blob: fbd48f9b797e32080d739dffa179fbbe27f76abe [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Nathan Scott7b718762005-11-02 14:58:39 +11002 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * 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.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * 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
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110019#include "xfs_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include "xfs_types.h"
Nathan Scotta844f452005-11-02 14:38:42 +110021#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110023#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_trans.h"
25#include "xfs_sb.h"
26#include "xfs_ag.h"
27#include "xfs_dir.h"
28#include "xfs_dir2.h"
Nathan Scotta844f452005-11-02 14:38:42 +110029#include "xfs_da_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110031#include "xfs_alloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include "xfs_ialloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "xfs_dir_sf.h"
34#include "xfs_dir2_sf.h"
Nathan Scotta844f452005-11-02 14:38:42 +110035#include "xfs_attr_sf.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include "xfs_dinode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include "xfs_inode.h"
Nathan Scotta844f452005-11-02 14:38:42 +110038#include "xfs_btree.h"
39#include "xfs_dmapi.h"
40#include "xfs_mount.h"
41#include "xfs_ialloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#include "xfs_itable.h"
Nathan Scotta844f452005-11-02 14:38:42 +110043#include "xfs_inode_item.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include "xfs_extfree_item.h"
45#include "xfs_alloc.h"
46#include "xfs_bmap.h"
47#include "xfs_rtalloc.h"
48#include "xfs_error.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#include "xfs_dir_leaf.h"
Nathan Scottd8cc8902005-11-02 10:34:53 +110050#include "xfs_attr_leaf.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070051#include "xfs_rw.h"
52#include "xfs_quota.h"
53#include "xfs_trans_space.h"
54#include "xfs_buf_item.h"
55
56
57#ifdef DEBUG
58STATIC void
59xfs_bmap_check_leaf_extents(xfs_btree_cur_t *cur, xfs_inode_t *ip, int whichfork);
60#endif
61
62kmem_zone_t *xfs_bmap_free_item_zone;
63
64/*
65 * Prototypes for internal bmap routines.
66 */
67
68
69/*
70 * Called from xfs_bmap_add_attrfork to handle extents format files.
71 */
72STATIC int /* error */
73xfs_bmap_add_attrfork_extents(
74 xfs_trans_t *tp, /* transaction pointer */
75 xfs_inode_t *ip, /* incore inode pointer */
76 xfs_fsblock_t *firstblock, /* first block allocated */
77 xfs_bmap_free_t *flist, /* blocks to free at commit */
78 int *flags); /* inode logging flags */
79
80/*
81 * Called from xfs_bmap_add_attrfork to handle local format files.
82 */
83STATIC int /* error */
84xfs_bmap_add_attrfork_local(
85 xfs_trans_t *tp, /* transaction pointer */
86 xfs_inode_t *ip, /* incore inode pointer */
87 xfs_fsblock_t *firstblock, /* first block allocated */
88 xfs_bmap_free_t *flist, /* blocks to free at commit */
89 int *flags); /* inode logging flags */
90
91/*
92 * Called by xfs_bmapi to update extent list structure and the btree
93 * after allocating space (or doing a delayed allocation).
94 */
95STATIC int /* error */
96xfs_bmap_add_extent(
97 xfs_inode_t *ip, /* incore inode pointer */
98 xfs_extnum_t idx, /* extent number to update/insert */
99 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
100 xfs_bmbt_irec_t *new, /* new data to put in extent list */
101 xfs_fsblock_t *first, /* pointer to firstblock variable */
102 xfs_bmap_free_t *flist, /* list of extents to be freed */
103 int *logflagsp, /* inode logging flags */
104 int whichfork, /* data or attr fork */
105 int rsvd); /* OK to allocate reserved blocks */
106
107/*
108 * Called by xfs_bmap_add_extent to handle cases converting a delayed
109 * allocation to a real allocation.
110 */
111STATIC int /* error */
112xfs_bmap_add_extent_delay_real(
113 xfs_inode_t *ip, /* incore inode pointer */
114 xfs_extnum_t idx, /* extent number to update/insert */
115 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
116 xfs_bmbt_irec_t *new, /* new data to put in extent list */
117 xfs_filblks_t *dnew, /* new delayed-alloc indirect blocks */
118 xfs_fsblock_t *first, /* pointer to firstblock variable */
119 xfs_bmap_free_t *flist, /* list of extents to be freed */
120 int *logflagsp, /* inode logging flags */
121 int rsvd); /* OK to allocate reserved blocks */
122
123/*
124 * Called by xfs_bmap_add_extent to handle cases converting a hole
125 * to a delayed allocation.
126 */
127STATIC int /* error */
128xfs_bmap_add_extent_hole_delay(
129 xfs_inode_t *ip, /* incore inode pointer */
130 xfs_extnum_t idx, /* extent number to update/insert */
131 xfs_btree_cur_t *cur, /* if null, not a btree */
132 xfs_bmbt_irec_t *new, /* new data to put in extent list */
133 int *logflagsp,/* inode logging flags */
134 int rsvd); /* OK to allocate reserved blocks */
135
136/*
137 * Called by xfs_bmap_add_extent to handle cases converting a hole
138 * to a real allocation.
139 */
140STATIC int /* error */
141xfs_bmap_add_extent_hole_real(
142 xfs_inode_t *ip, /* incore inode pointer */
143 xfs_extnum_t idx, /* extent number to update/insert */
144 xfs_btree_cur_t *cur, /* if null, not a btree */
145 xfs_bmbt_irec_t *new, /* new data to put in extent list */
146 int *logflagsp, /* inode logging flags */
147 int whichfork); /* data or attr fork */
148
149/*
150 * Called by xfs_bmap_add_extent to handle cases converting an unwritten
151 * allocation to a real allocation or vice versa.
152 */
153STATIC int /* error */
154xfs_bmap_add_extent_unwritten_real(
155 xfs_inode_t *ip, /* incore inode pointer */
156 xfs_extnum_t idx, /* extent number to update/insert */
157 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
158 xfs_bmbt_irec_t *new, /* new data to put in extent list */
159 int *logflagsp); /* inode logging flags */
160
161/*
162 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
163 * It figures out where to ask the underlying allocator to put the new extent.
164 */
165STATIC int /* error */
166xfs_bmap_alloc(
167 xfs_bmalloca_t *ap); /* bmap alloc argument struct */
168
169/*
170 * Transform a btree format file with only one leaf node, where the
171 * extents list will fit in the inode, into an extents format file.
172 * Since the extent list is already in-core, all we have to do is
173 * give up the space for the btree root and pitch the leaf block.
174 */
175STATIC int /* error */
176xfs_bmap_btree_to_extents(
177 xfs_trans_t *tp, /* transaction pointer */
178 xfs_inode_t *ip, /* incore inode pointer */
179 xfs_btree_cur_t *cur, /* btree cursor */
180 int *logflagsp, /* inode logging flags */
181 int whichfork); /* data or attr fork */
182
183#ifdef DEBUG
184/*
185 * Check that the extents list for the inode ip is in the right order.
186 */
187STATIC void
188xfs_bmap_check_extents(
189 xfs_inode_t *ip, /* incore inode pointer */
190 int whichfork); /* data or attr fork */
191#endif
192
193/*
194 * Called by xfs_bmapi to update extent list structure and the btree
195 * after removing space (or undoing a delayed allocation).
196 */
197STATIC int /* error */
198xfs_bmap_del_extent(
199 xfs_inode_t *ip, /* incore inode pointer */
200 xfs_trans_t *tp, /* current trans pointer */
201 xfs_extnum_t idx, /* extent number to update/insert */
202 xfs_bmap_free_t *flist, /* list of extents to be freed */
203 xfs_btree_cur_t *cur, /* if null, not a btree */
204 xfs_bmbt_irec_t *new, /* new data to put in extent list */
205 int *logflagsp,/* inode logging flags */
206 int whichfork, /* data or attr fork */
207 int rsvd); /* OK to allocate reserved blocks */
208
209/*
210 * Remove the entry "free" from the free item list. Prev points to the
211 * previous entry, unless "free" is the head of the list.
212 */
213STATIC void
214xfs_bmap_del_free(
215 xfs_bmap_free_t *flist, /* free item list header */
216 xfs_bmap_free_item_t *prev, /* previous item on list, if any */
217 xfs_bmap_free_item_t *free); /* list item to be freed */
218
219/*
220 * Remove count entries from the extents array for inode "ip", starting
221 * at index "idx". Copies the remaining items down over the deleted ones,
222 * and gives back the excess memory.
223 */
224STATIC void
225xfs_bmap_delete_exlist(
226 xfs_inode_t *ip, /* incode inode pointer */
227 xfs_extnum_t idx, /* starting delete index */
228 xfs_extnum_t count, /* count of items to delete */
229 int whichfork); /* data or attr fork */
230
231/*
232 * Convert an extents-format file into a btree-format file.
233 * The new file will have a root block (in the inode) and a single child block.
234 */
235STATIC int /* error */
236xfs_bmap_extents_to_btree(
237 xfs_trans_t *tp, /* transaction pointer */
238 xfs_inode_t *ip, /* incore inode pointer */
239 xfs_fsblock_t *firstblock, /* first-block-allocated */
240 xfs_bmap_free_t *flist, /* blocks freed in xaction */
241 xfs_btree_cur_t **curp, /* cursor returned to caller */
242 int wasdel, /* converting a delayed alloc */
243 int *logflagsp, /* inode logging flags */
244 int whichfork); /* data or attr fork */
245
246/*
247 * Insert new item(s) in the extent list for inode "ip".
248 * Count new items are inserted at offset idx.
249 */
250STATIC void
251xfs_bmap_insert_exlist(
252 xfs_inode_t *ip, /* incore inode pointer */
253 xfs_extnum_t idx, /* starting index of new items */
254 xfs_extnum_t count, /* number of inserted items */
255 xfs_bmbt_irec_t *new, /* items to insert */
256 int whichfork); /* data or attr fork */
257
258/*
259 * Convert a local file to an extents file.
260 * This code is sort of bogus, since the file data needs to get
261 * logged so it won't be lost. The bmap-level manipulations are ok, though.
262 */
263STATIC int /* error */
264xfs_bmap_local_to_extents(
265 xfs_trans_t *tp, /* transaction pointer */
266 xfs_inode_t *ip, /* incore inode pointer */
267 xfs_fsblock_t *firstblock, /* first block allocated in xaction */
268 xfs_extlen_t total, /* total blocks needed by transaction */
269 int *logflagsp, /* inode logging flags */
270 int whichfork); /* data or attr fork */
271
272/*
273 * Search the extents list for the inode, for the extent containing bno.
274 * If bno lies in a hole, point to the next entry. If bno lies past eof,
275 * *eofp will be set, and *prevp will contain the last entry (null if none).
276 * Else, *lastxp will be set to the index of the found
277 * entry; *gotp will contain the entry.
278 */
279STATIC xfs_bmbt_rec_t * /* pointer to found extent entry */
280xfs_bmap_search_extents(
281 xfs_inode_t *ip, /* incore inode pointer */
282 xfs_fileoff_t bno, /* block number searched for */
283 int whichfork, /* data or attr fork */
284 int *eofp, /* out: end of file found */
285 xfs_extnum_t *lastxp, /* out: last extent index */
286 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
287 xfs_bmbt_irec_t *prevp); /* out: previous extent entry found */
288
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000289/*
290 * Check the last inode extent to determine whether this allocation will result
291 * in blocks being allocated at the end of the file. When we allocate new data
292 * blocks at the end of the file which do not start at the previous data block,
293 * we will try to align the new blocks at stripe unit boundaries.
294 */
295STATIC int /* error */
296xfs_bmap_isaeof(
297 xfs_inode_t *ip, /* incore inode pointer */
298 xfs_fileoff_t off, /* file offset in fsblocks */
299 int whichfork, /* data or attribute fork */
300 char *aeof); /* return value */
301
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302#ifdef XFS_BMAP_TRACE
303/*
304 * Add a bmap trace buffer entry. Base routine for the others.
305 */
306STATIC void
307xfs_bmap_trace_addentry(
308 int opcode, /* operation */
309 char *fname, /* function name */
310 char *desc, /* operation description */
311 xfs_inode_t *ip, /* incore inode pointer */
312 xfs_extnum_t idx, /* index of entry(ies) */
313 xfs_extnum_t cnt, /* count of entries, 1 or 2 */
314 xfs_bmbt_rec_t *r1, /* first record */
315 xfs_bmbt_rec_t *r2, /* second record or null */
316 int whichfork); /* data or attr fork */
317
318/*
319 * Add bmap trace entry prior to a call to xfs_bmap_delete_exlist.
320 */
321STATIC void
322xfs_bmap_trace_delete(
323 char *fname, /* function name */
324 char *desc, /* operation description */
325 xfs_inode_t *ip, /* incore inode pointer */
326 xfs_extnum_t idx, /* index of entry(entries) deleted */
327 xfs_extnum_t cnt, /* count of entries deleted, 1 or 2 */
328 int whichfork); /* data or attr fork */
329
330/*
331 * Add bmap trace entry prior to a call to xfs_bmap_insert_exlist, or
332 * reading in the extents list from the disk (in the btree).
333 */
334STATIC void
335xfs_bmap_trace_insert(
336 char *fname, /* function name */
337 char *desc, /* operation description */
338 xfs_inode_t *ip, /* incore inode pointer */
339 xfs_extnum_t idx, /* index of entry(entries) inserted */
340 xfs_extnum_t cnt, /* count of entries inserted, 1 or 2 */
341 xfs_bmbt_irec_t *r1, /* inserted record 1 */
342 xfs_bmbt_irec_t *r2, /* inserted record 2 or null */
343 int whichfork); /* data or attr fork */
344
345/*
346 * Add bmap trace entry after updating an extent list entry in place.
347 */
348STATIC void
349xfs_bmap_trace_post_update(
350 char *fname, /* function name */
351 char *desc, /* operation description */
352 xfs_inode_t *ip, /* incore inode pointer */
353 xfs_extnum_t idx, /* index of entry updated */
354 int whichfork); /* data or attr fork */
355
356/*
357 * Add bmap trace entry prior to updating an extent list entry in place.
358 */
359STATIC void
360xfs_bmap_trace_pre_update(
361 char *fname, /* function name */
362 char *desc, /* operation description */
363 xfs_inode_t *ip, /* incore inode pointer */
364 xfs_extnum_t idx, /* index of entry to be updated */
365 int whichfork); /* data or attr fork */
366
367#else
368#define xfs_bmap_trace_delete(f,d,ip,i,c,w)
369#define xfs_bmap_trace_insert(f,d,ip,i,c,r1,r2,w)
370#define xfs_bmap_trace_post_update(f,d,ip,i,w)
371#define xfs_bmap_trace_pre_update(f,d,ip,i,w)
372#endif /* XFS_BMAP_TRACE */
373
374/*
375 * Compute the worst-case number of indirect blocks that will be used
376 * for ip's delayed extent of length "len".
377 */
378STATIC xfs_filblks_t
379xfs_bmap_worst_indlen(
380 xfs_inode_t *ip, /* incore inode pointer */
381 xfs_filblks_t len); /* delayed extent length */
382
383#ifdef DEBUG
384/*
385 * Perform various validation checks on the values being returned
386 * from xfs_bmapi().
387 */
388STATIC void
389xfs_bmap_validate_ret(
390 xfs_fileoff_t bno,
391 xfs_filblks_t len,
392 int flags,
393 xfs_bmbt_irec_t *mval,
394 int nmap,
395 int ret_nmap);
396#else
397#define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
398#endif /* DEBUG */
399
400#if defined(XFS_RW_TRACE)
401STATIC void
402xfs_bunmap_trace(
403 xfs_inode_t *ip,
404 xfs_fileoff_t bno,
405 xfs_filblks_t len,
406 int flags,
407 inst_t *ra);
408#else
409#define xfs_bunmap_trace(ip, bno, len, flags, ra)
410#endif /* XFS_RW_TRACE */
411
412STATIC int
413xfs_bmap_count_tree(
414 xfs_mount_t *mp,
415 xfs_trans_t *tp,
416 xfs_fsblock_t blockno,
417 int levelin,
418 int *count);
419
420STATIC int
421xfs_bmap_count_leaves(
422 xfs_bmbt_rec_t *frp,
423 int numrecs,
424 int *count);
425
426/*
427 * Bmap internal routines.
428 */
429
430/*
431 * Called from xfs_bmap_add_attrfork to handle btree format files.
432 */
433STATIC int /* error */
434xfs_bmap_add_attrfork_btree(
435 xfs_trans_t *tp, /* transaction pointer */
436 xfs_inode_t *ip, /* incore inode pointer */
437 xfs_fsblock_t *firstblock, /* first block allocated */
438 xfs_bmap_free_t *flist, /* blocks to free at commit */
439 int *flags) /* inode logging flags */
440{
441 xfs_btree_cur_t *cur; /* btree cursor */
442 int error; /* error return value */
443 xfs_mount_t *mp; /* file system mount struct */
444 int stat; /* newroot status */
445
446 mp = ip->i_mount;
447 if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
448 *flags |= XFS_ILOG_DBROOT;
449 else {
450 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
451 XFS_DATA_FORK);
452 cur->bc_private.b.flist = flist;
453 cur->bc_private.b.firstblock = *firstblock;
454 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
455 goto error0;
456 ASSERT(stat == 1); /* must be at least one entry */
457 if ((error = xfs_bmbt_newroot(cur, flags, &stat)))
458 goto error0;
459 if (stat == 0) {
460 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
461 return XFS_ERROR(ENOSPC);
462 }
463 *firstblock = cur->bc_private.b.firstblock;
464 cur->bc_private.b.allocated = 0;
465 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
466 }
467 return 0;
468error0:
469 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
470 return error;
471}
472
473/*
474 * Called from xfs_bmap_add_attrfork to handle extents format files.
475 */
476STATIC int /* error */
477xfs_bmap_add_attrfork_extents(
478 xfs_trans_t *tp, /* transaction pointer */
479 xfs_inode_t *ip, /* incore inode pointer */
480 xfs_fsblock_t *firstblock, /* first block allocated */
481 xfs_bmap_free_t *flist, /* blocks to free at commit */
482 int *flags) /* inode logging flags */
483{
484 xfs_btree_cur_t *cur; /* bmap btree cursor */
485 int error; /* error return value */
486
487 if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
488 return 0;
489 cur = NULL;
490 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
491 flags, XFS_DATA_FORK);
492 if (cur) {
493 cur->bc_private.b.allocated = 0;
494 xfs_btree_del_cursor(cur,
495 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
496 }
497 return error;
498}
499
500/*
501 * Called from xfs_bmap_add_attrfork to handle local format files.
502 */
503STATIC int /* error */
504xfs_bmap_add_attrfork_local(
505 xfs_trans_t *tp, /* transaction pointer */
506 xfs_inode_t *ip, /* incore inode pointer */
507 xfs_fsblock_t *firstblock, /* first block allocated */
508 xfs_bmap_free_t *flist, /* blocks to free at commit */
509 int *flags) /* inode logging flags */
510{
511 xfs_da_args_t dargs; /* args for dir/attr code */
512 int error; /* error return value */
513 xfs_mount_t *mp; /* mount structure pointer */
514
515 if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
516 return 0;
517 if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
518 mp = ip->i_mount;
519 memset(&dargs, 0, sizeof(dargs));
520 dargs.dp = ip;
521 dargs.firstblock = firstblock;
522 dargs.flist = flist;
523 dargs.total = mp->m_dirblkfsbs;
524 dargs.whichfork = XFS_DATA_FORK;
525 dargs.trans = tp;
526 error = XFS_DIR_SHORTFORM_TO_SINGLE(mp, &dargs);
527 } else
528 error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
529 XFS_DATA_FORK);
530 return error;
531}
532
533/*
534 * Called by xfs_bmapi to update extent list structure and the btree
535 * after allocating space (or doing a delayed allocation).
536 */
537STATIC int /* error */
538xfs_bmap_add_extent(
539 xfs_inode_t *ip, /* incore inode pointer */
540 xfs_extnum_t idx, /* extent number to update/insert */
541 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
542 xfs_bmbt_irec_t *new, /* new data to put in extent list */
543 xfs_fsblock_t *first, /* pointer to firstblock variable */
544 xfs_bmap_free_t *flist, /* list of extents to be freed */
545 int *logflagsp, /* inode logging flags */
546 int whichfork, /* data or attr fork */
547 int rsvd) /* OK to use reserved data blocks */
548{
549 xfs_btree_cur_t *cur; /* btree cursor or null */
550 xfs_filblks_t da_new; /* new count del alloc blocks used */
551 xfs_filblks_t da_old; /* old count del alloc blocks used */
552 int error; /* error return value */
553#ifdef XFS_BMAP_TRACE
554 static char fname[] = "xfs_bmap_add_extent";
555#endif
556 xfs_ifork_t *ifp; /* inode fork ptr */
557 int logflags; /* returned value */
558 xfs_extnum_t nextents; /* number of extents in file now */
559
560 XFS_STATS_INC(xs_add_exlist);
561 cur = *curp;
562 ifp = XFS_IFORK_PTR(ip, whichfork);
563 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
564 ASSERT(idx <= nextents);
565 da_old = da_new = 0;
566 error = 0;
567 /*
568 * This is the first extent added to a new/empty file.
569 * Special case this one, so other routines get to assume there are
570 * already extents in the list.
571 */
572 if (nextents == 0) {
573 xfs_bmap_trace_insert(fname, "insert empty", ip, 0, 1, new,
574 NULL, whichfork);
575 xfs_bmap_insert_exlist(ip, 0, 1, new, whichfork);
576 ASSERT(cur == NULL);
577 ifp->if_lastex = 0;
578 if (!ISNULLSTARTBLOCK(new->br_startblock)) {
579 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
580 logflags = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
581 } else
582 logflags = 0;
583 }
584 /*
585 * Any kind of new delayed allocation goes here.
586 */
587 else if (ISNULLSTARTBLOCK(new->br_startblock)) {
588 if (cur)
589 ASSERT((cur->bc_private.b.flags &
590 XFS_BTCUR_BPRV_WASDEL) == 0);
591 if ((error = xfs_bmap_add_extent_hole_delay(ip, idx, cur, new,
592 &logflags, rsvd)))
593 goto done;
594 }
595 /*
596 * Real allocation off the end of the file.
597 */
598 else if (idx == nextents) {
599 if (cur)
600 ASSERT((cur->bc_private.b.flags &
601 XFS_BTCUR_BPRV_WASDEL) == 0);
602 if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur, new,
603 &logflags, whichfork)))
604 goto done;
605 } else {
606 xfs_bmbt_irec_t prev; /* old extent at offset idx */
607
608 /*
609 * Get the record referred to by idx.
610 */
611 xfs_bmbt_get_all(&ifp->if_u1.if_extents[idx], &prev);
612 /*
613 * If it's a real allocation record, and the new allocation ends
614 * after the start of the referred to record, then we're filling
615 * in a delayed or unwritten allocation with a real one, or
616 * converting real back to unwritten.
617 */
618 if (!ISNULLSTARTBLOCK(new->br_startblock) &&
619 new->br_startoff + new->br_blockcount > prev.br_startoff) {
620 if (prev.br_state != XFS_EXT_UNWRITTEN &&
621 ISNULLSTARTBLOCK(prev.br_startblock)) {
622 da_old = STARTBLOCKVAL(prev.br_startblock);
623 if (cur)
624 ASSERT(cur->bc_private.b.flags &
625 XFS_BTCUR_BPRV_WASDEL);
626 if ((error = xfs_bmap_add_extent_delay_real(ip,
627 idx, &cur, new, &da_new, first, flist,
628 &logflags, rsvd)))
629 goto done;
630 } else if (new->br_state == XFS_EXT_NORM) {
631 ASSERT(new->br_state == XFS_EXT_NORM);
632 if ((error = xfs_bmap_add_extent_unwritten_real(
633 ip, idx, &cur, new, &logflags)))
634 goto done;
635 } else {
636 ASSERT(new->br_state == XFS_EXT_UNWRITTEN);
637 if ((error = xfs_bmap_add_extent_unwritten_real(
638 ip, idx, &cur, new, &logflags)))
639 goto done;
640 }
641 ASSERT(*curp == cur || *curp == NULL);
642 }
643 /*
644 * Otherwise we're filling in a hole with an allocation.
645 */
646 else {
647 if (cur)
648 ASSERT((cur->bc_private.b.flags &
649 XFS_BTCUR_BPRV_WASDEL) == 0);
650 if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur,
651 new, &logflags, whichfork)))
652 goto done;
653 }
654 }
655
656 ASSERT(*curp == cur || *curp == NULL);
657 /*
658 * Convert to a btree if necessary.
659 */
660 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
661 XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
662 int tmp_logflags; /* partial log flag return val */
663
664 ASSERT(cur == NULL);
665 error = xfs_bmap_extents_to_btree(ip->i_transp, ip, first,
666 flist, &cur, da_old > 0, &tmp_logflags, whichfork);
667 logflags |= tmp_logflags;
668 if (error)
669 goto done;
670 }
671 /*
672 * Adjust for changes in reserved delayed indirect blocks.
673 * Nothing to do for disk quotas here.
674 */
675 if (da_old || da_new) {
676 xfs_filblks_t nblks;
677
678 nblks = da_new;
679 if (cur)
680 nblks += cur->bc_private.b.allocated;
681 ASSERT(nblks <= da_old);
682 if (nblks < da_old)
683 xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
684 (int)(da_old - nblks), rsvd);
685 }
686 /*
687 * Clear out the allocated field, done with it now in any case.
688 */
689 if (cur) {
690 cur->bc_private.b.allocated = 0;
691 *curp = cur;
692 }
693done:
694#ifdef DEBUG
695 if (!error)
696 xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
697#endif
698 *logflagsp = logflags;
699 return error;
700}
701
702/*
703 * Called by xfs_bmap_add_extent to handle cases converting a delayed
704 * allocation to a real allocation.
705 */
706STATIC int /* error */
707xfs_bmap_add_extent_delay_real(
708 xfs_inode_t *ip, /* incore inode pointer */
709 xfs_extnum_t idx, /* extent number to update/insert */
710 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
711 xfs_bmbt_irec_t *new, /* new data to put in extent list */
712 xfs_filblks_t *dnew, /* new delayed-alloc indirect blocks */
713 xfs_fsblock_t *first, /* pointer to firstblock variable */
714 xfs_bmap_free_t *flist, /* list of extents to be freed */
715 int *logflagsp, /* inode logging flags */
716 int rsvd) /* OK to use reserved data block allocation */
717{
718 xfs_bmbt_rec_t *base; /* base of extent entry list */
719 xfs_btree_cur_t *cur; /* btree cursor */
720 int diff; /* temp value */
721 xfs_bmbt_rec_t *ep; /* extent entry for idx */
722 int error; /* error return value */
723#ifdef XFS_BMAP_TRACE
724 static char fname[] = "xfs_bmap_add_extent_delay_real";
725#endif
726 int i; /* temp state */
727 xfs_fileoff_t new_endoff; /* end offset of new entry */
728 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
729 /* left is 0, right is 1, prev is 2 */
730 int rval=0; /* return value (logging flags) */
731 int state = 0;/* state bits, accessed thru macros */
732 xfs_filblks_t temp; /* value for dnew calculations */
733 xfs_filblks_t temp2; /* value for dnew calculations */
734 int tmp_rval; /* partial logging flags */
735 enum { /* bit number definitions for state */
736 LEFT_CONTIG, RIGHT_CONTIG,
737 LEFT_FILLING, RIGHT_FILLING,
738 LEFT_DELAY, RIGHT_DELAY,
739 LEFT_VALID, RIGHT_VALID
740 };
741
742#define LEFT r[0]
743#define RIGHT r[1]
744#define PREV r[2]
745#define MASK(b) (1 << (b))
746#define MASK2(a,b) (MASK(a) | MASK(b))
747#define MASK3(a,b,c) (MASK2(a,b) | MASK(c))
748#define MASK4(a,b,c,d) (MASK3(a,b,c) | MASK(d))
749#define STATE_SET(b,v) ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
750#define STATE_TEST(b) (state & MASK(b))
751#define STATE_SET_TEST(b,v) ((v) ? ((state |= MASK(b)), 1) : \
752 ((state &= ~MASK(b)), 0))
753#define SWITCH_STATE \
754 (state & MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG))
755
756 /*
757 * Set up a bunch of variables to make the tests simpler.
758 */
759 cur = *curp;
760 base = ip->i_df.if_u1.if_extents;
761 ep = &base[idx];
762 xfs_bmbt_get_all(ep, &PREV);
763 new_endoff = new->br_startoff + new->br_blockcount;
764 ASSERT(PREV.br_startoff <= new->br_startoff);
765 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
766 /*
767 * Set flags determining what part of the previous delayed allocation
768 * extent is being replaced by a real allocation.
769 */
770 STATE_SET(LEFT_FILLING, PREV.br_startoff == new->br_startoff);
771 STATE_SET(RIGHT_FILLING,
772 PREV.br_startoff + PREV.br_blockcount == new_endoff);
773 /*
774 * Check and set flags if this segment has a left neighbor.
775 * Don't set contiguous if the combined extent would be too large.
776 */
777 if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
778 xfs_bmbt_get_all(ep - 1, &LEFT);
779 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(LEFT.br_startblock));
780 }
781 STATE_SET(LEFT_CONTIG,
782 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
783 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
784 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
785 LEFT.br_state == new->br_state &&
786 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN);
787 /*
788 * Check and set flags if this segment has a right neighbor.
789 * Don't set contiguous if the combined extent would be too large.
790 * Also check for all-three-contiguous being too large.
791 */
792 if (STATE_SET_TEST(RIGHT_VALID,
793 idx <
794 ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1)) {
795 xfs_bmbt_get_all(ep + 1, &RIGHT);
796 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(RIGHT.br_startblock));
797 }
798 STATE_SET(RIGHT_CONTIG,
799 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
800 new_endoff == RIGHT.br_startoff &&
801 new->br_startblock + new->br_blockcount ==
802 RIGHT.br_startblock &&
803 new->br_state == RIGHT.br_state &&
804 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
805 ((state & MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING)) !=
806 MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING) ||
807 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
808 <= MAXEXTLEN));
809 error = 0;
810 /*
811 * Switch out based on the FILLING and CONTIG state bits.
812 */
813 switch (SWITCH_STATE) {
814
815 case MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
816 /*
817 * Filling in all of a previously delayed allocation extent.
818 * The left and right neighbors are both contiguous with new.
819 */
820 xfs_bmap_trace_pre_update(fname, "LF|RF|LC|RC", ip, idx - 1,
821 XFS_DATA_FORK);
822 xfs_bmbt_set_blockcount(ep - 1,
823 LEFT.br_blockcount + PREV.br_blockcount +
824 RIGHT.br_blockcount);
825 xfs_bmap_trace_post_update(fname, "LF|RF|LC|RC", ip, idx - 1,
826 XFS_DATA_FORK);
827 xfs_bmap_trace_delete(fname, "LF|RF|LC|RC", ip, idx, 2,
828 XFS_DATA_FORK);
829 xfs_bmap_delete_exlist(ip, idx, 2, XFS_DATA_FORK);
830 ip->i_df.if_lastex = idx - 1;
831 ip->i_d.di_nextents--;
832 if (cur == NULL)
833 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
834 else {
835 rval = XFS_ILOG_CORE;
836 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
837 RIGHT.br_startblock,
838 RIGHT.br_blockcount, &i)))
839 goto done;
840 ASSERT(i == 1);
841 if ((error = xfs_bmbt_delete(cur, &i)))
842 goto done;
843 ASSERT(i == 1);
844 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
845 goto done;
846 ASSERT(i == 1);
847 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
848 LEFT.br_startblock,
849 LEFT.br_blockcount +
850 PREV.br_blockcount +
851 RIGHT.br_blockcount, LEFT.br_state)))
852 goto done;
853 }
854 *dnew = 0;
855 break;
856
857 case MASK3(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG):
858 /*
859 * Filling in all of a previously delayed allocation extent.
860 * The left neighbor is contiguous, the right is not.
861 */
862 xfs_bmap_trace_pre_update(fname, "LF|RF|LC", ip, idx - 1,
863 XFS_DATA_FORK);
864 xfs_bmbt_set_blockcount(ep - 1,
865 LEFT.br_blockcount + PREV.br_blockcount);
866 xfs_bmap_trace_post_update(fname, "LF|RF|LC", ip, idx - 1,
867 XFS_DATA_FORK);
868 ip->i_df.if_lastex = idx - 1;
869 xfs_bmap_trace_delete(fname, "LF|RF|LC", ip, idx, 1,
870 XFS_DATA_FORK);
871 xfs_bmap_delete_exlist(ip, idx, 1, XFS_DATA_FORK);
872 if (cur == NULL)
873 rval = XFS_ILOG_DEXT;
874 else {
875 rval = 0;
876 if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
877 LEFT.br_startblock, LEFT.br_blockcount,
878 &i)))
879 goto done;
880 ASSERT(i == 1);
881 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
882 LEFT.br_startblock,
883 LEFT.br_blockcount +
884 PREV.br_blockcount, LEFT.br_state)))
885 goto done;
886 }
887 *dnew = 0;
888 break;
889
890 case MASK3(LEFT_FILLING, RIGHT_FILLING, RIGHT_CONTIG):
891 /*
892 * Filling in all of a previously delayed allocation extent.
893 * The right neighbor is contiguous, the left is not.
894 */
895 xfs_bmap_trace_pre_update(fname, "LF|RF|RC", ip, idx,
896 XFS_DATA_FORK);
897 xfs_bmbt_set_startblock(ep, new->br_startblock);
898 xfs_bmbt_set_blockcount(ep,
899 PREV.br_blockcount + RIGHT.br_blockcount);
900 xfs_bmap_trace_post_update(fname, "LF|RF|RC", ip, idx,
901 XFS_DATA_FORK);
902 ip->i_df.if_lastex = idx;
903 xfs_bmap_trace_delete(fname, "LF|RF|RC", ip, idx + 1, 1,
904 XFS_DATA_FORK);
905 xfs_bmap_delete_exlist(ip, idx + 1, 1, XFS_DATA_FORK);
906 if (cur == NULL)
907 rval = XFS_ILOG_DEXT;
908 else {
909 rval = 0;
910 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
911 RIGHT.br_startblock,
912 RIGHT.br_blockcount, &i)))
913 goto done;
914 ASSERT(i == 1);
915 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
916 new->br_startblock,
917 PREV.br_blockcount +
918 RIGHT.br_blockcount, PREV.br_state)))
919 goto done;
920 }
921 *dnew = 0;
922 break;
923
924 case MASK2(LEFT_FILLING, RIGHT_FILLING):
925 /*
926 * Filling in all of a previously delayed allocation extent.
927 * Neither the left nor right neighbors are contiguous with
928 * the new one.
929 */
930 xfs_bmap_trace_pre_update(fname, "LF|RF", ip, idx,
931 XFS_DATA_FORK);
932 xfs_bmbt_set_startblock(ep, new->br_startblock);
933 xfs_bmap_trace_post_update(fname, "LF|RF", ip, idx,
934 XFS_DATA_FORK);
935 ip->i_df.if_lastex = idx;
936 ip->i_d.di_nextents++;
937 if (cur == NULL)
938 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
939 else {
940 rval = XFS_ILOG_CORE;
941 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
942 new->br_startblock, new->br_blockcount,
943 &i)))
944 goto done;
945 ASSERT(i == 0);
946 cur->bc_rec.b.br_state = XFS_EXT_NORM;
947 if ((error = xfs_bmbt_insert(cur, &i)))
948 goto done;
949 ASSERT(i == 1);
950 }
951 *dnew = 0;
952 break;
953
954 case MASK2(LEFT_FILLING, LEFT_CONTIG):
955 /*
956 * Filling in the first part of a previous delayed allocation.
957 * The left neighbor is contiguous.
958 */
959 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx - 1,
960 XFS_DATA_FORK);
961 xfs_bmbt_set_blockcount(ep - 1,
962 LEFT.br_blockcount + new->br_blockcount);
963 xfs_bmbt_set_startoff(ep,
964 PREV.br_startoff + new->br_blockcount);
965 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx - 1,
966 XFS_DATA_FORK);
967 temp = PREV.br_blockcount - new->br_blockcount;
968 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx,
969 XFS_DATA_FORK);
970 xfs_bmbt_set_blockcount(ep, temp);
971 ip->i_df.if_lastex = idx - 1;
972 if (cur == NULL)
973 rval = XFS_ILOG_DEXT;
974 else {
975 rval = 0;
976 if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
977 LEFT.br_startblock, LEFT.br_blockcount,
978 &i)))
979 goto done;
980 ASSERT(i == 1);
981 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
982 LEFT.br_startblock,
983 LEFT.br_blockcount +
984 new->br_blockcount,
985 LEFT.br_state)))
986 goto done;
987 }
988 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
989 STARTBLOCKVAL(PREV.br_startblock));
990 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
991 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx,
992 XFS_DATA_FORK);
993 *dnew = temp;
994 break;
995
996 case MASK(LEFT_FILLING):
997 /*
998 * Filling in the first part of a previous delayed allocation.
999 * The left neighbor is not contiguous.
1000 */
1001 xfs_bmap_trace_pre_update(fname, "LF", ip, idx, XFS_DATA_FORK);
1002 xfs_bmbt_set_startoff(ep, new_endoff);
1003 temp = PREV.br_blockcount - new->br_blockcount;
1004 xfs_bmbt_set_blockcount(ep, temp);
1005 xfs_bmap_trace_insert(fname, "LF", ip, idx, 1, new, NULL,
1006 XFS_DATA_FORK);
1007 xfs_bmap_insert_exlist(ip, idx, 1, new, XFS_DATA_FORK);
1008 ip->i_df.if_lastex = idx;
1009 ip->i_d.di_nextents++;
1010 if (cur == NULL)
1011 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1012 else {
1013 rval = XFS_ILOG_CORE;
1014 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1015 new->br_startblock, new->br_blockcount,
1016 &i)))
1017 goto done;
1018 ASSERT(i == 0);
1019 cur->bc_rec.b.br_state = XFS_EXT_NORM;
1020 if ((error = xfs_bmbt_insert(cur, &i)))
1021 goto done;
1022 ASSERT(i == 1);
1023 }
1024 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1025 ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1026 error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1027 first, flist, &cur, 1, &tmp_rval,
1028 XFS_DATA_FORK);
1029 rval |= tmp_rval;
1030 if (error)
1031 goto done;
1032 }
1033 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1034 STARTBLOCKVAL(PREV.br_startblock) -
1035 (cur ? cur->bc_private.b.allocated : 0));
1036 base = ip->i_df.if_u1.if_extents;
1037 ep = &base[idx + 1];
1038 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1039 xfs_bmap_trace_post_update(fname, "LF", ip, idx + 1,
1040 XFS_DATA_FORK);
1041 *dnew = temp;
1042 break;
1043
1044 case MASK2(RIGHT_FILLING, RIGHT_CONTIG):
1045 /*
1046 * Filling in the last part of a previous delayed allocation.
1047 * The right neighbor is contiguous with the new allocation.
1048 */
1049 temp = PREV.br_blockcount - new->br_blockcount;
1050 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx,
1051 XFS_DATA_FORK);
1052 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx + 1,
1053 XFS_DATA_FORK);
1054 xfs_bmbt_set_blockcount(ep, temp);
1055 xfs_bmbt_set_allf(ep + 1, new->br_startoff, new->br_startblock,
1056 new->br_blockcount + RIGHT.br_blockcount,
1057 RIGHT.br_state);
1058 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx + 1,
1059 XFS_DATA_FORK);
1060 ip->i_df.if_lastex = idx + 1;
1061 if (cur == NULL)
1062 rval = XFS_ILOG_DEXT;
1063 else {
1064 rval = 0;
1065 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1066 RIGHT.br_startblock,
1067 RIGHT.br_blockcount, &i)))
1068 goto done;
1069 ASSERT(i == 1);
1070 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1071 new->br_startblock,
1072 new->br_blockcount +
1073 RIGHT.br_blockcount,
1074 RIGHT.br_state)))
1075 goto done;
1076 }
1077 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1078 STARTBLOCKVAL(PREV.br_startblock));
1079 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1080 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx,
1081 XFS_DATA_FORK);
1082 *dnew = temp;
1083 break;
1084
1085 case MASK(RIGHT_FILLING):
1086 /*
1087 * Filling in the last part of a previous delayed allocation.
1088 * The right neighbor is not contiguous.
1089 */
1090 temp = PREV.br_blockcount - new->br_blockcount;
1091 xfs_bmap_trace_pre_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1092 xfs_bmbt_set_blockcount(ep, temp);
1093 xfs_bmap_trace_insert(fname, "RF", ip, idx + 1, 1,
1094 new, NULL, XFS_DATA_FORK);
1095 xfs_bmap_insert_exlist(ip, idx + 1, 1, new, XFS_DATA_FORK);
1096 ip->i_df.if_lastex = idx + 1;
1097 ip->i_d.di_nextents++;
1098 if (cur == NULL)
1099 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1100 else {
1101 rval = XFS_ILOG_CORE;
1102 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1103 new->br_startblock, new->br_blockcount,
1104 &i)))
1105 goto done;
1106 ASSERT(i == 0);
1107 cur->bc_rec.b.br_state = XFS_EXT_NORM;
1108 if ((error = xfs_bmbt_insert(cur, &i)))
1109 goto done;
1110 ASSERT(i == 1);
1111 }
1112 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1113 ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1114 error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1115 first, flist, &cur, 1, &tmp_rval,
1116 XFS_DATA_FORK);
1117 rval |= tmp_rval;
1118 if (error)
1119 goto done;
1120 }
1121 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1122 STARTBLOCKVAL(PREV.br_startblock) -
1123 (cur ? cur->bc_private.b.allocated : 0));
1124 base = ip->i_df.if_u1.if_extents;
1125 ep = &base[idx];
1126 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1127 xfs_bmap_trace_post_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1128 *dnew = temp;
1129 break;
1130
1131 case 0:
1132 /*
1133 * Filling in the middle part of a previous delayed allocation.
1134 * Contiguity is impossible here.
1135 * This case is avoided almost all the time.
1136 */
1137 temp = new->br_startoff - PREV.br_startoff;
1138 xfs_bmap_trace_pre_update(fname, "0", ip, idx, XFS_DATA_FORK);
1139 xfs_bmbt_set_blockcount(ep, temp);
1140 r[0] = *new;
1141 r[1].br_startoff = new_endoff;
1142 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
1143 r[1].br_blockcount = temp2;
1144 xfs_bmap_trace_insert(fname, "0", ip, idx + 1, 2, &r[0], &r[1],
1145 XFS_DATA_FORK);
1146 xfs_bmap_insert_exlist(ip, idx + 1, 2, &r[0], XFS_DATA_FORK);
1147 ip->i_df.if_lastex = idx + 1;
1148 ip->i_d.di_nextents++;
1149 if (cur == NULL)
1150 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1151 else {
1152 rval = XFS_ILOG_CORE;
1153 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1154 new->br_startblock, new->br_blockcount,
1155 &i)))
1156 goto done;
1157 ASSERT(i == 0);
1158 cur->bc_rec.b.br_state = XFS_EXT_NORM;
1159 if ((error = xfs_bmbt_insert(cur, &i)))
1160 goto done;
1161 ASSERT(i == 1);
1162 }
1163 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1164 ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1165 error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1166 first, flist, &cur, 1, &tmp_rval,
1167 XFS_DATA_FORK);
1168 rval |= tmp_rval;
1169 if (error)
1170 goto done;
1171 }
1172 temp = xfs_bmap_worst_indlen(ip, temp);
1173 temp2 = xfs_bmap_worst_indlen(ip, temp2);
1174 diff = (int)(temp + temp2 - STARTBLOCKVAL(PREV.br_startblock) -
1175 (cur ? cur->bc_private.b.allocated : 0));
1176 if (diff > 0 &&
1177 xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS, -diff, rsvd)) {
1178 /*
1179 * Ick gross gag me with a spoon.
1180 */
1181 ASSERT(0); /* want to see if this ever happens! */
1182 while (diff > 0) {
1183 if (temp) {
1184 temp--;
1185 diff--;
1186 if (!diff ||
1187 !xfs_mod_incore_sb(ip->i_mount,
1188 XFS_SBS_FDBLOCKS, -diff, rsvd))
1189 break;
1190 }
1191 if (temp2) {
1192 temp2--;
1193 diff--;
1194 if (!diff ||
1195 !xfs_mod_incore_sb(ip->i_mount,
1196 XFS_SBS_FDBLOCKS, -diff, rsvd))
1197 break;
1198 }
1199 }
1200 }
1201 base = ip->i_df.if_u1.if_extents;
1202 ep = &base[idx];
1203 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1204 xfs_bmap_trace_post_update(fname, "0", ip, idx, XFS_DATA_FORK);
1205 xfs_bmap_trace_pre_update(fname, "0", ip, idx + 2,
1206 XFS_DATA_FORK);
1207 xfs_bmbt_set_startblock(ep + 2, NULLSTARTBLOCK((int)temp2));
1208 xfs_bmap_trace_post_update(fname, "0", ip, idx + 2,
1209 XFS_DATA_FORK);
1210 *dnew = temp + temp2;
1211 break;
1212
1213 case MASK3(LEFT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1214 case MASK3(RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1215 case MASK2(LEFT_FILLING, RIGHT_CONTIG):
1216 case MASK2(RIGHT_FILLING, LEFT_CONTIG):
1217 case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1218 case MASK(LEFT_CONTIG):
1219 case MASK(RIGHT_CONTIG):
1220 /*
1221 * These cases are all impossible.
1222 */
1223 ASSERT(0);
1224 }
1225 *curp = cur;
1226done:
1227 *logflagsp = rval;
1228 return error;
1229#undef LEFT
1230#undef RIGHT
1231#undef PREV
1232#undef MASK
1233#undef MASK2
1234#undef MASK3
1235#undef MASK4
1236#undef STATE_SET
1237#undef STATE_TEST
1238#undef STATE_SET_TEST
1239#undef SWITCH_STATE
1240}
1241
1242/*
1243 * Called by xfs_bmap_add_extent to handle cases converting an unwritten
1244 * allocation to a real allocation or vice versa.
1245 */
1246STATIC int /* error */
1247xfs_bmap_add_extent_unwritten_real(
1248 xfs_inode_t *ip, /* incore inode pointer */
1249 xfs_extnum_t idx, /* extent number to update/insert */
1250 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
1251 xfs_bmbt_irec_t *new, /* new data to put in extent list */
1252 int *logflagsp) /* inode logging flags */
1253{
1254 xfs_bmbt_rec_t *base; /* base of extent entry list */
1255 xfs_btree_cur_t *cur; /* btree cursor */
1256 xfs_bmbt_rec_t *ep; /* extent entry for idx */
1257 int error; /* error return value */
1258#ifdef XFS_BMAP_TRACE
1259 static char fname[] = "xfs_bmap_add_extent_unwritten_real";
1260#endif
1261 int i; /* temp state */
1262 xfs_fileoff_t new_endoff; /* end offset of new entry */
1263 xfs_exntst_t newext; /* new extent state */
1264 xfs_exntst_t oldext; /* old extent state */
1265 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
1266 /* left is 0, right is 1, prev is 2 */
1267 int rval=0; /* return value (logging flags) */
1268 int state = 0;/* state bits, accessed thru macros */
1269 enum { /* bit number definitions for state */
1270 LEFT_CONTIG, RIGHT_CONTIG,
1271 LEFT_FILLING, RIGHT_FILLING,
1272 LEFT_DELAY, RIGHT_DELAY,
1273 LEFT_VALID, RIGHT_VALID
1274 };
1275
1276#define LEFT r[0]
1277#define RIGHT r[1]
1278#define PREV r[2]
1279#define MASK(b) (1 << (b))
1280#define MASK2(a,b) (MASK(a) | MASK(b))
1281#define MASK3(a,b,c) (MASK2(a,b) | MASK(c))
1282#define MASK4(a,b,c,d) (MASK3(a,b,c) | MASK(d))
1283#define STATE_SET(b,v) ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1284#define STATE_TEST(b) (state & MASK(b))
1285#define STATE_SET_TEST(b,v) ((v) ? ((state |= MASK(b)), 1) : \
1286 ((state &= ~MASK(b)), 0))
1287#define SWITCH_STATE \
1288 (state & MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG))
1289
1290 /*
1291 * Set up a bunch of variables to make the tests simpler.
1292 */
1293 error = 0;
1294 cur = *curp;
1295 base = ip->i_df.if_u1.if_extents;
1296 ep = &base[idx];
1297 xfs_bmbt_get_all(ep, &PREV);
1298 newext = new->br_state;
1299 oldext = (newext == XFS_EXT_UNWRITTEN) ?
1300 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
1301 ASSERT(PREV.br_state == oldext);
1302 new_endoff = new->br_startoff + new->br_blockcount;
1303 ASSERT(PREV.br_startoff <= new->br_startoff);
1304 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1305 /*
1306 * Set flags determining what part of the previous oldext allocation
1307 * extent is being replaced by a newext allocation.
1308 */
1309 STATE_SET(LEFT_FILLING, PREV.br_startoff == new->br_startoff);
1310 STATE_SET(RIGHT_FILLING,
1311 PREV.br_startoff + PREV.br_blockcount == new_endoff);
1312 /*
1313 * Check and set flags if this segment has a left neighbor.
1314 * Don't set contiguous if the combined extent would be too large.
1315 */
1316 if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1317 xfs_bmbt_get_all(ep - 1, &LEFT);
1318 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(LEFT.br_startblock));
1319 }
1320 STATE_SET(LEFT_CONTIG,
1321 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
1322 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1323 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1324 LEFT.br_state == newext &&
1325 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1326 /*
1327 * Check and set flags if this segment has a right neighbor.
1328 * Don't set contiguous if the combined extent would be too large.
1329 * Also check for all-three-contiguous being too large.
1330 */
1331 if (STATE_SET_TEST(RIGHT_VALID,
1332 idx <
1333 ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1)) {
1334 xfs_bmbt_get_all(ep + 1, &RIGHT);
1335 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(RIGHT.br_startblock));
1336 }
1337 STATE_SET(RIGHT_CONTIG,
1338 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
1339 new_endoff == RIGHT.br_startoff &&
1340 new->br_startblock + new->br_blockcount ==
1341 RIGHT.br_startblock &&
1342 newext == RIGHT.br_state &&
1343 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1344 ((state & MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING)) !=
1345 MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING) ||
1346 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1347 <= MAXEXTLEN));
1348 /*
1349 * Switch out based on the FILLING and CONTIG state bits.
1350 */
1351 switch (SWITCH_STATE) {
1352
1353 case MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1354 /*
1355 * Setting all of a previous oldext extent to newext.
1356 * The left and right neighbors are both contiguous with new.
1357 */
1358 xfs_bmap_trace_pre_update(fname, "LF|RF|LC|RC", ip, idx - 1,
1359 XFS_DATA_FORK);
1360 xfs_bmbt_set_blockcount(ep - 1,
1361 LEFT.br_blockcount + PREV.br_blockcount +
1362 RIGHT.br_blockcount);
1363 xfs_bmap_trace_post_update(fname, "LF|RF|LC|RC", ip, idx - 1,
1364 XFS_DATA_FORK);
1365 xfs_bmap_trace_delete(fname, "LF|RF|LC|RC", ip, idx, 2,
1366 XFS_DATA_FORK);
1367 xfs_bmap_delete_exlist(ip, idx, 2, XFS_DATA_FORK);
1368 ip->i_df.if_lastex = idx - 1;
1369 ip->i_d.di_nextents -= 2;
1370 if (cur == NULL)
1371 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1372 else {
1373 rval = XFS_ILOG_CORE;
1374 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1375 RIGHT.br_startblock,
1376 RIGHT.br_blockcount, &i)))
1377 goto done;
1378 ASSERT(i == 1);
1379 if ((error = xfs_bmbt_delete(cur, &i)))
1380 goto done;
1381 ASSERT(i == 1);
1382 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1383 goto done;
1384 ASSERT(i == 1);
1385 if ((error = xfs_bmbt_delete(cur, &i)))
1386 goto done;
1387 ASSERT(i == 1);
1388 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1389 goto done;
1390 ASSERT(i == 1);
1391 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1392 LEFT.br_startblock,
1393 LEFT.br_blockcount + PREV.br_blockcount +
1394 RIGHT.br_blockcount, LEFT.br_state)))
1395 goto done;
1396 }
1397 break;
1398
1399 case MASK3(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG):
1400 /*
1401 * Setting all of a previous oldext extent to newext.
1402 * The left neighbor is contiguous, the right is not.
1403 */
1404 xfs_bmap_trace_pre_update(fname, "LF|RF|LC", ip, idx - 1,
1405 XFS_DATA_FORK);
1406 xfs_bmbt_set_blockcount(ep - 1,
1407 LEFT.br_blockcount + PREV.br_blockcount);
1408 xfs_bmap_trace_post_update(fname, "LF|RF|LC", ip, idx - 1,
1409 XFS_DATA_FORK);
1410 ip->i_df.if_lastex = idx - 1;
1411 xfs_bmap_trace_delete(fname, "LF|RF|LC", ip, idx, 1,
1412 XFS_DATA_FORK);
1413 xfs_bmap_delete_exlist(ip, idx, 1, XFS_DATA_FORK);
1414 ip->i_d.di_nextents--;
1415 if (cur == NULL)
1416 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1417 else {
1418 rval = XFS_ILOG_CORE;
1419 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1420 PREV.br_startblock, PREV.br_blockcount,
1421 &i)))
1422 goto done;
1423 ASSERT(i == 1);
1424 if ((error = xfs_bmbt_delete(cur, &i)))
1425 goto done;
1426 ASSERT(i == 1);
1427 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1428 goto done;
1429 ASSERT(i == 1);
1430 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1431 LEFT.br_startblock,
1432 LEFT.br_blockcount + PREV.br_blockcount,
1433 LEFT.br_state)))
1434 goto done;
1435 }
1436 break;
1437
1438 case MASK3(LEFT_FILLING, RIGHT_FILLING, RIGHT_CONTIG):
1439 /*
1440 * Setting all of a previous oldext extent to newext.
1441 * The right neighbor is contiguous, the left is not.
1442 */
1443 xfs_bmap_trace_pre_update(fname, "LF|RF|RC", ip, idx,
1444 XFS_DATA_FORK);
1445 xfs_bmbt_set_blockcount(ep,
1446 PREV.br_blockcount + RIGHT.br_blockcount);
1447 xfs_bmbt_set_state(ep, newext);
1448 xfs_bmap_trace_post_update(fname, "LF|RF|RC", ip, idx,
1449 XFS_DATA_FORK);
1450 ip->i_df.if_lastex = idx;
1451 xfs_bmap_trace_delete(fname, "LF|RF|RC", ip, idx + 1, 1,
1452 XFS_DATA_FORK);
1453 xfs_bmap_delete_exlist(ip, idx + 1, 1, XFS_DATA_FORK);
1454 ip->i_d.di_nextents--;
1455 if (cur == NULL)
1456 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1457 else {
1458 rval = XFS_ILOG_CORE;
1459 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1460 RIGHT.br_startblock,
1461 RIGHT.br_blockcount, &i)))
1462 goto done;
1463 ASSERT(i == 1);
1464 if ((error = xfs_bmbt_delete(cur, &i)))
1465 goto done;
1466 ASSERT(i == 1);
1467 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1468 goto done;
1469 ASSERT(i == 1);
1470 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1471 new->br_startblock,
1472 new->br_blockcount + RIGHT.br_blockcount,
1473 newext)))
1474 goto done;
1475 }
1476 break;
1477
1478 case MASK2(LEFT_FILLING, RIGHT_FILLING):
1479 /*
1480 * Setting all of a previous oldext extent to newext.
1481 * Neither the left nor right neighbors are contiguous with
1482 * the new one.
1483 */
1484 xfs_bmap_trace_pre_update(fname, "LF|RF", ip, idx,
1485 XFS_DATA_FORK);
1486 xfs_bmbt_set_state(ep, newext);
1487 xfs_bmap_trace_post_update(fname, "LF|RF", ip, idx,
1488 XFS_DATA_FORK);
1489 ip->i_df.if_lastex = idx;
1490 if (cur == NULL)
1491 rval = XFS_ILOG_DEXT;
1492 else {
1493 rval = 0;
1494 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1495 new->br_startblock, new->br_blockcount,
1496 &i)))
1497 goto done;
1498 ASSERT(i == 1);
1499 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1500 new->br_startblock, new->br_blockcount,
1501 newext)))
1502 goto done;
1503 }
1504 break;
1505
1506 case MASK2(LEFT_FILLING, LEFT_CONTIG):
1507 /*
1508 * Setting the first part of a previous oldext extent to newext.
1509 * The left neighbor is contiguous.
1510 */
1511 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx - 1,
1512 XFS_DATA_FORK);
1513 xfs_bmbt_set_blockcount(ep - 1,
1514 LEFT.br_blockcount + new->br_blockcount);
1515 xfs_bmbt_set_startoff(ep,
1516 PREV.br_startoff + new->br_blockcount);
1517 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx - 1,
1518 XFS_DATA_FORK);
1519 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx,
1520 XFS_DATA_FORK);
1521 xfs_bmbt_set_startblock(ep,
1522 new->br_startblock + new->br_blockcount);
1523 xfs_bmbt_set_blockcount(ep,
1524 PREV.br_blockcount - new->br_blockcount);
1525 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx,
1526 XFS_DATA_FORK);
1527 ip->i_df.if_lastex = idx - 1;
1528 if (cur == NULL)
1529 rval = XFS_ILOG_DEXT;
1530 else {
1531 rval = 0;
1532 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1533 PREV.br_startblock, PREV.br_blockcount,
1534 &i)))
1535 goto done;
1536 ASSERT(i == 1);
1537 if ((error = xfs_bmbt_update(cur,
1538 PREV.br_startoff + new->br_blockcount,
1539 PREV.br_startblock + new->br_blockcount,
1540 PREV.br_blockcount - new->br_blockcount,
1541 oldext)))
1542 goto done;
1543 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1544 goto done;
1545 if (xfs_bmbt_update(cur, LEFT.br_startoff,
1546 LEFT.br_startblock,
1547 LEFT.br_blockcount + new->br_blockcount,
1548 LEFT.br_state))
1549 goto done;
1550 }
1551 break;
1552
1553 case MASK(LEFT_FILLING):
1554 /*
1555 * Setting the first part of a previous oldext extent to newext.
1556 * The left neighbor is not contiguous.
1557 */
1558 xfs_bmap_trace_pre_update(fname, "LF", ip, idx, XFS_DATA_FORK);
1559 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1560 xfs_bmbt_set_startoff(ep, new_endoff);
1561 xfs_bmbt_set_blockcount(ep,
1562 PREV.br_blockcount - new->br_blockcount);
1563 xfs_bmbt_set_startblock(ep,
1564 new->br_startblock + new->br_blockcount);
1565 xfs_bmap_trace_post_update(fname, "LF", ip, idx, XFS_DATA_FORK);
1566 xfs_bmap_trace_insert(fname, "LF", ip, idx, 1, new, NULL,
1567 XFS_DATA_FORK);
1568 xfs_bmap_insert_exlist(ip, idx, 1, new, XFS_DATA_FORK);
1569 ip->i_df.if_lastex = idx;
1570 ip->i_d.di_nextents++;
1571 if (cur == NULL)
1572 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1573 else {
1574 rval = XFS_ILOG_CORE;
1575 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1576 PREV.br_startblock, PREV.br_blockcount,
1577 &i)))
1578 goto done;
1579 ASSERT(i == 1);
1580 if ((error = xfs_bmbt_update(cur,
1581 PREV.br_startoff + new->br_blockcount,
1582 PREV.br_startblock + new->br_blockcount,
1583 PREV.br_blockcount - new->br_blockcount,
1584 oldext)))
1585 goto done;
1586 cur->bc_rec.b = *new;
1587 if ((error = xfs_bmbt_insert(cur, &i)))
1588 goto done;
1589 ASSERT(i == 1);
1590 }
1591 break;
1592
1593 case MASK2(RIGHT_FILLING, RIGHT_CONTIG):
1594 /*
1595 * Setting the last part of a previous oldext extent to newext.
1596 * The right neighbor is contiguous with the new allocation.
1597 */
1598 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx,
1599 XFS_DATA_FORK);
1600 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx + 1,
1601 XFS_DATA_FORK);
1602 xfs_bmbt_set_blockcount(ep,
1603 PREV.br_blockcount - new->br_blockcount);
1604 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx,
1605 XFS_DATA_FORK);
1606 xfs_bmbt_set_allf(ep + 1, new->br_startoff, new->br_startblock,
1607 new->br_blockcount + RIGHT.br_blockcount, newext);
1608 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx + 1,
1609 XFS_DATA_FORK);
1610 ip->i_df.if_lastex = idx + 1;
1611 if (cur == NULL)
1612 rval = XFS_ILOG_DEXT;
1613 else {
1614 rval = 0;
1615 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1616 PREV.br_startblock,
1617 PREV.br_blockcount, &i)))
1618 goto done;
1619 ASSERT(i == 1);
1620 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1621 PREV.br_startblock,
1622 PREV.br_blockcount - new->br_blockcount,
1623 oldext)))
1624 goto done;
1625 if ((error = xfs_bmbt_increment(cur, 0, &i)))
1626 goto done;
1627 if ((error = xfs_bmbt_update(cur, new->br_startoff,
1628 new->br_startblock,
1629 new->br_blockcount + RIGHT.br_blockcount,
1630 newext)))
1631 goto done;
1632 }
1633 break;
1634
1635 case MASK(RIGHT_FILLING):
1636 /*
1637 * Setting the last part of a previous oldext extent to newext.
1638 * The right neighbor is not contiguous.
1639 */
1640 xfs_bmap_trace_pre_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1641 xfs_bmbt_set_blockcount(ep,
1642 PREV.br_blockcount - new->br_blockcount);
1643 xfs_bmap_trace_post_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1644 xfs_bmap_trace_insert(fname, "RF", ip, idx + 1, 1,
1645 new, NULL, XFS_DATA_FORK);
1646 xfs_bmap_insert_exlist(ip, idx + 1, 1, new, XFS_DATA_FORK);
1647 ip->i_df.if_lastex = idx + 1;
1648 ip->i_d.di_nextents++;
1649 if (cur == NULL)
1650 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1651 else {
1652 rval = XFS_ILOG_CORE;
1653 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1654 PREV.br_startblock, PREV.br_blockcount,
1655 &i)))
1656 goto done;
1657 ASSERT(i == 1);
1658 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1659 PREV.br_startblock,
1660 PREV.br_blockcount - new->br_blockcount,
1661 oldext)))
1662 goto done;
1663 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1664 new->br_startblock, new->br_blockcount,
1665 &i)))
1666 goto done;
1667 ASSERT(i == 0);
1668 cur->bc_rec.b.br_state = XFS_EXT_NORM;
1669 if ((error = xfs_bmbt_insert(cur, &i)))
1670 goto done;
1671 ASSERT(i == 1);
1672 }
1673 break;
1674
1675 case 0:
1676 /*
1677 * Setting the middle part of a previous oldext extent to
1678 * newext. Contiguity is impossible here.
1679 * One extent becomes three extents.
1680 */
1681 xfs_bmap_trace_pre_update(fname, "0", ip, idx, XFS_DATA_FORK);
1682 xfs_bmbt_set_blockcount(ep,
1683 new->br_startoff - PREV.br_startoff);
1684 xfs_bmap_trace_post_update(fname, "0", ip, idx, XFS_DATA_FORK);
1685 r[0] = *new;
1686 r[1].br_startoff = new_endoff;
1687 r[1].br_blockcount =
1688 PREV.br_startoff + PREV.br_blockcount - new_endoff;
1689 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1690 r[1].br_state = oldext;
1691 xfs_bmap_trace_insert(fname, "0", ip, idx + 1, 2, &r[0], &r[1],
1692 XFS_DATA_FORK);
1693 xfs_bmap_insert_exlist(ip, idx + 1, 2, &r[0], XFS_DATA_FORK);
1694 ip->i_df.if_lastex = idx + 1;
1695 ip->i_d.di_nextents += 2;
1696 if (cur == NULL)
1697 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1698 else {
1699 rval = XFS_ILOG_CORE;
1700 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1701 PREV.br_startblock, PREV.br_blockcount,
1702 &i)))
1703 goto done;
1704 ASSERT(i == 1);
1705 /* new right extent - oldext */
1706 if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1707 r[1].br_startblock, r[1].br_blockcount,
1708 r[1].br_state)))
1709 goto done;
1710 /* new left extent - oldext */
1711 PREV.br_blockcount =
1712 new->br_startoff - PREV.br_startoff;
1713 cur->bc_rec.b = PREV;
1714 if ((error = xfs_bmbt_insert(cur, &i)))
1715 goto done;
1716 ASSERT(i == 1);
1717 if ((error = xfs_bmbt_increment(cur, 0, &i)))
1718 goto done;
1719 ASSERT(i == 1);
1720 /* new middle extent - newext */
1721 cur->bc_rec.b = *new;
1722 if ((error = xfs_bmbt_insert(cur, &i)))
1723 goto done;
1724 ASSERT(i == 1);
1725 }
1726 break;
1727
1728 case MASK3(LEFT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1729 case MASK3(RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1730 case MASK2(LEFT_FILLING, RIGHT_CONTIG):
1731 case MASK2(RIGHT_FILLING, LEFT_CONTIG):
1732 case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1733 case MASK(LEFT_CONTIG):
1734 case MASK(RIGHT_CONTIG):
1735 /*
1736 * These cases are all impossible.
1737 */
1738 ASSERT(0);
1739 }
1740 *curp = cur;
1741done:
1742 *logflagsp = rval;
1743 return error;
1744#undef LEFT
1745#undef RIGHT
1746#undef PREV
1747#undef MASK
1748#undef MASK2
1749#undef MASK3
1750#undef MASK4
1751#undef STATE_SET
1752#undef STATE_TEST
1753#undef STATE_SET_TEST
1754#undef SWITCH_STATE
1755}
1756
1757/*
1758 * Called by xfs_bmap_add_extent to handle cases converting a hole
1759 * to a delayed allocation.
1760 */
1761/*ARGSUSED*/
1762STATIC int /* error */
1763xfs_bmap_add_extent_hole_delay(
1764 xfs_inode_t *ip, /* incore inode pointer */
1765 xfs_extnum_t idx, /* extent number to update/insert */
1766 xfs_btree_cur_t *cur, /* if null, not a btree */
1767 xfs_bmbt_irec_t *new, /* new data to put in extent list */
1768 int *logflagsp, /* inode logging flags */
1769 int rsvd) /* OK to allocate reserved blocks */
1770{
1771 xfs_bmbt_rec_t *base; /* base of extent entry list */
1772 xfs_bmbt_rec_t *ep; /* extent list entry for idx */
1773#ifdef XFS_BMAP_TRACE
1774 static char fname[] = "xfs_bmap_add_extent_hole_delay";
1775#endif
1776 xfs_bmbt_irec_t left; /* left neighbor extent entry */
1777 xfs_filblks_t newlen=0; /* new indirect size */
1778 xfs_filblks_t oldlen=0; /* old indirect size */
1779 xfs_bmbt_irec_t right; /* right neighbor extent entry */
1780 int state; /* state bits, accessed thru macros */
1781 xfs_filblks_t temp; /* temp for indirect calculations */
1782 enum { /* bit number definitions for state */
1783 LEFT_CONTIG, RIGHT_CONTIG,
1784 LEFT_DELAY, RIGHT_DELAY,
1785 LEFT_VALID, RIGHT_VALID
1786 };
1787
1788#define MASK(b) (1 << (b))
1789#define MASK2(a,b) (MASK(a) | MASK(b))
1790#define STATE_SET(b,v) ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1791#define STATE_TEST(b) (state & MASK(b))
1792#define STATE_SET_TEST(b,v) ((v) ? ((state |= MASK(b)), 1) : \
1793 ((state &= ~MASK(b)), 0))
1794#define SWITCH_STATE (state & MASK2(LEFT_CONTIG, RIGHT_CONTIG))
1795
1796 base = ip->i_df.if_u1.if_extents;
1797 ep = &base[idx];
1798 state = 0;
1799 ASSERT(ISNULLSTARTBLOCK(new->br_startblock));
1800 /*
1801 * Check and set flags if this segment has a left neighbor
1802 */
1803 if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1804 xfs_bmbt_get_all(ep - 1, &left);
1805 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(left.br_startblock));
1806 }
1807 /*
1808 * Check and set flags if the current (right) segment exists.
1809 * If it doesn't exist, we're converting the hole at end-of-file.
1810 */
1811 if (STATE_SET_TEST(RIGHT_VALID,
1812 idx <
1813 ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1814 xfs_bmbt_get_all(ep, &right);
1815 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(right.br_startblock));
1816 }
1817 /*
1818 * Set contiguity flags on the left and right neighbors.
1819 * Don't let extents get too large, even if the pieces are contiguous.
1820 */
1821 STATE_SET(LEFT_CONTIG,
1822 STATE_TEST(LEFT_VALID) && STATE_TEST(LEFT_DELAY) &&
1823 left.br_startoff + left.br_blockcount == new->br_startoff &&
1824 left.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1825 STATE_SET(RIGHT_CONTIG,
1826 STATE_TEST(RIGHT_VALID) && STATE_TEST(RIGHT_DELAY) &&
1827 new->br_startoff + new->br_blockcount == right.br_startoff &&
1828 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1829 (!STATE_TEST(LEFT_CONTIG) ||
1830 (left.br_blockcount + new->br_blockcount +
1831 right.br_blockcount <= MAXEXTLEN)));
1832 /*
1833 * Switch out based on the contiguity flags.
1834 */
1835 switch (SWITCH_STATE) {
1836
1837 case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1838 /*
1839 * New allocation is contiguous with delayed allocations
1840 * on the left and on the right.
1841 * Merge all three into a single extent list entry.
1842 */
1843 temp = left.br_blockcount + new->br_blockcount +
1844 right.br_blockcount;
1845 xfs_bmap_trace_pre_update(fname, "LC|RC", ip, idx - 1,
1846 XFS_DATA_FORK);
1847 xfs_bmbt_set_blockcount(ep - 1, temp);
1848 oldlen = STARTBLOCKVAL(left.br_startblock) +
1849 STARTBLOCKVAL(new->br_startblock) +
1850 STARTBLOCKVAL(right.br_startblock);
1851 newlen = xfs_bmap_worst_indlen(ip, temp);
1852 xfs_bmbt_set_startblock(ep - 1, NULLSTARTBLOCK((int)newlen));
1853 xfs_bmap_trace_post_update(fname, "LC|RC", ip, idx - 1,
1854 XFS_DATA_FORK);
1855 xfs_bmap_trace_delete(fname, "LC|RC", ip, idx, 1,
1856 XFS_DATA_FORK);
1857 xfs_bmap_delete_exlist(ip, idx, 1, XFS_DATA_FORK);
1858 ip->i_df.if_lastex = idx - 1;
1859 break;
1860
1861 case MASK(LEFT_CONTIG):
1862 /*
1863 * New allocation is contiguous with a delayed allocation
1864 * on the left.
1865 * Merge the new allocation with the left neighbor.
1866 */
1867 temp = left.br_blockcount + new->br_blockcount;
1868 xfs_bmap_trace_pre_update(fname, "LC", ip, idx - 1,
1869 XFS_DATA_FORK);
1870 xfs_bmbt_set_blockcount(ep - 1, temp);
1871 oldlen = STARTBLOCKVAL(left.br_startblock) +
1872 STARTBLOCKVAL(new->br_startblock);
1873 newlen = xfs_bmap_worst_indlen(ip, temp);
1874 xfs_bmbt_set_startblock(ep - 1, NULLSTARTBLOCK((int)newlen));
1875 xfs_bmap_trace_post_update(fname, "LC", ip, idx - 1,
1876 XFS_DATA_FORK);
1877 ip->i_df.if_lastex = idx - 1;
1878 break;
1879
1880 case MASK(RIGHT_CONTIG):
1881 /*
1882 * New allocation is contiguous with a delayed allocation
1883 * on the right.
1884 * Merge the new allocation with the right neighbor.
1885 */
1886 xfs_bmap_trace_pre_update(fname, "RC", ip, idx, XFS_DATA_FORK);
1887 temp = new->br_blockcount + right.br_blockcount;
1888 oldlen = STARTBLOCKVAL(new->br_startblock) +
1889 STARTBLOCKVAL(right.br_startblock);
1890 newlen = xfs_bmap_worst_indlen(ip, temp);
1891 xfs_bmbt_set_allf(ep, new->br_startoff,
1892 NULLSTARTBLOCK((int)newlen), temp, right.br_state);
1893 xfs_bmap_trace_post_update(fname, "RC", ip, idx, XFS_DATA_FORK);
1894 ip->i_df.if_lastex = idx;
1895 break;
1896
1897 case 0:
1898 /*
1899 * New allocation is not contiguous with another
1900 * delayed allocation.
1901 * Insert a new entry.
1902 */
1903 oldlen = newlen = 0;
1904 xfs_bmap_trace_insert(fname, "0", ip, idx, 1, new, NULL,
1905 XFS_DATA_FORK);
1906 xfs_bmap_insert_exlist(ip, idx, 1, new, XFS_DATA_FORK);
1907 ip->i_df.if_lastex = idx;
1908 break;
1909 }
1910 if (oldlen != newlen) {
1911 ASSERT(oldlen > newlen);
1912 xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
1913 (int)(oldlen - newlen), rsvd);
1914 /*
1915 * Nothing to do for disk quota accounting here.
1916 */
1917 }
1918 *logflagsp = 0;
1919 return 0;
1920#undef MASK
1921#undef MASK2
1922#undef STATE_SET
1923#undef STATE_TEST
1924#undef STATE_SET_TEST
1925#undef SWITCH_STATE
1926}
1927
1928/*
1929 * Called by xfs_bmap_add_extent to handle cases converting a hole
1930 * to a real allocation.
1931 */
1932STATIC int /* error */
1933xfs_bmap_add_extent_hole_real(
1934 xfs_inode_t *ip, /* incore inode pointer */
1935 xfs_extnum_t idx, /* extent number to update/insert */
1936 xfs_btree_cur_t *cur, /* if null, not a btree */
1937 xfs_bmbt_irec_t *new, /* new data to put in extent list */
1938 int *logflagsp, /* inode logging flags */
1939 int whichfork) /* data or attr fork */
1940{
1941 xfs_bmbt_rec_t *ep; /* pointer to extent entry ins. point */
1942 int error; /* error return value */
1943#ifdef XFS_BMAP_TRACE
1944 static char fname[] = "xfs_bmap_add_extent_hole_real";
1945#endif
1946 int i; /* temp state */
1947 xfs_ifork_t *ifp; /* inode fork pointer */
1948 xfs_bmbt_irec_t left; /* left neighbor extent entry */
1949 xfs_bmbt_irec_t right; /* right neighbor extent entry */
1950 int state; /* state bits, accessed thru macros */
1951 enum { /* bit number definitions for state */
1952 LEFT_CONTIG, RIGHT_CONTIG,
1953 LEFT_DELAY, RIGHT_DELAY,
1954 LEFT_VALID, RIGHT_VALID
1955 };
1956
1957#define MASK(b) (1 << (b))
1958#define MASK2(a,b) (MASK(a) | MASK(b))
1959#define STATE_SET(b,v) ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1960#define STATE_TEST(b) (state & MASK(b))
1961#define STATE_SET_TEST(b,v) ((v) ? ((state |= MASK(b)), 1) : \
1962 ((state &= ~MASK(b)), 0))
1963#define SWITCH_STATE (state & MASK2(LEFT_CONTIG, RIGHT_CONTIG))
1964
1965 ifp = XFS_IFORK_PTR(ip, whichfork);
1966 ASSERT(idx <= ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
1967 ep = &ifp->if_u1.if_extents[idx];
1968 state = 0;
1969 /*
1970 * Check and set flags if this segment has a left neighbor.
1971 */
1972 if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1973 xfs_bmbt_get_all(ep - 1, &left);
1974 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(left.br_startblock));
1975 }
1976 /*
1977 * Check and set flags if this segment has a current value.
1978 * Not true if we're inserting into the "hole" at eof.
1979 */
1980 if (STATE_SET_TEST(RIGHT_VALID,
1981 idx <
1982 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1983 xfs_bmbt_get_all(ep, &right);
1984 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(right.br_startblock));
1985 }
1986 /*
1987 * We're inserting a real allocation between "left" and "right".
1988 * Set the contiguity flags. Don't let extents get too large.
1989 */
1990 STATE_SET(LEFT_CONTIG,
1991 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
1992 left.br_startoff + left.br_blockcount == new->br_startoff &&
1993 left.br_startblock + left.br_blockcount == new->br_startblock &&
1994 left.br_state == new->br_state &&
1995 left.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1996 STATE_SET(RIGHT_CONTIG,
1997 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
1998 new->br_startoff + new->br_blockcount == right.br_startoff &&
1999 new->br_startblock + new->br_blockcount ==
2000 right.br_startblock &&
2001 new->br_state == right.br_state &&
2002 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2003 (!STATE_TEST(LEFT_CONTIG) ||
2004 left.br_blockcount + new->br_blockcount +
2005 right.br_blockcount <= MAXEXTLEN));
2006
2007 /*
2008 * Select which case we're in here, and implement it.
2009 */
2010 switch (SWITCH_STATE) {
2011
2012 case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
2013 /*
2014 * New allocation is contiguous with real allocations on the
2015 * left and on the right.
2016 * Merge all three into a single extent list entry.
2017 */
2018 xfs_bmap_trace_pre_update(fname, "LC|RC", ip, idx - 1,
2019 whichfork);
2020 xfs_bmbt_set_blockcount(ep - 1,
2021 left.br_blockcount + new->br_blockcount +
2022 right.br_blockcount);
2023 xfs_bmap_trace_post_update(fname, "LC|RC", ip, idx - 1,
2024 whichfork);
2025 xfs_bmap_trace_delete(fname, "LC|RC", ip,
2026 idx, 1, whichfork);
2027 xfs_bmap_delete_exlist(ip, idx, 1, whichfork);
2028 ifp->if_lastex = idx - 1;
2029 XFS_IFORK_NEXT_SET(ip, whichfork,
2030 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2031 if (cur == NULL) {
2032 *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2033 return 0;
2034 }
2035 *logflagsp = XFS_ILOG_CORE;
2036 if ((error = xfs_bmbt_lookup_eq(cur, right.br_startoff,
2037 right.br_startblock, right.br_blockcount, &i)))
2038 return error;
2039 ASSERT(i == 1);
2040 if ((error = xfs_bmbt_delete(cur, &i)))
2041 return error;
2042 ASSERT(i == 1);
2043 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
2044 return error;
2045 ASSERT(i == 1);
2046 error = xfs_bmbt_update(cur, left.br_startoff,
2047 left.br_startblock,
2048 left.br_blockcount + new->br_blockcount +
2049 right.br_blockcount, left.br_state);
2050 return error;
2051
2052 case MASK(LEFT_CONTIG):
2053 /*
2054 * New allocation is contiguous with a real allocation
2055 * on the left.
2056 * Merge the new allocation with the left neighbor.
2057 */
2058 xfs_bmap_trace_pre_update(fname, "LC", ip, idx - 1, whichfork);
2059 xfs_bmbt_set_blockcount(ep - 1,
2060 left.br_blockcount + new->br_blockcount);
2061 xfs_bmap_trace_post_update(fname, "LC", ip, idx - 1, whichfork);
2062 ifp->if_lastex = idx - 1;
2063 if (cur == NULL) {
2064 *logflagsp = XFS_ILOG_FEXT(whichfork);
2065 return 0;
2066 }
2067 *logflagsp = 0;
2068 if ((error = xfs_bmbt_lookup_eq(cur, left.br_startoff,
2069 left.br_startblock, left.br_blockcount, &i)))
2070 return error;
2071 ASSERT(i == 1);
2072 error = xfs_bmbt_update(cur, left.br_startoff,
2073 left.br_startblock,
2074 left.br_blockcount + new->br_blockcount,
2075 left.br_state);
2076 return error;
2077
2078 case MASK(RIGHT_CONTIG):
2079 /*
2080 * New allocation is contiguous with a real allocation
2081 * on the right.
2082 * Merge the new allocation with the right neighbor.
2083 */
2084 xfs_bmap_trace_pre_update(fname, "RC", ip, idx, whichfork);
2085 xfs_bmbt_set_allf(ep, new->br_startoff, new->br_startblock,
2086 new->br_blockcount + right.br_blockcount,
2087 right.br_state);
2088 xfs_bmap_trace_post_update(fname, "RC", ip, idx, whichfork);
2089 ifp->if_lastex = idx;
2090 if (cur == NULL) {
2091 *logflagsp = XFS_ILOG_FEXT(whichfork);
2092 return 0;
2093 }
2094 *logflagsp = 0;
2095 if ((error = xfs_bmbt_lookup_eq(cur, right.br_startoff,
2096 right.br_startblock, right.br_blockcount, &i)))
2097 return error;
2098 ASSERT(i == 1);
2099 error = xfs_bmbt_update(cur, new->br_startoff,
2100 new->br_startblock,
2101 new->br_blockcount + right.br_blockcount,
2102 right.br_state);
2103 return error;
2104
2105 case 0:
2106 /*
2107 * New allocation is not contiguous with another
2108 * real allocation.
2109 * Insert a new entry.
2110 */
2111 xfs_bmap_trace_insert(fname, "0", ip, idx, 1, new, NULL,
2112 whichfork);
2113 xfs_bmap_insert_exlist(ip, idx, 1, new, whichfork);
2114 ifp->if_lastex = idx;
2115 XFS_IFORK_NEXT_SET(ip, whichfork,
2116 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2117 if (cur == NULL) {
2118 *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2119 return 0;
2120 }
2121 *logflagsp = XFS_ILOG_CORE;
2122 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2123 new->br_startblock, new->br_blockcount, &i)))
2124 return error;
2125 ASSERT(i == 0);
2126 cur->bc_rec.b.br_state = new->br_state;
2127 if ((error = xfs_bmbt_insert(cur, &i)))
2128 return error;
2129 ASSERT(i == 1);
2130 return 0;
2131 }
2132#undef MASK
2133#undef MASK2
2134#undef STATE_SET
2135#undef STATE_TEST
2136#undef STATE_SET_TEST
2137#undef SWITCH_STATE
2138 /* NOTREACHED */
2139 ASSERT(0);
2140 return 0; /* keep gcc quite */
2141}
2142
2143#define XFS_ALLOC_GAP_UNITS 4
2144
2145/*
2146 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2147 * It figures out where to ask the underlying allocator to put the new extent.
2148 */
2149STATIC int /* error */
2150xfs_bmap_alloc(
2151 xfs_bmalloca_t *ap) /* bmap alloc argument struct */
2152{
2153 xfs_fsblock_t adjust; /* adjustment to block numbers */
2154 xfs_alloctype_t atype=0; /* type for allocation routines */
2155 int error; /* error return value */
2156 xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
2157 xfs_mount_t *mp; /* mount point structure */
2158 int nullfb; /* true if ap->firstblock isn't set */
2159 int rt; /* true if inode is realtime */
2160#ifdef __KERNEL__
2161 xfs_extlen_t prod=0; /* product factor for allocators */
2162 xfs_extlen_t ralen=0; /* realtime allocation length */
2163#endif
2164
2165#define ISVALID(x,y) \
2166 (rt ? \
2167 (x) < mp->m_sb.sb_rblocks : \
2168 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2169 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2170 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2171
2172 /*
2173 * Set up variables.
2174 */
2175 mp = ap->ip->i_mount;
2176 nullfb = ap->firstblock == NULLFSBLOCK;
2177 rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2178 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2179#ifdef __KERNEL__
2180 if (rt) {
2181 xfs_extlen_t extsz; /* file extent size for rt */
2182 xfs_fileoff_t nexto; /* next file offset */
2183 xfs_extlen_t orig_alen; /* original ap->alen */
2184 xfs_fileoff_t orig_end; /* original off+len */
2185 xfs_fileoff_t orig_off; /* original ap->off */
2186 xfs_extlen_t mod_off; /* modulus calculations */
2187 xfs_fileoff_t prevo; /* previous file offset */
2188 xfs_rtblock_t rtx; /* realtime extent number */
2189 xfs_extlen_t temp; /* temp for rt calculations */
2190
2191 /*
2192 * Set prod to match the realtime extent size.
2193 */
2194 if (!(extsz = ap->ip->i_d.di_extsize))
2195 extsz = mp->m_sb.sb_rextsize;
2196 prod = extsz / mp->m_sb.sb_rextsize;
2197 orig_off = ap->off;
2198 orig_alen = ap->alen;
2199 orig_end = orig_off + orig_alen;
2200 /*
2201 * If the file offset is unaligned vs. the extent size
2202 * we need to align it. This will be possible unless
2203 * the file was previously written with a kernel that didn't
2204 * perform this alignment.
2205 */
2206 mod_off = do_mod(orig_off, extsz);
2207 if (mod_off) {
2208 ap->alen += mod_off;
2209 ap->off -= mod_off;
2210 }
2211 /*
2212 * Same adjustment for the end of the requested area.
2213 */
2214 if ((temp = (ap->alen % extsz)))
2215 ap->alen += extsz - temp;
2216 /*
2217 * If the previous block overlaps with this proposed allocation
2218 * then move the start forward without adjusting the length.
2219 */
2220 prevo =
2221 ap->prevp->br_startoff == NULLFILEOFF ?
2222 0 :
2223 (ap->prevp->br_startoff +
2224 ap->prevp->br_blockcount);
2225 if (ap->off != orig_off && ap->off < prevo)
2226 ap->off = prevo;
2227 /*
2228 * If the next block overlaps with this proposed allocation
2229 * then move the start back without adjusting the length,
2230 * but not before offset 0.
2231 * This may of course make the start overlap previous block,
2232 * and if we hit the offset 0 limit then the next block
2233 * can still overlap too.
2234 */
2235 nexto = (ap->eof || ap->gotp->br_startoff == NULLFILEOFF) ?
2236 NULLFILEOFF : ap->gotp->br_startoff;
2237 if (!ap->eof &&
2238 ap->off + ap->alen != orig_end &&
2239 ap->off + ap->alen > nexto)
2240 ap->off = nexto > ap->alen ? nexto - ap->alen : 0;
2241 /*
2242 * If we're now overlapping the next or previous extent that
2243 * means we can't fit an extsz piece in this hole. Just move
2244 * the start forward to the first valid spot and set
2245 * the length so we hit the end.
2246 */
2247 if ((ap->off != orig_off && ap->off < prevo) ||
2248 (ap->off + ap->alen != orig_end &&
2249 ap->off + ap->alen > nexto)) {
2250 ap->off = prevo;
2251 ap->alen = nexto - prevo;
2252 }
2253 /*
2254 * If the result isn't a multiple of rtextents we need to
2255 * remove blocks until it is.
2256 */
2257 if ((temp = (ap->alen % mp->m_sb.sb_rextsize))) {
2258 /*
2259 * We're not covering the original request, or
2260 * we won't be able to once we fix the length.
2261 */
2262 if (orig_off < ap->off ||
2263 orig_end > ap->off + ap->alen ||
2264 ap->alen - temp < orig_alen)
2265 return XFS_ERROR(EINVAL);
2266 /*
2267 * Try to fix it by moving the start up.
2268 */
2269 if (ap->off + temp <= orig_off) {
2270 ap->alen -= temp;
2271 ap->off += temp;
2272 }
2273 /*
2274 * Try to fix it by moving the end in.
2275 */
2276 else if (ap->off + ap->alen - temp >= orig_end)
2277 ap->alen -= temp;
2278 /*
2279 * Set the start to the minimum then trim the length.
2280 */
2281 else {
2282 ap->alen -= orig_off - ap->off;
2283 ap->off = orig_off;
2284 ap->alen -= ap->alen % mp->m_sb.sb_rextsize;
2285 }
2286 /*
2287 * Result doesn't cover the request, fail it.
2288 */
2289 if (orig_off < ap->off || orig_end > ap->off + ap->alen)
2290 return XFS_ERROR(EINVAL);
2291 }
2292 ASSERT(ap->alen % mp->m_sb.sb_rextsize == 0);
2293 /*
2294 * If the offset & length are not perfectly aligned
2295 * then kill prod, it will just get us in trouble.
2296 */
2297 if (do_mod(ap->off, extsz) || ap->alen % extsz)
2298 prod = 1;
2299 /*
2300 * Set ralen to be the actual requested length in rtextents.
2301 */
2302 ralen = ap->alen / mp->m_sb.sb_rextsize;
2303 /*
2304 * If the old value was close enough to MAXEXTLEN that
2305 * we rounded up to it, cut it back so it's valid again.
2306 * Note that if it's a really large request (bigger than
2307 * MAXEXTLEN), we don't hear about that number, and can't
2308 * adjust the starting point to match it.
2309 */
2310 if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2311 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2312 /*
2313 * If it's an allocation to an empty file at offset 0,
2314 * pick an extent that will space things out in the rt area.
2315 */
2316 if (ap->eof && ap->off == 0) {
2317 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2318 if (error)
2319 return error;
2320 ap->rval = rtx * mp->m_sb.sb_rextsize;
2321 } else
2322 ap->rval = 0;
2323 }
2324#else
2325 if (rt)
2326 ap->rval = 0;
2327#endif /* __KERNEL__ */
2328 else if (nullfb)
2329 ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2330 else
2331 ap->rval = ap->firstblock;
2332 /*
2333 * If allocating at eof, and there's a previous real block,
2334 * try to use it's last block as our starting point.
2335 */
2336 if (ap->eof && ap->prevp->br_startoff != NULLFILEOFF &&
2337 !ISNULLSTARTBLOCK(ap->prevp->br_startblock) &&
2338 ISVALID(ap->prevp->br_startblock + ap->prevp->br_blockcount,
2339 ap->prevp->br_startblock)) {
2340 ap->rval = ap->prevp->br_startblock + ap->prevp->br_blockcount;
2341 /*
2342 * Adjust for the gap between prevp and us.
2343 */
2344 adjust = ap->off -
2345 (ap->prevp->br_startoff + ap->prevp->br_blockcount);
2346 if (adjust &&
2347 ISVALID(ap->rval + adjust, ap->prevp->br_startblock))
2348 ap->rval += adjust;
2349 }
2350 /*
2351 * If not at eof, then compare the two neighbor blocks.
2352 * Figure out whether either one gives us a good starting point,
2353 * and pick the better one.
2354 */
2355 else if (!ap->eof) {
2356 xfs_fsblock_t gotbno; /* right side block number */
2357 xfs_fsblock_t gotdiff=0; /* right side difference */
2358 xfs_fsblock_t prevbno; /* left side block number */
2359 xfs_fsblock_t prevdiff=0; /* left side difference */
2360
2361 /*
2362 * If there's a previous (left) block, select a requested
2363 * start block based on it.
2364 */
2365 if (ap->prevp->br_startoff != NULLFILEOFF &&
2366 !ISNULLSTARTBLOCK(ap->prevp->br_startblock) &&
2367 (prevbno = ap->prevp->br_startblock +
2368 ap->prevp->br_blockcount) &&
2369 ISVALID(prevbno, ap->prevp->br_startblock)) {
2370 /*
2371 * Calculate gap to end of previous block.
2372 */
2373 adjust = prevdiff = ap->off -
2374 (ap->prevp->br_startoff +
2375 ap->prevp->br_blockcount);
2376 /*
2377 * Figure the startblock based on the previous block's
2378 * end and the gap size.
2379 * Heuristic!
2380 * If the gap is large relative to the piece we're
2381 * allocating, or using it gives us an invalid block
2382 * number, then just use the end of the previous block.
2383 */
2384 if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2385 ISVALID(prevbno + prevdiff,
2386 ap->prevp->br_startblock))
2387 prevbno += adjust;
2388 else
2389 prevdiff += adjust;
2390 /*
2391 * If the firstblock forbids it, can't use it,
2392 * must use default.
2393 */
2394 if (!rt && !nullfb &&
2395 XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2396 prevbno = NULLFSBLOCK;
2397 }
2398 /*
2399 * No previous block or can't follow it, just default.
2400 */
2401 else
2402 prevbno = NULLFSBLOCK;
2403 /*
2404 * If there's a following (right) block, select a requested
2405 * start block based on it.
2406 */
2407 if (!ISNULLSTARTBLOCK(ap->gotp->br_startblock)) {
2408 /*
2409 * Calculate gap to start of next block.
2410 */
2411 adjust = gotdiff = ap->gotp->br_startoff - ap->off;
2412 /*
2413 * Figure the startblock based on the next block's
2414 * start and the gap size.
2415 */
2416 gotbno = ap->gotp->br_startblock;
2417 /*
2418 * Heuristic!
2419 * If the gap is large relative to the piece we're
2420 * allocating, or using it gives us an invalid block
2421 * number, then just use the start of the next block
2422 * offset by our length.
2423 */
2424 if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2425 ISVALID(gotbno - gotdiff, gotbno))
2426 gotbno -= adjust;
2427 else if (ISVALID(gotbno - ap->alen, gotbno)) {
2428 gotbno -= ap->alen;
2429 gotdiff += adjust - ap->alen;
2430 } else
2431 gotdiff += adjust;
2432 /*
2433 * If the firstblock forbids it, can't use it,
2434 * must use default.
2435 */
2436 if (!rt && !nullfb &&
2437 XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2438 gotbno = NULLFSBLOCK;
2439 }
2440 /*
2441 * No next block, just default.
2442 */
2443 else
2444 gotbno = NULLFSBLOCK;
2445 /*
2446 * If both valid, pick the better one, else the only good
2447 * one, else ap->rval is already set (to 0 or the inode block).
2448 */
2449 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2450 ap->rval = prevdiff <= gotdiff ? prevbno : gotbno;
2451 else if (prevbno != NULLFSBLOCK)
2452 ap->rval = prevbno;
2453 else if (gotbno != NULLFSBLOCK)
2454 ap->rval = gotbno;
2455 }
2456 /*
2457 * If allowed, use ap->rval; otherwise must use firstblock since
2458 * it's in the right allocation group.
2459 */
2460 if (nullfb || rt || XFS_FSB_TO_AGNO(mp, ap->rval) == fb_agno)
2461 ;
2462 else
2463 ap->rval = ap->firstblock;
2464 /*
2465 * Realtime allocation, done through xfs_rtallocate_extent.
2466 */
2467 if (rt) {
2468#ifndef __KERNEL__
2469 ASSERT(0);
2470#else
2471 xfs_rtblock_t rtb;
2472
2473 atype = ap->rval == 0 ?
2474 XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2475 do_div(ap->rval, mp->m_sb.sb_rextsize);
2476 rtb = ap->rval;
2477 ap->alen = ralen;
2478 if ((error = xfs_rtallocate_extent(ap->tp, ap->rval, 1, ap->alen,
2479 &ralen, atype, ap->wasdel, prod, &rtb)))
2480 return error;
2481 if (rtb == NULLFSBLOCK && prod > 1 &&
2482 (error = xfs_rtallocate_extent(ap->tp, ap->rval, 1,
2483 ap->alen, &ralen, atype,
2484 ap->wasdel, 1, &rtb)))
2485 return error;
2486 ap->rval = rtb;
2487 if (ap->rval != NULLFSBLOCK) {
2488 ap->rval *= mp->m_sb.sb_rextsize;
2489 ralen *= mp->m_sb.sb_rextsize;
2490 ap->alen = ralen;
2491 ap->ip->i_d.di_nblocks += ralen;
2492 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2493 if (ap->wasdel)
2494 ap->ip->i_delayed_blks -= ralen;
2495 /*
2496 * Adjust the disk quota also. This was reserved
2497 * earlier.
2498 */
2499 XFS_TRANS_MOD_DQUOT_BYINO(mp, ap->tp, ap->ip,
2500 ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2501 XFS_TRANS_DQ_RTBCOUNT,
2502 (long) ralen);
2503 } else
2504 ap->alen = 0;
2505#endif /* __KERNEL__ */
2506 }
2507 /*
2508 * Normal allocation, done through xfs_alloc_vextent.
2509 */
2510 else {
2511 xfs_agnumber_t ag;
2512 xfs_alloc_arg_t args;
2513 xfs_extlen_t blen;
2514 xfs_extlen_t delta;
2515 int isaligned;
2516 xfs_extlen_t longest;
2517 xfs_extlen_t need;
2518 xfs_extlen_t nextminlen=0;
2519 int notinit;
2520 xfs_perag_t *pag;
2521 xfs_agnumber_t startag;
2522 int tryagain;
2523
2524 tryagain = isaligned = 0;
2525 args.tp = ap->tp;
2526 args.mp = mp;
2527 args.fsbno = ap->rval;
2528 args.maxlen = MIN(ap->alen, mp->m_sb.sb_agblocks);
2529 blen = 0;
2530 if (nullfb) {
2531 args.type = XFS_ALLOCTYPE_START_BNO;
2532 args.total = ap->total;
2533 /*
2534 * Find the longest available space.
2535 * We're going to try for the whole allocation at once.
2536 */
2537 startag = ag = XFS_FSB_TO_AGNO(mp, args.fsbno);
2538 notinit = 0;
2539 down_read(&mp->m_peraglock);
2540 while (blen < ap->alen) {
2541 pag = &mp->m_perag[ag];
2542 if (!pag->pagf_init &&
2543 (error = xfs_alloc_pagf_init(mp, args.tp,
2544 ag, XFS_ALLOC_FLAG_TRYLOCK))) {
2545 up_read(&mp->m_peraglock);
2546 return error;
2547 }
2548 /*
2549 * See xfs_alloc_fix_freelist...
2550 */
2551 if (pag->pagf_init) {
2552 need = XFS_MIN_FREELIST_PAG(pag, mp);
2553 delta = need > pag->pagf_flcount ?
2554 need - pag->pagf_flcount : 0;
2555 longest = (pag->pagf_longest > delta) ?
2556 (pag->pagf_longest - delta) :
2557 (pag->pagf_flcount > 0 ||
2558 pag->pagf_longest > 0);
2559 if (blen < longest)
2560 blen = longest;
2561 } else
2562 notinit = 1;
2563 if (++ag == mp->m_sb.sb_agcount)
2564 ag = 0;
2565 if (ag == startag)
2566 break;
2567 }
2568 up_read(&mp->m_peraglock);
2569 /*
2570 * Since the above loop did a BUF_TRYLOCK, it is
2571 * possible that there is space for this request.
2572 */
2573 if (notinit || blen < ap->minlen)
2574 args.minlen = ap->minlen;
2575 /*
2576 * If the best seen length is less than the request
2577 * length, use the best as the minimum.
2578 */
2579 else if (blen < ap->alen)
2580 args.minlen = blen;
2581 /*
2582 * Otherwise we've seen an extent as big as alen,
2583 * use that as the minimum.
2584 */
2585 else
2586 args.minlen = ap->alen;
2587 } else if (ap->low) {
2588 args.type = XFS_ALLOCTYPE_FIRST_AG;
2589 args.total = args.minlen = ap->minlen;
2590 } else {
2591 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2592 args.total = ap->total;
2593 args.minlen = ap->minlen;
2594 }
2595 if (ap->ip->i_d.di_extsize) {
2596 args.prod = ap->ip->i_d.di_extsize;
2597 if ((args.mod = (xfs_extlen_t)do_mod(ap->off, args.prod)))
2598 args.mod = (xfs_extlen_t)(args.prod - args.mod);
2599 } else if (mp->m_sb.sb_blocksize >= NBPP) {
2600 args.prod = 1;
2601 args.mod = 0;
2602 } else {
2603 args.prod = NBPP >> mp->m_sb.sb_blocklog;
2604 if ((args.mod = (xfs_extlen_t)(do_mod(ap->off, args.prod))))
2605 args.mod = (xfs_extlen_t)(args.prod - args.mod);
2606 }
2607 /*
2608 * If we are not low on available data blocks, and the
2609 * underlying logical volume manager is a stripe, and
2610 * the file offset is zero then try to allocate data
2611 * blocks on stripe unit boundary.
2612 * NOTE: ap->aeof is only set if the allocation length
2613 * is >= the stripe unit and the allocation offset is
2614 * at the end of file.
2615 */
2616 if (!ap->low && ap->aeof) {
2617 if (!ap->off) {
2618 args.alignment = mp->m_dalign;
2619 atype = args.type;
2620 isaligned = 1;
2621 /*
2622 * Adjust for alignment
2623 */
2624 if (blen > args.alignment && blen <= ap->alen)
2625 args.minlen = blen - args.alignment;
2626 args.minalignslop = 0;
2627 } else {
2628 /*
2629 * First try an exact bno allocation.
2630 * If it fails then do a near or start bno
2631 * allocation with alignment turned on.
2632 */
2633 atype = args.type;
2634 tryagain = 1;
2635 args.type = XFS_ALLOCTYPE_THIS_BNO;
2636 args.alignment = 1;
2637 /*
2638 * Compute the minlen+alignment for the
2639 * next case. Set slop so that the value
2640 * of minlen+alignment+slop doesn't go up
2641 * between the calls.
2642 */
2643 if (blen > mp->m_dalign && blen <= ap->alen)
2644 nextminlen = blen - mp->m_dalign;
2645 else
2646 nextminlen = args.minlen;
2647 if (nextminlen + mp->m_dalign > args.minlen + 1)
2648 args.minalignslop =
2649 nextminlen + mp->m_dalign -
2650 args.minlen - 1;
2651 else
2652 args.minalignslop = 0;
2653 }
2654 } else {
2655 args.alignment = 1;
2656 args.minalignslop = 0;
2657 }
2658 args.minleft = ap->minleft;
2659 args.wasdel = ap->wasdel;
2660 args.isfl = 0;
2661 args.userdata = ap->userdata;
2662 if ((error = xfs_alloc_vextent(&args)))
2663 return error;
2664 if (tryagain && args.fsbno == NULLFSBLOCK) {
2665 /*
2666 * Exact allocation failed. Now try with alignment
2667 * turned on.
2668 */
2669 args.type = atype;
2670 args.fsbno = ap->rval;
2671 args.alignment = mp->m_dalign;
2672 args.minlen = nextminlen;
2673 args.minalignslop = 0;
2674 isaligned = 1;
2675 if ((error = xfs_alloc_vextent(&args)))
2676 return error;
2677 }
2678 if (isaligned && args.fsbno == NULLFSBLOCK) {
2679 /*
2680 * allocation failed, so turn off alignment and
2681 * try again.
2682 */
2683 args.type = atype;
2684 args.fsbno = ap->rval;
2685 args.alignment = 0;
2686 if ((error = xfs_alloc_vextent(&args)))
2687 return error;
2688 }
2689 if (args.fsbno == NULLFSBLOCK && nullfb &&
2690 args.minlen > ap->minlen) {
2691 args.minlen = ap->minlen;
2692 args.type = XFS_ALLOCTYPE_START_BNO;
2693 args.fsbno = ap->rval;
2694 if ((error = xfs_alloc_vextent(&args)))
2695 return error;
2696 }
2697 if (args.fsbno == NULLFSBLOCK && nullfb) {
2698 args.fsbno = 0;
2699 args.type = XFS_ALLOCTYPE_FIRST_AG;
2700 args.total = ap->minlen;
2701 args.minleft = 0;
2702 if ((error = xfs_alloc_vextent(&args)))
2703 return error;
2704 ap->low = 1;
2705 }
2706 if (args.fsbno != NULLFSBLOCK) {
2707 ap->firstblock = ap->rval = args.fsbno;
2708 ASSERT(nullfb || fb_agno == args.agno ||
2709 (ap->low && fb_agno < args.agno));
2710 ap->alen = args.len;
2711 ap->ip->i_d.di_nblocks += args.len;
2712 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2713 if (ap->wasdel)
2714 ap->ip->i_delayed_blks -= args.len;
2715 /*
2716 * Adjust the disk quota also. This was reserved
2717 * earlier.
2718 */
2719 XFS_TRANS_MOD_DQUOT_BYINO(mp, ap->tp, ap->ip,
2720 ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2721 XFS_TRANS_DQ_BCOUNT,
2722 (long) args.len);
2723 } else {
2724 ap->rval = NULLFSBLOCK;
2725 ap->alen = 0;
2726 }
2727 }
2728 return 0;
2729#undef ISVALID
2730}
2731
2732/*
2733 * Transform a btree format file with only one leaf node, where the
2734 * extents list will fit in the inode, into an extents format file.
2735 * Since the extent list is already in-core, all we have to do is
2736 * give up the space for the btree root and pitch the leaf block.
2737 */
2738STATIC int /* error */
2739xfs_bmap_btree_to_extents(
2740 xfs_trans_t *tp, /* transaction pointer */
2741 xfs_inode_t *ip, /* incore inode pointer */
2742 xfs_btree_cur_t *cur, /* btree cursor */
2743 int *logflagsp, /* inode logging flags */
2744 int whichfork) /* data or attr fork */
2745{
2746 /* REFERENCED */
2747 xfs_bmbt_block_t *cblock;/* child btree block */
2748 xfs_fsblock_t cbno; /* child block number */
2749 xfs_buf_t *cbp; /* child block's buffer */
2750 int error; /* error return value */
2751 xfs_ifork_t *ifp; /* inode fork data */
2752 xfs_mount_t *mp; /* mount point structure */
2753 xfs_bmbt_ptr_t *pp; /* ptr to block address */
2754 xfs_bmbt_block_t *rblock;/* root btree block */
2755
2756 ifp = XFS_IFORK_PTR(ip, whichfork);
2757 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2758 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
2759 rblock = ifp->if_broot;
2760 ASSERT(INT_GET(rblock->bb_level, ARCH_CONVERT) == 1);
2761 ASSERT(INT_GET(rblock->bb_numrecs, ARCH_CONVERT) == 1);
2762 ASSERT(XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes) == 1);
2763 mp = ip->i_mount;
2764 pp = XFS_BMAP_BROOT_PTR_ADDR(rblock, 1, ifp->if_broot_bytes);
2765 *logflagsp = 0;
2766#ifdef DEBUG
2767 if ((error = xfs_btree_check_lptr(cur, INT_GET(*pp, ARCH_CONVERT), 1)))
2768 return error;
2769#endif
2770 cbno = INT_GET(*pp, ARCH_CONVERT);
2771 if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
2772 XFS_BMAP_BTREE_REF)))
2773 return error;
2774 cblock = XFS_BUF_TO_BMBT_BLOCK(cbp);
2775 if ((error = xfs_btree_check_lblock(cur, cblock, 0, cbp)))
2776 return error;
2777 xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
2778 ip->i_d.di_nblocks--;
2779 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
2780 xfs_trans_binval(tp, cbp);
2781 if (cur->bc_bufs[0] == cbp)
2782 cur->bc_bufs[0] = NULL;
2783 xfs_iroot_realloc(ip, -1, whichfork);
2784 ASSERT(ifp->if_broot == NULL);
2785 ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
2786 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
2787 *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2788 return 0;
2789}
2790
2791/*
2792 * Called by xfs_bmapi to update extent list structure and the btree
2793 * after removing space (or undoing a delayed allocation).
2794 */
2795STATIC int /* error */
2796xfs_bmap_del_extent(
2797 xfs_inode_t *ip, /* incore inode pointer */
2798 xfs_trans_t *tp, /* current transaction pointer */
2799 xfs_extnum_t idx, /* extent number to update/delete */
2800 xfs_bmap_free_t *flist, /* list of extents to be freed */
2801 xfs_btree_cur_t *cur, /* if null, not a btree */
2802 xfs_bmbt_irec_t *del, /* data to remove from extent list */
2803 int *logflagsp, /* inode logging flags */
2804 int whichfork, /* data or attr fork */
2805 int rsvd) /* OK to allocate reserved blocks */
2806{
2807 xfs_filblks_t da_new; /* new delay-alloc indirect blocks */
2808 xfs_filblks_t da_old; /* old delay-alloc indirect blocks */
2809 xfs_fsblock_t del_endblock=0; /* first block past del */
2810 xfs_fileoff_t del_endoff; /* first offset past del */
2811 int delay; /* current block is delayed allocated */
2812 int do_fx; /* free extent at end of routine */
2813 xfs_bmbt_rec_t *ep; /* current extent entry pointer */
2814 int error; /* error return value */
2815 int flags; /* inode logging flags */
2816#ifdef XFS_BMAP_TRACE
2817 static char fname[] = "xfs_bmap_del_extent";
2818#endif
2819 xfs_bmbt_irec_t got; /* current extent entry */
2820 xfs_fileoff_t got_endoff; /* first offset past got */
2821 int i; /* temp state */
2822 xfs_ifork_t *ifp; /* inode fork pointer */
2823 xfs_mount_t *mp; /* mount structure */
2824 xfs_filblks_t nblks; /* quota/sb block count */
2825 xfs_bmbt_irec_t new; /* new record to be inserted */
2826 /* REFERENCED */
2827 xfs_extnum_t nextents; /* number of extents in list */
2828 uint qfield; /* quota field to update */
2829 xfs_filblks_t temp; /* for indirect length calculations */
2830 xfs_filblks_t temp2; /* for indirect length calculations */
2831
2832 XFS_STATS_INC(xs_del_exlist);
2833 mp = ip->i_mount;
2834 ifp = XFS_IFORK_PTR(ip, whichfork);
2835 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2836 ASSERT(idx >= 0 && idx < nextents);
2837 ASSERT(del->br_blockcount > 0);
2838 ep = &ifp->if_u1.if_extents[idx];
2839 xfs_bmbt_get_all(ep, &got);
2840 ASSERT(got.br_startoff <= del->br_startoff);
2841 del_endoff = del->br_startoff + del->br_blockcount;
2842 got_endoff = got.br_startoff + got.br_blockcount;
2843 ASSERT(got_endoff >= del_endoff);
2844 delay = ISNULLSTARTBLOCK(got.br_startblock);
2845 ASSERT(ISNULLSTARTBLOCK(del->br_startblock) == delay);
2846 flags = 0;
2847 qfield = 0;
2848 error = 0;
2849 /*
2850 * If deleting a real allocation, must free up the disk space.
2851 */
2852 if (!delay) {
2853 flags = XFS_ILOG_CORE;
2854 /*
2855 * Realtime allocation. Free it and record di_nblocks update.
2856 */
2857 if (whichfork == XFS_DATA_FORK &&
2858 (ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
2859 xfs_fsblock_t bno;
2860 xfs_filblks_t len;
2861
2862 ASSERT(do_mod(del->br_blockcount,
2863 mp->m_sb.sb_rextsize) == 0);
2864 ASSERT(do_mod(del->br_startblock,
2865 mp->m_sb.sb_rextsize) == 0);
2866 bno = del->br_startblock;
2867 len = del->br_blockcount;
2868 do_div(bno, mp->m_sb.sb_rextsize);
2869 do_div(len, mp->m_sb.sb_rextsize);
2870 if ((error = xfs_rtfree_extent(ip->i_transp, bno,
2871 (xfs_extlen_t)len)))
2872 goto done;
2873 do_fx = 0;
2874 nblks = len * mp->m_sb.sb_rextsize;
2875 qfield = XFS_TRANS_DQ_RTBCOUNT;
2876 }
2877 /*
2878 * Ordinary allocation.
2879 */
2880 else {
2881 do_fx = 1;
2882 nblks = del->br_blockcount;
2883 qfield = XFS_TRANS_DQ_BCOUNT;
2884 }
2885 /*
2886 * Set up del_endblock and cur for later.
2887 */
2888 del_endblock = del->br_startblock + del->br_blockcount;
2889 if (cur) {
2890 if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
2891 got.br_startblock, got.br_blockcount,
2892 &i)))
2893 goto done;
2894 ASSERT(i == 1);
2895 }
2896 da_old = da_new = 0;
2897 } else {
2898 da_old = STARTBLOCKVAL(got.br_startblock);
2899 da_new = 0;
2900 nblks = 0;
2901 do_fx = 0;
2902 }
2903 /*
2904 * Set flag value to use in switch statement.
2905 * Left-contig is 2, right-contig is 1.
2906 */
2907 switch (((got.br_startoff == del->br_startoff) << 1) |
2908 (got_endoff == del_endoff)) {
2909 case 3:
2910 /*
2911 * Matches the whole extent. Delete the entry.
2912 */
2913 xfs_bmap_trace_delete(fname, "3", ip, idx, 1, whichfork);
2914 xfs_bmap_delete_exlist(ip, idx, 1, whichfork);
2915 ifp->if_lastex = idx;
2916 if (delay)
2917 break;
2918 XFS_IFORK_NEXT_SET(ip, whichfork,
2919 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2920 flags |= XFS_ILOG_CORE;
2921 if (!cur) {
2922 flags |= XFS_ILOG_FEXT(whichfork);
2923 break;
2924 }
2925 if ((error = xfs_bmbt_delete(cur, &i)))
2926 goto done;
2927 ASSERT(i == 1);
2928 break;
2929
2930 case 2:
2931 /*
2932 * Deleting the first part of the extent.
2933 */
2934 xfs_bmap_trace_pre_update(fname, "2", ip, idx, whichfork);
2935 xfs_bmbt_set_startoff(ep, del_endoff);
2936 temp = got.br_blockcount - del->br_blockcount;
2937 xfs_bmbt_set_blockcount(ep, temp);
2938 ifp->if_lastex = idx;
2939 if (delay) {
2940 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2941 da_old);
2942 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
2943 xfs_bmap_trace_post_update(fname, "2", ip, idx,
2944 whichfork);
2945 da_new = temp;
2946 break;
2947 }
2948 xfs_bmbt_set_startblock(ep, del_endblock);
2949 xfs_bmap_trace_post_update(fname, "2", ip, idx, whichfork);
2950 if (!cur) {
2951 flags |= XFS_ILOG_FEXT(whichfork);
2952 break;
2953 }
2954 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
2955 got.br_blockcount - del->br_blockcount,
2956 got.br_state)))
2957 goto done;
2958 break;
2959
2960 case 1:
2961 /*
2962 * Deleting the last part of the extent.
2963 */
2964 temp = got.br_blockcount - del->br_blockcount;
2965 xfs_bmap_trace_pre_update(fname, "1", ip, idx, whichfork);
2966 xfs_bmbt_set_blockcount(ep, temp);
2967 ifp->if_lastex = idx;
2968 if (delay) {
2969 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2970 da_old);
2971 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
2972 xfs_bmap_trace_post_update(fname, "1", ip, idx,
2973 whichfork);
2974 da_new = temp;
2975 break;
2976 }
2977 xfs_bmap_trace_post_update(fname, "1", ip, idx, whichfork);
2978 if (!cur) {
2979 flags |= XFS_ILOG_FEXT(whichfork);
2980 break;
2981 }
2982 if ((error = xfs_bmbt_update(cur, got.br_startoff,
2983 got.br_startblock,
2984 got.br_blockcount - del->br_blockcount,
2985 got.br_state)))
2986 goto done;
2987 break;
2988
2989 case 0:
2990 /*
2991 * Deleting the middle of the extent.
2992 */
2993 temp = del->br_startoff - got.br_startoff;
2994 xfs_bmap_trace_pre_update(fname, "0", ip, idx, whichfork);
2995 xfs_bmbt_set_blockcount(ep, temp);
2996 new.br_startoff = del_endoff;
2997 temp2 = got_endoff - del_endoff;
2998 new.br_blockcount = temp2;
2999 new.br_state = got.br_state;
3000 if (!delay) {
3001 new.br_startblock = del_endblock;
3002 flags |= XFS_ILOG_CORE;
3003 if (cur) {
3004 if ((error = xfs_bmbt_update(cur,
3005 got.br_startoff,
3006 got.br_startblock, temp,
3007 got.br_state)))
3008 goto done;
3009 if ((error = xfs_bmbt_increment(cur, 0, &i)))
3010 goto done;
3011 cur->bc_rec.b = new;
3012 error = xfs_bmbt_insert(cur, &i);
3013 if (error && error != ENOSPC)
3014 goto done;
3015 /*
3016 * If get no-space back from btree insert,
3017 * it tried a split, and we have a zero
3018 * block reservation.
3019 * Fix up our state and return the error.
3020 */
3021 if (error == ENOSPC) {
3022 /*
3023 * Reset the cursor, don't trust
3024 * it after any insert operation.
3025 */
3026 if ((error = xfs_bmbt_lookup_eq(cur,
3027 got.br_startoff,
3028 got.br_startblock,
3029 temp, &i)))
3030 goto done;
3031 ASSERT(i == 1);
3032 /*
3033 * Update the btree record back
3034 * to the original value.
3035 */
3036 if ((error = xfs_bmbt_update(cur,
3037 got.br_startoff,
3038 got.br_startblock,
3039 got.br_blockcount,
3040 got.br_state)))
3041 goto done;
3042 /*
3043 * Reset the extent record back
3044 * to the original value.
3045 */
3046 xfs_bmbt_set_blockcount(ep,
3047 got.br_blockcount);
3048 flags = 0;
3049 error = XFS_ERROR(ENOSPC);
3050 goto done;
3051 }
3052 ASSERT(i == 1);
3053 } else
3054 flags |= XFS_ILOG_FEXT(whichfork);
3055 XFS_IFORK_NEXT_SET(ip, whichfork,
3056 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
3057 } else {
3058 ASSERT(whichfork == XFS_DATA_FORK);
3059 temp = xfs_bmap_worst_indlen(ip, temp);
3060 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3061 temp2 = xfs_bmap_worst_indlen(ip, temp2);
3062 new.br_startblock = NULLSTARTBLOCK((int)temp2);
3063 da_new = temp + temp2;
3064 while (da_new > da_old) {
3065 if (temp) {
3066 temp--;
3067 da_new--;
3068 xfs_bmbt_set_startblock(ep,
3069 NULLSTARTBLOCK((int)temp));
3070 }
3071 if (da_new == da_old)
3072 break;
3073 if (temp2) {
3074 temp2--;
3075 da_new--;
3076 new.br_startblock =
3077 NULLSTARTBLOCK((int)temp2);
3078 }
3079 }
3080 }
3081 xfs_bmap_trace_post_update(fname, "0", ip, idx, whichfork);
3082 xfs_bmap_trace_insert(fname, "0", ip, idx + 1, 1, &new, NULL,
3083 whichfork);
3084 xfs_bmap_insert_exlist(ip, idx + 1, 1, &new, whichfork);
3085 ifp->if_lastex = idx + 1;
3086 break;
3087 }
3088 /*
3089 * If we need to, add to list of extents to delete.
3090 */
3091 if (do_fx)
3092 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
3093 mp);
3094 /*
3095 * Adjust inode # blocks in the file.
3096 */
3097 if (nblks)
3098 ip->i_d.di_nblocks -= nblks;
3099 /*
3100 * Adjust quota data.
3101 */
3102 if (qfield)
3103 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, qfield, (long)-nblks);
3104
3105 /*
3106 * Account for change in delayed indirect blocks.
3107 * Nothing to do for disk quota accounting here.
3108 */
3109 ASSERT(da_old >= da_new);
3110 if (da_old > da_new)
3111 xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS, (int)(da_old - da_new),
3112 rsvd);
3113done:
3114 *logflagsp = flags;
3115 return error;
3116}
3117
3118/*
3119 * Remove the entry "free" from the free item list. Prev points to the
3120 * previous entry, unless "free" is the head of the list.
3121 */
3122STATIC void
3123xfs_bmap_del_free(
3124 xfs_bmap_free_t *flist, /* free item list header */
3125 xfs_bmap_free_item_t *prev, /* previous item on list, if any */
3126 xfs_bmap_free_item_t *free) /* list item to be freed */
3127{
3128 if (prev)
3129 prev->xbfi_next = free->xbfi_next;
3130 else
3131 flist->xbf_first = free->xbfi_next;
3132 flist->xbf_count--;
3133 kmem_zone_free(xfs_bmap_free_item_zone, free);
3134}
3135
3136/*
3137 * Remove count entries from the extents array for inode "ip", starting
3138 * at index "idx". Copies the remaining items down over the deleted ones,
3139 * and gives back the excess memory.
3140 */
3141STATIC void
3142xfs_bmap_delete_exlist(
3143 xfs_inode_t *ip, /* incore inode pointer */
3144 xfs_extnum_t idx, /* starting delete index */
3145 xfs_extnum_t count, /* count of items to delete */
3146 int whichfork) /* data or attr fork */
3147{
3148 xfs_bmbt_rec_t *base; /* base of extent list */
3149 xfs_ifork_t *ifp; /* inode fork pointer */
3150 xfs_extnum_t nextents; /* number of extents in list after */
3151
3152 ifp = XFS_IFORK_PTR(ip, whichfork);
3153 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3154 base = ifp->if_u1.if_extents;
3155 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - count;
3156 memmove(&base[idx], &base[idx + count],
3157 (nextents - idx) * sizeof(*base));
3158 xfs_iext_realloc(ip, -count, whichfork);
3159}
3160
3161/*
3162 * Convert an extents-format file into a btree-format file.
3163 * The new file will have a root block (in the inode) and a single child block.
3164 */
3165STATIC int /* error */
3166xfs_bmap_extents_to_btree(
3167 xfs_trans_t *tp, /* transaction pointer */
3168 xfs_inode_t *ip, /* incore inode pointer */
3169 xfs_fsblock_t *firstblock, /* first-block-allocated */
3170 xfs_bmap_free_t *flist, /* blocks freed in xaction */
3171 xfs_btree_cur_t **curp, /* cursor returned to caller */
3172 int wasdel, /* converting a delayed alloc */
3173 int *logflagsp, /* inode logging flags */
3174 int whichfork) /* data or attr fork */
3175{
3176 xfs_bmbt_block_t *ablock; /* allocated (child) bt block */
3177 xfs_buf_t *abp; /* buffer for ablock */
3178 xfs_alloc_arg_t args; /* allocation arguments */
3179 xfs_bmbt_rec_t *arp; /* child record pointer */
3180 xfs_bmbt_block_t *block; /* btree root block */
3181 xfs_btree_cur_t *cur; /* bmap btree cursor */
3182 xfs_bmbt_rec_t *ep; /* extent list pointer */
3183 int error; /* error return value */
3184 xfs_extnum_t i, cnt; /* extent list index */
3185 xfs_ifork_t *ifp; /* inode fork pointer */
3186 xfs_bmbt_key_t *kp; /* root block key pointer */
3187 xfs_mount_t *mp; /* mount structure */
3188 xfs_extnum_t nextents; /* extent list size */
3189 xfs_bmbt_ptr_t *pp; /* root block address pointer */
3190
3191 ifp = XFS_IFORK_PTR(ip, whichfork);
3192 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3193 ASSERT(ifp->if_ext_max ==
3194 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
3195 /*
3196 * Make space in the inode incore.
3197 */
3198 xfs_iroot_realloc(ip, 1, whichfork);
3199 ifp->if_flags |= XFS_IFBROOT;
3200 /*
3201 * Fill in the root.
3202 */
3203 block = ifp->if_broot;
3204 INT_SET(block->bb_magic, ARCH_CONVERT, XFS_BMAP_MAGIC);
3205 INT_SET(block->bb_level, ARCH_CONVERT, 1);
3206 INT_SET(block->bb_numrecs, ARCH_CONVERT, 1);
3207 INT_SET(block->bb_leftsib, ARCH_CONVERT, NULLDFSBNO);
3208 INT_SET(block->bb_rightsib, ARCH_CONVERT, NULLDFSBNO);
3209 /*
3210 * Need a cursor. Can't allocate until bb_level is filled in.
3211 */
3212 mp = ip->i_mount;
3213 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
3214 whichfork);
3215 cur->bc_private.b.firstblock = *firstblock;
3216 cur->bc_private.b.flist = flist;
3217 cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3218 /*
3219 * Convert to a btree with two levels, one record in root.
3220 */
3221 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3222 args.tp = tp;
3223 args.mp = mp;
3224 if (*firstblock == NULLFSBLOCK) {
3225 args.type = XFS_ALLOCTYPE_START_BNO;
3226 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3227 } else if (flist->xbf_low) {
3228 args.type = XFS_ALLOCTYPE_START_BNO;
3229 args.fsbno = *firstblock;
3230 } else {
3231 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3232 args.fsbno = *firstblock;
3233 }
3234 args.minlen = args.maxlen = args.prod = 1;
3235 args.total = args.minleft = args.alignment = args.mod = args.isfl =
3236 args.minalignslop = 0;
3237 args.wasdel = wasdel;
3238 *logflagsp = 0;
3239 if ((error = xfs_alloc_vextent(&args))) {
3240 xfs_iroot_realloc(ip, -1, whichfork);
3241 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3242 return error;
3243 }
3244 /*
3245 * Allocation can't fail, the space was reserved.
3246 */
3247 ASSERT(args.fsbno != NULLFSBLOCK);
3248 ASSERT(*firstblock == NULLFSBLOCK ||
3249 args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3250 (flist->xbf_low &&
3251 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3252 *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3253 cur->bc_private.b.allocated++;
3254 ip->i_d.di_nblocks++;
3255 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3256 abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3257 /*
3258 * Fill in the child block.
3259 */
3260 ablock = XFS_BUF_TO_BMBT_BLOCK(abp);
3261 INT_SET(ablock->bb_magic, ARCH_CONVERT, XFS_BMAP_MAGIC);
3262 ablock->bb_level = 0;
3263 INT_SET(ablock->bb_leftsib, ARCH_CONVERT, NULLDFSBNO);
3264 INT_SET(ablock->bb_rightsib, ARCH_CONVERT, NULLDFSBNO);
3265 arp = XFS_BMAP_REC_IADDR(ablock, 1, cur);
3266 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3267 for (ep = ifp->if_u1.if_extents, cnt = i = 0; i < nextents; i++, ep++) {
3268 if (!ISNULLSTARTBLOCK(xfs_bmbt_get_startblock(ep))) {
3269 arp->l0 = INT_GET(ep->l0, ARCH_CONVERT);
3270 arp->l1 = INT_GET(ep->l1, ARCH_CONVERT);
3271 arp++; cnt++;
3272 }
3273 }
3274 INT_SET(ablock->bb_numrecs, ARCH_CONVERT, cnt);
3275 ASSERT(INT_GET(ablock->bb_numrecs, ARCH_CONVERT) == XFS_IFORK_NEXTENTS(ip, whichfork));
3276 /*
3277 * Fill in the root key and pointer.
3278 */
3279 kp = XFS_BMAP_KEY_IADDR(block, 1, cur);
3280 arp = XFS_BMAP_REC_IADDR(ablock, 1, cur);
3281 INT_SET(kp->br_startoff, ARCH_CONVERT, xfs_bmbt_disk_get_startoff(arp));
3282 pp = XFS_BMAP_PTR_IADDR(block, 1, cur);
3283 INT_SET(*pp, ARCH_CONVERT, args.fsbno);
3284 /*
3285 * Do all this logging at the end so that
3286 * the root is at the right level.
3287 */
3288 xfs_bmbt_log_block(cur, abp, XFS_BB_ALL_BITS);
3289 xfs_bmbt_log_recs(cur, abp, 1, INT_GET(ablock->bb_numrecs, ARCH_CONVERT));
3290 ASSERT(*curp == NULL);
3291 *curp = cur;
3292 *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FBROOT(whichfork);
3293 return 0;
3294}
3295
3296/*
3297 * Insert new item(s) in the extent list for inode "ip".
3298 * Count new items are inserted at offset idx.
3299 */
3300STATIC void
3301xfs_bmap_insert_exlist(
3302 xfs_inode_t *ip, /* incore inode pointer */
3303 xfs_extnum_t idx, /* starting index of new items */
3304 xfs_extnum_t count, /* number of inserted items */
3305 xfs_bmbt_irec_t *new, /* items to insert */
3306 int whichfork) /* data or attr fork */
3307{
3308 xfs_bmbt_rec_t *base; /* extent list base */
3309 xfs_ifork_t *ifp; /* inode fork pointer */
3310 xfs_extnum_t nextents; /* extent list size */
3311 xfs_extnum_t to; /* extent list index */
3312
3313 ifp = XFS_IFORK_PTR(ip, whichfork);
3314 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3315 xfs_iext_realloc(ip, count, whichfork);
3316 base = ifp->if_u1.if_extents;
3317 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3318 memmove(&base[idx + count], &base[idx],
3319 (nextents - (idx + count)) * sizeof(*base));
3320 for (to = idx; to < idx + count; to++, new++)
3321 xfs_bmbt_set_all(&base[to], new);
3322}
3323
3324/*
Nathan Scottd8cc8902005-11-02 10:34:53 +11003325 * Helper routine to reset inode di_forkoff field when switching
3326 * attribute fork from local to extent format - we reset it where
3327 * possible to make space available for inline data fork extents.
3328 */
3329STATIC void
3330xfs_bmap_forkoff_reset(
3331 xfs_mount_t *mp,
3332 xfs_inode_t *ip,
3333 int whichfork)
3334{
3335 if (whichfork == XFS_ATTR_FORK &&
3336 (ip->i_d.di_format != XFS_DINODE_FMT_DEV) &&
3337 (ip->i_d.di_format != XFS_DINODE_FMT_UUID) &&
3338 ((mp->m_attroffset >> 3) > ip->i_d.di_forkoff)) {
3339 ip->i_d.di_forkoff = mp->m_attroffset >> 3;
3340 ip->i_df.if_ext_max = XFS_IFORK_DSIZE(ip) /
3341 (uint)sizeof(xfs_bmbt_rec_t);
3342 ip->i_afp->if_ext_max = XFS_IFORK_ASIZE(ip) /
3343 (uint)sizeof(xfs_bmbt_rec_t);
3344 }
3345}
3346
3347/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003348 * Convert a local file to an extents file.
3349 * This code is out of bounds for data forks of regular files,
3350 * since the file data needs to get logged so things will stay consistent.
3351 * (The bmap-level manipulations are ok, though).
3352 */
3353STATIC int /* error */
3354xfs_bmap_local_to_extents(
3355 xfs_trans_t *tp, /* transaction pointer */
3356 xfs_inode_t *ip, /* incore inode pointer */
3357 xfs_fsblock_t *firstblock, /* first block allocated in xaction */
3358 xfs_extlen_t total, /* total blocks needed by transaction */
3359 int *logflagsp, /* inode logging flags */
3360 int whichfork) /* data or attr fork */
3361{
3362 int error; /* error return value */
3363 int flags; /* logging flags returned */
3364#ifdef XFS_BMAP_TRACE
3365 static char fname[] = "xfs_bmap_local_to_extents";
3366#endif
3367 xfs_ifork_t *ifp; /* inode fork pointer */
3368
3369 /*
3370 * We don't want to deal with the case of keeping inode data inline yet.
3371 * So sending the data fork of a regular inode is invalid.
3372 */
3373 ASSERT(!((ip->i_d.di_mode & S_IFMT) == S_IFREG &&
3374 whichfork == XFS_DATA_FORK));
3375 ifp = XFS_IFORK_PTR(ip, whichfork);
3376 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3377 flags = 0;
3378 error = 0;
3379 if (ifp->if_bytes) {
3380 xfs_alloc_arg_t args; /* allocation arguments */
3381 xfs_buf_t *bp; /* buffer for extent list block */
3382 xfs_bmbt_rec_t *ep; /* extent list pointer */
3383
3384 args.tp = tp;
3385 args.mp = ip->i_mount;
3386 ASSERT(ifp->if_flags & XFS_IFINLINE);
3387 /*
3388 * Allocate a block. We know we need only one, since the
3389 * file currently fits in an inode.
3390 */
3391 if (*firstblock == NULLFSBLOCK) {
3392 args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3393 args.type = XFS_ALLOCTYPE_START_BNO;
3394 } else {
3395 args.fsbno = *firstblock;
3396 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3397 }
3398 args.total = total;
3399 args.mod = args.minleft = args.alignment = args.wasdel =
3400 args.isfl = args.minalignslop = 0;
3401 args.minlen = args.maxlen = args.prod = 1;
3402 if ((error = xfs_alloc_vextent(&args)))
3403 goto done;
3404 /*
3405 * Can't fail, the space was reserved.
3406 */
3407 ASSERT(args.fsbno != NULLFSBLOCK);
3408 ASSERT(args.len == 1);
3409 *firstblock = args.fsbno;
3410 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3411 memcpy((char *)XFS_BUF_PTR(bp), ifp->if_u1.if_data,
3412 ifp->if_bytes);
3413 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
Nathan Scottd8cc8902005-11-02 10:34:53 +11003414 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003415 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3416 xfs_iext_realloc(ip, 1, whichfork);
3417 ep = ifp->if_u1.if_extents;
3418 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3419 xfs_bmap_trace_post_update(fname, "new", ip, 0, whichfork);
3420 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3421 ip->i_d.di_nblocks = 1;
3422 XFS_TRANS_MOD_DQUOT_BYINO(args.mp, tp, ip,
3423 XFS_TRANS_DQ_BCOUNT, 1L);
3424 flags |= XFS_ILOG_FEXT(whichfork);
Nathan Scottd8cc8902005-11-02 10:34:53 +11003425 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003426 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
Nathan Scottd8cc8902005-11-02 10:34:53 +11003427 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3428 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003429 ifp->if_flags &= ~XFS_IFINLINE;
3430 ifp->if_flags |= XFS_IFEXTENTS;
3431 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3432 flags |= XFS_ILOG_CORE;
3433done:
3434 *logflagsp = flags;
3435 return error;
3436}
3437
3438xfs_bmbt_rec_t * /* pointer to found extent entry */
3439xfs_bmap_do_search_extents(
3440 xfs_bmbt_rec_t *base, /* base of extent list */
3441 xfs_extnum_t lastx, /* last extent index used */
3442 xfs_extnum_t nextents, /* extent list size */
3443 xfs_fileoff_t bno, /* block number searched for */
3444 int *eofp, /* out: end of file found */
3445 xfs_extnum_t *lastxp, /* out: last extent index */
3446 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
3447 xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
3448{
3449 xfs_bmbt_rec_t *ep; /* extent list entry pointer */
3450 xfs_bmbt_irec_t got; /* extent list entry, decoded */
3451 int high; /* high index of binary search */
3452 int low; /* low index of binary search */
3453
3454 /*
3455 * Initialize the extent entry structure to catch access to
3456 * uninitialized br_startblock field.
3457 */
3458 got.br_startoff = 0xffa5a5a5a5a5a5a5LL;
3459 got.br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3460 got.br_state = XFS_EXT_INVALID;
3461
3462#if XFS_BIG_BLKNOS
3463 got.br_startblock = 0xffffa5a5a5a5a5a5LL;
3464#else
3465 got.br_startblock = 0xffffa5a5;
3466#endif
3467
3468 if (lastx != NULLEXTNUM && lastx < nextents)
3469 ep = base + lastx;
3470 else
3471 ep = NULL;
3472 prevp->br_startoff = NULLFILEOFF;
3473 if (ep && bno >= (got.br_startoff = xfs_bmbt_get_startoff(ep)) &&
3474 bno < got.br_startoff +
3475 (got.br_blockcount = xfs_bmbt_get_blockcount(ep)))
3476 *eofp = 0;
3477 else if (ep && lastx < nextents - 1 &&
3478 bno >= (got.br_startoff = xfs_bmbt_get_startoff(ep + 1)) &&
3479 bno < got.br_startoff +
3480 (got.br_blockcount = xfs_bmbt_get_blockcount(ep + 1))) {
3481 lastx++;
3482 ep++;
3483 *eofp = 0;
3484 } else if (nextents == 0)
3485 *eofp = 1;
3486 else if (bno == 0 &&
3487 (got.br_startoff = xfs_bmbt_get_startoff(base)) == 0) {
3488 ep = base;
3489 lastx = 0;
3490 got.br_blockcount = xfs_bmbt_get_blockcount(ep);
3491 *eofp = 0;
3492 } else {
3493 /* binary search the extents array */
3494 low = 0;
3495 high = nextents - 1;
3496 while (low <= high) {
3497 XFS_STATS_INC(xs_cmp_exlist);
3498 lastx = (low + high) >> 1;
3499 ep = base + lastx;
3500 got.br_startoff = xfs_bmbt_get_startoff(ep);
3501 got.br_blockcount = xfs_bmbt_get_blockcount(ep);
3502 if (bno < got.br_startoff)
3503 high = lastx - 1;
3504 else if (bno >= got.br_startoff + got.br_blockcount)
3505 low = lastx + 1;
3506 else {
3507 got.br_startblock = xfs_bmbt_get_startblock(ep);
3508 got.br_state = xfs_bmbt_get_state(ep);
3509 *eofp = 0;
3510 *lastxp = lastx;
3511 *gotp = got;
3512 return ep;
3513 }
3514 }
3515 if (bno >= got.br_startoff + got.br_blockcount) {
3516 lastx++;
3517 if (lastx == nextents) {
3518 *eofp = 1;
3519 got.br_startblock = xfs_bmbt_get_startblock(ep);
3520 got.br_state = xfs_bmbt_get_state(ep);
3521 *prevp = got;
3522 ep = NULL;
3523 } else {
3524 *eofp = 0;
3525 xfs_bmbt_get_all(ep, prevp);
3526 ep++;
3527 got.br_startoff = xfs_bmbt_get_startoff(ep);
3528 got.br_blockcount = xfs_bmbt_get_blockcount(ep);
3529 }
3530 } else {
3531 *eofp = 0;
3532 if (ep > base)
3533 xfs_bmbt_get_all(ep - 1, prevp);
3534 }
3535 }
3536 if (ep) {
3537 got.br_startblock = xfs_bmbt_get_startblock(ep);
3538 got.br_state = xfs_bmbt_get_state(ep);
3539 }
3540 *lastxp = lastx;
3541 *gotp = got;
3542 return ep;
3543}
3544
3545/*
3546 * Search the extents list for the inode, for the extent containing bno.
3547 * If bno lies in a hole, point to the next entry. If bno lies past eof,
3548 * *eofp will be set, and *prevp will contain the last entry (null if none).
3549 * Else, *lastxp will be set to the index of the found
3550 * entry; *gotp will contain the entry.
3551 */
3552STATIC xfs_bmbt_rec_t * /* pointer to found extent entry */
3553xfs_bmap_search_extents(
3554 xfs_inode_t *ip, /* incore inode pointer */
3555 xfs_fileoff_t bno, /* block number searched for */
3556 int whichfork, /* data or attr fork */
3557 int *eofp, /* out: end of file found */
3558 xfs_extnum_t *lastxp, /* out: last extent index */
3559 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
3560 xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
3561{
3562 xfs_ifork_t *ifp; /* inode fork pointer */
3563 xfs_bmbt_rec_t *base; /* base of extent list */
3564 xfs_extnum_t lastx; /* last extent index used */
3565 xfs_extnum_t nextents; /* extent list size */
3566 xfs_bmbt_rec_t *ep; /* extent list entry pointer */
3567 int rt; /* realtime flag */
3568
3569 XFS_STATS_INC(xs_look_exlist);
3570 ifp = XFS_IFORK_PTR(ip, whichfork);
3571 lastx = ifp->if_lastex;
3572 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3573 base = &ifp->if_u1.if_extents[0];
3574
3575 ep = xfs_bmap_do_search_extents(base, lastx, nextents, bno, eofp,
3576 lastxp, gotp, prevp);
3577 rt = ip->i_d.di_flags & XFS_DIFLAG_REALTIME;
3578 if(!rt && !gotp->br_startblock && (*lastxp != NULLEXTNUM)) {
3579 cmn_err(CE_PANIC,"Access to block zero: fs: <%s> inode: %lld "
3580 "start_block : %llx start_off : %llx blkcnt : %llx "
3581 "extent-state : %x \n",
3582 (ip->i_mount)->m_fsname,(long long)ip->i_ino,
3583 gotp->br_startblock, gotp->br_startoff,
3584 gotp->br_blockcount,gotp->br_state);
3585 }
3586 return ep;
3587}
3588
3589
3590#ifdef XFS_BMAP_TRACE
3591ktrace_t *xfs_bmap_trace_buf;
3592
3593/*
3594 * Add a bmap trace buffer entry. Base routine for the others.
3595 */
3596STATIC void
3597xfs_bmap_trace_addentry(
3598 int opcode, /* operation */
3599 char *fname, /* function name */
3600 char *desc, /* operation description */
3601 xfs_inode_t *ip, /* incore inode pointer */
3602 xfs_extnum_t idx, /* index of entry(ies) */
3603 xfs_extnum_t cnt, /* count of entries, 1 or 2 */
3604 xfs_bmbt_rec_t *r1, /* first record */
3605 xfs_bmbt_rec_t *r2, /* second record or null */
3606 int whichfork) /* data or attr fork */
3607{
3608 xfs_bmbt_rec_t tr2;
3609
3610 ASSERT(cnt == 1 || cnt == 2);
3611 ASSERT(r1 != NULL);
3612 if (cnt == 1) {
3613 ASSERT(r2 == NULL);
3614 r2 = &tr2;
3615 memset(&tr2, 0, sizeof(tr2));
3616 } else
3617 ASSERT(r2 != NULL);
3618 ktrace_enter(xfs_bmap_trace_buf,
3619 (void *)(__psint_t)(opcode | (whichfork << 16)),
3620 (void *)fname, (void *)desc, (void *)ip,
3621 (void *)(__psint_t)idx,
3622 (void *)(__psint_t)cnt,
3623 (void *)(__psunsigned_t)(ip->i_ino >> 32),
3624 (void *)(__psunsigned_t)(unsigned)ip->i_ino,
3625 (void *)(__psunsigned_t)(r1->l0 >> 32),
3626 (void *)(__psunsigned_t)(unsigned)(r1->l0),
3627 (void *)(__psunsigned_t)(r1->l1 >> 32),
3628 (void *)(__psunsigned_t)(unsigned)(r1->l1),
3629 (void *)(__psunsigned_t)(r2->l0 >> 32),
3630 (void *)(__psunsigned_t)(unsigned)(r2->l0),
3631 (void *)(__psunsigned_t)(r2->l1 >> 32),
3632 (void *)(__psunsigned_t)(unsigned)(r2->l1)
3633 );
3634 ASSERT(ip->i_xtrace);
3635 ktrace_enter(ip->i_xtrace,
3636 (void *)(__psint_t)(opcode | (whichfork << 16)),
3637 (void *)fname, (void *)desc, (void *)ip,
3638 (void *)(__psint_t)idx,
3639 (void *)(__psint_t)cnt,
3640 (void *)(__psunsigned_t)(ip->i_ino >> 32),
3641 (void *)(__psunsigned_t)(unsigned)ip->i_ino,
3642 (void *)(__psunsigned_t)(r1->l0 >> 32),
3643 (void *)(__psunsigned_t)(unsigned)(r1->l0),
3644 (void *)(__psunsigned_t)(r1->l1 >> 32),
3645 (void *)(__psunsigned_t)(unsigned)(r1->l1),
3646 (void *)(__psunsigned_t)(r2->l0 >> 32),
3647 (void *)(__psunsigned_t)(unsigned)(r2->l0),
3648 (void *)(__psunsigned_t)(r2->l1 >> 32),
3649 (void *)(__psunsigned_t)(unsigned)(r2->l1)
3650 );
3651}
3652
3653/*
3654 * Add bmap trace entry prior to a call to xfs_bmap_delete_exlist.
3655 */
3656STATIC void
3657xfs_bmap_trace_delete(
3658 char *fname, /* function name */
3659 char *desc, /* operation description */
3660 xfs_inode_t *ip, /* incore inode pointer */
3661 xfs_extnum_t idx, /* index of entry(entries) deleted */
3662 xfs_extnum_t cnt, /* count of entries deleted, 1 or 2 */
3663 int whichfork) /* data or attr fork */
3664{
3665 xfs_ifork_t *ifp; /* inode fork pointer */
3666
3667 ifp = XFS_IFORK_PTR(ip, whichfork);
3668 xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_DELETE, fname, desc, ip, idx,
3669 cnt, &ifp->if_u1.if_extents[idx],
3670 cnt == 2 ? &ifp->if_u1.if_extents[idx + 1] : NULL,
3671 whichfork);
3672}
3673
3674/*
3675 * Add bmap trace entry prior to a call to xfs_bmap_insert_exlist, or
3676 * reading in the extents list from the disk (in the btree).
3677 */
3678STATIC void
3679xfs_bmap_trace_insert(
3680 char *fname, /* function name */
3681 char *desc, /* operation description */
3682 xfs_inode_t *ip, /* incore inode pointer */
3683 xfs_extnum_t idx, /* index of entry(entries) inserted */
3684 xfs_extnum_t cnt, /* count of entries inserted, 1 or 2 */
3685 xfs_bmbt_irec_t *r1, /* inserted record 1 */
3686 xfs_bmbt_irec_t *r2, /* inserted record 2 or null */
3687 int whichfork) /* data or attr fork */
3688{
3689 xfs_bmbt_rec_t tr1; /* compressed record 1 */
3690 xfs_bmbt_rec_t tr2; /* compressed record 2 if needed */
3691
3692 xfs_bmbt_set_all(&tr1, r1);
3693 if (cnt == 2) {
3694 ASSERT(r2 != NULL);
3695 xfs_bmbt_set_all(&tr2, r2);
3696 } else {
3697 ASSERT(cnt == 1);
3698 ASSERT(r2 == NULL);
3699 }
3700 xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_INSERT, fname, desc, ip, idx,
3701 cnt, &tr1, cnt == 2 ? &tr2 : NULL, whichfork);
3702}
3703
3704/*
3705 * Add bmap trace entry after updating an extent list entry in place.
3706 */
3707STATIC void
3708xfs_bmap_trace_post_update(
3709 char *fname, /* function name */
3710 char *desc, /* operation description */
3711 xfs_inode_t *ip, /* incore inode pointer */
3712 xfs_extnum_t idx, /* index of entry updated */
3713 int whichfork) /* data or attr fork */
3714{
3715 xfs_ifork_t *ifp; /* inode fork pointer */
3716
3717 ifp = XFS_IFORK_PTR(ip, whichfork);
3718 xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_POST_UP, fname, desc, ip, idx,
3719 1, &ifp->if_u1.if_extents[idx], NULL, whichfork);
3720}
3721
3722/*
3723 * Add bmap trace entry prior to updating an extent list entry in place.
3724 */
3725STATIC void
3726xfs_bmap_trace_pre_update(
3727 char *fname, /* function name */
3728 char *desc, /* operation description */
3729 xfs_inode_t *ip, /* incore inode pointer */
3730 xfs_extnum_t idx, /* index of entry to be updated */
3731 int whichfork) /* data or attr fork */
3732{
3733 xfs_ifork_t *ifp; /* inode fork pointer */
3734
3735 ifp = XFS_IFORK_PTR(ip, whichfork);
3736 xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_PRE_UP, fname, desc, ip, idx, 1,
3737 &ifp->if_u1.if_extents[idx], NULL, whichfork);
3738}
3739#endif /* XFS_BMAP_TRACE */
3740
3741/*
3742 * Compute the worst-case number of indirect blocks that will be used
3743 * for ip's delayed extent of length "len".
3744 */
3745STATIC xfs_filblks_t
3746xfs_bmap_worst_indlen(
3747 xfs_inode_t *ip, /* incore inode pointer */
3748 xfs_filblks_t len) /* delayed extent length */
3749{
3750 int level; /* btree level number */
3751 int maxrecs; /* maximum record count at this level */
3752 xfs_mount_t *mp; /* mount structure */
3753 xfs_filblks_t rval; /* return value */
3754
3755 mp = ip->i_mount;
3756 maxrecs = mp->m_bmap_dmxr[0];
3757 for (level = 0, rval = 0;
3758 level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3759 level++) {
3760 len += maxrecs - 1;
3761 do_div(len, maxrecs);
3762 rval += len;
3763 if (len == 1)
3764 return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3765 level - 1;
3766 if (level == 0)
3767 maxrecs = mp->m_bmap_dmxr[1];
3768 }
3769 return rval;
3770}
3771
3772#if defined(XFS_RW_TRACE)
3773STATIC void
3774xfs_bunmap_trace(
3775 xfs_inode_t *ip,
3776 xfs_fileoff_t bno,
3777 xfs_filblks_t len,
3778 int flags,
3779 inst_t *ra)
3780{
3781 if (ip->i_rwtrace == NULL)
3782 return;
3783 ktrace_enter(ip->i_rwtrace,
3784 (void *)(__psint_t)XFS_BUNMAPI,
3785 (void *)ip,
3786 (void *)(__psint_t)((ip->i_d.di_size >> 32) & 0xffffffff),
3787 (void *)(__psint_t)(ip->i_d.di_size & 0xffffffff),
3788 (void *)(__psint_t)(((xfs_dfiloff_t)bno >> 32) & 0xffffffff),
3789 (void *)(__psint_t)((xfs_dfiloff_t)bno & 0xffffffff),
3790 (void *)(__psint_t)len,
3791 (void *)(__psint_t)flags,
3792 (void *)(unsigned long)current_cpu(),
3793 (void *)ra,
3794 (void *)0,
3795 (void *)0,
3796 (void *)0,
3797 (void *)0,
3798 (void *)0,
3799 (void *)0);
3800}
3801#endif
3802
3803/*
3804 * Convert inode from non-attributed to attributed.
3805 * Must not be in a transaction, ip must not be locked.
3806 */
3807int /* error code */
3808xfs_bmap_add_attrfork(
3809 xfs_inode_t *ip, /* incore inode pointer */
Nathan Scottd8cc8902005-11-02 10:34:53 +11003810 int size, /* space new attribute needs */
3811 int rsvd) /* xact may use reserved blks */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003812{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003813 xfs_fsblock_t firstblock; /* 1st block/ag allocated */
3814 xfs_bmap_free_t flist; /* freed extent list */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003815 xfs_mount_t *mp; /* mount structure */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003816 xfs_trans_t *tp; /* transaction pointer */
Nathan Scottd8cc8902005-11-02 10:34:53 +11003817 unsigned long s; /* spinlock spl value */
3818 int blks; /* space reservation */
3819 int version = 1; /* superblock attr version */
3820 int committed; /* xaction was committed */
3821 int logflags; /* logging flags */
3822 int error; /* error return value */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003823
Nathan Scottd8cc8902005-11-02 10:34:53 +11003824 ASSERT(XFS_IFORK_Q(ip) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003825 ASSERT(ip->i_df.if_ext_max ==
3826 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
Nathan Scottd8cc8902005-11-02 10:34:53 +11003827
Linus Torvalds1da177e2005-04-16 15:20:36 -07003828 mp = ip->i_mount;
3829 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3830 tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3831 blks = XFS_ADDAFORK_SPACE_RES(mp);
3832 if (rsvd)
3833 tp->t_flags |= XFS_TRANS_RESERVE;
3834 if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3835 XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3836 goto error0;
3837 xfs_ilock(ip, XFS_ILOCK_EXCL);
3838 error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip, blks, 0, rsvd ?
3839 XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3840 XFS_QMOPT_RES_REGBLKS);
3841 if (error) {
3842 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3843 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3844 return error;
3845 }
3846 if (XFS_IFORK_Q(ip))
3847 goto error1;
3848 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3849 /*
3850 * For inodes coming from pre-6.2 filesystems.
3851 */
3852 ASSERT(ip->i_d.di_aformat == 0);
3853 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
3854 }
3855 ASSERT(ip->i_d.di_anextents == 0);
3856 VN_HOLD(XFS_ITOV(ip));
3857 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3858 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3859 switch (ip->i_d.di_format) {
3860 case XFS_DINODE_FMT_DEV:
3861 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
3862 break;
3863 case XFS_DINODE_FMT_UUID:
3864 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
3865 break;
3866 case XFS_DINODE_FMT_LOCAL:
3867 case XFS_DINODE_FMT_EXTENTS:
3868 case XFS_DINODE_FMT_BTREE:
Nathan Scottd8cc8902005-11-02 10:34:53 +11003869 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
3870 if (!ip->i_d.di_forkoff)
3871 ip->i_d.di_forkoff = mp->m_attroffset >> 3;
3872 else if (!(mp->m_flags & XFS_MOUNT_COMPAT_ATTR))
3873 version = 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003874 break;
3875 default:
3876 ASSERT(0);
3877 error = XFS_ERROR(EINVAL);
3878 goto error1;
3879 }
3880 ip->i_df.if_ext_max =
3881 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3882 ASSERT(ip->i_afp == NULL);
3883 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
3884 ip->i_afp->if_ext_max =
3885 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3886 ip->i_afp->if_flags = XFS_IFEXTENTS;
3887 logflags = 0;
3888 XFS_BMAP_INIT(&flist, &firstblock);
3889 switch (ip->i_d.di_format) {
3890 case XFS_DINODE_FMT_LOCAL:
3891 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
3892 &logflags);
3893 break;
3894 case XFS_DINODE_FMT_EXTENTS:
3895 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
3896 &flist, &logflags);
3897 break;
3898 case XFS_DINODE_FMT_BTREE:
3899 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
3900 &logflags);
3901 break;
3902 default:
3903 error = 0;
3904 break;
3905 }
3906 if (logflags)
3907 xfs_trans_log_inode(tp, ip, logflags);
3908 if (error)
3909 goto error2;
Nathan Scottd8cc8902005-11-02 10:34:53 +11003910 if (!XFS_SB_VERSION_HASATTR(&mp->m_sb) ||
3911 (!XFS_SB_VERSION_HASATTR2(&mp->m_sb) && version == 2)) {
Nathan Scottda087ba2005-11-02 15:00:20 +11003912 __int64_t sbfields = 0;
3913
Linus Torvalds1da177e2005-04-16 15:20:36 -07003914 s = XFS_SB_LOCK(mp);
3915 if (!XFS_SB_VERSION_HASATTR(&mp->m_sb)) {
3916 XFS_SB_VERSION_ADDATTR(&mp->m_sb);
Nathan Scottda087ba2005-11-02 15:00:20 +11003917 sbfields |= XFS_SB_VERSIONNUM;
Nathan Scottd8cc8902005-11-02 10:34:53 +11003918 }
3919 if (!XFS_SB_VERSION_HASATTR2(&mp->m_sb) && version == 2) {
3920 XFS_SB_VERSION_ADDATTR2(&mp->m_sb);
Nathan Scottda087ba2005-11-02 15:00:20 +11003921 sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
Nathan Scottd8cc8902005-11-02 10:34:53 +11003922 }
Nathan Scottda087ba2005-11-02 15:00:20 +11003923 if (sbfields) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003924 XFS_SB_UNLOCK(mp, s);
Nathan Scottda087ba2005-11-02 15:00:20 +11003925 xfs_mod_sb(tp, sbfields);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003926 } else
3927 XFS_SB_UNLOCK(mp, s);
3928 }
3929 if ((error = xfs_bmap_finish(&tp, &flist, firstblock, &committed)))
3930 goto error2;
3931 error = xfs_trans_commit(tp, XFS_TRANS_PERM_LOG_RES, NULL);
3932 ASSERT(ip->i_df.if_ext_max ==
3933 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3934 return error;
3935error2:
3936 xfs_bmap_cancel(&flist);
3937error1:
3938 ASSERT(ismrlocked(&ip->i_lock,MR_UPDATE));
3939 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3940error0:
3941 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
3942 ASSERT(ip->i_df.if_ext_max ==
3943 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3944 return error;
3945}
3946
3947/*
3948 * Add the extent to the list of extents to be free at transaction end.
3949 * The list is maintained sorted (by block number).
3950 */
3951/* ARGSUSED */
3952void
3953xfs_bmap_add_free(
3954 xfs_fsblock_t bno, /* fs block number of extent */
3955 xfs_filblks_t len, /* length of extent */
3956 xfs_bmap_free_t *flist, /* list of extents */
3957 xfs_mount_t *mp) /* mount point structure */
3958{
3959 xfs_bmap_free_item_t *cur; /* current (next) element */
3960 xfs_bmap_free_item_t *new; /* new element */
3961 xfs_bmap_free_item_t *prev; /* previous element */
3962#ifdef DEBUG
3963 xfs_agnumber_t agno;
3964 xfs_agblock_t agbno;
3965
3966 ASSERT(bno != NULLFSBLOCK);
3967 ASSERT(len > 0);
3968 ASSERT(len <= MAXEXTLEN);
3969 ASSERT(!ISNULLSTARTBLOCK(bno));
3970 agno = XFS_FSB_TO_AGNO(mp, bno);
3971 agbno = XFS_FSB_TO_AGBNO(mp, bno);
3972 ASSERT(agno < mp->m_sb.sb_agcount);
3973 ASSERT(agbno < mp->m_sb.sb_agblocks);
3974 ASSERT(len < mp->m_sb.sb_agblocks);
3975 ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
3976#endif
3977 ASSERT(xfs_bmap_free_item_zone != NULL);
3978 new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
3979 new->xbfi_startblock = bno;
3980 new->xbfi_blockcount = (xfs_extlen_t)len;
3981 for (prev = NULL, cur = flist->xbf_first;
3982 cur != NULL;
3983 prev = cur, cur = cur->xbfi_next) {
3984 if (cur->xbfi_startblock >= bno)
3985 break;
3986 }
3987 if (prev)
3988 prev->xbfi_next = new;
3989 else
3990 flist->xbf_first = new;
3991 new->xbfi_next = cur;
3992 flist->xbf_count++;
3993}
3994
3995/*
3996 * Compute and fill in the value of the maximum depth of a bmap btree
3997 * in this filesystem. Done once, during mount.
3998 */
3999void
4000xfs_bmap_compute_maxlevels(
4001 xfs_mount_t *mp, /* file system mount structure */
4002 int whichfork) /* data or attr fork */
4003{
4004 int level; /* btree level */
4005 uint maxblocks; /* max blocks at this level */
4006 uint maxleafents; /* max leaf entries possible */
4007 int maxrootrecs; /* max records in root block */
4008 int minleafrecs; /* min records in leaf block */
4009 int minnoderecs; /* min records in node block */
4010 int sz; /* root block size */
4011
4012 /*
4013 * The maximum number of extents in a file, hence the maximum
4014 * number of leaf entries, is controlled by the type of di_nextents
4015 * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
4016 * (a signed 16-bit number, xfs_aextnum_t).
4017 */
Nathan Scottd8cc8902005-11-02 10:34:53 +11004018 if (whichfork == XFS_DATA_FORK) {
4019 maxleafents = MAXEXTNUM;
4020 sz = (mp->m_flags & XFS_MOUNT_COMPAT_ATTR) ?
4021 mp->m_attroffset : XFS_BMDR_SPACE_CALC(MINDBTPTRS);
4022 } else {
4023 maxleafents = MAXAEXTNUM;
4024 sz = (mp->m_flags & XFS_MOUNT_COMPAT_ATTR) ?
4025 mp->m_sb.sb_inodesize - mp->m_attroffset :
4026 XFS_BMDR_SPACE_CALC(MINABTPTRS);
4027 }
4028 maxrootrecs = (int)XFS_BTREE_BLOCK_MAXRECS(sz, xfs_bmdr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004029 minleafrecs = mp->m_bmap_dmnr[0];
4030 minnoderecs = mp->m_bmap_dmnr[1];
Linus Torvalds1da177e2005-04-16 15:20:36 -07004031 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
4032 for (level = 1; maxblocks > 1; level++) {
4033 if (maxblocks <= maxrootrecs)
4034 maxblocks = 1;
4035 else
4036 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
4037 }
4038 mp->m_bm_maxlevels[whichfork] = level;
4039}
4040
4041/*
4042 * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
4043 * caller. Frees all the extents that need freeing, which must be done
4044 * last due to locking considerations. We never free any extents in
4045 * the first transaction. This is to allow the caller to make the first
4046 * transaction a synchronous one so that the pointers to the data being
4047 * broken in this transaction will be permanent before the data is actually
4048 * freed. This is necessary to prevent blocks from being reallocated
4049 * and written to before the free and reallocation are actually permanent.
4050 * We do not just make the first transaction synchronous here, because
4051 * there are more efficient ways to gain the same protection in some cases
4052 * (see the file truncation code).
4053 *
4054 * Return 1 if the given transaction was committed and a new one
4055 * started, and 0 otherwise in the committed parameter.
4056 */
4057/*ARGSUSED*/
4058int /* error */
4059xfs_bmap_finish(
4060 xfs_trans_t **tp, /* transaction pointer addr */
4061 xfs_bmap_free_t *flist, /* i/o: list extents to free */
4062 xfs_fsblock_t firstblock, /* controlled ag for allocs */
4063 int *committed) /* xact committed or not */
4064{
4065 xfs_efd_log_item_t *efd; /* extent free data */
4066 xfs_efi_log_item_t *efi; /* extent free intention */
4067 int error; /* error return value */
4068 xfs_bmap_free_item_t *free; /* free extent list item */
4069 unsigned int logres; /* new log reservation */
4070 unsigned int logcount; /* new log count */
4071 xfs_mount_t *mp; /* filesystem mount structure */
4072 xfs_bmap_free_item_t *next; /* next item on free list */
4073 xfs_trans_t *ntp; /* new transaction pointer */
4074
4075 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
4076 if (flist->xbf_count == 0) {
4077 *committed = 0;
4078 return 0;
4079 }
4080 ntp = *tp;
4081 efi = xfs_trans_get_efi(ntp, flist->xbf_count);
4082 for (free = flist->xbf_first; free; free = free->xbfi_next)
4083 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
4084 free->xbfi_blockcount);
4085 logres = ntp->t_log_res;
4086 logcount = ntp->t_log_count;
4087 ntp = xfs_trans_dup(*tp);
4088 error = xfs_trans_commit(*tp, 0, NULL);
4089 *tp = ntp;
4090 *committed = 1;
4091 /*
4092 * We have a new transaction, so we should return committed=1,
4093 * even though we're returning an error.
4094 */
4095 if (error) {
4096 return error;
4097 }
4098 if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
4099 logcount)))
4100 return error;
4101 efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
4102 for (free = flist->xbf_first; free != NULL; free = next) {
4103 next = free->xbfi_next;
4104 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
4105 free->xbfi_blockcount))) {
4106 /*
4107 * The bmap free list will be cleaned up at a
4108 * higher level. The EFI will be canceled when
4109 * this transaction is aborted.
4110 * Need to force shutdown here to make sure it
4111 * happens, since this transaction may not be
4112 * dirty yet.
4113 */
4114 mp = ntp->t_mountp;
4115 if (!XFS_FORCED_SHUTDOWN(mp))
4116 xfs_force_shutdown(mp,
4117 (error == EFSCORRUPTED) ?
4118 XFS_CORRUPT_INCORE :
4119 XFS_METADATA_IO_ERROR);
4120 return error;
4121 }
4122 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
4123 free->xbfi_blockcount);
4124 xfs_bmap_del_free(flist, NULL, free);
4125 }
4126 return 0;
4127}
4128
4129/*
4130 * Free up any items left in the list.
4131 */
4132void
4133xfs_bmap_cancel(
4134 xfs_bmap_free_t *flist) /* list of bmap_free_items */
4135{
4136 xfs_bmap_free_item_t *free; /* free list item */
4137 xfs_bmap_free_item_t *next;
4138
4139 if (flist->xbf_count == 0)
4140 return;
4141 ASSERT(flist->xbf_first != NULL);
4142 for (free = flist->xbf_first; free; free = next) {
4143 next = free->xbfi_next;
4144 xfs_bmap_del_free(flist, NULL, free);
4145 }
4146 ASSERT(flist->xbf_count == 0);
4147}
4148
4149/*
4150 * Returns the file-relative block number of the first unused block(s)
4151 * in the file with at least "len" logically contiguous blocks free.
4152 * This is the lowest-address hole if the file has holes, else the first block
4153 * past the end of file.
4154 * Return 0 if the file is currently local (in-inode).
4155 */
4156int /* error */
4157xfs_bmap_first_unused(
4158 xfs_trans_t *tp, /* transaction pointer */
4159 xfs_inode_t *ip, /* incore inode */
4160 xfs_extlen_t len, /* size of hole to find */
4161 xfs_fileoff_t *first_unused, /* unused block */
4162 int whichfork) /* data or attr fork */
4163{
4164 xfs_bmbt_rec_t *base; /* base of extent array */
4165 xfs_bmbt_rec_t *ep; /* pointer to an extent entry */
4166 int error; /* error return value */
4167 xfs_ifork_t *ifp; /* inode fork pointer */
4168 xfs_fileoff_t lastaddr; /* last block number seen */
4169 xfs_fileoff_t lowest; /* lowest useful block */
4170 xfs_fileoff_t max; /* starting useful block */
4171 xfs_fileoff_t off; /* offset for this block */
4172 xfs_extnum_t nextents; /* number of extent entries */
4173
4174 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
4175 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
4176 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
4177 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4178 *first_unused = 0;
4179 return 0;
4180 }
4181 ifp = XFS_IFORK_PTR(ip, whichfork);
4182 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4183 (error = xfs_iread_extents(tp, ip, whichfork)))
4184 return error;
4185 lowest = *first_unused;
4186 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4187 base = &ifp->if_u1.if_extents[0];
4188 for (lastaddr = 0, max = lowest, ep = base;
4189 ep < &base[nextents];
4190 ep++) {
4191 off = xfs_bmbt_get_startoff(ep);
4192 /*
4193 * See if the hole before this extent will work.
4194 */
4195 if (off >= lowest + len && off - max >= len) {
4196 *first_unused = max;
4197 return 0;
4198 }
4199 lastaddr = off + xfs_bmbt_get_blockcount(ep);
4200 max = XFS_FILEOFF_MAX(lastaddr, lowest);
4201 }
4202 *first_unused = max;
4203 return 0;
4204}
4205
4206/*
4207 * Returns the file-relative block number of the last block + 1 before
4208 * last_block (input value) in the file.
4209 * This is not based on i_size, it is based on the extent list.
4210 * Returns 0 for local files, as they do not have an extent list.
4211 */
4212int /* error */
4213xfs_bmap_last_before(
4214 xfs_trans_t *tp, /* transaction pointer */
4215 xfs_inode_t *ip, /* incore inode */
4216 xfs_fileoff_t *last_block, /* last block */
4217 int whichfork) /* data or attr fork */
4218{
4219 xfs_fileoff_t bno; /* input file offset */
4220 int eof; /* hit end of file */
4221 xfs_bmbt_rec_t *ep; /* pointer to last extent */
4222 int error; /* error return value */
4223 xfs_bmbt_irec_t got; /* current extent value */
4224 xfs_ifork_t *ifp; /* inode fork pointer */
4225 xfs_extnum_t lastx; /* last extent used */
4226 xfs_bmbt_irec_t prev; /* previous extent value */
4227
4228 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4229 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4230 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4231 return XFS_ERROR(EIO);
4232 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4233 *last_block = 0;
4234 return 0;
4235 }
4236 ifp = XFS_IFORK_PTR(ip, whichfork);
4237 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4238 (error = xfs_iread_extents(tp, ip, whichfork)))
4239 return error;
4240 bno = *last_block - 1;
4241 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4242 &prev);
4243 if (eof || xfs_bmbt_get_startoff(ep) > bno) {
4244 if (prev.br_startoff == NULLFILEOFF)
4245 *last_block = 0;
4246 else
4247 *last_block = prev.br_startoff + prev.br_blockcount;
4248 }
4249 /*
4250 * Otherwise *last_block is already the right answer.
4251 */
4252 return 0;
4253}
4254
4255/*
4256 * Returns the file-relative block number of the first block past eof in
4257 * the file. This is not based on i_size, it is based on the extent list.
4258 * Returns 0 for local files, as they do not have an extent list.
4259 */
4260int /* error */
4261xfs_bmap_last_offset(
4262 xfs_trans_t *tp, /* transaction pointer */
4263 xfs_inode_t *ip, /* incore inode */
4264 xfs_fileoff_t *last_block, /* last block */
4265 int whichfork) /* data or attr fork */
4266{
4267 xfs_bmbt_rec_t *base; /* base of extent array */
4268 xfs_bmbt_rec_t *ep; /* pointer to last extent */
4269 int error; /* error return value */
4270 xfs_ifork_t *ifp; /* inode fork pointer */
4271 xfs_extnum_t nextents; /* number of extent entries */
4272
4273 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4274 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4275 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4276 return XFS_ERROR(EIO);
4277 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4278 *last_block = 0;
4279 return 0;
4280 }
4281 ifp = XFS_IFORK_PTR(ip, whichfork);
4282 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4283 (error = xfs_iread_extents(tp, ip, whichfork)))
4284 return error;
4285 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4286 if (!nextents) {
4287 *last_block = 0;
4288 return 0;
4289 }
4290 base = &ifp->if_u1.if_extents[0];
4291 ASSERT(base != NULL);
4292 ep = &base[nextents - 1];
4293 *last_block = xfs_bmbt_get_startoff(ep) + xfs_bmbt_get_blockcount(ep);
4294 return 0;
4295}
4296
4297/*
4298 * Returns whether the selected fork of the inode has exactly one
4299 * block or not. For the data fork we check this matches di_size,
4300 * implying the file's range is 0..bsize-1.
4301 */
4302int /* 1=>1 block, 0=>otherwise */
4303xfs_bmap_one_block(
4304 xfs_inode_t *ip, /* incore inode */
4305 int whichfork) /* data or attr fork */
4306{
4307 xfs_bmbt_rec_t *ep; /* ptr to fork's extent */
4308 xfs_ifork_t *ifp; /* inode fork pointer */
4309 int rval; /* return value */
4310 xfs_bmbt_irec_t s; /* internal version of extent */
4311
4312#ifndef DEBUG
4313 if (whichfork == XFS_DATA_FORK)
4314 return ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize;
4315#endif /* !DEBUG */
4316 if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4317 return 0;
4318 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4319 return 0;
4320 ifp = XFS_IFORK_PTR(ip, whichfork);
4321 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4322 ep = ifp->if_u1.if_extents;
4323 xfs_bmbt_get_all(ep, &s);
4324 rval = s.br_startoff == 0 && s.br_blockcount == 1;
4325 if (rval && whichfork == XFS_DATA_FORK)
4326 ASSERT(ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize);
4327 return rval;
4328}
4329
4330/*
4331 * Read in the extents to if_extents.
4332 * All inode fields are set up by caller, we just traverse the btree
4333 * and copy the records in. If the file system cannot contain unwritten
4334 * extents, the records are checked for no "state" flags.
4335 */
4336int /* error */
4337xfs_bmap_read_extents(
4338 xfs_trans_t *tp, /* transaction pointer */
4339 xfs_inode_t *ip, /* incore inode */
4340 int whichfork) /* data or attr fork */
4341{
4342 xfs_bmbt_block_t *block; /* current btree block */
4343 xfs_fsblock_t bno; /* block # of "block" */
4344 xfs_buf_t *bp; /* buffer for "block" */
4345 int error; /* error return value */
4346 xfs_exntfmt_t exntf; /* XFS_EXTFMT_NOSTATE, if checking */
4347#ifdef XFS_BMAP_TRACE
4348 static char fname[] = "xfs_bmap_read_extents";
4349#endif
4350 xfs_extnum_t i, j; /* index into the extents list */
4351 xfs_ifork_t *ifp; /* fork structure */
4352 int level; /* btree level, for checking */
4353 xfs_mount_t *mp; /* file system mount structure */
4354 xfs_bmbt_ptr_t *pp; /* pointer to block address */
4355 /* REFERENCED */
4356 xfs_extnum_t room; /* number of entries there's room for */
4357 xfs_bmbt_rec_t *trp; /* target record pointer */
4358
4359 bno = NULLFSBLOCK;
4360 mp = ip->i_mount;
4361 ifp = XFS_IFORK_PTR(ip, whichfork);
4362 exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4363 XFS_EXTFMT_INODE(ip);
4364 block = ifp->if_broot;
4365 /*
4366 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4367 */
4368 ASSERT(INT_GET(block->bb_level, ARCH_CONVERT) > 0);
4369 level = INT_GET(block->bb_level, ARCH_CONVERT);
4370 pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
4371 ASSERT(INT_GET(*pp, ARCH_CONVERT) != NULLDFSBNO);
4372 ASSERT(XFS_FSB_TO_AGNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agcount);
4373 ASSERT(XFS_FSB_TO_AGBNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agblocks);
4374 bno = INT_GET(*pp, ARCH_CONVERT);
4375 /*
4376 * Go down the tree until leaf level is reached, following the first
4377 * pointer (leftmost) at each level.
4378 */
4379 while (level-- > 0) {
4380 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4381 XFS_BMAP_BTREE_REF)))
4382 return error;
4383 block = XFS_BUF_TO_BMBT_BLOCK(bp);
4384 XFS_WANT_CORRUPTED_GOTO(
4385 XFS_BMAP_SANITY_CHECK(mp, block, level),
4386 error0);
4387 if (level == 0)
4388 break;
4389 pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt, block,
4390 1, mp->m_bmap_dmxr[1]);
4391 XFS_WANT_CORRUPTED_GOTO(
4392 XFS_FSB_SANITY_CHECK(mp, INT_GET(*pp, ARCH_CONVERT)),
4393 error0);
4394 bno = INT_GET(*pp, ARCH_CONVERT);
4395 xfs_trans_brelse(tp, bp);
4396 }
4397 /*
4398 * Here with bp and block set to the leftmost leaf node in the tree.
4399 */
4400 room = ifp->if_bytes / (uint)sizeof(*trp);
4401 trp = ifp->if_u1.if_extents;
4402 i = 0;
4403 /*
4404 * Loop over all leaf nodes. Copy information to the extent list.
4405 */
4406 for (;;) {
4407 xfs_bmbt_rec_t *frp, *temp;
4408 xfs_fsblock_t nextbno;
4409 xfs_extnum_t num_recs;
4410
4411
4412 num_recs = INT_GET(block->bb_numrecs, ARCH_CONVERT);
4413 if (unlikely(i + num_recs > room)) {
4414 ASSERT(i + num_recs <= room);
4415 xfs_fs_cmn_err(CE_WARN, ip->i_mount,
4416 "corrupt dinode %Lu, (btree extents). Unmount and run xfs_repair.",
4417 (unsigned long long) ip->i_ino);
4418 XFS_ERROR_REPORT("xfs_bmap_read_extents(1)",
4419 XFS_ERRLEVEL_LOW,
4420 ip->i_mount);
4421 goto error0;
4422 }
4423 XFS_WANT_CORRUPTED_GOTO(
4424 XFS_BMAP_SANITY_CHECK(mp, block, 0),
4425 error0);
4426 /*
4427 * Read-ahead the next leaf block, if any.
4428 */
4429 nextbno = INT_GET(block->bb_rightsib, ARCH_CONVERT);
4430 if (nextbno != NULLFSBLOCK)
4431 xfs_btree_reada_bufl(mp, nextbno, 1);
4432 /*
4433 * Copy records into the extent list.
4434 */
4435 frp = XFS_BTREE_REC_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt,
4436 block, 1, mp->m_bmap_dmxr[0]);
4437 temp = trp;
4438 for (j = 0; j < num_recs; j++, frp++, trp++) {
4439 trp->l0 = INT_GET(frp->l0, ARCH_CONVERT);
4440 trp->l1 = INT_GET(frp->l1, ARCH_CONVERT);
4441 }
4442 if (exntf == XFS_EXTFMT_NOSTATE) {
4443 /*
4444 * Check all attribute bmap btree records and
4445 * any "older" data bmap btree records for a
4446 * set bit in the "extent flag" position.
4447 */
4448 if (unlikely(xfs_check_nostate_extents(temp, num_recs))) {
4449 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4450 XFS_ERRLEVEL_LOW,
4451 ip->i_mount);
4452 goto error0;
4453 }
4454 }
4455 i += num_recs;
4456 xfs_trans_brelse(tp, bp);
4457 bno = nextbno;
4458 /*
4459 * If we've reached the end, stop.
4460 */
4461 if (bno == NULLFSBLOCK)
4462 break;
4463 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4464 XFS_BMAP_BTREE_REF)))
4465 return error;
4466 block = XFS_BUF_TO_BMBT_BLOCK(bp);
4467 }
4468 ASSERT(i == ifp->if_bytes / (uint)sizeof(*trp));
4469 ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4470 xfs_bmap_trace_exlist(fname, ip, i, whichfork);
4471 return 0;
4472error0:
4473 xfs_trans_brelse(tp, bp);
4474 return XFS_ERROR(EFSCORRUPTED);
4475}
4476
4477#ifdef XFS_BMAP_TRACE
4478/*
4479 * Add bmap trace insert entries for all the contents of the extent list.
4480 */
4481void
4482xfs_bmap_trace_exlist(
4483 char *fname, /* function name */
4484 xfs_inode_t *ip, /* incore inode pointer */
4485 xfs_extnum_t cnt, /* count of entries in the list */
4486 int whichfork) /* data or attr fork */
4487{
4488 xfs_bmbt_rec_t *base; /* base of extent list */
4489 xfs_bmbt_rec_t *ep; /* current entry in extent list */
4490 xfs_extnum_t idx; /* extent list entry number */
4491 xfs_ifork_t *ifp; /* inode fork pointer */
4492 xfs_bmbt_irec_t s; /* extent list record */
4493
4494 ifp = XFS_IFORK_PTR(ip, whichfork);
4495 ASSERT(cnt == ifp->if_bytes / (uint)sizeof(*base));
4496 base = ifp->if_u1.if_extents;
4497 for (idx = 0, ep = base; idx < cnt; idx++, ep++) {
4498 xfs_bmbt_get_all(ep, &s);
4499 xfs_bmap_trace_insert(fname, "exlist", ip, idx, 1, &s, NULL,
4500 whichfork);
4501 }
4502}
4503#endif
4504
4505#ifdef DEBUG
4506/*
4507 * Validate that the bmbt_irecs being returned from bmapi are valid
4508 * given the callers original parameters. Specifically check the
4509 * ranges of the returned irecs to ensure that they only extent beyond
4510 * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4511 */
4512STATIC void
4513xfs_bmap_validate_ret(
4514 xfs_fileoff_t bno,
4515 xfs_filblks_t len,
4516 int flags,
4517 xfs_bmbt_irec_t *mval,
4518 int nmap,
4519 int ret_nmap)
4520{
4521 int i; /* index to map values */
4522
4523 ASSERT(ret_nmap <= nmap);
4524
4525 for (i = 0; i < ret_nmap; i++) {
4526 ASSERT(mval[i].br_blockcount > 0);
4527 if (!(flags & XFS_BMAPI_ENTIRE)) {
4528 ASSERT(mval[i].br_startoff >= bno);
4529 ASSERT(mval[i].br_blockcount <= len);
4530 ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4531 bno + len);
4532 } else {
4533 ASSERT(mval[i].br_startoff < bno + len);
4534 ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4535 bno);
4536 }
4537 ASSERT(i == 0 ||
4538 mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4539 mval[i].br_startoff);
4540 if ((flags & XFS_BMAPI_WRITE) && !(flags & XFS_BMAPI_DELAY))
4541 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4542 mval[i].br_startblock != HOLESTARTBLOCK);
4543 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4544 mval[i].br_state == XFS_EXT_UNWRITTEN);
4545 }
4546}
4547#endif /* DEBUG */
4548
4549
4550/*
4551 * Map file blocks to filesystem blocks.
4552 * File range is given by the bno/len pair.
4553 * Adds blocks to file if a write ("flags & XFS_BMAPI_WRITE" set)
4554 * into a hole or past eof.
4555 * Only allocates blocks from a single allocation group,
4556 * to avoid locking problems.
4557 * The returned value in "firstblock" from the first call in a transaction
4558 * must be remembered and presented to subsequent calls in "firstblock".
4559 * An upper bound for the number of blocks to be allocated is supplied to
4560 * the first call in "total"; if no allocation group has that many free
4561 * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4562 */
4563int /* error */
4564xfs_bmapi(
4565 xfs_trans_t *tp, /* transaction pointer */
4566 xfs_inode_t *ip, /* incore inode */
4567 xfs_fileoff_t bno, /* starting file offs. mapped */
4568 xfs_filblks_t len, /* length to map in file */
4569 int flags, /* XFS_BMAPI_... */
4570 xfs_fsblock_t *firstblock, /* first allocated block
4571 controls a.g. for allocs */
4572 xfs_extlen_t total, /* total blocks needed */
4573 xfs_bmbt_irec_t *mval, /* output: map values */
4574 int *nmap, /* i/o: mval size/count */
4575 xfs_bmap_free_t *flist) /* i/o: list extents to free */
4576{
4577 xfs_fsblock_t abno; /* allocated block number */
4578 xfs_extlen_t alen; /* allocated extent length */
4579 xfs_fileoff_t aoff; /* allocated file offset */
4580 xfs_bmalloca_t bma; /* args for xfs_bmap_alloc */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004581 xfs_btree_cur_t *cur; /* bmap btree cursor */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004582 xfs_fileoff_t end; /* end of mapped file region */
4583 int eof; /* we've hit the end of extent list */
Nathan Scott06d10dd2005-06-21 15:48:47 +10004584 char contig; /* allocation must be one extent */
4585 char delay; /* this request is for delayed alloc */
4586 char exact; /* don't do all of wasdelayed extent */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004587 xfs_bmbt_rec_t *ep; /* extent list entry pointer */
4588 int error; /* error return */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004589 xfs_bmbt_irec_t got; /* current extent list record */
4590 xfs_ifork_t *ifp; /* inode fork pointer */
4591 xfs_extlen_t indlen; /* indirect blocks length */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004592 xfs_extnum_t lastx; /* last useful extent number */
4593 int logflags; /* flags for transaction logging */
4594 xfs_extlen_t minleft; /* min blocks left after allocation */
4595 xfs_extlen_t minlen; /* min allocation size */
4596 xfs_mount_t *mp; /* xfs mount structure */
4597 int n; /* current extent index */
4598 int nallocs; /* number of extents alloc\'d */
4599 xfs_extnum_t nextents; /* number of extents in file */
4600 xfs_fileoff_t obno; /* old block number (offset) */
4601 xfs_bmbt_irec_t prev; /* previous extent list record */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004602 int tmp_logflags; /* temp flags holder */
Nathan Scott06d10dd2005-06-21 15:48:47 +10004603 int whichfork; /* data or attr fork */
4604 char inhole; /* current location is hole in file */
4605 char stateless; /* ignore state flag set */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004606 char trim; /* output trimmed to match range */
4607 char userdata; /* allocating non-metadata */
4608 char wasdelay; /* old extent was delayed */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004609 char wr; /* this is a write request */
Nathan Scott06d10dd2005-06-21 15:48:47 +10004610 char rt; /* this is a realtime file */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004611 char rsvd; /* OK to allocate reserved blocks */
4612#ifdef DEBUG
4613 xfs_fileoff_t orig_bno; /* original block number value */
4614 int orig_flags; /* original flags arg value */
4615 xfs_filblks_t orig_len; /* original value of len arg */
4616 xfs_bmbt_irec_t *orig_mval; /* original value of mval */
4617 int orig_nmap; /* original value of *nmap */
4618
4619 orig_bno = bno;
4620 orig_len = len;
4621 orig_flags = flags;
4622 orig_mval = mval;
4623 orig_nmap = *nmap;
4624#endif
4625 ASSERT(*nmap >= 1);
4626 ASSERT(*nmap <= XFS_BMAP_MAX_NMAP || !(flags & XFS_BMAPI_WRITE));
4627 whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4628 XFS_ATTR_FORK : XFS_DATA_FORK;
4629 mp = ip->i_mount;
4630 if (unlikely(XFS_TEST_ERROR(
4631 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4632 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4633 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4634 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4635 XFS_ERROR_REPORT("xfs_bmapi", XFS_ERRLEVEL_LOW, mp);
4636 return XFS_ERROR(EFSCORRUPTED);
4637 }
4638 if (XFS_FORCED_SHUTDOWN(mp))
4639 return XFS_ERROR(EIO);
Nathan Scott06d10dd2005-06-21 15:48:47 +10004640 rt = XFS_IS_REALTIME_INODE(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004641 ifp = XFS_IFORK_PTR(ip, whichfork);
4642 ASSERT(ifp->if_ext_max ==
4643 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4644 if ((wr = (flags & XFS_BMAPI_WRITE)) != 0)
4645 XFS_STATS_INC(xs_blk_mapw);
4646 else
4647 XFS_STATS_INC(xs_blk_mapr);
4648 delay = (flags & XFS_BMAPI_DELAY) != 0;
4649 trim = (flags & XFS_BMAPI_ENTIRE) == 0;
4650 userdata = (flags & XFS_BMAPI_METADATA) == 0;
4651 exact = (flags & XFS_BMAPI_EXACT) != 0;
4652 rsvd = (flags & XFS_BMAPI_RSVBLOCKS) != 0;
4653 contig = (flags & XFS_BMAPI_CONTIG) != 0;
4654 /*
4655 * stateless is used to combine extents which
4656 * differ only due to the state of the extents.
4657 * This technique is used from xfs_getbmap()
4658 * when the caller does not wish to see the
4659 * separation (which is the default).
4660 *
4661 * This technique is also used when writing a
4662 * buffer which has been partially written,
4663 * (usually by being flushed during a chunkread),
4664 * to ensure one write takes place. This also
4665 * prevents a change in the xfs inode extents at
4666 * this time, intentionally. This change occurs
4667 * on completion of the write operation, in
4668 * xfs_strat_comp(), where the xfs_bmapi() call
4669 * is transactioned, and the extents combined.
4670 */
4671 stateless = (flags & XFS_BMAPI_IGSTATE) != 0;
4672 if (stateless && wr) /* if writing unwritten space, no */
4673 wr = 0; /* allocations are allowed */
4674 ASSERT(wr || !delay);
4675 logflags = 0;
4676 nallocs = 0;
4677 cur = NULL;
4678 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4679 ASSERT(wr && tp);
4680 if ((error = xfs_bmap_local_to_extents(tp, ip,
4681 firstblock, total, &logflags, whichfork)))
4682 goto error0;
4683 }
4684 if (wr && *firstblock == NULLFSBLOCK) {
4685 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4686 minleft = INT_GET(ifp->if_broot->bb_level, ARCH_CONVERT) + 1;
4687 else
4688 minleft = 1;
4689 } else
4690 minleft = 0;
4691 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4692 (error = xfs_iread_extents(tp, ip, whichfork)))
4693 goto error0;
4694 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4695 &prev);
4696 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4697 n = 0;
4698 end = bno + len;
4699 obno = bno;
4700 bma.ip = NULL;
4701 while (bno < end && n < *nmap) {
4702 /*
4703 * Reading past eof, act as though there's a hole
4704 * up to end.
4705 */
4706 if (eof && !wr)
4707 got.br_startoff = end;
4708 inhole = eof || got.br_startoff > bno;
4709 wasdelay = wr && !inhole && !delay &&
4710 ISNULLSTARTBLOCK(got.br_startblock);
4711 /*
4712 * First, deal with the hole before the allocated space
4713 * that we found, if any.
4714 */
4715 if (wr && (inhole || wasdelay)) {
4716 /*
4717 * For the wasdelay case, we could also just
4718 * allocate the stuff asked for in this bmap call
4719 * but that wouldn't be as good.
4720 */
4721 if (wasdelay && !exact) {
4722 alen = (xfs_extlen_t)got.br_blockcount;
4723 aoff = got.br_startoff;
4724 if (lastx != NULLEXTNUM && lastx) {
4725 ep = &ifp->if_u1.if_extents[lastx - 1];
4726 xfs_bmbt_get_all(ep, &prev);
4727 }
4728 } else if (wasdelay) {
4729 alen = (xfs_extlen_t)
4730 XFS_FILBLKS_MIN(len,
4731 (got.br_startoff +
4732 got.br_blockcount) - bno);
4733 aoff = bno;
4734 } else {
4735 alen = (xfs_extlen_t)
4736 XFS_FILBLKS_MIN(len, MAXEXTLEN);
4737 if (!eof)
4738 alen = (xfs_extlen_t)
4739 XFS_FILBLKS_MIN(alen,
4740 got.br_startoff - bno);
4741 aoff = bno;
4742 }
4743 minlen = contig ? alen : 1;
4744 if (delay) {
Nathan Scott06d10dd2005-06-21 15:48:47 +10004745 xfs_extlen_t extsz = 0;
4746
4747 /* Figure out the extent size, adjust alen */
4748 if (rt) {
4749 if (!(extsz = ip->i_d.di_extsize))
4750 extsz = mp->m_sb.sb_rextsize;
4751 alen = roundup(alen, extsz);
4752 extsz = alen / mp->m_sb.sb_rextsize;
4753 }
4754
Linus Torvalds1da177e2005-04-16 15:20:36 -07004755 /*
4756 * Make a transaction-less quota reservation for
4757 * delayed allocation blocks. This number gets
4758 * adjusted later.
4759 * We return EDQUOT if we haven't allocated
4760 * blks already inside this loop;
4761 */
Nathan Scott06d10dd2005-06-21 15:48:47 +10004762 if (XFS_TRANS_RESERVE_QUOTA_NBLKS(
4763 mp, NULL, ip, (long)alen, 0,
4764 rt ? XFS_QMOPT_RES_RTBLKS :
4765 XFS_QMOPT_RES_REGBLKS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004766 if (n == 0) {
4767 *nmap = 0;
4768 ASSERT(cur == NULL);
4769 return XFS_ERROR(EDQUOT);
4770 }
4771 break;
4772 }
4773
4774 /*
4775 * Split changing sb for alen and indlen since
4776 * they could be coming from different places.
4777 */
Nathan Scott06d10dd2005-06-21 15:48:47 +10004778 indlen = (xfs_extlen_t)
4779 xfs_bmap_worst_indlen(ip, alen);
4780 ASSERT(indlen > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004781
Nathan Scott06d10dd2005-06-21 15:48:47 +10004782 if (rt)
4783 error = xfs_mod_incore_sb(mp,
4784 XFS_SBS_FREXTENTS,
4785 -(extsz), rsvd);
4786 else
4787 error = xfs_mod_incore_sb(mp,
4788 XFS_SBS_FDBLOCKS,
4789 -(alen), rsvd);
David Chinner3bdbfb12005-09-02 16:40:47 +10004790 if (!error) {
Nathan Scott06d10dd2005-06-21 15:48:47 +10004791 error = xfs_mod_incore_sb(mp,
4792 XFS_SBS_FDBLOCKS,
4793 -(indlen), rsvd);
David Chinner3bdbfb12005-09-02 16:40:47 +10004794 if (error && rt) {
4795 xfs_mod_incore_sb(ip->i_mount,
4796 XFS_SBS_FREXTENTS,
4797 extsz, rsvd);
4798 } else if (error) {
4799 xfs_mod_incore_sb(ip->i_mount,
4800 XFS_SBS_FDBLOCKS,
4801 alen, rsvd);
4802 }
4803 }
Nathan Scott06d10dd2005-06-21 15:48:47 +10004804
4805 if (error) {
4806 if (XFS_IS_QUOTA_ON(ip->i_mount))
4807 /* unreserve the blocks now */
4808 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(
4809 mp, NULL, ip,
4810 (long)alen, 0, rt ?
4811 XFS_QMOPT_RES_RTBLKS :
4812 XFS_QMOPT_RES_REGBLKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004813 break;
4814 }
Nathan Scott06d10dd2005-06-21 15:48:47 +10004815
Linus Torvalds1da177e2005-04-16 15:20:36 -07004816 ip->i_delayed_blks += alen;
4817 abno = NULLSTARTBLOCK(indlen);
4818 } else {
4819 /*
4820 * If first time, allocate and fill in
4821 * once-only bma fields.
4822 */
4823 if (bma.ip == NULL) {
4824 bma.tp = tp;
4825 bma.ip = ip;
4826 bma.prevp = &prev;
4827 bma.gotp = &got;
4828 bma.total = total;
4829 bma.userdata = 0;
4830 }
4831 /* Indicate if this is the first user data
4832 * in the file, or just any user data.
4833 */
4834 if (userdata) {
4835 bma.userdata = (aoff == 0) ?
4836 XFS_ALLOC_INITIAL_USER_DATA :
4837 XFS_ALLOC_USERDATA;
4838 }
4839 /*
4840 * Fill in changeable bma fields.
4841 */
4842 bma.eof = eof;
4843 bma.firstblock = *firstblock;
4844 bma.alen = alen;
4845 bma.off = aoff;
4846 bma.wasdel = wasdelay;
4847 bma.minlen = minlen;
4848 bma.low = flist->xbf_low;
4849 bma.minleft = minleft;
4850 /*
4851 * Only want to do the alignment at the
4852 * eof if it is userdata and allocation length
4853 * is larger than a stripe unit.
4854 */
4855 if (mp->m_dalign && alen >= mp->m_dalign &&
4856 userdata && whichfork == XFS_DATA_FORK) {
4857 if ((error = xfs_bmap_isaeof(ip, aoff,
4858 whichfork, &bma.aeof)))
4859 goto error0;
4860 } else
4861 bma.aeof = 0;
4862 /*
4863 * Call allocator.
4864 */
4865 if ((error = xfs_bmap_alloc(&bma)))
4866 goto error0;
4867 /*
4868 * Copy out result fields.
4869 */
4870 abno = bma.rval;
4871 if ((flist->xbf_low = bma.low))
4872 minleft = 0;
4873 alen = bma.alen;
4874 aoff = bma.off;
4875 ASSERT(*firstblock == NULLFSBLOCK ||
4876 XFS_FSB_TO_AGNO(mp, *firstblock) ==
4877 XFS_FSB_TO_AGNO(mp, bma.firstblock) ||
4878 (flist->xbf_low &&
4879 XFS_FSB_TO_AGNO(mp, *firstblock) <
4880 XFS_FSB_TO_AGNO(mp, bma.firstblock)));
4881 *firstblock = bma.firstblock;
4882 if (cur)
4883 cur->bc_private.b.firstblock =
4884 *firstblock;
4885 if (abno == NULLFSBLOCK)
4886 break;
4887 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
4888 cur = xfs_btree_init_cursor(mp,
4889 tp, NULL, 0, XFS_BTNUM_BMAP,
4890 ip, whichfork);
4891 cur->bc_private.b.firstblock =
4892 *firstblock;
4893 cur->bc_private.b.flist = flist;
4894 }
4895 /*
4896 * Bump the number of extents we've allocated
4897 * in this call.
4898 */
4899 nallocs++;
4900 }
4901 if (cur)
4902 cur->bc_private.b.flags =
4903 wasdelay ? XFS_BTCUR_BPRV_WASDEL : 0;
4904 got.br_startoff = aoff;
4905 got.br_startblock = abno;
4906 got.br_blockcount = alen;
4907 got.br_state = XFS_EXT_NORM; /* assume normal */
4908 /*
4909 * Determine state of extent, and the filesystem.
4910 * A wasdelay extent has been initialized, so
4911 * shouldn't be flagged as unwritten.
4912 */
4913 if (wr && XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4914 if (!wasdelay && (flags & XFS_BMAPI_PREALLOC))
4915 got.br_state = XFS_EXT_UNWRITTEN;
4916 }
4917 error = xfs_bmap_add_extent(ip, lastx, &cur, &got,
4918 firstblock, flist, &tmp_logflags, whichfork,
4919 rsvd);
4920 logflags |= tmp_logflags;
4921 if (error)
4922 goto error0;
4923 lastx = ifp->if_lastex;
4924 ep = &ifp->if_u1.if_extents[lastx];
4925 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4926 xfs_bmbt_get_all(ep, &got);
4927 ASSERT(got.br_startoff <= aoff);
4928 ASSERT(got.br_startoff + got.br_blockcount >=
4929 aoff + alen);
4930#ifdef DEBUG
4931 if (delay) {
4932 ASSERT(ISNULLSTARTBLOCK(got.br_startblock));
4933 ASSERT(STARTBLOCKVAL(got.br_startblock) > 0);
4934 }
4935 ASSERT(got.br_state == XFS_EXT_NORM ||
4936 got.br_state == XFS_EXT_UNWRITTEN);
4937#endif
4938 /*
4939 * Fall down into the found allocated space case.
4940 */
4941 } else if (inhole) {
4942 /*
4943 * Reading in a hole.
4944 */
4945 mval->br_startoff = bno;
4946 mval->br_startblock = HOLESTARTBLOCK;
4947 mval->br_blockcount =
4948 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4949 mval->br_state = XFS_EXT_NORM;
4950 bno += mval->br_blockcount;
4951 len -= mval->br_blockcount;
4952 mval++;
4953 n++;
4954 continue;
4955 }
4956 /*
4957 * Then deal with the allocated space we found.
4958 */
4959 ASSERT(ep != NULL);
4960 if (trim && (got.br_startoff + got.br_blockcount > obno)) {
4961 if (obno > bno)
4962 bno = obno;
4963 ASSERT((bno >= obno) || (n == 0));
4964 ASSERT(bno < end);
4965 mval->br_startoff = bno;
4966 if (ISNULLSTARTBLOCK(got.br_startblock)) {
4967 ASSERT(!wr || delay);
4968 mval->br_startblock = DELAYSTARTBLOCK;
4969 } else
4970 mval->br_startblock =
4971 got.br_startblock +
4972 (bno - got.br_startoff);
4973 /*
4974 * Return the minimum of what we got and what we
4975 * asked for for the length. We can use the len
4976 * variable here because it is modified below
4977 * and we could have been there before coming
4978 * here if the first part of the allocation
4979 * didn't overlap what was asked for.
4980 */
4981 mval->br_blockcount =
4982 XFS_FILBLKS_MIN(end - bno, got.br_blockcount -
4983 (bno - got.br_startoff));
4984 mval->br_state = got.br_state;
4985 ASSERT(mval->br_blockcount <= len);
4986 } else {
4987 *mval = got;
4988 if (ISNULLSTARTBLOCK(mval->br_startblock)) {
4989 ASSERT(!wr || delay);
4990 mval->br_startblock = DELAYSTARTBLOCK;
4991 }
4992 }
4993
4994 /*
4995 * Check if writing previously allocated but
4996 * unwritten extents.
4997 */
4998 if (wr && mval->br_state == XFS_EXT_UNWRITTEN &&
4999 ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_DELAY)) == 0)) {
5000 /*
5001 * Modify (by adding) the state flag, if writing.
5002 */
5003 ASSERT(mval->br_blockcount <= len);
5004 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
5005 cur = xfs_btree_init_cursor(mp,
5006 tp, NULL, 0, XFS_BTNUM_BMAP,
5007 ip, whichfork);
5008 cur->bc_private.b.firstblock =
5009 *firstblock;
5010 cur->bc_private.b.flist = flist;
5011 }
5012 mval->br_state = XFS_EXT_NORM;
5013 error = xfs_bmap_add_extent(ip, lastx, &cur, mval,
5014 firstblock, flist, &tmp_logflags, whichfork,
5015 rsvd);
5016 logflags |= tmp_logflags;
5017 if (error)
5018 goto error0;
5019 lastx = ifp->if_lastex;
5020 ep = &ifp->if_u1.if_extents[lastx];
5021 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5022 xfs_bmbt_get_all(ep, &got);
5023 /*
5024 * We may have combined previously unwritten
5025 * space with written space, so generate
5026 * another request.
5027 */
5028 if (mval->br_blockcount < len)
5029 continue;
5030 }
5031
5032 ASSERT(!trim ||
5033 ((mval->br_startoff + mval->br_blockcount) <= end));
5034 ASSERT(!trim || (mval->br_blockcount <= len) ||
5035 (mval->br_startoff < obno));
5036 bno = mval->br_startoff + mval->br_blockcount;
5037 len = end - bno;
5038 if (n > 0 && mval->br_startoff == mval[-1].br_startoff) {
5039 ASSERT(mval->br_startblock == mval[-1].br_startblock);
5040 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
5041 ASSERT(mval->br_state == mval[-1].br_state);
5042 mval[-1].br_blockcount = mval->br_blockcount;
5043 mval[-1].br_state = mval->br_state;
5044 } else if (n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
5045 mval[-1].br_startblock != DELAYSTARTBLOCK &&
5046 mval[-1].br_startblock != HOLESTARTBLOCK &&
5047 mval->br_startblock ==
5048 mval[-1].br_startblock + mval[-1].br_blockcount &&
5049 (stateless || mval[-1].br_state == mval->br_state)) {
5050 ASSERT(mval->br_startoff ==
5051 mval[-1].br_startoff + mval[-1].br_blockcount);
5052 mval[-1].br_blockcount += mval->br_blockcount;
5053 } else if (n > 0 &&
5054 mval->br_startblock == DELAYSTARTBLOCK &&
5055 mval[-1].br_startblock == DELAYSTARTBLOCK &&
5056 mval->br_startoff ==
5057 mval[-1].br_startoff + mval[-1].br_blockcount) {
5058 mval[-1].br_blockcount += mval->br_blockcount;
5059 mval[-1].br_state = mval->br_state;
5060 } else if (!((n == 0) &&
5061 ((mval->br_startoff + mval->br_blockcount) <=
5062 obno))) {
5063 mval++;
5064 n++;
5065 }
5066 /*
5067 * If we're done, stop now. Stop when we've allocated
5068 * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise
5069 * the transaction may get too big.
5070 */
5071 if (bno >= end || n >= *nmap || nallocs >= *nmap)
5072 break;
5073 /*
5074 * Else go on to the next record.
5075 */
5076 ep++;
5077 lastx++;
5078 if (lastx >= nextents) {
5079 eof = 1;
5080 prev = got;
5081 } else
5082 xfs_bmbt_get_all(ep, &got);
5083 }
5084 ifp->if_lastex = lastx;
5085 *nmap = n;
5086 /*
5087 * Transform from btree to extents, give it cur.
5088 */
5089 if (tp && XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5090 XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5091 ASSERT(wr && cur);
5092 error = xfs_bmap_btree_to_extents(tp, ip, cur,
5093 &tmp_logflags, whichfork);
5094 logflags |= tmp_logflags;
5095 if (error)
5096 goto error0;
5097 }
5098 ASSERT(ifp->if_ext_max ==
5099 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5100 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
5101 XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max);
5102 error = 0;
5103
5104error0:
5105 /*
5106 * Log everything. Do this after conversion, there's no point in
5107 * logging the extent list if we've converted to btree format.
5108 */
5109 if ((logflags & XFS_ILOG_FEXT(whichfork)) &&
5110 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5111 logflags &= ~XFS_ILOG_FEXT(whichfork);
5112 else if ((logflags & XFS_ILOG_FBROOT(whichfork)) &&
5113 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5114 logflags &= ~XFS_ILOG_FBROOT(whichfork);
5115 /*
5116 * Log whatever the flags say, even if error. Otherwise we might miss
5117 * detecting a case where the data is changed, there's an error,
5118 * and it's not logged so we don't shutdown when we should.
5119 */
5120 if (logflags) {
5121 ASSERT(tp && wr);
5122 xfs_trans_log_inode(tp, ip, logflags);
5123 }
5124 if (cur) {
5125 if (!error) {
5126 ASSERT(*firstblock == NULLFSBLOCK ||
5127 XFS_FSB_TO_AGNO(mp, *firstblock) ==
5128 XFS_FSB_TO_AGNO(mp,
5129 cur->bc_private.b.firstblock) ||
5130 (flist->xbf_low &&
5131 XFS_FSB_TO_AGNO(mp, *firstblock) <
5132 XFS_FSB_TO_AGNO(mp,
5133 cur->bc_private.b.firstblock)));
5134 *firstblock = cur->bc_private.b.firstblock;
5135 }
5136 xfs_btree_del_cursor(cur,
5137 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5138 }
5139 if (!error)
5140 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
5141 orig_nmap, *nmap);
5142 return error;
5143}
5144
5145/*
5146 * Map file blocks to filesystem blocks, simple version.
5147 * One block (extent) only, read-only.
5148 * For flags, only the XFS_BMAPI_ATTRFORK flag is examined.
5149 * For the other flag values, the effect is as if XFS_BMAPI_METADATA
5150 * was set and all the others were clear.
5151 */
5152int /* error */
5153xfs_bmapi_single(
5154 xfs_trans_t *tp, /* transaction pointer */
5155 xfs_inode_t *ip, /* incore inode */
5156 int whichfork, /* data or attr fork */
5157 xfs_fsblock_t *fsb, /* output: mapped block */
5158 xfs_fileoff_t bno) /* starting file offs. mapped */
5159{
5160 int eof; /* we've hit the end of extent list */
5161 int error; /* error return */
5162 xfs_bmbt_irec_t got; /* current extent list record */
5163 xfs_ifork_t *ifp; /* inode fork pointer */
5164 xfs_extnum_t lastx; /* last useful extent number */
5165 xfs_bmbt_irec_t prev; /* previous extent list record */
5166
5167 ifp = XFS_IFORK_PTR(ip, whichfork);
5168 if (unlikely(
5169 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
5170 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)) {
5171 XFS_ERROR_REPORT("xfs_bmapi_single", XFS_ERRLEVEL_LOW,
5172 ip->i_mount);
5173 return XFS_ERROR(EFSCORRUPTED);
5174 }
5175 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
5176 return XFS_ERROR(EIO);
5177 XFS_STATS_INC(xs_blk_mapr);
5178 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5179 (error = xfs_iread_extents(tp, ip, whichfork)))
5180 return error;
5181 (void)xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5182 &prev);
5183 /*
5184 * Reading past eof, act as though there's a hole
5185 * up to end.
5186 */
5187 if (eof || got.br_startoff > bno) {
5188 *fsb = NULLFSBLOCK;
5189 return 0;
5190 }
5191 ASSERT(!ISNULLSTARTBLOCK(got.br_startblock));
5192 ASSERT(bno < got.br_startoff + got.br_blockcount);
5193 *fsb = got.br_startblock + (bno - got.br_startoff);
5194 ifp->if_lastex = lastx;
5195 return 0;
5196}
5197
5198/*
5199 * Unmap (remove) blocks from a file.
5200 * If nexts is nonzero then the number of extents to remove is limited to
5201 * that value. If not all extents in the block range can be removed then
5202 * *done is set.
5203 */
5204int /* error */
5205xfs_bunmapi(
5206 xfs_trans_t *tp, /* transaction pointer */
5207 struct xfs_inode *ip, /* incore inode */
5208 xfs_fileoff_t bno, /* starting offset to unmap */
5209 xfs_filblks_t len, /* length to unmap in file */
5210 int flags, /* misc flags */
5211 xfs_extnum_t nexts, /* number of extents max */
5212 xfs_fsblock_t *firstblock, /* first allocated block
5213 controls a.g. for allocs */
5214 xfs_bmap_free_t *flist, /* i/o: list extents to free */
5215 int *done) /* set if not done yet */
5216{
5217 xfs_btree_cur_t *cur; /* bmap btree cursor */
5218 xfs_bmbt_irec_t del; /* extent being deleted */
5219 int eof; /* is deleting at eof */
5220 xfs_bmbt_rec_t *ep; /* extent list entry pointer */
5221 int error; /* error return value */
5222 xfs_extnum_t extno; /* extent number in list */
5223 xfs_bmbt_irec_t got; /* current extent list entry */
5224 xfs_ifork_t *ifp; /* inode fork pointer */
5225 int isrt; /* freeing in rt area */
5226 xfs_extnum_t lastx; /* last extent index used */
5227 int logflags; /* transaction logging flags */
5228 xfs_extlen_t mod; /* rt extent offset */
5229 xfs_mount_t *mp; /* mount structure */
5230 xfs_extnum_t nextents; /* size of extent list */
5231 xfs_bmbt_irec_t prev; /* previous extent list entry */
5232 xfs_fileoff_t start; /* first file offset deleted */
5233 int tmp_logflags; /* partial logging flags */
5234 int wasdel; /* was a delayed alloc extent */
5235 int whichfork; /* data or attribute fork */
5236 int rsvd; /* OK to allocate reserved blocks */
5237 xfs_fsblock_t sum;
5238
5239 xfs_bunmap_trace(ip, bno, len, flags, (inst_t *)__return_address);
5240 whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5241 XFS_ATTR_FORK : XFS_DATA_FORK;
5242 ifp = XFS_IFORK_PTR(ip, whichfork);
5243 if (unlikely(
5244 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5245 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5246 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5247 ip->i_mount);
5248 return XFS_ERROR(EFSCORRUPTED);
5249 }
5250 mp = ip->i_mount;
5251 if (XFS_FORCED_SHUTDOWN(mp))
5252 return XFS_ERROR(EIO);
5253 rsvd = (flags & XFS_BMAPI_RSVBLOCKS) != 0;
5254 ASSERT(len > 0);
5255 ASSERT(nexts >= 0);
5256 ASSERT(ifp->if_ext_max ==
5257 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5258 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5259 (error = xfs_iread_extents(tp, ip, whichfork)))
5260 return error;
5261 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5262 if (nextents == 0) {
5263 *done = 1;
5264 return 0;
5265 }
5266 XFS_STATS_INC(xs_blk_unmap);
5267 isrt = (whichfork == XFS_DATA_FORK) &&
5268 (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
5269 start = bno;
5270 bno = start + len - 1;
5271 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5272 &prev);
5273 /*
5274 * Check to see if the given block number is past the end of the
5275 * file, back up to the last block if so...
5276 */
5277 if (eof) {
5278 ep = &ifp->if_u1.if_extents[--lastx];
5279 xfs_bmbt_get_all(ep, &got);
5280 bno = got.br_startoff + got.br_blockcount - 1;
5281 }
5282 logflags = 0;
5283 if (ifp->if_flags & XFS_IFBROOT) {
5284 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5285 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
5286 whichfork);
5287 cur->bc_private.b.firstblock = *firstblock;
5288 cur->bc_private.b.flist = flist;
5289 cur->bc_private.b.flags = 0;
5290 } else
5291 cur = NULL;
5292 extno = 0;
5293 while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5294 (nexts == 0 || extno < nexts)) {
5295 /*
5296 * Is the found extent after a hole in which bno lives?
5297 * Just back up to the previous extent, if so.
5298 */
5299 if (got.br_startoff > bno) {
5300 if (--lastx < 0)
5301 break;
5302 ep--;
5303 xfs_bmbt_get_all(ep, &got);
5304 }
5305 /*
5306 * Is the last block of this extent before the range
5307 * we're supposed to delete? If so, we're done.
5308 */
5309 bno = XFS_FILEOFF_MIN(bno,
5310 got.br_startoff + got.br_blockcount - 1);
5311 if (bno < start)
5312 break;
5313 /*
5314 * Then deal with the (possibly delayed) allocated space
5315 * we found.
5316 */
5317 ASSERT(ep != NULL);
5318 del = got;
5319 wasdel = ISNULLSTARTBLOCK(del.br_startblock);
5320 if (got.br_startoff < start) {
5321 del.br_startoff = start;
5322 del.br_blockcount -= start - got.br_startoff;
5323 if (!wasdel)
5324 del.br_startblock += start - got.br_startoff;
5325 }
5326 if (del.br_startoff + del.br_blockcount > bno + 1)
5327 del.br_blockcount = bno + 1 - del.br_startoff;
5328 sum = del.br_startblock + del.br_blockcount;
5329 if (isrt &&
5330 (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5331 /*
5332 * Realtime extent not lined up at the end.
5333 * The extent could have been split into written
5334 * and unwritten pieces, or we could just be
5335 * unmapping part of it. But we can't really
5336 * get rid of part of a realtime extent.
5337 */
5338 if (del.br_state == XFS_EXT_UNWRITTEN ||
5339 !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
5340 /*
5341 * This piece is unwritten, or we're not
5342 * using unwritten extents. Skip over it.
5343 */
5344 ASSERT(bno >= mod);
5345 bno -= mod > del.br_blockcount ?
5346 del.br_blockcount : mod;
5347 if (bno < got.br_startoff) {
5348 if (--lastx >= 0)
5349 xfs_bmbt_get_all(--ep, &got);
5350 }
5351 continue;
5352 }
5353 /*
5354 * It's written, turn it unwritten.
5355 * This is better than zeroing it.
5356 */
5357 ASSERT(del.br_state == XFS_EXT_NORM);
5358 ASSERT(xfs_trans_get_block_res(tp) > 0);
5359 /*
5360 * If this spans a realtime extent boundary,
5361 * chop it back to the start of the one we end at.
5362 */
5363 if (del.br_blockcount > mod) {
5364 del.br_startoff += del.br_blockcount - mod;
5365 del.br_startblock += del.br_blockcount - mod;
5366 del.br_blockcount = mod;
5367 }
5368 del.br_state = XFS_EXT_UNWRITTEN;
5369 error = xfs_bmap_add_extent(ip, lastx, &cur, &del,
5370 firstblock, flist, &logflags, XFS_DATA_FORK, 0);
5371 if (error)
5372 goto error0;
5373 goto nodelete;
5374 }
5375 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5376 /*
5377 * Realtime extent is lined up at the end but not
5378 * at the front. We'll get rid of full extents if
5379 * we can.
5380 */
5381 mod = mp->m_sb.sb_rextsize - mod;
5382 if (del.br_blockcount > mod) {
5383 del.br_blockcount -= mod;
5384 del.br_startoff += mod;
5385 del.br_startblock += mod;
5386 } else if ((del.br_startoff == start &&
5387 (del.br_state == XFS_EXT_UNWRITTEN ||
5388 xfs_trans_get_block_res(tp) == 0)) ||
5389 !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
5390 /*
5391 * Can't make it unwritten. There isn't
5392 * a full extent here so just skip it.
5393 */
5394 ASSERT(bno >= del.br_blockcount);
5395 bno -= del.br_blockcount;
5396 if (bno < got.br_startoff) {
5397 if (--lastx >= 0)
5398 xfs_bmbt_get_all(--ep, &got);
5399 }
5400 continue;
5401 } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5402 /*
5403 * This one is already unwritten.
5404 * It must have a written left neighbor.
5405 * Unwrite the killed part of that one and
5406 * try again.
5407 */
5408 ASSERT(lastx > 0);
5409 xfs_bmbt_get_all(ep - 1, &prev);
5410 ASSERT(prev.br_state == XFS_EXT_NORM);
5411 ASSERT(!ISNULLSTARTBLOCK(prev.br_startblock));
5412 ASSERT(del.br_startblock ==
5413 prev.br_startblock + prev.br_blockcount);
5414 if (prev.br_startoff < start) {
5415 mod = start - prev.br_startoff;
5416 prev.br_blockcount -= mod;
5417 prev.br_startblock += mod;
5418 prev.br_startoff = start;
5419 }
5420 prev.br_state = XFS_EXT_UNWRITTEN;
5421 error = xfs_bmap_add_extent(ip, lastx - 1, &cur,
5422 &prev, firstblock, flist, &logflags,
5423 XFS_DATA_FORK, 0);
5424 if (error)
5425 goto error0;
5426 goto nodelete;
5427 } else {
5428 ASSERT(del.br_state == XFS_EXT_NORM);
5429 del.br_state = XFS_EXT_UNWRITTEN;
5430 error = xfs_bmap_add_extent(ip, lastx, &cur,
5431 &del, firstblock, flist, &logflags,
5432 XFS_DATA_FORK, 0);
5433 if (error)
5434 goto error0;
5435 goto nodelete;
5436 }
5437 }
5438 if (wasdel) {
5439 ASSERT(STARTBLOCKVAL(del.br_startblock) > 0);
Nathan Scott06d10dd2005-06-21 15:48:47 +10005440 /* Update realtim/data freespace, unreserve quota */
5441 if (isrt) {
5442 xfs_filblks_t rtexts;
5443
5444 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5445 do_div(rtexts, mp->m_sb.sb_rextsize);
5446 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5447 (int)rtexts, rsvd);
5448 XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, NULL, ip,
5449 -((long)del.br_blockcount), 0,
5450 XFS_QMOPT_RES_RTBLKS);
5451 } else {
5452 xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS,
5453 (int)del.br_blockcount, rsvd);
5454 XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, NULL, ip,
5455 -((long)del.br_blockcount), 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005456 XFS_QMOPT_RES_REGBLKS);
Nathan Scott06d10dd2005-06-21 15:48:47 +10005457 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005458 ip->i_delayed_blks -= del.br_blockcount;
5459 if (cur)
5460 cur->bc_private.b.flags |=
5461 XFS_BTCUR_BPRV_WASDEL;
5462 } else if (cur)
5463 cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5464 /*
5465 * If it's the case where the directory code is running
5466 * with no block reservation, and the deleted block is in
5467 * the middle of its extent, and the resulting insert
5468 * of an extent would cause transformation to btree format,
5469 * then reject it. The calling code will then swap
5470 * blocks around instead.
5471 * We have to do this now, rather than waiting for the
5472 * conversion to btree format, since the transaction
5473 * will be dirty.
5474 */
5475 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5476 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5477 XFS_IFORK_NEXTENTS(ip, whichfork) >= ifp->if_ext_max &&
5478 del.br_startoff > got.br_startoff &&
5479 del.br_startoff + del.br_blockcount <
5480 got.br_startoff + got.br_blockcount) {
5481 error = XFS_ERROR(ENOSPC);
5482 goto error0;
5483 }
5484 error = xfs_bmap_del_extent(ip, tp, lastx, flist, cur, &del,
5485 &tmp_logflags, whichfork, rsvd);
5486 logflags |= tmp_logflags;
5487 if (error)
5488 goto error0;
5489 bno = del.br_startoff - 1;
5490nodelete:
5491 lastx = ifp->if_lastex;
5492 /*
5493 * If not done go on to the next (previous) record.
5494 * Reset ep in case the extents array was re-alloced.
5495 */
5496 ep = &ifp->if_u1.if_extents[lastx];
5497 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5498 if (lastx >= XFS_IFORK_NEXTENTS(ip, whichfork) ||
5499 xfs_bmbt_get_startoff(ep) > bno) {
5500 lastx--;
5501 ep--;
5502 }
5503 if (lastx >= 0)
5504 xfs_bmbt_get_all(ep, &got);
5505 extno++;
5506 }
5507 }
5508 ifp->if_lastex = lastx;
5509 *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5510 ASSERT(ifp->if_ext_max ==
5511 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5512 /*
5513 * Convert to a btree if necessary.
5514 */
5515 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5516 XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
5517 ASSERT(cur == NULL);
5518 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5519 &cur, 0, &tmp_logflags, whichfork);
5520 logflags |= tmp_logflags;
5521 if (error)
5522 goto error0;
5523 }
5524 /*
5525 * transform from btree to extents, give it cur
5526 */
5527 else if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5528 XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5529 ASSERT(cur != NULL);
5530 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5531 whichfork);
5532 logflags |= tmp_logflags;
5533 if (error)
5534 goto error0;
5535 }
5536 /*
5537 * transform from extents to local?
5538 */
5539 ASSERT(ifp->if_ext_max ==
5540 XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5541 error = 0;
5542error0:
5543 /*
5544 * Log everything. Do this after conversion, there's no point in
5545 * logging the extent list if we've converted to btree format.
5546 */
5547 if ((logflags & XFS_ILOG_FEXT(whichfork)) &&
5548 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5549 logflags &= ~XFS_ILOG_FEXT(whichfork);
5550 else if ((logflags & XFS_ILOG_FBROOT(whichfork)) &&
5551 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5552 logflags &= ~XFS_ILOG_FBROOT(whichfork);
5553 /*
5554 * Log inode even in the error case, if the transaction
5555 * is dirty we'll need to shut down the filesystem.
5556 */
5557 if (logflags)
5558 xfs_trans_log_inode(tp, ip, logflags);
5559 if (cur) {
5560 if (!error) {
5561 *firstblock = cur->bc_private.b.firstblock;
5562 cur->bc_private.b.allocated = 0;
5563 }
5564 xfs_btree_del_cursor(cur,
5565 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5566 }
5567 return error;
5568}
5569
5570/*
5571 * Fcntl interface to xfs_bmapi.
5572 */
5573int /* error code */
5574xfs_getbmap(
5575 bhv_desc_t *bdp, /* XFS behavior descriptor*/
5576 struct getbmap *bmv, /* user bmap structure */
5577 void __user *ap, /* pointer to user's array */
5578 int interface) /* interface flags */
5579{
5580 __int64_t bmvend; /* last block requested */
5581 int error; /* return value */
5582 __int64_t fixlen; /* length for -1 case */
5583 int i; /* extent number */
5584 xfs_inode_t *ip; /* xfs incore inode pointer */
5585 vnode_t *vp; /* corresponding vnode */
5586 int lock; /* lock state */
5587 xfs_bmbt_irec_t *map; /* buffer for user's data */
5588 xfs_mount_t *mp; /* file system mount point */
5589 int nex; /* # of user extents can do */
5590 int nexleft; /* # of user extents left */
5591 int subnex; /* # of bmapi's can do */
5592 int nmap; /* number of map entries */
5593 struct getbmap out; /* output structure */
5594 int whichfork; /* data or attr fork */
5595 int prealloced; /* this is a file with
5596 * preallocated data space */
5597 int sh_unwritten; /* true, if unwritten */
5598 /* extents listed separately */
5599 int bmapi_flags; /* flags for xfs_bmapi */
5600 __int32_t oflags; /* getbmapx bmv_oflags field */
5601
5602 vp = BHV_TO_VNODE(bdp);
5603 ip = XFS_BHVTOI(bdp);
5604 mp = ip->i_mount;
5605
5606 whichfork = interface & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5607 sh_unwritten = (interface & BMV_IF_PREALLOC) != 0;
5608
5609 /* If the BMV_IF_NO_DMAPI_READ interface bit specified, do not
5610 * generate a DMAPI read event. Otherwise, if the DM_EVENT_READ
5611 * bit is set for the file, generate a read event in order
5612 * that the DMAPI application may do its thing before we return
5613 * the extents. Usually this means restoring user file data to
5614 * regions of the file that look like holes.
5615 *
5616 * The "old behavior" (from XFS_IOC_GETBMAP) is to not specify
5617 * BMV_IF_NO_DMAPI_READ so that read events are generated.
5618 * If this were not true, callers of ioctl( XFS_IOC_GETBMAP )
5619 * could misinterpret holes in a DMAPI file as true holes,
5620 * when in fact they may represent offline user data.
5621 */
5622 if ( (interface & BMV_IF_NO_DMAPI_READ) == 0
5623 && DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_READ)
5624 && whichfork == XFS_DATA_FORK) {
5625
5626 error = XFS_SEND_DATA(mp, DM_EVENT_READ, vp, 0, 0, 0, NULL);
5627 if (error)
5628 return XFS_ERROR(error);
5629 }
5630
5631 if (whichfork == XFS_ATTR_FORK) {
5632 if (XFS_IFORK_Q(ip)) {
5633 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5634 ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5635 ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5636 return XFS_ERROR(EINVAL);
5637 } else if (unlikely(
5638 ip->i_d.di_aformat != 0 &&
5639 ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5640 XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5641 ip->i_mount);
5642 return XFS_ERROR(EFSCORRUPTED);
5643 }
5644 } else if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5645 ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5646 ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5647 return XFS_ERROR(EINVAL);
5648 if (whichfork == XFS_DATA_FORK) {
5649 if (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC) {
5650 prealloced = 1;
5651 fixlen = XFS_MAXIOFFSET(mp);
5652 } else {
5653 prealloced = 0;
5654 fixlen = ip->i_d.di_size;
5655 }
5656 } else {
5657 prealloced = 0;
5658 fixlen = 1LL << 32;
5659 }
5660
5661 if (bmv->bmv_length == -1) {
5662 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5663 bmv->bmv_length = MAX( (__int64_t)(fixlen - bmv->bmv_offset),
5664 (__int64_t)0);
5665 } else if (bmv->bmv_length < 0)
5666 return XFS_ERROR(EINVAL);
5667 if (bmv->bmv_length == 0) {
5668 bmv->bmv_entries = 0;
5669 return 0;
5670 }
5671 nex = bmv->bmv_count - 1;
5672 if (nex <= 0)
5673 return XFS_ERROR(EINVAL);
5674 bmvend = bmv->bmv_offset + bmv->bmv_length;
5675
5676 xfs_ilock(ip, XFS_IOLOCK_SHARED);
5677
5678 if (whichfork == XFS_DATA_FORK && ip->i_delayed_blks) {
5679 /* xfs_fsize_t last_byte = xfs_file_last_byte(ip); */
5680 VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1, 0, FI_REMAPF, error);
5681 }
5682
5683 ASSERT(whichfork == XFS_ATTR_FORK || ip->i_delayed_blks == 0);
5684
5685 lock = xfs_ilock_map_shared(ip);
5686
5687 /*
5688 * Don't let nex be bigger than the number of extents
5689 * we can have assuming alternating holes and real extents.
5690 */
5691 if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5692 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5693
5694 bmapi_flags = XFS_BMAPI_AFLAG(whichfork) |
5695 ((sh_unwritten) ? 0 : XFS_BMAPI_IGSTATE);
5696
5697 /*
5698 * Allocate enough space to handle "subnex" maps at a time.
5699 */
5700 subnex = 16;
5701 map = kmem_alloc(subnex * sizeof(*map), KM_SLEEP);
5702
5703 bmv->bmv_entries = 0;
5704
5705 if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0) {
5706 error = 0;
5707 goto unlock_and_return;
5708 }
5709
5710 nexleft = nex;
5711
5712 do {
5713 nmap = (nexleft > subnex) ? subnex : nexleft;
5714 error = xfs_bmapi(NULL, ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5715 XFS_BB_TO_FSB(mp, bmv->bmv_length),
5716 bmapi_flags, NULL, 0, map, &nmap, NULL);
5717 if (error)
5718 goto unlock_and_return;
5719 ASSERT(nmap <= subnex);
5720
5721 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5722 nexleft--;
5723 oflags = (map[i].br_state == XFS_EXT_UNWRITTEN) ?
5724 BMV_OF_PREALLOC : 0;
5725 out.bmv_offset = XFS_FSB_TO_BB(mp, map[i].br_startoff);
5726 out.bmv_length = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5727 ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
5728 if (prealloced &&
5729 map[i].br_startblock == HOLESTARTBLOCK &&
5730 out.bmv_offset + out.bmv_length == bmvend) {
5731 /*
5732 * came to hole at end of file
5733 */
5734 goto unlock_and_return;
5735 } else {
5736 out.bmv_block =
5737 (map[i].br_startblock == HOLESTARTBLOCK) ?
5738 -1 :
5739 XFS_FSB_TO_DB(ip, map[i].br_startblock);
5740
5741 /* return either getbmap/getbmapx structure. */
5742 if (interface & BMV_IF_EXTENDED) {
5743 struct getbmapx outx;
5744
5745 GETBMAP_CONVERT(out,outx);
5746 outx.bmv_oflags = oflags;
5747 outx.bmv_unused1 = outx.bmv_unused2 = 0;
5748 if (copy_to_user(ap, &outx,
5749 sizeof(outx))) {
5750 error = XFS_ERROR(EFAULT);
5751 goto unlock_and_return;
5752 }
5753 } else {
5754 if (copy_to_user(ap, &out,
5755 sizeof(out))) {
5756 error = XFS_ERROR(EFAULT);
5757 goto unlock_and_return;
5758 }
5759 }
5760 bmv->bmv_offset =
5761 out.bmv_offset + out.bmv_length;
5762 bmv->bmv_length = MAX((__int64_t)0,
5763 (__int64_t)(bmvend - bmv->bmv_offset));
5764 bmv->bmv_entries++;
5765 ap = (interface & BMV_IF_EXTENDED) ?
5766 (void __user *)
5767 ((struct getbmapx __user *)ap + 1) :
5768 (void __user *)
5769 ((struct getbmap __user *)ap + 1);
5770 }
5771 }
5772 } while (nmap && nexleft && bmv->bmv_length);
5773
5774unlock_and_return:
5775 xfs_iunlock_map_shared(ip, lock);
5776 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5777
5778 kmem_free(map, subnex * sizeof(*map));
5779
5780 return error;
5781}
5782
5783/*
5784 * Check the last inode extent to determine whether this allocation will result
5785 * in blocks being allocated at the end of the file. When we allocate new data
5786 * blocks at the end of the file which do not start at the previous data block,
5787 * we will try to align the new blocks at stripe unit boundaries.
5788 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10005789STATIC int /* error */
Linus Torvalds1da177e2005-04-16 15:20:36 -07005790xfs_bmap_isaeof(
5791 xfs_inode_t *ip, /* incore inode pointer */
5792 xfs_fileoff_t off, /* file offset in fsblocks */
5793 int whichfork, /* data or attribute fork */
5794 char *aeof) /* return value */
5795{
5796 int error; /* error return value */
5797 xfs_ifork_t *ifp; /* inode fork pointer */
5798 xfs_bmbt_rec_t *lastrec; /* extent list entry pointer */
5799 xfs_extnum_t nextents; /* size of extent list */
5800 xfs_bmbt_irec_t s; /* expanded extent list entry */
5801
5802 ASSERT(whichfork == XFS_DATA_FORK);
5803 ifp = XFS_IFORK_PTR(ip, whichfork);
5804 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5805 (error = xfs_iread_extents(NULL, ip, whichfork)))
5806 return error;
5807 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5808 if (nextents == 0) {
5809 *aeof = 1;
5810 return 0;
5811 }
5812 /*
5813 * Go to the last extent
5814 */
5815 lastrec = &ifp->if_u1.if_extents[nextents - 1];
5816 xfs_bmbt_get_all(lastrec, &s);
5817 /*
5818 * Check we are allocating in the last extent (for delayed allocations)
5819 * or past the last extent for non-delayed allocations.
5820 */
5821 *aeof = (off >= s.br_startoff &&
5822 off < s.br_startoff + s.br_blockcount &&
5823 ISNULLSTARTBLOCK(s.br_startblock)) ||
5824 off >= s.br_startoff + s.br_blockcount;
5825 return 0;
5826}
5827
5828/*
5829 * Check if the endoff is outside the last extent. If so the caller will grow
5830 * the allocation to a stripe unit boundary.
5831 */
5832int /* error */
5833xfs_bmap_eof(
5834 xfs_inode_t *ip, /* incore inode pointer */
5835 xfs_fileoff_t endoff, /* file offset in fsblocks */
5836 int whichfork, /* data or attribute fork */
5837 int *eof) /* result value */
5838{
5839 xfs_fsblock_t blockcount; /* extent block count */
5840 int error; /* error return value */
5841 xfs_ifork_t *ifp; /* inode fork pointer */
5842 xfs_bmbt_rec_t *lastrec; /* extent list entry pointer */
5843 xfs_extnum_t nextents; /* size of extent list */
5844 xfs_fileoff_t startoff; /* extent starting file offset */
5845
5846 ASSERT(whichfork == XFS_DATA_FORK);
5847 ifp = XFS_IFORK_PTR(ip, whichfork);
5848 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5849 (error = xfs_iread_extents(NULL, ip, whichfork)))
5850 return error;
5851 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5852 if (nextents == 0) {
5853 *eof = 1;
5854 return 0;
5855 }
5856 /*
5857 * Go to the last extent
5858 */
5859 lastrec = &ifp->if_u1.if_extents[nextents - 1];
5860 startoff = xfs_bmbt_get_startoff(lastrec);
5861 blockcount = xfs_bmbt_get_blockcount(lastrec);
5862 *eof = endoff >= startoff + blockcount;
5863 return 0;
5864}
5865
5866#ifdef DEBUG
5867/*
5868 * Check that the extents list for the inode ip is in the right order.
5869 */
5870STATIC void
5871xfs_bmap_check_extents(
5872 xfs_inode_t *ip, /* incore inode pointer */
5873 int whichfork) /* data or attr fork */
5874{
5875 xfs_bmbt_rec_t *base; /* base of extents list */
5876 xfs_bmbt_rec_t *ep; /* current extent entry */
5877 xfs_ifork_t *ifp; /* inode fork pointer */
5878 xfs_extnum_t nextents; /* number of extents in list */
5879
5880 ifp = XFS_IFORK_PTR(ip, whichfork);
5881 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
5882 base = ifp->if_u1.if_extents;
5883 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5884 for (ep = base; ep < &base[nextents - 1]; ep++) {
5885 xfs_btree_check_rec(XFS_BTNUM_BMAP, (void *)ep,
5886 (void *)(ep + 1));
5887 }
5888}
5889
5890STATIC
5891xfs_buf_t *
5892xfs_bmap_get_bp(
5893 xfs_btree_cur_t *cur,
5894 xfs_fsblock_t bno)
5895{
5896 int i;
5897 xfs_buf_t *bp;
5898
5899 if (!cur)
5900 return(NULL);
5901
5902 bp = NULL;
5903 for(i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
5904 bp = cur->bc_bufs[i];
5905 if (!bp) break;
5906 if (XFS_BUF_ADDR(bp) == bno)
5907 break; /* Found it */
5908 }
5909 if (i == XFS_BTREE_MAXLEVELS)
5910 bp = NULL;
5911
5912 if (!bp) { /* Chase down all the log items to see if the bp is there */
5913 xfs_log_item_chunk_t *licp;
5914 xfs_trans_t *tp;
5915
5916 tp = cur->bc_tp;
5917 licp = &tp->t_items;
5918 while (!bp && licp != NULL) {
5919 if (XFS_LIC_ARE_ALL_FREE(licp)) {
5920 licp = licp->lic_next;
5921 continue;
5922 }
5923 for (i = 0; i < licp->lic_unused; i++) {
5924 xfs_log_item_desc_t *lidp;
5925 xfs_log_item_t *lip;
5926 xfs_buf_log_item_t *bip;
5927 xfs_buf_t *lbp;
5928
5929 if (XFS_LIC_ISFREE(licp, i)) {
5930 continue;
5931 }
5932
5933 lidp = XFS_LIC_SLOT(licp, i);
5934 lip = lidp->lid_item;
5935 if (lip->li_type != XFS_LI_BUF)
5936 continue;
5937
5938 bip = (xfs_buf_log_item_t *)lip;
5939 lbp = bip->bli_buf;
5940
5941 if (XFS_BUF_ADDR(lbp) == bno) {
5942 bp = lbp;
5943 break; /* Found it */
5944 }
5945 }
5946 licp = licp->lic_next;
5947 }
5948 }
5949 return(bp);
5950}
5951
5952void
5953xfs_check_block(
5954 xfs_bmbt_block_t *block,
5955 xfs_mount_t *mp,
5956 int root,
5957 short sz)
5958{
5959 int i, j, dmxr;
5960 xfs_bmbt_ptr_t *pp, *thispa; /* pointer to block address */
5961 xfs_bmbt_key_t *prevp, *keyp;
5962
5963 ASSERT(INT_GET(block->bb_level, ARCH_CONVERT) > 0);
5964
5965 prevp = NULL;
5966 for( i = 1; i <= INT_GET(block->bb_numrecs, ARCH_CONVERT);i++) {
5967 dmxr = mp->m_bmap_dmxr[0];
5968
5969 if (root) {
5970 keyp = XFS_BMAP_BROOT_KEY_ADDR(block, i, sz);
5971 } else {
5972 keyp = XFS_BTREE_KEY_ADDR(mp->m_sb.sb_blocksize,
5973 xfs_bmbt, block, i, dmxr);
5974 }
5975
5976 if (prevp) {
5977 xfs_btree_check_key(XFS_BTNUM_BMAP, prevp, keyp);
5978 }
5979 prevp = keyp;
5980
5981 /*
5982 * Compare the block numbers to see if there are dups.
5983 */
5984
5985 if (root) {
5986 pp = XFS_BMAP_BROOT_PTR_ADDR(block, i, sz);
5987 } else {
5988 pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize,
5989 xfs_bmbt, block, i, dmxr);
5990 }
5991 for (j = i+1; j <= INT_GET(block->bb_numrecs, ARCH_CONVERT); j++) {
5992 if (root) {
5993 thispa = XFS_BMAP_BROOT_PTR_ADDR(block, j, sz);
5994 } else {
5995 thispa = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize,
5996 xfs_bmbt, block, j, dmxr);
5997 }
5998 if (INT_GET(*thispa, ARCH_CONVERT) ==
5999 INT_GET(*pp, ARCH_CONVERT)) {
6000 cmn_err(CE_WARN, "%s: thispa(%d) == pp(%d) %Ld",
6001 __FUNCTION__, j, i,
6002 INT_GET(*thispa, ARCH_CONVERT));
6003 panic("%s: ptrs are equal in node\n",
6004 __FUNCTION__);
6005 }
6006 }
6007 }
6008}
6009
6010/*
6011 * Check that the extents for the inode ip are in the right order in all
6012 * btree leaves.
6013 */
6014
6015STATIC void
6016xfs_bmap_check_leaf_extents(
6017 xfs_btree_cur_t *cur, /* btree cursor or null */
6018 xfs_inode_t *ip, /* incore inode pointer */
6019 int whichfork) /* data or attr fork */
6020{
6021 xfs_bmbt_block_t *block; /* current btree block */
6022 xfs_fsblock_t bno; /* block # of "block" */
6023 xfs_buf_t *bp; /* buffer for "block" */
6024 int error; /* error return value */
6025 xfs_extnum_t i=0; /* index into the extents list */
6026 xfs_ifork_t *ifp; /* fork structure */
6027 int level; /* btree level, for checking */
6028 xfs_mount_t *mp; /* file system mount structure */
6029 xfs_bmbt_ptr_t *pp; /* pointer to block address */
6030 xfs_bmbt_rec_t *ep, *lastp; /* extent pointers in block entry */
6031 int bp_release = 0;
6032
6033 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
6034 return;
6035 }
6036
6037 bno = NULLFSBLOCK;
6038 mp = ip->i_mount;
6039 ifp = XFS_IFORK_PTR(ip, whichfork);
6040 block = ifp->if_broot;
6041 /*
6042 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6043 */
6044 ASSERT(INT_GET(block->bb_level, ARCH_CONVERT) > 0);
6045 level = INT_GET(block->bb_level, ARCH_CONVERT);
6046 xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
6047 pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
6048 ASSERT(INT_GET(*pp, ARCH_CONVERT) != NULLDFSBNO);
6049 ASSERT(XFS_FSB_TO_AGNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agcount);
6050 ASSERT(XFS_FSB_TO_AGBNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agblocks);
6051 bno = INT_GET(*pp, ARCH_CONVERT);
6052 /*
6053 * Go down the tree until leaf level is reached, following the first
6054 * pointer (leftmost) at each level.
6055 */
6056 while (level-- > 0) {
6057 /* See if buf is in cur first */
6058 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
6059 if (bp) {
6060 bp_release = 0;
6061 } else {
6062 bp_release = 1;
6063 }
6064 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
6065 XFS_BMAP_BTREE_REF)))
6066 goto error_norelse;
6067 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6068 XFS_WANT_CORRUPTED_GOTO(
6069 XFS_BMAP_SANITY_CHECK(mp, block, level),
6070 error0);
6071 if (level == 0)
6072 break;
6073
6074 /*
6075 * Check this block for basic sanity (increasing keys and
6076 * no duplicate blocks).
6077 */
6078
6079 xfs_check_block(block, mp, 0, 0);
6080 pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt, block,
6081 1, mp->m_bmap_dmxr[1]);
6082 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, INT_GET(*pp, ARCH_CONVERT)), error0);
6083 bno = INT_GET(*pp, ARCH_CONVERT);
6084 if (bp_release) {
6085 bp_release = 0;
6086 xfs_trans_brelse(NULL, bp);
6087 }
6088 }
6089
6090 /*
6091 * Here with bp and block set to the leftmost leaf node in the tree.
6092 */
6093 i = 0;
6094
6095 /*
6096 * Loop over all leaf nodes checking that all extents are in the right order.
6097 */
6098 lastp = NULL;
6099 for (;;) {
6100 xfs_bmbt_rec_t *frp;
6101 xfs_fsblock_t nextbno;
6102 xfs_extnum_t num_recs;
6103
6104
6105 num_recs = INT_GET(block->bb_numrecs, ARCH_CONVERT);
6106
6107 /*
6108 * Read-ahead the next leaf block, if any.
6109 */
6110
6111 nextbno = INT_GET(block->bb_rightsib, ARCH_CONVERT);
6112
6113 /*
6114 * Check all the extents to make sure they are OK.
6115 * If we had a previous block, the last entry should
6116 * conform with the first entry in this one.
6117 */
6118
6119 frp = XFS_BTREE_REC_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt,
6120 block, 1, mp->m_bmap_dmxr[0]);
6121
6122 for (ep = frp;ep < frp + (num_recs - 1); ep++) {
6123 if (lastp) {
6124 xfs_btree_check_rec(XFS_BTNUM_BMAP,
6125 (void *)lastp, (void *)ep);
6126 }
6127 xfs_btree_check_rec(XFS_BTNUM_BMAP, (void *)ep,
6128 (void *)(ep + 1));
6129 }
6130 lastp = frp + num_recs - 1; /* For the next iteration */
6131
6132 i += num_recs;
6133 if (bp_release) {
6134 bp_release = 0;
6135 xfs_trans_brelse(NULL, bp);
6136 }
6137 bno = nextbno;
6138 /*
6139 * If we've reached the end, stop.
6140 */
6141 if (bno == NULLFSBLOCK)
6142 break;
6143
6144 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
6145 if (bp) {
6146 bp_release = 0;
6147 } else {
6148 bp_release = 1;
6149 }
6150 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
6151 XFS_BMAP_BTREE_REF)))
6152 goto error_norelse;
6153 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6154 }
6155 if (bp_release) {
6156 bp_release = 0;
6157 xfs_trans_brelse(NULL, bp);
6158 }
6159 return;
6160
6161error0:
6162 cmn_err(CE_WARN, "%s: at error0", __FUNCTION__);
6163 if (bp_release)
6164 xfs_trans_brelse(NULL, bp);
6165error_norelse:
6166 cmn_err(CE_WARN, "%s: BAD after btree leaves for %d extents",
Christoph Hellwigda1650a2005-11-02 10:21:35 +11006167 __FUNCTION__, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006168 panic("%s: CORRUPTED BTREE OR SOMETHING", __FUNCTION__);
6169 return;
6170}
6171#endif
6172
6173/*
6174 * Count fsblocks of the given fork.
6175 */
6176int /* error */
6177xfs_bmap_count_blocks(
6178 xfs_trans_t *tp, /* transaction pointer */
6179 xfs_inode_t *ip, /* incore inode */
6180 int whichfork, /* data or attr fork */
6181 int *count) /* out: count of blocks */
6182{
6183 xfs_bmbt_block_t *block; /* current btree block */
6184 xfs_fsblock_t bno; /* block # of "block" */
6185 xfs_ifork_t *ifp; /* fork structure */
6186 int level; /* btree level, for checking */
6187 xfs_mount_t *mp; /* file system mount structure */
6188 xfs_bmbt_ptr_t *pp; /* pointer to block address */
6189
6190 bno = NULLFSBLOCK;
6191 mp = ip->i_mount;
6192 ifp = XFS_IFORK_PTR(ip, whichfork);
6193 if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
6194 if (unlikely(xfs_bmap_count_leaves(ifp->if_u1.if_extents,
6195 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
6196 count) < 0)) {
6197 XFS_ERROR_REPORT("xfs_bmap_count_blocks(1)",
6198 XFS_ERRLEVEL_LOW, mp);
6199 return XFS_ERROR(EFSCORRUPTED);
6200 }
6201 return 0;
6202 }
6203
6204 /*
6205 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6206 */
6207 block = ifp->if_broot;
6208 ASSERT(INT_GET(block->bb_level, ARCH_CONVERT) > 0);
6209 level = INT_GET(block->bb_level, ARCH_CONVERT);
6210 pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
6211 ASSERT(INT_GET(*pp, ARCH_CONVERT) != NULLDFSBNO);
6212 ASSERT(XFS_FSB_TO_AGNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agcount);
6213 ASSERT(XFS_FSB_TO_AGBNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agblocks);
6214 bno = INT_GET(*pp, ARCH_CONVERT);
6215
6216 if (unlikely(xfs_bmap_count_tree(mp, tp, bno, level, count) < 0)) {
6217 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
6218 mp);
6219 return XFS_ERROR(EFSCORRUPTED);
6220 }
6221
6222 return 0;
6223}
6224
6225/*
6226 * Recursively walks each level of a btree
6227 * to count total fsblocks is use.
6228 */
6229int /* error */
6230xfs_bmap_count_tree(
6231 xfs_mount_t *mp, /* file system mount point */
6232 xfs_trans_t *tp, /* transaction pointer */
6233 xfs_fsblock_t blockno, /* file system block number */
6234 int levelin, /* level in btree */
6235 int *count) /* Count of blocks */
6236{
6237 int error;
6238 xfs_buf_t *bp, *nbp;
6239 int level = levelin;
6240 xfs_bmbt_ptr_t *pp;
6241 xfs_fsblock_t bno = blockno;
6242 xfs_fsblock_t nextbno;
6243 xfs_bmbt_block_t *block, *nextblock;
6244 int numrecs;
6245 xfs_bmbt_rec_t *frp;
6246
6247 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
6248 return error;
6249 *count += 1;
6250 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6251
6252 if (--level) {
6253 /* Not at node above leafs, count this level of nodes */
6254 nextbno = INT_GET(block->bb_rightsib, ARCH_CONVERT);
6255 while (nextbno != NULLFSBLOCK) {
6256 if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
6257 0, &nbp, XFS_BMAP_BTREE_REF)))
6258 return error;
6259 *count += 1;
6260 nextblock = XFS_BUF_TO_BMBT_BLOCK(nbp);
6261 nextbno = INT_GET(nextblock->bb_rightsib, ARCH_CONVERT);
6262 xfs_trans_brelse(tp, nbp);
6263 }
6264
6265 /* Dive to the next level */
6266 pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize,
6267 xfs_bmbt, block, 1, mp->m_bmap_dmxr[1]);
6268 bno = INT_GET(*pp, ARCH_CONVERT);
6269 if (unlikely((error =
6270 xfs_bmap_count_tree(mp, tp, bno, level, count)) < 0)) {
6271 xfs_trans_brelse(tp, bp);
6272 XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6273 XFS_ERRLEVEL_LOW, mp);
6274 return XFS_ERROR(EFSCORRUPTED);
6275 }
6276 xfs_trans_brelse(tp, bp);
6277 } else {
6278 /* count all level 1 nodes and their leaves */
6279 for (;;) {
6280 nextbno = INT_GET(block->bb_rightsib, ARCH_CONVERT);
6281 numrecs = INT_GET(block->bb_numrecs, ARCH_CONVERT);
6282 frp = XFS_BTREE_REC_ADDR(mp->m_sb.sb_blocksize,
6283 xfs_bmbt, block, 1, mp->m_bmap_dmxr[0]);
6284 if (unlikely(xfs_bmap_count_leaves(frp, numrecs, count) < 0)) {
6285 xfs_trans_brelse(tp, bp);
6286 XFS_ERROR_REPORT("xfs_bmap_count_tree(2)",
6287 XFS_ERRLEVEL_LOW, mp);
6288 return XFS_ERROR(EFSCORRUPTED);
6289 }
6290 xfs_trans_brelse(tp, bp);
6291 if (nextbno == NULLFSBLOCK)
6292 break;
6293 bno = nextbno;
6294 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6295 XFS_BMAP_BTREE_REF)))
6296 return error;
6297 *count += 1;
6298 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6299 }
6300 }
6301 return 0;
6302}
6303
6304/*
6305 * Count leaf blocks given a pointer to an extent list.
6306 */
6307int
6308xfs_bmap_count_leaves(
6309 xfs_bmbt_rec_t *frp,
6310 int numrecs,
6311 int *count)
6312{
6313 int b;
6314
6315 for ( b = 1; b <= numrecs; b++, frp++)
6316 *count += xfs_bmbt_disk_get_blockcount(frp);
6317 return 0;
6318}