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" |
a844f451 | 25 | #include "xfs_inum.h" |
1da177e4 LT |
26 | #include "xfs_sb.h" |
27 | #include "xfs_ag.h" | |
1da177e4 | 28 | #include "xfs_mount.h" |
1da177e4 | 29 | #include "xfs_inode.h" |
a844f451 NS |
30 | #include "xfs_btree.h" |
31 | #include "xfs_ialloc.h" | |
a4fbe6ab | 32 | #include "xfs_ialloc_btree.h" |
1da177e4 | 33 | #include "xfs_alloc.h" |
1da177e4 LT |
34 | #include "xfs_rtalloc.h" |
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" |
a4fbe6ab | 42 | #include "xfs_dinode.h" |
d123031a | 43 | #include "xfs_trace.h" |
1da177e4 | 44 | |
1da177e4 LT |
45 | |
46 | /* | |
47 | * Allocation group level functions. | |
48 | */ | |
75de2a91 DC |
49 | static inline int |
50 | xfs_ialloc_cluster_alignment( | |
51 | xfs_alloc_arg_t *args) | |
52 | { | |
53 | if (xfs_sb_version_hasalign(&args->mp->m_sb) && | |
54 | args->mp->m_sb.sb_inoalignmt >= | |
55 | XFS_B_TO_FSBT(args->mp, XFS_INODE_CLUSTER_SIZE(args->mp))) | |
56 | return args->mp->m_sb.sb_inoalignmt; | |
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; | |
21875505 CH |
71 | cur->bc_rec.i.ir_freecount = 0; |
72 | cur->bc_rec.i.ir_free = 0; | |
73 | return xfs_btree_lookup(cur, dir, stat); | |
fe033cc8 CH |
74 | } |
75 | ||
278d0ca1 | 76 | /* |
afabc24a | 77 | * Update the record referred to by cur to the value given. |
278d0ca1 CH |
78 | * This either works (return 0) or gets an EFSCORRUPTED error. |
79 | */ | |
80 | STATIC int /* error */ | |
81 | xfs_inobt_update( | |
82 | struct xfs_btree_cur *cur, /* btree cursor */ | |
afabc24a | 83 | xfs_inobt_rec_incore_t *irec) /* btree record */ |
278d0ca1 CH |
84 | { |
85 | union xfs_btree_rec rec; | |
86 | ||
afabc24a CH |
87 | rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino); |
88 | rec.inobt.ir_freecount = cpu_to_be32(irec->ir_freecount); | |
89 | rec.inobt.ir_free = cpu_to_be64(irec->ir_free); | |
278d0ca1 CH |
90 | return xfs_btree_update(cur, &rec); |
91 | } | |
92 | ||
8cc938fe CH |
93 | /* |
94 | * Get the data from the pointed-to record. | |
95 | */ | |
96 | int /* error */ | |
97 | xfs_inobt_get_rec( | |
98 | struct xfs_btree_cur *cur, /* btree cursor */ | |
2e287a73 | 99 | xfs_inobt_rec_incore_t *irec, /* btree record */ |
8cc938fe CH |
100 | int *stat) /* output: success/failure */ |
101 | { | |
102 | union xfs_btree_rec *rec; | |
103 | int error; | |
104 | ||
105 | error = xfs_btree_get_rec(cur, &rec, stat); | |
106 | if (!error && *stat == 1) { | |
2e287a73 CH |
107 | irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino); |
108 | irec->ir_freecount = be32_to_cpu(rec->inobt.ir_freecount); | |
109 | irec->ir_free = be64_to_cpu(rec->inobt.ir_free); | |
8cc938fe CH |
110 | } |
111 | return error; | |
112 | } | |
113 | ||
0b48db80 DC |
114 | /* |
115 | * Verify that the number of free inodes in the AGI is correct. | |
116 | */ | |
117 | #ifdef DEBUG | |
118 | STATIC int | |
119 | xfs_check_agi_freecount( | |
120 | struct xfs_btree_cur *cur, | |
121 | struct xfs_agi *agi) | |
122 | { | |
123 | if (cur->bc_nlevels == 1) { | |
124 | xfs_inobt_rec_incore_t rec; | |
125 | int freecount = 0; | |
126 | int error; | |
127 | int i; | |
128 | ||
21875505 | 129 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); |
0b48db80 DC |
130 | if (error) |
131 | return error; | |
132 | ||
133 | do { | |
134 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
135 | if (error) | |
136 | return error; | |
137 | ||
138 | if (i) { | |
139 | freecount += rec.ir_freecount; | |
140 | error = xfs_btree_increment(cur, 0, &i); | |
141 | if (error) | |
142 | return error; | |
143 | } | |
144 | } while (i == 1); | |
145 | ||
146 | if (!XFS_FORCED_SHUTDOWN(cur->bc_mp)) | |
147 | ASSERT(freecount == be32_to_cpu(agi->agi_freecount)); | |
148 | } | |
149 | return 0; | |
150 | } | |
151 | #else | |
152 | #define xfs_check_agi_freecount(cur, agi) 0 | |
153 | #endif | |
154 | ||
85c0b2ab | 155 | /* |
28c8e41a DC |
156 | * Initialise a new set of inodes. When called without a transaction context |
157 | * (e.g. from recovery) we initiate a delayed write of the inode buffers rather | |
158 | * than logging them (which in a transaction context puts them into the AIL | |
159 | * for writeback rather than the xfsbufd queue). | |
85c0b2ab | 160 | */ |
ddf6ad01 | 161 | int |
85c0b2ab DC |
162 | xfs_ialloc_inode_init( |
163 | struct xfs_mount *mp, | |
164 | struct xfs_trans *tp, | |
28c8e41a | 165 | struct list_head *buffer_list, |
85c0b2ab DC |
166 | xfs_agnumber_t agno, |
167 | xfs_agblock_t agbno, | |
168 | xfs_agblock_t length, | |
169 | unsigned int gen) | |
170 | { | |
171 | struct xfs_buf *fbuf; | |
172 | struct xfs_dinode *free; | |
173 | int blks_per_cluster, nbufs, ninodes; | |
174 | int version; | |
175 | int i, j; | |
176 | xfs_daddr_t d; | |
93848a99 | 177 | xfs_ino_t ino = 0; |
85c0b2ab DC |
178 | |
179 | /* | |
180 | * Loop over the new block(s), filling in the inodes. | |
181 | * For small block sizes, manipulate the inodes in buffers | |
182 | * which are multiples of the blocks size. | |
183 | */ | |
184 | if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) { | |
185 | blks_per_cluster = 1; | |
186 | nbufs = length; | |
187 | ninodes = mp->m_sb.sb_inopblock; | |
188 | } else { | |
189 | blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) / | |
190 | mp->m_sb.sb_blocksize; | |
191 | nbufs = length / blks_per_cluster; | |
192 | ninodes = blks_per_cluster * mp->m_sb.sb_inopblock; | |
193 | } | |
194 | ||
195 | /* | |
93848a99 CH |
196 | * Figure out what version number to use in the inodes we create. If |
197 | * the superblock version has caught up to the one that supports the new | |
198 | * inode format, then use the new inode version. Otherwise use the old | |
199 | * version so that old kernels will continue to be able to use the file | |
200 | * system. | |
201 | * | |
202 | * For v3 inodes, we also need to write the inode number into the inode, | |
203 | * so calculate the first inode number of the chunk here as | |
204 | * XFS_OFFBNO_TO_AGINO() only works within a filesystem block, not | |
205 | * across multiple filesystem blocks (such as a cluster) and so cannot | |
206 | * be used in the cluster buffer loop below. | |
207 | * | |
208 | * Further, because we are writing the inode directly into the buffer | |
209 | * and calculating a CRC on the entire inode, we have ot log the entire | |
210 | * inode so that the entire range the CRC covers is present in the log. | |
211 | * That means for v3 inode we log the entire buffer rather than just the | |
212 | * inode cores. | |
85c0b2ab | 213 | */ |
93848a99 CH |
214 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
215 | version = 3; | |
216 | ino = XFS_AGINO_TO_INO(mp, agno, | |
217 | XFS_OFFBNO_TO_AGINO(mp, agbno, 0)); | |
ddf6ad01 DC |
218 | |
219 | /* | |
220 | * log the initialisation that is about to take place as an | |
221 | * logical operation. This means the transaction does not | |
222 | * need to log the physical changes to the inode buffers as log | |
223 | * recovery will know what initialisation is actually needed. | |
224 | * Hence we only need to log the buffers as "ordered" buffers so | |
225 | * they track in the AIL as if they were physically logged. | |
226 | */ | |
227 | if (tp) | |
228 | xfs_icreate_log(tp, agno, agbno, XFS_IALLOC_INODES(mp), | |
229 | mp->m_sb.sb_inodesize, length, gen); | |
93848a99 | 230 | } else if (xfs_sb_version_hasnlink(&mp->m_sb)) |
85c0b2ab DC |
231 | version = 2; |
232 | else | |
233 | version = 1; | |
234 | ||
235 | for (j = 0; j < nbufs; j++) { | |
236 | /* | |
237 | * Get the block. | |
238 | */ | |
239 | d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster)); | |
240 | fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, | |
7c4cebe8 DC |
241 | mp->m_bsize * blks_per_cluster, |
242 | XBF_UNMAPPED); | |
2a30f36d CS |
243 | if (!fbuf) |
244 | return ENOMEM; | |
ddf6ad01 DC |
245 | |
246 | /* Initialize the inode buffers and log them appropriately. */ | |
1813dd64 | 247 | fbuf->b_ops = &xfs_inode_buf_ops; |
93848a99 | 248 | xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length)); |
85c0b2ab DC |
249 | for (i = 0; i < ninodes; i++) { |
250 | int ioffset = i << mp->m_sb.sb_inodelog; | |
93848a99 | 251 | uint isize = xfs_dinode_size(version); |
85c0b2ab DC |
252 | |
253 | free = xfs_make_iptr(mp, fbuf, i); | |
254 | free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC); | |
255 | free->di_version = version; | |
256 | free->di_gen = cpu_to_be32(gen); | |
257 | free->di_next_unlinked = cpu_to_be32(NULLAGINO); | |
93848a99 CH |
258 | |
259 | if (version == 3) { | |
260 | free->di_ino = cpu_to_be64(ino); | |
261 | ino++; | |
262 | uuid_copy(&free->di_uuid, &mp->m_sb.sb_uuid); | |
263 | xfs_dinode_calc_crc(mp, free); | |
28c8e41a | 264 | } else if (tp) { |
93848a99 CH |
265 | /* just log the inode core */ |
266 | xfs_trans_log_buf(tp, fbuf, ioffset, | |
267 | ioffset + isize - 1); | |
268 | } | |
269 | } | |
28c8e41a DC |
270 | |
271 | if (tp) { | |
272 | /* | |
273 | * Mark the buffer as an inode allocation buffer so it | |
274 | * sticks in AIL at the point of this allocation | |
275 | * transaction. This ensures the they are on disk before | |
276 | * the tail of the log can be moved past this | |
277 | * transaction (i.e. by preventing relogging from moving | |
278 | * it forward in the log). | |
279 | */ | |
280 | xfs_trans_inode_alloc_buf(tp, fbuf); | |
281 | if (version == 3) { | |
ddf6ad01 DC |
282 | /* |
283 | * Mark the buffer as ordered so that they are | |
284 | * not physically logged in the transaction but | |
285 | * still tracked in the AIL as part of the | |
286 | * transaction and pin the log appropriately. | |
287 | */ | |
288 | xfs_trans_ordered_buf(tp, fbuf); | |
28c8e41a DC |
289 | xfs_trans_log_buf(tp, fbuf, 0, |
290 | BBTOB(fbuf->b_length) - 1); | |
291 | } | |
292 | } else { | |
293 | fbuf->b_flags |= XBF_DONE; | |
294 | xfs_buf_delwri_queue(fbuf, buffer_list); | |
295 | xfs_buf_relse(fbuf); | |
85c0b2ab | 296 | } |
85c0b2ab | 297 | } |
2a30f36d | 298 | return 0; |
85c0b2ab DC |
299 | } |
300 | ||
1da177e4 LT |
301 | /* |
302 | * Allocate new inodes in the allocation group specified by agbp. | |
303 | * Return 0 for success, else error code. | |
304 | */ | |
305 | STATIC int /* error code or 0 */ | |
306 | xfs_ialloc_ag_alloc( | |
307 | xfs_trans_t *tp, /* transaction pointer */ | |
308 | xfs_buf_t *agbp, /* alloc group buffer */ | |
309 | int *alloc) | |
310 | { | |
311 | xfs_agi_t *agi; /* allocation group header */ | |
312 | xfs_alloc_arg_t args; /* allocation argument structure */ | |
1da177e4 | 313 | xfs_btree_cur_t *cur; /* inode btree cursor */ |
92821e2b | 314 | xfs_agnumber_t agno; |
1da177e4 | 315 | int error; |
85c0b2ab | 316 | int i; |
1da177e4 LT |
317 | xfs_agino_t newino; /* new first inode's number */ |
318 | xfs_agino_t newlen; /* new number of inodes */ | |
1da177e4 | 319 | xfs_agino_t thisino; /* current inode number, for loop */ |
3ccb8b5f | 320 | int isaligned = 0; /* inode allocation at stripe unit */ |
1da177e4 | 321 | /* boundary */ |
44b56e0a | 322 | struct xfs_perag *pag; |
1da177e4 | 323 | |
a0041684 | 324 | memset(&args, 0, sizeof(args)); |
1da177e4 LT |
325 | args.tp = tp; |
326 | args.mp = tp->t_mountp; | |
327 | ||
328 | /* | |
329 | * Locking will ensure that we don't have two callers in here | |
330 | * at one time. | |
331 | */ | |
332 | newlen = XFS_IALLOC_INODES(args.mp); | |
333 | if (args.mp->m_maxicount && | |
334 | args.mp->m_sb.sb_icount + newlen > args.mp->m_maxicount) | |
335 | return XFS_ERROR(ENOSPC); | |
336 | args.minlen = args.maxlen = XFS_IALLOC_BLOCKS(args.mp); | |
337 | /* | |
3ccb8b5f GO |
338 | * First try to allocate inodes contiguous with the last-allocated |
339 | * chunk of inodes. If the filesystem is striped, this will fill | |
340 | * an entire stripe unit with inodes. | |
28c8e41a | 341 | */ |
1da177e4 | 342 | agi = XFS_BUF_TO_AGI(agbp); |
3ccb8b5f | 343 | newino = be32_to_cpu(agi->agi_newino); |
85c0b2ab | 344 | agno = be32_to_cpu(agi->agi_seqno); |
019ff2d5 NS |
345 | args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) + |
346 | XFS_IALLOC_BLOCKS(args.mp); | |
347 | if (likely(newino != NULLAGINO && | |
348 | (args.agbno < be32_to_cpu(agi->agi_length)))) { | |
85c0b2ab | 349 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
3ccb8b5f | 350 | args.type = XFS_ALLOCTYPE_THIS_BNO; |
3ccb8b5f | 351 | args.prod = 1; |
75de2a91 | 352 | |
3ccb8b5f | 353 | /* |
75de2a91 DC |
354 | * We need to take into account alignment here to ensure that |
355 | * we don't modify the free list if we fail to have an exact | |
356 | * block. If we don't have an exact match, and every oher | |
357 | * attempt allocation attempt fails, we'll end up cancelling | |
358 | * a dirty transaction and shutting down. | |
359 | * | |
360 | * For an exact allocation, alignment must be 1, | |
361 | * however we need to take cluster alignment into account when | |
362 | * fixing up the freelist. Use the minalignslop field to | |
363 | * indicate that extra blocks might be required for alignment, | |
364 | * but not to use them in the actual exact allocation. | |
3ccb8b5f | 365 | */ |
75de2a91 DC |
366 | args.alignment = 1; |
367 | args.minalignslop = xfs_ialloc_cluster_alignment(&args) - 1; | |
368 | ||
369 | /* Allow space for the inode btree to split. */ | |
0d87e656 | 370 | args.minleft = args.mp->m_in_maxlevels - 1; |
3ccb8b5f GO |
371 | if ((error = xfs_alloc_vextent(&args))) |
372 | return error; | |
373 | } else | |
374 | args.fsbno = NULLFSBLOCK; | |
1da177e4 | 375 | |
3ccb8b5f GO |
376 | if (unlikely(args.fsbno == NULLFSBLOCK)) { |
377 | /* | |
378 | * Set the alignment for the allocation. | |
379 | * If stripe alignment is turned on then align at stripe unit | |
380 | * boundary. | |
019ff2d5 NS |
381 | * If the cluster size is smaller than a filesystem block |
382 | * then we're doing I/O for inodes in filesystem block size | |
3ccb8b5f GO |
383 | * pieces, so don't need alignment anyway. |
384 | */ | |
385 | isaligned = 0; | |
386 | if (args.mp->m_sinoalign) { | |
387 | ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN)); | |
388 | args.alignment = args.mp->m_dalign; | |
389 | isaligned = 1; | |
75de2a91 DC |
390 | } else |
391 | args.alignment = xfs_ialloc_cluster_alignment(&args); | |
3ccb8b5f GO |
392 | /* |
393 | * Need to figure out where to allocate the inode blocks. | |
394 | * Ideally they should be spaced out through the a.g. | |
395 | * For now, just allocate blocks up front. | |
396 | */ | |
397 | args.agbno = be32_to_cpu(agi->agi_root); | |
85c0b2ab | 398 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
3ccb8b5f GO |
399 | /* |
400 | * Allocate a fixed-size extent of inodes. | |
401 | */ | |
402 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
3ccb8b5f GO |
403 | args.prod = 1; |
404 | /* | |
405 | * Allow space for the inode btree to split. | |
406 | */ | |
0d87e656 | 407 | args.minleft = args.mp->m_in_maxlevels - 1; |
3ccb8b5f GO |
408 | if ((error = xfs_alloc_vextent(&args))) |
409 | return error; | |
410 | } | |
019ff2d5 | 411 | |
1da177e4 LT |
412 | /* |
413 | * If stripe alignment is turned on, then try again with cluster | |
414 | * alignment. | |
415 | */ | |
416 | if (isaligned && args.fsbno == NULLFSBLOCK) { | |
417 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
16259e7d | 418 | args.agbno = be32_to_cpu(agi->agi_root); |
85c0b2ab | 419 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
75de2a91 | 420 | args.alignment = xfs_ialloc_cluster_alignment(&args); |
1da177e4 LT |
421 | if ((error = xfs_alloc_vextent(&args))) |
422 | return error; | |
423 | } | |
424 | ||
425 | if (args.fsbno == NULLFSBLOCK) { | |
426 | *alloc = 0; | |
427 | return 0; | |
428 | } | |
429 | ASSERT(args.len == args.minlen); | |
1da177e4 | 430 | |
359346a9 | 431 | /* |
85c0b2ab DC |
432 | * Stamp and write the inode buffers. |
433 | * | |
359346a9 DC |
434 | * Seed the new inode cluster with a random generation number. This |
435 | * prevents short-term reuse of generation numbers if a chunk is | |
436 | * freed and then immediately reallocated. We use random numbers | |
437 | * rather than a linear progression to prevent the next generation | |
438 | * number from being easily guessable. | |
439 | */ | |
28c8e41a | 440 | error = xfs_ialloc_inode_init(args.mp, tp, NULL, agno, args.agbno, |
ecb3403d | 441 | args.len, prandom_u32()); |
d42f08f6 | 442 | |
2a30f36d CS |
443 | if (error) |
444 | return error; | |
85c0b2ab DC |
445 | /* |
446 | * Convert the results. | |
447 | */ | |
448 | newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0); | |
413d57c9 MS |
449 | be32_add_cpu(&agi->agi_count, newlen); |
450 | be32_add_cpu(&agi->agi_freecount, newlen); | |
44b56e0a DC |
451 | pag = xfs_perag_get(args.mp, agno); |
452 | pag->pagi_freecount += newlen; | |
453 | xfs_perag_put(pag); | |
16259e7d | 454 | agi->agi_newino = cpu_to_be32(newino); |
85c0b2ab | 455 | |
1da177e4 LT |
456 | /* |
457 | * Insert records describing the new inode chunk into the btree. | |
458 | */ | |
561f7d17 | 459 | cur = xfs_inobt_init_cursor(args.mp, tp, agbp, agno); |
1da177e4 LT |
460 | for (thisino = newino; |
461 | thisino < newino + newlen; | |
462 | thisino += XFS_INODES_PER_CHUNK) { | |
21875505 CH |
463 | cur->bc_rec.i.ir_startino = thisino; |
464 | cur->bc_rec.i.ir_freecount = XFS_INODES_PER_CHUNK; | |
465 | cur->bc_rec.i.ir_free = XFS_INOBT_ALL_FREE; | |
466 | error = xfs_btree_lookup(cur, XFS_LOOKUP_EQ, &i); | |
467 | if (error) { | |
1da177e4 LT |
468 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); |
469 | return error; | |
470 | } | |
471 | ASSERT(i == 0); | |
21875505 CH |
472 | error = xfs_btree_insert(cur, &i); |
473 | if (error) { | |
1da177e4 LT |
474 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); |
475 | return error; | |
476 | } | |
477 | ASSERT(i == 1); | |
478 | } | |
479 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
480 | /* | |
481 | * Log allocation group header fields | |
482 | */ | |
483 | xfs_ialloc_log_agi(tp, agbp, | |
484 | XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO); | |
485 | /* | |
486 | * Modify/log superblock values for inode count and inode free count. | |
487 | */ | |
488 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen); | |
489 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen); | |
490 | *alloc = 1; | |
491 | return 0; | |
492 | } | |
493 | ||
b8f82a4a | 494 | STATIC xfs_agnumber_t |
1da177e4 LT |
495 | xfs_ialloc_next_ag( |
496 | xfs_mount_t *mp) | |
497 | { | |
498 | xfs_agnumber_t agno; | |
499 | ||
500 | spin_lock(&mp->m_agirotor_lock); | |
501 | agno = mp->m_agirotor; | |
8aea3ff4 | 502 | if (++mp->m_agirotor >= mp->m_maxagi) |
1da177e4 LT |
503 | mp->m_agirotor = 0; |
504 | spin_unlock(&mp->m_agirotor_lock); | |
505 | ||
506 | return agno; | |
507 | } | |
508 | ||
509 | /* | |
510 | * Select an allocation group to look for a free inode in, based on the parent | |
2f21ff1c | 511 | * inode and the mode. Return the allocation group buffer. |
1da177e4 | 512 | */ |
55d6af64 | 513 | STATIC xfs_agnumber_t |
1da177e4 LT |
514 | xfs_ialloc_ag_select( |
515 | xfs_trans_t *tp, /* transaction pointer */ | |
516 | xfs_ino_t parent, /* parent directory inode number */ | |
576b1d67 | 517 | umode_t mode, /* bits set to indicate file type */ |
1da177e4 LT |
518 | int okalloc) /* ok to allocate more space */ |
519 | { | |
1da177e4 LT |
520 | xfs_agnumber_t agcount; /* number of ag's in the filesystem */ |
521 | xfs_agnumber_t agno; /* current ag number */ | |
522 | int flags; /* alloc buffer locking flags */ | |
523 | xfs_extlen_t ineed; /* blocks needed for inode allocation */ | |
524 | xfs_extlen_t longest = 0; /* longest extent available */ | |
525 | xfs_mount_t *mp; /* mount point structure */ | |
526 | int needspace; /* file mode implies space allocated */ | |
527 | xfs_perag_t *pag; /* per allocation group data */ | |
528 | xfs_agnumber_t pagno; /* parent (starting) ag number */ | |
55d6af64 | 529 | int error; |
1da177e4 LT |
530 | |
531 | /* | |
532 | * Files of these types need at least one block if length > 0 | |
533 | * (and they won't fit in the inode, but that's hard to figure out). | |
534 | */ | |
535 | needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode); | |
536 | mp = tp->t_mountp; | |
537 | agcount = mp->m_maxagi; | |
538 | if (S_ISDIR(mode)) | |
539 | pagno = xfs_ialloc_next_ag(mp); | |
540 | else { | |
541 | pagno = XFS_INO_TO_AGNO(mp, parent); | |
542 | if (pagno >= agcount) | |
543 | pagno = 0; | |
544 | } | |
55d6af64 | 545 | |
1da177e4 | 546 | ASSERT(pagno < agcount); |
55d6af64 | 547 | |
1da177e4 LT |
548 | /* |
549 | * Loop through allocation groups, looking for one with a little | |
550 | * free space in it. Note we don't look for free inodes, exactly. | |
551 | * Instead, we include whether there is a need to allocate inodes | |
552 | * to mean that blocks must be allocated for them, | |
553 | * if none are currently free. | |
554 | */ | |
555 | agno = pagno; | |
556 | flags = XFS_ALLOC_FLAG_TRYLOCK; | |
1da177e4 | 557 | for (;;) { |
44b56e0a | 558 | pag = xfs_perag_get(mp, agno); |
55d6af64 CH |
559 | if (!pag->pagi_inodeok) { |
560 | xfs_ialloc_next_ag(mp); | |
561 | goto nextag; | |
562 | } | |
563 | ||
1da177e4 | 564 | if (!pag->pagi_init) { |
55d6af64 CH |
565 | error = xfs_ialloc_pagi_init(mp, tp, agno); |
566 | if (error) | |
1da177e4 | 567 | goto nextag; |
55d6af64 | 568 | } |
1da177e4 | 569 | |
55d6af64 CH |
570 | if (pag->pagi_freecount) { |
571 | xfs_perag_put(pag); | |
572 | return agno; | |
1da177e4 LT |
573 | } |
574 | ||
55d6af64 CH |
575 | if (!okalloc) |
576 | goto nextag; | |
577 | ||
578 | if (!pag->pagf_init) { | |
579 | error = xfs_alloc_pagf_init(mp, tp, agno, flags); | |
580 | if (error) | |
1da177e4 | 581 | goto nextag; |
1da177e4 | 582 | } |
55d6af64 CH |
583 | |
584 | /* | |
585 | * Is there enough free space for the file plus a block of | |
586 | * inodes? (if we need to allocate some)? | |
587 | */ | |
588 | ineed = XFS_IALLOC_BLOCKS(mp); | |
589 | longest = pag->pagf_longest; | |
590 | if (!longest) | |
591 | longest = pag->pagf_flcount > 0; | |
592 | ||
593 | if (pag->pagf_freeblks >= needspace + ineed && | |
594 | longest >= ineed) { | |
595 | xfs_perag_put(pag); | |
596 | return agno; | |
1da177e4 | 597 | } |
1da177e4 | 598 | nextag: |
44b56e0a | 599 | xfs_perag_put(pag); |
1da177e4 LT |
600 | /* |
601 | * No point in iterating over the rest, if we're shutting | |
602 | * down. | |
603 | */ | |
1c1c6ebc | 604 | if (XFS_FORCED_SHUTDOWN(mp)) |
55d6af64 | 605 | return NULLAGNUMBER; |
1da177e4 LT |
606 | agno++; |
607 | if (agno >= agcount) | |
608 | agno = 0; | |
609 | if (agno == pagno) { | |
1c1c6ebc | 610 | if (flags == 0) |
55d6af64 | 611 | return NULLAGNUMBER; |
1da177e4 LT |
612 | flags = 0; |
613 | } | |
614 | } | |
615 | } | |
616 | ||
4254b0bb CH |
617 | /* |
618 | * Try to retrieve the next record to the left/right from the current one. | |
619 | */ | |
620 | STATIC int | |
621 | xfs_ialloc_next_rec( | |
622 | struct xfs_btree_cur *cur, | |
623 | xfs_inobt_rec_incore_t *rec, | |
624 | int *done, | |
625 | int left) | |
626 | { | |
627 | int error; | |
628 | int i; | |
629 | ||
630 | if (left) | |
631 | error = xfs_btree_decrement(cur, 0, &i); | |
632 | else | |
633 | error = xfs_btree_increment(cur, 0, &i); | |
634 | ||
635 | if (error) | |
636 | return error; | |
637 | *done = !i; | |
638 | if (i) { | |
639 | error = xfs_inobt_get_rec(cur, rec, &i); | |
640 | if (error) | |
641 | return error; | |
642 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
643 | } | |
644 | ||
645 | return 0; | |
646 | } | |
647 | ||
bd169565 DC |
648 | STATIC int |
649 | xfs_ialloc_get_rec( | |
650 | struct xfs_btree_cur *cur, | |
651 | xfs_agino_t agino, | |
652 | xfs_inobt_rec_incore_t *rec, | |
43df2ee6 | 653 | int *done) |
bd169565 DC |
654 | { |
655 | int error; | |
656 | int i; | |
657 | ||
658 | error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i); | |
659 | if (error) | |
660 | return error; | |
661 | *done = !i; | |
662 | if (i) { | |
663 | error = xfs_inobt_get_rec(cur, rec, &i); | |
664 | if (error) | |
665 | return error; | |
666 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
667 | } | |
668 | ||
669 | return 0; | |
670 | } | |
0b48db80 | 671 | |
1da177e4 | 672 | /* |
f2ecc5e4 | 673 | * Allocate an inode. |
1da177e4 | 674 | * |
f2ecc5e4 CH |
675 | * The caller selected an AG for us, and made sure that free inodes are |
676 | * available. | |
1da177e4 | 677 | */ |
f2ecc5e4 CH |
678 | STATIC int |
679 | xfs_dialloc_ag( | |
680 | struct xfs_trans *tp, | |
681 | struct xfs_buf *agbp, | |
682 | xfs_ino_t parent, | |
683 | xfs_ino_t *inop) | |
1da177e4 | 684 | { |
f2ecc5e4 CH |
685 | struct xfs_mount *mp = tp->t_mountp; |
686 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
687 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
688 | xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent); | |
689 | xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent); | |
690 | struct xfs_perag *pag; | |
691 | struct xfs_btree_cur *cur, *tcur; | |
692 | struct xfs_inobt_rec_incore rec, trec; | |
693 | xfs_ino_t ino; | |
694 | int error; | |
695 | int offset; | |
696 | int i, j; | |
1da177e4 | 697 | |
44b56e0a | 698 | pag = xfs_perag_get(mp, agno); |
bd169565 | 699 | |
4bb61069 CH |
700 | ASSERT(pag->pagi_init); |
701 | ASSERT(pag->pagi_inodeok); | |
702 | ASSERT(pag->pagi_freecount > 0); | |
703 | ||
bd169565 | 704 | restart_pagno: |
f2ecc5e4 | 705 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno); |
1da177e4 LT |
706 | /* |
707 | * If pagino is 0 (this is the root inode allocation) use newino. | |
708 | * This must work because we've just allocated some. | |
709 | */ | |
710 | if (!pagino) | |
16259e7d | 711 | pagino = be32_to_cpu(agi->agi_newino); |
1da177e4 | 712 | |
0b48db80 DC |
713 | error = xfs_check_agi_freecount(cur, agi); |
714 | if (error) | |
715 | goto error0; | |
1da177e4 | 716 | |
1da177e4 | 717 | /* |
4254b0bb | 718 | * If in the same AG as the parent, try to get near the parent. |
1da177e4 LT |
719 | */ |
720 | if (pagno == agno) { | |
4254b0bb CH |
721 | int doneleft; /* done, to the left */ |
722 | int doneright; /* done, to the right */ | |
bd169565 | 723 | int searchdistance = 10; |
4254b0bb | 724 | |
21875505 | 725 | error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i); |
4254b0bb | 726 | if (error) |
1da177e4 | 727 | goto error0; |
4254b0bb CH |
728 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); |
729 | ||
730 | error = xfs_inobt_get_rec(cur, &rec, &j); | |
731 | if (error) | |
732 | goto error0; | |
b121099d | 733 | XFS_WANT_CORRUPTED_GOTO(j == 1, error0); |
4254b0bb CH |
734 | |
735 | if (rec.ir_freecount > 0) { | |
1da177e4 LT |
736 | /* |
737 | * Found a free inode in the same chunk | |
4254b0bb | 738 | * as the parent, done. |
1da177e4 | 739 | */ |
4254b0bb | 740 | goto alloc_inode; |
1da177e4 | 741 | } |
4254b0bb CH |
742 | |
743 | ||
1da177e4 | 744 | /* |
4254b0bb | 745 | * In the same AG as parent, but parent's chunk is full. |
1da177e4 | 746 | */ |
1da177e4 | 747 | |
4254b0bb CH |
748 | /* duplicate the cursor, search left & right simultaneously */ |
749 | error = xfs_btree_dup_cursor(cur, &tcur); | |
750 | if (error) | |
751 | goto error0; | |
752 | ||
bd169565 DC |
753 | /* |
754 | * Skip to last blocks looked up if same parent inode. | |
755 | */ | |
756 | if (pagino != NULLAGINO && | |
757 | pag->pagl_pagino == pagino && | |
758 | pag->pagl_leftrec != NULLAGINO && | |
759 | pag->pagl_rightrec != NULLAGINO) { | |
760 | error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec, | |
43df2ee6 | 761 | &trec, &doneleft); |
bd169565 DC |
762 | if (error) |
763 | goto error1; | |
4254b0bb | 764 | |
bd169565 | 765 | error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec, |
43df2ee6 | 766 | &rec, &doneright); |
bd169565 DC |
767 | if (error) |
768 | goto error1; | |
769 | } else { | |
770 | /* search left with tcur, back up 1 record */ | |
771 | error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1); | |
772 | if (error) | |
773 | goto error1; | |
774 | ||
775 | /* search right with cur, go forward 1 record. */ | |
776 | error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0); | |
777 | if (error) | |
778 | goto error1; | |
779 | } | |
4254b0bb CH |
780 | |
781 | /* | |
782 | * Loop until we find an inode chunk with a free inode. | |
783 | */ | |
784 | while (!doneleft || !doneright) { | |
785 | int useleft; /* using left inode chunk this time */ | |
786 | ||
bd169565 DC |
787 | if (!--searchdistance) { |
788 | /* | |
789 | * Not in range - save last search | |
790 | * location and allocate a new inode | |
791 | */ | |
3b826386 | 792 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); |
bd169565 DC |
793 | pag->pagl_leftrec = trec.ir_startino; |
794 | pag->pagl_rightrec = rec.ir_startino; | |
795 | pag->pagl_pagino = pagino; | |
796 | goto newino; | |
797 | } | |
798 | ||
4254b0bb CH |
799 | /* figure out the closer block if both are valid. */ |
800 | if (!doneleft && !doneright) { | |
801 | useleft = pagino - | |
802 | (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) < | |
803 | rec.ir_startino - pagino; | |
804 | } else { | |
805 | useleft = !doneleft; | |
1da177e4 | 806 | } |
4254b0bb CH |
807 | |
808 | /* free inodes to the left? */ | |
809 | if (useleft && trec.ir_freecount) { | |
810 | rec = trec; | |
811 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
812 | cur = tcur; | |
bd169565 DC |
813 | |
814 | pag->pagl_leftrec = trec.ir_startino; | |
815 | pag->pagl_rightrec = rec.ir_startino; | |
816 | pag->pagl_pagino = pagino; | |
4254b0bb | 817 | goto alloc_inode; |
1da177e4 | 818 | } |
1da177e4 | 819 | |
4254b0bb CH |
820 | /* free inodes to the right? */ |
821 | if (!useleft && rec.ir_freecount) { | |
822 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
bd169565 DC |
823 | |
824 | pag->pagl_leftrec = trec.ir_startino; | |
825 | pag->pagl_rightrec = rec.ir_startino; | |
826 | pag->pagl_pagino = pagino; | |
4254b0bb | 827 | goto alloc_inode; |
1da177e4 | 828 | } |
4254b0bb CH |
829 | |
830 | /* get next record to check */ | |
831 | if (useleft) { | |
832 | error = xfs_ialloc_next_rec(tcur, &trec, | |
833 | &doneleft, 1); | |
834 | } else { | |
835 | error = xfs_ialloc_next_rec(cur, &rec, | |
836 | &doneright, 0); | |
837 | } | |
838 | if (error) | |
839 | goto error1; | |
1da177e4 | 840 | } |
bd169565 DC |
841 | |
842 | /* | |
843 | * We've reached the end of the btree. because | |
844 | * we are only searching a small chunk of the | |
845 | * btree each search, there is obviously free | |
846 | * inodes closer to the parent inode than we | |
847 | * are now. restart the search again. | |
848 | */ | |
849 | pag->pagl_pagino = NULLAGINO; | |
850 | pag->pagl_leftrec = NULLAGINO; | |
851 | pag->pagl_rightrec = NULLAGINO; | |
852 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
853 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
854 | goto restart_pagno; | |
1da177e4 | 855 | } |
4254b0bb | 856 | |
1da177e4 | 857 | /* |
4254b0bb | 858 | * In a different AG from the parent. |
1da177e4 LT |
859 | * See if the most recently allocated block has any free. |
860 | */ | |
bd169565 | 861 | newino: |
69ef921b | 862 | if (agi->agi_newino != cpu_to_be32(NULLAGINO)) { |
21875505 CH |
863 | error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino), |
864 | XFS_LOOKUP_EQ, &i); | |
4254b0bb | 865 | if (error) |
1da177e4 | 866 | goto error0; |
4254b0bb CH |
867 | |
868 | if (i == 1) { | |
869 | error = xfs_inobt_get_rec(cur, &rec, &j); | |
870 | if (error) | |
871 | goto error0; | |
872 | ||
873 | if (j == 1 && rec.ir_freecount > 0) { | |
874 | /* | |
875 | * The last chunk allocated in the group | |
876 | * still has a free inode. | |
877 | */ | |
878 | goto alloc_inode; | |
879 | } | |
1da177e4 | 880 | } |
bd169565 | 881 | } |
4254b0bb | 882 | |
bd169565 DC |
883 | /* |
884 | * None left in the last group, search the whole AG | |
885 | */ | |
886 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); | |
887 | if (error) | |
888 | goto error0; | |
889 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
890 | ||
891 | for (;;) { | |
892 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
893 | if (error) | |
894 | goto error0; | |
895 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
896 | if (rec.ir_freecount > 0) | |
897 | break; | |
898 | error = xfs_btree_increment(cur, 0, &i); | |
4254b0bb CH |
899 | if (error) |
900 | goto error0; | |
901 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1da177e4 | 902 | } |
4254b0bb CH |
903 | |
904 | alloc_inode: | |
824c3131 | 905 | offset = xfs_lowbit64(rec.ir_free); |
1da177e4 LT |
906 | ASSERT(offset >= 0); |
907 | ASSERT(offset < XFS_INODES_PER_CHUNK); | |
908 | ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) % | |
909 | XFS_INODES_PER_CHUNK) == 0); | |
910 | ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset); | |
0d87e656 | 911 | rec.ir_free &= ~XFS_INOBT_MASK(offset); |
1da177e4 | 912 | rec.ir_freecount--; |
afabc24a CH |
913 | error = xfs_inobt_update(cur, &rec); |
914 | if (error) | |
1da177e4 | 915 | goto error0; |
413d57c9 | 916 | be32_add_cpu(&agi->agi_freecount, -1); |
1da177e4 | 917 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); |
44b56e0a | 918 | pag->pagi_freecount--; |
1da177e4 | 919 | |
0b48db80 DC |
920 | error = xfs_check_agi_freecount(cur, agi); |
921 | if (error) | |
922 | goto error0; | |
923 | ||
1da177e4 LT |
924 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
925 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1); | |
44b56e0a | 926 | xfs_perag_put(pag); |
1da177e4 LT |
927 | *inop = ino; |
928 | return 0; | |
929 | error1: | |
930 | xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR); | |
931 | error0: | |
932 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
44b56e0a | 933 | xfs_perag_put(pag); |
1da177e4 LT |
934 | return error; |
935 | } | |
936 | ||
f2ecc5e4 CH |
937 | /* |
938 | * Allocate an inode on disk. | |
939 | * | |
940 | * Mode is used to tell whether the new inode will need space, and whether it | |
941 | * is a directory. | |
942 | * | |
943 | * This function is designed to be called twice if it has to do an allocation | |
944 | * to make more free inodes. On the first call, *IO_agbp should be set to NULL. | |
945 | * If an inode is available without having to performn an allocation, an inode | |
cd856db6 CM |
946 | * number is returned. In this case, *IO_agbp is set to NULL. If an allocation |
947 | * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp. | |
948 | * The caller should then commit the current transaction, allocate a | |
f2ecc5e4 CH |
949 | * new transaction, and call xfs_dialloc() again, passing in the previous value |
950 | * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI | |
951 | * buffer is locked across the two calls, the second call is guaranteed to have | |
952 | * a free inode available. | |
953 | * | |
954 | * Once we successfully pick an inode its number is returned and the on-disk | |
955 | * data structures are updated. The inode itself is not read in, since doing so | |
956 | * would break ordering constraints with xfs_reclaim. | |
957 | */ | |
958 | int | |
959 | xfs_dialloc( | |
960 | struct xfs_trans *tp, | |
961 | xfs_ino_t parent, | |
962 | umode_t mode, | |
963 | int okalloc, | |
964 | struct xfs_buf **IO_agbp, | |
f2ecc5e4 CH |
965 | xfs_ino_t *inop) |
966 | { | |
55d6af64 | 967 | struct xfs_mount *mp = tp->t_mountp; |
f2ecc5e4 CH |
968 | struct xfs_buf *agbp; |
969 | xfs_agnumber_t agno; | |
f2ecc5e4 CH |
970 | int error; |
971 | int ialloced; | |
972 | int noroom = 0; | |
be60fe54 | 973 | xfs_agnumber_t start_agno; |
f2ecc5e4 CH |
974 | struct xfs_perag *pag; |
975 | ||
4bb61069 | 976 | if (*IO_agbp) { |
f2ecc5e4 | 977 | /* |
4bb61069 CH |
978 | * If the caller passes in a pointer to the AGI buffer, |
979 | * continue where we left off before. In this case, we | |
f2ecc5e4 CH |
980 | * know that the allocation group has free inodes. |
981 | */ | |
982 | agbp = *IO_agbp; | |
4bb61069 | 983 | goto out_alloc; |
f2ecc5e4 | 984 | } |
4bb61069 CH |
985 | |
986 | /* | |
987 | * We do not have an agbp, so select an initial allocation | |
988 | * group for inode allocation. | |
989 | */ | |
be60fe54 CH |
990 | start_agno = xfs_ialloc_ag_select(tp, parent, mode, okalloc); |
991 | if (start_agno == NULLAGNUMBER) { | |
4bb61069 CH |
992 | *inop = NULLFSINO; |
993 | return 0; | |
994 | } | |
55d6af64 | 995 | |
f2ecc5e4 CH |
996 | /* |
997 | * If we have already hit the ceiling of inode blocks then clear | |
998 | * okalloc so we scan all available agi structures for a free | |
999 | * inode. | |
1000 | */ | |
f2ecc5e4 CH |
1001 | if (mp->m_maxicount && |
1002 | mp->m_sb.sb_icount + XFS_IALLOC_INODES(mp) > mp->m_maxicount) { | |
1003 | noroom = 1; | |
1004 | okalloc = 0; | |
1005 | } | |
1006 | ||
1007 | /* | |
1008 | * Loop until we find an allocation group that either has free inodes | |
1009 | * or in which we can allocate some inodes. Iterate through the | |
1010 | * allocation groups upward, wrapping at the end. | |
1011 | */ | |
be60fe54 CH |
1012 | agno = start_agno; |
1013 | for (;;) { | |
1014 | pag = xfs_perag_get(mp, agno); | |
1015 | if (!pag->pagi_inodeok) { | |
1016 | xfs_ialloc_next_ag(mp); | |
1017 | goto nextag; | |
1018 | } | |
1019 | ||
1020 | if (!pag->pagi_init) { | |
1021 | error = xfs_ialloc_pagi_init(mp, tp, agno); | |
1022 | if (error) | |
1023 | goto out_error; | |
f2ecc5e4 | 1024 | } |
be60fe54 | 1025 | |
f2ecc5e4 | 1026 | /* |
be60fe54 | 1027 | * Do a first racy fast path check if this AG is usable. |
f2ecc5e4 | 1028 | */ |
be60fe54 CH |
1029 | if (!pag->pagi_freecount && !okalloc) |
1030 | goto nextag; | |
1031 | ||
c4982110 CH |
1032 | /* |
1033 | * Then read in the AGI buffer and recheck with the AGI buffer | |
1034 | * lock held. | |
1035 | */ | |
be60fe54 CH |
1036 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); |
1037 | if (error) | |
1038 | goto out_error; | |
1039 | ||
be60fe54 CH |
1040 | if (pag->pagi_freecount) { |
1041 | xfs_perag_put(pag); | |
1042 | goto out_alloc; | |
1043 | } | |
1044 | ||
c4982110 CH |
1045 | if (!okalloc) |
1046 | goto nextag_relse_buffer; | |
1047 | ||
be60fe54 CH |
1048 | |
1049 | error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced); | |
1050 | if (error) { | |
1051 | xfs_trans_brelse(tp, agbp); | |
1052 | ||
1053 | if (error != ENOSPC) | |
1054 | goto out_error; | |
1055 | ||
1056 | xfs_perag_put(pag); | |
f2ecc5e4 | 1057 | *inop = NULLFSINO; |
be60fe54 | 1058 | return 0; |
f2ecc5e4 | 1059 | } |
be60fe54 CH |
1060 | |
1061 | if (ialloced) { | |
1062 | /* | |
1063 | * We successfully allocated some inodes, return | |
1064 | * the current context to the caller so that it | |
1065 | * can commit the current transaction and call | |
1066 | * us again where we left off. | |
1067 | */ | |
1068 | ASSERT(pag->pagi_freecount > 0); | |
f2ecc5e4 | 1069 | xfs_perag_put(pag); |
be60fe54 CH |
1070 | |
1071 | *IO_agbp = agbp; | |
1072 | *inop = NULLFSINO; | |
1073 | return 0; | |
f2ecc5e4 | 1074 | } |
be60fe54 | 1075 | |
c4982110 CH |
1076 | nextag_relse_buffer: |
1077 | xfs_trans_brelse(tp, agbp); | |
be60fe54 | 1078 | nextag: |
f2ecc5e4 | 1079 | xfs_perag_put(pag); |
be60fe54 CH |
1080 | if (++agno == mp->m_sb.sb_agcount) |
1081 | agno = 0; | |
1082 | if (agno == start_agno) { | |
1083 | *inop = NULLFSINO; | |
1084 | return noroom ? ENOSPC : 0; | |
1085 | } | |
f2ecc5e4 CH |
1086 | } |
1087 | ||
4bb61069 | 1088 | out_alloc: |
f2ecc5e4 CH |
1089 | *IO_agbp = NULL; |
1090 | return xfs_dialloc_ag(tp, agbp, parent, inop); | |
be60fe54 CH |
1091 | out_error: |
1092 | xfs_perag_put(pag); | |
1093 | return XFS_ERROR(error); | |
f2ecc5e4 CH |
1094 | } |
1095 | ||
1da177e4 LT |
1096 | /* |
1097 | * Free disk inode. Carefully avoids touching the incore inode, all | |
1098 | * manipulations incore are the caller's responsibility. | |
1099 | * The on-disk inode is not changed by this operation, only the | |
1100 | * btree (free inode mask) is changed. | |
1101 | */ | |
1102 | int | |
1103 | xfs_difree( | |
1104 | xfs_trans_t *tp, /* transaction pointer */ | |
1105 | xfs_ino_t inode, /* inode to be freed */ | |
1106 | xfs_bmap_free_t *flist, /* extents to free */ | |
1107 | int *delete, /* set if inode cluster was deleted */ | |
1108 | xfs_ino_t *first_ino) /* first inode in deleted cluster */ | |
1109 | { | |
1110 | /* REFERENCED */ | |
1111 | xfs_agblock_t agbno; /* block number containing inode */ | |
1112 | xfs_buf_t *agbp; /* buffer containing allocation group header */ | |
1113 | xfs_agino_t agino; /* inode number relative to allocation group */ | |
1114 | xfs_agnumber_t agno; /* allocation group number */ | |
1115 | xfs_agi_t *agi; /* allocation group header */ | |
1116 | xfs_btree_cur_t *cur; /* inode btree cursor */ | |
1117 | int error; /* error return value */ | |
1118 | int i; /* result code */ | |
1119 | int ilen; /* inodes in an inode cluster */ | |
1120 | xfs_mount_t *mp; /* mount structure for filesystem */ | |
1121 | int off; /* offset of inode in inode chunk */ | |
61a25848 | 1122 | xfs_inobt_rec_incore_t rec; /* btree record */ |
44b56e0a | 1123 | struct xfs_perag *pag; |
1da177e4 LT |
1124 | |
1125 | mp = tp->t_mountp; | |
1126 | ||
1127 | /* | |
1128 | * Break up inode number into its components. | |
1129 | */ | |
1130 | agno = XFS_INO_TO_AGNO(mp, inode); | |
1131 | if (agno >= mp->m_sb.sb_agcount) { | |
0b932ccc DC |
1132 | xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).", |
1133 | __func__, agno, mp->m_sb.sb_agcount); | |
1da177e4 LT |
1134 | ASSERT(0); |
1135 | return XFS_ERROR(EINVAL); | |
1136 | } | |
1137 | agino = XFS_INO_TO_AGINO(mp, inode); | |
1138 | if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
0b932ccc DC |
1139 | xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).", |
1140 | __func__, (unsigned long long)inode, | |
1141 | (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino)); | |
1da177e4 LT |
1142 | ASSERT(0); |
1143 | return XFS_ERROR(EINVAL); | |
1144 | } | |
1145 | agbno = XFS_AGINO_TO_AGBNO(mp, agino); | |
1146 | if (agbno >= mp->m_sb.sb_agblocks) { | |
0b932ccc DC |
1147 | xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).", |
1148 | __func__, agbno, mp->m_sb.sb_agblocks); | |
1da177e4 LT |
1149 | ASSERT(0); |
1150 | return XFS_ERROR(EINVAL); | |
1151 | } | |
1152 | /* | |
1153 | * Get the allocation group header. | |
1154 | */ | |
1da177e4 | 1155 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); |
1da177e4 | 1156 | if (error) { |
0b932ccc DC |
1157 | xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.", |
1158 | __func__, error); | |
1da177e4 LT |
1159 | return error; |
1160 | } | |
1161 | agi = XFS_BUF_TO_AGI(agbp); | |
69ef921b | 1162 | ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC)); |
16259e7d | 1163 | ASSERT(agbno < be32_to_cpu(agi->agi_length)); |
1da177e4 LT |
1164 | /* |
1165 | * Initialize the cursor. | |
1166 | */ | |
561f7d17 | 1167 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno); |
1da177e4 | 1168 | |
0b48db80 DC |
1169 | error = xfs_check_agi_freecount(cur, agi); |
1170 | if (error) | |
1171 | goto error0; | |
1172 | ||
1da177e4 LT |
1173 | /* |
1174 | * Look for the entry describing this inode. | |
1175 | */ | |
21875505 | 1176 | if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) { |
0b932ccc DC |
1177 | xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.", |
1178 | __func__, error); | |
1da177e4 LT |
1179 | goto error0; |
1180 | } | |
1181 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
2e287a73 CH |
1182 | error = xfs_inobt_get_rec(cur, &rec, &i); |
1183 | if (error) { | |
0b932ccc DC |
1184 | xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.", |
1185 | __func__, error); | |
1da177e4 LT |
1186 | goto error0; |
1187 | } | |
1188 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1189 | /* | |
1190 | * Get the offset in the inode chunk. | |
1191 | */ | |
1192 | off = agino - rec.ir_startino; | |
1193 | ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK); | |
0d87e656 | 1194 | ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off))); |
1da177e4 LT |
1195 | /* |
1196 | * Mark the inode free & increment the count. | |
1197 | */ | |
0d87e656 | 1198 | rec.ir_free |= XFS_INOBT_MASK(off); |
1da177e4 LT |
1199 | rec.ir_freecount++; |
1200 | ||
1201 | /* | |
c41564b5 | 1202 | * When an inode cluster is free, it becomes eligible for removal |
1da177e4 | 1203 | */ |
1bd960ee | 1204 | if (!(mp->m_flags & XFS_MOUNT_IKEEP) && |
1da177e4 LT |
1205 | (rec.ir_freecount == XFS_IALLOC_INODES(mp))) { |
1206 | ||
1207 | *delete = 1; | |
1208 | *first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino); | |
1209 | ||
1210 | /* | |
1211 | * Remove the inode cluster from the AGI B+Tree, adjust the | |
1212 | * AGI and Superblock inode counts, and mark the disk space | |
1213 | * to be freed when the transaction is committed. | |
1214 | */ | |
1215 | ilen = XFS_IALLOC_INODES(mp); | |
413d57c9 MS |
1216 | be32_add_cpu(&agi->agi_count, -ilen); |
1217 | be32_add_cpu(&agi->agi_freecount, -(ilen - 1)); | |
1da177e4 | 1218 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT); |
44b56e0a DC |
1219 | pag = xfs_perag_get(mp, agno); |
1220 | pag->pagi_freecount -= ilen - 1; | |
1221 | xfs_perag_put(pag); | |
1da177e4 LT |
1222 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen); |
1223 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1)); | |
1224 | ||
91cca5df | 1225 | if ((error = xfs_btree_delete(cur, &i))) { |
0b932ccc DC |
1226 | xfs_warn(mp, "%s: xfs_btree_delete returned error %d.", |
1227 | __func__, error); | |
1da177e4 LT |
1228 | goto error0; |
1229 | } | |
1230 | ||
1231 | xfs_bmap_add_free(XFS_AGB_TO_FSB(mp, | |
8e825e3a | 1232 | agno, XFS_AGINO_TO_AGBNO(mp, rec.ir_startino)), |
1da177e4 LT |
1233 | XFS_IALLOC_BLOCKS(mp), flist, mp); |
1234 | } else { | |
1235 | *delete = 0; | |
1236 | ||
afabc24a CH |
1237 | error = xfs_inobt_update(cur, &rec); |
1238 | if (error) { | |
0b932ccc DC |
1239 | xfs_warn(mp, "%s: xfs_inobt_update returned error %d.", |
1240 | __func__, error); | |
1da177e4 LT |
1241 | goto error0; |
1242 | } | |
afabc24a | 1243 | |
1da177e4 LT |
1244 | /* |
1245 | * Change the inode free counts and log the ag/sb changes. | |
1246 | */ | |
413d57c9 | 1247 | be32_add_cpu(&agi->agi_freecount, 1); |
1da177e4 | 1248 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); |
44b56e0a DC |
1249 | pag = xfs_perag_get(mp, agno); |
1250 | pag->pagi_freecount++; | |
1251 | xfs_perag_put(pag); | |
1da177e4 LT |
1252 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1); |
1253 | } | |
1254 | ||
0b48db80 DC |
1255 | error = xfs_check_agi_freecount(cur, agi); |
1256 | if (error) | |
1257 | goto error0; | |
1da177e4 | 1258 | |
1da177e4 LT |
1259 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
1260 | return 0; | |
1261 | ||
1262 | error0: | |
1263 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
1264 | return error; | |
1265 | } | |
1266 | ||
7124fe0a DC |
1267 | STATIC int |
1268 | xfs_imap_lookup( | |
1269 | struct xfs_mount *mp, | |
1270 | struct xfs_trans *tp, | |
1271 | xfs_agnumber_t agno, | |
1272 | xfs_agino_t agino, | |
1273 | xfs_agblock_t agbno, | |
1274 | xfs_agblock_t *chunk_agbno, | |
1275 | xfs_agblock_t *offset_agbno, | |
1276 | int flags) | |
1277 | { | |
1278 | struct xfs_inobt_rec_incore rec; | |
1279 | struct xfs_btree_cur *cur; | |
1280 | struct xfs_buf *agbp; | |
7124fe0a DC |
1281 | int error; |
1282 | int i; | |
1283 | ||
1284 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); | |
1285 | if (error) { | |
53487786 DC |
1286 | xfs_alert(mp, |
1287 | "%s: xfs_ialloc_read_agi() returned error %d, agno %d", | |
1288 | __func__, error, agno); | |
7124fe0a DC |
1289 | return error; |
1290 | } | |
1291 | ||
1292 | /* | |
4536f2ad DC |
1293 | * Lookup the inode record for the given agino. If the record cannot be |
1294 | * found, then it's an invalid inode number and we should abort. Once | |
1295 | * we have a record, we need to ensure it contains the inode number | |
1296 | * we are looking up. | |
7124fe0a DC |
1297 | */ |
1298 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno); | |
4536f2ad | 1299 | error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i); |
7124fe0a DC |
1300 | if (!error) { |
1301 | if (i) | |
1302 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
1303 | if (!error && i == 0) | |
1304 | error = EINVAL; | |
1305 | } | |
1306 | ||
1307 | xfs_trans_brelse(tp, agbp); | |
1308 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
1309 | if (error) | |
1310 | return error; | |
1311 | ||
4536f2ad DC |
1312 | /* check that the returned record contains the required inode */ |
1313 | if (rec.ir_startino > agino || | |
1314 | rec.ir_startino + XFS_IALLOC_INODES(mp) <= agino) | |
1315 | return EINVAL; | |
1316 | ||
7124fe0a | 1317 | /* for untrusted inodes check it is allocated first */ |
1920779e | 1318 | if ((flags & XFS_IGET_UNTRUSTED) && |
7124fe0a DC |
1319 | (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino))) |
1320 | return EINVAL; | |
1321 | ||
1322 | *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino); | |
1323 | *offset_agbno = agbno - *chunk_agbno; | |
1324 | return 0; | |
1325 | } | |
1326 | ||
1da177e4 | 1327 | /* |
94e1b69d | 1328 | * Return the location of the inode in imap, for mapping it into a buffer. |
1da177e4 | 1329 | */ |
1da177e4 | 1330 | int |
94e1b69d CH |
1331 | xfs_imap( |
1332 | xfs_mount_t *mp, /* file system mount structure */ | |
1333 | xfs_trans_t *tp, /* transaction pointer */ | |
1da177e4 | 1334 | xfs_ino_t ino, /* inode to locate */ |
94e1b69d CH |
1335 | struct xfs_imap *imap, /* location map structure */ |
1336 | uint flags) /* flags for inode btree lookup */ | |
1da177e4 LT |
1337 | { |
1338 | xfs_agblock_t agbno; /* block number of inode in the alloc group */ | |
1da177e4 LT |
1339 | xfs_agino_t agino; /* inode number within alloc group */ |
1340 | xfs_agnumber_t agno; /* allocation group number */ | |
1341 | int blks_per_cluster; /* num blocks per inode cluster */ | |
1342 | xfs_agblock_t chunk_agbno; /* first block in inode chunk */ | |
1da177e4 | 1343 | xfs_agblock_t cluster_agbno; /* first block in inode cluster */ |
1da177e4 | 1344 | int error; /* error code */ |
1da177e4 | 1345 | int offset; /* index of inode in its buffer */ |
836a94ad | 1346 | xfs_agblock_t offset_agbno; /* blks from chunk start to inode */ |
1da177e4 LT |
1347 | |
1348 | ASSERT(ino != NULLFSINO); | |
94e1b69d | 1349 | |
1da177e4 LT |
1350 | /* |
1351 | * Split up the inode number into its parts. | |
1352 | */ | |
1353 | agno = XFS_INO_TO_AGNO(mp, ino); | |
1354 | agino = XFS_INO_TO_AGINO(mp, ino); | |
1355 | agbno = XFS_AGINO_TO_AGBNO(mp, agino); | |
1356 | if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks || | |
1357 | ino != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
1358 | #ifdef DEBUG | |
1920779e DC |
1359 | /* |
1360 | * Don't output diagnostic information for untrusted inodes | |
1361 | * as they can be invalid without implying corruption. | |
1362 | */ | |
1363 | if (flags & XFS_IGET_UNTRUSTED) | |
4d1a2ed3 | 1364 | return XFS_ERROR(EINVAL); |
1da177e4 | 1365 | if (agno >= mp->m_sb.sb_agcount) { |
53487786 DC |
1366 | xfs_alert(mp, |
1367 | "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)", | |
1368 | __func__, agno, mp->m_sb.sb_agcount); | |
1da177e4 LT |
1369 | } |
1370 | if (agbno >= mp->m_sb.sb_agblocks) { | |
53487786 DC |
1371 | xfs_alert(mp, |
1372 | "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)", | |
1373 | __func__, (unsigned long long)agbno, | |
1374 | (unsigned long)mp->m_sb.sb_agblocks); | |
1da177e4 LT |
1375 | } |
1376 | if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
53487786 DC |
1377 | xfs_alert(mp, |
1378 | "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)", | |
1379 | __func__, ino, | |
1380 | XFS_AGINO_TO_INO(mp, agno, agino)); | |
1da177e4 | 1381 | } |
745b1f47 | 1382 | xfs_stack_trace(); |
1da177e4 LT |
1383 | #endif /* DEBUG */ |
1384 | return XFS_ERROR(EINVAL); | |
1385 | } | |
94e1b69d | 1386 | |
7124fe0a DC |
1387 | blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_blocklog; |
1388 | ||
1389 | /* | |
1390 | * For bulkstat and handle lookups, we have an untrusted inode number | |
1391 | * that we have to verify is valid. We cannot do this just by reading | |
1392 | * the inode buffer as it may have been unlinked and removed leaving | |
1393 | * inodes in stale state on disk. Hence we have to do a btree lookup | |
1394 | * in all cases where an untrusted inode number is passed. | |
1395 | */ | |
1920779e | 1396 | if (flags & XFS_IGET_UNTRUSTED) { |
7124fe0a DC |
1397 | error = xfs_imap_lookup(mp, tp, agno, agino, agbno, |
1398 | &chunk_agbno, &offset_agbno, flags); | |
1399 | if (error) | |
1400 | return error; | |
1401 | goto out_map; | |
1402 | } | |
1403 | ||
94e1b69d CH |
1404 | /* |
1405 | * If the inode cluster size is the same as the blocksize or | |
1406 | * smaller we get to the buffer by simple arithmetics. | |
1407 | */ | |
1408 | if (XFS_INODE_CLUSTER_SIZE(mp) <= mp->m_sb.sb_blocksize) { | |
1da177e4 LT |
1409 | offset = XFS_INO_TO_OFFSET(mp, ino); |
1410 | ASSERT(offset < mp->m_sb.sb_inopblock); | |
94e1b69d CH |
1411 | |
1412 | imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno); | |
1413 | imap->im_len = XFS_FSB_TO_BB(mp, 1); | |
1414 | imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog); | |
1da177e4 LT |
1415 | return 0; |
1416 | } | |
94e1b69d | 1417 | |
94e1b69d CH |
1418 | /* |
1419 | * If the inode chunks are aligned then use simple maths to | |
1420 | * find the location. Otherwise we have to do a btree | |
1421 | * lookup to find the location. | |
1422 | */ | |
1da177e4 LT |
1423 | if (mp->m_inoalign_mask) { |
1424 | offset_agbno = agbno & mp->m_inoalign_mask; | |
1425 | chunk_agbno = agbno - offset_agbno; | |
1426 | } else { | |
7124fe0a DC |
1427 | error = xfs_imap_lookup(mp, tp, agno, agino, agbno, |
1428 | &chunk_agbno, &offset_agbno, flags); | |
1da177e4 LT |
1429 | if (error) |
1430 | return error; | |
1da177e4 | 1431 | } |
94e1b69d | 1432 | |
7124fe0a | 1433 | out_map: |
1da177e4 LT |
1434 | ASSERT(agbno >= chunk_agbno); |
1435 | cluster_agbno = chunk_agbno + | |
1436 | ((offset_agbno / blks_per_cluster) * blks_per_cluster); | |
1437 | offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) + | |
1438 | XFS_INO_TO_OFFSET(mp, ino); | |
94e1b69d CH |
1439 | |
1440 | imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno); | |
1441 | imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster); | |
1442 | imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog); | |
1443 | ||
1444 | /* | |
1445 | * If the inode number maps to a block outside the bounds | |
1446 | * of the file system then return NULL rather than calling | |
1447 | * read_buf and panicing when we get an error from the | |
1448 | * driver. | |
1449 | */ | |
1450 | if ((imap->im_blkno + imap->im_len) > | |
1451 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) { | |
53487786 DC |
1452 | xfs_alert(mp, |
1453 | "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)", | |
1454 | __func__, (unsigned long long) imap->im_blkno, | |
94e1b69d CH |
1455 | (unsigned long long) imap->im_len, |
1456 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)); | |
1457 | return XFS_ERROR(EINVAL); | |
1458 | } | |
1da177e4 | 1459 | return 0; |
1da177e4 LT |
1460 | } |
1461 | ||
1462 | /* | |
1463 | * Compute and fill in value of m_in_maxlevels. | |
1464 | */ | |
1465 | void | |
1466 | xfs_ialloc_compute_maxlevels( | |
1467 | xfs_mount_t *mp) /* file system mount structure */ | |
1468 | { | |
1469 | int level; | |
1470 | uint maxblocks; | |
1471 | uint maxleafents; | |
1472 | int minleafrecs; | |
1473 | int minnoderecs; | |
1474 | ||
1475 | maxleafents = (1LL << XFS_INO_AGINO_BITS(mp)) >> | |
1476 | XFS_INODES_PER_CHUNK_LOG; | |
1477 | minleafrecs = mp->m_alloc_mnr[0]; | |
1478 | minnoderecs = mp->m_alloc_mnr[1]; | |
1479 | maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs; | |
1480 | for (level = 1; maxblocks > 1; level++) | |
1481 | maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs; | |
1482 | mp->m_in_maxlevels = level; | |
1483 | } | |
1484 | ||
1485 | /* | |
1486 | * Log specified fields for the ag hdr (inode section) | |
1487 | */ | |
1488 | void | |
1489 | xfs_ialloc_log_agi( | |
1490 | xfs_trans_t *tp, /* transaction pointer */ | |
1491 | xfs_buf_t *bp, /* allocation group header buffer */ | |
1492 | int fields) /* bitmask of fields to log */ | |
1493 | { | |
1494 | int first; /* first byte number */ | |
1495 | int last; /* last byte number */ | |
1496 | static const short offsets[] = { /* field starting offsets */ | |
1497 | /* keep in sync with bit definitions */ | |
1498 | offsetof(xfs_agi_t, agi_magicnum), | |
1499 | offsetof(xfs_agi_t, agi_versionnum), | |
1500 | offsetof(xfs_agi_t, agi_seqno), | |
1501 | offsetof(xfs_agi_t, agi_length), | |
1502 | offsetof(xfs_agi_t, agi_count), | |
1503 | offsetof(xfs_agi_t, agi_root), | |
1504 | offsetof(xfs_agi_t, agi_level), | |
1505 | offsetof(xfs_agi_t, agi_freecount), | |
1506 | offsetof(xfs_agi_t, agi_newino), | |
1507 | offsetof(xfs_agi_t, agi_dirino), | |
1508 | offsetof(xfs_agi_t, agi_unlinked), | |
1509 | sizeof(xfs_agi_t) | |
1510 | }; | |
1511 | #ifdef DEBUG | |
1512 | xfs_agi_t *agi; /* allocation group header */ | |
1513 | ||
1514 | agi = XFS_BUF_TO_AGI(bp); | |
69ef921b | 1515 | ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC)); |
1da177e4 LT |
1516 | #endif |
1517 | /* | |
1518 | * Compute byte offsets for the first and last fields. | |
1519 | */ | |
1520 | xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS, &first, &last); | |
1521 | /* | |
1522 | * Log the allocation group inode header buffer. | |
1523 | */ | |
61fe135c | 1524 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_AGI_BUF); |
1da177e4 LT |
1525 | xfs_trans_log_buf(tp, bp, first, last); |
1526 | } | |
1527 | ||
5e1be0fb CH |
1528 | #ifdef DEBUG |
1529 | STATIC void | |
1530 | xfs_check_agi_unlinked( | |
1531 | struct xfs_agi *agi) | |
1532 | { | |
1533 | int i; | |
1534 | ||
1535 | for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++) | |
1536 | ASSERT(agi->agi_unlinked[i]); | |
1537 | } | |
1538 | #else | |
1539 | #define xfs_check_agi_unlinked(agi) | |
1540 | #endif | |
1541 | ||
983d09ff | 1542 | static bool |
612cfbfe | 1543 | xfs_agi_verify( |
3702ce6e DC |
1544 | struct xfs_buf *bp) |
1545 | { | |
1546 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
1547 | struct xfs_agi *agi = XFS_BUF_TO_AGI(bp); | |
3702ce6e | 1548 | |
983d09ff DC |
1549 | if (xfs_sb_version_hascrc(&mp->m_sb) && |
1550 | !uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_uuid)) | |
1551 | return false; | |
3702ce6e DC |
1552 | /* |
1553 | * Validate the magic number of the agi block. | |
1554 | */ | |
983d09ff DC |
1555 | if (agi->agi_magicnum != cpu_to_be32(XFS_AGI_MAGIC)) |
1556 | return false; | |
1557 | if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum))) | |
1558 | return false; | |
3702ce6e DC |
1559 | |
1560 | /* | |
1561 | * during growfs operations, the perag is not fully initialised, | |
1562 | * so we can't use it for any useful checking. growfs ensures we can't | |
1563 | * use it by using uncached buffers that don't have the perag attached | |
1564 | * so we can detect and avoid this problem. | |
1565 | */ | |
983d09ff DC |
1566 | if (bp->b_pag && be32_to_cpu(agi->agi_seqno) != bp->b_pag->pag_agno) |
1567 | return false; | |
3702ce6e | 1568 | |
3702ce6e | 1569 | xfs_check_agi_unlinked(agi); |
983d09ff | 1570 | return true; |
612cfbfe DC |
1571 | } |
1572 | ||
1813dd64 DC |
1573 | static void |
1574 | xfs_agi_read_verify( | |
612cfbfe DC |
1575 | struct xfs_buf *bp) |
1576 | { | |
983d09ff DC |
1577 | struct xfs_mount *mp = bp->b_target->bt_mount; |
1578 | int agi_ok = 1; | |
1579 | ||
1580 | if (xfs_sb_version_hascrc(&mp->m_sb)) | |
1581 | agi_ok = xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length), | |
1582 | offsetof(struct xfs_agi, agi_crc)); | |
1583 | agi_ok = agi_ok && xfs_agi_verify(bp); | |
1584 | ||
1585 | if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IALLOC_READ_AGI, | |
1586 | XFS_RANDOM_IALLOC_READ_AGI))) { | |
1587 | XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr); | |
1588 | xfs_buf_ioerror(bp, EFSCORRUPTED); | |
1589 | } | |
612cfbfe DC |
1590 | } |
1591 | ||
b0f539de | 1592 | static void |
1813dd64 | 1593 | xfs_agi_write_verify( |
612cfbfe DC |
1594 | struct xfs_buf *bp) |
1595 | { | |
983d09ff DC |
1596 | struct xfs_mount *mp = bp->b_target->bt_mount; |
1597 | struct xfs_buf_log_item *bip = bp->b_fspriv; | |
1598 | ||
1599 | if (!xfs_agi_verify(bp)) { | |
1600 | XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr); | |
1601 | xfs_buf_ioerror(bp, EFSCORRUPTED); | |
1602 | return; | |
1603 | } | |
1604 | ||
1605 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
1606 | return; | |
1607 | ||
1608 | if (bip) | |
1609 | XFS_BUF_TO_AGI(bp)->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn); | |
1610 | xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length), | |
1611 | offsetof(struct xfs_agi, agi_crc)); | |
3702ce6e DC |
1612 | } |
1613 | ||
1813dd64 DC |
1614 | const struct xfs_buf_ops xfs_agi_buf_ops = { |
1615 | .verify_read = xfs_agi_read_verify, | |
1616 | .verify_write = xfs_agi_write_verify, | |
1617 | }; | |
1618 | ||
1da177e4 LT |
1619 | /* |
1620 | * Read in the allocation group header (inode allocation section) | |
1621 | */ | |
1622 | int | |
5e1be0fb CH |
1623 | xfs_read_agi( |
1624 | struct xfs_mount *mp, /* file system mount structure */ | |
1625 | struct xfs_trans *tp, /* transaction pointer */ | |
1626 | xfs_agnumber_t agno, /* allocation group number */ | |
1627 | struct xfs_buf **bpp) /* allocation group hdr buf */ | |
1da177e4 | 1628 | { |
5e1be0fb | 1629 | int error; |
1da177e4 | 1630 | |
d123031a | 1631 | trace_xfs_read_agi(mp, agno); |
5e1be0fb | 1632 | |
d123031a | 1633 | ASSERT(agno != NULLAGNUMBER); |
5e1be0fb | 1634 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, |
1da177e4 | 1635 | XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)), |
1813dd64 | 1636 | XFS_FSS_TO_BB(mp, 1), 0, bpp, &xfs_agi_buf_ops); |
1da177e4 LT |
1637 | if (error) |
1638 | return error; | |
5e1be0fb | 1639 | |
5a52c2a5 | 1640 | ASSERT(!xfs_buf_geterror(*bpp)); |
38f23232 | 1641 | xfs_buf_set_ref(*bpp, XFS_AGI_REF); |
5e1be0fb CH |
1642 | return 0; |
1643 | } | |
1644 | ||
1645 | int | |
1646 | xfs_ialloc_read_agi( | |
1647 | struct xfs_mount *mp, /* file system mount structure */ | |
1648 | struct xfs_trans *tp, /* transaction pointer */ | |
1649 | xfs_agnumber_t agno, /* allocation group number */ | |
1650 | struct xfs_buf **bpp) /* allocation group hdr buf */ | |
1651 | { | |
1652 | struct xfs_agi *agi; /* allocation group header */ | |
1653 | struct xfs_perag *pag; /* per allocation group data */ | |
1654 | int error; | |
1655 | ||
d123031a DC |
1656 | trace_xfs_ialloc_read_agi(mp, agno); |
1657 | ||
5e1be0fb CH |
1658 | error = xfs_read_agi(mp, tp, agno, bpp); |
1659 | if (error) | |
1660 | return error; | |
1661 | ||
1662 | agi = XFS_BUF_TO_AGI(*bpp); | |
44b56e0a | 1663 | pag = xfs_perag_get(mp, agno); |
1da177e4 | 1664 | if (!pag->pagi_init) { |
16259e7d | 1665 | pag->pagi_freecount = be32_to_cpu(agi->agi_freecount); |
92821e2b | 1666 | pag->pagi_count = be32_to_cpu(agi->agi_count); |
1da177e4 | 1667 | pag->pagi_init = 1; |
1da177e4 | 1668 | } |
1da177e4 | 1669 | |
5e1be0fb CH |
1670 | /* |
1671 | * It's possible for these to be out of sync if | |
1672 | * we are in the middle of a forced shutdown. | |
1673 | */ | |
1674 | ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) || | |
1675 | XFS_FORCED_SHUTDOWN(mp)); | |
44b56e0a | 1676 | xfs_perag_put(pag); |
1da177e4 LT |
1677 | return 0; |
1678 | } | |
92821e2b DC |
1679 | |
1680 | /* | |
1681 | * Read in the agi to initialise the per-ag data in the mount structure | |
1682 | */ | |
1683 | int | |
1684 | xfs_ialloc_pagi_init( | |
1685 | xfs_mount_t *mp, /* file system mount structure */ | |
1686 | xfs_trans_t *tp, /* transaction pointer */ | |
1687 | xfs_agnumber_t agno) /* allocation group number */ | |
1688 | { | |
1689 | xfs_buf_t *bp = NULL; | |
1690 | int error; | |
1691 | ||
1692 | error = xfs_ialloc_read_agi(mp, tp, agno, &bp); | |
1693 | if (error) | |
1694 | return error; | |
1695 | if (bp) | |
1696 | xfs_trans_brelse(tp, bp); | |
1697 | return 0; | |
1698 | } |