Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_fs.h" |
1da177e4 | 20 | #include "xfs_types.h" |
ef14f0c1 | 21 | #include "xfs_acl.h" |
a844f451 | 22 | #include "xfs_bit.h" |
1da177e4 | 23 | #include "xfs_log.h" |
a844f451 | 24 | #include "xfs_inum.h" |
1da177e4 LT |
25 | #include "xfs_trans.h" |
26 | #include "xfs_sb.h" | |
27 | #include "xfs_ag.h" | |
1da177e4 LT |
28 | #include "xfs_dir2.h" |
29 | #include "xfs_dmapi.h" | |
30 | #include "xfs_mount.h" | |
1da177e4 | 31 | #include "xfs_bmap_btree.h" |
a844f451 | 32 | #include "xfs_alloc_btree.h" |
1da177e4 | 33 | #include "xfs_ialloc_btree.h" |
1da177e4 | 34 | #include "xfs_dir2_sf.h" |
a844f451 | 35 | #include "xfs_attr_sf.h" |
1da177e4 LT |
36 | #include "xfs_dinode.h" |
37 | #include "xfs_inode.h" | |
a844f451 NS |
38 | #include "xfs_btree.h" |
39 | #include "xfs_ialloc.h" | |
1da177e4 LT |
40 | #include "xfs_quota.h" |
41 | #include "xfs_utils.h" | |
783a2f65 DC |
42 | #include "xfs_trans_priv.h" |
43 | #include "xfs_inode_item.h" | |
24f211ba CH |
44 | #include "xfs_bmap.h" |
45 | #include "xfs_btree_trace.h" | |
46 | #include "xfs_dir2_trace.h" | |
47 | ||
48 | ||
49 | /* | |
50 | * Allocate and initialise an xfs_inode. | |
51 | */ | |
52 | STATIC struct xfs_inode * | |
53 | xfs_inode_alloc( | |
54 | struct xfs_mount *mp, | |
55 | xfs_ino_t ino) | |
56 | { | |
57 | struct xfs_inode *ip; | |
58 | ||
59 | /* | |
60 | * if this didn't occur in transactions, we could use | |
61 | * KM_MAYFAIL and return NULL here on ENOMEM. Set the | |
62 | * code up to do this anyway. | |
63 | */ | |
64 | ip = kmem_zone_alloc(xfs_inode_zone, KM_SLEEP); | |
65 | if (!ip) | |
66 | return NULL; | |
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)); | |
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; | |
80 | ip->i_update_core = 0; | |
81 | ip->i_update_size = 0; | |
82 | ip->i_delayed_blks = 0; | |
83 | memset(&ip->i_d, 0, sizeof(xfs_icdinode_t)); | |
84 | ip->i_size = 0; | |
85 | ip->i_new_size = 0; | |
86 | ||
87 | /* | |
88 | * Initialize inode's trace buffers. | |
89 | */ | |
90 | #ifdef XFS_INODE_TRACE | |
91 | ip->i_trace = ktrace_alloc(INODE_TRACE_SIZE, KM_NOFS); | |
92 | #endif | |
93 | #ifdef XFS_BMAP_TRACE | |
94 | ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_NOFS); | |
95 | #endif | |
96 | #ifdef XFS_BTREE_TRACE | |
97 | ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_NOFS); | |
98 | #endif | |
99 | #ifdef XFS_RW_TRACE | |
100 | ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_NOFS); | |
101 | #endif | |
102 | #ifdef XFS_ILOCK_TRACE | |
103 | ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_NOFS); | |
104 | #endif | |
105 | #ifdef XFS_DIR2_TRACE | |
106 | ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_NOFS); | |
107 | #endif | |
705db3fd DC |
108 | /* |
109 | * Now initialise the VFS inode. We do this after the xfs_inode | |
110 | * initialisation as internal failures will result in ->destroy_inode | |
111 | * being called and that will pass down through the reclaim path and | |
112 | * free the XFS inode. This path requires the XFS inode to already be | |
113 | * initialised. Hence if this call fails, the xfs_inode has already | |
114 | * been freed and we should not reference it at all in the error | |
115 | * handling. | |
116 | */ | |
117 | if (!inode_init_always(mp->m_super, VFS_I(ip))) | |
118 | return NULL; | |
119 | ||
120 | /* prevent anyone from using this yet */ | |
121 | VFS_I(ip)->i_state = I_NEW|I_LOCK; | |
24f211ba CH |
122 | |
123 | return ip; | |
124 | } | |
1da177e4 | 125 | |
1da177e4 | 126 | /* |
6441e549 | 127 | * Check the validity of the inode we just found it the cache |
1da177e4 | 128 | */ |
6441e549 DC |
129 | static int |
130 | xfs_iget_cache_hit( | |
6441e549 DC |
131 | struct xfs_perag *pag, |
132 | struct xfs_inode *ip, | |
133 | int flags, | |
134 | int lock_flags) __releases(pag->pag_ici_lock) | |
1da177e4 | 135 | { |
6441e549 | 136 | struct xfs_mount *mp = ip->i_mount; |
6bfb3d06 | 137 | int error = EAGAIN; |
da353b0d | 138 | |
6441e549 DC |
139 | /* |
140 | * If INEW is set this inode is being set up | |
bf904248 | 141 | * If IRECLAIM is set this inode is being torn down |
6441e549 DC |
142 | * Pause and try again. |
143 | */ | |
bf904248 | 144 | if (xfs_iflags_test(ip, (XFS_INEW|XFS_IRECLAIM))) { |
6441e549 DC |
145 | XFS_STATS_INC(xs_ig_frecycle); |
146 | goto out_error; | |
147 | } | |
da353b0d | 148 | |
bf904248 DC |
149 | /* If IRECLAIMABLE is set, we've torn down the vfs inode part */ |
150 | if (xfs_iflags_test(ip, XFS_IRECLAIMABLE)) { | |
151 | ||
da353b0d | 152 | /* |
bf904248 DC |
153 | * If lookup is racing with unlink, then we should return an |
154 | * error immediately so we don't remove it from the reclaim | |
155 | * list and potentially leak the inode. | |
da353b0d | 156 | */ |
bf904248 DC |
157 | if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) { |
158 | error = ENOENT; | |
6441e549 DC |
159 | goto out_error; |
160 | } | |
bf904248 DC |
161 | |
162 | xfs_itrace_exit_tag(ip, "xfs_iget.alloc"); | |
da353b0d | 163 | |
6441e549 | 164 | /* |
bf904248 DC |
165 | * We need to re-initialise the VFS inode as it has been |
166 | * 'freed' by the VFS. Do this here so we can deal with | |
167 | * errors cleanly, then tag it so it can be set up correctly | |
168 | * later. | |
6441e549 | 169 | */ |
bf904248 DC |
170 | if (!inode_init_always(mp->m_super, VFS_I(ip))) { |
171 | error = ENOMEM; | |
6441e549 | 172 | goto out_error; |
da353b0d | 173 | } |
6bfb3d06 DC |
174 | |
175 | /* | |
176 | * We must set the XFS_INEW flag before clearing the | |
177 | * XFS_IRECLAIMABLE flag so that if a racing lookup does | |
178 | * not find the XFS_IRECLAIMABLE above but has the igrab() | |
179 | * below succeed we can safely check XFS_INEW to detect | |
180 | * that this inode is still being initialised. | |
181 | */ | |
bf904248 | 182 | xfs_iflags_set(ip, XFS_INEW); |
6441e549 | 183 | xfs_iflags_clear(ip, XFS_IRECLAIMABLE); |
396beb85 DC |
184 | |
185 | /* clear the radix tree reclaim flag as well. */ | |
186 | __xfs_inode_clear_reclaim_tag(mp, pag, ip); | |
bf904248 DC |
187 | } else if (!igrab(VFS_I(ip))) { |
188 | /* If the VFS inode is being torn down, pause and try again. */ | |
bf904248 DC |
189 | XFS_STATS_INC(xs_ig_frecycle); |
190 | goto out_error; | |
6bfb3d06 DC |
191 | } else if (xfs_iflags_test(ip, XFS_INEW)) { |
192 | /* | |
193 | * We are racing with another cache hit that is | |
194 | * currently recycling this inode out of the XFS_IRECLAIMABLE | |
195 | * state. Wait for the initialisation to complete before | |
196 | * continuing. | |
197 | */ | |
198 | wait_on_inode(VFS_I(ip)); | |
6441e549 | 199 | } |
1da177e4 | 200 | |
6441e549 DC |
201 | if (ip->i_d.di_mode == 0 && !(flags & XFS_IGET_CREATE)) { |
202 | error = ENOENT; | |
6bfb3d06 DC |
203 | iput(VFS_I(ip)); |
204 | goto out_error; | |
6441e549 | 205 | } |
da353b0d | 206 | |
6bfb3d06 DC |
207 | /* We've got a live one. */ |
208 | read_unlock(&pag->pag_ici_lock); | |
209 | ||
6441e549 DC |
210 | if (lock_flags != 0) |
211 | xfs_ilock(ip, lock_flags); | |
da353b0d | 212 | |
6441e549 DC |
213 | xfs_iflags_clear(ip, XFS_ISTALE); |
214 | xfs_itrace_exit_tag(ip, "xfs_iget.found"); | |
215 | XFS_STATS_INC(xs_ig_found); | |
216 | return 0; | |
1da177e4 | 217 | |
6441e549 | 218 | out_error: |
da353b0d | 219 | read_unlock(&pag->pag_ici_lock); |
6441e549 DC |
220 | return error; |
221 | } | |
222 | ||
223 | ||
224 | static int | |
225 | xfs_iget_cache_miss( | |
226 | struct xfs_mount *mp, | |
227 | struct xfs_perag *pag, | |
228 | xfs_trans_t *tp, | |
229 | xfs_ino_t ino, | |
230 | struct xfs_inode **ipp, | |
231 | xfs_daddr_t bno, | |
232 | int flags, | |
233 | int lock_flags) __releases(pag->pag_ici_lock) | |
234 | { | |
235 | struct xfs_inode *ip; | |
236 | int error; | |
237 | unsigned long first_index, mask; | |
238 | xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino); | |
1da177e4 | 239 | |
24f211ba CH |
240 | ip = xfs_inode_alloc(mp, ino); |
241 | if (!ip) | |
242 | return ENOMEM; | |
243 | ||
244 | error = xfs_iread(mp, tp, ip, bno, flags); | |
6441e549 | 245 | if (error) |
24f211ba | 246 | goto out_destroy; |
1da177e4 | 247 | |
15947f2d | 248 | xfs_itrace_exit_tag(ip, "xfs_iget.alloc"); |
1da177e4 | 249 | |
745b1f47 | 250 | if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) { |
6441e549 DC |
251 | error = ENOENT; |
252 | goto out_destroy; | |
1da177e4 LT |
253 | } |
254 | ||
255 | /* | |
bad55843 | 256 | * Preload the radix tree so we can insert safely under the |
56e73ec4 DC |
257 | * write spinlock. Note that we cannot sleep inside the preload |
258 | * region. | |
1da177e4 | 259 | */ |
da353b0d | 260 | if (radix_tree_preload(GFP_KERNEL)) { |
6441e549 | 261 | error = EAGAIN; |
ed93ec39 CH |
262 | goto out_destroy; |
263 | } | |
264 | ||
265 | /* | |
266 | * Because the inode hasn't been added to the radix-tree yet it can't | |
267 | * be found by another thread, so we can do the non-sleeping lock here. | |
268 | */ | |
269 | if (lock_flags) { | |
270 | if (!xfs_ilock_nowait(ip, lock_flags)) | |
271 | BUG(); | |
da353b0d | 272 | } |
f338f903 | 273 | |
da353b0d DC |
274 | mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1); |
275 | first_index = agino & mask; | |
276 | write_lock(&pag->pag_ici_lock); | |
6441e549 DC |
277 | |
278 | /* insert the new inode */ | |
da353b0d DC |
279 | error = radix_tree_insert(&pag->pag_ici_root, agino, ip); |
280 | if (unlikely(error)) { | |
6441e549 | 281 | WARN_ON(error != -EEXIST); |
da353b0d | 282 | XFS_STATS_INC(xs_ig_dup); |
6441e549 | 283 | error = EAGAIN; |
56e73ec4 | 284 | goto out_preload_end; |
1da177e4 LT |
285 | } |
286 | ||
6441e549 | 287 | /* These values _must_ be set before releasing the radix tree lock! */ |
1da177e4 | 288 | ip->i_udquot = ip->i_gdquot = NULL; |
7a18c386 | 289 | xfs_iflags_set(ip, XFS_INEW); |
1da177e4 | 290 | |
da353b0d DC |
291 | write_unlock(&pag->pag_ici_lock); |
292 | radix_tree_preload_end(); | |
6441e549 DC |
293 | *ipp = ip; |
294 | return 0; | |
295 | ||
56e73ec4 | 296 | out_preload_end: |
6441e549 DC |
297 | write_unlock(&pag->pag_ici_lock); |
298 | radix_tree_preload_end(); | |
56e73ec4 DC |
299 | if (lock_flags) |
300 | xfs_iunlock(ip, lock_flags); | |
6441e549 | 301 | out_destroy: |
9ed0451e | 302 | xfs_destroy_inode(ip); |
6441e549 DC |
303 | return error; |
304 | } | |
305 | ||
306 | /* | |
307 | * Look up an inode by number in the given file system. | |
308 | * The inode is looked up in the cache held in each AG. | |
bf904248 DC |
309 | * If the inode is found in the cache, initialise the vfs inode |
310 | * if necessary. | |
6441e549 DC |
311 | * |
312 | * If it is not in core, read it in from the file system's device, | |
bf904248 | 313 | * add it to the cache and initialise the vfs inode. |
6441e549 DC |
314 | * |
315 | * The inode is locked according to the value of the lock_flags parameter. | |
316 | * This flag parameter indicates how and if the inode's IO lock and inode lock | |
317 | * should be taken. | |
318 | * | |
319 | * mp -- the mount point structure for the current file system. It points | |
320 | * to the inode hash table. | |
321 | * tp -- a pointer to the current transaction if there is one. This is | |
322 | * simply passed through to the xfs_iread() call. | |
323 | * ino -- the number of the inode desired. This is the unique identifier | |
324 | * within the file system for the inode being requested. | |
325 | * lock_flags -- flags indicating how to lock the inode. See the comment | |
326 | * for xfs_ilock() for a list of valid values. | |
327 | * bno -- the block number starting the buffer containing the inode, | |
328 | * if known (as by bulkstat), else 0. | |
329 | */ | |
bf904248 DC |
330 | int |
331 | xfs_iget( | |
6441e549 DC |
332 | xfs_mount_t *mp, |
333 | xfs_trans_t *tp, | |
334 | xfs_ino_t ino, | |
335 | uint flags, | |
336 | uint lock_flags, | |
337 | xfs_inode_t **ipp, | |
338 | xfs_daddr_t bno) | |
339 | { | |
340 | xfs_inode_t *ip; | |
341 | int error; | |
342 | xfs_perag_t *pag; | |
343 | xfs_agino_t agino; | |
344 | ||
345 | /* the radix tree exists only in inode capable AGs */ | |
346 | if (XFS_INO_TO_AGNO(mp, ino) >= mp->m_maxagi) | |
347 | return EINVAL; | |
348 | ||
349 | /* get the perag structure and ensure that it's inode capable */ | |
350 | pag = xfs_get_perag(mp, ino); | |
351 | if (!pag->pagi_inodeok) | |
352 | return EINVAL; | |
353 | ASSERT(pag->pag_ici_init); | |
354 | agino = XFS_INO_TO_AGINO(mp, ino); | |
355 | ||
356 | again: | |
357 | error = 0; | |
358 | read_lock(&pag->pag_ici_lock); | |
359 | ip = radix_tree_lookup(&pag->pag_ici_root, agino); | |
360 | ||
361 | if (ip) { | |
bf904248 | 362 | error = xfs_iget_cache_hit(pag, ip, flags, lock_flags); |
6441e549 DC |
363 | if (error) |
364 | goto out_error_or_again; | |
365 | } else { | |
366 | read_unlock(&pag->pag_ici_lock); | |
367 | XFS_STATS_INC(xs_ig_missed); | |
368 | ||
369 | error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip, bno, | |
370 | flags, lock_flags); | |
371 | if (error) | |
372 | goto out_error_or_again; | |
373 | } | |
da353b0d | 374 | xfs_put_perag(mp, pag); |
1da177e4 | 375 | |
1da177e4 LT |
376 | *ipp = ip; |
377 | ||
bf904248 DC |
378 | ASSERT(ip->i_df.if_ext_max == |
379 | XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t)); | |
1da177e4 LT |
380 | /* |
381 | * If we have a real type for an on-disk inode, we can set ops(&unlock) | |
382 | * now. If it's a new inode being created, xfs_ialloc will handle it. | |
383 | */ | |
bf904248 | 384 | if (xfs_iflags_test(ip, XFS_INEW) && ip->i_d.di_mode != 0) |
41be8bed | 385 | xfs_setup_inode(ip); |
1da177e4 | 386 | return 0; |
6441e549 DC |
387 | |
388 | out_error_or_again: | |
389 | if (error == EAGAIN) { | |
390 | delay(1); | |
391 | goto again; | |
392 | } | |
393 | xfs_put_perag(mp, pag); | |
394 | return error; | |
1da177e4 LT |
395 | } |
396 | ||
397 | ||
1da177e4 LT |
398 | /* |
399 | * Look for the inode corresponding to the given ino in the hash table. | |
400 | * If it is there and its i_transp pointer matches tp, return it. | |
401 | * Otherwise, return NULL. | |
402 | */ | |
403 | xfs_inode_t * | |
404 | xfs_inode_incore(xfs_mount_t *mp, | |
405 | xfs_ino_t ino, | |
406 | xfs_trans_t *tp) | |
407 | { | |
1da177e4 | 408 | xfs_inode_t *ip; |
da353b0d DC |
409 | xfs_perag_t *pag; |
410 | ||
411 | pag = xfs_get_perag(mp, ino); | |
412 | read_lock(&pag->pag_ici_lock); | |
413 | ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ino)); | |
414 | read_unlock(&pag->pag_ici_lock); | |
415 | xfs_put_perag(mp, pag); | |
416 | ||
417 | /* the returned inode must match the transaction */ | |
418 | if (ip && (ip->i_transp != tp)) | |
419 | return NULL; | |
420 | return ip; | |
1da177e4 LT |
421 | } |
422 | ||
423 | /* | |
424 | * Decrement reference count of an inode structure and unlock it. | |
425 | * | |
426 | * ip -- the inode being released | |
427 | * lock_flags -- this parameter indicates the inode's locks to be | |
428 | * to be released. See the comment on xfs_iunlock() for a list | |
429 | * of valid values. | |
430 | */ | |
431 | void | |
432 | xfs_iput(xfs_inode_t *ip, | |
433 | uint lock_flags) | |
434 | { | |
cf441eeb | 435 | xfs_itrace_entry(ip); |
1da177e4 | 436 | xfs_iunlock(ip, lock_flags); |
10090be2 | 437 | IRELE(ip); |
1da177e4 LT |
438 | } |
439 | ||
440 | /* | |
441 | * Special iput for brand-new inodes that are still locked | |
442 | */ | |
443 | void | |
01651646 DC |
444 | xfs_iput_new( |
445 | xfs_inode_t *ip, | |
446 | uint lock_flags) | |
1da177e4 | 447 | { |
01651646 | 448 | struct inode *inode = VFS_I(ip); |
1da177e4 | 449 | |
cf441eeb | 450 | xfs_itrace_entry(ip); |
1da177e4 LT |
451 | |
452 | if ((ip->i_d.di_mode == 0)) { | |
7a18c386 | 453 | ASSERT(!xfs_iflags_test(ip, XFS_IRECLAIMABLE)); |
10090be2 | 454 | make_bad_inode(inode); |
1da177e4 LT |
455 | } |
456 | if (inode->i_state & I_NEW) | |
457 | unlock_new_inode(inode); | |
458 | if (lock_flags) | |
459 | xfs_iunlock(ip, lock_flags); | |
10090be2 | 460 | IRELE(ip); |
1da177e4 LT |
461 | } |
462 | ||
1da177e4 | 463 | /* |
5cafdeb2 CH |
464 | * This is called free all the memory associated with an inode. |
465 | * It must free the inode itself and any buffers allocated for | |
466 | * if_extents/if_data and if_broot. It must also free the lock | |
467 | * associated with the inode. | |
468 | * | |
469 | * Note: because we don't initialise everything on reallocation out | |
470 | * of the zone, we must ensure we nullify everything correctly before | |
471 | * freeing the structure. | |
1da177e4 LT |
472 | */ |
473 | void | |
5cafdeb2 CH |
474 | xfs_ireclaim( |
475 | struct xfs_inode *ip) | |
1da177e4 | 476 | { |
5cafdeb2 CH |
477 | struct xfs_mount *mp = ip->i_mount; |
478 | struct xfs_perag *pag; | |
1da177e4 | 479 | |
5cafdeb2 | 480 | XFS_STATS_INC(xs_ig_reclaims); |
1da177e4 LT |
481 | |
482 | /* | |
5cafdeb2 CH |
483 | * Remove the inode from the per-AG radix tree. It doesn't matter |
484 | * if it was never added to it because radix_tree_delete can deal | |
485 | * with that case just fine. | |
1da177e4 | 486 | */ |
5cafdeb2 CH |
487 | pag = xfs_get_perag(mp, ip->i_ino); |
488 | write_lock(&pag->pag_ici_lock); | |
489 | radix_tree_delete(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ip->i_ino)); | |
490 | write_unlock(&pag->pag_ici_lock); | |
491 | xfs_put_perag(mp, pag); | |
1da177e4 LT |
492 | |
493 | /* | |
5cafdeb2 CH |
494 | * Here we do an (almost) spurious inode lock in order to coordinate |
495 | * with inode cache radix tree lookups. This is because the lookup | |
496 | * can reference the inodes in the cache without taking references. | |
497 | * | |
498 | * We make that OK here by ensuring that we wait until the inode is | |
499 | * unlocked after the lookup before we go ahead and free it. We get | |
500 | * both the ilock and the iolock because the code may need to drop the | |
501 | * ilock one but will still hold the iolock. | |
1da177e4 | 502 | */ |
5cafdeb2 | 503 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); |
7d095257 | 504 | xfs_qm_dqdetach(ip); |
439b8434 | 505 | xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); |
1da177e4 | 506 | |
5cafdeb2 CH |
507 | switch (ip->i_d.di_mode & S_IFMT) { |
508 | case S_IFREG: | |
509 | case S_IFDIR: | |
510 | case S_IFLNK: | |
511 | xfs_idestroy_fork(ip, XFS_DATA_FORK); | |
512 | break; | |
513 | } | |
da353b0d | 514 | |
5cafdeb2 CH |
515 | if (ip->i_afp) |
516 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
1da177e4 | 517 | |
5cafdeb2 CH |
518 | #ifdef XFS_INODE_TRACE |
519 | ktrace_free(ip->i_trace); | |
520 | #endif | |
521 | #ifdef XFS_BMAP_TRACE | |
522 | ktrace_free(ip->i_xtrace); | |
523 | #endif | |
524 | #ifdef XFS_BTREE_TRACE | |
525 | ktrace_free(ip->i_btrace); | |
526 | #endif | |
527 | #ifdef XFS_RW_TRACE | |
528 | ktrace_free(ip->i_rwtrace); | |
529 | #endif | |
530 | #ifdef XFS_ILOCK_TRACE | |
531 | ktrace_free(ip->i_lock_trace); | |
532 | #endif | |
533 | #ifdef XFS_DIR2_TRACE | |
534 | ktrace_free(ip->i_dir_trace); | |
535 | #endif | |
536 | if (ip->i_itemp) { | |
537 | /* | |
538 | * Only if we are shutting down the fs will we see an | |
539 | * inode still in the AIL. If it is there, we should remove | |
540 | * it to prevent a use-after-free from occurring. | |
541 | */ | |
542 | xfs_log_item_t *lip = &ip->i_itemp->ili_item; | |
543 | struct xfs_ail *ailp = lip->li_ailp; | |
544 | ||
545 | ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) || | |
546 | XFS_FORCED_SHUTDOWN(ip->i_mount)); | |
547 | if (lip->li_flags & XFS_LI_IN_AIL) { | |
548 | spin_lock(&ailp->xa_lock); | |
549 | if (lip->li_flags & XFS_LI_IN_AIL) | |
550 | xfs_trans_ail_delete(ailp, lip); | |
551 | else | |
552 | spin_unlock(&ailp->xa_lock); | |
553 | } | |
554 | xfs_inode_item_destroy(ip); | |
555 | ip->i_itemp = NULL; | |
556 | } | |
557 | /* asserts to verify all state is correct here */ | |
558 | ASSERT(atomic_read(&ip->i_iocount) == 0); | |
559 | ASSERT(atomic_read(&ip->i_pincount) == 0); | |
560 | ASSERT(!spin_is_locked(&ip->i_flags_lock)); | |
561 | ASSERT(completion_done(&ip->i_flush)); | |
562 | kmem_zone_free(xfs_inode_zone, ip); | |
1da177e4 LT |
563 | } |
564 | ||
565 | /* | |
566 | * This is a wrapper routine around the xfs_ilock() routine | |
567 | * used to centralize some grungy code. It is used in places | |
568 | * that wish to lock the inode solely for reading the extents. | |
569 | * The reason these places can't just call xfs_ilock(SHARED) | |
570 | * is that the inode lock also guards to bringing in of the | |
571 | * extents from disk for a file in b-tree format. If the inode | |
572 | * is in b-tree format, then we need to lock the inode exclusively | |
573 | * until the extents are read in. Locking it exclusively all | |
574 | * the time would limit our parallelism unnecessarily, though. | |
575 | * What we do instead is check to see if the extents have been | |
576 | * read in yet, and only lock the inode exclusively if they | |
577 | * have not. | |
578 | * | |
579 | * The function returns a value which should be given to the | |
580 | * corresponding xfs_iunlock_map_shared(). This value is | |
581 | * the mode in which the lock was actually taken. | |
582 | */ | |
583 | uint | |
584 | xfs_ilock_map_shared( | |
585 | xfs_inode_t *ip) | |
586 | { | |
587 | uint lock_mode; | |
588 | ||
589 | if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) && | |
590 | ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) { | |
591 | lock_mode = XFS_ILOCK_EXCL; | |
592 | } else { | |
593 | lock_mode = XFS_ILOCK_SHARED; | |
594 | } | |
595 | ||
596 | xfs_ilock(ip, lock_mode); | |
597 | ||
598 | return lock_mode; | |
599 | } | |
600 | ||
601 | /* | |
602 | * This is simply the unlock routine to go with xfs_ilock_map_shared(). | |
603 | * All it does is call xfs_iunlock() with the given lock_mode. | |
604 | */ | |
605 | void | |
606 | xfs_iunlock_map_shared( | |
607 | xfs_inode_t *ip, | |
608 | unsigned int lock_mode) | |
609 | { | |
610 | xfs_iunlock(ip, lock_mode); | |
611 | } | |
612 | ||
613 | /* | |
614 | * The xfs inode contains 2 locks: a multi-reader lock called the | |
615 | * i_iolock and a multi-reader lock called the i_lock. This routine | |
616 | * allows either or both of the locks to be obtained. | |
617 | * | |
618 | * The 2 locks should always be ordered so that the IO lock is | |
619 | * obtained first in order to prevent deadlock. | |
620 | * | |
621 | * ip -- the inode being locked | |
622 | * lock_flags -- this parameter indicates the inode's locks | |
623 | * to be locked. It can be: | |
624 | * XFS_IOLOCK_SHARED, | |
625 | * XFS_IOLOCK_EXCL, | |
626 | * XFS_ILOCK_SHARED, | |
627 | * XFS_ILOCK_EXCL, | |
628 | * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED, | |
629 | * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL, | |
630 | * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED, | |
631 | * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL | |
632 | */ | |
633 | void | |
579aa9ca CH |
634 | xfs_ilock( |
635 | xfs_inode_t *ip, | |
636 | uint lock_flags) | |
1da177e4 LT |
637 | { |
638 | /* | |
639 | * You can't set both SHARED and EXCL for the same lock, | |
640 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
641 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
642 | */ | |
643 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
644 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
645 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
646 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
f7c66ce3 | 647 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); |
1da177e4 | 648 | |
579aa9ca | 649 | if (lock_flags & XFS_IOLOCK_EXCL) |
f7c66ce3 | 650 | mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); |
579aa9ca | 651 | else if (lock_flags & XFS_IOLOCK_SHARED) |
f7c66ce3 | 652 | mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); |
579aa9ca CH |
653 | |
654 | if (lock_flags & XFS_ILOCK_EXCL) | |
f7c66ce3 | 655 | mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); |
579aa9ca | 656 | else if (lock_flags & XFS_ILOCK_SHARED) |
f7c66ce3 | 657 | mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); |
579aa9ca | 658 | |
1da177e4 LT |
659 | xfs_ilock_trace(ip, 1, lock_flags, (inst_t *)__return_address); |
660 | } | |
661 | ||
662 | /* | |
663 | * This is just like xfs_ilock(), except that the caller | |
664 | * is guaranteed not to sleep. It returns 1 if it gets | |
665 | * the requested locks and 0 otherwise. If the IO lock is | |
666 | * obtained but the inode lock cannot be, then the IO lock | |
667 | * is dropped before returning. | |
668 | * | |
669 | * ip -- the inode being locked | |
670 | * lock_flags -- this parameter indicates the inode's locks to be | |
671 | * to be locked. See the comment for xfs_ilock() for a list | |
672 | * of valid values. | |
1da177e4 LT |
673 | */ |
674 | int | |
579aa9ca CH |
675 | xfs_ilock_nowait( |
676 | xfs_inode_t *ip, | |
677 | uint lock_flags) | |
1da177e4 | 678 | { |
1da177e4 LT |
679 | /* |
680 | * You can't set both SHARED and EXCL for the same lock, | |
681 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
682 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
683 | */ | |
684 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
685 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
686 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
687 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
f7c66ce3 | 688 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); |
1da177e4 | 689 | |
1da177e4 | 690 | if (lock_flags & XFS_IOLOCK_EXCL) { |
579aa9ca CH |
691 | if (!mrtryupdate(&ip->i_iolock)) |
692 | goto out; | |
1da177e4 | 693 | } else if (lock_flags & XFS_IOLOCK_SHARED) { |
579aa9ca CH |
694 | if (!mrtryaccess(&ip->i_iolock)) |
695 | goto out; | |
1da177e4 LT |
696 | } |
697 | if (lock_flags & XFS_ILOCK_EXCL) { | |
579aa9ca CH |
698 | if (!mrtryupdate(&ip->i_lock)) |
699 | goto out_undo_iolock; | |
1da177e4 | 700 | } else if (lock_flags & XFS_ILOCK_SHARED) { |
579aa9ca CH |
701 | if (!mrtryaccess(&ip->i_lock)) |
702 | goto out_undo_iolock; | |
1da177e4 LT |
703 | } |
704 | xfs_ilock_trace(ip, 2, lock_flags, (inst_t *)__return_address); | |
705 | return 1; | |
579aa9ca CH |
706 | |
707 | out_undo_iolock: | |
708 | if (lock_flags & XFS_IOLOCK_EXCL) | |
709 | mrunlock_excl(&ip->i_iolock); | |
710 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
711 | mrunlock_shared(&ip->i_iolock); | |
712 | out: | |
713 | return 0; | |
1da177e4 LT |
714 | } |
715 | ||
716 | /* | |
717 | * xfs_iunlock() is used to drop the inode locks acquired with | |
718 | * xfs_ilock() and xfs_ilock_nowait(). The caller must pass | |
719 | * in the flags given to xfs_ilock() or xfs_ilock_nowait() so | |
720 | * that we know which locks to drop. | |
721 | * | |
722 | * ip -- the inode being unlocked | |
723 | * lock_flags -- this parameter indicates the inode's locks to be | |
724 | * to be unlocked. See the comment for xfs_ilock() for a list | |
725 | * of valid values for this parameter. | |
726 | * | |
727 | */ | |
728 | void | |
579aa9ca CH |
729 | xfs_iunlock( |
730 | xfs_inode_t *ip, | |
731 | uint lock_flags) | |
1da177e4 LT |
732 | { |
733 | /* | |
734 | * You can't set both SHARED and EXCL for the same lock, | |
735 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
736 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
737 | */ | |
738 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
739 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
740 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
741 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
f7c66ce3 LM |
742 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_IUNLOCK_NONOTIFY | |
743 | XFS_LOCK_DEP_MASK)) == 0); | |
1da177e4 LT |
744 | ASSERT(lock_flags != 0); |
745 | ||
579aa9ca CH |
746 | if (lock_flags & XFS_IOLOCK_EXCL) |
747 | mrunlock_excl(&ip->i_iolock); | |
748 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
749 | mrunlock_shared(&ip->i_iolock); | |
1da177e4 | 750 | |
579aa9ca CH |
751 | if (lock_flags & XFS_ILOCK_EXCL) |
752 | mrunlock_excl(&ip->i_lock); | |
753 | else if (lock_flags & XFS_ILOCK_SHARED) | |
754 | mrunlock_shared(&ip->i_lock); | |
1da177e4 | 755 | |
579aa9ca CH |
756 | if ((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) && |
757 | !(lock_flags & XFS_IUNLOCK_NONOTIFY) && ip->i_itemp) { | |
1da177e4 LT |
758 | /* |
759 | * Let the AIL know that this item has been unlocked in case | |
760 | * it is in the AIL and anyone is waiting on it. Don't do | |
761 | * this if the caller has asked us not to. | |
762 | */ | |
783a2f65 | 763 | xfs_trans_unlocked_item(ip->i_itemp->ili_item.li_ailp, |
579aa9ca | 764 | (xfs_log_item_t*)(ip->i_itemp)); |
1da177e4 LT |
765 | } |
766 | xfs_ilock_trace(ip, 3, lock_flags, (inst_t *)__return_address); | |
767 | } | |
768 | ||
769 | /* | |
770 | * give up write locks. the i/o lock cannot be held nested | |
771 | * if it is being demoted. | |
772 | */ | |
773 | void | |
579aa9ca CH |
774 | xfs_ilock_demote( |
775 | xfs_inode_t *ip, | |
776 | uint lock_flags) | |
1da177e4 LT |
777 | { |
778 | ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)); | |
779 | ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0); | |
780 | ||
579aa9ca | 781 | if (lock_flags & XFS_ILOCK_EXCL) |
1da177e4 | 782 | mrdemote(&ip->i_lock); |
579aa9ca | 783 | if (lock_flags & XFS_IOLOCK_EXCL) |
1da177e4 | 784 | mrdemote(&ip->i_iolock); |
579aa9ca CH |
785 | } |
786 | ||
787 | #ifdef DEBUG | |
788 | /* | |
789 | * Debug-only routine, without additional rw_semaphore APIs, we can | |
790 | * now only answer requests regarding whether we hold the lock for write | |
791 | * (reader state is outside our visibility, we only track writer state). | |
792 | * | |
793 | * Note: this means !xfs_isilocked would give false positives, so don't do that. | |
794 | */ | |
795 | int | |
796 | xfs_isilocked( | |
797 | xfs_inode_t *ip, | |
798 | uint lock_flags) | |
799 | { | |
800 | if ((lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) == | |
801 | XFS_ILOCK_EXCL) { | |
802 | if (!ip->i_lock.mr_writer) | |
803 | return 0; | |
1da177e4 | 804 | } |
579aa9ca CH |
805 | |
806 | if ((lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) == | |
807 | XFS_IOLOCK_EXCL) { | |
808 | if (!ip->i_iolock.mr_writer) | |
809 | return 0; | |
810 | } | |
811 | ||
812 | return 1; | |
1da177e4 | 813 | } |
579aa9ca | 814 | #endif |
1da177e4 | 815 | |
5a8d0f3c CH |
816 | #ifdef XFS_INODE_TRACE |
817 | ||
818 | #define KTRACE_ENTER(ip, vk, s, line, ra) \ | |
819 | ktrace_enter((ip)->i_trace, \ | |
820 | /* 0 */ (void *)(__psint_t)(vk), \ | |
821 | /* 1 */ (void *)(s), \ | |
822 | /* 2 */ (void *)(__psint_t) line, \ | |
823 | /* 3 */ (void *)(__psint_t)atomic_read(&VFS_I(ip)->i_count), \ | |
824 | /* 4 */ (void *)(ra), \ | |
825 | /* 5 */ NULL, \ | |
826 | /* 6 */ (void *)(__psint_t)current_cpu(), \ | |
827 | /* 7 */ (void *)(__psint_t)current_pid(), \ | |
828 | /* 8 */ (void *)__return_address, \ | |
829 | /* 9 */ NULL, NULL, NULL, NULL, NULL, NULL, NULL) | |
830 | ||
831 | /* | |
832 | * Vnode tracing code. | |
833 | */ | |
834 | void | |
835 | _xfs_itrace_entry(xfs_inode_t *ip, const char *func, inst_t *ra) | |
836 | { | |
837 | KTRACE_ENTER(ip, INODE_KTRACE_ENTRY, func, 0, ra); | |
838 | } | |
839 | ||
840 | void | |
841 | _xfs_itrace_exit(xfs_inode_t *ip, const char *func, inst_t *ra) | |
842 | { | |
843 | KTRACE_ENTER(ip, INODE_KTRACE_EXIT, func, 0, ra); | |
844 | } | |
845 | ||
846 | void | |
847 | xfs_itrace_hold(xfs_inode_t *ip, char *file, int line, inst_t *ra) | |
848 | { | |
849 | KTRACE_ENTER(ip, INODE_KTRACE_HOLD, file, line, ra); | |
850 | } | |
851 | ||
852 | void | |
853 | _xfs_itrace_ref(xfs_inode_t *ip, char *file, int line, inst_t *ra) | |
854 | { | |
855 | KTRACE_ENTER(ip, INODE_KTRACE_REF, file, line, ra); | |
856 | } | |
857 | ||
858 | void | |
859 | xfs_itrace_rele(xfs_inode_t *ip, char *file, int line, inst_t *ra) | |
860 | { | |
861 | KTRACE_ENTER(ip, INODE_KTRACE_RELE, file, line, ra); | |
862 | } | |
863 | #endif /* XFS_INODE_TRACE */ |