blob: 478e587087feddcd975e6ab62fa5e1655136b59e [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"
David Chinner783a2f62008-10-30 17:39:58 +110041#include "xfs_trans_priv.h"
42#include "xfs_inode_item.h"
Christoph Hellwig24f211b2008-11-28 14:23:42 +110043#include "xfs_bmap.h"
44#include "xfs_btree_trace.h"
45#include "xfs_dir2_trace.h"
46
47
48/*
49 * Allocate and initialise an xfs_inode.
50 */
51STATIC struct xfs_inode *
52xfs_inode_alloc(
53 struct xfs_mount *mp,
54 xfs_ino_t ino)
55{
56 struct xfs_inode *ip;
57
58 /*
59 * if this didn't occur in transactions, we could use
60 * KM_MAYFAIL and return NULL here on ENOMEM. Set the
61 * code up to do this anyway.
62 */
63 ip = kmem_zone_alloc(xfs_inode_zone, KM_SLEEP);
64 if (!ip)
65 return NULL;
66
67 ASSERT(atomic_read(&ip->i_iocount) == 0);
68 ASSERT(atomic_read(&ip->i_pincount) == 0);
69 ASSERT(!spin_is_locked(&ip->i_flags_lock));
70 ASSERT(completion_done(&ip->i_flush));
71
72 /*
73 * initialise the VFS inode here to get failures
74 * out of the way early.
75 */
76 if (!inode_init_always(mp->m_super, VFS_I(ip))) {
77 kmem_zone_free(xfs_inode_zone, ip);
78 return NULL;
79 }
80
81 /* initialise the xfs inode */
82 ip->i_ino = ino;
83 ip->i_mount = mp;
84 memset(&ip->i_imap, 0, sizeof(struct xfs_imap));
85 ip->i_afp = NULL;
86 memset(&ip->i_df, 0, sizeof(xfs_ifork_t));
87 ip->i_flags = 0;
88 ip->i_update_core = 0;
89 ip->i_update_size = 0;
90 ip->i_delayed_blks = 0;
91 memset(&ip->i_d, 0, sizeof(xfs_icdinode_t));
92 ip->i_size = 0;
93 ip->i_new_size = 0;
94
95 /*
96 * Initialize inode's trace buffers.
97 */
98#ifdef XFS_INODE_TRACE
99 ip->i_trace = ktrace_alloc(INODE_TRACE_SIZE, KM_NOFS);
100#endif
101#ifdef XFS_BMAP_TRACE
102 ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_NOFS);
103#endif
104#ifdef XFS_BTREE_TRACE
105 ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_NOFS);
106#endif
107#ifdef XFS_RW_TRACE
108 ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_NOFS);
109#endif
110#ifdef XFS_ILOCK_TRACE
111 ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_NOFS);
112#endif
113#ifdef XFS_DIR2_TRACE
114 ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_NOFS);
115#endif
116
117 return ip;
118}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119
120/*
David Chinner6441e542008-10-30 17:21:19 +1100121 * Check the validity of the inode we just found it the cache
122 */
123static int
124xfs_iget_cache_hit(
David Chinner6441e542008-10-30 17:21:19 +1100125 struct xfs_perag *pag,
126 struct xfs_inode *ip,
127 int flags,
128 int lock_flags) __releases(pag->pag_ici_lock)
129{
130 struct xfs_mount *mp = ip->i_mount;
David Chinner6bfb3d02008-10-30 18:32:43 +1100131 int error = EAGAIN;
David Chinner6441e542008-10-30 17:21:19 +1100132
133 /*
134 * If INEW is set this inode is being set up
David Chinnerbf904242008-10-30 17:36:14 +1100135 * If IRECLAIM is set this inode is being torn down
David Chinner6441e542008-10-30 17:21:19 +1100136 * Pause and try again.
137 */
David Chinnerbf904242008-10-30 17:36:14 +1100138 if (xfs_iflags_test(ip, (XFS_INEW|XFS_IRECLAIM))) {
David Chinner6441e542008-10-30 17:21:19 +1100139 XFS_STATS_INC(xs_ig_frecycle);
140 goto out_error;
141 }
142
David Chinnerbf904242008-10-30 17:36:14 +1100143 /* If IRECLAIMABLE is set, we've torn down the vfs inode part */
144 if (xfs_iflags_test(ip, XFS_IRECLAIMABLE)) {
David Chinner6441e542008-10-30 17:21:19 +1100145
146 /*
David Chinnerbf904242008-10-30 17:36:14 +1100147 * If lookup is racing with unlink, then we should return an
148 * error immediately so we don't remove it from the reclaim
149 * list and potentially leak the inode.
David Chinner6441e542008-10-30 17:21:19 +1100150 */
David Chinnerbf904242008-10-30 17:36:14 +1100151 if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) {
David Chinner6441e542008-10-30 17:21:19 +1100152 error = ENOENT;
153 goto out_error;
154 }
David Chinnerbf904242008-10-30 17:36:14 +1100155
David Chinner6441e542008-10-30 17:21:19 +1100156 xfs_itrace_exit_tag(ip, "xfs_iget.alloc");
157
David Chinnerbf904242008-10-30 17:36:14 +1100158 /*
159 * We need to re-initialise the VFS inode as it has been
160 * 'freed' by the VFS. Do this here so we can deal with
161 * errors cleanly, then tag it so it can be set up correctly
162 * later.
163 */
164 if (!inode_init_always(mp->m_super, VFS_I(ip))) {
165 error = ENOMEM;
166 goto out_error;
167 }
David Chinner6bfb3d02008-10-30 18:32:43 +1100168
169 /*
170 * We must set the XFS_INEW flag before clearing the
171 * XFS_IRECLAIMABLE flag so that if a racing lookup does
172 * not find the XFS_IRECLAIMABLE above but has the igrab()
173 * below succeed we can safely check XFS_INEW to detect
174 * that this inode is still being initialised.
175 */
David Chinnerbf904242008-10-30 17:36:14 +1100176 xfs_iflags_set(ip, XFS_INEW);
David Chinner6441e542008-10-30 17:21:19 +1100177 xfs_iflags_clear(ip, XFS_IRECLAIMABLE);
David Chinner396beb82008-10-30 17:37:26 +1100178
179 /* clear the radix tree reclaim flag as well. */
180 __xfs_inode_clear_reclaim_tag(mp, pag, ip);
David Chinnerbf904242008-10-30 17:36:14 +1100181 } else if (!igrab(VFS_I(ip))) {
182 /* If the VFS inode is being torn down, pause and try again. */
David Chinnerbf904242008-10-30 17:36:14 +1100183 XFS_STATS_INC(xs_ig_frecycle);
184 goto out_error;
David Chinner6bfb3d02008-10-30 18:32:43 +1100185 } else if (xfs_iflags_test(ip, XFS_INEW)) {
186 /*
187 * We are racing with another cache hit that is
188 * currently recycling this inode out of the XFS_IRECLAIMABLE
189 * state. Wait for the initialisation to complete before
190 * continuing.
191 */
192 wait_on_inode(VFS_I(ip));
David Chinner6441e542008-10-30 17:21:19 +1100193 }
194
195 if (ip->i_d.di_mode == 0 && !(flags & XFS_IGET_CREATE)) {
196 error = ENOENT;
David Chinner6bfb3d02008-10-30 18:32:43 +1100197 iput(VFS_I(ip));
198 goto out_error;
David Chinner6441e542008-10-30 17:21:19 +1100199 }
200
David Chinner6bfb3d02008-10-30 18:32:43 +1100201 /* We've got a live one. */
202 read_unlock(&pag->pag_ici_lock);
203
David Chinner6441e542008-10-30 17:21:19 +1100204 if (lock_flags != 0)
205 xfs_ilock(ip, lock_flags);
206
207 xfs_iflags_clear(ip, XFS_ISTALE);
208 xfs_itrace_exit_tag(ip, "xfs_iget.found");
209 XFS_STATS_INC(xs_ig_found);
210 return 0;
211
212out_error:
213 read_unlock(&pag->pag_ici_lock);
David Chinner6441e542008-10-30 17:21:19 +1100214 return error;
215}
216
217
218static int
219xfs_iget_cache_miss(
220 struct xfs_mount *mp,
221 struct xfs_perag *pag,
222 xfs_trans_t *tp,
223 xfs_ino_t ino,
224 struct xfs_inode **ipp,
225 xfs_daddr_t bno,
226 int flags,
227 int lock_flags) __releases(pag->pag_ici_lock)
228{
229 struct xfs_inode *ip;
230 int error;
231 unsigned long first_index, mask;
232 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ino);
233
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100234 ip = xfs_inode_alloc(mp, ino);
235 if (!ip)
236 return ENOMEM;
237
238 error = xfs_iread(mp, tp, ip, bno, flags);
David Chinner6441e542008-10-30 17:21:19 +1100239 if (error)
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100240 goto out_destroy;
David Chinner6441e542008-10-30 17:21:19 +1100241
242 xfs_itrace_exit_tag(ip, "xfs_iget.alloc");
243
244 if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) {
245 error = ENOENT;
246 goto out_destroy;
247 }
248
249 /*
250 * Preload the radix tree so we can insert safely under the
David Chinner56e73ec2008-10-30 17:55:27 +1100251 * write spinlock. Note that we cannot sleep inside the preload
252 * region.
David Chinner6441e542008-10-30 17:21:19 +1100253 */
254 if (radix_tree_preload(GFP_KERNEL)) {
255 error = EAGAIN;
Christoph Hellwig7d46be42009-03-03 14:48:35 -0500256 goto out_destroy;
257 }
258
259 /*
260 * Because the inode hasn't been added to the radix-tree yet it can't
261 * be found by another thread, so we can do the non-sleeping lock here.
262 */
263 if (lock_flags) {
264 if (!xfs_ilock_nowait(ip, lock_flags))
265 BUG();
David Chinner6441e542008-10-30 17:21:19 +1100266 }
267
David Chinner6441e542008-10-30 17:21:19 +1100268 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
269 first_index = agino & mask;
270 write_lock(&pag->pag_ici_lock);
271
272 /* insert the new inode */
273 error = radix_tree_insert(&pag->pag_ici_root, agino, ip);
274 if (unlikely(error)) {
275 WARN_ON(error != -EEXIST);
276 XFS_STATS_INC(xs_ig_dup);
277 error = EAGAIN;
David Chinner56e73ec2008-10-30 17:55:27 +1100278 goto out_preload_end;
David Chinner6441e542008-10-30 17:21:19 +1100279 }
280
281 /* These values _must_ be set before releasing the radix tree lock! */
282 ip->i_udquot = ip->i_gdquot = NULL;
283 xfs_iflags_set(ip, XFS_INEW);
284
285 write_unlock(&pag->pag_ici_lock);
286 radix_tree_preload_end();
287 *ipp = ip;
288 return 0;
289
David Chinner56e73ec2008-10-30 17:55:27 +1100290out_preload_end:
David Chinner6441e542008-10-30 17:21:19 +1100291 write_unlock(&pag->pag_ici_lock);
292 radix_tree_preload_end();
David Chinner56e73ec2008-10-30 17:55:27 +1100293 if (lock_flags)
294 xfs_iunlock(ip, lock_flags);
David Chinner6441e542008-10-30 17:21:19 +1100295out_destroy:
Christoph Hellwig9ed04512008-10-30 18:26:04 +1100296 xfs_destroy_inode(ip);
David Chinner6441e542008-10-30 17:21:19 +1100297 return error;
298}
299
300/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 * Look up an inode by number in the given file system.
David Chinnerda353b02007-08-28 14:00:13 +1000302 * The inode is looked up in the cache held in each AG.
David Chinnerbf904242008-10-30 17:36:14 +1100303 * If the inode is found in the cache, initialise the vfs inode
304 * if necessary.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 *
David Chinnerda353b02007-08-28 14:00:13 +1000306 * If it is not in core, read it in from the file system's device,
David Chinnerbf904242008-10-30 17:36:14 +1100307 * add it to the cache and initialise the vfs inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 *
309 * The inode is locked according to the value of the lock_flags parameter.
310 * This flag parameter indicates how and if the inode's IO lock and inode lock
311 * should be taken.
312 *
313 * mp -- the mount point structure for the current file system. It points
314 * to the inode hash table.
315 * tp -- a pointer to the current transaction if there is one. This is
316 * simply passed through to the xfs_iread() call.
317 * ino -- the number of the inode desired. This is the unique identifier
318 * within the file system for the inode being requested.
319 * lock_flags -- flags indicating how to lock the inode. See the comment
320 * for xfs_ilock() for a list of valid values.
321 * bno -- the block number starting the buffer containing the inode,
322 * if known (as by bulkstat), else 0.
323 */
David Chinnerbf904242008-10-30 17:36:14 +1100324int
325xfs_iget(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 xfs_mount_t *mp,
327 xfs_trans_t *tp,
328 xfs_ino_t ino,
329 uint flags,
330 uint lock_flags,
331 xfs_inode_t **ipp,
332 xfs_daddr_t bno)
333{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 xfs_inode_t *ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 int error;
David Chinnerda353b02007-08-28 14:00:13 +1000336 xfs_perag_t *pag;
337 xfs_agino_t agino;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338
David Chinnerda353b02007-08-28 14:00:13 +1000339 /* the radix tree exists only in inode capable AGs */
340 if (XFS_INO_TO_AGNO(mp, ino) >= mp->m_maxagi)
341 return EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342
David Chinnerda353b02007-08-28 14:00:13 +1000343 /* get the perag structure and ensure that it's inode capable */
344 pag = xfs_get_perag(mp, ino);
345 if (!pag->pagi_inodeok)
346 return EINVAL;
347 ASSERT(pag->pag_ici_init);
348 agino = XFS_INO_TO_AGINO(mp, ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349
350again:
David Chinner6441e542008-10-30 17:21:19 +1100351 error = 0;
David Chinnerda353b02007-08-28 14:00:13 +1000352 read_lock(&pag->pag_ici_lock);
353 ip = radix_tree_lookup(&pag->pag_ici_root, agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354
David Chinner6441e542008-10-30 17:21:19 +1100355 if (ip) {
David Chinnerbf904242008-10-30 17:36:14 +1100356 error = xfs_iget_cache_hit(pag, ip, flags, lock_flags);
David Chinner6441e542008-10-30 17:21:19 +1100357 if (error)
358 goto out_error_or_again;
359 } else {
David Chinnerda353b02007-08-28 14:00:13 +1000360 read_unlock(&pag->pag_ici_lock);
David Chinner6441e542008-10-30 17:21:19 +1100361 XFS_STATS_INC(xs_ig_missed);
David Chinnerda353b02007-08-28 14:00:13 +1000362
David Chinner6441e542008-10-30 17:21:19 +1100363 error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip, bno,
364 flags, lock_flags);
365 if (error)
366 goto out_error_or_again;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 }
David Chinnerda353b02007-08-28 14:00:13 +1000368 xfs_put_perag(mp, pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 *ipp = ip;
371
David Chinnerbf904242008-10-30 17:36:14 +1100372 ASSERT(ip->i_df.if_ext_max ==
373 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t));
Christoph Hellwig41be8be2008-08-13 16:23:13 +1000374 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 * If we have a real type for an on-disk inode, we can set ops(&unlock)
376 * now. If it's a new inode being created, xfs_ialloc will handle it.
377 */
David Chinnerbf904242008-10-30 17:36:14 +1100378 if (xfs_iflags_test(ip, XFS_INEW) && ip->i_d.di_mode != 0)
Christoph Hellwig41be8be2008-08-13 16:23:13 +1000379 xfs_setup_inode(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 return 0;
David Chinner6441e542008-10-30 17:21:19 +1100381
382out_error_or_again:
383 if (error == EAGAIN) {
384 delay(1);
385 goto again;
386 }
387 xfs_put_perag(mp, pag);
388 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389}
390
391
392/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 * Look for the inode corresponding to the given ino in the hash table.
394 * If it is there and its i_transp pointer matches tp, return it.
395 * Otherwise, return NULL.
396 */
397xfs_inode_t *
398xfs_inode_incore(xfs_mount_t *mp,
399 xfs_ino_t ino,
400 xfs_trans_t *tp)
401{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 xfs_inode_t *ip;
David Chinnerda353b02007-08-28 14:00:13 +1000403 xfs_perag_t *pag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404
David Chinnerda353b02007-08-28 14:00:13 +1000405 pag = xfs_get_perag(mp, ino);
406 read_lock(&pag->pag_ici_lock);
407 ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ino));
408 read_unlock(&pag->pag_ici_lock);
409 xfs_put_perag(mp, pag);
410
411 /* the returned inode must match the transaction */
412 if (ip && (ip->i_transp != tp))
413 return NULL;
414 return ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415}
416
417/*
418 * Decrement reference count of an inode structure and unlock it.
419 *
420 * ip -- the inode being released
421 * lock_flags -- this parameter indicates the inode's locks to be
422 * to be released. See the comment on xfs_iunlock() for a list
423 * of valid values.
424 */
425void
426xfs_iput(xfs_inode_t *ip,
427 uint lock_flags)
428{
Lachlan McIlroycf441ee2008-02-07 16:42:19 +1100429 xfs_itrace_entry(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 xfs_iunlock(ip, lock_flags);
Christoph Hellwig10090be2007-10-11 18:11:03 +1000431 IRELE(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432}
433
434/*
435 * Special iput for brand-new inodes that are still locked
436 */
437void
David Chinner01651642008-08-13 15:45:15 +1000438xfs_iput_new(
439 xfs_inode_t *ip,
440 uint lock_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441{
David Chinner01651642008-08-13 15:45:15 +1000442 struct inode *inode = VFS_I(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443
Lachlan McIlroycf441ee2008-02-07 16:42:19 +1100444 xfs_itrace_entry(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445
446 if ((ip->i_d.di_mode == 0)) {
David Chinner7a18c382006-11-11 18:04:54 +1100447 ASSERT(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
Christoph Hellwig10090be2007-10-11 18:11:03 +1000448 make_bad_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 }
450 if (inode->i_state & I_NEW)
451 unlock_new_inode(inode);
452 if (lock_flags)
453 xfs_iunlock(ip, lock_flags);
Christoph Hellwig10090be2007-10-11 18:11:03 +1000454 IRELE(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455}
456
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457/*
Christoph Hellwig5cafdeb2008-12-03 12:20:25 +0100458 * This is called free all the memory associated with an inode.
459 * It must free the inode itself and any buffers allocated for
460 * if_extents/if_data and if_broot. It must also free the lock
461 * associated with the inode.
462 *
463 * Note: because we don't initialise everything on reallocation out
464 * of the zone, we must ensure we nullify everything correctly before
465 * freeing the structure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 */
467void
Christoph Hellwig5cafdeb2008-12-03 12:20:25 +0100468xfs_ireclaim(
469 struct xfs_inode *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470{
Christoph Hellwig5cafdeb2008-12-03 12:20:25 +0100471 struct xfs_mount *mp = ip->i_mount;
472 struct xfs_perag *pag;
473
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 XFS_STATS_INC(xs_ig_reclaims);
475
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 /*
Christoph Hellwig5cafdeb2008-12-03 12:20:25 +0100477 * Remove the inode from the per-AG radix tree. It doesn't matter
478 * if it was never added to it because radix_tree_delete can deal
479 * with that case just fine.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 */
Christoph Hellwig5cafdeb2008-12-03 12:20:25 +0100481 pag = xfs_get_perag(mp, ip->i_ino);
David Chinnerda353b02007-08-28 14:00:13 +1000482 write_lock(&pag->pag_ici_lock);
483 radix_tree_delete(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ip->i_ino));
484 write_unlock(&pag->pag_ici_lock);
485 xfs_put_perag(mp, pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486
Christoph Hellwig5cafdeb2008-12-03 12:20:25 +0100487 /*
488 * Here we do an (almost) spurious inode lock in order to coordinate
489 * with inode cache radix tree lookups. This is because the lookup
490 * can reference the inodes in the cache without taking references.
491 *
492 * We make that OK here by ensuring that we wait until the inode is
493 * unlocked after the lookup before we go ahead and free it. We get
494 * both the ilock and the iolock because the code may need to drop the
495 * ilock one but will still hold the iolock.
496 */
497 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
498 /*
499 * Release dquots (and their references) if any.
500 */
501 XFS_QM_DQDETACH(ip->i_mount, ip);
502 xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
503
504 switch (ip->i_d.di_mode & S_IFMT) {
505 case S_IFREG:
506 case S_IFDIR:
507 case S_IFLNK:
508 xfs_idestroy_fork(ip, XFS_DATA_FORK);
509 break;
510 }
511
512 if (ip->i_afp)
513 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
514
515#ifdef XFS_INODE_TRACE
516 ktrace_free(ip->i_trace);
517#endif
518#ifdef XFS_BMAP_TRACE
519 ktrace_free(ip->i_xtrace);
520#endif
521#ifdef XFS_BTREE_TRACE
522 ktrace_free(ip->i_btrace);
523#endif
524#ifdef XFS_RW_TRACE
525 ktrace_free(ip->i_rwtrace);
526#endif
527#ifdef XFS_ILOCK_TRACE
528 ktrace_free(ip->i_lock_trace);
529#endif
530#ifdef XFS_DIR2_TRACE
531 ktrace_free(ip->i_dir_trace);
532#endif
533 if (ip->i_itemp) {
534 /*
535 * Only if we are shutting down the fs will we see an
536 * inode still in the AIL. If it is there, we should remove
537 * it to prevent a use-after-free from occurring.
538 */
539 xfs_log_item_t *lip = &ip->i_itemp->ili_item;
540 struct xfs_ail *ailp = lip->li_ailp;
541
542 ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) ||
543 XFS_FORCED_SHUTDOWN(ip->i_mount));
544 if (lip->li_flags & XFS_LI_IN_AIL) {
545 spin_lock(&ailp->xa_lock);
546 if (lip->li_flags & XFS_LI_IN_AIL)
547 xfs_trans_ail_delete(ailp, lip);
548 else
549 spin_unlock(&ailp->xa_lock);
550 }
551 xfs_inode_item_destroy(ip);
552 ip->i_itemp = NULL;
553 }
554 /* asserts to verify all state is correct here */
555 ASSERT(atomic_read(&ip->i_iocount) == 0);
556 ASSERT(atomic_read(&ip->i_pincount) == 0);
557 ASSERT(!spin_is_locked(&ip->i_flags_lock));
558 ASSERT(completion_done(&ip->i_flush));
559 kmem_zone_free(xfs_inode_zone, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560}
561
562/*
563 * This is a wrapper routine around the xfs_ilock() routine
564 * used to centralize some grungy code. It is used in places
565 * that wish to lock the inode solely for reading the extents.
566 * The reason these places can't just call xfs_ilock(SHARED)
567 * is that the inode lock also guards to bringing in of the
568 * extents from disk for a file in b-tree format. If the inode
569 * is in b-tree format, then we need to lock the inode exclusively
570 * until the extents are read in. Locking it exclusively all
571 * the time would limit our parallelism unnecessarily, though.
572 * What we do instead is check to see if the extents have been
573 * read in yet, and only lock the inode exclusively if they
574 * have not.
575 *
576 * The function returns a value which should be given to the
577 * corresponding xfs_iunlock_map_shared(). This value is
578 * the mode in which the lock was actually taken.
579 */
580uint
581xfs_ilock_map_shared(
582 xfs_inode_t *ip)
583{
584 uint lock_mode;
585
586 if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) &&
587 ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) {
588 lock_mode = XFS_ILOCK_EXCL;
589 } else {
590 lock_mode = XFS_ILOCK_SHARED;
591 }
592
593 xfs_ilock(ip, lock_mode);
594
595 return lock_mode;
596}
597
598/*
599 * This is simply the unlock routine to go with xfs_ilock_map_shared().
600 * All it does is call xfs_iunlock() with the given lock_mode.
601 */
602void
603xfs_iunlock_map_shared(
604 xfs_inode_t *ip,
605 unsigned int lock_mode)
606{
607 xfs_iunlock(ip, lock_mode);
608}
609
610/*
611 * The xfs inode contains 2 locks: a multi-reader lock called the
612 * i_iolock and a multi-reader lock called the i_lock. This routine
613 * allows either or both of the locks to be obtained.
614 *
615 * The 2 locks should always be ordered so that the IO lock is
616 * obtained first in order to prevent deadlock.
617 *
618 * ip -- the inode being locked
619 * lock_flags -- this parameter indicates the inode's locks
620 * to be locked. It can be:
621 * XFS_IOLOCK_SHARED,
622 * XFS_IOLOCK_EXCL,
623 * XFS_ILOCK_SHARED,
624 * XFS_ILOCK_EXCL,
625 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
626 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
627 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
628 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
629 */
630void
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000631xfs_ilock(
632 xfs_inode_t *ip,
633 uint lock_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634{
635 /*
636 * You can't set both SHARED and EXCL for the same lock,
637 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
638 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
639 */
640 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
641 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
642 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
643 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000644 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000646 if (lock_flags & XFS_IOLOCK_EXCL)
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000647 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000648 else if (lock_flags & XFS_IOLOCK_SHARED)
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000649 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000650
651 if (lock_flags & XFS_ILOCK_EXCL)
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000652 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000653 else if (lock_flags & XFS_ILOCK_SHARED)
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000654 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000655
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 xfs_ilock_trace(ip, 1, lock_flags, (inst_t *)__return_address);
657}
658
659/*
660 * This is just like xfs_ilock(), except that the caller
661 * is guaranteed not to sleep. It returns 1 if it gets
662 * the requested locks and 0 otherwise. If the IO lock is
663 * obtained but the inode lock cannot be, then the IO lock
664 * is dropped before returning.
665 *
666 * ip -- the inode being locked
667 * lock_flags -- this parameter indicates the inode's locks to be
668 * to be locked. See the comment for xfs_ilock() for a list
669 * of valid values.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 */
671int
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000672xfs_ilock_nowait(
673 xfs_inode_t *ip,
674 uint lock_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 /*
677 * You can't set both SHARED and EXCL for the same lock,
678 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
679 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
680 */
681 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
682 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
683 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
684 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000685 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 if (lock_flags & XFS_IOLOCK_EXCL) {
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000688 if (!mrtryupdate(&ip->i_iolock))
689 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 } else if (lock_flags & XFS_IOLOCK_SHARED) {
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000691 if (!mrtryaccess(&ip->i_iolock))
692 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 }
694 if (lock_flags & XFS_ILOCK_EXCL) {
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000695 if (!mrtryupdate(&ip->i_lock))
696 goto out_undo_iolock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 } else if (lock_flags & XFS_ILOCK_SHARED) {
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000698 if (!mrtryaccess(&ip->i_lock))
699 goto out_undo_iolock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 }
701 xfs_ilock_trace(ip, 2, lock_flags, (inst_t *)__return_address);
702 return 1;
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000703
704 out_undo_iolock:
705 if (lock_flags & XFS_IOLOCK_EXCL)
706 mrunlock_excl(&ip->i_iolock);
707 else if (lock_flags & XFS_IOLOCK_SHARED)
708 mrunlock_shared(&ip->i_iolock);
709 out:
710 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711}
712
713/*
714 * xfs_iunlock() is used to drop the inode locks acquired with
715 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
716 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
717 * that we know which locks to drop.
718 *
719 * ip -- the inode being unlocked
720 * lock_flags -- this parameter indicates the inode's locks to be
721 * to be unlocked. See the comment for xfs_ilock() for a list
722 * of valid values for this parameter.
723 *
724 */
725void
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000726xfs_iunlock(
727 xfs_inode_t *ip,
728 uint lock_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729{
730 /*
731 * You can't set both SHARED and EXCL for the same lock,
732 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
733 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
734 */
735 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
736 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
737 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
738 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000739 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_IUNLOCK_NONOTIFY |
740 XFS_LOCK_DEP_MASK)) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 ASSERT(lock_flags != 0);
742
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000743 if (lock_flags & XFS_IOLOCK_EXCL)
744 mrunlock_excl(&ip->i_iolock);
745 else if (lock_flags & XFS_IOLOCK_SHARED)
746 mrunlock_shared(&ip->i_iolock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000748 if (lock_flags & XFS_ILOCK_EXCL)
749 mrunlock_excl(&ip->i_lock);
750 else if (lock_flags & XFS_ILOCK_SHARED)
751 mrunlock_shared(&ip->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000753 if ((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) &&
754 !(lock_flags & XFS_IUNLOCK_NONOTIFY) && ip->i_itemp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 /*
756 * Let the AIL know that this item has been unlocked in case
757 * it is in the AIL and anyone is waiting on it. Don't do
758 * this if the caller has asked us not to.
759 */
David Chinner783a2f62008-10-30 17:39:58 +1100760 xfs_trans_unlocked_item(ip->i_itemp->ili_item.li_ailp,
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000761 (xfs_log_item_t*)(ip->i_itemp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 }
763 xfs_ilock_trace(ip, 3, lock_flags, (inst_t *)__return_address);
764}
765
766/*
767 * give up write locks. the i/o lock cannot be held nested
768 * if it is being demoted.
769 */
770void
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000771xfs_ilock_demote(
772 xfs_inode_t *ip,
773 uint lock_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774{
775 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
776 ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
777
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000778 if (lock_flags & XFS_ILOCK_EXCL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 mrdemote(&ip->i_lock);
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000780 if (lock_flags & XFS_IOLOCK_EXCL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 mrdemote(&ip->i_iolock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782}
783
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000784#ifdef DEBUG
785/*
786 * Debug-only routine, without additional rw_semaphore APIs, we can
787 * now only answer requests regarding whether we hold the lock for write
788 * (reader state is outside our visibility, we only track writer state).
789 *
790 * Note: this means !xfs_isilocked would give false positives, so don't do that.
791 */
792int
793xfs_isilocked(
794 xfs_inode_t *ip,
795 uint lock_flags)
796{
797 if ((lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) ==
798 XFS_ILOCK_EXCL) {
799 if (!ip->i_lock.mr_writer)
800 return 0;
801 }
802
803 if ((lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) ==
804 XFS_IOLOCK_EXCL) {
805 if (!ip->i_iolock.mr_writer)
806 return 0;
807 }
808
809 return 1;
810}
811#endif
812
Christoph Hellwig5a8d0f32008-12-03 12:20:40 +0100813#ifdef XFS_INODE_TRACE
814
815#define KTRACE_ENTER(ip, vk, s, line, ra) \
816 ktrace_enter((ip)->i_trace, \
817/* 0 */ (void *)(__psint_t)(vk), \
818/* 1 */ (void *)(s), \
819/* 2 */ (void *)(__psint_t) line, \
820/* 3 */ (void *)(__psint_t)atomic_read(&VFS_I(ip)->i_count), \
821/* 4 */ (void *)(ra), \
822/* 5 */ NULL, \
823/* 6 */ (void *)(__psint_t)current_cpu(), \
824/* 7 */ (void *)(__psint_t)current_pid(), \
825/* 8 */ (void *)__return_address, \
826/* 9 */ NULL, NULL, NULL, NULL, NULL, NULL, NULL)
827
828/*
829 * Vnode tracing code.
830 */
831void
832_xfs_itrace_entry(xfs_inode_t *ip, const char *func, inst_t *ra)
833{
834 KTRACE_ENTER(ip, INODE_KTRACE_ENTRY, func, 0, ra);
835}
836
837void
838_xfs_itrace_exit(xfs_inode_t *ip, const char *func, inst_t *ra)
839{
840 KTRACE_ENTER(ip, INODE_KTRACE_EXIT, func, 0, ra);
841}
842
843void
844xfs_itrace_hold(xfs_inode_t *ip, char *file, int line, inst_t *ra)
845{
846 KTRACE_ENTER(ip, INODE_KTRACE_HOLD, file, line, ra);
847}
848
849void
850_xfs_itrace_ref(xfs_inode_t *ip, char *file, int line, inst_t *ra)
851{
852 KTRACE_ENTER(ip, INODE_KTRACE_REF, file, line, ra);
853}
854
855void
856xfs_itrace_rele(xfs_inode_t *ip, char *file, int line, inst_t *ra)
857{
858 KTRACE_ENTER(ip, INODE_KTRACE_RELE, file, line, ra);
859}
860#endif /* XFS_INODE_TRACE */