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" |
a844f451 | 21 | #include "xfs_bit.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 LT |
27 | #include "xfs_dir2.h" |
28 | #include "xfs_dmapi.h" | |
29 | #include "xfs_mount.h" | |
1da177e4 | 30 | #include "xfs_bmap_btree.h" |
a844f451 | 31 | #include "xfs_alloc_btree.h" |
1da177e4 | 32 | #include "xfs_ialloc_btree.h" |
1da177e4 | 33 | #include "xfs_dir2_sf.h" |
a844f451 | 34 | #include "xfs_attr_sf.h" |
1da177e4 LT |
35 | #include "xfs_dinode.h" |
36 | #include "xfs_inode.h" | |
a844f451 NS |
37 | #include "xfs_btree.h" |
38 | #include "xfs_ialloc.h" | |
1da177e4 LT |
39 | #include "xfs_quota.h" |
40 | #include "xfs_utils.h" | |
783a2f65 DC |
41 | #include "xfs_trans_priv.h" |
42 | #include "xfs_inode_item.h" | |
24f211ba CH |
43 | #include "xfs_bmap.h" |
44 | #include "xfs_btree_trace.h" | |
45 | #include "xfs_dir2_trace.h" | |
46 | ||
47 | ||
48 | /* | |
49 | * Allocate and initialise an xfs_inode. | |
50 | */ | |
51 | STATIC struct xfs_inode * | |
52 | xfs_inode_alloc( | |
53 | struct xfs_mount *mp, | |
54 | xfs_ino_t ino) | |
55 | { | |
56 | struct xfs_inode *ip; | |
57 | ||
58 | /* | |
59 | * if this didn't occur in transactions, we could use | |
60 | * KM_MAYFAIL and return NULL here on ENOMEM. Set the | |
61 | * code up to do this anyway. | |
62 | */ | |
63 | ip = kmem_zone_alloc(xfs_inode_zone, KM_SLEEP); | |
64 | if (!ip) | |
65 | return NULL; | |
66 | ||
67 | ASSERT(atomic_read(&ip->i_iocount) == 0); | |
68 | ASSERT(atomic_read(&ip->i_pincount) == 0); | |
69 | ASSERT(!spin_is_locked(&ip->i_flags_lock)); | |
70 | ASSERT(completion_done(&ip->i_flush)); | |
71 | ||
72 | /* | |
73 | * initialise the VFS inode here to get failures | |
74 | * out of the way early. | |
75 | */ | |
76 | if (!inode_init_always(mp->m_super, VFS_I(ip))) { | |
77 | kmem_zone_free(xfs_inode_zone, ip); | |
78 | return NULL; | |
79 | } | |
80 | ||
81 | /* initialise the xfs inode */ | |
82 | ip->i_ino = ino; | |
83 | ip->i_mount = mp; | |
84 | memset(&ip->i_imap, 0, sizeof(struct xfs_imap)); | |
85 | ip->i_afp = NULL; | |
86 | memset(&ip->i_df, 0, sizeof(xfs_ifork_t)); | |
87 | ip->i_flags = 0; | |
88 | ip->i_update_core = 0; | |
89 | ip->i_update_size = 0; | |
90 | ip->i_delayed_blks = 0; | |
91 | memset(&ip->i_d, 0, sizeof(xfs_icdinode_t)); | |
92 | ip->i_size = 0; | |
93 | ip->i_new_size = 0; | |
94 | ||
95 | /* | |
96 | * Initialize inode's trace buffers. | |
97 | */ | |
98 | #ifdef XFS_INODE_TRACE | |
99 | ip->i_trace = ktrace_alloc(INODE_TRACE_SIZE, KM_NOFS); | |
100 | #endif | |
101 | #ifdef XFS_BMAP_TRACE | |
102 | ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_NOFS); | |
103 | #endif | |
104 | #ifdef XFS_BTREE_TRACE | |
105 | ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_NOFS); | |
106 | #endif | |
107 | #ifdef XFS_RW_TRACE | |
108 | ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_NOFS); | |
109 | #endif | |
110 | #ifdef XFS_ILOCK_TRACE | |
111 | ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_NOFS); | |
112 | #endif | |
113 | #ifdef XFS_DIR2_TRACE | |
114 | ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_NOFS); | |
115 | #endif | |
116 | ||
117 | return ip; | |
118 | } | |
1da177e4 | 119 | |
1da177e4 | 120 | /* |
6441e549 | 121 | * Check the validity of the inode we just found it the cache |
1da177e4 | 122 | */ |
6441e549 DC |
123 | static int |
124 | xfs_iget_cache_hit( | |
6441e549 DC |
125 | struct xfs_perag *pag, |
126 | struct xfs_inode *ip, | |
127 | int flags, | |
128 | int lock_flags) __releases(pag->pag_ici_lock) | |
1da177e4 | 129 | { |
6441e549 | 130 | struct xfs_mount *mp = ip->i_mount; |
6bfb3d06 | 131 | int error = EAGAIN; |
da353b0d | 132 | |
6441e549 DC |
133 | /* |
134 | * If INEW is set this inode is being set up | |
bf904248 | 135 | * If IRECLAIM is set this inode is being torn down |
6441e549 DC |
136 | * Pause and try again. |
137 | */ | |
bf904248 | 138 | if (xfs_iflags_test(ip, (XFS_INEW|XFS_IRECLAIM))) { |
6441e549 DC |
139 | XFS_STATS_INC(xs_ig_frecycle); |
140 | goto out_error; | |
141 | } | |
da353b0d | 142 | |
bf904248 DC |
143 | /* If IRECLAIMABLE is set, we've torn down the vfs inode part */ |
144 | if (xfs_iflags_test(ip, XFS_IRECLAIMABLE)) { | |
145 | ||
da353b0d | 146 | /* |
bf904248 DC |
147 | * If lookup is racing with unlink, then we should return an |
148 | * error immediately so we don't remove it from the reclaim | |
149 | * list and potentially leak the inode. | |
da353b0d | 150 | */ |
bf904248 DC |
151 | if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) { |
152 | error = ENOENT; | |
6441e549 DC |
153 | goto out_error; |
154 | } | |
bf904248 DC |
155 | |
156 | xfs_itrace_exit_tag(ip, "xfs_iget.alloc"); | |
da353b0d | 157 | |
6441e549 | 158 | /* |
bf904248 DC |
159 | * We need to re-initialise the VFS inode as it has been |
160 | * 'freed' by the VFS. Do this here so we can deal with | |
161 | * errors cleanly, then tag it so it can be set up correctly | |
162 | * later. | |
6441e549 | 163 | */ |
bf904248 DC |
164 | if (!inode_init_always(mp->m_super, VFS_I(ip))) { |
165 | error = ENOMEM; | |
6441e549 | 166 | goto out_error; |
da353b0d | 167 | } |
6bfb3d06 DC |
168 | |
169 | /* | |
170 | * We must set the XFS_INEW flag before clearing the | |
171 | * XFS_IRECLAIMABLE flag so that if a racing lookup does | |
172 | * not find the XFS_IRECLAIMABLE above but has the igrab() | |
173 | * below succeed we can safely check XFS_INEW to detect | |
174 | * that this inode is still being initialised. | |
175 | */ | |
bf904248 | 176 | xfs_iflags_set(ip, XFS_INEW); |
6441e549 | 177 | xfs_iflags_clear(ip, XFS_IRECLAIMABLE); |
396beb85 DC |
178 | |
179 | /* clear the radix tree reclaim flag as well. */ | |
180 | __xfs_inode_clear_reclaim_tag(mp, pag, ip); | |
bf904248 DC |
181 | } else if (!igrab(VFS_I(ip))) { |
182 | /* If the VFS inode is being torn down, pause and try again. */ | |
bf904248 DC |
183 | XFS_STATS_INC(xs_ig_frecycle); |
184 | goto out_error; | |
6bfb3d06 DC |
185 | } else if (xfs_iflags_test(ip, XFS_INEW)) { |
186 | /* | |
187 | * We are racing with another cache hit that is | |
188 | * currently recycling this inode out of the XFS_IRECLAIMABLE | |
189 | * state. Wait for the initialisation to complete before | |
190 | * continuing. | |
191 | */ | |
192 | wait_on_inode(VFS_I(ip)); | |
6441e549 | 193 | } |
1da177e4 | 194 | |
6441e549 DC |
195 | if (ip->i_d.di_mode == 0 && !(flags & XFS_IGET_CREATE)) { |
196 | error = ENOENT; | |
6bfb3d06 DC |
197 | iput(VFS_I(ip)); |
198 | goto out_error; | |
6441e549 | 199 | } |
da353b0d | 200 | |
6bfb3d06 DC |
201 | /* We've got a live one. */ |
202 | read_unlock(&pag->pag_ici_lock); | |
203 | ||
6441e549 DC |
204 | if (lock_flags != 0) |
205 | xfs_ilock(ip, lock_flags); | |
da353b0d | 206 | |
6441e549 DC |
207 | xfs_iflags_clear(ip, XFS_ISTALE); |
208 | xfs_itrace_exit_tag(ip, "xfs_iget.found"); | |
209 | XFS_STATS_INC(xs_ig_found); | |
210 | return 0; | |
1da177e4 | 211 | |
6441e549 | 212 | out_error: |
da353b0d | 213 | read_unlock(&pag->pag_ici_lock); |
6441e549 DC |
214 | return error; |
215 | } | |
216 | ||
217 | ||
218 | static int | |
219 | xfs_iget_cache_miss( | |
220 | struct xfs_mount *mp, | |
221 | struct xfs_perag *pag, | |
222 | xfs_trans_t *tp, | |
223 | xfs_ino_t ino, | |
224 | struct xfs_inode **ipp, | |
225 | xfs_daddr_t bno, | |
226 | int flags, | |
227 | int lock_flags) __releases(pag->pag_ici_lock) | |
228 | { | |
229 | struct xfs_inode *ip; | |
230 | int error; | |
231 | unsigned long first_index, mask; | |
232 | xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino); | |
1da177e4 | 233 | |
24f211ba CH |
234 | ip = xfs_inode_alloc(mp, ino); |
235 | if (!ip) | |
236 | return ENOMEM; | |
237 | ||
238 | error = xfs_iread(mp, tp, ip, bno, flags); | |
6441e549 | 239 | if (error) |
24f211ba | 240 | goto out_destroy; |
1da177e4 | 241 | |
15947f2d | 242 | xfs_itrace_exit_tag(ip, "xfs_iget.alloc"); |
1da177e4 | 243 | |
745b1f47 | 244 | if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) { |
6441e549 DC |
245 | error = ENOENT; |
246 | goto out_destroy; | |
1da177e4 LT |
247 | } |
248 | ||
56e73ec4 DC |
249 | if (lock_flags) |
250 | xfs_ilock(ip, lock_flags); | |
251 | ||
1da177e4 | 252 | /* |
bad55843 | 253 | * Preload the radix tree so we can insert safely under the |
56e73ec4 DC |
254 | * write spinlock. Note that we cannot sleep inside the preload |
255 | * region. | |
1da177e4 | 256 | */ |
da353b0d | 257 | if (radix_tree_preload(GFP_KERNEL)) { |
6441e549 | 258 | error = EAGAIN; |
56e73ec4 | 259 | goto out_unlock; |
da353b0d | 260 | } |
f338f903 | 261 | |
da353b0d DC |
262 | mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1); |
263 | first_index = agino & mask; | |
264 | write_lock(&pag->pag_ici_lock); | |
6441e549 DC |
265 | |
266 | /* insert the new inode */ | |
da353b0d DC |
267 | error = radix_tree_insert(&pag->pag_ici_root, agino, ip); |
268 | if (unlikely(error)) { | |
6441e549 | 269 | WARN_ON(error != -EEXIST); |
da353b0d | 270 | XFS_STATS_INC(xs_ig_dup); |
6441e549 | 271 | error = EAGAIN; |
56e73ec4 | 272 | goto out_preload_end; |
1da177e4 LT |
273 | } |
274 | ||
6441e549 | 275 | /* These values _must_ be set before releasing the radix tree lock! */ |
1da177e4 | 276 | ip->i_udquot = ip->i_gdquot = NULL; |
7a18c386 | 277 | xfs_iflags_set(ip, XFS_INEW); |
1da177e4 | 278 | |
da353b0d DC |
279 | write_unlock(&pag->pag_ici_lock); |
280 | radix_tree_preload_end(); | |
6441e549 DC |
281 | *ipp = ip; |
282 | return 0; | |
283 | ||
56e73ec4 | 284 | out_preload_end: |
6441e549 DC |
285 | write_unlock(&pag->pag_ici_lock); |
286 | radix_tree_preload_end(); | |
56e73ec4 DC |
287 | out_unlock: |
288 | if (lock_flags) | |
289 | xfs_iunlock(ip, lock_flags); | |
6441e549 | 290 | out_destroy: |
9ed0451e | 291 | xfs_destroy_inode(ip); |
6441e549 DC |
292 | return error; |
293 | } | |
294 | ||
295 | /* | |
296 | * Look up an inode by number in the given file system. | |
297 | * The inode is looked up in the cache held in each AG. | |
bf904248 DC |
298 | * If the inode is found in the cache, initialise the vfs inode |
299 | * if necessary. | |
6441e549 DC |
300 | * |
301 | * If it is not in core, read it in from the file system's device, | |
bf904248 | 302 | * add it to the cache and initialise the vfs inode. |
6441e549 DC |
303 | * |
304 | * The inode is locked according to the value of the lock_flags parameter. | |
305 | * This flag parameter indicates how and if the inode's IO lock and inode lock | |
306 | * should be taken. | |
307 | * | |
308 | * mp -- the mount point structure for the current file system. It points | |
309 | * to the inode hash table. | |
310 | * tp -- a pointer to the current transaction if there is one. This is | |
311 | * simply passed through to the xfs_iread() call. | |
312 | * ino -- the number of the inode desired. This is the unique identifier | |
313 | * within the file system for the inode being requested. | |
314 | * lock_flags -- flags indicating how to lock the inode. See the comment | |
315 | * for xfs_ilock() for a list of valid values. | |
316 | * bno -- the block number starting the buffer containing the inode, | |
317 | * if known (as by bulkstat), else 0. | |
318 | */ | |
bf904248 DC |
319 | int |
320 | xfs_iget( | |
6441e549 DC |
321 | xfs_mount_t *mp, |
322 | xfs_trans_t *tp, | |
323 | xfs_ino_t ino, | |
324 | uint flags, | |
325 | uint lock_flags, | |
326 | xfs_inode_t **ipp, | |
327 | xfs_daddr_t bno) | |
328 | { | |
329 | xfs_inode_t *ip; | |
330 | int error; | |
331 | xfs_perag_t *pag; | |
332 | xfs_agino_t agino; | |
333 | ||
334 | /* the radix tree exists only in inode capable AGs */ | |
335 | if (XFS_INO_TO_AGNO(mp, ino) >= mp->m_maxagi) | |
336 | return EINVAL; | |
337 | ||
338 | /* get the perag structure and ensure that it's inode capable */ | |
339 | pag = xfs_get_perag(mp, ino); | |
340 | if (!pag->pagi_inodeok) | |
341 | return EINVAL; | |
342 | ASSERT(pag->pag_ici_init); | |
343 | agino = XFS_INO_TO_AGINO(mp, ino); | |
344 | ||
345 | again: | |
346 | error = 0; | |
347 | read_lock(&pag->pag_ici_lock); | |
348 | ip = radix_tree_lookup(&pag->pag_ici_root, agino); | |
349 | ||
350 | if (ip) { | |
bf904248 | 351 | error = xfs_iget_cache_hit(pag, ip, flags, lock_flags); |
6441e549 DC |
352 | if (error) |
353 | goto out_error_or_again; | |
354 | } else { | |
355 | read_unlock(&pag->pag_ici_lock); | |
356 | XFS_STATS_INC(xs_ig_missed); | |
357 | ||
358 | error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip, bno, | |
359 | flags, lock_flags); | |
360 | if (error) | |
361 | goto out_error_or_again; | |
362 | } | |
da353b0d | 363 | xfs_put_perag(mp, pag); |
1da177e4 | 364 | |
b3aea4ed | 365 | xfs_iflags_set(ip, XFS_IMODIFIED); |
1da177e4 LT |
366 | *ipp = ip; |
367 | ||
bf904248 DC |
368 | ASSERT(ip->i_df.if_ext_max == |
369 | XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t)); | |
1da177e4 LT |
370 | /* |
371 | * If we have a real type for an on-disk inode, we can set ops(&unlock) | |
372 | * now. If it's a new inode being created, xfs_ialloc will handle it. | |
373 | */ | |
bf904248 | 374 | if (xfs_iflags_test(ip, XFS_INEW) && ip->i_d.di_mode != 0) |
41be8bed | 375 | xfs_setup_inode(ip); |
1da177e4 | 376 | return 0; |
6441e549 DC |
377 | |
378 | out_error_or_again: | |
379 | if (error == EAGAIN) { | |
380 | delay(1); | |
381 | goto again; | |
382 | } | |
383 | xfs_put_perag(mp, pag); | |
384 | return error; | |
1da177e4 LT |
385 | } |
386 | ||
387 | ||
1da177e4 LT |
388 | /* |
389 | * Look for the inode corresponding to the given ino in the hash table. | |
390 | * If it is there and its i_transp pointer matches tp, return it. | |
391 | * Otherwise, return NULL. | |
392 | */ | |
393 | xfs_inode_t * | |
394 | xfs_inode_incore(xfs_mount_t *mp, | |
395 | xfs_ino_t ino, | |
396 | xfs_trans_t *tp) | |
397 | { | |
1da177e4 | 398 | xfs_inode_t *ip; |
da353b0d DC |
399 | xfs_perag_t *pag; |
400 | ||
401 | pag = xfs_get_perag(mp, ino); | |
402 | read_lock(&pag->pag_ici_lock); | |
403 | ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ino)); | |
404 | read_unlock(&pag->pag_ici_lock); | |
405 | xfs_put_perag(mp, pag); | |
406 | ||
407 | /* the returned inode must match the transaction */ | |
408 | if (ip && (ip->i_transp != tp)) | |
409 | return NULL; | |
410 | return ip; | |
1da177e4 LT |
411 | } |
412 | ||
413 | /* | |
414 | * Decrement reference count of an inode structure and unlock it. | |
415 | * | |
416 | * ip -- the inode being released | |
417 | * lock_flags -- this parameter indicates the inode's locks to be | |
418 | * to be released. See the comment on xfs_iunlock() for a list | |
419 | * of valid values. | |
420 | */ | |
421 | void | |
422 | xfs_iput(xfs_inode_t *ip, | |
423 | uint lock_flags) | |
424 | { | |
cf441eeb | 425 | xfs_itrace_entry(ip); |
1da177e4 | 426 | xfs_iunlock(ip, lock_flags); |
10090be2 | 427 | IRELE(ip); |
1da177e4 LT |
428 | } |
429 | ||
430 | /* | |
431 | * Special iput for brand-new inodes that are still locked | |
432 | */ | |
433 | void | |
01651646 DC |
434 | xfs_iput_new( |
435 | xfs_inode_t *ip, | |
436 | uint lock_flags) | |
1da177e4 | 437 | { |
01651646 | 438 | struct inode *inode = VFS_I(ip); |
1da177e4 | 439 | |
cf441eeb | 440 | xfs_itrace_entry(ip); |
1da177e4 LT |
441 | |
442 | if ((ip->i_d.di_mode == 0)) { | |
7a18c386 | 443 | ASSERT(!xfs_iflags_test(ip, XFS_IRECLAIMABLE)); |
10090be2 | 444 | make_bad_inode(inode); |
1da177e4 LT |
445 | } |
446 | if (inode->i_state & I_NEW) | |
447 | unlock_new_inode(inode); | |
448 | if (lock_flags) | |
449 | xfs_iunlock(ip, lock_flags); | |
10090be2 | 450 | IRELE(ip); |
1da177e4 LT |
451 | } |
452 | ||
453 | ||
454 | /* | |
455 | * This routine embodies the part of the reclaim code that pulls | |
456 | * the inode from the inode hash table and the mount structure's | |
457 | * inode list. | |
458 | * This should only be called from xfs_reclaim(). | |
459 | */ | |
460 | void | |
461 | xfs_ireclaim(xfs_inode_t *ip) | |
462 | { | |
1da177e4 LT |
463 | /* |
464 | * Remove from old hash list and mount list. | |
465 | */ | |
466 | XFS_STATS_INC(xs_ig_reclaims); | |
467 | ||
468 | xfs_iextract(ip); | |
469 | ||
470 | /* | |
a4e4c4f4 DC |
471 | * Here we do a spurious inode lock in order to coordinate with inode |
472 | * cache radix tree lookups. This is because the lookup can reference | |
473 | * the inodes in the cache without taking references. We make that OK | |
474 | * here by ensuring that we wait until the inode is unlocked after the | |
475 | * lookup before we go ahead and free it. We get both the ilock and | |
476 | * the iolock because the code may need to drop the ilock one but will | |
477 | * still hold the iolock. | |
1da177e4 LT |
478 | */ |
479 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
480 | ||
481 | /* | |
482 | * Release dquots (and their references) if any. An inode may escape | |
483 | * xfs_inactive and get here via vn_alloc->vn_reclaim path. | |
484 | */ | |
485 | XFS_QM_DQDETACH(ip->i_mount, ip); | |
486 | ||
1da177e4 LT |
487 | /* |
488 | * Free all memory associated with the inode. | |
489 | */ | |
439b8434 | 490 | xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); |
1da177e4 LT |
491 | xfs_idestroy(ip); |
492 | } | |
493 | ||
494 | /* | |
495 | * This routine removes an about-to-be-destroyed inode from | |
496 | * all of the lists in which it is located with the exception | |
497 | * of the behavior chain. | |
498 | */ | |
499 | void | |
500 | xfs_iextract( | |
501 | xfs_inode_t *ip) | |
502 | { | |
da353b0d DC |
503 | xfs_mount_t *mp = ip->i_mount; |
504 | xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino); | |
da353b0d DC |
505 | |
506 | write_lock(&pag->pag_ici_lock); | |
507 | radix_tree_delete(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ip->i_ino)); | |
508 | write_unlock(&pag->pag_ici_lock); | |
509 | xfs_put_perag(mp, pag); | |
1da177e4 | 510 | |
1da177e4 | 511 | mp->m_ireclaims++; |
1da177e4 LT |
512 | } |
513 | ||
514 | /* | |
515 | * This is a wrapper routine around the xfs_ilock() routine | |
516 | * used to centralize some grungy code. It is used in places | |
517 | * that wish to lock the inode solely for reading the extents. | |
518 | * The reason these places can't just call xfs_ilock(SHARED) | |
519 | * is that the inode lock also guards to bringing in of the | |
520 | * extents from disk for a file in b-tree format. If the inode | |
521 | * is in b-tree format, then we need to lock the inode exclusively | |
522 | * until the extents are read in. Locking it exclusively all | |
523 | * the time would limit our parallelism unnecessarily, though. | |
524 | * What we do instead is check to see if the extents have been | |
525 | * read in yet, and only lock the inode exclusively if they | |
526 | * have not. | |
527 | * | |
528 | * The function returns a value which should be given to the | |
529 | * corresponding xfs_iunlock_map_shared(). This value is | |
530 | * the mode in which the lock was actually taken. | |
531 | */ | |
532 | uint | |
533 | xfs_ilock_map_shared( | |
534 | xfs_inode_t *ip) | |
535 | { | |
536 | uint lock_mode; | |
537 | ||
538 | if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) && | |
539 | ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) { | |
540 | lock_mode = XFS_ILOCK_EXCL; | |
541 | } else { | |
542 | lock_mode = XFS_ILOCK_SHARED; | |
543 | } | |
544 | ||
545 | xfs_ilock(ip, lock_mode); | |
546 | ||
547 | return lock_mode; | |
548 | } | |
549 | ||
550 | /* | |
551 | * This is simply the unlock routine to go with xfs_ilock_map_shared(). | |
552 | * All it does is call xfs_iunlock() with the given lock_mode. | |
553 | */ | |
554 | void | |
555 | xfs_iunlock_map_shared( | |
556 | xfs_inode_t *ip, | |
557 | unsigned int lock_mode) | |
558 | { | |
559 | xfs_iunlock(ip, lock_mode); | |
560 | } | |
561 | ||
562 | /* | |
563 | * The xfs inode contains 2 locks: a multi-reader lock called the | |
564 | * i_iolock and a multi-reader lock called the i_lock. This routine | |
565 | * allows either or both of the locks to be obtained. | |
566 | * | |
567 | * The 2 locks should always be ordered so that the IO lock is | |
568 | * obtained first in order to prevent deadlock. | |
569 | * | |
570 | * ip -- the inode being locked | |
571 | * lock_flags -- this parameter indicates the inode's locks | |
572 | * to be locked. It can be: | |
573 | * XFS_IOLOCK_SHARED, | |
574 | * XFS_IOLOCK_EXCL, | |
575 | * XFS_ILOCK_SHARED, | |
576 | * XFS_ILOCK_EXCL, | |
577 | * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED, | |
578 | * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL, | |
579 | * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED, | |
580 | * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL | |
581 | */ | |
582 | void | |
579aa9ca CH |
583 | xfs_ilock( |
584 | xfs_inode_t *ip, | |
585 | uint lock_flags) | |
1da177e4 LT |
586 | { |
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 | |
579aa9ca | 598 | if (lock_flags & XFS_IOLOCK_EXCL) |
f7c66ce3 | 599 | mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); |
579aa9ca | 600 | else if (lock_flags & XFS_IOLOCK_SHARED) |
f7c66ce3 | 601 | mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); |
579aa9ca CH |
602 | |
603 | if (lock_flags & XFS_ILOCK_EXCL) | |
f7c66ce3 | 604 | mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); |
579aa9ca | 605 | else if (lock_flags & XFS_ILOCK_SHARED) |
f7c66ce3 | 606 | mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); |
579aa9ca | 607 | |
1da177e4 LT |
608 | xfs_ilock_trace(ip, 1, lock_flags, (inst_t *)__return_address); |
609 | } | |
610 | ||
611 | /* | |
612 | * This is just like xfs_ilock(), except that the caller | |
613 | * is guaranteed not to sleep. It returns 1 if it gets | |
614 | * the requested locks and 0 otherwise. If the IO lock is | |
615 | * obtained but the inode lock cannot be, then the IO lock | |
616 | * is dropped before returning. | |
617 | * | |
618 | * ip -- the inode being locked | |
619 | * lock_flags -- this parameter indicates the inode's locks to be | |
620 | * to be locked. See the comment for xfs_ilock() for a list | |
621 | * of valid values. | |
1da177e4 LT |
622 | */ |
623 | int | |
579aa9ca CH |
624 | xfs_ilock_nowait( |
625 | xfs_inode_t *ip, | |
626 | uint lock_flags) | |
1da177e4 | 627 | { |
1da177e4 LT |
628 | /* |
629 | * You can't set both SHARED and EXCL for the same lock, | |
630 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
631 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
632 | */ | |
633 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
634 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
635 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
636 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
f7c66ce3 | 637 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); |
1da177e4 | 638 | |
1da177e4 | 639 | if (lock_flags & XFS_IOLOCK_EXCL) { |
579aa9ca CH |
640 | if (!mrtryupdate(&ip->i_iolock)) |
641 | goto out; | |
1da177e4 | 642 | } else if (lock_flags & XFS_IOLOCK_SHARED) { |
579aa9ca CH |
643 | if (!mrtryaccess(&ip->i_iolock)) |
644 | goto out; | |
1da177e4 LT |
645 | } |
646 | if (lock_flags & XFS_ILOCK_EXCL) { | |
579aa9ca CH |
647 | if (!mrtryupdate(&ip->i_lock)) |
648 | goto out_undo_iolock; | |
1da177e4 | 649 | } else if (lock_flags & XFS_ILOCK_SHARED) { |
579aa9ca CH |
650 | if (!mrtryaccess(&ip->i_lock)) |
651 | goto out_undo_iolock; | |
1da177e4 LT |
652 | } |
653 | xfs_ilock_trace(ip, 2, lock_flags, (inst_t *)__return_address); | |
654 | return 1; | |
579aa9ca CH |
655 | |
656 | out_undo_iolock: | |
657 | if (lock_flags & XFS_IOLOCK_EXCL) | |
658 | mrunlock_excl(&ip->i_iolock); | |
659 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
660 | mrunlock_shared(&ip->i_iolock); | |
661 | out: | |
662 | return 0; | |
1da177e4 LT |
663 | } |
664 | ||
665 | /* | |
666 | * xfs_iunlock() is used to drop the inode locks acquired with | |
667 | * xfs_ilock() and xfs_ilock_nowait(). The caller must pass | |
668 | * in the flags given to xfs_ilock() or xfs_ilock_nowait() so | |
669 | * that we know which locks to drop. | |
670 | * | |
671 | * ip -- the inode being unlocked | |
672 | * lock_flags -- this parameter indicates the inode's locks to be | |
673 | * to be unlocked. See the comment for xfs_ilock() for a list | |
674 | * of valid values for this parameter. | |
675 | * | |
676 | */ | |
677 | void | |
579aa9ca CH |
678 | xfs_iunlock( |
679 | xfs_inode_t *ip, | |
680 | uint lock_flags) | |
1da177e4 LT |
681 | { |
682 | /* | |
683 | * You can't set both SHARED and EXCL for the same lock, | |
684 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
685 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
686 | */ | |
687 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
688 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
689 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
690 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
f7c66ce3 LM |
691 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_IUNLOCK_NONOTIFY | |
692 | XFS_LOCK_DEP_MASK)) == 0); | |
1da177e4 LT |
693 | ASSERT(lock_flags != 0); |
694 | ||
579aa9ca CH |
695 | if (lock_flags & XFS_IOLOCK_EXCL) |
696 | mrunlock_excl(&ip->i_iolock); | |
697 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
698 | mrunlock_shared(&ip->i_iolock); | |
1da177e4 | 699 | |
579aa9ca CH |
700 | if (lock_flags & XFS_ILOCK_EXCL) |
701 | mrunlock_excl(&ip->i_lock); | |
702 | else if (lock_flags & XFS_ILOCK_SHARED) | |
703 | mrunlock_shared(&ip->i_lock); | |
1da177e4 | 704 | |
579aa9ca CH |
705 | if ((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) && |
706 | !(lock_flags & XFS_IUNLOCK_NONOTIFY) && ip->i_itemp) { | |
1da177e4 LT |
707 | /* |
708 | * Let the AIL know that this item has been unlocked in case | |
709 | * it is in the AIL and anyone is waiting on it. Don't do | |
710 | * this if the caller has asked us not to. | |
711 | */ | |
783a2f65 | 712 | xfs_trans_unlocked_item(ip->i_itemp->ili_item.li_ailp, |
579aa9ca | 713 | (xfs_log_item_t*)(ip->i_itemp)); |
1da177e4 LT |
714 | } |
715 | xfs_ilock_trace(ip, 3, lock_flags, (inst_t *)__return_address); | |
716 | } | |
717 | ||
718 | /* | |
719 | * give up write locks. the i/o lock cannot be held nested | |
720 | * if it is being demoted. | |
721 | */ | |
722 | void | |
579aa9ca CH |
723 | xfs_ilock_demote( |
724 | xfs_inode_t *ip, | |
725 | uint lock_flags) | |
1da177e4 LT |
726 | { |
727 | ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)); | |
728 | ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0); | |
729 | ||
579aa9ca | 730 | if (lock_flags & XFS_ILOCK_EXCL) |
1da177e4 | 731 | mrdemote(&ip->i_lock); |
579aa9ca | 732 | if (lock_flags & XFS_IOLOCK_EXCL) |
1da177e4 | 733 | mrdemote(&ip->i_iolock); |
579aa9ca CH |
734 | } |
735 | ||
736 | #ifdef DEBUG | |
737 | /* | |
738 | * Debug-only routine, without additional rw_semaphore APIs, we can | |
739 | * now only answer requests regarding whether we hold the lock for write | |
740 | * (reader state is outside our visibility, we only track writer state). | |
741 | * | |
742 | * Note: this means !xfs_isilocked would give false positives, so don't do that. | |
743 | */ | |
744 | int | |
745 | xfs_isilocked( | |
746 | xfs_inode_t *ip, | |
747 | uint lock_flags) | |
748 | { | |
749 | if ((lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) == | |
750 | XFS_ILOCK_EXCL) { | |
751 | if (!ip->i_lock.mr_writer) | |
752 | return 0; | |
1da177e4 | 753 | } |
579aa9ca CH |
754 | |
755 | if ((lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) == | |
756 | XFS_IOLOCK_EXCL) { | |
757 | if (!ip->i_iolock.mr_writer) | |
758 | return 0; | |
759 | } | |
760 | ||
761 | return 1; | |
1da177e4 | 762 | } |
579aa9ca | 763 | #endif |
1da177e4 | 764 |