]>
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 | 28 | #include "xfs_mount.h" |
1da177e4 | 29 | #include "xfs_bmap_btree.h" |
a844f451 | 30 | #include "xfs_alloc_btree.h" |
1da177e4 | 31 | #include "xfs_ialloc_btree.h" |
1da177e4 LT |
32 | #include "xfs_dinode.h" |
33 | #include "xfs_inode.h" | |
a844f451 NS |
34 | #include "xfs_btree.h" |
35 | #include "xfs_ialloc.h" | |
1da177e4 LT |
36 | #include "xfs_quota.h" |
37 | #include "xfs_utils.h" | |
783a2f65 DC |
38 | #include "xfs_trans_priv.h" |
39 | #include "xfs_inode_item.h" | |
24f211ba CH |
40 | #include "xfs_bmap.h" |
41 | #include "xfs_btree_trace.h" | |
0b1b213f | 42 | #include "xfs_trace.h" |
24f211ba CH |
43 | |
44 | ||
45 | /* | |
46 | * Allocate and initialise an xfs_inode. | |
47 | */ | |
48 | STATIC struct xfs_inode * | |
49 | xfs_inode_alloc( | |
50 | struct xfs_mount *mp, | |
51 | xfs_ino_t ino) | |
52 | { | |
53 | struct xfs_inode *ip; | |
54 | ||
55 | /* | |
56 | * if this didn't occur in transactions, we could use | |
57 | * KM_MAYFAIL and return NULL here on ENOMEM. Set the | |
58 | * code up to do this anyway. | |
59 | */ | |
60 | ip = kmem_zone_alloc(xfs_inode_zone, KM_SLEEP); | |
61 | if (!ip) | |
62 | return NULL; | |
54e34621 CH |
63 | if (inode_init_always(mp->m_super, VFS_I(ip))) { |
64 | kmem_zone_free(xfs_inode_zone, ip); | |
65 | return NULL; | |
66 | } | |
24f211ba CH |
67 | |
68 | ASSERT(atomic_read(&ip->i_iocount) == 0); | |
69 | ASSERT(atomic_read(&ip->i_pincount) == 0); | |
70 | ASSERT(!spin_is_locked(&ip->i_flags_lock)); | |
71 | ASSERT(completion_done(&ip->i_flush)); | |
033da48f CH |
72 | |
73 | mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino); | |
24f211ba | 74 | |
24f211ba CH |
75 | /* initialise the xfs inode */ |
76 | ip->i_ino = ino; | |
77 | ip->i_mount = mp; | |
78 | memset(&ip->i_imap, 0, sizeof(struct xfs_imap)); | |
79 | ip->i_afp = NULL; | |
80 | memset(&ip->i_df, 0, sizeof(xfs_ifork_t)); | |
81 | ip->i_flags = 0; | |
82 | ip->i_update_core = 0; | |
24f211ba CH |
83 | ip->i_delayed_blks = 0; |
84 | memset(&ip->i_d, 0, sizeof(xfs_icdinode_t)); | |
85 | ip->i_size = 0; | |
86 | ip->i_new_size = 0; | |
87 | ||
705db3fd | 88 | /* prevent anyone from using this yet */ |
eaff8079 | 89 | VFS_I(ip)->i_state = I_NEW; |
24f211ba CH |
90 | |
91 | return ip; | |
92 | } | |
1da177e4 | 93 | |
b36ec042 CH |
94 | STATIC void |
95 | xfs_inode_free( | |
96 | struct xfs_inode *ip) | |
97 | { | |
98 | switch (ip->i_d.di_mode & S_IFMT) { | |
99 | case S_IFREG: | |
100 | case S_IFDIR: | |
101 | case S_IFLNK: | |
102 | xfs_idestroy_fork(ip, XFS_DATA_FORK); | |
103 | break; | |
104 | } | |
105 | ||
106 | if (ip->i_afp) | |
107 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
108 | ||
b36ec042 CH |
109 | if (ip->i_itemp) { |
110 | /* | |
111 | * Only if we are shutting down the fs will we see an | |
112 | * inode still in the AIL. If it is there, we should remove | |
113 | * it to prevent a use-after-free from occurring. | |
114 | */ | |
115 | xfs_log_item_t *lip = &ip->i_itemp->ili_item; | |
116 | struct xfs_ail *ailp = lip->li_ailp; | |
117 | ||
118 | ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) || | |
119 | XFS_FORCED_SHUTDOWN(ip->i_mount)); | |
120 | if (lip->li_flags & XFS_LI_IN_AIL) { | |
121 | spin_lock(&ailp->xa_lock); | |
122 | if (lip->li_flags & XFS_LI_IN_AIL) | |
123 | xfs_trans_ail_delete(ailp, lip); | |
124 | else | |
125 | spin_unlock(&ailp->xa_lock); | |
126 | } | |
127 | xfs_inode_item_destroy(ip); | |
128 | ip->i_itemp = NULL; | |
129 | } | |
130 | ||
131 | /* asserts to verify all state is correct here */ | |
132 | ASSERT(atomic_read(&ip->i_iocount) == 0); | |
133 | ASSERT(atomic_read(&ip->i_pincount) == 0); | |
134 | ASSERT(!spin_is_locked(&ip->i_flags_lock)); | |
135 | ASSERT(completion_done(&ip->i_flush)); | |
136 | ||
137 | kmem_zone_free(xfs_inode_zone, ip); | |
138 | } | |
139 | ||
1da177e4 | 140 | /* |
6441e549 | 141 | * Check the validity of the inode we just found it the cache |
1da177e4 | 142 | */ |
6441e549 DC |
143 | static int |
144 | xfs_iget_cache_hit( | |
6441e549 DC |
145 | struct xfs_perag *pag, |
146 | struct xfs_inode *ip, | |
147 | int flags, | |
148 | int lock_flags) __releases(pag->pag_ici_lock) | |
1da177e4 | 149 | { |
bc990f5c | 150 | struct inode *inode = VFS_I(ip); |
6441e549 | 151 | struct xfs_mount *mp = ip->i_mount; |
bc990f5c CH |
152 | int error; |
153 | ||
154 | spin_lock(&ip->i_flags_lock); | |
da353b0d | 155 | |
6441e549 | 156 | /* |
bc990f5c CH |
157 | * If we are racing with another cache hit that is currently |
158 | * instantiating this inode or currently recycling it out of | |
159 | * reclaimabe state, wait for the initialisation to complete | |
160 | * before continuing. | |
161 | * | |
162 | * XXX(hch): eventually we should do something equivalent to | |
163 | * wait_on_inode to wait for these flags to be cleared | |
164 | * instead of polling for it. | |
6441e549 | 165 | */ |
bc990f5c | 166 | if (ip->i_flags & (XFS_INEW|XFS_IRECLAIM)) { |
0b1b213f | 167 | trace_xfs_iget_skip(ip); |
6441e549 | 168 | XFS_STATS_INC(xs_ig_frecycle); |
bc990f5c | 169 | error = EAGAIN; |
6441e549 DC |
170 | goto out_error; |
171 | } | |
da353b0d | 172 | |
bc990f5c CH |
173 | /* |
174 | * If lookup is racing with unlink return an error immediately. | |
175 | */ | |
176 | if (ip->i_d.di_mode == 0 && !(flags & XFS_IGET_CREATE)) { | |
177 | error = ENOENT; | |
178 | goto out_error; | |
179 | } | |
bf904248 | 180 | |
bc990f5c CH |
181 | /* |
182 | * If IRECLAIMABLE is set, we've torn down the VFS inode already. | |
183 | * Need to carefully get it back into useable state. | |
184 | */ | |
185 | if (ip->i_flags & XFS_IRECLAIMABLE) { | |
0b1b213f | 186 | trace_xfs_iget_reclaim(ip); |
da353b0d | 187 | |
6441e549 | 188 | /* |
f1f724e4 CH |
189 | * We need to set XFS_IRECLAIM to prevent xfs_reclaim_inode |
190 | * from stomping over us while we recycle the inode. We can't | |
191 | * clear the radix tree reclaimable tag yet as it requires | |
192 | * pag_ici_lock to be held exclusive. | |
6441e549 | 193 | */ |
f1f724e4 | 194 | ip->i_flags |= XFS_IRECLAIM; |
bc990f5c CH |
195 | |
196 | spin_unlock(&ip->i_flags_lock); | |
197 | read_unlock(&pag->pag_ici_lock); | |
198 | ||
199 | error = -inode_init_always(mp->m_super, inode); | |
200 | if (error) { | |
201 | /* | |
202 | * Re-initializing the inode failed, and we are in deep | |
203 | * trouble. Try to re-add it to the reclaim list. | |
204 | */ | |
205 | read_lock(&pag->pag_ici_lock); | |
206 | spin_lock(&ip->i_flags_lock); | |
207 | ||
208 | ip->i_flags &= ~XFS_INEW; | |
209 | ip->i_flags |= XFS_IRECLAIMABLE; | |
210 | __xfs_inode_set_reclaim_tag(pag, ip); | |
d2e078c3 | 211 | trace_xfs_iget_reclaim_fail(ip); |
6441e549 | 212 | goto out_error; |
da353b0d | 213 | } |
f1f724e4 CH |
214 | |
215 | write_lock(&pag->pag_ici_lock); | |
216 | spin_lock(&ip->i_flags_lock); | |
217 | ip->i_flags &= ~(XFS_IRECLAIMABLE | XFS_IRECLAIM); | |
218 | ip->i_flags |= XFS_INEW; | |
219 | __xfs_inode_clear_reclaim_tag(mp, pag, ip); | |
eaff8079 | 220 | inode->i_state = I_NEW; |
f1f724e4 CH |
221 | spin_unlock(&ip->i_flags_lock); |
222 | write_unlock(&pag->pag_ici_lock); | |
bc990f5c | 223 | } else { |
bf904248 | 224 | /* If the VFS inode is being torn down, pause and try again. */ |
bc990f5c | 225 | if (!igrab(inode)) { |
d2e078c3 | 226 | trace_xfs_iget_skip(ip); |
bc990f5c CH |
227 | error = EAGAIN; |
228 | goto out_error; | |
229 | } | |
1da177e4 | 230 | |
bc990f5c CH |
231 | /* We've got a live one. */ |
232 | spin_unlock(&ip->i_flags_lock); | |
233 | read_unlock(&pag->pag_ici_lock); | |
d2e078c3 | 234 | trace_xfs_iget_hit(ip); |
6441e549 | 235 | } |
da353b0d | 236 | |
6441e549 DC |
237 | if (lock_flags != 0) |
238 | xfs_ilock(ip, lock_flags); | |
da353b0d | 239 | |
6441e549 | 240 | xfs_iflags_clear(ip, XFS_ISTALE); |
6441e549 | 241 | XFS_STATS_INC(xs_ig_found); |
0b1b213f | 242 | |
6441e549 | 243 | return 0; |
1da177e4 | 244 | |
6441e549 | 245 | out_error: |
bc990f5c | 246 | spin_unlock(&ip->i_flags_lock); |
da353b0d | 247 | read_unlock(&pag->pag_ici_lock); |
6441e549 DC |
248 | return error; |
249 | } | |
250 | ||
251 | ||
252 | static int | |
253 | xfs_iget_cache_miss( | |
254 | struct xfs_mount *mp, | |
255 | struct xfs_perag *pag, | |
256 | xfs_trans_t *tp, | |
257 | xfs_ino_t ino, | |
258 | struct xfs_inode **ipp, | |
6441e549 | 259 | int flags, |
0c3dc2b0 | 260 | int lock_flags) |
6441e549 DC |
261 | { |
262 | struct xfs_inode *ip; | |
263 | int error; | |
264 | unsigned long first_index, mask; | |
265 | xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino); | |
1da177e4 | 266 | |
24f211ba CH |
267 | ip = xfs_inode_alloc(mp, ino); |
268 | if (!ip) | |
269 | return ENOMEM; | |
270 | ||
7b6259e7 | 271 | error = xfs_iread(mp, tp, ip, flags); |
6441e549 | 272 | if (error) |
24f211ba | 273 | goto out_destroy; |
1da177e4 | 274 | |
d2e078c3 | 275 | trace_xfs_iget_miss(ip); |
1da177e4 | 276 | |
745b1f47 | 277 | if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) { |
6441e549 DC |
278 | error = ENOENT; |
279 | goto out_destroy; | |
1da177e4 LT |
280 | } |
281 | ||
282 | /* | |
bad55843 | 283 | * Preload the radix tree so we can insert safely under the |
56e73ec4 DC |
284 | * write spinlock. Note that we cannot sleep inside the preload |
285 | * region. | |
1da177e4 | 286 | */ |
da353b0d | 287 | if (radix_tree_preload(GFP_KERNEL)) { |
6441e549 | 288 | error = EAGAIN; |
ed93ec39 CH |
289 | goto out_destroy; |
290 | } | |
291 | ||
292 | /* | |
293 | * Because the inode hasn't been added to the radix-tree yet it can't | |
294 | * be found by another thread, so we can do the non-sleeping lock here. | |
295 | */ | |
296 | if (lock_flags) { | |
297 | if (!xfs_ilock_nowait(ip, lock_flags)) | |
298 | BUG(); | |
da353b0d | 299 | } |
f338f903 | 300 | |
da353b0d DC |
301 | mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1); |
302 | first_index = agino & mask; | |
303 | write_lock(&pag->pag_ici_lock); | |
6441e549 DC |
304 | |
305 | /* insert the new inode */ | |
da353b0d DC |
306 | error = radix_tree_insert(&pag->pag_ici_root, agino, ip); |
307 | if (unlikely(error)) { | |
6441e549 | 308 | WARN_ON(error != -EEXIST); |
da353b0d | 309 | XFS_STATS_INC(xs_ig_dup); |
6441e549 | 310 | error = EAGAIN; |
56e73ec4 | 311 | goto out_preload_end; |
1da177e4 LT |
312 | } |
313 | ||
6441e549 | 314 | /* These values _must_ be set before releasing the radix tree lock! */ |
1da177e4 | 315 | ip->i_udquot = ip->i_gdquot = NULL; |
7a18c386 | 316 | xfs_iflags_set(ip, XFS_INEW); |
1da177e4 | 317 | |
da353b0d DC |
318 | write_unlock(&pag->pag_ici_lock); |
319 | radix_tree_preload_end(); | |
0b1b213f | 320 | |
6441e549 DC |
321 | *ipp = ip; |
322 | return 0; | |
323 | ||
56e73ec4 | 324 | out_preload_end: |
6441e549 DC |
325 | write_unlock(&pag->pag_ici_lock); |
326 | radix_tree_preload_end(); | |
56e73ec4 DC |
327 | if (lock_flags) |
328 | xfs_iunlock(ip, lock_flags); | |
6441e549 | 329 | out_destroy: |
b36ec042 CH |
330 | __destroy_inode(VFS_I(ip)); |
331 | xfs_inode_free(ip); | |
6441e549 DC |
332 | return error; |
333 | } | |
334 | ||
335 | /* | |
336 | * Look up an inode by number in the given file system. | |
337 | * The inode is looked up in the cache held in each AG. | |
bf904248 DC |
338 | * If the inode is found in the cache, initialise the vfs inode |
339 | * if necessary. | |
6441e549 DC |
340 | * |
341 | * If it is not in core, read it in from the file system's device, | |
bf904248 | 342 | * add it to the cache and initialise the vfs inode. |
6441e549 DC |
343 | * |
344 | * The inode is locked according to the value of the lock_flags parameter. | |
345 | * This flag parameter indicates how and if the inode's IO lock and inode lock | |
346 | * should be taken. | |
347 | * | |
348 | * mp -- the mount point structure for the current file system. It points | |
349 | * to the inode hash table. | |
350 | * tp -- a pointer to the current transaction if there is one. This is | |
351 | * simply passed through to the xfs_iread() call. | |
352 | * ino -- the number of the inode desired. This is the unique identifier | |
353 | * within the file system for the inode being requested. | |
354 | * lock_flags -- flags indicating how to lock the inode. See the comment | |
355 | * for xfs_ilock() for a list of valid values. | |
6441e549 | 356 | */ |
bf904248 DC |
357 | int |
358 | xfs_iget( | |
6441e549 DC |
359 | xfs_mount_t *mp, |
360 | xfs_trans_t *tp, | |
361 | xfs_ino_t ino, | |
362 | uint flags, | |
363 | uint lock_flags, | |
7b6259e7 | 364 | xfs_inode_t **ipp) |
6441e549 DC |
365 | { |
366 | xfs_inode_t *ip; | |
367 | int error; | |
368 | xfs_perag_t *pag; | |
369 | xfs_agino_t agino; | |
370 | ||
371 | /* the radix tree exists only in inode capable AGs */ | |
372 | if (XFS_INO_TO_AGNO(mp, ino) >= mp->m_maxagi) | |
373 | return EINVAL; | |
374 | ||
375 | /* get the perag structure and ensure that it's inode capable */ | |
5017e97d | 376 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ino)); |
6441e549 DC |
377 | agino = XFS_INO_TO_AGINO(mp, ino); |
378 | ||
379 | again: | |
380 | error = 0; | |
381 | read_lock(&pag->pag_ici_lock); | |
382 | ip = radix_tree_lookup(&pag->pag_ici_root, agino); | |
383 | ||
384 | if (ip) { | |
bf904248 | 385 | error = xfs_iget_cache_hit(pag, ip, flags, lock_flags); |
6441e549 DC |
386 | if (error) |
387 | goto out_error_or_again; | |
388 | } else { | |
389 | read_unlock(&pag->pag_ici_lock); | |
390 | XFS_STATS_INC(xs_ig_missed); | |
391 | ||
7b6259e7 | 392 | error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip, |
6441e549 DC |
393 | flags, lock_flags); |
394 | if (error) | |
395 | goto out_error_or_again; | |
396 | } | |
5017e97d | 397 | xfs_perag_put(pag); |
1da177e4 | 398 | |
1da177e4 LT |
399 | *ipp = ip; |
400 | ||
bf904248 DC |
401 | ASSERT(ip->i_df.if_ext_max == |
402 | XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t)); | |
1da177e4 LT |
403 | /* |
404 | * If we have a real type for an on-disk inode, we can set ops(&unlock) | |
405 | * now. If it's a new inode being created, xfs_ialloc will handle it. | |
406 | */ | |
bf904248 | 407 | if (xfs_iflags_test(ip, XFS_INEW) && ip->i_d.di_mode != 0) |
41be8bed | 408 | xfs_setup_inode(ip); |
1da177e4 | 409 | return 0; |
6441e549 DC |
410 | |
411 | out_error_or_again: | |
412 | if (error == EAGAIN) { | |
413 | delay(1); | |
414 | goto again; | |
415 | } | |
5017e97d | 416 | xfs_perag_put(pag); |
6441e549 | 417 | return error; |
1da177e4 LT |
418 | } |
419 | ||
1da177e4 | 420 | /* |
5cafdeb2 CH |
421 | * This is called free all the memory associated with an inode. |
422 | * It must free the inode itself and any buffers allocated for | |
423 | * if_extents/if_data and if_broot. It must also free the lock | |
424 | * associated with the inode. | |
425 | * | |
426 | * Note: because we don't initialise everything on reallocation out | |
427 | * of the zone, we must ensure we nullify everything correctly before | |
428 | * freeing the structure. | |
1da177e4 LT |
429 | */ |
430 | void | |
5cafdeb2 CH |
431 | xfs_ireclaim( |
432 | struct xfs_inode *ip) | |
1da177e4 | 433 | { |
5cafdeb2 CH |
434 | struct xfs_mount *mp = ip->i_mount; |
435 | struct xfs_perag *pag; | |
b44b1126 | 436 | xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino); |
1da177e4 | 437 | |
5cafdeb2 | 438 | XFS_STATS_INC(xs_ig_reclaims); |
1da177e4 LT |
439 | |
440 | /* | |
b44b1126 CH |
441 | * Remove the inode from the per-AG radix tree. |
442 | * | |
443 | * Because radix_tree_delete won't complain even if the item was never | |
444 | * added to the tree assert that it's been there before to catch | |
445 | * problems with the inode life time early on. | |
1da177e4 | 446 | */ |
5017e97d | 447 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
5cafdeb2 | 448 | write_lock(&pag->pag_ici_lock); |
b44b1126 CH |
449 | if (!radix_tree_delete(&pag->pag_ici_root, agino)) |
450 | ASSERT(0); | |
5cafdeb2 | 451 | write_unlock(&pag->pag_ici_lock); |
5017e97d | 452 | xfs_perag_put(pag); |
1da177e4 LT |
453 | |
454 | /* | |
5cafdeb2 CH |
455 | * Here we do an (almost) spurious inode lock in order to coordinate |
456 | * with inode cache radix tree lookups. This is because the lookup | |
457 | * can reference the inodes in the cache without taking references. | |
458 | * | |
459 | * We make that OK here by ensuring that we wait until the inode is | |
460 | * unlocked after the lookup before we go ahead and free it. We get | |
461 | * both the ilock and the iolock because the code may need to drop the | |
462 | * ilock one but will still hold the iolock. | |
1da177e4 | 463 | */ |
5cafdeb2 | 464 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); |
7d095257 | 465 | xfs_qm_dqdetach(ip); |
439b8434 | 466 | xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); |
1da177e4 | 467 | |
b36ec042 | 468 | xfs_inode_free(ip); |
1da177e4 LT |
469 | } |
470 | ||
471 | /* | |
472 | * This is a wrapper routine around the xfs_ilock() routine | |
473 | * used to centralize some grungy code. It is used in places | |
474 | * that wish to lock the inode solely for reading the extents. | |
475 | * The reason these places can't just call xfs_ilock(SHARED) | |
476 | * is that the inode lock also guards to bringing in of the | |
477 | * extents from disk for a file in b-tree format. If the inode | |
478 | * is in b-tree format, then we need to lock the inode exclusively | |
479 | * until the extents are read in. Locking it exclusively all | |
480 | * the time would limit our parallelism unnecessarily, though. | |
481 | * What we do instead is check to see if the extents have been | |
482 | * read in yet, and only lock the inode exclusively if they | |
483 | * have not. | |
484 | * | |
485 | * The function returns a value which should be given to the | |
486 | * corresponding xfs_iunlock_map_shared(). This value is | |
487 | * the mode in which the lock was actually taken. | |
488 | */ | |
489 | uint | |
490 | xfs_ilock_map_shared( | |
491 | xfs_inode_t *ip) | |
492 | { | |
493 | uint lock_mode; | |
494 | ||
495 | if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) && | |
496 | ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) { | |
497 | lock_mode = XFS_ILOCK_EXCL; | |
498 | } else { | |
499 | lock_mode = XFS_ILOCK_SHARED; | |
500 | } | |
501 | ||
502 | xfs_ilock(ip, lock_mode); | |
503 | ||
504 | return lock_mode; | |
505 | } | |
506 | ||
507 | /* | |
508 | * This is simply the unlock routine to go with xfs_ilock_map_shared(). | |
509 | * All it does is call xfs_iunlock() with the given lock_mode. | |
510 | */ | |
511 | void | |
512 | xfs_iunlock_map_shared( | |
513 | xfs_inode_t *ip, | |
514 | unsigned int lock_mode) | |
515 | { | |
516 | xfs_iunlock(ip, lock_mode); | |
517 | } | |
518 | ||
519 | /* | |
520 | * The xfs inode contains 2 locks: a multi-reader lock called the | |
521 | * i_iolock and a multi-reader lock called the i_lock. This routine | |
522 | * allows either or both of the locks to be obtained. | |
523 | * | |
524 | * The 2 locks should always be ordered so that the IO lock is | |
525 | * obtained first in order to prevent deadlock. | |
526 | * | |
527 | * ip -- the inode being locked | |
528 | * lock_flags -- this parameter indicates the inode's locks | |
529 | * to be locked. It can be: | |
530 | * XFS_IOLOCK_SHARED, | |
531 | * XFS_IOLOCK_EXCL, | |
532 | * XFS_ILOCK_SHARED, | |
533 | * XFS_ILOCK_EXCL, | |
534 | * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED, | |
535 | * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL, | |
536 | * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED, | |
537 | * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL | |
538 | */ | |
539 | void | |
579aa9ca CH |
540 | xfs_ilock( |
541 | xfs_inode_t *ip, | |
542 | uint lock_flags) | |
1da177e4 LT |
543 | { |
544 | /* | |
545 | * You can't set both SHARED and EXCL for the same lock, | |
546 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
547 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
548 | */ | |
549 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
550 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
551 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
552 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
f7c66ce3 | 553 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); |
1da177e4 | 554 | |
579aa9ca | 555 | if (lock_flags & XFS_IOLOCK_EXCL) |
f7c66ce3 | 556 | mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); |
579aa9ca | 557 | else if (lock_flags & XFS_IOLOCK_SHARED) |
f7c66ce3 | 558 | mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); |
579aa9ca CH |
559 | |
560 | if (lock_flags & XFS_ILOCK_EXCL) | |
f7c66ce3 | 561 | mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); |
579aa9ca | 562 | else if (lock_flags & XFS_ILOCK_SHARED) |
f7c66ce3 | 563 | mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); |
579aa9ca | 564 | |
0b1b213f | 565 | trace_xfs_ilock(ip, lock_flags, _RET_IP_); |
1da177e4 LT |
566 | } |
567 | ||
568 | /* | |
569 | * This is just like xfs_ilock(), except that the caller | |
570 | * is guaranteed not to sleep. It returns 1 if it gets | |
571 | * the requested locks and 0 otherwise. If the IO lock is | |
572 | * obtained but the inode lock cannot be, then the IO lock | |
573 | * is dropped before returning. | |
574 | * | |
575 | * ip -- the inode being locked | |
576 | * lock_flags -- this parameter indicates the inode's locks to be | |
577 | * to be locked. See the comment for xfs_ilock() for a list | |
578 | * of valid values. | |
1da177e4 LT |
579 | */ |
580 | int | |
579aa9ca CH |
581 | xfs_ilock_nowait( |
582 | xfs_inode_t *ip, | |
583 | uint lock_flags) | |
1da177e4 | 584 | { |
1da177e4 LT |
585 | /* |
586 | * You can't set both SHARED and EXCL for the same lock, | |
587 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
588 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
589 | */ | |
590 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
591 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
592 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
593 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
f7c66ce3 | 594 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); |
1da177e4 | 595 | |
1da177e4 | 596 | if (lock_flags & XFS_IOLOCK_EXCL) { |
579aa9ca CH |
597 | if (!mrtryupdate(&ip->i_iolock)) |
598 | goto out; | |
1da177e4 | 599 | } else if (lock_flags & XFS_IOLOCK_SHARED) { |
579aa9ca CH |
600 | if (!mrtryaccess(&ip->i_iolock)) |
601 | goto out; | |
1da177e4 LT |
602 | } |
603 | if (lock_flags & XFS_ILOCK_EXCL) { | |
579aa9ca CH |
604 | if (!mrtryupdate(&ip->i_lock)) |
605 | goto out_undo_iolock; | |
1da177e4 | 606 | } else if (lock_flags & XFS_ILOCK_SHARED) { |
579aa9ca CH |
607 | if (!mrtryaccess(&ip->i_lock)) |
608 | goto out_undo_iolock; | |
1da177e4 | 609 | } |
0b1b213f | 610 | trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_); |
1da177e4 | 611 | return 1; |
579aa9ca CH |
612 | |
613 | out_undo_iolock: | |
614 | if (lock_flags & XFS_IOLOCK_EXCL) | |
615 | mrunlock_excl(&ip->i_iolock); | |
616 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
617 | mrunlock_shared(&ip->i_iolock); | |
618 | out: | |
619 | return 0; | |
1da177e4 LT |
620 | } |
621 | ||
622 | /* | |
623 | * xfs_iunlock() is used to drop the inode locks acquired with | |
624 | * xfs_ilock() and xfs_ilock_nowait(). The caller must pass | |
625 | * in the flags given to xfs_ilock() or xfs_ilock_nowait() so | |
626 | * that we know which locks to drop. | |
627 | * | |
628 | * ip -- the inode being unlocked | |
629 | * lock_flags -- this parameter indicates the inode's locks to be | |
630 | * to be unlocked. See the comment for xfs_ilock() for a list | |
631 | * of valid values for this parameter. | |
632 | * | |
633 | */ | |
634 | void | |
579aa9ca CH |
635 | xfs_iunlock( |
636 | xfs_inode_t *ip, | |
637 | uint lock_flags) | |
1da177e4 LT |
638 | { |
639 | /* | |
640 | * You can't set both SHARED and EXCL for the same lock, | |
641 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
642 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
643 | */ | |
644 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
645 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
646 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
647 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
f7c66ce3 LM |
648 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_IUNLOCK_NONOTIFY | |
649 | XFS_LOCK_DEP_MASK)) == 0); | |
1da177e4 LT |
650 | ASSERT(lock_flags != 0); |
651 | ||
579aa9ca CH |
652 | if (lock_flags & XFS_IOLOCK_EXCL) |
653 | mrunlock_excl(&ip->i_iolock); | |
654 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
655 | mrunlock_shared(&ip->i_iolock); | |
1da177e4 | 656 | |
579aa9ca CH |
657 | if (lock_flags & XFS_ILOCK_EXCL) |
658 | mrunlock_excl(&ip->i_lock); | |
659 | else if (lock_flags & XFS_ILOCK_SHARED) | |
660 | mrunlock_shared(&ip->i_lock); | |
1da177e4 | 661 | |
579aa9ca CH |
662 | if ((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) && |
663 | !(lock_flags & XFS_IUNLOCK_NONOTIFY) && ip->i_itemp) { | |
1da177e4 LT |
664 | /* |
665 | * Let the AIL know that this item has been unlocked in case | |
666 | * it is in the AIL and anyone is waiting on it. Don't do | |
667 | * this if the caller has asked us not to. | |
668 | */ | |
783a2f65 | 669 | xfs_trans_unlocked_item(ip->i_itemp->ili_item.li_ailp, |
579aa9ca | 670 | (xfs_log_item_t*)(ip->i_itemp)); |
1da177e4 | 671 | } |
0b1b213f | 672 | trace_xfs_iunlock(ip, lock_flags, _RET_IP_); |
1da177e4 LT |
673 | } |
674 | ||
675 | /* | |
676 | * give up write locks. the i/o lock cannot be held nested | |
677 | * if it is being demoted. | |
678 | */ | |
679 | void | |
579aa9ca CH |
680 | xfs_ilock_demote( |
681 | xfs_inode_t *ip, | |
682 | uint lock_flags) | |
1da177e4 LT |
683 | { |
684 | ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)); | |
685 | ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0); | |
686 | ||
579aa9ca | 687 | if (lock_flags & XFS_ILOCK_EXCL) |
1da177e4 | 688 | mrdemote(&ip->i_lock); |
579aa9ca | 689 | if (lock_flags & XFS_IOLOCK_EXCL) |
1da177e4 | 690 | mrdemote(&ip->i_iolock); |
0b1b213f CH |
691 | |
692 | trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_); | |
579aa9ca CH |
693 | } |
694 | ||
695 | #ifdef DEBUG | |
579aa9ca CH |
696 | int |
697 | xfs_isilocked( | |
698 | xfs_inode_t *ip, | |
699 | uint lock_flags) | |
700 | { | |
f9369729 CH |
701 | if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) { |
702 | if (!(lock_flags & XFS_ILOCK_SHARED)) | |
703 | return !!ip->i_lock.mr_writer; | |
704 | return rwsem_is_locked(&ip->i_lock.mr_lock); | |
1da177e4 | 705 | } |
579aa9ca | 706 | |
f9369729 CH |
707 | if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) { |
708 | if (!(lock_flags & XFS_IOLOCK_SHARED)) | |
709 | return !!ip->i_iolock.mr_writer; | |
710 | return rwsem_is_locked(&ip->i_iolock.mr_lock); | |
579aa9ca CH |
711 | } |
712 | ||
f9369729 CH |
713 | ASSERT(0); |
714 | return 0; | |
1da177e4 | 715 | } |
579aa9ca | 716 | #endif |