Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_fs.h" |
1da177e4 | 20 | #include "xfs_types.h" |
a844f451 | 21 | #include "xfs_bit.h" |
1da177e4 | 22 | #include "xfs_log.h" |
a844f451 | 23 | #include "xfs_inum.h" |
1da177e4 LT |
24 | #include "xfs_trans.h" |
25 | #include "xfs_sb.h" | |
26 | #include "xfs_ag.h" | |
1da177e4 LT |
27 | #include "xfs_dir2.h" |
28 | #include "xfs_dmapi.h" | |
29 | #include "xfs_mount.h" | |
1da177e4 | 30 | #include "xfs_bmap_btree.h" |
a844f451 | 31 | #include "xfs_alloc_btree.h" |
1da177e4 | 32 | #include "xfs_ialloc_btree.h" |
1da177e4 | 33 | #include "xfs_dir2_sf.h" |
a844f451 | 34 | #include "xfs_attr_sf.h" |
1da177e4 LT |
35 | #include "xfs_dinode.h" |
36 | #include "xfs_inode.h" | |
a844f451 NS |
37 | #include "xfs_btree.h" |
38 | #include "xfs_ialloc.h" | |
1da177e4 LT |
39 | #include "xfs_quota.h" |
40 | #include "xfs_utils.h" | |
1da177e4 LT |
41 | |
42 | /* | |
43 | * Initialize the inode hash table for the newly mounted file system. | |
44 | * Choose an initial table size based on user specified value, else | |
45 | * use a simple algorithm using the maximum number of inodes as an | |
46 | * indicator for table size, and clamp it between one and some large | |
47 | * number of pages. | |
48 | */ | |
49 | void | |
50 | xfs_ihash_init(xfs_mount_t *mp) | |
51 | { | |
52 | __uint64_t icount; | |
77e4635a | 53 | uint i; |
1da177e4 LT |
54 | |
55 | if (!mp->m_ihsize) { | |
56 | icount = mp->m_maxicount ? mp->m_maxicount : | |
57 | (mp->m_sb.sb_dblocks << mp->m_sb.sb_inopblog); | |
58 | mp->m_ihsize = 1 << max_t(uint, 8, | |
59 | (xfs_highbit64(icount) + 1) / 2); | |
60 | mp->m_ihsize = min_t(uint, mp->m_ihsize, | |
61 | (64 * NBPP) / sizeof(xfs_ihash_t)); | |
62 | } | |
63 | ||
77e4635a NS |
64 | mp->m_ihash = kmem_zalloc_greedy(&mp->m_ihsize, |
65 | NBPC * sizeof(xfs_ihash_t), | |
66 | mp->m_ihsize * sizeof(xfs_ihash_t), | |
67 | KM_SLEEP | KM_MAYFAIL | KM_LARGE); | |
68 | mp->m_ihsize /= sizeof(xfs_ihash_t); | |
69 | for (i = 0; i < mp->m_ihsize; i++) | |
1da177e4 | 70 | rwlock_init(&(mp->m_ihash[i].ih_lock)); |
1da177e4 LT |
71 | } |
72 | ||
73 | /* | |
74 | * Free up structures allocated by xfs_ihash_init, at unmount time. | |
75 | */ | |
76 | void | |
77 | xfs_ihash_free(xfs_mount_t *mp) | |
78 | { | |
77e4635a | 79 | kmem_free(mp->m_ihash, mp->m_ihsize * sizeof(xfs_ihash_t)); |
1da177e4 LT |
80 | mp->m_ihash = NULL; |
81 | } | |
82 | ||
83 | /* | |
84 | * Initialize the inode cluster hash table for the newly mounted file system. | |
85 | * Its size is derived from the ihash table size. | |
86 | */ | |
87 | void | |
88 | xfs_chash_init(xfs_mount_t *mp) | |
89 | { | |
90 | uint i; | |
91 | ||
92 | mp->m_chsize = max_t(uint, 1, mp->m_ihsize / | |
93 | (XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)); | |
94 | mp->m_chsize = min_t(uint, mp->m_chsize, mp->m_ihsize); | |
95 | mp->m_chash = (xfs_chash_t *)kmem_zalloc(mp->m_chsize | |
96 | * sizeof(xfs_chash_t), | |
efb8ad7e | 97 | KM_SLEEP | KM_LARGE); |
1da177e4 LT |
98 | for (i = 0; i < mp->m_chsize; i++) { |
99 | spinlock_init(&mp->m_chash[i].ch_lock,"xfshash"); | |
100 | } | |
101 | } | |
102 | ||
103 | /* | |
104 | * Free up structures allocated by xfs_chash_init, at unmount time. | |
105 | */ | |
106 | void | |
107 | xfs_chash_free(xfs_mount_t *mp) | |
108 | { | |
109 | int i; | |
110 | ||
111 | for (i = 0; i < mp->m_chsize; i++) { | |
112 | spinlock_destroy(&mp->m_chash[i].ch_lock); | |
113 | } | |
114 | ||
115 | kmem_free(mp->m_chash, mp->m_chsize*sizeof(xfs_chash_t)); | |
116 | mp->m_chash = NULL; | |
117 | } | |
118 | ||
71bce256 NS |
119 | /* |
120 | * Try to move an inode to the front of its hash list if possible | |
121 | * (and if its not there already). Called right after obtaining | |
122 | * the list version number and then dropping the read_lock on the | |
123 | * hash list in question (which is done right after looking up the | |
124 | * inode in question...). | |
125 | */ | |
126 | STATIC void | |
127 | xfs_ihash_promote( | |
128 | xfs_ihash_t *ih, | |
129 | xfs_inode_t *ip, | |
130 | ulong version) | |
131 | { | |
132 | xfs_inode_t *iq; | |
133 | ||
134 | if ((ip->i_prevp != &ih->ih_next) && write_trylock(&ih->ih_lock)) { | |
135 | if (likely(version == ih->ih_version)) { | |
136 | /* remove from list */ | |
137 | if ((iq = ip->i_next)) { | |
138 | iq->i_prevp = ip->i_prevp; | |
139 | } | |
140 | *ip->i_prevp = iq; | |
141 | ||
142 | /* insert at list head */ | |
143 | iq = ih->ih_next; | |
144 | iq->i_prevp = &ip->i_next; | |
145 | ip->i_next = iq; | |
146 | ip->i_prevp = &ih->ih_next; | |
147 | ih->ih_next = ip; | |
148 | } | |
149 | write_unlock(&ih->ih_lock); | |
150 | } | |
151 | } | |
152 | ||
1da177e4 LT |
153 | /* |
154 | * Look up an inode by number in the given file system. | |
155 | * The inode is looked up in the hash table for the file system | |
156 | * represented by the mount point parameter mp. Each bucket of | |
157 | * the hash table is guarded by an individual semaphore. | |
158 | * | |
159 | * If the inode is found in the hash table, its corresponding vnode | |
160 | * is obtained with a call to vn_get(). This call takes care of | |
161 | * coordination with the reclamation of the inode and vnode. Note | |
162 | * that the vmap structure is filled in while holding the hash lock. | |
163 | * This gives us the state of the inode/vnode when we found it and | |
164 | * is used for coordination in vn_get(). | |
165 | * | |
166 | * If it is not in core, read it in from the file system's device and | |
167 | * add the inode into the hash table. | |
168 | * | |
169 | * The inode is locked according to the value of the lock_flags parameter. | |
170 | * This flag parameter indicates how and if the inode's IO lock and inode lock | |
171 | * should be taken. | |
172 | * | |
173 | * mp -- the mount point structure for the current file system. It points | |
174 | * to the inode hash table. | |
175 | * tp -- a pointer to the current transaction if there is one. This is | |
176 | * simply passed through to the xfs_iread() call. | |
177 | * ino -- the number of the inode desired. This is the unique identifier | |
178 | * within the file system for the inode being requested. | |
179 | * lock_flags -- flags indicating how to lock the inode. See the comment | |
180 | * for xfs_ilock() for a list of valid values. | |
181 | * bno -- the block number starting the buffer containing the inode, | |
182 | * if known (as by bulkstat), else 0. | |
183 | */ | |
184 | STATIC int | |
185 | xfs_iget_core( | |
67fcaa73 | 186 | bhv_vnode_t *vp, |
1da177e4 LT |
187 | xfs_mount_t *mp, |
188 | xfs_trans_t *tp, | |
189 | xfs_ino_t ino, | |
190 | uint flags, | |
191 | uint lock_flags, | |
192 | xfs_inode_t **ipp, | |
193 | xfs_daddr_t bno) | |
194 | { | |
195 | xfs_ihash_t *ih; | |
196 | xfs_inode_t *ip; | |
197 | xfs_inode_t *iq; | |
67fcaa73 | 198 | bhv_vnode_t *inode_vp; |
1da177e4 LT |
199 | ulong version; |
200 | int error; | |
201 | /* REFERENCED */ | |
202 | xfs_chash_t *ch; | |
203 | xfs_chashlist_t *chl, *chlnew; | |
204 | SPLDECL(s); | |
205 | ||
206 | ||
207 | ih = XFS_IHASH(mp, ino); | |
208 | ||
209 | again: | |
210 | read_lock(&ih->ih_lock); | |
211 | ||
212 | for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) { | |
213 | if (ip->i_ino == ino) { | |
214 | /* | |
215 | * If INEW is set this inode is being set up | |
216 | * we need to pause and try again. | |
217 | */ | |
218 | if (ip->i_flags & XFS_INEW) { | |
219 | read_unlock(&ih->ih_lock); | |
220 | delay(1); | |
221 | XFS_STATS_INC(xs_ig_frecycle); | |
222 | ||
223 | goto again; | |
224 | } | |
225 | ||
226 | inode_vp = XFS_ITOV_NULL(ip); | |
227 | if (inode_vp == NULL) { | |
228 | /* | |
229 | * If IRECLAIM is set this inode is | |
230 | * on its way out of the system, | |
231 | * we need to pause and try again. | |
232 | */ | |
233 | if (ip->i_flags & XFS_IRECLAIM) { | |
234 | read_unlock(&ih->ih_lock); | |
235 | delay(1); | |
236 | XFS_STATS_INC(xs_ig_frecycle); | |
237 | ||
238 | goto again; | |
239 | } | |
240 | ||
241 | vn_trace_exit(vp, "xfs_iget.alloc", | |
242 | (inst_t *)__return_address); | |
243 | ||
244 | XFS_STATS_INC(xs_ig_found); | |
245 | ||
246 | ip->i_flags &= ~XFS_IRECLAIMABLE; | |
71bce256 | 247 | version = ih->ih_version; |
1da177e4 | 248 | read_unlock(&ih->ih_lock); |
71bce256 | 249 | xfs_ihash_promote(ih, ip, version); |
1da177e4 LT |
250 | |
251 | XFS_MOUNT_ILOCK(mp); | |
252 | list_del_init(&ip->i_reclaim); | |
253 | XFS_MOUNT_IUNLOCK(mp); | |
254 | ||
255 | goto finish_inode; | |
256 | ||
257 | } else if (vp != inode_vp) { | |
ec86dc02 | 258 | struct inode *inode = vn_to_inode(inode_vp); |
1da177e4 LT |
259 | |
260 | /* The inode is being torn down, pause and | |
261 | * try again. | |
262 | */ | |
263 | if (inode->i_state & (I_FREEING | I_CLEAR)) { | |
264 | read_unlock(&ih->ih_lock); | |
265 | delay(1); | |
266 | XFS_STATS_INC(xs_ig_frecycle); | |
267 | ||
268 | goto again; | |
269 | } | |
270 | /* Chances are the other vnode (the one in the inode) is being torn | |
271 | * down right now, and we landed on top of it. Question is, what do | |
272 | * we do? Unhook the old inode and hook up the new one? | |
273 | */ | |
274 | cmn_err(CE_PANIC, | |
275 | "xfs_iget_core: ambiguous vns: vp/0x%p, invp/0x%p", | |
276 | inode_vp, vp); | |
277 | } | |
278 | ||
71bce256 NS |
279 | /* |
280 | * Inode cache hit: if ip is not at the front of | |
281 | * its hash chain, move it there now. | |
282 | * Do this with the lock held for update, but | |
283 | * do statistics after releasing the lock. | |
284 | */ | |
285 | version = ih->ih_version; | |
1da177e4 | 286 | read_unlock(&ih->ih_lock); |
71bce256 | 287 | xfs_ihash_promote(ih, ip, version); |
1da177e4 LT |
288 | XFS_STATS_INC(xs_ig_found); |
289 | ||
290 | finish_inode: | |
291 | if (ip->i_d.di_mode == 0) { | |
745b1f47 | 292 | if (!(flags & XFS_IGET_CREATE)) |
1da177e4 LT |
293 | return ENOENT; |
294 | xfs_iocore_inode_reinit(ip); | |
295 | } | |
745b1f47 | 296 | |
1da177e4 LT |
297 | if (lock_flags != 0) |
298 | xfs_ilock(ip, lock_flags); | |
299 | ||
300 | ip->i_flags &= ~XFS_ISTALE; | |
301 | ||
302 | vn_trace_exit(vp, "xfs_iget.found", | |
303 | (inst_t *)__return_address); | |
304 | goto return_ip; | |
305 | } | |
306 | } | |
307 | ||
308 | /* | |
309 | * Inode cache miss: save the hash chain version stamp and unlock | |
310 | * the chain, so we don't deadlock in vn_alloc. | |
311 | */ | |
312 | XFS_STATS_INC(xs_ig_missed); | |
313 | ||
314 | version = ih->ih_version; | |
315 | ||
316 | read_unlock(&ih->ih_lock); | |
317 | ||
318 | /* | |
319 | * Read the disk inode attributes into a new inode structure and get | |
320 | * a new vnode for it. This should also initialize i_ino and i_mount. | |
321 | */ | |
745b1f47 NS |
322 | error = xfs_iread(mp, tp, ino, &ip, bno, |
323 | (flags & XFS_IGET_BULKSTAT) ? XFS_IMAP_BULKSTAT : 0); | |
324 | if (error) | |
1da177e4 | 325 | return error; |
1da177e4 LT |
326 | |
327 | vn_trace_exit(vp, "xfs_iget.alloc", (inst_t *)__return_address); | |
328 | ||
329 | xfs_inode_lock_init(ip, vp); | |
330 | xfs_iocore_inode_init(ip); | |
331 | ||
745b1f47 | 332 | if (lock_flags) |
1da177e4 | 333 | xfs_ilock(ip, lock_flags); |
745b1f47 NS |
334 | |
335 | if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) { | |
1da177e4 LT |
336 | xfs_idestroy(ip); |
337 | return ENOENT; | |
338 | } | |
339 | ||
340 | /* | |
341 | * Put ip on its hash chain, unless someone else hashed a duplicate | |
342 | * after we released the hash lock. | |
343 | */ | |
344 | write_lock(&ih->ih_lock); | |
345 | ||
346 | if (ih->ih_version != version) { | |
347 | for (iq = ih->ih_next; iq != NULL; iq = iq->i_next) { | |
348 | if (iq->i_ino == ino) { | |
349 | write_unlock(&ih->ih_lock); | |
350 | xfs_idestroy(ip); | |
351 | ||
352 | XFS_STATS_INC(xs_ig_dup); | |
353 | goto again; | |
354 | } | |
355 | } | |
356 | } | |
357 | ||
358 | /* | |
359 | * These values _must_ be set before releasing ihlock! | |
360 | */ | |
361 | ip->i_hash = ih; | |
362 | if ((iq = ih->ih_next)) { | |
363 | iq->i_prevp = &ip->i_next; | |
364 | } | |
365 | ip->i_next = iq; | |
366 | ip->i_prevp = &ih->ih_next; | |
367 | ih->ih_next = ip; | |
368 | ip->i_udquot = ip->i_gdquot = NULL; | |
369 | ih->ih_version++; | |
370 | ip->i_flags |= XFS_INEW; | |
371 | ||
372 | write_unlock(&ih->ih_lock); | |
373 | ||
374 | /* | |
375 | * put ip on its cluster's hash chain | |
376 | */ | |
377 | ASSERT(ip->i_chash == NULL && ip->i_cprev == NULL && | |
378 | ip->i_cnext == NULL); | |
379 | ||
380 | chlnew = NULL; | |
381 | ch = XFS_CHASH(mp, ip->i_blkno); | |
382 | chlredo: | |
383 | s = mutex_spinlock(&ch->ch_lock); | |
384 | for (chl = ch->ch_list; chl != NULL; chl = chl->chl_next) { | |
385 | if (chl->chl_blkno == ip->i_blkno) { | |
386 | ||
387 | /* insert this inode into the doubly-linked list | |
388 | * where chl points */ | |
389 | if ((iq = chl->chl_ip)) { | |
390 | ip->i_cprev = iq->i_cprev; | |
391 | iq->i_cprev->i_cnext = ip; | |
392 | iq->i_cprev = ip; | |
393 | ip->i_cnext = iq; | |
394 | } else { | |
395 | ip->i_cnext = ip; | |
396 | ip->i_cprev = ip; | |
397 | } | |
398 | chl->chl_ip = ip; | |
399 | ip->i_chash = chl; | |
400 | break; | |
401 | } | |
402 | } | |
403 | ||
404 | /* no hash list found for this block; add a new hash list */ | |
405 | if (chl == NULL) { | |
406 | if (chlnew == NULL) { | |
407 | mutex_spinunlock(&ch->ch_lock, s); | |
408 | ASSERT(xfs_chashlist_zone != NULL); | |
409 | chlnew = (xfs_chashlist_t *) | |
410 | kmem_zone_alloc(xfs_chashlist_zone, | |
411 | KM_SLEEP); | |
412 | ASSERT(chlnew != NULL); | |
413 | goto chlredo; | |
414 | } else { | |
415 | ip->i_cnext = ip; | |
416 | ip->i_cprev = ip; | |
417 | ip->i_chash = chlnew; | |
418 | chlnew->chl_ip = ip; | |
419 | chlnew->chl_blkno = ip->i_blkno; | |
1fc5d959 DC |
420 | if (ch->ch_list) |
421 | ch->ch_list->chl_prev = chlnew; | |
1da177e4 | 422 | chlnew->chl_next = ch->ch_list; |
1fc5d959 | 423 | chlnew->chl_prev = NULL; |
1da177e4 LT |
424 | ch->ch_list = chlnew; |
425 | chlnew = NULL; | |
426 | } | |
427 | } else { | |
428 | if (chlnew != NULL) { | |
429 | kmem_zone_free(xfs_chashlist_zone, chlnew); | |
430 | } | |
431 | } | |
432 | ||
433 | mutex_spinunlock(&ch->ch_lock, s); | |
434 | ||
435 | ||
436 | /* | |
437 | * Link ip to its mount and thread it on the mount's inode list. | |
438 | */ | |
439 | XFS_MOUNT_ILOCK(mp); | |
440 | if ((iq = mp->m_inodes)) { | |
441 | ASSERT(iq->i_mprev->i_mnext == iq); | |
442 | ip->i_mprev = iq->i_mprev; | |
443 | iq->i_mprev->i_mnext = ip; | |
444 | iq->i_mprev = ip; | |
445 | ip->i_mnext = iq; | |
446 | } else { | |
447 | ip->i_mnext = ip; | |
448 | ip->i_mprev = ip; | |
449 | } | |
450 | mp->m_inodes = ip; | |
451 | ||
452 | XFS_MOUNT_IUNLOCK(mp); | |
453 | ||
454 | return_ip: | |
455 | ASSERT(ip->i_df.if_ext_max == | |
456 | XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t)); | |
457 | ||
458 | ASSERT(((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) != 0) == | |
459 | ((ip->i_iocore.io_flags & XFS_IOCORE_RT) != 0)); | |
460 | ||
461 | *ipp = ip; | |
462 | ||
463 | /* | |
464 | * If we have a real type for an on-disk inode, we can set ops(&unlock) | |
465 | * now. If it's a new inode being created, xfs_ialloc will handle it. | |
466 | */ | |
b83bd138 | 467 | bhv_vfs_init_vnode(XFS_MTOVFS(mp), vp, XFS_ITOBHV(ip), 1); |
1da177e4 LT |
468 | |
469 | return 0; | |
470 | } | |
471 | ||
472 | ||
473 | /* | |
474 | * The 'normal' internal xfs_iget, if needed it will | |
475 | * 'allocate', or 'get', the vnode. | |
476 | */ | |
477 | int | |
478 | xfs_iget( | |
479 | xfs_mount_t *mp, | |
480 | xfs_trans_t *tp, | |
481 | xfs_ino_t ino, | |
482 | uint flags, | |
483 | uint lock_flags, | |
484 | xfs_inode_t **ipp, | |
485 | xfs_daddr_t bno) | |
486 | { | |
487 | struct inode *inode; | |
67fcaa73 | 488 | bhv_vnode_t *vp = NULL; |
1da177e4 LT |
489 | int error; |
490 | ||
1da177e4 LT |
491 | XFS_STATS_INC(xs_ig_attempts); |
492 | ||
ba403ab4 | 493 | retry: |
1da177e4 | 494 | if ((inode = iget_locked(XFS_MTOVFS(mp)->vfs_super, ino))) { |
1da177e4 | 495 | xfs_inode_t *ip; |
1da177e4 | 496 | |
ec86dc02 | 497 | vp = vn_from_inode(inode); |
1da177e4 | 498 | if (inode->i_state & I_NEW) { |
1da177e4 LT |
499 | vn_initialize(inode); |
500 | error = xfs_iget_core(vp, mp, tp, ino, flags, | |
501 | lock_flags, ipp, bno); | |
502 | if (error) { | |
503 | vn_mark_bad(vp); | |
504 | if (inode->i_state & I_NEW) | |
505 | unlock_new_inode(inode); | |
506 | iput(inode); | |
507 | } | |
508 | } else { | |
ba403ab4 CH |
509 | /* |
510 | * If the inode is not fully constructed due to | |
c41564b5 | 511 | * filehandle mismatches wait for the inode to go |
ba403ab4 CH |
512 | * away and try again. |
513 | * | |
514 | * iget_locked will call __wait_on_freeing_inode | |
515 | * to wait for the inode to go away. | |
516 | */ | |
517 | if (is_bad_inode(inode) || | |
75e17b3c | 518 | ((ip = xfs_vtoi(vp)) == NULL)) { |
1da177e4 | 519 | iput(inode); |
ba403ab4 CH |
520 | delay(1); |
521 | goto retry; | |
1da177e4 LT |
522 | } |
523 | ||
1da177e4 LT |
524 | if (lock_flags != 0) |
525 | xfs_ilock(ip, lock_flags); | |
1da177e4 LT |
526 | XFS_STATS_INC(xs_ig_found); |
527 | *ipp = ip; | |
528 | error = 0; | |
529 | } | |
530 | } else | |
531 | error = ENOMEM; /* If we got no inode we are out of memory */ | |
532 | ||
533 | return error; | |
534 | } | |
535 | ||
536 | /* | |
537 | * Do the setup for the various locks within the incore inode. | |
538 | */ | |
539 | void | |
540 | xfs_inode_lock_init( | |
541 | xfs_inode_t *ip, | |
67fcaa73 | 542 | bhv_vnode_t *vp) |
1da177e4 LT |
543 | { |
544 | mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER, | |
545 | "xfsino", (long)vp->v_number); | |
546 | mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", vp->v_number); | |
547 | init_waitqueue_head(&ip->i_ipin_wait); | |
548 | atomic_set(&ip->i_pincount, 0); | |
549 | init_sema(&ip->i_flock, 1, "xfsfino", vp->v_number); | |
550 | } | |
551 | ||
552 | /* | |
553 | * Look for the inode corresponding to the given ino in the hash table. | |
554 | * If it is there and its i_transp pointer matches tp, return it. | |
555 | * Otherwise, return NULL. | |
556 | */ | |
557 | xfs_inode_t * | |
558 | xfs_inode_incore(xfs_mount_t *mp, | |
559 | xfs_ino_t ino, | |
560 | xfs_trans_t *tp) | |
561 | { | |
562 | xfs_ihash_t *ih; | |
563 | xfs_inode_t *ip; | |
71bce256 | 564 | ulong version; |
1da177e4 LT |
565 | |
566 | ih = XFS_IHASH(mp, ino); | |
567 | read_lock(&ih->ih_lock); | |
568 | for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) { | |
569 | if (ip->i_ino == ino) { | |
570 | /* | |
571 | * If we find it and tp matches, return it. | |
71bce256 NS |
572 | * Also move it to the front of the hash list |
573 | * if we find it and it is not already there. | |
1da177e4 LT |
574 | * Otherwise break from the loop and return |
575 | * NULL. | |
576 | */ | |
577 | if (ip->i_transp == tp) { | |
71bce256 | 578 | version = ih->ih_version; |
1da177e4 | 579 | read_unlock(&ih->ih_lock); |
71bce256 | 580 | xfs_ihash_promote(ih, ip, version); |
1da177e4 LT |
581 | return (ip); |
582 | } | |
583 | break; | |
584 | } | |
585 | } | |
586 | read_unlock(&ih->ih_lock); | |
587 | return (NULL); | |
588 | } | |
589 | ||
590 | /* | |
591 | * Decrement reference count of an inode structure and unlock it. | |
592 | * | |
593 | * ip -- the inode being released | |
594 | * lock_flags -- this parameter indicates the inode's locks to be | |
595 | * to be released. See the comment on xfs_iunlock() for a list | |
596 | * of valid values. | |
597 | */ | |
598 | void | |
599 | xfs_iput(xfs_inode_t *ip, | |
600 | uint lock_flags) | |
601 | { | |
67fcaa73 | 602 | bhv_vnode_t *vp = XFS_ITOV(ip); |
1da177e4 LT |
603 | |
604 | vn_trace_entry(vp, "xfs_iput", (inst_t *)__return_address); | |
1da177e4 | 605 | xfs_iunlock(ip, lock_flags); |
1da177e4 LT |
606 | VN_RELE(vp); |
607 | } | |
608 | ||
609 | /* | |
610 | * Special iput for brand-new inodes that are still locked | |
611 | */ | |
612 | void | |
613 | xfs_iput_new(xfs_inode_t *ip, | |
614 | uint lock_flags) | |
615 | { | |
67fcaa73 | 616 | bhv_vnode_t *vp = XFS_ITOV(ip); |
ec86dc02 | 617 | struct inode *inode = vn_to_inode(vp); |
1da177e4 LT |
618 | |
619 | vn_trace_entry(vp, "xfs_iput_new", (inst_t *)__return_address); | |
620 | ||
621 | if ((ip->i_d.di_mode == 0)) { | |
622 | ASSERT(!(ip->i_flags & XFS_IRECLAIMABLE)); | |
623 | vn_mark_bad(vp); | |
624 | } | |
625 | if (inode->i_state & I_NEW) | |
626 | unlock_new_inode(inode); | |
627 | if (lock_flags) | |
628 | xfs_iunlock(ip, lock_flags); | |
629 | VN_RELE(vp); | |
630 | } | |
631 | ||
632 | ||
633 | /* | |
634 | * This routine embodies the part of the reclaim code that pulls | |
635 | * the inode from the inode hash table and the mount structure's | |
636 | * inode list. | |
637 | * This should only be called from xfs_reclaim(). | |
638 | */ | |
639 | void | |
640 | xfs_ireclaim(xfs_inode_t *ip) | |
641 | { | |
67fcaa73 | 642 | bhv_vnode_t *vp; |
1da177e4 LT |
643 | |
644 | /* | |
645 | * Remove from old hash list and mount list. | |
646 | */ | |
647 | XFS_STATS_INC(xs_ig_reclaims); | |
648 | ||
649 | xfs_iextract(ip); | |
650 | ||
651 | /* | |
652 | * Here we do a spurious inode lock in order to coordinate with | |
653 | * xfs_sync(). This is because xfs_sync() references the inodes | |
654 | * in the mount list without taking references on the corresponding | |
655 | * vnodes. We make that OK here by ensuring that we wait until | |
656 | * the inode is unlocked in xfs_sync() before we go ahead and | |
657 | * free it. We get both the regular lock and the io lock because | |
658 | * the xfs_sync() code may need to drop the regular one but will | |
659 | * still hold the io lock. | |
660 | */ | |
661 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
662 | ||
663 | /* | |
664 | * Release dquots (and their references) if any. An inode may escape | |
665 | * xfs_inactive and get here via vn_alloc->vn_reclaim path. | |
666 | */ | |
667 | XFS_QM_DQDETACH(ip->i_mount, ip); | |
668 | ||
669 | /* | |
670 | * Pull our behavior descriptor from the vnode chain. | |
671 | */ | |
672 | vp = XFS_ITOV_NULL(ip); | |
673 | if (vp) { | |
674 | vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip)); | |
675 | } | |
676 | ||
677 | /* | |
678 | * Free all memory associated with the inode. | |
679 | */ | |
680 | xfs_idestroy(ip); | |
681 | } | |
682 | ||
683 | /* | |
684 | * This routine removes an about-to-be-destroyed inode from | |
685 | * all of the lists in which it is located with the exception | |
686 | * of the behavior chain. | |
687 | */ | |
688 | void | |
689 | xfs_iextract( | |
690 | xfs_inode_t *ip) | |
691 | { | |
692 | xfs_ihash_t *ih; | |
693 | xfs_inode_t *iq; | |
694 | xfs_mount_t *mp; | |
695 | xfs_chash_t *ch; | |
696 | xfs_chashlist_t *chl, *chm; | |
697 | SPLDECL(s); | |
698 | ||
699 | ih = ip->i_hash; | |
700 | write_lock(&ih->ih_lock); | |
701 | if ((iq = ip->i_next)) { | |
702 | iq->i_prevp = ip->i_prevp; | |
703 | } | |
704 | *ip->i_prevp = iq; | |
71bce256 | 705 | ih->ih_version++; |
1da177e4 LT |
706 | write_unlock(&ih->ih_lock); |
707 | ||
708 | /* | |
709 | * Remove from cluster hash list | |
710 | * 1) delete the chashlist if this is the last inode on the chashlist | |
711 | * 2) unchain from list of inodes | |
712 | * 3) point chashlist->chl_ip to 'chl_next' if to this inode. | |
713 | */ | |
714 | mp = ip->i_mount; | |
715 | ch = XFS_CHASH(mp, ip->i_blkno); | |
716 | s = mutex_spinlock(&ch->ch_lock); | |
717 | ||
718 | if (ip->i_cnext == ip) { | |
719 | /* Last inode on chashlist */ | |
720 | ASSERT(ip->i_cnext == ip && ip->i_cprev == ip); | |
721 | ASSERT(ip->i_chash != NULL); | |
722 | chm=NULL; | |
1fc5d959 DC |
723 | chl = ip->i_chash; |
724 | if (chl->chl_prev) | |
725 | chl->chl_prev->chl_next = chl->chl_next; | |
726 | else | |
727 | ch->ch_list = chl->chl_next; | |
728 | if (chl->chl_next) | |
729 | chl->chl_next->chl_prev = chl->chl_prev; | |
730 | kmem_zone_free(xfs_chashlist_zone, chl); | |
731 | } else { | |
1da177e4 LT |
732 | /* delete one inode from a non-empty list */ |
733 | iq = ip->i_cnext; | |
734 | iq->i_cprev = ip->i_cprev; | |
735 | ip->i_cprev->i_cnext = iq; | |
736 | if (ip->i_chash->chl_ip == ip) { | |
737 | ip->i_chash->chl_ip = iq; | |
738 | } | |
739 | ip->i_chash = __return_address; | |
740 | ip->i_cprev = __return_address; | |
741 | ip->i_cnext = __return_address; | |
742 | } | |
743 | mutex_spinunlock(&ch->ch_lock, s); | |
744 | ||
745 | /* | |
746 | * Remove from mount's inode list. | |
747 | */ | |
748 | XFS_MOUNT_ILOCK(mp); | |
749 | ASSERT((ip->i_mnext != NULL) && (ip->i_mprev != NULL)); | |
750 | iq = ip->i_mnext; | |
751 | iq->i_mprev = ip->i_mprev; | |
752 | ip->i_mprev->i_mnext = iq; | |
753 | ||
754 | /* | |
755 | * Fix up the head pointer if it points to the inode being deleted. | |
756 | */ | |
757 | if (mp->m_inodes == ip) { | |
758 | if (ip == iq) { | |
759 | mp->m_inodes = NULL; | |
760 | } else { | |
761 | mp->m_inodes = iq; | |
762 | } | |
763 | } | |
764 | ||
765 | /* Deal with the deleted inodes list */ | |
766 | list_del_init(&ip->i_reclaim); | |
767 | ||
768 | mp->m_ireclaims++; | |
769 | XFS_MOUNT_IUNLOCK(mp); | |
770 | } | |
771 | ||
772 | /* | |
773 | * This is a wrapper routine around the xfs_ilock() routine | |
774 | * used to centralize some grungy code. It is used in places | |
775 | * that wish to lock the inode solely for reading the extents. | |
776 | * The reason these places can't just call xfs_ilock(SHARED) | |
777 | * is that the inode lock also guards to bringing in of the | |
778 | * extents from disk for a file in b-tree format. If the inode | |
779 | * is in b-tree format, then we need to lock the inode exclusively | |
780 | * until the extents are read in. Locking it exclusively all | |
781 | * the time would limit our parallelism unnecessarily, though. | |
782 | * What we do instead is check to see if the extents have been | |
783 | * read in yet, and only lock the inode exclusively if they | |
784 | * have not. | |
785 | * | |
786 | * The function returns a value which should be given to the | |
787 | * corresponding xfs_iunlock_map_shared(). This value is | |
788 | * the mode in which the lock was actually taken. | |
789 | */ | |
790 | uint | |
791 | xfs_ilock_map_shared( | |
792 | xfs_inode_t *ip) | |
793 | { | |
794 | uint lock_mode; | |
795 | ||
796 | if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) && | |
797 | ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) { | |
798 | lock_mode = XFS_ILOCK_EXCL; | |
799 | } else { | |
800 | lock_mode = XFS_ILOCK_SHARED; | |
801 | } | |
802 | ||
803 | xfs_ilock(ip, lock_mode); | |
804 | ||
805 | return lock_mode; | |
806 | } | |
807 | ||
808 | /* | |
809 | * This is simply the unlock routine to go with xfs_ilock_map_shared(). | |
810 | * All it does is call xfs_iunlock() with the given lock_mode. | |
811 | */ | |
812 | void | |
813 | xfs_iunlock_map_shared( | |
814 | xfs_inode_t *ip, | |
815 | unsigned int lock_mode) | |
816 | { | |
817 | xfs_iunlock(ip, lock_mode); | |
818 | } | |
819 | ||
820 | /* | |
821 | * The xfs inode contains 2 locks: a multi-reader lock called the | |
822 | * i_iolock and a multi-reader lock called the i_lock. This routine | |
823 | * allows either or both of the locks to be obtained. | |
824 | * | |
825 | * The 2 locks should always be ordered so that the IO lock is | |
826 | * obtained first in order to prevent deadlock. | |
827 | * | |
828 | * ip -- the inode being locked | |
829 | * lock_flags -- this parameter indicates the inode's locks | |
830 | * to be locked. It can be: | |
831 | * XFS_IOLOCK_SHARED, | |
832 | * XFS_IOLOCK_EXCL, | |
833 | * XFS_ILOCK_SHARED, | |
834 | * XFS_ILOCK_EXCL, | |
835 | * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED, | |
836 | * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL, | |
837 | * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED, | |
838 | * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL | |
839 | */ | |
840 | void | |
841 | xfs_ilock(xfs_inode_t *ip, | |
842 | uint lock_flags) | |
843 | { | |
844 | /* | |
845 | * You can't set both SHARED and EXCL for the same lock, | |
846 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
847 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
848 | */ | |
849 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
850 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
851 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
852 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
853 | ASSERT((lock_flags & ~XFS_LOCK_MASK) == 0); | |
854 | ||
855 | if (lock_flags & XFS_IOLOCK_EXCL) { | |
856 | mrupdate(&ip->i_iolock); | |
857 | } else if (lock_flags & XFS_IOLOCK_SHARED) { | |
858 | mraccess(&ip->i_iolock); | |
859 | } | |
860 | if (lock_flags & XFS_ILOCK_EXCL) { | |
861 | mrupdate(&ip->i_lock); | |
862 | } else if (lock_flags & XFS_ILOCK_SHARED) { | |
863 | mraccess(&ip->i_lock); | |
864 | } | |
865 | xfs_ilock_trace(ip, 1, lock_flags, (inst_t *)__return_address); | |
866 | } | |
867 | ||
868 | /* | |
869 | * This is just like xfs_ilock(), except that the caller | |
870 | * is guaranteed not to sleep. It returns 1 if it gets | |
871 | * the requested locks and 0 otherwise. If the IO lock is | |
872 | * obtained but the inode lock cannot be, then the IO lock | |
873 | * is dropped before returning. | |
874 | * | |
875 | * ip -- the inode being locked | |
876 | * lock_flags -- this parameter indicates the inode's locks to be | |
877 | * to be locked. See the comment for xfs_ilock() for a list | |
878 | * of valid values. | |
879 | * | |
880 | */ | |
881 | int | |
882 | xfs_ilock_nowait(xfs_inode_t *ip, | |
883 | uint lock_flags) | |
884 | { | |
885 | int iolocked; | |
886 | int ilocked; | |
887 | ||
888 | /* | |
889 | * You can't set both SHARED and EXCL for the same lock, | |
890 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
891 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
892 | */ | |
893 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
894 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
895 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
896 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
897 | ASSERT((lock_flags & ~XFS_LOCK_MASK) == 0); | |
898 | ||
899 | iolocked = 0; | |
900 | if (lock_flags & XFS_IOLOCK_EXCL) { | |
901 | iolocked = mrtryupdate(&ip->i_iolock); | |
902 | if (!iolocked) { | |
903 | return 0; | |
904 | } | |
905 | } else if (lock_flags & XFS_IOLOCK_SHARED) { | |
906 | iolocked = mrtryaccess(&ip->i_iolock); | |
907 | if (!iolocked) { | |
908 | return 0; | |
909 | } | |
910 | } | |
911 | if (lock_flags & XFS_ILOCK_EXCL) { | |
912 | ilocked = mrtryupdate(&ip->i_lock); | |
913 | if (!ilocked) { | |
914 | if (iolocked) { | |
915 | mrunlock(&ip->i_iolock); | |
916 | } | |
917 | return 0; | |
918 | } | |
919 | } else if (lock_flags & XFS_ILOCK_SHARED) { | |
920 | ilocked = mrtryaccess(&ip->i_lock); | |
921 | if (!ilocked) { | |
922 | if (iolocked) { | |
923 | mrunlock(&ip->i_iolock); | |
924 | } | |
925 | return 0; | |
926 | } | |
927 | } | |
928 | xfs_ilock_trace(ip, 2, lock_flags, (inst_t *)__return_address); | |
929 | return 1; | |
930 | } | |
931 | ||
932 | /* | |
933 | * xfs_iunlock() is used to drop the inode locks acquired with | |
934 | * xfs_ilock() and xfs_ilock_nowait(). The caller must pass | |
935 | * in the flags given to xfs_ilock() or xfs_ilock_nowait() so | |
936 | * that we know which locks to drop. | |
937 | * | |
938 | * ip -- the inode being unlocked | |
939 | * lock_flags -- this parameter indicates the inode's locks to be | |
940 | * to be unlocked. See the comment for xfs_ilock() for a list | |
941 | * of valid values for this parameter. | |
942 | * | |
943 | */ | |
944 | void | |
945 | xfs_iunlock(xfs_inode_t *ip, | |
946 | uint lock_flags) | |
947 | { | |
948 | /* | |
949 | * You can't set both SHARED and EXCL for the same lock, | |
950 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
951 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
952 | */ | |
953 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
954 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
955 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
956 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
957 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_IUNLOCK_NONOTIFY)) == 0); | |
958 | ASSERT(lock_flags != 0); | |
959 | ||
960 | if (lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) { | |
961 | ASSERT(!(lock_flags & XFS_IOLOCK_SHARED) || | |
962 | (ismrlocked(&ip->i_iolock, MR_ACCESS))); | |
963 | ASSERT(!(lock_flags & XFS_IOLOCK_EXCL) || | |
964 | (ismrlocked(&ip->i_iolock, MR_UPDATE))); | |
965 | mrunlock(&ip->i_iolock); | |
966 | } | |
967 | ||
968 | if (lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) { | |
969 | ASSERT(!(lock_flags & XFS_ILOCK_SHARED) || | |
970 | (ismrlocked(&ip->i_lock, MR_ACCESS))); | |
971 | ASSERT(!(lock_flags & XFS_ILOCK_EXCL) || | |
972 | (ismrlocked(&ip->i_lock, MR_UPDATE))); | |
973 | mrunlock(&ip->i_lock); | |
974 | ||
975 | /* | |
976 | * Let the AIL know that this item has been unlocked in case | |
977 | * it is in the AIL and anyone is waiting on it. Don't do | |
978 | * this if the caller has asked us not to. | |
979 | */ | |
980 | if (!(lock_flags & XFS_IUNLOCK_NONOTIFY) && | |
981 | ip->i_itemp != NULL) { | |
982 | xfs_trans_unlocked_item(ip->i_mount, | |
983 | (xfs_log_item_t*)(ip->i_itemp)); | |
984 | } | |
985 | } | |
986 | xfs_ilock_trace(ip, 3, lock_flags, (inst_t *)__return_address); | |
987 | } | |
988 | ||
989 | /* | |
990 | * give up write locks. the i/o lock cannot be held nested | |
991 | * if it is being demoted. | |
992 | */ | |
993 | void | |
994 | xfs_ilock_demote(xfs_inode_t *ip, | |
995 | uint lock_flags) | |
996 | { | |
997 | ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)); | |
998 | ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0); | |
999 | ||
1000 | if (lock_flags & XFS_ILOCK_EXCL) { | |
1001 | ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE)); | |
1002 | mrdemote(&ip->i_lock); | |
1003 | } | |
1004 | if (lock_flags & XFS_IOLOCK_EXCL) { | |
1005 | ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE)); | |
1006 | mrdemote(&ip->i_iolock); | |
1007 | } | |
1008 | } | |
1009 | ||
1010 | /* | |
1011 | * The following three routines simply manage the i_flock | |
1012 | * semaphore embedded in the inode. This semaphore synchronizes | |
1013 | * processes attempting to flush the in-core inode back to disk. | |
1014 | */ | |
1015 | void | |
1016 | xfs_iflock(xfs_inode_t *ip) | |
1017 | { | |
1018 | psema(&(ip->i_flock), PINOD|PLTWAIT); | |
1019 | } | |
1020 | ||
1021 | int | |
1022 | xfs_iflock_nowait(xfs_inode_t *ip) | |
1023 | { | |
1024 | return (cpsema(&(ip->i_flock))); | |
1025 | } | |
1026 | ||
1027 | void | |
1028 | xfs_ifunlock(xfs_inode_t *ip) | |
1029 | { | |
0d8fee32 | 1030 | ASSERT(issemalocked(&(ip->i_flock))); |
1da177e4 LT |
1031 | vsema(&(ip->i_flock)); |
1032 | } |