]>
Commit | Line | Data |
---|---|---|
2bd0ea18 | 1 | /* |
da23017d NS |
2 | * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
5000d01d | 4 | * |
da23017d 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 | |
2bd0ea18 | 7 | * published by the Free Software Foundation. |
5000d01d | 8 | * |
da23017d 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. | |
5000d01d | 13 | * |
da23017d 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 | |
2bd0ea18 | 17 | */ |
9c799827 | 18 | #include "libxfs_priv.h" |
b626fb59 DC |
19 | #include "xfs_fs.h" |
20 | #include "xfs_shared.h" | |
21 | #include "xfs_format.h" | |
22 | #include "xfs_log_format.h" | |
23 | #include "xfs_trans_resv.h" | |
24 | #include "xfs_bit.h" | |
25 | #include "xfs_sb.h" | |
26 | #include "xfs_mount.h" | |
f944d3d0 | 27 | #include "xfs_defer.h" |
b626fb59 DC |
28 | #include "xfs_inode.h" |
29 | #include "xfs_btree.h" | |
30 | #include "xfs_ialloc.h" | |
31 | #include "xfs_ialloc_btree.h" | |
32 | #include "xfs_alloc.h" | |
33 | #include "xfs_bmap.h" | |
34 | #include "xfs_cksum.h" | |
35 | #include "xfs_trans.h" | |
36 | #include "xfs_trace.h" | |
2bd0ea18 | 37 | |
2bd0ea18 NS |
38 | |
39 | /* | |
40 | * Allocation group level functions. | |
41 | */ | |
5e656dbb BN |
42 | static inline int |
43 | xfs_ialloc_cluster_alignment( | |
5a35bf2c | 44 | struct xfs_mount *mp) |
5e656dbb | 45 | { |
5a35bf2c DC |
46 | if (xfs_sb_version_hasalign(&mp->m_sb) && |
47 | mp->m_sb.sb_inoalignmt >= | |
48 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) | |
49 | return mp->m_sb.sb_inoalignmt; | |
5e656dbb BN |
50 | return 1; |
51 | } | |
2bd0ea18 | 52 | |
b194c7d8 | 53 | /* |
56b2de80 | 54 | * Lookup a record by ino in the btree given by cur. |
b194c7d8 BN |
55 | */ |
56 | int /* error */ | |
56b2de80 | 57 | xfs_inobt_lookup( |
b194c7d8 BN |
58 | struct xfs_btree_cur *cur, /* btree cursor */ |
59 | xfs_agino_t ino, /* starting inode of chunk */ | |
56b2de80 | 60 | xfs_lookup_t dir, /* <=, >=, == */ |
b194c7d8 BN |
61 | int *stat) /* success/failure */ |
62 | { | |
63 | cur->bc_rec.i.ir_startino = ino; | |
11640e30 BF |
64 | cur->bc_rec.i.ir_holemask = 0; |
65 | cur->bc_rec.i.ir_count = 0; | |
56b2de80 DC |
66 | cur->bc_rec.i.ir_freecount = 0; |
67 | cur->bc_rec.i.ir_free = 0; | |
68 | return xfs_btree_lookup(cur, dir, stat); | |
b194c7d8 BN |
69 | } |
70 | ||
71 | /* | |
56b2de80 | 72 | * Update the record referred to by cur to the value given. |
b194c7d8 BN |
73 | * This either works (return 0) or gets an EFSCORRUPTED error. |
74 | */ | |
75 | STATIC int /* error */ | |
76 | xfs_inobt_update( | |
77 | struct xfs_btree_cur *cur, /* btree cursor */ | |
56b2de80 | 78 | xfs_inobt_rec_incore_t *irec) /* btree record */ |
b194c7d8 BN |
79 | { |
80 | union xfs_btree_rec rec; | |
81 | ||
56b2de80 | 82 | rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino); |
11640e30 BF |
83 | if (xfs_sb_version_hassparseinodes(&cur->bc_mp->m_sb)) { |
84 | rec.inobt.ir_u.sp.ir_holemask = cpu_to_be16(irec->ir_holemask); | |
85 | rec.inobt.ir_u.sp.ir_count = irec->ir_count; | |
86 | rec.inobt.ir_u.sp.ir_freecount = irec->ir_freecount; | |
87 | } else { | |
88 | /* ir_holemask/ir_count not supported on-disk */ | |
89 | rec.inobt.ir_u.f.ir_freecount = cpu_to_be32(irec->ir_freecount); | |
90 | } | |
56b2de80 | 91 | rec.inobt.ir_free = cpu_to_be64(irec->ir_free); |
b194c7d8 BN |
92 | return xfs_btree_update(cur, &rec); |
93 | } | |
94 | ||
95 | /* | |
96 | * Get the data from the pointed-to record. | |
97 | */ | |
98 | int /* error */ | |
99 | xfs_inobt_get_rec( | |
100 | struct xfs_btree_cur *cur, /* btree cursor */ | |
56b2de80 | 101 | xfs_inobt_rec_incore_t *irec, /* btree record */ |
b194c7d8 BN |
102 | int *stat) /* output: success/failure */ |
103 | { | |
104 | union xfs_btree_rec *rec; | |
105 | int error; | |
106 | ||
107 | error = xfs_btree_get_rec(cur, &rec, stat); | |
11640e30 BF |
108 | if (error || *stat == 0) |
109 | return error; | |
110 | ||
111 | irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino); | |
112 | if (xfs_sb_version_hassparseinodes(&cur->bc_mp->m_sb)) { | |
113 | irec->ir_holemask = be16_to_cpu(rec->inobt.ir_u.sp.ir_holemask); | |
114 | irec->ir_count = rec->inobt.ir_u.sp.ir_count; | |
115 | irec->ir_freecount = rec->inobt.ir_u.sp.ir_freecount; | |
116 | } else { | |
117 | /* | |
118 | * ir_holemask/ir_count not supported on-disk. Fill in hardcoded | |
119 | * values for full inode chunks. | |
120 | */ | |
121 | irec->ir_holemask = XFS_INOBT_HOLEMASK_FULL; | |
122 | irec->ir_count = XFS_INODES_PER_CHUNK; | |
123 | irec->ir_freecount = | |
124 | be32_to_cpu(rec->inobt.ir_u.f.ir_freecount); | |
b194c7d8 | 125 | } |
11640e30 BF |
126 | irec->ir_free = be64_to_cpu(rec->inobt.ir_free); |
127 | ||
128 | return 0; | |
b194c7d8 BN |
129 | } |
130 | ||
3c699279 BF |
131 | /* |
132 | * Insert a single inobt record. Cursor must already point to desired location. | |
133 | */ | |
134 | STATIC int | |
135 | xfs_inobt_insert_rec( | |
136 | struct xfs_btree_cur *cur, | |
11640e30 BF |
137 | __uint16_t holemask, |
138 | __uint8_t count, | |
3c699279 BF |
139 | __int32_t freecount, |
140 | xfs_inofree_t free, | |
141 | int *stat) | |
142 | { | |
11640e30 BF |
143 | cur->bc_rec.i.ir_holemask = holemask; |
144 | cur->bc_rec.i.ir_count = count; | |
3c699279 BF |
145 | cur->bc_rec.i.ir_freecount = freecount; |
146 | cur->bc_rec.i.ir_free = free; | |
147 | return xfs_btree_insert(cur, stat); | |
148 | } | |
149 | ||
150 | /* | |
151 | * Insert records describing a newly allocated inode chunk into the inobt. | |
152 | */ | |
153 | STATIC int | |
154 | xfs_inobt_insert( | |
155 | struct xfs_mount *mp, | |
156 | struct xfs_trans *tp, | |
157 | struct xfs_buf *agbp, | |
158 | xfs_agino_t newino, | |
159 | xfs_agino_t newlen, | |
160 | xfs_btnum_t btnum) | |
161 | { | |
162 | struct xfs_btree_cur *cur; | |
163 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
164 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
165 | xfs_agino_t thisino; | |
166 | int i; | |
167 | int error; | |
168 | ||
169 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum); | |
170 | ||
171 | for (thisino = newino; | |
172 | thisino < newino + newlen; | |
173 | thisino += XFS_INODES_PER_CHUNK) { | |
174 | error = xfs_inobt_lookup(cur, thisino, XFS_LOOKUP_EQ, &i); | |
175 | if (error) { | |
176 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
177 | return error; | |
178 | } | |
179 | ASSERT(i == 0); | |
180 | ||
11640e30 BF |
181 | error = xfs_inobt_insert_rec(cur, XFS_INOBT_HOLEMASK_FULL, |
182 | XFS_INODES_PER_CHUNK, | |
183 | XFS_INODES_PER_CHUNK, | |
3c699279 BF |
184 | XFS_INOBT_ALL_FREE, &i); |
185 | if (error) { | |
186 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
187 | return error; | |
188 | } | |
189 | ASSERT(i == 1); | |
190 | } | |
191 | ||
192 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
193 | ||
194 | return 0; | |
195 | } | |
196 | ||
56b2de80 DC |
197 | /* |
198 | * Verify that the number of free inodes in the AGI is correct. | |
199 | */ | |
200 | #ifdef DEBUG | |
201 | STATIC int | |
202 | xfs_check_agi_freecount( | |
203 | struct xfs_btree_cur *cur, | |
204 | struct xfs_agi *agi) | |
205 | { | |
206 | if (cur->bc_nlevels == 1) { | |
207 | xfs_inobt_rec_incore_t rec; | |
208 | int freecount = 0; | |
209 | int error; | |
210 | int i; | |
211 | ||
212 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); | |
213 | if (error) | |
214 | return error; | |
215 | ||
216 | do { | |
217 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
218 | if (error) | |
219 | return error; | |
220 | ||
221 | if (i) { | |
222 | freecount += rec.ir_freecount; | |
223 | error = xfs_btree_increment(cur, 0, &i); | |
224 | if (error) | |
225 | return error; | |
226 | } | |
227 | } while (i == 1); | |
228 | ||
229 | if (!XFS_FORCED_SHUTDOWN(cur->bc_mp)) | |
230 | ASSERT(freecount == be32_to_cpu(agi->agi_freecount)); | |
231 | } | |
232 | return 0; | |
233 | } | |
234 | #else | |
235 | #define xfs_check_agi_freecount(cur, agi) 0 | |
236 | #endif | |
237 | ||
238 | /* | |
e9d35108 DC |
239 | * Initialise a new set of inodes. When called without a transaction context |
240 | * (e.g. from recovery) we initiate a delayed write of the inode buffers rather | |
241 | * than logging them (which in a transaction context puts them into the AIL | |
242 | * for writeback rather than the xfsbufd queue). | |
56b2de80 | 243 | */ |
e9d35108 | 244 | int |
56b2de80 DC |
245 | xfs_ialloc_inode_init( |
246 | struct xfs_mount *mp, | |
247 | struct xfs_trans *tp, | |
e9d35108 | 248 | struct list_head *buffer_list, |
fe8d48ac | 249 | int icount, |
56b2de80 DC |
250 | xfs_agnumber_t agno, |
251 | xfs_agblock_t agbno, | |
252 | xfs_agblock_t length, | |
253 | unsigned int gen) | |
254 | { | |
255 | struct xfs_buf *fbuf; | |
256 | struct xfs_dinode *free; | |
ff105f75 | 257 | int nbufs, blks_per_cluster, inodes_per_cluster; |
56b2de80 DC |
258 | int version; |
259 | int i, j; | |
260 | xfs_daddr_t d; | |
41ce5f36 | 261 | xfs_ino_t ino = 0; |
56b2de80 DC |
262 | |
263 | /* | |
ff105f75 DC |
264 | * Loop over the new block(s), filling in the inodes. For small block |
265 | * sizes, manipulate the inodes in buffers which are multiples of the | |
266 | * blocks size. | |
56b2de80 | 267 | */ |
ff105f75 DC |
268 | blks_per_cluster = xfs_icluster_size_fsb(mp); |
269 | inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog; | |
270 | nbufs = length / blks_per_cluster; | |
56b2de80 DC |
271 | |
272 | /* | |
e9d35108 DC |
273 | * Figure out what version number to use in the inodes we create. If |
274 | * the superblock version has caught up to the one that supports the new | |
275 | * inode format, then use the new inode version. Otherwise use the old | |
276 | * version so that old kernels will continue to be able to use the file | |
277 | * system. | |
41ce5f36 DC |
278 | * |
279 | * For v3 inodes, we also need to write the inode number into the inode, | |
280 | * so calculate the first inode number of the chunk here as | |
e9d35108 DC |
281 | * XFS_OFFBNO_TO_AGINO() only works within a filesystem block, not |
282 | * across multiple filesystem blocks (such as a cluster) and so cannot | |
283 | * be used in the cluster buffer loop below. | |
284 | * | |
285 | * Further, because we are writing the inode directly into the buffer | |
286 | * and calculating a CRC on the entire inode, we have ot log the entire | |
287 | * inode so that the entire range the CRC covers is present in the log. | |
288 | * That means for v3 inode we log the entire buffer rather than just the | |
289 | * inode cores. | |
56b2de80 | 290 | */ |
41ce5f36 DC |
291 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
292 | version = 3; | |
293 | ino = XFS_AGINO_TO_INO(mp, agno, | |
294 | XFS_OFFBNO_TO_AGINO(mp, agbno, 0)); | |
e9d35108 DC |
295 | |
296 | /* | |
297 | * log the initialisation that is about to take place as an | |
298 | * logical operation. This means the transaction does not | |
299 | * need to log the physical changes to the inode buffers as log | |
300 | * recovery will know what initialisation is actually needed. | |
301 | * Hence we only need to log the buffers as "ordered" buffers so | |
302 | * they track in the AIL as if they were physically logged. | |
303 | */ | |
304 | if (tp) | |
fe8d48ac | 305 | xfs_icreate_log(tp, agno, agbno, icount, |
e9d35108 | 306 | mp->m_sb.sb_inodesize, length, gen); |
ff105f75 | 307 | } else |
56b2de80 | 308 | version = 2; |
56b2de80 DC |
309 | |
310 | for (j = 0; j < nbufs; j++) { | |
311 | /* | |
312 | * Get the block. | |
313 | */ | |
314 | d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster)); | |
315 | fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, | |
316 | mp->m_bsize * blks_per_cluster, | |
a2ceac1f DC |
317 | XBF_UNMAPPED); |
318 | if (!fbuf) | |
12b53197 | 319 | return -ENOMEM; |
e9d35108 DC |
320 | |
321 | /* Initialize the inode buffers and log them appropriately. */ | |
a2ceac1f | 322 | fbuf->b_ops = &xfs_inode_buf_ops; |
e9d35108 | 323 | xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length)); |
ff105f75 | 324 | for (i = 0; i < inodes_per_cluster; i++) { |
56b2de80 | 325 | int ioffset = i << mp->m_sb.sb_inodelog; |
41ce5f36 | 326 | uint isize = xfs_dinode_size(version); |
56b2de80 DC |
327 | |
328 | free = xfs_make_iptr(mp, fbuf, i); | |
329 | free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC); | |
330 | free->di_version = version; | |
331 | free->di_gen = cpu_to_be32(gen); | |
332 | free->di_next_unlinked = cpu_to_be32(NULLAGINO); | |
41ce5f36 DC |
333 | |
334 | if (version == 3) { | |
335 | free->di_ino = cpu_to_be64(ino); | |
336 | ino++; | |
9c4e12fb ES |
337 | uuid_copy(&free->di_uuid, |
338 | &mp->m_sb.sb_meta_uuid); | |
41ce5f36 | 339 | xfs_dinode_calc_crc(mp, free); |
e9d35108 DC |
340 | } else if (tp) { |
341 | /* just log the inode core */ | |
342 | xfs_trans_log_buf(tp, fbuf, ioffset, | |
343 | ioffset + isize - 1); | |
41ce5f36 | 344 | } |
e9d35108 | 345 | } |
41ce5f36 | 346 | |
e9d35108 DC |
347 | if (tp) { |
348 | /* | |
349 | * Mark the buffer as an inode allocation buffer so it | |
350 | * sticks in AIL at the point of this allocation | |
351 | * transaction. This ensures the they are on disk before | |
352 | * the tail of the log can be moved past this | |
353 | * transaction (i.e. by preventing relogging from moving | |
354 | * it forward in the log). | |
355 | */ | |
356 | xfs_trans_inode_alloc_buf(tp, fbuf); | |
357 | if (version == 3) { | |
358 | /* | |
359 | * Mark the buffer as ordered so that they are | |
360 | * not physically logged in the transaction but | |
361 | * still tracked in the AIL as part of the | |
362 | * transaction and pin the log appropriately. | |
363 | */ | |
364 | xfs_trans_ordered_buf(tp, fbuf); | |
365 | xfs_trans_log_buf(tp, fbuf, 0, | |
366 | BBTOB(fbuf->b_length) - 1); | |
367 | } | |
368 | } else { | |
369 | fbuf->b_flags |= XBF_DONE; | |
370 | xfs_buf_delwri_queue(fbuf, buffer_list); | |
371 | xfs_buf_relse(fbuf); | |
56b2de80 | 372 | } |
56b2de80 | 373 | } |
a2ceac1f | 374 | return 0; |
56b2de80 DC |
375 | } |
376 | ||
6f4c54a4 BF |
377 | /* |
378 | * Align startino and allocmask for a recently allocated sparse chunk such that | |
379 | * they are fit for insertion (or merge) into the on-disk inode btrees. | |
380 | * | |
381 | * Background: | |
382 | * | |
383 | * When enabled, sparse inode support increases the inode alignment from cluster | |
384 | * size to inode chunk size. This means that the minimum range between two | |
385 | * non-adjacent inode records in the inobt is large enough for a full inode | |
386 | * record. This allows for cluster sized, cluster aligned block allocation | |
387 | * without need to worry about whether the resulting inode record overlaps with | |
388 | * another record in the tree. Without this basic rule, we would have to deal | |
389 | * with the consequences of overlap by potentially undoing recent allocations in | |
390 | * the inode allocation codepath. | |
391 | * | |
392 | * Because of this alignment rule (which is enforced on mount), there are two | |
393 | * inobt possibilities for newly allocated sparse chunks. One is that the | |
394 | * aligned inode record for the chunk covers a range of inodes not already | |
395 | * covered in the inobt (i.e., it is safe to insert a new sparse record). The | |
396 | * other is that a record already exists at the aligned startino that considers | |
397 | * the newly allocated range as sparse. In the latter case, record content is | |
398 | * merged in hope that sparse inode chunks fill to full chunks over time. | |
399 | */ | |
400 | STATIC void | |
401 | xfs_align_sparse_ino( | |
402 | struct xfs_mount *mp, | |
403 | xfs_agino_t *startino, | |
404 | uint16_t *allocmask) | |
405 | { | |
406 | xfs_agblock_t agbno; | |
407 | xfs_agblock_t mod; | |
408 | int offset; | |
409 | ||
410 | agbno = XFS_AGINO_TO_AGBNO(mp, *startino); | |
411 | mod = agbno % mp->m_sb.sb_inoalignmt; | |
412 | if (!mod) | |
413 | return; | |
414 | ||
415 | /* calculate the inode offset and align startino */ | |
416 | offset = mod << mp->m_sb.sb_inopblog; | |
417 | *startino -= offset; | |
418 | ||
419 | /* | |
420 | * Since startino has been aligned down, left shift allocmask such that | |
421 | * it continues to represent the same physical inodes relative to the | |
422 | * new startino. | |
423 | */ | |
424 | *allocmask <<= offset / XFS_INODES_PER_HOLEMASK_BIT; | |
425 | } | |
426 | ||
427 | /* | |
428 | * Determine whether the source inode record can merge into the target. Both | |
429 | * records must be sparse, the inode ranges must match and there must be no | |
430 | * allocation overlap between the records. | |
431 | */ | |
432 | STATIC bool | |
433 | __xfs_inobt_can_merge( | |
434 | struct xfs_inobt_rec_incore *trec, /* tgt record */ | |
435 | struct xfs_inobt_rec_incore *srec) /* src record */ | |
436 | { | |
437 | uint64_t talloc; | |
438 | uint64_t salloc; | |
439 | ||
440 | /* records must cover the same inode range */ | |
441 | if (trec->ir_startino != srec->ir_startino) | |
442 | return false; | |
443 | ||
444 | /* both records must be sparse */ | |
445 | if (!xfs_inobt_issparse(trec->ir_holemask) || | |
446 | !xfs_inobt_issparse(srec->ir_holemask)) | |
447 | return false; | |
448 | ||
449 | /* both records must track some inodes */ | |
450 | if (!trec->ir_count || !srec->ir_count) | |
451 | return false; | |
452 | ||
453 | /* can't exceed capacity of a full record */ | |
454 | if (trec->ir_count + srec->ir_count > XFS_INODES_PER_CHUNK) | |
455 | return false; | |
456 | ||
457 | /* verify there is no allocation overlap */ | |
458 | talloc = xfs_inobt_irec_to_allocmask(trec); | |
459 | salloc = xfs_inobt_irec_to_allocmask(srec); | |
460 | if (talloc & salloc) | |
461 | return false; | |
462 | ||
463 | return true; | |
464 | } | |
465 | ||
466 | /* | |
467 | * Merge the source inode record into the target. The caller must call | |
468 | * __xfs_inobt_can_merge() to ensure the merge is valid. | |
469 | */ | |
470 | STATIC void | |
471 | __xfs_inobt_rec_merge( | |
472 | struct xfs_inobt_rec_incore *trec, /* target */ | |
473 | struct xfs_inobt_rec_incore *srec) /* src */ | |
474 | { | |
475 | ASSERT(trec->ir_startino == srec->ir_startino); | |
476 | ||
477 | /* combine the counts */ | |
478 | trec->ir_count += srec->ir_count; | |
479 | trec->ir_freecount += srec->ir_freecount; | |
480 | ||
481 | /* | |
482 | * Merge the holemask and free mask. For both fields, 0 bits refer to | |
483 | * allocated inodes. We combine the allocated ranges with bitwise AND. | |
484 | */ | |
485 | trec->ir_holemask &= srec->ir_holemask; | |
486 | trec->ir_free &= srec->ir_free; | |
487 | } | |
488 | ||
489 | /* | |
490 | * Insert a new sparse inode chunk into the associated inode btree. The inode | |
491 | * record for the sparse chunk is pre-aligned to a startino that should match | |
492 | * any pre-existing sparse inode record in the tree. This allows sparse chunks | |
493 | * to fill over time. | |
494 | * | |
495 | * This function supports two modes of handling preexisting records depending on | |
496 | * the merge flag. If merge is true, the provided record is merged with the | |
497 | * existing record and updated in place. The merged record is returned in nrec. | |
498 | * If merge is false, an existing record is replaced with the provided record. | |
499 | * If no preexisting record exists, the provided record is always inserted. | |
500 | * | |
501 | * It is considered corruption if a merge is requested and not possible. Given | |
502 | * the sparse inode alignment constraints, this should never happen. | |
503 | */ | |
504 | STATIC int | |
505 | xfs_inobt_insert_sprec( | |
506 | struct xfs_mount *mp, | |
507 | struct xfs_trans *tp, | |
508 | struct xfs_buf *agbp, | |
509 | int btnum, | |
510 | struct xfs_inobt_rec_incore *nrec, /* in/out: new/merged rec. */ | |
511 | bool merge) /* merge or replace */ | |
512 | { | |
513 | struct xfs_btree_cur *cur; | |
514 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
515 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
516 | int error; | |
517 | int i; | |
518 | struct xfs_inobt_rec_incore rec; | |
519 | ||
520 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum); | |
521 | ||
522 | /* the new record is pre-aligned so we know where to look */ | |
523 | error = xfs_inobt_lookup(cur, nrec->ir_startino, XFS_LOOKUP_EQ, &i); | |
524 | if (error) | |
525 | goto error; | |
526 | /* if nothing there, insert a new record and return */ | |
527 | if (i == 0) { | |
528 | error = xfs_inobt_insert_rec(cur, nrec->ir_holemask, | |
529 | nrec->ir_count, nrec->ir_freecount, | |
530 | nrec->ir_free, &i); | |
531 | if (error) | |
532 | goto error; | |
533 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error); | |
534 | ||
535 | goto out; | |
536 | } | |
537 | ||
538 | /* | |
539 | * A record exists at this startino. Merge or replace the record | |
540 | * depending on what we've been asked to do. | |
541 | */ | |
542 | if (merge) { | |
543 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
544 | if (error) | |
545 | goto error; | |
546 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error); | |
547 | XFS_WANT_CORRUPTED_GOTO(mp, | |
548 | rec.ir_startino == nrec->ir_startino, | |
549 | error); | |
550 | ||
551 | /* | |
552 | * This should never fail. If we have coexisting records that | |
553 | * cannot merge, something is seriously wrong. | |
554 | */ | |
555 | XFS_WANT_CORRUPTED_GOTO(mp, __xfs_inobt_can_merge(nrec, &rec), | |
556 | error); | |
557 | ||
558 | trace_xfs_irec_merge_pre(mp, agno, rec.ir_startino, | |
559 | rec.ir_holemask, nrec->ir_startino, | |
560 | nrec->ir_holemask); | |
561 | ||
562 | /* merge to nrec to output the updated record */ | |
563 | __xfs_inobt_rec_merge(nrec, &rec); | |
564 | ||
565 | trace_xfs_irec_merge_post(mp, agno, nrec->ir_startino, | |
566 | nrec->ir_holemask); | |
567 | ||
568 | error = xfs_inobt_rec_check_count(mp, nrec); | |
569 | if (error) | |
570 | goto error; | |
571 | } | |
572 | ||
573 | error = xfs_inobt_update(cur, nrec); | |
574 | if (error) | |
575 | goto error; | |
576 | ||
577 | out: | |
578 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
579 | return 0; | |
580 | error: | |
581 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
582 | return error; | |
583 | } | |
584 | ||
2bd0ea18 NS |
585 | /* |
586 | * Allocate new inodes in the allocation group specified by agbp. | |
587 | * Return 0 for success, else error code. | |
588 | */ | |
589 | STATIC int /* error code or 0 */ | |
590 | xfs_ialloc_ag_alloc( | |
591 | xfs_trans_t *tp, /* transaction pointer */ | |
592 | xfs_buf_t *agbp, /* alloc group buffer */ | |
593 | int *alloc) | |
594 | { | |
595 | xfs_agi_t *agi; /* allocation group header */ | |
dfc130f3 | 596 | xfs_alloc_arg_t args; /* allocation argument structure */ |
5e656dbb | 597 | xfs_agnumber_t agno; |
2bd0ea18 | 598 | int error; |
2bd0ea18 NS |
599 | xfs_agino_t newino; /* new first inode's number */ |
600 | xfs_agino_t newlen; /* new number of inodes */ | |
5e656dbb | 601 | int isaligned = 0; /* inode allocation at stripe unit */ |
2bd0ea18 | 602 | /* boundary */ |
6f4c54a4 BF |
603 | uint16_t allocmask = (uint16_t) -1; /* init. to full chunk */ |
604 | struct xfs_inobt_rec_incore rec; | |
56b2de80 | 605 | struct xfs_perag *pag; |
c9005f41 BF |
606 | int do_sparse = 0; |
607 | ||
a2ceac1f | 608 | memset(&args, 0, sizeof(args)); |
2bd0ea18 NS |
609 | args.tp = tp; |
610 | args.mp = tp->t_mountp; | |
c9005f41 | 611 | args.fsbno = NULLFSBLOCK; |
2bd0ea18 | 612 | |
0d995d03 BF |
613 | #ifdef DEBUG |
614 | /* randomly do sparse inode allocations */ | |
615 | if (xfs_sb_version_hassparseinodes(&tp->t_mountp->m_sb) && | |
616 | args.mp->m_ialloc_min_blks < args.mp->m_ialloc_blks) | |
617 | do_sparse = prandom_u32() & 1; | |
618 | #endif | |
619 | ||
2bd0ea18 NS |
620 | /* |
621 | * Locking will ensure that we don't have two callers in here | |
622 | * at one time. | |
623 | */ | |
ff105f75 | 624 | newlen = args.mp->m_ialloc_inos; |
2bd0ea18 | 625 | if (args.mp->m_maxicount && |
cbf3beaa | 626 | percpu_counter_read_positive(&args.mp->m_icount) + newlen > |
19ebedcf | 627 | args.mp->m_maxicount) |
12b53197 | 628 | return -ENOSPC; |
ff105f75 | 629 | args.minlen = args.maxlen = args.mp->m_ialloc_blks; |
2bd0ea18 | 630 | /* |
5e656dbb BN |
631 | * First try to allocate inodes contiguous with the last-allocated |
632 | * chunk of inodes. If the filesystem is striped, this will fill | |
633 | * an entire stripe unit with inodes. | |
3439d03a | 634 | */ |
2bd0ea18 | 635 | agi = XFS_BUF_TO_AGI(agbp); |
5e656dbb | 636 | newino = be32_to_cpu(agi->agi_newino); |
56b2de80 | 637 | agno = be32_to_cpu(agi->agi_seqno); |
5e656dbb | 638 | args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) + |
ff105f75 | 639 | args.mp->m_ialloc_blks; |
c9005f41 BF |
640 | if (do_sparse) |
641 | goto sparse_alloc; | |
5e656dbb BN |
642 | if (likely(newino != NULLAGINO && |
643 | (args.agbno < be32_to_cpu(agi->agi_length)))) { | |
56b2de80 | 644 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
5e656dbb | 645 | args.type = XFS_ALLOCTYPE_THIS_BNO; |
5e656dbb BN |
646 | args.prod = 1; |
647 | ||
648 | /* | |
649 | * We need to take into account alignment here to ensure that | |
650 | * we don't modify the free list if we fail to have an exact | |
651 | * block. If we don't have an exact match, and every oher | |
652 | * attempt allocation attempt fails, we'll end up cancelling | |
653 | * a dirty transaction and shutting down. | |
654 | * | |
655 | * For an exact allocation, alignment must be 1, | |
656 | * however we need to take cluster alignment into account when | |
657 | * fixing up the freelist. Use the minalignslop field to | |
658 | * indicate that extra blocks might be required for alignment, | |
659 | * but not to use them in the actual exact allocation. | |
660 | */ | |
661 | args.alignment = 1; | |
5a35bf2c | 662 | args.minalignslop = xfs_ialloc_cluster_alignment(args.mp) - 1; |
5e656dbb BN |
663 | |
664 | /* Allow space for the inode btree to split. */ | |
56b2de80 | 665 | args.minleft = args.mp->m_in_maxlevels - 1; |
5e656dbb BN |
666 | if ((error = xfs_alloc_vextent(&args))) |
667 | return error; | |
ff105f75 DC |
668 | |
669 | /* | |
670 | * This request might have dirtied the transaction if the AG can | |
671 | * satisfy the request, but the exact block was not available. | |
672 | * If the allocation did fail, subsequent requests will relax | |
673 | * the exact agbno requirement and increase the alignment | |
674 | * instead. It is critical that the total size of the request | |
675 | * (len + alignment + slop) does not increase from this point | |
676 | * on, so reset minalignslop to ensure it is not included in | |
677 | * subsequent requests. | |
678 | */ | |
679 | args.minalignslop = 0; | |
c9005f41 | 680 | } |
5e656dbb BN |
681 | |
682 | if (unlikely(args.fsbno == NULLFSBLOCK)) { | |
683 | /* | |
684 | * Set the alignment for the allocation. | |
685 | * If stripe alignment is turned on then align at stripe unit | |
686 | * boundary. | |
687 | * If the cluster size is smaller than a filesystem block | |
688 | * then we're doing I/O for inodes in filesystem block size | |
689 | * pieces, so don't need alignment anyway. | |
690 | */ | |
691 | isaligned = 0; | |
692 | if (args.mp->m_sinoalign) { | |
693 | ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN)); | |
694 | args.alignment = args.mp->m_dalign; | |
695 | isaligned = 1; | |
696 | } else | |
5a35bf2c | 697 | args.alignment = xfs_ialloc_cluster_alignment(args.mp); |
5e656dbb BN |
698 | /* |
699 | * Need to figure out where to allocate the inode blocks. | |
700 | * Ideally they should be spaced out through the a.g. | |
701 | * For now, just allocate blocks up front. | |
702 | */ | |
703 | args.agbno = be32_to_cpu(agi->agi_root); | |
56b2de80 | 704 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
5e656dbb BN |
705 | /* |
706 | * Allocate a fixed-size extent of inodes. | |
707 | */ | |
708 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
5e656dbb BN |
709 | args.prod = 1; |
710 | /* | |
711 | * Allow space for the inode btree to split. | |
712 | */ | |
56b2de80 | 713 | args.minleft = args.mp->m_in_maxlevels - 1; |
5e656dbb BN |
714 | if ((error = xfs_alloc_vextent(&args))) |
715 | return error; | |
716 | } | |
2bd0ea18 NS |
717 | |
718 | /* | |
719 | * If stripe alignment is turned on, then try again with cluster | |
720 | * alignment. | |
721 | */ | |
722 | if (isaligned && args.fsbno == NULLFSBLOCK) { | |
723 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
6e3140c7 | 724 | args.agbno = be32_to_cpu(agi->agi_root); |
56b2de80 | 725 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
5a35bf2c | 726 | args.alignment = xfs_ialloc_cluster_alignment(args.mp); |
0e266570 | 727 | if ((error = xfs_alloc_vextent(&args))) |
dfc130f3 | 728 | return error; |
2bd0ea18 | 729 | } |
5000d01d | 730 | |
6f4c54a4 BF |
731 | /* |
732 | * Finally, try a sparse allocation if the filesystem supports it and | |
733 | * the sparse allocation length is smaller than a full chunk. | |
734 | */ | |
735 | if (xfs_sb_version_hassparseinodes(&args.mp->m_sb) && | |
736 | args.mp->m_ialloc_min_blks < args.mp->m_ialloc_blks && | |
737 | args.fsbno == NULLFSBLOCK) { | |
c9005f41 | 738 | sparse_alloc: |
6f4c54a4 BF |
739 | args.type = XFS_ALLOCTYPE_NEAR_BNO; |
740 | args.agbno = be32_to_cpu(agi->agi_root); | |
741 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); | |
742 | args.alignment = args.mp->m_sb.sb_spino_align; | |
743 | args.prod = 1; | |
744 | ||
745 | args.minlen = args.mp->m_ialloc_min_blks; | |
746 | args.maxlen = args.minlen; | |
747 | ||
748 | /* | |
749 | * The inode record will be aligned to full chunk size. We must | |
750 | * prevent sparse allocation from AG boundaries that result in | |
751 | * invalid inode records, such as records that start at agbno 0 | |
752 | * or extend beyond the AG. | |
753 | * | |
754 | * Set min agbno to the first aligned, non-zero agbno and max to | |
755 | * the last aligned agbno that is at least one full chunk from | |
756 | * the end of the AG. | |
757 | */ | |
758 | args.min_agbno = args.mp->m_sb.sb_inoalignmt; | |
759 | args.max_agbno = round_down(args.mp->m_sb.sb_agblocks, | |
760 | args.mp->m_sb.sb_inoalignmt) - | |
761 | args.mp->m_ialloc_blks; | |
762 | ||
763 | error = xfs_alloc_vextent(&args); | |
764 | if (error) | |
765 | return error; | |
766 | ||
767 | newlen = args.len << args.mp->m_sb.sb_inopblog; | |
0d995d03 | 768 | ASSERT(newlen <= XFS_INODES_PER_CHUNK); |
6f4c54a4 BF |
769 | allocmask = (1 << (newlen / XFS_INODES_PER_HOLEMASK_BIT)) - 1; |
770 | } | |
771 | ||
2bd0ea18 NS |
772 | if (args.fsbno == NULLFSBLOCK) { |
773 | *alloc = 0; | |
774 | return 0; | |
775 | } | |
776 | ASSERT(args.len == args.minlen); | |
a562a63b | 777 | |
5e656dbb | 778 | /* |
56b2de80 DC |
779 | * Stamp and write the inode buffers. |
780 | * | |
5e656dbb BN |
781 | * Seed the new inode cluster with a random generation number. This |
782 | * prevents short-term reuse of generation numbers if a chunk is | |
783 | * freed and then immediately reallocated. We use random numbers | |
784 | * rather than a linear progression to prevent the next generation | |
785 | * number from being easily guessable. | |
786 | */ | |
fe8d48ac BF |
787 | error = xfs_ialloc_inode_init(args.mp, tp, NULL, newlen, agno, |
788 | args.agbno, args.len, prandom_u32()); | |
56b2de80 | 789 | |
a2ceac1f DC |
790 | if (error) |
791 | return error; | |
56b2de80 DC |
792 | /* |
793 | * Convert the results. | |
794 | */ | |
795 | newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0); | |
6f4c54a4 BF |
796 | |
797 | if (xfs_inobt_issparse(~allocmask)) { | |
798 | /* | |
799 | * We've allocated a sparse chunk. Align the startino and mask. | |
800 | */ | |
801 | xfs_align_sparse_ino(args.mp, &newino, &allocmask); | |
802 | ||
803 | rec.ir_startino = newino; | |
804 | rec.ir_holemask = ~allocmask; | |
805 | rec.ir_count = newlen; | |
806 | rec.ir_freecount = newlen; | |
807 | rec.ir_free = XFS_INOBT_ALL_FREE; | |
808 | ||
809 | /* | |
810 | * Insert the sparse record into the inobt and allow for a merge | |
811 | * if necessary. If a merge does occur, rec is updated to the | |
812 | * merged record. | |
813 | */ | |
814 | error = xfs_inobt_insert_sprec(args.mp, tp, agbp, XFS_BTNUM_INO, | |
815 | &rec, true); | |
816 | if (error == -EFSCORRUPTED) { | |
817 | xfs_alert(args.mp, | |
818 | "invalid sparse inode record: ino 0x%llx holemask 0x%x count %u", | |
819 | XFS_AGINO_TO_INO(args.mp, agno, | |
820 | rec.ir_startino), | |
821 | rec.ir_holemask, rec.ir_count); | |
822 | xfs_force_shutdown(args.mp, SHUTDOWN_CORRUPT_INCORE); | |
823 | } | |
824 | if (error) | |
825 | return error; | |
826 | ||
827 | /* | |
828 | * We can't merge the part we've just allocated as for the inobt | |
829 | * due to finobt semantics. The original record may or may not | |
830 | * exist independent of whether physical inodes exist in this | |
831 | * sparse chunk. | |
832 | * | |
833 | * We must update the finobt record based on the inobt record. | |
834 | * rec contains the fully merged and up to date inobt record | |
835 | * from the previous call. Set merge false to replace any | |
836 | * existing record with this one. | |
837 | */ | |
838 | if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) { | |
839 | error = xfs_inobt_insert_sprec(args.mp, tp, agbp, | |
840 | XFS_BTNUM_FINO, &rec, | |
841 | false); | |
842 | if (error) | |
843 | return error; | |
844 | } | |
845 | } else { | |
846 | /* full chunk - insert new records to both btrees */ | |
847 | error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen, | |
848 | XFS_BTNUM_INO); | |
849 | if (error) | |
850 | return error; | |
851 | ||
852 | if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) { | |
853 | error = xfs_inobt_insert(args.mp, tp, agbp, newino, | |
854 | newlen, XFS_BTNUM_FINO); | |
855 | if (error) | |
856 | return error; | |
857 | } | |
858 | } | |
859 | ||
860 | /* | |
861 | * Update AGI counts and newino. | |
862 | */ | |
5e656dbb BN |
863 | be32_add_cpu(&agi->agi_count, newlen); |
864 | be32_add_cpu(&agi->agi_freecount, newlen); | |
56b2de80 DC |
865 | pag = xfs_perag_get(args.mp, agno); |
866 | pag->pagi_freecount += newlen; | |
867 | xfs_perag_put(pag); | |
6e3140c7 | 868 | agi->agi_newino = cpu_to_be32(newino); |
56b2de80 | 869 | |
2bd0ea18 NS |
870 | /* |
871 | * Log allocation group header fields | |
872 | */ | |
873 | xfs_ialloc_log_agi(tp, agbp, | |
874 | XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO); | |
875 | /* | |
876 | * Modify/log superblock values for inode count and inode free count. | |
877 | */ | |
878 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen); | |
879 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen); | |
880 | *alloc = 1; | |
881 | return 0; | |
882 | } | |
883 | ||
56b2de80 | 884 | STATIC xfs_agnumber_t |
321717ae NS |
885 | xfs_ialloc_next_ag( |
886 | xfs_mount_t *mp) | |
887 | { | |
888 | xfs_agnumber_t agno; | |
889 | ||
890 | spin_lock(&mp->m_agirotor_lock); | |
891 | agno = mp->m_agirotor; | |
a2ceac1f | 892 | if (++mp->m_agirotor >= mp->m_maxagi) |
321717ae NS |
893 | mp->m_agirotor = 0; |
894 | spin_unlock(&mp->m_agirotor_lock); | |
895 | ||
896 | return agno; | |
897 | } | |
898 | ||
2bd0ea18 NS |
899 | /* |
900 | * Select an allocation group to look for a free inode in, based on the parent | |
e6d77a21 | 901 | * inode and the mode. Return the allocation group buffer. |
2bd0ea18 | 902 | */ |
a2ceac1f | 903 | STATIC xfs_agnumber_t |
2bd0ea18 NS |
904 | xfs_ialloc_ag_select( |
905 | xfs_trans_t *tp, /* transaction pointer */ | |
906 | xfs_ino_t parent, /* parent directory inode number */ | |
a2ceac1f | 907 | umode_t mode, /* bits set to indicate file type */ |
2bd0ea18 NS |
908 | int okalloc) /* ok to allocate more space */ |
909 | { | |
2bd0ea18 NS |
910 | xfs_agnumber_t agcount; /* number of ag's in the filesystem */ |
911 | xfs_agnumber_t agno; /* current ag number */ | |
912 | int flags; /* alloc buffer locking flags */ | |
913 | xfs_extlen_t ineed; /* blocks needed for inode allocation */ | |
275ae71f | 914 | xfs_extlen_t longest = 0; /* longest extent available */ |
2bd0ea18 NS |
915 | xfs_mount_t *mp; /* mount point structure */ |
916 | int needspace; /* file mode implies space allocated */ | |
917 | xfs_perag_t *pag; /* per allocation group data */ | |
918 | xfs_agnumber_t pagno; /* parent (starting) ag number */ | |
a2ceac1f | 919 | int error; |
2bd0ea18 NS |
920 | |
921 | /* | |
922 | * Files of these types need at least one block if length > 0 | |
923 | * (and they won't fit in the inode, but that's hard to figure out). | |
924 | */ | |
925 | needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode); | |
926 | mp = tp->t_mountp; | |
34317449 | 927 | agcount = mp->m_maxagi; |
2bd0ea18 | 928 | if (S_ISDIR(mode)) |
321717ae | 929 | pagno = xfs_ialloc_next_ag(mp); |
5ce1d1f7 | 930 | else { |
2bd0ea18 | 931 | pagno = XFS_INO_TO_AGNO(mp, parent); |
5ce1d1f7 NS |
932 | if (pagno >= agcount) |
933 | pagno = 0; | |
934 | } | |
a2ceac1f | 935 | |
2bd0ea18 | 936 | ASSERT(pagno < agcount); |
a2ceac1f | 937 | |
2bd0ea18 NS |
938 | /* |
939 | * Loop through allocation groups, looking for one with a little | |
940 | * free space in it. Note we don't look for free inodes, exactly. | |
941 | * Instead, we include whether there is a need to allocate inodes | |
5000d01d | 942 | * to mean that blocks must be allocated for them, |
2bd0ea18 NS |
943 | * if none are currently free. |
944 | */ | |
945 | agno = pagno; | |
946 | flags = XFS_ALLOC_FLAG_TRYLOCK; | |
947 | for (;;) { | |
56b2de80 | 948 | pag = xfs_perag_get(mp, agno); |
a2ceac1f DC |
949 | if (!pag->pagi_inodeok) { |
950 | xfs_ialloc_next_ag(mp); | |
951 | goto nextag; | |
952 | } | |
953 | ||
2bd0ea18 | 954 | if (!pag->pagi_init) { |
a2ceac1f DC |
955 | error = xfs_ialloc_pagi_init(mp, tp, agno); |
956 | if (error) | |
2bd0ea18 | 957 | goto nextag; |
a2ceac1f | 958 | } |
34317449 | 959 | |
a2ceac1f DC |
960 | if (pag->pagi_freecount) { |
961 | xfs_perag_put(pag); | |
962 | return agno; | |
34317449 NS |
963 | } |
964 | ||
a2ceac1f DC |
965 | if (!okalloc) |
966 | goto nextag; | |
967 | ||
968 | if (!pag->pagf_init) { | |
969 | error = xfs_alloc_pagf_init(mp, tp, agno, flags); | |
970 | if (error) | |
2bd0ea18 | 971 | goto nextag; |
2bd0ea18 | 972 | } |
a2ceac1f DC |
973 | |
974 | /* | |
5a35bf2c DC |
975 | * Check that there is enough free space for the file plus a |
976 | * chunk of inodes if we need to allocate some. If this is the | |
977 | * first pass across the AGs, take into account the potential | |
978 | * space needed for alignment of inode chunks when checking the | |
979 | * longest contiguous free space in the AG - this prevents us | |
980 | * from getting ENOSPC because we have free space larger than | |
981 | * m_ialloc_blks but alignment constraints prevent us from using | |
982 | * it. | |
983 | * | |
984 | * If we can't find an AG with space for full alignment slack to | |
985 | * be taken into account, we must be near ENOSPC in all AGs. | |
986 | * Hence we don't include alignment for the second pass and so | |
987 | * if we fail allocation due to alignment issues then it is most | |
988 | * likely a real ENOSPC condition. | |
a2ceac1f | 989 | */ |
62dc6cdb | 990 | ineed = mp->m_ialloc_min_blks; |
5a35bf2c DC |
991 | if (flags && ineed > 1) |
992 | ineed += xfs_ialloc_cluster_alignment(mp); | |
a2ceac1f DC |
993 | longest = pag->pagf_longest; |
994 | if (!longest) | |
995 | longest = pag->pagf_flcount > 0; | |
996 | ||
997 | if (pag->pagf_freeblks >= needspace + ineed && | |
998 | longest >= ineed) { | |
999 | xfs_perag_put(pag); | |
1000 | return agno; | |
2bd0ea18 | 1001 | } |
5000d01d | 1002 | nextag: |
56b2de80 | 1003 | xfs_perag_put(pag); |
5000d01d | 1004 | /* |
2bd0ea18 NS |
1005 | * No point in iterating over the rest, if we're shutting |
1006 | * down. | |
1007 | */ | |
56b2de80 | 1008 | if (XFS_FORCED_SHUTDOWN(mp)) |
a2ceac1f | 1009 | return NULLAGNUMBER; |
2bd0ea18 | 1010 | agno++; |
5ce1d1f7 | 1011 | if (agno >= agcount) |
2bd0ea18 NS |
1012 | agno = 0; |
1013 | if (agno == pagno) { | |
56b2de80 | 1014 | if (flags == 0) |
a2ceac1f | 1015 | return NULLAGNUMBER; |
2bd0ea18 NS |
1016 | flags = 0; |
1017 | } | |
1018 | } | |
1019 | } | |
1020 | ||
56b2de80 DC |
1021 | /* |
1022 | * Try to retrieve the next record to the left/right from the current one. | |
1023 | */ | |
1024 | STATIC int | |
1025 | xfs_ialloc_next_rec( | |
1026 | struct xfs_btree_cur *cur, | |
1027 | xfs_inobt_rec_incore_t *rec, | |
1028 | int *done, | |
1029 | int left) | |
1030 | { | |
1031 | int error; | |
1032 | int i; | |
1033 | ||
1034 | if (left) | |
1035 | error = xfs_btree_decrement(cur, 0, &i); | |
1036 | else | |
1037 | error = xfs_btree_increment(cur, 0, &i); | |
1038 | ||
1039 | if (error) | |
1040 | return error; | |
1041 | *done = !i; | |
1042 | if (i) { | |
1043 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1044 | if (error) | |
1045 | return error; | |
19ebedcf | 1046 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
56b2de80 DC |
1047 | } |
1048 | ||
1049 | return 0; | |
1050 | } | |
1051 | ||
1052 | STATIC int | |
1053 | xfs_ialloc_get_rec( | |
1054 | struct xfs_btree_cur *cur, | |
1055 | xfs_agino_t agino, | |
1056 | xfs_inobt_rec_incore_t *rec, | |
3439d03a | 1057 | int *done) |
56b2de80 DC |
1058 | { |
1059 | int error; | |
1060 | int i; | |
1061 | ||
1062 | error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i); | |
1063 | if (error) | |
1064 | return error; | |
1065 | *done = !i; | |
1066 | if (i) { | |
1067 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1068 | if (error) | |
1069 | return error; | |
19ebedcf | 1070 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
56b2de80 DC |
1071 | } |
1072 | ||
1073 | return 0; | |
1074 | } | |
1075 | ||
01792d3b | 1076 | /* |
fa58183c BF |
1077 | * Return the offset of the first free inode in the record. If the inode chunk |
1078 | * is sparsely allocated, we convert the record holemask to inode granularity | |
1079 | * and mask off the unallocated regions from the inode free mask. | |
01792d3b BF |
1080 | */ |
1081 | STATIC int | |
1082 | xfs_inobt_first_free_inode( | |
1083 | struct xfs_inobt_rec_incore *rec) | |
1084 | { | |
fa58183c BF |
1085 | xfs_inofree_t realfree; |
1086 | ||
1087 | /* if there are no holes, return the first available offset */ | |
1088 | if (!xfs_inobt_issparse(rec->ir_holemask)) | |
1089 | return xfs_lowbit64(rec->ir_free); | |
1090 | ||
1091 | realfree = xfs_inobt_irec_to_allocmask(rec); | |
1092 | realfree &= rec->ir_free; | |
1093 | ||
1094 | return xfs_lowbit64(realfree); | |
01792d3b BF |
1095 | } |
1096 | ||
5000d01d | 1097 | /* |
ff105f75 | 1098 | * Allocate an inode using the inobt-only algorithm. |
2bd0ea18 | 1099 | */ |
a2ceac1f | 1100 | STATIC int |
ff105f75 | 1101 | xfs_dialloc_ag_inobt( |
a2ceac1f DC |
1102 | struct xfs_trans *tp, |
1103 | struct xfs_buf *agbp, | |
1104 | xfs_ino_t parent, | |
1105 | xfs_ino_t *inop) | |
2bd0ea18 | 1106 | { |
a2ceac1f DC |
1107 | struct xfs_mount *mp = tp->t_mountp; |
1108 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
1109 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
1110 | xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent); | |
1111 | xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent); | |
1112 | struct xfs_perag *pag; | |
1113 | struct xfs_btree_cur *cur, *tcur; | |
1114 | struct xfs_inobt_rec_incore rec, trec; | |
1115 | xfs_ino_t ino; | |
1116 | int error; | |
1117 | int offset; | |
1118 | int i, j; | |
2bd0ea18 | 1119 | |
56b2de80 DC |
1120 | pag = xfs_perag_get(mp, agno); |
1121 | ||
a2ceac1f DC |
1122 | ASSERT(pag->pagi_init); |
1123 | ASSERT(pag->pagi_inodeok); | |
1124 | ASSERT(pag->pagi_freecount > 0); | |
1125 | ||
56b2de80 | 1126 | restart_pagno: |
70eb7337 | 1127 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
2bd0ea18 NS |
1128 | /* |
1129 | * If pagino is 0 (this is the root inode allocation) use newino. | |
1130 | * This must work because we've just allocated some. | |
1131 | */ | |
1132 | if (!pagino) | |
6e3140c7 | 1133 | pagino = be32_to_cpu(agi->agi_newino); |
2bd0ea18 | 1134 | |
56b2de80 DC |
1135 | error = xfs_check_agi_freecount(cur, agi); |
1136 | if (error) | |
1137 | goto error0; | |
2bd0ea18 | 1138 | |
2bd0ea18 | 1139 | /* |
56b2de80 | 1140 | * If in the same AG as the parent, try to get near the parent. |
2bd0ea18 NS |
1141 | */ |
1142 | if (pagno == agno) { | |
56b2de80 DC |
1143 | int doneleft; /* done, to the left */ |
1144 | int doneright; /* done, to the right */ | |
1145 | int searchdistance = 10; | |
1146 | ||
1147 | error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i); | |
1148 | if (error) | |
2bd0ea18 | 1149 | goto error0; |
19ebedcf | 1150 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
56b2de80 DC |
1151 | |
1152 | error = xfs_inobt_get_rec(cur, &rec, &j); | |
1153 | if (error) | |
1154 | goto error0; | |
19ebedcf | 1155 | XFS_WANT_CORRUPTED_GOTO(mp, j == 1, error0); |
56b2de80 DC |
1156 | |
1157 | if (rec.ir_freecount > 0) { | |
2bd0ea18 NS |
1158 | /* |
1159 | * Found a free inode in the same chunk | |
56b2de80 | 1160 | * as the parent, done. |
2bd0ea18 | 1161 | */ |
56b2de80 | 1162 | goto alloc_inode; |
2bd0ea18 | 1163 | } |
56b2de80 DC |
1164 | |
1165 | ||
1166 | /* | |
1167 | * In the same AG as parent, but parent's chunk is full. | |
1168 | */ | |
1169 | ||
1170 | /* duplicate the cursor, search left & right simultaneously */ | |
1171 | error = xfs_btree_dup_cursor(cur, &tcur); | |
1172 | if (error) | |
1173 | goto error0; | |
1174 | ||
2bd0ea18 | 1175 | /* |
56b2de80 | 1176 | * Skip to last blocks looked up if same parent inode. |
2bd0ea18 | 1177 | */ |
56b2de80 DC |
1178 | if (pagino != NULLAGINO && |
1179 | pag->pagl_pagino == pagino && | |
1180 | pag->pagl_leftrec != NULLAGINO && | |
1181 | pag->pagl_rightrec != NULLAGINO) { | |
1182 | error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec, | |
3439d03a | 1183 | &trec, &doneleft); |
56b2de80 DC |
1184 | if (error) |
1185 | goto error1; | |
2bd0ea18 | 1186 | |
56b2de80 | 1187 | error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec, |
3439d03a | 1188 | &rec, &doneright); |
2bd0ea18 | 1189 | if (error) |
2bd0ea18 | 1190 | goto error1; |
56b2de80 DC |
1191 | } else { |
1192 | /* search left with tcur, back up 1 record */ | |
1193 | error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1); | |
1194 | if (error) | |
2bd0ea18 | 1195 | goto error1; |
2bd0ea18 | 1196 | |
56b2de80 DC |
1197 | /* search right with cur, go forward 1 record. */ |
1198 | error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0); | |
1199 | if (error) | |
1200 | goto error1; | |
1201 | } | |
1202 | ||
1203 | /* | |
1204 | * Loop until we find an inode chunk with a free inode. | |
1205 | */ | |
1206 | while (!doneleft || !doneright) { | |
1207 | int useleft; /* using left inode chunk this time */ | |
1208 | ||
1209 | if (!--searchdistance) { | |
2bd0ea18 | 1210 | /* |
56b2de80 DC |
1211 | * Not in range - save last search |
1212 | * location and allocate a new inode | |
2bd0ea18 | 1213 | */ |
56b2de80 DC |
1214 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); |
1215 | pag->pagl_leftrec = trec.ir_startino; | |
1216 | pag->pagl_rightrec = rec.ir_startino; | |
1217 | pag->pagl_pagino = pagino; | |
1218 | goto newino; | |
2bd0ea18 | 1219 | } |
56b2de80 DC |
1220 | |
1221 | /* figure out the closer block if both are valid. */ | |
1222 | if (!doneleft && !doneright) { | |
1223 | useleft = pagino - | |
1224 | (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) < | |
1225 | rec.ir_startino - pagino; | |
1226 | } else { | |
1227 | useleft = !doneleft; | |
1228 | } | |
1229 | ||
1230 | /* free inodes to the left? */ | |
1231 | if (useleft && trec.ir_freecount) { | |
1232 | rec = trec; | |
1233 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
1234 | cur = tcur; | |
1235 | ||
1236 | pag->pagl_leftrec = trec.ir_startino; | |
1237 | pag->pagl_rightrec = rec.ir_startino; | |
1238 | pag->pagl_pagino = pagino; | |
1239 | goto alloc_inode; | |
1240 | } | |
1241 | ||
1242 | /* free inodes to the right? */ | |
1243 | if (!useleft && rec.ir_freecount) { | |
1244 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
1245 | ||
1246 | pag->pagl_leftrec = trec.ir_startino; | |
1247 | pag->pagl_rightrec = rec.ir_startino; | |
1248 | pag->pagl_pagino = pagino; | |
1249 | goto alloc_inode; | |
1250 | } | |
1251 | ||
1252 | /* get next record to check */ | |
1253 | if (useleft) { | |
1254 | error = xfs_ialloc_next_rec(tcur, &trec, | |
1255 | &doneleft, 1); | |
1256 | } else { | |
1257 | error = xfs_ialloc_next_rec(cur, &rec, | |
1258 | &doneright, 0); | |
1259 | } | |
1260 | if (error) | |
1261 | goto error1; | |
2bd0ea18 | 1262 | } |
56b2de80 DC |
1263 | |
1264 | /* | |
1265 | * We've reached the end of the btree. because | |
1266 | * we are only searching a small chunk of the | |
1267 | * btree each search, there is obviously free | |
1268 | * inodes closer to the parent inode than we | |
1269 | * are now. restart the search again. | |
1270 | */ | |
1271 | pag->pagl_pagino = NULLAGINO; | |
1272 | pag->pagl_leftrec = NULLAGINO; | |
1273 | pag->pagl_rightrec = NULLAGINO; | |
1274 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
1275 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
1276 | goto restart_pagno; | |
2bd0ea18 | 1277 | } |
56b2de80 | 1278 | |
2bd0ea18 | 1279 | /* |
56b2de80 | 1280 | * In a different AG from the parent. |
2bd0ea18 NS |
1281 | * See if the most recently allocated block has any free. |
1282 | */ | |
56b2de80 | 1283 | newino: |
a2ceac1f | 1284 | if (agi->agi_newino != cpu_to_be32(NULLAGINO)) { |
56b2de80 DC |
1285 | error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino), |
1286 | XFS_LOOKUP_EQ, &i); | |
1287 | if (error) | |
2bd0ea18 | 1288 | goto error0; |
56b2de80 DC |
1289 | |
1290 | if (i == 1) { | |
1291 | error = xfs_inobt_get_rec(cur, &rec, &j); | |
2bd0ea18 NS |
1292 | if (error) |
1293 | goto error0; | |
56b2de80 DC |
1294 | |
1295 | if (j == 1 && rec.ir_freecount > 0) { | |
1296 | /* | |
1297 | * The last chunk allocated in the group | |
1298 | * still has a free inode. | |
1299 | */ | |
1300 | goto alloc_inode; | |
2bd0ea18 NS |
1301 | } |
1302 | } | |
1303 | } | |
56b2de80 DC |
1304 | |
1305 | /* | |
1306 | * None left in the last group, search the whole AG | |
1307 | */ | |
1308 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); | |
1309 | if (error) | |
1310 | goto error0; | |
19ebedcf | 1311 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
56b2de80 DC |
1312 | |
1313 | for (;;) { | |
1314 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
1315 | if (error) | |
1316 | goto error0; | |
19ebedcf | 1317 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
56b2de80 DC |
1318 | if (rec.ir_freecount > 0) |
1319 | break; | |
1320 | error = xfs_btree_increment(cur, 0, &i); | |
1321 | if (error) | |
1322 | goto error0; | |
19ebedcf | 1323 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
56b2de80 DC |
1324 | } |
1325 | ||
1326 | alloc_inode: | |
01792d3b | 1327 | offset = xfs_inobt_first_free_inode(&rec); |
2bd0ea18 NS |
1328 | ASSERT(offset >= 0); |
1329 | ASSERT(offset < XFS_INODES_PER_CHUNK); | |
1330 | ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) % | |
1331 | XFS_INODES_PER_CHUNK) == 0); | |
1332 | ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset); | |
56b2de80 | 1333 | rec.ir_free &= ~XFS_INOBT_MASK(offset); |
2bd0ea18 | 1334 | rec.ir_freecount--; |
56b2de80 DC |
1335 | error = xfs_inobt_update(cur, &rec); |
1336 | if (error) | |
2bd0ea18 | 1337 | goto error0; |
5e656dbb | 1338 | be32_add_cpu(&agi->agi_freecount, -1); |
2bd0ea18 | 1339 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); |
56b2de80 DC |
1340 | pag->pagi_freecount--; |
1341 | ||
1342 | error = xfs_check_agi_freecount(cur, agi); | |
1343 | if (error) | |
1344 | goto error0; | |
2bd0ea18 | 1345 | |
2bd0ea18 NS |
1346 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
1347 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1); | |
56b2de80 | 1348 | xfs_perag_put(pag); |
2bd0ea18 NS |
1349 | *inop = ino; |
1350 | return 0; | |
1351 | error1: | |
1352 | xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR); | |
1353 | error0: | |
1354 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
56b2de80 | 1355 | xfs_perag_put(pag); |
2bd0ea18 NS |
1356 | return error; |
1357 | } | |
1358 | ||
ff105f75 DC |
1359 | /* |
1360 | * Use the free inode btree to allocate an inode based on distance from the | |
1361 | * parent. Note that the provided cursor may be deleted and replaced. | |
1362 | */ | |
1363 | STATIC int | |
1364 | xfs_dialloc_ag_finobt_near( | |
1365 | xfs_agino_t pagino, | |
1366 | struct xfs_btree_cur **ocur, | |
1367 | struct xfs_inobt_rec_incore *rec) | |
1368 | { | |
1369 | struct xfs_btree_cur *lcur = *ocur; /* left search cursor */ | |
1370 | struct xfs_btree_cur *rcur; /* right search cursor */ | |
1371 | struct xfs_inobt_rec_incore rrec; | |
1372 | int error; | |
1373 | int i, j; | |
1374 | ||
1375 | error = xfs_inobt_lookup(lcur, pagino, XFS_LOOKUP_LE, &i); | |
1376 | if (error) | |
1377 | return error; | |
1378 | ||
1379 | if (i == 1) { | |
1380 | error = xfs_inobt_get_rec(lcur, rec, &i); | |
1381 | if (error) | |
1382 | return error; | |
19ebedcf | 1383 | XFS_WANT_CORRUPTED_RETURN(lcur->bc_mp, i == 1); |
ff105f75 DC |
1384 | |
1385 | /* | |
1386 | * See if we've landed in the parent inode record. The finobt | |
1387 | * only tracks chunks with at least one free inode, so record | |
1388 | * existence is enough. | |
1389 | */ | |
1390 | if (pagino >= rec->ir_startino && | |
1391 | pagino < (rec->ir_startino + XFS_INODES_PER_CHUNK)) | |
1392 | return 0; | |
1393 | } | |
1394 | ||
1395 | error = xfs_btree_dup_cursor(lcur, &rcur); | |
1396 | if (error) | |
1397 | return error; | |
1398 | ||
1399 | error = xfs_inobt_lookup(rcur, pagino, XFS_LOOKUP_GE, &j); | |
1400 | if (error) | |
1401 | goto error_rcur; | |
1402 | if (j == 1) { | |
1403 | error = xfs_inobt_get_rec(rcur, &rrec, &j); | |
1404 | if (error) | |
1405 | goto error_rcur; | |
19ebedcf | 1406 | XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, j == 1, error_rcur); |
ff105f75 DC |
1407 | } |
1408 | ||
19ebedcf | 1409 | XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, i == 1 || j == 1, error_rcur); |
ff105f75 DC |
1410 | if (i == 1 && j == 1) { |
1411 | /* | |
1412 | * Both the left and right records are valid. Choose the closer | |
1413 | * inode chunk to the target. | |
1414 | */ | |
1415 | if ((pagino - rec->ir_startino + XFS_INODES_PER_CHUNK - 1) > | |
1416 | (rrec.ir_startino - pagino)) { | |
1417 | *rec = rrec; | |
1418 | xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR); | |
1419 | *ocur = rcur; | |
1420 | } else { | |
1421 | xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR); | |
1422 | } | |
1423 | } else if (j == 1) { | |
1424 | /* only the right record is valid */ | |
1425 | *rec = rrec; | |
1426 | xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR); | |
1427 | *ocur = rcur; | |
1428 | } else if (i == 1) { | |
1429 | /* only the left record is valid */ | |
1430 | xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR); | |
1431 | } | |
1432 | ||
1433 | return 0; | |
1434 | ||
1435 | error_rcur: | |
1436 | xfs_btree_del_cursor(rcur, XFS_BTREE_ERROR); | |
1437 | return error; | |
1438 | } | |
1439 | ||
1440 | /* | |
1441 | * Use the free inode btree to find a free inode based on a newino hint. If | |
1442 | * the hint is NULL, find the first free inode in the AG. | |
1443 | */ | |
1444 | STATIC int | |
1445 | xfs_dialloc_ag_finobt_newino( | |
1446 | struct xfs_agi *agi, | |
1447 | struct xfs_btree_cur *cur, | |
1448 | struct xfs_inobt_rec_incore *rec) | |
1449 | { | |
1450 | int error; | |
1451 | int i; | |
1452 | ||
1453 | if (agi->agi_newino != cpu_to_be32(NULLAGINO)) { | |
1454 | error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino), | |
1455 | XFS_LOOKUP_EQ, &i); | |
1456 | if (error) | |
1457 | return error; | |
1458 | if (i == 1) { | |
1459 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1460 | if (error) | |
1461 | return error; | |
19ebedcf | 1462 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
ff105f75 DC |
1463 | return 0; |
1464 | } | |
1465 | } | |
1466 | ||
1467 | /* | |
1468 | * Find the first inode available in the AG. | |
1469 | */ | |
1470 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); | |
1471 | if (error) | |
1472 | return error; | |
19ebedcf | 1473 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
ff105f75 DC |
1474 | |
1475 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1476 | if (error) | |
1477 | return error; | |
19ebedcf | 1478 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
ff105f75 DC |
1479 | |
1480 | return 0; | |
1481 | } | |
1482 | ||
1483 | /* | |
1484 | * Update the inobt based on a modification made to the finobt. Also ensure that | |
1485 | * the records from both trees are equivalent post-modification. | |
1486 | */ | |
1487 | STATIC int | |
1488 | xfs_dialloc_ag_update_inobt( | |
1489 | struct xfs_btree_cur *cur, /* inobt cursor */ | |
1490 | struct xfs_inobt_rec_incore *frec, /* finobt record */ | |
1491 | int offset) /* inode offset */ | |
1492 | { | |
1493 | struct xfs_inobt_rec_incore rec; | |
1494 | int error; | |
1495 | int i; | |
1496 | ||
1497 | error = xfs_inobt_lookup(cur, frec->ir_startino, XFS_LOOKUP_EQ, &i); | |
1498 | if (error) | |
1499 | return error; | |
19ebedcf | 1500 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
ff105f75 DC |
1501 | |
1502 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
1503 | if (error) | |
1504 | return error; | |
19ebedcf | 1505 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
ff105f75 DC |
1506 | ASSERT((XFS_AGINO_TO_OFFSET(cur->bc_mp, rec.ir_startino) % |
1507 | XFS_INODES_PER_CHUNK) == 0); | |
1508 | ||
1509 | rec.ir_free &= ~XFS_INOBT_MASK(offset); | |
1510 | rec.ir_freecount--; | |
1511 | ||
19ebedcf | 1512 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, (rec.ir_free == frec->ir_free) && |
ff105f75 DC |
1513 | (rec.ir_freecount == frec->ir_freecount)); |
1514 | ||
5a35bf2c | 1515 | return xfs_inobt_update(cur, &rec); |
ff105f75 DC |
1516 | } |
1517 | ||
1518 | /* | |
1519 | * Allocate an inode using the free inode btree, if available. Otherwise, fall | |
1520 | * back to the inobt search algorithm. | |
1521 | * | |
1522 | * The caller selected an AG for us, and made sure that free inodes are | |
1523 | * available. | |
1524 | */ | |
88fc7306 BF |
1525 | STATIC int |
1526 | xfs_dialloc_ag( | |
1527 | struct xfs_trans *tp, | |
1528 | struct xfs_buf *agbp, | |
1529 | xfs_ino_t parent, | |
1530 | xfs_ino_t *inop) | |
1531 | { | |
1532 | struct xfs_mount *mp = tp->t_mountp; | |
1533 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
1534 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
1535 | xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent); | |
1536 | xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent); | |
1537 | struct xfs_perag *pag; | |
ff105f75 DC |
1538 | struct xfs_btree_cur *cur; /* finobt cursor */ |
1539 | struct xfs_btree_cur *icur; /* inobt cursor */ | |
88fc7306 | 1540 | struct xfs_inobt_rec_incore rec; |
88fc7306 BF |
1541 | xfs_ino_t ino; |
1542 | int error; | |
1543 | int offset; | |
ff105f75 | 1544 | int i; |
88fc7306 BF |
1545 | |
1546 | if (!xfs_sb_version_hasfinobt(&mp->m_sb)) | |
ff105f75 | 1547 | return xfs_dialloc_ag_inobt(tp, agbp, parent, inop); |
88fc7306 BF |
1548 | |
1549 | pag = xfs_perag_get(mp, agno); | |
1550 | ||
1551 | /* | |
1552 | * If pagino is 0 (this is the root inode allocation) use newino. | |
1553 | * This must work because we've just allocated some. | |
1554 | */ | |
1555 | if (!pagino) | |
1556 | pagino = be32_to_cpu(agi->agi_newino); | |
1557 | ||
1558 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO); | |
1559 | ||
1560 | error = xfs_check_agi_freecount(cur, agi); | |
1561 | if (error) | |
1562 | goto error_cur; | |
1563 | ||
ff105f75 DC |
1564 | /* |
1565 | * The search algorithm depends on whether we're in the same AG as the | |
1566 | * parent. If so, find the closest available inode to the parent. If | |
1567 | * not, consider the agi hint or find the first free inode in the AG. | |
1568 | */ | |
1569 | if (agno == pagno) | |
1570 | error = xfs_dialloc_ag_finobt_near(pagino, &cur, &rec); | |
1571 | else | |
1572 | error = xfs_dialloc_ag_finobt_newino(agi, cur, &rec); | |
1573 | if (error) | |
1574 | goto error_cur; | |
88fc7306 | 1575 | |
01792d3b | 1576 | offset = xfs_inobt_first_free_inode(&rec); |
88fc7306 BF |
1577 | ASSERT(offset >= 0); |
1578 | ASSERT(offset < XFS_INODES_PER_CHUNK); | |
1579 | ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) % | |
1580 | XFS_INODES_PER_CHUNK) == 0); | |
1581 | ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset); | |
1582 | ||
1583 | /* | |
1584 | * Modify or remove the finobt record. | |
1585 | */ | |
1586 | rec.ir_free &= ~XFS_INOBT_MASK(offset); | |
1587 | rec.ir_freecount--; | |
ff105f75 | 1588 | if (rec.ir_freecount) |
88fc7306 BF |
1589 | error = xfs_inobt_update(cur, &rec); |
1590 | else | |
1591 | error = xfs_btree_delete(cur, &i); | |
1592 | if (error) | |
1593 | goto error_cur; | |
1594 | ||
1595 | /* | |
ff105f75 DC |
1596 | * The finobt has now been updated appropriately. We haven't updated the |
1597 | * agi and superblock yet, so we can create an inobt cursor and validate | |
1598 | * the original freecount. If all is well, make the equivalent update to | |
1599 | * the inobt using the finobt record and offset information. | |
88fc7306 | 1600 | */ |
ff105f75 | 1601 | icur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
88fc7306 | 1602 | |
ff105f75 | 1603 | error = xfs_check_agi_freecount(icur, agi); |
88fc7306 | 1604 | if (error) |
ff105f75 | 1605 | goto error_icur; |
88fc7306 | 1606 | |
ff105f75 | 1607 | error = xfs_dialloc_ag_update_inobt(icur, &rec, offset); |
88fc7306 | 1608 | if (error) |
ff105f75 | 1609 | goto error_icur; |
88fc7306 BF |
1610 | |
1611 | /* | |
ff105f75 DC |
1612 | * Both trees have now been updated. We must update the perag and |
1613 | * superblock before we can check the freecount for each btree. | |
88fc7306 BF |
1614 | */ |
1615 | be32_add_cpu(&agi->agi_freecount, -1); | |
1616 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); | |
1617 | pag->pagi_freecount--; | |
1618 | ||
1619 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1); | |
1620 | ||
ff105f75 | 1621 | error = xfs_check_agi_freecount(icur, agi); |
88fc7306 | 1622 | if (error) |
ff105f75 | 1623 | goto error_icur; |
88fc7306 BF |
1624 | error = xfs_check_agi_freecount(cur, agi); |
1625 | if (error) | |
ff105f75 | 1626 | goto error_icur; |
88fc7306 | 1627 | |
ff105f75 | 1628 | xfs_btree_del_cursor(icur, XFS_BTREE_NOERROR); |
88fc7306 BF |
1629 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
1630 | xfs_perag_put(pag); | |
1631 | *inop = ino; | |
1632 | return 0; | |
1633 | ||
ff105f75 DC |
1634 | error_icur: |
1635 | xfs_btree_del_cursor(icur, XFS_BTREE_ERROR); | |
88fc7306 BF |
1636 | error_cur: |
1637 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
1638 | xfs_perag_put(pag); | |
1639 | return error; | |
1640 | } | |
1641 | ||
a2ceac1f DC |
1642 | /* |
1643 | * Allocate an inode on disk. | |
1644 | * | |
1645 | * Mode is used to tell whether the new inode will need space, and whether it | |
1646 | * is a directory. | |
1647 | * | |
1648 | * This function is designed to be called twice if it has to do an allocation | |
1649 | * to make more free inodes. On the first call, *IO_agbp should be set to NULL. | |
1650 | * If an inode is available without having to performn an allocation, an inode | |
1651 | * number is returned. In this case, *IO_agbp is set to NULL. If an allocation | |
1652 | * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp. | |
1653 | * The caller should then commit the current transaction, allocate a | |
1654 | * new transaction, and call xfs_dialloc() again, passing in the previous value | |
1655 | * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI | |
1656 | * buffer is locked across the two calls, the second call is guaranteed to have | |
1657 | * a free inode available. | |
1658 | * | |
1659 | * Once we successfully pick an inode its number is returned and the on-disk | |
1660 | * data structures are updated. The inode itself is not read in, since doing so | |
1661 | * would break ordering constraints with xfs_reclaim. | |
1662 | */ | |
1663 | int | |
1664 | xfs_dialloc( | |
1665 | struct xfs_trans *tp, | |
1666 | xfs_ino_t parent, | |
1667 | umode_t mode, | |
1668 | int okalloc, | |
1669 | struct xfs_buf **IO_agbp, | |
1670 | xfs_ino_t *inop) | |
1671 | { | |
1672 | struct xfs_mount *mp = tp->t_mountp; | |
1673 | struct xfs_buf *agbp; | |
1674 | xfs_agnumber_t agno; | |
1675 | int error; | |
1676 | int ialloced; | |
1677 | int noroom = 0; | |
1678 | xfs_agnumber_t start_agno; | |
1679 | struct xfs_perag *pag; | |
1680 | ||
1681 | if (*IO_agbp) { | |
1682 | /* | |
1683 | * If the caller passes in a pointer to the AGI buffer, | |
1684 | * continue where we left off before. In this case, we | |
1685 | * know that the allocation group has free inodes. | |
1686 | */ | |
1687 | agbp = *IO_agbp; | |
1688 | goto out_alloc; | |
1689 | } | |
1690 | ||
1691 | /* | |
1692 | * We do not have an agbp, so select an initial allocation | |
1693 | * group for inode allocation. | |
1694 | */ | |
1695 | start_agno = xfs_ialloc_ag_select(tp, parent, mode, okalloc); | |
1696 | if (start_agno == NULLAGNUMBER) { | |
1697 | *inop = NULLFSINO; | |
1698 | return 0; | |
1699 | } | |
1700 | ||
1701 | /* | |
1702 | * If we have already hit the ceiling of inode blocks then clear | |
1703 | * okalloc so we scan all available agi structures for a free | |
1704 | * inode. | |
cbf3beaa GW |
1705 | * |
1706 | * Read rough value of mp->m_icount by percpu_counter_read_positive, | |
1707 | * which will sacrifice the preciseness but improve the performance. | |
a2ceac1f DC |
1708 | */ |
1709 | if (mp->m_maxicount && | |
cbf3beaa GW |
1710 | percpu_counter_read_positive(&mp->m_icount) + mp->m_ialloc_inos |
1711 | > mp->m_maxicount) { | |
a2ceac1f DC |
1712 | noroom = 1; |
1713 | okalloc = 0; | |
1714 | } | |
1715 | ||
1716 | /* | |
1717 | * Loop until we find an allocation group that either has free inodes | |
1718 | * or in which we can allocate some inodes. Iterate through the | |
1719 | * allocation groups upward, wrapping at the end. | |
1720 | */ | |
1721 | agno = start_agno; | |
1722 | for (;;) { | |
1723 | pag = xfs_perag_get(mp, agno); | |
1724 | if (!pag->pagi_inodeok) { | |
1725 | xfs_ialloc_next_ag(mp); | |
1726 | goto nextag; | |
1727 | } | |
1728 | ||
1729 | if (!pag->pagi_init) { | |
1730 | error = xfs_ialloc_pagi_init(mp, tp, agno); | |
1731 | if (error) | |
1732 | goto out_error; | |
1733 | } | |
1734 | ||
1735 | /* | |
1736 | * Do a first racy fast path check if this AG is usable. | |
1737 | */ | |
1738 | if (!pag->pagi_freecount && !okalloc) | |
1739 | goto nextag; | |
1740 | ||
1741 | /* | |
1742 | * Then read in the AGI buffer and recheck with the AGI buffer | |
1743 | * lock held. | |
1744 | */ | |
1745 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); | |
1746 | if (error) | |
1747 | goto out_error; | |
1748 | ||
1749 | if (pag->pagi_freecount) { | |
1750 | xfs_perag_put(pag); | |
1751 | goto out_alloc; | |
1752 | } | |
1753 | ||
1754 | if (!okalloc) | |
1755 | goto nextag_relse_buffer; | |
1756 | ||
1757 | ||
1758 | error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced); | |
1759 | if (error) { | |
1760 | xfs_trans_brelse(tp, agbp); | |
1761 | ||
12b53197 | 1762 | if (error != -ENOSPC) |
a2ceac1f DC |
1763 | goto out_error; |
1764 | ||
1765 | xfs_perag_put(pag); | |
1766 | *inop = NULLFSINO; | |
1767 | return 0; | |
1768 | } | |
1769 | ||
1770 | if (ialloced) { | |
1771 | /* | |
1772 | * We successfully allocated some inodes, return | |
1773 | * the current context to the caller so that it | |
1774 | * can commit the current transaction and call | |
1775 | * us again where we left off. | |
1776 | */ | |
1777 | ASSERT(pag->pagi_freecount > 0); | |
1778 | xfs_perag_put(pag); | |
1779 | ||
1780 | *IO_agbp = agbp; | |
1781 | *inop = NULLFSINO; | |
1782 | return 0; | |
1783 | } | |
1784 | ||
1785 | nextag_relse_buffer: | |
1786 | xfs_trans_brelse(tp, agbp); | |
1787 | nextag: | |
1788 | xfs_perag_put(pag); | |
1789 | if (++agno == mp->m_sb.sb_agcount) | |
1790 | agno = 0; | |
1791 | if (agno == start_agno) { | |
1792 | *inop = NULLFSINO; | |
12b53197 | 1793 | return noroom ? -ENOSPC : 0; |
a2ceac1f DC |
1794 | } |
1795 | } | |
1796 | ||
1797 | out_alloc: | |
1798 | *IO_agbp = NULL; | |
1799 | return xfs_dialloc_ag(tp, agbp, parent, inop); | |
1800 | out_error: | |
1801 | xfs_perag_put(pag); | |
1e68581b | 1802 | return error; |
a2ceac1f DC |
1803 | } |
1804 | ||
7338c4b8 BF |
1805 | /* |
1806 | * Free the blocks of an inode chunk. We must consider that the inode chunk | |
1807 | * might be sparse and only free the regions that are allocated as part of the | |
1808 | * chunk. | |
1809 | */ | |
1810 | STATIC void | |
1811 | xfs_difree_inode_chunk( | |
1812 | struct xfs_mount *mp, | |
1813 | xfs_agnumber_t agno, | |
1814 | struct xfs_inobt_rec_incore *rec, | |
1815 | struct xfs_bmap_free *flist) | |
1816 | { | |
1817 | xfs_agblock_t sagbno = XFS_AGINO_TO_AGBNO(mp, rec->ir_startino); | |
1818 | int startidx, endidx; | |
1819 | int nextbit; | |
1820 | xfs_agblock_t agbno; | |
1821 | int contigblk; | |
1822 | DECLARE_BITMAP(holemask, XFS_INOBT_HOLEMASK_BITS); | |
1823 | ||
1824 | if (!xfs_inobt_issparse(rec->ir_holemask)) { | |
1825 | /* not sparse, calculate extent info directly */ | |
af2e7c6e | 1826 | xfs_bmap_add_free(mp, flist, XFS_AGB_TO_FSB(mp, agno, |
7338c4b8 | 1827 | XFS_AGINO_TO_AGBNO(mp, rec->ir_startino)), |
af2e7c6e | 1828 | mp->m_ialloc_blks); |
7338c4b8 BF |
1829 | return; |
1830 | } | |
1831 | ||
1832 | /* holemask is only 16-bits (fits in an unsigned long) */ | |
1833 | ASSERT(sizeof(rec->ir_holemask) <= sizeof(holemask[0])); | |
1834 | holemask[0] = rec->ir_holemask; | |
1835 | ||
1836 | /* | |
1837 | * Find contiguous ranges of zeroes (i.e., allocated regions) in the | |
1838 | * holemask and convert the start/end index of each range to an extent. | |
1839 | * We start with the start and end index both pointing at the first 0 in | |
1840 | * the mask. | |
1841 | */ | |
1842 | startidx = endidx = find_first_zero_bit(holemask, | |
1843 | XFS_INOBT_HOLEMASK_BITS); | |
1844 | nextbit = startidx + 1; | |
1845 | while (startidx < XFS_INOBT_HOLEMASK_BITS) { | |
1846 | nextbit = find_next_zero_bit(holemask, XFS_INOBT_HOLEMASK_BITS, | |
1847 | nextbit); | |
1848 | /* | |
1849 | * If the next zero bit is contiguous, update the end index of | |
1850 | * the current range and continue. | |
1851 | */ | |
1852 | if (nextbit != XFS_INOBT_HOLEMASK_BITS && | |
1853 | nextbit == endidx + 1) { | |
1854 | endidx = nextbit; | |
1855 | goto next; | |
1856 | } | |
1857 | ||
1858 | /* | |
1859 | * nextbit is not contiguous with the current end index. Convert | |
1860 | * the current start/end to an extent and add it to the free | |
1861 | * list. | |
1862 | */ | |
1863 | agbno = sagbno + (startidx * XFS_INODES_PER_HOLEMASK_BIT) / | |
1864 | mp->m_sb.sb_inopblock; | |
1865 | contigblk = ((endidx - startidx + 1) * | |
1866 | XFS_INODES_PER_HOLEMASK_BIT) / | |
1867 | mp->m_sb.sb_inopblock; | |
1868 | ||
1869 | ASSERT(agbno % mp->m_sb.sb_spino_align == 0); | |
1870 | ASSERT(contigblk % mp->m_sb.sb_spino_align == 0); | |
af2e7c6e DW |
1871 | xfs_bmap_add_free(mp, flist, XFS_AGB_TO_FSB(mp, agno, agbno), |
1872 | contigblk); | |
7338c4b8 BF |
1873 | |
1874 | /* reset range to current bit and carry on... */ | |
1875 | startidx = endidx = nextbit; | |
1876 | ||
1877 | next: | |
1878 | nextbit++; | |
1879 | } | |
1880 | } | |
1881 | ||
eb9a297a BF |
1882 | STATIC int |
1883 | xfs_difree_inobt( | |
1884 | struct xfs_mount *mp, | |
1885 | struct xfs_trans *tp, | |
1886 | struct xfs_buf *agbp, | |
1887 | xfs_agino_t agino, | |
1888 | struct xfs_bmap_free *flist, | |
5a3b2e0a | 1889 | struct xfs_icluster *xic, |
eb9a297a | 1890 | struct xfs_inobt_rec_incore *orec) |
3439d03a | 1891 | { |
eb9a297a BF |
1892 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); |
1893 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
1894 | struct xfs_perag *pag; | |
1895 | struct xfs_btree_cur *cur; | |
1896 | struct xfs_inobt_rec_incore rec; | |
1897 | int ilen; | |
1898 | int error; | |
1899 | int i; | |
1900 | int off; | |
3439d03a | 1901 | |
3439d03a | 1902 | ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC)); |
eb9a297a BF |
1903 | ASSERT(XFS_AGINO_TO_AGBNO(mp, agino) < be32_to_cpu(agi->agi_length)); |
1904 | ||
3439d03a DC |
1905 | /* |
1906 | * Initialize the cursor. | |
1907 | */ | |
70eb7337 | 1908 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
3439d03a DC |
1909 | |
1910 | error = xfs_check_agi_freecount(cur, agi); | |
1911 | if (error) | |
1912 | goto error0; | |
1913 | ||
1914 | /* | |
1915 | * Look for the entry describing this inode. | |
1916 | */ | |
1917 | if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) { | |
1918 | xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.", | |
1919 | __func__, error); | |
1920 | goto error0; | |
1921 | } | |
19ebedcf | 1922 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
3439d03a DC |
1923 | error = xfs_inobt_get_rec(cur, &rec, &i); |
1924 | if (error) { | |
1925 | xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.", | |
1926 | __func__, error); | |
1927 | goto error0; | |
1928 | } | |
19ebedcf | 1929 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
3439d03a DC |
1930 | /* |
1931 | * Get the offset in the inode chunk. | |
1932 | */ | |
1933 | off = agino - rec.ir_startino; | |
1934 | ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK); | |
1935 | ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off))); | |
1936 | /* | |
1937 | * Mark the inode free & increment the count. | |
1938 | */ | |
1939 | rec.ir_free |= XFS_INOBT_MASK(off); | |
1940 | rec.ir_freecount++; | |
1941 | ||
1942 | /* | |
f6580bcf BF |
1943 | * When an inode chunk is free, it becomes eligible for removal. Don't |
1944 | * remove the chunk if the block size is large enough for multiple inode | |
1945 | * chunks (that might not be free). | |
3439d03a DC |
1946 | */ |
1947 | if (!(mp->m_flags & XFS_MOUNT_IKEEP) && | |
f6580bcf BF |
1948 | rec.ir_free == XFS_INOBT_ALL_FREE && |
1949 | mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK) { | |
5a3b2e0a BF |
1950 | xic->deleted = 1; |
1951 | xic->first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino); | |
1952 | xic->alloc = xfs_inobt_irec_to_allocmask(&rec); | |
3439d03a DC |
1953 | |
1954 | /* | |
1955 | * Remove the inode cluster from the AGI B+Tree, adjust the | |
1956 | * AGI and Superblock inode counts, and mark the disk space | |
1957 | * to be freed when the transaction is committed. | |
1958 | */ | |
f6580bcf | 1959 | ilen = rec.ir_freecount; |
3439d03a DC |
1960 | be32_add_cpu(&agi->agi_count, -ilen); |
1961 | be32_add_cpu(&agi->agi_freecount, -(ilen - 1)); | |
1962 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT); | |
1963 | pag = xfs_perag_get(mp, agno); | |
1964 | pag->pagi_freecount -= ilen - 1; | |
1965 | xfs_perag_put(pag); | |
1966 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen); | |
1967 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1)); | |
1968 | ||
1969 | if ((error = xfs_btree_delete(cur, &i))) { | |
1970 | xfs_warn(mp, "%s: xfs_btree_delete returned error %d.", | |
1971 | __func__, error); | |
1972 | goto error0; | |
1973 | } | |
1974 | ||
7338c4b8 | 1975 | xfs_difree_inode_chunk(mp, agno, &rec, flist); |
3439d03a | 1976 | } else { |
5a3b2e0a | 1977 | xic->deleted = 0; |
3439d03a DC |
1978 | |
1979 | error = xfs_inobt_update(cur, &rec); | |
1980 | if (error) { | |
1981 | xfs_warn(mp, "%s: xfs_inobt_update returned error %d.", | |
1982 | __func__, error); | |
1983 | goto error0; | |
1984 | } | |
1985 | ||
f8149110 | 1986 | /* |
3439d03a DC |
1987 | * Change the inode free counts and log the ag/sb changes. |
1988 | */ | |
1989 | be32_add_cpu(&agi->agi_freecount, 1); | |
1990 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); | |
1991 | pag = xfs_perag_get(mp, agno); | |
1992 | pag->pagi_freecount++; | |
1993 | xfs_perag_put(pag); | |
1994 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1); | |
1995 | } | |
1996 | ||
1997 | error = xfs_check_agi_freecount(cur, agi); | |
1998 | if (error) | |
1999 | goto error0; | |
2000 | ||
eb9a297a | 2001 | *orec = rec; |
3439d03a DC |
2002 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
2003 | return 0; | |
2004 | ||
2005 | error0: | |
2006 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
2007 | return error; | |
2008 | } | |
2009 | ||
1bb93fd1 BF |
2010 | /* |
2011 | * Free an inode in the free inode btree. | |
2012 | */ | |
2013 | STATIC int | |
2014 | xfs_difree_finobt( | |
2015 | struct xfs_mount *mp, | |
2016 | struct xfs_trans *tp, | |
2017 | struct xfs_buf *agbp, | |
2018 | xfs_agino_t agino, | |
2019 | struct xfs_inobt_rec_incore *ibtrec) /* inobt record */ | |
2020 | { | |
2021 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
2022 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
2023 | struct xfs_btree_cur *cur; | |
2024 | struct xfs_inobt_rec_incore rec; | |
2025 | int offset = agino - ibtrec->ir_startino; | |
2026 | int error; | |
2027 | int i; | |
2028 | ||
2029 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO); | |
2030 | ||
2031 | error = xfs_inobt_lookup(cur, ibtrec->ir_startino, XFS_LOOKUP_EQ, &i); | |
2032 | if (error) | |
2033 | goto error; | |
2034 | if (i == 0) { | |
2035 | /* | |
2036 | * If the record does not exist in the finobt, we must have just | |
2037 | * freed an inode in a previously fully allocated chunk. If not, | |
2038 | * something is out of sync. | |
2039 | */ | |
19ebedcf | 2040 | XFS_WANT_CORRUPTED_GOTO(mp, ibtrec->ir_freecount == 1, error); |
1bb93fd1 | 2041 | |
11640e30 BF |
2042 | error = xfs_inobt_insert_rec(cur, ibtrec->ir_holemask, |
2043 | ibtrec->ir_count, | |
2044 | ibtrec->ir_freecount, | |
1bb93fd1 BF |
2045 | ibtrec->ir_free, &i); |
2046 | if (error) | |
2047 | goto error; | |
2048 | ASSERT(i == 1); | |
2049 | ||
2050 | goto out; | |
2051 | } | |
2052 | ||
2053 | /* | |
ff105f75 DC |
2054 | * Read and update the existing record. We could just copy the ibtrec |
2055 | * across here, but that would defeat the purpose of having redundant | |
2056 | * metadata. By making the modifications independently, we can catch | |
2057 | * corruptions that we wouldn't see if we just copied from one record | |
2058 | * to another. | |
1bb93fd1 BF |
2059 | */ |
2060 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
2061 | if (error) | |
2062 | goto error; | |
19ebedcf | 2063 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error); |
1bb93fd1 BF |
2064 | |
2065 | rec.ir_free |= XFS_INOBT_MASK(offset); | |
2066 | rec.ir_freecount++; | |
2067 | ||
19ebedcf | 2068 | XFS_WANT_CORRUPTED_GOTO(mp, (rec.ir_free == ibtrec->ir_free) && |
1bb93fd1 BF |
2069 | (rec.ir_freecount == ibtrec->ir_freecount), |
2070 | error); | |
2071 | ||
2072 | /* | |
2073 | * The content of inobt records should always match between the inobt | |
2074 | * and finobt. The lifecycle of records in the finobt is different from | |
2075 | * the inobt in that the finobt only tracks records with at least one | |
ff105f75 DC |
2076 | * free inode. Hence, if all of the inodes are free and we aren't |
2077 | * keeping inode chunks permanently on disk, remove the record. | |
2078 | * Otherwise, update the record with the new information. | |
f6580bcf BF |
2079 | * |
2080 | * Note that we currently can't free chunks when the block size is large | |
2081 | * enough for multiple chunks. Leave the finobt record to remain in sync | |
2082 | * with the inobt. | |
1bb93fd1 | 2083 | */ |
f6580bcf BF |
2084 | if (rec.ir_free == XFS_INOBT_ALL_FREE && |
2085 | mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK && | |
1bb93fd1 | 2086 | !(mp->m_flags & XFS_MOUNT_IKEEP)) { |
1bb93fd1 BF |
2087 | error = xfs_btree_delete(cur, &i); |
2088 | if (error) | |
2089 | goto error; | |
2090 | ASSERT(i == 1); | |
2091 | } else { | |
1bb93fd1 BF |
2092 | error = xfs_inobt_update(cur, &rec); |
2093 | if (error) | |
2094 | goto error; | |
2095 | } | |
2096 | ||
2097 | out: | |
2098 | error = xfs_check_agi_freecount(cur, agi); | |
2099 | if (error) | |
2100 | goto error; | |
2101 | ||
2102 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
2103 | return 0; | |
2104 | ||
2105 | error: | |
2106 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
2107 | return error; | |
2108 | } | |
2109 | ||
eb9a297a BF |
2110 | /* |
2111 | * Free disk inode. Carefully avoids touching the incore inode, all | |
2112 | * manipulations incore are the caller's responsibility. | |
2113 | * The on-disk inode is not changed by this operation, only the | |
2114 | * btree (free inode mask) is changed. | |
2115 | */ | |
2116 | int | |
2117 | xfs_difree( | |
2118 | struct xfs_trans *tp, /* transaction pointer */ | |
2119 | xfs_ino_t inode, /* inode to be freed */ | |
2120 | struct xfs_bmap_free *flist, /* extents to free */ | |
5a3b2e0a | 2121 | struct xfs_icluster *xic) /* cluster info if deleted */ |
eb9a297a BF |
2122 | { |
2123 | /* REFERENCED */ | |
2124 | xfs_agblock_t agbno; /* block number containing inode */ | |
2125 | struct xfs_buf *agbp; /* buffer for allocation group header */ | |
2126 | xfs_agino_t agino; /* allocation group inode number */ | |
2127 | xfs_agnumber_t agno; /* allocation group number */ | |
2128 | int error; /* error return value */ | |
2129 | struct xfs_mount *mp; /* mount structure for filesystem */ | |
2130 | struct xfs_inobt_rec_incore rec;/* btree record */ | |
2131 | ||
2132 | mp = tp->t_mountp; | |
2133 | ||
2134 | /* | |
2135 | * Break up inode number into its components. | |
2136 | */ | |
2137 | agno = XFS_INO_TO_AGNO(mp, inode); | |
2138 | if (agno >= mp->m_sb.sb_agcount) { | |
2139 | xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).", | |
2140 | __func__, agno, mp->m_sb.sb_agcount); | |
2141 | ASSERT(0); | |
12b53197 | 2142 | return -EINVAL; |
eb9a297a BF |
2143 | } |
2144 | agino = XFS_INO_TO_AGINO(mp, inode); | |
2145 | if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
2146 | xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).", | |
2147 | __func__, (unsigned long long)inode, | |
2148 | (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino)); | |
2149 | ASSERT(0); | |
12b53197 | 2150 | return -EINVAL; |
eb9a297a BF |
2151 | } |
2152 | agbno = XFS_AGINO_TO_AGBNO(mp, agino); | |
2153 | if (agbno >= mp->m_sb.sb_agblocks) { | |
2154 | xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).", | |
2155 | __func__, agbno, mp->m_sb.sb_agblocks); | |
2156 | ASSERT(0); | |
12b53197 | 2157 | return -EINVAL; |
eb9a297a BF |
2158 | } |
2159 | /* | |
2160 | * Get the allocation group header. | |
2161 | */ | |
2162 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); | |
2163 | if (error) { | |
2164 | xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.", | |
2165 | __func__, error); | |
2166 | return error; | |
2167 | } | |
2168 | ||
2169 | /* | |
2170 | * Fix up the inode allocation btree. | |
2171 | */ | |
5a3b2e0a | 2172 | error = xfs_difree_inobt(mp, tp, agbp, agino, flist, xic, &rec); |
eb9a297a BF |
2173 | if (error) |
2174 | goto error0; | |
2175 | ||
1bb93fd1 BF |
2176 | /* |
2177 | * Fix up the free inode btree. | |
2178 | */ | |
2179 | if (xfs_sb_version_hasfinobt(&mp->m_sb)) { | |
2180 | error = xfs_difree_finobt(mp, tp, agbp, agino, &rec); | |
2181 | if (error) | |
2182 | goto error0; | |
2183 | } | |
2184 | ||
eb9a297a BF |
2185 | return 0; |
2186 | ||
2187 | error0: | |
2188 | return error; | |
2189 | } | |
2190 | ||
56b2de80 DC |
2191 | STATIC int |
2192 | xfs_imap_lookup( | |
2193 | struct xfs_mount *mp, | |
2194 | struct xfs_trans *tp, | |
2195 | xfs_agnumber_t agno, | |
2196 | xfs_agino_t agino, | |
2197 | xfs_agblock_t agbno, | |
2198 | xfs_agblock_t *chunk_agbno, | |
2199 | xfs_agblock_t *offset_agbno, | |
2200 | int flags) | |
2201 | { | |
2202 | struct xfs_inobt_rec_incore rec; | |
2203 | struct xfs_btree_cur *cur; | |
2204 | struct xfs_buf *agbp; | |
2205 | int error; | |
2206 | int i; | |
2207 | ||
2208 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); | |
2209 | if (error) { | |
a2ceac1f DC |
2210 | xfs_alert(mp, |
2211 | "%s: xfs_ialloc_read_agi() returned error %d, agno %d", | |
2212 | __func__, error, agno); | |
56b2de80 DC |
2213 | return error; |
2214 | } | |
2215 | ||
2216 | /* | |
2217 | * Lookup the inode record for the given agino. If the record cannot be | |
2218 | * found, then it's an invalid inode number and we should abort. Once | |
2219 | * we have a record, we need to ensure it contains the inode number | |
2220 | * we are looking up. | |
2221 | */ | |
70eb7337 | 2222 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
56b2de80 DC |
2223 | error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i); |
2224 | if (!error) { | |
2225 | if (i) | |
2226 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
2227 | if (!error && i == 0) | |
12b53197 | 2228 | error = -EINVAL; |
56b2de80 DC |
2229 | } |
2230 | ||
2231 | xfs_trans_brelse(tp, agbp); | |
2232 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
2233 | if (error) | |
2234 | return error; | |
2235 | ||
2236 | /* check that the returned record contains the required inode */ | |
2237 | if (rec.ir_startino > agino || | |
ff105f75 | 2238 | rec.ir_startino + mp->m_ialloc_inos <= agino) |
12b53197 | 2239 | return -EINVAL; |
56b2de80 DC |
2240 | |
2241 | /* for untrusted inodes check it is allocated first */ | |
2242 | if ((flags & XFS_IGET_UNTRUSTED) && | |
2243 | (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino))) | |
12b53197 | 2244 | return -EINVAL; |
56b2de80 DC |
2245 | |
2246 | *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino); | |
2247 | *offset_agbno = agbno - *chunk_agbno; | |
2248 | return 0; | |
2249 | } | |
2bd0ea18 NS |
2250 | |
2251 | /* | |
56b2de80 | 2252 | * Return the location of the inode in imap, for mapping it into a buffer. |
2bd0ea18 | 2253 | */ |
2bd0ea18 | 2254 | int |
56b2de80 DC |
2255 | xfs_imap( |
2256 | xfs_mount_t *mp, /* file system mount structure */ | |
2257 | xfs_trans_t *tp, /* transaction pointer */ | |
2bd0ea18 | 2258 | xfs_ino_t ino, /* inode to locate */ |
56b2de80 DC |
2259 | struct xfs_imap *imap, /* location map structure */ |
2260 | uint flags) /* flags for inode btree lookup */ | |
2bd0ea18 NS |
2261 | { |
2262 | xfs_agblock_t agbno; /* block number of inode in the alloc group */ | |
2bd0ea18 NS |
2263 | xfs_agino_t agino; /* inode number within alloc group */ |
2264 | xfs_agnumber_t agno; /* allocation group number */ | |
2265 | int blks_per_cluster; /* num blocks per inode cluster */ | |
2266 | xfs_agblock_t chunk_agbno; /* first block in inode chunk */ | |
2bd0ea18 | 2267 | xfs_agblock_t cluster_agbno; /* first block in inode cluster */ |
2bd0ea18 | 2268 | int error; /* error code */ |
dfc130f3 | 2269 | int offset; /* index of inode in its buffer */ |
6bddecbc | 2270 | xfs_agblock_t offset_agbno; /* blks from chunk start to inode */ |
2bd0ea18 NS |
2271 | |
2272 | ASSERT(ino != NULLFSINO); | |
56b2de80 | 2273 | |
2bd0ea18 NS |
2274 | /* |
2275 | * Split up the inode number into its parts. | |
2276 | */ | |
2277 | agno = XFS_INO_TO_AGNO(mp, ino); | |
2278 | agino = XFS_INO_TO_AGINO(mp, ino); | |
2279 | agbno = XFS_AGINO_TO_AGBNO(mp, agino); | |
2280 | if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks || | |
63518810 NS |
2281 | ino != XFS_AGINO_TO_INO(mp, agno, agino)) { |
2282 | #ifdef DEBUG | |
56b2de80 DC |
2283 | /* |
2284 | * Don't output diagnostic information for untrusted inodes | |
2285 | * as they can be invalid without implying corruption. | |
2286 | */ | |
2287 | if (flags & XFS_IGET_UNTRUSTED) | |
12b53197 | 2288 | return -EINVAL; |
5000d01d | 2289 | if (agno >= mp->m_sb.sb_agcount) { |
a2ceac1f DC |
2290 | xfs_alert(mp, |
2291 | "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)", | |
2292 | __func__, agno, mp->m_sb.sb_agcount); | |
63518810 NS |
2293 | } |
2294 | if (agbno >= mp->m_sb.sb_agblocks) { | |
a2ceac1f DC |
2295 | xfs_alert(mp, |
2296 | "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)", | |
2297 | __func__, (unsigned long long)agbno, | |
2298 | (unsigned long)mp->m_sb.sb_agblocks); | |
63518810 NS |
2299 | } |
2300 | if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
a2ceac1f DC |
2301 | xfs_alert(mp, |
2302 | "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)", | |
2303 | __func__, ino, | |
2304 | XFS_AGINO_TO_INO(mp, agno, agino)); | |
63518810 | 2305 | } |
5e656dbb | 2306 | xfs_stack_trace(); |
63518810 | 2307 | #endif /* DEBUG */ |
12b53197 | 2308 | return -EINVAL; |
63518810 | 2309 | } |
56b2de80 | 2310 | |
ff105f75 | 2311 | blks_per_cluster = xfs_icluster_size_fsb(mp); |
56b2de80 DC |
2312 | |
2313 | /* | |
2314 | * For bulkstat and handle lookups, we have an untrusted inode number | |
2315 | * that we have to verify is valid. We cannot do this just by reading | |
2316 | * the inode buffer as it may have been unlinked and removed leaving | |
2317 | * inodes in stale state on disk. Hence we have to do a btree lookup | |
2318 | * in all cases where an untrusted inode number is passed. | |
2319 | */ | |
2320 | if (flags & XFS_IGET_UNTRUSTED) { | |
2321 | error = xfs_imap_lookup(mp, tp, agno, agino, agbno, | |
2322 | &chunk_agbno, &offset_agbno, flags); | |
2323 | if (error) | |
2324 | return error; | |
2325 | goto out_map; | |
2326 | } | |
2327 | ||
2328 | /* | |
2329 | * If the inode cluster size is the same as the blocksize or | |
2330 | * smaller we get to the buffer by simple arithmetics. | |
2331 | */ | |
ff105f75 | 2332 | if (blks_per_cluster == 1) { |
2bd0ea18 NS |
2333 | offset = XFS_INO_TO_OFFSET(mp, ino); |
2334 | ASSERT(offset < mp->m_sb.sb_inopblock); | |
56b2de80 DC |
2335 | |
2336 | imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno); | |
2337 | imap->im_len = XFS_FSB_TO_BB(mp, 1); | |
919dbc78 | 2338 | imap->im_boffset = (unsigned short)(offset << mp->m_sb.sb_inodelog); |
2bd0ea18 NS |
2339 | return 0; |
2340 | } | |
56b2de80 DC |
2341 | |
2342 | /* | |
2343 | * If the inode chunks are aligned then use simple maths to | |
2344 | * find the location. Otherwise we have to do a btree | |
2345 | * lookup to find the location. | |
2346 | */ | |
2bd0ea18 NS |
2347 | if (mp->m_inoalign_mask) { |
2348 | offset_agbno = agbno & mp->m_inoalign_mask; | |
2349 | chunk_agbno = agbno - offset_agbno; | |
2350 | } else { | |
56b2de80 DC |
2351 | error = xfs_imap_lookup(mp, tp, agno, agino, agbno, |
2352 | &chunk_agbno, &offset_agbno, flags); | |
2bd0ea18 NS |
2353 | if (error) |
2354 | return error; | |
2bd0ea18 | 2355 | } |
56b2de80 DC |
2356 | |
2357 | out_map: | |
2bd0ea18 NS |
2358 | ASSERT(agbno >= chunk_agbno); |
2359 | cluster_agbno = chunk_agbno + | |
2360 | ((offset_agbno / blks_per_cluster) * blks_per_cluster); | |
2361 | offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) + | |
2362 | XFS_INO_TO_OFFSET(mp, ino); | |
56b2de80 DC |
2363 | |
2364 | imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno); | |
2365 | imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster); | |
919dbc78 | 2366 | imap->im_boffset = (unsigned short)(offset << mp->m_sb.sb_inodelog); |
56b2de80 DC |
2367 | |
2368 | /* | |
2369 | * If the inode number maps to a block outside the bounds | |
2370 | * of the file system then return NULL rather than calling | |
2371 | * read_buf and panicing when we get an error from the | |
2372 | * driver. | |
2373 | */ | |
2374 | if ((imap->im_blkno + imap->im_len) > | |
2375 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) { | |
a2ceac1f DC |
2376 | xfs_alert(mp, |
2377 | "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)", | |
2378 | __func__, (unsigned long long) imap->im_blkno, | |
56b2de80 DC |
2379 | (unsigned long long) imap->im_len, |
2380 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)); | |
12b53197 | 2381 | return -EINVAL; |
56b2de80 | 2382 | } |
2bd0ea18 | 2383 | return 0; |
2bd0ea18 NS |
2384 | } |
2385 | ||
2386 | /* | |
2387 | * Compute and fill in value of m_in_maxlevels. | |
2388 | */ | |
2389 | void | |
2390 | xfs_ialloc_compute_maxlevels( | |
2391 | xfs_mount_t *mp) /* file system mount structure */ | |
2392 | { | |
730e2a19 DW |
2393 | uint inodes; |
2394 | ||
2395 | inodes = (1LL << XFS_INO_AGINO_BITS(mp)) >> XFS_INODES_PER_CHUNK_LOG; | |
2396 | mp->m_in_maxlevels = xfs_btree_compute_maxlevels(mp, mp->m_inobt_mnr, | |
2397 | inodes); | |
2bd0ea18 NS |
2398 | } |
2399 | ||
2400 | /* | |
ff105f75 DC |
2401 | * Log specified fields for the ag hdr (inode section). The growth of the agi |
2402 | * structure over time requires that we interpret the buffer as two logical | |
2403 | * regions delineated by the end of the unlinked list. This is due to the size | |
2404 | * of the hash table and its location in the middle of the agi. | |
2405 | * | |
2406 | * For example, a request to log a field before agi_unlinked and a field after | |
2407 | * agi_unlinked could cause us to log the entire hash table and use an excessive | |
2408 | * amount of log space. To avoid this behavior, log the region up through | |
2409 | * agi_unlinked in one call and the region after agi_unlinked through the end of | |
2410 | * the structure in another. | |
2bd0ea18 NS |
2411 | */ |
2412 | void | |
2413 | xfs_ialloc_log_agi( | |
2414 | xfs_trans_t *tp, /* transaction pointer */ | |
2415 | xfs_buf_t *bp, /* allocation group header buffer */ | |
2416 | int fields) /* bitmask of fields to log */ | |
2417 | { | |
2418 | int first; /* first byte number */ | |
2419 | int last; /* last byte number */ | |
2420 | static const short offsets[] = { /* field starting offsets */ | |
2421 | /* keep in sync with bit definitions */ | |
2422 | offsetof(xfs_agi_t, agi_magicnum), | |
2423 | offsetof(xfs_agi_t, agi_versionnum), | |
2424 | offsetof(xfs_agi_t, agi_seqno), | |
2425 | offsetof(xfs_agi_t, agi_length), | |
2426 | offsetof(xfs_agi_t, agi_count), | |
2427 | offsetof(xfs_agi_t, agi_root), | |
2428 | offsetof(xfs_agi_t, agi_level), | |
2429 | offsetof(xfs_agi_t, agi_freecount), | |
2430 | offsetof(xfs_agi_t, agi_newino), | |
2431 | offsetof(xfs_agi_t, agi_dirino), | |
2432 | offsetof(xfs_agi_t, agi_unlinked), | |
c0a4c227 BF |
2433 | offsetof(xfs_agi_t, agi_free_root), |
2434 | offsetof(xfs_agi_t, agi_free_level), | |
2bd0ea18 NS |
2435 | sizeof(xfs_agi_t) |
2436 | }; | |
2437 | #ifdef DEBUG | |
2438 | xfs_agi_t *agi; /* allocation group header */ | |
2439 | ||
2440 | agi = XFS_BUF_TO_AGI(bp); | |
a2ceac1f | 2441 | ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC)); |
2bd0ea18 | 2442 | #endif |
ff105f75 | 2443 | |
c0a4c227 BF |
2444 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_AGI_BUF); |
2445 | ||
2bd0ea18 | 2446 | /* |
c0a4c227 | 2447 | * Compute byte offsets for the first and last fields in the first |
ff105f75 DC |
2448 | * region and log the agi buffer. This only logs up through |
2449 | * agi_unlinked. | |
2bd0ea18 | 2450 | */ |
c0a4c227 BF |
2451 | if (fields & XFS_AGI_ALL_BITS_R1) { |
2452 | xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R1, | |
2453 | &first, &last); | |
2454 | xfs_trans_log_buf(tp, bp, first, last); | |
2455 | } | |
2456 | ||
2457 | /* | |
ff105f75 DC |
2458 | * Mask off the bits in the first region and calculate the first and |
2459 | * last field offsets for any bits in the second region. | |
c0a4c227 BF |
2460 | */ |
2461 | fields &= ~XFS_AGI_ALL_BITS_R1; | |
2462 | if (fields) { | |
2463 | xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R2, | |
2464 | &first, &last); | |
2465 | xfs_trans_log_buf(tp, bp, first, last); | |
2466 | } | |
2bd0ea18 NS |
2467 | } |
2468 | ||
56b2de80 DC |
2469 | #ifdef DEBUG |
2470 | STATIC void | |
2471 | xfs_check_agi_unlinked( | |
2472 | struct xfs_agi *agi) | |
2473 | { | |
2474 | int i; | |
2475 | ||
2476 | for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++) | |
2477 | ASSERT(agi->agi_unlinked[i]); | |
2478 | } | |
2479 | #else | |
2480 | #define xfs_check_agi_unlinked(agi) | |
2481 | #endif | |
2482 | ||
dd5b876e | 2483 | static bool |
a2ceac1f DC |
2484 | xfs_agi_verify( |
2485 | struct xfs_buf *bp) | |
2486 | { | |
2487 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
2488 | struct xfs_agi *agi = XFS_BUF_TO_AGI(bp); | |
a2ceac1f | 2489 | |
a65d8d29 BF |
2490 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
2491 | if (!uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid)) | |
2492 | return false; | |
2493 | if (!xfs_log_check_lsn(mp, | |
2494 | be64_to_cpu(XFS_BUF_TO_AGI(bp)->agi_lsn))) | |
dd5b876e | 2495 | return false; |
a65d8d29 BF |
2496 | } |
2497 | ||
a2ceac1f DC |
2498 | /* |
2499 | * Validate the magic number of the agi block. | |
2500 | */ | |
dd5b876e DC |
2501 | if (agi->agi_magicnum != cpu_to_be32(XFS_AGI_MAGIC)) |
2502 | return false; | |
2503 | if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum))) | |
2504 | return false; | |
a2ceac1f | 2505 | |
5a35bf2c DC |
2506 | if (be32_to_cpu(agi->agi_level) > XFS_BTREE_MAXLEVELS) |
2507 | return false; | |
a2ceac1f DC |
2508 | /* |
2509 | * during growfs operations, the perag is not fully initialised, | |
2510 | * so we can't use it for any useful checking. growfs ensures we can't | |
2511 | * use it by using uncached buffers that don't have the perag attached | |
2512 | * so we can detect and avoid this problem. | |
2513 | */ | |
dd5b876e DC |
2514 | if (bp->b_pag && be32_to_cpu(agi->agi_seqno) != bp->b_pag->pag_agno) |
2515 | return false; | |
a2ceac1f | 2516 | |
a2ceac1f | 2517 | xfs_check_agi_unlinked(agi); |
dd5b876e | 2518 | return true; |
a2ceac1f DC |
2519 | } |
2520 | ||
2521 | static void | |
2522 | xfs_agi_read_verify( | |
2523 | struct xfs_buf *bp) | |
2524 | { | |
dd5b876e | 2525 | struct xfs_mount *mp = bp->b_target->bt_mount; |
dd5b876e | 2526 | |
45922933 DC |
2527 | if (xfs_sb_version_hascrc(&mp->m_sb) && |
2528 | !xfs_buf_verify_cksum(bp, XFS_AGI_CRC_OFF)) | |
12b53197 | 2529 | xfs_buf_ioerror(bp, -EFSBADCRC); |
45922933 DC |
2530 | else if (XFS_TEST_ERROR(!xfs_agi_verify(bp), mp, |
2531 | XFS_ERRTAG_IALLOC_READ_AGI, | |
2532 | XFS_RANDOM_IALLOC_READ_AGI)) | |
12b53197 | 2533 | xfs_buf_ioerror(bp, -EFSCORRUPTED); |
45922933 DC |
2534 | |
2535 | if (bp->b_error) | |
2536 | xfs_verifier_error(bp); | |
a2ceac1f DC |
2537 | } |
2538 | ||
2539 | static void | |
2540 | xfs_agi_write_verify( | |
2541 | struct xfs_buf *bp) | |
2542 | { | |
dd5b876e DC |
2543 | struct xfs_mount *mp = bp->b_target->bt_mount; |
2544 | struct xfs_buf_log_item *bip = bp->b_fspriv; | |
2545 | ||
2546 | if (!xfs_agi_verify(bp)) { | |
12b53197 | 2547 | xfs_buf_ioerror(bp, -EFSCORRUPTED); |
45922933 | 2548 | xfs_verifier_error(bp); |
dd5b876e DC |
2549 | return; |
2550 | } | |
2551 | ||
2552 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
2553 | return; | |
2554 | ||
2555 | if (bip) | |
2556 | XFS_BUF_TO_AGI(bp)->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn); | |
43b5aeed | 2557 | xfs_buf_update_cksum(bp, XFS_AGI_CRC_OFF); |
a2ceac1f DC |
2558 | } |
2559 | ||
2560 | const struct xfs_buf_ops xfs_agi_buf_ops = { | |
a3fac935 | 2561 | .name = "xfs_agi", |
a2ceac1f DC |
2562 | .verify_read = xfs_agi_read_verify, |
2563 | .verify_write = xfs_agi_write_verify, | |
2564 | }; | |
2565 | ||
2bd0ea18 NS |
2566 | /* |
2567 | * Read in the allocation group header (inode allocation section) | |
2568 | */ | |
2569 | int | |
56b2de80 DC |
2570 | xfs_read_agi( |
2571 | struct xfs_mount *mp, /* file system mount structure */ | |
2572 | struct xfs_trans *tp, /* transaction pointer */ | |
2573 | xfs_agnumber_t agno, /* allocation group number */ | |
2574 | struct xfs_buf **bpp) /* allocation group hdr buf */ | |
2bd0ea18 | 2575 | { |
56b2de80 | 2576 | int error; |
2bd0ea18 | 2577 | |
ff105f75 | 2578 | trace_xfs_read_agi(mp, agno); |
56b2de80 | 2579 | |
ff105f75 | 2580 | ASSERT(agno != NULLAGNUMBER); |
56b2de80 | 2581 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, |
9440d84d | 2582 | XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)), |
a2ceac1f | 2583 | XFS_FSS_TO_BB(mp, 1), 0, bpp, &xfs_agi_buf_ops); |
9440d84d | 2584 | if (error) |
2bd0ea18 | 2585 | return error; |
56b2de80 | 2586 | |
a2ceac1f | 2587 | xfs_buf_set_ref(*bpp, XFS_AGI_REF); |
56b2de80 DC |
2588 | return 0; |
2589 | } | |
2590 | ||
2591 | int | |
2592 | xfs_ialloc_read_agi( | |
2593 | struct xfs_mount *mp, /* file system mount structure */ | |
2594 | struct xfs_trans *tp, /* transaction pointer */ | |
2595 | xfs_agnumber_t agno, /* allocation group number */ | |
2596 | struct xfs_buf **bpp) /* allocation group hdr buf */ | |
2597 | { | |
2598 | struct xfs_agi *agi; /* allocation group header */ | |
2599 | struct xfs_perag *pag; /* per allocation group data */ | |
2600 | int error; | |
2601 | ||
ff105f75 DC |
2602 | trace_xfs_ialloc_read_agi(mp, agno); |
2603 | ||
56b2de80 DC |
2604 | error = xfs_read_agi(mp, tp, agno, bpp); |
2605 | if (error) | |
2606 | return error; | |
2607 | ||
2608 | agi = XFS_BUF_TO_AGI(*bpp); | |
2609 | pag = xfs_perag_get(mp, agno); | |
2bd0ea18 | 2610 | if (!pag->pagi_init) { |
6e3140c7 | 2611 | pag->pagi_freecount = be32_to_cpu(agi->agi_freecount); |
cdded3d8 | 2612 | pag->pagi_count = be32_to_cpu(agi->agi_count); |
2bd0ea18 | 2613 | pag->pagi_init = 1; |
9440d84d | 2614 | } |
9440d84d | 2615 | |
56b2de80 DC |
2616 | /* |
2617 | * It's possible for these to be out of sync if | |
2618 | * we are in the middle of a forced shutdown. | |
2619 | */ | |
2620 | ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) || | |
2621 | XFS_FORCED_SHUTDOWN(mp)); | |
2622 | xfs_perag_put(pag); | |
2bd0ea18 NS |
2623 | return 0; |
2624 | } | |
cdded3d8 DC |
2625 | |
2626 | /* | |
2627 | * Read in the agi to initialise the per-ag data in the mount structure | |
2628 | */ | |
2629 | int | |
2630 | xfs_ialloc_pagi_init( | |
2631 | xfs_mount_t *mp, /* file system mount structure */ | |
2632 | xfs_trans_t *tp, /* transaction pointer */ | |
2633 | xfs_agnumber_t agno) /* allocation group number */ | |
2634 | { | |
2635 | xfs_buf_t *bp = NULL; | |
2636 | int error; | |
2637 | ||
5e656dbb BN |
2638 | error = xfs_ialloc_read_agi(mp, tp, agno, &bp); |
2639 | if (error) | |
cdded3d8 DC |
2640 | return error; |
2641 | if (bp) | |
2642 | xfs_trans_brelse(tp, bp); | |
2643 | return 0; | |
2644 | } |