?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
GUI: Fix Tomato RAF theme for all builds. Compilation typo.
[tomato.git] / release / src-rt-6.x.4708 / linux / linux-2.6.36 / fs / xfs / xfs_ialloc.c
blobe1072d34af71b572d5cd709f6cd504c98634a847
1 /*
2 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_inum.h"
24 #include "xfs_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_mount.h"
28 #include "xfs_bmap_btree.h"
29 #include "xfs_alloc_btree.h"
30 #include "xfs_ialloc_btree.h"
31 #include "xfs_dinode.h"
32 #include "xfs_inode.h"
33 #include "xfs_btree.h"
34 #include "xfs_ialloc.h"
35 #include "xfs_alloc.h"
36 #include "xfs_rtalloc.h"
37 #include "xfs_error.h"
38 #include "xfs_bmap.h"
39
40
41 /*
42 * Allocation group level functions.
43 */
44 static inline int
45 xfs_ialloc_cluster_alignment(
46 xfs_alloc_arg_t *args)
47 {
48 if (xfs_sb_version_hasalign(&args->mp->m_sb) &&
49 args->mp->m_sb.sb_inoalignmt >=
50 XFS_B_TO_FSBT(args->mp, XFS_INODE_CLUSTER_SIZE(args->mp)))
51 return args->mp->m_sb.sb_inoalignmt;
52 return 1;
53 }
54
55 /*
56 * Lookup a record by ino in the btree given by cur.
57 */
58 int /* error */
59 xfs_inobt_lookup(
60 struct xfs_btree_cur *cur, /* btree cursor */
61 xfs_agino_t ino, /* starting inode of chunk */
62 xfs_lookup_t dir, /* <=, >=, == */
63 int *stat) /* success/failure */
64 {
65 cur->bc_rec.i.ir_startino = ino;
66 cur->bc_rec.i.ir_freecount = 0;
67 cur->bc_rec.i.ir_free = 0;
68 return xfs_btree_lookup(cur, dir, stat);
69 }
70
71 /*
72 * Update the record referred to by cur to the value given.
73 * This either works (return 0) or gets an EFSCORRUPTED error.
74 */
75 STATIC int /* error */
76 xfs_inobt_update(
77 struct xfs_btree_cur *cur, /* btree cursor */
78 xfs_inobt_rec_incore_t *irec) /* btree record */
79 {
80 union xfs_btree_rec rec;
81
82 rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino);
83 rec.inobt.ir_freecount = cpu_to_be32(irec->ir_freecount);
84 rec.inobt.ir_free = cpu_to_be64(irec->ir_free);
85 return xfs_btree_update(cur, &rec);
86 }
87
88 /*
89 * Get the data from the pointed-to record.
90 */
91 int /* error */
92 xfs_inobt_get_rec(
93 struct xfs_btree_cur *cur, /* btree cursor */
94 xfs_inobt_rec_incore_t *irec, /* btree record */
95 int *stat) /* output: success/failure */
96 {
97 union xfs_btree_rec *rec;
98 int error;
99
100 error = xfs_btree_get_rec(cur, &rec, stat);
101 if (!error && *stat == 1) {
102 irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino);
103 irec->ir_freecount = be32_to_cpu(rec->inobt.ir_freecount);
104 irec->ir_free = be64_to_cpu(rec->inobt.ir_free);
105 }
106 return error;
107 }
108
109 /*
110 * Verify that the number of free inodes in the AGI is correct.
111 */
112 #ifdef DEBUG
113 STATIC int
114 xfs_check_agi_freecount(
115 struct xfs_btree_cur *cur,
116 struct xfs_agi *agi)
117 {
118 if (cur->bc_nlevels == 1) {
119 xfs_inobt_rec_incore_t rec;
120 int freecount = 0;
121 int error;
122 int i;
123
124 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
125 if (error)
126 return error;
127
128 do {
129 error = xfs_inobt_get_rec(cur, &rec, &i);
130 if (error)
131 return error;
132
133 if (i) {
134 freecount += rec.ir_freecount;
135 error = xfs_btree_increment(cur, 0, &i);
136 if (error)
137 return error;
138 }
139 } while (i == 1);
140
141 if (!XFS_FORCED_SHUTDOWN(cur->bc_mp))
142 ASSERT(freecount == be32_to_cpu(agi->agi_freecount));
143 }
144 return 0;
145 }
146 #else
147 #define xfs_check_agi_freecount(cur, agi) 0
148 #endif
149
150 /*
151 * Initialise a new set of inodes.
152 */
153 STATIC void
154 xfs_ialloc_inode_init(
155 struct xfs_mount *mp,
156 struct xfs_trans *tp,
157 xfs_agnumber_t agno,
158 xfs_agblock_t agbno,
159 xfs_agblock_t length,
160 unsigned int gen)
161 {
162 struct xfs_buf *fbuf;
163 struct xfs_dinode *free;
164 int blks_per_cluster, nbufs, ninodes;
165 int version;
166 int i, j;
167 xfs_daddr_t d;
168
169 /*
170 * Loop over the new block(s), filling in the inodes.
171 * For small block sizes, manipulate the inodes in buffers
172 * which are multiples of the blocks size.
173 */
174 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
175 blks_per_cluster = 1;
176 nbufs = length;
177 ninodes = mp->m_sb.sb_inopblock;
178 } else {
179 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
180 mp->m_sb.sb_blocksize;
181 nbufs = length / blks_per_cluster;
182 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
183 }
184
185 /*
186 * Figure out what version number to use in the inodes we create.
187 * If the superblock version has caught up to the one that supports
188 * the new inode format, then use the new inode version. Otherwise
189 * use the old version so that old kernels will continue to be
190 * able to use the file system.
191 */
192 if (xfs_sb_version_hasnlink(&mp->m_sb))
193 version = 2;
194 else
195 version = 1;
196
197 for (j = 0; j < nbufs; j++) {
198 /*
199 * Get the block.
200 */
201 d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster));
202 fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
203 mp->m_bsize * blks_per_cluster,
204 XBF_LOCK);
205 ASSERT(fbuf);
206 ASSERT(!XFS_BUF_GETERROR(fbuf));
207
208 xfs_biozero(fbuf, 0, ninodes << mp->m_sb.sb_inodelog);
209 for (i = 0; i < ninodes; i++) {
210 int ioffset = i << mp->m_sb.sb_inodelog;
211 uint isize = sizeof(struct xfs_dinode);
212
213 free = xfs_make_iptr(mp, fbuf, i);
214 free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
215 free->di_version = version;
216 free->di_gen = cpu_to_be32(gen);
217 free->di_next_unlinked = cpu_to_be32(NULLAGINO);
218 xfs_trans_log_buf(tp, fbuf, ioffset, ioffset + isize - 1);
219 }
220 xfs_trans_inode_alloc_buf(tp, fbuf);
221 }
222 }
223
224 /*
225 * Allocate new inodes in the allocation group specified by agbp.
226 * Return 0 for success, else error code.
227 */
228 STATIC int /* error code or 0 */
229 xfs_ialloc_ag_alloc(
230 xfs_trans_t *tp, /* transaction pointer */
231 xfs_buf_t *agbp, /* alloc group buffer */
232 int *alloc)
233 {
234 xfs_agi_t *agi; /* allocation group header */
235 xfs_alloc_arg_t args; /* allocation argument structure */
236 xfs_btree_cur_t *cur; /* inode btree cursor */
237 xfs_agnumber_t agno;
238 int error;
239 int i;
240 xfs_agino_t newino; /* new first inode's number */
241 xfs_agino_t newlen; /* new number of inodes */
242 xfs_agino_t thisino; /* current inode number, for loop */
243 int isaligned = 0; /* inode allocation at stripe unit */
244 /* boundary */
245 struct xfs_perag *pag;
246
247 args.tp = tp;
248 args.mp = tp->t_mountp;
249
250 /*
251 * Locking will ensure that we don't have two callers in here
252 * at one time.
253 */
254 newlen = XFS_IALLOC_INODES(args.mp);
255 if (args.mp->m_maxicount &&
256 args.mp->m_sb.sb_icount + newlen > args.mp->m_maxicount)
257 return XFS_ERROR(ENOSPC);
258 args.minlen = args.maxlen = XFS_IALLOC_BLOCKS(args.mp);
259 /*
260 * First try to allocate inodes contiguous with the last-allocated
261 * chunk of inodes. If the filesystem is striped, this will fill
262 * an entire stripe unit with inodes.
263 */
264 agi = XFS_BUF_TO_AGI(agbp);
265 newino = be32_to_cpu(agi->agi_newino);
266 agno = be32_to_cpu(agi->agi_seqno);
267 args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) +
268 XFS_IALLOC_BLOCKS(args.mp);
269 if (likely(newino != NULLAGINO &&
270 (args.agbno < be32_to_cpu(agi->agi_length)))) {
271 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
272 args.type = XFS_ALLOCTYPE_THIS_BNO;
273 args.mod = args.total = args.wasdel = args.isfl =
274 args.userdata = args.minalignslop = 0;
275 args.prod = 1;
276
277 /*
278 * We need to take into account alignment here to ensure that
279 * we don't modify the free list if we fail to have an exact
280 * block. If we don't have an exact match, and every oher
281 * attempt allocation attempt fails, we'll end up cancelling
282 * a dirty transaction and shutting down.
283 *
284 * For an exact allocation, alignment must be 1,
285 * however we need to take cluster alignment into account when
286 * fixing up the freelist. Use the minalignslop field to
287 * indicate that extra blocks might be required for alignment,
288 * but not to use them in the actual exact allocation.
289 */
290 args.alignment = 1;
291 args.minalignslop = xfs_ialloc_cluster_alignment(&args) - 1;
292
293 /* Allow space for the inode btree to split. */
294 args.minleft = args.mp->m_in_maxlevels - 1;
295 if ((error = xfs_alloc_vextent(&args)))
296 return error;
297 } else
298 args.fsbno = NULLFSBLOCK;
299
300 if (unlikely(args.fsbno == NULLFSBLOCK)) {
301 /*
302 * Set the alignment for the allocation.
303 * If stripe alignment is turned on then align at stripe unit
304 * boundary.
305 * If the cluster size is smaller than a filesystem block
306 * then we're doing I/O for inodes in filesystem block size
307 * pieces, so don't need alignment anyway.
308 */
309 isaligned = 0;
310 if (args.mp->m_sinoalign) {
311 ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN));
312 args.alignment = args.mp->m_dalign;
313 isaligned = 1;
314 } else
315 args.alignment = xfs_ialloc_cluster_alignment(&args);
316 /*
317 * Need to figure out where to allocate the inode blocks.
318 * Ideally they should be spaced out through the a.g.
319 * For now, just allocate blocks up front.
320 */
321 args.agbno = be32_to_cpu(agi->agi_root);
322 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
323 /*
324 * Allocate a fixed-size extent of inodes.
325 */
326 args.type = XFS_ALLOCTYPE_NEAR_BNO;
327 args.mod = args.total = args.wasdel = args.isfl =
328 args.userdata = args.minalignslop = 0;
329 args.prod = 1;
330 /*
331 * Allow space for the inode btree to split.
332 */
333 args.minleft = args.mp->m_in_maxlevels - 1;
334 if ((error = xfs_alloc_vextent(&args)))
335 return error;
336 }
337
338 /*
339 * If stripe alignment is turned on, then try again with cluster
340 * alignment.
341 */
342 if (isaligned && args.fsbno == NULLFSBLOCK) {
343 args.type = XFS_ALLOCTYPE_NEAR_BNO;
344 args.agbno = be32_to_cpu(agi->agi_root);
345 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
346 args.alignment = xfs_ialloc_cluster_alignment(&args);
347 if ((error = xfs_alloc_vextent(&args)))
348 return error;
349 }
350
351 if (args.fsbno == NULLFSBLOCK) {
352 *alloc = 0;
353 return 0;
354 }
355 ASSERT(args.len == args.minlen);
356
357 /*
358 * Stamp and write the inode buffers.
359 *
360 * Seed the new inode cluster with a random generation number. This
361 * prevents short-term reuse of generation numbers if a chunk is
362 * freed and then immediately reallocated. We use random numbers
363 * rather than a linear progression to prevent the next generation
364 * number from being easily guessable.
365 */
366 xfs_ialloc_inode_init(args.mp, tp, agno, args.agbno, args.len,
367 random32());
368
369 /*
370 * Convert the results.
371 */
372 newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0);
373 be32_add_cpu(&agi->agi_count, newlen);
374 be32_add_cpu(&agi->agi_freecount, newlen);
375 pag = xfs_perag_get(args.mp, agno);
376 pag->pagi_freecount += newlen;
377 xfs_perag_put(pag);
378 agi->agi_newino = cpu_to_be32(newino);
379
380 /*
381 * Insert records describing the new inode chunk into the btree.
382 */
383 cur = xfs_inobt_init_cursor(args.mp, tp, agbp, agno);
384 for (thisino = newino;
385 thisino < newino + newlen;
386 thisino += XFS_INODES_PER_CHUNK) {
387 cur->bc_rec.i.ir_startino = thisino;
388 cur->bc_rec.i.ir_freecount = XFS_INODES_PER_CHUNK;
389 cur->bc_rec.i.ir_free = XFS_INOBT_ALL_FREE;
390 error = xfs_btree_lookup(cur, XFS_LOOKUP_EQ, &i);
391 if (error) {
392 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
393 return error;
394 }
395 ASSERT(i == 0);
396 error = xfs_btree_insert(cur, &i);
397 if (error) {
398 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
399 return error;
400 }
401 ASSERT(i == 1);
402 }
403 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
404 /*
405 * Log allocation group header fields
406 */
407 xfs_ialloc_log_agi(tp, agbp,
408 XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO);
409 /*
410 * Modify/log superblock values for inode count and inode free count.
411 */
412 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen);
413 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen);
414 *alloc = 1;
415 return 0;
416 }
417
418 STATIC xfs_agnumber_t
419 xfs_ialloc_next_ag(
420 xfs_mount_t *mp)
421 {
422 xfs_agnumber_t agno;
423
424 spin_lock(&mp->m_agirotor_lock);
425 agno = mp->m_agirotor;
426 if (++mp->m_agirotor == mp->m_maxagi)
427 mp->m_agirotor = 0;
428 spin_unlock(&mp->m_agirotor_lock);
429
430 return agno;
431 }
432
433 /*
434 * Select an allocation group to look for a free inode in, based on the parent
435 * inode and then mode. Return the allocation group buffer.
436 */
437 STATIC xfs_buf_t * /* allocation group buffer */
438 xfs_ialloc_ag_select(
439 xfs_trans_t *tp, /* transaction pointer */
440 xfs_ino_t parent, /* parent directory inode number */
441 mode_t mode, /* bits set to indicate file type */
442 int okalloc) /* ok to allocate more space */
443 {
444 xfs_buf_t *agbp; /* allocation group header buffer */
445 xfs_agnumber_t agcount; /* number of ag's in the filesystem */
446 xfs_agnumber_t agno; /* current ag number */
447 int flags; /* alloc buffer locking flags */
448 xfs_extlen_t ineed; /* blocks needed for inode allocation */
449 xfs_extlen_t longest = 0; /* longest extent available */
450 xfs_mount_t *mp; /* mount point structure */
451 int needspace; /* file mode implies space allocated */
452 xfs_perag_t *pag; /* per allocation group data */
453 xfs_agnumber_t pagno; /* parent (starting) ag number */
454
455 /*
456 * Files of these types need at least one block if length > 0
457 * (and they won't fit in the inode, but that's hard to figure out).
458 */
459 needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode);
460 mp = tp->t_mountp;
461 agcount = mp->m_maxagi;
462 if (S_ISDIR(mode))
463 pagno = xfs_ialloc_next_ag(mp);
464 else {
465 pagno = XFS_INO_TO_AGNO(mp, parent);
466 if (pagno >= agcount)
467 pagno = 0;
468 }
469 ASSERT(pagno < agcount);
470 /*
471 * Loop through allocation groups, looking for one with a little
472 * free space in it. Note we don't look for free inodes, exactly.
473 * Instead, we include whether there is a need to allocate inodes
474 * to mean that blocks must be allocated for them,
475 * if none are currently free.
476 */
477 agno = pagno;
478 flags = XFS_ALLOC_FLAG_TRYLOCK;
479 for (;;) {
480 pag = xfs_perag_get(mp, agno);
481 if (!pag->pagi_init) {
482 if (xfs_ialloc_read_agi(mp, tp, agno, &agbp)) {
483 agbp = NULL;
484 goto nextag;
485 }
486 } else
487 agbp = NULL;
488
489 if (!pag->pagi_inodeok) {
490 xfs_ialloc_next_ag(mp);
491 goto unlock_nextag;
492 }
493
494 /*
495 * Is there enough free space for the file plus a block
496 * of inodes (if we need to allocate some)?
497 */
498 ineed = pag->pagi_freecount ? 0 : XFS_IALLOC_BLOCKS(mp);
499 if (ineed && !pag->pagf_init) {
500 if (agbp == NULL &&
501 xfs_ialloc_read_agi(mp, tp, agno, &agbp)) {
502 agbp = NULL;
503 goto nextag;
504 }
505 (void)xfs_alloc_pagf_init(mp, tp, agno, flags);
506 }
507 if (!ineed || pag->pagf_init) {
508 if (ineed && !(longest = pag->pagf_longest))
509 longest = pag->pagf_flcount > 0;
510 if (!ineed ||
511 (pag->pagf_freeblks >= needspace + ineed &&
512 longest >= ineed &&
513 okalloc)) {
514 if (agbp == NULL &&
515 xfs_ialloc_read_agi(mp, tp, agno, &agbp)) {
516 agbp = NULL;
517 goto nextag;
518 }
519 xfs_perag_put(pag);
520 return agbp;
521 }
522 }
523 unlock_nextag:
524 if (agbp)
525 xfs_trans_brelse(tp, agbp);
526 nextag:
527 xfs_perag_put(pag);
528 /*
529 * No point in iterating over the rest, if we're shutting
530 * down.
531 */
532 if (XFS_FORCED_SHUTDOWN(mp))
533 return NULL;
534 agno++;
535 if (agno >= agcount)
536 agno = 0;
537 if (agno == pagno) {
538 if (flags == 0)
539 return NULL;
540 flags = 0;
541 }
542 }
543 }
544
545 /*
546 * Try to retrieve the next record to the left/right from the current one.
547 */
548 STATIC int
549 xfs_ialloc_next_rec(
550 struct xfs_btree_cur *cur,
551 xfs_inobt_rec_incore_t *rec,
552 int *done,
553 int left)
554 {
555 int error;
556 int i;
557
558 if (left)
559 error = xfs_btree_decrement(cur, 0, &i);
560 else
561 error = xfs_btree_increment(cur, 0, &i);
562
563 if (error)
564 return error;
565 *done = !i;
566 if (i) {
567 error = xfs_inobt_get_rec(cur, rec, &i);
568 if (error)
569 return error;
570 XFS_WANT_CORRUPTED_RETURN(i == 1);
571 }
572
573 return 0;
574 }
575
576 STATIC int
577 xfs_ialloc_get_rec(
578 struct xfs_btree_cur *cur,
579 xfs_agino_t agino,
580 xfs_inobt_rec_incore_t *rec,
581 int *done,
582 int left)
583 {
584 int error;
585 int i;
586
587 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i);
588 if (error)
589 return error;
590 *done = !i;
591 if (i) {
592 error = xfs_inobt_get_rec(cur, rec, &i);
593 if (error)
594 return error;
595 XFS_WANT_CORRUPTED_RETURN(i == 1);
596 }
597
598 return 0;
599 }
600
601 /*
602 * Visible inode allocation functions.
603 */
604
605 /*
606 * Allocate an inode on disk.
607 * Mode is used to tell whether the new inode will need space, and whether
608 * it is a directory.
609 *
610 * The arguments IO_agbp and alloc_done are defined to work within
611 * the constraint of one allocation per transaction.
612 * xfs_dialloc() is designed to be called twice if it has to do an
613 * allocation to make more free inodes. On the first call,
614 * IO_agbp should be set to NULL. If an inode is available,
615 * i.e., xfs_dialloc() did not need to do an allocation, an inode
616 * number is returned. In this case, IO_agbp would be set to the
617 * current ag_buf and alloc_done set to false.
618 * If an allocation needed to be done, xfs_dialloc would return
619 * the current ag_buf in IO_agbp and set alloc_done to true.
620 * The caller should then commit the current transaction, allocate a new
621 * transaction, and call xfs_dialloc() again, passing in the previous
622 * value of IO_agbp. IO_agbp should be held across the transactions.
623 * Since the agbp is locked across the two calls, the second call is
624 * guaranteed to have a free inode available.
625 *
626 * Once we successfully pick an inode its number is returned and the
627 * on-disk data structures are updated. The inode itself is not read
628 * in, since doing so would break ordering constraints with xfs_reclaim.
629 */
630 int
631 xfs_dialloc(
632 xfs_trans_t *tp, /* transaction pointer */
633 xfs_ino_t parent, /* parent inode (directory) */
634 mode_t mode, /* mode bits for new inode */
635 int okalloc, /* ok to allocate more space */
636 xfs_buf_t **IO_agbp, /* in/out ag header's buffer */
637 boolean_t *alloc_done, /* true if we needed to replenish
638 inode freelist */
639 xfs_ino_t *inop) /* inode number allocated */
640 {
641 xfs_agnumber_t agcount; /* number of allocation groups */
642 xfs_buf_t *agbp; /* allocation group header's buffer */
643 xfs_agnumber_t agno; /* allocation group number */
644 xfs_agi_t *agi; /* allocation group header structure */
645 xfs_btree_cur_t *cur; /* inode allocation btree cursor */
646 int error; /* error return value */
647 int i; /* result code */
648 int ialloced; /* inode allocation status */
649 int noroom = 0; /* no space for inode blk allocation */
650 xfs_ino_t ino; /* fs-relative inode to be returned */
651 /* REFERENCED */
652 int j; /* result code */
653 xfs_mount_t *mp; /* file system mount structure */
654 int offset; /* index of inode in chunk */
655 xfs_agino_t pagino; /* parent's AG relative inode # */
656 xfs_agnumber_t pagno; /* parent's AG number */
657 xfs_inobt_rec_incore_t rec; /* inode allocation record */
658 xfs_agnumber_t tagno; /* testing allocation group number */
659 xfs_btree_cur_t *tcur; /* temp cursor */
660 xfs_inobt_rec_incore_t trec; /* temp inode allocation record */
661 struct xfs_perag *pag;
662
663
664 if (*IO_agbp == NULL) {
665 /*
666 * We do not have an agbp, so select an initial allocation
667 * group for inode allocation.
668 */
669 agbp = xfs_ialloc_ag_select(tp, parent, mode, okalloc);
670 /*
671 * Couldn't find an allocation group satisfying the
672 * criteria, give up.
673 */
674 if (!agbp) {
675 *inop = NULLFSINO;
676 return 0;
677 }
678 agi = XFS_BUF_TO_AGI(agbp);
679 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
680 } else {
681 /*
682 * Continue where we left off before. In this case, we
683 * know that the allocation group has free inodes.
684 */
685 agbp = *IO_agbp;
686 agi = XFS_BUF_TO_AGI(agbp);
687 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
688 ASSERT(be32_to_cpu(agi->agi_freecount) > 0);
689 }
690 mp = tp->t_mountp;
691 agcount = mp->m_sb.sb_agcount;
692 agno = be32_to_cpu(agi->agi_seqno);
693 tagno = agno;
694 pagno = XFS_INO_TO_AGNO(mp, parent);
695 pagino = XFS_INO_TO_AGINO(mp, parent);
696
697 /*
698 * If we have already hit the ceiling of inode blocks then clear
699 * okalloc so we scan all available agi structures for a free
700 * inode.
701 */
702
703 if (mp->m_maxicount &&
704 mp->m_sb.sb_icount + XFS_IALLOC_INODES(mp) > mp->m_maxicount) {
705 noroom = 1;
706 okalloc = 0;
707 }
708
709 /*
710 * Loop until we find an allocation group that either has free inodes
711 * or in which we can allocate some inodes. Iterate through the
712 * allocation groups upward, wrapping at the end.
713 */
714 *alloc_done = B_FALSE;
715 while (!agi->agi_freecount) {
716 /*
717 * Don't do anything if we're not supposed to allocate
718 * any blocks, just go on to the next ag.
719 */
720 if (okalloc) {
721 /*
722 * Try to allocate some new inodes in the allocation
723 * group.
724 */
725 if ((error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced))) {
726 xfs_trans_brelse(tp, agbp);
727 if (error == ENOSPC) {
728 *inop = NULLFSINO;
729 return 0;
730 } else
731 return error;
732 }
733 if (ialloced) {
734 /*
735 * We successfully allocated some inodes, return
736 * the current context to the caller so that it
737 * can commit the current transaction and call
738 * us again where we left off.
739 */
740 ASSERT(be32_to_cpu(agi->agi_freecount) > 0);
741 *alloc_done = B_TRUE;
742 *IO_agbp = agbp;
743 *inop = NULLFSINO;
744 return 0;
745 }
746 }
747 /*
748 * If it failed, give up on this ag.
749 */
750 xfs_trans_brelse(tp, agbp);
751 /*
752 * Go on to the next ag: get its ag header.
753 */
754 nextag:
755 if (++tagno == agcount)
756 tagno = 0;
757 if (tagno == agno) {
758 *inop = NULLFSINO;
759 return noroom ? ENOSPC : 0;
760 }
761 pag = xfs_perag_get(mp, tagno);
762 if (pag->pagi_inodeok == 0) {
763 xfs_perag_put(pag);
764 goto nextag;
765 }
766 error = xfs_ialloc_read_agi(mp, tp, tagno, &agbp);
767 xfs_perag_put(pag);
768 if (error)
769 goto nextag;
770 agi = XFS_BUF_TO_AGI(agbp);
771 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
772 }
773 /*
774 * Here with an allocation group that has a free inode.
775 * Reset agno since we may have chosen a new ag in the
776 * loop above.
777 */
778 agno = tagno;
779 *IO_agbp = NULL;
780 pag = xfs_perag_get(mp, agno);
781
782 restart_pagno:
783 cur = xfs_inobt_init_cursor(mp, tp, agbp, be32_to_cpu(agi->agi_seqno));
784 /*
785 * If pagino is 0 (this is the root inode allocation) use newino.
786 * This must work because we've just allocated some.
787 */
788 if (!pagino)
789 pagino = be32_to_cpu(agi->agi_newino);
790
791 error = xfs_check_agi_freecount(cur, agi);
792 if (error)
793 goto error0;
794
795 /*
796 * If in the same AG as the parent, try to get near the parent.
797 */
798 if (pagno == agno) {
799 int doneleft; /* done, to the left */
800 int doneright; /* done, to the right */
801 int searchdistance = 10;
802
803 error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i);
804 if (error)
805 goto error0;
806 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
807
808 error = xfs_inobt_get_rec(cur, &rec, &j);
809 if (error)
810 goto error0;
811 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
812
813 if (rec.ir_freecount > 0) {
814 /*
815 * Found a free inode in the same chunk
816 * as the parent, done.
817 */
818 goto alloc_inode;
819 }
820
821
822 /*
823 * In the same AG as parent, but parent's chunk is full.
824 */
825
826 /* duplicate the cursor, search left & right simultaneously */
827 error = xfs_btree_dup_cursor(cur, &tcur);
828 if (error)
829 goto error0;
830
831 /*
832 * Skip to last blocks looked up if same parent inode.
833 */
834 if (pagino != NULLAGINO &&
835 pag->pagl_pagino == pagino &&
836 pag->pagl_leftrec != NULLAGINO &&
837 pag->pagl_rightrec != NULLAGINO) {
838 error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec,
839 &trec, &doneleft, 1);
840 if (error)
841 goto error1;
842
843 error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec,
844 &rec, &doneright, 0);
845 if (error)
846 goto error1;
847 } else {
848 /* search left with tcur, back up 1 record */
849 error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1);
850 if (error)
851 goto error1;
852
853 /* search right with cur, go forward 1 record. */
854 error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0);
855 if (error)
856 goto error1;
857 }
858
859 /*
860 * Loop until we find an inode chunk with a free inode.
861 */
862 while (!doneleft || !doneright) {
863 int useleft; /* using left inode chunk this time */
864
865 if (!--searchdistance) {
866 /*
867 * Not in range - save last search
868 * location and allocate a new inode
869 */
870 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
871 pag->pagl_leftrec = trec.ir_startino;
872 pag->pagl_rightrec = rec.ir_startino;
873 pag->pagl_pagino = pagino;
874 goto newino;
875 }
876
877 /* figure out the closer block if both are valid. */
878 if (!doneleft && !doneright) {
879 useleft = pagino -
880 (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) <
881 rec.ir_startino - pagino;
882 } else {
883 useleft = !doneleft;
884 }
885
886 /* free inodes to the left? */
887 if (useleft && trec.ir_freecount) {
888 rec = trec;
889 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
890 cur = tcur;
891
892 pag->pagl_leftrec = trec.ir_startino;
893 pag->pagl_rightrec = rec.ir_startino;
894 pag->pagl_pagino = pagino;
895 goto alloc_inode;
896 }
897
898 /* free inodes to the right? */
899 if (!useleft && rec.ir_freecount) {
900 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
901
902 pag->pagl_leftrec = trec.ir_startino;
903 pag->pagl_rightrec = rec.ir_startino;
904 pag->pagl_pagino = pagino;
905 goto alloc_inode;
906 }
907
908 /* get next record to check */
909 if (useleft) {
910 error = xfs_ialloc_next_rec(tcur, &trec,
911 &doneleft, 1);
912 } else {
913 error = xfs_ialloc_next_rec(cur, &rec,
914 &doneright, 0);
915 }
916 if (error)
917 goto error1;
918 }
919
920 /*
921 * We've reached the end of the btree. because
922 * we are only searching a small chunk of the
923 * btree each search, there is obviously free
924 * inodes closer to the parent inode than we
925 * are now. restart the search again.
926 */
927 pag->pagl_pagino = NULLAGINO;
928 pag->pagl_leftrec = NULLAGINO;
929 pag->pagl_rightrec = NULLAGINO;
930 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
931 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
932 goto restart_pagno;
933 }
934
935 /*
936 * In a different AG from the parent.
937 * See if the most recently allocated block has any free.
938 */
939 newino:
940 if (be32_to_cpu(agi->agi_newino) != NULLAGINO) {
941 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
942 XFS_LOOKUP_EQ, &i);
943 if (error)
944 goto error0;
945
946 if (i == 1) {
947 error = xfs_inobt_get_rec(cur, &rec, &j);
948 if (error)
949 goto error0;
950
951 if (j == 1 && rec.ir_freecount > 0) {
952 /*
953 * The last chunk allocated in the group
954 * still has a free inode.
955 */
956 goto alloc_inode;
957 }
958 }
959 }
960
961 /*
962 * None left in the last group, search the whole AG
963 */
964 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
965 if (error)
966 goto error0;
967 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
968
969 for (;;) {
970 error = xfs_inobt_get_rec(cur, &rec, &i);
971 if (error)
972 goto error0;
973 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
974 if (rec.ir_freecount > 0)
975 break;
976 error = xfs_btree_increment(cur, 0, &i);
977 if (error)
978 goto error0;
979 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
980 }
981
982 alloc_inode:
983 offset = xfs_ialloc_find_free(&rec.ir_free);
984 ASSERT(offset >= 0);
985 ASSERT(offset < XFS_INODES_PER_CHUNK);
986 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
987 XFS_INODES_PER_CHUNK) == 0);
988 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
989 rec.ir_free &= ~XFS_INOBT_MASK(offset);
990 rec.ir_freecount--;
991 error = xfs_inobt_update(cur, &rec);
992 if (error)
993 goto error0;
994 be32_add_cpu(&agi->agi_freecount, -1);
995 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
996 pag->pagi_freecount--;
997
998 error = xfs_check_agi_freecount(cur, agi);
999 if (error)
1000 goto error0;
1001
1002 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1003 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1004 xfs_perag_put(pag);
1005 *inop = ino;
1006 return 0;
1007 error1:
1008 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
1009 error0:
1010 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1011 xfs_perag_put(pag);
1012 return error;
1013 }
1014
1015 /*
1016 * Free disk inode. Carefully avoids touching the incore inode, all
1017 * manipulations incore are the caller's responsibility.
1018 * The on-disk inode is not changed by this operation, only the
1019 * btree (free inode mask) is changed.
1020 */
1021 int
1022 xfs_difree(
1023 xfs_trans_t *tp, /* transaction pointer */
1024 xfs_ino_t inode, /* inode to be freed */
1025 xfs_bmap_free_t *flist, /* extents to free */
1026 int *delete, /* set if inode cluster was deleted */
1027 xfs_ino_t *first_ino) /* first inode in deleted cluster */
1028 {
1029 /* REFERENCED */
1030 xfs_agblock_t agbno; /* block number containing inode */
1031 xfs_buf_t *agbp; /* buffer containing allocation group header */
1032 xfs_agino_t agino; /* inode number relative to allocation group */
1033 xfs_agnumber_t agno; /* allocation group number */
1034 xfs_agi_t *agi; /* allocation group header */
1035 xfs_btree_cur_t *cur; /* inode btree cursor */
1036 int error; /* error return value */
1037 int i; /* result code */
1038 int ilen; /* inodes in an inode cluster */
1039 xfs_mount_t *mp; /* mount structure for filesystem */
1040 int off; /* offset of inode in inode chunk */
1041 xfs_inobt_rec_incore_t rec; /* btree record */
1042 struct xfs_perag *pag;
1043
1044 mp = tp->t_mountp;
1045
1046 /*
1047 * Break up inode number into its components.
1048 */
1049 agno = XFS_INO_TO_AGNO(mp, inode);
1050 if (agno >= mp->m_sb.sb_agcount) {
1051 cmn_err(CE_WARN,
1052 "xfs_difree: agno >= mp->m_sb.sb_agcount (%d >= %d) on %s. Returning EINVAL.",
1053 agno, mp->m_sb.sb_agcount, mp->m_fsname);
1054 ASSERT(0);
1055 return XFS_ERROR(EINVAL);
1056 }
1057 agino = XFS_INO_TO_AGINO(mp, inode);
1058 if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) {
1059 cmn_err(CE_WARN,
1060 "xfs_difree: inode != XFS_AGINO_TO_INO() "
1061 "(%llu != %llu) on %s. Returning EINVAL.",
1062 (unsigned long long)inode,
1063 (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino),
1064 mp->m_fsname);
1065 ASSERT(0);
1066 return XFS_ERROR(EINVAL);
1067 }
1068 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1069 if (agbno >= mp->m_sb.sb_agblocks) {
1070 cmn_err(CE_WARN,
1071 "xfs_difree: agbno >= mp->m_sb.sb_agblocks (%d >= %d) on %s. Returning EINVAL.",
1072 agbno, mp->m_sb.sb_agblocks, mp->m_fsname);
1073 ASSERT(0);
1074 return XFS_ERROR(EINVAL);
1075 }
1076 /*
1077 * Get the allocation group header.
1078 */
1079 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1080 if (error) {
1081 cmn_err(CE_WARN,
1082 "xfs_difree: xfs_ialloc_read_agi() returned an error %d on %s. Returning error.",
1083 error, mp->m_fsname);
1084 return error;
1085 }
1086 agi = XFS_BUF_TO_AGI(agbp);
1087 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
1088 ASSERT(agbno < be32_to_cpu(agi->agi_length));
1089 /*
1090 * Initialize the cursor.
1091 */
1092 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
1093
1094 error = xfs_check_agi_freecount(cur, agi);
1095 if (error)
1096 goto error0;
1097
1098 /*
1099 * Look for the entry describing this inode.
1100 */
1101 if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) {
1102 cmn_err(CE_WARN,
1103 "xfs_difree: xfs_inobt_lookup returned() an error %d on %s. Returning error.",
1104 error, mp->m_fsname);
1105 goto error0;
1106 }
1107 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1108 error = xfs_inobt_get_rec(cur, &rec, &i);
1109 if (error) {
1110 cmn_err(CE_WARN,
1111 "xfs_difree: xfs_inobt_get_rec() returned an error %d on %s. Returning error.",
1112 error, mp->m_fsname);
1113 goto error0;
1114 }
1115 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1116 /*
1117 * Get the offset in the inode chunk.
1118 */
1119 off = agino - rec.ir_startino;
1120 ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK);
1121 ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off)));
1122 /*
1123 * Mark the inode free & increment the count.
1124 */
1125 rec.ir_free |= XFS_INOBT_MASK(off);
1126 rec.ir_freecount++;
1127
1128 /*
1129 * When an inode cluster is free, it becomes eligible for removal
1130 */
1131 if (!(mp->m_flags & XFS_MOUNT_IKEEP) &&
1132 (rec.ir_freecount == XFS_IALLOC_INODES(mp))) {
1133
1134 *delete = 1;
1135 *first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino);
1136
1137 /*
1138 * Remove the inode cluster from the AGI B+Tree, adjust the
1139 * AGI and Superblock inode counts, and mark the disk space
1140 * to be freed when the transaction is committed.
1141 */
1142 ilen = XFS_IALLOC_INODES(mp);
1143 be32_add_cpu(&agi->agi_count, -ilen);
1144 be32_add_cpu(&agi->agi_freecount, -(ilen - 1));
1145 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
1146 pag = xfs_perag_get(mp, agno);
1147 pag->pagi_freecount -= ilen - 1;
1148 xfs_perag_put(pag);
1149 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen);
1150 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1));
1151
1152 if ((error = xfs_btree_delete(cur, &i))) {
1153 cmn_err(CE_WARN, "xfs_difree: xfs_btree_delete returned an error %d on %s.\n",
1154 error, mp->m_fsname);
1155 goto error0;
1156 }
1157
1158 xfs_bmap_add_free(XFS_AGB_TO_FSB(mp,
1159 agno, XFS_INO_TO_AGBNO(mp,rec.ir_startino)),
1160 XFS_IALLOC_BLOCKS(mp), flist, mp);
1161 } else {
1162 *delete = 0;
1163
1164 error = xfs_inobt_update(cur, &rec);
1165 if (error) {
1166 cmn_err(CE_WARN,
1167 "xfs_difree: xfs_inobt_update returned an error %d on %s.",
1168 error, mp->m_fsname);
1169 goto error0;
1170 }
1171
1172 /*
1173 * Change the inode free counts and log the ag/sb changes.
1174 */
1175 be32_add_cpu(&agi->agi_freecount, 1);
1176 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1177 pag = xfs_perag_get(mp, agno);
1178 pag->pagi_freecount++;
1179 xfs_perag_put(pag);
1180 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
1181 }
1182
1183 error = xfs_check_agi_freecount(cur, agi);
1184 if (error)
1185 goto error0;
1186
1187 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1188 return 0;
1189
1190 error0:
1191 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1192 return error;
1193 }
1194
1195 STATIC int
1196 xfs_imap_lookup(
1197 struct xfs_mount *mp,
1198 struct xfs_trans *tp,
1199 xfs_agnumber_t agno,
1200 xfs_agino_t agino,
1201 xfs_agblock_t agbno,
1202 xfs_agblock_t *chunk_agbno,
1203 xfs_agblock_t *offset_agbno,
1204 int flags)
1205 {
1206 struct xfs_inobt_rec_incore rec;
1207 struct xfs_btree_cur *cur;
1208 struct xfs_buf *agbp;
1209 int error;
1210 int i;
1211
1212 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1213 if (error) {
1214 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
1215 "xfs_ialloc_read_agi() returned "
1216 "error %d, agno %d",
1217 error, agno);
1218 return error;
1219 }
1220
1221 /*
1222 * Lookup the inode record for the given agino. If the record cannot be
1223 * found, then it's an invalid inode number and we should abort. Once
1224 * we have a record, we need to ensure it contains the inode number
1225 * we are looking up.
1226 */
1227 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
1228 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i);
1229 if (!error) {
1230 if (i)
1231 error = xfs_inobt_get_rec(cur, &rec, &i);
1232 if (!error && i == 0)
1233 error = EINVAL;
1234 }
1235
1236 xfs_trans_brelse(tp, agbp);
1237 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1238 if (error)
1239 return error;
1240
1241 /* check that the returned record contains the required inode */
1242 if (rec.ir_startino > agino ||
1243 rec.ir_startino + XFS_IALLOC_INODES(mp) <= agino)
1244 return EINVAL;
1245
1246 /* for untrusted inodes check it is allocated first */
1247 if ((flags & XFS_IGET_UNTRUSTED) &&
1248 (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino)))
1249 return EINVAL;
1250
1251 *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino);
1252 *offset_agbno = agbno - *chunk_agbno;
1253 return 0;
1254 }
1255
1256 /*
1257 * Return the location of the inode in imap, for mapping it into a buffer.
1258 */
1259 int
1260 xfs_imap(
1261 xfs_mount_t *mp, /* file system mount structure */
1262 xfs_trans_t *tp, /* transaction pointer */
1263 xfs_ino_t ino, /* inode to locate */
1264 struct xfs_imap *imap, /* location map structure */
1265 uint flags) /* flags for inode btree lookup */
1266 {
1267 xfs_agblock_t agbno; /* block number of inode in the alloc group */
1268 xfs_agino_t agino; /* inode number within alloc group */
1269 xfs_agnumber_t agno; /* allocation group number */
1270 int blks_per_cluster; /* num blocks per inode cluster */
1271 xfs_agblock_t chunk_agbno; /* first block in inode chunk */
1272 xfs_agblock_t cluster_agbno; /* first block in inode cluster */
1273 int error; /* error code */
1274 int offset; /* index of inode in its buffer */
1275 int offset_agbno; /* blks from chunk start to inode */
1276
1277 ASSERT(ino != NULLFSINO);
1278
1279 /*
1280 * Split up the inode number into its parts.
1281 */
1282 agno = XFS_INO_TO_AGNO(mp, ino);
1283 agino = XFS_INO_TO_AGINO(mp, ino);
1284 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1285 if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks ||
1286 ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1287 #ifdef DEBUG
1288 /*
1289 * Don't output diagnostic information for untrusted inodes
1290 * as they can be invalid without implying corruption.
1291 */
1292 if (flags & XFS_IGET_UNTRUSTED)
1293 return XFS_ERROR(EINVAL);
1294 if (agno >= mp->m_sb.sb_agcount) {
1295 xfs_fs_cmn_err(CE_ALERT, mp,
1296 "xfs_imap: agno (%d) >= "
1297 "mp->m_sb.sb_agcount (%d)",
1298 agno, mp->m_sb.sb_agcount);
1299 }
1300 if (agbno >= mp->m_sb.sb_agblocks) {
1301 xfs_fs_cmn_err(CE_ALERT, mp,
1302 "xfs_imap: agbno (0x%llx) >= "
1303 "mp->m_sb.sb_agblocks (0x%lx)",
1304 (unsigned long long) agbno,
1305 (unsigned long) mp->m_sb.sb_agblocks);
1306 }
1307 if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1308 xfs_fs_cmn_err(CE_ALERT, mp,
1309 "xfs_imap: ino (0x%llx) != "
1310 "XFS_AGINO_TO_INO(mp, agno, agino) "
1311 "(0x%llx)",
1312 ino, XFS_AGINO_TO_INO(mp, agno, agino));
1313 }
1314 xfs_stack_trace();
1315 #endif /* DEBUG */
1316 return XFS_ERROR(EINVAL);
1317 }
1318
1319 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_blocklog;
1320
1321 /*
1322 * For bulkstat and handle lookups, we have an untrusted inode number
1323 * that we have to verify is valid. We cannot do this just by reading
1324 * the inode buffer as it may have been unlinked and removed leaving
1325 * inodes in stale state on disk. Hence we have to do a btree lookup
1326 * in all cases where an untrusted inode number is passed.
1327 */
1328 if (flags & XFS_IGET_UNTRUSTED) {
1329 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
1330 &chunk_agbno, &offset_agbno, flags);
1331 if (error)
1332 return error;
1333 goto out_map;
1334 }
1335
1336 /*
1337 * If the inode cluster size is the same as the blocksize or
1338 * smaller we get to the buffer by simple arithmetics.
1339 */
1340 if (XFS_INODE_CLUSTER_SIZE(mp) <= mp->m_sb.sb_blocksize) {
1341 offset = XFS_INO_TO_OFFSET(mp, ino);
1342 ASSERT(offset < mp->m_sb.sb_inopblock);
1343
1344 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno);
1345 imap->im_len = XFS_FSB_TO_BB(mp, 1);
1346 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1347 return 0;
1348 }
1349
1350 /*
1351 * If the inode chunks are aligned then use simple maths to
1352 * find the location. Otherwise we have to do a btree
1353 * lookup to find the location.
1354 */
1355 if (mp->m_inoalign_mask) {
1356 offset_agbno = agbno & mp->m_inoalign_mask;
1357 chunk_agbno = agbno - offset_agbno;
1358 } else {
1359 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
1360 &chunk_agbno, &offset_agbno, flags);
1361 if (error)
1362 return error;
1363 }
1364
1365 out_map:
1366 ASSERT(agbno >= chunk_agbno);
1367 cluster_agbno = chunk_agbno +
1368 ((offset_agbno / blks_per_cluster) * blks_per_cluster);
1369 offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
1370 XFS_INO_TO_OFFSET(mp, ino);
1371
1372 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno);
1373 imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster);
1374 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1375
1376 /*
1377 * If the inode number maps to a block outside the bounds
1378 * of the file system then return NULL rather than calling
1379 * read_buf and panicing when we get an error from the
1380 * driver.
1381 */
1382 if ((imap->im_blkno + imap->im_len) >
1383 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
1384 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
1385 "(imap->im_blkno (0x%llx) + imap->im_len (0x%llx)) > "
1386 " XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks) (0x%llx)",
1387 (unsigned long long) imap->im_blkno,
1388 (unsigned long long) imap->im_len,
1389 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
1390 return XFS_ERROR(EINVAL);
1391 }
1392 return 0;
1393 }
1394
1395 /*
1396 * Compute and fill in value of m_in_maxlevels.
1397 */
1398 void
1399 xfs_ialloc_compute_maxlevels(
1400 xfs_mount_t *mp) /* file system mount structure */
1401 {
1402 int level;
1403 uint maxblocks;
1404 uint maxleafents;
1405 int minleafrecs;
1406 int minnoderecs;
1407
1408 maxleafents = (1LL << XFS_INO_AGINO_BITS(mp)) >>
1409 XFS_INODES_PER_CHUNK_LOG;
1410 minleafrecs = mp->m_alloc_mnr[0];
1411 minnoderecs = mp->m_alloc_mnr[1];
1412 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
1413 for (level = 1; maxblocks > 1; level++)
1414 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
1415 mp->m_in_maxlevels = level;
1416 }
1417
1418 /*
1419 * Log specified fields for the ag hdr (inode section)
1420 */
1421 void
1422 xfs_ialloc_log_agi(
1423 xfs_trans_t *tp, /* transaction pointer */
1424 xfs_buf_t *bp, /* allocation group header buffer */
1425 int fields) /* bitmask of fields to log */
1426 {
1427 int first; /* first byte number */
1428 int last; /* last byte number */
1429 static const short offsets[] = { /* field starting offsets */
1430 /* keep in sync with bit definitions */
1431 offsetof(xfs_agi_t, agi_magicnum),
1432 offsetof(xfs_agi_t, agi_versionnum),
1433 offsetof(xfs_agi_t, agi_seqno),
1434 offsetof(xfs_agi_t, agi_length),
1435 offsetof(xfs_agi_t, agi_count),
1436 offsetof(xfs_agi_t, agi_root),
1437 offsetof(xfs_agi_t, agi_level),
1438 offsetof(xfs_agi_t, agi_freecount),
1439 offsetof(xfs_agi_t, agi_newino),
1440 offsetof(xfs_agi_t, agi_dirino),
1441 offsetof(xfs_agi_t, agi_unlinked),
1442 sizeof(xfs_agi_t)
1443 };
1444 #ifdef DEBUG
1445 xfs_agi_t *agi; /* allocation group header */
1446
1447 agi = XFS_BUF_TO_AGI(bp);
1448 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
1449 #endif
1450 /*
1451 * Compute byte offsets for the first and last fields.
1452 */
1453 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS, &first, &last);
1454 /*
1455 * Log the allocation group inode header buffer.
1456 */
1457 xfs_trans_log_buf(tp, bp, first, last);
1458 }
1459
1460 #ifdef DEBUG
1461 STATIC void
1462 xfs_check_agi_unlinked(
1463 struct xfs_agi *agi)
1464 {
1465 int i;
1466
1467 for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++)
1468 ASSERT(agi->agi_unlinked[i]);
1469 }
1470 #else
1471 #define xfs_check_agi_unlinked(agi)
1472 #endif
1473
1474 /*
1475 * Read in the allocation group header (inode allocation section)
1476 */
1477 int
1478 xfs_read_agi(
1479 struct xfs_mount *mp, /* file system mount structure */
1480 struct xfs_trans *tp, /* transaction pointer */
1481 xfs_agnumber_t agno, /* allocation group number */
1482 struct xfs_buf **bpp) /* allocation group hdr buf */
1483 {
1484 struct xfs_agi *agi; /* allocation group header */
1485 int agi_ok; /* agi is consistent */
1486 int error;
1487
1488 ASSERT(agno != NULLAGNUMBER);
1489
1490 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
1491 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
1492 XFS_FSS_TO_BB(mp, 1), 0, bpp);
1493 if (error)
1494 return error;
1495
1496 ASSERT(*bpp && !XFS_BUF_GETERROR(*bpp));
1497 agi = XFS_BUF_TO_AGI(*bpp);
1498
1499 /*
1500 * Validate the magic number of the agi block.
1501 */
1502 agi_ok = be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1503 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)) &&
1504 be32_to_cpu(agi->agi_seqno) == agno;
1505 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IALLOC_READ_AGI,
1506 XFS_RANDOM_IALLOC_READ_AGI))) {
1507 XFS_CORRUPTION_ERROR("xfs_read_agi", XFS_ERRLEVEL_LOW,
1508 mp, agi);
1509 xfs_trans_brelse(tp, *bpp);
1510 return XFS_ERROR(EFSCORRUPTED);
1511 }
1512
1513 XFS_BUF_SET_VTYPE_REF(*bpp, B_FS_AGI, XFS_AGI_REF);
1514
1515 xfs_check_agi_unlinked(agi);
1516 return 0;
1517 }
1518
1519 int
1520 xfs_ialloc_read_agi(
1521 struct xfs_mount *mp, /* file system mount structure */
1522 struct xfs_trans *tp, /* transaction pointer */
1523 xfs_agnumber_t agno, /* allocation group number */
1524 struct xfs_buf **bpp) /* allocation group hdr buf */
1525 {
1526 struct xfs_agi *agi; /* allocation group header */
1527 struct xfs_perag *pag; /* per allocation group data */
1528 int error;
1529
1530 error = xfs_read_agi(mp, tp, agno, bpp);
1531 if (error)
1532 return error;
1533
1534 agi = XFS_BUF_TO_AGI(*bpp);
1535 pag = xfs_perag_get(mp, agno);
1536 if (!pag->pagi_init) {
1537 pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
1538 pag->pagi_count = be32_to_cpu(agi->agi_count);
1539 pag->pagi_init = 1;
1540 }
1541
1542 /*
1543 * It's possible for these to be out of sync if
1544 * we are in the middle of a forced shutdown.
1545 */
1546 ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) ||
1547 XFS_FORCED_SHUTDOWN(mp));
1548 xfs_perag_put(pag);
1549 return 0;
1550 }
1551
1552 /*
1553 * Read in the agi to initialise the per-ag data in the mount structure
1554 */
1555 int
1556 xfs_ialloc_pagi_init(
1557 xfs_mount_t *mp, /* file system mount structure */
1558 xfs_trans_t *tp, /* transaction pointer */
1559 xfs_agnumber_t agno) /* allocation group number */
1560 {
1561 xfs_buf_t *bp = NULL;
1562 int error;
1563
1564 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
1565 if (error)
1566 return error;
1567 if (bp)
1568 xfs_trans_brelse(tp, bp);
1569 return 0;
1570 }
Tomato firmware and modifications.