1// SPDX-License-Identifier: GPL-2.0-or-later
  2/* Handle vlserver selection and rotation.
  3 *
  4 * Copyright (C) 2018 Red Hat, Inc. All Rights Reserved.
  5 * Written by David Howells (dhowells@redhat.com)
  6 */
  7
  8#include <linux/kernel.h>
  9#include <linux/sched.h>
 10#include <linux/sched/signal.h>
 11#include "internal.h"
 12#include "afs_vl.h"
 13
 14/*
 15 * Begin an operation on a volume location server.
 16 */
 17bool afs_begin_vlserver_operation(struct afs_vl_cursor *vc, struct afs_cell *cell,
 18				  struct key *key)
 19{
 20	memset(vc, 0, sizeof(*vc));
 21	vc->cell = cell;
 22	vc->key = key;
 23	vc->error = -EDESTADDRREQ;
 24	vc->ac.error = SHRT_MAX;
 25
 26	if (signal_pending(current)) {
 27		vc->error = -EINTR;
 28		vc->flags |= AFS_VL_CURSOR_STOP;
 29		return false;
 30	}
 31
 32	return true;
 33}
 34
 35/*
 36 * Begin iteration through a server list, starting with the last used server if
 37 * possible, or the last recorded good server if not.
 38 */
 39static bool afs_start_vl_iteration(struct afs_vl_cursor *vc)
 40{
 41	struct afs_cell *cell = vc->cell;
 42	unsigned int dns_lookup_count;
 43
 44	if (cell->dns_source == DNS_RECORD_UNAVAILABLE ||
 45	    cell->dns_expiry <= ktime_get_real_seconds()) {
 46		dns_lookup_count = smp_load_acquire(&cell->dns_lookup_count);
 47		set_bit(AFS_CELL_FL_DO_LOOKUP, &cell->flags);
 48		queue_work(afs_wq, &cell->manager);
 49
 50		if (cell->dns_source == DNS_RECORD_UNAVAILABLE) {
 51			if (wait_var_event_interruptible(
 52				    &cell->dns_lookup_count,
 53				    smp_load_acquire(&cell->dns_lookup_count)
 54				    != dns_lookup_count) < 0) {
 55				vc->error = -ERESTARTSYS;
 56				return false;
 57			}
 58		}
 59
 60		/* Status load is ordered after lookup counter load */
 61		if (cell->dns_source == DNS_RECORD_UNAVAILABLE) {
 62			vc->error = -EDESTADDRREQ;
 63			return false;
 64		}
 65	}
 66
 67	read_lock(&cell->vl_servers_lock);
 68	vc->server_list = afs_get_vlserverlist(
 69		rcu_dereference_protected(cell->vl_servers,
 70					  lockdep_is_held(&cell->vl_servers_lock)));
 71	read_unlock(&cell->vl_servers_lock);
 72	if (!vc->server_list->nr_servers)
 73		return false;
 74
 75	vc->untried = (1UL << vc->server_list->nr_servers) - 1;
 76	vc->index = -1;
 77	return true;
 78}
 79
 80/*
 81 * Select the vlserver to use.  May be called multiple times to rotate
 82 * through the vlservers.
 83 */
 84bool afs_select_vlserver(struct afs_vl_cursor *vc)
 85{
 86	struct afs_addr_list *alist;
 87	struct afs_vlserver *vlserver;
 88	struct afs_error e;
 89	u32 rtt;
 90	int error = vc->ac.error, i;
 91
 92	_enter("%lx[%d],%lx[%d],%d,%d",
 93	       vc->untried, vc->index,
 94	       vc->ac.tried, vc->ac.index,
 95	       error, vc->ac.abort_code);
 96
 97	if (vc->flags & AFS_VL_CURSOR_STOP) {
 98		_leave(" = f [stopped]");
 99		return false;
100	}
101
102	vc->nr_iterations++;
103
104	/* Evaluate the result of the previous operation, if there was one. */
105	switch (error) {
106	case SHRT_MAX:
107		goto start;
108
109	default:
110	case 0:
111		/* Success or local failure.  Stop. */
112		vc->error = error;
113		vc->flags |= AFS_VL_CURSOR_STOP;
114		_leave(" = f [okay/local %d]", vc->ac.error);
115		return false;
116
117	case -ECONNABORTED:
118		/* The far side rejected the operation on some grounds.  This
119		 * might involve the server being busy or the volume having been moved.
120		 */
121		switch (vc->ac.abort_code) {
122		case AFSVL_IO:
123		case AFSVL_BADVOLOPER:
124		case AFSVL_NOMEM:
125			/* The server went weird. */
126			vc->error = -EREMOTEIO;
127			//write_lock(&vc->cell->vl_servers_lock);
128			//vc->server_list->weird_mask |= 1 << vc->index;
129			//write_unlock(&vc->cell->vl_servers_lock);
130			goto next_server;
131
132		default:
133			vc->error = afs_abort_to_error(vc->ac.abort_code);
134			goto failed;
135		}
136
137	case -ERFKILL:
138	case -EADDRNOTAVAIL:
139	case -ENETUNREACH:
140	case -EHOSTUNREACH:
141	case -EHOSTDOWN:
142	case -ECONNREFUSED:
143	case -ETIMEDOUT:
144	case -ETIME:
145		_debug("no conn %d", error);
146		vc->error = error;
147		goto iterate_address;
148
149	case -ECONNRESET:
150		_debug("call reset");
151		vc->error = error;
152		vc->flags |= AFS_VL_CURSOR_RETRY;
153		goto next_server;
154	}
155
156restart_from_beginning:
157	_debug("restart");
158	afs_end_cursor(&vc->ac);
159	afs_put_vlserverlist(vc->cell->net, vc->server_list);
160	vc->server_list = NULL;
161	if (vc->flags & AFS_VL_CURSOR_RETRIED)
162		goto failed;
163	vc->flags |= AFS_VL_CURSOR_RETRIED;
164start:
165	_debug("start");
166
167	if (!afs_start_vl_iteration(vc))
168		goto failed;
169
170	error = afs_send_vl_probes(vc->cell->net, vc->key, vc->server_list);
171	if (error < 0)
172		goto failed_set_error;
173
174pick_server:
175	_debug("pick [%lx]", vc->untried);
176
177	error = afs_wait_for_vl_probes(vc->server_list, vc->untried);
178	if (error < 0)
179		goto failed_set_error;
180
181	/* Pick the untried server with the lowest RTT. */
182	vc->index = vc->server_list->preferred;
183	if (test_bit(vc->index, &vc->untried))
184		goto selected_server;
185
186	vc->index = -1;
187	rtt = U32_MAX;
188	for (i = 0; i < vc->server_list->nr_servers; i++) {
189		struct afs_vlserver *s = vc->server_list->servers[i].server;
190
191		if (!test_bit(i, &vc->untried) || !s->probe.responded)
192			continue;
193		if (s->probe.rtt < rtt) {
194			vc->index = i;
195			rtt = s->probe.rtt;
196		}
197	}
198
199	if (vc->index == -1)
200		goto no_more_servers;
201
202selected_server:
203	_debug("use %d", vc->index);
204	__clear_bit(vc->index, &vc->untried);
205
206	/* We're starting on a different vlserver from the list.  We need to
207	 * check it, find its address list and probe its capabilities before we
208	 * use it.
209	 */
210	ASSERTCMP(vc->ac.alist, ==, NULL);
211	vlserver = vc->server_list->servers[vc->index].server;
212	vc->server = vlserver;
213
214	_debug("USING VLSERVER: %s", vlserver->name);
215
216	read_lock(&vlserver->lock);
217	alist = rcu_dereference_protected(vlserver->addresses,
218					  lockdep_is_held(&vlserver->lock));
219	afs_get_addrlist(alist);
220	read_unlock(&vlserver->lock);
221
222	memset(&vc->ac, 0, sizeof(vc->ac));
223
224	if (!vc->ac.alist)
225		vc->ac.alist = alist;
226	else
227		afs_put_addrlist(alist);
228
229	vc->ac.index = -1;
230
231iterate_address:
232	ASSERT(vc->ac.alist);
233	/* Iterate over the current server's address list to try and find an
234	 * address on which it will respond to us.
235	 */
236	if (!afs_iterate_addresses(&vc->ac))
237		goto next_server;
238
239	_debug("VL address %d/%d", vc->ac.index, vc->ac.alist->nr_addrs);
240
241	_leave(" = t %pISpc", &vc->ac.alist->addrs[vc->ac.index].transport);
242	return true;
243
244next_server:
245	_debug("next");
246	afs_end_cursor(&vc->ac);
247	goto pick_server;
248
249no_more_servers:
250	/* That's all the servers poked to no good effect.  Try again if some
251	 * of them were busy.
252	 */
253	if (vc->flags & AFS_VL_CURSOR_RETRY)
254		goto restart_from_beginning;
255
256	e.error = -EDESTADDRREQ;
257	e.responded = false;
258	for (i = 0; i < vc->server_list->nr_servers; i++) {
259		struct afs_vlserver *s = vc->server_list->servers[i].server;
260
261		afs_prioritise_error(&e, READ_ONCE(s->probe.error),
262				     s->probe.abort_code);
263	}
264
265failed_set_error:
266	vc->error = error;
267failed:
268	vc->flags |= AFS_VL_CURSOR_STOP;
269	afs_end_cursor(&vc->ac);
270	_leave(" = f [failed %d]", vc->error);
271	return false;
272}
273
274/*
275 * Dump cursor state in the case of the error being EDESTADDRREQ.
276 */
277static void afs_vl_dump_edestaddrreq(const struct afs_vl_cursor *vc)
278{
279	static int count;
280	int i;
281
282	if (!IS_ENABLED(CONFIG_AFS_DEBUG_CURSOR) || count > 3)
283		return;
284	count++;
285
286	rcu_read_lock();
287	pr_notice("EDESTADDR occurred\n");
288	pr_notice("VC: ut=%lx ix=%u ni=%hu fl=%hx err=%hd\n",
289		  vc->untried, vc->index, vc->nr_iterations, vc->flags, vc->error);
290
291	if (vc->server_list) {
292		const struct afs_vlserver_list *sl = vc->server_list;
293		pr_notice("VC: SL nr=%u ix=%u\n",
294			  sl->nr_servers, sl->index);
295		for (i = 0; i < sl->nr_servers; i++) {
296			const struct afs_vlserver *s = sl->servers[i].server;
297			pr_notice("VC: server %s+%hu fl=%lx E=%hd\n",
298				  s->name, s->port, s->flags, s->probe.error);
299			if (s->addresses) {
300				const struct afs_addr_list *a =
301					rcu_dereference(s->addresses);
302				pr_notice("VC:  - nr=%u/%u/%u pf=%u\n",
303					  a->nr_ipv4, a->nr_addrs, a->max_addrs,
304					  a->preferred);
305				pr_notice("VC:  - pr=%lx R=%lx F=%lx\n",
306					  a->probed, a->responded, a->failed);
307				if (a == vc->ac.alist)
308					pr_notice("VC:  - current\n");
309			}
310		}
311	}
312
313	pr_notice("AC: t=%lx ax=%u ac=%d er=%d r=%u ni=%u\n",
314		  vc->ac.tried, vc->ac.index, vc->ac.abort_code, vc->ac.error,
315		  vc->ac.responded, vc->ac.nr_iterations);
316	rcu_read_unlock();
317}
318
319/*
320 * Tidy up a volume location server cursor and unlock the vnode.
321 */
322int afs_end_vlserver_operation(struct afs_vl_cursor *vc)
323{
324	struct afs_net *net = vc->cell->net;
325
326	if (vc->error == -EDESTADDRREQ ||
327	    vc->error == -EADDRNOTAVAIL ||
328	    vc->error == -ENETUNREACH ||
329	    vc->error == -EHOSTUNREACH)
330		afs_vl_dump_edestaddrreq(vc);
331
332	afs_end_cursor(&vc->ac);
333	afs_put_vlserverlist(net, vc->server_list);
334
335	if (vc->error == -ECONNABORTED)
336		vc->error = afs_abort_to_error(vc->ac.abort_code);
337
338	return vc->error;
339}