diff options
Diffstat (limited to 'fs/afs/callback.c')
-rw-r--r-- | fs/afs/callback.c | 469 |
1 files changed, 373 insertions, 96 deletions
diff --git a/fs/afs/callback.c b/fs/afs/callback.c index 26a48fea42f4..611215547142 100644 --- a/fs/afs/callback.c +++ b/fs/afs/callback.c | |||
@@ -1,5 +1,5 @@ | |||
1 | /* | 1 | /* |
2 | * Copyright (c) 2002 Red Hat, Inc. All rights reserved. | 2 | * Copyright (c) 2002, 2007 Red Hat, Inc. All rights reserved. |
3 | * | 3 | * |
4 | * This software may be freely redistributed under the terms of the | 4 | * This software may be freely redistributed under the terms of the |
5 | * GNU General Public License. | 5 | * GNU General Public License. |
@@ -16,83 +16,182 @@ | |||
16 | #include <linux/kernel.h> | 16 | #include <linux/kernel.h> |
17 | #include <linux/module.h> | 17 | #include <linux/module.h> |
18 | #include <linux/init.h> | 18 | #include <linux/init.h> |
19 | #include "server.h" | 19 | #include <linux/circ_buf.h> |
20 | #include "vnode.h" | ||
21 | #include "internal.h" | 20 | #include "internal.h" |
22 | #include "cmservice.h" | 21 | |
22 | unsigned afs_vnode_update_timeout = 10; | ||
23 | |||
24 | #define afs_breakring_space(server) \ | ||
25 | CIRC_SPACE((server)->cb_break_head, (server)->cb_break_tail, \ | ||
26 | ARRAY_SIZE((server)->cb_break)) | ||
27 | |||
28 | //static void afs_callback_updater(struct work_struct *); | ||
29 | |||
30 | static struct workqueue_struct *afs_callback_update_worker; | ||
23 | 31 | ||
24 | /* | 32 | /* |
25 | * allow the fileserver to request callback state (re-)initialisation | 33 | * allow the fileserver to request callback state (re-)initialisation |
26 | */ | 34 | */ |
27 | int SRXAFSCM_InitCallBackState(struct afs_server *server) | 35 | void afs_init_callback_state(struct afs_server *server) |
28 | { | 36 | { |
29 | struct list_head callbacks; | 37 | struct afs_vnode *vnode; |
30 | 38 | ||
31 | _enter("%p", server); | 39 | _enter("{%p}", server); |
32 | 40 | ||
33 | INIT_LIST_HEAD(&callbacks); | ||
34 | |||
35 | /* transfer the callback list from the server to a temp holding area */ | ||
36 | spin_lock(&server->cb_lock); | 41 | spin_lock(&server->cb_lock); |
37 | 42 | ||
38 | list_add(&callbacks, &server->cb_promises); | 43 | /* kill all the promises on record from this server */ |
39 | list_del_init(&server->cb_promises); | 44 | while (!RB_EMPTY_ROOT(&server->cb_promises)) { |
45 | vnode = rb_entry(server->cb_promises.rb_node, | ||
46 | struct afs_vnode, cb_promise); | ||
47 | printk("\nUNPROMISE on %p\n", vnode); | ||
48 | rb_erase(&vnode->cb_promise, &server->cb_promises); | ||
49 | vnode->cb_promised = false; | ||
50 | } | ||
40 | 51 | ||
41 | /* munch our way through the list, grabbing the inode, dropping all the | 52 | spin_unlock(&server->cb_lock); |
42 | * locks and regetting them in the right order | 53 | _leave(""); |
43 | */ | 54 | } |
44 | while (!list_empty(&callbacks)) { | ||
45 | struct afs_vnode *vnode; | ||
46 | struct inode *inode; | ||
47 | 55 | ||
48 | vnode = list_entry(callbacks.next, struct afs_vnode, cb_link); | 56 | /* |
49 | list_del_init(&vnode->cb_link); | 57 | * handle the data invalidation side of a callback being broken |
58 | */ | ||
59 | void afs_broken_callback_work(struct work_struct *work) | ||
60 | { | ||
61 | struct afs_vnode *vnode = | ||
62 | container_of(work, struct afs_vnode, cb_broken_work); | ||
50 | 63 | ||
51 | /* try and grab the inode - may fail */ | 64 | _enter(""); |
52 | inode = igrab(AFS_VNODE_TO_I(vnode)); | ||
53 | if (inode) { | ||
54 | int release = 0; | ||
55 | 65 | ||
56 | spin_unlock(&server->cb_lock); | 66 | if (test_bit(AFS_VNODE_DELETED, &vnode->flags)) |
57 | spin_lock(&vnode->lock); | 67 | return; |
58 | 68 | ||
59 | if (vnode->cb_server == server) { | 69 | /* we're only interested in dealing with a broken callback on *this* |
60 | vnode->cb_server = NULL; | 70 | * vnode and only if no-one else has dealt with it yet */ |
61 | afs_kafstimod_del_timer(&vnode->cb_timeout); | 71 | if (!mutex_trylock(&vnode->cb_broken_lock)) |
62 | spin_lock(&afs_cb_hash_lock); | 72 | return; /* someone else is dealing with it */ |
63 | list_del_init(&vnode->cb_hash_link); | ||
64 | spin_unlock(&afs_cb_hash_lock); | ||
65 | release = 1; | ||
66 | } | ||
67 | 73 | ||
68 | spin_unlock(&vnode->lock); | 74 | if (test_bit(AFS_VNODE_CB_BROKEN, &vnode->flags)) { |
75 | if (afs_vnode_fetch_status(vnode) < 0) | ||
76 | goto out; | ||
69 | 77 | ||
70 | iput(inode); | 78 | if (test_bit(AFS_VNODE_DELETED, &vnode->flags)) |
71 | afs_put_server(server); | 79 | goto out; |
72 | 80 | ||
73 | spin_lock(&server->cb_lock); | 81 | /* if the vnode's data version number changed then its contents |
82 | * are different */ | ||
83 | if (test_and_clear_bit(AFS_VNODE_ZAP_DATA, &vnode->flags)) { | ||
84 | _debug("zap data"); | ||
85 | invalidate_remote_inode(&vnode->vfs_inode); | ||
74 | } | 86 | } |
75 | } | 87 | } |
76 | 88 | ||
77 | spin_unlock(&server->cb_lock); | 89 | out: |
90 | mutex_unlock(&vnode->cb_broken_lock); | ||
78 | 91 | ||
79 | _leave(" = 0"); | 92 | /* avoid the potential race whereby the mutex_trylock() in this |
80 | return 0; | 93 | * function happens again between the clear_bit() and the |
94 | * mutex_unlock() */ | ||
95 | if (test_bit(AFS_VNODE_CB_BROKEN, &vnode->flags)) { | ||
96 | _debug("requeue"); | ||
97 | queue_work(afs_callback_update_worker, &vnode->cb_broken_work); | ||
98 | } | ||
99 | _leave(""); | ||
100 | } | ||
101 | |||
102 | /* | ||
103 | * actually break a callback | ||
104 | */ | ||
105 | static void afs_break_callback(struct afs_server *server, | ||
106 | struct afs_vnode *vnode) | ||
107 | { | ||
108 | _enter(""); | ||
109 | |||
110 | set_bit(AFS_VNODE_CB_BROKEN, &vnode->flags); | ||
111 | |||
112 | if (vnode->cb_promised) { | ||
113 | spin_lock(&vnode->lock); | ||
114 | |||
115 | _debug("break callback"); | ||
116 | |||
117 | spin_lock(&server->cb_lock); | ||
118 | if (vnode->cb_promised) { | ||
119 | rb_erase(&vnode->cb_promise, &server->cb_promises); | ||
120 | vnode->cb_promised = false; | ||
121 | } | ||
122 | spin_unlock(&server->cb_lock); | ||
123 | |||
124 | queue_work(afs_callback_update_worker, &vnode->cb_broken_work); | ||
125 | spin_unlock(&vnode->lock); | ||
126 | } | ||
127 | } | ||
128 | |||
129 | /* | ||
130 | * allow the fileserver to explicitly break one callback | ||
131 | * - happens when | ||
132 | * - the backing file is changed | ||
133 | * - a lock is released | ||
134 | */ | ||
135 | static void afs_break_one_callback(struct afs_server *server, | ||
136 | struct afs_fid *fid) | ||
137 | { | ||
138 | struct afs_vnode *vnode; | ||
139 | struct rb_node *p; | ||
140 | |||
141 | _debug("find"); | ||
142 | spin_lock(&server->fs_lock); | ||
143 | p = server->fs_vnodes.rb_node; | ||
144 | while (p) { | ||
145 | vnode = rb_entry(p, struct afs_vnode, server_rb); | ||
146 | if (fid->vid < vnode->fid.vid) | ||
147 | p = p->rb_left; | ||
148 | else if (fid->vid > vnode->fid.vid) | ||
149 | p = p->rb_right; | ||
150 | else if (fid->vnode < vnode->fid.vnode) | ||
151 | p = p->rb_left; | ||
152 | else if (fid->vnode > vnode->fid.vnode) | ||
153 | p = p->rb_right; | ||
154 | else if (fid->unique < vnode->fid.unique) | ||
155 | p = p->rb_left; | ||
156 | else if (fid->unique > vnode->fid.unique) | ||
157 | p = p->rb_right; | ||
158 | else | ||
159 | goto found; | ||
160 | } | ||
161 | |||
162 | /* not found so we just ignore it (it may have moved to another | ||
163 | * server) */ | ||
164 | not_available: | ||
165 | _debug("not avail"); | ||
166 | spin_unlock(&server->fs_lock); | ||
167 | _leave(""); | ||
168 | return; | ||
169 | |||
170 | found: | ||
171 | _debug("found"); | ||
172 | ASSERTCMP(server, ==, vnode->server); | ||
173 | |||
174 | if (!igrab(AFS_VNODE_TO_I(vnode))) | ||
175 | goto not_available; | ||
176 | spin_unlock(&server->fs_lock); | ||
177 | |||
178 | afs_break_callback(server, vnode); | ||
179 | iput(&vnode->vfs_inode); | ||
180 | _leave(""); | ||
81 | } | 181 | } |
82 | 182 | ||
83 | /* | 183 | /* |
84 | * allow the fileserver to break callback promises | 184 | * allow the fileserver to break callback promises |
85 | */ | 185 | */ |
86 | int SRXAFSCM_CallBack(struct afs_server *server, size_t count, | 186 | void afs_break_callbacks(struct afs_server *server, size_t count, |
87 | struct afs_callback callbacks[]) | 187 | struct afs_callback callbacks[]) |
88 | { | 188 | { |
89 | _enter("%p,%u,", server, count); | 189 | _enter("%p,%zu,", server, count); |
90 | 190 | ||
91 | for (; count > 0; callbacks++, count--) { | 191 | ASSERT(server != NULL); |
92 | struct afs_vnode *vnode = NULL; | 192 | ASSERTCMP(count, <=, AFSCBMAX); |
93 | struct inode *inode = NULL; | ||
94 | int valid = 0; | ||
95 | 193 | ||
194 | for (; count > 0; callbacks++, count--) { | ||
96 | _debug("- Fid { vl=%08x n=%u u=%u } CB { v=%u x=%u t=%u }", | 195 | _debug("- Fid { vl=%08x n=%u u=%u } CB { v=%u x=%u t=%u }", |
97 | callbacks->fid.vid, | 196 | callbacks->fid.vid, |
98 | callbacks->fid.vnode, | 197 | callbacks->fid.vnode, |
@@ -101,66 +200,244 @@ int SRXAFSCM_CallBack(struct afs_server *server, size_t count, | |||
101 | callbacks->expiry, | 200 | callbacks->expiry, |
102 | callbacks->type | 201 | callbacks->type |
103 | ); | 202 | ); |
203 | afs_break_one_callback(server, &callbacks->fid); | ||
204 | } | ||
205 | |||
206 | _leave(""); | ||
207 | return; | ||
208 | } | ||
104 | 209 | ||
105 | /* find the inode for this fid */ | 210 | /* |
106 | spin_lock(&afs_cb_hash_lock); | 211 | * record the callback for breaking |
212 | * - the caller must hold server->cb_lock | ||
213 | */ | ||
214 | static void afs_do_give_up_callback(struct afs_server *server, | ||
215 | struct afs_vnode *vnode) | ||
216 | { | ||
217 | struct afs_callback *cb; | ||
107 | 218 | ||
108 | list_for_each_entry(vnode, | 219 | _enter("%p,%p", server, vnode); |
109 | &afs_cb_hash(server, &callbacks->fid), | ||
110 | cb_hash_link) { | ||
111 | if (memcmp(&vnode->fid, &callbacks->fid, | ||
112 | sizeof(struct afs_fid)) != 0) | ||
113 | continue; | ||
114 | 220 | ||
115 | /* right vnode, but is it same server? */ | 221 | cb = &server->cb_break[server->cb_break_head]; |
116 | if (vnode->cb_server != server) | 222 | cb->fid = vnode->fid; |
117 | break; /* no */ | 223 | cb->version = vnode->cb_version; |
224 | cb->expiry = vnode->cb_expiry; | ||
225 | cb->type = vnode->cb_type; | ||
226 | smp_wmb(); | ||
227 | server->cb_break_head = | ||
228 | (server->cb_break_head + 1) & | ||
229 | (ARRAY_SIZE(server->cb_break) - 1); | ||
118 | 230 | ||
119 | /* try and nail the inode down */ | 231 | /* defer the breaking of callbacks to try and collect as many as |
120 | inode = igrab(AFS_VNODE_TO_I(vnode)); | 232 | * possible to ship in one operation */ |
121 | break; | 233 | switch (atomic_inc_return(&server->cb_break_n)) { |
234 | case 1 ... AFSCBMAX - 1: | ||
235 | queue_delayed_work(afs_callback_update_worker, | ||
236 | &server->cb_break_work, HZ * 2); | ||
237 | break; | ||
238 | case AFSCBMAX: | ||
239 | afs_flush_callback_breaks(server); | ||
240 | break; | ||
241 | default: | ||
242 | break; | ||
243 | } | ||
244 | |||
245 | ASSERT(server->cb_promises.rb_node != NULL); | ||
246 | rb_erase(&vnode->cb_promise, &server->cb_promises); | ||
247 | vnode->cb_promised = false; | ||
248 | _leave(""); | ||
249 | } | ||
250 | |||
251 | /* | ||
252 | * give up the callback registered for a vnode on the file server when the | ||
253 | * inode is being cleared | ||
254 | */ | ||
255 | void afs_give_up_callback(struct afs_vnode *vnode) | ||
256 | { | ||
257 | struct afs_server *server = vnode->server; | ||
258 | |||
259 | DECLARE_WAITQUEUE(myself, current); | ||
260 | |||
261 | _enter("%d", vnode->cb_promised); | ||
262 | |||
263 | _debug("GIVE UP INODE %p", &vnode->vfs_inode); | ||
264 | |||
265 | if (!vnode->cb_promised) { | ||
266 | _leave(" [not promised]"); | ||
267 | return; | ||
268 | } | ||
269 | |||
270 | ASSERT(server != NULL); | ||
271 | |||
272 | spin_lock(&server->cb_lock); | ||
273 | if (vnode->cb_promised && afs_breakring_space(server) == 0) { | ||
274 | add_wait_queue(&server->cb_break_waitq, &myself); | ||
275 | for (;;) { | ||
276 | set_current_state(TASK_UNINTERRUPTIBLE); | ||
277 | if (!vnode->cb_promised || | ||
278 | afs_breakring_space(server) != 0) | ||
279 | break; | ||
280 | spin_unlock(&server->cb_lock); | ||
281 | schedule(); | ||
282 | spin_lock(&server->cb_lock); | ||
122 | } | 283 | } |
284 | remove_wait_queue(&server->cb_break_waitq, &myself); | ||
285 | __set_current_state(TASK_RUNNING); | ||
286 | } | ||
287 | |||
288 | /* of course, it's always possible for the server to break this vnode's | ||
289 | * callback first... */ | ||
290 | if (vnode->cb_promised) | ||
291 | afs_do_give_up_callback(server, vnode); | ||
292 | |||
293 | spin_unlock(&server->cb_lock); | ||
294 | _leave(""); | ||
295 | } | ||
296 | |||
297 | /* | ||
298 | * dispatch a deferred give up callbacks operation | ||
299 | */ | ||
300 | void afs_dispatch_give_up_callbacks(struct work_struct *work) | ||
301 | { | ||
302 | struct afs_server *server = | ||
303 | container_of(work, struct afs_server, cb_break_work.work); | ||
304 | |||
305 | _enter(""); | ||
306 | |||
307 | /* tell the fileserver to discard the callback promises it has | ||
308 | * - in the event of ENOMEM or some other error, we just forget that we | ||
309 | * had callbacks entirely, and the server will call us later to break | ||
310 | * them | ||
311 | */ | ||
312 | afs_fs_give_up_callbacks(server, &afs_async_call); | ||
313 | } | ||
314 | |||
315 | /* | ||
316 | * flush the outstanding callback breaks on a server | ||
317 | */ | ||
318 | void afs_flush_callback_breaks(struct afs_server *server) | ||
319 | { | ||
320 | cancel_delayed_work(&server->cb_break_work); | ||
321 | queue_delayed_work(afs_callback_update_worker, | ||
322 | &server->cb_break_work, 0); | ||
323 | } | ||
324 | |||
325 | #if 0 | ||
326 | /* | ||
327 | * update a bunch of callbacks | ||
328 | */ | ||
329 | static void afs_callback_updater(struct work_struct *work) | ||
330 | { | ||
331 | struct afs_server *server; | ||
332 | struct afs_vnode *vnode, *xvnode; | ||
333 | time_t now; | ||
334 | long timeout; | ||
335 | int ret; | ||
336 | |||
337 | server = container_of(work, struct afs_server, updater); | ||
338 | |||
339 | _enter(""); | ||
123 | 340 | ||
124 | spin_unlock(&afs_cb_hash_lock); | 341 | now = get_seconds(); |
125 | 342 | ||
126 | if (inode) { | 343 | /* find the first vnode to update */ |
127 | /* we've found the record for this vnode */ | 344 | spin_lock(&server->cb_lock); |
128 | spin_lock(&vnode->lock); | 345 | for (;;) { |
129 | if (vnode->cb_server == server) { | 346 | if (RB_EMPTY_ROOT(&server->cb_promises)) { |
130 | /* the callback _is_ on the calling server */ | 347 | spin_unlock(&server->cb_lock); |
131 | vnode->cb_server = NULL; | 348 | _leave(" [nothing]"); |
132 | valid = 1; | 349 | return; |
133 | |||
134 | afs_kafstimod_del_timer(&vnode->cb_timeout); | ||
135 | vnode->flags |= AFS_VNODE_CHANGED; | ||
136 | |||
137 | spin_lock(&server->cb_lock); | ||
138 | list_del_init(&vnode->cb_link); | ||
139 | spin_unlock(&server->cb_lock); | ||
140 | |||
141 | spin_lock(&afs_cb_hash_lock); | ||
142 | list_del_init(&vnode->cb_hash_link); | ||
143 | spin_unlock(&afs_cb_hash_lock); | ||
144 | } | ||
145 | spin_unlock(&vnode->lock); | ||
146 | |||
147 | if (valid) { | ||
148 | invalidate_remote_inode(inode); | ||
149 | afs_put_server(server); | ||
150 | } | ||
151 | iput(inode); | ||
152 | } | 350 | } |
351 | |||
352 | vnode = rb_entry(rb_first(&server->cb_promises), | ||
353 | struct afs_vnode, cb_promise); | ||
354 | if (atomic_read(&vnode->usage) > 0) | ||
355 | break; | ||
356 | rb_erase(&vnode->cb_promise, &server->cb_promises); | ||
357 | vnode->cb_promised = false; | ||
358 | } | ||
359 | |||
360 | timeout = vnode->update_at - now; | ||
361 | if (timeout > 0) { | ||
362 | queue_delayed_work(afs_vnode_update_worker, | ||
363 | &afs_vnode_update, timeout * HZ); | ||
364 | spin_unlock(&server->cb_lock); | ||
365 | _leave(" [nothing]"); | ||
366 | return; | ||
367 | } | ||
368 | |||
369 | list_del_init(&vnode->update); | ||
370 | atomic_inc(&vnode->usage); | ||
371 | spin_unlock(&server->cb_lock); | ||
372 | |||
373 | /* we can now perform the update */ | ||
374 | _debug("update %s", vnode->vldb.name); | ||
375 | vnode->state = AFS_VL_UPDATING; | ||
376 | vnode->upd_rej_cnt = 0; | ||
377 | vnode->upd_busy_cnt = 0; | ||
378 | |||
379 | ret = afs_vnode_update_record(vl, &vldb); | ||
380 | switch (ret) { | ||
381 | case 0: | ||
382 | afs_vnode_apply_update(vl, &vldb); | ||
383 | vnode->state = AFS_VL_UPDATING; | ||
384 | break; | ||
385 | case -ENOMEDIUM: | ||
386 | vnode->state = AFS_VL_VOLUME_DELETED; | ||
387 | break; | ||
388 | default: | ||
389 | vnode->state = AFS_VL_UNCERTAIN; | ||
390 | break; | ||
391 | } | ||
392 | |||
393 | /* and then reschedule */ | ||
394 | _debug("reschedule"); | ||
395 | vnode->update_at = get_seconds() + afs_vnode_update_timeout; | ||
396 | |||
397 | spin_lock(&server->cb_lock); | ||
398 | |||
399 | if (!list_empty(&server->cb_promises)) { | ||
400 | /* next update in 10 minutes, but wait at least 1 second more | ||
401 | * than the newest record already queued so that we don't spam | ||
402 | * the VL server suddenly with lots of requests | ||
403 | */ | ||
404 | xvnode = list_entry(server->cb_promises.prev, | ||
405 | struct afs_vnode, update); | ||
406 | if (vnode->update_at <= xvnode->update_at) | ||
407 | vnode->update_at = xvnode->update_at + 1; | ||
408 | xvnode = list_entry(server->cb_promises.next, | ||
409 | struct afs_vnode, update); | ||
410 | timeout = xvnode->update_at - now; | ||
411 | if (timeout < 0) | ||
412 | timeout = 0; | ||
413 | } else { | ||
414 | timeout = afs_vnode_update_timeout; | ||
153 | } | 415 | } |
154 | 416 | ||
155 | _leave(" = 0"); | 417 | list_add_tail(&vnode->update, &server->cb_promises); |
156 | return 0; | 418 | |
419 | _debug("timeout %ld", timeout); | ||
420 | queue_delayed_work(afs_vnode_update_worker, | ||
421 | &afs_vnode_update, timeout * HZ); | ||
422 | spin_unlock(&server->cb_lock); | ||
423 | afs_put_vnode(vl); | ||
424 | } | ||
425 | #endif | ||
426 | |||
427 | /* | ||
428 | * initialise the callback update process | ||
429 | */ | ||
430 | int __init afs_callback_update_init(void) | ||
431 | { | ||
432 | afs_callback_update_worker = | ||
433 | create_singlethread_workqueue("kafs_callbackd"); | ||
434 | return afs_callback_update_worker ? 0 : -ENOMEM; | ||
157 | } | 435 | } |
158 | 436 | ||
159 | /* | 437 | /* |
160 | * allow the fileserver to see if the cache manager is still alive | 438 | * shut down the callback update process |
161 | */ | 439 | */ |
162 | int SRXAFSCM_Probe(struct afs_server *server) | 440 | void __exit afs_callback_update_kill(void) |
163 | { | 441 | { |
164 | _debug("SRXAFSCM_Probe(%p)\n", server); | 442 | destroy_workqueue(afs_callback_update_worker); |
165 | return 0; | ||
166 | } | 443 | } |