Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_fs.h" |
70a9883c | 20 | #include "xfs_shared.h" |
239880ef DC |
21 | #include "xfs_format.h" |
22 | #include "xfs_log_format.h" | |
23 | #include "xfs_trans_resv.h" | |
a844f451 | 24 | #include "xfs_bit.h" |
1da177e4 | 25 | #include "xfs_sb.h" |
1da177e4 | 26 | #include "xfs_mount.h" |
1da177e4 | 27 | #include "xfs_inode.h" |
a844f451 NS |
28 | #include "xfs_btree.h" |
29 | #include "xfs_ialloc.h" | |
a4fbe6ab | 30 | #include "xfs_ialloc_btree.h" |
1da177e4 | 31 | #include "xfs_alloc.h" |
1da177e4 LT |
32 | #include "xfs_rtalloc.h" |
33 | #include "xfs_error.h" | |
34 | #include "xfs_bmap.h" | |
983d09ff | 35 | #include "xfs_cksum.h" |
239880ef | 36 | #include "xfs_trans.h" |
983d09ff | 37 | #include "xfs_buf_item.h" |
ddf6ad01 | 38 | #include "xfs_icreate_item.h" |
7bb85ef3 | 39 | #include "xfs_icache.h" |
d123031a | 40 | #include "xfs_trace.h" |
1da177e4 | 41 | |
1da177e4 LT |
42 | |
43 | /* | |
44 | * Allocation group level functions. | |
45 | */ | |
75de2a91 DC |
46 | static inline int |
47 | xfs_ialloc_cluster_alignment( | |
7a1df156 | 48 | struct xfs_mount *mp) |
75de2a91 | 49 | { |
7a1df156 DC |
50 | if (xfs_sb_version_hasalign(&mp->m_sb) && |
51 | mp->m_sb.sb_inoalignmt >= | |
52 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) | |
53 | return mp->m_sb.sb_inoalignmt; | |
75de2a91 DC |
54 | return 1; |
55 | } | |
1da177e4 | 56 | |
fe033cc8 | 57 | /* |
21875505 | 58 | * Lookup a record by ino in the btree given by cur. |
fe033cc8 | 59 | */ |
81e25176 | 60 | int /* error */ |
21875505 | 61 | xfs_inobt_lookup( |
fe033cc8 CH |
62 | struct xfs_btree_cur *cur, /* btree cursor */ |
63 | xfs_agino_t ino, /* starting inode of chunk */ | |
21875505 | 64 | xfs_lookup_t dir, /* <=, >=, == */ |
fe033cc8 CH |
65 | int *stat) /* success/failure */ |
66 | { | |
67 | cur->bc_rec.i.ir_startino = ino; | |
21875505 CH |
68 | cur->bc_rec.i.ir_freecount = 0; |
69 | cur->bc_rec.i.ir_free = 0; | |
70 | return xfs_btree_lookup(cur, dir, stat); | |
fe033cc8 CH |
71 | } |
72 | ||
278d0ca1 | 73 | /* |
afabc24a | 74 | * Update the record referred to by cur to the value given. |
278d0ca1 CH |
75 | * This either works (return 0) or gets an EFSCORRUPTED error. |
76 | */ | |
77 | STATIC int /* error */ | |
78 | xfs_inobt_update( | |
79 | struct xfs_btree_cur *cur, /* btree cursor */ | |
afabc24a | 80 | xfs_inobt_rec_incore_t *irec) /* btree record */ |
278d0ca1 CH |
81 | { |
82 | union xfs_btree_rec rec; | |
83 | ||
afabc24a CH |
84 | rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino); |
85 | rec.inobt.ir_freecount = cpu_to_be32(irec->ir_freecount); | |
86 | rec.inobt.ir_free = cpu_to_be64(irec->ir_free); | |
278d0ca1 CH |
87 | return xfs_btree_update(cur, &rec); |
88 | } | |
89 | ||
8cc938fe CH |
90 | /* |
91 | * Get the data from the pointed-to record. | |
92 | */ | |
93 | int /* error */ | |
94 | xfs_inobt_get_rec( | |
95 | struct xfs_btree_cur *cur, /* btree cursor */ | |
2e287a73 | 96 | xfs_inobt_rec_incore_t *irec, /* btree record */ |
8cc938fe CH |
97 | int *stat) /* output: success/failure */ |
98 | { | |
99 | union xfs_btree_rec *rec; | |
100 | int error; | |
101 | ||
102 | error = xfs_btree_get_rec(cur, &rec, stat); | |
103 | if (!error && *stat == 1) { | |
2e287a73 CH |
104 | irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino); |
105 | irec->ir_freecount = be32_to_cpu(rec->inobt.ir_freecount); | |
106 | irec->ir_free = be64_to_cpu(rec->inobt.ir_free); | |
8cc938fe CH |
107 | } |
108 | return error; | |
109 | } | |
110 | ||
0aa0a756 BF |
111 | /* |
112 | * Insert a single inobt record. Cursor must already point to desired location. | |
113 | */ | |
114 | STATIC int | |
115 | xfs_inobt_insert_rec( | |
116 | struct xfs_btree_cur *cur, | |
117 | __int32_t freecount, | |
118 | xfs_inofree_t free, | |
119 | int *stat) | |
120 | { | |
121 | cur->bc_rec.i.ir_freecount = freecount; | |
122 | cur->bc_rec.i.ir_free = free; | |
123 | return xfs_btree_insert(cur, stat); | |
124 | } | |
125 | ||
126 | /* | |
127 | * Insert records describing a newly allocated inode chunk into the inobt. | |
128 | */ | |
129 | STATIC int | |
130 | xfs_inobt_insert( | |
131 | struct xfs_mount *mp, | |
132 | struct xfs_trans *tp, | |
133 | struct xfs_buf *agbp, | |
134 | xfs_agino_t newino, | |
135 | xfs_agino_t newlen, | |
136 | xfs_btnum_t btnum) | |
137 | { | |
138 | struct xfs_btree_cur *cur; | |
139 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
140 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
141 | xfs_agino_t thisino; | |
142 | int i; | |
143 | int error; | |
144 | ||
145 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum); | |
146 | ||
147 | for (thisino = newino; | |
148 | thisino < newino + newlen; | |
149 | thisino += XFS_INODES_PER_CHUNK) { | |
150 | error = xfs_inobt_lookup(cur, thisino, XFS_LOOKUP_EQ, &i); | |
151 | if (error) { | |
152 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
153 | return error; | |
154 | } | |
155 | ASSERT(i == 0); | |
156 | ||
157 | error = xfs_inobt_insert_rec(cur, XFS_INODES_PER_CHUNK, | |
158 | XFS_INOBT_ALL_FREE, &i); | |
159 | if (error) { | |
160 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
161 | return error; | |
162 | } | |
163 | ASSERT(i == 1); | |
164 | } | |
165 | ||
166 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
167 | ||
168 | return 0; | |
169 | } | |
170 | ||
0b48db80 DC |
171 | /* |
172 | * Verify that the number of free inodes in the AGI is correct. | |
173 | */ | |
174 | #ifdef DEBUG | |
175 | STATIC int | |
176 | xfs_check_agi_freecount( | |
177 | struct xfs_btree_cur *cur, | |
178 | struct xfs_agi *agi) | |
179 | { | |
180 | if (cur->bc_nlevels == 1) { | |
181 | xfs_inobt_rec_incore_t rec; | |
182 | int freecount = 0; | |
183 | int error; | |
184 | int i; | |
185 | ||
21875505 | 186 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); |
0b48db80 DC |
187 | if (error) |
188 | return error; | |
189 | ||
190 | do { | |
191 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
192 | if (error) | |
193 | return error; | |
194 | ||
195 | if (i) { | |
196 | freecount += rec.ir_freecount; | |
197 | error = xfs_btree_increment(cur, 0, &i); | |
198 | if (error) | |
199 | return error; | |
200 | } | |
201 | } while (i == 1); | |
202 | ||
203 | if (!XFS_FORCED_SHUTDOWN(cur->bc_mp)) | |
204 | ASSERT(freecount == be32_to_cpu(agi->agi_freecount)); | |
205 | } | |
206 | return 0; | |
207 | } | |
208 | #else | |
209 | #define xfs_check_agi_freecount(cur, agi) 0 | |
210 | #endif | |
211 | ||
85c0b2ab | 212 | /* |
28c8e41a DC |
213 | * Initialise a new set of inodes. When called without a transaction context |
214 | * (e.g. from recovery) we initiate a delayed write of the inode buffers rather | |
215 | * than logging them (which in a transaction context puts them into the AIL | |
216 | * for writeback rather than the xfsbufd queue). | |
85c0b2ab | 217 | */ |
ddf6ad01 | 218 | int |
85c0b2ab DC |
219 | xfs_ialloc_inode_init( |
220 | struct xfs_mount *mp, | |
221 | struct xfs_trans *tp, | |
28c8e41a | 222 | struct list_head *buffer_list, |
85c0b2ab DC |
223 | xfs_agnumber_t agno, |
224 | xfs_agblock_t agbno, | |
225 | xfs_agblock_t length, | |
226 | unsigned int gen) | |
227 | { | |
228 | struct xfs_buf *fbuf; | |
229 | struct xfs_dinode *free; | |
6e0c7b8c | 230 | int nbufs, blks_per_cluster, inodes_per_cluster; |
85c0b2ab DC |
231 | int version; |
232 | int i, j; | |
233 | xfs_daddr_t d; | |
93848a99 | 234 | xfs_ino_t ino = 0; |
85c0b2ab DC |
235 | |
236 | /* | |
6e0c7b8c JL |
237 | * Loop over the new block(s), filling in the inodes. For small block |
238 | * sizes, manipulate the inodes in buffers which are multiples of the | |
239 | * blocks size. | |
85c0b2ab | 240 | */ |
6e0c7b8c JL |
241 | blks_per_cluster = xfs_icluster_size_fsb(mp); |
242 | inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog; | |
243 | nbufs = length / blks_per_cluster; | |
85c0b2ab DC |
244 | |
245 | /* | |
93848a99 CH |
246 | * Figure out what version number to use in the inodes we create. If |
247 | * the superblock version has caught up to the one that supports the new | |
248 | * inode format, then use the new inode version. Otherwise use the old | |
249 | * version so that old kernels will continue to be able to use the file | |
250 | * system. | |
251 | * | |
252 | * For v3 inodes, we also need to write the inode number into the inode, | |
253 | * so calculate the first inode number of the chunk here as | |
254 | * XFS_OFFBNO_TO_AGINO() only works within a filesystem block, not | |
255 | * across multiple filesystem blocks (such as a cluster) and so cannot | |
256 | * be used in the cluster buffer loop below. | |
257 | * | |
258 | * Further, because we are writing the inode directly into the buffer | |
259 | * and calculating a CRC on the entire inode, we have ot log the entire | |
260 | * inode so that the entire range the CRC covers is present in the log. | |
261 | * That means for v3 inode we log the entire buffer rather than just the | |
262 | * inode cores. | |
85c0b2ab | 263 | */ |
93848a99 CH |
264 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
265 | version = 3; | |
266 | ino = XFS_AGINO_TO_INO(mp, agno, | |
267 | XFS_OFFBNO_TO_AGINO(mp, agbno, 0)); | |
ddf6ad01 DC |
268 | |
269 | /* | |
270 | * log the initialisation that is about to take place as an | |
271 | * logical operation. This means the transaction does not | |
272 | * need to log the physical changes to the inode buffers as log | |
273 | * recovery will know what initialisation is actually needed. | |
274 | * Hence we only need to log the buffers as "ordered" buffers so | |
275 | * they track in the AIL as if they were physically logged. | |
276 | */ | |
277 | if (tp) | |
71783438 | 278 | xfs_icreate_log(tp, agno, agbno, mp->m_ialloc_inos, |
ddf6ad01 | 279 | mp->m_sb.sb_inodesize, length, gen); |
263997a6 | 280 | } else |
85c0b2ab | 281 | version = 2; |
85c0b2ab DC |
282 | |
283 | for (j = 0; j < nbufs; j++) { | |
284 | /* | |
285 | * Get the block. | |
286 | */ | |
287 | d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster)); | |
288 | fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, | |
7c4cebe8 DC |
289 | mp->m_bsize * blks_per_cluster, |
290 | XBF_UNMAPPED); | |
2a30f36d | 291 | if (!fbuf) |
2451337d | 292 | return -ENOMEM; |
ddf6ad01 DC |
293 | |
294 | /* Initialize the inode buffers and log them appropriately. */ | |
1813dd64 | 295 | fbuf->b_ops = &xfs_inode_buf_ops; |
93848a99 | 296 | xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length)); |
6e0c7b8c | 297 | for (i = 0; i < inodes_per_cluster; i++) { |
85c0b2ab | 298 | int ioffset = i << mp->m_sb.sb_inodelog; |
93848a99 | 299 | uint isize = xfs_dinode_size(version); |
85c0b2ab DC |
300 | |
301 | free = xfs_make_iptr(mp, fbuf, i); | |
302 | free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC); | |
303 | free->di_version = version; | |
304 | free->di_gen = cpu_to_be32(gen); | |
305 | free->di_next_unlinked = cpu_to_be32(NULLAGINO); | |
93848a99 CH |
306 | |
307 | if (version == 3) { | |
308 | free->di_ino = cpu_to_be64(ino); | |
309 | ino++; | |
310 | uuid_copy(&free->di_uuid, &mp->m_sb.sb_uuid); | |
311 | xfs_dinode_calc_crc(mp, free); | |
28c8e41a | 312 | } else if (tp) { |
93848a99 CH |
313 | /* just log the inode core */ |
314 | xfs_trans_log_buf(tp, fbuf, ioffset, | |
315 | ioffset + isize - 1); | |
316 | } | |
317 | } | |
28c8e41a DC |
318 | |
319 | if (tp) { | |
320 | /* | |
321 | * Mark the buffer as an inode allocation buffer so it | |
322 | * sticks in AIL at the point of this allocation | |
323 | * transaction. This ensures the they are on disk before | |
324 | * the tail of the log can be moved past this | |
325 | * transaction (i.e. by preventing relogging from moving | |
326 | * it forward in the log). | |
327 | */ | |
328 | xfs_trans_inode_alloc_buf(tp, fbuf); | |
329 | if (version == 3) { | |
ddf6ad01 DC |
330 | /* |
331 | * Mark the buffer as ordered so that they are | |
332 | * not physically logged in the transaction but | |
333 | * still tracked in the AIL as part of the | |
334 | * transaction and pin the log appropriately. | |
335 | */ | |
336 | xfs_trans_ordered_buf(tp, fbuf); | |
28c8e41a DC |
337 | xfs_trans_log_buf(tp, fbuf, 0, |
338 | BBTOB(fbuf->b_length) - 1); | |
339 | } | |
340 | } else { | |
341 | fbuf->b_flags |= XBF_DONE; | |
342 | xfs_buf_delwri_queue(fbuf, buffer_list); | |
343 | xfs_buf_relse(fbuf); | |
85c0b2ab | 344 | } |
85c0b2ab | 345 | } |
2a30f36d | 346 | return 0; |
85c0b2ab DC |
347 | } |
348 | ||
1da177e4 LT |
349 | /* |
350 | * Allocate new inodes in the allocation group specified by agbp. | |
351 | * Return 0 for success, else error code. | |
352 | */ | |
353 | STATIC int /* error code or 0 */ | |
354 | xfs_ialloc_ag_alloc( | |
355 | xfs_trans_t *tp, /* transaction pointer */ | |
356 | xfs_buf_t *agbp, /* alloc group buffer */ | |
357 | int *alloc) | |
358 | { | |
359 | xfs_agi_t *agi; /* allocation group header */ | |
360 | xfs_alloc_arg_t args; /* allocation argument structure */ | |
92821e2b | 361 | xfs_agnumber_t agno; |
1da177e4 | 362 | int error; |
1da177e4 LT |
363 | xfs_agino_t newino; /* new first inode's number */ |
364 | xfs_agino_t newlen; /* new number of inodes */ | |
3ccb8b5f | 365 | int isaligned = 0; /* inode allocation at stripe unit */ |
1da177e4 | 366 | /* boundary */ |
44b56e0a | 367 | struct xfs_perag *pag; |
1da177e4 | 368 | |
a0041684 | 369 | memset(&args, 0, sizeof(args)); |
1da177e4 LT |
370 | args.tp = tp; |
371 | args.mp = tp->t_mountp; | |
372 | ||
373 | /* | |
374 | * Locking will ensure that we don't have two callers in here | |
375 | * at one time. | |
376 | */ | |
71783438 | 377 | newlen = args.mp->m_ialloc_inos; |
1da177e4 LT |
378 | if (args.mp->m_maxicount && |
379 | args.mp->m_sb.sb_icount + newlen > args.mp->m_maxicount) | |
2451337d | 380 | return -ENOSPC; |
126cd105 | 381 | args.minlen = args.maxlen = args.mp->m_ialloc_blks; |
1da177e4 | 382 | /* |
3ccb8b5f GO |
383 | * First try to allocate inodes contiguous with the last-allocated |
384 | * chunk of inodes. If the filesystem is striped, this will fill | |
385 | * an entire stripe unit with inodes. | |
28c8e41a | 386 | */ |
1da177e4 | 387 | agi = XFS_BUF_TO_AGI(agbp); |
3ccb8b5f | 388 | newino = be32_to_cpu(agi->agi_newino); |
85c0b2ab | 389 | agno = be32_to_cpu(agi->agi_seqno); |
019ff2d5 | 390 | args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) + |
126cd105 | 391 | args.mp->m_ialloc_blks; |
019ff2d5 NS |
392 | if (likely(newino != NULLAGINO && |
393 | (args.agbno < be32_to_cpu(agi->agi_length)))) { | |
85c0b2ab | 394 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
3ccb8b5f | 395 | args.type = XFS_ALLOCTYPE_THIS_BNO; |
3ccb8b5f | 396 | args.prod = 1; |
75de2a91 | 397 | |
3ccb8b5f | 398 | /* |
75de2a91 DC |
399 | * We need to take into account alignment here to ensure that |
400 | * we don't modify the free list if we fail to have an exact | |
401 | * block. If we don't have an exact match, and every oher | |
402 | * attempt allocation attempt fails, we'll end up cancelling | |
403 | * a dirty transaction and shutting down. | |
404 | * | |
405 | * For an exact allocation, alignment must be 1, | |
406 | * however we need to take cluster alignment into account when | |
407 | * fixing up the freelist. Use the minalignslop field to | |
408 | * indicate that extra blocks might be required for alignment, | |
409 | * but not to use them in the actual exact allocation. | |
3ccb8b5f | 410 | */ |
75de2a91 | 411 | args.alignment = 1; |
7a1df156 | 412 | args.minalignslop = xfs_ialloc_cluster_alignment(args.mp) - 1; |
75de2a91 DC |
413 | |
414 | /* Allow space for the inode btree to split. */ | |
0d87e656 | 415 | args.minleft = args.mp->m_in_maxlevels - 1; |
3ccb8b5f GO |
416 | if ((error = xfs_alloc_vextent(&args))) |
417 | return error; | |
e480a723 BF |
418 | |
419 | /* | |
420 | * This request might have dirtied the transaction if the AG can | |
421 | * satisfy the request, but the exact block was not available. | |
422 | * If the allocation did fail, subsequent requests will relax | |
423 | * the exact agbno requirement and increase the alignment | |
424 | * instead. It is critical that the total size of the request | |
425 | * (len + alignment + slop) does not increase from this point | |
426 | * on, so reset minalignslop to ensure it is not included in | |
427 | * subsequent requests. | |
428 | */ | |
429 | args.minalignslop = 0; | |
3ccb8b5f GO |
430 | } else |
431 | args.fsbno = NULLFSBLOCK; | |
1da177e4 | 432 | |
3ccb8b5f GO |
433 | if (unlikely(args.fsbno == NULLFSBLOCK)) { |
434 | /* | |
435 | * Set the alignment for the allocation. | |
436 | * If stripe alignment is turned on then align at stripe unit | |
437 | * boundary. | |
019ff2d5 NS |
438 | * If the cluster size is smaller than a filesystem block |
439 | * then we're doing I/O for inodes in filesystem block size | |
3ccb8b5f GO |
440 | * pieces, so don't need alignment anyway. |
441 | */ | |
442 | isaligned = 0; | |
443 | if (args.mp->m_sinoalign) { | |
444 | ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN)); | |
445 | args.alignment = args.mp->m_dalign; | |
446 | isaligned = 1; | |
75de2a91 | 447 | } else |
7a1df156 | 448 | args.alignment = xfs_ialloc_cluster_alignment(args.mp); |
3ccb8b5f GO |
449 | /* |
450 | * Need to figure out where to allocate the inode blocks. | |
451 | * Ideally they should be spaced out through the a.g. | |
452 | * For now, just allocate blocks up front. | |
453 | */ | |
454 | args.agbno = be32_to_cpu(agi->agi_root); | |
85c0b2ab | 455 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
3ccb8b5f GO |
456 | /* |
457 | * Allocate a fixed-size extent of inodes. | |
458 | */ | |
459 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
3ccb8b5f GO |
460 | args.prod = 1; |
461 | /* | |
462 | * Allow space for the inode btree to split. | |
463 | */ | |
0d87e656 | 464 | args.minleft = args.mp->m_in_maxlevels - 1; |
3ccb8b5f GO |
465 | if ((error = xfs_alloc_vextent(&args))) |
466 | return error; | |
467 | } | |
019ff2d5 | 468 | |
1da177e4 LT |
469 | /* |
470 | * If stripe alignment is turned on, then try again with cluster | |
471 | * alignment. | |
472 | */ | |
473 | if (isaligned && args.fsbno == NULLFSBLOCK) { | |
474 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
16259e7d | 475 | args.agbno = be32_to_cpu(agi->agi_root); |
85c0b2ab | 476 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
7a1df156 | 477 | args.alignment = xfs_ialloc_cluster_alignment(args.mp); |
1da177e4 LT |
478 | if ((error = xfs_alloc_vextent(&args))) |
479 | return error; | |
480 | } | |
481 | ||
482 | if (args.fsbno == NULLFSBLOCK) { | |
483 | *alloc = 0; | |
484 | return 0; | |
485 | } | |
486 | ASSERT(args.len == args.minlen); | |
1da177e4 | 487 | |
359346a9 | 488 | /* |
85c0b2ab DC |
489 | * Stamp and write the inode buffers. |
490 | * | |
359346a9 DC |
491 | * Seed the new inode cluster with a random generation number. This |
492 | * prevents short-term reuse of generation numbers if a chunk is | |
493 | * freed and then immediately reallocated. We use random numbers | |
494 | * rather than a linear progression to prevent the next generation | |
495 | * number from being easily guessable. | |
496 | */ | |
28c8e41a | 497 | error = xfs_ialloc_inode_init(args.mp, tp, NULL, agno, args.agbno, |
ecb3403d | 498 | args.len, prandom_u32()); |
d42f08f6 | 499 | |
2a30f36d CS |
500 | if (error) |
501 | return error; | |
85c0b2ab DC |
502 | /* |
503 | * Convert the results. | |
504 | */ | |
505 | newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0); | |
413d57c9 MS |
506 | be32_add_cpu(&agi->agi_count, newlen); |
507 | be32_add_cpu(&agi->agi_freecount, newlen); | |
44b56e0a DC |
508 | pag = xfs_perag_get(args.mp, agno); |
509 | pag->pagi_freecount += newlen; | |
510 | xfs_perag_put(pag); | |
16259e7d | 511 | agi->agi_newino = cpu_to_be32(newino); |
85c0b2ab | 512 | |
1da177e4 | 513 | /* |
0aa0a756 | 514 | * Insert records describing the new inode chunk into the btrees. |
1da177e4 | 515 | */ |
0aa0a756 BF |
516 | error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen, |
517 | XFS_BTNUM_INO); | |
518 | if (error) | |
519 | return error; | |
520 | ||
521 | if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) { | |
522 | error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen, | |
523 | XFS_BTNUM_FINO); | |
524 | if (error) | |
1da177e4 | 525 | return error; |
1da177e4 | 526 | } |
1da177e4 LT |
527 | /* |
528 | * Log allocation group header fields | |
529 | */ | |
530 | xfs_ialloc_log_agi(tp, agbp, | |
531 | XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO); | |
532 | /* | |
533 | * Modify/log superblock values for inode count and inode free count. | |
534 | */ | |
535 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen); | |
536 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen); | |
537 | *alloc = 1; | |
538 | return 0; | |
539 | } | |
540 | ||
b8f82a4a | 541 | STATIC xfs_agnumber_t |
1da177e4 LT |
542 | xfs_ialloc_next_ag( |
543 | xfs_mount_t *mp) | |
544 | { | |
545 | xfs_agnumber_t agno; | |
546 | ||
547 | spin_lock(&mp->m_agirotor_lock); | |
548 | agno = mp->m_agirotor; | |
8aea3ff4 | 549 | if (++mp->m_agirotor >= mp->m_maxagi) |
1da177e4 LT |
550 | mp->m_agirotor = 0; |
551 | spin_unlock(&mp->m_agirotor_lock); | |
552 | ||
553 | return agno; | |
554 | } | |
555 | ||
556 | /* | |
557 | * Select an allocation group to look for a free inode in, based on the parent | |
2f21ff1c | 558 | * inode and the mode. Return the allocation group buffer. |
1da177e4 | 559 | */ |
55d6af64 | 560 | STATIC xfs_agnumber_t |
1da177e4 LT |
561 | xfs_ialloc_ag_select( |
562 | xfs_trans_t *tp, /* transaction pointer */ | |
563 | xfs_ino_t parent, /* parent directory inode number */ | |
576b1d67 | 564 | umode_t mode, /* bits set to indicate file type */ |
1da177e4 LT |
565 | int okalloc) /* ok to allocate more space */ |
566 | { | |
1da177e4 LT |
567 | xfs_agnumber_t agcount; /* number of ag's in the filesystem */ |
568 | xfs_agnumber_t agno; /* current ag number */ | |
569 | int flags; /* alloc buffer locking flags */ | |
570 | xfs_extlen_t ineed; /* blocks needed for inode allocation */ | |
571 | xfs_extlen_t longest = 0; /* longest extent available */ | |
572 | xfs_mount_t *mp; /* mount point structure */ | |
573 | int needspace; /* file mode implies space allocated */ | |
574 | xfs_perag_t *pag; /* per allocation group data */ | |
575 | xfs_agnumber_t pagno; /* parent (starting) ag number */ | |
55d6af64 | 576 | int error; |
1da177e4 LT |
577 | |
578 | /* | |
579 | * Files of these types need at least one block if length > 0 | |
580 | * (and they won't fit in the inode, but that's hard to figure out). | |
581 | */ | |
582 | needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode); | |
583 | mp = tp->t_mountp; | |
584 | agcount = mp->m_maxagi; | |
585 | if (S_ISDIR(mode)) | |
586 | pagno = xfs_ialloc_next_ag(mp); | |
587 | else { | |
588 | pagno = XFS_INO_TO_AGNO(mp, parent); | |
589 | if (pagno >= agcount) | |
590 | pagno = 0; | |
591 | } | |
55d6af64 | 592 | |
1da177e4 | 593 | ASSERT(pagno < agcount); |
55d6af64 | 594 | |
1da177e4 LT |
595 | /* |
596 | * Loop through allocation groups, looking for one with a little | |
597 | * free space in it. Note we don't look for free inodes, exactly. | |
598 | * Instead, we include whether there is a need to allocate inodes | |
599 | * to mean that blocks must be allocated for them, | |
600 | * if none are currently free. | |
601 | */ | |
602 | agno = pagno; | |
603 | flags = XFS_ALLOC_FLAG_TRYLOCK; | |
1da177e4 | 604 | for (;;) { |
44b56e0a | 605 | pag = xfs_perag_get(mp, agno); |
55d6af64 CH |
606 | if (!pag->pagi_inodeok) { |
607 | xfs_ialloc_next_ag(mp); | |
608 | goto nextag; | |
609 | } | |
610 | ||
1da177e4 | 611 | if (!pag->pagi_init) { |
55d6af64 CH |
612 | error = xfs_ialloc_pagi_init(mp, tp, agno); |
613 | if (error) | |
1da177e4 | 614 | goto nextag; |
55d6af64 | 615 | } |
1da177e4 | 616 | |
55d6af64 CH |
617 | if (pag->pagi_freecount) { |
618 | xfs_perag_put(pag); | |
619 | return agno; | |
1da177e4 LT |
620 | } |
621 | ||
55d6af64 CH |
622 | if (!okalloc) |
623 | goto nextag; | |
624 | ||
625 | if (!pag->pagf_init) { | |
626 | error = xfs_alloc_pagf_init(mp, tp, agno, flags); | |
627 | if (error) | |
1da177e4 | 628 | goto nextag; |
1da177e4 | 629 | } |
55d6af64 CH |
630 | |
631 | /* | |
7a1df156 DC |
632 | * Check that there is enough free space for the file plus a |
633 | * chunk of inodes if we need to allocate some. If this is the | |
634 | * first pass across the AGs, take into account the potential | |
635 | * space needed for alignment of inode chunks when checking the | |
636 | * longest contiguous free space in the AG - this prevents us | |
637 | * from getting ENOSPC because we have free space larger than | |
638 | * m_ialloc_blks but alignment constraints prevent us from using | |
639 | * it. | |
640 | * | |
641 | * If we can't find an AG with space for full alignment slack to | |
642 | * be taken into account, we must be near ENOSPC in all AGs. | |
643 | * Hence we don't include alignment for the second pass and so | |
644 | * if we fail allocation due to alignment issues then it is most | |
645 | * likely a real ENOSPC condition. | |
55d6af64 | 646 | */ |
126cd105 | 647 | ineed = mp->m_ialloc_blks; |
7a1df156 DC |
648 | if (flags && ineed > 1) |
649 | ineed += xfs_ialloc_cluster_alignment(mp); | |
55d6af64 CH |
650 | longest = pag->pagf_longest; |
651 | if (!longest) | |
652 | longest = pag->pagf_flcount > 0; | |
653 | ||
654 | if (pag->pagf_freeblks >= needspace + ineed && | |
655 | longest >= ineed) { | |
656 | xfs_perag_put(pag); | |
657 | return agno; | |
1da177e4 | 658 | } |
1da177e4 | 659 | nextag: |
44b56e0a | 660 | xfs_perag_put(pag); |
1da177e4 LT |
661 | /* |
662 | * No point in iterating over the rest, if we're shutting | |
663 | * down. | |
664 | */ | |
1c1c6ebc | 665 | if (XFS_FORCED_SHUTDOWN(mp)) |
55d6af64 | 666 | return NULLAGNUMBER; |
1da177e4 LT |
667 | agno++; |
668 | if (agno >= agcount) | |
669 | agno = 0; | |
670 | if (agno == pagno) { | |
1c1c6ebc | 671 | if (flags == 0) |
55d6af64 | 672 | return NULLAGNUMBER; |
1da177e4 LT |
673 | flags = 0; |
674 | } | |
675 | } | |
676 | } | |
677 | ||
4254b0bb CH |
678 | /* |
679 | * Try to retrieve the next record to the left/right from the current one. | |
680 | */ | |
681 | STATIC int | |
682 | xfs_ialloc_next_rec( | |
683 | struct xfs_btree_cur *cur, | |
684 | xfs_inobt_rec_incore_t *rec, | |
685 | int *done, | |
686 | int left) | |
687 | { | |
688 | int error; | |
689 | int i; | |
690 | ||
691 | if (left) | |
692 | error = xfs_btree_decrement(cur, 0, &i); | |
693 | else | |
694 | error = xfs_btree_increment(cur, 0, &i); | |
695 | ||
696 | if (error) | |
697 | return error; | |
698 | *done = !i; | |
699 | if (i) { | |
700 | error = xfs_inobt_get_rec(cur, rec, &i); | |
701 | if (error) | |
702 | return error; | |
703 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
704 | } | |
705 | ||
706 | return 0; | |
707 | } | |
708 | ||
bd169565 DC |
709 | STATIC int |
710 | xfs_ialloc_get_rec( | |
711 | struct xfs_btree_cur *cur, | |
712 | xfs_agino_t agino, | |
713 | xfs_inobt_rec_incore_t *rec, | |
43df2ee6 | 714 | int *done) |
bd169565 DC |
715 | { |
716 | int error; | |
717 | int i; | |
718 | ||
719 | error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i); | |
720 | if (error) | |
721 | return error; | |
722 | *done = !i; | |
723 | if (i) { | |
724 | error = xfs_inobt_get_rec(cur, rec, &i); | |
725 | if (error) | |
726 | return error; | |
727 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
728 | } | |
729 | ||
730 | return 0; | |
731 | } | |
0b48db80 | 732 | |
1da177e4 | 733 | /* |
6dd8638e | 734 | * Allocate an inode using the inobt-only algorithm. |
1da177e4 | 735 | */ |
f2ecc5e4 | 736 | STATIC int |
6dd8638e | 737 | xfs_dialloc_ag_inobt( |
f2ecc5e4 CH |
738 | struct xfs_trans *tp, |
739 | struct xfs_buf *agbp, | |
740 | xfs_ino_t parent, | |
741 | xfs_ino_t *inop) | |
1da177e4 | 742 | { |
f2ecc5e4 CH |
743 | struct xfs_mount *mp = tp->t_mountp; |
744 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
745 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
746 | xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent); | |
747 | xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent); | |
748 | struct xfs_perag *pag; | |
749 | struct xfs_btree_cur *cur, *tcur; | |
750 | struct xfs_inobt_rec_incore rec, trec; | |
751 | xfs_ino_t ino; | |
752 | int error; | |
753 | int offset; | |
754 | int i, j; | |
1da177e4 | 755 | |
44b56e0a | 756 | pag = xfs_perag_get(mp, agno); |
bd169565 | 757 | |
4bb61069 CH |
758 | ASSERT(pag->pagi_init); |
759 | ASSERT(pag->pagi_inodeok); | |
760 | ASSERT(pag->pagi_freecount > 0); | |
761 | ||
bd169565 | 762 | restart_pagno: |
57bd3dbe | 763 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
1da177e4 LT |
764 | /* |
765 | * If pagino is 0 (this is the root inode allocation) use newino. | |
766 | * This must work because we've just allocated some. | |
767 | */ | |
768 | if (!pagino) | |
16259e7d | 769 | pagino = be32_to_cpu(agi->agi_newino); |
1da177e4 | 770 | |
0b48db80 DC |
771 | error = xfs_check_agi_freecount(cur, agi); |
772 | if (error) | |
773 | goto error0; | |
1da177e4 | 774 | |
1da177e4 | 775 | /* |
4254b0bb | 776 | * If in the same AG as the parent, try to get near the parent. |
1da177e4 LT |
777 | */ |
778 | if (pagno == agno) { | |
4254b0bb CH |
779 | int doneleft; /* done, to the left */ |
780 | int doneright; /* done, to the right */ | |
bd169565 | 781 | int searchdistance = 10; |
4254b0bb | 782 | |
21875505 | 783 | error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i); |
4254b0bb | 784 | if (error) |
1da177e4 | 785 | goto error0; |
c29aad41 | 786 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
4254b0bb CH |
787 | |
788 | error = xfs_inobt_get_rec(cur, &rec, &j); | |
789 | if (error) | |
790 | goto error0; | |
c29aad41 | 791 | XFS_WANT_CORRUPTED_GOTO(mp, j == 1, error0); |
4254b0bb CH |
792 | |
793 | if (rec.ir_freecount > 0) { | |
1da177e4 LT |
794 | /* |
795 | * Found a free inode in the same chunk | |
4254b0bb | 796 | * as the parent, done. |
1da177e4 | 797 | */ |
4254b0bb | 798 | goto alloc_inode; |
1da177e4 | 799 | } |
4254b0bb CH |
800 | |
801 | ||
1da177e4 | 802 | /* |
4254b0bb | 803 | * In the same AG as parent, but parent's chunk is full. |
1da177e4 | 804 | */ |
1da177e4 | 805 | |
4254b0bb CH |
806 | /* duplicate the cursor, search left & right simultaneously */ |
807 | error = xfs_btree_dup_cursor(cur, &tcur); | |
808 | if (error) | |
809 | goto error0; | |
810 | ||
bd169565 DC |
811 | /* |
812 | * Skip to last blocks looked up if same parent inode. | |
813 | */ | |
814 | if (pagino != NULLAGINO && | |
815 | pag->pagl_pagino == pagino && | |
816 | pag->pagl_leftrec != NULLAGINO && | |
817 | pag->pagl_rightrec != NULLAGINO) { | |
818 | error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec, | |
43df2ee6 | 819 | &trec, &doneleft); |
bd169565 DC |
820 | if (error) |
821 | goto error1; | |
4254b0bb | 822 | |
bd169565 | 823 | error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec, |
43df2ee6 | 824 | &rec, &doneright); |
bd169565 DC |
825 | if (error) |
826 | goto error1; | |
827 | } else { | |
828 | /* search left with tcur, back up 1 record */ | |
829 | error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1); | |
830 | if (error) | |
831 | goto error1; | |
832 | ||
833 | /* search right with cur, go forward 1 record. */ | |
834 | error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0); | |
835 | if (error) | |
836 | goto error1; | |
837 | } | |
4254b0bb CH |
838 | |
839 | /* | |
840 | * Loop until we find an inode chunk with a free inode. | |
841 | */ | |
842 | while (!doneleft || !doneright) { | |
843 | int useleft; /* using left inode chunk this time */ | |
844 | ||
bd169565 DC |
845 | if (!--searchdistance) { |
846 | /* | |
847 | * Not in range - save last search | |
848 | * location and allocate a new inode | |
849 | */ | |
3b826386 | 850 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); |
bd169565 DC |
851 | pag->pagl_leftrec = trec.ir_startino; |
852 | pag->pagl_rightrec = rec.ir_startino; | |
853 | pag->pagl_pagino = pagino; | |
854 | goto newino; | |
855 | } | |
856 | ||
4254b0bb CH |
857 | /* figure out the closer block if both are valid. */ |
858 | if (!doneleft && !doneright) { | |
859 | useleft = pagino - | |
860 | (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) < | |
861 | rec.ir_startino - pagino; | |
862 | } else { | |
863 | useleft = !doneleft; | |
1da177e4 | 864 | } |
4254b0bb CH |
865 | |
866 | /* free inodes to the left? */ | |
867 | if (useleft && trec.ir_freecount) { | |
868 | rec = trec; | |
869 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
870 | cur = tcur; | |
bd169565 DC |
871 | |
872 | pag->pagl_leftrec = trec.ir_startino; | |
873 | pag->pagl_rightrec = rec.ir_startino; | |
874 | pag->pagl_pagino = pagino; | |
4254b0bb | 875 | goto alloc_inode; |
1da177e4 | 876 | } |
1da177e4 | 877 | |
4254b0bb CH |
878 | /* free inodes to the right? */ |
879 | if (!useleft && rec.ir_freecount) { | |
880 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
bd169565 DC |
881 | |
882 | pag->pagl_leftrec = trec.ir_startino; | |
883 | pag->pagl_rightrec = rec.ir_startino; | |
884 | pag->pagl_pagino = pagino; | |
4254b0bb | 885 | goto alloc_inode; |
1da177e4 | 886 | } |
4254b0bb CH |
887 | |
888 | /* get next record to check */ | |
889 | if (useleft) { | |
890 | error = xfs_ialloc_next_rec(tcur, &trec, | |
891 | &doneleft, 1); | |
892 | } else { | |
893 | error = xfs_ialloc_next_rec(cur, &rec, | |
894 | &doneright, 0); | |
895 | } | |
896 | if (error) | |
897 | goto error1; | |
1da177e4 | 898 | } |
bd169565 DC |
899 | |
900 | /* | |
901 | * We've reached the end of the btree. because | |
902 | * we are only searching a small chunk of the | |
903 | * btree each search, there is obviously free | |
904 | * inodes closer to the parent inode than we | |
905 | * are now. restart the search again. | |
906 | */ | |
907 | pag->pagl_pagino = NULLAGINO; | |
908 | pag->pagl_leftrec = NULLAGINO; | |
909 | pag->pagl_rightrec = NULLAGINO; | |
910 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
911 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
912 | goto restart_pagno; | |
1da177e4 | 913 | } |
4254b0bb | 914 | |
1da177e4 | 915 | /* |
4254b0bb | 916 | * In a different AG from the parent. |
1da177e4 LT |
917 | * See if the most recently allocated block has any free. |
918 | */ | |
bd169565 | 919 | newino: |
69ef921b | 920 | if (agi->agi_newino != cpu_to_be32(NULLAGINO)) { |
21875505 CH |
921 | error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino), |
922 | XFS_LOOKUP_EQ, &i); | |
4254b0bb | 923 | if (error) |
1da177e4 | 924 | goto error0; |
4254b0bb CH |
925 | |
926 | if (i == 1) { | |
927 | error = xfs_inobt_get_rec(cur, &rec, &j); | |
928 | if (error) | |
929 | goto error0; | |
930 | ||
931 | if (j == 1 && rec.ir_freecount > 0) { | |
932 | /* | |
933 | * The last chunk allocated in the group | |
934 | * still has a free inode. | |
935 | */ | |
936 | goto alloc_inode; | |
937 | } | |
1da177e4 | 938 | } |
bd169565 | 939 | } |
4254b0bb | 940 | |
bd169565 DC |
941 | /* |
942 | * None left in the last group, search the whole AG | |
943 | */ | |
944 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); | |
945 | if (error) | |
946 | goto error0; | |
c29aad41 | 947 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
bd169565 DC |
948 | |
949 | for (;;) { | |
950 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
951 | if (error) | |
952 | goto error0; | |
c29aad41 | 953 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
bd169565 DC |
954 | if (rec.ir_freecount > 0) |
955 | break; | |
956 | error = xfs_btree_increment(cur, 0, &i); | |
4254b0bb CH |
957 | if (error) |
958 | goto error0; | |
c29aad41 | 959 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 | 960 | } |
4254b0bb CH |
961 | |
962 | alloc_inode: | |
824c3131 | 963 | offset = xfs_lowbit64(rec.ir_free); |
1da177e4 LT |
964 | ASSERT(offset >= 0); |
965 | ASSERT(offset < XFS_INODES_PER_CHUNK); | |
966 | ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) % | |
967 | XFS_INODES_PER_CHUNK) == 0); | |
968 | ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset); | |
0d87e656 | 969 | rec.ir_free &= ~XFS_INOBT_MASK(offset); |
1da177e4 | 970 | rec.ir_freecount--; |
afabc24a CH |
971 | error = xfs_inobt_update(cur, &rec); |
972 | if (error) | |
1da177e4 | 973 | goto error0; |
413d57c9 | 974 | be32_add_cpu(&agi->agi_freecount, -1); |
1da177e4 | 975 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); |
44b56e0a | 976 | pag->pagi_freecount--; |
1da177e4 | 977 | |
0b48db80 DC |
978 | error = xfs_check_agi_freecount(cur, agi); |
979 | if (error) | |
980 | goto error0; | |
981 | ||
1da177e4 LT |
982 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
983 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1); | |
44b56e0a | 984 | xfs_perag_put(pag); |
1da177e4 LT |
985 | *inop = ino; |
986 | return 0; | |
987 | error1: | |
988 | xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR); | |
989 | error0: | |
990 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
44b56e0a | 991 | xfs_perag_put(pag); |
1da177e4 LT |
992 | return error; |
993 | } | |
994 | ||
6dd8638e BF |
995 | /* |
996 | * Use the free inode btree to allocate an inode based on distance from the | |
997 | * parent. Note that the provided cursor may be deleted and replaced. | |
998 | */ | |
999 | STATIC int | |
1000 | xfs_dialloc_ag_finobt_near( | |
1001 | xfs_agino_t pagino, | |
1002 | struct xfs_btree_cur **ocur, | |
1003 | struct xfs_inobt_rec_incore *rec) | |
1004 | { | |
1005 | struct xfs_btree_cur *lcur = *ocur; /* left search cursor */ | |
1006 | struct xfs_btree_cur *rcur; /* right search cursor */ | |
1007 | struct xfs_inobt_rec_incore rrec; | |
1008 | int error; | |
1009 | int i, j; | |
1010 | ||
1011 | error = xfs_inobt_lookup(lcur, pagino, XFS_LOOKUP_LE, &i); | |
1012 | if (error) | |
1013 | return error; | |
1014 | ||
1015 | if (i == 1) { | |
1016 | error = xfs_inobt_get_rec(lcur, rec, &i); | |
1017 | if (error) | |
1018 | return error; | |
1019 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
1020 | ||
1021 | /* | |
1022 | * See if we've landed in the parent inode record. The finobt | |
1023 | * only tracks chunks with at least one free inode, so record | |
1024 | * existence is enough. | |
1025 | */ | |
1026 | if (pagino >= rec->ir_startino && | |
1027 | pagino < (rec->ir_startino + XFS_INODES_PER_CHUNK)) | |
1028 | return 0; | |
1029 | } | |
1030 | ||
1031 | error = xfs_btree_dup_cursor(lcur, &rcur); | |
1032 | if (error) | |
1033 | return error; | |
1034 | ||
1035 | error = xfs_inobt_lookup(rcur, pagino, XFS_LOOKUP_GE, &j); | |
1036 | if (error) | |
1037 | goto error_rcur; | |
1038 | if (j == 1) { | |
1039 | error = xfs_inobt_get_rec(rcur, &rrec, &j); | |
1040 | if (error) | |
1041 | goto error_rcur; | |
c29aad41 | 1042 | XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, j == 1, error_rcur); |
6dd8638e BF |
1043 | } |
1044 | ||
c29aad41 | 1045 | XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, i == 1 || j == 1, error_rcur); |
6dd8638e BF |
1046 | if (i == 1 && j == 1) { |
1047 | /* | |
1048 | * Both the left and right records are valid. Choose the closer | |
1049 | * inode chunk to the target. | |
1050 | */ | |
1051 | if ((pagino - rec->ir_startino + XFS_INODES_PER_CHUNK - 1) > | |
1052 | (rrec.ir_startino - pagino)) { | |
1053 | *rec = rrec; | |
1054 | xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR); | |
1055 | *ocur = rcur; | |
1056 | } else { | |
1057 | xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR); | |
1058 | } | |
1059 | } else if (j == 1) { | |
1060 | /* only the right record is valid */ | |
1061 | *rec = rrec; | |
1062 | xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR); | |
1063 | *ocur = rcur; | |
1064 | } else if (i == 1) { | |
1065 | /* only the left record is valid */ | |
1066 | xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR); | |
1067 | } | |
1068 | ||
1069 | return 0; | |
1070 | ||
1071 | error_rcur: | |
1072 | xfs_btree_del_cursor(rcur, XFS_BTREE_ERROR); | |
1073 | return error; | |
1074 | } | |
1075 | ||
1076 | /* | |
1077 | * Use the free inode btree to find a free inode based on a newino hint. If | |
1078 | * the hint is NULL, find the first free inode in the AG. | |
1079 | */ | |
1080 | STATIC int | |
1081 | xfs_dialloc_ag_finobt_newino( | |
1082 | struct xfs_agi *agi, | |
1083 | struct xfs_btree_cur *cur, | |
1084 | struct xfs_inobt_rec_incore *rec) | |
1085 | { | |
1086 | int error; | |
1087 | int i; | |
1088 | ||
1089 | if (agi->agi_newino != cpu_to_be32(NULLAGINO)) { | |
e68ed775 DC |
1090 | error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino), |
1091 | XFS_LOOKUP_EQ, &i); | |
6dd8638e BF |
1092 | if (error) |
1093 | return error; | |
1094 | if (i == 1) { | |
1095 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1096 | if (error) | |
1097 | return error; | |
1098 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
6dd8638e BF |
1099 | return 0; |
1100 | } | |
1101 | } | |
1102 | ||
1103 | /* | |
1104 | * Find the first inode available in the AG. | |
1105 | */ | |
1106 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); | |
1107 | if (error) | |
1108 | return error; | |
1109 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
1110 | ||
1111 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1112 | if (error) | |
1113 | return error; | |
1114 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
1115 | ||
1116 | return 0; | |
1117 | } | |
1118 | ||
1119 | /* | |
1120 | * Update the inobt based on a modification made to the finobt. Also ensure that | |
1121 | * the records from both trees are equivalent post-modification. | |
1122 | */ | |
1123 | STATIC int | |
1124 | xfs_dialloc_ag_update_inobt( | |
1125 | struct xfs_btree_cur *cur, /* inobt cursor */ | |
1126 | struct xfs_inobt_rec_incore *frec, /* finobt record */ | |
1127 | int offset) /* inode offset */ | |
1128 | { | |
1129 | struct xfs_inobt_rec_incore rec; | |
1130 | int error; | |
1131 | int i; | |
1132 | ||
1133 | error = xfs_inobt_lookup(cur, frec->ir_startino, XFS_LOOKUP_EQ, &i); | |
1134 | if (error) | |
1135 | return error; | |
1136 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
1137 | ||
1138 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
1139 | if (error) | |
1140 | return error; | |
1141 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
1142 | ASSERT((XFS_AGINO_TO_OFFSET(cur->bc_mp, rec.ir_startino) % | |
1143 | XFS_INODES_PER_CHUNK) == 0); | |
1144 | ||
1145 | rec.ir_free &= ~XFS_INOBT_MASK(offset); | |
1146 | rec.ir_freecount--; | |
1147 | ||
1148 | XFS_WANT_CORRUPTED_RETURN((rec.ir_free == frec->ir_free) && | |
1149 | (rec.ir_freecount == frec->ir_freecount)); | |
1150 | ||
b72091f2 | 1151 | return xfs_inobt_update(cur, &rec); |
6dd8638e BF |
1152 | } |
1153 | ||
1154 | /* | |
1155 | * Allocate an inode using the free inode btree, if available. Otherwise, fall | |
1156 | * back to the inobt search algorithm. | |
1157 | * | |
1158 | * The caller selected an AG for us, and made sure that free inodes are | |
1159 | * available. | |
1160 | */ | |
1161 | STATIC int | |
1162 | xfs_dialloc_ag( | |
1163 | struct xfs_trans *tp, | |
1164 | struct xfs_buf *agbp, | |
1165 | xfs_ino_t parent, | |
1166 | xfs_ino_t *inop) | |
1167 | { | |
1168 | struct xfs_mount *mp = tp->t_mountp; | |
1169 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
1170 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
1171 | xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent); | |
1172 | xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent); | |
1173 | struct xfs_perag *pag; | |
1174 | struct xfs_btree_cur *cur; /* finobt cursor */ | |
1175 | struct xfs_btree_cur *icur; /* inobt cursor */ | |
1176 | struct xfs_inobt_rec_incore rec; | |
1177 | xfs_ino_t ino; | |
1178 | int error; | |
1179 | int offset; | |
1180 | int i; | |
1181 | ||
1182 | if (!xfs_sb_version_hasfinobt(&mp->m_sb)) | |
1183 | return xfs_dialloc_ag_inobt(tp, agbp, parent, inop); | |
1184 | ||
1185 | pag = xfs_perag_get(mp, agno); | |
1186 | ||
1187 | /* | |
1188 | * If pagino is 0 (this is the root inode allocation) use newino. | |
1189 | * This must work because we've just allocated some. | |
1190 | */ | |
1191 | if (!pagino) | |
1192 | pagino = be32_to_cpu(agi->agi_newino); | |
1193 | ||
1194 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO); | |
1195 | ||
1196 | error = xfs_check_agi_freecount(cur, agi); | |
1197 | if (error) | |
1198 | goto error_cur; | |
1199 | ||
1200 | /* | |
1201 | * The search algorithm depends on whether we're in the same AG as the | |
1202 | * parent. If so, find the closest available inode to the parent. If | |
1203 | * not, consider the agi hint or find the first free inode in the AG. | |
1204 | */ | |
1205 | if (agno == pagno) | |
1206 | error = xfs_dialloc_ag_finobt_near(pagino, &cur, &rec); | |
1207 | else | |
1208 | error = xfs_dialloc_ag_finobt_newino(agi, cur, &rec); | |
1209 | if (error) | |
1210 | goto error_cur; | |
1211 | ||
1212 | offset = xfs_lowbit64(rec.ir_free); | |
1213 | ASSERT(offset >= 0); | |
1214 | ASSERT(offset < XFS_INODES_PER_CHUNK); | |
1215 | ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) % | |
1216 | XFS_INODES_PER_CHUNK) == 0); | |
1217 | ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset); | |
1218 | ||
1219 | /* | |
1220 | * Modify or remove the finobt record. | |
1221 | */ | |
1222 | rec.ir_free &= ~XFS_INOBT_MASK(offset); | |
1223 | rec.ir_freecount--; | |
1224 | if (rec.ir_freecount) | |
1225 | error = xfs_inobt_update(cur, &rec); | |
1226 | else | |
1227 | error = xfs_btree_delete(cur, &i); | |
1228 | if (error) | |
1229 | goto error_cur; | |
1230 | ||
1231 | /* | |
1232 | * The finobt has now been updated appropriately. We haven't updated the | |
1233 | * agi and superblock yet, so we can create an inobt cursor and validate | |
1234 | * the original freecount. If all is well, make the equivalent update to | |
1235 | * the inobt using the finobt record and offset information. | |
1236 | */ | |
1237 | icur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); | |
1238 | ||
1239 | error = xfs_check_agi_freecount(icur, agi); | |
1240 | if (error) | |
1241 | goto error_icur; | |
1242 | ||
1243 | error = xfs_dialloc_ag_update_inobt(icur, &rec, offset); | |
1244 | if (error) | |
1245 | goto error_icur; | |
1246 | ||
1247 | /* | |
1248 | * Both trees have now been updated. We must update the perag and | |
1249 | * superblock before we can check the freecount for each btree. | |
1250 | */ | |
1251 | be32_add_cpu(&agi->agi_freecount, -1); | |
1252 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); | |
1253 | pag->pagi_freecount--; | |
1254 | ||
1255 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1); | |
1256 | ||
1257 | error = xfs_check_agi_freecount(icur, agi); | |
1258 | if (error) | |
1259 | goto error_icur; | |
1260 | error = xfs_check_agi_freecount(cur, agi); | |
1261 | if (error) | |
1262 | goto error_icur; | |
1263 | ||
1264 | xfs_btree_del_cursor(icur, XFS_BTREE_NOERROR); | |
1265 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
1266 | xfs_perag_put(pag); | |
1267 | *inop = ino; | |
1268 | return 0; | |
1269 | ||
1270 | error_icur: | |
1271 | xfs_btree_del_cursor(icur, XFS_BTREE_ERROR); | |
1272 | error_cur: | |
1273 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
1274 | xfs_perag_put(pag); | |
1275 | return error; | |
1276 | } | |
1277 | ||
f2ecc5e4 CH |
1278 | /* |
1279 | * Allocate an inode on disk. | |
1280 | * | |
1281 | * Mode is used to tell whether the new inode will need space, and whether it | |
1282 | * is a directory. | |
1283 | * | |
1284 | * This function is designed to be called twice if it has to do an allocation | |
1285 | * to make more free inodes. On the first call, *IO_agbp should be set to NULL. | |
1286 | * If an inode is available without having to performn an allocation, an inode | |
cd856db6 CM |
1287 | * number is returned. In this case, *IO_agbp is set to NULL. If an allocation |
1288 | * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp. | |
1289 | * The caller should then commit the current transaction, allocate a | |
f2ecc5e4 CH |
1290 | * new transaction, and call xfs_dialloc() again, passing in the previous value |
1291 | * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI | |
1292 | * buffer is locked across the two calls, the second call is guaranteed to have | |
1293 | * a free inode available. | |
1294 | * | |
1295 | * Once we successfully pick an inode its number is returned and the on-disk | |
1296 | * data structures are updated. The inode itself is not read in, since doing so | |
1297 | * would break ordering constraints with xfs_reclaim. | |
1298 | */ | |
1299 | int | |
1300 | xfs_dialloc( | |
1301 | struct xfs_trans *tp, | |
1302 | xfs_ino_t parent, | |
1303 | umode_t mode, | |
1304 | int okalloc, | |
1305 | struct xfs_buf **IO_agbp, | |
f2ecc5e4 CH |
1306 | xfs_ino_t *inop) |
1307 | { | |
55d6af64 | 1308 | struct xfs_mount *mp = tp->t_mountp; |
f2ecc5e4 CH |
1309 | struct xfs_buf *agbp; |
1310 | xfs_agnumber_t agno; | |
f2ecc5e4 CH |
1311 | int error; |
1312 | int ialloced; | |
1313 | int noroom = 0; | |
be60fe54 | 1314 | xfs_agnumber_t start_agno; |
f2ecc5e4 CH |
1315 | struct xfs_perag *pag; |
1316 | ||
4bb61069 | 1317 | if (*IO_agbp) { |
f2ecc5e4 | 1318 | /* |
4bb61069 CH |
1319 | * If the caller passes in a pointer to the AGI buffer, |
1320 | * continue where we left off before. In this case, we | |
f2ecc5e4 CH |
1321 | * know that the allocation group has free inodes. |
1322 | */ | |
1323 | agbp = *IO_agbp; | |
4bb61069 | 1324 | goto out_alloc; |
f2ecc5e4 | 1325 | } |
4bb61069 CH |
1326 | |
1327 | /* | |
1328 | * We do not have an agbp, so select an initial allocation | |
1329 | * group for inode allocation. | |
1330 | */ | |
be60fe54 CH |
1331 | start_agno = xfs_ialloc_ag_select(tp, parent, mode, okalloc); |
1332 | if (start_agno == NULLAGNUMBER) { | |
4bb61069 CH |
1333 | *inop = NULLFSINO; |
1334 | return 0; | |
1335 | } | |
55d6af64 | 1336 | |
f2ecc5e4 CH |
1337 | /* |
1338 | * If we have already hit the ceiling of inode blocks then clear | |
1339 | * okalloc so we scan all available agi structures for a free | |
1340 | * inode. | |
1341 | */ | |
f2ecc5e4 | 1342 | if (mp->m_maxicount && |
71783438 | 1343 | mp->m_sb.sb_icount + mp->m_ialloc_inos > mp->m_maxicount) { |
f2ecc5e4 CH |
1344 | noroom = 1; |
1345 | okalloc = 0; | |
1346 | } | |
1347 | ||
1348 | /* | |
1349 | * Loop until we find an allocation group that either has free inodes | |
1350 | * or in which we can allocate some inodes. Iterate through the | |
1351 | * allocation groups upward, wrapping at the end. | |
1352 | */ | |
be60fe54 CH |
1353 | agno = start_agno; |
1354 | for (;;) { | |
1355 | pag = xfs_perag_get(mp, agno); | |
1356 | if (!pag->pagi_inodeok) { | |
1357 | xfs_ialloc_next_ag(mp); | |
1358 | goto nextag; | |
1359 | } | |
1360 | ||
1361 | if (!pag->pagi_init) { | |
1362 | error = xfs_ialloc_pagi_init(mp, tp, agno); | |
1363 | if (error) | |
1364 | goto out_error; | |
f2ecc5e4 | 1365 | } |
be60fe54 | 1366 | |
f2ecc5e4 | 1367 | /* |
be60fe54 | 1368 | * Do a first racy fast path check if this AG is usable. |
f2ecc5e4 | 1369 | */ |
be60fe54 CH |
1370 | if (!pag->pagi_freecount && !okalloc) |
1371 | goto nextag; | |
1372 | ||
c4982110 CH |
1373 | /* |
1374 | * Then read in the AGI buffer and recheck with the AGI buffer | |
1375 | * lock held. | |
1376 | */ | |
be60fe54 CH |
1377 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); |
1378 | if (error) | |
1379 | goto out_error; | |
1380 | ||
be60fe54 CH |
1381 | if (pag->pagi_freecount) { |
1382 | xfs_perag_put(pag); | |
1383 | goto out_alloc; | |
1384 | } | |
1385 | ||
c4982110 CH |
1386 | if (!okalloc) |
1387 | goto nextag_relse_buffer; | |
1388 | ||
be60fe54 CH |
1389 | |
1390 | error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced); | |
1391 | if (error) { | |
1392 | xfs_trans_brelse(tp, agbp); | |
1393 | ||
2451337d | 1394 | if (error != -ENOSPC) |
be60fe54 CH |
1395 | goto out_error; |
1396 | ||
1397 | xfs_perag_put(pag); | |
f2ecc5e4 | 1398 | *inop = NULLFSINO; |
be60fe54 | 1399 | return 0; |
f2ecc5e4 | 1400 | } |
be60fe54 CH |
1401 | |
1402 | if (ialloced) { | |
1403 | /* | |
1404 | * We successfully allocated some inodes, return | |
1405 | * the current context to the caller so that it | |
1406 | * can commit the current transaction and call | |
1407 | * us again where we left off. | |
1408 | */ | |
1409 | ASSERT(pag->pagi_freecount > 0); | |
f2ecc5e4 | 1410 | xfs_perag_put(pag); |
be60fe54 CH |
1411 | |
1412 | *IO_agbp = agbp; | |
1413 | *inop = NULLFSINO; | |
1414 | return 0; | |
f2ecc5e4 | 1415 | } |
be60fe54 | 1416 | |
c4982110 CH |
1417 | nextag_relse_buffer: |
1418 | xfs_trans_brelse(tp, agbp); | |
be60fe54 | 1419 | nextag: |
f2ecc5e4 | 1420 | xfs_perag_put(pag); |
be60fe54 CH |
1421 | if (++agno == mp->m_sb.sb_agcount) |
1422 | agno = 0; | |
1423 | if (agno == start_agno) { | |
1424 | *inop = NULLFSINO; | |
2451337d | 1425 | return noroom ? -ENOSPC : 0; |
be60fe54 | 1426 | } |
f2ecc5e4 CH |
1427 | } |
1428 | ||
4bb61069 | 1429 | out_alloc: |
f2ecc5e4 CH |
1430 | *IO_agbp = NULL; |
1431 | return xfs_dialloc_ag(tp, agbp, parent, inop); | |
be60fe54 CH |
1432 | out_error: |
1433 | xfs_perag_put(pag); | |
b474c7ae | 1434 | return error; |
f2ecc5e4 CH |
1435 | } |
1436 | ||
2b64ee5c BF |
1437 | STATIC int |
1438 | xfs_difree_inobt( | |
1439 | struct xfs_mount *mp, | |
1440 | struct xfs_trans *tp, | |
1441 | struct xfs_buf *agbp, | |
1442 | xfs_agino_t agino, | |
1443 | struct xfs_bmap_free *flist, | |
0d907a3b | 1444 | int *deleted, |
2b64ee5c BF |
1445 | xfs_ino_t *first_ino, |
1446 | struct xfs_inobt_rec_incore *orec) | |
1da177e4 | 1447 | { |
2b64ee5c BF |
1448 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); |
1449 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
1450 | struct xfs_perag *pag; | |
1451 | struct xfs_btree_cur *cur; | |
1452 | struct xfs_inobt_rec_incore rec; | |
1453 | int ilen; | |
1454 | int error; | |
1455 | int i; | |
1456 | int off; | |
1da177e4 | 1457 | |
69ef921b | 1458 | ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC)); |
2b64ee5c BF |
1459 | ASSERT(XFS_AGINO_TO_AGBNO(mp, agino) < be32_to_cpu(agi->agi_length)); |
1460 | ||
1da177e4 LT |
1461 | /* |
1462 | * Initialize the cursor. | |
1463 | */ | |
57bd3dbe | 1464 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
1da177e4 | 1465 | |
0b48db80 DC |
1466 | error = xfs_check_agi_freecount(cur, agi); |
1467 | if (error) | |
1468 | goto error0; | |
1469 | ||
1da177e4 LT |
1470 | /* |
1471 | * Look for the entry describing this inode. | |
1472 | */ | |
21875505 | 1473 | if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) { |
0b932ccc DC |
1474 | xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.", |
1475 | __func__, error); | |
1da177e4 LT |
1476 | goto error0; |
1477 | } | |
c29aad41 | 1478 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
2e287a73 CH |
1479 | error = xfs_inobt_get_rec(cur, &rec, &i); |
1480 | if (error) { | |
0b932ccc DC |
1481 | xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.", |
1482 | __func__, error); | |
1da177e4 LT |
1483 | goto error0; |
1484 | } | |
c29aad41 | 1485 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 LT |
1486 | /* |
1487 | * Get the offset in the inode chunk. | |
1488 | */ | |
1489 | off = agino - rec.ir_startino; | |
1490 | ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK); | |
0d87e656 | 1491 | ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off))); |
1da177e4 LT |
1492 | /* |
1493 | * Mark the inode free & increment the count. | |
1494 | */ | |
0d87e656 | 1495 | rec.ir_free |= XFS_INOBT_MASK(off); |
1da177e4 LT |
1496 | rec.ir_freecount++; |
1497 | ||
1498 | /* | |
c41564b5 | 1499 | * When an inode cluster is free, it becomes eligible for removal |
1da177e4 | 1500 | */ |
1bd960ee | 1501 | if (!(mp->m_flags & XFS_MOUNT_IKEEP) && |
71783438 | 1502 | (rec.ir_freecount == mp->m_ialloc_inos)) { |
1da177e4 | 1503 | |
376c2f3a | 1504 | *deleted = 1; |
1da177e4 LT |
1505 | *first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino); |
1506 | ||
1507 | /* | |
1508 | * Remove the inode cluster from the AGI B+Tree, adjust the | |
1509 | * AGI and Superblock inode counts, and mark the disk space | |
1510 | * to be freed when the transaction is committed. | |
1511 | */ | |
71783438 | 1512 | ilen = mp->m_ialloc_inos; |
413d57c9 MS |
1513 | be32_add_cpu(&agi->agi_count, -ilen); |
1514 | be32_add_cpu(&agi->agi_freecount, -(ilen - 1)); | |
1da177e4 | 1515 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT); |
44b56e0a DC |
1516 | pag = xfs_perag_get(mp, agno); |
1517 | pag->pagi_freecount -= ilen - 1; | |
1518 | xfs_perag_put(pag); | |
1da177e4 LT |
1519 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen); |
1520 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1)); | |
1521 | ||
91cca5df | 1522 | if ((error = xfs_btree_delete(cur, &i))) { |
0b932ccc DC |
1523 | xfs_warn(mp, "%s: xfs_btree_delete returned error %d.", |
1524 | __func__, error); | |
1da177e4 LT |
1525 | goto error0; |
1526 | } | |
1527 | ||
126cd105 JL |
1528 | xfs_bmap_add_free(XFS_AGB_TO_FSB(mp, agno, |
1529 | XFS_AGINO_TO_AGBNO(mp, rec.ir_startino)), | |
1530 | mp->m_ialloc_blks, flist, mp); | |
1da177e4 | 1531 | } else { |
376c2f3a | 1532 | *deleted = 0; |
1da177e4 | 1533 | |
afabc24a CH |
1534 | error = xfs_inobt_update(cur, &rec); |
1535 | if (error) { | |
0b932ccc DC |
1536 | xfs_warn(mp, "%s: xfs_inobt_update returned error %d.", |
1537 | __func__, error); | |
1da177e4 LT |
1538 | goto error0; |
1539 | } | |
afabc24a | 1540 | |
1da177e4 LT |
1541 | /* |
1542 | * Change the inode free counts and log the ag/sb changes. | |
1543 | */ | |
413d57c9 | 1544 | be32_add_cpu(&agi->agi_freecount, 1); |
1da177e4 | 1545 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); |
44b56e0a DC |
1546 | pag = xfs_perag_get(mp, agno); |
1547 | pag->pagi_freecount++; | |
1548 | xfs_perag_put(pag); | |
1da177e4 LT |
1549 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1); |
1550 | } | |
1551 | ||
0b48db80 DC |
1552 | error = xfs_check_agi_freecount(cur, agi); |
1553 | if (error) | |
1554 | goto error0; | |
1da177e4 | 1555 | |
2b64ee5c | 1556 | *orec = rec; |
1da177e4 LT |
1557 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
1558 | return 0; | |
1559 | ||
1560 | error0: | |
1561 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
1562 | return error; | |
1563 | } | |
1564 | ||
3efa4ffd BF |
1565 | /* |
1566 | * Free an inode in the free inode btree. | |
1567 | */ | |
1568 | STATIC int | |
1569 | xfs_difree_finobt( | |
1570 | struct xfs_mount *mp, | |
1571 | struct xfs_trans *tp, | |
1572 | struct xfs_buf *agbp, | |
1573 | xfs_agino_t agino, | |
1574 | struct xfs_inobt_rec_incore *ibtrec) /* inobt record */ | |
1575 | { | |
1576 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
1577 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
1578 | struct xfs_btree_cur *cur; | |
1579 | struct xfs_inobt_rec_incore rec; | |
1580 | int offset = agino - ibtrec->ir_startino; | |
1581 | int error; | |
1582 | int i; | |
1583 | ||
1584 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO); | |
1585 | ||
1586 | error = xfs_inobt_lookup(cur, ibtrec->ir_startino, XFS_LOOKUP_EQ, &i); | |
1587 | if (error) | |
1588 | goto error; | |
1589 | if (i == 0) { | |
1590 | /* | |
1591 | * If the record does not exist in the finobt, we must have just | |
1592 | * freed an inode in a previously fully allocated chunk. If not, | |
1593 | * something is out of sync. | |
1594 | */ | |
c29aad41 | 1595 | XFS_WANT_CORRUPTED_GOTO(mp, ibtrec->ir_freecount == 1, error); |
3efa4ffd BF |
1596 | |
1597 | error = xfs_inobt_insert_rec(cur, ibtrec->ir_freecount, | |
1598 | ibtrec->ir_free, &i); | |
1599 | if (error) | |
1600 | goto error; | |
1601 | ASSERT(i == 1); | |
1602 | ||
1603 | goto out; | |
1604 | } | |
1605 | ||
1606 | /* | |
1607 | * Read and update the existing record. We could just copy the ibtrec | |
1608 | * across here, but that would defeat the purpose of having redundant | |
1609 | * metadata. By making the modifications independently, we can catch | |
1610 | * corruptions that we wouldn't see if we just copied from one record | |
1611 | * to another. | |
1612 | */ | |
1613 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
1614 | if (error) | |
1615 | goto error; | |
c29aad41 | 1616 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error); |
3efa4ffd BF |
1617 | |
1618 | rec.ir_free |= XFS_INOBT_MASK(offset); | |
1619 | rec.ir_freecount++; | |
1620 | ||
c29aad41 | 1621 | XFS_WANT_CORRUPTED_GOTO(mp, (rec.ir_free == ibtrec->ir_free) && |
3efa4ffd BF |
1622 | (rec.ir_freecount == ibtrec->ir_freecount), |
1623 | error); | |
1624 | ||
1625 | /* | |
1626 | * The content of inobt records should always match between the inobt | |
1627 | * and finobt. The lifecycle of records in the finobt is different from | |
1628 | * the inobt in that the finobt only tracks records with at least one | |
1629 | * free inode. Hence, if all of the inodes are free and we aren't | |
1630 | * keeping inode chunks permanently on disk, remove the record. | |
1631 | * Otherwise, update the record with the new information. | |
1632 | */ | |
1633 | if (rec.ir_freecount == mp->m_ialloc_inos && | |
1634 | !(mp->m_flags & XFS_MOUNT_IKEEP)) { | |
1635 | error = xfs_btree_delete(cur, &i); | |
1636 | if (error) | |
1637 | goto error; | |
1638 | ASSERT(i == 1); | |
1639 | } else { | |
1640 | error = xfs_inobt_update(cur, &rec); | |
1641 | if (error) | |
1642 | goto error; | |
1643 | } | |
1644 | ||
1645 | out: | |
1646 | error = xfs_check_agi_freecount(cur, agi); | |
1647 | if (error) | |
1648 | goto error; | |
1649 | ||
1650 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
1651 | return 0; | |
1652 | ||
1653 | error: | |
1654 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
1655 | return error; | |
1656 | } | |
1657 | ||
2b64ee5c BF |
1658 | /* |
1659 | * Free disk inode. Carefully avoids touching the incore inode, all | |
1660 | * manipulations incore are the caller's responsibility. | |
1661 | * The on-disk inode is not changed by this operation, only the | |
1662 | * btree (free inode mask) is changed. | |
1663 | */ | |
1664 | int | |
1665 | xfs_difree( | |
1666 | struct xfs_trans *tp, /* transaction pointer */ | |
1667 | xfs_ino_t inode, /* inode to be freed */ | |
1668 | struct xfs_bmap_free *flist, /* extents to free */ | |
0d907a3b | 1669 | int *deleted,/* set if inode cluster was deleted */ |
2b64ee5c BF |
1670 | xfs_ino_t *first_ino)/* first inode in deleted cluster */ |
1671 | { | |
1672 | /* REFERENCED */ | |
1673 | xfs_agblock_t agbno; /* block number containing inode */ | |
1674 | struct xfs_buf *agbp; /* buffer for allocation group header */ | |
1675 | xfs_agino_t agino; /* allocation group inode number */ | |
1676 | xfs_agnumber_t agno; /* allocation group number */ | |
1677 | int error; /* error return value */ | |
1678 | struct xfs_mount *mp; /* mount structure for filesystem */ | |
1679 | struct xfs_inobt_rec_incore rec;/* btree record */ | |
1680 | ||
1681 | mp = tp->t_mountp; | |
1682 | ||
1683 | /* | |
1684 | * Break up inode number into its components. | |
1685 | */ | |
1686 | agno = XFS_INO_TO_AGNO(mp, inode); | |
1687 | if (agno >= mp->m_sb.sb_agcount) { | |
1688 | xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).", | |
1689 | __func__, agno, mp->m_sb.sb_agcount); | |
1690 | ASSERT(0); | |
2451337d | 1691 | return -EINVAL; |
2b64ee5c BF |
1692 | } |
1693 | agino = XFS_INO_TO_AGINO(mp, inode); | |
1694 | if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
1695 | xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).", | |
1696 | __func__, (unsigned long long)inode, | |
1697 | (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino)); | |
1698 | ASSERT(0); | |
2451337d | 1699 | return -EINVAL; |
2b64ee5c BF |
1700 | } |
1701 | agbno = XFS_AGINO_TO_AGBNO(mp, agino); | |
1702 | if (agbno >= mp->m_sb.sb_agblocks) { | |
1703 | xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).", | |
1704 | __func__, agbno, mp->m_sb.sb_agblocks); | |
1705 | ASSERT(0); | |
2451337d | 1706 | return -EINVAL; |
2b64ee5c BF |
1707 | } |
1708 | /* | |
1709 | * Get the allocation group header. | |
1710 | */ | |
1711 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); | |
1712 | if (error) { | |
1713 | xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.", | |
1714 | __func__, error); | |
1715 | return error; | |
1716 | } | |
1717 | ||
1718 | /* | |
1719 | * Fix up the inode allocation btree. | |
1720 | */ | |
0d907a3b | 1721 | error = xfs_difree_inobt(mp, tp, agbp, agino, flist, deleted, first_ino, |
2b64ee5c BF |
1722 | &rec); |
1723 | if (error) | |
1724 | goto error0; | |
1725 | ||
3efa4ffd BF |
1726 | /* |
1727 | * Fix up the free inode btree. | |
1728 | */ | |
1729 | if (xfs_sb_version_hasfinobt(&mp->m_sb)) { | |
1730 | error = xfs_difree_finobt(mp, tp, agbp, agino, &rec); | |
1731 | if (error) | |
1732 | goto error0; | |
1733 | } | |
1734 | ||
2b64ee5c BF |
1735 | return 0; |
1736 | ||
1737 | error0: | |
1738 | return error; | |
1739 | } | |
1740 | ||
7124fe0a DC |
1741 | STATIC int |
1742 | xfs_imap_lookup( | |
1743 | struct xfs_mount *mp, | |
1744 | struct xfs_trans *tp, | |
1745 | xfs_agnumber_t agno, | |
1746 | xfs_agino_t agino, | |
1747 | xfs_agblock_t agbno, | |
1748 | xfs_agblock_t *chunk_agbno, | |
1749 | xfs_agblock_t *offset_agbno, | |
1750 | int flags) | |
1751 | { | |
1752 | struct xfs_inobt_rec_incore rec; | |
1753 | struct xfs_btree_cur *cur; | |
1754 | struct xfs_buf *agbp; | |
7124fe0a DC |
1755 | int error; |
1756 | int i; | |
1757 | ||
1758 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); | |
1759 | if (error) { | |
53487786 DC |
1760 | xfs_alert(mp, |
1761 | "%s: xfs_ialloc_read_agi() returned error %d, agno %d", | |
1762 | __func__, error, agno); | |
7124fe0a DC |
1763 | return error; |
1764 | } | |
1765 | ||
1766 | /* | |
4536f2ad DC |
1767 | * Lookup the inode record for the given agino. If the record cannot be |
1768 | * found, then it's an invalid inode number and we should abort. Once | |
1769 | * we have a record, we need to ensure it contains the inode number | |
1770 | * we are looking up. | |
7124fe0a | 1771 | */ |
57bd3dbe | 1772 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
4536f2ad | 1773 | error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i); |
7124fe0a DC |
1774 | if (!error) { |
1775 | if (i) | |
1776 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
1777 | if (!error && i == 0) | |
2451337d | 1778 | error = -EINVAL; |
7124fe0a DC |
1779 | } |
1780 | ||
1781 | xfs_trans_brelse(tp, agbp); | |
1782 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
1783 | if (error) | |
1784 | return error; | |
1785 | ||
4536f2ad DC |
1786 | /* check that the returned record contains the required inode */ |
1787 | if (rec.ir_startino > agino || | |
71783438 | 1788 | rec.ir_startino + mp->m_ialloc_inos <= agino) |
2451337d | 1789 | return -EINVAL; |
4536f2ad | 1790 | |
7124fe0a | 1791 | /* for untrusted inodes check it is allocated first */ |
1920779e | 1792 | if ((flags & XFS_IGET_UNTRUSTED) && |
7124fe0a | 1793 | (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino))) |
2451337d | 1794 | return -EINVAL; |
7124fe0a DC |
1795 | |
1796 | *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino); | |
1797 | *offset_agbno = agbno - *chunk_agbno; | |
1798 | return 0; | |
1799 | } | |
1800 | ||
1da177e4 | 1801 | /* |
94e1b69d | 1802 | * Return the location of the inode in imap, for mapping it into a buffer. |
1da177e4 | 1803 | */ |
1da177e4 | 1804 | int |
94e1b69d CH |
1805 | xfs_imap( |
1806 | xfs_mount_t *mp, /* file system mount structure */ | |
1807 | xfs_trans_t *tp, /* transaction pointer */ | |
1da177e4 | 1808 | xfs_ino_t ino, /* inode to locate */ |
94e1b69d CH |
1809 | struct xfs_imap *imap, /* location map structure */ |
1810 | uint flags) /* flags for inode btree lookup */ | |
1da177e4 LT |
1811 | { |
1812 | xfs_agblock_t agbno; /* block number of inode in the alloc group */ | |
1da177e4 LT |
1813 | xfs_agino_t agino; /* inode number within alloc group */ |
1814 | xfs_agnumber_t agno; /* allocation group number */ | |
1815 | int blks_per_cluster; /* num blocks per inode cluster */ | |
1816 | xfs_agblock_t chunk_agbno; /* first block in inode chunk */ | |
1da177e4 | 1817 | xfs_agblock_t cluster_agbno; /* first block in inode cluster */ |
1da177e4 | 1818 | int error; /* error code */ |
1da177e4 | 1819 | int offset; /* index of inode in its buffer */ |
836a94ad | 1820 | xfs_agblock_t offset_agbno; /* blks from chunk start to inode */ |
1da177e4 LT |
1821 | |
1822 | ASSERT(ino != NULLFSINO); | |
94e1b69d | 1823 | |
1da177e4 LT |
1824 | /* |
1825 | * Split up the inode number into its parts. | |
1826 | */ | |
1827 | agno = XFS_INO_TO_AGNO(mp, ino); | |
1828 | agino = XFS_INO_TO_AGINO(mp, ino); | |
1829 | agbno = XFS_AGINO_TO_AGBNO(mp, agino); | |
1830 | if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks || | |
1831 | ino != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
1832 | #ifdef DEBUG | |
1920779e DC |
1833 | /* |
1834 | * Don't output diagnostic information for untrusted inodes | |
1835 | * as they can be invalid without implying corruption. | |
1836 | */ | |
1837 | if (flags & XFS_IGET_UNTRUSTED) | |
2451337d | 1838 | return -EINVAL; |
1da177e4 | 1839 | if (agno >= mp->m_sb.sb_agcount) { |
53487786 DC |
1840 | xfs_alert(mp, |
1841 | "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)", | |
1842 | __func__, agno, mp->m_sb.sb_agcount); | |
1da177e4 LT |
1843 | } |
1844 | if (agbno >= mp->m_sb.sb_agblocks) { | |
53487786 DC |
1845 | xfs_alert(mp, |
1846 | "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)", | |
1847 | __func__, (unsigned long long)agbno, | |
1848 | (unsigned long)mp->m_sb.sb_agblocks); | |
1da177e4 LT |
1849 | } |
1850 | if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
53487786 DC |
1851 | xfs_alert(mp, |
1852 | "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)", | |
1853 | __func__, ino, | |
1854 | XFS_AGINO_TO_INO(mp, agno, agino)); | |
1da177e4 | 1855 | } |
745b1f47 | 1856 | xfs_stack_trace(); |
1da177e4 | 1857 | #endif /* DEBUG */ |
2451337d | 1858 | return -EINVAL; |
1da177e4 | 1859 | } |
94e1b69d | 1860 | |
f9e5abcf | 1861 | blks_per_cluster = xfs_icluster_size_fsb(mp); |
7124fe0a DC |
1862 | |
1863 | /* | |
1864 | * For bulkstat and handle lookups, we have an untrusted inode number | |
1865 | * that we have to verify is valid. We cannot do this just by reading | |
1866 | * the inode buffer as it may have been unlinked and removed leaving | |
1867 | * inodes in stale state on disk. Hence we have to do a btree lookup | |
1868 | * in all cases where an untrusted inode number is passed. | |
1869 | */ | |
1920779e | 1870 | if (flags & XFS_IGET_UNTRUSTED) { |
7124fe0a DC |
1871 | error = xfs_imap_lookup(mp, tp, agno, agino, agbno, |
1872 | &chunk_agbno, &offset_agbno, flags); | |
1873 | if (error) | |
1874 | return error; | |
1875 | goto out_map; | |
1876 | } | |
1877 | ||
94e1b69d CH |
1878 | /* |
1879 | * If the inode cluster size is the same as the blocksize or | |
1880 | * smaller we get to the buffer by simple arithmetics. | |
1881 | */ | |
f9e5abcf | 1882 | if (blks_per_cluster == 1) { |
1da177e4 LT |
1883 | offset = XFS_INO_TO_OFFSET(mp, ino); |
1884 | ASSERT(offset < mp->m_sb.sb_inopblock); | |
94e1b69d CH |
1885 | |
1886 | imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno); | |
1887 | imap->im_len = XFS_FSB_TO_BB(mp, 1); | |
1888 | imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog); | |
1da177e4 LT |
1889 | return 0; |
1890 | } | |
94e1b69d | 1891 | |
94e1b69d CH |
1892 | /* |
1893 | * If the inode chunks are aligned then use simple maths to | |
1894 | * find the location. Otherwise we have to do a btree | |
1895 | * lookup to find the location. | |
1896 | */ | |
1da177e4 LT |
1897 | if (mp->m_inoalign_mask) { |
1898 | offset_agbno = agbno & mp->m_inoalign_mask; | |
1899 | chunk_agbno = agbno - offset_agbno; | |
1900 | } else { | |
7124fe0a DC |
1901 | error = xfs_imap_lookup(mp, tp, agno, agino, agbno, |
1902 | &chunk_agbno, &offset_agbno, flags); | |
1da177e4 LT |
1903 | if (error) |
1904 | return error; | |
1da177e4 | 1905 | } |
94e1b69d | 1906 | |
7124fe0a | 1907 | out_map: |
1da177e4 LT |
1908 | ASSERT(agbno >= chunk_agbno); |
1909 | cluster_agbno = chunk_agbno + | |
1910 | ((offset_agbno / blks_per_cluster) * blks_per_cluster); | |
1911 | offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) + | |
1912 | XFS_INO_TO_OFFSET(mp, ino); | |
94e1b69d CH |
1913 | |
1914 | imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno); | |
1915 | imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster); | |
1916 | imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog); | |
1917 | ||
1918 | /* | |
1919 | * If the inode number maps to a block outside the bounds | |
1920 | * of the file system then return NULL rather than calling | |
1921 | * read_buf and panicing when we get an error from the | |
1922 | * driver. | |
1923 | */ | |
1924 | if ((imap->im_blkno + imap->im_len) > | |
1925 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) { | |
53487786 DC |
1926 | xfs_alert(mp, |
1927 | "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)", | |
1928 | __func__, (unsigned long long) imap->im_blkno, | |
94e1b69d CH |
1929 | (unsigned long long) imap->im_len, |
1930 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)); | |
2451337d | 1931 | return -EINVAL; |
94e1b69d | 1932 | } |
1da177e4 | 1933 | return 0; |
1da177e4 LT |
1934 | } |
1935 | ||
1936 | /* | |
1937 | * Compute and fill in value of m_in_maxlevels. | |
1938 | */ | |
1939 | void | |
1940 | xfs_ialloc_compute_maxlevels( | |
1941 | xfs_mount_t *mp) /* file system mount structure */ | |
1942 | { | |
1943 | int level; | |
1944 | uint maxblocks; | |
1945 | uint maxleafents; | |
1946 | int minleafrecs; | |
1947 | int minnoderecs; | |
1948 | ||
1949 | maxleafents = (1LL << XFS_INO_AGINO_BITS(mp)) >> | |
1950 | XFS_INODES_PER_CHUNK_LOG; | |
1951 | minleafrecs = mp->m_alloc_mnr[0]; | |
1952 | minnoderecs = mp->m_alloc_mnr[1]; | |
1953 | maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs; | |
1954 | for (level = 1; maxblocks > 1; level++) | |
1955 | maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs; | |
1956 | mp->m_in_maxlevels = level; | |
1957 | } | |
1958 | ||
1959 | /* | |
aafc3c24 BF |
1960 | * Log specified fields for the ag hdr (inode section). The growth of the agi |
1961 | * structure over time requires that we interpret the buffer as two logical | |
1962 | * regions delineated by the end of the unlinked list. This is due to the size | |
1963 | * of the hash table and its location in the middle of the agi. | |
1964 | * | |
1965 | * For example, a request to log a field before agi_unlinked and a field after | |
1966 | * agi_unlinked could cause us to log the entire hash table and use an excessive | |
1967 | * amount of log space. To avoid this behavior, log the region up through | |
1968 | * agi_unlinked in one call and the region after agi_unlinked through the end of | |
1969 | * the structure in another. | |
1da177e4 LT |
1970 | */ |
1971 | void | |
1972 | xfs_ialloc_log_agi( | |
1973 | xfs_trans_t *tp, /* transaction pointer */ | |
1974 | xfs_buf_t *bp, /* allocation group header buffer */ | |
1975 | int fields) /* bitmask of fields to log */ | |
1976 | { | |
1977 | int first; /* first byte number */ | |
1978 | int last; /* last byte number */ | |
1979 | static const short offsets[] = { /* field starting offsets */ | |
1980 | /* keep in sync with bit definitions */ | |
1981 | offsetof(xfs_agi_t, agi_magicnum), | |
1982 | offsetof(xfs_agi_t, agi_versionnum), | |
1983 | offsetof(xfs_agi_t, agi_seqno), | |
1984 | offsetof(xfs_agi_t, agi_length), | |
1985 | offsetof(xfs_agi_t, agi_count), | |
1986 | offsetof(xfs_agi_t, agi_root), | |
1987 | offsetof(xfs_agi_t, agi_level), | |
1988 | offsetof(xfs_agi_t, agi_freecount), | |
1989 | offsetof(xfs_agi_t, agi_newino), | |
1990 | offsetof(xfs_agi_t, agi_dirino), | |
1991 | offsetof(xfs_agi_t, agi_unlinked), | |
aafc3c24 BF |
1992 | offsetof(xfs_agi_t, agi_free_root), |
1993 | offsetof(xfs_agi_t, agi_free_level), | |
1da177e4 LT |
1994 | sizeof(xfs_agi_t) |
1995 | }; | |
1996 | #ifdef DEBUG | |
1997 | xfs_agi_t *agi; /* allocation group header */ | |
1998 | ||
1999 | agi = XFS_BUF_TO_AGI(bp); | |
69ef921b | 2000 | ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC)); |
1da177e4 | 2001 | #endif |
aafc3c24 BF |
2002 | |
2003 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_AGI_BUF); | |
2004 | ||
1da177e4 | 2005 | /* |
aafc3c24 BF |
2006 | * Compute byte offsets for the first and last fields in the first |
2007 | * region and log the agi buffer. This only logs up through | |
2008 | * agi_unlinked. | |
1da177e4 | 2009 | */ |
aafc3c24 BF |
2010 | if (fields & XFS_AGI_ALL_BITS_R1) { |
2011 | xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R1, | |
2012 | &first, &last); | |
2013 | xfs_trans_log_buf(tp, bp, first, last); | |
2014 | } | |
2015 | ||
1da177e4 | 2016 | /* |
aafc3c24 BF |
2017 | * Mask off the bits in the first region and calculate the first and |
2018 | * last field offsets for any bits in the second region. | |
1da177e4 | 2019 | */ |
aafc3c24 BF |
2020 | fields &= ~XFS_AGI_ALL_BITS_R1; |
2021 | if (fields) { | |
2022 | xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R2, | |
2023 | &first, &last); | |
2024 | xfs_trans_log_buf(tp, bp, first, last); | |
2025 | } | |
1da177e4 LT |
2026 | } |
2027 | ||
5e1be0fb CH |
2028 | #ifdef DEBUG |
2029 | STATIC void | |
2030 | xfs_check_agi_unlinked( | |
2031 | struct xfs_agi *agi) | |
2032 | { | |
2033 | int i; | |
2034 | ||
2035 | for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++) | |
2036 | ASSERT(agi->agi_unlinked[i]); | |
2037 | } | |
2038 | #else | |
2039 | #define xfs_check_agi_unlinked(agi) | |
2040 | #endif | |
2041 | ||
983d09ff | 2042 | static bool |
612cfbfe | 2043 | xfs_agi_verify( |
3702ce6e DC |
2044 | struct xfs_buf *bp) |
2045 | { | |
2046 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
2047 | struct xfs_agi *agi = XFS_BUF_TO_AGI(bp); | |
3702ce6e | 2048 | |
983d09ff DC |
2049 | if (xfs_sb_version_hascrc(&mp->m_sb) && |
2050 | !uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_uuid)) | |
2051 | return false; | |
3702ce6e DC |
2052 | /* |
2053 | * Validate the magic number of the agi block. | |
2054 | */ | |
983d09ff DC |
2055 | if (agi->agi_magicnum != cpu_to_be32(XFS_AGI_MAGIC)) |
2056 | return false; | |
2057 | if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum))) | |
2058 | return false; | |
3702ce6e | 2059 | |
e1b05723 ES |
2060 | if (be32_to_cpu(agi->agi_level) > XFS_BTREE_MAXLEVELS) |
2061 | return false; | |
3702ce6e DC |
2062 | /* |
2063 | * during growfs operations, the perag is not fully initialised, | |
2064 | * so we can't use it for any useful checking. growfs ensures we can't | |
2065 | * use it by using uncached buffers that don't have the perag attached | |
2066 | * so we can detect and avoid this problem. | |
2067 | */ | |
983d09ff DC |
2068 | if (bp->b_pag && be32_to_cpu(agi->agi_seqno) != bp->b_pag->pag_agno) |
2069 | return false; | |
3702ce6e | 2070 | |
3702ce6e | 2071 | xfs_check_agi_unlinked(agi); |
983d09ff | 2072 | return true; |
612cfbfe DC |
2073 | } |
2074 | ||
1813dd64 DC |
2075 | static void |
2076 | xfs_agi_read_verify( | |
612cfbfe DC |
2077 | struct xfs_buf *bp) |
2078 | { | |
983d09ff | 2079 | struct xfs_mount *mp = bp->b_target->bt_mount; |
983d09ff | 2080 | |
ce5028cf ES |
2081 | if (xfs_sb_version_hascrc(&mp->m_sb) && |
2082 | !xfs_buf_verify_cksum(bp, XFS_AGI_CRC_OFF)) | |
2451337d | 2083 | xfs_buf_ioerror(bp, -EFSBADCRC); |
ce5028cf ES |
2084 | else if (XFS_TEST_ERROR(!xfs_agi_verify(bp), mp, |
2085 | XFS_ERRTAG_IALLOC_READ_AGI, | |
2086 | XFS_RANDOM_IALLOC_READ_AGI)) | |
2451337d | 2087 | xfs_buf_ioerror(bp, -EFSCORRUPTED); |
ce5028cf ES |
2088 | |
2089 | if (bp->b_error) | |
2090 | xfs_verifier_error(bp); | |
612cfbfe DC |
2091 | } |
2092 | ||
b0f539de | 2093 | static void |
1813dd64 | 2094 | xfs_agi_write_verify( |
612cfbfe DC |
2095 | struct xfs_buf *bp) |
2096 | { | |
983d09ff DC |
2097 | struct xfs_mount *mp = bp->b_target->bt_mount; |
2098 | struct xfs_buf_log_item *bip = bp->b_fspriv; | |
2099 | ||
2100 | if (!xfs_agi_verify(bp)) { | |
2451337d | 2101 | xfs_buf_ioerror(bp, -EFSCORRUPTED); |
ce5028cf | 2102 | xfs_verifier_error(bp); |
983d09ff DC |
2103 | return; |
2104 | } | |
2105 | ||
2106 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
2107 | return; | |
2108 | ||
2109 | if (bip) | |
2110 | XFS_BUF_TO_AGI(bp)->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn); | |
f1dbcd7e | 2111 | xfs_buf_update_cksum(bp, XFS_AGI_CRC_OFF); |
3702ce6e DC |
2112 | } |
2113 | ||
1813dd64 DC |
2114 | const struct xfs_buf_ops xfs_agi_buf_ops = { |
2115 | .verify_read = xfs_agi_read_verify, | |
2116 | .verify_write = xfs_agi_write_verify, | |
2117 | }; | |
2118 | ||
1da177e4 LT |
2119 | /* |
2120 | * Read in the allocation group header (inode allocation section) | |
2121 | */ | |
2122 | int | |
5e1be0fb CH |
2123 | xfs_read_agi( |
2124 | struct xfs_mount *mp, /* file system mount structure */ | |
2125 | struct xfs_trans *tp, /* transaction pointer */ | |
2126 | xfs_agnumber_t agno, /* allocation group number */ | |
2127 | struct xfs_buf **bpp) /* allocation group hdr buf */ | |
1da177e4 | 2128 | { |
5e1be0fb | 2129 | int error; |
1da177e4 | 2130 | |
d123031a | 2131 | trace_xfs_read_agi(mp, agno); |
5e1be0fb | 2132 | |
d123031a | 2133 | ASSERT(agno != NULLAGNUMBER); |
5e1be0fb | 2134 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, |
1da177e4 | 2135 | XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)), |
1813dd64 | 2136 | XFS_FSS_TO_BB(mp, 1), 0, bpp, &xfs_agi_buf_ops); |
1da177e4 LT |
2137 | if (error) |
2138 | return error; | |
5e1be0fb | 2139 | |
38f23232 | 2140 | xfs_buf_set_ref(*bpp, XFS_AGI_REF); |
5e1be0fb CH |
2141 | return 0; |
2142 | } | |
2143 | ||
2144 | int | |
2145 | xfs_ialloc_read_agi( | |
2146 | struct xfs_mount *mp, /* file system mount structure */ | |
2147 | struct xfs_trans *tp, /* transaction pointer */ | |
2148 | xfs_agnumber_t agno, /* allocation group number */ | |
2149 | struct xfs_buf **bpp) /* allocation group hdr buf */ | |
2150 | { | |
2151 | struct xfs_agi *agi; /* allocation group header */ | |
2152 | struct xfs_perag *pag; /* per allocation group data */ | |
2153 | int error; | |
2154 | ||
d123031a DC |
2155 | trace_xfs_ialloc_read_agi(mp, agno); |
2156 | ||
5e1be0fb CH |
2157 | error = xfs_read_agi(mp, tp, agno, bpp); |
2158 | if (error) | |
2159 | return error; | |
2160 | ||
2161 | agi = XFS_BUF_TO_AGI(*bpp); | |
44b56e0a | 2162 | pag = xfs_perag_get(mp, agno); |
1da177e4 | 2163 | if (!pag->pagi_init) { |
16259e7d | 2164 | pag->pagi_freecount = be32_to_cpu(agi->agi_freecount); |
92821e2b | 2165 | pag->pagi_count = be32_to_cpu(agi->agi_count); |
1da177e4 | 2166 | pag->pagi_init = 1; |
1da177e4 | 2167 | } |
1da177e4 | 2168 | |
5e1be0fb CH |
2169 | /* |
2170 | * It's possible for these to be out of sync if | |
2171 | * we are in the middle of a forced shutdown. | |
2172 | */ | |
2173 | ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) || | |
2174 | XFS_FORCED_SHUTDOWN(mp)); | |
44b56e0a | 2175 | xfs_perag_put(pag); |
1da177e4 LT |
2176 | return 0; |
2177 | } | |
92821e2b DC |
2178 | |
2179 | /* | |
2180 | * Read in the agi to initialise the per-ag data in the mount structure | |
2181 | */ | |
2182 | int | |
2183 | xfs_ialloc_pagi_init( | |
2184 | xfs_mount_t *mp, /* file system mount structure */ | |
2185 | xfs_trans_t *tp, /* transaction pointer */ | |
2186 | xfs_agnumber_t agno) /* allocation group number */ | |
2187 | { | |
2188 | xfs_buf_t *bp = NULL; | |
2189 | int error; | |
2190 | ||
2191 | error = xfs_ialloc_read_agi(mp, tp, agno, &bp); | |
2192 | if (error) | |
2193 | return error; | |
2194 | if (bp) | |
2195 | xfs_trans_brelse(tp, bp); | |
2196 | return 0; | |
2197 | } |