]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
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 | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
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. | |
1da177e4 | 13 | * |
7b718769 NS |
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 | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_fs.h" |
70a9883c | 20 | #include "xfs_shared.h" |
239880ef DC |
21 | #include "xfs_format.h" |
22 | #include "xfs_log_format.h" | |
23 | #include "xfs_trans_resv.h" | |
a844f451 | 24 | #include "xfs_bit.h" |
1da177e4 | 25 | #include "xfs_sb.h" |
1da177e4 | 26 | #include "xfs_mount.h" |
1da177e4 | 27 | #include "xfs_inode.h" |
a844f451 NS |
28 | #include "xfs_btree.h" |
29 | #include "xfs_ialloc.h" | |
a4fbe6ab | 30 | #include "xfs_ialloc_btree.h" |
1da177e4 | 31 | #include "xfs_alloc.h" |
1da177e4 LT |
32 | #include "xfs_rtalloc.h" |
33 | #include "xfs_error.h" | |
34 | #include "xfs_bmap.h" | |
983d09ff | 35 | #include "xfs_cksum.h" |
239880ef | 36 | #include "xfs_trans.h" |
983d09ff | 37 | #include "xfs_buf_item.h" |
ddf6ad01 | 38 | #include "xfs_icreate_item.h" |
7bb85ef3 | 39 | #include "xfs_icache.h" |
d123031a | 40 | #include "xfs_trace.h" |
1da177e4 | 41 | |
1da177e4 LT |
42 | |
43 | /* | |
44 | * Allocation group level functions. | |
45 | */ | |
75de2a91 DC |
46 | static inline int |
47 | xfs_ialloc_cluster_alignment( | |
7a1df156 | 48 | struct xfs_mount *mp) |
75de2a91 | 49 | { |
7a1df156 DC |
50 | if (xfs_sb_version_hasalign(&mp->m_sb) && |
51 | mp->m_sb.sb_inoalignmt >= | |
52 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) | |
53 | return mp->m_sb.sb_inoalignmt; | |
75de2a91 DC |
54 | return 1; |
55 | } | |
1da177e4 | 56 | |
fe033cc8 | 57 | /* |
21875505 | 58 | * Lookup a record by ino in the btree given by cur. |
fe033cc8 | 59 | */ |
81e25176 | 60 | int /* error */ |
21875505 | 61 | xfs_inobt_lookup( |
fe033cc8 CH |
62 | struct xfs_btree_cur *cur, /* btree cursor */ |
63 | xfs_agino_t ino, /* starting inode of chunk */ | |
21875505 | 64 | xfs_lookup_t dir, /* <=, >=, == */ |
fe033cc8 CH |
65 | int *stat) /* success/failure */ |
66 | { | |
67 | cur->bc_rec.i.ir_startino = ino; | |
5419040f BF |
68 | cur->bc_rec.i.ir_holemask = 0; |
69 | cur->bc_rec.i.ir_count = 0; | |
21875505 CH |
70 | cur->bc_rec.i.ir_freecount = 0; |
71 | cur->bc_rec.i.ir_free = 0; | |
72 | return xfs_btree_lookup(cur, dir, stat); | |
fe033cc8 CH |
73 | } |
74 | ||
278d0ca1 | 75 | /* |
afabc24a | 76 | * Update the record referred to by cur to the value given. |
278d0ca1 CH |
77 | * This either works (return 0) or gets an EFSCORRUPTED error. |
78 | */ | |
79 | STATIC int /* error */ | |
80 | xfs_inobt_update( | |
81 | struct xfs_btree_cur *cur, /* btree cursor */ | |
afabc24a | 82 | xfs_inobt_rec_incore_t *irec) /* btree record */ |
278d0ca1 CH |
83 | { |
84 | union xfs_btree_rec rec; | |
85 | ||
afabc24a | 86 | rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino); |
5419040f BF |
87 | if (xfs_sb_version_hassparseinodes(&cur->bc_mp->m_sb)) { |
88 | rec.inobt.ir_u.sp.ir_holemask = cpu_to_be16(irec->ir_holemask); | |
89 | rec.inobt.ir_u.sp.ir_count = irec->ir_count; | |
90 | rec.inobt.ir_u.sp.ir_freecount = irec->ir_freecount; | |
91 | } else { | |
92 | /* ir_holemask/ir_count not supported on-disk */ | |
93 | rec.inobt.ir_u.f.ir_freecount = cpu_to_be32(irec->ir_freecount); | |
94 | } | |
afabc24a | 95 | rec.inobt.ir_free = cpu_to_be64(irec->ir_free); |
278d0ca1 CH |
96 | return xfs_btree_update(cur, &rec); |
97 | } | |
98 | ||
8cc938fe CH |
99 | /* |
100 | * Get the data from the pointed-to record. | |
101 | */ | |
102 | int /* error */ | |
103 | xfs_inobt_get_rec( | |
104 | struct xfs_btree_cur *cur, /* btree cursor */ | |
2e287a73 | 105 | xfs_inobt_rec_incore_t *irec, /* btree record */ |
8cc938fe CH |
106 | int *stat) /* output: success/failure */ |
107 | { | |
108 | union xfs_btree_rec *rec; | |
109 | int error; | |
110 | ||
111 | error = xfs_btree_get_rec(cur, &rec, stat); | |
5419040f BF |
112 | if (error || *stat == 0) |
113 | return error; | |
114 | ||
115 | irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino); | |
116 | if (xfs_sb_version_hassparseinodes(&cur->bc_mp->m_sb)) { | |
117 | irec->ir_holemask = be16_to_cpu(rec->inobt.ir_u.sp.ir_holemask); | |
118 | irec->ir_count = rec->inobt.ir_u.sp.ir_count; | |
119 | irec->ir_freecount = rec->inobt.ir_u.sp.ir_freecount; | |
120 | } else { | |
121 | /* | |
122 | * ir_holemask/ir_count not supported on-disk. Fill in hardcoded | |
123 | * values for full inode chunks. | |
124 | */ | |
125 | irec->ir_holemask = XFS_INOBT_HOLEMASK_FULL; | |
126 | irec->ir_count = XFS_INODES_PER_CHUNK; | |
127 | irec->ir_freecount = | |
128 | be32_to_cpu(rec->inobt.ir_u.f.ir_freecount); | |
8cc938fe | 129 | } |
5419040f BF |
130 | irec->ir_free = be64_to_cpu(rec->inobt.ir_free); |
131 | ||
132 | return 0; | |
8cc938fe CH |
133 | } |
134 | ||
0aa0a756 BF |
135 | /* |
136 | * Insert a single inobt record. Cursor must already point to desired location. | |
137 | */ | |
138 | STATIC int | |
139 | xfs_inobt_insert_rec( | |
140 | struct xfs_btree_cur *cur, | |
5419040f BF |
141 | __uint16_t holemask, |
142 | __uint8_t count, | |
0aa0a756 BF |
143 | __int32_t freecount, |
144 | xfs_inofree_t free, | |
145 | int *stat) | |
146 | { | |
5419040f BF |
147 | cur->bc_rec.i.ir_holemask = holemask; |
148 | cur->bc_rec.i.ir_count = count; | |
0aa0a756 BF |
149 | cur->bc_rec.i.ir_freecount = freecount; |
150 | cur->bc_rec.i.ir_free = free; | |
151 | return xfs_btree_insert(cur, stat); | |
152 | } | |
153 | ||
154 | /* | |
155 | * Insert records describing a newly allocated inode chunk into the inobt. | |
156 | */ | |
157 | STATIC int | |
158 | xfs_inobt_insert( | |
159 | struct xfs_mount *mp, | |
160 | struct xfs_trans *tp, | |
161 | struct xfs_buf *agbp, | |
162 | xfs_agino_t newino, | |
163 | xfs_agino_t newlen, | |
164 | xfs_btnum_t btnum) | |
165 | { | |
166 | struct xfs_btree_cur *cur; | |
167 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
168 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
169 | xfs_agino_t thisino; | |
170 | int i; | |
171 | int error; | |
172 | ||
173 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum); | |
174 | ||
175 | for (thisino = newino; | |
176 | thisino < newino + newlen; | |
177 | thisino += XFS_INODES_PER_CHUNK) { | |
178 | error = xfs_inobt_lookup(cur, thisino, XFS_LOOKUP_EQ, &i); | |
179 | if (error) { | |
180 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
181 | return error; | |
182 | } | |
183 | ASSERT(i == 0); | |
184 | ||
5419040f BF |
185 | error = xfs_inobt_insert_rec(cur, XFS_INOBT_HOLEMASK_FULL, |
186 | XFS_INODES_PER_CHUNK, | |
187 | XFS_INODES_PER_CHUNK, | |
0aa0a756 BF |
188 | XFS_INOBT_ALL_FREE, &i); |
189 | if (error) { | |
190 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
191 | return error; | |
192 | } | |
193 | ASSERT(i == 1); | |
194 | } | |
195 | ||
196 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
197 | ||
198 | return 0; | |
199 | } | |
200 | ||
0b48db80 DC |
201 | /* |
202 | * Verify that the number of free inodes in the AGI is correct. | |
203 | */ | |
204 | #ifdef DEBUG | |
205 | STATIC int | |
206 | xfs_check_agi_freecount( | |
207 | struct xfs_btree_cur *cur, | |
208 | struct xfs_agi *agi) | |
209 | { | |
210 | if (cur->bc_nlevels == 1) { | |
211 | xfs_inobt_rec_incore_t rec; | |
212 | int freecount = 0; | |
213 | int error; | |
214 | int i; | |
215 | ||
21875505 | 216 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); |
0b48db80 DC |
217 | if (error) |
218 | return error; | |
219 | ||
220 | do { | |
221 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
222 | if (error) | |
223 | return error; | |
224 | ||
225 | if (i) { | |
226 | freecount += rec.ir_freecount; | |
227 | error = xfs_btree_increment(cur, 0, &i); | |
228 | if (error) | |
229 | return error; | |
230 | } | |
231 | } while (i == 1); | |
232 | ||
233 | if (!XFS_FORCED_SHUTDOWN(cur->bc_mp)) | |
234 | ASSERT(freecount == be32_to_cpu(agi->agi_freecount)); | |
235 | } | |
236 | return 0; | |
237 | } | |
238 | #else | |
239 | #define xfs_check_agi_freecount(cur, agi) 0 | |
240 | #endif | |
241 | ||
85c0b2ab | 242 | /* |
28c8e41a DC |
243 | * Initialise a new set of inodes. When called without a transaction context |
244 | * (e.g. from recovery) we initiate a delayed write of the inode buffers rather | |
245 | * than logging them (which in a transaction context puts them into the AIL | |
246 | * for writeback rather than the xfsbufd queue). | |
85c0b2ab | 247 | */ |
ddf6ad01 | 248 | int |
85c0b2ab DC |
249 | xfs_ialloc_inode_init( |
250 | struct xfs_mount *mp, | |
251 | struct xfs_trans *tp, | |
28c8e41a | 252 | struct list_head *buffer_list, |
463958af | 253 | int icount, |
85c0b2ab DC |
254 | xfs_agnumber_t agno, |
255 | xfs_agblock_t agbno, | |
256 | xfs_agblock_t length, | |
257 | unsigned int gen) | |
258 | { | |
259 | struct xfs_buf *fbuf; | |
260 | struct xfs_dinode *free; | |
6e0c7b8c | 261 | int nbufs, blks_per_cluster, inodes_per_cluster; |
85c0b2ab DC |
262 | int version; |
263 | int i, j; | |
264 | xfs_daddr_t d; | |
93848a99 | 265 | xfs_ino_t ino = 0; |
85c0b2ab DC |
266 | |
267 | /* | |
6e0c7b8c JL |
268 | * Loop over the new block(s), filling in the inodes. For small block |
269 | * sizes, manipulate the inodes in buffers which are multiples of the | |
270 | * blocks size. | |
85c0b2ab | 271 | */ |
6e0c7b8c JL |
272 | blks_per_cluster = xfs_icluster_size_fsb(mp); |
273 | inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog; | |
274 | nbufs = length / blks_per_cluster; | |
85c0b2ab DC |
275 | |
276 | /* | |
93848a99 CH |
277 | * Figure out what version number to use in the inodes we create. If |
278 | * the superblock version has caught up to the one that supports the new | |
279 | * inode format, then use the new inode version. Otherwise use the old | |
280 | * version so that old kernels will continue to be able to use the file | |
281 | * system. | |
282 | * | |
283 | * For v3 inodes, we also need to write the inode number into the inode, | |
284 | * so calculate the first inode number of the chunk here as | |
285 | * XFS_OFFBNO_TO_AGINO() only works within a filesystem block, not | |
286 | * across multiple filesystem blocks (such as a cluster) and so cannot | |
287 | * be used in the cluster buffer loop below. | |
288 | * | |
289 | * Further, because we are writing the inode directly into the buffer | |
290 | * and calculating a CRC on the entire inode, we have ot log the entire | |
291 | * inode so that the entire range the CRC covers is present in the log. | |
292 | * That means for v3 inode we log the entire buffer rather than just the | |
293 | * inode cores. | |
85c0b2ab | 294 | */ |
93848a99 CH |
295 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
296 | version = 3; | |
297 | ino = XFS_AGINO_TO_INO(mp, agno, | |
298 | XFS_OFFBNO_TO_AGINO(mp, agbno, 0)); | |
ddf6ad01 DC |
299 | |
300 | /* | |
301 | * log the initialisation that is about to take place as an | |
302 | * logical operation. This means the transaction does not | |
303 | * need to log the physical changes to the inode buffers as log | |
304 | * recovery will know what initialisation is actually needed. | |
305 | * Hence we only need to log the buffers as "ordered" buffers so | |
306 | * they track in the AIL as if they were physically logged. | |
307 | */ | |
308 | if (tp) | |
463958af | 309 | xfs_icreate_log(tp, agno, agbno, icount, |
ddf6ad01 | 310 | mp->m_sb.sb_inodesize, length, gen); |
263997a6 | 311 | } else |
85c0b2ab | 312 | version = 2; |
85c0b2ab DC |
313 | |
314 | for (j = 0; j < nbufs; j++) { | |
315 | /* | |
316 | * Get the block. | |
317 | */ | |
318 | d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster)); | |
319 | fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, | |
7c4cebe8 DC |
320 | mp->m_bsize * blks_per_cluster, |
321 | XBF_UNMAPPED); | |
2a30f36d | 322 | if (!fbuf) |
2451337d | 323 | return -ENOMEM; |
ddf6ad01 DC |
324 | |
325 | /* Initialize the inode buffers and log them appropriately. */ | |
1813dd64 | 326 | fbuf->b_ops = &xfs_inode_buf_ops; |
93848a99 | 327 | xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length)); |
6e0c7b8c | 328 | for (i = 0; i < inodes_per_cluster; i++) { |
85c0b2ab | 329 | int ioffset = i << mp->m_sb.sb_inodelog; |
93848a99 | 330 | uint isize = xfs_dinode_size(version); |
85c0b2ab DC |
331 | |
332 | free = xfs_make_iptr(mp, fbuf, i); | |
333 | free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC); | |
334 | free->di_version = version; | |
335 | free->di_gen = cpu_to_be32(gen); | |
336 | free->di_next_unlinked = cpu_to_be32(NULLAGINO); | |
93848a99 CH |
337 | |
338 | if (version == 3) { | |
339 | free->di_ino = cpu_to_be64(ino); | |
340 | ino++; | |
341 | uuid_copy(&free->di_uuid, &mp->m_sb.sb_uuid); | |
342 | xfs_dinode_calc_crc(mp, free); | |
28c8e41a | 343 | } else if (tp) { |
93848a99 CH |
344 | /* just log the inode core */ |
345 | xfs_trans_log_buf(tp, fbuf, ioffset, | |
346 | ioffset + isize - 1); | |
347 | } | |
348 | } | |
28c8e41a DC |
349 | |
350 | if (tp) { | |
351 | /* | |
352 | * Mark the buffer as an inode allocation buffer so it | |
353 | * sticks in AIL at the point of this allocation | |
354 | * transaction. This ensures the they are on disk before | |
355 | * the tail of the log can be moved past this | |
356 | * transaction (i.e. by preventing relogging from moving | |
357 | * it forward in the log). | |
358 | */ | |
359 | xfs_trans_inode_alloc_buf(tp, fbuf); | |
360 | if (version == 3) { | |
ddf6ad01 DC |
361 | /* |
362 | * Mark the buffer as ordered so that they are | |
363 | * not physically logged in the transaction but | |
364 | * still tracked in the AIL as part of the | |
365 | * transaction and pin the log appropriately. | |
366 | */ | |
367 | xfs_trans_ordered_buf(tp, fbuf); | |
28c8e41a DC |
368 | xfs_trans_log_buf(tp, fbuf, 0, |
369 | BBTOB(fbuf->b_length) - 1); | |
370 | } | |
371 | } else { | |
372 | fbuf->b_flags |= XBF_DONE; | |
373 | xfs_buf_delwri_queue(fbuf, buffer_list); | |
374 | xfs_buf_relse(fbuf); | |
85c0b2ab | 375 | } |
85c0b2ab | 376 | } |
2a30f36d | 377 | return 0; |
85c0b2ab DC |
378 | } |
379 | ||
56d1115c BF |
380 | /* |
381 | * Align startino and allocmask for a recently allocated sparse chunk such that | |
382 | * they are fit for insertion (or merge) into the on-disk inode btrees. | |
383 | * | |
384 | * Background: | |
385 | * | |
386 | * When enabled, sparse inode support increases the inode alignment from cluster | |
387 | * size to inode chunk size. This means that the minimum range between two | |
388 | * non-adjacent inode records in the inobt is large enough for a full inode | |
389 | * record. This allows for cluster sized, cluster aligned block allocation | |
390 | * without need to worry about whether the resulting inode record overlaps with | |
391 | * another record in the tree. Without this basic rule, we would have to deal | |
392 | * with the consequences of overlap by potentially undoing recent allocations in | |
393 | * the inode allocation codepath. | |
394 | * | |
395 | * Because of this alignment rule (which is enforced on mount), there are two | |
396 | * inobt possibilities for newly allocated sparse chunks. One is that the | |
397 | * aligned inode record for the chunk covers a range of inodes not already | |
398 | * covered in the inobt (i.e., it is safe to insert a new sparse record). The | |
399 | * other is that a record already exists at the aligned startino that considers | |
400 | * the newly allocated range as sparse. In the latter case, record content is | |
401 | * merged in hope that sparse inode chunks fill to full chunks over time. | |
402 | */ | |
403 | STATIC void | |
404 | xfs_align_sparse_ino( | |
405 | struct xfs_mount *mp, | |
406 | xfs_agino_t *startino, | |
407 | uint16_t *allocmask) | |
408 | { | |
409 | xfs_agblock_t agbno; | |
410 | xfs_agblock_t mod; | |
411 | int offset; | |
412 | ||
413 | agbno = XFS_AGINO_TO_AGBNO(mp, *startino); | |
414 | mod = agbno % mp->m_sb.sb_inoalignmt; | |
415 | if (!mod) | |
416 | return; | |
417 | ||
418 | /* calculate the inode offset and align startino */ | |
419 | offset = mod << mp->m_sb.sb_inopblog; | |
420 | *startino -= offset; | |
421 | ||
422 | /* | |
423 | * Since startino has been aligned down, left shift allocmask such that | |
424 | * it continues to represent the same physical inodes relative to the | |
425 | * new startino. | |
426 | */ | |
427 | *allocmask <<= offset / XFS_INODES_PER_HOLEMASK_BIT; | |
428 | } | |
429 | ||
430 | /* | |
431 | * Determine whether the source inode record can merge into the target. Both | |
432 | * records must be sparse, the inode ranges must match and there must be no | |
433 | * allocation overlap between the records. | |
434 | */ | |
435 | STATIC bool | |
436 | __xfs_inobt_can_merge( | |
437 | struct xfs_inobt_rec_incore *trec, /* tgt record */ | |
438 | struct xfs_inobt_rec_incore *srec) /* src record */ | |
439 | { | |
440 | uint64_t talloc; | |
441 | uint64_t salloc; | |
442 | ||
443 | /* records must cover the same inode range */ | |
444 | if (trec->ir_startino != srec->ir_startino) | |
445 | return false; | |
446 | ||
447 | /* both records must be sparse */ | |
448 | if (!xfs_inobt_issparse(trec->ir_holemask) || | |
449 | !xfs_inobt_issparse(srec->ir_holemask)) | |
450 | return false; | |
451 | ||
452 | /* both records must track some inodes */ | |
453 | if (!trec->ir_count || !srec->ir_count) | |
454 | return false; | |
455 | ||
456 | /* can't exceed capacity of a full record */ | |
457 | if (trec->ir_count + srec->ir_count > XFS_INODES_PER_CHUNK) | |
458 | return false; | |
459 | ||
460 | /* verify there is no allocation overlap */ | |
461 | talloc = xfs_inobt_irec_to_allocmask(trec); | |
462 | salloc = xfs_inobt_irec_to_allocmask(srec); | |
463 | if (talloc & salloc) | |
464 | return false; | |
465 | ||
466 | return true; | |
467 | } | |
468 | ||
469 | /* | |
470 | * Merge the source inode record into the target. The caller must call | |
471 | * __xfs_inobt_can_merge() to ensure the merge is valid. | |
472 | */ | |
473 | STATIC void | |
474 | __xfs_inobt_rec_merge( | |
475 | struct xfs_inobt_rec_incore *trec, /* target */ | |
476 | struct xfs_inobt_rec_incore *srec) /* src */ | |
477 | { | |
478 | ASSERT(trec->ir_startino == srec->ir_startino); | |
479 | ||
480 | /* combine the counts */ | |
481 | trec->ir_count += srec->ir_count; | |
482 | trec->ir_freecount += srec->ir_freecount; | |
483 | ||
484 | /* | |
485 | * Merge the holemask and free mask. For both fields, 0 bits refer to | |
486 | * allocated inodes. We combine the allocated ranges with bitwise AND. | |
487 | */ | |
488 | trec->ir_holemask &= srec->ir_holemask; | |
489 | trec->ir_free &= srec->ir_free; | |
490 | } | |
491 | ||
492 | /* | |
493 | * Insert a new sparse inode chunk into the associated inode btree. The inode | |
494 | * record for the sparse chunk is pre-aligned to a startino that should match | |
495 | * any pre-existing sparse inode record in the tree. This allows sparse chunks | |
496 | * to fill over time. | |
497 | * | |
498 | * This function supports two modes of handling preexisting records depending on | |
499 | * the merge flag. If merge is true, the provided record is merged with the | |
500 | * existing record and updated in place. The merged record is returned in nrec. | |
501 | * If merge is false, an existing record is replaced with the provided record. | |
502 | * If no preexisting record exists, the provided record is always inserted. | |
503 | * | |
504 | * It is considered corruption if a merge is requested and not possible. Given | |
505 | * the sparse inode alignment constraints, this should never happen. | |
506 | */ | |
507 | STATIC int | |
508 | xfs_inobt_insert_sprec( | |
509 | struct xfs_mount *mp, | |
510 | struct xfs_trans *tp, | |
511 | struct xfs_buf *agbp, | |
512 | int btnum, | |
513 | struct xfs_inobt_rec_incore *nrec, /* in/out: new/merged rec. */ | |
514 | bool merge) /* merge or replace */ | |
515 | { | |
516 | struct xfs_btree_cur *cur; | |
517 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
518 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
519 | int error; | |
520 | int i; | |
521 | struct xfs_inobt_rec_incore rec; | |
522 | ||
523 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum); | |
524 | ||
525 | /* the new record is pre-aligned so we know where to look */ | |
526 | error = xfs_inobt_lookup(cur, nrec->ir_startino, XFS_LOOKUP_EQ, &i); | |
527 | if (error) | |
528 | goto error; | |
529 | /* if nothing there, insert a new record and return */ | |
530 | if (i == 0) { | |
531 | error = xfs_inobt_insert_rec(cur, nrec->ir_holemask, | |
532 | nrec->ir_count, nrec->ir_freecount, | |
533 | nrec->ir_free, &i); | |
534 | if (error) | |
535 | goto error; | |
536 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error); | |
537 | ||
538 | goto out; | |
539 | } | |
540 | ||
541 | /* | |
542 | * A record exists at this startino. Merge or replace the record | |
543 | * depending on what we've been asked to do. | |
544 | */ | |
545 | if (merge) { | |
546 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
547 | if (error) | |
548 | goto error; | |
549 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error); | |
550 | XFS_WANT_CORRUPTED_GOTO(mp, | |
551 | rec.ir_startino == nrec->ir_startino, | |
552 | error); | |
553 | ||
554 | /* | |
555 | * This should never fail. If we have coexisting records that | |
556 | * cannot merge, something is seriously wrong. | |
557 | */ | |
558 | XFS_WANT_CORRUPTED_GOTO(mp, __xfs_inobt_can_merge(nrec, &rec), | |
559 | error); | |
560 | ||
561 | trace_xfs_irec_merge_pre(mp, agno, rec.ir_startino, | |
562 | rec.ir_holemask, nrec->ir_startino, | |
563 | nrec->ir_holemask); | |
564 | ||
565 | /* merge to nrec to output the updated record */ | |
566 | __xfs_inobt_rec_merge(nrec, &rec); | |
567 | ||
568 | trace_xfs_irec_merge_post(mp, agno, nrec->ir_startino, | |
569 | nrec->ir_holemask); | |
570 | ||
571 | error = xfs_inobt_rec_check_count(mp, nrec); | |
572 | if (error) | |
573 | goto error; | |
574 | } | |
575 | ||
576 | error = xfs_inobt_update(cur, nrec); | |
577 | if (error) | |
578 | goto error; | |
579 | ||
580 | out: | |
581 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
582 | return 0; | |
583 | error: | |
584 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
585 | return error; | |
586 | } | |
587 | ||
1da177e4 LT |
588 | /* |
589 | * Allocate new inodes in the allocation group specified by agbp. | |
590 | * Return 0 for success, else error code. | |
591 | */ | |
592 | STATIC int /* error code or 0 */ | |
593 | xfs_ialloc_ag_alloc( | |
594 | xfs_trans_t *tp, /* transaction pointer */ | |
595 | xfs_buf_t *agbp, /* alloc group buffer */ | |
596 | int *alloc) | |
597 | { | |
598 | xfs_agi_t *agi; /* allocation group header */ | |
599 | xfs_alloc_arg_t args; /* allocation argument structure */ | |
92821e2b | 600 | xfs_agnumber_t agno; |
1da177e4 | 601 | int error; |
1da177e4 LT |
602 | xfs_agino_t newino; /* new first inode's number */ |
603 | xfs_agino_t newlen; /* new number of inodes */ | |
3ccb8b5f | 604 | int isaligned = 0; /* inode allocation at stripe unit */ |
1da177e4 | 605 | /* boundary */ |
56d1115c BF |
606 | uint16_t allocmask = (uint16_t) -1; /* init. to full chunk */ |
607 | struct xfs_inobt_rec_incore rec; | |
44b56e0a | 608 | struct xfs_perag *pag; |
1da177e4 | 609 | |
1cdadee1 BF |
610 | int do_sparse = 0; |
611 | ||
612 | #ifdef DEBUG | |
613 | /* randomly do sparse inode allocations */ | |
614 | if (xfs_sb_version_hassparseinodes(&tp->t_mountp->m_sb)) | |
615 | do_sparse = prandom_u32() & 1; | |
616 | #endif | |
617 | ||
a0041684 | 618 | memset(&args, 0, sizeof(args)); |
1da177e4 LT |
619 | args.tp = tp; |
620 | args.mp = tp->t_mountp; | |
1cdadee1 | 621 | args.fsbno = NULLFSBLOCK; |
1da177e4 LT |
622 | |
623 | /* | |
624 | * Locking will ensure that we don't have two callers in here | |
625 | * at one time. | |
626 | */ | |
71783438 | 627 | newlen = args.mp->m_ialloc_inos; |
1da177e4 | 628 | if (args.mp->m_maxicount && |
501ab323 DC |
629 | percpu_counter_read(&args.mp->m_icount) + newlen > |
630 | args.mp->m_maxicount) | |
2451337d | 631 | return -ENOSPC; |
126cd105 | 632 | args.minlen = args.maxlen = args.mp->m_ialloc_blks; |
1da177e4 | 633 | /* |
3ccb8b5f GO |
634 | * First try to allocate inodes contiguous with the last-allocated |
635 | * chunk of inodes. If the filesystem is striped, this will fill | |
636 | * an entire stripe unit with inodes. | |
28c8e41a | 637 | */ |
1da177e4 | 638 | agi = XFS_BUF_TO_AGI(agbp); |
3ccb8b5f | 639 | newino = be32_to_cpu(agi->agi_newino); |
85c0b2ab | 640 | agno = be32_to_cpu(agi->agi_seqno); |
019ff2d5 | 641 | args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) + |
126cd105 | 642 | args.mp->m_ialloc_blks; |
1cdadee1 BF |
643 | if (do_sparse) |
644 | goto sparse_alloc; | |
019ff2d5 NS |
645 | if (likely(newino != NULLAGINO && |
646 | (args.agbno < be32_to_cpu(agi->agi_length)))) { | |
85c0b2ab | 647 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
3ccb8b5f | 648 | args.type = XFS_ALLOCTYPE_THIS_BNO; |
3ccb8b5f | 649 | args.prod = 1; |
75de2a91 | 650 | |
3ccb8b5f | 651 | /* |
75de2a91 DC |
652 | * We need to take into account alignment here to ensure that |
653 | * we don't modify the free list if we fail to have an exact | |
654 | * block. If we don't have an exact match, and every oher | |
655 | * attempt allocation attempt fails, we'll end up cancelling | |
656 | * a dirty transaction and shutting down. | |
657 | * | |
658 | * For an exact allocation, alignment must be 1, | |
659 | * however we need to take cluster alignment into account when | |
660 | * fixing up the freelist. Use the minalignslop field to | |
661 | * indicate that extra blocks might be required for alignment, | |
662 | * but not to use them in the actual exact allocation. | |
3ccb8b5f | 663 | */ |
75de2a91 | 664 | args.alignment = 1; |
7a1df156 | 665 | args.minalignslop = xfs_ialloc_cluster_alignment(args.mp) - 1; |
75de2a91 DC |
666 | |
667 | /* Allow space for the inode btree to split. */ | |
0d87e656 | 668 | args.minleft = args.mp->m_in_maxlevels - 1; |
3ccb8b5f GO |
669 | if ((error = xfs_alloc_vextent(&args))) |
670 | return error; | |
e480a723 BF |
671 | |
672 | /* | |
673 | * This request might have dirtied the transaction if the AG can | |
674 | * satisfy the request, but the exact block was not available. | |
675 | * If the allocation did fail, subsequent requests will relax | |
676 | * the exact agbno requirement and increase the alignment | |
677 | * instead. It is critical that the total size of the request | |
678 | * (len + alignment + slop) does not increase from this point | |
679 | * on, so reset minalignslop to ensure it is not included in | |
680 | * subsequent requests. | |
681 | */ | |
682 | args.minalignslop = 0; | |
1cdadee1 | 683 | } |
1da177e4 | 684 | |
3ccb8b5f GO |
685 | if (unlikely(args.fsbno == NULLFSBLOCK)) { |
686 | /* | |
687 | * Set the alignment for the allocation. | |
688 | * If stripe alignment is turned on then align at stripe unit | |
689 | * boundary. | |
019ff2d5 NS |
690 | * If the cluster size is smaller than a filesystem block |
691 | * then we're doing I/O for inodes in filesystem block size | |
3ccb8b5f GO |
692 | * pieces, so don't need alignment anyway. |
693 | */ | |
694 | isaligned = 0; | |
695 | if (args.mp->m_sinoalign) { | |
696 | ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN)); | |
697 | args.alignment = args.mp->m_dalign; | |
698 | isaligned = 1; | |
75de2a91 | 699 | } else |
7a1df156 | 700 | args.alignment = xfs_ialloc_cluster_alignment(args.mp); |
3ccb8b5f GO |
701 | /* |
702 | * Need to figure out where to allocate the inode blocks. | |
703 | * Ideally they should be spaced out through the a.g. | |
704 | * For now, just allocate blocks up front. | |
705 | */ | |
706 | args.agbno = be32_to_cpu(agi->agi_root); | |
85c0b2ab | 707 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
3ccb8b5f GO |
708 | /* |
709 | * Allocate a fixed-size extent of inodes. | |
710 | */ | |
711 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
3ccb8b5f GO |
712 | args.prod = 1; |
713 | /* | |
714 | * Allow space for the inode btree to split. | |
715 | */ | |
0d87e656 | 716 | args.minleft = args.mp->m_in_maxlevels - 1; |
3ccb8b5f GO |
717 | if ((error = xfs_alloc_vextent(&args))) |
718 | return error; | |
719 | } | |
019ff2d5 | 720 | |
1da177e4 LT |
721 | /* |
722 | * If stripe alignment is turned on, then try again with cluster | |
723 | * alignment. | |
724 | */ | |
725 | if (isaligned && args.fsbno == NULLFSBLOCK) { | |
726 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
16259e7d | 727 | args.agbno = be32_to_cpu(agi->agi_root); |
85c0b2ab | 728 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
7a1df156 | 729 | args.alignment = xfs_ialloc_cluster_alignment(args.mp); |
1da177e4 LT |
730 | if ((error = xfs_alloc_vextent(&args))) |
731 | return error; | |
732 | } | |
733 | ||
56d1115c BF |
734 | /* |
735 | * Finally, try a sparse allocation if the filesystem supports it and | |
736 | * the sparse allocation length is smaller than a full chunk. | |
737 | */ | |
738 | if (xfs_sb_version_hassparseinodes(&args.mp->m_sb) && | |
739 | args.mp->m_ialloc_min_blks < args.mp->m_ialloc_blks && | |
740 | args.fsbno == NULLFSBLOCK) { | |
1cdadee1 | 741 | sparse_alloc: |
56d1115c BF |
742 | args.type = XFS_ALLOCTYPE_NEAR_BNO; |
743 | args.agbno = be32_to_cpu(agi->agi_root); | |
744 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); | |
745 | args.alignment = args.mp->m_sb.sb_spino_align; | |
746 | args.prod = 1; | |
747 | ||
748 | args.minlen = args.mp->m_ialloc_min_blks; | |
749 | args.maxlen = args.minlen; | |
750 | ||
751 | /* | |
752 | * The inode record will be aligned to full chunk size. We must | |
753 | * prevent sparse allocation from AG boundaries that result in | |
754 | * invalid inode records, such as records that start at agbno 0 | |
755 | * or extend beyond the AG. | |
756 | * | |
757 | * Set min agbno to the first aligned, non-zero agbno and max to | |
758 | * the last aligned agbno that is at least one full chunk from | |
759 | * the end of the AG. | |
760 | */ | |
761 | args.min_agbno = args.mp->m_sb.sb_inoalignmt; | |
762 | args.max_agbno = round_down(args.mp->m_sb.sb_agblocks, | |
763 | args.mp->m_sb.sb_inoalignmt) - | |
764 | args.mp->m_ialloc_blks; | |
765 | ||
766 | error = xfs_alloc_vextent(&args); | |
767 | if (error) | |
768 | return error; | |
769 | ||
770 | newlen = args.len << args.mp->m_sb.sb_inopblog; | |
771 | allocmask = (1 << (newlen / XFS_INODES_PER_HOLEMASK_BIT)) - 1; | |
772 | } | |
773 | ||
1da177e4 LT |
774 | if (args.fsbno == NULLFSBLOCK) { |
775 | *alloc = 0; | |
776 | return 0; | |
777 | } | |
778 | ASSERT(args.len == args.minlen); | |
1da177e4 | 779 | |
359346a9 | 780 | /* |
85c0b2ab DC |
781 | * Stamp and write the inode buffers. |
782 | * | |
359346a9 DC |
783 | * Seed the new inode cluster with a random generation number. This |
784 | * prevents short-term reuse of generation numbers if a chunk is | |
785 | * freed and then immediately reallocated. We use random numbers | |
786 | * rather than a linear progression to prevent the next generation | |
787 | * number from being easily guessable. | |
788 | */ | |
463958af BF |
789 | error = xfs_ialloc_inode_init(args.mp, tp, NULL, newlen, agno, |
790 | args.agbno, args.len, prandom_u32()); | |
d42f08f6 | 791 | |
2a30f36d CS |
792 | if (error) |
793 | return error; | |
85c0b2ab DC |
794 | /* |
795 | * Convert the results. | |
796 | */ | |
797 | newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0); | |
56d1115c BF |
798 | |
799 | if (xfs_inobt_issparse(~allocmask)) { | |
800 | /* | |
801 | * We've allocated a sparse chunk. Align the startino and mask. | |
802 | */ | |
803 | xfs_align_sparse_ino(args.mp, &newino, &allocmask); | |
804 | ||
805 | rec.ir_startino = newino; | |
806 | rec.ir_holemask = ~allocmask; | |
807 | rec.ir_count = newlen; | |
808 | rec.ir_freecount = newlen; | |
809 | rec.ir_free = XFS_INOBT_ALL_FREE; | |
810 | ||
811 | /* | |
812 | * Insert the sparse record into the inobt and allow for a merge | |
813 | * if necessary. If a merge does occur, rec is updated to the | |
814 | * merged record. | |
815 | */ | |
816 | error = xfs_inobt_insert_sprec(args.mp, tp, agbp, XFS_BTNUM_INO, | |
817 | &rec, true); | |
818 | if (error == -EFSCORRUPTED) { | |
819 | xfs_alert(args.mp, | |
820 | "invalid sparse inode record: ino 0x%llx holemask 0x%x count %u", | |
821 | XFS_AGINO_TO_INO(args.mp, agno, | |
822 | rec.ir_startino), | |
823 | rec.ir_holemask, rec.ir_count); | |
824 | xfs_force_shutdown(args.mp, SHUTDOWN_CORRUPT_INCORE); | |
825 | } | |
826 | if (error) | |
827 | return error; | |
828 | ||
829 | /* | |
830 | * We can't merge the part we've just allocated as for the inobt | |
831 | * due to finobt semantics. The original record may or may not | |
832 | * exist independent of whether physical inodes exist in this | |
833 | * sparse chunk. | |
834 | * | |
835 | * We must update the finobt record based on the inobt record. | |
836 | * rec contains the fully merged and up to date inobt record | |
837 | * from the previous call. Set merge false to replace any | |
838 | * existing record with this one. | |
839 | */ | |
840 | if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) { | |
841 | error = xfs_inobt_insert_sprec(args.mp, tp, agbp, | |
842 | XFS_BTNUM_FINO, &rec, | |
843 | false); | |
844 | if (error) | |
845 | return error; | |
846 | } | |
847 | } else { | |
848 | /* full chunk - insert new records to both btrees */ | |
849 | error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen, | |
850 | XFS_BTNUM_INO); | |
851 | if (error) | |
852 | return error; | |
853 | ||
854 | if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) { | |
855 | error = xfs_inobt_insert(args.mp, tp, agbp, newino, | |
856 | newlen, XFS_BTNUM_FINO); | |
857 | if (error) | |
858 | return error; | |
859 | } | |
860 | } | |
861 | ||
862 | /* | |
863 | * Update AGI counts and newino. | |
864 | */ | |
413d57c9 MS |
865 | be32_add_cpu(&agi->agi_count, newlen); |
866 | be32_add_cpu(&agi->agi_freecount, newlen); | |
44b56e0a DC |
867 | pag = xfs_perag_get(args.mp, agno); |
868 | pag->pagi_freecount += newlen; | |
869 | xfs_perag_put(pag); | |
16259e7d | 870 | agi->agi_newino = cpu_to_be32(newino); |
85c0b2ab | 871 | |
1da177e4 LT |
872 | /* |
873 | * Log allocation group header fields | |
874 | */ | |
875 | xfs_ialloc_log_agi(tp, agbp, | |
876 | XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO); | |
877 | /* | |
878 | * Modify/log superblock values for inode count and inode free count. | |
879 | */ | |
880 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen); | |
881 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen); | |
882 | *alloc = 1; | |
883 | return 0; | |
884 | } | |
885 | ||
b8f82a4a | 886 | STATIC xfs_agnumber_t |
1da177e4 LT |
887 | xfs_ialloc_next_ag( |
888 | xfs_mount_t *mp) | |
889 | { | |
890 | xfs_agnumber_t agno; | |
891 | ||
892 | spin_lock(&mp->m_agirotor_lock); | |
893 | agno = mp->m_agirotor; | |
8aea3ff4 | 894 | if (++mp->m_agirotor >= mp->m_maxagi) |
1da177e4 LT |
895 | mp->m_agirotor = 0; |
896 | spin_unlock(&mp->m_agirotor_lock); | |
897 | ||
898 | return agno; | |
899 | } | |
900 | ||
901 | /* | |
902 | * Select an allocation group to look for a free inode in, based on the parent | |
2f21ff1c | 903 | * inode and the mode. Return the allocation group buffer. |
1da177e4 | 904 | */ |
55d6af64 | 905 | STATIC xfs_agnumber_t |
1da177e4 LT |
906 | xfs_ialloc_ag_select( |
907 | xfs_trans_t *tp, /* transaction pointer */ | |
908 | xfs_ino_t parent, /* parent directory inode number */ | |
576b1d67 | 909 | umode_t mode, /* bits set to indicate file type */ |
1da177e4 LT |
910 | int okalloc) /* ok to allocate more space */ |
911 | { | |
1da177e4 LT |
912 | xfs_agnumber_t agcount; /* number of ag's in the filesystem */ |
913 | xfs_agnumber_t agno; /* current ag number */ | |
914 | int flags; /* alloc buffer locking flags */ | |
915 | xfs_extlen_t ineed; /* blocks needed for inode allocation */ | |
916 | xfs_extlen_t longest = 0; /* longest extent available */ | |
917 | xfs_mount_t *mp; /* mount point structure */ | |
918 | int needspace; /* file mode implies space allocated */ | |
919 | xfs_perag_t *pag; /* per allocation group data */ | |
920 | xfs_agnumber_t pagno; /* parent (starting) ag number */ | |
55d6af64 | 921 | int error; |
1da177e4 LT |
922 | |
923 | /* | |
924 | * Files of these types need at least one block if length > 0 | |
925 | * (and they won't fit in the inode, but that's hard to figure out). | |
926 | */ | |
927 | needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode); | |
928 | mp = tp->t_mountp; | |
929 | agcount = mp->m_maxagi; | |
930 | if (S_ISDIR(mode)) | |
931 | pagno = xfs_ialloc_next_ag(mp); | |
932 | else { | |
933 | pagno = XFS_INO_TO_AGNO(mp, parent); | |
934 | if (pagno >= agcount) | |
935 | pagno = 0; | |
936 | } | |
55d6af64 | 937 | |
1da177e4 | 938 | ASSERT(pagno < agcount); |
55d6af64 | 939 | |
1da177e4 LT |
940 | /* |
941 | * Loop through allocation groups, looking for one with a little | |
942 | * free space in it. Note we don't look for free inodes, exactly. | |
943 | * Instead, we include whether there is a need to allocate inodes | |
944 | * to mean that blocks must be allocated for them, | |
945 | * if none are currently free. | |
946 | */ | |
947 | agno = pagno; | |
948 | flags = XFS_ALLOC_FLAG_TRYLOCK; | |
1da177e4 | 949 | for (;;) { |
44b56e0a | 950 | pag = xfs_perag_get(mp, agno); |
55d6af64 CH |
951 | if (!pag->pagi_inodeok) { |
952 | xfs_ialloc_next_ag(mp); | |
953 | goto nextag; | |
954 | } | |
955 | ||
1da177e4 | 956 | if (!pag->pagi_init) { |
55d6af64 CH |
957 | error = xfs_ialloc_pagi_init(mp, tp, agno); |
958 | if (error) | |
1da177e4 | 959 | goto nextag; |
55d6af64 | 960 | } |
1da177e4 | 961 | |
55d6af64 CH |
962 | if (pag->pagi_freecount) { |
963 | xfs_perag_put(pag); | |
964 | return agno; | |
1da177e4 LT |
965 | } |
966 | ||
55d6af64 CH |
967 | if (!okalloc) |
968 | goto nextag; | |
969 | ||
970 | if (!pag->pagf_init) { | |
971 | error = xfs_alloc_pagf_init(mp, tp, agno, flags); | |
972 | if (error) | |
1da177e4 | 973 | goto nextag; |
1da177e4 | 974 | } |
55d6af64 CH |
975 | |
976 | /* | |
7a1df156 DC |
977 | * Check that there is enough free space for the file plus a |
978 | * chunk of inodes if we need to allocate some. If this is the | |
979 | * first pass across the AGs, take into account the potential | |
980 | * space needed for alignment of inode chunks when checking the | |
981 | * longest contiguous free space in the AG - this prevents us | |
982 | * from getting ENOSPC because we have free space larger than | |
983 | * m_ialloc_blks but alignment constraints prevent us from using | |
984 | * it. | |
985 | * | |
986 | * If we can't find an AG with space for full alignment slack to | |
987 | * be taken into account, we must be near ENOSPC in all AGs. | |
988 | * Hence we don't include alignment for the second pass and so | |
989 | * if we fail allocation due to alignment issues then it is most | |
990 | * likely a real ENOSPC condition. | |
55d6af64 | 991 | */ |
066a1884 | 992 | ineed = mp->m_ialloc_min_blks; |
7a1df156 DC |
993 | if (flags && ineed > 1) |
994 | ineed += xfs_ialloc_cluster_alignment(mp); | |
55d6af64 CH |
995 | longest = pag->pagf_longest; |
996 | if (!longest) | |
997 | longest = pag->pagf_flcount > 0; | |
998 | ||
999 | if (pag->pagf_freeblks >= needspace + ineed && | |
1000 | longest >= ineed) { | |
1001 | xfs_perag_put(pag); | |
1002 | return agno; | |
1da177e4 | 1003 | } |
1da177e4 | 1004 | nextag: |
44b56e0a | 1005 | xfs_perag_put(pag); |
1da177e4 LT |
1006 | /* |
1007 | * No point in iterating over the rest, if we're shutting | |
1008 | * down. | |
1009 | */ | |
1c1c6ebc | 1010 | if (XFS_FORCED_SHUTDOWN(mp)) |
55d6af64 | 1011 | return NULLAGNUMBER; |
1da177e4 LT |
1012 | agno++; |
1013 | if (agno >= agcount) | |
1014 | agno = 0; | |
1015 | if (agno == pagno) { | |
1c1c6ebc | 1016 | if (flags == 0) |
55d6af64 | 1017 | return NULLAGNUMBER; |
1da177e4 LT |
1018 | flags = 0; |
1019 | } | |
1020 | } | |
1021 | } | |
1022 | ||
4254b0bb CH |
1023 | /* |
1024 | * Try to retrieve the next record to the left/right from the current one. | |
1025 | */ | |
1026 | STATIC int | |
1027 | xfs_ialloc_next_rec( | |
1028 | struct xfs_btree_cur *cur, | |
1029 | xfs_inobt_rec_incore_t *rec, | |
1030 | int *done, | |
1031 | int left) | |
1032 | { | |
1033 | int error; | |
1034 | int i; | |
1035 | ||
1036 | if (left) | |
1037 | error = xfs_btree_decrement(cur, 0, &i); | |
1038 | else | |
1039 | error = xfs_btree_increment(cur, 0, &i); | |
1040 | ||
1041 | if (error) | |
1042 | return error; | |
1043 | *done = !i; | |
1044 | if (i) { | |
1045 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1046 | if (error) | |
1047 | return error; | |
5fb5aeee | 1048 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
4254b0bb CH |
1049 | } |
1050 | ||
1051 | return 0; | |
1052 | } | |
1053 | ||
bd169565 DC |
1054 | STATIC int |
1055 | xfs_ialloc_get_rec( | |
1056 | struct xfs_btree_cur *cur, | |
1057 | xfs_agino_t agino, | |
1058 | xfs_inobt_rec_incore_t *rec, | |
43df2ee6 | 1059 | int *done) |
bd169565 DC |
1060 | { |
1061 | int error; | |
1062 | int i; | |
1063 | ||
1064 | error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i); | |
1065 | if (error) | |
1066 | return error; | |
1067 | *done = !i; | |
1068 | if (i) { | |
1069 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1070 | if (error) | |
1071 | return error; | |
5fb5aeee | 1072 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
bd169565 DC |
1073 | } |
1074 | ||
1075 | return 0; | |
1076 | } | |
0b48db80 | 1077 | |
d4cc540b BF |
1078 | /* |
1079 | * Return the offset of the first free inode in the record. | |
1080 | */ | |
1081 | STATIC int | |
1082 | xfs_inobt_first_free_inode( | |
1083 | struct xfs_inobt_rec_incore *rec) | |
1084 | { | |
1085 | return xfs_lowbit64(rec->ir_free); | |
1086 | } | |
1087 | ||
1da177e4 | 1088 | /* |
6dd8638e | 1089 | * Allocate an inode using the inobt-only algorithm. |
1da177e4 | 1090 | */ |
f2ecc5e4 | 1091 | STATIC int |
6dd8638e | 1092 | xfs_dialloc_ag_inobt( |
f2ecc5e4 CH |
1093 | struct xfs_trans *tp, |
1094 | struct xfs_buf *agbp, | |
1095 | xfs_ino_t parent, | |
1096 | xfs_ino_t *inop) | |
1da177e4 | 1097 | { |
f2ecc5e4 CH |
1098 | struct xfs_mount *mp = tp->t_mountp; |
1099 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
1100 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
1101 | xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent); | |
1102 | xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent); | |
1103 | struct xfs_perag *pag; | |
1104 | struct xfs_btree_cur *cur, *tcur; | |
1105 | struct xfs_inobt_rec_incore rec, trec; | |
1106 | xfs_ino_t ino; | |
1107 | int error; | |
1108 | int offset; | |
1109 | int i, j; | |
1da177e4 | 1110 | |
44b56e0a | 1111 | pag = xfs_perag_get(mp, agno); |
bd169565 | 1112 | |
4bb61069 CH |
1113 | ASSERT(pag->pagi_init); |
1114 | ASSERT(pag->pagi_inodeok); | |
1115 | ASSERT(pag->pagi_freecount > 0); | |
1116 | ||
bd169565 | 1117 | restart_pagno: |
57bd3dbe | 1118 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
1da177e4 LT |
1119 | /* |
1120 | * If pagino is 0 (this is the root inode allocation) use newino. | |
1121 | * This must work because we've just allocated some. | |
1122 | */ | |
1123 | if (!pagino) | |
16259e7d | 1124 | pagino = be32_to_cpu(agi->agi_newino); |
1da177e4 | 1125 | |
0b48db80 DC |
1126 | error = xfs_check_agi_freecount(cur, agi); |
1127 | if (error) | |
1128 | goto error0; | |
1da177e4 | 1129 | |
1da177e4 | 1130 | /* |
4254b0bb | 1131 | * If in the same AG as the parent, try to get near the parent. |
1da177e4 LT |
1132 | */ |
1133 | if (pagno == agno) { | |
4254b0bb CH |
1134 | int doneleft; /* done, to the left */ |
1135 | int doneright; /* done, to the right */ | |
bd169565 | 1136 | int searchdistance = 10; |
4254b0bb | 1137 | |
21875505 | 1138 | error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i); |
4254b0bb | 1139 | if (error) |
1da177e4 | 1140 | goto error0; |
c29aad41 | 1141 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
4254b0bb CH |
1142 | |
1143 | error = xfs_inobt_get_rec(cur, &rec, &j); | |
1144 | if (error) | |
1145 | goto error0; | |
c29aad41 | 1146 | XFS_WANT_CORRUPTED_GOTO(mp, j == 1, error0); |
4254b0bb CH |
1147 | |
1148 | if (rec.ir_freecount > 0) { | |
1da177e4 LT |
1149 | /* |
1150 | * Found a free inode in the same chunk | |
4254b0bb | 1151 | * as the parent, done. |
1da177e4 | 1152 | */ |
4254b0bb | 1153 | goto alloc_inode; |
1da177e4 | 1154 | } |
4254b0bb CH |
1155 | |
1156 | ||
1da177e4 | 1157 | /* |
4254b0bb | 1158 | * In the same AG as parent, but parent's chunk is full. |
1da177e4 | 1159 | */ |
1da177e4 | 1160 | |
4254b0bb CH |
1161 | /* duplicate the cursor, search left & right simultaneously */ |
1162 | error = xfs_btree_dup_cursor(cur, &tcur); | |
1163 | if (error) | |
1164 | goto error0; | |
1165 | ||
bd169565 DC |
1166 | /* |
1167 | * Skip to last blocks looked up if same parent inode. | |
1168 | */ | |
1169 | if (pagino != NULLAGINO && | |
1170 | pag->pagl_pagino == pagino && | |
1171 | pag->pagl_leftrec != NULLAGINO && | |
1172 | pag->pagl_rightrec != NULLAGINO) { | |
1173 | error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec, | |
43df2ee6 | 1174 | &trec, &doneleft); |
bd169565 DC |
1175 | if (error) |
1176 | goto error1; | |
4254b0bb | 1177 | |
bd169565 | 1178 | error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec, |
43df2ee6 | 1179 | &rec, &doneright); |
bd169565 DC |
1180 | if (error) |
1181 | goto error1; | |
1182 | } else { | |
1183 | /* search left with tcur, back up 1 record */ | |
1184 | error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1); | |
1185 | if (error) | |
1186 | goto error1; | |
1187 | ||
1188 | /* search right with cur, go forward 1 record. */ | |
1189 | error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0); | |
1190 | if (error) | |
1191 | goto error1; | |
1192 | } | |
4254b0bb CH |
1193 | |
1194 | /* | |
1195 | * Loop until we find an inode chunk with a free inode. | |
1196 | */ | |
1197 | while (!doneleft || !doneright) { | |
1198 | int useleft; /* using left inode chunk this time */ | |
1199 | ||
bd169565 DC |
1200 | if (!--searchdistance) { |
1201 | /* | |
1202 | * Not in range - save last search | |
1203 | * location and allocate a new inode | |
1204 | */ | |
3b826386 | 1205 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); |
bd169565 DC |
1206 | pag->pagl_leftrec = trec.ir_startino; |
1207 | pag->pagl_rightrec = rec.ir_startino; | |
1208 | pag->pagl_pagino = pagino; | |
1209 | goto newino; | |
1210 | } | |
1211 | ||
4254b0bb CH |
1212 | /* figure out the closer block if both are valid. */ |
1213 | if (!doneleft && !doneright) { | |
1214 | useleft = pagino - | |
1215 | (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) < | |
1216 | rec.ir_startino - pagino; | |
1217 | } else { | |
1218 | useleft = !doneleft; | |
1da177e4 | 1219 | } |
4254b0bb CH |
1220 | |
1221 | /* free inodes to the left? */ | |
1222 | if (useleft && trec.ir_freecount) { | |
1223 | rec = trec; | |
1224 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
1225 | cur = tcur; | |
bd169565 DC |
1226 | |
1227 | pag->pagl_leftrec = trec.ir_startino; | |
1228 | pag->pagl_rightrec = rec.ir_startino; | |
1229 | pag->pagl_pagino = pagino; | |
4254b0bb | 1230 | goto alloc_inode; |
1da177e4 | 1231 | } |
1da177e4 | 1232 | |
4254b0bb CH |
1233 | /* free inodes to the right? */ |
1234 | if (!useleft && rec.ir_freecount) { | |
1235 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
bd169565 DC |
1236 | |
1237 | pag->pagl_leftrec = trec.ir_startino; | |
1238 | pag->pagl_rightrec = rec.ir_startino; | |
1239 | pag->pagl_pagino = pagino; | |
4254b0bb | 1240 | goto alloc_inode; |
1da177e4 | 1241 | } |
4254b0bb CH |
1242 | |
1243 | /* get next record to check */ | |
1244 | if (useleft) { | |
1245 | error = xfs_ialloc_next_rec(tcur, &trec, | |
1246 | &doneleft, 1); | |
1247 | } else { | |
1248 | error = xfs_ialloc_next_rec(cur, &rec, | |
1249 | &doneright, 0); | |
1250 | } | |
1251 | if (error) | |
1252 | goto error1; | |
1da177e4 | 1253 | } |
bd169565 DC |
1254 | |
1255 | /* | |
1256 | * We've reached the end of the btree. because | |
1257 | * we are only searching a small chunk of the | |
1258 | * btree each search, there is obviously free | |
1259 | * inodes closer to the parent inode than we | |
1260 | * are now. restart the search again. | |
1261 | */ | |
1262 | pag->pagl_pagino = NULLAGINO; | |
1263 | pag->pagl_leftrec = NULLAGINO; | |
1264 | pag->pagl_rightrec = NULLAGINO; | |
1265 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
1266 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
1267 | goto restart_pagno; | |
1da177e4 | 1268 | } |
4254b0bb | 1269 | |
1da177e4 | 1270 | /* |
4254b0bb | 1271 | * In a different AG from the parent. |
1da177e4 LT |
1272 | * See if the most recently allocated block has any free. |
1273 | */ | |
bd169565 | 1274 | newino: |
69ef921b | 1275 | if (agi->agi_newino != cpu_to_be32(NULLAGINO)) { |
21875505 CH |
1276 | error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino), |
1277 | XFS_LOOKUP_EQ, &i); | |
4254b0bb | 1278 | if (error) |
1da177e4 | 1279 | goto error0; |
4254b0bb CH |
1280 | |
1281 | if (i == 1) { | |
1282 | error = xfs_inobt_get_rec(cur, &rec, &j); | |
1283 | if (error) | |
1284 | goto error0; | |
1285 | ||
1286 | if (j == 1 && rec.ir_freecount > 0) { | |
1287 | /* | |
1288 | * The last chunk allocated in the group | |
1289 | * still has a free inode. | |
1290 | */ | |
1291 | goto alloc_inode; | |
1292 | } | |
1da177e4 | 1293 | } |
bd169565 | 1294 | } |
4254b0bb | 1295 | |
bd169565 DC |
1296 | /* |
1297 | * None left in the last group, search the whole AG | |
1298 | */ | |
1299 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); | |
1300 | if (error) | |
1301 | goto error0; | |
c29aad41 | 1302 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
bd169565 DC |
1303 | |
1304 | for (;;) { | |
1305 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
1306 | if (error) | |
1307 | goto error0; | |
c29aad41 | 1308 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
bd169565 DC |
1309 | if (rec.ir_freecount > 0) |
1310 | break; | |
1311 | error = xfs_btree_increment(cur, 0, &i); | |
4254b0bb CH |
1312 | if (error) |
1313 | goto error0; | |
c29aad41 | 1314 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 | 1315 | } |
4254b0bb CH |
1316 | |
1317 | alloc_inode: | |
d4cc540b | 1318 | offset = xfs_inobt_first_free_inode(&rec); |
1da177e4 LT |
1319 | ASSERT(offset >= 0); |
1320 | ASSERT(offset < XFS_INODES_PER_CHUNK); | |
1321 | ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) % | |
1322 | XFS_INODES_PER_CHUNK) == 0); | |
1323 | ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset); | |
0d87e656 | 1324 | rec.ir_free &= ~XFS_INOBT_MASK(offset); |
1da177e4 | 1325 | rec.ir_freecount--; |
afabc24a CH |
1326 | error = xfs_inobt_update(cur, &rec); |
1327 | if (error) | |
1da177e4 | 1328 | goto error0; |
413d57c9 | 1329 | be32_add_cpu(&agi->agi_freecount, -1); |
1da177e4 | 1330 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); |
44b56e0a | 1331 | pag->pagi_freecount--; |
1da177e4 | 1332 | |
0b48db80 DC |
1333 | error = xfs_check_agi_freecount(cur, agi); |
1334 | if (error) | |
1335 | goto error0; | |
1336 | ||
1da177e4 LT |
1337 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
1338 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1); | |
44b56e0a | 1339 | xfs_perag_put(pag); |
1da177e4 LT |
1340 | *inop = ino; |
1341 | return 0; | |
1342 | error1: | |
1343 | xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR); | |
1344 | error0: | |
1345 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
44b56e0a | 1346 | xfs_perag_put(pag); |
1da177e4 LT |
1347 | return error; |
1348 | } | |
1349 | ||
6dd8638e BF |
1350 | /* |
1351 | * Use the free inode btree to allocate an inode based on distance from the | |
1352 | * parent. Note that the provided cursor may be deleted and replaced. | |
1353 | */ | |
1354 | STATIC int | |
1355 | xfs_dialloc_ag_finobt_near( | |
1356 | xfs_agino_t pagino, | |
1357 | struct xfs_btree_cur **ocur, | |
1358 | struct xfs_inobt_rec_incore *rec) | |
1359 | { | |
1360 | struct xfs_btree_cur *lcur = *ocur; /* left search cursor */ | |
1361 | struct xfs_btree_cur *rcur; /* right search cursor */ | |
1362 | struct xfs_inobt_rec_incore rrec; | |
1363 | int error; | |
1364 | int i, j; | |
1365 | ||
1366 | error = xfs_inobt_lookup(lcur, pagino, XFS_LOOKUP_LE, &i); | |
1367 | if (error) | |
1368 | return error; | |
1369 | ||
1370 | if (i == 1) { | |
1371 | error = xfs_inobt_get_rec(lcur, rec, &i); | |
1372 | if (error) | |
1373 | return error; | |
5fb5aeee | 1374 | XFS_WANT_CORRUPTED_RETURN(lcur->bc_mp, i == 1); |
6dd8638e BF |
1375 | |
1376 | /* | |
1377 | * See if we've landed in the parent inode record. The finobt | |
1378 | * only tracks chunks with at least one free inode, so record | |
1379 | * existence is enough. | |
1380 | */ | |
1381 | if (pagino >= rec->ir_startino && | |
1382 | pagino < (rec->ir_startino + XFS_INODES_PER_CHUNK)) | |
1383 | return 0; | |
1384 | } | |
1385 | ||
1386 | error = xfs_btree_dup_cursor(lcur, &rcur); | |
1387 | if (error) | |
1388 | return error; | |
1389 | ||
1390 | error = xfs_inobt_lookup(rcur, pagino, XFS_LOOKUP_GE, &j); | |
1391 | if (error) | |
1392 | goto error_rcur; | |
1393 | if (j == 1) { | |
1394 | error = xfs_inobt_get_rec(rcur, &rrec, &j); | |
1395 | if (error) | |
1396 | goto error_rcur; | |
c29aad41 | 1397 | XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, j == 1, error_rcur); |
6dd8638e BF |
1398 | } |
1399 | ||
c29aad41 | 1400 | XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, i == 1 || j == 1, error_rcur); |
6dd8638e BF |
1401 | if (i == 1 && j == 1) { |
1402 | /* | |
1403 | * Both the left and right records are valid. Choose the closer | |
1404 | * inode chunk to the target. | |
1405 | */ | |
1406 | if ((pagino - rec->ir_startino + XFS_INODES_PER_CHUNK - 1) > | |
1407 | (rrec.ir_startino - pagino)) { | |
1408 | *rec = rrec; | |
1409 | xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR); | |
1410 | *ocur = rcur; | |
1411 | } else { | |
1412 | xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR); | |
1413 | } | |
1414 | } else if (j == 1) { | |
1415 | /* only the right record is valid */ | |
1416 | *rec = rrec; | |
1417 | xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR); | |
1418 | *ocur = rcur; | |
1419 | } else if (i == 1) { | |
1420 | /* only the left record is valid */ | |
1421 | xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR); | |
1422 | } | |
1423 | ||
1424 | return 0; | |
1425 | ||
1426 | error_rcur: | |
1427 | xfs_btree_del_cursor(rcur, XFS_BTREE_ERROR); | |
1428 | return error; | |
1429 | } | |
1430 | ||
1431 | /* | |
1432 | * Use the free inode btree to find a free inode based on a newino hint. If | |
1433 | * the hint is NULL, find the first free inode in the AG. | |
1434 | */ | |
1435 | STATIC int | |
1436 | xfs_dialloc_ag_finobt_newino( | |
1437 | struct xfs_agi *agi, | |
1438 | struct xfs_btree_cur *cur, | |
1439 | struct xfs_inobt_rec_incore *rec) | |
1440 | { | |
1441 | int error; | |
1442 | int i; | |
1443 | ||
1444 | if (agi->agi_newino != cpu_to_be32(NULLAGINO)) { | |
e68ed775 DC |
1445 | error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino), |
1446 | XFS_LOOKUP_EQ, &i); | |
6dd8638e BF |
1447 | if (error) |
1448 | return error; | |
1449 | if (i == 1) { | |
1450 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1451 | if (error) | |
1452 | return error; | |
5fb5aeee | 1453 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
6dd8638e BF |
1454 | return 0; |
1455 | } | |
1456 | } | |
1457 | ||
1458 | /* | |
1459 | * Find the first inode available in the AG. | |
1460 | */ | |
1461 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); | |
1462 | if (error) | |
1463 | return error; | |
5fb5aeee | 1464 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
6dd8638e BF |
1465 | |
1466 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1467 | if (error) | |
1468 | return error; | |
5fb5aeee | 1469 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
6dd8638e BF |
1470 | |
1471 | return 0; | |
1472 | } | |
1473 | ||
1474 | /* | |
1475 | * Update the inobt based on a modification made to the finobt. Also ensure that | |
1476 | * the records from both trees are equivalent post-modification. | |
1477 | */ | |
1478 | STATIC int | |
1479 | xfs_dialloc_ag_update_inobt( | |
1480 | struct xfs_btree_cur *cur, /* inobt cursor */ | |
1481 | struct xfs_inobt_rec_incore *frec, /* finobt record */ | |
1482 | int offset) /* inode offset */ | |
1483 | { | |
1484 | struct xfs_inobt_rec_incore rec; | |
1485 | int error; | |
1486 | int i; | |
1487 | ||
1488 | error = xfs_inobt_lookup(cur, frec->ir_startino, XFS_LOOKUP_EQ, &i); | |
1489 | if (error) | |
1490 | return error; | |
5fb5aeee | 1491 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
6dd8638e BF |
1492 | |
1493 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
1494 | if (error) | |
1495 | return error; | |
5fb5aeee | 1496 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
6dd8638e BF |
1497 | ASSERT((XFS_AGINO_TO_OFFSET(cur->bc_mp, rec.ir_startino) % |
1498 | XFS_INODES_PER_CHUNK) == 0); | |
1499 | ||
1500 | rec.ir_free &= ~XFS_INOBT_MASK(offset); | |
1501 | rec.ir_freecount--; | |
1502 | ||
5fb5aeee | 1503 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, (rec.ir_free == frec->ir_free) && |
6dd8638e BF |
1504 | (rec.ir_freecount == frec->ir_freecount)); |
1505 | ||
b72091f2 | 1506 | return xfs_inobt_update(cur, &rec); |
6dd8638e BF |
1507 | } |
1508 | ||
1509 | /* | |
1510 | * Allocate an inode using the free inode btree, if available. Otherwise, fall | |
1511 | * back to the inobt search algorithm. | |
1512 | * | |
1513 | * The caller selected an AG for us, and made sure that free inodes are | |
1514 | * available. | |
1515 | */ | |
1516 | STATIC int | |
1517 | xfs_dialloc_ag( | |
1518 | struct xfs_trans *tp, | |
1519 | struct xfs_buf *agbp, | |
1520 | xfs_ino_t parent, | |
1521 | xfs_ino_t *inop) | |
1522 | { | |
1523 | struct xfs_mount *mp = tp->t_mountp; | |
1524 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
1525 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
1526 | xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent); | |
1527 | xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent); | |
1528 | struct xfs_perag *pag; | |
1529 | struct xfs_btree_cur *cur; /* finobt cursor */ | |
1530 | struct xfs_btree_cur *icur; /* inobt cursor */ | |
1531 | struct xfs_inobt_rec_incore rec; | |
1532 | xfs_ino_t ino; | |
1533 | int error; | |
1534 | int offset; | |
1535 | int i; | |
1536 | ||
1537 | if (!xfs_sb_version_hasfinobt(&mp->m_sb)) | |
1538 | return xfs_dialloc_ag_inobt(tp, agbp, parent, inop); | |
1539 | ||
1540 | pag = xfs_perag_get(mp, agno); | |
1541 | ||
1542 | /* | |
1543 | * If pagino is 0 (this is the root inode allocation) use newino. | |
1544 | * This must work because we've just allocated some. | |
1545 | */ | |
1546 | if (!pagino) | |
1547 | pagino = be32_to_cpu(agi->agi_newino); | |
1548 | ||
1549 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO); | |
1550 | ||
1551 | error = xfs_check_agi_freecount(cur, agi); | |
1552 | if (error) | |
1553 | goto error_cur; | |
1554 | ||
1555 | /* | |
1556 | * The search algorithm depends on whether we're in the same AG as the | |
1557 | * parent. If so, find the closest available inode to the parent. If | |
1558 | * not, consider the agi hint or find the first free inode in the AG. | |
1559 | */ | |
1560 | if (agno == pagno) | |
1561 | error = xfs_dialloc_ag_finobt_near(pagino, &cur, &rec); | |
1562 | else | |
1563 | error = xfs_dialloc_ag_finobt_newino(agi, cur, &rec); | |
1564 | if (error) | |
1565 | goto error_cur; | |
1566 | ||
d4cc540b | 1567 | offset = xfs_inobt_first_free_inode(&rec); |
6dd8638e BF |
1568 | ASSERT(offset >= 0); |
1569 | ASSERT(offset < XFS_INODES_PER_CHUNK); | |
1570 | ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) % | |
1571 | XFS_INODES_PER_CHUNK) == 0); | |
1572 | ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset); | |
1573 | ||
1574 | /* | |
1575 | * Modify or remove the finobt record. | |
1576 | */ | |
1577 | rec.ir_free &= ~XFS_INOBT_MASK(offset); | |
1578 | rec.ir_freecount--; | |
1579 | if (rec.ir_freecount) | |
1580 | error = xfs_inobt_update(cur, &rec); | |
1581 | else | |
1582 | error = xfs_btree_delete(cur, &i); | |
1583 | if (error) | |
1584 | goto error_cur; | |
1585 | ||
1586 | /* | |
1587 | * The finobt has now been updated appropriately. We haven't updated the | |
1588 | * agi and superblock yet, so we can create an inobt cursor and validate | |
1589 | * the original freecount. If all is well, make the equivalent update to | |
1590 | * the inobt using the finobt record and offset information. | |
1591 | */ | |
1592 | icur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); | |
1593 | ||
1594 | error = xfs_check_agi_freecount(icur, agi); | |
1595 | if (error) | |
1596 | goto error_icur; | |
1597 | ||
1598 | error = xfs_dialloc_ag_update_inobt(icur, &rec, offset); | |
1599 | if (error) | |
1600 | goto error_icur; | |
1601 | ||
1602 | /* | |
1603 | * Both trees have now been updated. We must update the perag and | |
1604 | * superblock before we can check the freecount for each btree. | |
1605 | */ | |
1606 | be32_add_cpu(&agi->agi_freecount, -1); | |
1607 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); | |
1608 | pag->pagi_freecount--; | |
1609 | ||
1610 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1); | |
1611 | ||
1612 | error = xfs_check_agi_freecount(icur, agi); | |
1613 | if (error) | |
1614 | goto error_icur; | |
1615 | error = xfs_check_agi_freecount(cur, agi); | |
1616 | if (error) | |
1617 | goto error_icur; | |
1618 | ||
1619 | xfs_btree_del_cursor(icur, XFS_BTREE_NOERROR); | |
1620 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
1621 | xfs_perag_put(pag); | |
1622 | *inop = ino; | |
1623 | return 0; | |
1624 | ||
1625 | error_icur: | |
1626 | xfs_btree_del_cursor(icur, XFS_BTREE_ERROR); | |
1627 | error_cur: | |
1628 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
1629 | xfs_perag_put(pag); | |
1630 | return error; | |
1631 | } | |
1632 | ||
f2ecc5e4 CH |
1633 | /* |
1634 | * Allocate an inode on disk. | |
1635 | * | |
1636 | * Mode is used to tell whether the new inode will need space, and whether it | |
1637 | * is a directory. | |
1638 | * | |
1639 | * This function is designed to be called twice if it has to do an allocation | |
1640 | * to make more free inodes. On the first call, *IO_agbp should be set to NULL. | |
1641 | * If an inode is available without having to performn an allocation, an inode | |
cd856db6 CM |
1642 | * number is returned. In this case, *IO_agbp is set to NULL. If an allocation |
1643 | * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp. | |
1644 | * The caller should then commit the current transaction, allocate a | |
f2ecc5e4 CH |
1645 | * new transaction, and call xfs_dialloc() again, passing in the previous value |
1646 | * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI | |
1647 | * buffer is locked across the two calls, the second call is guaranteed to have | |
1648 | * a free inode available. | |
1649 | * | |
1650 | * Once we successfully pick an inode its number is returned and the on-disk | |
1651 | * data structures are updated. The inode itself is not read in, since doing so | |
1652 | * would break ordering constraints with xfs_reclaim. | |
1653 | */ | |
1654 | int | |
1655 | xfs_dialloc( | |
1656 | struct xfs_trans *tp, | |
1657 | xfs_ino_t parent, | |
1658 | umode_t mode, | |
1659 | int okalloc, | |
1660 | struct xfs_buf **IO_agbp, | |
f2ecc5e4 CH |
1661 | xfs_ino_t *inop) |
1662 | { | |
55d6af64 | 1663 | struct xfs_mount *mp = tp->t_mountp; |
f2ecc5e4 CH |
1664 | struct xfs_buf *agbp; |
1665 | xfs_agnumber_t agno; | |
f2ecc5e4 CH |
1666 | int error; |
1667 | int ialloced; | |
1668 | int noroom = 0; | |
be60fe54 | 1669 | xfs_agnumber_t start_agno; |
f2ecc5e4 CH |
1670 | struct xfs_perag *pag; |
1671 | ||
4bb61069 | 1672 | if (*IO_agbp) { |
f2ecc5e4 | 1673 | /* |
4bb61069 CH |
1674 | * If the caller passes in a pointer to the AGI buffer, |
1675 | * continue where we left off before. In this case, we | |
f2ecc5e4 CH |
1676 | * know that the allocation group has free inodes. |
1677 | */ | |
1678 | agbp = *IO_agbp; | |
4bb61069 | 1679 | goto out_alloc; |
f2ecc5e4 | 1680 | } |
4bb61069 CH |
1681 | |
1682 | /* | |
1683 | * We do not have an agbp, so select an initial allocation | |
1684 | * group for inode allocation. | |
1685 | */ | |
be60fe54 CH |
1686 | start_agno = xfs_ialloc_ag_select(tp, parent, mode, okalloc); |
1687 | if (start_agno == NULLAGNUMBER) { | |
4bb61069 CH |
1688 | *inop = NULLFSINO; |
1689 | return 0; | |
1690 | } | |
55d6af64 | 1691 | |
f2ecc5e4 CH |
1692 | /* |
1693 | * If we have already hit the ceiling of inode blocks then clear | |
1694 | * okalloc so we scan all available agi structures for a free | |
1695 | * inode. | |
1696 | */ | |
f2ecc5e4 | 1697 | if (mp->m_maxicount && |
501ab323 DC |
1698 | percpu_counter_read(&mp->m_icount) + mp->m_ialloc_inos > |
1699 | mp->m_maxicount) { | |
f2ecc5e4 CH |
1700 | noroom = 1; |
1701 | okalloc = 0; | |
1702 | } | |
1703 | ||
1704 | /* | |
1705 | * Loop until we find an allocation group that either has free inodes | |
1706 | * or in which we can allocate some inodes. Iterate through the | |
1707 | * allocation groups upward, wrapping at the end. | |
1708 | */ | |
be60fe54 CH |
1709 | agno = start_agno; |
1710 | for (;;) { | |
1711 | pag = xfs_perag_get(mp, agno); | |
1712 | if (!pag->pagi_inodeok) { | |
1713 | xfs_ialloc_next_ag(mp); | |
1714 | goto nextag; | |
1715 | } | |
1716 | ||
1717 | if (!pag->pagi_init) { | |
1718 | error = xfs_ialloc_pagi_init(mp, tp, agno); | |
1719 | if (error) | |
1720 | goto out_error; | |
f2ecc5e4 | 1721 | } |
be60fe54 | 1722 | |
f2ecc5e4 | 1723 | /* |
be60fe54 | 1724 | * Do a first racy fast path check if this AG is usable. |
f2ecc5e4 | 1725 | */ |
be60fe54 CH |
1726 | if (!pag->pagi_freecount && !okalloc) |
1727 | goto nextag; | |
1728 | ||
c4982110 CH |
1729 | /* |
1730 | * Then read in the AGI buffer and recheck with the AGI buffer | |
1731 | * lock held. | |
1732 | */ | |
be60fe54 CH |
1733 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); |
1734 | if (error) | |
1735 | goto out_error; | |
1736 | ||
be60fe54 CH |
1737 | if (pag->pagi_freecount) { |
1738 | xfs_perag_put(pag); | |
1739 | goto out_alloc; | |
1740 | } | |
1741 | ||
c4982110 CH |
1742 | if (!okalloc) |
1743 | goto nextag_relse_buffer; | |
1744 | ||
be60fe54 CH |
1745 | |
1746 | error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced); | |
1747 | if (error) { | |
1748 | xfs_trans_brelse(tp, agbp); | |
1749 | ||
2451337d | 1750 | if (error != -ENOSPC) |
be60fe54 CH |
1751 | goto out_error; |
1752 | ||
1753 | xfs_perag_put(pag); | |
f2ecc5e4 | 1754 | *inop = NULLFSINO; |
be60fe54 | 1755 | return 0; |
f2ecc5e4 | 1756 | } |
be60fe54 CH |
1757 | |
1758 | if (ialloced) { | |
1759 | /* | |
1760 | * We successfully allocated some inodes, return | |
1761 | * the current context to the caller so that it | |
1762 | * can commit the current transaction and call | |
1763 | * us again where we left off. | |
1764 | */ | |
1765 | ASSERT(pag->pagi_freecount > 0); | |
f2ecc5e4 | 1766 | xfs_perag_put(pag); |
be60fe54 CH |
1767 | |
1768 | *IO_agbp = agbp; | |
1769 | *inop = NULLFSINO; | |
1770 | return 0; | |
f2ecc5e4 | 1771 | } |
be60fe54 | 1772 | |
c4982110 CH |
1773 | nextag_relse_buffer: |
1774 | xfs_trans_brelse(tp, agbp); | |
be60fe54 | 1775 | nextag: |
f2ecc5e4 | 1776 | xfs_perag_put(pag); |
be60fe54 CH |
1777 | if (++agno == mp->m_sb.sb_agcount) |
1778 | agno = 0; | |
1779 | if (agno == start_agno) { | |
1780 | *inop = NULLFSINO; | |
2451337d | 1781 | return noroom ? -ENOSPC : 0; |
be60fe54 | 1782 | } |
f2ecc5e4 CH |
1783 | } |
1784 | ||
4bb61069 | 1785 | out_alloc: |
f2ecc5e4 CH |
1786 | *IO_agbp = NULL; |
1787 | return xfs_dialloc_ag(tp, agbp, parent, inop); | |
be60fe54 CH |
1788 | out_error: |
1789 | xfs_perag_put(pag); | |
b474c7ae | 1790 | return error; |
f2ecc5e4 CH |
1791 | } |
1792 | ||
2b64ee5c BF |
1793 | STATIC int |
1794 | xfs_difree_inobt( | |
1795 | struct xfs_mount *mp, | |
1796 | struct xfs_trans *tp, | |
1797 | struct xfs_buf *agbp, | |
1798 | xfs_agino_t agino, | |
1799 | struct xfs_bmap_free *flist, | |
0d907a3b | 1800 | int *deleted, |
2b64ee5c BF |
1801 | xfs_ino_t *first_ino, |
1802 | struct xfs_inobt_rec_incore *orec) | |
1da177e4 | 1803 | { |
2b64ee5c BF |
1804 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); |
1805 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
1806 | struct xfs_perag *pag; | |
1807 | struct xfs_btree_cur *cur; | |
1808 | struct xfs_inobt_rec_incore rec; | |
1809 | int ilen; | |
1810 | int error; | |
1811 | int i; | |
1812 | int off; | |
1da177e4 | 1813 | |
69ef921b | 1814 | ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC)); |
2b64ee5c BF |
1815 | ASSERT(XFS_AGINO_TO_AGBNO(mp, agino) < be32_to_cpu(agi->agi_length)); |
1816 | ||
1da177e4 LT |
1817 | /* |
1818 | * Initialize the cursor. | |
1819 | */ | |
57bd3dbe | 1820 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
1da177e4 | 1821 | |
0b48db80 DC |
1822 | error = xfs_check_agi_freecount(cur, agi); |
1823 | if (error) | |
1824 | goto error0; | |
1825 | ||
1da177e4 LT |
1826 | /* |
1827 | * Look for the entry describing this inode. | |
1828 | */ | |
21875505 | 1829 | if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) { |
0b932ccc DC |
1830 | xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.", |
1831 | __func__, error); | |
1da177e4 LT |
1832 | goto error0; |
1833 | } | |
c29aad41 | 1834 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
2e287a73 CH |
1835 | error = xfs_inobt_get_rec(cur, &rec, &i); |
1836 | if (error) { | |
0b932ccc DC |
1837 | xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.", |
1838 | __func__, error); | |
1da177e4 LT |
1839 | goto error0; |
1840 | } | |
c29aad41 | 1841 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 LT |
1842 | /* |
1843 | * Get the offset in the inode chunk. | |
1844 | */ | |
1845 | off = agino - rec.ir_startino; | |
1846 | ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK); | |
0d87e656 | 1847 | ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off))); |
1da177e4 LT |
1848 | /* |
1849 | * Mark the inode free & increment the count. | |
1850 | */ | |
0d87e656 | 1851 | rec.ir_free |= XFS_INOBT_MASK(off); |
1da177e4 LT |
1852 | rec.ir_freecount++; |
1853 | ||
1854 | /* | |
999633d3 BF |
1855 | * When an inode chunk is free, it becomes eligible for removal. Don't |
1856 | * remove the chunk if the block size is large enough for multiple inode | |
1857 | * chunks (that might not be free). | |
1da177e4 | 1858 | */ |
1bd960ee | 1859 | if (!(mp->m_flags & XFS_MOUNT_IKEEP) && |
999633d3 BF |
1860 | rec.ir_free == XFS_INOBT_ALL_FREE && |
1861 | mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK) { | |
1da177e4 | 1862 | |
376c2f3a | 1863 | *deleted = 1; |
1da177e4 LT |
1864 | *first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino); |
1865 | ||
1866 | /* | |
1867 | * Remove the inode cluster from the AGI B+Tree, adjust the | |
1868 | * AGI and Superblock inode counts, and mark the disk space | |
1869 | * to be freed when the transaction is committed. | |
1870 | */ | |
999633d3 | 1871 | ilen = rec.ir_freecount; |
413d57c9 MS |
1872 | be32_add_cpu(&agi->agi_count, -ilen); |
1873 | be32_add_cpu(&agi->agi_freecount, -(ilen - 1)); | |
1da177e4 | 1874 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT); |
44b56e0a DC |
1875 | pag = xfs_perag_get(mp, agno); |
1876 | pag->pagi_freecount -= ilen - 1; | |
1877 | xfs_perag_put(pag); | |
1da177e4 LT |
1878 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen); |
1879 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1)); | |
1880 | ||
91cca5df | 1881 | if ((error = xfs_btree_delete(cur, &i))) { |
0b932ccc DC |
1882 | xfs_warn(mp, "%s: xfs_btree_delete returned error %d.", |
1883 | __func__, error); | |
1da177e4 LT |
1884 | goto error0; |
1885 | } | |
1886 | ||
126cd105 JL |
1887 | xfs_bmap_add_free(XFS_AGB_TO_FSB(mp, agno, |
1888 | XFS_AGINO_TO_AGBNO(mp, rec.ir_startino)), | |
1889 | mp->m_ialloc_blks, flist, mp); | |
1da177e4 | 1890 | } else { |
376c2f3a | 1891 | *deleted = 0; |
1da177e4 | 1892 | |
afabc24a CH |
1893 | error = xfs_inobt_update(cur, &rec); |
1894 | if (error) { | |
0b932ccc DC |
1895 | xfs_warn(mp, "%s: xfs_inobt_update returned error %d.", |
1896 | __func__, error); | |
1da177e4 LT |
1897 | goto error0; |
1898 | } | |
afabc24a | 1899 | |
1da177e4 LT |
1900 | /* |
1901 | * Change the inode free counts and log the ag/sb changes. | |
1902 | */ | |
413d57c9 | 1903 | be32_add_cpu(&agi->agi_freecount, 1); |
1da177e4 | 1904 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); |
44b56e0a DC |
1905 | pag = xfs_perag_get(mp, agno); |
1906 | pag->pagi_freecount++; | |
1907 | xfs_perag_put(pag); | |
1da177e4 LT |
1908 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1); |
1909 | } | |
1910 | ||
0b48db80 DC |
1911 | error = xfs_check_agi_freecount(cur, agi); |
1912 | if (error) | |
1913 | goto error0; | |
1da177e4 | 1914 | |
2b64ee5c | 1915 | *orec = rec; |
1da177e4 LT |
1916 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
1917 | return 0; | |
1918 | ||
1919 | error0: | |
1920 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
1921 | return error; | |
1922 | } | |
1923 | ||
3efa4ffd BF |
1924 | /* |
1925 | * Free an inode in the free inode btree. | |
1926 | */ | |
1927 | STATIC int | |
1928 | xfs_difree_finobt( | |
1929 | struct xfs_mount *mp, | |
1930 | struct xfs_trans *tp, | |
1931 | struct xfs_buf *agbp, | |
1932 | xfs_agino_t agino, | |
1933 | struct xfs_inobt_rec_incore *ibtrec) /* inobt record */ | |
1934 | { | |
1935 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
1936 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
1937 | struct xfs_btree_cur *cur; | |
1938 | struct xfs_inobt_rec_incore rec; | |
1939 | int offset = agino - ibtrec->ir_startino; | |
1940 | int error; | |
1941 | int i; | |
1942 | ||
1943 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO); | |
1944 | ||
1945 | error = xfs_inobt_lookup(cur, ibtrec->ir_startino, XFS_LOOKUP_EQ, &i); | |
1946 | if (error) | |
1947 | goto error; | |
1948 | if (i == 0) { | |
1949 | /* | |
1950 | * If the record does not exist in the finobt, we must have just | |
1951 | * freed an inode in a previously fully allocated chunk. If not, | |
1952 | * something is out of sync. | |
1953 | */ | |
c29aad41 | 1954 | XFS_WANT_CORRUPTED_GOTO(mp, ibtrec->ir_freecount == 1, error); |
3efa4ffd | 1955 | |
5419040f BF |
1956 | error = xfs_inobt_insert_rec(cur, ibtrec->ir_holemask, |
1957 | ibtrec->ir_count, | |
1958 | ibtrec->ir_freecount, | |
3efa4ffd BF |
1959 | ibtrec->ir_free, &i); |
1960 | if (error) | |
1961 | goto error; | |
1962 | ASSERT(i == 1); | |
1963 | ||
1964 | goto out; | |
1965 | } | |
1966 | ||
1967 | /* | |
1968 | * Read and update the existing record. We could just copy the ibtrec | |
1969 | * across here, but that would defeat the purpose of having redundant | |
1970 | * metadata. By making the modifications independently, we can catch | |
1971 | * corruptions that we wouldn't see if we just copied from one record | |
1972 | * to another. | |
1973 | */ | |
1974 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
1975 | if (error) | |
1976 | goto error; | |
c29aad41 | 1977 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error); |
3efa4ffd BF |
1978 | |
1979 | rec.ir_free |= XFS_INOBT_MASK(offset); | |
1980 | rec.ir_freecount++; | |
1981 | ||
c29aad41 | 1982 | XFS_WANT_CORRUPTED_GOTO(mp, (rec.ir_free == ibtrec->ir_free) && |
3efa4ffd BF |
1983 | (rec.ir_freecount == ibtrec->ir_freecount), |
1984 | error); | |
1985 | ||
1986 | /* | |
1987 | * The content of inobt records should always match between the inobt | |
1988 | * and finobt. The lifecycle of records in the finobt is different from | |
1989 | * the inobt in that the finobt only tracks records with at least one | |
1990 | * free inode. Hence, if all of the inodes are free and we aren't | |
1991 | * keeping inode chunks permanently on disk, remove the record. | |
1992 | * Otherwise, update the record with the new information. | |
999633d3 BF |
1993 | * |
1994 | * Note that we currently can't free chunks when the block size is large | |
1995 | * enough for multiple chunks. Leave the finobt record to remain in sync | |
1996 | * with the inobt. | |
3efa4ffd | 1997 | */ |
999633d3 BF |
1998 | if (rec.ir_free == XFS_INOBT_ALL_FREE && |
1999 | mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK && | |
3efa4ffd BF |
2000 | !(mp->m_flags & XFS_MOUNT_IKEEP)) { |
2001 | error = xfs_btree_delete(cur, &i); | |
2002 | if (error) | |
2003 | goto error; | |
2004 | ASSERT(i == 1); | |
2005 | } else { | |
2006 | error = xfs_inobt_update(cur, &rec); | |
2007 | if (error) | |
2008 | goto error; | |
2009 | } | |
2010 | ||
2011 | out: | |
2012 | error = xfs_check_agi_freecount(cur, agi); | |
2013 | if (error) | |
2014 | goto error; | |
2015 | ||
2016 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
2017 | return 0; | |
2018 | ||
2019 | error: | |
2020 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
2021 | return error; | |
2022 | } | |
2023 | ||
2b64ee5c BF |
2024 | /* |
2025 | * Free disk inode. Carefully avoids touching the incore inode, all | |
2026 | * manipulations incore are the caller's responsibility. | |
2027 | * The on-disk inode is not changed by this operation, only the | |
2028 | * btree (free inode mask) is changed. | |
2029 | */ | |
2030 | int | |
2031 | xfs_difree( | |
2032 | struct xfs_trans *tp, /* transaction pointer */ | |
2033 | xfs_ino_t inode, /* inode to be freed */ | |
2034 | struct xfs_bmap_free *flist, /* extents to free */ | |
0d907a3b | 2035 | int *deleted,/* set if inode cluster was deleted */ |
2b64ee5c BF |
2036 | xfs_ino_t *first_ino)/* first inode in deleted cluster */ |
2037 | { | |
2038 | /* REFERENCED */ | |
2039 | xfs_agblock_t agbno; /* block number containing inode */ | |
2040 | struct xfs_buf *agbp; /* buffer for allocation group header */ | |
2041 | xfs_agino_t agino; /* allocation group inode number */ | |
2042 | xfs_agnumber_t agno; /* allocation group number */ | |
2043 | int error; /* error return value */ | |
2044 | struct xfs_mount *mp; /* mount structure for filesystem */ | |
2045 | struct xfs_inobt_rec_incore rec;/* btree record */ | |
2046 | ||
2047 | mp = tp->t_mountp; | |
2048 | ||
2049 | /* | |
2050 | * Break up inode number into its components. | |
2051 | */ | |
2052 | agno = XFS_INO_TO_AGNO(mp, inode); | |
2053 | if (agno >= mp->m_sb.sb_agcount) { | |
2054 | xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).", | |
2055 | __func__, agno, mp->m_sb.sb_agcount); | |
2056 | ASSERT(0); | |
2451337d | 2057 | return -EINVAL; |
2b64ee5c BF |
2058 | } |
2059 | agino = XFS_INO_TO_AGINO(mp, inode); | |
2060 | if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
2061 | xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).", | |
2062 | __func__, (unsigned long long)inode, | |
2063 | (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino)); | |
2064 | ASSERT(0); | |
2451337d | 2065 | return -EINVAL; |
2b64ee5c BF |
2066 | } |
2067 | agbno = XFS_AGINO_TO_AGBNO(mp, agino); | |
2068 | if (agbno >= mp->m_sb.sb_agblocks) { | |
2069 | xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).", | |
2070 | __func__, agbno, mp->m_sb.sb_agblocks); | |
2071 | ASSERT(0); | |
2451337d | 2072 | return -EINVAL; |
2b64ee5c BF |
2073 | } |
2074 | /* | |
2075 | * Get the allocation group header. | |
2076 | */ | |
2077 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); | |
2078 | if (error) { | |
2079 | xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.", | |
2080 | __func__, error); | |
2081 | return error; | |
2082 | } | |
2083 | ||
2084 | /* | |
2085 | * Fix up the inode allocation btree. | |
2086 | */ | |
0d907a3b | 2087 | error = xfs_difree_inobt(mp, tp, agbp, agino, flist, deleted, first_ino, |
2b64ee5c BF |
2088 | &rec); |
2089 | if (error) | |
2090 | goto error0; | |
2091 | ||
3efa4ffd BF |
2092 | /* |
2093 | * Fix up the free inode btree. | |
2094 | */ | |
2095 | if (xfs_sb_version_hasfinobt(&mp->m_sb)) { | |
2096 | error = xfs_difree_finobt(mp, tp, agbp, agino, &rec); | |
2097 | if (error) | |
2098 | goto error0; | |
2099 | } | |
2100 | ||
2b64ee5c BF |
2101 | return 0; |
2102 | ||
2103 | error0: | |
2104 | return error; | |
2105 | } | |
2106 | ||
7124fe0a DC |
2107 | STATIC int |
2108 | xfs_imap_lookup( | |
2109 | struct xfs_mount *mp, | |
2110 | struct xfs_trans *tp, | |
2111 | xfs_agnumber_t agno, | |
2112 | xfs_agino_t agino, | |
2113 | xfs_agblock_t agbno, | |
2114 | xfs_agblock_t *chunk_agbno, | |
2115 | xfs_agblock_t *offset_agbno, | |
2116 | int flags) | |
2117 | { | |
2118 | struct xfs_inobt_rec_incore rec; | |
2119 | struct xfs_btree_cur *cur; | |
2120 | struct xfs_buf *agbp; | |
7124fe0a DC |
2121 | int error; |
2122 | int i; | |
2123 | ||
2124 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); | |
2125 | if (error) { | |
53487786 DC |
2126 | xfs_alert(mp, |
2127 | "%s: xfs_ialloc_read_agi() returned error %d, agno %d", | |
2128 | __func__, error, agno); | |
7124fe0a DC |
2129 | return error; |
2130 | } | |
2131 | ||
2132 | /* | |
4536f2ad DC |
2133 | * Lookup the inode record for the given agino. If the record cannot be |
2134 | * found, then it's an invalid inode number and we should abort. Once | |
2135 | * we have a record, we need to ensure it contains the inode number | |
2136 | * we are looking up. | |
7124fe0a | 2137 | */ |
57bd3dbe | 2138 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
4536f2ad | 2139 | error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i); |
7124fe0a DC |
2140 | if (!error) { |
2141 | if (i) | |
2142 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
2143 | if (!error && i == 0) | |
2451337d | 2144 | error = -EINVAL; |
7124fe0a DC |
2145 | } |
2146 | ||
2147 | xfs_trans_brelse(tp, agbp); | |
2148 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
2149 | if (error) | |
2150 | return error; | |
2151 | ||
4536f2ad DC |
2152 | /* check that the returned record contains the required inode */ |
2153 | if (rec.ir_startino > agino || | |
71783438 | 2154 | rec.ir_startino + mp->m_ialloc_inos <= agino) |
2451337d | 2155 | return -EINVAL; |
4536f2ad | 2156 | |
7124fe0a | 2157 | /* for untrusted inodes check it is allocated first */ |
1920779e | 2158 | if ((flags & XFS_IGET_UNTRUSTED) && |
7124fe0a | 2159 | (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino))) |
2451337d | 2160 | return -EINVAL; |
7124fe0a DC |
2161 | |
2162 | *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino); | |
2163 | *offset_agbno = agbno - *chunk_agbno; | |
2164 | return 0; | |
2165 | } | |
2166 | ||
1da177e4 | 2167 | /* |
94e1b69d | 2168 | * Return the location of the inode in imap, for mapping it into a buffer. |
1da177e4 | 2169 | */ |
1da177e4 | 2170 | int |
94e1b69d CH |
2171 | xfs_imap( |
2172 | xfs_mount_t *mp, /* file system mount structure */ | |
2173 | xfs_trans_t *tp, /* transaction pointer */ | |
1da177e4 | 2174 | xfs_ino_t ino, /* inode to locate */ |
94e1b69d CH |
2175 | struct xfs_imap *imap, /* location map structure */ |
2176 | uint flags) /* flags for inode btree lookup */ | |
1da177e4 LT |
2177 | { |
2178 | xfs_agblock_t agbno; /* block number of inode in the alloc group */ | |
1da177e4 LT |
2179 | xfs_agino_t agino; /* inode number within alloc group */ |
2180 | xfs_agnumber_t agno; /* allocation group number */ | |
2181 | int blks_per_cluster; /* num blocks per inode cluster */ | |
2182 | xfs_agblock_t chunk_agbno; /* first block in inode chunk */ | |
1da177e4 | 2183 | xfs_agblock_t cluster_agbno; /* first block in inode cluster */ |
1da177e4 | 2184 | int error; /* error code */ |
1da177e4 | 2185 | int offset; /* index of inode in its buffer */ |
836a94ad | 2186 | xfs_agblock_t offset_agbno; /* blks from chunk start to inode */ |
1da177e4 LT |
2187 | |
2188 | ASSERT(ino != NULLFSINO); | |
94e1b69d | 2189 | |
1da177e4 LT |
2190 | /* |
2191 | * Split up the inode number into its parts. | |
2192 | */ | |
2193 | agno = XFS_INO_TO_AGNO(mp, ino); | |
2194 | agino = XFS_INO_TO_AGINO(mp, ino); | |
2195 | agbno = XFS_AGINO_TO_AGBNO(mp, agino); | |
2196 | if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks || | |
2197 | ino != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
2198 | #ifdef DEBUG | |
1920779e DC |
2199 | /* |
2200 | * Don't output diagnostic information for untrusted inodes | |
2201 | * as they can be invalid without implying corruption. | |
2202 | */ | |
2203 | if (flags & XFS_IGET_UNTRUSTED) | |
2451337d | 2204 | return -EINVAL; |
1da177e4 | 2205 | if (agno >= mp->m_sb.sb_agcount) { |
53487786 DC |
2206 | xfs_alert(mp, |
2207 | "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)", | |
2208 | __func__, agno, mp->m_sb.sb_agcount); | |
1da177e4 LT |
2209 | } |
2210 | if (agbno >= mp->m_sb.sb_agblocks) { | |
53487786 DC |
2211 | xfs_alert(mp, |
2212 | "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)", | |
2213 | __func__, (unsigned long long)agbno, | |
2214 | (unsigned long)mp->m_sb.sb_agblocks); | |
1da177e4 LT |
2215 | } |
2216 | if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
53487786 DC |
2217 | xfs_alert(mp, |
2218 | "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)", | |
2219 | __func__, ino, | |
2220 | XFS_AGINO_TO_INO(mp, agno, agino)); | |
1da177e4 | 2221 | } |
745b1f47 | 2222 | xfs_stack_trace(); |
1da177e4 | 2223 | #endif /* DEBUG */ |
2451337d | 2224 | return -EINVAL; |
1da177e4 | 2225 | } |
94e1b69d | 2226 | |
f9e5abcf | 2227 | blks_per_cluster = xfs_icluster_size_fsb(mp); |
7124fe0a DC |
2228 | |
2229 | /* | |
2230 | * For bulkstat and handle lookups, we have an untrusted inode number | |
2231 | * that we have to verify is valid. We cannot do this just by reading | |
2232 | * the inode buffer as it may have been unlinked and removed leaving | |
2233 | * inodes in stale state on disk. Hence we have to do a btree lookup | |
2234 | * in all cases where an untrusted inode number is passed. | |
2235 | */ | |
1920779e | 2236 | if (flags & XFS_IGET_UNTRUSTED) { |
7124fe0a DC |
2237 | error = xfs_imap_lookup(mp, tp, agno, agino, agbno, |
2238 | &chunk_agbno, &offset_agbno, flags); | |
2239 | if (error) | |
2240 | return error; | |
2241 | goto out_map; | |
2242 | } | |
2243 | ||
94e1b69d CH |
2244 | /* |
2245 | * If the inode cluster size is the same as the blocksize or | |
2246 | * smaller we get to the buffer by simple arithmetics. | |
2247 | */ | |
f9e5abcf | 2248 | if (blks_per_cluster == 1) { |
1da177e4 LT |
2249 | offset = XFS_INO_TO_OFFSET(mp, ino); |
2250 | ASSERT(offset < mp->m_sb.sb_inopblock); | |
94e1b69d CH |
2251 | |
2252 | imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno); | |
2253 | imap->im_len = XFS_FSB_TO_BB(mp, 1); | |
2254 | imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog); | |
1da177e4 LT |
2255 | return 0; |
2256 | } | |
94e1b69d | 2257 | |
94e1b69d CH |
2258 | /* |
2259 | * If the inode chunks are aligned then use simple maths to | |
2260 | * find the location. Otherwise we have to do a btree | |
2261 | * lookup to find the location. | |
2262 | */ | |
1da177e4 LT |
2263 | if (mp->m_inoalign_mask) { |
2264 | offset_agbno = agbno & mp->m_inoalign_mask; | |
2265 | chunk_agbno = agbno - offset_agbno; | |
2266 | } else { | |
7124fe0a DC |
2267 | error = xfs_imap_lookup(mp, tp, agno, agino, agbno, |
2268 | &chunk_agbno, &offset_agbno, flags); | |
1da177e4 LT |
2269 | if (error) |
2270 | return error; | |
1da177e4 | 2271 | } |
94e1b69d | 2272 | |
7124fe0a | 2273 | out_map: |
1da177e4 LT |
2274 | ASSERT(agbno >= chunk_agbno); |
2275 | cluster_agbno = chunk_agbno + | |
2276 | ((offset_agbno / blks_per_cluster) * blks_per_cluster); | |
2277 | offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) + | |
2278 | XFS_INO_TO_OFFSET(mp, ino); | |
94e1b69d CH |
2279 | |
2280 | imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno); | |
2281 | imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster); | |
2282 | imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog); | |
2283 | ||
2284 | /* | |
2285 | * If the inode number maps to a block outside the bounds | |
2286 | * of the file system then return NULL rather than calling | |
2287 | * read_buf and panicing when we get an error from the | |
2288 | * driver. | |
2289 | */ | |
2290 | if ((imap->im_blkno + imap->im_len) > | |
2291 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) { | |
53487786 DC |
2292 | xfs_alert(mp, |
2293 | "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)", | |
2294 | __func__, (unsigned long long) imap->im_blkno, | |
94e1b69d CH |
2295 | (unsigned long long) imap->im_len, |
2296 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)); | |
2451337d | 2297 | return -EINVAL; |
94e1b69d | 2298 | } |
1da177e4 | 2299 | return 0; |
1da177e4 LT |
2300 | } |
2301 | ||
2302 | /* | |
2303 | * Compute and fill in value of m_in_maxlevels. | |
2304 | */ | |
2305 | void | |
2306 | xfs_ialloc_compute_maxlevels( | |
2307 | xfs_mount_t *mp) /* file system mount structure */ | |
2308 | { | |
2309 | int level; | |
2310 | uint maxblocks; | |
2311 | uint maxleafents; | |
2312 | int minleafrecs; | |
2313 | int minnoderecs; | |
2314 | ||
2315 | maxleafents = (1LL << XFS_INO_AGINO_BITS(mp)) >> | |
2316 | XFS_INODES_PER_CHUNK_LOG; | |
2317 | minleafrecs = mp->m_alloc_mnr[0]; | |
2318 | minnoderecs = mp->m_alloc_mnr[1]; | |
2319 | maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs; | |
2320 | for (level = 1; maxblocks > 1; level++) | |
2321 | maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs; | |
2322 | mp->m_in_maxlevels = level; | |
2323 | } | |
2324 | ||
2325 | /* | |
aafc3c24 BF |
2326 | * Log specified fields for the ag hdr (inode section). The growth of the agi |
2327 | * structure over time requires that we interpret the buffer as two logical | |
2328 | * regions delineated by the end of the unlinked list. This is due to the size | |
2329 | * of the hash table and its location in the middle of the agi. | |
2330 | * | |
2331 | * For example, a request to log a field before agi_unlinked and a field after | |
2332 | * agi_unlinked could cause us to log the entire hash table and use an excessive | |
2333 | * amount of log space. To avoid this behavior, log the region up through | |
2334 | * agi_unlinked in one call and the region after agi_unlinked through the end of | |
2335 | * the structure in another. | |
1da177e4 LT |
2336 | */ |
2337 | void | |
2338 | xfs_ialloc_log_agi( | |
2339 | xfs_trans_t *tp, /* transaction pointer */ | |
2340 | xfs_buf_t *bp, /* allocation group header buffer */ | |
2341 | int fields) /* bitmask of fields to log */ | |
2342 | { | |
2343 | int first; /* first byte number */ | |
2344 | int last; /* last byte number */ | |
2345 | static const short offsets[] = { /* field starting offsets */ | |
2346 | /* keep in sync with bit definitions */ | |
2347 | offsetof(xfs_agi_t, agi_magicnum), | |
2348 | offsetof(xfs_agi_t, agi_versionnum), | |
2349 | offsetof(xfs_agi_t, agi_seqno), | |
2350 | offsetof(xfs_agi_t, agi_length), | |
2351 | offsetof(xfs_agi_t, agi_count), | |
2352 | offsetof(xfs_agi_t, agi_root), | |
2353 | offsetof(xfs_agi_t, agi_level), | |
2354 | offsetof(xfs_agi_t, agi_freecount), | |
2355 | offsetof(xfs_agi_t, agi_newino), | |
2356 | offsetof(xfs_agi_t, agi_dirino), | |
2357 | offsetof(xfs_agi_t, agi_unlinked), | |
aafc3c24 BF |
2358 | offsetof(xfs_agi_t, agi_free_root), |
2359 | offsetof(xfs_agi_t, agi_free_level), | |
1da177e4 LT |
2360 | sizeof(xfs_agi_t) |
2361 | }; | |
2362 | #ifdef DEBUG | |
2363 | xfs_agi_t *agi; /* allocation group header */ | |
2364 | ||
2365 | agi = XFS_BUF_TO_AGI(bp); | |
69ef921b | 2366 | ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC)); |
1da177e4 | 2367 | #endif |
aafc3c24 BF |
2368 | |
2369 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_AGI_BUF); | |
2370 | ||
1da177e4 | 2371 | /* |
aafc3c24 BF |
2372 | * Compute byte offsets for the first and last fields in the first |
2373 | * region and log the agi buffer. This only logs up through | |
2374 | * agi_unlinked. | |
1da177e4 | 2375 | */ |
aafc3c24 BF |
2376 | if (fields & XFS_AGI_ALL_BITS_R1) { |
2377 | xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R1, | |
2378 | &first, &last); | |
2379 | xfs_trans_log_buf(tp, bp, first, last); | |
2380 | } | |
2381 | ||
1da177e4 | 2382 | /* |
aafc3c24 BF |
2383 | * Mask off the bits in the first region and calculate the first and |
2384 | * last field offsets for any bits in the second region. | |
1da177e4 | 2385 | */ |
aafc3c24 BF |
2386 | fields &= ~XFS_AGI_ALL_BITS_R1; |
2387 | if (fields) { | |
2388 | xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R2, | |
2389 | &first, &last); | |
2390 | xfs_trans_log_buf(tp, bp, first, last); | |
2391 | } | |
1da177e4 LT |
2392 | } |
2393 | ||
5e1be0fb CH |
2394 | #ifdef DEBUG |
2395 | STATIC void | |
2396 | xfs_check_agi_unlinked( | |
2397 | struct xfs_agi *agi) | |
2398 | { | |
2399 | int i; | |
2400 | ||
2401 | for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++) | |
2402 | ASSERT(agi->agi_unlinked[i]); | |
2403 | } | |
2404 | #else | |
2405 | #define xfs_check_agi_unlinked(agi) | |
2406 | #endif | |
2407 | ||
983d09ff | 2408 | static bool |
612cfbfe | 2409 | xfs_agi_verify( |
3702ce6e DC |
2410 | struct xfs_buf *bp) |
2411 | { | |
2412 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
2413 | struct xfs_agi *agi = XFS_BUF_TO_AGI(bp); | |
3702ce6e | 2414 | |
983d09ff DC |
2415 | if (xfs_sb_version_hascrc(&mp->m_sb) && |
2416 | !uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_uuid)) | |
2417 | return false; | |
3702ce6e DC |
2418 | /* |
2419 | * Validate the magic number of the agi block. | |
2420 | */ | |
983d09ff DC |
2421 | if (agi->agi_magicnum != cpu_to_be32(XFS_AGI_MAGIC)) |
2422 | return false; | |
2423 | if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum))) | |
2424 | return false; | |
3702ce6e | 2425 | |
e1b05723 ES |
2426 | if (be32_to_cpu(agi->agi_level) > XFS_BTREE_MAXLEVELS) |
2427 | return false; | |
3702ce6e DC |
2428 | /* |
2429 | * during growfs operations, the perag is not fully initialised, | |
2430 | * so we can't use it for any useful checking. growfs ensures we can't | |
2431 | * use it by using uncached buffers that don't have the perag attached | |
2432 | * so we can detect and avoid this problem. | |
2433 | */ | |
983d09ff DC |
2434 | if (bp->b_pag && be32_to_cpu(agi->agi_seqno) != bp->b_pag->pag_agno) |
2435 | return false; | |
3702ce6e | 2436 | |
3702ce6e | 2437 | xfs_check_agi_unlinked(agi); |
983d09ff | 2438 | return true; |
612cfbfe DC |
2439 | } |
2440 | ||
1813dd64 DC |
2441 | static void |
2442 | xfs_agi_read_verify( | |
612cfbfe DC |
2443 | struct xfs_buf *bp) |
2444 | { | |
983d09ff | 2445 | struct xfs_mount *mp = bp->b_target->bt_mount; |
983d09ff | 2446 | |
ce5028cf ES |
2447 | if (xfs_sb_version_hascrc(&mp->m_sb) && |
2448 | !xfs_buf_verify_cksum(bp, XFS_AGI_CRC_OFF)) | |
2451337d | 2449 | xfs_buf_ioerror(bp, -EFSBADCRC); |
ce5028cf ES |
2450 | else if (XFS_TEST_ERROR(!xfs_agi_verify(bp), mp, |
2451 | XFS_ERRTAG_IALLOC_READ_AGI, | |
2452 | XFS_RANDOM_IALLOC_READ_AGI)) | |
2451337d | 2453 | xfs_buf_ioerror(bp, -EFSCORRUPTED); |
ce5028cf ES |
2454 | |
2455 | if (bp->b_error) | |
2456 | xfs_verifier_error(bp); | |
612cfbfe DC |
2457 | } |
2458 | ||
b0f539de | 2459 | static void |
1813dd64 | 2460 | xfs_agi_write_verify( |
612cfbfe DC |
2461 | struct xfs_buf *bp) |
2462 | { | |
983d09ff DC |
2463 | struct xfs_mount *mp = bp->b_target->bt_mount; |
2464 | struct xfs_buf_log_item *bip = bp->b_fspriv; | |
2465 | ||
2466 | if (!xfs_agi_verify(bp)) { | |
2451337d | 2467 | xfs_buf_ioerror(bp, -EFSCORRUPTED); |
ce5028cf | 2468 | xfs_verifier_error(bp); |
983d09ff DC |
2469 | return; |
2470 | } | |
2471 | ||
2472 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
2473 | return; | |
2474 | ||
2475 | if (bip) | |
2476 | XFS_BUF_TO_AGI(bp)->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn); | |
f1dbcd7e | 2477 | xfs_buf_update_cksum(bp, XFS_AGI_CRC_OFF); |
3702ce6e DC |
2478 | } |
2479 | ||
1813dd64 DC |
2480 | const struct xfs_buf_ops xfs_agi_buf_ops = { |
2481 | .verify_read = xfs_agi_read_verify, | |
2482 | .verify_write = xfs_agi_write_verify, | |
2483 | }; | |
2484 | ||
1da177e4 LT |
2485 | /* |
2486 | * Read in the allocation group header (inode allocation section) | |
2487 | */ | |
2488 | int | |
5e1be0fb CH |
2489 | xfs_read_agi( |
2490 | struct xfs_mount *mp, /* file system mount structure */ | |
2491 | struct xfs_trans *tp, /* transaction pointer */ | |
2492 | xfs_agnumber_t agno, /* allocation group number */ | |
2493 | struct xfs_buf **bpp) /* allocation group hdr buf */ | |
1da177e4 | 2494 | { |
5e1be0fb | 2495 | int error; |
1da177e4 | 2496 | |
d123031a | 2497 | trace_xfs_read_agi(mp, agno); |
5e1be0fb | 2498 | |
d123031a | 2499 | ASSERT(agno != NULLAGNUMBER); |
5e1be0fb | 2500 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, |
1da177e4 | 2501 | XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)), |
1813dd64 | 2502 | XFS_FSS_TO_BB(mp, 1), 0, bpp, &xfs_agi_buf_ops); |
1da177e4 LT |
2503 | if (error) |
2504 | return error; | |
5e1be0fb | 2505 | |
38f23232 | 2506 | xfs_buf_set_ref(*bpp, XFS_AGI_REF); |
5e1be0fb CH |
2507 | return 0; |
2508 | } | |
2509 | ||
2510 | int | |
2511 | xfs_ialloc_read_agi( | |
2512 | struct xfs_mount *mp, /* file system mount structure */ | |
2513 | struct xfs_trans *tp, /* transaction pointer */ | |
2514 | xfs_agnumber_t agno, /* allocation group number */ | |
2515 | struct xfs_buf **bpp) /* allocation group hdr buf */ | |
2516 | { | |
2517 | struct xfs_agi *agi; /* allocation group header */ | |
2518 | struct xfs_perag *pag; /* per allocation group data */ | |
2519 | int error; | |
2520 | ||
d123031a DC |
2521 | trace_xfs_ialloc_read_agi(mp, agno); |
2522 | ||
5e1be0fb CH |
2523 | error = xfs_read_agi(mp, tp, agno, bpp); |
2524 | if (error) | |
2525 | return error; | |
2526 | ||
2527 | agi = XFS_BUF_TO_AGI(*bpp); | |
44b56e0a | 2528 | pag = xfs_perag_get(mp, agno); |
1da177e4 | 2529 | if (!pag->pagi_init) { |
16259e7d | 2530 | pag->pagi_freecount = be32_to_cpu(agi->agi_freecount); |
92821e2b | 2531 | pag->pagi_count = be32_to_cpu(agi->agi_count); |
1da177e4 | 2532 | pag->pagi_init = 1; |
1da177e4 | 2533 | } |
1da177e4 | 2534 | |
5e1be0fb CH |
2535 | /* |
2536 | * It's possible for these to be out of sync if | |
2537 | * we are in the middle of a forced shutdown. | |
2538 | */ | |
2539 | ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) || | |
2540 | XFS_FORCED_SHUTDOWN(mp)); | |
44b56e0a | 2541 | xfs_perag_put(pag); |
1da177e4 LT |
2542 | return 0; |
2543 | } | |
92821e2b DC |
2544 | |
2545 | /* | |
2546 | * Read in the agi to initialise the per-ag data in the mount structure | |
2547 | */ | |
2548 | int | |
2549 | xfs_ialloc_pagi_init( | |
2550 | xfs_mount_t *mp, /* file system mount structure */ | |
2551 | xfs_trans_t *tp, /* transaction pointer */ | |
2552 | xfs_agnumber_t agno) /* allocation group number */ | |
2553 | { | |
2554 | xfs_buf_t *bp = NULL; | |
2555 | int error; | |
2556 | ||
2557 | error = xfs_ialloc_read_agi(mp, tp, agno, &bp); | |
2558 | if (error) | |
2559 | return error; | |
2560 | if (bp) | |
2561 | xfs_trans_brelse(tp, bp); | |
2562 | return 0; | |
2563 | } |