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" |
d123031a | 42 | #include "xfs_trace.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 >= | |
0f49efd8 | 54 | XFS_B_TO_FSBT(args->mp, args->mp->m_inode_cluster_size)) |
75de2a91 DC |
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 | ||
0aa0a756 BF |
113 | /* |
114 | * Insert a single inobt record. Cursor must already point to desired location. | |
115 | */ | |
116 | STATIC int | |
117 | xfs_inobt_insert_rec( | |
118 | struct xfs_btree_cur *cur, | |
119 | __int32_t freecount, | |
120 | xfs_inofree_t free, | |
121 | int *stat) | |
122 | { | |
123 | cur->bc_rec.i.ir_freecount = freecount; | |
124 | cur->bc_rec.i.ir_free = free; | |
125 | return xfs_btree_insert(cur, stat); | |
126 | } | |
127 | ||
128 | /* | |
129 | * Insert records describing a newly allocated inode chunk into the inobt. | |
130 | */ | |
131 | STATIC int | |
132 | xfs_inobt_insert( | |
133 | struct xfs_mount *mp, | |
134 | struct xfs_trans *tp, | |
135 | struct xfs_buf *agbp, | |
136 | xfs_agino_t newino, | |
137 | xfs_agino_t newlen, | |
138 | xfs_btnum_t btnum) | |
139 | { | |
140 | struct xfs_btree_cur *cur; | |
141 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
142 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
143 | xfs_agino_t thisino; | |
144 | int i; | |
145 | int error; | |
146 | ||
147 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum); | |
148 | ||
149 | for (thisino = newino; | |
150 | thisino < newino + newlen; | |
151 | thisino += XFS_INODES_PER_CHUNK) { | |
152 | error = xfs_inobt_lookup(cur, thisino, XFS_LOOKUP_EQ, &i); | |
153 | if (error) { | |
154 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
155 | return error; | |
156 | } | |
157 | ASSERT(i == 0); | |
158 | ||
159 | error = xfs_inobt_insert_rec(cur, XFS_INODES_PER_CHUNK, | |
160 | XFS_INOBT_ALL_FREE, &i); | |
161 | if (error) { | |
162 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
163 | return error; | |
164 | } | |
165 | ASSERT(i == 1); | |
166 | } | |
167 | ||
168 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
169 | ||
170 | return 0; | |
171 | } | |
172 | ||
0b48db80 DC |
173 | /* |
174 | * Verify that the number of free inodes in the AGI is correct. | |
175 | */ | |
176 | #ifdef DEBUG | |
177 | STATIC int | |
178 | xfs_check_agi_freecount( | |
179 | struct xfs_btree_cur *cur, | |
180 | struct xfs_agi *agi) | |
181 | { | |
182 | if (cur->bc_nlevels == 1) { | |
183 | xfs_inobt_rec_incore_t rec; | |
184 | int freecount = 0; | |
185 | int error; | |
186 | int i; | |
187 | ||
21875505 | 188 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); |
0b48db80 DC |
189 | if (error) |
190 | return error; | |
191 | ||
192 | do { | |
193 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
194 | if (error) | |
195 | return error; | |
196 | ||
197 | if (i) { | |
198 | freecount += rec.ir_freecount; | |
199 | error = xfs_btree_increment(cur, 0, &i); | |
200 | if (error) | |
201 | return error; | |
202 | } | |
203 | } while (i == 1); | |
204 | ||
205 | if (!XFS_FORCED_SHUTDOWN(cur->bc_mp)) | |
206 | ASSERT(freecount == be32_to_cpu(agi->agi_freecount)); | |
207 | } | |
208 | return 0; | |
209 | } | |
210 | #else | |
211 | #define xfs_check_agi_freecount(cur, agi) 0 | |
212 | #endif | |
213 | ||
85c0b2ab | 214 | /* |
28c8e41a DC |
215 | * Initialise a new set of inodes. When called without a transaction context |
216 | * (e.g. from recovery) we initiate a delayed write of the inode buffers rather | |
217 | * than logging them (which in a transaction context puts them into the AIL | |
218 | * for writeback rather than the xfsbufd queue). | |
85c0b2ab | 219 | */ |
ddf6ad01 | 220 | int |
85c0b2ab DC |
221 | xfs_ialloc_inode_init( |
222 | struct xfs_mount *mp, | |
223 | struct xfs_trans *tp, | |
28c8e41a | 224 | struct list_head *buffer_list, |
85c0b2ab DC |
225 | xfs_agnumber_t agno, |
226 | xfs_agblock_t agbno, | |
227 | xfs_agblock_t length, | |
228 | unsigned int gen) | |
229 | { | |
230 | struct xfs_buf *fbuf; | |
231 | struct xfs_dinode *free; | |
6e0c7b8c | 232 | int nbufs, blks_per_cluster, inodes_per_cluster; |
85c0b2ab DC |
233 | int version; |
234 | int i, j; | |
235 | xfs_daddr_t d; | |
93848a99 | 236 | xfs_ino_t ino = 0; |
85c0b2ab DC |
237 | |
238 | /* | |
6e0c7b8c JL |
239 | * Loop over the new block(s), filling in the inodes. For small block |
240 | * sizes, manipulate the inodes in buffers which are multiples of the | |
241 | * blocks size. | |
85c0b2ab | 242 | */ |
6e0c7b8c JL |
243 | blks_per_cluster = xfs_icluster_size_fsb(mp); |
244 | inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog; | |
245 | nbufs = length / blks_per_cluster; | |
85c0b2ab DC |
246 | |
247 | /* | |
93848a99 CH |
248 | * Figure out what version number to use in the inodes we create. If |
249 | * the superblock version has caught up to the one that supports the new | |
250 | * inode format, then use the new inode version. Otherwise use the old | |
251 | * version so that old kernels will continue to be able to use the file | |
252 | * system. | |
253 | * | |
254 | * For v3 inodes, we also need to write the inode number into the inode, | |
255 | * so calculate the first inode number of the chunk here as | |
256 | * XFS_OFFBNO_TO_AGINO() only works within a filesystem block, not | |
257 | * across multiple filesystem blocks (such as a cluster) and so cannot | |
258 | * be used in the cluster buffer loop below. | |
259 | * | |
260 | * Further, because we are writing the inode directly into the buffer | |
261 | * and calculating a CRC on the entire inode, we have ot log the entire | |
262 | * inode so that the entire range the CRC covers is present in the log. | |
263 | * That means for v3 inode we log the entire buffer rather than just the | |
264 | * inode cores. | |
85c0b2ab | 265 | */ |
93848a99 CH |
266 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
267 | version = 3; | |
268 | ino = XFS_AGINO_TO_INO(mp, agno, | |
269 | XFS_OFFBNO_TO_AGINO(mp, agbno, 0)); | |
ddf6ad01 DC |
270 | |
271 | /* | |
272 | * log the initialisation that is about to take place as an | |
273 | * logical operation. This means the transaction does not | |
274 | * need to log the physical changes to the inode buffers as log | |
275 | * recovery will know what initialisation is actually needed. | |
276 | * Hence we only need to log the buffers as "ordered" buffers so | |
277 | * they track in the AIL as if they were physically logged. | |
278 | */ | |
279 | if (tp) | |
71783438 | 280 | xfs_icreate_log(tp, agno, agbno, mp->m_ialloc_inos, |
ddf6ad01 | 281 | mp->m_sb.sb_inodesize, length, gen); |
263997a6 | 282 | } else |
85c0b2ab | 283 | version = 2; |
85c0b2ab DC |
284 | |
285 | for (j = 0; j < nbufs; j++) { | |
286 | /* | |
287 | * Get the block. | |
288 | */ | |
289 | d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster)); | |
290 | fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, | |
7c4cebe8 DC |
291 | mp->m_bsize * blks_per_cluster, |
292 | XBF_UNMAPPED); | |
2a30f36d | 293 | if (!fbuf) |
2451337d | 294 | return -ENOMEM; |
ddf6ad01 DC |
295 | |
296 | /* Initialize the inode buffers and log them appropriately. */ | |
1813dd64 | 297 | fbuf->b_ops = &xfs_inode_buf_ops; |
93848a99 | 298 | xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length)); |
6e0c7b8c | 299 | for (i = 0; i < inodes_per_cluster; i++) { |
85c0b2ab | 300 | int ioffset = i << mp->m_sb.sb_inodelog; |
93848a99 | 301 | uint isize = xfs_dinode_size(version); |
85c0b2ab DC |
302 | |
303 | free = xfs_make_iptr(mp, fbuf, i); | |
304 | free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC); | |
305 | free->di_version = version; | |
306 | free->di_gen = cpu_to_be32(gen); | |
307 | free->di_next_unlinked = cpu_to_be32(NULLAGINO); | |
93848a99 CH |
308 | |
309 | if (version == 3) { | |
310 | free->di_ino = cpu_to_be64(ino); | |
311 | ino++; | |
312 | uuid_copy(&free->di_uuid, &mp->m_sb.sb_uuid); | |
313 | xfs_dinode_calc_crc(mp, free); | |
28c8e41a | 314 | } else if (tp) { |
93848a99 CH |
315 | /* just log the inode core */ |
316 | xfs_trans_log_buf(tp, fbuf, ioffset, | |
317 | ioffset + isize - 1); | |
318 | } | |
319 | } | |
28c8e41a DC |
320 | |
321 | if (tp) { | |
322 | /* | |
323 | * Mark the buffer as an inode allocation buffer so it | |
324 | * sticks in AIL at the point of this allocation | |
325 | * transaction. This ensures the they are on disk before | |
326 | * the tail of the log can be moved past this | |
327 | * transaction (i.e. by preventing relogging from moving | |
328 | * it forward in the log). | |
329 | */ | |
330 | xfs_trans_inode_alloc_buf(tp, fbuf); | |
331 | if (version == 3) { | |
ddf6ad01 DC |
332 | /* |
333 | * Mark the buffer as ordered so that they are | |
334 | * not physically logged in the transaction but | |
335 | * still tracked in the AIL as part of the | |
336 | * transaction and pin the log appropriately. | |
337 | */ | |
338 | xfs_trans_ordered_buf(tp, fbuf); | |
28c8e41a DC |
339 | xfs_trans_log_buf(tp, fbuf, 0, |
340 | BBTOB(fbuf->b_length) - 1); | |
341 | } | |
342 | } else { | |
343 | fbuf->b_flags |= XBF_DONE; | |
344 | xfs_buf_delwri_queue(fbuf, buffer_list); | |
345 | xfs_buf_relse(fbuf); | |
85c0b2ab | 346 | } |
85c0b2ab | 347 | } |
2a30f36d | 348 | return 0; |
85c0b2ab DC |
349 | } |
350 | ||
1da177e4 LT |
351 | /* |
352 | * Allocate new inodes in the allocation group specified by agbp. | |
353 | * Return 0 for success, else error code. | |
354 | */ | |
355 | STATIC int /* error code or 0 */ | |
356 | xfs_ialloc_ag_alloc( | |
357 | xfs_trans_t *tp, /* transaction pointer */ | |
358 | xfs_buf_t *agbp, /* alloc group buffer */ | |
359 | int *alloc) | |
360 | { | |
361 | xfs_agi_t *agi; /* allocation group header */ | |
362 | xfs_alloc_arg_t args; /* allocation argument structure */ | |
92821e2b | 363 | xfs_agnumber_t agno; |
1da177e4 | 364 | int error; |
1da177e4 LT |
365 | xfs_agino_t newino; /* new first inode's number */ |
366 | xfs_agino_t newlen; /* new number of inodes */ | |
3ccb8b5f | 367 | int isaligned = 0; /* inode allocation at stripe unit */ |
1da177e4 | 368 | /* boundary */ |
44b56e0a | 369 | struct xfs_perag *pag; |
1da177e4 | 370 | |
a0041684 | 371 | memset(&args, 0, sizeof(args)); |
1da177e4 LT |
372 | args.tp = tp; |
373 | args.mp = tp->t_mountp; | |
374 | ||
375 | /* | |
376 | * Locking will ensure that we don't have two callers in here | |
377 | * at one time. | |
378 | */ | |
71783438 | 379 | newlen = args.mp->m_ialloc_inos; |
1da177e4 LT |
380 | if (args.mp->m_maxicount && |
381 | args.mp->m_sb.sb_icount + newlen > args.mp->m_maxicount) | |
2451337d | 382 | return -ENOSPC; |
126cd105 | 383 | args.minlen = args.maxlen = args.mp->m_ialloc_blks; |
1da177e4 | 384 | /* |
3ccb8b5f GO |
385 | * First try to allocate inodes contiguous with the last-allocated |
386 | * chunk of inodes. If the filesystem is striped, this will fill | |
387 | * an entire stripe unit with inodes. | |
28c8e41a | 388 | */ |
1da177e4 | 389 | agi = XFS_BUF_TO_AGI(agbp); |
3ccb8b5f | 390 | newino = be32_to_cpu(agi->agi_newino); |
85c0b2ab | 391 | agno = be32_to_cpu(agi->agi_seqno); |
019ff2d5 | 392 | args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) + |
126cd105 | 393 | args.mp->m_ialloc_blks; |
019ff2d5 NS |
394 | if (likely(newino != NULLAGINO && |
395 | (args.agbno < be32_to_cpu(agi->agi_length)))) { | |
85c0b2ab | 396 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
3ccb8b5f | 397 | args.type = XFS_ALLOCTYPE_THIS_BNO; |
3ccb8b5f | 398 | args.prod = 1; |
75de2a91 | 399 | |
3ccb8b5f | 400 | /* |
75de2a91 DC |
401 | * We need to take into account alignment here to ensure that |
402 | * we don't modify the free list if we fail to have an exact | |
403 | * block. If we don't have an exact match, and every oher | |
404 | * attempt allocation attempt fails, we'll end up cancelling | |
405 | * a dirty transaction and shutting down. | |
406 | * | |
407 | * For an exact allocation, alignment must be 1, | |
408 | * however we need to take cluster alignment into account when | |
409 | * fixing up the freelist. Use the minalignslop field to | |
410 | * indicate that extra blocks might be required for alignment, | |
411 | * but not to use them in the actual exact allocation. | |
3ccb8b5f | 412 | */ |
75de2a91 DC |
413 | args.alignment = 1; |
414 | args.minalignslop = xfs_ialloc_cluster_alignment(&args) - 1; | |
415 | ||
416 | /* Allow space for the inode btree to split. */ | |
0d87e656 | 417 | args.minleft = args.mp->m_in_maxlevels - 1; |
3ccb8b5f GO |
418 | if ((error = xfs_alloc_vextent(&args))) |
419 | return error; | |
e480a723 BF |
420 | |
421 | /* | |
422 | * This request might have dirtied the transaction if the AG can | |
423 | * satisfy the request, but the exact block was not available. | |
424 | * If the allocation did fail, subsequent requests will relax | |
425 | * the exact agbno requirement and increase the alignment | |
426 | * instead. It is critical that the total size of the request | |
427 | * (len + alignment + slop) does not increase from this point | |
428 | * on, so reset minalignslop to ensure it is not included in | |
429 | * subsequent requests. | |
430 | */ | |
431 | args.minalignslop = 0; | |
3ccb8b5f GO |
432 | } else |
433 | args.fsbno = NULLFSBLOCK; | |
1da177e4 | 434 | |
3ccb8b5f GO |
435 | if (unlikely(args.fsbno == NULLFSBLOCK)) { |
436 | /* | |
437 | * Set the alignment for the allocation. | |
438 | * If stripe alignment is turned on then align at stripe unit | |
439 | * boundary. | |
019ff2d5 NS |
440 | * If the cluster size is smaller than a filesystem block |
441 | * then we're doing I/O for inodes in filesystem block size | |
3ccb8b5f GO |
442 | * pieces, so don't need alignment anyway. |
443 | */ | |
444 | isaligned = 0; | |
445 | if (args.mp->m_sinoalign) { | |
446 | ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN)); | |
447 | args.alignment = args.mp->m_dalign; | |
448 | isaligned = 1; | |
75de2a91 DC |
449 | } else |
450 | args.alignment = xfs_ialloc_cluster_alignment(&args); | |
3ccb8b5f GO |
451 | /* |
452 | * Need to figure out where to allocate the inode blocks. | |
453 | * Ideally they should be spaced out through the a.g. | |
454 | * For now, just allocate blocks up front. | |
455 | */ | |
456 | args.agbno = be32_to_cpu(agi->agi_root); | |
85c0b2ab | 457 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
3ccb8b5f GO |
458 | /* |
459 | * Allocate a fixed-size extent of inodes. | |
460 | */ | |
461 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
3ccb8b5f GO |
462 | args.prod = 1; |
463 | /* | |
464 | * Allow space for the inode btree to split. | |
465 | */ | |
0d87e656 | 466 | args.minleft = args.mp->m_in_maxlevels - 1; |
3ccb8b5f GO |
467 | if ((error = xfs_alloc_vextent(&args))) |
468 | return error; | |
469 | } | |
019ff2d5 | 470 | |
1da177e4 LT |
471 | /* |
472 | * If stripe alignment is turned on, then try again with cluster | |
473 | * alignment. | |
474 | */ | |
475 | if (isaligned && args.fsbno == NULLFSBLOCK) { | |
476 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
16259e7d | 477 | args.agbno = be32_to_cpu(agi->agi_root); |
85c0b2ab | 478 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
75de2a91 | 479 | args.alignment = xfs_ialloc_cluster_alignment(&args); |
1da177e4 LT |
480 | if ((error = xfs_alloc_vextent(&args))) |
481 | return error; | |
482 | } | |
483 | ||
484 | if (args.fsbno == NULLFSBLOCK) { | |
485 | *alloc = 0; | |
486 | return 0; | |
487 | } | |
488 | ASSERT(args.len == args.minlen); | |
1da177e4 | 489 | |
359346a9 | 490 | /* |
85c0b2ab DC |
491 | * Stamp and write the inode buffers. |
492 | * | |
359346a9 DC |
493 | * Seed the new inode cluster with a random generation number. This |
494 | * prevents short-term reuse of generation numbers if a chunk is | |
495 | * freed and then immediately reallocated. We use random numbers | |
496 | * rather than a linear progression to prevent the next generation | |
497 | * number from being easily guessable. | |
498 | */ | |
28c8e41a | 499 | error = xfs_ialloc_inode_init(args.mp, tp, NULL, agno, args.agbno, |
ecb3403d | 500 | args.len, prandom_u32()); |
d42f08f6 | 501 | |
2a30f36d CS |
502 | if (error) |
503 | return error; | |
85c0b2ab DC |
504 | /* |
505 | * Convert the results. | |
506 | */ | |
507 | newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0); | |
413d57c9 MS |
508 | be32_add_cpu(&agi->agi_count, newlen); |
509 | be32_add_cpu(&agi->agi_freecount, newlen); | |
44b56e0a DC |
510 | pag = xfs_perag_get(args.mp, agno); |
511 | pag->pagi_freecount += newlen; | |
512 | xfs_perag_put(pag); | |
16259e7d | 513 | agi->agi_newino = cpu_to_be32(newino); |
85c0b2ab | 514 | |
1da177e4 | 515 | /* |
0aa0a756 | 516 | * Insert records describing the new inode chunk into the btrees. |
1da177e4 | 517 | */ |
0aa0a756 BF |
518 | error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen, |
519 | XFS_BTNUM_INO); | |
520 | if (error) | |
521 | return error; | |
522 | ||
523 | if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) { | |
524 | error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen, | |
525 | XFS_BTNUM_FINO); | |
526 | if (error) | |
1da177e4 | 527 | return error; |
1da177e4 | 528 | } |
1da177e4 LT |
529 | /* |
530 | * Log allocation group header fields | |
531 | */ | |
532 | xfs_ialloc_log_agi(tp, agbp, | |
533 | XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO); | |
534 | /* | |
535 | * Modify/log superblock values for inode count and inode free count. | |
536 | */ | |
537 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen); | |
538 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen); | |
539 | *alloc = 1; | |
540 | return 0; | |
541 | } | |
542 | ||
b8f82a4a | 543 | STATIC xfs_agnumber_t |
1da177e4 LT |
544 | xfs_ialloc_next_ag( |
545 | xfs_mount_t *mp) | |
546 | { | |
547 | xfs_agnumber_t agno; | |
548 | ||
549 | spin_lock(&mp->m_agirotor_lock); | |
550 | agno = mp->m_agirotor; | |
8aea3ff4 | 551 | if (++mp->m_agirotor >= mp->m_maxagi) |
1da177e4 LT |
552 | mp->m_agirotor = 0; |
553 | spin_unlock(&mp->m_agirotor_lock); | |
554 | ||
555 | return agno; | |
556 | } | |
557 | ||
558 | /* | |
559 | * Select an allocation group to look for a free inode in, based on the parent | |
2f21ff1c | 560 | * inode and the mode. Return the allocation group buffer. |
1da177e4 | 561 | */ |
55d6af64 | 562 | STATIC xfs_agnumber_t |
1da177e4 LT |
563 | xfs_ialloc_ag_select( |
564 | xfs_trans_t *tp, /* transaction pointer */ | |
565 | xfs_ino_t parent, /* parent directory inode number */ | |
576b1d67 | 566 | umode_t mode, /* bits set to indicate file type */ |
1da177e4 LT |
567 | int okalloc) /* ok to allocate more space */ |
568 | { | |
1da177e4 LT |
569 | xfs_agnumber_t agcount; /* number of ag's in the filesystem */ |
570 | xfs_agnumber_t agno; /* current ag number */ | |
571 | int flags; /* alloc buffer locking flags */ | |
572 | xfs_extlen_t ineed; /* blocks needed for inode allocation */ | |
573 | xfs_extlen_t longest = 0; /* longest extent available */ | |
574 | xfs_mount_t *mp; /* mount point structure */ | |
575 | int needspace; /* file mode implies space allocated */ | |
576 | xfs_perag_t *pag; /* per allocation group data */ | |
577 | xfs_agnumber_t pagno; /* parent (starting) ag number */ | |
55d6af64 | 578 | int error; |
1da177e4 LT |
579 | |
580 | /* | |
581 | * Files of these types need at least one block if length > 0 | |
582 | * (and they won't fit in the inode, but that's hard to figure out). | |
583 | */ | |
584 | needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode); | |
585 | mp = tp->t_mountp; | |
586 | agcount = mp->m_maxagi; | |
587 | if (S_ISDIR(mode)) | |
588 | pagno = xfs_ialloc_next_ag(mp); | |
589 | else { | |
590 | pagno = XFS_INO_TO_AGNO(mp, parent); | |
591 | if (pagno >= agcount) | |
592 | pagno = 0; | |
593 | } | |
55d6af64 | 594 | |
1da177e4 | 595 | ASSERT(pagno < agcount); |
55d6af64 | 596 | |
1da177e4 LT |
597 | /* |
598 | * Loop through allocation groups, looking for one with a little | |
599 | * free space in it. Note we don't look for free inodes, exactly. | |
600 | * Instead, we include whether there is a need to allocate inodes | |
601 | * to mean that blocks must be allocated for them, | |
602 | * if none are currently free. | |
603 | */ | |
604 | agno = pagno; | |
605 | flags = XFS_ALLOC_FLAG_TRYLOCK; | |
1da177e4 | 606 | for (;;) { |
44b56e0a | 607 | pag = xfs_perag_get(mp, agno); |
55d6af64 CH |
608 | if (!pag->pagi_inodeok) { |
609 | xfs_ialloc_next_ag(mp); | |
610 | goto nextag; | |
611 | } | |
612 | ||
1da177e4 | 613 | if (!pag->pagi_init) { |
55d6af64 CH |
614 | error = xfs_ialloc_pagi_init(mp, tp, agno); |
615 | if (error) | |
1da177e4 | 616 | goto nextag; |
55d6af64 | 617 | } |
1da177e4 | 618 | |
55d6af64 CH |
619 | if (pag->pagi_freecount) { |
620 | xfs_perag_put(pag); | |
621 | return agno; | |
1da177e4 LT |
622 | } |
623 | ||
55d6af64 CH |
624 | if (!okalloc) |
625 | goto nextag; | |
626 | ||
627 | if (!pag->pagf_init) { | |
628 | error = xfs_alloc_pagf_init(mp, tp, agno, flags); | |
629 | if (error) | |
1da177e4 | 630 | goto nextag; |
1da177e4 | 631 | } |
55d6af64 CH |
632 | |
633 | /* | |
634 | * Is there enough free space for the file plus a block of | |
635 | * inodes? (if we need to allocate some)? | |
636 | */ | |
126cd105 | 637 | ineed = mp->m_ialloc_blks; |
55d6af64 CH |
638 | longest = pag->pagf_longest; |
639 | if (!longest) | |
640 | longest = pag->pagf_flcount > 0; | |
641 | ||
642 | if (pag->pagf_freeblks >= needspace + ineed && | |
643 | longest >= ineed) { | |
644 | xfs_perag_put(pag); | |
645 | return agno; | |
1da177e4 | 646 | } |
1da177e4 | 647 | nextag: |
44b56e0a | 648 | xfs_perag_put(pag); |
1da177e4 LT |
649 | /* |
650 | * No point in iterating over the rest, if we're shutting | |
651 | * down. | |
652 | */ | |
1c1c6ebc | 653 | if (XFS_FORCED_SHUTDOWN(mp)) |
55d6af64 | 654 | return NULLAGNUMBER; |
1da177e4 LT |
655 | agno++; |
656 | if (agno >= agcount) | |
657 | agno = 0; | |
658 | if (agno == pagno) { | |
1c1c6ebc | 659 | if (flags == 0) |
55d6af64 | 660 | return NULLAGNUMBER; |
1da177e4 LT |
661 | flags = 0; |
662 | } | |
663 | } | |
664 | } | |
665 | ||
4254b0bb CH |
666 | /* |
667 | * Try to retrieve the next record to the left/right from the current one. | |
668 | */ | |
669 | STATIC int | |
670 | xfs_ialloc_next_rec( | |
671 | struct xfs_btree_cur *cur, | |
672 | xfs_inobt_rec_incore_t *rec, | |
673 | int *done, | |
674 | int left) | |
675 | { | |
676 | int error; | |
677 | int i; | |
678 | ||
679 | if (left) | |
680 | error = xfs_btree_decrement(cur, 0, &i); | |
681 | else | |
682 | error = xfs_btree_increment(cur, 0, &i); | |
683 | ||
684 | if (error) | |
685 | return error; | |
686 | *done = !i; | |
687 | if (i) { | |
688 | error = xfs_inobt_get_rec(cur, rec, &i); | |
689 | if (error) | |
690 | return error; | |
691 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
692 | } | |
693 | ||
694 | return 0; | |
695 | } | |
696 | ||
bd169565 DC |
697 | STATIC int |
698 | xfs_ialloc_get_rec( | |
699 | struct xfs_btree_cur *cur, | |
700 | xfs_agino_t agino, | |
701 | xfs_inobt_rec_incore_t *rec, | |
43df2ee6 | 702 | int *done) |
bd169565 DC |
703 | { |
704 | int error; | |
705 | int i; | |
706 | ||
707 | error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i); | |
708 | if (error) | |
709 | return error; | |
710 | *done = !i; | |
711 | if (i) { | |
712 | error = xfs_inobt_get_rec(cur, rec, &i); | |
713 | if (error) | |
714 | return error; | |
715 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
716 | } | |
717 | ||
718 | return 0; | |
719 | } | |
0b48db80 | 720 | |
1da177e4 | 721 | /* |
6dd8638e | 722 | * Allocate an inode using the inobt-only algorithm. |
1da177e4 | 723 | */ |
f2ecc5e4 | 724 | STATIC int |
6dd8638e | 725 | xfs_dialloc_ag_inobt( |
f2ecc5e4 CH |
726 | struct xfs_trans *tp, |
727 | struct xfs_buf *agbp, | |
728 | xfs_ino_t parent, | |
729 | xfs_ino_t *inop) | |
1da177e4 | 730 | { |
f2ecc5e4 CH |
731 | struct xfs_mount *mp = tp->t_mountp; |
732 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
733 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
734 | xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent); | |
735 | xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent); | |
736 | struct xfs_perag *pag; | |
737 | struct xfs_btree_cur *cur, *tcur; | |
738 | struct xfs_inobt_rec_incore rec, trec; | |
739 | xfs_ino_t ino; | |
740 | int error; | |
741 | int offset; | |
742 | int i, j; | |
1da177e4 | 743 | |
44b56e0a | 744 | pag = xfs_perag_get(mp, agno); |
bd169565 | 745 | |
4bb61069 CH |
746 | ASSERT(pag->pagi_init); |
747 | ASSERT(pag->pagi_inodeok); | |
748 | ASSERT(pag->pagi_freecount > 0); | |
749 | ||
bd169565 | 750 | restart_pagno: |
57bd3dbe | 751 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
1da177e4 LT |
752 | /* |
753 | * If pagino is 0 (this is the root inode allocation) use newino. | |
754 | * This must work because we've just allocated some. | |
755 | */ | |
756 | if (!pagino) | |
16259e7d | 757 | pagino = be32_to_cpu(agi->agi_newino); |
1da177e4 | 758 | |
0b48db80 DC |
759 | error = xfs_check_agi_freecount(cur, agi); |
760 | if (error) | |
761 | goto error0; | |
1da177e4 | 762 | |
1da177e4 | 763 | /* |
4254b0bb | 764 | * If in the same AG as the parent, try to get near the parent. |
1da177e4 LT |
765 | */ |
766 | if (pagno == agno) { | |
4254b0bb CH |
767 | int doneleft; /* done, to the left */ |
768 | int doneright; /* done, to the right */ | |
bd169565 | 769 | int searchdistance = 10; |
4254b0bb | 770 | |
21875505 | 771 | error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i); |
4254b0bb | 772 | if (error) |
1da177e4 | 773 | goto error0; |
4254b0bb CH |
774 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); |
775 | ||
776 | error = xfs_inobt_get_rec(cur, &rec, &j); | |
777 | if (error) | |
778 | goto error0; | |
b121099d | 779 | XFS_WANT_CORRUPTED_GOTO(j == 1, error0); |
4254b0bb CH |
780 | |
781 | if (rec.ir_freecount > 0) { | |
1da177e4 LT |
782 | /* |
783 | * Found a free inode in the same chunk | |
4254b0bb | 784 | * as the parent, done. |
1da177e4 | 785 | */ |
4254b0bb | 786 | goto alloc_inode; |
1da177e4 | 787 | } |
4254b0bb CH |
788 | |
789 | ||
1da177e4 | 790 | /* |
4254b0bb | 791 | * In the same AG as parent, but parent's chunk is full. |
1da177e4 | 792 | */ |
1da177e4 | 793 | |
4254b0bb CH |
794 | /* duplicate the cursor, search left & right simultaneously */ |
795 | error = xfs_btree_dup_cursor(cur, &tcur); | |
796 | if (error) | |
797 | goto error0; | |
798 | ||
bd169565 DC |
799 | /* |
800 | * Skip to last blocks looked up if same parent inode. | |
801 | */ | |
802 | if (pagino != NULLAGINO && | |
803 | pag->pagl_pagino == pagino && | |
804 | pag->pagl_leftrec != NULLAGINO && | |
805 | pag->pagl_rightrec != NULLAGINO) { | |
806 | error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec, | |
43df2ee6 | 807 | &trec, &doneleft); |
bd169565 DC |
808 | if (error) |
809 | goto error1; | |
4254b0bb | 810 | |
bd169565 | 811 | error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec, |
43df2ee6 | 812 | &rec, &doneright); |
bd169565 DC |
813 | if (error) |
814 | goto error1; | |
815 | } else { | |
816 | /* search left with tcur, back up 1 record */ | |
817 | error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1); | |
818 | if (error) | |
819 | goto error1; | |
820 | ||
821 | /* search right with cur, go forward 1 record. */ | |
822 | error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0); | |
823 | if (error) | |
824 | goto error1; | |
825 | } | |
4254b0bb CH |
826 | |
827 | /* | |
828 | * Loop until we find an inode chunk with a free inode. | |
829 | */ | |
830 | while (!doneleft || !doneright) { | |
831 | int useleft; /* using left inode chunk this time */ | |
832 | ||
bd169565 DC |
833 | if (!--searchdistance) { |
834 | /* | |
835 | * Not in range - save last search | |
836 | * location and allocate a new inode | |
837 | */ | |
3b826386 | 838 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); |
bd169565 DC |
839 | pag->pagl_leftrec = trec.ir_startino; |
840 | pag->pagl_rightrec = rec.ir_startino; | |
841 | pag->pagl_pagino = pagino; | |
842 | goto newino; | |
843 | } | |
844 | ||
4254b0bb CH |
845 | /* figure out the closer block if both are valid. */ |
846 | if (!doneleft && !doneright) { | |
847 | useleft = pagino - | |
848 | (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) < | |
849 | rec.ir_startino - pagino; | |
850 | } else { | |
851 | useleft = !doneleft; | |
1da177e4 | 852 | } |
4254b0bb CH |
853 | |
854 | /* free inodes to the left? */ | |
855 | if (useleft && trec.ir_freecount) { | |
856 | rec = trec; | |
857 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
858 | cur = tcur; | |
bd169565 DC |
859 | |
860 | pag->pagl_leftrec = trec.ir_startino; | |
861 | pag->pagl_rightrec = rec.ir_startino; | |
862 | pag->pagl_pagino = pagino; | |
4254b0bb | 863 | goto alloc_inode; |
1da177e4 | 864 | } |
1da177e4 | 865 | |
4254b0bb CH |
866 | /* free inodes to the right? */ |
867 | if (!useleft && rec.ir_freecount) { | |
868 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
bd169565 DC |
869 | |
870 | pag->pagl_leftrec = trec.ir_startino; | |
871 | pag->pagl_rightrec = rec.ir_startino; | |
872 | pag->pagl_pagino = pagino; | |
4254b0bb | 873 | goto alloc_inode; |
1da177e4 | 874 | } |
4254b0bb CH |
875 | |
876 | /* get next record to check */ | |
877 | if (useleft) { | |
878 | error = xfs_ialloc_next_rec(tcur, &trec, | |
879 | &doneleft, 1); | |
880 | } else { | |
881 | error = xfs_ialloc_next_rec(cur, &rec, | |
882 | &doneright, 0); | |
883 | } | |
884 | if (error) | |
885 | goto error1; | |
1da177e4 | 886 | } |
bd169565 DC |
887 | |
888 | /* | |
889 | * We've reached the end of the btree. because | |
890 | * we are only searching a small chunk of the | |
891 | * btree each search, there is obviously free | |
892 | * inodes closer to the parent inode than we | |
893 | * are now. restart the search again. | |
894 | */ | |
895 | pag->pagl_pagino = NULLAGINO; | |
896 | pag->pagl_leftrec = NULLAGINO; | |
897 | pag->pagl_rightrec = NULLAGINO; | |
898 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
899 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
900 | goto restart_pagno; | |
1da177e4 | 901 | } |
4254b0bb | 902 | |
1da177e4 | 903 | /* |
4254b0bb | 904 | * In a different AG from the parent. |
1da177e4 LT |
905 | * See if the most recently allocated block has any free. |
906 | */ | |
bd169565 | 907 | newino: |
69ef921b | 908 | if (agi->agi_newino != cpu_to_be32(NULLAGINO)) { |
21875505 CH |
909 | error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino), |
910 | XFS_LOOKUP_EQ, &i); | |
4254b0bb | 911 | if (error) |
1da177e4 | 912 | goto error0; |
4254b0bb CH |
913 | |
914 | if (i == 1) { | |
915 | error = xfs_inobt_get_rec(cur, &rec, &j); | |
916 | if (error) | |
917 | goto error0; | |
918 | ||
919 | if (j == 1 && rec.ir_freecount > 0) { | |
920 | /* | |
921 | * The last chunk allocated in the group | |
922 | * still has a free inode. | |
923 | */ | |
924 | goto alloc_inode; | |
925 | } | |
1da177e4 | 926 | } |
bd169565 | 927 | } |
4254b0bb | 928 | |
bd169565 DC |
929 | /* |
930 | * None left in the last group, search the whole AG | |
931 | */ | |
932 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); | |
933 | if (error) | |
934 | goto error0; | |
935 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
936 | ||
937 | for (;;) { | |
938 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
939 | if (error) | |
940 | goto error0; | |
941 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
942 | if (rec.ir_freecount > 0) | |
943 | break; | |
944 | error = xfs_btree_increment(cur, 0, &i); | |
4254b0bb CH |
945 | if (error) |
946 | goto error0; | |
947 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1da177e4 | 948 | } |
4254b0bb CH |
949 | |
950 | alloc_inode: | |
824c3131 | 951 | offset = xfs_lowbit64(rec.ir_free); |
1da177e4 LT |
952 | ASSERT(offset >= 0); |
953 | ASSERT(offset < XFS_INODES_PER_CHUNK); | |
954 | ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) % | |
955 | XFS_INODES_PER_CHUNK) == 0); | |
956 | ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset); | |
0d87e656 | 957 | rec.ir_free &= ~XFS_INOBT_MASK(offset); |
1da177e4 | 958 | rec.ir_freecount--; |
afabc24a CH |
959 | error = xfs_inobt_update(cur, &rec); |
960 | if (error) | |
1da177e4 | 961 | goto error0; |
413d57c9 | 962 | be32_add_cpu(&agi->agi_freecount, -1); |
1da177e4 | 963 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); |
44b56e0a | 964 | pag->pagi_freecount--; |
1da177e4 | 965 | |
0b48db80 DC |
966 | error = xfs_check_agi_freecount(cur, agi); |
967 | if (error) | |
968 | goto error0; | |
969 | ||
1da177e4 LT |
970 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
971 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1); | |
44b56e0a | 972 | xfs_perag_put(pag); |
1da177e4 LT |
973 | *inop = ino; |
974 | return 0; | |
975 | error1: | |
976 | xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR); | |
977 | error0: | |
978 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
44b56e0a | 979 | xfs_perag_put(pag); |
1da177e4 LT |
980 | return error; |
981 | } | |
982 | ||
6dd8638e BF |
983 | /* |
984 | * Use the free inode btree to allocate an inode based on distance from the | |
985 | * parent. Note that the provided cursor may be deleted and replaced. | |
986 | */ | |
987 | STATIC int | |
988 | xfs_dialloc_ag_finobt_near( | |
989 | xfs_agino_t pagino, | |
990 | struct xfs_btree_cur **ocur, | |
991 | struct xfs_inobt_rec_incore *rec) | |
992 | { | |
993 | struct xfs_btree_cur *lcur = *ocur; /* left search cursor */ | |
994 | struct xfs_btree_cur *rcur; /* right search cursor */ | |
995 | struct xfs_inobt_rec_incore rrec; | |
996 | int error; | |
997 | int i, j; | |
998 | ||
999 | error = xfs_inobt_lookup(lcur, pagino, XFS_LOOKUP_LE, &i); | |
1000 | if (error) | |
1001 | return error; | |
1002 | ||
1003 | if (i == 1) { | |
1004 | error = xfs_inobt_get_rec(lcur, rec, &i); | |
1005 | if (error) | |
1006 | return error; | |
1007 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
1008 | ||
1009 | /* | |
1010 | * See if we've landed in the parent inode record. The finobt | |
1011 | * only tracks chunks with at least one free inode, so record | |
1012 | * existence is enough. | |
1013 | */ | |
1014 | if (pagino >= rec->ir_startino && | |
1015 | pagino < (rec->ir_startino + XFS_INODES_PER_CHUNK)) | |
1016 | return 0; | |
1017 | } | |
1018 | ||
1019 | error = xfs_btree_dup_cursor(lcur, &rcur); | |
1020 | if (error) | |
1021 | return error; | |
1022 | ||
1023 | error = xfs_inobt_lookup(rcur, pagino, XFS_LOOKUP_GE, &j); | |
1024 | if (error) | |
1025 | goto error_rcur; | |
1026 | if (j == 1) { | |
1027 | error = xfs_inobt_get_rec(rcur, &rrec, &j); | |
1028 | if (error) | |
1029 | goto error_rcur; | |
1030 | XFS_WANT_CORRUPTED_GOTO(j == 1, error_rcur); | |
1031 | } | |
1032 | ||
1033 | XFS_WANT_CORRUPTED_GOTO(i == 1 || j == 1, error_rcur); | |
1034 | if (i == 1 && j == 1) { | |
1035 | /* | |
1036 | * Both the left and right records are valid. Choose the closer | |
1037 | * inode chunk to the target. | |
1038 | */ | |
1039 | if ((pagino - rec->ir_startino + XFS_INODES_PER_CHUNK - 1) > | |
1040 | (rrec.ir_startino - pagino)) { | |
1041 | *rec = rrec; | |
1042 | xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR); | |
1043 | *ocur = rcur; | |
1044 | } else { | |
1045 | xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR); | |
1046 | } | |
1047 | } else if (j == 1) { | |
1048 | /* only the right record is valid */ | |
1049 | *rec = rrec; | |
1050 | xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR); | |
1051 | *ocur = rcur; | |
1052 | } else if (i == 1) { | |
1053 | /* only the left record is valid */ | |
1054 | xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR); | |
1055 | } | |
1056 | ||
1057 | return 0; | |
1058 | ||
1059 | error_rcur: | |
1060 | xfs_btree_del_cursor(rcur, XFS_BTREE_ERROR); | |
1061 | return error; | |
1062 | } | |
1063 | ||
1064 | /* | |
1065 | * Use the free inode btree to find a free inode based on a newino hint. If | |
1066 | * the hint is NULL, find the first free inode in the AG. | |
1067 | */ | |
1068 | STATIC int | |
1069 | xfs_dialloc_ag_finobt_newino( | |
1070 | struct xfs_agi *agi, | |
1071 | struct xfs_btree_cur *cur, | |
1072 | struct xfs_inobt_rec_incore *rec) | |
1073 | { | |
1074 | int error; | |
1075 | int i; | |
1076 | ||
1077 | if (agi->agi_newino != cpu_to_be32(NULLAGINO)) { | |
e68ed775 DC |
1078 | error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino), |
1079 | XFS_LOOKUP_EQ, &i); | |
6dd8638e BF |
1080 | if (error) |
1081 | return error; | |
1082 | if (i == 1) { | |
1083 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1084 | if (error) | |
1085 | return error; | |
1086 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
6dd8638e BF |
1087 | return 0; |
1088 | } | |
1089 | } | |
1090 | ||
1091 | /* | |
1092 | * Find the first inode available in the AG. | |
1093 | */ | |
1094 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); | |
1095 | if (error) | |
1096 | return error; | |
1097 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
1098 | ||
1099 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1100 | if (error) | |
1101 | return error; | |
1102 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
1103 | ||
1104 | return 0; | |
1105 | } | |
1106 | ||
1107 | /* | |
1108 | * Update the inobt based on a modification made to the finobt. Also ensure that | |
1109 | * the records from both trees are equivalent post-modification. | |
1110 | */ | |
1111 | STATIC int | |
1112 | xfs_dialloc_ag_update_inobt( | |
1113 | struct xfs_btree_cur *cur, /* inobt cursor */ | |
1114 | struct xfs_inobt_rec_incore *frec, /* finobt record */ | |
1115 | int offset) /* inode offset */ | |
1116 | { | |
1117 | struct xfs_inobt_rec_incore rec; | |
1118 | int error; | |
1119 | int i; | |
1120 | ||
1121 | error = xfs_inobt_lookup(cur, frec->ir_startino, XFS_LOOKUP_EQ, &i); | |
1122 | if (error) | |
1123 | return error; | |
1124 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
1125 | ||
1126 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
1127 | if (error) | |
1128 | return error; | |
1129 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
1130 | ASSERT((XFS_AGINO_TO_OFFSET(cur->bc_mp, rec.ir_startino) % | |
1131 | XFS_INODES_PER_CHUNK) == 0); | |
1132 | ||
1133 | rec.ir_free &= ~XFS_INOBT_MASK(offset); | |
1134 | rec.ir_freecount--; | |
1135 | ||
1136 | XFS_WANT_CORRUPTED_RETURN((rec.ir_free == frec->ir_free) && | |
1137 | (rec.ir_freecount == frec->ir_freecount)); | |
1138 | ||
1139 | error = xfs_inobt_update(cur, &rec); | |
1140 | if (error) | |
1141 | return error; | |
1142 | ||
1143 | return 0; | |
1144 | } | |
1145 | ||
1146 | /* | |
1147 | * Allocate an inode using the free inode btree, if available. Otherwise, fall | |
1148 | * back to the inobt search algorithm. | |
1149 | * | |
1150 | * The caller selected an AG for us, and made sure that free inodes are | |
1151 | * available. | |
1152 | */ | |
1153 | STATIC int | |
1154 | xfs_dialloc_ag( | |
1155 | struct xfs_trans *tp, | |
1156 | struct xfs_buf *agbp, | |
1157 | xfs_ino_t parent, | |
1158 | xfs_ino_t *inop) | |
1159 | { | |
1160 | struct xfs_mount *mp = tp->t_mountp; | |
1161 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
1162 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
1163 | xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent); | |
1164 | xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent); | |
1165 | struct xfs_perag *pag; | |
1166 | struct xfs_btree_cur *cur; /* finobt cursor */ | |
1167 | struct xfs_btree_cur *icur; /* inobt cursor */ | |
1168 | struct xfs_inobt_rec_incore rec; | |
1169 | xfs_ino_t ino; | |
1170 | int error; | |
1171 | int offset; | |
1172 | int i; | |
1173 | ||
1174 | if (!xfs_sb_version_hasfinobt(&mp->m_sb)) | |
1175 | return xfs_dialloc_ag_inobt(tp, agbp, parent, inop); | |
1176 | ||
1177 | pag = xfs_perag_get(mp, agno); | |
1178 | ||
1179 | /* | |
1180 | * If pagino is 0 (this is the root inode allocation) use newino. | |
1181 | * This must work because we've just allocated some. | |
1182 | */ | |
1183 | if (!pagino) | |
1184 | pagino = be32_to_cpu(agi->agi_newino); | |
1185 | ||
1186 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO); | |
1187 | ||
1188 | error = xfs_check_agi_freecount(cur, agi); | |
1189 | if (error) | |
1190 | goto error_cur; | |
1191 | ||
1192 | /* | |
1193 | * The search algorithm depends on whether we're in the same AG as the | |
1194 | * parent. If so, find the closest available inode to the parent. If | |
1195 | * not, consider the agi hint or find the first free inode in the AG. | |
1196 | */ | |
1197 | if (agno == pagno) | |
1198 | error = xfs_dialloc_ag_finobt_near(pagino, &cur, &rec); | |
1199 | else | |
1200 | error = xfs_dialloc_ag_finobt_newino(agi, cur, &rec); | |
1201 | if (error) | |
1202 | goto error_cur; | |
1203 | ||
1204 | offset = xfs_lowbit64(rec.ir_free); | |
1205 | ASSERT(offset >= 0); | |
1206 | ASSERT(offset < XFS_INODES_PER_CHUNK); | |
1207 | ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) % | |
1208 | XFS_INODES_PER_CHUNK) == 0); | |
1209 | ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset); | |
1210 | ||
1211 | /* | |
1212 | * Modify or remove the finobt record. | |
1213 | */ | |
1214 | rec.ir_free &= ~XFS_INOBT_MASK(offset); | |
1215 | rec.ir_freecount--; | |
1216 | if (rec.ir_freecount) | |
1217 | error = xfs_inobt_update(cur, &rec); | |
1218 | else | |
1219 | error = xfs_btree_delete(cur, &i); | |
1220 | if (error) | |
1221 | goto error_cur; | |
1222 | ||
1223 | /* | |
1224 | * The finobt has now been updated appropriately. We haven't updated the | |
1225 | * agi and superblock yet, so we can create an inobt cursor and validate | |
1226 | * the original freecount. If all is well, make the equivalent update to | |
1227 | * the inobt using the finobt record and offset information. | |
1228 | */ | |
1229 | icur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); | |
1230 | ||
1231 | error = xfs_check_agi_freecount(icur, agi); | |
1232 | if (error) | |
1233 | goto error_icur; | |
1234 | ||
1235 | error = xfs_dialloc_ag_update_inobt(icur, &rec, offset); | |
1236 | if (error) | |
1237 | goto error_icur; | |
1238 | ||
1239 | /* | |
1240 | * Both trees have now been updated. We must update the perag and | |
1241 | * superblock before we can check the freecount for each btree. | |
1242 | */ | |
1243 | be32_add_cpu(&agi->agi_freecount, -1); | |
1244 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); | |
1245 | pag->pagi_freecount--; | |
1246 | ||
1247 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1); | |
1248 | ||
1249 | error = xfs_check_agi_freecount(icur, agi); | |
1250 | if (error) | |
1251 | goto error_icur; | |
1252 | error = xfs_check_agi_freecount(cur, agi); | |
1253 | if (error) | |
1254 | goto error_icur; | |
1255 | ||
1256 | xfs_btree_del_cursor(icur, XFS_BTREE_NOERROR); | |
1257 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
1258 | xfs_perag_put(pag); | |
1259 | *inop = ino; | |
1260 | return 0; | |
1261 | ||
1262 | error_icur: | |
1263 | xfs_btree_del_cursor(icur, XFS_BTREE_ERROR); | |
1264 | error_cur: | |
1265 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
1266 | xfs_perag_put(pag); | |
1267 | return error; | |
1268 | } | |
1269 | ||
f2ecc5e4 CH |
1270 | /* |
1271 | * Allocate an inode on disk. | |
1272 | * | |
1273 | * Mode is used to tell whether the new inode will need space, and whether it | |
1274 | * is a directory. | |
1275 | * | |
1276 | * This function is designed to be called twice if it has to do an allocation | |
1277 | * to make more free inodes. On the first call, *IO_agbp should be set to NULL. | |
1278 | * If an inode is available without having to performn an allocation, an inode | |
cd856db6 CM |
1279 | * number is returned. In this case, *IO_agbp is set to NULL. If an allocation |
1280 | * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp. | |
1281 | * The caller should then commit the current transaction, allocate a | |
f2ecc5e4 CH |
1282 | * new transaction, and call xfs_dialloc() again, passing in the previous value |
1283 | * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI | |
1284 | * buffer is locked across the two calls, the second call is guaranteed to have | |
1285 | * a free inode available. | |
1286 | * | |
1287 | * Once we successfully pick an inode its number is returned and the on-disk | |
1288 | * data structures are updated. The inode itself is not read in, since doing so | |
1289 | * would break ordering constraints with xfs_reclaim. | |
1290 | */ | |
1291 | int | |
1292 | xfs_dialloc( | |
1293 | struct xfs_trans *tp, | |
1294 | xfs_ino_t parent, | |
1295 | umode_t mode, | |
1296 | int okalloc, | |
1297 | struct xfs_buf **IO_agbp, | |
f2ecc5e4 CH |
1298 | xfs_ino_t *inop) |
1299 | { | |
55d6af64 | 1300 | struct xfs_mount *mp = tp->t_mountp; |
f2ecc5e4 CH |
1301 | struct xfs_buf *agbp; |
1302 | xfs_agnumber_t agno; | |
f2ecc5e4 CH |
1303 | int error; |
1304 | int ialloced; | |
1305 | int noroom = 0; | |
be60fe54 | 1306 | xfs_agnumber_t start_agno; |
f2ecc5e4 CH |
1307 | struct xfs_perag *pag; |
1308 | ||
4bb61069 | 1309 | if (*IO_agbp) { |
f2ecc5e4 | 1310 | /* |
4bb61069 CH |
1311 | * If the caller passes in a pointer to the AGI buffer, |
1312 | * continue where we left off before. In this case, we | |
f2ecc5e4 CH |
1313 | * know that the allocation group has free inodes. |
1314 | */ | |
1315 | agbp = *IO_agbp; | |
4bb61069 | 1316 | goto out_alloc; |
f2ecc5e4 | 1317 | } |
4bb61069 CH |
1318 | |
1319 | /* | |
1320 | * We do not have an agbp, so select an initial allocation | |
1321 | * group for inode allocation. | |
1322 | */ | |
be60fe54 CH |
1323 | start_agno = xfs_ialloc_ag_select(tp, parent, mode, okalloc); |
1324 | if (start_agno == NULLAGNUMBER) { | |
4bb61069 CH |
1325 | *inop = NULLFSINO; |
1326 | return 0; | |
1327 | } | |
55d6af64 | 1328 | |
f2ecc5e4 CH |
1329 | /* |
1330 | * If we have already hit the ceiling of inode blocks then clear | |
1331 | * okalloc so we scan all available agi structures for a free | |
1332 | * inode. | |
1333 | */ | |
f2ecc5e4 | 1334 | if (mp->m_maxicount && |
71783438 | 1335 | mp->m_sb.sb_icount + mp->m_ialloc_inos > mp->m_maxicount) { |
f2ecc5e4 CH |
1336 | noroom = 1; |
1337 | okalloc = 0; | |
1338 | } | |
1339 | ||
1340 | /* | |
1341 | * Loop until we find an allocation group that either has free inodes | |
1342 | * or in which we can allocate some inodes. Iterate through the | |
1343 | * allocation groups upward, wrapping at the end. | |
1344 | */ | |
be60fe54 CH |
1345 | agno = start_agno; |
1346 | for (;;) { | |
1347 | pag = xfs_perag_get(mp, agno); | |
1348 | if (!pag->pagi_inodeok) { | |
1349 | xfs_ialloc_next_ag(mp); | |
1350 | goto nextag; | |
1351 | } | |
1352 | ||
1353 | if (!pag->pagi_init) { | |
1354 | error = xfs_ialloc_pagi_init(mp, tp, agno); | |
1355 | if (error) | |
1356 | goto out_error; | |
f2ecc5e4 | 1357 | } |
be60fe54 | 1358 | |
f2ecc5e4 | 1359 | /* |
be60fe54 | 1360 | * Do a first racy fast path check if this AG is usable. |
f2ecc5e4 | 1361 | */ |
be60fe54 CH |
1362 | if (!pag->pagi_freecount && !okalloc) |
1363 | goto nextag; | |
1364 | ||
c4982110 CH |
1365 | /* |
1366 | * Then read in the AGI buffer and recheck with the AGI buffer | |
1367 | * lock held. | |
1368 | */ | |
be60fe54 CH |
1369 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); |
1370 | if (error) | |
1371 | goto out_error; | |
1372 | ||
be60fe54 CH |
1373 | if (pag->pagi_freecount) { |
1374 | xfs_perag_put(pag); | |
1375 | goto out_alloc; | |
1376 | } | |
1377 | ||
c4982110 CH |
1378 | if (!okalloc) |
1379 | goto nextag_relse_buffer; | |
1380 | ||
be60fe54 CH |
1381 | |
1382 | error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced); | |
1383 | if (error) { | |
1384 | xfs_trans_brelse(tp, agbp); | |
1385 | ||
2451337d | 1386 | if (error != -ENOSPC) |
be60fe54 CH |
1387 | goto out_error; |
1388 | ||
1389 | xfs_perag_put(pag); | |
f2ecc5e4 | 1390 | *inop = NULLFSINO; |
be60fe54 | 1391 | return 0; |
f2ecc5e4 | 1392 | } |
be60fe54 CH |
1393 | |
1394 | if (ialloced) { | |
1395 | /* | |
1396 | * We successfully allocated some inodes, return | |
1397 | * the current context to the caller so that it | |
1398 | * can commit the current transaction and call | |
1399 | * us again where we left off. | |
1400 | */ | |
1401 | ASSERT(pag->pagi_freecount > 0); | |
f2ecc5e4 | 1402 | xfs_perag_put(pag); |
be60fe54 CH |
1403 | |
1404 | *IO_agbp = agbp; | |
1405 | *inop = NULLFSINO; | |
1406 | return 0; | |
f2ecc5e4 | 1407 | } |
be60fe54 | 1408 | |
c4982110 CH |
1409 | nextag_relse_buffer: |
1410 | xfs_trans_brelse(tp, agbp); | |
be60fe54 | 1411 | nextag: |
f2ecc5e4 | 1412 | xfs_perag_put(pag); |
be60fe54 CH |
1413 | if (++agno == mp->m_sb.sb_agcount) |
1414 | agno = 0; | |
1415 | if (agno == start_agno) { | |
1416 | *inop = NULLFSINO; | |
2451337d | 1417 | return noroom ? -ENOSPC : 0; |
be60fe54 | 1418 | } |
f2ecc5e4 CH |
1419 | } |
1420 | ||
4bb61069 | 1421 | out_alloc: |
f2ecc5e4 CH |
1422 | *IO_agbp = NULL; |
1423 | return xfs_dialloc_ag(tp, agbp, parent, inop); | |
be60fe54 CH |
1424 | out_error: |
1425 | xfs_perag_put(pag); | |
b474c7ae | 1426 | return error; |
f2ecc5e4 CH |
1427 | } |
1428 | ||
2b64ee5c BF |
1429 | STATIC int |
1430 | xfs_difree_inobt( | |
1431 | struct xfs_mount *mp, | |
1432 | struct xfs_trans *tp, | |
1433 | struct xfs_buf *agbp, | |
1434 | xfs_agino_t agino, | |
1435 | struct xfs_bmap_free *flist, | |
0d907a3b | 1436 | int *deleted, |
2b64ee5c BF |
1437 | xfs_ino_t *first_ino, |
1438 | struct xfs_inobt_rec_incore *orec) | |
1da177e4 | 1439 | { |
2b64ee5c BF |
1440 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); |
1441 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
1442 | struct xfs_perag *pag; | |
1443 | struct xfs_btree_cur *cur; | |
1444 | struct xfs_inobt_rec_incore rec; | |
1445 | int ilen; | |
1446 | int error; | |
1447 | int i; | |
1448 | int off; | |
1da177e4 | 1449 | |
69ef921b | 1450 | ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC)); |
2b64ee5c BF |
1451 | ASSERT(XFS_AGINO_TO_AGBNO(mp, agino) < be32_to_cpu(agi->agi_length)); |
1452 | ||
1da177e4 LT |
1453 | /* |
1454 | * Initialize the cursor. | |
1455 | */ | |
57bd3dbe | 1456 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
1da177e4 | 1457 | |
0b48db80 DC |
1458 | error = xfs_check_agi_freecount(cur, agi); |
1459 | if (error) | |
1460 | goto error0; | |
1461 | ||
1da177e4 LT |
1462 | /* |
1463 | * Look for the entry describing this inode. | |
1464 | */ | |
21875505 | 1465 | if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) { |
0b932ccc DC |
1466 | xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.", |
1467 | __func__, error); | |
1da177e4 LT |
1468 | goto error0; |
1469 | } | |
1470 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
2e287a73 CH |
1471 | error = xfs_inobt_get_rec(cur, &rec, &i); |
1472 | if (error) { | |
0b932ccc DC |
1473 | xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.", |
1474 | __func__, error); | |
1da177e4 LT |
1475 | goto error0; |
1476 | } | |
1477 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1478 | /* | |
1479 | * Get the offset in the inode chunk. | |
1480 | */ | |
1481 | off = agino - rec.ir_startino; | |
1482 | ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK); | |
0d87e656 | 1483 | ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off))); |
1da177e4 LT |
1484 | /* |
1485 | * Mark the inode free & increment the count. | |
1486 | */ | |
0d87e656 | 1487 | rec.ir_free |= XFS_INOBT_MASK(off); |
1da177e4 LT |
1488 | rec.ir_freecount++; |
1489 | ||
1490 | /* | |
c41564b5 | 1491 | * When an inode cluster is free, it becomes eligible for removal |
1da177e4 | 1492 | */ |
1bd960ee | 1493 | if (!(mp->m_flags & XFS_MOUNT_IKEEP) && |
71783438 | 1494 | (rec.ir_freecount == mp->m_ialloc_inos)) { |
1da177e4 | 1495 | |
376c2f3a | 1496 | *deleted = 1; |
1da177e4 LT |
1497 | *first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino); |
1498 | ||
1499 | /* | |
1500 | * Remove the inode cluster from the AGI B+Tree, adjust the | |
1501 | * AGI and Superblock inode counts, and mark the disk space | |
1502 | * to be freed when the transaction is committed. | |
1503 | */ | |
71783438 | 1504 | ilen = mp->m_ialloc_inos; |
413d57c9 MS |
1505 | be32_add_cpu(&agi->agi_count, -ilen); |
1506 | be32_add_cpu(&agi->agi_freecount, -(ilen - 1)); | |
1da177e4 | 1507 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT); |
44b56e0a DC |
1508 | pag = xfs_perag_get(mp, agno); |
1509 | pag->pagi_freecount -= ilen - 1; | |
1510 | xfs_perag_put(pag); | |
1da177e4 LT |
1511 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen); |
1512 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1)); | |
1513 | ||
91cca5df | 1514 | if ((error = xfs_btree_delete(cur, &i))) { |
0b932ccc DC |
1515 | xfs_warn(mp, "%s: xfs_btree_delete returned error %d.", |
1516 | __func__, error); | |
1da177e4 LT |
1517 | goto error0; |
1518 | } | |
1519 | ||
126cd105 JL |
1520 | xfs_bmap_add_free(XFS_AGB_TO_FSB(mp, agno, |
1521 | XFS_AGINO_TO_AGBNO(mp, rec.ir_startino)), | |
1522 | mp->m_ialloc_blks, flist, mp); | |
1da177e4 | 1523 | } else { |
376c2f3a | 1524 | *deleted = 0; |
1da177e4 | 1525 | |
afabc24a CH |
1526 | error = xfs_inobt_update(cur, &rec); |
1527 | if (error) { | |
0b932ccc DC |
1528 | xfs_warn(mp, "%s: xfs_inobt_update returned error %d.", |
1529 | __func__, error); | |
1da177e4 LT |
1530 | goto error0; |
1531 | } | |
afabc24a | 1532 | |
1da177e4 LT |
1533 | /* |
1534 | * Change the inode free counts and log the ag/sb changes. | |
1535 | */ | |
413d57c9 | 1536 | be32_add_cpu(&agi->agi_freecount, 1); |
1da177e4 | 1537 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); |
44b56e0a DC |
1538 | pag = xfs_perag_get(mp, agno); |
1539 | pag->pagi_freecount++; | |
1540 | xfs_perag_put(pag); | |
1da177e4 LT |
1541 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1); |
1542 | } | |
1543 | ||
0b48db80 DC |
1544 | error = xfs_check_agi_freecount(cur, agi); |
1545 | if (error) | |
1546 | goto error0; | |
1da177e4 | 1547 | |
2b64ee5c | 1548 | *orec = rec; |
1da177e4 LT |
1549 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
1550 | return 0; | |
1551 | ||
1552 | error0: | |
1553 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
1554 | return error; | |
1555 | } | |
1556 | ||
3efa4ffd BF |
1557 | /* |
1558 | * Free an inode in the free inode btree. | |
1559 | */ | |
1560 | STATIC int | |
1561 | xfs_difree_finobt( | |
1562 | struct xfs_mount *mp, | |
1563 | struct xfs_trans *tp, | |
1564 | struct xfs_buf *agbp, | |
1565 | xfs_agino_t agino, | |
1566 | struct xfs_inobt_rec_incore *ibtrec) /* inobt record */ | |
1567 | { | |
1568 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
1569 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
1570 | struct xfs_btree_cur *cur; | |
1571 | struct xfs_inobt_rec_incore rec; | |
1572 | int offset = agino - ibtrec->ir_startino; | |
1573 | int error; | |
1574 | int i; | |
1575 | ||
1576 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO); | |
1577 | ||
1578 | error = xfs_inobt_lookup(cur, ibtrec->ir_startino, XFS_LOOKUP_EQ, &i); | |
1579 | if (error) | |
1580 | goto error; | |
1581 | if (i == 0) { | |
1582 | /* | |
1583 | * If the record does not exist in the finobt, we must have just | |
1584 | * freed an inode in a previously fully allocated chunk. If not, | |
1585 | * something is out of sync. | |
1586 | */ | |
1587 | XFS_WANT_CORRUPTED_GOTO(ibtrec->ir_freecount == 1, error); | |
1588 | ||
1589 | error = xfs_inobt_insert_rec(cur, ibtrec->ir_freecount, | |
1590 | ibtrec->ir_free, &i); | |
1591 | if (error) | |
1592 | goto error; | |
1593 | ASSERT(i == 1); | |
1594 | ||
1595 | goto out; | |
1596 | } | |
1597 | ||
1598 | /* | |
1599 | * Read and update the existing record. We could just copy the ibtrec | |
1600 | * across here, but that would defeat the purpose of having redundant | |
1601 | * metadata. By making the modifications independently, we can catch | |
1602 | * corruptions that we wouldn't see if we just copied from one record | |
1603 | * to another. | |
1604 | */ | |
1605 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
1606 | if (error) | |
1607 | goto error; | |
1608 | XFS_WANT_CORRUPTED_GOTO(i == 1, error); | |
1609 | ||
1610 | rec.ir_free |= XFS_INOBT_MASK(offset); | |
1611 | rec.ir_freecount++; | |
1612 | ||
1613 | XFS_WANT_CORRUPTED_GOTO((rec.ir_free == ibtrec->ir_free) && | |
1614 | (rec.ir_freecount == ibtrec->ir_freecount), | |
1615 | error); | |
1616 | ||
1617 | /* | |
1618 | * The content of inobt records should always match between the inobt | |
1619 | * and finobt. The lifecycle of records in the finobt is different from | |
1620 | * the inobt in that the finobt only tracks records with at least one | |
1621 | * free inode. Hence, if all of the inodes are free and we aren't | |
1622 | * keeping inode chunks permanently on disk, remove the record. | |
1623 | * Otherwise, update the record with the new information. | |
1624 | */ | |
1625 | if (rec.ir_freecount == mp->m_ialloc_inos && | |
1626 | !(mp->m_flags & XFS_MOUNT_IKEEP)) { | |
1627 | error = xfs_btree_delete(cur, &i); | |
1628 | if (error) | |
1629 | goto error; | |
1630 | ASSERT(i == 1); | |
1631 | } else { | |
1632 | error = xfs_inobt_update(cur, &rec); | |
1633 | if (error) | |
1634 | goto error; | |
1635 | } | |
1636 | ||
1637 | out: | |
1638 | error = xfs_check_agi_freecount(cur, agi); | |
1639 | if (error) | |
1640 | goto error; | |
1641 | ||
1642 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
1643 | return 0; | |
1644 | ||
1645 | error: | |
1646 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
1647 | return error; | |
1648 | } | |
1649 | ||
2b64ee5c BF |
1650 | /* |
1651 | * Free disk inode. Carefully avoids touching the incore inode, all | |
1652 | * manipulations incore are the caller's responsibility. | |
1653 | * The on-disk inode is not changed by this operation, only the | |
1654 | * btree (free inode mask) is changed. | |
1655 | */ | |
1656 | int | |
1657 | xfs_difree( | |
1658 | struct xfs_trans *tp, /* transaction pointer */ | |
1659 | xfs_ino_t inode, /* inode to be freed */ | |
1660 | struct xfs_bmap_free *flist, /* extents to free */ | |
0d907a3b | 1661 | int *deleted,/* set if inode cluster was deleted */ |
2b64ee5c BF |
1662 | xfs_ino_t *first_ino)/* first inode in deleted cluster */ |
1663 | { | |
1664 | /* REFERENCED */ | |
1665 | xfs_agblock_t agbno; /* block number containing inode */ | |
1666 | struct xfs_buf *agbp; /* buffer for allocation group header */ | |
1667 | xfs_agino_t agino; /* allocation group inode number */ | |
1668 | xfs_agnumber_t agno; /* allocation group number */ | |
1669 | int error; /* error return value */ | |
1670 | struct xfs_mount *mp; /* mount structure for filesystem */ | |
1671 | struct xfs_inobt_rec_incore rec;/* btree record */ | |
1672 | ||
1673 | mp = tp->t_mountp; | |
1674 | ||
1675 | /* | |
1676 | * Break up inode number into its components. | |
1677 | */ | |
1678 | agno = XFS_INO_TO_AGNO(mp, inode); | |
1679 | if (agno >= mp->m_sb.sb_agcount) { | |
1680 | xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).", | |
1681 | __func__, agno, mp->m_sb.sb_agcount); | |
1682 | ASSERT(0); | |
2451337d | 1683 | return -EINVAL; |
2b64ee5c BF |
1684 | } |
1685 | agino = XFS_INO_TO_AGINO(mp, inode); | |
1686 | if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
1687 | xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).", | |
1688 | __func__, (unsigned long long)inode, | |
1689 | (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino)); | |
1690 | ASSERT(0); | |
2451337d | 1691 | return -EINVAL; |
2b64ee5c BF |
1692 | } |
1693 | agbno = XFS_AGINO_TO_AGBNO(mp, agino); | |
1694 | if (agbno >= mp->m_sb.sb_agblocks) { | |
1695 | xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).", | |
1696 | __func__, agbno, mp->m_sb.sb_agblocks); | |
1697 | ASSERT(0); | |
2451337d | 1698 | return -EINVAL; |
2b64ee5c BF |
1699 | } |
1700 | /* | |
1701 | * Get the allocation group header. | |
1702 | */ | |
1703 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); | |
1704 | if (error) { | |
1705 | xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.", | |
1706 | __func__, error); | |
1707 | return error; | |
1708 | } | |
1709 | ||
1710 | /* | |
1711 | * Fix up the inode allocation btree. | |
1712 | */ | |
0d907a3b | 1713 | error = xfs_difree_inobt(mp, tp, agbp, agino, flist, deleted, first_ino, |
2b64ee5c BF |
1714 | &rec); |
1715 | if (error) | |
1716 | goto error0; | |
1717 | ||
3efa4ffd BF |
1718 | /* |
1719 | * Fix up the free inode btree. | |
1720 | */ | |
1721 | if (xfs_sb_version_hasfinobt(&mp->m_sb)) { | |
1722 | error = xfs_difree_finobt(mp, tp, agbp, agino, &rec); | |
1723 | if (error) | |
1724 | goto error0; | |
1725 | } | |
1726 | ||
2b64ee5c BF |
1727 | return 0; |
1728 | ||
1729 | error0: | |
1730 | return error; | |
1731 | } | |
1732 | ||
7124fe0a DC |
1733 | STATIC int |
1734 | xfs_imap_lookup( | |
1735 | struct xfs_mount *mp, | |
1736 | struct xfs_trans *tp, | |
1737 | xfs_agnumber_t agno, | |
1738 | xfs_agino_t agino, | |
1739 | xfs_agblock_t agbno, | |
1740 | xfs_agblock_t *chunk_agbno, | |
1741 | xfs_agblock_t *offset_agbno, | |
1742 | int flags) | |
1743 | { | |
1744 | struct xfs_inobt_rec_incore rec; | |
1745 | struct xfs_btree_cur *cur; | |
1746 | struct xfs_buf *agbp; | |
7124fe0a DC |
1747 | int error; |
1748 | int i; | |
1749 | ||
1750 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); | |
1751 | if (error) { | |
53487786 DC |
1752 | xfs_alert(mp, |
1753 | "%s: xfs_ialloc_read_agi() returned error %d, agno %d", | |
1754 | __func__, error, agno); | |
7124fe0a DC |
1755 | return error; |
1756 | } | |
1757 | ||
1758 | /* | |
4536f2ad DC |
1759 | * Lookup the inode record for the given agino. If the record cannot be |
1760 | * found, then it's an invalid inode number and we should abort. Once | |
1761 | * we have a record, we need to ensure it contains the inode number | |
1762 | * we are looking up. | |
7124fe0a | 1763 | */ |
57bd3dbe | 1764 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
4536f2ad | 1765 | error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i); |
7124fe0a DC |
1766 | if (!error) { |
1767 | if (i) | |
1768 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
1769 | if (!error && i == 0) | |
2451337d | 1770 | error = -EINVAL; |
7124fe0a DC |
1771 | } |
1772 | ||
1773 | xfs_trans_brelse(tp, agbp); | |
1774 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
1775 | if (error) | |
1776 | return error; | |
1777 | ||
4536f2ad DC |
1778 | /* check that the returned record contains the required inode */ |
1779 | if (rec.ir_startino > agino || | |
71783438 | 1780 | rec.ir_startino + mp->m_ialloc_inos <= agino) |
2451337d | 1781 | return -EINVAL; |
4536f2ad | 1782 | |
7124fe0a | 1783 | /* for untrusted inodes check it is allocated first */ |
1920779e | 1784 | if ((flags & XFS_IGET_UNTRUSTED) && |
7124fe0a | 1785 | (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino))) |
2451337d | 1786 | return -EINVAL; |
7124fe0a DC |
1787 | |
1788 | *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino); | |
1789 | *offset_agbno = agbno - *chunk_agbno; | |
1790 | return 0; | |
1791 | } | |
1792 | ||
1da177e4 | 1793 | /* |
94e1b69d | 1794 | * Return the location of the inode in imap, for mapping it into a buffer. |
1da177e4 | 1795 | */ |
1da177e4 | 1796 | int |
94e1b69d CH |
1797 | xfs_imap( |
1798 | xfs_mount_t *mp, /* file system mount structure */ | |
1799 | xfs_trans_t *tp, /* transaction pointer */ | |
1da177e4 | 1800 | xfs_ino_t ino, /* inode to locate */ |
94e1b69d CH |
1801 | struct xfs_imap *imap, /* location map structure */ |
1802 | uint flags) /* flags for inode btree lookup */ | |
1da177e4 LT |
1803 | { |
1804 | xfs_agblock_t agbno; /* block number of inode in the alloc group */ | |
1da177e4 LT |
1805 | xfs_agino_t agino; /* inode number within alloc group */ |
1806 | xfs_agnumber_t agno; /* allocation group number */ | |
1807 | int blks_per_cluster; /* num blocks per inode cluster */ | |
1808 | xfs_agblock_t chunk_agbno; /* first block in inode chunk */ | |
1da177e4 | 1809 | xfs_agblock_t cluster_agbno; /* first block in inode cluster */ |
1da177e4 | 1810 | int error; /* error code */ |
1da177e4 | 1811 | int offset; /* index of inode in its buffer */ |
836a94ad | 1812 | xfs_agblock_t offset_agbno; /* blks from chunk start to inode */ |
1da177e4 LT |
1813 | |
1814 | ASSERT(ino != NULLFSINO); | |
94e1b69d | 1815 | |
1da177e4 LT |
1816 | /* |
1817 | * Split up the inode number into its parts. | |
1818 | */ | |
1819 | agno = XFS_INO_TO_AGNO(mp, ino); | |
1820 | agino = XFS_INO_TO_AGINO(mp, ino); | |
1821 | agbno = XFS_AGINO_TO_AGBNO(mp, agino); | |
1822 | if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks || | |
1823 | ino != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
1824 | #ifdef DEBUG | |
1920779e DC |
1825 | /* |
1826 | * Don't output diagnostic information for untrusted inodes | |
1827 | * as they can be invalid without implying corruption. | |
1828 | */ | |
1829 | if (flags & XFS_IGET_UNTRUSTED) | |
2451337d | 1830 | return -EINVAL; |
1da177e4 | 1831 | if (agno >= mp->m_sb.sb_agcount) { |
53487786 DC |
1832 | xfs_alert(mp, |
1833 | "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)", | |
1834 | __func__, agno, mp->m_sb.sb_agcount); | |
1da177e4 LT |
1835 | } |
1836 | if (agbno >= mp->m_sb.sb_agblocks) { | |
53487786 DC |
1837 | xfs_alert(mp, |
1838 | "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)", | |
1839 | __func__, (unsigned long long)agbno, | |
1840 | (unsigned long)mp->m_sb.sb_agblocks); | |
1da177e4 LT |
1841 | } |
1842 | if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
53487786 DC |
1843 | xfs_alert(mp, |
1844 | "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)", | |
1845 | __func__, ino, | |
1846 | XFS_AGINO_TO_INO(mp, agno, agino)); | |
1da177e4 | 1847 | } |
745b1f47 | 1848 | xfs_stack_trace(); |
1da177e4 | 1849 | #endif /* DEBUG */ |
2451337d | 1850 | return -EINVAL; |
1da177e4 | 1851 | } |
94e1b69d | 1852 | |
f9e5abcf | 1853 | blks_per_cluster = xfs_icluster_size_fsb(mp); |
7124fe0a DC |
1854 | |
1855 | /* | |
1856 | * For bulkstat and handle lookups, we have an untrusted inode number | |
1857 | * that we have to verify is valid. We cannot do this just by reading | |
1858 | * the inode buffer as it may have been unlinked and removed leaving | |
1859 | * inodes in stale state on disk. Hence we have to do a btree lookup | |
1860 | * in all cases where an untrusted inode number is passed. | |
1861 | */ | |
1920779e | 1862 | if (flags & XFS_IGET_UNTRUSTED) { |
7124fe0a DC |
1863 | error = xfs_imap_lookup(mp, tp, agno, agino, agbno, |
1864 | &chunk_agbno, &offset_agbno, flags); | |
1865 | if (error) | |
1866 | return error; | |
1867 | goto out_map; | |
1868 | } | |
1869 | ||
94e1b69d CH |
1870 | /* |
1871 | * If the inode cluster size is the same as the blocksize or | |
1872 | * smaller we get to the buffer by simple arithmetics. | |
1873 | */ | |
f9e5abcf | 1874 | if (blks_per_cluster == 1) { |
1da177e4 LT |
1875 | offset = XFS_INO_TO_OFFSET(mp, ino); |
1876 | ASSERT(offset < mp->m_sb.sb_inopblock); | |
94e1b69d CH |
1877 | |
1878 | imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno); | |
1879 | imap->im_len = XFS_FSB_TO_BB(mp, 1); | |
1880 | imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog); | |
1da177e4 LT |
1881 | return 0; |
1882 | } | |
94e1b69d | 1883 | |
94e1b69d CH |
1884 | /* |
1885 | * If the inode chunks are aligned then use simple maths to | |
1886 | * find the location. Otherwise we have to do a btree | |
1887 | * lookup to find the location. | |
1888 | */ | |
1da177e4 LT |
1889 | if (mp->m_inoalign_mask) { |
1890 | offset_agbno = agbno & mp->m_inoalign_mask; | |
1891 | chunk_agbno = agbno - offset_agbno; | |
1892 | } else { | |
7124fe0a DC |
1893 | error = xfs_imap_lookup(mp, tp, agno, agino, agbno, |
1894 | &chunk_agbno, &offset_agbno, flags); | |
1da177e4 LT |
1895 | if (error) |
1896 | return error; | |
1da177e4 | 1897 | } |
94e1b69d | 1898 | |
7124fe0a | 1899 | out_map: |
1da177e4 LT |
1900 | ASSERT(agbno >= chunk_agbno); |
1901 | cluster_agbno = chunk_agbno + | |
1902 | ((offset_agbno / blks_per_cluster) * blks_per_cluster); | |
1903 | offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) + | |
1904 | XFS_INO_TO_OFFSET(mp, ino); | |
94e1b69d CH |
1905 | |
1906 | imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno); | |
1907 | imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster); | |
1908 | imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog); | |
1909 | ||
1910 | /* | |
1911 | * If the inode number maps to a block outside the bounds | |
1912 | * of the file system then return NULL rather than calling | |
1913 | * read_buf and panicing when we get an error from the | |
1914 | * driver. | |
1915 | */ | |
1916 | if ((imap->im_blkno + imap->im_len) > | |
1917 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) { | |
53487786 DC |
1918 | xfs_alert(mp, |
1919 | "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)", | |
1920 | __func__, (unsigned long long) imap->im_blkno, | |
94e1b69d CH |
1921 | (unsigned long long) imap->im_len, |
1922 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)); | |
2451337d | 1923 | return -EINVAL; |
94e1b69d | 1924 | } |
1da177e4 | 1925 | return 0; |
1da177e4 LT |
1926 | } |
1927 | ||
1928 | /* | |
1929 | * Compute and fill in value of m_in_maxlevels. | |
1930 | */ | |
1931 | void | |
1932 | xfs_ialloc_compute_maxlevels( | |
1933 | xfs_mount_t *mp) /* file system mount structure */ | |
1934 | { | |
1935 | int level; | |
1936 | uint maxblocks; | |
1937 | uint maxleafents; | |
1938 | int minleafrecs; | |
1939 | int minnoderecs; | |
1940 | ||
1941 | maxleafents = (1LL << XFS_INO_AGINO_BITS(mp)) >> | |
1942 | XFS_INODES_PER_CHUNK_LOG; | |
1943 | minleafrecs = mp->m_alloc_mnr[0]; | |
1944 | minnoderecs = mp->m_alloc_mnr[1]; | |
1945 | maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs; | |
1946 | for (level = 1; maxblocks > 1; level++) | |
1947 | maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs; | |
1948 | mp->m_in_maxlevels = level; | |
1949 | } | |
1950 | ||
1951 | /* | |
aafc3c24 BF |
1952 | * Log specified fields for the ag hdr (inode section). The growth of the agi |
1953 | * structure over time requires that we interpret the buffer as two logical | |
1954 | * regions delineated by the end of the unlinked list. This is due to the size | |
1955 | * of the hash table and its location in the middle of the agi. | |
1956 | * | |
1957 | * For example, a request to log a field before agi_unlinked and a field after | |
1958 | * agi_unlinked could cause us to log the entire hash table and use an excessive | |
1959 | * amount of log space. To avoid this behavior, log the region up through | |
1960 | * agi_unlinked in one call and the region after agi_unlinked through the end of | |
1961 | * the structure in another. | |
1da177e4 LT |
1962 | */ |
1963 | void | |
1964 | xfs_ialloc_log_agi( | |
1965 | xfs_trans_t *tp, /* transaction pointer */ | |
1966 | xfs_buf_t *bp, /* allocation group header buffer */ | |
1967 | int fields) /* bitmask of fields to log */ | |
1968 | { | |
1969 | int first; /* first byte number */ | |
1970 | int last; /* last byte number */ | |
1971 | static const short offsets[] = { /* field starting offsets */ | |
1972 | /* keep in sync with bit definitions */ | |
1973 | offsetof(xfs_agi_t, agi_magicnum), | |
1974 | offsetof(xfs_agi_t, agi_versionnum), | |
1975 | offsetof(xfs_agi_t, agi_seqno), | |
1976 | offsetof(xfs_agi_t, agi_length), | |
1977 | offsetof(xfs_agi_t, agi_count), | |
1978 | offsetof(xfs_agi_t, agi_root), | |
1979 | offsetof(xfs_agi_t, agi_level), | |
1980 | offsetof(xfs_agi_t, agi_freecount), | |
1981 | offsetof(xfs_agi_t, agi_newino), | |
1982 | offsetof(xfs_agi_t, agi_dirino), | |
1983 | offsetof(xfs_agi_t, agi_unlinked), | |
aafc3c24 BF |
1984 | offsetof(xfs_agi_t, agi_free_root), |
1985 | offsetof(xfs_agi_t, agi_free_level), | |
1da177e4 LT |
1986 | sizeof(xfs_agi_t) |
1987 | }; | |
1988 | #ifdef DEBUG | |
1989 | xfs_agi_t *agi; /* allocation group header */ | |
1990 | ||
1991 | agi = XFS_BUF_TO_AGI(bp); | |
69ef921b | 1992 | ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC)); |
1da177e4 | 1993 | #endif |
aafc3c24 BF |
1994 | |
1995 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_AGI_BUF); | |
1996 | ||
1da177e4 | 1997 | /* |
aafc3c24 BF |
1998 | * Compute byte offsets for the first and last fields in the first |
1999 | * region and log the agi buffer. This only logs up through | |
2000 | * agi_unlinked. | |
1da177e4 | 2001 | */ |
aafc3c24 BF |
2002 | if (fields & XFS_AGI_ALL_BITS_R1) { |
2003 | xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R1, | |
2004 | &first, &last); | |
2005 | xfs_trans_log_buf(tp, bp, first, last); | |
2006 | } | |
2007 | ||
1da177e4 | 2008 | /* |
aafc3c24 BF |
2009 | * Mask off the bits in the first region and calculate the first and |
2010 | * last field offsets for any bits in the second region. | |
1da177e4 | 2011 | */ |
aafc3c24 BF |
2012 | fields &= ~XFS_AGI_ALL_BITS_R1; |
2013 | if (fields) { | |
2014 | xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R2, | |
2015 | &first, &last); | |
2016 | xfs_trans_log_buf(tp, bp, first, last); | |
2017 | } | |
1da177e4 LT |
2018 | } |
2019 | ||
5e1be0fb CH |
2020 | #ifdef DEBUG |
2021 | STATIC void | |
2022 | xfs_check_agi_unlinked( | |
2023 | struct xfs_agi *agi) | |
2024 | { | |
2025 | int i; | |
2026 | ||
2027 | for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++) | |
2028 | ASSERT(agi->agi_unlinked[i]); | |
2029 | } | |
2030 | #else | |
2031 | #define xfs_check_agi_unlinked(agi) | |
2032 | #endif | |
2033 | ||
983d09ff | 2034 | static bool |
612cfbfe | 2035 | xfs_agi_verify( |
3702ce6e DC |
2036 | struct xfs_buf *bp) |
2037 | { | |
2038 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
2039 | struct xfs_agi *agi = XFS_BUF_TO_AGI(bp); | |
3702ce6e | 2040 | |
983d09ff DC |
2041 | if (xfs_sb_version_hascrc(&mp->m_sb) && |
2042 | !uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_uuid)) | |
2043 | return false; | |
3702ce6e DC |
2044 | /* |
2045 | * Validate the magic number of the agi block. | |
2046 | */ | |
983d09ff DC |
2047 | if (agi->agi_magicnum != cpu_to_be32(XFS_AGI_MAGIC)) |
2048 | return false; | |
2049 | if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum))) | |
2050 | return false; | |
3702ce6e | 2051 | |
e1b05723 ES |
2052 | if (be32_to_cpu(agi->agi_level) > XFS_BTREE_MAXLEVELS) |
2053 | return false; | |
3702ce6e DC |
2054 | /* |
2055 | * during growfs operations, the perag is not fully initialised, | |
2056 | * so we can't use it for any useful checking. growfs ensures we can't | |
2057 | * use it by using uncached buffers that don't have the perag attached | |
2058 | * so we can detect and avoid this problem. | |
2059 | */ | |
983d09ff DC |
2060 | if (bp->b_pag && be32_to_cpu(agi->agi_seqno) != bp->b_pag->pag_agno) |
2061 | return false; | |
3702ce6e | 2062 | |
3702ce6e | 2063 | xfs_check_agi_unlinked(agi); |
983d09ff | 2064 | return true; |
612cfbfe DC |
2065 | } |
2066 | ||
1813dd64 DC |
2067 | static void |
2068 | xfs_agi_read_verify( | |
612cfbfe DC |
2069 | struct xfs_buf *bp) |
2070 | { | |
983d09ff | 2071 | struct xfs_mount *mp = bp->b_target->bt_mount; |
983d09ff | 2072 | |
ce5028cf ES |
2073 | if (xfs_sb_version_hascrc(&mp->m_sb) && |
2074 | !xfs_buf_verify_cksum(bp, XFS_AGI_CRC_OFF)) | |
2451337d | 2075 | xfs_buf_ioerror(bp, -EFSBADCRC); |
ce5028cf ES |
2076 | else if (XFS_TEST_ERROR(!xfs_agi_verify(bp), mp, |
2077 | XFS_ERRTAG_IALLOC_READ_AGI, | |
2078 | XFS_RANDOM_IALLOC_READ_AGI)) | |
2451337d | 2079 | xfs_buf_ioerror(bp, -EFSCORRUPTED); |
ce5028cf ES |
2080 | |
2081 | if (bp->b_error) | |
2082 | xfs_verifier_error(bp); | |
612cfbfe DC |
2083 | } |
2084 | ||
b0f539de | 2085 | static void |
1813dd64 | 2086 | xfs_agi_write_verify( |
612cfbfe DC |
2087 | struct xfs_buf *bp) |
2088 | { | |
983d09ff DC |
2089 | struct xfs_mount *mp = bp->b_target->bt_mount; |
2090 | struct xfs_buf_log_item *bip = bp->b_fspriv; | |
2091 | ||
2092 | if (!xfs_agi_verify(bp)) { | |
2451337d | 2093 | xfs_buf_ioerror(bp, -EFSCORRUPTED); |
ce5028cf | 2094 | xfs_verifier_error(bp); |
983d09ff DC |
2095 | return; |
2096 | } | |
2097 | ||
2098 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
2099 | return; | |
2100 | ||
2101 | if (bip) | |
2102 | XFS_BUF_TO_AGI(bp)->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn); | |
f1dbcd7e | 2103 | xfs_buf_update_cksum(bp, XFS_AGI_CRC_OFF); |
3702ce6e DC |
2104 | } |
2105 | ||
1813dd64 DC |
2106 | const struct xfs_buf_ops xfs_agi_buf_ops = { |
2107 | .verify_read = xfs_agi_read_verify, | |
2108 | .verify_write = xfs_agi_write_verify, | |
2109 | }; | |
2110 | ||
1da177e4 LT |
2111 | /* |
2112 | * Read in the allocation group header (inode allocation section) | |
2113 | */ | |
2114 | int | |
5e1be0fb CH |
2115 | xfs_read_agi( |
2116 | struct xfs_mount *mp, /* file system mount structure */ | |
2117 | struct xfs_trans *tp, /* transaction pointer */ | |
2118 | xfs_agnumber_t agno, /* allocation group number */ | |
2119 | struct xfs_buf **bpp) /* allocation group hdr buf */ | |
1da177e4 | 2120 | { |
5e1be0fb | 2121 | int error; |
1da177e4 | 2122 | |
d123031a | 2123 | trace_xfs_read_agi(mp, agno); |
5e1be0fb | 2124 | |
d123031a | 2125 | ASSERT(agno != NULLAGNUMBER); |
5e1be0fb | 2126 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, |
1da177e4 | 2127 | XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)), |
1813dd64 | 2128 | XFS_FSS_TO_BB(mp, 1), 0, bpp, &xfs_agi_buf_ops); |
1da177e4 LT |
2129 | if (error) |
2130 | return error; | |
5e1be0fb | 2131 | |
38f23232 | 2132 | xfs_buf_set_ref(*bpp, XFS_AGI_REF); |
5e1be0fb CH |
2133 | return 0; |
2134 | } | |
2135 | ||
2136 | int | |
2137 | xfs_ialloc_read_agi( | |
2138 | struct xfs_mount *mp, /* file system mount structure */ | |
2139 | struct xfs_trans *tp, /* transaction pointer */ | |
2140 | xfs_agnumber_t agno, /* allocation group number */ | |
2141 | struct xfs_buf **bpp) /* allocation group hdr buf */ | |
2142 | { | |
2143 | struct xfs_agi *agi; /* allocation group header */ | |
2144 | struct xfs_perag *pag; /* per allocation group data */ | |
2145 | int error; | |
2146 | ||
d123031a DC |
2147 | trace_xfs_ialloc_read_agi(mp, agno); |
2148 | ||
5e1be0fb CH |
2149 | error = xfs_read_agi(mp, tp, agno, bpp); |
2150 | if (error) | |
2151 | return error; | |
2152 | ||
2153 | agi = XFS_BUF_TO_AGI(*bpp); | |
44b56e0a | 2154 | pag = xfs_perag_get(mp, agno); |
1da177e4 | 2155 | if (!pag->pagi_init) { |
16259e7d | 2156 | pag->pagi_freecount = be32_to_cpu(agi->agi_freecount); |
92821e2b | 2157 | pag->pagi_count = be32_to_cpu(agi->agi_count); |
1da177e4 | 2158 | pag->pagi_init = 1; |
1da177e4 | 2159 | } |
1da177e4 | 2160 | |
5e1be0fb CH |
2161 | /* |
2162 | * It's possible for these to be out of sync if | |
2163 | * we are in the middle of a forced shutdown. | |
2164 | */ | |
2165 | ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) || | |
2166 | XFS_FORCED_SHUTDOWN(mp)); | |
44b56e0a | 2167 | xfs_perag_put(pag); |
1da177e4 LT |
2168 | return 0; |
2169 | } | |
92821e2b DC |
2170 | |
2171 | /* | |
2172 | * Read in the agi to initialise the per-ag data in the mount structure | |
2173 | */ | |
2174 | int | |
2175 | xfs_ialloc_pagi_init( | |
2176 | xfs_mount_t *mp, /* file system mount structure */ | |
2177 | xfs_trans_t *tp, /* transaction pointer */ | |
2178 | xfs_agnumber_t agno) /* allocation group number */ | |
2179 | { | |
2180 | xfs_buf_t *bp = NULL; | |
2181 | int error; | |
2182 | ||
2183 | error = xfs_ialloc_read_agi(mp, tp, agno, &bp); | |
2184 | if (error) | |
2185 | return error; | |
2186 | if (bp) | |
2187 | xfs_trans_brelse(tp, bp); | |
2188 | return 0; | |
2189 | } |