blob: 03f32c68731f3abaa6bee1d58743dd1edf6e0321 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Nathan Scott7b718762005-11-02 14:58:39 +11002 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110019#include "xfs_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include "xfs_types.h"
Nathan Scotta844f452005-11-02 14:38:42 +110021#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110023#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_trans.h"
25#include "xfs_sb.h"
26#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_dir2.h"
28#include "xfs_dmapi.h"
29#include "xfs_mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110031#include "xfs_alloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include "xfs_ialloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "xfs_dir2_sf.h"
Nathan Scotta844f452005-11-02 14:38:42 +110034#include "xfs_attr_sf.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include "xfs_dinode.h"
36#include "xfs_inode.h"
Nathan Scotta844f452005-11-02 14:38:42 +110037#include "xfs_btree.h"
38#include "xfs_ialloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include "xfs_quota.h"
40#include "xfs_utils.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
42/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070043 * Look up an inode by number in the given file system.
David Chinnerda353b02007-08-28 14:00:13 +100044 * The inode is looked up in the cache held in each AG.
45 * If the inode is found in the cache, attach it to the provided
46 * vnode.
Linus Torvalds1da177e2005-04-16 15:20:36 -070047 *
David Chinnerda353b02007-08-28 14:00:13 +100048 * If it is not in core, read it in from the file system's device,
49 * add it to the cache and attach the provided vnode.
Linus Torvalds1da177e2005-04-16 15:20:36 -070050 *
51 * The inode is locked according to the value of the lock_flags parameter.
52 * This flag parameter indicates how and if the inode's IO lock and inode lock
53 * should be taken.
54 *
55 * mp -- the mount point structure for the current file system. It points
56 * to the inode hash table.
57 * tp -- a pointer to the current transaction if there is one. This is
58 * simply passed through to the xfs_iread() call.
59 * ino -- the number of the inode desired. This is the unique identifier
60 * within the file system for the inode being requested.
61 * lock_flags -- flags indicating how to lock the inode. See the comment
62 * for xfs_ilock() for a list of valid values.
63 * bno -- the block number starting the buffer containing the inode,
64 * if known (as by bulkstat), else 0.
65 */
66STATIC int
67xfs_iget_core(
Nathan Scott67fcaa72006-06-09 17:00:52 +100068 bhv_vnode_t *vp,
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 xfs_mount_t *mp,
70 xfs_trans_t *tp,
71 xfs_ino_t ino,
72 uint flags,
73 uint lock_flags,
74 xfs_inode_t **ipp,
75 xfs_daddr_t bno)
76{
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 xfs_inode_t *ip;
78 xfs_inode_t *iq;
Nathan Scott67fcaa72006-06-09 17:00:52 +100079 bhv_vnode_t *inode_vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 int error;
David Chinnerda353b02007-08-28 14:00:13 +100081 xfs_icluster_t *icl, *new_icl = NULL;
82 unsigned long first_index, mask;
83 xfs_perag_t *pag;
84 xfs_agino_t agino;
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
David Chinnerda353b02007-08-28 14:00:13 +100086 /* the radix tree exists only in inode capable AGs */
87 if (XFS_INO_TO_AGNO(mp, ino) >= mp->m_maxagi)
88 return EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070089
David Chinnerda353b02007-08-28 14:00:13 +100090 /* get the perag structure and ensure that it's inode capable */
91 pag = xfs_get_perag(mp, ino);
92 if (!pag->pagi_inodeok)
93 return EINVAL;
94 ASSERT(pag->pag_ici_init);
95 agino = XFS_INO_TO_AGINO(mp, ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -070096
97again:
David Chinnerda353b02007-08-28 14:00:13 +100098 read_lock(&pag->pag_ici_lock);
99 ip = radix_tree_lookup(&pag->pag_ici_root, agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100
David Chinnerda353b02007-08-28 14:00:13 +1000101 if (ip != NULL) {
102 /*
103 * If INEW is set this inode is being set up
104 * we need to pause and try again.
105 */
106 if (xfs_iflags_test(ip, XFS_INEW)) {
107 read_unlock(&pag->pag_ici_lock);
108 delay(1);
109 XFS_STATS_INC(xs_ig_frecycle);
110
111 goto again;
112 }
113
114 inode_vp = XFS_ITOV_NULL(ip);
115 if (inode_vp == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 /*
David Chinnerda353b02007-08-28 14:00:13 +1000117 * If IRECLAIM is set this inode is
118 * on its way out of the system,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 * we need to pause and try again.
120 */
David Chinnerda353b02007-08-28 14:00:13 +1000121 if (xfs_iflags_test(ip, XFS_IRECLAIM)) {
122 read_unlock(&pag->pag_ici_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 delay(1);
124 XFS_STATS_INC(xs_ig_frecycle);
125
126 goto again;
127 }
David Chinnerda353b02007-08-28 14:00:13 +1000128 ASSERT(xfs_iflags_test(ip, XFS_IRECLAIMABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129
David Chinnerda353b02007-08-28 14:00:13 +1000130 /*
131 * If lookup is racing with unlink, then we
132 * should return an error immediately so we
133 * don't remove it from the reclaim list and
134 * potentially leak the inode.
135 */
136 if ((ip->i_d.di_mode == 0) &&
137 !(flags & XFS_IGET_CREATE)) {
138 read_unlock(&pag->pag_ici_lock);
139 xfs_put_perag(mp, pag);
140 return ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 }
142
Nathan Scott71bce252005-05-05 13:23:27 -0700143 /*
David Chinnerda353b02007-08-28 14:00:13 +1000144 * There may be transactions sitting in the
145 * incore log buffers or being flushed to disk
146 * at this time. We can't clear the
147 * XFS_IRECLAIMABLE flag until these
148 * transactions have hit the disk, otherwise we
149 * will void the guarantee the flag provides
150 * xfs_iunpin()
Nathan Scott71bce252005-05-05 13:23:27 -0700151 */
David Chinnerda353b02007-08-28 14:00:13 +1000152 if (xfs_ipincount(ip)) {
153 read_unlock(&pag->pag_ici_lock);
154 xfs_log_force(mp, 0,
155 XFS_LOG_FORCE|XFS_LOG_SYNC);
156 XFS_STATS_INC(xs_ig_frecycle);
157 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 }
Nathan Scott745b1f472006-09-28 11:02:23 +1000159
Christoph Hellwig1543d792007-08-29 11:46:47 +1000160 vn_trace_exit(ip, "xfs_iget.alloc",
David Chinnerda353b02007-08-28 14:00:13 +1000161 (inst_t *)__return_address);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162
David Chinnerda353b02007-08-28 14:00:13 +1000163 XFS_STATS_INC(xs_ig_found);
164
165 xfs_iflags_clear(ip, XFS_IRECLAIMABLE);
166 read_unlock(&pag->pag_ici_lock);
167
168 XFS_MOUNT_ILOCK(mp);
169 list_del_init(&ip->i_reclaim);
170 XFS_MOUNT_IUNLOCK(mp);
171
172 goto finish_inode;
173
174 } else if (vp != inode_vp) {
175 struct inode *inode = vn_to_inode(inode_vp);
176
177 /* The inode is being torn down, pause and
178 * try again.
179 */
180 if (inode->i_state & (I_FREEING | I_CLEAR)) {
181 read_unlock(&pag->pag_ici_lock);
182 delay(1);
183 XFS_STATS_INC(xs_ig_frecycle);
184
185 goto again;
186 }
187/* Chances are the other vnode (the one in the inode) is being torn
188* down right now, and we landed on top of it. Question is, what do
189* we do? Unhook the old inode and hook up the new one?
190*/
191 cmn_err(CE_PANIC,
192 "xfs_iget_core: ambiguous vns: vp/0x%p, invp/0x%p",
193 inode_vp, vp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 }
David Chinnerda353b02007-08-28 14:00:13 +1000195
196 /*
197 * Inode cache hit
198 */
199 read_unlock(&pag->pag_ici_lock);
200 XFS_STATS_INC(xs_ig_found);
201
202finish_inode:
203 if (ip->i_d.di_mode == 0) {
204 if (!(flags & XFS_IGET_CREATE)) {
205 xfs_put_perag(mp, pag);
206 return ENOENT;
207 }
208 xfs_iocore_inode_reinit(ip);
209 }
210
211 if (lock_flags != 0)
212 xfs_ilock(ip, lock_flags);
213
214 xfs_iflags_clear(ip, XFS_ISTALE);
Christoph Hellwig1543d792007-08-29 11:46:47 +1000215 vn_trace_exit(ip, "xfs_iget.found",
David Chinnerda353b02007-08-28 14:00:13 +1000216 (inst_t *)__return_address);
217 goto return_ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 }
219
220 /*
David Chinnerda353b02007-08-28 14:00:13 +1000221 * Inode cache miss
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 */
David Chinnerda353b02007-08-28 14:00:13 +1000223 read_unlock(&pag->pag_ici_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 XFS_STATS_INC(xs_ig_missed);
225
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 /*
227 * Read the disk inode attributes into a new inode structure and get
228 * a new vnode for it. This should also initialize i_ino and i_mount.
229 */
Nathan Scott745b1f472006-09-28 11:02:23 +1000230 error = xfs_iread(mp, tp, ino, &ip, bno,
231 (flags & XFS_IGET_BULKSTAT) ? XFS_IMAP_BULKSTAT : 0);
David Chinnerda353b02007-08-28 14:00:13 +1000232 if (error) {
233 xfs_put_perag(mp, pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 return error;
David Chinnerda353b02007-08-28 14:00:13 +1000235 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236
Christoph Hellwig1543d792007-08-29 11:46:47 +1000237 vn_trace_exit(ip, "xfs_iget.alloc", (inst_t *)__return_address);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238
239 xfs_inode_lock_init(ip, vp);
240 xfs_iocore_inode_init(ip);
Nathan Scott745b1f472006-09-28 11:02:23 +1000241 if (lock_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 xfs_ilock(ip, lock_flags);
Nathan Scott745b1f472006-09-28 11:02:23 +1000243
244 if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 xfs_idestroy(ip);
David Chinnerda353b02007-08-28 14:00:13 +1000246 xfs_put_perag(mp, pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 return ENOENT;
248 }
249
250 /*
David Chinnerda353b02007-08-28 14:00:13 +1000251 * This is a bit messy - we preallocate everything we _might_
252 * need before we pick up the ici lock. That way we don't have to
253 * juggle locks and go all the way back to the start.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 */
David Chinnerda353b02007-08-28 14:00:13 +1000255 new_icl = kmem_zone_alloc(xfs_icluster_zone, KM_SLEEP);
256 if (radix_tree_preload(GFP_KERNEL)) {
257 delay(1);
258 goto again;
259 }
260 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
261 first_index = agino & mask;
262 write_lock(&pag->pag_ici_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263
David Chinnerda353b02007-08-28 14:00:13 +1000264 /*
265 * Find the cluster if it exists
266 */
267 icl = NULL;
268 if (radix_tree_gang_lookup(&pag->pag_ici_root, (void**)&iq,
269 first_index, 1)) {
270 if ((iq->i_ino & mask) == first_index)
271 icl = iq->i_cluster;
272 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273
David Chinnerda353b02007-08-28 14:00:13 +1000274 /*
275 * insert the new inode
276 */
277 error = radix_tree_insert(&pag->pag_ici_root, agino, ip);
278 if (unlikely(error)) {
279 BUG_ON(error != -EEXIST);
280 write_unlock(&pag->pag_ici_lock);
281 radix_tree_preload_end();
282 xfs_idestroy(ip);
283 XFS_STATS_INC(xs_ig_dup);
284 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 }
286
287 /*
288 * These values _must_ be set before releasing ihlock!
289 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 ip->i_udquot = ip->i_gdquot = NULL;
David Chinner7a18c382006-11-11 18:04:54 +1100291 xfs_iflags_set(ip, XFS_INEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292
David Chinnerda353b02007-08-28 14:00:13 +1000293 ASSERT(ip->i_cluster == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294
David Chinnerda353b02007-08-28 14:00:13 +1000295 if (!icl) {
296 spin_lock_init(&new_icl->icl_lock);
297 INIT_HLIST_HEAD(&new_icl->icl_inodes);
298 icl = new_icl;
299 new_icl = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 } else {
David Chinnerda353b02007-08-28 14:00:13 +1000301 ASSERT(!hlist_empty(&icl->icl_inodes));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 }
David Chinnerda353b02007-08-28 14:00:13 +1000303 spin_lock(&icl->icl_lock);
304 hlist_add_head(&ip->i_cnode, &icl->icl_inodes);
305 ip->i_cluster = icl;
306 spin_unlock(&icl->icl_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307
David Chinnerda353b02007-08-28 14:00:13 +1000308 write_unlock(&pag->pag_ici_lock);
309 radix_tree_preload_end();
310 if (new_icl)
311 kmem_zone_free(xfs_icluster_zone, new_icl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312
313 /*
314 * Link ip to its mount and thread it on the mount's inode list.
315 */
316 XFS_MOUNT_ILOCK(mp);
317 if ((iq = mp->m_inodes)) {
318 ASSERT(iq->i_mprev->i_mnext == iq);
319 ip->i_mprev = iq->i_mprev;
320 iq->i_mprev->i_mnext = ip;
321 iq->i_mprev = ip;
322 ip->i_mnext = iq;
323 } else {
324 ip->i_mnext = ip;
325 ip->i_mprev = ip;
326 }
327 mp->m_inodes = ip;
328
329 XFS_MOUNT_IUNLOCK(mp);
David Chinnerda353b02007-08-28 14:00:13 +1000330 xfs_put_perag(mp, pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331
332 return_ip:
333 ASSERT(ip->i_df.if_ext_max ==
334 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t));
335
336 ASSERT(((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) != 0) ==
337 ((ip->i_iocore.io_flags & XFS_IOCORE_RT) != 0));
338
Christoph Hellwigb3aea4e2007-08-29 11:44:37 +1000339 xfs_iflags_set(ip, XFS_IMODIFIED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 *ipp = ip;
341
342 /*
343 * If we have a real type for an on-disk inode, we can set ops(&unlock)
344 * now. If it's a new inode being created, xfs_ialloc will handle it.
345 */
Christoph Hellwig739bfb22007-08-29 10:58:01 +1000346 bhv_vfs_init_vnode(XFS_MTOVFS(mp), vp, ip, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347
348 return 0;
349}
350
351
352/*
353 * The 'normal' internal xfs_iget, if needed it will
354 * 'allocate', or 'get', the vnode.
355 */
356int
357xfs_iget(
358 xfs_mount_t *mp,
359 xfs_trans_t *tp,
360 xfs_ino_t ino,
361 uint flags,
362 uint lock_flags,
363 xfs_inode_t **ipp,
364 xfs_daddr_t bno)
365{
366 struct inode *inode;
Nathan Scott67fcaa72006-06-09 17:00:52 +1000367 bhv_vnode_t *vp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 int error;
369
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 XFS_STATS_INC(xs_ig_attempts);
371
Christoph Hellwigba403ab2005-09-05 08:33:00 +1000372retry:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 if ((inode = iget_locked(XFS_MTOVFS(mp)->vfs_super, ino))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 xfs_inode_t *ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375
Nathan Scottec86dc02006-03-17 17:25:36 +1100376 vp = vn_from_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 if (inode->i_state & I_NEW) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 vn_initialize(inode);
379 error = xfs_iget_core(vp, mp, tp, ino, flags,
380 lock_flags, ipp, bno);
381 if (error) {
382 vn_mark_bad(vp);
383 if (inode->i_state & I_NEW)
384 unlock_new_inode(inode);
385 iput(inode);
386 }
387 } else {
Christoph Hellwigba403ab2005-09-05 08:33:00 +1000388 /*
389 * If the inode is not fully constructed due to
Nathan Scottc41564b2006-03-29 08:55:14 +1000390 * filehandle mismatches wait for the inode to go
Christoph Hellwigba403ab2005-09-05 08:33:00 +1000391 * away and try again.
392 *
393 * iget_locked will call __wait_on_freeing_inode
394 * to wait for the inode to go away.
395 */
396 if (is_bad_inode(inode) ||
Christoph Hellwig75e17b32006-01-11 20:58:44 +1100397 ((ip = xfs_vtoi(vp)) == NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 iput(inode);
Christoph Hellwigba403ab2005-09-05 08:33:00 +1000399 delay(1);
400 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 }
402
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 if (lock_flags != 0)
404 xfs_ilock(ip, lock_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 XFS_STATS_INC(xs_ig_found);
406 *ipp = ip;
407 error = 0;
408 }
409 } else
410 error = ENOMEM; /* If we got no inode we are out of memory */
411
412 return error;
413}
414
415/*
416 * Do the setup for the various locks within the incore inode.
417 */
418void
419xfs_inode_lock_init(
420 xfs_inode_t *ip,
Nathan Scott67fcaa72006-06-09 17:00:52 +1000421 bhv_vnode_t *vp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422{
423 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
Christoph Hellwig2aeaa252007-08-29 11:46:57 +1000424 "xfsino", ip->i_ino);
425 mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 init_waitqueue_head(&ip->i_ipin_wait);
427 atomic_set(&ip->i_pincount, 0);
Eric Sandeen7ae67d72006-09-28 11:05:46 +1000428 initnsema(&ip->i_flock, 1, "xfsfino");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429}
430
431/*
432 * Look for the inode corresponding to the given ino in the hash table.
433 * If it is there and its i_transp pointer matches tp, return it.
434 * Otherwise, return NULL.
435 */
436xfs_inode_t *
437xfs_inode_incore(xfs_mount_t *mp,
438 xfs_ino_t ino,
439 xfs_trans_t *tp)
440{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 xfs_inode_t *ip;
David Chinnerda353b02007-08-28 14:00:13 +1000442 xfs_perag_t *pag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443
David Chinnerda353b02007-08-28 14:00:13 +1000444 pag = xfs_get_perag(mp, ino);
445 read_lock(&pag->pag_ici_lock);
446 ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ino));
447 read_unlock(&pag->pag_ici_lock);
448 xfs_put_perag(mp, pag);
449
450 /* the returned inode must match the transaction */
451 if (ip && (ip->i_transp != tp))
452 return NULL;
453 return ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454}
455
456/*
457 * Decrement reference count of an inode structure and unlock it.
458 *
459 * ip -- the inode being released
460 * lock_flags -- this parameter indicates the inode's locks to be
461 * to be released. See the comment on xfs_iunlock() for a list
462 * of valid values.
463 */
464void
465xfs_iput(xfs_inode_t *ip,
466 uint lock_flags)
467{
Nathan Scott67fcaa72006-06-09 17:00:52 +1000468 bhv_vnode_t *vp = XFS_ITOV(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469
Christoph Hellwig1543d792007-08-29 11:46:47 +1000470 vn_trace_entry(ip, "xfs_iput", (inst_t *)__return_address);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 xfs_iunlock(ip, lock_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 VN_RELE(vp);
473}
474
475/*
476 * Special iput for brand-new inodes that are still locked
477 */
478void
479xfs_iput_new(xfs_inode_t *ip,
480 uint lock_flags)
481{
Nathan Scott67fcaa72006-06-09 17:00:52 +1000482 bhv_vnode_t *vp = XFS_ITOV(ip);
Nathan Scottec86dc02006-03-17 17:25:36 +1100483 struct inode *inode = vn_to_inode(vp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484
Christoph Hellwig1543d792007-08-29 11:46:47 +1000485 vn_trace_entry(ip, "xfs_iput_new", (inst_t *)__return_address);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486
487 if ((ip->i_d.di_mode == 0)) {
David Chinner7a18c382006-11-11 18:04:54 +1100488 ASSERT(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 vn_mark_bad(vp);
490 }
491 if (inode->i_state & I_NEW)
492 unlock_new_inode(inode);
493 if (lock_flags)
494 xfs_iunlock(ip, lock_flags);
495 VN_RELE(vp);
496}
497
498
499/*
500 * This routine embodies the part of the reclaim code that pulls
501 * the inode from the inode hash table and the mount structure's
502 * inode list.
503 * This should only be called from xfs_reclaim().
504 */
505void
506xfs_ireclaim(xfs_inode_t *ip)
507{
Nathan Scott67fcaa72006-06-09 17:00:52 +1000508 bhv_vnode_t *vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509
510 /*
511 * Remove from old hash list and mount list.
512 */
513 XFS_STATS_INC(xs_ig_reclaims);
514
515 xfs_iextract(ip);
516
517 /*
518 * Here we do a spurious inode lock in order to coordinate with
519 * xfs_sync(). This is because xfs_sync() references the inodes
520 * in the mount list without taking references on the corresponding
521 * vnodes. We make that OK here by ensuring that we wait until
522 * the inode is unlocked in xfs_sync() before we go ahead and
523 * free it. We get both the regular lock and the io lock because
524 * the xfs_sync() code may need to drop the regular one but will
525 * still hold the io lock.
526 */
527 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
528
529 /*
530 * Release dquots (and their references) if any. An inode may escape
531 * xfs_inactive and get here via vn_alloc->vn_reclaim path.
532 */
533 XFS_QM_DQDETACH(ip->i_mount, ip);
534
535 /*
536 * Pull our behavior descriptor from the vnode chain.
537 */
538 vp = XFS_ITOV_NULL(ip);
539 if (vp) {
Christoph Hellwig739bfb22007-08-29 10:58:01 +1000540 vn_to_inode(vp)->i_private = NULL;
541 ip->i_vnode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 }
543
544 /*
545 * Free all memory associated with the inode.
546 */
Tim Shimmin439b8432006-11-11 18:04:34 +1100547 xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 xfs_idestroy(ip);
549}
550
551/*
552 * This routine removes an about-to-be-destroyed inode from
553 * all of the lists in which it is located with the exception
554 * of the behavior chain.
555 */
556void
557xfs_iextract(
558 xfs_inode_t *ip)
559{
David Chinnerda353b02007-08-28 14:00:13 +1000560 xfs_mount_t *mp = ip->i_mount;
561 xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 xfs_inode_t *iq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563
David Chinnerda353b02007-08-28 14:00:13 +1000564 write_lock(&pag->pag_ici_lock);
565 radix_tree_delete(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ip->i_ino));
566 write_unlock(&pag->pag_ici_lock);
567 xfs_put_perag(mp, pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568
569 /*
David Chinnerda353b02007-08-28 14:00:13 +1000570 * Remove from cluster list
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 */
572 mp = ip->i_mount;
David Chinnerda353b02007-08-28 14:00:13 +1000573 spin_lock(&ip->i_cluster->icl_lock);
574 hlist_del(&ip->i_cnode);
575 spin_unlock(&ip->i_cluster->icl_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576
David Chinnerda353b02007-08-28 14:00:13 +1000577 /* was last inode in cluster? */
578 if (hlist_empty(&ip->i_cluster->icl_inodes))
579 kmem_zone_free(xfs_icluster_zone, ip->i_cluster);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580
581 /*
582 * Remove from mount's inode list.
583 */
584 XFS_MOUNT_ILOCK(mp);
585 ASSERT((ip->i_mnext != NULL) && (ip->i_mprev != NULL));
586 iq = ip->i_mnext;
587 iq->i_mprev = ip->i_mprev;
588 ip->i_mprev->i_mnext = iq;
589
590 /*
591 * Fix up the head pointer if it points to the inode being deleted.
592 */
593 if (mp->m_inodes == ip) {
594 if (ip == iq) {
595 mp->m_inodes = NULL;
596 } else {
597 mp->m_inodes = iq;
598 }
599 }
600
601 /* Deal with the deleted inodes list */
602 list_del_init(&ip->i_reclaim);
603
604 mp->m_ireclaims++;
605 XFS_MOUNT_IUNLOCK(mp);
606}
607
608/*
609 * This is a wrapper routine around the xfs_ilock() routine
610 * used to centralize some grungy code. It is used in places
611 * that wish to lock the inode solely for reading the extents.
612 * The reason these places can't just call xfs_ilock(SHARED)
613 * is that the inode lock also guards to bringing in of the
614 * extents from disk for a file in b-tree format. If the inode
615 * is in b-tree format, then we need to lock the inode exclusively
616 * until the extents are read in. Locking it exclusively all
617 * the time would limit our parallelism unnecessarily, though.
618 * What we do instead is check to see if the extents have been
619 * read in yet, and only lock the inode exclusively if they
620 * have not.
621 *
622 * The function returns a value which should be given to the
623 * corresponding xfs_iunlock_map_shared(). This value is
624 * the mode in which the lock was actually taken.
625 */
626uint
627xfs_ilock_map_shared(
628 xfs_inode_t *ip)
629{
630 uint lock_mode;
631
632 if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) &&
633 ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) {
634 lock_mode = XFS_ILOCK_EXCL;
635 } else {
636 lock_mode = XFS_ILOCK_SHARED;
637 }
638
639 xfs_ilock(ip, lock_mode);
640
641 return lock_mode;
642}
643
644/*
645 * This is simply the unlock routine to go with xfs_ilock_map_shared().
646 * All it does is call xfs_iunlock() with the given lock_mode.
647 */
648void
649xfs_iunlock_map_shared(
650 xfs_inode_t *ip,
651 unsigned int lock_mode)
652{
653 xfs_iunlock(ip, lock_mode);
654}
655
656/*
657 * The xfs inode contains 2 locks: a multi-reader lock called the
658 * i_iolock and a multi-reader lock called the i_lock. This routine
659 * allows either or both of the locks to be obtained.
660 *
661 * The 2 locks should always be ordered so that the IO lock is
662 * obtained first in order to prevent deadlock.
663 *
664 * ip -- the inode being locked
665 * lock_flags -- this parameter indicates the inode's locks
666 * to be locked. It can be:
667 * XFS_IOLOCK_SHARED,
668 * XFS_IOLOCK_EXCL,
669 * XFS_ILOCK_SHARED,
670 * XFS_ILOCK_EXCL,
671 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
672 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
673 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
674 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
675 */
676void
677xfs_ilock(xfs_inode_t *ip,
678 uint lock_flags)
679{
680 /*
681 * You can't set both SHARED and EXCL for the same lock,
682 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
683 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
684 */
685 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
686 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
687 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
688 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000689 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690
691 if (lock_flags & XFS_IOLOCK_EXCL) {
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000692 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 } else if (lock_flags & XFS_IOLOCK_SHARED) {
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000694 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 }
696 if (lock_flags & XFS_ILOCK_EXCL) {
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000697 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 } else if (lock_flags & XFS_ILOCK_SHARED) {
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000699 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 }
701 xfs_ilock_trace(ip, 1, lock_flags, (inst_t *)__return_address);
702}
703
704/*
705 * This is just like xfs_ilock(), except that the caller
706 * is guaranteed not to sleep. It returns 1 if it gets
707 * the requested locks and 0 otherwise. If the IO lock is
708 * obtained but the inode lock cannot be, then the IO lock
709 * is dropped before returning.
710 *
711 * ip -- the inode being locked
712 * lock_flags -- this parameter indicates the inode's locks to be
713 * to be locked. See the comment for xfs_ilock() for a list
714 * of valid values.
715 *
716 */
717int
718xfs_ilock_nowait(xfs_inode_t *ip,
719 uint lock_flags)
720{
721 int iolocked;
722 int ilocked;
723
724 /*
725 * You can't set both SHARED and EXCL for the same lock,
726 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
727 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
728 */
729 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
730 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
731 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
732 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000733 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734
735 iolocked = 0;
736 if (lock_flags & XFS_IOLOCK_EXCL) {
737 iolocked = mrtryupdate(&ip->i_iolock);
738 if (!iolocked) {
739 return 0;
740 }
741 } else if (lock_flags & XFS_IOLOCK_SHARED) {
742 iolocked = mrtryaccess(&ip->i_iolock);
743 if (!iolocked) {
744 return 0;
745 }
746 }
747 if (lock_flags & XFS_ILOCK_EXCL) {
748 ilocked = mrtryupdate(&ip->i_lock);
749 if (!ilocked) {
750 if (iolocked) {
751 mrunlock(&ip->i_iolock);
752 }
753 return 0;
754 }
755 } else if (lock_flags & XFS_ILOCK_SHARED) {
756 ilocked = mrtryaccess(&ip->i_lock);
757 if (!ilocked) {
758 if (iolocked) {
759 mrunlock(&ip->i_iolock);
760 }
761 return 0;
762 }
763 }
764 xfs_ilock_trace(ip, 2, lock_flags, (inst_t *)__return_address);
765 return 1;
766}
767
768/*
769 * xfs_iunlock() is used to drop the inode locks acquired with
770 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
771 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
772 * that we know which locks to drop.
773 *
774 * ip -- the inode being unlocked
775 * lock_flags -- this parameter indicates the inode's locks to be
776 * to be unlocked. See the comment for xfs_ilock() for a list
777 * of valid values for this parameter.
778 *
779 */
780void
781xfs_iunlock(xfs_inode_t *ip,
782 uint lock_flags)
783{
784 /*
785 * You can't set both SHARED and EXCL for the same lock,
786 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
787 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
788 */
789 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
790 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
791 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
792 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +1000793 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_IUNLOCK_NONOTIFY |
794 XFS_LOCK_DEP_MASK)) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 ASSERT(lock_flags != 0);
796
797 if (lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) {
798 ASSERT(!(lock_flags & XFS_IOLOCK_SHARED) ||
799 (ismrlocked(&ip->i_iolock, MR_ACCESS)));
800 ASSERT(!(lock_flags & XFS_IOLOCK_EXCL) ||
801 (ismrlocked(&ip->i_iolock, MR_UPDATE)));
802 mrunlock(&ip->i_iolock);
803 }
804
805 if (lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) {
806 ASSERT(!(lock_flags & XFS_ILOCK_SHARED) ||
807 (ismrlocked(&ip->i_lock, MR_ACCESS)));
808 ASSERT(!(lock_flags & XFS_ILOCK_EXCL) ||
809 (ismrlocked(&ip->i_lock, MR_UPDATE)));
810 mrunlock(&ip->i_lock);
811
812 /*
813 * Let the AIL know that this item has been unlocked in case
814 * it is in the AIL and anyone is waiting on it. Don't do
815 * this if the caller has asked us not to.
816 */
817 if (!(lock_flags & XFS_IUNLOCK_NONOTIFY) &&
818 ip->i_itemp != NULL) {
819 xfs_trans_unlocked_item(ip->i_mount,
820 (xfs_log_item_t*)(ip->i_itemp));
821 }
822 }
823 xfs_ilock_trace(ip, 3, lock_flags, (inst_t *)__return_address);
824}
825
826/*
827 * give up write locks. the i/o lock cannot be held nested
828 * if it is being demoted.
829 */
830void
831xfs_ilock_demote(xfs_inode_t *ip,
832 uint lock_flags)
833{
834 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
835 ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
836
837 if (lock_flags & XFS_ILOCK_EXCL) {
838 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
839 mrdemote(&ip->i_lock);
840 }
841 if (lock_flags & XFS_IOLOCK_EXCL) {
842 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
843 mrdemote(&ip->i_iolock);
844 }
845}
846
847/*
848 * The following three routines simply manage the i_flock
849 * semaphore embedded in the inode. This semaphore synchronizes
850 * processes attempting to flush the in-core inode back to disk.
851 */
852void
853xfs_iflock(xfs_inode_t *ip)
854{
855 psema(&(ip->i_flock), PINOD|PLTWAIT);
856}
857
858int
859xfs_iflock_nowait(xfs_inode_t *ip)
860{
861 return (cpsema(&(ip->i_flock)));
862}
863
864void
865xfs_ifunlock(xfs_inode_t *ip)
866{
Al Viro0d8fee32006-06-19 08:41:30 +1000867 ASSERT(issemalocked(&(ip->i_flock)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 vsema(&(ip->i_flock));
869}