xref: /openbmc/linux/fs/afs/fs_probe.c (revision 9ac17575)
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_us = 0;
36 	bool have_result = false;
37 	int ret = call->error;
38 
39 	_enter("%pU,%u", &server->uuid, index);
40 
41 	spin_lock(&server->probe_lock);
42 
43 	switch (ret) {
44 	case 0:
45 		server->probe.error = 0;
46 		goto responded;
47 	case -ECONNABORTED:
48 		if (!server->probe.responded) {
49 			server->probe.abort_code = call->abort_code;
50 			server->probe.error = ret;
51 		}
52 		goto responded;
53 	case -ENOMEM:
54 	case -ENONET:
55 		server->probe.local_failure = true;
56 		afs_io_error(call, afs_io_error_fs_probe_fail);
57 		goto out;
58 	case -ECONNRESET: /* Responded, but call expired. */
59 	case -ERFKILL:
60 	case -EADDRNOTAVAIL:
61 	case -ENETUNREACH:
62 	case -EHOSTUNREACH:
63 	case -EHOSTDOWN:
64 	case -ECONNREFUSED:
65 	case -ETIMEDOUT:
66 	case -ETIME:
67 	default:
68 		clear_bit(index, &alist->responded);
69 		set_bit(index, &alist->failed);
70 		if (!server->probe.responded &&
71 		    (server->probe.error == 0 ||
72 		     server->probe.error == -ETIMEDOUT ||
73 		     server->probe.error == -ETIME))
74 			server->probe.error = ret;
75 		afs_io_error(call, afs_io_error_fs_probe_fail);
76 		goto out;
77 	}
78 
79 responded:
80 	set_bit(index, &alist->responded);
81 	clear_bit(index, &alist->failed);
82 
83 	if (call->service_id == YFS_FS_SERVICE) {
84 		server->probe.is_yfs = true;
85 		set_bit(AFS_SERVER_FL_IS_YFS, &server->flags);
86 		alist->addrs[index].srx_service = call->service_id;
87 	} else {
88 		server->probe.not_yfs = true;
89 		if (!server->probe.is_yfs) {
90 			clear_bit(AFS_SERVER_FL_IS_YFS, &server->flags);
91 			alist->addrs[index].srx_service = call->service_id;
92 		}
93 	}
94 
95 	rtt_us = rxrpc_kernel_get_srtt(call->net->socket, call->rxcall);
96 	if (rtt_us < server->probe.rtt) {
97 		server->probe.rtt = rtt_us;
98 		alist->preferred = index;
99 		have_result = true;
100 	}
101 
102 	smp_wmb(); /* Set rtt before responded. */
103 	server->probe.responded = true;
104 	set_bit(AFS_SERVER_FL_PROBED, &server->flags);
105 out:
106 	spin_unlock(&server->probe_lock);
107 
108 	_debug("probe [%u][%u] %pISpc rtt=%u ret=%d",
109 	       server_index, index, &alist->addrs[index].transport, rtt_us, ret);
110 
111 	have_result |= afs_fs_probe_done(server);
112 	if (have_result)
113 		wake_up_all(&server->probe_wq);
114 }
115 
116 /*
117  * Probe all of a fileserver's addresses to find out the best route and to
118  * query its capabilities.
119  */
120 static int afs_do_probe_fileserver(struct afs_net *net,
121 				   struct afs_server *server,
122 				   struct key *key,
123 				   unsigned int server_index,
124 				   struct afs_error *_e)
125 {
126 	struct afs_addr_cursor ac = {
127 		.index = 0,
128 	};
129 	struct afs_call *call;
130 	bool in_progress = false;
131 
132 	_enter("%pU", &server->uuid);
133 
134 	read_lock(&server->fs_lock);
135 	ac.alist = rcu_dereference_protected(server->addresses,
136 					     lockdep_is_held(&server->fs_lock));
137 	afs_get_addrlist(ac.alist);
138 	read_unlock(&server->fs_lock);
139 
140 	atomic_set(&server->probe_outstanding, ac.alist->nr_addrs);
141 	memset(&server->probe, 0, sizeof(server->probe));
142 	server->probe.rtt = UINT_MAX;
143 
144 	for (ac.index = 0; ac.index < ac.alist->nr_addrs; ac.index++) {
145 		call = afs_fs_get_capabilities(net, server, &ac, key, server_index);
146 		if (!IS_ERR(call)) {
147 			afs_put_call(call);
148 			in_progress = true;
149 		} else {
150 			afs_prioritise_error(_e, PTR_ERR(call), ac.abort_code);
151 		}
152 	}
153 
154 	if (!in_progress)
155 		afs_fs_probe_done(server);
156 	afs_put_addrlist(ac.alist);
157 	return in_progress;
158 }
159 
160 /*
161  * Send off probes to all unprobed servers.
162  */
163 int afs_probe_fileservers(struct afs_net *net, struct key *key,
164 			  struct afs_server_list *list)
165 {
166 	struct afs_server *server;
167 	struct afs_error e;
168 	bool in_progress = false;
169 	int i;
170 
171 	e.error = 0;
172 	e.responded = false;
173 	for (i = 0; i < list->nr_servers; i++) {
174 		server = list->servers[i].server;
175 		if (test_bit(AFS_SERVER_FL_PROBED, &server->flags))
176 			continue;
177 
178 		if (!test_and_set_bit_lock(AFS_SERVER_FL_PROBING, &server->flags) &&
179 		    afs_do_probe_fileserver(net, server, key, i, &e))
180 			in_progress = true;
181 	}
182 
183 	return in_progress ? 0 : e.error;
184 }
185 
186 /*
187  * Wait for the first as-yet untried fileserver to respond.
188  */
189 int afs_wait_for_fs_probes(struct afs_server_list *slist, unsigned long untried)
190 {
191 	struct wait_queue_entry *waits;
192 	struct afs_server *server;
193 	unsigned int rtt = UINT_MAX;
194 	bool have_responders = false;
195 	int pref = -1, i;
196 
197 	_enter("%u,%lx", slist->nr_servers, untried);
198 
199 	/* Only wait for servers that have a probe outstanding. */
200 	for (i = 0; i < slist->nr_servers; i++) {
201 		if (test_bit(i, &untried)) {
202 			server = slist->servers[i].server;
203 			if (!test_bit(AFS_SERVER_FL_PROBING, &server->flags))
204 				__clear_bit(i, &untried);
205 			if (server->probe.responded)
206 				have_responders = true;
207 		}
208 	}
209 	if (have_responders || !untried)
210 		return 0;
211 
212 	waits = kmalloc(array_size(slist->nr_servers, sizeof(*waits)), GFP_KERNEL);
213 	if (!waits)
214 		return -ENOMEM;
215 
216 	for (i = 0; i < slist->nr_servers; i++) {
217 		if (test_bit(i, &untried)) {
218 			server = slist->servers[i].server;
219 			init_waitqueue_entry(&waits[i], current);
220 			add_wait_queue(&server->probe_wq, &waits[i]);
221 		}
222 	}
223 
224 	for (;;) {
225 		bool still_probing = false;
226 
227 		set_current_state(TASK_INTERRUPTIBLE);
228 		for (i = 0; i < slist->nr_servers; i++) {
229 			if (test_bit(i, &untried)) {
230 				server = slist->servers[i].server;
231 				if (server->probe.responded)
232 					goto stop;
233 				if (test_bit(AFS_SERVER_FL_PROBING, &server->flags))
234 					still_probing = true;
235 			}
236 		}
237 
238 		if (!still_probing || signal_pending(current))
239 			goto stop;
240 		schedule();
241 	}
242 
243 stop:
244 	set_current_state(TASK_RUNNING);
245 
246 	for (i = 0; i < slist->nr_servers; i++) {
247 		if (test_bit(i, &untried)) {
248 			server = slist->servers[i].server;
249 			if (server->probe.responded &&
250 			    server->probe.rtt < rtt) {
251 				pref = i;
252 				rtt = server->probe.rtt;
253 			}
254 
255 			remove_wait_queue(&server->probe_wq, &waits[i]);
256 		}
257 	}
258 
259 	kfree(waits);
260 
261 	if (pref == -1 && signal_pending(current))
262 		return -ERESTARTSYS;
263 
264 	if (pref >= 0)
265 		slist->preferred = pref;
266 	return 0;
267 }
268