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 wake_up_all(&server->probe_wq); 122 } 123 124 /* 125 * Probe all of a fileserver's addresses to find out the best route and to 126 * query its capabilities. 127 */ 128 static int afs_do_probe_fileserver(struct afs_net *net, 129 struct afs_server *server, 130 struct key *key, 131 unsigned int server_index, 132 struct afs_error *_e) 133 { 134 struct afs_addr_cursor ac = { 135 .index = 0, 136 }; 137 struct afs_call *call; 138 bool in_progress = false; 139 140 _enter("%pU", &server->uuid); 141 142 read_lock(&server->fs_lock); 143 ac.alist = rcu_dereference_protected(server->addresses, 144 lockdep_is_held(&server->fs_lock)); 145 afs_get_addrlist(ac.alist); 146 read_unlock(&server->fs_lock); 147 148 atomic_set(&server->probe_outstanding, ac.alist->nr_addrs); 149 memset(&server->probe, 0, sizeof(server->probe)); 150 server->probe.rtt = UINT_MAX; 151 152 for (ac.index = 0; ac.index < ac.alist->nr_addrs; ac.index++) { 153 call = afs_fs_get_capabilities(net, server, &ac, key, server_index); 154 if (!IS_ERR(call)) { 155 afs_put_call(call); 156 in_progress = true; 157 } else { 158 afs_prioritise_error(_e, PTR_ERR(call), ac.abort_code); 159 } 160 } 161 162 if (!in_progress) 163 afs_fs_probe_done(server); 164 afs_put_addrlist(ac.alist); 165 return in_progress; 166 } 167 168 /* 169 * Send off probes to all unprobed servers. 170 */ 171 int afs_probe_fileservers(struct afs_net *net, struct key *key, 172 struct afs_server_list *list) 173 { 174 struct afs_server *server; 175 struct afs_error e; 176 bool in_progress = false; 177 int i; 178 179 e.error = 0; 180 e.responded = false; 181 for (i = 0; i < list->nr_servers; i++) { 182 server = list->servers[i].server; 183 if (test_bit(AFS_SERVER_FL_PROBED, &server->flags)) 184 continue; 185 186 if (!test_and_set_bit_lock(AFS_SERVER_FL_PROBING, &server->flags) && 187 afs_do_probe_fileserver(net, server, key, i, &e)) 188 in_progress = true; 189 } 190 191 return in_progress ? 0 : e.error; 192 } 193 194 /* 195 * Wait for the first as-yet untried fileserver to respond. 196 */ 197 int afs_wait_for_fs_probes(struct afs_server_list *slist, unsigned long untried) 198 { 199 struct wait_queue_entry *waits; 200 struct afs_server *server; 201 unsigned int rtt = UINT_MAX; 202 bool have_responders = false; 203 int pref = -1, i; 204 205 _enter("%u,%lx", slist->nr_servers, untried); 206 207 /* Only wait for servers that have a probe outstanding. */ 208 for (i = 0; i < slist->nr_servers; i++) { 209 if (test_bit(i, &untried)) { 210 server = slist->servers[i].server; 211 if (!test_bit(AFS_SERVER_FL_PROBING, &server->flags)) 212 __clear_bit(i, &untried); 213 if (server->probe.responded) 214 have_responders = true; 215 } 216 } 217 if (have_responders || !untried) 218 return 0; 219 220 waits = kmalloc(array_size(slist->nr_servers, sizeof(*waits)), GFP_KERNEL); 221 if (!waits) 222 return -ENOMEM; 223 224 for (i = 0; i < slist->nr_servers; i++) { 225 if (test_bit(i, &untried)) { 226 server = slist->servers[i].server; 227 init_waitqueue_entry(&waits[i], current); 228 add_wait_queue(&server->probe_wq, &waits[i]); 229 } 230 } 231 232 for (;;) { 233 bool still_probing = false; 234 235 set_current_state(TASK_INTERRUPTIBLE); 236 for (i = 0; i < slist->nr_servers; i++) { 237 if (test_bit(i, &untried)) { 238 server = slist->servers[i].server; 239 if (server->probe.responded) 240 goto stop; 241 if (test_bit(AFS_SERVER_FL_PROBING, &server->flags)) 242 still_probing = true; 243 } 244 } 245 246 if (!still_probing || signal_pending(current)) 247 goto stop; 248 schedule(); 249 } 250 251 stop: 252 set_current_state(TASK_RUNNING); 253 254 for (i = 0; i < slist->nr_servers; i++) { 255 if (test_bit(i, &untried)) { 256 server = slist->servers[i].server; 257 if (server->probe.responded && 258 server->probe.rtt < rtt) { 259 pref = i; 260 rtt = server->probe.rtt; 261 } 262 263 remove_wait_queue(&server->probe_wq, &waits[i]); 264 } 265 } 266 267 kfree(waits); 268 269 if (pref == -1 && signal_pending(current)) 270 return -ERESTARTSYS; 271 272 if (pref >= 0) 273 slist->preferred = pref; 274 return 0; 275 } 276