]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_fs.h" |
70a9883c | 20 | #include "xfs_shared.h" |
239880ef DC |
21 | #include "xfs_format.h" |
22 | #include "xfs_log_format.h" | |
23 | #include "xfs_trans_resv.h" | |
a844f451 | 24 | #include "xfs_bit.h" |
1da177e4 | 25 | #include "xfs_sb.h" |
1da177e4 | 26 | #include "xfs_mount.h" |
3ab78df2 | 27 | #include "xfs_defer.h" |
1da177e4 | 28 | #include "xfs_inode.h" |
a844f451 NS |
29 | #include "xfs_btree.h" |
30 | #include "xfs_ialloc.h" | |
a4fbe6ab | 31 | #include "xfs_ialloc_btree.h" |
1da177e4 | 32 | #include "xfs_alloc.h" |
1da177e4 | 33 | #include "xfs_rtalloc.h" |
e9e899a2 | 34 | #include "xfs_errortag.h" |
1da177e4 LT |
35 | #include "xfs_error.h" |
36 | #include "xfs_bmap.h" | |
983d09ff | 37 | #include "xfs_cksum.h" |
239880ef | 38 | #include "xfs_trans.h" |
983d09ff | 39 | #include "xfs_buf_item.h" |
ddf6ad01 | 40 | #include "xfs_icreate_item.h" |
7bb85ef3 | 41 | #include "xfs_icache.h" |
d123031a | 42 | #include "xfs_trace.h" |
a45086e2 | 43 | #include "xfs_log.h" |
340785cc | 44 | #include "xfs_rmap.h" |
1da177e4 | 45 | |
1da177e4 LT |
46 | |
47 | /* | |
48 | * Allocation group level functions. | |
49 | */ | |
e936945e | 50 | int |
75de2a91 | 51 | xfs_ialloc_cluster_alignment( |
7a1df156 | 52 | struct xfs_mount *mp) |
75de2a91 | 53 | { |
7a1df156 | 54 | if (xfs_sb_version_hasalign(&mp->m_sb) && |
8ee9fdbe | 55 | mp->m_sb.sb_inoalignmt >= xfs_icluster_size_fsb(mp)) |
7a1df156 | 56 | return mp->m_sb.sb_inoalignmt; |
75de2a91 DC |
57 | return 1; |
58 | } | |
1da177e4 | 59 | |
fe033cc8 | 60 | /* |
21875505 | 61 | * Lookup a record by ino in the btree given by cur. |
fe033cc8 | 62 | */ |
81e25176 | 63 | int /* error */ |
21875505 | 64 | xfs_inobt_lookup( |
fe033cc8 CH |
65 | struct xfs_btree_cur *cur, /* btree cursor */ |
66 | xfs_agino_t ino, /* starting inode of chunk */ | |
21875505 | 67 | xfs_lookup_t dir, /* <=, >=, == */ |
fe033cc8 CH |
68 | int *stat) /* success/failure */ |
69 | { | |
70 | cur->bc_rec.i.ir_startino = ino; | |
5419040f BF |
71 | cur->bc_rec.i.ir_holemask = 0; |
72 | cur->bc_rec.i.ir_count = 0; | |
21875505 CH |
73 | cur->bc_rec.i.ir_freecount = 0; |
74 | cur->bc_rec.i.ir_free = 0; | |
75 | return xfs_btree_lookup(cur, dir, stat); | |
fe033cc8 CH |
76 | } |
77 | ||
278d0ca1 | 78 | /* |
afabc24a | 79 | * Update the record referred to by cur to the value given. |
278d0ca1 CH |
80 | * This either works (return 0) or gets an EFSCORRUPTED error. |
81 | */ | |
82 | STATIC int /* error */ | |
83 | xfs_inobt_update( | |
84 | struct xfs_btree_cur *cur, /* btree cursor */ | |
afabc24a | 85 | xfs_inobt_rec_incore_t *irec) /* btree record */ |
278d0ca1 CH |
86 | { |
87 | union xfs_btree_rec rec; | |
88 | ||
afabc24a | 89 | rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino); |
5419040f BF |
90 | if (xfs_sb_version_hassparseinodes(&cur->bc_mp->m_sb)) { |
91 | rec.inobt.ir_u.sp.ir_holemask = cpu_to_be16(irec->ir_holemask); | |
92 | rec.inobt.ir_u.sp.ir_count = irec->ir_count; | |
93 | rec.inobt.ir_u.sp.ir_freecount = irec->ir_freecount; | |
94 | } else { | |
95 | /* ir_holemask/ir_count not supported on-disk */ | |
96 | rec.inobt.ir_u.f.ir_freecount = cpu_to_be32(irec->ir_freecount); | |
97 | } | |
afabc24a | 98 | rec.inobt.ir_free = cpu_to_be64(irec->ir_free); |
278d0ca1 CH |
99 | return xfs_btree_update(cur, &rec); |
100 | } | |
101 | ||
e936945e DW |
102 | /* Convert on-disk btree record to incore inobt record. */ |
103 | void | |
104 | xfs_inobt_btrec_to_irec( | |
105 | struct xfs_mount *mp, | |
106 | union xfs_btree_rec *rec, | |
107 | struct xfs_inobt_rec_incore *irec) | |
8cc938fe | 108 | { |
5419040f | 109 | irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino); |
e936945e | 110 | if (xfs_sb_version_hassparseinodes(&mp->m_sb)) { |
5419040f BF |
111 | irec->ir_holemask = be16_to_cpu(rec->inobt.ir_u.sp.ir_holemask); |
112 | irec->ir_count = rec->inobt.ir_u.sp.ir_count; | |
113 | irec->ir_freecount = rec->inobt.ir_u.sp.ir_freecount; | |
114 | } else { | |
115 | /* | |
116 | * ir_holemask/ir_count not supported on-disk. Fill in hardcoded | |
117 | * values for full inode chunks. | |
118 | */ | |
119 | irec->ir_holemask = XFS_INOBT_HOLEMASK_FULL; | |
120 | irec->ir_count = XFS_INODES_PER_CHUNK; | |
121 | irec->ir_freecount = | |
122 | be32_to_cpu(rec->inobt.ir_u.f.ir_freecount); | |
8cc938fe | 123 | } |
5419040f | 124 | irec->ir_free = be64_to_cpu(rec->inobt.ir_free); |
e936945e DW |
125 | } |
126 | ||
127 | /* | |
128 | * Get the data from the pointed-to record. | |
129 | */ | |
130 | int | |
131 | xfs_inobt_get_rec( | |
132 | struct xfs_btree_cur *cur, | |
133 | struct xfs_inobt_rec_incore *irec, | |
134 | int *stat) | |
135 | { | |
9e6c08d4 DC |
136 | struct xfs_mount *mp = cur->bc_mp; |
137 | xfs_agnumber_t agno = cur->bc_private.a.agno; | |
e936945e DW |
138 | union xfs_btree_rec *rec; |
139 | int error; | |
9e6c08d4 | 140 | uint64_t realfree; |
e936945e DW |
141 | |
142 | error = xfs_btree_get_rec(cur, &rec, stat); | |
143 | if (error || *stat == 0) | |
144 | return error; | |
145 | ||
9e6c08d4 DC |
146 | xfs_inobt_btrec_to_irec(mp, rec, irec); |
147 | ||
148 | if (!xfs_verify_agino(mp, agno, irec->ir_startino)) | |
149 | goto out_bad_rec; | |
150 | if (irec->ir_count < XFS_INODES_PER_HOLEMASK_BIT || | |
151 | irec->ir_count > XFS_INODES_PER_CHUNK) | |
152 | goto out_bad_rec; | |
153 | if (irec->ir_freecount > XFS_INODES_PER_CHUNK) | |
154 | goto out_bad_rec; | |
155 | ||
156 | /* if there are no holes, return the first available offset */ | |
157 | if (!xfs_inobt_issparse(irec->ir_holemask)) | |
158 | realfree = irec->ir_free; | |
159 | else | |
160 | realfree = irec->ir_free & xfs_inobt_irec_to_allocmask(irec); | |
161 | if (hweight64(realfree) != irec->ir_freecount) | |
162 | goto out_bad_rec; | |
5419040f BF |
163 | |
164 | return 0; | |
9e6c08d4 DC |
165 | |
166 | out_bad_rec: | |
167 | xfs_warn(mp, | |
168 | "%s Inode BTree record corruption in AG %d detected!", | |
169 | cur->bc_btnum == XFS_BTNUM_INO ? "Used" : "Free", agno); | |
170 | xfs_warn(mp, | |
171 | "start inode 0x%x, count 0x%x, free 0x%x freemask 0x%llx, holemask 0x%x", | |
172 | irec->ir_startino, irec->ir_count, irec->ir_freecount, | |
173 | irec->ir_free, irec->ir_holemask); | |
174 | return -EFSCORRUPTED; | |
8cc938fe CH |
175 | } |
176 | ||
0aa0a756 BF |
177 | /* |
178 | * Insert a single inobt record. Cursor must already point to desired location. | |
179 | */ | |
7f8f1313 | 180 | int |
0aa0a756 BF |
181 | xfs_inobt_insert_rec( |
182 | struct xfs_btree_cur *cur, | |
c8ce540d DW |
183 | uint16_t holemask, |
184 | uint8_t count, | |
185 | int32_t freecount, | |
0aa0a756 BF |
186 | xfs_inofree_t free, |
187 | int *stat) | |
188 | { | |
5419040f BF |
189 | cur->bc_rec.i.ir_holemask = holemask; |
190 | cur->bc_rec.i.ir_count = count; | |
0aa0a756 BF |
191 | cur->bc_rec.i.ir_freecount = freecount; |
192 | cur->bc_rec.i.ir_free = free; | |
193 | return xfs_btree_insert(cur, stat); | |
194 | } | |
195 | ||
196 | /* | |
197 | * Insert records describing a newly allocated inode chunk into the inobt. | |
198 | */ | |
199 | STATIC int | |
200 | xfs_inobt_insert( | |
201 | struct xfs_mount *mp, | |
202 | struct xfs_trans *tp, | |
203 | struct xfs_buf *agbp, | |
204 | xfs_agino_t newino, | |
205 | xfs_agino_t newlen, | |
206 | xfs_btnum_t btnum) | |
207 | { | |
208 | struct xfs_btree_cur *cur; | |
209 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
210 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
211 | xfs_agino_t thisino; | |
212 | int i; | |
213 | int error; | |
214 | ||
215 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum); | |
216 | ||
217 | for (thisino = newino; | |
218 | thisino < newino + newlen; | |
219 | thisino += XFS_INODES_PER_CHUNK) { | |
220 | error = xfs_inobt_lookup(cur, thisino, XFS_LOOKUP_EQ, &i); | |
221 | if (error) { | |
222 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
223 | return error; | |
224 | } | |
225 | ASSERT(i == 0); | |
226 | ||
5419040f BF |
227 | error = xfs_inobt_insert_rec(cur, XFS_INOBT_HOLEMASK_FULL, |
228 | XFS_INODES_PER_CHUNK, | |
229 | XFS_INODES_PER_CHUNK, | |
0aa0a756 BF |
230 | XFS_INOBT_ALL_FREE, &i); |
231 | if (error) { | |
232 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
233 | return error; | |
234 | } | |
235 | ASSERT(i == 1); | |
236 | } | |
237 | ||
238 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
239 | ||
240 | return 0; | |
241 | } | |
242 | ||
0b48db80 DC |
243 | /* |
244 | * Verify that the number of free inodes in the AGI is correct. | |
245 | */ | |
246 | #ifdef DEBUG | |
247 | STATIC int | |
248 | xfs_check_agi_freecount( | |
249 | struct xfs_btree_cur *cur, | |
250 | struct xfs_agi *agi) | |
251 | { | |
252 | if (cur->bc_nlevels == 1) { | |
253 | xfs_inobt_rec_incore_t rec; | |
254 | int freecount = 0; | |
255 | int error; | |
256 | int i; | |
257 | ||
21875505 | 258 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); |
0b48db80 DC |
259 | if (error) |
260 | return error; | |
261 | ||
262 | do { | |
263 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
264 | if (error) | |
265 | return error; | |
266 | ||
267 | if (i) { | |
268 | freecount += rec.ir_freecount; | |
269 | error = xfs_btree_increment(cur, 0, &i); | |
270 | if (error) | |
271 | return error; | |
272 | } | |
273 | } while (i == 1); | |
274 | ||
275 | if (!XFS_FORCED_SHUTDOWN(cur->bc_mp)) | |
276 | ASSERT(freecount == be32_to_cpu(agi->agi_freecount)); | |
277 | } | |
278 | return 0; | |
279 | } | |
280 | #else | |
281 | #define xfs_check_agi_freecount(cur, agi) 0 | |
282 | #endif | |
283 | ||
85c0b2ab | 284 | /* |
28c8e41a DC |
285 | * Initialise a new set of inodes. When called without a transaction context |
286 | * (e.g. from recovery) we initiate a delayed write of the inode buffers rather | |
287 | * than logging them (which in a transaction context puts them into the AIL | |
288 | * for writeback rather than the xfsbufd queue). | |
85c0b2ab | 289 | */ |
ddf6ad01 | 290 | int |
85c0b2ab DC |
291 | xfs_ialloc_inode_init( |
292 | struct xfs_mount *mp, | |
293 | struct xfs_trans *tp, | |
28c8e41a | 294 | struct list_head *buffer_list, |
463958af | 295 | int icount, |
85c0b2ab DC |
296 | xfs_agnumber_t agno, |
297 | xfs_agblock_t agbno, | |
298 | xfs_agblock_t length, | |
299 | unsigned int gen) | |
300 | { | |
301 | struct xfs_buf *fbuf; | |
302 | struct xfs_dinode *free; | |
6e0c7b8c | 303 | int nbufs, blks_per_cluster, inodes_per_cluster; |
85c0b2ab DC |
304 | int version; |
305 | int i, j; | |
306 | xfs_daddr_t d; | |
93848a99 | 307 | xfs_ino_t ino = 0; |
85c0b2ab DC |
308 | |
309 | /* | |
6e0c7b8c JL |
310 | * Loop over the new block(s), filling in the inodes. For small block |
311 | * sizes, manipulate the inodes in buffers which are multiples of the | |
312 | * blocks size. | |
85c0b2ab | 313 | */ |
6e0c7b8c JL |
314 | blks_per_cluster = xfs_icluster_size_fsb(mp); |
315 | inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog; | |
316 | nbufs = length / blks_per_cluster; | |
85c0b2ab DC |
317 | |
318 | /* | |
93848a99 CH |
319 | * Figure out what version number to use in the inodes we create. If |
320 | * the superblock version has caught up to the one that supports the new | |
321 | * inode format, then use the new inode version. Otherwise use the old | |
322 | * version so that old kernels will continue to be able to use the file | |
323 | * system. | |
324 | * | |
325 | * For v3 inodes, we also need to write the inode number into the inode, | |
326 | * so calculate the first inode number of the chunk here as | |
327 | * XFS_OFFBNO_TO_AGINO() only works within a filesystem block, not | |
328 | * across multiple filesystem blocks (such as a cluster) and so cannot | |
329 | * be used in the cluster buffer loop below. | |
330 | * | |
331 | * Further, because we are writing the inode directly into the buffer | |
332 | * and calculating a CRC on the entire inode, we have ot log the entire | |
333 | * inode so that the entire range the CRC covers is present in the log. | |
334 | * That means for v3 inode we log the entire buffer rather than just the | |
335 | * inode cores. | |
85c0b2ab | 336 | */ |
93848a99 CH |
337 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
338 | version = 3; | |
339 | ino = XFS_AGINO_TO_INO(mp, agno, | |
340 | XFS_OFFBNO_TO_AGINO(mp, agbno, 0)); | |
ddf6ad01 DC |
341 | |
342 | /* | |
343 | * log the initialisation that is about to take place as an | |
344 | * logical operation. This means the transaction does not | |
345 | * need to log the physical changes to the inode buffers as log | |
346 | * recovery will know what initialisation is actually needed. | |
347 | * Hence we only need to log the buffers as "ordered" buffers so | |
348 | * they track in the AIL as if they were physically logged. | |
349 | */ | |
350 | if (tp) | |
463958af | 351 | xfs_icreate_log(tp, agno, agbno, icount, |
ddf6ad01 | 352 | mp->m_sb.sb_inodesize, length, gen); |
263997a6 | 353 | } else |
85c0b2ab | 354 | version = 2; |
85c0b2ab DC |
355 | |
356 | for (j = 0; j < nbufs; j++) { | |
357 | /* | |
358 | * Get the block. | |
359 | */ | |
360 | d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster)); | |
361 | fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, | |
7c4cebe8 DC |
362 | mp->m_bsize * blks_per_cluster, |
363 | XBF_UNMAPPED); | |
2a30f36d | 364 | if (!fbuf) |
2451337d | 365 | return -ENOMEM; |
ddf6ad01 DC |
366 | |
367 | /* Initialize the inode buffers and log them appropriately. */ | |
1813dd64 | 368 | fbuf->b_ops = &xfs_inode_buf_ops; |
93848a99 | 369 | xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length)); |
6e0c7b8c | 370 | for (i = 0; i < inodes_per_cluster; i++) { |
85c0b2ab | 371 | int ioffset = i << mp->m_sb.sb_inodelog; |
93848a99 | 372 | uint isize = xfs_dinode_size(version); |
85c0b2ab DC |
373 | |
374 | free = xfs_make_iptr(mp, fbuf, i); | |
375 | free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC); | |
376 | free->di_version = version; | |
377 | free->di_gen = cpu_to_be32(gen); | |
378 | free->di_next_unlinked = cpu_to_be32(NULLAGINO); | |
93848a99 CH |
379 | |
380 | if (version == 3) { | |
381 | free->di_ino = cpu_to_be64(ino); | |
382 | ino++; | |
ce748eaa ES |
383 | uuid_copy(&free->di_uuid, |
384 | &mp->m_sb.sb_meta_uuid); | |
93848a99 | 385 | xfs_dinode_calc_crc(mp, free); |
28c8e41a | 386 | } else if (tp) { |
93848a99 CH |
387 | /* just log the inode core */ |
388 | xfs_trans_log_buf(tp, fbuf, ioffset, | |
389 | ioffset + isize - 1); | |
390 | } | |
391 | } | |
28c8e41a DC |
392 | |
393 | if (tp) { | |
394 | /* | |
395 | * Mark the buffer as an inode allocation buffer so it | |
396 | * sticks in AIL at the point of this allocation | |
397 | * transaction. This ensures the they are on disk before | |
398 | * the tail of the log can be moved past this | |
399 | * transaction (i.e. by preventing relogging from moving | |
400 | * it forward in the log). | |
401 | */ | |
402 | xfs_trans_inode_alloc_buf(tp, fbuf); | |
403 | if (version == 3) { | |
ddf6ad01 DC |
404 | /* |
405 | * Mark the buffer as ordered so that they are | |
406 | * not physically logged in the transaction but | |
407 | * still tracked in the AIL as part of the | |
408 | * transaction and pin the log appropriately. | |
409 | */ | |
410 | xfs_trans_ordered_buf(tp, fbuf); | |
28c8e41a DC |
411 | } |
412 | } else { | |
413 | fbuf->b_flags |= XBF_DONE; | |
414 | xfs_buf_delwri_queue(fbuf, buffer_list); | |
415 | xfs_buf_relse(fbuf); | |
85c0b2ab | 416 | } |
85c0b2ab | 417 | } |
2a30f36d | 418 | return 0; |
85c0b2ab DC |
419 | } |
420 | ||
56d1115c BF |
421 | /* |
422 | * Align startino and allocmask for a recently allocated sparse chunk such that | |
423 | * they are fit for insertion (or merge) into the on-disk inode btrees. | |
424 | * | |
425 | * Background: | |
426 | * | |
427 | * When enabled, sparse inode support increases the inode alignment from cluster | |
428 | * size to inode chunk size. This means that the minimum range between two | |
429 | * non-adjacent inode records in the inobt is large enough for a full inode | |
430 | * record. This allows for cluster sized, cluster aligned block allocation | |
431 | * without need to worry about whether the resulting inode record overlaps with | |
432 | * another record in the tree. Without this basic rule, we would have to deal | |
433 | * with the consequences of overlap by potentially undoing recent allocations in | |
434 | * the inode allocation codepath. | |
435 | * | |
436 | * Because of this alignment rule (which is enforced on mount), there are two | |
437 | * inobt possibilities for newly allocated sparse chunks. One is that the | |
438 | * aligned inode record for the chunk covers a range of inodes not already | |
439 | * covered in the inobt (i.e., it is safe to insert a new sparse record). The | |
440 | * other is that a record already exists at the aligned startino that considers | |
441 | * the newly allocated range as sparse. In the latter case, record content is | |
442 | * merged in hope that sparse inode chunks fill to full chunks over time. | |
443 | */ | |
444 | STATIC void | |
445 | xfs_align_sparse_ino( | |
446 | struct xfs_mount *mp, | |
447 | xfs_agino_t *startino, | |
448 | uint16_t *allocmask) | |
449 | { | |
450 | xfs_agblock_t agbno; | |
451 | xfs_agblock_t mod; | |
452 | int offset; | |
453 | ||
454 | agbno = XFS_AGINO_TO_AGBNO(mp, *startino); | |
455 | mod = agbno % mp->m_sb.sb_inoalignmt; | |
456 | if (!mod) | |
457 | return; | |
458 | ||
459 | /* calculate the inode offset and align startino */ | |
460 | offset = mod << mp->m_sb.sb_inopblog; | |
461 | *startino -= offset; | |
462 | ||
463 | /* | |
464 | * Since startino has been aligned down, left shift allocmask such that | |
465 | * it continues to represent the same physical inodes relative to the | |
466 | * new startino. | |
467 | */ | |
468 | *allocmask <<= offset / XFS_INODES_PER_HOLEMASK_BIT; | |
469 | } | |
470 | ||
471 | /* | |
472 | * Determine whether the source inode record can merge into the target. Both | |
473 | * records must be sparse, the inode ranges must match and there must be no | |
474 | * allocation overlap between the records. | |
475 | */ | |
476 | STATIC bool | |
477 | __xfs_inobt_can_merge( | |
478 | struct xfs_inobt_rec_incore *trec, /* tgt record */ | |
479 | struct xfs_inobt_rec_incore *srec) /* src record */ | |
480 | { | |
481 | uint64_t talloc; | |
482 | uint64_t salloc; | |
483 | ||
484 | /* records must cover the same inode range */ | |
485 | if (trec->ir_startino != srec->ir_startino) | |
486 | return false; | |
487 | ||
488 | /* both records must be sparse */ | |
489 | if (!xfs_inobt_issparse(trec->ir_holemask) || | |
490 | !xfs_inobt_issparse(srec->ir_holemask)) | |
491 | return false; | |
492 | ||
493 | /* both records must track some inodes */ | |
494 | if (!trec->ir_count || !srec->ir_count) | |
495 | return false; | |
496 | ||
497 | /* can't exceed capacity of a full record */ | |
498 | if (trec->ir_count + srec->ir_count > XFS_INODES_PER_CHUNK) | |
499 | return false; | |
500 | ||
501 | /* verify there is no allocation overlap */ | |
502 | talloc = xfs_inobt_irec_to_allocmask(trec); | |
503 | salloc = xfs_inobt_irec_to_allocmask(srec); | |
504 | if (talloc & salloc) | |
505 | return false; | |
506 | ||
507 | return true; | |
508 | } | |
509 | ||
510 | /* | |
511 | * Merge the source inode record into the target. The caller must call | |
512 | * __xfs_inobt_can_merge() to ensure the merge is valid. | |
513 | */ | |
514 | STATIC void | |
515 | __xfs_inobt_rec_merge( | |
516 | struct xfs_inobt_rec_incore *trec, /* target */ | |
517 | struct xfs_inobt_rec_incore *srec) /* src */ | |
518 | { | |
519 | ASSERT(trec->ir_startino == srec->ir_startino); | |
520 | ||
521 | /* combine the counts */ | |
522 | trec->ir_count += srec->ir_count; | |
523 | trec->ir_freecount += srec->ir_freecount; | |
524 | ||
525 | /* | |
526 | * Merge the holemask and free mask. For both fields, 0 bits refer to | |
527 | * allocated inodes. We combine the allocated ranges with bitwise AND. | |
528 | */ | |
529 | trec->ir_holemask &= srec->ir_holemask; | |
530 | trec->ir_free &= srec->ir_free; | |
531 | } | |
532 | ||
533 | /* | |
534 | * Insert a new sparse inode chunk into the associated inode btree. The inode | |
535 | * record for the sparse chunk is pre-aligned to a startino that should match | |
536 | * any pre-existing sparse inode record in the tree. This allows sparse chunks | |
537 | * to fill over time. | |
538 | * | |
539 | * This function supports two modes of handling preexisting records depending on | |
540 | * the merge flag. If merge is true, the provided record is merged with the | |
541 | * existing record and updated in place. The merged record is returned in nrec. | |
542 | * If merge is false, an existing record is replaced with the provided record. | |
543 | * If no preexisting record exists, the provided record is always inserted. | |
544 | * | |
545 | * It is considered corruption if a merge is requested and not possible. Given | |
546 | * the sparse inode alignment constraints, this should never happen. | |
547 | */ | |
548 | STATIC int | |
549 | xfs_inobt_insert_sprec( | |
550 | struct xfs_mount *mp, | |
551 | struct xfs_trans *tp, | |
552 | struct xfs_buf *agbp, | |
553 | int btnum, | |
554 | struct xfs_inobt_rec_incore *nrec, /* in/out: new/merged rec. */ | |
555 | bool merge) /* merge or replace */ | |
556 | { | |
557 | struct xfs_btree_cur *cur; | |
558 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
559 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
560 | int error; | |
561 | int i; | |
562 | struct xfs_inobt_rec_incore rec; | |
563 | ||
564 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum); | |
565 | ||
566 | /* the new record is pre-aligned so we know where to look */ | |
567 | error = xfs_inobt_lookup(cur, nrec->ir_startino, XFS_LOOKUP_EQ, &i); | |
568 | if (error) | |
569 | goto error; | |
570 | /* if nothing there, insert a new record and return */ | |
571 | if (i == 0) { | |
572 | error = xfs_inobt_insert_rec(cur, nrec->ir_holemask, | |
573 | nrec->ir_count, nrec->ir_freecount, | |
574 | nrec->ir_free, &i); | |
575 | if (error) | |
576 | goto error; | |
577 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error); | |
578 | ||
579 | goto out; | |
580 | } | |
581 | ||
582 | /* | |
583 | * A record exists at this startino. Merge or replace the record | |
584 | * depending on what we've been asked to do. | |
585 | */ | |
586 | if (merge) { | |
587 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
588 | if (error) | |
589 | goto error; | |
590 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error); | |
591 | XFS_WANT_CORRUPTED_GOTO(mp, | |
592 | rec.ir_startino == nrec->ir_startino, | |
593 | error); | |
594 | ||
595 | /* | |
596 | * This should never fail. If we have coexisting records that | |
597 | * cannot merge, something is seriously wrong. | |
598 | */ | |
599 | XFS_WANT_CORRUPTED_GOTO(mp, __xfs_inobt_can_merge(nrec, &rec), | |
600 | error); | |
601 | ||
602 | trace_xfs_irec_merge_pre(mp, agno, rec.ir_startino, | |
603 | rec.ir_holemask, nrec->ir_startino, | |
604 | nrec->ir_holemask); | |
605 | ||
606 | /* merge to nrec to output the updated record */ | |
607 | __xfs_inobt_rec_merge(nrec, &rec); | |
608 | ||
609 | trace_xfs_irec_merge_post(mp, agno, nrec->ir_startino, | |
610 | nrec->ir_holemask); | |
611 | ||
612 | error = xfs_inobt_rec_check_count(mp, nrec); | |
613 | if (error) | |
614 | goto error; | |
615 | } | |
616 | ||
617 | error = xfs_inobt_update(cur, nrec); | |
618 | if (error) | |
619 | goto error; | |
620 | ||
621 | out: | |
622 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
623 | return 0; | |
624 | error: | |
625 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
626 | return error; | |
627 | } | |
628 | ||
1da177e4 LT |
629 | /* |
630 | * Allocate new inodes in the allocation group specified by agbp. | |
631 | * Return 0 for success, else error code. | |
632 | */ | |
633 | STATIC int /* error code or 0 */ | |
634 | xfs_ialloc_ag_alloc( | |
635 | xfs_trans_t *tp, /* transaction pointer */ | |
636 | xfs_buf_t *agbp, /* alloc group buffer */ | |
637 | int *alloc) | |
638 | { | |
639 | xfs_agi_t *agi; /* allocation group header */ | |
640 | xfs_alloc_arg_t args; /* allocation argument structure */ | |
92821e2b | 641 | xfs_agnumber_t agno; |
1da177e4 | 642 | int error; |
1da177e4 LT |
643 | xfs_agino_t newino; /* new first inode's number */ |
644 | xfs_agino_t newlen; /* new number of inodes */ | |
3ccb8b5f | 645 | int isaligned = 0; /* inode allocation at stripe unit */ |
1da177e4 | 646 | /* boundary */ |
56d1115c BF |
647 | uint16_t allocmask = (uint16_t) -1; /* init. to full chunk */ |
648 | struct xfs_inobt_rec_incore rec; | |
44b56e0a | 649 | struct xfs_perag *pag; |
1cdadee1 BF |
650 | int do_sparse = 0; |
651 | ||
a0041684 | 652 | memset(&args, 0, sizeof(args)); |
1da177e4 LT |
653 | args.tp = tp; |
654 | args.mp = tp->t_mountp; | |
1cdadee1 | 655 | args.fsbno = NULLFSBLOCK; |
340785cc | 656 | xfs_rmap_ag_owner(&args.oinfo, XFS_RMAP_OWN_INODES); |
1da177e4 | 657 | |
46fc58da BF |
658 | #ifdef DEBUG |
659 | /* randomly do sparse inode allocations */ | |
660 | if (xfs_sb_version_hassparseinodes(&tp->t_mountp->m_sb) && | |
661 | args.mp->m_ialloc_min_blks < args.mp->m_ialloc_blks) | |
662 | do_sparse = prandom_u32() & 1; | |
663 | #endif | |
664 | ||
1da177e4 LT |
665 | /* |
666 | * Locking will ensure that we don't have two callers in here | |
667 | * at one time. | |
668 | */ | |
71783438 | 669 | newlen = args.mp->m_ialloc_inos; |
1da177e4 | 670 | if (args.mp->m_maxicount && |
74f9ce1c | 671 | percpu_counter_read_positive(&args.mp->m_icount) + newlen > |
501ab323 | 672 | args.mp->m_maxicount) |
2451337d | 673 | return -ENOSPC; |
126cd105 | 674 | args.minlen = args.maxlen = args.mp->m_ialloc_blks; |
1da177e4 | 675 | /* |
3ccb8b5f GO |
676 | * First try to allocate inodes contiguous with the last-allocated |
677 | * chunk of inodes. If the filesystem is striped, this will fill | |
678 | * an entire stripe unit with inodes. | |
28c8e41a | 679 | */ |
1da177e4 | 680 | agi = XFS_BUF_TO_AGI(agbp); |
3ccb8b5f | 681 | newino = be32_to_cpu(agi->agi_newino); |
85c0b2ab | 682 | agno = be32_to_cpu(agi->agi_seqno); |
019ff2d5 | 683 | args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) + |
126cd105 | 684 | args.mp->m_ialloc_blks; |
1cdadee1 BF |
685 | if (do_sparse) |
686 | goto sparse_alloc; | |
019ff2d5 NS |
687 | if (likely(newino != NULLAGINO && |
688 | (args.agbno < be32_to_cpu(agi->agi_length)))) { | |
85c0b2ab | 689 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
3ccb8b5f | 690 | args.type = XFS_ALLOCTYPE_THIS_BNO; |
3ccb8b5f | 691 | args.prod = 1; |
75de2a91 | 692 | |
3ccb8b5f | 693 | /* |
75de2a91 DC |
694 | * We need to take into account alignment here to ensure that |
695 | * we don't modify the free list if we fail to have an exact | |
696 | * block. If we don't have an exact match, and every oher | |
697 | * attempt allocation attempt fails, we'll end up cancelling | |
698 | * a dirty transaction and shutting down. | |
699 | * | |
700 | * For an exact allocation, alignment must be 1, | |
701 | * however we need to take cluster alignment into account when | |
702 | * fixing up the freelist. Use the minalignslop field to | |
703 | * indicate that extra blocks might be required for alignment, | |
704 | * but not to use them in the actual exact allocation. | |
3ccb8b5f | 705 | */ |
75de2a91 | 706 | args.alignment = 1; |
7a1df156 | 707 | args.minalignslop = xfs_ialloc_cluster_alignment(args.mp) - 1; |
75de2a91 DC |
708 | |
709 | /* Allow space for the inode btree to split. */ | |
0d87e656 | 710 | args.minleft = args.mp->m_in_maxlevels - 1; |
3ccb8b5f GO |
711 | if ((error = xfs_alloc_vextent(&args))) |
712 | return error; | |
e480a723 BF |
713 | |
714 | /* | |
715 | * This request might have dirtied the transaction if the AG can | |
716 | * satisfy the request, but the exact block was not available. | |
717 | * If the allocation did fail, subsequent requests will relax | |
718 | * the exact agbno requirement and increase the alignment | |
719 | * instead. It is critical that the total size of the request | |
720 | * (len + alignment + slop) does not increase from this point | |
721 | * on, so reset minalignslop to ensure it is not included in | |
722 | * subsequent requests. | |
723 | */ | |
724 | args.minalignslop = 0; | |
1cdadee1 | 725 | } |
1da177e4 | 726 | |
3ccb8b5f GO |
727 | if (unlikely(args.fsbno == NULLFSBLOCK)) { |
728 | /* | |
729 | * Set the alignment for the allocation. | |
730 | * If stripe alignment is turned on then align at stripe unit | |
731 | * boundary. | |
019ff2d5 NS |
732 | * If the cluster size is smaller than a filesystem block |
733 | * then we're doing I/O for inodes in filesystem block size | |
3ccb8b5f GO |
734 | * pieces, so don't need alignment anyway. |
735 | */ | |
736 | isaligned = 0; | |
737 | if (args.mp->m_sinoalign) { | |
738 | ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN)); | |
739 | args.alignment = args.mp->m_dalign; | |
740 | isaligned = 1; | |
75de2a91 | 741 | } else |
7a1df156 | 742 | args.alignment = xfs_ialloc_cluster_alignment(args.mp); |
3ccb8b5f GO |
743 | /* |
744 | * Need to figure out where to allocate the inode blocks. | |
745 | * Ideally they should be spaced out through the a.g. | |
746 | * For now, just allocate blocks up front. | |
747 | */ | |
748 | args.agbno = be32_to_cpu(agi->agi_root); | |
85c0b2ab | 749 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
3ccb8b5f GO |
750 | /* |
751 | * Allocate a fixed-size extent of inodes. | |
752 | */ | |
753 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
3ccb8b5f GO |
754 | args.prod = 1; |
755 | /* | |
756 | * Allow space for the inode btree to split. | |
757 | */ | |
0d87e656 | 758 | args.minleft = args.mp->m_in_maxlevels - 1; |
3ccb8b5f GO |
759 | if ((error = xfs_alloc_vextent(&args))) |
760 | return error; | |
761 | } | |
019ff2d5 | 762 | |
1da177e4 LT |
763 | /* |
764 | * If stripe alignment is turned on, then try again with cluster | |
765 | * alignment. | |
766 | */ | |
767 | if (isaligned && args.fsbno == NULLFSBLOCK) { | |
768 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
16259e7d | 769 | args.agbno = be32_to_cpu(agi->agi_root); |
85c0b2ab | 770 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
7a1df156 | 771 | args.alignment = xfs_ialloc_cluster_alignment(args.mp); |
1da177e4 LT |
772 | if ((error = xfs_alloc_vextent(&args))) |
773 | return error; | |
774 | } | |
775 | ||
56d1115c BF |
776 | /* |
777 | * Finally, try a sparse allocation if the filesystem supports it and | |
778 | * the sparse allocation length is smaller than a full chunk. | |
779 | */ | |
780 | if (xfs_sb_version_hassparseinodes(&args.mp->m_sb) && | |
781 | args.mp->m_ialloc_min_blks < args.mp->m_ialloc_blks && | |
782 | args.fsbno == NULLFSBLOCK) { | |
1cdadee1 | 783 | sparse_alloc: |
56d1115c BF |
784 | args.type = XFS_ALLOCTYPE_NEAR_BNO; |
785 | args.agbno = be32_to_cpu(agi->agi_root); | |
786 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); | |
787 | args.alignment = args.mp->m_sb.sb_spino_align; | |
788 | args.prod = 1; | |
789 | ||
790 | args.minlen = args.mp->m_ialloc_min_blks; | |
791 | args.maxlen = args.minlen; | |
792 | ||
793 | /* | |
794 | * The inode record will be aligned to full chunk size. We must | |
795 | * prevent sparse allocation from AG boundaries that result in | |
796 | * invalid inode records, such as records that start at agbno 0 | |
797 | * or extend beyond the AG. | |
798 | * | |
799 | * Set min agbno to the first aligned, non-zero agbno and max to | |
800 | * the last aligned agbno that is at least one full chunk from | |
801 | * the end of the AG. | |
802 | */ | |
803 | args.min_agbno = args.mp->m_sb.sb_inoalignmt; | |
804 | args.max_agbno = round_down(args.mp->m_sb.sb_agblocks, | |
805 | args.mp->m_sb.sb_inoalignmt) - | |
806 | args.mp->m_ialloc_blks; | |
807 | ||
808 | error = xfs_alloc_vextent(&args); | |
809 | if (error) | |
810 | return error; | |
811 | ||
812 | newlen = args.len << args.mp->m_sb.sb_inopblog; | |
46fc58da | 813 | ASSERT(newlen <= XFS_INODES_PER_CHUNK); |
56d1115c BF |
814 | allocmask = (1 << (newlen / XFS_INODES_PER_HOLEMASK_BIT)) - 1; |
815 | } | |
816 | ||
1da177e4 LT |
817 | if (args.fsbno == NULLFSBLOCK) { |
818 | *alloc = 0; | |
819 | return 0; | |
820 | } | |
821 | ASSERT(args.len == args.minlen); | |
1da177e4 | 822 | |
359346a9 | 823 | /* |
85c0b2ab DC |
824 | * Stamp and write the inode buffers. |
825 | * | |
359346a9 DC |
826 | * Seed the new inode cluster with a random generation number. This |
827 | * prevents short-term reuse of generation numbers if a chunk is | |
828 | * freed and then immediately reallocated. We use random numbers | |
829 | * rather than a linear progression to prevent the next generation | |
830 | * number from being easily guessable. | |
831 | */ | |
463958af BF |
832 | error = xfs_ialloc_inode_init(args.mp, tp, NULL, newlen, agno, |
833 | args.agbno, args.len, prandom_u32()); | |
d42f08f6 | 834 | |
2a30f36d CS |
835 | if (error) |
836 | return error; | |
85c0b2ab DC |
837 | /* |
838 | * Convert the results. | |
839 | */ | |
840 | newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0); | |
56d1115c BF |
841 | |
842 | if (xfs_inobt_issparse(~allocmask)) { | |
843 | /* | |
844 | * We've allocated a sparse chunk. Align the startino and mask. | |
845 | */ | |
846 | xfs_align_sparse_ino(args.mp, &newino, &allocmask); | |
847 | ||
848 | rec.ir_startino = newino; | |
849 | rec.ir_holemask = ~allocmask; | |
850 | rec.ir_count = newlen; | |
851 | rec.ir_freecount = newlen; | |
852 | rec.ir_free = XFS_INOBT_ALL_FREE; | |
853 | ||
854 | /* | |
855 | * Insert the sparse record into the inobt and allow for a merge | |
856 | * if necessary. If a merge does occur, rec is updated to the | |
857 | * merged record. | |
858 | */ | |
859 | error = xfs_inobt_insert_sprec(args.mp, tp, agbp, XFS_BTNUM_INO, | |
860 | &rec, true); | |
861 | if (error == -EFSCORRUPTED) { | |
862 | xfs_alert(args.mp, | |
863 | "invalid sparse inode record: ino 0x%llx holemask 0x%x count %u", | |
864 | XFS_AGINO_TO_INO(args.mp, agno, | |
865 | rec.ir_startino), | |
866 | rec.ir_holemask, rec.ir_count); | |
867 | xfs_force_shutdown(args.mp, SHUTDOWN_CORRUPT_INCORE); | |
868 | } | |
869 | if (error) | |
870 | return error; | |
871 | ||
872 | /* | |
873 | * We can't merge the part we've just allocated as for the inobt | |
874 | * due to finobt semantics. The original record may or may not | |
875 | * exist independent of whether physical inodes exist in this | |
876 | * sparse chunk. | |
877 | * | |
878 | * We must update the finobt record based on the inobt record. | |
879 | * rec contains the fully merged and up to date inobt record | |
880 | * from the previous call. Set merge false to replace any | |
881 | * existing record with this one. | |
882 | */ | |
883 | if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) { | |
884 | error = xfs_inobt_insert_sprec(args.mp, tp, agbp, | |
885 | XFS_BTNUM_FINO, &rec, | |
886 | false); | |
887 | if (error) | |
888 | return error; | |
889 | } | |
890 | } else { | |
891 | /* full chunk - insert new records to both btrees */ | |
892 | error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen, | |
893 | XFS_BTNUM_INO); | |
894 | if (error) | |
895 | return error; | |
896 | ||
897 | if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) { | |
898 | error = xfs_inobt_insert(args.mp, tp, agbp, newino, | |
899 | newlen, XFS_BTNUM_FINO); | |
900 | if (error) | |
901 | return error; | |
902 | } | |
903 | } | |
904 | ||
905 | /* | |
906 | * Update AGI counts and newino. | |
907 | */ | |
413d57c9 MS |
908 | be32_add_cpu(&agi->agi_count, newlen); |
909 | be32_add_cpu(&agi->agi_freecount, newlen); | |
44b56e0a DC |
910 | pag = xfs_perag_get(args.mp, agno); |
911 | pag->pagi_freecount += newlen; | |
912 | xfs_perag_put(pag); | |
16259e7d | 913 | agi->agi_newino = cpu_to_be32(newino); |
85c0b2ab | 914 | |
1da177e4 LT |
915 | /* |
916 | * Log allocation group header fields | |
917 | */ | |
918 | xfs_ialloc_log_agi(tp, agbp, | |
919 | XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO); | |
920 | /* | |
921 | * Modify/log superblock values for inode count and inode free count. | |
922 | */ | |
923 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen); | |
924 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen); | |
925 | *alloc = 1; | |
926 | return 0; | |
927 | } | |
928 | ||
b8f82a4a | 929 | STATIC xfs_agnumber_t |
1da177e4 LT |
930 | xfs_ialloc_next_ag( |
931 | xfs_mount_t *mp) | |
932 | { | |
933 | xfs_agnumber_t agno; | |
934 | ||
935 | spin_lock(&mp->m_agirotor_lock); | |
936 | agno = mp->m_agirotor; | |
8aea3ff4 | 937 | if (++mp->m_agirotor >= mp->m_maxagi) |
1da177e4 LT |
938 | mp->m_agirotor = 0; |
939 | spin_unlock(&mp->m_agirotor_lock); | |
940 | ||
941 | return agno; | |
942 | } | |
943 | ||
944 | /* | |
945 | * Select an allocation group to look for a free inode in, based on the parent | |
2f21ff1c | 946 | * inode and the mode. Return the allocation group buffer. |
1da177e4 | 947 | */ |
55d6af64 | 948 | STATIC xfs_agnumber_t |
1da177e4 LT |
949 | xfs_ialloc_ag_select( |
950 | xfs_trans_t *tp, /* transaction pointer */ | |
951 | xfs_ino_t parent, /* parent directory inode number */ | |
f59cf5c2 | 952 | umode_t mode) /* bits set to indicate file type */ |
1da177e4 | 953 | { |
1da177e4 LT |
954 | xfs_agnumber_t agcount; /* number of ag's in the filesystem */ |
955 | xfs_agnumber_t agno; /* current ag number */ | |
956 | int flags; /* alloc buffer locking flags */ | |
957 | xfs_extlen_t ineed; /* blocks needed for inode allocation */ | |
958 | xfs_extlen_t longest = 0; /* longest extent available */ | |
959 | xfs_mount_t *mp; /* mount point structure */ | |
960 | int needspace; /* file mode implies space allocated */ | |
961 | xfs_perag_t *pag; /* per allocation group data */ | |
962 | xfs_agnumber_t pagno; /* parent (starting) ag number */ | |
55d6af64 | 963 | int error; |
1da177e4 LT |
964 | |
965 | /* | |
966 | * Files of these types need at least one block if length > 0 | |
967 | * (and they won't fit in the inode, but that's hard to figure out). | |
968 | */ | |
969 | needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode); | |
970 | mp = tp->t_mountp; | |
971 | agcount = mp->m_maxagi; | |
972 | if (S_ISDIR(mode)) | |
973 | pagno = xfs_ialloc_next_ag(mp); | |
974 | else { | |
975 | pagno = XFS_INO_TO_AGNO(mp, parent); | |
976 | if (pagno >= agcount) | |
977 | pagno = 0; | |
978 | } | |
55d6af64 | 979 | |
1da177e4 | 980 | ASSERT(pagno < agcount); |
55d6af64 | 981 | |
1da177e4 LT |
982 | /* |
983 | * Loop through allocation groups, looking for one with a little | |
984 | * free space in it. Note we don't look for free inodes, exactly. | |
985 | * Instead, we include whether there is a need to allocate inodes | |
986 | * to mean that blocks must be allocated for them, | |
987 | * if none are currently free. | |
988 | */ | |
989 | agno = pagno; | |
990 | flags = XFS_ALLOC_FLAG_TRYLOCK; | |
1da177e4 | 991 | for (;;) { |
44b56e0a | 992 | pag = xfs_perag_get(mp, agno); |
55d6af64 CH |
993 | if (!pag->pagi_inodeok) { |
994 | xfs_ialloc_next_ag(mp); | |
995 | goto nextag; | |
996 | } | |
997 | ||
1da177e4 | 998 | if (!pag->pagi_init) { |
55d6af64 CH |
999 | error = xfs_ialloc_pagi_init(mp, tp, agno); |
1000 | if (error) | |
1da177e4 | 1001 | goto nextag; |
55d6af64 | 1002 | } |
1da177e4 | 1003 | |
55d6af64 CH |
1004 | if (pag->pagi_freecount) { |
1005 | xfs_perag_put(pag); | |
1006 | return agno; | |
1da177e4 LT |
1007 | } |
1008 | ||
55d6af64 CH |
1009 | if (!pag->pagf_init) { |
1010 | error = xfs_alloc_pagf_init(mp, tp, agno, flags); | |
1011 | if (error) | |
1da177e4 | 1012 | goto nextag; |
1da177e4 | 1013 | } |
55d6af64 CH |
1014 | |
1015 | /* | |
7a1df156 DC |
1016 | * Check that there is enough free space for the file plus a |
1017 | * chunk of inodes if we need to allocate some. If this is the | |
1018 | * first pass across the AGs, take into account the potential | |
1019 | * space needed for alignment of inode chunks when checking the | |
1020 | * longest contiguous free space in the AG - this prevents us | |
1021 | * from getting ENOSPC because we have free space larger than | |
1022 | * m_ialloc_blks but alignment constraints prevent us from using | |
1023 | * it. | |
1024 | * | |
1025 | * If we can't find an AG with space for full alignment slack to | |
1026 | * be taken into account, we must be near ENOSPC in all AGs. | |
1027 | * Hence we don't include alignment for the second pass and so | |
1028 | * if we fail allocation due to alignment issues then it is most | |
1029 | * likely a real ENOSPC condition. | |
55d6af64 | 1030 | */ |
066a1884 | 1031 | ineed = mp->m_ialloc_min_blks; |
7a1df156 DC |
1032 | if (flags && ineed > 1) |
1033 | ineed += xfs_ialloc_cluster_alignment(mp); | |
55d6af64 CH |
1034 | longest = pag->pagf_longest; |
1035 | if (!longest) | |
1036 | longest = pag->pagf_flcount > 0; | |
1037 | ||
1038 | if (pag->pagf_freeblks >= needspace + ineed && | |
1039 | longest >= ineed) { | |
1040 | xfs_perag_put(pag); | |
1041 | return agno; | |
1da177e4 | 1042 | } |
1da177e4 | 1043 | nextag: |
44b56e0a | 1044 | xfs_perag_put(pag); |
1da177e4 LT |
1045 | /* |
1046 | * No point in iterating over the rest, if we're shutting | |
1047 | * down. | |
1048 | */ | |
1c1c6ebc | 1049 | if (XFS_FORCED_SHUTDOWN(mp)) |
55d6af64 | 1050 | return NULLAGNUMBER; |
1da177e4 LT |
1051 | agno++; |
1052 | if (agno >= agcount) | |
1053 | agno = 0; | |
1054 | if (agno == pagno) { | |
1c1c6ebc | 1055 | if (flags == 0) |
55d6af64 | 1056 | return NULLAGNUMBER; |
1da177e4 LT |
1057 | flags = 0; |
1058 | } | |
1059 | } | |
1060 | } | |
1061 | ||
4254b0bb CH |
1062 | /* |
1063 | * Try to retrieve the next record to the left/right from the current one. | |
1064 | */ | |
1065 | STATIC int | |
1066 | xfs_ialloc_next_rec( | |
1067 | struct xfs_btree_cur *cur, | |
1068 | xfs_inobt_rec_incore_t *rec, | |
1069 | int *done, | |
1070 | int left) | |
1071 | { | |
1072 | int error; | |
1073 | int i; | |
1074 | ||
1075 | if (left) | |
1076 | error = xfs_btree_decrement(cur, 0, &i); | |
1077 | else | |
1078 | error = xfs_btree_increment(cur, 0, &i); | |
1079 | ||
1080 | if (error) | |
1081 | return error; | |
1082 | *done = !i; | |
1083 | if (i) { | |
1084 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1085 | if (error) | |
1086 | return error; | |
5fb5aeee | 1087 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
4254b0bb CH |
1088 | } |
1089 | ||
1090 | return 0; | |
1091 | } | |
1092 | ||
bd169565 DC |
1093 | STATIC int |
1094 | xfs_ialloc_get_rec( | |
1095 | struct xfs_btree_cur *cur, | |
1096 | xfs_agino_t agino, | |
1097 | xfs_inobt_rec_incore_t *rec, | |
43df2ee6 | 1098 | int *done) |
bd169565 DC |
1099 | { |
1100 | int error; | |
1101 | int i; | |
1102 | ||
1103 | error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i); | |
1104 | if (error) | |
1105 | return error; | |
1106 | *done = !i; | |
1107 | if (i) { | |
1108 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1109 | if (error) | |
1110 | return error; | |
5fb5aeee | 1111 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
bd169565 DC |
1112 | } |
1113 | ||
1114 | return 0; | |
1115 | } | |
0b48db80 | 1116 | |
d4cc540b | 1117 | /* |
26dd5217 BF |
1118 | * Return the offset of the first free inode in the record. If the inode chunk |
1119 | * is sparsely allocated, we convert the record holemask to inode granularity | |
1120 | * and mask off the unallocated regions from the inode free mask. | |
d4cc540b BF |
1121 | */ |
1122 | STATIC int | |
1123 | xfs_inobt_first_free_inode( | |
1124 | struct xfs_inobt_rec_incore *rec) | |
1125 | { | |
26dd5217 BF |
1126 | xfs_inofree_t realfree; |
1127 | ||
1128 | /* if there are no holes, return the first available offset */ | |
1129 | if (!xfs_inobt_issparse(rec->ir_holemask)) | |
1130 | return xfs_lowbit64(rec->ir_free); | |
1131 | ||
1132 | realfree = xfs_inobt_irec_to_allocmask(rec); | |
1133 | realfree &= rec->ir_free; | |
1134 | ||
1135 | return xfs_lowbit64(realfree); | |
d4cc540b BF |
1136 | } |
1137 | ||
1da177e4 | 1138 | /* |
6dd8638e | 1139 | * Allocate an inode using the inobt-only algorithm. |
1da177e4 | 1140 | */ |
f2ecc5e4 | 1141 | STATIC int |
6dd8638e | 1142 | xfs_dialloc_ag_inobt( |
f2ecc5e4 CH |
1143 | struct xfs_trans *tp, |
1144 | struct xfs_buf *agbp, | |
1145 | xfs_ino_t parent, | |
1146 | xfs_ino_t *inop) | |
1da177e4 | 1147 | { |
f2ecc5e4 CH |
1148 | struct xfs_mount *mp = tp->t_mountp; |
1149 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
1150 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
1151 | xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent); | |
1152 | xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent); | |
1153 | struct xfs_perag *pag; | |
1154 | struct xfs_btree_cur *cur, *tcur; | |
1155 | struct xfs_inobt_rec_incore rec, trec; | |
1156 | xfs_ino_t ino; | |
1157 | int error; | |
1158 | int offset; | |
1159 | int i, j; | |
2d32311c | 1160 | int searchdistance = 10; |
1da177e4 | 1161 | |
44b56e0a | 1162 | pag = xfs_perag_get(mp, agno); |
bd169565 | 1163 | |
4bb61069 CH |
1164 | ASSERT(pag->pagi_init); |
1165 | ASSERT(pag->pagi_inodeok); | |
1166 | ASSERT(pag->pagi_freecount > 0); | |
1167 | ||
bd169565 | 1168 | restart_pagno: |
57bd3dbe | 1169 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
1da177e4 LT |
1170 | /* |
1171 | * If pagino is 0 (this is the root inode allocation) use newino. | |
1172 | * This must work because we've just allocated some. | |
1173 | */ | |
1174 | if (!pagino) | |
16259e7d | 1175 | pagino = be32_to_cpu(agi->agi_newino); |
1da177e4 | 1176 | |
0b48db80 DC |
1177 | error = xfs_check_agi_freecount(cur, agi); |
1178 | if (error) | |
1179 | goto error0; | |
1da177e4 | 1180 | |
1da177e4 | 1181 | /* |
4254b0bb | 1182 | * If in the same AG as the parent, try to get near the parent. |
1da177e4 LT |
1183 | */ |
1184 | if (pagno == agno) { | |
4254b0bb CH |
1185 | int doneleft; /* done, to the left */ |
1186 | int doneright; /* done, to the right */ | |
1187 | ||
21875505 | 1188 | error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i); |
4254b0bb | 1189 | if (error) |
1da177e4 | 1190 | goto error0; |
c29aad41 | 1191 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
4254b0bb CH |
1192 | |
1193 | error = xfs_inobt_get_rec(cur, &rec, &j); | |
1194 | if (error) | |
1195 | goto error0; | |
c29aad41 | 1196 | XFS_WANT_CORRUPTED_GOTO(mp, j == 1, error0); |
4254b0bb CH |
1197 | |
1198 | if (rec.ir_freecount > 0) { | |
1da177e4 LT |
1199 | /* |
1200 | * Found a free inode in the same chunk | |
4254b0bb | 1201 | * as the parent, done. |
1da177e4 | 1202 | */ |
4254b0bb | 1203 | goto alloc_inode; |
1da177e4 | 1204 | } |
4254b0bb CH |
1205 | |
1206 | ||
1da177e4 | 1207 | /* |
4254b0bb | 1208 | * In the same AG as parent, but parent's chunk is full. |
1da177e4 | 1209 | */ |
1da177e4 | 1210 | |
4254b0bb CH |
1211 | /* duplicate the cursor, search left & right simultaneously */ |
1212 | error = xfs_btree_dup_cursor(cur, &tcur); | |
1213 | if (error) | |
1214 | goto error0; | |
1215 | ||
bd169565 DC |
1216 | /* |
1217 | * Skip to last blocks looked up if same parent inode. | |
1218 | */ | |
1219 | if (pagino != NULLAGINO && | |
1220 | pag->pagl_pagino == pagino && | |
1221 | pag->pagl_leftrec != NULLAGINO && | |
1222 | pag->pagl_rightrec != NULLAGINO) { | |
1223 | error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec, | |
43df2ee6 | 1224 | &trec, &doneleft); |
bd169565 DC |
1225 | if (error) |
1226 | goto error1; | |
4254b0bb | 1227 | |
bd169565 | 1228 | error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec, |
43df2ee6 | 1229 | &rec, &doneright); |
bd169565 DC |
1230 | if (error) |
1231 | goto error1; | |
1232 | } else { | |
1233 | /* search left with tcur, back up 1 record */ | |
1234 | error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1); | |
1235 | if (error) | |
1236 | goto error1; | |
1237 | ||
1238 | /* search right with cur, go forward 1 record. */ | |
1239 | error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0); | |
1240 | if (error) | |
1241 | goto error1; | |
1242 | } | |
4254b0bb CH |
1243 | |
1244 | /* | |
1245 | * Loop until we find an inode chunk with a free inode. | |
1246 | */ | |
2d32311c | 1247 | while (--searchdistance > 0 && (!doneleft || !doneright)) { |
4254b0bb CH |
1248 | int useleft; /* using left inode chunk this time */ |
1249 | ||
1250 | /* figure out the closer block if both are valid. */ | |
1251 | if (!doneleft && !doneright) { | |
1252 | useleft = pagino - | |
1253 | (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) < | |
1254 | rec.ir_startino - pagino; | |
1255 | } else { | |
1256 | useleft = !doneleft; | |
1da177e4 | 1257 | } |
4254b0bb CH |
1258 | |
1259 | /* free inodes to the left? */ | |
1260 | if (useleft && trec.ir_freecount) { | |
4254b0bb CH |
1261 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
1262 | cur = tcur; | |
bd169565 DC |
1263 | |
1264 | pag->pagl_leftrec = trec.ir_startino; | |
1265 | pag->pagl_rightrec = rec.ir_startino; | |
1266 | pag->pagl_pagino = pagino; | |
c44245b3 | 1267 | rec = trec; |
4254b0bb | 1268 | goto alloc_inode; |
1da177e4 | 1269 | } |
1da177e4 | 1270 | |
4254b0bb CH |
1271 | /* free inodes to the right? */ |
1272 | if (!useleft && rec.ir_freecount) { | |
1273 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
bd169565 DC |
1274 | |
1275 | pag->pagl_leftrec = trec.ir_startino; | |
1276 | pag->pagl_rightrec = rec.ir_startino; | |
1277 | pag->pagl_pagino = pagino; | |
4254b0bb | 1278 | goto alloc_inode; |
1da177e4 | 1279 | } |
4254b0bb CH |
1280 | |
1281 | /* get next record to check */ | |
1282 | if (useleft) { | |
1283 | error = xfs_ialloc_next_rec(tcur, &trec, | |
1284 | &doneleft, 1); | |
1285 | } else { | |
1286 | error = xfs_ialloc_next_rec(cur, &rec, | |
1287 | &doneright, 0); | |
1288 | } | |
1289 | if (error) | |
1290 | goto error1; | |
1da177e4 | 1291 | } |
bd169565 | 1292 | |
2d32311c CM |
1293 | if (searchdistance <= 0) { |
1294 | /* | |
1295 | * Not in range - save last search | |
1296 | * location and allocate a new inode | |
1297 | */ | |
1298 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
1299 | pag->pagl_leftrec = trec.ir_startino; | |
1300 | pag->pagl_rightrec = rec.ir_startino; | |
1301 | pag->pagl_pagino = pagino; | |
1302 | ||
1303 | } else { | |
1304 | /* | |
1305 | * We've reached the end of the btree. because | |
1306 | * we are only searching a small chunk of the | |
1307 | * btree each search, there is obviously free | |
1308 | * inodes closer to the parent inode than we | |
1309 | * are now. restart the search again. | |
1310 | */ | |
1311 | pag->pagl_pagino = NULLAGINO; | |
1312 | pag->pagl_leftrec = NULLAGINO; | |
1313 | pag->pagl_rightrec = NULLAGINO; | |
1314 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
1315 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
1316 | goto restart_pagno; | |
1317 | } | |
1da177e4 | 1318 | } |
4254b0bb | 1319 | |
1da177e4 | 1320 | /* |
4254b0bb | 1321 | * In a different AG from the parent. |
1da177e4 LT |
1322 | * See if the most recently allocated block has any free. |
1323 | */ | |
69ef921b | 1324 | if (agi->agi_newino != cpu_to_be32(NULLAGINO)) { |
21875505 CH |
1325 | error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino), |
1326 | XFS_LOOKUP_EQ, &i); | |
4254b0bb | 1327 | if (error) |
1da177e4 | 1328 | goto error0; |
4254b0bb CH |
1329 | |
1330 | if (i == 1) { | |
1331 | error = xfs_inobt_get_rec(cur, &rec, &j); | |
1332 | if (error) | |
1333 | goto error0; | |
1334 | ||
1335 | if (j == 1 && rec.ir_freecount > 0) { | |
1336 | /* | |
1337 | * The last chunk allocated in the group | |
1338 | * still has a free inode. | |
1339 | */ | |
1340 | goto alloc_inode; | |
1341 | } | |
1da177e4 | 1342 | } |
bd169565 | 1343 | } |
4254b0bb | 1344 | |
bd169565 DC |
1345 | /* |
1346 | * None left in the last group, search the whole AG | |
1347 | */ | |
1348 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); | |
1349 | if (error) | |
1350 | goto error0; | |
c29aad41 | 1351 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
bd169565 DC |
1352 | |
1353 | for (;;) { | |
1354 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
1355 | if (error) | |
1356 | goto error0; | |
c29aad41 | 1357 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
bd169565 DC |
1358 | if (rec.ir_freecount > 0) |
1359 | break; | |
1360 | error = xfs_btree_increment(cur, 0, &i); | |
4254b0bb CH |
1361 | if (error) |
1362 | goto error0; | |
c29aad41 | 1363 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 | 1364 | } |
4254b0bb CH |
1365 | |
1366 | alloc_inode: | |
d4cc540b | 1367 | offset = xfs_inobt_first_free_inode(&rec); |
1da177e4 LT |
1368 | ASSERT(offset >= 0); |
1369 | ASSERT(offset < XFS_INODES_PER_CHUNK); | |
1370 | ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) % | |
1371 | XFS_INODES_PER_CHUNK) == 0); | |
1372 | ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset); | |
0d87e656 | 1373 | rec.ir_free &= ~XFS_INOBT_MASK(offset); |
1da177e4 | 1374 | rec.ir_freecount--; |
afabc24a CH |
1375 | error = xfs_inobt_update(cur, &rec); |
1376 | if (error) | |
1da177e4 | 1377 | goto error0; |
413d57c9 | 1378 | be32_add_cpu(&agi->agi_freecount, -1); |
1da177e4 | 1379 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); |
44b56e0a | 1380 | pag->pagi_freecount--; |
1da177e4 | 1381 | |
0b48db80 DC |
1382 | error = xfs_check_agi_freecount(cur, agi); |
1383 | if (error) | |
1384 | goto error0; | |
1385 | ||
1da177e4 LT |
1386 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
1387 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1); | |
44b56e0a | 1388 | xfs_perag_put(pag); |
1da177e4 LT |
1389 | *inop = ino; |
1390 | return 0; | |
1391 | error1: | |
1392 | xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR); | |
1393 | error0: | |
1394 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
44b56e0a | 1395 | xfs_perag_put(pag); |
1da177e4 LT |
1396 | return error; |
1397 | } | |
1398 | ||
6dd8638e BF |
1399 | /* |
1400 | * Use the free inode btree to allocate an inode based on distance from the | |
1401 | * parent. Note that the provided cursor may be deleted and replaced. | |
1402 | */ | |
1403 | STATIC int | |
1404 | xfs_dialloc_ag_finobt_near( | |
1405 | xfs_agino_t pagino, | |
1406 | struct xfs_btree_cur **ocur, | |
1407 | struct xfs_inobt_rec_incore *rec) | |
1408 | { | |
1409 | struct xfs_btree_cur *lcur = *ocur; /* left search cursor */ | |
1410 | struct xfs_btree_cur *rcur; /* right search cursor */ | |
1411 | struct xfs_inobt_rec_incore rrec; | |
1412 | int error; | |
1413 | int i, j; | |
1414 | ||
1415 | error = xfs_inobt_lookup(lcur, pagino, XFS_LOOKUP_LE, &i); | |
1416 | if (error) | |
1417 | return error; | |
1418 | ||
1419 | if (i == 1) { | |
1420 | error = xfs_inobt_get_rec(lcur, rec, &i); | |
1421 | if (error) | |
1422 | return error; | |
5fb5aeee | 1423 | XFS_WANT_CORRUPTED_RETURN(lcur->bc_mp, i == 1); |
6dd8638e BF |
1424 | |
1425 | /* | |
1426 | * See if we've landed in the parent inode record. The finobt | |
1427 | * only tracks chunks with at least one free inode, so record | |
1428 | * existence is enough. | |
1429 | */ | |
1430 | if (pagino >= rec->ir_startino && | |
1431 | pagino < (rec->ir_startino + XFS_INODES_PER_CHUNK)) | |
1432 | return 0; | |
1433 | } | |
1434 | ||
1435 | error = xfs_btree_dup_cursor(lcur, &rcur); | |
1436 | if (error) | |
1437 | return error; | |
1438 | ||
1439 | error = xfs_inobt_lookup(rcur, pagino, XFS_LOOKUP_GE, &j); | |
1440 | if (error) | |
1441 | goto error_rcur; | |
1442 | if (j == 1) { | |
1443 | error = xfs_inobt_get_rec(rcur, &rrec, &j); | |
1444 | if (error) | |
1445 | goto error_rcur; | |
c29aad41 | 1446 | XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, j == 1, error_rcur); |
6dd8638e BF |
1447 | } |
1448 | ||
c29aad41 | 1449 | XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, i == 1 || j == 1, error_rcur); |
6dd8638e BF |
1450 | if (i == 1 && j == 1) { |
1451 | /* | |
1452 | * Both the left and right records are valid. Choose the closer | |
1453 | * inode chunk to the target. | |
1454 | */ | |
1455 | if ((pagino - rec->ir_startino + XFS_INODES_PER_CHUNK - 1) > | |
1456 | (rrec.ir_startino - pagino)) { | |
1457 | *rec = rrec; | |
1458 | xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR); | |
1459 | *ocur = rcur; | |
1460 | } else { | |
1461 | xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR); | |
1462 | } | |
1463 | } else if (j == 1) { | |
1464 | /* only the right record is valid */ | |
1465 | *rec = rrec; | |
1466 | xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR); | |
1467 | *ocur = rcur; | |
1468 | } else if (i == 1) { | |
1469 | /* only the left record is valid */ | |
1470 | xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR); | |
1471 | } | |
1472 | ||
1473 | return 0; | |
1474 | ||
1475 | error_rcur: | |
1476 | xfs_btree_del_cursor(rcur, XFS_BTREE_ERROR); | |
1477 | return error; | |
1478 | } | |
1479 | ||
1480 | /* | |
1481 | * Use the free inode btree to find a free inode based on a newino hint. If | |
1482 | * the hint is NULL, find the first free inode in the AG. | |
1483 | */ | |
1484 | STATIC int | |
1485 | xfs_dialloc_ag_finobt_newino( | |
1486 | struct xfs_agi *agi, | |
1487 | struct xfs_btree_cur *cur, | |
1488 | struct xfs_inobt_rec_incore *rec) | |
1489 | { | |
1490 | int error; | |
1491 | int i; | |
1492 | ||
1493 | if (agi->agi_newino != cpu_to_be32(NULLAGINO)) { | |
e68ed775 DC |
1494 | error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino), |
1495 | XFS_LOOKUP_EQ, &i); | |
6dd8638e BF |
1496 | if (error) |
1497 | return error; | |
1498 | if (i == 1) { | |
1499 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1500 | if (error) | |
1501 | return error; | |
5fb5aeee | 1502 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
6dd8638e BF |
1503 | return 0; |
1504 | } | |
1505 | } | |
1506 | ||
1507 | /* | |
1508 | * Find the first inode available in the AG. | |
1509 | */ | |
1510 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); | |
1511 | if (error) | |
1512 | return error; | |
5fb5aeee | 1513 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
6dd8638e BF |
1514 | |
1515 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1516 | if (error) | |
1517 | return error; | |
5fb5aeee | 1518 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
6dd8638e BF |
1519 | |
1520 | return 0; | |
1521 | } | |
1522 | ||
1523 | /* | |
1524 | * Update the inobt based on a modification made to the finobt. Also ensure that | |
1525 | * the records from both trees are equivalent post-modification. | |
1526 | */ | |
1527 | STATIC int | |
1528 | xfs_dialloc_ag_update_inobt( | |
1529 | struct xfs_btree_cur *cur, /* inobt cursor */ | |
1530 | struct xfs_inobt_rec_incore *frec, /* finobt record */ | |
1531 | int offset) /* inode offset */ | |
1532 | { | |
1533 | struct xfs_inobt_rec_incore rec; | |
1534 | int error; | |
1535 | int i; | |
1536 | ||
1537 | error = xfs_inobt_lookup(cur, frec->ir_startino, XFS_LOOKUP_EQ, &i); | |
1538 | if (error) | |
1539 | return error; | |
5fb5aeee | 1540 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
6dd8638e BF |
1541 | |
1542 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
1543 | if (error) | |
1544 | return error; | |
5fb5aeee | 1545 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
6dd8638e BF |
1546 | ASSERT((XFS_AGINO_TO_OFFSET(cur->bc_mp, rec.ir_startino) % |
1547 | XFS_INODES_PER_CHUNK) == 0); | |
1548 | ||
1549 | rec.ir_free &= ~XFS_INOBT_MASK(offset); | |
1550 | rec.ir_freecount--; | |
1551 | ||
5fb5aeee | 1552 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, (rec.ir_free == frec->ir_free) && |
6dd8638e BF |
1553 | (rec.ir_freecount == frec->ir_freecount)); |
1554 | ||
b72091f2 | 1555 | return xfs_inobt_update(cur, &rec); |
6dd8638e BF |
1556 | } |
1557 | ||
1558 | /* | |
1559 | * Allocate an inode using the free inode btree, if available. Otherwise, fall | |
1560 | * back to the inobt search algorithm. | |
1561 | * | |
1562 | * The caller selected an AG for us, and made sure that free inodes are | |
1563 | * available. | |
1564 | */ | |
1565 | STATIC int | |
1566 | xfs_dialloc_ag( | |
1567 | struct xfs_trans *tp, | |
1568 | struct xfs_buf *agbp, | |
1569 | xfs_ino_t parent, | |
1570 | xfs_ino_t *inop) | |
1571 | { | |
1572 | struct xfs_mount *mp = tp->t_mountp; | |
1573 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
1574 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
1575 | xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent); | |
1576 | xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent); | |
1577 | struct xfs_perag *pag; | |
1578 | struct xfs_btree_cur *cur; /* finobt cursor */ | |
1579 | struct xfs_btree_cur *icur; /* inobt cursor */ | |
1580 | struct xfs_inobt_rec_incore rec; | |
1581 | xfs_ino_t ino; | |
1582 | int error; | |
1583 | int offset; | |
1584 | int i; | |
1585 | ||
1586 | if (!xfs_sb_version_hasfinobt(&mp->m_sb)) | |
1587 | return xfs_dialloc_ag_inobt(tp, agbp, parent, inop); | |
1588 | ||
1589 | pag = xfs_perag_get(mp, agno); | |
1590 | ||
1591 | /* | |
1592 | * If pagino is 0 (this is the root inode allocation) use newino. | |
1593 | * This must work because we've just allocated some. | |
1594 | */ | |
1595 | if (!pagino) | |
1596 | pagino = be32_to_cpu(agi->agi_newino); | |
1597 | ||
1598 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO); | |
1599 | ||
1600 | error = xfs_check_agi_freecount(cur, agi); | |
1601 | if (error) | |
1602 | goto error_cur; | |
1603 | ||
1604 | /* | |
1605 | * The search algorithm depends on whether we're in the same AG as the | |
1606 | * parent. If so, find the closest available inode to the parent. If | |
1607 | * not, consider the agi hint or find the first free inode in the AG. | |
1608 | */ | |
1609 | if (agno == pagno) | |
1610 | error = xfs_dialloc_ag_finobt_near(pagino, &cur, &rec); | |
1611 | else | |
1612 | error = xfs_dialloc_ag_finobt_newino(agi, cur, &rec); | |
1613 | if (error) | |
1614 | goto error_cur; | |
1615 | ||
d4cc540b | 1616 | offset = xfs_inobt_first_free_inode(&rec); |
6dd8638e BF |
1617 | ASSERT(offset >= 0); |
1618 | ASSERT(offset < XFS_INODES_PER_CHUNK); | |
1619 | ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) % | |
1620 | XFS_INODES_PER_CHUNK) == 0); | |
1621 | ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset); | |
1622 | ||
1623 | /* | |
1624 | * Modify or remove the finobt record. | |
1625 | */ | |
1626 | rec.ir_free &= ~XFS_INOBT_MASK(offset); | |
1627 | rec.ir_freecount--; | |
1628 | if (rec.ir_freecount) | |
1629 | error = xfs_inobt_update(cur, &rec); | |
1630 | else | |
1631 | error = xfs_btree_delete(cur, &i); | |
1632 | if (error) | |
1633 | goto error_cur; | |
1634 | ||
1635 | /* | |
1636 | * The finobt has now been updated appropriately. We haven't updated the | |
1637 | * agi and superblock yet, so we can create an inobt cursor and validate | |
1638 | * the original freecount. If all is well, make the equivalent update to | |
1639 | * the inobt using the finobt record and offset information. | |
1640 | */ | |
1641 | icur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); | |
1642 | ||
1643 | error = xfs_check_agi_freecount(icur, agi); | |
1644 | if (error) | |
1645 | goto error_icur; | |
1646 | ||
1647 | error = xfs_dialloc_ag_update_inobt(icur, &rec, offset); | |
1648 | if (error) | |
1649 | goto error_icur; | |
1650 | ||
1651 | /* | |
1652 | * Both trees have now been updated. We must update the perag and | |
1653 | * superblock before we can check the freecount for each btree. | |
1654 | */ | |
1655 | be32_add_cpu(&agi->agi_freecount, -1); | |
1656 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); | |
1657 | pag->pagi_freecount--; | |
1658 | ||
1659 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1); | |
1660 | ||
1661 | error = xfs_check_agi_freecount(icur, agi); | |
1662 | if (error) | |
1663 | goto error_icur; | |
1664 | error = xfs_check_agi_freecount(cur, agi); | |
1665 | if (error) | |
1666 | goto error_icur; | |
1667 | ||
1668 | xfs_btree_del_cursor(icur, XFS_BTREE_NOERROR); | |
1669 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
1670 | xfs_perag_put(pag); | |
1671 | *inop = ino; | |
1672 | return 0; | |
1673 | ||
1674 | error_icur: | |
1675 | xfs_btree_del_cursor(icur, XFS_BTREE_ERROR); | |
1676 | error_cur: | |
1677 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
1678 | xfs_perag_put(pag); | |
1679 | return error; | |
1680 | } | |
1681 | ||
f2ecc5e4 CH |
1682 | /* |
1683 | * Allocate an inode on disk. | |
1684 | * | |
1685 | * Mode is used to tell whether the new inode will need space, and whether it | |
1686 | * is a directory. | |
1687 | * | |
1688 | * This function is designed to be called twice if it has to do an allocation | |
1689 | * to make more free inodes. On the first call, *IO_agbp should be set to NULL. | |
1690 | * If an inode is available without having to performn an allocation, an inode | |
cd856db6 CM |
1691 | * number is returned. In this case, *IO_agbp is set to NULL. If an allocation |
1692 | * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp. | |
1693 | * The caller should then commit the current transaction, allocate a | |
f2ecc5e4 CH |
1694 | * new transaction, and call xfs_dialloc() again, passing in the previous value |
1695 | * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI | |
1696 | * buffer is locked across the two calls, the second call is guaranteed to have | |
1697 | * a free inode available. | |
1698 | * | |
1699 | * Once we successfully pick an inode its number is returned and the on-disk | |
1700 | * data structures are updated. The inode itself is not read in, since doing so | |
1701 | * would break ordering constraints with xfs_reclaim. | |
1702 | */ | |
1703 | int | |
1704 | xfs_dialloc( | |
1705 | struct xfs_trans *tp, | |
1706 | xfs_ino_t parent, | |
1707 | umode_t mode, | |
f2ecc5e4 | 1708 | struct xfs_buf **IO_agbp, |
f2ecc5e4 CH |
1709 | xfs_ino_t *inop) |
1710 | { | |
55d6af64 | 1711 | struct xfs_mount *mp = tp->t_mountp; |
f2ecc5e4 CH |
1712 | struct xfs_buf *agbp; |
1713 | xfs_agnumber_t agno; | |
f2ecc5e4 CH |
1714 | int error; |
1715 | int ialloced; | |
1716 | int noroom = 0; | |
be60fe54 | 1717 | xfs_agnumber_t start_agno; |
f2ecc5e4 | 1718 | struct xfs_perag *pag; |
f59cf5c2 | 1719 | int okalloc = 1; |
f2ecc5e4 | 1720 | |
4bb61069 | 1721 | if (*IO_agbp) { |
f2ecc5e4 | 1722 | /* |
4bb61069 CH |
1723 | * If the caller passes in a pointer to the AGI buffer, |
1724 | * continue where we left off before. In this case, we | |
f2ecc5e4 CH |
1725 | * know that the allocation group has free inodes. |
1726 | */ | |
1727 | agbp = *IO_agbp; | |
4bb61069 | 1728 | goto out_alloc; |
f2ecc5e4 | 1729 | } |
4bb61069 CH |
1730 | |
1731 | /* | |
1732 | * We do not have an agbp, so select an initial allocation | |
1733 | * group for inode allocation. | |
1734 | */ | |
f59cf5c2 | 1735 | start_agno = xfs_ialloc_ag_select(tp, parent, mode); |
be60fe54 | 1736 | if (start_agno == NULLAGNUMBER) { |
4bb61069 CH |
1737 | *inop = NULLFSINO; |
1738 | return 0; | |
1739 | } | |
55d6af64 | 1740 | |
f2ecc5e4 CH |
1741 | /* |
1742 | * If we have already hit the ceiling of inode blocks then clear | |
1743 | * okalloc so we scan all available agi structures for a free | |
1744 | * inode. | |
74f9ce1c GW |
1745 | * |
1746 | * Read rough value of mp->m_icount by percpu_counter_read_positive, | |
1747 | * which will sacrifice the preciseness but improve the performance. | |
f2ecc5e4 | 1748 | */ |
f2ecc5e4 | 1749 | if (mp->m_maxicount && |
74f9ce1c GW |
1750 | percpu_counter_read_positive(&mp->m_icount) + mp->m_ialloc_inos |
1751 | > mp->m_maxicount) { | |
f2ecc5e4 CH |
1752 | noroom = 1; |
1753 | okalloc = 0; | |
1754 | } | |
1755 | ||
1756 | /* | |
1757 | * Loop until we find an allocation group that either has free inodes | |
1758 | * or in which we can allocate some inodes. Iterate through the | |
1759 | * allocation groups upward, wrapping at the end. | |
1760 | */ | |
be60fe54 CH |
1761 | agno = start_agno; |
1762 | for (;;) { | |
1763 | pag = xfs_perag_get(mp, agno); | |
1764 | if (!pag->pagi_inodeok) { | |
1765 | xfs_ialloc_next_ag(mp); | |
1766 | goto nextag; | |
1767 | } | |
1768 | ||
1769 | if (!pag->pagi_init) { | |
1770 | error = xfs_ialloc_pagi_init(mp, tp, agno); | |
1771 | if (error) | |
1772 | goto out_error; | |
f2ecc5e4 | 1773 | } |
be60fe54 | 1774 | |
f2ecc5e4 | 1775 | /* |
be60fe54 | 1776 | * Do a first racy fast path check if this AG is usable. |
f2ecc5e4 | 1777 | */ |
be60fe54 CH |
1778 | if (!pag->pagi_freecount && !okalloc) |
1779 | goto nextag; | |
1780 | ||
c4982110 CH |
1781 | /* |
1782 | * Then read in the AGI buffer and recheck with the AGI buffer | |
1783 | * lock held. | |
1784 | */ | |
be60fe54 CH |
1785 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); |
1786 | if (error) | |
1787 | goto out_error; | |
1788 | ||
be60fe54 CH |
1789 | if (pag->pagi_freecount) { |
1790 | xfs_perag_put(pag); | |
1791 | goto out_alloc; | |
1792 | } | |
1793 | ||
c4982110 CH |
1794 | if (!okalloc) |
1795 | goto nextag_relse_buffer; | |
1796 | ||
be60fe54 CH |
1797 | |
1798 | error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced); | |
1799 | if (error) { | |
1800 | xfs_trans_brelse(tp, agbp); | |
1801 | ||
2451337d | 1802 | if (error != -ENOSPC) |
be60fe54 CH |
1803 | goto out_error; |
1804 | ||
1805 | xfs_perag_put(pag); | |
f2ecc5e4 | 1806 | *inop = NULLFSINO; |
be60fe54 | 1807 | return 0; |
f2ecc5e4 | 1808 | } |
be60fe54 CH |
1809 | |
1810 | if (ialloced) { | |
1811 | /* | |
1812 | * We successfully allocated some inodes, return | |
1813 | * the current context to the caller so that it | |
1814 | * can commit the current transaction and call | |
1815 | * us again where we left off. | |
1816 | */ | |
1817 | ASSERT(pag->pagi_freecount > 0); | |
f2ecc5e4 | 1818 | xfs_perag_put(pag); |
be60fe54 CH |
1819 | |
1820 | *IO_agbp = agbp; | |
1821 | *inop = NULLFSINO; | |
1822 | return 0; | |
f2ecc5e4 | 1823 | } |
be60fe54 | 1824 | |
c4982110 CH |
1825 | nextag_relse_buffer: |
1826 | xfs_trans_brelse(tp, agbp); | |
be60fe54 | 1827 | nextag: |
f2ecc5e4 | 1828 | xfs_perag_put(pag); |
be60fe54 CH |
1829 | if (++agno == mp->m_sb.sb_agcount) |
1830 | agno = 0; | |
1831 | if (agno == start_agno) { | |
1832 | *inop = NULLFSINO; | |
2451337d | 1833 | return noroom ? -ENOSPC : 0; |
be60fe54 | 1834 | } |
f2ecc5e4 CH |
1835 | } |
1836 | ||
4bb61069 | 1837 | out_alloc: |
f2ecc5e4 CH |
1838 | *IO_agbp = NULL; |
1839 | return xfs_dialloc_ag(tp, agbp, parent, inop); | |
be60fe54 CH |
1840 | out_error: |
1841 | xfs_perag_put(pag); | |
b474c7ae | 1842 | return error; |
f2ecc5e4 CH |
1843 | } |
1844 | ||
10ae3dc7 BF |
1845 | /* |
1846 | * Free the blocks of an inode chunk. We must consider that the inode chunk | |
1847 | * might be sparse and only free the regions that are allocated as part of the | |
1848 | * chunk. | |
1849 | */ | |
1850 | STATIC void | |
1851 | xfs_difree_inode_chunk( | |
1852 | struct xfs_mount *mp, | |
1853 | xfs_agnumber_t agno, | |
1854 | struct xfs_inobt_rec_incore *rec, | |
2c3234d1 | 1855 | struct xfs_defer_ops *dfops) |
10ae3dc7 BF |
1856 | { |
1857 | xfs_agblock_t sagbno = XFS_AGINO_TO_AGBNO(mp, rec->ir_startino); | |
1858 | int startidx, endidx; | |
1859 | int nextbit; | |
1860 | xfs_agblock_t agbno; | |
1861 | int contigblk; | |
340785cc | 1862 | struct xfs_owner_info oinfo; |
10ae3dc7 | 1863 | DECLARE_BITMAP(holemask, XFS_INOBT_HOLEMASK_BITS); |
340785cc | 1864 | xfs_rmap_ag_owner(&oinfo, XFS_RMAP_OWN_INODES); |
10ae3dc7 BF |
1865 | |
1866 | if (!xfs_inobt_issparse(rec->ir_holemask)) { | |
1867 | /* not sparse, calculate extent info directly */ | |
2c3234d1 | 1868 | xfs_bmap_add_free(mp, dfops, XFS_AGB_TO_FSB(mp, agno, sagbno), |
340785cc | 1869 | mp->m_ialloc_blks, &oinfo); |
10ae3dc7 BF |
1870 | return; |
1871 | } | |
1872 | ||
1873 | /* holemask is only 16-bits (fits in an unsigned long) */ | |
1874 | ASSERT(sizeof(rec->ir_holemask) <= sizeof(holemask[0])); | |
1875 | holemask[0] = rec->ir_holemask; | |
1876 | ||
1877 | /* | |
1878 | * Find contiguous ranges of zeroes (i.e., allocated regions) in the | |
1879 | * holemask and convert the start/end index of each range to an extent. | |
1880 | * We start with the start and end index both pointing at the first 0 in | |
1881 | * the mask. | |
1882 | */ | |
1883 | startidx = endidx = find_first_zero_bit(holemask, | |
1884 | XFS_INOBT_HOLEMASK_BITS); | |
1885 | nextbit = startidx + 1; | |
1886 | while (startidx < XFS_INOBT_HOLEMASK_BITS) { | |
1887 | nextbit = find_next_zero_bit(holemask, XFS_INOBT_HOLEMASK_BITS, | |
1888 | nextbit); | |
1889 | /* | |
1890 | * If the next zero bit is contiguous, update the end index of | |
1891 | * the current range and continue. | |
1892 | */ | |
1893 | if (nextbit != XFS_INOBT_HOLEMASK_BITS && | |
1894 | nextbit == endidx + 1) { | |
1895 | endidx = nextbit; | |
1896 | goto next; | |
1897 | } | |
1898 | ||
1899 | /* | |
1900 | * nextbit is not contiguous with the current end index. Convert | |
1901 | * the current start/end to an extent and add it to the free | |
1902 | * list. | |
1903 | */ | |
1904 | agbno = sagbno + (startidx * XFS_INODES_PER_HOLEMASK_BIT) / | |
1905 | mp->m_sb.sb_inopblock; | |
1906 | contigblk = ((endidx - startidx + 1) * | |
1907 | XFS_INODES_PER_HOLEMASK_BIT) / | |
1908 | mp->m_sb.sb_inopblock; | |
1909 | ||
1910 | ASSERT(agbno % mp->m_sb.sb_spino_align == 0); | |
1911 | ASSERT(contigblk % mp->m_sb.sb_spino_align == 0); | |
2c3234d1 | 1912 | xfs_bmap_add_free(mp, dfops, XFS_AGB_TO_FSB(mp, agno, agbno), |
340785cc | 1913 | contigblk, &oinfo); |
10ae3dc7 BF |
1914 | |
1915 | /* reset range to current bit and carry on... */ | |
1916 | startidx = endidx = nextbit; | |
1917 | ||
1918 | next: | |
1919 | nextbit++; | |
1920 | } | |
1921 | } | |
1922 | ||
2b64ee5c BF |
1923 | STATIC int |
1924 | xfs_difree_inobt( | |
1925 | struct xfs_mount *mp, | |
1926 | struct xfs_trans *tp, | |
1927 | struct xfs_buf *agbp, | |
1928 | xfs_agino_t agino, | |
2c3234d1 | 1929 | struct xfs_defer_ops *dfops, |
09b56604 | 1930 | struct xfs_icluster *xic, |
2b64ee5c | 1931 | struct xfs_inobt_rec_incore *orec) |
1da177e4 | 1932 | { |
2b64ee5c BF |
1933 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); |
1934 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
1935 | struct xfs_perag *pag; | |
1936 | struct xfs_btree_cur *cur; | |
1937 | struct xfs_inobt_rec_incore rec; | |
1938 | int ilen; | |
1939 | int error; | |
1940 | int i; | |
1941 | int off; | |
1da177e4 | 1942 | |
69ef921b | 1943 | ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC)); |
2b64ee5c BF |
1944 | ASSERT(XFS_AGINO_TO_AGBNO(mp, agino) < be32_to_cpu(agi->agi_length)); |
1945 | ||
1da177e4 LT |
1946 | /* |
1947 | * Initialize the cursor. | |
1948 | */ | |
57bd3dbe | 1949 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
1da177e4 | 1950 | |
0b48db80 DC |
1951 | error = xfs_check_agi_freecount(cur, agi); |
1952 | if (error) | |
1953 | goto error0; | |
1954 | ||
1da177e4 LT |
1955 | /* |
1956 | * Look for the entry describing this inode. | |
1957 | */ | |
21875505 | 1958 | if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) { |
0b932ccc DC |
1959 | xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.", |
1960 | __func__, error); | |
1da177e4 LT |
1961 | goto error0; |
1962 | } | |
c29aad41 | 1963 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
2e287a73 CH |
1964 | error = xfs_inobt_get_rec(cur, &rec, &i); |
1965 | if (error) { | |
0b932ccc DC |
1966 | xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.", |
1967 | __func__, error); | |
1da177e4 LT |
1968 | goto error0; |
1969 | } | |
c29aad41 | 1970 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 LT |
1971 | /* |
1972 | * Get the offset in the inode chunk. | |
1973 | */ | |
1974 | off = agino - rec.ir_startino; | |
1975 | ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK); | |
0d87e656 | 1976 | ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off))); |
1da177e4 LT |
1977 | /* |
1978 | * Mark the inode free & increment the count. | |
1979 | */ | |
0d87e656 | 1980 | rec.ir_free |= XFS_INOBT_MASK(off); |
1da177e4 LT |
1981 | rec.ir_freecount++; |
1982 | ||
1983 | /* | |
999633d3 BF |
1984 | * When an inode chunk is free, it becomes eligible for removal. Don't |
1985 | * remove the chunk if the block size is large enough for multiple inode | |
1986 | * chunks (that might not be free). | |
1da177e4 | 1987 | */ |
1bd960ee | 1988 | if (!(mp->m_flags & XFS_MOUNT_IKEEP) && |
999633d3 BF |
1989 | rec.ir_free == XFS_INOBT_ALL_FREE && |
1990 | mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK) { | |
749f24f3 | 1991 | xic->deleted = true; |
09b56604 BF |
1992 | xic->first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino); |
1993 | xic->alloc = xfs_inobt_irec_to_allocmask(&rec); | |
1da177e4 LT |
1994 | |
1995 | /* | |
1996 | * Remove the inode cluster from the AGI B+Tree, adjust the | |
1997 | * AGI and Superblock inode counts, and mark the disk space | |
1998 | * to be freed when the transaction is committed. | |
1999 | */ | |
999633d3 | 2000 | ilen = rec.ir_freecount; |
413d57c9 MS |
2001 | be32_add_cpu(&agi->agi_count, -ilen); |
2002 | be32_add_cpu(&agi->agi_freecount, -(ilen - 1)); | |
1da177e4 | 2003 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT); |
44b56e0a DC |
2004 | pag = xfs_perag_get(mp, agno); |
2005 | pag->pagi_freecount -= ilen - 1; | |
2006 | xfs_perag_put(pag); | |
1da177e4 LT |
2007 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen); |
2008 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1)); | |
2009 | ||
91cca5df | 2010 | if ((error = xfs_btree_delete(cur, &i))) { |
0b932ccc DC |
2011 | xfs_warn(mp, "%s: xfs_btree_delete returned error %d.", |
2012 | __func__, error); | |
1da177e4 LT |
2013 | goto error0; |
2014 | } | |
2015 | ||
2c3234d1 | 2016 | xfs_difree_inode_chunk(mp, agno, &rec, dfops); |
1da177e4 | 2017 | } else { |
749f24f3 | 2018 | xic->deleted = false; |
1da177e4 | 2019 | |
afabc24a CH |
2020 | error = xfs_inobt_update(cur, &rec); |
2021 | if (error) { | |
0b932ccc DC |
2022 | xfs_warn(mp, "%s: xfs_inobt_update returned error %d.", |
2023 | __func__, error); | |
1da177e4 LT |
2024 | goto error0; |
2025 | } | |
afabc24a | 2026 | |
1da177e4 LT |
2027 | /* |
2028 | * Change the inode free counts and log the ag/sb changes. | |
2029 | */ | |
413d57c9 | 2030 | be32_add_cpu(&agi->agi_freecount, 1); |
1da177e4 | 2031 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); |
44b56e0a DC |
2032 | pag = xfs_perag_get(mp, agno); |
2033 | pag->pagi_freecount++; | |
2034 | xfs_perag_put(pag); | |
1da177e4 LT |
2035 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1); |
2036 | } | |
2037 | ||
0b48db80 DC |
2038 | error = xfs_check_agi_freecount(cur, agi); |
2039 | if (error) | |
2040 | goto error0; | |
1da177e4 | 2041 | |
2b64ee5c | 2042 | *orec = rec; |
1da177e4 LT |
2043 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
2044 | return 0; | |
2045 | ||
2046 | error0: | |
2047 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
2048 | return error; | |
2049 | } | |
2050 | ||
3efa4ffd BF |
2051 | /* |
2052 | * Free an inode in the free inode btree. | |
2053 | */ | |
2054 | STATIC int | |
2055 | xfs_difree_finobt( | |
2056 | struct xfs_mount *mp, | |
2057 | struct xfs_trans *tp, | |
2058 | struct xfs_buf *agbp, | |
2059 | xfs_agino_t agino, | |
2060 | struct xfs_inobt_rec_incore *ibtrec) /* inobt record */ | |
2061 | { | |
2062 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
2063 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
2064 | struct xfs_btree_cur *cur; | |
2065 | struct xfs_inobt_rec_incore rec; | |
2066 | int offset = agino - ibtrec->ir_startino; | |
2067 | int error; | |
2068 | int i; | |
2069 | ||
2070 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO); | |
2071 | ||
2072 | error = xfs_inobt_lookup(cur, ibtrec->ir_startino, XFS_LOOKUP_EQ, &i); | |
2073 | if (error) | |
2074 | goto error; | |
2075 | if (i == 0) { | |
2076 | /* | |
2077 | * If the record does not exist in the finobt, we must have just | |
2078 | * freed an inode in a previously fully allocated chunk. If not, | |
2079 | * something is out of sync. | |
2080 | */ | |
c29aad41 | 2081 | XFS_WANT_CORRUPTED_GOTO(mp, ibtrec->ir_freecount == 1, error); |
3efa4ffd | 2082 | |
5419040f BF |
2083 | error = xfs_inobt_insert_rec(cur, ibtrec->ir_holemask, |
2084 | ibtrec->ir_count, | |
2085 | ibtrec->ir_freecount, | |
3efa4ffd BF |
2086 | ibtrec->ir_free, &i); |
2087 | if (error) | |
2088 | goto error; | |
2089 | ASSERT(i == 1); | |
2090 | ||
2091 | goto out; | |
2092 | } | |
2093 | ||
2094 | /* | |
2095 | * Read and update the existing record. We could just copy the ibtrec | |
2096 | * across here, but that would defeat the purpose of having redundant | |
2097 | * metadata. By making the modifications independently, we can catch | |
2098 | * corruptions that we wouldn't see if we just copied from one record | |
2099 | * to another. | |
2100 | */ | |
2101 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
2102 | if (error) | |
2103 | goto error; | |
c29aad41 | 2104 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error); |
3efa4ffd BF |
2105 | |
2106 | rec.ir_free |= XFS_INOBT_MASK(offset); | |
2107 | rec.ir_freecount++; | |
2108 | ||
c29aad41 | 2109 | XFS_WANT_CORRUPTED_GOTO(mp, (rec.ir_free == ibtrec->ir_free) && |
3efa4ffd BF |
2110 | (rec.ir_freecount == ibtrec->ir_freecount), |
2111 | error); | |
2112 | ||
2113 | /* | |
2114 | * The content of inobt records should always match between the inobt | |
2115 | * and finobt. The lifecycle of records in the finobt is different from | |
2116 | * the inobt in that the finobt only tracks records with at least one | |
2117 | * free inode. Hence, if all of the inodes are free and we aren't | |
2118 | * keeping inode chunks permanently on disk, remove the record. | |
2119 | * Otherwise, update the record with the new information. | |
999633d3 BF |
2120 | * |
2121 | * Note that we currently can't free chunks when the block size is large | |
2122 | * enough for multiple chunks. Leave the finobt record to remain in sync | |
2123 | * with the inobt. | |
3efa4ffd | 2124 | */ |
999633d3 BF |
2125 | if (rec.ir_free == XFS_INOBT_ALL_FREE && |
2126 | mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK && | |
3efa4ffd BF |
2127 | !(mp->m_flags & XFS_MOUNT_IKEEP)) { |
2128 | error = xfs_btree_delete(cur, &i); | |
2129 | if (error) | |
2130 | goto error; | |
2131 | ASSERT(i == 1); | |
2132 | } else { | |
2133 | error = xfs_inobt_update(cur, &rec); | |
2134 | if (error) | |
2135 | goto error; | |
2136 | } | |
2137 | ||
2138 | out: | |
2139 | error = xfs_check_agi_freecount(cur, agi); | |
2140 | if (error) | |
2141 | goto error; | |
2142 | ||
2143 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
2144 | return 0; | |
2145 | ||
2146 | error: | |
2147 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
2148 | return error; | |
2149 | } | |
2150 | ||
2b64ee5c BF |
2151 | /* |
2152 | * Free disk inode. Carefully avoids touching the incore inode, all | |
2153 | * manipulations incore are the caller's responsibility. | |
2154 | * The on-disk inode is not changed by this operation, only the | |
2155 | * btree (free inode mask) is changed. | |
2156 | */ | |
2157 | int | |
2158 | xfs_difree( | |
2159 | struct xfs_trans *tp, /* transaction pointer */ | |
2160 | xfs_ino_t inode, /* inode to be freed */ | |
2c3234d1 | 2161 | struct xfs_defer_ops *dfops, /* extents to free */ |
09b56604 | 2162 | struct xfs_icluster *xic) /* cluster info if deleted */ |
2b64ee5c BF |
2163 | { |
2164 | /* REFERENCED */ | |
2165 | xfs_agblock_t agbno; /* block number containing inode */ | |
2166 | struct xfs_buf *agbp; /* buffer for allocation group header */ | |
2167 | xfs_agino_t agino; /* allocation group inode number */ | |
2168 | xfs_agnumber_t agno; /* allocation group number */ | |
2169 | int error; /* error return value */ | |
2170 | struct xfs_mount *mp; /* mount structure for filesystem */ | |
2171 | struct xfs_inobt_rec_incore rec;/* btree record */ | |
2172 | ||
2173 | mp = tp->t_mountp; | |
2174 | ||
2175 | /* | |
2176 | * Break up inode number into its components. | |
2177 | */ | |
2178 | agno = XFS_INO_TO_AGNO(mp, inode); | |
2179 | if (agno >= mp->m_sb.sb_agcount) { | |
2180 | xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).", | |
2181 | __func__, agno, mp->m_sb.sb_agcount); | |
2182 | ASSERT(0); | |
2451337d | 2183 | return -EINVAL; |
2b64ee5c BF |
2184 | } |
2185 | agino = XFS_INO_TO_AGINO(mp, inode); | |
2186 | if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
2187 | xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).", | |
2188 | __func__, (unsigned long long)inode, | |
2189 | (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino)); | |
2190 | ASSERT(0); | |
2451337d | 2191 | return -EINVAL; |
2b64ee5c BF |
2192 | } |
2193 | agbno = XFS_AGINO_TO_AGBNO(mp, agino); | |
2194 | if (agbno >= mp->m_sb.sb_agblocks) { | |
2195 | xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).", | |
2196 | __func__, agbno, mp->m_sb.sb_agblocks); | |
2197 | ASSERT(0); | |
2451337d | 2198 | return -EINVAL; |
2b64ee5c BF |
2199 | } |
2200 | /* | |
2201 | * Get the allocation group header. | |
2202 | */ | |
2203 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); | |
2204 | if (error) { | |
2205 | xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.", | |
2206 | __func__, error); | |
2207 | return error; | |
2208 | } | |
2209 | ||
2210 | /* | |
2211 | * Fix up the inode allocation btree. | |
2212 | */ | |
2c3234d1 | 2213 | error = xfs_difree_inobt(mp, tp, agbp, agino, dfops, xic, &rec); |
2b64ee5c BF |
2214 | if (error) |
2215 | goto error0; | |
2216 | ||
3efa4ffd BF |
2217 | /* |
2218 | * Fix up the free inode btree. | |
2219 | */ | |
2220 | if (xfs_sb_version_hasfinobt(&mp->m_sb)) { | |
2221 | error = xfs_difree_finobt(mp, tp, agbp, agino, &rec); | |
2222 | if (error) | |
2223 | goto error0; | |
2224 | } | |
2225 | ||
2b64ee5c BF |
2226 | return 0; |
2227 | ||
2228 | error0: | |
2229 | return error; | |
2230 | } | |
2231 | ||
7124fe0a DC |
2232 | STATIC int |
2233 | xfs_imap_lookup( | |
2234 | struct xfs_mount *mp, | |
2235 | struct xfs_trans *tp, | |
2236 | xfs_agnumber_t agno, | |
2237 | xfs_agino_t agino, | |
2238 | xfs_agblock_t agbno, | |
2239 | xfs_agblock_t *chunk_agbno, | |
2240 | xfs_agblock_t *offset_agbno, | |
2241 | int flags) | |
2242 | { | |
2243 | struct xfs_inobt_rec_incore rec; | |
2244 | struct xfs_btree_cur *cur; | |
2245 | struct xfs_buf *agbp; | |
7124fe0a DC |
2246 | int error; |
2247 | int i; | |
2248 | ||
2249 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); | |
2250 | if (error) { | |
53487786 DC |
2251 | xfs_alert(mp, |
2252 | "%s: xfs_ialloc_read_agi() returned error %d, agno %d", | |
2253 | __func__, error, agno); | |
7124fe0a DC |
2254 | return error; |
2255 | } | |
2256 | ||
2257 | /* | |
4536f2ad DC |
2258 | * Lookup the inode record for the given agino. If the record cannot be |
2259 | * found, then it's an invalid inode number and we should abort. Once | |
2260 | * we have a record, we need to ensure it contains the inode number | |
2261 | * we are looking up. | |
7124fe0a | 2262 | */ |
57bd3dbe | 2263 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
4536f2ad | 2264 | error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i); |
7124fe0a DC |
2265 | if (!error) { |
2266 | if (i) | |
2267 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
2268 | if (!error && i == 0) | |
2451337d | 2269 | error = -EINVAL; |
7124fe0a DC |
2270 | } |
2271 | ||
2272 | xfs_trans_brelse(tp, agbp); | |
f307080a | 2273 | xfs_btree_del_cursor(cur, error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); |
7124fe0a DC |
2274 | if (error) |
2275 | return error; | |
2276 | ||
4536f2ad DC |
2277 | /* check that the returned record contains the required inode */ |
2278 | if (rec.ir_startino > agino || | |
71783438 | 2279 | rec.ir_startino + mp->m_ialloc_inos <= agino) |
2451337d | 2280 | return -EINVAL; |
4536f2ad | 2281 | |
7124fe0a | 2282 | /* for untrusted inodes check it is allocated first */ |
1920779e | 2283 | if ((flags & XFS_IGET_UNTRUSTED) && |
7124fe0a | 2284 | (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino))) |
2451337d | 2285 | return -EINVAL; |
7124fe0a DC |
2286 | |
2287 | *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino); | |
2288 | *offset_agbno = agbno - *chunk_agbno; | |
2289 | return 0; | |
2290 | } | |
2291 | ||
1da177e4 | 2292 | /* |
94e1b69d | 2293 | * Return the location of the inode in imap, for mapping it into a buffer. |
1da177e4 | 2294 | */ |
1da177e4 | 2295 | int |
94e1b69d CH |
2296 | xfs_imap( |
2297 | xfs_mount_t *mp, /* file system mount structure */ | |
2298 | xfs_trans_t *tp, /* transaction pointer */ | |
1da177e4 | 2299 | xfs_ino_t ino, /* inode to locate */ |
94e1b69d CH |
2300 | struct xfs_imap *imap, /* location map structure */ |
2301 | uint flags) /* flags for inode btree lookup */ | |
1da177e4 LT |
2302 | { |
2303 | xfs_agblock_t agbno; /* block number of inode in the alloc group */ | |
1da177e4 LT |
2304 | xfs_agino_t agino; /* inode number within alloc group */ |
2305 | xfs_agnumber_t agno; /* allocation group number */ | |
2306 | int blks_per_cluster; /* num blocks per inode cluster */ | |
2307 | xfs_agblock_t chunk_agbno; /* first block in inode chunk */ | |
1da177e4 | 2308 | xfs_agblock_t cluster_agbno; /* first block in inode cluster */ |
1da177e4 | 2309 | int error; /* error code */ |
1da177e4 | 2310 | int offset; /* index of inode in its buffer */ |
836a94ad | 2311 | xfs_agblock_t offset_agbno; /* blks from chunk start to inode */ |
1da177e4 LT |
2312 | |
2313 | ASSERT(ino != NULLFSINO); | |
94e1b69d | 2314 | |
1da177e4 LT |
2315 | /* |
2316 | * Split up the inode number into its parts. | |
2317 | */ | |
2318 | agno = XFS_INO_TO_AGNO(mp, ino); | |
2319 | agino = XFS_INO_TO_AGINO(mp, ino); | |
2320 | agbno = XFS_AGINO_TO_AGBNO(mp, agino); | |
2321 | if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks || | |
2322 | ino != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
2323 | #ifdef DEBUG | |
1920779e DC |
2324 | /* |
2325 | * Don't output diagnostic information for untrusted inodes | |
2326 | * as they can be invalid without implying corruption. | |
2327 | */ | |
2328 | if (flags & XFS_IGET_UNTRUSTED) | |
2451337d | 2329 | return -EINVAL; |
1da177e4 | 2330 | if (agno >= mp->m_sb.sb_agcount) { |
53487786 DC |
2331 | xfs_alert(mp, |
2332 | "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)", | |
2333 | __func__, agno, mp->m_sb.sb_agcount); | |
1da177e4 LT |
2334 | } |
2335 | if (agbno >= mp->m_sb.sb_agblocks) { | |
53487786 DC |
2336 | xfs_alert(mp, |
2337 | "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)", | |
2338 | __func__, (unsigned long long)agbno, | |
2339 | (unsigned long)mp->m_sb.sb_agblocks); | |
1da177e4 LT |
2340 | } |
2341 | if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
53487786 DC |
2342 | xfs_alert(mp, |
2343 | "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)", | |
2344 | __func__, ino, | |
2345 | XFS_AGINO_TO_INO(mp, agno, agino)); | |
1da177e4 | 2346 | } |
745b1f47 | 2347 | xfs_stack_trace(); |
1da177e4 | 2348 | #endif /* DEBUG */ |
2451337d | 2349 | return -EINVAL; |
1da177e4 | 2350 | } |
94e1b69d | 2351 | |
f9e5abcf | 2352 | blks_per_cluster = xfs_icluster_size_fsb(mp); |
7124fe0a DC |
2353 | |
2354 | /* | |
2355 | * For bulkstat and handle lookups, we have an untrusted inode number | |
2356 | * that we have to verify is valid. We cannot do this just by reading | |
2357 | * the inode buffer as it may have been unlinked and removed leaving | |
2358 | * inodes in stale state on disk. Hence we have to do a btree lookup | |
2359 | * in all cases where an untrusted inode number is passed. | |
2360 | */ | |
1920779e | 2361 | if (flags & XFS_IGET_UNTRUSTED) { |
7124fe0a DC |
2362 | error = xfs_imap_lookup(mp, tp, agno, agino, agbno, |
2363 | &chunk_agbno, &offset_agbno, flags); | |
2364 | if (error) | |
2365 | return error; | |
2366 | goto out_map; | |
2367 | } | |
2368 | ||
94e1b69d CH |
2369 | /* |
2370 | * If the inode cluster size is the same as the blocksize or | |
2371 | * smaller we get to the buffer by simple arithmetics. | |
2372 | */ | |
f9e5abcf | 2373 | if (blks_per_cluster == 1) { |
1da177e4 LT |
2374 | offset = XFS_INO_TO_OFFSET(mp, ino); |
2375 | ASSERT(offset < mp->m_sb.sb_inopblock); | |
94e1b69d CH |
2376 | |
2377 | imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno); | |
2378 | imap->im_len = XFS_FSB_TO_BB(mp, 1); | |
755c7bf5 DW |
2379 | imap->im_boffset = (unsigned short)(offset << |
2380 | mp->m_sb.sb_inodelog); | |
1da177e4 LT |
2381 | return 0; |
2382 | } | |
94e1b69d | 2383 | |
94e1b69d CH |
2384 | /* |
2385 | * If the inode chunks are aligned then use simple maths to | |
2386 | * find the location. Otherwise we have to do a btree | |
2387 | * lookup to find the location. | |
2388 | */ | |
1da177e4 LT |
2389 | if (mp->m_inoalign_mask) { |
2390 | offset_agbno = agbno & mp->m_inoalign_mask; | |
2391 | chunk_agbno = agbno - offset_agbno; | |
2392 | } else { | |
7124fe0a DC |
2393 | error = xfs_imap_lookup(mp, tp, agno, agino, agbno, |
2394 | &chunk_agbno, &offset_agbno, flags); | |
1da177e4 LT |
2395 | if (error) |
2396 | return error; | |
1da177e4 | 2397 | } |
94e1b69d | 2398 | |
7124fe0a | 2399 | out_map: |
1da177e4 LT |
2400 | ASSERT(agbno >= chunk_agbno); |
2401 | cluster_agbno = chunk_agbno + | |
2402 | ((offset_agbno / blks_per_cluster) * blks_per_cluster); | |
2403 | offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) + | |
2404 | XFS_INO_TO_OFFSET(mp, ino); | |
94e1b69d CH |
2405 | |
2406 | imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno); | |
2407 | imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster); | |
755c7bf5 | 2408 | imap->im_boffset = (unsigned short)(offset << mp->m_sb.sb_inodelog); |
94e1b69d CH |
2409 | |
2410 | /* | |
2411 | * If the inode number maps to a block outside the bounds | |
2412 | * of the file system then return NULL rather than calling | |
2413 | * read_buf and panicing when we get an error from the | |
2414 | * driver. | |
2415 | */ | |
2416 | if ((imap->im_blkno + imap->im_len) > | |
2417 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) { | |
53487786 DC |
2418 | xfs_alert(mp, |
2419 | "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)", | |
2420 | __func__, (unsigned long long) imap->im_blkno, | |
94e1b69d CH |
2421 | (unsigned long long) imap->im_len, |
2422 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)); | |
2451337d | 2423 | return -EINVAL; |
94e1b69d | 2424 | } |
1da177e4 | 2425 | return 0; |
1da177e4 LT |
2426 | } |
2427 | ||
2428 | /* | |
2429 | * Compute and fill in value of m_in_maxlevels. | |
2430 | */ | |
2431 | void | |
2432 | xfs_ialloc_compute_maxlevels( | |
2433 | xfs_mount_t *mp) /* file system mount structure */ | |
2434 | { | |
19b54ee6 DW |
2435 | uint inodes; |
2436 | ||
2437 | inodes = (1LL << XFS_INO_AGINO_BITS(mp)) >> XFS_INODES_PER_CHUNK_LOG; | |
a1f69417 | 2438 | mp->m_in_maxlevels = xfs_btree_compute_maxlevels(mp->m_inobt_mnr, |
19b54ee6 | 2439 | inodes); |
1da177e4 LT |
2440 | } |
2441 | ||
2442 | /* | |
aafc3c24 BF |
2443 | * Log specified fields for the ag hdr (inode section). The growth of the agi |
2444 | * structure over time requires that we interpret the buffer as two logical | |
2445 | * regions delineated by the end of the unlinked list. This is due to the size | |
2446 | * of the hash table and its location in the middle of the agi. | |
2447 | * | |
2448 | * For example, a request to log a field before agi_unlinked and a field after | |
2449 | * agi_unlinked could cause us to log the entire hash table and use an excessive | |
2450 | * amount of log space. To avoid this behavior, log the region up through | |
2451 | * agi_unlinked in one call and the region after agi_unlinked through the end of | |
2452 | * the structure in another. | |
1da177e4 LT |
2453 | */ |
2454 | void | |
2455 | xfs_ialloc_log_agi( | |
2456 | xfs_trans_t *tp, /* transaction pointer */ | |
2457 | xfs_buf_t *bp, /* allocation group header buffer */ | |
2458 | int fields) /* bitmask of fields to log */ | |
2459 | { | |
2460 | int first; /* first byte number */ | |
2461 | int last; /* last byte number */ | |
2462 | static const short offsets[] = { /* field starting offsets */ | |
2463 | /* keep in sync with bit definitions */ | |
2464 | offsetof(xfs_agi_t, agi_magicnum), | |
2465 | offsetof(xfs_agi_t, agi_versionnum), | |
2466 | offsetof(xfs_agi_t, agi_seqno), | |
2467 | offsetof(xfs_agi_t, agi_length), | |
2468 | offsetof(xfs_agi_t, agi_count), | |
2469 | offsetof(xfs_agi_t, agi_root), | |
2470 | offsetof(xfs_agi_t, agi_level), | |
2471 | offsetof(xfs_agi_t, agi_freecount), | |
2472 | offsetof(xfs_agi_t, agi_newino), | |
2473 | offsetof(xfs_agi_t, agi_dirino), | |
2474 | offsetof(xfs_agi_t, agi_unlinked), | |
aafc3c24 BF |
2475 | offsetof(xfs_agi_t, agi_free_root), |
2476 | offsetof(xfs_agi_t, agi_free_level), | |
1da177e4 LT |
2477 | sizeof(xfs_agi_t) |
2478 | }; | |
2479 | #ifdef DEBUG | |
2480 | xfs_agi_t *agi; /* allocation group header */ | |
2481 | ||
2482 | agi = XFS_BUF_TO_AGI(bp); | |
69ef921b | 2483 | ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC)); |
1da177e4 | 2484 | #endif |
aafc3c24 | 2485 | |
1da177e4 | 2486 | /* |
aafc3c24 BF |
2487 | * Compute byte offsets for the first and last fields in the first |
2488 | * region and log the agi buffer. This only logs up through | |
2489 | * agi_unlinked. | |
1da177e4 | 2490 | */ |
aafc3c24 BF |
2491 | if (fields & XFS_AGI_ALL_BITS_R1) { |
2492 | xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R1, | |
2493 | &first, &last); | |
2494 | xfs_trans_log_buf(tp, bp, first, last); | |
2495 | } | |
2496 | ||
1da177e4 | 2497 | /* |
aafc3c24 BF |
2498 | * Mask off the bits in the first region and calculate the first and |
2499 | * last field offsets for any bits in the second region. | |
1da177e4 | 2500 | */ |
aafc3c24 BF |
2501 | fields &= ~XFS_AGI_ALL_BITS_R1; |
2502 | if (fields) { | |
2503 | xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R2, | |
2504 | &first, &last); | |
2505 | xfs_trans_log_buf(tp, bp, first, last); | |
2506 | } | |
1da177e4 LT |
2507 | } |
2508 | ||
a6a781a5 | 2509 | static xfs_failaddr_t |
612cfbfe | 2510 | xfs_agi_verify( |
3702ce6e DC |
2511 | struct xfs_buf *bp) |
2512 | { | |
2513 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
2514 | struct xfs_agi *agi = XFS_BUF_TO_AGI(bp); | |
9f96cc95 | 2515 | int i; |
3702ce6e | 2516 | |
a45086e2 BF |
2517 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
2518 | if (!uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid)) | |
a6a781a5 | 2519 | return __this_address; |
a45086e2 BF |
2520 | if (!xfs_log_check_lsn(mp, |
2521 | be64_to_cpu(XFS_BUF_TO_AGI(bp)->agi_lsn))) | |
a6a781a5 | 2522 | return __this_address; |
a45086e2 BF |
2523 | } |
2524 | ||
3702ce6e DC |
2525 | /* |
2526 | * Validate the magic number of the agi block. | |
2527 | */ | |
983d09ff | 2528 | if (agi->agi_magicnum != cpu_to_be32(XFS_AGI_MAGIC)) |
a6a781a5 | 2529 | return __this_address; |
983d09ff | 2530 | if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum))) |
a6a781a5 | 2531 | return __this_address; |
3702ce6e | 2532 | |
d2a047f3 DW |
2533 | if (be32_to_cpu(agi->agi_level) < 1 || |
2534 | be32_to_cpu(agi->agi_level) > XFS_BTREE_MAXLEVELS) | |
a6a781a5 | 2535 | return __this_address; |
d2a047f3 DW |
2536 | |
2537 | if (xfs_sb_version_hasfinobt(&mp->m_sb) && | |
2538 | (be32_to_cpu(agi->agi_free_level) < 1 || | |
2539 | be32_to_cpu(agi->agi_free_level) > XFS_BTREE_MAXLEVELS)) | |
a6a781a5 | 2540 | return __this_address; |
d2a047f3 | 2541 | |
3702ce6e DC |
2542 | /* |
2543 | * during growfs operations, the perag is not fully initialised, | |
2544 | * so we can't use it for any useful checking. growfs ensures we can't | |
2545 | * use it by using uncached buffers that don't have the perag attached | |
2546 | * so we can detect and avoid this problem. | |
2547 | */ | |
983d09ff | 2548 | if (bp->b_pag && be32_to_cpu(agi->agi_seqno) != bp->b_pag->pag_agno) |
a6a781a5 | 2549 | return __this_address; |
3702ce6e | 2550 | |
9f96cc95 DC |
2551 | for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++) { |
2552 | if (agi->agi_unlinked[i] == NULLAGINO) | |
2553 | continue; | |
2554 | if (!xfs_verify_ino(mp, be32_to_cpu(agi->agi_unlinked[i]))) | |
2555 | return __this_address; | |
2556 | } | |
2557 | ||
a6a781a5 | 2558 | return NULL; |
612cfbfe DC |
2559 | } |
2560 | ||
1813dd64 DC |
2561 | static void |
2562 | xfs_agi_read_verify( | |
612cfbfe DC |
2563 | struct xfs_buf *bp) |
2564 | { | |
983d09ff | 2565 | struct xfs_mount *mp = bp->b_target->bt_mount; |
bc1a09b8 | 2566 | xfs_failaddr_t fa; |
983d09ff | 2567 | |
ce5028cf ES |
2568 | if (xfs_sb_version_hascrc(&mp->m_sb) && |
2569 | !xfs_buf_verify_cksum(bp, XFS_AGI_CRC_OFF)) | |
bc1a09b8 DW |
2570 | xfs_verifier_error(bp, -EFSBADCRC, __this_address); |
2571 | else { | |
2572 | fa = xfs_agi_verify(bp); | |
2573 | if (XFS_TEST_ERROR(fa, mp, XFS_ERRTAG_IALLOC_READ_AGI)) | |
2574 | xfs_verifier_error(bp, -EFSCORRUPTED, fa); | |
2575 | } | |
612cfbfe DC |
2576 | } |
2577 | ||
b0f539de | 2578 | static void |
1813dd64 | 2579 | xfs_agi_write_verify( |
612cfbfe DC |
2580 | struct xfs_buf *bp) |
2581 | { | |
fb1755a6 CM |
2582 | struct xfs_mount *mp = bp->b_target->bt_mount; |
2583 | struct xfs_buf_log_item *bip = bp->b_log_item; | |
bc1a09b8 | 2584 | xfs_failaddr_t fa; |
983d09ff | 2585 | |
bc1a09b8 DW |
2586 | fa = xfs_agi_verify(bp); |
2587 | if (fa) { | |
2588 | xfs_verifier_error(bp, -EFSCORRUPTED, fa); | |
983d09ff DC |
2589 | return; |
2590 | } | |
2591 | ||
2592 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
2593 | return; | |
2594 | ||
2595 | if (bip) | |
2596 | XFS_BUF_TO_AGI(bp)->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn); | |
f1dbcd7e | 2597 | xfs_buf_update_cksum(bp, XFS_AGI_CRC_OFF); |
3702ce6e DC |
2598 | } |
2599 | ||
1813dd64 | 2600 | const struct xfs_buf_ops xfs_agi_buf_ops = { |
233135b7 | 2601 | .name = "xfs_agi", |
1813dd64 DC |
2602 | .verify_read = xfs_agi_read_verify, |
2603 | .verify_write = xfs_agi_write_verify, | |
b5572597 | 2604 | .verify_struct = xfs_agi_verify, |
1813dd64 DC |
2605 | }; |
2606 | ||
1da177e4 LT |
2607 | /* |
2608 | * Read in the allocation group header (inode allocation section) | |
2609 | */ | |
2610 | int | |
5e1be0fb CH |
2611 | xfs_read_agi( |
2612 | struct xfs_mount *mp, /* file system mount structure */ | |
2613 | struct xfs_trans *tp, /* transaction pointer */ | |
2614 | xfs_agnumber_t agno, /* allocation group number */ | |
2615 | struct xfs_buf **bpp) /* allocation group hdr buf */ | |
1da177e4 | 2616 | { |
5e1be0fb | 2617 | int error; |
1da177e4 | 2618 | |
d123031a | 2619 | trace_xfs_read_agi(mp, agno); |
5e1be0fb | 2620 | |
d123031a | 2621 | ASSERT(agno != NULLAGNUMBER); |
5e1be0fb | 2622 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, |
1da177e4 | 2623 | XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)), |
1813dd64 | 2624 | XFS_FSS_TO_BB(mp, 1), 0, bpp, &xfs_agi_buf_ops); |
1da177e4 LT |
2625 | if (error) |
2626 | return error; | |
200237d6 ES |
2627 | if (tp) |
2628 | xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_AGI_BUF); | |
5e1be0fb | 2629 | |
38f23232 | 2630 | xfs_buf_set_ref(*bpp, XFS_AGI_REF); |
5e1be0fb CH |
2631 | return 0; |
2632 | } | |
2633 | ||
2634 | int | |
2635 | xfs_ialloc_read_agi( | |
2636 | struct xfs_mount *mp, /* file system mount structure */ | |
2637 | struct xfs_trans *tp, /* transaction pointer */ | |
2638 | xfs_agnumber_t agno, /* allocation group number */ | |
2639 | struct xfs_buf **bpp) /* allocation group hdr buf */ | |
2640 | { | |
2641 | struct xfs_agi *agi; /* allocation group header */ | |
2642 | struct xfs_perag *pag; /* per allocation group data */ | |
2643 | int error; | |
2644 | ||
d123031a DC |
2645 | trace_xfs_ialloc_read_agi(mp, agno); |
2646 | ||
5e1be0fb CH |
2647 | error = xfs_read_agi(mp, tp, agno, bpp); |
2648 | if (error) | |
2649 | return error; | |
2650 | ||
2651 | agi = XFS_BUF_TO_AGI(*bpp); | |
44b56e0a | 2652 | pag = xfs_perag_get(mp, agno); |
1da177e4 | 2653 | if (!pag->pagi_init) { |
16259e7d | 2654 | pag->pagi_freecount = be32_to_cpu(agi->agi_freecount); |
92821e2b | 2655 | pag->pagi_count = be32_to_cpu(agi->agi_count); |
1da177e4 | 2656 | pag->pagi_init = 1; |
1da177e4 | 2657 | } |
1da177e4 | 2658 | |
5e1be0fb CH |
2659 | /* |
2660 | * It's possible for these to be out of sync if | |
2661 | * we are in the middle of a forced shutdown. | |
2662 | */ | |
2663 | ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) || | |
2664 | XFS_FORCED_SHUTDOWN(mp)); | |
44b56e0a | 2665 | xfs_perag_put(pag); |
1da177e4 LT |
2666 | return 0; |
2667 | } | |
92821e2b DC |
2668 | |
2669 | /* | |
2670 | * Read in the agi to initialise the per-ag data in the mount structure | |
2671 | */ | |
2672 | int | |
2673 | xfs_ialloc_pagi_init( | |
2674 | xfs_mount_t *mp, /* file system mount structure */ | |
2675 | xfs_trans_t *tp, /* transaction pointer */ | |
2676 | xfs_agnumber_t agno) /* allocation group number */ | |
2677 | { | |
2678 | xfs_buf_t *bp = NULL; | |
2679 | int error; | |
2680 | ||
2681 | error = xfs_ialloc_read_agi(mp, tp, agno, &bp); | |
2682 | if (error) | |
2683 | return error; | |
2684 | if (bp) | |
2685 | xfs_trans_brelse(tp, bp); | |
2686 | return 0; | |
2687 | } | |
91fb9afc DW |
2688 | |
2689 | /* Calculate the first and last possible inode number in an AG. */ | |
2690 | void | |
2691 | xfs_ialloc_agino_range( | |
2692 | struct xfs_mount *mp, | |
2693 | xfs_agnumber_t agno, | |
2694 | xfs_agino_t *first, | |
2695 | xfs_agino_t *last) | |
2696 | { | |
2697 | xfs_agblock_t bno; | |
2698 | xfs_agblock_t eoag; | |
2699 | ||
2700 | eoag = xfs_ag_block_count(mp, agno); | |
2701 | ||
2702 | /* | |
2703 | * Calculate the first inode, which will be in the first | |
2704 | * cluster-aligned block after the AGFL. | |
2705 | */ | |
2706 | bno = round_up(XFS_AGFL_BLOCK(mp) + 1, | |
2707 | xfs_ialloc_cluster_alignment(mp)); | |
2708 | *first = XFS_OFFBNO_TO_AGINO(mp, bno, 0); | |
2709 | ||
2710 | /* | |
2711 | * Calculate the last inode, which will be at the end of the | |
2712 | * last (aligned) cluster that can be allocated in the AG. | |
2713 | */ | |
2714 | bno = round_down(eoag, xfs_ialloc_cluster_alignment(mp)); | |
2715 | *last = XFS_OFFBNO_TO_AGINO(mp, bno, 0) - 1; | |
2716 | } | |
2717 | ||
2718 | /* | |
2719 | * Verify that an AG inode number pointer neither points outside the AG | |
2720 | * nor points at static metadata. | |
2721 | */ | |
2722 | bool | |
2723 | xfs_verify_agino( | |
2724 | struct xfs_mount *mp, | |
2725 | xfs_agnumber_t agno, | |
2726 | xfs_agino_t agino) | |
2727 | { | |
2728 | xfs_agino_t first; | |
2729 | xfs_agino_t last; | |
2730 | ||
2731 | xfs_ialloc_agino_range(mp, agno, &first, &last); | |
2732 | return agino >= first && agino <= last; | |
2733 | } | |
2734 | ||
2735 | /* | |
2736 | * Verify that an FS inode number pointer neither points outside the | |
2737 | * filesystem nor points at static AG metadata. | |
2738 | */ | |
2739 | bool | |
2740 | xfs_verify_ino( | |
2741 | struct xfs_mount *mp, | |
2742 | xfs_ino_t ino) | |
2743 | { | |
2744 | xfs_agnumber_t agno = XFS_INO_TO_AGNO(mp, ino); | |
2745 | xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino); | |
2746 | ||
2747 | if (agno >= mp->m_sb.sb_agcount) | |
2748 | return false; | |
2749 | if (XFS_AGINO_TO_INO(mp, agno, agino) != ino) | |
2750 | return false; | |
2751 | return xfs_verify_agino(mp, agno, agino); | |
2752 | } | |
2753 | ||
2754 | /* Is this an internal inode number? */ | |
2755 | bool | |
2756 | xfs_internal_inum( | |
2757 | struct xfs_mount *mp, | |
2758 | xfs_ino_t ino) | |
2759 | { | |
2760 | return ino == mp->m_sb.sb_rbmino || ino == mp->m_sb.sb_rsumino || | |
2761 | (xfs_sb_version_hasquota(&mp->m_sb) && | |
2762 | xfs_is_quota_inode(&mp->m_sb, ino)); | |
2763 | } | |
2764 | ||
2765 | /* | |
2766 | * Verify that a directory entry's inode number doesn't point at an internal | |
2767 | * inode, empty space, or static AG metadata. | |
2768 | */ | |
2769 | bool | |
2770 | xfs_verify_dir_ino( | |
2771 | struct xfs_mount *mp, | |
2772 | xfs_ino_t ino) | |
2773 | { | |
2774 | if (xfs_internal_inum(mp, ino)) | |
2775 | return false; | |
2776 | return xfs_verify_ino(mp, ino); | |
2777 | } | |
2e001266 DW |
2778 | |
2779 | /* Is there an inode record covering a given range of inode numbers? */ | |
2780 | int | |
2781 | xfs_ialloc_has_inode_record( | |
2782 | struct xfs_btree_cur *cur, | |
2783 | xfs_agino_t low, | |
2784 | xfs_agino_t high, | |
2785 | bool *exists) | |
2786 | { | |
2787 | struct xfs_inobt_rec_incore irec; | |
2788 | xfs_agino_t agino; | |
2789 | uint16_t holemask; | |
2790 | int has_record; | |
2791 | int i; | |
2792 | int error; | |
2793 | ||
2794 | *exists = false; | |
2795 | error = xfs_inobt_lookup(cur, low, XFS_LOOKUP_LE, &has_record); | |
2796 | while (error == 0 && has_record) { | |
2797 | error = xfs_inobt_get_rec(cur, &irec, &has_record); | |
2798 | if (error || irec.ir_startino > high) | |
2799 | break; | |
2800 | ||
2801 | agino = irec.ir_startino; | |
2802 | holemask = irec.ir_holemask; | |
2803 | for (i = 0; i < XFS_INOBT_HOLEMASK_BITS; holemask >>= 1, | |
2804 | i++, agino += XFS_INODES_PER_HOLEMASK_BIT) { | |
2805 | if (holemask & 1) | |
2806 | continue; | |
2807 | if (agino + XFS_INODES_PER_HOLEMASK_BIT > low && | |
2808 | agino <= high) { | |
2809 | *exists = true; | |
2810 | return 0; | |
2811 | } | |
2812 | } | |
2813 | ||
2814 | error = xfs_btree_increment(cur, 0, &has_record); | |
2815 | } | |
2816 | return error; | |
2817 | } | |
2818 | ||
2819 | /* Is there an inode record covering a given extent? */ | |
2820 | int | |
2821 | xfs_ialloc_has_inodes_at_extent( | |
2822 | struct xfs_btree_cur *cur, | |
2823 | xfs_agblock_t bno, | |
2824 | xfs_extlen_t len, | |
2825 | bool *exists) | |
2826 | { | |
2827 | xfs_agino_t low; | |
2828 | xfs_agino_t high; | |
2829 | ||
2830 | low = XFS_OFFBNO_TO_AGINO(cur->bc_mp, bno, 0); | |
2831 | high = XFS_OFFBNO_TO_AGINO(cur->bc_mp, bno + len, 0) - 1; | |
2832 | ||
2833 | return xfs_ialloc_has_inode_record(cur, low, high, exists); | |
2834 | } | |
2835 | ||
2836 | struct xfs_ialloc_count_inodes { | |
2837 | xfs_agino_t count; | |
2838 | xfs_agino_t freecount; | |
2839 | }; | |
2840 | ||
2841 | /* Record inode counts across all inobt records. */ | |
2842 | STATIC int | |
2843 | xfs_ialloc_count_inodes_rec( | |
2844 | struct xfs_btree_cur *cur, | |
2845 | union xfs_btree_rec *rec, | |
2846 | void *priv) | |
2847 | { | |
2848 | struct xfs_inobt_rec_incore irec; | |
2849 | struct xfs_ialloc_count_inodes *ci = priv; | |
2850 | ||
2851 | xfs_inobt_btrec_to_irec(cur->bc_mp, rec, &irec); | |
2852 | ci->count += irec.ir_count; | |
2853 | ci->freecount += irec.ir_freecount; | |
2854 | ||
2855 | return 0; | |
2856 | } | |
2857 | ||
2858 | /* Count allocated and free inodes under an inobt. */ | |
2859 | int | |
2860 | xfs_ialloc_count_inodes( | |
2861 | struct xfs_btree_cur *cur, | |
2862 | xfs_agino_t *count, | |
2863 | xfs_agino_t *freecount) | |
2864 | { | |
2865 | struct xfs_ialloc_count_inodes ci = {0}; | |
2866 | int error; | |
2867 | ||
2868 | ASSERT(cur->bc_btnum == XFS_BTNUM_INO); | |
2869 | error = xfs_btree_query_all(cur, xfs_ialloc_count_inodes_rec, &ci); | |
2870 | if (error) | |
2871 | return error; | |
2872 | ||
2873 | *count = ci.count; | |
2874 | *freecount = ci.freecount; | |
2875 | return 0; | |
2876 | } |