]>
Commit | Line | Data |
---|---|---|
2bd0ea18 | 1 | /* |
da23017d NS |
2 | * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
5000d01d | 4 | * |
da23017d 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 | |
2bd0ea18 | 7 | * published by the Free Software Foundation. |
5000d01d | 8 | * |
da23017d 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. | |
5000d01d | 13 | * |
da23017d 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 | |
2bd0ea18 | 17 | */ |
9c799827 | 18 | #include "libxfs_priv.h" |
b626fb59 DC |
19 | #include "xfs_fs.h" |
20 | #include "xfs_shared.h" | |
21 | #include "xfs_format.h" | |
22 | #include "xfs_log_format.h" | |
23 | #include "xfs_trans_resv.h" | |
24 | #include "xfs_bit.h" | |
25 | #include "xfs_sb.h" | |
26 | #include "xfs_mount.h" | |
27 | #include "xfs_inode.h" | |
28 | #include "xfs_btree.h" | |
29 | #include "xfs_ialloc.h" | |
30 | #include "xfs_ialloc_btree.h" | |
31 | #include "xfs_alloc.h" | |
32 | #include "xfs_bmap.h" | |
33 | #include "xfs_cksum.h" | |
34 | #include "xfs_trans.h" | |
35 | #include "xfs_trace.h" | |
2bd0ea18 | 36 | |
2bd0ea18 NS |
37 | |
38 | /* | |
39 | * Allocation group level functions. | |
40 | */ | |
5e656dbb BN |
41 | static inline int |
42 | xfs_ialloc_cluster_alignment( | |
5a35bf2c | 43 | struct xfs_mount *mp) |
5e656dbb | 44 | { |
5a35bf2c DC |
45 | if (xfs_sb_version_hasalign(&mp->m_sb) && |
46 | mp->m_sb.sb_inoalignmt >= | |
47 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) | |
48 | return mp->m_sb.sb_inoalignmt; | |
5e656dbb BN |
49 | return 1; |
50 | } | |
2bd0ea18 | 51 | |
b194c7d8 | 52 | /* |
56b2de80 | 53 | * Lookup a record by ino in the btree given by cur. |
b194c7d8 BN |
54 | */ |
55 | int /* error */ | |
56b2de80 | 56 | xfs_inobt_lookup( |
b194c7d8 BN |
57 | struct xfs_btree_cur *cur, /* btree cursor */ |
58 | xfs_agino_t ino, /* starting inode of chunk */ | |
56b2de80 | 59 | xfs_lookup_t dir, /* <=, >=, == */ |
b194c7d8 BN |
60 | int *stat) /* success/failure */ |
61 | { | |
62 | cur->bc_rec.i.ir_startino = ino; | |
56b2de80 DC |
63 | cur->bc_rec.i.ir_freecount = 0; |
64 | cur->bc_rec.i.ir_free = 0; | |
65 | return xfs_btree_lookup(cur, dir, stat); | |
b194c7d8 BN |
66 | } |
67 | ||
68 | /* | |
56b2de80 | 69 | * Update the record referred to by cur to the value given. |
b194c7d8 BN |
70 | * This either works (return 0) or gets an EFSCORRUPTED error. |
71 | */ | |
72 | STATIC int /* error */ | |
73 | xfs_inobt_update( | |
74 | struct xfs_btree_cur *cur, /* btree cursor */ | |
56b2de80 | 75 | xfs_inobt_rec_incore_t *irec) /* btree record */ |
b194c7d8 BN |
76 | { |
77 | union xfs_btree_rec rec; | |
78 | ||
56b2de80 DC |
79 | rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino); |
80 | rec.inobt.ir_freecount = cpu_to_be32(irec->ir_freecount); | |
81 | rec.inobt.ir_free = cpu_to_be64(irec->ir_free); | |
b194c7d8 BN |
82 | return xfs_btree_update(cur, &rec); |
83 | } | |
84 | ||
85 | /* | |
86 | * Get the data from the pointed-to record. | |
87 | */ | |
88 | int /* error */ | |
89 | xfs_inobt_get_rec( | |
90 | struct xfs_btree_cur *cur, /* btree cursor */ | |
56b2de80 | 91 | xfs_inobt_rec_incore_t *irec, /* btree record */ |
b194c7d8 BN |
92 | int *stat) /* output: success/failure */ |
93 | { | |
94 | union xfs_btree_rec *rec; | |
95 | int error; | |
96 | ||
97 | error = xfs_btree_get_rec(cur, &rec, stat); | |
98 | if (!error && *stat == 1) { | |
56b2de80 DC |
99 | irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino); |
100 | irec->ir_freecount = be32_to_cpu(rec->inobt.ir_freecount); | |
101 | irec->ir_free = be64_to_cpu(rec->inobt.ir_free); | |
b194c7d8 BN |
102 | } |
103 | return error; | |
104 | } | |
105 | ||
3c699279 BF |
106 | /* |
107 | * Insert a single inobt record. Cursor must already point to desired location. | |
108 | */ | |
109 | STATIC int | |
110 | xfs_inobt_insert_rec( | |
111 | struct xfs_btree_cur *cur, | |
112 | __int32_t freecount, | |
113 | xfs_inofree_t free, | |
114 | int *stat) | |
115 | { | |
116 | cur->bc_rec.i.ir_freecount = freecount; | |
117 | cur->bc_rec.i.ir_free = free; | |
118 | return xfs_btree_insert(cur, stat); | |
119 | } | |
120 | ||
121 | /* | |
122 | * Insert records describing a newly allocated inode chunk into the inobt. | |
123 | */ | |
124 | STATIC int | |
125 | xfs_inobt_insert( | |
126 | struct xfs_mount *mp, | |
127 | struct xfs_trans *tp, | |
128 | struct xfs_buf *agbp, | |
129 | xfs_agino_t newino, | |
130 | xfs_agino_t newlen, | |
131 | xfs_btnum_t btnum) | |
132 | { | |
133 | struct xfs_btree_cur *cur; | |
134 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
135 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
136 | xfs_agino_t thisino; | |
137 | int i; | |
138 | int error; | |
139 | ||
140 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum); | |
141 | ||
142 | for (thisino = newino; | |
143 | thisino < newino + newlen; | |
144 | thisino += XFS_INODES_PER_CHUNK) { | |
145 | error = xfs_inobt_lookup(cur, thisino, XFS_LOOKUP_EQ, &i); | |
146 | if (error) { | |
147 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
148 | return error; | |
149 | } | |
150 | ASSERT(i == 0); | |
151 | ||
152 | error = xfs_inobt_insert_rec(cur, XFS_INODES_PER_CHUNK, | |
153 | XFS_INOBT_ALL_FREE, &i); | |
154 | if (error) { | |
155 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
156 | return error; | |
157 | } | |
158 | ASSERT(i == 1); | |
159 | } | |
160 | ||
161 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
162 | ||
163 | return 0; | |
164 | } | |
165 | ||
56b2de80 DC |
166 | /* |
167 | * Verify that the number of free inodes in the AGI is correct. | |
168 | */ | |
169 | #ifdef DEBUG | |
170 | STATIC int | |
171 | xfs_check_agi_freecount( | |
172 | struct xfs_btree_cur *cur, | |
173 | struct xfs_agi *agi) | |
174 | { | |
175 | if (cur->bc_nlevels == 1) { | |
176 | xfs_inobt_rec_incore_t rec; | |
177 | int freecount = 0; | |
178 | int error; | |
179 | int i; | |
180 | ||
181 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); | |
182 | if (error) | |
183 | return error; | |
184 | ||
185 | do { | |
186 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
187 | if (error) | |
188 | return error; | |
189 | ||
190 | if (i) { | |
191 | freecount += rec.ir_freecount; | |
192 | error = xfs_btree_increment(cur, 0, &i); | |
193 | if (error) | |
194 | return error; | |
195 | } | |
196 | } while (i == 1); | |
197 | ||
198 | if (!XFS_FORCED_SHUTDOWN(cur->bc_mp)) | |
199 | ASSERT(freecount == be32_to_cpu(agi->agi_freecount)); | |
200 | } | |
201 | return 0; | |
202 | } | |
203 | #else | |
204 | #define xfs_check_agi_freecount(cur, agi) 0 | |
205 | #endif | |
206 | ||
207 | /* | |
e9d35108 DC |
208 | * Initialise a new set of inodes. When called without a transaction context |
209 | * (e.g. from recovery) we initiate a delayed write of the inode buffers rather | |
210 | * than logging them (which in a transaction context puts them into the AIL | |
211 | * for writeback rather than the xfsbufd queue). | |
56b2de80 | 212 | */ |
e9d35108 | 213 | int |
56b2de80 DC |
214 | xfs_ialloc_inode_init( |
215 | struct xfs_mount *mp, | |
216 | struct xfs_trans *tp, | |
e9d35108 | 217 | struct list_head *buffer_list, |
56b2de80 DC |
218 | xfs_agnumber_t agno, |
219 | xfs_agblock_t agbno, | |
220 | xfs_agblock_t length, | |
221 | unsigned int gen) | |
222 | { | |
223 | struct xfs_buf *fbuf; | |
224 | struct xfs_dinode *free; | |
ff105f75 | 225 | int nbufs, blks_per_cluster, inodes_per_cluster; |
56b2de80 DC |
226 | int version; |
227 | int i, j; | |
228 | xfs_daddr_t d; | |
41ce5f36 | 229 | xfs_ino_t ino = 0; |
56b2de80 DC |
230 | |
231 | /* | |
ff105f75 DC |
232 | * Loop over the new block(s), filling in the inodes. For small block |
233 | * sizes, manipulate the inodes in buffers which are multiples of the | |
234 | * blocks size. | |
56b2de80 | 235 | */ |
ff105f75 DC |
236 | blks_per_cluster = xfs_icluster_size_fsb(mp); |
237 | inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog; | |
238 | nbufs = length / blks_per_cluster; | |
56b2de80 DC |
239 | |
240 | /* | |
e9d35108 DC |
241 | * Figure out what version number to use in the inodes we create. If |
242 | * the superblock version has caught up to the one that supports the new | |
243 | * inode format, then use the new inode version. Otherwise use the old | |
244 | * version so that old kernels will continue to be able to use the file | |
245 | * system. | |
41ce5f36 DC |
246 | * |
247 | * For v3 inodes, we also need to write the inode number into the inode, | |
248 | * so calculate the first inode number of the chunk here as | |
e9d35108 DC |
249 | * XFS_OFFBNO_TO_AGINO() only works within a filesystem block, not |
250 | * across multiple filesystem blocks (such as a cluster) and so cannot | |
251 | * be used in the cluster buffer loop below. | |
252 | * | |
253 | * Further, because we are writing the inode directly into the buffer | |
254 | * and calculating a CRC on the entire inode, we have ot log the entire | |
255 | * inode so that the entire range the CRC covers is present in the log. | |
256 | * That means for v3 inode we log the entire buffer rather than just the | |
257 | * inode cores. | |
56b2de80 | 258 | */ |
41ce5f36 DC |
259 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
260 | version = 3; | |
261 | ino = XFS_AGINO_TO_INO(mp, agno, | |
262 | XFS_OFFBNO_TO_AGINO(mp, agbno, 0)); | |
e9d35108 DC |
263 | |
264 | /* | |
265 | * log the initialisation that is about to take place as an | |
266 | * logical operation. This means the transaction does not | |
267 | * need to log the physical changes to the inode buffers as log | |
268 | * recovery will know what initialisation is actually needed. | |
269 | * Hence we only need to log the buffers as "ordered" buffers so | |
270 | * they track in the AIL as if they were physically logged. | |
271 | */ | |
272 | if (tp) | |
ff105f75 | 273 | xfs_icreate_log(tp, agno, agbno, mp->m_ialloc_inos, |
e9d35108 | 274 | mp->m_sb.sb_inodesize, length, gen); |
ff105f75 | 275 | } else |
56b2de80 | 276 | version = 2; |
56b2de80 DC |
277 | |
278 | for (j = 0; j < nbufs; j++) { | |
279 | /* | |
280 | * Get the block. | |
281 | */ | |
282 | d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster)); | |
283 | fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, | |
284 | mp->m_bsize * blks_per_cluster, | |
a2ceac1f DC |
285 | XBF_UNMAPPED); |
286 | if (!fbuf) | |
12b53197 | 287 | return -ENOMEM; |
e9d35108 DC |
288 | |
289 | /* Initialize the inode buffers and log them appropriately. */ | |
a2ceac1f | 290 | fbuf->b_ops = &xfs_inode_buf_ops; |
e9d35108 | 291 | xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length)); |
ff105f75 | 292 | for (i = 0; i < inodes_per_cluster; i++) { |
56b2de80 | 293 | int ioffset = i << mp->m_sb.sb_inodelog; |
41ce5f36 | 294 | uint isize = xfs_dinode_size(version); |
56b2de80 DC |
295 | |
296 | free = xfs_make_iptr(mp, fbuf, i); | |
297 | free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC); | |
298 | free->di_version = version; | |
299 | free->di_gen = cpu_to_be32(gen); | |
300 | free->di_next_unlinked = cpu_to_be32(NULLAGINO); | |
41ce5f36 DC |
301 | |
302 | if (version == 3) { | |
303 | free->di_ino = cpu_to_be64(ino); | |
304 | ino++; | |
305 | uuid_copy(&free->di_uuid, &mp->m_sb.sb_uuid); | |
306 | xfs_dinode_calc_crc(mp, free); | |
e9d35108 DC |
307 | } else if (tp) { |
308 | /* just log the inode core */ | |
309 | xfs_trans_log_buf(tp, fbuf, ioffset, | |
310 | ioffset + isize - 1); | |
41ce5f36 | 311 | } |
e9d35108 | 312 | } |
41ce5f36 | 313 | |
e9d35108 DC |
314 | if (tp) { |
315 | /* | |
316 | * Mark the buffer as an inode allocation buffer so it | |
317 | * sticks in AIL at the point of this allocation | |
318 | * transaction. This ensures the they are on disk before | |
319 | * the tail of the log can be moved past this | |
320 | * transaction (i.e. by preventing relogging from moving | |
321 | * it forward in the log). | |
322 | */ | |
323 | xfs_trans_inode_alloc_buf(tp, fbuf); | |
324 | if (version == 3) { | |
325 | /* | |
326 | * Mark the buffer as ordered so that they are | |
327 | * not physically logged in the transaction but | |
328 | * still tracked in the AIL as part of the | |
329 | * transaction and pin the log appropriately. | |
330 | */ | |
331 | xfs_trans_ordered_buf(tp, fbuf); | |
332 | xfs_trans_log_buf(tp, fbuf, 0, | |
333 | BBTOB(fbuf->b_length) - 1); | |
334 | } | |
335 | } else { | |
336 | fbuf->b_flags |= XBF_DONE; | |
337 | xfs_buf_delwri_queue(fbuf, buffer_list); | |
338 | xfs_buf_relse(fbuf); | |
56b2de80 | 339 | } |
56b2de80 | 340 | } |
a2ceac1f | 341 | return 0; |
56b2de80 DC |
342 | } |
343 | ||
2bd0ea18 NS |
344 | /* |
345 | * Allocate new inodes in the allocation group specified by agbp. | |
346 | * Return 0 for success, else error code. | |
347 | */ | |
348 | STATIC int /* error code or 0 */ | |
349 | xfs_ialloc_ag_alloc( | |
350 | xfs_trans_t *tp, /* transaction pointer */ | |
351 | xfs_buf_t *agbp, /* alloc group buffer */ | |
352 | int *alloc) | |
353 | { | |
354 | xfs_agi_t *agi; /* allocation group header */ | |
dfc130f3 | 355 | xfs_alloc_arg_t args; /* allocation argument structure */ |
5e656dbb | 356 | xfs_agnumber_t agno; |
2bd0ea18 | 357 | int error; |
2bd0ea18 NS |
358 | xfs_agino_t newino; /* new first inode's number */ |
359 | xfs_agino_t newlen; /* new number of inodes */ | |
5e656dbb | 360 | int isaligned = 0; /* inode allocation at stripe unit */ |
2bd0ea18 | 361 | /* boundary */ |
56b2de80 | 362 | struct xfs_perag *pag; |
5000d01d | 363 | |
a2ceac1f | 364 | memset(&args, 0, sizeof(args)); |
2bd0ea18 NS |
365 | args.tp = tp; |
366 | args.mp = tp->t_mountp; | |
367 | ||
368 | /* | |
369 | * Locking will ensure that we don't have two callers in here | |
370 | * at one time. | |
371 | */ | |
ff105f75 | 372 | newlen = args.mp->m_ialloc_inos; |
2bd0ea18 | 373 | if (args.mp->m_maxicount && |
19ebedcf DC |
374 | percpu_counter_read(&args.mp->m_icount) + newlen > |
375 | args.mp->m_maxicount) | |
12b53197 | 376 | return -ENOSPC; |
ff105f75 | 377 | args.minlen = args.maxlen = args.mp->m_ialloc_blks; |
2bd0ea18 | 378 | /* |
5e656dbb BN |
379 | * First try to allocate inodes contiguous with the last-allocated |
380 | * chunk of inodes. If the filesystem is striped, this will fill | |
381 | * an entire stripe unit with inodes. | |
3439d03a | 382 | */ |
2bd0ea18 | 383 | agi = XFS_BUF_TO_AGI(agbp); |
5e656dbb | 384 | newino = be32_to_cpu(agi->agi_newino); |
56b2de80 | 385 | agno = be32_to_cpu(agi->agi_seqno); |
5e656dbb | 386 | args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) + |
ff105f75 | 387 | args.mp->m_ialloc_blks; |
5e656dbb BN |
388 | if (likely(newino != NULLAGINO && |
389 | (args.agbno < be32_to_cpu(agi->agi_length)))) { | |
56b2de80 | 390 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
5e656dbb | 391 | args.type = XFS_ALLOCTYPE_THIS_BNO; |
5e656dbb BN |
392 | args.prod = 1; |
393 | ||
394 | /* | |
395 | * We need to take into account alignment here to ensure that | |
396 | * we don't modify the free list if we fail to have an exact | |
397 | * block. If we don't have an exact match, and every oher | |
398 | * attempt allocation attempt fails, we'll end up cancelling | |
399 | * a dirty transaction and shutting down. | |
400 | * | |
401 | * For an exact allocation, alignment must be 1, | |
402 | * however we need to take cluster alignment into account when | |
403 | * fixing up the freelist. Use the minalignslop field to | |
404 | * indicate that extra blocks might be required for alignment, | |
405 | * but not to use them in the actual exact allocation. | |
406 | */ | |
407 | args.alignment = 1; | |
5a35bf2c | 408 | args.minalignslop = xfs_ialloc_cluster_alignment(args.mp) - 1; |
5e656dbb BN |
409 | |
410 | /* Allow space for the inode btree to split. */ | |
56b2de80 | 411 | args.minleft = args.mp->m_in_maxlevels - 1; |
5e656dbb BN |
412 | if ((error = xfs_alloc_vextent(&args))) |
413 | return error; | |
ff105f75 DC |
414 | |
415 | /* | |
416 | * This request might have dirtied the transaction if the AG can | |
417 | * satisfy the request, but the exact block was not available. | |
418 | * If the allocation did fail, subsequent requests will relax | |
419 | * the exact agbno requirement and increase the alignment | |
420 | * instead. It is critical that the total size of the request | |
421 | * (len + alignment + slop) does not increase from this point | |
422 | * on, so reset minalignslop to ensure it is not included in | |
423 | * subsequent requests. | |
424 | */ | |
425 | args.minalignslop = 0; | |
5e656dbb BN |
426 | } else |
427 | args.fsbno = NULLFSBLOCK; | |
428 | ||
429 | if (unlikely(args.fsbno == NULLFSBLOCK)) { | |
430 | /* | |
431 | * Set the alignment for the allocation. | |
432 | * If stripe alignment is turned on then align at stripe unit | |
433 | * boundary. | |
434 | * If the cluster size is smaller than a filesystem block | |
435 | * then we're doing I/O for inodes in filesystem block size | |
436 | * pieces, so don't need alignment anyway. | |
437 | */ | |
438 | isaligned = 0; | |
439 | if (args.mp->m_sinoalign) { | |
440 | ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN)); | |
441 | args.alignment = args.mp->m_dalign; | |
442 | isaligned = 1; | |
443 | } else | |
5a35bf2c | 444 | args.alignment = xfs_ialloc_cluster_alignment(args.mp); |
5e656dbb BN |
445 | /* |
446 | * Need to figure out where to allocate the inode blocks. | |
447 | * Ideally they should be spaced out through the a.g. | |
448 | * For now, just allocate blocks up front. | |
449 | */ | |
450 | args.agbno = be32_to_cpu(agi->agi_root); | |
56b2de80 | 451 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
5e656dbb BN |
452 | /* |
453 | * Allocate a fixed-size extent of inodes. | |
454 | */ | |
455 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
5e656dbb BN |
456 | args.prod = 1; |
457 | /* | |
458 | * Allow space for the inode btree to split. | |
459 | */ | |
56b2de80 | 460 | args.minleft = args.mp->m_in_maxlevels - 1; |
5e656dbb BN |
461 | if ((error = xfs_alloc_vextent(&args))) |
462 | return error; | |
463 | } | |
2bd0ea18 NS |
464 | |
465 | /* | |
466 | * If stripe alignment is turned on, then try again with cluster | |
467 | * alignment. | |
468 | */ | |
469 | if (isaligned && args.fsbno == NULLFSBLOCK) { | |
470 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
6e3140c7 | 471 | args.agbno = be32_to_cpu(agi->agi_root); |
56b2de80 | 472 | args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno); |
5a35bf2c | 473 | args.alignment = xfs_ialloc_cluster_alignment(args.mp); |
0e266570 | 474 | if ((error = xfs_alloc_vextent(&args))) |
dfc130f3 | 475 | return error; |
2bd0ea18 | 476 | } |
5000d01d | 477 | |
2bd0ea18 NS |
478 | if (args.fsbno == NULLFSBLOCK) { |
479 | *alloc = 0; | |
480 | return 0; | |
481 | } | |
482 | ASSERT(args.len == args.minlen); | |
a562a63b | 483 | |
5e656dbb | 484 | /* |
56b2de80 DC |
485 | * Stamp and write the inode buffers. |
486 | * | |
5e656dbb BN |
487 | * Seed the new inode cluster with a random generation number. This |
488 | * prevents short-term reuse of generation numbers if a chunk is | |
489 | * freed and then immediately reallocated. We use random numbers | |
490 | * rather than a linear progression to prevent the next generation | |
491 | * number from being easily guessable. | |
492 | */ | |
e9d35108 | 493 | error = xfs_ialloc_inode_init(args.mp, tp, NULL, agno, args.agbno, |
49f693fa | 494 | args.len, prandom_u32()); |
56b2de80 | 495 | |
a2ceac1f DC |
496 | if (error) |
497 | return error; | |
56b2de80 DC |
498 | /* |
499 | * Convert the results. | |
500 | */ | |
501 | newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0); | |
5e656dbb BN |
502 | be32_add_cpu(&agi->agi_count, newlen); |
503 | be32_add_cpu(&agi->agi_freecount, newlen); | |
56b2de80 DC |
504 | pag = xfs_perag_get(args.mp, agno); |
505 | pag->pagi_freecount += newlen; | |
506 | xfs_perag_put(pag); | |
6e3140c7 | 507 | agi->agi_newino = cpu_to_be32(newino); |
56b2de80 | 508 | |
2bd0ea18 | 509 | /* |
3c699279 | 510 | * Insert records describing the new inode chunk into the btrees. |
2bd0ea18 | 511 | */ |
3c699279 BF |
512 | error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen, |
513 | XFS_BTNUM_INO); | |
514 | if (error) | |
515 | return error; | |
516 | ||
517 | if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) { | |
518 | error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen, | |
519 | XFS_BTNUM_FINO); | |
520 | if (error) | |
2bd0ea18 | 521 | return error; |
2bd0ea18 | 522 | } |
2bd0ea18 NS |
523 | /* |
524 | * Log allocation group header fields | |
525 | */ | |
526 | xfs_ialloc_log_agi(tp, agbp, | |
527 | XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO); | |
528 | /* | |
529 | * Modify/log superblock values for inode count and inode free count. | |
530 | */ | |
531 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen); | |
532 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen); | |
533 | *alloc = 1; | |
534 | return 0; | |
535 | } | |
536 | ||
56b2de80 | 537 | STATIC xfs_agnumber_t |
321717ae NS |
538 | xfs_ialloc_next_ag( |
539 | xfs_mount_t *mp) | |
540 | { | |
541 | xfs_agnumber_t agno; | |
542 | ||
543 | spin_lock(&mp->m_agirotor_lock); | |
544 | agno = mp->m_agirotor; | |
a2ceac1f | 545 | if (++mp->m_agirotor >= mp->m_maxagi) |
321717ae NS |
546 | mp->m_agirotor = 0; |
547 | spin_unlock(&mp->m_agirotor_lock); | |
548 | ||
549 | return agno; | |
550 | } | |
551 | ||
2bd0ea18 NS |
552 | /* |
553 | * Select an allocation group to look for a free inode in, based on the parent | |
e6d77a21 | 554 | * inode and the mode. Return the allocation group buffer. |
2bd0ea18 | 555 | */ |
a2ceac1f | 556 | STATIC xfs_agnumber_t |
2bd0ea18 NS |
557 | xfs_ialloc_ag_select( |
558 | xfs_trans_t *tp, /* transaction pointer */ | |
559 | xfs_ino_t parent, /* parent directory inode number */ | |
a2ceac1f | 560 | umode_t mode, /* bits set to indicate file type */ |
2bd0ea18 NS |
561 | int okalloc) /* ok to allocate more space */ |
562 | { | |
2bd0ea18 NS |
563 | xfs_agnumber_t agcount; /* number of ag's in the filesystem */ |
564 | xfs_agnumber_t agno; /* current ag number */ | |
565 | int flags; /* alloc buffer locking flags */ | |
566 | xfs_extlen_t ineed; /* blocks needed for inode allocation */ | |
275ae71f | 567 | xfs_extlen_t longest = 0; /* longest extent available */ |
2bd0ea18 NS |
568 | xfs_mount_t *mp; /* mount point structure */ |
569 | int needspace; /* file mode implies space allocated */ | |
570 | xfs_perag_t *pag; /* per allocation group data */ | |
571 | xfs_agnumber_t pagno; /* parent (starting) ag number */ | |
a2ceac1f | 572 | int error; |
2bd0ea18 NS |
573 | |
574 | /* | |
575 | * Files of these types need at least one block if length > 0 | |
576 | * (and they won't fit in the inode, but that's hard to figure out). | |
577 | */ | |
578 | needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode); | |
579 | mp = tp->t_mountp; | |
34317449 | 580 | agcount = mp->m_maxagi; |
2bd0ea18 | 581 | if (S_ISDIR(mode)) |
321717ae | 582 | pagno = xfs_ialloc_next_ag(mp); |
5ce1d1f7 | 583 | else { |
2bd0ea18 | 584 | pagno = XFS_INO_TO_AGNO(mp, parent); |
5ce1d1f7 NS |
585 | if (pagno >= agcount) |
586 | pagno = 0; | |
587 | } | |
a2ceac1f | 588 | |
2bd0ea18 | 589 | ASSERT(pagno < agcount); |
a2ceac1f | 590 | |
2bd0ea18 NS |
591 | /* |
592 | * Loop through allocation groups, looking for one with a little | |
593 | * free space in it. Note we don't look for free inodes, exactly. | |
594 | * Instead, we include whether there is a need to allocate inodes | |
5000d01d | 595 | * to mean that blocks must be allocated for them, |
2bd0ea18 NS |
596 | * if none are currently free. |
597 | */ | |
598 | agno = pagno; | |
599 | flags = XFS_ALLOC_FLAG_TRYLOCK; | |
600 | for (;;) { | |
56b2de80 | 601 | pag = xfs_perag_get(mp, agno); |
a2ceac1f DC |
602 | if (!pag->pagi_inodeok) { |
603 | xfs_ialloc_next_ag(mp); | |
604 | goto nextag; | |
605 | } | |
606 | ||
2bd0ea18 | 607 | if (!pag->pagi_init) { |
a2ceac1f DC |
608 | error = xfs_ialloc_pagi_init(mp, tp, agno); |
609 | if (error) | |
2bd0ea18 | 610 | goto nextag; |
a2ceac1f | 611 | } |
34317449 | 612 | |
a2ceac1f DC |
613 | if (pag->pagi_freecount) { |
614 | xfs_perag_put(pag); | |
615 | return agno; | |
34317449 NS |
616 | } |
617 | ||
a2ceac1f DC |
618 | if (!okalloc) |
619 | goto nextag; | |
620 | ||
621 | if (!pag->pagf_init) { | |
622 | error = xfs_alloc_pagf_init(mp, tp, agno, flags); | |
623 | if (error) | |
2bd0ea18 | 624 | goto nextag; |
2bd0ea18 | 625 | } |
a2ceac1f DC |
626 | |
627 | /* | |
5a35bf2c DC |
628 | * Check that there is enough free space for the file plus a |
629 | * chunk of inodes if we need to allocate some. If this is the | |
630 | * first pass across the AGs, take into account the potential | |
631 | * space needed for alignment of inode chunks when checking the | |
632 | * longest contiguous free space in the AG - this prevents us | |
633 | * from getting ENOSPC because we have free space larger than | |
634 | * m_ialloc_blks but alignment constraints prevent us from using | |
635 | * it. | |
636 | * | |
637 | * If we can't find an AG with space for full alignment slack to | |
638 | * be taken into account, we must be near ENOSPC in all AGs. | |
639 | * Hence we don't include alignment for the second pass and so | |
640 | * if we fail allocation due to alignment issues then it is most | |
641 | * likely a real ENOSPC condition. | |
a2ceac1f | 642 | */ |
62dc6cdb | 643 | ineed = mp->m_ialloc_min_blks; |
5a35bf2c DC |
644 | if (flags && ineed > 1) |
645 | ineed += xfs_ialloc_cluster_alignment(mp); | |
a2ceac1f DC |
646 | longest = pag->pagf_longest; |
647 | if (!longest) | |
648 | longest = pag->pagf_flcount > 0; | |
649 | ||
650 | if (pag->pagf_freeblks >= needspace + ineed && | |
651 | longest >= ineed) { | |
652 | xfs_perag_put(pag); | |
653 | return agno; | |
2bd0ea18 | 654 | } |
5000d01d | 655 | nextag: |
56b2de80 | 656 | xfs_perag_put(pag); |
5000d01d | 657 | /* |
2bd0ea18 NS |
658 | * No point in iterating over the rest, if we're shutting |
659 | * down. | |
660 | */ | |
56b2de80 | 661 | if (XFS_FORCED_SHUTDOWN(mp)) |
a2ceac1f | 662 | return NULLAGNUMBER; |
2bd0ea18 | 663 | agno++; |
5ce1d1f7 | 664 | if (agno >= agcount) |
2bd0ea18 NS |
665 | agno = 0; |
666 | if (agno == pagno) { | |
56b2de80 | 667 | if (flags == 0) |
a2ceac1f | 668 | return NULLAGNUMBER; |
2bd0ea18 NS |
669 | flags = 0; |
670 | } | |
671 | } | |
672 | } | |
673 | ||
56b2de80 DC |
674 | /* |
675 | * Try to retrieve the next record to the left/right from the current one. | |
676 | */ | |
677 | STATIC int | |
678 | xfs_ialloc_next_rec( | |
679 | struct xfs_btree_cur *cur, | |
680 | xfs_inobt_rec_incore_t *rec, | |
681 | int *done, | |
682 | int left) | |
683 | { | |
684 | int error; | |
685 | int i; | |
686 | ||
687 | if (left) | |
688 | error = xfs_btree_decrement(cur, 0, &i); | |
689 | else | |
690 | error = xfs_btree_increment(cur, 0, &i); | |
691 | ||
692 | if (error) | |
693 | return error; | |
694 | *done = !i; | |
695 | if (i) { | |
696 | error = xfs_inobt_get_rec(cur, rec, &i); | |
697 | if (error) | |
698 | return error; | |
19ebedcf | 699 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
56b2de80 DC |
700 | } |
701 | ||
702 | return 0; | |
703 | } | |
704 | ||
705 | STATIC int | |
706 | xfs_ialloc_get_rec( | |
707 | struct xfs_btree_cur *cur, | |
708 | xfs_agino_t agino, | |
709 | xfs_inobt_rec_incore_t *rec, | |
3439d03a | 710 | int *done) |
56b2de80 DC |
711 | { |
712 | int error; | |
713 | int i; | |
714 | ||
715 | error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i); | |
716 | if (error) | |
717 | return error; | |
718 | *done = !i; | |
719 | if (i) { | |
720 | error = xfs_inobt_get_rec(cur, rec, &i); | |
721 | if (error) | |
722 | return error; | |
19ebedcf | 723 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
56b2de80 DC |
724 | } |
725 | ||
726 | return 0; | |
727 | } | |
728 | ||
01792d3b BF |
729 | /* |
730 | * Return the offset of the first free inode in the record. | |
731 | */ | |
732 | STATIC int | |
733 | xfs_inobt_first_free_inode( | |
734 | struct xfs_inobt_rec_incore *rec) | |
735 | { | |
736 | return xfs_lowbit64(rec->ir_free); | |
737 | } | |
738 | ||
5000d01d | 739 | /* |
ff105f75 | 740 | * Allocate an inode using the inobt-only algorithm. |
2bd0ea18 | 741 | */ |
a2ceac1f | 742 | STATIC int |
ff105f75 | 743 | xfs_dialloc_ag_inobt( |
a2ceac1f DC |
744 | struct xfs_trans *tp, |
745 | struct xfs_buf *agbp, | |
746 | xfs_ino_t parent, | |
747 | xfs_ino_t *inop) | |
2bd0ea18 | 748 | { |
a2ceac1f DC |
749 | struct xfs_mount *mp = tp->t_mountp; |
750 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
751 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
752 | xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent); | |
753 | xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent); | |
754 | struct xfs_perag *pag; | |
755 | struct xfs_btree_cur *cur, *tcur; | |
756 | struct xfs_inobt_rec_incore rec, trec; | |
757 | xfs_ino_t ino; | |
758 | int error; | |
759 | int offset; | |
760 | int i, j; | |
2bd0ea18 | 761 | |
56b2de80 DC |
762 | pag = xfs_perag_get(mp, agno); |
763 | ||
a2ceac1f DC |
764 | ASSERT(pag->pagi_init); |
765 | ASSERT(pag->pagi_inodeok); | |
766 | ASSERT(pag->pagi_freecount > 0); | |
767 | ||
56b2de80 | 768 | restart_pagno: |
70eb7337 | 769 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
2bd0ea18 NS |
770 | /* |
771 | * If pagino is 0 (this is the root inode allocation) use newino. | |
772 | * This must work because we've just allocated some. | |
773 | */ | |
774 | if (!pagino) | |
6e3140c7 | 775 | pagino = be32_to_cpu(agi->agi_newino); |
2bd0ea18 | 776 | |
56b2de80 DC |
777 | error = xfs_check_agi_freecount(cur, agi); |
778 | if (error) | |
779 | goto error0; | |
2bd0ea18 | 780 | |
2bd0ea18 | 781 | /* |
56b2de80 | 782 | * If in the same AG as the parent, try to get near the parent. |
2bd0ea18 NS |
783 | */ |
784 | if (pagno == agno) { | |
56b2de80 DC |
785 | int doneleft; /* done, to the left */ |
786 | int doneright; /* done, to the right */ | |
787 | int searchdistance = 10; | |
788 | ||
789 | error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i); | |
790 | if (error) | |
2bd0ea18 | 791 | goto error0; |
19ebedcf | 792 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
56b2de80 DC |
793 | |
794 | error = xfs_inobt_get_rec(cur, &rec, &j); | |
795 | if (error) | |
796 | goto error0; | |
19ebedcf | 797 | XFS_WANT_CORRUPTED_GOTO(mp, j == 1, error0); |
56b2de80 DC |
798 | |
799 | if (rec.ir_freecount > 0) { | |
2bd0ea18 NS |
800 | /* |
801 | * Found a free inode in the same chunk | |
56b2de80 | 802 | * as the parent, done. |
2bd0ea18 | 803 | */ |
56b2de80 | 804 | goto alloc_inode; |
2bd0ea18 | 805 | } |
56b2de80 DC |
806 | |
807 | ||
808 | /* | |
809 | * In the same AG as parent, but parent's chunk is full. | |
810 | */ | |
811 | ||
812 | /* duplicate the cursor, search left & right simultaneously */ | |
813 | error = xfs_btree_dup_cursor(cur, &tcur); | |
814 | if (error) | |
815 | goto error0; | |
816 | ||
2bd0ea18 | 817 | /* |
56b2de80 | 818 | * Skip to last blocks looked up if same parent inode. |
2bd0ea18 | 819 | */ |
56b2de80 DC |
820 | if (pagino != NULLAGINO && |
821 | pag->pagl_pagino == pagino && | |
822 | pag->pagl_leftrec != NULLAGINO && | |
823 | pag->pagl_rightrec != NULLAGINO) { | |
824 | error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec, | |
3439d03a | 825 | &trec, &doneleft); |
56b2de80 DC |
826 | if (error) |
827 | goto error1; | |
2bd0ea18 | 828 | |
56b2de80 | 829 | error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec, |
3439d03a | 830 | &rec, &doneright); |
2bd0ea18 | 831 | if (error) |
2bd0ea18 | 832 | goto error1; |
56b2de80 DC |
833 | } else { |
834 | /* search left with tcur, back up 1 record */ | |
835 | error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1); | |
836 | if (error) | |
2bd0ea18 | 837 | goto error1; |
2bd0ea18 | 838 | |
56b2de80 DC |
839 | /* search right with cur, go forward 1 record. */ |
840 | error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0); | |
841 | if (error) | |
842 | goto error1; | |
843 | } | |
844 | ||
845 | /* | |
846 | * Loop until we find an inode chunk with a free inode. | |
847 | */ | |
848 | while (!doneleft || !doneright) { | |
849 | int useleft; /* using left inode chunk this time */ | |
850 | ||
851 | if (!--searchdistance) { | |
2bd0ea18 | 852 | /* |
56b2de80 DC |
853 | * Not in range - save last search |
854 | * location and allocate a new inode | |
2bd0ea18 | 855 | */ |
56b2de80 DC |
856 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); |
857 | pag->pagl_leftrec = trec.ir_startino; | |
858 | pag->pagl_rightrec = rec.ir_startino; | |
859 | pag->pagl_pagino = pagino; | |
860 | goto newino; | |
2bd0ea18 | 861 | } |
56b2de80 DC |
862 | |
863 | /* figure out the closer block if both are valid. */ | |
864 | if (!doneleft && !doneright) { | |
865 | useleft = pagino - | |
866 | (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) < | |
867 | rec.ir_startino - pagino; | |
868 | } else { | |
869 | useleft = !doneleft; | |
870 | } | |
871 | ||
872 | /* free inodes to the left? */ | |
873 | if (useleft && trec.ir_freecount) { | |
874 | rec = trec; | |
875 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
876 | cur = tcur; | |
877 | ||
878 | pag->pagl_leftrec = trec.ir_startino; | |
879 | pag->pagl_rightrec = rec.ir_startino; | |
880 | pag->pagl_pagino = pagino; | |
881 | goto alloc_inode; | |
882 | } | |
883 | ||
884 | /* free inodes to the right? */ | |
885 | if (!useleft && rec.ir_freecount) { | |
886 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
887 | ||
888 | pag->pagl_leftrec = trec.ir_startino; | |
889 | pag->pagl_rightrec = rec.ir_startino; | |
890 | pag->pagl_pagino = pagino; | |
891 | goto alloc_inode; | |
892 | } | |
893 | ||
894 | /* get next record to check */ | |
895 | if (useleft) { | |
896 | error = xfs_ialloc_next_rec(tcur, &trec, | |
897 | &doneleft, 1); | |
898 | } else { | |
899 | error = xfs_ialloc_next_rec(cur, &rec, | |
900 | &doneright, 0); | |
901 | } | |
902 | if (error) | |
903 | goto error1; | |
2bd0ea18 | 904 | } |
56b2de80 DC |
905 | |
906 | /* | |
907 | * We've reached the end of the btree. because | |
908 | * we are only searching a small chunk of the | |
909 | * btree each search, there is obviously free | |
910 | * inodes closer to the parent inode than we | |
911 | * are now. restart the search again. | |
912 | */ | |
913 | pag->pagl_pagino = NULLAGINO; | |
914 | pag->pagl_leftrec = NULLAGINO; | |
915 | pag->pagl_rightrec = NULLAGINO; | |
916 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
917 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
918 | goto restart_pagno; | |
2bd0ea18 | 919 | } |
56b2de80 | 920 | |
2bd0ea18 | 921 | /* |
56b2de80 | 922 | * In a different AG from the parent. |
2bd0ea18 NS |
923 | * See if the most recently allocated block has any free. |
924 | */ | |
56b2de80 | 925 | newino: |
a2ceac1f | 926 | if (agi->agi_newino != cpu_to_be32(NULLAGINO)) { |
56b2de80 DC |
927 | error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino), |
928 | XFS_LOOKUP_EQ, &i); | |
929 | if (error) | |
2bd0ea18 | 930 | goto error0; |
56b2de80 DC |
931 | |
932 | if (i == 1) { | |
933 | error = xfs_inobt_get_rec(cur, &rec, &j); | |
2bd0ea18 NS |
934 | if (error) |
935 | goto error0; | |
56b2de80 DC |
936 | |
937 | if (j == 1 && rec.ir_freecount > 0) { | |
938 | /* | |
939 | * The last chunk allocated in the group | |
940 | * still has a free inode. | |
941 | */ | |
942 | goto alloc_inode; | |
2bd0ea18 NS |
943 | } |
944 | } | |
945 | } | |
56b2de80 DC |
946 | |
947 | /* | |
948 | * None left in the last group, search the whole AG | |
949 | */ | |
950 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); | |
951 | if (error) | |
952 | goto error0; | |
19ebedcf | 953 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
56b2de80 DC |
954 | |
955 | for (;;) { | |
956 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
957 | if (error) | |
958 | goto error0; | |
19ebedcf | 959 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
56b2de80 DC |
960 | if (rec.ir_freecount > 0) |
961 | break; | |
962 | error = xfs_btree_increment(cur, 0, &i); | |
963 | if (error) | |
964 | goto error0; | |
19ebedcf | 965 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
56b2de80 DC |
966 | } |
967 | ||
968 | alloc_inode: | |
01792d3b | 969 | offset = xfs_inobt_first_free_inode(&rec); |
2bd0ea18 NS |
970 | ASSERT(offset >= 0); |
971 | ASSERT(offset < XFS_INODES_PER_CHUNK); | |
972 | ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) % | |
973 | XFS_INODES_PER_CHUNK) == 0); | |
974 | ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset); | |
56b2de80 | 975 | rec.ir_free &= ~XFS_INOBT_MASK(offset); |
2bd0ea18 | 976 | rec.ir_freecount--; |
56b2de80 DC |
977 | error = xfs_inobt_update(cur, &rec); |
978 | if (error) | |
2bd0ea18 | 979 | goto error0; |
5e656dbb | 980 | be32_add_cpu(&agi->agi_freecount, -1); |
2bd0ea18 | 981 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); |
56b2de80 DC |
982 | pag->pagi_freecount--; |
983 | ||
984 | error = xfs_check_agi_freecount(cur, agi); | |
985 | if (error) | |
986 | goto error0; | |
2bd0ea18 | 987 | |
2bd0ea18 NS |
988 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
989 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1); | |
56b2de80 | 990 | xfs_perag_put(pag); |
2bd0ea18 NS |
991 | *inop = ino; |
992 | return 0; | |
993 | error1: | |
994 | xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR); | |
995 | error0: | |
996 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
56b2de80 | 997 | xfs_perag_put(pag); |
2bd0ea18 NS |
998 | return error; |
999 | } | |
1000 | ||
ff105f75 DC |
1001 | /* |
1002 | * Use the free inode btree to allocate an inode based on distance from the | |
1003 | * parent. Note that the provided cursor may be deleted and replaced. | |
1004 | */ | |
1005 | STATIC int | |
1006 | xfs_dialloc_ag_finobt_near( | |
1007 | xfs_agino_t pagino, | |
1008 | struct xfs_btree_cur **ocur, | |
1009 | struct xfs_inobt_rec_incore *rec) | |
1010 | { | |
1011 | struct xfs_btree_cur *lcur = *ocur; /* left search cursor */ | |
1012 | struct xfs_btree_cur *rcur; /* right search cursor */ | |
1013 | struct xfs_inobt_rec_incore rrec; | |
1014 | int error; | |
1015 | int i, j; | |
1016 | ||
1017 | error = xfs_inobt_lookup(lcur, pagino, XFS_LOOKUP_LE, &i); | |
1018 | if (error) | |
1019 | return error; | |
1020 | ||
1021 | if (i == 1) { | |
1022 | error = xfs_inobt_get_rec(lcur, rec, &i); | |
1023 | if (error) | |
1024 | return error; | |
19ebedcf | 1025 | XFS_WANT_CORRUPTED_RETURN(lcur->bc_mp, i == 1); |
ff105f75 DC |
1026 | |
1027 | /* | |
1028 | * See if we've landed in the parent inode record. The finobt | |
1029 | * only tracks chunks with at least one free inode, so record | |
1030 | * existence is enough. | |
1031 | */ | |
1032 | if (pagino >= rec->ir_startino && | |
1033 | pagino < (rec->ir_startino + XFS_INODES_PER_CHUNK)) | |
1034 | return 0; | |
1035 | } | |
1036 | ||
1037 | error = xfs_btree_dup_cursor(lcur, &rcur); | |
1038 | if (error) | |
1039 | return error; | |
1040 | ||
1041 | error = xfs_inobt_lookup(rcur, pagino, XFS_LOOKUP_GE, &j); | |
1042 | if (error) | |
1043 | goto error_rcur; | |
1044 | if (j == 1) { | |
1045 | error = xfs_inobt_get_rec(rcur, &rrec, &j); | |
1046 | if (error) | |
1047 | goto error_rcur; | |
19ebedcf | 1048 | XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, j == 1, error_rcur); |
ff105f75 DC |
1049 | } |
1050 | ||
19ebedcf | 1051 | XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, i == 1 || j == 1, error_rcur); |
ff105f75 DC |
1052 | if (i == 1 && j == 1) { |
1053 | /* | |
1054 | * Both the left and right records are valid. Choose the closer | |
1055 | * inode chunk to the target. | |
1056 | */ | |
1057 | if ((pagino - rec->ir_startino + XFS_INODES_PER_CHUNK - 1) > | |
1058 | (rrec.ir_startino - pagino)) { | |
1059 | *rec = rrec; | |
1060 | xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR); | |
1061 | *ocur = rcur; | |
1062 | } else { | |
1063 | xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR); | |
1064 | } | |
1065 | } else if (j == 1) { | |
1066 | /* only the right record is valid */ | |
1067 | *rec = rrec; | |
1068 | xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR); | |
1069 | *ocur = rcur; | |
1070 | } else if (i == 1) { | |
1071 | /* only the left record is valid */ | |
1072 | xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR); | |
1073 | } | |
1074 | ||
1075 | return 0; | |
1076 | ||
1077 | error_rcur: | |
1078 | xfs_btree_del_cursor(rcur, XFS_BTREE_ERROR); | |
1079 | return error; | |
1080 | } | |
1081 | ||
1082 | /* | |
1083 | * Use the free inode btree to find a free inode based on a newino hint. If | |
1084 | * the hint is NULL, find the first free inode in the AG. | |
1085 | */ | |
1086 | STATIC int | |
1087 | xfs_dialloc_ag_finobt_newino( | |
1088 | struct xfs_agi *agi, | |
1089 | struct xfs_btree_cur *cur, | |
1090 | struct xfs_inobt_rec_incore *rec) | |
1091 | { | |
1092 | int error; | |
1093 | int i; | |
1094 | ||
1095 | if (agi->agi_newino != cpu_to_be32(NULLAGINO)) { | |
1096 | error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino), | |
1097 | XFS_LOOKUP_EQ, &i); | |
1098 | if (error) | |
1099 | return error; | |
1100 | if (i == 1) { | |
1101 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1102 | if (error) | |
1103 | return error; | |
19ebedcf | 1104 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
ff105f75 DC |
1105 | return 0; |
1106 | } | |
1107 | } | |
1108 | ||
1109 | /* | |
1110 | * Find the first inode available in the AG. | |
1111 | */ | |
1112 | error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i); | |
1113 | if (error) | |
1114 | return error; | |
19ebedcf | 1115 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
ff105f75 DC |
1116 | |
1117 | error = xfs_inobt_get_rec(cur, rec, &i); | |
1118 | if (error) | |
1119 | return error; | |
19ebedcf | 1120 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
ff105f75 DC |
1121 | |
1122 | return 0; | |
1123 | } | |
1124 | ||
1125 | /* | |
1126 | * Update the inobt based on a modification made to the finobt. Also ensure that | |
1127 | * the records from both trees are equivalent post-modification. | |
1128 | */ | |
1129 | STATIC int | |
1130 | xfs_dialloc_ag_update_inobt( | |
1131 | struct xfs_btree_cur *cur, /* inobt cursor */ | |
1132 | struct xfs_inobt_rec_incore *frec, /* finobt record */ | |
1133 | int offset) /* inode offset */ | |
1134 | { | |
1135 | struct xfs_inobt_rec_incore rec; | |
1136 | int error; | |
1137 | int i; | |
1138 | ||
1139 | error = xfs_inobt_lookup(cur, frec->ir_startino, XFS_LOOKUP_EQ, &i); | |
1140 | if (error) | |
1141 | return error; | |
19ebedcf | 1142 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
ff105f75 DC |
1143 | |
1144 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
1145 | if (error) | |
1146 | return error; | |
19ebedcf | 1147 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
ff105f75 DC |
1148 | ASSERT((XFS_AGINO_TO_OFFSET(cur->bc_mp, rec.ir_startino) % |
1149 | XFS_INODES_PER_CHUNK) == 0); | |
1150 | ||
1151 | rec.ir_free &= ~XFS_INOBT_MASK(offset); | |
1152 | rec.ir_freecount--; | |
1153 | ||
19ebedcf | 1154 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, (rec.ir_free == frec->ir_free) && |
ff105f75 DC |
1155 | (rec.ir_freecount == frec->ir_freecount)); |
1156 | ||
5a35bf2c | 1157 | return xfs_inobt_update(cur, &rec); |
ff105f75 DC |
1158 | } |
1159 | ||
1160 | /* | |
1161 | * Allocate an inode using the free inode btree, if available. Otherwise, fall | |
1162 | * back to the inobt search algorithm. | |
1163 | * | |
1164 | * The caller selected an AG for us, and made sure that free inodes are | |
1165 | * available. | |
1166 | */ | |
88fc7306 BF |
1167 | STATIC int |
1168 | xfs_dialloc_ag( | |
1169 | struct xfs_trans *tp, | |
1170 | struct xfs_buf *agbp, | |
1171 | xfs_ino_t parent, | |
1172 | xfs_ino_t *inop) | |
1173 | { | |
1174 | struct xfs_mount *mp = tp->t_mountp; | |
1175 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
1176 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
1177 | xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent); | |
1178 | xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent); | |
1179 | struct xfs_perag *pag; | |
ff105f75 DC |
1180 | struct xfs_btree_cur *cur; /* finobt cursor */ |
1181 | struct xfs_btree_cur *icur; /* inobt cursor */ | |
88fc7306 | 1182 | struct xfs_inobt_rec_incore rec; |
88fc7306 BF |
1183 | xfs_ino_t ino; |
1184 | int error; | |
1185 | int offset; | |
ff105f75 | 1186 | int i; |
88fc7306 BF |
1187 | |
1188 | if (!xfs_sb_version_hasfinobt(&mp->m_sb)) | |
ff105f75 | 1189 | return xfs_dialloc_ag_inobt(tp, agbp, parent, inop); |
88fc7306 BF |
1190 | |
1191 | pag = xfs_perag_get(mp, agno); | |
1192 | ||
1193 | /* | |
1194 | * If pagino is 0 (this is the root inode allocation) use newino. | |
1195 | * This must work because we've just allocated some. | |
1196 | */ | |
1197 | if (!pagino) | |
1198 | pagino = be32_to_cpu(agi->agi_newino); | |
1199 | ||
1200 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO); | |
1201 | ||
1202 | error = xfs_check_agi_freecount(cur, agi); | |
1203 | if (error) | |
1204 | goto error_cur; | |
1205 | ||
ff105f75 DC |
1206 | /* |
1207 | * The search algorithm depends on whether we're in the same AG as the | |
1208 | * parent. If so, find the closest available inode to the parent. If | |
1209 | * not, consider the agi hint or find the first free inode in the AG. | |
1210 | */ | |
1211 | if (agno == pagno) | |
1212 | error = xfs_dialloc_ag_finobt_near(pagino, &cur, &rec); | |
1213 | else | |
1214 | error = xfs_dialloc_ag_finobt_newino(agi, cur, &rec); | |
1215 | if (error) | |
1216 | goto error_cur; | |
88fc7306 | 1217 | |
01792d3b | 1218 | offset = xfs_inobt_first_free_inode(&rec); |
88fc7306 BF |
1219 | ASSERT(offset >= 0); |
1220 | ASSERT(offset < XFS_INODES_PER_CHUNK); | |
1221 | ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) % | |
1222 | XFS_INODES_PER_CHUNK) == 0); | |
1223 | ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset); | |
1224 | ||
1225 | /* | |
1226 | * Modify or remove the finobt record. | |
1227 | */ | |
1228 | rec.ir_free &= ~XFS_INOBT_MASK(offset); | |
1229 | rec.ir_freecount--; | |
ff105f75 | 1230 | if (rec.ir_freecount) |
88fc7306 BF |
1231 | error = xfs_inobt_update(cur, &rec); |
1232 | else | |
1233 | error = xfs_btree_delete(cur, &i); | |
1234 | if (error) | |
1235 | goto error_cur; | |
1236 | ||
1237 | /* | |
ff105f75 DC |
1238 | * The finobt has now been updated appropriately. We haven't updated the |
1239 | * agi and superblock yet, so we can create an inobt cursor and validate | |
1240 | * the original freecount. If all is well, make the equivalent update to | |
1241 | * the inobt using the finobt record and offset information. | |
88fc7306 | 1242 | */ |
ff105f75 | 1243 | icur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
88fc7306 | 1244 | |
ff105f75 | 1245 | error = xfs_check_agi_freecount(icur, agi); |
88fc7306 | 1246 | if (error) |
ff105f75 | 1247 | goto error_icur; |
88fc7306 | 1248 | |
ff105f75 | 1249 | error = xfs_dialloc_ag_update_inobt(icur, &rec, offset); |
88fc7306 | 1250 | if (error) |
ff105f75 | 1251 | goto error_icur; |
88fc7306 BF |
1252 | |
1253 | /* | |
ff105f75 DC |
1254 | * Both trees have now been updated. We must update the perag and |
1255 | * superblock before we can check the freecount for each btree. | |
88fc7306 BF |
1256 | */ |
1257 | be32_add_cpu(&agi->agi_freecount, -1); | |
1258 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); | |
1259 | pag->pagi_freecount--; | |
1260 | ||
1261 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1); | |
1262 | ||
ff105f75 | 1263 | error = xfs_check_agi_freecount(icur, agi); |
88fc7306 | 1264 | if (error) |
ff105f75 | 1265 | goto error_icur; |
88fc7306 BF |
1266 | error = xfs_check_agi_freecount(cur, agi); |
1267 | if (error) | |
ff105f75 | 1268 | goto error_icur; |
88fc7306 | 1269 | |
ff105f75 | 1270 | xfs_btree_del_cursor(icur, XFS_BTREE_NOERROR); |
88fc7306 BF |
1271 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
1272 | xfs_perag_put(pag); | |
1273 | *inop = ino; | |
1274 | return 0; | |
1275 | ||
ff105f75 DC |
1276 | error_icur: |
1277 | xfs_btree_del_cursor(icur, XFS_BTREE_ERROR); | |
88fc7306 BF |
1278 | error_cur: |
1279 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
1280 | xfs_perag_put(pag); | |
1281 | return error; | |
1282 | } | |
1283 | ||
a2ceac1f DC |
1284 | /* |
1285 | * Allocate an inode on disk. | |
1286 | * | |
1287 | * Mode is used to tell whether the new inode will need space, and whether it | |
1288 | * is a directory. | |
1289 | * | |
1290 | * This function is designed to be called twice if it has to do an allocation | |
1291 | * to make more free inodes. On the first call, *IO_agbp should be set to NULL. | |
1292 | * If an inode is available without having to performn an allocation, an inode | |
1293 | * number is returned. In this case, *IO_agbp is set to NULL. If an allocation | |
1294 | * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp. | |
1295 | * The caller should then commit the current transaction, allocate a | |
1296 | * new transaction, and call xfs_dialloc() again, passing in the previous value | |
1297 | * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI | |
1298 | * buffer is locked across the two calls, the second call is guaranteed to have | |
1299 | * a free inode available. | |
1300 | * | |
1301 | * Once we successfully pick an inode its number is returned and the on-disk | |
1302 | * data structures are updated. The inode itself is not read in, since doing so | |
1303 | * would break ordering constraints with xfs_reclaim. | |
1304 | */ | |
1305 | int | |
1306 | xfs_dialloc( | |
1307 | struct xfs_trans *tp, | |
1308 | xfs_ino_t parent, | |
1309 | umode_t mode, | |
1310 | int okalloc, | |
1311 | struct xfs_buf **IO_agbp, | |
1312 | xfs_ino_t *inop) | |
1313 | { | |
1314 | struct xfs_mount *mp = tp->t_mountp; | |
1315 | struct xfs_buf *agbp; | |
1316 | xfs_agnumber_t agno; | |
1317 | int error; | |
1318 | int ialloced; | |
1319 | int noroom = 0; | |
1320 | xfs_agnumber_t start_agno; | |
1321 | struct xfs_perag *pag; | |
1322 | ||
1323 | if (*IO_agbp) { | |
1324 | /* | |
1325 | * If the caller passes in a pointer to the AGI buffer, | |
1326 | * continue where we left off before. In this case, we | |
1327 | * know that the allocation group has free inodes. | |
1328 | */ | |
1329 | agbp = *IO_agbp; | |
1330 | goto out_alloc; | |
1331 | } | |
1332 | ||
1333 | /* | |
1334 | * We do not have an agbp, so select an initial allocation | |
1335 | * group for inode allocation. | |
1336 | */ | |
1337 | start_agno = xfs_ialloc_ag_select(tp, parent, mode, okalloc); | |
1338 | if (start_agno == NULLAGNUMBER) { | |
1339 | *inop = NULLFSINO; | |
1340 | return 0; | |
1341 | } | |
1342 | ||
1343 | /* | |
1344 | * If we have already hit the ceiling of inode blocks then clear | |
1345 | * okalloc so we scan all available agi structures for a free | |
1346 | * inode. | |
1347 | */ | |
1348 | if (mp->m_maxicount && | |
19ebedcf DC |
1349 | percpu_counter_read(&mp->m_icount) + mp->m_ialloc_inos > |
1350 | mp->m_maxicount) { | |
a2ceac1f DC |
1351 | noroom = 1; |
1352 | okalloc = 0; | |
1353 | } | |
1354 | ||
1355 | /* | |
1356 | * Loop until we find an allocation group that either has free inodes | |
1357 | * or in which we can allocate some inodes. Iterate through the | |
1358 | * allocation groups upward, wrapping at the end. | |
1359 | */ | |
1360 | agno = start_agno; | |
1361 | for (;;) { | |
1362 | pag = xfs_perag_get(mp, agno); | |
1363 | if (!pag->pagi_inodeok) { | |
1364 | xfs_ialloc_next_ag(mp); | |
1365 | goto nextag; | |
1366 | } | |
1367 | ||
1368 | if (!pag->pagi_init) { | |
1369 | error = xfs_ialloc_pagi_init(mp, tp, agno); | |
1370 | if (error) | |
1371 | goto out_error; | |
1372 | } | |
1373 | ||
1374 | /* | |
1375 | * Do a first racy fast path check if this AG is usable. | |
1376 | */ | |
1377 | if (!pag->pagi_freecount && !okalloc) | |
1378 | goto nextag; | |
1379 | ||
1380 | /* | |
1381 | * Then read in the AGI buffer and recheck with the AGI buffer | |
1382 | * lock held. | |
1383 | */ | |
1384 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); | |
1385 | if (error) | |
1386 | goto out_error; | |
1387 | ||
1388 | if (pag->pagi_freecount) { | |
1389 | xfs_perag_put(pag); | |
1390 | goto out_alloc; | |
1391 | } | |
1392 | ||
1393 | if (!okalloc) | |
1394 | goto nextag_relse_buffer; | |
1395 | ||
1396 | ||
1397 | error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced); | |
1398 | if (error) { | |
1399 | xfs_trans_brelse(tp, agbp); | |
1400 | ||
12b53197 | 1401 | if (error != -ENOSPC) |
a2ceac1f DC |
1402 | goto out_error; |
1403 | ||
1404 | xfs_perag_put(pag); | |
1405 | *inop = NULLFSINO; | |
1406 | return 0; | |
1407 | } | |
1408 | ||
1409 | if (ialloced) { | |
1410 | /* | |
1411 | * We successfully allocated some inodes, return | |
1412 | * the current context to the caller so that it | |
1413 | * can commit the current transaction and call | |
1414 | * us again where we left off. | |
1415 | */ | |
1416 | ASSERT(pag->pagi_freecount > 0); | |
1417 | xfs_perag_put(pag); | |
1418 | ||
1419 | *IO_agbp = agbp; | |
1420 | *inop = NULLFSINO; | |
1421 | return 0; | |
1422 | } | |
1423 | ||
1424 | nextag_relse_buffer: | |
1425 | xfs_trans_brelse(tp, agbp); | |
1426 | nextag: | |
1427 | xfs_perag_put(pag); | |
1428 | if (++agno == mp->m_sb.sb_agcount) | |
1429 | agno = 0; | |
1430 | if (agno == start_agno) { | |
1431 | *inop = NULLFSINO; | |
12b53197 | 1432 | return noroom ? -ENOSPC : 0; |
a2ceac1f DC |
1433 | } |
1434 | } | |
1435 | ||
1436 | out_alloc: | |
1437 | *IO_agbp = NULL; | |
1438 | return xfs_dialloc_ag(tp, agbp, parent, inop); | |
1439 | out_error: | |
1440 | xfs_perag_put(pag); | |
1e68581b | 1441 | return error; |
a2ceac1f DC |
1442 | } |
1443 | ||
eb9a297a BF |
1444 | STATIC int |
1445 | xfs_difree_inobt( | |
1446 | struct xfs_mount *mp, | |
1447 | struct xfs_trans *tp, | |
1448 | struct xfs_buf *agbp, | |
1449 | xfs_agino_t agino, | |
1450 | struct xfs_bmap_free *flist, | |
1451 | int *deleted, | |
1452 | xfs_ino_t *first_ino, | |
1453 | struct xfs_inobt_rec_incore *orec) | |
3439d03a | 1454 | { |
eb9a297a BF |
1455 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); |
1456 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
1457 | struct xfs_perag *pag; | |
1458 | struct xfs_btree_cur *cur; | |
1459 | struct xfs_inobt_rec_incore rec; | |
1460 | int ilen; | |
1461 | int error; | |
1462 | int i; | |
1463 | int off; | |
3439d03a | 1464 | |
3439d03a | 1465 | ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC)); |
eb9a297a BF |
1466 | ASSERT(XFS_AGINO_TO_AGBNO(mp, agino) < be32_to_cpu(agi->agi_length)); |
1467 | ||
3439d03a DC |
1468 | /* |
1469 | * Initialize the cursor. | |
1470 | */ | |
70eb7337 | 1471 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
3439d03a DC |
1472 | |
1473 | error = xfs_check_agi_freecount(cur, agi); | |
1474 | if (error) | |
1475 | goto error0; | |
1476 | ||
1477 | /* | |
1478 | * Look for the entry describing this inode. | |
1479 | */ | |
1480 | if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) { | |
1481 | xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.", | |
1482 | __func__, error); | |
1483 | goto error0; | |
1484 | } | |
19ebedcf | 1485 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
3439d03a DC |
1486 | error = xfs_inobt_get_rec(cur, &rec, &i); |
1487 | if (error) { | |
1488 | xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.", | |
1489 | __func__, error); | |
1490 | goto error0; | |
1491 | } | |
19ebedcf | 1492 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
3439d03a DC |
1493 | /* |
1494 | * Get the offset in the inode chunk. | |
1495 | */ | |
1496 | off = agino - rec.ir_startino; | |
1497 | ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK); | |
1498 | ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off))); | |
1499 | /* | |
1500 | * Mark the inode free & increment the count. | |
1501 | */ | |
1502 | rec.ir_free |= XFS_INOBT_MASK(off); | |
1503 | rec.ir_freecount++; | |
1504 | ||
1505 | /* | |
f6580bcf BF |
1506 | * When an inode chunk is free, it becomes eligible for removal. Don't |
1507 | * remove the chunk if the block size is large enough for multiple inode | |
1508 | * chunks (that might not be free). | |
3439d03a DC |
1509 | */ |
1510 | if (!(mp->m_flags & XFS_MOUNT_IKEEP) && | |
f6580bcf BF |
1511 | rec.ir_free == XFS_INOBT_ALL_FREE && |
1512 | mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK) { | |
3439d03a | 1513 | |
66260204 | 1514 | *deleted = 1; |
3439d03a DC |
1515 | *first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino); |
1516 | ||
1517 | /* | |
1518 | * Remove the inode cluster from the AGI B+Tree, adjust the | |
1519 | * AGI and Superblock inode counts, and mark the disk space | |
1520 | * to be freed when the transaction is committed. | |
1521 | */ | |
f6580bcf | 1522 | ilen = rec.ir_freecount; |
3439d03a DC |
1523 | be32_add_cpu(&agi->agi_count, -ilen); |
1524 | be32_add_cpu(&agi->agi_freecount, -(ilen - 1)); | |
1525 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT); | |
1526 | pag = xfs_perag_get(mp, agno); | |
1527 | pag->pagi_freecount -= ilen - 1; | |
1528 | xfs_perag_put(pag); | |
1529 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen); | |
1530 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1)); | |
1531 | ||
1532 | if ((error = xfs_btree_delete(cur, &i))) { | |
1533 | xfs_warn(mp, "%s: xfs_btree_delete returned error %d.", | |
1534 | __func__, error); | |
1535 | goto error0; | |
1536 | } | |
1537 | ||
ff105f75 DC |
1538 | xfs_bmap_add_free(XFS_AGB_TO_FSB(mp, agno, |
1539 | XFS_AGINO_TO_AGBNO(mp, rec.ir_startino)), | |
1540 | mp->m_ialloc_blks, flist, mp); | |
3439d03a | 1541 | } else { |
66260204 | 1542 | *deleted = 0; |
3439d03a DC |
1543 | |
1544 | error = xfs_inobt_update(cur, &rec); | |
1545 | if (error) { | |
1546 | xfs_warn(mp, "%s: xfs_inobt_update returned error %d.", | |
1547 | __func__, error); | |
1548 | goto error0; | |
1549 | } | |
1550 | ||
1551 | /* | |
1552 | * Change the inode free counts and log the ag/sb changes. | |
1553 | */ | |
1554 | be32_add_cpu(&agi->agi_freecount, 1); | |
1555 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); | |
1556 | pag = xfs_perag_get(mp, agno); | |
1557 | pag->pagi_freecount++; | |
1558 | xfs_perag_put(pag); | |
1559 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1); | |
1560 | } | |
1561 | ||
1562 | error = xfs_check_agi_freecount(cur, agi); | |
1563 | if (error) | |
1564 | goto error0; | |
1565 | ||
eb9a297a | 1566 | *orec = rec; |
3439d03a DC |
1567 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
1568 | return 0; | |
1569 | ||
1570 | error0: | |
1571 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
1572 | return error; | |
1573 | } | |
1574 | ||
1bb93fd1 BF |
1575 | /* |
1576 | * Free an inode in the free inode btree. | |
1577 | */ | |
1578 | STATIC int | |
1579 | xfs_difree_finobt( | |
1580 | struct xfs_mount *mp, | |
1581 | struct xfs_trans *tp, | |
1582 | struct xfs_buf *agbp, | |
1583 | xfs_agino_t agino, | |
1584 | struct xfs_inobt_rec_incore *ibtrec) /* inobt record */ | |
1585 | { | |
1586 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); | |
1587 | xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno); | |
1588 | struct xfs_btree_cur *cur; | |
1589 | struct xfs_inobt_rec_incore rec; | |
1590 | int offset = agino - ibtrec->ir_startino; | |
1591 | int error; | |
1592 | int i; | |
1593 | ||
1594 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO); | |
1595 | ||
1596 | error = xfs_inobt_lookup(cur, ibtrec->ir_startino, XFS_LOOKUP_EQ, &i); | |
1597 | if (error) | |
1598 | goto error; | |
1599 | if (i == 0) { | |
1600 | /* | |
1601 | * If the record does not exist in the finobt, we must have just | |
1602 | * freed an inode in a previously fully allocated chunk. If not, | |
1603 | * something is out of sync. | |
1604 | */ | |
19ebedcf | 1605 | XFS_WANT_CORRUPTED_GOTO(mp, ibtrec->ir_freecount == 1, error); |
1bb93fd1 BF |
1606 | |
1607 | error = xfs_inobt_insert_rec(cur, ibtrec->ir_freecount, | |
1608 | ibtrec->ir_free, &i); | |
1609 | if (error) | |
1610 | goto error; | |
1611 | ASSERT(i == 1); | |
1612 | ||
1613 | goto out; | |
1614 | } | |
1615 | ||
1616 | /* | |
ff105f75 DC |
1617 | * Read and update the existing record. We could just copy the ibtrec |
1618 | * across here, but that would defeat the purpose of having redundant | |
1619 | * metadata. By making the modifications independently, we can catch | |
1620 | * corruptions that we wouldn't see if we just copied from one record | |
1621 | * to another. | |
1bb93fd1 BF |
1622 | */ |
1623 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
1624 | if (error) | |
1625 | goto error; | |
19ebedcf | 1626 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error); |
1bb93fd1 BF |
1627 | |
1628 | rec.ir_free |= XFS_INOBT_MASK(offset); | |
1629 | rec.ir_freecount++; | |
1630 | ||
19ebedcf | 1631 | XFS_WANT_CORRUPTED_GOTO(mp, (rec.ir_free == ibtrec->ir_free) && |
1bb93fd1 BF |
1632 | (rec.ir_freecount == ibtrec->ir_freecount), |
1633 | error); | |
1634 | ||
1635 | /* | |
1636 | * The content of inobt records should always match between the inobt | |
1637 | * and finobt. The lifecycle of records in the finobt is different from | |
1638 | * the inobt in that the finobt only tracks records with at least one | |
ff105f75 DC |
1639 | * free inode. Hence, if all of the inodes are free and we aren't |
1640 | * keeping inode chunks permanently on disk, remove the record. | |
1641 | * Otherwise, update the record with the new information. | |
f6580bcf BF |
1642 | * |
1643 | * Note that we currently can't free chunks when the block size is large | |
1644 | * enough for multiple chunks. Leave the finobt record to remain in sync | |
1645 | * with the inobt. | |
1bb93fd1 | 1646 | */ |
f6580bcf BF |
1647 | if (rec.ir_free == XFS_INOBT_ALL_FREE && |
1648 | mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK && | |
1bb93fd1 | 1649 | !(mp->m_flags & XFS_MOUNT_IKEEP)) { |
1bb93fd1 BF |
1650 | error = xfs_btree_delete(cur, &i); |
1651 | if (error) | |
1652 | goto error; | |
1653 | ASSERT(i == 1); | |
1654 | } else { | |
1bb93fd1 BF |
1655 | error = xfs_inobt_update(cur, &rec); |
1656 | if (error) | |
1657 | goto error; | |
1658 | } | |
1659 | ||
1660 | out: | |
1661 | error = xfs_check_agi_freecount(cur, agi); | |
1662 | if (error) | |
1663 | goto error; | |
1664 | ||
1665 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
1666 | return 0; | |
1667 | ||
1668 | error: | |
1669 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
1670 | return error; | |
1671 | } | |
1672 | ||
eb9a297a BF |
1673 | /* |
1674 | * Free disk inode. Carefully avoids touching the incore inode, all | |
1675 | * manipulations incore are the caller's responsibility. | |
1676 | * The on-disk inode is not changed by this operation, only the | |
1677 | * btree (free inode mask) is changed. | |
1678 | */ | |
1679 | int | |
1680 | xfs_difree( | |
1681 | struct xfs_trans *tp, /* transaction pointer */ | |
1682 | xfs_ino_t inode, /* inode to be freed */ | |
1683 | struct xfs_bmap_free *flist, /* extents to free */ | |
1684 | int *deleted,/* set if inode cluster was deleted */ | |
1685 | xfs_ino_t *first_ino)/* first inode in deleted cluster */ | |
1686 | { | |
1687 | /* REFERENCED */ | |
1688 | xfs_agblock_t agbno; /* block number containing inode */ | |
1689 | struct xfs_buf *agbp; /* buffer for allocation group header */ | |
1690 | xfs_agino_t agino; /* allocation group inode number */ | |
1691 | xfs_agnumber_t agno; /* allocation group number */ | |
1692 | int error; /* error return value */ | |
1693 | struct xfs_mount *mp; /* mount structure for filesystem */ | |
1694 | struct xfs_inobt_rec_incore rec;/* btree record */ | |
1695 | ||
1696 | mp = tp->t_mountp; | |
1697 | ||
1698 | /* | |
1699 | * Break up inode number into its components. | |
1700 | */ | |
1701 | agno = XFS_INO_TO_AGNO(mp, inode); | |
1702 | if (agno >= mp->m_sb.sb_agcount) { | |
1703 | xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).", | |
1704 | __func__, agno, mp->m_sb.sb_agcount); | |
1705 | ASSERT(0); | |
12b53197 | 1706 | return -EINVAL; |
eb9a297a BF |
1707 | } |
1708 | agino = XFS_INO_TO_AGINO(mp, inode); | |
1709 | if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
1710 | xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).", | |
1711 | __func__, (unsigned long long)inode, | |
1712 | (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino)); | |
1713 | ASSERT(0); | |
12b53197 | 1714 | return -EINVAL; |
eb9a297a BF |
1715 | } |
1716 | agbno = XFS_AGINO_TO_AGBNO(mp, agino); | |
1717 | if (agbno >= mp->m_sb.sb_agblocks) { | |
1718 | xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).", | |
1719 | __func__, agbno, mp->m_sb.sb_agblocks); | |
1720 | ASSERT(0); | |
12b53197 | 1721 | return -EINVAL; |
eb9a297a BF |
1722 | } |
1723 | /* | |
1724 | * Get the allocation group header. | |
1725 | */ | |
1726 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); | |
1727 | if (error) { | |
1728 | xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.", | |
1729 | __func__, error); | |
1730 | return error; | |
1731 | } | |
1732 | ||
1733 | /* | |
1734 | * Fix up the inode allocation btree. | |
1735 | */ | |
1736 | error = xfs_difree_inobt(mp, tp, agbp, agino, flist, deleted, first_ino, | |
1737 | &rec); | |
1738 | if (error) | |
1739 | goto error0; | |
1740 | ||
1bb93fd1 BF |
1741 | /* |
1742 | * Fix up the free inode btree. | |
1743 | */ | |
1744 | if (xfs_sb_version_hasfinobt(&mp->m_sb)) { | |
1745 | error = xfs_difree_finobt(mp, tp, agbp, agino, &rec); | |
1746 | if (error) | |
1747 | goto error0; | |
1748 | } | |
1749 | ||
eb9a297a BF |
1750 | return 0; |
1751 | ||
1752 | error0: | |
1753 | return error; | |
1754 | } | |
1755 | ||
56b2de80 DC |
1756 | STATIC int |
1757 | xfs_imap_lookup( | |
1758 | struct xfs_mount *mp, | |
1759 | struct xfs_trans *tp, | |
1760 | xfs_agnumber_t agno, | |
1761 | xfs_agino_t agino, | |
1762 | xfs_agblock_t agbno, | |
1763 | xfs_agblock_t *chunk_agbno, | |
1764 | xfs_agblock_t *offset_agbno, | |
1765 | int flags) | |
1766 | { | |
1767 | struct xfs_inobt_rec_incore rec; | |
1768 | struct xfs_btree_cur *cur; | |
1769 | struct xfs_buf *agbp; | |
1770 | int error; | |
1771 | int i; | |
1772 | ||
1773 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); | |
1774 | if (error) { | |
a2ceac1f DC |
1775 | xfs_alert(mp, |
1776 | "%s: xfs_ialloc_read_agi() returned error %d, agno %d", | |
1777 | __func__, error, agno); | |
56b2de80 DC |
1778 | return error; |
1779 | } | |
1780 | ||
1781 | /* | |
1782 | * Lookup the inode record for the given agino. If the record cannot be | |
1783 | * found, then it's an invalid inode number and we should abort. Once | |
1784 | * we have a record, we need to ensure it contains the inode number | |
1785 | * we are looking up. | |
1786 | */ | |
70eb7337 | 1787 | cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO); |
56b2de80 DC |
1788 | error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i); |
1789 | if (!error) { | |
1790 | if (i) | |
1791 | error = xfs_inobt_get_rec(cur, &rec, &i); | |
1792 | if (!error && i == 0) | |
12b53197 | 1793 | error = -EINVAL; |
56b2de80 DC |
1794 | } |
1795 | ||
1796 | xfs_trans_brelse(tp, agbp); | |
1797 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
1798 | if (error) | |
1799 | return error; | |
1800 | ||
1801 | /* check that the returned record contains the required inode */ | |
1802 | if (rec.ir_startino > agino || | |
ff105f75 | 1803 | rec.ir_startino + mp->m_ialloc_inos <= agino) |
12b53197 | 1804 | return -EINVAL; |
56b2de80 DC |
1805 | |
1806 | /* for untrusted inodes check it is allocated first */ | |
1807 | if ((flags & XFS_IGET_UNTRUSTED) && | |
1808 | (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino))) | |
12b53197 | 1809 | return -EINVAL; |
56b2de80 DC |
1810 | |
1811 | *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino); | |
1812 | *offset_agbno = agbno - *chunk_agbno; | |
1813 | return 0; | |
1814 | } | |
2bd0ea18 NS |
1815 | |
1816 | /* | |
56b2de80 | 1817 | * Return the location of the inode in imap, for mapping it into a buffer. |
2bd0ea18 | 1818 | */ |
2bd0ea18 | 1819 | int |
56b2de80 DC |
1820 | xfs_imap( |
1821 | xfs_mount_t *mp, /* file system mount structure */ | |
1822 | xfs_trans_t *tp, /* transaction pointer */ | |
2bd0ea18 | 1823 | xfs_ino_t ino, /* inode to locate */ |
56b2de80 DC |
1824 | struct xfs_imap *imap, /* location map structure */ |
1825 | uint flags) /* flags for inode btree lookup */ | |
2bd0ea18 NS |
1826 | { |
1827 | xfs_agblock_t agbno; /* block number of inode in the alloc group */ | |
2bd0ea18 NS |
1828 | xfs_agino_t agino; /* inode number within alloc group */ |
1829 | xfs_agnumber_t agno; /* allocation group number */ | |
1830 | int blks_per_cluster; /* num blocks per inode cluster */ | |
1831 | xfs_agblock_t chunk_agbno; /* first block in inode chunk */ | |
2bd0ea18 | 1832 | xfs_agblock_t cluster_agbno; /* first block in inode cluster */ |
2bd0ea18 | 1833 | int error; /* error code */ |
dfc130f3 | 1834 | int offset; /* index of inode in its buffer */ |
6bddecbc | 1835 | xfs_agblock_t offset_agbno; /* blks from chunk start to inode */ |
2bd0ea18 NS |
1836 | |
1837 | ASSERT(ino != NULLFSINO); | |
56b2de80 | 1838 | |
2bd0ea18 NS |
1839 | /* |
1840 | * Split up the inode number into its parts. | |
1841 | */ | |
1842 | agno = XFS_INO_TO_AGNO(mp, ino); | |
1843 | agino = XFS_INO_TO_AGINO(mp, ino); | |
1844 | agbno = XFS_AGINO_TO_AGBNO(mp, agino); | |
1845 | if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks || | |
63518810 NS |
1846 | ino != XFS_AGINO_TO_INO(mp, agno, agino)) { |
1847 | #ifdef DEBUG | |
56b2de80 DC |
1848 | /* |
1849 | * Don't output diagnostic information for untrusted inodes | |
1850 | * as they can be invalid without implying corruption. | |
1851 | */ | |
1852 | if (flags & XFS_IGET_UNTRUSTED) | |
12b53197 | 1853 | return -EINVAL; |
5000d01d | 1854 | if (agno >= mp->m_sb.sb_agcount) { |
a2ceac1f DC |
1855 | xfs_alert(mp, |
1856 | "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)", | |
1857 | __func__, agno, mp->m_sb.sb_agcount); | |
63518810 NS |
1858 | } |
1859 | if (agbno >= mp->m_sb.sb_agblocks) { | |
a2ceac1f DC |
1860 | xfs_alert(mp, |
1861 | "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)", | |
1862 | __func__, (unsigned long long)agbno, | |
1863 | (unsigned long)mp->m_sb.sb_agblocks); | |
63518810 NS |
1864 | } |
1865 | if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
a2ceac1f DC |
1866 | xfs_alert(mp, |
1867 | "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)", | |
1868 | __func__, ino, | |
1869 | XFS_AGINO_TO_INO(mp, agno, agino)); | |
63518810 | 1870 | } |
5e656dbb | 1871 | xfs_stack_trace(); |
63518810 | 1872 | #endif /* DEBUG */ |
12b53197 | 1873 | return -EINVAL; |
63518810 | 1874 | } |
56b2de80 | 1875 | |
ff105f75 | 1876 | blks_per_cluster = xfs_icluster_size_fsb(mp); |
56b2de80 DC |
1877 | |
1878 | /* | |
1879 | * For bulkstat and handle lookups, we have an untrusted inode number | |
1880 | * that we have to verify is valid. We cannot do this just by reading | |
1881 | * the inode buffer as it may have been unlinked and removed leaving | |
1882 | * inodes in stale state on disk. Hence we have to do a btree lookup | |
1883 | * in all cases where an untrusted inode number is passed. | |
1884 | */ | |
1885 | if (flags & XFS_IGET_UNTRUSTED) { | |
1886 | error = xfs_imap_lookup(mp, tp, agno, agino, agbno, | |
1887 | &chunk_agbno, &offset_agbno, flags); | |
1888 | if (error) | |
1889 | return error; | |
1890 | goto out_map; | |
1891 | } | |
1892 | ||
1893 | /* | |
1894 | * If the inode cluster size is the same as the blocksize or | |
1895 | * smaller we get to the buffer by simple arithmetics. | |
1896 | */ | |
ff105f75 | 1897 | if (blks_per_cluster == 1) { |
2bd0ea18 NS |
1898 | offset = XFS_INO_TO_OFFSET(mp, ino); |
1899 | ASSERT(offset < mp->m_sb.sb_inopblock); | |
56b2de80 DC |
1900 | |
1901 | imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno); | |
1902 | imap->im_len = XFS_FSB_TO_BB(mp, 1); | |
1903 | imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog); | |
2bd0ea18 NS |
1904 | return 0; |
1905 | } | |
56b2de80 DC |
1906 | |
1907 | /* | |
1908 | * If the inode chunks are aligned then use simple maths to | |
1909 | * find the location. Otherwise we have to do a btree | |
1910 | * lookup to find the location. | |
1911 | */ | |
2bd0ea18 NS |
1912 | if (mp->m_inoalign_mask) { |
1913 | offset_agbno = agbno & mp->m_inoalign_mask; | |
1914 | chunk_agbno = agbno - offset_agbno; | |
1915 | } else { | |
56b2de80 DC |
1916 | error = xfs_imap_lookup(mp, tp, agno, agino, agbno, |
1917 | &chunk_agbno, &offset_agbno, flags); | |
2bd0ea18 NS |
1918 | if (error) |
1919 | return error; | |
2bd0ea18 | 1920 | } |
56b2de80 DC |
1921 | |
1922 | out_map: | |
2bd0ea18 NS |
1923 | ASSERT(agbno >= chunk_agbno); |
1924 | cluster_agbno = chunk_agbno + | |
1925 | ((offset_agbno / blks_per_cluster) * blks_per_cluster); | |
1926 | offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) + | |
1927 | XFS_INO_TO_OFFSET(mp, ino); | |
56b2de80 DC |
1928 | |
1929 | imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno); | |
1930 | imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster); | |
1931 | imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog); | |
1932 | ||
1933 | /* | |
1934 | * If the inode number maps to a block outside the bounds | |
1935 | * of the file system then return NULL rather than calling | |
1936 | * read_buf and panicing when we get an error from the | |
1937 | * driver. | |
1938 | */ | |
1939 | if ((imap->im_blkno + imap->im_len) > | |
1940 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) { | |
a2ceac1f DC |
1941 | xfs_alert(mp, |
1942 | "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)", | |
1943 | __func__, (unsigned long long) imap->im_blkno, | |
56b2de80 DC |
1944 | (unsigned long long) imap->im_len, |
1945 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)); | |
12b53197 | 1946 | return -EINVAL; |
56b2de80 | 1947 | } |
2bd0ea18 | 1948 | return 0; |
2bd0ea18 NS |
1949 | } |
1950 | ||
1951 | /* | |
1952 | * Compute and fill in value of m_in_maxlevels. | |
1953 | */ | |
1954 | void | |
1955 | xfs_ialloc_compute_maxlevels( | |
1956 | xfs_mount_t *mp) /* file system mount structure */ | |
1957 | { | |
1958 | int level; | |
1959 | uint maxblocks; | |
1960 | uint maxleafents; | |
1961 | int minleafrecs; | |
1962 | int minnoderecs; | |
1963 | ||
1964 | maxleafents = (1LL << XFS_INO_AGINO_BITS(mp)) >> | |
1965 | XFS_INODES_PER_CHUNK_LOG; | |
1966 | minleafrecs = mp->m_alloc_mnr[0]; | |
1967 | minnoderecs = mp->m_alloc_mnr[1]; | |
1968 | maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs; | |
1969 | for (level = 1; maxblocks > 1; level++) | |
1970 | maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs; | |
1971 | mp->m_in_maxlevels = level; | |
1972 | } | |
1973 | ||
1974 | /* | |
ff105f75 DC |
1975 | * Log specified fields for the ag hdr (inode section). The growth of the agi |
1976 | * structure over time requires that we interpret the buffer as two logical | |
1977 | * regions delineated by the end of the unlinked list. This is due to the size | |
1978 | * of the hash table and its location in the middle of the agi. | |
1979 | * | |
1980 | * For example, a request to log a field before agi_unlinked and a field after | |
1981 | * agi_unlinked could cause us to log the entire hash table and use an excessive | |
1982 | * amount of log space. To avoid this behavior, log the region up through | |
1983 | * agi_unlinked in one call and the region after agi_unlinked through the end of | |
1984 | * the structure in another. | |
2bd0ea18 NS |
1985 | */ |
1986 | void | |
1987 | xfs_ialloc_log_agi( | |
1988 | xfs_trans_t *tp, /* transaction pointer */ | |
1989 | xfs_buf_t *bp, /* allocation group header buffer */ | |
1990 | int fields) /* bitmask of fields to log */ | |
1991 | { | |
1992 | int first; /* first byte number */ | |
1993 | int last; /* last byte number */ | |
1994 | static const short offsets[] = { /* field starting offsets */ | |
1995 | /* keep in sync with bit definitions */ | |
1996 | offsetof(xfs_agi_t, agi_magicnum), | |
1997 | offsetof(xfs_agi_t, agi_versionnum), | |
1998 | offsetof(xfs_agi_t, agi_seqno), | |
1999 | offsetof(xfs_agi_t, agi_length), | |
2000 | offsetof(xfs_agi_t, agi_count), | |
2001 | offsetof(xfs_agi_t, agi_root), | |
2002 | offsetof(xfs_agi_t, agi_level), | |
2003 | offsetof(xfs_agi_t, agi_freecount), | |
2004 | offsetof(xfs_agi_t, agi_newino), | |
2005 | offsetof(xfs_agi_t, agi_dirino), | |
2006 | offsetof(xfs_agi_t, agi_unlinked), | |
c0a4c227 BF |
2007 | offsetof(xfs_agi_t, agi_free_root), |
2008 | offsetof(xfs_agi_t, agi_free_level), | |
2bd0ea18 NS |
2009 | sizeof(xfs_agi_t) |
2010 | }; | |
2011 | #ifdef DEBUG | |
2012 | xfs_agi_t *agi; /* allocation group header */ | |
2013 | ||
2014 | agi = XFS_BUF_TO_AGI(bp); | |
a2ceac1f | 2015 | ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC)); |
2bd0ea18 | 2016 | #endif |
ff105f75 | 2017 | |
c0a4c227 BF |
2018 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_AGI_BUF); |
2019 | ||
2bd0ea18 | 2020 | /* |
c0a4c227 | 2021 | * Compute byte offsets for the first and last fields in the first |
ff105f75 DC |
2022 | * region and log the agi buffer. This only logs up through |
2023 | * agi_unlinked. | |
2bd0ea18 | 2024 | */ |
c0a4c227 BF |
2025 | if (fields & XFS_AGI_ALL_BITS_R1) { |
2026 | xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R1, | |
2027 | &first, &last); | |
2028 | xfs_trans_log_buf(tp, bp, first, last); | |
2029 | } | |
2030 | ||
2031 | /* | |
ff105f75 DC |
2032 | * Mask off the bits in the first region and calculate the first and |
2033 | * last field offsets for any bits in the second region. | |
c0a4c227 BF |
2034 | */ |
2035 | fields &= ~XFS_AGI_ALL_BITS_R1; | |
2036 | if (fields) { | |
2037 | xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R2, | |
2038 | &first, &last); | |
2039 | xfs_trans_log_buf(tp, bp, first, last); | |
2040 | } | |
2bd0ea18 NS |
2041 | } |
2042 | ||
56b2de80 DC |
2043 | #ifdef DEBUG |
2044 | STATIC void | |
2045 | xfs_check_agi_unlinked( | |
2046 | struct xfs_agi *agi) | |
2047 | { | |
2048 | int i; | |
2049 | ||
2050 | for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++) | |
2051 | ASSERT(agi->agi_unlinked[i]); | |
2052 | } | |
2053 | #else | |
2054 | #define xfs_check_agi_unlinked(agi) | |
2055 | #endif | |
2056 | ||
dd5b876e | 2057 | static bool |
a2ceac1f DC |
2058 | xfs_agi_verify( |
2059 | struct xfs_buf *bp) | |
2060 | { | |
2061 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
2062 | struct xfs_agi *agi = XFS_BUF_TO_AGI(bp); | |
a2ceac1f | 2063 | |
dd5b876e DC |
2064 | if (xfs_sb_version_hascrc(&mp->m_sb) && |
2065 | !uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_uuid)) | |
2066 | return false; | |
a2ceac1f DC |
2067 | /* |
2068 | * Validate the magic number of the agi block. | |
2069 | */ | |
dd5b876e DC |
2070 | if (agi->agi_magicnum != cpu_to_be32(XFS_AGI_MAGIC)) |
2071 | return false; | |
2072 | if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum))) | |
2073 | return false; | |
a2ceac1f | 2074 | |
5a35bf2c DC |
2075 | if (be32_to_cpu(agi->agi_level) > XFS_BTREE_MAXLEVELS) |
2076 | return false; | |
a2ceac1f DC |
2077 | /* |
2078 | * during growfs operations, the perag is not fully initialised, | |
2079 | * so we can't use it for any useful checking. growfs ensures we can't | |
2080 | * use it by using uncached buffers that don't have the perag attached | |
2081 | * so we can detect and avoid this problem. | |
2082 | */ | |
dd5b876e DC |
2083 | if (bp->b_pag && be32_to_cpu(agi->agi_seqno) != bp->b_pag->pag_agno) |
2084 | return false; | |
a2ceac1f | 2085 | |
a2ceac1f | 2086 | xfs_check_agi_unlinked(agi); |
dd5b876e | 2087 | return true; |
a2ceac1f DC |
2088 | } |
2089 | ||
2090 | static void | |
2091 | xfs_agi_read_verify( | |
2092 | struct xfs_buf *bp) | |
2093 | { | |
dd5b876e | 2094 | struct xfs_mount *mp = bp->b_target->bt_mount; |
dd5b876e | 2095 | |
45922933 DC |
2096 | if (xfs_sb_version_hascrc(&mp->m_sb) && |
2097 | !xfs_buf_verify_cksum(bp, XFS_AGI_CRC_OFF)) | |
12b53197 | 2098 | xfs_buf_ioerror(bp, -EFSBADCRC); |
45922933 DC |
2099 | else if (XFS_TEST_ERROR(!xfs_agi_verify(bp), mp, |
2100 | XFS_ERRTAG_IALLOC_READ_AGI, | |
2101 | XFS_RANDOM_IALLOC_READ_AGI)) | |
12b53197 | 2102 | xfs_buf_ioerror(bp, -EFSCORRUPTED); |
45922933 DC |
2103 | |
2104 | if (bp->b_error) | |
2105 | xfs_verifier_error(bp); | |
a2ceac1f DC |
2106 | } |
2107 | ||
2108 | static void | |
2109 | xfs_agi_write_verify( | |
2110 | struct xfs_buf *bp) | |
2111 | { | |
dd5b876e DC |
2112 | struct xfs_mount *mp = bp->b_target->bt_mount; |
2113 | struct xfs_buf_log_item *bip = bp->b_fspriv; | |
2114 | ||
2115 | if (!xfs_agi_verify(bp)) { | |
12b53197 | 2116 | xfs_buf_ioerror(bp, -EFSCORRUPTED); |
45922933 | 2117 | xfs_verifier_error(bp); |
dd5b876e DC |
2118 | return; |
2119 | } | |
2120 | ||
2121 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
2122 | return; | |
2123 | ||
2124 | if (bip) | |
2125 | XFS_BUF_TO_AGI(bp)->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn); | |
43b5aeed | 2126 | xfs_buf_update_cksum(bp, XFS_AGI_CRC_OFF); |
a2ceac1f DC |
2127 | } |
2128 | ||
2129 | const struct xfs_buf_ops xfs_agi_buf_ops = { | |
2130 | .verify_read = xfs_agi_read_verify, | |
2131 | .verify_write = xfs_agi_write_verify, | |
2132 | }; | |
2133 | ||
2bd0ea18 NS |
2134 | /* |
2135 | * Read in the allocation group header (inode allocation section) | |
2136 | */ | |
2137 | int | |
56b2de80 DC |
2138 | xfs_read_agi( |
2139 | struct xfs_mount *mp, /* file system mount structure */ | |
2140 | struct xfs_trans *tp, /* transaction pointer */ | |
2141 | xfs_agnumber_t agno, /* allocation group number */ | |
2142 | struct xfs_buf **bpp) /* allocation group hdr buf */ | |
2bd0ea18 | 2143 | { |
56b2de80 | 2144 | int error; |
2bd0ea18 | 2145 | |
ff105f75 | 2146 | trace_xfs_read_agi(mp, agno); |
56b2de80 | 2147 | |
ff105f75 | 2148 | ASSERT(agno != NULLAGNUMBER); |
56b2de80 | 2149 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, |
9440d84d | 2150 | XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)), |
a2ceac1f | 2151 | XFS_FSS_TO_BB(mp, 1), 0, bpp, &xfs_agi_buf_ops); |
9440d84d | 2152 | if (error) |
2bd0ea18 | 2153 | return error; |
56b2de80 | 2154 | |
a2ceac1f | 2155 | xfs_buf_set_ref(*bpp, XFS_AGI_REF); |
56b2de80 DC |
2156 | return 0; |
2157 | } | |
2158 | ||
2159 | int | |
2160 | xfs_ialloc_read_agi( | |
2161 | struct xfs_mount *mp, /* file system mount structure */ | |
2162 | struct xfs_trans *tp, /* transaction pointer */ | |
2163 | xfs_agnumber_t agno, /* allocation group number */ | |
2164 | struct xfs_buf **bpp) /* allocation group hdr buf */ | |
2165 | { | |
2166 | struct xfs_agi *agi; /* allocation group header */ | |
2167 | struct xfs_perag *pag; /* per allocation group data */ | |
2168 | int error; | |
2169 | ||
ff105f75 DC |
2170 | trace_xfs_ialloc_read_agi(mp, agno); |
2171 | ||
56b2de80 DC |
2172 | error = xfs_read_agi(mp, tp, agno, bpp); |
2173 | if (error) | |
2174 | return error; | |
2175 | ||
2176 | agi = XFS_BUF_TO_AGI(*bpp); | |
2177 | pag = xfs_perag_get(mp, agno); | |
2bd0ea18 | 2178 | if (!pag->pagi_init) { |
6e3140c7 | 2179 | pag->pagi_freecount = be32_to_cpu(agi->agi_freecount); |
cdded3d8 | 2180 | pag->pagi_count = be32_to_cpu(agi->agi_count); |
2bd0ea18 | 2181 | pag->pagi_init = 1; |
9440d84d | 2182 | } |
9440d84d | 2183 | |
56b2de80 DC |
2184 | /* |
2185 | * It's possible for these to be out of sync if | |
2186 | * we are in the middle of a forced shutdown. | |
2187 | */ | |
2188 | ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) || | |
2189 | XFS_FORCED_SHUTDOWN(mp)); | |
2190 | xfs_perag_put(pag); | |
2bd0ea18 NS |
2191 | return 0; |
2192 | } | |
cdded3d8 DC |
2193 | |
2194 | /* | |
2195 | * Read in the agi to initialise the per-ag data in the mount structure | |
2196 | */ | |
2197 | int | |
2198 | xfs_ialloc_pagi_init( | |
2199 | xfs_mount_t *mp, /* file system mount structure */ | |
2200 | xfs_trans_t *tp, /* transaction pointer */ | |
2201 | xfs_agnumber_t agno) /* allocation group number */ | |
2202 | { | |
2203 | xfs_buf_t *bp = NULL; | |
2204 | int error; | |
2205 | ||
5e656dbb BN |
2206 | error = xfs_ialloc_read_agi(mp, tp, agno, &bp); |
2207 | if (error) | |
cdded3d8 DC |
2208 | return error; |
2209 | if (bp) | |
2210 | xfs_trans_brelse(tp, bp); | |
2211 | return 0; | |
2212 | } |