blob: 155e798f30a166a9697fd6fad56f42163b6340c7 [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"
Christoph Hellwigef14f0c2009-06-10 17:07:47 +020021#include "xfs_acl.h"
Nathan Scotta844f452005-11-02 14:38:42 +110022#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110024#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include "xfs_trans.h"
26#include "xfs_sb.h"
27#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include "xfs_dir2.h"
29#include "xfs_dmapi.h"
30#include "xfs_mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110032#include "xfs_alloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "xfs_ialloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include "xfs_dir2_sf.h"
Nathan Scotta844f452005-11-02 14:38:42 +110035#include "xfs_attr_sf.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include "xfs_dinode.h"
37#include "xfs_inode.h"
Nathan Scotta844f452005-11-02 14:38:42 +110038#include "xfs_btree.h"
39#include "xfs_ialloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include "xfs_quota.h"
41#include "xfs_utils.h"
David Chinner783a2f62008-10-30 17:39:58 +110042#include "xfs_trans_priv.h"
43#include "xfs_inode_item.h"
Christoph Hellwig24f211b2008-11-28 14:23:42 +110044#include "xfs_bmap.h"
45#include "xfs_btree_trace.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000046#include "xfs_trace.h"
Christoph Hellwig24f211b2008-11-28 14:23:42 +110047
48
49/*
50 * Allocate and initialise an xfs_inode.
51 */
52STATIC struct xfs_inode *
53xfs_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;
Christoph Hellwig54e34622009-08-07 14:38:25 -030067 if (inode_init_always(mp->m_super, VFS_I(ip))) {
68 kmem_zone_free(xfs_inode_zone, ip);
69 return NULL;
70 }
Christoph Hellwig24f211b2008-11-28 14:23:42 +110071
72 ASSERT(atomic_read(&ip->i_iocount) == 0);
73 ASSERT(atomic_read(&ip->i_pincount) == 0);
74 ASSERT(!spin_is_locked(&ip->i_flags_lock));
75 ASSERT(completion_done(&ip->i_flush));
Christoph Hellwig033da482009-10-19 04:05:26 +000076
77 mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino);
Christoph Hellwig24f211b2008-11-28 14:23:42 +110078
Christoph Hellwig24f211b2008-11-28 14:23:42 +110079 /* initialise the xfs inode */
80 ip->i_ino = ino;
81 ip->i_mount = mp;
82 memset(&ip->i_imap, 0, sizeof(struct xfs_imap));
83 ip->i_afp = NULL;
84 memset(&ip->i_df, 0, sizeof(xfs_ifork_t));
85 ip->i_flags = 0;
86 ip->i_update_core = 0;
Christoph Hellwig24f211b2008-11-28 14:23:42 +110087 ip->i_delayed_blks = 0;
88 memset(&ip->i_d, 0, sizeof(xfs_icdinode_t));
89 ip->i_size = 0;
90 ip->i_new_size = 0;
91
Dave Chinner705db3f2009-04-06 18:40:17 +020092 /* prevent anyone from using this yet */
Christoph Hellwigeaff8072009-12-17 14:25:01 +010093 VFS_I(ip)->i_state = I_NEW;
Christoph Hellwig24f211b2008-11-28 14:23:42 +110094
95 return ip;
96}
Linus Torvalds1da177e2005-04-16 15:20:36 -070097
Christoph Hellwigb36ec042009-08-07 14:38:34 -030098STATIC void
99xfs_inode_free(
100 struct xfs_inode *ip)
101{
102 switch (ip->i_d.di_mode & S_IFMT) {
103 case S_IFREG:
104 case S_IFDIR:
105 case S_IFLNK:
106 xfs_idestroy_fork(ip, XFS_DATA_FORK);
107 break;
108 }
109
110 if (ip->i_afp)
111 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
112
Christoph Hellwigb36ec042009-08-07 14:38:34 -0300113 if (ip->i_itemp) {
114 /*
115 * Only if we are shutting down the fs will we see an
116 * inode still in the AIL. If it is there, we should remove
117 * it to prevent a use-after-free from occurring.
118 */
119 xfs_log_item_t *lip = &ip->i_itemp->ili_item;
120 struct xfs_ail *ailp = lip->li_ailp;
121
122 ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) ||
123 XFS_FORCED_SHUTDOWN(ip->i_mount));
124 if (lip->li_flags & XFS_LI_IN_AIL) {
125 spin_lock(&ailp->xa_lock);
126 if (lip->li_flags & XFS_LI_IN_AIL)
127 xfs_trans_ail_delete(ailp, lip);
128 else
129 spin_unlock(&ailp->xa_lock);
130 }
131 xfs_inode_item_destroy(ip);
132 ip->i_itemp = NULL;
133 }
134
135 /* asserts to verify all state is correct here */
136 ASSERT(atomic_read(&ip->i_iocount) == 0);
137 ASSERT(atomic_read(&ip->i_pincount) == 0);
138 ASSERT(!spin_is_locked(&ip->i_flags_lock));
139 ASSERT(completion_done(&ip->i_flush));
140
141 kmem_zone_free(xfs_inode_zone, ip);
142}
143
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144/*
David Chinner6441e542008-10-30 17:21:19 +1100145 * Check the validity of the inode we just found it the cache
146 */
147static int
148xfs_iget_cache_hit(
David Chinner6441e542008-10-30 17:21:19 +1100149 struct xfs_perag *pag,
150 struct xfs_inode *ip,
151 int flags,
152 int lock_flags) __releases(pag->pag_ici_lock)
153{
Christoph Hellwigbc990f52009-08-16 20:36:34 -0400154 struct inode *inode = VFS_I(ip);
David Chinner6441e542008-10-30 17:21:19 +1100155 struct xfs_mount *mp = ip->i_mount;
Christoph Hellwigbc990f52009-08-16 20:36:34 -0400156 int error;
157
158 spin_lock(&ip->i_flags_lock);
David Chinner6441e542008-10-30 17:21:19 +1100159
160 /*
Christoph Hellwigbc990f52009-08-16 20:36:34 -0400161 * If we are racing with another cache hit that is currently
162 * instantiating this inode or currently recycling it out of
163 * reclaimabe state, wait for the initialisation to complete
164 * before continuing.
165 *
166 * XXX(hch): eventually we should do something equivalent to
167 * wait_on_inode to wait for these flags to be cleared
168 * instead of polling for it.
David Chinner6441e542008-10-30 17:21:19 +1100169 */
Christoph Hellwigbc990f52009-08-16 20:36:34 -0400170 if (ip->i_flags & (XFS_INEW|XFS_IRECLAIM)) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000171 trace_xfs_iget_skip(ip);
David Chinner6441e542008-10-30 17:21:19 +1100172 XFS_STATS_INC(xs_ig_frecycle);
Christoph Hellwigbc990f52009-08-16 20:36:34 -0400173 error = EAGAIN;
David Chinner6441e542008-10-30 17:21:19 +1100174 goto out_error;
175 }
176
Christoph Hellwigbc990f52009-08-16 20:36:34 -0400177 /*
178 * If lookup is racing with unlink return an error immediately.
179 */
180 if (ip->i_d.di_mode == 0 && !(flags & XFS_IGET_CREATE)) {
181 error = ENOENT;
182 goto out_error;
183 }
David Chinner6441e542008-10-30 17:21:19 +1100184
Christoph Hellwigbc990f52009-08-16 20:36:34 -0400185 /*
186 * If IRECLAIMABLE is set, we've torn down the VFS inode already.
187 * Need to carefully get it back into useable state.
188 */
189 if (ip->i_flags & XFS_IRECLAIMABLE) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000190 trace_xfs_iget_reclaim(ip);
David Chinner6441e542008-10-30 17:21:19 +1100191
David Chinnerbf904242008-10-30 17:36:14 +1100192 /*
Christoph Hellwigbc990f52009-08-16 20:36:34 -0400193 * We need to set XFS_INEW atomically with clearing the
194 * reclaimable tag so that we do have an indicator of the
195 * inode still being initialized.
David Chinnerbf904242008-10-30 17:36:14 +1100196 */
Christoph Hellwigbc990f52009-08-16 20:36:34 -0400197 ip->i_flags |= XFS_INEW;
198 ip->i_flags &= ~XFS_IRECLAIMABLE;
199 __xfs_inode_clear_reclaim_tag(mp, pag, ip);
200
201 spin_unlock(&ip->i_flags_lock);
202 read_unlock(&pag->pag_ici_lock);
203
204 error = -inode_init_always(mp->m_super, inode);
205 if (error) {
206 /*
207 * Re-initializing the inode failed, and we are in deep
208 * trouble. Try to re-add it to the reclaim list.
209 */
210 read_lock(&pag->pag_ici_lock);
211 spin_lock(&ip->i_flags_lock);
212
213 ip->i_flags &= ~XFS_INEW;
214 ip->i_flags |= XFS_IRECLAIMABLE;
215 __xfs_inode_set_reclaim_tag(pag, ip);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000216 trace_xfs_iget_reclaim(ip);
Christoph Hellwigbc990f52009-08-16 20:36:34 -0400217 goto out_error;
218 }
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100219 inode->i_state = I_NEW;
Christoph Hellwigbc990f52009-08-16 20:36:34 -0400220 } else {
221 /* If the VFS inode is being torn down, pause and try again. */
222 if (!igrab(inode)) {
223 error = EAGAIN;
David Chinnerbf904242008-10-30 17:36:14 +1100224 goto out_error;
225 }
David Chinner6bfb3d02008-10-30 18:32:43 +1100226
Christoph Hellwigbc990f52009-08-16 20:36:34 -0400227 /* We've got a live one. */
228 spin_unlock(&ip->i_flags_lock);
229 read_unlock(&pag->pag_ici_lock);
David Chinner6441e542008-10-30 17:21:19 +1100230 }
231
David Chinner6441e542008-10-30 17:21:19 +1100232 if (lock_flags != 0)
233 xfs_ilock(ip, lock_flags);
234
235 xfs_iflags_clear(ip, XFS_ISTALE);
David Chinner6441e542008-10-30 17:21:19 +1100236 XFS_STATS_INC(xs_ig_found);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000237
238 trace_xfs_iget_found(ip);
David Chinner6441e542008-10-30 17:21:19 +1100239 return 0;
240
241out_error:
Christoph Hellwigbc990f52009-08-16 20:36:34 -0400242 spin_unlock(&ip->i_flags_lock);
David Chinner6441e542008-10-30 17:21:19 +1100243 read_unlock(&pag->pag_ici_lock);
David Chinner6441e542008-10-30 17:21:19 +1100244 return error;
245}
246
247
248static int
249xfs_iget_cache_miss(
250 struct xfs_mount *mp,
251 struct xfs_perag *pag,
252 xfs_trans_t *tp,
253 xfs_ino_t ino,
254 struct xfs_inode **ipp,
255 xfs_daddr_t bno,
256 int flags,
Christoph Hellwig0c3dc2b2009-11-14 16:17:23 +0000257 int lock_flags)
David Chinner6441e542008-10-30 17:21:19 +1100258{
259 struct xfs_inode *ip;
260 int error;
261 unsigned long first_index, mask;
262 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino);
263
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100264 ip = xfs_inode_alloc(mp, ino);
265 if (!ip)
266 return ENOMEM;
267
268 error = xfs_iread(mp, tp, ip, bno, flags);
David Chinner6441e542008-10-30 17:21:19 +1100269 if (error)
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100270 goto out_destroy;
David Chinner6441e542008-10-30 17:21:19 +1100271
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000272 xfs_itrace_entry(ip);
David Chinner6441e542008-10-30 17:21:19 +1100273
274 if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) {
275 error = ENOENT;
276 goto out_destroy;
277 }
278
279 /*
280 * Preload the radix tree so we can insert safely under the
David Chinner56e73ec2008-10-30 17:55:27 +1100281 * write spinlock. Note that we cannot sleep inside the preload
282 * region.
David Chinner6441e542008-10-30 17:21:19 +1100283 */
284 if (radix_tree_preload(GFP_KERNEL)) {
285 error = EAGAIN;
Christoph Hellwiged93ec32009-03-03 14:48:35 -0500286 goto out_destroy;
287 }
288
289 /*
290 * Because the inode hasn't been added to the radix-tree yet it can't
291 * be found by another thread, so we can do the non-sleeping lock here.
292 */
293 if (lock_flags) {
294 if (!xfs_ilock_nowait(ip, lock_flags))
295 BUG();
David Chinner6441e542008-10-30 17:21:19 +1100296 }
297
David Chinner6441e542008-10-30 17:21:19 +1100298 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
299 first_index = agino & mask;
300 write_lock(&pag->pag_ici_lock);
301
302 /* insert the new inode */
303 error = radix_tree_insert(&pag->pag_ici_root, agino, ip);
304 if (unlikely(error)) {
305 WARN_ON(error != -EEXIST);
306 XFS_STATS_INC(xs_ig_dup);
307 error = EAGAIN;
David Chinner56e73ec2008-10-30 17:55:27 +1100308 goto out_preload_end;
David Chinner6441e542008-10-30 17:21:19 +1100309 }
310
311 /* These values _must_ be set before releasing the radix tree lock! */
312 ip->i_udquot = ip->i_gdquot = NULL;
313 xfs_iflags_set(ip, XFS_INEW);
314
315 write_unlock(&pag->pag_ici_lock);
316 radix_tree_preload_end();
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000317
318 trace_xfs_iget_alloc(ip);
David Chinner6441e542008-10-30 17:21:19 +1100319 *ipp = ip;
320 return 0;
321
David Chinner56e73ec2008-10-30 17:55:27 +1100322out_preload_end:
David Chinner6441e542008-10-30 17:21:19 +1100323 write_unlock(&pag->pag_ici_lock);
324 radix_tree_preload_end();
David Chinner56e73ec2008-10-30 17:55:27 +1100325 if (lock_flags)
326 xfs_iunlock(ip, lock_flags);
David Chinner6441e542008-10-30 17:21:19 +1100327out_destroy:
Christoph Hellwigb36ec042009-08-07 14:38:34 -0300328 __destroy_inode(VFS_I(ip));
329 xfs_inode_free(ip);
David Chinner6441e542008-10-30 17:21:19 +1100330 return error;
331}
332
333/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 * Look up an inode by number in the given file system.
David Chinnerda353b02007-08-28 14:00:13 +1000335 * The inode is looked up in the cache held in each AG.
David Chinnerbf904242008-10-30 17:36:14 +1100336 * If the inode is found in the cache, initialise the vfs inode
337 * if necessary.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 *
David Chinnerda353b02007-08-28 14:00:13 +1000339 * If it is not in core, read it in from the file system's device,
David Chinnerbf904242008-10-30 17:36:14 +1100340 * add it to the cache and initialise the vfs inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 *
342 * The inode is locked according to the value of the lock_flags parameter.
343 * This flag parameter indicates how and if the inode's IO lock and inode lock
344 * should be taken.
345 *
346 * mp -- the mount point structure for the current file system. It points
347 * to the inode hash table.
348 * tp -- a pointer to the current transaction if there is one. This is
349 * simply passed through to the xfs_iread() call.
350 * ino -- the number of the inode desired. This is the unique identifier
351 * within the file system for the inode being requested.
352 * lock_flags -- flags indicating how to lock the inode. See the comment
353 * for xfs_ilock() for a list of valid values.
354 * bno -- the block number starting the buffer containing the inode,
355 * if known (as by bulkstat), else 0.
356 */
David Chinnerbf904242008-10-30 17:36:14 +1100357int
358xfs_iget(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 xfs_mount_t *mp,
360 xfs_trans_t *tp,
361 xfs_ino_t ino,
362 uint flags,
363 uint lock_flags,
364 xfs_inode_t **ipp,
365 xfs_daddr_t bno)
366{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 xfs_inode_t *ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 int error;
David Chinnerda353b02007-08-28 14:00:13 +1000369 xfs_perag_t *pag;
370 xfs_agino_t agino;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371
David Chinnerda353b02007-08-28 14:00:13 +1000372 /* the radix tree exists only in inode capable AGs */
373 if (XFS_INO_TO_AGNO(mp, ino) >= mp->m_maxagi)
374 return EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375
David Chinnerda353b02007-08-28 14:00:13 +1000376 /* get the perag structure and ensure that it's inode capable */
377 pag = xfs_get_perag(mp, ino);
378 if (!pag->pagi_inodeok)
379 return EINVAL;
380 ASSERT(pag->pag_ici_init);
381 agino = XFS_INO_TO_AGINO(mp, ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382
383again:
David Chinner6441e542008-10-30 17:21:19 +1100384 error = 0;
David Chinnerda353b02007-08-28 14:00:13 +1000385 read_lock(&pag->pag_ici_lock);
386 ip = radix_tree_lookup(&pag->pag_ici_root, agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387
David Chinner6441e542008-10-30 17:21:19 +1100388 if (ip) {
David Chinnerbf904242008-10-30 17:36:14 +1100389 error = xfs_iget_cache_hit(pag, ip, flags, lock_flags);
David Chinner6441e542008-10-30 17:21:19 +1100390 if (error)
391 goto out_error_or_again;
392 } else {
David Chinnerda353b02007-08-28 14:00:13 +1000393 read_unlock(&pag->pag_ici_lock);
David Chinner6441e542008-10-30 17:21:19 +1100394 XFS_STATS_INC(xs_ig_missed);
David Chinnerda353b02007-08-28 14:00:13 +1000395
David Chinner6441e542008-10-30 17:21:19 +1100396 error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip, bno,
397 flags, lock_flags);
398 if (error)
399 goto out_error_or_again;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 }
David Chinnerda353b02007-08-28 14:00:13 +1000401 xfs_put_perag(mp, pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 *ipp = ip;
404
David Chinnerbf904242008-10-30 17:36:14 +1100405 ASSERT(ip->i_df.if_ext_max ==
406 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t));
Christoph Hellwig41be8be2008-08-13 16:23:13 +1000407 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 * If we have a real type for an on-disk inode, we can set ops(&unlock)
409 * now. If it's a new inode being created, xfs_ialloc will handle it.
410 */
David Chinnerbf904242008-10-30 17:36:14 +1100411 if (xfs_iflags_test(ip, XFS_INEW) && ip->i_d.di_mode != 0)
Christoph Hellwig41be8be2008-08-13 16:23:13 +1000412 xfs_setup_inode(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 return 0;
David Chinner6441e542008-10-30 17:21:19 +1100414
415out_error_or_again:
416 if (error == EAGAIN) {
417 delay(1);
418 goto again;
419 }
420 xfs_put_perag(mp, pag);
421 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422}
423
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424/*
425 * Decrement reference count of an inode structure and unlock it.
426 *
427 * ip -- the inode being released
428 * lock_flags -- this parameter indicates the inode's locks to be
429 * to be released. See the comment on xfs_iunlock() for a list
430 * of valid values.
431 */
432void
433xfs_iput(xfs_inode_t *ip,
434 uint lock_flags)
435{
Lachlan McIlroycf441ee2008-02-07 16:42:19 +1100436 xfs_itrace_entry(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 xfs_iunlock(ip, lock_flags);
Christoph Hellwig10090be2007-10-11 18:11:03 +1000438 IRELE(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439}
440
441/*
442 * Special iput for brand-new inodes that are still locked
443 */
444void
David Chinner01651642008-08-13 15:45:15 +1000445xfs_iput_new(
446 xfs_inode_t *ip,
447 uint lock_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448{
David Chinner01651642008-08-13 15:45:15 +1000449 struct inode *inode = VFS_I(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450
Lachlan McIlroycf441ee2008-02-07 16:42:19 +1100451 xfs_itrace_entry(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452
453 if ((ip->i_d.di_mode == 0)) {
David Chinner7a18c382006-11-11 18:04:54 +1100454 ASSERT(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
Christoph Hellwig10090be2007-10-11 18:11:03 +1000455 make_bad_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 }
457 if (inode->i_state & I_NEW)
458 unlock_new_inode(inode);
459 if (lock_flags)
460 xfs_iunlock(ip, lock_flags);
Christoph Hellwig10090be2007-10-11 18:11:03 +1000461 IRELE(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462}
463
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464/*
Christoph Hellwig5cafdeb2008-12-03 12:20:25 +0100465 * This is called free all the memory associated with an inode.
466 * It must free the inode itself and any buffers allocated for
467 * if_extents/if_data and if_broot. It must also free the lock
468 * associated with the inode.
469 *
470 * Note: because we don't initialise everything on reallocation out
471 * of the zone, we must ensure we nullify everything correctly before
472 * freeing the structure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 */
474void
Christoph Hellwig5cafdeb2008-12-03 12:20:25 +0100475xfs_ireclaim(
476 struct xfs_inode *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477{
Christoph Hellwig5cafdeb2008-12-03 12:20:25 +0100478 struct xfs_mount *mp = ip->i_mount;
479 struct xfs_perag *pag;
Christoph Hellwigb44b1122009-12-01 18:12:29 +0000480 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
Christoph Hellwig5cafdeb2008-12-03 12:20:25 +0100481
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 XFS_STATS_INC(xs_ig_reclaims);
483
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 /*
Christoph Hellwigb44b1122009-12-01 18:12:29 +0000485 * Remove the inode from the per-AG radix tree.
486 *
487 * Because radix_tree_delete won't complain even if the item was never
488 * added to the tree assert that it's been there before to catch
489 * problems with the inode life time early on.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 */
Christoph Hellwig5cafdeb2008-12-03 12:20:25 +0100491 pag = xfs_get_perag(mp, ip->i_ino);
David Chinnerda353b02007-08-28 14:00:13 +1000492 write_lock(&pag->pag_ici_lock);
Christoph Hellwigb44b1122009-12-01 18:12:29 +0000493 if (!radix_tree_delete(&pag->pag_ici_root, agino))
494 ASSERT(0);
David Chinnerda353b02007-08-28 14:00:13 +1000495 write_unlock(&pag->pag_ici_lock);
496 xfs_put_perag(mp, pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497
Christoph Hellwig5cafdeb2008-12-03 12:20:25 +0100498 /*
499 * Here we do an (almost) spurious inode lock in order to coordinate
500 * with inode cache radix tree lookups. This is because the lookup
501 * can reference the inodes in the cache without taking references.
502 *
503 * We make that OK here by ensuring that we wait until the inode is
504 * unlocked after the lookup before we go ahead and free it. We get
505 * both the ilock and the iolock because the code may need to drop the
506 * ilock one but will still hold the iolock.
507 */
508 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
Christoph Hellwig7d095252009-06-08 15:33:32 +0200509 xfs_qm_dqdetach(ip);
Christoph Hellwig5cafdeb2008-12-03 12:20:25 +0100510 xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
511
Christoph Hellwigb36ec042009-08-07 14:38:34 -0300512 xfs_inode_free(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513}
514
515/*
516 * This is a wrapper routine around the xfs_ilock() routine
517 * used to centralize some grungy code. It is used in places
518 * that wish to lock the inode solely for reading the extents.
519 * The reason these places can't just call xfs_ilock(SHARED)
520 * is that the inode lock also guards to bringing in of the
521 * extents from disk for a file in b-tree format. If the inode
522 * is in b-tree format, then we need to lock the inode exclusively
523 * until the extents are read in. Locking it exclusively all
524 * the time would limit our parallelism unnecessarily, though.
525 * What we do instead is check to see if the extents have been
526 * read in yet, and only lock the inode exclusively if they
527 * have not.
528 *
529 * The function returns a value which should be given to the
530 * corresponding xfs_iunlock_map_shared(). This value is
531 * the mode in which the lock was actually taken.
532 */
533uint
534xfs_ilock_map_shared(
535 xfs_inode_t *ip)
536{
537 uint lock_mode;
538
539 if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) &&
540 ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) {
541 lock_mode = XFS_ILOCK_EXCL;
542 } else {
543 lock_mode = XFS_ILOCK_SHARED;
544 }
545
546 xfs_ilock(ip, lock_mode);
547
548 return lock_mode;
549}
550
551/*
552 * This is simply the unlock routine to go with xfs_ilock_map_shared().
553 * All it does is call xfs_iunlock() with the given lock_mode.
554 */
555void
556xfs_iunlock_map_shared(
557 xfs_inode_t *ip,
558 unsigned int lock_mode)
559{
560 xfs_iunlock(ip, lock_mode);
561}
562
563/*
564 * The xfs inode contains 2 locks: a multi-reader lock called the
565 * i_iolock and a multi-reader lock called the i_lock. This routine
566 * allows either or both of the locks to be obtained.
567 *
568 * The 2 locks should always be ordered so that the IO lock is
569 * obtained first in order to prevent deadlock.
570 *
571 * ip -- the inode being locked
572 * lock_flags -- this parameter indicates the inode's locks
573 * to be locked. It can be:
574 * XFS_IOLOCK_SHARED,
575 * XFS_IOLOCK_EXCL,
576 * XFS_ILOCK_SHARED,
577 * XFS_ILOCK_EXCL,
578 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
579 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
580 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
581 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
582 */
583void
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000584xfs_ilock(
585 xfs_inode_t *ip,
586 uint lock_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587{
588 /*
589 * You can't set both SHARED and EXCL for the same lock,
590 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
591 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
592 */
593 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
594 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
595 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
596 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000597 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000599 if (lock_flags & XFS_IOLOCK_EXCL)
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000600 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000601 else if (lock_flags & XFS_IOLOCK_SHARED)
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000602 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000603
604 if (lock_flags & XFS_ILOCK_EXCL)
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000605 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000606 else if (lock_flags & XFS_ILOCK_SHARED)
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000607 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000608
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000609 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610}
611
612/*
613 * This is just like xfs_ilock(), except that the caller
614 * is guaranteed not to sleep. It returns 1 if it gets
615 * the requested locks and 0 otherwise. If the IO lock is
616 * obtained but the inode lock cannot be, then the IO lock
617 * is dropped before returning.
618 *
619 * ip -- the inode being locked
620 * lock_flags -- this parameter indicates the inode's locks to be
621 * to be locked. See the comment for xfs_ilock() for a list
622 * of valid values.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 */
624int
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000625xfs_ilock_nowait(
626 xfs_inode_t *ip,
627 uint lock_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 /*
630 * You can't set both SHARED and EXCL for the same lock,
631 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
632 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
633 */
634 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
635 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
636 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
637 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000638 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 if (lock_flags & XFS_IOLOCK_EXCL) {
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000641 if (!mrtryupdate(&ip->i_iolock))
642 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 } else if (lock_flags & XFS_IOLOCK_SHARED) {
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000644 if (!mrtryaccess(&ip->i_iolock))
645 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 }
647 if (lock_flags & XFS_ILOCK_EXCL) {
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000648 if (!mrtryupdate(&ip->i_lock))
649 goto out_undo_iolock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 } else if (lock_flags & XFS_ILOCK_SHARED) {
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000651 if (!mrtryaccess(&ip->i_lock))
652 goto out_undo_iolock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 }
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000654 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 return 1;
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000656
657 out_undo_iolock:
658 if (lock_flags & XFS_IOLOCK_EXCL)
659 mrunlock_excl(&ip->i_iolock);
660 else if (lock_flags & XFS_IOLOCK_SHARED)
661 mrunlock_shared(&ip->i_iolock);
662 out:
663 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664}
665
666/*
667 * xfs_iunlock() is used to drop the inode locks acquired with
668 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
669 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
670 * that we know which locks to drop.
671 *
672 * ip -- the inode being unlocked
673 * lock_flags -- this parameter indicates the inode's locks to be
674 * to be unlocked. See the comment for xfs_ilock() for a list
675 * of valid values for this parameter.
676 *
677 */
678void
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000679xfs_iunlock(
680 xfs_inode_t *ip,
681 uint lock_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682{
683 /*
684 * You can't set both SHARED and EXCL for the same lock,
685 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
686 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
687 */
688 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
689 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
690 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
691 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000692 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_IUNLOCK_NONOTIFY |
693 XFS_LOCK_DEP_MASK)) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 ASSERT(lock_flags != 0);
695
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000696 if (lock_flags & XFS_IOLOCK_EXCL)
697 mrunlock_excl(&ip->i_iolock);
698 else if (lock_flags & XFS_IOLOCK_SHARED)
699 mrunlock_shared(&ip->i_iolock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000701 if (lock_flags & XFS_ILOCK_EXCL)
702 mrunlock_excl(&ip->i_lock);
703 else if (lock_flags & XFS_ILOCK_SHARED)
704 mrunlock_shared(&ip->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000706 if ((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) &&
707 !(lock_flags & XFS_IUNLOCK_NONOTIFY) && ip->i_itemp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 /*
709 * Let the AIL know that this item has been unlocked in case
710 * it is in the AIL and anyone is waiting on it. Don't do
711 * this if the caller has asked us not to.
712 */
David Chinner783a2f62008-10-30 17:39:58 +1100713 xfs_trans_unlocked_item(ip->i_itemp->ili_item.li_ailp,
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000714 (xfs_log_item_t*)(ip->i_itemp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 }
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000716 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717}
718
719/*
720 * give up write locks. the i/o lock cannot be held nested
721 * if it is being demoted.
722 */
723void
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000724xfs_ilock_demote(
725 xfs_inode_t *ip,
726 uint lock_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727{
728 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
729 ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
730
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000731 if (lock_flags & XFS_ILOCK_EXCL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 mrdemote(&ip->i_lock);
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000733 if (lock_flags & XFS_IOLOCK_EXCL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 mrdemote(&ip->i_iolock);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000735
736 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737}
738
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000739#ifdef DEBUG
740/*
741 * Debug-only routine, without additional rw_semaphore APIs, we can
742 * now only answer requests regarding whether we hold the lock for write
743 * (reader state is outside our visibility, we only track writer state).
744 *
745 * Note: this means !xfs_isilocked would give false positives, so don't do that.
746 */
747int
748xfs_isilocked(
749 xfs_inode_t *ip,
750 uint lock_flags)
751{
752 if ((lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) ==
753 XFS_ILOCK_EXCL) {
754 if (!ip->i_lock.mr_writer)
755 return 0;
756 }
757
758 if ((lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) ==
759 XFS_IOLOCK_EXCL) {
760 if (!ip->i_iolock.mr_writer)
761 return 0;
762 }
763
764 return 1;
765}
766#endif