server.c

  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 (dhowells@redhat.com)
  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(sizeof(struct afs_server), GFP_KERNEL);
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(rcu_access_pointer(server->endpoint_state));
246		kfree(candidate);
247		afs_dec_servers_outstanding(cell->net);
248	}
249	return server ?: ERR_PTR(ret);
250
251wait_for_creation:
252	afs_see_server(server, afs_server_trace_wait_create);
253	wait_on_bit(&server->flags, AFS_SERVER_FL_CREATING, TASK_UNINTERRUPTIBLE);
254	if (test_bit_acquire(AFS_SERVER_FL_UNCREATED, &server->flags)) {
255		/* Barrier: read flag before error */
256		ret = READ_ONCE(server->create_error);
257		afs_put_server(cell->net, server, afs_server_trace_unuse_create_fail);
258		server = NULL;
259		goto out;
260	}
261
262	ret = 0;
263	goto out;
264
265create_failed:
266	down_write(&cell->fs_lock);
267
268	WRITE_ONCE(server->create_error, ret);
269	smp_wmb(); /* Barrier: set error before flag. */
270	set_bit(AFS_SERVER_FL_UNCREATED, &server->flags);
271
272	clear_and_wake_up_bit(AFS_SERVER_FL_CREATING, &server->flags);
273
274	if (test_bit(AFS_SERVER_FL_UNCREATED, &server->flags)) {
275		clear_bit(AFS_SERVER_FL_UNCREATED, &server->flags);
276		creating = true;
277	}
278	afs_unuse_server(cell->net, server, afs_server_trace_unuse_create_fail);
279	server = NULL;
280
281	up_write(&cell->fs_lock);
282	goto out;
283}
284
285/*
286 * Set/reduce a server's timer.
287 */
288static void afs_set_server_timer(struct afs_server *server, unsigned int delay_secs)
289{
290	mod_timer(&server->timer, jiffies + delay_secs * HZ);
291}
292
293/*
294 * Get a reference on a server object.
295 */
296struct afs_server *afs_get_server(struct afs_server *server,
297				  enum afs_server_trace reason)
298{
299	unsigned int a;
300	int r;
301
302	__refcount_inc(&server->ref, &r);
303	a = atomic_read(&server->active);
304	trace_afs_server(server->debug_id, r + 1, a, reason);
305	return server;
306}
307
308/*
309 * Get an active count on a server object and maybe remove from the inactive
310 * list.
311 */
312struct afs_server *afs_use_server(struct afs_server *server, bool activate,
313				  enum afs_server_trace reason)
314{
315	unsigned int a;
316	int r;
317
318	__refcount_inc(&server->ref, &r);
319	a = atomic_inc_return(&server->active);
320	if (a == 1 && activate &&
321	    !test_bit(AFS_SERVER_FL_EXPIRED, &server->flags))
322		timer_delete(&server->timer);
323
324	trace_afs_server(server->debug_id, r + 1, a, reason);
325	return server;
326}
327
328/*
329 * Release a reference on a server record.
330 */
331void afs_put_server(struct afs_net *net, struct afs_server *server,
332		    enum afs_server_trace reason)
333{
334	unsigned int a, debug_id;
335	bool zero;
336	int r;
337
338	if (!server)
339		return;
340
341	debug_id = server->debug_id;
342	a = atomic_read(&server->active);
343	zero = __refcount_dec_and_test(&server->ref, &r);
344	trace_afs_server(debug_id, r - 1, a, reason);
345	if (unlikely(zero))
346		__afs_put_server(net, server);
347}
348
349/*
350 * Drop an active count on a server object without updating the last-unused
351 * time.
352 */
353void afs_unuse_server_notime(struct afs_net *net, struct afs_server *server,
354			     enum afs_server_trace reason)
355{
356	if (!server)
357		return;
358
359	if (atomic_dec_and_test(&server->active)) {
360		if (test_bit(AFS_SERVER_FL_EXPIRED, &server->flags) ||
361		    READ_ONCE(server->cell->state) >= AFS_CELL_REMOVING)
362			schedule_work(&server->destroyer);
363	}
364
365	afs_put_server(net, server, reason);
366}
367
368/*
369 * Drop an active count on a server object.
370 */
371void afs_unuse_server(struct afs_net *net, struct afs_server *server,
372		      enum afs_server_trace reason)
373{
374	if (!server)
375		return;
376
377	if (atomic_dec_and_test(&server->active)) {
378		if (!test_bit(AFS_SERVER_FL_EXPIRED, &server->flags) &&
379		    READ_ONCE(server->cell->state) < AFS_CELL_REMOVING) {
380			time64_t unuse_time = ktime_get_real_seconds();
381
382			server->unuse_time = unuse_time;
383			afs_set_server_timer(server, afs_server_gc_delay);
384		} else {
385			schedule_work(&server->destroyer);
386		}
387	}
388
389	afs_put_server(net, server, reason);
390}
391
392static void afs_server_rcu(struct rcu_head *rcu)
393{
394	struct afs_server *server = container_of(rcu, struct afs_server, rcu);
395
396	trace_afs_server(server->debug_id, refcount_read(&server->ref),
397			 atomic_read(&server->active), afs_server_trace_free);
398	afs_put_endpoint_state(rcu_access_pointer(server->endpoint_state),
399			       afs_estate_trace_put_server);
400	afs_put_cell(server->cell, afs_cell_trace_put_server);
401	kfree(server->cm_rxgk_appdata.data);
402	kfree(server);
403}
404
405static void __afs_put_server(struct afs_net *net, struct afs_server *server)
406{
407	call_rcu(&server->rcu, afs_server_rcu);
408	afs_dec_servers_outstanding(net);
409}
410
411static void afs_give_up_callbacks(struct afs_net *net, struct afs_server *server)
412{
413	struct afs_endpoint_state *estate = rcu_access_pointer(server->endpoint_state);
414	struct afs_addr_list *alist = estate->addresses;
415
416	afs_fs_give_up_all_callbacks(net, server, &alist->addrs[alist->preferred], NULL);
417}
418
419/*
420 * Check to see if the server record has expired.
421 */
422static bool afs_has_server_expired(const struct afs_server *server)
423{
424	time64_t expires_at;
425
426	if (atomic_read(&server->active))
427		return false;
428
429	if (server->cell->net->live ||
430	    server->cell->state >= AFS_CELL_REMOVING) {
431		trace_afs_server(server->debug_id, refcount_read(&server->ref),
432				 0, afs_server_trace_purging);
433		return true;
434	}
435
436	expires_at = server->unuse_time;
437	if (!test_bit(AFS_SERVER_FL_VL_FAIL, &server->flags) &&
438	    !test_bit(AFS_SERVER_FL_NOT_FOUND, &server->flags))
439		expires_at += afs_server_gc_delay;
440
441	return ktime_get_real_seconds() > expires_at;
442}
443
444/*
445 * Remove a server record from it's parent cell's database.
446 */
447static bool afs_remove_server_from_cell(struct afs_server *server)
448{
449	struct afs_cell *cell = server->cell;
450
451	down_write(&cell->fs_lock);
452
453	if (!afs_has_server_expired(server)) {
454		up_write(&cell->fs_lock);
455		return false;
456	}
457
458	set_bit(AFS_SERVER_FL_EXPIRED, &server->flags);
459	_debug("expire %pU %u", &server->uuid, atomic_read(&server->active));
460	afs_see_server(server, afs_server_trace_see_expired);
461	rb_erase(&server->uuid_rb, &cell->fs_servers);
462	up_write(&cell->fs_lock);
463	return true;
464}
465
466static void afs_server_destroyer(struct work_struct *work)
467{
468	struct afs_endpoint_state *estate;
469	struct afs_server *server = container_of(work, struct afs_server, destroyer);
470	struct afs_net *net = server->cell->net;
471
472	afs_see_server(server, afs_server_trace_see_destroyer);
473
474	if (test_bit(AFS_SERVER_FL_EXPIRED, &server->flags))
475		return;
476
477	if (!afs_remove_server_from_cell(server))
478		return;
479
480	timer_shutdown_sync(&server->timer);
481	cancel_work(&server->destroyer);
482
483	if (test_bit(AFS_SERVER_FL_MAY_HAVE_CB, &server->flags))
484		afs_give_up_callbacks(net, server);
485
486	/* Unbind the rxrpc_peer records from the server. */
487	estate = rcu_access_pointer(server->endpoint_state);
488	if (estate)
489		afs_set_peer_appdata(server, estate->addresses, NULL);
490
491	write_seqlock(&net->fs_lock);
492	list_del_init(&server->probe_link);
493	if (!hlist_unhashed(&server->proc_link))
494		hlist_del_rcu(&server->proc_link);
495	write_sequnlock(&net->fs_lock);
496
497	afs_put_server(net, server, afs_server_trace_destroy);
498}
499
500static void afs_server_timer(struct timer_list *timer)
501{
502	struct afs_server *server = container_of(timer, struct afs_server, timer);
503
504	afs_see_server(server, afs_server_trace_see_timer);
505	if (!test_bit(AFS_SERVER_FL_EXPIRED, &server->flags))
506		schedule_work(&server->destroyer);
507}
508
509/*
510 * Wake up all the servers in a cell so that they can purge themselves.
511 */
512void afs_purge_servers(struct afs_cell *cell)
513{
514	struct afs_server *server;
515	struct rb_node *rb;
516
517	down_read(&cell->fs_lock);
518	for (rb = rb_first(&cell->fs_servers); rb; rb = rb_next(rb)) {
519		server = rb_entry(rb, struct afs_server, uuid_rb);
520		afs_see_server(server, afs_server_trace_see_purge);
521		schedule_work(&server->destroyer);
522	}
523	up_read(&cell->fs_lock);
524}
525
526/*
527 * Wait for outstanding servers.
528 */
529void afs_wait_for_servers(struct afs_net *net)
530{
531	_enter("");
532
533	atomic_dec(&net->servers_outstanding);
534	wait_var_event(&net->servers_outstanding,
535		       !atomic_read(&net->servers_outstanding));
536	_leave("");
537}
538
539/*
540 * Get an update for a server's address list.
541 */
542static noinline bool afs_update_server_record(struct afs_operation *op,
543					      struct afs_server *server,
544					      struct key *key)
545{
546	struct afs_endpoint_state *estate;
547	struct afs_addr_list *alist;
548	bool has_addrs;
549
550	_enter("");
551
552	trace_afs_server(server->debug_id, refcount_read(&server->ref),
553			 atomic_read(&server->active),
554			 afs_server_trace_update);
555
556	alist = afs_vl_lookup_addrs(server, op->key);
557	if (IS_ERR(alist)) {
558		rcu_read_lock();
559		estate = rcu_dereference(server->endpoint_state);
560		has_addrs = estate->addresses;
561		rcu_read_unlock();
562
563		if ((PTR_ERR(alist) == -ERESTARTSYS ||
564		     PTR_ERR(alist) == -EINTR) &&
565		    (op->flags & AFS_OPERATION_UNINTR) &&
566		    has_addrs) {
567			_leave(" = t [intr]");
568			return true;
569		}
570		afs_op_set_error(op, PTR_ERR(alist));
571		_leave(" = f [%d]", afs_op_error(op));
572		return false;
573	}
574
575	if (server->addr_version != alist->version)
576		afs_fs_probe_fileserver(op->net, server, alist, key);
577
578	afs_put_addrlist(alist, afs_alist_trace_put_server_update);
579	_leave(" = t");
580	return true;
581}
582
583/*
584 * See if a server's address list needs updating.
585 */
586bool afs_check_server_record(struct afs_operation *op, struct afs_server *server,
587			     struct key *key)
588{
589	bool success;
590	int ret, retries = 0;
591
592	_enter("");
593
594	ASSERT(server);
595
596retry:
597	if (test_bit(AFS_SERVER_FL_UPDATING, &server->flags))
598		goto wait;
599	if (test_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags))
600		goto update;
601	_leave(" = t [good]");
602	return true;
603
604update:
605	if (!test_and_set_bit_lock(AFS_SERVER_FL_UPDATING, &server->flags)) {
606		clear_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags);
607		success = afs_update_server_record(op, server, key);
608		clear_bit_unlock(AFS_SERVER_FL_UPDATING, &server->flags);
609		wake_up_bit(&server->flags, AFS_SERVER_FL_UPDATING);
610		_leave(" = %d", success);
611		return success;
612	}
613
614wait:
615	ret = wait_on_bit(&server->flags, AFS_SERVER_FL_UPDATING,
616			  (op->flags & AFS_OPERATION_UNINTR) ?
617			  TASK_UNINTERRUPTIBLE : TASK_INTERRUPTIBLE);
618	if (ret == -ERESTARTSYS) {
619		afs_op_set_error(op, ret);
620		_leave(" = f [intr]");
621		return false;
622	}
623
624	retries++;
625	if (retries == 4) {
626		_leave(" = f [stale]");
627		ret = -ESTALE;
628		return false;
629	}
630	goto retry;
631}