[XFS] implement generic xfs_btree_lookup
[safe/jmp/linux-2.6] / fs / xfs / xfs_bmap.c
1 /*
2  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write the Free Software Foundation,
16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17  */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_inum.h"
24 #include "xfs_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_dir2.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_alloc_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_dir2_sf.h"
33 #include "xfs_attr_sf.h"
34 #include "xfs_dinode.h"
35 #include "xfs_inode.h"
36 #include "xfs_btree.h"
37 #include "xfs_dmapi.h"
38 #include "xfs_mount.h"
39 #include "xfs_ialloc.h"
40 #include "xfs_itable.h"
41 #include "xfs_dir2_data.h"
42 #include "xfs_dir2_leaf.h"
43 #include "xfs_dir2_block.h"
44 #include "xfs_inode_item.h"
45 #include "xfs_extfree_item.h"
46 #include "xfs_alloc.h"
47 #include "xfs_bmap.h"
48 #include "xfs_rtalloc.h"
49 #include "xfs_error.h"
50 #include "xfs_attr_leaf.h"
51 #include "xfs_rw.h"
52 #include "xfs_quota.h"
53 #include "xfs_trans_space.h"
54 #include "xfs_buf_item.h"
55 #include "xfs_filestream.h"
56 #include "xfs_vnodeops.h"
57
58
59 #ifdef DEBUG
60 STATIC void
61 xfs_bmap_check_leaf_extents(xfs_btree_cur_t *cur, xfs_inode_t *ip, int whichfork);
62 #endif
63
64 kmem_zone_t             *xfs_bmap_free_item_zone;
65
66 /*
67  * Prototypes for internal bmap routines.
68  */
69
70
71 /*
72  * Called from xfs_bmap_add_attrfork to handle extents format files.
73  */
74 STATIC int                                      /* error */
75 xfs_bmap_add_attrfork_extents(
76         xfs_trans_t             *tp,            /* transaction pointer */
77         xfs_inode_t             *ip,            /* incore inode pointer */
78         xfs_fsblock_t           *firstblock,    /* first block allocated */
79         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
80         int                     *flags);        /* inode logging flags */
81
82 /*
83  * Called from xfs_bmap_add_attrfork to handle local format files.
84  */
85 STATIC int                                      /* error */
86 xfs_bmap_add_attrfork_local(
87         xfs_trans_t             *tp,            /* transaction pointer */
88         xfs_inode_t             *ip,            /* incore inode pointer */
89         xfs_fsblock_t           *firstblock,    /* first block allocated */
90         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
91         int                     *flags);        /* inode logging flags */
92
93 /*
94  * Called by xfs_bmapi to update file extent records and the btree
95  * after allocating space (or doing a delayed allocation).
96  */
97 STATIC int                              /* error */
98 xfs_bmap_add_extent(
99         xfs_inode_t             *ip,    /* incore inode pointer */
100         xfs_extnum_t            idx,    /* extent number to update/insert */
101         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
102         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
103         xfs_fsblock_t           *first, /* pointer to firstblock variable */
104         xfs_bmap_free_t         *flist, /* list of extents to be freed */
105         int                     *logflagsp, /* inode logging flags */
106         xfs_extdelta_t          *delta, /* Change made to incore extents */
107         int                     whichfork, /* data or attr fork */
108         int                     rsvd);  /* OK to allocate reserved blocks */
109
110 /*
111  * Called by xfs_bmap_add_extent to handle cases converting a delayed
112  * allocation to a real allocation.
113  */
114 STATIC int                              /* error */
115 xfs_bmap_add_extent_delay_real(
116         xfs_inode_t             *ip,    /* incore inode pointer */
117         xfs_extnum_t            idx,    /* extent number to update/insert */
118         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
119         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
120         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
121         xfs_fsblock_t           *first, /* pointer to firstblock variable */
122         xfs_bmap_free_t         *flist, /* list of extents to be freed */
123         int                     *logflagsp, /* inode logging flags */
124         xfs_extdelta_t          *delta, /* Change made to incore extents */
125         int                     rsvd);  /* OK to allocate reserved blocks */
126
127 /*
128  * Called by xfs_bmap_add_extent to handle cases converting a hole
129  * to a delayed allocation.
130  */
131 STATIC int                              /* error */
132 xfs_bmap_add_extent_hole_delay(
133         xfs_inode_t             *ip,    /* incore inode pointer */
134         xfs_extnum_t            idx,    /* extent number to update/insert */
135         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
136         int                     *logflagsp,/* inode logging flags */
137         xfs_extdelta_t          *delta, /* Change made to incore extents */
138         int                     rsvd);  /* OK to allocate reserved blocks */
139
140 /*
141  * Called by xfs_bmap_add_extent to handle cases converting a hole
142  * to a real allocation.
143  */
144 STATIC int                              /* error */
145 xfs_bmap_add_extent_hole_real(
146         xfs_inode_t             *ip,    /* incore inode pointer */
147         xfs_extnum_t            idx,    /* extent number to update/insert */
148         xfs_btree_cur_t         *cur,   /* if null, not a btree */
149         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
150         int                     *logflagsp, /* inode logging flags */
151         xfs_extdelta_t          *delta, /* Change made to incore extents */
152         int                     whichfork); /* data or attr fork */
153
154 /*
155  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
156  * allocation to a real allocation or vice versa.
157  */
158 STATIC int                              /* error */
159 xfs_bmap_add_extent_unwritten_real(
160         xfs_inode_t             *ip,    /* incore inode pointer */
161         xfs_extnum_t            idx,    /* extent number to update/insert */
162         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
163         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
164         int                     *logflagsp, /* inode logging flags */
165         xfs_extdelta_t          *delta); /* Change made to incore extents */
166
167 /*
168  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
169  * It figures out where to ask the underlying allocator to put the new extent.
170  */
171 STATIC int                              /* error */
172 xfs_bmap_alloc(
173         xfs_bmalloca_t          *ap);   /* bmap alloc argument struct */
174
175 /*
176  * Transform a btree format file with only one leaf node, where the
177  * extents list will fit in the inode, into an extents format file.
178  * Since the file extents are already in-core, all we have to do is
179  * give up the space for the btree root and pitch the leaf block.
180  */
181 STATIC int                              /* error */
182 xfs_bmap_btree_to_extents(
183         xfs_trans_t             *tp,    /* transaction pointer */
184         xfs_inode_t             *ip,    /* incore inode pointer */
185         xfs_btree_cur_t         *cur,   /* btree cursor */
186         int                     *logflagsp, /* inode logging flags */
187         int                     whichfork); /* data or attr fork */
188
189 /*
190  * Called by xfs_bmapi to update file extent records and the btree
191  * after removing space (or undoing a delayed allocation).
192  */
193 STATIC int                              /* error */
194 xfs_bmap_del_extent(
195         xfs_inode_t             *ip,    /* incore inode pointer */
196         xfs_trans_t             *tp,    /* current trans pointer */
197         xfs_extnum_t            idx,    /* extent number to update/insert */
198         xfs_bmap_free_t         *flist, /* list of extents to be freed */
199         xfs_btree_cur_t         *cur,   /* if null, not a btree */
200         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
201         int                     *logflagsp,/* inode logging flags */
202         xfs_extdelta_t          *delta, /* Change made to incore extents */
203         int                     whichfork, /* data or attr fork */
204         int                     rsvd);   /* OK to allocate reserved blocks */
205
206 /*
207  * Remove the entry "free" from the free item list.  Prev points to the
208  * previous entry, unless "free" is the head of the list.
209  */
210 STATIC void
211 xfs_bmap_del_free(
212         xfs_bmap_free_t         *flist, /* free item list header */
213         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
214         xfs_bmap_free_item_t    *free); /* list item to be freed */
215
216 /*
217  * Convert an extents-format file into a btree-format file.
218  * The new file will have a root block (in the inode) and a single child block.
219  */
220 STATIC int                                      /* error */
221 xfs_bmap_extents_to_btree(
222         xfs_trans_t             *tp,            /* transaction pointer */
223         xfs_inode_t             *ip,            /* incore inode pointer */
224         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
225         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
226         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
227         int                     wasdel,         /* converting a delayed alloc */
228         int                     *logflagsp,     /* inode logging flags */
229         int                     whichfork);     /* data or attr fork */
230
231 /*
232  * Convert a local file to an extents file.
233  * This code is sort of bogus, since the file data needs to get
234  * logged so it won't be lost.  The bmap-level manipulations are ok, though.
235  */
236 STATIC int                              /* error */
237 xfs_bmap_local_to_extents(
238         xfs_trans_t     *tp,            /* transaction pointer */
239         xfs_inode_t     *ip,            /* incore inode pointer */
240         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
241         xfs_extlen_t    total,          /* total blocks needed by transaction */
242         int             *logflagsp,     /* inode logging flags */
243         int             whichfork);     /* data or attr fork */
244
245 /*
246  * Search the extents list for the inode, for the extent containing bno.
247  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
248  * *eofp will be set, and *prevp will contain the last entry (null if none).
249  * Else, *lastxp will be set to the index of the found
250  * entry; *gotp will contain the entry.
251  */
252 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
253 xfs_bmap_search_extents(
254         xfs_inode_t     *ip,            /* incore inode pointer */
255         xfs_fileoff_t   bno,            /* block number searched for */
256         int             whichfork,      /* data or attr fork */
257         int             *eofp,          /* out: end of file found */
258         xfs_extnum_t    *lastxp,        /* out: last extent index */
259         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
260         xfs_bmbt_irec_t *prevp);        /* out: previous extent entry found */
261
262 /*
263  * Check the last inode extent to determine whether this allocation will result
264  * in blocks being allocated at the end of the file. When we allocate new data
265  * blocks at the end of the file which do not start at the previous data block,
266  * we will try to align the new blocks at stripe unit boundaries.
267  */
268 STATIC int                              /* error */
269 xfs_bmap_isaeof(
270         xfs_inode_t     *ip,            /* incore inode pointer */
271         xfs_fileoff_t   off,            /* file offset in fsblocks */
272         int             whichfork,      /* data or attribute fork */
273         char            *aeof);         /* return value */
274
275 #ifdef XFS_BMAP_TRACE
276 /*
277  * Add bmap trace entry prior to a call to xfs_iext_remove.
278  */
279 STATIC void
280 xfs_bmap_trace_delete(
281         const char      *fname,         /* function name */
282         char            *desc,          /* operation description */
283         xfs_inode_t     *ip,            /* incore inode pointer */
284         xfs_extnum_t    idx,            /* index of entry(entries) deleted */
285         xfs_extnum_t    cnt,            /* count of entries deleted, 1 or 2 */
286         int             whichfork);     /* data or attr fork */
287
288 /*
289  * Add bmap trace entry prior to a call to xfs_iext_insert, or
290  * reading in the extents list from the disk (in the btree).
291  */
292 STATIC void
293 xfs_bmap_trace_insert(
294         const char      *fname,         /* function name */
295         char            *desc,          /* operation description */
296         xfs_inode_t     *ip,            /* incore inode pointer */
297         xfs_extnum_t    idx,            /* index of entry(entries) inserted */
298         xfs_extnum_t    cnt,            /* count of entries inserted, 1 or 2 */
299         xfs_bmbt_irec_t *r1,            /* inserted record 1 */
300         xfs_bmbt_irec_t *r2,            /* inserted record 2 or null */
301         int             whichfork);     /* data or attr fork */
302
303 /*
304  * Add bmap trace entry after updating an extent record in place.
305  */
306 STATIC void
307 xfs_bmap_trace_post_update(
308         const char      *fname,         /* function name */
309         char            *desc,          /* operation description */
310         xfs_inode_t     *ip,            /* incore inode pointer */
311         xfs_extnum_t    idx,            /* index of entry updated */
312         int             whichfork);     /* data or attr fork */
313
314 /*
315  * Add bmap trace entry prior to updating an extent record in place.
316  */
317 STATIC void
318 xfs_bmap_trace_pre_update(
319         const char      *fname,         /* function name */
320         char            *desc,          /* operation description */
321         xfs_inode_t     *ip,            /* incore inode pointer */
322         xfs_extnum_t    idx,            /* index of entry to be updated */
323         int             whichfork);     /* data or attr fork */
324
325 #define XFS_BMAP_TRACE_DELETE(d,ip,i,c,w)       \
326         xfs_bmap_trace_delete(__func__,d,ip,i,c,w)
327 #define XFS_BMAP_TRACE_INSERT(d,ip,i,c,r1,r2,w) \
328         xfs_bmap_trace_insert(__func__,d,ip,i,c,r1,r2,w)
329 #define XFS_BMAP_TRACE_POST_UPDATE(d,ip,i,w)    \
330         xfs_bmap_trace_post_update(__func__,d,ip,i,w)
331 #define XFS_BMAP_TRACE_PRE_UPDATE(d,ip,i,w)     \
332         xfs_bmap_trace_pre_update(__func__,d,ip,i,w)
333 #else
334 #define XFS_BMAP_TRACE_DELETE(d,ip,i,c,w)
335 #define XFS_BMAP_TRACE_INSERT(d,ip,i,c,r1,r2,w)
336 #define XFS_BMAP_TRACE_POST_UPDATE(d,ip,i,w)
337 #define XFS_BMAP_TRACE_PRE_UPDATE(d,ip,i,w)
338 #endif  /* XFS_BMAP_TRACE */
339
340 /*
341  * Compute the worst-case number of indirect blocks that will be used
342  * for ip's delayed extent of length "len".
343  */
344 STATIC xfs_filblks_t
345 xfs_bmap_worst_indlen(
346         xfs_inode_t             *ip,    /* incore inode pointer */
347         xfs_filblks_t           len);   /* delayed extent length */
348
349 #ifdef DEBUG
350 /*
351  * Perform various validation checks on the values being returned
352  * from xfs_bmapi().
353  */
354 STATIC void
355 xfs_bmap_validate_ret(
356         xfs_fileoff_t           bno,
357         xfs_filblks_t           len,
358         int                     flags,
359         xfs_bmbt_irec_t         *mval,
360         int                     nmap,
361         int                     ret_nmap);
362 #else
363 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
364 #endif /* DEBUG */
365
366 #if defined(XFS_RW_TRACE)
367 STATIC void
368 xfs_bunmap_trace(
369         xfs_inode_t             *ip,
370         xfs_fileoff_t           bno,
371         xfs_filblks_t           len,
372         int                     flags,
373         inst_t                  *ra);
374 #else
375 #define xfs_bunmap_trace(ip, bno, len, flags, ra)
376 #endif  /* XFS_RW_TRACE */
377
378 STATIC int
379 xfs_bmap_count_tree(
380         xfs_mount_t     *mp,
381         xfs_trans_t     *tp,
382         xfs_ifork_t     *ifp,
383         xfs_fsblock_t   blockno,
384         int             levelin,
385         int             *count);
386
387 STATIC void
388 xfs_bmap_count_leaves(
389         xfs_ifork_t             *ifp,
390         xfs_extnum_t            idx,
391         int                     numrecs,
392         int                     *count);
393
394 STATIC void
395 xfs_bmap_disk_count_leaves(
396         xfs_extnum_t            idx,
397         xfs_bmbt_block_t        *block,
398         int                     numrecs,
399         int                     *count);
400
401 /*
402  * Bmap internal routines.
403  */
404
405 STATIC int                              /* error */
406 xfs_bmbt_lookup_eq(
407         struct xfs_btree_cur    *cur,
408         xfs_fileoff_t           off,
409         xfs_fsblock_t           bno,
410         xfs_filblks_t           len,
411         int                     *stat)  /* success/failure */
412 {
413         cur->bc_rec.b.br_startoff = off;
414         cur->bc_rec.b.br_startblock = bno;
415         cur->bc_rec.b.br_blockcount = len;
416         return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
417 }
418
419 STATIC int                              /* error */
420 xfs_bmbt_lookup_ge(
421         struct xfs_btree_cur    *cur,
422         xfs_fileoff_t           off,
423         xfs_fsblock_t           bno,
424         xfs_filblks_t           len,
425         int                     *stat)  /* success/failure */
426 {
427         cur->bc_rec.b.br_startoff = off;
428         cur->bc_rec.b.br_startblock = bno;
429         cur->bc_rec.b.br_blockcount = len;
430         return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
431 }
432
433
434 /*
435  * Called from xfs_bmap_add_attrfork to handle btree format files.
436  */
437 STATIC int                                      /* error */
438 xfs_bmap_add_attrfork_btree(
439         xfs_trans_t             *tp,            /* transaction pointer */
440         xfs_inode_t             *ip,            /* incore inode pointer */
441         xfs_fsblock_t           *firstblock,    /* first block allocated */
442         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
443         int                     *flags)         /* inode logging flags */
444 {
445         xfs_btree_cur_t         *cur;           /* btree cursor */
446         int                     error;          /* error return value */
447         xfs_mount_t             *mp;            /* file system mount struct */
448         int                     stat;           /* newroot status */
449
450         mp = ip->i_mount;
451         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
452                 *flags |= XFS_ILOG_DBROOT;
453         else {
454                 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
455                 cur->bc_private.b.flist = flist;
456                 cur->bc_private.b.firstblock = *firstblock;
457                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
458                         goto error0;
459                 /* must be at least one entry */
460                 XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
461                 if ((error = xfs_bmbt_newroot(cur, flags, &stat)))
462                         goto error0;
463                 if (stat == 0) {
464                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
465                         return XFS_ERROR(ENOSPC);
466                 }
467                 *firstblock = cur->bc_private.b.firstblock;
468                 cur->bc_private.b.allocated = 0;
469                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
470         }
471         return 0;
472 error0:
473         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
474         return error;
475 }
476
477 /*
478  * Called from xfs_bmap_add_attrfork to handle extents format files.
479  */
480 STATIC int                                      /* error */
481 xfs_bmap_add_attrfork_extents(
482         xfs_trans_t             *tp,            /* transaction pointer */
483         xfs_inode_t             *ip,            /* incore inode pointer */
484         xfs_fsblock_t           *firstblock,    /* first block allocated */
485         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
486         int                     *flags)         /* inode logging flags */
487 {
488         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
489         int                     error;          /* error return value */
490
491         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
492                 return 0;
493         cur = NULL;
494         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
495                 flags, XFS_DATA_FORK);
496         if (cur) {
497                 cur->bc_private.b.allocated = 0;
498                 xfs_btree_del_cursor(cur,
499                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
500         }
501         return error;
502 }
503
504 /*
505  * Called from xfs_bmap_add_attrfork to handle local format files.
506  */
507 STATIC int                                      /* error */
508 xfs_bmap_add_attrfork_local(
509         xfs_trans_t             *tp,            /* transaction pointer */
510         xfs_inode_t             *ip,            /* incore inode pointer */
511         xfs_fsblock_t           *firstblock,    /* first block allocated */
512         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
513         int                     *flags)         /* inode logging flags */
514 {
515         xfs_da_args_t           dargs;          /* args for dir/attr code */
516         int                     error;          /* error return value */
517         xfs_mount_t             *mp;            /* mount structure pointer */
518
519         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
520                 return 0;
521         if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
522                 mp = ip->i_mount;
523                 memset(&dargs, 0, sizeof(dargs));
524                 dargs.dp = ip;
525                 dargs.firstblock = firstblock;
526                 dargs.flist = flist;
527                 dargs.total = mp->m_dirblkfsbs;
528                 dargs.whichfork = XFS_DATA_FORK;
529                 dargs.trans = tp;
530                 error = xfs_dir2_sf_to_block(&dargs);
531         } else
532                 error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
533                         XFS_DATA_FORK);
534         return error;
535 }
536
537 /*
538  * Called by xfs_bmapi to update file extent records and the btree
539  * after allocating space (or doing a delayed allocation).
540  */
541 STATIC int                              /* error */
542 xfs_bmap_add_extent(
543         xfs_inode_t             *ip,    /* incore inode pointer */
544         xfs_extnum_t            idx,    /* extent number to update/insert */
545         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
546         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
547         xfs_fsblock_t           *first, /* pointer to firstblock variable */
548         xfs_bmap_free_t         *flist, /* list of extents to be freed */
549         int                     *logflagsp, /* inode logging flags */
550         xfs_extdelta_t          *delta, /* Change made to incore extents */
551         int                     whichfork, /* data or attr fork */
552         int                     rsvd)   /* OK to use reserved data blocks */
553 {
554         xfs_btree_cur_t         *cur;   /* btree cursor or null */
555         xfs_filblks_t           da_new; /* new count del alloc blocks used */
556         xfs_filblks_t           da_old; /* old count del alloc blocks used */
557         int                     error;  /* error return value */
558         xfs_ifork_t             *ifp;   /* inode fork ptr */
559         int                     logflags; /* returned value */
560         xfs_extnum_t            nextents; /* number of extents in file now */
561
562         XFS_STATS_INC(xs_add_exlist);
563         cur = *curp;
564         ifp = XFS_IFORK_PTR(ip, whichfork);
565         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
566         ASSERT(idx <= nextents);
567         da_old = da_new = 0;
568         error = 0;
569         /*
570          * This is the first extent added to a new/empty file.
571          * Special case this one, so other routines get to assume there are
572          * already extents in the list.
573          */
574         if (nextents == 0) {
575                 XFS_BMAP_TRACE_INSERT("insert empty", ip, 0, 1, new, NULL,
576                         whichfork);
577                 xfs_iext_insert(ifp, 0, 1, new);
578                 ASSERT(cur == NULL);
579                 ifp->if_lastex = 0;
580                 if (!ISNULLSTARTBLOCK(new->br_startblock)) {
581                         XFS_IFORK_NEXT_SET(ip, whichfork, 1);
582                         logflags = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
583                 } else
584                         logflags = 0;
585                 /* DELTA: single new extent */
586                 if (delta) {
587                         if (delta->xed_startoff > new->br_startoff)
588                                 delta->xed_startoff = new->br_startoff;
589                         if (delta->xed_blockcount <
590                                         new->br_startoff + new->br_blockcount)
591                                 delta->xed_blockcount = new->br_startoff +
592                                                 new->br_blockcount;
593                 }
594         }
595         /*
596          * Any kind of new delayed allocation goes here.
597          */
598         else if (ISNULLSTARTBLOCK(new->br_startblock)) {
599                 if (cur)
600                         ASSERT((cur->bc_private.b.flags &
601                                 XFS_BTCUR_BPRV_WASDEL) == 0);
602                 if ((error = xfs_bmap_add_extent_hole_delay(ip, idx, new,
603                                 &logflags, delta, rsvd)))
604                         goto done;
605         }
606         /*
607          * Real allocation off the end of the file.
608          */
609         else if (idx == nextents) {
610                 if (cur)
611                         ASSERT((cur->bc_private.b.flags &
612                                 XFS_BTCUR_BPRV_WASDEL) == 0);
613                 if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur, new,
614                                 &logflags, delta, whichfork)))
615                         goto done;
616         } else {
617                 xfs_bmbt_irec_t prev;   /* old extent at offset idx */
618
619                 /*
620                  * Get the record referred to by idx.
621                  */
622                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx), &prev);
623                 /*
624                  * If it's a real allocation record, and the new allocation ends
625                  * after the start of the referred to record, then we're filling
626                  * in a delayed or unwritten allocation with a real one, or
627                  * converting real back to unwritten.
628                  */
629                 if (!ISNULLSTARTBLOCK(new->br_startblock) &&
630                     new->br_startoff + new->br_blockcount > prev.br_startoff) {
631                         if (prev.br_state != XFS_EXT_UNWRITTEN &&
632                             ISNULLSTARTBLOCK(prev.br_startblock)) {
633                                 da_old = STARTBLOCKVAL(prev.br_startblock);
634                                 if (cur)
635                                         ASSERT(cur->bc_private.b.flags &
636                                                 XFS_BTCUR_BPRV_WASDEL);
637                                 if ((error = xfs_bmap_add_extent_delay_real(ip,
638                                         idx, &cur, new, &da_new, first, flist,
639                                         &logflags, delta, rsvd)))
640                                         goto done;
641                         } else if (new->br_state == XFS_EXT_NORM) {
642                                 ASSERT(new->br_state == XFS_EXT_NORM);
643                                 if ((error = xfs_bmap_add_extent_unwritten_real(
644                                         ip, idx, &cur, new, &logflags, delta)))
645                                         goto done;
646                         } else {
647                                 ASSERT(new->br_state == XFS_EXT_UNWRITTEN);
648                                 if ((error = xfs_bmap_add_extent_unwritten_real(
649                                         ip, idx, &cur, new, &logflags, delta)))
650                                         goto done;
651                         }
652                         ASSERT(*curp == cur || *curp == NULL);
653                 }
654                 /*
655                  * Otherwise we're filling in a hole with an allocation.
656                  */
657                 else {
658                         if (cur)
659                                 ASSERT((cur->bc_private.b.flags &
660                                         XFS_BTCUR_BPRV_WASDEL) == 0);
661                         if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur,
662                                         new, &logflags, delta, whichfork)))
663                                 goto done;
664                 }
665         }
666
667         ASSERT(*curp == cur || *curp == NULL);
668         /*
669          * Convert to a btree if necessary.
670          */
671         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
672             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
673                 int     tmp_logflags;   /* partial log flag return val */
674
675                 ASSERT(cur == NULL);
676                 error = xfs_bmap_extents_to_btree(ip->i_transp, ip, first,
677                         flist, &cur, da_old > 0, &tmp_logflags, whichfork);
678                 logflags |= tmp_logflags;
679                 if (error)
680                         goto done;
681         }
682         /*
683          * Adjust for changes in reserved delayed indirect blocks.
684          * Nothing to do for disk quotas here.
685          */
686         if (da_old || da_new) {
687                 xfs_filblks_t   nblks;
688
689                 nblks = da_new;
690                 if (cur)
691                         nblks += cur->bc_private.b.allocated;
692                 ASSERT(nblks <= da_old);
693                 if (nblks < da_old)
694                         xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
695                                 (int64_t)(da_old - nblks), rsvd);
696         }
697         /*
698          * Clear out the allocated field, done with it now in any case.
699          */
700         if (cur) {
701                 cur->bc_private.b.allocated = 0;
702                 *curp = cur;
703         }
704 done:
705 #ifdef DEBUG
706         if (!error)
707                 xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
708 #endif
709         *logflagsp = logflags;
710         return error;
711 }
712
713 /*
714  * Called by xfs_bmap_add_extent to handle cases converting a delayed
715  * allocation to a real allocation.
716  */
717 STATIC int                              /* error */
718 xfs_bmap_add_extent_delay_real(
719         xfs_inode_t             *ip,    /* incore inode pointer */
720         xfs_extnum_t            idx,    /* extent number to update/insert */
721         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
722         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
723         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
724         xfs_fsblock_t           *first, /* pointer to firstblock variable */
725         xfs_bmap_free_t         *flist, /* list of extents to be freed */
726         int                     *logflagsp, /* inode logging flags */
727         xfs_extdelta_t          *delta, /* Change made to incore extents */
728         int                     rsvd)   /* OK to use reserved data block allocation */
729 {
730         xfs_btree_cur_t         *cur;   /* btree cursor */
731         int                     diff;   /* temp value */
732         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
733         int                     error;  /* error return value */
734         int                     i;      /* temp state */
735         xfs_ifork_t             *ifp;   /* inode fork pointer */
736         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
737         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
738                                         /* left is 0, right is 1, prev is 2 */
739         int                     rval=0; /* return value (logging flags) */
740         int                     state = 0;/* state bits, accessed thru macros */
741         xfs_filblks_t           temp=0; /* value for dnew calculations */
742         xfs_filblks_t           temp2=0;/* value for dnew calculations */
743         int                     tmp_rval;       /* partial logging flags */
744         enum {                          /* bit number definitions for state */
745                 LEFT_CONTIG,    RIGHT_CONTIG,
746                 LEFT_FILLING,   RIGHT_FILLING,
747                 LEFT_DELAY,     RIGHT_DELAY,
748                 LEFT_VALID,     RIGHT_VALID
749         };
750
751 #define LEFT            r[0]
752 #define RIGHT           r[1]
753 #define PREV            r[2]
754 #define MASK(b)         (1 << (b))
755 #define MASK2(a,b)      (MASK(a) | MASK(b))
756 #define MASK3(a,b,c)    (MASK2(a,b) | MASK(c))
757 #define MASK4(a,b,c,d)  (MASK3(a,b,c) | MASK(d))
758 #define STATE_SET(b,v)  ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
759 #define STATE_TEST(b)   (state & MASK(b))
760 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
761                                        ((state &= ~MASK(b)), 0))
762 #define SWITCH_STATE            \
763         (state & MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG))
764
765         /*
766          * Set up a bunch of variables to make the tests simpler.
767          */
768         cur = *curp;
769         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
770         ep = xfs_iext_get_ext(ifp, idx);
771         xfs_bmbt_get_all(ep, &PREV);
772         new_endoff = new->br_startoff + new->br_blockcount;
773         ASSERT(PREV.br_startoff <= new->br_startoff);
774         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
775         /*
776          * Set flags determining what part of the previous delayed allocation
777          * extent is being replaced by a real allocation.
778          */
779         STATE_SET(LEFT_FILLING, PREV.br_startoff == new->br_startoff);
780         STATE_SET(RIGHT_FILLING,
781                 PREV.br_startoff + PREV.br_blockcount == new_endoff);
782         /*
783          * Check and set flags if this segment has a left neighbor.
784          * Don't set contiguous if the combined extent would be too large.
785          */
786         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
787                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &LEFT);
788                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(LEFT.br_startblock));
789         }
790         STATE_SET(LEFT_CONTIG,
791                 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
792                 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
793                 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
794                 LEFT.br_state == new->br_state &&
795                 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN);
796         /*
797          * Check and set flags if this segment has a right neighbor.
798          * Don't set contiguous if the combined extent would be too large.
799          * Also check for all-three-contiguous being too large.
800          */
801         if (STATE_SET_TEST(RIGHT_VALID,
802                         idx <
803                         ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1)) {
804                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx + 1), &RIGHT);
805                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(RIGHT.br_startblock));
806         }
807         STATE_SET(RIGHT_CONTIG,
808                 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
809                 new_endoff == RIGHT.br_startoff &&
810                 new->br_startblock + new->br_blockcount ==
811                     RIGHT.br_startblock &&
812                 new->br_state == RIGHT.br_state &&
813                 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
814                 ((state & MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING)) !=
815                   MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING) ||
816                  LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
817                      <= MAXEXTLEN));
818         error = 0;
819         /*
820          * Switch out based on the FILLING and CONTIG state bits.
821          */
822         switch (SWITCH_STATE) {
823
824         case MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
825                 /*
826                  * Filling in all of a previously delayed allocation extent.
827                  * The left and right neighbors are both contiguous with new.
828                  */
829                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF|LC|RC", ip, idx - 1,
830                         XFS_DATA_FORK);
831                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
832                         LEFT.br_blockcount + PREV.br_blockcount +
833                         RIGHT.br_blockcount);
834                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF|LC|RC", ip, idx - 1,
835                         XFS_DATA_FORK);
836                 XFS_BMAP_TRACE_DELETE("LF|RF|LC|RC", ip, idx, 2, XFS_DATA_FORK);
837                 xfs_iext_remove(ifp, idx, 2);
838                 ip->i_df.if_lastex = idx - 1;
839                 ip->i_d.di_nextents--;
840                 if (cur == NULL)
841                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
842                 else {
843                         rval = XFS_ILOG_CORE;
844                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
845                                         RIGHT.br_startblock,
846                                         RIGHT.br_blockcount, &i)))
847                                 goto done;
848                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
849                         if ((error = xfs_bmbt_delete(cur, &i)))
850                                 goto done;
851                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
852                         if ((error = xfs_btree_decrement(cur, 0, &i)))
853                                 goto done;
854                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
855                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
856                                         LEFT.br_startblock,
857                                         LEFT.br_blockcount +
858                                         PREV.br_blockcount +
859                                         RIGHT.br_blockcount, LEFT.br_state)))
860                                 goto done;
861                 }
862                 *dnew = 0;
863                 /* DELTA: Three in-core extents are replaced by one. */
864                 temp = LEFT.br_startoff;
865                 temp2 = LEFT.br_blockcount +
866                         PREV.br_blockcount +
867                         RIGHT.br_blockcount;
868                 break;
869
870         case MASK3(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG):
871                 /*
872                  * Filling in all of a previously delayed allocation extent.
873                  * The left neighbor is contiguous, the right is not.
874                  */
875                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF|LC", ip, idx - 1,
876                         XFS_DATA_FORK);
877                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
878                         LEFT.br_blockcount + PREV.br_blockcount);
879                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF|LC", ip, idx - 1,
880                         XFS_DATA_FORK);
881                 ip->i_df.if_lastex = idx - 1;
882                 XFS_BMAP_TRACE_DELETE("LF|RF|LC", ip, idx, 1, XFS_DATA_FORK);
883                 xfs_iext_remove(ifp, idx, 1);
884                 if (cur == NULL)
885                         rval = XFS_ILOG_DEXT;
886                 else {
887                         rval = 0;
888                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
889                                         LEFT.br_startblock, LEFT.br_blockcount,
890                                         &i)))
891                                 goto done;
892                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
893                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
894                                         LEFT.br_startblock,
895                                         LEFT.br_blockcount +
896                                         PREV.br_blockcount, LEFT.br_state)))
897                                 goto done;
898                 }
899                 *dnew = 0;
900                 /* DELTA: Two in-core extents are replaced by one. */
901                 temp = LEFT.br_startoff;
902                 temp2 = LEFT.br_blockcount +
903                         PREV.br_blockcount;
904                 break;
905
906         case MASK3(LEFT_FILLING, RIGHT_FILLING, RIGHT_CONTIG):
907                 /*
908                  * Filling in all of a previously delayed allocation extent.
909                  * The right neighbor is contiguous, the left is not.
910                  */
911                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF|RC", ip, idx, 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("LF|RF|RC", ip, idx, XFS_DATA_FORK);
916                 ip->i_df.if_lastex = idx;
917                 XFS_BMAP_TRACE_DELETE("LF|RF|RC", ip, idx + 1, 1, XFS_DATA_FORK);
918                 xfs_iext_remove(ifp, idx + 1, 1);
919                 if (cur == NULL)
920                         rval = XFS_ILOG_DEXT;
921                 else {
922                         rval = 0;
923                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
924                                         RIGHT.br_startblock,
925                                         RIGHT.br_blockcount, &i)))
926                                 goto done;
927                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
928                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
929                                         new->br_startblock,
930                                         PREV.br_blockcount +
931                                         RIGHT.br_blockcount, PREV.br_state)))
932                                 goto done;
933                 }
934                 *dnew = 0;
935                 /* DELTA: Two in-core extents are replaced by one. */
936                 temp = PREV.br_startoff;
937                 temp2 = PREV.br_blockcount +
938                         RIGHT.br_blockcount;
939                 break;
940
941         case MASK2(LEFT_FILLING, RIGHT_FILLING):
942                 /*
943                  * Filling in all of a previously delayed allocation extent.
944                  * Neither the left nor right neighbors are contiguous with
945                  * the new one.
946                  */
947                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF", ip, idx, XFS_DATA_FORK);
948                 xfs_bmbt_set_startblock(ep, new->br_startblock);
949                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF", ip, idx, 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                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
961                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
962                         if ((error = xfs_bmbt_insert(cur, &i)))
963                                 goto done;
964                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
965                 }
966                 *dnew = 0;
967                 /* DELTA: The in-core extent described by new changed type. */
968                 temp = new->br_startoff;
969                 temp2 = new->br_blockcount;
970                 break;
971
972         case MASK2(LEFT_FILLING, LEFT_CONTIG):
973                 /*
974                  * Filling in the first part of a previous delayed allocation.
975                  * The left neighbor is contiguous.
976                  */
977                 XFS_BMAP_TRACE_PRE_UPDATE("LF|LC", ip, idx - 1, XFS_DATA_FORK);
978                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
979                         LEFT.br_blockcount + new->br_blockcount);
980                 xfs_bmbt_set_startoff(ep,
981                         PREV.br_startoff + new->br_blockcount);
982                 XFS_BMAP_TRACE_POST_UPDATE("LF|LC", ip, idx - 1, XFS_DATA_FORK);
983                 temp = PREV.br_blockcount - new->br_blockcount;
984                 XFS_BMAP_TRACE_PRE_UPDATE("LF|LC", ip, idx, 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                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
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("LF|LC", ip, idx, XFS_DATA_FORK);
1007                 *dnew = temp;
1008                 /* DELTA: The boundary between two in-core extents moved. */
1009                 temp = LEFT.br_startoff;
1010                 temp2 = LEFT.br_blockcount +
1011                         PREV.br_blockcount;
1012                 break;
1013
1014         case MASK(LEFT_FILLING):
1015                 /*
1016                  * Filling in the first part of a previous delayed allocation.
1017                  * The left neighbor is not contiguous.
1018                  */
1019                 XFS_BMAP_TRACE_PRE_UPDATE("LF", ip, idx, XFS_DATA_FORK);
1020                 xfs_bmbt_set_startoff(ep, new_endoff);
1021                 temp = PREV.br_blockcount - new->br_blockcount;
1022                 xfs_bmbt_set_blockcount(ep, temp);
1023                 XFS_BMAP_TRACE_INSERT("LF", ip, idx, 1, new, NULL,
1024                         XFS_DATA_FORK);
1025                 xfs_iext_insert(ifp, idx, 1, new);
1026                 ip->i_df.if_lastex = idx;
1027                 ip->i_d.di_nextents++;
1028                 if (cur == NULL)
1029                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1030                 else {
1031                         rval = XFS_ILOG_CORE;
1032                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1033                                         new->br_startblock, new->br_blockcount,
1034                                         &i)))
1035                                 goto done;
1036                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1037                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1038                         if ((error = xfs_bmbt_insert(cur, &i)))
1039                                 goto done;
1040                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1041                 }
1042                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1043                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1044                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1045                                         first, flist, &cur, 1, &tmp_rval,
1046                                         XFS_DATA_FORK);
1047                         rval |= tmp_rval;
1048                         if (error)
1049                                 goto done;
1050                 }
1051                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1052                         STARTBLOCKVAL(PREV.br_startblock) -
1053                         (cur ? cur->bc_private.b.allocated : 0));
1054                 ep = xfs_iext_get_ext(ifp, idx + 1);
1055                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1056                 XFS_BMAP_TRACE_POST_UPDATE("LF", ip, idx + 1, XFS_DATA_FORK);
1057                 *dnew = temp;
1058                 /* DELTA: One in-core extent is split in two. */
1059                 temp = PREV.br_startoff;
1060                 temp2 = PREV.br_blockcount;
1061                 break;
1062
1063         case MASK2(RIGHT_FILLING, RIGHT_CONTIG):
1064                 /*
1065                  * Filling in the last part of a previous delayed allocation.
1066                  * The right neighbor is contiguous with the new allocation.
1067                  */
1068                 temp = PREV.br_blockcount - new->br_blockcount;
1069                 XFS_BMAP_TRACE_PRE_UPDATE("RF|RC", ip, idx, XFS_DATA_FORK);
1070                 XFS_BMAP_TRACE_PRE_UPDATE("RF|RC", ip, idx + 1, XFS_DATA_FORK);
1071                 xfs_bmbt_set_blockcount(ep, temp);
1072                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, idx + 1),
1073                         new->br_startoff, new->br_startblock,
1074                         new->br_blockcount + RIGHT.br_blockcount,
1075                         RIGHT.br_state);
1076                 XFS_BMAP_TRACE_POST_UPDATE("RF|RC", ip, idx + 1, XFS_DATA_FORK);
1077                 ip->i_df.if_lastex = idx + 1;
1078                 if (cur == NULL)
1079                         rval = XFS_ILOG_DEXT;
1080                 else {
1081                         rval = 0;
1082                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1083                                         RIGHT.br_startblock,
1084                                         RIGHT.br_blockcount, &i)))
1085                                 goto done;
1086                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1087                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1088                                         new->br_startblock,
1089                                         new->br_blockcount +
1090                                         RIGHT.br_blockcount,
1091                                         RIGHT.br_state)))
1092                                 goto done;
1093                 }
1094                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1095                         STARTBLOCKVAL(PREV.br_startblock));
1096                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1097                 XFS_BMAP_TRACE_POST_UPDATE("RF|RC", ip, idx, XFS_DATA_FORK);
1098                 *dnew = temp;
1099                 /* DELTA: The boundary between two in-core extents moved. */
1100                 temp = PREV.br_startoff;
1101                 temp2 = PREV.br_blockcount +
1102                         RIGHT.br_blockcount;
1103                 break;
1104
1105         case MASK(RIGHT_FILLING):
1106                 /*
1107                  * Filling in the last part of a previous delayed allocation.
1108                  * The right neighbor is not contiguous.
1109                  */
1110                 temp = PREV.br_blockcount - new->br_blockcount;
1111                 XFS_BMAP_TRACE_PRE_UPDATE("RF", ip, idx, XFS_DATA_FORK);
1112                 xfs_bmbt_set_blockcount(ep, temp);
1113                 XFS_BMAP_TRACE_INSERT("RF", ip, idx + 1, 1, new, NULL,
1114                         XFS_DATA_FORK);
1115                 xfs_iext_insert(ifp, idx + 1, 1, new);
1116                 ip->i_df.if_lastex = idx + 1;
1117                 ip->i_d.di_nextents++;
1118                 if (cur == NULL)
1119                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1120                 else {
1121                         rval = XFS_ILOG_CORE;
1122                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1123                                         new->br_startblock, new->br_blockcount,
1124                                         &i)))
1125                                 goto done;
1126                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1127                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1128                         if ((error = xfs_bmbt_insert(cur, &i)))
1129                                 goto done;
1130                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1131                 }
1132                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1133                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1134                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1135                                 first, flist, &cur, 1, &tmp_rval,
1136                                 XFS_DATA_FORK);
1137                         rval |= tmp_rval;
1138                         if (error)
1139                                 goto done;
1140                 }
1141                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1142                         STARTBLOCKVAL(PREV.br_startblock) -
1143                         (cur ? cur->bc_private.b.allocated : 0));
1144                 ep = xfs_iext_get_ext(ifp, idx);
1145                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1146                 XFS_BMAP_TRACE_POST_UPDATE("RF", ip, idx, XFS_DATA_FORK);
1147                 *dnew = temp;
1148                 /* DELTA: One in-core extent is split in two. */
1149                 temp = PREV.br_startoff;
1150                 temp2 = PREV.br_blockcount;
1151                 break;
1152
1153         case 0:
1154                 /*
1155                  * Filling in the middle part of a previous delayed allocation.
1156                  * Contiguity is impossible here.
1157                  * This case is avoided almost all the time.
1158                  */
1159                 temp = new->br_startoff - PREV.br_startoff;
1160                 XFS_BMAP_TRACE_PRE_UPDATE("0", ip, idx, XFS_DATA_FORK);
1161                 xfs_bmbt_set_blockcount(ep, temp);
1162                 r[0] = *new;
1163                 r[1].br_state = PREV.br_state;
1164                 r[1].br_startblock = 0;
1165                 r[1].br_startoff = new_endoff;
1166                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
1167                 r[1].br_blockcount = temp2;
1168                 XFS_BMAP_TRACE_INSERT("0", ip, idx + 1, 2, &r[0], &r[1],
1169                         XFS_DATA_FORK);
1170                 xfs_iext_insert(ifp, idx + 1, 2, &r[0]);
1171                 ip->i_df.if_lastex = idx + 1;
1172                 ip->i_d.di_nextents++;
1173                 if (cur == NULL)
1174                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1175                 else {
1176                         rval = XFS_ILOG_CORE;
1177                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1178                                         new->br_startblock, new->br_blockcount,
1179                                         &i)))
1180                                 goto done;
1181                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1182                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1183                         if ((error = xfs_bmbt_insert(cur, &i)))
1184                                 goto done;
1185                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1186                 }
1187                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1188                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1189                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1190                                         first, flist, &cur, 1, &tmp_rval,
1191                                         XFS_DATA_FORK);
1192                         rval |= tmp_rval;
1193                         if (error)
1194                                 goto done;
1195                 }
1196                 temp = xfs_bmap_worst_indlen(ip, temp);
1197                 temp2 = xfs_bmap_worst_indlen(ip, temp2);
1198                 diff = (int)(temp + temp2 - STARTBLOCKVAL(PREV.br_startblock) -
1199                         (cur ? cur->bc_private.b.allocated : 0));
1200                 if (diff > 0 &&
1201                     xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS, -((int64_t)diff), rsvd)) {
1202                         /*
1203                          * Ick gross gag me with a spoon.
1204                          */
1205                         ASSERT(0);      /* want to see if this ever happens! */
1206                         while (diff > 0) {
1207                                 if (temp) {
1208                                         temp--;
1209                                         diff--;
1210                                         if (!diff ||
1211                                             !xfs_mod_incore_sb(ip->i_mount,
1212                                                     XFS_SBS_FDBLOCKS, -((int64_t)diff), rsvd))
1213                                                 break;
1214                                 }
1215                                 if (temp2) {
1216                                         temp2--;
1217                                         diff--;
1218                                         if (!diff ||
1219                                             !xfs_mod_incore_sb(ip->i_mount,
1220                                                     XFS_SBS_FDBLOCKS, -((int64_t)diff), rsvd))
1221                                                 break;
1222                                 }
1223                         }
1224                 }
1225                 ep = xfs_iext_get_ext(ifp, idx);
1226                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1227                 XFS_BMAP_TRACE_POST_UPDATE("0", ip, idx, XFS_DATA_FORK);
1228                 XFS_BMAP_TRACE_PRE_UPDATE("0", ip, idx + 2, XFS_DATA_FORK);
1229                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx + 2),
1230                         NULLSTARTBLOCK((int)temp2));
1231                 XFS_BMAP_TRACE_POST_UPDATE("0", ip, idx + 2, XFS_DATA_FORK);
1232                 *dnew = temp + temp2;
1233                 /* DELTA: One in-core extent is split in three. */
1234                 temp = PREV.br_startoff;
1235                 temp2 = PREV.br_blockcount;
1236                 break;
1237
1238         case MASK3(LEFT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1239         case MASK3(RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1240         case MASK2(LEFT_FILLING, RIGHT_CONTIG):
1241         case MASK2(RIGHT_FILLING, LEFT_CONTIG):
1242         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1243         case MASK(LEFT_CONTIG):
1244         case MASK(RIGHT_CONTIG):
1245                 /*
1246                  * These cases are all impossible.
1247                  */
1248                 ASSERT(0);
1249         }
1250         *curp = cur;
1251         if (delta) {
1252                 temp2 += temp;
1253                 if (delta->xed_startoff > temp)
1254                         delta->xed_startoff = temp;
1255                 if (delta->xed_blockcount < temp2)
1256                         delta->xed_blockcount = temp2;
1257         }
1258 done:
1259         *logflagsp = rval;
1260         return error;
1261 #undef  LEFT
1262 #undef  RIGHT
1263 #undef  PREV
1264 #undef  MASK
1265 #undef  MASK2
1266 #undef  MASK3
1267 #undef  MASK4
1268 #undef  STATE_SET
1269 #undef  STATE_TEST
1270 #undef  STATE_SET_TEST
1271 #undef  SWITCH_STATE
1272 }
1273
1274 /*
1275  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
1276  * allocation to a real allocation or vice versa.
1277  */
1278 STATIC int                              /* error */
1279 xfs_bmap_add_extent_unwritten_real(
1280         xfs_inode_t             *ip,    /* incore inode pointer */
1281         xfs_extnum_t            idx,    /* extent number to update/insert */
1282         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
1283         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1284         int                     *logflagsp, /* inode logging flags */
1285         xfs_extdelta_t          *delta) /* Change made to incore extents */
1286 {
1287         xfs_btree_cur_t         *cur;   /* btree cursor */
1288         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
1289         int                     error;  /* error return value */
1290         int                     i;      /* temp state */
1291         xfs_ifork_t             *ifp;   /* inode fork pointer */
1292         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1293         xfs_exntst_t            newext; /* new extent state */
1294         xfs_exntst_t            oldext; /* old extent state */
1295         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1296                                         /* left is 0, right is 1, prev is 2 */
1297         int                     rval=0; /* return value (logging flags) */
1298         int                     state = 0;/* state bits, accessed thru macros */
1299         xfs_filblks_t           temp=0;
1300         xfs_filblks_t           temp2=0;
1301         enum {                          /* bit number definitions for state */
1302                 LEFT_CONTIG,    RIGHT_CONTIG,
1303                 LEFT_FILLING,   RIGHT_FILLING,
1304                 LEFT_DELAY,     RIGHT_DELAY,
1305                 LEFT_VALID,     RIGHT_VALID
1306         };
1307
1308 #define LEFT            r[0]
1309 #define RIGHT           r[1]
1310 #define PREV            r[2]
1311 #define MASK(b)         (1 << (b))
1312 #define MASK2(a,b)      (MASK(a) | MASK(b))
1313 #define MASK3(a,b,c)    (MASK2(a,b) | MASK(c))
1314 #define MASK4(a,b,c,d)  (MASK3(a,b,c) | MASK(d))
1315 #define STATE_SET(b,v)  ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1316 #define STATE_TEST(b)   (state & MASK(b))
1317 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
1318                                        ((state &= ~MASK(b)), 0))
1319 #define SWITCH_STATE            \
1320         (state & MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG))
1321
1322         /*
1323          * Set up a bunch of variables to make the tests simpler.
1324          */
1325         error = 0;
1326         cur = *curp;
1327         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1328         ep = xfs_iext_get_ext(ifp, idx);
1329         xfs_bmbt_get_all(ep, &PREV);
1330         newext = new->br_state;
1331         oldext = (newext == XFS_EXT_UNWRITTEN) ?
1332                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
1333         ASSERT(PREV.br_state == oldext);
1334         new_endoff = new->br_startoff + new->br_blockcount;
1335         ASSERT(PREV.br_startoff <= new->br_startoff);
1336         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1337         /*
1338          * Set flags determining what part of the previous oldext allocation
1339          * extent is being replaced by a newext allocation.
1340          */
1341         STATE_SET(LEFT_FILLING, PREV.br_startoff == new->br_startoff);
1342         STATE_SET(RIGHT_FILLING,
1343                 PREV.br_startoff + PREV.br_blockcount == new_endoff);
1344         /*
1345          * Check and set flags if this segment has a left neighbor.
1346          * Don't set contiguous if the combined extent would be too large.
1347          */
1348         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1349                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &LEFT);
1350                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(LEFT.br_startblock));
1351         }
1352         STATE_SET(LEFT_CONTIG,
1353                 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
1354                 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1355                 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1356                 LEFT.br_state == newext &&
1357                 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1358         /*
1359          * Check and set flags if this segment has a right neighbor.
1360          * Don't set contiguous if the combined extent would be too large.
1361          * Also check for all-three-contiguous being too large.
1362          */
1363         if (STATE_SET_TEST(RIGHT_VALID,
1364                         idx <
1365                         ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1)) {
1366                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx + 1), &RIGHT);
1367                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(RIGHT.br_startblock));
1368         }
1369         STATE_SET(RIGHT_CONTIG,
1370                 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
1371                 new_endoff == RIGHT.br_startoff &&
1372                 new->br_startblock + new->br_blockcount ==
1373                     RIGHT.br_startblock &&
1374                 newext == RIGHT.br_state &&
1375                 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1376                 ((state & MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING)) !=
1377                   MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING) ||
1378                  LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1379                      <= MAXEXTLEN));
1380         /*
1381          * Switch out based on the FILLING and CONTIG state bits.
1382          */
1383         switch (SWITCH_STATE) {
1384
1385         case MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1386                 /*
1387                  * Setting all of a previous oldext extent to newext.
1388                  * The left and right neighbors are both contiguous with new.
1389                  */
1390                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF|LC|RC", ip, idx - 1,
1391                         XFS_DATA_FORK);
1392                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1393                         LEFT.br_blockcount + PREV.br_blockcount +
1394                         RIGHT.br_blockcount);
1395                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF|LC|RC", ip, idx - 1,
1396                         XFS_DATA_FORK);
1397                 XFS_BMAP_TRACE_DELETE("LF|RF|LC|RC", ip, idx, 2, XFS_DATA_FORK);
1398                 xfs_iext_remove(ifp, idx, 2);
1399                 ip->i_df.if_lastex = idx - 1;
1400                 ip->i_d.di_nextents -= 2;
1401                 if (cur == NULL)
1402                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1403                 else {
1404                         rval = XFS_ILOG_CORE;
1405                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1406                                         RIGHT.br_startblock,
1407                                         RIGHT.br_blockcount, &i)))
1408                                 goto done;
1409                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1410                         if ((error = xfs_bmbt_delete(cur, &i)))
1411                                 goto done;
1412                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1413                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1414                                 goto done;
1415                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1416                         if ((error = xfs_bmbt_delete(cur, &i)))
1417                                 goto done;
1418                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1419                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1420                                 goto done;
1421                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1422                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1423                                 LEFT.br_startblock,
1424                                 LEFT.br_blockcount + PREV.br_blockcount +
1425                                 RIGHT.br_blockcount, LEFT.br_state)))
1426                                 goto done;
1427                 }
1428                 /* DELTA: Three in-core extents are replaced by one. */
1429                 temp = LEFT.br_startoff;
1430                 temp2 = LEFT.br_blockcount +
1431                         PREV.br_blockcount +
1432                         RIGHT.br_blockcount;
1433                 break;
1434
1435         case MASK3(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG):
1436                 /*
1437                  * Setting all of a previous oldext extent to newext.
1438                  * The left neighbor is contiguous, the right is not.
1439                  */
1440                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF|LC", ip, idx - 1,
1441                         XFS_DATA_FORK);
1442                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1443                         LEFT.br_blockcount + PREV.br_blockcount);
1444                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF|LC", ip, idx - 1,
1445                         XFS_DATA_FORK);
1446                 ip->i_df.if_lastex = idx - 1;
1447                 XFS_BMAP_TRACE_DELETE("LF|RF|LC", ip, idx, 1, XFS_DATA_FORK);
1448                 xfs_iext_remove(ifp, idx, 1);
1449                 ip->i_d.di_nextents--;
1450                 if (cur == NULL)
1451                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1452                 else {
1453                         rval = XFS_ILOG_CORE;
1454                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1455                                         PREV.br_startblock, PREV.br_blockcount,
1456                                         &i)))
1457                                 goto done;
1458                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1459                         if ((error = xfs_bmbt_delete(cur, &i)))
1460                                 goto done;
1461                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1462                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1463                                 goto done;
1464                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1465                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1466                                 LEFT.br_startblock,
1467                                 LEFT.br_blockcount + PREV.br_blockcount,
1468                                 LEFT.br_state)))
1469                                 goto done;
1470                 }
1471                 /* DELTA: Two in-core extents are replaced by one. */
1472                 temp = LEFT.br_startoff;
1473                 temp2 = LEFT.br_blockcount +
1474                         PREV.br_blockcount;
1475                 break;
1476
1477         case MASK3(LEFT_FILLING, RIGHT_FILLING, RIGHT_CONTIG):
1478                 /*
1479                  * Setting all of a previous oldext extent to newext.
1480                  * The right neighbor is contiguous, the left is not.
1481                  */
1482                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF|RC", ip, idx,
1483                         XFS_DATA_FORK);
1484                 xfs_bmbt_set_blockcount(ep,
1485                         PREV.br_blockcount + RIGHT.br_blockcount);
1486                 xfs_bmbt_set_state(ep, newext);
1487                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF|RC", ip, idx,
1488                         XFS_DATA_FORK);
1489                 ip->i_df.if_lastex = idx;
1490                 XFS_BMAP_TRACE_DELETE("LF|RF|RC", ip, idx + 1, 1, XFS_DATA_FORK);
1491                 xfs_iext_remove(ifp, idx + 1, 1);
1492                 ip->i_d.di_nextents--;
1493                 if (cur == NULL)
1494                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1495                 else {
1496                         rval = XFS_ILOG_CORE;
1497                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1498                                         RIGHT.br_startblock,
1499                                         RIGHT.br_blockcount, &i)))
1500                                 goto done;
1501                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1502                         if ((error = xfs_bmbt_delete(cur, &i)))
1503                                 goto done;
1504                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1505                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1506                                 goto done;
1507                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1508                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1509                                 new->br_startblock,
1510                                 new->br_blockcount + RIGHT.br_blockcount,
1511                                 newext)))
1512                                 goto done;
1513                 }
1514                 /* DELTA: Two in-core extents are replaced by one. */
1515                 temp = PREV.br_startoff;
1516                 temp2 = PREV.br_blockcount +
1517                         RIGHT.br_blockcount;
1518                 break;
1519
1520         case MASK2(LEFT_FILLING, RIGHT_FILLING):
1521                 /*
1522                  * Setting all of a previous oldext extent to newext.
1523                  * Neither the left nor right neighbors are contiguous with
1524                  * the new one.
1525                  */
1526                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF", ip, idx,
1527                         XFS_DATA_FORK);
1528                 xfs_bmbt_set_state(ep, newext);
1529                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF", ip, idx,
1530                         XFS_DATA_FORK);
1531                 ip->i_df.if_lastex = idx;
1532                 if (cur == NULL)
1533                         rval = XFS_ILOG_DEXT;
1534                 else {
1535                         rval = 0;
1536                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1537                                         new->br_startblock, new->br_blockcount,
1538                                         &i)))
1539                                 goto done;
1540                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1541                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1542                                 new->br_startblock, new->br_blockcount,
1543                                 newext)))
1544                                 goto done;
1545                 }
1546                 /* DELTA: The in-core extent described by new changed type. */
1547                 temp = new->br_startoff;
1548                 temp2 = new->br_blockcount;
1549                 break;
1550
1551         case MASK2(LEFT_FILLING, LEFT_CONTIG):
1552                 /*
1553                  * Setting the first part of a previous oldext extent to newext.
1554                  * The left neighbor is contiguous.
1555                  */
1556                 XFS_BMAP_TRACE_PRE_UPDATE("LF|LC", ip, idx - 1,
1557                         XFS_DATA_FORK);
1558                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1559                         LEFT.br_blockcount + new->br_blockcount);
1560                 xfs_bmbt_set_startoff(ep,
1561                         PREV.br_startoff + new->br_blockcount);
1562                 XFS_BMAP_TRACE_POST_UPDATE("LF|LC", ip, idx - 1,
1563                         XFS_DATA_FORK);
1564                 XFS_BMAP_TRACE_PRE_UPDATE("LF|LC", ip, idx,
1565                         XFS_DATA_FORK);
1566                 xfs_bmbt_set_startblock(ep,
1567                         new->br_startblock + new->br_blockcount);
1568                 xfs_bmbt_set_blockcount(ep,
1569                         PREV.br_blockcount - new->br_blockcount);
1570                 XFS_BMAP_TRACE_POST_UPDATE("LF|LC", ip, idx,
1571                         XFS_DATA_FORK);
1572                 ip->i_df.if_lastex = idx - 1;
1573                 if (cur == NULL)
1574                         rval = XFS_ILOG_DEXT;
1575                 else {
1576                         rval = 0;
1577                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1578                                         PREV.br_startblock, PREV.br_blockcount,
1579                                         &i)))
1580                                 goto done;
1581                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1582                         if ((error = xfs_bmbt_update(cur,
1583                                 PREV.br_startoff + new->br_blockcount,
1584                                 PREV.br_startblock + new->br_blockcount,
1585                                 PREV.br_blockcount - new->br_blockcount,
1586                                 oldext)))
1587                                 goto done;
1588                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1589                                 goto done;
1590                         if (xfs_bmbt_update(cur, LEFT.br_startoff,
1591                                 LEFT.br_startblock,
1592                                 LEFT.br_blockcount + new->br_blockcount,
1593                                 LEFT.br_state))
1594                                 goto done;
1595                 }
1596                 /* DELTA: The boundary between two in-core extents moved. */
1597                 temp = LEFT.br_startoff;
1598                 temp2 = LEFT.br_blockcount +
1599                         PREV.br_blockcount;
1600                 break;
1601
1602         case MASK(LEFT_FILLING):
1603                 /*
1604                  * Setting the first part of a previous oldext extent to newext.
1605                  * The left neighbor is not contiguous.
1606                  */
1607                 XFS_BMAP_TRACE_PRE_UPDATE("LF", ip, idx, XFS_DATA_FORK);
1608                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1609                 xfs_bmbt_set_startoff(ep, new_endoff);
1610                 xfs_bmbt_set_blockcount(ep,
1611                         PREV.br_blockcount - new->br_blockcount);
1612                 xfs_bmbt_set_startblock(ep,
1613                         new->br_startblock + new->br_blockcount);
1614                 XFS_BMAP_TRACE_POST_UPDATE("LF", ip, idx, XFS_DATA_FORK);
1615                 XFS_BMAP_TRACE_INSERT("LF", ip, idx, 1, new, NULL,
1616                         XFS_DATA_FORK);
1617                 xfs_iext_insert(ifp, idx, 1, new);
1618                 ip->i_df.if_lastex = idx;
1619                 ip->i_d.di_nextents++;
1620                 if (cur == NULL)
1621                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1622                 else {
1623                         rval = XFS_ILOG_CORE;
1624                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1625                                         PREV.br_startblock, PREV.br_blockcount,
1626                                         &i)))
1627                                 goto done;
1628                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1629                         if ((error = xfs_bmbt_update(cur,
1630                                 PREV.br_startoff + new->br_blockcount,
1631                                 PREV.br_startblock + new->br_blockcount,
1632                                 PREV.br_blockcount - new->br_blockcount,
1633                                 oldext)))
1634                                 goto done;
1635                         cur->bc_rec.b = *new;
1636                         if ((error = xfs_bmbt_insert(cur, &i)))
1637                                 goto done;
1638                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1639                 }
1640                 /* DELTA: One in-core extent is split in two. */
1641                 temp = PREV.br_startoff;
1642                 temp2 = PREV.br_blockcount;
1643                 break;
1644
1645         case MASK2(RIGHT_FILLING, RIGHT_CONTIG):
1646                 /*
1647                  * Setting the last part of a previous oldext extent to newext.
1648                  * The right neighbor is contiguous with the new allocation.
1649                  */
1650                 XFS_BMAP_TRACE_PRE_UPDATE("RF|RC", ip, idx,
1651                         XFS_DATA_FORK);
1652                 XFS_BMAP_TRACE_PRE_UPDATE("RF|RC", ip, idx + 1,
1653                         XFS_DATA_FORK);
1654                 xfs_bmbt_set_blockcount(ep,
1655                         PREV.br_blockcount - new->br_blockcount);
1656                 XFS_BMAP_TRACE_POST_UPDATE("RF|RC", ip, idx,
1657                         XFS_DATA_FORK);
1658                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, idx + 1),
1659                         new->br_startoff, new->br_startblock,
1660                         new->br_blockcount + RIGHT.br_blockcount, newext);
1661                 XFS_BMAP_TRACE_POST_UPDATE("RF|RC", ip, idx + 1,
1662                         XFS_DATA_FORK);
1663                 ip->i_df.if_lastex = idx + 1;
1664                 if (cur == NULL)
1665                         rval = XFS_ILOG_DEXT;
1666                 else {
1667                         rval = 0;
1668                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1669                                         PREV.br_startblock,
1670                                         PREV.br_blockcount, &i)))
1671                                 goto done;
1672                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
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_btree_increment(cur, 0, &i)))
1679                                 goto done;
1680                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1681                                 new->br_startblock,
1682                                 new->br_blockcount + RIGHT.br_blockcount,
1683                                 newext)))
1684                                 goto done;
1685                 }
1686                 /* DELTA: The boundary between two in-core extents moved. */
1687                 temp = PREV.br_startoff;
1688                 temp2 = PREV.br_blockcount +
1689                         RIGHT.br_blockcount;
1690                 break;
1691
1692         case MASK(RIGHT_FILLING):
1693                 /*
1694                  * Setting the last part of a previous oldext extent to newext.
1695                  * The right neighbor is not contiguous.
1696                  */
1697                 XFS_BMAP_TRACE_PRE_UPDATE("RF", ip, idx, XFS_DATA_FORK);
1698                 xfs_bmbt_set_blockcount(ep,
1699                         PREV.br_blockcount - new->br_blockcount);
1700                 XFS_BMAP_TRACE_POST_UPDATE("RF", ip, idx, XFS_DATA_FORK);
1701                 XFS_BMAP_TRACE_INSERT("RF", ip, idx + 1, 1, new, NULL,
1702                         XFS_DATA_FORK);
1703                 xfs_iext_insert(ifp, idx + 1, 1, new);
1704                 ip->i_df.if_lastex = idx + 1;
1705                 ip->i_d.di_nextents++;
1706                 if (cur == NULL)
1707                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1708                 else {
1709                         rval = XFS_ILOG_CORE;
1710                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1711                                         PREV.br_startblock, PREV.br_blockcount,
1712                                         &i)))
1713                                 goto done;
1714                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1715                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1716                                 PREV.br_startblock,
1717                                 PREV.br_blockcount - new->br_blockcount,
1718                                 oldext)))
1719                                 goto done;
1720                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1721                                         new->br_startblock, new->br_blockcount,
1722                                         &i)))
1723                                 goto done;
1724                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1725                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1726                         if ((error = xfs_bmbt_insert(cur, &i)))
1727                                 goto done;
1728                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1729                 }
1730                 /* DELTA: One in-core extent is split in two. */
1731                 temp = PREV.br_startoff;
1732                 temp2 = PREV.br_blockcount;
1733                 break;
1734
1735         case 0:
1736                 /*
1737                  * Setting the middle part of a previous oldext extent to
1738                  * newext.  Contiguity is impossible here.
1739                  * One extent becomes three extents.
1740                  */
1741                 XFS_BMAP_TRACE_PRE_UPDATE("0", ip, idx, XFS_DATA_FORK);
1742                 xfs_bmbt_set_blockcount(ep,
1743                         new->br_startoff - PREV.br_startoff);
1744                 XFS_BMAP_TRACE_POST_UPDATE("0", ip, idx, XFS_DATA_FORK);
1745                 r[0] = *new;
1746                 r[1].br_startoff = new_endoff;
1747                 r[1].br_blockcount =
1748                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
1749                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1750                 r[1].br_state = oldext;
1751                 XFS_BMAP_TRACE_INSERT("0", ip, idx + 1, 2, &r[0], &r[1],
1752                         XFS_DATA_FORK);
1753                 xfs_iext_insert(ifp, idx + 1, 2, &r[0]);
1754                 ip->i_df.if_lastex = idx + 1;
1755                 ip->i_d.di_nextents += 2;
1756                 if (cur == NULL)
1757                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1758                 else {
1759                         rval = XFS_ILOG_CORE;
1760                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1761                                         PREV.br_startblock, PREV.br_blockcount,
1762                                         &i)))
1763                                 goto done;
1764                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1765                         /* new right extent - oldext */
1766                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1767                                 r[1].br_startblock, r[1].br_blockcount,
1768                                 r[1].br_state)))
1769                                 goto done;
1770                         /* new left extent - oldext */
1771                         cur->bc_rec.b = PREV;
1772                         cur->bc_rec.b.br_blockcount =
1773                                 new->br_startoff - PREV.br_startoff;
1774                         if ((error = xfs_bmbt_insert(cur, &i)))
1775                                 goto done;
1776                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1777                         /*
1778                          * Reset the cursor to the position of the new extent
1779                          * we are about to insert as we can't trust it after
1780                          * the previous insert.
1781                          */
1782                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1783                                         new->br_startblock, new->br_blockcount,
1784                                         &i)))
1785                                 goto done;
1786                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1787                         /* new middle extent - newext */
1788                         cur->bc_rec.b.br_state = new->br_state;
1789                         if ((error = xfs_bmbt_insert(cur, &i)))
1790                                 goto done;
1791                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1792                 }
1793                 /* DELTA: One in-core extent is split in three. */
1794                 temp = PREV.br_startoff;
1795                 temp2 = PREV.br_blockcount;
1796                 break;
1797
1798         case MASK3(LEFT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1799         case MASK3(RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1800         case MASK2(LEFT_FILLING, RIGHT_CONTIG):
1801         case MASK2(RIGHT_FILLING, LEFT_CONTIG):
1802         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1803         case MASK(LEFT_CONTIG):
1804         case MASK(RIGHT_CONTIG):
1805                 /*
1806                  * These cases are all impossible.
1807                  */
1808                 ASSERT(0);
1809         }
1810         *curp = cur;
1811         if (delta) {
1812                 temp2 += temp;
1813                 if (delta->xed_startoff > temp)
1814                         delta->xed_startoff = temp;
1815                 if (delta->xed_blockcount < temp2)
1816                         delta->xed_blockcount = temp2;
1817         }
1818 done:
1819         *logflagsp = rval;
1820         return error;
1821 #undef  LEFT
1822 #undef  RIGHT
1823 #undef  PREV
1824 #undef  MASK
1825 #undef  MASK2
1826 #undef  MASK3
1827 #undef  MASK4
1828 #undef  STATE_SET
1829 #undef  STATE_TEST
1830 #undef  STATE_SET_TEST
1831 #undef  SWITCH_STATE
1832 }
1833
1834 /*
1835  * Called by xfs_bmap_add_extent to handle cases converting a hole
1836  * to a delayed allocation.
1837  */
1838 /*ARGSUSED*/
1839 STATIC int                              /* error */
1840 xfs_bmap_add_extent_hole_delay(
1841         xfs_inode_t             *ip,    /* incore inode pointer */
1842         xfs_extnum_t            idx,    /* extent number to update/insert */
1843         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1844         int                     *logflagsp, /* inode logging flags */
1845         xfs_extdelta_t          *delta, /* Change made to incore extents */
1846         int                     rsvd)           /* OK to allocate reserved blocks */
1847 {
1848         xfs_bmbt_rec_host_t     *ep;    /* extent record for idx */
1849         xfs_ifork_t             *ifp;   /* inode fork pointer */
1850         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1851         xfs_filblks_t           newlen=0;       /* new indirect size */
1852         xfs_filblks_t           oldlen=0;       /* old indirect size */
1853         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1854         int                     state;  /* state bits, accessed thru macros */
1855         xfs_filblks_t           temp=0; /* temp for indirect calculations */
1856         xfs_filblks_t           temp2=0;
1857         enum {                          /* bit number definitions for state */
1858                 LEFT_CONTIG,    RIGHT_CONTIG,
1859                 LEFT_DELAY,     RIGHT_DELAY,
1860                 LEFT_VALID,     RIGHT_VALID
1861         };
1862
1863 #define MASK(b)                 (1 << (b))
1864 #define MASK2(a,b)              (MASK(a) | MASK(b))
1865 #define STATE_SET(b,v)          ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1866 #define STATE_TEST(b)           (state & MASK(b))
1867 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
1868                                        ((state &= ~MASK(b)), 0))
1869 #define SWITCH_STATE            (state & MASK2(LEFT_CONTIG, RIGHT_CONTIG))
1870
1871         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1872         ep = xfs_iext_get_ext(ifp, idx);
1873         state = 0;
1874         ASSERT(ISNULLSTARTBLOCK(new->br_startblock));
1875         /*
1876          * Check and set flags if this segment has a left neighbor
1877          */
1878         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1879                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &left);
1880                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(left.br_startblock));
1881         }
1882         /*
1883          * Check and set flags if the current (right) segment exists.
1884          * If it doesn't exist, we're converting the hole at end-of-file.
1885          */
1886         if (STATE_SET_TEST(RIGHT_VALID,
1887                            idx <
1888                            ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1889                 xfs_bmbt_get_all(ep, &right);
1890                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(right.br_startblock));
1891         }
1892         /*
1893          * Set contiguity flags on the left and right neighbors.
1894          * Don't let extents get too large, even if the pieces are contiguous.
1895          */
1896         STATE_SET(LEFT_CONTIG,
1897                 STATE_TEST(LEFT_VALID) && STATE_TEST(LEFT_DELAY) &&
1898                 left.br_startoff + left.br_blockcount == new->br_startoff &&
1899                 left.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1900         STATE_SET(RIGHT_CONTIG,
1901                 STATE_TEST(RIGHT_VALID) && STATE_TEST(RIGHT_DELAY) &&
1902                 new->br_startoff + new->br_blockcount == right.br_startoff &&
1903                 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1904                 (!STATE_TEST(LEFT_CONTIG) ||
1905                  (left.br_blockcount + new->br_blockcount +
1906                      right.br_blockcount <= MAXEXTLEN)));
1907         /*
1908          * Switch out based on the contiguity flags.
1909          */
1910         switch (SWITCH_STATE) {
1911
1912         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1913                 /*
1914                  * New allocation is contiguous with delayed allocations
1915                  * on the left and on the right.
1916                  * Merge all three into a single extent record.
1917                  */
1918                 temp = left.br_blockcount + new->br_blockcount +
1919                         right.br_blockcount;
1920                 XFS_BMAP_TRACE_PRE_UPDATE("LC|RC", ip, idx - 1,
1921                         XFS_DATA_FORK);
1922                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1), temp);
1923                 oldlen = STARTBLOCKVAL(left.br_startblock) +
1924                         STARTBLOCKVAL(new->br_startblock) +
1925                         STARTBLOCKVAL(right.br_startblock);
1926                 newlen = xfs_bmap_worst_indlen(ip, temp);
1927                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx - 1),
1928                         NULLSTARTBLOCK((int)newlen));
1929                 XFS_BMAP_TRACE_POST_UPDATE("LC|RC", ip, idx - 1,
1930                         XFS_DATA_FORK);
1931                 XFS_BMAP_TRACE_DELETE("LC|RC", ip, idx, 1, XFS_DATA_FORK);
1932                 xfs_iext_remove(ifp, idx, 1);
1933                 ip->i_df.if_lastex = idx - 1;
1934                 /* DELTA: Two in-core extents were replaced by one. */
1935                 temp2 = temp;
1936                 temp = left.br_startoff;
1937                 break;
1938
1939         case MASK(LEFT_CONTIG):
1940                 /*
1941                  * New allocation is contiguous with a delayed allocation
1942                  * on the left.
1943                  * Merge the new allocation with the left neighbor.
1944                  */
1945                 temp = left.br_blockcount + new->br_blockcount;
1946                 XFS_BMAP_TRACE_PRE_UPDATE("LC", ip, idx - 1,
1947                         XFS_DATA_FORK);
1948                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1), temp);
1949                 oldlen = STARTBLOCKVAL(left.br_startblock) +
1950                         STARTBLOCKVAL(new->br_startblock);
1951                 newlen = xfs_bmap_worst_indlen(ip, temp);
1952                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx - 1),
1953                         NULLSTARTBLOCK((int)newlen));
1954                 XFS_BMAP_TRACE_POST_UPDATE("LC", ip, idx - 1,
1955                         XFS_DATA_FORK);
1956                 ip->i_df.if_lastex = idx - 1;
1957                 /* DELTA: One in-core extent grew into a hole. */
1958                 temp2 = temp;
1959                 temp = left.br_startoff;
1960                 break;
1961
1962         case MASK(RIGHT_CONTIG):
1963                 /*
1964                  * New allocation is contiguous with a delayed allocation
1965                  * on the right.
1966                  * Merge the new allocation with the right neighbor.
1967                  */
1968                 XFS_BMAP_TRACE_PRE_UPDATE("RC", ip, idx, XFS_DATA_FORK);
1969                 temp = new->br_blockcount + right.br_blockcount;
1970                 oldlen = STARTBLOCKVAL(new->br_startblock) +
1971                         STARTBLOCKVAL(right.br_startblock);
1972                 newlen = xfs_bmap_worst_indlen(ip, temp);
1973                 xfs_bmbt_set_allf(ep, new->br_startoff,
1974                         NULLSTARTBLOCK((int)newlen), temp, right.br_state);
1975                 XFS_BMAP_TRACE_POST_UPDATE("RC", ip, idx, XFS_DATA_FORK);
1976                 ip->i_df.if_lastex = idx;
1977                 /* DELTA: One in-core extent grew into a hole. */
1978                 temp2 = temp;
1979                 temp = new->br_startoff;
1980                 break;
1981
1982         case 0:
1983                 /*
1984                  * New allocation is not contiguous with another
1985                  * delayed allocation.
1986                  * Insert a new entry.
1987                  */
1988                 oldlen = newlen = 0;
1989                 XFS_BMAP_TRACE_INSERT("0", ip, idx, 1, new, NULL,
1990                         XFS_DATA_FORK);
1991                 xfs_iext_insert(ifp, idx, 1, new);
1992                 ip->i_df.if_lastex = idx;
1993                 /* DELTA: A new in-core extent was added in a hole. */
1994                 temp2 = new->br_blockcount;
1995                 temp = new->br_startoff;
1996                 break;
1997         }
1998         if (oldlen != newlen) {
1999                 ASSERT(oldlen > newlen);
2000                 xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
2001                         (int64_t)(oldlen - newlen), rsvd);
2002                 /*
2003                  * Nothing to do for disk quota accounting here.
2004                  */
2005         }
2006         if (delta) {
2007                 temp2 += temp;
2008                 if (delta->xed_startoff > temp)
2009                         delta->xed_startoff = temp;
2010                 if (delta->xed_blockcount < temp2)
2011                         delta->xed_blockcount = temp2;
2012         }
2013         *logflagsp = 0;
2014         return 0;
2015 #undef  MASK
2016 #undef  MASK2
2017 #undef  STATE_SET
2018 #undef  STATE_TEST
2019 #undef  STATE_SET_TEST
2020 #undef  SWITCH_STATE
2021 }
2022
2023 /*
2024  * Called by xfs_bmap_add_extent to handle cases converting a hole
2025  * to a real allocation.
2026  */
2027 STATIC int                              /* error */
2028 xfs_bmap_add_extent_hole_real(
2029         xfs_inode_t             *ip,    /* incore inode pointer */
2030         xfs_extnum_t            idx,    /* extent number to update/insert */
2031         xfs_btree_cur_t         *cur,   /* if null, not a btree */
2032         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
2033         int                     *logflagsp, /* inode logging flags */
2034         xfs_extdelta_t          *delta, /* Change made to incore extents */
2035         int                     whichfork) /* data or attr fork */
2036 {
2037         xfs_bmbt_rec_host_t     *ep;    /* pointer to extent entry ins. point */
2038         int                     error;  /* error return value */
2039         int                     i;      /* temp state */
2040         xfs_ifork_t             *ifp;   /* inode fork pointer */
2041         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2042         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2043         int                     rval=0; /* return value (logging flags) */
2044         int                     state;  /* state bits, accessed thru macros */
2045         xfs_filblks_t           temp=0;
2046         xfs_filblks_t           temp2=0;
2047         enum {                          /* bit number definitions for state */
2048                 LEFT_CONTIG,    RIGHT_CONTIG,
2049                 LEFT_DELAY,     RIGHT_DELAY,
2050                 LEFT_VALID,     RIGHT_VALID
2051         };
2052
2053 #define MASK(b)                 (1 << (b))
2054 #define MASK2(a,b)              (MASK(a) | MASK(b))
2055 #define STATE_SET(b,v)          ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
2056 #define STATE_TEST(b)           (state & MASK(b))
2057 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
2058                                        ((state &= ~MASK(b)), 0))
2059 #define SWITCH_STATE            (state & MASK2(LEFT_CONTIG, RIGHT_CONTIG))
2060
2061         ifp = XFS_IFORK_PTR(ip, whichfork);
2062         ASSERT(idx <= ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
2063         ep = xfs_iext_get_ext(ifp, idx);
2064         state = 0;
2065         /*
2066          * Check and set flags if this segment has a left neighbor.
2067          */
2068         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
2069                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &left);
2070                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(left.br_startblock));
2071         }
2072         /*
2073          * Check and set flags if this segment has a current value.
2074          * Not true if we're inserting into the "hole" at eof.
2075          */
2076         if (STATE_SET_TEST(RIGHT_VALID,
2077                            idx <
2078                            ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
2079                 xfs_bmbt_get_all(ep, &right);
2080                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(right.br_startblock));
2081         }
2082         /*
2083          * We're inserting a real allocation between "left" and "right".
2084          * Set the contiguity flags.  Don't let extents get too large.
2085          */
2086         STATE_SET(LEFT_CONTIG,
2087                 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
2088                 left.br_startoff + left.br_blockcount == new->br_startoff &&
2089                 left.br_startblock + left.br_blockcount == new->br_startblock &&
2090                 left.br_state == new->br_state &&
2091                 left.br_blockcount + new->br_blockcount <= MAXEXTLEN);
2092         STATE_SET(RIGHT_CONTIG,
2093                 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
2094                 new->br_startoff + new->br_blockcount == right.br_startoff &&
2095                 new->br_startblock + new->br_blockcount ==
2096                     right.br_startblock &&
2097                 new->br_state == right.br_state &&
2098                 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2099                 (!STATE_TEST(LEFT_CONTIG) ||
2100                  left.br_blockcount + new->br_blockcount +
2101                      right.br_blockcount <= MAXEXTLEN));
2102
2103         error = 0;
2104         /*
2105          * Select which case we're in here, and implement it.
2106          */
2107         switch (SWITCH_STATE) {
2108
2109         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
2110                 /*
2111                  * New allocation is contiguous with real allocations on the
2112                  * left and on the right.
2113                  * Merge all three into a single extent record.
2114                  */
2115                 XFS_BMAP_TRACE_PRE_UPDATE("LC|RC", ip, idx - 1,
2116                         whichfork);
2117                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
2118                         left.br_blockcount + new->br_blockcount +
2119                         right.br_blockcount);
2120                 XFS_BMAP_TRACE_POST_UPDATE("LC|RC", ip, idx - 1,
2121                         whichfork);
2122                 XFS_BMAP_TRACE_DELETE("LC|RC", ip, idx, 1, whichfork);
2123                 xfs_iext_remove(ifp, idx, 1);
2124                 ifp->if_lastex = idx - 1;
2125                 XFS_IFORK_NEXT_SET(ip, whichfork,
2126                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2127                 if (cur == NULL) {
2128                         rval = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2129                 } else {
2130                         rval = XFS_ILOG_CORE;
2131                         if ((error = xfs_bmbt_lookup_eq(cur,
2132                                         right.br_startoff,
2133                                         right.br_startblock,
2134                                         right.br_blockcount, &i)))
2135                                 goto done;
2136                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2137                         if ((error = xfs_bmbt_delete(cur, &i)))
2138                                 goto done;
2139                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2140                         if ((error = xfs_btree_decrement(cur, 0, &i)))
2141                                 goto done;
2142                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2143                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
2144                                         left.br_startblock,
2145                                         left.br_blockcount +
2146                                                 new->br_blockcount +
2147                                                 right.br_blockcount,
2148                                         left.br_state)))
2149                                 goto done;
2150                 }
2151                 /* DELTA: Two in-core extents were replaced by one. */
2152                 temp = left.br_startoff;
2153                 temp2 = left.br_blockcount +
2154                         new->br_blockcount +
2155                         right.br_blockcount;
2156                 break;
2157
2158         case MASK(LEFT_CONTIG):
2159                 /*
2160                  * New allocation is contiguous with a real allocation
2161                  * on the left.
2162                  * Merge the new allocation with the left neighbor.
2163                  */
2164                 XFS_BMAP_TRACE_PRE_UPDATE("LC", ip, idx - 1, whichfork);
2165                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
2166                         left.br_blockcount + new->br_blockcount);
2167                 XFS_BMAP_TRACE_POST_UPDATE("LC", ip, idx - 1, whichfork);
2168                 ifp->if_lastex = idx - 1;
2169                 if (cur == NULL) {
2170                         rval = XFS_ILOG_FEXT(whichfork);
2171                 } else {
2172                         rval = 0;
2173                         if ((error = xfs_bmbt_lookup_eq(cur,
2174                                         left.br_startoff,
2175                                         left.br_startblock,
2176                                         left.br_blockcount, &i)))
2177                                 goto done;
2178                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2179                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
2180                                         left.br_startblock,
2181                                         left.br_blockcount +
2182                                                 new->br_blockcount,
2183                                         left.br_state)))
2184                                 goto done;
2185                 }
2186                 /* DELTA: One in-core extent grew. */
2187                 temp = left.br_startoff;
2188                 temp2 = left.br_blockcount +
2189                         new->br_blockcount;
2190                 break;
2191
2192         case MASK(RIGHT_CONTIG):
2193                 /*
2194                  * New allocation is contiguous with a real allocation
2195                  * on the right.
2196                  * Merge the new allocation with the right neighbor.
2197                  */
2198                 XFS_BMAP_TRACE_PRE_UPDATE("RC", ip, idx, whichfork);
2199                 xfs_bmbt_set_allf(ep, new->br_startoff, new->br_startblock,
2200                         new->br_blockcount + right.br_blockcount,
2201                         right.br_state);
2202                 XFS_BMAP_TRACE_POST_UPDATE("RC", ip, idx, whichfork);
2203                 ifp->if_lastex = idx;
2204                 if (cur == NULL) {
2205                         rval = XFS_ILOG_FEXT(whichfork);
2206                 } else {
2207                         rval = 0;
2208                         if ((error = xfs_bmbt_lookup_eq(cur,
2209                                         right.br_startoff,
2210                                         right.br_startblock,
2211                                         right.br_blockcount, &i)))
2212                                 goto done;
2213                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2214                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2215                                         new->br_startblock,
2216                                         new->br_blockcount +
2217                                                 right.br_blockcount,
2218                                         right.br_state)))
2219                                 goto done;
2220                 }
2221                 /* DELTA: One in-core extent grew. */
2222                 temp = new->br_startoff;
2223                 temp2 = new->br_blockcount +
2224                         right.br_blockcount;
2225                 break;
2226
2227         case 0:
2228                 /*
2229                  * New allocation is not contiguous with another
2230                  * real allocation.
2231                  * Insert a new entry.
2232                  */
2233                 XFS_BMAP_TRACE_INSERT("0", ip, idx, 1, new, NULL, whichfork);
2234                 xfs_iext_insert(ifp, idx, 1, new);
2235                 ifp->if_lastex = idx;
2236                 XFS_IFORK_NEXT_SET(ip, whichfork,
2237                         XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2238                 if (cur == NULL) {
2239                         rval = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2240                 } else {
2241                         rval = XFS_ILOG_CORE;
2242                         if ((error = xfs_bmbt_lookup_eq(cur,
2243                                         new->br_startoff,
2244                                         new->br_startblock,
2245                                         new->br_blockcount, &i)))
2246                                 goto done;
2247                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2248                         cur->bc_rec.b.br_state = new->br_state;
2249                         if ((error = xfs_bmbt_insert(cur, &i)))
2250                                 goto done;
2251                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2252                 }
2253                 /* DELTA: A new extent was added in a hole. */
2254                 temp = new->br_startoff;
2255                 temp2 = new->br_blockcount;
2256                 break;
2257         }
2258         if (delta) {
2259                 temp2 += temp;
2260                 if (delta->xed_startoff > temp)
2261                         delta->xed_startoff = temp;
2262                 if (delta->xed_blockcount < temp2)
2263                         delta->xed_blockcount = temp2;
2264         }
2265 done:
2266         *logflagsp = rval;
2267         return error;
2268 #undef  MASK
2269 #undef  MASK2
2270 #undef  STATE_SET
2271 #undef  STATE_TEST
2272 #undef  STATE_SET_TEST
2273 #undef  SWITCH_STATE
2274 }
2275
2276 /*
2277  * Adjust the size of the new extent based on di_extsize and rt extsize.
2278  */
2279 STATIC int
2280 xfs_bmap_extsize_align(
2281         xfs_mount_t     *mp,
2282         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
2283         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
2284         xfs_extlen_t    extsz,          /* align to this extent size */
2285         int             rt,             /* is this a realtime inode? */
2286         int             eof,            /* is extent at end-of-file? */
2287         int             delay,          /* creating delalloc extent? */
2288         int             convert,        /* overwriting unwritten extent? */
2289         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
2290         xfs_extlen_t    *lenp)          /* in/out: aligned length */
2291 {
2292         xfs_fileoff_t   orig_off;       /* original offset */
2293         xfs_extlen_t    orig_alen;      /* original length */
2294         xfs_fileoff_t   orig_end;       /* original off+len */
2295         xfs_fileoff_t   nexto;          /* next file offset */
2296         xfs_fileoff_t   prevo;          /* previous file offset */
2297         xfs_fileoff_t   align_off;      /* temp for offset */
2298         xfs_extlen_t    align_alen;     /* temp for length */
2299         xfs_extlen_t    temp;           /* temp for calculations */
2300
2301         if (convert)
2302                 return 0;
2303
2304         orig_off = align_off = *offp;
2305         orig_alen = align_alen = *lenp;
2306         orig_end = orig_off + orig_alen;
2307
2308         /*
2309          * If this request overlaps an existing extent, then don't
2310          * attempt to perform any additional alignment.
2311          */
2312         if (!delay && !eof &&
2313             (orig_off >= gotp->br_startoff) &&
2314             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2315                 return 0;
2316         }
2317
2318         /*
2319          * If the file offset is unaligned vs. the extent size
2320          * we need to align it.  This will be possible unless
2321          * the file was previously written with a kernel that didn't
2322          * perform this alignment, or if a truncate shot us in the
2323          * foot.
2324          */
2325         temp = do_mod(orig_off, extsz);
2326         if (temp) {
2327                 align_alen += temp;
2328                 align_off -= temp;
2329         }
2330         /*
2331          * Same adjustment for the end of the requested area.
2332          */
2333         if ((temp = (align_alen % extsz))) {
2334                 align_alen += extsz - temp;
2335         }
2336         /*
2337          * If the previous block overlaps with this proposed allocation
2338          * then move the start forward without adjusting the length.
2339          */
2340         if (prevp->br_startoff != NULLFILEOFF) {
2341                 if (prevp->br_startblock == HOLESTARTBLOCK)
2342                         prevo = prevp->br_startoff;
2343                 else
2344                         prevo = prevp->br_startoff + prevp->br_blockcount;
2345         } else
2346                 prevo = 0;
2347         if (align_off != orig_off && align_off < prevo)
2348                 align_off = prevo;
2349         /*
2350          * If the next block overlaps with this proposed allocation
2351          * then move the start back without adjusting the length,
2352          * but not before offset 0.
2353          * This may of course make the start overlap previous block,
2354          * and if we hit the offset 0 limit then the next block
2355          * can still overlap too.
2356          */
2357         if (!eof && gotp->br_startoff != NULLFILEOFF) {
2358                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
2359                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
2360                         nexto = gotp->br_startoff + gotp->br_blockcount;
2361                 else
2362                         nexto = gotp->br_startoff;
2363         } else
2364                 nexto = NULLFILEOFF;
2365         if (!eof &&
2366             align_off + align_alen != orig_end &&
2367             align_off + align_alen > nexto)
2368                 align_off = nexto > align_alen ? nexto - align_alen : 0;
2369         /*
2370          * If we're now overlapping the next or previous extent that
2371          * means we can't fit an extsz piece in this hole.  Just move
2372          * the start forward to the first valid spot and set
2373          * the length so we hit the end.
2374          */
2375         if (align_off != orig_off && align_off < prevo)
2376                 align_off = prevo;
2377         if (align_off + align_alen != orig_end &&
2378             align_off + align_alen > nexto &&
2379             nexto != NULLFILEOFF) {
2380                 ASSERT(nexto > prevo);
2381                 align_alen = nexto - align_off;
2382         }
2383
2384         /*
2385          * If realtime, and the result isn't a multiple of the realtime
2386          * extent size we need to remove blocks until it is.
2387          */
2388         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
2389                 /*
2390                  * We're not covering the original request, or
2391                  * we won't be able to once we fix the length.
2392                  */
2393                 if (orig_off < align_off ||
2394                     orig_end > align_off + align_alen ||
2395                     align_alen - temp < orig_alen)
2396                         return XFS_ERROR(EINVAL);
2397                 /*
2398                  * Try to fix it by moving the start up.
2399                  */
2400                 if (align_off + temp <= orig_off) {
2401                         align_alen -= temp;
2402                         align_off += temp;
2403                 }
2404                 /*
2405                  * Try to fix it by moving the end in.
2406                  */
2407                 else if (align_off + align_alen - temp >= orig_end)
2408                         align_alen -= temp;
2409                 /*
2410                  * Set the start to the minimum then trim the length.
2411                  */
2412                 else {
2413                         align_alen -= orig_off - align_off;
2414                         align_off = orig_off;
2415                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
2416                 }
2417                 /*
2418                  * Result doesn't cover the request, fail it.
2419                  */
2420                 if (orig_off < align_off || orig_end > align_off + align_alen)
2421                         return XFS_ERROR(EINVAL);
2422         } else {
2423                 ASSERT(orig_off >= align_off);
2424                 ASSERT(orig_end <= align_off + align_alen);
2425         }
2426
2427 #ifdef DEBUG
2428         if (!eof && gotp->br_startoff != NULLFILEOFF)
2429                 ASSERT(align_off + align_alen <= gotp->br_startoff);
2430         if (prevp->br_startoff != NULLFILEOFF)
2431                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
2432 #endif
2433
2434         *lenp = align_alen;
2435         *offp = align_off;
2436         return 0;
2437 }
2438
2439 #define XFS_ALLOC_GAP_UNITS     4
2440
2441 STATIC void
2442 xfs_bmap_adjacent(
2443         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2444 {
2445         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
2446         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2447         xfs_mount_t     *mp;            /* mount point structure */
2448         int             nullfb;         /* true if ap->firstblock isn't set */
2449         int             rt;             /* true if inode is realtime */
2450
2451 #define ISVALID(x,y)    \
2452         (rt ? \
2453                 (x) < mp->m_sb.sb_rblocks : \
2454                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2455                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2456                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2457
2458         mp = ap->ip->i_mount;
2459         nullfb = ap->firstblock == NULLFSBLOCK;
2460         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2461         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2462         /*
2463          * If allocating at eof, and there's a previous real block,
2464          * try to use it's last block as our starting point.
2465          */
2466         if (ap->eof && ap->prevp->br_startoff != NULLFILEOFF &&
2467             !ISNULLSTARTBLOCK(ap->prevp->br_startblock) &&
2468             ISVALID(ap->prevp->br_startblock + ap->prevp->br_blockcount,
2469                     ap->prevp->br_startblock)) {
2470                 ap->rval = ap->prevp->br_startblock + ap->prevp->br_blockcount;
2471                 /*
2472                  * Adjust for the gap between prevp and us.
2473                  */
2474                 adjust = ap->off -
2475                         (ap->prevp->br_startoff + ap->prevp->br_blockcount);
2476                 if (adjust &&
2477                     ISVALID(ap->rval + adjust, ap->prevp->br_startblock))
2478                         ap->rval += adjust;
2479         }
2480         /*
2481          * If not at eof, then compare the two neighbor blocks.
2482          * Figure out whether either one gives us a good starting point,
2483          * and pick the better one.
2484          */
2485         else if (!ap->eof) {
2486                 xfs_fsblock_t   gotbno;         /* right side block number */
2487                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
2488                 xfs_fsblock_t   prevbno;        /* left side block number */
2489                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
2490
2491                 /*
2492                  * If there's a previous (left) block, select a requested
2493                  * start block based on it.
2494                  */
2495                 if (ap->prevp->br_startoff != NULLFILEOFF &&
2496                     !ISNULLSTARTBLOCK(ap->prevp->br_startblock) &&
2497                     (prevbno = ap->prevp->br_startblock +
2498                                ap->prevp->br_blockcount) &&
2499                     ISVALID(prevbno, ap->prevp->br_startblock)) {
2500                         /*
2501                          * Calculate gap to end of previous block.
2502                          */
2503                         adjust = prevdiff = ap->off -
2504                                 (ap->prevp->br_startoff +
2505                                  ap->prevp->br_blockcount);
2506                         /*
2507                          * Figure the startblock based on the previous block's
2508                          * end and the gap size.
2509                          * Heuristic!
2510                          * If the gap is large relative to the piece we're
2511                          * allocating, or using it gives us an invalid block
2512                          * number, then just use the end of the previous block.
2513                          */
2514                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2515                             ISVALID(prevbno + prevdiff,
2516                                     ap->prevp->br_startblock))
2517                                 prevbno += adjust;
2518                         else
2519                                 prevdiff += adjust;
2520                         /*
2521                          * If the firstblock forbids it, can't use it,
2522                          * must use default.
2523                          */
2524                         if (!rt && !nullfb &&
2525                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2526                                 prevbno = NULLFSBLOCK;
2527                 }
2528                 /*
2529                  * No previous block or can't follow it, just default.
2530                  */
2531                 else
2532                         prevbno = NULLFSBLOCK;
2533                 /*
2534                  * If there's a following (right) block, select a requested
2535                  * start block based on it.
2536                  */
2537                 if (!ISNULLSTARTBLOCK(ap->gotp->br_startblock)) {
2538                         /*
2539                          * Calculate gap to start of next block.
2540                          */
2541                         adjust = gotdiff = ap->gotp->br_startoff - ap->off;
2542                         /*
2543                          * Figure the startblock based on the next block's
2544                          * start and the gap size.
2545                          */
2546                         gotbno = ap->gotp->br_startblock;
2547                         /*
2548                          * Heuristic!
2549                          * If the gap is large relative to the piece we're
2550                          * allocating, or using it gives us an invalid block
2551                          * number, then just use the start of the next block
2552                          * offset by our length.
2553                          */
2554                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2555                             ISVALID(gotbno - gotdiff, gotbno))
2556                                 gotbno -= adjust;
2557                         else if (ISVALID(gotbno - ap->alen, gotbno)) {
2558                                 gotbno -= ap->alen;
2559                                 gotdiff += adjust - ap->alen;
2560                         } else
2561                                 gotdiff += adjust;
2562                         /*
2563                          * If the firstblock forbids it, can't use it,
2564                          * must use default.
2565                          */
2566                         if (!rt && !nullfb &&
2567                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2568                                 gotbno = NULLFSBLOCK;
2569                 }
2570                 /*
2571                  * No next block, just default.
2572                  */
2573                 else
2574                         gotbno = NULLFSBLOCK;
2575                 /*
2576                  * If both valid, pick the better one, else the only good
2577                  * one, else ap->rval is already set (to 0 or the inode block).
2578                  */
2579                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2580                         ap->rval = prevdiff <= gotdiff ? prevbno : gotbno;
2581                 else if (prevbno != NULLFSBLOCK)
2582                         ap->rval = prevbno;
2583                 else if (gotbno != NULLFSBLOCK)
2584                         ap->rval = gotbno;
2585         }
2586 #undef ISVALID
2587 }
2588
2589 STATIC int
2590 xfs_bmap_rtalloc(
2591         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2592 {
2593         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2594         int             error;          /* error return value */
2595         xfs_mount_t     *mp;            /* mount point structure */
2596         xfs_extlen_t    prod = 0;       /* product factor for allocators */
2597         xfs_extlen_t    ralen = 0;      /* realtime allocation length */
2598         xfs_extlen_t    align;          /* minimum allocation alignment */
2599         xfs_rtblock_t   rtb;
2600
2601         mp = ap->ip->i_mount;
2602         align = xfs_get_extsz_hint(ap->ip);
2603         prod = align / mp->m_sb.sb_rextsize;
2604         error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2605                                         align, 1, ap->eof, 0,
2606                                         ap->conv, &ap->off, &ap->alen);
2607         if (error)
2608                 return error;
2609         ASSERT(ap->alen);
2610         ASSERT(ap->alen % mp->m_sb.sb_rextsize == 0);
2611
2612         /*
2613          * If the offset & length are not perfectly aligned
2614          * then kill prod, it will just get us in trouble.
2615          */
2616         if (do_mod(ap->off, align) || ap->alen % align)
2617                 prod = 1;
2618         /*
2619          * Set ralen to be the actual requested length in rtextents.
2620          */
2621         ralen = ap->alen / mp->m_sb.sb_rextsize;
2622         /*
2623          * If the old value was close enough to MAXEXTLEN that
2624          * we rounded up to it, cut it back so it's valid again.
2625          * Note that if it's a really large request (bigger than
2626          * MAXEXTLEN), we don't hear about that number, and can't
2627          * adjust the starting point to match it.
2628          */
2629         if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2630                 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2631         /*
2632          * If it's an allocation to an empty file at offset 0,
2633          * pick an extent that will space things out in the rt area.
2634          */
2635         if (ap->eof && ap->off == 0) {
2636                 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
2637
2638                 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2639                 if (error)
2640                         return error;
2641                 ap->rval = rtx * mp->m_sb.sb_rextsize;
2642         } else {
2643                 ap->rval = 0;
2644         }
2645
2646         xfs_bmap_adjacent(ap);
2647
2648         /*
2649          * Realtime allocation, done through xfs_rtallocate_extent.
2650          */
2651         atype = ap->rval == 0 ?  XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2652         do_div(ap->rval, mp->m_sb.sb_rextsize);
2653         rtb = ap->rval;
2654         ap->alen = ralen;
2655         if ((error = xfs_rtallocate_extent(ap->tp, ap->rval, 1, ap->alen,
2656                                 &ralen, atype, ap->wasdel, prod, &rtb)))
2657                 return error;
2658         if (rtb == NULLFSBLOCK && prod > 1 &&
2659             (error = xfs_rtallocate_extent(ap->tp, ap->rval, 1,
2660                                            ap->alen, &ralen, atype,
2661                                            ap->wasdel, 1, &rtb)))
2662                 return error;
2663         ap->rval = rtb;
2664         if (ap->rval != NULLFSBLOCK) {
2665                 ap->rval *= mp->m_sb.sb_rextsize;
2666                 ralen *= mp->m_sb.sb_rextsize;
2667                 ap->alen = ralen;
2668                 ap->ip->i_d.di_nblocks += ralen;
2669                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2670                 if (ap->wasdel)
2671                         ap->ip->i_delayed_blks -= ralen;
2672                 /*
2673                  * Adjust the disk quota also. This was reserved
2674                  * earlier.
2675                  */
2676                 XFS_TRANS_MOD_DQUOT_BYINO(mp, ap->tp, ap->ip,
2677                         ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2678                                         XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2679         } else {
2680                 ap->alen = 0;
2681         }
2682         return 0;
2683 }
2684
2685 STATIC int
2686 xfs_bmap_btalloc(
2687         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2688 {
2689         xfs_mount_t     *mp;            /* mount point structure */
2690         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2691         xfs_extlen_t    align;          /* minimum allocation alignment */
2692         xfs_agnumber_t  ag;
2693         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2694         xfs_agnumber_t  startag;
2695         xfs_alloc_arg_t args;
2696         xfs_extlen_t    blen;
2697         xfs_extlen_t    delta;
2698         xfs_extlen_t    longest;
2699         xfs_extlen_t    need;
2700         xfs_extlen_t    nextminlen = 0;
2701         xfs_perag_t     *pag;
2702         int             nullfb;         /* true if ap->firstblock isn't set */
2703         int             isaligned;
2704         int             notinit;
2705         int             tryagain;
2706         int             error;
2707
2708         mp = ap->ip->i_mount;
2709         align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
2710         if (unlikely(align)) {
2711                 error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2712                                                 align, 0, ap->eof, 0, ap->conv,
2713                                                 &ap->off, &ap->alen);
2714                 ASSERT(!error);
2715                 ASSERT(ap->alen);
2716         }
2717         nullfb = ap->firstblock == NULLFSBLOCK;
2718         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2719         if (nullfb) {
2720                 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
2721                         ag = xfs_filestream_lookup_ag(ap->ip);
2722                         ag = (ag != NULLAGNUMBER) ? ag : 0;
2723                         ap->rval = XFS_AGB_TO_FSB(mp, ag, 0);
2724                 } else {
2725                         ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2726                 }
2727         } else
2728                 ap->rval = ap->firstblock;
2729
2730         xfs_bmap_adjacent(ap);
2731
2732         /*
2733          * If allowed, use ap->rval; otherwise must use firstblock since
2734          * it's in the right allocation group.
2735          */
2736         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->rval) == fb_agno)
2737                 ;
2738         else
2739                 ap->rval = ap->firstblock;
2740         /*
2741          * Normal allocation, done through xfs_alloc_vextent.
2742          */
2743         tryagain = isaligned = 0;
2744         args.tp = ap->tp;
2745         args.mp = mp;
2746         args.fsbno = ap->rval;
2747         args.maxlen = MIN(ap->alen, mp->m_sb.sb_agblocks);
2748         args.firstblock = ap->firstblock;
2749         blen = 0;
2750         if (nullfb) {
2751                 if (ap->userdata && xfs_inode_is_filestream(ap->ip))
2752                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
2753                 else
2754                         args.type = XFS_ALLOCTYPE_START_BNO;
2755                 args.total = ap->total;
2756
2757                 /*
2758                  * Search for an allocation group with a single extent
2759                  * large enough for the request.
2760                  *
2761                  * If one isn't found, then adjust the minimum allocation
2762                  * size to the largest space found.
2763                  */
2764                 startag = ag = XFS_FSB_TO_AGNO(mp, args.fsbno);
2765                 if (startag == NULLAGNUMBER)
2766                         startag = ag = 0;
2767                 notinit = 0;
2768                 down_read(&mp->m_peraglock);
2769                 while (blen < ap->alen) {
2770                         pag = &mp->m_perag[ag];
2771                         if (!pag->pagf_init &&
2772                             (error = xfs_alloc_pagf_init(mp, args.tp,
2773                                     ag, XFS_ALLOC_FLAG_TRYLOCK))) {
2774                                 up_read(&mp->m_peraglock);
2775                                 return error;
2776                         }
2777                         /*
2778                          * See xfs_alloc_fix_freelist...
2779                          */
2780                         if (pag->pagf_init) {
2781                                 need = XFS_MIN_FREELIST_PAG(pag, mp);
2782                                 delta = need > pag->pagf_flcount ?
2783                                         need - pag->pagf_flcount : 0;
2784                                 longest = (pag->pagf_longest > delta) ?
2785                                         (pag->pagf_longest - delta) :
2786                                         (pag->pagf_flcount > 0 ||
2787                                          pag->pagf_longest > 0);
2788                                 if (blen < longest)
2789                                         blen = longest;
2790                         } else
2791                                 notinit = 1;
2792
2793                         if (xfs_inode_is_filestream(ap->ip)) {
2794                                 if (blen >= ap->alen)
2795                                         break;
2796
2797                                 if (ap->userdata) {
2798                                         /*
2799                                          * If startag is an invalid AG, we've
2800                                          * come here once before and
2801                                          * xfs_filestream_new_ag picked the
2802                                          * best currently available.
2803                                          *
2804                                          * Don't continue looping, since we
2805                                          * could loop forever.
2806                                          */
2807                                         if (startag == NULLAGNUMBER)
2808                                                 break;
2809
2810                                         error = xfs_filestream_new_ag(ap, &ag);
2811                                         if (error) {
2812                                                 up_read(&mp->m_peraglock);
2813                                                 return error;
2814                                         }
2815
2816                                         /* loop again to set 'blen'*/
2817                                         startag = NULLAGNUMBER;
2818                                         continue;
2819                                 }
2820                         }
2821                         if (++ag == mp->m_sb.sb_agcount)
2822                                 ag = 0;
2823                         if (ag == startag)
2824                                 break;
2825                 }
2826                 up_read(&mp->m_peraglock);
2827                 /*
2828                  * Since the above loop did a BUF_TRYLOCK, it is
2829                  * possible that there is space for this request.
2830                  */
2831                 if (notinit || blen < ap->minlen)
2832                         args.minlen = ap->minlen;
2833                 /*
2834                  * If the best seen length is less than the request
2835                  * length, use the best as the minimum.
2836                  */
2837                 else if (blen < ap->alen)
2838                         args.minlen = blen;
2839                 /*
2840                  * Otherwise we've seen an extent as big as alen,
2841                  * use that as the minimum.
2842                  */
2843                 else
2844                         args.minlen = ap->alen;
2845
2846                 /*
2847                  * set the failure fallback case to look in the selected
2848                  * AG as the stream may have moved.
2849                  */
2850                 if (xfs_inode_is_filestream(ap->ip))
2851                         ap->rval = args.fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
2852         } else if (ap->low) {
2853                 if (xfs_inode_is_filestream(ap->ip))
2854                         args.type = XFS_ALLOCTYPE_FIRST_AG;
2855                 else
2856                         args.type = XFS_ALLOCTYPE_START_BNO;
2857                 args.total = args.minlen = ap->minlen;
2858         } else {
2859                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2860                 args.total = ap->total;
2861                 args.minlen = ap->minlen;
2862         }
2863         /* apply extent size hints if obtained earlier */
2864         if (unlikely(align)) {
2865                 args.prod = align;
2866                 if ((args.mod = (xfs_extlen_t)do_mod(ap->off, args.prod)))
2867                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2868         } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
2869                 args.prod = 1;
2870                 args.mod = 0;
2871         } else {
2872                 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
2873                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->off, args.prod))))
2874                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2875         }
2876         /*
2877          * If we are not low on available data blocks, and the
2878          * underlying logical volume manager is a stripe, and
2879          * the file offset is zero then try to allocate data
2880          * blocks on stripe unit boundary.
2881          * NOTE: ap->aeof is only set if the allocation length
2882          * is >= the stripe unit and the allocation offset is
2883          * at the end of file.
2884          */
2885         if (!ap->low && ap->aeof) {
2886                 if (!ap->off) {
2887                         args.alignment = mp->m_dalign;
2888                         atype = args.type;
2889                         isaligned = 1;
2890                         /*
2891                          * Adjust for alignment
2892                          */
2893                         if (blen > args.alignment && blen <= ap->alen)
2894                                 args.minlen = blen - args.alignment;
2895                         args.minalignslop = 0;
2896                 } else {
2897                         /*
2898                          * First try an exact bno allocation.
2899                          * If it fails then do a near or start bno
2900                          * allocation with alignment turned on.
2901                          */
2902                         atype = args.type;
2903                         tryagain = 1;
2904                         args.type = XFS_ALLOCTYPE_THIS_BNO;
2905                         args.alignment = 1;
2906                         /*
2907                          * Compute the minlen+alignment for the
2908                          * next case.  Set slop so that the value
2909                          * of minlen+alignment+slop doesn't go up
2910                          * between the calls.
2911                          */
2912                         if (blen > mp->m_dalign && blen <= ap->alen)
2913                                 nextminlen = blen - mp->m_dalign;
2914                         else
2915                                 nextminlen = args.minlen;
2916                         if (nextminlen + mp->m_dalign > args.minlen + 1)
2917                                 args.minalignslop =
2918                                         nextminlen + mp->m_dalign -
2919                                         args.minlen - 1;
2920                         else
2921                                 args.minalignslop = 0;
2922                 }
2923         } else {
2924                 args.alignment = 1;
2925                 args.minalignslop = 0;
2926         }
2927         args.minleft = ap->minleft;
2928         args.wasdel = ap->wasdel;
2929         args.isfl = 0;
2930         args.userdata = ap->userdata;
2931         if ((error = xfs_alloc_vextent(&args)))
2932                 return error;
2933         if (tryagain && args.fsbno == NULLFSBLOCK) {
2934                 /*
2935                  * Exact allocation failed. Now try with alignment
2936                  * turned on.
2937                  */
2938                 args.type = atype;
2939                 args.fsbno = ap->rval;
2940                 args.alignment = mp->m_dalign;
2941                 args.minlen = nextminlen;
2942                 args.minalignslop = 0;
2943                 isaligned = 1;
2944                 if ((error = xfs_alloc_vextent(&args)))
2945                         return error;
2946         }
2947         if (isaligned && args.fsbno == NULLFSBLOCK) {
2948                 /*
2949                  * allocation failed, so turn off alignment and
2950                  * try again.
2951                  */
2952                 args.type = atype;
2953                 args.fsbno = ap->rval;
2954                 args.alignment = 0;
2955                 if ((error = xfs_alloc_vextent(&args)))
2956                         return error;
2957         }
2958         if (args.fsbno == NULLFSBLOCK && nullfb &&
2959             args.minlen > ap->minlen) {
2960                 args.minlen = ap->minlen;
2961                 args.type = XFS_ALLOCTYPE_START_BNO;
2962                 args.fsbno = ap->rval;
2963                 if ((error = xfs_alloc_vextent(&args)))
2964                         return error;
2965         }
2966         if (args.fsbno == NULLFSBLOCK && nullfb) {
2967                 args.fsbno = 0;
2968                 args.type = XFS_ALLOCTYPE_FIRST_AG;
2969                 args.total = ap->minlen;
2970                 args.minleft = 0;
2971                 if ((error = xfs_alloc_vextent(&args)))
2972                         return error;
2973                 ap->low = 1;
2974         }
2975         if (args.fsbno != NULLFSBLOCK) {
2976                 ap->firstblock = ap->rval = args.fsbno;
2977                 ASSERT(nullfb || fb_agno == args.agno ||
2978                        (ap->low && fb_agno < args.agno));
2979                 ap->alen = args.len;
2980                 ap->ip->i_d.di_nblocks += args.len;
2981                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2982                 if (ap->wasdel)
2983                         ap->ip->i_delayed_blks -= args.len;
2984                 /*
2985                  * Adjust the disk quota also. This was reserved
2986                  * earlier.
2987                  */
2988                 XFS_TRANS_MOD_DQUOT_BYINO(mp, ap->tp, ap->ip,
2989                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2990                                         XFS_TRANS_DQ_BCOUNT,
2991                         (long) args.len);
2992         } else {
2993                 ap->rval = NULLFSBLOCK;
2994                 ap->alen = 0;
2995         }
2996         return 0;
2997 }
2998
2999 /*
3000  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
3001  * It figures out where to ask the underlying allocator to put the new extent.
3002  */
3003 STATIC int
3004 xfs_bmap_alloc(
3005         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
3006 {
3007         if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
3008                 return xfs_bmap_rtalloc(ap);
3009         return xfs_bmap_btalloc(ap);
3010 }
3011
3012 /*
3013  * Transform a btree format file with only one leaf node, where the
3014  * extents list will fit in the inode, into an extents format file.
3015  * Since the file extents are already in-core, all we have to do is
3016  * give up the space for the btree root and pitch the leaf block.
3017  */
3018 STATIC int                              /* error */
3019 xfs_bmap_btree_to_extents(
3020         xfs_trans_t             *tp,    /* transaction pointer */
3021         xfs_inode_t             *ip,    /* incore inode pointer */
3022         xfs_btree_cur_t         *cur,   /* btree cursor */
3023         int                     *logflagsp, /* inode logging flags */
3024         int                     whichfork)  /* data or attr fork */
3025 {
3026         /* REFERENCED */
3027         xfs_bmbt_block_t        *cblock;/* child btree block */
3028         xfs_fsblock_t           cbno;   /* child block number */
3029         xfs_buf_t               *cbp;   /* child block's buffer */
3030         int                     error;  /* error return value */
3031         xfs_ifork_t             *ifp;   /* inode fork data */
3032         xfs_mount_t             *mp;    /* mount point structure */
3033         __be64                  *pp;    /* ptr to block address */
3034         xfs_bmbt_block_t        *rblock;/* root btree block */
3035
3036         ifp = XFS_IFORK_PTR(ip, whichfork);
3037         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3038         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
3039         rblock = ifp->if_broot;
3040         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
3041         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
3042         ASSERT(XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes) == 1);
3043         mp = ip->i_mount;
3044         pp = XFS_BMAP_BROOT_PTR_ADDR(rblock, 1, ifp->if_broot_bytes);
3045         cbno = be64_to_cpu(*pp);
3046         *logflagsp = 0;
3047 #ifdef DEBUG
3048         if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
3049                 return error;
3050 #endif
3051         if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
3052                         XFS_BMAP_BTREE_REF)))
3053                 return error;
3054         cblock = XFS_BUF_TO_BMBT_BLOCK(cbp);
3055         if ((error = xfs_btree_check_lblock(cur, cblock, 0, cbp)))
3056                 return error;
3057         xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
3058         ip->i_d.di_nblocks--;
3059         XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
3060         xfs_trans_binval(tp, cbp);
3061         if (cur->bc_bufs[0] == cbp)
3062                 cur->bc_bufs[0] = NULL;
3063         xfs_iroot_realloc(ip, -1, whichfork);
3064         ASSERT(ifp->if_broot == NULL);
3065         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
3066         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3067         *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
3068         return 0;
3069 }
3070
3071 /*
3072  * Called by xfs_bmapi to update file extent records and the btree
3073  * after removing space (or undoing a delayed allocation).
3074  */
3075 STATIC int                              /* error */
3076 xfs_bmap_del_extent(
3077         xfs_inode_t             *ip,    /* incore inode pointer */
3078         xfs_trans_t             *tp,    /* current transaction pointer */
3079         xfs_extnum_t            idx,    /* extent number to update/delete */
3080         xfs_bmap_free_t         *flist, /* list of extents to be freed */
3081         xfs_btree_cur_t         *cur,   /* if null, not a btree */
3082         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
3083         int                     *logflagsp, /* inode logging flags */
3084         xfs_extdelta_t          *delta, /* Change made to incore extents */
3085         int                     whichfork, /* data or attr fork */
3086         int                     rsvd)   /* OK to allocate reserved blocks */
3087 {
3088         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
3089         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
3090         xfs_fsblock_t           del_endblock=0; /* first block past del */
3091         xfs_fileoff_t           del_endoff;     /* first offset past del */
3092         int                     delay;  /* current block is delayed allocated */
3093         int                     do_fx;  /* free extent at end of routine */
3094         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
3095         int                     error;  /* error return value */
3096         int                     flags;  /* inode logging flags */
3097         xfs_bmbt_irec_t         got;    /* current extent entry */
3098         xfs_fileoff_t           got_endoff;     /* first offset past got */
3099         int                     i;      /* temp state */
3100         xfs_ifork_t             *ifp;   /* inode fork pointer */
3101         xfs_mount_t             *mp;    /* mount structure */
3102         xfs_filblks_t           nblks;  /* quota/sb block count */
3103         xfs_bmbt_irec_t         new;    /* new record to be inserted */
3104         /* REFERENCED */
3105         uint                    qfield; /* quota field to update */
3106         xfs_filblks_t           temp;   /* for indirect length calculations */
3107         xfs_filblks_t           temp2;  /* for indirect length calculations */
3108
3109         XFS_STATS_INC(xs_del_exlist);
3110         mp = ip->i_mount;
3111         ifp = XFS_IFORK_PTR(ip, whichfork);
3112         ASSERT((idx >= 0) && (idx < ifp->if_bytes /
3113                 (uint)sizeof(xfs_bmbt_rec_t)));
3114         ASSERT(del->br_blockcount > 0);
3115         ep = xfs_iext_get_ext(ifp, idx);
3116         xfs_bmbt_get_all(ep, &got);
3117         ASSERT(got.br_startoff <= del->br_startoff);
3118         del_endoff = del->br_startoff + del->br_blockcount;
3119         got_endoff = got.br_startoff + got.br_blockcount;
3120         ASSERT(got_endoff >= del_endoff);
3121         delay = ISNULLSTARTBLOCK(got.br_startblock);
3122         ASSERT(ISNULLSTARTBLOCK(del->br_startblock) == delay);
3123         flags = 0;
3124         qfield = 0;
3125         error = 0;
3126         /*
3127          * If deleting a real allocation, must free up the disk space.
3128          */
3129         if (!delay) {
3130                 flags = XFS_ILOG_CORE;
3131                 /*
3132                  * Realtime allocation.  Free it and record di_nblocks update.
3133                  */
3134                 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
3135                         xfs_fsblock_t   bno;
3136                         xfs_filblks_t   len;
3137
3138                         ASSERT(do_mod(del->br_blockcount,
3139                                       mp->m_sb.sb_rextsize) == 0);
3140                         ASSERT(do_mod(del->br_startblock,
3141                                       mp->m_sb.sb_rextsize) == 0);
3142                         bno = del->br_startblock;
3143                         len = del->br_blockcount;
3144                         do_div(bno, mp->m_sb.sb_rextsize);
3145                         do_div(len, mp->m_sb.sb_rextsize);
3146                         if ((error = xfs_rtfree_extent(ip->i_transp, bno,
3147                                         (xfs_extlen_t)len)))
3148                                 goto done;
3149                         do_fx = 0;
3150                         nblks = len * mp->m_sb.sb_rextsize;
3151                         qfield = XFS_TRANS_DQ_RTBCOUNT;
3152                 }
3153                 /*
3154                  * Ordinary allocation.
3155                  */
3156                 else {
3157                         do_fx = 1;
3158                         nblks = del->br_blockcount;
3159                         qfield = XFS_TRANS_DQ_BCOUNT;
3160                 }
3161                 /*
3162                  * Set up del_endblock and cur for later.
3163                  */
3164                 del_endblock = del->br_startblock + del->br_blockcount;
3165                 if (cur) {
3166                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
3167                                         got.br_startblock, got.br_blockcount,
3168                                         &i)))
3169                                 goto done;
3170                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3171                 }
3172                 da_old = da_new = 0;
3173         } else {
3174                 da_old = STARTBLOCKVAL(got.br_startblock);
3175                 da_new = 0;
3176                 nblks = 0;
3177                 do_fx = 0;
3178         }
3179         /*
3180          * Set flag value to use in switch statement.
3181          * Left-contig is 2, right-contig is 1.
3182          */
3183         switch (((got.br_startoff == del->br_startoff) << 1) |
3184                 (got_endoff == del_endoff)) {
3185         case 3:
3186                 /*
3187                  * Matches the whole extent.  Delete the entry.
3188                  */
3189                 XFS_BMAP_TRACE_DELETE("3", ip, idx, 1, whichfork);
3190                 xfs_iext_remove(ifp, idx, 1);
3191                 ifp->if_lastex = idx;
3192                 if (delay)
3193                         break;
3194                 XFS_IFORK_NEXT_SET(ip, whichfork,
3195                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
3196                 flags |= XFS_ILOG_CORE;
3197                 if (!cur) {
3198                         flags |= XFS_ILOG_FEXT(whichfork);
3199                         break;
3200                 }
3201                 if ((error = xfs_bmbt_delete(cur, &i)))
3202                         goto done;
3203                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3204                 break;
3205
3206         case 2:
3207                 /*
3208                  * Deleting the first part of the extent.
3209                  */
3210                 XFS_BMAP_TRACE_PRE_UPDATE("2", ip, idx, whichfork);
3211                 xfs_bmbt_set_startoff(ep, del_endoff);
3212                 temp = got.br_blockcount - del->br_blockcount;
3213                 xfs_bmbt_set_blockcount(ep, temp);
3214                 ifp->if_lastex = idx;
3215                 if (delay) {
3216                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
3217                                 da_old);
3218                         xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3219                         XFS_BMAP_TRACE_POST_UPDATE("2", ip, idx,
3220                                 whichfork);
3221                         da_new = temp;
3222                         break;
3223                 }
3224                 xfs_bmbt_set_startblock(ep, del_endblock);
3225                 XFS_BMAP_TRACE_POST_UPDATE("2", ip, idx, whichfork);
3226                 if (!cur) {
3227                         flags |= XFS_ILOG_FEXT(whichfork);
3228                         break;
3229                 }
3230                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
3231                                 got.br_blockcount - del->br_blockcount,
3232                                 got.br_state)))
3233                         goto done;
3234                 break;
3235
3236         case 1:
3237                 /*
3238                  * Deleting the last part of the extent.
3239                  */
3240                 temp = got.br_blockcount - del->br_blockcount;
3241                 XFS_BMAP_TRACE_PRE_UPDATE("1", ip, idx, whichfork);
3242                 xfs_bmbt_set_blockcount(ep, temp);
3243                 ifp->if_lastex = idx;
3244                 if (delay) {
3245                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
3246                                 da_old);
3247                         xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3248                         XFS_BMAP_TRACE_POST_UPDATE("1", ip, idx,
3249                                 whichfork);
3250                         da_new = temp;
3251                         break;
3252                 }
3253                 XFS_BMAP_TRACE_POST_UPDATE("1", ip, idx, whichfork);
3254                 if (!cur) {
3255                         flags |= XFS_ILOG_FEXT(whichfork);
3256                         break;
3257                 }
3258                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
3259                                 got.br_startblock,
3260                                 got.br_blockcount - del->br_blockcount,
3261                                 got.br_state)))
3262                         goto done;
3263                 break;
3264
3265         case 0:
3266                 /*
3267                  * Deleting the middle of the extent.
3268                  */
3269                 temp = del->br_startoff - got.br_startoff;
3270                 XFS_BMAP_TRACE_PRE_UPDATE("0", ip, idx, whichfork);
3271                 xfs_bmbt_set_blockcount(ep, temp);
3272                 new.br_startoff = del_endoff;
3273                 temp2 = got_endoff - del_endoff;
3274                 new.br_blockcount = temp2;
3275                 new.br_state = got.br_state;
3276                 if (!delay) {
3277                         new.br_startblock = del_endblock;
3278                         flags |= XFS_ILOG_CORE;
3279                         if (cur) {
3280                                 if ((error = xfs_bmbt_update(cur,
3281                                                 got.br_startoff,
3282                                                 got.br_startblock, temp,
3283                                                 got.br_state)))
3284                                         goto done;
3285                                 if ((error = xfs_btree_increment(cur, 0, &i)))
3286                                         goto done;
3287                                 cur->bc_rec.b = new;
3288                                 error = xfs_bmbt_insert(cur, &i);
3289                                 if (error && error != ENOSPC)
3290                                         goto done;
3291                                 /*
3292                                  * If get no-space back from btree insert,
3293                                  * it tried a split, and we have a zero
3294                                  * block reservation.
3295                                  * Fix up our state and return the error.
3296                                  */
3297                                 if (error == ENOSPC) {
3298                                         /*
3299                                          * Reset the cursor, don't trust
3300                                          * it after any insert operation.
3301                                          */
3302                                         if ((error = xfs_bmbt_lookup_eq(cur,
3303                                                         got.br_startoff,
3304                                                         got.br_startblock,
3305                                                         temp, &i)))
3306                                                 goto done;
3307                                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3308                                         /*
3309                                          * Update the btree record back
3310                                          * to the original value.
3311                                          */
3312                                         if ((error = xfs_bmbt_update(cur,
3313                                                         got.br_startoff,
3314                                                         got.br_startblock,
3315                                                         got.br_blockcount,
3316                                                         got.br_state)))
3317                                                 goto done;
3318                                         /*
3319                                          * Reset the extent record back
3320                                          * to the original value.
3321                                          */
3322                                         xfs_bmbt_set_blockcount(ep,
3323                                                 got.br_blockcount);
3324                                         flags = 0;
3325                                         error = XFS_ERROR(ENOSPC);
3326                                         goto done;
3327                                 }
3328                                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3329                         } else
3330                                 flags |= XFS_ILOG_FEXT(whichfork);
3331                         XFS_IFORK_NEXT_SET(ip, whichfork,
3332                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
3333                 } else {
3334                         ASSERT(whichfork == XFS_DATA_FORK);
3335                         temp = xfs_bmap_worst_indlen(ip, temp);
3336                         xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3337                         temp2 = xfs_bmap_worst_indlen(ip, temp2);
3338                         new.br_startblock = NULLSTARTBLOCK((int)temp2);
3339                         da_new = temp + temp2;
3340                         while (da_new > da_old) {
3341                                 if (temp) {
3342                                         temp--;
3343                                         da_new--;
3344                                         xfs_bmbt_set_startblock(ep,
3345                                                 NULLSTARTBLOCK((int)temp));
3346                                 }
3347                                 if (da_new == da_old)
3348                                         break;
3349                                 if (temp2) {
3350                                         temp2--;
3351                                         da_new--;
3352                                         new.br_startblock =
3353                                                 NULLSTARTBLOCK((int)temp2);
3354                                 }
3355                         }
3356                 }
3357                 XFS_BMAP_TRACE_POST_UPDATE("0", ip, idx, whichfork);
3358                 XFS_BMAP_TRACE_INSERT("0", ip, idx + 1, 1, &new, NULL,
3359                         whichfork);
3360                 xfs_iext_insert(ifp, idx + 1, 1, &new);
3361                 ifp->if_lastex = idx + 1;
3362                 break;
3363         }
3364         /*
3365          * If we need to, add to list of extents to delete.
3366          */
3367         if (do_fx)
3368                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
3369                         mp);
3370         /*
3371          * Adjust inode # blocks in the file.
3372          */
3373         if (nblks)
3374                 ip->i_d.di_nblocks -= nblks;
3375         /*
3376          * Adjust quota data.
3377          */
3378         if (qfield)
3379                 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, qfield, (long)-nblks);
3380
3381         /*
3382          * Account for change in delayed indirect blocks.
3383          * Nothing to do for disk quota accounting here.
3384          */
3385         ASSERT(da_old >= da_new);
3386         if (da_old > da_new)
3387                 xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS, (int64_t)(da_old - da_new),
3388                         rsvd);
3389         if (delta) {
3390                 /* DELTA: report the original extent. */
3391                 if (delta->xed_startoff > got.br_startoff)
3392                         delta->xed_startoff = got.br_startoff;
3393                 if (delta->xed_blockcount < got.br_startoff+got.br_blockcount)
3394                         delta->xed_blockcount = got.br_startoff +
3395                                                         got.br_blockcount;
3396         }
3397 done:
3398         *logflagsp = flags;
3399         return error;
3400 }
3401
3402 /*
3403  * Remove the entry "free" from the free item list.  Prev points to the
3404  * previous entry, unless "free" is the head of the list.
3405  */
3406 STATIC void
3407 xfs_bmap_del_free(
3408         xfs_bmap_free_t         *flist, /* free item list header */
3409         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
3410         xfs_bmap_free_item_t    *free)  /* list item to be freed */
3411 {
3412         if (prev)
3413                 prev->xbfi_next = free->xbfi_next;
3414         else
3415                 flist->xbf_first = free->xbfi_next;
3416         flist->xbf_count--;
3417         kmem_zone_free(xfs_bmap_free_item_zone, free);
3418 }
3419
3420 /*
3421  * Convert an extents-format file into a btree-format file.
3422  * The new file will have a root block (in the inode) and a single child block.
3423  */
3424 STATIC int                                      /* error */
3425 xfs_bmap_extents_to_btree(
3426         xfs_trans_t             *tp,            /* transaction pointer */
3427         xfs_inode_t             *ip,            /* incore inode pointer */
3428         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
3429         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
3430         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
3431         int                     wasdel,         /* converting a delayed alloc */
3432         int                     *logflagsp,     /* inode logging flags */
3433         int                     whichfork)      /* data or attr fork */
3434 {
3435         xfs_bmbt_block_t        *ablock;        /* allocated (child) bt block */
3436         xfs_buf_t               *abp;           /* buffer for ablock */
3437         xfs_alloc_arg_t         args;           /* allocation arguments */
3438         xfs_bmbt_rec_t          *arp;           /* child record pointer */
3439         xfs_bmbt_block_t        *block;         /* btree root block */
3440         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
3441         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
3442         int                     error;          /* error return value */
3443         xfs_extnum_t            i, cnt;         /* extent record index */
3444         xfs_ifork_t             *ifp;           /* inode fork pointer */
3445         xfs_bmbt_key_t          *kp;            /* root block key pointer */
3446         xfs_mount_t             *mp;            /* mount structure */
3447         xfs_extnum_t            nextents;       /* number of file extents */
3448         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
3449
3450         ifp = XFS_IFORK_PTR(ip, whichfork);
3451         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3452         ASSERT(ifp->if_ext_max ==
3453                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
3454         /*
3455          * Make space in the inode incore.
3456          */
3457         xfs_iroot_realloc(ip, 1, whichfork);
3458         ifp->if_flags |= XFS_IFBROOT;
3459         /*
3460          * Fill in the root.
3461          */
3462         block = ifp->if_broot;
3463         block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3464         block->bb_level = cpu_to_be16(1);
3465         block->bb_numrecs = cpu_to_be16(1);
3466         block->bb_leftsib = cpu_to_be64(NULLDFSBNO);
3467         block->bb_rightsib = cpu_to_be64(NULLDFSBNO);
3468         /*
3469          * Need a cursor.  Can't allocate until bb_level is filled in.
3470          */
3471         mp = ip->i_mount;
3472         cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
3473         cur->bc_private.b.firstblock = *firstblock;
3474         cur->bc_private.b.flist = flist;
3475         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3476         /*
3477          * Convert to a btree with two levels, one record in root.
3478          */
3479         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3480         args.tp = tp;
3481         args.mp = mp;
3482         args.firstblock = *firstblock;
3483         if (*firstblock == NULLFSBLOCK) {
3484                 args.type = XFS_ALLOCTYPE_START_BNO;
3485                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3486         } else if (flist->xbf_low) {
3487                 args.type = XFS_ALLOCTYPE_START_BNO;
3488                 args.fsbno = *firstblock;
3489         } else {
3490                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3491                 args.fsbno = *firstblock;
3492         }
3493         args.minlen = args.maxlen = args.prod = 1;
3494         args.total = args.minleft = args.alignment = args.mod = args.isfl =
3495                 args.minalignslop = 0;
3496         args.wasdel = wasdel;
3497         *logflagsp = 0;
3498         if ((error = xfs_alloc_vextent(&args))) {
3499                 xfs_iroot_realloc(ip, -1, whichfork);
3500                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3501                 return error;
3502         }
3503         /*
3504          * Allocation can't fail, the space was reserved.
3505          */
3506         ASSERT(args.fsbno != NULLFSBLOCK);
3507         ASSERT(*firstblock == NULLFSBLOCK ||
3508                args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3509                (flist->xbf_low &&
3510                 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3511         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3512         cur->bc_private.b.allocated++;
3513         ip->i_d.di_nblocks++;
3514         XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3515         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3516         /*
3517          * Fill in the child block.
3518          */
3519         ablock = XFS_BUF_TO_BMBT_BLOCK(abp);
3520         ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3521         ablock->bb_level = 0;
3522         ablock->bb_leftsib = cpu_to_be64(NULLDFSBNO);
3523         ablock->bb_rightsib = cpu_to_be64(NULLDFSBNO);
3524         arp = XFS_BMAP_REC_IADDR(ablock, 1, cur);
3525         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3526         for (cnt = i = 0; i < nextents; i++) {
3527                 ep = xfs_iext_get_ext(ifp, i);
3528                 if (!ISNULLSTARTBLOCK(xfs_bmbt_get_startblock(ep))) {
3529                         arp->l0 = cpu_to_be64(ep->l0);
3530                         arp->l1 = cpu_to_be64(ep->l1);
3531                         arp++; cnt++;
3532                 }
3533         }
3534         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3535         ablock->bb_numrecs = cpu_to_be16(cnt);
3536         /*
3537          * Fill in the root key and pointer.
3538          */
3539         kp = XFS_BMAP_KEY_IADDR(block, 1, cur);
3540         arp = XFS_BMAP_REC_IADDR(ablock, 1, cur);
3541         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
3542         pp = XFS_BMAP_PTR_IADDR(block, 1, cur);
3543         *pp = cpu_to_be64(args.fsbno);
3544         /*
3545          * Do all this logging at the end so that
3546          * the root is at the right level.
3547          */
3548         xfs_bmbt_log_block(cur, abp, XFS_BB_ALL_BITS);
3549         xfs_bmbt_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3550         ASSERT(*curp == NULL);
3551         *curp = cur;
3552         *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FBROOT(whichfork);
3553         return 0;
3554 }
3555
3556 /*
3557  * Helper routine to reset inode di_forkoff field when switching
3558  * attribute fork from local to extent format - we reset it where
3559  * possible to make space available for inline data fork extents.
3560  */
3561 STATIC void
3562 xfs_bmap_forkoff_reset(
3563         xfs_mount_t     *mp,
3564         xfs_inode_t     *ip,
3565         int             whichfork)
3566 {
3567         if (whichfork == XFS_ATTR_FORK &&
3568             (ip->i_d.di_format != XFS_DINODE_FMT_DEV) &&
3569             (ip->i_d.di_format != XFS_DINODE_FMT_UUID) &&
3570             (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3571             ((mp->m_attroffset >> 3) > ip->i_d.di_forkoff)) {
3572                 ip->i_d.di_forkoff = mp->m_attroffset >> 3;
3573                 ip->i_df.if_ext_max = XFS_IFORK_DSIZE(ip) /
3574                                         (uint)sizeof(xfs_bmbt_rec_t);
3575                 ip->i_afp->if_ext_max = XFS_IFORK_ASIZE(ip) /
3576                                         (uint)sizeof(xfs_bmbt_rec_t);
3577         }
3578 }
3579
3580 /*
3581  * Convert a local file to an extents file.
3582  * This code is out of bounds for data forks of regular files,
3583  * since the file data needs to get logged so things will stay consistent.
3584  * (The bmap-level manipulations are ok, though).
3585  */
3586 STATIC int                              /* error */
3587 xfs_bmap_local_to_extents(
3588         xfs_trans_t     *tp,            /* transaction pointer */
3589         xfs_inode_t     *ip,            /* incore inode pointer */
3590         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
3591         xfs_extlen_t    total,          /* total blocks needed by transaction */
3592         int             *logflagsp,     /* inode logging flags */
3593         int             whichfork)      /* data or attr fork */
3594 {
3595         int             error;          /* error return value */
3596         int             flags;          /* logging flags returned */
3597         xfs_ifork_t     *ifp;           /* inode fork pointer */
3598
3599         /*
3600          * We don't want to deal with the case of keeping inode data inline yet.
3601          * So sending the data fork of a regular inode is invalid.
3602          */
3603         ASSERT(!((ip->i_d.di_mode & S_IFMT) == S_IFREG &&
3604                  whichfork == XFS_DATA_FORK));
3605         ifp = XFS_IFORK_PTR(ip, whichfork);
3606         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3607         flags = 0;
3608         error = 0;
3609         if (ifp->if_bytes) {
3610                 xfs_alloc_arg_t args;   /* allocation arguments */
3611                 xfs_buf_t       *bp;    /* buffer for extent block */
3612                 xfs_bmbt_rec_host_t *ep;/* extent record pointer */
3613
3614                 args.tp = tp;
3615                 args.mp = ip->i_mount;
3616                 args.firstblock = *firstblock;
3617                 ASSERT((ifp->if_flags &
3618                         (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == XFS_IFINLINE);
3619                 /*
3620                  * Allocate a block.  We know we need only one, since the
3621                  * file currently fits in an inode.
3622                  */
3623                 if (*firstblock == NULLFSBLOCK) {
3624                         args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3625                         args.type = XFS_ALLOCTYPE_START_BNO;
3626                 } else {
3627                         args.fsbno = *firstblock;
3628                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
3629                 }
3630                 args.total = total;
3631                 args.mod = args.minleft = args.alignment = args.wasdel =
3632                         args.isfl = args.minalignslop = 0;
3633                 args.minlen = args.maxlen = args.prod = 1;
3634                 if ((error = xfs_alloc_vextent(&args)))
3635                         goto done;
3636                 /*
3637                  * Can't fail, the space was reserved.
3638                  */
3639                 ASSERT(args.fsbno != NULLFSBLOCK);
3640                 ASSERT(args.len == 1);
3641                 *firstblock = args.fsbno;
3642                 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3643                 memcpy((char *)XFS_BUF_PTR(bp), ifp->if_u1.if_data,
3644                         ifp->if_bytes);
3645                 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3646                 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3647                 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3648                 xfs_iext_add(ifp, 0, 1);
3649                 ep = xfs_iext_get_ext(ifp, 0);
3650                 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3651                 XFS_BMAP_TRACE_POST_UPDATE("new", ip, 0, whichfork);
3652                 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3653                 ip->i_d.di_nblocks = 1;
3654                 XFS_TRANS_MOD_DQUOT_BYINO(args.mp, tp, ip,
3655                         XFS_TRANS_DQ_BCOUNT, 1L);
3656                 flags |= XFS_ILOG_FEXT(whichfork);
3657         } else {
3658                 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3659                 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3660         }
3661         ifp->if_flags &= ~XFS_IFINLINE;
3662         ifp->if_flags |= XFS_IFEXTENTS;
3663         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3664         flags |= XFS_ILOG_CORE;
3665 done:
3666         *logflagsp = flags;
3667         return error;
3668 }
3669
3670 /*
3671  * Search the extent records for the entry containing block bno.
3672  * If bno lies in a hole, point to the next entry.  If bno lies
3673  * past eof, *eofp will be set, and *prevp will contain the last
3674  * entry (null if none).  Else, *lastxp will be set to the index
3675  * of the found entry; *gotp will contain the entry.
3676  */
3677 xfs_bmbt_rec_host_t *                   /* pointer to found extent entry */
3678 xfs_bmap_search_multi_extents(
3679         xfs_ifork_t     *ifp,           /* inode fork pointer */
3680         xfs_fileoff_t   bno,            /* block number searched for */
3681         int             *eofp,          /* out: end of file found */
3682         xfs_extnum_t    *lastxp,        /* out: last extent index */
3683         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3684         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3685 {
3686         xfs_bmbt_rec_host_t *ep;                /* extent record pointer */
3687         xfs_extnum_t    lastx;          /* last extent index */
3688
3689         /*
3690          * Initialize the extent entry structure to catch access to
3691          * uninitialized br_startblock field.
3692          */
3693         gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
3694         gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3695         gotp->br_state = XFS_EXT_INVALID;
3696 #if XFS_BIG_BLKNOS
3697         gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
3698 #else
3699         gotp->br_startblock = 0xffffa5a5;
3700 #endif
3701         prevp->br_startoff = NULLFILEOFF;
3702
3703         ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
3704         if (lastx > 0) {
3705                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
3706         }
3707         if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
3708                 xfs_bmbt_get_all(ep, gotp);
3709                 *eofp = 0;
3710         } else {
3711                 if (lastx > 0) {
3712                         *gotp = *prevp;
3713                 }
3714                 *eofp = 1;
3715                 ep = NULL;
3716         }
3717         *lastxp = lastx;
3718         return ep;
3719 }
3720
3721 /*
3722  * Search the extents list for the inode, for the extent containing bno.
3723  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
3724  * *eofp will be set, and *prevp will contain the last entry (null if none).
3725  * Else, *lastxp will be set to the index of the found
3726  * entry; *gotp will contain the entry.
3727  */
3728 STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
3729 xfs_bmap_search_extents(
3730         xfs_inode_t     *ip,            /* incore inode pointer */
3731         xfs_fileoff_t   bno,            /* block number searched for */
3732         int             fork,           /* data or attr fork */
3733         int             *eofp,          /* out: end of file found */
3734         xfs_extnum_t    *lastxp,        /* out: last extent index */
3735         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3736         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3737 {
3738         xfs_ifork_t     *ifp;           /* inode fork pointer */
3739         xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
3740
3741         XFS_STATS_INC(xs_look_exlist);
3742         ifp = XFS_IFORK_PTR(ip, fork);
3743
3744         ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
3745
3746         if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
3747                      !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
3748                 xfs_cmn_err(XFS_PTAG_FSBLOCK_ZERO, CE_ALERT, ip->i_mount,
3749                                 "Access to block zero in inode %llu "
3750                                 "start_block: %llx start_off: %llx "
3751                                 "blkcnt: %llx extent-state: %x lastx: %x\n",
3752                         (unsigned long long)ip->i_ino,
3753                         (unsigned long long)gotp->br_startblock,
3754                         (unsigned long long)gotp->br_startoff,
3755                         (unsigned long long)gotp->br_blockcount,
3756                         gotp->br_state, *lastxp);
3757                 *lastxp = NULLEXTNUM;
3758                 *eofp = 1;
3759                 return NULL;
3760         }
3761         return ep;
3762 }
3763
3764
3765 #ifdef XFS_BMAP_TRACE
3766 ktrace_t        *xfs_bmap_trace_buf;
3767
3768 /*
3769  * Add a bmap trace buffer entry.  Base routine for the others.
3770  */
3771 STATIC void
3772 xfs_bmap_trace_addentry(
3773         int             opcode,         /* operation */
3774         const char      *fname,         /* function name */
3775         char            *desc,          /* operation description */
3776         xfs_inode_t     *ip,            /* incore inode pointer */
3777         xfs_extnum_t    idx,            /* index of entry(ies) */
3778         xfs_extnum_t    cnt,            /* count of entries, 1 or 2 */
3779         xfs_bmbt_rec_host_t *r1,        /* first record */
3780         xfs_bmbt_rec_host_t *r2,        /* second record or null */
3781         int             whichfork)      /* data or attr fork */
3782 {
3783         xfs_bmbt_rec_host_t tr2;
3784
3785         ASSERT(cnt == 1 || cnt == 2);
3786         ASSERT(r1 != NULL);
3787         if (cnt == 1) {
3788                 ASSERT(r2 == NULL);
3789                 r2 = &tr2;
3790                 memset(&tr2, 0, sizeof(tr2));
3791         } else
3792                 ASSERT(r2 != NULL);
3793         ktrace_enter(xfs_bmap_trace_buf,
3794                 (void *)(__psint_t)(opcode | (whichfork << 16)),
3795                 (void *)fname, (void *)desc, (void *)ip,
3796                 (void *)(__psint_t)idx,
3797                 (void *)(__psint_t)cnt,
3798                 (void *)(__psunsigned_t)(ip->i_ino >> 32),
3799                 (void *)(__psunsigned_t)(unsigned)ip->i_ino,
3800                 (void *)(__psunsigned_t)(r1->l0 >> 32),
3801                 (void *)(__psunsigned_t)(unsigned)(r1->l0),
3802                 (void *)(__psunsigned_t)(r1->l1 >> 32),
3803                 (void *)(__psunsigned_t)(unsigned)(r1->l1),
3804                 (void *)(__psunsigned_t)(r2->l0 >> 32),
3805                 (void *)(__psunsigned_t)(unsigned)(r2->l0),
3806                 (void *)(__psunsigned_t)(r2->l1 >> 32),
3807                 (void *)(__psunsigned_t)(unsigned)(r2->l1)
3808                 );
3809         ASSERT(ip->i_xtrace);
3810         ktrace_enter(ip->i_xtrace,
3811                 (void *)(__psint_t)(opcode | (whichfork << 16)),
3812                 (void *)fname, (void *)desc, (void *)ip,
3813                 (void *)(__psint_t)idx,
3814                 (void *)(__psint_t)cnt,
3815                 (void *)(__psunsigned_t)(ip->i_ino >> 32),
3816                 (void *)(__psunsigned_t)(unsigned)ip->i_ino,
3817                 (void *)(__psunsigned_t)(r1->l0 >> 32),
3818                 (void *)(__psunsigned_t)(unsigned)(r1->l0),
3819                 (void *)(__psunsigned_t)(r1->l1 >> 32),
3820                 (void *)(__psunsigned_t)(unsigned)(r1->l1),
3821                 (void *)(__psunsigned_t)(r2->l0 >> 32),
3822                 (void *)(__psunsigned_t)(unsigned)(r2->l0),
3823                 (void *)(__psunsigned_t)(r2->l1 >> 32),
3824                 (void *)(__psunsigned_t)(unsigned)(r2->l1)
3825                 );
3826 }
3827
3828 /*
3829  * Add bmap trace entry prior to a call to xfs_iext_remove.
3830  */
3831 STATIC void
3832 xfs_bmap_trace_delete(
3833         const char      *fname,         /* function name */
3834         char            *desc,          /* operation description */
3835         xfs_inode_t     *ip,            /* incore inode pointer */
3836         xfs_extnum_t    idx,            /* index of entry(entries) deleted */
3837         xfs_extnum_t    cnt,            /* count of entries deleted, 1 or 2 */
3838         int             whichfork)      /* data or attr fork */
3839 {
3840         xfs_ifork_t     *ifp;           /* inode fork pointer */
3841
3842         ifp = XFS_IFORK_PTR(ip, whichfork);
3843         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_DELETE, fname, desc, ip, idx,
3844                 cnt, xfs_iext_get_ext(ifp, idx),
3845                 cnt == 2 ? xfs_iext_get_ext(ifp, idx + 1) : NULL,
3846                 whichfork);
3847 }
3848
3849 /*
3850  * Add bmap trace entry prior to a call to xfs_iext_insert, or
3851  * reading in the extents list from the disk (in the btree).
3852  */
3853 STATIC void
3854 xfs_bmap_trace_insert(
3855         const char      *fname,         /* function name */
3856         char            *desc,          /* operation description */
3857         xfs_inode_t     *ip,            /* incore inode pointer */
3858         xfs_extnum_t    idx,            /* index of entry(entries) inserted */
3859         xfs_extnum_t    cnt,            /* count of entries inserted, 1 or 2 */
3860         xfs_bmbt_irec_t *r1,            /* inserted record 1 */
3861         xfs_bmbt_irec_t *r2,            /* inserted record 2 or null */
3862         int             whichfork)      /* data or attr fork */
3863 {
3864         xfs_bmbt_rec_host_t tr1;        /* compressed record 1 */
3865         xfs_bmbt_rec_host_t tr2;        /* compressed record 2 if needed */
3866
3867         xfs_bmbt_set_all(&tr1, r1);
3868         if (cnt == 2) {
3869                 ASSERT(r2 != NULL);
3870                 xfs_bmbt_set_all(&tr2, r2);
3871         } else {
3872                 ASSERT(cnt == 1);
3873                 ASSERT(r2 == NULL);
3874         }
3875         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_INSERT, fname, desc, ip, idx,
3876                 cnt, &tr1, cnt == 2 ? &tr2 : NULL, whichfork);
3877 }
3878
3879 /*
3880  * Add bmap trace entry after updating an extent record in place.
3881  */
3882 STATIC void
3883 xfs_bmap_trace_post_update(
3884         const char      *fname,         /* function name */
3885         char            *desc,          /* operation description */
3886         xfs_inode_t     *ip,            /* incore inode pointer */
3887         xfs_extnum_t    idx,            /* index of entry updated */
3888         int             whichfork)      /* data or attr fork */
3889 {
3890         xfs_ifork_t     *ifp;           /* inode fork pointer */
3891
3892         ifp = XFS_IFORK_PTR(ip, whichfork);
3893         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_POST_UP, fname, desc, ip, idx,
3894                 1, xfs_iext_get_ext(ifp, idx), NULL, whichfork);
3895 }
3896
3897 /*
3898  * Add bmap trace entry prior to updating an extent record in place.
3899  */
3900 STATIC void
3901 xfs_bmap_trace_pre_update(
3902         const char      *fname,         /* function name */
3903         char            *desc,          /* operation description */
3904         xfs_inode_t     *ip,            /* incore inode pointer */
3905         xfs_extnum_t    idx,            /* index of entry to be updated */
3906         int             whichfork)      /* data or attr fork */
3907 {
3908         xfs_ifork_t     *ifp;           /* inode fork pointer */
3909
3910         ifp = XFS_IFORK_PTR(ip, whichfork);
3911         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_PRE_UP, fname, desc, ip, idx, 1,
3912                 xfs_iext_get_ext(ifp, idx), NULL, whichfork);
3913 }
3914 #endif  /* XFS_BMAP_TRACE */
3915
3916 /*
3917  * Compute the worst-case number of indirect blocks that will be used
3918  * for ip's delayed extent of length "len".
3919  */
3920 STATIC xfs_filblks_t
3921 xfs_bmap_worst_indlen(
3922         xfs_inode_t     *ip,            /* incore inode pointer */
3923         xfs_filblks_t   len)            /* delayed extent length */
3924 {
3925         int             level;          /* btree level number */
3926         int             maxrecs;        /* maximum record count at this level */
3927         xfs_mount_t     *mp;            /* mount structure */
3928         xfs_filblks_t   rval;           /* return value */
3929
3930         mp = ip->i_mount;
3931         maxrecs = mp->m_bmap_dmxr[0];
3932         for (level = 0, rval = 0;
3933              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3934              level++) {
3935                 len += maxrecs - 1;
3936                 do_div(len, maxrecs);
3937                 rval += len;
3938                 if (len == 1)
3939                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3940                                 level - 1;
3941                 if (level == 0)
3942                         maxrecs = mp->m_bmap_dmxr[1];
3943         }
3944         return rval;
3945 }
3946
3947 #if defined(XFS_RW_TRACE)
3948 STATIC void
3949 xfs_bunmap_trace(
3950         xfs_inode_t             *ip,
3951         xfs_fileoff_t           bno,
3952         xfs_filblks_t           len,
3953         int                     flags,
3954         inst_t                  *ra)
3955 {
3956         if (ip->i_rwtrace == NULL)
3957                 return;
3958         ktrace_enter(ip->i_rwtrace,
3959                 (void *)(__psint_t)XFS_BUNMAP,
3960                 (void *)ip,
3961                 (void *)(__psint_t)((ip->i_d.di_size >> 32) & 0xffffffff),
3962                 (void *)(__psint_t)(ip->i_d.di_size & 0xffffffff),
3963                 (void *)(__psint_t)(((xfs_dfiloff_t)bno >> 32) & 0xffffffff),
3964                 (void *)(__psint_t)((xfs_dfiloff_t)bno & 0xffffffff),
3965                 (void *)(__psint_t)len,
3966                 (void *)(__psint_t)flags,
3967                 (void *)(unsigned long)current_cpu(),
3968                 (void *)ra,
3969                 (void *)0,
3970                 (void *)0,
3971                 (void *)0,
3972                 (void *)0,
3973                 (void *)0,
3974                 (void *)0);
3975 }
3976 #endif
3977
3978 /*
3979  * Convert inode from non-attributed to attributed.
3980  * Must not be in a transaction, ip must not be locked.
3981  */
3982 int                                             /* error code */
3983 xfs_bmap_add_attrfork(
3984         xfs_inode_t             *ip,            /* incore inode pointer */
3985         int                     size,           /* space new attribute needs */
3986         int                     rsvd)           /* xact may use reserved blks */
3987 {
3988         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
3989         xfs_bmap_free_t         flist;          /* freed extent records */
3990         xfs_mount_t             *mp;            /* mount structure */
3991         xfs_trans_t             *tp;            /* transaction pointer */
3992         int                     blks;           /* space reservation */
3993         int                     version = 1;    /* superblock attr version */
3994         int                     committed;      /* xaction was committed */
3995         int                     logflags;       /* logging flags */
3996         int                     error;          /* error return value */
3997
3998         ASSERT(XFS_IFORK_Q(ip) == 0);
3999         ASSERT(ip->i_df.if_ext_max ==
4000                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
4001
4002         mp = ip->i_mount;
4003         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
4004         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
4005         blks = XFS_ADDAFORK_SPACE_RES(mp);
4006         if (rsvd)
4007                 tp->t_flags |= XFS_TRANS_RESERVE;
4008         if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
4009                         XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
4010                 goto error0;
4011         xfs_ilock(ip, XFS_ILOCK_EXCL);
4012         error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip, blks, 0, rsvd ?
4013                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
4014                         XFS_QMOPT_RES_REGBLKS);
4015         if (error) {
4016                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4017                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
4018                 return error;
4019         }
4020         if (XFS_IFORK_Q(ip))
4021                 goto error1;
4022         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
4023                 /*
4024                  * For inodes coming from pre-6.2 filesystems.
4025                  */
4026                 ASSERT(ip->i_d.di_aformat == 0);
4027                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
4028         }
4029         ASSERT(ip->i_d.di_anextents == 0);
4030         IHOLD(ip);
4031         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4032         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4033         switch (ip->i_d.di_format) {
4034         case XFS_DINODE_FMT_DEV:
4035                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
4036                 break;
4037         case XFS_DINODE_FMT_UUID:
4038                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
4039                 break;
4040         case XFS_DINODE_FMT_LOCAL:
4041         case XFS_DINODE_FMT_EXTENTS:
4042         case XFS_DINODE_FMT_BTREE:
4043                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
4044                 if (!ip->i_d.di_forkoff)
4045                         ip->i_d.di_forkoff = mp->m_attroffset >> 3;
4046                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
4047                         version = 2;
4048                 break;
4049         default:
4050                 ASSERT(0);
4051                 error = XFS_ERROR(EINVAL);
4052                 goto error1;
4053         }
4054         ip->i_df.if_ext_max =
4055                 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
4056         ASSERT(ip->i_afp == NULL);
4057         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
4058         ip->i_afp->if_ext_max =
4059                 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
4060         ip->i_afp->if_flags = XFS_IFEXTENTS;
4061         logflags = 0;
4062         XFS_BMAP_INIT(&flist, &firstblock);
4063         switch (ip->i_d.di_format) {
4064         case XFS_DINODE_FMT_LOCAL:
4065                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
4066                         &logflags);
4067                 break;
4068         case XFS_DINODE_FMT_EXTENTS:
4069                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
4070                         &flist, &logflags);
4071                 break;
4072         case XFS_DINODE_FMT_BTREE:
4073                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
4074                         &logflags);
4075                 break;
4076         default:
4077                 error = 0;
4078                 break;
4079         }
4080         if (logflags)
4081                 xfs_trans_log_inode(tp, ip, logflags);
4082         if (error)
4083                 goto error2;
4084         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
4085            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
4086                 __int64_t sbfields = 0;
4087
4088                 spin_lock(&mp->m_sb_lock);
4089                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
4090                         xfs_sb_version_addattr(&mp->m_sb);
4091                         sbfields |= XFS_SB_VERSIONNUM;
4092                 }
4093                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
4094                         xfs_sb_version_addattr2(&mp->m_sb);
4095                         sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
4096                 }
4097                 if (sbfields) {
4098                         spin_unlock(&mp->m_sb_lock);
4099                         xfs_mod_sb(tp, sbfields);
4100                 } else
4101                         spin_unlock(&mp->m_sb_lock);
4102         }
4103         if ((error = xfs_bmap_finish(&tp, &flist, &committed)))
4104                 goto error2;
4105         error = xfs_trans_commit(tp, XFS_TRANS_PERM_LOG_RES);
4106         ASSERT(ip->i_df.if_ext_max ==
4107                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
4108         return error;
4109 error2:
4110         xfs_bmap_cancel(&flist);
4111 error1:
4112         xfs_iunlock(ip, XFS_ILOCK_EXCL);
4113 error0:
4114         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
4115         ASSERT(ip->i_df.if_ext_max ==
4116                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
4117         return error;
4118 }
4119
4120 /*
4121  * Add the extent to the list of extents to be free at transaction end.
4122  * The list is maintained sorted (by block number).
4123  */
4124 /* ARGSUSED */
4125 void
4126 xfs_bmap_add_free(
4127         xfs_fsblock_t           bno,            /* fs block number of extent */
4128         xfs_filblks_t           len,            /* length of extent */
4129         xfs_bmap_free_t         *flist,         /* list of extents */
4130         xfs_mount_t             *mp)            /* mount point structure */
4131 {
4132         xfs_bmap_free_item_t    *cur;           /* current (next) element */
4133         xfs_bmap_free_item_t    *new;           /* new element */
4134         xfs_bmap_free_item_t    *prev;          /* previous element */
4135 #ifdef DEBUG
4136         xfs_agnumber_t          agno;
4137         xfs_agblock_t           agbno;
4138
4139         ASSERT(bno != NULLFSBLOCK);
4140         ASSERT(len > 0);
4141         ASSERT(len <= MAXEXTLEN);
4142         ASSERT(!ISNULLSTARTBLOCK(bno));
4143         agno = XFS_FSB_TO_AGNO(mp, bno);
4144         agbno = XFS_FSB_TO_AGBNO(mp, bno);
4145         ASSERT(agno < mp->m_sb.sb_agcount);
4146         ASSERT(agbno < mp->m_sb.sb_agblocks);
4147         ASSERT(len < mp->m_sb.sb_agblocks);
4148         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
4149 #endif
4150         ASSERT(xfs_bmap_free_item_zone != NULL);
4151         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
4152         new->xbfi_startblock = bno;
4153         new->xbfi_blockcount = (xfs_extlen_t)len;
4154         for (prev = NULL, cur = flist->xbf_first;
4155              cur != NULL;
4156              prev = cur, cur = cur->xbfi_next) {
4157                 if (cur->xbfi_startblock >= bno)
4158                         break;
4159         }
4160         if (prev)
4161                 prev->xbfi_next = new;
4162         else
4163                 flist->xbf_first = new;
4164         new->xbfi_next = cur;
4165         flist->xbf_count++;
4166 }
4167
4168 /*
4169  * Compute and fill in the value of the maximum depth of a bmap btree
4170  * in this filesystem.  Done once, during mount.
4171  */
4172 void
4173 xfs_bmap_compute_maxlevels(
4174         xfs_mount_t     *mp,            /* file system mount structure */
4175         int             whichfork)      /* data or attr fork */
4176 {
4177         int             level;          /* btree level */
4178         uint            maxblocks;      /* max blocks at this level */
4179         uint            maxleafents;    /* max leaf entries possible */
4180         int             maxrootrecs;    /* max records in root block */
4181         int             minleafrecs;    /* min records in leaf block */
4182         int             minnoderecs;    /* min records in node block */
4183         int             sz;             /* root block size */
4184
4185         /*
4186          * The maximum number of extents in a file, hence the maximum
4187          * number of leaf entries, is controlled by the type of di_nextents
4188          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
4189          * (a signed 16-bit number, xfs_aextnum_t).
4190          *
4191          * Note that we can no longer assume that if we are in ATTR1 that
4192          * the fork offset of all the inodes will be (m_attroffset >> 3)
4193          * because we could have mounted with ATTR2 and then mounted back
4194          * with ATTR1, keeping the di_forkoff's fixed but probably at
4195          * various positions. Therefore, for both ATTR1 and ATTR2
4196          * we have to assume the worst case scenario of a minimum size
4197          * available.
4198          */
4199         if (whichfork == XFS_DATA_FORK) {
4200                 maxleafents = MAXEXTNUM;
4201                 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
4202         } else {
4203                 maxleafents = MAXAEXTNUM;
4204                 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
4205         }
4206         maxrootrecs = (int)XFS_BTREE_BLOCK_MAXRECS(sz, xfs_bmdr, 0);
4207         minleafrecs = mp->m_bmap_dmnr[0];
4208         minnoderecs = mp->m_bmap_dmnr[1];
4209         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
4210         for (level = 1; maxblocks > 1; level++) {
4211                 if (maxblocks <= maxrootrecs)
4212                         maxblocks = 1;
4213                 else
4214                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
4215         }
4216         mp->m_bm_maxlevels[whichfork] = level;
4217 }
4218
4219 /*
4220  * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
4221  * caller.  Frees all the extents that need freeing, which must be done
4222  * last due to locking considerations.  We never free any extents in
4223  * the first transaction.  This is to allow the caller to make the first
4224  * transaction a synchronous one so that the pointers to the data being
4225  * broken in this transaction will be permanent before the data is actually
4226  * freed.  This is necessary to prevent blocks from being reallocated
4227  * and written to before the free and reallocation are actually permanent.
4228  * We do not just make the first transaction synchronous here, because
4229  * there are more efficient ways to gain the same protection in some cases
4230  * (see the file truncation code).
4231  *
4232  * Return 1 if the given transaction was committed and a new one
4233  * started, and 0 otherwise in the committed parameter.
4234  */
4235 /*ARGSUSED*/
4236 int                                             /* error */
4237 xfs_bmap_finish(
4238         xfs_trans_t             **tp,           /* transaction pointer addr */
4239         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
4240         int                     *committed)     /* xact committed or not */
4241 {
4242         xfs_efd_log_item_t      *efd;           /* extent free data */
4243         xfs_efi_log_item_t      *efi;           /* extent free intention */
4244         int                     error;          /* error return value */
4245         xfs_bmap_free_item_t    *free;          /* free extent item */
4246         unsigned int            logres;         /* new log reservation */
4247         unsigned int            logcount;       /* new log count */
4248         xfs_mount_t             *mp;            /* filesystem mount structure */
4249         xfs_bmap_free_item_t    *next;          /* next item on free list */
4250         xfs_trans_t             *ntp;           /* new transaction pointer */
4251
4252         ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
4253         if (flist->xbf_count == 0) {
4254                 *committed = 0;
4255                 return 0;
4256         }
4257         ntp = *tp;
4258         efi = xfs_trans_get_efi(ntp, flist->xbf_count);
4259         for (free = flist->xbf_first; free; free = free->xbfi_next)
4260                 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
4261                         free->xbfi_blockcount);
4262         logres = ntp->t_log_res;
4263         logcount = ntp->t_log_count;
4264         ntp = xfs_trans_dup(*tp);
4265         error = xfs_trans_commit(*tp, 0);
4266         *tp = ntp;
4267         *committed = 1;
4268         /*
4269          * We have a new transaction, so we should return committed=1,
4270          * even though we're returning an error.
4271          */
4272         if (error) {
4273                 return error;
4274         }
4275         if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
4276                         logcount)))
4277                 return error;
4278         efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
4279         for (free = flist->xbf_first; free != NULL; free = next) {
4280                 next = free->xbfi_next;
4281                 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
4282                                 free->xbfi_blockcount))) {
4283                         /*
4284                          * The bmap free list will be cleaned up at a
4285                          * higher level.  The EFI will be canceled when
4286                          * this transaction is aborted.
4287                          * Need to force shutdown here to make sure it
4288                          * happens, since this transaction may not be
4289                          * dirty yet.
4290                          */
4291                         mp = ntp->t_mountp;
4292                         if (!XFS_FORCED_SHUTDOWN(mp))
4293                                 xfs_force_shutdown(mp,
4294                                                    (error == EFSCORRUPTED) ?
4295                                                    SHUTDOWN_CORRUPT_INCORE :
4296                                                    SHUTDOWN_META_IO_ERROR);
4297                         return error;
4298                 }
4299                 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
4300                         free->xbfi_blockcount);
4301                 xfs_bmap_del_free(flist, NULL, free);
4302         }
4303         return 0;
4304 }
4305
4306 /*
4307  * Free up any items left in the list.
4308  */
4309 void
4310 xfs_bmap_cancel(
4311         xfs_bmap_free_t         *flist) /* list of bmap_free_items */
4312 {
4313         xfs_bmap_free_item_t    *free;  /* free list item */
4314         xfs_bmap_free_item_t    *next;
4315
4316         if (flist->xbf_count == 0)
4317                 return;
4318         ASSERT(flist->xbf_first != NULL);
4319         for (free = flist->xbf_first; free; free = next) {
4320                 next = free->xbfi_next;
4321                 xfs_bmap_del_free(flist, NULL, free);
4322         }
4323         ASSERT(flist->xbf_count == 0);
4324 }
4325
4326 /*
4327  * Returns the file-relative block number of the first unused block(s)
4328  * in the file with at least "len" logically contiguous blocks free.
4329  * This is the lowest-address hole if the file has holes, else the first block
4330  * past the end of file.
4331  * Return 0 if the file is currently local (in-inode).
4332  */
4333 int                                             /* error */
4334 xfs_bmap_first_unused(
4335         xfs_trans_t     *tp,                    /* transaction pointer */
4336         xfs_inode_t     *ip,                    /* incore inode */
4337         xfs_extlen_t    len,                    /* size of hole to find */
4338         xfs_fileoff_t   *first_unused,          /* unused block */
4339         int             whichfork)              /* data or attr fork */
4340 {
4341         int             error;                  /* error return value */
4342         int             idx;                    /* extent record index */
4343         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4344         xfs_fileoff_t   lastaddr;               /* last block number seen */
4345         xfs_fileoff_t   lowest;                 /* lowest useful block */
4346         xfs_fileoff_t   max;                    /* starting useful block */
4347         xfs_fileoff_t   off;                    /* offset for this block */
4348         xfs_extnum_t    nextents;               /* number of extent entries */
4349
4350         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
4351                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
4352                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
4353         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4354                 *first_unused = 0;
4355                 return 0;
4356         }
4357         ifp = XFS_IFORK_PTR(ip, whichfork);
4358         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4359             (error = xfs_iread_extents(tp, ip, whichfork)))
4360                 return error;
4361         lowest = *first_unused;
4362         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4363         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
4364                 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
4365                 off = xfs_bmbt_get_startoff(ep);
4366                 /*
4367                  * See if the hole before this extent will work.
4368                  */
4369                 if (off >= lowest + len && off - max >= len) {
4370                         *first_unused = max;
4371                         return 0;
4372                 }
4373                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
4374                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
4375         }
4376         *first_unused = max;
4377         return 0;
4378 }
4379
4380 /*
4381  * Returns the file-relative block number of the last block + 1 before
4382  * last_block (input value) in the file.
4383  * This is not based on i_size, it is based on the extent records.
4384  * Returns 0 for local files, as they do not have extent records.
4385  */
4386 int                                             /* error */
4387 xfs_bmap_last_before(
4388         xfs_trans_t     *tp,                    /* transaction pointer */
4389         xfs_inode_t     *ip,                    /* incore inode */
4390         xfs_fileoff_t   *last_block,            /* last block */
4391         int             whichfork)              /* data or attr fork */
4392 {
4393         xfs_fileoff_t   bno;                    /* input file offset */
4394         int             eof;                    /* hit end of file */
4395         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
4396         int             error;                  /* error return value */
4397         xfs_bmbt_irec_t got;                    /* current extent value */
4398         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4399         xfs_extnum_t    lastx;                  /* last extent used */
4400         xfs_bmbt_irec_t prev;                   /* previous extent value */
4401
4402         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4403             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4404             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4405                return XFS_ERROR(EIO);
4406         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4407                 *last_block = 0;
4408                 return 0;
4409         }
4410         ifp = XFS_IFORK_PTR(ip, whichfork);
4411         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4412             (error = xfs_iread_extents(tp, ip, whichfork)))
4413                 return error;
4414         bno = *last_block - 1;
4415         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4416                 &prev);
4417         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
4418                 if (prev.br_startoff == NULLFILEOFF)
4419                         *last_block = 0;
4420                 else
4421                         *last_block = prev.br_startoff + prev.br_blockcount;
4422         }
4423         /*
4424          * Otherwise *last_block is already the right answer.
4425          */
4426         return 0;
4427 }
4428
4429 /*
4430  * Returns the file-relative block number of the first block past eof in
4431  * the file.  This is not based on i_size, it is based on the extent records.
4432  * Returns 0 for local files, as they do not have extent records.
4433  */
4434 int                                             /* error */
4435 xfs_bmap_last_offset(
4436         xfs_trans_t     *tp,                    /* transaction pointer */
4437         xfs_inode_t     *ip,                    /* incore inode */
4438         xfs_fileoff_t   *last_block,            /* last block */
4439         int             whichfork)              /* data or attr fork */
4440 {
4441         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
4442         int             error;                  /* error return value */
4443         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4444         xfs_extnum_t    nextents;               /* number of extent entries */
4445
4446         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4447             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4448             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4449                return XFS_ERROR(EIO);
4450         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4451                 *last_block = 0;
4452                 return 0;
4453         }
4454         ifp = XFS_IFORK_PTR(ip, whichfork);
4455         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4456             (error = xfs_iread_extents(tp, ip, whichfork)))
4457                 return error;
4458         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4459         if (!nextents) {
4460                 *last_block = 0;
4461                 return 0;
4462         }
4463         ep = xfs_iext_get_ext(ifp, nextents - 1);
4464         *last_block = xfs_bmbt_get_startoff(ep) + xfs_bmbt_get_blockcount(ep);
4465         return 0;
4466 }
4467
4468 /*
4469  * Returns whether the selected fork of the inode has exactly one
4470  * block or not.  For the data fork we check this matches di_size,
4471  * implying the file's range is 0..bsize-1.
4472  */
4473 int                                     /* 1=>1 block, 0=>otherwise */
4474 xfs_bmap_one_block(
4475         xfs_inode_t     *ip,            /* incore inode */
4476         int             whichfork)      /* data or attr fork */
4477 {
4478         xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
4479         xfs_ifork_t     *ifp;           /* inode fork pointer */
4480         int             rval;           /* return value */
4481         xfs_bmbt_irec_t s;              /* internal version of extent */
4482
4483 #ifndef DEBUG
4484         if (whichfork == XFS_DATA_FORK) {
4485                 return ((ip->i_d.di_mode & S_IFMT) == S_IFREG) ?
4486                         (ip->i_size == ip->i_mount->m_sb.sb_blocksize) :
4487                         (ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize);
4488         }
4489 #endif  /* !DEBUG */
4490         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4491                 return 0;
4492         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4493                 return 0;
4494         ifp = XFS_IFORK_PTR(ip, whichfork);
4495         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4496         ep = xfs_iext_get_ext(ifp, 0);
4497         xfs_bmbt_get_all(ep, &s);
4498         rval = s.br_startoff == 0 && s.br_blockcount == 1;
4499         if (rval && whichfork == XFS_DATA_FORK)
4500                 ASSERT(ip->i_size == ip->i_mount->m_sb.sb_blocksize);
4501         return rval;
4502 }
4503
4504 /*
4505  * Read in the extents to if_extents.
4506  * All inode fields are set up by caller, we just traverse the btree
4507  * and copy the records in. If the file system cannot contain unwritten
4508  * extents, the records are checked for no "state" flags.
4509  */
4510 int                                     /* error */
4511 xfs_bmap_read_extents(
4512         xfs_trans_t             *tp,    /* transaction pointer */
4513         xfs_inode_t             *ip,    /* incore inode */
4514         int                     whichfork) /* data or attr fork */
4515 {
4516         xfs_bmbt_block_t        *block; /* current btree block */
4517         xfs_fsblock_t           bno;    /* block # of "block" */
4518         xfs_buf_t               *bp;    /* buffer for "block" */
4519         int                     error;  /* error return value */
4520         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
4521         xfs_extnum_t            i, j;   /* index into the extents list */
4522         xfs_ifork_t             *ifp;   /* fork structure */
4523         int                     level;  /* btree level, for checking */
4524         xfs_mount_t             *mp;    /* file system mount structure */
4525         __be64                  *pp;    /* pointer to block address */
4526         /* REFERENCED */
4527         xfs_extnum_t            room;   /* number of entries there's room for */
4528
4529         bno = NULLFSBLOCK;
4530         mp = ip->i_mount;
4531         ifp = XFS_IFORK_PTR(ip, whichfork);
4532         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4533                                         XFS_EXTFMT_INODE(ip);
4534         block = ifp->if_broot;
4535         /*
4536          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4537          */
4538         level = be16_to_cpu(block->bb_level);
4539         ASSERT(level > 0);
4540         pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
4541         bno = be64_to_cpu(*pp);
4542         ASSERT(bno != NULLDFSBNO);
4543         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
4544         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
4545         /*
4546          * Go down the tree until leaf level is reached, following the first
4547          * pointer (leftmost) at each level.
4548          */
4549         while (level-- > 0) {
4550                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4551                                 XFS_BMAP_BTREE_REF)))
4552                         return error;
4553                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
4554                 XFS_WANT_CORRUPTED_GOTO(
4555                         XFS_BMAP_SANITY_CHECK(mp, block, level),
4556                         error0);
4557                 if (level == 0)
4558                         break;
4559                 pp = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, 1, mp->m_bmap_dmxr[1]);
4560                 bno = be64_to_cpu(*pp);
4561                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
4562                 xfs_trans_brelse(tp, bp);
4563         }
4564         /*
4565          * Here with bp and block set to the leftmost leaf node in the tree.
4566          */
4567         room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4568         i = 0;
4569         /*
4570          * Loop over all leaf nodes.  Copy information to the extent records.
4571          */
4572         for (;;) {
4573                 xfs_bmbt_rec_t  *frp;
4574                 xfs_fsblock_t   nextbno;
4575                 xfs_extnum_t    num_recs;
4576                 xfs_extnum_t    start;
4577
4578
4579                 num_recs = be16_to_cpu(block->bb_numrecs);
4580                 if (unlikely(i + num_recs > room)) {
4581                         ASSERT(i + num_recs <= room);
4582                         xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
4583                                 "corrupt dinode %Lu, (btree extents).",
4584                                 (unsigned long long) ip->i_ino);
4585                         XFS_ERROR_REPORT("xfs_bmap_read_extents(1)",
4586                                          XFS_ERRLEVEL_LOW,
4587                                         ip->i_mount);
4588                         goto error0;
4589                 }
4590                 XFS_WANT_CORRUPTED_GOTO(
4591                         XFS_BMAP_SANITY_CHECK(mp, block, 0),
4592                         error0);
4593                 /*
4594                  * Read-ahead the next leaf block, if any.
4595                  */
4596                 nextbno = be64_to_cpu(block->bb_rightsib);
4597                 if (nextbno != NULLFSBLOCK)
4598                         xfs_btree_reada_bufl(mp, nextbno, 1);
4599                 /*
4600                  * Copy records into the extent records.
4601                  */
4602                 frp = XFS_BTREE_REC_ADDR(xfs_bmbt, block, 1);
4603                 start = i;
4604                 for (j = 0; j < num_recs; j++, i++, frp++) {
4605                         xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
4606                         trp->l0 = be64_to_cpu(frp->l0);
4607                         trp->l1 = be64_to_cpu(frp->l1);
4608                 }
4609                 if (exntf == XFS_EXTFMT_NOSTATE) {
4610                         /*
4611                          * Check all attribute bmap btree records and
4612                          * any "older" data bmap btree records for a
4613                          * set bit in the "extent flag" position.
4614                          */
4615                         if (unlikely(xfs_check_nostate_extents(ifp,
4616                                         start, num_recs))) {
4617                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4618                                                  XFS_ERRLEVEL_LOW,
4619                                                  ip->i_mount);
4620                                 goto error0;
4621                         }
4622                 }
4623                 xfs_trans_brelse(tp, bp);
4624                 bno = nextbno;
4625                 /*
4626                  * If we've reached the end, stop.
4627                  */
4628                 if (bno == NULLFSBLOCK)
4629                         break;
4630                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4631                                 XFS_BMAP_BTREE_REF)))
4632                         return error;
4633                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
4634         }
4635         ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4636         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4637         XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
4638         return 0;
4639 error0:
4640         xfs_trans_brelse(tp, bp);
4641         return XFS_ERROR(EFSCORRUPTED);
4642 }
4643
4644 #ifdef XFS_BMAP_TRACE
4645 /*
4646  * Add bmap trace insert entries for all the contents of the extent records.
4647  */
4648 void
4649 xfs_bmap_trace_exlist(
4650         const char      *fname,         /* function name */
4651         xfs_inode_t     *ip,            /* incore inode pointer */
4652         xfs_extnum_t    cnt,            /* count of entries in the list */
4653         int             whichfork)      /* data or attr fork */
4654 {
4655         xfs_bmbt_rec_host_t *ep;        /* current extent record */
4656         xfs_extnum_t    idx;            /* extent record index */
4657         xfs_ifork_t     *ifp;           /* inode fork pointer */
4658         xfs_bmbt_irec_t s;              /* file extent record */
4659
4660         ifp = XFS_IFORK_PTR(ip, whichfork);
4661         ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4662         for (idx = 0; idx < cnt; idx++) {
4663                 ep = xfs_iext_get_ext(ifp, idx);
4664                 xfs_bmbt_get_all(ep, &s);
4665                 XFS_BMAP_TRACE_INSERT("exlist", ip, idx, 1, &s, NULL,
4666                         whichfork);
4667         }
4668 }
4669 #endif
4670
4671 #ifdef DEBUG
4672 /*
4673  * Validate that the bmbt_irecs being returned from bmapi are valid
4674  * given the callers original parameters.  Specifically check the
4675  * ranges of the returned irecs to ensure that they only extent beyond
4676  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4677  */
4678 STATIC void
4679 xfs_bmap_validate_ret(
4680         xfs_fileoff_t           bno,
4681         xfs_filblks_t           len,
4682         int                     flags,
4683         xfs_bmbt_irec_t         *mval,
4684         int                     nmap,
4685         int                     ret_nmap)
4686 {
4687         int                     i;              /* index to map values */
4688
4689         ASSERT(ret_nmap <= nmap);
4690
4691         for (i = 0; i < ret_nmap; i++) {
4692                 ASSERT(mval[i].br_blockcount > 0);
4693                 if (!(flags & XFS_BMAPI_ENTIRE)) {
4694                         ASSERT(mval[i].br_startoff >= bno);
4695                         ASSERT(mval[i].br_blockcount <= len);
4696                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4697                                bno + len);
4698                 } else {
4699                         ASSERT(mval[i].br_startoff < bno + len);
4700                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4701                                bno);
4702                 }
4703                 ASSERT(i == 0 ||
4704                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4705                        mval[i].br_startoff);
4706                 if ((flags & XFS_BMAPI_WRITE) && !(flags & XFS_BMAPI_DELAY))
4707                         ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4708                                mval[i].br_startblock != HOLESTARTBLOCK);
4709                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4710                        mval[i].br_state == XFS_EXT_UNWRITTEN);
4711         }
4712 }
4713 #endif /* DEBUG */
4714
4715
4716 /*
4717  * Map file blocks to filesystem blocks.
4718  * File range is given by the bno/len pair.
4719  * Adds blocks to file if a write ("flags & XFS_BMAPI_WRITE" set)
4720  * into a hole or past eof.
4721  * Only allocates blocks from a single allocation group,
4722  * to avoid locking problems.
4723  * The returned value in "firstblock" from the first call in a transaction
4724  * must be remembered and presented to subsequent calls in "firstblock".
4725  * An upper bound for the number of blocks to be allocated is supplied to
4726  * the first call in "total"; if no allocation group has that many free
4727  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4728  */
4729 int                                     /* error */
4730 xfs_bmapi(
4731         xfs_trans_t     *tp,            /* transaction pointer */
4732         xfs_inode_t     *ip,            /* incore inode */
4733         xfs_fileoff_t   bno,            /* starting file offs. mapped */
4734         xfs_filblks_t   len,            /* length to map in file */
4735         int             flags,          /* XFS_BMAPI_... */
4736         xfs_fsblock_t   *firstblock,    /* first allocated block
4737                                            controls a.g. for allocs */
4738         xfs_extlen_t    total,          /* total blocks needed */
4739         xfs_bmbt_irec_t *mval,          /* output: map values */
4740         int             *nmap,          /* i/o: mval size/count */
4741         xfs_bmap_free_t *flist,         /* i/o: list extents to free */
4742         xfs_extdelta_t  *delta)         /* o: change made to incore extents */
4743 {
4744         xfs_fsblock_t   abno;           /* allocated block number */
4745         xfs_extlen_t    alen;           /* allocated extent length */
4746         xfs_fileoff_t   aoff;           /* allocated file offset */
4747         xfs_bmalloca_t  bma;            /* args for xfs_bmap_alloc */
4748         xfs_btree_cur_t *cur;           /* bmap btree cursor */
4749         xfs_fileoff_t   end;            /* end of mapped file region */
4750         int             eof;            /* we've hit the end of extents */
4751         xfs_bmbt_rec_host_t *ep;        /* extent record pointer */
4752         int             error;          /* error return */
4753         xfs_bmbt_irec_t got;            /* current file extent record */
4754         xfs_ifork_t     *ifp;           /* inode fork pointer */
4755         xfs_extlen_t    indlen;         /* indirect blocks length */
4756         xfs_extnum_t    lastx;          /* last useful extent number */
4757         int             logflags;       /* flags for transaction logging */
4758         xfs_extlen_t    minleft;        /* min blocks left after allocation */
4759         xfs_extlen_t    minlen;         /* min allocation size */
4760         xfs_mount_t     *mp;            /* xfs mount structure */
4761         int             n;              /* current extent index */
4762         int             nallocs;        /* number of extents alloc\'d */
4763         xfs_extnum_t    nextents;       /* number of extents in file */
4764         xfs_fileoff_t   obno;           /* old block number (offset) */
4765         xfs_bmbt_irec_t prev;           /* previous file extent record */
4766         int             tmp_logflags;   /* temp flags holder */
4767         int             whichfork;      /* data or attr fork */
4768         char            inhole;         /* current location is hole in file */
4769         char            wasdelay;       /* old extent was delayed */
4770         char            wr;             /* this is a write request */
4771         char            rt;             /* this is a realtime file */
4772 #ifdef DEBUG
4773         xfs_fileoff_t   orig_bno;       /* original block number value */
4774         int             orig_flags;     /* original flags arg value */
4775         xfs_filblks_t   orig_len;       /* original value of len arg */
4776         xfs_bmbt_irec_t *orig_mval;     /* original value of mval */
4777         int             orig_nmap;      /* original value of *nmap */
4778
4779         orig_bno = bno;
4780         orig_len = len;
4781         orig_flags = flags;
4782         orig_mval = mval;
4783         orig_nmap = *nmap;
4784 #endif
4785         ASSERT(*nmap >= 1);
4786         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP || !(flags & XFS_BMAPI_WRITE));
4787         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4788                 XFS_ATTR_FORK : XFS_DATA_FORK;
4789         mp = ip->i_mount;
4790         if (unlikely(XFS_TEST_ERROR(
4791             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4792              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4793              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4794              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4795                 XFS_ERROR_REPORT("xfs_bmapi", XFS_ERRLEVEL_LOW, mp);
4796                 return XFS_ERROR(EFSCORRUPTED);
4797         }
4798         if (XFS_FORCED_SHUTDOWN(mp))
4799                 return XFS_ERROR(EIO);
4800         rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4801         ifp = XFS_IFORK_PTR(ip, whichfork);
4802         ASSERT(ifp->if_ext_max ==
4803                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4804         if ((wr = (flags & XFS_BMAPI_WRITE)) != 0)
4805                 XFS_STATS_INC(xs_blk_mapw);
4806         else
4807                 XFS_STATS_INC(xs_blk_mapr);
4808         /*
4809          * IGSTATE flag is used to combine extents which
4810          * differ only due to the state of the extents.
4811          * This technique is used from xfs_getbmap()
4812          * when the caller does not wish to see the
4813          * separation (which is the default).
4814          *
4815          * This technique is also used when writing a
4816          * buffer which has been partially written,
4817          * (usually by being flushed during a chunkread),
4818          * to ensure one write takes place. This also
4819          * prevents a change in the xfs inode extents at
4820          * this time, intentionally. This change occurs
4821          * on completion of the write operation, in
4822          * xfs_strat_comp(), where the xfs_bmapi() call
4823          * is transactioned, and the extents combined.
4824          */
4825         if ((flags & XFS_BMAPI_IGSTATE) && wr)  /* if writing unwritten space */
4826                 wr = 0;                         /* no allocations are allowed */
4827         ASSERT(wr || !(flags & XFS_BMAPI_DELAY));
4828         logflags = 0;
4829         nallocs = 0;
4830         cur = NULL;
4831         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4832                 ASSERT(wr && tp);
4833                 if ((error = xfs_bmap_local_to_extents(tp, ip,
4834                                 firstblock, total, &logflags, whichfork)))
4835                         goto error0;
4836         }
4837         if (wr && *firstblock == NULLFSBLOCK) {
4838                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4839                         minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4840                 else
4841                         minleft = 1;
4842         } else
4843                 minleft = 0;
4844         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4845             (error = xfs_iread_extents(tp, ip, whichfork)))
4846                 goto error0;
4847         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4848                 &prev);
4849         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4850         n = 0;
4851         end = bno + len;
4852         obno = bno;
4853         bma.ip = NULL;
4854         if (delta) {
4855                 delta->xed_startoff = NULLFILEOFF;
4856                 delta->xed_blockcount = 0;
4857         }
4858         while (bno < end && n < *nmap) {
4859                 /*
4860                  * Reading past eof, act as though there's a hole
4861                  * up to end.
4862                  */
4863                 if (eof && !wr)
4864                         got.br_startoff = end;
4865                 inhole = eof || got.br_startoff > bno;
4866                 wasdelay = wr && !inhole && !(flags & XFS_BMAPI_DELAY) &&
4867                         ISNULLSTARTBLOCK(got.br_startblock);
4868                 /*
4869                  * First, deal with the hole before the allocated space
4870                  * that we found, if any.
4871                  */
4872                 if (wr && (inhole || wasdelay)) {
4873                         /*
4874                          * For the wasdelay case, we could also just
4875                          * allocate the stuff asked for in this bmap call
4876                          * but that wouldn't be as good.
4877                          */
4878                         if (wasdelay && !(flags & XFS_BMAPI_EXACT)) {
4879                                 alen = (xfs_extlen_t)got.br_blockcount;
4880                                 aoff = got.br_startoff;
4881                                 if (lastx != NULLEXTNUM && lastx) {
4882                                         ep = xfs_iext_get_ext(ifp, lastx - 1);
4883                                         xfs_bmbt_get_all(ep, &prev);
4884                                 }
4885                         } else if (wasdelay) {
4886                                 alen = (xfs_extlen_t)
4887                                         XFS_FILBLKS_MIN(len,
4888                                                 (got.br_startoff +
4889                                                  got.br_blockcount) - bno);
4890                                 aoff = bno;
4891                         } else {
4892                                 alen = (xfs_extlen_t)
4893                                         XFS_FILBLKS_MIN(len, MAXEXTLEN);
4894                                 if (!eof)
4895                                         alen = (xfs_extlen_t)
4896                                                 XFS_FILBLKS_MIN(alen,
4897                                                         got.br_startoff - bno);
4898                                 aoff = bno;
4899                         }
4900                         minlen = (flags & XFS_BMAPI_CONTIG) ? alen : 1;
4901                         if (flags & XFS_BMAPI_DELAY) {
4902                                 xfs_extlen_t    extsz;
4903
4904                                 /* Figure out the extent size, adjust alen */
4905                                 extsz = xfs_get_extsz_hint(ip);
4906                                 if (extsz) {
4907                                         error = xfs_bmap_extsize_align(mp,
4908                                                         &got, &prev, extsz,
4909                                                         rt, eof,
4910                                                         flags&XFS_BMAPI_DELAY,
4911                                                         flags&XFS_BMAPI_CONVERT,
4912                                                         &aoff, &alen);
4913                                         ASSERT(!error);
4914                                 }
4915
4916                                 if (rt)
4917                                         extsz = alen / mp->m_sb.sb_rextsize;
4918
4919                                 /*
4920                                  * Make a transaction-less quota reservation for
4921                                  * delayed allocation blocks. This number gets
4922                                  * adjusted later.  We return if we haven't
4923                                  * allocated blocks already inside this loop.
4924                                  */
4925                                 if ((error = XFS_TRANS_RESERVE_QUOTA_NBLKS(
4926                                                 mp, NULL, ip, (long)alen, 0,
4927                                                 rt ? XFS_QMOPT_RES_RTBLKS :
4928                                                      XFS_QMOPT_RES_REGBLKS))) {
4929                                         if (n == 0) {
4930                                                 *nmap = 0;
4931                                                 ASSERT(cur == NULL);
4932                                                 return error;
4933                                         }
4934                                         break;
4935                                 }
4936
4937                                 /*
4938                                  * Split changing sb for alen and indlen since
4939                                  * they could be coming from different places.
4940                                  */
4941                                 indlen = (xfs_extlen_t)
4942                                         xfs_bmap_worst_indlen(ip, alen);
4943                                 ASSERT(indlen > 0);
4944
4945                                 if (rt) {
4946                                         error = xfs_mod_incore_sb(mp,
4947                                                         XFS_SBS_FREXTENTS,
4948                                                         -((int64_t)extsz), (flags &
4949                                                         XFS_BMAPI_RSVBLOCKS));
4950                                 } else {
4951                                         error = xfs_mod_incore_sb(mp,
4952                                                         XFS_SBS_FDBLOCKS,
4953                                                         -((int64_t)alen), (flags &
4954                                                         XFS_BMAPI_RSVBLOCKS));
4955                                 }
4956                                 if (!error) {
4957                                         error = xfs_mod_incore_sb(mp,
4958                                                         XFS_SBS_FDBLOCKS,
4959                                                         -((int64_t)indlen), (flags &
4960                                                         XFS_BMAPI_RSVBLOCKS));
4961                                         if (error && rt)
4962                                                 xfs_mod_incore_sb(mp,
4963                                                         XFS_SBS_FREXTENTS,
4964                                                         (int64_t)extsz, (flags &
4965                                                         XFS_BMAPI_RSVBLOCKS));
4966                                         else if (error)
4967                                                 xfs_mod_incore_sb(mp,
4968                                                         XFS_SBS_FDBLOCKS,
4969                                                         (int64_t)alen, (flags &
4970                                                         XFS_BMAPI_RSVBLOCKS));
4971                                 }
4972
4973                                 if (error) {
4974                                         if (XFS_IS_QUOTA_ON(mp))
4975                                                 /* unreserve the blocks now */
4976                                                 (void)
4977                                                 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(
4978                                                         mp, NULL, ip,
4979                                                         (long)alen, 0, rt ?
4980                                                         XFS_QMOPT_RES_RTBLKS :
4981                                                         XFS_QMOPT_RES_REGBLKS);
4982                                         break;
4983                                 }
4984
4985                                 ip->i_delayed_blks += alen;
4986                                 abno = NULLSTARTBLOCK(indlen);
4987                         } else {
4988                                 /*
4989                                  * If first time, allocate and fill in
4990                                  * once-only bma fields.
4991                                  */
4992                                 if (bma.ip == NULL) {
4993                                         bma.tp = tp;
4994                                         bma.ip = ip;
4995                                         bma.prevp = &prev;
4996                                         bma.gotp = &got;
4997                                         bma.total = total;
4998                                         bma.userdata = 0;
4999                                 }
5000                                 /* Indicate if this is the first user data
5001                                  * in the file, or just any user data.
5002                                  */
5003                                 if (!(flags & XFS_BMAPI_METADATA)) {
5004                                         bma.userdata = (aoff == 0) ?
5005                                                 XFS_ALLOC_INITIAL_USER_DATA :
5006                                                 XFS_ALLOC_USERDATA;
5007                                 }
5008                                 /*
5009                                  * Fill in changeable bma fields.
5010                                  */
5011                                 bma.eof = eof;
5012                                 bma.firstblock = *firstblock;
5013                                 bma.alen = alen;
5014                                 bma.off = aoff;
5015                                 bma.conv = !!(flags & XFS_BMAPI_CONVERT);
5016                                 bma.wasdel = wasdelay;
5017                                 bma.minlen = minlen;
5018                                 bma.low = flist->xbf_low;
5019                                 bma.minleft = minleft;
5020                                 /*
5021                                  * Only want to do the alignment at the
5022                                  * eof if it is userdata and allocation length
5023                                  * is larger than a stripe unit.
5024                                  */
5025                                 if (mp->m_dalign && alen >= mp->m_dalign &&
5026                                     (!(flags & XFS_BMAPI_METADATA)) &&
5027                                     (whichfork == XFS_DATA_FORK)) {
5028                                         if ((error = xfs_bmap_isaeof(ip, aoff,
5029                                                         whichfork, &bma.aeof)))
5030                                                 goto error0;
5031                                 } else
5032                                         bma.aeof = 0;
5033                                 /*
5034                                  * Call allocator.
5035                                  */
5036                                 if ((error = xfs_bmap_alloc(&bma)))
5037                                         goto error0;
5038                                 /*
5039                                  * Copy out result fields.
5040                                  */
5041                                 abno = bma.rval;
5042                                 if ((flist->xbf_low = bma.low))
5043                                         minleft = 0;
5044                                 alen = bma.alen;
5045                                 aoff = bma.off;
5046                                 ASSERT(*firstblock == NULLFSBLOCK ||
5047                                        XFS_FSB_TO_AGNO(mp, *firstblock) ==
5048                                        XFS_FSB_TO_AGNO(mp, bma.firstblock) ||
5049                                        (flist->xbf_low &&
5050                                         XFS_FSB_TO_AGNO(mp, *firstblock) <
5051                                         XFS_FSB_TO_AGNO(mp, bma.firstblock)));
5052                                 *firstblock = bma.firstblock;
5053                                 if (cur)
5054                                         cur->bc_private.b.firstblock =
5055                                                 *firstblock;
5056                                 if (abno == NULLFSBLOCK)
5057                                         break;
5058                                 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
5059                                         cur = xfs_bmbt_init_cursor(mp, tp,
5060                                                 ip, whichfork);
5061                                         cur->bc_private.b.firstblock =
5062                                                 *firstblock;
5063                                         cur->bc_private.b.flist = flist;
5064                                 }
5065                                 /*
5066                                  * Bump the number of extents we've allocated
5067                                  * in this call.
5068                                  */
5069                                 nallocs++;
5070                         }
5071                         if (cur)
5072                                 cur->bc_private.b.flags =
5073                                         wasdelay ? XFS_BTCUR_BPRV_WASDEL : 0;
5074                         got.br_startoff = aoff;
5075                         got.br_startblock = abno;
5076                         got.br_blockcount = alen;
5077                         got.br_state = XFS_EXT_NORM;    /* assume normal */
5078                         /*
5079                          * Determine state of extent, and the filesystem.
5080                          * A wasdelay extent has been initialized, so
5081                          * shouldn't be flagged as unwritten.
5082                          */
5083                         if (wr && xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5084                                 if (!wasdelay && (flags & XFS_BMAPI_PREALLOC))
5085                                         got.br_state = XFS_EXT_UNWRITTEN;
5086                         }
5087                         error = xfs_bmap_add_extent(ip, lastx, &cur, &got,
5088                                 firstblock, flist, &tmp_logflags, delta,
5089                                 whichfork, (flags & XFS_BMAPI_RSVBLOCKS));
5090                         logflags |= tmp_logflags;
5091                         if (error)
5092                                 goto error0;
5093                         lastx = ifp->if_lastex;
5094                         ep = xfs_iext_get_ext(ifp, lastx);
5095                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5096                         xfs_bmbt_get_all(ep, &got);
5097                         ASSERT(got.br_startoff <= aoff);
5098                         ASSERT(got.br_startoff + got.br_blockcount >=
5099                                 aoff + alen);
5100 #ifdef DEBUG
5101                         if (flags & XFS_BMAPI_DELAY) {
5102                                 ASSERT(ISNULLSTARTBLOCK(got.br_startblock));
5103                                 ASSERT(STARTBLOCKVAL(got.br_startblock) > 0);
5104                         }
5105                         ASSERT(got.br_state == XFS_EXT_NORM ||
5106                                got.br_state == XFS_EXT_UNWRITTEN);
5107 #endif
5108                         /*
5109                          * Fall down into the found allocated space case.
5110                          */
5111                 } else if (inhole) {
5112                         /*
5113                          * Reading in a hole.
5114                          */
5115                         mval->br_startoff = bno;
5116                         mval->br_startblock = HOLESTARTBLOCK;
5117                         mval->br_blockcount =
5118                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
5119                         mval->br_state = XFS_EXT_NORM;
5120                         bno += mval->br_blockcount;
5121                         len -= mval->br_blockcount;
5122                         mval++;
5123                         n++;
5124                         continue;
5125                 }
5126                 /*
5127                  * Then deal with the allocated space we found.
5128                  */
5129                 ASSERT(ep != NULL);
5130                 if (!(flags & XFS_BMAPI_ENTIRE) &&
5131                     (got.br_startoff + got.br_blockcount > obno)) {
5132                         if (obno > bno)
5133                                 bno = obno;
5134                         ASSERT((bno >= obno) || (n == 0));
5135                         ASSERT(bno < end);
5136                         mval->br_startoff = bno;
5137                         if (ISNULLSTARTBLOCK(got.br_startblock)) {
5138                                 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
5139                                 mval->br_startblock = DELAYSTARTBLOCK;
5140                         } else
5141                                 mval->br_startblock =
5142                                         got.br_startblock +
5143                                         (bno - got.br_startoff);
5144                         /*
5145                          * Return the minimum of what we got and what we
5146                          * asked for for the length.  We can use the len
5147                          * variable here because it is modified below
5148                          * and we could have been there before coming
5149                          * here if the first part of the allocation
5150                          * didn't overlap what was asked for.
5151                          */
5152                         mval->br_blockcount =
5153                                 XFS_FILBLKS_MIN(end - bno, got.br_blockcount -
5154                                         (bno - got.br_startoff));
5155                         mval->br_state = got.br_state;
5156                         ASSERT(mval->br_blockcount <= len);
5157                 } else {
5158                         *mval = got;
5159                         if (ISNULLSTARTBLOCK(mval->br_startblock)) {
5160                                 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
5161                                 mval->br_startblock = DELAYSTARTBLOCK;
5162                         }
5163                 }
5164
5165                 /*
5166                  * Check if writing previously allocated but
5167                  * unwritten extents.
5168                  */
5169                 if (wr && mval->br_state == XFS_EXT_UNWRITTEN &&
5170                     ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_DELAY)) == 0)) {
5171                         /*
5172                          * Modify (by adding) the state flag, if writing.
5173                          */
5174                         ASSERT(mval->br_blockcount <= len);
5175                         if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
5176                                 cur = xfs_bmbt_init_cursor(mp,
5177                                         tp, ip, whichfork);
5178                                 cur->bc_private.b.firstblock =
5179                                         *firstblock;
5180                                 cur->bc_private.b.flist = flist;
5181                         }
5182                         mval->br_state = XFS_EXT_NORM;
5183                         error = xfs_bmap_add_extent(ip, lastx, &cur, mval,
5184                                 firstblock, flist, &tmp_logflags, delta,
5185                                 whichfork, (flags & XFS_BMAPI_RSVBLOCKS));
5186                         logflags |= tmp_logflags;
5187                         if (error)
5188                                 goto error0;
5189                         lastx = ifp->if_lastex;
5190                         ep = xfs_iext_get_ext(ifp, lastx);
5191                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5192                         xfs_bmbt_get_all(ep, &got);
5193                         /*
5194                          * We may have combined previously unwritten
5195                          * space with written space, so generate
5196                          * another request.
5197                          */
5198                         if (mval->br_blockcount < len)
5199                                 continue;
5200                 }
5201
5202                 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
5203                        ((mval->br_startoff + mval->br_blockcount) <= end));
5204                 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
5205                        (mval->br_blockcount <= len) ||
5206                        (mval->br_startoff < obno));
5207                 bno = mval->br_startoff + mval->br_blockcount;
5208                 len = end - bno;
5209                 if (n > 0 && mval->br_startoff == mval[-1].br_startoff) {
5210                         ASSERT(mval->br_startblock == mval[-1].br_startblock);
5211                         ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
5212                         ASSERT(mval->br_state == mval[-1].br_state);
5213                         mval[-1].br_blockcount = mval->br_blockcount;
5214                         mval[-1].br_state = mval->br_state;
5215                 } else if (n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
5216                            mval[-1].br_startblock != DELAYSTARTBLOCK &&
5217                            mval[-1].br_startblock != HOLESTARTBLOCK &&
5218                            mval->br_startblock ==
5219                            mval[-1].br_startblock + mval[-1].br_blockcount &&
5220                            ((flags & XFS_BMAPI_IGSTATE) ||
5221                                 mval[-1].br_state == mval->br_state)) {
5222                         ASSERT(mval->br_startoff ==
5223                                mval[-1].br_startoff + mval[-1].br_blockcount);
5224                         mval[-1].br_blockcount += mval->br_blockcount;
5225                 } else if (n > 0 &&
5226                            mval->br_startblock == DELAYSTARTBLOCK &&
5227                            mval[-1].br_startblock == DELAYSTARTBLOCK &&
5228                            mval->br_startoff ==
5229                            mval[-1].br_startoff + mval[-1].br_blockcount) {
5230                         mval[-1].br_blockcount += mval->br_blockcount;
5231                         mval[-1].br_state = mval->br_state;
5232                 } else if (!((n == 0) &&
5233                              ((mval->br_startoff + mval->br_blockcount) <=
5234                               obno))) {
5235                         mval++;
5236                         n++;
5237                 }
5238                 /*
5239                  * If we're done, stop now.  Stop when we've allocated
5240                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
5241                  * the transaction may get too big.
5242                  */
5243                 if (bno >= end || n >= *nmap || nallocs >= *nmap)
5244                         break;
5245                 /*
5246                  * Else go on to the next record.
5247                  */
5248                 ep = xfs_iext_get_ext(ifp, ++lastx);
5249                 prev = got;
5250                 if (lastx >= nextents)
5251                         eof = 1;
5252                 else
5253                         xfs_bmbt_get_all(ep, &got);
5254         }
5255         ifp->if_lastex = lastx;
5256         *nmap = n;
5257         /*
5258          * Transform from btree to extents, give it cur.
5259          */
5260         if (tp && XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5261             XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5262                 ASSERT(wr && cur);
5263                 error = xfs_bmap_btree_to_extents(tp, ip, cur,
5264                         &tmp_logflags, whichfork);
5265                 logflags |= tmp_logflags;
5266                 if (error)
5267                         goto error0;
5268         }
5269         ASSERT(ifp->if_ext_max ==
5270                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5271         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
5272                XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max);
5273         error = 0;
5274         if (delta && delta->xed_startoff != NULLFILEOFF) {
5275                 /* A change was actually made.
5276                  * Note that delta->xed_blockount is an offset at this
5277                  * point and needs to be converted to a block count.
5278                  */
5279                 ASSERT(delta->xed_blockcount > delta->xed_startoff);
5280                 delta->xed_blockcount -= delta->xed_startoff;
5281         }
5282 error0:
5283         /*
5284          * Log everything.  Do this after conversion, there's no point in
5285          * logging the extent records if we've converted to btree format.
5286          */
5287         if ((logflags & XFS_ILOG_FEXT(whichfork)) &&
5288             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5289                 logflags &= ~XFS_ILOG_FEXT(whichfork);
5290         else if ((logflags & XFS_ILOG_FBROOT(whichfork)) &&
5291                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5292                 logflags &= ~XFS_ILOG_FBROOT(whichfork);
5293         /*
5294          * Log whatever the flags say, even if error.  Otherwise we might miss
5295          * detecting a case where the data is changed, there's an error,
5296          * and it's not logged so we don't shutdown when we should.
5297          */
5298         if (logflags) {
5299                 ASSERT(tp && wr);
5300                 xfs_trans_log_inode(tp, ip, logflags);
5301         }
5302         if (cur) {
5303                 if (!error) {
5304                         ASSERT(*firstblock == NULLFSBLOCK ||
5305                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
5306                                XFS_FSB_TO_AGNO(mp,
5307                                        cur->bc_private.b.firstblock) ||
5308                                (flist->xbf_low &&
5309                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
5310                                 XFS_FSB_TO_AGNO(mp,
5311                                         cur->bc_private.b.firstblock)));
5312                         *firstblock = cur->bc_private.b.firstblock;
5313                 }
5314                 xfs_btree_del_cursor(cur,
5315                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5316         }
5317         if (!error)
5318                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
5319                         orig_nmap, *nmap);
5320         return error;
5321 }
5322
5323 /*
5324  * Map file blocks to filesystem blocks, simple version.
5325  * One block (extent) only, read-only.
5326  * For flags, only the XFS_BMAPI_ATTRFORK flag is examined.
5327  * For the other flag values, the effect is as if XFS_BMAPI_METADATA
5328  * was set and all the others were clear.
5329  */
5330 int                                             /* error */
5331 xfs_bmapi_single(
5332         xfs_trans_t     *tp,            /* transaction pointer */
5333         xfs_inode_t     *ip,            /* incore inode */
5334         int             whichfork,      /* data or attr fork */
5335         xfs_fsblock_t   *fsb,           /* output: mapped block */
5336         xfs_fileoff_t   bno)            /* starting file offs. mapped */
5337 {
5338         int             eof;            /* we've hit the end of extents */
5339         int             error;          /* error return */
5340         xfs_bmbt_irec_t got;            /* current file extent record */
5341         xfs_ifork_t     *ifp;           /* inode fork pointer */
5342         xfs_extnum_t    lastx;          /* last useful extent number */
5343         xfs_bmbt_irec_t prev;           /* previous file extent record */
5344
5345         ifp = XFS_IFORK_PTR(ip, whichfork);
5346         if (unlikely(
5347             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
5348             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)) {
5349                XFS_ERROR_REPORT("xfs_bmapi_single", XFS_ERRLEVEL_LOW,
5350                                 ip->i_mount);
5351                return XFS_ERROR(EFSCORRUPTED);
5352         }
5353         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
5354                 return XFS_ERROR(EIO);
5355         XFS_STATS_INC(xs_blk_mapr);
5356         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5357             (error = xfs_iread_extents(tp, ip, whichfork)))
5358                 return error;
5359         (void)xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5360                 &prev);
5361         /*
5362          * Reading past eof, act as though there's a hole
5363          * up to end.
5364          */
5365         if (eof || got.br_startoff > bno) {
5366                 *fsb = NULLFSBLOCK;
5367                 return 0;
5368         }
5369         ASSERT(!ISNULLSTARTBLOCK(got.br_startblock));
5370         ASSERT(bno < got.br_startoff + got.br_blockcount);
5371         *fsb = got.br_startblock + (bno - got.br_startoff);
5372         ifp->if_lastex = lastx;
5373         return 0;
5374 }
5375
5376 /*
5377  * Unmap (remove) blocks from a file.
5378  * If nexts is nonzero then the number of extents to remove is limited to
5379  * that value.  If not all extents in the block range can be removed then
5380  * *done is set.
5381  */
5382 int                                             /* error */
5383 xfs_bunmapi(
5384         xfs_trans_t             *tp,            /* transaction pointer */
5385         struct xfs_inode        *ip,            /* incore inode */
5386         xfs_fileoff_t           bno,            /* starting offset to unmap */
5387         xfs_filblks_t           len,            /* length to unmap in file */
5388         int                     flags,          /* misc flags */
5389         xfs_extnum_t            nexts,          /* number of extents max */
5390         xfs_fsblock_t           *firstblock,    /* first allocated block
5391                                                    controls a.g. for allocs */
5392         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
5393         xfs_extdelta_t          *delta,         /* o: change made to incore
5394                                                    extents */
5395         int                     *done)          /* set if not done yet */
5396 {
5397         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
5398         xfs_bmbt_irec_t         del;            /* extent being deleted */
5399         int                     eof;            /* is deleting at eof */
5400         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
5401         int                     error;          /* error return value */
5402         xfs_extnum_t            extno;          /* extent number in list */
5403         xfs_bmbt_irec_t         got;            /* current extent record */
5404         xfs_ifork_t             *ifp;           /* inode fork pointer */
5405         int                     isrt;           /* freeing in rt area */
5406         xfs_extnum_t            lastx;          /* last extent index used */
5407         int                     logflags;       /* transaction logging flags */
5408         xfs_extlen_t            mod;            /* rt extent offset */
5409         xfs_mount_t             *mp;            /* mount structure */
5410         xfs_extnum_t            nextents;       /* number of file extents */
5411         xfs_bmbt_irec_t         prev;           /* previous extent record */
5412         xfs_fileoff_t           start;          /* first file offset deleted */
5413         int                     tmp_logflags;   /* partial logging flags */
5414         int                     wasdel;         /* was a delayed alloc extent */
5415         int                     whichfork;      /* data or attribute fork */
5416         int                     rsvd;           /* OK to allocate reserved blocks */
5417         xfs_fsblock_t           sum;
5418
5419         xfs_bunmap_trace(ip, bno, len, flags, (inst_t *)__return_address);
5420         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5421                 XFS_ATTR_FORK : XFS_DATA_FORK;
5422         ifp = XFS_IFORK_PTR(ip, whichfork);
5423         if (unlikely(
5424             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5425             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5426                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5427                                  ip->i_mount);
5428                 return XFS_ERROR(EFSCORRUPTED);
5429         }
5430         mp = ip->i_mount;
5431         if (XFS_FORCED_SHUTDOWN(mp))
5432                 return XFS_ERROR(EIO);
5433         rsvd = (flags & XFS_BMAPI_RSVBLOCKS) != 0;
5434         ASSERT(len > 0);
5435         ASSERT(nexts >= 0);
5436         ASSERT(ifp->if_ext_max ==
5437                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5438         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5439             (error = xfs_iread_extents(tp, ip, whichfork)))
5440                 return error;
5441         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5442         if (nextents == 0) {
5443                 *done = 1;
5444                 return 0;
5445         }
5446         XFS_STATS_INC(xs_blk_unmap);
5447         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5448         start = bno;
5449         bno = start + len - 1;
5450         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5451                 &prev);
5452         if (delta) {
5453                 delta->xed_startoff = NULLFILEOFF;
5454                 delta->xed_blockcount = 0;
5455         }
5456         /*
5457          * Check to see if the given block number is past the end of the
5458          * file, back up to the last block if so...
5459          */
5460         if (eof) {
5461                 ep = xfs_iext_get_ext(ifp, --lastx);
5462                 xfs_bmbt_get_all(ep, &got);
5463                 bno = got.br_startoff + got.br_blockcount - 1;
5464         }
5465         logflags = 0;
5466         if (ifp->if_flags & XFS_IFBROOT) {
5467                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5468                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5469                 cur->bc_private.b.firstblock = *firstblock;
5470                 cur->bc_private.b.flist = flist;
5471                 cur->bc_private.b.flags = 0;
5472         } else
5473                 cur = NULL;
5474         extno = 0;
5475         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5476                (nexts == 0 || extno < nexts)) {
5477                 /*
5478                  * Is the found extent after a hole in which bno lives?
5479                  * Just back up to the previous extent, if so.
5480                  */
5481                 if (got.br_startoff > bno) {
5482                         if (--lastx < 0)
5483                                 break;
5484                         ep = xfs_iext_get_ext(ifp, lastx);
5485                         xfs_bmbt_get_all(ep, &got);
5486                 }
5487                 /*
5488                  * Is the last block of this extent before the range
5489                  * we're supposed to delete?  If so, we're done.
5490                  */
5491                 bno = XFS_FILEOFF_MIN(bno,
5492                         got.br_startoff + got.br_blockcount - 1);
5493                 if (bno < start)
5494                         break;
5495                 /*
5496                  * Then deal with the (possibly delayed) allocated space
5497                  * we found.
5498                  */
5499                 ASSERT(ep != NULL);
5500                 del = got;
5501                 wasdel = ISNULLSTARTBLOCK(del.br_startblock);
5502                 if (got.br_startoff < start) {
5503                         del.br_startoff = start;
5504                         del.br_blockcount -= start - got.br_startoff;
5505                         if (!wasdel)
5506                                 del.br_startblock += start - got.br_startoff;
5507                 }
5508                 if (del.br_startoff + del.br_blockcount > bno + 1)
5509                         del.br_blockcount = bno + 1 - del.br_startoff;
5510                 sum = del.br_startblock + del.br_blockcount;
5511                 if (isrt &&
5512                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5513                         /*
5514                          * Realtime extent not lined up at the end.
5515                          * The extent could have been split into written
5516                          * and unwritten pieces, or we could just be
5517                          * unmapping part of it.  But we can't really
5518                          * get rid of part of a realtime extent.
5519                          */
5520                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5521                             !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5522                                 /*
5523                                  * This piece is unwritten, or we're not
5524                                  * using unwritten extents.  Skip over it.
5525                                  */
5526                                 ASSERT(bno >= mod);
5527                                 bno -= mod > del.br_blockcount ?
5528                                         del.br_blockcount : mod;
5529                                 if (bno < got.br_startoff) {
5530                                         if (--lastx >= 0)
5531                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5532                                                         ifp, lastx), &got);
5533                                 }
5534                                 continue;
5535                         }
5536                         /*
5537                          * It's written, turn it unwritten.
5538                          * This is better than zeroing it.
5539                          */
5540                         ASSERT(del.br_state == XFS_EXT_NORM);
5541                         ASSERT(xfs_trans_get_block_res(tp) > 0);
5542                         /*
5543                          * If this spans a realtime extent boundary,
5544                          * chop it back to the start of the one we end at.
5545                          */
5546                         if (del.br_blockcount > mod) {
5547                                 del.br_startoff += del.br_blockcount - mod;
5548                                 del.br_startblock += del.br_blockcount - mod;
5549                                 del.br_blockcount = mod;
5550                         }
5551                         del.br_state = XFS_EXT_UNWRITTEN;
5552                         error = xfs_bmap_add_extent(ip, lastx, &cur, &del,
5553                                 firstblock, flist, &logflags, delta,
5554                                 XFS_DATA_FORK, 0);
5555                         if (error)
5556                                 goto error0;
5557                         goto nodelete;
5558                 }
5559                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5560                         /*
5561                          * Realtime extent is lined up at the end but not
5562                          * at the front.  We'll get rid of full extents if
5563                          * we can.
5564                          */
5565                         mod = mp->m_sb.sb_rextsize - mod;
5566                         if (del.br_blockcount > mod) {
5567                                 del.br_blockcount -= mod;
5568                                 del.br_startoff += mod;
5569                                 del.br_startblock += mod;
5570                         } else if ((del.br_startoff == start &&
5571                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5572                                      xfs_trans_get_block_res(tp) == 0)) ||
5573                                    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5574                                 /*
5575                                  * Can't make it unwritten.  There isn't
5576                                  * a full extent here so just skip it.
5577                                  */
5578                                 ASSERT(bno >= del.br_blockcount);
5579                                 bno -= del.br_blockcount;
5580                                 if (bno < got.br_startoff) {
5581                                         if (--lastx >= 0)
5582                                                 xfs_bmbt_get_all(--ep, &got);
5583                                 }
5584                                 continue;
5585                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5586                                 /*
5587                                  * This one is already unwritten.
5588                                  * It must have a written left neighbor.
5589                                  * Unwrite the killed part of that one and
5590                                  * try again.
5591                                  */
5592                                 ASSERT(lastx > 0);
5593                                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5594                                                 lastx - 1), &prev);
5595                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5596                                 ASSERT(!ISNULLSTARTBLOCK(prev.br_startblock));
5597                                 ASSERT(del.br_startblock ==
5598                                        prev.br_startblock + prev.br_blockcount);
5599                                 if (prev.br_startoff < start) {
5600                                         mod = start - prev.br_startoff;
5601                                         prev.br_blockcount -= mod;
5602                                         prev.br_startblock += mod;
5603                                         prev.br_startoff = start;
5604                                 }
5605                                 prev.br_state = XFS_EXT_UNWRITTEN;
5606                                 error = xfs_bmap_add_extent(ip, lastx - 1, &cur,
5607                                         &prev, firstblock, flist, &logflags,
5608                                         delta, XFS_DATA_FORK, 0);
5609                                 if (error)
5610                                         goto error0;
5611                                 goto nodelete;
5612                         } else {
5613                                 ASSERT(del.br_state == XFS_EXT_NORM);
5614                                 del.br_state = XFS_EXT_UNWRITTEN;
5615                                 error = xfs_bmap_add_extent(ip, lastx, &cur,
5616                                         &del, firstblock, flist, &logflags,
5617                                         delta, XFS_DATA_FORK, 0);
5618                                 if (error)
5619                                         goto error0;
5620                                 goto nodelete;
5621                         }
5622                 }
5623                 if (wasdel) {
5624                         ASSERT(STARTBLOCKVAL(del.br_startblock) > 0);
5625                         /* Update realtime/data freespace, unreserve quota */
5626                         if (isrt) {
5627                                 xfs_filblks_t rtexts;
5628
5629                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5630                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5631                                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5632                                                 (int64_t)rtexts, rsvd);
5633                                 (void)XFS_TRANS_RESERVE_QUOTA_NBLKS(mp,
5634                                         NULL, ip, -((long)del.br_blockcount), 0,
5635                                         XFS_QMOPT_RES_RTBLKS);
5636                         } else {
5637                                 xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS,
5638                                                 (int64_t)del.br_blockcount, rsvd);
5639                                 (void)XFS_TRANS_RESERVE_QUOTA_NBLKS(mp,
5640                                         NULL, ip, -((long)del.br_blockcount), 0,
5641                                         XFS_QMOPT_RES_REGBLKS);
5642                         }
5643                         ip->i_delayed_blks -= del.br_blockcount;
5644                         if (cur)
5645                                 cur->bc_private.b.flags |=
5646                                         XFS_BTCUR_BPRV_WASDEL;
5647                 } else if (cur)
5648                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5649                 /*
5650                  * If it's the case where the directory code is running
5651                  * with no block reservation, and the deleted block is in
5652                  * the middle of its extent, and the resulting insert
5653                  * of an extent would cause transformation to btree format,
5654                  * then reject it.  The calling code will then swap
5655                  * blocks around instead.
5656                  * We have to do this now, rather than waiting for the
5657                  * conversion to btree format, since the transaction
5658                  * will be dirty.
5659                  */
5660                 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5661                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5662                     XFS_IFORK_NEXTENTS(ip, whichfork) >= ifp->if_ext_max &&
5663                     del.br_startoff > got.br_startoff &&
5664                     del.br_startoff + del.br_blockcount <
5665                     got.br_startoff + got.br_blockcount) {
5666                         error = XFS_ERROR(ENOSPC);
5667                         goto error0;
5668                 }
5669                 error = xfs_bmap_del_extent(ip, tp, lastx, flist, cur, &del,
5670                                 &tmp_logflags, delta, whichfork, rsvd);
5671                 logflags |= tmp_logflags;
5672                 if (error)
5673                         goto error0;
5674                 bno = del.br_startoff - 1;
5675 nodelete:
5676                 lastx = ifp->if_lastex;
5677                 /*
5678                  * If not done go on to the next (previous) record.
5679                  * Reset ep in case the extents array was re-alloced.
5680                  */
5681                 ep = xfs_iext_get_ext(ifp, lastx);
5682                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5683                         if (lastx >= XFS_IFORK_NEXTENTS(ip, whichfork) ||
5684                             xfs_bmbt_get_startoff(ep) > bno) {
5685                                 if (--lastx >= 0)
5686                                         ep = xfs_iext_get_ext(ifp, lastx);
5687                         }
5688                         if (lastx >= 0)
5689                                 xfs_bmbt_get_all(ep, &got);
5690                         extno++;
5691                 }
5692         }
5693         ifp->if_lastex = lastx;
5694         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5695         ASSERT(ifp->if_ext_max ==
5696                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5697         /*
5698          * Convert to a btree if necessary.
5699          */
5700         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5701             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
5702                 ASSERT(cur == NULL);
5703                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5704                         &cur, 0, &tmp_logflags, whichfork);
5705                 logflags |= tmp_logflags;
5706                 if (error)
5707                         goto error0;
5708         }
5709         /*
5710          * transform from btree to extents, give it cur
5711          */
5712         else if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5713                  XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5714                 ASSERT(cur != NULL);
5715                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5716                         whichfork);
5717                 logflags |= tmp_logflags;
5718                 if (error)
5719                         goto error0;
5720         }
5721         /*
5722          * transform from extents to local?
5723          */
5724         ASSERT(ifp->if_ext_max ==
5725                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5726         error = 0;
5727         if (delta && delta->xed_startoff != NULLFILEOFF) {
5728                 /* A change was actually made.
5729                  * Note that delta->xed_blockount is an offset at this
5730                  * point and needs to be converted to a block count.
5731                  */
5732                 ASSERT(delta->xed_blockcount > delta->xed_startoff);
5733                 delta->xed_blockcount -= delta->xed_startoff;
5734         }
5735 error0:
5736         /*
5737          * Log everything.  Do this after conversion, there's no point in
5738          * logging the extent records if we've converted to btree format.
5739          */
5740         if ((logflags & XFS_ILOG_FEXT(whichfork)) &&
5741             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5742                 logflags &= ~XFS_ILOG_FEXT(whichfork);
5743         else if ((logflags & XFS_ILOG_FBROOT(whichfork)) &&
5744                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5745                 logflags &= ~XFS_ILOG_FBROOT(whichfork);
5746         /*
5747          * Log inode even in the error case, if the transaction
5748          * is dirty we'll need to shut down the filesystem.
5749          */
5750         if (logflags)
5751                 xfs_trans_log_inode(tp, ip, logflags);
5752         if (cur) {
5753                 if (!error) {
5754                         *firstblock = cur->bc_private.b.firstblock;
5755                         cur->bc_private.b.allocated = 0;
5756                 }
5757                 xfs_btree_del_cursor(cur,
5758                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5759         }
5760         return error;
5761 }
5762
5763 /*
5764  * returns 1 for success, 0 if we failed to map the extent.
5765  */
5766 STATIC int
5767 xfs_getbmapx_fix_eof_hole(
5768         xfs_inode_t             *ip,            /* xfs incore inode pointer */
5769         struct getbmap          *out,           /* output structure */
5770         int                     prealloced,     /* this is a file with
5771                                                 * preallocated data space */
5772         __int64_t               end,            /* last block requested */
5773         xfs_fsblock_t           startblock)
5774 {
5775         __int64_t               fixlen;
5776         xfs_mount_t             *mp;            /* file system mount point */
5777
5778         if (startblock == HOLESTARTBLOCK) {
5779                 mp = ip->i_mount;
5780                 out->bmv_block = -1;
5781                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, ip->i_size));
5782                 fixlen -= out->bmv_offset;
5783                 if (prealloced && out->bmv_offset + out->bmv_length == end) {
5784                         /* Came to hole at EOF. Trim it. */
5785                         if (fixlen <= 0)
5786                                 return 0;
5787                         out->bmv_length = fixlen;
5788                 }
5789         } else {
5790                 out->bmv_block = XFS_FSB_TO_DB(ip, startblock);
5791         }
5792
5793         return 1;
5794 }
5795
5796 /*
5797  * Fcntl interface to xfs_bmapi.
5798  */
5799 int                                             /* error code */
5800 xfs_getbmap(
5801         xfs_inode_t             *ip,
5802         struct getbmap          *bmv,           /* user bmap structure */
5803         void                    __user *ap,     /* pointer to user's array */
5804         int                     interface)      /* interface flags */
5805 {
5806         __int64_t               bmvend;         /* last block requested */
5807         int                     error;          /* return value */
5808         __int64_t               fixlen;         /* length for -1 case */
5809         int                     i;              /* extent number */
5810         int                     lock;           /* lock state */
5811         xfs_bmbt_irec_t         *map;           /* buffer for user's data */
5812         xfs_mount_t             *mp;            /* file system mount point */
5813         int                     nex;            /* # of user extents can do */
5814         int                     nexleft;        /* # of user extents left */
5815         int                     subnex;         /* # of bmapi's can do */
5816         int                     nmap;           /* number of map entries */
5817         struct getbmap          out;            /* output structure */
5818         int                     whichfork;      /* data or attr fork */
5819         int                     prealloced;     /* this is a file with
5820                                                  * preallocated data space */
5821         int                     sh_unwritten;   /* true, if unwritten */
5822                                                 /* extents listed separately */
5823         int                     bmapi_flags;    /* flags for xfs_bmapi */
5824         __int32_t               oflags;         /* getbmapx bmv_oflags field */
5825
5826         mp = ip->i_mount;
5827
5828         whichfork = interface & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5829         sh_unwritten = (interface & BMV_IF_PREALLOC) != 0;
5830
5831         /*      If the BMV_IF_NO_DMAPI_READ interface bit specified, do not
5832          *      generate a DMAPI read event.  Otherwise, if the DM_EVENT_READ
5833          *      bit is set for the file, generate a read event in order
5834          *      that the DMAPI application may do its thing before we return
5835          *      the extents.  Usually this means restoring user file data to
5836          *      regions of the file that look like holes.
5837          *
5838          *      The "old behavior" (from XFS_IOC_GETBMAP) is to not specify
5839          *      BMV_IF_NO_DMAPI_READ so that read events are generated.
5840          *      If this were not true, callers of ioctl( XFS_IOC_GETBMAP )
5841          *      could misinterpret holes in a DMAPI file as true holes,
5842          *      when in fact they may represent offline user data.
5843          */
5844         if ((interface & BMV_IF_NO_DMAPI_READ) == 0 &&
5845             DM_EVENT_ENABLED(ip, DM_EVENT_READ) &&
5846             whichfork == XFS_DATA_FORK) {
5847                 error = XFS_SEND_DATA(mp, DM_EVENT_READ, ip, 0, 0, 0, NULL);
5848                 if (error)
5849                         return XFS_ERROR(error);
5850         }
5851
5852         if (whichfork == XFS_ATTR_FORK) {
5853                 if (XFS_IFORK_Q(ip)) {
5854                         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5855                             ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5856                             ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5857                                 return XFS_ERROR(EINVAL);
5858                 } else if (unlikely(
5859                            ip->i_d.di_aformat != 0 &&
5860                            ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5861                         XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5862                                          ip->i_mount);
5863                         return XFS_ERROR(EFSCORRUPTED);
5864                 }
5865         } else if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5866                    ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5867                    ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5868                 return XFS_ERROR(EINVAL);
5869         if (whichfork == XFS_DATA_FORK) {
5870                 if (xfs_get_extsz_hint(ip) ||
5871                     ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5872                         prealloced = 1;
5873                         fixlen = XFS_MAXIOFFSET(mp);
5874                 } else {
5875                         prealloced = 0;
5876                         fixlen = ip->i_size;
5877                 }
5878         } else {
5879                 prealloced = 0;
5880                 fixlen = 1LL << 32;
5881         }
5882
5883         if (bmv->bmv_length == -1) {
5884                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5885                 bmv->bmv_length = MAX( (__int64_t)(fixlen - bmv->bmv_offset),
5886                                         (__int64_t)0);
5887         } else if (bmv->bmv_length < 0)
5888                 return XFS_ERROR(EINVAL);
5889         if (bmv->bmv_length == 0) {
5890                 bmv->bmv_entries = 0;
5891                 return 0;
5892         }
5893         nex = bmv->bmv_count - 1;
5894         if (nex <= 0)
5895                 return XFS_ERROR(EINVAL);
5896         bmvend = bmv->bmv_offset + bmv->bmv_length;
5897
5898         xfs_ilock(ip, XFS_IOLOCK_SHARED);
5899
5900         if (whichfork == XFS_DATA_FORK &&
5901                 (ip->i_delayed_blks || ip->i_size > ip->i_d.di_size)) {
5902                 /* xfs_fsize_t last_byte = xfs_file_last_byte(ip); */
5903                 error = xfs_flush_pages(ip, (xfs_off_t)0,
5904                                                -1, 0, FI_REMAPF);
5905                 if (error) {
5906                         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5907                 return error;
5908                 }
5909         }
5910
5911         ASSERT(whichfork == XFS_ATTR_FORK || ip->i_delayed_blks == 0);
5912
5913         lock = xfs_ilock_map_shared(ip);
5914
5915         /*
5916          * Don't let nex be bigger than the number of extents
5917          * we can have assuming alternating holes and real extents.
5918          */
5919         if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5920                 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5921
5922         bmapi_flags = XFS_BMAPI_AFLAG(whichfork) |
5923                         ((sh_unwritten) ? 0 : XFS_BMAPI_IGSTATE);
5924
5925         /*
5926          * Allocate enough space to handle "subnex" maps at a time.
5927          */
5928         subnex = 16;
5929         map = kmem_alloc(subnex * sizeof(*map), KM_SLEEP);
5930
5931         bmv->bmv_entries = 0;
5932
5933         if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0) {
5934                 error = 0;
5935                 goto unlock_and_return;
5936         }
5937
5938         nexleft = nex;
5939
5940         do {
5941                 nmap = (nexleft > subnex) ? subnex : nexleft;
5942                 error = xfs_bmapi(NULL, ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5943                                   XFS_BB_TO_FSB(mp, bmv->bmv_length),
5944                                   bmapi_flags, NULL, 0, map, &nmap,
5945                                   NULL, NULL);
5946                 if (error)
5947                         goto unlock_and_return;
5948                 ASSERT(nmap <= subnex);
5949
5950                 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5951                         nexleft--;
5952                         oflags = (map[i].br_state == XFS_EXT_UNWRITTEN) ?
5953                                         BMV_OF_PREALLOC : 0;
5954                         out.bmv_offset = XFS_FSB_TO_BB(mp, map[i].br_startoff);
5955                         out.bmv_length = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5956                         ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
5957                         if (map[i].br_startblock == HOLESTARTBLOCK &&
5958                             whichfork == XFS_ATTR_FORK) {
5959                                 /* came to the end of attribute fork */
5960                                 goto unlock_and_return;
5961                         } else {
5962                                 if (!xfs_getbmapx_fix_eof_hole(ip, &out,
5963                                                         prealloced, bmvend,
5964                                                         map[i].br_startblock)) {
5965                                         goto unlock_and_return;
5966                                 }
5967
5968                                 /* return either getbmap/getbmapx structure. */
5969                                 if (interface & BMV_IF_EXTENDED) {
5970                                         struct  getbmapx        outx;
5971
5972                                         GETBMAP_CONVERT(out,outx);
5973                                         outx.bmv_oflags = oflags;
5974                                         outx.bmv_unused1 = outx.bmv_unused2 = 0;
5975                                         if (copy_to_user(ap, &outx,
5976                                                         sizeof(outx))) {
5977                                                 error = XFS_ERROR(EFAULT);
5978                                                 goto unlock_and_return;
5979                                         }
5980                                 } else {
5981                                         if (copy_to_user(ap, &out,
5982                                                         sizeof(out))) {
5983                                                 error = XFS_ERROR(EFAULT);
5984                                                 goto unlock_and_return;
5985                                         }
5986                                 }
5987                                 bmv->bmv_offset =
5988                                         out.bmv_offset + out.bmv_length;
5989                                 bmv->bmv_length = MAX((__int64_t)0,
5990                                         (__int64_t)(bmvend - bmv->bmv_offset));
5991                                 bmv->bmv_entries++;
5992                                 ap = (interface & BMV_IF_EXTENDED) ?
5993                                                 (void __user *)
5994                                         ((struct getbmapx __user *)ap + 1) :
5995                                                 (void __user *)
5996                                         ((struct getbmap __user *)ap + 1);
5997                         }
5998                 }
5999         } while (nmap && nexleft && bmv->bmv_length);
6000
6001 unlock_and_return:
6002         xfs_iunlock_map_shared(ip, lock);
6003         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
6004
6005         kmem_free(map);
6006
6007         return error;
6008 }
6009
6010 /*
6011  * Check the last inode extent to determine whether this allocation will result
6012  * in blocks being allocated at the end of the file. When we allocate new data
6013  * blocks at the end of the file which do not start at the previous data block,
6014  * we will try to align the new blocks at stripe unit boundaries.
6015  */
6016 STATIC int                              /* error */
6017 xfs_bmap_isaeof(
6018         xfs_inode_t     *ip,            /* incore inode pointer */
6019         xfs_fileoff_t   off,            /* file offset in fsblocks */
6020         int             whichfork,      /* data or attribute fork */
6021         char            *aeof)          /* return value */
6022 {
6023         int             error;          /* error return value */
6024         xfs_ifork_t     *ifp;           /* inode fork pointer */
6025         xfs_bmbt_rec_host_t *lastrec;   /* extent record pointer */
6026         xfs_extnum_t    nextents;       /* number of file extents */
6027         xfs_bmbt_irec_t s;              /* expanded extent record */
6028
6029         ASSERT(whichfork == XFS_DATA_FORK);
6030         ifp = XFS_IFORK_PTR(ip, whichfork);
6031         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
6032             (error = xfs_iread_extents(NULL, ip, whichfork)))
6033                 return error;
6034         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
6035         if (nextents == 0) {
6036                 *aeof = 1;
6037                 return 0;
6038         }
6039         /*
6040          * Go to the last extent
6041          */
6042         lastrec = xfs_iext_get_ext(ifp, nextents - 1);
6043         xfs_bmbt_get_all(lastrec, &s);
6044         /*
6045          * Check we are allocating in the last extent (for delayed allocations)
6046          * or past the last extent for non-delayed allocations.
6047          */
6048         *aeof = (off >= s.br_startoff &&
6049                  off < s.br_startoff + s.br_blockcount &&
6050                  ISNULLSTARTBLOCK(s.br_startblock)) ||
6051                 off >= s.br_startoff + s.br_blockcount;
6052         return 0;
6053 }
6054
6055 /*
6056  * Check if the endoff is outside the last extent. If so the caller will grow
6057  * the allocation to a stripe unit boundary.
6058  */
6059 int                                     /* error */
6060 xfs_bmap_eof(
6061         xfs_inode_t     *ip,            /* incore inode pointer */
6062         xfs_fileoff_t   endoff,         /* file offset in fsblocks */
6063         int             whichfork,      /* data or attribute fork */
6064         int             *eof)           /* result value */
6065 {
6066         xfs_fsblock_t   blockcount;     /* extent block count */
6067         int             error;          /* error return value */
6068         xfs_ifork_t     *ifp;           /* inode fork pointer */
6069         xfs_bmbt_rec_host_t *lastrec;   /* extent record pointer */
6070         xfs_extnum_t    nextents;       /* number of file extents */
6071         xfs_fileoff_t   startoff;       /* extent starting file offset */
6072
6073         ASSERT(whichfork == XFS_DATA_FORK);
6074         ifp = XFS_IFORK_PTR(ip, whichfork);
6075         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
6076             (error = xfs_iread_extents(NULL, ip, whichfork)))
6077                 return error;
6078         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
6079         if (nextents == 0) {
6080                 *eof = 1;
6081                 return 0;
6082         }
6083         /*
6084          * Go to the last extent
6085          */
6086         lastrec = xfs_iext_get_ext(ifp, nextents - 1);
6087         startoff = xfs_bmbt_get_startoff(lastrec);
6088         blockcount = xfs_bmbt_get_blockcount(lastrec);
6089         *eof = endoff >= startoff + blockcount;
6090         return 0;
6091 }
6092
6093 #ifdef DEBUG
6094 STATIC
6095 xfs_buf_t *
6096 xfs_bmap_get_bp(
6097         xfs_btree_cur_t         *cur,
6098         xfs_fsblock_t           bno)
6099 {
6100         int i;
6101         xfs_buf_t *bp;
6102
6103         if (!cur)
6104                 return(NULL);
6105
6106         bp = NULL;
6107         for(i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
6108                 bp = cur->bc_bufs[i];
6109                 if (!bp) break;
6110                 if (XFS_BUF_ADDR(bp) == bno)
6111                         break;  /* Found it */
6112         }
6113         if (i == XFS_BTREE_MAXLEVELS)
6114                 bp = NULL;
6115
6116         if (!bp) { /* Chase down all the log items to see if the bp is there */
6117                 xfs_log_item_chunk_t    *licp;
6118                 xfs_trans_t             *tp;
6119
6120                 tp = cur->bc_tp;
6121                 licp = &tp->t_items;
6122                 while (!bp && licp != NULL) {
6123                         if (xfs_lic_are_all_free(licp)) {
6124                                 licp = licp->lic_next;
6125                                 continue;
6126                         }
6127                         for (i = 0; i < licp->lic_unused; i++) {
6128                                 xfs_log_item_desc_t     *lidp;
6129                                 xfs_log_item_t          *lip;
6130                                 xfs_buf_log_item_t      *bip;
6131                                 xfs_buf_t               *lbp;
6132
6133                                 if (xfs_lic_isfree(licp, i)) {
6134                                         continue;
6135                                 }
6136
6137                                 lidp = xfs_lic_slot(licp, i);
6138                                 lip = lidp->lid_item;
6139                                 if (lip->li_type != XFS_LI_BUF)
6140                                         continue;
6141
6142                                 bip = (xfs_buf_log_item_t *)lip;
6143                                 lbp = bip->bli_buf;
6144
6145                                 if (XFS_BUF_ADDR(lbp) == bno) {
6146                                         bp = lbp;
6147                                         break; /* Found it */
6148                                 }
6149                         }
6150                         licp = licp->lic_next;
6151                 }
6152         }
6153         return(bp);
6154 }
6155
6156 void
6157 xfs_check_block(
6158         xfs_bmbt_block_t        *block,
6159         xfs_mount_t             *mp,
6160         int                     root,
6161         short                   sz)
6162 {
6163         int                     i, j, dmxr;
6164         __be64                  *pp, *thispa;   /* pointer to block address */
6165         xfs_bmbt_key_t          *prevp, *keyp;
6166
6167         ASSERT(be16_to_cpu(block->bb_level) > 0);
6168
6169         prevp = NULL;
6170         for( i = 1; i <= be16_to_cpu(block->bb_numrecs); i++) {
6171                 dmxr = mp->m_bmap_dmxr[0];
6172
6173                 if (root) {
6174                         keyp = XFS_BMAP_BROOT_KEY_ADDR(block, i, sz);
6175                 } else {
6176                         keyp = XFS_BTREE_KEY_ADDR(xfs_bmbt, block, i);
6177                 }
6178
6179                 if (prevp) {
6180                         xfs_btree_check_key(XFS_BTNUM_BMAP, prevp, keyp);
6181                 }
6182                 prevp = keyp;
6183
6184                 /*
6185                  * Compare the block numbers to see if there are dups.
6186                  */
6187
6188                 if (root) {
6189                         pp = XFS_BMAP_BROOT_PTR_ADDR(block, i, sz);
6190                 } else {
6191                         pp = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, i, dmxr);
6192                 }
6193                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
6194                         if (root) {
6195                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(block, j, sz);
6196                         } else {
6197                                 thispa = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, j,
6198                                                             dmxr);
6199                         }
6200                         if (*thispa == *pp) {
6201                                 cmn_err(CE_WARN, "%s: thispa(%d) == pp(%d) %Ld",
6202                                         __func__, j, i,
6203                                         (unsigned long long)be64_to_cpu(*thispa));
6204                                 panic("%s: ptrs are equal in node\n",
6205                                         __func__);
6206                         }
6207                 }
6208         }
6209 }
6210
6211 /*
6212  * Check that the extents for the inode ip are in the right order in all
6213  * btree leaves.
6214  */
6215
6216 STATIC void
6217 xfs_bmap_check_leaf_extents(
6218         xfs_btree_cur_t         *cur,   /* btree cursor or null */
6219         xfs_inode_t             *ip,            /* incore inode pointer */
6220         int                     whichfork)      /* data or attr fork */
6221 {
6222         xfs_bmbt_block_t        *block; /* current btree block */
6223         xfs_fsblock_t           bno;    /* block # of "block" */
6224         xfs_buf_t               *bp;    /* buffer for "block" */
6225         int                     error;  /* error return value */
6226         xfs_extnum_t            i=0, j; /* index into the extents list */
6227         xfs_ifork_t             *ifp;   /* fork structure */
6228         int                     level;  /* btree level, for checking */
6229         xfs_mount_t             *mp;    /* file system mount structure */
6230         __be64                  *pp;    /* pointer to block address */
6231         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
6232         xfs_bmbt_rec_t          last = {0, 0}; /* last extent in prev block */
6233         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
6234         int                     bp_release = 0;
6235
6236         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
6237                 return;
6238         }
6239
6240         bno = NULLFSBLOCK;
6241         mp = ip->i_mount;
6242         ifp = XFS_IFORK_PTR(ip, whichfork);
6243         block = ifp->if_broot;
6244         /*
6245          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6246          */
6247         level = be16_to_cpu(block->bb_level);
6248         ASSERT(level > 0);
6249         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
6250         pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
6251         bno = be64_to_cpu(*pp);
6252
6253         ASSERT(bno != NULLDFSBNO);
6254         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
6255         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
6256
6257         /*
6258          * Go down the tree until leaf level is reached, following the first
6259          * pointer (leftmost) at each level.
6260          */
6261         while (level-- > 0) {
6262                 /* See if buf is in cur first */
6263                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
6264                 if (bp) {
6265                         bp_release = 0;
6266                 } else {
6267                         bp_release = 1;
6268                 }
6269                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
6270                                 XFS_BMAP_BTREE_REF)))
6271                         goto error_norelse;
6272                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6273                 XFS_WANT_CORRUPTED_GOTO(
6274                         XFS_BMAP_SANITY_CHECK(mp, block, level),
6275                         error0);
6276                 if (level == 0)
6277                         break;
6278
6279                 /*
6280                  * Check this block for basic sanity (increasing keys and
6281                  * no duplicate blocks).
6282                  */
6283
6284                 xfs_check_block(block, mp, 0, 0);
6285                 pp = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, 1, mp->m_bmap_dmxr[1]);
6286                 bno = be64_to_cpu(*pp);
6287                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
6288                 if (bp_release) {
6289                         bp_release = 0;
6290                         xfs_trans_brelse(NULL, bp);
6291                 }
6292         }
6293
6294         /*
6295          * Here with bp and block set to the leftmost leaf node in the tree.
6296          */
6297         i = 0;
6298
6299         /*
6300          * Loop over all leaf nodes checking that all extents are in the right order.
6301          */
6302         for (;;) {
6303                 xfs_fsblock_t   nextbno;
6304                 xfs_extnum_t    num_recs;
6305
6306
6307                 num_recs = be16_to_cpu(block->bb_numrecs);
6308
6309                 /*
6310                  * Read-ahead the next leaf block, if any.
6311                  */
6312
6313                 nextbno = be64_to_cpu(block->bb_rightsib);
6314
6315                 /*
6316                  * Check all the extents to make sure they are OK.
6317                  * If we had a previous block, the last entry should
6318                  * conform with the first entry in this one.
6319                  */
6320
6321                 ep = XFS_BTREE_REC_ADDR(xfs_bmbt, block, 1);
6322                 if (i) {
6323                         xfs_btree_check_rec(XFS_BTNUM_BMAP, &last, ep);
6324                 }
6325                 for (j = 1; j < num_recs; j++) {
6326                         nextp = XFS_BTREE_REC_ADDR(xfs_bmbt, block, j + 1);
6327                         xfs_btree_check_rec(XFS_BTNUM_BMAP, ep, nextp);
6328                         ep = nextp;
6329                 }
6330
6331                 last = *ep;
6332                 i += num_recs;
6333                 if (bp_release) {
6334                         bp_release = 0;
6335                         xfs_trans_brelse(NULL, bp);
6336                 }
6337                 bno = nextbno;
6338                 /*
6339                  * If we've reached the end, stop.
6340                  */
6341                 if (bno == NULLFSBLOCK)
6342                         break;
6343
6344                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
6345                 if (bp) {
6346                         bp_release = 0;
6347                 } else {
6348                         bp_release = 1;
6349                 }
6350                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
6351                                 XFS_BMAP_BTREE_REF)))
6352                         goto error_norelse;
6353                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6354         }
6355         if (bp_release) {
6356                 bp_release = 0;
6357                 xfs_trans_brelse(NULL, bp);
6358         }
6359         return;
6360
6361 error0:
6362         cmn_err(CE_WARN, "%s: at error0", __func__);
6363         if (bp_release)
6364                 xfs_trans_brelse(NULL, bp);
6365 error_norelse:
6366         cmn_err(CE_WARN, "%s: BAD after btree leaves for %d extents",
6367                 __func__, i);
6368         panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
6369         return;
6370 }
6371 #endif
6372
6373 /*
6374  * Count fsblocks of the given fork.
6375  */
6376 int                                             /* error */
6377 xfs_bmap_count_blocks(
6378         xfs_trans_t             *tp,            /* transaction pointer */
6379         xfs_inode_t             *ip,            /* incore inode */
6380         int                     whichfork,      /* data or attr fork */
6381         int                     *count)         /* out: count of blocks */
6382 {
6383         xfs_bmbt_block_t        *block; /* current btree block */
6384         xfs_fsblock_t           bno;    /* block # of "block" */
6385         xfs_ifork_t             *ifp;   /* fork structure */
6386         int                     level;  /* btree level, for checking */
6387         xfs_mount_t             *mp;    /* file system mount structure */
6388         __be64                  *pp;    /* pointer to block address */
6389
6390         bno = NULLFSBLOCK;
6391         mp = ip->i_mount;
6392         ifp = XFS_IFORK_PTR(ip, whichfork);
6393         if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
6394                 xfs_bmap_count_leaves(ifp, 0,
6395                         ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
6396                         count);
6397                 return 0;
6398         }
6399
6400         /*
6401          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6402          */
6403         block = ifp->if_broot;
6404         level = be16_to_cpu(block->bb_level);
6405         ASSERT(level > 0);
6406         pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
6407         bno = be64_to_cpu(*pp);
6408         ASSERT(bno != NULLDFSBNO);
6409         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
6410         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
6411
6412         if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
6413                 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
6414                                  mp);
6415                 return XFS_ERROR(EFSCORRUPTED);
6416         }
6417
6418         return 0;
6419 }
6420
6421 /*
6422  * Recursively walks each level of a btree
6423  * to count total fsblocks is use.
6424  */
6425 STATIC int                                     /* error */
6426 xfs_bmap_count_tree(
6427         xfs_mount_t     *mp,            /* file system mount point */
6428         xfs_trans_t     *tp,            /* transaction pointer */
6429         xfs_ifork_t     *ifp,           /* inode fork pointer */
6430         xfs_fsblock_t   blockno,        /* file system block number */
6431         int             levelin,        /* level in btree */
6432         int             *count)         /* Count of blocks */
6433 {
6434         int                     error;
6435         xfs_buf_t               *bp, *nbp;
6436         int                     level = levelin;
6437         __be64                  *pp;
6438         xfs_fsblock_t           bno = blockno;
6439         xfs_fsblock_t           nextbno;
6440         xfs_bmbt_block_t        *block, *nextblock;
6441         int                     numrecs;
6442
6443         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
6444                 return error;
6445         *count += 1;
6446         block = XFS_BUF_TO_BMBT_BLOCK(bp);
6447
6448         if (--level) {
6449                 /* Not at node above leafs, count this level of nodes */
6450                 nextbno = be64_to_cpu(block->bb_rightsib);
6451                 while (nextbno != NULLFSBLOCK) {
6452                         if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
6453                                 0, &nbp, XFS_BMAP_BTREE_REF)))
6454                                 return error;
6455                         *count += 1;
6456                         nextblock = XFS_BUF_TO_BMBT_BLOCK(nbp);
6457                         nextbno = be64_to_cpu(nextblock->bb_rightsib);
6458                         xfs_trans_brelse(tp, nbp);
6459                 }
6460
6461                 /* Dive to the next level */
6462                 pp = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, 1, mp->m_bmap_dmxr[1]);
6463                 bno = be64_to_cpu(*pp);
6464                 if (unlikely((error =
6465                      xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
6466                         xfs_trans_brelse(tp, bp);
6467                         XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6468                                          XFS_ERRLEVEL_LOW, mp);
6469                         return XFS_ERROR(EFSCORRUPTED);
6470                 }
6471                 xfs_trans_brelse(tp, bp);
6472         } else {
6473                 /* count all level 1 nodes and their leaves */
6474                 for (;;) {
6475                         nextbno = be64_to_cpu(block->bb_rightsib);
6476                         numrecs = be16_to_cpu(block->bb_numrecs);
6477                         xfs_bmap_disk_count_leaves(0, block, numrecs, count);
6478                         xfs_trans_brelse(tp, bp);
6479                         if (nextbno == NULLFSBLOCK)
6480                                 break;
6481                         bno = nextbno;
6482                         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6483                                 XFS_BMAP_BTREE_REF)))
6484                                 return error;
6485                         *count += 1;
6486                         block = XFS_BUF_TO_BMBT_BLOCK(bp);
6487                 }
6488         }
6489         return 0;
6490 }
6491
6492 /*
6493  * Count leaf blocks given a range of extent records.
6494  */
6495 STATIC void
6496 xfs_bmap_count_leaves(
6497         xfs_ifork_t             *ifp,
6498         xfs_extnum_t            idx,
6499         int                     numrecs,
6500         int                     *count)
6501 {
6502         int             b;
6503
6504         for (b = 0; b < numrecs; b++) {
6505                 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
6506                 *count += xfs_bmbt_get_blockcount(frp);
6507         }
6508 }
6509
6510 /*
6511  * Count leaf blocks given a range of extent records originally
6512  * in btree format.
6513  */
6514 STATIC void
6515 xfs_bmap_disk_count_leaves(
6516         xfs_extnum_t            idx,
6517         xfs_bmbt_block_t        *block,
6518         int                     numrecs,
6519         int                     *count)
6520 {
6521         int             b;
6522         xfs_bmbt_rec_t  *frp;
6523
6524         for (b = 1; b <= numrecs; b++) {
6525                 frp = XFS_BTREE_REC_ADDR(xfs_bmbt, block, idx + b);
6526                 *count += xfs_bmbt_disk_get_blockcount(frp);
6527         }
6528 }