]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_fs.h" |
70a9883c | 20 | #include "xfs_shared.h" |
239880ef DC |
21 | #include "xfs_format.h" |
22 | #include "xfs_log_format.h" | |
23 | #include "xfs_trans_resv.h" | |
a844f451 | 24 | #include "xfs_bit.h" |
1da177e4 | 25 | #include "xfs_sb.h" |
1da177e4 | 26 | #include "xfs_mount.h" |
3ab78df2 | 27 | #include "xfs_defer.h" |
1da177e4 | 28 | #include "xfs_inode.h" |
a844f451 NS |
29 | #include "xfs_btree.h" |
30 | #include "xfs_ialloc.h" | |
a4fbe6ab | 31 | #include "xfs_ialloc_btree.h" |
1da177e4 | 32 | #include "xfs_alloc.h" |
1da177e4 LT |
33 | #include "xfs_rtalloc.h" |
34 | #include "xfs_error.h" | |
35 | #include "xfs_bmap.h" | |
983d09ff | 36 | #include "xfs_cksum.h" |
239880ef | 37 | #include "xfs_trans.h" |
983d09ff | 38 | #include "xfs_buf_item.h" |
ddf6ad01 | 39 | #include "xfs_icreate_item.h" |
7bb85ef3 | 40 | #include "xfs_icache.h" |
d123031a | 41 | #include "xfs_trace.h" |
a45086e2 | 42 | #include "xfs_log.h" |
1da177e4 | 43 | |
1da177e4 LT |
44 | |
45 | /* | |
46 | * Allocation group level functions. | |
47 | */ | |
75de2a91 DC |
48 | static inline int |
49 | xfs_ialloc_cluster_alignment( | |
7a1df156 | 50 | struct xfs_mount *mp) |
75de2a91 | 51 | { |
7a1df156 DC |
52 | if (xfs_sb_version_hasalign(&mp->m_sb) && |
53 | mp->m_sb.sb_inoalignmt >= | |
54 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) | |
55 | return mp->m_sb.sb_inoalignmt; | |
75de2a91 DC |
56 | return 1; |
57 | } | |
1da177e4 | 58 | |
fe033cc8 | 59 | /* |
21875505 | 60 | * Lookup a record by ino in the btree given by cur. |
fe033cc8 | 61 | */ |
81e25176 | 62 | int /* error */ |
21875505 | 63 | xfs_inobt_lookup( |
fe033cc8 CH |
64 | struct xfs_btree_cur *cur, /* btree cursor */ |
65 | xfs_agino_t ino, /* starting inode of chunk */ | |
21875505 | 66 | xfs_lookup_t dir, /* <=, >=, == */ |
fe033cc8 CH |
67 | int *stat) /* success/failure */ |
68 | { | |
69 | cur->bc_rec.i.ir_startino = ino; | |
5419040f BF |
70 | cur->bc_rec.i.ir_holemask = 0; |
71 | cur->bc_rec.i.ir_count = 0; | |
21875505 CH |
72 | cur->bc_rec.i.ir_freecount = 0; |
73 | cur->bc_rec.i.ir_free = 0; | |
74 | return xfs_btree_lookup(cur, dir, stat); | |
fe033cc8 CH |
75 | } |
76 | ||
278d0ca1 | 77 | /* |
afabc24a | 78 | * Update the record referred to by cur to the value given. |
278d0ca1 CH |
79 | * This either works (return 0) or gets an EFSCORRUPTED error. |
80 | */ | |
81 | STATIC int /* error */ | |
82 | xfs_inobt_update( | |
83 | struct xfs_btree_cur *cur, /* btree cursor */ | |
afabc24a | 84 | xfs_inobt_rec_incore_t *irec) /* btree record */ |
278d0ca1 CH |
85 | { |
86 | union xfs_btree_rec rec; | |
87 | ||
afabc24a | 88 | rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino); |
5419040f BF |
89 | if (xfs_sb_version_hassparseinodes(&cur->bc_mp->m_sb)) { |
90 | rec.inobt.ir_u.sp.ir_holemask = cpu_to_be16(irec->ir_holemask); | |
91 | rec.inobt.ir_u.sp.ir_count = irec->ir_count; | |
92 | rec.inobt.ir_u.sp.ir_freecount = irec->ir_freecount; | |
93 | } else { | |
94 | /* ir_holemask/ir_count not supported on-disk */ | |
95 | rec.inobt.ir_u.f.ir_freecount = cpu_to_be32(irec->ir_freecount); | |
96 | } | |
afabc24a | 97 | rec.inobt.ir_free = cpu_to_be64(irec->ir_free); |
278d0ca1 CH |
98 | return xfs_btree_update(cur, &rec); |
99 | } | |
100 | ||
8cc938fe CH |
101 | /* |
102 | * Get the data from the pointed-to record. | |
103 | */ | |
104 | int /* error */ | |
105 | xfs_inobt_get_rec( | |
106 | struct xfs_btree_cur *cur, /* btree cursor */ | |
2e287a73 | 107 | xfs_inobt_rec_incore_t *irec, /* btree record */ |
8cc938fe CH |
108 | int *stat) /* output: success/failure */ |
109 | { | |
110 | union xfs_btree_rec *rec; | |
111 | int error; | |
112 | ||
113 | error = xfs_btree_get_rec(cur, &rec, stat); | |
5419040f BF |
114 | if (error || *stat == 0) |
115 | return error; | |
116 | ||
117 | irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino); | |
118 | if (xfs_sb_version_hassparseinodes(&cur->bc_mp->m_sb)) { | |
119 | irec->ir_holemask = be16_to_cpu(rec->inobt.ir_u.sp.ir_holemask); | |
120 | irec->ir_count = rec->inobt.ir_u.sp.ir_count; | |
121 | irec->ir_freecount = rec->inobt.ir_u.sp.ir_freecount; | |
122 | } else { | |
123 | /* | |
124 | * ir_holemask/ir_count not supported on-disk. Fill in hardcoded | |
125 | * values for full inode chunks. | |
126 | */ | |
127 | irec->ir_holemask = XFS_INOBT_HOLEMASK_FULL; | |
128 | irec->ir_count = XFS_INODES_PER_CHUNK; | |
129 | irec->ir_freecount = | |
130 | be32_to_cpu(rec->inobt.ir_u.f.ir_freecount); | |
8cc938fe | 131 | } |
5419040f BF |
132 | irec->ir_free = be64_to_cpu(rec->inobt.ir_free); |
133 | ||
134 | return 0; | |
8cc938fe CH |
135 | } |
136 | ||
0aa0a756 BF |
137 | /* |
138 | * Insert a single inobt record. Cursor must already point to desired location. | |
139 | */ | |
140 | STATIC int | |
141 | xfs_inobt_insert_rec( | |
142 | struct xfs_btree_cur *cur, | |
5419040f BF |
143 | __uint16_t holemask, |
144 | __uint8_t count, | |
0aa0a756 BF |
145 | __int32_t freecount, |
146 | xfs_inofree_t free, | |
147 | int *stat) | |
148 | { | |
5419040f BF |
149 | cur->bc_rec.i.ir_holemask = holemask; |
150 | cur->bc_rec.i.ir_count = count; | |
0aa0a756 BF |
151 | cur->bc_rec.i.ir_freecount = freecount; |
152 | cur->bc_rec.i.ir_free = free; | |
153 | return xfs_btree_insert(cur, stat); | |
154 | } | |
155 | ||
156 | /* | |
157 | * Insert records describing a newly allocated inode chunk into the inobt. | |
158 | */ | |
159 | STATIC int | |
160 | xfs_inobt_insert( | |
161 | struct xfs_mount *mp, | |
162 | struct xfs_trans *tp, | |
163 | struct xfs_buf *agbp, | |
164 | xfs_agino_t newino, | |
165 | xfs_agino_t newlen, | |
166 | xfs_btnum_t btnum) | |
167 | { | |
168 | struct xfs_btree_cur *cur; | |
169 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
170 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
171 | xfs_agino_t thisino; | |
172 | int i; | |
173 | int error; | |
174 | ||
175 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum); | |
176 | ||
177 | for (thisino = newino; | |
178 | thisino < newino + newlen; | |
179 | thisino += XFS_INODES_PER_CHUNK) { | |
180 | error = xfs_inobt_lookup(cur, thisino, XFS_LOOKUP_EQ, &i); | |
181 | if (error) { | |
182 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
183 | return error; | |
184 | } | |
185 | ASSERT(i == 0); | |
186 | ||
5419040f BF |
187 | error = xfs_inobt_insert_rec(cur, XFS_INOBT_HOLEMASK_FULL, |
188 | XFS_INODES_PER_CHUNK, | |
189 | XFS_INODES_PER_CHUNK, | |
0aa0a756 BF |
190 | XFS_INOBT_ALL_FREE, &i); |
191 | if (error) { | |
192 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
193 | return error; | |
194 | } | |
195 | ASSERT(i == 1); | |
196 | } | |
197 | ||
198 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
199 | ||
200 | return 0; | |
201 | } | |
202 | ||
0b48db80 DC |
203 | /* |
204 | * Verify that the number of free inodes in the AGI is correct. | |
205 | */ | |
206 | #ifdef DEBUG | |
207 | STATIC int | |
208 | xfs_check_agi_freecount( | |
209 | struct xfs_btree_cur *cur, | |
210 | struct xfs_agi *agi) | |
211 | { | |
212 | if (cur->bc_nlevels == 1) { | |
213 | xfs_inobt_rec_incore_t rec; | |
214 | int freecount = 0; | |
215 | int error; | |
216 | int i; | |
217 | ||
21875505 | 218 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); |
0b48db80 DC |
219 | if (error) |
220 | return error; | |
221 | ||
222 | do { | |
223 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
224 | if (error) | |
225 | return error; | |
226 | ||
227 | if (i) { | |
228 | freecount += rec.ir_freecount; | |
229 | error = xfs_btree_increment(cur, 0, &i); | |
230 | if (error) | |
231 | return error; | |
232 | } | |
233 | } while (i == 1); | |
234 | ||
235 | if (!XFS_FORCED_SHUTDOWN(cur->bc_mp)) | |
236 | ASSERT(freecount == be32_to_cpu(agi->agi_freecount)); | |
237 | } | |
238 | return 0; | |
239 | } | |
240 | #else | |
241 | #define xfs_check_agi_freecount(cur, agi) 0 | |
242 | #endif | |
243 | ||
85c0b2ab | 244 | /* |
28c8e41a DC |
245 | * Initialise a new set of inodes. When called without a transaction context |
246 | * (e.g. from recovery) we initiate a delayed write of the inode buffers rather | |
247 | * than logging them (which in a transaction context puts them into the AIL | |
248 | * for writeback rather than the xfsbufd queue). | |
85c0b2ab | 249 | */ |
ddf6ad01 | 250 | int |
85c0b2ab DC |
251 | xfs_ialloc_inode_init( |
252 | struct xfs_mount *mp, | |
253 | struct xfs_trans *tp, | |
28c8e41a | 254 | struct list_head *buffer_list, |
463958af | 255 | int icount, |
85c0b2ab DC |
256 | xfs_agnumber_t agno, |
257 | xfs_agblock_t agbno, | |
258 | xfs_agblock_t length, | |
259 | unsigned int gen) | |
260 | { | |
261 | struct xfs_buf *fbuf; | |
262 | struct xfs_dinode *free; | |
6e0c7b8c | 263 | int nbufs, blks_per_cluster, inodes_per_cluster; |
85c0b2ab DC |
264 | int version; |
265 | int i, j; | |
266 | xfs_daddr_t d; | |
93848a99 | 267 | xfs_ino_t ino = 0; |
85c0b2ab DC |
268 | |
269 | /* | |
6e0c7b8c JL |
270 | * Loop over the new block(s), filling in the inodes. For small block |
271 | * sizes, manipulate the inodes in buffers which are multiples of the | |
272 | * blocks size. | |
85c0b2ab | 273 | */ |
6e0c7b8c JL |
274 | blks_per_cluster = xfs_icluster_size_fsb(mp); |
275 | inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog; | |
276 | nbufs = length / blks_per_cluster; | |
85c0b2ab DC |
277 | |
278 | /* | |
93848a99 CH |
279 | * Figure out what version number to use in the inodes we create. If |
280 | * the superblock version has caught up to the one that supports the new | |
281 | * inode format, then use the new inode version. Otherwise use the old | |
282 | * version so that old kernels will continue to be able to use the file | |
283 | * system. | |
284 | * | |
285 | * For v3 inodes, we also need to write the inode number into the inode, | |
286 | * so calculate the first inode number of the chunk here as | |
287 | * XFS_OFFBNO_TO_AGINO() only works within a filesystem block, not | |
288 | * across multiple filesystem blocks (such as a cluster) and so cannot | |
289 | * be used in the cluster buffer loop below. | |
290 | * | |
291 | * Further, because we are writing the inode directly into the buffer | |
292 | * and calculating a CRC on the entire inode, we have ot log the entire | |
293 | * inode so that the entire range the CRC covers is present in the log. | |
294 | * That means for v3 inode we log the entire buffer rather than just the | |
295 | * inode cores. | |
85c0b2ab | 296 | */ |
93848a99 CH |
297 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
298 | version = 3; | |
299 | ino = XFS_AGINO_TO_INO(mp, agno, | |
300 | XFS_OFFBNO_TO_AGINO(mp, agbno, 0)); | |
ddf6ad01 DC |
301 | |
302 | /* | |
303 | * log the initialisation that is about to take place as an | |
304 | * logical operation. This means the transaction does not | |
305 | * need to log the physical changes to the inode buffers as log | |
306 | * recovery will know what initialisation is actually needed. | |
307 | * Hence we only need to log the buffers as "ordered" buffers so | |
308 | * they track in the AIL as if they were physically logged. | |
309 | */ | |
310 | if (tp) | |
463958af | 311 | xfs_icreate_log(tp, agno, agbno, icount, |
ddf6ad01 | 312 | mp->m_sb.sb_inodesize, length, gen); |
263997a6 | 313 | } else |
85c0b2ab | 314 | version = 2; |
85c0b2ab DC |
315 | |
316 | for (j = 0; j < nbufs; j++) { | |
317 | /* | |
318 | * Get the block. | |
319 | */ | |
320 | d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster)); | |
321 | fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, | |
7c4cebe8 DC |
322 | mp->m_bsize * blks_per_cluster, |
323 | XBF_UNMAPPED); | |
2a30f36d | 324 | if (!fbuf) |
2451337d | 325 | return -ENOMEM; |
ddf6ad01 DC |
326 | |
327 | /* Initialize the inode buffers and log them appropriately. */ | |
1813dd64 | 328 | fbuf->b_ops = &xfs_inode_buf_ops; |
93848a99 | 329 | xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length)); |
6e0c7b8c | 330 | for (i = 0; i < inodes_per_cluster; i++) { |
85c0b2ab | 331 | int ioffset = i << mp->m_sb.sb_inodelog; |
93848a99 | 332 | uint isize = xfs_dinode_size(version); |
85c0b2ab DC |
333 | |
334 | free = xfs_make_iptr(mp, fbuf, i); | |
335 | free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC); | |
336 | free->di_version = version; | |
337 | free->di_gen = cpu_to_be32(gen); | |
338 | free->di_next_unlinked = cpu_to_be32(NULLAGINO); | |
93848a99 CH |
339 | |
340 | if (version == 3) { | |
341 | free->di_ino = cpu_to_be64(ino); | |
342 | ino++; | |
ce748eaa ES |
343 | uuid_copy(&free->di_uuid, |
344 | &mp->m_sb.sb_meta_uuid); | |
93848a99 | 345 | xfs_dinode_calc_crc(mp, free); |
28c8e41a | 346 | } else if (tp) { |
93848a99 CH |
347 | /* just log the inode core */ |
348 | xfs_trans_log_buf(tp, fbuf, ioffset, | |
349 | ioffset + isize - 1); | |
350 | } | |
351 | } | |
28c8e41a DC |
352 | |
353 | if (tp) { | |
354 | /* | |
355 | * Mark the buffer as an inode allocation buffer so it | |
356 | * sticks in AIL at the point of this allocation | |
357 | * transaction. This ensures the they are on disk before | |
358 | * the tail of the log can be moved past this | |
359 | * transaction (i.e. by preventing relogging from moving | |
360 | * it forward in the log). | |
361 | */ | |
362 | xfs_trans_inode_alloc_buf(tp, fbuf); | |
363 | if (version == 3) { | |
ddf6ad01 DC |
364 | /* |
365 | * Mark the buffer as ordered so that they are | |
366 | * not physically logged in the transaction but | |
367 | * still tracked in the AIL as part of the | |
368 | * transaction and pin the log appropriately. | |
369 | */ | |
370 | xfs_trans_ordered_buf(tp, fbuf); | |
28c8e41a DC |
371 | xfs_trans_log_buf(tp, fbuf, 0, |
372 | BBTOB(fbuf->b_length) - 1); | |
373 | } | |
374 | } else { | |
375 | fbuf->b_flags |= XBF_DONE; | |
376 | xfs_buf_delwri_queue(fbuf, buffer_list); | |
377 | xfs_buf_relse(fbuf); | |
85c0b2ab | 378 | } |
85c0b2ab | 379 | } |
2a30f36d | 380 | return 0; |
85c0b2ab DC |
381 | } |
382 | ||
56d1115c BF |
383 | /* |
384 | * Align startino and allocmask for a recently allocated sparse chunk such that | |
385 | * they are fit for insertion (or merge) into the on-disk inode btrees. | |
386 | * | |
387 | * Background: | |
388 | * | |
389 | * When enabled, sparse inode support increases the inode alignment from cluster | |
390 | * size to inode chunk size. This means that the minimum range between two | |
391 | * non-adjacent inode records in the inobt is large enough for a full inode | |
392 | * record. This allows for cluster sized, cluster aligned block allocation | |
393 | * without need to worry about whether the resulting inode record overlaps with | |
394 | * another record in the tree. Without this basic rule, we would have to deal | |
395 | * with the consequences of overlap by potentially undoing recent allocations in | |
396 | * the inode allocation codepath. | |
397 | * | |
398 | * Because of this alignment rule (which is enforced on mount), there are two | |
399 | * inobt possibilities for newly allocated sparse chunks. One is that the | |
400 | * aligned inode record for the chunk covers a range of inodes not already | |
401 | * covered in the inobt (i.e., it is safe to insert a new sparse record). The | |
402 | * other is that a record already exists at the aligned startino that considers | |
403 | * the newly allocated range as sparse. In the latter case, record content is | |
404 | * merged in hope that sparse inode chunks fill to full chunks over time. | |
405 | */ | |
406 | STATIC void | |
407 | xfs_align_sparse_ino( | |
408 | struct xfs_mount *mp, | |
409 | xfs_agino_t *startino, | |
410 | uint16_t *allocmask) | |
411 | { | |
412 | xfs_agblock_t agbno; | |
413 | xfs_agblock_t mod; | |
414 | int offset; | |
415 | ||
416 | agbno = XFS_AGINO_TO_AGBNO(mp, *startino); | |
417 | mod = agbno % mp->m_sb.sb_inoalignmt; | |
418 | if (!mod) | |
419 | return; | |
420 | ||
421 | /* calculate the inode offset and align startino */ | |
422 | offset = mod << mp->m_sb.sb_inopblog; | |
423 | *startino -= offset; | |
424 | ||
425 | /* | |
426 | * Since startino has been aligned down, left shift allocmask such that | |
427 | * it continues to represent the same physical inodes relative to the | |
428 | * new startino. | |
429 | */ | |
430 | *allocmask <<= offset / XFS_INODES_PER_HOLEMASK_BIT; | |
431 | } | |
432 | ||
433 | /* | |
434 | * Determine whether the source inode record can merge into the target. Both | |
435 | * records must be sparse, the inode ranges must match and there must be no | |
436 | * allocation overlap between the records. | |
437 | */ | |
438 | STATIC bool | |
439 | __xfs_inobt_can_merge( | |
440 | struct xfs_inobt_rec_incore *trec, /* tgt record */ | |
441 | struct xfs_inobt_rec_incore *srec) /* src record */ | |
442 | { | |
443 | uint64_t talloc; | |
444 | uint64_t salloc; | |
445 | ||
446 | /* records must cover the same inode range */ | |
447 | if (trec->ir_startino != srec->ir_startino) | |
448 | return false; | |
449 | ||
450 | /* both records must be sparse */ | |
451 | if (!xfs_inobt_issparse(trec->ir_holemask) || | |
452 | !xfs_inobt_issparse(srec->ir_holemask)) | |
453 | return false; | |
454 | ||
455 | /* both records must track some inodes */ | |
456 | if (!trec->ir_count || !srec->ir_count) | |
457 | return false; | |
458 | ||
459 | /* can't exceed capacity of a full record */ | |
460 | if (trec->ir_count + srec->ir_count > XFS_INODES_PER_CHUNK) | |
461 | return false; | |
462 | ||
463 | /* verify there is no allocation overlap */ | |
464 | talloc = xfs_inobt_irec_to_allocmask(trec); | |
465 | salloc = xfs_inobt_irec_to_allocmask(srec); | |
466 | if (talloc & salloc) | |
467 | return false; | |
468 | ||
469 | return true; | |
470 | } | |
471 | ||
472 | /* | |
473 | * Merge the source inode record into the target. The caller must call | |
474 | * __xfs_inobt_can_merge() to ensure the merge is valid. | |
475 | */ | |
476 | STATIC void | |
477 | __xfs_inobt_rec_merge( | |
478 | struct xfs_inobt_rec_incore *trec, /* target */ | |
479 | struct xfs_inobt_rec_incore *srec) /* src */ | |
480 | { | |
481 | ASSERT(trec->ir_startino == srec->ir_startino); | |
482 | ||
483 | /* combine the counts */ | |
484 | trec->ir_count += srec->ir_count; | |
485 | trec->ir_freecount += srec->ir_freecount; | |
486 | ||
487 | /* | |
488 | * Merge the holemask and free mask. For both fields, 0 bits refer to | |
489 | * allocated inodes. We combine the allocated ranges with bitwise AND. | |
490 | */ | |
491 | trec->ir_holemask &= srec->ir_holemask; | |
492 | trec->ir_free &= srec->ir_free; | |
493 | } | |
494 | ||
495 | /* | |
496 | * Insert a new sparse inode chunk into the associated inode btree. The inode | |
497 | * record for the sparse chunk is pre-aligned to a startino that should match | |
498 | * any pre-existing sparse inode record in the tree. This allows sparse chunks | |
499 | * to fill over time. | |
500 | * | |
501 | * This function supports two modes of handling preexisting records depending on | |
502 | * the merge flag. If merge is true, the provided record is merged with the | |
503 | * existing record and updated in place. The merged record is returned in nrec. | |
504 | * If merge is false, an existing record is replaced with the provided record. | |
505 | * If no preexisting record exists, the provided record is always inserted. | |
506 | * | |
507 | * It is considered corruption if a merge is requested and not possible. Given | |
508 | * the sparse inode alignment constraints, this should never happen. | |
509 | */ | |
510 | STATIC int | |
511 | xfs_inobt_insert_sprec( | |
512 | struct xfs_mount *mp, | |
513 | struct xfs_trans *tp, | |
514 | struct xfs_buf *agbp, | |
515 | int btnum, | |
516 | struct xfs_inobt_rec_incore *nrec, /* in/out: new/merged rec. */ | |
517 | bool merge) /* merge or replace */ | |
518 | { | |
519 | struct xfs_btree_cur *cur; | |
520 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
521 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
522 | int error; | |
523 | int i; | |
524 | struct xfs_inobt_rec_incore rec; | |
525 | ||
526 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum); | |
527 | ||
528 | /* the new record is pre-aligned so we know where to look */ | |
529 | error = xfs_inobt_lookup(cur, nrec->ir_startino, XFS_LOOKUP_EQ, &i); | |
530 | if (error) | |
531 | goto error; | |
532 | /* if nothing there, insert a new record and return */ | |
533 | if (i == 0) { | |
534 | error = xfs_inobt_insert_rec(cur, nrec->ir_holemask, | |
535 | nrec->ir_count, nrec->ir_freecount, | |
536 | nrec->ir_free, &i); | |
537 | if (error) | |
538 | goto error; | |
539 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error); | |
540 | ||
541 | goto out; | |
542 | } | |
543 | ||
544 | /* | |
545 | * A record exists at this startino. Merge or replace the record | |
546 | * depending on what we've been asked to do. | |
547 | */ | |
548 | if (merge) { | |
549 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
550 | if (error) | |
551 | goto error; | |
552 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error); | |
553 | XFS_WANT_CORRUPTED_GOTO(mp, | |
554 | rec.ir_startino == nrec->ir_startino, | |
555 | error); | |
556 | ||
557 | /* | |
558 | * This should never fail. If we have coexisting records that | |
559 | * cannot merge, something is seriously wrong. | |
560 | */ | |
561 | XFS_WANT_CORRUPTED_GOTO(mp, __xfs_inobt_can_merge(nrec, &rec), | |
562 | error); | |
563 | ||
564 | trace_xfs_irec_merge_pre(mp, agno, rec.ir_startino, | |
565 | rec.ir_holemask, nrec->ir_startino, | |
566 | nrec->ir_holemask); | |
567 | ||
568 | /* merge to nrec to output the updated record */ | |
569 | __xfs_inobt_rec_merge(nrec, &rec); | |
570 | ||
571 | trace_xfs_irec_merge_post(mp, agno, nrec->ir_startino, | |
572 | nrec->ir_holemask); | |
573 | ||
574 | error = xfs_inobt_rec_check_count(mp, nrec); | |
575 | if (error) | |
576 | goto error; | |
577 | } | |
578 | ||
579 | error = xfs_inobt_update(cur, nrec); | |
580 | if (error) | |
581 | goto error; | |
582 | ||
583 | out: | |
584 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
585 | return 0; | |
586 | error: | |
587 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
588 | return error; | |
589 | } | |
590 | ||
1da177e4 LT |
591 | /* |
592 | * Allocate new inodes in the allocation group specified by agbp. | |
593 | * Return 0 for success, else error code. | |
594 | */ | |
595 | STATIC int /* error code or 0 */ | |
596 | xfs_ialloc_ag_alloc( | |
597 | xfs_trans_t *tp, /* transaction pointer */ | |
598 | xfs_buf_t *agbp, /* alloc group buffer */ | |
599 | int *alloc) | |
600 | { | |
601 | xfs_agi_t *agi; /* allocation group header */ | |
602 | xfs_alloc_arg_t args; /* allocation argument structure */ | |
92821e2b | 603 | xfs_agnumber_t agno; |
1da177e4 | 604 | int error; |
1da177e4 LT |
605 | xfs_agino_t newino; /* new first inode's number */ |
606 | xfs_agino_t newlen; /* new number of inodes */ | |
3ccb8b5f | 607 | int isaligned = 0; /* inode allocation at stripe unit */ |
1da177e4 | 608 | /* boundary */ |
56d1115c BF |
609 | uint16_t allocmask = (uint16_t) -1; /* init. to full chunk */ |
610 | struct xfs_inobt_rec_incore rec; | |
44b56e0a | 611 | struct xfs_perag *pag; |
1cdadee1 BF |
612 | int do_sparse = 0; |
613 | ||
a0041684 | 614 | memset(&args, 0, sizeof(args)); |
1da177e4 LT |
615 | args.tp = tp; |
616 | args.mp = tp->t_mountp; | |
1cdadee1 | 617 | args.fsbno = NULLFSBLOCK; |
1da177e4 | 618 | |
46fc58da BF |
619 | #ifdef DEBUG |
620 | /* randomly do sparse inode allocations */ | |
621 | if (xfs_sb_version_hassparseinodes(&tp->t_mountp->m_sb) && | |
622 | args.mp->m_ialloc_min_blks < args.mp->m_ialloc_blks) | |
623 | do_sparse = prandom_u32() & 1; | |
624 | #endif | |
625 | ||
1da177e4 LT |
626 | /* |
627 | * Locking will ensure that we don't have two callers in here | |
628 | * at one time. | |
629 | */ | |
71783438 | 630 | newlen = args.mp->m_ialloc_inos; |
1da177e4 | 631 | if (args.mp->m_maxicount && |
74f9ce1c | 632 | percpu_counter_read_positive(&args.mp->m_icount) + newlen > |
501ab323 | 633 | args.mp->m_maxicount) |
2451337d | 634 | return -ENOSPC; |
126cd105 | 635 | args.minlen = args.maxlen = args.mp->m_ialloc_blks; |
1da177e4 | 636 | /* |
3ccb8b5f GO |
637 | * First try to allocate inodes contiguous with the last-allocated |
638 | * chunk of inodes. If the filesystem is striped, this will fill | |
639 | * an entire stripe unit with inodes. | |
28c8e41a | 640 | */ |
1da177e4 | 641 | agi = XFS_BUF_TO_AGI(agbp); |
3ccb8b5f | 642 | newino = be32_to_cpu(agi->agi_newino); |
85c0b2ab | 643 | agno = be32_to_cpu(agi->agi_seqno); |
019ff2d5 | 644 | args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) + |
126cd105 | 645 | args.mp->m_ialloc_blks; |
1cdadee1 BF |
646 | if (do_sparse) |
647 | goto sparse_alloc; | |
019ff2d5 NS |
648 | if (likely(newino != NULLAGINO && |
649 | (args.agbno < be32_to_cpu(agi->agi_length)))) { | |
85c0b2ab | 650 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
3ccb8b5f | 651 | args.type = XFS_ALLOCTYPE_THIS_BNO; |
3ccb8b5f | 652 | args.prod = 1; |
75de2a91 | 653 | |
3ccb8b5f | 654 | /* |
75de2a91 DC |
655 | * We need to take into account alignment here to ensure that |
656 | * we don't modify the free list if we fail to have an exact | |
657 | * block. If we don't have an exact match, and every oher | |
658 | * attempt allocation attempt fails, we'll end up cancelling | |
659 | * a dirty transaction and shutting down. | |
660 | * | |
661 | * For an exact allocation, alignment must be 1, | |
662 | * however we need to take cluster alignment into account when | |
663 | * fixing up the freelist. Use the minalignslop field to | |
664 | * indicate that extra blocks might be required for alignment, | |
665 | * but not to use them in the actual exact allocation. | |
3ccb8b5f | 666 | */ |
75de2a91 | 667 | args.alignment = 1; |
7a1df156 | 668 | args.minalignslop = xfs_ialloc_cluster_alignment(args.mp) - 1; |
75de2a91 DC |
669 | |
670 | /* Allow space for the inode btree to split. */ | |
0d87e656 | 671 | args.minleft = args.mp->m_in_maxlevels - 1; |
3ccb8b5f GO |
672 | if ((error = xfs_alloc_vextent(&args))) |
673 | return error; | |
e480a723 BF |
674 | |
675 | /* | |
676 | * This request might have dirtied the transaction if the AG can | |
677 | * satisfy the request, but the exact block was not available. | |
678 | * If the allocation did fail, subsequent requests will relax | |
679 | * the exact agbno requirement and increase the alignment | |
680 | * instead. It is critical that the total size of the request | |
681 | * (len + alignment + slop) does not increase from this point | |
682 | * on, so reset minalignslop to ensure it is not included in | |
683 | * subsequent requests. | |
684 | */ | |
685 | args.minalignslop = 0; | |
1cdadee1 | 686 | } |
1da177e4 | 687 | |
3ccb8b5f GO |
688 | if (unlikely(args.fsbno == NULLFSBLOCK)) { |
689 | /* | |
690 | * Set the alignment for the allocation. | |
691 | * If stripe alignment is turned on then align at stripe unit | |
692 | * boundary. | |
019ff2d5 NS |
693 | * If the cluster size is smaller than a filesystem block |
694 | * then we're doing I/O for inodes in filesystem block size | |
3ccb8b5f GO |
695 | * pieces, so don't need alignment anyway. |
696 | */ | |
697 | isaligned = 0; | |
698 | if (args.mp->m_sinoalign) { | |
699 | ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN)); | |
700 | args.alignment = args.mp->m_dalign; | |
701 | isaligned = 1; | |
75de2a91 | 702 | } else |
7a1df156 | 703 | args.alignment = xfs_ialloc_cluster_alignment(args.mp); |
3ccb8b5f GO |
704 | /* |
705 | * Need to figure out where to allocate the inode blocks. | |
706 | * Ideally they should be spaced out through the a.g. | |
707 | * For now, just allocate blocks up front. | |
708 | */ | |
709 | args.agbno = be32_to_cpu(agi->agi_root); | |
85c0b2ab | 710 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
3ccb8b5f GO |
711 | /* |
712 | * Allocate a fixed-size extent of inodes. | |
713 | */ | |
714 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
3ccb8b5f GO |
715 | args.prod = 1; |
716 | /* | |
717 | * Allow space for the inode btree to split. | |
718 | */ | |
0d87e656 | 719 | args.minleft = args.mp->m_in_maxlevels - 1; |
3ccb8b5f GO |
720 | if ((error = xfs_alloc_vextent(&args))) |
721 | return error; | |
722 | } | |
019ff2d5 | 723 | |
1da177e4 LT |
724 | /* |
725 | * If stripe alignment is turned on, then try again with cluster | |
726 | * alignment. | |
727 | */ | |
728 | if (isaligned && args.fsbno == NULLFSBLOCK) { | |
729 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
16259e7d | 730 | args.agbno = be32_to_cpu(agi->agi_root); |
85c0b2ab | 731 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
7a1df156 | 732 | args.alignment = xfs_ialloc_cluster_alignment(args.mp); |
1da177e4 LT |
733 | if ((error = xfs_alloc_vextent(&args))) |
734 | return error; | |
735 | } | |
736 | ||
56d1115c BF |
737 | /* |
738 | * Finally, try a sparse allocation if the filesystem supports it and | |
739 | * the sparse allocation length is smaller than a full chunk. | |
740 | */ | |
741 | if (xfs_sb_version_hassparseinodes(&args.mp->m_sb) && | |
742 | args.mp->m_ialloc_min_blks < args.mp->m_ialloc_blks && | |
743 | args.fsbno == NULLFSBLOCK) { | |
1cdadee1 | 744 | sparse_alloc: |
56d1115c BF |
745 | args.type = XFS_ALLOCTYPE_NEAR_BNO; |
746 | args.agbno = be32_to_cpu(agi->agi_root); | |
747 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); | |
748 | args.alignment = args.mp->m_sb.sb_spino_align; | |
749 | args.prod = 1; | |
750 | ||
751 | args.minlen = args.mp->m_ialloc_min_blks; | |
752 | args.maxlen = args.minlen; | |
753 | ||
754 | /* | |
755 | * The inode record will be aligned to full chunk size. We must | |
756 | * prevent sparse allocation from AG boundaries that result in | |
757 | * invalid inode records, such as records that start at agbno 0 | |
758 | * or extend beyond the AG. | |
759 | * | |
760 | * Set min agbno to the first aligned, non-zero agbno and max to | |
761 | * the last aligned agbno that is at least one full chunk from | |
762 | * the end of the AG. | |
763 | */ | |
764 | args.min_agbno = args.mp->m_sb.sb_inoalignmt; | |
765 | args.max_agbno = round_down(args.mp->m_sb.sb_agblocks, | |
766 | args.mp->m_sb.sb_inoalignmt) - | |
767 | args.mp->m_ialloc_blks; | |
768 | ||
769 | error = xfs_alloc_vextent(&args); | |
770 | if (error) | |
771 | return error; | |
772 | ||
773 | newlen = args.len << args.mp->m_sb.sb_inopblog; | |
46fc58da | 774 | ASSERT(newlen <= XFS_INODES_PER_CHUNK); |
56d1115c BF |
775 | allocmask = (1 << (newlen / XFS_INODES_PER_HOLEMASK_BIT)) - 1; |
776 | } | |
777 | ||
1da177e4 LT |
778 | if (args.fsbno == NULLFSBLOCK) { |
779 | *alloc = 0; | |
780 | return 0; | |
781 | } | |
782 | ASSERT(args.len == args.minlen); | |
1da177e4 | 783 | |
359346a9 | 784 | /* |
85c0b2ab DC |
785 | * Stamp and write the inode buffers. |
786 | * | |
359346a9 DC |
787 | * Seed the new inode cluster with a random generation number. This |
788 | * prevents short-term reuse of generation numbers if a chunk is | |
789 | * freed and then immediately reallocated. We use random numbers | |
790 | * rather than a linear progression to prevent the next generation | |
791 | * number from being easily guessable. | |
792 | */ | |
463958af BF |
793 | error = xfs_ialloc_inode_init(args.mp, tp, NULL, newlen, agno, |
794 | args.agbno, args.len, prandom_u32()); | |
d42f08f6 | 795 | |
2a30f36d CS |
796 | if (error) |
797 | return error; | |
85c0b2ab DC |
798 | /* |
799 | * Convert the results. | |
800 | */ | |
801 | newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0); | |
56d1115c BF |
802 | |
803 | if (xfs_inobt_issparse(~allocmask)) { | |
804 | /* | |
805 | * We've allocated a sparse chunk. Align the startino and mask. | |
806 | */ | |
807 | xfs_align_sparse_ino(args.mp, &newino, &allocmask); | |
808 | ||
809 | rec.ir_startino = newino; | |
810 | rec.ir_holemask = ~allocmask; | |
811 | rec.ir_count = newlen; | |
812 | rec.ir_freecount = newlen; | |
813 | rec.ir_free = XFS_INOBT_ALL_FREE; | |
814 | ||
815 | /* | |
816 | * Insert the sparse record into the inobt and allow for a merge | |
817 | * if necessary. If a merge does occur, rec is updated to the | |
818 | * merged record. | |
819 | */ | |
820 | error = xfs_inobt_insert_sprec(args.mp, tp, agbp, XFS_BTNUM_INO, | |
821 | &rec, true); | |
822 | if (error == -EFSCORRUPTED) { | |
823 | xfs_alert(args.mp, | |
824 | "invalid sparse inode record: ino 0x%llx holemask 0x%x count %u", | |
825 | XFS_AGINO_TO_INO(args.mp, agno, | |
826 | rec.ir_startino), | |
827 | rec.ir_holemask, rec.ir_count); | |
828 | xfs_force_shutdown(args.mp, SHUTDOWN_CORRUPT_INCORE); | |
829 | } | |
830 | if (error) | |
831 | return error; | |
832 | ||
833 | /* | |
834 | * We can't merge the part we've just allocated as for the inobt | |
835 | * due to finobt semantics. The original record may or may not | |
836 | * exist independent of whether physical inodes exist in this | |
837 | * sparse chunk. | |
838 | * | |
839 | * We must update the finobt record based on the inobt record. | |
840 | * rec contains the fully merged and up to date inobt record | |
841 | * from the previous call. Set merge false to replace any | |
842 | * existing record with this one. | |
843 | */ | |
844 | if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) { | |
845 | error = xfs_inobt_insert_sprec(args.mp, tp, agbp, | |
846 | XFS_BTNUM_FINO, &rec, | |
847 | false); | |
848 | if (error) | |
849 | return error; | |
850 | } | |
851 | } else { | |
852 | /* full chunk - insert new records to both btrees */ | |
853 | error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen, | |
854 | XFS_BTNUM_INO); | |
855 | if (error) | |
856 | return error; | |
857 | ||
858 | if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) { | |
859 | error = xfs_inobt_insert(args.mp, tp, agbp, newino, | |
860 | newlen, XFS_BTNUM_FINO); | |
861 | if (error) | |
862 | return error; | |
863 | } | |
864 | } | |
865 | ||
866 | /* | |
867 | * Update AGI counts and newino. | |
868 | */ | |
413d57c9 MS |
869 | be32_add_cpu(&agi->agi_count, newlen); |
870 | be32_add_cpu(&agi->agi_freecount, newlen); | |
44b56e0a DC |
871 | pag = xfs_perag_get(args.mp, agno); |
872 | pag->pagi_freecount += newlen; | |
873 | xfs_perag_put(pag); | |
16259e7d | 874 | agi->agi_newino = cpu_to_be32(newino); |
85c0b2ab | 875 | |
1da177e4 LT |
876 | /* |
877 | * Log allocation group header fields | |
878 | */ | |
879 | xfs_ialloc_log_agi(tp, agbp, | |
880 | XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO); | |
881 | /* | |
882 | * Modify/log superblock values for inode count and inode free count. | |
883 | */ | |
884 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen); | |
885 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen); | |
886 | *alloc = 1; | |
887 | return 0; | |
888 | } | |
889 | ||
b8f82a4a | 890 | STATIC xfs_agnumber_t |
1da177e4 LT |
891 | xfs_ialloc_next_ag( |
892 | xfs_mount_t *mp) | |
893 | { | |
894 | xfs_agnumber_t agno; | |
895 | ||
896 | spin_lock(&mp->m_agirotor_lock); | |
897 | agno = mp->m_agirotor; | |
8aea3ff4 | 898 | if (++mp->m_agirotor >= mp->m_maxagi) |
1da177e4 LT |
899 | mp->m_agirotor = 0; |
900 | spin_unlock(&mp->m_agirotor_lock); | |
901 | ||
902 | return agno; | |
903 | } | |
904 | ||
905 | /* | |
906 | * Select an allocation group to look for a free inode in, based on the parent | |
2f21ff1c | 907 | * inode and the mode. Return the allocation group buffer. |
1da177e4 | 908 | */ |
55d6af64 | 909 | STATIC xfs_agnumber_t |
1da177e4 LT |
910 | xfs_ialloc_ag_select( |
911 | xfs_trans_t *tp, /* transaction pointer */ | |
912 | xfs_ino_t parent, /* parent directory inode number */ | |
576b1d67 | 913 | umode_t mode, /* bits set to indicate file type */ |
1da177e4 LT |
914 | int okalloc) /* ok to allocate more space */ |
915 | { | |
1da177e4 LT |
916 | xfs_agnumber_t agcount; /* number of ag's in the filesystem */ |
917 | xfs_agnumber_t agno; /* current ag number */ | |
918 | int flags; /* alloc buffer locking flags */ | |
919 | xfs_extlen_t ineed; /* blocks needed for inode allocation */ | |
920 | xfs_extlen_t longest = 0; /* longest extent available */ | |
921 | xfs_mount_t *mp; /* mount point structure */ | |
922 | int needspace; /* file mode implies space allocated */ | |
923 | xfs_perag_t *pag; /* per allocation group data */ | |
924 | xfs_agnumber_t pagno; /* parent (starting) ag number */ | |
55d6af64 | 925 | int error; |
1da177e4 LT |
926 | |
927 | /* | |
928 | * Files of these types need at least one block if length > 0 | |
929 | * (and they won't fit in the inode, but that's hard to figure out). | |
930 | */ | |
931 | needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode); | |
932 | mp = tp->t_mountp; | |
933 | agcount = mp->m_maxagi; | |
934 | if (S_ISDIR(mode)) | |
935 | pagno = xfs_ialloc_next_ag(mp); | |
936 | else { | |
937 | pagno = XFS_INO_TO_AGNO(mp, parent); | |
938 | if (pagno >= agcount) | |
939 | pagno = 0; | |
940 | } | |
55d6af64 | 941 | |
1da177e4 | 942 | ASSERT(pagno < agcount); |
55d6af64 | 943 | |
1da177e4 LT |
944 | /* |
945 | * Loop through allocation groups, looking for one with a little | |
946 | * free space in it. Note we don't look for free inodes, exactly. | |
947 | * Instead, we include whether there is a need to allocate inodes | |
948 | * to mean that blocks must be allocated for them, | |
949 | * if none are currently free. | |
950 | */ | |
951 | agno = pagno; | |
952 | flags = XFS_ALLOC_FLAG_TRYLOCK; | |
1da177e4 | 953 | for (;;) { |
44b56e0a | 954 | pag = xfs_perag_get(mp, agno); |
55d6af64 CH |
955 | if (!pag->pagi_inodeok) { |
956 | xfs_ialloc_next_ag(mp); | |
957 | goto nextag; | |
958 | } | |
959 | ||
1da177e4 | 960 | if (!pag->pagi_init) { |
55d6af64 CH |
961 | error = xfs_ialloc_pagi_init(mp, tp, agno); |
962 | if (error) | |
1da177e4 | 963 | goto nextag; |
55d6af64 | 964 | } |
1da177e4 | 965 | |
55d6af64 CH |
966 | if (pag->pagi_freecount) { |
967 | xfs_perag_put(pag); | |
968 | return agno; | |
1da177e4 LT |
969 | } |
970 | ||
55d6af64 CH |
971 | if (!okalloc) |
972 | goto nextag; | |
973 | ||
974 | if (!pag->pagf_init) { | |
975 | error = xfs_alloc_pagf_init(mp, tp, agno, flags); | |
976 | if (error) | |
1da177e4 | 977 | goto nextag; |
1da177e4 | 978 | } |
55d6af64 CH |
979 | |
980 | /* | |
7a1df156 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. | |
55d6af64 | 995 | */ |
066a1884 | 996 | ineed = mp->m_ialloc_min_blks; |
7a1df156 DC |
997 | if (flags && ineed > 1) |
998 | ineed += xfs_ialloc_cluster_alignment(mp); | |
55d6af64 CH |
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; | |
1da177e4 | 1007 | } |
1da177e4 | 1008 | nextag: |
44b56e0a | 1009 | xfs_perag_put(pag); |
1da177e4 LT |
1010 | /* |
1011 | * No point in iterating over the rest, if we're shutting | |
1012 | * down. | |
1013 | */ | |
1c1c6ebc | 1014 | if (XFS_FORCED_SHUTDOWN(mp)) |
55d6af64 | 1015 | return NULLAGNUMBER; |
1da177e4 LT |
1016 | agno++; |
1017 | if (agno >= agcount) | |
1018 | agno = 0; | |
1019 | if (agno == pagno) { | |
1c1c6ebc | 1020 | if (flags == 0) |
55d6af64 | 1021 | return NULLAGNUMBER; |
1da177e4 LT |
1022 | flags = 0; |
1023 | } | |
1024 | } | |
1025 | } | |
1026 | ||
4254b0bb CH |
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; | |
5fb5aeee | 1052 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
4254b0bb CH |
1053 | } |
1054 | ||
1055 | return 0; | |
1056 | } | |
1057 | ||
bd169565 DC |
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, | |
43df2ee6 | 1063 | int *done) |
bd169565 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; | |
5fb5aeee | 1076 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
bd169565 DC |
1077 | } |
1078 | ||
1079 | return 0; | |
1080 | } | |
0b48db80 | 1081 | |
d4cc540b | 1082 | /* |
26dd5217 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. | |
d4cc540b BF |
1086 | */ |
1087 | STATIC int | |
1088 | xfs_inobt_first_free_inode( | |
1089 | struct xfs_inobt_rec_incore *rec) | |
1090 | { | |
26dd5217 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); | |
d4cc540b BF |
1101 | } |
1102 | ||
1da177e4 | 1103 | /* |
6dd8638e | 1104 | * Allocate an inode using the inobt-only algorithm. |
1da177e4 | 1105 | */ |
f2ecc5e4 | 1106 | STATIC int |
6dd8638e | 1107 | xfs_dialloc_ag_inobt( |
f2ecc5e4 CH |
1108 | struct xfs_trans *tp, |
1109 | struct xfs_buf *agbp, | |
1110 | xfs_ino_t parent, | |
1111 | xfs_ino_t *inop) | |
1da177e4 | 1112 | { |
f2ecc5e4 CH |
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; | |
1da177e4 | 1125 | |
44b56e0a | 1126 | pag = xfs_perag_get(mp, agno); |
bd169565 | 1127 | |
4bb61069 CH |
1128 | ASSERT(pag->pagi_init); |
1129 | ASSERT(pag->pagi_inodeok); | |
1130 | ASSERT(pag->pagi_freecount > 0); | |
1131 | ||
bd169565 | 1132 | restart_pagno: |
57bd3dbe | 1133 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
1da177e4 LT |
1134 | /* |
1135 | * If pagino is 0 (this is the root inode allocation) use newino. | |
1136 | * This must work because we've just allocated some. | |
1137 | */ | |
1138 | if (!pagino) | |
16259e7d | 1139 | pagino = be32_to_cpu(agi->agi_newino); |
1da177e4 | 1140 | |
0b48db80 DC |
1141 | error = xfs_check_agi_freecount(cur, agi); |
1142 | if (error) | |
1143 | goto error0; | |
1da177e4 | 1144 | |
1da177e4 | 1145 | /* |
4254b0bb | 1146 | * If in the same AG as the parent, try to get near the parent. |
1da177e4 LT |
1147 | */ |
1148 | if (pagno == agno) { | |
4254b0bb CH |
1149 | int doneleft; /* done, to the left */ |
1150 | int doneright; /* done, to the right */ | |
bd169565 | 1151 | int searchdistance = 10; |
4254b0bb | 1152 | |
21875505 | 1153 | error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i); |
4254b0bb | 1154 | if (error) |
1da177e4 | 1155 | goto error0; |
c29aad41 | 1156 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
4254b0bb CH |
1157 | |
1158 | error = xfs_inobt_get_rec(cur, &rec, &j); | |
1159 | if (error) | |
1160 | goto error0; | |
c29aad41 | 1161 | XFS_WANT_CORRUPTED_GOTO(mp, j == 1, error0); |
4254b0bb CH |
1162 | |
1163 | if (rec.ir_freecount > 0) { | |
1da177e4 LT |
1164 | /* |
1165 | * Found a free inode in the same chunk | |
4254b0bb | 1166 | * as the parent, done. |
1da177e4 | 1167 | */ |
4254b0bb | 1168 | goto alloc_inode; |
1da177e4 | 1169 | } |
4254b0bb CH |
1170 | |
1171 | ||
1da177e4 | 1172 | /* |
4254b0bb | 1173 | * In the same AG as parent, but parent's chunk is full. |
1da177e4 | 1174 | */ |
1da177e4 | 1175 | |
4254b0bb CH |
1176 | /* duplicate the cursor, search left & right simultaneously */ |
1177 | error = xfs_btree_dup_cursor(cur, &tcur); | |
1178 | if (error) | |
1179 | goto error0; | |
1180 | ||
bd169565 DC |
1181 | /* |
1182 | * Skip to last blocks looked up if same parent inode. | |
1183 | */ | |
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, | |
43df2ee6 | 1189 | &trec, &doneleft); |
bd169565 DC |
1190 | if (error) |
1191 | goto error1; | |
4254b0bb | 1192 | |
bd169565 | 1193 | error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec, |
43df2ee6 | 1194 | &rec, &doneright); |
bd169565 DC |
1195 | if (error) |
1196 | goto error1; | |
1197 | } else { | |
1198 | /* search left with tcur, back up 1 record */ | |
1199 | error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1); | |
1200 | if (error) | |
1201 | goto error1; | |
1202 | ||
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 | } | |
4254b0bb CH |
1208 | |
1209 | /* | |
1210 | * Loop until we find an inode chunk with a free inode. | |
1211 | */ | |
1212 | while (!doneleft || !doneright) { | |
1213 | int useleft; /* using left inode chunk this time */ | |
1214 | ||
bd169565 DC |
1215 | if (!--searchdistance) { |
1216 | /* | |
1217 | * Not in range - save last search | |
1218 | * location and allocate a new inode | |
1219 | */ | |
3b826386 | 1220 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); |
bd169565 DC |
1221 | pag->pagl_leftrec = trec.ir_startino; |
1222 | pag->pagl_rightrec = rec.ir_startino; | |
1223 | pag->pagl_pagino = pagino; | |
1224 | goto newino; | |
1225 | } | |
1226 | ||
4254b0bb CH |
1227 | /* figure out the closer block if both are valid. */ |
1228 | if (!doneleft && !doneright) { | |
1229 | useleft = pagino - | |
1230 | (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) < | |
1231 | rec.ir_startino - pagino; | |
1232 | } else { | |
1233 | useleft = !doneleft; | |
1da177e4 | 1234 | } |
4254b0bb CH |
1235 | |
1236 | /* free inodes to the left? */ | |
1237 | if (useleft && trec.ir_freecount) { | |
1238 | rec = trec; | |
1239 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
1240 | cur = tcur; | |
bd169565 DC |
1241 | |
1242 | pag->pagl_leftrec = trec.ir_startino; | |
1243 | pag->pagl_rightrec = rec.ir_startino; | |
1244 | pag->pagl_pagino = pagino; | |
4254b0bb | 1245 | goto alloc_inode; |
1da177e4 | 1246 | } |
1da177e4 | 1247 | |
4254b0bb CH |
1248 | /* free inodes to the right? */ |
1249 | if (!useleft && rec.ir_freecount) { | |
1250 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
bd169565 DC |
1251 | |
1252 | pag->pagl_leftrec = trec.ir_startino; | |
1253 | pag->pagl_rightrec = rec.ir_startino; | |
1254 | pag->pagl_pagino = pagino; | |
4254b0bb | 1255 | goto alloc_inode; |
1da177e4 | 1256 | } |
4254b0bb CH |
1257 | |
1258 | /* get next record to check */ | |
1259 | if (useleft) { | |
1260 | error = xfs_ialloc_next_rec(tcur, &trec, | |
1261 | &doneleft, 1); | |
1262 | } else { | |
1263 | error = xfs_ialloc_next_rec(cur, &rec, | |
1264 | &doneright, 0); | |
1265 | } | |
1266 | if (error) | |
1267 | goto error1; | |
1da177e4 | 1268 | } |
bd169565 DC |
1269 | |
1270 | /* | |
1271 | * We've reached the end of the btree. because | |
1272 | * we are only searching a small chunk of the | |
1273 | * btree each search, there is obviously free | |
1274 | * inodes closer to the parent inode than we | |
1275 | * are now. restart the search again. | |
1276 | */ | |
1277 | pag->pagl_pagino = NULLAGINO; | |
1278 | pag->pagl_leftrec = NULLAGINO; | |
1279 | pag->pagl_rightrec = NULLAGINO; | |
1280 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
1281 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
1282 | goto restart_pagno; | |
1da177e4 | 1283 | } |
4254b0bb | 1284 | |
1da177e4 | 1285 | /* |
4254b0bb | 1286 | * In a different AG from the parent. |
1da177e4 LT |
1287 | * See if the most recently allocated block has any free. |
1288 | */ | |
bd169565 | 1289 | newino: |
69ef921b | 1290 | if (agi->agi_newino != cpu_to_be32(NULLAGINO)) { |
21875505 CH |
1291 | error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino), |
1292 | XFS_LOOKUP_EQ, &i); | |
4254b0bb | 1293 | if (error) |
1da177e4 | 1294 | goto error0; |
4254b0bb CH |
1295 | |
1296 | if (i == 1) { | |
1297 | error = xfs_inobt_get_rec(cur, &rec, &j); | |
1298 | if (error) | |
1299 | goto error0; | |
1300 | ||
1301 | if (j == 1 && rec.ir_freecount > 0) { | |
1302 | /* | |
1303 | * The last chunk allocated in the group | |
1304 | * still has a free inode. | |
1305 | */ | |
1306 | goto alloc_inode; | |
1307 | } | |
1da177e4 | 1308 | } |
bd169565 | 1309 | } |
4254b0bb | 1310 | |
bd169565 DC |
1311 | /* |
1312 | * None left in the last group, search the whole AG | |
1313 | */ | |
1314 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); | |
1315 | if (error) | |
1316 | goto error0; | |
c29aad41 | 1317 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
bd169565 DC |
1318 | |
1319 | for (;;) { | |
1320 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
1321 | if (error) | |
1322 | goto error0; | |
c29aad41 | 1323 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
bd169565 DC |
1324 | if (rec.ir_freecount > 0) |
1325 | break; | |
1326 | error = xfs_btree_increment(cur, 0, &i); | |
4254b0bb CH |
1327 | if (error) |
1328 | goto error0; | |
c29aad41 | 1329 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 | 1330 | } |
4254b0bb CH |
1331 | |
1332 | alloc_inode: | |
d4cc540b | 1333 | offset = xfs_inobt_first_free_inode(&rec); |
1da177e4 LT |
1334 | ASSERT(offset >= 0); |
1335 | ASSERT(offset < XFS_INODES_PER_CHUNK); | |
1336 | ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) % | |
1337 | XFS_INODES_PER_CHUNK) == 0); | |
1338 | ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset); | |
0d87e656 | 1339 | rec.ir_free &= ~XFS_INOBT_MASK(offset); |
1da177e4 | 1340 | rec.ir_freecount--; |
afabc24a CH |
1341 | error = xfs_inobt_update(cur, &rec); |
1342 | if (error) | |
1da177e4 | 1343 | goto error0; |
413d57c9 | 1344 | be32_add_cpu(&agi->agi_freecount, -1); |
1da177e4 | 1345 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); |
44b56e0a | 1346 | pag->pagi_freecount--; |
1da177e4 | 1347 | |
0b48db80 DC |
1348 | error = xfs_check_agi_freecount(cur, agi); |
1349 | if (error) | |
1350 | goto error0; | |
1351 | ||
1da177e4 LT |
1352 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
1353 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1); | |
44b56e0a | 1354 | xfs_perag_put(pag); |
1da177e4 LT |
1355 | *inop = ino; |
1356 | return 0; | |
1357 | error1: | |
1358 | xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR); | |
1359 | error0: | |
1360 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
44b56e0a | 1361 | xfs_perag_put(pag); |
1da177e4 LT |
1362 | return error; |
1363 | } | |
1364 | ||
6dd8638e BF |
1365 | /* |
1366 | * Use the free inode btree to allocate an inode based on distance from the | |
1367 | * parent. Note that the provided cursor may be deleted and replaced. | |
1368 | */ | |
1369 | STATIC int | |
1370 | xfs_dialloc_ag_finobt_near( | |
1371 | xfs_agino_t pagino, | |
1372 | struct xfs_btree_cur **ocur, | |
1373 | struct xfs_inobt_rec_incore *rec) | |
1374 | { | |
1375 | struct xfs_btree_cur *lcur = *ocur; /* left search cursor */ | |
1376 | struct xfs_btree_cur *rcur; /* right search cursor */ | |
1377 | struct xfs_inobt_rec_incore rrec; | |
1378 | int error; | |
1379 | int i, j; | |
1380 | ||
1381 | error = xfs_inobt_lookup(lcur, pagino, XFS_LOOKUP_LE, &i); | |
1382 | if (error) | |
1383 | return error; | |
1384 | ||
1385 | if (i == 1) { | |
1386 | error = xfs_inobt_get_rec(lcur, rec, &i); | |
1387 | if (error) | |
1388 | return error; | |
5fb5aeee | 1389 | XFS_WANT_CORRUPTED_RETURN(lcur->bc_mp, i == 1); |
6dd8638e BF |
1390 | |
1391 | /* | |
1392 | * See if we've landed in the parent inode record. The finobt | |
1393 | * only tracks chunks with at least one free inode, so record | |
1394 | * existence is enough. | |
1395 | */ | |
1396 | if (pagino >= rec->ir_startino && | |
1397 | pagino < (rec->ir_startino + XFS_INODES_PER_CHUNK)) | |
1398 | return 0; | |
1399 | } | |
1400 | ||
1401 | error = xfs_btree_dup_cursor(lcur, &rcur); | |
1402 | if (error) | |
1403 | return error; | |
1404 | ||
1405 | error = xfs_inobt_lookup(rcur, pagino, XFS_LOOKUP_GE, &j); | |
1406 | if (error) | |
1407 | goto error_rcur; | |
1408 | if (j == 1) { | |
1409 | error = xfs_inobt_get_rec(rcur, &rrec, &j); | |
1410 | if (error) | |
1411 | goto error_rcur; | |
c29aad41 | 1412 | XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, j == 1, error_rcur); |
6dd8638e BF |
1413 | } |
1414 | ||
c29aad41 | 1415 | XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, i == 1 || j == 1, error_rcur); |
6dd8638e BF |
1416 | if (i == 1 && j == 1) { |
1417 | /* | |
1418 | * Both the left and right records are valid. Choose the closer | |
1419 | * inode chunk to the target. | |
1420 | */ | |
1421 | if ((pagino - rec->ir_startino + XFS_INODES_PER_CHUNK - 1) > | |
1422 | (rrec.ir_startino - pagino)) { | |
1423 | *rec = rrec; | |
1424 | xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR); | |
1425 | *ocur = rcur; | |
1426 | } else { | |
1427 | xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR); | |
1428 | } | |
1429 | } else if (j == 1) { | |
1430 | /* only the right record is valid */ | |
1431 | *rec = rrec; | |
1432 | xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR); | |
1433 | *ocur = rcur; | |
1434 | } else if (i == 1) { | |
1435 | /* only the left record is valid */ | |
1436 | xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR); | |
1437 | } | |
1438 | ||
1439 | return 0; | |
1440 | ||
1441 | error_rcur: | |
1442 | xfs_btree_del_cursor(rcur, XFS_BTREE_ERROR); | |
1443 | return error; | |
1444 | } | |
1445 | ||
1446 | /* | |
1447 | * Use the free inode btree to find a free inode based on a newino hint. If | |
1448 | * the hint is NULL, find the first free inode in the AG. | |
1449 | */ | |
1450 | STATIC int | |
1451 | xfs_dialloc_ag_finobt_newino( | |
1452 | struct xfs_agi *agi, | |
1453 | struct xfs_btree_cur *cur, | |
1454 | struct xfs_inobt_rec_incore *rec) | |
1455 | { | |
1456 | int error; | |
1457 | int i; | |
1458 | ||
1459 | if (agi->agi_newino != cpu_to_be32(NULLAGINO)) { | |
e68ed775 DC |
1460 | error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino), |
1461 | XFS_LOOKUP_EQ, &i); | |
6dd8638e BF |
1462 | if (error) |
1463 | return error; | |
1464 | if (i == 1) { | |
1465 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1466 | if (error) | |
1467 | return error; | |
5fb5aeee | 1468 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
6dd8638e BF |
1469 | return 0; |
1470 | } | |
1471 | } | |
1472 | ||
1473 | /* | |
1474 | * Find the first inode available in the AG. | |
1475 | */ | |
1476 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); | |
1477 | if (error) | |
1478 | return error; | |
5fb5aeee | 1479 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
6dd8638e BF |
1480 | |
1481 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1482 | if (error) | |
1483 | return error; | |
5fb5aeee | 1484 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
6dd8638e BF |
1485 | |
1486 | return 0; | |
1487 | } | |
1488 | ||
1489 | /* | |
1490 | * Update the inobt based on a modification made to the finobt. Also ensure that | |
1491 | * the records from both trees are equivalent post-modification. | |
1492 | */ | |
1493 | STATIC int | |
1494 | xfs_dialloc_ag_update_inobt( | |
1495 | struct xfs_btree_cur *cur, /* inobt cursor */ | |
1496 | struct xfs_inobt_rec_incore *frec, /* finobt record */ | |
1497 | int offset) /* inode offset */ | |
1498 | { | |
1499 | struct xfs_inobt_rec_incore rec; | |
1500 | int error; | |
1501 | int i; | |
1502 | ||
1503 | error = xfs_inobt_lookup(cur, frec->ir_startino, XFS_LOOKUP_EQ, &i); | |
1504 | if (error) | |
1505 | return error; | |
5fb5aeee | 1506 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
6dd8638e BF |
1507 | |
1508 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
1509 | if (error) | |
1510 | return error; | |
5fb5aeee | 1511 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
6dd8638e BF |
1512 | ASSERT((XFS_AGINO_TO_OFFSET(cur->bc_mp, rec.ir_startino) % |
1513 | XFS_INODES_PER_CHUNK) == 0); | |
1514 | ||
1515 | rec.ir_free &= ~XFS_INOBT_MASK(offset); | |
1516 | rec.ir_freecount--; | |
1517 | ||
5fb5aeee | 1518 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, (rec.ir_free == frec->ir_free) && |
6dd8638e BF |
1519 | (rec.ir_freecount == frec->ir_freecount)); |
1520 | ||
b72091f2 | 1521 | return xfs_inobt_update(cur, &rec); |
6dd8638e BF |
1522 | } |
1523 | ||
1524 | /* | |
1525 | * Allocate an inode using the free inode btree, if available. Otherwise, fall | |
1526 | * back to the inobt search algorithm. | |
1527 | * | |
1528 | * The caller selected an AG for us, and made sure that free inodes are | |
1529 | * available. | |
1530 | */ | |
1531 | STATIC int | |
1532 | xfs_dialloc_ag( | |
1533 | struct xfs_trans *tp, | |
1534 | struct xfs_buf *agbp, | |
1535 | xfs_ino_t parent, | |
1536 | xfs_ino_t *inop) | |
1537 | { | |
1538 | struct xfs_mount *mp = tp->t_mountp; | |
1539 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
1540 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
1541 | xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent); | |
1542 | xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent); | |
1543 | struct xfs_perag *pag; | |
1544 | struct xfs_btree_cur *cur; /* finobt cursor */ | |
1545 | struct xfs_btree_cur *icur; /* inobt cursor */ | |
1546 | struct xfs_inobt_rec_incore rec; | |
1547 | xfs_ino_t ino; | |
1548 | int error; | |
1549 | int offset; | |
1550 | int i; | |
1551 | ||
1552 | if (!xfs_sb_version_hasfinobt(&mp->m_sb)) | |
1553 | return xfs_dialloc_ag_inobt(tp, agbp, parent, inop); | |
1554 | ||
1555 | pag = xfs_perag_get(mp, agno); | |
1556 | ||
1557 | /* | |
1558 | * If pagino is 0 (this is the root inode allocation) use newino. | |
1559 | * This must work because we've just allocated some. | |
1560 | */ | |
1561 | if (!pagino) | |
1562 | pagino = be32_to_cpu(agi->agi_newino); | |
1563 | ||
1564 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO); | |
1565 | ||
1566 | error = xfs_check_agi_freecount(cur, agi); | |
1567 | if (error) | |
1568 | goto error_cur; | |
1569 | ||
1570 | /* | |
1571 | * The search algorithm depends on whether we're in the same AG as the | |
1572 | * parent. If so, find the closest available inode to the parent. If | |
1573 | * not, consider the agi hint or find the first free inode in the AG. | |
1574 | */ | |
1575 | if (agno == pagno) | |
1576 | error = xfs_dialloc_ag_finobt_near(pagino, &cur, &rec); | |
1577 | else | |
1578 | error = xfs_dialloc_ag_finobt_newino(agi, cur, &rec); | |
1579 | if (error) | |
1580 | goto error_cur; | |
1581 | ||
d4cc540b | 1582 | offset = xfs_inobt_first_free_inode(&rec); |
6dd8638e BF |
1583 | ASSERT(offset >= 0); |
1584 | ASSERT(offset < XFS_INODES_PER_CHUNK); | |
1585 | ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) % | |
1586 | XFS_INODES_PER_CHUNK) == 0); | |
1587 | ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset); | |
1588 | ||
1589 | /* | |
1590 | * Modify or remove the finobt record. | |
1591 | */ | |
1592 | rec.ir_free &= ~XFS_INOBT_MASK(offset); | |
1593 | rec.ir_freecount--; | |
1594 | if (rec.ir_freecount) | |
1595 | error = xfs_inobt_update(cur, &rec); | |
1596 | else | |
1597 | error = xfs_btree_delete(cur, &i); | |
1598 | if (error) | |
1599 | goto error_cur; | |
1600 | ||
1601 | /* | |
1602 | * The finobt has now been updated appropriately. We haven't updated the | |
1603 | * agi and superblock yet, so we can create an inobt cursor and validate | |
1604 | * the original freecount. If all is well, make the equivalent update to | |
1605 | * the inobt using the finobt record and offset information. | |
1606 | */ | |
1607 | icur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); | |
1608 | ||
1609 | error = xfs_check_agi_freecount(icur, agi); | |
1610 | if (error) | |
1611 | goto error_icur; | |
1612 | ||
1613 | error = xfs_dialloc_ag_update_inobt(icur, &rec, offset); | |
1614 | if (error) | |
1615 | goto error_icur; | |
1616 | ||
1617 | /* | |
1618 | * Both trees have now been updated. We must update the perag and | |
1619 | * superblock before we can check the freecount for each btree. | |
1620 | */ | |
1621 | be32_add_cpu(&agi->agi_freecount, -1); | |
1622 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); | |
1623 | pag->pagi_freecount--; | |
1624 | ||
1625 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1); | |
1626 | ||
1627 | error = xfs_check_agi_freecount(icur, agi); | |
1628 | if (error) | |
1629 | goto error_icur; | |
1630 | error = xfs_check_agi_freecount(cur, agi); | |
1631 | if (error) | |
1632 | goto error_icur; | |
1633 | ||
1634 | xfs_btree_del_cursor(icur, XFS_BTREE_NOERROR); | |
1635 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
1636 | xfs_perag_put(pag); | |
1637 | *inop = ino; | |
1638 | return 0; | |
1639 | ||
1640 | error_icur: | |
1641 | xfs_btree_del_cursor(icur, XFS_BTREE_ERROR); | |
1642 | error_cur: | |
1643 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
1644 | xfs_perag_put(pag); | |
1645 | return error; | |
1646 | } | |
1647 | ||
f2ecc5e4 CH |
1648 | /* |
1649 | * Allocate an inode on disk. | |
1650 | * | |
1651 | * Mode is used to tell whether the new inode will need space, and whether it | |
1652 | * is a directory. | |
1653 | * | |
1654 | * This function is designed to be called twice if it has to do an allocation | |
1655 | * to make more free inodes. On the first call, *IO_agbp should be set to NULL. | |
1656 | * If an inode is available without having to performn an allocation, an inode | |
cd856db6 CM |
1657 | * number is returned. In this case, *IO_agbp is set to NULL. If an allocation |
1658 | * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp. | |
1659 | * The caller should then commit the current transaction, allocate a | |
f2ecc5e4 CH |
1660 | * new transaction, and call xfs_dialloc() again, passing in the previous value |
1661 | * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI | |
1662 | * buffer is locked across the two calls, the second call is guaranteed to have | |
1663 | * a free inode available. | |
1664 | * | |
1665 | * Once we successfully pick an inode its number is returned and the on-disk | |
1666 | * data structures are updated. The inode itself is not read in, since doing so | |
1667 | * would break ordering constraints with xfs_reclaim. | |
1668 | */ | |
1669 | int | |
1670 | xfs_dialloc( | |
1671 | struct xfs_trans *tp, | |
1672 | xfs_ino_t parent, | |
1673 | umode_t mode, | |
1674 | int okalloc, | |
1675 | struct xfs_buf **IO_agbp, | |
f2ecc5e4 CH |
1676 | xfs_ino_t *inop) |
1677 | { | |
55d6af64 | 1678 | struct xfs_mount *mp = tp->t_mountp; |
f2ecc5e4 CH |
1679 | struct xfs_buf *agbp; |
1680 | xfs_agnumber_t agno; | |
f2ecc5e4 CH |
1681 | int error; |
1682 | int ialloced; | |
1683 | int noroom = 0; | |
be60fe54 | 1684 | xfs_agnumber_t start_agno; |
f2ecc5e4 CH |
1685 | struct xfs_perag *pag; |
1686 | ||
4bb61069 | 1687 | if (*IO_agbp) { |
f2ecc5e4 | 1688 | /* |
4bb61069 CH |
1689 | * If the caller passes in a pointer to the AGI buffer, |
1690 | * continue where we left off before. In this case, we | |
f2ecc5e4 CH |
1691 | * know that the allocation group has free inodes. |
1692 | */ | |
1693 | agbp = *IO_agbp; | |
4bb61069 | 1694 | goto out_alloc; |
f2ecc5e4 | 1695 | } |
4bb61069 CH |
1696 | |
1697 | /* | |
1698 | * We do not have an agbp, so select an initial allocation | |
1699 | * group for inode allocation. | |
1700 | */ | |
be60fe54 CH |
1701 | start_agno = xfs_ialloc_ag_select(tp, parent, mode, okalloc); |
1702 | if (start_agno == NULLAGNUMBER) { | |
4bb61069 CH |
1703 | *inop = NULLFSINO; |
1704 | return 0; | |
1705 | } | |
55d6af64 | 1706 | |
f2ecc5e4 CH |
1707 | /* |
1708 | * If we have already hit the ceiling of inode blocks then clear | |
1709 | * okalloc so we scan all available agi structures for a free | |
1710 | * inode. | |
74f9ce1c GW |
1711 | * |
1712 | * Read rough value of mp->m_icount by percpu_counter_read_positive, | |
1713 | * which will sacrifice the preciseness but improve the performance. | |
f2ecc5e4 | 1714 | */ |
f2ecc5e4 | 1715 | if (mp->m_maxicount && |
74f9ce1c GW |
1716 | percpu_counter_read_positive(&mp->m_icount) + mp->m_ialloc_inos |
1717 | > mp->m_maxicount) { | |
f2ecc5e4 CH |
1718 | noroom = 1; |
1719 | okalloc = 0; | |
1720 | } | |
1721 | ||
1722 | /* | |
1723 | * Loop until we find an allocation group that either has free inodes | |
1724 | * or in which we can allocate some inodes. Iterate through the | |
1725 | * allocation groups upward, wrapping at the end. | |
1726 | */ | |
be60fe54 CH |
1727 | agno = start_agno; |
1728 | for (;;) { | |
1729 | pag = xfs_perag_get(mp, agno); | |
1730 | if (!pag->pagi_inodeok) { | |
1731 | xfs_ialloc_next_ag(mp); | |
1732 | goto nextag; | |
1733 | } | |
1734 | ||
1735 | if (!pag->pagi_init) { | |
1736 | error = xfs_ialloc_pagi_init(mp, tp, agno); | |
1737 | if (error) | |
1738 | goto out_error; | |
f2ecc5e4 | 1739 | } |
be60fe54 | 1740 | |
f2ecc5e4 | 1741 | /* |
be60fe54 | 1742 | * Do a first racy fast path check if this AG is usable. |
f2ecc5e4 | 1743 | */ |
be60fe54 CH |
1744 | if (!pag->pagi_freecount && !okalloc) |
1745 | goto nextag; | |
1746 | ||
c4982110 CH |
1747 | /* |
1748 | * Then read in the AGI buffer and recheck with the AGI buffer | |
1749 | * lock held. | |
1750 | */ | |
be60fe54 CH |
1751 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); |
1752 | if (error) | |
1753 | goto out_error; | |
1754 | ||
be60fe54 CH |
1755 | if (pag->pagi_freecount) { |
1756 | xfs_perag_put(pag); | |
1757 | goto out_alloc; | |
1758 | } | |
1759 | ||
c4982110 CH |
1760 | if (!okalloc) |
1761 | goto nextag_relse_buffer; | |
1762 | ||
be60fe54 CH |
1763 | |
1764 | error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced); | |
1765 | if (error) { | |
1766 | xfs_trans_brelse(tp, agbp); | |
1767 | ||
2451337d | 1768 | if (error != -ENOSPC) |
be60fe54 CH |
1769 | goto out_error; |
1770 | ||
1771 | xfs_perag_put(pag); | |
f2ecc5e4 | 1772 | *inop = NULLFSINO; |
be60fe54 | 1773 | return 0; |
f2ecc5e4 | 1774 | } |
be60fe54 CH |
1775 | |
1776 | if (ialloced) { | |
1777 | /* | |
1778 | * We successfully allocated some inodes, return | |
1779 | * the current context to the caller so that it | |
1780 | * can commit the current transaction and call | |
1781 | * us again where we left off. | |
1782 | */ | |
1783 | ASSERT(pag->pagi_freecount > 0); | |
f2ecc5e4 | 1784 | xfs_perag_put(pag); |
be60fe54 CH |
1785 | |
1786 | *IO_agbp = agbp; | |
1787 | *inop = NULLFSINO; | |
1788 | return 0; | |
f2ecc5e4 | 1789 | } |
be60fe54 | 1790 | |
c4982110 CH |
1791 | nextag_relse_buffer: |
1792 | xfs_trans_brelse(tp, agbp); | |
be60fe54 | 1793 | nextag: |
f2ecc5e4 | 1794 | xfs_perag_put(pag); |
be60fe54 CH |
1795 | if (++agno == mp->m_sb.sb_agcount) |
1796 | agno = 0; | |
1797 | if (agno == start_agno) { | |
1798 | *inop = NULLFSINO; | |
2451337d | 1799 | return noroom ? -ENOSPC : 0; |
be60fe54 | 1800 | } |
f2ecc5e4 CH |
1801 | } |
1802 | ||
4bb61069 | 1803 | out_alloc: |
f2ecc5e4 CH |
1804 | *IO_agbp = NULL; |
1805 | return xfs_dialloc_ag(tp, agbp, parent, inop); | |
be60fe54 CH |
1806 | out_error: |
1807 | xfs_perag_put(pag); | |
b474c7ae | 1808 | return error; |
f2ecc5e4 CH |
1809 | } |
1810 | ||
10ae3dc7 BF |
1811 | /* |
1812 | * Free the blocks of an inode chunk. We must consider that the inode chunk | |
1813 | * might be sparse and only free the regions that are allocated as part of the | |
1814 | * chunk. | |
1815 | */ | |
1816 | STATIC void | |
1817 | xfs_difree_inode_chunk( | |
1818 | struct xfs_mount *mp, | |
1819 | xfs_agnumber_t agno, | |
1820 | struct xfs_inobt_rec_incore *rec, | |
310a75a3 | 1821 | struct xfs_defer_ops *flist) |
10ae3dc7 BF |
1822 | { |
1823 | xfs_agblock_t sagbno = XFS_AGINO_TO_AGBNO(mp, rec->ir_startino); | |
1824 | int startidx, endidx; | |
1825 | int nextbit; | |
1826 | xfs_agblock_t agbno; | |
1827 | int contigblk; | |
1828 | DECLARE_BITMAP(holemask, XFS_INOBT_HOLEMASK_BITS); | |
1829 | ||
1830 | if (!xfs_inobt_issparse(rec->ir_holemask)) { | |
1831 | /* not sparse, calculate extent info directly */ | |
59bad075 DW |
1832 | xfs_bmap_add_free(mp, flist, XFS_AGB_TO_FSB(mp, agno, sagbno), |
1833 | mp->m_ialloc_blks); | |
10ae3dc7 BF |
1834 | return; |
1835 | } | |
1836 | ||
1837 | /* holemask is only 16-bits (fits in an unsigned long) */ | |
1838 | ASSERT(sizeof(rec->ir_holemask) <= sizeof(holemask[0])); | |
1839 | holemask[0] = rec->ir_holemask; | |
1840 | ||
1841 | /* | |
1842 | * Find contiguous ranges of zeroes (i.e., allocated regions) in the | |
1843 | * holemask and convert the start/end index of each range to an extent. | |
1844 | * We start with the start and end index both pointing at the first 0 in | |
1845 | * the mask. | |
1846 | */ | |
1847 | startidx = endidx = find_first_zero_bit(holemask, | |
1848 | XFS_INOBT_HOLEMASK_BITS); | |
1849 | nextbit = startidx + 1; | |
1850 | while (startidx < XFS_INOBT_HOLEMASK_BITS) { | |
1851 | nextbit = find_next_zero_bit(holemask, XFS_INOBT_HOLEMASK_BITS, | |
1852 | nextbit); | |
1853 | /* | |
1854 | * If the next zero bit is contiguous, update the end index of | |
1855 | * the current range and continue. | |
1856 | */ | |
1857 | if (nextbit != XFS_INOBT_HOLEMASK_BITS && | |
1858 | nextbit == endidx + 1) { | |
1859 | endidx = nextbit; | |
1860 | goto next; | |
1861 | } | |
1862 | ||
1863 | /* | |
1864 | * nextbit is not contiguous with the current end index. Convert | |
1865 | * the current start/end to an extent and add it to the free | |
1866 | * list. | |
1867 | */ | |
1868 | agbno = sagbno + (startidx * XFS_INODES_PER_HOLEMASK_BIT) / | |
1869 | mp->m_sb.sb_inopblock; | |
1870 | contigblk = ((endidx - startidx + 1) * | |
1871 | XFS_INODES_PER_HOLEMASK_BIT) / | |
1872 | mp->m_sb.sb_inopblock; | |
1873 | ||
1874 | ASSERT(agbno % mp->m_sb.sb_spino_align == 0); | |
1875 | ASSERT(contigblk % mp->m_sb.sb_spino_align == 0); | |
59bad075 DW |
1876 | xfs_bmap_add_free(mp, flist, XFS_AGB_TO_FSB(mp, agno, agbno), |
1877 | contigblk); | |
10ae3dc7 BF |
1878 | |
1879 | /* reset range to current bit and carry on... */ | |
1880 | startidx = endidx = nextbit; | |
1881 | ||
1882 | next: | |
1883 | nextbit++; | |
1884 | } | |
1885 | } | |
1886 | ||
2b64ee5c BF |
1887 | STATIC int |
1888 | xfs_difree_inobt( | |
1889 | struct xfs_mount *mp, | |
1890 | struct xfs_trans *tp, | |
1891 | struct xfs_buf *agbp, | |
1892 | xfs_agino_t agino, | |
310a75a3 | 1893 | struct xfs_defer_ops *flist, |
09b56604 | 1894 | struct xfs_icluster *xic, |
2b64ee5c | 1895 | struct xfs_inobt_rec_incore *orec) |
1da177e4 | 1896 | { |
2b64ee5c BF |
1897 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); |
1898 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
1899 | struct xfs_perag *pag; | |
1900 | struct xfs_btree_cur *cur; | |
1901 | struct xfs_inobt_rec_incore rec; | |
1902 | int ilen; | |
1903 | int error; | |
1904 | int i; | |
1905 | int off; | |
1da177e4 | 1906 | |
69ef921b | 1907 | ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC)); |
2b64ee5c BF |
1908 | ASSERT(XFS_AGINO_TO_AGBNO(mp, agino) < be32_to_cpu(agi->agi_length)); |
1909 | ||
1da177e4 LT |
1910 | /* |
1911 | * Initialize the cursor. | |
1912 | */ | |
57bd3dbe | 1913 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
1da177e4 | 1914 | |
0b48db80 DC |
1915 | error = xfs_check_agi_freecount(cur, agi); |
1916 | if (error) | |
1917 | goto error0; | |
1918 | ||
1da177e4 LT |
1919 | /* |
1920 | * Look for the entry describing this inode. | |
1921 | */ | |
21875505 | 1922 | if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) { |
0b932ccc DC |
1923 | xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.", |
1924 | __func__, error); | |
1da177e4 LT |
1925 | goto error0; |
1926 | } | |
c29aad41 | 1927 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
2e287a73 CH |
1928 | error = xfs_inobt_get_rec(cur, &rec, &i); |
1929 | if (error) { | |
0b932ccc DC |
1930 | xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.", |
1931 | __func__, error); | |
1da177e4 LT |
1932 | goto error0; |
1933 | } | |
c29aad41 | 1934 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 LT |
1935 | /* |
1936 | * Get the offset in the inode chunk. | |
1937 | */ | |
1938 | off = agino - rec.ir_startino; | |
1939 | ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK); | |
0d87e656 | 1940 | ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off))); |
1da177e4 LT |
1941 | /* |
1942 | * Mark the inode free & increment the count. | |
1943 | */ | |
0d87e656 | 1944 | rec.ir_free |= XFS_INOBT_MASK(off); |
1da177e4 LT |
1945 | rec.ir_freecount++; |
1946 | ||
1947 | /* | |
999633d3 BF |
1948 | * When an inode chunk is free, it becomes eligible for removal. Don't |
1949 | * remove the chunk if the block size is large enough for multiple inode | |
1950 | * chunks (that might not be free). | |
1da177e4 | 1951 | */ |
1bd960ee | 1952 | if (!(mp->m_flags & XFS_MOUNT_IKEEP) && |
999633d3 BF |
1953 | rec.ir_free == XFS_INOBT_ALL_FREE && |
1954 | mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK) { | |
09b56604 BF |
1955 | xic->deleted = 1; |
1956 | xic->first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino); | |
1957 | xic->alloc = xfs_inobt_irec_to_allocmask(&rec); | |
1da177e4 LT |
1958 | |
1959 | /* | |
1960 | * Remove the inode cluster from the AGI B+Tree, adjust the | |
1961 | * AGI and Superblock inode counts, and mark the disk space | |
1962 | * to be freed when the transaction is committed. | |
1963 | */ | |
999633d3 | 1964 | ilen = rec.ir_freecount; |
413d57c9 MS |
1965 | be32_add_cpu(&agi->agi_count, -ilen); |
1966 | be32_add_cpu(&agi->agi_freecount, -(ilen - 1)); | |
1da177e4 | 1967 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT); |
44b56e0a DC |
1968 | pag = xfs_perag_get(mp, agno); |
1969 | pag->pagi_freecount -= ilen - 1; | |
1970 | xfs_perag_put(pag); | |
1da177e4 LT |
1971 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen); |
1972 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1)); | |
1973 | ||
91cca5df | 1974 | if ((error = xfs_btree_delete(cur, &i))) { |
0b932ccc DC |
1975 | xfs_warn(mp, "%s: xfs_btree_delete returned error %d.", |
1976 | __func__, error); | |
1da177e4 LT |
1977 | goto error0; |
1978 | } | |
1979 | ||
10ae3dc7 | 1980 | xfs_difree_inode_chunk(mp, agno, &rec, flist); |
1da177e4 | 1981 | } else { |
09b56604 | 1982 | xic->deleted = 0; |
1da177e4 | 1983 | |
afabc24a CH |
1984 | error = xfs_inobt_update(cur, &rec); |
1985 | if (error) { | |
0b932ccc DC |
1986 | xfs_warn(mp, "%s: xfs_inobt_update returned error %d.", |
1987 | __func__, error); | |
1da177e4 LT |
1988 | goto error0; |
1989 | } | |
afabc24a | 1990 | |
1da177e4 LT |
1991 | /* |
1992 | * Change the inode free counts and log the ag/sb changes. | |
1993 | */ | |
413d57c9 | 1994 | be32_add_cpu(&agi->agi_freecount, 1); |
1da177e4 | 1995 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); |
44b56e0a DC |
1996 | pag = xfs_perag_get(mp, agno); |
1997 | pag->pagi_freecount++; | |
1998 | xfs_perag_put(pag); | |
1da177e4 LT |
1999 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1); |
2000 | } | |
2001 | ||
0b48db80 DC |
2002 | error = xfs_check_agi_freecount(cur, agi); |
2003 | if (error) | |
2004 | goto error0; | |
1da177e4 | 2005 | |
2b64ee5c | 2006 | *orec = rec; |
1da177e4 LT |
2007 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
2008 | return 0; | |
2009 | ||
2010 | error0: | |
2011 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
2012 | return error; | |
2013 | } | |
2014 | ||
3efa4ffd BF |
2015 | /* |
2016 | * Free an inode in the free inode btree. | |
2017 | */ | |
2018 | STATIC int | |
2019 | xfs_difree_finobt( | |
2020 | struct xfs_mount *mp, | |
2021 | struct xfs_trans *tp, | |
2022 | struct xfs_buf *agbp, | |
2023 | xfs_agino_t agino, | |
2024 | struct xfs_inobt_rec_incore *ibtrec) /* inobt record */ | |
2025 | { | |
2026 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
2027 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
2028 | struct xfs_btree_cur *cur; | |
2029 | struct xfs_inobt_rec_incore rec; | |
2030 | int offset = agino - ibtrec->ir_startino; | |
2031 | int error; | |
2032 | int i; | |
2033 | ||
2034 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO); | |
2035 | ||
2036 | error = xfs_inobt_lookup(cur, ibtrec->ir_startino, XFS_LOOKUP_EQ, &i); | |
2037 | if (error) | |
2038 | goto error; | |
2039 | if (i == 0) { | |
2040 | /* | |
2041 | * If the record does not exist in the finobt, we must have just | |
2042 | * freed an inode in a previously fully allocated chunk. If not, | |
2043 | * something is out of sync. | |
2044 | */ | |
c29aad41 | 2045 | XFS_WANT_CORRUPTED_GOTO(mp, ibtrec->ir_freecount == 1, error); |
3efa4ffd | 2046 | |
5419040f BF |
2047 | error = xfs_inobt_insert_rec(cur, ibtrec->ir_holemask, |
2048 | ibtrec->ir_count, | |
2049 | ibtrec->ir_freecount, | |
3efa4ffd BF |
2050 | ibtrec->ir_free, &i); |
2051 | if (error) | |
2052 | goto error; | |
2053 | ASSERT(i == 1); | |
2054 | ||
2055 | goto out; | |
2056 | } | |
2057 | ||
2058 | /* | |
2059 | * Read and update the existing record. We could just copy the ibtrec | |
2060 | * across here, but that would defeat the purpose of having redundant | |
2061 | * metadata. By making the modifications independently, we can catch | |
2062 | * corruptions that we wouldn't see if we just copied from one record | |
2063 | * to another. | |
2064 | */ | |
2065 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
2066 | if (error) | |
2067 | goto error; | |
c29aad41 | 2068 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error); |
3efa4ffd BF |
2069 | |
2070 | rec.ir_free |= XFS_INOBT_MASK(offset); | |
2071 | rec.ir_freecount++; | |
2072 | ||
c29aad41 | 2073 | XFS_WANT_CORRUPTED_GOTO(mp, (rec.ir_free == ibtrec->ir_free) && |
3efa4ffd BF |
2074 | (rec.ir_freecount == ibtrec->ir_freecount), |
2075 | error); | |
2076 | ||
2077 | /* | |
2078 | * The content of inobt records should always match between the inobt | |
2079 | * and finobt. The lifecycle of records in the finobt is different from | |
2080 | * the inobt in that the finobt only tracks records with at least one | |
2081 | * free inode. Hence, if all of the inodes are free and we aren't | |
2082 | * keeping inode chunks permanently on disk, remove the record. | |
2083 | * Otherwise, update the record with the new information. | |
999633d3 BF |
2084 | * |
2085 | * Note that we currently can't free chunks when the block size is large | |
2086 | * enough for multiple chunks. Leave the finobt record to remain in sync | |
2087 | * with the inobt. | |
3efa4ffd | 2088 | */ |
999633d3 BF |
2089 | if (rec.ir_free == XFS_INOBT_ALL_FREE && |
2090 | mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK && | |
3efa4ffd BF |
2091 | !(mp->m_flags & XFS_MOUNT_IKEEP)) { |
2092 | error = xfs_btree_delete(cur, &i); | |
2093 | if (error) | |
2094 | goto error; | |
2095 | ASSERT(i == 1); | |
2096 | } else { | |
2097 | error = xfs_inobt_update(cur, &rec); | |
2098 | if (error) | |
2099 | goto error; | |
2100 | } | |
2101 | ||
2102 | out: | |
2103 | error = xfs_check_agi_freecount(cur, agi); | |
2104 | if (error) | |
2105 | goto error; | |
2106 | ||
2107 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
2108 | return 0; | |
2109 | ||
2110 | error: | |
2111 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
2112 | return error; | |
2113 | } | |
2114 | ||
2b64ee5c BF |
2115 | /* |
2116 | * Free disk inode. Carefully avoids touching the incore inode, all | |
2117 | * manipulations incore are the caller's responsibility. | |
2118 | * The on-disk inode is not changed by this operation, only the | |
2119 | * btree (free inode mask) is changed. | |
2120 | */ | |
2121 | int | |
2122 | xfs_difree( | |
2123 | struct xfs_trans *tp, /* transaction pointer */ | |
2124 | xfs_ino_t inode, /* inode to be freed */ | |
310a75a3 | 2125 | struct xfs_defer_ops *flist, /* extents to free */ |
09b56604 | 2126 | struct xfs_icluster *xic) /* cluster info if deleted */ |
2b64ee5c BF |
2127 | { |
2128 | /* REFERENCED */ | |
2129 | xfs_agblock_t agbno; /* block number containing inode */ | |
2130 | struct xfs_buf *agbp; /* buffer for allocation group header */ | |
2131 | xfs_agino_t agino; /* allocation group inode number */ | |
2132 | xfs_agnumber_t agno; /* allocation group number */ | |
2133 | int error; /* error return value */ | |
2134 | struct xfs_mount *mp; /* mount structure for filesystem */ | |
2135 | struct xfs_inobt_rec_incore rec;/* btree record */ | |
2136 | ||
2137 | mp = tp->t_mountp; | |
2138 | ||
2139 | /* | |
2140 | * Break up inode number into its components. | |
2141 | */ | |
2142 | agno = XFS_INO_TO_AGNO(mp, inode); | |
2143 | if (agno >= mp->m_sb.sb_agcount) { | |
2144 | xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).", | |
2145 | __func__, agno, mp->m_sb.sb_agcount); | |
2146 | ASSERT(0); | |
2451337d | 2147 | return -EINVAL; |
2b64ee5c BF |
2148 | } |
2149 | agino = XFS_INO_TO_AGINO(mp, inode); | |
2150 | if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
2151 | xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).", | |
2152 | __func__, (unsigned long long)inode, | |
2153 | (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino)); | |
2154 | ASSERT(0); | |
2451337d | 2155 | return -EINVAL; |
2b64ee5c BF |
2156 | } |
2157 | agbno = XFS_AGINO_TO_AGBNO(mp, agino); | |
2158 | if (agbno >= mp->m_sb.sb_agblocks) { | |
2159 | xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).", | |
2160 | __func__, agbno, mp->m_sb.sb_agblocks); | |
2161 | ASSERT(0); | |
2451337d | 2162 | return -EINVAL; |
2b64ee5c BF |
2163 | } |
2164 | /* | |
2165 | * Get the allocation group header. | |
2166 | */ | |
2167 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); | |
2168 | if (error) { | |
2169 | xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.", | |
2170 | __func__, error); | |
2171 | return error; | |
2172 | } | |
2173 | ||
2174 | /* | |
2175 | * Fix up the inode allocation btree. | |
2176 | */ | |
09b56604 | 2177 | error = xfs_difree_inobt(mp, tp, agbp, agino, flist, xic, &rec); |
2b64ee5c BF |
2178 | if (error) |
2179 | goto error0; | |
2180 | ||
3efa4ffd BF |
2181 | /* |
2182 | * Fix up the free inode btree. | |
2183 | */ | |
2184 | if (xfs_sb_version_hasfinobt(&mp->m_sb)) { | |
2185 | error = xfs_difree_finobt(mp, tp, agbp, agino, &rec); | |
2186 | if (error) | |
2187 | goto error0; | |
2188 | } | |
2189 | ||
2b64ee5c BF |
2190 | return 0; |
2191 | ||
2192 | error0: | |
2193 | return error; | |
2194 | } | |
2195 | ||
7124fe0a DC |
2196 | STATIC int |
2197 | xfs_imap_lookup( | |
2198 | struct xfs_mount *mp, | |
2199 | struct xfs_trans *tp, | |
2200 | xfs_agnumber_t agno, | |
2201 | xfs_agino_t agino, | |
2202 | xfs_agblock_t agbno, | |
2203 | xfs_agblock_t *chunk_agbno, | |
2204 | xfs_agblock_t *offset_agbno, | |
2205 | int flags) | |
2206 | { | |
2207 | struct xfs_inobt_rec_incore rec; | |
2208 | struct xfs_btree_cur *cur; | |
2209 | struct xfs_buf *agbp; | |
7124fe0a DC |
2210 | int error; |
2211 | int i; | |
2212 | ||
2213 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); | |
2214 | if (error) { | |
53487786 DC |
2215 | xfs_alert(mp, |
2216 | "%s: xfs_ialloc_read_agi() returned error %d, agno %d", | |
2217 | __func__, error, agno); | |
7124fe0a DC |
2218 | return error; |
2219 | } | |
2220 | ||
2221 | /* | |
4536f2ad DC |
2222 | * Lookup the inode record for the given agino. If the record cannot be |
2223 | * found, then it's an invalid inode number and we should abort. Once | |
2224 | * we have a record, we need to ensure it contains the inode number | |
2225 | * we are looking up. | |
7124fe0a | 2226 | */ |
57bd3dbe | 2227 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
4536f2ad | 2228 | error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i); |
7124fe0a DC |
2229 | if (!error) { |
2230 | if (i) | |
2231 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
2232 | if (!error && i == 0) | |
2451337d | 2233 | error = -EINVAL; |
7124fe0a DC |
2234 | } |
2235 | ||
2236 | xfs_trans_brelse(tp, agbp); | |
f307080a | 2237 | xfs_btree_del_cursor(cur, error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); |
7124fe0a DC |
2238 | if (error) |
2239 | return error; | |
2240 | ||
4536f2ad DC |
2241 | /* check that the returned record contains the required inode */ |
2242 | if (rec.ir_startino > agino || | |
71783438 | 2243 | rec.ir_startino + mp->m_ialloc_inos <= agino) |
2451337d | 2244 | return -EINVAL; |
4536f2ad | 2245 | |
7124fe0a | 2246 | /* for untrusted inodes check it is allocated first */ |
1920779e | 2247 | if ((flags & XFS_IGET_UNTRUSTED) && |
7124fe0a | 2248 | (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino))) |
2451337d | 2249 | return -EINVAL; |
7124fe0a DC |
2250 | |
2251 | *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino); | |
2252 | *offset_agbno = agbno - *chunk_agbno; | |
2253 | return 0; | |
2254 | } | |
2255 | ||
1da177e4 | 2256 | /* |
94e1b69d | 2257 | * Return the location of the inode in imap, for mapping it into a buffer. |
1da177e4 | 2258 | */ |
1da177e4 | 2259 | int |
94e1b69d CH |
2260 | xfs_imap( |
2261 | xfs_mount_t *mp, /* file system mount structure */ | |
2262 | xfs_trans_t *tp, /* transaction pointer */ | |
1da177e4 | 2263 | xfs_ino_t ino, /* inode to locate */ |
94e1b69d CH |
2264 | struct xfs_imap *imap, /* location map structure */ |
2265 | uint flags) /* flags for inode btree lookup */ | |
1da177e4 LT |
2266 | { |
2267 | xfs_agblock_t agbno; /* block number of inode in the alloc group */ | |
1da177e4 LT |
2268 | xfs_agino_t agino; /* inode number within alloc group */ |
2269 | xfs_agnumber_t agno; /* allocation group number */ | |
2270 | int blks_per_cluster; /* num blocks per inode cluster */ | |
2271 | xfs_agblock_t chunk_agbno; /* first block in inode chunk */ | |
1da177e4 | 2272 | xfs_agblock_t cluster_agbno; /* first block in inode cluster */ |
1da177e4 | 2273 | int error; /* error code */ |
1da177e4 | 2274 | int offset; /* index of inode in its buffer */ |
836a94ad | 2275 | xfs_agblock_t offset_agbno; /* blks from chunk start to inode */ |
1da177e4 LT |
2276 | |
2277 | ASSERT(ino != NULLFSINO); | |
94e1b69d | 2278 | |
1da177e4 LT |
2279 | /* |
2280 | * Split up the inode number into its parts. | |
2281 | */ | |
2282 | agno = XFS_INO_TO_AGNO(mp, ino); | |
2283 | agino = XFS_INO_TO_AGINO(mp, ino); | |
2284 | agbno = XFS_AGINO_TO_AGBNO(mp, agino); | |
2285 | if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks || | |
2286 | ino != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
2287 | #ifdef DEBUG | |
1920779e DC |
2288 | /* |
2289 | * Don't output diagnostic information for untrusted inodes | |
2290 | * as they can be invalid without implying corruption. | |
2291 | */ | |
2292 | if (flags & XFS_IGET_UNTRUSTED) | |
2451337d | 2293 | return -EINVAL; |
1da177e4 | 2294 | if (agno >= mp->m_sb.sb_agcount) { |
53487786 DC |
2295 | xfs_alert(mp, |
2296 | "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)", | |
2297 | __func__, agno, mp->m_sb.sb_agcount); | |
1da177e4 LT |
2298 | } |
2299 | if (agbno >= mp->m_sb.sb_agblocks) { | |
53487786 DC |
2300 | xfs_alert(mp, |
2301 | "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)", | |
2302 | __func__, (unsigned long long)agbno, | |
2303 | (unsigned long)mp->m_sb.sb_agblocks); | |
1da177e4 LT |
2304 | } |
2305 | if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
53487786 DC |
2306 | xfs_alert(mp, |
2307 | "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)", | |
2308 | __func__, ino, | |
2309 | XFS_AGINO_TO_INO(mp, agno, agino)); | |
1da177e4 | 2310 | } |
745b1f47 | 2311 | xfs_stack_trace(); |
1da177e4 | 2312 | #endif /* DEBUG */ |
2451337d | 2313 | return -EINVAL; |
1da177e4 | 2314 | } |
94e1b69d | 2315 | |
f9e5abcf | 2316 | blks_per_cluster = xfs_icluster_size_fsb(mp); |
7124fe0a DC |
2317 | |
2318 | /* | |
2319 | * For bulkstat and handle lookups, we have an untrusted inode number | |
2320 | * that we have to verify is valid. We cannot do this just by reading | |
2321 | * the inode buffer as it may have been unlinked and removed leaving | |
2322 | * inodes in stale state on disk. Hence we have to do a btree lookup | |
2323 | * in all cases where an untrusted inode number is passed. | |
2324 | */ | |
1920779e | 2325 | if (flags & XFS_IGET_UNTRUSTED) { |
7124fe0a DC |
2326 | error = xfs_imap_lookup(mp, tp, agno, agino, agbno, |
2327 | &chunk_agbno, &offset_agbno, flags); | |
2328 | if (error) | |
2329 | return error; | |
2330 | goto out_map; | |
2331 | } | |
2332 | ||
94e1b69d CH |
2333 | /* |
2334 | * If the inode cluster size is the same as the blocksize or | |
2335 | * smaller we get to the buffer by simple arithmetics. | |
2336 | */ | |
f9e5abcf | 2337 | if (blks_per_cluster == 1) { |
1da177e4 LT |
2338 | offset = XFS_INO_TO_OFFSET(mp, ino); |
2339 | ASSERT(offset < mp->m_sb.sb_inopblock); | |
94e1b69d CH |
2340 | |
2341 | imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno); | |
2342 | imap->im_len = XFS_FSB_TO_BB(mp, 1); | |
2343 | imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog); | |
1da177e4 LT |
2344 | return 0; |
2345 | } | |
94e1b69d | 2346 | |
94e1b69d CH |
2347 | /* |
2348 | * If the inode chunks are aligned then use simple maths to | |
2349 | * find the location. Otherwise we have to do a btree | |
2350 | * lookup to find the location. | |
2351 | */ | |
1da177e4 LT |
2352 | if (mp->m_inoalign_mask) { |
2353 | offset_agbno = agbno & mp->m_inoalign_mask; | |
2354 | chunk_agbno = agbno - offset_agbno; | |
2355 | } else { | |
7124fe0a DC |
2356 | error = xfs_imap_lookup(mp, tp, agno, agino, agbno, |
2357 | &chunk_agbno, &offset_agbno, flags); | |
1da177e4 LT |
2358 | if (error) |
2359 | return error; | |
1da177e4 | 2360 | } |
94e1b69d | 2361 | |
7124fe0a | 2362 | out_map: |
1da177e4 LT |
2363 | ASSERT(agbno >= chunk_agbno); |
2364 | cluster_agbno = chunk_agbno + | |
2365 | ((offset_agbno / blks_per_cluster) * blks_per_cluster); | |
2366 | offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) + | |
2367 | XFS_INO_TO_OFFSET(mp, ino); | |
94e1b69d CH |
2368 | |
2369 | imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno); | |
2370 | imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster); | |
2371 | imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog); | |
2372 | ||
2373 | /* | |
2374 | * If the inode number maps to a block outside the bounds | |
2375 | * of the file system then return NULL rather than calling | |
2376 | * read_buf and panicing when we get an error from the | |
2377 | * driver. | |
2378 | */ | |
2379 | if ((imap->im_blkno + imap->im_len) > | |
2380 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) { | |
53487786 DC |
2381 | xfs_alert(mp, |
2382 | "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)", | |
2383 | __func__, (unsigned long long) imap->im_blkno, | |
94e1b69d CH |
2384 | (unsigned long long) imap->im_len, |
2385 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)); | |
2451337d | 2386 | return -EINVAL; |
94e1b69d | 2387 | } |
1da177e4 | 2388 | return 0; |
1da177e4 LT |
2389 | } |
2390 | ||
2391 | /* | |
2392 | * Compute and fill in value of m_in_maxlevels. | |
2393 | */ | |
2394 | void | |
2395 | xfs_ialloc_compute_maxlevels( | |
2396 | xfs_mount_t *mp) /* file system mount structure */ | |
2397 | { | |
19b54ee6 DW |
2398 | uint inodes; |
2399 | ||
2400 | inodes = (1LL << XFS_INO_AGINO_BITS(mp)) >> XFS_INODES_PER_CHUNK_LOG; | |
2401 | mp->m_in_maxlevels = xfs_btree_compute_maxlevels(mp, mp->m_inobt_mnr, | |
2402 | inodes); | |
1da177e4 LT |
2403 | } |
2404 | ||
2405 | /* | |
aafc3c24 BF |
2406 | * Log specified fields for the ag hdr (inode section). The growth of the agi |
2407 | * structure over time requires that we interpret the buffer as two logical | |
2408 | * regions delineated by the end of the unlinked list. This is due to the size | |
2409 | * of the hash table and its location in the middle of the agi. | |
2410 | * | |
2411 | * For example, a request to log a field before agi_unlinked and a field after | |
2412 | * agi_unlinked could cause us to log the entire hash table and use an excessive | |
2413 | * amount of log space. To avoid this behavior, log the region up through | |
2414 | * agi_unlinked in one call and the region after agi_unlinked through the end of | |
2415 | * the structure in another. | |
1da177e4 LT |
2416 | */ |
2417 | void | |
2418 | xfs_ialloc_log_agi( | |
2419 | xfs_trans_t *tp, /* transaction pointer */ | |
2420 | xfs_buf_t *bp, /* allocation group header buffer */ | |
2421 | int fields) /* bitmask of fields to log */ | |
2422 | { | |
2423 | int first; /* first byte number */ | |
2424 | int last; /* last byte number */ | |
2425 | static const short offsets[] = { /* field starting offsets */ | |
2426 | /* keep in sync with bit definitions */ | |
2427 | offsetof(xfs_agi_t, agi_magicnum), | |
2428 | offsetof(xfs_agi_t, agi_versionnum), | |
2429 | offsetof(xfs_agi_t, agi_seqno), | |
2430 | offsetof(xfs_agi_t, agi_length), | |
2431 | offsetof(xfs_agi_t, agi_count), | |
2432 | offsetof(xfs_agi_t, agi_root), | |
2433 | offsetof(xfs_agi_t, agi_level), | |
2434 | offsetof(xfs_agi_t, agi_freecount), | |
2435 | offsetof(xfs_agi_t, agi_newino), | |
2436 | offsetof(xfs_agi_t, agi_dirino), | |
2437 | offsetof(xfs_agi_t, agi_unlinked), | |
aafc3c24 BF |
2438 | offsetof(xfs_agi_t, agi_free_root), |
2439 | offsetof(xfs_agi_t, agi_free_level), | |
1da177e4 LT |
2440 | sizeof(xfs_agi_t) |
2441 | }; | |
2442 | #ifdef DEBUG | |
2443 | xfs_agi_t *agi; /* allocation group header */ | |
2444 | ||
2445 | agi = XFS_BUF_TO_AGI(bp); | |
69ef921b | 2446 | ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC)); |
1da177e4 | 2447 | #endif |
aafc3c24 BF |
2448 | |
2449 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_AGI_BUF); | |
2450 | ||
1da177e4 | 2451 | /* |
aafc3c24 BF |
2452 | * Compute byte offsets for the first and last fields in the first |
2453 | * region and log the agi buffer. This only logs up through | |
2454 | * agi_unlinked. | |
1da177e4 | 2455 | */ |
aafc3c24 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 | ||
1da177e4 | 2462 | /* |
aafc3c24 BF |
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. | |
1da177e4 | 2465 | */ |
aafc3c24 BF |
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 | } | |
1da177e4 LT |
2472 | } |
2473 | ||
5e1be0fb CH |
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 | ||
983d09ff | 2488 | static bool |
612cfbfe | 2489 | xfs_agi_verify( |
3702ce6e DC |
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); | |
3702ce6e | 2494 | |
a45086e2 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))) | |
983d09ff | 2500 | return false; |
a45086e2 BF |
2501 | } |
2502 | ||
3702ce6e DC |
2503 | /* |
2504 | * Validate the magic number of the agi block. | |
2505 | */ | |
983d09ff 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; | |
3702ce6e | 2510 | |
e1b05723 ES |
2511 | if (be32_to_cpu(agi->agi_level) > XFS_BTREE_MAXLEVELS) |
2512 | return false; | |
3702ce6e DC |
2513 | /* |
2514 | * during growfs operations, the perag is not fully initialised, | |
2515 | * so we can't use it for any useful checking. growfs ensures we can't | |
2516 | * use it by using uncached buffers that don't have the perag attached | |
2517 | * so we can detect and avoid this problem. | |
2518 | */ | |
983d09ff DC |
2519 | if (bp->b_pag && be32_to_cpu(agi->agi_seqno) != bp->b_pag->pag_agno) |
2520 | return false; | |
3702ce6e | 2521 | |
3702ce6e | 2522 | xfs_check_agi_unlinked(agi); |
983d09ff | 2523 | return true; |
612cfbfe DC |
2524 | } |
2525 | ||
1813dd64 DC |
2526 | static void |
2527 | xfs_agi_read_verify( | |
612cfbfe DC |
2528 | struct xfs_buf *bp) |
2529 | { | |
983d09ff | 2530 | struct xfs_mount *mp = bp->b_target->bt_mount; |
983d09ff | 2531 | |
ce5028cf ES |
2532 | if (xfs_sb_version_hascrc(&mp->m_sb) && |
2533 | !xfs_buf_verify_cksum(bp, XFS_AGI_CRC_OFF)) | |
2451337d | 2534 | xfs_buf_ioerror(bp, -EFSBADCRC); |
ce5028cf ES |
2535 | else if (XFS_TEST_ERROR(!xfs_agi_verify(bp), mp, |
2536 | XFS_ERRTAG_IALLOC_READ_AGI, | |
2537 | XFS_RANDOM_IALLOC_READ_AGI)) | |
2451337d | 2538 | xfs_buf_ioerror(bp, -EFSCORRUPTED); |
ce5028cf ES |
2539 | |
2540 | if (bp->b_error) | |
2541 | xfs_verifier_error(bp); | |
612cfbfe DC |
2542 | } |
2543 | ||
b0f539de | 2544 | static void |
1813dd64 | 2545 | xfs_agi_write_verify( |
612cfbfe DC |
2546 | struct xfs_buf *bp) |
2547 | { | |
983d09ff DC |
2548 | struct xfs_mount *mp = bp->b_target->bt_mount; |
2549 | struct xfs_buf_log_item *bip = bp->b_fspriv; | |
2550 | ||
2551 | if (!xfs_agi_verify(bp)) { | |
2451337d | 2552 | xfs_buf_ioerror(bp, -EFSCORRUPTED); |
ce5028cf | 2553 | xfs_verifier_error(bp); |
983d09ff DC |
2554 | return; |
2555 | } | |
2556 | ||
2557 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
2558 | return; | |
2559 | ||
2560 | if (bip) | |
2561 | XFS_BUF_TO_AGI(bp)->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn); | |
f1dbcd7e | 2562 | xfs_buf_update_cksum(bp, XFS_AGI_CRC_OFF); |
3702ce6e DC |
2563 | } |
2564 | ||
1813dd64 | 2565 | const struct xfs_buf_ops xfs_agi_buf_ops = { |
233135b7 | 2566 | .name = "xfs_agi", |
1813dd64 DC |
2567 | .verify_read = xfs_agi_read_verify, |
2568 | .verify_write = xfs_agi_write_verify, | |
2569 | }; | |
2570 | ||
1da177e4 LT |
2571 | /* |
2572 | * Read in the allocation group header (inode allocation section) | |
2573 | */ | |
2574 | int | |
5e1be0fb CH |
2575 | xfs_read_agi( |
2576 | struct xfs_mount *mp, /* file system mount structure */ | |
2577 | struct xfs_trans *tp, /* transaction pointer */ | |
2578 | xfs_agnumber_t agno, /* allocation group number */ | |
2579 | struct xfs_buf **bpp) /* allocation group hdr buf */ | |
1da177e4 | 2580 | { |
5e1be0fb | 2581 | int error; |
1da177e4 | 2582 | |
d123031a | 2583 | trace_xfs_read_agi(mp, agno); |
5e1be0fb | 2584 | |
d123031a | 2585 | ASSERT(agno != NULLAGNUMBER); |
5e1be0fb | 2586 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, |
1da177e4 | 2587 | XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)), |
1813dd64 | 2588 | XFS_FSS_TO_BB(mp, 1), 0, bpp, &xfs_agi_buf_ops); |
1da177e4 LT |
2589 | if (error) |
2590 | return error; | |
5e1be0fb | 2591 | |
38f23232 | 2592 | xfs_buf_set_ref(*bpp, XFS_AGI_REF); |
5e1be0fb CH |
2593 | return 0; |
2594 | } | |
2595 | ||
2596 | int | |
2597 | xfs_ialloc_read_agi( | |
2598 | struct xfs_mount *mp, /* file system mount structure */ | |
2599 | struct xfs_trans *tp, /* transaction pointer */ | |
2600 | xfs_agnumber_t agno, /* allocation group number */ | |
2601 | struct xfs_buf **bpp) /* allocation group hdr buf */ | |
2602 | { | |
2603 | struct xfs_agi *agi; /* allocation group header */ | |
2604 | struct xfs_perag *pag; /* per allocation group data */ | |
2605 | int error; | |
2606 | ||
d123031a DC |
2607 | trace_xfs_ialloc_read_agi(mp, agno); |
2608 | ||
5e1be0fb CH |
2609 | error = xfs_read_agi(mp, tp, agno, bpp); |
2610 | if (error) | |
2611 | return error; | |
2612 | ||
2613 | agi = XFS_BUF_TO_AGI(*bpp); | |
44b56e0a | 2614 | pag = xfs_perag_get(mp, agno); |
1da177e4 | 2615 | if (!pag->pagi_init) { |
16259e7d | 2616 | pag->pagi_freecount = be32_to_cpu(agi->agi_freecount); |
92821e2b | 2617 | pag->pagi_count = be32_to_cpu(agi->agi_count); |
1da177e4 | 2618 | pag->pagi_init = 1; |
1da177e4 | 2619 | } |
1da177e4 | 2620 | |
5e1be0fb CH |
2621 | /* |
2622 | * It's possible for these to be out of sync if | |
2623 | * we are in the middle of a forced shutdown. | |
2624 | */ | |
2625 | ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) || | |
2626 | XFS_FORCED_SHUTDOWN(mp)); | |
44b56e0a | 2627 | xfs_perag_put(pag); |
1da177e4 LT |
2628 | return 0; |
2629 | } | |
92821e2b DC |
2630 | |
2631 | /* | |
2632 | * Read in the agi to initialise the per-ag data in the mount structure | |
2633 | */ | |
2634 | int | |
2635 | xfs_ialloc_pagi_init( | |
2636 | xfs_mount_t *mp, /* file system mount structure */ | |
2637 | xfs_trans_t *tp, /* transaction pointer */ | |
2638 | xfs_agnumber_t agno) /* allocation group number */ | |
2639 | { | |
2640 | xfs_buf_t *bp = NULL; | |
2641 | int error; | |
2642 | ||
2643 | error = xfs_ialloc_read_agi(mp, tp, agno, &bp); | |
2644 | if (error) | |
2645 | return error; | |
2646 | if (bp) | |
2647 | xfs_trans_brelse(tp, bp); | |
2648 | return 0; | |
2649 | } |