hvc_console: Add tty driver flag TTY_DRIVER_RESET_TERMIOS
[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 /*
406  * Called from xfs_bmap_add_attrfork to handle btree format files.
407  */
408 STATIC int                                      /* error */
409 xfs_bmap_add_attrfork_btree(
410         xfs_trans_t             *tp,            /* transaction pointer */
411         xfs_inode_t             *ip,            /* incore inode pointer */
412         xfs_fsblock_t           *firstblock,    /* first block allocated */
413         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
414         int                     *flags)         /* inode logging flags */
415 {
416         xfs_btree_cur_t         *cur;           /* btree cursor */
417         int                     error;          /* error return value */
418         xfs_mount_t             *mp;            /* file system mount struct */
419         int                     stat;           /* newroot status */
420
421         mp = ip->i_mount;
422         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
423                 *flags |= XFS_ILOG_DBROOT;
424         else {
425                 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
426                         XFS_DATA_FORK);
427                 cur->bc_private.b.flist = flist;
428                 cur->bc_private.b.firstblock = *firstblock;
429                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
430                         goto error0;
431                 /* must be at least one entry */
432                 XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
433                 if ((error = xfs_bmbt_newroot(cur, flags, &stat)))
434                         goto error0;
435                 if (stat == 0) {
436                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
437                         return XFS_ERROR(ENOSPC);
438                 }
439                 *firstblock = cur->bc_private.b.firstblock;
440                 cur->bc_private.b.allocated = 0;
441                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
442         }
443         return 0;
444 error0:
445         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
446         return error;
447 }
448
449 /*
450  * Called from xfs_bmap_add_attrfork to handle extents format files.
451  */
452 STATIC int                                      /* error */
453 xfs_bmap_add_attrfork_extents(
454         xfs_trans_t             *tp,            /* transaction pointer */
455         xfs_inode_t             *ip,            /* incore inode pointer */
456         xfs_fsblock_t           *firstblock,    /* first block allocated */
457         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
458         int                     *flags)         /* inode logging flags */
459 {
460         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
461         int                     error;          /* error return value */
462
463         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
464                 return 0;
465         cur = NULL;
466         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
467                 flags, XFS_DATA_FORK);
468         if (cur) {
469                 cur->bc_private.b.allocated = 0;
470                 xfs_btree_del_cursor(cur,
471                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
472         }
473         return error;
474 }
475
476 /*
477  * Called from xfs_bmap_add_attrfork to handle local format files.
478  */
479 STATIC int                                      /* error */
480 xfs_bmap_add_attrfork_local(
481         xfs_trans_t             *tp,            /* transaction pointer */
482         xfs_inode_t             *ip,            /* incore inode pointer */
483         xfs_fsblock_t           *firstblock,    /* first block allocated */
484         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
485         int                     *flags)         /* inode logging flags */
486 {
487         xfs_da_args_t           dargs;          /* args for dir/attr code */
488         int                     error;          /* error return value */
489         xfs_mount_t             *mp;            /* mount structure pointer */
490
491         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
492                 return 0;
493         if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
494                 mp = ip->i_mount;
495                 memset(&dargs, 0, sizeof(dargs));
496                 dargs.dp = ip;
497                 dargs.firstblock = firstblock;
498                 dargs.flist = flist;
499                 dargs.total = mp->m_dirblkfsbs;
500                 dargs.whichfork = XFS_DATA_FORK;
501                 dargs.trans = tp;
502                 error = xfs_dir2_sf_to_block(&dargs);
503         } else
504                 error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
505                         XFS_DATA_FORK);
506         return error;
507 }
508
509 /*
510  * Called by xfs_bmapi to update file extent records and the btree
511  * after allocating space (or doing a delayed allocation).
512  */
513 STATIC int                              /* error */
514 xfs_bmap_add_extent(
515         xfs_inode_t             *ip,    /* incore inode pointer */
516         xfs_extnum_t            idx,    /* extent number to update/insert */
517         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
518         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
519         xfs_fsblock_t           *first, /* pointer to firstblock variable */
520         xfs_bmap_free_t         *flist, /* list of extents to be freed */
521         int                     *logflagsp, /* inode logging flags */
522         xfs_extdelta_t          *delta, /* Change made to incore extents */
523         int                     whichfork, /* data or attr fork */
524         int                     rsvd)   /* OK to use reserved data blocks */
525 {
526         xfs_btree_cur_t         *cur;   /* btree cursor or null */
527         xfs_filblks_t           da_new; /* new count del alloc blocks used */
528         xfs_filblks_t           da_old; /* old count del alloc blocks used */
529         int                     error;  /* error return value */
530         xfs_ifork_t             *ifp;   /* inode fork ptr */
531         int                     logflags; /* returned value */
532         xfs_extnum_t            nextents; /* number of extents in file now */
533
534         XFS_STATS_INC(xs_add_exlist);
535         cur = *curp;
536         ifp = XFS_IFORK_PTR(ip, whichfork);
537         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
538         ASSERT(idx <= nextents);
539         da_old = da_new = 0;
540         error = 0;
541         /*
542          * This is the first extent added to a new/empty file.
543          * Special case this one, so other routines get to assume there are
544          * already extents in the list.
545          */
546         if (nextents == 0) {
547                 XFS_BMAP_TRACE_INSERT("insert empty", ip, 0, 1, new, NULL,
548                         whichfork);
549                 xfs_iext_insert(ifp, 0, 1, new);
550                 ASSERT(cur == NULL);
551                 ifp->if_lastex = 0;
552                 if (!ISNULLSTARTBLOCK(new->br_startblock)) {
553                         XFS_IFORK_NEXT_SET(ip, whichfork, 1);
554                         logflags = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
555                 } else
556                         logflags = 0;
557                 /* DELTA: single new extent */
558                 if (delta) {
559                         if (delta->xed_startoff > new->br_startoff)
560                                 delta->xed_startoff = new->br_startoff;
561                         if (delta->xed_blockcount <
562                                         new->br_startoff + new->br_blockcount)
563                                 delta->xed_blockcount = new->br_startoff +
564                                                 new->br_blockcount;
565                 }
566         }
567         /*
568          * Any kind of new delayed allocation goes here.
569          */
570         else if (ISNULLSTARTBLOCK(new->br_startblock)) {
571                 if (cur)
572                         ASSERT((cur->bc_private.b.flags &
573                                 XFS_BTCUR_BPRV_WASDEL) == 0);
574                 if ((error = xfs_bmap_add_extent_hole_delay(ip, idx, new,
575                                 &logflags, delta, rsvd)))
576                         goto done;
577         }
578         /*
579          * Real allocation off the end of the file.
580          */
581         else if (idx == nextents) {
582                 if (cur)
583                         ASSERT((cur->bc_private.b.flags &
584                                 XFS_BTCUR_BPRV_WASDEL) == 0);
585                 if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur, new,
586                                 &logflags, delta, whichfork)))
587                         goto done;
588         } else {
589                 xfs_bmbt_irec_t prev;   /* old extent at offset idx */
590
591                 /*
592                  * Get the record referred to by idx.
593                  */
594                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx), &prev);
595                 /*
596                  * If it's a real allocation record, and the new allocation ends
597                  * after the start of the referred to record, then we're filling
598                  * in a delayed or unwritten allocation with a real one, or
599                  * converting real back to unwritten.
600                  */
601                 if (!ISNULLSTARTBLOCK(new->br_startblock) &&
602                     new->br_startoff + new->br_blockcount > prev.br_startoff) {
603                         if (prev.br_state != XFS_EXT_UNWRITTEN &&
604                             ISNULLSTARTBLOCK(prev.br_startblock)) {
605                                 da_old = STARTBLOCKVAL(prev.br_startblock);
606                                 if (cur)
607                                         ASSERT(cur->bc_private.b.flags &
608                                                 XFS_BTCUR_BPRV_WASDEL);
609                                 if ((error = xfs_bmap_add_extent_delay_real(ip,
610                                         idx, &cur, new, &da_new, first, flist,
611                                         &logflags, delta, rsvd)))
612                                         goto done;
613                         } else if (new->br_state == XFS_EXT_NORM) {
614                                 ASSERT(new->br_state == XFS_EXT_NORM);
615                                 if ((error = xfs_bmap_add_extent_unwritten_real(
616                                         ip, idx, &cur, new, &logflags, delta)))
617                                         goto done;
618                         } else {
619                                 ASSERT(new->br_state == XFS_EXT_UNWRITTEN);
620                                 if ((error = xfs_bmap_add_extent_unwritten_real(
621                                         ip, idx, &cur, new, &logflags, delta)))
622                                         goto done;
623                         }
624                         ASSERT(*curp == cur || *curp == NULL);
625                 }
626                 /*
627                  * Otherwise we're filling in a hole with an allocation.
628                  */
629                 else {
630                         if (cur)
631                                 ASSERT((cur->bc_private.b.flags &
632                                         XFS_BTCUR_BPRV_WASDEL) == 0);
633                         if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur,
634                                         new, &logflags, delta, whichfork)))
635                                 goto done;
636                 }
637         }
638
639         ASSERT(*curp == cur || *curp == NULL);
640         /*
641          * Convert to a btree if necessary.
642          */
643         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
644             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
645                 int     tmp_logflags;   /* partial log flag return val */
646
647                 ASSERT(cur == NULL);
648                 error = xfs_bmap_extents_to_btree(ip->i_transp, ip, first,
649                         flist, &cur, da_old > 0, &tmp_logflags, whichfork);
650                 logflags |= tmp_logflags;
651                 if (error)
652                         goto done;
653         }
654         /*
655          * Adjust for changes in reserved delayed indirect blocks.
656          * Nothing to do for disk quotas here.
657          */
658         if (da_old || da_new) {
659                 xfs_filblks_t   nblks;
660
661                 nblks = da_new;
662                 if (cur)
663                         nblks += cur->bc_private.b.allocated;
664                 ASSERT(nblks <= da_old);
665                 if (nblks < da_old)
666                         xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
667                                 (int64_t)(da_old - nblks), rsvd);
668         }
669         /*
670          * Clear out the allocated field, done with it now in any case.
671          */
672         if (cur) {
673                 cur->bc_private.b.allocated = 0;
674                 *curp = cur;
675         }
676 done:
677 #ifdef DEBUG
678         if (!error)
679                 xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
680 #endif
681         *logflagsp = logflags;
682         return error;
683 }
684
685 /*
686  * Called by xfs_bmap_add_extent to handle cases converting a delayed
687  * allocation to a real allocation.
688  */
689 STATIC int                              /* error */
690 xfs_bmap_add_extent_delay_real(
691         xfs_inode_t             *ip,    /* incore inode pointer */
692         xfs_extnum_t            idx,    /* extent number to update/insert */
693         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
694         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
695         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
696         xfs_fsblock_t           *first, /* pointer to firstblock variable */
697         xfs_bmap_free_t         *flist, /* list of extents to be freed */
698         int                     *logflagsp, /* inode logging flags */
699         xfs_extdelta_t          *delta, /* Change made to incore extents */
700         int                     rsvd)   /* OK to use reserved data block allocation */
701 {
702         xfs_btree_cur_t         *cur;   /* btree cursor */
703         int                     diff;   /* temp value */
704         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
705         int                     error;  /* error return value */
706         int                     i;      /* temp state */
707         xfs_ifork_t             *ifp;   /* inode fork pointer */
708         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
709         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
710                                         /* left is 0, right is 1, prev is 2 */
711         int                     rval=0; /* return value (logging flags) */
712         int                     state = 0;/* state bits, accessed thru macros */
713         xfs_filblks_t           temp=0; /* value for dnew calculations */
714         xfs_filblks_t           temp2=0;/* value for dnew calculations */
715         int                     tmp_rval;       /* partial logging flags */
716         enum {                          /* bit number definitions for state */
717                 LEFT_CONTIG,    RIGHT_CONTIG,
718                 LEFT_FILLING,   RIGHT_FILLING,
719                 LEFT_DELAY,     RIGHT_DELAY,
720                 LEFT_VALID,     RIGHT_VALID
721         };
722
723 #define LEFT            r[0]
724 #define RIGHT           r[1]
725 #define PREV            r[2]
726 #define MASK(b)         (1 << (b))
727 #define MASK2(a,b)      (MASK(a) | MASK(b))
728 #define MASK3(a,b,c)    (MASK2(a,b) | MASK(c))
729 #define MASK4(a,b,c,d)  (MASK3(a,b,c) | MASK(d))
730 #define STATE_SET(b,v)  ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
731 #define STATE_TEST(b)   (state & MASK(b))
732 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
733                                        ((state &= ~MASK(b)), 0))
734 #define SWITCH_STATE            \
735         (state & MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG))
736
737         /*
738          * Set up a bunch of variables to make the tests simpler.
739          */
740         cur = *curp;
741         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
742         ep = xfs_iext_get_ext(ifp, idx);
743         xfs_bmbt_get_all(ep, &PREV);
744         new_endoff = new->br_startoff + new->br_blockcount;
745         ASSERT(PREV.br_startoff <= new->br_startoff);
746         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
747         /*
748          * Set flags determining what part of the previous delayed allocation
749          * extent is being replaced by a real allocation.
750          */
751         STATE_SET(LEFT_FILLING, PREV.br_startoff == new->br_startoff);
752         STATE_SET(RIGHT_FILLING,
753                 PREV.br_startoff + PREV.br_blockcount == new_endoff);
754         /*
755          * Check and set flags if this segment has a left neighbor.
756          * Don't set contiguous if the combined extent would be too large.
757          */
758         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
759                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &LEFT);
760                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(LEFT.br_startblock));
761         }
762         STATE_SET(LEFT_CONTIG,
763                 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
764                 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
765                 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
766                 LEFT.br_state == new->br_state &&
767                 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN);
768         /*
769          * Check and set flags if this segment has a right neighbor.
770          * Don't set contiguous if the combined extent would be too large.
771          * Also check for all-three-contiguous being too large.
772          */
773         if (STATE_SET_TEST(RIGHT_VALID,
774                         idx <
775                         ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1)) {
776                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx + 1), &RIGHT);
777                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(RIGHT.br_startblock));
778         }
779         STATE_SET(RIGHT_CONTIG,
780                 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
781                 new_endoff == RIGHT.br_startoff &&
782                 new->br_startblock + new->br_blockcount ==
783                     RIGHT.br_startblock &&
784                 new->br_state == RIGHT.br_state &&
785                 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
786                 ((state & MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING)) !=
787                   MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING) ||
788                  LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
789                      <= MAXEXTLEN));
790         error = 0;
791         /*
792          * Switch out based on the FILLING and CONTIG state bits.
793          */
794         switch (SWITCH_STATE) {
795
796         case MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
797                 /*
798                  * Filling in all of a previously delayed allocation extent.
799                  * The left and right neighbors are both contiguous with new.
800                  */
801                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF|LC|RC", ip, idx - 1,
802                         XFS_DATA_FORK);
803                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
804                         LEFT.br_blockcount + PREV.br_blockcount +
805                         RIGHT.br_blockcount);
806                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF|LC|RC", ip, idx - 1,
807                         XFS_DATA_FORK);
808                 XFS_BMAP_TRACE_DELETE("LF|RF|LC|RC", ip, idx, 2, XFS_DATA_FORK);
809                 xfs_iext_remove(ifp, idx, 2);
810                 ip->i_df.if_lastex = idx - 1;
811                 ip->i_d.di_nextents--;
812                 if (cur == NULL)
813                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
814                 else {
815                         rval = XFS_ILOG_CORE;
816                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
817                                         RIGHT.br_startblock,
818                                         RIGHT.br_blockcount, &i)))
819                                 goto done;
820                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
821                         if ((error = xfs_bmbt_delete(cur, &i)))
822                                 goto done;
823                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
824                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
825                                 goto done;
826                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
827                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
828                                         LEFT.br_startblock,
829                                         LEFT.br_blockcount +
830                                         PREV.br_blockcount +
831                                         RIGHT.br_blockcount, LEFT.br_state)))
832                                 goto done;
833                 }
834                 *dnew = 0;
835                 /* DELTA: Three in-core extents are replaced by one. */
836                 temp = LEFT.br_startoff;
837                 temp2 = LEFT.br_blockcount +
838                         PREV.br_blockcount +
839                         RIGHT.br_blockcount;
840                 break;
841
842         case MASK3(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG):
843                 /*
844                  * Filling in all of a previously delayed allocation extent.
845                  * The left neighbor is contiguous, the right is not.
846                  */
847                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF|LC", ip, idx - 1,
848                         XFS_DATA_FORK);
849                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
850                         LEFT.br_blockcount + PREV.br_blockcount);
851                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF|LC", ip, idx - 1,
852                         XFS_DATA_FORK);
853                 ip->i_df.if_lastex = idx - 1;
854                 XFS_BMAP_TRACE_DELETE("LF|RF|LC", ip, idx, 1, XFS_DATA_FORK);
855                 xfs_iext_remove(ifp, idx, 1);
856                 if (cur == NULL)
857                         rval = XFS_ILOG_DEXT;
858                 else {
859                         rval = 0;
860                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
861                                         LEFT.br_startblock, LEFT.br_blockcount,
862                                         &i)))
863                                 goto done;
864                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
865                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
866                                         LEFT.br_startblock,
867                                         LEFT.br_blockcount +
868                                         PREV.br_blockcount, LEFT.br_state)))
869                                 goto done;
870                 }
871                 *dnew = 0;
872                 /* DELTA: Two in-core extents are replaced by one. */
873                 temp = LEFT.br_startoff;
874                 temp2 = LEFT.br_blockcount +
875                         PREV.br_blockcount;
876                 break;
877
878         case MASK3(LEFT_FILLING, RIGHT_FILLING, RIGHT_CONTIG):
879                 /*
880                  * Filling in all of a previously delayed allocation extent.
881                  * The right neighbor is contiguous, the left is not.
882                  */
883                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF|RC", ip, idx, XFS_DATA_FORK);
884                 xfs_bmbt_set_startblock(ep, new->br_startblock);
885                 xfs_bmbt_set_blockcount(ep,
886                         PREV.br_blockcount + RIGHT.br_blockcount);
887                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF|RC", ip, idx, XFS_DATA_FORK);
888                 ip->i_df.if_lastex = idx;
889                 XFS_BMAP_TRACE_DELETE("LF|RF|RC", ip, idx + 1, 1, XFS_DATA_FORK);
890                 xfs_iext_remove(ifp, idx + 1, 1);
891                 if (cur == NULL)
892                         rval = XFS_ILOG_DEXT;
893                 else {
894                         rval = 0;
895                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
896                                         RIGHT.br_startblock,
897                                         RIGHT.br_blockcount, &i)))
898                                 goto done;
899                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
900                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
901                                         new->br_startblock,
902                                         PREV.br_blockcount +
903                                         RIGHT.br_blockcount, PREV.br_state)))
904                                 goto done;
905                 }
906                 *dnew = 0;
907                 /* DELTA: Two in-core extents are replaced by one. */
908                 temp = PREV.br_startoff;
909                 temp2 = PREV.br_blockcount +
910                         RIGHT.br_blockcount;
911                 break;
912
913         case MASK2(LEFT_FILLING, RIGHT_FILLING):
914                 /*
915                  * Filling in all of a previously delayed allocation extent.
916                  * Neither the left nor right neighbors are contiguous with
917                  * the new one.
918                  */
919                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF", ip, idx, XFS_DATA_FORK);
920                 xfs_bmbt_set_startblock(ep, new->br_startblock);
921                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF", ip, idx, XFS_DATA_FORK);
922                 ip->i_df.if_lastex = idx;
923                 ip->i_d.di_nextents++;
924                 if (cur == NULL)
925                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
926                 else {
927                         rval = XFS_ILOG_CORE;
928                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
929                                         new->br_startblock, new->br_blockcount,
930                                         &i)))
931                                 goto done;
932                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
933                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
934                         if ((error = xfs_bmbt_insert(cur, &i)))
935                                 goto done;
936                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
937                 }
938                 *dnew = 0;
939                 /* DELTA: The in-core extent described by new changed type. */
940                 temp = new->br_startoff;
941                 temp2 = new->br_blockcount;
942                 break;
943
944         case MASK2(LEFT_FILLING, LEFT_CONTIG):
945                 /*
946                  * Filling in the first part of a previous delayed allocation.
947                  * The left neighbor is contiguous.
948                  */
949                 XFS_BMAP_TRACE_PRE_UPDATE("LF|LC", ip, idx - 1, XFS_DATA_FORK);
950                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
951                         LEFT.br_blockcount + new->br_blockcount);
952                 xfs_bmbt_set_startoff(ep,
953                         PREV.br_startoff + new->br_blockcount);
954                 XFS_BMAP_TRACE_POST_UPDATE("LF|LC", ip, idx - 1, XFS_DATA_FORK);
955                 temp = PREV.br_blockcount - new->br_blockcount;
956                 XFS_BMAP_TRACE_PRE_UPDATE("LF|LC", ip, idx, XFS_DATA_FORK);
957                 xfs_bmbt_set_blockcount(ep, temp);
958                 ip->i_df.if_lastex = idx - 1;
959                 if (cur == NULL)
960                         rval = XFS_ILOG_DEXT;
961                 else {
962                         rval = 0;
963                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
964                                         LEFT.br_startblock, LEFT.br_blockcount,
965                                         &i)))
966                                 goto done;
967                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
968                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
969                                         LEFT.br_startblock,
970                                         LEFT.br_blockcount +
971                                         new->br_blockcount,
972                                         LEFT.br_state)))
973                                 goto done;
974                 }
975                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
976                         STARTBLOCKVAL(PREV.br_startblock));
977                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
978                 XFS_BMAP_TRACE_POST_UPDATE("LF|LC", ip, idx, XFS_DATA_FORK);
979                 *dnew = temp;
980                 /* DELTA: The boundary between two in-core extents moved. */
981                 temp = LEFT.br_startoff;
982                 temp2 = LEFT.br_blockcount +
983                         PREV.br_blockcount;
984                 break;
985
986         case MASK(LEFT_FILLING):
987                 /*
988                  * Filling in the first part of a previous delayed allocation.
989                  * The left neighbor is not contiguous.
990                  */
991                 XFS_BMAP_TRACE_PRE_UPDATE("LF", ip, idx, XFS_DATA_FORK);
992                 xfs_bmbt_set_startoff(ep, new_endoff);
993                 temp = PREV.br_blockcount - new->br_blockcount;
994                 xfs_bmbt_set_blockcount(ep, temp);
995                 XFS_BMAP_TRACE_INSERT("LF", ip, idx, 1, new, NULL,
996                         XFS_DATA_FORK);
997                 xfs_iext_insert(ifp, idx, 1, new);
998                 ip->i_df.if_lastex = idx;
999                 ip->i_d.di_nextents++;
1000                 if (cur == NULL)
1001                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1002                 else {
1003                         rval = XFS_ILOG_CORE;
1004                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1005                                         new->br_startblock, new->br_blockcount,
1006                                         &i)))
1007                                 goto done;
1008                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1009                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1010                         if ((error = xfs_bmbt_insert(cur, &i)))
1011                                 goto done;
1012                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1013                 }
1014                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1015                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1016                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1017                                         first, flist, &cur, 1, &tmp_rval,
1018                                         XFS_DATA_FORK);
1019                         rval |= tmp_rval;
1020                         if (error)
1021                                 goto done;
1022                 }
1023                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1024                         STARTBLOCKVAL(PREV.br_startblock) -
1025                         (cur ? cur->bc_private.b.allocated : 0));
1026                 ep = xfs_iext_get_ext(ifp, idx + 1);
1027                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1028                 XFS_BMAP_TRACE_POST_UPDATE("LF", ip, idx + 1, XFS_DATA_FORK);
1029                 *dnew = temp;
1030                 /* DELTA: One in-core extent is split in two. */
1031                 temp = PREV.br_startoff;
1032                 temp2 = PREV.br_blockcount;
1033                 break;
1034
1035         case MASK2(RIGHT_FILLING, RIGHT_CONTIG):
1036                 /*
1037                  * Filling in the last part of a previous delayed allocation.
1038                  * The right neighbor is contiguous with the new allocation.
1039                  */
1040                 temp = PREV.br_blockcount - new->br_blockcount;
1041                 XFS_BMAP_TRACE_PRE_UPDATE("RF|RC", ip, idx, XFS_DATA_FORK);
1042                 XFS_BMAP_TRACE_PRE_UPDATE("RF|RC", ip, idx + 1, XFS_DATA_FORK);
1043                 xfs_bmbt_set_blockcount(ep, temp);
1044                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, idx + 1),
1045                         new->br_startoff, new->br_startblock,
1046                         new->br_blockcount + RIGHT.br_blockcount,
1047                         RIGHT.br_state);
1048                 XFS_BMAP_TRACE_POST_UPDATE("RF|RC", ip, idx + 1, XFS_DATA_FORK);
1049                 ip->i_df.if_lastex = idx + 1;
1050                 if (cur == NULL)
1051                         rval = XFS_ILOG_DEXT;
1052                 else {
1053                         rval = 0;
1054                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1055                                         RIGHT.br_startblock,
1056                                         RIGHT.br_blockcount, &i)))
1057                                 goto done;
1058                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1059                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1060                                         new->br_startblock,
1061                                         new->br_blockcount +
1062                                         RIGHT.br_blockcount,
1063                                         RIGHT.br_state)))
1064                                 goto done;
1065                 }
1066                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1067                         STARTBLOCKVAL(PREV.br_startblock));
1068                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1069                 XFS_BMAP_TRACE_POST_UPDATE("RF|RC", ip, idx, XFS_DATA_FORK);
1070                 *dnew = temp;
1071                 /* DELTA: The boundary between two in-core extents moved. */
1072                 temp = PREV.br_startoff;
1073                 temp2 = PREV.br_blockcount +
1074                         RIGHT.br_blockcount;
1075                 break;
1076
1077         case MASK(RIGHT_FILLING):
1078                 /*
1079                  * Filling in the last part of a previous delayed allocation.
1080                  * The right neighbor is not contiguous.
1081                  */
1082                 temp = PREV.br_blockcount - new->br_blockcount;
1083                 XFS_BMAP_TRACE_PRE_UPDATE("RF", ip, idx, XFS_DATA_FORK);
1084                 xfs_bmbt_set_blockcount(ep, temp);
1085                 XFS_BMAP_TRACE_INSERT("RF", ip, idx + 1, 1, new, NULL,
1086                         XFS_DATA_FORK);
1087                 xfs_iext_insert(ifp, idx + 1, 1, new);
1088                 ip->i_df.if_lastex = idx + 1;
1089                 ip->i_d.di_nextents++;
1090                 if (cur == NULL)
1091                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1092                 else {
1093                         rval = XFS_ILOG_CORE;
1094                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1095                                         new->br_startblock, new->br_blockcount,
1096                                         &i)))
1097                                 goto done;
1098                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1099                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1100                         if ((error = xfs_bmbt_insert(cur, &i)))
1101                                 goto done;
1102                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1103                 }
1104                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1105                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1106                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1107                                 first, flist, &cur, 1, &tmp_rval,
1108                                 XFS_DATA_FORK);
1109                         rval |= tmp_rval;
1110                         if (error)
1111                                 goto done;
1112                 }
1113                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1114                         STARTBLOCKVAL(PREV.br_startblock) -
1115                         (cur ? cur->bc_private.b.allocated : 0));
1116                 ep = xfs_iext_get_ext(ifp, idx);
1117                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1118                 XFS_BMAP_TRACE_POST_UPDATE("RF", ip, idx, XFS_DATA_FORK);
1119                 *dnew = temp;
1120                 /* DELTA: One in-core extent is split in two. */
1121                 temp = PREV.br_startoff;
1122                 temp2 = PREV.br_blockcount;
1123                 break;
1124
1125         case 0:
1126                 /*
1127                  * Filling in the middle part of a previous delayed allocation.
1128                  * Contiguity is impossible here.
1129                  * This case is avoided almost all the time.
1130                  */
1131                 temp = new->br_startoff - PREV.br_startoff;
1132                 XFS_BMAP_TRACE_PRE_UPDATE("0", ip, idx, XFS_DATA_FORK);
1133                 xfs_bmbt_set_blockcount(ep, temp);
1134                 r[0] = *new;
1135                 r[1].br_state = PREV.br_state;
1136                 r[1].br_startblock = 0;
1137                 r[1].br_startoff = new_endoff;
1138                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
1139                 r[1].br_blockcount = temp2;
1140                 XFS_BMAP_TRACE_INSERT("0", ip, idx + 1, 2, &r[0], &r[1],
1141                         XFS_DATA_FORK);
1142                 xfs_iext_insert(ifp, idx + 1, 2, &r[0]);
1143                 ip->i_df.if_lastex = idx + 1;
1144                 ip->i_d.di_nextents++;
1145                 if (cur == NULL)
1146                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1147                 else {
1148                         rval = XFS_ILOG_CORE;
1149                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1150                                         new->br_startblock, new->br_blockcount,
1151                                         &i)))
1152                                 goto done;
1153                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1154                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1155                         if ((error = xfs_bmbt_insert(cur, &i)))
1156                                 goto done;
1157                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1158                 }
1159                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1160                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1161                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1162                                         first, flist, &cur, 1, &tmp_rval,
1163                                         XFS_DATA_FORK);
1164                         rval |= tmp_rval;
1165                         if (error)
1166                                 goto done;
1167                 }
1168                 temp = xfs_bmap_worst_indlen(ip, temp);
1169                 temp2 = xfs_bmap_worst_indlen(ip, temp2);
1170                 diff = (int)(temp + temp2 - STARTBLOCKVAL(PREV.br_startblock) -
1171                         (cur ? cur->bc_private.b.allocated : 0));
1172                 if (diff > 0 &&
1173                     xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS, -((int64_t)diff), rsvd)) {
1174                         /*
1175                          * Ick gross gag me with a spoon.
1176                          */
1177                         ASSERT(0);      /* want to see if this ever happens! */
1178                         while (diff > 0) {
1179                                 if (temp) {
1180                                         temp--;
1181                                         diff--;
1182                                         if (!diff ||
1183                                             !xfs_mod_incore_sb(ip->i_mount,
1184                                                     XFS_SBS_FDBLOCKS, -((int64_t)diff), rsvd))
1185                                                 break;
1186                                 }
1187                                 if (temp2) {
1188                                         temp2--;
1189                                         diff--;
1190                                         if (!diff ||
1191                                             !xfs_mod_incore_sb(ip->i_mount,
1192                                                     XFS_SBS_FDBLOCKS, -((int64_t)diff), rsvd))
1193                                                 break;
1194                                 }
1195                         }
1196                 }
1197                 ep = xfs_iext_get_ext(ifp, idx);
1198                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1199                 XFS_BMAP_TRACE_POST_UPDATE("0", ip, idx, XFS_DATA_FORK);
1200                 XFS_BMAP_TRACE_PRE_UPDATE("0", ip, idx + 2, XFS_DATA_FORK);
1201                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx + 2),
1202                         NULLSTARTBLOCK((int)temp2));
1203                 XFS_BMAP_TRACE_POST_UPDATE("0", ip, idx + 2, XFS_DATA_FORK);
1204                 *dnew = temp + temp2;
1205                 /* DELTA: One in-core extent is split in three. */
1206                 temp = PREV.br_startoff;
1207                 temp2 = PREV.br_blockcount;
1208                 break;
1209
1210         case MASK3(LEFT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1211         case MASK3(RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1212         case MASK2(LEFT_FILLING, RIGHT_CONTIG):
1213         case MASK2(RIGHT_FILLING, LEFT_CONTIG):
1214         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1215         case MASK(LEFT_CONTIG):
1216         case MASK(RIGHT_CONTIG):
1217                 /*
1218                  * These cases are all impossible.
1219                  */
1220                 ASSERT(0);
1221         }
1222         *curp = cur;
1223         if (delta) {
1224                 temp2 += temp;
1225                 if (delta->xed_startoff > temp)
1226                         delta->xed_startoff = temp;
1227                 if (delta->xed_blockcount < temp2)
1228                         delta->xed_blockcount = temp2;
1229         }
1230 done:
1231         *logflagsp = rval;
1232         return error;
1233 #undef  LEFT
1234 #undef  RIGHT
1235 #undef  PREV
1236 #undef  MASK
1237 #undef  MASK2
1238 #undef  MASK3
1239 #undef  MASK4
1240 #undef  STATE_SET
1241 #undef  STATE_TEST
1242 #undef  STATE_SET_TEST
1243 #undef  SWITCH_STATE
1244 }
1245
1246 /*
1247  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
1248  * allocation to a real allocation or vice versa.
1249  */
1250 STATIC int                              /* error */
1251 xfs_bmap_add_extent_unwritten_real(
1252         xfs_inode_t             *ip,    /* incore inode pointer */
1253         xfs_extnum_t            idx,    /* extent number to update/insert */
1254         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
1255         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1256         int                     *logflagsp, /* inode logging flags */
1257         xfs_extdelta_t          *delta) /* Change made to incore extents */
1258 {
1259         xfs_btree_cur_t         *cur;   /* btree cursor */
1260         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
1261         int                     error;  /* error return value */
1262         int                     i;      /* temp state */
1263         xfs_ifork_t             *ifp;   /* inode fork pointer */
1264         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1265         xfs_exntst_t            newext; /* new extent state */
1266         xfs_exntst_t            oldext; /* old extent state */
1267         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1268                                         /* left is 0, right is 1, prev is 2 */
1269         int                     rval=0; /* return value (logging flags) */
1270         int                     state = 0;/* state bits, accessed thru macros */
1271         xfs_filblks_t           temp=0;
1272         xfs_filblks_t           temp2=0;
1273         enum {                          /* bit number definitions for state */
1274                 LEFT_CONTIG,    RIGHT_CONTIG,
1275                 LEFT_FILLING,   RIGHT_FILLING,
1276                 LEFT_DELAY,     RIGHT_DELAY,
1277                 LEFT_VALID,     RIGHT_VALID
1278         };
1279
1280 #define LEFT            r[0]
1281 #define RIGHT           r[1]
1282 #define PREV            r[2]
1283 #define MASK(b)         (1 << (b))
1284 #define MASK2(a,b)      (MASK(a) | MASK(b))
1285 #define MASK3(a,b,c)    (MASK2(a,b) | MASK(c))
1286 #define MASK4(a,b,c,d)  (MASK3(a,b,c) | MASK(d))
1287 #define STATE_SET(b,v)  ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1288 #define STATE_TEST(b)   (state & MASK(b))
1289 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
1290                                        ((state &= ~MASK(b)), 0))
1291 #define SWITCH_STATE            \
1292         (state & MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG))
1293
1294         /*
1295          * Set up a bunch of variables to make the tests simpler.
1296          */
1297         error = 0;
1298         cur = *curp;
1299         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1300         ep = xfs_iext_get_ext(ifp, idx);
1301         xfs_bmbt_get_all(ep, &PREV);
1302         newext = new->br_state;
1303         oldext = (newext == XFS_EXT_UNWRITTEN) ?
1304                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
1305         ASSERT(PREV.br_state == oldext);
1306         new_endoff = new->br_startoff + new->br_blockcount;
1307         ASSERT(PREV.br_startoff <= new->br_startoff);
1308         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1309         /*
1310          * Set flags determining what part of the previous oldext allocation
1311          * extent is being replaced by a newext allocation.
1312          */
1313         STATE_SET(LEFT_FILLING, PREV.br_startoff == new->br_startoff);
1314         STATE_SET(RIGHT_FILLING,
1315                 PREV.br_startoff + PREV.br_blockcount == new_endoff);
1316         /*
1317          * Check and set flags if this segment has a left neighbor.
1318          * Don't set contiguous if the combined extent would be too large.
1319          */
1320         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1321                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &LEFT);
1322                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(LEFT.br_startblock));
1323         }
1324         STATE_SET(LEFT_CONTIG,
1325                 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
1326                 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1327                 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1328                 LEFT.br_state == newext &&
1329                 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1330         /*
1331          * Check and set flags if this segment has a right neighbor.
1332          * Don't set contiguous if the combined extent would be too large.
1333          * Also check for all-three-contiguous being too large.
1334          */
1335         if (STATE_SET_TEST(RIGHT_VALID,
1336                         idx <
1337                         ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1)) {
1338                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx + 1), &RIGHT);
1339                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(RIGHT.br_startblock));
1340         }
1341         STATE_SET(RIGHT_CONTIG,
1342                 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
1343                 new_endoff == RIGHT.br_startoff &&
1344                 new->br_startblock + new->br_blockcount ==
1345                     RIGHT.br_startblock &&
1346                 newext == RIGHT.br_state &&
1347                 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1348                 ((state & MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING)) !=
1349                   MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING) ||
1350                  LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1351                      <= MAXEXTLEN));
1352         /*
1353          * Switch out based on the FILLING and CONTIG state bits.
1354          */
1355         switch (SWITCH_STATE) {
1356
1357         case MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1358                 /*
1359                  * Setting all of a previous oldext extent to newext.
1360                  * The left and right neighbors are both contiguous with new.
1361                  */
1362                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF|LC|RC", ip, idx - 1,
1363                         XFS_DATA_FORK);
1364                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1365                         LEFT.br_blockcount + PREV.br_blockcount +
1366                         RIGHT.br_blockcount);
1367                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF|LC|RC", ip, idx - 1,
1368                         XFS_DATA_FORK);
1369                 XFS_BMAP_TRACE_DELETE("LF|RF|LC|RC", ip, idx, 2, XFS_DATA_FORK);
1370                 xfs_iext_remove(ifp, idx, 2);
1371                 ip->i_df.if_lastex = idx - 1;
1372                 ip->i_d.di_nextents -= 2;
1373                 if (cur == NULL)
1374                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1375                 else {
1376                         rval = XFS_ILOG_CORE;
1377                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1378                                         RIGHT.br_startblock,
1379                                         RIGHT.br_blockcount, &i)))
1380                                 goto done;
1381                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1382                         if ((error = xfs_bmbt_delete(cur, &i)))
1383                                 goto done;
1384                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1385                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1386                                 goto done;
1387                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1388                         if ((error = xfs_bmbt_delete(cur, &i)))
1389                                 goto done;
1390                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1391                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1392                                 goto done;
1393                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1394                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1395                                 LEFT.br_startblock,
1396                                 LEFT.br_blockcount + PREV.br_blockcount +
1397                                 RIGHT.br_blockcount, LEFT.br_state)))
1398                                 goto done;
1399                 }
1400                 /* DELTA: Three in-core extents are replaced by one. */
1401                 temp = LEFT.br_startoff;
1402                 temp2 = LEFT.br_blockcount +
1403                         PREV.br_blockcount +
1404                         RIGHT.br_blockcount;
1405                 break;
1406
1407         case MASK3(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG):
1408                 /*
1409                  * Setting all of a previous oldext extent to newext.
1410                  * The left neighbor is contiguous, the right is not.
1411                  */
1412                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF|LC", ip, idx - 1,
1413                         XFS_DATA_FORK);
1414                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1415                         LEFT.br_blockcount + PREV.br_blockcount);
1416                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF|LC", ip, idx - 1,
1417                         XFS_DATA_FORK);
1418                 ip->i_df.if_lastex = idx - 1;
1419                 XFS_BMAP_TRACE_DELETE("LF|RF|LC", ip, idx, 1, XFS_DATA_FORK);
1420                 xfs_iext_remove(ifp, idx, 1);
1421                 ip->i_d.di_nextents--;
1422                 if (cur == NULL)
1423                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1424                 else {
1425                         rval = XFS_ILOG_CORE;
1426                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1427                                         PREV.br_startblock, PREV.br_blockcount,
1428                                         &i)))
1429                                 goto done;
1430                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1431                         if ((error = xfs_bmbt_delete(cur, &i)))
1432                                 goto done;
1433                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1434                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1435                                 goto done;
1436                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1437                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1438                                 LEFT.br_startblock,
1439                                 LEFT.br_blockcount + PREV.br_blockcount,
1440                                 LEFT.br_state)))
1441                                 goto done;
1442                 }
1443                 /* DELTA: Two in-core extents are replaced by one. */
1444                 temp = LEFT.br_startoff;
1445                 temp2 = LEFT.br_blockcount +
1446                         PREV.br_blockcount;
1447                 break;
1448
1449         case MASK3(LEFT_FILLING, RIGHT_FILLING, RIGHT_CONTIG):
1450                 /*
1451                  * Setting all of a previous oldext extent to newext.
1452                  * The right neighbor is contiguous, the left is not.
1453                  */
1454                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF|RC", ip, idx,
1455                         XFS_DATA_FORK);
1456                 xfs_bmbt_set_blockcount(ep,
1457                         PREV.br_blockcount + RIGHT.br_blockcount);
1458                 xfs_bmbt_set_state(ep, newext);
1459                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF|RC", ip, idx,
1460                         XFS_DATA_FORK);
1461                 ip->i_df.if_lastex = idx;
1462                 XFS_BMAP_TRACE_DELETE("LF|RF|RC", ip, idx + 1, 1, XFS_DATA_FORK);
1463                 xfs_iext_remove(ifp, idx + 1, 1);
1464                 ip->i_d.di_nextents--;
1465                 if (cur == NULL)
1466                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1467                 else {
1468                         rval = XFS_ILOG_CORE;
1469                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1470                                         RIGHT.br_startblock,
1471                                         RIGHT.br_blockcount, &i)))
1472                                 goto done;
1473                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1474                         if ((error = xfs_bmbt_delete(cur, &i)))
1475                                 goto done;
1476                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1477                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1478                                 goto done;
1479                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1480                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1481                                 new->br_startblock,
1482                                 new->br_blockcount + RIGHT.br_blockcount,
1483                                 newext)))
1484                                 goto done;
1485                 }
1486                 /* DELTA: Two in-core extents are replaced by one. */
1487                 temp = PREV.br_startoff;
1488                 temp2 = PREV.br_blockcount +
1489                         RIGHT.br_blockcount;
1490                 break;
1491
1492         case MASK2(LEFT_FILLING, RIGHT_FILLING):
1493                 /*
1494                  * Setting all of a previous oldext extent to newext.
1495                  * Neither the left nor right neighbors are contiguous with
1496                  * the new one.
1497                  */
1498                 XFS_BMAP_TRACE_PRE_UPDATE("LF|RF", ip, idx,
1499                         XFS_DATA_FORK);
1500                 xfs_bmbt_set_state(ep, newext);
1501                 XFS_BMAP_TRACE_POST_UPDATE("LF|RF", ip, idx,
1502                         XFS_DATA_FORK);
1503                 ip->i_df.if_lastex = idx;
1504                 if (cur == NULL)
1505                         rval = XFS_ILOG_DEXT;
1506                 else {
1507                         rval = 0;
1508                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1509                                         new->br_startblock, new->br_blockcount,
1510                                         &i)))
1511                                 goto done;
1512                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1513                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1514                                 new->br_startblock, new->br_blockcount,
1515                                 newext)))
1516                                 goto done;
1517                 }
1518                 /* DELTA: The in-core extent described by new changed type. */
1519                 temp = new->br_startoff;
1520                 temp2 = new->br_blockcount;
1521                 break;
1522
1523         case MASK2(LEFT_FILLING, LEFT_CONTIG):
1524                 /*
1525                  * Setting the first part of a previous oldext extent to newext.
1526                  * The left neighbor is contiguous.
1527                  */
1528                 XFS_BMAP_TRACE_PRE_UPDATE("LF|LC", ip, idx - 1,
1529                         XFS_DATA_FORK);
1530                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1531                         LEFT.br_blockcount + new->br_blockcount);
1532                 xfs_bmbt_set_startoff(ep,
1533                         PREV.br_startoff + new->br_blockcount);
1534                 XFS_BMAP_TRACE_POST_UPDATE("LF|LC", ip, idx - 1,
1535                         XFS_DATA_FORK);
1536                 XFS_BMAP_TRACE_PRE_UPDATE("LF|LC", ip, idx,
1537                         XFS_DATA_FORK);
1538                 xfs_bmbt_set_startblock(ep,
1539                         new->br_startblock + new->br_blockcount);
1540                 xfs_bmbt_set_blockcount(ep,
1541                         PREV.br_blockcount - new->br_blockcount);
1542                 XFS_BMAP_TRACE_POST_UPDATE("LF|LC", ip, idx,
1543                         XFS_DATA_FORK);
1544                 ip->i_df.if_lastex = idx - 1;
1545                 if (cur == NULL)
1546                         rval = XFS_ILOG_DEXT;
1547                 else {
1548                         rval = 0;
1549                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1550                                         PREV.br_startblock, PREV.br_blockcount,
1551                                         &i)))
1552                                 goto done;
1553                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1554                         if ((error = xfs_bmbt_update(cur,
1555                                 PREV.br_startoff + new->br_blockcount,
1556                                 PREV.br_startblock + new->br_blockcount,
1557                                 PREV.br_blockcount - new->br_blockcount,
1558                                 oldext)))
1559                                 goto done;
1560                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1561                                 goto done;
1562                         if (xfs_bmbt_update(cur, LEFT.br_startoff,
1563                                 LEFT.br_startblock,
1564                                 LEFT.br_blockcount + new->br_blockcount,
1565                                 LEFT.br_state))
1566                                 goto done;
1567                 }
1568                 /* DELTA: The boundary between two in-core extents moved. */
1569                 temp = LEFT.br_startoff;
1570                 temp2 = LEFT.br_blockcount +
1571                         PREV.br_blockcount;
1572                 break;
1573
1574         case MASK(LEFT_FILLING):
1575                 /*
1576                  * Setting the first part of a previous oldext extent to newext.
1577                  * The left neighbor is not contiguous.
1578                  */
1579                 XFS_BMAP_TRACE_PRE_UPDATE("LF", ip, idx, XFS_DATA_FORK);
1580                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1581                 xfs_bmbt_set_startoff(ep, new_endoff);
1582                 xfs_bmbt_set_blockcount(ep,
1583                         PREV.br_blockcount - new->br_blockcount);
1584                 xfs_bmbt_set_startblock(ep,
1585                         new->br_startblock + new->br_blockcount);
1586                 XFS_BMAP_TRACE_POST_UPDATE("LF", ip, idx, XFS_DATA_FORK);
1587                 XFS_BMAP_TRACE_INSERT("LF", ip, idx, 1, new, NULL,
1588                         XFS_DATA_FORK);
1589                 xfs_iext_insert(ifp, idx, 1, new);
1590                 ip->i_df.if_lastex = idx;
1591                 ip->i_d.di_nextents++;
1592                 if (cur == NULL)
1593                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1594                 else {
1595                         rval = XFS_ILOG_CORE;
1596                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1597                                         PREV.br_startblock, PREV.br_blockcount,
1598                                         &i)))
1599                                 goto done;
1600                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1601                         if ((error = xfs_bmbt_update(cur,
1602                                 PREV.br_startoff + new->br_blockcount,
1603                                 PREV.br_startblock + new->br_blockcount,
1604                                 PREV.br_blockcount - new->br_blockcount,
1605                                 oldext)))
1606                                 goto done;
1607                         cur->bc_rec.b = *new;
1608                         if ((error = xfs_bmbt_insert(cur, &i)))
1609                                 goto done;
1610                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1611                 }
1612                 /* DELTA: One in-core extent is split in two. */
1613                 temp = PREV.br_startoff;
1614                 temp2 = PREV.br_blockcount;
1615                 break;
1616
1617         case MASK2(RIGHT_FILLING, RIGHT_CONTIG):
1618                 /*
1619                  * Setting the last part of a previous oldext extent to newext.
1620                  * The right neighbor is contiguous with the new allocation.
1621                  */
1622                 XFS_BMAP_TRACE_PRE_UPDATE("RF|RC", ip, idx,
1623                         XFS_DATA_FORK);
1624                 XFS_BMAP_TRACE_PRE_UPDATE("RF|RC", ip, idx + 1,
1625                         XFS_DATA_FORK);
1626                 xfs_bmbt_set_blockcount(ep,
1627                         PREV.br_blockcount - new->br_blockcount);
1628                 XFS_BMAP_TRACE_POST_UPDATE("RF|RC", ip, idx,
1629                         XFS_DATA_FORK);
1630                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, idx + 1),
1631                         new->br_startoff, new->br_startblock,
1632                         new->br_blockcount + RIGHT.br_blockcount, newext);
1633                 XFS_BMAP_TRACE_POST_UPDATE("RF|RC", ip, idx + 1,
1634                         XFS_DATA_FORK);
1635                 ip->i_df.if_lastex = idx + 1;
1636                 if (cur == NULL)
1637                         rval = XFS_ILOG_DEXT;
1638                 else {
1639                         rval = 0;
1640                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1641                                         PREV.br_startblock,
1642                                         PREV.br_blockcount, &i)))
1643                                 goto done;
1644                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1645                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1646                                 PREV.br_startblock,
1647                                 PREV.br_blockcount - new->br_blockcount,
1648                                 oldext)))
1649                                 goto done;
1650                         if ((error = xfs_bmbt_increment(cur, 0, &i)))
1651                                 goto done;
1652                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1653                                 new->br_startblock,
1654                                 new->br_blockcount + RIGHT.br_blockcount,
1655                                 newext)))
1656                                 goto done;
1657                 }
1658                 /* DELTA: The boundary between two in-core extents moved. */
1659                 temp = PREV.br_startoff;
1660                 temp2 = PREV.br_blockcount +
1661                         RIGHT.br_blockcount;
1662                 break;
1663
1664         case MASK(RIGHT_FILLING):
1665                 /*
1666                  * Setting the last part of a previous oldext extent to newext.
1667                  * The right neighbor is not contiguous.
1668                  */
1669                 XFS_BMAP_TRACE_PRE_UPDATE("RF", ip, idx, XFS_DATA_FORK);
1670                 xfs_bmbt_set_blockcount(ep,
1671                         PREV.br_blockcount - new->br_blockcount);
1672                 XFS_BMAP_TRACE_POST_UPDATE("RF", ip, idx, XFS_DATA_FORK);
1673                 XFS_BMAP_TRACE_INSERT("RF", ip, idx + 1, 1, new, NULL,
1674                         XFS_DATA_FORK);
1675                 xfs_iext_insert(ifp, idx + 1, 1, new);
1676                 ip->i_df.if_lastex = idx + 1;
1677                 ip->i_d.di_nextents++;
1678                 if (cur == NULL)
1679                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1680                 else {
1681                         rval = XFS_ILOG_CORE;
1682                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1683                                         PREV.br_startblock, PREV.br_blockcount,
1684                                         &i)))
1685                                 goto done;
1686                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1687                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1688                                 PREV.br_startblock,
1689                                 PREV.br_blockcount - new->br_blockcount,
1690                                 oldext)))
1691                                 goto done;
1692                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1693                                         new->br_startblock, new->br_blockcount,
1694                                         &i)))
1695                                 goto done;
1696                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1697                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1698                         if ((error = xfs_bmbt_insert(cur, &i)))
1699                                 goto done;
1700                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1701                 }
1702                 /* DELTA: One in-core extent is split in two. */
1703                 temp = PREV.br_startoff;
1704                 temp2 = PREV.br_blockcount;
1705                 break;
1706
1707         case 0:
1708                 /*
1709                  * Setting the middle part of a previous oldext extent to
1710                  * newext.  Contiguity is impossible here.
1711                  * One extent becomes three extents.
1712                  */
1713                 XFS_BMAP_TRACE_PRE_UPDATE("0", ip, idx, XFS_DATA_FORK);
1714                 xfs_bmbt_set_blockcount(ep,
1715                         new->br_startoff - PREV.br_startoff);
1716                 XFS_BMAP_TRACE_POST_UPDATE("0", ip, idx, XFS_DATA_FORK);
1717                 r[0] = *new;
1718                 r[1].br_startoff = new_endoff;
1719                 r[1].br_blockcount =
1720                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
1721                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1722                 r[1].br_state = oldext;
1723                 XFS_BMAP_TRACE_INSERT("0", ip, idx + 1, 2, &r[0], &r[1],
1724                         XFS_DATA_FORK);
1725                 xfs_iext_insert(ifp, idx + 1, 2, &r[0]);
1726                 ip->i_df.if_lastex = idx + 1;
1727                 ip->i_d.di_nextents += 2;
1728                 if (cur == NULL)
1729                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1730                 else {
1731                         rval = XFS_ILOG_CORE;
1732                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1733                                         PREV.br_startblock, PREV.br_blockcount,
1734                                         &i)))
1735                                 goto done;
1736                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1737                         /* new right extent - oldext */
1738                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1739                                 r[1].br_startblock, r[1].br_blockcount,
1740                                 r[1].br_state)))
1741                                 goto done;
1742                         /* new left extent - oldext */
1743                         cur->bc_rec.b = PREV;
1744                         cur->bc_rec.b.br_blockcount =
1745                                 new->br_startoff - PREV.br_startoff;
1746                         if ((error = xfs_bmbt_insert(cur, &i)))
1747                                 goto done;
1748                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1749                         /*
1750                          * Reset the cursor to the position of the new extent
1751                          * we are about to insert as we can't trust it after
1752                          * the previous insert.
1753                          */
1754                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1755                                         new->br_startblock, new->br_blockcount,
1756                                         &i)))
1757                                 goto done;
1758                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1759                         /* new middle extent - newext */
1760                         cur->bc_rec.b.br_state = new->br_state;
1761                         if ((error = xfs_bmbt_insert(cur, &i)))
1762                                 goto done;
1763                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1764                 }
1765                 /* DELTA: One in-core extent is split in three. */
1766                 temp = PREV.br_startoff;
1767                 temp2 = PREV.br_blockcount;
1768                 break;
1769
1770         case MASK3(LEFT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1771         case MASK3(RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1772         case MASK2(LEFT_FILLING, RIGHT_CONTIG):
1773         case MASK2(RIGHT_FILLING, LEFT_CONTIG):
1774         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1775         case MASK(LEFT_CONTIG):
1776         case MASK(RIGHT_CONTIG):
1777                 /*
1778                  * These cases are all impossible.
1779                  */
1780                 ASSERT(0);
1781         }
1782         *curp = cur;
1783         if (delta) {
1784                 temp2 += temp;
1785                 if (delta->xed_startoff > temp)
1786                         delta->xed_startoff = temp;
1787                 if (delta->xed_blockcount < temp2)
1788                         delta->xed_blockcount = temp2;
1789         }
1790 done:
1791         *logflagsp = rval;
1792         return error;
1793 #undef  LEFT
1794 #undef  RIGHT
1795 #undef  PREV
1796 #undef  MASK
1797 #undef  MASK2
1798 #undef  MASK3
1799 #undef  MASK4
1800 #undef  STATE_SET
1801 #undef  STATE_TEST
1802 #undef  STATE_SET_TEST
1803 #undef  SWITCH_STATE
1804 }
1805
1806 /*
1807  * Called by xfs_bmap_add_extent to handle cases converting a hole
1808  * to a delayed allocation.
1809  */
1810 /*ARGSUSED*/
1811 STATIC int                              /* error */
1812 xfs_bmap_add_extent_hole_delay(
1813         xfs_inode_t             *ip,    /* incore inode pointer */
1814         xfs_extnum_t            idx,    /* extent number to update/insert */
1815         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1816         int                     *logflagsp, /* inode logging flags */
1817         xfs_extdelta_t          *delta, /* Change made to incore extents */
1818         int                     rsvd)           /* OK to allocate reserved blocks */
1819 {
1820         xfs_bmbt_rec_host_t     *ep;    /* extent record for idx */
1821         xfs_ifork_t             *ifp;   /* inode fork pointer */
1822         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1823         xfs_filblks_t           newlen=0;       /* new indirect size */
1824         xfs_filblks_t           oldlen=0;       /* old indirect size */
1825         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1826         int                     state;  /* state bits, accessed thru macros */
1827         xfs_filblks_t           temp=0; /* temp for indirect calculations */
1828         xfs_filblks_t           temp2=0;
1829         enum {                          /* bit number definitions for state */
1830                 LEFT_CONTIG,    RIGHT_CONTIG,
1831                 LEFT_DELAY,     RIGHT_DELAY,
1832                 LEFT_VALID,     RIGHT_VALID
1833         };
1834
1835 #define MASK(b)                 (1 << (b))
1836 #define MASK2(a,b)              (MASK(a) | MASK(b))
1837 #define STATE_SET(b,v)          ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1838 #define STATE_TEST(b)           (state & MASK(b))
1839 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
1840                                        ((state &= ~MASK(b)), 0))
1841 #define SWITCH_STATE            (state & MASK2(LEFT_CONTIG, RIGHT_CONTIG))
1842
1843         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1844         ep = xfs_iext_get_ext(ifp, idx);
1845         state = 0;
1846         ASSERT(ISNULLSTARTBLOCK(new->br_startblock));
1847         /*
1848          * Check and set flags if this segment has a left neighbor
1849          */
1850         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1851                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &left);
1852                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(left.br_startblock));
1853         }
1854         /*
1855          * Check and set flags if the current (right) segment exists.
1856          * If it doesn't exist, we're converting the hole at end-of-file.
1857          */
1858         if (STATE_SET_TEST(RIGHT_VALID,
1859                            idx <
1860                            ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1861                 xfs_bmbt_get_all(ep, &right);
1862                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(right.br_startblock));
1863         }
1864         /*
1865          * Set contiguity flags on the left and right neighbors.
1866          * Don't let extents get too large, even if the pieces are contiguous.
1867          */
1868         STATE_SET(LEFT_CONTIG,
1869                 STATE_TEST(LEFT_VALID) && STATE_TEST(LEFT_DELAY) &&
1870                 left.br_startoff + left.br_blockcount == new->br_startoff &&
1871                 left.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1872         STATE_SET(RIGHT_CONTIG,
1873                 STATE_TEST(RIGHT_VALID) && STATE_TEST(RIGHT_DELAY) &&
1874                 new->br_startoff + new->br_blockcount == right.br_startoff &&
1875                 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1876                 (!STATE_TEST(LEFT_CONTIG) ||
1877                  (left.br_blockcount + new->br_blockcount +
1878                      right.br_blockcount <= MAXEXTLEN)));
1879         /*
1880          * Switch out based on the contiguity flags.
1881          */
1882         switch (SWITCH_STATE) {
1883
1884         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1885                 /*
1886                  * New allocation is contiguous with delayed allocations
1887                  * on the left and on the right.
1888                  * Merge all three into a single extent record.
1889                  */
1890                 temp = left.br_blockcount + new->br_blockcount +
1891                         right.br_blockcount;
1892                 XFS_BMAP_TRACE_PRE_UPDATE("LC|RC", ip, idx - 1,
1893                         XFS_DATA_FORK);
1894                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1), temp);
1895                 oldlen = STARTBLOCKVAL(left.br_startblock) +
1896                         STARTBLOCKVAL(new->br_startblock) +
1897                         STARTBLOCKVAL(right.br_startblock);
1898                 newlen = xfs_bmap_worst_indlen(ip, temp);
1899                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx - 1),
1900                         NULLSTARTBLOCK((int)newlen));
1901                 XFS_BMAP_TRACE_POST_UPDATE("LC|RC", ip, idx - 1,
1902                         XFS_DATA_FORK);
1903                 XFS_BMAP_TRACE_DELETE("LC|RC", ip, idx, 1, XFS_DATA_FORK);
1904                 xfs_iext_remove(ifp, idx, 1);
1905                 ip->i_df.if_lastex = idx - 1;
1906                 /* DELTA: Two in-core extents were replaced by one. */
1907                 temp2 = temp;
1908                 temp = left.br_startoff;
1909                 break;
1910
1911         case MASK(LEFT_CONTIG):
1912                 /*
1913                  * New allocation is contiguous with a delayed allocation
1914                  * on the left.
1915                  * Merge the new allocation with the left neighbor.
1916                  */
1917                 temp = left.br_blockcount + new->br_blockcount;
1918                 XFS_BMAP_TRACE_PRE_UPDATE("LC", ip, idx - 1,
1919                         XFS_DATA_FORK);
1920                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1), temp);
1921                 oldlen = STARTBLOCKVAL(left.br_startblock) +
1922                         STARTBLOCKVAL(new->br_startblock);
1923                 newlen = xfs_bmap_worst_indlen(ip, temp);
1924                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx - 1),
1925                         NULLSTARTBLOCK((int)newlen));
1926                 XFS_BMAP_TRACE_POST_UPDATE("LC", ip, idx - 1,
1927                         XFS_DATA_FORK);
1928                 ip->i_df.if_lastex = idx - 1;
1929                 /* DELTA: One in-core extent grew into a hole. */
1930                 temp2 = temp;
1931                 temp = left.br_startoff;
1932                 break;
1933
1934         case MASK(RIGHT_CONTIG):
1935                 /*
1936                  * New allocation is contiguous with a delayed allocation
1937                  * on the right.
1938                  * Merge the new allocation with the right neighbor.
1939                  */
1940                 XFS_BMAP_TRACE_PRE_UPDATE("RC", ip, idx, XFS_DATA_FORK);
1941                 temp = new->br_blockcount + right.br_blockcount;
1942                 oldlen = STARTBLOCKVAL(new->br_startblock) +
1943                         STARTBLOCKVAL(right.br_startblock);
1944                 newlen = xfs_bmap_worst_indlen(ip, temp);
1945                 xfs_bmbt_set_allf(ep, new->br_startoff,
1946                         NULLSTARTBLOCK((int)newlen), temp, right.br_state);
1947                 XFS_BMAP_TRACE_POST_UPDATE("RC", ip, idx, XFS_DATA_FORK);
1948                 ip->i_df.if_lastex = idx;
1949                 /* DELTA: One in-core extent grew into a hole. */
1950                 temp2 = temp;
1951                 temp = new->br_startoff;
1952                 break;
1953
1954         case 0:
1955                 /*
1956                  * New allocation is not contiguous with another
1957                  * delayed allocation.
1958                  * Insert a new entry.
1959                  */
1960                 oldlen = newlen = 0;
1961                 XFS_BMAP_TRACE_INSERT("0", ip, idx, 1, new, NULL,
1962                         XFS_DATA_FORK);
1963                 xfs_iext_insert(ifp, idx, 1, new);
1964                 ip->i_df.if_lastex = idx;
1965                 /* DELTA: A new in-core extent was added in a hole. */
1966                 temp2 = new->br_blockcount;
1967                 temp = new->br_startoff;
1968                 break;
1969         }
1970         if (oldlen != newlen) {
1971                 ASSERT(oldlen > newlen);
1972                 xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
1973                         (int64_t)(oldlen - newlen), rsvd);
1974                 /*
1975                  * Nothing to do for disk quota accounting here.
1976                  */
1977         }
1978         if (delta) {
1979                 temp2 += temp;
1980                 if (delta->xed_startoff > temp)
1981                         delta->xed_startoff = temp;
1982                 if (delta->xed_blockcount < temp2)
1983                         delta->xed_blockcount = temp2;
1984         }
1985         *logflagsp = 0;
1986         return 0;
1987 #undef  MASK
1988 #undef  MASK2
1989 #undef  STATE_SET
1990 #undef  STATE_TEST
1991 #undef  STATE_SET_TEST
1992 #undef  SWITCH_STATE
1993 }
1994
1995 /*
1996  * Called by xfs_bmap_add_extent to handle cases converting a hole
1997  * to a real allocation.
1998  */
1999 STATIC int                              /* error */
2000 xfs_bmap_add_extent_hole_real(
2001         xfs_inode_t             *ip,    /* incore inode pointer */
2002         xfs_extnum_t            idx,    /* extent number to update/insert */
2003         xfs_btree_cur_t         *cur,   /* if null, not a btree */
2004         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
2005         int                     *logflagsp, /* inode logging flags */
2006         xfs_extdelta_t          *delta, /* Change made to incore extents */
2007         int                     whichfork) /* data or attr fork */
2008 {
2009         xfs_bmbt_rec_host_t     *ep;    /* pointer to extent entry ins. point */
2010         int                     error;  /* error return value */
2011         int                     i;      /* temp state */
2012         xfs_ifork_t             *ifp;   /* inode fork pointer */
2013         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
2014         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
2015         int                     rval=0; /* return value (logging flags) */
2016         int                     state;  /* state bits, accessed thru macros */
2017         xfs_filblks_t           temp=0;
2018         xfs_filblks_t           temp2=0;
2019         enum {                          /* bit number definitions for state */
2020                 LEFT_CONTIG,    RIGHT_CONTIG,
2021                 LEFT_DELAY,     RIGHT_DELAY,
2022                 LEFT_VALID,     RIGHT_VALID
2023         };
2024
2025 #define MASK(b)                 (1 << (b))
2026 #define MASK2(a,b)              (MASK(a) | MASK(b))
2027 #define STATE_SET(b,v)          ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
2028 #define STATE_TEST(b)           (state & MASK(b))
2029 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
2030                                        ((state &= ~MASK(b)), 0))
2031 #define SWITCH_STATE            (state & MASK2(LEFT_CONTIG, RIGHT_CONTIG))
2032
2033         ifp = XFS_IFORK_PTR(ip, whichfork);
2034         ASSERT(idx <= ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
2035         ep = xfs_iext_get_ext(ifp, idx);
2036         state = 0;
2037         /*
2038          * Check and set flags if this segment has a left neighbor.
2039          */
2040         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
2041                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &left);
2042                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(left.br_startblock));
2043         }
2044         /*
2045          * Check and set flags if this segment has a current value.
2046          * Not true if we're inserting into the "hole" at eof.
2047          */
2048         if (STATE_SET_TEST(RIGHT_VALID,
2049                            idx <
2050                            ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
2051                 xfs_bmbt_get_all(ep, &right);
2052                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(right.br_startblock));
2053         }
2054         /*
2055          * We're inserting a real allocation between "left" and "right".
2056          * Set the contiguity flags.  Don't let extents get too large.
2057          */
2058         STATE_SET(LEFT_CONTIG,
2059                 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
2060                 left.br_startoff + left.br_blockcount == new->br_startoff &&
2061                 left.br_startblock + left.br_blockcount == new->br_startblock &&
2062                 left.br_state == new->br_state &&
2063                 left.br_blockcount + new->br_blockcount <= MAXEXTLEN);
2064         STATE_SET(RIGHT_CONTIG,
2065                 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
2066                 new->br_startoff + new->br_blockcount == right.br_startoff &&
2067                 new->br_startblock + new->br_blockcount ==
2068                     right.br_startblock &&
2069                 new->br_state == right.br_state &&
2070                 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2071                 (!STATE_TEST(LEFT_CONTIG) ||
2072                  left.br_blockcount + new->br_blockcount +
2073                      right.br_blockcount <= MAXEXTLEN));
2074
2075         error = 0;
2076         /*
2077          * Select which case we're in here, and implement it.
2078          */
2079         switch (SWITCH_STATE) {
2080
2081         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
2082                 /*
2083                  * New allocation is contiguous with real allocations on the
2084                  * left and on the right.
2085                  * Merge all three into a single extent record.
2086                  */
2087                 XFS_BMAP_TRACE_PRE_UPDATE("LC|RC", ip, idx - 1,
2088                         whichfork);
2089                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
2090                         left.br_blockcount + new->br_blockcount +
2091                         right.br_blockcount);
2092                 XFS_BMAP_TRACE_POST_UPDATE("LC|RC", ip, idx - 1,
2093                         whichfork);
2094                 XFS_BMAP_TRACE_DELETE("LC|RC", ip, idx, 1, whichfork);
2095                 xfs_iext_remove(ifp, idx, 1);
2096                 ifp->if_lastex = idx - 1;
2097                 XFS_IFORK_NEXT_SET(ip, whichfork,
2098                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2099                 if (cur == NULL) {
2100                         rval = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2101                 } else {
2102                         rval = XFS_ILOG_CORE;
2103                         if ((error = xfs_bmbt_lookup_eq(cur,
2104                                         right.br_startoff,
2105                                         right.br_startblock,
2106                                         right.br_blockcount, &i)))
2107                                 goto done;
2108                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2109                         if ((error = xfs_bmbt_delete(cur, &i)))
2110                                 goto done;
2111                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2112                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
2113                                 goto done;
2114                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2115                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
2116                                         left.br_startblock,
2117                                         left.br_blockcount +
2118                                                 new->br_blockcount +
2119                                                 right.br_blockcount,
2120                                         left.br_state)))
2121                                 goto done;
2122                 }
2123                 /* DELTA: Two in-core extents were replaced by one. */
2124                 temp = left.br_startoff;
2125                 temp2 = left.br_blockcount +
2126                         new->br_blockcount +
2127                         right.br_blockcount;
2128                 break;
2129
2130         case MASK(LEFT_CONTIG):
2131                 /*
2132                  * New allocation is contiguous with a real allocation
2133                  * on the left.
2134                  * Merge the new allocation with the left neighbor.
2135                  */
2136                 XFS_BMAP_TRACE_PRE_UPDATE("LC", ip, idx - 1, whichfork);
2137                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
2138                         left.br_blockcount + new->br_blockcount);
2139                 XFS_BMAP_TRACE_POST_UPDATE("LC", ip, idx - 1, whichfork);
2140                 ifp->if_lastex = idx - 1;
2141                 if (cur == NULL) {
2142                         rval = XFS_ILOG_FEXT(whichfork);
2143                 } else {
2144                         rval = 0;
2145                         if ((error = xfs_bmbt_lookup_eq(cur,
2146                                         left.br_startoff,
2147                                         left.br_startblock,
2148                                         left.br_blockcount, &i)))
2149                                 goto done;
2150                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2151                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
2152                                         left.br_startblock,
2153                                         left.br_blockcount +
2154                                                 new->br_blockcount,
2155                                         left.br_state)))
2156                                 goto done;
2157                 }
2158                 /* DELTA: One in-core extent grew. */
2159                 temp = left.br_startoff;
2160                 temp2 = left.br_blockcount +
2161                         new->br_blockcount;
2162                 break;
2163
2164         case MASK(RIGHT_CONTIG):
2165                 /*
2166                  * New allocation is contiguous with a real allocation
2167                  * on the right.
2168                  * Merge the new allocation with the right neighbor.
2169                  */
2170                 XFS_BMAP_TRACE_PRE_UPDATE("RC", ip, idx, whichfork);
2171                 xfs_bmbt_set_allf(ep, new->br_startoff, new->br_startblock,
2172                         new->br_blockcount + right.br_blockcount,
2173                         right.br_state);
2174                 XFS_BMAP_TRACE_POST_UPDATE("RC", ip, idx, whichfork);
2175                 ifp->if_lastex = idx;
2176                 if (cur == NULL) {
2177                         rval = XFS_ILOG_FEXT(whichfork);
2178                 } else {
2179                         rval = 0;
2180                         if ((error = xfs_bmbt_lookup_eq(cur,
2181                                         right.br_startoff,
2182                                         right.br_startblock,
2183                                         right.br_blockcount, &i)))
2184                                 goto done;
2185                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2186                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
2187                                         new->br_startblock,
2188                                         new->br_blockcount +
2189                                                 right.br_blockcount,
2190                                         right.br_state)))
2191                                 goto done;
2192                 }
2193                 /* DELTA: One in-core extent grew. */
2194                 temp = new->br_startoff;
2195                 temp2 = new->br_blockcount +
2196                         right.br_blockcount;
2197                 break;
2198
2199         case 0:
2200                 /*
2201                  * New allocation is not contiguous with another
2202                  * real allocation.
2203                  * Insert a new entry.
2204                  */
2205                 XFS_BMAP_TRACE_INSERT("0", ip, idx, 1, new, NULL, whichfork);
2206                 xfs_iext_insert(ifp, idx, 1, new);
2207                 ifp->if_lastex = idx;
2208                 XFS_IFORK_NEXT_SET(ip, whichfork,
2209                         XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2210                 if (cur == NULL) {
2211                         rval = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2212                 } else {
2213                         rval = XFS_ILOG_CORE;
2214                         if ((error = xfs_bmbt_lookup_eq(cur,
2215                                         new->br_startoff,
2216                                         new->br_startblock,
2217                                         new->br_blockcount, &i)))
2218                                 goto done;
2219                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2220                         cur->bc_rec.b.br_state = new->br_state;
2221                         if ((error = xfs_bmbt_insert(cur, &i)))
2222                                 goto done;
2223                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2224                 }
2225                 /* DELTA: A new extent was added in a hole. */
2226                 temp = new->br_startoff;
2227                 temp2 = new->br_blockcount;
2228                 break;
2229         }
2230         if (delta) {
2231                 temp2 += temp;
2232                 if (delta->xed_startoff > temp)
2233                         delta->xed_startoff = temp;
2234                 if (delta->xed_blockcount < temp2)
2235                         delta->xed_blockcount = temp2;
2236         }
2237 done:
2238         *logflagsp = rval;
2239         return error;
2240 #undef  MASK
2241 #undef  MASK2
2242 #undef  STATE_SET
2243 #undef  STATE_TEST
2244 #undef  STATE_SET_TEST
2245 #undef  SWITCH_STATE
2246 }
2247
2248 /*
2249  * Adjust the size of the new extent based on di_extsize and rt extsize.
2250  */
2251 STATIC int
2252 xfs_bmap_extsize_align(
2253         xfs_mount_t     *mp,
2254         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
2255         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
2256         xfs_extlen_t    extsz,          /* align to this extent size */
2257         int             rt,             /* is this a realtime inode? */
2258         int             eof,            /* is extent at end-of-file? */
2259         int             delay,          /* creating delalloc extent? */
2260         int             convert,        /* overwriting unwritten extent? */
2261         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
2262         xfs_extlen_t    *lenp)          /* in/out: aligned length */
2263 {
2264         xfs_fileoff_t   orig_off;       /* original offset */
2265         xfs_extlen_t    orig_alen;      /* original length */
2266         xfs_fileoff_t   orig_end;       /* original off+len */
2267         xfs_fileoff_t   nexto;          /* next file offset */
2268         xfs_fileoff_t   prevo;          /* previous file offset */
2269         xfs_fileoff_t   align_off;      /* temp for offset */
2270         xfs_extlen_t    align_alen;     /* temp for length */
2271         xfs_extlen_t    temp;           /* temp for calculations */
2272
2273         if (convert)
2274                 return 0;
2275
2276         orig_off = align_off = *offp;
2277         orig_alen = align_alen = *lenp;
2278         orig_end = orig_off + orig_alen;
2279
2280         /*
2281          * If this request overlaps an existing extent, then don't
2282          * attempt to perform any additional alignment.
2283          */
2284         if (!delay && !eof &&
2285             (orig_off >= gotp->br_startoff) &&
2286             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2287                 return 0;
2288         }
2289
2290         /*
2291          * If the file offset is unaligned vs. the extent size
2292          * we need to align it.  This will be possible unless
2293          * the file was previously written with a kernel that didn't
2294          * perform this alignment, or if a truncate shot us in the
2295          * foot.
2296          */
2297         temp = do_mod(orig_off, extsz);
2298         if (temp) {
2299                 align_alen += temp;
2300                 align_off -= temp;
2301         }
2302         /*
2303          * Same adjustment for the end of the requested area.
2304          */
2305         if ((temp = (align_alen % extsz))) {
2306                 align_alen += extsz - temp;
2307         }
2308         /*
2309          * If the previous block overlaps with this proposed allocation
2310          * then move the start forward without adjusting the length.
2311          */
2312         if (prevp->br_startoff != NULLFILEOFF) {
2313                 if (prevp->br_startblock == HOLESTARTBLOCK)
2314                         prevo = prevp->br_startoff;
2315                 else
2316                         prevo = prevp->br_startoff + prevp->br_blockcount;
2317         } else
2318                 prevo = 0;
2319         if (align_off != orig_off && align_off < prevo)
2320                 align_off = prevo;
2321         /*
2322          * If the next block overlaps with this proposed allocation
2323          * then move the start back without adjusting the length,
2324          * but not before offset 0.
2325          * This may of course make the start overlap previous block,
2326          * and if we hit the offset 0 limit then the next block
2327          * can still overlap too.
2328          */
2329         if (!eof && gotp->br_startoff != NULLFILEOFF) {
2330                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
2331                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
2332                         nexto = gotp->br_startoff + gotp->br_blockcount;
2333                 else
2334                         nexto = gotp->br_startoff;
2335         } else
2336                 nexto = NULLFILEOFF;
2337         if (!eof &&
2338             align_off + align_alen != orig_end &&
2339             align_off + align_alen > nexto)
2340                 align_off = nexto > align_alen ? nexto - align_alen : 0;
2341         /*
2342          * If we're now overlapping the next or previous extent that
2343          * means we can't fit an extsz piece in this hole.  Just move
2344          * the start forward to the first valid spot and set
2345          * the length so we hit the end.
2346          */
2347         if (align_off != orig_off && align_off < prevo)
2348                 align_off = prevo;
2349         if (align_off + align_alen != orig_end &&
2350             align_off + align_alen > nexto &&
2351             nexto != NULLFILEOFF) {
2352                 ASSERT(nexto > prevo);
2353                 align_alen = nexto - align_off;
2354         }
2355
2356         /*
2357          * If realtime, and the result isn't a multiple of the realtime
2358          * extent size we need to remove blocks until it is.
2359          */
2360         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
2361                 /*
2362                  * We're not covering the original request, or
2363                  * we won't be able to once we fix the length.
2364                  */
2365                 if (orig_off < align_off ||
2366                     orig_end > align_off + align_alen ||
2367                     align_alen - temp < orig_alen)
2368                         return XFS_ERROR(EINVAL);
2369                 /*
2370                  * Try to fix it by moving the start up.
2371                  */
2372                 if (align_off + temp <= orig_off) {
2373                         align_alen -= temp;
2374                         align_off += temp;
2375                 }
2376                 /*
2377                  * Try to fix it by moving the end in.
2378                  */
2379                 else if (align_off + align_alen - temp >= orig_end)
2380                         align_alen -= temp;
2381                 /*
2382                  * Set the start to the minimum then trim the length.
2383                  */
2384                 else {
2385                         align_alen -= orig_off - align_off;
2386                         align_off = orig_off;
2387                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
2388                 }
2389                 /*
2390                  * Result doesn't cover the request, fail it.
2391                  */
2392                 if (orig_off < align_off || orig_end > align_off + align_alen)
2393                         return XFS_ERROR(EINVAL);
2394         } else {
2395                 ASSERT(orig_off >= align_off);
2396                 ASSERT(orig_end <= align_off + align_alen);
2397         }
2398
2399 #ifdef DEBUG
2400         if (!eof && gotp->br_startoff != NULLFILEOFF)
2401                 ASSERT(align_off + align_alen <= gotp->br_startoff);
2402         if (prevp->br_startoff != NULLFILEOFF)
2403                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
2404 #endif
2405
2406         *lenp = align_alen;
2407         *offp = align_off;
2408         return 0;
2409 }
2410
2411 #define XFS_ALLOC_GAP_UNITS     4
2412
2413 STATIC void
2414 xfs_bmap_adjacent(
2415         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2416 {
2417         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
2418         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2419         xfs_mount_t     *mp;            /* mount point structure */
2420         int             nullfb;         /* true if ap->firstblock isn't set */
2421         int             rt;             /* true if inode is realtime */
2422
2423 #define ISVALID(x,y)    \
2424         (rt ? \
2425                 (x) < mp->m_sb.sb_rblocks : \
2426                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2427                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2428                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2429
2430         mp = ap->ip->i_mount;
2431         nullfb = ap->firstblock == NULLFSBLOCK;
2432         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2433         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2434         /*
2435          * If allocating at eof, and there's a previous real block,
2436          * try to use it's last block as our starting point.
2437          */
2438         if (ap->eof && ap->prevp->br_startoff != NULLFILEOFF &&
2439             !ISNULLSTARTBLOCK(ap->prevp->br_startblock) &&
2440             ISVALID(ap->prevp->br_startblock + ap->prevp->br_blockcount,
2441                     ap->prevp->br_startblock)) {
2442                 ap->rval = ap->prevp->br_startblock + ap->prevp->br_blockcount;
2443                 /*
2444                  * Adjust for the gap between prevp and us.
2445                  */
2446                 adjust = ap->off -
2447                         (ap->prevp->br_startoff + ap->prevp->br_blockcount);
2448                 if (adjust &&
2449                     ISVALID(ap->rval + adjust, ap->prevp->br_startblock))
2450                         ap->rval += adjust;
2451         }
2452         /*
2453          * If not at eof, then compare the two neighbor blocks.
2454          * Figure out whether either one gives us a good starting point,
2455          * and pick the better one.
2456          */
2457         else if (!ap->eof) {
2458                 xfs_fsblock_t   gotbno;         /* right side block number */
2459                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
2460                 xfs_fsblock_t   prevbno;        /* left side block number */
2461                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
2462
2463                 /*
2464                  * If there's a previous (left) block, select a requested
2465                  * start block based on it.
2466                  */
2467                 if (ap->prevp->br_startoff != NULLFILEOFF &&
2468                     !ISNULLSTARTBLOCK(ap->prevp->br_startblock) &&
2469                     (prevbno = ap->prevp->br_startblock +
2470                                ap->prevp->br_blockcount) &&
2471                     ISVALID(prevbno, ap->prevp->br_startblock)) {
2472                         /*
2473                          * Calculate gap to end of previous block.
2474                          */
2475                         adjust = prevdiff = ap->off -
2476                                 (ap->prevp->br_startoff +
2477                                  ap->prevp->br_blockcount);
2478                         /*
2479                          * Figure the startblock based on the previous block's
2480                          * end and the gap size.
2481                          * Heuristic!
2482                          * If the gap is large relative to the piece we're
2483                          * allocating, or using it gives us an invalid block
2484                          * number, then just use the end of the previous block.
2485                          */
2486                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2487                             ISVALID(prevbno + prevdiff,
2488                                     ap->prevp->br_startblock))
2489                                 prevbno += adjust;
2490                         else
2491                                 prevdiff += adjust;
2492                         /*
2493                          * If the firstblock forbids it, can't use it,
2494                          * must use default.
2495                          */
2496                         if (!rt && !nullfb &&
2497                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2498                                 prevbno = NULLFSBLOCK;
2499                 }
2500                 /*
2501                  * No previous block or can't follow it, just default.
2502                  */
2503                 else
2504                         prevbno = NULLFSBLOCK;
2505                 /*
2506                  * If there's a following (right) block, select a requested
2507                  * start block based on it.
2508                  */
2509                 if (!ISNULLSTARTBLOCK(ap->gotp->br_startblock)) {
2510                         /*
2511                          * Calculate gap to start of next block.
2512                          */
2513                         adjust = gotdiff = ap->gotp->br_startoff - ap->off;
2514                         /*
2515                          * Figure the startblock based on the next block's
2516                          * start and the gap size.
2517                          */
2518                         gotbno = ap->gotp->br_startblock;
2519                         /*
2520                          * Heuristic!
2521                          * If the gap is large relative to the piece we're
2522                          * allocating, or using it gives us an invalid block
2523                          * number, then just use the start of the next block
2524                          * offset by our length.
2525                          */
2526                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2527                             ISVALID(gotbno - gotdiff, gotbno))
2528                                 gotbno -= adjust;
2529                         else if (ISVALID(gotbno - ap->alen, gotbno)) {
2530                                 gotbno -= ap->alen;
2531                                 gotdiff += adjust - ap->alen;
2532                         } else
2533                                 gotdiff += adjust;
2534                         /*
2535                          * If the firstblock forbids it, can't use it,
2536                          * must use default.
2537                          */
2538                         if (!rt && !nullfb &&
2539                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2540                                 gotbno = NULLFSBLOCK;
2541                 }
2542                 /*
2543                  * No next block, just default.
2544                  */
2545                 else
2546                         gotbno = NULLFSBLOCK;
2547                 /*
2548                  * If both valid, pick the better one, else the only good
2549                  * one, else ap->rval is already set (to 0 or the inode block).
2550                  */
2551                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2552                         ap->rval = prevdiff <= gotdiff ? prevbno : gotbno;
2553                 else if (prevbno != NULLFSBLOCK)
2554                         ap->rval = prevbno;
2555                 else if (gotbno != NULLFSBLOCK)
2556                         ap->rval = gotbno;
2557         }
2558 #undef ISVALID
2559 }
2560
2561 STATIC int
2562 xfs_bmap_rtalloc(
2563         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2564 {
2565         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2566         int             error;          /* error return value */
2567         xfs_mount_t     *mp;            /* mount point structure */
2568         xfs_extlen_t    prod = 0;       /* product factor for allocators */
2569         xfs_extlen_t    ralen = 0;      /* realtime allocation length */
2570         xfs_extlen_t    align;          /* minimum allocation alignment */
2571         xfs_rtblock_t   rtb;
2572
2573         mp = ap->ip->i_mount;
2574         align = xfs_get_extsz_hint(ap->ip);
2575         prod = align / mp->m_sb.sb_rextsize;
2576         error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2577                                         align, 1, ap->eof, 0,
2578                                         ap->conv, &ap->off, &ap->alen);
2579         if (error)
2580                 return error;
2581         ASSERT(ap->alen);
2582         ASSERT(ap->alen % mp->m_sb.sb_rextsize == 0);
2583
2584         /*
2585          * If the offset & length are not perfectly aligned
2586          * then kill prod, it will just get us in trouble.
2587          */
2588         if (do_mod(ap->off, align) || ap->alen % align)
2589                 prod = 1;
2590         /*
2591          * Set ralen to be the actual requested length in rtextents.
2592          */
2593         ralen = ap->alen / mp->m_sb.sb_rextsize;
2594         /*
2595          * If the old value was close enough to MAXEXTLEN that
2596          * we rounded up to it, cut it back so it's valid again.
2597          * Note that if it's a really large request (bigger than
2598          * MAXEXTLEN), we don't hear about that number, and can't
2599          * adjust the starting point to match it.
2600          */
2601         if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2602                 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2603         /*
2604          * If it's an allocation to an empty file at offset 0,
2605          * pick an extent that will space things out in the rt area.
2606          */
2607         if (ap->eof && ap->off == 0) {
2608                 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
2609
2610                 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2611                 if (error)
2612                         return error;
2613                 ap->rval = rtx * mp->m_sb.sb_rextsize;
2614         } else {
2615                 ap->rval = 0;
2616         }
2617
2618         xfs_bmap_adjacent(ap);
2619
2620         /*
2621          * Realtime allocation, done through xfs_rtallocate_extent.
2622          */
2623         atype = ap->rval == 0 ?  XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2624         do_div(ap->rval, mp->m_sb.sb_rextsize);
2625         rtb = ap->rval;
2626         ap->alen = ralen;
2627         if ((error = xfs_rtallocate_extent(ap->tp, ap->rval, 1, ap->alen,
2628                                 &ralen, atype, ap->wasdel, prod, &rtb)))
2629                 return error;
2630         if (rtb == NULLFSBLOCK && prod > 1 &&
2631             (error = xfs_rtallocate_extent(ap->tp, ap->rval, 1,
2632                                            ap->alen, &ralen, atype,
2633                                            ap->wasdel, 1, &rtb)))
2634                 return error;
2635         ap->rval = rtb;
2636         if (ap->rval != NULLFSBLOCK) {
2637                 ap->rval *= mp->m_sb.sb_rextsize;
2638                 ralen *= mp->m_sb.sb_rextsize;
2639                 ap->alen = ralen;
2640                 ap->ip->i_d.di_nblocks += ralen;
2641                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2642                 if (ap->wasdel)
2643                         ap->ip->i_delayed_blks -= ralen;
2644                 /*
2645                  * Adjust the disk quota also. This was reserved
2646                  * earlier.
2647                  */
2648                 XFS_TRANS_MOD_DQUOT_BYINO(mp, ap->tp, ap->ip,
2649                         ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2650                                         XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2651         } else {
2652                 ap->alen = 0;
2653         }
2654         return 0;
2655 }
2656
2657 STATIC int
2658 xfs_bmap_btalloc(
2659         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2660 {
2661         xfs_mount_t     *mp;            /* mount point structure */
2662         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2663         xfs_extlen_t    align;          /* minimum allocation alignment */
2664         xfs_agnumber_t  ag;
2665         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2666         xfs_agnumber_t  startag;
2667         xfs_alloc_arg_t args;
2668         xfs_extlen_t    blen;
2669         xfs_extlen_t    delta;
2670         xfs_extlen_t    longest;
2671         xfs_extlen_t    need;
2672         xfs_extlen_t    nextminlen = 0;
2673         xfs_perag_t     *pag;
2674         int             nullfb;         /* true if ap->firstblock isn't set */
2675         int             isaligned;
2676         int             notinit;
2677         int             tryagain;
2678         int             error;
2679
2680         mp = ap->ip->i_mount;
2681         align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
2682         if (unlikely(align)) {
2683                 error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2684                                                 align, 0, ap->eof, 0, ap->conv,
2685                                                 &ap->off, &ap->alen);
2686                 ASSERT(!error);
2687                 ASSERT(ap->alen);
2688         }
2689         nullfb = ap->firstblock == NULLFSBLOCK;
2690         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2691         if (nullfb) {
2692                 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
2693                         ag = xfs_filestream_lookup_ag(ap->ip);
2694                         ag = (ag != NULLAGNUMBER) ? ag : 0;
2695                         ap->rval = XFS_AGB_TO_FSB(mp, ag, 0);
2696                 } else {
2697                         ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2698                 }
2699         } else
2700                 ap->rval = ap->firstblock;
2701
2702         xfs_bmap_adjacent(ap);
2703
2704         /*
2705          * If allowed, use ap->rval; otherwise must use firstblock since
2706          * it's in the right allocation group.
2707          */
2708         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->rval) == fb_agno)
2709                 ;
2710         else
2711                 ap->rval = ap->firstblock;
2712         /*
2713          * Normal allocation, done through xfs_alloc_vextent.
2714          */
2715         tryagain = isaligned = 0;
2716         args.tp = ap->tp;
2717         args.mp = mp;
2718         args.fsbno = ap->rval;
2719         args.maxlen = MIN(ap->alen, mp->m_sb.sb_agblocks);
2720         args.firstblock = ap->firstblock;
2721         blen = 0;
2722         if (nullfb) {
2723                 if (ap->userdata && xfs_inode_is_filestream(ap->ip))
2724                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
2725                 else
2726                         args.type = XFS_ALLOCTYPE_START_BNO;
2727                 args.total = ap->total;
2728
2729                 /*
2730                  * Search for an allocation group with a single extent
2731                  * large enough for the request.
2732                  *
2733                  * If one isn't found, then adjust the minimum allocation
2734                  * size to the largest space found.
2735                  */
2736                 startag = ag = XFS_FSB_TO_AGNO(mp, args.fsbno);
2737                 if (startag == NULLAGNUMBER)
2738                         startag = ag = 0;
2739                 notinit = 0;
2740                 down_read(&mp->m_peraglock);
2741                 while (blen < ap->alen) {
2742                         pag = &mp->m_perag[ag];
2743                         if (!pag->pagf_init &&
2744                             (error = xfs_alloc_pagf_init(mp, args.tp,
2745                                     ag, XFS_ALLOC_FLAG_TRYLOCK))) {
2746                                 up_read(&mp->m_peraglock);
2747                                 return error;
2748                         }
2749                         /*
2750                          * See xfs_alloc_fix_freelist...
2751                          */
2752                         if (pag->pagf_init) {
2753                                 need = XFS_MIN_FREELIST_PAG(pag, mp);
2754                                 delta = need > pag->pagf_flcount ?
2755                                         need - pag->pagf_flcount : 0;
2756                                 longest = (pag->pagf_longest > delta) ?
2757                                         (pag->pagf_longest - delta) :
2758                                         (pag->pagf_flcount > 0 ||
2759                                          pag->pagf_longest > 0);
2760                                 if (blen < longest)
2761                                         blen = longest;
2762                         } else
2763                                 notinit = 1;
2764
2765                         if (xfs_inode_is_filestream(ap->ip)) {
2766                                 if (blen >= ap->alen)
2767                                         break;
2768
2769                                 if (ap->userdata) {
2770                                         /*
2771                                          * If startag is an invalid AG, we've
2772                                          * come here once before and
2773                                          * xfs_filestream_new_ag picked the
2774                                          * best currently available.
2775                                          *
2776                                          * Don't continue looping, since we
2777                                          * could loop forever.
2778                                          */
2779                                         if (startag == NULLAGNUMBER)
2780                                                 break;
2781
2782                                         error = xfs_filestream_new_ag(ap, &ag);
2783                                         if (error) {
2784                                                 up_read(&mp->m_peraglock);
2785                                                 return error;
2786                                         }
2787
2788                                         /* loop again to set 'blen'*/
2789                                         startag = NULLAGNUMBER;
2790                                         continue;
2791                                 }
2792                         }
2793                         if (++ag == mp->m_sb.sb_agcount)
2794                                 ag = 0;
2795                         if (ag == startag)
2796                                 break;
2797                 }
2798                 up_read(&mp->m_peraglock);
2799                 /*
2800                  * Since the above loop did a BUF_TRYLOCK, it is
2801                  * possible that there is space for this request.
2802                  */
2803                 if (notinit || blen < ap->minlen)
2804                         args.minlen = ap->minlen;
2805                 /*
2806                  * If the best seen length is less than the request
2807                  * length, use the best as the minimum.
2808                  */
2809                 else if (blen < ap->alen)
2810                         args.minlen = blen;
2811                 /*
2812                  * Otherwise we've seen an extent as big as alen,
2813                  * use that as the minimum.
2814                  */
2815                 else
2816                         args.minlen = ap->alen;
2817
2818                 /*
2819                  * set the failure fallback case to look in the selected
2820                  * AG as the stream may have moved.
2821                  */
2822                 if (xfs_inode_is_filestream(ap->ip))
2823                         ap->rval = args.fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
2824         } else if (ap->low) {
2825                 if (xfs_inode_is_filestream(ap->ip))
2826                         args.type = XFS_ALLOCTYPE_FIRST_AG;
2827                 else
2828                         args.type = XFS_ALLOCTYPE_START_BNO;
2829                 args.total = args.minlen = ap->minlen;
2830         } else {
2831                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2832                 args.total = ap->total;
2833                 args.minlen = ap->minlen;
2834         }
2835         /* apply extent size hints if obtained earlier */
2836         if (unlikely(align)) {
2837                 args.prod = align;
2838                 if ((args.mod = (xfs_extlen_t)do_mod(ap->off, args.prod)))
2839                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2840         } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
2841                 args.prod = 1;
2842                 args.mod = 0;
2843         } else {
2844                 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
2845                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->off, args.prod))))
2846                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2847         }
2848         /*
2849          * If we are not low on available data blocks, and the
2850          * underlying logical volume manager is a stripe, and
2851          * the file offset is zero then try to allocate data
2852          * blocks on stripe unit boundary.
2853          * NOTE: ap->aeof is only set if the allocation length
2854          * is >= the stripe unit and the allocation offset is
2855          * at the end of file.
2856          */
2857         if (!ap->low && ap->aeof) {
2858                 if (!ap->off) {
2859                         args.alignment = mp->m_dalign;
2860                         atype = args.type;
2861                         isaligned = 1;
2862                         /*
2863                          * Adjust for alignment
2864                          */
2865                         if (blen > args.alignment && blen <= ap->alen)
2866                                 args.minlen = blen - args.alignment;
2867                         args.minalignslop = 0;
2868                 } else {
2869                         /*
2870                          * First try an exact bno allocation.
2871                          * If it fails then do a near or start bno
2872                          * allocation with alignment turned on.
2873                          */
2874                         atype = args.type;
2875                         tryagain = 1;
2876                         args.type = XFS_ALLOCTYPE_THIS_BNO;
2877                         args.alignment = 1;
2878                         /*
2879                          * Compute the minlen+alignment for the
2880                          * next case.  Set slop so that the value
2881                          * of minlen+alignment+slop doesn't go up
2882                          * between the calls.
2883                          */
2884                         if (blen > mp->m_dalign && blen <= ap->alen)
2885                                 nextminlen = blen - mp->m_dalign;
2886                         else
2887                                 nextminlen = args.minlen;
2888                         if (nextminlen + mp->m_dalign > args.minlen + 1)
2889                                 args.minalignslop =
2890                                         nextminlen + mp->m_dalign -
2891                                         args.minlen - 1;
2892                         else
2893                                 args.minalignslop = 0;
2894                 }
2895         } else {
2896                 args.alignment = 1;
2897                 args.minalignslop = 0;
2898         }
2899         args.minleft = ap->minleft;
2900         args.wasdel = ap->wasdel;
2901         args.isfl = 0;
2902         args.userdata = ap->userdata;
2903         if ((error = xfs_alloc_vextent(&args)))
2904                 return error;
2905         if (tryagain && args.fsbno == NULLFSBLOCK) {
2906                 /*
2907                  * Exact allocation failed. Now try with alignment
2908                  * turned on.
2909                  */
2910                 args.type = atype;
2911                 args.fsbno = ap->rval;
2912                 args.alignment = mp->m_dalign;
2913                 args.minlen = nextminlen;
2914                 args.minalignslop = 0;
2915                 isaligned = 1;
2916                 if ((error = xfs_alloc_vextent(&args)))
2917                         return error;
2918         }
2919         if (isaligned && args.fsbno == NULLFSBLOCK) {
2920                 /*
2921                  * allocation failed, so turn off alignment and
2922                  * try again.
2923                  */
2924                 args.type = atype;
2925                 args.fsbno = ap->rval;
2926                 args.alignment = 0;
2927                 if ((error = xfs_alloc_vextent(&args)))
2928                         return error;
2929         }
2930         if (args.fsbno == NULLFSBLOCK && nullfb &&
2931             args.minlen > ap->minlen) {
2932                 args.minlen = ap->minlen;
2933                 args.type = XFS_ALLOCTYPE_START_BNO;
2934                 args.fsbno = ap->rval;
2935                 if ((error = xfs_alloc_vextent(&args)))
2936                         return error;
2937         }
2938         if (args.fsbno == NULLFSBLOCK && nullfb) {
2939                 args.fsbno = 0;
2940                 args.type = XFS_ALLOCTYPE_FIRST_AG;
2941                 args.total = ap->minlen;
2942                 args.minleft = 0;
2943                 if ((error = xfs_alloc_vextent(&args)))
2944                         return error;
2945                 ap->low = 1;
2946         }
2947         if (args.fsbno != NULLFSBLOCK) {
2948                 ap->firstblock = ap->rval = args.fsbno;
2949                 ASSERT(nullfb || fb_agno == args.agno ||
2950                        (ap->low && fb_agno < args.agno));
2951                 ap->alen = args.len;
2952                 ap->ip->i_d.di_nblocks += args.len;
2953                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2954                 if (ap->wasdel)
2955                         ap->ip->i_delayed_blks -= args.len;
2956                 /*
2957                  * Adjust the disk quota also. This was reserved
2958                  * earlier.
2959                  */
2960                 XFS_TRANS_MOD_DQUOT_BYINO(mp, ap->tp, ap->ip,
2961                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2962                                         XFS_TRANS_DQ_BCOUNT,
2963                         (long) args.len);
2964         } else {
2965                 ap->rval = NULLFSBLOCK;
2966                 ap->alen = 0;
2967         }
2968         return 0;
2969 }
2970
2971 /*
2972  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2973  * It figures out where to ask the underlying allocator to put the new extent.
2974  */
2975 STATIC int
2976 xfs_bmap_alloc(
2977         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2978 {
2979         if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
2980                 return xfs_bmap_rtalloc(ap);
2981         return xfs_bmap_btalloc(ap);
2982 }
2983
2984 /*
2985  * Transform a btree format file with only one leaf node, where the
2986  * extents list will fit in the inode, into an extents format file.
2987  * Since the file extents are already in-core, all we have to do is
2988  * give up the space for the btree root and pitch the leaf block.
2989  */
2990 STATIC int                              /* error */
2991 xfs_bmap_btree_to_extents(
2992         xfs_trans_t             *tp,    /* transaction pointer */
2993         xfs_inode_t             *ip,    /* incore inode pointer */
2994         xfs_btree_cur_t         *cur,   /* btree cursor */
2995         int                     *logflagsp, /* inode logging flags */
2996         int                     whichfork)  /* data or attr fork */
2997 {
2998         /* REFERENCED */
2999         xfs_bmbt_block_t        *cblock;/* child btree block */
3000         xfs_fsblock_t           cbno;   /* child block number */
3001         xfs_buf_t               *cbp;   /* child block's buffer */
3002         int                     error;  /* error return value */
3003         xfs_ifork_t             *ifp;   /* inode fork data */
3004         xfs_mount_t             *mp;    /* mount point structure */
3005         __be64                  *pp;    /* ptr to block address */
3006         xfs_bmbt_block_t        *rblock;/* root btree block */
3007
3008         ifp = XFS_IFORK_PTR(ip, whichfork);
3009         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3010         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
3011         rblock = ifp->if_broot;
3012         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
3013         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
3014         ASSERT(XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes) == 1);
3015         mp = ip->i_mount;
3016         pp = XFS_BMAP_BROOT_PTR_ADDR(rblock, 1, ifp->if_broot_bytes);
3017         cbno = be64_to_cpu(*pp);
3018         *logflagsp = 0;
3019 #ifdef DEBUG
3020         if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
3021                 return error;
3022 #endif
3023         if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
3024                         XFS_BMAP_BTREE_REF)))
3025                 return error;
3026         cblock = XFS_BUF_TO_BMBT_BLOCK(cbp);
3027         if ((error = xfs_btree_check_lblock(cur, cblock, 0, cbp)))
3028                 return error;
3029         xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
3030         ip->i_d.di_nblocks--;
3031         XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
3032         xfs_trans_binval(tp, cbp);
3033         if (cur->bc_bufs[0] == cbp)
3034                 cur->bc_bufs[0] = NULL;
3035         xfs_iroot_realloc(ip, -1, whichfork);
3036         ASSERT(ifp->if_broot == NULL);
3037         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
3038         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3039         *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
3040         return 0;
3041 }
3042
3043 /*
3044  * Called by xfs_bmapi to update file extent records and the btree
3045  * after removing space (or undoing a delayed allocation).
3046  */
3047 STATIC int                              /* error */
3048 xfs_bmap_del_extent(
3049         xfs_inode_t             *ip,    /* incore inode pointer */
3050         xfs_trans_t             *tp,    /* current transaction pointer */
3051         xfs_extnum_t            idx,    /* extent number to update/delete */
3052         xfs_bmap_free_t         *flist, /* list of extents to be freed */
3053         xfs_btree_cur_t         *cur,   /* if null, not a btree */
3054         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
3055         int                     *logflagsp, /* inode logging flags */
3056         xfs_extdelta_t          *delta, /* Change made to incore extents */
3057         int                     whichfork, /* data or attr fork */
3058         int                     rsvd)   /* OK to allocate reserved blocks */
3059 {
3060         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
3061         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
3062         xfs_fsblock_t           del_endblock=0; /* first block past del */
3063         xfs_fileoff_t           del_endoff;     /* first offset past del */
3064         int                     delay;  /* current block is delayed allocated */
3065         int                     do_fx;  /* free extent at end of routine */
3066         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
3067         int                     error;  /* error return value */
3068         int                     flags;  /* inode logging flags */
3069         xfs_bmbt_irec_t         got;    /* current extent entry */
3070         xfs_fileoff_t           got_endoff;     /* first offset past got */
3071         int                     i;      /* temp state */
3072         xfs_ifork_t             *ifp;   /* inode fork pointer */
3073         xfs_mount_t             *mp;    /* mount structure */
3074         xfs_filblks_t           nblks;  /* quota/sb block count */
3075         xfs_bmbt_irec_t         new;    /* new record to be inserted */
3076         /* REFERENCED */
3077         uint                    qfield; /* quota field to update */
3078         xfs_filblks_t           temp;   /* for indirect length calculations */
3079         xfs_filblks_t           temp2;  /* for indirect length calculations */
3080
3081         XFS_STATS_INC(xs_del_exlist);
3082         mp = ip->i_mount;
3083         ifp = XFS_IFORK_PTR(ip, whichfork);
3084         ASSERT((idx >= 0) && (idx < ifp->if_bytes /
3085                 (uint)sizeof(xfs_bmbt_rec_t)));
3086         ASSERT(del->br_blockcount > 0);
3087         ep = xfs_iext_get_ext(ifp, idx);
3088         xfs_bmbt_get_all(ep, &got);
3089         ASSERT(got.br_startoff <= del->br_startoff);
3090         del_endoff = del->br_startoff + del->br_blockcount;
3091         got_endoff = got.br_startoff + got.br_blockcount;
3092         ASSERT(got_endoff >= del_endoff);
3093         delay = ISNULLSTARTBLOCK(got.br_startblock);
3094         ASSERT(ISNULLSTARTBLOCK(del->br_startblock) == delay);
3095         flags = 0;
3096         qfield = 0;
3097         error = 0;
3098         /*
3099          * If deleting a real allocation, must free up the disk space.
3100          */
3101         if (!delay) {
3102                 flags = XFS_ILOG_CORE;
3103                 /*
3104                  * Realtime allocation.  Free it and record di_nblocks update.
3105                  */
3106                 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
3107                         xfs_fsblock_t   bno;
3108                         xfs_filblks_t   len;
3109
3110                         ASSERT(do_mod(del->br_blockcount,
3111                                       mp->m_sb.sb_rextsize) == 0);
3112                         ASSERT(do_mod(del->br_startblock,
3113                                       mp->m_sb.sb_rextsize) == 0);
3114                         bno = del->br_startblock;
3115                         len = del->br_blockcount;
3116                         do_div(bno, mp->m_sb.sb_rextsize);
3117                         do_div(len, mp->m_sb.sb_rextsize);
3118                         if ((error = xfs_rtfree_extent(ip->i_transp, bno,
3119                                         (xfs_extlen_t)len)))
3120                                 goto done;
3121                         do_fx = 0;
3122                         nblks = len * mp->m_sb.sb_rextsize;
3123                         qfield = XFS_TRANS_DQ_RTBCOUNT;
3124                 }
3125                 /*
3126                  * Ordinary allocation.
3127                  */
3128                 else {
3129                         do_fx = 1;
3130                         nblks = del->br_blockcount;
3131                         qfield = XFS_TRANS_DQ_BCOUNT;
3132                 }
3133                 /*
3134                  * Set up del_endblock and cur for later.
3135                  */
3136                 del_endblock = del->br_startblock + del->br_blockcount;
3137                 if (cur) {
3138                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
3139                                         got.br_startblock, got.br_blockcount,
3140                                         &i)))
3141                                 goto done;
3142                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3143                 }
3144                 da_old = da_new = 0;
3145         } else {
3146                 da_old = STARTBLOCKVAL(got.br_startblock);
3147                 da_new = 0;
3148                 nblks = 0;
3149                 do_fx = 0;
3150         }
3151         /*
3152          * Set flag value to use in switch statement.
3153          * Left-contig is 2, right-contig is 1.
3154          */
3155         switch (((got.br_startoff == del->br_startoff) << 1) |
3156                 (got_endoff == del_endoff)) {
3157         case 3:
3158                 /*
3159                  * Matches the whole extent.  Delete the entry.
3160                  */
3161                 XFS_BMAP_TRACE_DELETE("3", ip, idx, 1, whichfork);
3162                 xfs_iext_remove(ifp, idx, 1);
3163                 ifp->if_lastex = idx;
3164                 if (delay)
3165                         break;
3166                 XFS_IFORK_NEXT_SET(ip, whichfork,
3167                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
3168                 flags |= XFS_ILOG_CORE;
3169                 if (!cur) {
3170                         flags |= XFS_ILOG_FEXT(whichfork);
3171                         break;
3172                 }
3173                 if ((error = xfs_bmbt_delete(cur, &i)))
3174                         goto done;
3175                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3176                 break;
3177
3178         case 2:
3179                 /*
3180                  * Deleting the first part of the extent.
3181                  */
3182                 XFS_BMAP_TRACE_PRE_UPDATE("2", ip, idx, whichfork);
3183                 xfs_bmbt_set_startoff(ep, del_endoff);
3184                 temp = got.br_blockcount - del->br_blockcount;
3185                 xfs_bmbt_set_blockcount(ep, temp);
3186                 ifp->if_lastex = idx;
3187                 if (delay) {
3188                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
3189                                 da_old);
3190                         xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3191                         XFS_BMAP_TRACE_POST_UPDATE("2", ip, idx,
3192                                 whichfork);
3193                         da_new = temp;
3194                         break;
3195                 }
3196                 xfs_bmbt_set_startblock(ep, del_endblock);
3197                 XFS_BMAP_TRACE_POST_UPDATE("2", ip, idx, whichfork);
3198                 if (!cur) {
3199                         flags |= XFS_ILOG_FEXT(whichfork);
3200                         break;
3201                 }
3202                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
3203                                 got.br_blockcount - del->br_blockcount,
3204                                 got.br_state)))
3205                         goto done;
3206                 break;
3207
3208         case 1:
3209                 /*
3210                  * Deleting the last part of the extent.
3211                  */
3212                 temp = got.br_blockcount - del->br_blockcount;
3213                 XFS_BMAP_TRACE_PRE_UPDATE("1", ip, idx, whichfork);
3214                 xfs_bmbt_set_blockcount(ep, temp);
3215                 ifp->if_lastex = idx;
3216                 if (delay) {
3217                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
3218                                 da_old);
3219                         xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3220                         XFS_BMAP_TRACE_POST_UPDATE("1", ip, idx,
3221                                 whichfork);
3222                         da_new = temp;
3223                         break;
3224                 }
3225                 XFS_BMAP_TRACE_POST_UPDATE("1", ip, idx, whichfork);
3226                 if (!cur) {
3227                         flags |= XFS_ILOG_FEXT(whichfork);
3228                         break;
3229                 }
3230                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
3231                                 got.br_startblock,
3232                                 got.br_blockcount - del->br_blockcount,
3233                                 got.br_state)))
3234                         goto done;
3235                 break;
3236
3237         case 0:
3238                 /*
3239                  * Deleting the middle of the extent.
3240                  */
3241                 temp = del->br_startoff - got.br_startoff;
3242                 XFS_BMAP_TRACE_PRE_UPDATE("0", ip, idx, whichfork);
3243                 xfs_bmbt_set_blockcount(ep, temp);
3244                 new.br_startoff = del_endoff;
3245                 temp2 = got_endoff - del_endoff;
3246                 new.br_blockcount = temp2;
3247                 new.br_state = got.br_state;
3248                 if (!delay) {
3249                         new.br_startblock = del_endblock;
3250                         flags |= XFS_ILOG_CORE;
3251                         if (cur) {
3252                                 if ((error = xfs_bmbt_update(cur,
3253                                                 got.br_startoff,
3254                                                 got.br_startblock, temp,
3255                                                 got.br_state)))
3256                                         goto done;
3257                                 if ((error = xfs_bmbt_increment(cur, 0, &i)))
3258                                         goto done;
3259                                 cur->bc_rec.b = new;
3260                                 error = xfs_bmbt_insert(cur, &i);
3261                                 if (error && error != ENOSPC)
3262                                         goto done;
3263                                 /*
3264                                  * If get no-space back from btree insert,
3265                                  * it tried a split, and we have a zero
3266                                  * block reservation.
3267                                  * Fix up our state and return the error.
3268                                  */
3269                                 if (error == ENOSPC) {
3270                                         /*
3271                                          * Reset the cursor, don't trust
3272                                          * it after any insert operation.
3273                                          */
3274                                         if ((error = xfs_bmbt_lookup_eq(cur,
3275                                                         got.br_startoff,
3276                                                         got.br_startblock,
3277                                                         temp, &i)))
3278                                                 goto done;
3279                                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3280                                         /*
3281                                          * Update the btree record back
3282                                          * to the original value.
3283                                          */
3284                                         if ((error = xfs_bmbt_update(cur,
3285                                                         got.br_startoff,
3286                                                         got.br_startblock,
3287                                                         got.br_blockcount,
3288                                                         got.br_state)))
3289                                                 goto done;
3290                                         /*
3291                                          * Reset the extent record back
3292                                          * to the original value.
3293                                          */
3294                                         xfs_bmbt_set_blockcount(ep,
3295                                                 got.br_blockcount);
3296                                         flags = 0;
3297                                         error = XFS_ERROR(ENOSPC);
3298                                         goto done;
3299                                 }
3300                                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3301                         } else
3302                                 flags |= XFS_ILOG_FEXT(whichfork);
3303                         XFS_IFORK_NEXT_SET(ip, whichfork,
3304                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
3305                 } else {
3306                         ASSERT(whichfork == XFS_DATA_FORK);
3307                         temp = xfs_bmap_worst_indlen(ip, temp);
3308                         xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3309                         temp2 = xfs_bmap_worst_indlen(ip, temp2);
3310                         new.br_startblock = NULLSTARTBLOCK((int)temp2);
3311                         da_new = temp + temp2;
3312                         while (da_new > da_old) {
3313                                 if (temp) {
3314                                         temp--;
3315                                         da_new--;
3316                                         xfs_bmbt_set_startblock(ep,
3317                                                 NULLSTARTBLOCK((int)temp));
3318                                 }
3319                                 if (da_new == da_old)
3320                                         break;
3321                                 if (temp2) {
3322                                         temp2--;
3323                                         da_new--;
3324                                         new.br_startblock =
3325                                                 NULLSTARTBLOCK((int)temp2);
3326                                 }
3327                         }
3328                 }
3329                 XFS_BMAP_TRACE_POST_UPDATE("0", ip, idx, whichfork);
3330                 XFS_BMAP_TRACE_INSERT("0", ip, idx + 1, 1, &new, NULL,
3331                         whichfork);
3332                 xfs_iext_insert(ifp, idx + 1, 1, &new);
3333                 ifp->if_lastex = idx + 1;
3334                 break;
3335         }
3336         /*
3337          * If we need to, add to list of extents to delete.
3338          */
3339         if (do_fx)
3340                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
3341                         mp);
3342         /*
3343          * Adjust inode # blocks in the file.
3344          */
3345         if (nblks)
3346                 ip->i_d.di_nblocks -= nblks;
3347         /*
3348          * Adjust quota data.
3349          */
3350         if (qfield)
3351                 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, qfield, (long)-nblks);
3352
3353         /*
3354          * Account for change in delayed indirect blocks.
3355          * Nothing to do for disk quota accounting here.
3356          */
3357         ASSERT(da_old >= da_new);
3358         if (da_old > da_new)
3359                 xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS, (int64_t)(da_old - da_new),
3360                         rsvd);
3361         if (delta) {
3362                 /* DELTA: report the original extent. */
3363                 if (delta->xed_startoff > got.br_startoff)
3364                         delta->xed_startoff = got.br_startoff;
3365                 if (delta->xed_blockcount < got.br_startoff+got.br_blockcount)
3366                         delta->xed_blockcount = got.br_startoff +
3367                                                         got.br_blockcount;
3368         }
3369 done:
3370         *logflagsp = flags;
3371         return error;
3372 }
3373
3374 /*
3375  * Remove the entry "free" from the free item list.  Prev points to the
3376  * previous entry, unless "free" is the head of the list.
3377  */
3378 STATIC void
3379 xfs_bmap_del_free(
3380         xfs_bmap_free_t         *flist, /* free item list header */
3381         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
3382         xfs_bmap_free_item_t    *free)  /* list item to be freed */
3383 {
3384         if (prev)
3385                 prev->xbfi_next = free->xbfi_next;
3386         else
3387                 flist->xbf_first = free->xbfi_next;
3388         flist->xbf_count--;
3389         kmem_zone_free(xfs_bmap_free_item_zone, free);
3390 }
3391
3392 /*
3393  * Convert an extents-format file into a btree-format file.
3394  * The new file will have a root block (in the inode) and a single child block.
3395  */
3396 STATIC int                                      /* error */
3397 xfs_bmap_extents_to_btree(
3398         xfs_trans_t             *tp,            /* transaction pointer */
3399         xfs_inode_t             *ip,            /* incore inode pointer */
3400         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
3401         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
3402         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
3403         int                     wasdel,         /* converting a delayed alloc */
3404         int                     *logflagsp,     /* inode logging flags */
3405         int                     whichfork)      /* data or attr fork */
3406 {
3407         xfs_bmbt_block_t        *ablock;        /* allocated (child) bt block */
3408         xfs_buf_t               *abp;           /* buffer for ablock */
3409         xfs_alloc_arg_t         args;           /* allocation arguments */
3410         xfs_bmbt_rec_t          *arp;           /* child record pointer */
3411         xfs_bmbt_block_t        *block;         /* btree root block */
3412         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
3413         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
3414         int                     error;          /* error return value */
3415         xfs_extnum_t            i, cnt;         /* extent record index */
3416         xfs_ifork_t             *ifp;           /* inode fork pointer */
3417         xfs_bmbt_key_t          *kp;            /* root block key pointer */
3418         xfs_mount_t             *mp;            /* mount structure */
3419         xfs_extnum_t            nextents;       /* number of file extents */
3420         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
3421
3422         ifp = XFS_IFORK_PTR(ip, whichfork);
3423         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3424         ASSERT(ifp->if_ext_max ==
3425                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
3426         /*
3427          * Make space in the inode incore.
3428          */
3429         xfs_iroot_realloc(ip, 1, whichfork);
3430         ifp->if_flags |= XFS_IFBROOT;
3431         /*
3432          * Fill in the root.
3433          */
3434         block = ifp->if_broot;
3435         block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3436         block->bb_level = cpu_to_be16(1);
3437         block->bb_numrecs = cpu_to_be16(1);
3438         block->bb_leftsib = cpu_to_be64(NULLDFSBNO);
3439         block->bb_rightsib = cpu_to_be64(NULLDFSBNO);
3440         /*
3441          * Need a cursor.  Can't allocate until bb_level is filled in.
3442          */
3443         mp = ip->i_mount;
3444         cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
3445                 whichfork);
3446         cur->bc_private.b.firstblock = *firstblock;
3447         cur->bc_private.b.flist = flist;
3448         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3449         /*
3450          * Convert to a btree with two levels, one record in root.
3451          */
3452         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3453         args.tp = tp;
3454         args.mp = mp;
3455         args.firstblock = *firstblock;
3456         if (*firstblock == NULLFSBLOCK) {
3457                 args.type = XFS_ALLOCTYPE_START_BNO;
3458                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3459         } else if (flist->xbf_low) {
3460                 args.type = XFS_ALLOCTYPE_START_BNO;
3461                 args.fsbno = *firstblock;
3462         } else {
3463                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3464                 args.fsbno = *firstblock;
3465         }
3466         args.minlen = args.maxlen = args.prod = 1;
3467         args.total = args.minleft = args.alignment = args.mod = args.isfl =
3468                 args.minalignslop = 0;
3469         args.wasdel = wasdel;
3470         *logflagsp = 0;
3471         if ((error = xfs_alloc_vextent(&args))) {
3472                 xfs_iroot_realloc(ip, -1, whichfork);
3473                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3474                 return error;
3475         }
3476         /*
3477          * Allocation can't fail, the space was reserved.
3478          */
3479         ASSERT(args.fsbno != NULLFSBLOCK);
3480         ASSERT(*firstblock == NULLFSBLOCK ||
3481                args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3482                (flist->xbf_low &&
3483                 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3484         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3485         cur->bc_private.b.allocated++;
3486         ip->i_d.di_nblocks++;
3487         XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3488         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3489         /*
3490          * Fill in the child block.
3491          */
3492         ablock = XFS_BUF_TO_BMBT_BLOCK(abp);
3493         ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3494         ablock->bb_level = 0;
3495         ablock->bb_leftsib = cpu_to_be64(NULLDFSBNO);
3496         ablock->bb_rightsib = cpu_to_be64(NULLDFSBNO);
3497         arp = XFS_BMAP_REC_IADDR(ablock, 1, cur);
3498         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3499         for (cnt = i = 0; i < nextents; i++) {
3500                 ep = xfs_iext_get_ext(ifp, i);
3501                 if (!ISNULLSTARTBLOCK(xfs_bmbt_get_startblock(ep))) {
3502                         arp->l0 = cpu_to_be64(ep->l0);
3503                         arp->l1 = cpu_to_be64(ep->l1);
3504                         arp++; cnt++;
3505                 }
3506         }
3507         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3508         ablock->bb_numrecs = cpu_to_be16(cnt);
3509         /*
3510          * Fill in the root key and pointer.
3511          */
3512         kp = XFS_BMAP_KEY_IADDR(block, 1, cur);
3513         arp = XFS_BMAP_REC_IADDR(ablock, 1, cur);
3514         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
3515         pp = XFS_BMAP_PTR_IADDR(block, 1, cur);
3516         *pp = cpu_to_be64(args.fsbno);
3517         /*
3518          * Do all this logging at the end so that
3519          * the root is at the right level.
3520          */
3521         xfs_bmbt_log_block(cur, abp, XFS_BB_ALL_BITS);
3522         xfs_bmbt_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3523         ASSERT(*curp == NULL);
3524         *curp = cur;
3525         *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FBROOT(whichfork);
3526         return 0;
3527 }
3528
3529 /*
3530  * Helper routine to reset inode di_forkoff field when switching
3531  * attribute fork from local to extent format - we reset it where
3532  * possible to make space available for inline data fork extents.
3533  */
3534 STATIC void
3535 xfs_bmap_forkoff_reset(
3536         xfs_mount_t     *mp,
3537         xfs_inode_t     *ip,
3538         int             whichfork)
3539 {
3540         if (whichfork == XFS_ATTR_FORK &&
3541             (ip->i_d.di_format != XFS_DINODE_FMT_DEV) &&
3542             (ip->i_d.di_format != XFS_DINODE_FMT_UUID) &&
3543             (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3544             ((mp->m_attroffset >> 3) > ip->i_d.di_forkoff)) {
3545                 ip->i_d.di_forkoff = mp->m_attroffset >> 3;
3546                 ip->i_df.if_ext_max = XFS_IFORK_DSIZE(ip) /
3547                                         (uint)sizeof(xfs_bmbt_rec_t);
3548                 ip->i_afp->if_ext_max = XFS_IFORK_ASIZE(ip) /
3549                                         (uint)sizeof(xfs_bmbt_rec_t);
3550         }
3551 }
3552
3553 /*
3554  * Convert a local file to an extents file.
3555  * This code is out of bounds for data forks of regular files,
3556  * since the file data needs to get logged so things will stay consistent.
3557  * (The bmap-level manipulations are ok, though).
3558  */
3559 STATIC int                              /* error */
3560 xfs_bmap_local_to_extents(
3561         xfs_trans_t     *tp,            /* transaction pointer */
3562         xfs_inode_t     *ip,            /* incore inode pointer */
3563         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
3564         xfs_extlen_t    total,          /* total blocks needed by transaction */
3565         int             *logflagsp,     /* inode logging flags */
3566         int             whichfork)      /* data or attr fork */
3567 {
3568         int             error;          /* error return value */
3569         int             flags;          /* logging flags returned */
3570         xfs_ifork_t     *ifp;           /* inode fork pointer */
3571
3572         /*
3573          * We don't want to deal with the case of keeping inode data inline yet.
3574          * So sending the data fork of a regular inode is invalid.
3575          */
3576         ASSERT(!((ip->i_d.di_mode & S_IFMT) == S_IFREG &&
3577                  whichfork == XFS_DATA_FORK));
3578         ifp = XFS_IFORK_PTR(ip, whichfork);
3579         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3580         flags = 0;
3581         error = 0;
3582         if (ifp->if_bytes) {
3583                 xfs_alloc_arg_t args;   /* allocation arguments */
3584                 xfs_buf_t       *bp;    /* buffer for extent block */
3585                 xfs_bmbt_rec_host_t *ep;/* extent record pointer */
3586
3587                 args.tp = tp;
3588                 args.mp = ip->i_mount;
3589                 args.firstblock = *firstblock;
3590                 ASSERT((ifp->if_flags &
3591                         (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == XFS_IFINLINE);
3592                 /*
3593                  * Allocate a block.  We know we need only one, since the
3594                  * file currently fits in an inode.
3595                  */
3596                 if (*firstblock == NULLFSBLOCK) {
3597                         args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3598                         args.type = XFS_ALLOCTYPE_START_BNO;
3599                 } else {
3600                         args.fsbno = *firstblock;
3601                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
3602                 }
3603                 args.total = total;
3604                 args.mod = args.minleft = args.alignment = args.wasdel =
3605                         args.isfl = args.minalignslop = 0;
3606                 args.minlen = args.maxlen = args.prod = 1;
3607                 if ((error = xfs_alloc_vextent(&args)))
3608                         goto done;
3609                 /*
3610                  * Can't fail, the space was reserved.
3611                  */
3612                 ASSERT(args.fsbno != NULLFSBLOCK);
3613                 ASSERT(args.len == 1);
3614                 *firstblock = args.fsbno;
3615                 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3616                 memcpy((char *)XFS_BUF_PTR(bp), ifp->if_u1.if_data,
3617                         ifp->if_bytes);
3618                 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3619                 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3620                 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3621                 xfs_iext_add(ifp, 0, 1);
3622                 ep = xfs_iext_get_ext(ifp, 0);
3623                 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3624                 XFS_BMAP_TRACE_POST_UPDATE("new", ip, 0, whichfork);
3625                 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3626                 ip->i_d.di_nblocks = 1;
3627                 XFS_TRANS_MOD_DQUOT_BYINO(args.mp, tp, ip,
3628                         XFS_TRANS_DQ_BCOUNT, 1L);
3629                 flags |= XFS_ILOG_FEXT(whichfork);
3630         } else {
3631                 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3632                 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3633         }
3634         ifp->if_flags &= ~XFS_IFINLINE;
3635         ifp->if_flags |= XFS_IFEXTENTS;
3636         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3637         flags |= XFS_ILOG_CORE;
3638 done:
3639         *logflagsp = flags;
3640         return error;
3641 }
3642
3643 /*
3644  * Search the extent records for the entry containing block bno.
3645  * If bno lies in a hole, point to the next entry.  If bno lies
3646  * past eof, *eofp will be set, and *prevp will contain the last
3647  * entry (null if none).  Else, *lastxp will be set to the index
3648  * of the found entry; *gotp will contain the entry.
3649  */
3650 xfs_bmbt_rec_host_t *                   /* pointer to found extent entry */
3651 xfs_bmap_search_multi_extents(
3652         xfs_ifork_t     *ifp,           /* inode fork pointer */
3653         xfs_fileoff_t   bno,            /* block number searched for */
3654         int             *eofp,          /* out: end of file found */
3655         xfs_extnum_t    *lastxp,        /* out: last extent index */
3656         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3657         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3658 {
3659         xfs_bmbt_rec_host_t *ep;                /* extent record pointer */
3660         xfs_extnum_t    lastx;          /* last extent index */
3661
3662         /*
3663          * Initialize the extent entry structure to catch access to
3664          * uninitialized br_startblock field.
3665          */
3666         gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
3667         gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3668         gotp->br_state = XFS_EXT_INVALID;
3669 #if XFS_BIG_BLKNOS
3670         gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
3671 #else
3672         gotp->br_startblock = 0xffffa5a5;
3673 #endif
3674         prevp->br_startoff = NULLFILEOFF;
3675
3676         ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
3677         if (lastx > 0) {
3678                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
3679         }
3680         if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
3681                 xfs_bmbt_get_all(ep, gotp);
3682                 *eofp = 0;
3683         } else {
3684                 if (lastx > 0) {
3685                         *gotp = *prevp;
3686                 }
3687                 *eofp = 1;
3688                 ep = NULL;
3689         }
3690         *lastxp = lastx;
3691         return ep;
3692 }
3693
3694 /*
3695  * Search the extents list for the inode, for the extent containing bno.
3696  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
3697  * *eofp will be set, and *prevp will contain the last entry (null if none).
3698  * Else, *lastxp will be set to the index of the found
3699  * entry; *gotp will contain the entry.
3700  */
3701 STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
3702 xfs_bmap_search_extents(
3703         xfs_inode_t     *ip,            /* incore inode pointer */
3704         xfs_fileoff_t   bno,            /* block number searched for */
3705         int             fork,           /* data or attr fork */
3706         int             *eofp,          /* out: end of file found */
3707         xfs_extnum_t    *lastxp,        /* out: last extent index */
3708         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3709         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3710 {
3711         xfs_ifork_t     *ifp;           /* inode fork pointer */
3712         xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
3713
3714         XFS_STATS_INC(xs_look_exlist);
3715         ifp = XFS_IFORK_PTR(ip, fork);
3716
3717         ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
3718
3719         if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
3720                      !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
3721                 xfs_cmn_err(XFS_PTAG_FSBLOCK_ZERO, CE_ALERT, ip->i_mount,
3722                                 "Access to block zero in inode %llu "
3723                                 "start_block: %llx start_off: %llx "
3724                                 "blkcnt: %llx extent-state: %x lastx: %x\n",
3725                         (unsigned long long)ip->i_ino,
3726                         (unsigned long long)gotp->br_startblock,
3727                         (unsigned long long)gotp->br_startoff,
3728                         (unsigned long long)gotp->br_blockcount,
3729                         gotp->br_state, *lastxp);
3730                 *lastxp = NULLEXTNUM;
3731                 *eofp = 1;
3732                 return NULL;
3733         }
3734         return ep;
3735 }
3736
3737
3738 #ifdef XFS_BMAP_TRACE
3739 ktrace_t        *xfs_bmap_trace_buf;
3740
3741 /*
3742  * Add a bmap trace buffer entry.  Base routine for the others.
3743  */
3744 STATIC void
3745 xfs_bmap_trace_addentry(
3746         int             opcode,         /* operation */
3747         const char      *fname,         /* function name */
3748         char            *desc,          /* operation description */
3749         xfs_inode_t     *ip,            /* incore inode pointer */
3750         xfs_extnum_t    idx,            /* index of entry(ies) */
3751         xfs_extnum_t    cnt,            /* count of entries, 1 or 2 */
3752         xfs_bmbt_rec_host_t *r1,        /* first record */
3753         xfs_bmbt_rec_host_t *r2,        /* second record or null */
3754         int             whichfork)      /* data or attr fork */
3755 {
3756         xfs_bmbt_rec_host_t tr2;
3757
3758         ASSERT(cnt == 1 || cnt == 2);
3759         ASSERT(r1 != NULL);
3760         if (cnt == 1) {
3761                 ASSERT(r2 == NULL);
3762                 r2 = &tr2;
3763                 memset(&tr2, 0, sizeof(tr2));
3764         } else
3765                 ASSERT(r2 != NULL);
3766         ktrace_enter(xfs_bmap_trace_buf,
3767                 (void *)(__psint_t)(opcode | (whichfork << 16)),
3768                 (void *)fname, (void *)desc, (void *)ip,
3769                 (void *)(__psint_t)idx,
3770                 (void *)(__psint_t)cnt,
3771                 (void *)(__psunsigned_t)(ip->i_ino >> 32),
3772                 (void *)(__psunsigned_t)(unsigned)ip->i_ino,
3773                 (void *)(__psunsigned_t)(r1->l0 >> 32),
3774                 (void *)(__psunsigned_t)(unsigned)(r1->l0),
3775                 (void *)(__psunsigned_t)(r1->l1 >> 32),
3776                 (void *)(__psunsigned_t)(unsigned)(r1->l1),
3777                 (void *)(__psunsigned_t)(r2->l0 >> 32),
3778                 (void *)(__psunsigned_t)(unsigned)(r2->l0),
3779                 (void *)(__psunsigned_t)(r2->l1 >> 32),
3780                 (void *)(__psunsigned_t)(unsigned)(r2->l1)
3781                 );
3782         ASSERT(ip->i_xtrace);
3783         ktrace_enter(ip->i_xtrace,
3784                 (void *)(__psint_t)(opcode | (whichfork << 16)),
3785                 (void *)fname, (void *)desc, (void *)ip,
3786                 (void *)(__psint_t)idx,
3787                 (void *)(__psint_t)cnt,
3788                 (void *)(__psunsigned_t)(ip->i_ino >> 32),
3789                 (void *)(__psunsigned_t)(unsigned)ip->i_ino,
3790                 (void *)(__psunsigned_t)(r1->l0 >> 32),
3791                 (void *)(__psunsigned_t)(unsigned)(r1->l0),
3792                 (void *)(__psunsigned_t)(r1->l1 >> 32),
3793                 (void *)(__psunsigned_t)(unsigned)(r1->l1),
3794                 (void *)(__psunsigned_t)(r2->l0 >> 32),
3795                 (void *)(__psunsigned_t)(unsigned)(r2->l0),
3796                 (void *)(__psunsigned_t)(r2->l1 >> 32),
3797                 (void *)(__psunsigned_t)(unsigned)(r2->l1)
3798                 );
3799 }
3800
3801 /*
3802  * Add bmap trace entry prior to a call to xfs_iext_remove.
3803  */
3804 STATIC void
3805 xfs_bmap_trace_delete(
3806         const char      *fname,         /* function name */
3807         char            *desc,          /* operation description */
3808         xfs_inode_t     *ip,            /* incore inode pointer */
3809         xfs_extnum_t    idx,            /* index of entry(entries) deleted */
3810         xfs_extnum_t    cnt,            /* count of entries deleted, 1 or 2 */
3811         int             whichfork)      /* data or attr fork */
3812 {
3813         xfs_ifork_t     *ifp;           /* inode fork pointer */
3814
3815         ifp = XFS_IFORK_PTR(ip, whichfork);
3816         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_DELETE, fname, desc, ip, idx,
3817                 cnt, xfs_iext_get_ext(ifp, idx),
3818                 cnt == 2 ? xfs_iext_get_ext(ifp, idx + 1) : NULL,
3819                 whichfork);
3820 }
3821
3822 /*
3823  * Add bmap trace entry prior to a call to xfs_iext_insert, or
3824  * reading in the extents list from the disk (in the btree).
3825  */
3826 STATIC void
3827 xfs_bmap_trace_insert(
3828         const char      *fname,         /* function name */
3829         char            *desc,          /* operation description */
3830         xfs_inode_t     *ip,            /* incore inode pointer */
3831         xfs_extnum_t    idx,            /* index of entry(entries) inserted */
3832         xfs_extnum_t    cnt,            /* count of entries inserted, 1 or 2 */
3833         xfs_bmbt_irec_t *r1,            /* inserted record 1 */
3834         xfs_bmbt_irec_t *r2,            /* inserted record 2 or null */
3835         int             whichfork)      /* data or attr fork */
3836 {
3837         xfs_bmbt_rec_host_t tr1;        /* compressed record 1 */
3838         xfs_bmbt_rec_host_t tr2;        /* compressed record 2 if needed */
3839
3840         xfs_bmbt_set_all(&tr1, r1);
3841         if (cnt == 2) {
3842                 ASSERT(r2 != NULL);
3843                 xfs_bmbt_set_all(&tr2, r2);
3844         } else {
3845                 ASSERT(cnt == 1);
3846                 ASSERT(r2 == NULL);
3847         }
3848         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_INSERT, fname, desc, ip, idx,
3849                 cnt, &tr1, cnt == 2 ? &tr2 : NULL, whichfork);
3850 }
3851
3852 /*
3853  * Add bmap trace entry after updating an extent record in place.
3854  */
3855 STATIC void
3856 xfs_bmap_trace_post_update(
3857         const char      *fname,         /* function name */
3858         char            *desc,          /* operation description */
3859         xfs_inode_t     *ip,            /* incore inode pointer */
3860         xfs_extnum_t    idx,            /* index of entry updated */
3861         int             whichfork)      /* data or attr fork */
3862 {
3863         xfs_ifork_t     *ifp;           /* inode fork pointer */
3864
3865         ifp = XFS_IFORK_PTR(ip, whichfork);
3866         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_POST_UP, fname, desc, ip, idx,
3867                 1, xfs_iext_get_ext(ifp, idx), NULL, whichfork);
3868 }
3869
3870 /*
3871  * Add bmap trace entry prior to updating an extent record in place.
3872  */
3873 STATIC void
3874 xfs_bmap_trace_pre_update(
3875         const char      *fname,         /* function name */
3876         char            *desc,          /* operation description */
3877         xfs_inode_t     *ip,            /* incore inode pointer */
3878         xfs_extnum_t    idx,            /* index of entry to be updated */
3879         int             whichfork)      /* data or attr fork */
3880 {
3881         xfs_ifork_t     *ifp;           /* inode fork pointer */
3882
3883         ifp = XFS_IFORK_PTR(ip, whichfork);
3884         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_PRE_UP, fname, desc, ip, idx, 1,
3885                 xfs_iext_get_ext(ifp, idx), NULL, whichfork);
3886 }
3887 #endif  /* XFS_BMAP_TRACE */
3888
3889 /*
3890  * Compute the worst-case number of indirect blocks that will be used
3891  * for ip's delayed extent of length "len".
3892  */
3893 STATIC xfs_filblks_t
3894 xfs_bmap_worst_indlen(
3895         xfs_inode_t     *ip,            /* incore inode pointer */
3896         xfs_filblks_t   len)            /* delayed extent length */
3897 {
3898         int             level;          /* btree level number */
3899         int             maxrecs;        /* maximum record count at this level */
3900         xfs_mount_t     *mp;            /* mount structure */
3901         xfs_filblks_t   rval;           /* return value */
3902
3903         mp = ip->i_mount;
3904         maxrecs = mp->m_bmap_dmxr[0];
3905         for (level = 0, rval = 0;
3906              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3907              level++) {
3908                 len += maxrecs - 1;
3909                 do_div(len, maxrecs);
3910                 rval += len;
3911                 if (len == 1)
3912                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3913                                 level - 1;
3914                 if (level == 0)
3915                         maxrecs = mp->m_bmap_dmxr[1];
3916         }
3917         return rval;
3918 }
3919
3920 #if defined(XFS_RW_TRACE)
3921 STATIC void
3922 xfs_bunmap_trace(
3923         xfs_inode_t             *ip,
3924         xfs_fileoff_t           bno,
3925         xfs_filblks_t           len,
3926         int                     flags,
3927         inst_t                  *ra)
3928 {
3929         if (ip->i_rwtrace == NULL)
3930                 return;
3931         ktrace_enter(ip->i_rwtrace,
3932                 (void *)(__psint_t)XFS_BUNMAP,
3933                 (void *)ip,
3934                 (void *)(__psint_t)((ip->i_d.di_size >> 32) & 0xffffffff),
3935                 (void *)(__psint_t)(ip->i_d.di_size & 0xffffffff),
3936                 (void *)(__psint_t)(((xfs_dfiloff_t)bno >> 32) & 0xffffffff),
3937                 (void *)(__psint_t)((xfs_dfiloff_t)bno & 0xffffffff),
3938                 (void *)(__psint_t)len,
3939                 (void *)(__psint_t)flags,
3940                 (void *)(unsigned long)current_cpu(),
3941                 (void *)ra,
3942                 (void *)0,
3943                 (void *)0,
3944                 (void *)0,
3945                 (void *)0,
3946                 (void *)0,
3947                 (void *)0);
3948 }
3949 #endif
3950
3951 /*
3952  * Convert inode from non-attributed to attributed.
3953  * Must not be in a transaction, ip must not be locked.
3954  */
3955 int                                             /* error code */
3956 xfs_bmap_add_attrfork(
3957         xfs_inode_t             *ip,            /* incore inode pointer */
3958         int                     size,           /* space new attribute needs */
3959         int                     rsvd)           /* xact may use reserved blks */
3960 {
3961         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
3962         xfs_bmap_free_t         flist;          /* freed extent records */
3963         xfs_mount_t             *mp;            /* mount structure */
3964         xfs_trans_t             *tp;            /* transaction pointer */
3965         int                     blks;           /* space reservation */
3966         int                     version = 1;    /* superblock attr version */
3967         int                     committed;      /* xaction was committed */
3968         int                     logflags;       /* logging flags */
3969         int                     error;          /* error return value */
3970
3971         ASSERT(XFS_IFORK_Q(ip) == 0);
3972         ASSERT(ip->i_df.if_ext_max ==
3973                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3974
3975         mp = ip->i_mount;
3976         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3977         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3978         blks = XFS_ADDAFORK_SPACE_RES(mp);
3979         if (rsvd)
3980                 tp->t_flags |= XFS_TRANS_RESERVE;
3981         if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3982                         XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3983                 goto error0;
3984         xfs_ilock(ip, XFS_ILOCK_EXCL);
3985         error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip, blks, 0, rsvd ?
3986                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3987                         XFS_QMOPT_RES_REGBLKS);
3988         if (error) {
3989                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3990                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3991                 return error;
3992         }
3993         if (XFS_IFORK_Q(ip))
3994                 goto error1;
3995         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3996                 /*
3997                  * For inodes coming from pre-6.2 filesystems.
3998                  */
3999                 ASSERT(ip->i_d.di_aformat == 0);
4000                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
4001         }
4002         ASSERT(ip->i_d.di_anextents == 0);
4003         IHOLD(ip);
4004         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4005         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4006         switch (ip->i_d.di_format) {
4007         case XFS_DINODE_FMT_DEV:
4008                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
4009                 break;
4010         case XFS_DINODE_FMT_UUID:
4011                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
4012                 break;
4013         case XFS_DINODE_FMT_LOCAL:
4014         case XFS_DINODE_FMT_EXTENTS:
4015         case XFS_DINODE_FMT_BTREE:
4016                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
4017                 if (!ip->i_d.di_forkoff)
4018                         ip->i_d.di_forkoff = mp->m_attroffset >> 3;
4019                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
4020                         version = 2;
4021                 break;
4022         default:
4023                 ASSERT(0);
4024                 error = XFS_ERROR(EINVAL);
4025                 goto error1;
4026         }
4027         ip->i_df.if_ext_max =
4028                 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
4029         ASSERT(ip->i_afp == NULL);
4030         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
4031         ip->i_afp->if_ext_max =
4032                 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
4033         ip->i_afp->if_flags = XFS_IFEXTENTS;
4034         logflags = 0;
4035         XFS_BMAP_INIT(&flist, &firstblock);
4036         switch (ip->i_d.di_format) {
4037         case XFS_DINODE_FMT_LOCAL:
4038                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
4039                         &logflags);
4040                 break;
4041         case XFS_DINODE_FMT_EXTENTS:
4042                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
4043                         &flist, &logflags);
4044                 break;
4045         case XFS_DINODE_FMT_BTREE:
4046                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
4047                         &logflags);
4048                 break;
4049         default:
4050                 error = 0;
4051                 break;
4052         }
4053         if (logflags)
4054                 xfs_trans_log_inode(tp, ip, logflags);
4055         if (error)
4056                 goto error2;
4057         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
4058            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
4059                 __int64_t sbfields = 0;
4060
4061                 spin_lock(&mp->m_sb_lock);
4062                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
4063                         xfs_sb_version_addattr(&mp->m_sb);
4064                         sbfields |= XFS_SB_VERSIONNUM;
4065                 }
4066                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
4067                         xfs_sb_version_addattr2(&mp->m_sb);
4068                         sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
4069                 }
4070                 if (sbfields) {
4071                         spin_unlock(&mp->m_sb_lock);
4072                         xfs_mod_sb(tp, sbfields);
4073                 } else
4074                         spin_unlock(&mp->m_sb_lock);
4075         }
4076         if ((error = xfs_bmap_finish(&tp, &flist, &committed)))
4077                 goto error2;
4078         error = xfs_trans_commit(tp, XFS_TRANS_PERM_LOG_RES);
4079         ASSERT(ip->i_df.if_ext_max ==
4080                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
4081         return error;
4082 error2:
4083         xfs_bmap_cancel(&flist);
4084 error1:
4085         xfs_iunlock(ip, XFS_ILOCK_EXCL);
4086 error0:
4087         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
4088         ASSERT(ip->i_df.if_ext_max ==
4089                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
4090         return error;
4091 }
4092
4093 /*
4094  * Add the extent to the list of extents to be free at transaction end.
4095  * The list is maintained sorted (by block number).
4096  */
4097 /* ARGSUSED */
4098 void
4099 xfs_bmap_add_free(
4100         xfs_fsblock_t           bno,            /* fs block number of extent */
4101         xfs_filblks_t           len,            /* length of extent */
4102         xfs_bmap_free_t         *flist,         /* list of extents */
4103         xfs_mount_t             *mp)            /* mount point structure */
4104 {
4105         xfs_bmap_free_item_t    *cur;           /* current (next) element */
4106         xfs_bmap_free_item_t    *new;           /* new element */
4107         xfs_bmap_free_item_t    *prev;          /* previous element */
4108 #ifdef DEBUG
4109         xfs_agnumber_t          agno;
4110         xfs_agblock_t           agbno;
4111
4112         ASSERT(bno != NULLFSBLOCK);
4113         ASSERT(len > 0);
4114         ASSERT(len <= MAXEXTLEN);
4115         ASSERT(!ISNULLSTARTBLOCK(bno));
4116         agno = XFS_FSB_TO_AGNO(mp, bno);
4117         agbno = XFS_FSB_TO_AGBNO(mp, bno);
4118         ASSERT(agno < mp->m_sb.sb_agcount);
4119         ASSERT(agbno < mp->m_sb.sb_agblocks);
4120         ASSERT(len < mp->m_sb.sb_agblocks);
4121         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
4122 #endif
4123         ASSERT(xfs_bmap_free_item_zone != NULL);
4124         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
4125         new->xbfi_startblock = bno;
4126         new->xbfi_blockcount = (xfs_extlen_t)len;
4127         for (prev = NULL, cur = flist->xbf_first;
4128              cur != NULL;
4129              prev = cur, cur = cur->xbfi_next) {
4130                 if (cur->xbfi_startblock >= bno)
4131                         break;
4132         }
4133         if (prev)
4134                 prev->xbfi_next = new;
4135         else
4136                 flist->xbf_first = new;
4137         new->xbfi_next = cur;
4138         flist->xbf_count++;
4139 }
4140
4141 /*
4142  * Compute and fill in the value of the maximum depth of a bmap btree
4143  * in this filesystem.  Done once, during mount.
4144  */
4145 void
4146 xfs_bmap_compute_maxlevels(
4147         xfs_mount_t     *mp,            /* file system mount structure */
4148         int             whichfork)      /* data or attr fork */
4149 {
4150         int             level;          /* btree level */
4151         uint            maxblocks;      /* max blocks at this level */
4152         uint            maxleafents;    /* max leaf entries possible */
4153         int             maxrootrecs;    /* max records in root block */
4154         int             minleafrecs;    /* min records in leaf block */
4155         int             minnoderecs;    /* min records in node block */
4156         int             sz;             /* root block size */
4157
4158         /*
4159          * The maximum number of extents in a file, hence the maximum
4160          * number of leaf entries, is controlled by the type of di_nextents
4161          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
4162          * (a signed 16-bit number, xfs_aextnum_t).
4163          *
4164          * Note that we can no longer assume that if we are in ATTR1 that
4165          * the fork offset of all the inodes will be (m_attroffset >> 3)
4166          * because we could have mounted with ATTR2 and then mounted back
4167          * with ATTR1, keeping the di_forkoff's fixed but probably at
4168          * various positions. Therefore, for both ATTR1 and ATTR2
4169          * we have to assume the worst case scenario of a minimum size
4170          * available.
4171          */
4172         if (whichfork == XFS_DATA_FORK) {
4173                 maxleafents = MAXEXTNUM;
4174                 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
4175         } else {
4176                 maxleafents = MAXAEXTNUM;
4177                 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
4178         }
4179         maxrootrecs = (int)XFS_BTREE_BLOCK_MAXRECS(sz, xfs_bmdr, 0);
4180         minleafrecs = mp->m_bmap_dmnr[0];
4181         minnoderecs = mp->m_bmap_dmnr[1];
4182         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
4183         for (level = 1; maxblocks > 1; level++) {
4184                 if (maxblocks <= maxrootrecs)
4185                         maxblocks = 1;
4186                 else
4187                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
4188         }
4189         mp->m_bm_maxlevels[whichfork] = level;
4190 }
4191
4192 /*
4193  * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
4194  * caller.  Frees all the extents that need freeing, which must be done
4195  * last due to locking considerations.  We never free any extents in
4196  * the first transaction.  This is to allow the caller to make the first
4197  * transaction a synchronous one so that the pointers to the data being
4198  * broken in this transaction will be permanent before the data is actually
4199  * freed.  This is necessary to prevent blocks from being reallocated
4200  * and written to before the free and reallocation are actually permanent.
4201  * We do not just make the first transaction synchronous here, because
4202  * there are more efficient ways to gain the same protection in some cases
4203  * (see the file truncation code).
4204  *
4205  * Return 1 if the given transaction was committed and a new one
4206  * started, and 0 otherwise in the committed parameter.
4207  */
4208 /*ARGSUSED*/
4209 int                                             /* error */
4210 xfs_bmap_finish(
4211         xfs_trans_t             **tp,           /* transaction pointer addr */
4212         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
4213         int                     *committed)     /* xact committed or not */
4214 {
4215         xfs_efd_log_item_t      *efd;           /* extent free data */
4216         xfs_efi_log_item_t      *efi;           /* extent free intention */
4217         int                     error;          /* error return value */
4218         xfs_bmap_free_item_t    *free;          /* free extent item */
4219         unsigned int            logres;         /* new log reservation */
4220         unsigned int            logcount;       /* new log count */
4221         xfs_mount_t             *mp;            /* filesystem mount structure */
4222         xfs_bmap_free_item_t    *next;          /* next item on free list */
4223         xfs_trans_t             *ntp;           /* new transaction pointer */
4224
4225         ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
4226         if (flist->xbf_count == 0) {
4227                 *committed = 0;
4228                 return 0;
4229         }
4230         ntp = *tp;
4231         efi = xfs_trans_get_efi(ntp, flist->xbf_count);
4232         for (free = flist->xbf_first; free; free = free->xbfi_next)
4233                 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
4234                         free->xbfi_blockcount);
4235         logres = ntp->t_log_res;
4236         logcount = ntp->t_log_count;
4237         ntp = xfs_trans_dup(*tp);
4238         error = xfs_trans_commit(*tp, 0);
4239         *tp = ntp;
4240         *committed = 1;
4241         /*
4242          * We have a new transaction, so we should return committed=1,
4243          * even though we're returning an error.
4244          */
4245         if (error) {
4246                 return error;
4247         }
4248         if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
4249                         logcount)))
4250                 return error;
4251         efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
4252         for (free = flist->xbf_first; free != NULL; free = next) {
4253                 next = free->xbfi_next;
4254                 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
4255                                 free->xbfi_blockcount))) {
4256                         /*
4257                          * The bmap free list will be cleaned up at a
4258                          * higher level.  The EFI will be canceled when
4259                          * this transaction is aborted.
4260                          * Need to force shutdown here to make sure it
4261                          * happens, since this transaction may not be
4262                          * dirty yet.
4263                          */
4264                         mp = ntp->t_mountp;
4265                         if (!XFS_FORCED_SHUTDOWN(mp))
4266                                 xfs_force_shutdown(mp,
4267                                                    (error == EFSCORRUPTED) ?
4268                                                    SHUTDOWN_CORRUPT_INCORE :
4269                                                    SHUTDOWN_META_IO_ERROR);
4270                         return error;
4271                 }
4272                 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
4273                         free->xbfi_blockcount);
4274                 xfs_bmap_del_free(flist, NULL, free);
4275         }
4276         return 0;
4277 }
4278
4279 /*
4280  * Free up any items left in the list.
4281  */
4282 void
4283 xfs_bmap_cancel(
4284         xfs_bmap_free_t         *flist) /* list of bmap_free_items */
4285 {
4286         xfs_bmap_free_item_t    *free;  /* free list item */
4287         xfs_bmap_free_item_t    *next;
4288
4289         if (flist->xbf_count == 0)
4290                 return;
4291         ASSERT(flist->xbf_first != NULL);
4292         for (free = flist->xbf_first; free; free = next) {
4293                 next = free->xbfi_next;
4294                 xfs_bmap_del_free(flist, NULL, free);
4295         }
4296         ASSERT(flist->xbf_count == 0);
4297 }
4298
4299 /*
4300  * Returns the file-relative block number of the first unused block(s)
4301  * in the file with at least "len" logically contiguous blocks free.
4302  * This is the lowest-address hole if the file has holes, else the first block
4303  * past the end of file.
4304  * Return 0 if the file is currently local (in-inode).
4305  */
4306 int                                             /* error */
4307 xfs_bmap_first_unused(
4308         xfs_trans_t     *tp,                    /* transaction pointer */
4309         xfs_inode_t     *ip,                    /* incore inode */
4310         xfs_extlen_t    len,                    /* size of hole to find */
4311         xfs_fileoff_t   *first_unused,          /* unused block */
4312         int             whichfork)              /* data or attr fork */
4313 {
4314         int             error;                  /* error return value */
4315         int             idx;                    /* extent record index */
4316         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4317         xfs_fileoff_t   lastaddr;               /* last block number seen */
4318         xfs_fileoff_t   lowest;                 /* lowest useful block */
4319         xfs_fileoff_t   max;                    /* starting useful block */
4320         xfs_fileoff_t   off;                    /* offset for this block */
4321         xfs_extnum_t    nextents;               /* number of extent entries */
4322
4323         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
4324                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
4325                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
4326         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4327                 *first_unused = 0;
4328                 return 0;
4329         }
4330         ifp = XFS_IFORK_PTR(ip, whichfork);
4331         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4332             (error = xfs_iread_extents(tp, ip, whichfork)))
4333                 return error;
4334         lowest = *first_unused;
4335         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4336         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
4337                 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
4338                 off = xfs_bmbt_get_startoff(ep);
4339                 /*
4340                  * See if the hole before this extent will work.
4341                  */
4342                 if (off >= lowest + len && off - max >= len) {
4343                         *first_unused = max;
4344                         return 0;
4345                 }
4346                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
4347                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
4348         }
4349         *first_unused = max;
4350         return 0;
4351 }
4352
4353 /*
4354  * Returns the file-relative block number of the last block + 1 before
4355  * last_block (input value) in the file.
4356  * This is not based on i_size, it is based on the extent records.
4357  * Returns 0 for local files, as they do not have extent records.
4358  */
4359 int                                             /* error */
4360 xfs_bmap_last_before(
4361         xfs_trans_t     *tp,                    /* transaction pointer */
4362         xfs_inode_t     *ip,                    /* incore inode */
4363         xfs_fileoff_t   *last_block,            /* last block */
4364         int             whichfork)              /* data or attr fork */
4365 {
4366         xfs_fileoff_t   bno;                    /* input file offset */
4367         int             eof;                    /* hit end of file */
4368         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
4369         int             error;                  /* error return value */
4370         xfs_bmbt_irec_t got;                    /* current extent value */
4371         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4372         xfs_extnum_t    lastx;                  /* last extent used */
4373         xfs_bmbt_irec_t prev;                   /* previous extent value */
4374
4375         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4376             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4377             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4378                return XFS_ERROR(EIO);
4379         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4380                 *last_block = 0;
4381                 return 0;
4382         }
4383         ifp = XFS_IFORK_PTR(ip, whichfork);
4384         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4385             (error = xfs_iread_extents(tp, ip, whichfork)))
4386                 return error;
4387         bno = *last_block - 1;
4388         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4389                 &prev);
4390         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
4391                 if (prev.br_startoff == NULLFILEOFF)
4392                         *last_block = 0;
4393                 else
4394                         *last_block = prev.br_startoff + prev.br_blockcount;
4395         }
4396         /*
4397          * Otherwise *last_block is already the right answer.
4398          */
4399         return 0;
4400 }
4401
4402 /*
4403  * Returns the file-relative block number of the first block past eof in
4404  * the file.  This is not based on i_size, it is based on the extent records.
4405  * Returns 0 for local files, as they do not have extent records.
4406  */
4407 int                                             /* error */
4408 xfs_bmap_last_offset(
4409         xfs_trans_t     *tp,                    /* transaction pointer */
4410         xfs_inode_t     *ip,                    /* incore inode */
4411         xfs_fileoff_t   *last_block,            /* last block */
4412         int             whichfork)              /* data or attr fork */
4413 {
4414         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
4415         int             error;                  /* error return value */
4416         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4417         xfs_extnum_t    nextents;               /* number of extent entries */
4418
4419         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4420             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4421             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4422                return XFS_ERROR(EIO);
4423         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4424                 *last_block = 0;
4425                 return 0;
4426         }
4427         ifp = XFS_IFORK_PTR(ip, whichfork);
4428         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4429             (error = xfs_iread_extents(tp, ip, whichfork)))
4430                 return error;
4431         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4432         if (!nextents) {
4433                 *last_block = 0;
4434                 return 0;
4435         }
4436         ep = xfs_iext_get_ext(ifp, nextents - 1);
4437         *last_block = xfs_bmbt_get_startoff(ep) + xfs_bmbt_get_blockcount(ep);
4438         return 0;
4439 }
4440
4441 /*
4442  * Returns whether the selected fork of the inode has exactly one
4443  * block or not.  For the data fork we check this matches di_size,
4444  * implying the file's range is 0..bsize-1.
4445  */
4446 int                                     /* 1=>1 block, 0=>otherwise */
4447 xfs_bmap_one_block(
4448         xfs_inode_t     *ip,            /* incore inode */
4449         int             whichfork)      /* data or attr fork */
4450 {
4451         xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
4452         xfs_ifork_t     *ifp;           /* inode fork pointer */
4453         int             rval;           /* return value */
4454         xfs_bmbt_irec_t s;              /* internal version of extent */
4455
4456 #ifndef DEBUG
4457         if (whichfork == XFS_DATA_FORK) {
4458                 return ((ip->i_d.di_mode & S_IFMT) == S_IFREG) ?
4459                         (ip->i_size == ip->i_mount->m_sb.sb_blocksize) :
4460                         (ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize);
4461         }
4462 #endif  /* !DEBUG */
4463         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4464                 return 0;
4465         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4466                 return 0;
4467         ifp = XFS_IFORK_PTR(ip, whichfork);
4468         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4469         ep = xfs_iext_get_ext(ifp, 0);
4470         xfs_bmbt_get_all(ep, &s);
4471         rval = s.br_startoff == 0 && s.br_blockcount == 1;
4472         if (rval && whichfork == XFS_DATA_FORK)
4473                 ASSERT(ip->i_size == ip->i_mount->m_sb.sb_blocksize);
4474         return rval;
4475 }
4476
4477 /*
4478  * Read in the extents to if_extents.
4479  * All inode fields are set up by caller, we just traverse the btree
4480  * and copy the records in. If the file system cannot contain unwritten
4481  * extents, the records are checked for no "state" flags.
4482  */
4483 int                                     /* error */
4484 xfs_bmap_read_extents(
4485         xfs_trans_t             *tp,    /* transaction pointer */
4486         xfs_inode_t             *ip,    /* incore inode */
4487         int                     whichfork) /* data or attr fork */
4488 {
4489         xfs_bmbt_block_t        *block; /* current btree block */
4490         xfs_fsblock_t           bno;    /* block # of "block" */
4491         xfs_buf_t               *bp;    /* buffer for "block" */
4492         int                     error;  /* error return value */
4493         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
4494         xfs_extnum_t            i, j;   /* index into the extents list */
4495         xfs_ifork_t             *ifp;   /* fork structure */
4496         int                     level;  /* btree level, for checking */
4497         xfs_mount_t             *mp;    /* file system mount structure */
4498         __be64                  *pp;    /* pointer to block address */
4499         /* REFERENCED */
4500         xfs_extnum_t            room;   /* number of entries there's room for */
4501
4502         bno = NULLFSBLOCK;
4503         mp = ip->i_mount;
4504         ifp = XFS_IFORK_PTR(ip, whichfork);
4505         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4506                                         XFS_EXTFMT_INODE(ip);
4507         block = ifp->if_broot;
4508         /*
4509          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4510          */
4511         level = be16_to_cpu(block->bb_level);
4512         ASSERT(level > 0);
4513         pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
4514         bno = be64_to_cpu(*pp);
4515         ASSERT(bno != NULLDFSBNO);
4516         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
4517         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
4518         /*
4519          * Go down the tree until leaf level is reached, following the first
4520          * pointer (leftmost) at each level.
4521          */
4522         while (level-- > 0) {
4523                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4524                                 XFS_BMAP_BTREE_REF)))
4525                         return error;
4526                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
4527                 XFS_WANT_CORRUPTED_GOTO(
4528                         XFS_BMAP_SANITY_CHECK(mp, block, level),
4529                         error0);
4530                 if (level == 0)
4531                         break;
4532                 pp = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, 1, mp->m_bmap_dmxr[1]);
4533                 bno = be64_to_cpu(*pp);
4534                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
4535                 xfs_trans_brelse(tp, bp);
4536         }
4537         /*
4538          * Here with bp and block set to the leftmost leaf node in the tree.
4539          */
4540         room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4541         i = 0;
4542         /*
4543          * Loop over all leaf nodes.  Copy information to the extent records.
4544          */
4545         for (;;) {
4546                 xfs_bmbt_rec_t  *frp;
4547                 xfs_fsblock_t   nextbno;
4548                 xfs_extnum_t    num_recs;
4549                 xfs_extnum_t    start;
4550
4551
4552                 num_recs = be16_to_cpu(block->bb_numrecs);
4553                 if (unlikely(i + num_recs > room)) {
4554                         ASSERT(i + num_recs <= room);
4555                         xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
4556                                 "corrupt dinode %Lu, (btree extents).",
4557                                 (unsigned long long) ip->i_ino);
4558                         XFS_ERROR_REPORT("xfs_bmap_read_extents(1)",
4559                                          XFS_ERRLEVEL_LOW,
4560                                         ip->i_mount);
4561                         goto error0;
4562                 }
4563                 XFS_WANT_CORRUPTED_GOTO(
4564                         XFS_BMAP_SANITY_CHECK(mp, block, 0),
4565                         error0);
4566                 /*
4567                  * Read-ahead the next leaf block, if any.
4568                  */
4569                 nextbno = be64_to_cpu(block->bb_rightsib);
4570                 if (nextbno != NULLFSBLOCK)
4571                         xfs_btree_reada_bufl(mp, nextbno, 1);
4572                 /*
4573                  * Copy records into the extent records.
4574                  */
4575                 frp = XFS_BTREE_REC_ADDR(xfs_bmbt, block, 1);
4576                 start = i;
4577                 for (j = 0; j < num_recs; j++, i++, frp++) {
4578                         xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
4579                         trp->l0 = be64_to_cpu(frp->l0);
4580                         trp->l1 = be64_to_cpu(frp->l1);
4581                 }
4582                 if (exntf == XFS_EXTFMT_NOSTATE) {
4583                         /*
4584                          * Check all attribute bmap btree records and
4585                          * any "older" data bmap btree records for a
4586                          * set bit in the "extent flag" position.
4587                          */
4588                         if (unlikely(xfs_check_nostate_extents(ifp,
4589                                         start, num_recs))) {
4590                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4591                                                  XFS_ERRLEVEL_LOW,
4592                                                  ip->i_mount);
4593                                 goto error0;
4594                         }
4595                 }
4596                 xfs_trans_brelse(tp, bp);
4597                 bno = nextbno;
4598                 /*
4599                  * If we've reached the end, stop.
4600                  */
4601                 if (bno == NULLFSBLOCK)
4602                         break;
4603                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4604                                 XFS_BMAP_BTREE_REF)))
4605                         return error;
4606                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
4607         }
4608         ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4609         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4610         XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
4611         return 0;
4612 error0:
4613         xfs_trans_brelse(tp, bp);
4614         return XFS_ERROR(EFSCORRUPTED);
4615 }
4616
4617 #ifdef XFS_BMAP_TRACE
4618 /*
4619  * Add bmap trace insert entries for all the contents of the extent records.
4620  */
4621 void
4622 xfs_bmap_trace_exlist(
4623         const char      *fname,         /* function name */
4624         xfs_inode_t     *ip,            /* incore inode pointer */
4625         xfs_extnum_t    cnt,            /* count of entries in the list */
4626         int             whichfork)      /* data or attr fork */
4627 {
4628         xfs_bmbt_rec_host_t *ep;        /* current extent record */
4629         xfs_extnum_t    idx;            /* extent record index */
4630         xfs_ifork_t     *ifp;           /* inode fork pointer */
4631         xfs_bmbt_irec_t s;              /* file extent record */
4632
4633         ifp = XFS_IFORK_PTR(ip, whichfork);
4634         ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4635         for (idx = 0; idx < cnt; idx++) {
4636                 ep = xfs_iext_get_ext(ifp, idx);
4637                 xfs_bmbt_get_all(ep, &s);
4638                 XFS_BMAP_TRACE_INSERT("exlist", ip, idx, 1, &s, NULL,
4639                         whichfork);
4640         }
4641 }
4642 #endif
4643
4644 #ifdef DEBUG
4645 /*
4646  * Validate that the bmbt_irecs being returned from bmapi are valid
4647  * given the callers original parameters.  Specifically check the
4648  * ranges of the returned irecs to ensure that they only extent beyond
4649  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4650  */
4651 STATIC void
4652 xfs_bmap_validate_ret(
4653         xfs_fileoff_t           bno,
4654         xfs_filblks_t           len,
4655         int                     flags,
4656         xfs_bmbt_irec_t         *mval,
4657         int                     nmap,
4658         int                     ret_nmap)
4659 {
4660         int                     i;              /* index to map values */
4661
4662         ASSERT(ret_nmap <= nmap);
4663
4664         for (i = 0; i < ret_nmap; i++) {
4665                 ASSERT(mval[i].br_blockcount > 0);
4666                 if (!(flags & XFS_BMAPI_ENTIRE)) {
4667                         ASSERT(mval[i].br_startoff >= bno);
4668                         ASSERT(mval[i].br_blockcount <= len);
4669                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4670                                bno + len);
4671                 } else {
4672                         ASSERT(mval[i].br_startoff < bno + len);
4673                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4674                                bno);
4675                 }
4676                 ASSERT(i == 0 ||
4677                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4678                        mval[i].br_startoff);
4679                 if ((flags & XFS_BMAPI_WRITE) && !(flags & XFS_BMAPI_DELAY))
4680                         ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4681                                mval[i].br_startblock != HOLESTARTBLOCK);
4682                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4683                        mval[i].br_state == XFS_EXT_UNWRITTEN);
4684         }
4685 }
4686 #endif /* DEBUG */
4687
4688
4689 /*
4690  * Map file blocks to filesystem blocks.
4691  * File range is given by the bno/len pair.
4692  * Adds blocks to file if a write ("flags & XFS_BMAPI_WRITE" set)
4693  * into a hole or past eof.
4694  * Only allocates blocks from a single allocation group,
4695  * to avoid locking problems.
4696  * The returned value in "firstblock" from the first call in a transaction
4697  * must be remembered and presented to subsequent calls in "firstblock".
4698  * An upper bound for the number of blocks to be allocated is supplied to
4699  * the first call in "total"; if no allocation group has that many free
4700  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4701  */
4702 int                                     /* error */
4703 xfs_bmapi(
4704         xfs_trans_t     *tp,            /* transaction pointer */
4705         xfs_inode_t     *ip,            /* incore inode */
4706         xfs_fileoff_t   bno,            /* starting file offs. mapped */
4707         xfs_filblks_t   len,            /* length to map in file */
4708         int             flags,          /* XFS_BMAPI_... */
4709         xfs_fsblock_t   *firstblock,    /* first allocated block
4710                                            controls a.g. for allocs */
4711         xfs_extlen_t    total,          /* total blocks needed */
4712         xfs_bmbt_irec_t *mval,          /* output: map values */
4713         int             *nmap,          /* i/o: mval size/count */
4714         xfs_bmap_free_t *flist,         /* i/o: list extents to free */
4715         xfs_extdelta_t  *delta)         /* o: change made to incore extents */
4716 {
4717         xfs_fsblock_t   abno;           /* allocated block number */
4718         xfs_extlen_t    alen;           /* allocated extent length */
4719         xfs_fileoff_t   aoff;           /* allocated file offset */
4720         xfs_bmalloca_t  bma;            /* args for xfs_bmap_alloc */
4721         xfs_btree_cur_t *cur;           /* bmap btree cursor */
4722         xfs_fileoff_t   end;            /* end of mapped file region */
4723         int             eof;            /* we've hit the end of extents */
4724         xfs_bmbt_rec_host_t *ep;        /* extent record pointer */
4725         int             error;          /* error return */
4726         xfs_bmbt_irec_t got;            /* current file extent record */
4727         xfs_ifork_t     *ifp;           /* inode fork pointer */
4728         xfs_extlen_t    indlen;         /* indirect blocks length */
4729         xfs_extnum_t    lastx;          /* last useful extent number */
4730         int             logflags;       /* flags for transaction logging */
4731         xfs_extlen_t    minleft;        /* min blocks left after allocation */
4732         xfs_extlen_t    minlen;         /* min allocation size */
4733         xfs_mount_t     *mp;            /* xfs mount structure */
4734         int             n;              /* current extent index */
4735         int             nallocs;        /* number of extents alloc\'d */
4736         xfs_extnum_t    nextents;       /* number of extents in file */
4737         xfs_fileoff_t   obno;           /* old block number (offset) */
4738         xfs_bmbt_irec_t prev;           /* previous file extent record */
4739         int             tmp_logflags;   /* temp flags holder */
4740         int             whichfork;      /* data or attr fork */
4741         char            inhole;         /* current location is hole in file */
4742         char            wasdelay;       /* old extent was delayed */
4743         char            wr;             /* this is a write request */
4744         char            rt;             /* this is a realtime file */
4745 #ifdef DEBUG
4746         xfs_fileoff_t   orig_bno;       /* original block number value */
4747         int             orig_flags;     /* original flags arg value */
4748         xfs_filblks_t   orig_len;       /* original value of len arg */
4749         xfs_bmbt_irec_t *orig_mval;     /* original value of mval */
4750         int             orig_nmap;      /* original value of *nmap */
4751
4752         orig_bno = bno;
4753         orig_len = len;
4754         orig_flags = flags;
4755         orig_mval = mval;
4756         orig_nmap = *nmap;
4757 #endif
4758         ASSERT(*nmap >= 1);
4759         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP || !(flags & XFS_BMAPI_WRITE));
4760         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4761                 XFS_ATTR_FORK : XFS_DATA_FORK;
4762         mp = ip->i_mount;
4763         if (unlikely(XFS_TEST_ERROR(
4764             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4765              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4766              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4767              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4768                 XFS_ERROR_REPORT("xfs_bmapi", XFS_ERRLEVEL_LOW, mp);
4769                 return XFS_ERROR(EFSCORRUPTED);
4770         }
4771         if (XFS_FORCED_SHUTDOWN(mp))
4772                 return XFS_ERROR(EIO);
4773         rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4774         ifp = XFS_IFORK_PTR(ip, whichfork);
4775         ASSERT(ifp->if_ext_max ==
4776                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4777         if ((wr = (flags & XFS_BMAPI_WRITE)) != 0)
4778                 XFS_STATS_INC(xs_blk_mapw);
4779         else
4780                 XFS_STATS_INC(xs_blk_mapr);
4781         /*
4782          * IGSTATE flag is used to combine extents which
4783          * differ only due to the state of the extents.
4784          * This technique is used from xfs_getbmap()
4785          * when the caller does not wish to see the
4786          * separation (which is the default).
4787          *
4788          * This technique is also used when writing a
4789          * buffer which has been partially written,
4790          * (usually by being flushed during a chunkread),
4791          * to ensure one write takes place. This also
4792          * prevents a change in the xfs inode extents at
4793          * this time, intentionally. This change occurs
4794          * on completion of the write operation, in
4795          * xfs_strat_comp(), where the xfs_bmapi() call
4796          * is transactioned, and the extents combined.
4797          */
4798         if ((flags & XFS_BMAPI_IGSTATE) && wr)  /* if writing unwritten space */
4799                 wr = 0;                         /* no allocations are allowed */
4800         ASSERT(wr || !(flags & XFS_BMAPI_DELAY));
4801         logflags = 0;
4802         nallocs = 0;
4803         cur = NULL;
4804         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4805                 ASSERT(wr && tp);
4806                 if ((error = xfs_bmap_local_to_extents(tp, ip,
4807                                 firstblock, total, &logflags, whichfork)))
4808                         goto error0;
4809         }
4810         if (wr && *firstblock == NULLFSBLOCK) {
4811                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4812                         minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4813                 else
4814                         minleft = 1;
4815         } else
4816                 minleft = 0;
4817         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4818             (error = xfs_iread_extents(tp, ip, whichfork)))
4819                 goto error0;
4820         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4821                 &prev);
4822         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4823         n = 0;
4824         end = bno + len;
4825         obno = bno;
4826         bma.ip = NULL;
4827         if (delta) {
4828                 delta->xed_startoff = NULLFILEOFF;
4829                 delta->xed_blockcount = 0;
4830         }
4831         while (bno < end && n < *nmap) {
4832                 /*
4833                  * Reading past eof, act as though there's a hole
4834                  * up to end.
4835                  */
4836                 if (eof && !wr)
4837                         got.br_startoff = end;
4838                 inhole = eof || got.br_startoff > bno;
4839                 wasdelay = wr && !inhole && !(flags & XFS_BMAPI_DELAY) &&
4840                         ISNULLSTARTBLOCK(got.br_startblock);
4841                 /*
4842                  * First, deal with the hole before the allocated space
4843                  * that we found, if any.
4844                  */
4845                 if (wr && (inhole || wasdelay)) {
4846                         /*
4847                          * For the wasdelay case, we could also just
4848                          * allocate the stuff asked for in this bmap call
4849                          * but that wouldn't be as good.
4850                          */
4851                         if (wasdelay && !(flags & XFS_BMAPI_EXACT)) {
4852                                 alen = (xfs_extlen_t)got.br_blockcount;
4853                                 aoff = got.br_startoff;
4854                                 if (lastx != NULLEXTNUM && lastx) {
4855                                         ep = xfs_iext_get_ext(ifp, lastx - 1);
4856                                         xfs_bmbt_get_all(ep, &prev);
4857                                 }
4858                         } else if (wasdelay) {
4859                                 alen = (xfs_extlen_t)
4860                                         XFS_FILBLKS_MIN(len,
4861                                                 (got.br_startoff +
4862                                                  got.br_blockcount) - bno);
4863                                 aoff = bno;
4864                         } else {
4865                                 alen = (xfs_extlen_t)
4866                                         XFS_FILBLKS_MIN(len, MAXEXTLEN);
4867                                 if (!eof)
4868                                         alen = (xfs_extlen_t)
4869                                                 XFS_FILBLKS_MIN(alen,
4870                                                         got.br_startoff - bno);
4871                                 aoff = bno;
4872                         }
4873                         minlen = (flags & XFS_BMAPI_CONTIG) ? alen : 1;
4874                         if (flags & XFS_BMAPI_DELAY) {
4875                                 xfs_extlen_t    extsz;
4876
4877                                 /* Figure out the extent size, adjust alen */
4878                                 extsz = xfs_get_extsz_hint(ip);
4879                                 if (extsz) {
4880                                         error = xfs_bmap_extsize_align(mp,
4881                                                         &got, &prev, extsz,
4882                                                         rt, eof,
4883                                                         flags&XFS_BMAPI_DELAY,
4884                                                         flags&XFS_BMAPI_CONVERT,
4885                                                         &aoff, &alen);
4886                                         ASSERT(!error);
4887                                 }
4888
4889                                 if (rt)
4890                                         extsz = alen / mp->m_sb.sb_rextsize;
4891
4892                                 /*
4893                                  * Make a transaction-less quota reservation for
4894                                  * delayed allocation blocks. This number gets
4895                                  * adjusted later.  We return if we haven't
4896                                  * allocated blocks already inside this loop.
4897                                  */
4898                                 if ((error = XFS_TRANS_RESERVE_QUOTA_NBLKS(
4899                                                 mp, NULL, ip, (long)alen, 0,
4900                                                 rt ? XFS_QMOPT_RES_RTBLKS :
4901                                                      XFS_QMOPT_RES_REGBLKS))) {
4902                                         if (n == 0) {
4903                                                 *nmap = 0;
4904                                                 ASSERT(cur == NULL);
4905                                                 return error;
4906                                         }
4907                                         break;
4908                                 }
4909
4910                                 /*
4911                                  * Split changing sb for alen and indlen since
4912                                  * they could be coming from different places.
4913                                  */
4914                                 indlen = (xfs_extlen_t)
4915                                         xfs_bmap_worst_indlen(ip, alen);
4916                                 ASSERT(indlen > 0);
4917
4918                                 if (rt) {
4919                                         error = xfs_mod_incore_sb(mp,
4920                                                         XFS_SBS_FREXTENTS,
4921                                                         -((int64_t)extsz), (flags &
4922                                                         XFS_BMAPI_RSVBLOCKS));
4923                                 } else {
4924                                         error = xfs_mod_incore_sb(mp,
4925                                                         XFS_SBS_FDBLOCKS,
4926                                                         -((int64_t)alen), (flags &
4927                                                         XFS_BMAPI_RSVBLOCKS));
4928                                 }
4929                                 if (!error) {
4930                                         error = xfs_mod_incore_sb(mp,
4931                                                         XFS_SBS_FDBLOCKS,
4932                                                         -((int64_t)indlen), (flags &
4933                                                         XFS_BMAPI_RSVBLOCKS));
4934                                         if (error && rt)
4935                                                 xfs_mod_incore_sb(mp,
4936                                                         XFS_SBS_FREXTENTS,
4937                                                         (int64_t)extsz, (flags &
4938                                                         XFS_BMAPI_RSVBLOCKS));
4939                                         else if (error)
4940                                                 xfs_mod_incore_sb(mp,
4941                                                         XFS_SBS_FDBLOCKS,
4942                                                         (int64_t)alen, (flags &
4943                                                         XFS_BMAPI_RSVBLOCKS));
4944                                 }
4945
4946                                 if (error) {
4947                                         if (XFS_IS_QUOTA_ON(mp))
4948                                                 /* unreserve the blocks now */
4949                                                 (void)
4950                                                 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(
4951                                                         mp, NULL, ip,
4952                                                         (long)alen, 0, rt ?
4953                                                         XFS_QMOPT_RES_RTBLKS :
4954                                                         XFS_QMOPT_RES_REGBLKS);
4955                                         break;
4956                                 }
4957
4958                                 ip->i_delayed_blks += alen;
4959                                 abno = NULLSTARTBLOCK(indlen);
4960                         } else {
4961                                 /*
4962                                  * If first time, allocate and fill in
4963                                  * once-only bma fields.
4964                                  */
4965                                 if (bma.ip == NULL) {
4966                                         bma.tp = tp;
4967                                         bma.ip = ip;
4968                                         bma.prevp = &prev;
4969                                         bma.gotp = &got;
4970                                         bma.total = total;
4971                                         bma.userdata = 0;
4972                                 }
4973                                 /* Indicate if this is the first user data
4974                                  * in the file, or just any user data.
4975                                  */
4976                                 if (!(flags & XFS_BMAPI_METADATA)) {
4977                                         bma.userdata = (aoff == 0) ?
4978                                                 XFS_ALLOC_INITIAL_USER_DATA :
4979                                                 XFS_ALLOC_USERDATA;
4980                                 }
4981                                 /*
4982                                  * Fill in changeable bma fields.
4983                                  */
4984                                 bma.eof = eof;
4985                                 bma.firstblock = *firstblock;
4986                                 bma.alen = alen;
4987                                 bma.off = aoff;
4988                                 bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4989                                 bma.wasdel = wasdelay;
4990                                 bma.minlen = minlen;
4991                                 bma.low = flist->xbf_low;
4992                                 bma.minleft = minleft;
4993                                 /*
4994                                  * Only want to do the alignment at the
4995                                  * eof if it is userdata and allocation length
4996                                  * is larger than a stripe unit.
4997                                  */
4998                                 if (mp->m_dalign && alen >= mp->m_dalign &&
4999                                     (!(flags & XFS_BMAPI_METADATA)) &&
5000                                     (whichfork == XFS_DATA_FORK)) {
5001                                         if ((error = xfs_bmap_isaeof(ip, aoff,
5002                                                         whichfork, &bma.aeof)))
5003                                                 goto error0;
5004                                 } else
5005                                         bma.aeof = 0;
5006                                 /*
5007                                  * Call allocator.
5008                                  */
5009                                 if ((error = xfs_bmap_alloc(&bma)))
5010                                         goto error0;
5011                                 /*
5012                                  * Copy out result fields.
5013                                  */
5014                                 abno = bma.rval;
5015                                 if ((flist->xbf_low = bma.low))
5016                                         minleft = 0;
5017                                 alen = bma.alen;
5018                                 aoff = bma.off;
5019                                 ASSERT(*firstblock == NULLFSBLOCK ||
5020                                        XFS_FSB_TO_AGNO(mp, *firstblock) ==
5021                                        XFS_FSB_TO_AGNO(mp, bma.firstblock) ||
5022                                        (flist->xbf_low &&
5023                                         XFS_FSB_TO_AGNO(mp, *firstblock) <
5024                                         XFS_FSB_TO_AGNO(mp, bma.firstblock)));
5025                                 *firstblock = bma.firstblock;
5026                                 if (cur)
5027                                         cur->bc_private.b.firstblock =
5028                                                 *firstblock;
5029                                 if (abno == NULLFSBLOCK)
5030                                         break;
5031                                 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
5032                                         cur = xfs_btree_init_cursor(mp,
5033                                                 tp, NULL, 0, XFS_BTNUM_BMAP,
5034                                                 ip, whichfork);
5035                                         cur->bc_private.b.firstblock =
5036                                                 *firstblock;
5037                                         cur->bc_private.b.flist = flist;
5038                                 }
5039                                 /*
5040                                  * Bump the number of extents we've allocated
5041                                  * in this call.
5042                                  */
5043                                 nallocs++;
5044                         }
5045                         if (cur)
5046                                 cur->bc_private.b.flags =
5047                                         wasdelay ? XFS_BTCUR_BPRV_WASDEL : 0;
5048                         got.br_startoff = aoff;
5049                         got.br_startblock = abno;
5050                         got.br_blockcount = alen;
5051                         got.br_state = XFS_EXT_NORM;    /* assume normal */
5052                         /*
5053                          * Determine state of extent, and the filesystem.
5054                          * A wasdelay extent has been initialized, so
5055                          * shouldn't be flagged as unwritten.
5056                          */
5057                         if (wr && xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5058                                 if (!wasdelay && (flags & XFS_BMAPI_PREALLOC))
5059                                         got.br_state = XFS_EXT_UNWRITTEN;
5060                         }
5061                         error = xfs_bmap_add_extent(ip, lastx, &cur, &got,
5062                                 firstblock, flist, &tmp_logflags, delta,
5063                                 whichfork, (flags & XFS_BMAPI_RSVBLOCKS));
5064                         logflags |= tmp_logflags;
5065                         if (error)
5066                                 goto error0;
5067                         lastx = ifp->if_lastex;
5068                         ep = xfs_iext_get_ext(ifp, lastx);
5069                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5070                         xfs_bmbt_get_all(ep, &got);
5071                         ASSERT(got.br_startoff <= aoff);
5072                         ASSERT(got.br_startoff + got.br_blockcount >=
5073                                 aoff + alen);
5074 #ifdef DEBUG
5075                         if (flags & XFS_BMAPI_DELAY) {
5076                                 ASSERT(ISNULLSTARTBLOCK(got.br_startblock));
5077                                 ASSERT(STARTBLOCKVAL(got.br_startblock) > 0);
5078                         }
5079                         ASSERT(got.br_state == XFS_EXT_NORM ||
5080                                got.br_state == XFS_EXT_UNWRITTEN);
5081 #endif
5082                         /*
5083                          * Fall down into the found allocated space case.
5084                          */
5085                 } else if (inhole) {
5086                         /*
5087                          * Reading in a hole.
5088                          */
5089                         mval->br_startoff = bno;
5090                         mval->br_startblock = HOLESTARTBLOCK;
5091                         mval->br_blockcount =
5092                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
5093                         mval->br_state = XFS_EXT_NORM;
5094                         bno += mval->br_blockcount;
5095                         len -= mval->br_blockcount;
5096                         mval++;
5097                         n++;
5098                         continue;
5099                 }
5100                 /*
5101                  * Then deal with the allocated space we found.
5102                  */
5103                 ASSERT(ep != NULL);
5104                 if (!(flags & XFS_BMAPI_ENTIRE) &&
5105                     (got.br_startoff + got.br_blockcount > obno)) {
5106                         if (obno > bno)
5107                                 bno = obno;
5108                         ASSERT((bno >= obno) || (n == 0));
5109                         ASSERT(bno < end);
5110                         mval->br_startoff = bno;
5111                         if (ISNULLSTARTBLOCK(got.br_startblock)) {
5112                                 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
5113                                 mval->br_startblock = DELAYSTARTBLOCK;
5114                         } else
5115                                 mval->br_startblock =
5116                                         got.br_startblock +
5117                                         (bno - got.br_startoff);
5118                         /*
5119                          * Return the minimum of what we got and what we
5120                          * asked for for the length.  We can use the len
5121                          * variable here because it is modified below
5122                          * and we could have been there before coming
5123                          * here if the first part of the allocation
5124                          * didn't overlap what was asked for.
5125                          */
5126                         mval->br_blockcount =
5127                                 XFS_FILBLKS_MIN(end - bno, got.br_blockcount -
5128                                         (bno - got.br_startoff));
5129                         mval->br_state = got.br_state;
5130                         ASSERT(mval->br_blockcount <= len);
5131                 } else {
5132                         *mval = got;
5133                         if (ISNULLSTARTBLOCK(mval->br_startblock)) {
5134                                 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
5135                                 mval->br_startblock = DELAYSTARTBLOCK;
5136                         }
5137                 }
5138
5139                 /*
5140                  * Check if writing previously allocated but
5141                  * unwritten extents.
5142                  */
5143                 if (wr && mval->br_state == XFS_EXT_UNWRITTEN &&
5144                     ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_DELAY)) == 0)) {
5145                         /*
5146                          * Modify (by adding) the state flag, if writing.
5147                          */
5148                         ASSERT(mval->br_blockcount <= len);
5149                         if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
5150                                 cur = xfs_btree_init_cursor(mp,
5151                                         tp, NULL, 0, XFS_BTNUM_BMAP,
5152                                         ip, whichfork);
5153                                 cur->bc_private.b.firstblock =
5154                                         *firstblock;
5155                                 cur->bc_private.b.flist = flist;
5156                         }
5157                         mval->br_state = XFS_EXT_NORM;
5158                         error = xfs_bmap_add_extent(ip, lastx, &cur, mval,
5159                                 firstblock, flist, &tmp_logflags, delta,
5160                                 whichfork, (flags & XFS_BMAPI_RSVBLOCKS));
5161                         logflags |= tmp_logflags;
5162                         if (error)
5163                                 goto error0;
5164                         lastx = ifp->if_lastex;
5165                         ep = xfs_iext_get_ext(ifp, lastx);
5166                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5167                         xfs_bmbt_get_all(ep, &got);
5168                         /*
5169                          * We may have combined previously unwritten
5170                          * space with written space, so generate
5171                          * another request.
5172                          */
5173                         if (mval->br_blockcount < len)
5174                                 continue;
5175                 }
5176
5177                 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
5178                        ((mval->br_startoff + mval->br_blockcount) <= end));
5179                 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
5180                        (mval->br_blockcount <= len) ||
5181                        (mval->br_startoff < obno));
5182                 bno = mval->br_startoff + mval->br_blockcount;
5183                 len = end - bno;
5184                 if (n > 0 && mval->br_startoff == mval[-1].br_startoff) {
5185                         ASSERT(mval->br_startblock == mval[-1].br_startblock);
5186                         ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
5187                         ASSERT(mval->br_state == mval[-1].br_state);
5188                         mval[-1].br_blockcount = mval->br_blockcount;
5189                         mval[-1].br_state = mval->br_state;
5190                 } else if (n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
5191                            mval[-1].br_startblock != DELAYSTARTBLOCK &&
5192                            mval[-1].br_startblock != HOLESTARTBLOCK &&
5193                            mval->br_startblock ==
5194                            mval[-1].br_startblock + mval[-1].br_blockcount &&
5195                            ((flags & XFS_BMAPI_IGSTATE) ||
5196                                 mval[-1].br_state == mval->br_state)) {
5197                         ASSERT(mval->br_startoff ==
5198                                mval[-1].br_startoff + mval[-1].br_blockcount);
5199                         mval[-1].br_blockcount += mval->br_blockcount;
5200                 } else if (n > 0 &&
5201                            mval->br_startblock == DELAYSTARTBLOCK &&
5202                            mval[-1].br_startblock == DELAYSTARTBLOCK &&
5203                            mval->br_startoff ==
5204                            mval[-1].br_startoff + mval[-1].br_blockcount) {
5205                         mval[-1].br_blockcount += mval->br_blockcount;
5206                         mval[-1].br_state = mval->br_state;
5207                 } else if (!((n == 0) &&
5208                              ((mval->br_startoff + mval->br_blockcount) <=
5209                               obno))) {
5210                         mval++;
5211                         n++;
5212                 }
5213                 /*
5214                  * If we're done, stop now.  Stop when we've allocated
5215                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
5216                  * the transaction may get too big.
5217                  */
5218                 if (bno >= end || n >= *nmap || nallocs >= *nmap)
5219                         break;
5220                 /*
5221                  * Else go on to the next record.
5222                  */
5223                 ep = xfs_iext_get_ext(ifp, ++lastx);
5224                 prev = got;
5225                 if (lastx >= nextents)
5226                         eof = 1;
5227                 else
5228                         xfs_bmbt_get_all(ep, &got);
5229         }
5230         ifp->if_lastex = lastx;
5231         *nmap = n;
5232         /*
5233          * Transform from btree to extents, give it cur.
5234          */
5235         if (tp && XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5236             XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5237                 ASSERT(wr && cur);
5238                 error = xfs_bmap_btree_to_extents(tp, ip, cur,
5239                         &tmp_logflags, whichfork);
5240                 logflags |= tmp_logflags;
5241                 if (error)
5242                         goto error0;
5243         }
5244         ASSERT(ifp->if_ext_max ==
5245                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5246         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
5247                XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max);
5248         error = 0;
5249         if (delta && delta->xed_startoff != NULLFILEOFF) {
5250                 /* A change was actually made.
5251                  * Note that delta->xed_blockount is an offset at this
5252                  * point and needs to be converted to a block count.
5253                  */
5254                 ASSERT(delta->xed_blockcount > delta->xed_startoff);
5255                 delta->xed_blockcount -= delta->xed_startoff;
5256         }
5257 error0:
5258         /*
5259          * Log everything.  Do this after conversion, there's no point in
5260          * logging the extent records if we've converted to btree format.
5261          */
5262         if ((logflags & XFS_ILOG_FEXT(whichfork)) &&
5263             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5264                 logflags &= ~XFS_ILOG_FEXT(whichfork);
5265         else if ((logflags & XFS_ILOG_FBROOT(whichfork)) &&
5266                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5267                 logflags &= ~XFS_ILOG_FBROOT(whichfork);
5268         /*
5269          * Log whatever the flags say, even if error.  Otherwise we might miss
5270          * detecting a case where the data is changed, there's an error,
5271          * and it's not logged so we don't shutdown when we should.
5272          */
5273         if (logflags) {
5274                 ASSERT(tp && wr);
5275                 xfs_trans_log_inode(tp, ip, logflags);
5276         }
5277         if (cur) {
5278                 if (!error) {
5279                         ASSERT(*firstblock == NULLFSBLOCK ||
5280                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
5281                                XFS_FSB_TO_AGNO(mp,
5282                                        cur->bc_private.b.firstblock) ||
5283                                (flist->xbf_low &&
5284                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
5285                                 XFS_FSB_TO_AGNO(mp,
5286                                         cur->bc_private.b.firstblock)));
5287                         *firstblock = cur->bc_private.b.firstblock;
5288                 }
5289                 xfs_btree_del_cursor(cur,
5290                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5291         }
5292         if (!error)
5293                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
5294                         orig_nmap, *nmap);
5295         return error;
5296 }
5297
5298 /*
5299  * Map file blocks to filesystem blocks, simple version.
5300  * One block (extent) only, read-only.
5301  * For flags, only the XFS_BMAPI_ATTRFORK flag is examined.
5302  * For the other flag values, the effect is as if XFS_BMAPI_METADATA
5303  * was set and all the others were clear.
5304  */
5305 int                                             /* error */
5306 xfs_bmapi_single(
5307         xfs_trans_t     *tp,            /* transaction pointer */
5308         xfs_inode_t     *ip,            /* incore inode */
5309         int             whichfork,      /* data or attr fork */
5310         xfs_fsblock_t   *fsb,           /* output: mapped block */
5311         xfs_fileoff_t   bno)            /* starting file offs. mapped */
5312 {
5313         int             eof;            /* we've hit the end of extents */
5314         int             error;          /* error return */
5315         xfs_bmbt_irec_t got;            /* current file extent record */
5316         xfs_ifork_t     *ifp;           /* inode fork pointer */
5317         xfs_extnum_t    lastx;          /* last useful extent number */
5318         xfs_bmbt_irec_t prev;           /* previous file extent record */
5319
5320         ifp = XFS_IFORK_PTR(ip, whichfork);
5321         if (unlikely(
5322             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
5323             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)) {
5324                XFS_ERROR_REPORT("xfs_bmapi_single", XFS_ERRLEVEL_LOW,
5325                                 ip->i_mount);
5326                return XFS_ERROR(EFSCORRUPTED);
5327         }
5328         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
5329                 return XFS_ERROR(EIO);
5330         XFS_STATS_INC(xs_blk_mapr);
5331         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5332             (error = xfs_iread_extents(tp, ip, whichfork)))
5333                 return error;
5334         (void)xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5335                 &prev);
5336         /*
5337          * Reading past eof, act as though there's a hole
5338          * up to end.
5339          */
5340         if (eof || got.br_startoff > bno) {
5341                 *fsb = NULLFSBLOCK;
5342                 return 0;
5343         }
5344         ASSERT(!ISNULLSTARTBLOCK(got.br_startblock));
5345         ASSERT(bno < got.br_startoff + got.br_blockcount);
5346         *fsb = got.br_startblock + (bno - got.br_startoff);
5347         ifp->if_lastex = lastx;
5348         return 0;
5349 }
5350
5351 /*
5352  * Unmap (remove) blocks from a file.
5353  * If nexts is nonzero then the number of extents to remove is limited to
5354  * that value.  If not all extents in the block range can be removed then
5355  * *done is set.
5356  */
5357 int                                             /* error */
5358 xfs_bunmapi(
5359         xfs_trans_t             *tp,            /* transaction pointer */
5360         struct xfs_inode        *ip,            /* incore inode */
5361         xfs_fileoff_t           bno,            /* starting offset to unmap */
5362         xfs_filblks_t           len,            /* length to unmap in file */
5363         int                     flags,          /* misc flags */
5364         xfs_extnum_t            nexts,          /* number of extents max */
5365         xfs_fsblock_t           *firstblock,    /* first allocated block
5366                                                    controls a.g. for allocs */
5367         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
5368         xfs_extdelta_t          *delta,         /* o: change made to incore
5369                                                    extents */
5370         int                     *done)          /* set if not done yet */
5371 {
5372         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
5373         xfs_bmbt_irec_t         del;            /* extent being deleted */
5374         int                     eof;            /* is deleting at eof */
5375         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
5376         int                     error;          /* error return value */
5377         xfs_extnum_t            extno;          /* extent number in list */
5378         xfs_bmbt_irec_t         got;            /* current extent record */
5379         xfs_ifork_t             *ifp;           /* inode fork pointer */
5380         int                     isrt;           /* freeing in rt area */
5381         xfs_extnum_t            lastx;          /* last extent index used */
5382         int                     logflags;       /* transaction logging flags */
5383         xfs_extlen_t            mod;            /* rt extent offset */
5384         xfs_mount_t             *mp;            /* mount structure */
5385         xfs_extnum_t            nextents;       /* number of file extents */
5386         xfs_bmbt_irec_t         prev;           /* previous extent record */
5387         xfs_fileoff_t           start;          /* first file offset deleted */
5388         int                     tmp_logflags;   /* partial logging flags */
5389         int                     wasdel;         /* was a delayed alloc extent */
5390         int                     whichfork;      /* data or attribute fork */
5391         int                     rsvd;           /* OK to allocate reserved blocks */
5392         xfs_fsblock_t           sum;
5393
5394         xfs_bunmap_trace(ip, bno, len, flags, (inst_t *)__return_address);
5395         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5396                 XFS_ATTR_FORK : XFS_DATA_FORK;
5397         ifp = XFS_IFORK_PTR(ip, whichfork);
5398         if (unlikely(
5399             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5400             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5401                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5402                                  ip->i_mount);
5403                 return XFS_ERROR(EFSCORRUPTED);
5404         }
5405         mp = ip->i_mount;
5406         if (XFS_FORCED_SHUTDOWN(mp))
5407                 return XFS_ERROR(EIO);
5408         rsvd = (flags & XFS_BMAPI_RSVBLOCKS) != 0;
5409         ASSERT(len > 0);
5410         ASSERT(nexts >= 0);
5411         ASSERT(ifp->if_ext_max ==
5412                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5413         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5414             (error = xfs_iread_extents(tp, ip, whichfork)))
5415                 return error;
5416         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5417         if (nextents == 0) {
5418                 *done = 1;
5419                 return 0;
5420         }
5421         XFS_STATS_INC(xs_blk_unmap);
5422         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5423         start = bno;
5424         bno = start + len - 1;
5425         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5426                 &prev);
5427         if (delta) {
5428                 delta->xed_startoff = NULLFILEOFF;
5429                 delta->xed_blockcount = 0;
5430         }
5431         /*
5432          * Check to see if the given block number is past the end of the
5433          * file, back up to the last block if so...
5434          */
5435         if (eof) {
5436                 ep = xfs_iext_get_ext(ifp, --lastx);
5437                 xfs_bmbt_get_all(ep, &got);
5438                 bno = got.br_startoff + got.br_blockcount - 1;
5439         }
5440         logflags = 0;
5441         if (ifp->if_flags & XFS_IFBROOT) {
5442                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5443                 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
5444                         whichfork);
5445                 cur->bc_private.b.firstblock = *firstblock;
5446                 cur->bc_private.b.flist = flist;
5447                 cur->bc_private.b.flags = 0;
5448         } else
5449                 cur = NULL;
5450         extno = 0;
5451         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5452                (nexts == 0 || extno < nexts)) {
5453                 /*
5454                  * Is the found extent after a hole in which bno lives?
5455                  * Just back up to the previous extent, if so.
5456                  */
5457                 if (got.br_startoff > bno) {
5458                         if (--lastx < 0)
5459                                 break;
5460                         ep = xfs_iext_get_ext(ifp, lastx);
5461                         xfs_bmbt_get_all(ep, &got);
5462                 }
5463                 /*
5464                  * Is the last block of this extent before the range
5465                  * we're supposed to delete?  If so, we're done.
5466                  */
5467                 bno = XFS_FILEOFF_MIN(bno,
5468                         got.br_startoff + got.br_blockcount - 1);
5469                 if (bno < start)
5470                         break;
5471                 /*
5472                  * Then deal with the (possibly delayed) allocated space
5473                  * we found.
5474                  */
5475                 ASSERT(ep != NULL);
5476                 del = got;
5477                 wasdel = ISNULLSTARTBLOCK(del.br_startblock);
5478                 if (got.br_startoff < start) {
5479                         del.br_startoff = start;
5480                         del.br_blockcount -= start - got.br_startoff;
5481                         if (!wasdel)
5482                                 del.br_startblock += start - got.br_startoff;
5483                 }
5484                 if (del.br_startoff + del.br_blockcount > bno + 1)
5485                         del.br_blockcount = bno + 1 - del.br_startoff;
5486                 sum = del.br_startblock + del.br_blockcount;
5487                 if (isrt &&
5488                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5489                         /*
5490                          * Realtime extent not lined up at the end.
5491                          * The extent could have been split into written
5492                          * and unwritten pieces, or we could just be
5493                          * unmapping part of it.  But we can't really
5494                          * get rid of part of a realtime extent.
5495                          */
5496                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5497                             !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5498                                 /*
5499                                  * This piece is unwritten, or we're not
5500                                  * using unwritten extents.  Skip over it.
5501                                  */
5502                                 ASSERT(bno >= mod);
5503                                 bno -= mod > del.br_blockcount ?
5504                                         del.br_blockcount : mod;
5505                                 if (bno < got.br_startoff) {
5506                                         if (--lastx >= 0)
5507                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5508                                                         ifp, lastx), &got);
5509                                 }
5510                                 continue;
5511                         }
5512                         /*
5513                          * It's written, turn it unwritten.
5514                          * This is better than zeroing it.
5515                          */
5516                         ASSERT(del.br_state == XFS_EXT_NORM);
5517                         ASSERT(xfs_trans_get_block_res(tp) > 0);
5518                         /*
5519                          * If this spans a realtime extent boundary,
5520                          * chop it back to the start of the one we end at.
5521                          */
5522                         if (del.br_blockcount > mod) {
5523                                 del.br_startoff += del.br_blockcount - mod;
5524                                 del.br_startblock += del.br_blockcount - mod;
5525                                 del.br_blockcount = mod;
5526                         }
5527                         del.br_state = XFS_EXT_UNWRITTEN;
5528                         error = xfs_bmap_add_extent(ip, lastx, &cur, &del,
5529                                 firstblock, flist, &logflags, delta,
5530                                 XFS_DATA_FORK, 0);
5531                         if (error)
5532                                 goto error0;
5533                         goto nodelete;
5534                 }
5535                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5536                         /*
5537                          * Realtime extent is lined up at the end but not
5538                          * at the front.  We'll get rid of full extents if
5539                          * we can.
5540                          */
5541                         mod = mp->m_sb.sb_rextsize - mod;
5542                         if (del.br_blockcount > mod) {
5543                                 del.br_blockcount -= mod;
5544                                 del.br_startoff += mod;
5545                                 del.br_startblock += mod;
5546                         } else if ((del.br_startoff == start &&
5547                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5548                                      xfs_trans_get_block_res(tp) == 0)) ||
5549                                    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5550                                 /*
5551                                  * Can't make it unwritten.  There isn't
5552                                  * a full extent here so just skip it.
5553                                  */
5554                                 ASSERT(bno >= del.br_blockcount);
5555                                 bno -= del.br_blockcount;
5556                                 if (bno < got.br_startoff) {
5557                                         if (--lastx >= 0)
5558                                                 xfs_bmbt_get_all(--ep, &got);
5559                                 }
5560                                 continue;
5561                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5562                                 /*
5563                                  * This one is already unwritten.
5564                                  * It must have a written left neighbor.
5565                                  * Unwrite the killed part of that one and
5566                                  * try again.
5567                                  */
5568                                 ASSERT(lastx > 0);
5569                                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5570                                                 lastx - 1), &prev);
5571                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5572                                 ASSERT(!ISNULLSTARTBLOCK(prev.br_startblock));
5573                                 ASSERT(del.br_startblock ==
5574                                        prev.br_startblock + prev.br_blockcount);
5575                                 if (prev.br_startoff < start) {
5576                                         mod = start - prev.br_startoff;
5577                                         prev.br_blockcount -= mod;
5578                                         prev.br_startblock += mod;
5579                                         prev.br_startoff = start;
5580                                 }
5581                                 prev.br_state = XFS_EXT_UNWRITTEN;
5582                                 error = xfs_bmap_add_extent(ip, lastx - 1, &cur,
5583                                         &prev, firstblock, flist, &logflags,
5584                                         delta, XFS_DATA_FORK, 0);
5585                                 if (error)
5586                                         goto error0;
5587                                 goto nodelete;
5588                         } else {
5589                                 ASSERT(del.br_state == XFS_EXT_NORM);
5590                                 del.br_state = XFS_EXT_UNWRITTEN;
5591                                 error = xfs_bmap_add_extent(ip, lastx, &cur,
5592                                         &del, firstblock, flist, &logflags,
5593                                         delta, XFS_DATA_FORK, 0);
5594                                 if (error)
5595                                         goto error0;
5596                                 goto nodelete;
5597                         }
5598                 }
5599                 if (wasdel) {
5600                         ASSERT(STARTBLOCKVAL(del.br_startblock) > 0);
5601                         /* Update realtime/data freespace, unreserve quota */
5602                         if (isrt) {
5603                                 xfs_filblks_t rtexts;
5604
5605                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5606                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5607                                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5608                                                 (int64_t)rtexts, rsvd);
5609                                 (void)XFS_TRANS_RESERVE_QUOTA_NBLKS(mp,
5610                                         NULL, ip, -((long)del.br_blockcount), 0,
5611                                         XFS_QMOPT_RES_RTBLKS);
5612                         } else {
5613                                 xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS,
5614                                                 (int64_t)del.br_blockcount, rsvd);
5615                                 (void)XFS_TRANS_RESERVE_QUOTA_NBLKS(mp,
5616                                         NULL, ip, -((long)del.br_blockcount), 0,
5617                                         XFS_QMOPT_RES_REGBLKS);
5618                         }
5619                         ip->i_delayed_blks -= del.br_blockcount;
5620                         if (cur)
5621                                 cur->bc_private.b.flags |=
5622                                         XFS_BTCUR_BPRV_WASDEL;
5623                 } else if (cur)
5624                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5625                 /*
5626                  * If it's the case where the directory code is running
5627                  * with no block reservation, and the deleted block is in
5628                  * the middle of its extent, and the resulting insert
5629                  * of an extent would cause transformation to btree format,
5630                  * then reject it.  The calling code will then swap
5631                  * blocks around instead.
5632                  * We have to do this now, rather than waiting for the
5633                  * conversion to btree format, since the transaction
5634                  * will be dirty.
5635                  */
5636                 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5637                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5638                     XFS_IFORK_NEXTENTS(ip, whichfork) >= ifp->if_ext_max &&
5639                     del.br_startoff > got.br_startoff &&
5640                     del.br_startoff + del.br_blockcount <
5641                     got.br_startoff + got.br_blockcount) {
5642                         error = XFS_ERROR(ENOSPC);
5643                         goto error0;
5644                 }
5645                 error = xfs_bmap_del_extent(ip, tp, lastx, flist, cur, &del,
5646                                 &tmp_logflags, delta, whichfork, rsvd);
5647                 logflags |= tmp_logflags;
5648                 if (error)
5649                         goto error0;
5650                 bno = del.br_startoff - 1;
5651 nodelete:
5652                 lastx = ifp->if_lastex;
5653                 /*
5654                  * If not done go on to the next (previous) record.
5655                  * Reset ep in case the extents array was re-alloced.
5656                  */
5657                 ep = xfs_iext_get_ext(ifp, lastx);
5658                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5659                         if (lastx >= XFS_IFORK_NEXTENTS(ip, whichfork) ||
5660                             xfs_bmbt_get_startoff(ep) > bno) {
5661                                 if (--lastx >= 0)
5662                                         ep = xfs_iext_get_ext(ifp, lastx);
5663                         }
5664                         if (lastx >= 0)
5665                                 xfs_bmbt_get_all(ep, &got);
5666                         extno++;
5667                 }
5668         }
5669         ifp->if_lastex = lastx;
5670         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5671         ASSERT(ifp->if_ext_max ==
5672                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5673         /*
5674          * Convert to a btree if necessary.
5675          */
5676         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5677             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
5678                 ASSERT(cur == NULL);
5679                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5680                         &cur, 0, &tmp_logflags, whichfork);
5681                 logflags |= tmp_logflags;
5682                 if (error)
5683                         goto error0;
5684         }
5685         /*
5686          * transform from btree to extents, give it cur
5687          */
5688         else if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5689                  XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5690                 ASSERT(cur != NULL);
5691                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5692                         whichfork);
5693                 logflags |= tmp_logflags;
5694                 if (error)
5695                         goto error0;
5696         }
5697         /*
5698          * transform from extents to local?
5699          */
5700         ASSERT(ifp->if_ext_max ==
5701                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5702         error = 0;
5703         if (delta && delta->xed_startoff != NULLFILEOFF) {
5704                 /* A change was actually made.
5705                  * Note that delta->xed_blockount is an offset at this
5706                  * point and needs to be converted to a block count.
5707                  */
5708                 ASSERT(delta->xed_blockcount > delta->xed_startoff);
5709                 delta->xed_blockcount -= delta->xed_startoff;
5710         }
5711 error0:
5712         /*
5713          * Log everything.  Do this after conversion, there's no point in
5714          * logging the extent records if we've converted to btree format.
5715          */
5716         if ((logflags & XFS_ILOG_FEXT(whichfork)) &&
5717             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5718                 logflags &= ~XFS_ILOG_FEXT(whichfork);
5719         else if ((logflags & XFS_ILOG_FBROOT(whichfork)) &&
5720                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5721                 logflags &= ~XFS_ILOG_FBROOT(whichfork);
5722         /*
5723          * Log inode even in the error case, if the transaction
5724          * is dirty we'll need to shut down the filesystem.
5725          */
5726         if (logflags)
5727                 xfs_trans_log_inode(tp, ip, logflags);
5728         if (cur) {
5729                 if (!error) {
5730                         *firstblock = cur->bc_private.b.firstblock;
5731                         cur->bc_private.b.allocated = 0;
5732                 }
5733                 xfs_btree_del_cursor(cur,
5734                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5735         }
5736         return error;
5737 }
5738
5739 /*
5740  * returns 1 for success, 0 if we failed to map the extent.
5741  */
5742 STATIC int
5743 xfs_getbmapx_fix_eof_hole(
5744         xfs_inode_t             *ip,            /* xfs incore inode pointer */
5745         struct getbmap          *out,           /* output structure */
5746         int                     prealloced,     /* this is a file with
5747                                                 * preallocated data space */
5748         __int64_t               end,            /* last block requested */
5749         xfs_fsblock_t           startblock)
5750 {
5751         __int64_t               fixlen;
5752         xfs_mount_t             *mp;            /* file system mount point */
5753
5754         if (startblock == HOLESTARTBLOCK) {
5755                 mp = ip->i_mount;
5756                 out->bmv_block = -1;
5757                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, ip->i_size));
5758                 fixlen -= out->bmv_offset;
5759                 if (prealloced && out->bmv_offset + out->bmv_length == end) {
5760                         /* Came to hole at EOF. Trim it. */
5761                         if (fixlen <= 0)
5762                                 return 0;
5763                         out->bmv_length = fixlen;
5764                 }
5765         } else {
5766                 out->bmv_block = XFS_FSB_TO_DB(ip, startblock);
5767         }
5768
5769         return 1;
5770 }
5771
5772 /*
5773  * Fcntl interface to xfs_bmapi.
5774  */
5775 int                                             /* error code */
5776 xfs_getbmap(
5777         xfs_inode_t             *ip,
5778         struct getbmap          *bmv,           /* user bmap structure */
5779         void                    __user *ap,     /* pointer to user's array */
5780         int                     interface)      /* interface flags */
5781 {
5782         __int64_t               bmvend;         /* last block requested */
5783         int                     error;          /* return value */
5784         __int64_t               fixlen;         /* length for -1 case */
5785         int                     i;              /* extent number */
5786         int                     lock;           /* lock state */
5787         xfs_bmbt_irec_t         *map;           /* buffer for user's data */
5788         xfs_mount_t             *mp;            /* file system mount point */
5789         int                     nex;            /* # of user extents can do */
5790         int                     nexleft;        /* # of user extents left */
5791         int                     subnex;         /* # of bmapi's can do */
5792         int                     nmap;           /* number of map entries */
5793         struct getbmap          out;            /* output structure */
5794         int                     whichfork;      /* data or attr fork */
5795         int                     prealloced;     /* this is a file with
5796                                                  * preallocated data space */
5797         int                     sh_unwritten;   /* true, if unwritten */
5798                                                 /* extents listed separately */
5799         int                     bmapi_flags;    /* flags for xfs_bmapi */
5800         __int32_t               oflags;         /* getbmapx bmv_oflags field */
5801
5802         mp = ip->i_mount;
5803
5804         whichfork = interface & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5805         sh_unwritten = (interface & BMV_IF_PREALLOC) != 0;
5806
5807         /*      If the BMV_IF_NO_DMAPI_READ interface bit specified, do not
5808          *      generate a DMAPI read event.  Otherwise, if the DM_EVENT_READ
5809          *      bit is set for the file, generate a read event in order
5810          *      that the DMAPI application may do its thing before we return
5811          *      the extents.  Usually this means restoring user file data to
5812          *      regions of the file that look like holes.
5813          *
5814          *      The "old behavior" (from XFS_IOC_GETBMAP) is to not specify
5815          *      BMV_IF_NO_DMAPI_READ so that read events are generated.
5816          *      If this were not true, callers of ioctl( XFS_IOC_GETBMAP )
5817          *      could misinterpret holes in a DMAPI file as true holes,
5818          *      when in fact they may represent offline user data.
5819          */
5820         if ((interface & BMV_IF_NO_DMAPI_READ) == 0 &&
5821             DM_EVENT_ENABLED(ip, DM_EVENT_READ) &&
5822             whichfork == XFS_DATA_FORK) {
5823                 error = XFS_SEND_DATA(mp, DM_EVENT_READ, ip, 0, 0, 0, NULL);
5824                 if (error)
5825                         return XFS_ERROR(error);
5826         }
5827
5828         if (whichfork == XFS_ATTR_FORK) {
5829                 if (XFS_IFORK_Q(ip)) {
5830                         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5831                             ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5832                             ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5833                                 return XFS_ERROR(EINVAL);
5834                 } else if (unlikely(
5835                            ip->i_d.di_aformat != 0 &&
5836                            ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5837                         XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5838                                          ip->i_mount);
5839                         return XFS_ERROR(EFSCORRUPTED);
5840                 }
5841         } else if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5842                    ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5843                    ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5844                 return XFS_ERROR(EINVAL);
5845         if (whichfork == XFS_DATA_FORK) {
5846                 if (xfs_get_extsz_hint(ip) ||
5847                     ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5848                         prealloced = 1;
5849                         fixlen = XFS_MAXIOFFSET(mp);
5850                 } else {
5851                         prealloced = 0;
5852                         fixlen = ip->i_size;
5853                 }
5854         } else {
5855                 prealloced = 0;
5856                 fixlen = 1LL << 32;
5857         }
5858
5859         if (bmv->bmv_length == -1) {
5860                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5861                 bmv->bmv_length = MAX( (__int64_t)(fixlen - bmv->bmv_offset),
5862                                         (__int64_t)0);
5863         } else if (bmv->bmv_length < 0)
5864                 return XFS_ERROR(EINVAL);
5865         if (bmv->bmv_length == 0) {
5866                 bmv->bmv_entries = 0;
5867                 return 0;
5868         }
5869         nex = bmv->bmv_count - 1;
5870         if (nex <= 0)
5871                 return XFS_ERROR(EINVAL);
5872         bmvend = bmv->bmv_offset + bmv->bmv_length;
5873
5874         xfs_ilock(ip, XFS_IOLOCK_SHARED);
5875
5876         if (whichfork == XFS_DATA_FORK &&
5877                 (ip->i_delayed_blks || ip->i_size > ip->i_d.di_size)) {
5878                 /* xfs_fsize_t last_byte = xfs_file_last_byte(ip); */
5879                 error = xfs_flush_pages(ip, (xfs_off_t)0,
5880                                                -1, 0, FI_REMAPF);
5881                 if (error) {
5882                         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5883                 return error;
5884                 }
5885         }
5886
5887         ASSERT(whichfork == XFS_ATTR_FORK || ip->i_delayed_blks == 0);
5888
5889         lock = xfs_ilock_map_shared(ip);
5890
5891         /*
5892          * Don't let nex be bigger than the number of extents
5893          * we can have assuming alternating holes and real extents.
5894          */
5895         if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5896                 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5897
5898         bmapi_flags = XFS_BMAPI_AFLAG(whichfork) |
5899                         ((sh_unwritten) ? 0 : XFS_BMAPI_IGSTATE);
5900
5901         /*
5902          * Allocate enough space to handle "subnex" maps at a time.
5903          */
5904         subnex = 16;
5905         map = kmem_alloc(subnex * sizeof(*map), KM_SLEEP);
5906
5907         bmv->bmv_entries = 0;
5908
5909         if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0) {
5910                 error = 0;
5911                 goto unlock_and_return;
5912         }
5913
5914         nexleft = nex;
5915
5916         do {
5917                 nmap = (nexleft > subnex) ? subnex : nexleft;
5918                 error = xfs_bmapi(NULL, ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5919                                   XFS_BB_TO_FSB(mp, bmv->bmv_length),
5920                                   bmapi_flags, NULL, 0, map, &nmap,
5921                                   NULL, NULL);
5922                 if (error)
5923                         goto unlock_and_return;
5924                 ASSERT(nmap <= subnex);
5925
5926                 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5927                         nexleft--;
5928                         oflags = (map[i].br_state == XFS_EXT_UNWRITTEN) ?
5929                                         BMV_OF_PREALLOC : 0;
5930                         out.bmv_offset = XFS_FSB_TO_BB(mp, map[i].br_startoff);
5931                         out.bmv_length = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5932                         ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
5933                         if (map[i].br_startblock == HOLESTARTBLOCK &&
5934                             whichfork == XFS_ATTR_FORK) {
5935                                 /* came to the end of attribute fork */
5936                                 goto unlock_and_return;
5937                         } else {
5938                                 if (!xfs_getbmapx_fix_eof_hole(ip, &out,
5939                                                         prealloced, bmvend,
5940                                                         map[i].br_startblock)) {
5941                                         goto unlock_and_return;
5942                                 }
5943
5944                                 /* return either getbmap/getbmapx structure. */
5945                                 if (interface & BMV_IF_EXTENDED) {
5946                                         struct  getbmapx        outx;
5947
5948                                         GETBMAP_CONVERT(out,outx);
5949                                         outx.bmv_oflags = oflags;
5950                                         outx.bmv_unused1 = outx.bmv_unused2 = 0;
5951                                         if (copy_to_user(ap, &outx,
5952                                                         sizeof(outx))) {
5953                                                 error = XFS_ERROR(EFAULT);
5954                                                 goto unlock_and_return;
5955                                         }
5956                                 } else {
5957                                         if (copy_to_user(ap, &out,
5958                                                         sizeof(out))) {
5959                                                 error = XFS_ERROR(EFAULT);
5960                                                 goto unlock_and_return;
5961                                         }
5962                                 }
5963                                 bmv->bmv_offset =
5964                                         out.bmv_offset + out.bmv_length;
5965                                 bmv->bmv_length = MAX((__int64_t)0,
5966                                         (__int64_t)(bmvend - bmv->bmv_offset));
5967                                 bmv->bmv_entries++;
5968                                 ap = (interface & BMV_IF_EXTENDED) ?
5969                                                 (void __user *)
5970                                         ((struct getbmapx __user *)ap + 1) :
5971                                                 (void __user *)
5972                                         ((struct getbmap __user *)ap + 1);
5973                         }
5974                 }
5975         } while (nmap && nexleft && bmv->bmv_length);
5976
5977 unlock_and_return:
5978         xfs_iunlock_map_shared(ip, lock);
5979         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5980
5981         kmem_free(map);
5982
5983         return error;
5984 }
5985
5986 /*
5987  * Check the last inode extent to determine whether this allocation will result
5988  * in blocks being allocated at the end of the file. When we allocate new data
5989  * blocks at the end of the file which do not start at the previous data block,
5990  * we will try to align the new blocks at stripe unit boundaries.
5991  */
5992 STATIC int                              /* error */
5993 xfs_bmap_isaeof(
5994         xfs_inode_t     *ip,            /* incore inode pointer */
5995         xfs_fileoff_t   off,            /* file offset in fsblocks */
5996         int             whichfork,      /* data or attribute fork */
5997         char            *aeof)          /* return value */
5998 {
5999         int             error;          /* error return value */
6000         xfs_ifork_t     *ifp;           /* inode fork pointer */
6001         xfs_bmbt_rec_host_t *lastrec;   /* extent record pointer */
6002         xfs_extnum_t    nextents;       /* number of file extents */
6003         xfs_bmbt_irec_t s;              /* expanded extent record */
6004
6005         ASSERT(whichfork == XFS_DATA_FORK);
6006         ifp = XFS_IFORK_PTR(ip, whichfork);
6007         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
6008             (error = xfs_iread_extents(NULL, ip, whichfork)))
6009                 return error;
6010         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
6011         if (nextents == 0) {
6012                 *aeof = 1;
6013                 return 0;
6014         }
6015         /*
6016          * Go to the last extent
6017          */
6018         lastrec = xfs_iext_get_ext(ifp, nextents - 1);
6019         xfs_bmbt_get_all(lastrec, &s);
6020         /*
6021          * Check we are allocating in the last extent (for delayed allocations)
6022          * or past the last extent for non-delayed allocations.
6023          */
6024         *aeof = (off >= s.br_startoff &&
6025                  off < s.br_startoff + s.br_blockcount &&
6026                  ISNULLSTARTBLOCK(s.br_startblock)) ||
6027                 off >= s.br_startoff + s.br_blockcount;
6028         return 0;
6029 }
6030
6031 /*
6032  * Check if the endoff is outside the last extent. If so the caller will grow
6033  * the allocation to a stripe unit boundary.
6034  */
6035 int                                     /* error */
6036 xfs_bmap_eof(
6037         xfs_inode_t     *ip,            /* incore inode pointer */
6038         xfs_fileoff_t   endoff,         /* file offset in fsblocks */
6039         int             whichfork,      /* data or attribute fork */
6040         int             *eof)           /* result value */
6041 {
6042         xfs_fsblock_t   blockcount;     /* extent block count */
6043         int             error;          /* error return value */
6044         xfs_ifork_t     *ifp;           /* inode fork pointer */
6045         xfs_bmbt_rec_host_t *lastrec;   /* extent record pointer */
6046         xfs_extnum_t    nextents;       /* number of file extents */
6047         xfs_fileoff_t   startoff;       /* extent starting file offset */
6048
6049         ASSERT(whichfork == XFS_DATA_FORK);
6050         ifp = XFS_IFORK_PTR(ip, whichfork);
6051         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
6052             (error = xfs_iread_extents(NULL, ip, whichfork)))
6053                 return error;
6054         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
6055         if (nextents == 0) {
6056                 *eof = 1;
6057                 return 0;
6058         }
6059         /*
6060          * Go to the last extent
6061          */
6062         lastrec = xfs_iext_get_ext(ifp, nextents - 1);
6063         startoff = xfs_bmbt_get_startoff(lastrec);
6064         blockcount = xfs_bmbt_get_blockcount(lastrec);
6065         *eof = endoff >= startoff + blockcount;
6066         return 0;
6067 }
6068
6069 #ifdef DEBUG
6070 STATIC
6071 xfs_buf_t *
6072 xfs_bmap_get_bp(
6073         xfs_btree_cur_t         *cur,
6074         xfs_fsblock_t           bno)
6075 {
6076         int i;
6077         xfs_buf_t *bp;
6078
6079         if (!cur)
6080                 return(NULL);
6081
6082         bp = NULL;
6083         for(i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
6084                 bp = cur->bc_bufs[i];
6085                 if (!bp) break;
6086                 if (XFS_BUF_ADDR(bp) == bno)
6087                         break;  /* Found it */
6088         }
6089         if (i == XFS_BTREE_MAXLEVELS)
6090                 bp = NULL;
6091
6092         if (!bp) { /* Chase down all the log items to see if the bp is there */
6093                 xfs_log_item_chunk_t    *licp;
6094                 xfs_trans_t             *tp;
6095
6096                 tp = cur->bc_tp;
6097                 licp = &tp->t_items;
6098                 while (!bp && licp != NULL) {
6099                         if (xfs_lic_are_all_free(licp)) {
6100                                 licp = licp->lic_next;
6101                                 continue;
6102                         }
6103                         for (i = 0; i < licp->lic_unused; i++) {
6104                                 xfs_log_item_desc_t     *lidp;
6105                                 xfs_log_item_t          *lip;
6106                                 xfs_buf_log_item_t      *bip;
6107                                 xfs_buf_t               *lbp;
6108
6109                                 if (xfs_lic_isfree(licp, i)) {
6110                                         continue;
6111                                 }
6112
6113                                 lidp = xfs_lic_slot(licp, i);
6114                                 lip = lidp->lid_item;
6115                                 if (lip->li_type != XFS_LI_BUF)
6116                                         continue;
6117
6118                                 bip = (xfs_buf_log_item_t *)lip;
6119                                 lbp = bip->bli_buf;
6120
6121                                 if (XFS_BUF_ADDR(lbp) == bno) {
6122                                         bp = lbp;
6123                                         break; /* Found it */
6124                                 }
6125                         }
6126                         licp = licp->lic_next;
6127                 }
6128         }
6129         return(bp);
6130 }
6131
6132 void
6133 xfs_check_block(
6134         xfs_bmbt_block_t        *block,
6135         xfs_mount_t             *mp,
6136         int                     root,
6137         short                   sz)
6138 {
6139         int                     i, j, dmxr;
6140         __be64                  *pp, *thispa;   /* pointer to block address */
6141         xfs_bmbt_key_t          *prevp, *keyp;
6142
6143         ASSERT(be16_to_cpu(block->bb_level) > 0);
6144
6145         prevp = NULL;
6146         for( i = 1; i <= be16_to_cpu(block->bb_numrecs); i++) {
6147                 dmxr = mp->m_bmap_dmxr[0];
6148
6149                 if (root) {
6150                         keyp = XFS_BMAP_BROOT_KEY_ADDR(block, i, sz);
6151                 } else {
6152                         keyp = XFS_BTREE_KEY_ADDR(xfs_bmbt, block, i);
6153                 }
6154
6155                 if (prevp) {
6156                         xfs_btree_check_key(XFS_BTNUM_BMAP, prevp, keyp);
6157                 }
6158                 prevp = keyp;
6159
6160                 /*
6161                  * Compare the block numbers to see if there are dups.
6162                  */
6163
6164                 if (root) {
6165                         pp = XFS_BMAP_BROOT_PTR_ADDR(block, i, sz);
6166                 } else {
6167                         pp = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, i, dmxr);
6168                 }
6169                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
6170                         if (root) {
6171                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(block, j, sz);
6172                         } else {
6173                                 thispa = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, j,
6174                                                             dmxr);
6175                         }
6176                         if (*thispa == *pp) {
6177                                 cmn_err(CE_WARN, "%s: thispa(%d) == pp(%d) %Ld",
6178                                         __func__, j, i,
6179                                         (unsigned long long)be64_to_cpu(*thispa));
6180                                 panic("%s: ptrs are equal in node\n",
6181                                         __func__);
6182                         }
6183                 }
6184         }
6185 }
6186
6187 /*
6188  * Check that the extents for the inode ip are in the right order in all
6189  * btree leaves.
6190  */
6191
6192 STATIC void
6193 xfs_bmap_check_leaf_extents(
6194         xfs_btree_cur_t         *cur,   /* btree cursor or null */
6195         xfs_inode_t             *ip,            /* incore inode pointer */
6196         int                     whichfork)      /* data or attr fork */
6197 {
6198         xfs_bmbt_block_t        *block; /* current btree block */
6199         xfs_fsblock_t           bno;    /* block # of "block" */
6200         xfs_buf_t               *bp;    /* buffer for "block" */
6201         int                     error;  /* error return value */
6202         xfs_extnum_t            i=0, j; /* index into the extents list */
6203         xfs_ifork_t             *ifp;   /* fork structure */
6204         int                     level;  /* btree level, for checking */
6205         xfs_mount_t             *mp;    /* file system mount structure */
6206         __be64                  *pp;    /* pointer to block address */
6207         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
6208         xfs_bmbt_rec_t          last = {0, 0}; /* last extent in prev block */
6209         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
6210         int                     bp_release = 0;
6211
6212         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
6213                 return;
6214         }
6215
6216         bno = NULLFSBLOCK;
6217         mp = ip->i_mount;
6218         ifp = XFS_IFORK_PTR(ip, whichfork);
6219         block = ifp->if_broot;
6220         /*
6221          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6222          */
6223         level = be16_to_cpu(block->bb_level);
6224         ASSERT(level > 0);
6225         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
6226         pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
6227         bno = be64_to_cpu(*pp);
6228
6229         ASSERT(bno != NULLDFSBNO);
6230         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
6231         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
6232
6233         /*
6234          * Go down the tree until leaf level is reached, following the first
6235          * pointer (leftmost) at each level.
6236          */
6237         while (level-- > 0) {
6238                 /* See if buf is in cur first */
6239                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
6240                 if (bp) {
6241                         bp_release = 0;
6242                 } else {
6243                         bp_release = 1;
6244                 }
6245                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
6246                                 XFS_BMAP_BTREE_REF)))
6247                         goto error_norelse;
6248                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6249                 XFS_WANT_CORRUPTED_GOTO(
6250                         XFS_BMAP_SANITY_CHECK(mp, block, level),
6251                         error0);
6252                 if (level == 0)
6253                         break;
6254
6255                 /*
6256                  * Check this block for basic sanity (increasing keys and
6257                  * no duplicate blocks).
6258                  */
6259
6260                 xfs_check_block(block, mp, 0, 0);
6261                 pp = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, 1, mp->m_bmap_dmxr[1]);
6262                 bno = be64_to_cpu(*pp);
6263                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
6264                 if (bp_release) {
6265                         bp_release = 0;
6266                         xfs_trans_brelse(NULL, bp);
6267                 }
6268         }
6269
6270         /*
6271          * Here with bp and block set to the leftmost leaf node in the tree.
6272          */
6273         i = 0;
6274
6275         /*
6276          * Loop over all leaf nodes checking that all extents are in the right order.
6277          */
6278         for (;;) {
6279                 xfs_fsblock_t   nextbno;
6280                 xfs_extnum_t    num_recs;
6281
6282
6283                 num_recs = be16_to_cpu(block->bb_numrecs);
6284
6285                 /*
6286                  * Read-ahead the next leaf block, if any.
6287                  */
6288
6289                 nextbno = be64_to_cpu(block->bb_rightsib);
6290
6291                 /*
6292                  * Check all the extents to make sure they are OK.
6293                  * If we had a previous block, the last entry should
6294                  * conform with the first entry in this one.
6295                  */
6296
6297                 ep = XFS_BTREE_REC_ADDR(xfs_bmbt, block, 1);
6298                 if (i) {
6299                         xfs_btree_check_rec(XFS_BTNUM_BMAP, &last, ep);
6300                 }
6301                 for (j = 1; j < num_recs; j++) {
6302                         nextp = XFS_BTREE_REC_ADDR(xfs_bmbt, block, j + 1);
6303                         xfs_btree_check_rec(XFS_BTNUM_BMAP, ep, nextp);
6304                         ep = nextp;
6305                 }
6306
6307                 last = *ep;
6308                 i += num_recs;
6309                 if (bp_release) {
6310                         bp_release = 0;
6311                         xfs_trans_brelse(NULL, bp);
6312                 }
6313                 bno = nextbno;
6314                 /*
6315                  * If we've reached the end, stop.
6316                  */
6317                 if (bno == NULLFSBLOCK)
6318                         break;
6319
6320                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
6321                 if (bp) {
6322                         bp_release = 0;
6323                 } else {
6324                         bp_release = 1;
6325                 }
6326                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
6327                                 XFS_BMAP_BTREE_REF)))
6328                         goto error_norelse;
6329                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6330         }
6331         if (bp_release) {
6332                 bp_release = 0;
6333                 xfs_trans_brelse(NULL, bp);
6334         }
6335         return;
6336
6337 error0:
6338         cmn_err(CE_WARN, "%s: at error0", __func__);
6339         if (bp_release)
6340                 xfs_trans_brelse(NULL, bp);
6341 error_norelse:
6342         cmn_err(CE_WARN, "%s: BAD after btree leaves for %d extents",
6343                 __func__, i);
6344         panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
6345         return;
6346 }
6347 #endif
6348
6349 /*
6350  * Count fsblocks of the given fork.
6351  */
6352 int                                             /* error */
6353 xfs_bmap_count_blocks(
6354         xfs_trans_t             *tp,            /* transaction pointer */
6355         xfs_inode_t             *ip,            /* incore inode */
6356         int                     whichfork,      /* data or attr fork */
6357         int                     *count)         /* out: count of blocks */
6358 {
6359         xfs_bmbt_block_t        *block; /* current btree block */
6360         xfs_fsblock_t           bno;    /* block # of "block" */
6361         xfs_ifork_t             *ifp;   /* fork structure */
6362         int                     level;  /* btree level, for checking */
6363         xfs_mount_t             *mp;    /* file system mount structure */
6364         __be64                  *pp;    /* pointer to block address */
6365
6366         bno = NULLFSBLOCK;
6367         mp = ip->i_mount;
6368         ifp = XFS_IFORK_PTR(ip, whichfork);
6369         if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
6370                 xfs_bmap_count_leaves(ifp, 0,
6371                         ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
6372                         count);
6373                 return 0;
6374         }
6375
6376         /*
6377          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6378          */
6379         block = ifp->if_broot;
6380         level = be16_to_cpu(block->bb_level);
6381         ASSERT(level > 0);
6382         pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
6383         bno = be64_to_cpu(*pp);
6384         ASSERT(bno != NULLDFSBNO);
6385         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
6386         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
6387
6388         if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
6389                 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
6390                                  mp);
6391                 return XFS_ERROR(EFSCORRUPTED);
6392         }
6393
6394         return 0;
6395 }
6396
6397 /*
6398  * Recursively walks each level of a btree
6399  * to count total fsblocks is use.
6400  */
6401 STATIC int                                     /* error */
6402 xfs_bmap_count_tree(
6403         xfs_mount_t     *mp,            /* file system mount point */
6404         xfs_trans_t     *tp,            /* transaction pointer */
6405         xfs_ifork_t     *ifp,           /* inode fork pointer */
6406         xfs_fsblock_t   blockno,        /* file system block number */
6407         int             levelin,        /* level in btree */
6408         int             *count)         /* Count of blocks */
6409 {
6410         int                     error;
6411         xfs_buf_t               *bp, *nbp;
6412         int                     level = levelin;
6413         __be64                  *pp;
6414         xfs_fsblock_t           bno = blockno;
6415         xfs_fsblock_t           nextbno;
6416         xfs_bmbt_block_t        *block, *nextblock;
6417         int                     numrecs;
6418
6419         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
6420                 return error;
6421         *count += 1;
6422         block = XFS_BUF_TO_BMBT_BLOCK(bp);
6423
6424         if (--level) {
6425                 /* Not at node above leafs, count this level of nodes */
6426                 nextbno = be64_to_cpu(block->bb_rightsib);
6427                 while (nextbno != NULLFSBLOCK) {
6428                         if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
6429                                 0, &nbp, XFS_BMAP_BTREE_REF)))
6430                                 return error;
6431                         *count += 1;
6432                         nextblock = XFS_BUF_TO_BMBT_BLOCK(nbp);
6433                         nextbno = be64_to_cpu(nextblock->bb_rightsib);
6434                         xfs_trans_brelse(tp, nbp);
6435                 }
6436
6437                 /* Dive to the next level */
6438                 pp = XFS_BTREE_PTR_ADDR(xfs_bmbt, block, 1, mp->m_bmap_dmxr[1]);
6439                 bno = be64_to_cpu(*pp);
6440                 if (unlikely((error =
6441                      xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
6442                         xfs_trans_brelse(tp, bp);
6443                         XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6444                                          XFS_ERRLEVEL_LOW, mp);
6445                         return XFS_ERROR(EFSCORRUPTED);
6446                 }
6447                 xfs_trans_brelse(tp, bp);
6448         } else {
6449                 /* count all level 1 nodes and their leaves */
6450                 for (;;) {
6451                         nextbno = be64_to_cpu(block->bb_rightsib);
6452                         numrecs = be16_to_cpu(block->bb_numrecs);
6453                         xfs_bmap_disk_count_leaves(0, block, numrecs, count);
6454                         xfs_trans_brelse(tp, bp);
6455                         if (nextbno == NULLFSBLOCK)
6456                                 break;
6457                         bno = nextbno;
6458                         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6459                                 XFS_BMAP_BTREE_REF)))
6460                                 return error;
6461                         *count += 1;
6462                         block = XFS_BUF_TO_BMBT_BLOCK(bp);
6463                 }
6464         }
6465         return 0;
6466 }
6467
6468 /*
6469  * Count leaf blocks given a range of extent records.
6470  */
6471 STATIC void
6472 xfs_bmap_count_leaves(
6473         xfs_ifork_t             *ifp,
6474         xfs_extnum_t            idx,
6475         int                     numrecs,
6476         int                     *count)
6477 {
6478         int             b;
6479
6480         for (b = 0; b < numrecs; b++) {
6481                 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
6482                 *count += xfs_bmbt_get_blockcount(frp);
6483         }
6484 }
6485
6486 /*
6487  * Count leaf blocks given a range of extent records originally
6488  * in btree format.
6489  */
6490 STATIC void
6491 xfs_bmap_disk_count_leaves(
6492         xfs_extnum_t            idx,
6493         xfs_bmbt_block_t        *block,
6494         int                     numrecs,
6495         int                     *count)
6496 {
6497         int             b;
6498         xfs_bmbt_rec_t  *frp;
6499
6500         for (b = 1; b <= numrecs; b++) {
6501                 frp = XFS_BTREE_REC_ADDR(xfs_bmbt, block, idx + b);
6502                 *count += xfs_bmbt_disk_get_blockcount(frp);
6503         }
6504 }