projects/linux·blob master · refs

630 lines · 16.1 KiB

1// SPDX-License-Identifier: GPL-2.0-or-later
2/* AFS server record management
3 *
4 * Copyright (C) 2002, 2007 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells ([email protected])
6 */
7
8#include <linux/sched.h>
9#include <linux/slab.h>
10#include "afs_fs.h"
11#include "internal.h"
12#include "protocol_yfs.h"
13
14static unsigned afs_server_gc_delay = 10; /* Server record timeout in seconds */
15static atomic_t afs_server_debug_id;
16
17static void __afs_put_server(struct afs_net *, struct afs_server *);
18static void afs_server_timer(struct timer_list *timer);
19static void afs_server_destroyer(struct work_struct *work);
20
21/*
22 * Find a server by one of its addresses.
23 */
24struct afs_server *afs_find_server(const struct rxrpc_peer *peer)
25{
26 struct afs_server *server = (struct afs_server *)rxrpc_kernel_get_peer_data(peer);
27
28 if (!server)
29 return NULL;
30 return afs_use_server(server, false, afs_server_trace_use_cm_call);
31}
32
33/*
34 * Look up a server by its UUID and mark it active. The caller must hold
35 * cell->fs_lock.
36 */
37static struct afs_server *afs_find_server_by_uuid(struct afs_cell *cell, const uuid_t *uuid)
38{
39 struct afs_server *server;
40 struct rb_node *p;
41 int diff;
42
43 _enter("%pU", uuid);
44
45 p = cell->fs_servers.rb_node;
46 while (p) {
47 server = rb_entry(p, struct afs_server, uuid_rb);
48
49 diff = memcmp(uuid, &server->uuid, sizeof(*uuid));
50 if (diff < 0) {
51 p = p->rb_left;
52 } else if (diff > 0) {
53 p = p->rb_right;
54 } else {
55 if (test_bit(AFS_SERVER_FL_UNCREATED, &server->flags))
56 return NULL; /* Need a write lock */
57 afs_use_server(server, true, afs_server_trace_use_by_uuid);
58 return server;
59 }
60 }
61
62 return NULL;
63}
64
65/*
66 * Install a server record in the cell tree. The caller must hold an exclusive
67 * lock on cell->fs_lock.
68 */
69static struct afs_server *afs_install_server(struct afs_cell *cell,
70 struct afs_server **candidate)
71{
72 struct afs_server *server;
73 struct afs_net *net = cell->net;
74 struct rb_node **pp, *p;
75 int diff;
76
77 _enter("%p", candidate);
78
79 /* Firstly install the server in the UUID lookup tree */
80 pp = &cell->fs_servers.rb_node;
81 p = NULL;
82 while (*pp) {
83 p = *pp;
84 _debug("- consider %p", p);
85 server = rb_entry(p, struct afs_server, uuid_rb);
86 diff = memcmp(&(*candidate)->uuid, &server->uuid, sizeof(uuid_t));
87 if (diff < 0)
88 pp = &(*pp)->rb_left;
89 else if (diff > 0)
90 pp = &(*pp)->rb_right;
91 else
92 goto exists;
93 }
94
95 server = *candidate;
96 *candidate = NULL;
97 rb_link_node(&server->uuid_rb, p, pp);
98 rb_insert_color(&server->uuid_rb, &cell->fs_servers);
99 write_seqlock(&net->fs_lock);
100 hlist_add_head_rcu(&server->proc_link, &net->fs_proc);
101 write_sequnlock(&net->fs_lock);
102
103 afs_get_cell(cell, afs_cell_trace_get_server);
104
105exists:
106 afs_use_server(server, true, afs_server_trace_use_install);
107 return server;
108}
109
110/*
111 * Allocate a new server record and mark it as active but uncreated.
112 */
113static struct afs_server *afs_alloc_server(struct afs_cell *cell, const uuid_t *uuid)
114{
115 struct afs_server *server;
116 struct afs_net *net = cell->net;
117
118 _enter("");
119
120 server = kzalloc_obj(struct afs_server);
121 if (!server)
122 return NULL;
123
124 refcount_set(&server->ref, 1);
125 atomic_set(&server->active, 0);
126 __set_bit(AFS_SERVER_FL_UNCREATED, &server->flags);
127 server->debug_id = atomic_inc_return(&afs_server_debug_id);
128 server->uuid = *uuid;
129 rwlock_init(&server->fs_lock);
130 INIT_WORK(&server->destroyer, &afs_server_destroyer);
131 timer_setup(&server->timer, afs_server_timer, 0);
132 INIT_LIST_HEAD(&server->volumes);
133 init_waitqueue_head(&server->probe_wq);
134 mutex_init(&server->cm_token_lock);
135 INIT_LIST_HEAD(&server->probe_link);
136 INIT_HLIST_NODE(&server->proc_link);
137 spin_lock_init(&server->probe_lock);
138 server->cell = cell;
139 server->rtt = UINT_MAX;
140 server->service_id = FS_SERVICE;
141 server->probe_counter = 1;
142 server->probed_at = jiffies - LONG_MAX / 2;
143
144 afs_inc_servers_outstanding(net);
145 _leave(" = %p", server);
146 return server;
147}
148
149/*
150 * Look up an address record for a server
151 */
152static struct afs_addr_list *afs_vl_lookup_addrs(struct afs_server *server,
153 struct key *key)
154{
155 struct afs_vl_cursor vc;
156 struct afs_addr_list *alist = NULL;
157 int ret;
158
159 ret = -ERESTARTSYS;
160 if (afs_begin_vlserver_operation(&vc, server->cell, key)) {
161 while (afs_select_vlserver(&vc)) {
162 if (test_bit(AFS_VLSERVER_FL_IS_YFS, &vc.server->flags))
163 alist = afs_yfsvl_get_endpoints(&vc, &server->uuid);
164 else
165 alist = afs_vl_get_addrs_u(&vc, &server->uuid);
166 }
167
168 ret = afs_end_vlserver_operation(&vc);
169 }
170
171 return ret < 0 ? ERR_PTR(ret) : alist;
172}
173
174/*
175 * Get or create a fileserver record and return it with an active-use count on
176 * it.
177 */
178struct afs_server *afs_lookup_server(struct afs_cell *cell, struct key *key,
179 const uuid_t *uuid, u32 addr_version)
180{
181 struct afs_addr_list *alist = NULL;
182 struct afs_server *server, *candidate = NULL;
183 bool creating = false;
184 int ret;
185
186 _enter("%p,%pU", cell->net, uuid);
187
188 down_read(&cell->fs_lock);
189 server = afs_find_server_by_uuid(cell, uuid);
190 /* Won't see servers marked uncreated. */
191 up_read(&cell->fs_lock);
192
193 if (server) {
194 timer_delete_sync(&server->timer);
195 if (test_bit(AFS_SERVER_FL_CREATING, &server->flags))
196 goto wait_for_creation;
197 if (server->addr_version != addr_version)
198 set_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags);
199 return server;
200 }
201
202 candidate = afs_alloc_server(cell, uuid);
203 if (!candidate) {
204 afs_put_addrlist(alist, afs_alist_trace_put_server_oom);
205 return ERR_PTR(-ENOMEM);
206 }
207
208 down_write(&cell->fs_lock);
209 server = afs_install_server(cell, &candidate);
210 if (test_bit(AFS_SERVER_FL_CREATING, &server->flags)) {
211 /* We need to wait for creation to complete. */
212 up_write(&cell->fs_lock);
213 goto wait_for_creation;
214 }
215 if (test_bit(AFS_SERVER_FL_UNCREATED, &server->flags)) {
216 set_bit(AFS_SERVER_FL_CREATING, &server->flags);
217 clear_bit(AFS_SERVER_FL_UNCREATED, &server->flags);
218 creating = true;
219 }
220 up_write(&cell->fs_lock);
221 timer_delete_sync(&server->timer);
222
223 /* If we get to create the server, we look up the addresses and then
224 * immediately dispatch an asynchronous probe to each interface on the
225 * fileserver. This will make sure the repeat-probing service is
226 * started.
227 */
228 if (creating) {
229 alist = afs_vl_lookup_addrs(server, key);
230 if (IS_ERR(alist)) {
231 ret = PTR_ERR(alist);
232 goto create_failed;
233 }
234
235 ret = afs_fs_probe_fileserver(cell->net, server, alist, key);
236 if (ret)
237 goto create_failed;
238
239 clear_and_wake_up_bit(AFS_SERVER_FL_CREATING, &server->flags);
240 }
241
242out:
243 afs_put_addrlist(alist, afs_alist_trace_put_server_create);
244 if (candidate) {
245 kfree(candidate);
246 afs_dec_servers_outstanding(cell->net);
247 }
248 return server ?: ERR_PTR(ret);
249
250wait_for_creation:
251 afs_see_server(server, afs_server_trace_wait_create);
252 wait_on_bit(&server->flags, AFS_SERVER_FL_CREATING, TASK_UNINTERRUPTIBLE);
253 if (test_bit_acquire(AFS_SERVER_FL_UNCREATED, &server->flags)) {
254 /* Barrier: read flag before error */
255 ret = READ_ONCE(server->create_error);
256 afs_put_server(cell->net, server, afs_server_trace_unuse_create_fail);
257 server = NULL;
258 goto out;
259 }
260
261 ret = 0;
262 goto out;
263
264create_failed:
265 down_write(&cell->fs_lock);
266
267 WRITE_ONCE(server->create_error, ret);
268 smp_wmb(); /* Barrier: set error before flag. */
269 set_bit(AFS_SERVER_FL_UNCREATED, &server->flags);
270
271 clear_and_wake_up_bit(AFS_SERVER_FL_CREATING, &server->flags);
272
273 if (test_bit(AFS_SERVER_FL_UNCREATED, &server->flags)) {
274 clear_bit(AFS_SERVER_FL_UNCREATED, &server->flags);
275 creating = true;
276 }
277 afs_unuse_server(cell->net, server, afs_server_trace_unuse_create_fail);
278 server = NULL;
279
280 up_write(&cell->fs_lock);
281 goto out;
282}
283
284/*
285 * Set/reduce a server's timer.
286 */
287static void afs_set_server_timer(struct afs_server *server, unsigned int delay_secs)
288{
289 mod_timer(&server->timer, jiffies + delay_secs * HZ);
290}
291
292/*
293 * Get a reference on a server object.
294 */
295struct afs_server *afs_get_server(struct afs_server *server,
296 enum afs_server_trace reason)
297{
298 unsigned int a;
299 int r;
300
301 __refcount_inc(&server->ref, &r);
302 a = atomic_read(&server->active);
303 trace_afs_server(server->debug_id, r + 1, a, reason);
304 return server;
305}
306
307/*
308 * Get an active count on a server object and maybe remove from the inactive
309 * list.
310 */
311struct afs_server *afs_use_server(struct afs_server *server, bool activate,
312 enum afs_server_trace reason)
313{
314 unsigned int a;
315 int r;
316
317 __refcount_inc(&server->ref, &r);
318 a = atomic_inc_return(&server->active);
319 if (a == 1 && activate &&
320 !test_bit(AFS_SERVER_FL_EXPIRED, &server->flags))
321 timer_delete(&server->timer);
322
323 trace_afs_server(server->debug_id, r + 1, a, reason);
324 return server;
325}
326
327/*
328 * Release a reference on a server record.
329 */
330void afs_put_server(struct afs_net *net, struct afs_server *server,
331 enum afs_server_trace reason)
332{
333 unsigned int a, debug_id;
334 bool zero;
335 int r;
336
337 if (!server)
338 return;
339
340 debug_id = server->debug_id;
341 a = atomic_read(&server->active);
342 zero = __refcount_dec_and_test(&server->ref, &r);
343 trace_afs_server(debug_id, r - 1, a, reason);
344 if (unlikely(zero))
345 __afs_put_server(net, server);
346}
347
348/*
349 * Drop an active count on a server object without updating the last-unused
350 * time.
351 */
352void afs_unuse_server_notime(struct afs_net *net, struct afs_server *server,
353 enum afs_server_trace reason)
354{
355 if (!server)
356 return;
357
358 if (atomic_dec_and_test(&server->active)) {
359 if (test_bit(AFS_SERVER_FL_EXPIRED, &server->flags) ||
360 READ_ONCE(server->cell->state) >= AFS_CELL_REMOVING)
361 schedule_work(&server->destroyer);
362 }
363
364 afs_put_server(net, server, reason);
365}
366
367/*
368 * Drop an active count on a server object.
369 */
370void afs_unuse_server(struct afs_net *net, struct afs_server *server,
371 enum afs_server_trace reason)
372{
373 if (!server)
374 return;
375
376 if (atomic_dec_and_test(&server->active)) {
377 if (!test_bit(AFS_SERVER_FL_EXPIRED, &server->flags) &&
378 READ_ONCE(server->cell->state) < AFS_CELL_REMOVING) {
379 time64_t unuse_time = ktime_get_real_seconds();
380
381 server->unuse_time = unuse_time;
382 afs_set_server_timer(server, afs_server_gc_delay);
383 } else {
384 schedule_work(&server->destroyer);
385 }
386 }
387
388 afs_put_server(net, server, reason);
389}
390
391static void afs_server_rcu(struct rcu_head *rcu)
392{
393 struct afs_server *server = container_of(rcu, struct afs_server, rcu);
394
395 trace_afs_server(server->debug_id, refcount_read(&server->ref),
396 atomic_read(&server->active), afs_server_trace_free);
397 afs_put_endpoint_state(rcu_access_pointer(server->endpoint_state),
398 afs_estate_trace_put_server);
399 afs_put_cell(server->cell, afs_cell_trace_put_server);
400 kfree(server->cm_rxgk_appdata.data);
401 kfree(server);
402}
403
404static void __afs_put_server(struct afs_net *net, struct afs_server *server)
405{
406 call_rcu(&server->rcu, afs_server_rcu);
407 afs_dec_servers_outstanding(net);
408}
409
410static void afs_give_up_callbacks(struct afs_net *net, struct afs_server *server)
411{
412 struct afs_endpoint_state *estate = rcu_access_pointer(server->endpoint_state);
413 struct afs_addr_list *alist = estate->addresses;
414
415 afs_fs_give_up_all_callbacks(net, server, &alist->addrs[alist->preferred], NULL);
416}
417
418/*
419 * Check to see if the server record has expired.
420 */
421static bool afs_has_server_expired(const struct afs_server *server)
422{
423 time64_t expires_at;
424
425 if (atomic_read(&server->active))
426 return false;
427
428 if (server->cell->net->live ||
429 server->cell->state >= AFS_CELL_REMOVING) {
430 trace_afs_server(server->debug_id, refcount_read(&server->ref),
431 0, afs_server_trace_purging);
432 return true;
433 }
434
435 expires_at = server->unuse_time;
436 if (!test_bit(AFS_SERVER_FL_VL_FAIL, &server->flags) &&
437 !test_bit(AFS_SERVER_FL_NOT_FOUND, &server->flags))
438 expires_at += afs_server_gc_delay;
439
440 return ktime_get_real_seconds() > expires_at;
441}
442
443/*
444 * Remove a server record from it's parent cell's database.
445 */
446static bool afs_remove_server_from_cell(struct afs_server *server)
447{
448 struct afs_cell *cell = server->cell;
449
450 down_write(&cell->fs_lock);
451
452 if (!afs_has_server_expired(server)) {
453 up_write(&cell->fs_lock);
454 return false;
455 }
456
457 set_bit(AFS_SERVER_FL_EXPIRED, &server->flags);
458 _debug("expire %pU %u", &server->uuid, atomic_read(&server->active));
459 afs_see_server(server, afs_server_trace_see_expired);
460 rb_erase(&server->uuid_rb, &cell->fs_servers);
461 up_write(&cell->fs_lock);
462 return true;
463}
464
465static void afs_server_destroyer(struct work_struct *work)
466{
467 struct afs_endpoint_state *estate;
468 struct afs_server *server = container_of(work, struct afs_server, destroyer);
469 struct afs_net *net = server->cell->net;
470
471 afs_see_server(server, afs_server_trace_see_destroyer);
472
473 if (test_bit(AFS_SERVER_FL_EXPIRED, &server->flags))
474 return;
475
476 if (!afs_remove_server_from_cell(server))
477 return;
478
479 timer_shutdown_sync(&server->timer);
480 cancel_work(&server->destroyer);
481
482 if (test_bit(AFS_SERVER_FL_MAY_HAVE_CB, &server->flags))
483 afs_give_up_callbacks(net, server);
484
485 /* Unbind the rxrpc_peer records from the server. */
486 estate = rcu_access_pointer(server->endpoint_state);
487 if (estate)
488 afs_set_peer_appdata(server, estate->addresses, NULL);
489
490 write_seqlock(&net->fs_lock);
491 list_del_init(&server->probe_link);
492 if (!hlist_unhashed(&server->proc_link))
493 hlist_del_rcu(&server->proc_link);
494 write_sequnlock(&net->fs_lock);
495
496 afs_put_server(net, server, afs_server_trace_destroy);
497}
498
499static void afs_server_timer(struct timer_list *timer)
500{
501 struct afs_server *server = container_of(timer, struct afs_server, timer);
502
503 afs_see_server(server, afs_server_trace_see_timer);
504 if (!test_bit(AFS_SERVER_FL_EXPIRED, &server->flags))
505 schedule_work(&server->destroyer);
506}
507
508/*
509 * Wake up all the servers in a cell so that they can purge themselves.
510 */
511void afs_purge_servers(struct afs_cell *cell)
512{
513 struct afs_server *server;
514 struct rb_node *rb;
515
516 down_read(&cell->fs_lock);
517 for (rb = rb_first(&cell->fs_servers); rb; rb = rb_next(rb)) {
518 server = rb_entry(rb, struct afs_server, uuid_rb);
519 afs_see_server(server, afs_server_trace_see_purge);
520 schedule_work(&server->destroyer);
521 }
522 up_read(&cell->fs_lock);
523}
524
525/*
526 * Wait for outstanding servers.
527 */
528void afs_wait_for_servers(struct afs_net *net)
529{
530 _enter("");
531
532 atomic_dec(&net->servers_outstanding);
533 wait_var_event(&net->servers_outstanding,
534 !atomic_read(&net->servers_outstanding));
535 _leave("");
536}
537
538/*
539 * Get an update for a server's address list.
540 */
541static noinline bool afs_update_server_record(struct afs_operation *op,
542 struct afs_server *server,
543 struct key *key)
544{
545 struct afs_endpoint_state *estate;
546 struct afs_addr_list *alist;
547 bool has_addrs;
548
549 _enter("");
550
551 trace_afs_server(server->debug_id, refcount_read(&server->ref),
552 atomic_read(&server->active),
553 afs_server_trace_update);
554
555 alist = afs_vl_lookup_addrs(server, op->key);
556 if (IS_ERR(alist)) {
557 rcu_read_lock();
558 estate = rcu_dereference(server->endpoint_state);
559 has_addrs = estate->addresses;
560 rcu_read_unlock();
561
562 if ((PTR_ERR(alist) == -ERESTARTSYS ||
563 PTR_ERR(alist) == -EINTR) &&
564 (op->flags & AFS_OPERATION_UNINTR) &&
565 has_addrs) {
566 _leave(" = t [intr]");
567 return true;
568 }
569 afs_op_set_error(op, PTR_ERR(alist));
570 _leave(" = f [%d]", afs_op_error(op));
571 return false;
572 }
573
574 if (server->addr_version != alist->version)
575 afs_fs_probe_fileserver(op->net, server, alist, key);
576
577 afs_put_addrlist(alist, afs_alist_trace_put_server_update);
578 _leave(" = t");
579 return true;
580}
581
582/*
583 * See if a server's address list needs updating.
584 */
585bool afs_check_server_record(struct afs_operation *op, struct afs_server *server,
586 struct key *key)
587{
588 bool success;
589 int ret, retries = 0;
590
591 _enter("");
592
593 ASSERT(server);
594
595retry:
596 if (test_bit(AFS_SERVER_FL_UPDATING, &server->flags))
597 goto wait;
598 if (test_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags))
599 goto update;
600 _leave(" = t [good]");
601 return true;
602
603update:
604 if (!test_and_set_bit_lock(AFS_SERVER_FL_UPDATING, &server->flags)) {
605 clear_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags);
606 success = afs_update_server_record(op, server, key);
607 clear_bit_unlock(AFS_SERVER_FL_UPDATING, &server->flags);
608 wake_up_bit(&server->flags, AFS_SERVER_FL_UPDATING);
609 _leave(" = %d", success);
610 return success;
611 }
612
613wait:
614 ret = wait_on_bit(&server->flags, AFS_SERVER_FL_UPDATING,
615 (op->flags & AFS_OPERATION_UNINTR) ?
616 TASK_UNINTERRUPTIBLE : TASK_INTERRUPTIBLE);
617 if (ret == -ERESTARTSYS) {
618 afs_op_set_error(op, ret);
619 _leave(" = f [intr]");
620 return false;
621 }
622
623 retries++;
624 if (retries == 4) {
625 _leave(" = f [stale]");
626 ret = -ESTALE;
627 return false;
628 }
629 goto retry;
630}