]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. | |
3 | * All Rights Reserved. | |
4 | * | |
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 | |
7 | * published by the Free Software Foundation. | |
8 | * | |
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. | |
13 | * | |
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 | |
17 | */ | |
18 | #include "xfs.h" | |
19 | #include "xfs_fs.h" | |
20 | #include "xfs_types.h" | |
21 | #include "xfs_acl.h" | |
22 | #include "xfs_bit.h" | |
23 | #include "xfs_log.h" | |
24 | #include "xfs_inum.h" | |
25 | #include "xfs_trans.h" | |
26 | #include "xfs_sb.h" | |
27 | #include "xfs_ag.h" | |
28 | #include "xfs_dir2.h" | |
29 | #include "xfs_dmapi.h" | |
30 | #include "xfs_mount.h" | |
31 | #include "xfs_bmap_btree.h" | |
32 | #include "xfs_alloc_btree.h" | |
33 | #include "xfs_ialloc_btree.h" | |
34 | #include "xfs_dir2_sf.h" | |
35 | #include "xfs_attr_sf.h" | |
36 | #include "xfs_dinode.h" | |
37 | #include "xfs_inode.h" | |
38 | #include "xfs_btree.h" | |
39 | #include "xfs_ialloc.h" | |
40 | #include "xfs_quota.h" | |
41 | #include "xfs_utils.h" | |
42 | #include "xfs_trans_priv.h" | |
43 | #include "xfs_inode_item.h" | |
44 | #include "xfs_bmap.h" | |
45 | #include "xfs_btree_trace.h" | |
46 | #include "xfs_dir2_trace.h" | |
47 | ||
48 | ||
49 | /* | |
50 | * Allocate and initialise an xfs_inode. | |
51 | */ | |
52 | STATIC struct xfs_inode * | |
53 | xfs_inode_alloc( | |
54 | struct xfs_mount *mp, | |
55 | xfs_ino_t ino) | |
56 | { | |
57 | struct xfs_inode *ip; | |
58 | ||
59 | /* | |
60 | * if this didn't occur in transactions, we could use | |
61 | * KM_MAYFAIL and return NULL here on ENOMEM. Set the | |
62 | * code up to do this anyway. | |
63 | */ | |
64 | ip = kmem_zone_alloc(xfs_inode_zone, KM_SLEEP); | |
65 | if (!ip) | |
66 | return NULL; | |
67 | if (inode_init_always(mp->m_super, VFS_I(ip))) { | |
68 | kmem_zone_free(xfs_inode_zone, ip); | |
69 | return NULL; | |
70 | } | |
71 | ||
72 | ASSERT(atomic_read(&ip->i_iocount) == 0); | |
73 | ASSERT(atomic_read(&ip->i_pincount) == 0); | |
74 | ASSERT(!spin_is_locked(&ip->i_flags_lock)); | |
75 | ASSERT(completion_done(&ip->i_flush)); | |
76 | ||
77 | /* initialise the xfs inode */ | |
78 | ip->i_ino = ino; | |
79 | ip->i_mount = mp; | |
80 | memset(&ip->i_imap, 0, sizeof(struct xfs_imap)); | |
81 | ip->i_afp = NULL; | |
82 | memset(&ip->i_df, 0, sizeof(xfs_ifork_t)); | |
83 | ip->i_flags = 0; | |
84 | ip->i_update_core = 0; | |
85 | ip->i_update_size = 0; | |
86 | ip->i_delayed_blks = 0; | |
87 | memset(&ip->i_d, 0, sizeof(xfs_icdinode_t)); | |
88 | ip->i_size = 0; | |
89 | ip->i_new_size = 0; | |
90 | ||
91 | /* | |
92 | * Initialize inode's trace buffers. | |
93 | */ | |
94 | #ifdef XFS_INODE_TRACE | |
95 | ip->i_trace = ktrace_alloc(INODE_TRACE_SIZE, KM_NOFS); | |
96 | #endif | |
97 | #ifdef XFS_BMAP_TRACE | |
98 | ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_NOFS); | |
99 | #endif | |
100 | #ifdef XFS_BTREE_TRACE | |
101 | ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_NOFS); | |
102 | #endif | |
103 | #ifdef XFS_RW_TRACE | |
104 | ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_NOFS); | |
105 | #endif | |
106 | #ifdef XFS_ILOCK_TRACE | |
107 | ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_NOFS); | |
108 | #endif | |
109 | #ifdef XFS_DIR2_TRACE | |
110 | ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_NOFS); | |
111 | #endif | |
112 | ||
113 | /* prevent anyone from using this yet */ | |
114 | VFS_I(ip)->i_state = I_NEW|I_LOCK; | |
115 | ||
116 | return ip; | |
117 | } | |
118 | ||
119 | STATIC void | |
120 | xfs_inode_free( | |
121 | struct xfs_inode *ip) | |
122 | { | |
123 | switch (ip->i_d.di_mode & S_IFMT) { | |
124 | case S_IFREG: | |
125 | case S_IFDIR: | |
126 | case S_IFLNK: | |
127 | xfs_idestroy_fork(ip, XFS_DATA_FORK); | |
128 | break; | |
129 | } | |
130 | ||
131 | if (ip->i_afp) | |
132 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
133 | ||
134 | #ifdef XFS_INODE_TRACE | |
135 | ktrace_free(ip->i_trace); | |
136 | #endif | |
137 | #ifdef XFS_BMAP_TRACE | |
138 | ktrace_free(ip->i_xtrace); | |
139 | #endif | |
140 | #ifdef XFS_BTREE_TRACE | |
141 | ktrace_free(ip->i_btrace); | |
142 | #endif | |
143 | #ifdef XFS_RW_TRACE | |
144 | ktrace_free(ip->i_rwtrace); | |
145 | #endif | |
146 | #ifdef XFS_ILOCK_TRACE | |
147 | ktrace_free(ip->i_lock_trace); | |
148 | #endif | |
149 | #ifdef XFS_DIR2_TRACE | |
150 | ktrace_free(ip->i_dir_trace); | |
151 | #endif | |
152 | ||
153 | if (ip->i_itemp) { | |
154 | /* | |
155 | * Only if we are shutting down the fs will we see an | |
156 | * inode still in the AIL. If it is there, we should remove | |
157 | * it to prevent a use-after-free from occurring. | |
158 | */ | |
159 | xfs_log_item_t *lip = &ip->i_itemp->ili_item; | |
160 | struct xfs_ail *ailp = lip->li_ailp; | |
161 | ||
162 | ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) || | |
163 | XFS_FORCED_SHUTDOWN(ip->i_mount)); | |
164 | if (lip->li_flags & XFS_LI_IN_AIL) { | |
165 | spin_lock(&ailp->xa_lock); | |
166 | if (lip->li_flags & XFS_LI_IN_AIL) | |
167 | xfs_trans_ail_delete(ailp, lip); | |
168 | else | |
169 | spin_unlock(&ailp->xa_lock); | |
170 | } | |
171 | xfs_inode_item_destroy(ip); | |
172 | ip->i_itemp = NULL; | |
173 | } | |
174 | ||
175 | /* asserts to verify all state is correct here */ | |
176 | ASSERT(atomic_read(&ip->i_iocount) == 0); | |
177 | ASSERT(atomic_read(&ip->i_pincount) == 0); | |
178 | ASSERT(!spin_is_locked(&ip->i_flags_lock)); | |
179 | ASSERT(completion_done(&ip->i_flush)); | |
180 | ||
181 | kmem_zone_free(xfs_inode_zone, ip); | |
182 | } | |
183 | ||
184 | /* | |
185 | * Check the validity of the inode we just found it the cache | |
186 | */ | |
187 | static int | |
188 | xfs_iget_cache_hit( | |
189 | struct xfs_perag *pag, | |
190 | struct xfs_inode *ip, | |
191 | int flags, | |
192 | int lock_flags) __releases(pag->pag_ici_lock) | |
193 | { | |
194 | struct inode *inode = VFS_I(ip); | |
195 | struct xfs_mount *mp = ip->i_mount; | |
196 | int error; | |
197 | ||
198 | spin_lock(&ip->i_flags_lock); | |
199 | ||
200 | /* | |
201 | * If we are racing with another cache hit that is currently | |
202 | * instantiating this inode or currently recycling it out of | |
203 | * reclaimabe state, wait for the initialisation to complete | |
204 | * before continuing. | |
205 | * | |
206 | * XXX(hch): eventually we should do something equivalent to | |
207 | * wait_on_inode to wait for these flags to be cleared | |
208 | * instead of polling for it. | |
209 | */ | |
210 | if (ip->i_flags & (XFS_INEW|XFS_IRECLAIM)) { | |
211 | XFS_STATS_INC(xs_ig_frecycle); | |
212 | error = EAGAIN; | |
213 | goto out_error; | |
214 | } | |
215 | ||
216 | /* | |
217 | * If lookup is racing with unlink return an error immediately. | |
218 | */ | |
219 | if (ip->i_d.di_mode == 0 && !(flags & XFS_IGET_CREATE)) { | |
220 | error = ENOENT; | |
221 | goto out_error; | |
222 | } | |
223 | ||
224 | /* | |
225 | * If IRECLAIMABLE is set, we've torn down the VFS inode already. | |
226 | * Need to carefully get it back into useable state. | |
227 | */ | |
228 | if (ip->i_flags & XFS_IRECLAIMABLE) { | |
229 | xfs_itrace_exit_tag(ip, "xfs_iget.alloc"); | |
230 | ||
231 | /* | |
232 | * We need to set XFS_INEW atomically with clearing the | |
233 | * reclaimable tag so that we do have an indicator of the | |
234 | * inode still being initialized. | |
235 | */ | |
236 | ip->i_flags |= XFS_INEW; | |
237 | ip->i_flags &= ~XFS_IRECLAIMABLE; | |
238 | __xfs_inode_clear_reclaim_tag(mp, pag, ip); | |
239 | ||
240 | spin_unlock(&ip->i_flags_lock); | |
241 | read_unlock(&pag->pag_ici_lock); | |
242 | ||
243 | error = -inode_init_always(mp->m_super, inode); | |
244 | if (error) { | |
245 | /* | |
246 | * Re-initializing the inode failed, and we are in deep | |
247 | * trouble. Try to re-add it to the reclaim list. | |
248 | */ | |
249 | read_lock(&pag->pag_ici_lock); | |
250 | spin_lock(&ip->i_flags_lock); | |
251 | ||
252 | ip->i_flags &= ~XFS_INEW; | |
253 | ip->i_flags |= XFS_IRECLAIMABLE; | |
254 | __xfs_inode_set_reclaim_tag(pag, ip); | |
255 | goto out_error; | |
256 | } | |
257 | inode->i_state = I_LOCK|I_NEW; | |
258 | } else { | |
259 | /* If the VFS inode is being torn down, pause and try again. */ | |
260 | if (!igrab(inode)) { | |
261 | error = EAGAIN; | |
262 | goto out_error; | |
263 | } | |
264 | ||
265 | /* We've got a live one. */ | |
266 | spin_unlock(&ip->i_flags_lock); | |
267 | read_unlock(&pag->pag_ici_lock); | |
268 | } | |
269 | ||
270 | if (lock_flags != 0) | |
271 | xfs_ilock(ip, lock_flags); | |
272 | ||
273 | xfs_iflags_clear(ip, XFS_ISTALE); | |
274 | xfs_itrace_exit_tag(ip, "xfs_iget.found"); | |
275 | XFS_STATS_INC(xs_ig_found); | |
276 | return 0; | |
277 | ||
278 | out_error: | |
279 | spin_unlock(&ip->i_flags_lock); | |
280 | read_unlock(&pag->pag_ici_lock); | |
281 | return error; | |
282 | } | |
283 | ||
284 | ||
285 | static int | |
286 | xfs_iget_cache_miss( | |
287 | struct xfs_mount *mp, | |
288 | struct xfs_perag *pag, | |
289 | xfs_trans_t *tp, | |
290 | xfs_ino_t ino, | |
291 | struct xfs_inode **ipp, | |
292 | xfs_daddr_t bno, | |
293 | int flags, | |
294 | int lock_flags) __releases(pag->pag_ici_lock) | |
295 | { | |
296 | struct xfs_inode *ip; | |
297 | int error; | |
298 | unsigned long first_index, mask; | |
299 | xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino); | |
300 | ||
301 | ip = xfs_inode_alloc(mp, ino); | |
302 | if (!ip) | |
303 | return ENOMEM; | |
304 | ||
305 | error = xfs_iread(mp, tp, ip, bno, flags); | |
306 | if (error) | |
307 | goto out_destroy; | |
308 | ||
309 | xfs_itrace_exit_tag(ip, "xfs_iget.alloc"); | |
310 | ||
311 | if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) { | |
312 | error = ENOENT; | |
313 | goto out_destroy; | |
314 | } | |
315 | ||
316 | /* | |
317 | * Preload the radix tree so we can insert safely under the | |
318 | * write spinlock. Note that we cannot sleep inside the preload | |
319 | * region. | |
320 | */ | |
321 | if (radix_tree_preload(GFP_KERNEL)) { | |
322 | error = EAGAIN; | |
323 | goto out_destroy; | |
324 | } | |
325 | ||
326 | /* | |
327 | * Because the inode hasn't been added to the radix-tree yet it can't | |
328 | * be found by another thread, so we can do the non-sleeping lock here. | |
329 | */ | |
330 | if (lock_flags) { | |
331 | if (!xfs_ilock_nowait(ip, lock_flags)) | |
332 | BUG(); | |
333 | } | |
334 | ||
335 | mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1); | |
336 | first_index = agino & mask; | |
337 | write_lock(&pag->pag_ici_lock); | |
338 | ||
339 | /* insert the new inode */ | |
340 | error = radix_tree_insert(&pag->pag_ici_root, agino, ip); | |
341 | if (unlikely(error)) { | |
342 | WARN_ON(error != -EEXIST); | |
343 | XFS_STATS_INC(xs_ig_dup); | |
344 | error = EAGAIN; | |
345 | goto out_preload_end; | |
346 | } | |
347 | ||
348 | /* These values _must_ be set before releasing the radix tree lock! */ | |
349 | ip->i_udquot = ip->i_gdquot = NULL; | |
350 | xfs_iflags_set(ip, XFS_INEW); | |
351 | ||
352 | write_unlock(&pag->pag_ici_lock); | |
353 | radix_tree_preload_end(); | |
354 | *ipp = ip; | |
355 | return 0; | |
356 | ||
357 | out_preload_end: | |
358 | write_unlock(&pag->pag_ici_lock); | |
359 | radix_tree_preload_end(); | |
360 | if (lock_flags) | |
361 | xfs_iunlock(ip, lock_flags); | |
362 | out_destroy: | |
363 | __destroy_inode(VFS_I(ip)); | |
364 | xfs_inode_free(ip); | |
365 | return error; | |
366 | } | |
367 | ||
368 | /* | |
369 | * Look up an inode by number in the given file system. | |
370 | * The inode is looked up in the cache held in each AG. | |
371 | * If the inode is found in the cache, initialise the vfs inode | |
372 | * if necessary. | |
373 | * | |
374 | * If it is not in core, read it in from the file system's device, | |
375 | * add it to the cache and initialise the vfs inode. | |
376 | * | |
377 | * The inode is locked according to the value of the lock_flags parameter. | |
378 | * This flag parameter indicates how and if the inode's IO lock and inode lock | |
379 | * should be taken. | |
380 | * | |
381 | * mp -- the mount point structure for the current file system. It points | |
382 | * to the inode hash table. | |
383 | * tp -- a pointer to the current transaction if there is one. This is | |
384 | * simply passed through to the xfs_iread() call. | |
385 | * ino -- the number of the inode desired. This is the unique identifier | |
386 | * within the file system for the inode being requested. | |
387 | * lock_flags -- flags indicating how to lock the inode. See the comment | |
388 | * for xfs_ilock() for a list of valid values. | |
389 | * bno -- the block number starting the buffer containing the inode, | |
390 | * if known (as by bulkstat), else 0. | |
391 | */ | |
392 | int | |
393 | xfs_iget( | |
394 | xfs_mount_t *mp, | |
395 | xfs_trans_t *tp, | |
396 | xfs_ino_t ino, | |
397 | uint flags, | |
398 | uint lock_flags, | |
399 | xfs_inode_t **ipp, | |
400 | xfs_daddr_t bno) | |
401 | { | |
402 | xfs_inode_t *ip; | |
403 | int error; | |
404 | xfs_perag_t *pag; | |
405 | xfs_agino_t agino; | |
406 | ||
407 | /* the radix tree exists only in inode capable AGs */ | |
408 | if (XFS_INO_TO_AGNO(mp, ino) >= mp->m_maxagi) | |
409 | return EINVAL; | |
410 | ||
411 | /* get the perag structure and ensure that it's inode capable */ | |
412 | pag = xfs_get_perag(mp, ino); | |
413 | if (!pag->pagi_inodeok) | |
414 | return EINVAL; | |
415 | ASSERT(pag->pag_ici_init); | |
416 | agino = XFS_INO_TO_AGINO(mp, ino); | |
417 | ||
418 | again: | |
419 | error = 0; | |
420 | read_lock(&pag->pag_ici_lock); | |
421 | ip = radix_tree_lookup(&pag->pag_ici_root, agino); | |
422 | ||
423 | if (ip) { | |
424 | error = xfs_iget_cache_hit(pag, ip, flags, lock_flags); | |
425 | if (error) | |
426 | goto out_error_or_again; | |
427 | } else { | |
428 | read_unlock(&pag->pag_ici_lock); | |
429 | XFS_STATS_INC(xs_ig_missed); | |
430 | ||
431 | error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip, bno, | |
432 | flags, lock_flags); | |
433 | if (error) | |
434 | goto out_error_or_again; | |
435 | } | |
436 | xfs_put_perag(mp, pag); | |
437 | ||
438 | *ipp = ip; | |
439 | ||
440 | ASSERT(ip->i_df.if_ext_max == | |
441 | XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t)); | |
442 | /* | |
443 | * If we have a real type for an on-disk inode, we can set ops(&unlock) | |
444 | * now. If it's a new inode being created, xfs_ialloc will handle it. | |
445 | */ | |
446 | if (xfs_iflags_test(ip, XFS_INEW) && ip->i_d.di_mode != 0) | |
447 | xfs_setup_inode(ip); | |
448 | return 0; | |
449 | ||
450 | out_error_or_again: | |
451 | if (error == EAGAIN) { | |
452 | delay(1); | |
453 | goto again; | |
454 | } | |
455 | xfs_put_perag(mp, pag); | |
456 | return error; | |
457 | } | |
458 | ||
459 | ||
460 | /* | |
461 | * Look for the inode corresponding to the given ino in the hash table. | |
462 | * If it is there and its i_transp pointer matches tp, return it. | |
463 | * Otherwise, return NULL. | |
464 | */ | |
465 | xfs_inode_t * | |
466 | xfs_inode_incore(xfs_mount_t *mp, | |
467 | xfs_ino_t ino, | |
468 | xfs_trans_t *tp) | |
469 | { | |
470 | xfs_inode_t *ip; | |
471 | xfs_perag_t *pag; | |
472 | ||
473 | pag = xfs_get_perag(mp, ino); | |
474 | read_lock(&pag->pag_ici_lock); | |
475 | ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ino)); | |
476 | read_unlock(&pag->pag_ici_lock); | |
477 | xfs_put_perag(mp, pag); | |
478 | ||
479 | /* the returned inode must match the transaction */ | |
480 | if (ip && (ip->i_transp != tp)) | |
481 | return NULL; | |
482 | return ip; | |
483 | } | |
484 | ||
485 | /* | |
486 | * Decrement reference count of an inode structure and unlock it. | |
487 | * | |
488 | * ip -- the inode being released | |
489 | * lock_flags -- this parameter indicates the inode's locks to be | |
490 | * to be released. See the comment on xfs_iunlock() for a list | |
491 | * of valid values. | |
492 | */ | |
493 | void | |
494 | xfs_iput(xfs_inode_t *ip, | |
495 | uint lock_flags) | |
496 | { | |
497 | xfs_itrace_entry(ip); | |
498 | xfs_iunlock(ip, lock_flags); | |
499 | IRELE(ip); | |
500 | } | |
501 | ||
502 | /* | |
503 | * Special iput for brand-new inodes that are still locked | |
504 | */ | |
505 | void | |
506 | xfs_iput_new( | |
507 | xfs_inode_t *ip, | |
508 | uint lock_flags) | |
509 | { | |
510 | struct inode *inode = VFS_I(ip); | |
511 | ||
512 | xfs_itrace_entry(ip); | |
513 | ||
514 | if ((ip->i_d.di_mode == 0)) { | |
515 | ASSERT(!xfs_iflags_test(ip, XFS_IRECLAIMABLE)); | |
516 | make_bad_inode(inode); | |
517 | } | |
518 | if (inode->i_state & I_NEW) | |
519 | unlock_new_inode(inode); | |
520 | if (lock_flags) | |
521 | xfs_iunlock(ip, lock_flags); | |
522 | IRELE(ip); | |
523 | } | |
524 | ||
525 | /* | |
526 | * This is called free all the memory associated with an inode. | |
527 | * It must free the inode itself and any buffers allocated for | |
528 | * if_extents/if_data and if_broot. It must also free the lock | |
529 | * associated with the inode. | |
530 | * | |
531 | * Note: because we don't initialise everything on reallocation out | |
532 | * of the zone, we must ensure we nullify everything correctly before | |
533 | * freeing the structure. | |
534 | */ | |
535 | void | |
536 | xfs_ireclaim( | |
537 | struct xfs_inode *ip) | |
538 | { | |
539 | struct xfs_mount *mp = ip->i_mount; | |
540 | struct xfs_perag *pag; | |
541 | ||
542 | XFS_STATS_INC(xs_ig_reclaims); | |
543 | ||
544 | /* | |
545 | * Remove the inode from the per-AG radix tree. It doesn't matter | |
546 | * if it was never added to it because radix_tree_delete can deal | |
547 | * with that case just fine. | |
548 | */ | |
549 | pag = xfs_get_perag(mp, ip->i_ino); | |
550 | write_lock(&pag->pag_ici_lock); | |
551 | radix_tree_delete(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ip->i_ino)); | |
552 | write_unlock(&pag->pag_ici_lock); | |
553 | xfs_put_perag(mp, pag); | |
554 | ||
555 | /* | |
556 | * Here we do an (almost) spurious inode lock in order to coordinate | |
557 | * with inode cache radix tree lookups. This is because the lookup | |
558 | * can reference the inodes in the cache without taking references. | |
559 | * | |
560 | * We make that OK here by ensuring that we wait until the inode is | |
561 | * unlocked after the lookup before we go ahead and free it. We get | |
562 | * both the ilock and the iolock because the code may need to drop the | |
563 | * ilock one but will still hold the iolock. | |
564 | */ | |
565 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
566 | xfs_qm_dqdetach(ip); | |
567 | xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
568 | ||
569 | xfs_inode_free(ip); | |
570 | } | |
571 | ||
572 | /* | |
573 | * This is a wrapper routine around the xfs_ilock() routine | |
574 | * used to centralize some grungy code. It is used in places | |
575 | * that wish to lock the inode solely for reading the extents. | |
576 | * The reason these places can't just call xfs_ilock(SHARED) | |
577 | * is that the inode lock also guards to bringing in of the | |
578 | * extents from disk for a file in b-tree format. If the inode | |
579 | * is in b-tree format, then we need to lock the inode exclusively | |
580 | * until the extents are read in. Locking it exclusively all | |
581 | * the time would limit our parallelism unnecessarily, though. | |
582 | * What we do instead is check to see if the extents have been | |
583 | * read in yet, and only lock the inode exclusively if they | |
584 | * have not. | |
585 | * | |
586 | * The function returns a value which should be given to the | |
587 | * corresponding xfs_iunlock_map_shared(). This value is | |
588 | * the mode in which the lock was actually taken. | |
589 | */ | |
590 | uint | |
591 | xfs_ilock_map_shared( | |
592 | xfs_inode_t *ip) | |
593 | { | |
594 | uint lock_mode; | |
595 | ||
596 | if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) && | |
597 | ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) { | |
598 | lock_mode = XFS_ILOCK_EXCL; | |
599 | } else { | |
600 | lock_mode = XFS_ILOCK_SHARED; | |
601 | } | |
602 | ||
603 | xfs_ilock(ip, lock_mode); | |
604 | ||
605 | return lock_mode; | |
606 | } | |
607 | ||
608 | /* | |
609 | * This is simply the unlock routine to go with xfs_ilock_map_shared(). | |
610 | * All it does is call xfs_iunlock() with the given lock_mode. | |
611 | */ | |
612 | void | |
613 | xfs_iunlock_map_shared( | |
614 | xfs_inode_t *ip, | |
615 | unsigned int lock_mode) | |
616 | { | |
617 | xfs_iunlock(ip, lock_mode); | |
618 | } | |
619 | ||
620 | /* | |
621 | * The xfs inode contains 2 locks: a multi-reader lock called the | |
622 | * i_iolock and a multi-reader lock called the i_lock. This routine | |
623 | * allows either or both of the locks to be obtained. | |
624 | * | |
625 | * The 2 locks should always be ordered so that the IO lock is | |
626 | * obtained first in order to prevent deadlock. | |
627 | * | |
628 | * ip -- the inode being locked | |
629 | * lock_flags -- this parameter indicates the inode's locks | |
630 | * to be locked. It can be: | |
631 | * XFS_IOLOCK_SHARED, | |
632 | * XFS_IOLOCK_EXCL, | |
633 | * XFS_ILOCK_SHARED, | |
634 | * XFS_ILOCK_EXCL, | |
635 | * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED, | |
636 | * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL, | |
637 | * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED, | |
638 | * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL | |
639 | */ | |
640 | void | |
641 | xfs_ilock( | |
642 | xfs_inode_t *ip, | |
643 | uint lock_flags) | |
644 | { | |
645 | /* | |
646 | * You can't set both SHARED and EXCL for the same lock, | |
647 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
648 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
649 | */ | |
650 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
651 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
652 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
653 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
654 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); | |
655 | ||
656 | if (lock_flags & XFS_IOLOCK_EXCL) | |
657 | mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); | |
658 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
659 | mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); | |
660 | ||
661 | if (lock_flags & XFS_ILOCK_EXCL) | |
662 | mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); | |
663 | else if (lock_flags & XFS_ILOCK_SHARED) | |
664 | mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); | |
665 | ||
666 | xfs_ilock_trace(ip, 1, lock_flags, (inst_t *)__return_address); | |
667 | } | |
668 | ||
669 | /* | |
670 | * This is just like xfs_ilock(), except that the caller | |
671 | * is guaranteed not to sleep. It returns 1 if it gets | |
672 | * the requested locks and 0 otherwise. If the IO lock is | |
673 | * obtained but the inode lock cannot be, then the IO lock | |
674 | * is dropped before returning. | |
675 | * | |
676 | * ip -- the inode being locked | |
677 | * lock_flags -- this parameter indicates the inode's locks to be | |
678 | * to be locked. See the comment for xfs_ilock() for a list | |
679 | * of valid values. | |
680 | */ | |
681 | int | |
682 | xfs_ilock_nowait( | |
683 | xfs_inode_t *ip, | |
684 | uint lock_flags) | |
685 | { | |
686 | /* | |
687 | * You can't set both SHARED and EXCL for the same lock, | |
688 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
689 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
690 | */ | |
691 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
692 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
693 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
694 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
695 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); | |
696 | ||
697 | if (lock_flags & XFS_IOLOCK_EXCL) { | |
698 | if (!mrtryupdate(&ip->i_iolock)) | |
699 | goto out; | |
700 | } else if (lock_flags & XFS_IOLOCK_SHARED) { | |
701 | if (!mrtryaccess(&ip->i_iolock)) | |
702 | goto out; | |
703 | } | |
704 | if (lock_flags & XFS_ILOCK_EXCL) { | |
705 | if (!mrtryupdate(&ip->i_lock)) | |
706 | goto out_undo_iolock; | |
707 | } else if (lock_flags & XFS_ILOCK_SHARED) { | |
708 | if (!mrtryaccess(&ip->i_lock)) | |
709 | goto out_undo_iolock; | |
710 | } | |
711 | xfs_ilock_trace(ip, 2, lock_flags, (inst_t *)__return_address); | |
712 | return 1; | |
713 | ||
714 | out_undo_iolock: | |
715 | if (lock_flags & XFS_IOLOCK_EXCL) | |
716 | mrunlock_excl(&ip->i_iolock); | |
717 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
718 | mrunlock_shared(&ip->i_iolock); | |
719 | out: | |
720 | return 0; | |
721 | } | |
722 | ||
723 | /* | |
724 | * xfs_iunlock() is used to drop the inode locks acquired with | |
725 | * xfs_ilock() and xfs_ilock_nowait(). The caller must pass | |
726 | * in the flags given to xfs_ilock() or xfs_ilock_nowait() so | |
727 | * that we know which locks to drop. | |
728 | * | |
729 | * ip -- the inode being unlocked | |
730 | * lock_flags -- this parameter indicates the inode's locks to be | |
731 | * to be unlocked. See the comment for xfs_ilock() for a list | |
732 | * of valid values for this parameter. | |
733 | * | |
734 | */ | |
735 | void | |
736 | xfs_iunlock( | |
737 | xfs_inode_t *ip, | |
738 | uint lock_flags) | |
739 | { | |
740 | /* | |
741 | * You can't set both SHARED and EXCL for the same lock, | |
742 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
743 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
744 | */ | |
745 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
746 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
747 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
748 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
749 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_IUNLOCK_NONOTIFY | | |
750 | XFS_LOCK_DEP_MASK)) == 0); | |
751 | ASSERT(lock_flags != 0); | |
752 | ||
753 | if (lock_flags & XFS_IOLOCK_EXCL) | |
754 | mrunlock_excl(&ip->i_iolock); | |
755 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
756 | mrunlock_shared(&ip->i_iolock); | |
757 | ||
758 | if (lock_flags & XFS_ILOCK_EXCL) | |
759 | mrunlock_excl(&ip->i_lock); | |
760 | else if (lock_flags & XFS_ILOCK_SHARED) | |
761 | mrunlock_shared(&ip->i_lock); | |
762 | ||
763 | if ((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) && | |
764 | !(lock_flags & XFS_IUNLOCK_NONOTIFY) && ip->i_itemp) { | |
765 | /* | |
766 | * Let the AIL know that this item has been unlocked in case | |
767 | * it is in the AIL and anyone is waiting on it. Don't do | |
768 | * this if the caller has asked us not to. | |
769 | */ | |
770 | xfs_trans_unlocked_item(ip->i_itemp->ili_item.li_ailp, | |
771 | (xfs_log_item_t*)(ip->i_itemp)); | |
772 | } | |
773 | xfs_ilock_trace(ip, 3, lock_flags, (inst_t *)__return_address); | |
774 | } | |
775 | ||
776 | /* | |
777 | * give up write locks. the i/o lock cannot be held nested | |
778 | * if it is being demoted. | |
779 | */ | |
780 | void | |
781 | xfs_ilock_demote( | |
782 | xfs_inode_t *ip, | |
783 | uint lock_flags) | |
784 | { | |
785 | ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)); | |
786 | ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0); | |
787 | ||
788 | if (lock_flags & XFS_ILOCK_EXCL) | |
789 | mrdemote(&ip->i_lock); | |
790 | if (lock_flags & XFS_IOLOCK_EXCL) | |
791 | mrdemote(&ip->i_iolock); | |
792 | } | |
793 | ||
794 | #ifdef DEBUG | |
795 | /* | |
796 | * Debug-only routine, without additional rw_semaphore APIs, we can | |
797 | * now only answer requests regarding whether we hold the lock for write | |
798 | * (reader state is outside our visibility, we only track writer state). | |
799 | * | |
800 | * Note: this means !xfs_isilocked would give false positives, so don't do that. | |
801 | */ | |
802 | int | |
803 | xfs_isilocked( | |
804 | xfs_inode_t *ip, | |
805 | uint lock_flags) | |
806 | { | |
807 | if ((lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) == | |
808 | XFS_ILOCK_EXCL) { | |
809 | if (!ip->i_lock.mr_writer) | |
810 | return 0; | |
811 | } | |
812 | ||
813 | if ((lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) == | |
814 | XFS_IOLOCK_EXCL) { | |
815 | if (!ip->i_iolock.mr_writer) | |
816 | return 0; | |
817 | } | |
818 | ||
819 | return 1; | |
820 | } | |
821 | #endif | |
822 | ||
823 | #ifdef XFS_INODE_TRACE | |
824 | ||
825 | #define KTRACE_ENTER(ip, vk, s, line, ra) \ | |
826 | ktrace_enter((ip)->i_trace, \ | |
827 | /* 0 */ (void *)(__psint_t)(vk), \ | |
828 | /* 1 */ (void *)(s), \ | |
829 | /* 2 */ (void *)(__psint_t) line, \ | |
830 | /* 3 */ (void *)(__psint_t)atomic_read(&VFS_I(ip)->i_count), \ | |
831 | /* 4 */ (void *)(ra), \ | |
832 | /* 5 */ NULL, \ | |
833 | /* 6 */ (void *)(__psint_t)current_cpu(), \ | |
834 | /* 7 */ (void *)(__psint_t)current_pid(), \ | |
835 | /* 8 */ (void *)__return_address, \ | |
836 | /* 9 */ NULL, NULL, NULL, NULL, NULL, NULL, NULL) | |
837 | ||
838 | /* | |
839 | * Vnode tracing code. | |
840 | */ | |
841 | void | |
842 | _xfs_itrace_entry(xfs_inode_t *ip, const char *func, inst_t *ra) | |
843 | { | |
844 | KTRACE_ENTER(ip, INODE_KTRACE_ENTRY, func, 0, ra); | |
845 | } | |
846 | ||
847 | void | |
848 | _xfs_itrace_exit(xfs_inode_t *ip, const char *func, inst_t *ra) | |
849 | { | |
850 | KTRACE_ENTER(ip, INODE_KTRACE_EXIT, func, 0, ra); | |
851 | } | |
852 | ||
853 | void | |
854 | xfs_itrace_hold(xfs_inode_t *ip, char *file, int line, inst_t *ra) | |
855 | { | |
856 | KTRACE_ENTER(ip, INODE_KTRACE_HOLD, file, line, ra); | |
857 | } | |
858 | ||
859 | void | |
860 | _xfs_itrace_ref(xfs_inode_t *ip, char *file, int line, inst_t *ra) | |
861 | { | |
862 | KTRACE_ENTER(ip, INODE_KTRACE_REF, file, line, ra); | |
863 | } | |
864 | ||
865 | void | |
866 | xfs_itrace_rele(xfs_inode_t *ip, char *file, int line, inst_t *ra) | |
867 | { | |
868 | KTRACE_ENTER(ip, INODE_KTRACE_RELE, file, line, ra); | |
869 | } | |
870 | #endif /* XFS_INODE_TRACE */ |