xref: /linux/fs/afs/fs_probe.c (revision aa23aa55166c2865ac430168c4b9d405cf8c6980)
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 	read_unlock(&server->fs_lock);
149 
150 	atomic_set(&server->probe_outstanding, ac.alist->nr_addrs);
151 	memset(&server->probe, 0, sizeof(server->probe));
152 	server->probe.rtt = UINT_MAX;
153 
154 	for (ac.index = 0; ac.index < ac.alist->nr_addrs; ac.index++) {
155 		call = afs_fs_get_capabilities(net, server, &ac, key, server_index);
156 		if (!IS_ERR(call)) {
157 			afs_put_call(call);
158 			in_progress = true;
159 		} else {
160 			afs_prioritise_error(_e, PTR_ERR(call), ac.abort_code);
161 		}
162 	}
163 
164 	if (!in_progress)
165 		afs_fs_probe_done(server);
166 	return in_progress;
167 }
168 
169 /*
170  * Send off probes to all unprobed servers.
171  */
172 int afs_probe_fileservers(struct afs_net *net, struct key *key,
173 			  struct afs_server_list *list)
174 {
175 	struct afs_server *server;
176 	struct afs_error e;
177 	bool in_progress = false;
178 	int i;
179 
180 	e.error = 0;
181 	e.responded = false;
182 	for (i = 0; i < list->nr_servers; i++) {
183 		server = list->servers[i].server;
184 		if (test_bit(AFS_SERVER_FL_PROBED, &server->flags))
185 			continue;
186 
187 		if (!test_and_set_bit_lock(AFS_SERVER_FL_PROBING, &server->flags) &&
188 		    afs_do_probe_fileserver(net, server, key, i, &e))
189 			in_progress = true;
190 	}
191 
192 	return in_progress ? 0 : e.error;
193 }
194 
195 /*
196  * Wait for the first as-yet untried fileserver to respond.
197  */
198 int afs_wait_for_fs_probes(struct afs_server_list *slist, unsigned long untried)
199 {
200 	struct wait_queue_entry *waits;
201 	struct afs_server *server;
202 	unsigned int rtt = UINT_MAX;
203 	bool have_responders = false;
204 	int pref = -1, i;
205 
206 	_enter("%u,%lx", slist->nr_servers, untried);
207 
208 	/* Only wait for servers that have a probe outstanding. */
209 	for (i = 0; i < slist->nr_servers; i++) {
210 		if (test_bit(i, &untried)) {
211 			server = slist->servers[i].server;
212 			if (!test_bit(AFS_SERVER_FL_PROBING, &server->flags))
213 				__clear_bit(i, &untried);
214 			if (server->probe.responded)
215 				have_responders = true;
216 		}
217 	}
218 	if (have_responders || !untried)
219 		return 0;
220 
221 	waits = kmalloc(array_size(slist->nr_servers, sizeof(*waits)), GFP_KERNEL);
222 	if (!waits)
223 		return -ENOMEM;
224 
225 	for (i = 0; i < slist->nr_servers; i++) {
226 		if (test_bit(i, &untried)) {
227 			server = slist->servers[i].server;
228 			init_waitqueue_entry(&waits[i], current);
229 			add_wait_queue(&server->probe_wq, &waits[i]);
230 		}
231 	}
232 
233 	for (;;) {
234 		bool still_probing = false;
235 
236 		set_current_state(TASK_INTERRUPTIBLE);
237 		for (i = 0; i < slist->nr_servers; i++) {
238 			if (test_bit(i, &untried)) {
239 				server = slist->servers[i].server;
240 				if (server->probe.responded)
241 					goto stop;
242 				if (test_bit(AFS_SERVER_FL_PROBING, &server->flags))
243 					still_probing = true;
244 			}
245 		}
246 
247 		if (!still_probing || signal_pending(current))
248 			goto stop;
249 		schedule();
250 	}
251 
252 stop:
253 	set_current_state(TASK_RUNNING);
254 
255 	for (i = 0; i < slist->nr_servers; i++) {
256 		if (test_bit(i, &untried)) {
257 			server = slist->servers[i].server;
258 			if (server->probe.responded &&
259 			    server->probe.rtt < rtt) {
260 				pref = i;
261 				rtt = server->probe.rtt;
262 			}
263 
264 			remove_wait_queue(&server->probe_wq, &waits[i]);
265 		}
266 	}
267 
268 	kfree(waits);
269 
270 	if (pref == -1 && signal_pending(current))
271 		return -ERESTARTSYS;
272 
273 	if (pref >= 0)
274 		slist->preferred = pref;
275 	return 0;
276 }
277