Loading...
1/*
2 * Copyright (c) 2002, 2007 Red Hat, Inc. All rights reserved.
3 *
4 * This software may be freely redistributed under the terms of the
5 * GNU General Public License.
6 *
7 * You should have received a copy of the GNU General Public License
8 * along with this program; if not, write to the Free Software
9 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
10 *
11 * Authors: David Woodhouse <dwmw2@infradead.org>
12 * David Howells <dhowells@redhat.com>
13 *
14 */
15
16#include <linux/kernel.h>
17#include <linux/module.h>
18#include <linux/init.h>
19#include <linux/circ_buf.h>
20#include <linux/sched.h>
21#include "internal.h"
22
23/*
24 * Allow the fileserver to request callback state (re-)initialisation.
25 * Unfortunately, UUIDs are not guaranteed unique.
26 */
27void afs_init_callback_state(struct afs_server *server)
28{
29 rcu_read_lock();
30 do {
31 server->cb_s_break++;
32 server = rcu_dereference(server->uuid_next);
33 } while (0);
34 rcu_read_unlock();
35}
36
37/*
38 * actually break a callback
39 */
40void __afs_break_callback(struct afs_vnode *vnode, enum afs_cb_break_reason reason)
41{
42 _enter("");
43
44 clear_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags);
45 if (test_and_clear_bit(AFS_VNODE_CB_PROMISED, &vnode->flags)) {
46 vnode->cb_break++;
47 afs_clear_permits(vnode);
48
49 if (vnode->lock_state == AFS_VNODE_LOCK_WAITING_FOR_CB)
50 afs_lock_may_be_available(vnode);
51
52 trace_afs_cb_break(&vnode->fid, vnode->cb_break, reason, true);
53 } else {
54 trace_afs_cb_break(&vnode->fid, vnode->cb_break, reason, false);
55 }
56}
57
58void afs_break_callback(struct afs_vnode *vnode, enum afs_cb_break_reason reason)
59{
60 write_seqlock(&vnode->cb_lock);
61 __afs_break_callback(vnode, reason);
62 write_sequnlock(&vnode->cb_lock);
63}
64
65/*
66 * Look up a volume by volume ID under RCU conditions.
67 */
68static struct afs_volume *afs_lookup_volume_rcu(struct afs_cell *cell,
69 afs_volid_t vid)
70{
71 struct afs_volume *volume = NULL;
72 struct rb_node *p;
73 int seq = 0;
74
75 do {
76 /* Unfortunately, rbtree walking doesn't give reliable results
77 * under just the RCU read lock, so we have to check for
78 * changes.
79 */
80 read_seqbegin_or_lock(&cell->volume_lock, &seq);
81
82 p = rcu_dereference_raw(cell->volumes.rb_node);
83 while (p) {
84 volume = rb_entry(p, struct afs_volume, cell_node);
85
86 if (volume->vid < vid)
87 p = rcu_dereference_raw(p->rb_left);
88 else if (volume->vid > vid)
89 p = rcu_dereference_raw(p->rb_right);
90 else
91 break;
92 volume = NULL;
93 }
94
95 } while (need_seqretry(&cell->volume_lock, seq));
96
97 done_seqretry(&cell->volume_lock, seq);
98 return volume;
99}
100
101/*
102 * allow the fileserver to explicitly break one callback
103 * - happens when
104 * - the backing file is changed
105 * - a lock is released
106 */
107static void afs_break_one_callback(struct afs_volume *volume,
108 struct afs_fid *fid)
109{
110 struct super_block *sb;
111 struct afs_vnode *vnode;
112 struct inode *inode;
113
114 if (fid->vnode == 0 && fid->unique == 0) {
115 /* The callback break applies to an entire volume. */
116 write_lock(&volume->cb_v_break_lock);
117 volume->cb_v_break++;
118 trace_afs_cb_break(fid, volume->cb_v_break,
119 afs_cb_break_for_volume_callback, false);
120 write_unlock(&volume->cb_v_break_lock);
121 return;
122 }
123
124 /* See if we can find a matching inode - even an I_NEW inode needs to
125 * be marked as it can have its callback broken before we finish
126 * setting up the local inode.
127 */
128 sb = rcu_dereference(volume->sb);
129 if (!sb)
130 return;
131
132 inode = find_inode_rcu(sb, fid->vnode, afs_ilookup5_test_by_fid, fid);
133 if (inode) {
134 vnode = AFS_FS_I(inode);
135 afs_break_callback(vnode, afs_cb_break_for_callback);
136 } else {
137 trace_afs_cb_miss(fid, afs_cb_break_for_callback);
138 }
139}
140
141static void afs_break_some_callbacks(struct afs_server *server,
142 struct afs_callback_break *cbb,
143 size_t *_count)
144{
145 struct afs_callback_break *residue = cbb;
146 struct afs_volume *volume;
147 afs_volid_t vid = cbb->fid.vid;
148 size_t i;
149
150 volume = afs_lookup_volume_rcu(server->cell, vid);
151
152 /* TODO: Find all matching volumes if we couldn't match the server and
153 * break them anyway.
154 */
155
156 for (i = *_count; i > 0; cbb++, i--) {
157 if (cbb->fid.vid == vid) {
158 _debug("- Fid { vl=%08llx n=%llu u=%u }",
159 cbb->fid.vid,
160 cbb->fid.vnode,
161 cbb->fid.unique);
162 --*_count;
163 if (volume)
164 afs_break_one_callback(volume, &cbb->fid);
165 } else {
166 *residue++ = *cbb;
167 }
168 }
169}
170
171/*
172 * allow the fileserver to break callback promises
173 */
174void afs_break_callbacks(struct afs_server *server, size_t count,
175 struct afs_callback_break *callbacks)
176{
177 _enter("%p,%zu,", server, count);
178
179 ASSERT(server != NULL);
180
181 rcu_read_lock();
182
183 while (count > 0)
184 afs_break_some_callbacks(server, callbacks, &count);
185
186 rcu_read_unlock();
187 return;
188}
1/*
2 * Copyright (c) 2002, 2007 Red Hat, Inc. All rights reserved.
3 *
4 * This software may be freely redistributed under the terms of the
5 * GNU General Public License.
6 *
7 * You should have received a copy of the GNU General Public License
8 * along with this program; if not, write to the Free Software
9 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
10 *
11 * Authors: David Woodhouse <dwmw2@infradead.org>
12 * David Howells <dhowells@redhat.com>
13 *
14 */
15
16#include <linux/kernel.h>
17#include <linux/module.h>
18#include <linux/init.h>
19#include <linux/circ_buf.h>
20#include <linux/sched.h>
21#include "internal.h"
22
23#if 0
24unsigned afs_vnode_update_timeout = 10;
25#endif /* 0 */
26
27#define afs_breakring_space(server) \
28 CIRC_SPACE((server)->cb_break_head, (server)->cb_break_tail, \
29 ARRAY_SIZE((server)->cb_break))
30
31//static void afs_callback_updater(struct work_struct *);
32
33static struct workqueue_struct *afs_callback_update_worker;
34
35/*
36 * allow the fileserver to request callback state (re-)initialisation
37 */
38void afs_init_callback_state(struct afs_server *server)
39{
40 struct afs_vnode *vnode;
41
42 _enter("{%p}", server);
43
44 spin_lock(&server->cb_lock);
45
46 /* kill all the promises on record from this server */
47 while (!RB_EMPTY_ROOT(&server->cb_promises)) {
48 vnode = rb_entry(server->cb_promises.rb_node,
49 struct afs_vnode, cb_promise);
50 _debug("UNPROMISE { vid=%x:%u uq=%u}",
51 vnode->fid.vid, vnode->fid.vnode, vnode->fid.unique);
52 rb_erase(&vnode->cb_promise, &server->cb_promises);
53 vnode->cb_promised = false;
54 }
55
56 spin_unlock(&server->cb_lock);
57 _leave("");
58}
59
60/*
61 * handle the data invalidation side of a callback being broken
62 */
63void afs_broken_callback_work(struct work_struct *work)
64{
65 struct afs_vnode *vnode =
66 container_of(work, struct afs_vnode, cb_broken_work);
67
68 _enter("");
69
70 if (test_bit(AFS_VNODE_DELETED, &vnode->flags))
71 return;
72
73 /* we're only interested in dealing with a broken callback on *this*
74 * vnode and only if no-one else has dealt with it yet */
75 if (!mutex_trylock(&vnode->validate_lock))
76 return; /* someone else is dealing with it */
77
78 if (test_bit(AFS_VNODE_CB_BROKEN, &vnode->flags)) {
79 if (S_ISDIR(vnode->vfs_inode.i_mode))
80 afs_clear_permits(vnode);
81
82 if (afs_vnode_fetch_status(vnode, NULL, NULL) < 0)
83 goto out;
84
85 if (test_bit(AFS_VNODE_DELETED, &vnode->flags))
86 goto out;
87
88 /* if the vnode's data version number changed then its contents
89 * are different */
90 if (test_and_clear_bit(AFS_VNODE_ZAP_DATA, &vnode->flags))
91 afs_zap_data(vnode);
92 }
93
94out:
95 mutex_unlock(&vnode->validate_lock);
96
97 /* avoid the potential race whereby the mutex_trylock() in this
98 * function happens again between the clear_bit() and the
99 * mutex_unlock() */
100 if (test_bit(AFS_VNODE_CB_BROKEN, &vnode->flags)) {
101 _debug("requeue");
102 queue_work(afs_callback_update_worker, &vnode->cb_broken_work);
103 }
104 _leave("");
105}
106
107/*
108 * actually break a callback
109 */
110static void afs_break_callback(struct afs_server *server,
111 struct afs_vnode *vnode)
112{
113 _enter("");
114
115 set_bit(AFS_VNODE_CB_BROKEN, &vnode->flags);
116
117 if (vnode->cb_promised) {
118 spin_lock(&vnode->lock);
119
120 _debug("break callback");
121
122 spin_lock(&server->cb_lock);
123 if (vnode->cb_promised) {
124 rb_erase(&vnode->cb_promise, &server->cb_promises);
125 vnode->cb_promised = false;
126 }
127 spin_unlock(&server->cb_lock);
128
129 queue_work(afs_callback_update_worker, &vnode->cb_broken_work);
130 if (list_empty(&vnode->granted_locks) &&
131 !list_empty(&vnode->pending_locks))
132 afs_lock_may_be_available(vnode);
133 spin_unlock(&vnode->lock);
134 }
135}
136
137/*
138 * allow the fileserver to explicitly break one callback
139 * - happens when
140 * - the backing file is changed
141 * - a lock is released
142 */
143static void afs_break_one_callback(struct afs_server *server,
144 struct afs_fid *fid)
145{
146 struct afs_vnode *vnode;
147 struct rb_node *p;
148
149 _debug("find");
150 spin_lock(&server->fs_lock);
151 p = server->fs_vnodes.rb_node;
152 while (p) {
153 vnode = rb_entry(p, struct afs_vnode, server_rb);
154 if (fid->vid < vnode->fid.vid)
155 p = p->rb_left;
156 else if (fid->vid > vnode->fid.vid)
157 p = p->rb_right;
158 else if (fid->vnode < vnode->fid.vnode)
159 p = p->rb_left;
160 else if (fid->vnode > vnode->fid.vnode)
161 p = p->rb_right;
162 else if (fid->unique < vnode->fid.unique)
163 p = p->rb_left;
164 else if (fid->unique > vnode->fid.unique)
165 p = p->rb_right;
166 else
167 goto found;
168 }
169
170 /* not found so we just ignore it (it may have moved to another
171 * server) */
172not_available:
173 _debug("not avail");
174 spin_unlock(&server->fs_lock);
175 _leave("");
176 return;
177
178found:
179 _debug("found");
180 ASSERTCMP(server, ==, vnode->server);
181
182 if (!igrab(AFS_VNODE_TO_I(vnode)))
183 goto not_available;
184 spin_unlock(&server->fs_lock);
185
186 afs_break_callback(server, vnode);
187 iput(&vnode->vfs_inode);
188 _leave("");
189}
190
191/*
192 * allow the fileserver to break callback promises
193 */
194void afs_break_callbacks(struct afs_server *server, size_t count,
195 struct afs_callback callbacks[])
196{
197 _enter("%p,%zu,", server, count);
198
199 ASSERT(server != NULL);
200 ASSERTCMP(count, <=, AFSCBMAX);
201
202 for (; count > 0; callbacks++, count--) {
203 _debug("- Fid { vl=%08x n=%u u=%u } CB { v=%u x=%u t=%u }",
204 callbacks->fid.vid,
205 callbacks->fid.vnode,
206 callbacks->fid.unique,
207 callbacks->version,
208 callbacks->expiry,
209 callbacks->type
210 );
211 afs_break_one_callback(server, &callbacks->fid);
212 }
213
214 _leave("");
215 return;
216}
217
218/*
219 * record the callback for breaking
220 * - the caller must hold server->cb_lock
221 */
222static void afs_do_give_up_callback(struct afs_server *server,
223 struct afs_vnode *vnode)
224{
225 struct afs_callback *cb;
226
227 _enter("%p,%p", server, vnode);
228
229 cb = &server->cb_break[server->cb_break_head];
230 cb->fid = vnode->fid;
231 cb->version = vnode->cb_version;
232 cb->expiry = vnode->cb_expiry;
233 cb->type = vnode->cb_type;
234 smp_wmb();
235 server->cb_break_head =
236 (server->cb_break_head + 1) &
237 (ARRAY_SIZE(server->cb_break) - 1);
238
239 /* defer the breaking of callbacks to try and collect as many as
240 * possible to ship in one operation */
241 switch (atomic_inc_return(&server->cb_break_n)) {
242 case 1 ... AFSCBMAX - 1:
243 queue_delayed_work(afs_callback_update_worker,
244 &server->cb_break_work, HZ * 2);
245 break;
246 case AFSCBMAX:
247 afs_flush_callback_breaks(server);
248 break;
249 default:
250 break;
251 }
252
253 ASSERT(server->cb_promises.rb_node != NULL);
254 rb_erase(&vnode->cb_promise, &server->cb_promises);
255 vnode->cb_promised = false;
256 _leave("");
257}
258
259/*
260 * discard the callback on a deleted item
261 */
262void afs_discard_callback_on_delete(struct afs_vnode *vnode)
263{
264 struct afs_server *server = vnode->server;
265
266 _enter("%d", vnode->cb_promised);
267
268 if (!vnode->cb_promised) {
269 _leave(" [not promised]");
270 return;
271 }
272
273 ASSERT(server != NULL);
274
275 spin_lock(&server->cb_lock);
276 if (vnode->cb_promised) {
277 ASSERT(server->cb_promises.rb_node != NULL);
278 rb_erase(&vnode->cb_promise, &server->cb_promises);
279 vnode->cb_promised = false;
280 }
281 spin_unlock(&server->cb_lock);
282 _leave("");
283}
284
285/*
286 * give up the callback registered for a vnode on the file server when the
287 * inode is being cleared
288 */
289void afs_give_up_callback(struct afs_vnode *vnode)
290{
291 struct afs_server *server = vnode->server;
292
293 DECLARE_WAITQUEUE(myself, current);
294
295 _enter("%d", vnode->cb_promised);
296
297 _debug("GIVE UP INODE %p", &vnode->vfs_inode);
298
299 if (!vnode->cb_promised) {
300 _leave(" [not promised]");
301 return;
302 }
303
304 ASSERT(server != NULL);
305
306 spin_lock(&server->cb_lock);
307 if (vnode->cb_promised && afs_breakring_space(server) == 0) {
308 add_wait_queue(&server->cb_break_waitq, &myself);
309 for (;;) {
310 set_current_state(TASK_UNINTERRUPTIBLE);
311 if (!vnode->cb_promised ||
312 afs_breakring_space(server) != 0)
313 break;
314 spin_unlock(&server->cb_lock);
315 schedule();
316 spin_lock(&server->cb_lock);
317 }
318 remove_wait_queue(&server->cb_break_waitq, &myself);
319 __set_current_state(TASK_RUNNING);
320 }
321
322 /* of course, it's always possible for the server to break this vnode's
323 * callback first... */
324 if (vnode->cb_promised)
325 afs_do_give_up_callback(server, vnode);
326
327 spin_unlock(&server->cb_lock);
328 _leave("");
329}
330
331/*
332 * dispatch a deferred give up callbacks operation
333 */
334void afs_dispatch_give_up_callbacks(struct work_struct *work)
335{
336 struct afs_server *server =
337 container_of(work, struct afs_server, cb_break_work.work);
338
339 _enter("");
340
341 /* tell the fileserver to discard the callback promises it has
342 * - in the event of ENOMEM or some other error, we just forget that we
343 * had callbacks entirely, and the server will call us later to break
344 * them
345 */
346 afs_fs_give_up_callbacks(server, &afs_async_call);
347}
348
349/*
350 * flush the outstanding callback breaks on a server
351 */
352void afs_flush_callback_breaks(struct afs_server *server)
353{
354 mod_delayed_work(afs_callback_update_worker, &server->cb_break_work, 0);
355}
356
357#if 0
358/*
359 * update a bunch of callbacks
360 */
361static void afs_callback_updater(struct work_struct *work)
362{
363 struct afs_server *server;
364 struct afs_vnode *vnode, *xvnode;
365 time_t now;
366 long timeout;
367 int ret;
368
369 server = container_of(work, struct afs_server, updater);
370
371 _enter("");
372
373 now = get_seconds();
374
375 /* find the first vnode to update */
376 spin_lock(&server->cb_lock);
377 for (;;) {
378 if (RB_EMPTY_ROOT(&server->cb_promises)) {
379 spin_unlock(&server->cb_lock);
380 _leave(" [nothing]");
381 return;
382 }
383
384 vnode = rb_entry(rb_first(&server->cb_promises),
385 struct afs_vnode, cb_promise);
386 if (atomic_read(&vnode->usage) > 0)
387 break;
388 rb_erase(&vnode->cb_promise, &server->cb_promises);
389 vnode->cb_promised = false;
390 }
391
392 timeout = vnode->update_at - now;
393 if (timeout > 0) {
394 queue_delayed_work(afs_vnode_update_worker,
395 &afs_vnode_update, timeout * HZ);
396 spin_unlock(&server->cb_lock);
397 _leave(" [nothing]");
398 return;
399 }
400
401 list_del_init(&vnode->update);
402 atomic_inc(&vnode->usage);
403 spin_unlock(&server->cb_lock);
404
405 /* we can now perform the update */
406 _debug("update %s", vnode->vldb.name);
407 vnode->state = AFS_VL_UPDATING;
408 vnode->upd_rej_cnt = 0;
409 vnode->upd_busy_cnt = 0;
410
411 ret = afs_vnode_update_record(vl, &vldb);
412 switch (ret) {
413 case 0:
414 afs_vnode_apply_update(vl, &vldb);
415 vnode->state = AFS_VL_UPDATING;
416 break;
417 case -ENOMEDIUM:
418 vnode->state = AFS_VL_VOLUME_DELETED;
419 break;
420 default:
421 vnode->state = AFS_VL_UNCERTAIN;
422 break;
423 }
424
425 /* and then reschedule */
426 _debug("reschedule");
427 vnode->update_at = get_seconds() + afs_vnode_update_timeout;
428
429 spin_lock(&server->cb_lock);
430
431 if (!list_empty(&server->cb_promises)) {
432 /* next update in 10 minutes, but wait at least 1 second more
433 * than the newest record already queued so that we don't spam
434 * the VL server suddenly with lots of requests
435 */
436 xvnode = list_entry(server->cb_promises.prev,
437 struct afs_vnode, update);
438 if (vnode->update_at <= xvnode->update_at)
439 vnode->update_at = xvnode->update_at + 1;
440 xvnode = list_entry(server->cb_promises.next,
441 struct afs_vnode, update);
442 timeout = xvnode->update_at - now;
443 if (timeout < 0)
444 timeout = 0;
445 } else {
446 timeout = afs_vnode_update_timeout;
447 }
448
449 list_add_tail(&vnode->update, &server->cb_promises);
450
451 _debug("timeout %ld", timeout);
452 queue_delayed_work(afs_vnode_update_worker,
453 &afs_vnode_update, timeout * HZ);
454 spin_unlock(&server->cb_lock);
455 afs_put_vnode(vl);
456}
457#endif
458
459/*
460 * initialise the callback update process
461 */
462int __init afs_callback_update_init(void)
463{
464 afs_callback_update_worker =
465 create_singlethread_workqueue("kafs_callbackd");
466 return afs_callback_update_worker ? 0 : -ENOMEM;
467}
468
469/*
470 * shut down the callback update process
471 */
472void afs_callback_update_kill(void)
473{
474 destroy_workqueue(afs_callback_update_worker);
475}