blob: b07604b94d9f32c2f15e7eb08c5e83b458b3fd42 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Nathan Scott7b718762005-11-02 14:58:39 +11002 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * 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.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * 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
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110019#include "xfs_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include "xfs_types.h"
Nathan Scotta844f452005-11-02 14:38:42 +110021#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110023#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_trans.h"
25#include "xfs_sb.h"
26#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_dir2.h"
28#include "xfs_dmapi.h"
29#include "xfs_mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110031#include "xfs_alloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include "xfs_ialloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "xfs_dir2_sf.h"
Nathan Scotta844f452005-11-02 14:38:42 +110034#include "xfs_attr_sf.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include "xfs_dinode.h"
36#include "xfs_inode.h"
Nathan Scotta844f452005-11-02 14:38:42 +110037#include "xfs_btree.h"
38#include "xfs_ialloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include "xfs_quota.h"
40#include "xfs_utils.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
42/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070043 * Look up an inode by number in the given file system.
David Chinnerda353b02007-08-28 14:00:13 +100044 * 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.
Linus Torvalds1da177e2005-04-16 15:20:36 -070047 *
David Chinnerda353b02007-08-28 14:00:13 +100048 * 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.
Linus Torvalds1da177e2005-04-16 15:20:36 -070050 *
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 */
66STATIC int
67xfs_iget_core(
Christoph Hellwig10090be2007-10-11 18:11:03 +100068 struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 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{
Christoph Hellwig10090be2007-10-11 18:11:03 +100077 struct inode *old_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -070078 xfs_inode_t *ip;
79 xfs_inode_t *iq;
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 int error;
David Chinnerda353b02007-08-28 14:00:13 +100081 unsigned long first_index, mask;
82 xfs_perag_t *pag;
83 xfs_agino_t agino;
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
David Chinnerda353b02007-08-28 14:00:13 +100085 /* the radix tree exists only in inode capable AGs */
86 if (XFS_INO_TO_AGNO(mp, ino) >= mp->m_maxagi)
87 return EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
David Chinnerda353b02007-08-28 14:00:13 +100089 /* get the perag structure and ensure that it's inode capable */
90 pag = xfs_get_perag(mp, ino);
91 if (!pag->pagi_inodeok)
92 return EINVAL;
93 ASSERT(pag->pag_ici_init);
94 agino = XFS_INO_TO_AGINO(mp, ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -070095
96again:
David Chinnerda353b02007-08-28 14:00:13 +100097 read_lock(&pag->pag_ici_lock);
98 ip = radix_tree_lookup(&pag->pag_ici_root, agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -070099
David Chinnerda353b02007-08-28 14:00:13 +1000100 if (ip != NULL) {
101 /*
102 * If INEW is set this inode is being set up
103 * we need to pause and try again.
104 */
105 if (xfs_iflags_test(ip, XFS_INEW)) {
106 read_unlock(&pag->pag_ici_lock);
107 delay(1);
108 XFS_STATS_INC(xs_ig_frecycle);
109
110 goto again;
111 }
112
Christoph Hellwig10090be2007-10-11 18:11:03 +1000113 old_inode = ip->i_vnode;
114 if (old_inode == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 /*
David Chinnerda353b02007-08-28 14:00:13 +1000116 * If IRECLAIM is set this inode is
117 * on its way out of the system,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 * we need to pause and try again.
119 */
David Chinnerda353b02007-08-28 14:00:13 +1000120 if (xfs_iflags_test(ip, XFS_IRECLAIM)) {
121 read_unlock(&pag->pag_ici_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 delay(1);
123 XFS_STATS_INC(xs_ig_frecycle);
124
125 goto again;
126 }
David Chinnerda353b02007-08-28 14:00:13 +1000127 ASSERT(xfs_iflags_test(ip, XFS_IRECLAIMABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128
David Chinnerda353b02007-08-28 14:00:13 +1000129 /*
130 * If lookup is racing with unlink, then we
131 * should return an error immediately so we
132 * don't remove it from the reclaim list and
133 * potentially leak the inode.
134 */
135 if ((ip->i_d.di_mode == 0) &&
136 !(flags & XFS_IGET_CREATE)) {
137 read_unlock(&pag->pag_ici_lock);
138 xfs_put_perag(mp, pag);
139 return ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 }
141
Lachlan McIlroy15947f22007-09-17 13:11:58 +1000142 xfs_itrace_exit_tag(ip, "xfs_iget.alloc");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143
David Chinnerda353b02007-08-28 14:00:13 +1000144 XFS_STATS_INC(xs_ig_found);
David Chinnerda353b02007-08-28 14:00:13 +1000145 xfs_iflags_clear(ip, XFS_IRECLAIMABLE);
146 read_unlock(&pag->pag_ici_lock);
147
148 XFS_MOUNT_ILOCK(mp);
149 list_del_init(&ip->i_reclaim);
150 XFS_MOUNT_IUNLOCK(mp);
151
152 goto finish_inode;
153
Christoph Hellwig10090be2007-10-11 18:11:03 +1000154 } else if (inode != old_inode) {
David Chinnerda353b02007-08-28 14:00:13 +1000155 /* The inode is being torn down, pause and
156 * try again.
157 */
Christoph Hellwig10090be2007-10-11 18:11:03 +1000158 if (old_inode->i_state & (I_FREEING | I_CLEAR)) {
David Chinnerda353b02007-08-28 14:00:13 +1000159 read_unlock(&pag->pag_ici_lock);
160 delay(1);
161 XFS_STATS_INC(xs_ig_frecycle);
162
163 goto again;
164 }
165/* Chances are the other vnode (the one in the inode) is being torn
166* down right now, and we landed on top of it. Question is, what do
167* we do? Unhook the old inode and hook up the new one?
168*/
169 cmn_err(CE_PANIC,
170 "xfs_iget_core: ambiguous vns: vp/0x%p, invp/0x%p",
Christoph Hellwig10090be2007-10-11 18:11:03 +1000171 old_inode, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 }
David Chinnerda353b02007-08-28 14:00:13 +1000173
174 /*
175 * Inode cache hit
176 */
177 read_unlock(&pag->pag_ici_lock);
178 XFS_STATS_INC(xs_ig_found);
179
180finish_inode:
Christoph Hellwig613d7042007-10-11 17:44:08 +1000181 if (ip->i_d.di_mode == 0 && !(flags & XFS_IGET_CREATE)) {
182 xfs_put_perag(mp, pag);
183 return ENOENT;
David Chinnerda353b02007-08-28 14:00:13 +1000184 }
185
186 if (lock_flags != 0)
187 xfs_ilock(ip, lock_flags);
188
189 xfs_iflags_clear(ip, XFS_ISTALE);
Lachlan McIlroy15947f22007-09-17 13:11:58 +1000190 xfs_itrace_exit_tag(ip, "xfs_iget.found");
David Chinnerda353b02007-08-28 14:00:13 +1000191 goto return_ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 }
193
194 /*
David Chinnerda353b02007-08-28 14:00:13 +1000195 * Inode cache miss
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 */
David Chinnerda353b02007-08-28 14:00:13 +1000197 read_unlock(&pag->pag_ici_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 XFS_STATS_INC(xs_ig_missed);
199
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 /*
201 * Read the disk inode attributes into a new inode structure and get
202 * a new vnode for it. This should also initialize i_ino and i_mount.
203 */
Nathan Scott745b1f472006-09-28 11:02:23 +1000204 error = xfs_iread(mp, tp, ino, &ip, bno,
205 (flags & XFS_IGET_BULKSTAT) ? XFS_IMAP_BULKSTAT : 0);
David Chinnerda353b02007-08-28 14:00:13 +1000206 if (error) {
207 xfs_put_perag(mp, pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 return error;
David Chinnerda353b02007-08-28 14:00:13 +1000209 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210
Lachlan McIlroy15947f22007-09-17 13:11:58 +1000211 xfs_itrace_exit_tag(ip, "xfs_iget.alloc");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212
Christoph Hellwig10090be2007-10-11 18:11:03 +1000213
214 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
215 "xfsino", ip->i_ino);
216 mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino);
217 init_waitqueue_head(&ip->i_ipin_wait);
218 atomic_set(&ip->i_pincount, 0);
219 initnsema(&ip->i_flock, 1, "xfsfino");
220
Nathan Scott745b1f472006-09-28 11:02:23 +1000221 if (lock_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 xfs_ilock(ip, lock_flags);
Nathan Scott745b1f472006-09-28 11:02:23 +1000223
224 if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 xfs_idestroy(ip);
David Chinnerda353b02007-08-28 14:00:13 +1000226 xfs_put_perag(mp, pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 return ENOENT;
228 }
229
230 /*
David Chinnerbad55842008-03-06 13:43:49 +1100231 * Preload the radix tree so we can insert safely under the
232 * write spinlock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 */
David Chinnerda353b02007-08-28 14:00:13 +1000234 if (radix_tree_preload(GFP_KERNEL)) {
David Chinner72772a32008-03-06 13:49:43 +1100235 xfs_idestroy(ip);
David Chinnerda353b02007-08-28 14:00:13 +1000236 delay(1);
237 goto again;
238 }
239 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
240 first_index = agino & mask;
241 write_lock(&pag->pag_ici_lock);
David Chinnerda353b02007-08-28 14:00:13 +1000242 /*
243 * insert the new inode
244 */
245 error = radix_tree_insert(&pag->pag_ici_root, agino, ip);
246 if (unlikely(error)) {
247 BUG_ON(error != -EEXIST);
248 write_unlock(&pag->pag_ici_lock);
249 radix_tree_preload_end();
250 xfs_idestroy(ip);
251 XFS_STATS_INC(xs_ig_dup);
252 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 }
254
255 /*
David Chinnerbad55842008-03-06 13:43:49 +1100256 * These values _must_ be set before releasing the radix tree lock!
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 ip->i_udquot = ip->i_gdquot = NULL;
David Chinner7a18c382006-11-11 18:04:54 +1100259 xfs_iflags_set(ip, XFS_INEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260
David Chinnerda353b02007-08-28 14:00:13 +1000261 write_unlock(&pag->pag_ici_lock);
262 radix_tree_preload_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263
264 /*
265 * Link ip to its mount and thread it on the mount's inode list.
266 */
267 XFS_MOUNT_ILOCK(mp);
268 if ((iq = mp->m_inodes)) {
269 ASSERT(iq->i_mprev->i_mnext == iq);
270 ip->i_mprev = iq->i_mprev;
271 iq->i_mprev->i_mnext = ip;
272 iq->i_mprev = ip;
273 ip->i_mnext = iq;
274 } else {
275 ip->i_mnext = ip;
276 ip->i_mprev = ip;
277 }
278 mp->m_inodes = ip;
279
280 XFS_MOUNT_IUNLOCK(mp);
David Chinnerda353b02007-08-28 14:00:13 +1000281 xfs_put_perag(mp, pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282
283 return_ip:
284 ASSERT(ip->i_df.if_ext_max ==
285 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t));
286
Christoph Hellwigb3aea4e2007-08-29 11:44:37 +1000287 xfs_iflags_set(ip, XFS_IMODIFIED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 *ipp = ip;
289
290 /*
291 * If we have a real type for an on-disk inode, we can set ops(&unlock)
292 * now. If it's a new inode being created, xfs_ialloc will handle it.
293 */
Christoph Hellwig10090be2007-10-11 18:11:03 +1000294 xfs_initialize_vnode(mp, inode, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 return 0;
296}
297
298
299/*
300 * The 'normal' internal xfs_iget, if needed it will
301 * 'allocate', or 'get', the vnode.
302 */
303int
304xfs_iget(
305 xfs_mount_t *mp,
306 xfs_trans_t *tp,
307 xfs_ino_t ino,
308 uint flags,
309 uint lock_flags,
310 xfs_inode_t **ipp,
311 xfs_daddr_t bno)
312{
313 struct inode *inode;
Christoph Hellwig10090be2007-10-11 18:11:03 +1000314 xfs_inode_t *ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 int error;
316
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 XFS_STATS_INC(xs_ig_attempts);
318
Christoph Hellwigba403ab2005-09-05 08:33:00 +1000319retry:
Christoph Hellwigb267ce92007-08-30 17:21:30 +1000320 inode = iget_locked(mp->m_super, ino);
Christoph Hellwig10090be2007-10-11 18:11:03 +1000321 if (!inode)
322 /* If we got no inode we are out of memory */
323 return ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324
Christoph Hellwig10090be2007-10-11 18:11:03 +1000325 if (inode->i_state & I_NEW) {
326 XFS_STATS_INC(vn_active);
327 XFS_STATS_INC(vn_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328
Christoph Hellwig10090be2007-10-11 18:11:03 +1000329 error = xfs_iget_core(inode, mp, tp, ino, flags,
330 lock_flags, ipp, bno);
331 if (error) {
332 make_bad_inode(inode);
333 if (inode->i_state & I_NEW)
334 unlock_new_inode(inode);
335 iput(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 }
Christoph Hellwig10090be2007-10-11 18:11:03 +1000337 return error;
338 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339
Christoph Hellwig10090be2007-10-11 18:11:03 +1000340 /*
341 * If the inode is not fully constructed due to
342 * filehandle mismatches wait for the inode to go
343 * away and try again.
344 *
345 * iget_locked will call __wait_on_freeing_inode
346 * to wait for the inode to go away.
347 */
348 if (is_bad_inode(inode)) {
349 iput(inode);
350 delay(1);
351 goto retry;
352 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353
Christoph Hellwig10090be2007-10-11 18:11:03 +1000354 ip = XFS_I(inode);
355 if (!ip) {
356 iput(inode);
357 delay(1);
358 goto retry;
359 }
360
361 if (lock_flags != 0)
362 xfs_ilock(ip, lock_flags);
363 XFS_STATS_INC(xs_ig_found);
364 *ipp = ip;
365 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366}
367
368/*
369 * Look for the inode corresponding to the given ino in the hash table.
370 * If it is there and its i_transp pointer matches tp, return it.
371 * Otherwise, return NULL.
372 */
373xfs_inode_t *
374xfs_inode_incore(xfs_mount_t *mp,
375 xfs_ino_t ino,
376 xfs_trans_t *tp)
377{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 xfs_inode_t *ip;
David Chinnerda353b02007-08-28 14:00:13 +1000379 xfs_perag_t *pag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380
David Chinnerda353b02007-08-28 14:00:13 +1000381 pag = xfs_get_perag(mp, ino);
382 read_lock(&pag->pag_ici_lock);
383 ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ino));
384 read_unlock(&pag->pag_ici_lock);
385 xfs_put_perag(mp, pag);
386
387 /* the returned inode must match the transaction */
388 if (ip && (ip->i_transp != tp))
389 return NULL;
390 return ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391}
392
393/*
394 * Decrement reference count of an inode structure and unlock it.
395 *
396 * ip -- the inode being released
397 * lock_flags -- this parameter indicates the inode's locks to be
398 * to be released. See the comment on xfs_iunlock() for a list
399 * of valid values.
400 */
401void
402xfs_iput(xfs_inode_t *ip,
403 uint lock_flags)
404{
Lachlan McIlroycf441ee2008-02-07 16:42:19 +1100405 xfs_itrace_entry(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 xfs_iunlock(ip, lock_flags);
Christoph Hellwig10090be2007-10-11 18:11:03 +1000407 IRELE(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408}
409
410/*
411 * Special iput for brand-new inodes that are still locked
412 */
413void
414xfs_iput_new(xfs_inode_t *ip,
415 uint lock_flags)
416{
Christoph Hellwig10090be2007-10-11 18:11:03 +1000417 struct inode *inode = ip->i_vnode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418
Lachlan McIlroycf441ee2008-02-07 16:42:19 +1100419 xfs_itrace_entry(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420
421 if ((ip->i_d.di_mode == 0)) {
David Chinner7a18c382006-11-11 18:04:54 +1100422 ASSERT(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
Christoph Hellwig10090be2007-10-11 18:11:03 +1000423 make_bad_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 }
425 if (inode->i_state & I_NEW)
426 unlock_new_inode(inode);
427 if (lock_flags)
428 xfs_iunlock(ip, lock_flags);
Christoph Hellwig10090be2007-10-11 18:11:03 +1000429 IRELE(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430}
431
432
433/*
434 * This routine embodies the part of the reclaim code that pulls
435 * the inode from the inode hash table and the mount structure's
436 * inode list.
437 * This should only be called from xfs_reclaim().
438 */
439void
440xfs_ireclaim(xfs_inode_t *ip)
441{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 /*
443 * Remove from old hash list and mount list.
444 */
445 XFS_STATS_INC(xs_ig_reclaims);
446
447 xfs_iextract(ip);
448
449 /*
450 * Here we do a spurious inode lock in order to coordinate with
451 * xfs_sync(). This is because xfs_sync() references the inodes
452 * in the mount list without taking references on the corresponding
453 * vnodes. We make that OK here by ensuring that we wait until
454 * the inode is unlocked in xfs_sync() before we go ahead and
455 * free it. We get both the regular lock and the io lock because
456 * the xfs_sync() code may need to drop the regular one but will
457 * still hold the io lock.
458 */
459 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
460
461 /*
462 * Release dquots (and their references) if any. An inode may escape
463 * xfs_inactive and get here via vn_alloc->vn_reclaim path.
464 */
465 XFS_QM_DQDETACH(ip->i_mount, ip);
466
467 /*
468 * Pull our behavior descriptor from the vnode chain.
469 */
Christoph Hellwig10090be2007-10-11 18:11:03 +1000470 if (ip->i_vnode) {
471 ip->i_vnode->i_private = NULL;
Christoph Hellwig739bfb22007-08-29 10:58:01 +1000472 ip->i_vnode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 }
474
475 /*
476 * Free all memory associated with the inode.
477 */
Tim Shimmin439b8432006-11-11 18:04:34 +1100478 xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 xfs_idestroy(ip);
480}
481
482/*
483 * This routine removes an about-to-be-destroyed inode from
484 * all of the lists in which it is located with the exception
485 * of the behavior chain.
486 */
487void
488xfs_iextract(
489 xfs_inode_t *ip)
490{
David Chinnerda353b02007-08-28 14:00:13 +1000491 xfs_mount_t *mp = ip->i_mount;
492 xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 xfs_inode_t *iq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494
David Chinnerda353b02007-08-28 14:00:13 +1000495 write_lock(&pag->pag_ici_lock);
496 radix_tree_delete(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ip->i_ino));
497 write_unlock(&pag->pag_ici_lock);
498 xfs_put_perag(mp, pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499
500 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 * Remove from mount's inode list.
502 */
503 XFS_MOUNT_ILOCK(mp);
504 ASSERT((ip->i_mnext != NULL) && (ip->i_mprev != NULL));
505 iq = ip->i_mnext;
506 iq->i_mprev = ip->i_mprev;
507 ip->i_mprev->i_mnext = iq;
508
509 /*
510 * Fix up the head pointer if it points to the inode being deleted.
511 */
512 if (mp->m_inodes == ip) {
513 if (ip == iq) {
514 mp->m_inodes = NULL;
515 } else {
516 mp->m_inodes = iq;
517 }
518 }
519
520 /* Deal with the deleted inodes list */
521 list_del_init(&ip->i_reclaim);
522
523 mp->m_ireclaims++;
524 XFS_MOUNT_IUNLOCK(mp);
525}
526
527/*
528 * This is a wrapper routine around the xfs_ilock() routine
529 * used to centralize some grungy code. It is used in places
530 * that wish to lock the inode solely for reading the extents.
531 * The reason these places can't just call xfs_ilock(SHARED)
532 * is that the inode lock also guards to bringing in of the
533 * extents from disk for a file in b-tree format. If the inode
534 * is in b-tree format, then we need to lock the inode exclusively
535 * until the extents are read in. Locking it exclusively all
536 * the time would limit our parallelism unnecessarily, though.
537 * What we do instead is check to see if the extents have been
538 * read in yet, and only lock the inode exclusively if they
539 * have not.
540 *
541 * The function returns a value which should be given to the
542 * corresponding xfs_iunlock_map_shared(). This value is
543 * the mode in which the lock was actually taken.
544 */
545uint
546xfs_ilock_map_shared(
547 xfs_inode_t *ip)
548{
549 uint lock_mode;
550
551 if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) &&
552 ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) {
553 lock_mode = XFS_ILOCK_EXCL;
554 } else {
555 lock_mode = XFS_ILOCK_SHARED;
556 }
557
558 xfs_ilock(ip, lock_mode);
559
560 return lock_mode;
561}
562
563/*
564 * This is simply the unlock routine to go with xfs_ilock_map_shared().
565 * All it does is call xfs_iunlock() with the given lock_mode.
566 */
567void
568xfs_iunlock_map_shared(
569 xfs_inode_t *ip,
570 unsigned int lock_mode)
571{
572 xfs_iunlock(ip, lock_mode);
573}
574
575/*
576 * The xfs inode contains 2 locks: a multi-reader lock called the
577 * i_iolock and a multi-reader lock called the i_lock. This routine
578 * allows either or both of the locks to be obtained.
579 *
580 * The 2 locks should always be ordered so that the IO lock is
581 * obtained first in order to prevent deadlock.
582 *
583 * ip -- the inode being locked
584 * lock_flags -- this parameter indicates the inode's locks
585 * to be locked. It can be:
586 * XFS_IOLOCK_SHARED,
587 * XFS_IOLOCK_EXCL,
588 * XFS_ILOCK_SHARED,
589 * XFS_ILOCK_EXCL,
590 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
591 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
592 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
593 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
594 */
595void
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000596xfs_ilock(
597 xfs_inode_t *ip,
598 uint lock_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599{
600 /*
601 * You can't set both SHARED and EXCL for the same lock,
602 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
603 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
604 */
605 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
606 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
607 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
608 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000609 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000611 if (lock_flags & XFS_IOLOCK_EXCL)
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000612 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000613 else if (lock_flags & XFS_IOLOCK_SHARED)
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000614 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000615
616 if (lock_flags & XFS_ILOCK_EXCL)
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000617 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000618 else if (lock_flags & XFS_ILOCK_SHARED)
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000619 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000620
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 xfs_ilock_trace(ip, 1, lock_flags, (inst_t *)__return_address);
622}
623
624/*
625 * This is just like xfs_ilock(), except that the caller
626 * is guaranteed not to sleep. It returns 1 if it gets
627 * the requested locks and 0 otherwise. If the IO lock is
628 * obtained but the inode lock cannot be, then the IO lock
629 * is dropped before returning.
630 *
631 * ip -- the inode being locked
632 * lock_flags -- this parameter indicates the inode's locks to be
633 * to be locked. See the comment for xfs_ilock() for a list
634 * of valid values.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 */
636int
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000637xfs_ilock_nowait(
638 xfs_inode_t *ip,
639 uint lock_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 /*
642 * You can't set both SHARED and EXCL for the same lock,
643 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
644 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
645 */
646 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
647 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
648 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
649 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000650 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 if (lock_flags & XFS_IOLOCK_EXCL) {
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000653 if (!mrtryupdate(&ip->i_iolock))
654 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 } else if (lock_flags & XFS_IOLOCK_SHARED) {
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000656 if (!mrtryaccess(&ip->i_iolock))
657 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 }
659 if (lock_flags & XFS_ILOCK_EXCL) {
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000660 if (!mrtryupdate(&ip->i_lock))
661 goto out_undo_iolock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 } else if (lock_flags & XFS_ILOCK_SHARED) {
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000663 if (!mrtryaccess(&ip->i_lock))
664 goto out_undo_iolock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 }
666 xfs_ilock_trace(ip, 2, lock_flags, (inst_t *)__return_address);
667 return 1;
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000668
669 out_undo_iolock:
670 if (lock_flags & XFS_IOLOCK_EXCL)
671 mrunlock_excl(&ip->i_iolock);
672 else if (lock_flags & XFS_IOLOCK_SHARED)
673 mrunlock_shared(&ip->i_iolock);
674 out:
675 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676}
677
678/*
679 * xfs_iunlock() is used to drop the inode locks acquired with
680 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
681 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
682 * that we know which locks to drop.
683 *
684 * ip -- the inode being unlocked
685 * lock_flags -- this parameter indicates the inode's locks to be
686 * to be unlocked. See the comment for xfs_ilock() for a list
687 * of valid values for this parameter.
688 *
689 */
690void
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000691xfs_iunlock(
692 xfs_inode_t *ip,
693 uint lock_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694{
695 /*
696 * You can't set both SHARED and EXCL for the same lock,
697 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
698 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
699 */
700 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
701 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
702 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
703 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000704 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_IUNLOCK_NONOTIFY |
705 XFS_LOCK_DEP_MASK)) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 ASSERT(lock_flags != 0);
707
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000708 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);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000713 if (lock_flags & XFS_ILOCK_EXCL)
714 mrunlock_excl(&ip->i_lock);
715 else if (lock_flags & XFS_ILOCK_SHARED)
716 mrunlock_shared(&ip->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000718 if ((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) &&
719 !(lock_flags & XFS_IUNLOCK_NONOTIFY) && ip->i_itemp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 /*
721 * Let the AIL know that this item has been unlocked in case
722 * it is in the AIL and anyone is waiting on it. Don't do
723 * this if the caller has asked us not to.
724 */
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000725 xfs_trans_unlocked_item(ip->i_mount,
726 (xfs_log_item_t*)(ip->i_itemp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 }
728 xfs_ilock_trace(ip, 3, lock_flags, (inst_t *)__return_address);
729}
730
731/*
732 * give up write locks. the i/o lock cannot be held nested
733 * if it is being demoted.
734 */
735void
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000736xfs_ilock_demote(
737 xfs_inode_t *ip,
738 uint lock_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739{
740 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
741 ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
742
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000743 if (lock_flags & XFS_ILOCK_EXCL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 mrdemote(&ip->i_lock);
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000745 if (lock_flags & XFS_IOLOCK_EXCL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 mrdemote(&ip->i_iolock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747}
748
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000749#ifdef DEBUG
750/*
751 * Debug-only routine, without additional rw_semaphore APIs, we can
752 * now only answer requests regarding whether we hold the lock for write
753 * (reader state is outside our visibility, we only track writer state).
754 *
755 * Note: this means !xfs_isilocked would give false positives, so don't do that.
756 */
757int
758xfs_isilocked(
759 xfs_inode_t *ip,
760 uint lock_flags)
761{
762 if ((lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) ==
763 XFS_ILOCK_EXCL) {
764 if (!ip->i_lock.mr_writer)
765 return 0;
766 }
767
768 if ((lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) ==
769 XFS_IOLOCK_EXCL) {
770 if (!ip->i_iolock.mr_writer)
771 return 0;
772 }
773
774 return 1;
775}
776#endif
777
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778/*
779 * The following three routines simply manage the i_flock
780 * semaphore embedded in the inode. This semaphore synchronizes
781 * processes attempting to flush the in-core inode back to disk.
782 */
783void
784xfs_iflock(xfs_inode_t *ip)
785{
786 psema(&(ip->i_flock), PINOD|PLTWAIT);
787}
788
789int
790xfs_iflock_nowait(xfs_inode_t *ip)
791{
792 return (cpsema(&(ip->i_flock)));
793}
794
795void
796xfs_ifunlock(xfs_inode_t *ip)
797{
Al Viro0d8fee32006-06-19 08:41:30 +1000798 ASSERT(issemalocked(&(ip->i_flock)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 vsema(&(ip->i_flock));
800}