]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_fs.h" |
70a9883c | 20 | #include "xfs_shared.h" |
239880ef DC |
21 | #include "xfs_format.h" |
22 | #include "xfs_log_format.h" | |
23 | #include "xfs_trans_resv.h" | |
a844f451 | 24 | #include "xfs_bit.h" |
1da177e4 | 25 | #include "xfs_sb.h" |
1da177e4 | 26 | #include "xfs_mount.h" |
3ab78df2 | 27 | #include "xfs_defer.h" |
1da177e4 | 28 | #include "xfs_inode.h" |
a844f451 NS |
29 | #include "xfs_btree.h" |
30 | #include "xfs_ialloc.h" | |
a4fbe6ab | 31 | #include "xfs_ialloc_btree.h" |
1da177e4 | 32 | #include "xfs_alloc.h" |
1da177e4 | 33 | #include "xfs_rtalloc.h" |
e9e899a2 | 34 | #include "xfs_errortag.h" |
1da177e4 LT |
35 | #include "xfs_error.h" |
36 | #include "xfs_bmap.h" | |
983d09ff | 37 | #include "xfs_cksum.h" |
239880ef | 38 | #include "xfs_trans.h" |
983d09ff | 39 | #include "xfs_buf_item.h" |
ddf6ad01 | 40 | #include "xfs_icreate_item.h" |
7bb85ef3 | 41 | #include "xfs_icache.h" |
d123031a | 42 | #include "xfs_trace.h" |
a45086e2 | 43 | #include "xfs_log.h" |
340785cc | 44 | #include "xfs_rmap.h" |
1da177e4 | 45 | |
1da177e4 LT |
46 | |
47 | /* | |
48 | * Allocation group level functions. | |
49 | */ | |
e936945e | 50 | int |
75de2a91 | 51 | xfs_ialloc_cluster_alignment( |
7a1df156 | 52 | struct xfs_mount *mp) |
75de2a91 | 53 | { |
7a1df156 | 54 | if (xfs_sb_version_hasalign(&mp->m_sb) && |
8ee9fdbe | 55 | mp->m_sb.sb_inoalignmt >= xfs_icluster_size_fsb(mp)) |
7a1df156 | 56 | return mp->m_sb.sb_inoalignmt; |
75de2a91 DC |
57 | return 1; |
58 | } | |
1da177e4 | 59 | |
fe033cc8 | 60 | /* |
21875505 | 61 | * Lookup a record by ino in the btree given by cur. |
fe033cc8 | 62 | */ |
81e25176 | 63 | int /* error */ |
21875505 | 64 | xfs_inobt_lookup( |
fe033cc8 CH |
65 | struct xfs_btree_cur *cur, /* btree cursor */ |
66 | xfs_agino_t ino, /* starting inode of chunk */ | |
21875505 | 67 | xfs_lookup_t dir, /* <=, >=, == */ |
fe033cc8 CH |
68 | int *stat) /* success/failure */ |
69 | { | |
70 | cur->bc_rec.i.ir_startino = ino; | |
5419040f BF |
71 | cur->bc_rec.i.ir_holemask = 0; |
72 | cur->bc_rec.i.ir_count = 0; | |
21875505 CH |
73 | cur->bc_rec.i.ir_freecount = 0; |
74 | cur->bc_rec.i.ir_free = 0; | |
75 | return xfs_btree_lookup(cur, dir, stat); | |
fe033cc8 CH |
76 | } |
77 | ||
278d0ca1 | 78 | /* |
afabc24a | 79 | * Update the record referred to by cur to the value given. |
278d0ca1 CH |
80 | * This either works (return 0) or gets an EFSCORRUPTED error. |
81 | */ | |
82 | STATIC int /* error */ | |
83 | xfs_inobt_update( | |
84 | struct xfs_btree_cur *cur, /* btree cursor */ | |
afabc24a | 85 | xfs_inobt_rec_incore_t *irec) /* btree record */ |
278d0ca1 CH |
86 | { |
87 | union xfs_btree_rec rec; | |
88 | ||
afabc24a | 89 | rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino); |
5419040f BF |
90 | if (xfs_sb_version_hassparseinodes(&cur->bc_mp->m_sb)) { |
91 | rec.inobt.ir_u.sp.ir_holemask = cpu_to_be16(irec->ir_holemask); | |
92 | rec.inobt.ir_u.sp.ir_count = irec->ir_count; | |
93 | rec.inobt.ir_u.sp.ir_freecount = irec->ir_freecount; | |
94 | } else { | |
95 | /* ir_holemask/ir_count not supported on-disk */ | |
96 | rec.inobt.ir_u.f.ir_freecount = cpu_to_be32(irec->ir_freecount); | |
97 | } | |
afabc24a | 98 | rec.inobt.ir_free = cpu_to_be64(irec->ir_free); |
278d0ca1 CH |
99 | return xfs_btree_update(cur, &rec); |
100 | } | |
101 | ||
e936945e DW |
102 | /* Convert on-disk btree record to incore inobt record. */ |
103 | void | |
104 | xfs_inobt_btrec_to_irec( | |
105 | struct xfs_mount *mp, | |
106 | union xfs_btree_rec *rec, | |
107 | struct xfs_inobt_rec_incore *irec) | |
8cc938fe | 108 | { |
5419040f | 109 | irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino); |
e936945e | 110 | if (xfs_sb_version_hassparseinodes(&mp->m_sb)) { |
5419040f BF |
111 | irec->ir_holemask = be16_to_cpu(rec->inobt.ir_u.sp.ir_holemask); |
112 | irec->ir_count = rec->inobt.ir_u.sp.ir_count; | |
113 | irec->ir_freecount = rec->inobt.ir_u.sp.ir_freecount; | |
114 | } else { | |
115 | /* | |
116 | * ir_holemask/ir_count not supported on-disk. Fill in hardcoded | |
117 | * values for full inode chunks. | |
118 | */ | |
119 | irec->ir_holemask = XFS_INOBT_HOLEMASK_FULL; | |
120 | irec->ir_count = XFS_INODES_PER_CHUNK; | |
121 | irec->ir_freecount = | |
122 | be32_to_cpu(rec->inobt.ir_u.f.ir_freecount); | |
8cc938fe | 123 | } |
5419040f | 124 | irec->ir_free = be64_to_cpu(rec->inobt.ir_free); |
e936945e DW |
125 | } |
126 | ||
127 | /* | |
128 | * Get the data from the pointed-to record. | |
129 | */ | |
130 | int | |
131 | xfs_inobt_get_rec( | |
132 | struct xfs_btree_cur *cur, | |
133 | struct xfs_inobt_rec_incore *irec, | |
134 | int *stat) | |
135 | { | |
136 | union xfs_btree_rec *rec; | |
137 | int error; | |
138 | ||
139 | error = xfs_btree_get_rec(cur, &rec, stat); | |
140 | if (error || *stat == 0) | |
141 | return error; | |
142 | ||
143 | xfs_inobt_btrec_to_irec(cur->bc_mp, rec, irec); | |
5419040f BF |
144 | |
145 | return 0; | |
8cc938fe CH |
146 | } |
147 | ||
0aa0a756 BF |
148 | /* |
149 | * Insert a single inobt record. Cursor must already point to desired location. | |
150 | */ | |
151 | STATIC int | |
152 | xfs_inobt_insert_rec( | |
153 | struct xfs_btree_cur *cur, | |
c8ce540d DW |
154 | uint16_t holemask, |
155 | uint8_t count, | |
156 | int32_t freecount, | |
0aa0a756 BF |
157 | xfs_inofree_t free, |
158 | int *stat) | |
159 | { | |
5419040f BF |
160 | cur->bc_rec.i.ir_holemask = holemask; |
161 | cur->bc_rec.i.ir_count = count; | |
0aa0a756 BF |
162 | cur->bc_rec.i.ir_freecount = freecount; |
163 | cur->bc_rec.i.ir_free = free; | |
164 | return xfs_btree_insert(cur, stat); | |
165 | } | |
166 | ||
167 | /* | |
168 | * Insert records describing a newly allocated inode chunk into the inobt. | |
169 | */ | |
170 | STATIC int | |
171 | xfs_inobt_insert( | |
172 | struct xfs_mount *mp, | |
173 | struct xfs_trans *tp, | |
174 | struct xfs_buf *agbp, | |
175 | xfs_agino_t newino, | |
176 | xfs_agino_t newlen, | |
177 | xfs_btnum_t btnum) | |
178 | { | |
179 | struct xfs_btree_cur *cur; | |
180 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
181 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
182 | xfs_agino_t thisino; | |
183 | int i; | |
184 | int error; | |
185 | ||
186 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum); | |
187 | ||
188 | for (thisino = newino; | |
189 | thisino < newino + newlen; | |
190 | thisino += XFS_INODES_PER_CHUNK) { | |
191 | error = xfs_inobt_lookup(cur, thisino, XFS_LOOKUP_EQ, &i); | |
192 | if (error) { | |
193 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
194 | return error; | |
195 | } | |
196 | ASSERT(i == 0); | |
197 | ||
5419040f BF |
198 | error = xfs_inobt_insert_rec(cur, XFS_INOBT_HOLEMASK_FULL, |
199 | XFS_INODES_PER_CHUNK, | |
200 | XFS_INODES_PER_CHUNK, | |
0aa0a756 BF |
201 | XFS_INOBT_ALL_FREE, &i); |
202 | if (error) { | |
203 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
204 | return error; | |
205 | } | |
206 | ASSERT(i == 1); | |
207 | } | |
208 | ||
209 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
210 | ||
211 | return 0; | |
212 | } | |
213 | ||
0b48db80 DC |
214 | /* |
215 | * Verify that the number of free inodes in the AGI is correct. | |
216 | */ | |
217 | #ifdef DEBUG | |
218 | STATIC int | |
219 | xfs_check_agi_freecount( | |
220 | struct xfs_btree_cur *cur, | |
221 | struct xfs_agi *agi) | |
222 | { | |
223 | if (cur->bc_nlevels == 1) { | |
224 | xfs_inobt_rec_incore_t rec; | |
225 | int freecount = 0; | |
226 | int error; | |
227 | int i; | |
228 | ||
21875505 | 229 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); |
0b48db80 DC |
230 | if (error) |
231 | return error; | |
232 | ||
233 | do { | |
234 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
235 | if (error) | |
236 | return error; | |
237 | ||
238 | if (i) { | |
239 | freecount += rec.ir_freecount; | |
240 | error = xfs_btree_increment(cur, 0, &i); | |
241 | if (error) | |
242 | return error; | |
243 | } | |
244 | } while (i == 1); | |
245 | ||
246 | if (!XFS_FORCED_SHUTDOWN(cur->bc_mp)) | |
247 | ASSERT(freecount == be32_to_cpu(agi->agi_freecount)); | |
248 | } | |
249 | return 0; | |
250 | } | |
251 | #else | |
252 | #define xfs_check_agi_freecount(cur, agi) 0 | |
253 | #endif | |
254 | ||
85c0b2ab | 255 | /* |
28c8e41a DC |
256 | * Initialise a new set of inodes. When called without a transaction context |
257 | * (e.g. from recovery) we initiate a delayed write of the inode buffers rather | |
258 | * than logging them (which in a transaction context puts them into the AIL | |
259 | * for writeback rather than the xfsbufd queue). | |
85c0b2ab | 260 | */ |
ddf6ad01 | 261 | int |
85c0b2ab DC |
262 | xfs_ialloc_inode_init( |
263 | struct xfs_mount *mp, | |
264 | struct xfs_trans *tp, | |
28c8e41a | 265 | struct list_head *buffer_list, |
463958af | 266 | int icount, |
85c0b2ab DC |
267 | xfs_agnumber_t agno, |
268 | xfs_agblock_t agbno, | |
269 | xfs_agblock_t length, | |
270 | unsigned int gen) | |
271 | { | |
272 | struct xfs_buf *fbuf; | |
273 | struct xfs_dinode *free; | |
6e0c7b8c | 274 | int nbufs, blks_per_cluster, inodes_per_cluster; |
85c0b2ab DC |
275 | int version; |
276 | int i, j; | |
277 | xfs_daddr_t d; | |
93848a99 | 278 | xfs_ino_t ino = 0; |
85c0b2ab DC |
279 | |
280 | /* | |
6e0c7b8c JL |
281 | * Loop over the new block(s), filling in the inodes. For small block |
282 | * sizes, manipulate the inodes in buffers which are multiples of the | |
283 | * blocks size. | |
85c0b2ab | 284 | */ |
6e0c7b8c JL |
285 | blks_per_cluster = xfs_icluster_size_fsb(mp); |
286 | inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog; | |
287 | nbufs = length / blks_per_cluster; | |
85c0b2ab DC |
288 | |
289 | /* | |
93848a99 CH |
290 | * Figure out what version number to use in the inodes we create. If |
291 | * the superblock version has caught up to the one that supports the new | |
292 | * inode format, then use the new inode version. Otherwise use the old | |
293 | * version so that old kernels will continue to be able to use the file | |
294 | * system. | |
295 | * | |
296 | * For v3 inodes, we also need to write the inode number into the inode, | |
297 | * so calculate the first inode number of the chunk here as | |
298 | * XFS_OFFBNO_TO_AGINO() only works within a filesystem block, not | |
299 | * across multiple filesystem blocks (such as a cluster) and so cannot | |
300 | * be used in the cluster buffer loop below. | |
301 | * | |
302 | * Further, because we are writing the inode directly into the buffer | |
303 | * and calculating a CRC on the entire inode, we have ot log the entire | |
304 | * inode so that the entire range the CRC covers is present in the log. | |
305 | * That means for v3 inode we log the entire buffer rather than just the | |
306 | * inode cores. | |
85c0b2ab | 307 | */ |
93848a99 CH |
308 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
309 | version = 3; | |
310 | ino = XFS_AGINO_TO_INO(mp, agno, | |
311 | XFS_OFFBNO_TO_AGINO(mp, agbno, 0)); | |
ddf6ad01 DC |
312 | |
313 | /* | |
314 | * log the initialisation that is about to take place as an | |
315 | * logical operation. This means the transaction does not | |
316 | * need to log the physical changes to the inode buffers as log | |
317 | * recovery will know what initialisation is actually needed. | |
318 | * Hence we only need to log the buffers as "ordered" buffers so | |
319 | * they track in the AIL as if they were physically logged. | |
320 | */ | |
321 | if (tp) | |
463958af | 322 | xfs_icreate_log(tp, agno, agbno, icount, |
ddf6ad01 | 323 | mp->m_sb.sb_inodesize, length, gen); |
263997a6 | 324 | } else |
85c0b2ab | 325 | version = 2; |
85c0b2ab DC |
326 | |
327 | for (j = 0; j < nbufs; j++) { | |
328 | /* | |
329 | * Get the block. | |
330 | */ | |
331 | d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster)); | |
332 | fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, | |
7c4cebe8 DC |
333 | mp->m_bsize * blks_per_cluster, |
334 | XBF_UNMAPPED); | |
2a30f36d | 335 | if (!fbuf) |
2451337d | 336 | return -ENOMEM; |
ddf6ad01 DC |
337 | |
338 | /* Initialize the inode buffers and log them appropriately. */ | |
1813dd64 | 339 | fbuf->b_ops = &xfs_inode_buf_ops; |
93848a99 | 340 | xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length)); |
6e0c7b8c | 341 | for (i = 0; i < inodes_per_cluster; i++) { |
85c0b2ab | 342 | int ioffset = i << mp->m_sb.sb_inodelog; |
93848a99 | 343 | uint isize = xfs_dinode_size(version); |
85c0b2ab DC |
344 | |
345 | free = xfs_make_iptr(mp, fbuf, i); | |
346 | free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC); | |
347 | free->di_version = version; | |
348 | free->di_gen = cpu_to_be32(gen); | |
349 | free->di_next_unlinked = cpu_to_be32(NULLAGINO); | |
93848a99 CH |
350 | |
351 | if (version == 3) { | |
352 | free->di_ino = cpu_to_be64(ino); | |
353 | ino++; | |
ce748eaa ES |
354 | uuid_copy(&free->di_uuid, |
355 | &mp->m_sb.sb_meta_uuid); | |
93848a99 | 356 | xfs_dinode_calc_crc(mp, free); |
28c8e41a | 357 | } else if (tp) { |
93848a99 CH |
358 | /* just log the inode core */ |
359 | xfs_trans_log_buf(tp, fbuf, ioffset, | |
360 | ioffset + isize - 1); | |
361 | } | |
362 | } | |
28c8e41a DC |
363 | |
364 | if (tp) { | |
365 | /* | |
366 | * Mark the buffer as an inode allocation buffer so it | |
367 | * sticks in AIL at the point of this allocation | |
368 | * transaction. This ensures the they are on disk before | |
369 | * the tail of the log can be moved past this | |
370 | * transaction (i.e. by preventing relogging from moving | |
371 | * it forward in the log). | |
372 | */ | |
373 | xfs_trans_inode_alloc_buf(tp, fbuf); | |
374 | if (version == 3) { | |
ddf6ad01 DC |
375 | /* |
376 | * Mark the buffer as ordered so that they are | |
377 | * not physically logged in the transaction but | |
378 | * still tracked in the AIL as part of the | |
379 | * transaction and pin the log appropriately. | |
380 | */ | |
381 | xfs_trans_ordered_buf(tp, fbuf); | |
28c8e41a DC |
382 | } |
383 | } else { | |
384 | fbuf->b_flags |= XBF_DONE; | |
385 | xfs_buf_delwri_queue(fbuf, buffer_list); | |
386 | xfs_buf_relse(fbuf); | |
85c0b2ab | 387 | } |
85c0b2ab | 388 | } |
2a30f36d | 389 | return 0; |
85c0b2ab DC |
390 | } |
391 | ||
56d1115c BF |
392 | /* |
393 | * Align startino and allocmask for a recently allocated sparse chunk such that | |
394 | * they are fit for insertion (or merge) into the on-disk inode btrees. | |
395 | * | |
396 | * Background: | |
397 | * | |
398 | * When enabled, sparse inode support increases the inode alignment from cluster | |
399 | * size to inode chunk size. This means that the minimum range between two | |
400 | * non-adjacent inode records in the inobt is large enough for a full inode | |
401 | * record. This allows for cluster sized, cluster aligned block allocation | |
402 | * without need to worry about whether the resulting inode record overlaps with | |
403 | * another record in the tree. Without this basic rule, we would have to deal | |
404 | * with the consequences of overlap by potentially undoing recent allocations in | |
405 | * the inode allocation codepath. | |
406 | * | |
407 | * Because of this alignment rule (which is enforced on mount), there are two | |
408 | * inobt possibilities for newly allocated sparse chunks. One is that the | |
409 | * aligned inode record for the chunk covers a range of inodes not already | |
410 | * covered in the inobt (i.e., it is safe to insert a new sparse record). The | |
411 | * other is that a record already exists at the aligned startino that considers | |
412 | * the newly allocated range as sparse. In the latter case, record content is | |
413 | * merged in hope that sparse inode chunks fill to full chunks over time. | |
414 | */ | |
415 | STATIC void | |
416 | xfs_align_sparse_ino( | |
417 | struct xfs_mount *mp, | |
418 | xfs_agino_t *startino, | |
419 | uint16_t *allocmask) | |
420 | { | |
421 | xfs_agblock_t agbno; | |
422 | xfs_agblock_t mod; | |
423 | int offset; | |
424 | ||
425 | agbno = XFS_AGINO_TO_AGBNO(mp, *startino); | |
426 | mod = agbno % mp->m_sb.sb_inoalignmt; | |
427 | if (!mod) | |
428 | return; | |
429 | ||
430 | /* calculate the inode offset and align startino */ | |
431 | offset = mod << mp->m_sb.sb_inopblog; | |
432 | *startino -= offset; | |
433 | ||
434 | /* | |
435 | * Since startino has been aligned down, left shift allocmask such that | |
436 | * it continues to represent the same physical inodes relative to the | |
437 | * new startino. | |
438 | */ | |
439 | *allocmask <<= offset / XFS_INODES_PER_HOLEMASK_BIT; | |
440 | } | |
441 | ||
442 | /* | |
443 | * Determine whether the source inode record can merge into the target. Both | |
444 | * records must be sparse, the inode ranges must match and there must be no | |
445 | * allocation overlap between the records. | |
446 | */ | |
447 | STATIC bool | |
448 | __xfs_inobt_can_merge( | |
449 | struct xfs_inobt_rec_incore *trec, /* tgt record */ | |
450 | struct xfs_inobt_rec_incore *srec) /* src record */ | |
451 | { | |
452 | uint64_t talloc; | |
453 | uint64_t salloc; | |
454 | ||
455 | /* records must cover the same inode range */ | |
456 | if (trec->ir_startino != srec->ir_startino) | |
457 | return false; | |
458 | ||
459 | /* both records must be sparse */ | |
460 | if (!xfs_inobt_issparse(trec->ir_holemask) || | |
461 | !xfs_inobt_issparse(srec->ir_holemask)) | |
462 | return false; | |
463 | ||
464 | /* both records must track some inodes */ | |
465 | if (!trec->ir_count || !srec->ir_count) | |
466 | return false; | |
467 | ||
468 | /* can't exceed capacity of a full record */ | |
469 | if (trec->ir_count + srec->ir_count > XFS_INODES_PER_CHUNK) | |
470 | return false; | |
471 | ||
472 | /* verify there is no allocation overlap */ | |
473 | talloc = xfs_inobt_irec_to_allocmask(trec); | |
474 | salloc = xfs_inobt_irec_to_allocmask(srec); | |
475 | if (talloc & salloc) | |
476 | return false; | |
477 | ||
478 | return true; | |
479 | } | |
480 | ||
481 | /* | |
482 | * Merge the source inode record into the target. The caller must call | |
483 | * __xfs_inobt_can_merge() to ensure the merge is valid. | |
484 | */ | |
485 | STATIC void | |
486 | __xfs_inobt_rec_merge( | |
487 | struct xfs_inobt_rec_incore *trec, /* target */ | |
488 | struct xfs_inobt_rec_incore *srec) /* src */ | |
489 | { | |
490 | ASSERT(trec->ir_startino == srec->ir_startino); | |
491 | ||
492 | /* combine the counts */ | |
493 | trec->ir_count += srec->ir_count; | |
494 | trec->ir_freecount += srec->ir_freecount; | |
495 | ||
496 | /* | |
497 | * Merge the holemask and free mask. For both fields, 0 bits refer to | |
498 | * allocated inodes. We combine the allocated ranges with bitwise AND. | |
499 | */ | |
500 | trec->ir_holemask &= srec->ir_holemask; | |
501 | trec->ir_free &= srec->ir_free; | |
502 | } | |
503 | ||
504 | /* | |
505 | * Insert a new sparse inode chunk into the associated inode btree. The inode | |
506 | * record for the sparse chunk is pre-aligned to a startino that should match | |
507 | * any pre-existing sparse inode record in the tree. This allows sparse chunks | |
508 | * to fill over time. | |
509 | * | |
510 | * This function supports two modes of handling preexisting records depending on | |
511 | * the merge flag. If merge is true, the provided record is merged with the | |
512 | * existing record and updated in place. The merged record is returned in nrec. | |
513 | * If merge is false, an existing record is replaced with the provided record. | |
514 | * If no preexisting record exists, the provided record is always inserted. | |
515 | * | |
516 | * It is considered corruption if a merge is requested and not possible. Given | |
517 | * the sparse inode alignment constraints, this should never happen. | |
518 | */ | |
519 | STATIC int | |
520 | xfs_inobt_insert_sprec( | |
521 | struct xfs_mount *mp, | |
522 | struct xfs_trans *tp, | |
523 | struct xfs_buf *agbp, | |
524 | int btnum, | |
525 | struct xfs_inobt_rec_incore *nrec, /* in/out: new/merged rec. */ | |
526 | bool merge) /* merge or replace */ | |
527 | { | |
528 | struct xfs_btree_cur *cur; | |
529 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
530 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
531 | int error; | |
532 | int i; | |
533 | struct xfs_inobt_rec_incore rec; | |
534 | ||
535 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum); | |
536 | ||
537 | /* the new record is pre-aligned so we know where to look */ | |
538 | error = xfs_inobt_lookup(cur, nrec->ir_startino, XFS_LOOKUP_EQ, &i); | |
539 | if (error) | |
540 | goto error; | |
541 | /* if nothing there, insert a new record and return */ | |
542 | if (i == 0) { | |
543 | error = xfs_inobt_insert_rec(cur, nrec->ir_holemask, | |
544 | nrec->ir_count, nrec->ir_freecount, | |
545 | nrec->ir_free, &i); | |
546 | if (error) | |
547 | goto error; | |
548 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error); | |
549 | ||
550 | goto out; | |
551 | } | |
552 | ||
553 | /* | |
554 | * A record exists at this startino. Merge or replace the record | |
555 | * depending on what we've been asked to do. | |
556 | */ | |
557 | if (merge) { | |
558 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
559 | if (error) | |
560 | goto error; | |
561 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error); | |
562 | XFS_WANT_CORRUPTED_GOTO(mp, | |
563 | rec.ir_startino == nrec->ir_startino, | |
564 | error); | |
565 | ||
566 | /* | |
567 | * This should never fail. If we have coexisting records that | |
568 | * cannot merge, something is seriously wrong. | |
569 | */ | |
570 | XFS_WANT_CORRUPTED_GOTO(mp, __xfs_inobt_can_merge(nrec, &rec), | |
571 | error); | |
572 | ||
573 | trace_xfs_irec_merge_pre(mp, agno, rec.ir_startino, | |
574 | rec.ir_holemask, nrec->ir_startino, | |
575 | nrec->ir_holemask); | |
576 | ||
577 | /* merge to nrec to output the updated record */ | |
578 | __xfs_inobt_rec_merge(nrec, &rec); | |
579 | ||
580 | trace_xfs_irec_merge_post(mp, agno, nrec->ir_startino, | |
581 | nrec->ir_holemask); | |
582 | ||
583 | error = xfs_inobt_rec_check_count(mp, nrec); | |
584 | if (error) | |
585 | goto error; | |
586 | } | |
587 | ||
588 | error = xfs_inobt_update(cur, nrec); | |
589 | if (error) | |
590 | goto error; | |
591 | ||
592 | out: | |
593 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
594 | return 0; | |
595 | error: | |
596 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
597 | return error; | |
598 | } | |
599 | ||
1da177e4 LT |
600 | /* |
601 | * Allocate new inodes in the allocation group specified by agbp. | |
602 | * Return 0 for success, else error code. | |
603 | */ | |
604 | STATIC int /* error code or 0 */ | |
605 | xfs_ialloc_ag_alloc( | |
606 | xfs_trans_t *tp, /* transaction pointer */ | |
607 | xfs_buf_t *agbp, /* alloc group buffer */ | |
608 | int *alloc) | |
609 | { | |
610 | xfs_agi_t *agi; /* allocation group header */ | |
611 | xfs_alloc_arg_t args; /* allocation argument structure */ | |
92821e2b | 612 | xfs_agnumber_t agno; |
1da177e4 | 613 | int error; |
1da177e4 LT |
614 | xfs_agino_t newino; /* new first inode's number */ |
615 | xfs_agino_t newlen; /* new number of inodes */ | |
3ccb8b5f | 616 | int isaligned = 0; /* inode allocation at stripe unit */ |
1da177e4 | 617 | /* boundary */ |
56d1115c BF |
618 | uint16_t allocmask = (uint16_t) -1; /* init. to full chunk */ |
619 | struct xfs_inobt_rec_incore rec; | |
44b56e0a | 620 | struct xfs_perag *pag; |
1cdadee1 BF |
621 | int do_sparse = 0; |
622 | ||
a0041684 | 623 | memset(&args, 0, sizeof(args)); |
1da177e4 LT |
624 | args.tp = tp; |
625 | args.mp = tp->t_mountp; | |
1cdadee1 | 626 | args.fsbno = NULLFSBLOCK; |
340785cc | 627 | xfs_rmap_ag_owner(&args.oinfo, XFS_RMAP_OWN_INODES); |
1da177e4 | 628 | |
46fc58da BF |
629 | #ifdef DEBUG |
630 | /* randomly do sparse inode allocations */ | |
631 | if (xfs_sb_version_hassparseinodes(&tp->t_mountp->m_sb) && | |
632 | args.mp->m_ialloc_min_blks < args.mp->m_ialloc_blks) | |
633 | do_sparse = prandom_u32() & 1; | |
634 | #endif | |
635 | ||
1da177e4 LT |
636 | /* |
637 | * Locking will ensure that we don't have two callers in here | |
638 | * at one time. | |
639 | */ | |
71783438 | 640 | newlen = args.mp->m_ialloc_inos; |
1da177e4 | 641 | if (args.mp->m_maxicount && |
74f9ce1c | 642 | percpu_counter_read_positive(&args.mp->m_icount) + newlen > |
501ab323 | 643 | args.mp->m_maxicount) |
2451337d | 644 | return -ENOSPC; |
126cd105 | 645 | args.minlen = args.maxlen = args.mp->m_ialloc_blks; |
1da177e4 | 646 | /* |
3ccb8b5f GO |
647 | * First try to allocate inodes contiguous with the last-allocated |
648 | * chunk of inodes. If the filesystem is striped, this will fill | |
649 | * an entire stripe unit with inodes. | |
28c8e41a | 650 | */ |
1da177e4 | 651 | agi = XFS_BUF_TO_AGI(agbp); |
3ccb8b5f | 652 | newino = be32_to_cpu(agi->agi_newino); |
85c0b2ab | 653 | agno = be32_to_cpu(agi->agi_seqno); |
019ff2d5 | 654 | args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) + |
126cd105 | 655 | args.mp->m_ialloc_blks; |
1cdadee1 BF |
656 | if (do_sparse) |
657 | goto sparse_alloc; | |
019ff2d5 NS |
658 | if (likely(newino != NULLAGINO && |
659 | (args.agbno < be32_to_cpu(agi->agi_length)))) { | |
85c0b2ab | 660 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
3ccb8b5f | 661 | args.type = XFS_ALLOCTYPE_THIS_BNO; |
3ccb8b5f | 662 | args.prod = 1; |
75de2a91 | 663 | |
3ccb8b5f | 664 | /* |
75de2a91 DC |
665 | * We need to take into account alignment here to ensure that |
666 | * we don't modify the free list if we fail to have an exact | |
667 | * block. If we don't have an exact match, and every oher | |
668 | * attempt allocation attempt fails, we'll end up cancelling | |
669 | * a dirty transaction and shutting down. | |
670 | * | |
671 | * For an exact allocation, alignment must be 1, | |
672 | * however we need to take cluster alignment into account when | |
673 | * fixing up the freelist. Use the minalignslop field to | |
674 | * indicate that extra blocks might be required for alignment, | |
675 | * but not to use them in the actual exact allocation. | |
3ccb8b5f | 676 | */ |
75de2a91 | 677 | args.alignment = 1; |
7a1df156 | 678 | args.minalignslop = xfs_ialloc_cluster_alignment(args.mp) - 1; |
75de2a91 DC |
679 | |
680 | /* Allow space for the inode btree to split. */ | |
0d87e656 | 681 | args.minleft = args.mp->m_in_maxlevels - 1; |
3ccb8b5f GO |
682 | if ((error = xfs_alloc_vextent(&args))) |
683 | return error; | |
e480a723 BF |
684 | |
685 | /* | |
686 | * This request might have dirtied the transaction if the AG can | |
687 | * satisfy the request, but the exact block was not available. | |
688 | * If the allocation did fail, subsequent requests will relax | |
689 | * the exact agbno requirement and increase the alignment | |
690 | * instead. It is critical that the total size of the request | |
691 | * (len + alignment + slop) does not increase from this point | |
692 | * on, so reset minalignslop to ensure it is not included in | |
693 | * subsequent requests. | |
694 | */ | |
695 | args.minalignslop = 0; | |
1cdadee1 | 696 | } |
1da177e4 | 697 | |
3ccb8b5f GO |
698 | if (unlikely(args.fsbno == NULLFSBLOCK)) { |
699 | /* | |
700 | * Set the alignment for the allocation. | |
701 | * If stripe alignment is turned on then align at stripe unit | |
702 | * boundary. | |
019ff2d5 NS |
703 | * If the cluster size is smaller than a filesystem block |
704 | * then we're doing I/O for inodes in filesystem block size | |
3ccb8b5f GO |
705 | * pieces, so don't need alignment anyway. |
706 | */ | |
707 | isaligned = 0; | |
708 | if (args.mp->m_sinoalign) { | |
709 | ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN)); | |
710 | args.alignment = args.mp->m_dalign; | |
711 | isaligned = 1; | |
75de2a91 | 712 | } else |
7a1df156 | 713 | args.alignment = xfs_ialloc_cluster_alignment(args.mp); |
3ccb8b5f GO |
714 | /* |
715 | * Need to figure out where to allocate the inode blocks. | |
716 | * Ideally they should be spaced out through the a.g. | |
717 | * For now, just allocate blocks up front. | |
718 | */ | |
719 | args.agbno = be32_to_cpu(agi->agi_root); | |
85c0b2ab | 720 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
3ccb8b5f GO |
721 | /* |
722 | * Allocate a fixed-size extent of inodes. | |
723 | */ | |
724 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
3ccb8b5f GO |
725 | args.prod = 1; |
726 | /* | |
727 | * Allow space for the inode btree to split. | |
728 | */ | |
0d87e656 | 729 | args.minleft = args.mp->m_in_maxlevels - 1; |
3ccb8b5f GO |
730 | if ((error = xfs_alloc_vextent(&args))) |
731 | return error; | |
732 | } | |
019ff2d5 | 733 | |
1da177e4 LT |
734 | /* |
735 | * If stripe alignment is turned on, then try again with cluster | |
736 | * alignment. | |
737 | */ | |
738 | if (isaligned && args.fsbno == NULLFSBLOCK) { | |
739 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
16259e7d | 740 | args.agbno = be32_to_cpu(agi->agi_root); |
85c0b2ab | 741 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
7a1df156 | 742 | args.alignment = xfs_ialloc_cluster_alignment(args.mp); |
1da177e4 LT |
743 | if ((error = xfs_alloc_vextent(&args))) |
744 | return error; | |
745 | } | |
746 | ||
56d1115c BF |
747 | /* |
748 | * Finally, try a sparse allocation if the filesystem supports it and | |
749 | * the sparse allocation length is smaller than a full chunk. | |
750 | */ | |
751 | if (xfs_sb_version_hassparseinodes(&args.mp->m_sb) && | |
752 | args.mp->m_ialloc_min_blks < args.mp->m_ialloc_blks && | |
753 | args.fsbno == NULLFSBLOCK) { | |
1cdadee1 | 754 | sparse_alloc: |
56d1115c BF |
755 | args.type = XFS_ALLOCTYPE_NEAR_BNO; |
756 | args.agbno = be32_to_cpu(agi->agi_root); | |
757 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); | |
758 | args.alignment = args.mp->m_sb.sb_spino_align; | |
759 | args.prod = 1; | |
760 | ||
761 | args.minlen = args.mp->m_ialloc_min_blks; | |
762 | args.maxlen = args.minlen; | |
763 | ||
764 | /* | |
765 | * The inode record will be aligned to full chunk size. We must | |
766 | * prevent sparse allocation from AG boundaries that result in | |
767 | * invalid inode records, such as records that start at agbno 0 | |
768 | * or extend beyond the AG. | |
769 | * | |
770 | * Set min agbno to the first aligned, non-zero agbno and max to | |
771 | * the last aligned agbno that is at least one full chunk from | |
772 | * the end of the AG. | |
773 | */ | |
774 | args.min_agbno = args.mp->m_sb.sb_inoalignmt; | |
775 | args.max_agbno = round_down(args.mp->m_sb.sb_agblocks, | |
776 | args.mp->m_sb.sb_inoalignmt) - | |
777 | args.mp->m_ialloc_blks; | |
778 | ||
779 | error = xfs_alloc_vextent(&args); | |
780 | if (error) | |
781 | return error; | |
782 | ||
783 | newlen = args.len << args.mp->m_sb.sb_inopblog; | |
46fc58da | 784 | ASSERT(newlen <= XFS_INODES_PER_CHUNK); |
56d1115c BF |
785 | allocmask = (1 << (newlen / XFS_INODES_PER_HOLEMASK_BIT)) - 1; |
786 | } | |
787 | ||
1da177e4 LT |
788 | if (args.fsbno == NULLFSBLOCK) { |
789 | *alloc = 0; | |
790 | return 0; | |
791 | } | |
792 | ASSERT(args.len == args.minlen); | |
1da177e4 | 793 | |
359346a9 | 794 | /* |
85c0b2ab DC |
795 | * Stamp and write the inode buffers. |
796 | * | |
359346a9 DC |
797 | * Seed the new inode cluster with a random generation number. This |
798 | * prevents short-term reuse of generation numbers if a chunk is | |
799 | * freed and then immediately reallocated. We use random numbers | |
800 | * rather than a linear progression to prevent the next generation | |
801 | * number from being easily guessable. | |
802 | */ | |
463958af BF |
803 | error = xfs_ialloc_inode_init(args.mp, tp, NULL, newlen, agno, |
804 | args.agbno, args.len, prandom_u32()); | |
d42f08f6 | 805 | |
2a30f36d CS |
806 | if (error) |
807 | return error; | |
85c0b2ab DC |
808 | /* |
809 | * Convert the results. | |
810 | */ | |
811 | newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0); | |
56d1115c BF |
812 | |
813 | if (xfs_inobt_issparse(~allocmask)) { | |
814 | /* | |
815 | * We've allocated a sparse chunk. Align the startino and mask. | |
816 | */ | |
817 | xfs_align_sparse_ino(args.mp, &newino, &allocmask); | |
818 | ||
819 | rec.ir_startino = newino; | |
820 | rec.ir_holemask = ~allocmask; | |
821 | rec.ir_count = newlen; | |
822 | rec.ir_freecount = newlen; | |
823 | rec.ir_free = XFS_INOBT_ALL_FREE; | |
824 | ||
825 | /* | |
826 | * Insert the sparse record into the inobt and allow for a merge | |
827 | * if necessary. If a merge does occur, rec is updated to the | |
828 | * merged record. | |
829 | */ | |
830 | error = xfs_inobt_insert_sprec(args.mp, tp, agbp, XFS_BTNUM_INO, | |
831 | &rec, true); | |
832 | if (error == -EFSCORRUPTED) { | |
833 | xfs_alert(args.mp, | |
834 | "invalid sparse inode record: ino 0x%llx holemask 0x%x count %u", | |
835 | XFS_AGINO_TO_INO(args.mp, agno, | |
836 | rec.ir_startino), | |
837 | rec.ir_holemask, rec.ir_count); | |
838 | xfs_force_shutdown(args.mp, SHUTDOWN_CORRUPT_INCORE); | |
839 | } | |
840 | if (error) | |
841 | return error; | |
842 | ||
843 | /* | |
844 | * We can't merge the part we've just allocated as for the inobt | |
845 | * due to finobt semantics. The original record may or may not | |
846 | * exist independent of whether physical inodes exist in this | |
847 | * sparse chunk. | |
848 | * | |
849 | * We must update the finobt record based on the inobt record. | |
850 | * rec contains the fully merged and up to date inobt record | |
851 | * from the previous call. Set merge false to replace any | |
852 | * existing record with this one. | |
853 | */ | |
854 | if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) { | |
855 | error = xfs_inobt_insert_sprec(args.mp, tp, agbp, | |
856 | XFS_BTNUM_FINO, &rec, | |
857 | false); | |
858 | if (error) | |
859 | return error; | |
860 | } | |
861 | } else { | |
862 | /* full chunk - insert new records to both btrees */ | |
863 | error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen, | |
864 | XFS_BTNUM_INO); | |
865 | if (error) | |
866 | return error; | |
867 | ||
868 | if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) { | |
869 | error = xfs_inobt_insert(args.mp, tp, agbp, newino, | |
870 | newlen, XFS_BTNUM_FINO); | |
871 | if (error) | |
872 | return error; | |
873 | } | |
874 | } | |
875 | ||
876 | /* | |
877 | * Update AGI counts and newino. | |
878 | */ | |
413d57c9 MS |
879 | be32_add_cpu(&agi->agi_count, newlen); |
880 | be32_add_cpu(&agi->agi_freecount, newlen); | |
44b56e0a DC |
881 | pag = xfs_perag_get(args.mp, agno); |
882 | pag->pagi_freecount += newlen; | |
883 | xfs_perag_put(pag); | |
16259e7d | 884 | agi->agi_newino = cpu_to_be32(newino); |
85c0b2ab | 885 | |
1da177e4 LT |
886 | /* |
887 | * Log allocation group header fields | |
888 | */ | |
889 | xfs_ialloc_log_agi(tp, agbp, | |
890 | XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO); | |
891 | /* | |
892 | * Modify/log superblock values for inode count and inode free count. | |
893 | */ | |
894 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen); | |
895 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen); | |
896 | *alloc = 1; | |
897 | return 0; | |
898 | } | |
899 | ||
b8f82a4a | 900 | STATIC xfs_agnumber_t |
1da177e4 LT |
901 | xfs_ialloc_next_ag( |
902 | xfs_mount_t *mp) | |
903 | { | |
904 | xfs_agnumber_t agno; | |
905 | ||
906 | spin_lock(&mp->m_agirotor_lock); | |
907 | agno = mp->m_agirotor; | |
8aea3ff4 | 908 | if (++mp->m_agirotor >= mp->m_maxagi) |
1da177e4 LT |
909 | mp->m_agirotor = 0; |
910 | spin_unlock(&mp->m_agirotor_lock); | |
911 | ||
912 | return agno; | |
913 | } | |
914 | ||
915 | /* | |
916 | * Select an allocation group to look for a free inode in, based on the parent | |
2f21ff1c | 917 | * inode and the mode. Return the allocation group buffer. |
1da177e4 | 918 | */ |
55d6af64 | 919 | STATIC xfs_agnumber_t |
1da177e4 LT |
920 | xfs_ialloc_ag_select( |
921 | xfs_trans_t *tp, /* transaction pointer */ | |
922 | xfs_ino_t parent, /* parent directory inode number */ | |
f59cf5c2 | 923 | umode_t mode) /* bits set to indicate file type */ |
1da177e4 | 924 | { |
1da177e4 LT |
925 | xfs_agnumber_t agcount; /* number of ag's in the filesystem */ |
926 | xfs_agnumber_t agno; /* current ag number */ | |
927 | int flags; /* alloc buffer locking flags */ | |
928 | xfs_extlen_t ineed; /* blocks needed for inode allocation */ | |
929 | xfs_extlen_t longest = 0; /* longest extent available */ | |
930 | xfs_mount_t *mp; /* mount point structure */ | |
931 | int needspace; /* file mode implies space allocated */ | |
932 | xfs_perag_t *pag; /* per allocation group data */ | |
933 | xfs_agnumber_t pagno; /* parent (starting) ag number */ | |
55d6af64 | 934 | int error; |
1da177e4 LT |
935 | |
936 | /* | |
937 | * Files of these types need at least one block if length > 0 | |
938 | * (and they won't fit in the inode, but that's hard to figure out). | |
939 | */ | |
940 | needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode); | |
941 | mp = tp->t_mountp; | |
942 | agcount = mp->m_maxagi; | |
943 | if (S_ISDIR(mode)) | |
944 | pagno = xfs_ialloc_next_ag(mp); | |
945 | else { | |
946 | pagno = XFS_INO_TO_AGNO(mp, parent); | |
947 | if (pagno >= agcount) | |
948 | pagno = 0; | |
949 | } | |
55d6af64 | 950 | |
1da177e4 | 951 | ASSERT(pagno < agcount); |
55d6af64 | 952 | |
1da177e4 LT |
953 | /* |
954 | * Loop through allocation groups, looking for one with a little | |
955 | * free space in it. Note we don't look for free inodes, exactly. | |
956 | * Instead, we include whether there is a need to allocate inodes | |
957 | * to mean that blocks must be allocated for them, | |
958 | * if none are currently free. | |
959 | */ | |
960 | agno = pagno; | |
961 | flags = XFS_ALLOC_FLAG_TRYLOCK; | |
1da177e4 | 962 | for (;;) { |
44b56e0a | 963 | pag = xfs_perag_get(mp, agno); |
55d6af64 CH |
964 | if (!pag->pagi_inodeok) { |
965 | xfs_ialloc_next_ag(mp); | |
966 | goto nextag; | |
967 | } | |
968 | ||
1da177e4 | 969 | if (!pag->pagi_init) { |
55d6af64 CH |
970 | error = xfs_ialloc_pagi_init(mp, tp, agno); |
971 | if (error) | |
1da177e4 | 972 | goto nextag; |
55d6af64 | 973 | } |
1da177e4 | 974 | |
55d6af64 CH |
975 | if (pag->pagi_freecount) { |
976 | xfs_perag_put(pag); | |
977 | return agno; | |
1da177e4 LT |
978 | } |
979 | ||
55d6af64 CH |
980 | if (!pag->pagf_init) { |
981 | error = xfs_alloc_pagf_init(mp, tp, agno, flags); | |
982 | if (error) | |
1da177e4 | 983 | goto nextag; |
1da177e4 | 984 | } |
55d6af64 CH |
985 | |
986 | /* | |
7a1df156 DC |
987 | * Check that there is enough free space for the file plus a |
988 | * chunk of inodes if we need to allocate some. If this is the | |
989 | * first pass across the AGs, take into account the potential | |
990 | * space needed for alignment of inode chunks when checking the | |
991 | * longest contiguous free space in the AG - this prevents us | |
992 | * from getting ENOSPC because we have free space larger than | |
993 | * m_ialloc_blks but alignment constraints prevent us from using | |
994 | * it. | |
995 | * | |
996 | * If we can't find an AG with space for full alignment slack to | |
997 | * be taken into account, we must be near ENOSPC in all AGs. | |
998 | * Hence we don't include alignment for the second pass and so | |
999 | * if we fail allocation due to alignment issues then it is most | |
1000 | * likely a real ENOSPC condition. | |
55d6af64 | 1001 | */ |
066a1884 | 1002 | ineed = mp->m_ialloc_min_blks; |
7a1df156 DC |
1003 | if (flags && ineed > 1) |
1004 | ineed += xfs_ialloc_cluster_alignment(mp); | |
55d6af64 CH |
1005 | longest = pag->pagf_longest; |
1006 | if (!longest) | |
1007 | longest = pag->pagf_flcount > 0; | |
1008 | ||
1009 | if (pag->pagf_freeblks >= needspace + ineed && | |
1010 | longest >= ineed) { | |
1011 | xfs_perag_put(pag); | |
1012 | return agno; | |
1da177e4 | 1013 | } |
1da177e4 | 1014 | nextag: |
44b56e0a | 1015 | xfs_perag_put(pag); |
1da177e4 LT |
1016 | /* |
1017 | * No point in iterating over the rest, if we're shutting | |
1018 | * down. | |
1019 | */ | |
1c1c6ebc | 1020 | if (XFS_FORCED_SHUTDOWN(mp)) |
55d6af64 | 1021 | return NULLAGNUMBER; |
1da177e4 LT |
1022 | agno++; |
1023 | if (agno >= agcount) | |
1024 | agno = 0; | |
1025 | if (agno == pagno) { | |
1c1c6ebc | 1026 | if (flags == 0) |
55d6af64 | 1027 | return NULLAGNUMBER; |
1da177e4 LT |
1028 | flags = 0; |
1029 | } | |
1030 | } | |
1031 | } | |
1032 | ||
4254b0bb CH |
1033 | /* |
1034 | * Try to retrieve the next record to the left/right from the current one. | |
1035 | */ | |
1036 | STATIC int | |
1037 | xfs_ialloc_next_rec( | |
1038 | struct xfs_btree_cur *cur, | |
1039 | xfs_inobt_rec_incore_t *rec, | |
1040 | int *done, | |
1041 | int left) | |
1042 | { | |
1043 | int error; | |
1044 | int i; | |
1045 | ||
1046 | if (left) | |
1047 | error = xfs_btree_decrement(cur, 0, &i); | |
1048 | else | |
1049 | error = xfs_btree_increment(cur, 0, &i); | |
1050 | ||
1051 | if (error) | |
1052 | return error; | |
1053 | *done = !i; | |
1054 | if (i) { | |
1055 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1056 | if (error) | |
1057 | return error; | |
5fb5aeee | 1058 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
4254b0bb CH |
1059 | } |
1060 | ||
1061 | return 0; | |
1062 | } | |
1063 | ||
bd169565 DC |
1064 | STATIC int |
1065 | xfs_ialloc_get_rec( | |
1066 | struct xfs_btree_cur *cur, | |
1067 | xfs_agino_t agino, | |
1068 | xfs_inobt_rec_incore_t *rec, | |
43df2ee6 | 1069 | int *done) |
bd169565 DC |
1070 | { |
1071 | int error; | |
1072 | int i; | |
1073 | ||
1074 | error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i); | |
1075 | if (error) | |
1076 | return error; | |
1077 | *done = !i; | |
1078 | if (i) { | |
1079 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1080 | if (error) | |
1081 | return error; | |
5fb5aeee | 1082 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
bd169565 DC |
1083 | } |
1084 | ||
1085 | return 0; | |
1086 | } | |
0b48db80 | 1087 | |
d4cc540b | 1088 | /* |
26dd5217 BF |
1089 | * Return the offset of the first free inode in the record. If the inode chunk |
1090 | * is sparsely allocated, we convert the record holemask to inode granularity | |
1091 | * and mask off the unallocated regions from the inode free mask. | |
d4cc540b BF |
1092 | */ |
1093 | STATIC int | |
1094 | xfs_inobt_first_free_inode( | |
1095 | struct xfs_inobt_rec_incore *rec) | |
1096 | { | |
26dd5217 BF |
1097 | xfs_inofree_t realfree; |
1098 | ||
1099 | /* if there are no holes, return the first available offset */ | |
1100 | if (!xfs_inobt_issparse(rec->ir_holemask)) | |
1101 | return xfs_lowbit64(rec->ir_free); | |
1102 | ||
1103 | realfree = xfs_inobt_irec_to_allocmask(rec); | |
1104 | realfree &= rec->ir_free; | |
1105 | ||
1106 | return xfs_lowbit64(realfree); | |
d4cc540b BF |
1107 | } |
1108 | ||
1da177e4 | 1109 | /* |
6dd8638e | 1110 | * Allocate an inode using the inobt-only algorithm. |
1da177e4 | 1111 | */ |
f2ecc5e4 | 1112 | STATIC int |
6dd8638e | 1113 | xfs_dialloc_ag_inobt( |
f2ecc5e4 CH |
1114 | struct xfs_trans *tp, |
1115 | struct xfs_buf *agbp, | |
1116 | xfs_ino_t parent, | |
1117 | xfs_ino_t *inop) | |
1da177e4 | 1118 | { |
f2ecc5e4 CH |
1119 | struct xfs_mount *mp = tp->t_mountp; |
1120 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
1121 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
1122 | xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent); | |
1123 | xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent); | |
1124 | struct xfs_perag *pag; | |
1125 | struct xfs_btree_cur *cur, *tcur; | |
1126 | struct xfs_inobt_rec_incore rec, trec; | |
1127 | xfs_ino_t ino; | |
1128 | int error; | |
1129 | int offset; | |
1130 | int i, j; | |
2d32311c | 1131 | int searchdistance = 10; |
1da177e4 | 1132 | |
44b56e0a | 1133 | pag = xfs_perag_get(mp, agno); |
bd169565 | 1134 | |
4bb61069 CH |
1135 | ASSERT(pag->pagi_init); |
1136 | ASSERT(pag->pagi_inodeok); | |
1137 | ASSERT(pag->pagi_freecount > 0); | |
1138 | ||
bd169565 | 1139 | restart_pagno: |
57bd3dbe | 1140 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
1da177e4 LT |
1141 | /* |
1142 | * If pagino is 0 (this is the root inode allocation) use newino. | |
1143 | * This must work because we've just allocated some. | |
1144 | */ | |
1145 | if (!pagino) | |
16259e7d | 1146 | pagino = be32_to_cpu(agi->agi_newino); |
1da177e4 | 1147 | |
0b48db80 DC |
1148 | error = xfs_check_agi_freecount(cur, agi); |
1149 | if (error) | |
1150 | goto error0; | |
1da177e4 | 1151 | |
1da177e4 | 1152 | /* |
4254b0bb | 1153 | * If in the same AG as the parent, try to get near the parent. |
1da177e4 LT |
1154 | */ |
1155 | if (pagno == agno) { | |
4254b0bb CH |
1156 | int doneleft; /* done, to the left */ |
1157 | int doneright; /* done, to the right */ | |
1158 | ||
21875505 | 1159 | error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i); |
4254b0bb | 1160 | if (error) |
1da177e4 | 1161 | goto error0; |
c29aad41 | 1162 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
4254b0bb CH |
1163 | |
1164 | error = xfs_inobt_get_rec(cur, &rec, &j); | |
1165 | if (error) | |
1166 | goto error0; | |
c29aad41 | 1167 | XFS_WANT_CORRUPTED_GOTO(mp, j == 1, error0); |
4254b0bb CH |
1168 | |
1169 | if (rec.ir_freecount > 0) { | |
1da177e4 LT |
1170 | /* |
1171 | * Found a free inode in the same chunk | |
4254b0bb | 1172 | * as the parent, done. |
1da177e4 | 1173 | */ |
4254b0bb | 1174 | goto alloc_inode; |
1da177e4 | 1175 | } |
4254b0bb CH |
1176 | |
1177 | ||
1da177e4 | 1178 | /* |
4254b0bb | 1179 | * In the same AG as parent, but parent's chunk is full. |
1da177e4 | 1180 | */ |
1da177e4 | 1181 | |
4254b0bb CH |
1182 | /* duplicate the cursor, search left & right simultaneously */ |
1183 | error = xfs_btree_dup_cursor(cur, &tcur); | |
1184 | if (error) | |
1185 | goto error0; | |
1186 | ||
bd169565 DC |
1187 | /* |
1188 | * Skip to last blocks looked up if same parent inode. | |
1189 | */ | |
1190 | if (pagino != NULLAGINO && | |
1191 | pag->pagl_pagino == pagino && | |
1192 | pag->pagl_leftrec != NULLAGINO && | |
1193 | pag->pagl_rightrec != NULLAGINO) { | |
1194 | error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec, | |
43df2ee6 | 1195 | &trec, &doneleft); |
bd169565 DC |
1196 | if (error) |
1197 | goto error1; | |
4254b0bb | 1198 | |
bd169565 | 1199 | error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec, |
43df2ee6 | 1200 | &rec, &doneright); |
bd169565 DC |
1201 | if (error) |
1202 | goto error1; | |
1203 | } else { | |
1204 | /* search left with tcur, back up 1 record */ | |
1205 | error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1); | |
1206 | if (error) | |
1207 | goto error1; | |
1208 | ||
1209 | /* search right with cur, go forward 1 record. */ | |
1210 | error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0); | |
1211 | if (error) | |
1212 | goto error1; | |
1213 | } | |
4254b0bb CH |
1214 | |
1215 | /* | |
1216 | * Loop until we find an inode chunk with a free inode. | |
1217 | */ | |
2d32311c | 1218 | while (--searchdistance > 0 && (!doneleft || !doneright)) { |
4254b0bb CH |
1219 | int useleft; /* using left inode chunk this time */ |
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; | |
1da177e4 | 1228 | } |
4254b0bb CH |
1229 | |
1230 | /* free inodes to the left? */ | |
1231 | if (useleft && trec.ir_freecount) { | |
4254b0bb CH |
1232 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
1233 | cur = tcur; | |
bd169565 DC |
1234 | |
1235 | pag->pagl_leftrec = trec.ir_startino; | |
1236 | pag->pagl_rightrec = rec.ir_startino; | |
1237 | pag->pagl_pagino = pagino; | |
c44245b3 | 1238 | rec = trec; |
4254b0bb | 1239 | goto alloc_inode; |
1da177e4 | 1240 | } |
1da177e4 | 1241 | |
4254b0bb CH |
1242 | /* free inodes to the right? */ |
1243 | if (!useleft && rec.ir_freecount) { | |
1244 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
bd169565 DC |
1245 | |
1246 | pag->pagl_leftrec = trec.ir_startino; | |
1247 | pag->pagl_rightrec = rec.ir_startino; | |
1248 | pag->pagl_pagino = pagino; | |
4254b0bb | 1249 | goto alloc_inode; |
1da177e4 | 1250 | } |
4254b0bb CH |
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; | |
1da177e4 | 1262 | } |
bd169565 | 1263 | |
2d32311c CM |
1264 | if (searchdistance <= 0) { |
1265 | /* | |
1266 | * Not in range - save last search | |
1267 | * location and allocate a new inode | |
1268 | */ | |
1269 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
1270 | pag->pagl_leftrec = trec.ir_startino; | |
1271 | pag->pagl_rightrec = rec.ir_startino; | |
1272 | pag->pagl_pagino = pagino; | |
1273 | ||
1274 | } else { | |
1275 | /* | |
1276 | * We've reached the end of the btree. because | |
1277 | * we are only searching a small chunk of the | |
1278 | * btree each search, there is obviously free | |
1279 | * inodes closer to the parent inode than we | |
1280 | * are now. restart the search again. | |
1281 | */ | |
1282 | pag->pagl_pagino = NULLAGINO; | |
1283 | pag->pagl_leftrec = NULLAGINO; | |
1284 | pag->pagl_rightrec = NULLAGINO; | |
1285 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
1286 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
1287 | goto restart_pagno; | |
1288 | } | |
1da177e4 | 1289 | } |
4254b0bb | 1290 | |
1da177e4 | 1291 | /* |
4254b0bb | 1292 | * In a different AG from the parent. |
1da177e4 LT |
1293 | * See if the most recently allocated block has any free. |
1294 | */ | |
69ef921b | 1295 | if (agi->agi_newino != cpu_to_be32(NULLAGINO)) { |
21875505 CH |
1296 | error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino), |
1297 | XFS_LOOKUP_EQ, &i); | |
4254b0bb | 1298 | if (error) |
1da177e4 | 1299 | goto error0; |
4254b0bb CH |
1300 | |
1301 | if (i == 1) { | |
1302 | error = xfs_inobt_get_rec(cur, &rec, &j); | |
1303 | if (error) | |
1304 | goto error0; | |
1305 | ||
1306 | if (j == 1 && rec.ir_freecount > 0) { | |
1307 | /* | |
1308 | * The last chunk allocated in the group | |
1309 | * still has a free inode. | |
1310 | */ | |
1311 | goto alloc_inode; | |
1312 | } | |
1da177e4 | 1313 | } |
bd169565 | 1314 | } |
4254b0bb | 1315 | |
bd169565 DC |
1316 | /* |
1317 | * None left in the last group, search the whole AG | |
1318 | */ | |
1319 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); | |
1320 | if (error) | |
1321 | goto error0; | |
c29aad41 | 1322 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
bd169565 DC |
1323 | |
1324 | for (;;) { | |
1325 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
1326 | if (error) | |
1327 | goto error0; | |
c29aad41 | 1328 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
bd169565 DC |
1329 | if (rec.ir_freecount > 0) |
1330 | break; | |
1331 | error = xfs_btree_increment(cur, 0, &i); | |
4254b0bb CH |
1332 | if (error) |
1333 | goto error0; | |
c29aad41 | 1334 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 | 1335 | } |
4254b0bb CH |
1336 | |
1337 | alloc_inode: | |
d4cc540b | 1338 | offset = xfs_inobt_first_free_inode(&rec); |
1da177e4 LT |
1339 | ASSERT(offset >= 0); |
1340 | ASSERT(offset < XFS_INODES_PER_CHUNK); | |
1341 | ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) % | |
1342 | XFS_INODES_PER_CHUNK) == 0); | |
1343 | ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset); | |
0d87e656 | 1344 | rec.ir_free &= ~XFS_INOBT_MASK(offset); |
1da177e4 | 1345 | rec.ir_freecount--; |
afabc24a CH |
1346 | error = xfs_inobt_update(cur, &rec); |
1347 | if (error) | |
1da177e4 | 1348 | goto error0; |
413d57c9 | 1349 | be32_add_cpu(&agi->agi_freecount, -1); |
1da177e4 | 1350 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); |
44b56e0a | 1351 | pag->pagi_freecount--; |
1da177e4 | 1352 | |
0b48db80 DC |
1353 | error = xfs_check_agi_freecount(cur, agi); |
1354 | if (error) | |
1355 | goto error0; | |
1356 | ||
1da177e4 LT |
1357 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
1358 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1); | |
44b56e0a | 1359 | xfs_perag_put(pag); |
1da177e4 LT |
1360 | *inop = ino; |
1361 | return 0; | |
1362 | error1: | |
1363 | xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR); | |
1364 | error0: | |
1365 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
44b56e0a | 1366 | xfs_perag_put(pag); |
1da177e4 LT |
1367 | return error; |
1368 | } | |
1369 | ||
6dd8638e BF |
1370 | /* |
1371 | * Use the free inode btree to allocate an inode based on distance from the | |
1372 | * parent. Note that the provided cursor may be deleted and replaced. | |
1373 | */ | |
1374 | STATIC int | |
1375 | xfs_dialloc_ag_finobt_near( | |
1376 | xfs_agino_t pagino, | |
1377 | struct xfs_btree_cur **ocur, | |
1378 | struct xfs_inobt_rec_incore *rec) | |
1379 | { | |
1380 | struct xfs_btree_cur *lcur = *ocur; /* left search cursor */ | |
1381 | struct xfs_btree_cur *rcur; /* right search cursor */ | |
1382 | struct xfs_inobt_rec_incore rrec; | |
1383 | int error; | |
1384 | int i, j; | |
1385 | ||
1386 | error = xfs_inobt_lookup(lcur, pagino, XFS_LOOKUP_LE, &i); | |
1387 | if (error) | |
1388 | return error; | |
1389 | ||
1390 | if (i == 1) { | |
1391 | error = xfs_inobt_get_rec(lcur, rec, &i); | |
1392 | if (error) | |
1393 | return error; | |
5fb5aeee | 1394 | XFS_WANT_CORRUPTED_RETURN(lcur->bc_mp, i == 1); |
6dd8638e BF |
1395 | |
1396 | /* | |
1397 | * See if we've landed in the parent inode record. The finobt | |
1398 | * only tracks chunks with at least one free inode, so record | |
1399 | * existence is enough. | |
1400 | */ | |
1401 | if (pagino >= rec->ir_startino && | |
1402 | pagino < (rec->ir_startino + XFS_INODES_PER_CHUNK)) | |
1403 | return 0; | |
1404 | } | |
1405 | ||
1406 | error = xfs_btree_dup_cursor(lcur, &rcur); | |
1407 | if (error) | |
1408 | return error; | |
1409 | ||
1410 | error = xfs_inobt_lookup(rcur, pagino, XFS_LOOKUP_GE, &j); | |
1411 | if (error) | |
1412 | goto error_rcur; | |
1413 | if (j == 1) { | |
1414 | error = xfs_inobt_get_rec(rcur, &rrec, &j); | |
1415 | if (error) | |
1416 | goto error_rcur; | |
c29aad41 | 1417 | XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, j == 1, error_rcur); |
6dd8638e BF |
1418 | } |
1419 | ||
c29aad41 | 1420 | XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, i == 1 || j == 1, error_rcur); |
6dd8638e BF |
1421 | if (i == 1 && j == 1) { |
1422 | /* | |
1423 | * Both the left and right records are valid. Choose the closer | |
1424 | * inode chunk to the target. | |
1425 | */ | |
1426 | if ((pagino - rec->ir_startino + XFS_INODES_PER_CHUNK - 1) > | |
1427 | (rrec.ir_startino - pagino)) { | |
1428 | *rec = rrec; | |
1429 | xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR); | |
1430 | *ocur = rcur; | |
1431 | } else { | |
1432 | xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR); | |
1433 | } | |
1434 | } else if (j == 1) { | |
1435 | /* only the right record is valid */ | |
1436 | *rec = rrec; | |
1437 | xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR); | |
1438 | *ocur = rcur; | |
1439 | } else if (i == 1) { | |
1440 | /* only the left record is valid */ | |
1441 | xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR); | |
1442 | } | |
1443 | ||
1444 | return 0; | |
1445 | ||
1446 | error_rcur: | |
1447 | xfs_btree_del_cursor(rcur, XFS_BTREE_ERROR); | |
1448 | return error; | |
1449 | } | |
1450 | ||
1451 | /* | |
1452 | * Use the free inode btree to find a free inode based on a newino hint. If | |
1453 | * the hint is NULL, find the first free inode in the AG. | |
1454 | */ | |
1455 | STATIC int | |
1456 | xfs_dialloc_ag_finobt_newino( | |
1457 | struct xfs_agi *agi, | |
1458 | struct xfs_btree_cur *cur, | |
1459 | struct xfs_inobt_rec_incore *rec) | |
1460 | { | |
1461 | int error; | |
1462 | int i; | |
1463 | ||
1464 | if (agi->agi_newino != cpu_to_be32(NULLAGINO)) { | |
e68ed775 DC |
1465 | error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino), |
1466 | XFS_LOOKUP_EQ, &i); | |
6dd8638e BF |
1467 | if (error) |
1468 | return error; | |
1469 | if (i == 1) { | |
1470 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1471 | if (error) | |
1472 | return error; | |
5fb5aeee | 1473 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
6dd8638e BF |
1474 | return 0; |
1475 | } | |
1476 | } | |
1477 | ||
1478 | /* | |
1479 | * Find the first inode available in the AG. | |
1480 | */ | |
1481 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); | |
1482 | if (error) | |
1483 | return error; | |
5fb5aeee | 1484 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
6dd8638e BF |
1485 | |
1486 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1487 | if (error) | |
1488 | return error; | |
5fb5aeee | 1489 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
6dd8638e BF |
1490 | |
1491 | return 0; | |
1492 | } | |
1493 | ||
1494 | /* | |
1495 | * Update the inobt based on a modification made to the finobt. Also ensure that | |
1496 | * the records from both trees are equivalent post-modification. | |
1497 | */ | |
1498 | STATIC int | |
1499 | xfs_dialloc_ag_update_inobt( | |
1500 | struct xfs_btree_cur *cur, /* inobt cursor */ | |
1501 | struct xfs_inobt_rec_incore *frec, /* finobt record */ | |
1502 | int offset) /* inode offset */ | |
1503 | { | |
1504 | struct xfs_inobt_rec_incore rec; | |
1505 | int error; | |
1506 | int i; | |
1507 | ||
1508 | error = xfs_inobt_lookup(cur, frec->ir_startino, XFS_LOOKUP_EQ, &i); | |
1509 | if (error) | |
1510 | return error; | |
5fb5aeee | 1511 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
6dd8638e BF |
1512 | |
1513 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
1514 | if (error) | |
1515 | return error; | |
5fb5aeee | 1516 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
6dd8638e BF |
1517 | ASSERT((XFS_AGINO_TO_OFFSET(cur->bc_mp, rec.ir_startino) % |
1518 | XFS_INODES_PER_CHUNK) == 0); | |
1519 | ||
1520 | rec.ir_free &= ~XFS_INOBT_MASK(offset); | |
1521 | rec.ir_freecount--; | |
1522 | ||
5fb5aeee | 1523 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, (rec.ir_free == frec->ir_free) && |
6dd8638e BF |
1524 | (rec.ir_freecount == frec->ir_freecount)); |
1525 | ||
b72091f2 | 1526 | return xfs_inobt_update(cur, &rec); |
6dd8638e BF |
1527 | } |
1528 | ||
1529 | /* | |
1530 | * Allocate an inode using the free inode btree, if available. Otherwise, fall | |
1531 | * back to the inobt search algorithm. | |
1532 | * | |
1533 | * The caller selected an AG for us, and made sure that free inodes are | |
1534 | * available. | |
1535 | */ | |
1536 | STATIC int | |
1537 | xfs_dialloc_ag( | |
1538 | struct xfs_trans *tp, | |
1539 | struct xfs_buf *agbp, | |
1540 | xfs_ino_t parent, | |
1541 | xfs_ino_t *inop) | |
1542 | { | |
1543 | struct xfs_mount *mp = tp->t_mountp; | |
1544 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
1545 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
1546 | xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent); | |
1547 | xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent); | |
1548 | struct xfs_perag *pag; | |
1549 | struct xfs_btree_cur *cur; /* finobt cursor */ | |
1550 | struct xfs_btree_cur *icur; /* inobt cursor */ | |
1551 | struct xfs_inobt_rec_incore rec; | |
1552 | xfs_ino_t ino; | |
1553 | int error; | |
1554 | int offset; | |
1555 | int i; | |
1556 | ||
1557 | if (!xfs_sb_version_hasfinobt(&mp->m_sb)) | |
1558 | return xfs_dialloc_ag_inobt(tp, agbp, parent, inop); | |
1559 | ||
1560 | pag = xfs_perag_get(mp, agno); | |
1561 | ||
1562 | /* | |
1563 | * If pagino is 0 (this is the root inode allocation) use newino. | |
1564 | * This must work because we've just allocated some. | |
1565 | */ | |
1566 | if (!pagino) | |
1567 | pagino = be32_to_cpu(agi->agi_newino); | |
1568 | ||
1569 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO); | |
1570 | ||
1571 | error = xfs_check_agi_freecount(cur, agi); | |
1572 | if (error) | |
1573 | goto error_cur; | |
1574 | ||
1575 | /* | |
1576 | * The search algorithm depends on whether we're in the same AG as the | |
1577 | * parent. If so, find the closest available inode to the parent. If | |
1578 | * not, consider the agi hint or find the first free inode in the AG. | |
1579 | */ | |
1580 | if (agno == pagno) | |
1581 | error = xfs_dialloc_ag_finobt_near(pagino, &cur, &rec); | |
1582 | else | |
1583 | error = xfs_dialloc_ag_finobt_newino(agi, cur, &rec); | |
1584 | if (error) | |
1585 | goto error_cur; | |
1586 | ||
d4cc540b | 1587 | offset = xfs_inobt_first_free_inode(&rec); |
6dd8638e BF |
1588 | ASSERT(offset >= 0); |
1589 | ASSERT(offset < XFS_INODES_PER_CHUNK); | |
1590 | ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) % | |
1591 | XFS_INODES_PER_CHUNK) == 0); | |
1592 | ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset); | |
1593 | ||
1594 | /* | |
1595 | * Modify or remove the finobt record. | |
1596 | */ | |
1597 | rec.ir_free &= ~XFS_INOBT_MASK(offset); | |
1598 | rec.ir_freecount--; | |
1599 | if (rec.ir_freecount) | |
1600 | error = xfs_inobt_update(cur, &rec); | |
1601 | else | |
1602 | error = xfs_btree_delete(cur, &i); | |
1603 | if (error) | |
1604 | goto error_cur; | |
1605 | ||
1606 | /* | |
1607 | * The finobt has now been updated appropriately. We haven't updated the | |
1608 | * agi and superblock yet, so we can create an inobt cursor and validate | |
1609 | * the original freecount. If all is well, make the equivalent update to | |
1610 | * the inobt using the finobt record and offset information. | |
1611 | */ | |
1612 | icur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); | |
1613 | ||
1614 | error = xfs_check_agi_freecount(icur, agi); | |
1615 | if (error) | |
1616 | goto error_icur; | |
1617 | ||
1618 | error = xfs_dialloc_ag_update_inobt(icur, &rec, offset); | |
1619 | if (error) | |
1620 | goto error_icur; | |
1621 | ||
1622 | /* | |
1623 | * Both trees have now been updated. We must update the perag and | |
1624 | * superblock before we can check the freecount for each btree. | |
1625 | */ | |
1626 | be32_add_cpu(&agi->agi_freecount, -1); | |
1627 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); | |
1628 | pag->pagi_freecount--; | |
1629 | ||
1630 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1); | |
1631 | ||
1632 | error = xfs_check_agi_freecount(icur, agi); | |
1633 | if (error) | |
1634 | goto error_icur; | |
1635 | error = xfs_check_agi_freecount(cur, agi); | |
1636 | if (error) | |
1637 | goto error_icur; | |
1638 | ||
1639 | xfs_btree_del_cursor(icur, XFS_BTREE_NOERROR); | |
1640 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
1641 | xfs_perag_put(pag); | |
1642 | *inop = ino; | |
1643 | return 0; | |
1644 | ||
1645 | error_icur: | |
1646 | xfs_btree_del_cursor(icur, XFS_BTREE_ERROR); | |
1647 | error_cur: | |
1648 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
1649 | xfs_perag_put(pag); | |
1650 | return error; | |
1651 | } | |
1652 | ||
f2ecc5e4 CH |
1653 | /* |
1654 | * Allocate an inode on disk. | |
1655 | * | |
1656 | * Mode is used to tell whether the new inode will need space, and whether it | |
1657 | * is a directory. | |
1658 | * | |
1659 | * This function is designed to be called twice if it has to do an allocation | |
1660 | * to make more free inodes. On the first call, *IO_agbp should be set to NULL. | |
1661 | * If an inode is available without having to performn an allocation, an inode | |
cd856db6 CM |
1662 | * number is returned. In this case, *IO_agbp is set to NULL. If an allocation |
1663 | * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp. | |
1664 | * The caller should then commit the current transaction, allocate a | |
f2ecc5e4 CH |
1665 | * new transaction, and call xfs_dialloc() again, passing in the previous value |
1666 | * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI | |
1667 | * buffer is locked across the two calls, the second call is guaranteed to have | |
1668 | * a free inode available. | |
1669 | * | |
1670 | * Once we successfully pick an inode its number is returned and the on-disk | |
1671 | * data structures are updated. The inode itself is not read in, since doing so | |
1672 | * would break ordering constraints with xfs_reclaim. | |
1673 | */ | |
1674 | int | |
1675 | xfs_dialloc( | |
1676 | struct xfs_trans *tp, | |
1677 | xfs_ino_t parent, | |
1678 | umode_t mode, | |
f2ecc5e4 | 1679 | struct xfs_buf **IO_agbp, |
f2ecc5e4 CH |
1680 | xfs_ino_t *inop) |
1681 | { | |
55d6af64 | 1682 | struct xfs_mount *mp = tp->t_mountp; |
f2ecc5e4 CH |
1683 | struct xfs_buf *agbp; |
1684 | xfs_agnumber_t agno; | |
f2ecc5e4 CH |
1685 | int error; |
1686 | int ialloced; | |
1687 | int noroom = 0; | |
be60fe54 | 1688 | xfs_agnumber_t start_agno; |
f2ecc5e4 | 1689 | struct xfs_perag *pag; |
f59cf5c2 | 1690 | int okalloc = 1; |
f2ecc5e4 | 1691 | |
4bb61069 | 1692 | if (*IO_agbp) { |
f2ecc5e4 | 1693 | /* |
4bb61069 CH |
1694 | * If the caller passes in a pointer to the AGI buffer, |
1695 | * continue where we left off before. In this case, we | |
f2ecc5e4 CH |
1696 | * know that the allocation group has free inodes. |
1697 | */ | |
1698 | agbp = *IO_agbp; | |
4bb61069 | 1699 | goto out_alloc; |
f2ecc5e4 | 1700 | } |
4bb61069 CH |
1701 | |
1702 | /* | |
1703 | * We do not have an agbp, so select an initial allocation | |
1704 | * group for inode allocation. | |
1705 | */ | |
f59cf5c2 | 1706 | start_agno = xfs_ialloc_ag_select(tp, parent, mode); |
be60fe54 | 1707 | if (start_agno == NULLAGNUMBER) { |
4bb61069 CH |
1708 | *inop = NULLFSINO; |
1709 | return 0; | |
1710 | } | |
55d6af64 | 1711 | |
f2ecc5e4 CH |
1712 | /* |
1713 | * If we have already hit the ceiling of inode blocks then clear | |
1714 | * okalloc so we scan all available agi structures for a free | |
1715 | * inode. | |
74f9ce1c GW |
1716 | * |
1717 | * Read rough value of mp->m_icount by percpu_counter_read_positive, | |
1718 | * which will sacrifice the preciseness but improve the performance. | |
f2ecc5e4 | 1719 | */ |
f2ecc5e4 | 1720 | if (mp->m_maxicount && |
74f9ce1c GW |
1721 | percpu_counter_read_positive(&mp->m_icount) + mp->m_ialloc_inos |
1722 | > mp->m_maxicount) { | |
f2ecc5e4 CH |
1723 | noroom = 1; |
1724 | okalloc = 0; | |
1725 | } | |
1726 | ||
1727 | /* | |
1728 | * Loop until we find an allocation group that either has free inodes | |
1729 | * or in which we can allocate some inodes. Iterate through the | |
1730 | * allocation groups upward, wrapping at the end. | |
1731 | */ | |
be60fe54 CH |
1732 | agno = start_agno; |
1733 | for (;;) { | |
1734 | pag = xfs_perag_get(mp, agno); | |
1735 | if (!pag->pagi_inodeok) { | |
1736 | xfs_ialloc_next_ag(mp); | |
1737 | goto nextag; | |
1738 | } | |
1739 | ||
1740 | if (!pag->pagi_init) { | |
1741 | error = xfs_ialloc_pagi_init(mp, tp, agno); | |
1742 | if (error) | |
1743 | goto out_error; | |
f2ecc5e4 | 1744 | } |
be60fe54 | 1745 | |
f2ecc5e4 | 1746 | /* |
be60fe54 | 1747 | * Do a first racy fast path check if this AG is usable. |
f2ecc5e4 | 1748 | */ |
be60fe54 CH |
1749 | if (!pag->pagi_freecount && !okalloc) |
1750 | goto nextag; | |
1751 | ||
c4982110 CH |
1752 | /* |
1753 | * Then read in the AGI buffer and recheck with the AGI buffer | |
1754 | * lock held. | |
1755 | */ | |
be60fe54 CH |
1756 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); |
1757 | if (error) | |
1758 | goto out_error; | |
1759 | ||
be60fe54 CH |
1760 | if (pag->pagi_freecount) { |
1761 | xfs_perag_put(pag); | |
1762 | goto out_alloc; | |
1763 | } | |
1764 | ||
c4982110 CH |
1765 | if (!okalloc) |
1766 | goto nextag_relse_buffer; | |
1767 | ||
be60fe54 CH |
1768 | |
1769 | error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced); | |
1770 | if (error) { | |
1771 | xfs_trans_brelse(tp, agbp); | |
1772 | ||
2451337d | 1773 | if (error != -ENOSPC) |
be60fe54 CH |
1774 | goto out_error; |
1775 | ||
1776 | xfs_perag_put(pag); | |
f2ecc5e4 | 1777 | *inop = NULLFSINO; |
be60fe54 | 1778 | return 0; |
f2ecc5e4 | 1779 | } |
be60fe54 CH |
1780 | |
1781 | if (ialloced) { | |
1782 | /* | |
1783 | * We successfully allocated some inodes, return | |
1784 | * the current context to the caller so that it | |
1785 | * can commit the current transaction and call | |
1786 | * us again where we left off. | |
1787 | */ | |
1788 | ASSERT(pag->pagi_freecount > 0); | |
f2ecc5e4 | 1789 | xfs_perag_put(pag); |
be60fe54 CH |
1790 | |
1791 | *IO_agbp = agbp; | |
1792 | *inop = NULLFSINO; | |
1793 | return 0; | |
f2ecc5e4 | 1794 | } |
be60fe54 | 1795 | |
c4982110 CH |
1796 | nextag_relse_buffer: |
1797 | xfs_trans_brelse(tp, agbp); | |
be60fe54 | 1798 | nextag: |
f2ecc5e4 | 1799 | xfs_perag_put(pag); |
be60fe54 CH |
1800 | if (++agno == mp->m_sb.sb_agcount) |
1801 | agno = 0; | |
1802 | if (agno == start_agno) { | |
1803 | *inop = NULLFSINO; | |
2451337d | 1804 | return noroom ? -ENOSPC : 0; |
be60fe54 | 1805 | } |
f2ecc5e4 CH |
1806 | } |
1807 | ||
4bb61069 | 1808 | out_alloc: |
f2ecc5e4 CH |
1809 | *IO_agbp = NULL; |
1810 | return xfs_dialloc_ag(tp, agbp, parent, inop); | |
be60fe54 CH |
1811 | out_error: |
1812 | xfs_perag_put(pag); | |
b474c7ae | 1813 | return error; |
f2ecc5e4 CH |
1814 | } |
1815 | ||
10ae3dc7 BF |
1816 | /* |
1817 | * Free the blocks of an inode chunk. We must consider that the inode chunk | |
1818 | * might be sparse and only free the regions that are allocated as part of the | |
1819 | * chunk. | |
1820 | */ | |
1821 | STATIC void | |
1822 | xfs_difree_inode_chunk( | |
1823 | struct xfs_mount *mp, | |
1824 | xfs_agnumber_t agno, | |
1825 | struct xfs_inobt_rec_incore *rec, | |
2c3234d1 | 1826 | struct xfs_defer_ops *dfops) |
10ae3dc7 BF |
1827 | { |
1828 | xfs_agblock_t sagbno = XFS_AGINO_TO_AGBNO(mp, rec->ir_startino); | |
1829 | int startidx, endidx; | |
1830 | int nextbit; | |
1831 | xfs_agblock_t agbno; | |
1832 | int contigblk; | |
340785cc | 1833 | struct xfs_owner_info oinfo; |
10ae3dc7 | 1834 | DECLARE_BITMAP(holemask, XFS_INOBT_HOLEMASK_BITS); |
340785cc | 1835 | xfs_rmap_ag_owner(&oinfo, XFS_RMAP_OWN_INODES); |
10ae3dc7 BF |
1836 | |
1837 | if (!xfs_inobt_issparse(rec->ir_holemask)) { | |
1838 | /* not sparse, calculate extent info directly */ | |
2c3234d1 | 1839 | xfs_bmap_add_free(mp, dfops, XFS_AGB_TO_FSB(mp, agno, sagbno), |
340785cc | 1840 | mp->m_ialloc_blks, &oinfo); |
10ae3dc7 BF |
1841 | return; |
1842 | } | |
1843 | ||
1844 | /* holemask is only 16-bits (fits in an unsigned long) */ | |
1845 | ASSERT(sizeof(rec->ir_holemask) <= sizeof(holemask[0])); | |
1846 | holemask[0] = rec->ir_holemask; | |
1847 | ||
1848 | /* | |
1849 | * Find contiguous ranges of zeroes (i.e., allocated regions) in the | |
1850 | * holemask and convert the start/end index of each range to an extent. | |
1851 | * We start with the start and end index both pointing at the first 0 in | |
1852 | * the mask. | |
1853 | */ | |
1854 | startidx = endidx = find_first_zero_bit(holemask, | |
1855 | XFS_INOBT_HOLEMASK_BITS); | |
1856 | nextbit = startidx + 1; | |
1857 | while (startidx < XFS_INOBT_HOLEMASK_BITS) { | |
1858 | nextbit = find_next_zero_bit(holemask, XFS_INOBT_HOLEMASK_BITS, | |
1859 | nextbit); | |
1860 | /* | |
1861 | * If the next zero bit is contiguous, update the end index of | |
1862 | * the current range and continue. | |
1863 | */ | |
1864 | if (nextbit != XFS_INOBT_HOLEMASK_BITS && | |
1865 | nextbit == endidx + 1) { | |
1866 | endidx = nextbit; | |
1867 | goto next; | |
1868 | } | |
1869 | ||
1870 | /* | |
1871 | * nextbit is not contiguous with the current end index. Convert | |
1872 | * the current start/end to an extent and add it to the free | |
1873 | * list. | |
1874 | */ | |
1875 | agbno = sagbno + (startidx * XFS_INODES_PER_HOLEMASK_BIT) / | |
1876 | mp->m_sb.sb_inopblock; | |
1877 | contigblk = ((endidx - startidx + 1) * | |
1878 | XFS_INODES_PER_HOLEMASK_BIT) / | |
1879 | mp->m_sb.sb_inopblock; | |
1880 | ||
1881 | ASSERT(agbno % mp->m_sb.sb_spino_align == 0); | |
1882 | ASSERT(contigblk % mp->m_sb.sb_spino_align == 0); | |
2c3234d1 | 1883 | xfs_bmap_add_free(mp, dfops, XFS_AGB_TO_FSB(mp, agno, agbno), |
340785cc | 1884 | contigblk, &oinfo); |
10ae3dc7 BF |
1885 | |
1886 | /* reset range to current bit and carry on... */ | |
1887 | startidx = endidx = nextbit; | |
1888 | ||
1889 | next: | |
1890 | nextbit++; | |
1891 | } | |
1892 | } | |
1893 | ||
2b64ee5c BF |
1894 | STATIC int |
1895 | xfs_difree_inobt( | |
1896 | struct xfs_mount *mp, | |
1897 | struct xfs_trans *tp, | |
1898 | struct xfs_buf *agbp, | |
1899 | xfs_agino_t agino, | |
2c3234d1 | 1900 | struct xfs_defer_ops *dfops, |
09b56604 | 1901 | struct xfs_icluster *xic, |
2b64ee5c | 1902 | struct xfs_inobt_rec_incore *orec) |
1da177e4 | 1903 | { |
2b64ee5c BF |
1904 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); |
1905 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
1906 | struct xfs_perag *pag; | |
1907 | struct xfs_btree_cur *cur; | |
1908 | struct xfs_inobt_rec_incore rec; | |
1909 | int ilen; | |
1910 | int error; | |
1911 | int i; | |
1912 | int off; | |
1da177e4 | 1913 | |
69ef921b | 1914 | ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC)); |
2b64ee5c BF |
1915 | ASSERT(XFS_AGINO_TO_AGBNO(mp, agino) < be32_to_cpu(agi->agi_length)); |
1916 | ||
1da177e4 LT |
1917 | /* |
1918 | * Initialize the cursor. | |
1919 | */ | |
57bd3dbe | 1920 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
1da177e4 | 1921 | |
0b48db80 DC |
1922 | error = xfs_check_agi_freecount(cur, agi); |
1923 | if (error) | |
1924 | goto error0; | |
1925 | ||
1da177e4 LT |
1926 | /* |
1927 | * Look for the entry describing this inode. | |
1928 | */ | |
21875505 | 1929 | if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) { |
0b932ccc DC |
1930 | xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.", |
1931 | __func__, error); | |
1da177e4 LT |
1932 | goto error0; |
1933 | } | |
c29aad41 | 1934 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
2e287a73 CH |
1935 | error = xfs_inobt_get_rec(cur, &rec, &i); |
1936 | if (error) { | |
0b932ccc DC |
1937 | xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.", |
1938 | __func__, error); | |
1da177e4 LT |
1939 | goto error0; |
1940 | } | |
c29aad41 | 1941 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 LT |
1942 | /* |
1943 | * Get the offset in the inode chunk. | |
1944 | */ | |
1945 | off = agino - rec.ir_startino; | |
1946 | ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK); | |
0d87e656 | 1947 | ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off))); |
1da177e4 LT |
1948 | /* |
1949 | * Mark the inode free & increment the count. | |
1950 | */ | |
0d87e656 | 1951 | rec.ir_free |= XFS_INOBT_MASK(off); |
1da177e4 LT |
1952 | rec.ir_freecount++; |
1953 | ||
1954 | /* | |
999633d3 BF |
1955 | * When an inode chunk is free, it becomes eligible for removal. Don't |
1956 | * remove the chunk if the block size is large enough for multiple inode | |
1957 | * chunks (that might not be free). | |
1da177e4 | 1958 | */ |
1bd960ee | 1959 | if (!(mp->m_flags & XFS_MOUNT_IKEEP) && |
999633d3 BF |
1960 | rec.ir_free == XFS_INOBT_ALL_FREE && |
1961 | mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK) { | |
749f24f3 | 1962 | xic->deleted = true; |
09b56604 BF |
1963 | xic->first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino); |
1964 | xic->alloc = xfs_inobt_irec_to_allocmask(&rec); | |
1da177e4 LT |
1965 | |
1966 | /* | |
1967 | * Remove the inode cluster from the AGI B+Tree, adjust the | |
1968 | * AGI and Superblock inode counts, and mark the disk space | |
1969 | * to be freed when the transaction is committed. | |
1970 | */ | |
999633d3 | 1971 | ilen = rec.ir_freecount; |
413d57c9 MS |
1972 | be32_add_cpu(&agi->agi_count, -ilen); |
1973 | be32_add_cpu(&agi->agi_freecount, -(ilen - 1)); | |
1da177e4 | 1974 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT); |
44b56e0a DC |
1975 | pag = xfs_perag_get(mp, agno); |
1976 | pag->pagi_freecount -= ilen - 1; | |
1977 | xfs_perag_put(pag); | |
1da177e4 LT |
1978 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen); |
1979 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1)); | |
1980 | ||
91cca5df | 1981 | if ((error = xfs_btree_delete(cur, &i))) { |
0b932ccc DC |
1982 | xfs_warn(mp, "%s: xfs_btree_delete returned error %d.", |
1983 | __func__, error); | |
1da177e4 LT |
1984 | goto error0; |
1985 | } | |
1986 | ||
2c3234d1 | 1987 | xfs_difree_inode_chunk(mp, agno, &rec, dfops); |
1da177e4 | 1988 | } else { |
749f24f3 | 1989 | xic->deleted = false; |
1da177e4 | 1990 | |
afabc24a CH |
1991 | error = xfs_inobt_update(cur, &rec); |
1992 | if (error) { | |
0b932ccc DC |
1993 | xfs_warn(mp, "%s: xfs_inobt_update returned error %d.", |
1994 | __func__, error); | |
1da177e4 LT |
1995 | goto error0; |
1996 | } | |
afabc24a | 1997 | |
1da177e4 LT |
1998 | /* |
1999 | * Change the inode free counts and log the ag/sb changes. | |
2000 | */ | |
413d57c9 | 2001 | be32_add_cpu(&agi->agi_freecount, 1); |
1da177e4 | 2002 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); |
44b56e0a DC |
2003 | pag = xfs_perag_get(mp, agno); |
2004 | pag->pagi_freecount++; | |
2005 | xfs_perag_put(pag); | |
1da177e4 LT |
2006 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1); |
2007 | } | |
2008 | ||
0b48db80 DC |
2009 | error = xfs_check_agi_freecount(cur, agi); |
2010 | if (error) | |
2011 | goto error0; | |
1da177e4 | 2012 | |
2b64ee5c | 2013 | *orec = rec; |
1da177e4 LT |
2014 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
2015 | return 0; | |
2016 | ||
2017 | error0: | |
2018 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
2019 | return error; | |
2020 | } | |
2021 | ||
3efa4ffd BF |
2022 | /* |
2023 | * Free an inode in the free inode btree. | |
2024 | */ | |
2025 | STATIC int | |
2026 | xfs_difree_finobt( | |
2027 | struct xfs_mount *mp, | |
2028 | struct xfs_trans *tp, | |
2029 | struct xfs_buf *agbp, | |
2030 | xfs_agino_t agino, | |
2031 | struct xfs_inobt_rec_incore *ibtrec) /* inobt record */ | |
2032 | { | |
2033 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
2034 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
2035 | struct xfs_btree_cur *cur; | |
2036 | struct xfs_inobt_rec_incore rec; | |
2037 | int offset = agino - ibtrec->ir_startino; | |
2038 | int error; | |
2039 | int i; | |
2040 | ||
2041 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO); | |
2042 | ||
2043 | error = xfs_inobt_lookup(cur, ibtrec->ir_startino, XFS_LOOKUP_EQ, &i); | |
2044 | if (error) | |
2045 | goto error; | |
2046 | if (i == 0) { | |
2047 | /* | |
2048 | * If the record does not exist in the finobt, we must have just | |
2049 | * freed an inode in a previously fully allocated chunk. If not, | |
2050 | * something is out of sync. | |
2051 | */ | |
c29aad41 | 2052 | XFS_WANT_CORRUPTED_GOTO(mp, ibtrec->ir_freecount == 1, error); |
3efa4ffd | 2053 | |
5419040f BF |
2054 | error = xfs_inobt_insert_rec(cur, ibtrec->ir_holemask, |
2055 | ibtrec->ir_count, | |
2056 | ibtrec->ir_freecount, | |
3efa4ffd BF |
2057 | ibtrec->ir_free, &i); |
2058 | if (error) | |
2059 | goto error; | |
2060 | ASSERT(i == 1); | |
2061 | ||
2062 | goto out; | |
2063 | } | |
2064 | ||
2065 | /* | |
2066 | * Read and update the existing record. We could just copy the ibtrec | |
2067 | * across here, but that would defeat the purpose of having redundant | |
2068 | * metadata. By making the modifications independently, we can catch | |
2069 | * corruptions that we wouldn't see if we just copied from one record | |
2070 | * to another. | |
2071 | */ | |
2072 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
2073 | if (error) | |
2074 | goto error; | |
c29aad41 | 2075 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error); |
3efa4ffd BF |
2076 | |
2077 | rec.ir_free |= XFS_INOBT_MASK(offset); | |
2078 | rec.ir_freecount++; | |
2079 | ||
c29aad41 | 2080 | XFS_WANT_CORRUPTED_GOTO(mp, (rec.ir_free == ibtrec->ir_free) && |
3efa4ffd BF |
2081 | (rec.ir_freecount == ibtrec->ir_freecount), |
2082 | error); | |
2083 | ||
2084 | /* | |
2085 | * The content of inobt records should always match between the inobt | |
2086 | * and finobt. The lifecycle of records in the finobt is different from | |
2087 | * the inobt in that the finobt only tracks records with at least one | |
2088 | * free inode. Hence, if all of the inodes are free and we aren't | |
2089 | * keeping inode chunks permanently on disk, remove the record. | |
2090 | * Otherwise, update the record with the new information. | |
999633d3 BF |
2091 | * |
2092 | * Note that we currently can't free chunks when the block size is large | |
2093 | * enough for multiple chunks. Leave the finobt record to remain in sync | |
2094 | * with the inobt. | |
3efa4ffd | 2095 | */ |
999633d3 BF |
2096 | if (rec.ir_free == XFS_INOBT_ALL_FREE && |
2097 | mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK && | |
3efa4ffd BF |
2098 | !(mp->m_flags & XFS_MOUNT_IKEEP)) { |
2099 | error = xfs_btree_delete(cur, &i); | |
2100 | if (error) | |
2101 | goto error; | |
2102 | ASSERT(i == 1); | |
2103 | } else { | |
2104 | error = xfs_inobt_update(cur, &rec); | |
2105 | if (error) | |
2106 | goto error; | |
2107 | } | |
2108 | ||
2109 | out: | |
2110 | error = xfs_check_agi_freecount(cur, agi); | |
2111 | if (error) | |
2112 | goto error; | |
2113 | ||
2114 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
2115 | return 0; | |
2116 | ||
2117 | error: | |
2118 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
2119 | return error; | |
2120 | } | |
2121 | ||
2b64ee5c BF |
2122 | /* |
2123 | * Free disk inode. Carefully avoids touching the incore inode, all | |
2124 | * manipulations incore are the caller's responsibility. | |
2125 | * The on-disk inode is not changed by this operation, only the | |
2126 | * btree (free inode mask) is changed. | |
2127 | */ | |
2128 | int | |
2129 | xfs_difree( | |
2130 | struct xfs_trans *tp, /* transaction pointer */ | |
2131 | xfs_ino_t inode, /* inode to be freed */ | |
2c3234d1 | 2132 | struct xfs_defer_ops *dfops, /* extents to free */ |
09b56604 | 2133 | struct xfs_icluster *xic) /* cluster info if deleted */ |
2b64ee5c BF |
2134 | { |
2135 | /* REFERENCED */ | |
2136 | xfs_agblock_t agbno; /* block number containing inode */ | |
2137 | struct xfs_buf *agbp; /* buffer for allocation group header */ | |
2138 | xfs_agino_t agino; /* allocation group inode number */ | |
2139 | xfs_agnumber_t agno; /* allocation group number */ | |
2140 | int error; /* error return value */ | |
2141 | struct xfs_mount *mp; /* mount structure for filesystem */ | |
2142 | struct xfs_inobt_rec_incore rec;/* btree record */ | |
2143 | ||
2144 | mp = tp->t_mountp; | |
2145 | ||
2146 | /* | |
2147 | * Break up inode number into its components. | |
2148 | */ | |
2149 | agno = XFS_INO_TO_AGNO(mp, inode); | |
2150 | if (agno >= mp->m_sb.sb_agcount) { | |
2151 | xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).", | |
2152 | __func__, agno, mp->m_sb.sb_agcount); | |
2153 | ASSERT(0); | |
2451337d | 2154 | return -EINVAL; |
2b64ee5c BF |
2155 | } |
2156 | agino = XFS_INO_TO_AGINO(mp, inode); | |
2157 | if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
2158 | xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).", | |
2159 | __func__, (unsigned long long)inode, | |
2160 | (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino)); | |
2161 | ASSERT(0); | |
2451337d | 2162 | return -EINVAL; |
2b64ee5c BF |
2163 | } |
2164 | agbno = XFS_AGINO_TO_AGBNO(mp, agino); | |
2165 | if (agbno >= mp->m_sb.sb_agblocks) { | |
2166 | xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).", | |
2167 | __func__, agbno, mp->m_sb.sb_agblocks); | |
2168 | ASSERT(0); | |
2451337d | 2169 | return -EINVAL; |
2b64ee5c BF |
2170 | } |
2171 | /* | |
2172 | * Get the allocation group header. | |
2173 | */ | |
2174 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); | |
2175 | if (error) { | |
2176 | xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.", | |
2177 | __func__, error); | |
2178 | return error; | |
2179 | } | |
2180 | ||
2181 | /* | |
2182 | * Fix up the inode allocation btree. | |
2183 | */ | |
2c3234d1 | 2184 | error = xfs_difree_inobt(mp, tp, agbp, agino, dfops, xic, &rec); |
2b64ee5c BF |
2185 | if (error) |
2186 | goto error0; | |
2187 | ||
3efa4ffd BF |
2188 | /* |
2189 | * Fix up the free inode btree. | |
2190 | */ | |
2191 | if (xfs_sb_version_hasfinobt(&mp->m_sb)) { | |
2192 | error = xfs_difree_finobt(mp, tp, agbp, agino, &rec); | |
2193 | if (error) | |
2194 | goto error0; | |
2195 | } | |
2196 | ||
2b64ee5c BF |
2197 | return 0; |
2198 | ||
2199 | error0: | |
2200 | return error; | |
2201 | } | |
2202 | ||
7124fe0a DC |
2203 | STATIC int |
2204 | xfs_imap_lookup( | |
2205 | struct xfs_mount *mp, | |
2206 | struct xfs_trans *tp, | |
2207 | xfs_agnumber_t agno, | |
2208 | xfs_agino_t agino, | |
2209 | xfs_agblock_t agbno, | |
2210 | xfs_agblock_t *chunk_agbno, | |
2211 | xfs_agblock_t *offset_agbno, | |
2212 | int flags) | |
2213 | { | |
2214 | struct xfs_inobt_rec_incore rec; | |
2215 | struct xfs_btree_cur *cur; | |
2216 | struct xfs_buf *agbp; | |
7124fe0a DC |
2217 | int error; |
2218 | int i; | |
2219 | ||
2220 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); | |
2221 | if (error) { | |
53487786 DC |
2222 | xfs_alert(mp, |
2223 | "%s: xfs_ialloc_read_agi() returned error %d, agno %d", | |
2224 | __func__, error, agno); | |
7124fe0a DC |
2225 | return error; |
2226 | } | |
2227 | ||
2228 | /* | |
4536f2ad DC |
2229 | * Lookup the inode record for the given agino. If the record cannot be |
2230 | * found, then it's an invalid inode number and we should abort. Once | |
2231 | * we have a record, we need to ensure it contains the inode number | |
2232 | * we are looking up. | |
7124fe0a | 2233 | */ |
57bd3dbe | 2234 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
4536f2ad | 2235 | error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i); |
7124fe0a DC |
2236 | if (!error) { |
2237 | if (i) | |
2238 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
2239 | if (!error && i == 0) | |
2451337d | 2240 | error = -EINVAL; |
7124fe0a DC |
2241 | } |
2242 | ||
2243 | xfs_trans_brelse(tp, agbp); | |
f307080a | 2244 | xfs_btree_del_cursor(cur, error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); |
7124fe0a DC |
2245 | if (error) |
2246 | return error; | |
2247 | ||
4536f2ad DC |
2248 | /* check that the returned record contains the required inode */ |
2249 | if (rec.ir_startino > agino || | |
71783438 | 2250 | rec.ir_startino + mp->m_ialloc_inos <= agino) |
2451337d | 2251 | return -EINVAL; |
4536f2ad | 2252 | |
7124fe0a | 2253 | /* for untrusted inodes check it is allocated first */ |
1920779e | 2254 | if ((flags & XFS_IGET_UNTRUSTED) && |
7124fe0a | 2255 | (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino))) |
2451337d | 2256 | return -EINVAL; |
7124fe0a DC |
2257 | |
2258 | *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino); | |
2259 | *offset_agbno = agbno - *chunk_agbno; | |
2260 | return 0; | |
2261 | } | |
2262 | ||
1da177e4 | 2263 | /* |
94e1b69d | 2264 | * Return the location of the inode in imap, for mapping it into a buffer. |
1da177e4 | 2265 | */ |
1da177e4 | 2266 | int |
94e1b69d CH |
2267 | xfs_imap( |
2268 | xfs_mount_t *mp, /* file system mount structure */ | |
2269 | xfs_trans_t *tp, /* transaction pointer */ | |
1da177e4 | 2270 | xfs_ino_t ino, /* inode to locate */ |
94e1b69d CH |
2271 | struct xfs_imap *imap, /* location map structure */ |
2272 | uint flags) /* flags for inode btree lookup */ | |
1da177e4 LT |
2273 | { |
2274 | xfs_agblock_t agbno; /* block number of inode in the alloc group */ | |
1da177e4 LT |
2275 | xfs_agino_t agino; /* inode number within alloc group */ |
2276 | xfs_agnumber_t agno; /* allocation group number */ | |
2277 | int blks_per_cluster; /* num blocks per inode cluster */ | |
2278 | xfs_agblock_t chunk_agbno; /* first block in inode chunk */ | |
1da177e4 | 2279 | xfs_agblock_t cluster_agbno; /* first block in inode cluster */ |
1da177e4 | 2280 | int error; /* error code */ |
1da177e4 | 2281 | int offset; /* index of inode in its buffer */ |
836a94ad | 2282 | xfs_agblock_t offset_agbno; /* blks from chunk start to inode */ |
1da177e4 LT |
2283 | |
2284 | ASSERT(ino != NULLFSINO); | |
94e1b69d | 2285 | |
1da177e4 LT |
2286 | /* |
2287 | * Split up the inode number into its parts. | |
2288 | */ | |
2289 | agno = XFS_INO_TO_AGNO(mp, ino); | |
2290 | agino = XFS_INO_TO_AGINO(mp, ino); | |
2291 | agbno = XFS_AGINO_TO_AGBNO(mp, agino); | |
2292 | if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks || | |
2293 | ino != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
2294 | #ifdef DEBUG | |
1920779e DC |
2295 | /* |
2296 | * Don't output diagnostic information for untrusted inodes | |
2297 | * as they can be invalid without implying corruption. | |
2298 | */ | |
2299 | if (flags & XFS_IGET_UNTRUSTED) | |
2451337d | 2300 | return -EINVAL; |
1da177e4 | 2301 | if (agno >= mp->m_sb.sb_agcount) { |
53487786 DC |
2302 | xfs_alert(mp, |
2303 | "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)", | |
2304 | __func__, agno, mp->m_sb.sb_agcount); | |
1da177e4 LT |
2305 | } |
2306 | if (agbno >= mp->m_sb.sb_agblocks) { | |
53487786 DC |
2307 | xfs_alert(mp, |
2308 | "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)", | |
2309 | __func__, (unsigned long long)agbno, | |
2310 | (unsigned long)mp->m_sb.sb_agblocks); | |
1da177e4 LT |
2311 | } |
2312 | if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
53487786 DC |
2313 | xfs_alert(mp, |
2314 | "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)", | |
2315 | __func__, ino, | |
2316 | XFS_AGINO_TO_INO(mp, agno, agino)); | |
1da177e4 | 2317 | } |
745b1f47 | 2318 | xfs_stack_trace(); |
1da177e4 | 2319 | #endif /* DEBUG */ |
2451337d | 2320 | return -EINVAL; |
1da177e4 | 2321 | } |
94e1b69d | 2322 | |
f9e5abcf | 2323 | blks_per_cluster = xfs_icluster_size_fsb(mp); |
7124fe0a DC |
2324 | |
2325 | /* | |
2326 | * For bulkstat and handle lookups, we have an untrusted inode number | |
2327 | * that we have to verify is valid. We cannot do this just by reading | |
2328 | * the inode buffer as it may have been unlinked and removed leaving | |
2329 | * inodes in stale state on disk. Hence we have to do a btree lookup | |
2330 | * in all cases where an untrusted inode number is passed. | |
2331 | */ | |
1920779e | 2332 | if (flags & XFS_IGET_UNTRUSTED) { |
7124fe0a DC |
2333 | error = xfs_imap_lookup(mp, tp, agno, agino, agbno, |
2334 | &chunk_agbno, &offset_agbno, flags); | |
2335 | if (error) | |
2336 | return error; | |
2337 | goto out_map; | |
2338 | } | |
2339 | ||
94e1b69d CH |
2340 | /* |
2341 | * If the inode cluster size is the same as the blocksize or | |
2342 | * smaller we get to the buffer by simple arithmetics. | |
2343 | */ | |
f9e5abcf | 2344 | if (blks_per_cluster == 1) { |
1da177e4 LT |
2345 | offset = XFS_INO_TO_OFFSET(mp, ino); |
2346 | ASSERT(offset < mp->m_sb.sb_inopblock); | |
94e1b69d CH |
2347 | |
2348 | imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno); | |
2349 | imap->im_len = XFS_FSB_TO_BB(mp, 1); | |
755c7bf5 DW |
2350 | imap->im_boffset = (unsigned short)(offset << |
2351 | mp->m_sb.sb_inodelog); | |
1da177e4 LT |
2352 | return 0; |
2353 | } | |
94e1b69d | 2354 | |
94e1b69d CH |
2355 | /* |
2356 | * If the inode chunks are aligned then use simple maths to | |
2357 | * find the location. Otherwise we have to do a btree | |
2358 | * lookup to find the location. | |
2359 | */ | |
1da177e4 LT |
2360 | if (mp->m_inoalign_mask) { |
2361 | offset_agbno = agbno & mp->m_inoalign_mask; | |
2362 | chunk_agbno = agbno - offset_agbno; | |
2363 | } else { | |
7124fe0a DC |
2364 | error = xfs_imap_lookup(mp, tp, agno, agino, agbno, |
2365 | &chunk_agbno, &offset_agbno, flags); | |
1da177e4 LT |
2366 | if (error) |
2367 | return error; | |
1da177e4 | 2368 | } |
94e1b69d | 2369 | |
7124fe0a | 2370 | out_map: |
1da177e4 LT |
2371 | ASSERT(agbno >= chunk_agbno); |
2372 | cluster_agbno = chunk_agbno + | |
2373 | ((offset_agbno / blks_per_cluster) * blks_per_cluster); | |
2374 | offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) + | |
2375 | XFS_INO_TO_OFFSET(mp, ino); | |
94e1b69d CH |
2376 | |
2377 | imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno); | |
2378 | imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster); | |
755c7bf5 | 2379 | imap->im_boffset = (unsigned short)(offset << mp->m_sb.sb_inodelog); |
94e1b69d CH |
2380 | |
2381 | /* | |
2382 | * If the inode number maps to a block outside the bounds | |
2383 | * of the file system then return NULL rather than calling | |
2384 | * read_buf and panicing when we get an error from the | |
2385 | * driver. | |
2386 | */ | |
2387 | if ((imap->im_blkno + imap->im_len) > | |
2388 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) { | |
53487786 DC |
2389 | xfs_alert(mp, |
2390 | "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)", | |
2391 | __func__, (unsigned long long) imap->im_blkno, | |
94e1b69d CH |
2392 | (unsigned long long) imap->im_len, |
2393 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)); | |
2451337d | 2394 | return -EINVAL; |
94e1b69d | 2395 | } |
1da177e4 | 2396 | return 0; |
1da177e4 LT |
2397 | } |
2398 | ||
2399 | /* | |
2400 | * Compute and fill in value of m_in_maxlevels. | |
2401 | */ | |
2402 | void | |
2403 | xfs_ialloc_compute_maxlevels( | |
2404 | xfs_mount_t *mp) /* file system mount structure */ | |
2405 | { | |
19b54ee6 DW |
2406 | uint inodes; |
2407 | ||
2408 | inodes = (1LL << XFS_INO_AGINO_BITS(mp)) >> XFS_INODES_PER_CHUNK_LOG; | |
2409 | mp->m_in_maxlevels = xfs_btree_compute_maxlevels(mp, mp->m_inobt_mnr, | |
2410 | inodes); | |
1da177e4 LT |
2411 | } |
2412 | ||
2413 | /* | |
aafc3c24 BF |
2414 | * Log specified fields for the ag hdr (inode section). The growth of the agi |
2415 | * structure over time requires that we interpret the buffer as two logical | |
2416 | * regions delineated by the end of the unlinked list. This is due to the size | |
2417 | * of the hash table and its location in the middle of the agi. | |
2418 | * | |
2419 | * For example, a request to log a field before agi_unlinked and a field after | |
2420 | * agi_unlinked could cause us to log the entire hash table and use an excessive | |
2421 | * amount of log space. To avoid this behavior, log the region up through | |
2422 | * agi_unlinked in one call and the region after agi_unlinked through the end of | |
2423 | * the structure in another. | |
1da177e4 LT |
2424 | */ |
2425 | void | |
2426 | xfs_ialloc_log_agi( | |
2427 | xfs_trans_t *tp, /* transaction pointer */ | |
2428 | xfs_buf_t *bp, /* allocation group header buffer */ | |
2429 | int fields) /* bitmask of fields to log */ | |
2430 | { | |
2431 | int first; /* first byte number */ | |
2432 | int last; /* last byte number */ | |
2433 | static const short offsets[] = { /* field starting offsets */ | |
2434 | /* keep in sync with bit definitions */ | |
2435 | offsetof(xfs_agi_t, agi_magicnum), | |
2436 | offsetof(xfs_agi_t, agi_versionnum), | |
2437 | offsetof(xfs_agi_t, agi_seqno), | |
2438 | offsetof(xfs_agi_t, agi_length), | |
2439 | offsetof(xfs_agi_t, agi_count), | |
2440 | offsetof(xfs_agi_t, agi_root), | |
2441 | offsetof(xfs_agi_t, agi_level), | |
2442 | offsetof(xfs_agi_t, agi_freecount), | |
2443 | offsetof(xfs_agi_t, agi_newino), | |
2444 | offsetof(xfs_agi_t, agi_dirino), | |
2445 | offsetof(xfs_agi_t, agi_unlinked), | |
aafc3c24 BF |
2446 | offsetof(xfs_agi_t, agi_free_root), |
2447 | offsetof(xfs_agi_t, agi_free_level), | |
1da177e4 LT |
2448 | sizeof(xfs_agi_t) |
2449 | }; | |
2450 | #ifdef DEBUG | |
2451 | xfs_agi_t *agi; /* allocation group header */ | |
2452 | ||
2453 | agi = XFS_BUF_TO_AGI(bp); | |
69ef921b | 2454 | ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC)); |
1da177e4 | 2455 | #endif |
aafc3c24 | 2456 | |
1da177e4 | 2457 | /* |
aafc3c24 BF |
2458 | * Compute byte offsets for the first and last fields in the first |
2459 | * region and log the agi buffer. This only logs up through | |
2460 | * agi_unlinked. | |
1da177e4 | 2461 | */ |
aafc3c24 BF |
2462 | if (fields & XFS_AGI_ALL_BITS_R1) { |
2463 | xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R1, | |
2464 | &first, &last); | |
2465 | xfs_trans_log_buf(tp, bp, first, last); | |
2466 | } | |
2467 | ||
1da177e4 | 2468 | /* |
aafc3c24 BF |
2469 | * Mask off the bits in the first region and calculate the first and |
2470 | * last field offsets for any bits in the second region. | |
1da177e4 | 2471 | */ |
aafc3c24 BF |
2472 | fields &= ~XFS_AGI_ALL_BITS_R1; |
2473 | if (fields) { | |
2474 | xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R2, | |
2475 | &first, &last); | |
2476 | xfs_trans_log_buf(tp, bp, first, last); | |
2477 | } | |
1da177e4 LT |
2478 | } |
2479 | ||
5e1be0fb CH |
2480 | #ifdef DEBUG |
2481 | STATIC void | |
2482 | xfs_check_agi_unlinked( | |
2483 | struct xfs_agi *agi) | |
2484 | { | |
2485 | int i; | |
2486 | ||
2487 | for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++) | |
2488 | ASSERT(agi->agi_unlinked[i]); | |
2489 | } | |
2490 | #else | |
2491 | #define xfs_check_agi_unlinked(agi) | |
2492 | #endif | |
2493 | ||
983d09ff | 2494 | static bool |
612cfbfe | 2495 | xfs_agi_verify( |
3702ce6e DC |
2496 | struct xfs_buf *bp) |
2497 | { | |
2498 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
2499 | struct xfs_agi *agi = XFS_BUF_TO_AGI(bp); | |
3702ce6e | 2500 | |
a45086e2 BF |
2501 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
2502 | if (!uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid)) | |
2503 | return false; | |
2504 | if (!xfs_log_check_lsn(mp, | |
2505 | be64_to_cpu(XFS_BUF_TO_AGI(bp)->agi_lsn))) | |
983d09ff | 2506 | return false; |
a45086e2 BF |
2507 | } |
2508 | ||
3702ce6e DC |
2509 | /* |
2510 | * Validate the magic number of the agi block. | |
2511 | */ | |
983d09ff DC |
2512 | if (agi->agi_magicnum != cpu_to_be32(XFS_AGI_MAGIC)) |
2513 | return false; | |
2514 | if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum))) | |
2515 | return false; | |
3702ce6e | 2516 | |
d2a047f3 DW |
2517 | if (be32_to_cpu(agi->agi_level) < 1 || |
2518 | be32_to_cpu(agi->agi_level) > XFS_BTREE_MAXLEVELS) | |
e1b05723 | 2519 | return false; |
d2a047f3 DW |
2520 | |
2521 | if (xfs_sb_version_hasfinobt(&mp->m_sb) && | |
2522 | (be32_to_cpu(agi->agi_free_level) < 1 || | |
2523 | be32_to_cpu(agi->agi_free_level) > XFS_BTREE_MAXLEVELS)) | |
e1b05723 | 2524 | return false; |
d2a047f3 | 2525 | |
3702ce6e DC |
2526 | /* |
2527 | * during growfs operations, the perag is not fully initialised, | |
2528 | * so we can't use it for any useful checking. growfs ensures we can't | |
2529 | * use it by using uncached buffers that don't have the perag attached | |
2530 | * so we can detect and avoid this problem. | |
2531 | */ | |
983d09ff DC |
2532 | if (bp->b_pag && be32_to_cpu(agi->agi_seqno) != bp->b_pag->pag_agno) |
2533 | return false; | |
3702ce6e | 2534 | |
3702ce6e | 2535 | xfs_check_agi_unlinked(agi); |
983d09ff | 2536 | return true; |
612cfbfe DC |
2537 | } |
2538 | ||
1813dd64 DC |
2539 | static void |
2540 | xfs_agi_read_verify( | |
612cfbfe DC |
2541 | struct xfs_buf *bp) |
2542 | { | |
983d09ff | 2543 | struct xfs_mount *mp = bp->b_target->bt_mount; |
983d09ff | 2544 | |
ce5028cf ES |
2545 | if (xfs_sb_version_hascrc(&mp->m_sb) && |
2546 | !xfs_buf_verify_cksum(bp, XFS_AGI_CRC_OFF)) | |
2451337d | 2547 | xfs_buf_ioerror(bp, -EFSBADCRC); |
ce5028cf | 2548 | else if (XFS_TEST_ERROR(!xfs_agi_verify(bp), mp, |
9e24cfd0 | 2549 | XFS_ERRTAG_IALLOC_READ_AGI)) |
2451337d | 2550 | xfs_buf_ioerror(bp, -EFSCORRUPTED); |
ce5028cf ES |
2551 | |
2552 | if (bp->b_error) | |
2553 | xfs_verifier_error(bp); | |
612cfbfe DC |
2554 | } |
2555 | ||
b0f539de | 2556 | static void |
1813dd64 | 2557 | xfs_agi_write_verify( |
612cfbfe DC |
2558 | struct xfs_buf *bp) |
2559 | { | |
983d09ff DC |
2560 | struct xfs_mount *mp = bp->b_target->bt_mount; |
2561 | struct xfs_buf_log_item *bip = bp->b_fspriv; | |
2562 | ||
2563 | if (!xfs_agi_verify(bp)) { | |
2451337d | 2564 | xfs_buf_ioerror(bp, -EFSCORRUPTED); |
ce5028cf | 2565 | xfs_verifier_error(bp); |
983d09ff DC |
2566 | return; |
2567 | } | |
2568 | ||
2569 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
2570 | return; | |
2571 | ||
2572 | if (bip) | |
2573 | XFS_BUF_TO_AGI(bp)->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn); | |
f1dbcd7e | 2574 | xfs_buf_update_cksum(bp, XFS_AGI_CRC_OFF); |
3702ce6e DC |
2575 | } |
2576 | ||
1813dd64 | 2577 | const struct xfs_buf_ops xfs_agi_buf_ops = { |
233135b7 | 2578 | .name = "xfs_agi", |
1813dd64 DC |
2579 | .verify_read = xfs_agi_read_verify, |
2580 | .verify_write = xfs_agi_write_verify, | |
2581 | }; | |
2582 | ||
1da177e4 LT |
2583 | /* |
2584 | * Read in the allocation group header (inode allocation section) | |
2585 | */ | |
2586 | int | |
5e1be0fb CH |
2587 | xfs_read_agi( |
2588 | struct xfs_mount *mp, /* file system mount structure */ | |
2589 | struct xfs_trans *tp, /* transaction pointer */ | |
2590 | xfs_agnumber_t agno, /* allocation group number */ | |
2591 | struct xfs_buf **bpp) /* allocation group hdr buf */ | |
1da177e4 | 2592 | { |
5e1be0fb | 2593 | int error; |
1da177e4 | 2594 | |
d123031a | 2595 | trace_xfs_read_agi(mp, agno); |
5e1be0fb | 2596 | |
d123031a | 2597 | ASSERT(agno != NULLAGNUMBER); |
5e1be0fb | 2598 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, |
1da177e4 | 2599 | XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)), |
1813dd64 | 2600 | XFS_FSS_TO_BB(mp, 1), 0, bpp, &xfs_agi_buf_ops); |
1da177e4 LT |
2601 | if (error) |
2602 | return error; | |
200237d6 ES |
2603 | if (tp) |
2604 | xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_AGI_BUF); | |
5e1be0fb | 2605 | |
38f23232 | 2606 | xfs_buf_set_ref(*bpp, XFS_AGI_REF); |
5e1be0fb CH |
2607 | return 0; |
2608 | } | |
2609 | ||
2610 | int | |
2611 | xfs_ialloc_read_agi( | |
2612 | struct xfs_mount *mp, /* file system mount structure */ | |
2613 | struct xfs_trans *tp, /* transaction pointer */ | |
2614 | xfs_agnumber_t agno, /* allocation group number */ | |
2615 | struct xfs_buf **bpp) /* allocation group hdr buf */ | |
2616 | { | |
2617 | struct xfs_agi *agi; /* allocation group header */ | |
2618 | struct xfs_perag *pag; /* per allocation group data */ | |
2619 | int error; | |
2620 | ||
d123031a DC |
2621 | trace_xfs_ialloc_read_agi(mp, agno); |
2622 | ||
5e1be0fb CH |
2623 | error = xfs_read_agi(mp, tp, agno, bpp); |
2624 | if (error) | |
2625 | return error; | |
2626 | ||
2627 | agi = XFS_BUF_TO_AGI(*bpp); | |
44b56e0a | 2628 | pag = xfs_perag_get(mp, agno); |
1da177e4 | 2629 | if (!pag->pagi_init) { |
16259e7d | 2630 | pag->pagi_freecount = be32_to_cpu(agi->agi_freecount); |
92821e2b | 2631 | pag->pagi_count = be32_to_cpu(agi->agi_count); |
1da177e4 | 2632 | pag->pagi_init = 1; |
1da177e4 | 2633 | } |
1da177e4 | 2634 | |
5e1be0fb CH |
2635 | /* |
2636 | * It's possible for these to be out of sync if | |
2637 | * we are in the middle of a forced shutdown. | |
2638 | */ | |
2639 | ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) || | |
2640 | XFS_FORCED_SHUTDOWN(mp)); | |
44b56e0a | 2641 | xfs_perag_put(pag); |
1da177e4 LT |
2642 | return 0; |
2643 | } | |
92821e2b DC |
2644 | |
2645 | /* | |
2646 | * Read in the agi to initialise the per-ag data in the mount structure | |
2647 | */ | |
2648 | int | |
2649 | xfs_ialloc_pagi_init( | |
2650 | xfs_mount_t *mp, /* file system mount structure */ | |
2651 | xfs_trans_t *tp, /* transaction pointer */ | |
2652 | xfs_agnumber_t agno) /* allocation group number */ | |
2653 | { | |
2654 | xfs_buf_t *bp = NULL; | |
2655 | int error; | |
2656 | ||
2657 | error = xfs_ialloc_read_agi(mp, tp, agno, &bp); | |
2658 | if (error) | |
2659 | return error; | |
2660 | if (bp) | |
2661 | xfs_trans_brelse(tp, bp); | |
2662 | return 0; | |
2663 | } | |
91fb9afc DW |
2664 | |
2665 | /* Calculate the first and last possible inode number in an AG. */ | |
2666 | void | |
2667 | xfs_ialloc_agino_range( | |
2668 | struct xfs_mount *mp, | |
2669 | xfs_agnumber_t agno, | |
2670 | xfs_agino_t *first, | |
2671 | xfs_agino_t *last) | |
2672 | { | |
2673 | xfs_agblock_t bno; | |
2674 | xfs_agblock_t eoag; | |
2675 | ||
2676 | eoag = xfs_ag_block_count(mp, agno); | |
2677 | ||
2678 | /* | |
2679 | * Calculate the first inode, which will be in the first | |
2680 | * cluster-aligned block after the AGFL. | |
2681 | */ | |
2682 | bno = round_up(XFS_AGFL_BLOCK(mp) + 1, | |
2683 | xfs_ialloc_cluster_alignment(mp)); | |
2684 | *first = XFS_OFFBNO_TO_AGINO(mp, bno, 0); | |
2685 | ||
2686 | /* | |
2687 | * Calculate the last inode, which will be at the end of the | |
2688 | * last (aligned) cluster that can be allocated in the AG. | |
2689 | */ | |
2690 | bno = round_down(eoag, xfs_ialloc_cluster_alignment(mp)); | |
2691 | *last = XFS_OFFBNO_TO_AGINO(mp, bno, 0) - 1; | |
2692 | } | |
2693 | ||
2694 | /* | |
2695 | * Verify that an AG inode number pointer neither points outside the AG | |
2696 | * nor points at static metadata. | |
2697 | */ | |
2698 | bool | |
2699 | xfs_verify_agino( | |
2700 | struct xfs_mount *mp, | |
2701 | xfs_agnumber_t agno, | |
2702 | xfs_agino_t agino) | |
2703 | { | |
2704 | xfs_agino_t first; | |
2705 | xfs_agino_t last; | |
2706 | ||
2707 | xfs_ialloc_agino_range(mp, agno, &first, &last); | |
2708 | return agino >= first && agino <= last; | |
2709 | } | |
2710 | ||
2711 | /* | |
2712 | * Verify that an FS inode number pointer neither points outside the | |
2713 | * filesystem nor points at static AG metadata. | |
2714 | */ | |
2715 | bool | |
2716 | xfs_verify_ino( | |
2717 | struct xfs_mount *mp, | |
2718 | xfs_ino_t ino) | |
2719 | { | |
2720 | xfs_agnumber_t agno = XFS_INO_TO_AGNO(mp, ino); | |
2721 | xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino); | |
2722 | ||
2723 | if (agno >= mp->m_sb.sb_agcount) | |
2724 | return false; | |
2725 | if (XFS_AGINO_TO_INO(mp, agno, agino) != ino) | |
2726 | return false; | |
2727 | return xfs_verify_agino(mp, agno, agino); | |
2728 | } | |
2729 | ||
2730 | /* Is this an internal inode number? */ | |
2731 | bool | |
2732 | xfs_internal_inum( | |
2733 | struct xfs_mount *mp, | |
2734 | xfs_ino_t ino) | |
2735 | { | |
2736 | return ino == mp->m_sb.sb_rbmino || ino == mp->m_sb.sb_rsumino || | |
2737 | (xfs_sb_version_hasquota(&mp->m_sb) && | |
2738 | xfs_is_quota_inode(&mp->m_sb, ino)); | |
2739 | } | |
2740 | ||
2741 | /* | |
2742 | * Verify that a directory entry's inode number doesn't point at an internal | |
2743 | * inode, empty space, or static AG metadata. | |
2744 | */ | |
2745 | bool | |
2746 | xfs_verify_dir_ino( | |
2747 | struct xfs_mount *mp, | |
2748 | xfs_ino_t ino) | |
2749 | { | |
2750 | if (xfs_internal_inum(mp, ino)) | |
2751 | return false; | |
2752 | return xfs_verify_ino(mp, ino); | |
2753 | } |