xref: /linux/fs/afs/vl_probe.c (revision 4b660dbd9ee2059850fd30e0df420ca7a38a1856)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* AFS vlserver probing
3  *
4  * Copyright (C) 2018 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 
14 
15 /*
16  * Handle the completion of a set of probes.
17  */
18 static void afs_finished_vl_probe(struct afs_vlserver *server)
19 {
20 	if (!(server->probe.flags & AFS_VLSERVER_PROBE_RESPONDED)) {
21 		server->rtt = UINT_MAX;
22 		clear_bit(AFS_VLSERVER_FL_RESPONDING, &server->flags);
23 	}
24 
25 	clear_bit_unlock(AFS_VLSERVER_FL_PROBING, &server->flags);
26 	wake_up_bit(&server->flags, AFS_VLSERVER_FL_PROBING);
27 }
28 
29 /*
30  * Handle the completion of a probe RPC call.
31  */
32 static void afs_done_one_vl_probe(struct afs_vlserver *server, bool wake_up)
33 {
34 	if (atomic_dec_and_test(&server->probe_outstanding)) {
35 		afs_finished_vl_probe(server);
36 		wake_up = true;
37 	}
38 
39 	if (wake_up)
40 		wake_up_all(&server->probe_wq);
41 }
42 
43 /*
44  * Process the result of probing a vlserver.  This is called after successful
45  * or failed delivery of an VL.GetCapabilities operation.
46  */
47 void afs_vlserver_probe_result(struct afs_call *call)
48 {
49 	struct afs_addr_list *alist = call->vl_probe;
50 	struct afs_vlserver *server = call->vlserver;
51 	struct afs_address *addr = &alist->addrs[call->probe_index];
52 	unsigned int server_index = call->server_index;
53 	unsigned int rtt_us = 0;
54 	unsigned int index = call->probe_index;
55 	bool have_result = false;
56 	int ret = call->error;
57 
58 	_enter("%s,%u,%u,%d,%d", server->name, server_index, index, ret, call->abort_code);
59 
60 	spin_lock(&server->probe_lock);
61 
62 	switch (ret) {
63 	case 0:
64 		server->probe.error = 0;
65 		goto responded;
66 	case -ECONNABORTED:
67 		if (!(server->probe.flags & AFS_VLSERVER_PROBE_RESPONDED)) {
68 			server->probe.abort_code = call->abort_code;
69 			server->probe.error = ret;
70 		}
71 		goto responded;
72 	case -ENOMEM:
73 	case -ENONET:
74 	case -EKEYEXPIRED:
75 	case -EKEYREVOKED:
76 	case -EKEYREJECTED:
77 		server->probe.flags |= AFS_VLSERVER_PROBE_LOCAL_FAILURE;
78 		if (server->probe.error == 0)
79 			server->probe.error = ret;
80 		trace_afs_io_error(call->debug_id, ret, afs_io_error_vl_probe_fail);
81 		goto out;
82 	case -ECONNRESET: /* Responded, but call expired. */
83 	case -ERFKILL:
84 	case -EADDRNOTAVAIL:
85 	case -ENETUNREACH:
86 	case -EHOSTUNREACH:
87 	case -EHOSTDOWN:
88 	case -ECONNREFUSED:
89 	case -ETIMEDOUT:
90 	case -ETIME:
91 	default:
92 		clear_bit(index, &alist->responded);
93 		set_bit(index, &alist->probe_failed);
94 		if (!(server->probe.flags & AFS_VLSERVER_PROBE_RESPONDED) &&
95 		    (server->probe.error == 0 ||
96 		     server->probe.error == -ETIMEDOUT ||
97 		     server->probe.error == -ETIME))
98 			server->probe.error = ret;
99 		trace_afs_io_error(call->debug_id, ret, afs_io_error_vl_probe_fail);
100 		goto out;
101 	}
102 
103 responded:
104 	set_bit(index, &alist->responded);
105 	clear_bit(index, &alist->probe_failed);
106 
107 	if (call->service_id == YFS_VL_SERVICE) {
108 		server->probe.flags |= AFS_VLSERVER_PROBE_IS_YFS;
109 		set_bit(AFS_VLSERVER_FL_IS_YFS, &server->flags);
110 		server->service_id = call->service_id;
111 	} else {
112 		server->probe.flags |= AFS_VLSERVER_PROBE_NOT_YFS;
113 		if (!(server->probe.flags & AFS_VLSERVER_PROBE_IS_YFS)) {
114 			clear_bit(AFS_VLSERVER_FL_IS_YFS, &server->flags);
115 			server->service_id = call->service_id;
116 		}
117 	}
118 
119 	rtt_us = rxrpc_kernel_get_srtt(addr->peer);
120 	if (rtt_us < server->probe.rtt) {
121 		server->probe.rtt = rtt_us;
122 		server->rtt = rtt_us;
123 		alist->preferred = index;
124 	}
125 
126 	smp_wmb(); /* Set rtt before responded. */
127 	server->probe.flags |= AFS_VLSERVER_PROBE_RESPONDED;
128 	set_bit(AFS_VLSERVER_FL_PROBED, &server->flags);
129 	set_bit(AFS_VLSERVER_FL_RESPONDING, &server->flags);
130 	have_result = true;
131 out:
132 	spin_unlock(&server->probe_lock);
133 
134 	trace_afs_vl_probe(server, false, alist, index, call->error, call->abort_code, rtt_us);
135 	_debug("probe [%u][%u] %pISpc rtt=%d ret=%d",
136 	       server_index, index, rxrpc_kernel_remote_addr(addr->peer),
137 	       rtt_us, ret);
138 
139 	afs_done_one_vl_probe(server, have_result);
140 }
141 
142 /*
143  * Probe all of a vlserver's addresses to find out the best route and to
144  * query its capabilities.
145  */
146 static bool afs_do_probe_vlserver(struct afs_net *net,
147 				  struct afs_vlserver *server,
148 				  struct key *key,
149 				  unsigned int server_index,
150 				  struct afs_error *_e)
151 {
152 	struct afs_addr_list *alist;
153 	struct afs_call *call;
154 	unsigned long unprobed;
155 	unsigned int index, i;
156 	bool in_progress = false;
157 	int best_prio;
158 
159 	_enter("%s", server->name);
160 
161 	read_lock(&server->lock);
162 	alist = rcu_dereference_protected(server->addresses,
163 					  lockdep_is_held(&server->lock));
164 	afs_get_addrlist(alist, afs_alist_trace_get_vlprobe);
165 	read_unlock(&server->lock);
166 
167 	atomic_set(&server->probe_outstanding, alist->nr_addrs);
168 	memset(&server->probe, 0, sizeof(server->probe));
169 	server->probe.rtt = UINT_MAX;
170 
171 	unprobed = (1UL << alist->nr_addrs) - 1;
172 	while (unprobed) {
173 		best_prio = -1;
174 		index = 0;
175 		for (i = 0; i < alist->nr_addrs; i++) {
176 			if (test_bit(i, &unprobed) &&
177 			    alist->addrs[i].prio > best_prio) {
178 				index = i;
179 				best_prio = alist->addrs[i].prio;
180 			}
181 		}
182 		__clear_bit(index, &unprobed);
183 
184 		trace_afs_vl_probe(server, true, alist, index, 0, 0, 0);
185 		call = afs_vl_get_capabilities(net, alist, index, key, server,
186 					       server_index);
187 		if (!IS_ERR(call)) {
188 			afs_prioritise_error(_e, call->error, call->abort_code);
189 			afs_put_call(call);
190 			in_progress = true;
191 		} else {
192 			afs_prioritise_error(_e, PTR_ERR(call), 0);
193 			afs_done_one_vl_probe(server, false);
194 		}
195 	}
196 
197 	afs_put_addrlist(alist, afs_alist_trace_put_vlprobe);
198 	return in_progress;
199 }
200 
201 /*
202  * Send off probes to all unprobed servers.
203  */
204 int afs_send_vl_probes(struct afs_net *net, struct key *key,
205 		       struct afs_vlserver_list *vllist)
206 {
207 	struct afs_vlserver *server;
208 	struct afs_error e = {};
209 	bool in_progress = false;
210 	int i;
211 
212 	for (i = 0; i < vllist->nr_servers; i++) {
213 		server = vllist->servers[i].server;
214 		if (test_bit(AFS_VLSERVER_FL_PROBED, &server->flags))
215 			continue;
216 
217 		if (!test_and_set_bit_lock(AFS_VLSERVER_FL_PROBING, &server->flags) &&
218 		    afs_do_probe_vlserver(net, server, key, i, &e))
219 			in_progress = true;
220 	}
221 
222 	return in_progress ? 0 : e.error;
223 }
224 
225 /*
226  * Wait for the first as-yet untried server to respond.
227  */
228 int afs_wait_for_vl_probes(struct afs_vlserver_list *vllist,
229 			   unsigned long untried)
230 {
231 	struct wait_queue_entry *waits;
232 	struct afs_vlserver *server;
233 	unsigned int rtt = UINT_MAX, rtt_s;
234 	bool have_responders = false;
235 	int pref = -1, i;
236 
237 	_enter("%u,%lx", vllist->nr_servers, untried);
238 
239 	/* Only wait for servers that have a probe outstanding. */
240 	for (i = 0; i < vllist->nr_servers; i++) {
241 		if (test_bit(i, &untried)) {
242 			server = vllist->servers[i].server;
243 			if (!test_bit(AFS_VLSERVER_FL_PROBING, &server->flags))
244 				__clear_bit(i, &untried);
245 			if (server->probe.flags & AFS_VLSERVER_PROBE_RESPONDED)
246 				have_responders = true;
247 		}
248 	}
249 	if (have_responders || !untried)
250 		return 0;
251 
252 	waits = kmalloc(array_size(vllist->nr_servers, sizeof(*waits)), GFP_KERNEL);
253 	if (!waits)
254 		return -ENOMEM;
255 
256 	for (i = 0; i < vllist->nr_servers; i++) {
257 		if (test_bit(i, &untried)) {
258 			server = vllist->servers[i].server;
259 			init_waitqueue_entry(&waits[i], current);
260 			add_wait_queue(&server->probe_wq, &waits[i]);
261 		}
262 	}
263 
264 	for (;;) {
265 		bool still_probing = false;
266 
267 		set_current_state(TASK_INTERRUPTIBLE);
268 		for (i = 0; i < vllist->nr_servers; i++) {
269 			if (test_bit(i, &untried)) {
270 				server = vllist->servers[i].server;
271 				if (server->probe.flags & AFS_VLSERVER_PROBE_RESPONDED)
272 					goto stop;
273 				if (test_bit(AFS_VLSERVER_FL_PROBING, &server->flags))
274 					still_probing = true;
275 			}
276 		}
277 
278 		if (!still_probing || signal_pending(current))
279 			goto stop;
280 		schedule();
281 	}
282 
283 stop:
284 	set_current_state(TASK_RUNNING);
285 
286 	for (i = 0; i < vllist->nr_servers; i++) {
287 		if (test_bit(i, &untried)) {
288 			server = vllist->servers[i].server;
289 			rtt_s = READ_ONCE(server->rtt);
290 			if (test_bit(AFS_VLSERVER_FL_RESPONDING, &server->flags) &&
291 			    rtt_s < rtt) {
292 				pref = i;
293 				rtt = rtt_s;
294 			}
295 
296 			remove_wait_queue(&server->probe_wq, &waits[i]);
297 		}
298 	}
299 
300 	kfree(waits);
301 
302 	if (pref == -1 && signal_pending(current))
303 		return -ERESTARTSYS;
304 
305 	if (pref >= 0)
306 		vllist->preferred = pref;
307 
308 	_leave(" = 0 [%u]", pref);
309 	return 0;
310 }
311