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" | |
1da177e4 | 41 | |
1da177e4 LT |
42 | /* |
43 | * Look up an inode by number in the given file system. | |
da353b0d DC |
44 | * The inode is looked up in the cache held in each AG. |
45 | * If the inode is found in the cache, attach it to the provided | |
46 | * vnode. | |
1da177e4 | 47 | * |
da353b0d DC |
48 | * If it is not in core, read it in from the file system's device, |
49 | * add it to the cache and attach the provided vnode. | |
1da177e4 LT |
50 | * |
51 | * The inode is locked according to the value of the lock_flags parameter. | |
52 | * This flag parameter indicates how and if the inode's IO lock and inode lock | |
53 | * should be taken. | |
54 | * | |
55 | * mp -- the mount point structure for the current file system. It points | |
56 | * to the inode hash table. | |
57 | * tp -- a pointer to the current transaction if there is one. This is | |
58 | * simply passed through to the xfs_iread() call. | |
59 | * ino -- the number of the inode desired. This is the unique identifier | |
60 | * within the file system for the inode being requested. | |
61 | * lock_flags -- flags indicating how to lock the inode. See the comment | |
62 | * for xfs_ilock() for a list of valid values. | |
63 | * bno -- the block number starting the buffer containing the inode, | |
64 | * if known (as by bulkstat), else 0. | |
65 | */ | |
66 | STATIC int | |
67 | xfs_iget_core( | |
67fcaa73 | 68 | bhv_vnode_t *vp, |
1da177e4 LT |
69 | xfs_mount_t *mp, |
70 | xfs_trans_t *tp, | |
71 | xfs_ino_t ino, | |
72 | uint flags, | |
73 | uint lock_flags, | |
74 | xfs_inode_t **ipp, | |
75 | xfs_daddr_t bno) | |
76 | { | |
1da177e4 LT |
77 | xfs_inode_t *ip; |
78 | xfs_inode_t *iq; | |
67fcaa73 | 79 | bhv_vnode_t *inode_vp; |
1da177e4 | 80 | int error; |
da353b0d DC |
81 | xfs_icluster_t *icl, *new_icl = NULL; |
82 | unsigned long first_index, mask; | |
83 | xfs_perag_t *pag; | |
84 | xfs_agino_t agino; | |
85 | ||
86 | /* the radix tree exists only in inode capable AGs */ | |
87 | if (XFS_INO_TO_AGNO(mp, ino) >= mp->m_maxagi) | |
88 | return EINVAL; | |
89 | ||
90 | /* get the perag structure and ensure that it's inode capable */ | |
91 | pag = xfs_get_perag(mp, ino); | |
92 | if (!pag->pagi_inodeok) | |
93 | return EINVAL; | |
94 | ASSERT(pag->pag_ici_init); | |
95 | agino = XFS_INO_TO_AGINO(mp, ino); | |
1da177e4 LT |
96 | |
97 | again: | |
da353b0d DC |
98 | read_lock(&pag->pag_ici_lock); |
99 | ip = radix_tree_lookup(&pag->pag_ici_root, agino); | |
100 | ||
101 | if (ip != NULL) { | |
102 | /* | |
103 | * If INEW is set this inode is being set up | |
104 | * we need to pause and try again. | |
105 | */ | |
106 | if (xfs_iflags_test(ip, XFS_INEW)) { | |
107 | read_unlock(&pag->pag_ici_lock); | |
108 | delay(1); | |
109 | XFS_STATS_INC(xs_ig_frecycle); | |
110 | ||
111 | goto again; | |
112 | } | |
1da177e4 | 113 | |
da353b0d DC |
114 | inode_vp = XFS_ITOV_NULL(ip); |
115 | if (inode_vp == NULL) { | |
1da177e4 | 116 | /* |
da353b0d DC |
117 | * If IRECLAIM is set this inode is |
118 | * on its way out of the system, | |
1da177e4 LT |
119 | * we need to pause and try again. |
120 | */ | |
da353b0d DC |
121 | if (xfs_iflags_test(ip, XFS_IRECLAIM)) { |
122 | read_unlock(&pag->pag_ici_lock); | |
1da177e4 LT |
123 | delay(1); |
124 | XFS_STATS_INC(xs_ig_frecycle); | |
125 | ||
126 | goto again; | |
127 | } | |
da353b0d | 128 | ASSERT(xfs_iflags_test(ip, XFS_IRECLAIMABLE)); |
1da177e4 | 129 | |
da353b0d DC |
130 | /* |
131 | * If lookup is racing with unlink, then we | |
132 | * should return an error immediately so we | |
133 | * don't remove it from the reclaim list and | |
134 | * potentially leak the inode. | |
135 | */ | |
136 | if ((ip->i_d.di_mode == 0) && | |
137 | !(flags & XFS_IGET_CREATE)) { | |
138 | read_unlock(&pag->pag_ici_lock); | |
139 | xfs_put_perag(mp, pag); | |
140 | return ENOENT; | |
141 | } | |
1da177e4 | 142 | |
da353b0d DC |
143 | /* |
144 | * There may be transactions sitting in the | |
145 | * incore log buffers or being flushed to disk | |
146 | * at this time. We can't clear the | |
147 | * XFS_IRECLAIMABLE flag until these | |
148 | * transactions have hit the disk, otherwise we | |
149 | * will void the guarantee the flag provides | |
150 | * xfs_iunpin() | |
151 | */ | |
152 | if (xfs_ipincount(ip)) { | |
153 | read_unlock(&pag->pag_ici_lock); | |
154 | xfs_log_force(mp, 0, | |
155 | XFS_LOG_FORCE|XFS_LOG_SYNC); | |
156 | XFS_STATS_INC(xs_ig_frecycle); | |
157 | goto again; | |
158 | } | |
1da177e4 | 159 | |
15947f2d | 160 | xfs_itrace_exit_tag(ip, "xfs_iget.alloc"); |
1da177e4 | 161 | |
da353b0d | 162 | XFS_STATS_INC(xs_ig_found); |
1da177e4 | 163 | |
da353b0d DC |
164 | xfs_iflags_clear(ip, XFS_IRECLAIMABLE); |
165 | read_unlock(&pag->pag_ici_lock); | |
1da177e4 | 166 | |
da353b0d DC |
167 | XFS_MOUNT_ILOCK(mp); |
168 | list_del_init(&ip->i_reclaim); | |
169 | XFS_MOUNT_IUNLOCK(mp); | |
1da177e4 | 170 | |
da353b0d | 171 | goto finish_inode; |
1da177e4 | 172 | |
da353b0d DC |
173 | } else if (vp != inode_vp) { |
174 | struct inode *inode = vn_to_inode(inode_vp); | |
1da177e4 | 175 | |
da353b0d DC |
176 | /* The inode is being torn down, pause and |
177 | * try again. | |
71bce256 | 178 | */ |
da353b0d DC |
179 | if (inode->i_state & (I_FREEING | I_CLEAR)) { |
180 | read_unlock(&pag->pag_ici_lock); | |
181 | delay(1); | |
182 | XFS_STATS_INC(xs_ig_frecycle); | |
1da177e4 | 183 | |
da353b0d | 184 | goto again; |
1da177e4 | 185 | } |
da353b0d DC |
186 | /* Chances are the other vnode (the one in the inode) is being torn |
187 | * down right now, and we landed on top of it. Question is, what do | |
188 | * we do? Unhook the old inode and hook up the new one? | |
189 | */ | |
190 | cmn_err(CE_PANIC, | |
191 | "xfs_iget_core: ambiguous vns: vp/0x%p, invp/0x%p", | |
192 | inode_vp, vp); | |
193 | } | |
745b1f47 | 194 | |
da353b0d DC |
195 | /* |
196 | * Inode cache hit | |
197 | */ | |
198 | read_unlock(&pag->pag_ici_lock); | |
199 | XFS_STATS_INC(xs_ig_found); | |
1da177e4 | 200 | |
da353b0d | 201 | finish_inode: |
613d7043 CH |
202 | if (ip->i_d.di_mode == 0 && !(flags & XFS_IGET_CREATE)) { |
203 | xfs_put_perag(mp, pag); | |
204 | return ENOENT; | |
1da177e4 | 205 | } |
da353b0d DC |
206 | |
207 | if (lock_flags != 0) | |
208 | xfs_ilock(ip, lock_flags); | |
209 | ||
210 | xfs_iflags_clear(ip, XFS_ISTALE); | |
15947f2d | 211 | xfs_itrace_exit_tag(ip, "xfs_iget.found"); |
da353b0d | 212 | goto return_ip; |
1da177e4 LT |
213 | } |
214 | ||
215 | /* | |
da353b0d | 216 | * Inode cache miss |
1da177e4 | 217 | */ |
da353b0d | 218 | read_unlock(&pag->pag_ici_lock); |
1da177e4 LT |
219 | XFS_STATS_INC(xs_ig_missed); |
220 | ||
1da177e4 LT |
221 | /* |
222 | * Read the disk inode attributes into a new inode structure and get | |
223 | * a new vnode for it. This should also initialize i_ino and i_mount. | |
224 | */ | |
745b1f47 NS |
225 | error = xfs_iread(mp, tp, ino, &ip, bno, |
226 | (flags & XFS_IGET_BULKSTAT) ? XFS_IMAP_BULKSTAT : 0); | |
da353b0d DC |
227 | if (error) { |
228 | xfs_put_perag(mp, pag); | |
1da177e4 | 229 | return error; |
da353b0d | 230 | } |
1da177e4 | 231 | |
15947f2d | 232 | xfs_itrace_exit_tag(ip, "xfs_iget.alloc"); |
1da177e4 LT |
233 | |
234 | xfs_inode_lock_init(ip, vp); | |
745b1f47 | 235 | if (lock_flags) |
1da177e4 | 236 | xfs_ilock(ip, lock_flags); |
745b1f47 NS |
237 | |
238 | if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) { | |
1da177e4 | 239 | xfs_idestroy(ip); |
da353b0d | 240 | xfs_put_perag(mp, pag); |
1da177e4 LT |
241 | return ENOENT; |
242 | } | |
243 | ||
244 | /* | |
da353b0d DC |
245 | * This is a bit messy - we preallocate everything we _might_ |
246 | * need before we pick up the ici lock. That way we don't have to | |
247 | * juggle locks and go all the way back to the start. | |
1da177e4 | 248 | */ |
da353b0d DC |
249 | new_icl = kmem_zone_alloc(xfs_icluster_zone, KM_SLEEP); |
250 | if (radix_tree_preload(GFP_KERNEL)) { | |
251 | delay(1); | |
252 | goto again; | |
253 | } | |
254 | mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1); | |
255 | first_index = agino & mask; | |
256 | write_lock(&pag->pag_ici_lock); | |
1da177e4 | 257 | |
da353b0d DC |
258 | /* |
259 | * Find the cluster if it exists | |
260 | */ | |
261 | icl = NULL; | |
262 | if (radix_tree_gang_lookup(&pag->pag_ici_root, (void**)&iq, | |
263 | first_index, 1)) { | |
a7430847 | 264 | if ((XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) == first_index) |
da353b0d DC |
265 | icl = iq->i_cluster; |
266 | } | |
1da177e4 | 267 | |
da353b0d DC |
268 | /* |
269 | * insert the new inode | |
270 | */ | |
271 | error = radix_tree_insert(&pag->pag_ici_root, agino, ip); | |
272 | if (unlikely(error)) { | |
273 | BUG_ON(error != -EEXIST); | |
274 | write_unlock(&pag->pag_ici_lock); | |
275 | radix_tree_preload_end(); | |
276 | xfs_idestroy(ip); | |
277 | XFS_STATS_INC(xs_ig_dup); | |
278 | goto again; | |
1da177e4 LT |
279 | } |
280 | ||
281 | /* | |
282 | * These values _must_ be set before releasing ihlock! | |
283 | */ | |
1da177e4 | 284 | ip->i_udquot = ip->i_gdquot = NULL; |
7a18c386 | 285 | xfs_iflags_set(ip, XFS_INEW); |
1da177e4 | 286 | |
da353b0d | 287 | ASSERT(ip->i_cluster == NULL); |
1da177e4 | 288 | |
da353b0d DC |
289 | if (!icl) { |
290 | spin_lock_init(&new_icl->icl_lock); | |
291 | INIT_HLIST_HEAD(&new_icl->icl_inodes); | |
292 | icl = new_icl; | |
293 | new_icl = NULL; | |
1da177e4 | 294 | } else { |
da353b0d | 295 | ASSERT(!hlist_empty(&icl->icl_inodes)); |
1da177e4 | 296 | } |
da353b0d DC |
297 | spin_lock(&icl->icl_lock); |
298 | hlist_add_head(&ip->i_cnode, &icl->icl_inodes); | |
299 | ip->i_cluster = icl; | |
300 | spin_unlock(&icl->icl_lock); | |
1da177e4 | 301 | |
da353b0d DC |
302 | write_unlock(&pag->pag_ici_lock); |
303 | radix_tree_preload_end(); | |
304 | if (new_icl) | |
305 | kmem_zone_free(xfs_icluster_zone, new_icl); | |
1da177e4 LT |
306 | |
307 | /* | |
308 | * Link ip to its mount and thread it on the mount's inode list. | |
309 | */ | |
310 | XFS_MOUNT_ILOCK(mp); | |
311 | if ((iq = mp->m_inodes)) { | |
312 | ASSERT(iq->i_mprev->i_mnext == iq); | |
313 | ip->i_mprev = iq->i_mprev; | |
314 | iq->i_mprev->i_mnext = ip; | |
315 | iq->i_mprev = ip; | |
316 | ip->i_mnext = iq; | |
317 | } else { | |
318 | ip->i_mnext = ip; | |
319 | ip->i_mprev = ip; | |
320 | } | |
321 | mp->m_inodes = ip; | |
322 | ||
323 | XFS_MOUNT_IUNLOCK(mp); | |
da353b0d | 324 | xfs_put_perag(mp, pag); |
1da177e4 LT |
325 | |
326 | return_ip: | |
327 | ASSERT(ip->i_df.if_ext_max == | |
328 | XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t)); | |
329 | ||
b3aea4ed | 330 | xfs_iflags_set(ip, XFS_IMODIFIED); |
1da177e4 LT |
331 | *ipp = ip; |
332 | ||
333 | /* | |
334 | * If we have a real type for an on-disk inode, we can set ops(&unlock) | |
335 | * now. If it's a new inode being created, xfs_ialloc will handle it. | |
336 | */ | |
745f6919 | 337 | xfs_initialize_vnode(mp, vp, ip); |
1da177e4 LT |
338 | return 0; |
339 | } | |
340 | ||
341 | ||
342 | /* | |
343 | * The 'normal' internal xfs_iget, if needed it will | |
344 | * 'allocate', or 'get', the vnode. | |
345 | */ | |
346 | int | |
347 | xfs_iget( | |
348 | xfs_mount_t *mp, | |
349 | xfs_trans_t *tp, | |
350 | xfs_ino_t ino, | |
351 | uint flags, | |
352 | uint lock_flags, | |
353 | xfs_inode_t **ipp, | |
354 | xfs_daddr_t bno) | |
355 | { | |
356 | struct inode *inode; | |
67fcaa73 | 357 | bhv_vnode_t *vp = NULL; |
1da177e4 LT |
358 | int error; |
359 | ||
1da177e4 LT |
360 | XFS_STATS_INC(xs_ig_attempts); |
361 | ||
ba403ab4 | 362 | retry: |
b267ce99 | 363 | inode = iget_locked(mp->m_super, ino); |
745f6919 | 364 | if (inode) { |
1da177e4 | 365 | xfs_inode_t *ip; |
1da177e4 | 366 | |
ec86dc02 | 367 | vp = vn_from_inode(inode); |
1da177e4 | 368 | if (inode->i_state & I_NEW) { |
1da177e4 LT |
369 | vn_initialize(inode); |
370 | error = xfs_iget_core(vp, mp, tp, ino, flags, | |
371 | lock_flags, ipp, bno); | |
372 | if (error) { | |
373 | vn_mark_bad(vp); | |
374 | if (inode->i_state & I_NEW) | |
375 | unlock_new_inode(inode); | |
376 | iput(inode); | |
377 | } | |
378 | } else { | |
ba403ab4 CH |
379 | /* |
380 | * If the inode is not fully constructed due to | |
c41564b5 | 381 | * filehandle mismatches wait for the inode to go |
ba403ab4 CH |
382 | * away and try again. |
383 | * | |
384 | * iget_locked will call __wait_on_freeing_inode | |
385 | * to wait for the inode to go away. | |
386 | */ | |
387 | if (is_bad_inode(inode) || | |
75e17b3c | 388 | ((ip = xfs_vtoi(vp)) == NULL)) { |
1da177e4 | 389 | iput(inode); |
ba403ab4 CH |
390 | delay(1); |
391 | goto retry; | |
1da177e4 LT |
392 | } |
393 | ||
1da177e4 LT |
394 | if (lock_flags != 0) |
395 | xfs_ilock(ip, lock_flags); | |
1da177e4 LT |
396 | XFS_STATS_INC(xs_ig_found); |
397 | *ipp = ip; | |
398 | error = 0; | |
399 | } | |
400 | } else | |
401 | error = ENOMEM; /* If we got no inode we are out of memory */ | |
402 | ||
403 | return error; | |
404 | } | |
405 | ||
406 | /* | |
407 | * Do the setup for the various locks within the incore inode. | |
408 | */ | |
409 | void | |
410 | xfs_inode_lock_init( | |
411 | xfs_inode_t *ip, | |
67fcaa73 | 412 | bhv_vnode_t *vp) |
1da177e4 LT |
413 | { |
414 | mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER, | |
2aeaa258 CH |
415 | "xfsino", ip->i_ino); |
416 | mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino); | |
1da177e4 LT |
417 | init_waitqueue_head(&ip->i_ipin_wait); |
418 | atomic_set(&ip->i_pincount, 0); | |
7ae67d78 | 419 | initnsema(&ip->i_flock, 1, "xfsfino"); |
1da177e4 LT |
420 | } |
421 | ||
422 | /* | |
423 | * Look for the inode corresponding to the given ino in the hash table. | |
424 | * If it is there and its i_transp pointer matches tp, return it. | |
425 | * Otherwise, return NULL. | |
426 | */ | |
427 | xfs_inode_t * | |
428 | xfs_inode_incore(xfs_mount_t *mp, | |
429 | xfs_ino_t ino, | |
430 | xfs_trans_t *tp) | |
431 | { | |
1da177e4 | 432 | xfs_inode_t *ip; |
da353b0d DC |
433 | xfs_perag_t *pag; |
434 | ||
435 | pag = xfs_get_perag(mp, ino); | |
436 | read_lock(&pag->pag_ici_lock); | |
437 | ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ino)); | |
438 | read_unlock(&pag->pag_ici_lock); | |
439 | xfs_put_perag(mp, pag); | |
440 | ||
441 | /* the returned inode must match the transaction */ | |
442 | if (ip && (ip->i_transp != tp)) | |
443 | return NULL; | |
444 | return ip; | |
1da177e4 LT |
445 | } |
446 | ||
447 | /* | |
448 | * Decrement reference count of an inode structure and unlock it. | |
449 | * | |
450 | * ip -- the inode being released | |
451 | * lock_flags -- this parameter indicates the inode's locks to be | |
452 | * to be released. See the comment on xfs_iunlock() for a list | |
453 | * of valid values. | |
454 | */ | |
455 | void | |
456 | xfs_iput(xfs_inode_t *ip, | |
457 | uint lock_flags) | |
458 | { | |
67fcaa73 | 459 | bhv_vnode_t *vp = XFS_ITOV(ip); |
1da177e4 | 460 | |
cf441eeb | 461 | xfs_itrace_entry(ip); |
1da177e4 | 462 | xfs_iunlock(ip, lock_flags); |
1da177e4 LT |
463 | VN_RELE(vp); |
464 | } | |
465 | ||
466 | /* | |
467 | * Special iput for brand-new inodes that are still locked | |
468 | */ | |
469 | void | |
470 | xfs_iput_new(xfs_inode_t *ip, | |
471 | uint lock_flags) | |
472 | { | |
67fcaa73 | 473 | bhv_vnode_t *vp = XFS_ITOV(ip); |
ec86dc02 | 474 | struct inode *inode = vn_to_inode(vp); |
1da177e4 | 475 | |
cf441eeb | 476 | xfs_itrace_entry(ip); |
1da177e4 LT |
477 | |
478 | if ((ip->i_d.di_mode == 0)) { | |
7a18c386 | 479 | ASSERT(!xfs_iflags_test(ip, XFS_IRECLAIMABLE)); |
1da177e4 LT |
480 | vn_mark_bad(vp); |
481 | } | |
482 | if (inode->i_state & I_NEW) | |
483 | unlock_new_inode(inode); | |
484 | if (lock_flags) | |
485 | xfs_iunlock(ip, lock_flags); | |
486 | VN_RELE(vp); | |
487 | } | |
488 | ||
489 | ||
490 | /* | |
491 | * This routine embodies the part of the reclaim code that pulls | |
492 | * the inode from the inode hash table and the mount structure's | |
493 | * inode list. | |
494 | * This should only be called from xfs_reclaim(). | |
495 | */ | |
496 | void | |
497 | xfs_ireclaim(xfs_inode_t *ip) | |
498 | { | |
67fcaa73 | 499 | bhv_vnode_t *vp; |
1da177e4 LT |
500 | |
501 | /* | |
502 | * Remove from old hash list and mount list. | |
503 | */ | |
504 | XFS_STATS_INC(xs_ig_reclaims); | |
505 | ||
506 | xfs_iextract(ip); | |
507 | ||
508 | /* | |
509 | * Here we do a spurious inode lock in order to coordinate with | |
510 | * xfs_sync(). This is because xfs_sync() references the inodes | |
511 | * in the mount list without taking references on the corresponding | |
512 | * vnodes. We make that OK here by ensuring that we wait until | |
513 | * the inode is unlocked in xfs_sync() before we go ahead and | |
514 | * free it. We get both the regular lock and the io lock because | |
515 | * the xfs_sync() code may need to drop the regular one but will | |
516 | * still hold the io lock. | |
517 | */ | |
518 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
519 | ||
520 | /* | |
521 | * Release dquots (and their references) if any. An inode may escape | |
522 | * xfs_inactive and get here via vn_alloc->vn_reclaim path. | |
523 | */ | |
524 | XFS_QM_DQDETACH(ip->i_mount, ip); | |
525 | ||
526 | /* | |
527 | * Pull our behavior descriptor from the vnode chain. | |
528 | */ | |
529 | vp = XFS_ITOV_NULL(ip); | |
530 | if (vp) { | |
739bfb2a CH |
531 | vn_to_inode(vp)->i_private = NULL; |
532 | ip->i_vnode = NULL; | |
1da177e4 LT |
533 | } |
534 | ||
535 | /* | |
536 | * Free all memory associated with the inode. | |
537 | */ | |
439b8434 | 538 | xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); |
1da177e4 LT |
539 | xfs_idestroy(ip); |
540 | } | |
541 | ||
542 | /* | |
543 | * This routine removes an about-to-be-destroyed inode from | |
544 | * all of the lists in which it is located with the exception | |
545 | * of the behavior chain. | |
546 | */ | |
547 | void | |
548 | xfs_iextract( | |
549 | xfs_inode_t *ip) | |
550 | { | |
da353b0d DC |
551 | xfs_mount_t *mp = ip->i_mount; |
552 | xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino); | |
1da177e4 | 553 | xfs_inode_t *iq; |
da353b0d DC |
554 | |
555 | write_lock(&pag->pag_ici_lock); | |
556 | radix_tree_delete(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ip->i_ino)); | |
557 | write_unlock(&pag->pag_ici_lock); | |
558 | xfs_put_perag(mp, pag); | |
1da177e4 LT |
559 | |
560 | /* | |
da353b0d | 561 | * Remove from cluster list |
1da177e4 LT |
562 | */ |
563 | mp = ip->i_mount; | |
da353b0d DC |
564 | spin_lock(&ip->i_cluster->icl_lock); |
565 | hlist_del(&ip->i_cnode); | |
566 | spin_unlock(&ip->i_cluster->icl_lock); | |
567 | ||
568 | /* was last inode in cluster? */ | |
569 | if (hlist_empty(&ip->i_cluster->icl_inodes)) | |
570 | kmem_zone_free(xfs_icluster_zone, ip->i_cluster); | |
1da177e4 LT |
571 | |
572 | /* | |
573 | * Remove from mount's inode list. | |
574 | */ | |
575 | XFS_MOUNT_ILOCK(mp); | |
576 | ASSERT((ip->i_mnext != NULL) && (ip->i_mprev != NULL)); | |
577 | iq = ip->i_mnext; | |
578 | iq->i_mprev = ip->i_mprev; | |
579 | ip->i_mprev->i_mnext = iq; | |
580 | ||
581 | /* | |
582 | * Fix up the head pointer if it points to the inode being deleted. | |
583 | */ | |
584 | if (mp->m_inodes == ip) { | |
585 | if (ip == iq) { | |
586 | mp->m_inodes = NULL; | |
587 | } else { | |
588 | mp->m_inodes = iq; | |
589 | } | |
590 | } | |
591 | ||
592 | /* Deal with the deleted inodes list */ | |
593 | list_del_init(&ip->i_reclaim); | |
594 | ||
595 | mp->m_ireclaims++; | |
596 | XFS_MOUNT_IUNLOCK(mp); | |
597 | } | |
598 | ||
599 | /* | |
600 | * This is a wrapper routine around the xfs_ilock() routine | |
601 | * used to centralize some grungy code. It is used in places | |
602 | * that wish to lock the inode solely for reading the extents. | |
603 | * The reason these places can't just call xfs_ilock(SHARED) | |
604 | * is that the inode lock also guards to bringing in of the | |
605 | * extents from disk for a file in b-tree format. If the inode | |
606 | * is in b-tree format, then we need to lock the inode exclusively | |
607 | * until the extents are read in. Locking it exclusively all | |
608 | * the time would limit our parallelism unnecessarily, though. | |
609 | * What we do instead is check to see if the extents have been | |
610 | * read in yet, and only lock the inode exclusively if they | |
611 | * have not. | |
612 | * | |
613 | * The function returns a value which should be given to the | |
614 | * corresponding xfs_iunlock_map_shared(). This value is | |
615 | * the mode in which the lock was actually taken. | |
616 | */ | |
617 | uint | |
618 | xfs_ilock_map_shared( | |
619 | xfs_inode_t *ip) | |
620 | { | |
621 | uint lock_mode; | |
622 | ||
623 | if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) && | |
624 | ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) { | |
625 | lock_mode = XFS_ILOCK_EXCL; | |
626 | } else { | |
627 | lock_mode = XFS_ILOCK_SHARED; | |
628 | } | |
629 | ||
630 | xfs_ilock(ip, lock_mode); | |
631 | ||
632 | return lock_mode; | |
633 | } | |
634 | ||
635 | /* | |
636 | * This is simply the unlock routine to go with xfs_ilock_map_shared(). | |
637 | * All it does is call xfs_iunlock() with the given lock_mode. | |
638 | */ | |
639 | void | |
640 | xfs_iunlock_map_shared( | |
641 | xfs_inode_t *ip, | |
642 | unsigned int lock_mode) | |
643 | { | |
644 | xfs_iunlock(ip, lock_mode); | |
645 | } | |
646 | ||
647 | /* | |
648 | * The xfs inode contains 2 locks: a multi-reader lock called the | |
649 | * i_iolock and a multi-reader lock called the i_lock. This routine | |
650 | * allows either or both of the locks to be obtained. | |
651 | * | |
652 | * The 2 locks should always be ordered so that the IO lock is | |
653 | * obtained first in order to prevent deadlock. | |
654 | * | |
655 | * ip -- the inode being locked | |
656 | * lock_flags -- this parameter indicates the inode's locks | |
657 | * to be locked. It can be: | |
658 | * XFS_IOLOCK_SHARED, | |
659 | * XFS_IOLOCK_EXCL, | |
660 | * XFS_ILOCK_SHARED, | |
661 | * XFS_ILOCK_EXCL, | |
662 | * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED, | |
663 | * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL, | |
664 | * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED, | |
665 | * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL | |
666 | */ | |
667 | void | |
668 | xfs_ilock(xfs_inode_t *ip, | |
669 | uint lock_flags) | |
670 | { | |
671 | /* | |
672 | * You can't set both SHARED and EXCL for the same lock, | |
673 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
674 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
675 | */ | |
676 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
677 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
678 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
679 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
f7c66ce3 | 680 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); |
1da177e4 LT |
681 | |
682 | if (lock_flags & XFS_IOLOCK_EXCL) { | |
f7c66ce3 | 683 | mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); |
1da177e4 | 684 | } else if (lock_flags & XFS_IOLOCK_SHARED) { |
f7c66ce3 | 685 | mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); |
1da177e4 LT |
686 | } |
687 | if (lock_flags & XFS_ILOCK_EXCL) { | |
f7c66ce3 | 688 | mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); |
1da177e4 | 689 | } else if (lock_flags & XFS_ILOCK_SHARED) { |
f7c66ce3 | 690 | mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); |
1da177e4 LT |
691 | } |
692 | xfs_ilock_trace(ip, 1, lock_flags, (inst_t *)__return_address); | |
693 | } | |
694 | ||
695 | /* | |
696 | * This is just like xfs_ilock(), except that the caller | |
697 | * is guaranteed not to sleep. It returns 1 if it gets | |
698 | * the requested locks and 0 otherwise. If the IO lock is | |
699 | * obtained but the inode lock cannot be, then the IO lock | |
700 | * is dropped before returning. | |
701 | * | |
702 | * ip -- the inode being locked | |
703 | * lock_flags -- this parameter indicates the inode's locks to be | |
704 | * to be locked. See the comment for xfs_ilock() for a list | |
705 | * of valid values. | |
706 | * | |
707 | */ | |
708 | int | |
709 | xfs_ilock_nowait(xfs_inode_t *ip, | |
710 | uint lock_flags) | |
711 | { | |
712 | int iolocked; | |
713 | int ilocked; | |
714 | ||
715 | /* | |
716 | * You can't set both SHARED and EXCL for the same lock, | |
717 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
718 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
719 | */ | |
720 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
721 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
722 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
723 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
f7c66ce3 | 724 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); |
1da177e4 LT |
725 | |
726 | iolocked = 0; | |
727 | if (lock_flags & XFS_IOLOCK_EXCL) { | |
728 | iolocked = mrtryupdate(&ip->i_iolock); | |
729 | if (!iolocked) { | |
730 | return 0; | |
731 | } | |
732 | } else if (lock_flags & XFS_IOLOCK_SHARED) { | |
733 | iolocked = mrtryaccess(&ip->i_iolock); | |
734 | if (!iolocked) { | |
735 | return 0; | |
736 | } | |
737 | } | |
738 | if (lock_flags & XFS_ILOCK_EXCL) { | |
739 | ilocked = mrtryupdate(&ip->i_lock); | |
740 | if (!ilocked) { | |
741 | if (iolocked) { | |
742 | mrunlock(&ip->i_iolock); | |
743 | } | |
744 | return 0; | |
745 | } | |
746 | } else if (lock_flags & XFS_ILOCK_SHARED) { | |
747 | ilocked = mrtryaccess(&ip->i_lock); | |
748 | if (!ilocked) { | |
749 | if (iolocked) { | |
750 | mrunlock(&ip->i_iolock); | |
751 | } | |
752 | return 0; | |
753 | } | |
754 | } | |
755 | xfs_ilock_trace(ip, 2, lock_flags, (inst_t *)__return_address); | |
756 | return 1; | |
757 | } | |
758 | ||
759 | /* | |
760 | * xfs_iunlock() is used to drop the inode locks acquired with | |
761 | * xfs_ilock() and xfs_ilock_nowait(). The caller must pass | |
762 | * in the flags given to xfs_ilock() or xfs_ilock_nowait() so | |
763 | * that we know which locks to drop. | |
764 | * | |
765 | * ip -- the inode being unlocked | |
766 | * lock_flags -- this parameter indicates the inode's locks to be | |
767 | * to be unlocked. See the comment for xfs_ilock() for a list | |
768 | * of valid values for this parameter. | |
769 | * | |
770 | */ | |
771 | void | |
772 | xfs_iunlock(xfs_inode_t *ip, | |
773 | uint lock_flags) | |
774 | { | |
775 | /* | |
776 | * You can't set both SHARED and EXCL for the same lock, | |
777 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
778 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
779 | */ | |
780 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
781 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
782 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
783 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
f7c66ce3 LM |
784 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_IUNLOCK_NONOTIFY | |
785 | XFS_LOCK_DEP_MASK)) == 0); | |
1da177e4 LT |
786 | ASSERT(lock_flags != 0); |
787 | ||
788 | if (lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) { | |
789 | ASSERT(!(lock_flags & XFS_IOLOCK_SHARED) || | |
790 | (ismrlocked(&ip->i_iolock, MR_ACCESS))); | |
791 | ASSERT(!(lock_flags & XFS_IOLOCK_EXCL) || | |
792 | (ismrlocked(&ip->i_iolock, MR_UPDATE))); | |
793 | mrunlock(&ip->i_iolock); | |
794 | } | |
795 | ||
796 | if (lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) { | |
797 | ASSERT(!(lock_flags & XFS_ILOCK_SHARED) || | |
798 | (ismrlocked(&ip->i_lock, MR_ACCESS))); | |
799 | ASSERT(!(lock_flags & XFS_ILOCK_EXCL) || | |
800 | (ismrlocked(&ip->i_lock, MR_UPDATE))); | |
801 | mrunlock(&ip->i_lock); | |
802 | ||
803 | /* | |
804 | * Let the AIL know that this item has been unlocked in case | |
805 | * it is in the AIL and anyone is waiting on it. Don't do | |
806 | * this if the caller has asked us not to. | |
807 | */ | |
808 | if (!(lock_flags & XFS_IUNLOCK_NONOTIFY) && | |
809 | ip->i_itemp != NULL) { | |
810 | xfs_trans_unlocked_item(ip->i_mount, | |
811 | (xfs_log_item_t*)(ip->i_itemp)); | |
812 | } | |
813 | } | |
814 | xfs_ilock_trace(ip, 3, lock_flags, (inst_t *)__return_address); | |
815 | } | |
816 | ||
817 | /* | |
818 | * give up write locks. the i/o lock cannot be held nested | |
819 | * if it is being demoted. | |
820 | */ | |
821 | void | |
822 | xfs_ilock_demote(xfs_inode_t *ip, | |
823 | uint lock_flags) | |
824 | { | |
825 | ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)); | |
826 | ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0); | |
827 | ||
828 | if (lock_flags & XFS_ILOCK_EXCL) { | |
829 | ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE)); | |
830 | mrdemote(&ip->i_lock); | |
831 | } | |
832 | if (lock_flags & XFS_IOLOCK_EXCL) { | |
833 | ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE)); | |
834 | mrdemote(&ip->i_iolock); | |
835 | } | |
836 | } | |
837 | ||
838 | /* | |
839 | * The following three routines simply manage the i_flock | |
840 | * semaphore embedded in the inode. This semaphore synchronizes | |
841 | * processes attempting to flush the in-core inode back to disk. | |
842 | */ | |
843 | void | |
844 | xfs_iflock(xfs_inode_t *ip) | |
845 | { | |
846 | psema(&(ip->i_flock), PINOD|PLTWAIT); | |
847 | } | |
848 | ||
849 | int | |
850 | xfs_iflock_nowait(xfs_inode_t *ip) | |
851 | { | |
852 | return (cpsema(&(ip->i_flock))); | |
853 | } | |
854 | ||
855 | void | |
856 | xfs_ifunlock(xfs_inode_t *ip) | |
857 | { | |
0d8fee32 | 858 | ASSERT(issemalocked(&(ip->i_flock))); |
1da177e4 LT |
859 | vsema(&(ip->i_flock)); |
860 | } |