projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
xfs: convert remaining direct references to m_perag
[deliverable/linux.git] / fs / xfs / xfs_ialloc.c
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 XFS_BUF_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 down_read(&args.mp->m_peraglock);
387 pag = xfs_perag_get(args.mp, agno);
388 pag->pagi_freecount += newlen;
389 xfs_perag_put(pag);
390 up_read(&args.mp->m_peraglock);
391 agi->agi_newino = cpu_to_be32(newino);
392
393 /*
394 * Insert records describing the new inode chunk into the btree.
395 */
396 cur = xfs_inobt_init_cursor(args.mp, tp, agbp, agno);
397 for (thisino = newino;
398 thisino < newino + newlen;
399 thisino += XFS_INODES_PER_CHUNK) {
400 cur->bc_rec.i.ir_startino = thisino;
401 cur->bc_rec.i.ir_freecount = XFS_INODES_PER_CHUNK;
402 cur->bc_rec.i.ir_free = XFS_INOBT_ALL_FREE;
403 error = xfs_btree_lookup(cur, XFS_LOOKUP_EQ, &i);
404 if (error) {
405 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
406 return error;
407 }
408 ASSERT(i == 0);
409 error = xfs_btree_insert(cur, &i);
410 if (error) {
411 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
412 return error;
413 }
414 ASSERT(i == 1);
415 }
416 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
417 /*
418 * Log allocation group header fields
419 */
420 xfs_ialloc_log_agi(tp, agbp,
421 XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO);
422 /*
423 * Modify/log superblock values for inode count and inode free count.
424 */
425 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen);
426 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen);
427 *alloc = 1;
428 return 0;
429 }
430
431 STATIC xfs_agnumber_t
432 xfs_ialloc_next_ag(
433 xfs_mount_t *mp)
434 {
435 xfs_agnumber_t agno;
436
437 spin_lock(&mp->m_agirotor_lock);
438 agno = mp->m_agirotor;
439 if (++mp->m_agirotor == mp->m_maxagi)
440 mp->m_agirotor = 0;
441 spin_unlock(&mp->m_agirotor_lock);
442
443 return agno;
444 }
445
446 /*
447 * Select an allocation group to look for a free inode in, based on the parent
448 * inode and then mode. Return the allocation group buffer.
449 */
450 STATIC xfs_buf_t * /* allocation group buffer */
451 xfs_ialloc_ag_select(
452 xfs_trans_t *tp, /* transaction pointer */
453 xfs_ino_t parent, /* parent directory inode number */
454 mode_t mode, /* bits set to indicate file type */
455 int okalloc) /* ok to allocate more space */
456 {
457 xfs_buf_t *agbp; /* allocation group header buffer */
458 xfs_agnumber_t agcount; /* number of ag's in the filesystem */
459 xfs_agnumber_t agno; /* current ag number */
460 int flags; /* alloc buffer locking flags */
461 xfs_extlen_t ineed; /* blocks needed for inode allocation */
462 xfs_extlen_t longest = 0; /* longest extent available */
463 xfs_mount_t *mp; /* mount point structure */
464 int needspace; /* file mode implies space allocated */
465 xfs_perag_t *pag; /* per allocation group data */
466 xfs_agnumber_t pagno; /* parent (starting) ag number */
467
468 /*
469 * Files of these types need at least one block if length > 0
470 * (and they won't fit in the inode, but that's hard to figure out).
471 */
472 needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode);
473 mp = tp->t_mountp;
474 agcount = mp->m_maxagi;
475 if (S_ISDIR(mode))
476 pagno = xfs_ialloc_next_ag(mp);
477 else {
478 pagno = XFS_INO_TO_AGNO(mp, parent);
479 if (pagno >= agcount)
480 pagno = 0;
481 }
482 ASSERT(pagno < agcount);
483 /*
484 * Loop through allocation groups, looking for one with a little
485 * free space in it. Note we don't look for free inodes, exactly.
486 * Instead, we include whether there is a need to allocate inodes
487 * to mean that blocks must be allocated for them,
488 * if none are currently free.
489 */
490 agno = pagno;
491 flags = XFS_ALLOC_FLAG_TRYLOCK;
492 down_read(&mp->m_peraglock);
493 for (;;) {
494 pag = xfs_perag_get(mp, agno);
495 if (!pag->pagi_init) {
496 if (xfs_ialloc_read_agi(mp, tp, agno, &agbp)) {
497 agbp = NULL;
498 goto nextag;
499 }
500 } else
501 agbp = NULL;
502
503 if (!pag->pagi_inodeok) {
504 xfs_ialloc_next_ag(mp);
505 goto unlock_nextag;
506 }
507
508 /*
509 * Is there enough free space for the file plus a block
510 * of inodes (if we need to allocate some)?
511 */
512 ineed = pag->pagi_freecount ? 0 : XFS_IALLOC_BLOCKS(mp);
513 if (ineed && !pag->pagf_init) {
514 if (agbp == NULL &&
515 xfs_ialloc_read_agi(mp, tp, agno, &agbp)) {
516 agbp = NULL;
517 goto nextag;
518 }
519 (void)xfs_alloc_pagf_init(mp, tp, agno, flags);
520 }
521 if (!ineed || pag->pagf_init) {
522 if (ineed && !(longest = pag->pagf_longest))
523 longest = pag->pagf_flcount > 0;
524 if (!ineed ||
525 (pag->pagf_freeblks >= needspace + ineed &&
526 longest >= ineed &&
527 okalloc)) {
528 if (agbp == NULL &&
529 xfs_ialloc_read_agi(mp, tp, agno, &agbp)) {
530 agbp = NULL;
531 goto nextag;
532 }
533 xfs_perag_put(pag);
534 up_read(&mp->m_peraglock);
535 return agbp;
536 }
537 }
538 unlock_nextag:
539 if (agbp)
540 xfs_trans_brelse(tp, agbp);
541 nextag:
542 xfs_perag_put(pag);
543 /*
544 * No point in iterating over the rest, if we're shutting
545 * down.
546 */
547 if (XFS_FORCED_SHUTDOWN(mp)) {
548 up_read(&mp->m_peraglock);
549 return NULL;
550 }
551 agno++;
552 if (agno >= agcount)
553 agno = 0;
554 if (agno == pagno) {
555 if (flags == 0) {
556 up_read(&mp->m_peraglock);
557 return NULL;
558 }
559 flags = 0;
560 }
561 }
562 }
563
564 /*
565 * Try to retrieve the next record to the left/right from the current one.
566 */
567 STATIC int
568 xfs_ialloc_next_rec(
569 struct xfs_btree_cur *cur,
570 xfs_inobt_rec_incore_t *rec,
571 int *done,
572 int left)
573 {
574 int error;
575 int i;
576
577 if (left)
578 error = xfs_btree_decrement(cur, 0, &i);
579 else
580 error = xfs_btree_increment(cur, 0, &i);
581
582 if (error)
583 return error;
584 *done = !i;
585 if (i) {
586 error = xfs_inobt_get_rec(cur, rec, &i);
587 if (error)
588 return error;
589 XFS_WANT_CORRUPTED_RETURN(i == 1);
590 }
591
592 return 0;
593 }
594
595 STATIC int
596 xfs_ialloc_get_rec(
597 struct xfs_btree_cur *cur,
598 xfs_agino_t agino,
599 xfs_inobt_rec_incore_t *rec,
600 int *done,
601 int left)
602 {
603 int error;
604 int i;
605
606 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i);
607 if (error)
608 return error;
609 *done = !i;
610 if (i) {
611 error = xfs_inobt_get_rec(cur, rec, &i);
612 if (error)
613 return error;
614 XFS_WANT_CORRUPTED_RETURN(i == 1);
615 }
616
617 return 0;
618 }
619
620 /*
621 * Visible inode allocation functions.
622 */
623
624 /*
625 * Allocate an inode on disk.
626 * Mode is used to tell whether the new inode will need space, and whether
627 * it is a directory.
628 *
629 * The arguments IO_agbp and alloc_done are defined to work within
630 * the constraint of one allocation per transaction.
631 * xfs_dialloc() is designed to be called twice if it has to do an
632 * allocation to make more free inodes. On the first call,
633 * IO_agbp should be set to NULL. If an inode is available,
634 * i.e., xfs_dialloc() did not need to do an allocation, an inode
635 * number is returned. In this case, IO_agbp would be set to the
636 * current ag_buf and alloc_done set to false.
637 * If an allocation needed to be done, xfs_dialloc would return
638 * the current ag_buf in IO_agbp and set alloc_done to true.
639 * The caller should then commit the current transaction, allocate a new
640 * transaction, and call xfs_dialloc() again, passing in the previous
641 * value of IO_agbp. IO_agbp should be held across the transactions.
642 * Since the agbp is locked across the two calls, the second call is
643 * guaranteed to have a free inode available.
644 *
645 * Once we successfully pick an inode its number is returned and the
646 * on-disk data structures are updated. The inode itself is not read
647 * in, since doing so would break ordering constraints with xfs_reclaim.
648 */
649 int
650 xfs_dialloc(
651 xfs_trans_t *tp, /* transaction pointer */
652 xfs_ino_t parent, /* parent inode (directory) */
653 mode_t mode, /* mode bits for new inode */
654 int okalloc, /* ok to allocate more space */
655 xfs_buf_t **IO_agbp, /* in/out ag header's buffer */
656 boolean_t *alloc_done, /* true if we needed to replenish
657 inode freelist */
658 xfs_ino_t *inop) /* inode number allocated */
659 {
660 xfs_agnumber_t agcount; /* number of allocation groups */
661 xfs_buf_t *agbp; /* allocation group header's buffer */
662 xfs_agnumber_t agno; /* allocation group number */
663 xfs_agi_t *agi; /* allocation group header structure */
664 xfs_btree_cur_t *cur; /* inode allocation btree cursor */
665 int error; /* error return value */
666 int i; /* result code */
667 int ialloced; /* inode allocation status */
668 int noroom = 0; /* no space for inode blk allocation */
669 xfs_ino_t ino; /* fs-relative inode to be returned */
670 /* REFERENCED */
671 int j; /* result code */
672 xfs_mount_t *mp; /* file system mount structure */
673 int offset; /* index of inode in chunk */
674 xfs_agino_t pagino; /* parent's AG relative inode # */
675 xfs_agnumber_t pagno; /* parent's AG number */
676 xfs_inobt_rec_incore_t rec; /* inode allocation record */
677 xfs_agnumber_t tagno; /* testing allocation group number */
678 xfs_btree_cur_t *tcur; /* temp cursor */
679 xfs_inobt_rec_incore_t trec; /* temp inode allocation record */
680 struct xfs_perag *pag;
681
682
683 if (*IO_agbp == NULL) {
684 /*
685 * We do not have an agbp, so select an initial allocation
686 * group for inode allocation.
687 */
688 agbp = xfs_ialloc_ag_select(tp, parent, mode, okalloc);
689 /*
690 * Couldn't find an allocation group satisfying the
691 * criteria, give up.
692 */
693 if (!agbp) {
694 *inop = NULLFSINO;
695 return 0;
696 }
697 agi = XFS_BUF_TO_AGI(agbp);
698 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
699 } else {
700 /*
701 * Continue where we left off before. In this case, we
702 * know that the allocation group has free inodes.
703 */
704 agbp = *IO_agbp;
705 agi = XFS_BUF_TO_AGI(agbp);
706 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
707 ASSERT(be32_to_cpu(agi->agi_freecount) > 0);
708 }
709 mp = tp->t_mountp;
710 agcount = mp->m_sb.sb_agcount;
711 agno = be32_to_cpu(agi->agi_seqno);
712 tagno = agno;
713 pagno = XFS_INO_TO_AGNO(mp, parent);
714 pagino = XFS_INO_TO_AGINO(mp, parent);
715
716 /*
717 * If we have already hit the ceiling of inode blocks then clear
718 * okalloc so we scan all available agi structures for a free
719 * inode.
720 */
721
722 if (mp->m_maxicount &&
723 mp->m_sb.sb_icount + XFS_IALLOC_INODES(mp) > mp->m_maxicount) {
724 noroom = 1;
725 okalloc = 0;
726 }
727
728 /*
729 * Loop until we find an allocation group that either has free inodes
730 * or in which we can allocate some inodes. Iterate through the
731 * allocation groups upward, wrapping at the end.
732 */
733 *alloc_done = B_FALSE;
734 while (!agi->agi_freecount) {
735 /*
736 * Don't do anything if we're not supposed to allocate
737 * any blocks, just go on to the next ag.
738 */
739 if (okalloc) {
740 /*
741 * Try to allocate some new inodes in the allocation
742 * group.
743 */
744 if ((error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced))) {
745 xfs_trans_brelse(tp, agbp);
746 if (error == ENOSPC) {
747 *inop = NULLFSINO;
748 return 0;
749 } else
750 return error;
751 }
752 if (ialloced) {
753 /*
754 * We successfully allocated some inodes, return
755 * the current context to the caller so that it
756 * can commit the current transaction and call
757 * us again where we left off.
758 */
759 ASSERT(be32_to_cpu(agi->agi_freecount) > 0);
760 *alloc_done = B_TRUE;
761 *IO_agbp = agbp;
762 *inop = NULLFSINO;
763 return 0;
764 }
765 }
766 /*
767 * If it failed, give up on this ag.
768 */
769 xfs_trans_brelse(tp, agbp);
770 /*
771 * Go on to the next ag: get its ag header.
772 */
773 nextag:
774 if (++tagno == agcount)
775 tagno = 0;
776 if (tagno == agno) {
777 *inop = NULLFSINO;
778 return noroom ? ENOSPC : 0;
779 }
780 down_read(&mp->m_peraglock);
781 pag = xfs_perag_get(mp, tagno);
782 if (pag->pagi_inodeok == 0) {
783 xfs_perag_put(pag);
784 up_read(&mp->m_peraglock);
785 goto nextag;
786 }
787 error = xfs_ialloc_read_agi(mp, tp, tagno, &agbp);
788 xfs_perag_put(pag);
789 up_read(&mp->m_peraglock);
790 if (error)
791 goto nextag;
792 agi = XFS_BUF_TO_AGI(agbp);
793 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
794 }
795 /*
796 * Here with an allocation group that has a free inode.
797 * Reset agno since we may have chosen a new ag in the
798 * loop above.
799 */
800 agno = tagno;
801 *IO_agbp = NULL;
802 pag = xfs_perag_get(mp, agno);
803
804 restart_pagno:
805 cur = xfs_inobt_init_cursor(mp, tp, agbp, be32_to_cpu(agi->agi_seqno));
806 /*
807 * If pagino is 0 (this is the root inode allocation) use newino.
808 * This must work because we've just allocated some.
809 */
810 if (!pagino)
811 pagino = be32_to_cpu(agi->agi_newino);
812
813 error = xfs_check_agi_freecount(cur, agi);
814 if (error)
815 goto error0;
816
817 /*
818 * If in the same AG as the parent, try to get near the parent.
819 */
820 if (pagno == agno) {
821 int doneleft; /* done, to the left */
822 int doneright; /* done, to the right */
823 int searchdistance = 10;
824
825 error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i);
826 if (error)
827 goto error0;
828 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
829
830 error = xfs_inobt_get_rec(cur, &rec, &j);
831 if (error)
832 goto error0;
833 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
834
835 if (rec.ir_freecount > 0) {
836 /*
837 * Found a free inode in the same chunk
838 * as the parent, done.
839 */
840 goto alloc_inode;
841 }
842
843
844 /*
845 * In the same AG as parent, but parent's chunk is full.
846 */
847
848 /* duplicate the cursor, search left & right simultaneously */
849 error = xfs_btree_dup_cursor(cur, &tcur);
850 if (error)
851 goto error0;
852
853 /*
854 * Skip to last blocks looked up if same parent inode.
855 */
856 if (pagino != NULLAGINO &&
857 pag->pagl_pagino == pagino &&
858 pag->pagl_leftrec != NULLAGINO &&
859 pag->pagl_rightrec != NULLAGINO) {
860 error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec,
861 &trec, &doneleft, 1);
862 if (error)
863 goto error1;
864
865 error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec,
866 &rec, &doneright, 0);
867 if (error)
868 goto error1;
869 } else {
870 /* search left with tcur, back up 1 record */
871 error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1);
872 if (error)
873 goto error1;
874
875 /* search right with cur, go forward 1 record. */
876 error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0);
877 if (error)
878 goto error1;
879 }
880
881 /*
882 * Loop until we find an inode chunk with a free inode.
883 */
884 while (!doneleft || !doneright) {
885 int useleft; /* using left inode chunk this time */
886
887 if (!--searchdistance) {
888 /*
889 * Not in range - save last search
890 * location and allocate a new inode
891 */
892 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
893 pag->pagl_leftrec = trec.ir_startino;
894 pag->pagl_rightrec = rec.ir_startino;
895 pag->pagl_pagino = pagino;
896 goto newino;
897 }
898
899 /* figure out the closer block if both are valid. */
900 if (!doneleft && !doneright) {
901 useleft = pagino -
902 (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) <
903 rec.ir_startino - pagino;
904 } else {
905 useleft = !doneleft;
906 }
907
908 /* free inodes to the left? */
909 if (useleft && trec.ir_freecount) {
910 rec = trec;
911 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
912 cur = tcur;
913
914 pag->pagl_leftrec = trec.ir_startino;
915 pag->pagl_rightrec = rec.ir_startino;
916 pag->pagl_pagino = pagino;
917 goto alloc_inode;
918 }
919
920 /* free inodes to the right? */
921 if (!useleft && rec.ir_freecount) {
922 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
923
924 pag->pagl_leftrec = trec.ir_startino;
925 pag->pagl_rightrec = rec.ir_startino;
926 pag->pagl_pagino = pagino;
927 goto alloc_inode;
928 }
929
930 /* get next record to check */
931 if (useleft) {
932 error = xfs_ialloc_next_rec(tcur, &trec,
933 &doneleft, 1);
934 } else {
935 error = xfs_ialloc_next_rec(cur, &rec,
936 &doneright, 0);
937 }
938 if (error)
939 goto error1;
940 }
941
942 /*
943 * We've reached the end of the btree. because
944 * we are only searching a small chunk of the
945 * btree each search, there is obviously free
946 * inodes closer to the parent inode than we
947 * are now. restart the search again.
948 */
949 pag->pagl_pagino = NULLAGINO;
950 pag->pagl_leftrec = NULLAGINO;
951 pag->pagl_rightrec = NULLAGINO;
952 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
953 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
954 goto restart_pagno;
955 }
956
957 /*
958 * In a different AG from the parent.
959 * See if the most recently allocated block has any free.
960 */
961 newino:
962 if (be32_to_cpu(agi->agi_newino) != NULLAGINO) {
963 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
964 XFS_LOOKUP_EQ, &i);
965 if (error)
966 goto error0;
967
968 if (i == 1) {
969 error = xfs_inobt_get_rec(cur, &rec, &j);
970 if (error)
971 goto error0;
972
973 if (j == 1 && rec.ir_freecount > 0) {
974 /*
975 * The last chunk allocated in the group
976 * still has a free inode.
977 */
978 goto alloc_inode;
979 }
980 }
981 }
982
983 /*
984 * None left in the last group, search the whole AG
985 */
986 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
987 if (error)
988 goto error0;
989 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
990
991 for (;;) {
992 error = xfs_inobt_get_rec(cur, &rec, &i);
993 if (error)
994 goto error0;
995 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
996 if (rec.ir_freecount > 0)
997 break;
998 error = xfs_btree_increment(cur, 0, &i);
999 if (error)
1000 goto error0;
1001 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1002 }
1003
1004 alloc_inode:
1005 offset = xfs_ialloc_find_free(&rec.ir_free);
1006 ASSERT(offset >= 0);
1007 ASSERT(offset < XFS_INODES_PER_CHUNK);
1008 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
1009 XFS_INODES_PER_CHUNK) == 0);
1010 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
1011 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1012 rec.ir_freecount--;
1013 error = xfs_inobt_update(cur, &rec);
1014 if (error)
1015 goto error0;
1016 be32_add_cpu(&agi->agi_freecount, -1);
1017 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1018 down_read(&mp->m_peraglock);
1019 pag->pagi_freecount--;
1020 up_read(&mp->m_peraglock);
1021
1022 error = xfs_check_agi_freecount(cur, agi);
1023 if (error)
1024 goto error0;
1025
1026 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1027 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1028 xfs_perag_put(pag);
1029 *inop = ino;
1030 return 0;
1031 error1:
1032 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
1033 error0:
1034 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1035 xfs_perag_put(pag);
1036 return error;
1037 }
1038
1039 /*
1040 * Free disk inode. Carefully avoids touching the incore inode, all
1041 * manipulations incore are the caller's responsibility.
1042 * The on-disk inode is not changed by this operation, only the
1043 * btree (free inode mask) is changed.
1044 */
1045 int
1046 xfs_difree(
1047 xfs_trans_t *tp, /* transaction pointer */
1048 xfs_ino_t inode, /* inode to be freed */
1049 xfs_bmap_free_t *flist, /* extents to free */
1050 int *delete, /* set if inode cluster was deleted */
1051 xfs_ino_t *first_ino) /* first inode in deleted cluster */
1052 {
1053 /* REFERENCED */
1054 xfs_agblock_t agbno; /* block number containing inode */
1055 xfs_buf_t *agbp; /* buffer containing allocation group header */
1056 xfs_agino_t agino; /* inode number relative to allocation group */
1057 xfs_agnumber_t agno; /* allocation group number */
1058 xfs_agi_t *agi; /* allocation group header */
1059 xfs_btree_cur_t *cur; /* inode btree cursor */
1060 int error; /* error return value */
1061 int i; /* result code */
1062 int ilen; /* inodes in an inode cluster */
1063 xfs_mount_t *mp; /* mount structure for filesystem */
1064 int off; /* offset of inode in inode chunk */
1065 xfs_inobt_rec_incore_t rec; /* btree record */
1066 struct xfs_perag *pag;
1067
1068 mp = tp->t_mountp;
1069
1070 /*
1071 * Break up inode number into its components.
1072 */
1073 agno = XFS_INO_TO_AGNO(mp, inode);
1074 if (agno >= mp->m_sb.sb_agcount) {
1075 cmn_err(CE_WARN,
1076 "xfs_difree: agno >= mp->m_sb.sb_agcount (%d >= %d) on %s. Returning EINVAL.",
1077 agno, mp->m_sb.sb_agcount, mp->m_fsname);
1078 ASSERT(0);
1079 return XFS_ERROR(EINVAL);
1080 }
1081 agino = XFS_INO_TO_AGINO(mp, inode);
1082 if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) {
1083 cmn_err(CE_WARN,
1084 "xfs_difree: inode != XFS_AGINO_TO_INO() "
1085 "(%llu != %llu) on %s. Returning EINVAL.",
1086 (unsigned long long)inode,
1087 (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino),
1088 mp->m_fsname);
1089 ASSERT(0);
1090 return XFS_ERROR(EINVAL);
1091 }
1092 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1093 if (agbno >= mp->m_sb.sb_agblocks) {
1094 cmn_err(CE_WARN,
1095 "xfs_difree: agbno >= mp->m_sb.sb_agblocks (%d >= %d) on %s. Returning EINVAL.",
1096 agbno, mp->m_sb.sb_agblocks, mp->m_fsname);
1097 ASSERT(0);
1098 return XFS_ERROR(EINVAL);
1099 }
1100 /*
1101 * Get the allocation group header.
1102 */
1103 down_read(&mp->m_peraglock);
1104 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1105 up_read(&mp->m_peraglock);
1106 if (error) {
1107 cmn_err(CE_WARN,
1108 "xfs_difree: xfs_ialloc_read_agi() returned an error %d on %s. Returning error.",
1109 error, mp->m_fsname);
1110 return error;
1111 }
1112 agi = XFS_BUF_TO_AGI(agbp);
1113 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
1114 ASSERT(agbno < be32_to_cpu(agi->agi_length));
1115 /*
1116 * Initialize the cursor.
1117 */
1118 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
1119
1120 error = xfs_check_agi_freecount(cur, agi);
1121 if (error)
1122 goto error0;
1123
1124 /*
1125 * Look for the entry describing this inode.
1126 */
1127 if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) {
1128 cmn_err(CE_WARN,
1129 "xfs_difree: xfs_inobt_lookup returned() an error %d on %s. Returning error.",
1130 error, mp->m_fsname);
1131 goto error0;
1132 }
1133 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1134 error = xfs_inobt_get_rec(cur, &rec, &i);
1135 if (error) {
1136 cmn_err(CE_WARN,
1137 "xfs_difree: xfs_inobt_get_rec() returned an error %d on %s. Returning error.",
1138 error, mp->m_fsname);
1139 goto error0;
1140 }
1141 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1142 /*
1143 * Get the offset in the inode chunk.
1144 */
1145 off = agino - rec.ir_startino;
1146 ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK);
1147 ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off)));
1148 /*
1149 * Mark the inode free & increment the count.
1150 */
1151 rec.ir_free |= XFS_INOBT_MASK(off);
1152 rec.ir_freecount++;
1153
1154 /*
1155 * When an inode cluster is free, it becomes eligible for removal
1156 */
1157 if (!(mp->m_flags & XFS_MOUNT_IKEEP) &&
1158 (rec.ir_freecount == XFS_IALLOC_INODES(mp))) {
1159
1160 *delete = 1;
1161 *first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino);
1162
1163 /*
1164 * Remove the inode cluster from the AGI B+Tree, adjust the
1165 * AGI and Superblock inode counts, and mark the disk space
1166 * to be freed when the transaction is committed.
1167 */
1168 ilen = XFS_IALLOC_INODES(mp);
1169 be32_add_cpu(&agi->agi_count, -ilen);
1170 be32_add_cpu(&agi->agi_freecount, -(ilen - 1));
1171 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
1172 down_read(&mp->m_peraglock);
1173 pag = xfs_perag_get(mp, agno);
1174 pag->pagi_freecount -= ilen - 1;
1175 xfs_perag_put(pag);
1176 up_read(&mp->m_peraglock);
1177 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen);
1178 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1));
1179
1180 if ((error = xfs_btree_delete(cur, &i))) {
1181 cmn_err(CE_WARN, "xfs_difree: xfs_btree_delete returned an error %d on %s.\n",
1182 error, mp->m_fsname);
1183 goto error0;
1184 }
1185
1186 xfs_bmap_add_free(XFS_AGB_TO_FSB(mp,
1187 agno, XFS_INO_TO_AGBNO(mp,rec.ir_startino)),
1188 XFS_IALLOC_BLOCKS(mp), flist, mp);
1189 } else {
1190 *delete = 0;
1191
1192 error = xfs_inobt_update(cur, &rec);
1193 if (error) {
1194 cmn_err(CE_WARN,
1195 "xfs_difree: xfs_inobt_update returned an error %d on %s.",
1196 error, mp->m_fsname);
1197 goto error0;
1198 }
1199
1200 /*
1201 * Change the inode free counts and log the ag/sb changes.
1202 */
1203 be32_add_cpu(&agi->agi_freecount, 1);
1204 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1205 down_read(&mp->m_peraglock);
1206 pag = xfs_perag_get(mp, agno);
1207 pag->pagi_freecount++;
1208 xfs_perag_put(pag);
1209 up_read(&mp->m_peraglock);
1210 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
1211 }
1212
1213 error = xfs_check_agi_freecount(cur, agi);
1214 if (error)
1215 goto error0;
1216
1217 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1218 return 0;
1219
1220 error0:
1221 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1222 return error;
1223 }
1224
1225 /*
1226 * Return the location of the inode in imap, for mapping it into a buffer.
1227 */
1228 int
1229 xfs_imap(
1230 xfs_mount_t *mp, /* file system mount structure */
1231 xfs_trans_t *tp, /* transaction pointer */
1232 xfs_ino_t ino, /* inode to locate */
1233 struct xfs_imap *imap, /* location map structure */
1234 uint flags) /* flags for inode btree lookup */
1235 {
1236 xfs_agblock_t agbno; /* block number of inode in the alloc group */
1237 xfs_agino_t agino; /* inode number within alloc group */
1238 xfs_agnumber_t agno; /* allocation group number */
1239 int blks_per_cluster; /* num blocks per inode cluster */
1240 xfs_agblock_t chunk_agbno; /* first block in inode chunk */
1241 xfs_agblock_t cluster_agbno; /* first block in inode cluster */
1242 int error; /* error code */
1243 int offset; /* index of inode in its buffer */
1244 int offset_agbno; /* blks from chunk start to inode */
1245
1246 ASSERT(ino != NULLFSINO);
1247
1248 /*
1249 * Split up the inode number into its parts.
1250 */
1251 agno = XFS_INO_TO_AGNO(mp, ino);
1252 agino = XFS_INO_TO_AGINO(mp, ino);
1253 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1254 if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks ||
1255 ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1256 #ifdef DEBUG
1257 /* no diagnostics for bulkstat, ino comes from userspace */
1258 if (flags & XFS_IGET_BULKSTAT)
1259 return XFS_ERROR(EINVAL);
1260 if (agno >= mp->m_sb.sb_agcount) {
1261 xfs_fs_cmn_err(CE_ALERT, mp,
1262 "xfs_imap: agno (%d) >= "
1263 "mp->m_sb.sb_agcount (%d)",
1264 agno, mp->m_sb.sb_agcount);
1265 }
1266 if (agbno >= mp->m_sb.sb_agblocks) {
1267 xfs_fs_cmn_err(CE_ALERT, mp,
1268 "xfs_imap: agbno (0x%llx) >= "
1269 "mp->m_sb.sb_agblocks (0x%lx)",
1270 (unsigned long long) agbno,
1271 (unsigned long) mp->m_sb.sb_agblocks);
1272 }
1273 if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1274 xfs_fs_cmn_err(CE_ALERT, mp,
1275 "xfs_imap: ino (0x%llx) != "
1276 "XFS_AGINO_TO_INO(mp, agno, agino) "
1277 "(0x%llx)",
1278 ino, XFS_AGINO_TO_INO(mp, agno, agino));
1279 }
1280 xfs_stack_trace();
1281 #endif /* DEBUG */
1282 return XFS_ERROR(EINVAL);
1283 }
1284
1285 /*
1286 * If the inode cluster size is the same as the blocksize or
1287 * smaller we get to the buffer by simple arithmetics.
1288 */
1289 if (XFS_INODE_CLUSTER_SIZE(mp) <= mp->m_sb.sb_blocksize) {
1290 offset = XFS_INO_TO_OFFSET(mp, ino);
1291 ASSERT(offset < mp->m_sb.sb_inopblock);
1292
1293 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno);
1294 imap->im_len = XFS_FSB_TO_BB(mp, 1);
1295 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1296 return 0;
1297 }
1298
1299 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_blocklog;
1300
1301 /*
1302 * If we get a block number passed from bulkstat we can use it to
1303 * find the buffer easily.
1304 */
1305 if (imap->im_blkno) {
1306 offset = XFS_INO_TO_OFFSET(mp, ino);
1307 ASSERT(offset < mp->m_sb.sb_inopblock);
1308
1309 cluster_agbno = xfs_daddr_to_agbno(mp, imap->im_blkno);
1310 offset += (agbno - cluster_agbno) * mp->m_sb.sb_inopblock;
1311
1312 imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster);
1313 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1314 return 0;
1315 }
1316
1317 /*
1318 * If the inode chunks are aligned then use simple maths to
1319 * find the location. Otherwise we have to do a btree
1320 * lookup to find the location.
1321 */
1322 if (mp->m_inoalign_mask) {
1323 offset_agbno = agbno & mp->m_inoalign_mask;
1324 chunk_agbno = agbno - offset_agbno;
1325 } else {
1326 xfs_btree_cur_t *cur; /* inode btree cursor */
1327 xfs_inobt_rec_incore_t chunk_rec;
1328 xfs_buf_t *agbp; /* agi buffer */
1329 int i; /* temp state */
1330
1331 down_read(&mp->m_peraglock);
1332 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1333 up_read(&mp->m_peraglock);
1334 if (error) {
1335 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
1336 "xfs_ialloc_read_agi() returned "
1337 "error %d, agno %d",
1338 error, agno);
1339 return error;
1340 }
1341
1342 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
1343 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i);
1344 if (error) {
1345 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
1346 "xfs_inobt_lookup() failed");
1347 goto error0;
1348 }
1349
1350 error = xfs_inobt_get_rec(cur, &chunk_rec, &i);
1351 if (error) {
1352 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
1353 "xfs_inobt_get_rec() failed");
1354 goto error0;
1355 }
1356 if (i == 0) {
1357 #ifdef DEBUG
1358 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
1359 "xfs_inobt_get_rec() failed");
1360 #endif /* DEBUG */
1361 error = XFS_ERROR(EINVAL);
1362 }
1363 error0:
1364 xfs_trans_brelse(tp, agbp);
1365 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1366 if (error)
1367 return error;
1368 chunk_agbno = XFS_AGINO_TO_AGBNO(mp, chunk_rec.ir_startino);
1369 offset_agbno = agbno - chunk_agbno;
1370 }
1371
1372 ASSERT(agbno >= chunk_agbno);
1373 cluster_agbno = chunk_agbno +
1374 ((offset_agbno / blks_per_cluster) * blks_per_cluster);
1375 offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
1376 XFS_INO_TO_OFFSET(mp, ino);
1377
1378 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno);
1379 imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster);
1380 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1381
1382 /*
1383 * If the inode number maps to a block outside the bounds
1384 * of the file system then return NULL rather than calling
1385 * read_buf and panicing when we get an error from the
1386 * driver.
1387 */
1388 if ((imap->im_blkno + imap->im_len) >
1389 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
1390 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
1391 "(imap->im_blkno (0x%llx) + imap->im_len (0x%llx)) > "
1392 " XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks) (0x%llx)",
1393 (unsigned long long) imap->im_blkno,
1394 (unsigned long long) imap->im_len,
1395 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
1396 return XFS_ERROR(EINVAL);
1397 }
1398 return 0;
1399 }
1400
1401 /*
1402 * Compute and fill in value of m_in_maxlevels.
1403 */
1404 void
1405 xfs_ialloc_compute_maxlevels(
1406 xfs_mount_t *mp) /* file system mount structure */
1407 {
1408 int level;
1409 uint maxblocks;
1410 uint maxleafents;
1411 int minleafrecs;
1412 int minnoderecs;
1413
1414 maxleafents = (1LL << XFS_INO_AGINO_BITS(mp)) >>
1415 XFS_INODES_PER_CHUNK_LOG;
1416 minleafrecs = mp->m_alloc_mnr[0];
1417 minnoderecs = mp->m_alloc_mnr[1];
1418 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
1419 for (level = 1; maxblocks > 1; level++)
1420 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
1421 mp->m_in_maxlevels = level;
1422 }
1423
1424 /*
1425 * Log specified fields for the ag hdr (inode section)
1426 */
1427 void
1428 xfs_ialloc_log_agi(
1429 xfs_trans_t *tp, /* transaction pointer */
1430 xfs_buf_t *bp, /* allocation group header buffer */
1431 int fields) /* bitmask of fields to log */
1432 {
1433 int first; /* first byte number */
1434 int last; /* last byte number */
1435 static const short offsets[] = { /* field starting offsets */
1436 /* keep in sync with bit definitions */
1437 offsetof(xfs_agi_t, agi_magicnum),
1438 offsetof(xfs_agi_t, agi_versionnum),
1439 offsetof(xfs_agi_t, agi_seqno),
1440 offsetof(xfs_agi_t, agi_length),
1441 offsetof(xfs_agi_t, agi_count),
1442 offsetof(xfs_agi_t, agi_root),
1443 offsetof(xfs_agi_t, agi_level),
1444 offsetof(xfs_agi_t, agi_freecount),
1445 offsetof(xfs_agi_t, agi_newino),
1446 offsetof(xfs_agi_t, agi_dirino),
1447 offsetof(xfs_agi_t, agi_unlinked),
1448 sizeof(xfs_agi_t)
1449 };
1450 #ifdef DEBUG
1451 xfs_agi_t *agi; /* allocation group header */
1452
1453 agi = XFS_BUF_TO_AGI(bp);
1454 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
1455 #endif
1456 /*
1457 * Compute byte offsets for the first and last fields.
1458 */
1459 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS, &first, &last);
1460 /*
1461 * Log the allocation group inode header buffer.
1462 */
1463 xfs_trans_log_buf(tp, bp, first, last);
1464 }
1465
1466 #ifdef DEBUG
1467 STATIC void
1468 xfs_check_agi_unlinked(
1469 struct xfs_agi *agi)
1470 {
1471 int i;
1472
1473 for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++)
1474 ASSERT(agi->agi_unlinked[i]);
1475 }
1476 #else
1477 #define xfs_check_agi_unlinked(agi)
1478 #endif
1479
1480 /*
1481 * Read in the allocation group header (inode allocation section)
1482 */
1483 int
1484 xfs_read_agi(
1485 struct xfs_mount *mp, /* file system mount structure */
1486 struct xfs_trans *tp, /* transaction pointer */
1487 xfs_agnumber_t agno, /* allocation group number */
1488 struct xfs_buf **bpp) /* allocation group hdr buf */
1489 {
1490 struct xfs_agi *agi; /* allocation group header */
1491 int agi_ok; /* agi is consistent */
1492 int error;
1493
1494 ASSERT(agno != NULLAGNUMBER);
1495
1496 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
1497 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
1498 XFS_FSS_TO_BB(mp, 1), 0, bpp);
1499 if (error)
1500 return error;
1501
1502 ASSERT(*bpp && !XFS_BUF_GETERROR(*bpp));
1503 agi = XFS_BUF_TO_AGI(*bpp);
1504
1505 /*
1506 * Validate the magic number of the agi block.
1507 */
1508 agi_ok = be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1509 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)) &&
1510 be32_to_cpu(agi->agi_seqno) == agno;
1511 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IALLOC_READ_AGI,
1512 XFS_RANDOM_IALLOC_READ_AGI))) {
1513 XFS_CORRUPTION_ERROR("xfs_read_agi", XFS_ERRLEVEL_LOW,
1514 mp, agi);
1515 xfs_trans_brelse(tp, *bpp);
1516 return XFS_ERROR(EFSCORRUPTED);
1517 }
1518
1519 XFS_BUF_SET_VTYPE_REF(*bpp, B_FS_AGI, XFS_AGI_REF);
1520
1521 xfs_check_agi_unlinked(agi);
1522 return 0;
1523 }
1524
1525 int
1526 xfs_ialloc_read_agi(
1527 struct xfs_mount *mp, /* file system mount structure */
1528 struct xfs_trans *tp, /* transaction pointer */
1529 xfs_agnumber_t agno, /* allocation group number */
1530 struct xfs_buf **bpp) /* allocation group hdr buf */
1531 {
1532 struct xfs_agi *agi; /* allocation group header */
1533 struct xfs_perag *pag; /* per allocation group data */
1534 int error;
1535
1536 error = xfs_read_agi(mp, tp, agno, bpp);
1537 if (error)
1538 return error;
1539
1540 agi = XFS_BUF_TO_AGI(*bpp);
1541 pag = xfs_perag_get(mp, agno);
1542 if (!pag->pagi_init) {
1543 pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
1544 pag->pagi_count = be32_to_cpu(agi->agi_count);
1545 pag->pagi_init = 1;
1546 }
1547
1548 /*
1549 * It's possible for these to be out of sync if
1550 * we are in the middle of a forced shutdown.
1551 */
1552 ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) ||
1553 XFS_FORCED_SHUTDOWN(mp));
1554 xfs_perag_put(pag);
1555 return 0;
1556 }
1557
1558 /*
1559 * Read in the agi to initialise the per-ag data in the mount structure
1560 */
1561 int
1562 xfs_ialloc_pagi_init(
1563 xfs_mount_t *mp, /* file system mount structure */
1564 xfs_trans_t *tp, /* transaction pointer */
1565 xfs_agnumber_t agno) /* allocation group number */
1566 {
1567 xfs_buf_t *bp = NULL;
1568 int error;
1569
1570 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
1571 if (error)
1572 return error;
1573 if (bp)
1574 xfs_trans_brelse(tp, bp);
1575 return 0;
1576 }
Linux kernel - Deliverables
This page took 0.066281 seconds and 6 git commands to generate.