?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Luca's patch ported
[cbs-scheduler.git] / fs / xfs / xfs_ialloc.c
blobab016e5ae7be1567e6782bb17725e8b52c501ed8
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 the record equal to ino in the btree given by cur.
61 */
62 STATIC int /* error */
63 xfs_inobt_lookup_eq(
64 struct xfs_btree_cur *cur, /* btree cursor */
65 xfs_agino_t ino, /* starting inode of chunk */
66 __int32_t fcnt, /* free inode count */
67 xfs_inofree_t free, /* free inode mask */
68 int *stat) /* success/failure */
69 {
70 cur->bc_rec.i.ir_startino = ino;
71 cur->bc_rec.i.ir_freecount = fcnt;
72 cur->bc_rec.i.ir_free = free;
73 return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
74 }
75
76 /*
77 * Lookup the first record greater than or equal to ino
78 * in the btree given by cur.
79 */
80 int /* error */
81 xfs_inobt_lookup_ge(
82 struct xfs_btree_cur *cur, /* btree cursor */
83 xfs_agino_t ino, /* starting inode of chunk */
84 __int32_t fcnt, /* free inode count */
85 xfs_inofree_t free, /* free inode mask */
86 int *stat) /* success/failure */
87 {
88 cur->bc_rec.i.ir_startino = ino;
89 cur->bc_rec.i.ir_freecount = fcnt;
90 cur->bc_rec.i.ir_free = free;
91 return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
92 }
93
94 /*
95 * Lookup the first record less than or equal to ino
96 * in the btree given by cur.
97 */
98 int /* error */
99 xfs_inobt_lookup_le(
100 struct xfs_btree_cur *cur, /* btree cursor */
101 xfs_agino_t ino, /* starting inode of chunk */
102 __int32_t fcnt, /* free inode count */
103 xfs_inofree_t free, /* free inode mask */
104 int *stat) /* success/failure */
105 {
106 cur->bc_rec.i.ir_startino = ino;
107 cur->bc_rec.i.ir_freecount = fcnt;
108 cur->bc_rec.i.ir_free = free;
109 return xfs_btree_lookup(cur, XFS_LOOKUP_LE, stat);
110 }
111
112 /*
113 * Update the record referred to by cur to the value given
114 * by [ino, fcnt, free].
115 * This either works (return 0) or gets an EFSCORRUPTED error.
116 */
117 STATIC int /* error */
118 xfs_inobt_update(
119 struct xfs_btree_cur *cur, /* btree cursor */
120 xfs_agino_t ino, /* starting inode of chunk */
121 __int32_t fcnt, /* free inode count */
122 xfs_inofree_t free) /* free inode mask */
123 {
124 union xfs_btree_rec rec;
125
126 rec.inobt.ir_startino = cpu_to_be32(ino);
127 rec.inobt.ir_freecount = cpu_to_be32(fcnt);
128 rec.inobt.ir_free = cpu_to_be64(free);
129 return xfs_btree_update(cur, &rec);
130 }
131
132 /*
133 * Get the data from the pointed-to record.
134 */
135 int /* error */
136 xfs_inobt_get_rec(
137 struct xfs_btree_cur *cur, /* btree cursor */
138 xfs_agino_t *ino, /* output: starting inode of chunk */
139 __int32_t *fcnt, /* output: number of free inodes */
140 xfs_inofree_t *free, /* output: free inode mask */
141 int *stat) /* output: success/failure */
142 {
143 union xfs_btree_rec *rec;
144 int error;
145
146 error = xfs_btree_get_rec(cur, &rec, stat);
147 if (!error && *stat == 1) {
148 *ino = be32_to_cpu(rec->inobt.ir_startino);
149 *fcnt = be32_to_cpu(rec->inobt.ir_freecount);
150 *free = be64_to_cpu(rec->inobt.ir_free);
151 }
152 return error;
153 }
154
155 /*
156 * Allocate new inodes in the allocation group specified by agbp.
157 * Return 0 for success, else error code.
158 */
159 STATIC int /* error code or 0 */
160 xfs_ialloc_ag_alloc(
161 xfs_trans_t *tp, /* transaction pointer */
162 xfs_buf_t *agbp, /* alloc group buffer */
163 int *alloc)
164 {
165 xfs_agi_t *agi; /* allocation group header */
166 xfs_alloc_arg_t args; /* allocation argument structure */
167 int blks_per_cluster; /* fs blocks per inode cluster */
168 xfs_btree_cur_t *cur; /* inode btree cursor */
169 xfs_daddr_t d; /* disk addr of buffer */
170 xfs_agnumber_t agno;
171 int error;
172 xfs_buf_t *fbuf; /* new free inodes' buffer */
173 xfs_dinode_t *free; /* new free inode structure */
174 int i; /* inode counter */
175 int j; /* block counter */
176 int nbufs; /* num bufs of new inodes */
177 xfs_agino_t newino; /* new first inode's number */
178 xfs_agino_t newlen; /* new number of inodes */
179 int ninodes; /* num inodes per buf */
180 xfs_agino_t thisino; /* current inode number, for loop */
181 int version; /* inode version number to use */
182 int isaligned = 0; /* inode allocation at stripe unit */
183 /* boundary */
184 unsigned int gen;
185
186 args.tp = tp;
187 args.mp = tp->t_mountp;
188
189 /*
190 * Locking will ensure that we don't have two callers in here
191 * at one time.
192 */
193 newlen = XFS_IALLOC_INODES(args.mp);
194 if (args.mp->m_maxicount &&
195 args.mp->m_sb.sb_icount + newlen > args.mp->m_maxicount)
196 return XFS_ERROR(ENOSPC);
197 args.minlen = args.maxlen = XFS_IALLOC_BLOCKS(args.mp);
198 /*
199 * First try to allocate inodes contiguous with the last-allocated
200 * chunk of inodes. If the filesystem is striped, this will fill
201 * an entire stripe unit with inodes.
202 */
203 agi = XFS_BUF_TO_AGI(agbp);
204 newino = be32_to_cpu(agi->agi_newino);
205 args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) +
206 XFS_IALLOC_BLOCKS(args.mp);
207 if (likely(newino != NULLAGINO &&
208 (args.agbno < be32_to_cpu(agi->agi_length)))) {
209 args.fsbno = XFS_AGB_TO_FSB(args.mp,
210 be32_to_cpu(agi->agi_seqno), args.agbno);
211 args.type = XFS_ALLOCTYPE_THIS_BNO;
212 args.mod = args.total = args.wasdel = args.isfl =
213 args.userdata = args.minalignslop = 0;
214 args.prod = 1;
215
216 /*
217 * We need to take into account alignment here to ensure that
218 * we don't modify the free list if we fail to have an exact
219 * block. If we don't have an exact match, and every oher
220 * attempt allocation attempt fails, we'll end up cancelling
221 * a dirty transaction and shutting down.
222 *
223 * For an exact allocation, alignment must be 1,
224 * however we need to take cluster alignment into account when
225 * fixing up the freelist. Use the minalignslop field to
226 * indicate that extra blocks might be required for alignment,
227 * but not to use them in the actual exact allocation.
228 */
229 args.alignment = 1;
230 args.minalignslop = xfs_ialloc_cluster_alignment(&args) - 1;
231
232 /* Allow space for the inode btree to split. */
233 args.minleft = XFS_IN_MAXLEVELS(args.mp) - 1;
234 if ((error = xfs_alloc_vextent(&args)))
235 return error;
236 } else
237 args.fsbno = NULLFSBLOCK;
238
239 if (unlikely(args.fsbno == NULLFSBLOCK)) {
240 /*
241 * Set the alignment for the allocation.
242 * If stripe alignment is turned on then align at stripe unit
243 * boundary.
244 * If the cluster size is smaller than a filesystem block
245 * then we're doing I/O for inodes in filesystem block size
246 * pieces, so don't need alignment anyway.
247 */
248 isaligned = 0;
249 if (args.mp->m_sinoalign) {
250 ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN));
251 args.alignment = args.mp->m_dalign;
252 isaligned = 1;
253 } else
254 args.alignment = xfs_ialloc_cluster_alignment(&args);
255 /*
256 * Need to figure out where to allocate the inode blocks.
257 * Ideally they should be spaced out through the a.g.
258 * For now, just allocate blocks up front.
259 */
260 args.agbno = be32_to_cpu(agi->agi_root);
261 args.fsbno = XFS_AGB_TO_FSB(args.mp,
262 be32_to_cpu(agi->agi_seqno), args.agbno);
263 /*
264 * Allocate a fixed-size extent of inodes.
265 */
266 args.type = XFS_ALLOCTYPE_NEAR_BNO;
267 args.mod = args.total = args.wasdel = args.isfl =
268 args.userdata = args.minalignslop = 0;
269 args.prod = 1;
270 /*
271 * Allow space for the inode btree to split.
272 */
273 args.minleft = XFS_IN_MAXLEVELS(args.mp) - 1;
274 if ((error = xfs_alloc_vextent(&args)))
275 return error;
276 }
277
278 /*
279 * If stripe alignment is turned on, then try again with cluster
280 * alignment.
281 */
282 if (isaligned && args.fsbno == NULLFSBLOCK) {
283 args.type = XFS_ALLOCTYPE_NEAR_BNO;
284 args.agbno = be32_to_cpu(agi->agi_root);
285 args.fsbno = XFS_AGB_TO_FSB(args.mp,
286 be32_to_cpu(agi->agi_seqno), args.agbno);
287 args.alignment = xfs_ialloc_cluster_alignment(&args);
288 if ((error = xfs_alloc_vextent(&args)))
289 return error;
290 }
291
292 if (args.fsbno == NULLFSBLOCK) {
293 *alloc = 0;
294 return 0;
295 }
296 ASSERT(args.len == args.minlen);
297 /*
298 * Convert the results.
299 */
300 newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0);
301 /*
302 * Loop over the new block(s), filling in the inodes.
303 * For small block sizes, manipulate the inodes in buffers
304 * which are multiples of the blocks size.
305 */
306 if (args.mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(args.mp)) {
307 blks_per_cluster = 1;
308 nbufs = (int)args.len;
309 ninodes = args.mp->m_sb.sb_inopblock;
310 } else {
311 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(args.mp) /
312 args.mp->m_sb.sb_blocksize;
313 nbufs = (int)args.len / blks_per_cluster;
314 ninodes = blks_per_cluster * args.mp->m_sb.sb_inopblock;
315 }
316 /*
317 * Figure out what version number to use in the inodes we create.
318 * If the superblock version has caught up to the one that supports
319 * the new inode format, then use the new inode version. Otherwise
320 * use the old version so that old kernels will continue to be
321 * able to use the file system.
322 */
323 if (xfs_sb_version_hasnlink(&args.mp->m_sb))
324 version = 2;
325 else
326 version = 1;
327
328 /*
329 * Seed the new inode cluster with a random generation number. This
330 * prevents short-term reuse of generation numbers if a chunk is
331 * freed and then immediately reallocated. We use random numbers
332 * rather than a linear progression to prevent the next generation
333 * number from being easily guessable.
334 */
335 gen = random32();
336 for (j = 0; j < nbufs; j++) {
337 /*
338 * Get the block.
339 */
340 d = XFS_AGB_TO_DADDR(args.mp, be32_to_cpu(agi->agi_seqno),
341 args.agbno + (j * blks_per_cluster));
342 fbuf = xfs_trans_get_buf(tp, args.mp->m_ddev_targp, d,
343 args.mp->m_bsize * blks_per_cluster,
344 XFS_BUF_LOCK);
345 ASSERT(fbuf);
346 ASSERT(!XFS_BUF_GETERROR(fbuf));
347
348 /*
349 * Initialize all inodes in this buffer and then log them.
350 *
351 * XXX: It would be much better if we had just one transaction to
352 * log a whole cluster of inodes instead of all the indivdual
353 * transactions causing a lot of log traffic.
354 */
355 xfs_biozero(fbuf, 0, ninodes << args.mp->m_sb.sb_inodelog);
356 for (i = 0; i < ninodes; i++) {
357 int ioffset = i << args.mp->m_sb.sb_inodelog;
358 uint isize = sizeof(struct xfs_dinode);
359
360 free = xfs_make_iptr(args.mp, fbuf, i);
361 free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
362 free->di_version = version;
363 free->di_gen = cpu_to_be32(gen);
364 free->di_next_unlinked = cpu_to_be32(NULLAGINO);
365 xfs_trans_log_buf(tp, fbuf, ioffset, ioffset + isize - 1);
366 }
367 xfs_trans_inode_alloc_buf(tp, fbuf);
368 }
369 be32_add_cpu(&agi->agi_count, newlen);
370 be32_add_cpu(&agi->agi_freecount, newlen);
371 agno = be32_to_cpu(agi->agi_seqno);
372 down_read(&args.mp->m_peraglock);
373 args.mp->m_perag[agno].pagi_freecount += newlen;
374 up_read(&args.mp->m_peraglock);
375 agi->agi_newino = cpu_to_be32(newino);
376 /*
377 * Insert records describing the new inode chunk into the btree.
378 */
379 cur = xfs_inobt_init_cursor(args.mp, tp, agbp, agno);
380 for (thisino = newino;
381 thisino < newino + newlen;
382 thisino += XFS_INODES_PER_CHUNK) {
383 if ((error = xfs_inobt_lookup_eq(cur, thisino,
384 XFS_INODES_PER_CHUNK, XFS_INOBT_ALL_FREE, &i))) {
385 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
386 return error;
387 }
388 ASSERT(i == 0);
389 if ((error = xfs_btree_insert(cur, &i))) {
390 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
391 return error;
392 }
393 ASSERT(i == 1);
394 }
395 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
396 /*
397 * Log allocation group header fields
398 */
399 xfs_ialloc_log_agi(tp, agbp,
400 XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO);
401 /*
402 * Modify/log superblock values for inode count and inode free count.
403 */
404 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen);
405 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen);
406 *alloc = 1;
407 return 0;
408 }
409
410 STATIC_INLINE xfs_agnumber_t
411 xfs_ialloc_next_ag(
412 xfs_mount_t *mp)
413 {
414 xfs_agnumber_t agno;
415
416 spin_lock(&mp->m_agirotor_lock);
417 agno = mp->m_agirotor;
418 if (++mp->m_agirotor == mp->m_maxagi)
419 mp->m_agirotor = 0;
420 spin_unlock(&mp->m_agirotor_lock);
421
422 return agno;
423 }
424
425 /*
426 * Select an allocation group to look for a free inode in, based on the parent
427 * inode and then mode. Return the allocation group buffer.
428 */
429 STATIC xfs_buf_t * /* allocation group buffer */
430 xfs_ialloc_ag_select(
431 xfs_trans_t *tp, /* transaction pointer */
432 xfs_ino_t parent, /* parent directory inode number */
433 mode_t mode, /* bits set to indicate file type */
434 int okalloc) /* ok to allocate more space */
435 {
436 xfs_buf_t *agbp; /* allocation group header buffer */
437 xfs_agnumber_t agcount; /* number of ag's in the filesystem */
438 xfs_agnumber_t agno; /* current ag number */
439 int flags; /* alloc buffer locking flags */
440 xfs_extlen_t ineed; /* blocks needed for inode allocation */
441 xfs_extlen_t longest = 0; /* longest extent available */
442 xfs_mount_t *mp; /* mount point structure */
443 int needspace; /* file mode implies space allocated */
444 xfs_perag_t *pag; /* per allocation group data */
445 xfs_agnumber_t pagno; /* parent (starting) ag number */
446
447 /*
448 * Files of these types need at least one block if length > 0
449 * (and they won't fit in the inode, but that's hard to figure out).
450 */
451 needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode);
452 mp = tp->t_mountp;
453 agcount = mp->m_maxagi;
454 if (S_ISDIR(mode))
455 pagno = xfs_ialloc_next_ag(mp);
456 else {
457 pagno = XFS_INO_TO_AGNO(mp, parent);
458 if (pagno >= agcount)
459 pagno = 0;
460 }
461 ASSERT(pagno < agcount);
462 /*
463 * Loop through allocation groups, looking for one with a little
464 * free space in it. Note we don't look for free inodes, exactly.
465 * Instead, we include whether there is a need to allocate inodes
466 * to mean that blocks must be allocated for them,
467 * if none are currently free.
468 */
469 agno = pagno;
470 flags = XFS_ALLOC_FLAG_TRYLOCK;
471 down_read(&mp->m_peraglock);
472 for (;;) {
473 pag = &mp->m_perag[agno];
474 if (!pag->pagi_init) {
475 if (xfs_ialloc_read_agi(mp, tp, agno, &agbp)) {
476 agbp = NULL;
477 goto nextag;
478 }
479 } else
480 agbp = NULL;
481
482 if (!pag->pagi_inodeok) {
483 xfs_ialloc_next_ag(mp);
484 goto unlock_nextag;
485 }
486
487 /*
488 * Is there enough free space for the file plus a block
489 * of inodes (if we need to allocate some)?
490 */
491 ineed = pag->pagi_freecount ? 0 : XFS_IALLOC_BLOCKS(mp);
492 if (ineed && !pag->pagf_init) {
493 if (agbp == NULL &&
494 xfs_ialloc_read_agi(mp, tp, agno, &agbp)) {
495 agbp = NULL;
496 goto nextag;
497 }
498 (void)xfs_alloc_pagf_init(mp, tp, agno, flags);
499 }
500 if (!ineed || pag->pagf_init) {
501 if (ineed && !(longest = pag->pagf_longest))
502 longest = pag->pagf_flcount > 0;
503 if (!ineed ||
504 (pag->pagf_freeblks >= needspace + ineed &&
505 longest >= ineed &&
506 okalloc)) {
507 if (agbp == NULL &&
508 xfs_ialloc_read_agi(mp, tp, agno, &agbp)) {
509 agbp = NULL;
510 goto nextag;
511 }
512 up_read(&mp->m_peraglock);
513 return agbp;
514 }
515 }
516 unlock_nextag:
517 if (agbp)
518 xfs_trans_brelse(tp, agbp);
519 nextag:
520 /*
521 * No point in iterating over the rest, if we're shutting
522 * down.
523 */
524 if (XFS_FORCED_SHUTDOWN(mp)) {
525 up_read(&mp->m_peraglock);
526 return NULL;
527 }
528 agno++;
529 if (agno >= agcount)
530 agno = 0;
531 if (agno == pagno) {
532 if (flags == 0) {
533 up_read(&mp->m_peraglock);
534 return NULL;
535 }
536 flags = 0;
537 }
538 }
539 }
540
541 /*
542 * Visible inode allocation functions.
543 */
544
545 /*
546 * Allocate an inode on disk.
547 * Mode is used to tell whether the new inode will need space, and whether
548 * it is a directory.
549 *
550 * The arguments IO_agbp and alloc_done are defined to work within
551 * the constraint of one allocation per transaction.
552 * xfs_dialloc() is designed to be called twice if it has to do an
553 * allocation to make more free inodes. On the first call,
554 * IO_agbp should be set to NULL. If an inode is available,
555 * i.e., xfs_dialloc() did not need to do an allocation, an inode
556 * number is returned. In this case, IO_agbp would be set to the
557 * current ag_buf and alloc_done set to false.
558 * If an allocation needed to be done, xfs_dialloc would return
559 * the current ag_buf in IO_agbp and set alloc_done to true.
560 * The caller should then commit the current transaction, allocate a new
561 * transaction, and call xfs_dialloc() again, passing in the previous
562 * value of IO_agbp. IO_agbp should be held across the transactions.
563 * Since the agbp is locked across the two calls, the second call is
564 * guaranteed to have a free inode available.
565 *
566 * Once we successfully pick an inode its number is returned and the
567 * on-disk data structures are updated. The inode itself is not read
568 * in, since doing so would break ordering constraints with xfs_reclaim.
569 */
570 int
571 xfs_dialloc(
572 xfs_trans_t *tp, /* transaction pointer */
573 xfs_ino_t parent, /* parent inode (directory) */
574 mode_t mode, /* mode bits for new inode */
575 int okalloc, /* ok to allocate more space */
576 xfs_buf_t **IO_agbp, /* in/out ag header's buffer */
577 boolean_t *alloc_done, /* true if we needed to replenish
578 inode freelist */
579 xfs_ino_t *inop) /* inode number allocated */
580 {
581 xfs_agnumber_t agcount; /* number of allocation groups */
582 xfs_buf_t *agbp; /* allocation group header's buffer */
583 xfs_agnumber_t agno; /* allocation group number */
584 xfs_agi_t *agi; /* allocation group header structure */
585 xfs_btree_cur_t *cur; /* inode allocation btree cursor */
586 int error; /* error return value */
587 int i; /* result code */
588 int ialloced; /* inode allocation status */
589 int noroom = 0; /* no space for inode blk allocation */
590 xfs_ino_t ino; /* fs-relative inode to be returned */
591 /* REFERENCED */
592 int j; /* result code */
593 xfs_mount_t *mp; /* file system mount structure */
594 int offset; /* index of inode in chunk */
595 xfs_agino_t pagino; /* parent's a.g. relative inode # */
596 xfs_agnumber_t pagno; /* parent's allocation group number */
597 xfs_inobt_rec_incore_t rec; /* inode allocation record */
598 xfs_agnumber_t tagno; /* testing allocation group number */
599 xfs_btree_cur_t *tcur; /* temp cursor */
600 xfs_inobt_rec_incore_t trec; /* temp inode allocation record */
601
602
603 if (*IO_agbp == NULL) {
604 /*
605 * We do not have an agbp, so select an initial allocation
606 * group for inode allocation.
607 */
608 agbp = xfs_ialloc_ag_select(tp, parent, mode, okalloc);
609 /*
610 * Couldn't find an allocation group satisfying the
611 * criteria, give up.
612 */
613 if (!agbp) {
614 *inop = NULLFSINO;
615 return 0;
616 }
617 agi = XFS_BUF_TO_AGI(agbp);
618 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
619 } else {
620 /*
621 * Continue where we left off before. In this case, we
622 * know that the allocation group has free inodes.
623 */
624 agbp = *IO_agbp;
625 agi = XFS_BUF_TO_AGI(agbp);
626 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
627 ASSERT(be32_to_cpu(agi->agi_freecount) > 0);
628 }
629 mp = tp->t_mountp;
630 agcount = mp->m_sb.sb_agcount;
631 agno = be32_to_cpu(agi->agi_seqno);
632 tagno = agno;
633 pagno = XFS_INO_TO_AGNO(mp, parent);
634 pagino = XFS_INO_TO_AGINO(mp, parent);
635
636 /*
637 * If we have already hit the ceiling of inode blocks then clear
638 * okalloc so we scan all available agi structures for a free
639 * inode.
640 */
641
642 if (mp->m_maxicount &&
643 mp->m_sb.sb_icount + XFS_IALLOC_INODES(mp) > mp->m_maxicount) {
644 noroom = 1;
645 okalloc = 0;
646 }
647
648 /*
649 * Loop until we find an allocation group that either has free inodes
650 * or in which we can allocate some inodes. Iterate through the
651 * allocation groups upward, wrapping at the end.
652 */
653 *alloc_done = B_FALSE;
654 while (!agi->agi_freecount) {
655 /*
656 * Don't do anything if we're not supposed to allocate
657 * any blocks, just go on to the next ag.
658 */
659 if (okalloc) {
660 /*
661 * Try to allocate some new inodes in the allocation
662 * group.
663 */
664 if ((error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced))) {
665 xfs_trans_brelse(tp, agbp);
666 if (error == ENOSPC) {
667 *inop = NULLFSINO;
668 return 0;
669 } else
670 return error;
671 }
672 if (ialloced) {
673 /*
674 * We successfully allocated some inodes, return
675 * the current context to the caller so that it
676 * can commit the current transaction and call
677 * us again where we left off.
678 */
679 ASSERT(be32_to_cpu(agi->agi_freecount) > 0);
680 *alloc_done = B_TRUE;
681 *IO_agbp = agbp;
682 *inop = NULLFSINO;
683 return 0;
684 }
685 }
686 /*
687 * If it failed, give up on this ag.
688 */
689 xfs_trans_brelse(tp, agbp);
690 /*
691 * Go on to the next ag: get its ag header.
692 */
693 nextag:
694 if (++tagno == agcount)
695 tagno = 0;
696 if (tagno == agno) {
697 *inop = NULLFSINO;
698 return noroom ? ENOSPC : 0;
699 }
700 down_read(&mp->m_peraglock);
701 if (mp->m_perag[tagno].pagi_inodeok == 0) {
702 up_read(&mp->m_peraglock);
703 goto nextag;
704 }
705 error = xfs_ialloc_read_agi(mp, tp, tagno, &agbp);
706 up_read(&mp->m_peraglock);
707 if (error)
708 goto nextag;
709 agi = XFS_BUF_TO_AGI(agbp);
710 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
711 }
712 /*
713 * Here with an allocation group that has a free inode.
714 * Reset agno since we may have chosen a new ag in the
715 * loop above.
716 */
717 agno = tagno;
718 *IO_agbp = NULL;
719 cur = xfs_inobt_init_cursor(mp, tp, agbp, be32_to_cpu(agi->agi_seqno));
720 /*
721 * If pagino is 0 (this is the root inode allocation) use newino.
722 * This must work because we've just allocated some.
723 */
724 if (!pagino)
725 pagino = be32_to_cpu(agi->agi_newino);
726 #ifdef DEBUG
727 if (cur->bc_nlevels == 1) {
728 int freecount = 0;
729
730 if ((error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &i)))
731 goto error0;
732 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
733 do {
734 if ((error = xfs_inobt_get_rec(cur, &rec.ir_startino,
735 &rec.ir_freecount, &rec.ir_free, &i)))
736 goto error0;
737 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
738 freecount += rec.ir_freecount;
739 if ((error = xfs_btree_increment(cur, 0, &i)))
740 goto error0;
741 } while (i == 1);
742
743 ASSERT(freecount == be32_to_cpu(agi->agi_freecount) ||
744 XFS_FORCED_SHUTDOWN(mp));
745 }
746 #endif
747 /*
748 * If in the same a.g. as the parent, try to get near the parent.
749 */
750 if (pagno == agno) {
751 if ((error = xfs_inobt_lookup_le(cur, pagino, 0, 0, &i)))
752 goto error0;
753 if (i != 0 &&
754 (error = xfs_inobt_get_rec(cur, &rec.ir_startino,
755 &rec.ir_freecount, &rec.ir_free, &j)) == 0 &&
756 j == 1 &&
757 rec.ir_freecount > 0) {
758 /*
759 * Found a free inode in the same chunk
760 * as parent, done.
761 */
762 }
763 /*
764 * In the same a.g. as parent, but parent's chunk is full.
765 */
766 else {
767 int doneleft; /* done, to the left */
768 int doneright; /* done, to the right */
769
770 if (error)
771 goto error0;
772 ASSERT(i == 1);
773 ASSERT(j == 1);
774 /*
775 * Duplicate the cursor, search left & right
776 * simultaneously.
777 */
778 if ((error = xfs_btree_dup_cursor(cur, &tcur)))
779 goto error0;
780 /*
781 * Search left with tcur, back up 1 record.
782 */
783 if ((error = xfs_btree_decrement(tcur, 0, &i)))
784 goto error1;
785 doneleft = !i;
786 if (!doneleft) {
787 if ((error = xfs_inobt_get_rec(tcur,
788 &trec.ir_startino,
789 &trec.ir_freecount,
790 &trec.ir_free, &i)))
791 goto error1;
792 XFS_WANT_CORRUPTED_GOTO(i == 1, error1);
793 }
794 /*
795 * Search right with cur, go forward 1 record.
796 */
797 if ((error = xfs_btree_increment(cur, 0, &i)))
798 goto error1;
799 doneright = !i;
800 if (!doneright) {
801 if ((error = xfs_inobt_get_rec(cur,
802 &rec.ir_startino,
803 &rec.ir_freecount,
804 &rec.ir_free, &i)))
805 goto error1;
806 XFS_WANT_CORRUPTED_GOTO(i == 1, error1);
807 }
808 /*
809 * Loop until we find the closest inode chunk
810 * with a free one.
811 */
812 while (!doneleft || !doneright) {
813 int useleft; /* using left inode
814 chunk this time */
815
816 /*
817 * Figure out which block is closer,
818 * if both are valid.
819 */
820 if (!doneleft && !doneright)
821 useleft =
822 pagino -
823 (trec.ir_startino +
824 XFS_INODES_PER_CHUNK - 1) <
825 rec.ir_startino - pagino;
826 else
827 useleft = !doneleft;
828 /*
829 * If checking the left, does it have
830 * free inodes?
831 */
832 if (useleft && trec.ir_freecount) {
833 /*
834 * Yes, set it up as the chunk to use.
835 */
836 rec = trec;
837 xfs_btree_del_cursor(cur,
838 XFS_BTREE_NOERROR);
839 cur = tcur;
840 break;
841 }
842 /*
843 * If checking the right, does it have
844 * free inodes?
845 */
846 if (!useleft && rec.ir_freecount) {
847 /*
848 * Yes, it's already set up.
849 */
850 xfs_btree_del_cursor(tcur,
851 XFS_BTREE_NOERROR);
852 break;
853 }
854 /*
855 * If used the left, get another one
856 * further left.
857 */
858 if (useleft) {
859 if ((error = xfs_btree_decrement(tcur, 0,
860 &i)))
861 goto error1;
862 doneleft = !i;
863 if (!doneleft) {
864 if ((error = xfs_inobt_get_rec(
865 tcur,
866 &trec.ir_startino,
867 &trec.ir_freecount,
868 &trec.ir_free, &i)))
869 goto error1;
870 XFS_WANT_CORRUPTED_GOTO(i == 1,
871 error1);
872 }
873 }
874 /*
875 * If used the right, get another one
876 * further right.
877 */
878 else {
879 if ((error = xfs_btree_increment(cur, 0,
880 &i)))
881 goto error1;
882 doneright = !i;
883 if (!doneright) {
884 if ((error = xfs_inobt_get_rec(
885 cur,
886 &rec.ir_startino,
887 &rec.ir_freecount,
888 &rec.ir_free, &i)))
889 goto error1;
890 XFS_WANT_CORRUPTED_GOTO(i == 1,
891 error1);
892 }
893 }
894 }
895 ASSERT(!doneleft || !doneright);
896 }
897 }
898 /*
899 * In a different a.g. from the parent.
900 * See if the most recently allocated block has any free.
901 */
902 else if (be32_to_cpu(agi->agi_newino) != NULLAGINO) {
903 if ((error = xfs_inobt_lookup_eq(cur,
904 be32_to_cpu(agi->agi_newino), 0, 0, &i)))
905 goto error0;
906 if (i == 1 &&
907 (error = xfs_inobt_get_rec(cur, &rec.ir_startino,
908 &rec.ir_freecount, &rec.ir_free, &j)) == 0 &&
909 j == 1 &&
910 rec.ir_freecount > 0) {
911 /*
912 * The last chunk allocated in the group still has
913 * a free inode.
914 */
915 }
916 /*
917 * None left in the last group, search the whole a.g.
918 */
919 else {
920 if (error)
921 goto error0;
922 if ((error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &i)))
923 goto error0;
924 ASSERT(i == 1);
925 for (;;) {
926 if ((error = xfs_inobt_get_rec(cur,
927 &rec.ir_startino,
928 &rec.ir_freecount, &rec.ir_free,
929 &i)))
930 goto error0;
931 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
932 if (rec.ir_freecount > 0)
933 break;
934 if ((error = xfs_btree_increment(cur, 0, &i)))
935 goto error0;
936 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
937 }
938 }
939 }
940 offset = xfs_ialloc_find_free(&rec.ir_free);
941 ASSERT(offset >= 0);
942 ASSERT(offset < XFS_INODES_PER_CHUNK);
943 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
944 XFS_INODES_PER_CHUNK) == 0);
945 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
946 XFS_INOBT_CLR_FREE(&rec, offset);
947 rec.ir_freecount--;
948 if ((error = xfs_inobt_update(cur, rec.ir_startino, rec.ir_freecount,
949 rec.ir_free)))
950 goto error0;
951 be32_add_cpu(&agi->agi_freecount, -1);
952 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
953 down_read(&mp->m_peraglock);
954 mp->m_perag[tagno].pagi_freecount--;
955 up_read(&mp->m_peraglock);
956 #ifdef DEBUG
957 if (cur->bc_nlevels == 1) {
958 int freecount = 0;
959
960 if ((error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &i)))
961 goto error0;
962 do {
963 if ((error = xfs_inobt_get_rec(cur, &rec.ir_startino,
964 &rec.ir_freecount, &rec.ir_free, &i)))
965 goto error0;
966 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
967 freecount += rec.ir_freecount;
968 if ((error = xfs_btree_increment(cur, 0, &i)))
969 goto error0;
970 } while (i == 1);
971 ASSERT(freecount == be32_to_cpu(agi->agi_freecount) ||
972 XFS_FORCED_SHUTDOWN(mp));
973 }
974 #endif
975 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
976 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
977 *inop = ino;
978 return 0;
979 error1:
980 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
981 error0:
982 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
983 return error;
984 }
985
986 /*
987 * Free disk inode. Carefully avoids touching the incore inode, all
988 * manipulations incore are the caller's responsibility.
989 * The on-disk inode is not changed by this operation, only the
990 * btree (free inode mask) is changed.
991 */
992 int
993 xfs_difree(
994 xfs_trans_t *tp, /* transaction pointer */
995 xfs_ino_t inode, /* inode to be freed */
996 xfs_bmap_free_t *flist, /* extents to free */
997 int *delete, /* set if inode cluster was deleted */
998 xfs_ino_t *first_ino) /* first inode in deleted cluster */
999 {
1000 /* REFERENCED */
1001 xfs_agblock_t agbno; /* block number containing inode */
1002 xfs_buf_t *agbp; /* buffer containing allocation group header */
1003 xfs_agino_t agino; /* inode number relative to allocation group */
1004 xfs_agnumber_t agno; /* allocation group number */
1005 xfs_agi_t *agi; /* allocation group header */
1006 xfs_btree_cur_t *cur; /* inode btree cursor */
1007 int error; /* error return value */
1008 int i; /* result code */
1009 int ilen; /* inodes in an inode cluster */
1010 xfs_mount_t *mp; /* mount structure for filesystem */
1011 int off; /* offset of inode in inode chunk */
1012 xfs_inobt_rec_incore_t rec; /* btree record */
1013
1014 mp = tp->t_mountp;
1015
1016 /*
1017 * Break up inode number into its components.
1018 */
1019 agno = XFS_INO_TO_AGNO(mp, inode);
1020 if (agno >= mp->m_sb.sb_agcount) {
1021 cmn_err(CE_WARN,
1022 "xfs_difree: agno >= mp->m_sb.sb_agcount (%d >= %d) on %s. Returning EINVAL.",
1023 agno, mp->m_sb.sb_agcount, mp->m_fsname);
1024 ASSERT(0);
1025 return XFS_ERROR(EINVAL);
1026 }
1027 agino = XFS_INO_TO_AGINO(mp, inode);
1028 if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) {
1029 cmn_err(CE_WARN,
1030 "xfs_difree: inode != XFS_AGINO_TO_INO() "
1031 "(%llu != %llu) on %s. Returning EINVAL.",
1032 (unsigned long long)inode,
1033 (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino),
1034 mp->m_fsname);
1035 ASSERT(0);
1036 return XFS_ERROR(EINVAL);
1037 }
1038 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1039 if (agbno >= mp->m_sb.sb_agblocks) {
1040 cmn_err(CE_WARN,
1041 "xfs_difree: agbno >= mp->m_sb.sb_agblocks (%d >= %d) on %s. Returning EINVAL.",
1042 agbno, mp->m_sb.sb_agblocks, mp->m_fsname);
1043 ASSERT(0);
1044 return XFS_ERROR(EINVAL);
1045 }
1046 /*
1047 * Get the allocation group header.
1048 */
1049 down_read(&mp->m_peraglock);
1050 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1051 up_read(&mp->m_peraglock);
1052 if (error) {
1053 cmn_err(CE_WARN,
1054 "xfs_difree: xfs_ialloc_read_agi() returned an error %d on %s. Returning error.",
1055 error, mp->m_fsname);
1056 return error;
1057 }
1058 agi = XFS_BUF_TO_AGI(agbp);
1059 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
1060 ASSERT(agbno < be32_to_cpu(agi->agi_length));
1061 /*
1062 * Initialize the cursor.
1063 */
1064 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
1065 #ifdef DEBUG
1066 if (cur->bc_nlevels == 1) {
1067 int freecount = 0;
1068
1069 if ((error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &i)))
1070 goto error0;
1071 do {
1072 if ((error = xfs_inobt_get_rec(cur, &rec.ir_startino,
1073 &rec.ir_freecount, &rec.ir_free, &i)))
1074 goto error0;
1075 if (i) {
1076 freecount += rec.ir_freecount;
1077 if ((error = xfs_btree_increment(cur, 0, &i)))
1078 goto error0;
1079 }
1080 } while (i == 1);
1081 ASSERT(freecount == be32_to_cpu(agi->agi_freecount) ||
1082 XFS_FORCED_SHUTDOWN(mp));
1083 }
1084 #endif
1085 /*
1086 * Look for the entry describing this inode.
1087 */
1088 if ((error = xfs_inobt_lookup_le(cur, agino, 0, 0, &i))) {
1089 cmn_err(CE_WARN,
1090 "xfs_difree: xfs_inobt_lookup_le returned() an error %d on %s. Returning error.",
1091 error, mp->m_fsname);
1092 goto error0;
1093 }
1094 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1095 if ((error = xfs_inobt_get_rec(cur, &rec.ir_startino, &rec.ir_freecount,
1096 &rec.ir_free, &i))) {
1097 cmn_err(CE_WARN,
1098 "xfs_difree: xfs_inobt_get_rec() returned an error %d on %s. Returning error.",
1099 error, mp->m_fsname);
1100 goto error0;
1101 }
1102 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1103 /*
1104 * Get the offset in the inode chunk.
1105 */
1106 off = agino - rec.ir_startino;
1107 ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK);
1108 ASSERT(!XFS_INOBT_IS_FREE(&rec, off));
1109 /*
1110 * Mark the inode free & increment the count.
1111 */
1112 XFS_INOBT_SET_FREE(&rec, off);
1113 rec.ir_freecount++;
1114
1115 /*
1116 * When an inode cluster is free, it becomes eligible for removal
1117 */
1118 if (!(mp->m_flags & XFS_MOUNT_IKEEP) &&
1119 (rec.ir_freecount == XFS_IALLOC_INODES(mp))) {
1120
1121 *delete = 1;
1122 *first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino);
1123
1124 /*
1125 * Remove the inode cluster from the AGI B+Tree, adjust the
1126 * AGI and Superblock inode counts, and mark the disk space
1127 * to be freed when the transaction is committed.
1128 */
1129 ilen = XFS_IALLOC_INODES(mp);
1130 be32_add_cpu(&agi->agi_count, -ilen);
1131 be32_add_cpu(&agi->agi_freecount, -(ilen - 1));
1132 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
1133 down_read(&mp->m_peraglock);
1134 mp->m_perag[agno].pagi_freecount -= ilen - 1;
1135 up_read(&mp->m_peraglock);
1136 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen);
1137 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1));
1138
1139 if ((error = xfs_btree_delete(cur, &i))) {
1140 cmn_err(CE_WARN, "xfs_difree: xfs_btree_delete returned an error %d on %s.\n",
1141 error, mp->m_fsname);
1142 goto error0;
1143 }
1144
1145 xfs_bmap_add_free(XFS_AGB_TO_FSB(mp,
1146 agno, XFS_INO_TO_AGBNO(mp,rec.ir_startino)),
1147 XFS_IALLOC_BLOCKS(mp), flist, mp);
1148 } else {
1149 *delete = 0;
1150
1151 if ((error = xfs_inobt_update(cur, rec.ir_startino, rec.ir_freecount, rec.ir_free))) {
1152 cmn_err(CE_WARN,
1153 "xfs_difree: xfs_inobt_update() returned an error %d on %s. Returning error.",
1154 error, mp->m_fsname);
1155 goto error0;
1156 }
1157 /*
1158 * Change the inode free counts and log the ag/sb changes.
1159 */
1160 be32_add_cpu(&agi->agi_freecount, 1);
1161 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1162 down_read(&mp->m_peraglock);
1163 mp->m_perag[agno].pagi_freecount++;
1164 up_read(&mp->m_peraglock);
1165 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
1166 }
1167
1168 #ifdef DEBUG
1169 if (cur->bc_nlevels == 1) {
1170 int freecount = 0;
1171
1172 if ((error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &i)))
1173 goto error0;
1174 do {
1175 if ((error = xfs_inobt_get_rec(cur,
1176 &rec.ir_startino,
1177 &rec.ir_freecount,
1178 &rec.ir_free, &i)))
1179 goto error0;
1180 if (i) {
1181 freecount += rec.ir_freecount;
1182 if ((error = xfs_btree_increment(cur, 0, &i)))
1183 goto error0;
1184 }
1185 } while (i == 1);
1186 ASSERT(freecount == be32_to_cpu(agi->agi_freecount) ||
1187 XFS_FORCED_SHUTDOWN(mp));
1188 }
1189 #endif
1190 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1191 return 0;
1192
1193 error0:
1194 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1195 return error;
1196 }
1197
1198 /*
1199 * Return the location of the inode in imap, for mapping it into a buffer.
1200 */
1201 int
1202 xfs_imap(
1203 xfs_mount_t *mp, /* file system mount structure */
1204 xfs_trans_t *tp, /* transaction pointer */
1205 xfs_ino_t ino, /* inode to locate */
1206 struct xfs_imap *imap, /* location map structure */
1207 uint flags) /* flags for inode btree lookup */
1208 {
1209 xfs_agblock_t agbno; /* block number of inode in the alloc group */
1210 xfs_agino_t agino; /* inode number within alloc group */
1211 xfs_agnumber_t agno; /* allocation group number */
1212 int blks_per_cluster; /* num blocks per inode cluster */
1213 xfs_agblock_t chunk_agbno; /* first block in inode chunk */
1214 xfs_agblock_t cluster_agbno; /* first block in inode cluster */
1215 int error; /* error code */
1216 int offset; /* index of inode in its buffer */
1217 int offset_agbno; /* blks from chunk start to inode */
1218
1219 ASSERT(ino != NULLFSINO);
1220
1221 /*
1222 * Split up the inode number into its parts.
1223 */
1224 agno = XFS_INO_TO_AGNO(mp, ino);
1225 agino = XFS_INO_TO_AGINO(mp, ino);
1226 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1227 if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks ||
1228 ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1229 #ifdef DEBUG
1230 /* no diagnostics for bulkstat, ino comes from userspace */
1231 if (flags & XFS_IGET_BULKSTAT)
1232 return XFS_ERROR(EINVAL);
1233 if (agno >= mp->m_sb.sb_agcount) {
1234 xfs_fs_cmn_err(CE_ALERT, mp,
1235 "xfs_imap: agno (%d) >= "
1236 "mp->m_sb.sb_agcount (%d)",
1237 agno, mp->m_sb.sb_agcount);
1238 }
1239 if (agbno >= mp->m_sb.sb_agblocks) {
1240 xfs_fs_cmn_err(CE_ALERT, mp,
1241 "xfs_imap: agbno (0x%llx) >= "
1242 "mp->m_sb.sb_agblocks (0x%lx)",
1243 (unsigned long long) agbno,
1244 (unsigned long) mp->m_sb.sb_agblocks);
1245 }
1246 if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1247 xfs_fs_cmn_err(CE_ALERT, mp,
1248 "xfs_imap: ino (0x%llx) != "
1249 "XFS_AGINO_TO_INO(mp, agno, agino) "
1250 "(0x%llx)",
1251 ino, XFS_AGINO_TO_INO(mp, agno, agino));
1252 }
1253 xfs_stack_trace();
1254 #endif /* DEBUG */
1255 return XFS_ERROR(EINVAL);
1256 }
1257
1258 /*
1259 * If the inode cluster size is the same as the blocksize or
1260 * smaller we get to the buffer by simple arithmetics.
1261 */
1262 if (XFS_INODE_CLUSTER_SIZE(mp) <= mp->m_sb.sb_blocksize) {
1263 offset = XFS_INO_TO_OFFSET(mp, ino);
1264 ASSERT(offset < mp->m_sb.sb_inopblock);
1265
1266 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno);
1267 imap->im_len = XFS_FSB_TO_BB(mp, 1);
1268 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1269 return 0;
1270 }
1271
1272 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_blocklog;
1273
1274 /*
1275 * If we get a block number passed from bulkstat we can use it to
1276 * find the buffer easily.
1277 */
1278 if (imap->im_blkno) {
1279 offset = XFS_INO_TO_OFFSET(mp, ino);
1280 ASSERT(offset < mp->m_sb.sb_inopblock);
1281
1282 cluster_agbno = xfs_daddr_to_agbno(mp, imap->im_blkno);
1283 offset += (agbno - cluster_agbno) * mp->m_sb.sb_inopblock;
1284
1285 imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster);
1286 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1287 return 0;
1288 }
1289
1290 /*
1291 * If the inode chunks are aligned then use simple maths to
1292 * find the location. Otherwise we have to do a btree
1293 * lookup to find the location.
1294 */
1295 if (mp->m_inoalign_mask) {
1296 offset_agbno = agbno & mp->m_inoalign_mask;
1297 chunk_agbno = agbno - offset_agbno;
1298 } else {
1299 xfs_btree_cur_t *cur; /* inode btree cursor */
1300 xfs_agino_t chunk_agino; /* first agino in inode chunk */
1301 __int32_t chunk_cnt; /* count of free inodes in chunk */
1302 xfs_inofree_t chunk_free; /* mask of free inodes in chunk */
1303 xfs_buf_t *agbp; /* agi buffer */
1304 int i; /* temp state */
1305
1306 down_read(&mp->m_peraglock);
1307 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1308 up_read(&mp->m_peraglock);
1309 if (error) {
1310 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
1311 "xfs_ialloc_read_agi() returned "
1312 "error %d, agno %d",
1313 error, agno);
1314 return error;
1315 }
1316
1317 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
1318 error = xfs_inobt_lookup_le(cur, agino, 0, 0, &i);
1319 if (error) {
1320 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
1321 "xfs_inobt_lookup_le() failed");
1322 goto error0;
1323 }
1324
1325 error = xfs_inobt_get_rec(cur, &chunk_agino, &chunk_cnt,
1326 &chunk_free, &i);
1327 if (error) {
1328 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
1329 "xfs_inobt_get_rec() failed");
1330 goto error0;
1331 }
1332 if (i == 0) {
1333 #ifdef DEBUG
1334 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
1335 "xfs_inobt_get_rec() failed");
1336 #endif /* DEBUG */
1337 error = XFS_ERROR(EINVAL);
1338 }
1339 error0:
1340 xfs_trans_brelse(tp, agbp);
1341 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1342 if (error)
1343 return error;
1344 chunk_agbno = XFS_AGINO_TO_AGBNO(mp, chunk_agino);
1345 offset_agbno = agbno - chunk_agbno;
1346 }
1347
1348 ASSERT(agbno >= chunk_agbno);
1349 cluster_agbno = chunk_agbno +
1350 ((offset_agbno / blks_per_cluster) * blks_per_cluster);
1351 offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
1352 XFS_INO_TO_OFFSET(mp, ino);
1353
1354 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno);
1355 imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster);
1356 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1357
1358 /*
1359 * If the inode number maps to a block outside the bounds
1360 * of the file system then return NULL rather than calling
1361 * read_buf and panicing when we get an error from the
1362 * driver.
1363 */
1364 if ((imap->im_blkno + imap->im_len) >
1365 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
1366 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
1367 "(imap->im_blkno (0x%llx) + imap->im_len (0x%llx)) > "
1368 " XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks) (0x%llx)",
1369 (unsigned long long) imap->im_blkno,
1370 (unsigned long long) imap->im_len,
1371 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
1372 return XFS_ERROR(EINVAL);
1373 }
1374
1375 return 0;
1376 }
1377
1378 /*
1379 * Compute and fill in value of m_in_maxlevels.
1380 */
1381 void
1382 xfs_ialloc_compute_maxlevels(
1383 xfs_mount_t *mp) /* file system mount structure */
1384 {
1385 int level;
1386 uint maxblocks;
1387 uint maxleafents;
1388 int minleafrecs;
1389 int minnoderecs;
1390
1391 maxleafents = (1LL << XFS_INO_AGINO_BITS(mp)) >>
1392 XFS_INODES_PER_CHUNK_LOG;
1393 minleafrecs = mp->m_alloc_mnr[0];
1394 minnoderecs = mp->m_alloc_mnr[1];
1395 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
1396 for (level = 1; maxblocks > 1; level++)
1397 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
1398 mp->m_in_maxlevels = level;
1399 }
1400
1401 /*
1402 * Log specified fields for the ag hdr (inode section)
1403 */
1404 void
1405 xfs_ialloc_log_agi(
1406 xfs_trans_t *tp, /* transaction pointer */
1407 xfs_buf_t *bp, /* allocation group header buffer */
1408 int fields) /* bitmask of fields to log */
1409 {
1410 int first; /* first byte number */
1411 int last; /* last byte number */
1412 static const short offsets[] = { /* field starting offsets */
1413 /* keep in sync with bit definitions */
1414 offsetof(xfs_agi_t, agi_magicnum),
1415 offsetof(xfs_agi_t, agi_versionnum),
1416 offsetof(xfs_agi_t, agi_seqno),
1417 offsetof(xfs_agi_t, agi_length),
1418 offsetof(xfs_agi_t, agi_count),
1419 offsetof(xfs_agi_t, agi_root),
1420 offsetof(xfs_agi_t, agi_level),
1421 offsetof(xfs_agi_t, agi_freecount),
1422 offsetof(xfs_agi_t, agi_newino),
1423 offsetof(xfs_agi_t, agi_dirino),
1424 offsetof(xfs_agi_t, agi_unlinked),
1425 sizeof(xfs_agi_t)
1426 };
1427 #ifdef DEBUG
1428 xfs_agi_t *agi; /* allocation group header */
1429
1430 agi = XFS_BUF_TO_AGI(bp);
1431 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
1432 #endif
1433 /*
1434 * Compute byte offsets for the first and last fields.
1435 */
1436 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS, &first, &last);
1437 /*
1438 * Log the allocation group inode header buffer.
1439 */
1440 xfs_trans_log_buf(tp, bp, first, last);
1441 }
1442
1443 #ifdef DEBUG
1444 STATIC void
1445 xfs_check_agi_unlinked(
1446 struct xfs_agi *agi)
1447 {
1448 int i;
1449
1450 for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++)
1451 ASSERT(agi->agi_unlinked[i]);
1452 }
1453 #else
1454 #define xfs_check_agi_unlinked(agi)
1455 #endif
1456
1457 /*
1458 * Read in the allocation group header (inode allocation section)
1459 */
1460 int
1461 xfs_read_agi(
1462 struct xfs_mount *mp, /* file system mount structure */
1463 struct xfs_trans *tp, /* transaction pointer */
1464 xfs_agnumber_t agno, /* allocation group number */
1465 struct xfs_buf **bpp) /* allocation group hdr buf */
1466 {
1467 struct xfs_agi *agi; /* allocation group header */
1468 int agi_ok; /* agi is consistent */
1469 int error;
1470
1471 ASSERT(agno != NULLAGNUMBER);
1472
1473 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
1474 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
1475 XFS_FSS_TO_BB(mp, 1), 0, bpp);
1476 if (error)
1477 return error;
1478
1479 ASSERT(*bpp && !XFS_BUF_GETERROR(*bpp));
1480 agi = XFS_BUF_TO_AGI(*bpp);
1481
1482 /*
1483 * Validate the magic number of the agi block.
1484 */
1485 agi_ok = be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1486 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)) &&
1487 be32_to_cpu(agi->agi_seqno) == agno;
1488 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IALLOC_READ_AGI,
1489 XFS_RANDOM_IALLOC_READ_AGI))) {
1490 XFS_CORRUPTION_ERROR("xfs_read_agi", XFS_ERRLEVEL_LOW,
1491 mp, agi);
1492 xfs_trans_brelse(tp, *bpp);
1493 return XFS_ERROR(EFSCORRUPTED);
1494 }
1495
1496 XFS_BUF_SET_VTYPE_REF(*bpp, B_FS_AGI, XFS_AGI_REF);
1497
1498 xfs_check_agi_unlinked(agi);
1499 return 0;
1500 }
1501
1502 int
1503 xfs_ialloc_read_agi(
1504 struct xfs_mount *mp, /* file system mount structure */
1505 struct xfs_trans *tp, /* transaction pointer */
1506 xfs_agnumber_t agno, /* allocation group number */
1507 struct xfs_buf **bpp) /* allocation group hdr buf */
1508 {
1509 struct xfs_agi *agi; /* allocation group header */
1510 struct xfs_perag *pag; /* per allocation group data */
1511 int error;
1512
1513 error = xfs_read_agi(mp, tp, agno, bpp);
1514 if (error)
1515 return error;
1516
1517 agi = XFS_BUF_TO_AGI(*bpp);
1518 pag = &mp->m_perag[agno];
1519
1520 if (!pag->pagi_init) {
1521 pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
1522 pag->pagi_count = be32_to_cpu(agi->agi_count);
1523 pag->pagi_init = 1;
1524 }
1525
1526 /*
1527 * It's possible for these to be out of sync if
1528 * we are in the middle of a forced shutdown.
1529 */
1530 ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) ||
1531 XFS_FORCED_SHUTDOWN(mp));
1532 return 0;
1533 }
1534
1535 /*
1536 * Read in the agi to initialise the per-ag data in the mount structure
1537 */
1538 int
1539 xfs_ialloc_pagi_init(
1540 xfs_mount_t *mp, /* file system mount structure */
1541 xfs_trans_t *tp, /* transaction pointer */
1542 xfs_agnumber_t agno) /* allocation group number */
1543 {
1544 xfs_buf_t *bp = NULL;
1545 int error;
1546
1547 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
1548 if (error)
1549 return error;
1550 if (bp)
1551 xfs_trans_brelse(tp, bp);
1552 return 0;
1553 }
A cbs scheduler for rt-kernel