?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
staging: iio: adis16300 clean out some unused code
[linux-2.6/kvm.git] / fs / xfs / xfs_ialloc.c
blob9d884c127bb980c9c24b934113be222a4fc1093d
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_dir2.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_alloc_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_dir2_sf.h"
34 #include "xfs_attr_sf.h"
35 #include "xfs_dinode.h"
36 #include "xfs_inode.h"
37 #include "xfs_btree.h"
38 #include "xfs_ialloc.h"
39 #include "xfs_alloc.h"
40 #include "xfs_rtalloc.h"
41 #include "xfs_error.h"
42 #include "xfs_bmap.h"
43
44
45 /*
46 * Allocation group level functions.
47 */
48 static inline int
49 xfs_ialloc_cluster_alignment(
50 xfs_alloc_arg_t *args)
51 {
52 if (xfs_sb_version_hasalign(&args->mp->m_sb) &&
53 args->mp->m_sb.sb_inoalignmt >=
54 XFS_B_TO_FSBT(args->mp, XFS_INODE_CLUSTER_SIZE(args->mp)))
55 return args->mp->m_sb.sb_inoalignmt;
56 return 1;
57 }
58
59 /*
60 * Lookup a record by ino in the btree given by cur.
61 */
62 int /* error */
63 xfs_inobt_lookup(
64 struct xfs_btree_cur *cur, /* btree cursor */
65 xfs_agino_t ino, /* starting inode of chunk */
66 xfs_lookup_t dir, /* <=, >=, == */
67 int *stat) /* success/failure */
68 {
69 cur->bc_rec.i.ir_startino = ino;
70 cur->bc_rec.i.ir_freecount = 0;
71 cur->bc_rec.i.ir_free = 0;
72 return xfs_btree_lookup(cur, dir, stat);
73 }
74
75 /*
76 * Update the record referred to by cur to the value given.
77 * This either works (return 0) or gets an EFSCORRUPTED error.
78 */
79 STATIC int /* error */
80 xfs_inobt_update(
81 struct xfs_btree_cur *cur, /* btree cursor */
82 xfs_inobt_rec_incore_t *irec) /* btree record */
83 {
84 union xfs_btree_rec rec;
85
86 rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino);
87 rec.inobt.ir_freecount = cpu_to_be32(irec->ir_freecount);
88 rec.inobt.ir_free = cpu_to_be64(irec->ir_free);
89 return xfs_btree_update(cur, &rec);
90 }
91
92 /*
93 * Get the data from the pointed-to record.
94 */
95 int /* error */
96 xfs_inobt_get_rec(
97 struct xfs_btree_cur *cur, /* btree cursor */
98 xfs_inobt_rec_incore_t *irec, /* btree record */
99 int *stat) /* output: success/failure */
100 {
101 union xfs_btree_rec *rec;
102 int error;
103
104 error = xfs_btree_get_rec(cur, &rec, stat);
105 if (!error && *stat == 1) {
106 irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino);
107 irec->ir_freecount = be32_to_cpu(rec->inobt.ir_freecount);
108 irec->ir_free = be64_to_cpu(rec->inobt.ir_free);
109 }
110 return error;
111 }
112
113 /*
114 * Verify that the number of free inodes in the AGI is correct.
115 */
116 #ifdef DEBUG
117 STATIC int
118 xfs_check_agi_freecount(
119 struct xfs_btree_cur *cur,
120 struct xfs_agi *agi)
121 {
122 if (cur->bc_nlevels == 1) {
123 xfs_inobt_rec_incore_t rec;
124 int freecount = 0;
125 int error;
126 int i;
127
128 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
129 if (error)
130 return error;
131
132 do {
133 error = xfs_inobt_get_rec(cur, &rec, &i);
134 if (error)
135 return error;
136
137 if (i) {
138 freecount += rec.ir_freecount;
139 error = xfs_btree_increment(cur, 0, &i);
140 if (error)
141 return error;
142 }
143 } while (i == 1);
144
145 if (!XFS_FORCED_SHUTDOWN(cur->bc_mp))
146 ASSERT(freecount == be32_to_cpu(agi->agi_freecount));
147 }
148 return 0;
149 }
150 #else
151 #define xfs_check_agi_freecount(cur, agi) 0
152 #endif
153
154 /*
155 * Initialise a new set of inodes.
156 */
157 STATIC void
158 xfs_ialloc_inode_init(
159 struct xfs_mount *mp,
160 struct xfs_trans *tp,
161 xfs_agnumber_t agno,
162 xfs_agblock_t agbno,
163 xfs_agblock_t length,
164 unsigned int gen)
165 {
166 struct xfs_buf *fbuf;
167 struct xfs_dinode *free;
168 int blks_per_cluster, nbufs, ninodes;
169 int version;
170 int i, j;
171 xfs_daddr_t d;
172
173 /*
174 * Loop over the new block(s), filling in the inodes.
175 * For small block sizes, manipulate the inodes in buffers
176 * which are multiples of the blocks size.
177 */
178 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
179 blks_per_cluster = 1;
180 nbufs = length;
181 ninodes = mp->m_sb.sb_inopblock;
182 } else {
183 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
184 mp->m_sb.sb_blocksize;
185 nbufs = length / blks_per_cluster;
186 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
187 }
188
189 /*
190 * Figure out what version number to use in the inodes we create.
191 * If the superblock version has caught up to the one that supports
192 * the new inode format, then use the new inode version. Otherwise
193 * use the old version so that old kernels will continue to be
194 * able to use the file system.
195 */
196 if (xfs_sb_version_hasnlink(&mp->m_sb))
197 version = 2;
198 else
199 version = 1;
200
201 for (j = 0; j < nbufs; j++) {
202 /*
203 * Get the block.
204 */
205 d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster));
206 fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
207 mp->m_bsize * blks_per_cluster,
208 XBF_LOCK);
209 ASSERT(fbuf);
210 ASSERT(!XFS_BUF_GETERROR(fbuf));
211
212 /*
213 * Initialize all inodes in this buffer and then log them.
214 *
215 * XXX: It would be much better if we had just one transaction
216 * to log a whole cluster of inodes instead of all the
217 * individual transactions causing a lot of log traffic.
218 */
219 xfs_biozero(fbuf, 0, ninodes << mp->m_sb.sb_inodelog);
220 for (i = 0; i < ninodes; i++) {
221 int ioffset = i << mp->m_sb.sb_inodelog;
222 uint isize = sizeof(struct xfs_dinode);
223
224 free = xfs_make_iptr(mp, fbuf, i);
225 free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
226 free->di_version = version;
227 free->di_gen = cpu_to_be32(gen);
228 free->di_next_unlinked = cpu_to_be32(NULLAGINO);
229 xfs_trans_log_buf(tp, fbuf, ioffset, ioffset + isize - 1);
230 }
231 xfs_trans_inode_alloc_buf(tp, fbuf);
232 }
233 }
234
235 /*
236 * Allocate new inodes in the allocation group specified by agbp.
237 * Return 0 for success, else error code.
238 */
239 STATIC int /* error code or 0 */
240 xfs_ialloc_ag_alloc(
241 xfs_trans_t *tp, /* transaction pointer */
242 xfs_buf_t *agbp, /* alloc group buffer */
243 int *alloc)
244 {
245 xfs_agi_t *agi; /* allocation group header */
246 xfs_alloc_arg_t args; /* allocation argument structure */
247 xfs_btree_cur_t *cur; /* inode btree cursor */
248 xfs_agnumber_t agno;
249 int error;
250 int i;
251 xfs_agino_t newino; /* new first inode's number */
252 xfs_agino_t newlen; /* new number of inodes */
253 xfs_agino_t thisino; /* current inode number, for loop */
254 int isaligned = 0; /* inode allocation at stripe unit */
255 /* boundary */
256 struct xfs_perag *pag;
257
258 args.tp = tp;
259 args.mp = tp->t_mountp;
260
261 /*
262 * Locking will ensure that we don't have two callers in here
263 * at one time.
264 */
265 newlen = XFS_IALLOC_INODES(args.mp);
266 if (args.mp->m_maxicount &&
267 args.mp->m_sb.sb_icount + newlen > args.mp->m_maxicount)
268 return XFS_ERROR(ENOSPC);
269 args.minlen = args.maxlen = XFS_IALLOC_BLOCKS(args.mp);
270 /*
271 * First try to allocate inodes contiguous with the last-allocated
272 * chunk of inodes. If the filesystem is striped, this will fill
273 * an entire stripe unit with inodes.
274 */
275 agi = XFS_BUF_TO_AGI(agbp);
276 newino = be32_to_cpu(agi->agi_newino);
277 agno = be32_to_cpu(agi->agi_seqno);
278 args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) +
279 XFS_IALLOC_BLOCKS(args.mp);
280 if (likely(newino != NULLAGINO &&
281 (args.agbno < be32_to_cpu(agi->agi_length)))) {
282 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
283 args.type = XFS_ALLOCTYPE_THIS_BNO;
284 args.mod = args.total = args.wasdel = args.isfl =
285 args.userdata = args.minalignslop = 0;
286 args.prod = 1;
287
288 /*
289 * We need to take into account alignment here to ensure that
290 * we don't modify the free list if we fail to have an exact
291 * block. If we don't have an exact match, and every oher
292 * attempt allocation attempt fails, we'll end up cancelling
293 * a dirty transaction and shutting down.
294 *
295 * For an exact allocation, alignment must be 1,
296 * however we need to take cluster alignment into account when
297 * fixing up the freelist. Use the minalignslop field to
298 * indicate that extra blocks might be required for alignment,
299 * but not to use them in the actual exact allocation.
300 */
301 args.alignment = 1;
302 args.minalignslop = xfs_ialloc_cluster_alignment(&args) - 1;
303
304 /* Allow space for the inode btree to split. */
305 args.minleft = args.mp->m_in_maxlevels - 1;
306 if ((error = xfs_alloc_vextent(&args)))
307 return error;
308 } else
309 args.fsbno = NULLFSBLOCK;
310
311 if (unlikely(args.fsbno == NULLFSBLOCK)) {
312 /*
313 * Set the alignment for the allocation.
314 * If stripe alignment is turned on then align at stripe unit
315 * boundary.
316 * If the cluster size is smaller than a filesystem block
317 * then we're doing I/O for inodes in filesystem block size
318 * pieces, so don't need alignment anyway.
319 */
320 isaligned = 0;
321 if (args.mp->m_sinoalign) {
322 ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN));
323 args.alignment = args.mp->m_dalign;
324 isaligned = 1;
325 } else
326 args.alignment = xfs_ialloc_cluster_alignment(&args);
327 /*
328 * Need to figure out where to allocate the inode blocks.
329 * Ideally they should be spaced out through the a.g.
330 * For now, just allocate blocks up front.
331 */
332 args.agbno = be32_to_cpu(agi->agi_root);
333 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
334 /*
335 * Allocate a fixed-size extent of inodes.
336 */
337 args.type = XFS_ALLOCTYPE_NEAR_BNO;
338 args.mod = args.total = args.wasdel = args.isfl =
339 args.userdata = args.minalignslop = 0;
340 args.prod = 1;
341 /*
342 * Allow space for the inode btree to split.
343 */
344 args.minleft = args.mp->m_in_maxlevels - 1;
345 if ((error = xfs_alloc_vextent(&args)))
346 return error;
347 }
348
349 /*
350 * If stripe alignment is turned on, then try again with cluster
351 * alignment.
352 */
353 if (isaligned && args.fsbno == NULLFSBLOCK) {
354 args.type = XFS_ALLOCTYPE_NEAR_BNO;
355 args.agbno = be32_to_cpu(agi->agi_root);
356 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
357 args.alignment = xfs_ialloc_cluster_alignment(&args);
358 if ((error = xfs_alloc_vextent(&args)))
359 return error;
360 }
361
362 if (args.fsbno == NULLFSBLOCK) {
363 *alloc = 0;
364 return 0;
365 }
366 ASSERT(args.len == args.minlen);
367
368 /*
369 * Stamp and write the inode buffers.
370 *
371 * Seed the new inode cluster with a random generation number. This
372 * prevents short-term reuse of generation numbers if a chunk is
373 * freed and then immediately reallocated. We use random numbers
374 * rather than a linear progression to prevent the next generation
375 * number from being easily guessable.
376 */
377 xfs_ialloc_inode_init(args.mp, tp, agno, args.agbno, args.len,
378 random32());
379
380 /*
381 * Convert the results.
382 */
383 newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0);
384 be32_add_cpu(&agi->agi_count, newlen);
385 be32_add_cpu(&agi->agi_freecount, newlen);
386 pag = xfs_perag_get(args.mp, agno);
387 pag->pagi_freecount += newlen;
388 xfs_perag_put(pag);
389 agi->agi_newino = cpu_to_be32(newino);
390
391 /*
392 * Insert records describing the new inode chunk into the btree.
393 */
394 cur = xfs_inobt_init_cursor(args.mp, tp, agbp, agno);
395 for (thisino = newino;
396 thisino < newino + newlen;
397 thisino += XFS_INODES_PER_CHUNK) {
398 cur->bc_rec.i.ir_startino = thisino;
399 cur->bc_rec.i.ir_freecount = XFS_INODES_PER_CHUNK;
400 cur->bc_rec.i.ir_free = XFS_INOBT_ALL_FREE;
401 error = xfs_btree_lookup(cur, XFS_LOOKUP_EQ, &i);
402 if (error) {
403 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
404 return error;
405 }
406 ASSERT(i == 0);
407 error = xfs_btree_insert(cur, &i);
408 if (error) {
409 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
410 return error;
411 }
412 ASSERT(i == 1);
413 }
414 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
415 /*
416 * Log allocation group header fields
417 */
418 xfs_ialloc_log_agi(tp, agbp,
419 XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO);
420 /*
421 * Modify/log superblock values for inode count and inode free count.
422 */
423 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen);
424 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen);
425 *alloc = 1;
426 return 0;
427 }
428
429 STATIC xfs_agnumber_t
430 xfs_ialloc_next_ag(
431 xfs_mount_t *mp)
432 {
433 xfs_agnumber_t agno;
434
435 spin_lock(&mp->m_agirotor_lock);
436 agno = mp->m_agirotor;
437 if (++mp->m_agirotor == mp->m_maxagi)
438 mp->m_agirotor = 0;
439 spin_unlock(&mp->m_agirotor_lock);
440
441 return agno;
442 }
443
444 /*
445 * Select an allocation group to look for a free inode in, based on the parent
446 * inode and then mode. Return the allocation group buffer.
447 */
448 STATIC xfs_buf_t * /* allocation group buffer */
449 xfs_ialloc_ag_select(
450 xfs_trans_t *tp, /* transaction pointer */
451 xfs_ino_t parent, /* parent directory inode number */
452 mode_t mode, /* bits set to indicate file type */
453 int okalloc) /* ok to allocate more space */
454 {
455 xfs_buf_t *agbp; /* allocation group header buffer */
456 xfs_agnumber_t agcount; /* number of ag's in the filesystem */
457 xfs_agnumber_t agno; /* current ag number */
458 int flags; /* alloc buffer locking flags */
459 xfs_extlen_t ineed; /* blocks needed for inode allocation */
460 xfs_extlen_t longest = 0; /* longest extent available */
461 xfs_mount_t *mp; /* mount point structure */
462 int needspace; /* file mode implies space allocated */
463 xfs_perag_t *pag; /* per allocation group data */
464 xfs_agnumber_t pagno; /* parent (starting) ag number */
465
466 /*
467 * Files of these types need at least one block if length > 0
468 * (and they won't fit in the inode, but that's hard to figure out).
469 */
470 needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode);
471 mp = tp->t_mountp;
472 agcount = mp->m_maxagi;
473 if (S_ISDIR(mode))
474 pagno = xfs_ialloc_next_ag(mp);
475 else {
476 pagno = XFS_INO_TO_AGNO(mp, parent);
477 if (pagno >= agcount)
478 pagno = 0;
479 }
480 ASSERT(pagno < agcount);
481 /*
482 * Loop through allocation groups, looking for one with a little
483 * free space in it. Note we don't look for free inodes, exactly.
484 * Instead, we include whether there is a need to allocate inodes
485 * to mean that blocks must be allocated for them,
486 * if none are currently free.
487 */
488 agno = pagno;
489 flags = XFS_ALLOC_FLAG_TRYLOCK;
490 for (;;) {
491 pag = xfs_perag_get(mp, agno);
492 if (!pag->pagi_init) {
493 if (xfs_ialloc_read_agi(mp, tp, agno, &agbp)) {
494 agbp = NULL;
495 goto nextag;
496 }
497 } else
498 agbp = NULL;
499
500 if (!pag->pagi_inodeok) {
501 xfs_ialloc_next_ag(mp);
502 goto unlock_nextag;
503 }
504
505 /*
506 * Is there enough free space for the file plus a block
507 * of inodes (if we need to allocate some)?
508 */
509 ineed = pag->pagi_freecount ? 0 : XFS_IALLOC_BLOCKS(mp);
510 if (ineed && !pag->pagf_init) {
511 if (agbp == NULL &&
512 xfs_ialloc_read_agi(mp, tp, agno, &agbp)) {
513 agbp = NULL;
514 goto nextag;
515 }
516 (void)xfs_alloc_pagf_init(mp, tp, agno, flags);
517 }
518 if (!ineed || pag->pagf_init) {
519 if (ineed && !(longest = pag->pagf_longest))
520 longest = pag->pagf_flcount > 0;
521 if (!ineed ||
522 (pag->pagf_freeblks >= needspace + ineed &&
523 longest >= ineed &&
524 okalloc)) {
525 if (agbp == NULL &&
526 xfs_ialloc_read_agi(mp, tp, agno, &agbp)) {
527 agbp = NULL;
528 goto nextag;
529 }
530 xfs_perag_put(pag);
531 return agbp;
532 }
533 }
534 unlock_nextag:
535 if (agbp)
536 xfs_trans_brelse(tp, agbp);
537 nextag:
538 xfs_perag_put(pag);
539 /*
540 * No point in iterating over the rest, if we're shutting
541 * down.
542 */
543 if (XFS_FORCED_SHUTDOWN(mp))
544 return NULL;
545 agno++;
546 if (agno >= agcount)
547 agno = 0;
548 if (agno == pagno) {
549 if (flags == 0)
550 return NULL;
551 flags = 0;
552 }
553 }
554 }
555
556 /*
557 * Try to retrieve the next record to the left/right from the current one.
558 */
559 STATIC int
560 xfs_ialloc_next_rec(
561 struct xfs_btree_cur *cur,
562 xfs_inobt_rec_incore_t *rec,
563 int *done,
564 int left)
565 {
566 int error;
567 int i;
568
569 if (left)
570 error = xfs_btree_decrement(cur, 0, &i);
571 else
572 error = xfs_btree_increment(cur, 0, &i);
573
574 if (error)
575 return error;
576 *done = !i;
577 if (i) {
578 error = xfs_inobt_get_rec(cur, rec, &i);
579 if (error)
580 return error;
581 XFS_WANT_CORRUPTED_RETURN(i == 1);
582 }
583
584 return 0;
585 }
586
587 STATIC int
588 xfs_ialloc_get_rec(
589 struct xfs_btree_cur *cur,
590 xfs_agino_t agino,
591 xfs_inobt_rec_incore_t *rec,
592 int *done,
593 int left)
594 {
595 int error;
596 int i;
597
598 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i);
599 if (error)
600 return error;
601 *done = !i;
602 if (i) {
603 error = xfs_inobt_get_rec(cur, rec, &i);
604 if (error)
605 return error;
606 XFS_WANT_CORRUPTED_RETURN(i == 1);
607 }
608
609 return 0;
610 }
611
612 /*
613 * Visible inode allocation functions.
614 */
615
616 /*
617 * Allocate an inode on disk.
618 * Mode is used to tell whether the new inode will need space, and whether
619 * it is a directory.
620 *
621 * The arguments IO_agbp and alloc_done are defined to work within
622 * the constraint of one allocation per transaction.
623 * xfs_dialloc() is designed to be called twice if it has to do an
624 * allocation to make more free inodes. On the first call,
625 * IO_agbp should be set to NULL. If an inode is available,
626 * i.e., xfs_dialloc() did not need to do an allocation, an inode
627 * number is returned. In this case, IO_agbp would be set to the
628 * current ag_buf and alloc_done set to false.
629 * If an allocation needed to be done, xfs_dialloc would return
630 * the current ag_buf in IO_agbp and set alloc_done to true.
631 * The caller should then commit the current transaction, allocate a new
632 * transaction, and call xfs_dialloc() again, passing in the previous
633 * value of IO_agbp. IO_agbp should be held across the transactions.
634 * Since the agbp is locked across the two calls, the second call is
635 * guaranteed to have a free inode available.
636 *
637 * Once we successfully pick an inode its number is returned and the
638 * on-disk data structures are updated. The inode itself is not read
639 * in, since doing so would break ordering constraints with xfs_reclaim.
640 */
641 int
642 xfs_dialloc(
643 xfs_trans_t *tp, /* transaction pointer */
644 xfs_ino_t parent, /* parent inode (directory) */
645 mode_t mode, /* mode bits for new inode */
646 int okalloc, /* ok to allocate more space */
647 xfs_buf_t **IO_agbp, /* in/out ag header's buffer */
648 boolean_t *alloc_done, /* true if we needed to replenish
649 inode freelist */
650 xfs_ino_t *inop) /* inode number allocated */
651 {
652 xfs_agnumber_t agcount; /* number of allocation groups */
653 xfs_buf_t *agbp; /* allocation group header's buffer */
654 xfs_agnumber_t agno; /* allocation group number */
655 xfs_agi_t *agi; /* allocation group header structure */
656 xfs_btree_cur_t *cur; /* inode allocation btree cursor */
657 int error; /* error return value */
658 int i; /* result code */
659 int ialloced; /* inode allocation status */
660 int noroom = 0; /* no space for inode blk allocation */
661 xfs_ino_t ino; /* fs-relative inode to be returned */
662 /* REFERENCED */
663 int j; /* result code */
664 xfs_mount_t *mp; /* file system mount structure */
665 int offset; /* index of inode in chunk */
666 xfs_agino_t pagino; /* parent's AG relative inode # */
667 xfs_agnumber_t pagno; /* parent's AG number */
668 xfs_inobt_rec_incore_t rec; /* inode allocation record */
669 xfs_agnumber_t tagno; /* testing allocation group number */
670 xfs_btree_cur_t *tcur; /* temp cursor */
671 xfs_inobt_rec_incore_t trec; /* temp inode allocation record */
672 struct xfs_perag *pag;
673
674
675 if (*IO_agbp == NULL) {
676 /*
677 * We do not have an agbp, so select an initial allocation
678 * group for inode allocation.
679 */
680 agbp = xfs_ialloc_ag_select(tp, parent, mode, okalloc);
681 /*
682 * Couldn't find an allocation group satisfying the
683 * criteria, give up.
684 */
685 if (!agbp) {
686 *inop = NULLFSINO;
687 return 0;
688 }
689 agi = XFS_BUF_TO_AGI(agbp);
690 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
691 } else {
692 /*
693 * Continue where we left off before. In this case, we
694 * know that the allocation group has free inodes.
695 */
696 agbp = *IO_agbp;
697 agi = XFS_BUF_TO_AGI(agbp);
698 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
699 ASSERT(be32_to_cpu(agi->agi_freecount) > 0);
700 }
701 mp = tp->t_mountp;
702 agcount = mp->m_sb.sb_agcount;
703 agno = be32_to_cpu(agi->agi_seqno);
704 tagno = agno;
705 pagno = XFS_INO_TO_AGNO(mp, parent);
706 pagino = XFS_INO_TO_AGINO(mp, parent);
707
708 /*
709 * If we have already hit the ceiling of inode blocks then clear
710 * okalloc so we scan all available agi structures for a free
711 * inode.
712 */
713
714 if (mp->m_maxicount &&
715 mp->m_sb.sb_icount + XFS_IALLOC_INODES(mp) > mp->m_maxicount) {
716 noroom = 1;
717 okalloc = 0;
718 }
719
720 /*
721 * Loop until we find an allocation group that either has free inodes
722 * or in which we can allocate some inodes. Iterate through the
723 * allocation groups upward, wrapping at the end.
724 */
725 *alloc_done = B_FALSE;
726 while (!agi->agi_freecount) {
727 /*
728 * Don't do anything if we're not supposed to allocate
729 * any blocks, just go on to the next ag.
730 */
731 if (okalloc) {
732 /*
733 * Try to allocate some new inodes in the allocation
734 * group.
735 */
736 if ((error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced))) {
737 xfs_trans_brelse(tp, agbp);
738 if (error == ENOSPC) {
739 *inop = NULLFSINO;
740 return 0;
741 } else
742 return error;
743 }
744 if (ialloced) {
745 /*
746 * We successfully allocated some inodes, return
747 * the current context to the caller so that it
748 * can commit the current transaction and call
749 * us again where we left off.
750 */
751 ASSERT(be32_to_cpu(agi->agi_freecount) > 0);
752 *alloc_done = B_TRUE;
753 *IO_agbp = agbp;
754 *inop = NULLFSINO;
755 return 0;
756 }
757 }
758 /*
759 * If it failed, give up on this ag.
760 */
761 xfs_trans_brelse(tp, agbp);
762 /*
763 * Go on to the next ag: get its ag header.
764 */
765 nextag:
766 if (++tagno == agcount)
767 tagno = 0;
768 if (tagno == agno) {
769 *inop = NULLFSINO;
770 return noroom ? ENOSPC : 0;
771 }
772 pag = xfs_perag_get(mp, tagno);
773 if (pag->pagi_inodeok == 0) {
774 xfs_perag_put(pag);
775 goto nextag;
776 }
777 error = xfs_ialloc_read_agi(mp, tp, tagno, &agbp);
778 xfs_perag_put(pag);
779 if (error)
780 goto nextag;
781 agi = XFS_BUF_TO_AGI(agbp);
782 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
783 }
784 /*
785 * Here with an allocation group that has a free inode.
786 * Reset agno since we may have chosen a new ag in the
787 * loop above.
788 */
789 agno = tagno;
790 *IO_agbp = NULL;
791 pag = xfs_perag_get(mp, agno);
792
793 restart_pagno:
794 cur = xfs_inobt_init_cursor(mp, tp, agbp, be32_to_cpu(agi->agi_seqno));
795 /*
796 * If pagino is 0 (this is the root inode allocation) use newino.
797 * This must work because we've just allocated some.
798 */
799 if (!pagino)
800 pagino = be32_to_cpu(agi->agi_newino);
801
802 error = xfs_check_agi_freecount(cur, agi);
803 if (error)
804 goto error0;
805
806 /*
807 * If in the same AG as the parent, try to get near the parent.
808 */
809 if (pagno == agno) {
810 int doneleft; /* done, to the left */
811 int doneright; /* done, to the right */
812 int searchdistance = 10;
813
814 error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i);
815 if (error)
816 goto error0;
817 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
818
819 error = xfs_inobt_get_rec(cur, &rec, &j);
820 if (error)
821 goto error0;
822 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
823
824 if (rec.ir_freecount > 0) {
825 /*
826 * Found a free inode in the same chunk
827 * as the parent, done.
828 */
829 goto alloc_inode;
830 }
831
832
833 /*
834 * In the same AG as parent, but parent's chunk is full.
835 */
836
837 /* duplicate the cursor, search left & right simultaneously */
838 error = xfs_btree_dup_cursor(cur, &tcur);
839 if (error)
840 goto error0;
841
842 /*
843 * Skip to last blocks looked up if same parent inode.
844 */
845 if (pagino != NULLAGINO &&
846 pag->pagl_pagino == pagino &&
847 pag->pagl_leftrec != NULLAGINO &&
848 pag->pagl_rightrec != NULLAGINO) {
849 error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec,
850 &trec, &doneleft, 1);
851 if (error)
852 goto error1;
853
854 error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec,
855 &rec, &doneright, 0);
856 if (error)
857 goto error1;
858 } else {
859 /* search left with tcur, back up 1 record */
860 error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1);
861 if (error)
862 goto error1;
863
864 /* search right with cur, go forward 1 record. */
865 error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0);
866 if (error)
867 goto error1;
868 }
869
870 /*
871 * Loop until we find an inode chunk with a free inode.
872 */
873 while (!doneleft || !doneright) {
874 int useleft; /* using left inode chunk this time */
875
876 if (!--searchdistance) {
877 /*
878 * Not in range - save last search
879 * location and allocate a new inode
880 */
881 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
882 pag->pagl_leftrec = trec.ir_startino;
883 pag->pagl_rightrec = rec.ir_startino;
884 pag->pagl_pagino = pagino;
885 goto newino;
886 }
887
888 /* figure out the closer block if both are valid. */
889 if (!doneleft && !doneright) {
890 useleft = pagino -
891 (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) <
892 rec.ir_startino - pagino;
893 } else {
894 useleft = !doneleft;
895 }
896
897 /* free inodes to the left? */
898 if (useleft && trec.ir_freecount) {
899 rec = trec;
900 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
901 cur = tcur;
902
903 pag->pagl_leftrec = trec.ir_startino;
904 pag->pagl_rightrec = rec.ir_startino;
905 pag->pagl_pagino = pagino;
906 goto alloc_inode;
907 }
908
909 /* free inodes to the right? */
910 if (!useleft && rec.ir_freecount) {
911 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
912
913 pag->pagl_leftrec = trec.ir_startino;
914 pag->pagl_rightrec = rec.ir_startino;
915 pag->pagl_pagino = pagino;
916 goto alloc_inode;
917 }
918
919 /* get next record to check */
920 if (useleft) {
921 error = xfs_ialloc_next_rec(tcur, &trec,
922 &doneleft, 1);
923 } else {
924 error = xfs_ialloc_next_rec(cur, &rec,
925 &doneright, 0);
926 }
927 if (error)
928 goto error1;
929 }
930
931 /*
932 * We've reached the end of the btree. because
933 * we are only searching a small chunk of the
934 * btree each search, there is obviously free
935 * inodes closer to the parent inode than we
936 * are now. restart the search again.
937 */
938 pag->pagl_pagino = NULLAGINO;
939 pag->pagl_leftrec = NULLAGINO;
940 pag->pagl_rightrec = NULLAGINO;
941 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
942 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
943 goto restart_pagno;
944 }
945
946 /*
947 * In a different AG from the parent.
948 * See if the most recently allocated block has any free.
949 */
950 newino:
951 if (be32_to_cpu(agi->agi_newino) != NULLAGINO) {
952 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
953 XFS_LOOKUP_EQ, &i);
954 if (error)
955 goto error0;
956
957 if (i == 1) {
958 error = xfs_inobt_get_rec(cur, &rec, &j);
959 if (error)
960 goto error0;
961
962 if (j == 1 && rec.ir_freecount > 0) {
963 /*
964 * The last chunk allocated in the group
965 * still has a free inode.
966 */
967 goto alloc_inode;
968 }
969 }
970 }
971
972 /*
973 * None left in the last group, search the whole AG
974 */
975 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
976 if (error)
977 goto error0;
978 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
979
980 for (;;) {
981 error = xfs_inobt_get_rec(cur, &rec, &i);
982 if (error)
983 goto error0;
984 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
985 if (rec.ir_freecount > 0)
986 break;
987 error = xfs_btree_increment(cur, 0, &i);
988 if (error)
989 goto error0;
990 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
991 }
992
993 alloc_inode:
994 offset = xfs_ialloc_find_free(&rec.ir_free);
995 ASSERT(offset >= 0);
996 ASSERT(offset < XFS_INODES_PER_CHUNK);
997 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
998 XFS_INODES_PER_CHUNK) == 0);
999 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
1000 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1001 rec.ir_freecount--;
1002 error = xfs_inobt_update(cur, &rec);
1003 if (error)
1004 goto error0;
1005 be32_add_cpu(&agi->agi_freecount, -1);
1006 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1007 pag->pagi_freecount--;
1008
1009 error = xfs_check_agi_freecount(cur, agi);
1010 if (error)
1011 goto error0;
1012
1013 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1014 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1015 xfs_perag_put(pag);
1016 *inop = ino;
1017 return 0;
1018 error1:
1019 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
1020 error0:
1021 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1022 xfs_perag_put(pag);
1023 return error;
1024 }
1025
1026 /*
1027 * Free disk inode. Carefully avoids touching the incore inode, all
1028 * manipulations incore are the caller's responsibility.
1029 * The on-disk inode is not changed by this operation, only the
1030 * btree (free inode mask) is changed.
1031 */
1032 int
1033 xfs_difree(
1034 xfs_trans_t *tp, /* transaction pointer */
1035 xfs_ino_t inode, /* inode to be freed */
1036 xfs_bmap_free_t *flist, /* extents to free */
1037 int *delete, /* set if inode cluster was deleted */
1038 xfs_ino_t *first_ino) /* first inode in deleted cluster */
1039 {
1040 /* REFERENCED */
1041 xfs_agblock_t agbno; /* block number containing inode */
1042 xfs_buf_t *agbp; /* buffer containing allocation group header */
1043 xfs_agino_t agino; /* inode number relative to allocation group */
1044 xfs_agnumber_t agno; /* allocation group number */
1045 xfs_agi_t *agi; /* allocation group header */
1046 xfs_btree_cur_t *cur; /* inode btree cursor */
1047 int error; /* error return value */
1048 int i; /* result code */
1049 int ilen; /* inodes in an inode cluster */
1050 xfs_mount_t *mp; /* mount structure for filesystem */
1051 int off; /* offset of inode in inode chunk */
1052 xfs_inobt_rec_incore_t rec; /* btree record */
1053 struct xfs_perag *pag;
1054
1055 mp = tp->t_mountp;
1056
1057 /*
1058 * Break up inode number into its components.
1059 */
1060 agno = XFS_INO_TO_AGNO(mp, inode);
1061 if (agno >= mp->m_sb.sb_agcount) {
1062 cmn_err(CE_WARN,
1063 "xfs_difree: agno >= mp->m_sb.sb_agcount (%d >= %d) on %s. Returning EINVAL.",
1064 agno, mp->m_sb.sb_agcount, mp->m_fsname);
1065 ASSERT(0);
1066 return XFS_ERROR(EINVAL);
1067 }
1068 agino = XFS_INO_TO_AGINO(mp, inode);
1069 if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) {
1070 cmn_err(CE_WARN,
1071 "xfs_difree: inode != XFS_AGINO_TO_INO() "
1072 "(%llu != %llu) on %s. Returning EINVAL.",
1073 (unsigned long long)inode,
1074 (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino),
1075 mp->m_fsname);
1076 ASSERT(0);
1077 return XFS_ERROR(EINVAL);
1078 }
1079 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1080 if (agbno >= mp->m_sb.sb_agblocks) {
1081 cmn_err(CE_WARN,
1082 "xfs_difree: agbno >= mp->m_sb.sb_agblocks (%d >= %d) on %s. Returning EINVAL.",
1083 agbno, mp->m_sb.sb_agblocks, mp->m_fsname);
1084 ASSERT(0);
1085 return XFS_ERROR(EINVAL);
1086 }
1087 /*
1088 * Get the allocation group header.
1089 */
1090 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1091 if (error) {
1092 cmn_err(CE_WARN,
1093 "xfs_difree: xfs_ialloc_read_agi() returned an error %d on %s. Returning error.",
1094 error, mp->m_fsname);
1095 return error;
1096 }
1097 agi = XFS_BUF_TO_AGI(agbp);
1098 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
1099 ASSERT(agbno < be32_to_cpu(agi->agi_length));
1100 /*
1101 * Initialize the cursor.
1102 */
1103 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
1104
1105 error = xfs_check_agi_freecount(cur, agi);
1106 if (error)
1107 goto error0;
1108
1109 /*
1110 * Look for the entry describing this inode.
1111 */
1112 if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) {
1113 cmn_err(CE_WARN,
1114 "xfs_difree: xfs_inobt_lookup returned() an error %d on %s. Returning error.",
1115 error, mp->m_fsname);
1116 goto error0;
1117 }
1118 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1119 error = xfs_inobt_get_rec(cur, &rec, &i);
1120 if (error) {
1121 cmn_err(CE_WARN,
1122 "xfs_difree: xfs_inobt_get_rec() returned an error %d on %s. Returning error.",
1123 error, mp->m_fsname);
1124 goto error0;
1125 }
1126 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1127 /*
1128 * Get the offset in the inode chunk.
1129 */
1130 off = agino - rec.ir_startino;
1131 ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK);
1132 ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off)));
1133 /*
1134 * Mark the inode free & increment the count.
1135 */
1136 rec.ir_free |= XFS_INOBT_MASK(off);
1137 rec.ir_freecount++;
1138
1139 /*
1140 * When an inode cluster is free, it becomes eligible for removal
1141 */
1142 if (!(mp->m_flags & XFS_MOUNT_IKEEP) &&
1143 (rec.ir_freecount == XFS_IALLOC_INODES(mp))) {
1144
1145 *delete = 1;
1146 *first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino);
1147
1148 /*
1149 * Remove the inode cluster from the AGI B+Tree, adjust the
1150 * AGI and Superblock inode counts, and mark the disk space
1151 * to be freed when the transaction is committed.
1152 */
1153 ilen = XFS_IALLOC_INODES(mp);
1154 be32_add_cpu(&agi->agi_count, -ilen);
1155 be32_add_cpu(&agi->agi_freecount, -(ilen - 1));
1156 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
1157 pag = xfs_perag_get(mp, agno);
1158 pag->pagi_freecount -= ilen - 1;
1159 xfs_perag_put(pag);
1160 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen);
1161 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1));
1162
1163 if ((error = xfs_btree_delete(cur, &i))) {
1164 cmn_err(CE_WARN, "xfs_difree: xfs_btree_delete returned an error %d on %s.\n",
1165 error, mp->m_fsname);
1166 goto error0;
1167 }
1168
1169 xfs_bmap_add_free(XFS_AGB_TO_FSB(mp,
1170 agno, XFS_INO_TO_AGBNO(mp,rec.ir_startino)),
1171 XFS_IALLOC_BLOCKS(mp), flist, mp);
1172 } else {
1173 *delete = 0;
1174
1175 error = xfs_inobt_update(cur, &rec);
1176 if (error) {
1177 cmn_err(CE_WARN,
1178 "xfs_difree: xfs_inobt_update returned an error %d on %s.",
1179 error, mp->m_fsname);
1180 goto error0;
1181 }
1182
1183 /*
1184 * Change the inode free counts and log the ag/sb changes.
1185 */
1186 be32_add_cpu(&agi->agi_freecount, 1);
1187 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1188 pag = xfs_perag_get(mp, agno);
1189 pag->pagi_freecount++;
1190 xfs_perag_put(pag);
1191 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
1192 }
1193
1194 error = xfs_check_agi_freecount(cur, agi);
1195 if (error)
1196 goto error0;
1197
1198 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1199 return 0;
1200
1201 error0:
1202 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1203 return error;
1204 }
1205
1206 /*
1207 * Return the location of the inode in imap, for mapping it into a buffer.
1208 */
1209 int
1210 xfs_imap(
1211 xfs_mount_t *mp, /* file system mount structure */
1212 xfs_trans_t *tp, /* transaction pointer */
1213 xfs_ino_t ino, /* inode to locate */
1214 struct xfs_imap *imap, /* location map structure */
1215 uint flags) /* flags for inode btree lookup */
1216 {
1217 xfs_agblock_t agbno; /* block number of inode in the alloc group */
1218 xfs_agino_t agino; /* inode number within alloc group */
1219 xfs_agnumber_t agno; /* allocation group number */
1220 int blks_per_cluster; /* num blocks per inode cluster */
1221 xfs_agblock_t chunk_agbno; /* first block in inode chunk */
1222 xfs_agblock_t cluster_agbno; /* first block in inode cluster */
1223 int error; /* error code */
1224 int offset; /* index of inode in its buffer */
1225 int offset_agbno; /* blks from chunk start to inode */
1226
1227 ASSERT(ino != NULLFSINO);
1228
1229 /*
1230 * Split up the inode number into its parts.
1231 */
1232 agno = XFS_INO_TO_AGNO(mp, ino);
1233 agino = XFS_INO_TO_AGINO(mp, ino);
1234 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1235 if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks ||
1236 ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1237 #ifdef DEBUG
1238 /* no diagnostics for bulkstat, ino comes from userspace */
1239 if (flags & XFS_IGET_BULKSTAT)
1240 return XFS_ERROR(EINVAL);
1241 if (agno >= mp->m_sb.sb_agcount) {
1242 xfs_fs_cmn_err(CE_ALERT, mp,
1243 "xfs_imap: agno (%d) >= "
1244 "mp->m_sb.sb_agcount (%d)",
1245 agno, mp->m_sb.sb_agcount);
1246 }
1247 if (agbno >= mp->m_sb.sb_agblocks) {
1248 xfs_fs_cmn_err(CE_ALERT, mp,
1249 "xfs_imap: agbno (0x%llx) >= "
1250 "mp->m_sb.sb_agblocks (0x%lx)",
1251 (unsigned long long) agbno,
1252 (unsigned long) mp->m_sb.sb_agblocks);
1253 }
1254 if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1255 xfs_fs_cmn_err(CE_ALERT, mp,
1256 "xfs_imap: ino (0x%llx) != "
1257 "XFS_AGINO_TO_INO(mp, agno, agino) "
1258 "(0x%llx)",
1259 ino, XFS_AGINO_TO_INO(mp, agno, agino));
1260 }
1261 xfs_stack_trace();
1262 #endif /* DEBUG */
1263 return XFS_ERROR(EINVAL);
1264 }
1265
1266 /*
1267 * If the inode cluster size is the same as the blocksize or
1268 * smaller we get to the buffer by simple arithmetics.
1269 */
1270 if (XFS_INODE_CLUSTER_SIZE(mp) <= mp->m_sb.sb_blocksize) {
1271 offset = XFS_INO_TO_OFFSET(mp, ino);
1272 ASSERT(offset < mp->m_sb.sb_inopblock);
1273
1274 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno);
1275 imap->im_len = XFS_FSB_TO_BB(mp, 1);
1276 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1277 return 0;
1278 }
1279
1280 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_blocklog;
1281
1282 /*
1283 * If we get a block number passed from bulkstat we can use it to
1284 * find the buffer easily.
1285 */
1286 if (imap->im_blkno) {
1287 offset = XFS_INO_TO_OFFSET(mp, ino);
1288 ASSERT(offset < mp->m_sb.sb_inopblock);
1289
1290 cluster_agbno = xfs_daddr_to_agbno(mp, imap->im_blkno);
1291 offset += (agbno - cluster_agbno) * mp->m_sb.sb_inopblock;
1292
1293 imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster);
1294 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1295 return 0;
1296 }
1297
1298 /*
1299 * If the inode chunks are aligned then use simple maths to
1300 * find the location. Otherwise we have to do a btree
1301 * lookup to find the location.
1302 */
1303 if (mp->m_inoalign_mask) {
1304 offset_agbno = agbno & mp->m_inoalign_mask;
1305 chunk_agbno = agbno - offset_agbno;
1306 } else {
1307 xfs_btree_cur_t *cur; /* inode btree cursor */
1308 xfs_inobt_rec_incore_t chunk_rec;
1309 xfs_buf_t *agbp; /* agi buffer */
1310 int i; /* temp state */
1311
1312 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1313 if (error) {
1314 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
1315 "xfs_ialloc_read_agi() returned "
1316 "error %d, agno %d",
1317 error, agno);
1318 return error;
1319 }
1320
1321 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
1322 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i);
1323 if (error) {
1324 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
1325 "xfs_inobt_lookup() failed");
1326 goto error0;
1327 }
1328
1329 error = xfs_inobt_get_rec(cur, &chunk_rec, &i);
1330 if (error) {
1331 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
1332 "xfs_inobt_get_rec() failed");
1333 goto error0;
1334 }
1335 if (i == 0) {
1336 #ifdef DEBUG
1337 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
1338 "xfs_inobt_get_rec() failed");
1339 #endif /* DEBUG */
1340 error = XFS_ERROR(EINVAL);
1341 }
1342 error0:
1343 xfs_trans_brelse(tp, agbp);
1344 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1345 if (error)
1346 return error;
1347 chunk_agbno = XFS_AGINO_TO_AGBNO(mp, chunk_rec.ir_startino);
1348 offset_agbno = agbno - chunk_agbno;
1349 }
1350
1351 ASSERT(agbno >= chunk_agbno);
1352 cluster_agbno = chunk_agbno +
1353 ((offset_agbno / blks_per_cluster) * blks_per_cluster);
1354 offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
1355 XFS_INO_TO_OFFSET(mp, ino);
1356
1357 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno);
1358 imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster);
1359 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1360
1361 /*
1362 * If the inode number maps to a block outside the bounds
1363 * of the file system then return NULL rather than calling
1364 * read_buf and panicing when we get an error from the
1365 * driver.
1366 */
1367 if ((imap->im_blkno + imap->im_len) >
1368 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
1369 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
1370 "(imap->im_blkno (0x%llx) + imap->im_len (0x%llx)) > "
1371 " XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks) (0x%llx)",
1372 (unsigned long long) imap->im_blkno,
1373 (unsigned long long) imap->im_len,
1374 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
1375 return XFS_ERROR(EINVAL);
1376 }
1377 return 0;
1378 }
1379
1380 /*
1381 * Compute and fill in value of m_in_maxlevels.
1382 */
1383 void
1384 xfs_ialloc_compute_maxlevels(
1385 xfs_mount_t *mp) /* file system mount structure */
1386 {
1387 int level;
1388 uint maxblocks;
1389 uint maxleafents;
1390 int minleafrecs;
1391 int minnoderecs;
1392
1393 maxleafents = (1LL << XFS_INO_AGINO_BITS(mp)) >>
1394 XFS_INODES_PER_CHUNK_LOG;
1395 minleafrecs = mp->m_alloc_mnr[0];
1396 minnoderecs = mp->m_alloc_mnr[1];
1397 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
1398 for (level = 1; maxblocks > 1; level++)
1399 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
1400 mp->m_in_maxlevels = level;
1401 }
1402
1403 /*
1404 * Log specified fields for the ag hdr (inode section)
1405 */
1406 void
1407 xfs_ialloc_log_agi(
1408 xfs_trans_t *tp, /* transaction pointer */
1409 xfs_buf_t *bp, /* allocation group header buffer */
1410 int fields) /* bitmask of fields to log */
1411 {
1412 int first; /* first byte number */
1413 int last; /* last byte number */
1414 static const short offsets[] = { /* field starting offsets */
1415 /* keep in sync with bit definitions */
1416 offsetof(xfs_agi_t, agi_magicnum),
1417 offsetof(xfs_agi_t, agi_versionnum),
1418 offsetof(xfs_agi_t, agi_seqno),
1419 offsetof(xfs_agi_t, agi_length),
1420 offsetof(xfs_agi_t, agi_count),
1421 offsetof(xfs_agi_t, agi_root),
1422 offsetof(xfs_agi_t, agi_level),
1423 offsetof(xfs_agi_t, agi_freecount),
1424 offsetof(xfs_agi_t, agi_newino),
1425 offsetof(xfs_agi_t, agi_dirino),
1426 offsetof(xfs_agi_t, agi_unlinked),
1427 sizeof(xfs_agi_t)
1428 };
1429 #ifdef DEBUG
1430 xfs_agi_t *agi; /* allocation group header */
1431
1432 agi = XFS_BUF_TO_AGI(bp);
1433 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
1434 #endif
1435 /*
1436 * Compute byte offsets for the first and last fields.
1437 */
1438 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS, &first, &last);
1439 /*
1440 * Log the allocation group inode header buffer.
1441 */
1442 xfs_trans_log_buf(tp, bp, first, last);
1443 }
1444
1445 #ifdef DEBUG
1446 STATIC void
1447 xfs_check_agi_unlinked(
1448 struct xfs_agi *agi)
1449 {
1450 int i;
1451
1452 for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++)
1453 ASSERT(agi->agi_unlinked[i]);
1454 }
1455 #else
1456 #define xfs_check_agi_unlinked(agi)
1457 #endif
1458
1459 /*
1460 * Read in the allocation group header (inode allocation section)
1461 */
1462 int
1463 xfs_read_agi(
1464 struct xfs_mount *mp, /* file system mount structure */
1465 struct xfs_trans *tp, /* transaction pointer */
1466 xfs_agnumber_t agno, /* allocation group number */
1467 struct xfs_buf **bpp) /* allocation group hdr buf */
1468 {
1469 struct xfs_agi *agi; /* allocation group header */
1470 int agi_ok; /* agi is consistent */
1471 int error;
1472
1473 ASSERT(agno != NULLAGNUMBER);
1474
1475 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
1476 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
1477 XFS_FSS_TO_BB(mp, 1), 0, bpp);
1478 if (error)
1479 return error;
1480
1481 ASSERT(*bpp && !XFS_BUF_GETERROR(*bpp));
1482 agi = XFS_BUF_TO_AGI(*bpp);
1483
1484 /*
1485 * Validate the magic number of the agi block.
1486 */
1487 agi_ok = be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1488 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)) &&
1489 be32_to_cpu(agi->agi_seqno) == agno;
1490 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IALLOC_READ_AGI,
1491 XFS_RANDOM_IALLOC_READ_AGI))) {
1492 XFS_CORRUPTION_ERROR("xfs_read_agi", XFS_ERRLEVEL_LOW,
1493 mp, agi);
1494 xfs_trans_brelse(tp, *bpp);
1495 return XFS_ERROR(EFSCORRUPTED);
1496 }
1497
1498 XFS_BUF_SET_VTYPE_REF(*bpp, B_FS_AGI, XFS_AGI_REF);
1499
1500 xfs_check_agi_unlinked(agi);
1501 return 0;
1502 }
1503
1504 int
1505 xfs_ialloc_read_agi(
1506 struct xfs_mount *mp, /* file system mount structure */
1507 struct xfs_trans *tp, /* transaction pointer */
1508 xfs_agnumber_t agno, /* allocation group number */
1509 struct xfs_buf **bpp) /* allocation group hdr buf */
1510 {
1511 struct xfs_agi *agi; /* allocation group header */
1512 struct xfs_perag *pag; /* per allocation group data */
1513 int error;
1514
1515 error = xfs_read_agi(mp, tp, agno, bpp);
1516 if (error)
1517 return error;
1518
1519 agi = XFS_BUF_TO_AGI(*bpp);
1520 pag = xfs_perag_get(mp, agno);
1521 if (!pag->pagi_init) {
1522 pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
1523 pag->pagi_count = be32_to_cpu(agi->agi_count);
1524 pag->pagi_init = 1;
1525 }
1526
1527 /*
1528 * It's possible for these to be out of sync if
1529 * we are in the middle of a forced shutdown.
1530 */
1531 ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) ||
1532 XFS_FORCED_SHUTDOWN(mp));
1533 xfs_perag_put(pag);
1534 return 0;
1535 }
1536
1537 /*
1538 * Read in the agi to initialise the per-ag data in the mount structure
1539 */
1540 int
1541 xfs_ialloc_pagi_init(
1542 xfs_mount_t *mp, /* file system mount structure */
1543 xfs_trans_t *tp, /* transaction pointer */
1544 xfs_agnumber_t agno) /* allocation group number */
1545 {
1546 xfs_buf_t *bp = NULL;
1547 int error;
1548
1549 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
1550 if (error)
1551 return error;
1552 if (bp)
1553 xfs_trans_brelse(tp, bp);
1554 return 0;
1555 }
kernel-based virtual machine