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" |
ef14f0c1 | 21 | #include "xfs_acl.h" |
a844f451 | 22 | #include "xfs_bit.h" |
1da177e4 | 23 | #include "xfs_log.h" |
a844f451 | 24 | #include "xfs_inum.h" |
1da177e4 LT |
25 | #include "xfs_trans.h" |
26 | #include "xfs_sb.h" | |
27 | #include "xfs_ag.h" | |
1da177e4 LT |
28 | #include "xfs_dir2.h" |
29 | #include "xfs_dmapi.h" | |
30 | #include "xfs_mount.h" | |
1da177e4 | 31 | #include "xfs_bmap_btree.h" |
a844f451 | 32 | #include "xfs_alloc_btree.h" |
1da177e4 | 33 | #include "xfs_ialloc_btree.h" |
1da177e4 | 34 | #include "xfs_dir2_sf.h" |
a844f451 | 35 | #include "xfs_attr_sf.h" |
1da177e4 LT |
36 | #include "xfs_dinode.h" |
37 | #include "xfs_inode.h" | |
a844f451 NS |
38 | #include "xfs_btree.h" |
39 | #include "xfs_ialloc.h" | |
1da177e4 LT |
40 | #include "xfs_quota.h" |
41 | #include "xfs_utils.h" | |
783a2f65 DC |
42 | #include "xfs_trans_priv.h" |
43 | #include "xfs_inode_item.h" | |
24f211ba CH |
44 | #include "xfs_bmap.h" |
45 | #include "xfs_btree_trace.h" | |
0b1b213f | 46 | #include "xfs_trace.h" |
24f211ba CH |
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; | |
54e34621 CH |
67 | if (inode_init_always(mp->m_super, VFS_I(ip))) { |
68 | kmem_zone_free(xfs_inode_zone, ip); | |
69 | return NULL; | |
70 | } | |
24f211ba CH |
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)); | |
033da48f CH |
76 | ASSERT(!rwsem_is_locked(&ip->i_iolock.mr_lock)); |
77 | ||
78 | mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino); | |
24f211ba | 79 | |
24f211ba CH |
80 | /* initialise the xfs inode */ |
81 | ip->i_ino = ino; | |
82 | ip->i_mount = mp; | |
83 | memset(&ip->i_imap, 0, sizeof(struct xfs_imap)); | |
84 | ip->i_afp = NULL; | |
85 | memset(&ip->i_df, 0, sizeof(xfs_ifork_t)); | |
86 | ip->i_flags = 0; | |
87 | ip->i_update_core = 0; | |
24f211ba CH |
88 | ip->i_delayed_blks = 0; |
89 | memset(&ip->i_d, 0, sizeof(xfs_icdinode_t)); | |
90 | ip->i_size = 0; | |
91 | ip->i_new_size = 0; | |
92 | ||
705db3fd | 93 | /* prevent anyone from using this yet */ |
eaff8079 | 94 | VFS_I(ip)->i_state = I_NEW; |
24f211ba CH |
95 | |
96 | return ip; | |
97 | } | |
1da177e4 | 98 | |
b36ec042 CH |
99 | STATIC void |
100 | xfs_inode_free( | |
101 | struct xfs_inode *ip) | |
102 | { | |
103 | switch (ip->i_d.di_mode & S_IFMT) { | |
104 | case S_IFREG: | |
105 | case S_IFDIR: | |
106 | case S_IFLNK: | |
107 | xfs_idestroy_fork(ip, XFS_DATA_FORK); | |
108 | break; | |
109 | } | |
110 | ||
111 | if (ip->i_afp) | |
112 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
113 | ||
b36ec042 CH |
114 | if (ip->i_itemp) { |
115 | /* | |
116 | * Only if we are shutting down the fs will we see an | |
117 | * inode still in the AIL. If it is there, we should remove | |
118 | * it to prevent a use-after-free from occurring. | |
119 | */ | |
120 | xfs_log_item_t *lip = &ip->i_itemp->ili_item; | |
121 | struct xfs_ail *ailp = lip->li_ailp; | |
122 | ||
123 | ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) || | |
124 | XFS_FORCED_SHUTDOWN(ip->i_mount)); | |
125 | if (lip->li_flags & XFS_LI_IN_AIL) { | |
126 | spin_lock(&ailp->xa_lock); | |
127 | if (lip->li_flags & XFS_LI_IN_AIL) | |
128 | xfs_trans_ail_delete(ailp, lip); | |
129 | else | |
130 | spin_unlock(&ailp->xa_lock); | |
131 | } | |
132 | xfs_inode_item_destroy(ip); | |
133 | ip->i_itemp = NULL; | |
134 | } | |
135 | ||
136 | /* asserts to verify all state is correct here */ | |
137 | ASSERT(atomic_read(&ip->i_iocount) == 0); | |
138 | ASSERT(atomic_read(&ip->i_pincount) == 0); | |
139 | ASSERT(!spin_is_locked(&ip->i_flags_lock)); | |
140 | ASSERT(completion_done(&ip->i_flush)); | |
141 | ||
142 | kmem_zone_free(xfs_inode_zone, ip); | |
143 | } | |
144 | ||
1da177e4 | 145 | /* |
6441e549 | 146 | * Check the validity of the inode we just found it the cache |
1da177e4 | 147 | */ |
6441e549 DC |
148 | static int |
149 | xfs_iget_cache_hit( | |
6441e549 DC |
150 | struct xfs_perag *pag, |
151 | struct xfs_inode *ip, | |
152 | int flags, | |
153 | int lock_flags) __releases(pag->pag_ici_lock) | |
1da177e4 | 154 | { |
bc990f5c | 155 | struct inode *inode = VFS_I(ip); |
6441e549 | 156 | struct xfs_mount *mp = ip->i_mount; |
bc990f5c CH |
157 | int error; |
158 | ||
159 | spin_lock(&ip->i_flags_lock); | |
da353b0d | 160 | |
6441e549 | 161 | /* |
bc990f5c CH |
162 | * If we are racing with another cache hit that is currently |
163 | * instantiating this inode or currently recycling it out of | |
164 | * reclaimabe state, wait for the initialisation to complete | |
165 | * before continuing. | |
166 | * | |
167 | * XXX(hch): eventually we should do something equivalent to | |
168 | * wait_on_inode to wait for these flags to be cleared | |
169 | * instead of polling for it. | |
6441e549 | 170 | */ |
bc990f5c | 171 | if (ip->i_flags & (XFS_INEW|XFS_IRECLAIM)) { |
0b1b213f | 172 | trace_xfs_iget_skip(ip); |
6441e549 | 173 | XFS_STATS_INC(xs_ig_frecycle); |
bc990f5c | 174 | error = EAGAIN; |
6441e549 DC |
175 | goto out_error; |
176 | } | |
da353b0d | 177 | |
bc990f5c CH |
178 | /* |
179 | * If lookup is racing with unlink return an error immediately. | |
180 | */ | |
181 | if (ip->i_d.di_mode == 0 && !(flags & XFS_IGET_CREATE)) { | |
182 | error = ENOENT; | |
183 | goto out_error; | |
184 | } | |
bf904248 | 185 | |
bc990f5c CH |
186 | /* |
187 | * If IRECLAIMABLE is set, we've torn down the VFS inode already. | |
188 | * Need to carefully get it back into useable state. | |
189 | */ | |
190 | if (ip->i_flags & XFS_IRECLAIMABLE) { | |
0b1b213f | 191 | trace_xfs_iget_reclaim(ip); |
da353b0d | 192 | |
6441e549 | 193 | /* |
bc990f5c CH |
194 | * We need to set XFS_INEW atomically with clearing the |
195 | * reclaimable tag so that we do have an indicator of the | |
196 | * inode still being initialized. | |
6441e549 | 197 | */ |
bc990f5c CH |
198 | ip->i_flags |= XFS_INEW; |
199 | ip->i_flags &= ~XFS_IRECLAIMABLE; | |
200 | __xfs_inode_clear_reclaim_tag(mp, pag, ip); | |
201 | ||
202 | spin_unlock(&ip->i_flags_lock); | |
203 | read_unlock(&pag->pag_ici_lock); | |
204 | ||
205 | error = -inode_init_always(mp->m_super, inode); | |
206 | if (error) { | |
207 | /* | |
208 | * Re-initializing the inode failed, and we are in deep | |
209 | * trouble. Try to re-add it to the reclaim list. | |
210 | */ | |
211 | read_lock(&pag->pag_ici_lock); | |
212 | spin_lock(&ip->i_flags_lock); | |
213 | ||
214 | ip->i_flags &= ~XFS_INEW; | |
215 | ip->i_flags |= XFS_IRECLAIMABLE; | |
216 | __xfs_inode_set_reclaim_tag(pag, ip); | |
0b1b213f | 217 | trace_xfs_iget_reclaim(ip); |
6441e549 | 218 | goto out_error; |
da353b0d | 219 | } |
eaff8079 | 220 | inode->i_state = I_NEW; |
bc990f5c | 221 | } else { |
bf904248 | 222 | /* If the VFS inode is being torn down, pause and try again. */ |
bc990f5c CH |
223 | if (!igrab(inode)) { |
224 | error = EAGAIN; | |
225 | goto out_error; | |
226 | } | |
1da177e4 | 227 | |
bc990f5c CH |
228 | /* We've got a live one. */ |
229 | spin_unlock(&ip->i_flags_lock); | |
230 | read_unlock(&pag->pag_ici_lock); | |
6441e549 | 231 | } |
da353b0d | 232 | |
6441e549 DC |
233 | if (lock_flags != 0) |
234 | xfs_ilock(ip, lock_flags); | |
da353b0d | 235 | |
6441e549 | 236 | xfs_iflags_clear(ip, XFS_ISTALE); |
6441e549 | 237 | XFS_STATS_INC(xs_ig_found); |
0b1b213f CH |
238 | |
239 | trace_xfs_iget_found(ip); | |
6441e549 | 240 | return 0; |
1da177e4 | 241 | |
6441e549 | 242 | out_error: |
bc990f5c | 243 | spin_unlock(&ip->i_flags_lock); |
da353b0d | 244 | read_unlock(&pag->pag_ici_lock); |
6441e549 DC |
245 | return error; |
246 | } | |
247 | ||
248 | ||
249 | static int | |
250 | xfs_iget_cache_miss( | |
251 | struct xfs_mount *mp, | |
252 | struct xfs_perag *pag, | |
253 | xfs_trans_t *tp, | |
254 | xfs_ino_t ino, | |
255 | struct xfs_inode **ipp, | |
256 | xfs_daddr_t bno, | |
257 | int flags, | |
0c3dc2b0 | 258 | int lock_flags) |
6441e549 DC |
259 | { |
260 | struct xfs_inode *ip; | |
261 | int error; | |
262 | unsigned long first_index, mask; | |
263 | xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino); | |
1da177e4 | 264 | |
24f211ba CH |
265 | ip = xfs_inode_alloc(mp, ino); |
266 | if (!ip) | |
267 | return ENOMEM; | |
268 | ||
269 | error = xfs_iread(mp, tp, ip, bno, flags); | |
6441e549 | 270 | if (error) |
24f211ba | 271 | goto out_destroy; |
1da177e4 | 272 | |
0b1b213f | 273 | xfs_itrace_entry(ip); |
1da177e4 | 274 | |
745b1f47 | 275 | if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) { |
6441e549 DC |
276 | error = ENOENT; |
277 | goto out_destroy; | |
1da177e4 LT |
278 | } |
279 | ||
280 | /* | |
bad55843 | 281 | * Preload the radix tree so we can insert safely under the |
56e73ec4 DC |
282 | * write spinlock. Note that we cannot sleep inside the preload |
283 | * region. | |
1da177e4 | 284 | */ |
da353b0d | 285 | if (radix_tree_preload(GFP_KERNEL)) { |
6441e549 | 286 | error = EAGAIN; |
ed93ec39 CH |
287 | goto out_destroy; |
288 | } | |
289 | ||
290 | /* | |
291 | * Because the inode hasn't been added to the radix-tree yet it can't | |
292 | * be found by another thread, so we can do the non-sleeping lock here. | |
293 | */ | |
294 | if (lock_flags) { | |
295 | if (!xfs_ilock_nowait(ip, lock_flags)) | |
296 | BUG(); | |
da353b0d | 297 | } |
f338f903 | 298 | |
da353b0d DC |
299 | mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1); |
300 | first_index = agino & mask; | |
301 | write_lock(&pag->pag_ici_lock); | |
6441e549 DC |
302 | |
303 | /* insert the new inode */ | |
da353b0d DC |
304 | error = radix_tree_insert(&pag->pag_ici_root, agino, ip); |
305 | if (unlikely(error)) { | |
6441e549 | 306 | WARN_ON(error != -EEXIST); |
da353b0d | 307 | XFS_STATS_INC(xs_ig_dup); |
6441e549 | 308 | error = EAGAIN; |
56e73ec4 | 309 | goto out_preload_end; |
1da177e4 LT |
310 | } |
311 | ||
6441e549 | 312 | /* These values _must_ be set before releasing the radix tree lock! */ |
1da177e4 | 313 | ip->i_udquot = ip->i_gdquot = NULL; |
7a18c386 | 314 | xfs_iflags_set(ip, XFS_INEW); |
1da177e4 | 315 | |
da353b0d DC |
316 | write_unlock(&pag->pag_ici_lock); |
317 | radix_tree_preload_end(); | |
0b1b213f CH |
318 | |
319 | trace_xfs_iget_alloc(ip); | |
6441e549 DC |
320 | *ipp = ip; |
321 | return 0; | |
322 | ||
56e73ec4 | 323 | out_preload_end: |
6441e549 DC |
324 | write_unlock(&pag->pag_ici_lock); |
325 | radix_tree_preload_end(); | |
56e73ec4 DC |
326 | if (lock_flags) |
327 | xfs_iunlock(ip, lock_flags); | |
6441e549 | 328 | out_destroy: |
b36ec042 CH |
329 | __destroy_inode(VFS_I(ip)); |
330 | xfs_inode_free(ip); | |
6441e549 DC |
331 | return error; |
332 | } | |
333 | ||
334 | /* | |
335 | * Look up an inode by number in the given file system. | |
336 | * The inode is looked up in the cache held in each AG. | |
bf904248 DC |
337 | * If the inode is found in the cache, initialise the vfs inode |
338 | * if necessary. | |
6441e549 DC |
339 | * |
340 | * If it is not in core, read it in from the file system's device, | |
bf904248 | 341 | * add it to the cache and initialise the vfs inode. |
6441e549 DC |
342 | * |
343 | * The inode is locked according to the value of the lock_flags parameter. | |
344 | * This flag parameter indicates how and if the inode's IO lock and inode lock | |
345 | * should be taken. | |
346 | * | |
347 | * mp -- the mount point structure for the current file system. It points | |
348 | * to the inode hash table. | |
349 | * tp -- a pointer to the current transaction if there is one. This is | |
350 | * simply passed through to the xfs_iread() call. | |
351 | * ino -- the number of the inode desired. This is the unique identifier | |
352 | * within the file system for the inode being requested. | |
353 | * lock_flags -- flags indicating how to lock the inode. See the comment | |
354 | * for xfs_ilock() for a list of valid values. | |
355 | * bno -- the block number starting the buffer containing the inode, | |
356 | * if known (as by bulkstat), else 0. | |
357 | */ | |
bf904248 DC |
358 | int |
359 | xfs_iget( | |
6441e549 DC |
360 | xfs_mount_t *mp, |
361 | xfs_trans_t *tp, | |
362 | xfs_ino_t ino, | |
363 | uint flags, | |
364 | uint lock_flags, | |
365 | xfs_inode_t **ipp, | |
366 | xfs_daddr_t bno) | |
367 | { | |
368 | xfs_inode_t *ip; | |
369 | int error; | |
370 | xfs_perag_t *pag; | |
371 | xfs_agino_t agino; | |
372 | ||
373 | /* the radix tree exists only in inode capable AGs */ | |
374 | if (XFS_INO_TO_AGNO(mp, ino) >= mp->m_maxagi) | |
375 | return EINVAL; | |
376 | ||
377 | /* get the perag structure and ensure that it's inode capable */ | |
378 | pag = xfs_get_perag(mp, ino); | |
379 | if (!pag->pagi_inodeok) | |
380 | return EINVAL; | |
381 | ASSERT(pag->pag_ici_init); | |
382 | agino = XFS_INO_TO_AGINO(mp, ino); | |
383 | ||
384 | again: | |
385 | error = 0; | |
386 | read_lock(&pag->pag_ici_lock); | |
387 | ip = radix_tree_lookup(&pag->pag_ici_root, agino); | |
388 | ||
389 | if (ip) { | |
bf904248 | 390 | error = xfs_iget_cache_hit(pag, ip, flags, lock_flags); |
6441e549 DC |
391 | if (error) |
392 | goto out_error_or_again; | |
393 | } else { | |
394 | read_unlock(&pag->pag_ici_lock); | |
395 | XFS_STATS_INC(xs_ig_missed); | |
396 | ||
397 | error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip, bno, | |
398 | flags, lock_flags); | |
399 | if (error) | |
400 | goto out_error_or_again; | |
401 | } | |
da353b0d | 402 | xfs_put_perag(mp, pag); |
1da177e4 | 403 | |
1da177e4 LT |
404 | *ipp = ip; |
405 | ||
bf904248 DC |
406 | ASSERT(ip->i_df.if_ext_max == |
407 | XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t)); | |
1da177e4 LT |
408 | /* |
409 | * If we have a real type for an on-disk inode, we can set ops(&unlock) | |
410 | * now. If it's a new inode being created, xfs_ialloc will handle it. | |
411 | */ | |
bf904248 | 412 | if (xfs_iflags_test(ip, XFS_INEW) && ip->i_d.di_mode != 0) |
41be8bed | 413 | xfs_setup_inode(ip); |
1da177e4 | 414 | return 0; |
6441e549 DC |
415 | |
416 | out_error_or_again: | |
417 | if (error == EAGAIN) { | |
418 | delay(1); | |
419 | goto again; | |
420 | } | |
421 | xfs_put_perag(mp, pag); | |
422 | return error; | |
1da177e4 LT |
423 | } |
424 | ||
1da177e4 LT |
425 | /* |
426 | * Decrement reference count of an inode structure and unlock it. | |
427 | * | |
428 | * ip -- the inode being released | |
429 | * lock_flags -- this parameter indicates the inode's locks to be | |
430 | * to be released. See the comment on xfs_iunlock() for a list | |
431 | * of valid values. | |
432 | */ | |
433 | void | |
434 | xfs_iput(xfs_inode_t *ip, | |
435 | uint lock_flags) | |
436 | { | |
cf441eeb | 437 | xfs_itrace_entry(ip); |
1da177e4 | 438 | xfs_iunlock(ip, lock_flags); |
10090be2 | 439 | IRELE(ip); |
1da177e4 LT |
440 | } |
441 | ||
442 | /* | |
443 | * Special iput for brand-new inodes that are still locked | |
444 | */ | |
445 | void | |
01651646 DC |
446 | xfs_iput_new( |
447 | xfs_inode_t *ip, | |
448 | uint lock_flags) | |
1da177e4 | 449 | { |
01651646 | 450 | struct inode *inode = VFS_I(ip); |
1da177e4 | 451 | |
cf441eeb | 452 | xfs_itrace_entry(ip); |
1da177e4 LT |
453 | |
454 | if ((ip->i_d.di_mode == 0)) { | |
7a18c386 | 455 | ASSERT(!xfs_iflags_test(ip, XFS_IRECLAIMABLE)); |
10090be2 | 456 | make_bad_inode(inode); |
1da177e4 LT |
457 | } |
458 | if (inode->i_state & I_NEW) | |
459 | unlock_new_inode(inode); | |
460 | if (lock_flags) | |
461 | xfs_iunlock(ip, lock_flags); | |
10090be2 | 462 | IRELE(ip); |
1da177e4 LT |
463 | } |
464 | ||
1da177e4 | 465 | /* |
5cafdeb2 CH |
466 | * This is called free all the memory associated with an inode. |
467 | * It must free the inode itself and any buffers allocated for | |
468 | * if_extents/if_data and if_broot. It must also free the lock | |
469 | * associated with the inode. | |
470 | * | |
471 | * Note: because we don't initialise everything on reallocation out | |
472 | * of the zone, we must ensure we nullify everything correctly before | |
473 | * freeing the structure. | |
1da177e4 LT |
474 | */ |
475 | void | |
5cafdeb2 CH |
476 | xfs_ireclaim( |
477 | struct xfs_inode *ip) | |
1da177e4 | 478 | { |
5cafdeb2 CH |
479 | struct xfs_mount *mp = ip->i_mount; |
480 | struct xfs_perag *pag; | |
b44b1126 | 481 | xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino); |
1da177e4 | 482 | |
5cafdeb2 | 483 | XFS_STATS_INC(xs_ig_reclaims); |
1da177e4 LT |
484 | |
485 | /* | |
b44b1126 CH |
486 | * Remove the inode from the per-AG radix tree. |
487 | * | |
488 | * Because radix_tree_delete won't complain even if the item was never | |
489 | * added to the tree assert that it's been there before to catch | |
490 | * problems with the inode life time early on. | |
1da177e4 | 491 | */ |
5cafdeb2 CH |
492 | pag = xfs_get_perag(mp, ip->i_ino); |
493 | write_lock(&pag->pag_ici_lock); | |
b44b1126 CH |
494 | if (!radix_tree_delete(&pag->pag_ici_root, agino)) |
495 | ASSERT(0); | |
5cafdeb2 CH |
496 | write_unlock(&pag->pag_ici_lock); |
497 | xfs_put_perag(mp, pag); | |
1da177e4 LT |
498 | |
499 | /* | |
5cafdeb2 CH |
500 | * Here we do an (almost) spurious inode lock in order to coordinate |
501 | * with inode cache radix tree lookups. This is because the lookup | |
502 | * can reference the inodes in the cache without taking references. | |
503 | * | |
504 | * We make that OK here by ensuring that we wait until the inode is | |
505 | * unlocked after the lookup before we go ahead and free it. We get | |
506 | * both the ilock and the iolock because the code may need to drop the | |
507 | * ilock one but will still hold the iolock. | |
1da177e4 | 508 | */ |
5cafdeb2 | 509 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); |
7d095257 | 510 | xfs_qm_dqdetach(ip); |
439b8434 | 511 | xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); |
1da177e4 | 512 | |
b36ec042 | 513 | xfs_inode_free(ip); |
1da177e4 LT |
514 | } |
515 | ||
516 | /* | |
517 | * This is a wrapper routine around the xfs_ilock() routine | |
518 | * used to centralize some grungy code. It is used in places | |
519 | * that wish to lock the inode solely for reading the extents. | |
520 | * The reason these places can't just call xfs_ilock(SHARED) | |
521 | * is that the inode lock also guards to bringing in of the | |
522 | * extents from disk for a file in b-tree format. If the inode | |
523 | * is in b-tree format, then we need to lock the inode exclusively | |
524 | * until the extents are read in. Locking it exclusively all | |
525 | * the time would limit our parallelism unnecessarily, though. | |
526 | * What we do instead is check to see if the extents have been | |
527 | * read in yet, and only lock the inode exclusively if they | |
528 | * have not. | |
529 | * | |
530 | * The function returns a value which should be given to the | |
531 | * corresponding xfs_iunlock_map_shared(). This value is | |
532 | * the mode in which the lock was actually taken. | |
533 | */ | |
534 | uint | |
535 | xfs_ilock_map_shared( | |
536 | xfs_inode_t *ip) | |
537 | { | |
538 | uint lock_mode; | |
539 | ||
540 | if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) && | |
541 | ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) { | |
542 | lock_mode = XFS_ILOCK_EXCL; | |
543 | } else { | |
544 | lock_mode = XFS_ILOCK_SHARED; | |
545 | } | |
546 | ||
547 | xfs_ilock(ip, lock_mode); | |
548 | ||
549 | return lock_mode; | |
550 | } | |
551 | ||
552 | /* | |
553 | * This is simply the unlock routine to go with xfs_ilock_map_shared(). | |
554 | * All it does is call xfs_iunlock() with the given lock_mode. | |
555 | */ | |
556 | void | |
557 | xfs_iunlock_map_shared( | |
558 | xfs_inode_t *ip, | |
559 | unsigned int lock_mode) | |
560 | { | |
561 | xfs_iunlock(ip, lock_mode); | |
562 | } | |
563 | ||
564 | /* | |
565 | * The xfs inode contains 2 locks: a multi-reader lock called the | |
566 | * i_iolock and a multi-reader lock called the i_lock. This routine | |
567 | * allows either or both of the locks to be obtained. | |
568 | * | |
569 | * The 2 locks should always be ordered so that the IO lock is | |
570 | * obtained first in order to prevent deadlock. | |
571 | * | |
572 | * ip -- the inode being locked | |
573 | * lock_flags -- this parameter indicates the inode's locks | |
574 | * to be locked. It can be: | |
575 | * XFS_IOLOCK_SHARED, | |
576 | * XFS_IOLOCK_EXCL, | |
577 | * XFS_ILOCK_SHARED, | |
578 | * XFS_ILOCK_EXCL, | |
579 | * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED, | |
580 | * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL, | |
581 | * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED, | |
582 | * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL | |
583 | */ | |
584 | void | |
579aa9ca CH |
585 | xfs_ilock( |
586 | xfs_inode_t *ip, | |
587 | uint lock_flags) | |
1da177e4 LT |
588 | { |
589 | /* | |
590 | * You can't set both SHARED and EXCL for the same lock, | |
591 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
592 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
593 | */ | |
594 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
595 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
596 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
597 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
f7c66ce3 | 598 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); |
1da177e4 | 599 | |
579aa9ca | 600 | if (lock_flags & XFS_IOLOCK_EXCL) |
f7c66ce3 | 601 | mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); |
579aa9ca | 602 | else if (lock_flags & XFS_IOLOCK_SHARED) |
f7c66ce3 | 603 | mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); |
579aa9ca CH |
604 | |
605 | if (lock_flags & XFS_ILOCK_EXCL) | |
f7c66ce3 | 606 | mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); |
579aa9ca | 607 | else if (lock_flags & XFS_ILOCK_SHARED) |
f7c66ce3 | 608 | mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); |
579aa9ca | 609 | |
0b1b213f | 610 | trace_xfs_ilock(ip, lock_flags, _RET_IP_); |
1da177e4 LT |
611 | } |
612 | ||
613 | /* | |
614 | * This is just like xfs_ilock(), except that the caller | |
615 | * is guaranteed not to sleep. It returns 1 if it gets | |
616 | * the requested locks and 0 otherwise. If the IO lock is | |
617 | * obtained but the inode lock cannot be, then the IO lock | |
618 | * is dropped before returning. | |
619 | * | |
620 | * ip -- the inode being locked | |
621 | * lock_flags -- this parameter indicates the inode's locks to be | |
622 | * to be locked. See the comment for xfs_ilock() for a list | |
623 | * of valid values. | |
1da177e4 LT |
624 | */ |
625 | int | |
579aa9ca CH |
626 | xfs_ilock_nowait( |
627 | xfs_inode_t *ip, | |
628 | uint lock_flags) | |
1da177e4 | 629 | { |
1da177e4 LT |
630 | /* |
631 | * You can't set both SHARED and EXCL for the same lock, | |
632 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
633 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
634 | */ | |
635 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
636 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
637 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
638 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
f7c66ce3 | 639 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); |
1da177e4 | 640 | |
1da177e4 | 641 | if (lock_flags & XFS_IOLOCK_EXCL) { |
579aa9ca CH |
642 | if (!mrtryupdate(&ip->i_iolock)) |
643 | goto out; | |
1da177e4 | 644 | } else if (lock_flags & XFS_IOLOCK_SHARED) { |
579aa9ca CH |
645 | if (!mrtryaccess(&ip->i_iolock)) |
646 | goto out; | |
1da177e4 LT |
647 | } |
648 | if (lock_flags & XFS_ILOCK_EXCL) { | |
579aa9ca CH |
649 | if (!mrtryupdate(&ip->i_lock)) |
650 | goto out_undo_iolock; | |
1da177e4 | 651 | } else if (lock_flags & XFS_ILOCK_SHARED) { |
579aa9ca CH |
652 | if (!mrtryaccess(&ip->i_lock)) |
653 | goto out_undo_iolock; | |
1da177e4 | 654 | } |
0b1b213f | 655 | trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_); |
1da177e4 | 656 | return 1; |
579aa9ca CH |
657 | |
658 | out_undo_iolock: | |
659 | if (lock_flags & XFS_IOLOCK_EXCL) | |
660 | mrunlock_excl(&ip->i_iolock); | |
661 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
662 | mrunlock_shared(&ip->i_iolock); | |
663 | out: | |
664 | return 0; | |
1da177e4 LT |
665 | } |
666 | ||
667 | /* | |
668 | * xfs_iunlock() is used to drop the inode locks acquired with | |
669 | * xfs_ilock() and xfs_ilock_nowait(). The caller must pass | |
670 | * in the flags given to xfs_ilock() or xfs_ilock_nowait() so | |
671 | * that we know which locks to drop. | |
672 | * | |
673 | * ip -- the inode being unlocked | |
674 | * lock_flags -- this parameter indicates the inode's locks to be | |
675 | * to be unlocked. See the comment for xfs_ilock() for a list | |
676 | * of valid values for this parameter. | |
677 | * | |
678 | */ | |
679 | void | |
579aa9ca CH |
680 | xfs_iunlock( |
681 | xfs_inode_t *ip, | |
682 | uint lock_flags) | |
1da177e4 LT |
683 | { |
684 | /* | |
685 | * You can't set both SHARED and EXCL for the same lock, | |
686 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
687 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
688 | */ | |
689 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
690 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
691 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
692 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
f7c66ce3 LM |
693 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_IUNLOCK_NONOTIFY | |
694 | XFS_LOCK_DEP_MASK)) == 0); | |
1da177e4 LT |
695 | ASSERT(lock_flags != 0); |
696 | ||
579aa9ca CH |
697 | if (lock_flags & XFS_IOLOCK_EXCL) |
698 | mrunlock_excl(&ip->i_iolock); | |
699 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
700 | mrunlock_shared(&ip->i_iolock); | |
1da177e4 | 701 | |
579aa9ca CH |
702 | if (lock_flags & XFS_ILOCK_EXCL) |
703 | mrunlock_excl(&ip->i_lock); | |
704 | else if (lock_flags & XFS_ILOCK_SHARED) | |
705 | mrunlock_shared(&ip->i_lock); | |
1da177e4 | 706 | |
579aa9ca CH |
707 | if ((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) && |
708 | !(lock_flags & XFS_IUNLOCK_NONOTIFY) && ip->i_itemp) { | |
1da177e4 LT |
709 | /* |
710 | * Let the AIL know that this item has been unlocked in case | |
711 | * it is in the AIL and anyone is waiting on it. Don't do | |
712 | * this if the caller has asked us not to. | |
713 | */ | |
783a2f65 | 714 | xfs_trans_unlocked_item(ip->i_itemp->ili_item.li_ailp, |
579aa9ca | 715 | (xfs_log_item_t*)(ip->i_itemp)); |
1da177e4 | 716 | } |
0b1b213f | 717 | trace_xfs_iunlock(ip, lock_flags, _RET_IP_); |
1da177e4 LT |
718 | } |
719 | ||
720 | /* | |
721 | * give up write locks. the i/o lock cannot be held nested | |
722 | * if it is being demoted. | |
723 | */ | |
724 | void | |
579aa9ca CH |
725 | xfs_ilock_demote( |
726 | xfs_inode_t *ip, | |
727 | uint lock_flags) | |
1da177e4 LT |
728 | { |
729 | ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)); | |
730 | ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0); | |
731 | ||
579aa9ca | 732 | if (lock_flags & XFS_ILOCK_EXCL) |
1da177e4 | 733 | mrdemote(&ip->i_lock); |
579aa9ca | 734 | if (lock_flags & XFS_IOLOCK_EXCL) |
1da177e4 | 735 | mrdemote(&ip->i_iolock); |
0b1b213f CH |
736 | |
737 | trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_); | |
579aa9ca CH |
738 | } |
739 | ||
740 | #ifdef DEBUG | |
741 | /* | |
742 | * Debug-only routine, without additional rw_semaphore APIs, we can | |
743 | * now only answer requests regarding whether we hold the lock for write | |
744 | * (reader state is outside our visibility, we only track writer state). | |
745 | * | |
746 | * Note: this means !xfs_isilocked would give false positives, so don't do that. | |
747 | */ | |
748 | int | |
749 | xfs_isilocked( | |
750 | xfs_inode_t *ip, | |
751 | uint lock_flags) | |
752 | { | |
753 | if ((lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) == | |
754 | XFS_ILOCK_EXCL) { | |
755 | if (!ip->i_lock.mr_writer) | |
756 | return 0; | |
1da177e4 | 757 | } |
579aa9ca CH |
758 | |
759 | if ((lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) == | |
760 | XFS_IOLOCK_EXCL) { | |
761 | if (!ip->i_iolock.mr_writer) | |
762 | return 0; | |
763 | } | |
764 | ||
765 | return 1; | |
1da177e4 | 766 | } |
579aa9ca | 767 | #endif |