xref: /openbmc/linux/fs/afs/fs_probe.c (revision 15e3ae36)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* AFS fileserver 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 static bool afs_fs_probe_done(struct afs_server *server)
15 {
16 	if (!atomic_dec_and_test(&server->probe_outstanding))
17 		return false;
18 
19 	wake_up_var(&server->probe_outstanding);
20 	clear_bit_unlock(AFS_SERVER_FL_PROBING, &server->flags);
21 	wake_up_bit(&server->flags, AFS_SERVER_FL_PROBING);
22 	return true;
23 }
24 
25 /*
26  * Process the result of probing a fileserver.  This is called after successful
27  * or failed delivery of an FS.GetCapabilities operation.
28  */
29 void afs_fileserver_probe_result(struct afs_call *call)
30 {
31 	struct afs_addr_list *alist = call->alist;
32 	struct afs_server *server = call->server;
33 	unsigned int server_index = call->server_index;
34 	unsigned int index = call->addr_ix;
35 	unsigned int rtt = UINT_MAX;
36 	bool have_result = false;
37 	u64 _rtt;
38 	int ret = call->error;
39 
40 	_enter("%pU,%u", &server->uuid, index);
41 
42 	spin_lock(&server->probe_lock);
43 
44 	switch (ret) {
45 	case 0:
46 		server->probe.error = 0;
47 		goto responded;
48 	case -ECONNABORTED:
49 		if (!server->probe.responded) {
50 			server->probe.abort_code = call->abort_code;
51 			server->probe.error = ret;
52 		}
53 		goto responded;
54 	case -ENOMEM:
55 	case -ENONET:
56 		server->probe.local_failure = true;
57 		afs_io_error(call, afs_io_error_fs_probe_fail);
58 		goto out;
59 	case -ECONNRESET: /* Responded, but call expired. */
60 	case -ERFKILL:
61 	case -EADDRNOTAVAIL:
62 	case -ENETUNREACH:
63 	case -EHOSTUNREACH:
64 	case -EHOSTDOWN:
65 	case -ECONNREFUSED:
66 	case -ETIMEDOUT:
67 	case -ETIME:
68 	default:
69 		clear_bit(index, &alist->responded);
70 		set_bit(index, &alist->failed);
71 		if (!server->probe.responded &&
72 		    (server->probe.error == 0 ||
73 		     server->probe.error == -ETIMEDOUT ||
74 		     server->probe.error == -ETIME))
75 			server->probe.error = ret;
76 		afs_io_error(call, afs_io_error_fs_probe_fail);
77 		goto out;
78 	}
79 
80 responded:
81 	set_bit(index, &alist->responded);
82 	clear_bit(index, &alist->failed);
83 
84 	if (call->service_id == YFS_FS_SERVICE) {
85 		server->probe.is_yfs = true;
86 		set_bit(AFS_SERVER_FL_IS_YFS, &server->flags);
87 		alist->addrs[index].srx_service = call->service_id;
88 	} else {
89 		server->probe.not_yfs = true;
90 		if (!server->probe.is_yfs) {
91 			clear_bit(AFS_SERVER_FL_IS_YFS, &server->flags);
92 			alist->addrs[index].srx_service = call->service_id;
93 		}
94 	}
95 
96 	/* Get the RTT and scale it to fit into a 32-bit value that represents
97 	 * over a minute of time so that we can access it with one instruction
98 	 * on a 32-bit system.
99 	 */
100 	_rtt = rxrpc_kernel_get_rtt(call->net->socket, call->rxcall);
101 	_rtt /= 64;
102 	rtt = (_rtt > UINT_MAX) ? UINT_MAX : _rtt;
103 	if (rtt < server->probe.rtt) {
104 		server->probe.rtt = rtt;
105 		alist->preferred = index;
106 		have_result = true;
107 	}
108 
109 	smp_wmb(); /* Set rtt before responded. */
110 	server->probe.responded = true;
111 	set_bit(AFS_SERVER_FL_PROBED, &server->flags);
112 out:
113 	spin_unlock(&server->probe_lock);
114 
115 	_debug("probe [%u][%u] %pISpc rtt=%u ret=%d",
116 	       server_index, index, &alist->addrs[index].transport,
117 	       (unsigned int)rtt, ret);
118 
119 	have_result |= afs_fs_probe_done(server);
120 	if (have_result) {
121 		server->probe.have_result = true;
122 		wake_up_var(&server->probe.have_result);
123 		wake_up_all(&server->probe_wq);
124 	}
125 }
126 
127 /*
128  * Probe all of a fileserver's addresses to find out the best route and to
129  * query its capabilities.
130  */
131 static int afs_do_probe_fileserver(struct afs_net *net,
132 				   struct afs_server *server,
133 				   struct key *key,
134 				   unsigned int server_index,
135 				   struct afs_error *_e)
136 {
137 	struct afs_addr_cursor ac = {
138 		.index = 0,
139 	};
140 	struct afs_call *call;
141 	bool in_progress = false;
142 
143 	_enter("%pU", &server->uuid);
144 
145 	read_lock(&server->fs_lock);
146 	ac.alist = rcu_dereference_protected(server->addresses,
147 					     lockdep_is_held(&server->fs_lock));
148 	afs_get_addrlist(ac.alist);
149 	read_unlock(&server->fs_lock);
150 
151 	atomic_set(&server->probe_outstanding, ac.alist->nr_addrs);
152 	memset(&server->probe, 0, sizeof(server->probe));
153 	server->probe.rtt = UINT_MAX;
154 
155 	for (ac.index = 0; ac.index < ac.alist->nr_addrs; ac.index++) {
156 		call = afs_fs_get_capabilities(net, server, &ac, key, server_index);
157 		if (!IS_ERR(call)) {
158 			afs_put_call(call);
159 			in_progress = true;
160 		} else {
161 			afs_prioritise_error(_e, PTR_ERR(call), ac.abort_code);
162 		}
163 	}
164 
165 	if (!in_progress)
166 		afs_fs_probe_done(server);
167 	afs_put_addrlist(ac.alist);
168 	return in_progress;
169 }
170 
171 /*
172  * Send off probes to all unprobed servers.
173  */
174 int afs_probe_fileservers(struct afs_net *net, struct key *key,
175 			  struct afs_server_list *list)
176 {
177 	struct afs_server *server;
178 	struct afs_error e;
179 	bool in_progress = false;
180 	int i;
181 
182 	e.error = 0;
183 	e.responded = false;
184 	for (i = 0; i < list->nr_servers; i++) {
185 		server = list->servers[i].server;
186 		if (test_bit(AFS_SERVER_FL_PROBED, &server->flags))
187 			continue;
188 
189 		if (!test_and_set_bit_lock(AFS_SERVER_FL_PROBING, &server->flags) &&
190 		    afs_do_probe_fileserver(net, server, key, i, &e))
191 			in_progress = true;
192 	}
193 
194 	return in_progress ? 0 : e.error;
195 }
196 
197 /*
198  * Wait for the first as-yet untried fileserver to respond.
199  */
200 int afs_wait_for_fs_probes(struct afs_server_list *slist, unsigned long untried)
201 {
202 	struct wait_queue_entry *waits;
203 	struct afs_server *server;
204 	unsigned int rtt = UINT_MAX;
205 	bool have_responders = false;
206 	int pref = -1, i;
207 
208 	_enter("%u,%lx", slist->nr_servers, untried);
209 
210 	/* Only wait for servers that have a probe outstanding. */
211 	for (i = 0; i < slist->nr_servers; i++) {
212 		if (test_bit(i, &untried)) {
213 			server = slist->servers[i].server;
214 			if (!test_bit(AFS_SERVER_FL_PROBING, &server->flags))
215 				__clear_bit(i, &untried);
216 			if (server->probe.responded)
217 				have_responders = true;
218 		}
219 	}
220 	if (have_responders || !untried)
221 		return 0;
222 
223 	waits = kmalloc(array_size(slist->nr_servers, sizeof(*waits)), GFP_KERNEL);
224 	if (!waits)
225 		return -ENOMEM;
226 
227 	for (i = 0; i < slist->nr_servers; i++) {
228 		if (test_bit(i, &untried)) {
229 			server = slist->servers[i].server;
230 			init_waitqueue_entry(&waits[i], current);
231 			add_wait_queue(&server->probe_wq, &waits[i]);
232 		}
233 	}
234 
235 	for (;;) {
236 		bool still_probing = false;
237 
238 		set_current_state(TASK_INTERRUPTIBLE);
239 		for (i = 0; i < slist->nr_servers; i++) {
240 			if (test_bit(i, &untried)) {
241 				server = slist->servers[i].server;
242 				if (server->probe.responded)
243 					goto stop;
244 				if (test_bit(AFS_SERVER_FL_PROBING, &server->flags))
245 					still_probing = true;
246 			}
247 		}
248 
249 		if (!still_probing || signal_pending(current))
250 			goto stop;
251 		schedule();
252 	}
253 
254 stop:
255 	set_current_state(TASK_RUNNING);
256 
257 	for (i = 0; i < slist->nr_servers; i++) {
258 		if (test_bit(i, &untried)) {
259 			server = slist->servers[i].server;
260 			if (server->probe.responded &&
261 			    server->probe.rtt < rtt) {
262 				pref = i;
263 				rtt = server->probe.rtt;
264 			}
265 
266 			remove_wait_queue(&server->probe_wq, &waits[i]);
267 		}
268 	}
269 
270 	kfree(waits);
271 
272 	if (pref == -1 && signal_pending(current))
273 		return -ERESTARTSYS;
274 
275 	if (pref >= 0)
276 		slist->preferred = pref;
277 	return 0;
278 }
279