blob: cdb1c2505fc672db69eb1ef9b1ee3519995bbdf9 [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_mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110030#include "xfs_alloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include "xfs_ialloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include "xfs_dinode.h"
33#include "xfs_inode.h"
Nathan Scotta844f452005-11-02 14:38:42 +110034#include "xfs_btree.h"
35#include "xfs_ialloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include "xfs_quota.h"
37#include "xfs_utils.h"
David Chinner783a2f62008-10-30 17:39:58 +110038#include "xfs_trans_priv.h"
39#include "xfs_inode_item.h"
Christoph Hellwig24f211b2008-11-28 14:23:42 +110040#include "xfs_bmap.h"
41#include "xfs_btree_trace.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000042#include "xfs_trace.h"
Christoph Hellwig24f211b2008-11-28 14:23:42 +110043
44
45/*
Dave Chinnerdcfcf202010-12-23 11:57:13 +110046 * Define xfs inode iolock lockdep classes. We need to ensure that all active
47 * inodes are considered the same for lockdep purposes, including inodes that
48 * are recycled through the XFS_IRECLAIMABLE state. This is the the only way to
49 * guarantee the locks are considered the same when there are multiple lock
50 * initialisation siteѕ. Also, define a reclaimable inode class so it is
51 * obvious in lockdep reports which class the report is against.
52 */
53static struct lock_class_key xfs_iolock_active;
54struct lock_class_key xfs_iolock_reclaimable;
55
56/*
Christoph Hellwig24f211b2008-11-28 14:23:42 +110057 * Allocate and initialise an xfs_inode.
58 */
59STATIC struct xfs_inode *
60xfs_inode_alloc(
61 struct xfs_mount *mp,
62 xfs_ino_t ino)
63{
64 struct xfs_inode *ip;
65
66 /*
67 * if this didn't occur in transactions, we could use
68 * KM_MAYFAIL and return NULL here on ENOMEM. Set the
69 * code up to do this anyway.
70 */
71 ip = kmem_zone_alloc(xfs_inode_zone, KM_SLEEP);
72 if (!ip)
73 return NULL;
Christoph Hellwig54e34622009-08-07 14:38:25 -030074 if (inode_init_always(mp->m_super, VFS_I(ip))) {
75 kmem_zone_free(xfs_inode_zone, ip);
76 return NULL;
77 }
Christoph Hellwig24f211b2008-11-28 14:23:42 +110078
79 ASSERT(atomic_read(&ip->i_iocount) == 0);
80 ASSERT(atomic_read(&ip->i_pincount) == 0);
81 ASSERT(!spin_is_locked(&ip->i_flags_lock));
82 ASSERT(completion_done(&ip->i_flush));
Christoph Hellwig033da482009-10-19 04:05:26 +000083
84 mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino);
Dave Chinnerdcfcf202010-12-23 11:57:13 +110085 lockdep_set_class_and_name(&ip->i_iolock.mr_lock,
86 &xfs_iolock_active, "xfs_iolock_active");
Christoph Hellwig24f211b2008-11-28 14:23:42 +110087
Christoph Hellwig24f211b2008-11-28 14:23:42 +110088 /* initialise the xfs inode */
89 ip->i_ino = ino;
90 ip->i_mount = mp;
91 memset(&ip->i_imap, 0, sizeof(struct xfs_imap));
92 ip->i_afp = NULL;
93 memset(&ip->i_df, 0, sizeof(xfs_ifork_t));
94 ip->i_flags = 0;
95 ip->i_update_core = 0;
Christoph Hellwig24f211b2008-11-28 14:23:42 +110096 ip->i_delayed_blks = 0;
97 memset(&ip->i_d, 0, sizeof(xfs_icdinode_t));
98 ip->i_size = 0;
99 ip->i_new_size = 0;
100
Dave Chinner705db3f2009-04-06 18:40:17 +0200101 /* prevent anyone from using this yet */
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100102 VFS_I(ip)->i_state = I_NEW;
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100103
104 return ip;
105}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106
Dave Chinner2f11fea2010-07-20 17:53:25 +1000107void
Christoph Hellwigb36ec042009-08-07 14:38:34 -0300108xfs_inode_free(
109 struct xfs_inode *ip)
110{
111 switch (ip->i_d.di_mode & S_IFMT) {
112 case S_IFREG:
113 case S_IFDIR:
114 case S_IFLNK:
115 xfs_idestroy_fork(ip, XFS_DATA_FORK);
116 break;
117 }
118
119 if (ip->i_afp)
120 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
121
Christoph Hellwigb36ec042009-08-07 14:38:34 -0300122 if (ip->i_itemp) {
123 /*
124 * Only if we are shutting down the fs will we see an
125 * inode still in the AIL. If it is there, we should remove
126 * it to prevent a use-after-free from occurring.
127 */
128 xfs_log_item_t *lip = &ip->i_itemp->ili_item;
129 struct xfs_ail *ailp = lip->li_ailp;
130
131 ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) ||
132 XFS_FORCED_SHUTDOWN(ip->i_mount));
133 if (lip->li_flags & XFS_LI_IN_AIL) {
134 spin_lock(&ailp->xa_lock);
135 if (lip->li_flags & XFS_LI_IN_AIL)
136 xfs_trans_ail_delete(ailp, lip);
137 else
138 spin_unlock(&ailp->xa_lock);
139 }
140 xfs_inode_item_destroy(ip);
141 ip->i_itemp = NULL;
142 }
143
144 /* asserts to verify all state is correct here */
145 ASSERT(atomic_read(&ip->i_iocount) == 0);
146 ASSERT(atomic_read(&ip->i_pincount) == 0);
147 ASSERT(!spin_is_locked(&ip->i_flags_lock));
148 ASSERT(completion_done(&ip->i_flush));
149
150 kmem_zone_free(xfs_inode_zone, ip);
151}
152
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153/*
David Chinner6441e542008-10-30 17:21:19 +1100154 * Check the validity of the inode we just found it the cache
155 */
156static int
157xfs_iget_cache_hit(
David Chinner6441e542008-10-30 17:21:19 +1100158 struct xfs_perag *pag,
159 struct xfs_inode *ip,
160 int flags,
161 int lock_flags) __releases(pag->pag_ici_lock)
162{
Christoph Hellwigbc990f52009-08-16 20:36:34 -0400163 struct inode *inode = VFS_I(ip);
David Chinner6441e542008-10-30 17:21:19 +1100164 struct xfs_mount *mp = ip->i_mount;
Christoph Hellwigbc990f52009-08-16 20:36:34 -0400165 int error;
166
167 spin_lock(&ip->i_flags_lock);
David Chinner6441e542008-10-30 17:21:19 +1100168
169 /*
Christoph Hellwigbc990f52009-08-16 20:36:34 -0400170 * If we are racing with another cache hit that is currently
171 * instantiating this inode or currently recycling it out of
172 * reclaimabe state, wait for the initialisation to complete
173 * before continuing.
174 *
175 * XXX(hch): eventually we should do something equivalent to
176 * wait_on_inode to wait for these flags to be cleared
177 * instead of polling for it.
David Chinner6441e542008-10-30 17:21:19 +1100178 */
Christoph Hellwigbc990f52009-08-16 20:36:34 -0400179 if (ip->i_flags & (XFS_INEW|XFS_IRECLAIM)) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000180 trace_xfs_iget_skip(ip);
David Chinner6441e542008-10-30 17:21:19 +1100181 XFS_STATS_INC(xs_ig_frecycle);
Christoph Hellwigbc990f52009-08-16 20:36:34 -0400182 error = EAGAIN;
David Chinner6441e542008-10-30 17:21:19 +1100183 goto out_error;
184 }
185
Christoph Hellwigbc990f52009-08-16 20:36:34 -0400186 /*
187 * If lookup is racing with unlink return an error immediately.
188 */
189 if (ip->i_d.di_mode == 0 && !(flags & XFS_IGET_CREATE)) {
190 error = ENOENT;
191 goto out_error;
192 }
David Chinner6441e542008-10-30 17:21:19 +1100193
Christoph Hellwigbc990f52009-08-16 20:36:34 -0400194 /*
195 * If IRECLAIMABLE is set, we've torn down the VFS inode already.
196 * Need to carefully get it back into useable state.
197 */
198 if (ip->i_flags & XFS_IRECLAIMABLE) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000199 trace_xfs_iget_reclaim(ip);
David Chinner6441e542008-10-30 17:21:19 +1100200
David Chinnerbf904242008-10-30 17:36:14 +1100201 /*
Christoph Hellwigf1f724e2010-03-01 11:30:31 +0000202 * We need to set XFS_IRECLAIM to prevent xfs_reclaim_inode
203 * from stomping over us while we recycle the inode. We can't
204 * clear the radix tree reclaimable tag yet as it requires
205 * pag_ici_lock to be held exclusive.
David Chinnerbf904242008-10-30 17:36:14 +1100206 */
Christoph Hellwigf1f724e2010-03-01 11:30:31 +0000207 ip->i_flags |= XFS_IRECLAIM;
Christoph Hellwigbc990f52009-08-16 20:36:34 -0400208
209 spin_unlock(&ip->i_flags_lock);
210 read_unlock(&pag->pag_ici_lock);
211
212 error = -inode_init_always(mp->m_super, inode);
213 if (error) {
214 /*
215 * Re-initializing the inode failed, and we are in deep
216 * trouble. Try to re-add it to the reclaim list.
217 */
218 read_lock(&pag->pag_ici_lock);
219 spin_lock(&ip->i_flags_lock);
220
221 ip->i_flags &= ~XFS_INEW;
222 ip->i_flags |= XFS_IRECLAIMABLE;
223 __xfs_inode_set_reclaim_tag(pag, ip);
Christoph Hellwigd2e078c2010-06-24 11:50:22 +1000224 trace_xfs_iget_reclaim_fail(ip);
Christoph Hellwigbc990f52009-08-16 20:36:34 -0400225 goto out_error;
226 }
Christoph Hellwigf1f724e2010-03-01 11:30:31 +0000227
228 write_lock(&pag->pag_ici_lock);
229 spin_lock(&ip->i_flags_lock);
230 ip->i_flags &= ~(XFS_IRECLAIMABLE | XFS_IRECLAIM);
231 ip->i_flags |= XFS_INEW;
232 __xfs_inode_clear_reclaim_tag(mp, pag, ip);
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100233 inode->i_state = I_NEW;
Dave Chinnerdcfcf202010-12-23 11:57:13 +1100234
235 ASSERT(!rwsem_is_locked(&ip->i_iolock.mr_lock));
236 mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino);
237 lockdep_set_class_and_name(&ip->i_iolock.mr_lock,
238 &xfs_iolock_active, "xfs_iolock_active");
239
Christoph Hellwigf1f724e2010-03-01 11:30:31 +0000240 spin_unlock(&ip->i_flags_lock);
241 write_unlock(&pag->pag_ici_lock);
Christoph Hellwigbc990f52009-08-16 20:36:34 -0400242 } else {
243 /* If the VFS inode is being torn down, pause and try again. */
244 if (!igrab(inode)) {
Christoph Hellwigd2e078c2010-06-24 11:50:22 +1000245 trace_xfs_iget_skip(ip);
Christoph Hellwigbc990f52009-08-16 20:36:34 -0400246 error = EAGAIN;
David Chinnerbf904242008-10-30 17:36:14 +1100247 goto out_error;
248 }
David Chinner6bfb3d02008-10-30 18:32:43 +1100249
Christoph Hellwigbc990f52009-08-16 20:36:34 -0400250 /* We've got a live one. */
251 spin_unlock(&ip->i_flags_lock);
252 read_unlock(&pag->pag_ici_lock);
Christoph Hellwigd2e078c2010-06-24 11:50:22 +1000253 trace_xfs_iget_hit(ip);
David Chinner6441e542008-10-30 17:21:19 +1100254 }
255
David Chinner6441e542008-10-30 17:21:19 +1100256 if (lock_flags != 0)
257 xfs_ilock(ip, lock_flags);
258
259 xfs_iflags_clear(ip, XFS_ISTALE);
David Chinner6441e542008-10-30 17:21:19 +1100260 XFS_STATS_INC(xs_ig_found);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000261
David Chinner6441e542008-10-30 17:21:19 +1100262 return 0;
263
264out_error:
Christoph Hellwigbc990f52009-08-16 20:36:34 -0400265 spin_unlock(&ip->i_flags_lock);
David Chinner6441e542008-10-30 17:21:19 +1100266 read_unlock(&pag->pag_ici_lock);
David Chinner6441e542008-10-30 17:21:19 +1100267 return error;
268}
269
270
271static int
272xfs_iget_cache_miss(
273 struct xfs_mount *mp,
274 struct xfs_perag *pag,
275 xfs_trans_t *tp,
276 xfs_ino_t ino,
277 struct xfs_inode **ipp,
David Chinner6441e542008-10-30 17:21:19 +1100278 int flags,
Christoph Hellwig0c3dc2b2009-11-14 16:17:23 +0000279 int lock_flags)
David Chinner6441e542008-10-30 17:21:19 +1100280{
281 struct xfs_inode *ip;
282 int error;
David Chinner6441e542008-10-30 17:21:19 +1100283 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino);
284
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100285 ip = xfs_inode_alloc(mp, ino);
286 if (!ip)
287 return ENOMEM;
288
Dave Chinner7b6259e2010-06-24 11:35:17 +1000289 error = xfs_iread(mp, tp, ip, flags);
David Chinner6441e542008-10-30 17:21:19 +1100290 if (error)
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100291 goto out_destroy;
David Chinner6441e542008-10-30 17:21:19 +1100292
Christoph Hellwigd2e078c2010-06-24 11:50:22 +1000293 trace_xfs_iget_miss(ip);
David Chinner6441e542008-10-30 17:21:19 +1100294
295 if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) {
296 error = ENOENT;
297 goto out_destroy;
298 }
299
300 /*
301 * Preload the radix tree so we can insert safely under the
David Chinner56e73ec2008-10-30 17:55:27 +1100302 * write spinlock. Note that we cannot sleep inside the preload
303 * region.
David Chinner6441e542008-10-30 17:21:19 +1100304 */
305 if (radix_tree_preload(GFP_KERNEL)) {
306 error = EAGAIN;
Christoph Hellwiged93ec32009-03-03 14:48:35 -0500307 goto out_destroy;
308 }
309
310 /*
311 * Because the inode hasn't been added to the radix-tree yet it can't
312 * be found by another thread, so we can do the non-sleeping lock here.
313 */
314 if (lock_flags) {
315 if (!xfs_ilock_nowait(ip, lock_flags))
316 BUG();
David Chinner6441e542008-10-30 17:21:19 +1100317 }
318
David Chinner6441e542008-10-30 17:21:19 +1100319 write_lock(&pag->pag_ici_lock);
320
321 /* insert the new inode */
322 error = radix_tree_insert(&pag->pag_ici_root, agino, ip);
323 if (unlikely(error)) {
324 WARN_ON(error != -EEXIST);
325 XFS_STATS_INC(xs_ig_dup);
326 error = EAGAIN;
David Chinner56e73ec2008-10-30 17:55:27 +1100327 goto out_preload_end;
David Chinner6441e542008-10-30 17:21:19 +1100328 }
329
330 /* These values _must_ be set before releasing the radix tree lock! */
331 ip->i_udquot = ip->i_gdquot = NULL;
332 xfs_iflags_set(ip, XFS_INEW);
333
334 write_unlock(&pag->pag_ici_lock);
335 radix_tree_preload_end();
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000336
David Chinner6441e542008-10-30 17:21:19 +1100337 *ipp = ip;
338 return 0;
339
David Chinner56e73ec2008-10-30 17:55:27 +1100340out_preload_end:
David Chinner6441e542008-10-30 17:21:19 +1100341 write_unlock(&pag->pag_ici_lock);
342 radix_tree_preload_end();
David Chinner56e73ec2008-10-30 17:55:27 +1100343 if (lock_flags)
344 xfs_iunlock(ip, lock_flags);
David Chinner6441e542008-10-30 17:21:19 +1100345out_destroy:
Christoph Hellwigb36ec042009-08-07 14:38:34 -0300346 __destroy_inode(VFS_I(ip));
347 xfs_inode_free(ip);
David Chinner6441e542008-10-30 17:21:19 +1100348 return error;
349}
350
351/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 * Look up an inode by number in the given file system.
David Chinnerda353b02007-08-28 14:00:13 +1000353 * The inode is looked up in the cache held in each AG.
David Chinnerbf904242008-10-30 17:36:14 +1100354 * If the inode is found in the cache, initialise the vfs inode
355 * if necessary.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 *
David Chinnerda353b02007-08-28 14:00:13 +1000357 * If it is not in core, read it in from the file system's device,
David Chinnerbf904242008-10-30 17:36:14 +1100358 * add it to the cache and initialise the vfs inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 *
360 * The inode is locked according to the value of the lock_flags parameter.
361 * This flag parameter indicates how and if the inode's IO lock and inode lock
362 * should be taken.
363 *
364 * mp -- the mount point structure for the current file system. It points
365 * to the inode hash table.
366 * tp -- a pointer to the current transaction if there is one. This is
367 * simply passed through to the xfs_iread() call.
368 * ino -- the number of the inode desired. This is the unique identifier
369 * within the file system for the inode being requested.
370 * lock_flags -- flags indicating how to lock the inode. See the comment
371 * for xfs_ilock() for a list of valid values.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 */
David Chinnerbf904242008-10-30 17:36:14 +1100373int
374xfs_iget(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 xfs_mount_t *mp,
376 xfs_trans_t *tp,
377 xfs_ino_t ino,
378 uint flags,
379 uint lock_flags,
Dave Chinner7b6259e2010-06-24 11:35:17 +1000380 xfs_inode_t **ipp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 xfs_inode_t *ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 int error;
David Chinnerda353b02007-08-28 14:00:13 +1000384 xfs_perag_t *pag;
385 xfs_agino_t agino;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386
Christoph Hellwigd2767342010-10-06 18:31:23 +0000387 /* reject inode numbers outside existing AGs */
388 if (XFS_INO_TO_AGNO(mp, ino) >= mp->m_sb.sb_agcount)
David Chinnerda353b02007-08-28 14:00:13 +1000389 return EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390
David Chinnerda353b02007-08-28 14:00:13 +1000391 /* get the perag structure and ensure that it's inode capable */
Dave Chinner5017e972010-01-11 11:47:40 +0000392 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ino));
David Chinnerda353b02007-08-28 14:00:13 +1000393 agino = XFS_INO_TO_AGINO(mp, ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394
395again:
David Chinner6441e542008-10-30 17:21:19 +1100396 error = 0;
David Chinnerda353b02007-08-28 14:00:13 +1000397 read_lock(&pag->pag_ici_lock);
398 ip = radix_tree_lookup(&pag->pag_ici_root, agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399
David Chinner6441e542008-10-30 17:21:19 +1100400 if (ip) {
David Chinnerbf904242008-10-30 17:36:14 +1100401 error = xfs_iget_cache_hit(pag, ip, flags, lock_flags);
David Chinner6441e542008-10-30 17:21:19 +1100402 if (error)
403 goto out_error_or_again;
404 } else {
David Chinnerda353b02007-08-28 14:00:13 +1000405 read_unlock(&pag->pag_ici_lock);
David Chinner6441e542008-10-30 17:21:19 +1100406 XFS_STATS_INC(xs_ig_missed);
David Chinnerda353b02007-08-28 14:00:13 +1000407
Dave Chinner7b6259e2010-06-24 11:35:17 +1000408 error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip,
David Chinner6441e542008-10-30 17:21:19 +1100409 flags, lock_flags);
410 if (error)
411 goto out_error_or_again;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 }
Dave Chinner5017e972010-01-11 11:47:40 +0000413 xfs_perag_put(pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 *ipp = ip;
416
David Chinnerbf904242008-10-30 17:36:14 +1100417 ASSERT(ip->i_df.if_ext_max ==
418 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t));
Christoph Hellwig41be8be2008-08-13 16:23:13 +1000419 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 * If we have a real type for an on-disk inode, we can set ops(&unlock)
421 * now. If it's a new inode being created, xfs_ialloc will handle it.
422 */
David Chinnerbf904242008-10-30 17:36:14 +1100423 if (xfs_iflags_test(ip, XFS_INEW) && ip->i_d.di_mode != 0)
Christoph Hellwig41be8be2008-08-13 16:23:13 +1000424 xfs_setup_inode(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 return 0;
David Chinner6441e542008-10-30 17:21:19 +1100426
427out_error_or_again:
428 if (error == EAGAIN) {
429 delay(1);
430 goto again;
431 }
Dave Chinner5017e972010-01-11 11:47:40 +0000432 xfs_perag_put(pag);
David Chinner6441e542008-10-30 17:21:19 +1100433 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434}
435
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 * This is a wrapper routine around the xfs_ilock() routine
438 * used to centralize some grungy code. It is used in places
439 * that wish to lock the inode solely for reading the extents.
440 * The reason these places can't just call xfs_ilock(SHARED)
441 * is that the inode lock also guards to bringing in of the
442 * extents from disk for a file in b-tree format. If the inode
443 * is in b-tree format, then we need to lock the inode exclusively
444 * until the extents are read in. Locking it exclusively all
445 * the time would limit our parallelism unnecessarily, though.
446 * What we do instead is check to see if the extents have been
447 * read in yet, and only lock the inode exclusively if they
448 * have not.
449 *
450 * The function returns a value which should be given to the
451 * corresponding xfs_iunlock_map_shared(). This value is
452 * the mode in which the lock was actually taken.
453 */
454uint
455xfs_ilock_map_shared(
456 xfs_inode_t *ip)
457{
458 uint lock_mode;
459
460 if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) &&
461 ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) {
462 lock_mode = XFS_ILOCK_EXCL;
463 } else {
464 lock_mode = XFS_ILOCK_SHARED;
465 }
466
467 xfs_ilock(ip, lock_mode);
468
469 return lock_mode;
470}
471
472/*
473 * This is simply the unlock routine to go with xfs_ilock_map_shared().
474 * All it does is call xfs_iunlock() with the given lock_mode.
475 */
476void
477xfs_iunlock_map_shared(
478 xfs_inode_t *ip,
479 unsigned int lock_mode)
480{
481 xfs_iunlock(ip, lock_mode);
482}
483
484/*
485 * The xfs inode contains 2 locks: a multi-reader lock called the
486 * i_iolock and a multi-reader lock called the i_lock. This routine
487 * allows either or both of the locks to be obtained.
488 *
489 * The 2 locks should always be ordered so that the IO lock is
490 * obtained first in order to prevent deadlock.
491 *
492 * ip -- the inode being locked
493 * lock_flags -- this parameter indicates the inode's locks
494 * to be locked. It can be:
495 * XFS_IOLOCK_SHARED,
496 * XFS_IOLOCK_EXCL,
497 * XFS_ILOCK_SHARED,
498 * XFS_ILOCK_EXCL,
499 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
500 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
501 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
502 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
503 */
504void
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000505xfs_ilock(
506 xfs_inode_t *ip,
507 uint lock_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508{
509 /*
510 * You can't set both SHARED and EXCL for the same lock,
511 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
512 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
513 */
514 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
515 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
516 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
517 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000518 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000520 if (lock_flags & XFS_IOLOCK_EXCL)
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000521 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000522 else if (lock_flags & XFS_IOLOCK_SHARED)
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000523 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000524
525 if (lock_flags & XFS_ILOCK_EXCL)
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000526 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000527 else if (lock_flags & XFS_ILOCK_SHARED)
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000528 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000529
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000530 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531}
532
533/*
534 * This is just like xfs_ilock(), except that the caller
535 * is guaranteed not to sleep. It returns 1 if it gets
536 * the requested locks and 0 otherwise. If the IO lock is
537 * obtained but the inode lock cannot be, then the IO lock
538 * is dropped before returning.
539 *
540 * ip -- the inode being locked
541 * lock_flags -- this parameter indicates the inode's locks to be
542 * to be locked. See the comment for xfs_ilock() for a list
543 * of valid values.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 */
545int
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000546xfs_ilock_nowait(
547 xfs_inode_t *ip,
548 uint lock_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 /*
551 * You can't set both SHARED and EXCL for the same lock,
552 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
553 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
554 */
555 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
556 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
557 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
558 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000559 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 if (lock_flags & XFS_IOLOCK_EXCL) {
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000562 if (!mrtryupdate(&ip->i_iolock))
563 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 } else if (lock_flags & XFS_IOLOCK_SHARED) {
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000565 if (!mrtryaccess(&ip->i_iolock))
566 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 }
568 if (lock_flags & XFS_ILOCK_EXCL) {
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000569 if (!mrtryupdate(&ip->i_lock))
570 goto out_undo_iolock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 } else if (lock_flags & XFS_ILOCK_SHARED) {
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000572 if (!mrtryaccess(&ip->i_lock))
573 goto out_undo_iolock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 }
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000575 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 return 1;
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000577
578 out_undo_iolock:
579 if (lock_flags & XFS_IOLOCK_EXCL)
580 mrunlock_excl(&ip->i_iolock);
581 else if (lock_flags & XFS_IOLOCK_SHARED)
582 mrunlock_shared(&ip->i_iolock);
583 out:
584 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585}
586
587/*
588 * xfs_iunlock() is used to drop the inode locks acquired with
589 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
590 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
591 * that we know which locks to drop.
592 *
593 * ip -- the inode being unlocked
594 * lock_flags -- this parameter indicates the inode's locks to be
595 * to be unlocked. See the comment for xfs_ilock() for a list
596 * of valid values for this parameter.
597 *
598 */
599void
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000600xfs_iunlock(
601 xfs_inode_t *ip,
602 uint lock_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603{
604 /*
605 * You can't set both SHARED and EXCL for the same lock,
606 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
607 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
608 */
609 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
610 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
611 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
612 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000613 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_IUNLOCK_NONOTIFY |
614 XFS_LOCK_DEP_MASK)) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 ASSERT(lock_flags != 0);
616
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000617 if (lock_flags & XFS_IOLOCK_EXCL)
618 mrunlock_excl(&ip->i_iolock);
619 else if (lock_flags & XFS_IOLOCK_SHARED)
620 mrunlock_shared(&ip->i_iolock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000622 if (lock_flags & XFS_ILOCK_EXCL)
623 mrunlock_excl(&ip->i_lock);
624 else if (lock_flags & XFS_ILOCK_SHARED)
625 mrunlock_shared(&ip->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000627 if ((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) &&
628 !(lock_flags & XFS_IUNLOCK_NONOTIFY) && ip->i_itemp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 /*
630 * Let the AIL know that this item has been unlocked in case
631 * it is in the AIL and anyone is waiting on it. Don't do
632 * this if the caller has asked us not to.
633 */
David Chinner783a2f62008-10-30 17:39:58 +1100634 xfs_trans_unlocked_item(ip->i_itemp->ili_item.li_ailp,
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000635 (xfs_log_item_t*)(ip->i_itemp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 }
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000637 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638}
639
640/*
641 * give up write locks. the i/o lock cannot be held nested
642 * if it is being demoted.
643 */
644void
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000645xfs_ilock_demote(
646 xfs_inode_t *ip,
647 uint lock_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648{
649 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
650 ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
651
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000652 if (lock_flags & XFS_ILOCK_EXCL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 mrdemote(&ip->i_lock);
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000654 if (lock_flags & XFS_IOLOCK_EXCL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 mrdemote(&ip->i_iolock);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000656
657 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658}
659
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000660#ifdef DEBUG
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000661int
662xfs_isilocked(
663 xfs_inode_t *ip,
664 uint lock_flags)
665{
Christoph Hellwigf9369722010-06-03 16:22:29 +1000666 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
667 if (!(lock_flags & XFS_ILOCK_SHARED))
668 return !!ip->i_lock.mr_writer;
669 return rwsem_is_locked(&ip->i_lock.mr_lock);
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000670 }
671
Christoph Hellwigf9369722010-06-03 16:22:29 +1000672 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
673 if (!(lock_flags & XFS_IOLOCK_SHARED))
674 return !!ip->i_iolock.mr_writer;
675 return rwsem_is_locked(&ip->i_iolock.mr_lock);
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000676 }
677
Christoph Hellwigf9369722010-06-03 16:22:29 +1000678 ASSERT(0);
679 return 0;
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000680}
681#endif