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" |
1da177e4 | 20 | #include "xfs_types.h" |
a844f451 | 21 | #include "xfs_bit.h" |
1da177e4 | 22 | #include "xfs_log.h" |
a844f451 | 23 | #include "xfs_inum.h" |
1da177e4 LT |
24 | #include "xfs_trans.h" |
25 | #include "xfs_sb.h" | |
26 | #include "xfs_ag.h" | |
1da177e4 | 27 | #include "xfs_mount.h" |
1da177e4 | 28 | #include "xfs_bmap_btree.h" |
a844f451 | 29 | #include "xfs_alloc_btree.h" |
1da177e4 | 30 | #include "xfs_ialloc_btree.h" |
1da177e4 LT |
31 | #include "xfs_dinode.h" |
32 | #include "xfs_inode.h" | |
a844f451 NS |
33 | #include "xfs_btree.h" |
34 | #include "xfs_ialloc.h" | |
1da177e4 | 35 | #include "xfs_alloc.h" |
1da177e4 LT |
36 | #include "xfs_rtalloc.h" |
37 | #include "xfs_error.h" | |
38 | #include "xfs_bmap.h" | |
983d09ff DC |
39 | #include "xfs_cksum.h" |
40 | #include "xfs_buf_item.h" | |
ddf6ad01 | 41 | #include "xfs_icreate_item.h" |
7bb85ef3 | 42 | #include "xfs_icache.h" |
1da177e4 | 43 | |
1da177e4 LT |
44 | |
45 | /* | |
46 | * Allocation group level functions. | |
47 | */ | |
75de2a91 DC |
48 | static inline int |
49 | xfs_ialloc_cluster_alignment( | |
50 | xfs_alloc_arg_t *args) | |
51 | { | |
52 | if (xfs_sb_version_hasalign(&args->mp->m_sb) && | |
53 | args->mp->m_sb.sb_inoalignmt >= | |
54 | XFS_B_TO_FSBT(args->mp, XFS_INODE_CLUSTER_SIZE(args->mp))) | |
55 | return args->mp->m_sb.sb_inoalignmt; | |
56 | return 1; | |
57 | } | |
1da177e4 | 58 | |
fe033cc8 | 59 | /* |
21875505 | 60 | * Lookup a record by ino in the btree given by cur. |
fe033cc8 | 61 | */ |
81e25176 | 62 | int /* error */ |
21875505 | 63 | xfs_inobt_lookup( |
fe033cc8 CH |
64 | struct xfs_btree_cur *cur, /* btree cursor */ |
65 | xfs_agino_t ino, /* starting inode of chunk */ | |
21875505 | 66 | xfs_lookup_t dir, /* <=, >=, == */ |
fe033cc8 CH |
67 | int *stat) /* success/failure */ |
68 | { | |
69 | cur->bc_rec.i.ir_startino = ino; | |
21875505 CH |
70 | cur->bc_rec.i.ir_freecount = 0; |
71 | cur->bc_rec.i.ir_free = 0; | |
72 | return xfs_btree_lookup(cur, dir, stat); | |
fe033cc8 CH |
73 | } |
74 | ||
278d0ca1 | 75 | /* |
afabc24a | 76 | * Update the record referred to by cur to the value given. |
278d0ca1 CH |
77 | * This either works (return 0) or gets an EFSCORRUPTED error. |
78 | */ | |
79 | STATIC int /* error */ | |
80 | xfs_inobt_update( | |
81 | struct xfs_btree_cur *cur, /* btree cursor */ | |
afabc24a | 82 | xfs_inobt_rec_incore_t *irec) /* btree record */ |
278d0ca1 CH |
83 | { |
84 | union xfs_btree_rec rec; | |
85 | ||
afabc24a CH |
86 | rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino); |
87 | rec.inobt.ir_freecount = cpu_to_be32(irec->ir_freecount); | |
88 | rec.inobt.ir_free = cpu_to_be64(irec->ir_free); | |
278d0ca1 CH |
89 | return xfs_btree_update(cur, &rec); |
90 | } | |
91 | ||
8cc938fe CH |
92 | /* |
93 | * Get the data from the pointed-to record. | |
94 | */ | |
95 | int /* error */ | |
96 | xfs_inobt_get_rec( | |
97 | struct xfs_btree_cur *cur, /* btree cursor */ | |
2e287a73 | 98 | xfs_inobt_rec_incore_t *irec, /* btree record */ |
8cc938fe CH |
99 | int *stat) /* output: success/failure */ |
100 | { | |
101 | union xfs_btree_rec *rec; | |
102 | int error; | |
103 | ||
104 | error = xfs_btree_get_rec(cur, &rec, stat); | |
105 | if (!error && *stat == 1) { | |
2e287a73 CH |
106 | irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino); |
107 | irec->ir_freecount = be32_to_cpu(rec->inobt.ir_freecount); | |
108 | irec->ir_free = be64_to_cpu(rec->inobt.ir_free); | |
8cc938fe CH |
109 | } |
110 | return error; | |
111 | } | |
112 | ||
0b48db80 DC |
113 | /* |
114 | * Verify that the number of free inodes in the AGI is correct. | |
115 | */ | |
116 | #ifdef DEBUG | |
117 | STATIC int | |
118 | xfs_check_agi_freecount( | |
119 | struct xfs_btree_cur *cur, | |
120 | struct xfs_agi *agi) | |
121 | { | |
122 | if (cur->bc_nlevels == 1) { | |
123 | xfs_inobt_rec_incore_t rec; | |
124 | int freecount = 0; | |
125 | int error; | |
126 | int i; | |
127 | ||
21875505 | 128 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); |
0b48db80 DC |
129 | if (error) |
130 | return error; | |
131 | ||
132 | do { | |
133 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
134 | if (error) | |
135 | return error; | |
136 | ||
137 | if (i) { | |
138 | freecount += rec.ir_freecount; | |
139 | error = xfs_btree_increment(cur, 0, &i); | |
140 | if (error) | |
141 | return error; | |
142 | } | |
143 | } while (i == 1); | |
144 | ||
145 | if (!XFS_FORCED_SHUTDOWN(cur->bc_mp)) | |
146 | ASSERT(freecount == be32_to_cpu(agi->agi_freecount)); | |
147 | } | |
148 | return 0; | |
149 | } | |
150 | #else | |
151 | #define xfs_check_agi_freecount(cur, agi) 0 | |
152 | #endif | |
153 | ||
85c0b2ab | 154 | /* |
28c8e41a DC |
155 | * Initialise a new set of inodes. When called without a transaction context |
156 | * (e.g. from recovery) we initiate a delayed write of the inode buffers rather | |
157 | * than logging them (which in a transaction context puts them into the AIL | |
158 | * for writeback rather than the xfsbufd queue). | |
85c0b2ab | 159 | */ |
ddf6ad01 | 160 | int |
85c0b2ab DC |
161 | xfs_ialloc_inode_init( |
162 | struct xfs_mount *mp, | |
163 | struct xfs_trans *tp, | |
28c8e41a | 164 | struct list_head *buffer_list, |
85c0b2ab DC |
165 | xfs_agnumber_t agno, |
166 | xfs_agblock_t agbno, | |
167 | xfs_agblock_t length, | |
168 | unsigned int gen) | |
169 | { | |
170 | struct xfs_buf *fbuf; | |
171 | struct xfs_dinode *free; | |
172 | int blks_per_cluster, nbufs, ninodes; | |
173 | int version; | |
174 | int i, j; | |
175 | xfs_daddr_t d; | |
93848a99 | 176 | xfs_ino_t ino = 0; |
85c0b2ab DC |
177 | |
178 | /* | |
179 | * Loop over the new block(s), filling in the inodes. | |
180 | * For small block sizes, manipulate the inodes in buffers | |
181 | * which are multiples of the blocks size. | |
182 | */ | |
183 | if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) { | |
184 | blks_per_cluster = 1; | |
185 | nbufs = length; | |
186 | ninodes = mp->m_sb.sb_inopblock; | |
187 | } else { | |
188 | blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) / | |
189 | mp->m_sb.sb_blocksize; | |
190 | nbufs = length / blks_per_cluster; | |
191 | ninodes = blks_per_cluster * mp->m_sb.sb_inopblock; | |
192 | } | |
193 | ||
194 | /* | |
93848a99 CH |
195 | * Figure out what version number to use in the inodes we create. If |
196 | * the superblock version has caught up to the one that supports the new | |
197 | * inode format, then use the new inode version. Otherwise use the old | |
198 | * version so that old kernels will continue to be able to use the file | |
199 | * system. | |
200 | * | |
201 | * For v3 inodes, we also need to write the inode number into the inode, | |
202 | * so calculate the first inode number of the chunk here as | |
203 | * XFS_OFFBNO_TO_AGINO() only works within a filesystem block, not | |
204 | * across multiple filesystem blocks (such as a cluster) and so cannot | |
205 | * be used in the cluster buffer loop below. | |
206 | * | |
207 | * Further, because we are writing the inode directly into the buffer | |
208 | * and calculating a CRC on the entire inode, we have ot log the entire | |
209 | * inode so that the entire range the CRC covers is present in the log. | |
210 | * That means for v3 inode we log the entire buffer rather than just the | |
211 | * inode cores. | |
85c0b2ab | 212 | */ |
93848a99 CH |
213 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
214 | version = 3; | |
215 | ino = XFS_AGINO_TO_INO(mp, agno, | |
216 | XFS_OFFBNO_TO_AGINO(mp, agbno, 0)); | |
ddf6ad01 DC |
217 | |
218 | /* | |
219 | * log the initialisation that is about to take place as an | |
220 | * logical operation. This means the transaction does not | |
221 | * need to log the physical changes to the inode buffers as log | |
222 | * recovery will know what initialisation is actually needed. | |
223 | * Hence we only need to log the buffers as "ordered" buffers so | |
224 | * they track in the AIL as if they were physically logged. | |
225 | */ | |
226 | if (tp) | |
227 | xfs_icreate_log(tp, agno, agbno, XFS_IALLOC_INODES(mp), | |
228 | mp->m_sb.sb_inodesize, length, gen); | |
93848a99 | 229 | } else if (xfs_sb_version_hasnlink(&mp->m_sb)) |
85c0b2ab DC |
230 | version = 2; |
231 | else | |
232 | version = 1; | |
233 | ||
234 | for (j = 0; j < nbufs; j++) { | |
235 | /* | |
236 | * Get the block. | |
237 | */ | |
238 | d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster)); | |
239 | fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, | |
7c4cebe8 DC |
240 | mp->m_bsize * blks_per_cluster, |
241 | XBF_UNMAPPED); | |
2a30f36d CS |
242 | if (!fbuf) |
243 | return ENOMEM; | |
ddf6ad01 DC |
244 | |
245 | /* Initialize the inode buffers and log them appropriately. */ | |
1813dd64 | 246 | fbuf->b_ops = &xfs_inode_buf_ops; |
93848a99 | 247 | xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length)); |
85c0b2ab DC |
248 | for (i = 0; i < ninodes; i++) { |
249 | int ioffset = i << mp->m_sb.sb_inodelog; | |
93848a99 | 250 | uint isize = xfs_dinode_size(version); |
85c0b2ab DC |
251 | |
252 | free = xfs_make_iptr(mp, fbuf, i); | |
253 | free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC); | |
254 | free->di_version = version; | |
255 | free->di_gen = cpu_to_be32(gen); | |
256 | free->di_next_unlinked = cpu_to_be32(NULLAGINO); | |
93848a99 CH |
257 | |
258 | if (version == 3) { | |
259 | free->di_ino = cpu_to_be64(ino); | |
260 | ino++; | |
261 | uuid_copy(&free->di_uuid, &mp->m_sb.sb_uuid); | |
262 | xfs_dinode_calc_crc(mp, free); | |
28c8e41a | 263 | } else if (tp) { |
93848a99 CH |
264 | /* just log the inode core */ |
265 | xfs_trans_log_buf(tp, fbuf, ioffset, | |
266 | ioffset + isize - 1); | |
267 | } | |
268 | } | |
28c8e41a DC |
269 | |
270 | if (tp) { | |
271 | /* | |
272 | * Mark the buffer as an inode allocation buffer so it | |
273 | * sticks in AIL at the point of this allocation | |
274 | * transaction. This ensures the they are on disk before | |
275 | * the tail of the log can be moved past this | |
276 | * transaction (i.e. by preventing relogging from moving | |
277 | * it forward in the log). | |
278 | */ | |
279 | xfs_trans_inode_alloc_buf(tp, fbuf); | |
280 | if (version == 3) { | |
ddf6ad01 DC |
281 | /* |
282 | * Mark the buffer as ordered so that they are | |
283 | * not physically logged in the transaction but | |
284 | * still tracked in the AIL as part of the | |
285 | * transaction and pin the log appropriately. | |
286 | */ | |
287 | xfs_trans_ordered_buf(tp, fbuf); | |
28c8e41a DC |
288 | xfs_trans_log_buf(tp, fbuf, 0, |
289 | BBTOB(fbuf->b_length) - 1); | |
290 | } | |
291 | } else { | |
292 | fbuf->b_flags |= XBF_DONE; | |
293 | xfs_buf_delwri_queue(fbuf, buffer_list); | |
294 | xfs_buf_relse(fbuf); | |
85c0b2ab | 295 | } |
85c0b2ab | 296 | } |
2a30f36d | 297 | return 0; |
85c0b2ab DC |
298 | } |
299 | ||
1da177e4 LT |
300 | /* |
301 | * Allocate new inodes in the allocation group specified by agbp. | |
302 | * Return 0 for success, else error code. | |
303 | */ | |
304 | STATIC int /* error code or 0 */ | |
305 | xfs_ialloc_ag_alloc( | |
306 | xfs_trans_t *tp, /* transaction pointer */ | |
307 | xfs_buf_t *agbp, /* alloc group buffer */ | |
308 | int *alloc) | |
309 | { | |
310 | xfs_agi_t *agi; /* allocation group header */ | |
311 | xfs_alloc_arg_t args; /* allocation argument structure */ | |
1da177e4 | 312 | xfs_btree_cur_t *cur; /* inode btree cursor */ |
92821e2b | 313 | xfs_agnumber_t agno; |
1da177e4 | 314 | int error; |
85c0b2ab | 315 | int i; |
1da177e4 LT |
316 | xfs_agino_t newino; /* new first inode's number */ |
317 | xfs_agino_t newlen; /* new number of inodes */ | |
1da177e4 | 318 | xfs_agino_t thisino; /* current inode number, for loop */ |
3ccb8b5f | 319 | int isaligned = 0; /* inode allocation at stripe unit */ |
1da177e4 | 320 | /* boundary */ |
44b56e0a | 321 | struct xfs_perag *pag; |
1da177e4 | 322 | |
a0041684 | 323 | memset(&args, 0, sizeof(args)); |
1da177e4 LT |
324 | args.tp = tp; |
325 | args.mp = tp->t_mountp; | |
326 | ||
327 | /* | |
328 | * Locking will ensure that we don't have two callers in here | |
329 | * at one time. | |
330 | */ | |
331 | newlen = XFS_IALLOC_INODES(args.mp); | |
332 | if (args.mp->m_maxicount && | |
333 | args.mp->m_sb.sb_icount + newlen > args.mp->m_maxicount) | |
334 | return XFS_ERROR(ENOSPC); | |
335 | args.minlen = args.maxlen = XFS_IALLOC_BLOCKS(args.mp); | |
336 | /* | |
3ccb8b5f GO |
337 | * First try to allocate inodes contiguous with the last-allocated |
338 | * chunk of inodes. If the filesystem is striped, this will fill | |
339 | * an entire stripe unit with inodes. | |
28c8e41a | 340 | */ |
1da177e4 | 341 | agi = XFS_BUF_TO_AGI(agbp); |
3ccb8b5f | 342 | newino = be32_to_cpu(agi->agi_newino); |
85c0b2ab | 343 | agno = be32_to_cpu(agi->agi_seqno); |
019ff2d5 NS |
344 | args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) + |
345 | XFS_IALLOC_BLOCKS(args.mp); | |
346 | if (likely(newino != NULLAGINO && | |
347 | (args.agbno < be32_to_cpu(agi->agi_length)))) { | |
85c0b2ab | 348 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
3ccb8b5f | 349 | args.type = XFS_ALLOCTYPE_THIS_BNO; |
3ccb8b5f | 350 | args.prod = 1; |
75de2a91 | 351 | |
3ccb8b5f | 352 | /* |
75de2a91 DC |
353 | * We need to take into account alignment here to ensure that |
354 | * we don't modify the free list if we fail to have an exact | |
355 | * block. If we don't have an exact match, and every oher | |
356 | * attempt allocation attempt fails, we'll end up cancelling | |
357 | * a dirty transaction and shutting down. | |
358 | * | |
359 | * For an exact allocation, alignment must be 1, | |
360 | * however we need to take cluster alignment into account when | |
361 | * fixing up the freelist. Use the minalignslop field to | |
362 | * indicate that extra blocks might be required for alignment, | |
363 | * but not to use them in the actual exact allocation. | |
3ccb8b5f | 364 | */ |
75de2a91 DC |
365 | args.alignment = 1; |
366 | args.minalignslop = xfs_ialloc_cluster_alignment(&args) - 1; | |
367 | ||
368 | /* Allow space for the inode btree to split. */ | |
0d87e656 | 369 | args.minleft = args.mp->m_in_maxlevels - 1; |
3ccb8b5f GO |
370 | if ((error = xfs_alloc_vextent(&args))) |
371 | return error; | |
372 | } else | |
373 | args.fsbno = NULLFSBLOCK; | |
1da177e4 | 374 | |
3ccb8b5f GO |
375 | if (unlikely(args.fsbno == NULLFSBLOCK)) { |
376 | /* | |
377 | * Set the alignment for the allocation. | |
378 | * If stripe alignment is turned on then align at stripe unit | |
379 | * boundary. | |
019ff2d5 NS |
380 | * If the cluster size is smaller than a filesystem block |
381 | * then we're doing I/O for inodes in filesystem block size | |
3ccb8b5f GO |
382 | * pieces, so don't need alignment anyway. |
383 | */ | |
384 | isaligned = 0; | |
385 | if (args.mp->m_sinoalign) { | |
386 | ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN)); | |
387 | args.alignment = args.mp->m_dalign; | |
388 | isaligned = 1; | |
75de2a91 DC |
389 | } else |
390 | args.alignment = xfs_ialloc_cluster_alignment(&args); | |
3ccb8b5f GO |
391 | /* |
392 | * Need to figure out where to allocate the inode blocks. | |
393 | * Ideally they should be spaced out through the a.g. | |
394 | * For now, just allocate blocks up front. | |
395 | */ | |
396 | args.agbno = be32_to_cpu(agi->agi_root); | |
85c0b2ab | 397 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
3ccb8b5f GO |
398 | /* |
399 | * Allocate a fixed-size extent of inodes. | |
400 | */ | |
401 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
3ccb8b5f GO |
402 | args.prod = 1; |
403 | /* | |
404 | * Allow space for the inode btree to split. | |
405 | */ | |
0d87e656 | 406 | args.minleft = args.mp->m_in_maxlevels - 1; |
3ccb8b5f GO |
407 | if ((error = xfs_alloc_vextent(&args))) |
408 | return error; | |
409 | } | |
019ff2d5 | 410 | |
1da177e4 LT |
411 | /* |
412 | * If stripe alignment is turned on, then try again with cluster | |
413 | * alignment. | |
414 | */ | |
415 | if (isaligned && args.fsbno == NULLFSBLOCK) { | |
416 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
16259e7d | 417 | args.agbno = be32_to_cpu(agi->agi_root); |
85c0b2ab | 418 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
75de2a91 | 419 | args.alignment = xfs_ialloc_cluster_alignment(&args); |
1da177e4 LT |
420 | if ((error = xfs_alloc_vextent(&args))) |
421 | return error; | |
422 | } | |
423 | ||
424 | if (args.fsbno == NULLFSBLOCK) { | |
425 | *alloc = 0; | |
426 | return 0; | |
427 | } | |
428 | ASSERT(args.len == args.minlen); | |
1da177e4 | 429 | |
359346a9 | 430 | /* |
85c0b2ab DC |
431 | * Stamp and write the inode buffers. |
432 | * | |
359346a9 DC |
433 | * Seed the new inode cluster with a random generation number. This |
434 | * prevents short-term reuse of generation numbers if a chunk is | |
435 | * freed and then immediately reallocated. We use random numbers | |
436 | * rather than a linear progression to prevent the next generation | |
437 | * number from being easily guessable. | |
438 | */ | |
28c8e41a | 439 | error = xfs_ialloc_inode_init(args.mp, tp, NULL, agno, args.agbno, |
ecb3403d | 440 | args.len, prandom_u32()); |
d42f08f6 | 441 | |
2a30f36d CS |
442 | if (error) |
443 | return error; | |
85c0b2ab DC |
444 | /* |
445 | * Convert the results. | |
446 | */ | |
447 | newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0); | |
413d57c9 MS |
448 | be32_add_cpu(&agi->agi_count, newlen); |
449 | be32_add_cpu(&agi->agi_freecount, newlen); | |
44b56e0a DC |
450 | pag = xfs_perag_get(args.mp, agno); |
451 | pag->pagi_freecount += newlen; | |
452 | xfs_perag_put(pag); | |
16259e7d | 453 | agi->agi_newino = cpu_to_be32(newino); |
85c0b2ab | 454 | |
1da177e4 LT |
455 | /* |
456 | * Insert records describing the new inode chunk into the btree. | |
457 | */ | |
561f7d17 | 458 | cur = xfs_inobt_init_cursor(args.mp, tp, agbp, agno); |
1da177e4 LT |
459 | for (thisino = newino; |
460 | thisino < newino + newlen; | |
461 | thisino += XFS_INODES_PER_CHUNK) { | |
21875505 CH |
462 | cur->bc_rec.i.ir_startino = thisino; |
463 | cur->bc_rec.i.ir_freecount = XFS_INODES_PER_CHUNK; | |
464 | cur->bc_rec.i.ir_free = XFS_INOBT_ALL_FREE; | |
465 | error = xfs_btree_lookup(cur, XFS_LOOKUP_EQ, &i); | |
466 | if (error) { | |
1da177e4 LT |
467 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); |
468 | return error; | |
469 | } | |
470 | ASSERT(i == 0); | |
21875505 CH |
471 | error = xfs_btree_insert(cur, &i); |
472 | if (error) { | |
1da177e4 LT |
473 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); |
474 | return error; | |
475 | } | |
476 | ASSERT(i == 1); | |
477 | } | |
478 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
479 | /* | |
480 | * Log allocation group header fields | |
481 | */ | |
482 | xfs_ialloc_log_agi(tp, agbp, | |
483 | XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO); | |
484 | /* | |
485 | * Modify/log superblock values for inode count and inode free count. | |
486 | */ | |
487 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen); | |
488 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen); | |
489 | *alloc = 1; | |
490 | return 0; | |
491 | } | |
492 | ||
b8f82a4a | 493 | STATIC xfs_agnumber_t |
1da177e4 LT |
494 | xfs_ialloc_next_ag( |
495 | xfs_mount_t *mp) | |
496 | { | |
497 | xfs_agnumber_t agno; | |
498 | ||
499 | spin_lock(&mp->m_agirotor_lock); | |
500 | agno = mp->m_agirotor; | |
8aea3ff4 | 501 | if (++mp->m_agirotor >= mp->m_maxagi) |
1da177e4 LT |
502 | mp->m_agirotor = 0; |
503 | spin_unlock(&mp->m_agirotor_lock); | |
504 | ||
505 | return agno; | |
506 | } | |
507 | ||
508 | /* | |
509 | * Select an allocation group to look for a free inode in, based on the parent | |
2f21ff1c | 510 | * inode and the mode. Return the allocation group buffer. |
1da177e4 | 511 | */ |
55d6af64 | 512 | STATIC xfs_agnumber_t |
1da177e4 LT |
513 | xfs_ialloc_ag_select( |
514 | xfs_trans_t *tp, /* transaction pointer */ | |
515 | xfs_ino_t parent, /* parent directory inode number */ | |
576b1d67 | 516 | umode_t mode, /* bits set to indicate file type */ |
1da177e4 LT |
517 | int okalloc) /* ok to allocate more space */ |
518 | { | |
1da177e4 LT |
519 | xfs_agnumber_t agcount; /* number of ag's in the filesystem */ |
520 | xfs_agnumber_t agno; /* current ag number */ | |
521 | int flags; /* alloc buffer locking flags */ | |
522 | xfs_extlen_t ineed; /* blocks needed for inode allocation */ | |
523 | xfs_extlen_t longest = 0; /* longest extent available */ | |
524 | xfs_mount_t *mp; /* mount point structure */ | |
525 | int needspace; /* file mode implies space allocated */ | |
526 | xfs_perag_t *pag; /* per allocation group data */ | |
527 | xfs_agnumber_t pagno; /* parent (starting) ag number */ | |
55d6af64 | 528 | int error; |
1da177e4 LT |
529 | |
530 | /* | |
531 | * Files of these types need at least one block if length > 0 | |
532 | * (and they won't fit in the inode, but that's hard to figure out). | |
533 | */ | |
534 | needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode); | |
535 | mp = tp->t_mountp; | |
536 | agcount = mp->m_maxagi; | |
537 | if (S_ISDIR(mode)) | |
538 | pagno = xfs_ialloc_next_ag(mp); | |
539 | else { | |
540 | pagno = XFS_INO_TO_AGNO(mp, parent); | |
541 | if (pagno >= agcount) | |
542 | pagno = 0; | |
543 | } | |
55d6af64 | 544 | |
1da177e4 | 545 | ASSERT(pagno < agcount); |
55d6af64 | 546 | |
1da177e4 LT |
547 | /* |
548 | * Loop through allocation groups, looking for one with a little | |
549 | * free space in it. Note we don't look for free inodes, exactly. | |
550 | * Instead, we include whether there is a need to allocate inodes | |
551 | * to mean that blocks must be allocated for them, | |
552 | * if none are currently free. | |
553 | */ | |
554 | agno = pagno; | |
555 | flags = XFS_ALLOC_FLAG_TRYLOCK; | |
1da177e4 | 556 | for (;;) { |
44b56e0a | 557 | pag = xfs_perag_get(mp, agno); |
55d6af64 CH |
558 | if (!pag->pagi_inodeok) { |
559 | xfs_ialloc_next_ag(mp); | |
560 | goto nextag; | |
561 | } | |
562 | ||
1da177e4 | 563 | if (!pag->pagi_init) { |
55d6af64 CH |
564 | error = xfs_ialloc_pagi_init(mp, tp, agno); |
565 | if (error) | |
1da177e4 | 566 | goto nextag; |
55d6af64 | 567 | } |
1da177e4 | 568 | |
55d6af64 CH |
569 | if (pag->pagi_freecount) { |
570 | xfs_perag_put(pag); | |
571 | return agno; | |
1da177e4 LT |
572 | } |
573 | ||
55d6af64 CH |
574 | if (!okalloc) |
575 | goto nextag; | |
576 | ||
577 | if (!pag->pagf_init) { | |
578 | error = xfs_alloc_pagf_init(mp, tp, agno, flags); | |
579 | if (error) | |
1da177e4 | 580 | goto nextag; |
1da177e4 | 581 | } |
55d6af64 CH |
582 | |
583 | /* | |
584 | * Is there enough free space for the file plus a block of | |
585 | * inodes? (if we need to allocate some)? | |
586 | */ | |
587 | ineed = XFS_IALLOC_BLOCKS(mp); | |
588 | longest = pag->pagf_longest; | |
589 | if (!longest) | |
590 | longest = pag->pagf_flcount > 0; | |
591 | ||
592 | if (pag->pagf_freeblks >= needspace + ineed && | |
593 | longest >= ineed) { | |
594 | xfs_perag_put(pag); | |
595 | return agno; | |
1da177e4 | 596 | } |
1da177e4 | 597 | nextag: |
44b56e0a | 598 | xfs_perag_put(pag); |
1da177e4 LT |
599 | /* |
600 | * No point in iterating over the rest, if we're shutting | |
601 | * down. | |
602 | */ | |
1c1c6ebc | 603 | if (XFS_FORCED_SHUTDOWN(mp)) |
55d6af64 | 604 | return NULLAGNUMBER; |
1da177e4 LT |
605 | agno++; |
606 | if (agno >= agcount) | |
607 | agno = 0; | |
608 | if (agno == pagno) { | |
1c1c6ebc | 609 | if (flags == 0) |
55d6af64 | 610 | return NULLAGNUMBER; |
1da177e4 LT |
611 | flags = 0; |
612 | } | |
613 | } | |
614 | } | |
615 | ||
4254b0bb CH |
616 | /* |
617 | * Try to retrieve the next record to the left/right from the current one. | |
618 | */ | |
619 | STATIC int | |
620 | xfs_ialloc_next_rec( | |
621 | struct xfs_btree_cur *cur, | |
622 | xfs_inobt_rec_incore_t *rec, | |
623 | int *done, | |
624 | int left) | |
625 | { | |
626 | int error; | |
627 | int i; | |
628 | ||
629 | if (left) | |
630 | error = xfs_btree_decrement(cur, 0, &i); | |
631 | else | |
632 | error = xfs_btree_increment(cur, 0, &i); | |
633 | ||
634 | if (error) | |
635 | return error; | |
636 | *done = !i; | |
637 | if (i) { | |
638 | error = xfs_inobt_get_rec(cur, rec, &i); | |
639 | if (error) | |
640 | return error; | |
641 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
642 | } | |
643 | ||
644 | return 0; | |
645 | } | |
646 | ||
bd169565 DC |
647 | STATIC int |
648 | xfs_ialloc_get_rec( | |
649 | struct xfs_btree_cur *cur, | |
650 | xfs_agino_t agino, | |
651 | xfs_inobt_rec_incore_t *rec, | |
43df2ee6 | 652 | int *done) |
bd169565 DC |
653 | { |
654 | int error; | |
655 | int i; | |
656 | ||
657 | error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i); | |
658 | if (error) | |
659 | return error; | |
660 | *done = !i; | |
661 | if (i) { | |
662 | error = xfs_inobt_get_rec(cur, rec, &i); | |
663 | if (error) | |
664 | return error; | |
665 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
666 | } | |
667 | ||
668 | return 0; | |
669 | } | |
0b48db80 | 670 | |
1da177e4 | 671 | /* |
f2ecc5e4 | 672 | * Allocate an inode. |
1da177e4 | 673 | * |
f2ecc5e4 CH |
674 | * The caller selected an AG for us, and made sure that free inodes are |
675 | * available. | |
1da177e4 | 676 | */ |
f2ecc5e4 CH |
677 | STATIC int |
678 | xfs_dialloc_ag( | |
679 | struct xfs_trans *tp, | |
680 | struct xfs_buf *agbp, | |
681 | xfs_ino_t parent, | |
682 | xfs_ino_t *inop) | |
1da177e4 | 683 | { |
f2ecc5e4 CH |
684 | struct xfs_mount *mp = tp->t_mountp; |
685 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
686 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
687 | xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent); | |
688 | xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent); | |
689 | struct xfs_perag *pag; | |
690 | struct xfs_btree_cur *cur, *tcur; | |
691 | struct xfs_inobt_rec_incore rec, trec; | |
692 | xfs_ino_t ino; | |
693 | int error; | |
694 | int offset; | |
695 | int i, j; | |
1da177e4 | 696 | |
44b56e0a | 697 | pag = xfs_perag_get(mp, agno); |
bd169565 | 698 | |
4bb61069 CH |
699 | ASSERT(pag->pagi_init); |
700 | ASSERT(pag->pagi_inodeok); | |
701 | ASSERT(pag->pagi_freecount > 0); | |
702 | ||
bd169565 | 703 | restart_pagno: |
f2ecc5e4 | 704 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno); |
1da177e4 LT |
705 | /* |
706 | * If pagino is 0 (this is the root inode allocation) use newino. | |
707 | * This must work because we've just allocated some. | |
708 | */ | |
709 | if (!pagino) | |
16259e7d | 710 | pagino = be32_to_cpu(agi->agi_newino); |
1da177e4 | 711 | |
0b48db80 DC |
712 | error = xfs_check_agi_freecount(cur, agi); |
713 | if (error) | |
714 | goto error0; | |
1da177e4 | 715 | |
1da177e4 | 716 | /* |
4254b0bb | 717 | * If in the same AG as the parent, try to get near the parent. |
1da177e4 LT |
718 | */ |
719 | if (pagno == agno) { | |
4254b0bb CH |
720 | int doneleft; /* done, to the left */ |
721 | int doneright; /* done, to the right */ | |
bd169565 | 722 | int searchdistance = 10; |
4254b0bb | 723 | |
21875505 | 724 | error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i); |
4254b0bb | 725 | if (error) |
1da177e4 | 726 | goto error0; |
4254b0bb CH |
727 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); |
728 | ||
729 | error = xfs_inobt_get_rec(cur, &rec, &j); | |
730 | if (error) | |
731 | goto error0; | |
732 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
733 | ||
734 | if (rec.ir_freecount > 0) { | |
1da177e4 LT |
735 | /* |
736 | * Found a free inode in the same chunk | |
4254b0bb | 737 | * as the parent, done. |
1da177e4 | 738 | */ |
4254b0bb | 739 | goto alloc_inode; |
1da177e4 | 740 | } |
4254b0bb CH |
741 | |
742 | ||
1da177e4 | 743 | /* |
4254b0bb | 744 | * In the same AG as parent, but parent's chunk is full. |
1da177e4 | 745 | */ |
1da177e4 | 746 | |
4254b0bb CH |
747 | /* duplicate the cursor, search left & right simultaneously */ |
748 | error = xfs_btree_dup_cursor(cur, &tcur); | |
749 | if (error) | |
750 | goto error0; | |
751 | ||
bd169565 DC |
752 | /* |
753 | * Skip to last blocks looked up if same parent inode. | |
754 | */ | |
755 | if (pagino != NULLAGINO && | |
756 | pag->pagl_pagino == pagino && | |
757 | pag->pagl_leftrec != NULLAGINO && | |
758 | pag->pagl_rightrec != NULLAGINO) { | |
759 | error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec, | |
43df2ee6 | 760 | &trec, &doneleft); |
bd169565 DC |
761 | if (error) |
762 | goto error1; | |
4254b0bb | 763 | |
bd169565 | 764 | error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec, |
43df2ee6 | 765 | &rec, &doneright); |
bd169565 DC |
766 | if (error) |
767 | goto error1; | |
768 | } else { | |
769 | /* search left with tcur, back up 1 record */ | |
770 | error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1); | |
771 | if (error) | |
772 | goto error1; | |
773 | ||
774 | /* search right with cur, go forward 1 record. */ | |
775 | error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0); | |
776 | if (error) | |
777 | goto error1; | |
778 | } | |
4254b0bb CH |
779 | |
780 | /* | |
781 | * Loop until we find an inode chunk with a free inode. | |
782 | */ | |
783 | while (!doneleft || !doneright) { | |
784 | int useleft; /* using left inode chunk this time */ | |
785 | ||
bd169565 DC |
786 | if (!--searchdistance) { |
787 | /* | |
788 | * Not in range - save last search | |
789 | * location and allocate a new inode | |
790 | */ | |
3b826386 | 791 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); |
bd169565 DC |
792 | pag->pagl_leftrec = trec.ir_startino; |
793 | pag->pagl_rightrec = rec.ir_startino; | |
794 | pag->pagl_pagino = pagino; | |
795 | goto newino; | |
796 | } | |
797 | ||
4254b0bb CH |
798 | /* figure out the closer block if both are valid. */ |
799 | if (!doneleft && !doneright) { | |
800 | useleft = pagino - | |
801 | (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) < | |
802 | rec.ir_startino - pagino; | |
803 | } else { | |
804 | useleft = !doneleft; | |
1da177e4 | 805 | } |
4254b0bb CH |
806 | |
807 | /* free inodes to the left? */ | |
808 | if (useleft && trec.ir_freecount) { | |
809 | rec = trec; | |
810 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
811 | cur = tcur; | |
bd169565 DC |
812 | |
813 | pag->pagl_leftrec = trec.ir_startino; | |
814 | pag->pagl_rightrec = rec.ir_startino; | |
815 | pag->pagl_pagino = pagino; | |
4254b0bb | 816 | goto alloc_inode; |
1da177e4 | 817 | } |
1da177e4 | 818 | |
4254b0bb CH |
819 | /* free inodes to the right? */ |
820 | if (!useleft && rec.ir_freecount) { | |
821 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
bd169565 DC |
822 | |
823 | pag->pagl_leftrec = trec.ir_startino; | |
824 | pag->pagl_rightrec = rec.ir_startino; | |
825 | pag->pagl_pagino = pagino; | |
4254b0bb | 826 | goto alloc_inode; |
1da177e4 | 827 | } |
4254b0bb CH |
828 | |
829 | /* get next record to check */ | |
830 | if (useleft) { | |
831 | error = xfs_ialloc_next_rec(tcur, &trec, | |
832 | &doneleft, 1); | |
833 | } else { | |
834 | error = xfs_ialloc_next_rec(cur, &rec, | |
835 | &doneright, 0); | |
836 | } | |
837 | if (error) | |
838 | goto error1; | |
1da177e4 | 839 | } |
bd169565 DC |
840 | |
841 | /* | |
842 | * We've reached the end of the btree. because | |
843 | * we are only searching a small chunk of the | |
844 | * btree each search, there is obviously free | |
845 | * inodes closer to the parent inode than we | |
846 | * are now. restart the search again. | |
847 | */ | |
848 | pag->pagl_pagino = NULLAGINO; | |
849 | pag->pagl_leftrec = NULLAGINO; | |
850 | pag->pagl_rightrec = NULLAGINO; | |
851 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
852 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
853 | goto restart_pagno; | |
1da177e4 | 854 | } |
4254b0bb | 855 | |
1da177e4 | 856 | /* |
4254b0bb | 857 | * In a different AG from the parent. |
1da177e4 LT |
858 | * See if the most recently allocated block has any free. |
859 | */ | |
bd169565 | 860 | newino: |
69ef921b | 861 | if (agi->agi_newino != cpu_to_be32(NULLAGINO)) { |
21875505 CH |
862 | error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino), |
863 | XFS_LOOKUP_EQ, &i); | |
4254b0bb | 864 | if (error) |
1da177e4 | 865 | goto error0; |
4254b0bb CH |
866 | |
867 | if (i == 1) { | |
868 | error = xfs_inobt_get_rec(cur, &rec, &j); | |
869 | if (error) | |
870 | goto error0; | |
871 | ||
872 | if (j == 1 && rec.ir_freecount > 0) { | |
873 | /* | |
874 | * The last chunk allocated in the group | |
875 | * still has a free inode. | |
876 | */ | |
877 | goto alloc_inode; | |
878 | } | |
1da177e4 | 879 | } |
bd169565 | 880 | } |
4254b0bb | 881 | |
bd169565 DC |
882 | /* |
883 | * None left in the last group, search the whole AG | |
884 | */ | |
885 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); | |
886 | if (error) | |
887 | goto error0; | |
888 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
889 | ||
890 | for (;;) { | |
891 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
892 | if (error) | |
893 | goto error0; | |
894 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
895 | if (rec.ir_freecount > 0) | |
896 | break; | |
897 | error = xfs_btree_increment(cur, 0, &i); | |
4254b0bb CH |
898 | if (error) |
899 | goto error0; | |
900 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1da177e4 | 901 | } |
4254b0bb CH |
902 | |
903 | alloc_inode: | |
824c3131 | 904 | offset = xfs_lowbit64(rec.ir_free); |
1da177e4 LT |
905 | ASSERT(offset >= 0); |
906 | ASSERT(offset < XFS_INODES_PER_CHUNK); | |
907 | ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) % | |
908 | XFS_INODES_PER_CHUNK) == 0); | |
909 | ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset); | |
0d87e656 | 910 | rec.ir_free &= ~XFS_INOBT_MASK(offset); |
1da177e4 | 911 | rec.ir_freecount--; |
afabc24a CH |
912 | error = xfs_inobt_update(cur, &rec); |
913 | if (error) | |
1da177e4 | 914 | goto error0; |
413d57c9 | 915 | be32_add_cpu(&agi->agi_freecount, -1); |
1da177e4 | 916 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); |
44b56e0a | 917 | pag->pagi_freecount--; |
1da177e4 | 918 | |
0b48db80 DC |
919 | error = xfs_check_agi_freecount(cur, agi); |
920 | if (error) | |
921 | goto error0; | |
922 | ||
1da177e4 LT |
923 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
924 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1); | |
44b56e0a | 925 | xfs_perag_put(pag); |
1da177e4 LT |
926 | *inop = ino; |
927 | return 0; | |
928 | error1: | |
929 | xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR); | |
930 | error0: | |
931 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
44b56e0a | 932 | xfs_perag_put(pag); |
1da177e4 LT |
933 | return error; |
934 | } | |
935 | ||
f2ecc5e4 CH |
936 | /* |
937 | * Allocate an inode on disk. | |
938 | * | |
939 | * Mode is used to tell whether the new inode will need space, and whether it | |
940 | * is a directory. | |
941 | * | |
942 | * This function is designed to be called twice if it has to do an allocation | |
943 | * to make more free inodes. On the first call, *IO_agbp should be set to NULL. | |
944 | * If an inode is available without having to performn an allocation, an inode | |
cd856db6 CM |
945 | * number is returned. In this case, *IO_agbp is set to NULL. If an allocation |
946 | * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp. | |
947 | * The caller should then commit the current transaction, allocate a | |
f2ecc5e4 CH |
948 | * new transaction, and call xfs_dialloc() again, passing in the previous value |
949 | * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI | |
950 | * buffer is locked across the two calls, the second call is guaranteed to have | |
951 | * a free inode available. | |
952 | * | |
953 | * Once we successfully pick an inode its number is returned and the on-disk | |
954 | * data structures are updated. The inode itself is not read in, since doing so | |
955 | * would break ordering constraints with xfs_reclaim. | |
956 | */ | |
957 | int | |
958 | xfs_dialloc( | |
959 | struct xfs_trans *tp, | |
960 | xfs_ino_t parent, | |
961 | umode_t mode, | |
962 | int okalloc, | |
963 | struct xfs_buf **IO_agbp, | |
f2ecc5e4 CH |
964 | xfs_ino_t *inop) |
965 | { | |
55d6af64 | 966 | struct xfs_mount *mp = tp->t_mountp; |
f2ecc5e4 CH |
967 | struct xfs_buf *agbp; |
968 | xfs_agnumber_t agno; | |
f2ecc5e4 CH |
969 | int error; |
970 | int ialloced; | |
971 | int noroom = 0; | |
be60fe54 | 972 | xfs_agnumber_t start_agno; |
f2ecc5e4 CH |
973 | struct xfs_perag *pag; |
974 | ||
4bb61069 | 975 | if (*IO_agbp) { |
f2ecc5e4 | 976 | /* |
4bb61069 CH |
977 | * If the caller passes in a pointer to the AGI buffer, |
978 | * continue where we left off before. In this case, we | |
f2ecc5e4 CH |
979 | * know that the allocation group has free inodes. |
980 | */ | |
981 | agbp = *IO_agbp; | |
4bb61069 | 982 | goto out_alloc; |
f2ecc5e4 | 983 | } |
4bb61069 CH |
984 | |
985 | /* | |
986 | * We do not have an agbp, so select an initial allocation | |
987 | * group for inode allocation. | |
988 | */ | |
be60fe54 CH |
989 | start_agno = xfs_ialloc_ag_select(tp, parent, mode, okalloc); |
990 | if (start_agno == NULLAGNUMBER) { | |
4bb61069 CH |
991 | *inop = NULLFSINO; |
992 | return 0; | |
993 | } | |
55d6af64 | 994 | |
f2ecc5e4 CH |
995 | /* |
996 | * If we have already hit the ceiling of inode blocks then clear | |
997 | * okalloc so we scan all available agi structures for a free | |
998 | * inode. | |
999 | */ | |
f2ecc5e4 CH |
1000 | if (mp->m_maxicount && |
1001 | mp->m_sb.sb_icount + XFS_IALLOC_INODES(mp) > mp->m_maxicount) { | |
1002 | noroom = 1; | |
1003 | okalloc = 0; | |
1004 | } | |
1005 | ||
1006 | /* | |
1007 | * Loop until we find an allocation group that either has free inodes | |
1008 | * or in which we can allocate some inodes. Iterate through the | |
1009 | * allocation groups upward, wrapping at the end. | |
1010 | */ | |
be60fe54 CH |
1011 | agno = start_agno; |
1012 | for (;;) { | |
1013 | pag = xfs_perag_get(mp, agno); | |
1014 | if (!pag->pagi_inodeok) { | |
1015 | xfs_ialloc_next_ag(mp); | |
1016 | goto nextag; | |
1017 | } | |
1018 | ||
1019 | if (!pag->pagi_init) { | |
1020 | error = xfs_ialloc_pagi_init(mp, tp, agno); | |
1021 | if (error) | |
1022 | goto out_error; | |
f2ecc5e4 | 1023 | } |
be60fe54 | 1024 | |
f2ecc5e4 | 1025 | /* |
be60fe54 | 1026 | * Do a first racy fast path check if this AG is usable. |
f2ecc5e4 | 1027 | */ |
be60fe54 CH |
1028 | if (!pag->pagi_freecount && !okalloc) |
1029 | goto nextag; | |
1030 | ||
c4982110 CH |
1031 | /* |
1032 | * Then read in the AGI buffer and recheck with the AGI buffer | |
1033 | * lock held. | |
1034 | */ | |
be60fe54 CH |
1035 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); |
1036 | if (error) | |
1037 | goto out_error; | |
1038 | ||
be60fe54 CH |
1039 | if (pag->pagi_freecount) { |
1040 | xfs_perag_put(pag); | |
1041 | goto out_alloc; | |
1042 | } | |
1043 | ||
c4982110 CH |
1044 | if (!okalloc) |
1045 | goto nextag_relse_buffer; | |
1046 | ||
be60fe54 CH |
1047 | |
1048 | error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced); | |
1049 | if (error) { | |
1050 | xfs_trans_brelse(tp, agbp); | |
1051 | ||
1052 | if (error != ENOSPC) | |
1053 | goto out_error; | |
1054 | ||
1055 | xfs_perag_put(pag); | |
f2ecc5e4 | 1056 | *inop = NULLFSINO; |
be60fe54 | 1057 | return 0; |
f2ecc5e4 | 1058 | } |
be60fe54 CH |
1059 | |
1060 | if (ialloced) { | |
1061 | /* | |
1062 | * We successfully allocated some inodes, return | |
1063 | * the current context to the caller so that it | |
1064 | * can commit the current transaction and call | |
1065 | * us again where we left off. | |
1066 | */ | |
1067 | ASSERT(pag->pagi_freecount > 0); | |
f2ecc5e4 | 1068 | xfs_perag_put(pag); |
be60fe54 CH |
1069 | |
1070 | *IO_agbp = agbp; | |
1071 | *inop = NULLFSINO; | |
1072 | return 0; | |
f2ecc5e4 | 1073 | } |
be60fe54 | 1074 | |
c4982110 CH |
1075 | nextag_relse_buffer: |
1076 | xfs_trans_brelse(tp, agbp); | |
be60fe54 | 1077 | nextag: |
f2ecc5e4 | 1078 | xfs_perag_put(pag); |
be60fe54 CH |
1079 | if (++agno == mp->m_sb.sb_agcount) |
1080 | agno = 0; | |
1081 | if (agno == start_agno) { | |
1082 | *inop = NULLFSINO; | |
1083 | return noroom ? ENOSPC : 0; | |
1084 | } | |
f2ecc5e4 CH |
1085 | } |
1086 | ||
4bb61069 | 1087 | out_alloc: |
f2ecc5e4 CH |
1088 | *IO_agbp = NULL; |
1089 | return xfs_dialloc_ag(tp, agbp, parent, inop); | |
be60fe54 CH |
1090 | out_error: |
1091 | xfs_perag_put(pag); | |
1092 | return XFS_ERROR(error); | |
f2ecc5e4 CH |
1093 | } |
1094 | ||
1da177e4 LT |
1095 | /* |
1096 | * Free disk inode. Carefully avoids touching the incore inode, all | |
1097 | * manipulations incore are the caller's responsibility. | |
1098 | * The on-disk inode is not changed by this operation, only the | |
1099 | * btree (free inode mask) is changed. | |
1100 | */ | |
1101 | int | |
1102 | xfs_difree( | |
1103 | xfs_trans_t *tp, /* transaction pointer */ | |
1104 | xfs_ino_t inode, /* inode to be freed */ | |
1105 | xfs_bmap_free_t *flist, /* extents to free */ | |
1106 | int *delete, /* set if inode cluster was deleted */ | |
1107 | xfs_ino_t *first_ino) /* first inode in deleted cluster */ | |
1108 | { | |
1109 | /* REFERENCED */ | |
1110 | xfs_agblock_t agbno; /* block number containing inode */ | |
1111 | xfs_buf_t *agbp; /* buffer containing allocation group header */ | |
1112 | xfs_agino_t agino; /* inode number relative to allocation group */ | |
1113 | xfs_agnumber_t agno; /* allocation group number */ | |
1114 | xfs_agi_t *agi; /* allocation group header */ | |
1115 | xfs_btree_cur_t *cur; /* inode btree cursor */ | |
1116 | int error; /* error return value */ | |
1117 | int i; /* result code */ | |
1118 | int ilen; /* inodes in an inode cluster */ | |
1119 | xfs_mount_t *mp; /* mount structure for filesystem */ | |
1120 | int off; /* offset of inode in inode chunk */ | |
61a25848 | 1121 | xfs_inobt_rec_incore_t rec; /* btree record */ |
44b56e0a | 1122 | struct xfs_perag *pag; |
1da177e4 LT |
1123 | |
1124 | mp = tp->t_mountp; | |
1125 | ||
1126 | /* | |
1127 | * Break up inode number into its components. | |
1128 | */ | |
1129 | agno = XFS_INO_TO_AGNO(mp, inode); | |
1130 | if (agno >= mp->m_sb.sb_agcount) { | |
0b932ccc DC |
1131 | xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).", |
1132 | __func__, agno, mp->m_sb.sb_agcount); | |
1da177e4 LT |
1133 | ASSERT(0); |
1134 | return XFS_ERROR(EINVAL); | |
1135 | } | |
1136 | agino = XFS_INO_TO_AGINO(mp, inode); | |
1137 | if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
0b932ccc DC |
1138 | xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).", |
1139 | __func__, (unsigned long long)inode, | |
1140 | (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino)); | |
1da177e4 LT |
1141 | ASSERT(0); |
1142 | return XFS_ERROR(EINVAL); | |
1143 | } | |
1144 | agbno = XFS_AGINO_TO_AGBNO(mp, agino); | |
1145 | if (agbno >= mp->m_sb.sb_agblocks) { | |
0b932ccc DC |
1146 | xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).", |
1147 | __func__, agbno, mp->m_sb.sb_agblocks); | |
1da177e4 LT |
1148 | ASSERT(0); |
1149 | return XFS_ERROR(EINVAL); | |
1150 | } | |
1151 | /* | |
1152 | * Get the allocation group header. | |
1153 | */ | |
1da177e4 | 1154 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); |
1da177e4 | 1155 | if (error) { |
0b932ccc DC |
1156 | xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.", |
1157 | __func__, error); | |
1da177e4 LT |
1158 | return error; |
1159 | } | |
1160 | agi = XFS_BUF_TO_AGI(agbp); | |
69ef921b | 1161 | ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC)); |
16259e7d | 1162 | ASSERT(agbno < be32_to_cpu(agi->agi_length)); |
1da177e4 LT |
1163 | /* |
1164 | * Initialize the cursor. | |
1165 | */ | |
561f7d17 | 1166 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno); |
1da177e4 | 1167 | |
0b48db80 DC |
1168 | error = xfs_check_agi_freecount(cur, agi); |
1169 | if (error) | |
1170 | goto error0; | |
1171 | ||
1da177e4 LT |
1172 | /* |
1173 | * Look for the entry describing this inode. | |
1174 | */ | |
21875505 | 1175 | if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) { |
0b932ccc DC |
1176 | xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.", |
1177 | __func__, error); | |
1da177e4 LT |
1178 | goto error0; |
1179 | } | |
1180 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
2e287a73 CH |
1181 | error = xfs_inobt_get_rec(cur, &rec, &i); |
1182 | if (error) { | |
0b932ccc DC |
1183 | xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.", |
1184 | __func__, error); | |
1da177e4 LT |
1185 | goto error0; |
1186 | } | |
1187 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1188 | /* | |
1189 | * Get the offset in the inode chunk. | |
1190 | */ | |
1191 | off = agino - rec.ir_startino; | |
1192 | ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK); | |
0d87e656 | 1193 | ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off))); |
1da177e4 LT |
1194 | /* |
1195 | * Mark the inode free & increment the count. | |
1196 | */ | |
0d87e656 | 1197 | rec.ir_free |= XFS_INOBT_MASK(off); |
1da177e4 LT |
1198 | rec.ir_freecount++; |
1199 | ||
1200 | /* | |
c41564b5 | 1201 | * When an inode cluster is free, it becomes eligible for removal |
1da177e4 | 1202 | */ |
1bd960ee | 1203 | if (!(mp->m_flags & XFS_MOUNT_IKEEP) && |
1da177e4 LT |
1204 | (rec.ir_freecount == XFS_IALLOC_INODES(mp))) { |
1205 | ||
1206 | *delete = 1; | |
1207 | *first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino); | |
1208 | ||
1209 | /* | |
1210 | * Remove the inode cluster from the AGI B+Tree, adjust the | |
1211 | * AGI and Superblock inode counts, and mark the disk space | |
1212 | * to be freed when the transaction is committed. | |
1213 | */ | |
1214 | ilen = XFS_IALLOC_INODES(mp); | |
413d57c9 MS |
1215 | be32_add_cpu(&agi->agi_count, -ilen); |
1216 | be32_add_cpu(&agi->agi_freecount, -(ilen - 1)); | |
1da177e4 | 1217 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT); |
44b56e0a DC |
1218 | pag = xfs_perag_get(mp, agno); |
1219 | pag->pagi_freecount -= ilen - 1; | |
1220 | xfs_perag_put(pag); | |
1da177e4 LT |
1221 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen); |
1222 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1)); | |
1223 | ||
91cca5df | 1224 | if ((error = xfs_btree_delete(cur, &i))) { |
0b932ccc DC |
1225 | xfs_warn(mp, "%s: xfs_btree_delete returned error %d.", |
1226 | __func__, error); | |
1da177e4 LT |
1227 | goto error0; |
1228 | } | |
1229 | ||
1230 | xfs_bmap_add_free(XFS_AGB_TO_FSB(mp, | |
1231 | agno, XFS_INO_TO_AGBNO(mp,rec.ir_startino)), | |
1232 | XFS_IALLOC_BLOCKS(mp), flist, mp); | |
1233 | } else { | |
1234 | *delete = 0; | |
1235 | ||
afabc24a CH |
1236 | error = xfs_inobt_update(cur, &rec); |
1237 | if (error) { | |
0b932ccc DC |
1238 | xfs_warn(mp, "%s: xfs_inobt_update returned error %d.", |
1239 | __func__, error); | |
1da177e4 LT |
1240 | goto error0; |
1241 | } | |
afabc24a | 1242 | |
1da177e4 LT |
1243 | /* |
1244 | * Change the inode free counts and log the ag/sb changes. | |
1245 | */ | |
413d57c9 | 1246 | be32_add_cpu(&agi->agi_freecount, 1); |
1da177e4 | 1247 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); |
44b56e0a DC |
1248 | pag = xfs_perag_get(mp, agno); |
1249 | pag->pagi_freecount++; | |
1250 | xfs_perag_put(pag); | |
1da177e4 LT |
1251 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1); |
1252 | } | |
1253 | ||
0b48db80 DC |
1254 | error = xfs_check_agi_freecount(cur, agi); |
1255 | if (error) | |
1256 | goto error0; | |
1da177e4 | 1257 | |
1da177e4 LT |
1258 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
1259 | return 0; | |
1260 | ||
1261 | error0: | |
1262 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
1263 | return error; | |
1264 | } | |
1265 | ||
7124fe0a DC |
1266 | STATIC int |
1267 | xfs_imap_lookup( | |
1268 | struct xfs_mount *mp, | |
1269 | struct xfs_trans *tp, | |
1270 | xfs_agnumber_t agno, | |
1271 | xfs_agino_t agino, | |
1272 | xfs_agblock_t agbno, | |
1273 | xfs_agblock_t *chunk_agbno, | |
1274 | xfs_agblock_t *offset_agbno, | |
1275 | int flags) | |
1276 | { | |
1277 | struct xfs_inobt_rec_incore rec; | |
1278 | struct xfs_btree_cur *cur; | |
1279 | struct xfs_buf *agbp; | |
7124fe0a DC |
1280 | int error; |
1281 | int i; | |
1282 | ||
1283 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); | |
1284 | if (error) { | |
53487786 DC |
1285 | xfs_alert(mp, |
1286 | "%s: xfs_ialloc_read_agi() returned error %d, agno %d", | |
1287 | __func__, error, agno); | |
7124fe0a DC |
1288 | return error; |
1289 | } | |
1290 | ||
1291 | /* | |
4536f2ad DC |
1292 | * Lookup the inode record for the given agino. If the record cannot be |
1293 | * found, then it's an invalid inode number and we should abort. Once | |
1294 | * we have a record, we need to ensure it contains the inode number | |
1295 | * we are looking up. | |
7124fe0a DC |
1296 | */ |
1297 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno); | |
4536f2ad | 1298 | error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i); |
7124fe0a DC |
1299 | if (!error) { |
1300 | if (i) | |
1301 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
1302 | if (!error && i == 0) | |
1303 | error = EINVAL; | |
1304 | } | |
1305 | ||
1306 | xfs_trans_brelse(tp, agbp); | |
1307 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
1308 | if (error) | |
1309 | return error; | |
1310 | ||
4536f2ad DC |
1311 | /* check that the returned record contains the required inode */ |
1312 | if (rec.ir_startino > agino || | |
1313 | rec.ir_startino + XFS_IALLOC_INODES(mp) <= agino) | |
1314 | return EINVAL; | |
1315 | ||
7124fe0a | 1316 | /* for untrusted inodes check it is allocated first */ |
1920779e | 1317 | if ((flags & XFS_IGET_UNTRUSTED) && |
7124fe0a DC |
1318 | (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino))) |
1319 | return EINVAL; | |
1320 | ||
1321 | *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino); | |
1322 | *offset_agbno = agbno - *chunk_agbno; | |
1323 | return 0; | |
1324 | } | |
1325 | ||
1da177e4 | 1326 | /* |
94e1b69d | 1327 | * Return the location of the inode in imap, for mapping it into a buffer. |
1da177e4 | 1328 | */ |
1da177e4 | 1329 | int |
94e1b69d CH |
1330 | xfs_imap( |
1331 | xfs_mount_t *mp, /* file system mount structure */ | |
1332 | xfs_trans_t *tp, /* transaction pointer */ | |
1da177e4 | 1333 | xfs_ino_t ino, /* inode to locate */ |
94e1b69d CH |
1334 | struct xfs_imap *imap, /* location map structure */ |
1335 | uint flags) /* flags for inode btree lookup */ | |
1da177e4 LT |
1336 | { |
1337 | xfs_agblock_t agbno; /* block number of inode in the alloc group */ | |
1da177e4 LT |
1338 | xfs_agino_t agino; /* inode number within alloc group */ |
1339 | xfs_agnumber_t agno; /* allocation group number */ | |
1340 | int blks_per_cluster; /* num blocks per inode cluster */ | |
1341 | xfs_agblock_t chunk_agbno; /* first block in inode chunk */ | |
1da177e4 | 1342 | xfs_agblock_t cluster_agbno; /* first block in inode cluster */ |
1da177e4 | 1343 | int error; /* error code */ |
1da177e4 | 1344 | int offset; /* index of inode in its buffer */ |
836a94ad | 1345 | xfs_agblock_t offset_agbno; /* blks from chunk start to inode */ |
1da177e4 LT |
1346 | |
1347 | ASSERT(ino != NULLFSINO); | |
94e1b69d | 1348 | |
1da177e4 LT |
1349 | /* |
1350 | * Split up the inode number into its parts. | |
1351 | */ | |
1352 | agno = XFS_INO_TO_AGNO(mp, ino); | |
1353 | agino = XFS_INO_TO_AGINO(mp, ino); | |
1354 | agbno = XFS_AGINO_TO_AGBNO(mp, agino); | |
1355 | if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks || | |
1356 | ino != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
1357 | #ifdef DEBUG | |
1920779e DC |
1358 | /* |
1359 | * Don't output diagnostic information for untrusted inodes | |
1360 | * as they can be invalid without implying corruption. | |
1361 | */ | |
1362 | if (flags & XFS_IGET_UNTRUSTED) | |
4d1a2ed3 | 1363 | return XFS_ERROR(EINVAL); |
1da177e4 | 1364 | if (agno >= mp->m_sb.sb_agcount) { |
53487786 DC |
1365 | xfs_alert(mp, |
1366 | "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)", | |
1367 | __func__, agno, mp->m_sb.sb_agcount); | |
1da177e4 LT |
1368 | } |
1369 | if (agbno >= mp->m_sb.sb_agblocks) { | |
53487786 DC |
1370 | xfs_alert(mp, |
1371 | "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)", | |
1372 | __func__, (unsigned long long)agbno, | |
1373 | (unsigned long)mp->m_sb.sb_agblocks); | |
1da177e4 LT |
1374 | } |
1375 | if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
53487786 DC |
1376 | xfs_alert(mp, |
1377 | "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)", | |
1378 | __func__, ino, | |
1379 | XFS_AGINO_TO_INO(mp, agno, agino)); | |
1da177e4 | 1380 | } |
745b1f47 | 1381 | xfs_stack_trace(); |
1da177e4 LT |
1382 | #endif /* DEBUG */ |
1383 | return XFS_ERROR(EINVAL); | |
1384 | } | |
94e1b69d | 1385 | |
7124fe0a DC |
1386 | blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_blocklog; |
1387 | ||
1388 | /* | |
1389 | * For bulkstat and handle lookups, we have an untrusted inode number | |
1390 | * that we have to verify is valid. We cannot do this just by reading | |
1391 | * the inode buffer as it may have been unlinked and removed leaving | |
1392 | * inodes in stale state on disk. Hence we have to do a btree lookup | |
1393 | * in all cases where an untrusted inode number is passed. | |
1394 | */ | |
1920779e | 1395 | if (flags & XFS_IGET_UNTRUSTED) { |
7124fe0a DC |
1396 | error = xfs_imap_lookup(mp, tp, agno, agino, agbno, |
1397 | &chunk_agbno, &offset_agbno, flags); | |
1398 | if (error) | |
1399 | return error; | |
1400 | goto out_map; | |
1401 | } | |
1402 | ||
94e1b69d CH |
1403 | /* |
1404 | * If the inode cluster size is the same as the blocksize or | |
1405 | * smaller we get to the buffer by simple arithmetics. | |
1406 | */ | |
1407 | if (XFS_INODE_CLUSTER_SIZE(mp) <= mp->m_sb.sb_blocksize) { | |
1da177e4 LT |
1408 | offset = XFS_INO_TO_OFFSET(mp, ino); |
1409 | ASSERT(offset < mp->m_sb.sb_inopblock); | |
94e1b69d CH |
1410 | |
1411 | imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno); | |
1412 | imap->im_len = XFS_FSB_TO_BB(mp, 1); | |
1413 | imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog); | |
1da177e4 LT |
1414 | return 0; |
1415 | } | |
94e1b69d | 1416 | |
94e1b69d CH |
1417 | /* |
1418 | * If the inode chunks are aligned then use simple maths to | |
1419 | * find the location. Otherwise we have to do a btree | |
1420 | * lookup to find the location. | |
1421 | */ | |
1da177e4 LT |
1422 | if (mp->m_inoalign_mask) { |
1423 | offset_agbno = agbno & mp->m_inoalign_mask; | |
1424 | chunk_agbno = agbno - offset_agbno; | |
1425 | } else { | |
7124fe0a DC |
1426 | error = xfs_imap_lookup(mp, tp, agno, agino, agbno, |
1427 | &chunk_agbno, &offset_agbno, flags); | |
1da177e4 LT |
1428 | if (error) |
1429 | return error; | |
1da177e4 | 1430 | } |
94e1b69d | 1431 | |
7124fe0a | 1432 | out_map: |
1da177e4 LT |
1433 | ASSERT(agbno >= chunk_agbno); |
1434 | cluster_agbno = chunk_agbno + | |
1435 | ((offset_agbno / blks_per_cluster) * blks_per_cluster); | |
1436 | offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) + | |
1437 | XFS_INO_TO_OFFSET(mp, ino); | |
94e1b69d CH |
1438 | |
1439 | imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno); | |
1440 | imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster); | |
1441 | imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog); | |
1442 | ||
1443 | /* | |
1444 | * If the inode number maps to a block outside the bounds | |
1445 | * of the file system then return NULL rather than calling | |
1446 | * read_buf and panicing when we get an error from the | |
1447 | * driver. | |
1448 | */ | |
1449 | if ((imap->im_blkno + imap->im_len) > | |
1450 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) { | |
53487786 DC |
1451 | xfs_alert(mp, |
1452 | "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)", | |
1453 | __func__, (unsigned long long) imap->im_blkno, | |
94e1b69d CH |
1454 | (unsigned long long) imap->im_len, |
1455 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)); | |
1456 | return XFS_ERROR(EINVAL); | |
1457 | } | |
1da177e4 | 1458 | return 0; |
1da177e4 LT |
1459 | } |
1460 | ||
1461 | /* | |
1462 | * Compute and fill in value of m_in_maxlevels. | |
1463 | */ | |
1464 | void | |
1465 | xfs_ialloc_compute_maxlevels( | |
1466 | xfs_mount_t *mp) /* file system mount structure */ | |
1467 | { | |
1468 | int level; | |
1469 | uint maxblocks; | |
1470 | uint maxleafents; | |
1471 | int minleafrecs; | |
1472 | int minnoderecs; | |
1473 | ||
1474 | maxleafents = (1LL << XFS_INO_AGINO_BITS(mp)) >> | |
1475 | XFS_INODES_PER_CHUNK_LOG; | |
1476 | minleafrecs = mp->m_alloc_mnr[0]; | |
1477 | minnoderecs = mp->m_alloc_mnr[1]; | |
1478 | maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs; | |
1479 | for (level = 1; maxblocks > 1; level++) | |
1480 | maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs; | |
1481 | mp->m_in_maxlevels = level; | |
1482 | } | |
1483 | ||
1484 | /* | |
1485 | * Log specified fields for the ag hdr (inode section) | |
1486 | */ | |
1487 | void | |
1488 | xfs_ialloc_log_agi( | |
1489 | xfs_trans_t *tp, /* transaction pointer */ | |
1490 | xfs_buf_t *bp, /* allocation group header buffer */ | |
1491 | int fields) /* bitmask of fields to log */ | |
1492 | { | |
1493 | int first; /* first byte number */ | |
1494 | int last; /* last byte number */ | |
1495 | static const short offsets[] = { /* field starting offsets */ | |
1496 | /* keep in sync with bit definitions */ | |
1497 | offsetof(xfs_agi_t, agi_magicnum), | |
1498 | offsetof(xfs_agi_t, agi_versionnum), | |
1499 | offsetof(xfs_agi_t, agi_seqno), | |
1500 | offsetof(xfs_agi_t, agi_length), | |
1501 | offsetof(xfs_agi_t, agi_count), | |
1502 | offsetof(xfs_agi_t, agi_root), | |
1503 | offsetof(xfs_agi_t, agi_level), | |
1504 | offsetof(xfs_agi_t, agi_freecount), | |
1505 | offsetof(xfs_agi_t, agi_newino), | |
1506 | offsetof(xfs_agi_t, agi_dirino), | |
1507 | offsetof(xfs_agi_t, agi_unlinked), | |
1508 | sizeof(xfs_agi_t) | |
1509 | }; | |
1510 | #ifdef DEBUG | |
1511 | xfs_agi_t *agi; /* allocation group header */ | |
1512 | ||
1513 | agi = XFS_BUF_TO_AGI(bp); | |
69ef921b | 1514 | ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC)); |
1da177e4 LT |
1515 | #endif |
1516 | /* | |
1517 | * Compute byte offsets for the first and last fields. | |
1518 | */ | |
1519 | xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS, &first, &last); | |
1520 | /* | |
1521 | * Log the allocation group inode header buffer. | |
1522 | */ | |
61fe135c | 1523 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_AGI_BUF); |
1da177e4 LT |
1524 | xfs_trans_log_buf(tp, bp, first, last); |
1525 | } | |
1526 | ||
5e1be0fb CH |
1527 | #ifdef DEBUG |
1528 | STATIC void | |
1529 | xfs_check_agi_unlinked( | |
1530 | struct xfs_agi *agi) | |
1531 | { | |
1532 | int i; | |
1533 | ||
1534 | for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++) | |
1535 | ASSERT(agi->agi_unlinked[i]); | |
1536 | } | |
1537 | #else | |
1538 | #define xfs_check_agi_unlinked(agi) | |
1539 | #endif | |
1540 | ||
983d09ff | 1541 | static bool |
612cfbfe | 1542 | xfs_agi_verify( |
3702ce6e DC |
1543 | struct xfs_buf *bp) |
1544 | { | |
1545 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
1546 | struct xfs_agi *agi = XFS_BUF_TO_AGI(bp); | |
3702ce6e | 1547 | |
983d09ff DC |
1548 | if (xfs_sb_version_hascrc(&mp->m_sb) && |
1549 | !uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_uuid)) | |
1550 | return false; | |
3702ce6e DC |
1551 | /* |
1552 | * Validate the magic number of the agi block. | |
1553 | */ | |
983d09ff DC |
1554 | if (agi->agi_magicnum != cpu_to_be32(XFS_AGI_MAGIC)) |
1555 | return false; | |
1556 | if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum))) | |
1557 | return false; | |
3702ce6e DC |
1558 | |
1559 | /* | |
1560 | * during growfs operations, the perag is not fully initialised, | |
1561 | * so we can't use it for any useful checking. growfs ensures we can't | |
1562 | * use it by using uncached buffers that don't have the perag attached | |
1563 | * so we can detect and avoid this problem. | |
1564 | */ | |
983d09ff DC |
1565 | if (bp->b_pag && be32_to_cpu(agi->agi_seqno) != bp->b_pag->pag_agno) |
1566 | return false; | |
3702ce6e | 1567 | |
3702ce6e | 1568 | xfs_check_agi_unlinked(agi); |
983d09ff | 1569 | return true; |
612cfbfe DC |
1570 | } |
1571 | ||
1813dd64 DC |
1572 | static void |
1573 | xfs_agi_read_verify( | |
612cfbfe DC |
1574 | struct xfs_buf *bp) |
1575 | { | |
983d09ff DC |
1576 | struct xfs_mount *mp = bp->b_target->bt_mount; |
1577 | int agi_ok = 1; | |
1578 | ||
1579 | if (xfs_sb_version_hascrc(&mp->m_sb)) | |
1580 | agi_ok = xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length), | |
1581 | offsetof(struct xfs_agi, agi_crc)); | |
1582 | agi_ok = agi_ok && xfs_agi_verify(bp); | |
1583 | ||
1584 | if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IALLOC_READ_AGI, | |
1585 | XFS_RANDOM_IALLOC_READ_AGI))) { | |
1586 | XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr); | |
1587 | xfs_buf_ioerror(bp, EFSCORRUPTED); | |
1588 | } | |
612cfbfe DC |
1589 | } |
1590 | ||
b0f539de | 1591 | static void |
1813dd64 | 1592 | xfs_agi_write_verify( |
612cfbfe DC |
1593 | struct xfs_buf *bp) |
1594 | { | |
983d09ff DC |
1595 | struct xfs_mount *mp = bp->b_target->bt_mount; |
1596 | struct xfs_buf_log_item *bip = bp->b_fspriv; | |
1597 | ||
1598 | if (!xfs_agi_verify(bp)) { | |
1599 | XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr); | |
1600 | xfs_buf_ioerror(bp, EFSCORRUPTED); | |
1601 | return; | |
1602 | } | |
1603 | ||
1604 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
1605 | return; | |
1606 | ||
1607 | if (bip) | |
1608 | XFS_BUF_TO_AGI(bp)->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn); | |
1609 | xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length), | |
1610 | offsetof(struct xfs_agi, agi_crc)); | |
3702ce6e DC |
1611 | } |
1612 | ||
1813dd64 DC |
1613 | const struct xfs_buf_ops xfs_agi_buf_ops = { |
1614 | .verify_read = xfs_agi_read_verify, | |
1615 | .verify_write = xfs_agi_write_verify, | |
1616 | }; | |
1617 | ||
1da177e4 LT |
1618 | /* |
1619 | * Read in the allocation group header (inode allocation section) | |
1620 | */ | |
1621 | int | |
5e1be0fb CH |
1622 | xfs_read_agi( |
1623 | struct xfs_mount *mp, /* file system mount structure */ | |
1624 | struct xfs_trans *tp, /* transaction pointer */ | |
1625 | xfs_agnumber_t agno, /* allocation group number */ | |
1626 | struct xfs_buf **bpp) /* allocation group hdr buf */ | |
1da177e4 | 1627 | { |
5e1be0fb | 1628 | int error; |
1da177e4 LT |
1629 | |
1630 | ASSERT(agno != NULLAGNUMBER); | |
5e1be0fb CH |
1631 | |
1632 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, | |
1da177e4 | 1633 | XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)), |
1813dd64 | 1634 | XFS_FSS_TO_BB(mp, 1), 0, bpp, &xfs_agi_buf_ops); |
1da177e4 LT |
1635 | if (error) |
1636 | return error; | |
5e1be0fb | 1637 | |
5a52c2a5 | 1638 | ASSERT(!xfs_buf_geterror(*bpp)); |
38f23232 | 1639 | xfs_buf_set_ref(*bpp, XFS_AGI_REF); |
5e1be0fb CH |
1640 | return 0; |
1641 | } | |
1642 | ||
1643 | int | |
1644 | xfs_ialloc_read_agi( | |
1645 | struct xfs_mount *mp, /* file system mount structure */ | |
1646 | struct xfs_trans *tp, /* transaction pointer */ | |
1647 | xfs_agnumber_t agno, /* allocation group number */ | |
1648 | struct xfs_buf **bpp) /* allocation group hdr buf */ | |
1649 | { | |
1650 | struct xfs_agi *agi; /* allocation group header */ | |
1651 | struct xfs_perag *pag; /* per allocation group data */ | |
1652 | int error; | |
1653 | ||
1654 | error = xfs_read_agi(mp, tp, agno, bpp); | |
1655 | if (error) | |
1656 | return error; | |
1657 | ||
1658 | agi = XFS_BUF_TO_AGI(*bpp); | |
44b56e0a | 1659 | pag = xfs_perag_get(mp, agno); |
1da177e4 | 1660 | if (!pag->pagi_init) { |
16259e7d | 1661 | pag->pagi_freecount = be32_to_cpu(agi->agi_freecount); |
92821e2b | 1662 | pag->pagi_count = be32_to_cpu(agi->agi_count); |
1da177e4 | 1663 | pag->pagi_init = 1; |
1da177e4 | 1664 | } |
1da177e4 | 1665 | |
5e1be0fb CH |
1666 | /* |
1667 | * It's possible for these to be out of sync if | |
1668 | * we are in the middle of a forced shutdown. | |
1669 | */ | |
1670 | ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) || | |
1671 | XFS_FORCED_SHUTDOWN(mp)); | |
44b56e0a | 1672 | xfs_perag_put(pag); |
1da177e4 LT |
1673 | return 0; |
1674 | } | |
92821e2b DC |
1675 | |
1676 | /* | |
1677 | * Read in the agi to initialise the per-ag data in the mount structure | |
1678 | */ | |
1679 | int | |
1680 | xfs_ialloc_pagi_init( | |
1681 | xfs_mount_t *mp, /* file system mount structure */ | |
1682 | xfs_trans_t *tp, /* transaction pointer */ | |
1683 | xfs_agnumber_t agno) /* allocation group number */ | |
1684 | { | |
1685 | xfs_buf_t *bp = NULL; | |
1686 | int error; | |
1687 | ||
1688 | error = xfs_ialloc_read_agi(mp, tp, agno, &bp); | |
1689 | if (error) | |
1690 | return error; | |
1691 | if (bp) | |
1692 | xfs_trans_brelse(tp, bp); | |
1693 | return 0; | |
1694 | } |