1;
4058         /*
4059          * We have a new transaction, so we should return committed=1,
4060          * even though we're returning an error.
4061          */
4062         if (error) {
4063                 return error;
4064         }
4065         if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
4066                         logcount)))
4067                 return error;
4068         efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
4069         for (free = flist->xbf_first; free != NULL; free = next) {
4070                 next = free->xbfi_next;
4071                 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
4072                                 free->xbfi_blockcount))) {
4073                         /*
4074                          * The bmap free list will be cleaned up at a
4075                          * higher level.  The EFI will be canceled when
4076                          * this transaction is aborted.
4077                          * Need to force shutdown here to make sure it
4078                          * happens, since this transaction may not be
4079                          * dirty yet.
4080                          */
4081                         mp = ntp->t_mountp;
4082                         if (!XFS_FORCED_SHUTDOWN(mp))
4083                                 xfs_force_shutdown(mp,
4084                                                    (error == EFSCORRUPTED) ?
4085                                                    XFS_CORRUPT_INCORE :
4086                                                    XFS_METADATA_IO_ERROR);
4087                         return error;
4088                 }
4089                 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
4090                         free->xbfi_blockcount);
4091                 xfs_bmap_del_free(flist, NULL, free);
4092         }
4093         return 0;
4094 }
4095
4096 /*
4097  * Free up any items left in the list.
4098  */
4099 void
4100 xfs_bmap_cancel(
4101         xfs_bmap_free_t         *flist) /* list of bmap_free_items */
4102 {
4103         xfs_bmap_free_item_t    *free;  /* free list item */
4104         xfs_bmap_free_item_t    *next;
4105
4106         if (flist->xbf_count == 0)
4107                 return;
4108         ASSERT(flist->xbf_first != NULL);
4109         for (free = flist->xbf_first; free; free = next) {
4110                 next = free->xbfi_next;
4111                 xfs_bmap_del_free(flist, NULL, free);
4112         }
4113         ASSERT(flist->xbf_count == 0);
4114 }
4115
4116 /*
4117  * Returns the file-relative block number of the first unused block(s)
4118  * in the file with at least "len" logically contiguous blocks free.
4119  * This is the lowest-address hole if the file has holes, else the first block
4120  * past the end of file.
4121  * Return 0 if the file is currently local (in-inode).
4122  */
4123 int                                             /* error */
4124 xfs_bmap_first_unused(
4125         xfs_trans_t     *tp,                    /* transaction pointer */
4126         xfs_inode_t     *ip,                    /* incore inode */
4127         xfs_extlen_t    len,                    /* size of hole to find */
4128         xfs_fileoff_t   *first_unused,          /* unused block */
4129         int             whichfork)              /* data or attr fork */
4130 {
4131         xfs_bmbt_rec_t  *base;                  /* base of extent array */
4132         xfs_bmbt_rec_t  *ep;                    /* pointer to an extent entry */
4133         int             error;                  /* error return value */
4134         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4135         xfs_fileoff_t   lastaddr;               /* last block number seen */
4136         xfs_fileoff_t   lowest;                 /* lowest useful block */
4137         xfs_fileoff_t   max;                    /* starting useful block */
4138         xfs_fileoff_t   off;                    /* offset for this block */
4139         xfs_extnum_t    nextents;               /* number of extent entries */
4140
4141         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
4142                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
4143                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
4144         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4145                 *first_unused = 0;
4146                 return 0;
4147         }
4148         ifp = XFS_IFORK_PTR(ip, whichfork);
4149         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4150             (error = xfs_iread_extents(tp, ip, whichfork)))
4151                 return error;
4152         lowest = *first_unused;
4153         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4154         base = &ifp->if_u1.if_extents[0];
4155         for (lastaddr = 0, max = lowest, ep = base;
4156              ep < &base[nextents];
4157              ep++) {
4158                 off = xfs_bmbt_get_startoff(ep);
4159                 /*
4160                  * See if the hole before this extent will work.
4161                  */
4162                 if (off >= lowest + len && off - max >= len) {
4163                         *first_unused = max;
4164                         return 0;
4165                 }
4166                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
4167                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
4168         }
4169         *first_unused = max;
4170         return 0;
4171 }
4172
4173 /*
4174  * Returns the file-relative block number of the last block + 1 before
4175  * last_block (input value) in the file.
4176  * This is not based on i_size, it is based on the extent list.
4177  * Returns 0 for local files, as they do not have an extent list.
4178  */
4179 int                                             /* error */
4180 xfs_bmap_last_before(
4181         xfs_trans_t     *tp,                    /* transaction pointer */
4182         xfs_inode_t     *ip,                    /* incore inode */
4183         xfs_fileoff_t   *last_block,            /* last block */
4184         int             whichfork)              /* data or attr fork */
4185 {
4186         xfs_fileoff_t   bno;                    /* input file offset */
4187         int             eof;                    /* hit end of file */
4188         xfs_bmbt_rec_t  *ep;                    /* pointer to last extent */
4189         int             error;                  /* error return value */
4190         xfs_bmbt_irec_t got;                    /* current extent value */
4191         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4192         xfs_extnum_t    lastx;                  /* last extent used */
4193         xfs_bmbt_irec_t prev;                   /* previous extent value */
4194
4195         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4196             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4197             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4198                return XFS_ERROR(EIO);
4199         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4200                 *last_block = 0;
4201                 return 0;
4202         }
4203         ifp = XFS_IFORK_PTR(ip, whichfork);
4204         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4205             (error = xfs_iread_extents(tp, ip, whichfork)))
4206                 return error;
4207         bno = *last_block - 1;
4208         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4209                 &prev);
4210         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
4211                 if (prev.br_startoff == NULLFILEOFF)
4212                         *last_block = 0;
4213                 else
4214                         *last_block = prev.br_startoff + prev.br_blockcount;
4215         }
4216         /*
4217          * Otherwise *last_block is already the right answer.
4218          */
4219         return 0;
4220 }
4221
4222 /*
4223  * Returns the file-relative block number of the first block past eof in
4224  * the file.  This is not based on i_size, it is based on the extent list.
4225  * Returns 0 for local files, as they do not have an extent list.
4226  */
4227 int                                             /* error */
4228 xfs_bmap_last_offset(
4229         xfs_trans_t     *tp,                    /* transaction pointer */
4230         xfs_inode_t     *ip,                    /* incore inode */
4231         xfs_fileoff_t   *last_block,            /* last block */
4232         int             whichfork)              /* data or attr fork */
4233 {
4234         xfs_bmbt_rec_t  *base;                  /* base of extent array */
4235         xfs_bmbt_rec_t  *ep;                    /* pointer to last extent */
4236         int             error;                  /* error return value */
4237         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4238         xfs_extnum_t    nextents;               /* number of extent entries */
4239
4240         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4241             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4242             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4243                return XFS_ERROR(EIO);
4244         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4245                 *last_block = 0;
4246                 return 0;
4247         }
4248         ifp = XFS_IFORK_PTR(ip, whichfork);
4249         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4250             (error = xfs_iread_extents(tp, ip, whichfork)))
4251                 return error;
4252         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4253         if (!nextents) {
4254                 *last_block = 0;
4255                 return 0;
4256         }
4257         base = &ifp->if_u1.if_extents[0];
4258         ASSERT(base != NULL);
4259         ep = &base[nextents - 1];
4260         *last_block = xfs_bmbt_get_startoff(ep) + xfs_bmbt_get_blockcount(ep);
4261         return 0;
4262 }
4263
4264 /*
4265  * Returns whether the selected fork of the inode has exactly one
4266  * block or not.  For the data fork we check this matches di_size,
4267  * implying the file's range is 0..bsize-1.
4268  */
4269 int                                     /* 1=>1 block, 0=>otherwise */
4270 xfs_bmap_one_block(
4271         xfs_inode_t     *ip,            /* incore inode */
4272         int             whichfork)      /* data or attr fork */
4273 {
4274         xfs_bmbt_rec_t  *ep;            /* ptr to fork's extent */
4275         xfs_ifork_t     *ifp;           /* inode fork pointer */
4276         int             rval;           /* return value */
4277         xfs_bmbt_irec_t s;              /* internal version of extent */
4278
4279 #ifndef DEBUG
4280         if (whichfork == XFS_DATA_FORK)
4281                 return ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize;
4282 #endif  /* !DEBUG */
4283         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4284                 return 0;
4285         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4286                 return 0;
4287         ifp = XFS_IFORK_PTR(ip, whichfork);
4288         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4289         ep = ifp->if_u1.if_extents;
4290         xfs_bmbt_get_all(ep, &s);
4291         rval = s.br_startoff == 0 && s.br_blockcount == 1;
4292         if (rval && whichfork == XFS_DATA_FORK)
4293                 ASSERT(ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize);
4294         return rval;
4295 }
4296
4297 /*
4298  * Read in the extents to if_extents.
4299  * All inode fields are set up by caller, we just traverse the btree
4300  * and copy the records in. If the file system cannot contain unwritten
4301  * extents, the records are checked for no "state" flags.
4302  */
4303 int                                     /* error */
4304 xfs_bmap_read_extents(
4305         xfs_trans_t             *tp,    /* transaction pointer */
4306         xfs_inode_t             *ip,    /* incore inode */
4307         int                     whichfork) /* data or attr fork */
4308 {
4309         xfs_bmbt_block_t        *block; /* current btree block */
4310         xfs_fsblock_t           bno;    /* block # of "block" */
4311         xfs_buf_t               *bp;    /* buffer for "block" */
4312         int                     error;  /* error return value */
4313         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
4314 #ifdef XFS_BMAP_TRACE
4315         static char             fname[] = "xfs_bmap_read_extents";
4316 #endif
4317         xfs_extnum_t            i, j;   /* index into the extents list */
4318         xfs_ifork_t             *ifp;   /* fork structure */
4319         int                     level;  /* btree level, for checking */
4320         xfs_mount_t             *mp;    /* file system mount structure */
4321         xfs_bmbt_ptr_t          *pp;    /* pointer to block address */
4322         /* REFERENCED */
4323         xfs_extnum_t            room;   /* number of entries there's room for */
4324         xfs_bmbt_rec_t          *trp;   /* target record pointer */
4325
4326         bno = NULLFSBLOCK;
4327         mp = ip->i_mount;
4328         ifp = XFS_IFORK_PTR(ip, whichfork);
4329         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4330                                         XFS_EXTFMT_INODE(ip);
4331         block = ifp->if_broot;
4332         /*
4333          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4334          */
4335         ASSERT(INT_GET(block->bb_level, ARCH_CONVERT) > 0);
4336         level = INT_GET(block->bb_level, ARCH_CONVERT);
4337         pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
4338         ASSERT(INT_GET(*pp, ARCH_CONVERT) != NULLDFSBNO);
4339         ASSERT(XFS_FSB_TO_AGNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agcount);
4340         ASSERT(XFS_FSB_TO_AGBNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agblocks);
4341         bno = INT_GET(*pp, ARCH_CONVERT);
4342         /*
4343          * Go down the tree until leaf level is reached, following the first
4344          * pointer (leftmost) at each level.
4345          */
4346         while (level-- > 0) {
4347                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4348                                 XFS_BMAP_BTREE_REF)))
4349                         return error;
4350                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
4351                 XFS_WANT_CORRUPTED_GOTO(
4352                         XFS_BMAP_SANITY_CHECK(mp, block, level),
4353                         error0);
4354                 if (level == 0)
4355                         break;
4356                 pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt, block,
4357                         1, mp->m_bmap_dmxr[1]);
4358                 XFS_WANT_CORRUPTED_GOTO(
4359                         XFS_FSB_SANITY_CHECK(mp, INT_GET(*pp, ARCH_CONVERT)),
4360                         error0);
4361                 bno = INT_GET(*pp, ARCH_CONVERT);
4362                 xfs_trans_brelse(tp, bp);
4363         }
4364         /*
4365          * Here with bp and block set to the leftmost leaf node in the tree.
4366          */
4367         room = ifp->if_bytes / (uint)sizeof(*trp);
4368         trp = ifp->if_u1.if_extents;
4369         i = 0;
4370         /*
4371          * Loop over all leaf nodes.  Copy information to the extent list.
4372          */
4373         for (;;) {
4374                 xfs_bmbt_rec_t  *frp, *temp;
4375                 xfs_fsblock_t   nextbno;
4376                 xfs_extnum_t    num_recs;
4377
4378
4379                 num_recs = INT_GET(block->bb_numrecs, ARCH_CONVERT);
4380                 if (unlikely(i + num_recs > room)) {
4381                         ASSERT(i + num_recs <= room);
4382                         xfs_fs_cmn_err(CE_WARN, ip->i_mount,
4383                                 "corrupt dinode %Lu, (btree extents).  Unmount and run xfs_repair.",
4384                                 (unsigned long long) ip->i_ino);
4385                         XFS_ERROR_REPORT("xfs_bmap_read_extents(1)",
4386                                          XFS_ERRLEVEL_LOW,
4387                                         ip->i_mount);
4388                         goto error0;
4389                 }
4390                 XFS_WANT_CORRUPTED_GOTO(
4391                         XFS_BMAP_SANITY_CHECK(mp, block, 0),
4392                         error0);
4393                 /*
4394                  * Read-ahead the next leaf block, if any.
4395                  */
4396                 nextbno = INT_GET(block->bb_rightsib, ARCH_CONVERT);
4397                 if (nextbno != NULLFSBLOCK)
4398                         xfs_btree_reada_bufl(mp, nextbno, 1);
4399                 /*
4400                  * Copy records into the extent list.
4401                  */
4402                 frp = XFS_BTREE_REC_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt,
4403                         block, 1, mp->m_bmap_dmxr[0]);
4404                 temp = trp;
4405                 for (j = 0; j < num_recs; j++, frp++, trp++) {
4406                         trp->l0 = INT_GET(frp->l0, ARCH_CONVERT);
4407                         trp->l1 = INT_GET(frp->l1, ARCH_CONVERT);
4408                 }
4409                 if (exntf == XFS_EXTFMT_NOSTATE) {
4410                         /*
4411                          * Check all attribute bmap btree records and
4412                          * any "older" data bmap btree records for a
4413                          * set bit in the "extent flag" position.
4414                          */
4415                         if (unlikely(xfs_check_nostate_extents(temp, num_recs))) {
4416                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4417                                                  XFS_ERRLEVEL_LOW,
4418                                                  ip->i_mount);
4419                                 goto error0;
4420                         }
4421                 }
4422                 i += num_recs;
4423                 xfs_trans_brelse(tp, bp);
4424                 bno = nextbno;
4425                 /*
4426                  * If we've reached the end, stop.
4427                  */
4428                 if (bno == NULLFSBLOCK)
4429                         break;
4430                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4431                                 XFS_BMAP_BTREE_REF)))
4432                         return error;
4433                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
4434         }
4435         ASSERT(i == ifp->if_bytes / (uint)sizeof(*trp));
4436         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4437         xfs_bmap_trace_exlist(fname, ip, i, whichfork);
4438         return 0;
4439 error0:
4440         xfs_trans_brelse(tp, bp);
4441         return XFS_ERROR(EFSCORRUPTED);
4442 }
4443
4444 #ifdef XFS_BMAP_TRACE
4445 /*
4446  * Add bmap trace insert entries for all the contents of the extent list.
4447  */
4448 void
4449 xfs_bmap_trace_exlist(
4450         char            *fname,         /* function name */
4451         xfs_inode_t     *ip,            /* incore inode pointer */
4452         xfs_extnum_t    cnt,            /* count of entries in the list */
4453         int             whichfork)      /* data or attr fork */
4454 {
4455         xfs_bmbt_rec_t  *base;          /* base of extent list */
4456         xfs_bmbt_rec_t  *ep;            /* current entry in extent list */
4457         xfs_extnum_t    idx;            /* extent list entry number */
4458         xfs_ifork_t     *ifp;           /* inode fork pointer */
4459         xfs_bmbt_irec_t s;              /* extent list record */
4460
4461         ifp = XFS_IFORK_PTR(ip, whichfork);
4462         ASSERT(cnt == ifp->if_bytes / (uint)sizeof(*base));
4463         base = ifp->if_u1.if_extents;
4464         for (idx = 0, ep = base; idx < cnt; idx++, ep++) {
4465                 xfs_bmbt_get_all(ep, &s);
4466                 xfs_bmap_trace_insert(fname, "exlist", ip, idx, 1, &s, NULL,
4467                         whichfork);
4468         }
4469 }
4470 #endif
4471
4472 #ifdef DEBUG
4473 /*
4474  * Validate that the bmbt_irecs being returned from bmapi are valid
4475  * given the callers original parameters.  Specifically check the
4476  * ranges of the returned irecs to ensure that they only extent beyond
4477  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4478  */
4479 STATIC void
4480 xfs_bmap_validate_ret(
4481         xfs_fileoff_t           bno,
4482         xfs_filblks_t           len,
4483         int                     flags,
4484         xfs_bmbt_irec_t         *mval,
4485         int                     nmap,
4486         int                     ret_nmap)
4487 {
4488         int                     i;              /* index to map values */
4489
4490         ASSERT(ret_nmap <= nmap);
4491
4492         for (i = 0; i < ret_nmap; i++) {
4493                 ASSERT(mval[i].br_blockcount > 0);
4494                 if (!(flags & XFS_BMAPI_ENTIRE)) {
4495                         ASSERT(mval[i].br_startoff >= bno);
4496                         ASSERT(mval[i].br_blockcount <= len);
4497                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4498                                bno + len);
4499                 } else {
4500                         ASSERT(mval[i].br_startoff < bno + len);
4501                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4502                                bno);
4503                 }
4504                 ASSERT(i == 0 ||
4505                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4506                        mval[i].br_startoff);
4507                 if ((flags & XFS_BMAPI_WRITE) && !(flags & XFS_BMAPI_DELAY))
4508                         ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4509                                mval[i].br_startblock != HOLESTARTBLOCK);
4510                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4511                        mval[i].br_state == XFS_EXT_UNWRITTEN);
4512         }
4513 }
4514 #endif /* DEBUG */
4515
4516
4517 /*
4518  * Map file blocks to filesystem blocks.
4519  * File range is given by the bno/len pair.
4520  * Adds blocks to file if a write ("flags & XFS_BMAPI_WRITE" set)
4521  * into a hole or past eof.
4522  * Only allocates blocks from a single allocation group,
4523  * to avoid locking problems.
4524  * The returned value in "firstblock" from the first call in a transaction
4525  * must be remembered and presented to subsequent calls in "firstblock".
4526  * An upper bound for the number of blocks to be allocated is supplied to
4527  * the first call in "total"; if no allocation group has that many free
4528  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4529  */
4530 int                                     /* error */
4531 xfs_bmapi(
4532         xfs_trans_t     *tp,            /* transaction pointer */
4533         xfs_inode_t     *ip,            /* incore inode */
4534         xfs_fileoff_t   bno,            /* starting file offs. mapped */
4535         xfs_filblks_t   len,            /* length to map in file */
4536         int             flags,          /* XFS_BMAPI_... */
4537         xfs_fsblock_t   *firstblock,    /* first allocated block
4538                                            controls a.g. for allocs */
4539         xfs_extlen_t    total,          /* total blocks needed */
4540         xfs_bmbt_irec_t *mval,          /* output: map values */
4541         int             *nmap,          /* i/o: mval size/count */
4542         xfs_bmap_free_t *flist)         /* i/o: list extents to free */
4543 {
4544         xfs_fsblock_t   abno;           /* allocated block number */
4545         xfs_extlen_t    alen;           /* allocated extent length */
4546         xfs_fileoff_t   aoff;           /* allocated file offset */
4547         xfs_bmalloca_t  bma;            /* args for xfs_bmap_alloc */
4548         xfs_btree_cur_t *cur;           /* bmap btree cursor */
4549         xfs_fileoff_t   end;            /* end of mapped file region */
4550         int             eof;            /* we've hit the end of extent list */
4551         char            contig;         /* allocation must be one extent */
4552         char            delay;          /* this request is for delayed alloc */
4553         char            exact;          /* don't do all of wasdelayed extent */
4554         xfs_bmbt_rec_t  *ep;            /* extent list entry pointer */
4555         int             error;          /* error return */
4556         xfs_bmbt_irec_t got;            /* current extent list record */
4557         xfs_ifork_t     *ifp;           /* inode fork pointer */
4558         xfs_extlen_t    indlen;         /* indirect blocks length */
4559         xfs_extnum_t    lastx;          /* last useful extent number */
4560         int             logflags;       /* flags for transaction logging */
4561         xfs_extlen_t    minleft;        /* min blocks left after allocation */
4562         xfs_extlen_t    minlen;         /* min allocation size */
4563         xfs_mount_t     *mp;            /* xfs mount structure */
4564         int             n;              /* current extent index */
4565         int             nallocs;        /* number of extents alloc\'d */
4566         xfs_extnum_t    nextents;       /* number of extents in file */
4567         xfs_fileoff_t   obno;           /* old block number (offset) */
4568         xfs_bmbt_irec_t prev;           /* previous extent list record */
4569         int             tmp_logflags;   /* temp flags holder */
4570         int             whichfork;      /* data or attr fork */
4571         char            inhole;         /* current location is hole in file */
4572         char            stateless;      /* ignore state flag set */
4573         char            trim;           /* output trimmed to match range */
4574         char            userdata;       /* allocating non-metadata */
4575         char            wasdelay;       /* old extent was delayed */
4576         char            wr;             /* this is a write request */
4577         char            rt;             /* this is a realtime file */
4578         char            rsvd;           /* OK to allocate reserved blocks */
4579 #ifdef DEBUG
4580         xfs_fileoff_t   orig_bno;       /* original block number value */
4581         int             orig_flags;     /* original flags arg value */
4582         xfs_filblks_t   orig_len;       /* original value of len arg */
4583         xfs_bmbt_irec_t *orig_mval;     /* original value of mval */
4584         int             orig_nmap;      /* original value of *nmap */
4585
4586         orig_bno = bno;
4587         orig_len = len;
4588         orig_flags = flags;
4589         orig_mval = mval;
4590         orig_nmap = *nmap;
4591 #endif
4592         ASSERT(*nmap >= 1);
4593         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP || !(flags & XFS_BMAPI_WRITE));
4594         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4595                 XFS_ATTR_FORK : XFS_DATA_FORK;
4596         mp = ip->i_mount;
4597         if (unlikely(XFS_TEST_ERROR(
4598             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4599              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4600              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4601              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4602                 XFS_ERROR_REPORT("xfs_bmapi", XFS_ERRLEVEL_LOW, mp);
4603                 return XFS_ERROR(EFSCORRUPTED);
4604         }
4605         if (XFS_FORCED_SHUTDOWN(mp))
4606                 return XFS_ERROR(EIO);
4607         rt = XFS_IS_REALTIME_INODE(ip);
4608         ifp = XFS_IFORK_PTR(ip, whichfork);
4609         ASSERT(ifp->if_ext_max ==
4610                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4611         if ((wr = (flags & XFS_BMAPI_WRITE)) != 0)
4612                 XFS_STATS_INC(xs_blk_mapw);
4613         else
4614                 XFS_STATS_INC(xs_blk_mapr);
4615         delay = (flags & XFS_BMAPI_DELAY) != 0;
4616         trim = (flags & XFS_BMAPI_ENTIRE) == 0;
4617         userdata = (flags & XFS_BMAPI_METADATA) == 0;
4618         exact = (flags & XFS_BMAPI_EXACT) != 0;
4619         rsvd = (flags & XFS_BMAPI_RSVBLOCKS) != 0;
4620         contig = (flags & XFS_BMAPI_CONTIG) != 0;
4621         /*
4622          * stateless is used to combine extents which
4623          * differ only due to the state of the extents.
4624          * This technique is used from xfs_getbmap()
4625          * when the caller does not wish to see the
4626          * separation (which is the default).
4627          *
4628          * This technique is also used when writing a
4629          * buffer which has been partially written,
4630          * (usually by being flushed during a chunkread),
4631          * to ensure one write takes place. This also
4632          * prevents a change in the xfs inode extents at
4633          * this time, intentionally. This change occurs
4634          * on completion of the write operation, in
4635          * xfs_strat_comp(), where the xfs_bmapi() call
4636          * is transactioned, and the extents combined.
4637          */
4638         stateless = (flags & XFS_BMAPI_IGSTATE) != 0;
4639         if (stateless && wr)    /* if writing unwritten space, no */
4640                 wr = 0;         /* allocations are allowed */
4641         ASSERT(wr || !delay);
4642         logflags = 0;
4643         nallocs = 0;
4644         cur = NULL;
4645         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4646                 ASSERT(wr && tp);
4647                 if ((error = xfs_bmap_local_to_extents(tp, ip,
4648                                 firstblock, total, &logflags, whichfork)))
4649                         goto error0;
4650         }
4651         if (wr && *firstblock == NULLFSBLOCK) {
4652                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4653                         minleft = INT_GET(ifp->if_broot->bb_level, ARCH_CONVERT) + 1;
4654                 else
4655                         minleft = 1;
4656         } else
4657                 minleft = 0;
4658         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4659             (error = xfs_iread_extents(tp, ip, whichfork)))
4660                 goto error0;
4661         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4662                 &prev);
4663         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4664         n = 0;
4665         end = bno + len;
4666         obno = bno;
4667         bma.ip = NULL;
4668         while (bno < end && n < *nmap) {
4669                 /*
4670                  * Reading past eof, act as though there's a hole
4671                  * up to end.
4672                  */
4673                 if (eof && !wr)
4674                         got.br_startoff = end;
4675                 inhole = eof || got.br_startoff > bno;
4676                 wasdelay = wr && !inhole && !delay &&
4677                         ISNULLSTARTBLOCK(got.br_startblock);
4678                 /*
4679                  * First, deal with the hole before the allocated space
4680                  * that we found, if any.
4681                  */
4682                 if (wr && (inhole || wasdelay)) {
4683                         /*
4684                          * For the wasdelay case, we could also just
4685                          * allocate the stuff asked for in this bmap call
4686                          * but that wouldn't be as good.
4687                          */
4688                         if (wasdelay && !exact) {
4689                                 alen = (xfs_extlen_t)got.br_blockcount;
4690                                 aoff = got.br_startoff;
4691                                 if (lastx != NULLEXTNUM && lastx) {
4692                                         ep = &ifp->if_u1.if_extents[lastx - 1];
4693                                         xfs_bmbt_get_all(ep, &prev);
4694                                 }
4695                         } else if (wasdelay) {
4696                                 alen = (xfs_extlen_t)
4697                                         XFS_FILBLKS_MIN(len,
4698                                                 (got.br_startoff +
4699                                                  got.br_blockcount) - bno);
4700                                 aoff = bno;
4701                         } else {
4702                                 alen = (xfs_extlen_t)
4703                                         XFS_FILBLKS_MIN(len, MAXEXTLEN);
4704                                 if (!eof)
4705                                         alen = (xfs_extlen_t)
4706                                                 XFS_FILBLKS_MIN(alen,
4707                                                         got.br_startoff - bno);
4708                                 aoff = bno;
4709                         }
4710                         minlen = contig ? alen : 1;
4711                         if (delay) {
4712                                 xfs_extlen_t    extsz = 0;
4713
4714                                 /* Figure out the extent size, adjust alen */
4715                                 if (rt) {
4716                                         if (!(extsz = ip->i_d.di_extsize))
4717                                                 extsz = mp->m_sb.sb_rextsize;
4718                                         alen = roundup(alen, extsz);
4719                                         extsz = alen / mp->m_sb.sb_rextsize;
4720                                 }
4721
4722                                 /*
4723                                  * Make a transaction-less quota reservation for
4724                                  * delayed allocation blocks. This number gets
4725                                  * adjusted later.
4726                                  * We return EDQUOT if we haven't allocated
4727                                  * blks already inside this loop;
4728                                  */
4729                                 if (XFS_TRANS_RESERVE_QUOTA_NBLKS(
4730                                                 mp, NULL, ip, (long)alen, 0,
4731                                                 rt ? XFS_QMOPT_RES_RTBLKS :
4732                                                      XFS_QMOPT_RES_REGBLKS)) {
4733                                         if (n == 0) {
4734                                                 *nmap = 0;
4735                                                 ASSERT(cur == NULL);
4736                                                 return XFS_ERROR(EDQUOT);
4737                                         }
4738                                         break;
4739                                 }
4740
4741                                 /*
4742                                  * Split changing sb for alen and indlen since
4743                                  * they could be coming from different places.
4744                                  */
4745                                 indlen = (xfs_extlen_t)
4746                                         xfs_bmap_worst_indlen(ip, alen);
4747                                 ASSERT(indlen > 0);
4748
4749                                 if (rt)
4750                                         error = xfs_mod_incore_sb(mp,
4751                                                         XFS_SBS_FREXTENTS,
4752                                                         -(extsz), rsvd);
4753                                 else
4754                                         error = xfs_mod_incore_sb(mp,
4755                                                         XFS_SBS_FDBLOCKS,
4756                                                         -(alen), rsvd);
4757                                 if (!error)
4758                                         error = xfs_mod_incore_sb(mp,
4759                                                         XFS_SBS_FDBLOCKS,
4760                                                         -(indlen), rsvd);
4761
4762                                 if (error) {
4763                                         if (XFS_IS_QUOTA_ON(ip->i_mount))
4764                                                 /* unreserve the blocks now */
4765                                                 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(
4766                                                         mp, NULL, ip,
4767                                                         (long)alen, 0, rt ?
4768                                                         XFS_QMOPT_RES_RTBLKS :
4769                                                         XFS_QMOPT_RES_REGBLKS);
4770                                         break;
4771                                 }
4772
4773                                 ip->i_delayed_blks += alen;
4774                                 abno = NULLSTARTBLOCK(indlen);
4775                         } else {
4776                                 /*
4777                                  * If first time, allocate and fill in
4778                                  * once-only bma fields.
4779                                  */
4780                                 if (bma.ip == NULL) {
4781                                         bma.tp = tp;
4782                                         bma.ip = ip;
4783                                         bma.prevp = &prev;
4784                                         bma.gotp = &got;
4785                                         bma.total = total;
4786                                         bma.userdata = 0;
4787                                 }
4788                                 /* Indicate if this is the first user data
4789                                  * in the file, or just any user data.
4790                                  */
4791                                 if (userdata) {
4792                                         bma.userdata = (aoff == 0) ?
4793                                                 XFS_ALLOC_INITIAL_USER_DATA :
4794                                                 XFS_ALLOC_USERDATA;
4795                                 }
4796                                 /*
4797                                  * Fill in changeable bma fields.
4798                                  */
4799                                 bma.eof = eof;
4800                                 bma.firstblock = *firstblock;
4801                                 bma.alen = alen;
4802                                 bma.off = aoff;
4803                                 bma.wasdel = wasdelay;
4804                                 bma.minlen = minlen;
4805                                 bma.low = flist->xbf_low;
4806                                 bma.minleft = minleft;
4807                                 /*
4808                                  * Only want to do the alignment at the
4809                                  * eof if it is userdata and allocation length
4810                                  * is larger than a stripe unit.
4811                                  */
4812                                 if (mp->m_dalign && alen >= mp->m_dalign &&
4813                                     userdata && whichfork == XFS_DATA_FORK) {
4814                                         if ((error = xfs_bmap_isaeof(ip, aoff,
4815                                                         whichfork, &bma.aeof)))
4816                                                 goto error0;
4817                                 } else
4818                                         bma.aeof = 0;
4819                                 /*
4820                                  * Call allocator.
4821                                  */
4822                                 if ((error = xfs_bmap_alloc(&bma)))
4823                                         goto error0;
4824                                 /*
4825                                  * Copy out result fields.
4826                                  */
4827                                 abno = bma.rval;
4828                                 if ((flist->xbf_low = bma.low))
4829                                         minleft = 0;
4830                                 alen = bma.alen;
4831                                 aoff = bma.off;
4832                                 ASSERT(*firstblock == NULLFSBLOCK ||
4833                                        XFS_FSB_TO_AGNO(mp, *firstblock) ==
4834                                        XFS_FSB_TO_AGNO(mp, bma.firstblock) ||
4835                                        (flist->xbf_low &&
4836                                         XFS_FSB_TO_AGNO(mp, *firstblock) <
4837                                         XFS_FSB_TO_AGNO(mp, bma.firstblock)));
4838                                 *firstblock = bma.firstblock;
4839                                 if (cur)
4840                                         cur->bc_private.b.firstblock =
4841                                                 *firstblock;
4842                                 if (abno == NULLFSBLOCK)
4843                                         break;
4844                                 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
4845                                         cur = xfs_btree_init_cursor(mp,
4846                                                 tp, NULL, 0, XFS_BTNUM_BMAP,
4847                                                 ip, whichfork);
4848                                         cur->bc_private.b.firstblock =
4849                                                 *firstblock;
4850                                         cur->bc_private.b.flist = flist;
4851                                 }
4852                                 /*
4853                                  * Bump the number of extents we've allocated
4854                                  * in this call.
4855                                  */
4856                                 nallocs++;
4857                         }
4858                         if (cur)
4859                                 cur->bc_private.b.flags =
4860                                         wasdelay ? XFS_BTCUR_BPRV_WASDEL : 0;
4861                         got.br_startoff = aoff;
4862                         got.br_startblock = abno;
4863                         got.br_blockcount = alen;
4864                         got.br_state = XFS_EXT_NORM;    /* assume normal */
4865                         /*
4866                          * Determine state of extent, and the filesystem.
4867                          * A wasdelay extent has been initialized, so
4868                          * shouldn't be flagged as unwritten.
4869                          */
4870                         if (wr && XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4871                                 if (!wasdelay && (flags & XFS_BMAPI_PREALLOC))
4872                                         got.br_state = XFS_EXT_UNWRITTEN;
4873                         }
4874                         error = xfs_bmap_add_extent(ip, lastx, &cur, &got,
4875                                 firstblock, flist, &tmp_logflags, whichfork,
4876                                 rsvd);
4877                         logflags |= tmp_logflags;
4878                         if (error)
4879                                 goto error0;
4880                         lastx = ifp->if_lastex;
4881                         ep = &ifp->if_u1.if_extents[lastx];
4882                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4883                         xfs_bmbt_get_all(ep, &got);
4884                         ASSERT(got.br_startoff <= aoff);
4885                         ASSERT(got.br_startoff + got.br_blockcount >=
4886                                 aoff + alen);
4887 #ifdef DEBUG
4888                         if (delay) {
4889                                 ASSERT(ISNULLSTARTBLOCK(got.br_startblock));
4890                                 ASSERT(STARTBLOCKVAL(got.br_startblock) > 0);
4891                         }
4892                         ASSERT(got.br_state == XFS_EXT_NORM ||
4893                                got.br_state == XFS_EXT_UNWRITTEN);
4894 #endif
4895                         /*
4896                          * Fall down into the found allocated space case.
4897                          */
4898                 } else if (inhole) {
4899                         /*
4900                          * Reading in a hole.
4901                          */
4902                         mval->br_startoff = bno;
4903                         mval->br_startblock = HOLESTARTBLOCK;
4904                         mval->br_blockcount =
4905                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4906                         mval->br_state = XFS_EXT_NORM;
4907                         bno += mval->br_blockcount;
4908                         len -= mval->br_blockcount;
4909                         mval++;
4910                         n++;
4911                         continue;
4912                 }
4913                 /*
4914                  * Then deal with the allocated space we found.
4915                  */
4916                 ASSERT(ep != NULL);
4917                 if (trim && (got.br_startoff + got.br_blockcount > obno)) {
4918                         if (obno > bno)
4919                                 bno = obno;
4920                         ASSERT((bno >= obno) || (n == 0));
4921                         ASSERT(bno < end);
4922                         mval->br_startoff = bno;
4923                         if (ISNULLSTARTBLOCK(got.br_startblock)) {
4924                                 ASSERT(!wr || delay);
4925                                 mval->br_startblock = DELAYSTARTBLOCK;
4926                         } else
4927                                 mval->br_startblock =
4928                                         got.br_startblock +
4929                                         (bno - got.br_startoff);
4930                         /*
4931                          * Return the minimum of what we got and what we
4932                          * asked for for the length.  We can use the len
4933                          * variable here because it is modified below
4934                          * and we could have been there before coming
4935                          * here if the first part of the allocation
4936                          * didn't overlap what was asked for.
4937                          */
4938                         mval->br_blockcount =
4939                                 XFS_FILBLKS_MIN(end - bno, got.br_blockcount -
4940                                         (bno - got.br_startoff));
4941                         mval->br_state = got.br_state;
4942                         ASSERT(mval->br_blockcount <= len);
4943                 } else {
4944                         *mval = got;
4945                         if (ISNULLSTARTBLOCK(mval->br_startblock)) {
4946                                 ASSERT(!wr || delay);
4947                                 mval->br_startblock = DELAYSTARTBLOCK;
4948                         }
4949                 }
4950
4951                 /*
4952                  * Check if writing previously allocated but
4953                  * unwritten extents.
4954                  */
4955                 if (wr && mval->br_state == XFS_EXT_UNWRITTEN &&
4956                     ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_DELAY)) == 0)) {
4957                         /*
4958                          * Modify (by adding) the state flag, if writing.
4959                          */
4960                         ASSERT(mval->br_blockcount <= len);
4961                         if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
4962                                 cur = xfs_btree_init_cursor(mp,
4963                                         tp, NULL, 0, XFS_BTNUM_BMAP,
4964                                         ip, whichfork);
4965                                 cur->bc_private.b.firstblock =
4966                                         *firstblock;
4967                                 cur->bc_private.b.flist = flist;
4968                         }
4969                         mval->br_state = XFS_EXT_NORM;
4970                         error = xfs_bmap_add_extent(ip, lastx, &cur, mval,
4971                                 firstblock, flist, &tmp_logflags, whichfork,
4972                                 rsvd);
4973                         logflags |= tmp_logflags;
4974                         if (error)
4975                                 goto error0;
4976                         lastx = ifp->if_lastex;
4977                         ep = &ifp->if_u1.if_extents[lastx];
4978                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4979                         xfs_bmbt_get_all(ep, &got);
4980                         /*
4981                          * We may have combined previously unwritten
4982                          * space with written space, so generate
4983                          * another request.
4984                          */
4985                         if (mval->br_blockcount < len)
4986                                 continue;
4987                 }
4988
4989                 ASSERT(!trim ||
4990                        ((mval->br_startoff + mval->br_blockcount) <= end));
4991                 ASSERT(!trim || (mval->br_blockcount <= len) ||
4992                        (mval->br_startoff < obno));
4993                 bno = mval->br_startoff + mval->br_blockcount;
4994                 len = end - bno;
4995                 if (n > 0 && mval->br_startoff == mval[-1].br_startoff) {
4996                         ASSERT(mval->br_startblock == mval[-1].br_startblock);
4997                         ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
4998                         ASSERT(mval->br_state == mval[-1].br_state);
4999                         mval[-1].br_blockcount = mval->br_blockcount;
5000                         mval[-1].br_state = mval->br_state;
5001                 } else if (n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
5002                            mval[-1].br_startblock != DELAYSTARTBLOCK &&
5003                            mval[-1].br_startblock != HOLESTARTBLOCK &&
5004                            mval->br_startblock ==
5005                            mval[-1].br_startblock + mval[-1].br_blockcount &&
5006                            (stateless || mval[-1].br_state == mval->br_state)) {
5007                         ASSERT(mval->br_startoff ==
5008                                mval[-1].br_startoff + mval[-1].br_blockcount);
5009                         mval[-1].br_blockcount += mval->br_blockcount;
5010                 } else if (n > 0 &&
5011                            mval->br_startblock == DELAYSTARTBLOCK &&
5012                            mval[-1].br_startblock == DELAYSTARTBLOCK &&
5013                            mval->br_startoff ==
5014                            mval[-1].br_startoff + mval[-1].br_blockcount) {
5015                         mval[-1].br_blockcount += mval->br_blockcount;
5016                         mval[-1].br_state = mval->br_state;
5017                 } else if (!((n == 0) &&
5018                              ((mval->br_startoff + mval->br_blockcount) <=
5019                               obno))) {
5020                         mval++;
5021                         n++;
5022                 }
5023                 /*
5024                  * If we're done, stop now.  Stop when we've allocated
5025                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
5026                  * the transaction may get too big.
5027                  */
5028                 if (bno >= end || n >= *nmap || nallocs >= *nmap)
5029                         break;
5030                 /*
5031                  * Else go on to the next record.
5032                  */
5033                 ep++;
5034                 lastx++;
5035                 if (lastx >= nextents) {
5036                         eof = 1;
5037                         prev = got;
5038                 } else
5039                         xfs_bmbt_get_all(ep, &got);
5040         }
5041         ifp->if_lastex = lastx;
5042         *nmap = n;
5043         /*
5044          * Transform from btree to extents, give it cur.
5045          */
5046         if (tp && XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5047             XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5048                 ASSERT(wr && cur);
5049                 error = xfs_bmap_btree_to_extents(tp, ip, cur,
5050                         &tmp_logflags, whichfork);
5051                 logflags |= tmp_logflags;
5052                 if (error)
5053                         goto error0;
5054         }
5055         ASSERT(ifp->if_ext_max ==
5056                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5057         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
5058                XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max);
5059         error = 0;
5060
5061 error0:
5062         /*
5063          * Log everything.  Do this after conversion, there's no point in
5064          * logging the extent list if we've converted to btree format.
5065          */
5066         if ((logflags & XFS_ILOG_FEXT(whichfork)) &&
5067             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5068                 logflags &= ~XFS_ILOG_FEXT(whichfork);
5069         else if ((logflags & XFS_ILOG_FBROOT(whichfork)) &&
5070                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5071                 logflags &= ~XFS_ILOG_FBROOT(whichfork);
5072         /*
5073          * Log whatever the flags say, even if error.  Otherwise we might miss
5074          * detecting a case where the data is changed, there's an error,
5075          * and it's not logged so we don't shutdown when we should.
5076          */
5077         if (logflags) {
5078                 ASSERT(tp && wr);
5079                 xfs_trans_log_inode(tp, ip, logflags);
5080         }
5081         if (cur) {
5082                 if (!error) {
5083                         ASSERT(*firstblock == NULLFSBLOCK ||
5084                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
5085                                XFS_FSB_TO_AGNO(mp,
5086                                        cur->bc_private.b.firstblock) ||
5087                                (flist->xbf_low &&
5088                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
5089                                 XFS_FSB_TO_AGNO(mp,
5090                                         cur->bc_private.b.firstblock)));
5091                         *firstblock = cur->bc_private.b.firstblock;
5092                 }
5093                 xfs_btree_del_cursor(cur,
5094                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5095         }
5096         if (!error)
5097                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
5098                         orig_nmap, *nmap);
5099         return error;
5100 }
5101
5102 /*
5103  * Map file blocks to filesystem blocks, simple version.
5104  * One block (extent) only, read-only.
5105  * For flags, only the XFS_BMAPI_ATTRFORK flag is examined.
5106  * For the other flag values, the effect is as if XFS_BMAPI_METADATA
5107  * was set and all the others were clear.
5108  */
5109 int                                             /* error */
5110 xfs_bmapi_single(
5111         xfs_trans_t     *tp,            /* transaction pointer */
5112         xfs_inode_t     *ip,            /* incore inode */
5113         int             whichfork,      /* data or attr fork */
5114         xfs_fsblock_t   *fsb,           /* output: mapped block */
5115         xfs_fileoff_t   bno)            /* starting file offs. mapped */
5116 {
5117         int             eof;            /* we've hit the end of extent list */
5118         int             error;          /* error return */
5119         xfs_bmbt_irec_t got;            /* current extent list record */
5120         xfs_ifork_t     *ifp;           /* inode fork pointer */
5121         xfs_extnum_t    lastx;          /* last useful extent number */
5122         xfs_bmbt_irec_t prev;           /* previous extent list record */
5123
5124         ifp = XFS_IFORK_PTR(ip, whichfork);
5125         if (unlikely(
5126             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
5127             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)) {
5128                XFS_ERROR_REPORT("xfs_bmapi_single", XFS_ERRLEVEL_LOW,
5129                                 ip->i_mount);
5130                return XFS_ERROR(EFSCORRUPTED);
5131         }
5132         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
5133                 return XFS_ERROR(EIO);
5134         XFS_STATS_INC(xs_blk_mapr);
5135         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5136             (error = xfs_iread_extents(tp, ip, whichfork)))
5137                 return error;
5138         (void)xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5139                 &prev);
5140         /*
5141          * Reading past eof, act as though there's a hole
5142          * up to end.
5143          */
5144         if (eof || got.br_startoff > bno) {
5145                 *fsb = NULLFSBLOCK;
5146                 return 0;
5147         }
5148         ASSERT(!ISNULLSTARTBLOCK(got.br_startblock));
5149         ASSERT(bno < got.br_startoff + got.br_blockcount);
5150         *fsb = got.br_startblock + (bno - got.br_startoff);
5151         ifp->if_lastex = lastx;
5152         return 0;
5153 }
5154
5155 /*
5156  * Unmap (remove) blocks from a file.
5157  * If nexts is nonzero then the number of extents to remove is limited to
5158  * that value.  If not all extents in the block range can be removed then
5159  * *done is set.
5160  */
5161 int                                             /* error */
5162 xfs_bunmapi(
5163         xfs_trans_t             *tp,            /* transaction pointer */
5164         struct xfs_inode        *ip,            /* incore inode */
5165         xfs_fileoff_t           bno,            /* starting offset to unmap */
5166         xfs_filblks_t           len,            /* length to unmap in file */
5167         int                     flags,          /* misc flags */
5168         xfs_extnum_t            nexts,          /* number of extents max */
5169         xfs_fsblock_t           *firstblock,    /* first allocated block
5170                                                    controls a.g. for allocs */
5171         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
5172         int                     *done)          /* set if not done yet */
5173 {
5174         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
5175         xfs_bmbt_irec_t         del;            /* extent being deleted */
5176         int                     eof;            /* is deleting at eof */
5177         xfs_bmbt_rec_t          *ep;            /* extent list entry pointer */
5178         int                     error;          /* error return value */
5179         xfs_extnum_t            extno;          /* extent number in list */
5180         xfs_bmbt_irec_t         got;            /* current extent list entry */
5181         xfs_ifork_t             *ifp;           /* inode fork pointer */
5182         int                     isrt;           /* freeing in rt area */
5183         xfs_extnum_t            lastx;          /* last extent index used */
5184         int                     logflags;       /* transaction logging flags */
5185         xfs_extlen_t            mod;            /* rt extent offset */
5186         xfs_mount_t             *mp;            /* mount structure */
5187         xfs_extnum_t            nextents;       /* size of extent list */
5188         xfs_bmbt_irec_t         prev;           /* previous extent list entry */
5189         xfs_fileoff_t           start;          /* first file offset deleted */
5190         int                     tmp_logflags;   /* partial logging flags */
5191         int                     wasdel;         /* was a delayed alloc extent */
5192         int                     whichfork;      /* data or attribute fork */
5193         int                     rsvd;           /* OK to allocate reserved blocks */
5194         xfs_fsblock_t           sum;
5195
5196         xfs_bunmap_trace(ip, bno, len, flags, (inst_t *)__return_address);
5197         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5198                 XFS_ATTR_FORK : XFS_DATA_FORK;
5199         ifp = XFS_IFORK_PTR(ip, whichfork);
5200         if (unlikely(
5201             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5202             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5203                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5204                                  ip->i_mount);
5205                 return XFS_ERROR(EFSCORRUPTED);
5206         }
5207         mp = ip->i_mount;
5208         if (XFS_FORCED_SHUTDOWN(mp))
5209                 return XFS_ERROR(EIO);
5210         rsvd = (flags & XFS_BMAPI_RSVBLOCKS) != 0;
5211         ASSERT(len > 0);
5212         ASSERT(nexts >= 0);
5213         ASSERT(ifp->if_ext_max ==
5214                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5215         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5216             (error = xfs_iread_extents(tp, ip, whichfork)))
5217                 return error;
5218         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5219         if (nextents == 0) {
5220                 *done = 1;
5221                 return 0;
5222         }
5223         XFS_STATS_INC(xs_blk_unmap);
5224         isrt = (whichfork == XFS_DATA_FORK) &&
5225                (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
5226         start = bno;
5227         bno = start + len - 1;
5228         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5229                 &prev);
5230         /*
5231          * Check to see if the given block number is past the end of the
5232          * file, back up to the last block if so...
5233          */
5234         if (eof) {
5235                 ep = &ifp->if_u1.if_extents[--lastx];
5236                 xfs_bmbt_get_all(ep, &got);
5237                 bno = got.br_startoff + got.br_blockcount - 1;
5238         }
5239         logflags = 0;
5240         if (ifp->if_flags & XFS_IFBROOT) {
5241                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5242                 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
5243                         whichfork);
5244                 cur->bc_private.b.firstblock = *firstblock;
5245                 cur->bc_private.b.flist = flist;
5246                 cur->bc_private.b.flags = 0;
5247         } else
5248                 cur = NULL;
5249         extno = 0;
5250         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5251                (nexts == 0 || extno < nexts)) {
5252                 /*
5253                  * Is the found extent after a hole in which bno lives?
5254                  * Just back up to the previous extent, if so.
5255                  */
5256                 if (got.br_startoff > bno) {
5257                         if (--lastx < 0)
5258                                 break;
5259                         ep--;
5260                         xfs_bmbt_get_all(ep, &got);
5261                 }
5262                 /*
5263                  * Is the last block of this extent before the range
5264                  * we're supposed to delete?  If so, we're done.
5265                  */
5266                 bno = XFS_FILEOFF_MIN(bno,
5267                         got.br_startoff + got.br_blockcount - 1);
5268                 if (bno < start)
5269                         break;
5270                 /*
5271                  * Then deal with the (possibly delayed) allocated space
5272                  * we found.
5273                  */
5274                 ASSERT(ep != NULL);
5275                 del = got;
5276                 wasdel = ISNULLSTARTBLOCK(del.br_startblock);
5277                 if (got.br_startoff < start) {
5278                         del.br_startoff = start;
5279                         del.br_blockcount -= start - got.br_startoff;
5280                         if (!wasdel)
5281                                 del.br_startblock += start - got.br_startoff;
5282                 }
5283                 if (del.br_startoff + del.br_blockcount > bno + 1)
5284                         del.br_blockcount = bno + 1 - del.br_startoff;
5285                 sum = del.br_startblock + del.br_blockcount;
5286                 if (isrt &&
5287                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5288                         /*
5289                          * Realtime extent not lined up at the end.
5290                          * The extent could have been split into written
5291                          * and unwritten pieces, or we could just be
5292                          * unmapping part of it.  But we can't really
5293                          * get rid of part of a realtime extent.
5294                          */
5295                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5296                             !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
5297                                 /*
5298                                  * This piece is unwritten, or we're not
5299                                  * using unwritten extents.  Skip over it.
5300                                  */
5301                                 ASSERT(bno >= mod);
5302                                 bno -= mod > del.br_blockcount ?
5303                                         del.br_blockcount : mod;
5304                                 if (bno < got.br_startoff) {
5305                                         if (--lastx >= 0)
5306                                                 xfs_bmbt_get_all(--ep, &got);
5307                                 }
5308                                 continue;
5309                         }
5310                         /*
5311                          * It's written, turn it unwritten.
5312                          * This is better than zeroing it.
5313                          */
5314                         ASSERT(del.br_state == XFS_EXT_NORM);
5315                         ASSERT(xfs_trans_get_block_res(tp) > 0);
5316                         /*
5317                          * If this spans a realtime extent boundary,
5318                          * chop it back to the start of the one we end at.
5319                          */
5320                         if (del.br_blockcount > mod) {
5321                                 del.br_startoff += del.br_blockcount - mod;
5322                                 del.br_startblock += del.br_blockcount - mod;
5323                                 del.br_blockcount = mod;
5324                         }
5325                         del.br_state = XFS_EXT_UNWRITTEN;
5326                         error = xfs_bmap_add_extent(ip, lastx, &cur, &del,
5327                                 firstblock, flist, &logflags, XFS_DATA_FORK, 0);
5328                         if (error)
5329                                 goto error0;
5330                         goto nodelete;
5331                 }
5332                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5333                         /*
5334                          * Realtime extent is lined up at the end but not
5335                          * at the front.  We'll get rid of full extents if
5336                          * we can.
5337                          */
5338                         mod = mp->m_sb.sb_rextsize - mod;
5339                         if (del.br_blockcount > mod) {
5340                                 del.br_blockcount -= mod;
5341                                 del.br_startoff += mod;
5342                                 del.br_startblock += mod;
5343                         } else if ((del.br_startoff == start &&
5344                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5345                                      xfs_trans_get_block_res(tp) == 0)) ||
5346                                    !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
5347                                 /*
5348                                  * Can't make it unwritten.  There isn't
5349                                  * a full extent here so just skip it.
5350                                  */
5351                                 ASSERT(bno >= del.br_blockcount);
5352                                 bno -= del.br_blockcount;
5353                                 if (bno < got.br_startoff) {
5354                                         if (--lastx >= 0)
5355                                                 xfs_bmbt_get_all(--ep, &got);
5356                                 }
5357                                 continue;
5358                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5359                                 /*
5360                                  * This one is already unwritten.
5361                                  * It must have a written left neighbor.
5362                                  * Unwrite the killed part of that one and
5363                                  * try again.
5364                                  */
5365                                 ASSERT(lastx > 0);
5366                                 xfs_bmbt_get_all(ep - 1, &prev);
5367                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5368                                 ASSERT(!ISNULLSTARTBLOCK(prev.br_startblock));
5369                                 ASSERT(del.br_startblock ==
5370                                        prev.br_startblock + prev.br_blockcount);
5371                                 if (prev.br_startoff < start) {
5372                                         mod = start - prev.br_startoff;
5373                                         prev.br_blockcount -= mod;
5374                                         prev.br_startblock += mod;
5375                                         prev.br_startoff = start;
5376                                 }
5377                                 prev.br_state = XFS_EXT_UNWRITTEN;
5378                                 error = xfs_bmap_add_extent(ip, lastx - 1, &cur,
5379                                         &prev, firstblock, flist, &logflags,
5380                                         XFS_DATA_FORK, 0);
5381                                 if (error)
5382                                         goto error0;
5383                                 goto nodelete;
5384                         } else {
5385                                 ASSERT(del.br_state == XFS_EXT_NORM);
5386                                 del.br_state = XFS_EXT_UNWRITTEN;
5387                                 error = xfs_bmap_add_extent(ip, lastx, &cur,
5388                                         &del, firstblock, flist, &logflags,
5389                                         XFS_DATA_FORK, 0);
5390                                 if (error)
5391                                         goto error0;
5392                                 goto nodelete;
5393                         }
5394                 }
5395                 if (wasdel) {
5396                         ASSERT(STARTBLOCKVAL(del.br_startblock) > 0);
5397                         /* Update realtim/data freespace, unreserve quota */
5398                         if (isrt) {
5399                                 xfs_filblks_t rtexts;
5400
5401                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5402                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5403                                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5404                                                 (int)rtexts, rsvd);
5405                                 XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, NULL, ip,
5406                                         -((long)del.br_blockcount), 0,
5407                                         XFS_QMOPT_RES_RTBLKS);
5408                         } else {
5409                                 xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS,
5410                                                 (int)del.br_blockcount, rsvd);
5411                                 XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, NULL, ip,
5412                                         -((long)del.br_blockcount), 0,
5413                                         XFS_QMOPT_RES_REGBLKS);
5414                         }
5415                         ip->i_delayed_blks -= del.br_blockcount;
5416                         if (cur)
5417                                 cur->bc_private.b.flags |=
5418                                         XFS_BTCUR_BPRV_WASDEL;
5419                 } else if (cur)
5420                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5421                 /*
5422                  * If it's the case where the directory code is running
5423                  * with no block reservation, and the deleted block is in
5424                  * the middle of its extent, and the resulting insert
5425                  * of an extent would cause transformation to btree format,
5426                  * then reject it.  The calling code will then swap
5427                  * blocks around instead.
5428                  * We have to do this now, rather than waiting for the
5429                  * conversion to btree format, since the transaction
5430                  * will be dirty.
5431                  */
5432                 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5433                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5434                     XFS_IFORK_NEXTENTS(ip, whichfork) >= ifp->if_ext_max &&
5435                     del.br_startoff > got.br_startoff &&
5436                     del.br_startoff + del.br_blockcount <
5437                     got.br_startoff + got.br_blockcount) {
5438                         error = XFS_ERROR(ENOSPC);
5439                         goto error0;
5440                 }
5441                 error = xfs_bmap_del_extent(ip, tp, lastx, flist, cur, &del,
5442                         &tmp_logflags, whichfork, rsvd);
5443                 logflags |= tmp_logflags;
5444                 if (error)
5445                         goto error0;
5446                 bno = del.br_startoff - 1;
5447 nodelete:
5448                 lastx = ifp->if_lastex;
5449                 /*
5450                  * If not done go on to the next (previous) record.
5451                  * Reset ep in case the extents array was re-alloced.
5452                  */
5453                 ep = &ifp->if_u1.if_extents[lastx];
5454                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5455                         if (lastx >= XFS_IFORK_NEXTENTS(ip, whichfork) ||
5456                             xfs_bmbt_get_startoff(ep) > bno) {
5457                                 lastx--;
5458                                 ep--;
5459                         }
5460                         if (lastx >= 0)
5461                                 xfs_bmbt_get_all(ep, &got);
5462                         extno++;
5463                 }
5464         }
5465         ifp->if_lastex = lastx;
5466         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5467         ASSERT(ifp->if_ext_max ==
5468                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5469         /*
5470          * Convert to a btree if necessary.
5471          */
5472         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5473             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
5474                 ASSERT(cur == NULL);
5475                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5476                         &cur, 0, &tmp_logflags, whichfork);
5477                 logflags |= tmp_logflags;
5478                 if (error)
5479                         goto error0;
5480         }
5481         /*
5482          * transform from btree to extents, give it cur
5483          */
5484         else if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5485                  XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5486                 ASSERT(cur != NULL);
5487                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5488                         whichfork);
5489                 logflags |= tmp_logflags;
5490                 if (error)
5491                         goto error0;
5492         }
5493         /*
5494          * transform from extents to local?
5495          */
5496         ASSERT(ifp->if_ext_max ==
5497                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5498         error = 0;
5499 error0:
5500         /*
5501          * Log everything.  Do this after conversion, there's no point in
5502          * logging the extent list if we've converted to btree format.
5503          */
5504         if ((logflags & XFS_ILOG_FEXT(whichfork)) &&
5505             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5506                 logflags &= ~XFS_ILOG_FEXT(whichfork);
5507         else if ((logflags & XFS_ILOG_FBROOT(whichfork)) &&
5508                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5509                 logflags &= ~XFS_ILOG_FBROOT(whichfork);
5510         /*
5511          * Log inode even in the error case, if the transaction
5512          * is dirty we'll need to shut down the filesystem.
5513          */
5514         if (logflags)
5515                 xfs_trans_log_inode(tp, ip, logflags);
5516         if (cur) {
5517                 if (!error) {
5518                         *firstblock = cur->bc_private.b.firstblock;
5519                         cur->bc_private.b.allocated = 0;
5520                 }
5521                 xfs_btree_del_cursor(cur,
5522                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5523         }
5524         return error;
5525 }
5526
5527 /*
5528  * Fcntl interface to xfs_bmapi.
5529  */
5530 int                                             /* error code */
5531 xfs_getbmap(
5532         bhv_desc_t              *bdp,           /* XFS behavior descriptor*/
5533         struct getbmap          *bmv,           /* user bmap structure */
5534         void                    __user *ap,     /* pointer to user's array */
5535         int                     interface)      /* interface flags */
5536 {
5537         __int64_t               bmvend;         /* last block requested */
5538         int                     error;          /* return value */
5539         __int64_t               fixlen;         /* length for -1 case */
5540         int                     i;              /* extent number */
5541         xfs_inode_t             *ip;            /* xfs incore inode pointer */
5542         vnode_t                 *vp;            /* corresponding vnode */
5543         int                     lock;           /* lock state */
5544         xfs_bmbt_irec_t         *map;           /* buffer for user's data */
5545         xfs_mount_t             *mp;            /* file system mount point */
5546         int                     nex;            /* # of user extents can do */
5547         int                     nexleft;        /* # of user extents left */
5548         int                     subnex;         /* # of bmapi's can do */
5549         int                     nmap;           /* number of map entries */
5550         struct getbmap          out;            /* output structure */
5551         int                     whichfork;      /* data or attr fork */
5552         int                     prealloced;     /* this is a file with
5553                                                  * preallocated data space */
5554         int                     sh_unwritten;   /* true, if unwritten */
5555                                                 /* extents listed separately */
5556         int                     bmapi_flags;    /* flags for xfs_bmapi */
5557         __int32_t               oflags;         /* getbmapx bmv_oflags field */
5558
5559         vp = BHV_TO_VNODE(bdp);
5560         ip = XFS_BHVTOI(bdp);
5561         mp = ip->i_mount;
5562
5563         whichfork = interface & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5564         sh_unwritten = (interface & BMV_IF_PREALLOC) != 0;
5565
5566         /*      If the BMV_IF_NO_DMAPI_READ interface bit specified, do not
5567          *      generate a DMAPI read event.  Otherwise, if the DM_EVENT_READ
5568          *      bit is set for the file, generate a read event in order
5569          *      that the DMAPI application may do its thing before we return
5570          *      the extents.  Usually this means restoring user file data to
5571          *      regions of the file that look like holes.
5572          *
5573          *      The "old behavior" (from XFS_IOC_GETBMAP) is to not specify
5574          *      BMV_IF_NO_DMAPI_READ so that read events are generated.
5575          *      If this were not true, callers of ioctl( XFS_IOC_GETBMAP )
5576          *      could misinterpret holes in a DMAPI file as true holes,
5577          *      when in fact they may represent offline user data.
5578          */
5579         if (   (interface & BMV_IF_NO_DMAPI_READ) == 0
5580             && DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_READ)
5581             && whichfork == XFS_DATA_FORK) {
5582
5583                 error = XFS_SEND_DATA(mp, DM_EVENT_READ, vp, 0, 0, 0, NULL);
5584                 if (error)
5585                         return XFS_ERROR(error);
5586         }
5587
5588         if (whichfork == XFS_ATTR_FORK) {
5589                 if (XFS_IFORK_Q(ip)) {
5590                         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5591                             ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5592                             ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5593                                 return XFS_ERROR(EINVAL);
5594                 } else if (unlikely(
5595                            ip->i_d.di_aformat != 0 &&
5596                            ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5597                         XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5598                                          ip->i_mount);
5599                         return XFS_ERROR(EFSCORRUPTED);
5600                 }
5601         } else if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5602                    ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5603                    ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5604                 return XFS_ERROR(EINVAL);
5605         if (whichfork == XFS_DATA_FORK) {
5606                 if (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC) {
5607                         prealloced = 1;
5608                         fixlen = XFS_MAXIOFFSET(mp);
5609                 } else {
5610                         prealloced = 0;
5611                         fixlen = ip->i_d.di_size;
5612                 }
5613         } else {
5614                 prealloced = 0;
5615                 fixlen = 1LL << 32;
5616         }
5617
5618         if (bmv->bmv_length == -1) {
5619                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5620                 bmv->bmv_length = MAX( (__int64_t)(fixlen - bmv->bmv_offset),
5621                                         (__int64_t)0);
5622         } else if (bmv->bmv_length < 0)
5623                 return XFS_ERROR(EINVAL);
5624         if (bmv->bmv_length == 0) {
5625                 bmv->bmv_entries = 0;
5626                 return 0;
5627         }
5628         nex = bmv->bmv_count - 1;
5629         if (nex <= 0)
5630                 return XFS_ERROR(EINVAL);
5631         bmvend = bmv->bmv_offset + bmv->bmv_length;
5632
5633         xfs_ilock(ip, XFS_IOLOCK_SHARED);
5634
5635         if (whichfork == XFS_DATA_FORK && ip->i_delayed_blks) {
5636                 /* xfs_fsize_t last_byte = xfs_file_last_byte(ip); */
5637                 VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1, 0, FI_REMAPF, error);
5638         }
5639
5640         ASSERT(whichfork == XFS_ATTR_FORK || ip->i_delayed_blks == 0);
5641
5642         lock = xfs_ilock_map_shared(ip);
5643
5644         /*
5645          * Don't let nex be bigger than the number of extents
5646          * we can have assuming alternating holes and real extents.
5647          */
5648         if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5649                 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5650
5651         bmapi_flags = XFS_BMAPI_AFLAG(whichfork) |
5652                         ((sh_unwritten) ? 0 : XFS_BMAPI_IGSTATE);
5653
5654         /*
5655          * Allocate enough space to handle "subnex" maps at a time.
5656          */
5657         subnex = 16;
5658         map = kmem_alloc(subnex * sizeof(*map), KM_SLEEP);
5659
5660         bmv->bmv_entries = 0;
5661
5662         if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0) {
5663                 error = 0;
5664                 goto unlock_and_return;
5665         }
5666
5667         nexleft = nex;
5668
5669         do {
5670                 nmap = (nexleft > subnex) ? subnex : nexleft;
5671                 error = xfs_bmapi(NULL, ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5672                                   XFS_BB_TO_FSB(mp, bmv->bmv_length),
5673                                   bmapi_flags, NULL, 0, map, &nmap, NULL);
5674                 if (error)
5675                         goto unlock_and_return;
5676                 ASSERT(nmap <= subnex);
5677
5678                 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5679                         nexleft--;
5680                         oflags = (map[i].br_state == XFS_EXT_UNWRITTEN) ?
5681                                         BMV_OF_PREALLOC : 0;
5682                         out.bmv_offset = XFS_FSB_TO_BB(mp, map[i].br_startoff);
5683                         out.bmv_length = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5684                         ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
5685                         if (prealloced &&
5686                             map[i].br_startblock == HOLESTARTBLOCK &&
5687                             out.bmv_offset + out.bmv_length == bmvend) {
5688                                 /*
5689                                  * came to hole at end of file
5690                                  */
5691                                 goto unlock_and_return;
5692                         } else {
5693                                 out.bmv_block =
5694                                     (map[i].br_startblock == HOLESTARTBLOCK) ?
5695                                         -1 :
5696                                         XFS_FSB_TO_DB(ip, map[i].br_startblock);
5697
5698                                 /* return either getbmap/getbmapx structure. */
5699                                 if (interface & BMV_IF_EXTENDED) {
5700                                         struct  getbmapx        outx;
5701
5702                                         GETBMAP_CONVERT(out,outx);
5703                                         outx.bmv_oflags = oflags;
5704                                         outx.bmv_unused1 = outx.bmv_unused2 = 0;
5705                                         if (copy_to_user(ap, &outx,
5706                                                         sizeof(outx))) {
5707                                                 error = XFS_ERROR(EFAULT);
5708                                                 goto unlock_and_return;
5709                                         }
5710                                 } else {
5711                                         if (copy_to_user(ap, &out,
5712                                                         sizeof(out))) {
5713                                                 error = XFS_ERROR(EFAULT);
5714                                                 goto unlock_and_return;
5715                                         }
5716                                 }
5717                                 bmv->bmv_offset =
5718                                         out.bmv_offset + out.bmv_length;
5719                                 bmv->bmv_length = MAX((__int64_t)0,
5720                                         (__int64_t)(bmvend - bmv->bmv_offset));
5721                                 bmv->bmv_entries++;
5722                                 ap = (interface & BMV_IF_EXTENDED) ?
5723                                                 (void __user *)
5724                                         ((struct getbmapx __user *)ap + 1) :
5725                                                 (void __user *)
5726                                         ((struct getbmap __user *)ap + 1);
5727                         }
5728                 }
5729         } while (nmap && nexleft && bmv->bmv_length);
5730
5731 unlock_and_return:
5732         xfs_iunlock_map_shared(ip, lock);
5733         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5734
5735         kmem_free(map, subnex * sizeof(*map));
5736
5737         return error;
5738 }
5739
5740 /*
5741  * Check the last inode extent to determine whether this allocation will result
5742  * in blocks being allocated at the end of the file. When we allocate new data
5743  * blocks at the end of the file which do not start at the previous data block,
5744  * we will try to align the new blocks at stripe unit boundaries.
5745  */
5746 STATIC int                              /* error */
5747 xfs_bmap_isaeof(
5748         xfs_inode_t     *ip,            /* incore inode pointer */
5749         xfs_fileoff_t   off,            /* file offset in fsblocks */
5750         int             whichfork,      /* data or attribute fork */
5751         char            *aeof)          /* return value */
5752 {
5753         int             error;          /* error return value */
5754         xfs_ifork_t     *ifp;           /* inode fork pointer */
5755         xfs_bmbt_rec_t  *lastrec;       /* extent list entry pointer */
5756         xfs_extnum_t    nextents;       /* size of extent list */
5757         xfs_bmbt_irec_t s;              /* expanded extent list entry */
5758
5759         ASSERT(whichfork == XFS_DATA_FORK);
5760         ifp = XFS_IFORK_PTR(ip, whichfork);
5761         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5762             (error = xfs_iread_extents(NULL, ip, whichfork)))
5763                 return error;
5764         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5765         if (nextents == 0) {
5766                 *aeof = 1;
5767                 return 0;
5768         }
5769         /*
5770          * Go to the last extent
5771          */
5772         lastrec = &ifp->if_u1.if_extents[nextents - 1];
5773         xfs_bmbt_get_all(lastrec, &s);
5774         /*
5775          * Check we are allocating in the last extent (for delayed allocations)
5776          * or past the last extent for non-delayed allocations.
5777          */
5778         *aeof = (off >= s.br_startoff &&
5779                  off < s.br_startoff + s.br_blockcount &&
5780                  ISNULLSTARTBLOCK(s.br_startblock)) ||
5781                 off >= s.br_startoff + s.br_blockcount;
5782         return 0;
5783 }
5784
5785 /*
5786  * Check if the endoff is outside the last extent. If so the caller will grow
5787  * the allocation to a stripe unit boundary.
5788  */
5789 int                                     /* error */
5790 xfs_bmap_eof(
5791         xfs_inode_t     *ip,            /* incore inode pointer */
5792         xfs_fileoff_t   endoff,         /* file offset in fsblocks */
5793         int             whichfork,      /* data or attribute fork */
5794         int             *eof)           /* result value */
5795 {
5796         xfs_fsblock_t   blockcount;     /* extent block count */
5797         int             error;          /* error return value */
5798         xfs_ifork_t     *ifp;           /* inode fork pointer */
5799         xfs_bmbt_rec_t  *lastrec;       /* extent list entry pointer */
5800         xfs_extnum_t    nextents;       /* size of extent list */
5801         xfs_fileoff_t   startoff;       /* extent starting file offset */
5802
5803         ASSERT(whichfork == XFS_DATA_FORK);
5804         ifp = XFS_IFORK_PTR(ip, whichfork);
5805         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5806             (error = xfs_iread_extents(NULL, ip, whichfork)))
5807                 return error;
5808         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5809         if (nextents == 0) {
5810                 *eof = 1;
5811                 return 0;
5812         }
5813         /*
5814          * Go to the last extent
5815          */
5816         lastrec = &ifp->if_u1.if_extents[nextents - 1];
5817         startoff = xfs_bmbt_get_startoff(lastrec);
5818         blockcount = xfs_bmbt_get_blockcount(lastrec);
5819         *eof = endoff >= startoff + blockcount;
5820         return 0;
5821 }
5822
5823 #ifdef DEBUG
5824 /*
5825  * Check that the extents list for the inode ip is in the right order.
5826  */
5827 STATIC void
5828 xfs_bmap_check_extents(
5829         xfs_inode_t             *ip,            /* incore inode pointer */
5830         int                     whichfork)      /* data or attr fork */
5831 {
5832         xfs_bmbt_rec_t          *base;          /* base of extents list */
5833         xfs_bmbt_rec_t          *ep;            /* current extent entry */
5834         xfs_ifork_t             *ifp;           /* inode fork pointer */
5835         xfs_extnum_t            nextents;       /* number of extents in list */
5836
5837         ifp = XFS_IFORK_PTR(ip, whichfork);
5838         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
5839         base = ifp->if_u1.if_extents;
5840         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5841         for (ep = base; ep < &base[nextents - 1]; ep++) {
5842                 xfs_btree_check_rec(XFS_BTNUM_BMAP, (void *)ep,
5843                         (void *)(ep + 1));
5844         }
5845 }
5846
5847 STATIC
5848 xfs_buf_t *
5849 xfs_bmap_get_bp(
5850         xfs_btree_cur_t         *cur,
5851         xfs_fsblock_t           bno)
5852 {
5853         int i;
5854         xfs_buf_t *bp;
5855
5856         if (!cur)
5857                 return(NULL);
5858
5859         bp = NULL;
5860         for(i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
5861                 bp = cur->bc_bufs[i];
5862                 if (!bp) break;
5863                 if (XFS_BUF_ADDR(bp) == bno)
5864                         break;  /* Found it */
5865         }
5866         if (i == XFS_BTREE_MAXLEVELS)
5867                 bp = NULL;
5868
5869         if (!bp) { /* Chase down all the log items to see if the bp is there */
5870                 xfs_log_item_chunk_t    *licp;
5871                 xfs_trans_t             *tp;
5872
5873                 tp = cur->bc_tp;
5874                 licp = &tp->t_items;
5875                 while (!bp && licp != NULL) {
5876                         if (XFS_LIC_ARE_ALL_FREE(licp)) {
5877                                 licp = licp->lic_next;
5878                                 continue;
5879                         }
5880                         for (i = 0; i < licp->lic_unused; i++) {
5881                                 xfs_log_item_desc_t     *lidp;
5882                                 xfs_log_item_t          *lip;
5883                                 xfs_buf_log_item_t      *bip;
5884                                 xfs_buf_t               *lbp;
5885
5886                                 if (XFS_LIC_ISFREE(licp, i)) {
5887                                         continue;
5888                                 }
5889
5890                                 lidp = XFS_LIC_SLOT(licp, i);
5891                                 lip = lidp->lid_item;
5892                                 if (lip->li_type != XFS_LI_BUF)
5893                                         continue;
5894
5895                                 bip = (xfs_buf_log_item_t *)lip;
5896                                 lbp = bip->bli_buf;
5897
5898                                 if (XFS_BUF_ADDR(lbp) == bno) {
5899                                         bp = lbp;
5900                                         break; /* Found it */
5901                                 }
5902                         }
5903                         licp = licp->lic_next;
5904                 }
5905         }
5906         return(bp);
5907 }
5908
5909 void
5910 xfs_check_block(
5911         xfs_bmbt_block_t        *block,
5912         xfs_mount_t             *mp,
5913         int                     root,
5914         short                   sz)
5915 {
5916         int                     i, j, dmxr;
5917         xfs_bmbt_ptr_t          *pp, *thispa;   /* pointer to block address */
5918         xfs_bmbt_key_t          *prevp, *keyp;
5919
5920         ASSERT(INT_GET(block->bb_level, ARCH_CONVERT) > 0);
5921
5922         prevp = NULL;
5923         for( i = 1; i <= INT_GET(block->bb_numrecs, ARCH_CONVERT);i++) {
5924                 dmxr = mp->m_bmap_dmxr[0];
5925
5926                 if (root) {
5927                         keyp = XFS_BMAP_BROOT_KEY_ADDR(block, i, sz);
5928                 } else {
5929                         keyp = XFS_BTREE_KEY_ADDR(mp->m_sb.sb_blocksize,
5930                                 xfs_bmbt, block, i, dmxr);
5931                 }
5932
5933                 if (prevp) {
5934                         xfs_btree_check_key(XFS_BTNUM_BMAP, prevp, keyp);
5935                 }
5936                 prevp = keyp;
5937
5938                 /*
5939                  * Compare the block numbers to see if there are dups.
5940                  */
5941
5942                 if (root) {
5943                         pp = XFS_BMAP_BROOT_PTR_ADDR(block, i, sz);
5944                 } else {
5945                         pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize,
5946                                 xfs_bmbt, block, i, dmxr);
5947                 }
5948                 for (j = i+1; j <= INT_GET(block->bb_numrecs, ARCH_CONVERT); j++) {
5949                         if (root) {
5950                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(block, j, sz);
5951                         } else {
5952                                 thispa = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize,
5953                                         xfs_bmbt, block, j, dmxr);
5954                         }
5955                         if (INT_GET(*thispa, ARCH_CONVERT) ==
5956                             INT_GET(*pp, ARCH_CONVERT)) {
5957                                 cmn_err(CE_WARN, "%s: thispa(%d) == pp(%d) %Ld",
5958                                         __FUNCTION__, j, i,
5959                                         INT_GET(*thispa, ARCH_CONVERT));
5960                                 panic("%s: ptrs are equal in node\n",
5961                                         __FUNCTION__);
5962                         }
5963                 }
5964         }
5965 }
5966
5967 /*
5968  * Check that the extents for the inode ip are in the right order in all
5969  * btree leaves.
5970  */
5971
5972 STATIC void
5973 xfs_bmap_check_leaf_extents(
5974         xfs_btree_cur_t         *cur,   /* btree cursor or null */
5975         xfs_inode_t             *ip,            /* incore inode pointer */
5976         int                     whichfork)      /* data or attr fork */
5977 {
5978         xfs_bmbt_block_t        *block; /* current btree block */
5979         xfs_fsblock_t           bno;    /* block # of "block" */
5980         xfs_buf_t               *bp;    /* buffer for "block" */
5981         int                     error;  /* error return value */
5982         xfs_extnum_t            i=0;    /* index into the extents list */
5983         xfs_ifork_t             *ifp;   /* fork structure */
5984         int                     level;  /* btree level, for checking */
5985         xfs_mount_t             *mp;    /* file system mount structure */
5986         xfs_bmbt_ptr_t          *pp;    /* pointer to block address */
5987         xfs_bmbt_rec_t          *ep, *lastp;    /* extent pointers in block entry */
5988         int                     bp_release = 0;
5989
5990         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
5991                 return;
5992         }
5993
5994         bno = NULLFSBLOCK;
5995         mp = ip->i_mount;
5996         ifp = XFS_IFORK_PTR(ip, whichfork);
5997         block = ifp->if_broot;
5998         /*
5999          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6000          */
6001         ASSERT(INT_GET(block->bb_level, ARCH_CONVERT) > 0);
6002         level = INT_GET(block->bb_level, ARCH_CONVERT);
6003         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
6004         pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
6005         ASSERT(INT_GET(*pp, ARCH_CONVERT) != NULLDFSBNO);
6006         ASSERT(XFS_FSB_TO_AGNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agcount);
6007         ASSERT(XFS_FSB_TO_AGBNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agblocks);
6008         bno = INT_GET(*pp, ARCH_CONVERT);
6009         /*
6010          * Go down the tree until leaf level is reached, following the first
6011          * pointer (leftmost) at each level.
6012          */
6013         while (level-- > 0) {
6014                 /* See if buf is in cur first */
6015                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
6016                 if (bp) {
6017                         bp_release = 0;
6018                 } else {
6019                         bp_release = 1;
6020                 }
6021                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
6022                                 XFS_BMAP_BTREE_REF)))
6023                         goto error_norelse;
6024                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6025                 XFS_WANT_CORRUPTED_GOTO(
6026                         XFS_BMAP_SANITY_CHECK(mp, block, level),
6027                         error0);
6028                 if (level == 0)
6029                         break;
6030
6031                 /*
6032                  * Check this block for basic sanity (increasing keys and
6033                  * no duplicate blocks).
6034                  */
6035
6036                 xfs_check_block(block, mp, 0, 0);
6037                 pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt, block,
6038                         1, mp->m_bmap_dmxr[1]);
6039                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, INT_GET(*pp, ARCH_CONVERT)), error0);
6040                 bno = INT_GET(*pp, ARCH_CONVERT);
6041                 if (bp_release) {
6042                         bp_release = 0;
6043                         xfs_trans_brelse(NULL, bp);
6044                 }
6045         }
6046
6047         /*
6048          * Here with bp and block set to the leftmost leaf node in the tree.
6049          */
6050         i = 0;
6051
6052         /*
6053          * Loop over all leaf nodes checking that all extents are in the right order.
6054          */
6055         lastp = NULL;
6056         for (;;) {
6057                 xfs_bmbt_rec_t  *frp;
6058                 xfs_fsblock_t   nextbno;
6059                 xfs_extnum_t    num_recs;
6060
6061
6062                 num_recs = INT_GET(block->bb_numrecs, ARCH_CONVERT);
6063
6064                 /*
6065                  * Read-ahead the next leaf block, if any.
6066                  */
6067
6068                 nextbno = INT_GET(block->bb_rightsib, ARCH_CONVERT);
6069
6070                 /*
6071                  * Check all the extents to make sure they are OK.
6072                  * If we had a previous block, the last entry should
6073                  * conform with the first entry in this one.
6074                  */
6075
6076                 frp = XFS_BTREE_REC_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt,
6077                         block, 1, mp->m_bmap_dmxr[0]);
6078
6079                 for (ep = frp;ep < frp + (num_recs - 1); ep++) {
6080                         if (lastp) {
6081                                 xfs_btree_check_rec(XFS_BTNUM_BMAP,
6082                                         (void *)lastp, (void *)ep);
6083                         }
6084                         xfs_btree_check_rec(XFS_BTNUM_BMAP, (void *)ep,
6085                                 (void *)(ep + 1));
6086                 }
6087                 lastp = frp + num_recs - 1; /* For the next iteration */
6088
6089                 i += num_recs;
6090                 if (bp_release) {
6091                         bp_release = 0;
6092                         xfs_trans_brelse(NULL, bp);
6093                 }
6094                 bno = nextbno;
6095                 /*
6096                  * If we've reached the end, stop.
6097                  */
6098                 if (bno == NULLFSBLOCK)
6099                         break;
6100
6101                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
6102                 if (bp) {
6103                         bp_release = 0;
6104                 } else {
6105                         bp_release = 1;
6106                 }
6107                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
6108                                 XFS_BMAP_BTREE_REF)))
6109                         goto error_norelse;
6110                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6111         }
6112         if (bp_release) {
6113                 bp_release = 0;
6114                 xfs_trans_brelse(NULL, bp);
6115         }
6116         return;
6117
6118 error0:
6119         cmn_err(CE_WARN, "%s: at error0", __FUNCTION__);
6120         if (bp_release)
6121                 xfs_trans_brelse(NULL, bp);
6122 error_norelse:
6123         cmn_err(CE_WARN, "%s: BAD after btree leaves for %d extents",
6124                 i, __FUNCTION__);
6125         panic("%s: CORRUPTED BTREE OR SOMETHING", __FUNCTION__);
6126         return;
6127 }
6128 #endif
6129
6130 /*
6131  * Count fsblocks of the given fork.
6132  */
6133 int                                             /* error */
6134 xfs_bmap_count_blocks(
6135         xfs_trans_t             *tp,            /* transaction pointer */
6136         xfs_inode_t             *ip,            /* incore inode */
6137         int                     whichfork,      /* data or attr fork */
6138         int                     *count)         /* out: count of blocks */
6139 {
6140         xfs_bmbt_block_t        *block; /* current btree block */
6141         xfs_fsblock_t           bno;    /* block # of "block" */
6142         xfs_ifork_t             *ifp;   /* fork structure */
6143         int                     level;  /* btree level, for checking */
6144         xfs_mount_t             *mp;    /* file system mount structure */
6145         xfs_bmbt_ptr_t          *pp;    /* pointer to block address */
6146
6147         bno = NULLFSBLOCK;
6148         mp = ip->i_mount;
6149         ifp = XFS_IFORK_PTR(ip, whichfork);
6150         if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
6151                 if (unlikely(xfs_bmap_count_leaves(ifp->if_u1.if_extents,
6152                         ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
6153                         count) < 0)) {
6154                         XFS_ERROR_REPORT("xfs_bmap_count_blocks(1)",
6155                                          XFS_ERRLEVEL_LOW, mp);
6156                         return XFS_ERROR(EFSCORRUPTED);
6157                 }
6158                 return 0;
6159         }
6160
6161         /*
6162          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6163          */
6164         block = ifp->if_broot;
6165         ASSERT(INT_GET(block->bb_level, ARCH_CONVERT) > 0);
6166         level = INT_GET(block->bb_level, ARCH_CONVERT);
6167         pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
6168         ASSERT(INT_GET(*pp, ARCH_CONVERT) != NULLDFSBNO);
6169         ASSERT(XFS_FSB_TO_AGNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agcount);
6170         ASSERT(XFS_FSB_TO_AGBNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agblocks);
6171         bno = INT_GET(*pp, ARCH_CONVERT);
6172
6173         if (unlikely(xfs_bmap_count_tree(mp, tp, bno, level, count) < 0)) {
6174                 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
6175                                  mp);
6176                 return XFS_ERROR(EFSCORRUPTED);
6177         }
6178
6179         return 0;
6180 }
6181
6182 /*
6183  * Recursively walks each level of a btree
6184  * to count total fsblocks is use.
6185  */
6186 int                                     /* error */
6187 xfs_bmap_count_tree(
6188         xfs_mount_t     *mp,            /* file system mount point */
6189         xfs_trans_t     *tp,            /* transaction pointer */
6190         xfs_fsblock_t   blockno,        /* file system block number */
6191         int             levelin,        /* level in btree */
6192         int             *count)         /* Count of blocks */
6193 {
6194         int                     error;
6195         xfs_buf_t               *bp, *nbp;
6196         int                     level = levelin;
6197         xfs_bmbt_ptr_t          *pp;
6198         xfs_fsblock_t           bno = blockno;
6199         xfs_fsblock_t           nextbno;
6200         xfs_bmbt_block_t        *block, *nextblock;
6201         int                     numrecs;
6202         xfs_bmbt_rec_t          *frp;
6203
6204         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
6205                 return error;
6206         *count += 1;
6207         block = XFS_BUF_TO_BMBT_BLOCK(bp);
6208
6209         if (--level) {
6210                 /* Not at node above leafs, count this level of nodes */
6211                 nextbno = INT_GET(block->bb_rightsib, ARCH_CONVERT);
6212                 while (nextbno != NULLFSBLOCK) {
6213                         if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
6214                                 0, &nbp, XFS_BMAP_BTREE_REF)))
6215                                 return error;
6216                         *count += 1;
6217                         nextblock = XFS_BUF_TO_BMBT_BLOCK(nbp);
6218                         nextbno = INT_GET(nextblock->bb_rightsib, ARCH_CONVERT);
6219                         xfs_trans_brelse(tp, nbp);
6220                 }
6221
6222                 /* Dive to the next level */
6223                 pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize,
6224                         xfs_bmbt, block, 1, mp->m_bmap_dmxr[1]);
6225                 bno = INT_GET(*pp, ARCH_CONVERT);
6226                 if (unlikely((error =
6227                      xfs_bmap_count_tree(mp, tp, bno, level, count)) < 0)) {
6228                         xfs_trans_brelse(tp, bp);
6229                         XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6230                                          XFS_ERRLEVEL_LOW, mp);
6231                         return XFS_ERROR(EFSCORRUPTED);
6232                 }
6233                 xfs_trans_brelse(tp, bp);
6234         } else {
6235                 /* count all level 1 nodes and their leaves */
6236                 for (;;) {
6237                         nextbno = INT_GET(block->bb_rightsib, ARCH_CONVERT);
6238                         numrecs = INT_GET(block->bb_numrecs, ARCH_CONVERT);
6239                         frp = XFS_BTREE_REC_ADDR(mp->m_sb.sb_blocksize,
6240                                 xfs_bmbt, block, 1, mp->m_bmap_dmxr[0]);
6241                         if (unlikely(xfs_bmap_count_leaves(frp, numrecs, count) < 0)) {
6242                                 xfs_trans_brelse(tp, bp);
6243                                 XFS_ERROR_REPORT("xfs_bmap_count_tree(2)",
6244                                                  XFS_ERRLEVEL_LOW, mp);
6245                                 return XFS_ERROR(EFSCORRUPTED);
6246                         }
6247                         xfs_trans_brelse(tp, bp);
6248                         if (nextbno == NULLFSBLOCK)
6249                                 break;
6250                         bno = nextbno;
6251                         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6252                                 XFS_BMAP_BTREE_REF)))
6253                                 return error;
6254                         *count += 1;
6255                         block = XFS_BUF_TO_BMBT_BLOCK(bp);
6256                 }
6257         }
6258         return 0;
6259 }
6260
6261 /*
6262  * Count leaf blocks given a pointer to an extent list.
6263  */
6264 int
6265 xfs_bmap_count_leaves(
6266         xfs_bmbt_rec_t          *frp,
6267         int                     numrecs,
6268         int                     *count)
6269 {
6270         int             b;
6271
6272         for ( b = 1; b <= numrecs; b++, frp++)
6273                 *count += xfs_bmbt_disk_get_blockcount(frp);
6274         return 0;
6275 }