]>
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" | |
56d3fc2b | 33 | #include "xfs_errortag.h" |
b626fb59 DC |
34 | #include "xfs_bmap.h" |
35 | #include "xfs_cksum.h" | |
36 | #include "xfs_trans.h" | |
37 | #include "xfs_trace.h" | |
85aec44f | 38 | #include "xfs_rmap.h" |
2bd0ea18 | 39 | |
2bd0ea18 NS |
40 | |
41 | /* | |
42 | * Allocation group level functions. | |
43 | */ | |
b5f6747c | 44 | int |
5e656dbb | 45 | xfs_ialloc_cluster_alignment( |
5a35bf2c | 46 | struct xfs_mount *mp) |
5e656dbb | 47 | { |
5a35bf2c | 48 | if (xfs_sb_version_hasalign(&mp->m_sb) && |
e32fb241 | 49 | mp->m_sb.sb_inoalignmt >= xfs_icluster_size_fsb(mp)) |
5a35bf2c | 50 | return mp->m_sb.sb_inoalignmt; |
5e656dbb BN |
51 | return 1; |
52 | } | |
2bd0ea18 | 53 | |
b194c7d8 | 54 | /* |
56b2de80 | 55 | * Lookup a record by ino in the btree given by cur. |
b194c7d8 BN |
56 | */ |
57 | int /* error */ | |
56b2de80 | 58 | xfs_inobt_lookup( |
b194c7d8 BN |
59 | struct xfs_btree_cur *cur, /* btree cursor */ |
60 | xfs_agino_t ino, /* starting inode of chunk */ | |
56b2de80 | 61 | xfs_lookup_t dir, /* <=, >=, == */ |
b194c7d8 BN |
62 | int *stat) /* success/failure */ |
63 | { | |
64 | cur->bc_rec.i.ir_startino = ino; | |
11640e30 BF |
65 | cur->bc_rec.i.ir_holemask = 0; |
66 | cur->bc_rec.i.ir_count = 0; | |
56b2de80 DC |
67 | cur->bc_rec.i.ir_freecount = 0; |
68 | cur->bc_rec.i.ir_free = 0; | |
69 | return xfs_btree_lookup(cur, dir, stat); | |
b194c7d8 BN |
70 | } |
71 | ||
72 | /* | |
56b2de80 | 73 | * Update the record referred to by cur to the value given. |
b194c7d8 BN |
74 | * This either works (return 0) or gets an EFSCORRUPTED error. |
75 | */ | |
76 | STATIC int /* error */ | |
77 | xfs_inobt_update( | |
78 | struct xfs_btree_cur *cur, /* btree cursor */ | |
56b2de80 | 79 | xfs_inobt_rec_incore_t *irec) /* btree record */ |
b194c7d8 BN |
80 | { |
81 | union xfs_btree_rec rec; | |
82 | ||
56b2de80 | 83 | rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino); |
11640e30 BF |
84 | if (xfs_sb_version_hassparseinodes(&cur->bc_mp->m_sb)) { |
85 | rec.inobt.ir_u.sp.ir_holemask = cpu_to_be16(irec->ir_holemask); | |
86 | rec.inobt.ir_u.sp.ir_count = irec->ir_count; | |
87 | rec.inobt.ir_u.sp.ir_freecount = irec->ir_freecount; | |
88 | } else { | |
89 | /* ir_holemask/ir_count not supported on-disk */ | |
90 | rec.inobt.ir_u.f.ir_freecount = cpu_to_be32(irec->ir_freecount); | |
91 | } | |
56b2de80 | 92 | rec.inobt.ir_free = cpu_to_be64(irec->ir_free); |
b194c7d8 BN |
93 | return xfs_btree_update(cur, &rec); |
94 | } | |
95 | ||
b5f6747c DW |
96 | /* Convert on-disk btree record to incore inobt record. */ |
97 | void | |
98 | xfs_inobt_btrec_to_irec( | |
99 | struct xfs_mount *mp, | |
100 | union xfs_btree_rec *rec, | |
101 | struct xfs_inobt_rec_incore *irec) | |
b194c7d8 | 102 | { |
11640e30 | 103 | irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino); |
b5f6747c | 104 | if (xfs_sb_version_hassparseinodes(&mp->m_sb)) { |
11640e30 BF |
105 | irec->ir_holemask = be16_to_cpu(rec->inobt.ir_u.sp.ir_holemask); |
106 | irec->ir_count = rec->inobt.ir_u.sp.ir_count; | |
107 | irec->ir_freecount = rec->inobt.ir_u.sp.ir_freecount; | |
108 | } else { | |
109 | /* | |
110 | * ir_holemask/ir_count not supported on-disk. Fill in hardcoded | |
111 | * values for full inode chunks. | |
112 | */ | |
113 | irec->ir_holemask = XFS_INOBT_HOLEMASK_FULL; | |
114 | irec->ir_count = XFS_INODES_PER_CHUNK; | |
115 | irec->ir_freecount = | |
116 | be32_to_cpu(rec->inobt.ir_u.f.ir_freecount); | |
b194c7d8 | 117 | } |
11640e30 | 118 | irec->ir_free = be64_to_cpu(rec->inobt.ir_free); |
b5f6747c DW |
119 | } |
120 | ||
121 | /* | |
122 | * Get the data from the pointed-to record. | |
123 | */ | |
124 | int | |
125 | xfs_inobt_get_rec( | |
126 | struct xfs_btree_cur *cur, | |
127 | struct xfs_inobt_rec_incore *irec, | |
128 | int *stat) | |
129 | { | |
130 | union xfs_btree_rec *rec; | |
131 | int error; | |
132 | ||
133 | error = xfs_btree_get_rec(cur, &rec, stat); | |
134 | if (error || *stat == 0) | |
135 | return error; | |
136 | ||
137 | xfs_inobt_btrec_to_irec(cur->bc_mp, rec, irec); | |
11640e30 BF |
138 | |
139 | return 0; | |
b194c7d8 BN |
140 | } |
141 | ||
3c699279 BF |
142 | /* |
143 | * Insert a single inobt record. Cursor must already point to desired location. | |
144 | */ | |
145 | STATIC int | |
146 | xfs_inobt_insert_rec( | |
147 | struct xfs_btree_cur *cur, | |
4a492e72 DW |
148 | uint16_t holemask, |
149 | uint8_t count, | |
150 | int32_t freecount, | |
3c699279 BF |
151 | xfs_inofree_t free, |
152 | int *stat) | |
153 | { | |
11640e30 BF |
154 | cur->bc_rec.i.ir_holemask = holemask; |
155 | cur->bc_rec.i.ir_count = count; | |
3c699279 BF |
156 | cur->bc_rec.i.ir_freecount = freecount; |
157 | cur->bc_rec.i.ir_free = free; | |
158 | return xfs_btree_insert(cur, stat); | |
159 | } | |
160 | ||
161 | /* | |
162 | * Insert records describing a newly allocated inode chunk into the inobt. | |
163 | */ | |
164 | STATIC int | |
165 | xfs_inobt_insert( | |
166 | struct xfs_mount *mp, | |
167 | struct xfs_trans *tp, | |
168 | struct xfs_buf *agbp, | |
169 | xfs_agino_t newino, | |
170 | xfs_agino_t newlen, | |
171 | xfs_btnum_t btnum) | |
172 | { | |
173 | struct xfs_btree_cur *cur; | |
174 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
175 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
176 | xfs_agino_t thisino; | |
177 | int i; | |
178 | int error; | |
179 | ||
180 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum); | |
181 | ||
182 | for (thisino = newino; | |
183 | thisino < newino + newlen; | |
184 | thisino += XFS_INODES_PER_CHUNK) { | |
185 | error = xfs_inobt_lookup(cur, thisino, XFS_LOOKUP_EQ, &i); | |
186 | if (error) { | |
187 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
188 | return error; | |
189 | } | |
190 | ASSERT(i == 0); | |
191 | ||
11640e30 BF |
192 | error = xfs_inobt_insert_rec(cur, XFS_INOBT_HOLEMASK_FULL, |
193 | XFS_INODES_PER_CHUNK, | |
194 | XFS_INODES_PER_CHUNK, | |
3c699279 BF |
195 | XFS_INOBT_ALL_FREE, &i); |
196 | if (error) { | |
197 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
198 | return error; | |
199 | } | |
200 | ASSERT(i == 1); | |
201 | } | |
202 | ||
203 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
204 | ||
205 | return 0; | |
206 | } | |
207 | ||
56b2de80 DC |
208 | /* |
209 | * Verify that the number of free inodes in the AGI is correct. | |
210 | */ | |
211 | #ifdef DEBUG | |
212 | STATIC int | |
213 | xfs_check_agi_freecount( | |
214 | struct xfs_btree_cur *cur, | |
215 | struct xfs_agi *agi) | |
216 | { | |
217 | if (cur->bc_nlevels == 1) { | |
218 | xfs_inobt_rec_incore_t rec; | |
219 | int freecount = 0; | |
220 | int error; | |
221 | int i; | |
222 | ||
223 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); | |
224 | if (error) | |
225 | return error; | |
226 | ||
227 | do { | |
228 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
229 | if (error) | |
230 | return error; | |
231 | ||
232 | if (i) { | |
233 | freecount += rec.ir_freecount; | |
234 | error = xfs_btree_increment(cur, 0, &i); | |
235 | if (error) | |
236 | return error; | |
237 | } | |
238 | } while (i == 1); | |
239 | ||
240 | if (!XFS_FORCED_SHUTDOWN(cur->bc_mp)) | |
241 | ASSERT(freecount == be32_to_cpu(agi->agi_freecount)); | |
242 | } | |
243 | return 0; | |
244 | } | |
245 | #else | |
246 | #define xfs_check_agi_freecount(cur, agi) 0 | |
247 | #endif | |
248 | ||
249 | /* | |
e9d35108 DC |
250 | * Initialise a new set of inodes. When called without a transaction context |
251 | * (e.g. from recovery) we initiate a delayed write of the inode buffers rather | |
252 | * than logging them (which in a transaction context puts them into the AIL | |
253 | * for writeback rather than the xfsbufd queue). | |
56b2de80 | 254 | */ |
e9d35108 | 255 | int |
56b2de80 DC |
256 | xfs_ialloc_inode_init( |
257 | struct xfs_mount *mp, | |
258 | struct xfs_trans *tp, | |
e9d35108 | 259 | struct list_head *buffer_list, |
fe8d48ac | 260 | int icount, |
56b2de80 DC |
261 | xfs_agnumber_t agno, |
262 | xfs_agblock_t agbno, | |
263 | xfs_agblock_t length, | |
264 | unsigned int gen) | |
265 | { | |
266 | struct xfs_buf *fbuf; | |
267 | struct xfs_dinode *free; | |
ff105f75 | 268 | int nbufs, blks_per_cluster, inodes_per_cluster; |
56b2de80 DC |
269 | int version; |
270 | int i, j; | |
271 | xfs_daddr_t d; | |
41ce5f36 | 272 | xfs_ino_t ino = 0; |
56b2de80 DC |
273 | |
274 | /* | |
ff105f75 DC |
275 | * Loop over the new block(s), filling in the inodes. For small block |
276 | * sizes, manipulate the inodes in buffers which are multiples of the | |
277 | * blocks size. | |
56b2de80 | 278 | */ |
ff105f75 DC |
279 | blks_per_cluster = xfs_icluster_size_fsb(mp); |
280 | inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog; | |
281 | nbufs = length / blks_per_cluster; | |
56b2de80 DC |
282 | |
283 | /* | |
e9d35108 DC |
284 | * Figure out what version number to use in the inodes we create. If |
285 | * the superblock version has caught up to the one that supports the new | |
286 | * inode format, then use the new inode version. Otherwise use the old | |
287 | * version so that old kernels will continue to be able to use the file | |
288 | * system. | |
41ce5f36 DC |
289 | * |
290 | * For v3 inodes, we also need to write the inode number into the inode, | |
291 | * so calculate the first inode number of the chunk here as | |
e9d35108 DC |
292 | * XFS_OFFBNO_TO_AGINO() only works within a filesystem block, not |
293 | * across multiple filesystem blocks (such as a cluster) and so cannot | |
294 | * be used in the cluster buffer loop below. | |
295 | * | |
296 | * Further, because we are writing the inode directly into the buffer | |
297 | * and calculating a CRC on the entire inode, we have ot log the entire | |
298 | * inode so that the entire range the CRC covers is present in the log. | |
299 | * That means for v3 inode we log the entire buffer rather than just the | |
300 | * inode cores. | |
56b2de80 | 301 | */ |
41ce5f36 DC |
302 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
303 | version = 3; | |
304 | ino = XFS_AGINO_TO_INO(mp, agno, | |
305 | XFS_OFFBNO_TO_AGINO(mp, agbno, 0)); | |
e9d35108 DC |
306 | |
307 | /* | |
308 | * log the initialisation that is about to take place as an | |
309 | * logical operation. This means the transaction does not | |
310 | * need to log the physical changes to the inode buffers as log | |
311 | * recovery will know what initialisation is actually needed. | |
312 | * Hence we only need to log the buffers as "ordered" buffers so | |
313 | * they track in the AIL as if they were physically logged. | |
314 | */ | |
315 | if (tp) | |
fe8d48ac | 316 | xfs_icreate_log(tp, agno, agbno, icount, |
e9d35108 | 317 | mp->m_sb.sb_inodesize, length, gen); |
ff105f75 | 318 | } else |
56b2de80 | 319 | version = 2; |
56b2de80 DC |
320 | |
321 | for (j = 0; j < nbufs; j++) { | |
322 | /* | |
323 | * Get the block. | |
324 | */ | |
325 | d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster)); | |
326 | fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, | |
327 | mp->m_bsize * blks_per_cluster, | |
a2ceac1f DC |
328 | XBF_UNMAPPED); |
329 | if (!fbuf) | |
12b53197 | 330 | return -ENOMEM; |
e9d35108 DC |
331 | |
332 | /* Initialize the inode buffers and log them appropriately. */ | |
a2ceac1f | 333 | fbuf->b_ops = &xfs_inode_buf_ops; |
e9d35108 | 334 | xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length)); |
ff105f75 | 335 | for (i = 0; i < inodes_per_cluster; i++) { |
56b2de80 | 336 | int ioffset = i << mp->m_sb.sb_inodelog; |
41ce5f36 | 337 | uint isize = xfs_dinode_size(version); |
56b2de80 DC |
338 | |
339 | free = xfs_make_iptr(mp, fbuf, i); | |
340 | free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC); | |
341 | free->di_version = version; | |
342 | free->di_gen = cpu_to_be32(gen); | |
343 | free->di_next_unlinked = cpu_to_be32(NULLAGINO); | |
41ce5f36 DC |
344 | |
345 | if (version == 3) { | |
346 | free->di_ino = cpu_to_be64(ino); | |
347 | ino++; | |
9c4e12fb ES |
348 | uuid_copy(&free->di_uuid, |
349 | &mp->m_sb.sb_meta_uuid); | |
41ce5f36 | 350 | xfs_dinode_calc_crc(mp, free); |
e9d35108 DC |
351 | } else if (tp) { |
352 | /* just log the inode core */ | |
353 | xfs_trans_log_buf(tp, fbuf, ioffset, | |
354 | ioffset + isize - 1); | |
41ce5f36 | 355 | } |
e9d35108 | 356 | } |
41ce5f36 | 357 | |
e9d35108 DC |
358 | if (tp) { |
359 | /* | |
360 | * Mark the buffer as an inode allocation buffer so it | |
361 | * sticks in AIL at the point of this allocation | |
362 | * transaction. This ensures the they are on disk before | |
363 | * the tail of the log can be moved past this | |
364 | * transaction (i.e. by preventing relogging from moving | |
365 | * it forward in the log). | |
366 | */ | |
367 | xfs_trans_inode_alloc_buf(tp, fbuf); | |
368 | if (version == 3) { | |
369 | /* | |
370 | * Mark the buffer as ordered so that they are | |
371 | * not physically logged in the transaction but | |
372 | * still tracked in the AIL as part of the | |
373 | * transaction and pin the log appropriately. | |
374 | */ | |
375 | xfs_trans_ordered_buf(tp, fbuf); | |
e9d35108 DC |
376 | } |
377 | } else { | |
378 | fbuf->b_flags |= XBF_DONE; | |
379 | xfs_buf_delwri_queue(fbuf, buffer_list); | |
380 | xfs_buf_relse(fbuf); | |
56b2de80 | 381 | } |
56b2de80 | 382 | } |
a2ceac1f | 383 | return 0; |
56b2de80 DC |
384 | } |
385 | ||
6f4c54a4 BF |
386 | /* |
387 | * Align startino and allocmask for a recently allocated sparse chunk such that | |
388 | * they are fit for insertion (or merge) into the on-disk inode btrees. | |
389 | * | |
390 | * Background: | |
391 | * | |
392 | * When enabled, sparse inode support increases the inode alignment from cluster | |
393 | * size to inode chunk size. This means that the minimum range between two | |
394 | * non-adjacent inode records in the inobt is large enough for a full inode | |
395 | * record. This allows for cluster sized, cluster aligned block allocation | |
396 | * without need to worry about whether the resulting inode record overlaps with | |
397 | * another record in the tree. Without this basic rule, we would have to deal | |
398 | * with the consequences of overlap by potentially undoing recent allocations in | |
399 | * the inode allocation codepath. | |
400 | * | |
401 | * Because of this alignment rule (which is enforced on mount), there are two | |
402 | * inobt possibilities for newly allocated sparse chunks. One is that the | |
403 | * aligned inode record for the chunk covers a range of inodes not already | |
404 | * covered in the inobt (i.e., it is safe to insert a new sparse record). The | |
405 | * other is that a record already exists at the aligned startino that considers | |
406 | * the newly allocated range as sparse. In the latter case, record content is | |
407 | * merged in hope that sparse inode chunks fill to full chunks over time. | |
408 | */ | |
409 | STATIC void | |
410 | xfs_align_sparse_ino( | |
411 | struct xfs_mount *mp, | |
412 | xfs_agino_t *startino, | |
413 | uint16_t *allocmask) | |
414 | { | |
415 | xfs_agblock_t agbno; | |
416 | xfs_agblock_t mod; | |
417 | int offset; | |
418 | ||
419 | agbno = XFS_AGINO_TO_AGBNO(mp, *startino); | |
420 | mod = agbno % mp->m_sb.sb_inoalignmt; | |
421 | if (!mod) | |
422 | return; | |
423 | ||
424 | /* calculate the inode offset and align startino */ | |
425 | offset = mod << mp->m_sb.sb_inopblog; | |
426 | *startino -= offset; | |
427 | ||
428 | /* | |
429 | * Since startino has been aligned down, left shift allocmask such that | |
430 | * it continues to represent the same physical inodes relative to the | |
431 | * new startino. | |
432 | */ | |
433 | *allocmask <<= offset / XFS_INODES_PER_HOLEMASK_BIT; | |
434 | } | |
435 | ||
436 | /* | |
437 | * Determine whether the source inode record can merge into the target. Both | |
438 | * records must be sparse, the inode ranges must match and there must be no | |
439 | * allocation overlap between the records. | |
440 | */ | |
441 | STATIC bool | |
442 | __xfs_inobt_can_merge( | |
443 | struct xfs_inobt_rec_incore *trec, /* tgt record */ | |
444 | struct xfs_inobt_rec_incore *srec) /* src record */ | |
445 | { | |
446 | uint64_t talloc; | |
447 | uint64_t salloc; | |
448 | ||
449 | /* records must cover the same inode range */ | |
450 | if (trec->ir_startino != srec->ir_startino) | |
451 | return false; | |
452 | ||
453 | /* both records must be sparse */ | |
454 | if (!xfs_inobt_issparse(trec->ir_holemask) || | |
455 | !xfs_inobt_issparse(srec->ir_holemask)) | |
456 | return false; | |
457 | ||
458 | /* both records must track some inodes */ | |
459 | if (!trec->ir_count || !srec->ir_count) | |
460 | return false; | |
461 | ||
462 | /* can't exceed capacity of a full record */ | |
463 | if (trec->ir_count + srec->ir_count > XFS_INODES_PER_CHUNK) | |
464 | return false; | |
465 | ||
466 | /* verify there is no allocation overlap */ | |
467 | talloc = xfs_inobt_irec_to_allocmask(trec); | |
468 | salloc = xfs_inobt_irec_to_allocmask(srec); | |
469 | if (talloc & salloc) | |
470 | return false; | |
471 | ||
472 | return true; | |
473 | } | |
474 | ||
475 | /* | |
476 | * Merge the source inode record into the target. The caller must call | |
477 | * __xfs_inobt_can_merge() to ensure the merge is valid. | |
478 | */ | |
479 | STATIC void | |
480 | __xfs_inobt_rec_merge( | |
481 | struct xfs_inobt_rec_incore *trec, /* target */ | |
482 | struct xfs_inobt_rec_incore *srec) /* src */ | |
483 | { | |
484 | ASSERT(trec->ir_startino == srec->ir_startino); | |
485 | ||
486 | /* combine the counts */ | |
487 | trec->ir_count += srec->ir_count; | |
488 | trec->ir_freecount += srec->ir_freecount; | |
489 | ||
490 | /* | |
491 | * Merge the holemask and free mask. For both fields, 0 bits refer to | |
492 | * allocated inodes. We combine the allocated ranges with bitwise AND. | |
493 | */ | |
494 | trec->ir_holemask &= srec->ir_holemask; | |
495 | trec->ir_free &= srec->ir_free; | |
496 | } | |
497 | ||
498 | /* | |
499 | * Insert a new sparse inode chunk into the associated inode btree. The inode | |
500 | * record for the sparse chunk is pre-aligned to a startino that should match | |
501 | * any pre-existing sparse inode record in the tree. This allows sparse chunks | |
502 | * to fill over time. | |
503 | * | |
504 | * This function supports two modes of handling preexisting records depending on | |
505 | * the merge flag. If merge is true, the provided record is merged with the | |
506 | * existing record and updated in place. The merged record is returned in nrec. | |
507 | * If merge is false, an existing record is replaced with the provided record. | |
508 | * If no preexisting record exists, the provided record is always inserted. | |
509 | * | |
510 | * It is considered corruption if a merge is requested and not possible. Given | |
511 | * the sparse inode alignment constraints, this should never happen. | |
512 | */ | |
513 | STATIC int | |
514 | xfs_inobt_insert_sprec( | |
515 | struct xfs_mount *mp, | |
516 | struct xfs_trans *tp, | |
517 | struct xfs_buf *agbp, | |
518 | int btnum, | |
519 | struct xfs_inobt_rec_incore *nrec, /* in/out: new/merged rec. */ | |
520 | bool merge) /* merge or replace */ | |
521 | { | |
522 | struct xfs_btree_cur *cur; | |
523 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
524 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
525 | int error; | |
526 | int i; | |
527 | struct xfs_inobt_rec_incore rec; | |
528 | ||
529 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum); | |
530 | ||
531 | /* the new record is pre-aligned so we know where to look */ | |
532 | error = xfs_inobt_lookup(cur, nrec->ir_startino, XFS_LOOKUP_EQ, &i); | |
533 | if (error) | |
534 | goto error; | |
535 | /* if nothing there, insert a new record and return */ | |
536 | if (i == 0) { | |
537 | error = xfs_inobt_insert_rec(cur, nrec->ir_holemask, | |
538 | nrec->ir_count, nrec->ir_freecount, | |
539 | nrec->ir_free, &i); | |
540 | if (error) | |
541 | goto error; | |
542 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error); | |
543 | ||
544 | goto out; | |
545 | } | |
546 | ||
547 | /* | |
548 | * A record exists at this startino. Merge or replace the record | |
549 | * depending on what we've been asked to do. | |
550 | */ | |
551 | if (merge) { | |
552 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
553 | if (error) | |
554 | goto error; | |
555 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error); | |
556 | XFS_WANT_CORRUPTED_GOTO(mp, | |
557 | rec.ir_startino == nrec->ir_startino, | |
558 | error); | |
559 | ||
560 | /* | |
561 | * This should never fail. If we have coexisting records that | |
562 | * cannot merge, something is seriously wrong. | |
563 | */ | |
564 | XFS_WANT_CORRUPTED_GOTO(mp, __xfs_inobt_can_merge(nrec, &rec), | |
565 | error); | |
566 | ||
567 | trace_xfs_irec_merge_pre(mp, agno, rec.ir_startino, | |
568 | rec.ir_holemask, nrec->ir_startino, | |
569 | nrec->ir_holemask); | |
570 | ||
571 | /* merge to nrec to output the updated record */ | |
572 | __xfs_inobt_rec_merge(nrec, &rec); | |
573 | ||
574 | trace_xfs_irec_merge_post(mp, agno, nrec->ir_startino, | |
575 | nrec->ir_holemask); | |
576 | ||
577 | error = xfs_inobt_rec_check_count(mp, nrec); | |
578 | if (error) | |
579 | goto error; | |
580 | } | |
581 | ||
582 | error = xfs_inobt_update(cur, nrec); | |
583 | if (error) | |
584 | goto error; | |
585 | ||
586 | out: | |
587 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
588 | return 0; | |
589 | error: | |
590 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
591 | return error; | |
592 | } | |
593 | ||
2bd0ea18 NS |
594 | /* |
595 | * Allocate new inodes in the allocation group specified by agbp. | |
596 | * Return 0 for success, else error code. | |
597 | */ | |
598 | STATIC int /* error code or 0 */ | |
599 | xfs_ialloc_ag_alloc( | |
600 | xfs_trans_t *tp, /* transaction pointer */ | |
601 | xfs_buf_t *agbp, /* alloc group buffer */ | |
602 | int *alloc) | |
603 | { | |
604 | xfs_agi_t *agi; /* allocation group header */ | |
dfc130f3 | 605 | xfs_alloc_arg_t args; /* allocation argument structure */ |
5e656dbb | 606 | xfs_agnumber_t agno; |
2bd0ea18 | 607 | int error; |
2bd0ea18 NS |
608 | xfs_agino_t newino; /* new first inode's number */ |
609 | xfs_agino_t newlen; /* new number of inodes */ | |
5e656dbb | 610 | int isaligned = 0; /* inode allocation at stripe unit */ |
2bd0ea18 | 611 | /* boundary */ |
6f4c54a4 BF |
612 | uint16_t allocmask = (uint16_t) -1; /* init. to full chunk */ |
613 | struct xfs_inobt_rec_incore rec; | |
56b2de80 | 614 | struct xfs_perag *pag; |
c9005f41 BF |
615 | int do_sparse = 0; |
616 | ||
a2ceac1f | 617 | memset(&args, 0, sizeof(args)); |
2bd0ea18 NS |
618 | args.tp = tp; |
619 | args.mp = tp->t_mountp; | |
c9005f41 | 620 | args.fsbno = NULLFSBLOCK; |
85aec44f | 621 | xfs_rmap_ag_owner(&args.oinfo, XFS_RMAP_OWN_INODES); |
2bd0ea18 | 622 | |
0d995d03 BF |
623 | #ifdef DEBUG |
624 | /* randomly do sparse inode allocations */ | |
625 | if (xfs_sb_version_hassparseinodes(&tp->t_mountp->m_sb) && | |
626 | args.mp->m_ialloc_min_blks < args.mp->m_ialloc_blks) | |
627 | do_sparse = prandom_u32() & 1; | |
628 | #endif | |
629 | ||
2bd0ea18 NS |
630 | /* |
631 | * Locking will ensure that we don't have two callers in here | |
632 | * at one time. | |
633 | */ | |
ff105f75 | 634 | newlen = args.mp->m_ialloc_inos; |
2bd0ea18 | 635 | if (args.mp->m_maxicount && |
cbf3beaa | 636 | percpu_counter_read_positive(&args.mp->m_icount) + newlen > |
19ebedcf | 637 | args.mp->m_maxicount) |
12b53197 | 638 | return -ENOSPC; |
ff105f75 | 639 | args.minlen = args.maxlen = args.mp->m_ialloc_blks; |
2bd0ea18 | 640 | /* |
5e656dbb BN |
641 | * First try to allocate inodes contiguous with the last-allocated |
642 | * chunk of inodes. If the filesystem is striped, this will fill | |
643 | * an entire stripe unit with inodes. | |
3439d03a | 644 | */ |
2bd0ea18 | 645 | agi = XFS_BUF_TO_AGI(agbp); |
5e656dbb | 646 | newino = be32_to_cpu(agi->agi_newino); |
56b2de80 | 647 | agno = be32_to_cpu(agi->agi_seqno); |
5e656dbb | 648 | args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) + |
ff105f75 | 649 | args.mp->m_ialloc_blks; |
c9005f41 BF |
650 | if (do_sparse) |
651 | goto sparse_alloc; | |
5e656dbb BN |
652 | if (likely(newino != NULLAGINO && |
653 | (args.agbno < be32_to_cpu(agi->agi_length)))) { | |
56b2de80 | 654 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
5e656dbb | 655 | args.type = XFS_ALLOCTYPE_THIS_BNO; |
5e656dbb BN |
656 | args.prod = 1; |
657 | ||
658 | /* | |
659 | * We need to take into account alignment here to ensure that | |
660 | * we don't modify the free list if we fail to have an exact | |
661 | * block. If we don't have an exact match, and every oher | |
662 | * attempt allocation attempt fails, we'll end up cancelling | |
663 | * a dirty transaction and shutting down. | |
664 | * | |
665 | * For an exact allocation, alignment must be 1, | |
666 | * however we need to take cluster alignment into account when | |
667 | * fixing up the freelist. Use the minalignslop field to | |
668 | * indicate that extra blocks might be required for alignment, | |
669 | * but not to use them in the actual exact allocation. | |
670 | */ | |
671 | args.alignment = 1; | |
5a35bf2c | 672 | args.minalignslop = xfs_ialloc_cluster_alignment(args.mp) - 1; |
5e656dbb BN |
673 | |
674 | /* Allow space for the inode btree to split. */ | |
56b2de80 | 675 | args.minleft = args.mp->m_in_maxlevels - 1; |
5e656dbb BN |
676 | if ((error = xfs_alloc_vextent(&args))) |
677 | return error; | |
ff105f75 DC |
678 | |
679 | /* | |
680 | * This request might have dirtied the transaction if the AG can | |
681 | * satisfy the request, but the exact block was not available. | |
682 | * If the allocation did fail, subsequent requests will relax | |
683 | * the exact agbno requirement and increase the alignment | |
684 | * instead. It is critical that the total size of the request | |
685 | * (len + alignment + slop) does not increase from this point | |
686 | * on, so reset minalignslop to ensure it is not included in | |
687 | * subsequent requests. | |
688 | */ | |
689 | args.minalignslop = 0; | |
c9005f41 | 690 | } |
5e656dbb BN |
691 | |
692 | if (unlikely(args.fsbno == NULLFSBLOCK)) { | |
693 | /* | |
694 | * Set the alignment for the allocation. | |
695 | * If stripe alignment is turned on then align at stripe unit | |
696 | * boundary. | |
697 | * If the cluster size is smaller than a filesystem block | |
698 | * then we're doing I/O for inodes in filesystem block size | |
699 | * pieces, so don't need alignment anyway. | |
700 | */ | |
701 | isaligned = 0; | |
702 | if (args.mp->m_sinoalign) { | |
703 | ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN)); | |
704 | args.alignment = args.mp->m_dalign; | |
705 | isaligned = 1; | |
706 | } else | |
5a35bf2c | 707 | args.alignment = xfs_ialloc_cluster_alignment(args.mp); |
5e656dbb BN |
708 | /* |
709 | * Need to figure out where to allocate the inode blocks. | |
710 | * Ideally they should be spaced out through the a.g. | |
711 | * For now, just allocate blocks up front. | |
712 | */ | |
713 | args.agbno = be32_to_cpu(agi->agi_root); | |
56b2de80 | 714 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
5e656dbb BN |
715 | /* |
716 | * Allocate a fixed-size extent of inodes. | |
717 | */ | |
718 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
5e656dbb BN |
719 | args.prod = 1; |
720 | /* | |
721 | * Allow space for the inode btree to split. | |
722 | */ | |
56b2de80 | 723 | args.minleft = args.mp->m_in_maxlevels - 1; |
5e656dbb BN |
724 | if ((error = xfs_alloc_vextent(&args))) |
725 | return error; | |
726 | } | |
2bd0ea18 NS |
727 | |
728 | /* | |
729 | * If stripe alignment is turned on, then try again with cluster | |
730 | * alignment. | |
731 | */ | |
732 | if (isaligned && args.fsbno == NULLFSBLOCK) { | |
733 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
6e3140c7 | 734 | args.agbno = be32_to_cpu(agi->agi_root); |
56b2de80 | 735 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
5a35bf2c | 736 | args.alignment = xfs_ialloc_cluster_alignment(args.mp); |
0e266570 | 737 | if ((error = xfs_alloc_vextent(&args))) |
dfc130f3 | 738 | return error; |
2bd0ea18 | 739 | } |
5000d01d | 740 | |
6f4c54a4 BF |
741 | /* |
742 | * Finally, try a sparse allocation if the filesystem supports it and | |
743 | * the sparse allocation length is smaller than a full chunk. | |
744 | */ | |
745 | if (xfs_sb_version_hassparseinodes(&args.mp->m_sb) && | |
746 | args.mp->m_ialloc_min_blks < args.mp->m_ialloc_blks && | |
747 | args.fsbno == NULLFSBLOCK) { | |
c9005f41 | 748 | sparse_alloc: |
6f4c54a4 BF |
749 | args.type = XFS_ALLOCTYPE_NEAR_BNO; |
750 | args.agbno = be32_to_cpu(agi->agi_root); | |
751 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); | |
752 | args.alignment = args.mp->m_sb.sb_spino_align; | |
753 | args.prod = 1; | |
754 | ||
755 | args.minlen = args.mp->m_ialloc_min_blks; | |
756 | args.maxlen = args.minlen; | |
757 | ||
758 | /* | |
759 | * The inode record will be aligned to full chunk size. We must | |
760 | * prevent sparse allocation from AG boundaries that result in | |
761 | * invalid inode records, such as records that start at agbno 0 | |
762 | * or extend beyond the AG. | |
763 | * | |
764 | * Set min agbno to the first aligned, non-zero agbno and max to | |
765 | * the last aligned agbno that is at least one full chunk from | |
766 | * the end of the AG. | |
767 | */ | |
768 | args.min_agbno = args.mp->m_sb.sb_inoalignmt; | |
769 | args.max_agbno = round_down(args.mp->m_sb.sb_agblocks, | |
770 | args.mp->m_sb.sb_inoalignmt) - | |
771 | args.mp->m_ialloc_blks; | |
772 | ||
773 | error = xfs_alloc_vextent(&args); | |
774 | if (error) | |
775 | return error; | |
776 | ||
777 | newlen = args.len << args.mp->m_sb.sb_inopblog; | |
0d995d03 | 778 | ASSERT(newlen <= XFS_INODES_PER_CHUNK); |
6f4c54a4 BF |
779 | allocmask = (1 << (newlen / XFS_INODES_PER_HOLEMASK_BIT)) - 1; |
780 | } | |
781 | ||
2bd0ea18 NS |
782 | if (args.fsbno == NULLFSBLOCK) { |
783 | *alloc = 0; | |
784 | return 0; | |
785 | } | |
786 | ASSERT(args.len == args.minlen); | |
a562a63b | 787 | |
5e656dbb | 788 | /* |
56b2de80 DC |
789 | * Stamp and write the inode buffers. |
790 | * | |
5e656dbb BN |
791 | * Seed the new inode cluster with a random generation number. This |
792 | * prevents short-term reuse of generation numbers if a chunk is | |
793 | * freed and then immediately reallocated. We use random numbers | |
794 | * rather than a linear progression to prevent the next generation | |
795 | * number from being easily guessable. | |
796 | */ | |
fe8d48ac BF |
797 | error = xfs_ialloc_inode_init(args.mp, tp, NULL, newlen, agno, |
798 | args.agbno, args.len, prandom_u32()); | |
56b2de80 | 799 | |
a2ceac1f DC |
800 | if (error) |
801 | return error; | |
56b2de80 DC |
802 | /* |
803 | * Convert the results. | |
804 | */ | |
805 | newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0); | |
6f4c54a4 BF |
806 | |
807 | if (xfs_inobt_issparse(~allocmask)) { | |
808 | /* | |
809 | * We've allocated a sparse chunk. Align the startino and mask. | |
810 | */ | |
811 | xfs_align_sparse_ino(args.mp, &newino, &allocmask); | |
812 | ||
813 | rec.ir_startino = newino; | |
814 | rec.ir_holemask = ~allocmask; | |
815 | rec.ir_count = newlen; | |
816 | rec.ir_freecount = newlen; | |
817 | rec.ir_free = XFS_INOBT_ALL_FREE; | |
818 | ||
819 | /* | |
820 | * Insert the sparse record into the inobt and allow for a merge | |
821 | * if necessary. If a merge does occur, rec is updated to the | |
822 | * merged record. | |
823 | */ | |
824 | error = xfs_inobt_insert_sprec(args.mp, tp, agbp, XFS_BTNUM_INO, | |
825 | &rec, true); | |
826 | if (error == -EFSCORRUPTED) { | |
827 | xfs_alert(args.mp, | |
828 | "invalid sparse inode record: ino 0x%llx holemask 0x%x count %u", | |
829 | XFS_AGINO_TO_INO(args.mp, agno, | |
830 | rec.ir_startino), | |
831 | rec.ir_holemask, rec.ir_count); | |
832 | xfs_force_shutdown(args.mp, SHUTDOWN_CORRUPT_INCORE); | |
833 | } | |
834 | if (error) | |
835 | return error; | |
836 | ||
837 | /* | |
838 | * We can't merge the part we've just allocated as for the inobt | |
839 | * due to finobt semantics. The original record may or may not | |
840 | * exist independent of whether physical inodes exist in this | |
841 | * sparse chunk. | |
842 | * | |
843 | * We must update the finobt record based on the inobt record. | |
844 | * rec contains the fully merged and up to date inobt record | |
845 | * from the previous call. Set merge false to replace any | |
846 | * existing record with this one. | |
847 | */ | |
848 | if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) { | |
849 | error = xfs_inobt_insert_sprec(args.mp, tp, agbp, | |
850 | XFS_BTNUM_FINO, &rec, | |
851 | false); | |
852 | if (error) | |
853 | return error; | |
854 | } | |
855 | } else { | |
856 | /* full chunk - insert new records to both btrees */ | |
857 | error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen, | |
858 | XFS_BTNUM_INO); | |
859 | if (error) | |
860 | return error; | |
861 | ||
862 | if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) { | |
863 | error = xfs_inobt_insert(args.mp, tp, agbp, newino, | |
864 | newlen, XFS_BTNUM_FINO); | |
865 | if (error) | |
866 | return error; | |
867 | } | |
868 | } | |
869 | ||
870 | /* | |
871 | * Update AGI counts and newino. | |
872 | */ | |
5e656dbb BN |
873 | be32_add_cpu(&agi->agi_count, newlen); |
874 | be32_add_cpu(&agi->agi_freecount, newlen); | |
56b2de80 DC |
875 | pag = xfs_perag_get(args.mp, agno); |
876 | pag->pagi_freecount += newlen; | |
877 | xfs_perag_put(pag); | |
6e3140c7 | 878 | agi->agi_newino = cpu_to_be32(newino); |
56b2de80 | 879 | |
2bd0ea18 NS |
880 | /* |
881 | * Log allocation group header fields | |
882 | */ | |
883 | xfs_ialloc_log_agi(tp, agbp, | |
884 | XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO); | |
885 | /* | |
886 | * Modify/log superblock values for inode count and inode free count. | |
887 | */ | |
888 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen); | |
889 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen); | |
890 | *alloc = 1; | |
891 | return 0; | |
892 | } | |
893 | ||
56b2de80 | 894 | STATIC xfs_agnumber_t |
321717ae NS |
895 | xfs_ialloc_next_ag( |
896 | xfs_mount_t *mp) | |
897 | { | |
898 | xfs_agnumber_t agno; | |
899 | ||
900 | spin_lock(&mp->m_agirotor_lock); | |
901 | agno = mp->m_agirotor; | |
a2ceac1f | 902 | if (++mp->m_agirotor >= mp->m_maxagi) |
321717ae NS |
903 | mp->m_agirotor = 0; |
904 | spin_unlock(&mp->m_agirotor_lock); | |
905 | ||
906 | return agno; | |
907 | } | |
908 | ||
2bd0ea18 NS |
909 | /* |
910 | * Select an allocation group to look for a free inode in, based on the parent | |
e6d77a21 | 911 | * inode and the mode. Return the allocation group buffer. |
2bd0ea18 | 912 | */ |
a2ceac1f | 913 | STATIC xfs_agnumber_t |
2bd0ea18 NS |
914 | xfs_ialloc_ag_select( |
915 | xfs_trans_t *tp, /* transaction pointer */ | |
916 | xfs_ino_t parent, /* parent directory inode number */ | |
c1a29e92 | 917 | umode_t mode) /* bits set to indicate file type */ |
2bd0ea18 | 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 (!pag->pagf_init) { |
975 | error = xfs_alloc_pagf_init(mp, tp, agno, flags); | |
976 | if (error) | |
2bd0ea18 | 977 | goto nextag; |
2bd0ea18 | 978 | } |
a2ceac1f DC |
979 | |
980 | /* | |
5a35bf2c DC |
981 | * Check that there is enough free space for the file plus a |
982 | * chunk of inodes if we need to allocate some. If this is the | |
983 | * first pass across the AGs, take into account the potential | |
984 | * space needed for alignment of inode chunks when checking the | |
985 | * longest contiguous free space in the AG - this prevents us | |
986 | * from getting ENOSPC because we have free space larger than | |
987 | * m_ialloc_blks but alignment constraints prevent us from using | |
988 | * it. | |
989 | * | |
990 | * If we can't find an AG with space for full alignment slack to | |
991 | * be taken into account, we must be near ENOSPC in all AGs. | |
992 | * Hence we don't include alignment for the second pass and so | |
993 | * if we fail allocation due to alignment issues then it is most | |
994 | * likely a real ENOSPC condition. | |
a2ceac1f | 995 | */ |
62dc6cdb | 996 | ineed = mp->m_ialloc_min_blks; |
5a35bf2c DC |
997 | if (flags && ineed > 1) |
998 | ineed += xfs_ialloc_cluster_alignment(mp); | |
a2ceac1f DC |
999 | longest = pag->pagf_longest; |
1000 | if (!longest) | |
1001 | longest = pag->pagf_flcount > 0; | |
1002 | ||
1003 | if (pag->pagf_freeblks >= needspace + ineed && | |
1004 | longest >= ineed) { | |
1005 | xfs_perag_put(pag); | |
1006 | return agno; | |
2bd0ea18 | 1007 | } |
5000d01d | 1008 | nextag: |
56b2de80 | 1009 | xfs_perag_put(pag); |
5000d01d | 1010 | /* |
2bd0ea18 NS |
1011 | * No point in iterating over the rest, if we're shutting |
1012 | * down. | |
1013 | */ | |
56b2de80 | 1014 | if (XFS_FORCED_SHUTDOWN(mp)) |
a2ceac1f | 1015 | return NULLAGNUMBER; |
2bd0ea18 | 1016 | agno++; |
5ce1d1f7 | 1017 | if (agno >= agcount) |
2bd0ea18 NS |
1018 | agno = 0; |
1019 | if (agno == pagno) { | |
56b2de80 | 1020 | if (flags == 0) |
a2ceac1f | 1021 | return NULLAGNUMBER; |
2bd0ea18 NS |
1022 | flags = 0; |
1023 | } | |
1024 | } | |
1025 | } | |
1026 | ||
56b2de80 DC |
1027 | /* |
1028 | * Try to retrieve the next record to the left/right from the current one. | |
1029 | */ | |
1030 | STATIC int | |
1031 | xfs_ialloc_next_rec( | |
1032 | struct xfs_btree_cur *cur, | |
1033 | xfs_inobt_rec_incore_t *rec, | |
1034 | int *done, | |
1035 | int left) | |
1036 | { | |
1037 | int error; | |
1038 | int i; | |
1039 | ||
1040 | if (left) | |
1041 | error = xfs_btree_decrement(cur, 0, &i); | |
1042 | else | |
1043 | error = xfs_btree_increment(cur, 0, &i); | |
1044 | ||
1045 | if (error) | |
1046 | return error; | |
1047 | *done = !i; | |
1048 | if (i) { | |
1049 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1050 | if (error) | |
1051 | return error; | |
19ebedcf | 1052 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
56b2de80 DC |
1053 | } |
1054 | ||
1055 | return 0; | |
1056 | } | |
1057 | ||
1058 | STATIC int | |
1059 | xfs_ialloc_get_rec( | |
1060 | struct xfs_btree_cur *cur, | |
1061 | xfs_agino_t agino, | |
1062 | xfs_inobt_rec_incore_t *rec, | |
3439d03a | 1063 | int *done) |
56b2de80 DC |
1064 | { |
1065 | int error; | |
1066 | int i; | |
1067 | ||
1068 | error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i); | |
1069 | if (error) | |
1070 | return error; | |
1071 | *done = !i; | |
1072 | if (i) { | |
1073 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1074 | if (error) | |
1075 | return error; | |
19ebedcf | 1076 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
56b2de80 DC |
1077 | } |
1078 | ||
1079 | return 0; | |
1080 | } | |
1081 | ||
01792d3b | 1082 | /* |
fa58183c BF |
1083 | * Return the offset of the first free inode in the record. If the inode chunk |
1084 | * is sparsely allocated, we convert the record holemask to inode granularity | |
1085 | * and mask off the unallocated regions from the inode free mask. | |
01792d3b BF |
1086 | */ |
1087 | STATIC int | |
1088 | xfs_inobt_first_free_inode( | |
1089 | struct xfs_inobt_rec_incore *rec) | |
1090 | { | |
fa58183c BF |
1091 | xfs_inofree_t realfree; |
1092 | ||
1093 | /* if there are no holes, return the first available offset */ | |
1094 | if (!xfs_inobt_issparse(rec->ir_holemask)) | |
1095 | return xfs_lowbit64(rec->ir_free); | |
1096 | ||
1097 | realfree = xfs_inobt_irec_to_allocmask(rec); | |
1098 | realfree &= rec->ir_free; | |
1099 | ||
1100 | return xfs_lowbit64(realfree); | |
01792d3b BF |
1101 | } |
1102 | ||
5000d01d | 1103 | /* |
ff105f75 | 1104 | * Allocate an inode using the inobt-only algorithm. |
2bd0ea18 | 1105 | */ |
a2ceac1f | 1106 | STATIC int |
ff105f75 | 1107 | xfs_dialloc_ag_inobt( |
a2ceac1f DC |
1108 | struct xfs_trans *tp, |
1109 | struct xfs_buf *agbp, | |
1110 | xfs_ino_t parent, | |
1111 | xfs_ino_t *inop) | |
2bd0ea18 | 1112 | { |
a2ceac1f DC |
1113 | struct xfs_mount *mp = tp->t_mountp; |
1114 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
1115 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
1116 | xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent); | |
1117 | xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent); | |
1118 | struct xfs_perag *pag; | |
1119 | struct xfs_btree_cur *cur, *tcur; | |
1120 | struct xfs_inobt_rec_incore rec, trec; | |
1121 | xfs_ino_t ino; | |
1122 | int error; | |
1123 | int offset; | |
1124 | int i, j; | |
d8c47890 | 1125 | int searchdistance = 10; |
2bd0ea18 | 1126 | |
56b2de80 DC |
1127 | pag = xfs_perag_get(mp, agno); |
1128 | ||
a2ceac1f DC |
1129 | ASSERT(pag->pagi_init); |
1130 | ASSERT(pag->pagi_inodeok); | |
1131 | ASSERT(pag->pagi_freecount > 0); | |
1132 | ||
56b2de80 | 1133 | restart_pagno: |
70eb7337 | 1134 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
2bd0ea18 NS |
1135 | /* |
1136 | * If pagino is 0 (this is the root inode allocation) use newino. | |
1137 | * This must work because we've just allocated some. | |
1138 | */ | |
1139 | if (!pagino) | |
6e3140c7 | 1140 | pagino = be32_to_cpu(agi->agi_newino); |
2bd0ea18 | 1141 | |
56b2de80 DC |
1142 | error = xfs_check_agi_freecount(cur, agi); |
1143 | if (error) | |
1144 | goto error0; | |
2bd0ea18 | 1145 | |
2bd0ea18 | 1146 | /* |
56b2de80 | 1147 | * If in the same AG as the parent, try to get near the parent. |
2bd0ea18 NS |
1148 | */ |
1149 | if (pagno == agno) { | |
56b2de80 DC |
1150 | int doneleft; /* done, to the left */ |
1151 | int doneright; /* done, to the right */ | |
56b2de80 DC |
1152 | |
1153 | error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i); | |
1154 | if (error) | |
2bd0ea18 | 1155 | goto error0; |
19ebedcf | 1156 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
56b2de80 DC |
1157 | |
1158 | error = xfs_inobt_get_rec(cur, &rec, &j); | |
1159 | if (error) | |
1160 | goto error0; | |
19ebedcf | 1161 | XFS_WANT_CORRUPTED_GOTO(mp, j == 1, error0); |
56b2de80 DC |
1162 | |
1163 | if (rec.ir_freecount > 0) { | |
2bd0ea18 NS |
1164 | /* |
1165 | * Found a free inode in the same chunk | |
56b2de80 | 1166 | * as the parent, done. |
2bd0ea18 | 1167 | */ |
56b2de80 | 1168 | goto alloc_inode; |
2bd0ea18 | 1169 | } |
56b2de80 DC |
1170 | |
1171 | ||
1172 | /* | |
1173 | * In the same AG as parent, but parent's chunk is full. | |
1174 | */ | |
1175 | ||
1176 | /* duplicate the cursor, search left & right simultaneously */ | |
1177 | error = xfs_btree_dup_cursor(cur, &tcur); | |
1178 | if (error) | |
1179 | goto error0; | |
1180 | ||
2bd0ea18 | 1181 | /* |
56b2de80 | 1182 | * Skip to last blocks looked up if same parent inode. |
2bd0ea18 | 1183 | */ |
56b2de80 DC |
1184 | if (pagino != NULLAGINO && |
1185 | pag->pagl_pagino == pagino && | |
1186 | pag->pagl_leftrec != NULLAGINO && | |
1187 | pag->pagl_rightrec != NULLAGINO) { | |
1188 | error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec, | |
3439d03a | 1189 | &trec, &doneleft); |
56b2de80 DC |
1190 | if (error) |
1191 | goto error1; | |
2bd0ea18 | 1192 | |
56b2de80 | 1193 | error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec, |
3439d03a | 1194 | &rec, &doneright); |
2bd0ea18 | 1195 | if (error) |
2bd0ea18 | 1196 | goto error1; |
56b2de80 DC |
1197 | } else { |
1198 | /* search left with tcur, back up 1 record */ | |
1199 | error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1); | |
1200 | if (error) | |
2bd0ea18 | 1201 | goto error1; |
2bd0ea18 | 1202 | |
56b2de80 DC |
1203 | /* search right with cur, go forward 1 record. */ |
1204 | error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0); | |
1205 | if (error) | |
1206 | goto error1; | |
1207 | } | |
1208 | ||
1209 | /* | |
1210 | * Loop until we find an inode chunk with a free inode. | |
1211 | */ | |
d8c47890 | 1212 | while (--searchdistance > 0 && (!doneleft || !doneright)) { |
56b2de80 DC |
1213 | int useleft; /* using left inode chunk this time */ |
1214 | ||
56b2de80 DC |
1215 | /* figure out the closer block if both are valid. */ |
1216 | if (!doneleft && !doneright) { | |
1217 | useleft = pagino - | |
1218 | (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) < | |
1219 | rec.ir_startino - pagino; | |
1220 | } else { | |
1221 | useleft = !doneleft; | |
1222 | } | |
1223 | ||
1224 | /* free inodes to the left? */ | |
1225 | if (useleft && trec.ir_freecount) { | |
56b2de80 DC |
1226 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
1227 | cur = tcur; | |
1228 | ||
1229 | pag->pagl_leftrec = trec.ir_startino; | |
1230 | pag->pagl_rightrec = rec.ir_startino; | |
1231 | pag->pagl_pagino = pagino; | |
ef54efb5 | 1232 | rec = trec; |
56b2de80 DC |
1233 | goto alloc_inode; |
1234 | } | |
1235 | ||
1236 | /* free inodes to the right? */ | |
1237 | if (!useleft && rec.ir_freecount) { | |
1238 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
1239 | ||
1240 | pag->pagl_leftrec = trec.ir_startino; | |
1241 | pag->pagl_rightrec = rec.ir_startino; | |
1242 | pag->pagl_pagino = pagino; | |
1243 | goto alloc_inode; | |
1244 | } | |
1245 | ||
1246 | /* get next record to check */ | |
1247 | if (useleft) { | |
1248 | error = xfs_ialloc_next_rec(tcur, &trec, | |
1249 | &doneleft, 1); | |
1250 | } else { | |
1251 | error = xfs_ialloc_next_rec(cur, &rec, | |
1252 | &doneright, 0); | |
1253 | } | |
1254 | if (error) | |
1255 | goto error1; | |
2bd0ea18 | 1256 | } |
56b2de80 | 1257 | |
d8c47890 CM |
1258 | if (searchdistance <= 0) { |
1259 | /* | |
1260 | * Not in range - save last search | |
1261 | * location and allocate a new inode | |
1262 | */ | |
1263 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
1264 | pag->pagl_leftrec = trec.ir_startino; | |
1265 | pag->pagl_rightrec = rec.ir_startino; | |
1266 | pag->pagl_pagino = pagino; | |
1267 | ||
1268 | } else { | |
1269 | /* | |
1270 | * We've reached the end of the btree. because | |
1271 | * we are only searching a small chunk of the | |
1272 | * btree each search, there is obviously free | |
1273 | * inodes closer to the parent inode than we | |
1274 | * are now. restart the search again. | |
1275 | */ | |
1276 | pag->pagl_pagino = NULLAGINO; | |
1277 | pag->pagl_leftrec = NULLAGINO; | |
1278 | pag->pagl_rightrec = NULLAGINO; | |
1279 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
1280 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
1281 | goto restart_pagno; | |
1282 | } | |
2bd0ea18 | 1283 | } |
56b2de80 | 1284 | |
2bd0ea18 | 1285 | /* |
56b2de80 | 1286 | * In a different AG from the parent. |
2bd0ea18 NS |
1287 | * See if the most recently allocated block has any free. |
1288 | */ | |
a2ceac1f | 1289 | if (agi->agi_newino != cpu_to_be32(NULLAGINO)) { |
56b2de80 DC |
1290 | error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino), |
1291 | XFS_LOOKUP_EQ, &i); | |
1292 | if (error) | |
2bd0ea18 | 1293 | goto error0; |
56b2de80 DC |
1294 | |
1295 | if (i == 1) { | |
1296 | error = xfs_inobt_get_rec(cur, &rec, &j); | |
2bd0ea18 NS |
1297 | if (error) |
1298 | goto error0; | |
56b2de80 DC |
1299 | |
1300 | if (j == 1 && rec.ir_freecount > 0) { | |
1301 | /* | |
1302 | * The last chunk allocated in the group | |
1303 | * still has a free inode. | |
1304 | */ | |
1305 | goto alloc_inode; | |
2bd0ea18 NS |
1306 | } |
1307 | } | |
1308 | } | |
56b2de80 DC |
1309 | |
1310 | /* | |
1311 | * None left in the last group, search the whole AG | |
1312 | */ | |
1313 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); | |
1314 | if (error) | |
1315 | goto error0; | |
19ebedcf | 1316 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
56b2de80 DC |
1317 | |
1318 | for (;;) { | |
1319 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
1320 | if (error) | |
1321 | goto error0; | |
19ebedcf | 1322 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
56b2de80 DC |
1323 | if (rec.ir_freecount > 0) |
1324 | break; | |
1325 | error = xfs_btree_increment(cur, 0, &i); | |
1326 | if (error) | |
1327 | goto error0; | |
19ebedcf | 1328 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
56b2de80 DC |
1329 | } |
1330 | ||
1331 | alloc_inode: | |
01792d3b | 1332 | offset = xfs_inobt_first_free_inode(&rec); |
2bd0ea18 NS |
1333 | ASSERT(offset >= 0); |
1334 | ASSERT(offset < XFS_INODES_PER_CHUNK); | |
1335 | ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) % | |
1336 | XFS_INODES_PER_CHUNK) == 0); | |
1337 | ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset); | |
56b2de80 | 1338 | rec.ir_free &= ~XFS_INOBT_MASK(offset); |
2bd0ea18 | 1339 | rec.ir_freecount--; |
56b2de80 DC |
1340 | error = xfs_inobt_update(cur, &rec); |
1341 | if (error) | |
2bd0ea18 | 1342 | goto error0; |
5e656dbb | 1343 | be32_add_cpu(&agi->agi_freecount, -1); |
2bd0ea18 | 1344 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); |
56b2de80 DC |
1345 | pag->pagi_freecount--; |
1346 | ||
1347 | error = xfs_check_agi_freecount(cur, agi); | |
1348 | if (error) | |
1349 | goto error0; | |
2bd0ea18 | 1350 | |
2bd0ea18 NS |
1351 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
1352 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1); | |
56b2de80 | 1353 | xfs_perag_put(pag); |
2bd0ea18 NS |
1354 | *inop = ino; |
1355 | return 0; | |
1356 | error1: | |
1357 | xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR); | |
1358 | error0: | |
1359 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
56b2de80 | 1360 | xfs_perag_put(pag); |
2bd0ea18 NS |
1361 | return error; |
1362 | } | |
1363 | ||
ff105f75 DC |
1364 | /* |
1365 | * Use the free inode btree to allocate an inode based on distance from the | |
1366 | * parent. Note that the provided cursor may be deleted and replaced. | |
1367 | */ | |
1368 | STATIC int | |
1369 | xfs_dialloc_ag_finobt_near( | |
1370 | xfs_agino_t pagino, | |
1371 | struct xfs_btree_cur **ocur, | |
1372 | struct xfs_inobt_rec_incore *rec) | |
1373 | { | |
1374 | struct xfs_btree_cur *lcur = *ocur; /* left search cursor */ | |
1375 | struct xfs_btree_cur *rcur; /* right search cursor */ | |
1376 | struct xfs_inobt_rec_incore rrec; | |
1377 | int error; | |
1378 | int i, j; | |
1379 | ||
1380 | error = xfs_inobt_lookup(lcur, pagino, XFS_LOOKUP_LE, &i); | |
1381 | if (error) | |
1382 | return error; | |
1383 | ||
1384 | if (i == 1) { | |
1385 | error = xfs_inobt_get_rec(lcur, rec, &i); | |
1386 | if (error) | |
1387 | return error; | |
19ebedcf | 1388 | XFS_WANT_CORRUPTED_RETURN(lcur->bc_mp, i == 1); |
ff105f75 DC |
1389 | |
1390 | /* | |
1391 | * See if we've landed in the parent inode record. The finobt | |
1392 | * only tracks chunks with at least one free inode, so record | |
1393 | * existence is enough. | |
1394 | */ | |
1395 | if (pagino >= rec->ir_startino && | |
1396 | pagino < (rec->ir_startino + XFS_INODES_PER_CHUNK)) | |
1397 | return 0; | |
1398 | } | |
1399 | ||
1400 | error = xfs_btree_dup_cursor(lcur, &rcur); | |
1401 | if (error) | |
1402 | return error; | |
1403 | ||
1404 | error = xfs_inobt_lookup(rcur, pagino, XFS_LOOKUP_GE, &j); | |
1405 | if (error) | |
1406 | goto error_rcur; | |
1407 | if (j == 1) { | |
1408 | error = xfs_inobt_get_rec(rcur, &rrec, &j); | |
1409 | if (error) | |
1410 | goto error_rcur; | |
19ebedcf | 1411 | XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, j == 1, error_rcur); |
ff105f75 DC |
1412 | } |
1413 | ||
19ebedcf | 1414 | XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, i == 1 || j == 1, error_rcur); |
ff105f75 DC |
1415 | if (i == 1 && j == 1) { |
1416 | /* | |
1417 | * Both the left and right records are valid. Choose the closer | |
1418 | * inode chunk to the target. | |
1419 | */ | |
1420 | if ((pagino - rec->ir_startino + XFS_INODES_PER_CHUNK - 1) > | |
1421 | (rrec.ir_startino - pagino)) { | |
1422 | *rec = rrec; | |
1423 | xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR); | |
1424 | *ocur = rcur; | |
1425 | } else { | |
1426 | xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR); | |
1427 | } | |
1428 | } else if (j == 1) { | |
1429 | /* only the right record is valid */ | |
1430 | *rec = rrec; | |
1431 | xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR); | |
1432 | *ocur = rcur; | |
1433 | } else if (i == 1) { | |
1434 | /* only the left record is valid */ | |
1435 | xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR); | |
1436 | } | |
1437 | ||
1438 | return 0; | |
1439 | ||
1440 | error_rcur: | |
1441 | xfs_btree_del_cursor(rcur, XFS_BTREE_ERROR); | |
1442 | return error; | |
1443 | } | |
1444 | ||
1445 | /* | |
1446 | * Use the free inode btree to find a free inode based on a newino hint. If | |
1447 | * the hint is NULL, find the first free inode in the AG. | |
1448 | */ | |
1449 | STATIC int | |
1450 | xfs_dialloc_ag_finobt_newino( | |
1451 | struct xfs_agi *agi, | |
1452 | struct xfs_btree_cur *cur, | |
1453 | struct xfs_inobt_rec_incore *rec) | |
1454 | { | |
1455 | int error; | |
1456 | int i; | |
1457 | ||
1458 | if (agi->agi_newino != cpu_to_be32(NULLAGINO)) { | |
1459 | error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino), | |
1460 | XFS_LOOKUP_EQ, &i); | |
1461 | if (error) | |
1462 | return error; | |
1463 | if (i == 1) { | |
1464 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1465 | if (error) | |
1466 | return error; | |
19ebedcf | 1467 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
ff105f75 DC |
1468 | return 0; |
1469 | } | |
1470 | } | |
1471 | ||
1472 | /* | |
1473 | * Find the first inode available in the AG. | |
1474 | */ | |
1475 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); | |
1476 | if (error) | |
1477 | return error; | |
19ebedcf | 1478 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
ff105f75 DC |
1479 | |
1480 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1481 | if (error) | |
1482 | return error; | |
19ebedcf | 1483 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
ff105f75 DC |
1484 | |
1485 | return 0; | |
1486 | } | |
1487 | ||
1488 | /* | |
1489 | * Update the inobt based on a modification made to the finobt. Also ensure that | |
1490 | * the records from both trees are equivalent post-modification. | |
1491 | */ | |
1492 | STATIC int | |
1493 | xfs_dialloc_ag_update_inobt( | |
1494 | struct xfs_btree_cur *cur, /* inobt cursor */ | |
1495 | struct xfs_inobt_rec_incore *frec, /* finobt record */ | |
1496 | int offset) /* inode offset */ | |
1497 | { | |
1498 | struct xfs_inobt_rec_incore rec; | |
1499 | int error; | |
1500 | int i; | |
1501 | ||
1502 | error = xfs_inobt_lookup(cur, frec->ir_startino, XFS_LOOKUP_EQ, &i); | |
1503 | if (error) | |
1504 | return error; | |
19ebedcf | 1505 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
ff105f75 DC |
1506 | |
1507 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
1508 | if (error) | |
1509 | return error; | |
19ebedcf | 1510 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
ff105f75 DC |
1511 | ASSERT((XFS_AGINO_TO_OFFSET(cur->bc_mp, rec.ir_startino) % |
1512 | XFS_INODES_PER_CHUNK) == 0); | |
1513 | ||
1514 | rec.ir_free &= ~XFS_INOBT_MASK(offset); | |
1515 | rec.ir_freecount--; | |
1516 | ||
19ebedcf | 1517 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, (rec.ir_free == frec->ir_free) && |
ff105f75 DC |
1518 | (rec.ir_freecount == frec->ir_freecount)); |
1519 | ||
5a35bf2c | 1520 | return xfs_inobt_update(cur, &rec); |
ff105f75 DC |
1521 | } |
1522 | ||
1523 | /* | |
1524 | * Allocate an inode using the free inode btree, if available. Otherwise, fall | |
1525 | * back to the inobt search algorithm. | |
1526 | * | |
1527 | * The caller selected an AG for us, and made sure that free inodes are | |
1528 | * available. | |
1529 | */ | |
88fc7306 BF |
1530 | STATIC int |
1531 | xfs_dialloc_ag( | |
1532 | struct xfs_trans *tp, | |
1533 | struct xfs_buf *agbp, | |
1534 | xfs_ino_t parent, | |
1535 | xfs_ino_t *inop) | |
1536 | { | |
1537 | struct xfs_mount *mp = tp->t_mountp; | |
1538 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
1539 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
1540 | xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent); | |
1541 | xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent); | |
1542 | struct xfs_perag *pag; | |
ff105f75 DC |
1543 | struct xfs_btree_cur *cur; /* finobt cursor */ |
1544 | struct xfs_btree_cur *icur; /* inobt cursor */ | |
88fc7306 | 1545 | struct xfs_inobt_rec_incore rec; |
88fc7306 BF |
1546 | xfs_ino_t ino; |
1547 | int error; | |
1548 | int offset; | |
ff105f75 | 1549 | int i; |
88fc7306 BF |
1550 | |
1551 | if (!xfs_sb_version_hasfinobt(&mp->m_sb)) | |
ff105f75 | 1552 | return xfs_dialloc_ag_inobt(tp, agbp, parent, inop); |
88fc7306 BF |
1553 | |
1554 | pag = xfs_perag_get(mp, agno); | |
1555 | ||
1556 | /* | |
1557 | * If pagino is 0 (this is the root inode allocation) use newino. | |
1558 | * This must work because we've just allocated some. | |
1559 | */ | |
1560 | if (!pagino) | |
1561 | pagino = be32_to_cpu(agi->agi_newino); | |
1562 | ||
1563 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO); | |
1564 | ||
1565 | error = xfs_check_agi_freecount(cur, agi); | |
1566 | if (error) | |
1567 | goto error_cur; | |
1568 | ||
ff105f75 DC |
1569 | /* |
1570 | * The search algorithm depends on whether we're in the same AG as the | |
1571 | * parent. If so, find the closest available inode to the parent. If | |
1572 | * not, consider the agi hint or find the first free inode in the AG. | |
1573 | */ | |
1574 | if (agno == pagno) | |
1575 | error = xfs_dialloc_ag_finobt_near(pagino, &cur, &rec); | |
1576 | else | |
1577 | error = xfs_dialloc_ag_finobt_newino(agi, cur, &rec); | |
1578 | if (error) | |
1579 | goto error_cur; | |
88fc7306 | 1580 | |
01792d3b | 1581 | offset = xfs_inobt_first_free_inode(&rec); |
88fc7306 BF |
1582 | ASSERT(offset >= 0); |
1583 | ASSERT(offset < XFS_INODES_PER_CHUNK); | |
1584 | ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) % | |
1585 | XFS_INODES_PER_CHUNK) == 0); | |
1586 | ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset); | |
1587 | ||
1588 | /* | |
1589 | * Modify or remove the finobt record. | |
1590 | */ | |
1591 | rec.ir_free &= ~XFS_INOBT_MASK(offset); | |
1592 | rec.ir_freecount--; | |
ff105f75 | 1593 | if (rec.ir_freecount) |
88fc7306 BF |
1594 | error = xfs_inobt_update(cur, &rec); |
1595 | else | |
1596 | error = xfs_btree_delete(cur, &i); | |
1597 | if (error) | |
1598 | goto error_cur; | |
1599 | ||
1600 | /* | |
ff105f75 DC |
1601 | * The finobt has now been updated appropriately. We haven't updated the |
1602 | * agi and superblock yet, so we can create an inobt cursor and validate | |
1603 | * the original freecount. If all is well, make the equivalent update to | |
1604 | * the inobt using the finobt record and offset information. | |
88fc7306 | 1605 | */ |
ff105f75 | 1606 | icur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
88fc7306 | 1607 | |
ff105f75 | 1608 | error = xfs_check_agi_freecount(icur, agi); |
88fc7306 | 1609 | if (error) |
ff105f75 | 1610 | goto error_icur; |
88fc7306 | 1611 | |
ff105f75 | 1612 | error = xfs_dialloc_ag_update_inobt(icur, &rec, offset); |
88fc7306 | 1613 | if (error) |
ff105f75 | 1614 | goto error_icur; |
88fc7306 BF |
1615 | |
1616 | /* | |
ff105f75 DC |
1617 | * Both trees have now been updated. We must update the perag and |
1618 | * superblock before we can check the freecount for each btree. | |
88fc7306 BF |
1619 | */ |
1620 | be32_add_cpu(&agi->agi_freecount, -1); | |
1621 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); | |
1622 | pag->pagi_freecount--; | |
1623 | ||
1624 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1); | |
1625 | ||
ff105f75 | 1626 | error = xfs_check_agi_freecount(icur, agi); |
88fc7306 | 1627 | if (error) |
ff105f75 | 1628 | goto error_icur; |
88fc7306 BF |
1629 | error = xfs_check_agi_freecount(cur, agi); |
1630 | if (error) | |
ff105f75 | 1631 | goto error_icur; |
88fc7306 | 1632 | |
ff105f75 | 1633 | xfs_btree_del_cursor(icur, XFS_BTREE_NOERROR); |
88fc7306 BF |
1634 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
1635 | xfs_perag_put(pag); | |
1636 | *inop = ino; | |
1637 | return 0; | |
1638 | ||
ff105f75 DC |
1639 | error_icur: |
1640 | xfs_btree_del_cursor(icur, XFS_BTREE_ERROR); | |
88fc7306 BF |
1641 | error_cur: |
1642 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
1643 | xfs_perag_put(pag); | |
1644 | return error; | |
1645 | } | |
1646 | ||
a2ceac1f DC |
1647 | /* |
1648 | * Allocate an inode on disk. | |
1649 | * | |
1650 | * Mode is used to tell whether the new inode will need space, and whether it | |
1651 | * is a directory. | |
1652 | * | |
1653 | * This function is designed to be called twice if it has to do an allocation | |
1654 | * to make more free inodes. On the first call, *IO_agbp should be set to NULL. | |
1655 | * If an inode is available without having to performn an allocation, an inode | |
1656 | * number is returned. In this case, *IO_agbp is set to NULL. If an allocation | |
1657 | * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp. | |
1658 | * The caller should then commit the current transaction, allocate a | |
1659 | * new transaction, and call xfs_dialloc() again, passing in the previous value | |
1660 | * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI | |
1661 | * buffer is locked across the two calls, the second call is guaranteed to have | |
1662 | * a free inode available. | |
1663 | * | |
1664 | * Once we successfully pick an inode its number is returned and the on-disk | |
1665 | * data structures are updated. The inode itself is not read in, since doing so | |
1666 | * would break ordering constraints with xfs_reclaim. | |
1667 | */ | |
1668 | int | |
1669 | xfs_dialloc( | |
1670 | struct xfs_trans *tp, | |
1671 | xfs_ino_t parent, | |
1672 | umode_t mode, | |
a2ceac1f DC |
1673 | struct xfs_buf **IO_agbp, |
1674 | xfs_ino_t *inop) | |
1675 | { | |
1676 | struct xfs_mount *mp = tp->t_mountp; | |
1677 | struct xfs_buf *agbp; | |
1678 | xfs_agnumber_t agno; | |
1679 | int error; | |
1680 | int ialloced; | |
1681 | int noroom = 0; | |
1682 | xfs_agnumber_t start_agno; | |
1683 | struct xfs_perag *pag; | |
c1a29e92 | 1684 | int okalloc = 1; |
a2ceac1f DC |
1685 | |
1686 | if (*IO_agbp) { | |
1687 | /* | |
1688 | * If the caller passes in a pointer to the AGI buffer, | |
1689 | * continue where we left off before. In this case, we | |
1690 | * know that the allocation group has free inodes. | |
1691 | */ | |
1692 | agbp = *IO_agbp; | |
1693 | goto out_alloc; | |
1694 | } | |
1695 | ||
1696 | /* | |
1697 | * We do not have an agbp, so select an initial allocation | |
1698 | * group for inode allocation. | |
1699 | */ | |
c1a29e92 | 1700 | start_agno = xfs_ialloc_ag_select(tp, parent, mode); |
a2ceac1f DC |
1701 | if (start_agno == NULLAGNUMBER) { |
1702 | *inop = NULLFSINO; | |
1703 | return 0; | |
1704 | } | |
1705 | ||
1706 | /* | |
1707 | * If we have already hit the ceiling of inode blocks then clear | |
1708 | * okalloc so we scan all available agi structures for a free | |
1709 | * inode. | |
cbf3beaa GW |
1710 | * |
1711 | * Read rough value of mp->m_icount by percpu_counter_read_positive, | |
1712 | * which will sacrifice the preciseness but improve the performance. | |
a2ceac1f DC |
1713 | */ |
1714 | if (mp->m_maxicount && | |
cbf3beaa GW |
1715 | percpu_counter_read_positive(&mp->m_icount) + mp->m_ialloc_inos |
1716 | > mp->m_maxicount) { | |
a2ceac1f DC |
1717 | noroom = 1; |
1718 | okalloc = 0; | |
1719 | } | |
1720 | ||
1721 | /* | |
1722 | * Loop until we find an allocation group that either has free inodes | |
1723 | * or in which we can allocate some inodes. Iterate through the | |
1724 | * allocation groups upward, wrapping at the end. | |
1725 | */ | |
1726 | agno = start_agno; | |
1727 | for (;;) { | |
1728 | pag = xfs_perag_get(mp, agno); | |
1729 | if (!pag->pagi_inodeok) { | |
1730 | xfs_ialloc_next_ag(mp); | |
1731 | goto nextag; | |
1732 | } | |
1733 | ||
1734 | if (!pag->pagi_init) { | |
1735 | error = xfs_ialloc_pagi_init(mp, tp, agno); | |
1736 | if (error) | |
1737 | goto out_error; | |
1738 | } | |
1739 | ||
1740 | /* | |
1741 | * Do a first racy fast path check if this AG is usable. | |
1742 | */ | |
1743 | if (!pag->pagi_freecount && !okalloc) | |
1744 | goto nextag; | |
1745 | ||
1746 | /* | |
1747 | * Then read in the AGI buffer and recheck with the AGI buffer | |
1748 | * lock held. | |
1749 | */ | |
1750 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); | |
1751 | if (error) | |
1752 | goto out_error; | |
1753 | ||
1754 | if (pag->pagi_freecount) { | |
1755 | xfs_perag_put(pag); | |
1756 | goto out_alloc; | |
1757 | } | |
1758 | ||
1759 | if (!okalloc) | |
1760 | goto nextag_relse_buffer; | |
1761 | ||
1762 | ||
1763 | error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced); | |
1764 | if (error) { | |
1765 | xfs_trans_brelse(tp, agbp); | |
1766 | ||
12b53197 | 1767 | if (error != -ENOSPC) |
a2ceac1f DC |
1768 | goto out_error; |
1769 | ||
1770 | xfs_perag_put(pag); | |
1771 | *inop = NULLFSINO; | |
1772 | return 0; | |
1773 | } | |
1774 | ||
1775 | if (ialloced) { | |
1776 | /* | |
1777 | * We successfully allocated some inodes, return | |
1778 | * the current context to the caller so that it | |
1779 | * can commit the current transaction and call | |
1780 | * us again where we left off. | |
1781 | */ | |
1782 | ASSERT(pag->pagi_freecount > 0); | |
1783 | xfs_perag_put(pag); | |
1784 | ||
1785 | *IO_agbp = agbp; | |
1786 | *inop = NULLFSINO; | |
1787 | return 0; | |
1788 | } | |
1789 | ||
1790 | nextag_relse_buffer: | |
1791 | xfs_trans_brelse(tp, agbp); | |
1792 | nextag: | |
1793 | xfs_perag_put(pag); | |
1794 | if (++agno == mp->m_sb.sb_agcount) | |
1795 | agno = 0; | |
1796 | if (agno == start_agno) { | |
1797 | *inop = NULLFSINO; | |
12b53197 | 1798 | return noroom ? -ENOSPC : 0; |
a2ceac1f DC |
1799 | } |
1800 | } | |
1801 | ||
1802 | out_alloc: | |
1803 | *IO_agbp = NULL; | |
1804 | return xfs_dialloc_ag(tp, agbp, parent, inop); | |
1805 | out_error: | |
1806 | xfs_perag_put(pag); | |
1e68581b | 1807 | return error; |
a2ceac1f DC |
1808 | } |
1809 | ||
7338c4b8 BF |
1810 | /* |
1811 | * Free the blocks of an inode chunk. We must consider that the inode chunk | |
1812 | * might be sparse and only free the regions that are allocated as part of the | |
1813 | * chunk. | |
1814 | */ | |
1815 | STATIC void | |
1816 | xfs_difree_inode_chunk( | |
1817 | struct xfs_mount *mp, | |
1818 | xfs_agnumber_t agno, | |
1819 | struct xfs_inobt_rec_incore *rec, | |
f33cea1a | 1820 | struct xfs_defer_ops *dfops) |
7338c4b8 BF |
1821 | { |
1822 | xfs_agblock_t sagbno = XFS_AGINO_TO_AGBNO(mp, rec->ir_startino); | |
1823 | int startidx, endidx; | |
1824 | int nextbit; | |
1825 | xfs_agblock_t agbno; | |
1826 | int contigblk; | |
85aec44f | 1827 | struct xfs_owner_info oinfo; |
7338c4b8 | 1828 | DECLARE_BITMAP(holemask, XFS_INOBT_HOLEMASK_BITS); |
85aec44f | 1829 | xfs_rmap_ag_owner(&oinfo, XFS_RMAP_OWN_INODES); |
7338c4b8 BF |
1830 | |
1831 | if (!xfs_inobt_issparse(rec->ir_holemask)) { | |
1832 | /* not sparse, calculate extent info directly */ | |
f33cea1a | 1833 | xfs_bmap_add_free(mp, dfops, XFS_AGB_TO_FSB(mp, agno, sagbno), |
85aec44f | 1834 | mp->m_ialloc_blks, &oinfo); |
7338c4b8 BF |
1835 | return; |
1836 | } | |
1837 | ||
1838 | /* holemask is only 16-bits (fits in an unsigned long) */ | |
1839 | ASSERT(sizeof(rec->ir_holemask) <= sizeof(holemask[0])); | |
1840 | holemask[0] = rec->ir_holemask; | |
1841 | ||
1842 | /* | |
1843 | * Find contiguous ranges of zeroes (i.e., allocated regions) in the | |
1844 | * holemask and convert the start/end index of each range to an extent. | |
1845 | * We start with the start and end index both pointing at the first 0 in | |
1846 | * the mask. | |
1847 | */ | |
1848 | startidx = endidx = find_first_zero_bit(holemask, | |
1849 | XFS_INOBT_HOLEMASK_BITS); | |
1850 | nextbit = startidx + 1; | |
1851 | while (startidx < XFS_INOBT_HOLEMASK_BITS) { | |
1852 | nextbit = find_next_zero_bit(holemask, XFS_INOBT_HOLEMASK_BITS, | |
1853 | nextbit); | |
1854 | /* | |
1855 | * If the next zero bit is contiguous, update the end index of | |
1856 | * the current range and continue. | |
1857 | */ | |
1858 | if (nextbit != XFS_INOBT_HOLEMASK_BITS && | |
1859 | nextbit == endidx + 1) { | |
1860 | endidx = nextbit; | |
1861 | goto next; | |
1862 | } | |
1863 | ||
1864 | /* | |
1865 | * nextbit is not contiguous with the current end index. Convert | |
1866 | * the current start/end to an extent and add it to the free | |
1867 | * list. | |
1868 | */ | |
1869 | agbno = sagbno + (startidx * XFS_INODES_PER_HOLEMASK_BIT) / | |
1870 | mp->m_sb.sb_inopblock; | |
1871 | contigblk = ((endidx - startidx + 1) * | |
1872 | XFS_INODES_PER_HOLEMASK_BIT) / | |
1873 | mp->m_sb.sb_inopblock; | |
1874 | ||
1875 | ASSERT(agbno % mp->m_sb.sb_spino_align == 0); | |
1876 | ASSERT(contigblk % mp->m_sb.sb_spino_align == 0); | |
f33cea1a | 1877 | xfs_bmap_add_free(mp, dfops, XFS_AGB_TO_FSB(mp, agno, agbno), |
85aec44f | 1878 | contigblk, &oinfo); |
7338c4b8 BF |
1879 | |
1880 | /* reset range to current bit and carry on... */ | |
1881 | startidx = endidx = nextbit; | |
1882 | ||
1883 | next: | |
1884 | nextbit++; | |
1885 | } | |
1886 | } | |
1887 | ||
eb9a297a BF |
1888 | STATIC int |
1889 | xfs_difree_inobt( | |
1890 | struct xfs_mount *mp, | |
1891 | struct xfs_trans *tp, | |
1892 | struct xfs_buf *agbp, | |
1893 | xfs_agino_t agino, | |
f33cea1a | 1894 | struct xfs_defer_ops *dfops, |
5a3b2e0a | 1895 | struct xfs_icluster *xic, |
eb9a297a | 1896 | struct xfs_inobt_rec_incore *orec) |
3439d03a | 1897 | { |
eb9a297a BF |
1898 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); |
1899 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
1900 | struct xfs_perag *pag; | |
1901 | struct xfs_btree_cur *cur; | |
1902 | struct xfs_inobt_rec_incore rec; | |
1903 | int ilen; | |
1904 | int error; | |
1905 | int i; | |
1906 | int off; | |
3439d03a | 1907 | |
3439d03a | 1908 | ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC)); |
eb9a297a BF |
1909 | ASSERT(XFS_AGINO_TO_AGBNO(mp, agino) < be32_to_cpu(agi->agi_length)); |
1910 | ||
3439d03a DC |
1911 | /* |
1912 | * Initialize the cursor. | |
1913 | */ | |
70eb7337 | 1914 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
3439d03a DC |
1915 | |
1916 | error = xfs_check_agi_freecount(cur, agi); | |
1917 | if (error) | |
1918 | goto error0; | |
1919 | ||
1920 | /* | |
1921 | * Look for the entry describing this inode. | |
1922 | */ | |
1923 | if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) { | |
1924 | xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.", | |
1925 | __func__, error); | |
1926 | goto error0; | |
1927 | } | |
19ebedcf | 1928 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
3439d03a DC |
1929 | error = xfs_inobt_get_rec(cur, &rec, &i); |
1930 | if (error) { | |
1931 | xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.", | |
1932 | __func__, error); | |
1933 | goto error0; | |
1934 | } | |
19ebedcf | 1935 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
3439d03a DC |
1936 | /* |
1937 | * Get the offset in the inode chunk. | |
1938 | */ | |
1939 | off = agino - rec.ir_startino; | |
1940 | ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK); | |
1941 | ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off))); | |
1942 | /* | |
1943 | * Mark the inode free & increment the count. | |
1944 | */ | |
1945 | rec.ir_free |= XFS_INOBT_MASK(off); | |
1946 | rec.ir_freecount++; | |
1947 | ||
1948 | /* | |
f6580bcf BF |
1949 | * When an inode chunk is free, it becomes eligible for removal. Don't |
1950 | * remove the chunk if the block size is large enough for multiple inode | |
1951 | * chunks (that might not be free). | |
3439d03a DC |
1952 | */ |
1953 | if (!(mp->m_flags & XFS_MOUNT_IKEEP) && | |
f6580bcf BF |
1954 | rec.ir_free == XFS_INOBT_ALL_FREE && |
1955 | mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK) { | |
180d3cd8 | 1956 | xic->deleted = true; |
5a3b2e0a BF |
1957 | xic->first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino); |
1958 | xic->alloc = xfs_inobt_irec_to_allocmask(&rec); | |
3439d03a DC |
1959 | |
1960 | /* | |
1961 | * Remove the inode cluster from the AGI B+Tree, adjust the | |
1962 | * AGI and Superblock inode counts, and mark the disk space | |
1963 | * to be freed when the transaction is committed. | |
1964 | */ | |
f6580bcf | 1965 | ilen = rec.ir_freecount; |
3439d03a DC |
1966 | be32_add_cpu(&agi->agi_count, -ilen); |
1967 | be32_add_cpu(&agi->agi_freecount, -(ilen - 1)); | |
1968 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT); | |
1969 | pag = xfs_perag_get(mp, agno); | |
1970 | pag->pagi_freecount -= ilen - 1; | |
1971 | xfs_perag_put(pag); | |
1972 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen); | |
1973 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1)); | |
1974 | ||
1975 | if ((error = xfs_btree_delete(cur, &i))) { | |
1976 | xfs_warn(mp, "%s: xfs_btree_delete returned error %d.", | |
1977 | __func__, error); | |
1978 | goto error0; | |
1979 | } | |
1980 | ||
f33cea1a | 1981 | xfs_difree_inode_chunk(mp, agno, &rec, dfops); |
3439d03a | 1982 | } else { |
180d3cd8 | 1983 | xic->deleted = false; |
3439d03a DC |
1984 | |
1985 | error = xfs_inobt_update(cur, &rec); | |
1986 | if (error) { | |
1987 | xfs_warn(mp, "%s: xfs_inobt_update returned error %d.", | |
1988 | __func__, error); | |
1989 | goto error0; | |
1990 | } | |
1991 | ||
f8149110 | 1992 | /* |
3439d03a DC |
1993 | * Change the inode free counts and log the ag/sb changes. |
1994 | */ | |
1995 | be32_add_cpu(&agi->agi_freecount, 1); | |
1996 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); | |
1997 | pag = xfs_perag_get(mp, agno); | |
1998 | pag->pagi_freecount++; | |
1999 | xfs_perag_put(pag); | |
2000 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1); | |
2001 | } | |
2002 | ||
2003 | error = xfs_check_agi_freecount(cur, agi); | |
2004 | if (error) | |
2005 | goto error0; | |
2006 | ||
eb9a297a | 2007 | *orec = rec; |
3439d03a DC |
2008 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
2009 | return 0; | |
2010 | ||
2011 | error0: | |
2012 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
2013 | return error; | |
2014 | } | |
2015 | ||
1bb93fd1 BF |
2016 | /* |
2017 | * Free an inode in the free inode btree. | |
2018 | */ | |
2019 | STATIC int | |
2020 | xfs_difree_finobt( | |
2021 | struct xfs_mount *mp, | |
2022 | struct xfs_trans *tp, | |
2023 | struct xfs_buf *agbp, | |
2024 | xfs_agino_t agino, | |
2025 | struct xfs_inobt_rec_incore *ibtrec) /* inobt record */ | |
2026 | { | |
2027 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
2028 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
2029 | struct xfs_btree_cur *cur; | |
2030 | struct xfs_inobt_rec_incore rec; | |
2031 | int offset = agino - ibtrec->ir_startino; | |
2032 | int error; | |
2033 | int i; | |
2034 | ||
2035 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO); | |
2036 | ||
2037 | error = xfs_inobt_lookup(cur, ibtrec->ir_startino, XFS_LOOKUP_EQ, &i); | |
2038 | if (error) | |
2039 | goto error; | |
2040 | if (i == 0) { | |
2041 | /* | |
2042 | * If the record does not exist in the finobt, we must have just | |
2043 | * freed an inode in a previously fully allocated chunk. If not, | |
2044 | * something is out of sync. | |
2045 | */ | |
19ebedcf | 2046 | XFS_WANT_CORRUPTED_GOTO(mp, ibtrec->ir_freecount == 1, error); |
1bb93fd1 | 2047 | |
11640e30 BF |
2048 | error = xfs_inobt_insert_rec(cur, ibtrec->ir_holemask, |
2049 | ibtrec->ir_count, | |
2050 | ibtrec->ir_freecount, | |
1bb93fd1 BF |
2051 | ibtrec->ir_free, &i); |
2052 | if (error) | |
2053 | goto error; | |
2054 | ASSERT(i == 1); | |
2055 | ||
2056 | goto out; | |
2057 | } | |
2058 | ||
2059 | /* | |
ff105f75 DC |
2060 | * Read and update the existing record. We could just copy the ibtrec |
2061 | * across here, but that would defeat the purpose of having redundant | |
2062 | * metadata. By making the modifications independently, we can catch | |
2063 | * corruptions that we wouldn't see if we just copied from one record | |
2064 | * to another. | |
1bb93fd1 BF |
2065 | */ |
2066 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
2067 | if (error) | |
2068 | goto error; | |
19ebedcf | 2069 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error); |
1bb93fd1 BF |
2070 | |
2071 | rec.ir_free |= XFS_INOBT_MASK(offset); | |
2072 | rec.ir_freecount++; | |
2073 | ||
19ebedcf | 2074 | XFS_WANT_CORRUPTED_GOTO(mp, (rec.ir_free == ibtrec->ir_free) && |
1bb93fd1 BF |
2075 | (rec.ir_freecount == ibtrec->ir_freecount), |
2076 | error); | |
2077 | ||
2078 | /* | |
2079 | * The content of inobt records should always match between the inobt | |
2080 | * and finobt. The lifecycle of records in the finobt is different from | |
2081 | * the inobt in that the finobt only tracks records with at least one | |
ff105f75 DC |
2082 | * free inode. Hence, if all of the inodes are free and we aren't |
2083 | * keeping inode chunks permanently on disk, remove the record. | |
2084 | * Otherwise, update the record with the new information. | |
f6580bcf BF |
2085 | * |
2086 | * Note that we currently can't free chunks when the block size is large | |
2087 | * enough for multiple chunks. Leave the finobt record to remain in sync | |
2088 | * with the inobt. | |
1bb93fd1 | 2089 | */ |
f6580bcf BF |
2090 | if (rec.ir_free == XFS_INOBT_ALL_FREE && |
2091 | mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK && | |
1bb93fd1 | 2092 | !(mp->m_flags & XFS_MOUNT_IKEEP)) { |
1bb93fd1 BF |
2093 | error = xfs_btree_delete(cur, &i); |
2094 | if (error) | |
2095 | goto error; | |
2096 | ASSERT(i == 1); | |
2097 | } else { | |
1bb93fd1 BF |
2098 | error = xfs_inobt_update(cur, &rec); |
2099 | if (error) | |
2100 | goto error; | |
2101 | } | |
2102 | ||
2103 | out: | |
2104 | error = xfs_check_agi_freecount(cur, agi); | |
2105 | if (error) | |
2106 | goto error; | |
2107 | ||
2108 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
2109 | return 0; | |
2110 | ||
2111 | error: | |
2112 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
2113 | return error; | |
2114 | } | |
2115 | ||
eb9a297a BF |
2116 | /* |
2117 | * Free disk inode. Carefully avoids touching the incore inode, all | |
2118 | * manipulations incore are the caller's responsibility. | |
2119 | * The on-disk inode is not changed by this operation, only the | |
2120 | * btree (free inode mask) is changed. | |
2121 | */ | |
2122 | int | |
2123 | xfs_difree( | |
2124 | struct xfs_trans *tp, /* transaction pointer */ | |
2125 | xfs_ino_t inode, /* inode to be freed */ | |
f33cea1a | 2126 | struct xfs_defer_ops *dfops, /* extents to free */ |
5a3b2e0a | 2127 | struct xfs_icluster *xic) /* cluster info if deleted */ |
eb9a297a BF |
2128 | { |
2129 | /* REFERENCED */ | |
2130 | xfs_agblock_t agbno; /* block number containing inode */ | |
2131 | struct xfs_buf *agbp; /* buffer for allocation group header */ | |
2132 | xfs_agino_t agino; /* allocation group inode number */ | |
2133 | xfs_agnumber_t agno; /* allocation group number */ | |
2134 | int error; /* error return value */ | |
2135 | struct xfs_mount *mp; /* mount structure for filesystem */ | |
2136 | struct xfs_inobt_rec_incore rec;/* btree record */ | |
2137 | ||
2138 | mp = tp->t_mountp; | |
2139 | ||
2140 | /* | |
2141 | * Break up inode number into its components. | |
2142 | */ | |
2143 | agno = XFS_INO_TO_AGNO(mp, inode); | |
2144 | if (agno >= mp->m_sb.sb_agcount) { | |
2145 | xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).", | |
2146 | __func__, agno, mp->m_sb.sb_agcount); | |
2147 | ASSERT(0); | |
12b53197 | 2148 | return -EINVAL; |
eb9a297a BF |
2149 | } |
2150 | agino = XFS_INO_TO_AGINO(mp, inode); | |
2151 | if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
2152 | xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).", | |
2153 | __func__, (unsigned long long)inode, | |
2154 | (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino)); | |
2155 | ASSERT(0); | |
12b53197 | 2156 | return -EINVAL; |
eb9a297a BF |
2157 | } |
2158 | agbno = XFS_AGINO_TO_AGBNO(mp, agino); | |
2159 | if (agbno >= mp->m_sb.sb_agblocks) { | |
2160 | xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).", | |
2161 | __func__, agbno, mp->m_sb.sb_agblocks); | |
2162 | ASSERT(0); | |
12b53197 | 2163 | return -EINVAL; |
eb9a297a BF |
2164 | } |
2165 | /* | |
2166 | * Get the allocation group header. | |
2167 | */ | |
2168 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); | |
2169 | if (error) { | |
2170 | xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.", | |
2171 | __func__, error); | |
2172 | return error; | |
2173 | } | |
2174 | ||
2175 | /* | |
2176 | * Fix up the inode allocation btree. | |
2177 | */ | |
f33cea1a | 2178 | error = xfs_difree_inobt(mp, tp, agbp, agino, dfops, xic, &rec); |
eb9a297a BF |
2179 | if (error) |
2180 | goto error0; | |
2181 | ||
1bb93fd1 BF |
2182 | /* |
2183 | * Fix up the free inode btree. | |
2184 | */ | |
2185 | if (xfs_sb_version_hasfinobt(&mp->m_sb)) { | |
2186 | error = xfs_difree_finobt(mp, tp, agbp, agino, &rec); | |
2187 | if (error) | |
2188 | goto error0; | |
2189 | } | |
2190 | ||
eb9a297a BF |
2191 | return 0; |
2192 | ||
2193 | error0: | |
2194 | return error; | |
2195 | } | |
2196 | ||
56b2de80 DC |
2197 | STATIC int |
2198 | xfs_imap_lookup( | |
2199 | struct xfs_mount *mp, | |
2200 | struct xfs_trans *tp, | |
2201 | xfs_agnumber_t agno, | |
2202 | xfs_agino_t agino, | |
2203 | xfs_agblock_t agbno, | |
2204 | xfs_agblock_t *chunk_agbno, | |
2205 | xfs_agblock_t *offset_agbno, | |
2206 | int flags) | |
2207 | { | |
2208 | struct xfs_inobt_rec_incore rec; | |
2209 | struct xfs_btree_cur *cur; | |
2210 | struct xfs_buf *agbp; | |
2211 | int error; | |
2212 | int i; | |
2213 | ||
2214 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); | |
2215 | if (error) { | |
a2ceac1f DC |
2216 | xfs_alert(mp, |
2217 | "%s: xfs_ialloc_read_agi() returned error %d, agno %d", | |
2218 | __func__, error, agno); | |
56b2de80 DC |
2219 | return error; |
2220 | } | |
2221 | ||
2222 | /* | |
2223 | * Lookup the inode record for the given agino. If the record cannot be | |
2224 | * found, then it's an invalid inode number and we should abort. Once | |
2225 | * we have a record, we need to ensure it contains the inode number | |
2226 | * we are looking up. | |
2227 | */ | |
70eb7337 | 2228 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
56b2de80 DC |
2229 | error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i); |
2230 | if (!error) { | |
2231 | if (i) | |
2232 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
2233 | if (!error && i == 0) | |
12b53197 | 2234 | error = -EINVAL; |
56b2de80 DC |
2235 | } |
2236 | ||
2237 | xfs_trans_brelse(tp, agbp); | |
d4fa7e6c | 2238 | xfs_btree_del_cursor(cur, error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); |
56b2de80 DC |
2239 | if (error) |
2240 | return error; | |
2241 | ||
2242 | /* check that the returned record contains the required inode */ | |
2243 | if (rec.ir_startino > agino || | |
ff105f75 | 2244 | rec.ir_startino + mp->m_ialloc_inos <= agino) |
12b53197 | 2245 | return -EINVAL; |
56b2de80 DC |
2246 | |
2247 | /* for untrusted inodes check it is allocated first */ | |
2248 | if ((flags & XFS_IGET_UNTRUSTED) && | |
2249 | (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino))) | |
12b53197 | 2250 | return -EINVAL; |
56b2de80 DC |
2251 | |
2252 | *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino); | |
2253 | *offset_agbno = agbno - *chunk_agbno; | |
2254 | return 0; | |
2255 | } | |
2bd0ea18 NS |
2256 | |
2257 | /* | |
56b2de80 | 2258 | * Return the location of the inode in imap, for mapping it into a buffer. |
2bd0ea18 | 2259 | */ |
2bd0ea18 | 2260 | int |
56b2de80 DC |
2261 | xfs_imap( |
2262 | xfs_mount_t *mp, /* file system mount structure */ | |
2263 | xfs_trans_t *tp, /* transaction pointer */ | |
2bd0ea18 | 2264 | xfs_ino_t ino, /* inode to locate */ |
56b2de80 DC |
2265 | struct xfs_imap *imap, /* location map structure */ |
2266 | uint flags) /* flags for inode btree lookup */ | |
2bd0ea18 NS |
2267 | { |
2268 | xfs_agblock_t agbno; /* block number of inode in the alloc group */ | |
2bd0ea18 NS |
2269 | xfs_agino_t agino; /* inode number within alloc group */ |
2270 | xfs_agnumber_t agno; /* allocation group number */ | |
2271 | int blks_per_cluster; /* num blocks per inode cluster */ | |
2272 | xfs_agblock_t chunk_agbno; /* first block in inode chunk */ | |
2bd0ea18 | 2273 | xfs_agblock_t cluster_agbno; /* first block in inode cluster */ |
2bd0ea18 | 2274 | int error; /* error code */ |
dfc130f3 | 2275 | int offset; /* index of inode in its buffer */ |
6bddecbc | 2276 | xfs_agblock_t offset_agbno; /* blks from chunk start to inode */ |
2bd0ea18 NS |
2277 | |
2278 | ASSERT(ino != NULLFSINO); | |
56b2de80 | 2279 | |
2bd0ea18 NS |
2280 | /* |
2281 | * Split up the inode number into its parts. | |
2282 | */ | |
2283 | agno = XFS_INO_TO_AGNO(mp, ino); | |
2284 | agino = XFS_INO_TO_AGINO(mp, ino); | |
2285 | agbno = XFS_AGINO_TO_AGBNO(mp, agino); | |
2286 | if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks || | |
63518810 NS |
2287 | ino != XFS_AGINO_TO_INO(mp, agno, agino)) { |
2288 | #ifdef DEBUG | |
56b2de80 DC |
2289 | /* |
2290 | * Don't output diagnostic information for untrusted inodes | |
2291 | * as they can be invalid without implying corruption. | |
2292 | */ | |
2293 | if (flags & XFS_IGET_UNTRUSTED) | |
12b53197 | 2294 | return -EINVAL; |
5000d01d | 2295 | if (agno >= mp->m_sb.sb_agcount) { |
a2ceac1f DC |
2296 | xfs_alert(mp, |
2297 | "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)", | |
2298 | __func__, agno, mp->m_sb.sb_agcount); | |
63518810 NS |
2299 | } |
2300 | if (agbno >= mp->m_sb.sb_agblocks) { | |
a2ceac1f DC |
2301 | xfs_alert(mp, |
2302 | "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)", | |
2303 | __func__, (unsigned long long)agbno, | |
2304 | (unsigned long)mp->m_sb.sb_agblocks); | |
63518810 NS |
2305 | } |
2306 | if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
a2ceac1f DC |
2307 | xfs_alert(mp, |
2308 | "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)", | |
2309 | __func__, ino, | |
2310 | XFS_AGINO_TO_INO(mp, agno, agino)); | |
63518810 | 2311 | } |
5e656dbb | 2312 | xfs_stack_trace(); |
63518810 | 2313 | #endif /* DEBUG */ |
12b53197 | 2314 | return -EINVAL; |
63518810 | 2315 | } |
56b2de80 | 2316 | |
ff105f75 | 2317 | blks_per_cluster = xfs_icluster_size_fsb(mp); |
56b2de80 DC |
2318 | |
2319 | /* | |
2320 | * For bulkstat and handle lookups, we have an untrusted inode number | |
2321 | * that we have to verify is valid. We cannot do this just by reading | |
2322 | * the inode buffer as it may have been unlinked and removed leaving | |
2323 | * inodes in stale state on disk. Hence we have to do a btree lookup | |
2324 | * in all cases where an untrusted inode number is passed. | |
2325 | */ | |
2326 | if (flags & XFS_IGET_UNTRUSTED) { | |
2327 | error = xfs_imap_lookup(mp, tp, agno, agino, agbno, | |
2328 | &chunk_agbno, &offset_agbno, flags); | |
2329 | if (error) | |
2330 | return error; | |
2331 | goto out_map; | |
2332 | } | |
2333 | ||
2334 | /* | |
2335 | * If the inode cluster size is the same as the blocksize or | |
2336 | * smaller we get to the buffer by simple arithmetics. | |
2337 | */ | |
ff105f75 | 2338 | if (blks_per_cluster == 1) { |
2bd0ea18 NS |
2339 | offset = XFS_INO_TO_OFFSET(mp, ino); |
2340 | ASSERT(offset < mp->m_sb.sb_inopblock); | |
56b2de80 DC |
2341 | |
2342 | imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno); | |
2343 | imap->im_len = XFS_FSB_TO_BB(mp, 1); | |
187f8c22 DW |
2344 | imap->im_boffset = (unsigned short)(offset << |
2345 | mp->m_sb.sb_inodelog); | |
2bd0ea18 NS |
2346 | return 0; |
2347 | } | |
56b2de80 DC |
2348 | |
2349 | /* | |
2350 | * If the inode chunks are aligned then use simple maths to | |
2351 | * find the location. Otherwise we have to do a btree | |
2352 | * lookup to find the location. | |
2353 | */ | |
2bd0ea18 NS |
2354 | if (mp->m_inoalign_mask) { |
2355 | offset_agbno = agbno & mp->m_inoalign_mask; | |
2356 | chunk_agbno = agbno - offset_agbno; | |
2357 | } else { | |
56b2de80 DC |
2358 | error = xfs_imap_lookup(mp, tp, agno, agino, agbno, |
2359 | &chunk_agbno, &offset_agbno, flags); | |
2bd0ea18 NS |
2360 | if (error) |
2361 | return error; | |
2bd0ea18 | 2362 | } |
56b2de80 DC |
2363 | |
2364 | out_map: | |
2bd0ea18 NS |
2365 | ASSERT(agbno >= chunk_agbno); |
2366 | cluster_agbno = chunk_agbno + | |
2367 | ((offset_agbno / blks_per_cluster) * blks_per_cluster); | |
2368 | offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) + | |
2369 | XFS_INO_TO_OFFSET(mp, ino); | |
56b2de80 DC |
2370 | |
2371 | imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno); | |
2372 | imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster); | |
919dbc78 | 2373 | imap->im_boffset = (unsigned short)(offset << mp->m_sb.sb_inodelog); |
56b2de80 DC |
2374 | |
2375 | /* | |
2376 | * If the inode number maps to a block outside the bounds | |
2377 | * of the file system then return NULL rather than calling | |
2378 | * read_buf and panicing when we get an error from the | |
2379 | * driver. | |
2380 | */ | |
2381 | if ((imap->im_blkno + imap->im_len) > | |
2382 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) { | |
a2ceac1f DC |
2383 | xfs_alert(mp, |
2384 | "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)", | |
2385 | __func__, (unsigned long long) imap->im_blkno, | |
56b2de80 DC |
2386 | (unsigned long long) imap->im_len, |
2387 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)); | |
12b53197 | 2388 | return -EINVAL; |
56b2de80 | 2389 | } |
2bd0ea18 | 2390 | return 0; |
2bd0ea18 NS |
2391 | } |
2392 | ||
2393 | /* | |
2394 | * Compute and fill in value of m_in_maxlevels. | |
2395 | */ | |
2396 | void | |
2397 | xfs_ialloc_compute_maxlevels( | |
2398 | xfs_mount_t *mp) /* file system mount structure */ | |
2399 | { | |
730e2a19 DW |
2400 | uint inodes; |
2401 | ||
2402 | inodes = (1LL << XFS_INO_AGINO_BITS(mp)) >> XFS_INODES_PER_CHUNK_LOG; | |
2403 | mp->m_in_maxlevels = xfs_btree_compute_maxlevels(mp, mp->m_inobt_mnr, | |
2404 | inodes); | |
2bd0ea18 NS |
2405 | } |
2406 | ||
2407 | /* | |
ff105f75 DC |
2408 | * Log specified fields for the ag hdr (inode section). The growth of the agi |
2409 | * structure over time requires that we interpret the buffer as two logical | |
2410 | * regions delineated by the end of the unlinked list. This is due to the size | |
2411 | * of the hash table and its location in the middle of the agi. | |
2412 | * | |
2413 | * For example, a request to log a field before agi_unlinked and a field after | |
2414 | * agi_unlinked could cause us to log the entire hash table and use an excessive | |
2415 | * amount of log space. To avoid this behavior, log the region up through | |
2416 | * agi_unlinked in one call and the region after agi_unlinked through the end of | |
2417 | * the structure in another. | |
2bd0ea18 NS |
2418 | */ |
2419 | void | |
2420 | xfs_ialloc_log_agi( | |
2421 | xfs_trans_t *tp, /* transaction pointer */ | |
2422 | xfs_buf_t *bp, /* allocation group header buffer */ | |
2423 | int fields) /* bitmask of fields to log */ | |
2424 | { | |
2425 | int first; /* first byte number */ | |
2426 | int last; /* last byte number */ | |
2427 | static const short offsets[] = { /* field starting offsets */ | |
2428 | /* keep in sync with bit definitions */ | |
2429 | offsetof(xfs_agi_t, agi_magicnum), | |
2430 | offsetof(xfs_agi_t, agi_versionnum), | |
2431 | offsetof(xfs_agi_t, agi_seqno), | |
2432 | offsetof(xfs_agi_t, agi_length), | |
2433 | offsetof(xfs_agi_t, agi_count), | |
2434 | offsetof(xfs_agi_t, agi_root), | |
2435 | offsetof(xfs_agi_t, agi_level), | |
2436 | offsetof(xfs_agi_t, agi_freecount), | |
2437 | offsetof(xfs_agi_t, agi_newino), | |
2438 | offsetof(xfs_agi_t, agi_dirino), | |
2439 | offsetof(xfs_agi_t, agi_unlinked), | |
c0a4c227 BF |
2440 | offsetof(xfs_agi_t, agi_free_root), |
2441 | offsetof(xfs_agi_t, agi_free_level), | |
2bd0ea18 NS |
2442 | sizeof(xfs_agi_t) |
2443 | }; | |
2444 | #ifdef DEBUG | |
2445 | xfs_agi_t *agi; /* allocation group header */ | |
2446 | ||
2447 | agi = XFS_BUF_TO_AGI(bp); | |
a2ceac1f | 2448 | ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC)); |
2bd0ea18 | 2449 | #endif |
ff105f75 | 2450 | |
2bd0ea18 | 2451 | /* |
c0a4c227 | 2452 | * Compute byte offsets for the first and last fields in the first |
ff105f75 DC |
2453 | * region and log the agi buffer. This only logs up through |
2454 | * agi_unlinked. | |
2bd0ea18 | 2455 | */ |
c0a4c227 BF |
2456 | if (fields & XFS_AGI_ALL_BITS_R1) { |
2457 | xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R1, | |
2458 | &first, &last); | |
2459 | xfs_trans_log_buf(tp, bp, first, last); | |
2460 | } | |
2461 | ||
2462 | /* | |
ff105f75 DC |
2463 | * Mask off the bits in the first region and calculate the first and |
2464 | * last field offsets for any bits in the second region. | |
c0a4c227 BF |
2465 | */ |
2466 | fields &= ~XFS_AGI_ALL_BITS_R1; | |
2467 | if (fields) { | |
2468 | xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R2, | |
2469 | &first, &last); | |
2470 | xfs_trans_log_buf(tp, bp, first, last); | |
2471 | } | |
2bd0ea18 NS |
2472 | } |
2473 | ||
56b2de80 DC |
2474 | #ifdef DEBUG |
2475 | STATIC void | |
2476 | xfs_check_agi_unlinked( | |
2477 | struct xfs_agi *agi) | |
2478 | { | |
2479 | int i; | |
2480 | ||
2481 | for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++) | |
2482 | ASSERT(agi->agi_unlinked[i]); | |
2483 | } | |
2484 | #else | |
2485 | #define xfs_check_agi_unlinked(agi) | |
2486 | #endif | |
2487 | ||
dd5b876e | 2488 | static bool |
a2ceac1f DC |
2489 | xfs_agi_verify( |
2490 | struct xfs_buf *bp) | |
2491 | { | |
2492 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
2493 | struct xfs_agi *agi = XFS_BUF_TO_AGI(bp); | |
a2ceac1f | 2494 | |
a65d8d29 BF |
2495 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
2496 | if (!uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid)) | |
2497 | return false; | |
2498 | if (!xfs_log_check_lsn(mp, | |
2499 | be64_to_cpu(XFS_BUF_TO_AGI(bp)->agi_lsn))) | |
dd5b876e | 2500 | return false; |
a65d8d29 BF |
2501 | } |
2502 | ||
a2ceac1f DC |
2503 | /* |
2504 | * Validate the magic number of the agi block. | |
2505 | */ | |
dd5b876e DC |
2506 | if (agi->agi_magicnum != cpu_to_be32(XFS_AGI_MAGIC)) |
2507 | return false; | |
2508 | if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum))) | |
2509 | return false; | |
a2ceac1f | 2510 | |
00795aae DW |
2511 | if (be32_to_cpu(agi->agi_level) < 1 || |
2512 | be32_to_cpu(agi->agi_level) > XFS_BTREE_MAXLEVELS) | |
5a35bf2c | 2513 | return false; |
00795aae DW |
2514 | |
2515 | if (xfs_sb_version_hasfinobt(&mp->m_sb) && | |
2516 | (be32_to_cpu(agi->agi_free_level) < 1 || | |
2517 | be32_to_cpu(agi->agi_free_level) > XFS_BTREE_MAXLEVELS)) | |
2518 | return false; | |
2519 | ||
a2ceac1f DC |
2520 | /* |
2521 | * during growfs operations, the perag is not fully initialised, | |
2522 | * so we can't use it for any useful checking. growfs ensures we can't | |
2523 | * use it by using uncached buffers that don't have the perag attached | |
2524 | * so we can detect and avoid this problem. | |
2525 | */ | |
dd5b876e DC |
2526 | if (bp->b_pag && be32_to_cpu(agi->agi_seqno) != bp->b_pag->pag_agno) |
2527 | return false; | |
a2ceac1f | 2528 | |
a2ceac1f | 2529 | xfs_check_agi_unlinked(agi); |
dd5b876e | 2530 | return true; |
a2ceac1f DC |
2531 | } |
2532 | ||
2533 | static void | |
2534 | xfs_agi_read_verify( | |
2535 | struct xfs_buf *bp) | |
2536 | { | |
dd5b876e | 2537 | struct xfs_mount *mp = bp->b_target->bt_mount; |
dd5b876e | 2538 | |
45922933 DC |
2539 | if (xfs_sb_version_hascrc(&mp->m_sb) && |
2540 | !xfs_buf_verify_cksum(bp, XFS_AGI_CRC_OFF)) | |
7e6c95f1 | 2541 | xfs_verifier_error(bp, -EFSBADCRC); |
45922933 | 2542 | else if (XFS_TEST_ERROR(!xfs_agi_verify(bp), mp, |
e2a190dd | 2543 | XFS_ERRTAG_IALLOC_READ_AGI)) |
7e6c95f1 | 2544 | xfs_verifier_error(bp, -EFSCORRUPTED); |
a2ceac1f DC |
2545 | } |
2546 | ||
2547 | static void | |
2548 | xfs_agi_write_verify( | |
2549 | struct xfs_buf *bp) | |
2550 | { | |
dd5b876e DC |
2551 | struct xfs_mount *mp = bp->b_target->bt_mount; |
2552 | struct xfs_buf_log_item *bip = bp->b_fspriv; | |
2553 | ||
2554 | if (!xfs_agi_verify(bp)) { | |
7e6c95f1 | 2555 | xfs_verifier_error(bp, -EFSCORRUPTED); |
dd5b876e DC |
2556 | return; |
2557 | } | |
2558 | ||
2559 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
2560 | return; | |
2561 | ||
2562 | if (bip) | |
2563 | XFS_BUF_TO_AGI(bp)->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn); | |
43b5aeed | 2564 | xfs_buf_update_cksum(bp, XFS_AGI_CRC_OFF); |
a2ceac1f DC |
2565 | } |
2566 | ||
2567 | const struct xfs_buf_ops xfs_agi_buf_ops = { | |
a3fac935 | 2568 | .name = "xfs_agi", |
a2ceac1f DC |
2569 | .verify_read = xfs_agi_read_verify, |
2570 | .verify_write = xfs_agi_write_verify, | |
2571 | }; | |
2572 | ||
2bd0ea18 NS |
2573 | /* |
2574 | * Read in the allocation group header (inode allocation section) | |
2575 | */ | |
2576 | int | |
56b2de80 DC |
2577 | xfs_read_agi( |
2578 | struct xfs_mount *mp, /* file system mount structure */ | |
2579 | struct xfs_trans *tp, /* transaction pointer */ | |
2580 | xfs_agnumber_t agno, /* allocation group number */ | |
2581 | struct xfs_buf **bpp) /* allocation group hdr buf */ | |
2bd0ea18 | 2582 | { |
56b2de80 | 2583 | int error; |
2bd0ea18 | 2584 | |
ff105f75 | 2585 | trace_xfs_read_agi(mp, agno); |
56b2de80 | 2586 | |
ff105f75 | 2587 | ASSERT(agno != NULLAGNUMBER); |
56b2de80 | 2588 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, |
9440d84d | 2589 | XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)), |
a2ceac1f | 2590 | XFS_FSS_TO_BB(mp, 1), 0, bpp, &xfs_agi_buf_ops); |
9440d84d | 2591 | if (error) |
2bd0ea18 | 2592 | return error; |
1b521475 ES |
2593 | if (tp) |
2594 | xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_AGI_BUF); | |
56b2de80 | 2595 | |
a2ceac1f | 2596 | xfs_buf_set_ref(*bpp, XFS_AGI_REF); |
56b2de80 DC |
2597 | return 0; |
2598 | } | |
2599 | ||
2600 | int | |
2601 | xfs_ialloc_read_agi( | |
2602 | struct xfs_mount *mp, /* file system mount structure */ | |
2603 | struct xfs_trans *tp, /* transaction pointer */ | |
2604 | xfs_agnumber_t agno, /* allocation group number */ | |
2605 | struct xfs_buf **bpp) /* allocation group hdr buf */ | |
2606 | { | |
2607 | struct xfs_agi *agi; /* allocation group header */ | |
2608 | struct xfs_perag *pag; /* per allocation group data */ | |
2609 | int error; | |
2610 | ||
ff105f75 DC |
2611 | trace_xfs_ialloc_read_agi(mp, agno); |
2612 | ||
56b2de80 DC |
2613 | error = xfs_read_agi(mp, tp, agno, bpp); |
2614 | if (error) | |
2615 | return error; | |
2616 | ||
2617 | agi = XFS_BUF_TO_AGI(*bpp); | |
2618 | pag = xfs_perag_get(mp, agno); | |
2bd0ea18 | 2619 | if (!pag->pagi_init) { |
6e3140c7 | 2620 | pag->pagi_freecount = be32_to_cpu(agi->agi_freecount); |
cdded3d8 | 2621 | pag->pagi_count = be32_to_cpu(agi->agi_count); |
2bd0ea18 | 2622 | pag->pagi_init = 1; |
9440d84d | 2623 | } |
9440d84d | 2624 | |
56b2de80 DC |
2625 | /* |
2626 | * It's possible for these to be out of sync if | |
2627 | * we are in the middle of a forced shutdown. | |
2628 | */ | |
2629 | ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) || | |
2630 | XFS_FORCED_SHUTDOWN(mp)); | |
2631 | xfs_perag_put(pag); | |
2bd0ea18 NS |
2632 | return 0; |
2633 | } | |
cdded3d8 DC |
2634 | |
2635 | /* | |
2636 | * Read in the agi to initialise the per-ag data in the mount structure | |
2637 | */ | |
2638 | int | |
2639 | xfs_ialloc_pagi_init( | |
2640 | xfs_mount_t *mp, /* file system mount structure */ | |
2641 | xfs_trans_t *tp, /* transaction pointer */ | |
2642 | xfs_agnumber_t agno) /* allocation group number */ | |
2643 | { | |
2644 | xfs_buf_t *bp = NULL; | |
2645 | int error; | |
2646 | ||
5e656dbb BN |
2647 | error = xfs_ialloc_read_agi(mp, tp, agno, &bp); |
2648 | if (error) | |
cdded3d8 DC |
2649 | return error; |
2650 | if (bp) | |
2651 | xfs_trans_brelse(tp, bp); | |
2652 | return 0; | |
2653 | } | |
be3bf25c DW |
2654 | |
2655 | /* Calculate the first and last possible inode number in an AG. */ | |
2656 | void | |
2657 | xfs_ialloc_agino_range( | |
2658 | struct xfs_mount *mp, | |
2659 | xfs_agnumber_t agno, | |
2660 | xfs_agino_t *first, | |
2661 | xfs_agino_t *last) | |
2662 | { | |
2663 | xfs_agblock_t bno; | |
2664 | xfs_agblock_t eoag; | |
2665 | ||
2666 | eoag = xfs_ag_block_count(mp, agno); | |
2667 | ||
2668 | /* | |
2669 | * Calculate the first inode, which will be in the first | |
2670 | * cluster-aligned block after the AGFL. | |
2671 | */ | |
2672 | bno = round_up(XFS_AGFL_BLOCK(mp) + 1, | |
2673 | xfs_ialloc_cluster_alignment(mp)); | |
2674 | *first = XFS_OFFBNO_TO_AGINO(mp, bno, 0); | |
2675 | ||
2676 | /* | |
2677 | * Calculate the last inode, which will be at the end of the | |
2678 | * last (aligned) cluster that can be allocated in the AG. | |
2679 | */ | |
2680 | bno = round_down(eoag, xfs_ialloc_cluster_alignment(mp)); | |
2681 | *last = XFS_OFFBNO_TO_AGINO(mp, bno, 0) - 1; | |
2682 | } | |
2683 | ||
2684 | /* | |
2685 | * Verify that an AG inode number pointer neither points outside the AG | |
2686 | * nor points at static metadata. | |
2687 | */ | |
2688 | bool | |
2689 | xfs_verify_agino( | |
2690 | struct xfs_mount *mp, | |
2691 | xfs_agnumber_t agno, | |
2692 | xfs_agino_t agino) | |
2693 | { | |
2694 | xfs_agino_t first; | |
2695 | xfs_agino_t last; | |
2696 | ||
2697 | xfs_ialloc_agino_range(mp, agno, &first, &last); | |
2698 | return agino >= first && agino <= last; | |
2699 | } | |
2700 | ||
2701 | /* | |
2702 | * Verify that an FS inode number pointer neither points outside the | |
2703 | * filesystem nor points at static AG metadata. | |
2704 | */ | |
2705 | bool | |
2706 | xfs_verify_ino( | |
2707 | struct xfs_mount *mp, | |
2708 | xfs_ino_t ino) | |
2709 | { | |
2710 | xfs_agnumber_t agno = XFS_INO_TO_AGNO(mp, ino); | |
2711 | xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino); | |
2712 | ||
2713 | if (agno >= mp->m_sb.sb_agcount) | |
2714 | return false; | |
2715 | if (XFS_AGINO_TO_INO(mp, agno, agino) != ino) | |
2716 | return false; | |
2717 | return xfs_verify_agino(mp, agno, agino); | |
2718 | } | |
2719 | ||
2720 | /* Is this an internal inode number? */ | |
2721 | bool | |
2722 | xfs_internal_inum( | |
2723 | struct xfs_mount *mp, | |
2724 | xfs_ino_t ino) | |
2725 | { | |
2726 | return ino == mp->m_sb.sb_rbmino || ino == mp->m_sb.sb_rsumino || | |
2727 | (xfs_sb_version_hasquota(&mp->m_sb) && | |
2728 | xfs_is_quota_inode(&mp->m_sb, ino)); | |
2729 | } | |
2730 | ||
2731 | /* | |
2732 | * Verify that a directory entry's inode number doesn't point at an internal | |
2733 | * inode, empty space, or static AG metadata. | |
2734 | */ | |
2735 | bool | |
2736 | xfs_verify_dir_ino( | |
2737 | struct xfs_mount *mp, | |
2738 | xfs_ino_t ino) | |
2739 | { | |
2740 | if (xfs_internal_inum(mp, ino)) | |
2741 | return false; | |
2742 | return xfs_verify_ino(mp, ino); | |
2743 | } |