]>
Commit | Line | Data |
---|---|---|
2bd0ea18 | 1 | /* |
da23017d NS |
2 | * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
5000d01d | 4 | * |
da23017d NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
2bd0ea18 | 7 | * published by the Free Software Foundation. |
5000d01d | 8 | * |
da23017d NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
5000d01d | 13 | * |
da23017d NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
2bd0ea18 | 17 | */ |
9c799827 | 18 | #include "libxfs_priv.h" |
b626fb59 DC |
19 | #include "xfs_fs.h" |
20 | #include "xfs_shared.h" | |
21 | #include "xfs_format.h" | |
22 | #include "xfs_log_format.h" | |
23 | #include "xfs_trans_resv.h" | |
24 | #include "xfs_bit.h" | |
25 | #include "xfs_sb.h" | |
26 | #include "xfs_mount.h" | |
f944d3d0 | 27 | #include "xfs_defer.h" |
b626fb59 DC |
28 | #include "xfs_inode.h" |
29 | #include "xfs_btree.h" | |
30 | #include "xfs_ialloc.h" | |
31 | #include "xfs_ialloc_btree.h" | |
32 | #include "xfs_alloc.h" | |
33 | #include "xfs_bmap.h" | |
34 | #include "xfs_cksum.h" | |
35 | #include "xfs_trans.h" | |
36 | #include "xfs_trace.h" | |
85aec44f | 37 | #include "xfs_rmap.h" |
2bd0ea18 | 38 | |
2bd0ea18 NS |
39 | |
40 | /* | |
41 | * Allocation group level functions. | |
42 | */ | |
b5f6747c | 43 | int |
5e656dbb | 44 | xfs_ialloc_cluster_alignment( |
5a35bf2c | 45 | struct xfs_mount *mp) |
5e656dbb | 46 | { |
5a35bf2c | 47 | if (xfs_sb_version_hasalign(&mp->m_sb) && |
e32fb241 | 48 | mp->m_sb.sb_inoalignmt >= xfs_icluster_size_fsb(mp)) |
5a35bf2c | 49 | return mp->m_sb.sb_inoalignmt; |
5e656dbb BN |
50 | return 1; |
51 | } | |
2bd0ea18 | 52 | |
b194c7d8 | 53 | /* |
56b2de80 | 54 | * Lookup a record by ino in the btree given by cur. |
b194c7d8 BN |
55 | */ |
56 | int /* error */ | |
56b2de80 | 57 | xfs_inobt_lookup( |
b194c7d8 BN |
58 | struct xfs_btree_cur *cur, /* btree cursor */ |
59 | xfs_agino_t ino, /* starting inode of chunk */ | |
56b2de80 | 60 | xfs_lookup_t dir, /* <=, >=, == */ |
b194c7d8 BN |
61 | int *stat) /* success/failure */ |
62 | { | |
63 | cur->bc_rec.i.ir_startino = ino; | |
11640e30 BF |
64 | cur->bc_rec.i.ir_holemask = 0; |
65 | cur->bc_rec.i.ir_count = 0; | |
56b2de80 DC |
66 | cur->bc_rec.i.ir_freecount = 0; |
67 | cur->bc_rec.i.ir_free = 0; | |
68 | return xfs_btree_lookup(cur, dir, stat); | |
b194c7d8 BN |
69 | } |
70 | ||
71 | /* | |
56b2de80 | 72 | * Update the record referred to by cur to the value given. |
b194c7d8 BN |
73 | * This either works (return 0) or gets an EFSCORRUPTED error. |
74 | */ | |
75 | STATIC int /* error */ | |
76 | xfs_inobt_update( | |
77 | struct xfs_btree_cur *cur, /* btree cursor */ | |
56b2de80 | 78 | xfs_inobt_rec_incore_t *irec) /* btree record */ |
b194c7d8 BN |
79 | { |
80 | union xfs_btree_rec rec; | |
81 | ||
56b2de80 | 82 | rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino); |
11640e30 BF |
83 | if (xfs_sb_version_hassparseinodes(&cur->bc_mp->m_sb)) { |
84 | rec.inobt.ir_u.sp.ir_holemask = cpu_to_be16(irec->ir_holemask); | |
85 | rec.inobt.ir_u.sp.ir_count = irec->ir_count; | |
86 | rec.inobt.ir_u.sp.ir_freecount = irec->ir_freecount; | |
87 | } else { | |
88 | /* ir_holemask/ir_count not supported on-disk */ | |
89 | rec.inobt.ir_u.f.ir_freecount = cpu_to_be32(irec->ir_freecount); | |
90 | } | |
56b2de80 | 91 | rec.inobt.ir_free = cpu_to_be64(irec->ir_free); |
b194c7d8 BN |
92 | return xfs_btree_update(cur, &rec); |
93 | } | |
94 | ||
b5f6747c DW |
95 | /* Convert on-disk btree record to incore inobt record. */ |
96 | void | |
97 | xfs_inobt_btrec_to_irec( | |
98 | struct xfs_mount *mp, | |
99 | union xfs_btree_rec *rec, | |
100 | struct xfs_inobt_rec_incore *irec) | |
b194c7d8 | 101 | { |
11640e30 | 102 | irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino); |
b5f6747c | 103 | if (xfs_sb_version_hassparseinodes(&mp->m_sb)) { |
11640e30 BF |
104 | irec->ir_holemask = be16_to_cpu(rec->inobt.ir_u.sp.ir_holemask); |
105 | irec->ir_count = rec->inobt.ir_u.sp.ir_count; | |
106 | irec->ir_freecount = rec->inobt.ir_u.sp.ir_freecount; | |
107 | } else { | |
108 | /* | |
109 | * ir_holemask/ir_count not supported on-disk. Fill in hardcoded | |
110 | * values for full inode chunks. | |
111 | */ | |
112 | irec->ir_holemask = XFS_INOBT_HOLEMASK_FULL; | |
113 | irec->ir_count = XFS_INODES_PER_CHUNK; | |
114 | irec->ir_freecount = | |
115 | be32_to_cpu(rec->inobt.ir_u.f.ir_freecount); | |
b194c7d8 | 116 | } |
11640e30 | 117 | irec->ir_free = be64_to_cpu(rec->inobt.ir_free); |
b5f6747c DW |
118 | } |
119 | ||
120 | /* | |
121 | * Get the data from the pointed-to record. | |
122 | */ | |
123 | int | |
124 | xfs_inobt_get_rec( | |
125 | struct xfs_btree_cur *cur, | |
126 | struct xfs_inobt_rec_incore *irec, | |
127 | int *stat) | |
128 | { | |
129 | union xfs_btree_rec *rec; | |
130 | int error; | |
131 | ||
132 | error = xfs_btree_get_rec(cur, &rec, stat); | |
133 | if (error || *stat == 0) | |
134 | return error; | |
135 | ||
136 | xfs_inobt_btrec_to_irec(cur->bc_mp, rec, irec); | |
11640e30 BF |
137 | |
138 | return 0; | |
b194c7d8 BN |
139 | } |
140 | ||
3c699279 BF |
141 | /* |
142 | * Insert a single inobt record. Cursor must already point to desired location. | |
143 | */ | |
144 | STATIC int | |
145 | xfs_inobt_insert_rec( | |
146 | struct xfs_btree_cur *cur, | |
4a492e72 DW |
147 | uint16_t holemask, |
148 | uint8_t count, | |
149 | int32_t freecount, | |
3c699279 BF |
150 | xfs_inofree_t free, |
151 | int *stat) | |
152 | { | |
11640e30 BF |
153 | cur->bc_rec.i.ir_holemask = holemask; |
154 | cur->bc_rec.i.ir_count = count; | |
3c699279 BF |
155 | cur->bc_rec.i.ir_freecount = freecount; |
156 | cur->bc_rec.i.ir_free = free; | |
157 | return xfs_btree_insert(cur, stat); | |
158 | } | |
159 | ||
160 | /* | |
161 | * Insert records describing a newly allocated inode chunk into the inobt. | |
162 | */ | |
163 | STATIC int | |
164 | xfs_inobt_insert( | |
165 | struct xfs_mount *mp, | |
166 | struct xfs_trans *tp, | |
167 | struct xfs_buf *agbp, | |
168 | xfs_agino_t newino, | |
169 | xfs_agino_t newlen, | |
170 | xfs_btnum_t btnum) | |
171 | { | |
172 | struct xfs_btree_cur *cur; | |
173 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
174 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
175 | xfs_agino_t thisino; | |
176 | int i; | |
177 | int error; | |
178 | ||
179 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum); | |
180 | ||
181 | for (thisino = newino; | |
182 | thisino < newino + newlen; | |
183 | thisino += XFS_INODES_PER_CHUNK) { | |
184 | error = xfs_inobt_lookup(cur, thisino, XFS_LOOKUP_EQ, &i); | |
185 | if (error) { | |
186 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
187 | return error; | |
188 | } | |
189 | ASSERT(i == 0); | |
190 | ||
11640e30 BF |
191 | error = xfs_inobt_insert_rec(cur, XFS_INOBT_HOLEMASK_FULL, |
192 | XFS_INODES_PER_CHUNK, | |
193 | XFS_INODES_PER_CHUNK, | |
3c699279 BF |
194 | XFS_INOBT_ALL_FREE, &i); |
195 | if (error) { | |
196 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
197 | return error; | |
198 | } | |
199 | ASSERT(i == 1); | |
200 | } | |
201 | ||
202 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
203 | ||
204 | return 0; | |
205 | } | |
206 | ||
56b2de80 DC |
207 | /* |
208 | * Verify that the number of free inodes in the AGI is correct. | |
209 | */ | |
210 | #ifdef DEBUG | |
211 | STATIC int | |
212 | xfs_check_agi_freecount( | |
213 | struct xfs_btree_cur *cur, | |
214 | struct xfs_agi *agi) | |
215 | { | |
216 | if (cur->bc_nlevels == 1) { | |
217 | xfs_inobt_rec_incore_t rec; | |
218 | int freecount = 0; | |
219 | int error; | |
220 | int i; | |
221 | ||
222 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); | |
223 | if (error) | |
224 | return error; | |
225 | ||
226 | do { | |
227 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
228 | if (error) | |
229 | return error; | |
230 | ||
231 | if (i) { | |
232 | freecount += rec.ir_freecount; | |
233 | error = xfs_btree_increment(cur, 0, &i); | |
234 | if (error) | |
235 | return error; | |
236 | } | |
237 | } while (i == 1); | |
238 | ||
239 | if (!XFS_FORCED_SHUTDOWN(cur->bc_mp)) | |
240 | ASSERT(freecount == be32_to_cpu(agi->agi_freecount)); | |
241 | } | |
242 | return 0; | |
243 | } | |
244 | #else | |
245 | #define xfs_check_agi_freecount(cur, agi) 0 | |
246 | #endif | |
247 | ||
248 | /* | |
e9d35108 DC |
249 | * Initialise a new set of inodes. When called without a transaction context |
250 | * (e.g. from recovery) we initiate a delayed write of the inode buffers rather | |
251 | * than logging them (which in a transaction context puts them into the AIL | |
252 | * for writeback rather than the xfsbufd queue). | |
56b2de80 | 253 | */ |
e9d35108 | 254 | int |
56b2de80 DC |
255 | xfs_ialloc_inode_init( |
256 | struct xfs_mount *mp, | |
257 | struct xfs_trans *tp, | |
e9d35108 | 258 | struct list_head *buffer_list, |
fe8d48ac | 259 | int icount, |
56b2de80 DC |
260 | xfs_agnumber_t agno, |
261 | xfs_agblock_t agbno, | |
262 | xfs_agblock_t length, | |
263 | unsigned int gen) | |
264 | { | |
265 | struct xfs_buf *fbuf; | |
266 | struct xfs_dinode *free; | |
ff105f75 | 267 | int nbufs, blks_per_cluster, inodes_per_cluster; |
56b2de80 DC |
268 | int version; |
269 | int i, j; | |
270 | xfs_daddr_t d; | |
41ce5f36 | 271 | xfs_ino_t ino = 0; |
56b2de80 DC |
272 | |
273 | /* | |
ff105f75 DC |
274 | * Loop over the new block(s), filling in the inodes. For small block |
275 | * sizes, manipulate the inodes in buffers which are multiples of the | |
276 | * blocks size. | |
56b2de80 | 277 | */ |
ff105f75 DC |
278 | blks_per_cluster = xfs_icluster_size_fsb(mp); |
279 | inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog; | |
280 | nbufs = length / blks_per_cluster; | |
56b2de80 DC |
281 | |
282 | /* | |
e9d35108 DC |
283 | * Figure out what version number to use in the inodes we create. If |
284 | * the superblock version has caught up to the one that supports the new | |
285 | * inode format, then use the new inode version. Otherwise use the old | |
286 | * version so that old kernels will continue to be able to use the file | |
287 | * system. | |
41ce5f36 DC |
288 | * |
289 | * For v3 inodes, we also need to write the inode number into the inode, | |
290 | * so calculate the first inode number of the chunk here as | |
e9d35108 DC |
291 | * XFS_OFFBNO_TO_AGINO() only works within a filesystem block, not |
292 | * across multiple filesystem blocks (such as a cluster) and so cannot | |
293 | * be used in the cluster buffer loop below. | |
294 | * | |
295 | * Further, because we are writing the inode directly into the buffer | |
296 | * and calculating a CRC on the entire inode, we have ot log the entire | |
297 | * inode so that the entire range the CRC covers is present in the log. | |
298 | * That means for v3 inode we log the entire buffer rather than just the | |
299 | * inode cores. | |
56b2de80 | 300 | */ |
41ce5f36 DC |
301 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
302 | version = 3; | |
303 | ino = XFS_AGINO_TO_INO(mp, agno, | |
304 | XFS_OFFBNO_TO_AGINO(mp, agbno, 0)); | |
e9d35108 DC |
305 | |
306 | /* | |
307 | * log the initialisation that is about to take place as an | |
308 | * logical operation. This means the transaction does not | |
309 | * need to log the physical changes to the inode buffers as log | |
310 | * recovery will know what initialisation is actually needed. | |
311 | * Hence we only need to log the buffers as "ordered" buffers so | |
312 | * they track in the AIL as if they were physically logged. | |
313 | */ | |
314 | if (tp) | |
fe8d48ac | 315 | xfs_icreate_log(tp, agno, agbno, icount, |
e9d35108 | 316 | mp->m_sb.sb_inodesize, length, gen); |
ff105f75 | 317 | } else |
56b2de80 | 318 | version = 2; |
56b2de80 DC |
319 | |
320 | for (j = 0; j < nbufs; j++) { | |
321 | /* | |
322 | * Get the block. | |
323 | */ | |
324 | d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster)); | |
325 | fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, | |
326 | mp->m_bsize * blks_per_cluster, | |
a2ceac1f DC |
327 | XBF_UNMAPPED); |
328 | if (!fbuf) | |
12b53197 | 329 | return -ENOMEM; |
e9d35108 DC |
330 | |
331 | /* Initialize the inode buffers and log them appropriately. */ | |
a2ceac1f | 332 | fbuf->b_ops = &xfs_inode_buf_ops; |
e9d35108 | 333 | xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length)); |
ff105f75 | 334 | for (i = 0; i < inodes_per_cluster; i++) { |
56b2de80 | 335 | int ioffset = i << mp->m_sb.sb_inodelog; |
41ce5f36 | 336 | uint isize = xfs_dinode_size(version); |
56b2de80 DC |
337 | |
338 | free = xfs_make_iptr(mp, fbuf, i); | |
339 | free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC); | |
340 | free->di_version = version; | |
341 | free->di_gen = cpu_to_be32(gen); | |
342 | free->di_next_unlinked = cpu_to_be32(NULLAGINO); | |
41ce5f36 DC |
343 | |
344 | if (version == 3) { | |
345 | free->di_ino = cpu_to_be64(ino); | |
346 | ino++; | |
9c4e12fb ES |
347 | uuid_copy(&free->di_uuid, |
348 | &mp->m_sb.sb_meta_uuid); | |
41ce5f36 | 349 | xfs_dinode_calc_crc(mp, free); |
e9d35108 DC |
350 | } else if (tp) { |
351 | /* just log the inode core */ | |
352 | xfs_trans_log_buf(tp, fbuf, ioffset, | |
353 | ioffset + isize - 1); | |
41ce5f36 | 354 | } |
e9d35108 | 355 | } |
41ce5f36 | 356 | |
e9d35108 DC |
357 | if (tp) { |
358 | /* | |
359 | * Mark the buffer as an inode allocation buffer so it | |
360 | * sticks in AIL at the point of this allocation | |
361 | * transaction. This ensures the they are on disk before | |
362 | * the tail of the log can be moved past this | |
363 | * transaction (i.e. by preventing relogging from moving | |
364 | * it forward in the log). | |
365 | */ | |
366 | xfs_trans_inode_alloc_buf(tp, fbuf); | |
367 | if (version == 3) { | |
368 | /* | |
369 | * Mark the buffer as ordered so that they are | |
370 | * not physically logged in the transaction but | |
371 | * still tracked in the AIL as part of the | |
372 | * transaction and pin the log appropriately. | |
373 | */ | |
374 | xfs_trans_ordered_buf(tp, fbuf); | |
e9d35108 DC |
375 | } |
376 | } else { | |
377 | fbuf->b_flags |= XBF_DONE; | |
378 | xfs_buf_delwri_queue(fbuf, buffer_list); | |
379 | xfs_buf_relse(fbuf); | |
56b2de80 | 380 | } |
56b2de80 | 381 | } |
a2ceac1f | 382 | return 0; |
56b2de80 DC |
383 | } |
384 | ||
6f4c54a4 BF |
385 | /* |
386 | * Align startino and allocmask for a recently allocated sparse chunk such that | |
387 | * they are fit for insertion (or merge) into the on-disk inode btrees. | |
388 | * | |
389 | * Background: | |
390 | * | |
391 | * When enabled, sparse inode support increases the inode alignment from cluster | |
392 | * size to inode chunk size. This means that the minimum range between two | |
393 | * non-adjacent inode records in the inobt is large enough for a full inode | |
394 | * record. This allows for cluster sized, cluster aligned block allocation | |
395 | * without need to worry about whether the resulting inode record overlaps with | |
396 | * another record in the tree. Without this basic rule, we would have to deal | |
397 | * with the consequences of overlap by potentially undoing recent allocations in | |
398 | * the inode allocation codepath. | |
399 | * | |
400 | * Because of this alignment rule (which is enforced on mount), there are two | |
401 | * inobt possibilities for newly allocated sparse chunks. One is that the | |
402 | * aligned inode record for the chunk covers a range of inodes not already | |
403 | * covered in the inobt (i.e., it is safe to insert a new sparse record). The | |
404 | * other is that a record already exists at the aligned startino that considers | |
405 | * the newly allocated range as sparse. In the latter case, record content is | |
406 | * merged in hope that sparse inode chunks fill to full chunks over time. | |
407 | */ | |
408 | STATIC void | |
409 | xfs_align_sparse_ino( | |
410 | struct xfs_mount *mp, | |
411 | xfs_agino_t *startino, | |
412 | uint16_t *allocmask) | |
413 | { | |
414 | xfs_agblock_t agbno; | |
415 | xfs_agblock_t mod; | |
416 | int offset; | |
417 | ||
418 | agbno = XFS_AGINO_TO_AGBNO(mp, *startino); | |
419 | mod = agbno % mp->m_sb.sb_inoalignmt; | |
420 | if (!mod) | |
421 | return; | |
422 | ||
423 | /* calculate the inode offset and align startino */ | |
424 | offset = mod << mp->m_sb.sb_inopblog; | |
425 | *startino -= offset; | |
426 | ||
427 | /* | |
428 | * Since startino has been aligned down, left shift allocmask such that | |
429 | * it continues to represent the same physical inodes relative to the | |
430 | * new startino. | |
431 | */ | |
432 | *allocmask <<= offset / XFS_INODES_PER_HOLEMASK_BIT; | |
433 | } | |
434 | ||
435 | /* | |
436 | * Determine whether the source inode record can merge into the target. Both | |
437 | * records must be sparse, the inode ranges must match and there must be no | |
438 | * allocation overlap between the records. | |
439 | */ | |
440 | STATIC bool | |
441 | __xfs_inobt_can_merge( | |
442 | struct xfs_inobt_rec_incore *trec, /* tgt record */ | |
443 | struct xfs_inobt_rec_incore *srec) /* src record */ | |
444 | { | |
445 | uint64_t talloc; | |
446 | uint64_t salloc; | |
447 | ||
448 | /* records must cover the same inode range */ | |
449 | if (trec->ir_startino != srec->ir_startino) | |
450 | return false; | |
451 | ||
452 | /* both records must be sparse */ | |
453 | if (!xfs_inobt_issparse(trec->ir_holemask) || | |
454 | !xfs_inobt_issparse(srec->ir_holemask)) | |
455 | return false; | |
456 | ||
457 | /* both records must track some inodes */ | |
458 | if (!trec->ir_count || !srec->ir_count) | |
459 | return false; | |
460 | ||
461 | /* can't exceed capacity of a full record */ | |
462 | if (trec->ir_count + srec->ir_count > XFS_INODES_PER_CHUNK) | |
463 | return false; | |
464 | ||
465 | /* verify there is no allocation overlap */ | |
466 | talloc = xfs_inobt_irec_to_allocmask(trec); | |
467 | salloc = xfs_inobt_irec_to_allocmask(srec); | |
468 | if (talloc & salloc) | |
469 | return false; | |
470 | ||
471 | return true; | |
472 | } | |
473 | ||
474 | /* | |
475 | * Merge the source inode record into the target. The caller must call | |
476 | * __xfs_inobt_can_merge() to ensure the merge is valid. | |
477 | */ | |
478 | STATIC void | |
479 | __xfs_inobt_rec_merge( | |
480 | struct xfs_inobt_rec_incore *trec, /* target */ | |
481 | struct xfs_inobt_rec_incore *srec) /* src */ | |
482 | { | |
483 | ASSERT(trec->ir_startino == srec->ir_startino); | |
484 | ||
485 | /* combine the counts */ | |
486 | trec->ir_count += srec->ir_count; | |
487 | trec->ir_freecount += srec->ir_freecount; | |
488 | ||
489 | /* | |
490 | * Merge the holemask and free mask. For both fields, 0 bits refer to | |
491 | * allocated inodes. We combine the allocated ranges with bitwise AND. | |
492 | */ | |
493 | trec->ir_holemask &= srec->ir_holemask; | |
494 | trec->ir_free &= srec->ir_free; | |
495 | } | |
496 | ||
497 | /* | |
498 | * Insert a new sparse inode chunk into the associated inode btree. The inode | |
499 | * record for the sparse chunk is pre-aligned to a startino that should match | |
500 | * any pre-existing sparse inode record in the tree. This allows sparse chunks | |
501 | * to fill over time. | |
502 | * | |
503 | * This function supports two modes of handling preexisting records depending on | |
504 | * the merge flag. If merge is true, the provided record is merged with the | |
505 | * existing record and updated in place. The merged record is returned in nrec. | |
506 | * If merge is false, an existing record is replaced with the provided record. | |
507 | * If no preexisting record exists, the provided record is always inserted. | |
508 | * | |
509 | * It is considered corruption if a merge is requested and not possible. Given | |
510 | * the sparse inode alignment constraints, this should never happen. | |
511 | */ | |
512 | STATIC int | |
513 | xfs_inobt_insert_sprec( | |
514 | struct xfs_mount *mp, | |
515 | struct xfs_trans *tp, | |
516 | struct xfs_buf *agbp, | |
517 | int btnum, | |
518 | struct xfs_inobt_rec_incore *nrec, /* in/out: new/merged rec. */ | |
519 | bool merge) /* merge or replace */ | |
520 | { | |
521 | struct xfs_btree_cur *cur; | |
522 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
523 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
524 | int error; | |
525 | int i; | |
526 | struct xfs_inobt_rec_incore rec; | |
527 | ||
528 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum); | |
529 | ||
530 | /* the new record is pre-aligned so we know where to look */ | |
531 | error = xfs_inobt_lookup(cur, nrec->ir_startino, XFS_LOOKUP_EQ, &i); | |
532 | if (error) | |
533 | goto error; | |
534 | /* if nothing there, insert a new record and return */ | |
535 | if (i == 0) { | |
536 | error = xfs_inobt_insert_rec(cur, nrec->ir_holemask, | |
537 | nrec->ir_count, nrec->ir_freecount, | |
538 | nrec->ir_free, &i); | |
539 | if (error) | |
540 | goto error; | |
541 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error); | |
542 | ||
543 | goto out; | |
544 | } | |
545 | ||
546 | /* | |
547 | * A record exists at this startino. Merge or replace the record | |
548 | * depending on what we've been asked to do. | |
549 | */ | |
550 | if (merge) { | |
551 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
552 | if (error) | |
553 | goto error; | |
554 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error); | |
555 | XFS_WANT_CORRUPTED_GOTO(mp, | |
556 | rec.ir_startino == nrec->ir_startino, | |
557 | error); | |
558 | ||
559 | /* | |
560 | * This should never fail. If we have coexisting records that | |
561 | * cannot merge, something is seriously wrong. | |
562 | */ | |
563 | XFS_WANT_CORRUPTED_GOTO(mp, __xfs_inobt_can_merge(nrec, &rec), | |
564 | error); | |
565 | ||
566 | trace_xfs_irec_merge_pre(mp, agno, rec.ir_startino, | |
567 | rec.ir_holemask, nrec->ir_startino, | |
568 | nrec->ir_holemask); | |
569 | ||
570 | /* merge to nrec to output the updated record */ | |
571 | __xfs_inobt_rec_merge(nrec, &rec); | |
572 | ||
573 | trace_xfs_irec_merge_post(mp, agno, nrec->ir_startino, | |
574 | nrec->ir_holemask); | |
575 | ||
576 | error = xfs_inobt_rec_check_count(mp, nrec); | |
577 | if (error) | |
578 | goto error; | |
579 | } | |
580 | ||
581 | error = xfs_inobt_update(cur, nrec); | |
582 | if (error) | |
583 | goto error; | |
584 | ||
585 | out: | |
586 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
587 | return 0; | |
588 | error: | |
589 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
590 | return error; | |
591 | } | |
592 | ||
2bd0ea18 NS |
593 | /* |
594 | * Allocate new inodes in the allocation group specified by agbp. | |
595 | * Return 0 for success, else error code. | |
596 | */ | |
597 | STATIC int /* error code or 0 */ | |
598 | xfs_ialloc_ag_alloc( | |
599 | xfs_trans_t *tp, /* transaction pointer */ | |
600 | xfs_buf_t *agbp, /* alloc group buffer */ | |
601 | int *alloc) | |
602 | { | |
603 | xfs_agi_t *agi; /* allocation group header */ | |
dfc130f3 | 604 | xfs_alloc_arg_t args; /* allocation argument structure */ |
5e656dbb | 605 | xfs_agnumber_t agno; |
2bd0ea18 | 606 | int error; |
2bd0ea18 NS |
607 | xfs_agino_t newino; /* new first inode's number */ |
608 | xfs_agino_t newlen; /* new number of inodes */ | |
5e656dbb | 609 | int isaligned = 0; /* inode allocation at stripe unit */ |
2bd0ea18 | 610 | /* boundary */ |
6f4c54a4 BF |
611 | uint16_t allocmask = (uint16_t) -1; /* init. to full chunk */ |
612 | struct xfs_inobt_rec_incore rec; | |
56b2de80 | 613 | struct xfs_perag *pag; |
c9005f41 BF |
614 | int do_sparse = 0; |
615 | ||
a2ceac1f | 616 | memset(&args, 0, sizeof(args)); |
2bd0ea18 NS |
617 | args.tp = tp; |
618 | args.mp = tp->t_mountp; | |
c9005f41 | 619 | args.fsbno = NULLFSBLOCK; |
85aec44f | 620 | xfs_rmap_ag_owner(&args.oinfo, XFS_RMAP_OWN_INODES); |
2bd0ea18 | 621 | |
0d995d03 BF |
622 | #ifdef DEBUG |
623 | /* randomly do sparse inode allocations */ | |
624 | if (xfs_sb_version_hassparseinodes(&tp->t_mountp->m_sb) && | |
625 | args.mp->m_ialloc_min_blks < args.mp->m_ialloc_blks) | |
626 | do_sparse = prandom_u32() & 1; | |
627 | #endif | |
628 | ||
2bd0ea18 NS |
629 | /* |
630 | * Locking will ensure that we don't have two callers in here | |
631 | * at one time. | |
632 | */ | |
ff105f75 | 633 | newlen = args.mp->m_ialloc_inos; |
2bd0ea18 | 634 | if (args.mp->m_maxicount && |
cbf3beaa | 635 | percpu_counter_read_positive(&args.mp->m_icount) + newlen > |
19ebedcf | 636 | args.mp->m_maxicount) |
12b53197 | 637 | return -ENOSPC; |
ff105f75 | 638 | args.minlen = args.maxlen = args.mp->m_ialloc_blks; |
2bd0ea18 | 639 | /* |
5e656dbb BN |
640 | * First try to allocate inodes contiguous with the last-allocated |
641 | * chunk of inodes. If the filesystem is striped, this will fill | |
642 | * an entire stripe unit with inodes. | |
3439d03a | 643 | */ |
2bd0ea18 | 644 | agi = XFS_BUF_TO_AGI(agbp); |
5e656dbb | 645 | newino = be32_to_cpu(agi->agi_newino); |
56b2de80 | 646 | agno = be32_to_cpu(agi->agi_seqno); |
5e656dbb | 647 | args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) + |
ff105f75 | 648 | args.mp->m_ialloc_blks; |
c9005f41 BF |
649 | if (do_sparse) |
650 | goto sparse_alloc; | |
5e656dbb BN |
651 | if (likely(newino != NULLAGINO && |
652 | (args.agbno < be32_to_cpu(agi->agi_length)))) { | |
56b2de80 | 653 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
5e656dbb | 654 | args.type = XFS_ALLOCTYPE_THIS_BNO; |
5e656dbb BN |
655 | args.prod = 1; |
656 | ||
657 | /* | |
658 | * We need to take into account alignment here to ensure that | |
659 | * we don't modify the free list if we fail to have an exact | |
660 | * block. If we don't have an exact match, and every oher | |
661 | * attempt allocation attempt fails, we'll end up cancelling | |
662 | * a dirty transaction and shutting down. | |
663 | * | |
664 | * For an exact allocation, alignment must be 1, | |
665 | * however we need to take cluster alignment into account when | |
666 | * fixing up the freelist. Use the minalignslop field to | |
667 | * indicate that extra blocks might be required for alignment, | |
668 | * but not to use them in the actual exact allocation. | |
669 | */ | |
670 | args.alignment = 1; | |
5a35bf2c | 671 | args.minalignslop = xfs_ialloc_cluster_alignment(args.mp) - 1; |
5e656dbb BN |
672 | |
673 | /* Allow space for the inode btree to split. */ | |
56b2de80 | 674 | args.minleft = args.mp->m_in_maxlevels - 1; |
5e656dbb BN |
675 | if ((error = xfs_alloc_vextent(&args))) |
676 | return error; | |
ff105f75 DC |
677 | |
678 | /* | |
679 | * This request might have dirtied the transaction if the AG can | |
680 | * satisfy the request, but the exact block was not available. | |
681 | * If the allocation did fail, subsequent requests will relax | |
682 | * the exact agbno requirement and increase the alignment | |
683 | * instead. It is critical that the total size of the request | |
684 | * (len + alignment + slop) does not increase from this point | |
685 | * on, so reset minalignslop to ensure it is not included in | |
686 | * subsequent requests. | |
687 | */ | |
688 | args.minalignslop = 0; | |
c9005f41 | 689 | } |
5e656dbb BN |
690 | |
691 | if (unlikely(args.fsbno == NULLFSBLOCK)) { | |
692 | /* | |
693 | * Set the alignment for the allocation. | |
694 | * If stripe alignment is turned on then align at stripe unit | |
695 | * boundary. | |
696 | * If the cluster size is smaller than a filesystem block | |
697 | * then we're doing I/O for inodes in filesystem block size | |
698 | * pieces, so don't need alignment anyway. | |
699 | */ | |
700 | isaligned = 0; | |
701 | if (args.mp->m_sinoalign) { | |
702 | ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN)); | |
703 | args.alignment = args.mp->m_dalign; | |
704 | isaligned = 1; | |
705 | } else | |
5a35bf2c | 706 | args.alignment = xfs_ialloc_cluster_alignment(args.mp); |
5e656dbb BN |
707 | /* |
708 | * Need to figure out where to allocate the inode blocks. | |
709 | * Ideally they should be spaced out through the a.g. | |
710 | * For now, just allocate blocks up front. | |
711 | */ | |
712 | args.agbno = be32_to_cpu(agi->agi_root); | |
56b2de80 | 713 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
5e656dbb BN |
714 | /* |
715 | * Allocate a fixed-size extent of inodes. | |
716 | */ | |
717 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
5e656dbb BN |
718 | args.prod = 1; |
719 | /* | |
720 | * Allow space for the inode btree to split. | |
721 | */ | |
56b2de80 | 722 | args.minleft = args.mp->m_in_maxlevels - 1; |
5e656dbb BN |
723 | if ((error = xfs_alloc_vextent(&args))) |
724 | return error; | |
725 | } | |
2bd0ea18 NS |
726 | |
727 | /* | |
728 | * If stripe alignment is turned on, then try again with cluster | |
729 | * alignment. | |
730 | */ | |
731 | if (isaligned && args.fsbno == NULLFSBLOCK) { | |
732 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
6e3140c7 | 733 | args.agbno = be32_to_cpu(agi->agi_root); |
56b2de80 | 734 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
5a35bf2c | 735 | args.alignment = xfs_ialloc_cluster_alignment(args.mp); |
0e266570 | 736 | if ((error = xfs_alloc_vextent(&args))) |
dfc130f3 | 737 | return error; |
2bd0ea18 | 738 | } |
5000d01d | 739 | |
6f4c54a4 BF |
740 | /* |
741 | * Finally, try a sparse allocation if the filesystem supports it and | |
742 | * the sparse allocation length is smaller than a full chunk. | |
743 | */ | |
744 | if (xfs_sb_version_hassparseinodes(&args.mp->m_sb) && | |
745 | args.mp->m_ialloc_min_blks < args.mp->m_ialloc_blks && | |
746 | args.fsbno == NULLFSBLOCK) { | |
c9005f41 | 747 | sparse_alloc: |
6f4c54a4 BF |
748 | args.type = XFS_ALLOCTYPE_NEAR_BNO; |
749 | args.agbno = be32_to_cpu(agi->agi_root); | |
750 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); | |
751 | args.alignment = args.mp->m_sb.sb_spino_align; | |
752 | args.prod = 1; | |
753 | ||
754 | args.minlen = args.mp->m_ialloc_min_blks; | |
755 | args.maxlen = args.minlen; | |
756 | ||
757 | /* | |
758 | * The inode record will be aligned to full chunk size. We must | |
759 | * prevent sparse allocation from AG boundaries that result in | |
760 | * invalid inode records, such as records that start at agbno 0 | |
761 | * or extend beyond the AG. | |
762 | * | |
763 | * Set min agbno to the first aligned, non-zero agbno and max to | |
764 | * the last aligned agbno that is at least one full chunk from | |
765 | * the end of the AG. | |
766 | */ | |
767 | args.min_agbno = args.mp->m_sb.sb_inoalignmt; | |
768 | args.max_agbno = round_down(args.mp->m_sb.sb_agblocks, | |
769 | args.mp->m_sb.sb_inoalignmt) - | |
770 | args.mp->m_ialloc_blks; | |
771 | ||
772 | error = xfs_alloc_vextent(&args); | |
773 | if (error) | |
774 | return error; | |
775 | ||
776 | newlen = args.len << args.mp->m_sb.sb_inopblog; | |
0d995d03 | 777 | ASSERT(newlen <= XFS_INODES_PER_CHUNK); |
6f4c54a4 BF |
778 | allocmask = (1 << (newlen / XFS_INODES_PER_HOLEMASK_BIT)) - 1; |
779 | } | |
780 | ||
2bd0ea18 NS |
781 | if (args.fsbno == NULLFSBLOCK) { |
782 | *alloc = 0; | |
783 | return 0; | |
784 | } | |
785 | ASSERT(args.len == args.minlen); | |
a562a63b | 786 | |
5e656dbb | 787 | /* |
56b2de80 DC |
788 | * Stamp and write the inode buffers. |
789 | * | |
5e656dbb BN |
790 | * Seed the new inode cluster with a random generation number. This |
791 | * prevents short-term reuse of generation numbers if a chunk is | |
792 | * freed and then immediately reallocated. We use random numbers | |
793 | * rather than a linear progression to prevent the next generation | |
794 | * number from being easily guessable. | |
795 | */ | |
fe8d48ac BF |
796 | error = xfs_ialloc_inode_init(args.mp, tp, NULL, newlen, agno, |
797 | args.agbno, args.len, prandom_u32()); | |
56b2de80 | 798 | |
a2ceac1f DC |
799 | if (error) |
800 | return error; | |
56b2de80 DC |
801 | /* |
802 | * Convert the results. | |
803 | */ | |
804 | newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0); | |
6f4c54a4 BF |
805 | |
806 | if (xfs_inobt_issparse(~allocmask)) { | |
807 | /* | |
808 | * We've allocated a sparse chunk. Align the startino and mask. | |
809 | */ | |
810 | xfs_align_sparse_ino(args.mp, &newino, &allocmask); | |
811 | ||
812 | rec.ir_startino = newino; | |
813 | rec.ir_holemask = ~allocmask; | |
814 | rec.ir_count = newlen; | |
815 | rec.ir_freecount = newlen; | |
816 | rec.ir_free = XFS_INOBT_ALL_FREE; | |
817 | ||
818 | /* | |
819 | * Insert the sparse record into the inobt and allow for a merge | |
820 | * if necessary. If a merge does occur, rec is updated to the | |
821 | * merged record. | |
822 | */ | |
823 | error = xfs_inobt_insert_sprec(args.mp, tp, agbp, XFS_BTNUM_INO, | |
824 | &rec, true); | |
825 | if (error == -EFSCORRUPTED) { | |
826 | xfs_alert(args.mp, | |
827 | "invalid sparse inode record: ino 0x%llx holemask 0x%x count %u", | |
828 | XFS_AGINO_TO_INO(args.mp, agno, | |
829 | rec.ir_startino), | |
830 | rec.ir_holemask, rec.ir_count); | |
831 | xfs_force_shutdown(args.mp, SHUTDOWN_CORRUPT_INCORE); | |
832 | } | |
833 | if (error) | |
834 | return error; | |
835 | ||
836 | /* | |
837 | * We can't merge the part we've just allocated as for the inobt | |
838 | * due to finobt semantics. The original record may or may not | |
839 | * exist independent of whether physical inodes exist in this | |
840 | * sparse chunk. | |
841 | * | |
842 | * We must update the finobt record based on the inobt record. | |
843 | * rec contains the fully merged and up to date inobt record | |
844 | * from the previous call. Set merge false to replace any | |
845 | * existing record with this one. | |
846 | */ | |
847 | if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) { | |
848 | error = xfs_inobt_insert_sprec(args.mp, tp, agbp, | |
849 | XFS_BTNUM_FINO, &rec, | |
850 | false); | |
851 | if (error) | |
852 | return error; | |
853 | } | |
854 | } else { | |
855 | /* full chunk - insert new records to both btrees */ | |
856 | error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen, | |
857 | XFS_BTNUM_INO); | |
858 | if (error) | |
859 | return error; | |
860 | ||
861 | if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) { | |
862 | error = xfs_inobt_insert(args.mp, tp, agbp, newino, | |
863 | newlen, XFS_BTNUM_FINO); | |
864 | if (error) | |
865 | return error; | |
866 | } | |
867 | } | |
868 | ||
869 | /* | |
870 | * Update AGI counts and newino. | |
871 | */ | |
5e656dbb BN |
872 | be32_add_cpu(&agi->agi_count, newlen); |
873 | be32_add_cpu(&agi->agi_freecount, newlen); | |
56b2de80 DC |
874 | pag = xfs_perag_get(args.mp, agno); |
875 | pag->pagi_freecount += newlen; | |
876 | xfs_perag_put(pag); | |
6e3140c7 | 877 | agi->agi_newino = cpu_to_be32(newino); |
56b2de80 | 878 | |
2bd0ea18 NS |
879 | /* |
880 | * Log allocation group header fields | |
881 | */ | |
882 | xfs_ialloc_log_agi(tp, agbp, | |
883 | XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO); | |
884 | /* | |
885 | * Modify/log superblock values for inode count and inode free count. | |
886 | */ | |
887 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen); | |
888 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen); | |
889 | *alloc = 1; | |
890 | return 0; | |
891 | } | |
892 | ||
56b2de80 | 893 | STATIC xfs_agnumber_t |
321717ae NS |
894 | xfs_ialloc_next_ag( |
895 | xfs_mount_t *mp) | |
896 | { | |
897 | xfs_agnumber_t agno; | |
898 | ||
899 | spin_lock(&mp->m_agirotor_lock); | |
900 | agno = mp->m_agirotor; | |
a2ceac1f | 901 | if (++mp->m_agirotor >= mp->m_maxagi) |
321717ae NS |
902 | mp->m_agirotor = 0; |
903 | spin_unlock(&mp->m_agirotor_lock); | |
904 | ||
905 | return agno; | |
906 | } | |
907 | ||
2bd0ea18 NS |
908 | /* |
909 | * Select an allocation group to look for a free inode in, based on the parent | |
e6d77a21 | 910 | * inode and the mode. Return the allocation group buffer. |
2bd0ea18 | 911 | */ |
a2ceac1f | 912 | STATIC xfs_agnumber_t |
2bd0ea18 NS |
913 | xfs_ialloc_ag_select( |
914 | xfs_trans_t *tp, /* transaction pointer */ | |
915 | xfs_ino_t parent, /* parent directory inode number */ | |
a2ceac1f | 916 | umode_t mode, /* bits set to indicate file type */ |
2bd0ea18 NS |
917 | int okalloc) /* ok to allocate more space */ |
918 | { | |
2bd0ea18 NS |
919 | xfs_agnumber_t agcount; /* number of ag's in the filesystem */ |
920 | xfs_agnumber_t agno; /* current ag number */ | |
921 | int flags; /* alloc buffer locking flags */ | |
922 | xfs_extlen_t ineed; /* blocks needed for inode allocation */ | |
275ae71f | 923 | xfs_extlen_t longest = 0; /* longest extent available */ |
2bd0ea18 NS |
924 | xfs_mount_t *mp; /* mount point structure */ |
925 | int needspace; /* file mode implies space allocated */ | |
926 | xfs_perag_t *pag; /* per allocation group data */ | |
927 | xfs_agnumber_t pagno; /* parent (starting) ag number */ | |
a2ceac1f | 928 | int error; |
2bd0ea18 NS |
929 | |
930 | /* | |
931 | * Files of these types need at least one block if length > 0 | |
932 | * (and they won't fit in the inode, but that's hard to figure out). | |
933 | */ | |
934 | needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode); | |
935 | mp = tp->t_mountp; | |
34317449 | 936 | agcount = mp->m_maxagi; |
2bd0ea18 | 937 | if (S_ISDIR(mode)) |
321717ae | 938 | pagno = xfs_ialloc_next_ag(mp); |
5ce1d1f7 | 939 | else { |
2bd0ea18 | 940 | pagno = XFS_INO_TO_AGNO(mp, parent); |
5ce1d1f7 NS |
941 | if (pagno >= agcount) |
942 | pagno = 0; | |
943 | } | |
a2ceac1f | 944 | |
2bd0ea18 | 945 | ASSERT(pagno < agcount); |
a2ceac1f | 946 | |
2bd0ea18 NS |
947 | /* |
948 | * Loop through allocation groups, looking for one with a little | |
949 | * free space in it. Note we don't look for free inodes, exactly. | |
950 | * Instead, we include whether there is a need to allocate inodes | |
5000d01d | 951 | * to mean that blocks must be allocated for them, |
2bd0ea18 NS |
952 | * if none are currently free. |
953 | */ | |
954 | agno = pagno; | |
955 | flags = XFS_ALLOC_FLAG_TRYLOCK; | |
956 | for (;;) { | |
56b2de80 | 957 | pag = xfs_perag_get(mp, agno); |
a2ceac1f DC |
958 | if (!pag->pagi_inodeok) { |
959 | xfs_ialloc_next_ag(mp); | |
960 | goto nextag; | |
961 | } | |
962 | ||
2bd0ea18 | 963 | if (!pag->pagi_init) { |
a2ceac1f DC |
964 | error = xfs_ialloc_pagi_init(mp, tp, agno); |
965 | if (error) | |
2bd0ea18 | 966 | goto nextag; |
a2ceac1f | 967 | } |
34317449 | 968 | |
a2ceac1f DC |
969 | if (pag->pagi_freecount) { |
970 | xfs_perag_put(pag); | |
971 | return agno; | |
34317449 NS |
972 | } |
973 | ||
a2ceac1f DC |
974 | if (!okalloc) |
975 | goto nextag; | |
976 | ||
977 | if (!pag->pagf_init) { | |
978 | error = xfs_alloc_pagf_init(mp, tp, agno, flags); | |
979 | if (error) | |
2bd0ea18 | 980 | goto nextag; |
2bd0ea18 | 981 | } |
a2ceac1f DC |
982 | |
983 | /* | |
5a35bf2c DC |
984 | * Check that there is enough free space for the file plus a |
985 | * chunk of inodes if we need to allocate some. If this is the | |
986 | * first pass across the AGs, take into account the potential | |
987 | * space needed for alignment of inode chunks when checking the | |
988 | * longest contiguous free space in the AG - this prevents us | |
989 | * from getting ENOSPC because we have free space larger than | |
990 | * m_ialloc_blks but alignment constraints prevent us from using | |
991 | * it. | |
992 | * | |
993 | * If we can't find an AG with space for full alignment slack to | |
994 | * be taken into account, we must be near ENOSPC in all AGs. | |
995 | * Hence we don't include alignment for the second pass and so | |
996 | * if we fail allocation due to alignment issues then it is most | |
997 | * likely a real ENOSPC condition. | |
a2ceac1f | 998 | */ |
62dc6cdb | 999 | ineed = mp->m_ialloc_min_blks; |
5a35bf2c DC |
1000 | if (flags && ineed > 1) |
1001 | ineed += xfs_ialloc_cluster_alignment(mp); | |
a2ceac1f DC |
1002 | longest = pag->pagf_longest; |
1003 | if (!longest) | |
1004 | longest = pag->pagf_flcount > 0; | |
1005 | ||
1006 | if (pag->pagf_freeblks >= needspace + ineed && | |
1007 | longest >= ineed) { | |
1008 | xfs_perag_put(pag); | |
1009 | return agno; | |
2bd0ea18 | 1010 | } |
5000d01d | 1011 | nextag: |
56b2de80 | 1012 | xfs_perag_put(pag); |
5000d01d | 1013 | /* |
2bd0ea18 NS |
1014 | * No point in iterating over the rest, if we're shutting |
1015 | * down. | |
1016 | */ | |
56b2de80 | 1017 | if (XFS_FORCED_SHUTDOWN(mp)) |
a2ceac1f | 1018 | return NULLAGNUMBER; |
2bd0ea18 | 1019 | agno++; |
5ce1d1f7 | 1020 | if (agno >= agcount) |
2bd0ea18 NS |
1021 | agno = 0; |
1022 | if (agno == pagno) { | |
56b2de80 | 1023 | if (flags == 0) |
a2ceac1f | 1024 | return NULLAGNUMBER; |
2bd0ea18 NS |
1025 | flags = 0; |
1026 | } | |
1027 | } | |
1028 | } | |
1029 | ||
56b2de80 DC |
1030 | /* |
1031 | * Try to retrieve the next record to the left/right from the current one. | |
1032 | */ | |
1033 | STATIC int | |
1034 | xfs_ialloc_next_rec( | |
1035 | struct xfs_btree_cur *cur, | |
1036 | xfs_inobt_rec_incore_t *rec, | |
1037 | int *done, | |
1038 | int left) | |
1039 | { | |
1040 | int error; | |
1041 | int i; | |
1042 | ||
1043 | if (left) | |
1044 | error = xfs_btree_decrement(cur, 0, &i); | |
1045 | else | |
1046 | error = xfs_btree_increment(cur, 0, &i); | |
1047 | ||
1048 | if (error) | |
1049 | return error; | |
1050 | *done = !i; | |
1051 | if (i) { | |
1052 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1053 | if (error) | |
1054 | return error; | |
19ebedcf | 1055 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
56b2de80 DC |
1056 | } |
1057 | ||
1058 | return 0; | |
1059 | } | |
1060 | ||
1061 | STATIC int | |
1062 | xfs_ialloc_get_rec( | |
1063 | struct xfs_btree_cur *cur, | |
1064 | xfs_agino_t agino, | |
1065 | xfs_inobt_rec_incore_t *rec, | |
3439d03a | 1066 | int *done) |
56b2de80 DC |
1067 | { |
1068 | int error; | |
1069 | int i; | |
1070 | ||
1071 | error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i); | |
1072 | if (error) | |
1073 | return error; | |
1074 | *done = !i; | |
1075 | if (i) { | |
1076 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1077 | if (error) | |
1078 | return error; | |
19ebedcf | 1079 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
56b2de80 DC |
1080 | } |
1081 | ||
1082 | return 0; | |
1083 | } | |
1084 | ||
01792d3b | 1085 | /* |
fa58183c BF |
1086 | * Return the offset of the first free inode in the record. If the inode chunk |
1087 | * is sparsely allocated, we convert the record holemask to inode granularity | |
1088 | * and mask off the unallocated regions from the inode free mask. | |
01792d3b BF |
1089 | */ |
1090 | STATIC int | |
1091 | xfs_inobt_first_free_inode( | |
1092 | struct xfs_inobt_rec_incore *rec) | |
1093 | { | |
fa58183c BF |
1094 | xfs_inofree_t realfree; |
1095 | ||
1096 | /* if there are no holes, return the first available offset */ | |
1097 | if (!xfs_inobt_issparse(rec->ir_holemask)) | |
1098 | return xfs_lowbit64(rec->ir_free); | |
1099 | ||
1100 | realfree = xfs_inobt_irec_to_allocmask(rec); | |
1101 | realfree &= rec->ir_free; | |
1102 | ||
1103 | return xfs_lowbit64(realfree); | |
01792d3b BF |
1104 | } |
1105 | ||
5000d01d | 1106 | /* |
ff105f75 | 1107 | * Allocate an inode using the inobt-only algorithm. |
2bd0ea18 | 1108 | */ |
a2ceac1f | 1109 | STATIC int |
ff105f75 | 1110 | xfs_dialloc_ag_inobt( |
a2ceac1f DC |
1111 | struct xfs_trans *tp, |
1112 | struct xfs_buf *agbp, | |
1113 | xfs_ino_t parent, | |
1114 | xfs_ino_t *inop) | |
2bd0ea18 | 1115 | { |
a2ceac1f DC |
1116 | struct xfs_mount *mp = tp->t_mountp; |
1117 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
1118 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
1119 | xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent); | |
1120 | xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent); | |
1121 | struct xfs_perag *pag; | |
1122 | struct xfs_btree_cur *cur, *tcur; | |
1123 | struct xfs_inobt_rec_incore rec, trec; | |
1124 | xfs_ino_t ino; | |
1125 | int error; | |
1126 | int offset; | |
1127 | int i, j; | |
d8c47890 | 1128 | int searchdistance = 10; |
2bd0ea18 | 1129 | |
56b2de80 DC |
1130 | pag = xfs_perag_get(mp, agno); |
1131 | ||
a2ceac1f DC |
1132 | ASSERT(pag->pagi_init); |
1133 | ASSERT(pag->pagi_inodeok); | |
1134 | ASSERT(pag->pagi_freecount > 0); | |
1135 | ||
56b2de80 | 1136 | restart_pagno: |
70eb7337 | 1137 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
2bd0ea18 NS |
1138 | /* |
1139 | * If pagino is 0 (this is the root inode allocation) use newino. | |
1140 | * This must work because we've just allocated some. | |
1141 | */ | |
1142 | if (!pagino) | |
6e3140c7 | 1143 | pagino = be32_to_cpu(agi->agi_newino); |
2bd0ea18 | 1144 | |
56b2de80 DC |
1145 | error = xfs_check_agi_freecount(cur, agi); |
1146 | if (error) | |
1147 | goto error0; | |
2bd0ea18 | 1148 | |
2bd0ea18 | 1149 | /* |
56b2de80 | 1150 | * If in the same AG as the parent, try to get near the parent. |
2bd0ea18 NS |
1151 | */ |
1152 | if (pagno == agno) { | |
56b2de80 DC |
1153 | int doneleft; /* done, to the left */ |
1154 | int doneright; /* done, to the right */ | |
56b2de80 DC |
1155 | |
1156 | error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i); | |
1157 | if (error) | |
2bd0ea18 | 1158 | goto error0; |
19ebedcf | 1159 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
56b2de80 DC |
1160 | |
1161 | error = xfs_inobt_get_rec(cur, &rec, &j); | |
1162 | if (error) | |
1163 | goto error0; | |
19ebedcf | 1164 | XFS_WANT_CORRUPTED_GOTO(mp, j == 1, error0); |
56b2de80 DC |
1165 | |
1166 | if (rec.ir_freecount > 0) { | |
2bd0ea18 NS |
1167 | /* |
1168 | * Found a free inode in the same chunk | |
56b2de80 | 1169 | * as the parent, done. |
2bd0ea18 | 1170 | */ |
56b2de80 | 1171 | goto alloc_inode; |
2bd0ea18 | 1172 | } |
56b2de80 DC |
1173 | |
1174 | ||
1175 | /* | |
1176 | * In the same AG as parent, but parent's chunk is full. | |
1177 | */ | |
1178 | ||
1179 | /* duplicate the cursor, search left & right simultaneously */ | |
1180 | error = xfs_btree_dup_cursor(cur, &tcur); | |
1181 | if (error) | |
1182 | goto error0; | |
1183 | ||
2bd0ea18 | 1184 | /* |
56b2de80 | 1185 | * Skip to last blocks looked up if same parent inode. |
2bd0ea18 | 1186 | */ |
56b2de80 DC |
1187 | if (pagino != NULLAGINO && |
1188 | pag->pagl_pagino == pagino && | |
1189 | pag->pagl_leftrec != NULLAGINO && | |
1190 | pag->pagl_rightrec != NULLAGINO) { | |
1191 | error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec, | |
3439d03a | 1192 | &trec, &doneleft); |
56b2de80 DC |
1193 | if (error) |
1194 | goto error1; | |
2bd0ea18 | 1195 | |
56b2de80 | 1196 | error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec, |
3439d03a | 1197 | &rec, &doneright); |
2bd0ea18 | 1198 | if (error) |
2bd0ea18 | 1199 | goto error1; |
56b2de80 DC |
1200 | } else { |
1201 | /* search left with tcur, back up 1 record */ | |
1202 | error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1); | |
1203 | if (error) | |
2bd0ea18 | 1204 | goto error1; |
2bd0ea18 | 1205 | |
56b2de80 DC |
1206 | /* search right with cur, go forward 1 record. */ |
1207 | error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0); | |
1208 | if (error) | |
1209 | goto error1; | |
1210 | } | |
1211 | ||
1212 | /* | |
1213 | * Loop until we find an inode chunk with a free inode. | |
1214 | */ | |
d8c47890 | 1215 | while (--searchdistance > 0 && (!doneleft || !doneright)) { |
56b2de80 DC |
1216 | int useleft; /* using left inode chunk this time */ |
1217 | ||
56b2de80 DC |
1218 | /* figure out the closer block if both are valid. */ |
1219 | if (!doneleft && !doneright) { | |
1220 | useleft = pagino - | |
1221 | (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) < | |
1222 | rec.ir_startino - pagino; | |
1223 | } else { | |
1224 | useleft = !doneleft; | |
1225 | } | |
1226 | ||
1227 | /* free inodes to the left? */ | |
1228 | if (useleft && trec.ir_freecount) { | |
56b2de80 DC |
1229 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
1230 | cur = tcur; | |
1231 | ||
1232 | pag->pagl_leftrec = trec.ir_startino; | |
1233 | pag->pagl_rightrec = rec.ir_startino; | |
1234 | pag->pagl_pagino = pagino; | |
ef54efb5 | 1235 | rec = trec; |
56b2de80 DC |
1236 | goto alloc_inode; |
1237 | } | |
1238 | ||
1239 | /* free inodes to the right? */ | |
1240 | if (!useleft && rec.ir_freecount) { | |
1241 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
1242 | ||
1243 | pag->pagl_leftrec = trec.ir_startino; | |
1244 | pag->pagl_rightrec = rec.ir_startino; | |
1245 | pag->pagl_pagino = pagino; | |
1246 | goto alloc_inode; | |
1247 | } | |
1248 | ||
1249 | /* get next record to check */ | |
1250 | if (useleft) { | |
1251 | error = xfs_ialloc_next_rec(tcur, &trec, | |
1252 | &doneleft, 1); | |
1253 | } else { | |
1254 | error = xfs_ialloc_next_rec(cur, &rec, | |
1255 | &doneright, 0); | |
1256 | } | |
1257 | if (error) | |
1258 | goto error1; | |
2bd0ea18 | 1259 | } |
56b2de80 | 1260 | |
d8c47890 CM |
1261 | if (searchdistance <= 0) { |
1262 | /* | |
1263 | * Not in range - save last search | |
1264 | * location and allocate a new inode | |
1265 | */ | |
1266 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
1267 | pag->pagl_leftrec = trec.ir_startino; | |
1268 | pag->pagl_rightrec = rec.ir_startino; | |
1269 | pag->pagl_pagino = pagino; | |
1270 | ||
1271 | } else { | |
1272 | /* | |
1273 | * We've reached the end of the btree. because | |
1274 | * we are only searching a small chunk of the | |
1275 | * btree each search, there is obviously free | |
1276 | * inodes closer to the parent inode than we | |
1277 | * are now. restart the search again. | |
1278 | */ | |
1279 | pag->pagl_pagino = NULLAGINO; | |
1280 | pag->pagl_leftrec = NULLAGINO; | |
1281 | pag->pagl_rightrec = NULLAGINO; | |
1282 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
1283 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
1284 | goto restart_pagno; | |
1285 | } | |
2bd0ea18 | 1286 | } |
56b2de80 | 1287 | |
2bd0ea18 | 1288 | /* |
56b2de80 | 1289 | * In a different AG from the parent. |
2bd0ea18 NS |
1290 | * See if the most recently allocated block has any free. |
1291 | */ | |
a2ceac1f | 1292 | if (agi->agi_newino != cpu_to_be32(NULLAGINO)) { |
56b2de80 DC |
1293 | error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino), |
1294 | XFS_LOOKUP_EQ, &i); | |
1295 | if (error) | |
2bd0ea18 | 1296 | goto error0; |
56b2de80 DC |
1297 | |
1298 | if (i == 1) { | |
1299 | error = xfs_inobt_get_rec(cur, &rec, &j); | |
2bd0ea18 NS |
1300 | if (error) |
1301 | goto error0; | |
56b2de80 DC |
1302 | |
1303 | if (j == 1 && rec.ir_freecount > 0) { | |
1304 | /* | |
1305 | * The last chunk allocated in the group | |
1306 | * still has a free inode. | |
1307 | */ | |
1308 | goto alloc_inode; | |
2bd0ea18 NS |
1309 | } |
1310 | } | |
1311 | } | |
56b2de80 DC |
1312 | |
1313 | /* | |
1314 | * None left in the last group, search the whole AG | |
1315 | */ | |
1316 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); | |
1317 | if (error) | |
1318 | goto error0; | |
19ebedcf | 1319 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
56b2de80 DC |
1320 | |
1321 | for (;;) { | |
1322 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
1323 | if (error) | |
1324 | goto error0; | |
19ebedcf | 1325 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
56b2de80 DC |
1326 | if (rec.ir_freecount > 0) |
1327 | break; | |
1328 | error = xfs_btree_increment(cur, 0, &i); | |
1329 | if (error) | |
1330 | goto error0; | |
19ebedcf | 1331 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
56b2de80 DC |
1332 | } |
1333 | ||
1334 | alloc_inode: | |
01792d3b | 1335 | offset = xfs_inobt_first_free_inode(&rec); |
2bd0ea18 NS |
1336 | ASSERT(offset >= 0); |
1337 | ASSERT(offset < XFS_INODES_PER_CHUNK); | |
1338 | ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) % | |
1339 | XFS_INODES_PER_CHUNK) == 0); | |
1340 | ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset); | |
56b2de80 | 1341 | rec.ir_free &= ~XFS_INOBT_MASK(offset); |
2bd0ea18 | 1342 | rec.ir_freecount--; |
56b2de80 DC |
1343 | error = xfs_inobt_update(cur, &rec); |
1344 | if (error) | |
2bd0ea18 | 1345 | goto error0; |
5e656dbb | 1346 | be32_add_cpu(&agi->agi_freecount, -1); |
2bd0ea18 | 1347 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); |
56b2de80 DC |
1348 | pag->pagi_freecount--; |
1349 | ||
1350 | error = xfs_check_agi_freecount(cur, agi); | |
1351 | if (error) | |
1352 | goto error0; | |
2bd0ea18 | 1353 | |
2bd0ea18 NS |
1354 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
1355 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1); | |
56b2de80 | 1356 | xfs_perag_put(pag); |
2bd0ea18 NS |
1357 | *inop = ino; |
1358 | return 0; | |
1359 | error1: | |
1360 | xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR); | |
1361 | error0: | |
1362 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
56b2de80 | 1363 | xfs_perag_put(pag); |
2bd0ea18 NS |
1364 | return error; |
1365 | } | |
1366 | ||
ff105f75 DC |
1367 | /* |
1368 | * Use the free inode btree to allocate an inode based on distance from the | |
1369 | * parent. Note that the provided cursor may be deleted and replaced. | |
1370 | */ | |
1371 | STATIC int | |
1372 | xfs_dialloc_ag_finobt_near( | |
1373 | xfs_agino_t pagino, | |
1374 | struct xfs_btree_cur **ocur, | |
1375 | struct xfs_inobt_rec_incore *rec) | |
1376 | { | |
1377 | struct xfs_btree_cur *lcur = *ocur; /* left search cursor */ | |
1378 | struct xfs_btree_cur *rcur; /* right search cursor */ | |
1379 | struct xfs_inobt_rec_incore rrec; | |
1380 | int error; | |
1381 | int i, j; | |
1382 | ||
1383 | error = xfs_inobt_lookup(lcur, pagino, XFS_LOOKUP_LE, &i); | |
1384 | if (error) | |
1385 | return error; | |
1386 | ||
1387 | if (i == 1) { | |
1388 | error = xfs_inobt_get_rec(lcur, rec, &i); | |
1389 | if (error) | |
1390 | return error; | |
19ebedcf | 1391 | XFS_WANT_CORRUPTED_RETURN(lcur->bc_mp, i == 1); |
ff105f75 DC |
1392 | |
1393 | /* | |
1394 | * See if we've landed in the parent inode record. The finobt | |
1395 | * only tracks chunks with at least one free inode, so record | |
1396 | * existence is enough. | |
1397 | */ | |
1398 | if (pagino >= rec->ir_startino && | |
1399 | pagino < (rec->ir_startino + XFS_INODES_PER_CHUNK)) | |
1400 | return 0; | |
1401 | } | |
1402 | ||
1403 | error = xfs_btree_dup_cursor(lcur, &rcur); | |
1404 | if (error) | |
1405 | return error; | |
1406 | ||
1407 | error = xfs_inobt_lookup(rcur, pagino, XFS_LOOKUP_GE, &j); | |
1408 | if (error) | |
1409 | goto error_rcur; | |
1410 | if (j == 1) { | |
1411 | error = xfs_inobt_get_rec(rcur, &rrec, &j); | |
1412 | if (error) | |
1413 | goto error_rcur; | |
19ebedcf | 1414 | XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, j == 1, error_rcur); |
ff105f75 DC |
1415 | } |
1416 | ||
19ebedcf | 1417 | XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, i == 1 || j == 1, error_rcur); |
ff105f75 DC |
1418 | if (i == 1 && j == 1) { |
1419 | /* | |
1420 | * Both the left and right records are valid. Choose the closer | |
1421 | * inode chunk to the target. | |
1422 | */ | |
1423 | if ((pagino - rec->ir_startino + XFS_INODES_PER_CHUNK - 1) > | |
1424 | (rrec.ir_startino - pagino)) { | |
1425 | *rec = rrec; | |
1426 | xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR); | |
1427 | *ocur = rcur; | |
1428 | } else { | |
1429 | xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR); | |
1430 | } | |
1431 | } else if (j == 1) { | |
1432 | /* only the right record is valid */ | |
1433 | *rec = rrec; | |
1434 | xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR); | |
1435 | *ocur = rcur; | |
1436 | } else if (i == 1) { | |
1437 | /* only the left record is valid */ | |
1438 | xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR); | |
1439 | } | |
1440 | ||
1441 | return 0; | |
1442 | ||
1443 | error_rcur: | |
1444 | xfs_btree_del_cursor(rcur, XFS_BTREE_ERROR); | |
1445 | return error; | |
1446 | } | |
1447 | ||
1448 | /* | |
1449 | * Use the free inode btree to find a free inode based on a newino hint. If | |
1450 | * the hint is NULL, find the first free inode in the AG. | |
1451 | */ | |
1452 | STATIC int | |
1453 | xfs_dialloc_ag_finobt_newino( | |
1454 | struct xfs_agi *agi, | |
1455 | struct xfs_btree_cur *cur, | |
1456 | struct xfs_inobt_rec_incore *rec) | |
1457 | { | |
1458 | int error; | |
1459 | int i; | |
1460 | ||
1461 | if (agi->agi_newino != cpu_to_be32(NULLAGINO)) { | |
1462 | error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino), | |
1463 | XFS_LOOKUP_EQ, &i); | |
1464 | if (error) | |
1465 | return error; | |
1466 | if (i == 1) { | |
1467 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1468 | if (error) | |
1469 | return error; | |
19ebedcf | 1470 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
ff105f75 DC |
1471 | return 0; |
1472 | } | |
1473 | } | |
1474 | ||
1475 | /* | |
1476 | * Find the first inode available in the AG. | |
1477 | */ | |
1478 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); | |
1479 | if (error) | |
1480 | return error; | |
19ebedcf | 1481 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
ff105f75 DC |
1482 | |
1483 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1484 | if (error) | |
1485 | return error; | |
19ebedcf | 1486 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
ff105f75 DC |
1487 | |
1488 | return 0; | |
1489 | } | |
1490 | ||
1491 | /* | |
1492 | * Update the inobt based on a modification made to the finobt. Also ensure that | |
1493 | * the records from both trees are equivalent post-modification. | |
1494 | */ | |
1495 | STATIC int | |
1496 | xfs_dialloc_ag_update_inobt( | |
1497 | struct xfs_btree_cur *cur, /* inobt cursor */ | |
1498 | struct xfs_inobt_rec_incore *frec, /* finobt record */ | |
1499 | int offset) /* inode offset */ | |
1500 | { | |
1501 | struct xfs_inobt_rec_incore rec; | |
1502 | int error; | |
1503 | int i; | |
1504 | ||
1505 | error = xfs_inobt_lookup(cur, frec->ir_startino, XFS_LOOKUP_EQ, &i); | |
1506 | if (error) | |
1507 | return error; | |
19ebedcf | 1508 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
ff105f75 DC |
1509 | |
1510 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
1511 | if (error) | |
1512 | return error; | |
19ebedcf | 1513 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
ff105f75 DC |
1514 | ASSERT((XFS_AGINO_TO_OFFSET(cur->bc_mp, rec.ir_startino) % |
1515 | XFS_INODES_PER_CHUNK) == 0); | |
1516 | ||
1517 | rec.ir_free &= ~XFS_INOBT_MASK(offset); | |
1518 | rec.ir_freecount--; | |
1519 | ||
19ebedcf | 1520 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, (rec.ir_free == frec->ir_free) && |
ff105f75 DC |
1521 | (rec.ir_freecount == frec->ir_freecount)); |
1522 | ||
5a35bf2c | 1523 | return xfs_inobt_update(cur, &rec); |
ff105f75 DC |
1524 | } |
1525 | ||
1526 | /* | |
1527 | * Allocate an inode using the free inode btree, if available. Otherwise, fall | |
1528 | * back to the inobt search algorithm. | |
1529 | * | |
1530 | * The caller selected an AG for us, and made sure that free inodes are | |
1531 | * available. | |
1532 | */ | |
88fc7306 BF |
1533 | STATIC int |
1534 | xfs_dialloc_ag( | |
1535 | struct xfs_trans *tp, | |
1536 | struct xfs_buf *agbp, | |
1537 | xfs_ino_t parent, | |
1538 | xfs_ino_t *inop) | |
1539 | { | |
1540 | struct xfs_mount *mp = tp->t_mountp; | |
1541 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
1542 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
1543 | xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent); | |
1544 | xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent); | |
1545 | struct xfs_perag *pag; | |
ff105f75 DC |
1546 | struct xfs_btree_cur *cur; /* finobt cursor */ |
1547 | struct xfs_btree_cur *icur; /* inobt cursor */ | |
88fc7306 | 1548 | struct xfs_inobt_rec_incore rec; |
88fc7306 BF |
1549 | xfs_ino_t ino; |
1550 | int error; | |
1551 | int offset; | |
ff105f75 | 1552 | int i; |
88fc7306 BF |
1553 | |
1554 | if (!xfs_sb_version_hasfinobt(&mp->m_sb)) | |
ff105f75 | 1555 | return xfs_dialloc_ag_inobt(tp, agbp, parent, inop); |
88fc7306 BF |
1556 | |
1557 | pag = xfs_perag_get(mp, agno); | |
1558 | ||
1559 | /* | |
1560 | * If pagino is 0 (this is the root inode allocation) use newino. | |
1561 | * This must work because we've just allocated some. | |
1562 | */ | |
1563 | if (!pagino) | |
1564 | pagino = be32_to_cpu(agi->agi_newino); | |
1565 | ||
1566 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO); | |
1567 | ||
1568 | error = xfs_check_agi_freecount(cur, agi); | |
1569 | if (error) | |
1570 | goto error_cur; | |
1571 | ||
ff105f75 DC |
1572 | /* |
1573 | * The search algorithm depends on whether we're in the same AG as the | |
1574 | * parent. If so, find the closest available inode to the parent. If | |
1575 | * not, consider the agi hint or find the first free inode in the AG. | |
1576 | */ | |
1577 | if (agno == pagno) | |
1578 | error = xfs_dialloc_ag_finobt_near(pagino, &cur, &rec); | |
1579 | else | |
1580 | error = xfs_dialloc_ag_finobt_newino(agi, cur, &rec); | |
1581 | if (error) | |
1582 | goto error_cur; | |
88fc7306 | 1583 | |
01792d3b | 1584 | offset = xfs_inobt_first_free_inode(&rec); |
88fc7306 BF |
1585 | ASSERT(offset >= 0); |
1586 | ASSERT(offset < XFS_INODES_PER_CHUNK); | |
1587 | ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) % | |
1588 | XFS_INODES_PER_CHUNK) == 0); | |
1589 | ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset); | |
1590 | ||
1591 | /* | |
1592 | * Modify or remove the finobt record. | |
1593 | */ | |
1594 | rec.ir_free &= ~XFS_INOBT_MASK(offset); | |
1595 | rec.ir_freecount--; | |
ff105f75 | 1596 | if (rec.ir_freecount) |
88fc7306 BF |
1597 | error = xfs_inobt_update(cur, &rec); |
1598 | else | |
1599 | error = xfs_btree_delete(cur, &i); | |
1600 | if (error) | |
1601 | goto error_cur; | |
1602 | ||
1603 | /* | |
ff105f75 DC |
1604 | * The finobt has now been updated appropriately. We haven't updated the |
1605 | * agi and superblock yet, so we can create an inobt cursor and validate | |
1606 | * the original freecount. If all is well, make the equivalent update to | |
1607 | * the inobt using the finobt record and offset information. | |
88fc7306 | 1608 | */ |
ff105f75 | 1609 | icur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
88fc7306 | 1610 | |
ff105f75 | 1611 | error = xfs_check_agi_freecount(icur, agi); |
88fc7306 | 1612 | if (error) |
ff105f75 | 1613 | goto error_icur; |
88fc7306 | 1614 | |
ff105f75 | 1615 | error = xfs_dialloc_ag_update_inobt(icur, &rec, offset); |
88fc7306 | 1616 | if (error) |
ff105f75 | 1617 | goto error_icur; |
88fc7306 BF |
1618 | |
1619 | /* | |
ff105f75 DC |
1620 | * Both trees have now been updated. We must update the perag and |
1621 | * superblock before we can check the freecount for each btree. | |
88fc7306 BF |
1622 | */ |
1623 | be32_add_cpu(&agi->agi_freecount, -1); | |
1624 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); | |
1625 | pag->pagi_freecount--; | |
1626 | ||
1627 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1); | |
1628 | ||
ff105f75 | 1629 | error = xfs_check_agi_freecount(icur, agi); |
88fc7306 | 1630 | if (error) |
ff105f75 | 1631 | goto error_icur; |
88fc7306 BF |
1632 | error = xfs_check_agi_freecount(cur, agi); |
1633 | if (error) | |
ff105f75 | 1634 | goto error_icur; |
88fc7306 | 1635 | |
ff105f75 | 1636 | xfs_btree_del_cursor(icur, XFS_BTREE_NOERROR); |
88fc7306 BF |
1637 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
1638 | xfs_perag_put(pag); | |
1639 | *inop = ino; | |
1640 | return 0; | |
1641 | ||
ff105f75 DC |
1642 | error_icur: |
1643 | xfs_btree_del_cursor(icur, XFS_BTREE_ERROR); | |
88fc7306 BF |
1644 | error_cur: |
1645 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
1646 | xfs_perag_put(pag); | |
1647 | return error; | |
1648 | } | |
1649 | ||
a2ceac1f DC |
1650 | /* |
1651 | * Allocate an inode on disk. | |
1652 | * | |
1653 | * Mode is used to tell whether the new inode will need space, and whether it | |
1654 | * is a directory. | |
1655 | * | |
1656 | * This function is designed to be called twice if it has to do an allocation | |
1657 | * to make more free inodes. On the first call, *IO_agbp should be set to NULL. | |
1658 | * If an inode is available without having to performn an allocation, an inode | |
1659 | * number is returned. In this case, *IO_agbp is set to NULL. If an allocation | |
1660 | * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp. | |
1661 | * The caller should then commit the current transaction, allocate a | |
1662 | * new transaction, and call xfs_dialloc() again, passing in the previous value | |
1663 | * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI | |
1664 | * buffer is locked across the two calls, the second call is guaranteed to have | |
1665 | * a free inode available. | |
1666 | * | |
1667 | * Once we successfully pick an inode its number is returned and the on-disk | |
1668 | * data structures are updated. The inode itself is not read in, since doing so | |
1669 | * would break ordering constraints with xfs_reclaim. | |
1670 | */ | |
1671 | int | |
1672 | xfs_dialloc( | |
1673 | struct xfs_trans *tp, | |
1674 | xfs_ino_t parent, | |
1675 | umode_t mode, | |
1676 | int okalloc, | |
1677 | struct xfs_buf **IO_agbp, | |
1678 | xfs_ino_t *inop) | |
1679 | { | |
1680 | struct xfs_mount *mp = tp->t_mountp; | |
1681 | struct xfs_buf *agbp; | |
1682 | xfs_agnumber_t agno; | |
1683 | int error; | |
1684 | int ialloced; | |
1685 | int noroom = 0; | |
1686 | xfs_agnumber_t start_agno; | |
1687 | struct xfs_perag *pag; | |
1688 | ||
1689 | if (*IO_agbp) { | |
1690 | /* | |
1691 | * If the caller passes in a pointer to the AGI buffer, | |
1692 | * continue where we left off before. In this case, we | |
1693 | * know that the allocation group has free inodes. | |
1694 | */ | |
1695 | agbp = *IO_agbp; | |
1696 | goto out_alloc; | |
1697 | } | |
1698 | ||
1699 | /* | |
1700 | * We do not have an agbp, so select an initial allocation | |
1701 | * group for inode allocation. | |
1702 | */ | |
1703 | start_agno = xfs_ialloc_ag_select(tp, parent, mode, okalloc); | |
1704 | if (start_agno == NULLAGNUMBER) { | |
1705 | *inop = NULLFSINO; | |
1706 | return 0; | |
1707 | } | |
1708 | ||
1709 | /* | |
1710 | * If we have already hit the ceiling of inode blocks then clear | |
1711 | * okalloc so we scan all available agi structures for a free | |
1712 | * inode. | |
cbf3beaa GW |
1713 | * |
1714 | * Read rough value of mp->m_icount by percpu_counter_read_positive, | |
1715 | * which will sacrifice the preciseness but improve the performance. | |
a2ceac1f DC |
1716 | */ |
1717 | if (mp->m_maxicount && | |
cbf3beaa GW |
1718 | percpu_counter_read_positive(&mp->m_icount) + mp->m_ialloc_inos |
1719 | > mp->m_maxicount) { | |
a2ceac1f DC |
1720 | noroom = 1; |
1721 | okalloc = 0; | |
1722 | } | |
1723 | ||
1724 | /* | |
1725 | * Loop until we find an allocation group that either has free inodes | |
1726 | * or in which we can allocate some inodes. Iterate through the | |
1727 | * allocation groups upward, wrapping at the end. | |
1728 | */ | |
1729 | agno = start_agno; | |
1730 | for (;;) { | |
1731 | pag = xfs_perag_get(mp, agno); | |
1732 | if (!pag->pagi_inodeok) { | |
1733 | xfs_ialloc_next_ag(mp); | |
1734 | goto nextag; | |
1735 | } | |
1736 | ||
1737 | if (!pag->pagi_init) { | |
1738 | error = xfs_ialloc_pagi_init(mp, tp, agno); | |
1739 | if (error) | |
1740 | goto out_error; | |
1741 | } | |
1742 | ||
1743 | /* | |
1744 | * Do a first racy fast path check if this AG is usable. | |
1745 | */ | |
1746 | if (!pag->pagi_freecount && !okalloc) | |
1747 | goto nextag; | |
1748 | ||
1749 | /* | |
1750 | * Then read in the AGI buffer and recheck with the AGI buffer | |
1751 | * lock held. | |
1752 | */ | |
1753 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); | |
1754 | if (error) | |
1755 | goto out_error; | |
1756 | ||
1757 | if (pag->pagi_freecount) { | |
1758 | xfs_perag_put(pag); | |
1759 | goto out_alloc; | |
1760 | } | |
1761 | ||
1762 | if (!okalloc) | |
1763 | goto nextag_relse_buffer; | |
1764 | ||
1765 | ||
1766 | error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced); | |
1767 | if (error) { | |
1768 | xfs_trans_brelse(tp, agbp); | |
1769 | ||
12b53197 | 1770 | if (error != -ENOSPC) |
a2ceac1f DC |
1771 | goto out_error; |
1772 | ||
1773 | xfs_perag_put(pag); | |
1774 | *inop = NULLFSINO; | |
1775 | return 0; | |
1776 | } | |
1777 | ||
1778 | if (ialloced) { | |
1779 | /* | |
1780 | * We successfully allocated some inodes, return | |
1781 | * the current context to the caller so that it | |
1782 | * can commit the current transaction and call | |
1783 | * us again where we left off. | |
1784 | */ | |
1785 | ASSERT(pag->pagi_freecount > 0); | |
1786 | xfs_perag_put(pag); | |
1787 | ||
1788 | *IO_agbp = agbp; | |
1789 | *inop = NULLFSINO; | |
1790 | return 0; | |
1791 | } | |
1792 | ||
1793 | nextag_relse_buffer: | |
1794 | xfs_trans_brelse(tp, agbp); | |
1795 | nextag: | |
1796 | xfs_perag_put(pag); | |
1797 | if (++agno == mp->m_sb.sb_agcount) | |
1798 | agno = 0; | |
1799 | if (agno == start_agno) { | |
1800 | *inop = NULLFSINO; | |
12b53197 | 1801 | return noroom ? -ENOSPC : 0; |
a2ceac1f DC |
1802 | } |
1803 | } | |
1804 | ||
1805 | out_alloc: | |
1806 | *IO_agbp = NULL; | |
1807 | return xfs_dialloc_ag(tp, agbp, parent, inop); | |
1808 | out_error: | |
1809 | xfs_perag_put(pag); | |
1e68581b | 1810 | return error; |
a2ceac1f DC |
1811 | } |
1812 | ||
7338c4b8 BF |
1813 | /* |
1814 | * Free the blocks of an inode chunk. We must consider that the inode chunk | |
1815 | * might be sparse and only free the regions that are allocated as part of the | |
1816 | * chunk. | |
1817 | */ | |
1818 | STATIC void | |
1819 | xfs_difree_inode_chunk( | |
1820 | struct xfs_mount *mp, | |
1821 | xfs_agnumber_t agno, | |
1822 | struct xfs_inobt_rec_incore *rec, | |
f33cea1a | 1823 | struct xfs_defer_ops *dfops) |
7338c4b8 BF |
1824 | { |
1825 | xfs_agblock_t sagbno = XFS_AGINO_TO_AGBNO(mp, rec->ir_startino); | |
1826 | int startidx, endidx; | |
1827 | int nextbit; | |
1828 | xfs_agblock_t agbno; | |
1829 | int contigblk; | |
85aec44f | 1830 | struct xfs_owner_info oinfo; |
7338c4b8 | 1831 | DECLARE_BITMAP(holemask, XFS_INOBT_HOLEMASK_BITS); |
85aec44f | 1832 | xfs_rmap_ag_owner(&oinfo, XFS_RMAP_OWN_INODES); |
7338c4b8 BF |
1833 | |
1834 | if (!xfs_inobt_issparse(rec->ir_holemask)) { | |
1835 | /* not sparse, calculate extent info directly */ | |
f33cea1a | 1836 | xfs_bmap_add_free(mp, dfops, XFS_AGB_TO_FSB(mp, agno, sagbno), |
85aec44f | 1837 | mp->m_ialloc_blks, &oinfo); |
7338c4b8 BF |
1838 | return; |
1839 | } | |
1840 | ||
1841 | /* holemask is only 16-bits (fits in an unsigned long) */ | |
1842 | ASSERT(sizeof(rec->ir_holemask) <= sizeof(holemask[0])); | |
1843 | holemask[0] = rec->ir_holemask; | |
1844 | ||
1845 | /* | |
1846 | * Find contiguous ranges of zeroes (i.e., allocated regions) in the | |
1847 | * holemask and convert the start/end index of each range to an extent. | |
1848 | * We start with the start and end index both pointing at the first 0 in | |
1849 | * the mask. | |
1850 | */ | |
1851 | startidx = endidx = find_first_zero_bit(holemask, | |
1852 | XFS_INOBT_HOLEMASK_BITS); | |
1853 | nextbit = startidx + 1; | |
1854 | while (startidx < XFS_INOBT_HOLEMASK_BITS) { | |
1855 | nextbit = find_next_zero_bit(holemask, XFS_INOBT_HOLEMASK_BITS, | |
1856 | nextbit); | |
1857 | /* | |
1858 | * If the next zero bit is contiguous, update the end index of | |
1859 | * the current range and continue. | |
1860 | */ | |
1861 | if (nextbit != XFS_INOBT_HOLEMASK_BITS && | |
1862 | nextbit == endidx + 1) { | |
1863 | endidx = nextbit; | |
1864 | goto next; | |
1865 | } | |
1866 | ||
1867 | /* | |
1868 | * nextbit is not contiguous with the current end index. Convert | |
1869 | * the current start/end to an extent and add it to the free | |
1870 | * list. | |
1871 | */ | |
1872 | agbno = sagbno + (startidx * XFS_INODES_PER_HOLEMASK_BIT) / | |
1873 | mp->m_sb.sb_inopblock; | |
1874 | contigblk = ((endidx - startidx + 1) * | |
1875 | XFS_INODES_PER_HOLEMASK_BIT) / | |
1876 | mp->m_sb.sb_inopblock; | |
1877 | ||
1878 | ASSERT(agbno % mp->m_sb.sb_spino_align == 0); | |
1879 | ASSERT(contigblk % mp->m_sb.sb_spino_align == 0); | |
f33cea1a | 1880 | xfs_bmap_add_free(mp, dfops, XFS_AGB_TO_FSB(mp, agno, agbno), |
85aec44f | 1881 | contigblk, &oinfo); |
7338c4b8 BF |
1882 | |
1883 | /* reset range to current bit and carry on... */ | |
1884 | startidx = endidx = nextbit; | |
1885 | ||
1886 | next: | |
1887 | nextbit++; | |
1888 | } | |
1889 | } | |
1890 | ||
eb9a297a BF |
1891 | STATIC int |
1892 | xfs_difree_inobt( | |
1893 | struct xfs_mount *mp, | |
1894 | struct xfs_trans *tp, | |
1895 | struct xfs_buf *agbp, | |
1896 | xfs_agino_t agino, | |
f33cea1a | 1897 | struct xfs_defer_ops *dfops, |
5a3b2e0a | 1898 | struct xfs_icluster *xic, |
eb9a297a | 1899 | struct xfs_inobt_rec_incore *orec) |
3439d03a | 1900 | { |
eb9a297a BF |
1901 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); |
1902 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
1903 | struct xfs_perag *pag; | |
1904 | struct xfs_btree_cur *cur; | |
1905 | struct xfs_inobt_rec_incore rec; | |
1906 | int ilen; | |
1907 | int error; | |
1908 | int i; | |
1909 | int off; | |
3439d03a | 1910 | |
3439d03a | 1911 | ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC)); |
eb9a297a BF |
1912 | ASSERT(XFS_AGINO_TO_AGBNO(mp, agino) < be32_to_cpu(agi->agi_length)); |
1913 | ||
3439d03a DC |
1914 | /* |
1915 | * Initialize the cursor. | |
1916 | */ | |
70eb7337 | 1917 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
3439d03a DC |
1918 | |
1919 | error = xfs_check_agi_freecount(cur, agi); | |
1920 | if (error) | |
1921 | goto error0; | |
1922 | ||
1923 | /* | |
1924 | * Look for the entry describing this inode. | |
1925 | */ | |
1926 | if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) { | |
1927 | xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.", | |
1928 | __func__, error); | |
1929 | goto error0; | |
1930 | } | |
19ebedcf | 1931 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
3439d03a DC |
1932 | error = xfs_inobt_get_rec(cur, &rec, &i); |
1933 | if (error) { | |
1934 | xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.", | |
1935 | __func__, error); | |
1936 | goto error0; | |
1937 | } | |
19ebedcf | 1938 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
3439d03a DC |
1939 | /* |
1940 | * Get the offset in the inode chunk. | |
1941 | */ | |
1942 | off = agino - rec.ir_startino; | |
1943 | ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK); | |
1944 | ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off))); | |
1945 | /* | |
1946 | * Mark the inode free & increment the count. | |
1947 | */ | |
1948 | rec.ir_free |= XFS_INOBT_MASK(off); | |
1949 | rec.ir_freecount++; | |
1950 | ||
1951 | /* | |
f6580bcf BF |
1952 | * When an inode chunk is free, it becomes eligible for removal. Don't |
1953 | * remove the chunk if the block size is large enough for multiple inode | |
1954 | * chunks (that might not be free). | |
3439d03a DC |
1955 | */ |
1956 | if (!(mp->m_flags & XFS_MOUNT_IKEEP) && | |
f6580bcf BF |
1957 | rec.ir_free == XFS_INOBT_ALL_FREE && |
1958 | mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK) { | |
180d3cd8 | 1959 | xic->deleted = true; |
5a3b2e0a BF |
1960 | xic->first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino); |
1961 | xic->alloc = xfs_inobt_irec_to_allocmask(&rec); | |
3439d03a DC |
1962 | |
1963 | /* | |
1964 | * Remove the inode cluster from the AGI B+Tree, adjust the | |
1965 | * AGI and Superblock inode counts, and mark the disk space | |
1966 | * to be freed when the transaction is committed. | |
1967 | */ | |
f6580bcf | 1968 | ilen = rec.ir_freecount; |
3439d03a DC |
1969 | be32_add_cpu(&agi->agi_count, -ilen); |
1970 | be32_add_cpu(&agi->agi_freecount, -(ilen - 1)); | |
1971 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT); | |
1972 | pag = xfs_perag_get(mp, agno); | |
1973 | pag->pagi_freecount -= ilen - 1; | |
1974 | xfs_perag_put(pag); | |
1975 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen); | |
1976 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1)); | |
1977 | ||
1978 | if ((error = xfs_btree_delete(cur, &i))) { | |
1979 | xfs_warn(mp, "%s: xfs_btree_delete returned error %d.", | |
1980 | __func__, error); | |
1981 | goto error0; | |
1982 | } | |
1983 | ||
f33cea1a | 1984 | xfs_difree_inode_chunk(mp, agno, &rec, dfops); |
3439d03a | 1985 | } else { |
180d3cd8 | 1986 | xic->deleted = false; |
3439d03a DC |
1987 | |
1988 | error = xfs_inobt_update(cur, &rec); | |
1989 | if (error) { | |
1990 | xfs_warn(mp, "%s: xfs_inobt_update returned error %d.", | |
1991 | __func__, error); | |
1992 | goto error0; | |
1993 | } | |
1994 | ||
f8149110 | 1995 | /* |
3439d03a DC |
1996 | * Change the inode free counts and log the ag/sb changes. |
1997 | */ | |
1998 | be32_add_cpu(&agi->agi_freecount, 1); | |
1999 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); | |
2000 | pag = xfs_perag_get(mp, agno); | |
2001 | pag->pagi_freecount++; | |
2002 | xfs_perag_put(pag); | |
2003 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1); | |
2004 | } | |
2005 | ||
2006 | error = xfs_check_agi_freecount(cur, agi); | |
2007 | if (error) | |
2008 | goto error0; | |
2009 | ||
eb9a297a | 2010 | *orec = rec; |
3439d03a DC |
2011 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
2012 | return 0; | |
2013 | ||
2014 | error0: | |
2015 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
2016 | return error; | |
2017 | } | |
2018 | ||
1bb93fd1 BF |
2019 | /* |
2020 | * Free an inode in the free inode btree. | |
2021 | */ | |
2022 | STATIC int | |
2023 | xfs_difree_finobt( | |
2024 | struct xfs_mount *mp, | |
2025 | struct xfs_trans *tp, | |
2026 | struct xfs_buf *agbp, | |
2027 | xfs_agino_t agino, | |
2028 | struct xfs_inobt_rec_incore *ibtrec) /* inobt record */ | |
2029 | { | |
2030 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
2031 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
2032 | struct xfs_btree_cur *cur; | |
2033 | struct xfs_inobt_rec_incore rec; | |
2034 | int offset = agino - ibtrec->ir_startino; | |
2035 | int error; | |
2036 | int i; | |
2037 | ||
2038 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO); | |
2039 | ||
2040 | error = xfs_inobt_lookup(cur, ibtrec->ir_startino, XFS_LOOKUP_EQ, &i); | |
2041 | if (error) | |
2042 | goto error; | |
2043 | if (i == 0) { | |
2044 | /* | |
2045 | * If the record does not exist in the finobt, we must have just | |
2046 | * freed an inode in a previously fully allocated chunk. If not, | |
2047 | * something is out of sync. | |
2048 | */ | |
19ebedcf | 2049 | XFS_WANT_CORRUPTED_GOTO(mp, ibtrec->ir_freecount == 1, error); |
1bb93fd1 | 2050 | |
11640e30 BF |
2051 | error = xfs_inobt_insert_rec(cur, ibtrec->ir_holemask, |
2052 | ibtrec->ir_count, | |
2053 | ibtrec->ir_freecount, | |
1bb93fd1 BF |
2054 | ibtrec->ir_free, &i); |
2055 | if (error) | |
2056 | goto error; | |
2057 | ASSERT(i == 1); | |
2058 | ||
2059 | goto out; | |
2060 | } | |
2061 | ||
2062 | /* | |
ff105f75 DC |
2063 | * Read and update the existing record. We could just copy the ibtrec |
2064 | * across here, but that would defeat the purpose of having redundant | |
2065 | * metadata. By making the modifications independently, we can catch | |
2066 | * corruptions that we wouldn't see if we just copied from one record | |
2067 | * to another. | |
1bb93fd1 BF |
2068 | */ |
2069 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
2070 | if (error) | |
2071 | goto error; | |
19ebedcf | 2072 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error); |
1bb93fd1 BF |
2073 | |
2074 | rec.ir_free |= XFS_INOBT_MASK(offset); | |
2075 | rec.ir_freecount++; | |
2076 | ||
19ebedcf | 2077 | XFS_WANT_CORRUPTED_GOTO(mp, (rec.ir_free == ibtrec->ir_free) && |
1bb93fd1 BF |
2078 | (rec.ir_freecount == ibtrec->ir_freecount), |
2079 | error); | |
2080 | ||
2081 | /* | |
2082 | * The content of inobt records should always match between the inobt | |
2083 | * and finobt. The lifecycle of records in the finobt is different from | |
2084 | * the inobt in that the finobt only tracks records with at least one | |
ff105f75 DC |
2085 | * free inode. Hence, if all of the inodes are free and we aren't |
2086 | * keeping inode chunks permanently on disk, remove the record. | |
2087 | * Otherwise, update the record with the new information. | |
f6580bcf BF |
2088 | * |
2089 | * Note that we currently can't free chunks when the block size is large | |
2090 | * enough for multiple chunks. Leave the finobt record to remain in sync | |
2091 | * with the inobt. | |
1bb93fd1 | 2092 | */ |
f6580bcf BF |
2093 | if (rec.ir_free == XFS_INOBT_ALL_FREE && |
2094 | mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK && | |
1bb93fd1 | 2095 | !(mp->m_flags & XFS_MOUNT_IKEEP)) { |
1bb93fd1 BF |
2096 | error = xfs_btree_delete(cur, &i); |
2097 | if (error) | |
2098 | goto error; | |
2099 | ASSERT(i == 1); | |
2100 | } else { | |
1bb93fd1 BF |
2101 | error = xfs_inobt_update(cur, &rec); |
2102 | if (error) | |
2103 | goto error; | |
2104 | } | |
2105 | ||
2106 | out: | |
2107 | error = xfs_check_agi_freecount(cur, agi); | |
2108 | if (error) | |
2109 | goto error; | |
2110 | ||
2111 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
2112 | return 0; | |
2113 | ||
2114 | error: | |
2115 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
2116 | return error; | |
2117 | } | |
2118 | ||
eb9a297a BF |
2119 | /* |
2120 | * Free disk inode. Carefully avoids touching the incore inode, all | |
2121 | * manipulations incore are the caller's responsibility. | |
2122 | * The on-disk inode is not changed by this operation, only the | |
2123 | * btree (free inode mask) is changed. | |
2124 | */ | |
2125 | int | |
2126 | xfs_difree( | |
2127 | struct xfs_trans *tp, /* transaction pointer */ | |
2128 | xfs_ino_t inode, /* inode to be freed */ | |
f33cea1a | 2129 | struct xfs_defer_ops *dfops, /* extents to free */ |
5a3b2e0a | 2130 | struct xfs_icluster *xic) /* cluster info if deleted */ |
eb9a297a BF |
2131 | { |
2132 | /* REFERENCED */ | |
2133 | xfs_agblock_t agbno; /* block number containing inode */ | |
2134 | struct xfs_buf *agbp; /* buffer for allocation group header */ | |
2135 | xfs_agino_t agino; /* allocation group inode number */ | |
2136 | xfs_agnumber_t agno; /* allocation group number */ | |
2137 | int error; /* error return value */ | |
2138 | struct xfs_mount *mp; /* mount structure for filesystem */ | |
2139 | struct xfs_inobt_rec_incore rec;/* btree record */ | |
2140 | ||
2141 | mp = tp->t_mountp; | |
2142 | ||
2143 | /* | |
2144 | * Break up inode number into its components. | |
2145 | */ | |
2146 | agno = XFS_INO_TO_AGNO(mp, inode); | |
2147 | if (agno >= mp->m_sb.sb_agcount) { | |
2148 | xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).", | |
2149 | __func__, agno, mp->m_sb.sb_agcount); | |
2150 | ASSERT(0); | |
12b53197 | 2151 | return -EINVAL; |
eb9a297a BF |
2152 | } |
2153 | agino = XFS_INO_TO_AGINO(mp, inode); | |
2154 | if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
2155 | xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).", | |
2156 | __func__, (unsigned long long)inode, | |
2157 | (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino)); | |
2158 | ASSERT(0); | |
12b53197 | 2159 | return -EINVAL; |
eb9a297a BF |
2160 | } |
2161 | agbno = XFS_AGINO_TO_AGBNO(mp, agino); | |
2162 | if (agbno >= mp->m_sb.sb_agblocks) { | |
2163 | xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).", | |
2164 | __func__, agbno, mp->m_sb.sb_agblocks); | |
2165 | ASSERT(0); | |
12b53197 | 2166 | return -EINVAL; |
eb9a297a BF |
2167 | } |
2168 | /* | |
2169 | * Get the allocation group header. | |
2170 | */ | |
2171 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); | |
2172 | if (error) { | |
2173 | xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.", | |
2174 | __func__, error); | |
2175 | return error; | |
2176 | } | |
2177 | ||
2178 | /* | |
2179 | * Fix up the inode allocation btree. | |
2180 | */ | |
f33cea1a | 2181 | error = xfs_difree_inobt(mp, tp, agbp, agino, dfops, xic, &rec); |
eb9a297a BF |
2182 | if (error) |
2183 | goto error0; | |
2184 | ||
1bb93fd1 BF |
2185 | /* |
2186 | * Fix up the free inode btree. | |
2187 | */ | |
2188 | if (xfs_sb_version_hasfinobt(&mp->m_sb)) { | |
2189 | error = xfs_difree_finobt(mp, tp, agbp, agino, &rec); | |
2190 | if (error) | |
2191 | goto error0; | |
2192 | } | |
2193 | ||
eb9a297a BF |
2194 | return 0; |
2195 | ||
2196 | error0: | |
2197 | return error; | |
2198 | } | |
2199 | ||
56b2de80 DC |
2200 | STATIC int |
2201 | xfs_imap_lookup( | |
2202 | struct xfs_mount *mp, | |
2203 | struct xfs_trans *tp, | |
2204 | xfs_agnumber_t agno, | |
2205 | xfs_agino_t agino, | |
2206 | xfs_agblock_t agbno, | |
2207 | xfs_agblock_t *chunk_agbno, | |
2208 | xfs_agblock_t *offset_agbno, | |
2209 | int flags) | |
2210 | { | |
2211 | struct xfs_inobt_rec_incore rec; | |
2212 | struct xfs_btree_cur *cur; | |
2213 | struct xfs_buf *agbp; | |
2214 | int error; | |
2215 | int i; | |
2216 | ||
2217 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); | |
2218 | if (error) { | |
a2ceac1f DC |
2219 | xfs_alert(mp, |
2220 | "%s: xfs_ialloc_read_agi() returned error %d, agno %d", | |
2221 | __func__, error, agno); | |
56b2de80 DC |
2222 | return error; |
2223 | } | |
2224 | ||
2225 | /* | |
2226 | * Lookup the inode record for the given agino. If the record cannot be | |
2227 | * found, then it's an invalid inode number and we should abort. Once | |
2228 | * we have a record, we need to ensure it contains the inode number | |
2229 | * we are looking up. | |
2230 | */ | |
70eb7337 | 2231 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
56b2de80 DC |
2232 | error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i); |
2233 | if (!error) { | |
2234 | if (i) | |
2235 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
2236 | if (!error && i == 0) | |
12b53197 | 2237 | error = -EINVAL; |
56b2de80 DC |
2238 | } |
2239 | ||
2240 | xfs_trans_brelse(tp, agbp); | |
d4fa7e6c | 2241 | xfs_btree_del_cursor(cur, error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); |
56b2de80 DC |
2242 | if (error) |
2243 | return error; | |
2244 | ||
2245 | /* check that the returned record contains the required inode */ | |
2246 | if (rec.ir_startino > agino || | |
ff105f75 | 2247 | rec.ir_startino + mp->m_ialloc_inos <= agino) |
12b53197 | 2248 | return -EINVAL; |
56b2de80 DC |
2249 | |
2250 | /* for untrusted inodes check it is allocated first */ | |
2251 | if ((flags & XFS_IGET_UNTRUSTED) && | |
2252 | (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino))) | |
12b53197 | 2253 | return -EINVAL; |
56b2de80 DC |
2254 | |
2255 | *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino); | |
2256 | *offset_agbno = agbno - *chunk_agbno; | |
2257 | return 0; | |
2258 | } | |
2bd0ea18 NS |
2259 | |
2260 | /* | |
56b2de80 | 2261 | * Return the location of the inode in imap, for mapping it into a buffer. |
2bd0ea18 | 2262 | */ |
2bd0ea18 | 2263 | int |
56b2de80 DC |
2264 | xfs_imap( |
2265 | xfs_mount_t *mp, /* file system mount structure */ | |
2266 | xfs_trans_t *tp, /* transaction pointer */ | |
2bd0ea18 | 2267 | xfs_ino_t ino, /* inode to locate */ |
56b2de80 DC |
2268 | struct xfs_imap *imap, /* location map structure */ |
2269 | uint flags) /* flags for inode btree lookup */ | |
2bd0ea18 NS |
2270 | { |
2271 | xfs_agblock_t agbno; /* block number of inode in the alloc group */ | |
2bd0ea18 NS |
2272 | xfs_agino_t agino; /* inode number within alloc group */ |
2273 | xfs_agnumber_t agno; /* allocation group number */ | |
2274 | int blks_per_cluster; /* num blocks per inode cluster */ | |
2275 | xfs_agblock_t chunk_agbno; /* first block in inode chunk */ | |
2bd0ea18 | 2276 | xfs_agblock_t cluster_agbno; /* first block in inode cluster */ |
2bd0ea18 | 2277 | int error; /* error code */ |
dfc130f3 | 2278 | int offset; /* index of inode in its buffer */ |
6bddecbc | 2279 | xfs_agblock_t offset_agbno; /* blks from chunk start to inode */ |
2bd0ea18 NS |
2280 | |
2281 | ASSERT(ino != NULLFSINO); | |
56b2de80 | 2282 | |
2bd0ea18 NS |
2283 | /* |
2284 | * Split up the inode number into its parts. | |
2285 | */ | |
2286 | agno = XFS_INO_TO_AGNO(mp, ino); | |
2287 | agino = XFS_INO_TO_AGINO(mp, ino); | |
2288 | agbno = XFS_AGINO_TO_AGBNO(mp, agino); | |
2289 | if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks || | |
63518810 NS |
2290 | ino != XFS_AGINO_TO_INO(mp, agno, agino)) { |
2291 | #ifdef DEBUG | |
56b2de80 DC |
2292 | /* |
2293 | * Don't output diagnostic information for untrusted inodes | |
2294 | * as they can be invalid without implying corruption. | |
2295 | */ | |
2296 | if (flags & XFS_IGET_UNTRUSTED) | |
12b53197 | 2297 | return -EINVAL; |
5000d01d | 2298 | if (agno >= mp->m_sb.sb_agcount) { |
a2ceac1f DC |
2299 | xfs_alert(mp, |
2300 | "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)", | |
2301 | __func__, agno, mp->m_sb.sb_agcount); | |
63518810 NS |
2302 | } |
2303 | if (agbno >= mp->m_sb.sb_agblocks) { | |
a2ceac1f DC |
2304 | xfs_alert(mp, |
2305 | "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)", | |
2306 | __func__, (unsigned long long)agbno, | |
2307 | (unsigned long)mp->m_sb.sb_agblocks); | |
63518810 NS |
2308 | } |
2309 | if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
a2ceac1f DC |
2310 | xfs_alert(mp, |
2311 | "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)", | |
2312 | __func__, ino, | |
2313 | XFS_AGINO_TO_INO(mp, agno, agino)); | |
63518810 | 2314 | } |
5e656dbb | 2315 | xfs_stack_trace(); |
63518810 | 2316 | #endif /* DEBUG */ |
12b53197 | 2317 | return -EINVAL; |
63518810 | 2318 | } |
56b2de80 | 2319 | |
ff105f75 | 2320 | blks_per_cluster = xfs_icluster_size_fsb(mp); |
56b2de80 DC |
2321 | |
2322 | /* | |
2323 | * For bulkstat and handle lookups, we have an untrusted inode number | |
2324 | * that we have to verify is valid. We cannot do this just by reading | |
2325 | * the inode buffer as it may have been unlinked and removed leaving | |
2326 | * inodes in stale state on disk. Hence we have to do a btree lookup | |
2327 | * in all cases where an untrusted inode number is passed. | |
2328 | */ | |
2329 | if (flags & XFS_IGET_UNTRUSTED) { | |
2330 | error = xfs_imap_lookup(mp, tp, agno, agino, agbno, | |
2331 | &chunk_agbno, &offset_agbno, flags); | |
2332 | if (error) | |
2333 | return error; | |
2334 | goto out_map; | |
2335 | } | |
2336 | ||
2337 | /* | |
2338 | * If the inode cluster size is the same as the blocksize or | |
2339 | * smaller we get to the buffer by simple arithmetics. | |
2340 | */ | |
ff105f75 | 2341 | if (blks_per_cluster == 1) { |
2bd0ea18 NS |
2342 | offset = XFS_INO_TO_OFFSET(mp, ino); |
2343 | ASSERT(offset < mp->m_sb.sb_inopblock); | |
56b2de80 DC |
2344 | |
2345 | imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno); | |
2346 | imap->im_len = XFS_FSB_TO_BB(mp, 1); | |
187f8c22 DW |
2347 | imap->im_boffset = (unsigned short)(offset << |
2348 | mp->m_sb.sb_inodelog); | |
2bd0ea18 NS |
2349 | return 0; |
2350 | } | |
56b2de80 DC |
2351 | |
2352 | /* | |
2353 | * If the inode chunks are aligned then use simple maths to | |
2354 | * find the location. Otherwise we have to do a btree | |
2355 | * lookup to find the location. | |
2356 | */ | |
2bd0ea18 NS |
2357 | if (mp->m_inoalign_mask) { |
2358 | offset_agbno = agbno & mp->m_inoalign_mask; | |
2359 | chunk_agbno = agbno - offset_agbno; | |
2360 | } else { | |
56b2de80 DC |
2361 | error = xfs_imap_lookup(mp, tp, agno, agino, agbno, |
2362 | &chunk_agbno, &offset_agbno, flags); | |
2bd0ea18 NS |
2363 | if (error) |
2364 | return error; | |
2bd0ea18 | 2365 | } |
56b2de80 DC |
2366 | |
2367 | out_map: | |
2bd0ea18 NS |
2368 | ASSERT(agbno >= chunk_agbno); |
2369 | cluster_agbno = chunk_agbno + | |
2370 | ((offset_agbno / blks_per_cluster) * blks_per_cluster); | |
2371 | offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) + | |
2372 | XFS_INO_TO_OFFSET(mp, ino); | |
56b2de80 DC |
2373 | |
2374 | imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno); | |
2375 | imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster); | |
919dbc78 | 2376 | imap->im_boffset = (unsigned short)(offset << mp->m_sb.sb_inodelog); |
56b2de80 DC |
2377 | |
2378 | /* | |
2379 | * If the inode number maps to a block outside the bounds | |
2380 | * of the file system then return NULL rather than calling | |
2381 | * read_buf and panicing when we get an error from the | |
2382 | * driver. | |
2383 | */ | |
2384 | if ((imap->im_blkno + imap->im_len) > | |
2385 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) { | |
a2ceac1f DC |
2386 | xfs_alert(mp, |
2387 | "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)", | |
2388 | __func__, (unsigned long long) imap->im_blkno, | |
56b2de80 DC |
2389 | (unsigned long long) imap->im_len, |
2390 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)); | |
12b53197 | 2391 | return -EINVAL; |
56b2de80 | 2392 | } |
2bd0ea18 | 2393 | return 0; |
2bd0ea18 NS |
2394 | } |
2395 | ||
2396 | /* | |
2397 | * Compute and fill in value of m_in_maxlevels. | |
2398 | */ | |
2399 | void | |
2400 | xfs_ialloc_compute_maxlevels( | |
2401 | xfs_mount_t *mp) /* file system mount structure */ | |
2402 | { | |
730e2a19 DW |
2403 | uint inodes; |
2404 | ||
2405 | inodes = (1LL << XFS_INO_AGINO_BITS(mp)) >> XFS_INODES_PER_CHUNK_LOG; | |
2406 | mp->m_in_maxlevels = xfs_btree_compute_maxlevels(mp, mp->m_inobt_mnr, | |
2407 | inodes); | |
2bd0ea18 NS |
2408 | } |
2409 | ||
2410 | /* | |
ff105f75 DC |
2411 | * Log specified fields for the ag hdr (inode section). The growth of the agi |
2412 | * structure over time requires that we interpret the buffer as two logical | |
2413 | * regions delineated by the end of the unlinked list. This is due to the size | |
2414 | * of the hash table and its location in the middle of the agi. | |
2415 | * | |
2416 | * For example, a request to log a field before agi_unlinked and a field after | |
2417 | * agi_unlinked could cause us to log the entire hash table and use an excessive | |
2418 | * amount of log space. To avoid this behavior, log the region up through | |
2419 | * agi_unlinked in one call and the region after agi_unlinked through the end of | |
2420 | * the structure in another. | |
2bd0ea18 NS |
2421 | */ |
2422 | void | |
2423 | xfs_ialloc_log_agi( | |
2424 | xfs_trans_t *tp, /* transaction pointer */ | |
2425 | xfs_buf_t *bp, /* allocation group header buffer */ | |
2426 | int fields) /* bitmask of fields to log */ | |
2427 | { | |
2428 | int first; /* first byte number */ | |
2429 | int last; /* last byte number */ | |
2430 | static const short offsets[] = { /* field starting offsets */ | |
2431 | /* keep in sync with bit definitions */ | |
2432 | offsetof(xfs_agi_t, agi_magicnum), | |
2433 | offsetof(xfs_agi_t, agi_versionnum), | |
2434 | offsetof(xfs_agi_t, agi_seqno), | |
2435 | offsetof(xfs_agi_t, agi_length), | |
2436 | offsetof(xfs_agi_t, agi_count), | |
2437 | offsetof(xfs_agi_t, agi_root), | |
2438 | offsetof(xfs_agi_t, agi_level), | |
2439 | offsetof(xfs_agi_t, agi_freecount), | |
2440 | offsetof(xfs_agi_t, agi_newino), | |
2441 | offsetof(xfs_agi_t, agi_dirino), | |
2442 | offsetof(xfs_agi_t, agi_unlinked), | |
c0a4c227 BF |
2443 | offsetof(xfs_agi_t, agi_free_root), |
2444 | offsetof(xfs_agi_t, agi_free_level), | |
2bd0ea18 NS |
2445 | sizeof(xfs_agi_t) |
2446 | }; | |
2447 | #ifdef DEBUG | |
2448 | xfs_agi_t *agi; /* allocation group header */ | |
2449 | ||
2450 | agi = XFS_BUF_TO_AGI(bp); | |
a2ceac1f | 2451 | ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC)); |
2bd0ea18 | 2452 | #endif |
ff105f75 | 2453 | |
2bd0ea18 | 2454 | /* |
c0a4c227 | 2455 | * Compute byte offsets for the first and last fields in the first |
ff105f75 DC |
2456 | * region and log the agi buffer. This only logs up through |
2457 | * agi_unlinked. | |
2bd0ea18 | 2458 | */ |
c0a4c227 BF |
2459 | if (fields & XFS_AGI_ALL_BITS_R1) { |
2460 | xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R1, | |
2461 | &first, &last); | |
2462 | xfs_trans_log_buf(tp, bp, first, last); | |
2463 | } | |
2464 | ||
2465 | /* | |
ff105f75 DC |
2466 | * Mask off the bits in the first region and calculate the first and |
2467 | * last field offsets for any bits in the second region. | |
c0a4c227 BF |
2468 | */ |
2469 | fields &= ~XFS_AGI_ALL_BITS_R1; | |
2470 | if (fields) { | |
2471 | xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R2, | |
2472 | &first, &last); | |
2473 | xfs_trans_log_buf(tp, bp, first, last); | |
2474 | } | |
2bd0ea18 NS |
2475 | } |
2476 | ||
56b2de80 DC |
2477 | #ifdef DEBUG |
2478 | STATIC void | |
2479 | xfs_check_agi_unlinked( | |
2480 | struct xfs_agi *agi) | |
2481 | { | |
2482 | int i; | |
2483 | ||
2484 | for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++) | |
2485 | ASSERT(agi->agi_unlinked[i]); | |
2486 | } | |
2487 | #else | |
2488 | #define xfs_check_agi_unlinked(agi) | |
2489 | #endif | |
2490 | ||
dd5b876e | 2491 | static bool |
a2ceac1f DC |
2492 | xfs_agi_verify( |
2493 | struct xfs_buf *bp) | |
2494 | { | |
2495 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
2496 | struct xfs_agi *agi = XFS_BUF_TO_AGI(bp); | |
a2ceac1f | 2497 | |
a65d8d29 BF |
2498 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
2499 | if (!uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid)) | |
2500 | return false; | |
2501 | if (!xfs_log_check_lsn(mp, | |
2502 | be64_to_cpu(XFS_BUF_TO_AGI(bp)->agi_lsn))) | |
dd5b876e | 2503 | return false; |
a65d8d29 BF |
2504 | } |
2505 | ||
a2ceac1f DC |
2506 | /* |
2507 | * Validate the magic number of the agi block. | |
2508 | */ | |
dd5b876e DC |
2509 | if (agi->agi_magicnum != cpu_to_be32(XFS_AGI_MAGIC)) |
2510 | return false; | |
2511 | if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum))) | |
2512 | return false; | |
a2ceac1f | 2513 | |
00795aae DW |
2514 | if (be32_to_cpu(agi->agi_level) < 1 || |
2515 | be32_to_cpu(agi->agi_level) > XFS_BTREE_MAXLEVELS) | |
5a35bf2c | 2516 | return false; |
00795aae DW |
2517 | |
2518 | if (xfs_sb_version_hasfinobt(&mp->m_sb) && | |
2519 | (be32_to_cpu(agi->agi_free_level) < 1 || | |
2520 | be32_to_cpu(agi->agi_free_level) > XFS_BTREE_MAXLEVELS)) | |
2521 | return false; | |
2522 | ||
a2ceac1f DC |
2523 | /* |
2524 | * during growfs operations, the perag is not fully initialised, | |
2525 | * so we can't use it for any useful checking. growfs ensures we can't | |
2526 | * use it by using uncached buffers that don't have the perag attached | |
2527 | * so we can detect and avoid this problem. | |
2528 | */ | |
dd5b876e DC |
2529 | if (bp->b_pag && be32_to_cpu(agi->agi_seqno) != bp->b_pag->pag_agno) |
2530 | return false; | |
a2ceac1f | 2531 | |
a2ceac1f | 2532 | xfs_check_agi_unlinked(agi); |
dd5b876e | 2533 | return true; |
a2ceac1f DC |
2534 | } |
2535 | ||
2536 | static void | |
2537 | xfs_agi_read_verify( | |
2538 | struct xfs_buf *bp) | |
2539 | { | |
dd5b876e | 2540 | struct xfs_mount *mp = bp->b_target->bt_mount; |
dd5b876e | 2541 | |
45922933 DC |
2542 | if (xfs_sb_version_hascrc(&mp->m_sb) && |
2543 | !xfs_buf_verify_cksum(bp, XFS_AGI_CRC_OFF)) | |
12b53197 | 2544 | xfs_buf_ioerror(bp, -EFSBADCRC); |
45922933 | 2545 | else if (XFS_TEST_ERROR(!xfs_agi_verify(bp), mp, |
e2a190dd | 2546 | XFS_ERRTAG_IALLOC_READ_AGI)) |
12b53197 | 2547 | xfs_buf_ioerror(bp, -EFSCORRUPTED); |
45922933 DC |
2548 | |
2549 | if (bp->b_error) | |
2550 | xfs_verifier_error(bp); | |
a2ceac1f DC |
2551 | } |
2552 | ||
2553 | static void | |
2554 | xfs_agi_write_verify( | |
2555 | struct xfs_buf *bp) | |
2556 | { | |
dd5b876e DC |
2557 | struct xfs_mount *mp = bp->b_target->bt_mount; |
2558 | struct xfs_buf_log_item *bip = bp->b_fspriv; | |
2559 | ||
2560 | if (!xfs_agi_verify(bp)) { | |
12b53197 | 2561 | xfs_buf_ioerror(bp, -EFSCORRUPTED); |
45922933 | 2562 | xfs_verifier_error(bp); |
dd5b876e DC |
2563 | return; |
2564 | } | |
2565 | ||
2566 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
2567 | return; | |
2568 | ||
2569 | if (bip) | |
2570 | XFS_BUF_TO_AGI(bp)->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn); | |
43b5aeed | 2571 | xfs_buf_update_cksum(bp, XFS_AGI_CRC_OFF); |
a2ceac1f DC |
2572 | } |
2573 | ||
2574 | const struct xfs_buf_ops xfs_agi_buf_ops = { | |
a3fac935 | 2575 | .name = "xfs_agi", |
a2ceac1f DC |
2576 | .verify_read = xfs_agi_read_verify, |
2577 | .verify_write = xfs_agi_write_verify, | |
2578 | }; | |
2579 | ||
2bd0ea18 NS |
2580 | /* |
2581 | * Read in the allocation group header (inode allocation section) | |
2582 | */ | |
2583 | int | |
56b2de80 DC |
2584 | xfs_read_agi( |
2585 | struct xfs_mount *mp, /* file system mount structure */ | |
2586 | struct xfs_trans *tp, /* transaction pointer */ | |
2587 | xfs_agnumber_t agno, /* allocation group number */ | |
2588 | struct xfs_buf **bpp) /* allocation group hdr buf */ | |
2bd0ea18 | 2589 | { |
56b2de80 | 2590 | int error; |
2bd0ea18 | 2591 | |
ff105f75 | 2592 | trace_xfs_read_agi(mp, agno); |
56b2de80 | 2593 | |
ff105f75 | 2594 | ASSERT(agno != NULLAGNUMBER); |
56b2de80 | 2595 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, |
9440d84d | 2596 | XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)), |
a2ceac1f | 2597 | XFS_FSS_TO_BB(mp, 1), 0, bpp, &xfs_agi_buf_ops); |
9440d84d | 2598 | if (error) |
2bd0ea18 | 2599 | return error; |
1b521475 ES |
2600 | if (tp) |
2601 | xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_AGI_BUF); | |
56b2de80 | 2602 | |
a2ceac1f | 2603 | xfs_buf_set_ref(*bpp, XFS_AGI_REF); |
56b2de80 DC |
2604 | return 0; |
2605 | } | |
2606 | ||
2607 | int | |
2608 | xfs_ialloc_read_agi( | |
2609 | struct xfs_mount *mp, /* file system mount structure */ | |
2610 | struct xfs_trans *tp, /* transaction pointer */ | |
2611 | xfs_agnumber_t agno, /* allocation group number */ | |
2612 | struct xfs_buf **bpp) /* allocation group hdr buf */ | |
2613 | { | |
2614 | struct xfs_agi *agi; /* allocation group header */ | |
2615 | struct xfs_perag *pag; /* per allocation group data */ | |
2616 | int error; | |
2617 | ||
ff105f75 DC |
2618 | trace_xfs_ialloc_read_agi(mp, agno); |
2619 | ||
56b2de80 DC |
2620 | error = xfs_read_agi(mp, tp, agno, bpp); |
2621 | if (error) | |
2622 | return error; | |
2623 | ||
2624 | agi = XFS_BUF_TO_AGI(*bpp); | |
2625 | pag = xfs_perag_get(mp, agno); | |
2bd0ea18 | 2626 | if (!pag->pagi_init) { |
6e3140c7 | 2627 | pag->pagi_freecount = be32_to_cpu(agi->agi_freecount); |
cdded3d8 | 2628 | pag->pagi_count = be32_to_cpu(agi->agi_count); |
2bd0ea18 | 2629 | pag->pagi_init = 1; |
9440d84d | 2630 | } |
9440d84d | 2631 | |
56b2de80 DC |
2632 | /* |
2633 | * It's possible for these to be out of sync if | |
2634 | * we are in the middle of a forced shutdown. | |
2635 | */ | |
2636 | ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) || | |
2637 | XFS_FORCED_SHUTDOWN(mp)); | |
2638 | xfs_perag_put(pag); | |
2bd0ea18 NS |
2639 | return 0; |
2640 | } | |
cdded3d8 DC |
2641 | |
2642 | /* | |
2643 | * Read in the agi to initialise the per-ag data in the mount structure | |
2644 | */ | |
2645 | int | |
2646 | xfs_ialloc_pagi_init( | |
2647 | xfs_mount_t *mp, /* file system mount structure */ | |
2648 | xfs_trans_t *tp, /* transaction pointer */ | |
2649 | xfs_agnumber_t agno) /* allocation group number */ | |
2650 | { | |
2651 | xfs_buf_t *bp = NULL; | |
2652 | int error; | |
2653 | ||
5e656dbb BN |
2654 | error = xfs_ialloc_read_agi(mp, tp, agno, &bp); |
2655 | if (error) | |
cdded3d8 DC |
2656 | return error; |
2657 | if (bp) | |
2658 | xfs_trans_brelse(tp, bp); | |
2659 | return 0; | |
2660 | } | |
be3bf25c DW |
2661 | |
2662 | /* Calculate the first and last possible inode number in an AG. */ | |
2663 | void | |
2664 | xfs_ialloc_agino_range( | |
2665 | struct xfs_mount *mp, | |
2666 | xfs_agnumber_t agno, | |
2667 | xfs_agino_t *first, | |
2668 | xfs_agino_t *last) | |
2669 | { | |
2670 | xfs_agblock_t bno; | |
2671 | xfs_agblock_t eoag; | |
2672 | ||
2673 | eoag = xfs_ag_block_count(mp, agno); | |
2674 | ||
2675 | /* | |
2676 | * Calculate the first inode, which will be in the first | |
2677 | * cluster-aligned block after the AGFL. | |
2678 | */ | |
2679 | bno = round_up(XFS_AGFL_BLOCK(mp) + 1, | |
2680 | xfs_ialloc_cluster_alignment(mp)); | |
2681 | *first = XFS_OFFBNO_TO_AGINO(mp, bno, 0); | |
2682 | ||
2683 | /* | |
2684 | * Calculate the last inode, which will be at the end of the | |
2685 | * last (aligned) cluster that can be allocated in the AG. | |
2686 | */ | |
2687 | bno = round_down(eoag, xfs_ialloc_cluster_alignment(mp)); | |
2688 | *last = XFS_OFFBNO_TO_AGINO(mp, bno, 0) - 1; | |
2689 | } | |
2690 | ||
2691 | /* | |
2692 | * Verify that an AG inode number pointer neither points outside the AG | |
2693 | * nor points at static metadata. | |
2694 | */ | |
2695 | bool | |
2696 | xfs_verify_agino( | |
2697 | struct xfs_mount *mp, | |
2698 | xfs_agnumber_t agno, | |
2699 | xfs_agino_t agino) | |
2700 | { | |
2701 | xfs_agino_t first; | |
2702 | xfs_agino_t last; | |
2703 | ||
2704 | xfs_ialloc_agino_range(mp, agno, &first, &last); | |
2705 | return agino >= first && agino <= last; | |
2706 | } | |
2707 | ||
2708 | /* | |
2709 | * Verify that an FS inode number pointer neither points outside the | |
2710 | * filesystem nor points at static AG metadata. | |
2711 | */ | |
2712 | bool | |
2713 | xfs_verify_ino( | |
2714 | struct xfs_mount *mp, | |
2715 | xfs_ino_t ino) | |
2716 | { | |
2717 | xfs_agnumber_t agno = XFS_INO_TO_AGNO(mp, ino); | |
2718 | xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino); | |
2719 | ||
2720 | if (agno >= mp->m_sb.sb_agcount) | |
2721 | return false; | |
2722 | if (XFS_AGINO_TO_INO(mp, agno, agino) != ino) | |
2723 | return false; | |
2724 | return xfs_verify_agino(mp, agno, agino); | |
2725 | } | |
2726 | ||
2727 | /* Is this an internal inode number? */ | |
2728 | bool | |
2729 | xfs_internal_inum( | |
2730 | struct xfs_mount *mp, | |
2731 | xfs_ino_t ino) | |
2732 | { | |
2733 | return ino == mp->m_sb.sb_rbmino || ino == mp->m_sb.sb_rsumino || | |
2734 | (xfs_sb_version_hasquota(&mp->m_sb) && | |
2735 | xfs_is_quota_inode(&mp->m_sb, ino)); | |
2736 | } | |
2737 | ||
2738 | /* | |
2739 | * Verify that a directory entry's inode number doesn't point at an internal | |
2740 | * inode, empty space, or static AG metadata. | |
2741 | */ | |
2742 | bool | |
2743 | xfs_verify_dir_ino( | |
2744 | struct xfs_mount *mp, | |
2745 | xfs_ino_t ino) | |
2746 | { | |
2747 | if (xfs_internal_inum(mp, ino)) | |
2748 | return false; | |
2749 | return xfs_verify_ino(mp, ino); | |
2750 | } |