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" |
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 | 27 | #include "xfs_mount.h" |
1da177e4 | 28 | #include "xfs_bmap_btree.h" |
a844f451 | 29 | #include "xfs_alloc_btree.h" |
1da177e4 | 30 | #include "xfs_ialloc_btree.h" |
1da177e4 LT |
31 | #include "xfs_dinode.h" |
32 | #include "xfs_inode.h" | |
a844f451 NS |
33 | #include "xfs_btree.h" |
34 | #include "xfs_ialloc.h" | |
1da177e4 LT |
35 | #include "xfs_quota.h" |
36 | #include "xfs_utils.h" | |
783a2f65 DC |
37 | #include "xfs_trans_priv.h" |
38 | #include "xfs_inode_item.h" | |
24f211ba | 39 | #include "xfs_bmap.h" |
0b1b213f | 40 | #include "xfs_trace.h" |
24f211ba CH |
41 | |
42 | ||
43 | /* | |
44 | * Allocate and initialise an xfs_inode. | |
45 | */ | |
46 | STATIC struct xfs_inode * | |
47 | xfs_inode_alloc( | |
48 | struct xfs_mount *mp, | |
49 | xfs_ino_t ino) | |
50 | { | |
51 | struct xfs_inode *ip; | |
52 | ||
53 | /* | |
54 | * if this didn't occur in transactions, we could use | |
55 | * KM_MAYFAIL and return NULL here on ENOMEM. Set the | |
56 | * code up to do this anyway. | |
57 | */ | |
58 | ip = kmem_zone_alloc(xfs_inode_zone, KM_SLEEP); | |
59 | if (!ip) | |
60 | return NULL; | |
54e34621 CH |
61 | if (inode_init_always(mp->m_super, VFS_I(ip))) { |
62 | kmem_zone_free(xfs_inode_zone, ip); | |
63 | return NULL; | |
64 | } | |
24f211ba | 65 | |
24f211ba CH |
66 | ASSERT(atomic_read(&ip->i_pincount) == 0); |
67 | ASSERT(!spin_is_locked(&ip->i_flags_lock)); | |
474fce06 | 68 | ASSERT(!xfs_isiflocked(ip)); |
1a3e8f3d | 69 | ASSERT(ip->i_ino == 0); |
033da48f CH |
70 | |
71 | mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino); | |
24f211ba | 72 | |
24f211ba CH |
73 | /* initialise the xfs inode */ |
74 | ip->i_ino = ino; | |
75 | ip->i_mount = mp; | |
76 | memset(&ip->i_imap, 0, sizeof(struct xfs_imap)); | |
77 | ip->i_afp = NULL; | |
78 | memset(&ip->i_df, 0, sizeof(xfs_ifork_t)); | |
79 | ip->i_flags = 0; | |
24f211ba CH |
80 | ip->i_delayed_blks = 0; |
81 | memset(&ip->i_d, 0, sizeof(xfs_icdinode_t)); | |
24f211ba | 82 | |
24f211ba CH |
83 | return ip; |
84 | } | |
1da177e4 | 85 | |
fa0d7e3d NP |
86 | STATIC void |
87 | xfs_inode_free_callback( | |
88 | struct rcu_head *head) | |
89 | { | |
90 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
91 | struct xfs_inode *ip = XFS_I(inode); | |
92 | ||
fa0d7e3d NP |
93 | kmem_zone_free(xfs_inode_zone, ip); |
94 | } | |
95 | ||
2f11feab | 96 | void |
b36ec042 CH |
97 | xfs_inode_free( |
98 | struct xfs_inode *ip) | |
99 | { | |
100 | switch (ip->i_d.di_mode & S_IFMT) { | |
101 | case S_IFREG: | |
102 | case S_IFDIR: | |
103 | case S_IFLNK: | |
104 | xfs_idestroy_fork(ip, XFS_DATA_FORK); | |
105 | break; | |
106 | } | |
107 | ||
108 | if (ip->i_afp) | |
109 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
110 | ||
b36ec042 | 111 | if (ip->i_itemp) { |
32ce90a4 | 112 | ASSERT(!(ip->i_itemp->ili_item.li_flags & XFS_LI_IN_AIL)); |
b36ec042 CH |
113 | xfs_inode_item_destroy(ip); |
114 | ip->i_itemp = NULL; | |
115 | } | |
116 | ||
117 | /* asserts to verify all state is correct here */ | |
b36ec042 CH |
118 | ASSERT(atomic_read(&ip->i_pincount) == 0); |
119 | ASSERT(!spin_is_locked(&ip->i_flags_lock)); | |
474fce06 | 120 | ASSERT(!xfs_isiflocked(ip)); |
b36ec042 | 121 | |
1a3e8f3d DC |
122 | /* |
123 | * Because we use RCU freeing we need to ensure the inode always | |
124 | * appears to be reclaimed with an invalid inode number when in the | |
125 | * free state. The ip->i_flags_lock provides the barrier against lookup | |
126 | * races. | |
127 | */ | |
128 | spin_lock(&ip->i_flags_lock); | |
129 | ip->i_flags = XFS_IRECLAIM; | |
130 | ip->i_ino = 0; | |
131 | spin_unlock(&ip->i_flags_lock); | |
92f1c008 AE |
132 | |
133 | call_rcu(&VFS_I(ip)->i_rcu, xfs_inode_free_callback); | |
b36ec042 CH |
134 | } |
135 | ||
1da177e4 | 136 | /* |
6441e549 | 137 | * Check the validity of the inode we just found it the cache |
1da177e4 | 138 | */ |
6441e549 DC |
139 | static int |
140 | xfs_iget_cache_hit( | |
6441e549 DC |
141 | struct xfs_perag *pag, |
142 | struct xfs_inode *ip, | |
1a3e8f3d | 143 | xfs_ino_t ino, |
6441e549 | 144 | int flags, |
1a3e8f3d | 145 | int lock_flags) __releases(RCU) |
1da177e4 | 146 | { |
bc990f5c | 147 | struct inode *inode = VFS_I(ip); |
6441e549 | 148 | struct xfs_mount *mp = ip->i_mount; |
bc990f5c CH |
149 | int error; |
150 | ||
1a3e8f3d DC |
151 | /* |
152 | * check for re-use of an inode within an RCU grace period due to the | |
153 | * radix tree nodes not being updated yet. We monitor for this by | |
154 | * setting the inode number to zero before freeing the inode structure. | |
155 | * If the inode has been reallocated and set up, then the inode number | |
156 | * will not match, so check for that, too. | |
157 | */ | |
bc990f5c | 158 | spin_lock(&ip->i_flags_lock); |
1a3e8f3d DC |
159 | if (ip->i_ino != ino) { |
160 | trace_xfs_iget_skip(ip); | |
161 | XFS_STATS_INC(xs_ig_frecycle); | |
162 | error = EAGAIN; | |
163 | goto out_error; | |
164 | } | |
165 | ||
da353b0d | 166 | |
6441e549 | 167 | /* |
bc990f5c CH |
168 | * If we are racing with another cache hit that is currently |
169 | * instantiating this inode or currently recycling it out of | |
170 | * reclaimabe state, wait for the initialisation to complete | |
171 | * before continuing. | |
172 | * | |
173 | * XXX(hch): eventually we should do something equivalent to | |
174 | * wait_on_inode to wait for these flags to be cleared | |
175 | * instead of polling for it. | |
6441e549 | 176 | */ |
bc990f5c | 177 | if (ip->i_flags & (XFS_INEW|XFS_IRECLAIM)) { |
0b1b213f | 178 | trace_xfs_iget_skip(ip); |
6441e549 | 179 | XFS_STATS_INC(xs_ig_frecycle); |
bc990f5c | 180 | error = EAGAIN; |
6441e549 DC |
181 | goto out_error; |
182 | } | |
da353b0d | 183 | |
bc990f5c CH |
184 | /* |
185 | * If lookup is racing with unlink return an error immediately. | |
186 | */ | |
187 | if (ip->i_d.di_mode == 0 && !(flags & XFS_IGET_CREATE)) { | |
188 | error = ENOENT; | |
189 | goto out_error; | |
190 | } | |
bf904248 | 191 | |
bc990f5c CH |
192 | /* |
193 | * If IRECLAIMABLE is set, we've torn down the VFS inode already. | |
194 | * Need to carefully get it back into useable state. | |
195 | */ | |
196 | if (ip->i_flags & XFS_IRECLAIMABLE) { | |
0b1b213f | 197 | trace_xfs_iget_reclaim(ip); |
da353b0d | 198 | |
6441e549 | 199 | /* |
f1f724e4 CH |
200 | * We need to set XFS_IRECLAIM to prevent xfs_reclaim_inode |
201 | * from stomping over us while we recycle the inode. We can't | |
202 | * clear the radix tree reclaimable tag yet as it requires | |
203 | * pag_ici_lock to be held exclusive. | |
6441e549 | 204 | */ |
f1f724e4 | 205 | ip->i_flags |= XFS_IRECLAIM; |
bc990f5c CH |
206 | |
207 | spin_unlock(&ip->i_flags_lock); | |
1a3e8f3d | 208 | rcu_read_unlock(); |
bc990f5c CH |
209 | |
210 | error = -inode_init_always(mp->m_super, inode); | |
211 | if (error) { | |
212 | /* | |
213 | * Re-initializing the inode failed, and we are in deep | |
214 | * trouble. Try to re-add it to the reclaim list. | |
215 | */ | |
1a3e8f3d | 216 | rcu_read_lock(); |
bc990f5c CH |
217 | spin_lock(&ip->i_flags_lock); |
218 | ||
778e24bb DC |
219 | ip->i_flags &= ~(XFS_INEW | XFS_IRECLAIM); |
220 | ASSERT(ip->i_flags & XFS_IRECLAIMABLE); | |
d2e078c3 | 221 | trace_xfs_iget_reclaim_fail(ip); |
6441e549 | 222 | goto out_error; |
da353b0d | 223 | } |
f1f724e4 | 224 | |
1a427ab0 | 225 | spin_lock(&pag->pag_ici_lock); |
f1f724e4 | 226 | spin_lock(&ip->i_flags_lock); |
778e24bb DC |
227 | |
228 | /* | |
229 | * Clear the per-lifetime state in the inode as we are now | |
230 | * effectively a new inode and need to return to the initial | |
231 | * state before reuse occurs. | |
232 | */ | |
233 | ip->i_flags &= ~XFS_IRECLAIM_RESET_FLAGS; | |
f1f724e4 CH |
234 | ip->i_flags |= XFS_INEW; |
235 | __xfs_inode_clear_reclaim_tag(mp, pag, ip); | |
eaff8079 | 236 | inode->i_state = I_NEW; |
dcfcf205 DC |
237 | |
238 | ASSERT(!rwsem_is_locked(&ip->i_iolock.mr_lock)); | |
239 | mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino); | |
dcfcf205 | 240 | |
f1f724e4 | 241 | spin_unlock(&ip->i_flags_lock); |
1a427ab0 | 242 | spin_unlock(&pag->pag_ici_lock); |
bc990f5c | 243 | } else { |
bf904248 | 244 | /* If the VFS inode is being torn down, pause and try again. */ |
bc990f5c | 245 | if (!igrab(inode)) { |
d2e078c3 | 246 | trace_xfs_iget_skip(ip); |
bc990f5c CH |
247 | error = EAGAIN; |
248 | goto out_error; | |
249 | } | |
1da177e4 | 250 | |
bc990f5c CH |
251 | /* We've got a live one. */ |
252 | spin_unlock(&ip->i_flags_lock); | |
1a3e8f3d | 253 | rcu_read_unlock(); |
d2e078c3 | 254 | trace_xfs_iget_hit(ip); |
6441e549 | 255 | } |
da353b0d | 256 | |
6441e549 DC |
257 | if (lock_flags != 0) |
258 | xfs_ilock(ip, lock_flags); | |
da353b0d | 259 | |
5132ba8f | 260 | xfs_iflags_clear(ip, XFS_ISTALE | XFS_IDONTCACHE); |
6441e549 | 261 | XFS_STATS_INC(xs_ig_found); |
0b1b213f | 262 | |
6441e549 | 263 | return 0; |
1da177e4 | 264 | |
6441e549 | 265 | out_error: |
bc990f5c | 266 | spin_unlock(&ip->i_flags_lock); |
1a3e8f3d | 267 | rcu_read_unlock(); |
6441e549 DC |
268 | return error; |
269 | } | |
270 | ||
271 | ||
272 | static int | |
273 | xfs_iget_cache_miss( | |
274 | struct xfs_mount *mp, | |
275 | struct xfs_perag *pag, | |
276 | xfs_trans_t *tp, | |
277 | xfs_ino_t ino, | |
278 | struct xfs_inode **ipp, | |
6441e549 | 279 | int flags, |
0c3dc2b0 | 280 | int lock_flags) |
6441e549 DC |
281 | { |
282 | struct xfs_inode *ip; | |
283 | int error; | |
6441e549 | 284 | xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino); |
5132ba8f | 285 | int iflags; |
1da177e4 | 286 | |
24f211ba CH |
287 | ip = xfs_inode_alloc(mp, ino); |
288 | if (!ip) | |
289 | return ENOMEM; | |
290 | ||
7b6259e7 | 291 | error = xfs_iread(mp, tp, ip, flags); |
6441e549 | 292 | if (error) |
24f211ba | 293 | goto out_destroy; |
1da177e4 | 294 | |
d2e078c3 | 295 | trace_xfs_iget_miss(ip); |
1da177e4 | 296 | |
745b1f47 | 297 | if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) { |
6441e549 DC |
298 | error = ENOENT; |
299 | goto out_destroy; | |
1da177e4 LT |
300 | } |
301 | ||
302 | /* | |
bad55843 | 303 | * Preload the radix tree so we can insert safely under the |
56e73ec4 | 304 | * write spinlock. Note that we cannot sleep inside the preload |
3ba31603 PW |
305 | * region. Since we can be called from transaction context, don't |
306 | * recurse into the file system. | |
1da177e4 | 307 | */ |
3ba31603 | 308 | if (radix_tree_preload(GFP_NOFS)) { |
6441e549 | 309 | error = EAGAIN; |
ed93ec39 CH |
310 | goto out_destroy; |
311 | } | |
312 | ||
313 | /* | |
314 | * Because the inode hasn't been added to the radix-tree yet it can't | |
315 | * be found by another thread, so we can do the non-sleeping lock here. | |
316 | */ | |
317 | if (lock_flags) { | |
318 | if (!xfs_ilock_nowait(ip, lock_flags)) | |
319 | BUG(); | |
da353b0d | 320 | } |
f338f903 | 321 | |
f30d500f DC |
322 | /* |
323 | * These values must be set before inserting the inode into the radix | |
324 | * tree as the moment it is inserted a concurrent lookup (allowed by the | |
325 | * RCU locking mechanism) can find it and that lookup must see that this | |
326 | * is an inode currently under construction (i.e. that XFS_INEW is set). | |
327 | * The ip->i_flags_lock that protects the XFS_INEW flag forms the | |
328 | * memory barrier that ensures this detection works correctly at lookup | |
329 | * time. | |
330 | */ | |
5132ba8f DC |
331 | iflags = XFS_INEW; |
332 | if (flags & XFS_IGET_DONTCACHE) | |
333 | iflags |= XFS_IDONTCACHE; | |
f30d500f | 334 | ip->i_udquot = ip->i_gdquot = NULL; |
5132ba8f | 335 | xfs_iflags_set(ip, iflags); |
6441e549 DC |
336 | |
337 | /* insert the new inode */ | |
f30d500f | 338 | spin_lock(&pag->pag_ici_lock); |
da353b0d DC |
339 | error = radix_tree_insert(&pag->pag_ici_root, agino, ip); |
340 | if (unlikely(error)) { | |
6441e549 | 341 | WARN_ON(error != -EEXIST); |
da353b0d | 342 | XFS_STATS_INC(xs_ig_dup); |
6441e549 | 343 | error = EAGAIN; |
56e73ec4 | 344 | goto out_preload_end; |
1da177e4 | 345 | } |
1a427ab0 | 346 | spin_unlock(&pag->pag_ici_lock); |
da353b0d | 347 | radix_tree_preload_end(); |
0b1b213f | 348 | |
6441e549 DC |
349 | *ipp = ip; |
350 | return 0; | |
351 | ||
56e73ec4 | 352 | out_preload_end: |
1a427ab0 | 353 | spin_unlock(&pag->pag_ici_lock); |
6441e549 | 354 | radix_tree_preload_end(); |
56e73ec4 DC |
355 | if (lock_flags) |
356 | xfs_iunlock(ip, lock_flags); | |
6441e549 | 357 | out_destroy: |
b36ec042 CH |
358 | __destroy_inode(VFS_I(ip)); |
359 | xfs_inode_free(ip); | |
6441e549 DC |
360 | return error; |
361 | } | |
362 | ||
363 | /* | |
364 | * Look up an inode by number in the given file system. | |
365 | * The inode is looked up in the cache held in each AG. | |
bf904248 DC |
366 | * If the inode is found in the cache, initialise the vfs inode |
367 | * if necessary. | |
6441e549 DC |
368 | * |
369 | * If it is not in core, read it in from the file system's device, | |
bf904248 | 370 | * add it to the cache and initialise the vfs inode. |
6441e549 DC |
371 | * |
372 | * The inode is locked according to the value of the lock_flags parameter. | |
373 | * This flag parameter indicates how and if the inode's IO lock and inode lock | |
374 | * should be taken. | |
375 | * | |
376 | * mp -- the mount point structure for the current file system. It points | |
377 | * to the inode hash table. | |
378 | * tp -- a pointer to the current transaction if there is one. This is | |
379 | * simply passed through to the xfs_iread() call. | |
380 | * ino -- the number of the inode desired. This is the unique identifier | |
381 | * within the file system for the inode being requested. | |
382 | * lock_flags -- flags indicating how to lock the inode. See the comment | |
383 | * for xfs_ilock() for a list of valid values. | |
6441e549 | 384 | */ |
bf904248 DC |
385 | int |
386 | xfs_iget( | |
6441e549 DC |
387 | xfs_mount_t *mp, |
388 | xfs_trans_t *tp, | |
389 | xfs_ino_t ino, | |
390 | uint flags, | |
391 | uint lock_flags, | |
7b6259e7 | 392 | xfs_inode_t **ipp) |
6441e549 DC |
393 | { |
394 | xfs_inode_t *ip; | |
395 | int error; | |
396 | xfs_perag_t *pag; | |
397 | xfs_agino_t agino; | |
398 | ||
ad637a10 AE |
399 | /* |
400 | * xfs_reclaim_inode() uses the ILOCK to ensure an inode | |
401 | * doesn't get freed while it's being referenced during a | |
402 | * radix tree traversal here. It assumes this function | |
403 | * aqcuires only the ILOCK (and therefore it has no need to | |
404 | * involve the IOLOCK in this synchronization). | |
405 | */ | |
406 | ASSERT((lock_flags & (XFS_IOLOCK_EXCL | XFS_IOLOCK_SHARED)) == 0); | |
407 | ||
d276734d | 408 | /* reject inode numbers outside existing AGs */ |
1a3e8f3d | 409 | if (!ino || XFS_INO_TO_AGNO(mp, ino) >= mp->m_sb.sb_agcount) |
6441e549 DC |
410 | return EINVAL; |
411 | ||
412 | /* get the perag structure and ensure that it's inode capable */ | |
5017e97d | 413 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ino)); |
6441e549 DC |
414 | agino = XFS_INO_TO_AGINO(mp, ino); |
415 | ||
416 | again: | |
417 | error = 0; | |
1a3e8f3d | 418 | rcu_read_lock(); |
6441e549 DC |
419 | ip = radix_tree_lookup(&pag->pag_ici_root, agino); |
420 | ||
421 | if (ip) { | |
1a3e8f3d | 422 | error = xfs_iget_cache_hit(pag, ip, ino, flags, lock_flags); |
6441e549 DC |
423 | if (error) |
424 | goto out_error_or_again; | |
425 | } else { | |
1a3e8f3d | 426 | rcu_read_unlock(); |
6441e549 DC |
427 | XFS_STATS_INC(xs_ig_missed); |
428 | ||
7b6259e7 | 429 | error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip, |
6441e549 DC |
430 | flags, lock_flags); |
431 | if (error) | |
432 | goto out_error_or_again; | |
433 | } | |
5017e97d | 434 | xfs_perag_put(pag); |
1da177e4 | 435 | |
1da177e4 LT |
436 | *ipp = ip; |
437 | ||
438 | /* | |
439 | * If we have a real type for an on-disk inode, we can set ops(&unlock) | |
440 | * now. If it's a new inode being created, xfs_ialloc will handle it. | |
441 | */ | |
bf904248 | 442 | if (xfs_iflags_test(ip, XFS_INEW) && ip->i_d.di_mode != 0) |
41be8bed | 443 | xfs_setup_inode(ip); |
1da177e4 | 444 | return 0; |
6441e549 DC |
445 | |
446 | out_error_or_again: | |
447 | if (error == EAGAIN) { | |
448 | delay(1); | |
449 | goto again; | |
450 | } | |
5017e97d | 451 | xfs_perag_put(pag); |
6441e549 | 452 | return error; |
1da177e4 LT |
453 | } |
454 | ||
1da177e4 LT |
455 | /* |
456 | * This is a wrapper routine around the xfs_ilock() routine | |
457 | * used to centralize some grungy code. It is used in places | |
458 | * that wish to lock the inode solely for reading the extents. | |
459 | * The reason these places can't just call xfs_ilock(SHARED) | |
460 | * is that the inode lock also guards to bringing in of the | |
461 | * extents from disk for a file in b-tree format. If the inode | |
462 | * is in b-tree format, then we need to lock the inode exclusively | |
463 | * until the extents are read in. Locking it exclusively all | |
464 | * the time would limit our parallelism unnecessarily, though. | |
465 | * What we do instead is check to see if the extents have been | |
466 | * read in yet, and only lock the inode exclusively if they | |
467 | * have not. | |
468 | * | |
469 | * The function returns a value which should be given to the | |
470 | * corresponding xfs_iunlock_map_shared(). This value is | |
471 | * the mode in which the lock was actually taken. | |
472 | */ | |
473 | uint | |
474 | xfs_ilock_map_shared( | |
475 | xfs_inode_t *ip) | |
476 | { | |
477 | uint lock_mode; | |
478 | ||
479 | if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) && | |
480 | ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) { | |
481 | lock_mode = XFS_ILOCK_EXCL; | |
482 | } else { | |
483 | lock_mode = XFS_ILOCK_SHARED; | |
484 | } | |
485 | ||
486 | xfs_ilock(ip, lock_mode); | |
487 | ||
488 | return lock_mode; | |
489 | } | |
490 | ||
491 | /* | |
492 | * This is simply the unlock routine to go with xfs_ilock_map_shared(). | |
493 | * All it does is call xfs_iunlock() with the given lock_mode. | |
494 | */ | |
495 | void | |
496 | xfs_iunlock_map_shared( | |
497 | xfs_inode_t *ip, | |
498 | unsigned int lock_mode) | |
499 | { | |
500 | xfs_iunlock(ip, lock_mode); | |
501 | } | |
502 | ||
503 | /* | |
504 | * The xfs inode contains 2 locks: a multi-reader lock called the | |
505 | * i_iolock and a multi-reader lock called the i_lock. This routine | |
506 | * allows either or both of the locks to be obtained. | |
507 | * | |
508 | * The 2 locks should always be ordered so that the IO lock is | |
509 | * obtained first in order to prevent deadlock. | |
510 | * | |
511 | * ip -- the inode being locked | |
512 | * lock_flags -- this parameter indicates the inode's locks | |
513 | * to be locked. It can be: | |
514 | * XFS_IOLOCK_SHARED, | |
515 | * XFS_IOLOCK_EXCL, | |
516 | * XFS_ILOCK_SHARED, | |
517 | * XFS_ILOCK_EXCL, | |
518 | * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED, | |
519 | * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL, | |
520 | * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED, | |
521 | * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL | |
522 | */ | |
523 | void | |
579aa9ca CH |
524 | xfs_ilock( |
525 | xfs_inode_t *ip, | |
526 | uint lock_flags) | |
1da177e4 LT |
527 | { |
528 | /* | |
529 | * You can't set both SHARED and EXCL for the same lock, | |
530 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
531 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
532 | */ | |
533 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
534 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
535 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
536 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
f7c66ce3 | 537 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); |
1da177e4 | 538 | |
579aa9ca | 539 | if (lock_flags & XFS_IOLOCK_EXCL) |
f7c66ce3 | 540 | mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); |
579aa9ca | 541 | else if (lock_flags & XFS_IOLOCK_SHARED) |
f7c66ce3 | 542 | mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); |
579aa9ca CH |
543 | |
544 | if (lock_flags & XFS_ILOCK_EXCL) | |
f7c66ce3 | 545 | mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); |
579aa9ca | 546 | else if (lock_flags & XFS_ILOCK_SHARED) |
f7c66ce3 | 547 | mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); |
579aa9ca | 548 | |
0b1b213f | 549 | trace_xfs_ilock(ip, lock_flags, _RET_IP_); |
1da177e4 LT |
550 | } |
551 | ||
552 | /* | |
553 | * This is just like xfs_ilock(), except that the caller | |
554 | * is guaranteed not to sleep. It returns 1 if it gets | |
555 | * the requested locks and 0 otherwise. If the IO lock is | |
556 | * obtained but the inode lock cannot be, then the IO lock | |
557 | * is dropped before returning. | |
558 | * | |
559 | * ip -- the inode being locked | |
560 | * lock_flags -- this parameter indicates the inode's locks to be | |
561 | * to be locked. See the comment for xfs_ilock() for a list | |
562 | * of valid values. | |
1da177e4 LT |
563 | */ |
564 | int | |
579aa9ca CH |
565 | xfs_ilock_nowait( |
566 | xfs_inode_t *ip, | |
567 | uint lock_flags) | |
1da177e4 | 568 | { |
1da177e4 LT |
569 | /* |
570 | * You can't set both SHARED and EXCL for the same lock, | |
571 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
572 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
573 | */ | |
574 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
575 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
576 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
577 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
f7c66ce3 | 578 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); |
1da177e4 | 579 | |
1da177e4 | 580 | if (lock_flags & XFS_IOLOCK_EXCL) { |
579aa9ca CH |
581 | if (!mrtryupdate(&ip->i_iolock)) |
582 | goto out; | |
1da177e4 | 583 | } else if (lock_flags & XFS_IOLOCK_SHARED) { |
579aa9ca CH |
584 | if (!mrtryaccess(&ip->i_iolock)) |
585 | goto out; | |
1da177e4 LT |
586 | } |
587 | if (lock_flags & XFS_ILOCK_EXCL) { | |
579aa9ca CH |
588 | if (!mrtryupdate(&ip->i_lock)) |
589 | goto out_undo_iolock; | |
1da177e4 | 590 | } else if (lock_flags & XFS_ILOCK_SHARED) { |
579aa9ca CH |
591 | if (!mrtryaccess(&ip->i_lock)) |
592 | goto out_undo_iolock; | |
1da177e4 | 593 | } |
0b1b213f | 594 | trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_); |
1da177e4 | 595 | return 1; |
579aa9ca CH |
596 | |
597 | out_undo_iolock: | |
598 | if (lock_flags & XFS_IOLOCK_EXCL) | |
599 | mrunlock_excl(&ip->i_iolock); | |
600 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
601 | mrunlock_shared(&ip->i_iolock); | |
602 | out: | |
603 | return 0; | |
1da177e4 LT |
604 | } |
605 | ||
606 | /* | |
607 | * xfs_iunlock() is used to drop the inode locks acquired with | |
608 | * xfs_ilock() and xfs_ilock_nowait(). The caller must pass | |
609 | * in the flags given to xfs_ilock() or xfs_ilock_nowait() so | |
610 | * that we know which locks to drop. | |
611 | * | |
612 | * ip -- the inode being unlocked | |
613 | * lock_flags -- this parameter indicates the inode's locks to be | |
614 | * to be unlocked. See the comment for xfs_ilock() for a list | |
615 | * of valid values for this parameter. | |
616 | * | |
617 | */ | |
618 | void | |
579aa9ca CH |
619 | xfs_iunlock( |
620 | xfs_inode_t *ip, | |
621 | uint lock_flags) | |
1da177e4 LT |
622 | { |
623 | /* | |
624 | * You can't set both SHARED and EXCL for the same lock, | |
625 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
626 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
627 | */ | |
628 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
629 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
630 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
631 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
5b03ff1b | 632 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); |
1da177e4 LT |
633 | ASSERT(lock_flags != 0); |
634 | ||
579aa9ca CH |
635 | if (lock_flags & XFS_IOLOCK_EXCL) |
636 | mrunlock_excl(&ip->i_iolock); | |
637 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
638 | mrunlock_shared(&ip->i_iolock); | |
1da177e4 | 639 | |
579aa9ca CH |
640 | if (lock_flags & XFS_ILOCK_EXCL) |
641 | mrunlock_excl(&ip->i_lock); | |
642 | else if (lock_flags & XFS_ILOCK_SHARED) | |
643 | mrunlock_shared(&ip->i_lock); | |
1da177e4 | 644 | |
0b1b213f | 645 | trace_xfs_iunlock(ip, lock_flags, _RET_IP_); |
1da177e4 LT |
646 | } |
647 | ||
648 | /* | |
649 | * give up write locks. the i/o lock cannot be held nested | |
650 | * if it is being demoted. | |
651 | */ | |
652 | void | |
579aa9ca CH |
653 | xfs_ilock_demote( |
654 | xfs_inode_t *ip, | |
655 | uint lock_flags) | |
1da177e4 LT |
656 | { |
657 | ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)); | |
658 | ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0); | |
659 | ||
579aa9ca | 660 | if (lock_flags & XFS_ILOCK_EXCL) |
1da177e4 | 661 | mrdemote(&ip->i_lock); |
579aa9ca | 662 | if (lock_flags & XFS_IOLOCK_EXCL) |
1da177e4 | 663 | mrdemote(&ip->i_iolock); |
0b1b213f CH |
664 | |
665 | trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_); | |
579aa9ca CH |
666 | } |
667 | ||
668 | #ifdef DEBUG | |
579aa9ca CH |
669 | int |
670 | xfs_isilocked( | |
671 | xfs_inode_t *ip, | |
672 | uint lock_flags) | |
673 | { | |
f9369729 CH |
674 | if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) { |
675 | if (!(lock_flags & XFS_ILOCK_SHARED)) | |
676 | return !!ip->i_lock.mr_writer; | |
677 | return rwsem_is_locked(&ip->i_lock.mr_lock); | |
1da177e4 | 678 | } |
579aa9ca | 679 | |
f9369729 CH |
680 | if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) { |
681 | if (!(lock_flags & XFS_IOLOCK_SHARED)) | |
682 | return !!ip->i_iolock.mr_writer; | |
683 | return rwsem_is_locked(&ip->i_iolock.mr_lock); | |
579aa9ca CH |
684 | } |
685 | ||
f9369729 CH |
686 | ASSERT(0); |
687 | return 0; | |
1da177e4 | 688 | } |
579aa9ca | 689 | #endif |
474fce06 CH |
690 | |
691 | void | |
692 | __xfs_iflock( | |
693 | struct xfs_inode *ip) | |
694 | { | |
695 | wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT); | |
696 | DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT); | |
697 | ||
698 | do { | |
699 | prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE); | |
700 | if (xfs_isiflocked(ip)) | |
701 | io_schedule(); | |
702 | } while (!xfs_iflock_nowait(ip)); | |
703 | ||
704 | finish_wait(wq, &wait.wait); | |
705 | } |