]>
Commit | Line | Data |
---|---|---|
0b61f8a4 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 | 2 | /* |
7b718769 NS |
3 | * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc. |
4 | * All Rights Reserved. | |
1da177e4 | 5 | */ |
1da177e4 | 6 | #include "xfs.h" |
a844f451 | 7 | #include "xfs_fs.h" |
70a9883c | 8 | #include "xfs_shared.h" |
239880ef DC |
9 | #include "xfs_format.h" |
10 | #include "xfs_log_format.h" | |
11 | #include "xfs_trans_resv.h" | |
a844f451 | 12 | #include "xfs_bit.h" |
1da177e4 | 13 | #include "xfs_mount.h" |
1da177e4 | 14 | #include "xfs_inode.h" |
a844f451 NS |
15 | #include "xfs_btree.h" |
16 | #include "xfs_ialloc.h" | |
a4fbe6ab | 17 | #include "xfs_ialloc_btree.h" |
1da177e4 | 18 | #include "xfs_alloc.h" |
e9e899a2 | 19 | #include "xfs_errortag.h" |
1da177e4 LT |
20 | #include "xfs_error.h" |
21 | #include "xfs_bmap.h" | |
239880ef | 22 | #include "xfs_trans.h" |
983d09ff | 23 | #include "xfs_buf_item.h" |
ddf6ad01 | 24 | #include "xfs_icreate_item.h" |
7bb85ef3 | 25 | #include "xfs_icache.h" |
d123031a | 26 | #include "xfs_trace.h" |
a45086e2 | 27 | #include "xfs_log.h" |
340785cc | 28 | #include "xfs_rmap.h" |
9bbafc71 | 29 | #include "xfs_ag.h" |
1da177e4 | 30 | |
fe033cc8 | 31 | /* |
21875505 | 32 | * Lookup a record by ino in the btree given by cur. |
fe033cc8 | 33 | */ |
81e25176 | 34 | int /* error */ |
21875505 | 35 | xfs_inobt_lookup( |
fe033cc8 CH |
36 | struct xfs_btree_cur *cur, /* btree cursor */ |
37 | xfs_agino_t ino, /* starting inode of chunk */ | |
21875505 | 38 | xfs_lookup_t dir, /* <=, >=, == */ |
fe033cc8 CH |
39 | int *stat) /* success/failure */ |
40 | { | |
41 | cur->bc_rec.i.ir_startino = ino; | |
5419040f BF |
42 | cur->bc_rec.i.ir_holemask = 0; |
43 | cur->bc_rec.i.ir_count = 0; | |
21875505 CH |
44 | cur->bc_rec.i.ir_freecount = 0; |
45 | cur->bc_rec.i.ir_free = 0; | |
46 | return xfs_btree_lookup(cur, dir, stat); | |
fe033cc8 CH |
47 | } |
48 | ||
278d0ca1 | 49 | /* |
afabc24a | 50 | * Update the record referred to by cur to the value given. |
278d0ca1 CH |
51 | * This either works (return 0) or gets an EFSCORRUPTED error. |
52 | */ | |
53 | STATIC int /* error */ | |
54 | xfs_inobt_update( | |
55 | struct xfs_btree_cur *cur, /* btree cursor */ | |
afabc24a | 56 | xfs_inobt_rec_incore_t *irec) /* btree record */ |
278d0ca1 CH |
57 | { |
58 | union xfs_btree_rec rec; | |
59 | ||
afabc24a | 60 | rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino); |
38c26bfd | 61 | if (xfs_has_sparseinodes(cur->bc_mp)) { |
5419040f BF |
62 | rec.inobt.ir_u.sp.ir_holemask = cpu_to_be16(irec->ir_holemask); |
63 | rec.inobt.ir_u.sp.ir_count = irec->ir_count; | |
64 | rec.inobt.ir_u.sp.ir_freecount = irec->ir_freecount; | |
65 | } else { | |
66 | /* ir_holemask/ir_count not supported on-disk */ | |
67 | rec.inobt.ir_u.f.ir_freecount = cpu_to_be32(irec->ir_freecount); | |
68 | } | |
afabc24a | 69 | rec.inobt.ir_free = cpu_to_be64(irec->ir_free); |
278d0ca1 CH |
70 | return xfs_btree_update(cur, &rec); |
71 | } | |
72 | ||
e936945e DW |
73 | /* Convert on-disk btree record to incore inobt record. */ |
74 | void | |
75 | xfs_inobt_btrec_to_irec( | |
76 | struct xfs_mount *mp, | |
159eb69d | 77 | const union xfs_btree_rec *rec, |
e936945e | 78 | struct xfs_inobt_rec_incore *irec) |
8cc938fe | 79 | { |
5419040f | 80 | irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino); |
38c26bfd | 81 | if (xfs_has_sparseinodes(mp)) { |
5419040f BF |
82 | irec->ir_holemask = be16_to_cpu(rec->inobt.ir_u.sp.ir_holemask); |
83 | irec->ir_count = rec->inobt.ir_u.sp.ir_count; | |
84 | irec->ir_freecount = rec->inobt.ir_u.sp.ir_freecount; | |
85 | } else { | |
86 | /* | |
87 | * ir_holemask/ir_count not supported on-disk. Fill in hardcoded | |
88 | * values for full inode chunks. | |
89 | */ | |
90 | irec->ir_holemask = XFS_INOBT_HOLEMASK_FULL; | |
91 | irec->ir_count = XFS_INODES_PER_CHUNK; | |
92 | irec->ir_freecount = | |
93 | be32_to_cpu(rec->inobt.ir_u.f.ir_freecount); | |
8cc938fe | 94 | } |
5419040f | 95 | irec->ir_free = be64_to_cpu(rec->inobt.ir_free); |
e936945e DW |
96 | } |
97 | ||
98 | /* | |
99 | * Get the data from the pointed-to record. | |
100 | */ | |
101 | int | |
102 | xfs_inobt_get_rec( | |
103 | struct xfs_btree_cur *cur, | |
104 | struct xfs_inobt_rec_incore *irec, | |
105 | int *stat) | |
106 | { | |
9e6c08d4 | 107 | struct xfs_mount *mp = cur->bc_mp; |
e936945e DW |
108 | union xfs_btree_rec *rec; |
109 | int error; | |
9e6c08d4 | 110 | uint64_t realfree; |
e936945e DW |
111 | |
112 | error = xfs_btree_get_rec(cur, &rec, stat); | |
113 | if (error || *stat == 0) | |
114 | return error; | |
115 | ||
9e6c08d4 DC |
116 | xfs_inobt_btrec_to_irec(mp, rec, irec); |
117 | ||
2d6ca832 | 118 | if (!xfs_verify_agino(cur->bc_ag.pag, irec->ir_startino)) |
9e6c08d4 DC |
119 | goto out_bad_rec; |
120 | if (irec->ir_count < XFS_INODES_PER_HOLEMASK_BIT || | |
121 | irec->ir_count > XFS_INODES_PER_CHUNK) | |
122 | goto out_bad_rec; | |
123 | if (irec->ir_freecount > XFS_INODES_PER_CHUNK) | |
124 | goto out_bad_rec; | |
125 | ||
126 | /* if there are no holes, return the first available offset */ | |
127 | if (!xfs_inobt_issparse(irec->ir_holemask)) | |
128 | realfree = irec->ir_free; | |
129 | else | |
130 | realfree = irec->ir_free & xfs_inobt_irec_to_allocmask(irec); | |
131 | if (hweight64(realfree) != irec->ir_freecount) | |
132 | goto out_bad_rec; | |
5419040f BF |
133 | |
134 | return 0; | |
9e6c08d4 DC |
135 | |
136 | out_bad_rec: | |
137 | xfs_warn(mp, | |
138 | "%s Inode BTree record corruption in AG %d detected!", | |
2d6ca832 DC |
139 | cur->bc_btnum == XFS_BTNUM_INO ? "Used" : "Free", |
140 | cur->bc_ag.pag->pag_agno); | |
9e6c08d4 DC |
141 | xfs_warn(mp, |
142 | "start inode 0x%x, count 0x%x, free 0x%x freemask 0x%llx, holemask 0x%x", | |
143 | irec->ir_startino, irec->ir_count, irec->ir_freecount, | |
144 | irec->ir_free, irec->ir_holemask); | |
145 | return -EFSCORRUPTED; | |
8cc938fe CH |
146 | } |
147 | ||
0aa0a756 BF |
148 | /* |
149 | * Insert a single inobt record. Cursor must already point to desired location. | |
150 | */ | |
7f8f1313 | 151 | int |
0aa0a756 BF |
152 | xfs_inobt_insert_rec( |
153 | struct xfs_btree_cur *cur, | |
c8ce540d DW |
154 | uint16_t holemask, |
155 | uint8_t count, | |
156 | int32_t freecount, | |
0aa0a756 BF |
157 | xfs_inofree_t free, |
158 | int *stat) | |
159 | { | |
5419040f BF |
160 | cur->bc_rec.i.ir_holemask = holemask; |
161 | cur->bc_rec.i.ir_count = count; | |
0aa0a756 BF |
162 | cur->bc_rec.i.ir_freecount = freecount; |
163 | cur->bc_rec.i.ir_free = free; | |
164 | return xfs_btree_insert(cur, stat); | |
165 | } | |
166 | ||
167 | /* | |
168 | * Insert records describing a newly allocated inode chunk into the inobt. | |
169 | */ | |
170 | STATIC int | |
171 | xfs_inobt_insert( | |
172 | struct xfs_mount *mp, | |
173 | struct xfs_trans *tp, | |
174 | struct xfs_buf *agbp, | |
7b13c515 | 175 | struct xfs_perag *pag, |
0aa0a756 BF |
176 | xfs_agino_t newino, |
177 | xfs_agino_t newlen, | |
178 | xfs_btnum_t btnum) | |
179 | { | |
180 | struct xfs_btree_cur *cur; | |
0aa0a756 BF |
181 | xfs_agino_t thisino; |
182 | int i; | |
183 | int error; | |
184 | ||
7b13c515 | 185 | cur = xfs_inobt_init_cursor(mp, tp, agbp, pag, btnum); |
0aa0a756 BF |
186 | |
187 | for (thisino = newino; | |
188 | thisino < newino + newlen; | |
189 | thisino += XFS_INODES_PER_CHUNK) { | |
190 | error = xfs_inobt_lookup(cur, thisino, XFS_LOOKUP_EQ, &i); | |
191 | if (error) { | |
192 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
193 | return error; | |
194 | } | |
195 | ASSERT(i == 0); | |
196 | ||
5419040f BF |
197 | error = xfs_inobt_insert_rec(cur, XFS_INOBT_HOLEMASK_FULL, |
198 | XFS_INODES_PER_CHUNK, | |
199 | XFS_INODES_PER_CHUNK, | |
0aa0a756 BF |
200 | XFS_INOBT_ALL_FREE, &i); |
201 | if (error) { | |
202 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
203 | return error; | |
204 | } | |
205 | ASSERT(i == 1); | |
206 | } | |
207 | ||
208 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
209 | ||
210 | return 0; | |
211 | } | |
212 | ||
0b48db80 DC |
213 | /* |
214 | * Verify that the number of free inodes in the AGI is correct. | |
215 | */ | |
216 | #ifdef DEBUG | |
9ba0889e | 217 | static int |
0b48db80 | 218 | xfs_check_agi_freecount( |
9ba0889e | 219 | struct xfs_btree_cur *cur) |
0b48db80 DC |
220 | { |
221 | if (cur->bc_nlevels == 1) { | |
222 | xfs_inobt_rec_incore_t rec; | |
223 | int freecount = 0; | |
224 | int error; | |
225 | int i; | |
226 | ||
21875505 | 227 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); |
0b48db80 DC |
228 | if (error) |
229 | return error; | |
230 | ||
231 | do { | |
232 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
233 | if (error) | |
234 | return error; | |
235 | ||
236 | if (i) { | |
237 | freecount += rec.ir_freecount; | |
238 | error = xfs_btree_increment(cur, 0, &i); | |
239 | if (error) | |
240 | return error; | |
241 | } | |
242 | } while (i == 1); | |
243 | ||
75c8c50f | 244 | if (!xfs_is_shutdown(cur->bc_mp)) |
9ba0889e | 245 | ASSERT(freecount == cur->bc_ag.pag->pagi_freecount); |
0b48db80 DC |
246 | } |
247 | return 0; | |
248 | } | |
249 | #else | |
9ba0889e | 250 | #define xfs_check_agi_freecount(cur) 0 |
0b48db80 DC |
251 | #endif |
252 | ||
85c0b2ab | 253 | /* |
28c8e41a DC |
254 | * Initialise a new set of inodes. When called without a transaction context |
255 | * (e.g. from recovery) we initiate a delayed write of the inode buffers rather | |
256 | * than logging them (which in a transaction context puts them into the AIL | |
257 | * for writeback rather than the xfsbufd queue). | |
85c0b2ab | 258 | */ |
ddf6ad01 | 259 | int |
85c0b2ab DC |
260 | xfs_ialloc_inode_init( |
261 | struct xfs_mount *mp, | |
262 | struct xfs_trans *tp, | |
28c8e41a | 263 | struct list_head *buffer_list, |
463958af | 264 | int icount, |
85c0b2ab DC |
265 | xfs_agnumber_t agno, |
266 | xfs_agblock_t agbno, | |
267 | xfs_agblock_t length, | |
268 | unsigned int gen) | |
269 | { | |
270 | struct xfs_buf *fbuf; | |
271 | struct xfs_dinode *free; | |
83dcdb44 | 272 | int nbufs; |
85c0b2ab DC |
273 | int version; |
274 | int i, j; | |
275 | xfs_daddr_t d; | |
93848a99 | 276 | xfs_ino_t ino = 0; |
ce92464c | 277 | int error; |
85c0b2ab DC |
278 | |
279 | /* | |
6e0c7b8c JL |
280 | * Loop over the new block(s), filling in the inodes. For small block |
281 | * sizes, manipulate the inodes in buffers which are multiples of the | |
282 | * blocks size. | |
85c0b2ab | 283 | */ |
ef325959 | 284 | nbufs = length / M_IGEO(mp)->blocks_per_cluster; |
85c0b2ab DC |
285 | |
286 | /* | |
93848a99 CH |
287 | * Figure out what version number to use in the inodes we create. If |
288 | * the superblock version has caught up to the one that supports the new | |
289 | * inode format, then use the new inode version. Otherwise use the old | |
290 | * version so that old kernels will continue to be able to use the file | |
291 | * system. | |
292 | * | |
293 | * For v3 inodes, we also need to write the inode number into the inode, | |
294 | * so calculate the first inode number of the chunk here as | |
43004b2a | 295 | * XFS_AGB_TO_AGINO() only works within a filesystem block, not |
93848a99 CH |
296 | * across multiple filesystem blocks (such as a cluster) and so cannot |
297 | * be used in the cluster buffer loop below. | |
298 | * | |
299 | * Further, because we are writing the inode directly into the buffer | |
300 | * and calculating a CRC on the entire inode, we have ot log the entire | |
301 | * inode so that the entire range the CRC covers is present in the log. | |
302 | * That means for v3 inode we log the entire buffer rather than just the | |
303 | * inode cores. | |
85c0b2ab | 304 | */ |
ebd9027d | 305 | if (xfs_has_v3inodes(mp)) { |
93848a99 | 306 | version = 3; |
43004b2a | 307 | ino = XFS_AGINO_TO_INO(mp, agno, XFS_AGB_TO_AGINO(mp, agbno)); |
ddf6ad01 DC |
308 | |
309 | /* | |
310 | * log the initialisation that is about to take place as an | |
311 | * logical operation. This means the transaction does not | |
312 | * need to log the physical changes to the inode buffers as log | |
313 | * recovery will know what initialisation is actually needed. | |
314 | * Hence we only need to log the buffers as "ordered" buffers so | |
315 | * they track in the AIL as if they were physically logged. | |
316 | */ | |
317 | if (tp) | |
463958af | 318 | xfs_icreate_log(tp, agno, agbno, icount, |
ddf6ad01 | 319 | mp->m_sb.sb_inodesize, length, gen); |
263997a6 | 320 | } else |
85c0b2ab | 321 | version = 2; |
85c0b2ab DC |
322 | |
323 | for (j = 0; j < nbufs; j++) { | |
324 | /* | |
325 | * Get the block. | |
326 | */ | |
83dcdb44 | 327 | d = XFS_AGB_TO_DADDR(mp, agno, agbno + |
ef325959 | 328 | (j * M_IGEO(mp)->blocks_per_cluster)); |
ce92464c DW |
329 | error = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, |
330 | mp->m_bsize * M_IGEO(mp)->blocks_per_cluster, | |
331 | XBF_UNMAPPED, &fbuf); | |
332 | if (error) | |
333 | return error; | |
ddf6ad01 DC |
334 | |
335 | /* Initialize the inode buffers and log them appropriately. */ | |
1813dd64 | 336 | fbuf->b_ops = &xfs_inode_buf_ops; |
93848a99 | 337 | xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length)); |
ef325959 | 338 | for (i = 0; i < M_IGEO(mp)->inodes_per_cluster; i++) { |
85c0b2ab | 339 | int ioffset = i << mp->m_sb.sb_inodelog; |
85c0b2ab DC |
340 | |
341 | free = xfs_make_iptr(mp, fbuf, i); | |
342 | free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC); | |
343 | free->di_version = version; | |
344 | free->di_gen = cpu_to_be32(gen); | |
345 | free->di_next_unlinked = cpu_to_be32(NULLAGINO); | |
93848a99 CH |
346 | |
347 | if (version == 3) { | |
348 | free->di_ino = cpu_to_be64(ino); | |
349 | ino++; | |
ce748eaa ES |
350 | uuid_copy(&free->di_uuid, |
351 | &mp->m_sb.sb_meta_uuid); | |
93848a99 | 352 | xfs_dinode_calc_crc(mp, free); |
28c8e41a | 353 | } else if (tp) { |
93848a99 CH |
354 | /* just log the inode core */ |
355 | xfs_trans_log_buf(tp, fbuf, ioffset, | |
cf28e17c | 356 | ioffset + XFS_DINODE_SIZE(mp) - 1); |
93848a99 CH |
357 | } |
358 | } | |
28c8e41a DC |
359 | |
360 | if (tp) { | |
361 | /* | |
362 | * Mark the buffer as an inode allocation buffer so it | |
363 | * sticks in AIL at the point of this allocation | |
364 | * transaction. This ensures the they are on disk before | |
365 | * the tail of the log can be moved past this | |
366 | * transaction (i.e. by preventing relogging from moving | |
367 | * it forward in the log). | |
368 | */ | |
369 | xfs_trans_inode_alloc_buf(tp, fbuf); | |
370 | if (version == 3) { | |
ddf6ad01 DC |
371 | /* |
372 | * Mark the buffer as ordered so that they are | |
373 | * not physically logged in the transaction but | |
374 | * still tracked in the AIL as part of the | |
375 | * transaction and pin the log appropriately. | |
376 | */ | |
377 | xfs_trans_ordered_buf(tp, fbuf); | |
28c8e41a DC |
378 | } |
379 | } else { | |
380 | fbuf->b_flags |= XBF_DONE; | |
381 | xfs_buf_delwri_queue(fbuf, buffer_list); | |
382 | xfs_buf_relse(fbuf); | |
85c0b2ab | 383 | } |
85c0b2ab | 384 | } |
2a30f36d | 385 | return 0; |
85c0b2ab DC |
386 | } |
387 | ||
56d1115c BF |
388 | /* |
389 | * Align startino and allocmask for a recently allocated sparse chunk such that | |
390 | * they are fit for insertion (or merge) into the on-disk inode btrees. | |
391 | * | |
392 | * Background: | |
393 | * | |
394 | * When enabled, sparse inode support increases the inode alignment from cluster | |
395 | * size to inode chunk size. This means that the minimum range between two | |
396 | * non-adjacent inode records in the inobt is large enough for a full inode | |
397 | * record. This allows for cluster sized, cluster aligned block allocation | |
398 | * without need to worry about whether the resulting inode record overlaps with | |
399 | * another record in the tree. Without this basic rule, we would have to deal | |
400 | * with the consequences of overlap by potentially undoing recent allocations in | |
401 | * the inode allocation codepath. | |
402 | * | |
403 | * Because of this alignment rule (which is enforced on mount), there are two | |
404 | * inobt possibilities for newly allocated sparse chunks. One is that the | |
405 | * aligned inode record for the chunk covers a range of inodes not already | |
406 | * covered in the inobt (i.e., it is safe to insert a new sparse record). The | |
407 | * other is that a record already exists at the aligned startino that considers | |
408 | * the newly allocated range as sparse. In the latter case, record content is | |
409 | * merged in hope that sparse inode chunks fill to full chunks over time. | |
410 | */ | |
411 | STATIC void | |
412 | xfs_align_sparse_ino( | |
413 | struct xfs_mount *mp, | |
414 | xfs_agino_t *startino, | |
415 | uint16_t *allocmask) | |
416 | { | |
417 | xfs_agblock_t agbno; | |
418 | xfs_agblock_t mod; | |
419 | int offset; | |
420 | ||
421 | agbno = XFS_AGINO_TO_AGBNO(mp, *startino); | |
422 | mod = agbno % mp->m_sb.sb_inoalignmt; | |
423 | if (!mod) | |
424 | return; | |
425 | ||
426 | /* calculate the inode offset and align startino */ | |
43004b2a | 427 | offset = XFS_AGB_TO_AGINO(mp, mod); |
56d1115c BF |
428 | *startino -= offset; |
429 | ||
430 | /* | |
431 | * Since startino has been aligned down, left shift allocmask such that | |
432 | * it continues to represent the same physical inodes relative to the | |
433 | * new startino. | |
434 | */ | |
435 | *allocmask <<= offset / XFS_INODES_PER_HOLEMASK_BIT; | |
436 | } | |
437 | ||
438 | /* | |
439 | * Determine whether the source inode record can merge into the target. Both | |
440 | * records must be sparse, the inode ranges must match and there must be no | |
441 | * allocation overlap between the records. | |
442 | */ | |
443 | STATIC bool | |
444 | __xfs_inobt_can_merge( | |
445 | struct xfs_inobt_rec_incore *trec, /* tgt record */ | |
446 | struct xfs_inobt_rec_incore *srec) /* src record */ | |
447 | { | |
448 | uint64_t talloc; | |
449 | uint64_t salloc; | |
450 | ||
451 | /* records must cover the same inode range */ | |
452 | if (trec->ir_startino != srec->ir_startino) | |
453 | return false; | |
454 | ||
455 | /* both records must be sparse */ | |
456 | if (!xfs_inobt_issparse(trec->ir_holemask) || | |
457 | !xfs_inobt_issparse(srec->ir_holemask)) | |
458 | return false; | |
459 | ||
460 | /* both records must track some inodes */ | |
461 | if (!trec->ir_count || !srec->ir_count) | |
462 | return false; | |
463 | ||
464 | /* can't exceed capacity of a full record */ | |
465 | if (trec->ir_count + srec->ir_count > XFS_INODES_PER_CHUNK) | |
466 | return false; | |
467 | ||
468 | /* verify there is no allocation overlap */ | |
469 | talloc = xfs_inobt_irec_to_allocmask(trec); | |
470 | salloc = xfs_inobt_irec_to_allocmask(srec); | |
471 | if (talloc & salloc) | |
472 | return false; | |
473 | ||
474 | return true; | |
475 | } | |
476 | ||
477 | /* | |
478 | * Merge the source inode record into the target. The caller must call | |
479 | * __xfs_inobt_can_merge() to ensure the merge is valid. | |
480 | */ | |
481 | STATIC void | |
482 | __xfs_inobt_rec_merge( | |
483 | struct xfs_inobt_rec_incore *trec, /* target */ | |
484 | struct xfs_inobt_rec_incore *srec) /* src */ | |
485 | { | |
486 | ASSERT(trec->ir_startino == srec->ir_startino); | |
487 | ||
488 | /* combine the counts */ | |
489 | trec->ir_count += srec->ir_count; | |
490 | trec->ir_freecount += srec->ir_freecount; | |
491 | ||
492 | /* | |
493 | * Merge the holemask and free mask. For both fields, 0 bits refer to | |
494 | * allocated inodes. We combine the allocated ranges with bitwise AND. | |
495 | */ | |
496 | trec->ir_holemask &= srec->ir_holemask; | |
497 | trec->ir_free &= srec->ir_free; | |
498 | } | |
499 | ||
500 | /* | |
501 | * Insert a new sparse inode chunk into the associated inode btree. The inode | |
502 | * record for the sparse chunk is pre-aligned to a startino that should match | |
503 | * any pre-existing sparse inode record in the tree. This allows sparse chunks | |
504 | * to fill over time. | |
505 | * | |
506 | * This function supports two modes of handling preexisting records depending on | |
507 | * the merge flag. If merge is true, the provided record is merged with the | |
508 | * existing record and updated in place. The merged record is returned in nrec. | |
509 | * If merge is false, an existing record is replaced with the provided record. | |
510 | * If no preexisting record exists, the provided record is always inserted. | |
511 | * | |
512 | * It is considered corruption if a merge is requested and not possible. Given | |
513 | * the sparse inode alignment constraints, this should never happen. | |
514 | */ | |
515 | STATIC int | |
516 | xfs_inobt_insert_sprec( | |
517 | struct xfs_mount *mp, | |
518 | struct xfs_trans *tp, | |
519 | struct xfs_buf *agbp, | |
7b13c515 | 520 | struct xfs_perag *pag, |
56d1115c BF |
521 | int btnum, |
522 | struct xfs_inobt_rec_incore *nrec, /* in/out: new/merged rec. */ | |
523 | bool merge) /* merge or replace */ | |
524 | { | |
525 | struct xfs_btree_cur *cur; | |
56d1115c BF |
526 | int error; |
527 | int i; | |
528 | struct xfs_inobt_rec_incore rec; | |
529 | ||
7b13c515 | 530 | cur = xfs_inobt_init_cursor(mp, tp, agbp, pag, btnum); |
56d1115c BF |
531 | |
532 | /* the new record is pre-aligned so we know where to look */ | |
533 | error = xfs_inobt_lookup(cur, nrec->ir_startino, XFS_LOOKUP_EQ, &i); | |
534 | if (error) | |
535 | goto error; | |
536 | /* if nothing there, insert a new record and return */ | |
537 | if (i == 0) { | |
538 | error = xfs_inobt_insert_rec(cur, nrec->ir_holemask, | |
539 | nrec->ir_count, nrec->ir_freecount, | |
540 | nrec->ir_free, &i); | |
541 | if (error) | |
542 | goto error; | |
f9e03706 DW |
543 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
544 | error = -EFSCORRUPTED; | |
545 | goto error; | |
546 | } | |
56d1115c BF |
547 | |
548 | goto out; | |
549 | } | |
550 | ||
551 | /* | |
552 | * A record exists at this startino. Merge or replace the record | |
553 | * depending on what we've been asked to do. | |
554 | */ | |
555 | if (merge) { | |
556 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
557 | if (error) | |
558 | goto error; | |
f9e03706 DW |
559 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
560 | error = -EFSCORRUPTED; | |
561 | goto error; | |
562 | } | |
563 | if (XFS_IS_CORRUPT(mp, rec.ir_startino != nrec->ir_startino)) { | |
564 | error = -EFSCORRUPTED; | |
565 | goto error; | |
566 | } | |
56d1115c BF |
567 | |
568 | /* | |
569 | * This should never fail. If we have coexisting records that | |
570 | * cannot merge, something is seriously wrong. | |
571 | */ | |
f9e03706 DW |
572 | if (XFS_IS_CORRUPT(mp, !__xfs_inobt_can_merge(nrec, &rec))) { |
573 | error = -EFSCORRUPTED; | |
574 | goto error; | |
575 | } | |
56d1115c | 576 | |
7b13c515 | 577 | trace_xfs_irec_merge_pre(mp, pag->pag_agno, rec.ir_startino, |
56d1115c BF |
578 | rec.ir_holemask, nrec->ir_startino, |
579 | nrec->ir_holemask); | |
580 | ||
581 | /* merge to nrec to output the updated record */ | |
582 | __xfs_inobt_rec_merge(nrec, &rec); | |
583 | ||
7b13c515 | 584 | trace_xfs_irec_merge_post(mp, pag->pag_agno, nrec->ir_startino, |
56d1115c BF |
585 | nrec->ir_holemask); |
586 | ||
587 | error = xfs_inobt_rec_check_count(mp, nrec); | |
588 | if (error) | |
589 | goto error; | |
590 | } | |
591 | ||
592 | error = xfs_inobt_update(cur, nrec); | |
593 | if (error) | |
594 | goto error; | |
595 | ||
596 | out: | |
597 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
598 | return 0; | |
599 | error: | |
600 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
601 | return error; | |
602 | } | |
603 | ||
1da177e4 | 604 | /* |
8237fbf5 DC |
605 | * Allocate new inodes in the allocation group specified by agbp. Returns 0 if |
606 | * inodes were allocated in this AG; -EAGAIN if there was no space in this AG so | |
607 | * the caller knows it can try another AG, a hard -ENOSPC when over the maximum | |
608 | * inode count threshold, or the usual negative error code for other errors. | |
1da177e4 | 609 | */ |
ef325959 | 610 | STATIC int |
1da177e4 | 611 | xfs_ialloc_ag_alloc( |
ef325959 | 612 | struct xfs_trans *tp, |
7b13c515 DC |
613 | struct xfs_buf *agbp, |
614 | struct xfs_perag *pag) | |
1da177e4 | 615 | { |
ef325959 DW |
616 | struct xfs_agi *agi; |
617 | struct xfs_alloc_arg args; | |
ef325959 DW |
618 | int error; |
619 | xfs_agino_t newino; /* new first inode's number */ | |
620 | xfs_agino_t newlen; /* new number of inodes */ | |
621 | int isaligned = 0; /* inode allocation at stripe */ | |
622 | /* unit boundary */ | |
623 | /* init. to full chunk */ | |
56d1115c | 624 | struct xfs_inobt_rec_incore rec; |
ef325959 | 625 | struct xfs_ino_geometry *igeo = M_IGEO(tp->t_mountp); |
7b13c515 | 626 | uint16_t allocmask = (uint16_t) -1; |
ef325959 | 627 | int do_sparse = 0; |
1cdadee1 | 628 | |
a0041684 | 629 | memset(&args, 0, sizeof(args)); |
1da177e4 LT |
630 | args.tp = tp; |
631 | args.mp = tp->t_mountp; | |
1cdadee1 | 632 | args.fsbno = NULLFSBLOCK; |
7280feda | 633 | args.oinfo = XFS_RMAP_OINFO_INODES; |
1da177e4 | 634 | |
46fc58da BF |
635 | #ifdef DEBUG |
636 | /* randomly do sparse inode allocations */ | |
ebd9027d | 637 | if (xfs_has_sparseinodes(tp->t_mountp) && |
ef325959 | 638 | igeo->ialloc_min_blks < igeo->ialloc_blks) |
46fc58da BF |
639 | do_sparse = prandom_u32() & 1; |
640 | #endif | |
641 | ||
1da177e4 LT |
642 | /* |
643 | * Locking will ensure that we don't have two callers in here | |
644 | * at one time. | |
645 | */ | |
ef325959 DW |
646 | newlen = igeo->ialloc_inos; |
647 | if (igeo->maxicount && | |
74f9ce1c | 648 | percpu_counter_read_positive(&args.mp->m_icount) + newlen > |
ef325959 | 649 | igeo->maxicount) |
2451337d | 650 | return -ENOSPC; |
ef325959 | 651 | args.minlen = args.maxlen = igeo->ialloc_blks; |
1da177e4 | 652 | /* |
3ccb8b5f GO |
653 | * First try to allocate inodes contiguous with the last-allocated |
654 | * chunk of inodes. If the filesystem is striped, this will fill | |
655 | * an entire stripe unit with inodes. | |
28c8e41a | 656 | */ |
370c782b | 657 | agi = agbp->b_addr; |
3ccb8b5f | 658 | newino = be32_to_cpu(agi->agi_newino); |
019ff2d5 | 659 | args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) + |
ef325959 | 660 | igeo->ialloc_blks; |
1cdadee1 BF |
661 | if (do_sparse) |
662 | goto sparse_alloc; | |
019ff2d5 NS |
663 | if (likely(newino != NULLAGINO && |
664 | (args.agbno < be32_to_cpu(agi->agi_length)))) { | |
7b13c515 | 665 | args.fsbno = XFS_AGB_TO_FSB(args.mp, pag->pag_agno, args.agbno); |
3ccb8b5f | 666 | args.type = XFS_ALLOCTYPE_THIS_BNO; |
3ccb8b5f | 667 | args.prod = 1; |
75de2a91 | 668 | |
3ccb8b5f | 669 | /* |
75de2a91 DC |
670 | * We need to take into account alignment here to ensure that |
671 | * we don't modify the free list if we fail to have an exact | |
672 | * block. If we don't have an exact match, and every oher | |
673 | * attempt allocation attempt fails, we'll end up cancelling | |
674 | * a dirty transaction and shutting down. | |
675 | * | |
676 | * For an exact allocation, alignment must be 1, | |
677 | * however we need to take cluster alignment into account when | |
678 | * fixing up the freelist. Use the minalignslop field to | |
679 | * indicate that extra blocks might be required for alignment, | |
680 | * but not to use them in the actual exact allocation. | |
3ccb8b5f | 681 | */ |
75de2a91 | 682 | args.alignment = 1; |
ef325959 | 683 | args.minalignslop = igeo->cluster_align - 1; |
75de2a91 DC |
684 | |
685 | /* Allow space for the inode btree to split. */ | |
657f1019 | 686 | args.minleft = igeo->inobt_maxlevels; |
3ccb8b5f GO |
687 | if ((error = xfs_alloc_vextent(&args))) |
688 | return error; | |
e480a723 BF |
689 | |
690 | /* | |
691 | * This request might have dirtied the transaction if the AG can | |
692 | * satisfy the request, but the exact block was not available. | |
693 | * If the allocation did fail, subsequent requests will relax | |
694 | * the exact agbno requirement and increase the alignment | |
695 | * instead. It is critical that the total size of the request | |
696 | * (len + alignment + slop) does not increase from this point | |
697 | * on, so reset minalignslop to ensure it is not included in | |
698 | * subsequent requests. | |
699 | */ | |
700 | args.minalignslop = 0; | |
1cdadee1 | 701 | } |
1da177e4 | 702 | |
3ccb8b5f GO |
703 | if (unlikely(args.fsbno == NULLFSBLOCK)) { |
704 | /* | |
705 | * Set the alignment for the allocation. | |
706 | * If stripe alignment is turned on then align at stripe unit | |
707 | * boundary. | |
019ff2d5 NS |
708 | * If the cluster size is smaller than a filesystem block |
709 | * then we're doing I/O for inodes in filesystem block size | |
3ccb8b5f GO |
710 | * pieces, so don't need alignment anyway. |
711 | */ | |
712 | isaligned = 0; | |
ef325959 | 713 | if (igeo->ialloc_align) { |
0560f31a | 714 | ASSERT(!xfs_has_noalign(args.mp)); |
3ccb8b5f GO |
715 | args.alignment = args.mp->m_dalign; |
716 | isaligned = 1; | |
75de2a91 | 717 | } else |
ef325959 | 718 | args.alignment = igeo->cluster_align; |
3ccb8b5f GO |
719 | /* |
720 | * Need to figure out where to allocate the inode blocks. | |
721 | * Ideally they should be spaced out through the a.g. | |
722 | * For now, just allocate blocks up front. | |
723 | */ | |
724 | args.agbno = be32_to_cpu(agi->agi_root); | |
7b13c515 | 725 | args.fsbno = XFS_AGB_TO_FSB(args.mp, pag->pag_agno, args.agbno); |
3ccb8b5f GO |
726 | /* |
727 | * Allocate a fixed-size extent of inodes. | |
728 | */ | |
729 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
3ccb8b5f GO |
730 | args.prod = 1; |
731 | /* | |
732 | * Allow space for the inode btree to split. | |
733 | */ | |
657f1019 | 734 | args.minleft = igeo->inobt_maxlevels; |
3ccb8b5f GO |
735 | if ((error = xfs_alloc_vextent(&args))) |
736 | return error; | |
737 | } | |
019ff2d5 | 738 | |
1da177e4 LT |
739 | /* |
740 | * If stripe alignment is turned on, then try again with cluster | |
741 | * alignment. | |
742 | */ | |
743 | if (isaligned && args.fsbno == NULLFSBLOCK) { | |
744 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
16259e7d | 745 | args.agbno = be32_to_cpu(agi->agi_root); |
7b13c515 | 746 | args.fsbno = XFS_AGB_TO_FSB(args.mp, pag->pag_agno, args.agbno); |
ef325959 | 747 | args.alignment = igeo->cluster_align; |
1da177e4 LT |
748 | if ((error = xfs_alloc_vextent(&args))) |
749 | return error; | |
750 | } | |
751 | ||
56d1115c BF |
752 | /* |
753 | * Finally, try a sparse allocation if the filesystem supports it and | |
754 | * the sparse allocation length is smaller than a full chunk. | |
755 | */ | |
ebd9027d | 756 | if (xfs_has_sparseinodes(args.mp) && |
ef325959 | 757 | igeo->ialloc_min_blks < igeo->ialloc_blks && |
56d1115c | 758 | args.fsbno == NULLFSBLOCK) { |
1cdadee1 | 759 | sparse_alloc: |
56d1115c BF |
760 | args.type = XFS_ALLOCTYPE_NEAR_BNO; |
761 | args.agbno = be32_to_cpu(agi->agi_root); | |
7b13c515 | 762 | args.fsbno = XFS_AGB_TO_FSB(args.mp, pag->pag_agno, args.agbno); |
56d1115c BF |
763 | args.alignment = args.mp->m_sb.sb_spino_align; |
764 | args.prod = 1; | |
765 | ||
ef325959 | 766 | args.minlen = igeo->ialloc_min_blks; |
56d1115c BF |
767 | args.maxlen = args.minlen; |
768 | ||
769 | /* | |
770 | * The inode record will be aligned to full chunk size. We must | |
771 | * prevent sparse allocation from AG boundaries that result in | |
772 | * invalid inode records, such as records that start at agbno 0 | |
773 | * or extend beyond the AG. | |
774 | * | |
775 | * Set min agbno to the first aligned, non-zero agbno and max to | |
776 | * the last aligned agbno that is at least one full chunk from | |
777 | * the end of the AG. | |
778 | */ | |
779 | args.min_agbno = args.mp->m_sb.sb_inoalignmt; | |
780 | args.max_agbno = round_down(args.mp->m_sb.sb_agblocks, | |
781 | args.mp->m_sb.sb_inoalignmt) - | |
ef325959 | 782 | igeo->ialloc_blks; |
56d1115c BF |
783 | |
784 | error = xfs_alloc_vextent(&args); | |
785 | if (error) | |
786 | return error; | |
787 | ||
43004b2a | 788 | newlen = XFS_AGB_TO_AGINO(args.mp, args.len); |
46fc58da | 789 | ASSERT(newlen <= XFS_INODES_PER_CHUNK); |
56d1115c BF |
790 | allocmask = (1 << (newlen / XFS_INODES_PER_HOLEMASK_BIT)) - 1; |
791 | } | |
792 | ||
3937493c | 793 | if (args.fsbno == NULLFSBLOCK) |
8237fbf5 | 794 | return -EAGAIN; |
3937493c | 795 | |
1da177e4 | 796 | ASSERT(args.len == args.minlen); |
1da177e4 | 797 | |
359346a9 | 798 | /* |
85c0b2ab DC |
799 | * Stamp and write the inode buffers. |
800 | * | |
359346a9 DC |
801 | * Seed the new inode cluster with a random generation number. This |
802 | * prevents short-term reuse of generation numbers if a chunk is | |
803 | * freed and then immediately reallocated. We use random numbers | |
804 | * rather than a linear progression to prevent the next generation | |
805 | * number from being easily guessable. | |
806 | */ | |
7b13c515 | 807 | error = xfs_ialloc_inode_init(args.mp, tp, NULL, newlen, pag->pag_agno, |
463958af | 808 | args.agbno, args.len, prandom_u32()); |
d42f08f6 | 809 | |
2a30f36d CS |
810 | if (error) |
811 | return error; | |
85c0b2ab DC |
812 | /* |
813 | * Convert the results. | |
814 | */ | |
43004b2a | 815 | newino = XFS_AGB_TO_AGINO(args.mp, args.agbno); |
56d1115c BF |
816 | |
817 | if (xfs_inobt_issparse(~allocmask)) { | |
818 | /* | |
819 | * We've allocated a sparse chunk. Align the startino and mask. | |
820 | */ | |
821 | xfs_align_sparse_ino(args.mp, &newino, &allocmask); | |
822 | ||
823 | rec.ir_startino = newino; | |
824 | rec.ir_holemask = ~allocmask; | |
825 | rec.ir_count = newlen; | |
826 | rec.ir_freecount = newlen; | |
827 | rec.ir_free = XFS_INOBT_ALL_FREE; | |
828 | ||
829 | /* | |
830 | * Insert the sparse record into the inobt and allow for a merge | |
831 | * if necessary. If a merge does occur, rec is updated to the | |
832 | * merged record. | |
833 | */ | |
7b13c515 DC |
834 | error = xfs_inobt_insert_sprec(args.mp, tp, agbp, pag, |
835 | XFS_BTNUM_INO, &rec, true); | |
56d1115c BF |
836 | if (error == -EFSCORRUPTED) { |
837 | xfs_alert(args.mp, | |
838 | "invalid sparse inode record: ino 0x%llx holemask 0x%x count %u", | |
7b13c515 | 839 | XFS_AGINO_TO_INO(args.mp, pag->pag_agno, |
56d1115c BF |
840 | rec.ir_startino), |
841 | rec.ir_holemask, rec.ir_count); | |
842 | xfs_force_shutdown(args.mp, SHUTDOWN_CORRUPT_INCORE); | |
843 | } | |
844 | if (error) | |
845 | return error; | |
846 | ||
847 | /* | |
848 | * We can't merge the part we've just allocated as for the inobt | |
849 | * due to finobt semantics. The original record may or may not | |
850 | * exist independent of whether physical inodes exist in this | |
851 | * sparse chunk. | |
852 | * | |
853 | * We must update the finobt record based on the inobt record. | |
854 | * rec contains the fully merged and up to date inobt record | |
855 | * from the previous call. Set merge false to replace any | |
856 | * existing record with this one. | |
857 | */ | |
ebd9027d | 858 | if (xfs_has_finobt(args.mp)) { |
7b13c515 DC |
859 | error = xfs_inobt_insert_sprec(args.mp, tp, agbp, pag, |
860 | XFS_BTNUM_FINO, &rec, false); | |
56d1115c BF |
861 | if (error) |
862 | return error; | |
863 | } | |
864 | } else { | |
865 | /* full chunk - insert new records to both btrees */ | |
7b13c515 | 866 | error = xfs_inobt_insert(args.mp, tp, agbp, pag, newino, newlen, |
56d1115c BF |
867 | XFS_BTNUM_INO); |
868 | if (error) | |
869 | return error; | |
870 | ||
ebd9027d | 871 | if (xfs_has_finobt(args.mp)) { |
7b13c515 | 872 | error = xfs_inobt_insert(args.mp, tp, agbp, pag, newino, |
56d1115c BF |
873 | newlen, XFS_BTNUM_FINO); |
874 | if (error) | |
875 | return error; | |
876 | } | |
877 | } | |
878 | ||
879 | /* | |
880 | * Update AGI counts and newino. | |
881 | */ | |
413d57c9 MS |
882 | be32_add_cpu(&agi->agi_count, newlen); |
883 | be32_add_cpu(&agi->agi_freecount, newlen); | |
44b56e0a | 884 | pag->pagi_freecount += newlen; |
89e9b5c0 | 885 | pag->pagi_count += newlen; |
16259e7d | 886 | agi->agi_newino = cpu_to_be32(newino); |
85c0b2ab | 887 | |
1da177e4 LT |
888 | /* |
889 | * Log allocation group header fields | |
890 | */ | |
891 | xfs_ialloc_log_agi(tp, agbp, | |
892 | XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO); | |
893 | /* | |
894 | * Modify/log superblock values for inode count and inode free count. | |
895 | */ | |
896 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen); | |
897 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen); | |
1da177e4 LT |
898 | return 0; |
899 | } | |
900 | ||
4254b0bb CH |
901 | /* |
902 | * Try to retrieve the next record to the left/right from the current one. | |
903 | */ | |
904 | STATIC int | |
905 | xfs_ialloc_next_rec( | |
906 | struct xfs_btree_cur *cur, | |
907 | xfs_inobt_rec_incore_t *rec, | |
908 | int *done, | |
909 | int left) | |
910 | { | |
911 | int error; | |
912 | int i; | |
913 | ||
914 | if (left) | |
915 | error = xfs_btree_decrement(cur, 0, &i); | |
916 | else | |
917 | error = xfs_btree_increment(cur, 0, &i); | |
918 | ||
919 | if (error) | |
920 | return error; | |
921 | *done = !i; | |
922 | if (i) { | |
923 | error = xfs_inobt_get_rec(cur, rec, &i); | |
924 | if (error) | |
925 | return error; | |
f9e03706 DW |
926 | if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) |
927 | return -EFSCORRUPTED; | |
4254b0bb CH |
928 | } |
929 | ||
930 | return 0; | |
931 | } | |
932 | ||
bd169565 DC |
933 | STATIC int |
934 | xfs_ialloc_get_rec( | |
935 | struct xfs_btree_cur *cur, | |
936 | xfs_agino_t agino, | |
937 | xfs_inobt_rec_incore_t *rec, | |
43df2ee6 | 938 | int *done) |
bd169565 DC |
939 | { |
940 | int error; | |
941 | int i; | |
942 | ||
943 | error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i); | |
944 | if (error) | |
945 | return error; | |
946 | *done = !i; | |
947 | if (i) { | |
948 | error = xfs_inobt_get_rec(cur, rec, &i); | |
949 | if (error) | |
950 | return error; | |
f9e03706 DW |
951 | if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) |
952 | return -EFSCORRUPTED; | |
bd169565 DC |
953 | } |
954 | ||
955 | return 0; | |
956 | } | |
0b48db80 | 957 | |
d4cc540b | 958 | /* |
26dd5217 BF |
959 | * Return the offset of the first free inode in the record. If the inode chunk |
960 | * is sparsely allocated, we convert the record holemask to inode granularity | |
961 | * and mask off the unallocated regions from the inode free mask. | |
d4cc540b BF |
962 | */ |
963 | STATIC int | |
964 | xfs_inobt_first_free_inode( | |
965 | struct xfs_inobt_rec_incore *rec) | |
966 | { | |
26dd5217 BF |
967 | xfs_inofree_t realfree; |
968 | ||
969 | /* if there are no holes, return the first available offset */ | |
970 | if (!xfs_inobt_issparse(rec->ir_holemask)) | |
971 | return xfs_lowbit64(rec->ir_free); | |
972 | ||
973 | realfree = xfs_inobt_irec_to_allocmask(rec); | |
974 | realfree &= rec->ir_free; | |
975 | ||
976 | return xfs_lowbit64(realfree); | |
d4cc540b BF |
977 | } |
978 | ||
1da177e4 | 979 | /* |
6dd8638e | 980 | * Allocate an inode using the inobt-only algorithm. |
1da177e4 | 981 | */ |
f2ecc5e4 | 982 | STATIC int |
6dd8638e | 983 | xfs_dialloc_ag_inobt( |
f2ecc5e4 CH |
984 | struct xfs_trans *tp, |
985 | struct xfs_buf *agbp, | |
7b13c515 | 986 | struct xfs_perag *pag, |
f2ecc5e4 CH |
987 | xfs_ino_t parent, |
988 | xfs_ino_t *inop) | |
1da177e4 | 989 | { |
f2ecc5e4 | 990 | struct xfs_mount *mp = tp->t_mountp; |
370c782b | 991 | struct xfs_agi *agi = agbp->b_addr; |
f2ecc5e4 CH |
992 | xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent); |
993 | xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent); | |
f2ecc5e4 CH |
994 | struct xfs_btree_cur *cur, *tcur; |
995 | struct xfs_inobt_rec_incore rec, trec; | |
996 | xfs_ino_t ino; | |
997 | int error; | |
998 | int offset; | |
999 | int i, j; | |
2d32311c | 1000 | int searchdistance = 10; |
1da177e4 | 1001 | |
4bb61069 CH |
1002 | ASSERT(pag->pagi_init); |
1003 | ASSERT(pag->pagi_inodeok); | |
1004 | ASSERT(pag->pagi_freecount > 0); | |
1005 | ||
bd169565 | 1006 | restart_pagno: |
7b13c515 | 1007 | cur = xfs_inobt_init_cursor(mp, tp, agbp, pag, XFS_BTNUM_INO); |
1da177e4 LT |
1008 | /* |
1009 | * If pagino is 0 (this is the root inode allocation) use newino. | |
1010 | * This must work because we've just allocated some. | |
1011 | */ | |
1012 | if (!pagino) | |
16259e7d | 1013 | pagino = be32_to_cpu(agi->agi_newino); |
1da177e4 | 1014 | |
9ba0889e | 1015 | error = xfs_check_agi_freecount(cur); |
0b48db80 DC |
1016 | if (error) |
1017 | goto error0; | |
1da177e4 | 1018 | |
1da177e4 | 1019 | /* |
4254b0bb | 1020 | * If in the same AG as the parent, try to get near the parent. |
1da177e4 | 1021 | */ |
7b13c515 | 1022 | if (pagno == pag->pag_agno) { |
4254b0bb CH |
1023 | int doneleft; /* done, to the left */ |
1024 | int doneright; /* done, to the right */ | |
1025 | ||
21875505 | 1026 | error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i); |
4254b0bb | 1027 | if (error) |
1da177e4 | 1028 | goto error0; |
f9e03706 DW |
1029 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
1030 | error = -EFSCORRUPTED; | |
1031 | goto error0; | |
1032 | } | |
4254b0bb CH |
1033 | |
1034 | error = xfs_inobt_get_rec(cur, &rec, &j); | |
1035 | if (error) | |
1036 | goto error0; | |
f9e03706 DW |
1037 | if (XFS_IS_CORRUPT(mp, j != 1)) { |
1038 | error = -EFSCORRUPTED; | |
1039 | goto error0; | |
1040 | } | |
4254b0bb CH |
1041 | |
1042 | if (rec.ir_freecount > 0) { | |
1da177e4 LT |
1043 | /* |
1044 | * Found a free inode in the same chunk | |
4254b0bb | 1045 | * as the parent, done. |
1da177e4 | 1046 | */ |
4254b0bb | 1047 | goto alloc_inode; |
1da177e4 | 1048 | } |
4254b0bb CH |
1049 | |
1050 | ||
1da177e4 | 1051 | /* |
4254b0bb | 1052 | * In the same AG as parent, but parent's chunk is full. |
1da177e4 | 1053 | */ |
1da177e4 | 1054 | |
4254b0bb CH |
1055 | /* duplicate the cursor, search left & right simultaneously */ |
1056 | error = xfs_btree_dup_cursor(cur, &tcur); | |
1057 | if (error) | |
1058 | goto error0; | |
1059 | ||
bd169565 DC |
1060 | /* |
1061 | * Skip to last blocks looked up if same parent inode. | |
1062 | */ | |
1063 | if (pagino != NULLAGINO && | |
1064 | pag->pagl_pagino == pagino && | |
1065 | pag->pagl_leftrec != NULLAGINO && | |
1066 | pag->pagl_rightrec != NULLAGINO) { | |
1067 | error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec, | |
43df2ee6 | 1068 | &trec, &doneleft); |
bd169565 DC |
1069 | if (error) |
1070 | goto error1; | |
4254b0bb | 1071 | |
bd169565 | 1072 | error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec, |
43df2ee6 | 1073 | &rec, &doneright); |
bd169565 DC |
1074 | if (error) |
1075 | goto error1; | |
1076 | } else { | |
1077 | /* search left with tcur, back up 1 record */ | |
1078 | error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1); | |
1079 | if (error) | |
1080 | goto error1; | |
1081 | ||
1082 | /* search right with cur, go forward 1 record. */ | |
1083 | error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0); | |
1084 | if (error) | |
1085 | goto error1; | |
1086 | } | |
4254b0bb CH |
1087 | |
1088 | /* | |
1089 | * Loop until we find an inode chunk with a free inode. | |
1090 | */ | |
2d32311c | 1091 | while (--searchdistance > 0 && (!doneleft || !doneright)) { |
4254b0bb CH |
1092 | int useleft; /* using left inode chunk this time */ |
1093 | ||
1094 | /* figure out the closer block if both are valid. */ | |
1095 | if (!doneleft && !doneright) { | |
1096 | useleft = pagino - | |
1097 | (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) < | |
1098 | rec.ir_startino - pagino; | |
1099 | } else { | |
1100 | useleft = !doneleft; | |
1da177e4 | 1101 | } |
4254b0bb CH |
1102 | |
1103 | /* free inodes to the left? */ | |
1104 | if (useleft && trec.ir_freecount) { | |
4254b0bb CH |
1105 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
1106 | cur = tcur; | |
bd169565 DC |
1107 | |
1108 | pag->pagl_leftrec = trec.ir_startino; | |
1109 | pag->pagl_rightrec = rec.ir_startino; | |
1110 | pag->pagl_pagino = pagino; | |
c44245b3 | 1111 | rec = trec; |
4254b0bb | 1112 | goto alloc_inode; |
1da177e4 | 1113 | } |
1da177e4 | 1114 | |
4254b0bb CH |
1115 | /* free inodes to the right? */ |
1116 | if (!useleft && rec.ir_freecount) { | |
1117 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
bd169565 DC |
1118 | |
1119 | pag->pagl_leftrec = trec.ir_startino; | |
1120 | pag->pagl_rightrec = rec.ir_startino; | |
1121 | pag->pagl_pagino = pagino; | |
4254b0bb | 1122 | goto alloc_inode; |
1da177e4 | 1123 | } |
4254b0bb CH |
1124 | |
1125 | /* get next record to check */ | |
1126 | if (useleft) { | |
1127 | error = xfs_ialloc_next_rec(tcur, &trec, | |
1128 | &doneleft, 1); | |
1129 | } else { | |
1130 | error = xfs_ialloc_next_rec(cur, &rec, | |
1131 | &doneright, 0); | |
1132 | } | |
1133 | if (error) | |
1134 | goto error1; | |
1da177e4 | 1135 | } |
bd169565 | 1136 | |
2d32311c CM |
1137 | if (searchdistance <= 0) { |
1138 | /* | |
1139 | * Not in range - save last search | |
1140 | * location and allocate a new inode | |
1141 | */ | |
1142 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
1143 | pag->pagl_leftrec = trec.ir_startino; | |
1144 | pag->pagl_rightrec = rec.ir_startino; | |
1145 | pag->pagl_pagino = pagino; | |
1146 | ||
1147 | } else { | |
1148 | /* | |
1149 | * We've reached the end of the btree. because | |
1150 | * we are only searching a small chunk of the | |
1151 | * btree each search, there is obviously free | |
1152 | * inodes closer to the parent inode than we | |
1153 | * are now. restart the search again. | |
1154 | */ | |
1155 | pag->pagl_pagino = NULLAGINO; | |
1156 | pag->pagl_leftrec = NULLAGINO; | |
1157 | pag->pagl_rightrec = NULLAGINO; | |
1158 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
1159 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
1160 | goto restart_pagno; | |
1161 | } | |
1da177e4 | 1162 | } |
4254b0bb | 1163 | |
1da177e4 | 1164 | /* |
4254b0bb | 1165 | * In a different AG from the parent. |
1da177e4 LT |
1166 | * See if the most recently allocated block has any free. |
1167 | */ | |
69ef921b | 1168 | if (agi->agi_newino != cpu_to_be32(NULLAGINO)) { |
21875505 CH |
1169 | error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino), |
1170 | XFS_LOOKUP_EQ, &i); | |
4254b0bb | 1171 | if (error) |
1da177e4 | 1172 | goto error0; |
4254b0bb CH |
1173 | |
1174 | if (i == 1) { | |
1175 | error = xfs_inobt_get_rec(cur, &rec, &j); | |
1176 | if (error) | |
1177 | goto error0; | |
1178 | ||
1179 | if (j == 1 && rec.ir_freecount > 0) { | |
1180 | /* | |
1181 | * The last chunk allocated in the group | |
1182 | * still has a free inode. | |
1183 | */ | |
1184 | goto alloc_inode; | |
1185 | } | |
1da177e4 | 1186 | } |
bd169565 | 1187 | } |
4254b0bb | 1188 | |
bd169565 DC |
1189 | /* |
1190 | * None left in the last group, search the whole AG | |
1191 | */ | |
1192 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); | |
1193 | if (error) | |
1194 | goto error0; | |
f9e03706 DW |
1195 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
1196 | error = -EFSCORRUPTED; | |
1197 | goto error0; | |
1198 | } | |
bd169565 DC |
1199 | |
1200 | for (;;) { | |
1201 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
1202 | if (error) | |
1203 | goto error0; | |
f9e03706 DW |
1204 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
1205 | error = -EFSCORRUPTED; | |
1206 | goto error0; | |
1207 | } | |
bd169565 DC |
1208 | if (rec.ir_freecount > 0) |
1209 | break; | |
1210 | error = xfs_btree_increment(cur, 0, &i); | |
4254b0bb CH |
1211 | if (error) |
1212 | goto error0; | |
f9e03706 DW |
1213 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
1214 | error = -EFSCORRUPTED; | |
1215 | goto error0; | |
1216 | } | |
1da177e4 | 1217 | } |
4254b0bb CH |
1218 | |
1219 | alloc_inode: | |
d4cc540b | 1220 | offset = xfs_inobt_first_free_inode(&rec); |
1da177e4 LT |
1221 | ASSERT(offset >= 0); |
1222 | ASSERT(offset < XFS_INODES_PER_CHUNK); | |
1223 | ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) % | |
1224 | XFS_INODES_PER_CHUNK) == 0); | |
7b13c515 | 1225 | ino = XFS_AGINO_TO_INO(mp, pag->pag_agno, rec.ir_startino + offset); |
0d87e656 | 1226 | rec.ir_free &= ~XFS_INOBT_MASK(offset); |
1da177e4 | 1227 | rec.ir_freecount--; |
afabc24a CH |
1228 | error = xfs_inobt_update(cur, &rec); |
1229 | if (error) | |
1da177e4 | 1230 | goto error0; |
413d57c9 | 1231 | be32_add_cpu(&agi->agi_freecount, -1); |
1da177e4 | 1232 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); |
44b56e0a | 1233 | pag->pagi_freecount--; |
1da177e4 | 1234 | |
9ba0889e | 1235 | error = xfs_check_agi_freecount(cur); |
0b48db80 DC |
1236 | if (error) |
1237 | goto error0; | |
1238 | ||
1da177e4 LT |
1239 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
1240 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1); | |
1241 | *inop = ino; | |
1242 | return 0; | |
1243 | error1: | |
1244 | xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR); | |
1245 | error0: | |
1246 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
1247 | return error; | |
1248 | } | |
1249 | ||
6dd8638e BF |
1250 | /* |
1251 | * Use the free inode btree to allocate an inode based on distance from the | |
1252 | * parent. Note that the provided cursor may be deleted and replaced. | |
1253 | */ | |
1254 | STATIC int | |
1255 | xfs_dialloc_ag_finobt_near( | |
1256 | xfs_agino_t pagino, | |
1257 | struct xfs_btree_cur **ocur, | |
1258 | struct xfs_inobt_rec_incore *rec) | |
1259 | { | |
1260 | struct xfs_btree_cur *lcur = *ocur; /* left search cursor */ | |
1261 | struct xfs_btree_cur *rcur; /* right search cursor */ | |
1262 | struct xfs_inobt_rec_incore rrec; | |
1263 | int error; | |
1264 | int i, j; | |
1265 | ||
1266 | error = xfs_inobt_lookup(lcur, pagino, XFS_LOOKUP_LE, &i); | |
1267 | if (error) | |
1268 | return error; | |
1269 | ||
1270 | if (i == 1) { | |
1271 | error = xfs_inobt_get_rec(lcur, rec, &i); | |
1272 | if (error) | |
1273 | return error; | |
f9e03706 DW |
1274 | if (XFS_IS_CORRUPT(lcur->bc_mp, i != 1)) |
1275 | return -EFSCORRUPTED; | |
6dd8638e BF |
1276 | |
1277 | /* | |
1278 | * See if we've landed in the parent inode record. The finobt | |
1279 | * only tracks chunks with at least one free inode, so record | |
1280 | * existence is enough. | |
1281 | */ | |
1282 | if (pagino >= rec->ir_startino && | |
1283 | pagino < (rec->ir_startino + XFS_INODES_PER_CHUNK)) | |
1284 | return 0; | |
1285 | } | |
1286 | ||
1287 | error = xfs_btree_dup_cursor(lcur, &rcur); | |
1288 | if (error) | |
1289 | return error; | |
1290 | ||
1291 | error = xfs_inobt_lookup(rcur, pagino, XFS_LOOKUP_GE, &j); | |
1292 | if (error) | |
1293 | goto error_rcur; | |
1294 | if (j == 1) { | |
1295 | error = xfs_inobt_get_rec(rcur, &rrec, &j); | |
1296 | if (error) | |
1297 | goto error_rcur; | |
f9e03706 DW |
1298 | if (XFS_IS_CORRUPT(lcur->bc_mp, j != 1)) { |
1299 | error = -EFSCORRUPTED; | |
1300 | goto error_rcur; | |
1301 | } | |
6dd8638e BF |
1302 | } |
1303 | ||
f9e03706 DW |
1304 | if (XFS_IS_CORRUPT(lcur->bc_mp, i != 1 && j != 1)) { |
1305 | error = -EFSCORRUPTED; | |
1306 | goto error_rcur; | |
1307 | } | |
6dd8638e BF |
1308 | if (i == 1 && j == 1) { |
1309 | /* | |
1310 | * Both the left and right records are valid. Choose the closer | |
1311 | * inode chunk to the target. | |
1312 | */ | |
1313 | if ((pagino - rec->ir_startino + XFS_INODES_PER_CHUNK - 1) > | |
1314 | (rrec.ir_startino - pagino)) { | |
1315 | *rec = rrec; | |
1316 | xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR); | |
1317 | *ocur = rcur; | |
1318 | } else { | |
1319 | xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR); | |
1320 | } | |
1321 | } else if (j == 1) { | |
1322 | /* only the right record is valid */ | |
1323 | *rec = rrec; | |
1324 | xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR); | |
1325 | *ocur = rcur; | |
1326 | } else if (i == 1) { | |
1327 | /* only the left record is valid */ | |
1328 | xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR); | |
1329 | } | |
1330 | ||
1331 | return 0; | |
1332 | ||
1333 | error_rcur: | |
1334 | xfs_btree_del_cursor(rcur, XFS_BTREE_ERROR); | |
1335 | return error; | |
1336 | } | |
1337 | ||
1338 | /* | |
1339 | * Use the free inode btree to find a free inode based on a newino hint. If | |
1340 | * the hint is NULL, find the first free inode in the AG. | |
1341 | */ | |
1342 | STATIC int | |
1343 | xfs_dialloc_ag_finobt_newino( | |
1344 | struct xfs_agi *agi, | |
1345 | struct xfs_btree_cur *cur, | |
1346 | struct xfs_inobt_rec_incore *rec) | |
1347 | { | |
1348 | int error; | |
1349 | int i; | |
1350 | ||
1351 | if (agi->agi_newino != cpu_to_be32(NULLAGINO)) { | |
e68ed775 DC |
1352 | error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino), |
1353 | XFS_LOOKUP_EQ, &i); | |
6dd8638e BF |
1354 | if (error) |
1355 | return error; | |
1356 | if (i == 1) { | |
1357 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1358 | if (error) | |
1359 | return error; | |
f9e03706 DW |
1360 | if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) |
1361 | return -EFSCORRUPTED; | |
6dd8638e BF |
1362 | return 0; |
1363 | } | |
1364 | } | |
1365 | ||
1366 | /* | |
1367 | * Find the first inode available in the AG. | |
1368 | */ | |
1369 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); | |
1370 | if (error) | |
1371 | return error; | |
f9e03706 DW |
1372 | if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) |
1373 | return -EFSCORRUPTED; | |
6dd8638e BF |
1374 | |
1375 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1376 | if (error) | |
1377 | return error; | |
f9e03706 DW |
1378 | if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) |
1379 | return -EFSCORRUPTED; | |
6dd8638e BF |
1380 | |
1381 | return 0; | |
1382 | } | |
1383 | ||
1384 | /* | |
1385 | * Update the inobt based on a modification made to the finobt. Also ensure that | |
1386 | * the records from both trees are equivalent post-modification. | |
1387 | */ | |
1388 | STATIC int | |
1389 | xfs_dialloc_ag_update_inobt( | |
1390 | struct xfs_btree_cur *cur, /* inobt cursor */ | |
1391 | struct xfs_inobt_rec_incore *frec, /* finobt record */ | |
1392 | int offset) /* inode offset */ | |
1393 | { | |
1394 | struct xfs_inobt_rec_incore rec; | |
1395 | int error; | |
1396 | int i; | |
1397 | ||
1398 | error = xfs_inobt_lookup(cur, frec->ir_startino, XFS_LOOKUP_EQ, &i); | |
1399 | if (error) | |
1400 | return error; | |
f9e03706 DW |
1401 | if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) |
1402 | return -EFSCORRUPTED; | |
6dd8638e BF |
1403 | |
1404 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
1405 | if (error) | |
1406 | return error; | |
f9e03706 DW |
1407 | if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) |
1408 | return -EFSCORRUPTED; | |
6dd8638e BF |
1409 | ASSERT((XFS_AGINO_TO_OFFSET(cur->bc_mp, rec.ir_startino) % |
1410 | XFS_INODES_PER_CHUNK) == 0); | |
1411 | ||
1412 | rec.ir_free &= ~XFS_INOBT_MASK(offset); | |
1413 | rec.ir_freecount--; | |
1414 | ||
f9e03706 DW |
1415 | if (XFS_IS_CORRUPT(cur->bc_mp, |
1416 | rec.ir_free != frec->ir_free || | |
1417 | rec.ir_freecount != frec->ir_freecount)) | |
1418 | return -EFSCORRUPTED; | |
6dd8638e | 1419 | |
b72091f2 | 1420 | return xfs_inobt_update(cur, &rec); |
6dd8638e BF |
1421 | } |
1422 | ||
1423 | /* | |
1424 | * Allocate an inode using the free inode btree, if available. Otherwise, fall | |
1425 | * back to the inobt search algorithm. | |
1426 | * | |
1427 | * The caller selected an AG for us, and made sure that free inodes are | |
1428 | * available. | |
1429 | */ | |
b652afd9 | 1430 | static int |
6dd8638e BF |
1431 | xfs_dialloc_ag( |
1432 | struct xfs_trans *tp, | |
1433 | struct xfs_buf *agbp, | |
309161f6 | 1434 | struct xfs_perag *pag, |
6dd8638e BF |
1435 | xfs_ino_t parent, |
1436 | xfs_ino_t *inop) | |
1437 | { | |
1438 | struct xfs_mount *mp = tp->t_mountp; | |
370c782b | 1439 | struct xfs_agi *agi = agbp->b_addr; |
6dd8638e BF |
1440 | xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent); |
1441 | xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent); | |
6dd8638e BF |
1442 | struct xfs_btree_cur *cur; /* finobt cursor */ |
1443 | struct xfs_btree_cur *icur; /* inobt cursor */ | |
1444 | struct xfs_inobt_rec_incore rec; | |
1445 | xfs_ino_t ino; | |
1446 | int error; | |
1447 | int offset; | |
1448 | int i; | |
1449 | ||
ebd9027d | 1450 | if (!xfs_has_finobt(mp)) |
7b13c515 | 1451 | return xfs_dialloc_ag_inobt(tp, agbp, pag, parent, inop); |
6dd8638e | 1452 | |
6dd8638e BF |
1453 | /* |
1454 | * If pagino is 0 (this is the root inode allocation) use newino. | |
1455 | * This must work because we've just allocated some. | |
1456 | */ | |
1457 | if (!pagino) | |
1458 | pagino = be32_to_cpu(agi->agi_newino); | |
1459 | ||
7b13c515 | 1460 | cur = xfs_inobt_init_cursor(mp, tp, agbp, pag, XFS_BTNUM_FINO); |
6dd8638e | 1461 | |
9ba0889e | 1462 | error = xfs_check_agi_freecount(cur); |
6dd8638e BF |
1463 | if (error) |
1464 | goto error_cur; | |
1465 | ||
1466 | /* | |
1467 | * The search algorithm depends on whether we're in the same AG as the | |
1468 | * parent. If so, find the closest available inode to the parent. If | |
1469 | * not, consider the agi hint or find the first free inode in the AG. | |
1470 | */ | |
7b13c515 | 1471 | if (pag->pag_agno == pagno) |
6dd8638e BF |
1472 | error = xfs_dialloc_ag_finobt_near(pagino, &cur, &rec); |
1473 | else | |
1474 | error = xfs_dialloc_ag_finobt_newino(agi, cur, &rec); | |
1475 | if (error) | |
1476 | goto error_cur; | |
1477 | ||
d4cc540b | 1478 | offset = xfs_inobt_first_free_inode(&rec); |
6dd8638e BF |
1479 | ASSERT(offset >= 0); |
1480 | ASSERT(offset < XFS_INODES_PER_CHUNK); | |
1481 | ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) % | |
1482 | XFS_INODES_PER_CHUNK) == 0); | |
7b13c515 | 1483 | ino = XFS_AGINO_TO_INO(mp, pag->pag_agno, rec.ir_startino + offset); |
6dd8638e BF |
1484 | |
1485 | /* | |
1486 | * Modify or remove the finobt record. | |
1487 | */ | |
1488 | rec.ir_free &= ~XFS_INOBT_MASK(offset); | |
1489 | rec.ir_freecount--; | |
1490 | if (rec.ir_freecount) | |
1491 | error = xfs_inobt_update(cur, &rec); | |
1492 | else | |
1493 | error = xfs_btree_delete(cur, &i); | |
1494 | if (error) | |
1495 | goto error_cur; | |
1496 | ||
1497 | /* | |
1498 | * The finobt has now been updated appropriately. We haven't updated the | |
1499 | * agi and superblock yet, so we can create an inobt cursor and validate | |
1500 | * the original freecount. If all is well, make the equivalent update to | |
1501 | * the inobt using the finobt record and offset information. | |
1502 | */ | |
7b13c515 | 1503 | icur = xfs_inobt_init_cursor(mp, tp, agbp, pag, XFS_BTNUM_INO); |
6dd8638e | 1504 | |
9ba0889e | 1505 | error = xfs_check_agi_freecount(icur); |
6dd8638e BF |
1506 | if (error) |
1507 | goto error_icur; | |
1508 | ||
1509 | error = xfs_dialloc_ag_update_inobt(icur, &rec, offset); | |
1510 | if (error) | |
1511 | goto error_icur; | |
1512 | ||
1513 | /* | |
1514 | * Both trees have now been updated. We must update the perag and | |
1515 | * superblock before we can check the freecount for each btree. | |
1516 | */ | |
1517 | be32_add_cpu(&agi->agi_freecount, -1); | |
1518 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); | |
7b13c515 | 1519 | pag->pagi_freecount--; |
6dd8638e BF |
1520 | |
1521 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1); | |
1522 | ||
9ba0889e | 1523 | error = xfs_check_agi_freecount(icur); |
6dd8638e BF |
1524 | if (error) |
1525 | goto error_icur; | |
9ba0889e | 1526 | error = xfs_check_agi_freecount(cur); |
6dd8638e BF |
1527 | if (error) |
1528 | goto error_icur; | |
1529 | ||
1530 | xfs_btree_del_cursor(icur, XFS_BTREE_NOERROR); | |
1531 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
6dd8638e BF |
1532 | *inop = ino; |
1533 | return 0; | |
1534 | ||
1535 | error_icur: | |
1536 | xfs_btree_del_cursor(icur, XFS_BTREE_ERROR); | |
1537 | error_cur: | |
1538 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
6dd8638e BF |
1539 | return error; |
1540 | } | |
1541 | ||
f3bf6e0f | 1542 | static int |
aececc9f DC |
1543 | xfs_dialloc_roll( |
1544 | struct xfs_trans **tpp, | |
1545 | struct xfs_buf *agibp) | |
1546 | { | |
1547 | struct xfs_trans *tp = *tpp; | |
1548 | struct xfs_dquot_acct *dqinfo; | |
1549 | int error; | |
1550 | ||
1551 | /* | |
1552 | * Hold to on to the agibp across the commit so no other allocation can | |
1553 | * come in and take the free inodes we just allocated for our caller. | |
1554 | */ | |
1555 | xfs_trans_bhold(tp, agibp); | |
1556 | ||
1557 | /* | |
1558 | * We want the quota changes to be associated with the next transaction, | |
1559 | * NOT this one. So, detach the dqinfo from this and attach it to the | |
1560 | * next transaction. | |
1561 | */ | |
1562 | dqinfo = tp->t_dqinfo; | |
1563 | tp->t_dqinfo = NULL; | |
1564 | ||
1565 | error = xfs_trans_roll(&tp); | |
1566 | ||
1567 | /* Re-attach the quota info that we detached from prev trx. */ | |
1568 | tp->t_dqinfo = dqinfo; | |
1569 | ||
8237fbf5 DC |
1570 | /* |
1571 | * Join the buffer even on commit error so that the buffer is released | |
1572 | * when the caller cancels the transaction and doesn't have to handle | |
1573 | * this error case specially. | |
1574 | */ | |
aececc9f | 1575 | xfs_trans_bjoin(tp, agibp); |
8237fbf5 DC |
1576 | *tpp = tp; |
1577 | return error; | |
aececc9f DC |
1578 | } |
1579 | ||
8237fbf5 | 1580 | static xfs_agnumber_t |
89b1f55a DC |
1581 | xfs_ialloc_next_ag( |
1582 | xfs_mount_t *mp) | |
1583 | { | |
1584 | xfs_agnumber_t agno; | |
1585 | ||
1586 | spin_lock(&mp->m_agirotor_lock); | |
1587 | agno = mp->m_agirotor; | |
1588 | if (++mp->m_agirotor >= mp->m_maxagi) | |
1589 | mp->m_agirotor = 0; | |
1590 | spin_unlock(&mp->m_agirotor_lock); | |
1591 | ||
1592 | return agno; | |
1593 | } | |
1594 | ||
8237fbf5 DC |
1595 | static bool |
1596 | xfs_dialloc_good_ag( | |
1597 | struct xfs_trans *tp, | |
1598 | struct xfs_perag *pag, | |
1599 | umode_t mode, | |
1600 | int flags, | |
1601 | bool ok_alloc) | |
1602 | { | |
1603 | struct xfs_mount *mp = tp->t_mountp; | |
1604 | xfs_extlen_t ineed; | |
1605 | xfs_extlen_t longest = 0; | |
1606 | int needspace; | |
1607 | int error; | |
1608 | ||
1609 | if (!pag->pagi_inodeok) | |
1610 | return false; | |
1611 | ||
1612 | if (!pag->pagi_init) { | |
99b13c7f | 1613 | error = xfs_ialloc_read_agi(pag, tp, NULL); |
8237fbf5 DC |
1614 | if (error) |
1615 | return false; | |
1616 | } | |
1617 | ||
1618 | if (pag->pagi_freecount) | |
1619 | return true; | |
1620 | if (!ok_alloc) | |
1621 | return false; | |
1622 | ||
1623 | if (!pag->pagf_init) { | |
08d3e84f | 1624 | error = xfs_alloc_read_agf(pag, tp, flags, NULL); |
8237fbf5 DC |
1625 | if (error) |
1626 | return false; | |
1627 | } | |
1628 | ||
1629 | /* | |
1630 | * Check that there is enough free space for the file plus a chunk of | |
1631 | * inodes if we need to allocate some. If this is the first pass across | |
1632 | * the AGs, take into account the potential space needed for alignment | |
1633 | * of inode chunks when checking the longest contiguous free space in | |
1634 | * the AG - this prevents us from getting ENOSPC because we have free | |
1635 | * space larger than ialloc_blks but alignment constraints prevent us | |
1636 | * from using it. | |
1637 | * | |
1638 | * If we can't find an AG with space for full alignment slack to be | |
1639 | * taken into account, we must be near ENOSPC in all AGs. Hence we | |
1640 | * don't include alignment for the second pass and so if we fail | |
1641 | * allocation due to alignment issues then it is most likely a real | |
1642 | * ENOSPC condition. | |
1643 | * | |
1644 | * XXX(dgc): this calculation is now bogus thanks to the per-ag | |
1645 | * reservations that xfs_alloc_fix_freelist() now does via | |
1646 | * xfs_alloc_space_available(). When the AG fills up, pagf_freeblks will | |
1647 | * be more than large enough for the check below to succeed, but | |
1648 | * xfs_alloc_space_available() will fail because of the non-zero | |
1649 | * metadata reservation and hence we won't actually be able to allocate | |
1650 | * more inodes in this AG. We do soooo much unnecessary work near ENOSPC | |
1651 | * because of this. | |
1652 | */ | |
1653 | ineed = M_IGEO(mp)->ialloc_min_blks; | |
1654 | if (flags && ineed > 1) | |
1655 | ineed += M_IGEO(mp)->cluster_align; | |
1656 | longest = pag->pagf_longest; | |
1657 | if (!longest) | |
1658 | longest = pag->pagf_flcount > 0; | |
1659 | needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode); | |
1660 | ||
1661 | if (pag->pagf_freeblks < needspace + ineed || longest < ineed) | |
1662 | return false; | |
1663 | return true; | |
1664 | } | |
1665 | ||
1666 | static int | |
1667 | xfs_dialloc_try_ag( | |
1668 | struct xfs_trans **tpp, | |
1669 | struct xfs_perag *pag, | |
1670 | xfs_ino_t parent, | |
1671 | xfs_ino_t *new_ino, | |
1672 | bool ok_alloc) | |
1673 | { | |
1674 | struct xfs_buf *agbp; | |
1675 | xfs_ino_t ino; | |
1676 | int error; | |
1677 | ||
1678 | /* | |
1679 | * Then read in the AGI buffer and recheck with the AGI buffer | |
1680 | * lock held. | |
1681 | */ | |
99b13c7f | 1682 | error = xfs_ialloc_read_agi(pag, *tpp, &agbp); |
8237fbf5 DC |
1683 | if (error) |
1684 | return error; | |
1685 | ||
1686 | if (!pag->pagi_freecount) { | |
1687 | if (!ok_alloc) { | |
1688 | error = -EAGAIN; | |
1689 | goto out_release; | |
1690 | } | |
1691 | ||
1692 | error = xfs_ialloc_ag_alloc(*tpp, agbp, pag); | |
1693 | if (error < 0) | |
1694 | goto out_release; | |
1695 | ||
1696 | /* | |
1697 | * We successfully allocated space for an inode cluster in this | |
1698 | * AG. Roll the transaction so that we can allocate one of the | |
1699 | * new inodes. | |
1700 | */ | |
1701 | ASSERT(pag->pagi_freecount > 0); | |
1702 | error = xfs_dialloc_roll(tpp, agbp); | |
1703 | if (error) | |
1704 | goto out_release; | |
1705 | } | |
1706 | ||
1707 | /* Allocate an inode in the found AG */ | |
1708 | error = xfs_dialloc_ag(*tpp, agbp, pag, parent, &ino); | |
1709 | if (!error) | |
1710 | *new_ino = ino; | |
1711 | return error; | |
1712 | ||
1713 | out_release: | |
1714 | xfs_trans_brelse(*tpp, agbp); | |
1715 | return error; | |
1716 | } | |
1717 | ||
f2ecc5e4 | 1718 | /* |
8237fbf5 | 1719 | * Allocate an on-disk inode. |
f2ecc5e4 | 1720 | * |
8d822dc3 | 1721 | * Mode is used to tell whether the new inode is a directory and hence where to |
8237fbf5 DC |
1722 | * locate it. The on-disk inode that is allocated will be returned in @new_ino |
1723 | * on success, otherwise an error will be set to indicate the failure (e.g. | |
1724 | * -ENOSPC). | |
f2ecc5e4 CH |
1725 | */ |
1726 | int | |
b652afd9 | 1727 | xfs_dialloc( |
f3bf6e0f | 1728 | struct xfs_trans **tpp, |
f2ecc5e4 CH |
1729 | xfs_ino_t parent, |
1730 | umode_t mode, | |
b652afd9 | 1731 | xfs_ino_t *new_ino) |
f2ecc5e4 | 1732 | { |
f3bf6e0f | 1733 | struct xfs_mount *mp = (*tpp)->t_mountp; |
f2ecc5e4 | 1734 | xfs_agnumber_t agno; |
b652afd9 | 1735 | int error = 0; |
be60fe54 | 1736 | xfs_agnumber_t start_agno; |
f2ecc5e4 | 1737 | struct xfs_perag *pag; |
ef325959 | 1738 | struct xfs_ino_geometry *igeo = M_IGEO(mp); |
8237fbf5 | 1739 | bool ok_alloc = true; |
89b1f55a | 1740 | int flags; |
b652afd9 | 1741 | xfs_ino_t ino; |
8d822dc3 | 1742 | |
4bb61069 | 1743 | /* |
89b1f55a DC |
1744 | * Directories, symlinks, and regular files frequently allocate at least |
1745 | * one block, so factor that potential expansion when we examine whether | |
1746 | * an AG has enough space for file creation. | |
4bb61069 | 1747 | */ |
89b1f55a DC |
1748 | if (S_ISDIR(mode)) |
1749 | start_agno = xfs_ialloc_next_ag(mp); | |
1750 | else { | |
1751 | start_agno = XFS_INO_TO_AGNO(mp, parent); | |
1752 | if (start_agno >= mp->m_maxagi) | |
1753 | start_agno = 0; | |
1754 | } | |
55d6af64 | 1755 | |
f2ecc5e4 CH |
1756 | /* |
1757 | * If we have already hit the ceiling of inode blocks then clear | |
8237fbf5 | 1758 | * ok_alloc so we scan all available agi structures for a free |
f2ecc5e4 | 1759 | * inode. |
74f9ce1c GW |
1760 | * |
1761 | * Read rough value of mp->m_icount by percpu_counter_read_positive, | |
1762 | * which will sacrifice the preciseness but improve the performance. | |
f2ecc5e4 | 1763 | */ |
ef325959 DW |
1764 | if (igeo->maxicount && |
1765 | percpu_counter_read_positive(&mp->m_icount) + igeo->ialloc_inos | |
1766 | > igeo->maxicount) { | |
8237fbf5 | 1767 | ok_alloc = false; |
f2ecc5e4 CH |
1768 | } |
1769 | ||
1770 | /* | |
1771 | * Loop until we find an allocation group that either has free inodes | |
1772 | * or in which we can allocate some inodes. Iterate through the | |
1773 | * allocation groups upward, wrapping at the end. | |
1774 | */ | |
be60fe54 | 1775 | agno = start_agno; |
89b1f55a | 1776 | flags = XFS_ALLOC_FLAG_TRYLOCK; |
be60fe54 CH |
1777 | for (;;) { |
1778 | pag = xfs_perag_get(mp, agno); | |
8237fbf5 DC |
1779 | if (xfs_dialloc_good_ag(*tpp, pag, mode, flags, ok_alloc)) { |
1780 | error = xfs_dialloc_try_ag(tpp, pag, parent, | |
1781 | &ino, ok_alloc); | |
1782 | if (error != -EAGAIN) | |
42685473 | 1783 | break; |
f2ecc5e4 | 1784 | } |
be60fe54 | 1785 | |
75c8c50f | 1786 | if (xfs_is_shutdown(mp)) { |
89b1f55a | 1787 | error = -EFSCORRUPTED; |
42685473 | 1788 | break; |
89b1f55a DC |
1789 | } |
1790 | if (++agno == mp->m_maxagi) | |
1791 | agno = 0; | |
1792 | if (agno == start_agno) { | |
8237fbf5 DC |
1793 | if (!flags) { |
1794 | error = -ENOSPC; | |
89b1f55a | 1795 | break; |
8237fbf5 | 1796 | } |
89b1f55a DC |
1797 | flags = 0; |
1798 | } | |
42685473 | 1799 | xfs_perag_put(pag); |
f2ecc5e4 CH |
1800 | } |
1801 | ||
8237fbf5 DC |
1802 | if (!error) |
1803 | *new_ino = ino; | |
be60fe54 | 1804 | xfs_perag_put(pag); |
8237fbf5 | 1805 | return error; |
f2ecc5e4 CH |
1806 | } |
1807 | ||
10ae3dc7 BF |
1808 | /* |
1809 | * Free the blocks of an inode chunk. We must consider that the inode chunk | |
1810 | * might be sparse and only free the regions that are allocated as part of the | |
1811 | * chunk. | |
1812 | */ | |
1813 | STATIC void | |
1814 | xfs_difree_inode_chunk( | |
0f37d178 | 1815 | struct xfs_trans *tp, |
10ae3dc7 | 1816 | xfs_agnumber_t agno, |
0f37d178 | 1817 | struct xfs_inobt_rec_incore *rec) |
10ae3dc7 | 1818 | { |
0f37d178 BF |
1819 | struct xfs_mount *mp = tp->t_mountp; |
1820 | xfs_agblock_t sagbno = XFS_AGINO_TO_AGBNO(mp, | |
1821 | rec->ir_startino); | |
1822 | int startidx, endidx; | |
1823 | int nextbit; | |
1824 | xfs_agblock_t agbno; | |
1825 | int contigblk; | |
10ae3dc7 BF |
1826 | DECLARE_BITMAP(holemask, XFS_INOBT_HOLEMASK_BITS); |
1827 | ||
1828 | if (!xfs_inobt_issparse(rec->ir_holemask)) { | |
1829 | /* not sparse, calculate extent info directly */ | |
c201d9ca | 1830 | xfs_free_extent_later(tp, XFS_AGB_TO_FSB(mp, agno, sagbno), |
ef325959 DW |
1831 | M_IGEO(mp)->ialloc_blks, |
1832 | &XFS_RMAP_OINFO_INODES); | |
10ae3dc7 BF |
1833 | return; |
1834 | } | |
1835 | ||
1836 | /* holemask is only 16-bits (fits in an unsigned long) */ | |
1837 | ASSERT(sizeof(rec->ir_holemask) <= sizeof(holemask[0])); | |
1838 | holemask[0] = rec->ir_holemask; | |
1839 | ||
1840 | /* | |
1841 | * Find contiguous ranges of zeroes (i.e., allocated regions) in the | |
1842 | * holemask and convert the start/end index of each range to an extent. | |
1843 | * We start with the start and end index both pointing at the first 0 in | |
1844 | * the mask. | |
1845 | */ | |
1846 | startidx = endidx = find_first_zero_bit(holemask, | |
1847 | XFS_INOBT_HOLEMASK_BITS); | |
1848 | nextbit = startidx + 1; | |
1849 | while (startidx < XFS_INOBT_HOLEMASK_BITS) { | |
1850 | nextbit = find_next_zero_bit(holemask, XFS_INOBT_HOLEMASK_BITS, | |
1851 | nextbit); | |
1852 | /* | |
1853 | * If the next zero bit is contiguous, update the end index of | |
1854 | * the current range and continue. | |
1855 | */ | |
1856 | if (nextbit != XFS_INOBT_HOLEMASK_BITS && | |
1857 | nextbit == endidx + 1) { | |
1858 | endidx = nextbit; | |
1859 | goto next; | |
1860 | } | |
1861 | ||
1862 | /* | |
1863 | * nextbit is not contiguous with the current end index. Convert | |
1864 | * the current start/end to an extent and add it to the free | |
1865 | * list. | |
1866 | */ | |
1867 | agbno = sagbno + (startidx * XFS_INODES_PER_HOLEMASK_BIT) / | |
1868 | mp->m_sb.sb_inopblock; | |
1869 | contigblk = ((endidx - startidx + 1) * | |
1870 | XFS_INODES_PER_HOLEMASK_BIT) / | |
1871 | mp->m_sb.sb_inopblock; | |
1872 | ||
1873 | ASSERT(agbno % mp->m_sb.sb_spino_align == 0); | |
1874 | ASSERT(contigblk % mp->m_sb.sb_spino_align == 0); | |
c201d9ca | 1875 | xfs_free_extent_later(tp, XFS_AGB_TO_FSB(mp, agno, agbno), |
7280feda | 1876 | contigblk, &XFS_RMAP_OINFO_INODES); |
10ae3dc7 BF |
1877 | |
1878 | /* reset range to current bit and carry on... */ | |
1879 | startidx = endidx = nextbit; | |
1880 | ||
1881 | next: | |
1882 | nextbit++; | |
1883 | } | |
1884 | } | |
1885 | ||
2b64ee5c BF |
1886 | STATIC int |
1887 | xfs_difree_inobt( | |
1888 | struct xfs_mount *mp, | |
1889 | struct xfs_trans *tp, | |
1890 | struct xfs_buf *agbp, | |
7b13c515 | 1891 | struct xfs_perag *pag, |
2b64ee5c | 1892 | xfs_agino_t agino, |
09b56604 | 1893 | struct xfs_icluster *xic, |
2b64ee5c | 1894 | struct xfs_inobt_rec_incore *orec) |
1da177e4 | 1895 | { |
370c782b | 1896 | struct xfs_agi *agi = agbp->b_addr; |
2b64ee5c BF |
1897 | struct xfs_btree_cur *cur; |
1898 | struct xfs_inobt_rec_incore rec; | |
1899 | int ilen; | |
1900 | int error; | |
1901 | int i; | |
1902 | int off; | |
1da177e4 | 1903 | |
69ef921b | 1904 | ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC)); |
2b64ee5c BF |
1905 | ASSERT(XFS_AGINO_TO_AGBNO(mp, agino) < be32_to_cpu(agi->agi_length)); |
1906 | ||
1da177e4 LT |
1907 | /* |
1908 | * Initialize the cursor. | |
1909 | */ | |
7b13c515 | 1910 | cur = xfs_inobt_init_cursor(mp, tp, agbp, pag, XFS_BTNUM_INO); |
1da177e4 | 1911 | |
9ba0889e | 1912 | error = xfs_check_agi_freecount(cur); |
0b48db80 DC |
1913 | if (error) |
1914 | goto error0; | |
1915 | ||
1da177e4 LT |
1916 | /* |
1917 | * Look for the entry describing this inode. | |
1918 | */ | |
21875505 | 1919 | if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) { |
0b932ccc DC |
1920 | xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.", |
1921 | __func__, error); | |
1da177e4 LT |
1922 | goto error0; |
1923 | } | |
f9e03706 DW |
1924 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
1925 | error = -EFSCORRUPTED; | |
1926 | goto error0; | |
1927 | } | |
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 | } | |
f9e03706 DW |
1934 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
1935 | error = -EFSCORRUPTED; | |
1936 | goto error0; | |
1937 | } | |
1da177e4 LT |
1938 | /* |
1939 | * Get the offset in the inode chunk. | |
1940 | */ | |
1941 | off = agino - rec.ir_startino; | |
1942 | ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK); | |
0d87e656 | 1943 | ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off))); |
1da177e4 LT |
1944 | /* |
1945 | * Mark the inode free & increment the count. | |
1946 | */ | |
0d87e656 | 1947 | rec.ir_free |= XFS_INOBT_MASK(off); |
1da177e4 LT |
1948 | rec.ir_freecount++; |
1949 | ||
1950 | /* | |
999633d3 BF |
1951 | * When an inode chunk is free, it becomes eligible for removal. Don't |
1952 | * remove the chunk if the block size is large enough for multiple inode | |
1953 | * chunks (that might not be free). | |
1da177e4 | 1954 | */ |
0560f31a | 1955 | if (!xfs_has_ikeep(mp) && rec.ir_free == XFS_INOBT_ALL_FREE && |
999633d3 | 1956 | mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK) { |
92a00544 GX |
1957 | struct xfs_perag *pag = agbp->b_pag; |
1958 | ||
749f24f3 | 1959 | xic->deleted = true; |
7b13c515 DC |
1960 | xic->first_ino = XFS_AGINO_TO_INO(mp, pag->pag_agno, |
1961 | rec.ir_startino); | |
09b56604 | 1962 | xic->alloc = xfs_inobt_irec_to_allocmask(&rec); |
1da177e4 LT |
1963 | |
1964 | /* | |
1965 | * Remove the inode cluster from the AGI B+Tree, adjust the | |
1966 | * AGI and Superblock inode counts, and mark the disk space | |
1967 | * to be freed when the transaction is committed. | |
1968 | */ | |
999633d3 | 1969 | ilen = rec.ir_freecount; |
413d57c9 MS |
1970 | be32_add_cpu(&agi->agi_count, -ilen); |
1971 | be32_add_cpu(&agi->agi_freecount, -(ilen - 1)); | |
1da177e4 | 1972 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT); |
44b56e0a | 1973 | pag->pagi_freecount -= ilen - 1; |
89e9b5c0 | 1974 | pag->pagi_count -= ilen; |
1da177e4 LT |
1975 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen); |
1976 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1)); | |
1977 | ||
91cca5df | 1978 | if ((error = xfs_btree_delete(cur, &i))) { |
0b932ccc DC |
1979 | xfs_warn(mp, "%s: xfs_btree_delete returned error %d.", |
1980 | __func__, error); | |
1da177e4 LT |
1981 | goto error0; |
1982 | } | |
1983 | ||
7b13c515 | 1984 | xfs_difree_inode_chunk(tp, pag->pag_agno, &rec); |
1da177e4 | 1985 | } else { |
749f24f3 | 1986 | xic->deleted = false; |
1da177e4 | 1987 | |
afabc24a CH |
1988 | error = xfs_inobt_update(cur, &rec); |
1989 | if (error) { | |
0b932ccc DC |
1990 | xfs_warn(mp, "%s: xfs_inobt_update returned error %d.", |
1991 | __func__, error); | |
1da177e4 LT |
1992 | goto error0; |
1993 | } | |
afabc24a | 1994 | |
b7df7630 | 1995 | /* |
1da177e4 LT |
1996 | * Change the inode free counts and log the ag/sb changes. |
1997 | */ | |
413d57c9 | 1998 | be32_add_cpu(&agi->agi_freecount, 1); |
1da177e4 | 1999 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); |
7b13c515 | 2000 | pag->pagi_freecount++; |
1da177e4 LT |
2001 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1); |
2002 | } | |
2003 | ||
9ba0889e | 2004 | error = xfs_check_agi_freecount(cur); |
0b48db80 DC |
2005 | if (error) |
2006 | goto error0; | |
1da177e4 | 2007 | |
2b64ee5c | 2008 | *orec = rec; |
1da177e4 LT |
2009 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
2010 | return 0; | |
2011 | ||
2012 | error0: | |
2013 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
2014 | return error; | |
2015 | } | |
2016 | ||
3efa4ffd BF |
2017 | /* |
2018 | * Free an inode in the free inode btree. | |
2019 | */ | |
2020 | STATIC int | |
2021 | xfs_difree_finobt( | |
2022 | struct xfs_mount *mp, | |
2023 | struct xfs_trans *tp, | |
2024 | struct xfs_buf *agbp, | |
7b13c515 | 2025 | struct xfs_perag *pag, |
3efa4ffd BF |
2026 | xfs_agino_t agino, |
2027 | struct xfs_inobt_rec_incore *ibtrec) /* inobt record */ | |
2028 | { | |
3efa4ffd BF |
2029 | struct xfs_btree_cur *cur; |
2030 | struct xfs_inobt_rec_incore rec; | |
2031 | int offset = agino - ibtrec->ir_startino; | |
2032 | int error; | |
2033 | int i; | |
2034 | ||
7b13c515 | 2035 | cur = xfs_inobt_init_cursor(mp, tp, agbp, pag, XFS_BTNUM_FINO); |
3efa4ffd BF |
2036 | |
2037 | error = xfs_inobt_lookup(cur, ibtrec->ir_startino, XFS_LOOKUP_EQ, &i); | |
2038 | if (error) | |
2039 | goto error; | |
2040 | if (i == 0) { | |
2041 | /* | |
2042 | * If the record does not exist in the finobt, we must have just | |
2043 | * freed an inode in a previously fully allocated chunk. If not, | |
2044 | * something is out of sync. | |
2045 | */ | |
f9e03706 DW |
2046 | if (XFS_IS_CORRUPT(mp, ibtrec->ir_freecount != 1)) { |
2047 | error = -EFSCORRUPTED; | |
2048 | goto error; | |
2049 | } | |
3efa4ffd | 2050 | |
5419040f BF |
2051 | error = xfs_inobt_insert_rec(cur, ibtrec->ir_holemask, |
2052 | ibtrec->ir_count, | |
2053 | ibtrec->ir_freecount, | |
3efa4ffd BF |
2054 | ibtrec->ir_free, &i); |
2055 | if (error) | |
2056 | goto error; | |
2057 | ASSERT(i == 1); | |
2058 | ||
2059 | goto out; | |
2060 | } | |
2061 | ||
2062 | /* | |
2063 | * Read and update the existing record. We could just copy the ibtrec | |
2064 | * across here, but that would defeat the purpose of having redundant | |
2065 | * metadata. By making the modifications independently, we can catch | |
2066 | * corruptions that we wouldn't see if we just copied from one record | |
2067 | * to another. | |
2068 | */ | |
2069 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
2070 | if (error) | |
2071 | goto error; | |
f9e03706 DW |
2072 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2073 | error = -EFSCORRUPTED; | |
2074 | goto error; | |
2075 | } | |
3efa4ffd BF |
2076 | |
2077 | rec.ir_free |= XFS_INOBT_MASK(offset); | |
2078 | rec.ir_freecount++; | |
2079 | ||
f9e03706 DW |
2080 | if (XFS_IS_CORRUPT(mp, |
2081 | rec.ir_free != ibtrec->ir_free || | |
2082 | rec.ir_freecount != ibtrec->ir_freecount)) { | |
2083 | error = -EFSCORRUPTED; | |
2084 | goto error; | |
2085 | } | |
3efa4ffd BF |
2086 | |
2087 | /* | |
2088 | * The content of inobt records should always match between the inobt | |
2089 | * and finobt. The lifecycle of records in the finobt is different from | |
2090 | * the inobt in that the finobt only tracks records with at least one | |
2091 | * free inode. Hence, if all of the inodes are free and we aren't | |
2092 | * keeping inode chunks permanently on disk, remove the record. | |
2093 | * Otherwise, update the record with the new information. | |
999633d3 BF |
2094 | * |
2095 | * Note that we currently can't free chunks when the block size is large | |
2096 | * enough for multiple chunks. Leave the finobt record to remain in sync | |
2097 | * with the inobt. | |
3efa4ffd | 2098 | */ |
0560f31a DC |
2099 | if (!xfs_has_ikeep(mp) && rec.ir_free == XFS_INOBT_ALL_FREE && |
2100 | mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK) { | |
3efa4ffd BF |
2101 | error = xfs_btree_delete(cur, &i); |
2102 | if (error) | |
2103 | goto error; | |
2104 | ASSERT(i == 1); | |
2105 | } else { | |
2106 | error = xfs_inobt_update(cur, &rec); | |
2107 | if (error) | |
2108 | goto error; | |
2109 | } | |
2110 | ||
2111 | out: | |
9ba0889e | 2112 | error = xfs_check_agi_freecount(cur); |
3efa4ffd BF |
2113 | if (error) |
2114 | goto error; | |
2115 | ||
2116 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
2117 | return 0; | |
2118 | ||
2119 | error: | |
2120 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
2121 | return error; | |
2122 | } | |
2123 | ||
2b64ee5c BF |
2124 | /* |
2125 | * Free disk inode. Carefully avoids touching the incore inode, all | |
2126 | * manipulations incore are the caller's responsibility. | |
2127 | * The on-disk inode is not changed by this operation, only the | |
2128 | * btree (free inode mask) is changed. | |
2129 | */ | |
2130 | int | |
2131 | xfs_difree( | |
f40aadb2 DC |
2132 | struct xfs_trans *tp, |
2133 | struct xfs_perag *pag, | |
2134 | xfs_ino_t inode, | |
2135 | struct xfs_icluster *xic) | |
2b64ee5c BF |
2136 | { |
2137 | /* REFERENCED */ | |
2138 | xfs_agblock_t agbno; /* block number containing inode */ | |
2139 | struct xfs_buf *agbp; /* buffer for allocation group header */ | |
2140 | xfs_agino_t agino; /* allocation group inode number */ | |
2b64ee5c | 2141 | int error; /* error return value */ |
7b13c515 | 2142 | struct xfs_mount *mp = tp->t_mountp; |
2b64ee5c | 2143 | struct xfs_inobt_rec_incore rec;/* btree record */ |
2b64ee5c BF |
2144 | |
2145 | /* | |
2146 | * Break up inode number into its components. | |
2147 | */ | |
f40aadb2 DC |
2148 | if (pag->pag_agno != XFS_INO_TO_AGNO(mp, inode)) { |
2149 | xfs_warn(mp, "%s: agno != pag->pag_agno (%d != %d).", | |
2150 | __func__, XFS_INO_TO_AGNO(mp, inode), pag->pag_agno); | |
2b64ee5c | 2151 | ASSERT(0); |
2451337d | 2152 | return -EINVAL; |
2b64ee5c BF |
2153 | } |
2154 | agino = XFS_INO_TO_AGINO(mp, inode); | |
f40aadb2 | 2155 | if (inode != XFS_AGINO_TO_INO(mp, pag->pag_agno, agino)) { |
2b64ee5c BF |
2156 | xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).", |
2157 | __func__, (unsigned long long)inode, | |
f40aadb2 | 2158 | (unsigned long long)XFS_AGINO_TO_INO(mp, pag->pag_agno, agino)); |
2b64ee5c | 2159 | ASSERT(0); |
2451337d | 2160 | return -EINVAL; |
2b64ee5c BF |
2161 | } |
2162 | agbno = XFS_AGINO_TO_AGBNO(mp, agino); | |
2163 | if (agbno >= mp->m_sb.sb_agblocks) { | |
2164 | xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).", | |
2165 | __func__, agbno, mp->m_sb.sb_agblocks); | |
2166 | ASSERT(0); | |
2451337d | 2167 | return -EINVAL; |
2b64ee5c BF |
2168 | } |
2169 | /* | |
2170 | * Get the allocation group header. | |
2171 | */ | |
99b13c7f | 2172 | error = xfs_ialloc_read_agi(pag, tp, &agbp); |
2b64ee5c BF |
2173 | if (error) { |
2174 | xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.", | |
2175 | __func__, error); | |
2176 | return error; | |
2177 | } | |
2178 | ||
2179 | /* | |
2180 | * Fix up the inode allocation btree. | |
2181 | */ | |
7b13c515 | 2182 | error = xfs_difree_inobt(mp, tp, agbp, pag, agino, xic, &rec); |
2b64ee5c BF |
2183 | if (error) |
2184 | goto error0; | |
2185 | ||
3efa4ffd BF |
2186 | /* |
2187 | * Fix up the free inode btree. | |
2188 | */ | |
ebd9027d | 2189 | if (xfs_has_finobt(mp)) { |
7b13c515 | 2190 | error = xfs_difree_finobt(mp, tp, agbp, pag, agino, &rec); |
3efa4ffd BF |
2191 | if (error) |
2192 | goto error0; | |
2193 | } | |
2194 | ||
2b64ee5c BF |
2195 | return 0; |
2196 | ||
2197 | error0: | |
2198 | return error; | |
2199 | } | |
2200 | ||
7124fe0a DC |
2201 | STATIC int |
2202 | xfs_imap_lookup( | |
2203 | struct xfs_mount *mp, | |
2204 | struct xfs_trans *tp, | |
7b13c515 | 2205 | struct xfs_perag *pag, |
7124fe0a DC |
2206 | xfs_agino_t agino, |
2207 | xfs_agblock_t agbno, | |
2208 | xfs_agblock_t *chunk_agbno, | |
2209 | xfs_agblock_t *offset_agbno, | |
2210 | int flags) | |
2211 | { | |
2212 | struct xfs_inobt_rec_incore rec; | |
2213 | struct xfs_btree_cur *cur; | |
2214 | struct xfs_buf *agbp; | |
7124fe0a DC |
2215 | int error; |
2216 | int i; | |
2217 | ||
99b13c7f | 2218 | error = xfs_ialloc_read_agi(pag, tp, &agbp); |
7124fe0a | 2219 | if (error) { |
53487786 DC |
2220 | xfs_alert(mp, |
2221 | "%s: xfs_ialloc_read_agi() returned error %d, agno %d", | |
7b13c515 | 2222 | __func__, error, pag->pag_agno); |
7124fe0a DC |
2223 | return error; |
2224 | } | |
2225 | ||
2226 | /* | |
4536f2ad DC |
2227 | * Lookup the inode record for the given agino. If the record cannot be |
2228 | * found, then it's an invalid inode number and we should abort. Once | |
2229 | * we have a record, we need to ensure it contains the inode number | |
2230 | * we are looking up. | |
7124fe0a | 2231 | */ |
7b13c515 | 2232 | cur = xfs_inobt_init_cursor(mp, tp, agbp, pag, XFS_BTNUM_INO); |
4536f2ad | 2233 | error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i); |
7124fe0a DC |
2234 | if (!error) { |
2235 | if (i) | |
2236 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
2237 | if (!error && i == 0) | |
2451337d | 2238 | error = -EINVAL; |
7124fe0a DC |
2239 | } |
2240 | ||
2241 | xfs_trans_brelse(tp, agbp); | |
0b04b6b8 | 2242 | xfs_btree_del_cursor(cur, error); |
7124fe0a DC |
2243 | if (error) |
2244 | return error; | |
2245 | ||
4536f2ad DC |
2246 | /* check that the returned record contains the required inode */ |
2247 | if (rec.ir_startino > agino || | |
ef325959 | 2248 | rec.ir_startino + M_IGEO(mp)->ialloc_inos <= agino) |
2451337d | 2249 | return -EINVAL; |
4536f2ad | 2250 | |
7124fe0a | 2251 | /* for untrusted inodes check it is allocated first */ |
1920779e | 2252 | if ((flags & XFS_IGET_UNTRUSTED) && |
7124fe0a | 2253 | (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino))) |
2451337d | 2254 | return -EINVAL; |
7124fe0a DC |
2255 | |
2256 | *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino); | |
2257 | *offset_agbno = agbno - *chunk_agbno; | |
2258 | return 0; | |
2259 | } | |
2260 | ||
1da177e4 | 2261 | /* |
94e1b69d | 2262 | * Return the location of the inode in imap, for mapping it into a buffer. |
1da177e4 | 2263 | */ |
1da177e4 | 2264 | int |
94e1b69d | 2265 | xfs_imap( |
7b13c515 DC |
2266 | struct xfs_mount *mp, /* file system mount structure */ |
2267 | struct xfs_trans *tp, /* transaction pointer */ | |
2268 | xfs_ino_t ino, /* inode to locate */ | |
2269 | struct xfs_imap *imap, /* location map structure */ | |
2270 | uint flags) /* flags for inode btree lookup */ | |
1da177e4 | 2271 | { |
7b13c515 DC |
2272 | xfs_agblock_t agbno; /* block number of inode in the alloc group */ |
2273 | xfs_agino_t agino; /* inode number within alloc group */ | |
2274 | xfs_agblock_t chunk_agbno; /* first block in inode chunk */ | |
2275 | xfs_agblock_t cluster_agbno; /* first block in inode cluster */ | |
2276 | int error; /* error code */ | |
2277 | int offset; /* index of inode in its buffer */ | |
2278 | xfs_agblock_t offset_agbno; /* blks from chunk start to inode */ | |
2279 | struct xfs_perag *pag; | |
1da177e4 LT |
2280 | |
2281 | ASSERT(ino != NULLFSINO); | |
94e1b69d | 2282 | |
1da177e4 LT |
2283 | /* |
2284 | * Split up the inode number into its parts. | |
2285 | */ | |
7b13c515 | 2286 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ino)); |
1da177e4 LT |
2287 | agino = XFS_INO_TO_AGINO(mp, ino); |
2288 | agbno = XFS_AGINO_TO_AGBNO(mp, agino); | |
7b13c515 DC |
2289 | if (!pag || agbno >= mp->m_sb.sb_agblocks || |
2290 | ino != XFS_AGINO_TO_INO(mp, pag->pag_agno, agino)) { | |
2291 | error = -EINVAL; | |
1da177e4 | 2292 | #ifdef DEBUG |
1920779e DC |
2293 | /* |
2294 | * Don't output diagnostic information for untrusted inodes | |
2295 | * as they can be invalid without implying corruption. | |
2296 | */ | |
2297 | if (flags & XFS_IGET_UNTRUSTED) | |
7b13c515 DC |
2298 | goto out_drop; |
2299 | if (!pag) { | |
53487786 DC |
2300 | xfs_alert(mp, |
2301 | "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)", | |
7b13c515 DC |
2302 | __func__, XFS_INO_TO_AGNO(mp, ino), |
2303 | mp->m_sb.sb_agcount); | |
1da177e4 LT |
2304 | } |
2305 | if (agbno >= mp->m_sb.sb_agblocks) { | |
53487786 DC |
2306 | xfs_alert(mp, |
2307 | "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)", | |
2308 | __func__, (unsigned long long)agbno, | |
2309 | (unsigned long)mp->m_sb.sb_agblocks); | |
1da177e4 | 2310 | } |
7b13c515 | 2311 | if (pag && ino != XFS_AGINO_TO_INO(mp, pag->pag_agno, agino)) { |
53487786 DC |
2312 | xfs_alert(mp, |
2313 | "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)", | |
2314 | __func__, ino, | |
7b13c515 | 2315 | XFS_AGINO_TO_INO(mp, pag->pag_agno, agino)); |
1da177e4 | 2316 | } |
745b1f47 | 2317 | xfs_stack_trace(); |
1da177e4 | 2318 | #endif /* DEBUG */ |
7b13c515 | 2319 | goto out_drop; |
1da177e4 | 2320 | } |
94e1b69d | 2321 | |
7124fe0a DC |
2322 | /* |
2323 | * For bulkstat and handle lookups, we have an untrusted inode number | |
2324 | * that we have to verify is valid. We cannot do this just by reading | |
2325 | * the inode buffer as it may have been unlinked and removed leaving | |
2326 | * inodes in stale state on disk. Hence we have to do a btree lookup | |
2327 | * in all cases where an untrusted inode number is passed. | |
2328 | */ | |
1920779e | 2329 | if (flags & XFS_IGET_UNTRUSTED) { |
7b13c515 | 2330 | error = xfs_imap_lookup(mp, tp, pag, agino, agbno, |
7124fe0a DC |
2331 | &chunk_agbno, &offset_agbno, flags); |
2332 | if (error) | |
7b13c515 | 2333 | goto out_drop; |
7124fe0a DC |
2334 | goto out_map; |
2335 | } | |
2336 | ||
94e1b69d CH |
2337 | /* |
2338 | * If the inode cluster size is the same as the blocksize or | |
2339 | * smaller we get to the buffer by simple arithmetics. | |
2340 | */ | |
ef325959 | 2341 | if (M_IGEO(mp)->blocks_per_cluster == 1) { |
1da177e4 LT |
2342 | offset = XFS_INO_TO_OFFSET(mp, ino); |
2343 | ASSERT(offset < mp->m_sb.sb_inopblock); | |
94e1b69d | 2344 | |
7b13c515 | 2345 | imap->im_blkno = XFS_AGB_TO_DADDR(mp, pag->pag_agno, agbno); |
94e1b69d | 2346 | imap->im_len = XFS_FSB_TO_BB(mp, 1); |
755c7bf5 DW |
2347 | imap->im_boffset = (unsigned short)(offset << |
2348 | mp->m_sb.sb_inodelog); | |
7b13c515 DC |
2349 | error = 0; |
2350 | goto out_drop; | |
1da177e4 | 2351 | } |
94e1b69d | 2352 | |
94e1b69d CH |
2353 | /* |
2354 | * If the inode chunks are aligned then use simple maths to | |
2355 | * find the location. Otherwise we have to do a btree | |
2356 | * lookup to find the location. | |
2357 | */ | |
ef325959 DW |
2358 | if (M_IGEO(mp)->inoalign_mask) { |
2359 | offset_agbno = agbno & M_IGEO(mp)->inoalign_mask; | |
1da177e4 LT |
2360 | chunk_agbno = agbno - offset_agbno; |
2361 | } else { | |
7b13c515 | 2362 | error = xfs_imap_lookup(mp, tp, pag, agino, agbno, |
7124fe0a | 2363 | &chunk_agbno, &offset_agbno, flags); |
1da177e4 | 2364 | if (error) |
7b13c515 | 2365 | goto out_drop; |
1da177e4 | 2366 | } |
94e1b69d | 2367 | |
7124fe0a | 2368 | out_map: |
1da177e4 LT |
2369 | ASSERT(agbno >= chunk_agbno); |
2370 | cluster_agbno = chunk_agbno + | |
ef325959 DW |
2371 | ((offset_agbno / M_IGEO(mp)->blocks_per_cluster) * |
2372 | M_IGEO(mp)->blocks_per_cluster); | |
1da177e4 LT |
2373 | offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) + |
2374 | XFS_INO_TO_OFFSET(mp, ino); | |
94e1b69d | 2375 | |
7b13c515 | 2376 | imap->im_blkno = XFS_AGB_TO_DADDR(mp, pag->pag_agno, cluster_agbno); |
ef325959 | 2377 | imap->im_len = XFS_FSB_TO_BB(mp, M_IGEO(mp)->blocks_per_cluster); |
755c7bf5 | 2378 | imap->im_boffset = (unsigned short)(offset << mp->m_sb.sb_inodelog); |
94e1b69d CH |
2379 | |
2380 | /* | |
2381 | * If the inode number maps to a block outside the bounds | |
2382 | * of the file system then return NULL rather than calling | |
2383 | * read_buf and panicing when we get an error from the | |
2384 | * driver. | |
2385 | */ | |
2386 | if ((imap->im_blkno + imap->im_len) > | |
2387 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) { | |
53487786 DC |
2388 | xfs_alert(mp, |
2389 | "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)", | |
2390 | __func__, (unsigned long long) imap->im_blkno, | |
94e1b69d CH |
2391 | (unsigned long long) imap->im_len, |
2392 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)); | |
7b13c515 DC |
2393 | error = -EINVAL; |
2394 | goto out_drop; | |
94e1b69d | 2395 | } |
7b13c515 DC |
2396 | error = 0; |
2397 | out_drop: | |
90e2c1c2 DC |
2398 | if (pag) |
2399 | xfs_perag_put(pag); | |
7b13c515 | 2400 | return error; |
1da177e4 LT |
2401 | } |
2402 | ||
1da177e4 | 2403 | /* |
aafc3c24 BF |
2404 | * Log specified fields for the ag hdr (inode section). The growth of the agi |
2405 | * structure over time requires that we interpret the buffer as two logical | |
2406 | * regions delineated by the end of the unlinked list. This is due to the size | |
2407 | * of the hash table and its location in the middle of the agi. | |
2408 | * | |
2409 | * For example, a request to log a field before agi_unlinked and a field after | |
2410 | * agi_unlinked could cause us to log the entire hash table and use an excessive | |
2411 | * amount of log space. To avoid this behavior, log the region up through | |
2412 | * agi_unlinked in one call and the region after agi_unlinked through the end of | |
2413 | * the structure in another. | |
1da177e4 LT |
2414 | */ |
2415 | void | |
2416 | xfs_ialloc_log_agi( | |
0d1b9769 DC |
2417 | struct xfs_trans *tp, |
2418 | struct xfs_buf *bp, | |
2419 | uint32_t fields) | |
1da177e4 LT |
2420 | { |
2421 | int first; /* first byte number */ | |
2422 | int last; /* last byte number */ | |
2423 | static const short offsets[] = { /* field starting offsets */ | |
2424 | /* keep in sync with bit definitions */ | |
2425 | offsetof(xfs_agi_t, agi_magicnum), | |
2426 | offsetof(xfs_agi_t, agi_versionnum), | |
2427 | offsetof(xfs_agi_t, agi_seqno), | |
2428 | offsetof(xfs_agi_t, agi_length), | |
2429 | offsetof(xfs_agi_t, agi_count), | |
2430 | offsetof(xfs_agi_t, agi_root), | |
2431 | offsetof(xfs_agi_t, agi_level), | |
2432 | offsetof(xfs_agi_t, agi_freecount), | |
2433 | offsetof(xfs_agi_t, agi_newino), | |
2434 | offsetof(xfs_agi_t, agi_dirino), | |
2435 | offsetof(xfs_agi_t, agi_unlinked), | |
aafc3c24 BF |
2436 | offsetof(xfs_agi_t, agi_free_root), |
2437 | offsetof(xfs_agi_t, agi_free_level), | |
2a39946c | 2438 | offsetof(xfs_agi_t, agi_iblocks), |
1da177e4 LT |
2439 | sizeof(xfs_agi_t) |
2440 | }; | |
2441 | #ifdef DEBUG | |
370c782b | 2442 | struct xfs_agi *agi = bp->b_addr; |
1da177e4 | 2443 | |
69ef921b | 2444 | ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC)); |
1da177e4 | 2445 | #endif |
aafc3c24 | 2446 | |
1da177e4 | 2447 | /* |
aafc3c24 BF |
2448 | * Compute byte offsets for the first and last fields in the first |
2449 | * region and log the agi buffer. This only logs up through | |
2450 | * agi_unlinked. | |
1da177e4 | 2451 | */ |
aafc3c24 BF |
2452 | if (fields & XFS_AGI_ALL_BITS_R1) { |
2453 | xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R1, | |
2454 | &first, &last); | |
2455 | xfs_trans_log_buf(tp, bp, first, last); | |
2456 | } | |
2457 | ||
1da177e4 | 2458 | /* |
aafc3c24 BF |
2459 | * Mask off the bits in the first region and calculate the first and |
2460 | * last field offsets for any bits in the second region. | |
1da177e4 | 2461 | */ |
aafc3c24 BF |
2462 | fields &= ~XFS_AGI_ALL_BITS_R1; |
2463 | if (fields) { | |
2464 | xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R2, | |
2465 | &first, &last); | |
2466 | xfs_trans_log_buf(tp, bp, first, last); | |
2467 | } | |
1da177e4 LT |
2468 | } |
2469 | ||
a6a781a5 | 2470 | static xfs_failaddr_t |
612cfbfe | 2471 | xfs_agi_verify( |
3702ce6e DC |
2472 | struct xfs_buf *bp) |
2473 | { | |
dbd329f1 | 2474 | struct xfs_mount *mp = bp->b_mount; |
370c782b | 2475 | struct xfs_agi *agi = bp->b_addr; |
9f96cc95 | 2476 | int i; |
3702ce6e | 2477 | |
38c26bfd | 2478 | if (xfs_has_crc(mp)) { |
a45086e2 | 2479 | if (!uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid)) |
a6a781a5 | 2480 | return __this_address; |
370c782b | 2481 | if (!xfs_log_check_lsn(mp, be64_to_cpu(agi->agi_lsn))) |
a6a781a5 | 2482 | return __this_address; |
a45086e2 BF |
2483 | } |
2484 | ||
3702ce6e DC |
2485 | /* |
2486 | * Validate the magic number of the agi block. | |
2487 | */ | |
39708c20 | 2488 | if (!xfs_verify_magic(bp, agi->agi_magicnum)) |
a6a781a5 | 2489 | return __this_address; |
983d09ff | 2490 | if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum))) |
a6a781a5 | 2491 | return __this_address; |
3702ce6e | 2492 | |
d2a047f3 | 2493 | if (be32_to_cpu(agi->agi_level) < 1 || |
973975b7 | 2494 | be32_to_cpu(agi->agi_level) > M_IGEO(mp)->inobt_maxlevels) |
a6a781a5 | 2495 | return __this_address; |
d2a047f3 | 2496 | |
38c26bfd | 2497 | if (xfs_has_finobt(mp) && |
d2a047f3 | 2498 | (be32_to_cpu(agi->agi_free_level) < 1 || |
973975b7 | 2499 | be32_to_cpu(agi->agi_free_level) > M_IGEO(mp)->inobt_maxlevels)) |
a6a781a5 | 2500 | return __this_address; |
d2a047f3 | 2501 | |
3702ce6e DC |
2502 | /* |
2503 | * during growfs operations, the perag is not fully initialised, | |
2504 | * so we can't use it for any useful checking. growfs ensures we can't | |
2505 | * use it by using uncached buffers that don't have the perag attached | |
2506 | * so we can detect and avoid this problem. | |
2507 | */ | |
983d09ff | 2508 | if (bp->b_pag && be32_to_cpu(agi->agi_seqno) != bp->b_pag->pag_agno) |
a6a781a5 | 2509 | return __this_address; |
3702ce6e | 2510 | |
9f96cc95 | 2511 | for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++) { |
5089eaff | 2512 | if (agi->agi_unlinked[i] == cpu_to_be32(NULLAGINO)) |
9f96cc95 DC |
2513 | continue; |
2514 | if (!xfs_verify_ino(mp, be32_to_cpu(agi->agi_unlinked[i]))) | |
2515 | return __this_address; | |
2516 | } | |
2517 | ||
a6a781a5 | 2518 | return NULL; |
612cfbfe DC |
2519 | } |
2520 | ||
1813dd64 DC |
2521 | static void |
2522 | xfs_agi_read_verify( | |
612cfbfe DC |
2523 | struct xfs_buf *bp) |
2524 | { | |
dbd329f1 | 2525 | struct xfs_mount *mp = bp->b_mount; |
bc1a09b8 | 2526 | xfs_failaddr_t fa; |
983d09ff | 2527 | |
38c26bfd | 2528 | if (xfs_has_crc(mp) && |
ce5028cf | 2529 | !xfs_buf_verify_cksum(bp, XFS_AGI_CRC_OFF)) |
bc1a09b8 DW |
2530 | xfs_verifier_error(bp, -EFSBADCRC, __this_address); |
2531 | else { | |
2532 | fa = xfs_agi_verify(bp); | |
2533 | if (XFS_TEST_ERROR(fa, mp, XFS_ERRTAG_IALLOC_READ_AGI)) | |
2534 | xfs_verifier_error(bp, -EFSCORRUPTED, fa); | |
2535 | } | |
612cfbfe DC |
2536 | } |
2537 | ||
b0f539de | 2538 | static void |
1813dd64 | 2539 | xfs_agi_write_verify( |
612cfbfe DC |
2540 | struct xfs_buf *bp) |
2541 | { | |
dbd329f1 | 2542 | struct xfs_mount *mp = bp->b_mount; |
fb1755a6 | 2543 | struct xfs_buf_log_item *bip = bp->b_log_item; |
370c782b | 2544 | struct xfs_agi *agi = bp->b_addr; |
bc1a09b8 | 2545 | xfs_failaddr_t fa; |
983d09ff | 2546 | |
bc1a09b8 DW |
2547 | fa = xfs_agi_verify(bp); |
2548 | if (fa) { | |
2549 | xfs_verifier_error(bp, -EFSCORRUPTED, fa); | |
983d09ff DC |
2550 | return; |
2551 | } | |
2552 | ||
38c26bfd | 2553 | if (!xfs_has_crc(mp)) |
983d09ff DC |
2554 | return; |
2555 | ||
2556 | if (bip) | |
370c782b | 2557 | agi->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn); |
f1dbcd7e | 2558 | xfs_buf_update_cksum(bp, XFS_AGI_CRC_OFF); |
3702ce6e DC |
2559 | } |
2560 | ||
1813dd64 | 2561 | const struct xfs_buf_ops xfs_agi_buf_ops = { |
233135b7 | 2562 | .name = "xfs_agi", |
39708c20 | 2563 | .magic = { cpu_to_be32(XFS_AGI_MAGIC), cpu_to_be32(XFS_AGI_MAGIC) }, |
1813dd64 DC |
2564 | .verify_read = xfs_agi_read_verify, |
2565 | .verify_write = xfs_agi_write_verify, | |
b5572597 | 2566 | .verify_struct = xfs_agi_verify, |
1813dd64 DC |
2567 | }; |
2568 | ||
1da177e4 LT |
2569 | /* |
2570 | * Read in the allocation group header (inode allocation section) | |
2571 | */ | |
2572 | int | |
5e1be0fb | 2573 | xfs_read_agi( |
61021deb DC |
2574 | struct xfs_perag *pag, |
2575 | struct xfs_trans *tp, | |
2576 | struct xfs_buf **agibpp) | |
1da177e4 | 2577 | { |
61021deb | 2578 | struct xfs_mount *mp = pag->pag_mount; |
5e1be0fb | 2579 | int error; |
1da177e4 | 2580 | |
61021deb | 2581 | trace_xfs_read_agi(pag->pag_mount, pag->pag_agno); |
5e1be0fb CH |
2582 | |
2583 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, | |
61021deb DC |
2584 | XFS_AG_DADDR(mp, pag->pag_agno, XFS_AGI_DADDR(mp)), |
2585 | XFS_FSS_TO_BB(mp, 1), 0, agibpp, &xfs_agi_buf_ops); | |
1da177e4 LT |
2586 | if (error) |
2587 | return error; | |
200237d6 | 2588 | if (tp) |
61021deb | 2589 | xfs_trans_buf_set_type(tp, *agibpp, XFS_BLFT_AGI_BUF); |
5e1be0fb | 2590 | |
61021deb | 2591 | xfs_buf_set_ref(*agibpp, XFS_AGI_REF); |
5e1be0fb CH |
2592 | return 0; |
2593 | } | |
2594 | ||
a95fee40 DC |
2595 | /* |
2596 | * Read in the agi and initialise the per-ag data. If the caller supplies a | |
2597 | * @agibpp, return the locked AGI buffer to them, otherwise release it. | |
2598 | */ | |
5e1be0fb CH |
2599 | int |
2600 | xfs_ialloc_read_agi( | |
99b13c7f DC |
2601 | struct xfs_perag *pag, |
2602 | struct xfs_trans *tp, | |
a95fee40 | 2603 | struct xfs_buf **agibpp) |
5e1be0fb | 2604 | { |
a95fee40 | 2605 | struct xfs_buf *agibp; |
99b13c7f | 2606 | struct xfs_agi *agi; |
5e1be0fb CH |
2607 | int error; |
2608 | ||
99b13c7f | 2609 | trace_xfs_ialloc_read_agi(pag->pag_mount, pag->pag_agno); |
d123031a | 2610 | |
61021deb | 2611 | error = xfs_read_agi(pag, tp, &agibp); |
5e1be0fb CH |
2612 | if (error) |
2613 | return error; | |
2614 | ||
a95fee40 | 2615 | agi = agibp->b_addr; |
1da177e4 | 2616 | if (!pag->pagi_init) { |
16259e7d | 2617 | pag->pagi_freecount = be32_to_cpu(agi->agi_freecount); |
92821e2b | 2618 | pag->pagi_count = be32_to_cpu(agi->agi_count); |
1da177e4 | 2619 | pag->pagi_init = 1; |
1da177e4 | 2620 | } |
1da177e4 | 2621 | |
5e1be0fb CH |
2622 | /* |
2623 | * It's possible for these to be out of sync if | |
2624 | * we are in the middle of a forced shutdown. | |
2625 | */ | |
2626 | ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) || | |
99b13c7f | 2627 | xfs_is_shutdown(pag->pag_mount)); |
a95fee40 DC |
2628 | if (agibpp) |
2629 | *agibpp = agibp; | |
2630 | else | |
2631 | xfs_trans_brelse(tp, agibp); | |
92821e2b DC |
2632 | return 0; |
2633 | } | |
91fb9afc | 2634 | |
2e001266 DW |
2635 | /* Is there an inode record covering a given range of inode numbers? */ |
2636 | int | |
2637 | xfs_ialloc_has_inode_record( | |
2638 | struct xfs_btree_cur *cur, | |
2639 | xfs_agino_t low, | |
2640 | xfs_agino_t high, | |
2641 | bool *exists) | |
2642 | { | |
2643 | struct xfs_inobt_rec_incore irec; | |
2644 | xfs_agino_t agino; | |
2645 | uint16_t holemask; | |
2646 | int has_record; | |
2647 | int i; | |
2648 | int error; | |
2649 | ||
2650 | *exists = false; | |
2651 | error = xfs_inobt_lookup(cur, low, XFS_LOOKUP_LE, &has_record); | |
2652 | while (error == 0 && has_record) { | |
2653 | error = xfs_inobt_get_rec(cur, &irec, &has_record); | |
2654 | if (error || irec.ir_startino > high) | |
2655 | break; | |
2656 | ||
2657 | agino = irec.ir_startino; | |
2658 | holemask = irec.ir_holemask; | |
2659 | for (i = 0; i < XFS_INOBT_HOLEMASK_BITS; holemask >>= 1, | |
2660 | i++, agino += XFS_INODES_PER_HOLEMASK_BIT) { | |
2661 | if (holemask & 1) | |
2662 | continue; | |
2663 | if (agino + XFS_INODES_PER_HOLEMASK_BIT > low && | |
2664 | agino <= high) { | |
2665 | *exists = true; | |
2666 | return 0; | |
2667 | } | |
2668 | } | |
2669 | ||
2670 | error = xfs_btree_increment(cur, 0, &has_record); | |
2671 | } | |
2672 | return error; | |
2673 | } | |
2674 | ||
2675 | /* Is there an inode record covering a given extent? */ | |
2676 | int | |
2677 | xfs_ialloc_has_inodes_at_extent( | |
2678 | struct xfs_btree_cur *cur, | |
2679 | xfs_agblock_t bno, | |
2680 | xfs_extlen_t len, | |
2681 | bool *exists) | |
2682 | { | |
2683 | xfs_agino_t low; | |
2684 | xfs_agino_t high; | |
2685 | ||
43004b2a DW |
2686 | low = XFS_AGB_TO_AGINO(cur->bc_mp, bno); |
2687 | high = XFS_AGB_TO_AGINO(cur->bc_mp, bno + len) - 1; | |
2e001266 DW |
2688 | |
2689 | return xfs_ialloc_has_inode_record(cur, low, high, exists); | |
2690 | } | |
2691 | ||
2692 | struct xfs_ialloc_count_inodes { | |
2693 | xfs_agino_t count; | |
2694 | xfs_agino_t freecount; | |
2695 | }; | |
2696 | ||
2697 | /* Record inode counts across all inobt records. */ | |
2698 | STATIC int | |
2699 | xfs_ialloc_count_inodes_rec( | |
2700 | struct xfs_btree_cur *cur, | |
159eb69d | 2701 | const union xfs_btree_rec *rec, |
2e001266 DW |
2702 | void *priv) |
2703 | { | |
2704 | struct xfs_inobt_rec_incore irec; | |
2705 | struct xfs_ialloc_count_inodes *ci = priv; | |
2706 | ||
2707 | xfs_inobt_btrec_to_irec(cur->bc_mp, rec, &irec); | |
2708 | ci->count += irec.ir_count; | |
2709 | ci->freecount += irec.ir_freecount; | |
2710 | ||
2711 | return 0; | |
2712 | } | |
2713 | ||
2714 | /* Count allocated and free inodes under an inobt. */ | |
2715 | int | |
2716 | xfs_ialloc_count_inodes( | |
2717 | struct xfs_btree_cur *cur, | |
2718 | xfs_agino_t *count, | |
2719 | xfs_agino_t *freecount) | |
2720 | { | |
2721 | struct xfs_ialloc_count_inodes ci = {0}; | |
2722 | int error; | |
2723 | ||
2724 | ASSERT(cur->bc_btnum == XFS_BTNUM_INO); | |
2725 | error = xfs_btree_query_all(cur, xfs_ialloc_count_inodes_rec, &ci); | |
2726 | if (error) | |
2727 | return error; | |
2728 | ||
2729 | *count = ci.count; | |
2730 | *freecount = ci.freecount; | |
2731 | return 0; | |
2732 | } | |
494dba7b DW |
2733 | |
2734 | /* | |
2735 | * Initialize inode-related geometry information. | |
2736 | * | |
2737 | * Compute the inode btree min and max levels and set maxicount. | |
2738 | * | |
2739 | * Set the inode cluster size. This may still be overridden by the file | |
2740 | * system block size if it is larger than the chosen cluster size. | |
2741 | * | |
2742 | * For v5 filesystems, scale the cluster size with the inode size to keep a | |
2743 | * constant ratio of inode per cluster buffer, but only if mkfs has set the | |
2744 | * inode alignment value appropriately for larger cluster sizes. | |
2745 | * | |
2746 | * Then compute the inode cluster alignment information. | |
2747 | */ | |
2748 | void | |
2749 | xfs_ialloc_setup_geometry( | |
2750 | struct xfs_mount *mp) | |
2751 | { | |
2752 | struct xfs_sb *sbp = &mp->m_sb; | |
2753 | struct xfs_ino_geometry *igeo = M_IGEO(mp); | |
2754 | uint64_t icount; | |
2755 | uint inodes; | |
2756 | ||
f93e5436 | 2757 | igeo->new_diflags2 = 0; |
ebd9027d | 2758 | if (xfs_has_bigtime(mp)) |
f93e5436 | 2759 | igeo->new_diflags2 |= XFS_DIFLAG2_BIGTIME; |
9b7d16e3 CB |
2760 | if (xfs_has_large_extent_counts(mp)) |
2761 | igeo->new_diflags2 |= XFS_DIFLAG2_NREXT64; | |
f93e5436 | 2762 | |
494dba7b DW |
2763 | /* Compute inode btree geometry. */ |
2764 | igeo->agino_log = sbp->sb_inopblog + sbp->sb_agblklog; | |
2765 | igeo->inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1); | |
2766 | igeo->inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0); | |
2767 | igeo->inobt_mnr[0] = igeo->inobt_mxr[0] / 2; | |
2768 | igeo->inobt_mnr[1] = igeo->inobt_mxr[1] / 2; | |
2769 | ||
2770 | igeo->ialloc_inos = max_t(uint16_t, XFS_INODES_PER_CHUNK, | |
2771 | sbp->sb_inopblock); | |
2772 | igeo->ialloc_blks = igeo->ialloc_inos >> sbp->sb_inopblog; | |
2773 | ||
2774 | if (sbp->sb_spino_align) | |
2775 | igeo->ialloc_min_blks = sbp->sb_spino_align; | |
2776 | else | |
2777 | igeo->ialloc_min_blks = igeo->ialloc_blks; | |
2778 | ||
2779 | /* Compute and fill in value of m_ino_geo.inobt_maxlevels. */ | |
2780 | inodes = (1LL << XFS_INO_AGINO_BITS(mp)) >> XFS_INODES_PER_CHUNK_LOG; | |
2781 | igeo->inobt_maxlevels = xfs_btree_compute_maxlevels(igeo->inobt_mnr, | |
2782 | inodes); | |
0ed5f735 | 2783 | ASSERT(igeo->inobt_maxlevels <= xfs_iallocbt_maxlevels_ondisk()); |
494dba7b | 2784 | |
c94613fe DW |
2785 | /* |
2786 | * Set the maximum inode count for this filesystem, being careful not | |
2787 | * to use obviously garbage sb_inopblog/sb_inopblock values. Regular | |
2788 | * users should never get here due to failing sb verification, but | |
2789 | * certain users (xfs_db) need to be usable even with corrupt metadata. | |
2790 | */ | |
2791 | if (sbp->sb_imax_pct && igeo->ialloc_blks) { | |
494dba7b DW |
2792 | /* |
2793 | * Make sure the maximum inode count is a multiple | |
2794 | * of the units we allocate inodes in. | |
2795 | */ | |
2796 | icount = sbp->sb_dblocks * sbp->sb_imax_pct; | |
2797 | do_div(icount, 100); | |
2798 | do_div(icount, igeo->ialloc_blks); | |
2799 | igeo->maxicount = XFS_FSB_TO_INO(mp, | |
2800 | icount * igeo->ialloc_blks); | |
2801 | } else { | |
2802 | igeo->maxicount = 0; | |
2803 | } | |
2804 | ||
490d451f DW |
2805 | /* |
2806 | * Compute the desired size of an inode cluster buffer size, which | |
2807 | * starts at 8K and (on v5 filesystems) scales up with larger inode | |
2808 | * sizes. | |
2809 | * | |
2810 | * Preserve the desired inode cluster size because the sparse inodes | |
2811 | * feature uses that desired size (not the actual size) to compute the | |
2812 | * sparse inode alignment. The mount code validates this value, so we | |
2813 | * cannot change the behavior. | |
2814 | */ | |
2815 | igeo->inode_cluster_size_raw = XFS_INODE_BIG_CLUSTER_SIZE; | |
ebd9027d | 2816 | if (xfs_has_v3inodes(mp)) { |
490d451f | 2817 | int new_size = igeo->inode_cluster_size_raw; |
494dba7b DW |
2818 | |
2819 | new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE; | |
2820 | if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size)) | |
490d451f | 2821 | igeo->inode_cluster_size_raw = new_size; |
494dba7b DW |
2822 | } |
2823 | ||
2824 | /* Calculate inode cluster ratios. */ | |
490d451f | 2825 | if (igeo->inode_cluster_size_raw > mp->m_sb.sb_blocksize) |
494dba7b | 2826 | igeo->blocks_per_cluster = XFS_B_TO_FSBT(mp, |
490d451f | 2827 | igeo->inode_cluster_size_raw); |
494dba7b DW |
2828 | else |
2829 | igeo->blocks_per_cluster = 1; | |
490d451f | 2830 | igeo->inode_cluster_size = XFS_FSB_TO_B(mp, igeo->blocks_per_cluster); |
494dba7b DW |
2831 | igeo->inodes_per_cluster = XFS_FSB_TO_INO(mp, igeo->blocks_per_cluster); |
2832 | ||
2833 | /* Calculate inode cluster alignment. */ | |
ebd9027d | 2834 | if (xfs_has_align(mp) && |
494dba7b DW |
2835 | mp->m_sb.sb_inoalignmt >= igeo->blocks_per_cluster) |
2836 | igeo->cluster_align = mp->m_sb.sb_inoalignmt; | |
2837 | else | |
2838 | igeo->cluster_align = 1; | |
2839 | igeo->inoalign_mask = igeo->cluster_align - 1; | |
2840 | igeo->cluster_align_inodes = XFS_FSB_TO_INO(mp, igeo->cluster_align); | |
2841 | ||
2842 | /* | |
2843 | * If we are using stripe alignment, check whether | |
2844 | * the stripe unit is a multiple of the inode alignment | |
2845 | */ | |
2846 | if (mp->m_dalign && igeo->inoalign_mask && | |
2847 | !(mp->m_dalign & igeo->inoalign_mask)) | |
2848 | igeo->ialloc_align = mp->m_dalign; | |
2849 | else | |
2850 | igeo->ialloc_align = 0; | |
2851 | } | |
13eaec4b DW |
2852 | |
2853 | /* Compute the location of the root directory inode that is laid out by mkfs. */ | |
2854 | xfs_ino_t | |
2855 | xfs_ialloc_calc_rootino( | |
2856 | struct xfs_mount *mp, | |
2857 | int sunit) | |
2858 | { | |
2859 | struct xfs_ino_geometry *igeo = M_IGEO(mp); | |
2860 | xfs_agblock_t first_bno; | |
2861 | ||
2862 | /* | |
2863 | * Pre-calculate the geometry of AG 0. We know what it looks like | |
2864 | * because libxfs knows how to create allocation groups now. | |
2865 | * | |
2866 | * first_bno is the first block in which mkfs could possibly have | |
2867 | * allocated the root directory inode, once we factor in the metadata | |
2868 | * that mkfs formats before it. Namely, the four AG headers... | |
2869 | */ | |
2870 | first_bno = howmany(4 * mp->m_sb.sb_sectsize, mp->m_sb.sb_blocksize); | |
2871 | ||
2872 | /* ...the two free space btree roots... */ | |
2873 | first_bno += 2; | |
2874 | ||
2875 | /* ...the inode btree root... */ | |
2876 | first_bno += 1; | |
2877 | ||
2878 | /* ...the initial AGFL... */ | |
2879 | first_bno += xfs_alloc_min_freelist(mp, NULL); | |
2880 | ||
2881 | /* ...the free inode btree root... */ | |
ebd9027d | 2882 | if (xfs_has_finobt(mp)) |
13eaec4b DW |
2883 | first_bno++; |
2884 | ||
2885 | /* ...the reverse mapping btree root... */ | |
ebd9027d | 2886 | if (xfs_has_rmapbt(mp)) |
13eaec4b DW |
2887 | first_bno++; |
2888 | ||
2889 | /* ...the reference count btree... */ | |
ebd9027d | 2890 | if (xfs_has_reflink(mp)) |
13eaec4b DW |
2891 | first_bno++; |
2892 | ||
2893 | /* | |
2894 | * ...and the log, if it is allocated in the first allocation group. | |
2895 | * | |
2896 | * This can happen with filesystems that only have a single | |
2897 | * allocation group, or very odd geometries created by old mkfs | |
2898 | * versions on very small filesystems. | |
2899 | */ | |
36029dee | 2900 | if (xfs_ag_contains_log(mp, 0)) |
13eaec4b DW |
2901 | first_bno += mp->m_sb.sb_logblocks; |
2902 | ||
2903 | /* | |
2904 | * Now round first_bno up to whatever allocation alignment is given | |
2905 | * by the filesystem or was passed in. | |
2906 | */ | |
ebd9027d | 2907 | if (xfs_has_dalign(mp) && igeo->ialloc_align > 0) |
13eaec4b | 2908 | first_bno = roundup(first_bno, sunit); |
ebd9027d | 2909 | else if (xfs_has_align(mp) && |
13eaec4b DW |
2910 | mp->m_sb.sb_inoalignmt > 1) |
2911 | first_bno = roundup(first_bno, mp->m_sb.sb_inoalignmt); | |
2912 | ||
2913 | return XFS_AGINO_TO_INO(mp, 0, XFS_AGB_TO_AGINO(mp, first_bno)); | |
2914 | } | |
da062d16 DW |
2915 | |
2916 | /* | |
2917 | * Ensure there are not sparse inode clusters that cross the new EOAG. | |
2918 | * | |
2919 | * This is a no-op for non-spinode filesystems since clusters are always fully | |
2920 | * allocated and checking the bnobt suffices. However, a spinode filesystem | |
2921 | * could have a record where the upper inodes are free blocks. If those blocks | |
2922 | * were removed from the filesystem, the inode record would extend beyond EOAG, | |
2923 | * which will be flagged as corruption. | |
2924 | */ | |
2925 | int | |
2926 | xfs_ialloc_check_shrink( | |
2927 | struct xfs_trans *tp, | |
2928 | xfs_agnumber_t agno, | |
2929 | struct xfs_buf *agibp, | |
2930 | xfs_agblock_t new_length) | |
2931 | { | |
2932 | struct xfs_inobt_rec_incore rec; | |
2933 | struct xfs_btree_cur *cur; | |
2934 | struct xfs_mount *mp = tp->t_mountp; | |
2935 | struct xfs_perag *pag; | |
2936 | xfs_agino_t agino = XFS_AGB_TO_AGINO(mp, new_length); | |
2937 | int has; | |
2938 | int error; | |
2939 | ||
38c26bfd | 2940 | if (!xfs_has_sparseinodes(mp)) |
da062d16 DW |
2941 | return 0; |
2942 | ||
2943 | pag = xfs_perag_get(mp, agno); | |
2944 | cur = xfs_inobt_init_cursor(mp, tp, agibp, pag, XFS_BTNUM_INO); | |
2945 | ||
2946 | /* Look up the inobt record that would correspond to the new EOFS. */ | |
2947 | error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &has); | |
2948 | if (error || !has) | |
2949 | goto out; | |
2950 | ||
2951 | error = xfs_inobt_get_rec(cur, &rec, &has); | |
2952 | if (error) | |
2953 | goto out; | |
2954 | ||
2955 | if (!has) { | |
2956 | error = -EFSCORRUPTED; | |
2957 | goto out; | |
2958 | } | |
2959 | ||
2960 | /* If the record covers inodes that would be beyond EOFS, bail out. */ | |
2961 | if (rec.ir_startino + XFS_INODES_PER_CHUNK > agino) { | |
2962 | error = -ENOSPC; | |
2963 | goto out; | |
2964 | } | |
2965 | out: | |
2966 | xfs_btree_del_cursor(cur, error); | |
2967 | xfs_perag_put(pag); | |
2968 | return error; | |
2969 | } |