1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6 #include "xfs.h"
7 #include "xfs_fs.h"
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_bit.h"
13 #include "xfs_sb.h"
14 #include "xfs_mount.h"
15 #include "xfs_defer.h"
16 #include "xfs_inode.h"
17 #include "xfs_btree.h"
18 #include "xfs_ialloc.h"
19 #include "xfs_ialloc_btree.h"
20 #include "xfs_alloc.h"
21 #include "xfs_rtalloc.h"
22 #include "xfs_errortag.h"
23 #include "xfs_error.h"
24 #include "xfs_bmap.h"
25 #include "xfs_cksum.h"
26 #include "xfs_trans.h"
27 #include "xfs_buf_item.h"
28 #include "xfs_icreate_item.h"
29 #include "xfs_icache.h"
30 #include "xfs_trace.h"
31 #include "xfs_log.h"
32 #include "xfs_rmap.h"
33
34
35 /*
36 * Allocation group level functions.
37 */
38 int
xfs_ialloc_cluster_alignment(struct xfs_mount * mp)39 xfs_ialloc_cluster_alignment(
40 struct xfs_mount *mp)
41 {
42 if (xfs_sb_version_hasalign(&mp->m_sb) &&
43 mp->m_sb.sb_inoalignmt >= xfs_icluster_size_fsb(mp))
44 return mp->m_sb.sb_inoalignmt;
45 return 1;
46 }
47
48 /*
49 * Lookup a record by ino in the btree given by cur.
50 */
51 int /* error */
xfs_inobt_lookup(struct xfs_btree_cur * cur,xfs_agino_t ino,xfs_lookup_t dir,int * stat)52 xfs_inobt_lookup(
53 struct xfs_btree_cur *cur, /* btree cursor */
54 xfs_agino_t ino, /* starting inode of chunk */
55 xfs_lookup_t dir, /* <=, >=, == */
56 int *stat) /* success/failure */
57 {
58 cur->bc_rec.i.ir_startino = ino;
59 cur->bc_rec.i.ir_holemask = 0;
60 cur->bc_rec.i.ir_count = 0;
61 cur->bc_rec.i.ir_freecount = 0;
62 cur->bc_rec.i.ir_free = 0;
63 return xfs_btree_lookup(cur, dir, stat);
64 }
65
66 /*
67 * Update the record referred to by cur to the value given.
68 * This either works (return 0) or gets an EFSCORRUPTED error.
69 */
70 STATIC int /* error */
xfs_inobt_update(struct xfs_btree_cur * cur,xfs_inobt_rec_incore_t * irec)71 xfs_inobt_update(
72 struct xfs_btree_cur *cur, /* btree cursor */
73 xfs_inobt_rec_incore_t *irec) /* btree record */
74 {
75 union xfs_btree_rec rec;
76
77 rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino);
78 if (xfs_sb_version_hassparseinodes(&cur->bc_mp->m_sb)) {
79 rec.inobt.ir_u.sp.ir_holemask = cpu_to_be16(irec->ir_holemask);
80 rec.inobt.ir_u.sp.ir_count = irec->ir_count;
81 rec.inobt.ir_u.sp.ir_freecount = irec->ir_freecount;
82 } else {
83 /* ir_holemask/ir_count not supported on-disk */
84 rec.inobt.ir_u.f.ir_freecount = cpu_to_be32(irec->ir_freecount);
85 }
86 rec.inobt.ir_free = cpu_to_be64(irec->ir_free);
87 return xfs_btree_update(cur, &rec);
88 }
89
90 /* Convert on-disk btree record to incore inobt record. */
91 void
xfs_inobt_btrec_to_irec(struct xfs_mount * mp,union xfs_btree_rec * rec,struct xfs_inobt_rec_incore * irec)92 xfs_inobt_btrec_to_irec(
93 struct xfs_mount *mp,
94 union xfs_btree_rec *rec,
95 struct xfs_inobt_rec_incore *irec)
96 {
97 irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino);
98 if (xfs_sb_version_hassparseinodes(&mp->m_sb)) {
99 irec->ir_holemask = be16_to_cpu(rec->inobt.ir_u.sp.ir_holemask);
100 irec->ir_count = rec->inobt.ir_u.sp.ir_count;
101 irec->ir_freecount = rec->inobt.ir_u.sp.ir_freecount;
102 } else {
103 /*
104 * ir_holemask/ir_count not supported on-disk. Fill in hardcoded
105 * values for full inode chunks.
106 */
107 irec->ir_holemask = XFS_INOBT_HOLEMASK_FULL;
108 irec->ir_count = XFS_INODES_PER_CHUNK;
109 irec->ir_freecount =
110 be32_to_cpu(rec->inobt.ir_u.f.ir_freecount);
111 }
112 irec->ir_free = be64_to_cpu(rec->inobt.ir_free);
113 }
114
115 /*
116 * Get the data from the pointed-to record.
117 */
118 int
xfs_inobt_get_rec(struct xfs_btree_cur * cur,struct xfs_inobt_rec_incore * irec,int * stat)119 xfs_inobt_get_rec(
120 struct xfs_btree_cur *cur,
121 struct xfs_inobt_rec_incore *irec,
122 int *stat)
123 {
124 struct xfs_mount *mp = cur->bc_mp;
125 xfs_agnumber_t agno = cur->bc_private.a.agno;
126 union xfs_btree_rec *rec;
127 int error;
128 uint64_t realfree;
129
130 error = xfs_btree_get_rec(cur, &rec, stat);
131 if (error || *stat == 0)
132 return error;
133
134 xfs_inobt_btrec_to_irec(mp, rec, irec);
135
136 if (!xfs_verify_agino(mp, agno, irec->ir_startino))
137 goto out_bad_rec;
138 if (irec->ir_count < XFS_INODES_PER_HOLEMASK_BIT ||
139 irec->ir_count > XFS_INODES_PER_CHUNK)
140 goto out_bad_rec;
141 if (irec->ir_freecount > XFS_INODES_PER_CHUNK)
142 goto out_bad_rec;
143
144 /* if there are no holes, return the first available offset */
145 if (!xfs_inobt_issparse(irec->ir_holemask))
146 realfree = irec->ir_free;
147 else
148 realfree = irec->ir_free & xfs_inobt_irec_to_allocmask(irec);
149 if (hweight64(realfree) != irec->ir_freecount)
150 goto out_bad_rec;
151
152 return 0;
153
154 out_bad_rec:
155 xfs_warn(mp,
156 "%s Inode BTree record corruption in AG %d detected!",
157 cur->bc_btnum == XFS_BTNUM_INO ? "Used" : "Free", agno);
158 xfs_warn(mp,
159 "start inode 0x%x, count 0x%x, free 0x%x freemask 0x%llx, holemask 0x%x",
160 irec->ir_startino, irec->ir_count, irec->ir_freecount,
161 irec->ir_free, irec->ir_holemask);
162 return -EFSCORRUPTED;
163 }
164
165 /*
166 * Insert a single inobt record. Cursor must already point to desired location.
167 */
168 int
xfs_inobt_insert_rec(struct xfs_btree_cur * cur,uint16_t holemask,uint8_t count,int32_t freecount,xfs_inofree_t free,int * stat)169 xfs_inobt_insert_rec(
170 struct xfs_btree_cur *cur,
171 uint16_t holemask,
172 uint8_t count,
173 int32_t freecount,
174 xfs_inofree_t free,
175 int *stat)
176 {
177 cur->bc_rec.i.ir_holemask = holemask;
178 cur->bc_rec.i.ir_count = count;
179 cur->bc_rec.i.ir_freecount = freecount;
180 cur->bc_rec.i.ir_free = free;
181 return xfs_btree_insert(cur, stat);
182 }
183
184 /*
185 * Insert records describing a newly allocated inode chunk into the inobt.
186 */
187 STATIC int
xfs_inobt_insert(struct xfs_mount * mp,struct xfs_trans * tp,struct xfs_buf * agbp,xfs_agino_t newino,xfs_agino_t newlen,xfs_btnum_t btnum)188 xfs_inobt_insert(
189 struct xfs_mount *mp,
190 struct xfs_trans *tp,
191 struct xfs_buf *agbp,
192 xfs_agino_t newino,
193 xfs_agino_t newlen,
194 xfs_btnum_t btnum)
195 {
196 struct xfs_btree_cur *cur;
197 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
198 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
199 xfs_agino_t thisino;
200 int i;
201 int error;
202
203 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum);
204
205 for (thisino = newino;
206 thisino < newino + newlen;
207 thisino += XFS_INODES_PER_CHUNK) {
208 error = xfs_inobt_lookup(cur, thisino, XFS_LOOKUP_EQ, &i);
209 if (error) {
210 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
211 return error;
212 }
213 ASSERT(i == 0);
214
215 error = xfs_inobt_insert_rec(cur, XFS_INOBT_HOLEMASK_FULL,
216 XFS_INODES_PER_CHUNK,
217 XFS_INODES_PER_CHUNK,
218 XFS_INOBT_ALL_FREE, &i);
219 if (error) {
220 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
221 return error;
222 }
223 ASSERT(i == 1);
224 }
225
226 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
227
228 return 0;
229 }
230
231 /*
232 * Verify that the number of free inodes in the AGI is correct.
233 */
234 #ifdef DEBUG
235 STATIC int
xfs_check_agi_freecount(struct xfs_btree_cur * cur,struct xfs_agi * agi)236 xfs_check_agi_freecount(
237 struct xfs_btree_cur *cur,
238 struct xfs_agi *agi)
239 {
240 if (cur->bc_nlevels == 1) {
241 xfs_inobt_rec_incore_t rec;
242 int freecount = 0;
243 int error;
244 int i;
245
246 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
247 if (error)
248 return error;
249
250 do {
251 error = xfs_inobt_get_rec(cur, &rec, &i);
252 if (error)
253 return error;
254
255 if (i) {
256 freecount += rec.ir_freecount;
257 error = xfs_btree_increment(cur, 0, &i);
258 if (error)
259 return error;
260 }
261 } while (i == 1);
262
263 if (!XFS_FORCED_SHUTDOWN(cur->bc_mp))
264 ASSERT(freecount == be32_to_cpu(agi->agi_freecount));
265 }
266 return 0;
267 }
268 #else
269 #define xfs_check_agi_freecount(cur, agi) 0
270 #endif
271
272 /*
273 * Initialise a new set of inodes. When called without a transaction context
274 * (e.g. from recovery) we initiate a delayed write of the inode buffers rather
275 * than logging them (which in a transaction context puts them into the AIL
276 * for writeback rather than the xfsbufd queue).
277 */
278 int
xfs_ialloc_inode_init(struct xfs_mount * mp,struct xfs_trans * tp,struct list_head * buffer_list,int icount,xfs_agnumber_t agno,xfs_agblock_t agbno,xfs_agblock_t length,unsigned int gen)279 xfs_ialloc_inode_init(
280 struct xfs_mount *mp,
281 struct xfs_trans *tp,
282 struct list_head *buffer_list,
283 int icount,
284 xfs_agnumber_t agno,
285 xfs_agblock_t agbno,
286 xfs_agblock_t length,
287 unsigned int gen)
288 {
289 struct xfs_buf *fbuf;
290 struct xfs_dinode *free;
291 int nbufs, blks_per_cluster, inodes_per_cluster;
292 int version;
293 int i, j;
294 xfs_daddr_t d;
295 xfs_ino_t ino = 0;
296
297 /*
298 * Loop over the new block(s), filling in the inodes. For small block
299 * sizes, manipulate the inodes in buffers which are multiples of the
300 * blocks size.
301 */
302 blks_per_cluster = xfs_icluster_size_fsb(mp);
303 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
304 nbufs = length / blks_per_cluster;
305
306 /*
307 * Figure out what version number to use in the inodes we create. If
308 * the superblock version has caught up to the one that supports the new
309 * inode format, then use the new inode version. Otherwise use the old
310 * version so that old kernels will continue to be able to use the file
311 * system.
312 *
313 * For v3 inodes, we also need to write the inode number into the inode,
314 * so calculate the first inode number of the chunk here as
315 * XFS_OFFBNO_TO_AGINO() only works within a filesystem block, not
316 * across multiple filesystem blocks (such as a cluster) and so cannot
317 * be used in the cluster buffer loop below.
318 *
319 * Further, because we are writing the inode directly into the buffer
320 * and calculating a CRC on the entire inode, we have ot log the entire
321 * inode so that the entire range the CRC covers is present in the log.
322 * That means for v3 inode we log the entire buffer rather than just the
323 * inode cores.
324 */
325 if (xfs_sb_version_hascrc(&mp->m_sb)) {
326 version = 3;
327 ino = XFS_AGINO_TO_INO(mp, agno,
328 XFS_OFFBNO_TO_AGINO(mp, agbno, 0));
329
330 /*
331 * log the initialisation that is about to take place as an
332 * logical operation. This means the transaction does not
333 * need to log the physical changes to the inode buffers as log
334 * recovery will know what initialisation is actually needed.
335 * Hence we only need to log the buffers as "ordered" buffers so
336 * they track in the AIL as if they were physically logged.
337 */
338 if (tp)
339 xfs_icreate_log(tp, agno, agbno, icount,
340 mp->m_sb.sb_inodesize, length, gen);
341 } else
342 version = 2;
343
344 for (j = 0; j < nbufs; j++) {
345 /*
346 * Get the block.
347 */
348 d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster));
349 fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
350 mp->m_bsize * blks_per_cluster,
351 XBF_UNMAPPED);
352 if (!fbuf)
353 return -ENOMEM;
354
355 /* Initialize the inode buffers and log them appropriately. */
356 fbuf->b_ops = &xfs_inode_buf_ops;
357 xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length));
358 for (i = 0; i < inodes_per_cluster; i++) {
359 int ioffset = i << mp->m_sb.sb_inodelog;
360 uint isize = xfs_dinode_size(version);
361
362 free = xfs_make_iptr(mp, fbuf, i);
363 free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
364 free->di_version = version;
365 free->di_gen = cpu_to_be32(gen);
366 free->di_next_unlinked = cpu_to_be32(NULLAGINO);
367
368 if (version == 3) {
369 free->di_ino = cpu_to_be64(ino);
370 ino++;
371 uuid_copy(&free->di_uuid,
372 &mp->m_sb.sb_meta_uuid);
373 xfs_dinode_calc_crc(mp, free);
374 } else if (tp) {
375 /* just log the inode core */
376 xfs_trans_log_buf(tp, fbuf, ioffset,
377 ioffset + isize - 1);
378 }
379 }
380
381 if (tp) {
382 /*
383 * Mark the buffer as an inode allocation buffer so it
384 * sticks in AIL at the point of this allocation
385 * transaction. This ensures the they are on disk before
386 * the tail of the log can be moved past this
387 * transaction (i.e. by preventing relogging from moving
388 * it forward in the log).
389 */
390 xfs_trans_inode_alloc_buf(tp, fbuf);
391 if (version == 3) {
392 /*
393 * Mark the buffer as ordered so that they are
394 * not physically logged in the transaction but
395 * still tracked in the AIL as part of the
396 * transaction and pin the log appropriately.
397 */
398 xfs_trans_ordered_buf(tp, fbuf);
399 }
400 } else {
401 fbuf->b_flags |= XBF_DONE;
402 xfs_buf_delwri_queue(fbuf, buffer_list);
403 xfs_buf_relse(fbuf);
404 }
405 }
406 return 0;
407 }
408
409 /*
410 * Align startino and allocmask for a recently allocated sparse chunk such that
411 * they are fit for insertion (or merge) into the on-disk inode btrees.
412 *
413 * Background:
414 *
415 * When enabled, sparse inode support increases the inode alignment from cluster
416 * size to inode chunk size. This means that the minimum range between two
417 * non-adjacent inode records in the inobt is large enough for a full inode
418 * record. This allows for cluster sized, cluster aligned block allocation
419 * without need to worry about whether the resulting inode record overlaps with
420 * another record in the tree. Without this basic rule, we would have to deal
421 * with the consequences of overlap by potentially undoing recent allocations in
422 * the inode allocation codepath.
423 *
424 * Because of this alignment rule (which is enforced on mount), there are two
425 * inobt possibilities for newly allocated sparse chunks. One is that the
426 * aligned inode record for the chunk covers a range of inodes not already
427 * covered in the inobt (i.e., it is safe to insert a new sparse record). The
428 * other is that a record already exists at the aligned startino that considers
429 * the newly allocated range as sparse. In the latter case, record content is
430 * merged in hope that sparse inode chunks fill to full chunks over time.
431 */
432 STATIC void
xfs_align_sparse_ino(struct xfs_mount * mp,xfs_agino_t * startino,uint16_t * allocmask)433 xfs_align_sparse_ino(
434 struct xfs_mount *mp,
435 xfs_agino_t *startino,
436 uint16_t *allocmask)
437 {
438 xfs_agblock_t agbno;
439 xfs_agblock_t mod;
440 int offset;
441
442 agbno = XFS_AGINO_TO_AGBNO(mp, *startino);
443 mod = agbno % mp->m_sb.sb_inoalignmt;
444 if (!mod)
445 return;
446
447 /* calculate the inode offset and align startino */
448 offset = mod << mp->m_sb.sb_inopblog;
449 *startino -= offset;
450
451 /*
452 * Since startino has been aligned down, left shift allocmask such that
453 * it continues to represent the same physical inodes relative to the
454 * new startino.
455 */
456 *allocmask <<= offset / XFS_INODES_PER_HOLEMASK_BIT;
457 }
458
459 /*
460 * Determine whether the source inode record can merge into the target. Both
461 * records must be sparse, the inode ranges must match and there must be no
462 * allocation overlap between the records.
463 */
464 STATIC bool
__xfs_inobt_can_merge(struct xfs_inobt_rec_incore * trec,struct xfs_inobt_rec_incore * srec)465 __xfs_inobt_can_merge(
466 struct xfs_inobt_rec_incore *trec, /* tgt record */
467 struct xfs_inobt_rec_incore *srec) /* src record */
468 {
469 uint64_t talloc;
470 uint64_t salloc;
471
472 /* records must cover the same inode range */
473 if (trec->ir_startino != srec->ir_startino)
474 return false;
475
476 /* both records must be sparse */
477 if (!xfs_inobt_issparse(trec->ir_holemask) ||
478 !xfs_inobt_issparse(srec->ir_holemask))
479 return false;
480
481 /* both records must track some inodes */
482 if (!trec->ir_count || !srec->ir_count)
483 return false;
484
485 /* can't exceed capacity of a full record */
486 if (trec->ir_count + srec->ir_count > XFS_INODES_PER_CHUNK)
487 return false;
488
489 /* verify there is no allocation overlap */
490 talloc = xfs_inobt_irec_to_allocmask(trec);
491 salloc = xfs_inobt_irec_to_allocmask(srec);
492 if (talloc & salloc)
493 return false;
494
495 return true;
496 }
497
498 /*
499 * Merge the source inode record into the target. The caller must call
500 * __xfs_inobt_can_merge() to ensure the merge is valid.
501 */
502 STATIC void
__xfs_inobt_rec_merge(struct xfs_inobt_rec_incore * trec,struct xfs_inobt_rec_incore * srec)503 __xfs_inobt_rec_merge(
504 struct xfs_inobt_rec_incore *trec, /* target */
505 struct xfs_inobt_rec_incore *srec) /* src */
506 {
507 ASSERT(trec->ir_startino == srec->ir_startino);
508
509 /* combine the counts */
510 trec->ir_count += srec->ir_count;
511 trec->ir_freecount += srec->ir_freecount;
512
513 /*
514 * Merge the holemask and free mask. For both fields, 0 bits refer to
515 * allocated inodes. We combine the allocated ranges with bitwise AND.
516 */
517 trec->ir_holemask &= srec->ir_holemask;
518 trec->ir_free &= srec->ir_free;
519 }
520
521 /*
522 * Insert a new sparse inode chunk into the associated inode btree. The inode
523 * record for the sparse chunk is pre-aligned to a startino that should match
524 * any pre-existing sparse inode record in the tree. This allows sparse chunks
525 * to fill over time.
526 *
527 * This function supports two modes of handling preexisting records depending on
528 * the merge flag. If merge is true, the provided record is merged with the
529 * existing record and updated in place. The merged record is returned in nrec.
530 * If merge is false, an existing record is replaced with the provided record.
531 * If no preexisting record exists, the provided record is always inserted.
532 *
533 * It is considered corruption if a merge is requested and not possible. Given
534 * the sparse inode alignment constraints, this should never happen.
535 */
536 STATIC int
xfs_inobt_insert_sprec(struct xfs_mount * mp,struct xfs_trans * tp,struct xfs_buf * agbp,int btnum,struct xfs_inobt_rec_incore * nrec,bool merge)537 xfs_inobt_insert_sprec(
538 struct xfs_mount *mp,
539 struct xfs_trans *tp,
540 struct xfs_buf *agbp,
541 int btnum,
542 struct xfs_inobt_rec_incore *nrec, /* in/out: new/merged rec. */
543 bool merge) /* merge or replace */
544 {
545 struct xfs_btree_cur *cur;
546 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
547 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
548 int error;
549 int i;
550 struct xfs_inobt_rec_incore rec;
551
552 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum);
553
554 /* the new record is pre-aligned so we know where to look */
555 error = xfs_inobt_lookup(cur, nrec->ir_startino, XFS_LOOKUP_EQ, &i);
556 if (error)
557 goto error;
558 /* if nothing there, insert a new record and return */
559 if (i == 0) {
560 error = xfs_inobt_insert_rec(cur, nrec->ir_holemask,
561 nrec->ir_count, nrec->ir_freecount,
562 nrec->ir_free, &i);
563 if (error)
564 goto error;
565 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error);
566
567 goto out;
568 }
569
570 /*
571 * A record exists at this startino. Merge or replace the record
572 * depending on what we've been asked to do.
573 */
574 if (merge) {
575 error = xfs_inobt_get_rec(cur, &rec, &i);
576 if (error)
577 goto error;
578 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error);
579 XFS_WANT_CORRUPTED_GOTO(mp,
580 rec.ir_startino == nrec->ir_startino,
581 error);
582
583 /*
584 * This should never fail. If we have coexisting records that
585 * cannot merge, something is seriously wrong.
586 */
587 XFS_WANT_CORRUPTED_GOTO(mp, __xfs_inobt_can_merge(nrec, &rec),
588 error);
589
590 trace_xfs_irec_merge_pre(mp, agno, rec.ir_startino,
591 rec.ir_holemask, nrec->ir_startino,
592 nrec->ir_holemask);
593
594 /* merge to nrec to output the updated record */
595 __xfs_inobt_rec_merge(nrec, &rec);
596
597 trace_xfs_irec_merge_post(mp, agno, nrec->ir_startino,
598 nrec->ir_holemask);
599
600 error = xfs_inobt_rec_check_count(mp, nrec);
601 if (error)
602 goto error;
603 }
604
605 error = xfs_inobt_update(cur, nrec);
606 if (error)
607 goto error;
608
609 out:
610 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
611 return 0;
612 error:
613 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
614 return error;
615 }
616
617 /*
618 * Allocate new inodes in the allocation group specified by agbp.
619 * Return 0 for success, else error code.
620 */
621 STATIC int /* error code or 0 */
xfs_ialloc_ag_alloc(xfs_trans_t * tp,xfs_buf_t * agbp,int * alloc)622 xfs_ialloc_ag_alloc(
623 xfs_trans_t *tp, /* transaction pointer */
624 xfs_buf_t *agbp, /* alloc group buffer */
625 int *alloc)
626 {
627 xfs_agi_t *agi; /* allocation group header */
628 xfs_alloc_arg_t args; /* allocation argument structure */
629 xfs_agnumber_t agno;
630 int error;
631 xfs_agino_t newino; /* new first inode's number */
632 xfs_agino_t newlen; /* new number of inodes */
633 int isaligned = 0; /* inode allocation at stripe unit */
634 /* boundary */
635 uint16_t allocmask = (uint16_t) -1; /* init. to full chunk */
636 struct xfs_inobt_rec_incore rec;
637 struct xfs_perag *pag;
638 int do_sparse = 0;
639
640 memset(&args, 0, sizeof(args));
641 args.tp = tp;
642 args.mp = tp->t_mountp;
643 args.fsbno = NULLFSBLOCK;
644 xfs_rmap_ag_owner(&args.oinfo, XFS_RMAP_OWN_INODES);
645
646 #ifdef DEBUG
647 /* randomly do sparse inode allocations */
648 if (xfs_sb_version_hassparseinodes(&tp->t_mountp->m_sb) &&
649 args.mp->m_ialloc_min_blks < args.mp->m_ialloc_blks)
650 do_sparse = prandom_u32() & 1;
651 #endif
652
653 /*
654 * Locking will ensure that we don't have two callers in here
655 * at one time.
656 */
657 newlen = args.mp->m_ialloc_inos;
658 if (args.mp->m_maxicount &&
659 percpu_counter_read_positive(&args.mp->m_icount) + newlen >
660 args.mp->m_maxicount)
661 return -ENOSPC;
662 args.minlen = args.maxlen = args.mp->m_ialloc_blks;
663 /*
664 * First try to allocate inodes contiguous with the last-allocated
665 * chunk of inodes. If the filesystem is striped, this will fill
666 * an entire stripe unit with inodes.
667 */
668 agi = XFS_BUF_TO_AGI(agbp);
669 newino = be32_to_cpu(agi->agi_newino);
670 agno = be32_to_cpu(agi->agi_seqno);
671 args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) +
672 args.mp->m_ialloc_blks;
673 if (do_sparse)
674 goto sparse_alloc;
675 if (likely(newino != NULLAGINO &&
676 (args.agbno < be32_to_cpu(agi->agi_length)))) {
677 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
678 args.type = XFS_ALLOCTYPE_THIS_BNO;
679 args.prod = 1;
680
681 /*
682 * We need to take into account alignment here to ensure that
683 * we don't modify the free list if we fail to have an exact
684 * block. If we don't have an exact match, and every oher
685 * attempt allocation attempt fails, we'll end up cancelling
686 * a dirty transaction and shutting down.
687 *
688 * For an exact allocation, alignment must be 1,
689 * however we need to take cluster alignment into account when
690 * fixing up the freelist. Use the minalignslop field to
691 * indicate that extra blocks might be required for alignment,
692 * but not to use them in the actual exact allocation.
693 */
694 args.alignment = 1;
695 args.minalignslop = xfs_ialloc_cluster_alignment(args.mp) - 1;
696
697 /* Allow space for the inode btree to split. */
698 args.minleft = args.mp->m_in_maxlevels - 1;
699 if ((error = xfs_alloc_vextent(&args)))
700 return error;
701
702 /*
703 * This request might have dirtied the transaction if the AG can
704 * satisfy the request, but the exact block was not available.
705 * If the allocation did fail, subsequent requests will relax
706 * the exact agbno requirement and increase the alignment
707 * instead. It is critical that the total size of the request
708 * (len + alignment + slop) does not increase from this point
709 * on, so reset minalignslop to ensure it is not included in
710 * subsequent requests.
711 */
712 args.minalignslop = 0;
713 }
714
715 if (unlikely(args.fsbno == NULLFSBLOCK)) {
716 /*
717 * Set the alignment for the allocation.
718 * If stripe alignment is turned on then align at stripe unit
719 * boundary.
720 * If the cluster size is smaller than a filesystem block
721 * then we're doing I/O for inodes in filesystem block size
722 * pieces, so don't need alignment anyway.
723 */
724 isaligned = 0;
725 if (args.mp->m_sinoalign) {
726 ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN));
727 args.alignment = args.mp->m_dalign;
728 isaligned = 1;
729 } else
730 args.alignment = xfs_ialloc_cluster_alignment(args.mp);
731 /*
732 * Need to figure out where to allocate the inode blocks.
733 * Ideally they should be spaced out through the a.g.
734 * For now, just allocate blocks up front.
735 */
736 args.agbno = be32_to_cpu(agi->agi_root);
737 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
738 /*
739 * Allocate a fixed-size extent of inodes.
740 */
741 args.type = XFS_ALLOCTYPE_NEAR_BNO;
742 args.prod = 1;
743 /*
744 * Allow space for the inode btree to split.
745 */
746 args.minleft = args.mp->m_in_maxlevels - 1;
747 if ((error = xfs_alloc_vextent(&args)))
748 return error;
749 }
750
751 /*
752 * If stripe alignment is turned on, then try again with cluster
753 * alignment.
754 */
755 if (isaligned && args.fsbno == NULLFSBLOCK) {
756 args.type = XFS_ALLOCTYPE_NEAR_BNO;
757 args.agbno = be32_to_cpu(agi->agi_root);
758 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
759 args.alignment = xfs_ialloc_cluster_alignment(args.mp);
760 if ((error = xfs_alloc_vextent(&args)))
761 return error;
762 }
763
764 /*
765 * Finally, try a sparse allocation if the filesystem supports it and
766 * the sparse allocation length is smaller than a full chunk.
767 */
768 if (xfs_sb_version_hassparseinodes(&args.mp->m_sb) &&
769 args.mp->m_ialloc_min_blks < args.mp->m_ialloc_blks &&
770 args.fsbno == NULLFSBLOCK) {
771 sparse_alloc:
772 args.type = XFS_ALLOCTYPE_NEAR_BNO;
773 args.agbno = be32_to_cpu(agi->agi_root);
774 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
775 args.alignment = args.mp->m_sb.sb_spino_align;
776 args.prod = 1;
777
778 args.minlen = args.mp->m_ialloc_min_blks;
779 args.maxlen = args.minlen;
780
781 /*
782 * The inode record will be aligned to full chunk size. We must
783 * prevent sparse allocation from AG boundaries that result in
784 * invalid inode records, such as records that start at agbno 0
785 * or extend beyond the AG.
786 *
787 * Set min agbno to the first aligned, non-zero agbno and max to
788 * the last aligned agbno that is at least one full chunk from
789 * the end of the AG.
790 */
791 args.min_agbno = args.mp->m_sb.sb_inoalignmt;
792 args.max_agbno = round_down(args.mp->m_sb.sb_agblocks,
793 args.mp->m_sb.sb_inoalignmt) -
794 args.mp->m_ialloc_blks;
795
796 error = xfs_alloc_vextent(&args);
797 if (error)
798 return error;
799
800 newlen = args.len << args.mp->m_sb.sb_inopblog;
801 ASSERT(newlen <= XFS_INODES_PER_CHUNK);
802 allocmask = (1 << (newlen / XFS_INODES_PER_HOLEMASK_BIT)) - 1;
803 }
804
805 if (args.fsbno == NULLFSBLOCK) {
806 *alloc = 0;
807 return 0;
808 }
809 ASSERT(args.len == args.minlen);
810
811 /*
812 * Stamp and write the inode buffers.
813 *
814 * Seed the new inode cluster with a random generation number. This
815 * prevents short-term reuse of generation numbers if a chunk is
816 * freed and then immediately reallocated. We use random numbers
817 * rather than a linear progression to prevent the next generation
818 * number from being easily guessable.
819 */
820 error = xfs_ialloc_inode_init(args.mp, tp, NULL, newlen, agno,
821 args.agbno, args.len, prandom_u32());
822
823 if (error)
824 return error;
825 /*
826 * Convert the results.
827 */
828 newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0);
829
830 if (xfs_inobt_issparse(~allocmask)) {
831 /*
832 * We've allocated a sparse chunk. Align the startino and mask.
833 */
834 xfs_align_sparse_ino(args.mp, &newino, &allocmask);
835
836 rec.ir_startino = newino;
837 rec.ir_holemask = ~allocmask;
838 rec.ir_count = newlen;
839 rec.ir_freecount = newlen;
840 rec.ir_free = XFS_INOBT_ALL_FREE;
841
842 /*
843 * Insert the sparse record into the inobt and allow for a merge
844 * if necessary. If a merge does occur, rec is updated to the
845 * merged record.
846 */
847 error = xfs_inobt_insert_sprec(args.mp, tp, agbp, XFS_BTNUM_INO,
848 &rec, true);
849 if (error == -EFSCORRUPTED) {
850 xfs_alert(args.mp,
851 "invalid sparse inode record: ino 0x%llx holemask 0x%x count %u",
852 XFS_AGINO_TO_INO(args.mp, agno,
853 rec.ir_startino),
854 rec.ir_holemask, rec.ir_count);
855 xfs_force_shutdown(args.mp, SHUTDOWN_CORRUPT_INCORE);
856 }
857 if (error)
858 return error;
859
860 /*
861 * We can't merge the part we've just allocated as for the inobt
862 * due to finobt semantics. The original record may or may not
863 * exist independent of whether physical inodes exist in this
864 * sparse chunk.
865 *
866 * We must update the finobt record based on the inobt record.
867 * rec contains the fully merged and up to date inobt record
868 * from the previous call. Set merge false to replace any
869 * existing record with this one.
870 */
871 if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) {
872 error = xfs_inobt_insert_sprec(args.mp, tp, agbp,
873 XFS_BTNUM_FINO, &rec,
874 false);
875 if (error)
876 return error;
877 }
878 } else {
879 /* full chunk - insert new records to both btrees */
880 error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen,
881 XFS_BTNUM_INO);
882 if (error)
883 return error;
884
885 if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) {
886 error = xfs_inobt_insert(args.mp, tp, agbp, newino,
887 newlen, XFS_BTNUM_FINO);
888 if (error)
889 return error;
890 }
891 }
892
893 /*
894 * Update AGI counts and newino.
895 */
896 be32_add_cpu(&agi->agi_count, newlen);
897 be32_add_cpu(&agi->agi_freecount, newlen);
898 pag = xfs_perag_get(args.mp, agno);
899 pag->pagi_freecount += newlen;
900 pag->pagi_count += newlen;
901 xfs_perag_put(pag);
902 agi->agi_newino = cpu_to_be32(newino);
903
904 /*
905 * Log allocation group header fields
906 */
907 xfs_ialloc_log_agi(tp, agbp,
908 XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO);
909 /*
910 * Modify/log superblock values for inode count and inode free count.
911 */
912 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen);
913 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen);
914 *alloc = 1;
915 return 0;
916 }
917
918 STATIC xfs_agnumber_t
xfs_ialloc_next_ag(xfs_mount_t * mp)919 xfs_ialloc_next_ag(
920 xfs_mount_t *mp)
921 {
922 xfs_agnumber_t agno;
923
924 spin_lock(&mp->m_agirotor_lock);
925 agno = mp->m_agirotor;
926 if (++mp->m_agirotor >= mp->m_maxagi)
927 mp->m_agirotor = 0;
928 spin_unlock(&mp->m_agirotor_lock);
929
930 return agno;
931 }
932
933 /*
934 * Select an allocation group to look for a free inode in, based on the parent
935 * inode and the mode. Return the allocation group buffer.
936 */
937 STATIC xfs_agnumber_t
xfs_ialloc_ag_select(xfs_trans_t * tp,xfs_ino_t parent,umode_t mode)938 xfs_ialloc_ag_select(
939 xfs_trans_t *tp, /* transaction pointer */
940 xfs_ino_t parent, /* parent directory inode number */
941 umode_t mode) /* bits set to indicate file type */
942 {
943 xfs_agnumber_t agcount; /* number of ag's in the filesystem */
944 xfs_agnumber_t agno; /* current ag number */
945 int flags; /* alloc buffer locking flags */
946 xfs_extlen_t ineed; /* blocks needed for inode allocation */
947 xfs_extlen_t longest = 0; /* longest extent available */
948 xfs_mount_t *mp; /* mount point structure */
949 int needspace; /* file mode implies space allocated */
950 xfs_perag_t *pag; /* per allocation group data */
951 xfs_agnumber_t pagno; /* parent (starting) ag number */
952 int error;
953
954 /*
955 * Files of these types need at least one block if length > 0
956 * (and they won't fit in the inode, but that's hard to figure out).
957 */
958 needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode);
959 mp = tp->t_mountp;
960 agcount = mp->m_maxagi;
961 if (S_ISDIR(mode))
962 pagno = xfs_ialloc_next_ag(mp);
963 else {
964 pagno = XFS_INO_TO_AGNO(mp, parent);
965 if (pagno >= agcount)
966 pagno = 0;
967 }
968
969 ASSERT(pagno < agcount);
970
971 /*
972 * Loop through allocation groups, looking for one with a little
973 * free space in it. Note we don't look for free inodes, exactly.
974 * Instead, we include whether there is a need to allocate inodes
975 * to mean that blocks must be allocated for them,
976 * if none are currently free.
977 */
978 agno = pagno;
979 flags = XFS_ALLOC_FLAG_TRYLOCK;
980 for (;;) {
981 pag = xfs_perag_get(mp, agno);
982 if (!pag->pagi_inodeok) {
983 xfs_ialloc_next_ag(mp);
984 goto nextag;
985 }
986
987 if (!pag->pagi_init) {
988 error = xfs_ialloc_pagi_init(mp, tp, agno);
989 if (error)
990 goto nextag;
991 }
992
993 if (pag->pagi_freecount) {
994 xfs_perag_put(pag);
995 return agno;
996 }
997
998 if (!pag->pagf_init) {
999 error = xfs_alloc_pagf_init(mp, tp, agno, flags);
1000 if (error)
1001 goto nextag;
1002 }
1003
1004 /*
1005 * Check that there is enough free space for the file plus a
1006 * chunk of inodes if we need to allocate some. If this is the
1007 * first pass across the AGs, take into account the potential
1008 * space needed for alignment of inode chunks when checking the
1009 * longest contiguous free space in the AG - this prevents us
1010 * from getting ENOSPC because we have free space larger than
1011 * m_ialloc_blks but alignment constraints prevent us from using
1012 * it.
1013 *
1014 * If we can't find an AG with space for full alignment slack to
1015 * be taken into account, we must be near ENOSPC in all AGs.
1016 * Hence we don't include alignment for the second pass and so
1017 * if we fail allocation due to alignment issues then it is most
1018 * likely a real ENOSPC condition.
1019 */
1020 ineed = mp->m_ialloc_min_blks;
1021 if (flags && ineed > 1)
1022 ineed += xfs_ialloc_cluster_alignment(mp);
1023 longest = pag->pagf_longest;
1024 if (!longest)
1025 longest = pag->pagf_flcount > 0;
1026
1027 if (pag->pagf_freeblks >= needspace + ineed &&
1028 longest >= ineed) {
1029 xfs_perag_put(pag);
1030 return agno;
1031 }
1032 nextag:
1033 xfs_perag_put(pag);
1034 /*
1035 * No point in iterating over the rest, if we're shutting
1036 * down.
1037 */
1038 if (XFS_FORCED_SHUTDOWN(mp))
1039 return NULLAGNUMBER;
1040 agno++;
1041 if (agno >= agcount)
1042 agno = 0;
1043 if (agno == pagno) {
1044 if (flags == 0)
1045 return NULLAGNUMBER;
1046 flags = 0;
1047 }
1048 }
1049 }
1050
1051 /*
1052 * Try to retrieve the next record to the left/right from the current one.
1053 */
1054 STATIC int
xfs_ialloc_next_rec(struct xfs_btree_cur * cur,xfs_inobt_rec_incore_t * rec,int * done,int left)1055 xfs_ialloc_next_rec(
1056 struct xfs_btree_cur *cur,
1057 xfs_inobt_rec_incore_t *rec,
1058 int *done,
1059 int left)
1060 {
1061 int error;
1062 int i;
1063
1064 if (left)
1065 error = xfs_btree_decrement(cur, 0, &i);
1066 else
1067 error = xfs_btree_increment(cur, 0, &i);
1068
1069 if (error)
1070 return error;
1071 *done = !i;
1072 if (i) {
1073 error = xfs_inobt_get_rec(cur, rec, &i);
1074 if (error)
1075 return error;
1076 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1077 }
1078
1079 return 0;
1080 }
1081
1082 STATIC int
xfs_ialloc_get_rec(struct xfs_btree_cur * cur,xfs_agino_t agino,xfs_inobt_rec_incore_t * rec,int * done)1083 xfs_ialloc_get_rec(
1084 struct xfs_btree_cur *cur,
1085 xfs_agino_t agino,
1086 xfs_inobt_rec_incore_t *rec,
1087 int *done)
1088 {
1089 int error;
1090 int i;
1091
1092 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i);
1093 if (error)
1094 return error;
1095 *done = !i;
1096 if (i) {
1097 error = xfs_inobt_get_rec(cur, rec, &i);
1098 if (error)
1099 return error;
1100 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1101 }
1102
1103 return 0;
1104 }
1105
1106 /*
1107 * Return the offset of the first free inode in the record. If the inode chunk
1108 * is sparsely allocated, we convert the record holemask to inode granularity
1109 * and mask off the unallocated regions from the inode free mask.
1110 */
1111 STATIC int
xfs_inobt_first_free_inode(struct xfs_inobt_rec_incore * rec)1112 xfs_inobt_first_free_inode(
1113 struct xfs_inobt_rec_incore *rec)
1114 {
1115 xfs_inofree_t realfree;
1116
1117 /* if there are no holes, return the first available offset */
1118 if (!xfs_inobt_issparse(rec->ir_holemask))
1119 return xfs_lowbit64(rec->ir_free);
1120
1121 realfree = xfs_inobt_irec_to_allocmask(rec);
1122 realfree &= rec->ir_free;
1123
1124 return xfs_lowbit64(realfree);
1125 }
1126
1127 /*
1128 * Allocate an inode using the inobt-only algorithm.
1129 */
1130 STATIC int
xfs_dialloc_ag_inobt(struct xfs_trans * tp,struct xfs_buf * agbp,xfs_ino_t parent,xfs_ino_t * inop)1131 xfs_dialloc_ag_inobt(
1132 struct xfs_trans *tp,
1133 struct xfs_buf *agbp,
1134 xfs_ino_t parent,
1135 xfs_ino_t *inop)
1136 {
1137 struct xfs_mount *mp = tp->t_mountp;
1138 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
1139 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
1140 xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent);
1141 xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent);
1142 struct xfs_perag *pag;
1143 struct xfs_btree_cur *cur, *tcur;
1144 struct xfs_inobt_rec_incore rec, trec;
1145 xfs_ino_t ino;
1146 int error;
1147 int offset;
1148 int i, j;
1149 int searchdistance = 10;
1150
1151 pag = xfs_perag_get(mp, agno);
1152
1153 ASSERT(pag->pagi_init);
1154 ASSERT(pag->pagi_inodeok);
1155 ASSERT(pag->pagi_freecount > 0);
1156
1157 restart_pagno:
1158 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1159 /*
1160 * If pagino is 0 (this is the root inode allocation) use newino.
1161 * This must work because we've just allocated some.
1162 */
1163 if (!pagino)
1164 pagino = be32_to_cpu(agi->agi_newino);
1165
1166 error = xfs_check_agi_freecount(cur, agi);
1167 if (error)
1168 goto error0;
1169
1170 /*
1171 * If in the same AG as the parent, try to get near the parent.
1172 */
1173 if (pagno == agno) {
1174 int doneleft; /* done, to the left */
1175 int doneright; /* done, to the right */
1176
1177 error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i);
1178 if (error)
1179 goto error0;
1180 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1181
1182 error = xfs_inobt_get_rec(cur, &rec, &j);
1183 if (error)
1184 goto error0;
1185 XFS_WANT_CORRUPTED_GOTO(mp, j == 1, error0);
1186
1187 if (rec.ir_freecount > 0) {
1188 /*
1189 * Found a free inode in the same chunk
1190 * as the parent, done.
1191 */
1192 goto alloc_inode;
1193 }
1194
1195
1196 /*
1197 * In the same AG as parent, but parent's chunk is full.
1198 */
1199
1200 /* duplicate the cursor, search left & right simultaneously */
1201 error = xfs_btree_dup_cursor(cur, &tcur);
1202 if (error)
1203 goto error0;
1204
1205 /*
1206 * Skip to last blocks looked up if same parent inode.
1207 */
1208 if (pagino != NULLAGINO &&
1209 pag->pagl_pagino == pagino &&
1210 pag->pagl_leftrec != NULLAGINO &&
1211 pag->pagl_rightrec != NULLAGINO) {
1212 error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec,
1213 &trec, &doneleft);
1214 if (error)
1215 goto error1;
1216
1217 error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec,
1218 &rec, &doneright);
1219 if (error)
1220 goto error1;
1221 } else {
1222 /* search left with tcur, back up 1 record */
1223 error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1);
1224 if (error)
1225 goto error1;
1226
1227 /* search right with cur, go forward 1 record. */
1228 error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0);
1229 if (error)
1230 goto error1;
1231 }
1232
1233 /*
1234 * Loop until we find an inode chunk with a free inode.
1235 */
1236 while (--searchdistance > 0 && (!doneleft || !doneright)) {
1237 int useleft; /* using left inode chunk this time */
1238
1239 /* figure out the closer block if both are valid. */
1240 if (!doneleft && !doneright) {
1241 useleft = pagino -
1242 (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) <
1243 rec.ir_startino - pagino;
1244 } else {
1245 useleft = !doneleft;
1246 }
1247
1248 /* free inodes to the left? */
1249 if (useleft && trec.ir_freecount) {
1250 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1251 cur = tcur;
1252
1253 pag->pagl_leftrec = trec.ir_startino;
1254 pag->pagl_rightrec = rec.ir_startino;
1255 pag->pagl_pagino = pagino;
1256 rec = trec;
1257 goto alloc_inode;
1258 }
1259
1260 /* free inodes to the right? */
1261 if (!useleft && rec.ir_freecount) {
1262 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
1263
1264 pag->pagl_leftrec = trec.ir_startino;
1265 pag->pagl_rightrec = rec.ir_startino;
1266 pag->pagl_pagino = pagino;
1267 goto alloc_inode;
1268 }
1269
1270 /* get next record to check */
1271 if (useleft) {
1272 error = xfs_ialloc_next_rec(tcur, &trec,
1273 &doneleft, 1);
1274 } else {
1275 error = xfs_ialloc_next_rec(cur, &rec,
1276 &doneright, 0);
1277 }
1278 if (error)
1279 goto error1;
1280 }
1281
1282 if (searchdistance <= 0) {
1283 /*
1284 * Not in range - save last search
1285 * location and allocate a new inode
1286 */
1287 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
1288 pag->pagl_leftrec = trec.ir_startino;
1289 pag->pagl_rightrec = rec.ir_startino;
1290 pag->pagl_pagino = pagino;
1291
1292 } else {
1293 /*
1294 * We've reached the end of the btree. because
1295 * we are only searching a small chunk of the
1296 * btree each search, there is obviously free
1297 * inodes closer to the parent inode than we
1298 * are now. restart the search again.
1299 */
1300 pag->pagl_pagino = NULLAGINO;
1301 pag->pagl_leftrec = NULLAGINO;
1302 pag->pagl_rightrec = NULLAGINO;
1303 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
1304 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1305 goto restart_pagno;
1306 }
1307 }
1308
1309 /*
1310 * In a different AG from the parent.
1311 * See if the most recently allocated block has any free.
1312 */
1313 if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
1314 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
1315 XFS_LOOKUP_EQ, &i);
1316 if (error)
1317 goto error0;
1318
1319 if (i == 1) {
1320 error = xfs_inobt_get_rec(cur, &rec, &j);
1321 if (error)
1322 goto error0;
1323
1324 if (j == 1 && rec.ir_freecount > 0) {
1325 /*
1326 * The last chunk allocated in the group
1327 * still has a free inode.
1328 */
1329 goto alloc_inode;
1330 }
1331 }
1332 }
1333
1334 /*
1335 * None left in the last group, search the whole AG
1336 */
1337 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
1338 if (error)
1339 goto error0;
1340 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1341
1342 for (;;) {
1343 error = xfs_inobt_get_rec(cur, &rec, &i);
1344 if (error)
1345 goto error0;
1346 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1347 if (rec.ir_freecount > 0)
1348 break;
1349 error = xfs_btree_increment(cur, 0, &i);
1350 if (error)
1351 goto error0;
1352 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1353 }
1354
1355 alloc_inode:
1356 offset = xfs_inobt_first_free_inode(&rec);
1357 ASSERT(offset >= 0);
1358 ASSERT(offset < XFS_INODES_PER_CHUNK);
1359 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
1360 XFS_INODES_PER_CHUNK) == 0);
1361 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
1362 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1363 rec.ir_freecount--;
1364 error = xfs_inobt_update(cur, &rec);
1365 if (error)
1366 goto error0;
1367 be32_add_cpu(&agi->agi_freecount, -1);
1368 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1369 pag->pagi_freecount--;
1370
1371 error = xfs_check_agi_freecount(cur, agi);
1372 if (error)
1373 goto error0;
1374
1375 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1376 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1377 xfs_perag_put(pag);
1378 *inop = ino;
1379 return 0;
1380 error1:
1381 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
1382 error0:
1383 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1384 xfs_perag_put(pag);
1385 return error;
1386 }
1387
1388 /*
1389 * Use the free inode btree to allocate an inode based on distance from the
1390 * parent. Note that the provided cursor may be deleted and replaced.
1391 */
1392 STATIC int
xfs_dialloc_ag_finobt_near(xfs_agino_t pagino,struct xfs_btree_cur ** ocur,struct xfs_inobt_rec_incore * rec)1393 xfs_dialloc_ag_finobt_near(
1394 xfs_agino_t pagino,
1395 struct xfs_btree_cur **ocur,
1396 struct xfs_inobt_rec_incore *rec)
1397 {
1398 struct xfs_btree_cur *lcur = *ocur; /* left search cursor */
1399 struct xfs_btree_cur *rcur; /* right search cursor */
1400 struct xfs_inobt_rec_incore rrec;
1401 int error;
1402 int i, j;
1403
1404 error = xfs_inobt_lookup(lcur, pagino, XFS_LOOKUP_LE, &i);
1405 if (error)
1406 return error;
1407
1408 if (i == 1) {
1409 error = xfs_inobt_get_rec(lcur, rec, &i);
1410 if (error)
1411 return error;
1412 XFS_WANT_CORRUPTED_RETURN(lcur->bc_mp, i == 1);
1413
1414 /*
1415 * See if we've landed in the parent inode record. The finobt
1416 * only tracks chunks with at least one free inode, so record
1417 * existence is enough.
1418 */
1419 if (pagino >= rec->ir_startino &&
1420 pagino < (rec->ir_startino + XFS_INODES_PER_CHUNK))
1421 return 0;
1422 }
1423
1424 error = xfs_btree_dup_cursor(lcur, &rcur);
1425 if (error)
1426 return error;
1427
1428 error = xfs_inobt_lookup(rcur, pagino, XFS_LOOKUP_GE, &j);
1429 if (error)
1430 goto error_rcur;
1431 if (j == 1) {
1432 error = xfs_inobt_get_rec(rcur, &rrec, &j);
1433 if (error)
1434 goto error_rcur;
1435 XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, j == 1, error_rcur);
1436 }
1437
1438 XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, i == 1 || j == 1, error_rcur);
1439 if (i == 1 && j == 1) {
1440 /*
1441 * Both the left and right records are valid. Choose the closer
1442 * inode chunk to the target.
1443 */
1444 if ((pagino - rec->ir_startino + XFS_INODES_PER_CHUNK - 1) >
1445 (rrec.ir_startino - pagino)) {
1446 *rec = rrec;
1447 xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1448 *ocur = rcur;
1449 } else {
1450 xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1451 }
1452 } else if (j == 1) {
1453 /* only the right record is valid */
1454 *rec = rrec;
1455 xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1456 *ocur = rcur;
1457 } else if (i == 1) {
1458 /* only the left record is valid */
1459 xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1460 }
1461
1462 return 0;
1463
1464 error_rcur:
1465 xfs_btree_del_cursor(rcur, XFS_BTREE_ERROR);
1466 return error;
1467 }
1468
1469 /*
1470 * Use the free inode btree to find a free inode based on a newino hint. If
1471 * the hint is NULL, find the first free inode in the AG.
1472 */
1473 STATIC int
xfs_dialloc_ag_finobt_newino(struct xfs_agi * agi,struct xfs_btree_cur * cur,struct xfs_inobt_rec_incore * rec)1474 xfs_dialloc_ag_finobt_newino(
1475 struct xfs_agi *agi,
1476 struct xfs_btree_cur *cur,
1477 struct xfs_inobt_rec_incore *rec)
1478 {
1479 int error;
1480 int i;
1481
1482 if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
1483 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
1484 XFS_LOOKUP_EQ, &i);
1485 if (error)
1486 return error;
1487 if (i == 1) {
1488 error = xfs_inobt_get_rec(cur, rec, &i);
1489 if (error)
1490 return error;
1491 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1492 return 0;
1493 }
1494 }
1495
1496 /*
1497 * Find the first inode available in the AG.
1498 */
1499 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
1500 if (error)
1501 return error;
1502 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1503
1504 error = xfs_inobt_get_rec(cur, rec, &i);
1505 if (error)
1506 return error;
1507 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1508
1509 return 0;
1510 }
1511
1512 /*
1513 * Update the inobt based on a modification made to the finobt. Also ensure that
1514 * the records from both trees are equivalent post-modification.
1515 */
1516 STATIC int
xfs_dialloc_ag_update_inobt(struct xfs_btree_cur * cur,struct xfs_inobt_rec_incore * frec,int offset)1517 xfs_dialloc_ag_update_inobt(
1518 struct xfs_btree_cur *cur, /* inobt cursor */
1519 struct xfs_inobt_rec_incore *frec, /* finobt record */
1520 int offset) /* inode offset */
1521 {
1522 struct xfs_inobt_rec_incore rec;
1523 int error;
1524 int i;
1525
1526 error = xfs_inobt_lookup(cur, frec->ir_startino, XFS_LOOKUP_EQ, &i);
1527 if (error)
1528 return error;
1529 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1530
1531 error = xfs_inobt_get_rec(cur, &rec, &i);
1532 if (error)
1533 return error;
1534 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1535 ASSERT((XFS_AGINO_TO_OFFSET(cur->bc_mp, rec.ir_startino) %
1536 XFS_INODES_PER_CHUNK) == 0);
1537
1538 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1539 rec.ir_freecount--;
1540
1541 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, (rec.ir_free == frec->ir_free) &&
1542 (rec.ir_freecount == frec->ir_freecount));
1543
1544 return xfs_inobt_update(cur, &rec);
1545 }
1546
1547 /*
1548 * Allocate an inode using the free inode btree, if available. Otherwise, fall
1549 * back to the inobt search algorithm.
1550 *
1551 * The caller selected an AG for us, and made sure that free inodes are
1552 * available.
1553 */
1554 STATIC int
xfs_dialloc_ag(struct xfs_trans * tp,struct xfs_buf * agbp,xfs_ino_t parent,xfs_ino_t * inop)1555 xfs_dialloc_ag(
1556 struct xfs_trans *tp,
1557 struct xfs_buf *agbp,
1558 xfs_ino_t parent,
1559 xfs_ino_t *inop)
1560 {
1561 struct xfs_mount *mp = tp->t_mountp;
1562 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
1563 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
1564 xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent);
1565 xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent);
1566 struct xfs_perag *pag;
1567 struct xfs_btree_cur *cur; /* finobt cursor */
1568 struct xfs_btree_cur *icur; /* inobt cursor */
1569 struct xfs_inobt_rec_incore rec;
1570 xfs_ino_t ino;
1571 int error;
1572 int offset;
1573 int i;
1574
1575 if (!xfs_sb_version_hasfinobt(&mp->m_sb))
1576 return xfs_dialloc_ag_inobt(tp, agbp, parent, inop);
1577
1578 pag = xfs_perag_get(mp, agno);
1579
1580 /*
1581 * If pagino is 0 (this is the root inode allocation) use newino.
1582 * This must work because we've just allocated some.
1583 */
1584 if (!pagino)
1585 pagino = be32_to_cpu(agi->agi_newino);
1586
1587 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO);
1588
1589 error = xfs_check_agi_freecount(cur, agi);
1590 if (error)
1591 goto error_cur;
1592
1593 /*
1594 * The search algorithm depends on whether we're in the same AG as the
1595 * parent. If so, find the closest available inode to the parent. If
1596 * not, consider the agi hint or find the first free inode in the AG.
1597 */
1598 if (agno == pagno)
1599 error = xfs_dialloc_ag_finobt_near(pagino, &cur, &rec);
1600 else
1601 error = xfs_dialloc_ag_finobt_newino(agi, cur, &rec);
1602 if (error)
1603 goto error_cur;
1604
1605 offset = xfs_inobt_first_free_inode(&rec);
1606 ASSERT(offset >= 0);
1607 ASSERT(offset < XFS_INODES_PER_CHUNK);
1608 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
1609 XFS_INODES_PER_CHUNK) == 0);
1610 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
1611
1612 /*
1613 * Modify or remove the finobt record.
1614 */
1615 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1616 rec.ir_freecount--;
1617 if (rec.ir_freecount)
1618 error = xfs_inobt_update(cur, &rec);
1619 else
1620 error = xfs_btree_delete(cur, &i);
1621 if (error)
1622 goto error_cur;
1623
1624 /*
1625 * The finobt has now been updated appropriately. We haven't updated the
1626 * agi and superblock yet, so we can create an inobt cursor and validate
1627 * the original freecount. If all is well, make the equivalent update to
1628 * the inobt using the finobt record and offset information.
1629 */
1630 icur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1631
1632 error = xfs_check_agi_freecount(icur, agi);
1633 if (error)
1634 goto error_icur;
1635
1636 error = xfs_dialloc_ag_update_inobt(icur, &rec, offset);
1637 if (error)
1638 goto error_icur;
1639
1640 /*
1641 * Both trees have now been updated. We must update the perag and
1642 * superblock before we can check the freecount for each btree.
1643 */
1644 be32_add_cpu(&agi->agi_freecount, -1);
1645 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1646 pag->pagi_freecount--;
1647
1648 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1649
1650 error = xfs_check_agi_freecount(icur, agi);
1651 if (error)
1652 goto error_icur;
1653 error = xfs_check_agi_freecount(cur, agi);
1654 if (error)
1655 goto error_icur;
1656
1657 xfs_btree_del_cursor(icur, XFS_BTREE_NOERROR);
1658 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1659 xfs_perag_put(pag);
1660 *inop = ino;
1661 return 0;
1662
1663 error_icur:
1664 xfs_btree_del_cursor(icur, XFS_BTREE_ERROR);
1665 error_cur:
1666 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1667 xfs_perag_put(pag);
1668 return error;
1669 }
1670
1671 /*
1672 * Allocate an inode on disk.
1673 *
1674 * Mode is used to tell whether the new inode will need space, and whether it
1675 * is a directory.
1676 *
1677 * This function is designed to be called twice if it has to do an allocation
1678 * to make more free inodes. On the first call, *IO_agbp should be set to NULL.
1679 * If an inode is available without having to performn an allocation, an inode
1680 * number is returned. In this case, *IO_agbp is set to NULL. If an allocation
1681 * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp.
1682 * The caller should then commit the current transaction, allocate a
1683 * new transaction, and call xfs_dialloc() again, passing in the previous value
1684 * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI
1685 * buffer is locked across the two calls, the second call is guaranteed to have
1686 * a free inode available.
1687 *
1688 * Once we successfully pick an inode its number is returned and the on-disk
1689 * data structures are updated. The inode itself is not read in, since doing so
1690 * would break ordering constraints with xfs_reclaim.
1691 */
1692 int
xfs_dialloc(struct xfs_trans * tp,xfs_ino_t parent,umode_t mode,struct xfs_buf ** IO_agbp,xfs_ino_t * inop)1693 xfs_dialloc(
1694 struct xfs_trans *tp,
1695 xfs_ino_t parent,
1696 umode_t mode,
1697 struct xfs_buf **IO_agbp,
1698 xfs_ino_t *inop)
1699 {
1700 struct xfs_mount *mp = tp->t_mountp;
1701 struct xfs_buf *agbp;
1702 xfs_agnumber_t agno;
1703 int error;
1704 int ialloced;
1705 int noroom = 0;
1706 xfs_agnumber_t start_agno;
1707 struct xfs_perag *pag;
1708 int okalloc = 1;
1709
1710 if (*IO_agbp) {
1711 /*
1712 * If the caller passes in a pointer to the AGI buffer,
1713 * continue where we left off before. In this case, we
1714 * know that the allocation group has free inodes.
1715 */
1716 agbp = *IO_agbp;
1717 goto out_alloc;
1718 }
1719
1720 /*
1721 * We do not have an agbp, so select an initial allocation
1722 * group for inode allocation.
1723 */
1724 start_agno = xfs_ialloc_ag_select(tp, parent, mode);
1725 if (start_agno == NULLAGNUMBER) {
1726 *inop = NULLFSINO;
1727 return 0;
1728 }
1729
1730 /*
1731 * If we have already hit the ceiling of inode blocks then clear
1732 * okalloc so we scan all available agi structures for a free
1733 * inode.
1734 *
1735 * Read rough value of mp->m_icount by percpu_counter_read_positive,
1736 * which will sacrifice the preciseness but improve the performance.
1737 */
1738 if (mp->m_maxicount &&
1739 percpu_counter_read_positive(&mp->m_icount) + mp->m_ialloc_inos
1740 > mp->m_maxicount) {
1741 noroom = 1;
1742 okalloc = 0;
1743 }
1744
1745 /*
1746 * Loop until we find an allocation group that either has free inodes
1747 * or in which we can allocate some inodes. Iterate through the
1748 * allocation groups upward, wrapping at the end.
1749 */
1750 agno = start_agno;
1751 for (;;) {
1752 pag = xfs_perag_get(mp, agno);
1753 if (!pag->pagi_inodeok) {
1754 xfs_ialloc_next_ag(mp);
1755 goto nextag;
1756 }
1757
1758 if (!pag->pagi_init) {
1759 error = xfs_ialloc_pagi_init(mp, tp, agno);
1760 if (error)
1761 goto out_error;
1762 }
1763
1764 /*
1765 * Do a first racy fast path check if this AG is usable.
1766 */
1767 if (!pag->pagi_freecount && !okalloc)
1768 goto nextag;
1769
1770 /*
1771 * Then read in the AGI buffer and recheck with the AGI buffer
1772 * lock held.
1773 */
1774 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1775 if (error)
1776 goto out_error;
1777
1778 if (pag->pagi_freecount) {
1779 xfs_perag_put(pag);
1780 goto out_alloc;
1781 }
1782
1783 if (!okalloc)
1784 goto nextag_relse_buffer;
1785
1786
1787 error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced);
1788 if (error) {
1789 xfs_trans_brelse(tp, agbp);
1790
1791 if (error != -ENOSPC)
1792 goto out_error;
1793
1794 xfs_perag_put(pag);
1795 *inop = NULLFSINO;
1796 return 0;
1797 }
1798
1799 if (ialloced) {
1800 /*
1801 * We successfully allocated some inodes, return
1802 * the current context to the caller so that it
1803 * can commit the current transaction and call
1804 * us again where we left off.
1805 */
1806 ASSERT(pag->pagi_freecount > 0);
1807 xfs_perag_put(pag);
1808
1809 *IO_agbp = agbp;
1810 *inop = NULLFSINO;
1811 return 0;
1812 }
1813
1814 nextag_relse_buffer:
1815 xfs_trans_brelse(tp, agbp);
1816 nextag:
1817 xfs_perag_put(pag);
1818 if (++agno == mp->m_sb.sb_agcount)
1819 agno = 0;
1820 if (agno == start_agno) {
1821 *inop = NULLFSINO;
1822 return noroom ? -ENOSPC : 0;
1823 }
1824 }
1825
1826 out_alloc:
1827 *IO_agbp = NULL;
1828 return xfs_dialloc_ag(tp, agbp, parent, inop);
1829 out_error:
1830 xfs_perag_put(pag);
1831 return error;
1832 }
1833
1834 /*
1835 * Free the blocks of an inode chunk. We must consider that the inode chunk
1836 * might be sparse and only free the regions that are allocated as part of the
1837 * chunk.
1838 */
1839 STATIC void
xfs_difree_inode_chunk(struct xfs_trans * tp,xfs_agnumber_t agno,struct xfs_inobt_rec_incore * rec)1840 xfs_difree_inode_chunk(
1841 struct xfs_trans *tp,
1842 xfs_agnumber_t agno,
1843 struct xfs_inobt_rec_incore *rec)
1844 {
1845 struct xfs_mount *mp = tp->t_mountp;
1846 xfs_agblock_t sagbno = XFS_AGINO_TO_AGBNO(mp,
1847 rec->ir_startino);
1848 int startidx, endidx;
1849 int nextbit;
1850 xfs_agblock_t agbno;
1851 int contigblk;
1852 struct xfs_owner_info oinfo;
1853 DECLARE_BITMAP(holemask, XFS_INOBT_HOLEMASK_BITS);
1854 xfs_rmap_ag_owner(&oinfo, XFS_RMAP_OWN_INODES);
1855
1856 if (!xfs_inobt_issparse(rec->ir_holemask)) {
1857 /* not sparse, calculate extent info directly */
1858 xfs_bmap_add_free(tp, XFS_AGB_TO_FSB(mp, agno, sagbno),
1859 mp->m_ialloc_blks, &oinfo);
1860 return;
1861 }
1862
1863 /* holemask is only 16-bits (fits in an unsigned long) */
1864 ASSERT(sizeof(rec->ir_holemask) <= sizeof(holemask[0]));
1865 holemask[0] = rec->ir_holemask;
1866
1867 /*
1868 * Find contiguous ranges of zeroes (i.e., allocated regions) in the
1869 * holemask and convert the start/end index of each range to an extent.
1870 * We start with the start and end index both pointing at the first 0 in
1871 * the mask.
1872 */
1873 startidx = endidx = find_first_zero_bit(holemask,
1874 XFS_INOBT_HOLEMASK_BITS);
1875 nextbit = startidx + 1;
1876 while (startidx < XFS_INOBT_HOLEMASK_BITS) {
1877 nextbit = find_next_zero_bit(holemask, XFS_INOBT_HOLEMASK_BITS,
1878 nextbit);
1879 /*
1880 * If the next zero bit is contiguous, update the end index of
1881 * the current range and continue.
1882 */
1883 if (nextbit != XFS_INOBT_HOLEMASK_BITS &&
1884 nextbit == endidx + 1) {
1885 endidx = nextbit;
1886 goto next;
1887 }
1888
1889 /*
1890 * nextbit is not contiguous with the current end index. Convert
1891 * the current start/end to an extent and add it to the free
1892 * list.
1893 */
1894 agbno = sagbno + (startidx * XFS_INODES_PER_HOLEMASK_BIT) /
1895 mp->m_sb.sb_inopblock;
1896 contigblk = ((endidx - startidx + 1) *
1897 XFS_INODES_PER_HOLEMASK_BIT) /
1898 mp->m_sb.sb_inopblock;
1899
1900 ASSERT(agbno % mp->m_sb.sb_spino_align == 0);
1901 ASSERT(contigblk % mp->m_sb.sb_spino_align == 0);
1902 xfs_bmap_add_free(tp, XFS_AGB_TO_FSB(mp, agno, agbno),
1903 contigblk, &oinfo);
1904
1905 /* reset range to current bit and carry on... */
1906 startidx = endidx = nextbit;
1907
1908 next:
1909 nextbit++;
1910 }
1911 }
1912
1913 STATIC int
xfs_difree_inobt(struct xfs_mount * mp,struct xfs_trans * tp,struct xfs_buf * agbp,xfs_agino_t agino,struct xfs_icluster * xic,struct xfs_inobt_rec_incore * orec)1914 xfs_difree_inobt(
1915 struct xfs_mount *mp,
1916 struct xfs_trans *tp,
1917 struct xfs_buf *agbp,
1918 xfs_agino_t agino,
1919 struct xfs_icluster *xic,
1920 struct xfs_inobt_rec_incore *orec)
1921 {
1922 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
1923 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
1924 struct xfs_perag *pag;
1925 struct xfs_btree_cur *cur;
1926 struct xfs_inobt_rec_incore rec;
1927 int ilen;
1928 int error;
1929 int i;
1930 int off;
1931
1932 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
1933 ASSERT(XFS_AGINO_TO_AGBNO(mp, agino) < be32_to_cpu(agi->agi_length));
1934
1935 /*
1936 * Initialize the cursor.
1937 */
1938 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1939
1940 error = xfs_check_agi_freecount(cur, agi);
1941 if (error)
1942 goto error0;
1943
1944 /*
1945 * Look for the entry describing this inode.
1946 */
1947 if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) {
1948 xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.",
1949 __func__, error);
1950 goto error0;
1951 }
1952 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1953 error = xfs_inobt_get_rec(cur, &rec, &i);
1954 if (error) {
1955 xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.",
1956 __func__, error);
1957 goto error0;
1958 }
1959 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1960 /*
1961 * Get the offset in the inode chunk.
1962 */
1963 off = agino - rec.ir_startino;
1964 ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK);
1965 ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off)));
1966 /*
1967 * Mark the inode free & increment the count.
1968 */
1969 rec.ir_free |= XFS_INOBT_MASK(off);
1970 rec.ir_freecount++;
1971
1972 /*
1973 * When an inode chunk is free, it becomes eligible for removal. Don't
1974 * remove the chunk if the block size is large enough for multiple inode
1975 * chunks (that might not be free).
1976 */
1977 if (!(mp->m_flags & XFS_MOUNT_IKEEP) &&
1978 rec.ir_free == XFS_INOBT_ALL_FREE &&
1979 mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK) {
1980 xic->deleted = true;
1981 xic->first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino);
1982 xic->alloc = xfs_inobt_irec_to_allocmask(&rec);
1983
1984 /*
1985 * Remove the inode cluster from the AGI B+Tree, adjust the
1986 * AGI and Superblock inode counts, and mark the disk space
1987 * to be freed when the transaction is committed.
1988 */
1989 ilen = rec.ir_freecount;
1990 be32_add_cpu(&agi->agi_count, -ilen);
1991 be32_add_cpu(&agi->agi_freecount, -(ilen - 1));
1992 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
1993 pag = xfs_perag_get(mp, agno);
1994 pag->pagi_freecount -= ilen - 1;
1995 pag->pagi_count -= ilen;
1996 xfs_perag_put(pag);
1997 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen);
1998 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1));
1999
2000 if ((error = xfs_btree_delete(cur, &i))) {
2001 xfs_warn(mp, "%s: xfs_btree_delete returned error %d.",
2002 __func__, error);
2003 goto error0;
2004 }
2005
2006 xfs_difree_inode_chunk(tp, agno, &rec);
2007 } else {
2008 xic->deleted = false;
2009
2010 error = xfs_inobt_update(cur, &rec);
2011 if (error) {
2012 xfs_warn(mp, "%s: xfs_inobt_update returned error %d.",
2013 __func__, error);
2014 goto error0;
2015 }
2016
2017 /*
2018 * Change the inode free counts and log the ag/sb changes.
2019 */
2020 be32_add_cpu(&agi->agi_freecount, 1);
2021 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
2022 pag = xfs_perag_get(mp, agno);
2023 pag->pagi_freecount++;
2024 xfs_perag_put(pag);
2025 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
2026 }
2027
2028 error = xfs_check_agi_freecount(cur, agi);
2029 if (error)
2030 goto error0;
2031
2032 *orec = rec;
2033 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
2034 return 0;
2035
2036 error0:
2037 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
2038 return error;
2039 }
2040
2041 /*
2042 * Free an inode in the free inode btree.
2043 */
2044 STATIC int
xfs_difree_finobt(struct xfs_mount * mp,struct xfs_trans * tp,struct xfs_buf * agbp,xfs_agino_t agino,struct xfs_inobt_rec_incore * ibtrec)2045 xfs_difree_finobt(
2046 struct xfs_mount *mp,
2047 struct xfs_trans *tp,
2048 struct xfs_buf *agbp,
2049 xfs_agino_t agino,
2050 struct xfs_inobt_rec_incore *ibtrec) /* inobt record */
2051 {
2052 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
2053 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
2054 struct xfs_btree_cur *cur;
2055 struct xfs_inobt_rec_incore rec;
2056 int offset = agino - ibtrec->ir_startino;
2057 int error;
2058 int i;
2059
2060 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO);
2061
2062 error = xfs_inobt_lookup(cur, ibtrec->ir_startino, XFS_LOOKUP_EQ, &i);
2063 if (error)
2064 goto error;
2065 if (i == 0) {
2066 /*
2067 * If the record does not exist in the finobt, we must have just
2068 * freed an inode in a previously fully allocated chunk. If not,
2069 * something is out of sync.
2070 */
2071 XFS_WANT_CORRUPTED_GOTO(mp, ibtrec->ir_freecount == 1, error);
2072
2073 error = xfs_inobt_insert_rec(cur, ibtrec->ir_holemask,
2074 ibtrec->ir_count,
2075 ibtrec->ir_freecount,
2076 ibtrec->ir_free, &i);
2077 if (error)
2078 goto error;
2079 ASSERT(i == 1);
2080
2081 goto out;
2082 }
2083
2084 /*
2085 * Read and update the existing record. We could just copy the ibtrec
2086 * across here, but that would defeat the purpose of having redundant
2087 * metadata. By making the modifications independently, we can catch
2088 * corruptions that we wouldn't see if we just copied from one record
2089 * to another.
2090 */
2091 error = xfs_inobt_get_rec(cur, &rec, &i);
2092 if (error)
2093 goto error;
2094 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error);
2095
2096 rec.ir_free |= XFS_INOBT_MASK(offset);
2097 rec.ir_freecount++;
2098
2099 XFS_WANT_CORRUPTED_GOTO(mp, (rec.ir_free == ibtrec->ir_free) &&
2100 (rec.ir_freecount == ibtrec->ir_freecount),
2101 error);
2102
2103 /*
2104 * The content of inobt records should always match between the inobt
2105 * and finobt. The lifecycle of records in the finobt is different from
2106 * the inobt in that the finobt only tracks records with at least one
2107 * free inode. Hence, if all of the inodes are free and we aren't
2108 * keeping inode chunks permanently on disk, remove the record.
2109 * Otherwise, update the record with the new information.
2110 *
2111 * Note that we currently can't free chunks when the block size is large
2112 * enough for multiple chunks. Leave the finobt record to remain in sync
2113 * with the inobt.
2114 */
2115 if (rec.ir_free == XFS_INOBT_ALL_FREE &&
2116 mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK &&
2117 !(mp->m_flags & XFS_MOUNT_IKEEP)) {
2118 error = xfs_btree_delete(cur, &i);
2119 if (error)
2120 goto error;
2121 ASSERT(i == 1);
2122 } else {
2123 error = xfs_inobt_update(cur, &rec);
2124 if (error)
2125 goto error;
2126 }
2127
2128 out:
2129 error = xfs_check_agi_freecount(cur, agi);
2130 if (error)
2131 goto error;
2132
2133 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
2134 return 0;
2135
2136 error:
2137 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
2138 return error;
2139 }
2140
2141 /*
2142 * Free disk inode. Carefully avoids touching the incore inode, all
2143 * manipulations incore are the caller's responsibility.
2144 * The on-disk inode is not changed by this operation, only the
2145 * btree (free inode mask) is changed.
2146 */
2147 int
xfs_difree(struct xfs_trans * tp,xfs_ino_t inode,struct xfs_icluster * xic)2148 xfs_difree(
2149 struct xfs_trans *tp, /* transaction pointer */
2150 xfs_ino_t inode, /* inode to be freed */
2151 struct xfs_icluster *xic) /* cluster info if deleted */
2152 {
2153 /* REFERENCED */
2154 xfs_agblock_t agbno; /* block number containing inode */
2155 struct xfs_buf *agbp; /* buffer for allocation group header */
2156 xfs_agino_t agino; /* allocation group inode number */
2157 xfs_agnumber_t agno; /* allocation group number */
2158 int error; /* error return value */
2159 struct xfs_mount *mp; /* mount structure for filesystem */
2160 struct xfs_inobt_rec_incore rec;/* btree record */
2161
2162 mp = tp->t_mountp;
2163
2164 /*
2165 * Break up inode number into its components.
2166 */
2167 agno = XFS_INO_TO_AGNO(mp, inode);
2168 if (agno >= mp->m_sb.sb_agcount) {
2169 xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).",
2170 __func__, agno, mp->m_sb.sb_agcount);
2171 ASSERT(0);
2172 return -EINVAL;
2173 }
2174 agino = XFS_INO_TO_AGINO(mp, inode);
2175 if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) {
2176 xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).",
2177 __func__, (unsigned long long)inode,
2178 (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino));
2179 ASSERT(0);
2180 return -EINVAL;
2181 }
2182 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
2183 if (agbno >= mp->m_sb.sb_agblocks) {
2184 xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).",
2185 __func__, agbno, mp->m_sb.sb_agblocks);
2186 ASSERT(0);
2187 return -EINVAL;
2188 }
2189 /*
2190 * Get the allocation group header.
2191 */
2192 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
2193 if (error) {
2194 xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.",
2195 __func__, error);
2196 return error;
2197 }
2198
2199 /*
2200 * Fix up the inode allocation btree.
2201 */
2202 error = xfs_difree_inobt(mp, tp, agbp, agino, xic, &rec);
2203 if (error)
2204 goto error0;
2205
2206 /*
2207 * Fix up the free inode btree.
2208 */
2209 if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
2210 error = xfs_difree_finobt(mp, tp, agbp, agino, &rec);
2211 if (error)
2212 goto error0;
2213 }
2214
2215 return 0;
2216
2217 error0:
2218 return error;
2219 }
2220
2221 STATIC int
xfs_imap_lookup(struct xfs_mount * mp,struct xfs_trans * tp,xfs_agnumber_t agno,xfs_agino_t agino,xfs_agblock_t agbno,xfs_agblock_t * chunk_agbno,xfs_agblock_t * offset_agbno,int flags)2222 xfs_imap_lookup(
2223 struct xfs_mount *mp,
2224 struct xfs_trans *tp,
2225 xfs_agnumber_t agno,
2226 xfs_agino_t agino,
2227 xfs_agblock_t agbno,
2228 xfs_agblock_t *chunk_agbno,
2229 xfs_agblock_t *offset_agbno,
2230 int flags)
2231 {
2232 struct xfs_inobt_rec_incore rec;
2233 struct xfs_btree_cur *cur;
2234 struct xfs_buf *agbp;
2235 int error;
2236 int i;
2237
2238 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
2239 if (error) {
2240 xfs_alert(mp,
2241 "%s: xfs_ialloc_read_agi() returned error %d, agno %d",
2242 __func__, error, agno);
2243 return error;
2244 }
2245
2246 /*
2247 * Lookup the inode record for the given agino. If the record cannot be
2248 * found, then it's an invalid inode number and we should abort. Once
2249 * we have a record, we need to ensure it contains the inode number
2250 * we are looking up.
2251 */
2252 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
2253 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i);
2254 if (!error) {
2255 if (i)
2256 error = xfs_inobt_get_rec(cur, &rec, &i);
2257 if (!error && i == 0)
2258 error = -EINVAL;
2259 }
2260
2261 xfs_trans_brelse(tp, agbp);
2262 xfs_btree_del_cursor(cur, error);
2263 if (error)
2264 return error;
2265
2266 /* check that the returned record contains the required inode */
2267 if (rec.ir_startino > agino ||
2268 rec.ir_startino + mp->m_ialloc_inos <= agino)
2269 return -EINVAL;
2270
2271 /* for untrusted inodes check it is allocated first */
2272 if ((flags & XFS_IGET_UNTRUSTED) &&
2273 (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino)))
2274 return -EINVAL;
2275
2276 *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino);
2277 *offset_agbno = agbno - *chunk_agbno;
2278 return 0;
2279 }
2280
2281 /*
2282 * Return the location of the inode in imap, for mapping it into a buffer.
2283 */
2284 int
xfs_imap(xfs_mount_t * mp,xfs_trans_t * tp,xfs_ino_t ino,struct xfs_imap * imap,uint flags)2285 xfs_imap(
2286 xfs_mount_t *mp, /* file system mount structure */
2287 xfs_trans_t *tp, /* transaction pointer */
2288 xfs_ino_t ino, /* inode to locate */
2289 struct xfs_imap *imap, /* location map structure */
2290 uint flags) /* flags for inode btree lookup */
2291 {
2292 xfs_agblock_t agbno; /* block number of inode in the alloc group */
2293 xfs_agino_t agino; /* inode number within alloc group */
2294 xfs_agnumber_t agno; /* allocation group number */
2295 int blks_per_cluster; /* num blocks per inode cluster */
2296 xfs_agblock_t chunk_agbno; /* first block in inode chunk */
2297 xfs_agblock_t cluster_agbno; /* first block in inode cluster */
2298 int error; /* error code */
2299 int offset; /* index of inode in its buffer */
2300 xfs_agblock_t offset_agbno; /* blks from chunk start to inode */
2301
2302 ASSERT(ino != NULLFSINO);
2303
2304 /*
2305 * Split up the inode number into its parts.
2306 */
2307 agno = XFS_INO_TO_AGNO(mp, ino);
2308 agino = XFS_INO_TO_AGINO(mp, ino);
2309 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
2310 if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks ||
2311 ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
2312 #ifdef DEBUG
2313 /*
2314 * Don't output diagnostic information for untrusted inodes
2315 * as they can be invalid without implying corruption.
2316 */
2317 if (flags & XFS_IGET_UNTRUSTED)
2318 return -EINVAL;
2319 if (agno >= mp->m_sb.sb_agcount) {
2320 xfs_alert(mp,
2321 "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)",
2322 __func__, agno, mp->m_sb.sb_agcount);
2323 }
2324 if (agbno >= mp->m_sb.sb_agblocks) {
2325 xfs_alert(mp,
2326 "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)",
2327 __func__, (unsigned long long)agbno,
2328 (unsigned long)mp->m_sb.sb_agblocks);
2329 }
2330 if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
2331 xfs_alert(mp,
2332 "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)",
2333 __func__, ino,
2334 XFS_AGINO_TO_INO(mp, agno, agino));
2335 }
2336 xfs_stack_trace();
2337 #endif /* DEBUG */
2338 return -EINVAL;
2339 }
2340
2341 blks_per_cluster = xfs_icluster_size_fsb(mp);
2342
2343 /*
2344 * For bulkstat and handle lookups, we have an untrusted inode number
2345 * that we have to verify is valid. We cannot do this just by reading
2346 * the inode buffer as it may have been unlinked and removed leaving
2347 * inodes in stale state on disk. Hence we have to do a btree lookup
2348 * in all cases where an untrusted inode number is passed.
2349 */
2350 if (flags & XFS_IGET_UNTRUSTED) {
2351 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
2352 &chunk_agbno, &offset_agbno, flags);
2353 if (error)
2354 return error;
2355 goto out_map;
2356 }
2357
2358 /*
2359 * If the inode cluster size is the same as the blocksize or
2360 * smaller we get to the buffer by simple arithmetics.
2361 */
2362 if (blks_per_cluster == 1) {
2363 offset = XFS_INO_TO_OFFSET(mp, ino);
2364 ASSERT(offset < mp->m_sb.sb_inopblock);
2365
2366 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno);
2367 imap->im_len = XFS_FSB_TO_BB(mp, 1);
2368 imap->im_boffset = (unsigned short)(offset <<
2369 mp->m_sb.sb_inodelog);
2370 return 0;
2371 }
2372
2373 /*
2374 * If the inode chunks are aligned then use simple maths to
2375 * find the location. Otherwise we have to do a btree
2376 * lookup to find the location.
2377 */
2378 if (mp->m_inoalign_mask) {
2379 offset_agbno = agbno & mp->m_inoalign_mask;
2380 chunk_agbno = agbno - offset_agbno;
2381 } else {
2382 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
2383 &chunk_agbno, &offset_agbno, flags);
2384 if (error)
2385 return error;
2386 }
2387
2388 out_map:
2389 ASSERT(agbno >= chunk_agbno);
2390 cluster_agbno = chunk_agbno +
2391 ((offset_agbno / blks_per_cluster) * blks_per_cluster);
2392 offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
2393 XFS_INO_TO_OFFSET(mp, ino);
2394
2395 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno);
2396 imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster);
2397 imap->im_boffset = (unsigned short)(offset << mp->m_sb.sb_inodelog);
2398
2399 /*
2400 * If the inode number maps to a block outside the bounds
2401 * of the file system then return NULL rather than calling
2402 * read_buf and panicing when we get an error from the
2403 * driver.
2404 */
2405 if ((imap->im_blkno + imap->im_len) >
2406 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
2407 xfs_alert(mp,
2408 "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)",
2409 __func__, (unsigned long long) imap->im_blkno,
2410 (unsigned long long) imap->im_len,
2411 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
2412 return -EINVAL;
2413 }
2414 return 0;
2415 }
2416
2417 /*
2418 * Compute and fill in value of m_in_maxlevels.
2419 */
2420 void
xfs_ialloc_compute_maxlevels(xfs_mount_t * mp)2421 xfs_ialloc_compute_maxlevels(
2422 xfs_mount_t *mp) /* file system mount structure */
2423 {
2424 uint inodes;
2425
2426 inodes = (1LL << XFS_INO_AGINO_BITS(mp)) >> XFS_INODES_PER_CHUNK_LOG;
2427 mp->m_in_maxlevels = xfs_btree_compute_maxlevels(mp->m_inobt_mnr,
2428 inodes);
2429 }
2430
2431 /*
2432 * Log specified fields for the ag hdr (inode section). The growth of the agi
2433 * structure over time requires that we interpret the buffer as two logical
2434 * regions delineated by the end of the unlinked list. This is due to the size
2435 * of the hash table and its location in the middle of the agi.
2436 *
2437 * For example, a request to log a field before agi_unlinked and a field after
2438 * agi_unlinked could cause us to log the entire hash table and use an excessive
2439 * amount of log space. To avoid this behavior, log the region up through
2440 * agi_unlinked in one call and the region after agi_unlinked through the end of
2441 * the structure in another.
2442 */
2443 void
xfs_ialloc_log_agi(xfs_trans_t * tp,xfs_buf_t * bp,int fields)2444 xfs_ialloc_log_agi(
2445 xfs_trans_t *tp, /* transaction pointer */
2446 xfs_buf_t *bp, /* allocation group header buffer */
2447 int fields) /* bitmask of fields to log */
2448 {
2449 int first; /* first byte number */
2450 int last; /* last byte number */
2451 static const short offsets[] = { /* field starting offsets */
2452 /* keep in sync with bit definitions */
2453 offsetof(xfs_agi_t, agi_magicnum),
2454 offsetof(xfs_agi_t, agi_versionnum),
2455 offsetof(xfs_agi_t, agi_seqno),
2456 offsetof(xfs_agi_t, agi_length),
2457 offsetof(xfs_agi_t, agi_count),
2458 offsetof(xfs_agi_t, agi_root),
2459 offsetof(xfs_agi_t, agi_level),
2460 offsetof(xfs_agi_t, agi_freecount),
2461 offsetof(xfs_agi_t, agi_newino),
2462 offsetof(xfs_agi_t, agi_dirino),
2463 offsetof(xfs_agi_t, agi_unlinked),
2464 offsetof(xfs_agi_t, agi_free_root),
2465 offsetof(xfs_agi_t, agi_free_level),
2466 sizeof(xfs_agi_t)
2467 };
2468 #ifdef DEBUG
2469 xfs_agi_t *agi; /* allocation group header */
2470
2471 agi = XFS_BUF_TO_AGI(bp);
2472 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
2473 #endif
2474
2475 /*
2476 * Compute byte offsets for the first and last fields in the first
2477 * region and log the agi buffer. This only logs up through
2478 * agi_unlinked.
2479 */
2480 if (fields & XFS_AGI_ALL_BITS_R1) {
2481 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R1,
2482 &first, &last);
2483 xfs_trans_log_buf(tp, bp, first, last);
2484 }
2485
2486 /*
2487 * Mask off the bits in the first region and calculate the first and
2488 * last field offsets for any bits in the second region.
2489 */
2490 fields &= ~XFS_AGI_ALL_BITS_R1;
2491 if (fields) {
2492 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R2,
2493 &first, &last);
2494 xfs_trans_log_buf(tp, bp, first, last);
2495 }
2496 }
2497
2498 static xfs_failaddr_t
xfs_agi_verify(struct xfs_buf * bp)2499 xfs_agi_verify(
2500 struct xfs_buf *bp)
2501 {
2502 struct xfs_mount *mp = bp->b_target->bt_mount;
2503 struct xfs_agi *agi = XFS_BUF_TO_AGI(bp);
2504 int i;
2505
2506 if (xfs_sb_version_hascrc(&mp->m_sb)) {
2507 if (!uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid))
2508 return __this_address;
2509 if (!xfs_log_check_lsn(mp,
2510 be64_to_cpu(XFS_BUF_TO_AGI(bp)->agi_lsn)))
2511 return __this_address;
2512 }
2513
2514 /*
2515 * Validate the magic number of the agi block.
2516 */
2517 if (agi->agi_magicnum != cpu_to_be32(XFS_AGI_MAGIC))
2518 return __this_address;
2519 if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)))
2520 return __this_address;
2521
2522 if (be32_to_cpu(agi->agi_level) < 1 ||
2523 be32_to_cpu(agi->agi_level) > XFS_BTREE_MAXLEVELS)
2524 return __this_address;
2525
2526 if (xfs_sb_version_hasfinobt(&mp->m_sb) &&
2527 (be32_to_cpu(agi->agi_free_level) < 1 ||
2528 be32_to_cpu(agi->agi_free_level) > XFS_BTREE_MAXLEVELS))
2529 return __this_address;
2530
2531 /*
2532 * during growfs operations, the perag is not fully initialised,
2533 * so we can't use it for any useful checking. growfs ensures we can't
2534 * use it by using uncached buffers that don't have the perag attached
2535 * so we can detect and avoid this problem.
2536 */
2537 if (bp->b_pag && be32_to_cpu(agi->agi_seqno) != bp->b_pag->pag_agno)
2538 return __this_address;
2539
2540 for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++) {
2541 if (agi->agi_unlinked[i] == cpu_to_be32(NULLAGINO))
2542 continue;
2543 if (!xfs_verify_ino(mp, be32_to_cpu(agi->agi_unlinked[i])))
2544 return __this_address;
2545 }
2546
2547 return NULL;
2548 }
2549
2550 static void
xfs_agi_read_verify(struct xfs_buf * bp)2551 xfs_agi_read_verify(
2552 struct xfs_buf *bp)
2553 {
2554 struct xfs_mount *mp = bp->b_target->bt_mount;
2555 xfs_failaddr_t fa;
2556
2557 if (xfs_sb_version_hascrc(&mp->m_sb) &&
2558 !xfs_buf_verify_cksum(bp, XFS_AGI_CRC_OFF))
2559 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
2560 else {
2561 fa = xfs_agi_verify(bp);
2562 if (XFS_TEST_ERROR(fa, mp, XFS_ERRTAG_IALLOC_READ_AGI))
2563 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
2564 }
2565 }
2566
2567 static void
xfs_agi_write_verify(struct xfs_buf * bp)2568 xfs_agi_write_verify(
2569 struct xfs_buf *bp)
2570 {
2571 struct xfs_mount *mp = bp->b_target->bt_mount;
2572 struct xfs_buf_log_item *bip = bp->b_log_item;
2573 xfs_failaddr_t fa;
2574
2575 fa = xfs_agi_verify(bp);
2576 if (fa) {
2577 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
2578 return;
2579 }
2580
2581 if (!xfs_sb_version_hascrc(&mp->m_sb))
2582 return;
2583
2584 if (bip)
2585 XFS_BUF_TO_AGI(bp)->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn);
2586 xfs_buf_update_cksum(bp, XFS_AGI_CRC_OFF);
2587 }
2588
2589 const struct xfs_buf_ops xfs_agi_buf_ops = {
2590 .name = "xfs_agi",
2591 .verify_read = xfs_agi_read_verify,
2592 .verify_write = xfs_agi_write_verify,
2593 .verify_struct = xfs_agi_verify,
2594 };
2595
2596 /*
2597 * Read in the allocation group header (inode allocation section)
2598 */
2599 int
xfs_read_agi(struct xfs_mount * mp,struct xfs_trans * tp,xfs_agnumber_t agno,struct xfs_buf ** bpp)2600 xfs_read_agi(
2601 struct xfs_mount *mp, /* file system mount structure */
2602 struct xfs_trans *tp, /* transaction pointer */
2603 xfs_agnumber_t agno, /* allocation group number */
2604 struct xfs_buf **bpp) /* allocation group hdr buf */
2605 {
2606 int error;
2607
2608 trace_xfs_read_agi(mp, agno);
2609
2610 ASSERT(agno != NULLAGNUMBER);
2611 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
2612 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
2613 XFS_FSS_TO_BB(mp, 1), 0, bpp, &xfs_agi_buf_ops);
2614 if (error)
2615 return error;
2616 if (tp)
2617 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_AGI_BUF);
2618
2619 xfs_buf_set_ref(*bpp, XFS_AGI_REF);
2620 return 0;
2621 }
2622
2623 int
xfs_ialloc_read_agi(struct xfs_mount * mp,struct xfs_trans * tp,xfs_agnumber_t agno,struct xfs_buf ** bpp)2624 xfs_ialloc_read_agi(
2625 struct xfs_mount *mp, /* file system mount structure */
2626 struct xfs_trans *tp, /* transaction pointer */
2627 xfs_agnumber_t agno, /* allocation group number */
2628 struct xfs_buf **bpp) /* allocation group hdr buf */
2629 {
2630 struct xfs_agi *agi; /* allocation group header */
2631 struct xfs_perag *pag; /* per allocation group data */
2632 int error;
2633
2634 trace_xfs_ialloc_read_agi(mp, agno);
2635
2636 error = xfs_read_agi(mp, tp, agno, bpp);
2637 if (error)
2638 return error;
2639
2640 agi = XFS_BUF_TO_AGI(*bpp);
2641 pag = xfs_perag_get(mp, agno);
2642 if (!pag->pagi_init) {
2643 pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
2644 pag->pagi_count = be32_to_cpu(agi->agi_count);
2645 pag->pagi_init = 1;
2646 }
2647
2648 /*
2649 * It's possible for these to be out of sync if
2650 * we are in the middle of a forced shutdown.
2651 */
2652 ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) ||
2653 XFS_FORCED_SHUTDOWN(mp));
2654 xfs_perag_put(pag);
2655 return 0;
2656 }
2657
2658 /*
2659 * Read in the agi to initialise the per-ag data in the mount structure
2660 */
2661 int
xfs_ialloc_pagi_init(xfs_mount_t * mp,xfs_trans_t * tp,xfs_agnumber_t agno)2662 xfs_ialloc_pagi_init(
2663 xfs_mount_t *mp, /* file system mount structure */
2664 xfs_trans_t *tp, /* transaction pointer */
2665 xfs_agnumber_t agno) /* allocation group number */
2666 {
2667 xfs_buf_t *bp = NULL;
2668 int error;
2669
2670 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
2671 if (error)
2672 return error;
2673 if (bp)
2674 xfs_trans_brelse(tp, bp);
2675 return 0;
2676 }
2677
2678 /* Is there an inode record covering a given range of inode numbers? */
2679 int
xfs_ialloc_has_inode_record(struct xfs_btree_cur * cur,xfs_agino_t low,xfs_agino_t high,bool * exists)2680 xfs_ialloc_has_inode_record(
2681 struct xfs_btree_cur *cur,
2682 xfs_agino_t low,
2683 xfs_agino_t high,
2684 bool *exists)
2685 {
2686 struct xfs_inobt_rec_incore irec;
2687 xfs_agino_t agino;
2688 uint16_t holemask;
2689 int has_record;
2690 int i;
2691 int error;
2692
2693 *exists = false;
2694 error = xfs_inobt_lookup(cur, low, XFS_LOOKUP_LE, &has_record);
2695 while (error == 0 && has_record) {
2696 error = xfs_inobt_get_rec(cur, &irec, &has_record);
2697 if (error || irec.ir_startino > high)
2698 break;
2699
2700 agino = irec.ir_startino;
2701 holemask = irec.ir_holemask;
2702 for (i = 0; i < XFS_INOBT_HOLEMASK_BITS; holemask >>= 1,
2703 i++, agino += XFS_INODES_PER_HOLEMASK_BIT) {
2704 if (holemask & 1)
2705 continue;
2706 if (agino + XFS_INODES_PER_HOLEMASK_BIT > low &&
2707 agino <= high) {
2708 *exists = true;
2709 return 0;
2710 }
2711 }
2712
2713 error = xfs_btree_increment(cur, 0, &has_record);
2714 }
2715 return error;
2716 }
2717
2718 /* Is there an inode record covering a given extent? */
2719 int
xfs_ialloc_has_inodes_at_extent(struct xfs_btree_cur * cur,xfs_agblock_t bno,xfs_extlen_t len,bool * exists)2720 xfs_ialloc_has_inodes_at_extent(
2721 struct xfs_btree_cur *cur,
2722 xfs_agblock_t bno,
2723 xfs_extlen_t len,
2724 bool *exists)
2725 {
2726 xfs_agino_t low;
2727 xfs_agino_t high;
2728
2729 low = XFS_OFFBNO_TO_AGINO(cur->bc_mp, bno, 0);
2730 high = XFS_OFFBNO_TO_AGINO(cur->bc_mp, bno + len, 0) - 1;
2731
2732 return xfs_ialloc_has_inode_record(cur, low, high, exists);
2733 }
2734
2735 struct xfs_ialloc_count_inodes {
2736 xfs_agino_t count;
2737 xfs_agino_t freecount;
2738 };
2739
2740 /* Record inode counts across all inobt records. */
2741 STATIC int
xfs_ialloc_count_inodes_rec(struct xfs_btree_cur * cur,union xfs_btree_rec * rec,void * priv)2742 xfs_ialloc_count_inodes_rec(
2743 struct xfs_btree_cur *cur,
2744 union xfs_btree_rec *rec,
2745 void *priv)
2746 {
2747 struct xfs_inobt_rec_incore irec;
2748 struct xfs_ialloc_count_inodes *ci = priv;
2749
2750 xfs_inobt_btrec_to_irec(cur->bc_mp, rec, &irec);
2751 ci->count += irec.ir_count;
2752 ci->freecount += irec.ir_freecount;
2753
2754 return 0;
2755 }
2756
2757 /* Count allocated and free inodes under an inobt. */
2758 int
xfs_ialloc_count_inodes(struct xfs_btree_cur * cur,xfs_agino_t * count,xfs_agino_t * freecount)2759 xfs_ialloc_count_inodes(
2760 struct xfs_btree_cur *cur,
2761 xfs_agino_t *count,
2762 xfs_agino_t *freecount)
2763 {
2764 struct xfs_ialloc_count_inodes ci = {0};
2765 int error;
2766
2767 ASSERT(cur->bc_btnum == XFS_BTNUM_INO);
2768 error = xfs_btree_query_all(cur, xfs_ialloc_count_inodes_rec, &ci);
2769 if (error)
2770 return error;
2771
2772 *count = ci.count;
2773 *freecount = ci.freecount;
2774 return 0;
2775 }
2776