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 | 40 | #include "xfs_bmap.h" |
0b1b213f | 41 | #include "xfs_trace.h" |
24f211ba CH |
42 | |
43 | ||
dcfcf205 DC |
44 | /* |
45 | * Define xfs inode iolock lockdep classes. We need to ensure that all active | |
46 | * inodes are considered the same for lockdep purposes, including inodes that | |
47 | * are recycled through the XFS_IRECLAIMABLE state. This is the the only way to | |
48 | * guarantee the locks are considered the same when there are multiple lock | |
49 | * initialisation siteѕ. Also, define a reclaimable inode class so it is | |
50 | * obvious in lockdep reports which class the report is against. | |
51 | */ | |
52 | static struct lock_class_key xfs_iolock_active; | |
53 | struct lock_class_key xfs_iolock_reclaimable; | |
54 | ||
24f211ba CH |
55 | /* |
56 | * Allocate and initialise an xfs_inode. | |
57 | */ | |
58 | STATIC struct xfs_inode * | |
59 | xfs_inode_alloc( | |
60 | struct xfs_mount *mp, | |
61 | xfs_ino_t ino) | |
62 | { | |
63 | struct xfs_inode *ip; | |
64 | ||
65 | /* | |
66 | * if this didn't occur in transactions, we could use | |
67 | * KM_MAYFAIL and return NULL here on ENOMEM. Set the | |
68 | * code up to do this anyway. | |
69 | */ | |
70 | ip = kmem_zone_alloc(xfs_inode_zone, KM_SLEEP); | |
71 | if (!ip) | |
72 | return NULL; | |
54e34621 CH |
73 | if (inode_init_always(mp->m_super, VFS_I(ip))) { |
74 | kmem_zone_free(xfs_inode_zone, ip); | |
75 | return NULL; | |
76 | } | |
24f211ba | 77 | |
24f211ba CH |
78 | ASSERT(atomic_read(&ip->i_pincount) == 0); |
79 | ASSERT(!spin_is_locked(&ip->i_flags_lock)); | |
80 | ASSERT(completion_done(&ip->i_flush)); | |
1a3e8f3d | 81 | ASSERT(ip->i_ino == 0); |
033da48f CH |
82 | |
83 | mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino); | |
dcfcf205 DC |
84 | lockdep_set_class_and_name(&ip->i_iolock.mr_lock, |
85 | &xfs_iolock_active, "xfs_iolock_active"); | |
24f211ba | 86 | |
24f211ba CH |
87 | /* initialise the xfs inode */ |
88 | ip->i_ino = ino; | |
89 | ip->i_mount = mp; | |
90 | memset(&ip->i_imap, 0, sizeof(struct xfs_imap)); | |
91 | ip->i_afp = NULL; | |
92 | memset(&ip->i_df, 0, sizeof(xfs_ifork_t)); | |
93 | ip->i_flags = 0; | |
94 | ip->i_update_core = 0; | |
24f211ba CH |
95 | ip->i_delayed_blks = 0; |
96 | memset(&ip->i_d, 0, sizeof(xfs_icdinode_t)); | |
97 | ip->i_size = 0; | |
98 | ip->i_new_size = 0; | |
99 | ||
24f211ba CH |
100 | return ip; |
101 | } | |
1da177e4 | 102 | |
fa0d7e3d NP |
103 | STATIC void |
104 | xfs_inode_free_callback( | |
105 | struct rcu_head *head) | |
106 | { | |
107 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
108 | struct xfs_inode *ip = XFS_I(inode); | |
109 | ||
110 | INIT_LIST_HEAD(&inode->i_dentry); | |
111 | kmem_zone_free(xfs_inode_zone, ip); | |
112 | } | |
113 | ||
2f11feab | 114 | void |
b36ec042 CH |
115 | xfs_inode_free( |
116 | struct xfs_inode *ip) | |
117 | { | |
118 | switch (ip->i_d.di_mode & S_IFMT) { | |
119 | case S_IFREG: | |
120 | case S_IFDIR: | |
121 | case S_IFLNK: | |
122 | xfs_idestroy_fork(ip, XFS_DATA_FORK); | |
123 | break; | |
124 | } | |
125 | ||
126 | if (ip->i_afp) | |
127 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
128 | ||
b36ec042 CH |
129 | if (ip->i_itemp) { |
130 | /* | |
131 | * Only if we are shutting down the fs will we see an | |
132 | * inode still in the AIL. If it is there, we should remove | |
133 | * it to prevent a use-after-free from occurring. | |
134 | */ | |
135 | xfs_log_item_t *lip = &ip->i_itemp->ili_item; | |
136 | struct xfs_ail *ailp = lip->li_ailp; | |
137 | ||
138 | ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) || | |
139 | XFS_FORCED_SHUTDOWN(ip->i_mount)); | |
140 | if (lip->li_flags & XFS_LI_IN_AIL) { | |
141 | spin_lock(&ailp->xa_lock); | |
142 | if (lip->li_flags & XFS_LI_IN_AIL) | |
143 | xfs_trans_ail_delete(ailp, lip); | |
144 | else | |
145 | spin_unlock(&ailp->xa_lock); | |
146 | } | |
147 | xfs_inode_item_destroy(ip); | |
148 | ip->i_itemp = NULL; | |
149 | } | |
150 | ||
151 | /* asserts to verify all state is correct here */ | |
b36ec042 CH |
152 | ASSERT(atomic_read(&ip->i_pincount) == 0); |
153 | ASSERT(!spin_is_locked(&ip->i_flags_lock)); | |
154 | ASSERT(completion_done(&ip->i_flush)); | |
155 | ||
1a3e8f3d DC |
156 | /* |
157 | * Because we use RCU freeing we need to ensure the inode always | |
158 | * appears to be reclaimed with an invalid inode number when in the | |
159 | * free state. The ip->i_flags_lock provides the barrier against lookup | |
160 | * races. | |
161 | */ | |
162 | spin_lock(&ip->i_flags_lock); | |
163 | ip->i_flags = XFS_IRECLAIM; | |
164 | ip->i_ino = 0; | |
165 | spin_unlock(&ip->i_flags_lock); | |
92f1c008 AE |
166 | |
167 | call_rcu(&VFS_I(ip)->i_rcu, xfs_inode_free_callback); | |
b36ec042 CH |
168 | } |
169 | ||
1da177e4 | 170 | /* |
6441e549 | 171 | * Check the validity of the inode we just found it the cache |
1da177e4 | 172 | */ |
6441e549 DC |
173 | static int |
174 | xfs_iget_cache_hit( | |
6441e549 DC |
175 | struct xfs_perag *pag, |
176 | struct xfs_inode *ip, | |
1a3e8f3d | 177 | xfs_ino_t ino, |
6441e549 | 178 | int flags, |
1a3e8f3d | 179 | int lock_flags) __releases(RCU) |
1da177e4 | 180 | { |
bc990f5c | 181 | struct inode *inode = VFS_I(ip); |
6441e549 | 182 | struct xfs_mount *mp = ip->i_mount; |
bc990f5c CH |
183 | int error; |
184 | ||
1a3e8f3d DC |
185 | /* |
186 | * check for re-use of an inode within an RCU grace period due to the | |
187 | * radix tree nodes not being updated yet. We monitor for this by | |
188 | * setting the inode number to zero before freeing the inode structure. | |
189 | * If the inode has been reallocated and set up, then the inode number | |
190 | * will not match, so check for that, too. | |
191 | */ | |
bc990f5c | 192 | spin_lock(&ip->i_flags_lock); |
1a3e8f3d DC |
193 | if (ip->i_ino != ino) { |
194 | trace_xfs_iget_skip(ip); | |
195 | XFS_STATS_INC(xs_ig_frecycle); | |
196 | error = EAGAIN; | |
197 | goto out_error; | |
198 | } | |
199 | ||
da353b0d | 200 | |
6441e549 | 201 | /* |
bc990f5c CH |
202 | * If we are racing with another cache hit that is currently |
203 | * instantiating this inode or currently recycling it out of | |
204 | * reclaimabe state, wait for the initialisation to complete | |
205 | * before continuing. | |
206 | * | |
207 | * XXX(hch): eventually we should do something equivalent to | |
208 | * wait_on_inode to wait for these flags to be cleared | |
209 | * instead of polling for it. | |
6441e549 | 210 | */ |
bc990f5c | 211 | if (ip->i_flags & (XFS_INEW|XFS_IRECLAIM)) { |
0b1b213f | 212 | trace_xfs_iget_skip(ip); |
6441e549 | 213 | XFS_STATS_INC(xs_ig_frecycle); |
bc990f5c | 214 | error = EAGAIN; |
6441e549 DC |
215 | goto out_error; |
216 | } | |
da353b0d | 217 | |
bc990f5c CH |
218 | /* |
219 | * If lookup is racing with unlink return an error immediately. | |
220 | */ | |
221 | if (ip->i_d.di_mode == 0 && !(flags & XFS_IGET_CREATE)) { | |
222 | error = ENOENT; | |
223 | goto out_error; | |
224 | } | |
bf904248 | 225 | |
bc990f5c CH |
226 | /* |
227 | * If IRECLAIMABLE is set, we've torn down the VFS inode already. | |
228 | * Need to carefully get it back into useable state. | |
229 | */ | |
230 | if (ip->i_flags & XFS_IRECLAIMABLE) { | |
0b1b213f | 231 | trace_xfs_iget_reclaim(ip); |
da353b0d | 232 | |
6441e549 | 233 | /* |
f1f724e4 CH |
234 | * We need to set XFS_IRECLAIM to prevent xfs_reclaim_inode |
235 | * from stomping over us while we recycle the inode. We can't | |
236 | * clear the radix tree reclaimable tag yet as it requires | |
237 | * pag_ici_lock to be held exclusive. | |
6441e549 | 238 | */ |
f1f724e4 | 239 | ip->i_flags |= XFS_IRECLAIM; |
bc990f5c CH |
240 | |
241 | spin_unlock(&ip->i_flags_lock); | |
1a3e8f3d | 242 | rcu_read_unlock(); |
bc990f5c CH |
243 | |
244 | error = -inode_init_always(mp->m_super, inode); | |
245 | if (error) { | |
246 | /* | |
247 | * Re-initializing the inode failed, and we are in deep | |
248 | * trouble. Try to re-add it to the reclaim list. | |
249 | */ | |
1a3e8f3d | 250 | rcu_read_lock(); |
bc990f5c CH |
251 | spin_lock(&ip->i_flags_lock); |
252 | ||
778e24bb DC |
253 | ip->i_flags &= ~(XFS_INEW | XFS_IRECLAIM); |
254 | ASSERT(ip->i_flags & XFS_IRECLAIMABLE); | |
d2e078c3 | 255 | trace_xfs_iget_reclaim_fail(ip); |
6441e549 | 256 | goto out_error; |
da353b0d | 257 | } |
f1f724e4 | 258 | |
1a427ab0 | 259 | spin_lock(&pag->pag_ici_lock); |
f1f724e4 | 260 | spin_lock(&ip->i_flags_lock); |
778e24bb DC |
261 | |
262 | /* | |
263 | * Clear the per-lifetime state in the inode as we are now | |
264 | * effectively a new inode and need to return to the initial | |
265 | * state before reuse occurs. | |
266 | */ | |
267 | ip->i_flags &= ~XFS_IRECLAIM_RESET_FLAGS; | |
f1f724e4 CH |
268 | ip->i_flags |= XFS_INEW; |
269 | __xfs_inode_clear_reclaim_tag(mp, pag, ip); | |
eaff8079 | 270 | inode->i_state = I_NEW; |
dcfcf205 DC |
271 | |
272 | ASSERT(!rwsem_is_locked(&ip->i_iolock.mr_lock)); | |
273 | mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino); | |
274 | lockdep_set_class_and_name(&ip->i_iolock.mr_lock, | |
275 | &xfs_iolock_active, "xfs_iolock_active"); | |
276 | ||
f1f724e4 | 277 | spin_unlock(&ip->i_flags_lock); |
1a427ab0 | 278 | spin_unlock(&pag->pag_ici_lock); |
bc990f5c | 279 | } else { |
bf904248 | 280 | /* If the VFS inode is being torn down, pause and try again. */ |
bc990f5c | 281 | if (!igrab(inode)) { |
d2e078c3 | 282 | trace_xfs_iget_skip(ip); |
bc990f5c CH |
283 | error = EAGAIN; |
284 | goto out_error; | |
285 | } | |
1da177e4 | 286 | |
bc990f5c CH |
287 | /* We've got a live one. */ |
288 | spin_unlock(&ip->i_flags_lock); | |
1a3e8f3d | 289 | rcu_read_unlock(); |
d2e078c3 | 290 | trace_xfs_iget_hit(ip); |
6441e549 | 291 | } |
da353b0d | 292 | |
6441e549 DC |
293 | if (lock_flags != 0) |
294 | xfs_ilock(ip, lock_flags); | |
da353b0d | 295 | |
6441e549 | 296 | xfs_iflags_clear(ip, XFS_ISTALE); |
6441e549 | 297 | XFS_STATS_INC(xs_ig_found); |
0b1b213f | 298 | |
6441e549 | 299 | return 0; |
1da177e4 | 300 | |
6441e549 | 301 | out_error: |
bc990f5c | 302 | spin_unlock(&ip->i_flags_lock); |
1a3e8f3d | 303 | rcu_read_unlock(); |
6441e549 DC |
304 | return error; |
305 | } | |
306 | ||
307 | ||
308 | static int | |
309 | xfs_iget_cache_miss( | |
310 | struct xfs_mount *mp, | |
311 | struct xfs_perag *pag, | |
312 | xfs_trans_t *tp, | |
313 | xfs_ino_t ino, | |
314 | struct xfs_inode **ipp, | |
6441e549 | 315 | int flags, |
0c3dc2b0 | 316 | int lock_flags) |
6441e549 DC |
317 | { |
318 | struct xfs_inode *ip; | |
319 | int error; | |
6441e549 | 320 | xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino); |
1da177e4 | 321 | |
24f211ba CH |
322 | ip = xfs_inode_alloc(mp, ino); |
323 | if (!ip) | |
324 | return ENOMEM; | |
325 | ||
7b6259e7 | 326 | error = xfs_iread(mp, tp, ip, flags); |
6441e549 | 327 | if (error) |
24f211ba | 328 | goto out_destroy; |
1da177e4 | 329 | |
d2e078c3 | 330 | trace_xfs_iget_miss(ip); |
1da177e4 | 331 | |
745b1f47 | 332 | if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) { |
6441e549 DC |
333 | error = ENOENT; |
334 | goto out_destroy; | |
1da177e4 LT |
335 | } |
336 | ||
337 | /* | |
bad55843 | 338 | * Preload the radix tree so we can insert safely under the |
56e73ec4 DC |
339 | * write spinlock. Note that we cannot sleep inside the preload |
340 | * region. | |
1da177e4 | 341 | */ |
da353b0d | 342 | if (radix_tree_preload(GFP_KERNEL)) { |
6441e549 | 343 | error = EAGAIN; |
ed93ec39 CH |
344 | goto out_destroy; |
345 | } | |
346 | ||
347 | /* | |
348 | * Because the inode hasn't been added to the radix-tree yet it can't | |
349 | * be found by another thread, so we can do the non-sleeping lock here. | |
350 | */ | |
351 | if (lock_flags) { | |
352 | if (!xfs_ilock_nowait(ip, lock_flags)) | |
353 | BUG(); | |
da353b0d | 354 | } |
f338f903 | 355 | |
1a427ab0 | 356 | spin_lock(&pag->pag_ici_lock); |
6441e549 DC |
357 | |
358 | /* insert the new inode */ | |
da353b0d DC |
359 | error = radix_tree_insert(&pag->pag_ici_root, agino, ip); |
360 | if (unlikely(error)) { | |
6441e549 | 361 | WARN_ON(error != -EEXIST); |
da353b0d | 362 | XFS_STATS_INC(xs_ig_dup); |
6441e549 | 363 | error = EAGAIN; |
56e73ec4 | 364 | goto out_preload_end; |
1da177e4 LT |
365 | } |
366 | ||
6441e549 | 367 | /* These values _must_ be set before releasing the radix tree lock! */ |
1da177e4 | 368 | ip->i_udquot = ip->i_gdquot = NULL; |
7a18c386 | 369 | xfs_iflags_set(ip, XFS_INEW); |
1da177e4 | 370 | |
1a427ab0 | 371 | spin_unlock(&pag->pag_ici_lock); |
da353b0d | 372 | radix_tree_preload_end(); |
0b1b213f | 373 | |
6441e549 DC |
374 | *ipp = ip; |
375 | return 0; | |
376 | ||
56e73ec4 | 377 | out_preload_end: |
1a427ab0 | 378 | spin_unlock(&pag->pag_ici_lock); |
6441e549 | 379 | radix_tree_preload_end(); |
56e73ec4 DC |
380 | if (lock_flags) |
381 | xfs_iunlock(ip, lock_flags); | |
6441e549 | 382 | out_destroy: |
b36ec042 CH |
383 | __destroy_inode(VFS_I(ip)); |
384 | xfs_inode_free(ip); | |
6441e549 DC |
385 | return error; |
386 | } | |
387 | ||
388 | /* | |
389 | * Look up an inode by number in the given file system. | |
390 | * The inode is looked up in the cache held in each AG. | |
bf904248 DC |
391 | * If the inode is found in the cache, initialise the vfs inode |
392 | * if necessary. | |
6441e549 DC |
393 | * |
394 | * If it is not in core, read it in from the file system's device, | |
bf904248 | 395 | * add it to the cache and initialise the vfs inode. |
6441e549 DC |
396 | * |
397 | * The inode is locked according to the value of the lock_flags parameter. | |
398 | * This flag parameter indicates how and if the inode's IO lock and inode lock | |
399 | * should be taken. | |
400 | * | |
401 | * mp -- the mount point structure for the current file system. It points | |
402 | * to the inode hash table. | |
403 | * tp -- a pointer to the current transaction if there is one. This is | |
404 | * simply passed through to the xfs_iread() call. | |
405 | * ino -- the number of the inode desired. This is the unique identifier | |
406 | * within the file system for the inode being requested. | |
407 | * lock_flags -- flags indicating how to lock the inode. See the comment | |
408 | * for xfs_ilock() for a list of valid values. | |
6441e549 | 409 | */ |
bf904248 DC |
410 | int |
411 | xfs_iget( | |
6441e549 DC |
412 | xfs_mount_t *mp, |
413 | xfs_trans_t *tp, | |
414 | xfs_ino_t ino, | |
415 | uint flags, | |
416 | uint lock_flags, | |
7b6259e7 | 417 | xfs_inode_t **ipp) |
6441e549 DC |
418 | { |
419 | xfs_inode_t *ip; | |
420 | int error; | |
421 | xfs_perag_t *pag; | |
422 | xfs_agino_t agino; | |
423 | ||
d276734d | 424 | /* reject inode numbers outside existing AGs */ |
1a3e8f3d | 425 | if (!ino || XFS_INO_TO_AGNO(mp, ino) >= mp->m_sb.sb_agcount) |
6441e549 DC |
426 | return EINVAL; |
427 | ||
428 | /* get the perag structure and ensure that it's inode capable */ | |
5017e97d | 429 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ino)); |
6441e549 DC |
430 | agino = XFS_INO_TO_AGINO(mp, ino); |
431 | ||
432 | again: | |
433 | error = 0; | |
1a3e8f3d | 434 | rcu_read_lock(); |
6441e549 DC |
435 | ip = radix_tree_lookup(&pag->pag_ici_root, agino); |
436 | ||
437 | if (ip) { | |
1a3e8f3d | 438 | error = xfs_iget_cache_hit(pag, ip, ino, flags, lock_flags); |
6441e549 DC |
439 | if (error) |
440 | goto out_error_or_again; | |
441 | } else { | |
1a3e8f3d | 442 | rcu_read_unlock(); |
6441e549 DC |
443 | XFS_STATS_INC(xs_ig_missed); |
444 | ||
7b6259e7 | 445 | error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip, |
6441e549 DC |
446 | flags, lock_flags); |
447 | if (error) | |
448 | goto out_error_or_again; | |
449 | } | |
5017e97d | 450 | xfs_perag_put(pag); |
1da177e4 | 451 | |
1da177e4 LT |
452 | *ipp = ip; |
453 | ||
bf904248 DC |
454 | ASSERT(ip->i_df.if_ext_max == |
455 | XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t)); | |
1da177e4 LT |
456 | /* |
457 | * If we have a real type for an on-disk inode, we can set ops(&unlock) | |
458 | * now. If it's a new inode being created, xfs_ialloc will handle it. | |
459 | */ | |
bf904248 | 460 | if (xfs_iflags_test(ip, XFS_INEW) && ip->i_d.di_mode != 0) |
41be8bed | 461 | xfs_setup_inode(ip); |
1da177e4 | 462 | return 0; |
6441e549 DC |
463 | |
464 | out_error_or_again: | |
465 | if (error == EAGAIN) { | |
466 | delay(1); | |
467 | goto again; | |
468 | } | |
5017e97d | 469 | xfs_perag_put(pag); |
6441e549 | 470 | return error; |
1da177e4 LT |
471 | } |
472 | ||
1da177e4 LT |
473 | /* |
474 | * This is a wrapper routine around the xfs_ilock() routine | |
475 | * used to centralize some grungy code. It is used in places | |
476 | * that wish to lock the inode solely for reading the extents. | |
477 | * The reason these places can't just call xfs_ilock(SHARED) | |
478 | * is that the inode lock also guards to bringing in of the | |
479 | * extents from disk for a file in b-tree format. If the inode | |
480 | * is in b-tree format, then we need to lock the inode exclusively | |
481 | * until the extents are read in. Locking it exclusively all | |
482 | * the time would limit our parallelism unnecessarily, though. | |
483 | * What we do instead is check to see if the extents have been | |
484 | * read in yet, and only lock the inode exclusively if they | |
485 | * have not. | |
486 | * | |
487 | * The function returns a value which should be given to the | |
488 | * corresponding xfs_iunlock_map_shared(). This value is | |
489 | * the mode in which the lock was actually taken. | |
490 | */ | |
491 | uint | |
492 | xfs_ilock_map_shared( | |
493 | xfs_inode_t *ip) | |
494 | { | |
495 | uint lock_mode; | |
496 | ||
497 | if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) && | |
498 | ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) { | |
499 | lock_mode = XFS_ILOCK_EXCL; | |
500 | } else { | |
501 | lock_mode = XFS_ILOCK_SHARED; | |
502 | } | |
503 | ||
504 | xfs_ilock(ip, lock_mode); | |
505 | ||
506 | return lock_mode; | |
507 | } | |
508 | ||
509 | /* | |
510 | * This is simply the unlock routine to go with xfs_ilock_map_shared(). | |
511 | * All it does is call xfs_iunlock() with the given lock_mode. | |
512 | */ | |
513 | void | |
514 | xfs_iunlock_map_shared( | |
515 | xfs_inode_t *ip, | |
516 | unsigned int lock_mode) | |
517 | { | |
518 | xfs_iunlock(ip, lock_mode); | |
519 | } | |
520 | ||
521 | /* | |
522 | * The xfs inode contains 2 locks: a multi-reader lock called the | |
523 | * i_iolock and a multi-reader lock called the i_lock. This routine | |
524 | * allows either or both of the locks to be obtained. | |
525 | * | |
526 | * The 2 locks should always be ordered so that the IO lock is | |
527 | * obtained first in order to prevent deadlock. | |
528 | * | |
529 | * ip -- the inode being locked | |
530 | * lock_flags -- this parameter indicates the inode's locks | |
531 | * to be locked. It can be: | |
532 | * XFS_IOLOCK_SHARED, | |
533 | * XFS_IOLOCK_EXCL, | |
534 | * XFS_ILOCK_SHARED, | |
535 | * XFS_ILOCK_EXCL, | |
536 | * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED, | |
537 | * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL, | |
538 | * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED, | |
539 | * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL | |
540 | */ | |
541 | void | |
579aa9ca CH |
542 | xfs_ilock( |
543 | xfs_inode_t *ip, | |
544 | uint lock_flags) | |
1da177e4 LT |
545 | { |
546 | /* | |
547 | * You can't set both SHARED and EXCL for the same lock, | |
548 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
549 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
550 | */ | |
551 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
552 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
553 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
554 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
f7c66ce3 | 555 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); |
1da177e4 | 556 | |
579aa9ca | 557 | if (lock_flags & XFS_IOLOCK_EXCL) |
f7c66ce3 | 558 | mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); |
579aa9ca | 559 | else if (lock_flags & XFS_IOLOCK_SHARED) |
f7c66ce3 | 560 | mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); |
579aa9ca CH |
561 | |
562 | if (lock_flags & XFS_ILOCK_EXCL) | |
f7c66ce3 | 563 | mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); |
579aa9ca | 564 | else if (lock_flags & XFS_ILOCK_SHARED) |
f7c66ce3 | 565 | mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); |
579aa9ca | 566 | |
0b1b213f | 567 | trace_xfs_ilock(ip, lock_flags, _RET_IP_); |
1da177e4 LT |
568 | } |
569 | ||
570 | /* | |
571 | * This is just like xfs_ilock(), except that the caller | |
572 | * is guaranteed not to sleep. It returns 1 if it gets | |
573 | * the requested locks and 0 otherwise. If the IO lock is | |
574 | * obtained but the inode lock cannot be, then the IO lock | |
575 | * is dropped before returning. | |
576 | * | |
577 | * ip -- the inode being locked | |
578 | * lock_flags -- this parameter indicates the inode's locks to be | |
579 | * to be locked. See the comment for xfs_ilock() for a list | |
580 | * of valid values. | |
1da177e4 LT |
581 | */ |
582 | int | |
579aa9ca CH |
583 | xfs_ilock_nowait( |
584 | xfs_inode_t *ip, | |
585 | uint lock_flags) | |
1da177e4 | 586 | { |
1da177e4 LT |
587 | /* |
588 | * You can't set both SHARED and EXCL for the same lock, | |
589 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
590 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
591 | */ | |
592 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
593 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
594 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
595 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
f7c66ce3 | 596 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); |
1da177e4 | 597 | |
1da177e4 | 598 | if (lock_flags & XFS_IOLOCK_EXCL) { |
579aa9ca CH |
599 | if (!mrtryupdate(&ip->i_iolock)) |
600 | goto out; | |
1da177e4 | 601 | } else if (lock_flags & XFS_IOLOCK_SHARED) { |
579aa9ca CH |
602 | if (!mrtryaccess(&ip->i_iolock)) |
603 | goto out; | |
1da177e4 LT |
604 | } |
605 | if (lock_flags & XFS_ILOCK_EXCL) { | |
579aa9ca CH |
606 | if (!mrtryupdate(&ip->i_lock)) |
607 | goto out_undo_iolock; | |
1da177e4 | 608 | } else if (lock_flags & XFS_ILOCK_SHARED) { |
579aa9ca CH |
609 | if (!mrtryaccess(&ip->i_lock)) |
610 | goto out_undo_iolock; | |
1da177e4 | 611 | } |
0b1b213f | 612 | trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_); |
1da177e4 | 613 | return 1; |
579aa9ca CH |
614 | |
615 | out_undo_iolock: | |
616 | if (lock_flags & XFS_IOLOCK_EXCL) | |
617 | mrunlock_excl(&ip->i_iolock); | |
618 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
619 | mrunlock_shared(&ip->i_iolock); | |
620 | out: | |
621 | return 0; | |
1da177e4 LT |
622 | } |
623 | ||
624 | /* | |
625 | * xfs_iunlock() is used to drop the inode locks acquired with | |
626 | * xfs_ilock() and xfs_ilock_nowait(). The caller must pass | |
627 | * in the flags given to xfs_ilock() or xfs_ilock_nowait() so | |
628 | * that we know which locks to drop. | |
629 | * | |
630 | * ip -- the inode being unlocked | |
631 | * lock_flags -- this parameter indicates the inode's locks to be | |
632 | * to be unlocked. See the comment for xfs_ilock() for a list | |
633 | * of valid values for this parameter. | |
634 | * | |
635 | */ | |
636 | void | |
579aa9ca CH |
637 | xfs_iunlock( |
638 | xfs_inode_t *ip, | |
639 | uint lock_flags) | |
1da177e4 LT |
640 | { |
641 | /* | |
642 | * You can't set both SHARED and EXCL for the same lock, | |
643 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
644 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
645 | */ | |
646 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
647 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
648 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
649 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
f7c66ce3 LM |
650 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_IUNLOCK_NONOTIFY | |
651 | XFS_LOCK_DEP_MASK)) == 0); | |
1da177e4 LT |
652 | ASSERT(lock_flags != 0); |
653 | ||
579aa9ca CH |
654 | if (lock_flags & XFS_IOLOCK_EXCL) |
655 | mrunlock_excl(&ip->i_iolock); | |
656 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
657 | mrunlock_shared(&ip->i_iolock); | |
1da177e4 | 658 | |
579aa9ca CH |
659 | if (lock_flags & XFS_ILOCK_EXCL) |
660 | mrunlock_excl(&ip->i_lock); | |
661 | else if (lock_flags & XFS_ILOCK_SHARED) | |
662 | mrunlock_shared(&ip->i_lock); | |
1da177e4 | 663 | |
579aa9ca CH |
664 | if ((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) && |
665 | !(lock_flags & XFS_IUNLOCK_NONOTIFY) && ip->i_itemp) { | |
1da177e4 LT |
666 | /* |
667 | * Let the AIL know that this item has been unlocked in case | |
668 | * it is in the AIL and anyone is waiting on it. Don't do | |
669 | * this if the caller has asked us not to. | |
670 | */ | |
783a2f65 | 671 | xfs_trans_unlocked_item(ip->i_itemp->ili_item.li_ailp, |
579aa9ca | 672 | (xfs_log_item_t*)(ip->i_itemp)); |
1da177e4 | 673 | } |
0b1b213f | 674 | trace_xfs_iunlock(ip, lock_flags, _RET_IP_); |
1da177e4 LT |
675 | } |
676 | ||
677 | /* | |
678 | * give up write locks. the i/o lock cannot be held nested | |
679 | * if it is being demoted. | |
680 | */ | |
681 | void | |
579aa9ca CH |
682 | xfs_ilock_demote( |
683 | xfs_inode_t *ip, | |
684 | uint lock_flags) | |
1da177e4 LT |
685 | { |
686 | ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)); | |
687 | ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0); | |
688 | ||
579aa9ca | 689 | if (lock_flags & XFS_ILOCK_EXCL) |
1da177e4 | 690 | mrdemote(&ip->i_lock); |
579aa9ca | 691 | if (lock_flags & XFS_IOLOCK_EXCL) |
1da177e4 | 692 | mrdemote(&ip->i_iolock); |
0b1b213f CH |
693 | |
694 | trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_); | |
579aa9ca CH |
695 | } |
696 | ||
697 | #ifdef DEBUG | |
579aa9ca CH |
698 | int |
699 | xfs_isilocked( | |
700 | xfs_inode_t *ip, | |
701 | uint lock_flags) | |
702 | { | |
f9369729 CH |
703 | if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) { |
704 | if (!(lock_flags & XFS_ILOCK_SHARED)) | |
705 | return !!ip->i_lock.mr_writer; | |
706 | return rwsem_is_locked(&ip->i_lock.mr_lock); | |
1da177e4 | 707 | } |
579aa9ca | 708 | |
f9369729 CH |
709 | if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) { |
710 | if (!(lock_flags & XFS_IOLOCK_SHARED)) | |
711 | return !!ip->i_iolock.mr_writer; | |
712 | return rwsem_is_locked(&ip->i_iolock.mr_lock); | |
579aa9ca CH |
713 | } |
714 | ||
f9369729 CH |
715 | ASSERT(0); |
716 | return 0; | |
1da177e4 | 717 | } |
579aa9ca | 718 | #endif |