Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (c) 2000-2002 Silicon Graphics, Inc. All Rights Reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or modify it | |
5 | * under the terms of version 2 of the GNU General Public License as | |
6 | * published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it would be useful, but | |
9 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. | |
11 | * | |
12 | * Further, this software is distributed without any warranty that it is | |
13 | * free of the rightful claim of any third person regarding infringement | |
14 | * or the like. Any license provided herein, whether implied or | |
15 | * otherwise, applies only to this software file. Patent licenses, if | |
16 | * any, provided herein do not apply to combinations of this program with | |
17 | * other software, or any other product whatsoever. | |
18 | * | |
19 | * You should have received a copy of the GNU General Public License along | |
20 | * with this program; if not, write the Free Software Foundation, Inc., 59 | |
21 | * Temple Place - Suite 330, Boston MA 02111-1307, USA. | |
22 | * | |
23 | * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy, | |
24 | * Mountain View, CA 94043, or: | |
25 | * | |
26 | * http://www.sgi.com | |
27 | * | |
28 | * For further information regarding this notice, see: | |
29 | * | |
30 | * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/ | |
31 | */ | |
1da177e4 | 32 | #include "xfs.h" |
a844f451 | 33 | #include "xfs_fs.h" |
1da177e4 | 34 | #include "xfs_types.h" |
a844f451 | 35 | #include "xfs_bit.h" |
1da177e4 | 36 | #include "xfs_log.h" |
a844f451 | 37 | #include "xfs_inum.h" |
1da177e4 LT |
38 | #include "xfs_trans.h" |
39 | #include "xfs_sb.h" | |
40 | #include "xfs_ag.h" | |
41 | #include "xfs_dir.h" | |
42 | #include "xfs_dir2.h" | |
43 | #include "xfs_dmapi.h" | |
44 | #include "xfs_mount.h" | |
1da177e4 | 45 | #include "xfs_bmap_btree.h" |
a844f451 | 46 | #include "xfs_alloc_btree.h" |
1da177e4 | 47 | #include "xfs_ialloc_btree.h" |
1da177e4 LT |
48 | #include "xfs_dir_sf.h" |
49 | #include "xfs_dir2_sf.h" | |
a844f451 | 50 | #include "xfs_attr_sf.h" |
1da177e4 LT |
51 | #include "xfs_dinode.h" |
52 | #include "xfs_inode.h" | |
a844f451 NS |
53 | #include "xfs_btree.h" |
54 | #include "xfs_ialloc.h" | |
1da177e4 | 55 | #include "xfs_alloc.h" |
1da177e4 LT |
56 | #include "xfs_rtalloc.h" |
57 | #include "xfs_error.h" | |
58 | #include "xfs_bmap.h" | |
59 | ||
60 | /* | |
61 | * Log specified fields for the inode given by bp and off. | |
62 | */ | |
63 | STATIC void | |
64 | xfs_ialloc_log_di( | |
65 | xfs_trans_t *tp, /* transaction pointer */ | |
66 | xfs_buf_t *bp, /* inode buffer */ | |
67 | int off, /* index of inode in buffer */ | |
68 | int fields) /* bitmask of fields to log */ | |
69 | { | |
70 | int first; /* first byte number */ | |
71 | int ioffset; /* off in bytes */ | |
72 | int last; /* last byte number */ | |
73 | xfs_mount_t *mp; /* mount point structure */ | |
74 | static const short offsets[] = { /* field offsets */ | |
75 | /* keep in sync with bits */ | |
76 | offsetof(xfs_dinode_core_t, di_magic), | |
77 | offsetof(xfs_dinode_core_t, di_mode), | |
78 | offsetof(xfs_dinode_core_t, di_version), | |
79 | offsetof(xfs_dinode_core_t, di_format), | |
80 | offsetof(xfs_dinode_core_t, di_onlink), | |
81 | offsetof(xfs_dinode_core_t, di_uid), | |
82 | offsetof(xfs_dinode_core_t, di_gid), | |
83 | offsetof(xfs_dinode_core_t, di_nlink), | |
84 | offsetof(xfs_dinode_core_t, di_projid), | |
85 | offsetof(xfs_dinode_core_t, di_pad), | |
86 | offsetof(xfs_dinode_core_t, di_atime), | |
87 | offsetof(xfs_dinode_core_t, di_mtime), | |
88 | offsetof(xfs_dinode_core_t, di_ctime), | |
89 | offsetof(xfs_dinode_core_t, di_size), | |
90 | offsetof(xfs_dinode_core_t, di_nblocks), | |
91 | offsetof(xfs_dinode_core_t, di_extsize), | |
92 | offsetof(xfs_dinode_core_t, di_nextents), | |
93 | offsetof(xfs_dinode_core_t, di_anextents), | |
94 | offsetof(xfs_dinode_core_t, di_forkoff), | |
95 | offsetof(xfs_dinode_core_t, di_aformat), | |
96 | offsetof(xfs_dinode_core_t, di_dmevmask), | |
97 | offsetof(xfs_dinode_core_t, di_dmstate), | |
98 | offsetof(xfs_dinode_core_t, di_flags), | |
99 | offsetof(xfs_dinode_core_t, di_gen), | |
100 | offsetof(xfs_dinode_t, di_next_unlinked), | |
101 | offsetof(xfs_dinode_t, di_u), | |
102 | offsetof(xfs_dinode_t, di_a), | |
103 | sizeof(xfs_dinode_t) | |
104 | }; | |
105 | ||
106 | ||
107 | ASSERT(offsetof(xfs_dinode_t, di_core) == 0); | |
108 | ASSERT((fields & (XFS_DI_U|XFS_DI_A)) == 0); | |
109 | mp = tp->t_mountp; | |
110 | /* | |
111 | * Get the inode-relative first and last bytes for these fields | |
112 | */ | |
113 | xfs_btree_offsets(fields, offsets, XFS_DI_NUM_BITS, &first, &last); | |
114 | /* | |
115 | * Convert to buffer offsets and log it. | |
116 | */ | |
117 | ioffset = off << mp->m_sb.sb_inodelog; | |
118 | first += ioffset; | |
119 | last += ioffset; | |
120 | xfs_trans_log_buf(tp, bp, first, last); | |
121 | } | |
122 | ||
123 | /* | |
124 | * Allocation group level functions. | |
125 | */ | |
126 | ||
127 | /* | |
128 | * Allocate new inodes in the allocation group specified by agbp. | |
129 | * Return 0 for success, else error code. | |
130 | */ | |
131 | STATIC int /* error code or 0 */ | |
132 | xfs_ialloc_ag_alloc( | |
133 | xfs_trans_t *tp, /* transaction pointer */ | |
134 | xfs_buf_t *agbp, /* alloc group buffer */ | |
135 | int *alloc) | |
136 | { | |
137 | xfs_agi_t *agi; /* allocation group header */ | |
138 | xfs_alloc_arg_t args; /* allocation argument structure */ | |
139 | int blks_per_cluster; /* fs blocks per inode cluster */ | |
140 | xfs_btree_cur_t *cur; /* inode btree cursor */ | |
141 | xfs_daddr_t d; /* disk addr of buffer */ | |
142 | int error; | |
143 | xfs_buf_t *fbuf; /* new free inodes' buffer */ | |
144 | xfs_dinode_t *free; /* new free inode structure */ | |
145 | int i; /* inode counter */ | |
146 | int j; /* block counter */ | |
147 | int nbufs; /* num bufs of new inodes */ | |
148 | xfs_agino_t newino; /* new first inode's number */ | |
149 | xfs_agino_t newlen; /* new number of inodes */ | |
150 | int ninodes; /* num inodes per buf */ | |
151 | xfs_agino_t thisino; /* current inode number, for loop */ | |
152 | int version; /* inode version number to use */ | |
153 | int isaligned; /* inode allocation at stripe unit */ | |
154 | /* boundary */ | |
155 | xfs_dinode_core_t dic; /* a dinode_core to copy to new */ | |
156 | /* inodes */ | |
157 | ||
158 | args.tp = tp; | |
159 | args.mp = tp->t_mountp; | |
160 | ||
161 | /* | |
162 | * Locking will ensure that we don't have two callers in here | |
163 | * at one time. | |
164 | */ | |
165 | newlen = XFS_IALLOC_INODES(args.mp); | |
166 | if (args.mp->m_maxicount && | |
167 | args.mp->m_sb.sb_icount + newlen > args.mp->m_maxicount) | |
168 | return XFS_ERROR(ENOSPC); | |
169 | args.minlen = args.maxlen = XFS_IALLOC_BLOCKS(args.mp); | |
170 | /* | |
171 | * Set the alignment for the allocation. | |
172 | * If stripe alignment is turned on then align at stripe unit | |
173 | * boundary. | |
174 | * If the cluster size is smaller than a filesystem block | |
175 | * then we're doing I/O for inodes in filesystem block size pieces, | |
176 | * so don't need alignment anyway. | |
177 | */ | |
178 | isaligned = 0; | |
179 | if (args.mp->m_sinoalign) { | |
180 | ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN)); | |
181 | args.alignment = args.mp->m_dalign; | |
182 | isaligned = 1; | |
183 | } else if (XFS_SB_VERSION_HASALIGN(&args.mp->m_sb) && | |
184 | args.mp->m_sb.sb_inoalignmt >= | |
185 | XFS_B_TO_FSBT(args.mp, XFS_INODE_CLUSTER_SIZE(args.mp))) | |
186 | args.alignment = args.mp->m_sb.sb_inoalignmt; | |
187 | else | |
188 | args.alignment = 1; | |
189 | agi = XFS_BUF_TO_AGI(agbp); | |
190 | /* | |
191 | * Need to figure out where to allocate the inode blocks. | |
192 | * Ideally they should be spaced out through the a.g. | |
193 | * For now, just allocate blocks up front. | |
194 | */ | |
195 | args.agbno = INT_GET(agi->agi_root, ARCH_CONVERT); | |
196 | args.fsbno = XFS_AGB_TO_FSB(args.mp, INT_GET(agi->agi_seqno, ARCH_CONVERT), | |
197 | args.agbno); | |
198 | /* | |
199 | * Allocate a fixed-size extent of inodes. | |
200 | */ | |
201 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
202 | args.mod = args.total = args.wasdel = args.isfl = args.userdata = | |
203 | args.minalignslop = 0; | |
204 | args.prod = 1; | |
205 | /* | |
206 | * Allow space for the inode btree to split. | |
207 | */ | |
208 | args.minleft = XFS_IN_MAXLEVELS(args.mp) - 1; | |
209 | if ((error = xfs_alloc_vextent(&args))) | |
210 | return error; | |
211 | ||
212 | /* | |
213 | * If stripe alignment is turned on, then try again with cluster | |
214 | * alignment. | |
215 | */ | |
216 | if (isaligned && args.fsbno == NULLFSBLOCK) { | |
217 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
218 | args.agbno = INT_GET(agi->agi_root, ARCH_CONVERT); | |
219 | args.fsbno = XFS_AGB_TO_FSB(args.mp, | |
220 | INT_GET(agi->agi_seqno, ARCH_CONVERT), args.agbno); | |
221 | if (XFS_SB_VERSION_HASALIGN(&args.mp->m_sb) && | |
222 | args.mp->m_sb.sb_inoalignmt >= | |
223 | XFS_B_TO_FSBT(args.mp, XFS_INODE_CLUSTER_SIZE(args.mp))) | |
224 | args.alignment = args.mp->m_sb.sb_inoalignmt; | |
225 | else | |
226 | args.alignment = 1; | |
227 | if ((error = xfs_alloc_vextent(&args))) | |
228 | return error; | |
229 | } | |
230 | ||
231 | if (args.fsbno == NULLFSBLOCK) { | |
232 | *alloc = 0; | |
233 | return 0; | |
234 | } | |
235 | ASSERT(args.len == args.minlen); | |
236 | /* | |
237 | * Convert the results. | |
238 | */ | |
239 | newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0); | |
240 | /* | |
241 | * Loop over the new block(s), filling in the inodes. | |
242 | * For small block sizes, manipulate the inodes in buffers | |
243 | * which are multiples of the blocks size. | |
244 | */ | |
245 | if (args.mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(args.mp)) { | |
246 | blks_per_cluster = 1; | |
247 | nbufs = (int)args.len; | |
248 | ninodes = args.mp->m_sb.sb_inopblock; | |
249 | } else { | |
250 | blks_per_cluster = XFS_INODE_CLUSTER_SIZE(args.mp) / | |
251 | args.mp->m_sb.sb_blocksize; | |
252 | nbufs = (int)args.len / blks_per_cluster; | |
253 | ninodes = blks_per_cluster * args.mp->m_sb.sb_inopblock; | |
254 | } | |
255 | /* | |
256 | * Figure out what version number to use in the inodes we create. | |
257 | * If the superblock version has caught up to the one that supports | |
258 | * the new inode format, then use the new inode version. Otherwise | |
259 | * use the old version so that old kernels will continue to be | |
260 | * able to use the file system. | |
261 | */ | |
262 | if (XFS_SB_VERSION_HASNLINK(&args.mp->m_sb)) | |
263 | version = XFS_DINODE_VERSION_2; | |
264 | else | |
265 | version = XFS_DINODE_VERSION_1; | |
266 | ||
267 | memset(&dic, 0, sizeof(xfs_dinode_core_t)); | |
268 | INT_SET(dic.di_magic, ARCH_CONVERT, XFS_DINODE_MAGIC); | |
269 | INT_SET(dic.di_version, ARCH_CONVERT, version); | |
270 | ||
271 | for (j = 0; j < nbufs; j++) { | |
272 | /* | |
273 | * Get the block. | |
274 | */ | |
275 | d = XFS_AGB_TO_DADDR(args.mp, INT_GET(agi->agi_seqno, ARCH_CONVERT), | |
276 | args.agbno + (j * blks_per_cluster)); | |
277 | fbuf = xfs_trans_get_buf(tp, args.mp->m_ddev_targp, d, | |
278 | args.mp->m_bsize * blks_per_cluster, | |
279 | XFS_BUF_LOCK); | |
280 | ASSERT(fbuf); | |
281 | ASSERT(!XFS_BUF_GETERROR(fbuf)); | |
282 | /* | |
283 | * Loop over the inodes in this buffer. | |
284 | */ | |
285 | ||
286 | for (i = 0; i < ninodes; i++) { | |
287 | free = XFS_MAKE_IPTR(args.mp, fbuf, i); | |
288 | memcpy(&(free->di_core), &dic, sizeof(xfs_dinode_core_t)); | |
289 | INT_SET(free->di_next_unlinked, ARCH_CONVERT, NULLAGINO); | |
290 | xfs_ialloc_log_di(tp, fbuf, i, | |
291 | XFS_DI_CORE_BITS | XFS_DI_NEXT_UNLINKED); | |
292 | } | |
293 | xfs_trans_inode_alloc_buf(tp, fbuf); | |
294 | } | |
295 | INT_MOD(agi->agi_count, ARCH_CONVERT, newlen); | |
296 | INT_MOD(agi->agi_freecount, ARCH_CONVERT, newlen); | |
297 | down_read(&args.mp->m_peraglock); | |
298 | args.mp->m_perag[INT_GET(agi->agi_seqno, ARCH_CONVERT)].pagi_freecount += newlen; | |
299 | up_read(&args.mp->m_peraglock); | |
300 | INT_SET(agi->agi_newino, ARCH_CONVERT, newino); | |
301 | /* | |
302 | * Insert records describing the new inode chunk into the btree. | |
303 | */ | |
304 | cur = xfs_btree_init_cursor(args.mp, tp, agbp, | |
305 | INT_GET(agi->agi_seqno, ARCH_CONVERT), | |
306 | XFS_BTNUM_INO, (xfs_inode_t *)0, 0); | |
307 | for (thisino = newino; | |
308 | thisino < newino + newlen; | |
309 | thisino += XFS_INODES_PER_CHUNK) { | |
310 | if ((error = xfs_inobt_lookup_eq(cur, thisino, | |
311 | XFS_INODES_PER_CHUNK, XFS_INOBT_ALL_FREE, &i))) { | |
312 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
313 | return error; | |
314 | } | |
315 | ASSERT(i == 0); | |
316 | if ((error = xfs_inobt_insert(cur, &i))) { | |
317 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
318 | return error; | |
319 | } | |
320 | ASSERT(i == 1); | |
321 | } | |
322 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
323 | /* | |
324 | * Log allocation group header fields | |
325 | */ | |
326 | xfs_ialloc_log_agi(tp, agbp, | |
327 | XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO); | |
328 | /* | |
329 | * Modify/log superblock values for inode count and inode free count. | |
330 | */ | |
331 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen); | |
332 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen); | |
333 | *alloc = 1; | |
334 | return 0; | |
335 | } | |
336 | ||
337 | STATIC __inline xfs_agnumber_t | |
338 | xfs_ialloc_next_ag( | |
339 | xfs_mount_t *mp) | |
340 | { | |
341 | xfs_agnumber_t agno; | |
342 | ||
343 | spin_lock(&mp->m_agirotor_lock); | |
344 | agno = mp->m_agirotor; | |
345 | if (++mp->m_agirotor == mp->m_maxagi) | |
346 | mp->m_agirotor = 0; | |
347 | spin_unlock(&mp->m_agirotor_lock); | |
348 | ||
349 | return agno; | |
350 | } | |
351 | ||
352 | /* | |
353 | * Select an allocation group to look for a free inode in, based on the parent | |
354 | * inode and then mode. Return the allocation group buffer. | |
355 | */ | |
356 | STATIC xfs_buf_t * /* allocation group buffer */ | |
357 | xfs_ialloc_ag_select( | |
358 | xfs_trans_t *tp, /* transaction pointer */ | |
359 | xfs_ino_t parent, /* parent directory inode number */ | |
360 | mode_t mode, /* bits set to indicate file type */ | |
361 | int okalloc) /* ok to allocate more space */ | |
362 | { | |
363 | xfs_buf_t *agbp; /* allocation group header buffer */ | |
364 | xfs_agnumber_t agcount; /* number of ag's in the filesystem */ | |
365 | xfs_agnumber_t agno; /* current ag number */ | |
366 | int flags; /* alloc buffer locking flags */ | |
367 | xfs_extlen_t ineed; /* blocks needed for inode allocation */ | |
368 | xfs_extlen_t longest = 0; /* longest extent available */ | |
369 | xfs_mount_t *mp; /* mount point structure */ | |
370 | int needspace; /* file mode implies space allocated */ | |
371 | xfs_perag_t *pag; /* per allocation group data */ | |
372 | xfs_agnumber_t pagno; /* parent (starting) ag number */ | |
373 | ||
374 | /* | |
375 | * Files of these types need at least one block if length > 0 | |
376 | * (and they won't fit in the inode, but that's hard to figure out). | |
377 | */ | |
378 | needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode); | |
379 | mp = tp->t_mountp; | |
380 | agcount = mp->m_maxagi; | |
381 | if (S_ISDIR(mode)) | |
382 | pagno = xfs_ialloc_next_ag(mp); | |
383 | else { | |
384 | pagno = XFS_INO_TO_AGNO(mp, parent); | |
385 | if (pagno >= agcount) | |
386 | pagno = 0; | |
387 | } | |
388 | ASSERT(pagno < agcount); | |
389 | /* | |
390 | * Loop through allocation groups, looking for one with a little | |
391 | * free space in it. Note we don't look for free inodes, exactly. | |
392 | * Instead, we include whether there is a need to allocate inodes | |
393 | * to mean that blocks must be allocated for them, | |
394 | * if none are currently free. | |
395 | */ | |
396 | agno = pagno; | |
397 | flags = XFS_ALLOC_FLAG_TRYLOCK; | |
398 | down_read(&mp->m_peraglock); | |
399 | for (;;) { | |
400 | pag = &mp->m_perag[agno]; | |
401 | if (!pag->pagi_init) { | |
402 | if (xfs_ialloc_read_agi(mp, tp, agno, &agbp)) { | |
403 | agbp = NULL; | |
404 | goto nextag; | |
405 | } | |
406 | } else | |
407 | agbp = NULL; | |
408 | ||
409 | if (!pag->pagi_inodeok) { | |
410 | xfs_ialloc_next_ag(mp); | |
411 | goto unlock_nextag; | |
412 | } | |
413 | ||
414 | /* | |
415 | * Is there enough free space for the file plus a block | |
416 | * of inodes (if we need to allocate some)? | |
417 | */ | |
418 | ineed = pag->pagi_freecount ? 0 : XFS_IALLOC_BLOCKS(mp); | |
419 | if (ineed && !pag->pagf_init) { | |
420 | if (agbp == NULL && | |
421 | xfs_ialloc_read_agi(mp, tp, agno, &agbp)) { | |
422 | agbp = NULL; | |
423 | goto nextag; | |
424 | } | |
425 | (void)xfs_alloc_pagf_init(mp, tp, agno, flags); | |
426 | } | |
427 | if (!ineed || pag->pagf_init) { | |
428 | if (ineed && !(longest = pag->pagf_longest)) | |
429 | longest = pag->pagf_flcount > 0; | |
430 | if (!ineed || | |
431 | (pag->pagf_freeblks >= needspace + ineed && | |
432 | longest >= ineed && | |
433 | okalloc)) { | |
434 | if (agbp == NULL && | |
435 | xfs_ialloc_read_agi(mp, tp, agno, &agbp)) { | |
436 | agbp = NULL; | |
437 | goto nextag; | |
438 | } | |
439 | up_read(&mp->m_peraglock); | |
440 | return agbp; | |
441 | } | |
442 | } | |
443 | unlock_nextag: | |
444 | if (agbp) | |
445 | xfs_trans_brelse(tp, agbp); | |
446 | nextag: | |
447 | /* | |
448 | * No point in iterating over the rest, if we're shutting | |
449 | * down. | |
450 | */ | |
451 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
452 | up_read(&mp->m_peraglock); | |
453 | return (xfs_buf_t *)0; | |
454 | } | |
455 | agno++; | |
456 | if (agno >= agcount) | |
457 | agno = 0; | |
458 | if (agno == pagno) { | |
459 | if (flags == 0) { | |
460 | up_read(&mp->m_peraglock); | |
461 | return (xfs_buf_t *)0; | |
462 | } | |
463 | flags = 0; | |
464 | } | |
465 | } | |
466 | } | |
467 | ||
468 | /* | |
469 | * Visible inode allocation functions. | |
470 | */ | |
471 | ||
472 | /* | |
473 | * Allocate an inode on disk. | |
474 | * Mode is used to tell whether the new inode will need space, and whether | |
475 | * it is a directory. | |
476 | * | |
477 | * The arguments IO_agbp and alloc_done are defined to work within | |
478 | * the constraint of one allocation per transaction. | |
479 | * xfs_dialloc() is designed to be called twice if it has to do an | |
480 | * allocation to make more free inodes. On the first call, | |
481 | * IO_agbp should be set to NULL. If an inode is available, | |
482 | * i.e., xfs_dialloc() did not need to do an allocation, an inode | |
483 | * number is returned. In this case, IO_agbp would be set to the | |
484 | * current ag_buf and alloc_done set to false. | |
485 | * If an allocation needed to be done, xfs_dialloc would return | |
486 | * the current ag_buf in IO_agbp and set alloc_done to true. | |
487 | * The caller should then commit the current transaction, allocate a new | |
488 | * transaction, and call xfs_dialloc() again, passing in the previous | |
489 | * value of IO_agbp. IO_agbp should be held across the transactions. | |
490 | * Since the agbp is locked across the two calls, the second call is | |
491 | * guaranteed to have a free inode available. | |
492 | * | |
493 | * Once we successfully pick an inode its number is returned and the | |
494 | * on-disk data structures are updated. The inode itself is not read | |
495 | * in, since doing so would break ordering constraints with xfs_reclaim. | |
496 | */ | |
497 | int | |
498 | xfs_dialloc( | |
499 | xfs_trans_t *tp, /* transaction pointer */ | |
500 | xfs_ino_t parent, /* parent inode (directory) */ | |
501 | mode_t mode, /* mode bits for new inode */ | |
502 | int okalloc, /* ok to allocate more space */ | |
503 | xfs_buf_t **IO_agbp, /* in/out ag header's buffer */ | |
504 | boolean_t *alloc_done, /* true if we needed to replenish | |
505 | inode freelist */ | |
506 | xfs_ino_t *inop) /* inode number allocated */ | |
507 | { | |
508 | xfs_agnumber_t agcount; /* number of allocation groups */ | |
509 | xfs_buf_t *agbp; /* allocation group header's buffer */ | |
510 | xfs_agnumber_t agno; /* allocation group number */ | |
511 | xfs_agi_t *agi; /* allocation group header structure */ | |
512 | xfs_btree_cur_t *cur; /* inode allocation btree cursor */ | |
513 | int error; /* error return value */ | |
514 | int i; /* result code */ | |
515 | int ialloced; /* inode allocation status */ | |
516 | int noroom = 0; /* no space for inode blk allocation */ | |
517 | xfs_ino_t ino; /* fs-relative inode to be returned */ | |
518 | /* REFERENCED */ | |
519 | int j; /* result code */ | |
520 | xfs_mount_t *mp; /* file system mount structure */ | |
521 | int offset; /* index of inode in chunk */ | |
522 | xfs_agino_t pagino; /* parent's a.g. relative inode # */ | |
523 | xfs_agnumber_t pagno; /* parent's allocation group number */ | |
524 | xfs_inobt_rec_t rec; /* inode allocation record */ | |
525 | xfs_agnumber_t tagno; /* testing allocation group number */ | |
526 | xfs_btree_cur_t *tcur; /* temp cursor */ | |
527 | xfs_inobt_rec_t trec; /* temp inode allocation record */ | |
528 | ||
529 | ||
530 | if (*IO_agbp == NULL) { | |
531 | /* | |
532 | * We do not have an agbp, so select an initial allocation | |
533 | * group for inode allocation. | |
534 | */ | |
535 | agbp = xfs_ialloc_ag_select(tp, parent, mode, okalloc); | |
536 | /* | |
537 | * Couldn't find an allocation group satisfying the | |
538 | * criteria, give up. | |
539 | */ | |
540 | if (!agbp) { | |
541 | *inop = NULLFSINO; | |
542 | return 0; | |
543 | } | |
544 | agi = XFS_BUF_TO_AGI(agbp); | |
545 | ASSERT(INT_GET(agi->agi_magicnum, ARCH_CONVERT) == XFS_AGI_MAGIC); | |
546 | } else { | |
547 | /* | |
548 | * Continue where we left off before. In this case, we | |
549 | * know that the allocation group has free inodes. | |
550 | */ | |
551 | agbp = *IO_agbp; | |
552 | agi = XFS_BUF_TO_AGI(agbp); | |
553 | ASSERT(INT_GET(agi->agi_magicnum, ARCH_CONVERT) == XFS_AGI_MAGIC); | |
554 | ASSERT(INT_GET(agi->agi_freecount, ARCH_CONVERT) > 0); | |
555 | } | |
556 | mp = tp->t_mountp; | |
557 | agcount = mp->m_sb.sb_agcount; | |
558 | agno = INT_GET(agi->agi_seqno, ARCH_CONVERT); | |
559 | tagno = agno; | |
560 | pagno = XFS_INO_TO_AGNO(mp, parent); | |
561 | pagino = XFS_INO_TO_AGINO(mp, parent); | |
562 | ||
563 | /* | |
564 | * If we have already hit the ceiling of inode blocks then clear | |
565 | * okalloc so we scan all available agi structures for a free | |
566 | * inode. | |
567 | */ | |
568 | ||
569 | if (mp->m_maxicount && | |
570 | mp->m_sb.sb_icount + XFS_IALLOC_INODES(mp) > mp->m_maxicount) { | |
571 | noroom = 1; | |
572 | okalloc = 0; | |
573 | } | |
574 | ||
575 | /* | |
576 | * Loop until we find an allocation group that either has free inodes | |
577 | * or in which we can allocate some inodes. Iterate through the | |
578 | * allocation groups upward, wrapping at the end. | |
579 | */ | |
580 | *alloc_done = B_FALSE; | |
581 | while (!agi->agi_freecount) { | |
582 | /* | |
583 | * Don't do anything if we're not supposed to allocate | |
584 | * any blocks, just go on to the next ag. | |
585 | */ | |
586 | if (okalloc) { | |
587 | /* | |
588 | * Try to allocate some new inodes in the allocation | |
589 | * group. | |
590 | */ | |
591 | if ((error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced))) { | |
592 | xfs_trans_brelse(tp, agbp); | |
593 | if (error == ENOSPC) { | |
594 | *inop = NULLFSINO; | |
595 | return 0; | |
596 | } else | |
597 | return error; | |
598 | } | |
599 | if (ialloced) { | |
600 | /* | |
601 | * We successfully allocated some inodes, return | |
602 | * the current context to the caller so that it | |
603 | * can commit the current transaction and call | |
604 | * us again where we left off. | |
605 | */ | |
606 | ASSERT(INT_GET(agi->agi_freecount, ARCH_CONVERT) > 0); | |
607 | *alloc_done = B_TRUE; | |
608 | *IO_agbp = agbp; | |
609 | *inop = NULLFSINO; | |
610 | return 0; | |
611 | } | |
612 | } | |
613 | /* | |
614 | * If it failed, give up on this ag. | |
615 | */ | |
616 | xfs_trans_brelse(tp, agbp); | |
617 | /* | |
618 | * Go on to the next ag: get its ag header. | |
619 | */ | |
620 | nextag: | |
621 | if (++tagno == agcount) | |
622 | tagno = 0; | |
623 | if (tagno == agno) { | |
624 | *inop = NULLFSINO; | |
625 | return noroom ? ENOSPC : 0; | |
626 | } | |
627 | down_read(&mp->m_peraglock); | |
628 | if (mp->m_perag[tagno].pagi_inodeok == 0) { | |
629 | up_read(&mp->m_peraglock); | |
630 | goto nextag; | |
631 | } | |
632 | error = xfs_ialloc_read_agi(mp, tp, tagno, &agbp); | |
633 | up_read(&mp->m_peraglock); | |
634 | if (error) | |
635 | goto nextag; | |
636 | agi = XFS_BUF_TO_AGI(agbp); | |
637 | ASSERT(INT_GET(agi->agi_magicnum, ARCH_CONVERT) == XFS_AGI_MAGIC); | |
638 | } | |
639 | /* | |
640 | * Here with an allocation group that has a free inode. | |
641 | * Reset agno since we may have chosen a new ag in the | |
642 | * loop above. | |
643 | */ | |
644 | agno = tagno; | |
645 | *IO_agbp = NULL; | |
646 | cur = xfs_btree_init_cursor(mp, tp, agbp, INT_GET(agi->agi_seqno, ARCH_CONVERT), | |
647 | XFS_BTNUM_INO, (xfs_inode_t *)0, 0); | |
648 | /* | |
649 | * If pagino is 0 (this is the root inode allocation) use newino. | |
650 | * This must work because we've just allocated some. | |
651 | */ | |
652 | if (!pagino) | |
653 | pagino = INT_GET(agi->agi_newino, ARCH_CONVERT); | |
654 | #ifdef DEBUG | |
655 | if (cur->bc_nlevels == 1) { | |
656 | int freecount = 0; | |
657 | ||
658 | if ((error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &i))) | |
659 | goto error0; | |
660 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
661 | do { | |
662 | if ((error = xfs_inobt_get_rec(cur, &rec.ir_startino, | |
663 | &rec.ir_freecount, &rec.ir_free, &i))) | |
664 | goto error0; | |
665 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
666 | freecount += rec.ir_freecount; | |
667 | if ((error = xfs_inobt_increment(cur, 0, &i))) | |
668 | goto error0; | |
669 | } while (i == 1); | |
670 | ||
671 | ASSERT(freecount == INT_GET(agi->agi_freecount, ARCH_CONVERT) || | |
672 | XFS_FORCED_SHUTDOWN(mp)); | |
673 | } | |
674 | #endif | |
675 | /* | |
676 | * If in the same a.g. as the parent, try to get near the parent. | |
677 | */ | |
678 | if (pagno == agno) { | |
679 | if ((error = xfs_inobt_lookup_le(cur, pagino, 0, 0, &i))) | |
680 | goto error0; | |
681 | if (i != 0 && | |
682 | (error = xfs_inobt_get_rec(cur, &rec.ir_startino, | |
683 | &rec.ir_freecount, &rec.ir_free, &j)) == 0 && | |
684 | j == 1 && | |
685 | rec.ir_freecount > 0) { | |
686 | /* | |
687 | * Found a free inode in the same chunk | |
688 | * as parent, done. | |
689 | */ | |
690 | } | |
691 | /* | |
692 | * In the same a.g. as parent, but parent's chunk is full. | |
693 | */ | |
694 | else { | |
695 | int doneleft; /* done, to the left */ | |
696 | int doneright; /* done, to the right */ | |
697 | ||
698 | if (error) | |
699 | goto error0; | |
700 | ASSERT(i == 1); | |
701 | ASSERT(j == 1); | |
702 | /* | |
703 | * Duplicate the cursor, search left & right | |
704 | * simultaneously. | |
705 | */ | |
706 | if ((error = xfs_btree_dup_cursor(cur, &tcur))) | |
707 | goto error0; | |
708 | /* | |
709 | * Search left with tcur, back up 1 record. | |
710 | */ | |
711 | if ((error = xfs_inobt_decrement(tcur, 0, &i))) | |
712 | goto error1; | |
713 | doneleft = !i; | |
714 | if (!doneleft) { | |
715 | if ((error = xfs_inobt_get_rec(tcur, | |
716 | &trec.ir_startino, | |
717 | &trec.ir_freecount, | |
718 | &trec.ir_free, &i))) | |
719 | goto error1; | |
720 | XFS_WANT_CORRUPTED_GOTO(i == 1, error1); | |
721 | } | |
722 | /* | |
723 | * Search right with cur, go forward 1 record. | |
724 | */ | |
725 | if ((error = xfs_inobt_increment(cur, 0, &i))) | |
726 | goto error1; | |
727 | doneright = !i; | |
728 | if (!doneright) { | |
729 | if ((error = xfs_inobt_get_rec(cur, | |
730 | &rec.ir_startino, | |
731 | &rec.ir_freecount, | |
732 | &rec.ir_free, &i))) | |
733 | goto error1; | |
734 | XFS_WANT_CORRUPTED_GOTO(i == 1, error1); | |
735 | } | |
736 | /* | |
737 | * Loop until we find the closest inode chunk | |
738 | * with a free one. | |
739 | */ | |
740 | while (!doneleft || !doneright) { | |
741 | int useleft; /* using left inode | |
742 | chunk this time */ | |
743 | ||
744 | /* | |
745 | * Figure out which block is closer, | |
746 | * if both are valid. | |
747 | */ | |
748 | if (!doneleft && !doneright) | |
749 | useleft = | |
750 | pagino - | |
751 | (trec.ir_startino + | |
752 | XFS_INODES_PER_CHUNK - 1) < | |
753 | rec.ir_startino - pagino; | |
754 | else | |
755 | useleft = !doneleft; | |
756 | /* | |
757 | * If checking the left, does it have | |
758 | * free inodes? | |
759 | */ | |
760 | if (useleft && trec.ir_freecount) { | |
761 | /* | |
762 | * Yes, set it up as the chunk to use. | |
763 | */ | |
764 | rec = trec; | |
765 | xfs_btree_del_cursor(cur, | |
766 | XFS_BTREE_NOERROR); | |
767 | cur = tcur; | |
768 | break; | |
769 | } | |
770 | /* | |
771 | * If checking the right, does it have | |
772 | * free inodes? | |
773 | */ | |
774 | if (!useleft && rec.ir_freecount) { | |
775 | /* | |
776 | * Yes, it's already set up. | |
777 | */ | |
778 | xfs_btree_del_cursor(tcur, | |
779 | XFS_BTREE_NOERROR); | |
780 | break; | |
781 | } | |
782 | /* | |
783 | * If used the left, get another one | |
784 | * further left. | |
785 | */ | |
786 | if (useleft) { | |
787 | if ((error = xfs_inobt_decrement(tcur, 0, | |
788 | &i))) | |
789 | goto error1; | |
790 | doneleft = !i; | |
791 | if (!doneleft) { | |
792 | if ((error = xfs_inobt_get_rec( | |
793 | tcur, | |
794 | &trec.ir_startino, | |
795 | &trec.ir_freecount, | |
796 | &trec.ir_free, &i))) | |
797 | goto error1; | |
798 | XFS_WANT_CORRUPTED_GOTO(i == 1, | |
799 | error1); | |
800 | } | |
801 | } | |
802 | /* | |
803 | * If used the right, get another one | |
804 | * further right. | |
805 | */ | |
806 | else { | |
807 | if ((error = xfs_inobt_increment(cur, 0, | |
808 | &i))) | |
809 | goto error1; | |
810 | doneright = !i; | |
811 | if (!doneright) { | |
812 | if ((error = xfs_inobt_get_rec( | |
813 | cur, | |
814 | &rec.ir_startino, | |
815 | &rec.ir_freecount, | |
816 | &rec.ir_free, &i))) | |
817 | goto error1; | |
818 | XFS_WANT_CORRUPTED_GOTO(i == 1, | |
819 | error1); | |
820 | } | |
821 | } | |
822 | } | |
823 | ASSERT(!doneleft || !doneright); | |
824 | } | |
825 | } | |
826 | /* | |
827 | * In a different a.g. from the parent. | |
828 | * See if the most recently allocated block has any free. | |
829 | */ | |
830 | else if (INT_GET(agi->agi_newino, ARCH_CONVERT) != NULLAGINO) { | |
831 | if ((error = xfs_inobt_lookup_eq(cur, | |
832 | INT_GET(agi->agi_newino, ARCH_CONVERT), 0, 0, &i))) | |
833 | goto error0; | |
834 | if (i == 1 && | |
835 | (error = xfs_inobt_get_rec(cur, &rec.ir_startino, | |
836 | &rec.ir_freecount, &rec.ir_free, &j)) == 0 && | |
837 | j == 1 && | |
838 | rec.ir_freecount > 0) { | |
839 | /* | |
840 | * The last chunk allocated in the group still has | |
841 | * a free inode. | |
842 | */ | |
843 | } | |
844 | /* | |
845 | * None left in the last group, search the whole a.g. | |
846 | */ | |
847 | else { | |
848 | if (error) | |
849 | goto error0; | |
850 | if ((error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &i))) | |
851 | goto error0; | |
852 | ASSERT(i == 1); | |
853 | for (;;) { | |
854 | if ((error = xfs_inobt_get_rec(cur, | |
855 | &rec.ir_startino, | |
856 | &rec.ir_freecount, &rec.ir_free, | |
857 | &i))) | |
858 | goto error0; | |
859 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
860 | if (rec.ir_freecount > 0) | |
861 | break; | |
862 | if ((error = xfs_inobt_increment(cur, 0, &i))) | |
863 | goto error0; | |
864 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
865 | } | |
866 | } | |
867 | } | |
868 | offset = XFS_IALLOC_FIND_FREE(&rec.ir_free); | |
869 | ASSERT(offset >= 0); | |
870 | ASSERT(offset < XFS_INODES_PER_CHUNK); | |
871 | ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) % | |
872 | XFS_INODES_PER_CHUNK) == 0); | |
873 | ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset); | |
874 | XFS_INOBT_CLR_FREE(&rec, offset); | |
875 | rec.ir_freecount--; | |
876 | if ((error = xfs_inobt_update(cur, rec.ir_startino, rec.ir_freecount, | |
877 | rec.ir_free))) | |
878 | goto error0; | |
879 | INT_MOD(agi->agi_freecount, ARCH_CONVERT, -1); | |
880 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); | |
881 | down_read(&mp->m_peraglock); | |
882 | mp->m_perag[tagno].pagi_freecount--; | |
883 | up_read(&mp->m_peraglock); | |
884 | #ifdef DEBUG | |
885 | if (cur->bc_nlevels == 1) { | |
886 | int freecount = 0; | |
887 | ||
888 | if ((error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &i))) | |
889 | goto error0; | |
890 | do { | |
891 | if ((error = xfs_inobt_get_rec(cur, &rec.ir_startino, | |
892 | &rec.ir_freecount, &rec.ir_free, &i))) | |
893 | goto error0; | |
894 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
895 | freecount += rec.ir_freecount; | |
896 | if ((error = xfs_inobt_increment(cur, 0, &i))) | |
897 | goto error0; | |
898 | } while (i == 1); | |
899 | ASSERT(freecount == INT_GET(agi->agi_freecount, ARCH_CONVERT) || | |
900 | XFS_FORCED_SHUTDOWN(mp)); | |
901 | } | |
902 | #endif | |
903 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
904 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1); | |
905 | *inop = ino; | |
906 | return 0; | |
907 | error1: | |
908 | xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR); | |
909 | error0: | |
910 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
911 | return error; | |
912 | } | |
913 | ||
914 | /* | |
915 | * Free disk inode. Carefully avoids touching the incore inode, all | |
916 | * manipulations incore are the caller's responsibility. | |
917 | * The on-disk inode is not changed by this operation, only the | |
918 | * btree (free inode mask) is changed. | |
919 | */ | |
920 | int | |
921 | xfs_difree( | |
922 | xfs_trans_t *tp, /* transaction pointer */ | |
923 | xfs_ino_t inode, /* inode to be freed */ | |
924 | xfs_bmap_free_t *flist, /* extents to free */ | |
925 | int *delete, /* set if inode cluster was deleted */ | |
926 | xfs_ino_t *first_ino) /* first inode in deleted cluster */ | |
927 | { | |
928 | /* REFERENCED */ | |
929 | xfs_agblock_t agbno; /* block number containing inode */ | |
930 | xfs_buf_t *agbp; /* buffer containing allocation group header */ | |
931 | xfs_agino_t agino; /* inode number relative to allocation group */ | |
932 | xfs_agnumber_t agno; /* allocation group number */ | |
933 | xfs_agi_t *agi; /* allocation group header */ | |
934 | xfs_btree_cur_t *cur; /* inode btree cursor */ | |
935 | int error; /* error return value */ | |
936 | int i; /* result code */ | |
937 | int ilen; /* inodes in an inode cluster */ | |
938 | xfs_mount_t *mp; /* mount structure for filesystem */ | |
939 | int off; /* offset of inode in inode chunk */ | |
940 | xfs_inobt_rec_t rec; /* btree record */ | |
941 | ||
942 | mp = tp->t_mountp; | |
943 | ||
944 | /* | |
945 | * Break up inode number into its components. | |
946 | */ | |
947 | agno = XFS_INO_TO_AGNO(mp, inode); | |
948 | if (agno >= mp->m_sb.sb_agcount) { | |
949 | cmn_err(CE_WARN, | |
950 | "xfs_difree: agno >= mp->m_sb.sb_agcount (%d >= %d) on %s. Returning EINVAL.", | |
951 | agno, mp->m_sb.sb_agcount, mp->m_fsname); | |
952 | ASSERT(0); | |
953 | return XFS_ERROR(EINVAL); | |
954 | } | |
955 | agino = XFS_INO_TO_AGINO(mp, inode); | |
956 | if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
957 | cmn_err(CE_WARN, | |
da1650a5 CH |
958 | "xfs_difree: inode != XFS_AGINO_TO_INO() " |
959 | "(%llu != %llu) on %s. Returning EINVAL.", | |
960 | (unsigned long long)inode, | |
961 | (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino), | |
962 | mp->m_fsname); | |
1da177e4 LT |
963 | ASSERT(0); |
964 | return XFS_ERROR(EINVAL); | |
965 | } | |
966 | agbno = XFS_AGINO_TO_AGBNO(mp, agino); | |
967 | if (agbno >= mp->m_sb.sb_agblocks) { | |
968 | cmn_err(CE_WARN, | |
969 | "xfs_difree: agbno >= mp->m_sb.sb_agblocks (%d >= %d) on %s. Returning EINVAL.", | |
970 | agbno, mp->m_sb.sb_agblocks, mp->m_fsname); | |
971 | ASSERT(0); | |
972 | return XFS_ERROR(EINVAL); | |
973 | } | |
974 | /* | |
975 | * Get the allocation group header. | |
976 | */ | |
977 | down_read(&mp->m_peraglock); | |
978 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); | |
979 | up_read(&mp->m_peraglock); | |
980 | if (error) { | |
981 | cmn_err(CE_WARN, | |
982 | "xfs_difree: xfs_ialloc_read_agi() returned an error %d on %s. Returning error.", | |
983 | error, mp->m_fsname); | |
984 | return error; | |
985 | } | |
986 | agi = XFS_BUF_TO_AGI(agbp); | |
987 | ASSERT(INT_GET(agi->agi_magicnum, ARCH_CONVERT) == XFS_AGI_MAGIC); | |
988 | ASSERT(agbno < INT_GET(agi->agi_length, ARCH_CONVERT)); | |
989 | /* | |
990 | * Initialize the cursor. | |
991 | */ | |
992 | cur = xfs_btree_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO, | |
993 | (xfs_inode_t *)0, 0); | |
994 | #ifdef DEBUG | |
995 | if (cur->bc_nlevels == 1) { | |
996 | int freecount = 0; | |
997 | ||
998 | if ((error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &i))) | |
999 | goto error0; | |
1000 | do { | |
1001 | if ((error = xfs_inobt_get_rec(cur, &rec.ir_startino, | |
1002 | &rec.ir_freecount, &rec.ir_free, &i))) | |
1003 | goto error0; | |
1004 | if (i) { | |
1005 | freecount += rec.ir_freecount; | |
1006 | if ((error = xfs_inobt_increment(cur, 0, &i))) | |
1007 | goto error0; | |
1008 | } | |
1009 | } while (i == 1); | |
1010 | ASSERT(freecount == INT_GET(agi->agi_freecount, ARCH_CONVERT) || | |
1011 | XFS_FORCED_SHUTDOWN(mp)); | |
1012 | } | |
1013 | #endif | |
1014 | /* | |
1015 | * Look for the entry describing this inode. | |
1016 | */ | |
1017 | if ((error = xfs_inobt_lookup_le(cur, agino, 0, 0, &i))) { | |
1018 | cmn_err(CE_WARN, | |
1019 | "xfs_difree: xfs_inobt_lookup_le returned() an error %d on %s. Returning error.", | |
1020 | error, mp->m_fsname); | |
1021 | goto error0; | |
1022 | } | |
1023 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1024 | if ((error = xfs_inobt_get_rec(cur, &rec.ir_startino, &rec.ir_freecount, | |
1025 | &rec.ir_free, &i))) { | |
1026 | cmn_err(CE_WARN, | |
1027 | "xfs_difree: xfs_inobt_get_rec() returned an error %d on %s. Returning error.", | |
1028 | error, mp->m_fsname); | |
1029 | goto error0; | |
1030 | } | |
1031 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1032 | /* | |
1033 | * Get the offset in the inode chunk. | |
1034 | */ | |
1035 | off = agino - rec.ir_startino; | |
1036 | ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK); | |
1037 | ASSERT(!XFS_INOBT_IS_FREE(&rec, off)); | |
1038 | /* | |
1039 | * Mark the inode free & increment the count. | |
1040 | */ | |
1041 | XFS_INOBT_SET_FREE(&rec, off); | |
1042 | rec.ir_freecount++; | |
1043 | ||
1044 | /* | |
1045 | * When an inode cluster is free, it becomes elgible for removal | |
1046 | */ | |
1047 | if ((mp->m_flags & XFS_MOUNT_IDELETE) && | |
1048 | (rec.ir_freecount == XFS_IALLOC_INODES(mp))) { | |
1049 | ||
1050 | *delete = 1; | |
1051 | *first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino); | |
1052 | ||
1053 | /* | |
1054 | * Remove the inode cluster from the AGI B+Tree, adjust the | |
1055 | * AGI and Superblock inode counts, and mark the disk space | |
1056 | * to be freed when the transaction is committed. | |
1057 | */ | |
1058 | ilen = XFS_IALLOC_INODES(mp); | |
1059 | INT_MOD(agi->agi_count, ARCH_CONVERT, -ilen); | |
1060 | INT_MOD(agi->agi_freecount, ARCH_CONVERT, -(ilen - 1)); | |
1061 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT); | |
1062 | down_read(&mp->m_peraglock); | |
1063 | mp->m_perag[agno].pagi_freecount -= ilen - 1; | |
1064 | up_read(&mp->m_peraglock); | |
1065 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen); | |
1066 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1)); | |
1067 | ||
1068 | if ((error = xfs_inobt_delete(cur, &i))) { | |
1069 | cmn_err(CE_WARN, "xfs_difree: xfs_inobt_delete returned an error %d on %s.\n", | |
1070 | error, mp->m_fsname); | |
1071 | goto error0; | |
1072 | } | |
1073 | ||
1074 | xfs_bmap_add_free(XFS_AGB_TO_FSB(mp, | |
1075 | agno, XFS_INO_TO_AGBNO(mp,rec.ir_startino)), | |
1076 | XFS_IALLOC_BLOCKS(mp), flist, mp); | |
1077 | } else { | |
1078 | *delete = 0; | |
1079 | ||
1080 | if ((error = xfs_inobt_update(cur, rec.ir_startino, rec.ir_freecount, rec.ir_free))) { | |
1081 | cmn_err(CE_WARN, | |
1082 | "xfs_difree: xfs_inobt_update() returned an error %d on %s. Returning error.", | |
1083 | error, mp->m_fsname); | |
1084 | goto error0; | |
1085 | } | |
1086 | /* | |
1087 | * Change the inode free counts and log the ag/sb changes. | |
1088 | */ | |
1089 | INT_MOD(agi->agi_freecount, ARCH_CONVERT, 1); | |
1090 | xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); | |
1091 | down_read(&mp->m_peraglock); | |
1092 | mp->m_perag[agno].pagi_freecount++; | |
1093 | up_read(&mp->m_peraglock); | |
1094 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1); | |
1095 | } | |
1096 | ||
1097 | #ifdef DEBUG | |
1098 | if (cur->bc_nlevels == 1) { | |
1099 | int freecount = 0; | |
1100 | ||
1101 | if ((error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &i))) | |
1102 | goto error0; | |
1103 | do { | |
1104 | if ((error = xfs_inobt_get_rec(cur, | |
1105 | &rec.ir_startino, | |
1106 | &rec.ir_freecount, | |
1107 | &rec.ir_free, &i))) | |
1108 | goto error0; | |
1109 | if (i) { | |
1110 | freecount += rec.ir_freecount; | |
1111 | if ((error = xfs_inobt_increment(cur, 0, &i))) | |
1112 | goto error0; | |
1113 | } | |
1114 | } while (i == 1); | |
1115 | ASSERT(freecount == INT_GET(agi->agi_freecount, ARCH_CONVERT) || | |
1116 | XFS_FORCED_SHUTDOWN(mp)); | |
1117 | } | |
1118 | #endif | |
1119 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
1120 | return 0; | |
1121 | ||
1122 | error0: | |
1123 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
1124 | return error; | |
1125 | } | |
1126 | ||
1127 | /* | |
1128 | * Return the location of the inode in bno/off, for mapping it into a buffer. | |
1129 | */ | |
1130 | /*ARGSUSED*/ | |
1131 | int | |
1132 | xfs_dilocate( | |
1133 | xfs_mount_t *mp, /* file system mount structure */ | |
1134 | xfs_trans_t *tp, /* transaction pointer */ | |
1135 | xfs_ino_t ino, /* inode to locate */ | |
1136 | xfs_fsblock_t *bno, /* output: block containing inode */ | |
1137 | int *len, /* output: num blocks in inode cluster */ | |
1138 | int *off, /* output: index in block of inode */ | |
1139 | uint flags) /* flags concerning inode lookup */ | |
1140 | { | |
1141 | xfs_agblock_t agbno; /* block number of inode in the alloc group */ | |
1142 | xfs_buf_t *agbp; /* agi buffer */ | |
1143 | xfs_agino_t agino; /* inode number within alloc group */ | |
1144 | xfs_agnumber_t agno; /* allocation group number */ | |
1145 | int blks_per_cluster; /* num blocks per inode cluster */ | |
1146 | xfs_agblock_t chunk_agbno; /* first block in inode chunk */ | |
1147 | xfs_agino_t chunk_agino; /* first agino in inode chunk */ | |
1148 | __int32_t chunk_cnt; /* count of free inodes in chunk */ | |
1149 | xfs_inofree_t chunk_free; /* mask of free inodes in chunk */ | |
1150 | xfs_agblock_t cluster_agbno; /* first block in inode cluster */ | |
1151 | xfs_btree_cur_t *cur; /* inode btree cursor */ | |
1152 | int error; /* error code */ | |
1153 | int i; /* temp state */ | |
1154 | int offset; /* index of inode in its buffer */ | |
1155 | int offset_agbno; /* blks from chunk start to inode */ | |
1156 | ||
1157 | ASSERT(ino != NULLFSINO); | |
1158 | /* | |
1159 | * Split up the inode number into its parts. | |
1160 | */ | |
1161 | agno = XFS_INO_TO_AGNO(mp, ino); | |
1162 | agino = XFS_INO_TO_AGINO(mp, ino); | |
1163 | agbno = XFS_AGINO_TO_AGBNO(mp, agino); | |
1164 | if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks || | |
1165 | ino != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
1166 | #ifdef DEBUG | |
1167 | if (agno >= mp->m_sb.sb_agcount) { | |
1168 | xfs_fs_cmn_err(CE_ALERT, mp, | |
1169 | "xfs_dilocate: agno (%d) >= " | |
1170 | "mp->m_sb.sb_agcount (%d)", | |
1171 | agno, mp->m_sb.sb_agcount); | |
1172 | } | |
1173 | if (agbno >= mp->m_sb.sb_agblocks) { | |
1174 | xfs_fs_cmn_err(CE_ALERT, mp, | |
1175 | "xfs_dilocate: agbno (0x%llx) >= " | |
1176 | "mp->m_sb.sb_agblocks (0x%lx)", | |
1177 | (unsigned long long) agbno, | |
1178 | (unsigned long) mp->m_sb.sb_agblocks); | |
1179 | } | |
1180 | if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) { | |
1181 | xfs_fs_cmn_err(CE_ALERT, mp, | |
1182 | "xfs_dilocate: ino (0x%llx) != " | |
1183 | "XFS_AGINO_TO_INO(mp, agno, agino) " | |
1184 | "(0x%llx)", | |
1185 | ino, XFS_AGINO_TO_INO(mp, agno, agino)); | |
1186 | } | |
1187 | #endif /* DEBUG */ | |
1188 | return XFS_ERROR(EINVAL); | |
1189 | } | |
1190 | if ((mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) || | |
1191 | !(flags & XFS_IMAP_LOOKUP)) { | |
1192 | offset = XFS_INO_TO_OFFSET(mp, ino); | |
1193 | ASSERT(offset < mp->m_sb.sb_inopblock); | |
1194 | *bno = XFS_AGB_TO_FSB(mp, agno, agbno); | |
1195 | *off = offset; | |
1196 | *len = 1; | |
1197 | return 0; | |
1198 | } | |
1199 | blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_blocklog; | |
1200 | if (*bno != NULLFSBLOCK) { | |
1201 | offset = XFS_INO_TO_OFFSET(mp, ino); | |
1202 | ASSERT(offset < mp->m_sb.sb_inopblock); | |
1203 | cluster_agbno = XFS_FSB_TO_AGBNO(mp, *bno); | |
1204 | *off = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) + | |
1205 | offset; | |
1206 | *len = blks_per_cluster; | |
1207 | return 0; | |
1208 | } | |
1209 | if (mp->m_inoalign_mask) { | |
1210 | offset_agbno = agbno & mp->m_inoalign_mask; | |
1211 | chunk_agbno = agbno - offset_agbno; | |
1212 | } else { | |
1213 | down_read(&mp->m_peraglock); | |
1214 | error = xfs_ialloc_read_agi(mp, tp, agno, &agbp); | |
1215 | up_read(&mp->m_peraglock); | |
1216 | if (error) { | |
1217 | #ifdef DEBUG | |
1218 | xfs_fs_cmn_err(CE_ALERT, mp, "xfs_dilocate: " | |
1219 | "xfs_ialloc_read_agi() returned " | |
1220 | "error %d, agno %d", | |
1221 | error, agno); | |
1222 | #endif /* DEBUG */ | |
1223 | return error; | |
1224 | } | |
1225 | cur = xfs_btree_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO, | |
1226 | (xfs_inode_t *)0, 0); | |
1227 | if ((error = xfs_inobt_lookup_le(cur, agino, 0, 0, &i))) { | |
1228 | #ifdef DEBUG | |
1229 | xfs_fs_cmn_err(CE_ALERT, mp, "xfs_dilocate: " | |
1230 | "xfs_inobt_lookup_le() failed"); | |
1231 | #endif /* DEBUG */ | |
1232 | goto error0; | |
1233 | } | |
1234 | if ((error = xfs_inobt_get_rec(cur, &chunk_agino, &chunk_cnt, | |
1235 | &chunk_free, &i))) { | |
1236 | #ifdef DEBUG | |
1237 | xfs_fs_cmn_err(CE_ALERT, mp, "xfs_dilocate: " | |
1238 | "xfs_inobt_get_rec() failed"); | |
1239 | #endif /* DEBUG */ | |
1240 | goto error0; | |
1241 | } | |
1242 | if (i == 0) { | |
1243 | #ifdef DEBUG | |
1244 | xfs_fs_cmn_err(CE_ALERT, mp, "xfs_dilocate: " | |
1245 | "xfs_inobt_get_rec() failed"); | |
1246 | #endif /* DEBUG */ | |
1247 | error = XFS_ERROR(EINVAL); | |
1248 | } | |
1249 | xfs_trans_brelse(tp, agbp); | |
1250 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
1251 | if (error) | |
1252 | return error; | |
1253 | chunk_agbno = XFS_AGINO_TO_AGBNO(mp, chunk_agino); | |
1254 | offset_agbno = agbno - chunk_agbno; | |
1255 | } | |
1256 | ASSERT(agbno >= chunk_agbno); | |
1257 | cluster_agbno = chunk_agbno + | |
1258 | ((offset_agbno / blks_per_cluster) * blks_per_cluster); | |
1259 | offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) + | |
1260 | XFS_INO_TO_OFFSET(mp, ino); | |
1261 | *bno = XFS_AGB_TO_FSB(mp, agno, cluster_agbno); | |
1262 | *off = offset; | |
1263 | *len = blks_per_cluster; | |
1264 | return 0; | |
1265 | error0: | |
1266 | xfs_trans_brelse(tp, agbp); | |
1267 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
1268 | return error; | |
1269 | } | |
1270 | ||
1271 | /* | |
1272 | * Compute and fill in value of m_in_maxlevels. | |
1273 | */ | |
1274 | void | |
1275 | xfs_ialloc_compute_maxlevels( | |
1276 | xfs_mount_t *mp) /* file system mount structure */ | |
1277 | { | |
1278 | int level; | |
1279 | uint maxblocks; | |
1280 | uint maxleafents; | |
1281 | int minleafrecs; | |
1282 | int minnoderecs; | |
1283 | ||
1284 | maxleafents = (1LL << XFS_INO_AGINO_BITS(mp)) >> | |
1285 | XFS_INODES_PER_CHUNK_LOG; | |
1286 | minleafrecs = mp->m_alloc_mnr[0]; | |
1287 | minnoderecs = mp->m_alloc_mnr[1]; | |
1288 | maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs; | |
1289 | for (level = 1; maxblocks > 1; level++) | |
1290 | maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs; | |
1291 | mp->m_in_maxlevels = level; | |
1292 | } | |
1293 | ||
1294 | /* | |
1295 | * Log specified fields for the ag hdr (inode section) | |
1296 | */ | |
1297 | void | |
1298 | xfs_ialloc_log_agi( | |
1299 | xfs_trans_t *tp, /* transaction pointer */ | |
1300 | xfs_buf_t *bp, /* allocation group header buffer */ | |
1301 | int fields) /* bitmask of fields to log */ | |
1302 | { | |
1303 | int first; /* first byte number */ | |
1304 | int last; /* last byte number */ | |
1305 | static const short offsets[] = { /* field starting offsets */ | |
1306 | /* keep in sync with bit definitions */ | |
1307 | offsetof(xfs_agi_t, agi_magicnum), | |
1308 | offsetof(xfs_agi_t, agi_versionnum), | |
1309 | offsetof(xfs_agi_t, agi_seqno), | |
1310 | offsetof(xfs_agi_t, agi_length), | |
1311 | offsetof(xfs_agi_t, agi_count), | |
1312 | offsetof(xfs_agi_t, agi_root), | |
1313 | offsetof(xfs_agi_t, agi_level), | |
1314 | offsetof(xfs_agi_t, agi_freecount), | |
1315 | offsetof(xfs_agi_t, agi_newino), | |
1316 | offsetof(xfs_agi_t, agi_dirino), | |
1317 | offsetof(xfs_agi_t, agi_unlinked), | |
1318 | sizeof(xfs_agi_t) | |
1319 | }; | |
1320 | #ifdef DEBUG | |
1321 | xfs_agi_t *agi; /* allocation group header */ | |
1322 | ||
1323 | agi = XFS_BUF_TO_AGI(bp); | |
1324 | ASSERT(INT_GET(agi->agi_magicnum, ARCH_CONVERT) == XFS_AGI_MAGIC); | |
1325 | #endif | |
1326 | /* | |
1327 | * Compute byte offsets for the first and last fields. | |
1328 | */ | |
1329 | xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS, &first, &last); | |
1330 | /* | |
1331 | * Log the allocation group inode header buffer. | |
1332 | */ | |
1333 | xfs_trans_log_buf(tp, bp, first, last); | |
1334 | } | |
1335 | ||
1336 | /* | |
1337 | * Read in the allocation group header (inode allocation section) | |
1338 | */ | |
1339 | int | |
1340 | xfs_ialloc_read_agi( | |
1341 | xfs_mount_t *mp, /* file system mount structure */ | |
1342 | xfs_trans_t *tp, /* transaction pointer */ | |
1343 | xfs_agnumber_t agno, /* allocation group number */ | |
1344 | xfs_buf_t **bpp) /* allocation group hdr buf */ | |
1345 | { | |
1346 | xfs_agi_t *agi; /* allocation group header */ | |
1347 | int agi_ok; /* agi is consistent */ | |
1348 | xfs_buf_t *bp; /* allocation group hdr buf */ | |
1349 | xfs_perag_t *pag; /* per allocation group data */ | |
1350 | int error; | |
1351 | ||
1352 | ASSERT(agno != NULLAGNUMBER); | |
1353 | error = xfs_trans_read_buf( | |
1354 | mp, tp, mp->m_ddev_targp, | |
1355 | XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)), | |
1356 | XFS_FSS_TO_BB(mp, 1), 0, &bp); | |
1357 | if (error) | |
1358 | return error; | |
1359 | ASSERT(bp && !XFS_BUF_GETERROR(bp)); | |
1360 | ||
1361 | /* | |
1362 | * Validate the magic number of the agi block. | |
1363 | */ | |
1364 | agi = XFS_BUF_TO_AGI(bp); | |
1365 | agi_ok = | |
1366 | INT_GET(agi->agi_magicnum, ARCH_CONVERT) == XFS_AGI_MAGIC && | |
1367 | XFS_AGI_GOOD_VERSION( | |
1368 | INT_GET(agi->agi_versionnum, ARCH_CONVERT)); | |
1369 | if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IALLOC_READ_AGI, | |
1370 | XFS_RANDOM_IALLOC_READ_AGI))) { | |
1371 | XFS_CORRUPTION_ERROR("xfs_ialloc_read_agi", XFS_ERRLEVEL_LOW, | |
1372 | mp, agi); | |
1373 | xfs_trans_brelse(tp, bp); | |
1374 | return XFS_ERROR(EFSCORRUPTED); | |
1375 | } | |
1376 | pag = &mp->m_perag[agno]; | |
1377 | if (!pag->pagi_init) { | |
1378 | pag->pagi_freecount = INT_GET(agi->agi_freecount, ARCH_CONVERT); | |
1379 | pag->pagi_init = 1; | |
1380 | } else { | |
1381 | /* | |
1382 | * It's possible for these to be out of sync if | |
1383 | * we are in the middle of a forced shutdown. | |
1384 | */ | |
1385 | ASSERT(pag->pagi_freecount == | |
1386 | INT_GET(agi->agi_freecount, ARCH_CONVERT) | |
1387 | || XFS_FORCED_SHUTDOWN(mp)); | |
1388 | } | |
1389 | ||
1390 | #ifdef DEBUG | |
1391 | { | |
1392 | int i; | |
1393 | ||
1394 | for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++) | |
1395 | ASSERT(agi->agi_unlinked[i]); | |
1396 | } | |
1397 | #endif | |
1398 | ||
1399 | XFS_BUF_SET_VTYPE_REF(bp, B_FS_AGI, XFS_AGI_REF); | |
1400 | *bpp = bp; | |
1401 | return 0; | |
1402 | } |