1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * linux/net/sunrpc/svc_xprt.c 4 * 5 * Author: Tom Tucker <tom@opengridcomputing.com> 6 */ 7 8 #include <linux/sched.h> 9 #include <linux/errno.h> 10 #include <linux/freezer.h> 11 #include <linux/kthread.h> 12 #include <linux/slab.h> 13 #include <net/sock.h> 14 #include <linux/sunrpc/addr.h> 15 #include <linux/sunrpc/stats.h> 16 #include <linux/sunrpc/svc_xprt.h> 17 #include <linux/sunrpc/svcsock.h> 18 #include <linux/sunrpc/xprt.h> 19 #include <linux/module.h> 20 #include <linux/netdevice.h> 21 #include <trace/events/sunrpc.h> 22 23 #define RPCDBG_FACILITY RPCDBG_SVCXPRT 24 25 static unsigned int svc_rpc_per_connection_limit __read_mostly; 26 module_param(svc_rpc_per_connection_limit, uint, 0644); 27 28 29 static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt); 30 static int svc_deferred_recv(struct svc_rqst *rqstp); 31 static struct cache_deferred_req *svc_defer(struct cache_req *req); 32 static void svc_age_temp_xprts(struct timer_list *t); 33 static void svc_delete_xprt(struct svc_xprt *xprt); 34 35 /* apparently the "standard" is that clients close 36 * idle connections after 5 minutes, servers after 37 * 6 minutes 38 * http://nfsv4bat.org/Documents/ConnectAThon/1996/nfstcp.pdf 39 */ 40 static int svc_conn_age_period = 6*60; 41 42 /* List of registered transport classes */ 43 static DEFINE_SPINLOCK(svc_xprt_class_lock); 44 static LIST_HEAD(svc_xprt_class_list); 45 46 /* SMP locking strategy: 47 * 48 * svc_pool->sp_lock protects most of the fields of that pool. 49 * svc_serv->sv_lock protects sv_tempsocks, sv_permsocks, sv_tmpcnt. 50 * when both need to be taken (rare), svc_serv->sv_lock is first. 51 * The "service mutex" protects svc_serv->sv_nrthread. 52 * svc_sock->sk_lock protects the svc_sock->sk_deferred list 53 * and the ->sk_info_authunix cache. 54 * 55 * The XPT_BUSY bit in xprt->xpt_flags prevents a transport being 56 * enqueued multiply. During normal transport processing this bit 57 * is set by svc_xprt_enqueue and cleared by svc_xprt_received. 58 * Providers should not manipulate this bit directly. 59 * 60 * Some flags can be set to certain values at any time 61 * providing that certain rules are followed: 62 * 63 * XPT_CONN, XPT_DATA: 64 * - Can be set or cleared at any time. 65 * - After a set, svc_xprt_enqueue must be called to enqueue 66 * the transport for processing. 67 * - After a clear, the transport must be read/accepted. 68 * If this succeeds, it must be set again. 69 * XPT_CLOSE: 70 * - Can set at any time. It is never cleared. 71 * XPT_DEAD: 72 * - Can only be set while XPT_BUSY is held which ensures 73 * that no other thread will be using the transport or will 74 * try to set XPT_DEAD. 75 */ 76 int svc_reg_xprt_class(struct svc_xprt_class *xcl) 77 { 78 struct svc_xprt_class *cl; 79 int res = -EEXIST; 80 81 dprintk("svc: Adding svc transport class '%s'\n", xcl->xcl_name); 82 83 INIT_LIST_HEAD(&xcl->xcl_list); 84 spin_lock(&svc_xprt_class_lock); 85 /* Make sure there isn't already a class with the same name */ 86 list_for_each_entry(cl, &svc_xprt_class_list, xcl_list) { 87 if (strcmp(xcl->xcl_name, cl->xcl_name) == 0) 88 goto out; 89 } 90 list_add_tail(&xcl->xcl_list, &svc_xprt_class_list); 91 res = 0; 92 out: 93 spin_unlock(&svc_xprt_class_lock); 94 return res; 95 } 96 EXPORT_SYMBOL_GPL(svc_reg_xprt_class); 97 98 void svc_unreg_xprt_class(struct svc_xprt_class *xcl) 99 { 100 dprintk("svc: Removing svc transport class '%s'\n", xcl->xcl_name); 101 spin_lock(&svc_xprt_class_lock); 102 list_del_init(&xcl->xcl_list); 103 spin_unlock(&svc_xprt_class_lock); 104 } 105 EXPORT_SYMBOL_GPL(svc_unreg_xprt_class); 106 107 /** 108 * svc_print_xprts - Format the transport list for printing 109 * @buf: target buffer for formatted address 110 * @maxlen: length of target buffer 111 * 112 * Fills in @buf with a string containing a list of transport names, each name 113 * terminated with '\n'. If the buffer is too small, some entries may be 114 * missing, but it is guaranteed that all lines in the output buffer are 115 * complete. 116 * 117 * Returns positive length of the filled-in string. 118 */ 119 int svc_print_xprts(char *buf, int maxlen) 120 { 121 struct svc_xprt_class *xcl; 122 char tmpstr[80]; 123 int len = 0; 124 buf[0] = '\0'; 125 126 spin_lock(&svc_xprt_class_lock); 127 list_for_each_entry(xcl, &svc_xprt_class_list, xcl_list) { 128 int slen; 129 130 slen = snprintf(tmpstr, sizeof(tmpstr), "%s %d\n", 131 xcl->xcl_name, xcl->xcl_max_payload); 132 if (slen >= sizeof(tmpstr) || len + slen >= maxlen) 133 break; 134 len += slen; 135 strcat(buf, tmpstr); 136 } 137 spin_unlock(&svc_xprt_class_lock); 138 139 return len; 140 } 141 142 static void svc_xprt_free(struct kref *kref) 143 { 144 struct svc_xprt *xprt = 145 container_of(kref, struct svc_xprt, xpt_ref); 146 struct module *owner = xprt->xpt_class->xcl_owner; 147 if (test_bit(XPT_CACHE_AUTH, &xprt->xpt_flags)) 148 svcauth_unix_info_release(xprt); 149 put_cred(xprt->xpt_cred); 150 put_net(xprt->xpt_net); 151 /* See comment on corresponding get in xs_setup_bc_tcp(): */ 152 if (xprt->xpt_bc_xprt) 153 xprt_put(xprt->xpt_bc_xprt); 154 if (xprt->xpt_bc_xps) 155 xprt_switch_put(xprt->xpt_bc_xps); 156 xprt->xpt_ops->xpo_free(xprt); 157 module_put(owner); 158 } 159 160 void svc_xprt_put(struct svc_xprt *xprt) 161 { 162 kref_put(&xprt->xpt_ref, svc_xprt_free); 163 } 164 EXPORT_SYMBOL_GPL(svc_xprt_put); 165 166 /* 167 * Called by transport drivers to initialize the transport independent 168 * portion of the transport instance. 169 */ 170 void svc_xprt_init(struct net *net, struct svc_xprt_class *xcl, 171 struct svc_xprt *xprt, struct svc_serv *serv) 172 { 173 memset(xprt, 0, sizeof(*xprt)); 174 xprt->xpt_class = xcl; 175 xprt->xpt_ops = xcl->xcl_ops; 176 kref_init(&xprt->xpt_ref); 177 xprt->xpt_server = serv; 178 INIT_LIST_HEAD(&xprt->xpt_list); 179 INIT_LIST_HEAD(&xprt->xpt_ready); 180 INIT_LIST_HEAD(&xprt->xpt_deferred); 181 INIT_LIST_HEAD(&xprt->xpt_users); 182 mutex_init(&xprt->xpt_mutex); 183 spin_lock_init(&xprt->xpt_lock); 184 set_bit(XPT_BUSY, &xprt->xpt_flags); 185 xprt->xpt_net = get_net(net); 186 strcpy(xprt->xpt_remotebuf, "uninitialized"); 187 } 188 EXPORT_SYMBOL_GPL(svc_xprt_init); 189 190 static struct svc_xprt *__svc_xpo_create(struct svc_xprt_class *xcl, 191 struct svc_serv *serv, 192 struct net *net, 193 const int family, 194 const unsigned short port, 195 int flags) 196 { 197 struct sockaddr_in sin = { 198 .sin_family = AF_INET, 199 .sin_addr.s_addr = htonl(INADDR_ANY), 200 .sin_port = htons(port), 201 }; 202 #if IS_ENABLED(CONFIG_IPV6) 203 struct sockaddr_in6 sin6 = { 204 .sin6_family = AF_INET6, 205 .sin6_addr = IN6ADDR_ANY_INIT, 206 .sin6_port = htons(port), 207 }; 208 #endif 209 struct sockaddr *sap; 210 size_t len; 211 212 switch (family) { 213 case PF_INET: 214 sap = (struct sockaddr *)&sin; 215 len = sizeof(sin); 216 break; 217 #if IS_ENABLED(CONFIG_IPV6) 218 case PF_INET6: 219 sap = (struct sockaddr *)&sin6; 220 len = sizeof(sin6); 221 break; 222 #endif 223 default: 224 return ERR_PTR(-EAFNOSUPPORT); 225 } 226 227 return xcl->xcl_ops->xpo_create(serv, net, sap, len, flags); 228 } 229 230 /* 231 * svc_xprt_received conditionally queues the transport for processing 232 * by another thread. The caller must hold the XPT_BUSY bit and must 233 * not thereafter touch transport data. 234 * 235 * Note: XPT_DATA only gets cleared when a read-attempt finds no (or 236 * insufficient) data. 237 */ 238 static void svc_xprt_received(struct svc_xprt *xprt) 239 { 240 if (!test_bit(XPT_BUSY, &xprt->xpt_flags)) { 241 WARN_ONCE(1, "xprt=0x%p already busy!", xprt); 242 return; 243 } 244 245 /* As soon as we clear busy, the xprt could be closed and 246 * 'put', so we need a reference to call svc_enqueue_xprt with: 247 */ 248 svc_xprt_get(xprt); 249 smp_mb__before_atomic(); 250 clear_bit(XPT_BUSY, &xprt->xpt_flags); 251 xprt->xpt_server->sv_ops->svo_enqueue_xprt(xprt); 252 svc_xprt_put(xprt); 253 } 254 255 void svc_add_new_perm_xprt(struct svc_serv *serv, struct svc_xprt *new) 256 { 257 clear_bit(XPT_TEMP, &new->xpt_flags); 258 spin_lock_bh(&serv->sv_lock); 259 list_add(&new->xpt_list, &serv->sv_permsocks); 260 spin_unlock_bh(&serv->sv_lock); 261 svc_xprt_received(new); 262 } 263 264 static int _svc_create_xprt(struct svc_serv *serv, const char *xprt_name, 265 struct net *net, const int family, 266 const unsigned short port, int flags, 267 const struct cred *cred) 268 { 269 struct svc_xprt_class *xcl; 270 271 spin_lock(&svc_xprt_class_lock); 272 list_for_each_entry(xcl, &svc_xprt_class_list, xcl_list) { 273 struct svc_xprt *newxprt; 274 unsigned short newport; 275 276 if (strcmp(xprt_name, xcl->xcl_name)) 277 continue; 278 279 if (!try_module_get(xcl->xcl_owner)) 280 goto err; 281 282 spin_unlock(&svc_xprt_class_lock); 283 newxprt = __svc_xpo_create(xcl, serv, net, family, port, flags); 284 if (IS_ERR(newxprt)) { 285 module_put(xcl->xcl_owner); 286 return PTR_ERR(newxprt); 287 } 288 newxprt->xpt_cred = get_cred(cred); 289 svc_add_new_perm_xprt(serv, newxprt); 290 newport = svc_xprt_local_port(newxprt); 291 return newport; 292 } 293 err: 294 spin_unlock(&svc_xprt_class_lock); 295 /* This errno is exposed to user space. Provide a reasonable 296 * perror msg for a bad transport. */ 297 return -EPROTONOSUPPORT; 298 } 299 300 int svc_create_xprt(struct svc_serv *serv, const char *xprt_name, 301 struct net *net, const int family, 302 const unsigned short port, int flags, 303 const struct cred *cred) 304 { 305 int err; 306 307 dprintk("svc: creating transport %s[%d]\n", xprt_name, port); 308 err = _svc_create_xprt(serv, xprt_name, net, family, port, flags, cred); 309 if (err == -EPROTONOSUPPORT) { 310 request_module("svc%s", xprt_name); 311 err = _svc_create_xprt(serv, xprt_name, net, family, port, flags, cred); 312 } 313 if (err < 0) 314 dprintk("svc: transport %s not found, err %d\n", 315 xprt_name, -err); 316 return err; 317 } 318 EXPORT_SYMBOL_GPL(svc_create_xprt); 319 320 /* 321 * Copy the local and remote xprt addresses to the rqstp structure 322 */ 323 void svc_xprt_copy_addrs(struct svc_rqst *rqstp, struct svc_xprt *xprt) 324 { 325 memcpy(&rqstp->rq_addr, &xprt->xpt_remote, xprt->xpt_remotelen); 326 rqstp->rq_addrlen = xprt->xpt_remotelen; 327 328 /* 329 * Destination address in request is needed for binding the 330 * source address in RPC replies/callbacks later. 331 */ 332 memcpy(&rqstp->rq_daddr, &xprt->xpt_local, xprt->xpt_locallen); 333 rqstp->rq_daddrlen = xprt->xpt_locallen; 334 } 335 EXPORT_SYMBOL_GPL(svc_xprt_copy_addrs); 336 337 /** 338 * svc_print_addr - Format rq_addr field for printing 339 * @rqstp: svc_rqst struct containing address to print 340 * @buf: target buffer for formatted address 341 * @len: length of target buffer 342 * 343 */ 344 char *svc_print_addr(struct svc_rqst *rqstp, char *buf, size_t len) 345 { 346 return __svc_print_addr(svc_addr(rqstp), buf, len); 347 } 348 EXPORT_SYMBOL_GPL(svc_print_addr); 349 350 static bool svc_xprt_slots_in_range(struct svc_xprt *xprt) 351 { 352 unsigned int limit = svc_rpc_per_connection_limit; 353 int nrqsts = atomic_read(&xprt->xpt_nr_rqsts); 354 355 return limit == 0 || (nrqsts >= 0 && nrqsts < limit); 356 } 357 358 static bool svc_xprt_reserve_slot(struct svc_rqst *rqstp, struct svc_xprt *xprt) 359 { 360 if (!test_bit(RQ_DATA, &rqstp->rq_flags)) { 361 if (!svc_xprt_slots_in_range(xprt)) 362 return false; 363 atomic_inc(&xprt->xpt_nr_rqsts); 364 set_bit(RQ_DATA, &rqstp->rq_flags); 365 } 366 return true; 367 } 368 369 static void svc_xprt_release_slot(struct svc_rqst *rqstp) 370 { 371 struct svc_xprt *xprt = rqstp->rq_xprt; 372 if (test_and_clear_bit(RQ_DATA, &rqstp->rq_flags)) { 373 atomic_dec(&xprt->xpt_nr_rqsts); 374 smp_wmb(); /* See smp_rmb() in svc_xprt_ready() */ 375 svc_xprt_enqueue(xprt); 376 } 377 } 378 379 static bool svc_xprt_ready(struct svc_xprt *xprt) 380 { 381 unsigned long xpt_flags; 382 383 /* 384 * If another cpu has recently updated xpt_flags, 385 * sk_sock->flags, xpt_reserved, or xpt_nr_rqsts, we need to 386 * know about it; otherwise it's possible that both that cpu and 387 * this one could call svc_xprt_enqueue() without either 388 * svc_xprt_enqueue() recognizing that the conditions below 389 * are satisfied, and we could stall indefinitely: 390 */ 391 smp_rmb(); 392 xpt_flags = READ_ONCE(xprt->xpt_flags); 393 394 if (xpt_flags & (BIT(XPT_CONN) | BIT(XPT_CLOSE))) 395 return true; 396 if (xpt_flags & (BIT(XPT_DATA) | BIT(XPT_DEFERRED))) { 397 if (xprt->xpt_ops->xpo_has_wspace(xprt) && 398 svc_xprt_slots_in_range(xprt)) 399 return true; 400 trace_svc_xprt_no_write_space(xprt); 401 return false; 402 } 403 return false; 404 } 405 406 void svc_xprt_do_enqueue(struct svc_xprt *xprt) 407 { 408 struct svc_pool *pool; 409 struct svc_rqst *rqstp = NULL; 410 int cpu; 411 412 if (!svc_xprt_ready(xprt)) 413 return; 414 415 /* Mark transport as busy. It will remain in this state until 416 * the provider calls svc_xprt_received. We update XPT_BUSY 417 * atomically because it also guards against trying to enqueue 418 * the transport twice. 419 */ 420 if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags)) 421 return; 422 423 cpu = get_cpu(); 424 pool = svc_pool_for_cpu(xprt->xpt_server, cpu); 425 426 atomic_long_inc(&pool->sp_stats.packets); 427 428 spin_lock_bh(&pool->sp_lock); 429 list_add_tail(&xprt->xpt_ready, &pool->sp_sockets); 430 pool->sp_stats.sockets_queued++; 431 spin_unlock_bh(&pool->sp_lock); 432 433 /* find a thread for this xprt */ 434 rcu_read_lock(); 435 list_for_each_entry_rcu(rqstp, &pool->sp_all_threads, rq_all) { 436 if (test_and_set_bit(RQ_BUSY, &rqstp->rq_flags)) 437 continue; 438 atomic_long_inc(&pool->sp_stats.threads_woken); 439 rqstp->rq_qtime = ktime_get(); 440 wake_up_process(rqstp->rq_task); 441 goto out_unlock; 442 } 443 set_bit(SP_CONGESTED, &pool->sp_flags); 444 rqstp = NULL; 445 out_unlock: 446 rcu_read_unlock(); 447 put_cpu(); 448 trace_svc_xprt_do_enqueue(xprt, rqstp); 449 } 450 EXPORT_SYMBOL_GPL(svc_xprt_do_enqueue); 451 452 /* 453 * Queue up a transport with data pending. If there are idle nfsd 454 * processes, wake 'em up. 455 * 456 */ 457 void svc_xprt_enqueue(struct svc_xprt *xprt) 458 { 459 if (test_bit(XPT_BUSY, &xprt->xpt_flags)) 460 return; 461 xprt->xpt_server->sv_ops->svo_enqueue_xprt(xprt); 462 } 463 EXPORT_SYMBOL_GPL(svc_xprt_enqueue); 464 465 /* 466 * Dequeue the first transport, if there is one. 467 */ 468 static struct svc_xprt *svc_xprt_dequeue(struct svc_pool *pool) 469 { 470 struct svc_xprt *xprt = NULL; 471 472 if (list_empty(&pool->sp_sockets)) 473 goto out; 474 475 spin_lock_bh(&pool->sp_lock); 476 if (likely(!list_empty(&pool->sp_sockets))) { 477 xprt = list_first_entry(&pool->sp_sockets, 478 struct svc_xprt, xpt_ready); 479 list_del_init(&xprt->xpt_ready); 480 svc_xprt_get(xprt); 481 } 482 spin_unlock_bh(&pool->sp_lock); 483 out: 484 return xprt; 485 } 486 487 /** 488 * svc_reserve - change the space reserved for the reply to a request. 489 * @rqstp: The request in question 490 * @space: new max space to reserve 491 * 492 * Each request reserves some space on the output queue of the transport 493 * to make sure the reply fits. This function reduces that reserved 494 * space to be the amount of space used already, plus @space. 495 * 496 */ 497 void svc_reserve(struct svc_rqst *rqstp, int space) 498 { 499 struct svc_xprt *xprt = rqstp->rq_xprt; 500 501 space += rqstp->rq_res.head[0].iov_len; 502 503 if (xprt && space < rqstp->rq_reserved) { 504 atomic_sub((rqstp->rq_reserved - space), &xprt->xpt_reserved); 505 rqstp->rq_reserved = space; 506 smp_wmb(); /* See smp_rmb() in svc_xprt_ready() */ 507 svc_xprt_enqueue(xprt); 508 } 509 } 510 EXPORT_SYMBOL_GPL(svc_reserve); 511 512 static void svc_xprt_release(struct svc_rqst *rqstp) 513 { 514 struct svc_xprt *xprt = rqstp->rq_xprt; 515 516 xprt->xpt_ops->xpo_release_rqst(rqstp); 517 518 kfree(rqstp->rq_deferred); 519 rqstp->rq_deferred = NULL; 520 521 svc_free_res_pages(rqstp); 522 rqstp->rq_res.page_len = 0; 523 rqstp->rq_res.page_base = 0; 524 525 /* Reset response buffer and release 526 * the reservation. 527 * But first, check that enough space was reserved 528 * for the reply, otherwise we have a bug! 529 */ 530 if ((rqstp->rq_res.len) > rqstp->rq_reserved) 531 printk(KERN_ERR "RPC request reserved %d but used %d\n", 532 rqstp->rq_reserved, 533 rqstp->rq_res.len); 534 535 rqstp->rq_res.head[0].iov_len = 0; 536 svc_reserve(rqstp, 0); 537 svc_xprt_release_slot(rqstp); 538 rqstp->rq_xprt = NULL; 539 svc_xprt_put(xprt); 540 } 541 542 /* 543 * Some svc_serv's will have occasional work to do, even when a xprt is not 544 * waiting to be serviced. This function is there to "kick" a task in one of 545 * those services so that it can wake up and do that work. Note that we only 546 * bother with pool 0 as we don't need to wake up more than one thread for 547 * this purpose. 548 */ 549 void svc_wake_up(struct svc_serv *serv) 550 { 551 struct svc_rqst *rqstp; 552 struct svc_pool *pool; 553 554 pool = &serv->sv_pools[0]; 555 556 rcu_read_lock(); 557 list_for_each_entry_rcu(rqstp, &pool->sp_all_threads, rq_all) { 558 /* skip any that aren't queued */ 559 if (test_bit(RQ_BUSY, &rqstp->rq_flags)) 560 continue; 561 rcu_read_unlock(); 562 wake_up_process(rqstp->rq_task); 563 trace_svc_wake_up(rqstp->rq_task->pid); 564 return; 565 } 566 rcu_read_unlock(); 567 568 /* No free entries available */ 569 set_bit(SP_TASK_PENDING, &pool->sp_flags); 570 smp_wmb(); 571 trace_svc_wake_up(0); 572 } 573 EXPORT_SYMBOL_GPL(svc_wake_up); 574 575 int svc_port_is_privileged(struct sockaddr *sin) 576 { 577 switch (sin->sa_family) { 578 case AF_INET: 579 return ntohs(((struct sockaddr_in *)sin)->sin_port) 580 < PROT_SOCK; 581 case AF_INET6: 582 return ntohs(((struct sockaddr_in6 *)sin)->sin6_port) 583 < PROT_SOCK; 584 default: 585 return 0; 586 } 587 } 588 589 /* 590 * Make sure that we don't have too many active connections. If we have, 591 * something must be dropped. It's not clear what will happen if we allow 592 * "too many" connections, but when dealing with network-facing software, 593 * we have to code defensively. Here we do that by imposing hard limits. 594 * 595 * There's no point in trying to do random drop here for DoS 596 * prevention. The NFS clients does 1 reconnect in 15 seconds. An 597 * attacker can easily beat that. 598 * 599 * The only somewhat efficient mechanism would be if drop old 600 * connections from the same IP first. But right now we don't even 601 * record the client IP in svc_sock. 602 * 603 * single-threaded services that expect a lot of clients will probably 604 * need to set sv_maxconn to override the default value which is based 605 * on the number of threads 606 */ 607 static void svc_check_conn_limits(struct svc_serv *serv) 608 { 609 unsigned int limit = serv->sv_maxconn ? serv->sv_maxconn : 610 (serv->sv_nrthreads+3) * 20; 611 612 if (serv->sv_tmpcnt > limit) { 613 struct svc_xprt *xprt = NULL; 614 spin_lock_bh(&serv->sv_lock); 615 if (!list_empty(&serv->sv_tempsocks)) { 616 /* Try to help the admin */ 617 net_notice_ratelimited("%s: too many open connections, consider increasing the %s\n", 618 serv->sv_name, serv->sv_maxconn ? 619 "max number of connections" : 620 "number of threads"); 621 /* 622 * Always select the oldest connection. It's not fair, 623 * but so is life 624 */ 625 xprt = list_entry(serv->sv_tempsocks.prev, 626 struct svc_xprt, 627 xpt_list); 628 set_bit(XPT_CLOSE, &xprt->xpt_flags); 629 svc_xprt_get(xprt); 630 } 631 spin_unlock_bh(&serv->sv_lock); 632 633 if (xprt) { 634 svc_xprt_enqueue(xprt); 635 svc_xprt_put(xprt); 636 } 637 } 638 } 639 640 static int svc_alloc_arg(struct svc_rqst *rqstp) 641 { 642 struct svc_serv *serv = rqstp->rq_server; 643 struct xdr_buf *arg; 644 int pages; 645 int i; 646 647 /* now allocate needed pages. If we get a failure, sleep briefly */ 648 pages = (serv->sv_max_mesg + 2 * PAGE_SIZE) >> PAGE_SHIFT; 649 if (pages > RPCSVC_MAXPAGES) { 650 pr_warn_once("svc: warning: pages=%u > RPCSVC_MAXPAGES=%lu\n", 651 pages, RPCSVC_MAXPAGES); 652 /* use as many pages as possible */ 653 pages = RPCSVC_MAXPAGES; 654 } 655 for (i = 0; i < pages ; i++) 656 while (rqstp->rq_pages[i] == NULL) { 657 struct page *p = alloc_page(GFP_KERNEL); 658 if (!p) { 659 set_current_state(TASK_INTERRUPTIBLE); 660 if (signalled() || kthread_should_stop()) { 661 set_current_state(TASK_RUNNING); 662 return -EINTR; 663 } 664 schedule_timeout(msecs_to_jiffies(500)); 665 } 666 rqstp->rq_pages[i] = p; 667 } 668 rqstp->rq_page_end = &rqstp->rq_pages[i]; 669 rqstp->rq_pages[i++] = NULL; /* this might be seen in nfs_read_actor */ 670 671 /* Make arg->head point to first page and arg->pages point to rest */ 672 arg = &rqstp->rq_arg; 673 arg->head[0].iov_base = page_address(rqstp->rq_pages[0]); 674 arg->head[0].iov_len = PAGE_SIZE; 675 arg->pages = rqstp->rq_pages + 1; 676 arg->page_base = 0; 677 /* save at least one page for response */ 678 arg->page_len = (pages-2)*PAGE_SIZE; 679 arg->len = (pages-1)*PAGE_SIZE; 680 arg->tail[0].iov_len = 0; 681 return 0; 682 } 683 684 static bool 685 rqst_should_sleep(struct svc_rqst *rqstp) 686 { 687 struct svc_pool *pool = rqstp->rq_pool; 688 689 /* did someone call svc_wake_up? */ 690 if (test_and_clear_bit(SP_TASK_PENDING, &pool->sp_flags)) 691 return false; 692 693 /* was a socket queued? */ 694 if (!list_empty(&pool->sp_sockets)) 695 return false; 696 697 /* are we shutting down? */ 698 if (signalled() || kthread_should_stop()) 699 return false; 700 701 /* are we freezing? */ 702 if (freezing(current)) 703 return false; 704 705 return true; 706 } 707 708 static struct svc_xprt *svc_get_next_xprt(struct svc_rqst *rqstp, long timeout) 709 { 710 struct svc_pool *pool = rqstp->rq_pool; 711 long time_left = 0; 712 713 /* rq_xprt should be clear on entry */ 714 WARN_ON_ONCE(rqstp->rq_xprt); 715 716 rqstp->rq_xprt = svc_xprt_dequeue(pool); 717 if (rqstp->rq_xprt) 718 goto out_found; 719 720 /* 721 * We have to be able to interrupt this wait 722 * to bring down the daemons ... 723 */ 724 set_current_state(TASK_INTERRUPTIBLE); 725 smp_mb__before_atomic(); 726 clear_bit(SP_CONGESTED, &pool->sp_flags); 727 clear_bit(RQ_BUSY, &rqstp->rq_flags); 728 smp_mb__after_atomic(); 729 730 if (likely(rqst_should_sleep(rqstp))) 731 time_left = schedule_timeout(timeout); 732 else 733 __set_current_state(TASK_RUNNING); 734 735 try_to_freeze(); 736 737 set_bit(RQ_BUSY, &rqstp->rq_flags); 738 smp_mb__after_atomic(); 739 rqstp->rq_xprt = svc_xprt_dequeue(pool); 740 if (rqstp->rq_xprt) 741 goto out_found; 742 743 if (!time_left) 744 atomic_long_inc(&pool->sp_stats.threads_timedout); 745 746 if (signalled() || kthread_should_stop()) 747 return ERR_PTR(-EINTR); 748 return ERR_PTR(-EAGAIN); 749 out_found: 750 /* Normally we will wait up to 5 seconds for any required 751 * cache information to be provided. 752 */ 753 if (!test_bit(SP_CONGESTED, &pool->sp_flags)) 754 rqstp->rq_chandle.thread_wait = 5*HZ; 755 else 756 rqstp->rq_chandle.thread_wait = 1*HZ; 757 trace_svc_xprt_dequeue(rqstp); 758 return rqstp->rq_xprt; 759 } 760 761 static void svc_add_new_temp_xprt(struct svc_serv *serv, struct svc_xprt *newxpt) 762 { 763 spin_lock_bh(&serv->sv_lock); 764 set_bit(XPT_TEMP, &newxpt->xpt_flags); 765 list_add(&newxpt->xpt_list, &serv->sv_tempsocks); 766 serv->sv_tmpcnt++; 767 if (serv->sv_temptimer.function == NULL) { 768 /* setup timer to age temp transports */ 769 serv->sv_temptimer.function = svc_age_temp_xprts; 770 mod_timer(&serv->sv_temptimer, 771 jiffies + svc_conn_age_period * HZ); 772 } 773 spin_unlock_bh(&serv->sv_lock); 774 svc_xprt_received(newxpt); 775 } 776 777 static int svc_handle_xprt(struct svc_rqst *rqstp, struct svc_xprt *xprt) 778 { 779 struct svc_serv *serv = rqstp->rq_server; 780 int len = 0; 781 782 if (test_bit(XPT_CLOSE, &xprt->xpt_flags)) { 783 dprintk("svc_recv: found XPT_CLOSE\n"); 784 if (test_and_clear_bit(XPT_KILL_TEMP, &xprt->xpt_flags)) 785 xprt->xpt_ops->xpo_kill_temp_xprt(xprt); 786 svc_delete_xprt(xprt); 787 /* Leave XPT_BUSY set on the dead xprt: */ 788 goto out; 789 } 790 if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) { 791 struct svc_xprt *newxpt; 792 /* 793 * We know this module_get will succeed because the 794 * listener holds a reference too 795 */ 796 __module_get(xprt->xpt_class->xcl_owner); 797 svc_check_conn_limits(xprt->xpt_server); 798 newxpt = xprt->xpt_ops->xpo_accept(xprt); 799 if (newxpt) { 800 newxpt->xpt_cred = get_cred(xprt->xpt_cred); 801 svc_add_new_temp_xprt(serv, newxpt); 802 } else 803 module_put(xprt->xpt_class->xcl_owner); 804 } else if (svc_xprt_reserve_slot(rqstp, xprt)) { 805 /* XPT_DATA|XPT_DEFERRED case: */ 806 dprintk("svc: server %p, pool %u, transport %p, inuse=%d\n", 807 rqstp, rqstp->rq_pool->sp_id, xprt, 808 kref_read(&xprt->xpt_ref)); 809 rqstp->rq_deferred = svc_deferred_dequeue(xprt); 810 if (rqstp->rq_deferred) 811 len = svc_deferred_recv(rqstp); 812 else 813 len = xprt->xpt_ops->xpo_recvfrom(rqstp); 814 if (len > 0) 815 trace_svc_recvfrom(&rqstp->rq_arg); 816 rqstp->rq_stime = ktime_get(); 817 rqstp->rq_reserved = serv->sv_max_mesg; 818 atomic_add(rqstp->rq_reserved, &xprt->xpt_reserved); 819 } 820 /* clear XPT_BUSY: */ 821 svc_xprt_received(xprt); 822 out: 823 trace_svc_handle_xprt(xprt, len); 824 return len; 825 } 826 827 /* 828 * Receive the next request on any transport. This code is carefully 829 * organised not to touch any cachelines in the shared svc_serv 830 * structure, only cachelines in the local svc_pool. 831 */ 832 int svc_recv(struct svc_rqst *rqstp, long timeout) 833 { 834 struct svc_xprt *xprt = NULL; 835 struct svc_serv *serv = rqstp->rq_server; 836 int len, err; 837 838 dprintk("svc: server %p waiting for data (to = %ld)\n", 839 rqstp, timeout); 840 841 if (rqstp->rq_xprt) 842 printk(KERN_ERR 843 "svc_recv: service %p, transport not NULL!\n", 844 rqstp); 845 846 err = svc_alloc_arg(rqstp); 847 if (err) 848 goto out; 849 850 try_to_freeze(); 851 cond_resched(); 852 err = -EINTR; 853 if (signalled() || kthread_should_stop()) 854 goto out; 855 856 xprt = svc_get_next_xprt(rqstp, timeout); 857 if (IS_ERR(xprt)) { 858 err = PTR_ERR(xprt); 859 goto out; 860 } 861 862 len = svc_handle_xprt(rqstp, xprt); 863 864 /* No data, incomplete (TCP) read, or accept() */ 865 err = -EAGAIN; 866 if (len <= 0) 867 goto out_release; 868 869 clear_bit(XPT_OLD, &xprt->xpt_flags); 870 871 xprt->xpt_ops->xpo_secure_port(rqstp); 872 rqstp->rq_chandle.defer = svc_defer; 873 rqstp->rq_xid = svc_getu32(&rqstp->rq_arg.head[0]); 874 875 if (serv->sv_stats) 876 serv->sv_stats->netcnt++; 877 trace_svc_recv(rqstp, len); 878 return len; 879 out_release: 880 rqstp->rq_res.len = 0; 881 svc_xprt_release(rqstp); 882 out: 883 return err; 884 } 885 EXPORT_SYMBOL_GPL(svc_recv); 886 887 /* 888 * Drop request 889 */ 890 void svc_drop(struct svc_rqst *rqstp) 891 { 892 trace_svc_drop(rqstp); 893 dprintk("svc: xprt %p dropped request\n", rqstp->rq_xprt); 894 svc_xprt_release(rqstp); 895 } 896 EXPORT_SYMBOL_GPL(svc_drop); 897 898 /* 899 * Return reply to client. 900 */ 901 int svc_send(struct svc_rqst *rqstp) 902 { 903 struct svc_xprt *xprt; 904 int len = -EFAULT; 905 struct xdr_buf *xb; 906 907 xprt = rqstp->rq_xprt; 908 if (!xprt) 909 goto out; 910 911 /* release the receive skb before sending the reply */ 912 xprt->xpt_ops->xpo_release_rqst(rqstp); 913 914 /* calculate over-all length */ 915 xb = &rqstp->rq_res; 916 xb->len = xb->head[0].iov_len + 917 xb->page_len + 918 xb->tail[0].iov_len; 919 trace_svc_sendto(xb); 920 921 /* Grab mutex to serialize outgoing data. */ 922 mutex_lock(&xprt->xpt_mutex); 923 trace_svc_stats_latency(rqstp); 924 if (test_bit(XPT_DEAD, &xprt->xpt_flags) 925 || test_bit(XPT_CLOSE, &xprt->xpt_flags)) 926 len = -ENOTCONN; 927 else 928 len = xprt->xpt_ops->xpo_sendto(rqstp); 929 mutex_unlock(&xprt->xpt_mutex); 930 trace_svc_send(rqstp, len); 931 svc_xprt_release(rqstp); 932 933 if (len == -ECONNREFUSED || len == -ENOTCONN || len == -EAGAIN) 934 len = 0; 935 out: 936 return len; 937 } 938 939 /* 940 * Timer function to close old temporary transports, using 941 * a mark-and-sweep algorithm. 942 */ 943 static void svc_age_temp_xprts(struct timer_list *t) 944 { 945 struct svc_serv *serv = from_timer(serv, t, sv_temptimer); 946 struct svc_xprt *xprt; 947 struct list_head *le, *next; 948 949 dprintk("svc_age_temp_xprts\n"); 950 951 if (!spin_trylock_bh(&serv->sv_lock)) { 952 /* busy, try again 1 sec later */ 953 dprintk("svc_age_temp_xprts: busy\n"); 954 mod_timer(&serv->sv_temptimer, jiffies + HZ); 955 return; 956 } 957 958 list_for_each_safe(le, next, &serv->sv_tempsocks) { 959 xprt = list_entry(le, struct svc_xprt, xpt_list); 960 961 /* First time through, just mark it OLD. Second time 962 * through, close it. */ 963 if (!test_and_set_bit(XPT_OLD, &xprt->xpt_flags)) 964 continue; 965 if (kref_read(&xprt->xpt_ref) > 1 || 966 test_bit(XPT_BUSY, &xprt->xpt_flags)) 967 continue; 968 list_del_init(le); 969 set_bit(XPT_CLOSE, &xprt->xpt_flags); 970 dprintk("queuing xprt %p for closing\n", xprt); 971 972 /* a thread will dequeue and close it soon */ 973 svc_xprt_enqueue(xprt); 974 } 975 spin_unlock_bh(&serv->sv_lock); 976 977 mod_timer(&serv->sv_temptimer, jiffies + svc_conn_age_period * HZ); 978 } 979 980 /* Close temporary transports whose xpt_local matches server_addr immediately 981 * instead of waiting for them to be picked up by the timer. 982 * 983 * This is meant to be called from a notifier_block that runs when an ip 984 * address is deleted. 985 */ 986 void svc_age_temp_xprts_now(struct svc_serv *serv, struct sockaddr *server_addr) 987 { 988 struct svc_xprt *xprt; 989 struct list_head *le, *next; 990 LIST_HEAD(to_be_closed); 991 992 spin_lock_bh(&serv->sv_lock); 993 list_for_each_safe(le, next, &serv->sv_tempsocks) { 994 xprt = list_entry(le, struct svc_xprt, xpt_list); 995 if (rpc_cmp_addr(server_addr, (struct sockaddr *) 996 &xprt->xpt_local)) { 997 dprintk("svc_age_temp_xprts_now: found %p\n", xprt); 998 list_move(le, &to_be_closed); 999 } 1000 } 1001 spin_unlock_bh(&serv->sv_lock); 1002 1003 while (!list_empty(&to_be_closed)) { 1004 le = to_be_closed.next; 1005 list_del_init(le); 1006 xprt = list_entry(le, struct svc_xprt, xpt_list); 1007 set_bit(XPT_CLOSE, &xprt->xpt_flags); 1008 set_bit(XPT_KILL_TEMP, &xprt->xpt_flags); 1009 dprintk("svc_age_temp_xprts_now: queuing xprt %p for closing\n", 1010 xprt); 1011 svc_xprt_enqueue(xprt); 1012 } 1013 } 1014 EXPORT_SYMBOL_GPL(svc_age_temp_xprts_now); 1015 1016 static void call_xpt_users(struct svc_xprt *xprt) 1017 { 1018 struct svc_xpt_user *u; 1019 1020 spin_lock(&xprt->xpt_lock); 1021 while (!list_empty(&xprt->xpt_users)) { 1022 u = list_first_entry(&xprt->xpt_users, struct svc_xpt_user, list); 1023 list_del_init(&u->list); 1024 u->callback(u); 1025 } 1026 spin_unlock(&xprt->xpt_lock); 1027 } 1028 1029 /* 1030 * Remove a dead transport 1031 */ 1032 static void svc_delete_xprt(struct svc_xprt *xprt) 1033 { 1034 struct svc_serv *serv = xprt->xpt_server; 1035 struct svc_deferred_req *dr; 1036 1037 /* Only do this once */ 1038 if (test_and_set_bit(XPT_DEAD, &xprt->xpt_flags)) 1039 BUG(); 1040 1041 dprintk("svc: svc_delete_xprt(%p)\n", xprt); 1042 xprt->xpt_ops->xpo_detach(xprt); 1043 1044 spin_lock_bh(&serv->sv_lock); 1045 list_del_init(&xprt->xpt_list); 1046 WARN_ON_ONCE(!list_empty(&xprt->xpt_ready)); 1047 if (test_bit(XPT_TEMP, &xprt->xpt_flags)) 1048 serv->sv_tmpcnt--; 1049 spin_unlock_bh(&serv->sv_lock); 1050 1051 while ((dr = svc_deferred_dequeue(xprt)) != NULL) 1052 kfree(dr); 1053 1054 call_xpt_users(xprt); 1055 svc_xprt_put(xprt); 1056 } 1057 1058 void svc_close_xprt(struct svc_xprt *xprt) 1059 { 1060 set_bit(XPT_CLOSE, &xprt->xpt_flags); 1061 if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags)) 1062 /* someone else will have to effect the close */ 1063 return; 1064 /* 1065 * We expect svc_close_xprt() to work even when no threads are 1066 * running (e.g., while configuring the server before starting 1067 * any threads), so if the transport isn't busy, we delete 1068 * it ourself: 1069 */ 1070 svc_delete_xprt(xprt); 1071 } 1072 EXPORT_SYMBOL_GPL(svc_close_xprt); 1073 1074 static int svc_close_list(struct svc_serv *serv, struct list_head *xprt_list, struct net *net) 1075 { 1076 struct svc_xprt *xprt; 1077 int ret = 0; 1078 1079 spin_lock(&serv->sv_lock); 1080 list_for_each_entry(xprt, xprt_list, xpt_list) { 1081 if (xprt->xpt_net != net) 1082 continue; 1083 ret++; 1084 set_bit(XPT_CLOSE, &xprt->xpt_flags); 1085 svc_xprt_enqueue(xprt); 1086 } 1087 spin_unlock(&serv->sv_lock); 1088 return ret; 1089 } 1090 1091 static struct svc_xprt *svc_dequeue_net(struct svc_serv *serv, struct net *net) 1092 { 1093 struct svc_pool *pool; 1094 struct svc_xprt *xprt; 1095 struct svc_xprt *tmp; 1096 int i; 1097 1098 for (i = 0; i < serv->sv_nrpools; i++) { 1099 pool = &serv->sv_pools[i]; 1100 1101 spin_lock_bh(&pool->sp_lock); 1102 list_for_each_entry_safe(xprt, tmp, &pool->sp_sockets, xpt_ready) { 1103 if (xprt->xpt_net != net) 1104 continue; 1105 list_del_init(&xprt->xpt_ready); 1106 spin_unlock_bh(&pool->sp_lock); 1107 return xprt; 1108 } 1109 spin_unlock_bh(&pool->sp_lock); 1110 } 1111 return NULL; 1112 } 1113 1114 static void svc_clean_up_xprts(struct svc_serv *serv, struct net *net) 1115 { 1116 struct svc_xprt *xprt; 1117 1118 while ((xprt = svc_dequeue_net(serv, net))) { 1119 set_bit(XPT_CLOSE, &xprt->xpt_flags); 1120 svc_delete_xprt(xprt); 1121 } 1122 } 1123 1124 /* 1125 * Server threads may still be running (especially in the case where the 1126 * service is still running in other network namespaces). 1127 * 1128 * So we shut down sockets the same way we would on a running server, by 1129 * setting XPT_CLOSE, enqueuing, and letting a thread pick it up to do 1130 * the close. In the case there are no such other threads, 1131 * threads running, svc_clean_up_xprts() does a simple version of a 1132 * server's main event loop, and in the case where there are other 1133 * threads, we may need to wait a little while and then check again to 1134 * see if they're done. 1135 */ 1136 void svc_close_net(struct svc_serv *serv, struct net *net) 1137 { 1138 int delay = 0; 1139 1140 while (svc_close_list(serv, &serv->sv_permsocks, net) + 1141 svc_close_list(serv, &serv->sv_tempsocks, net)) { 1142 1143 svc_clean_up_xprts(serv, net); 1144 msleep(delay++); 1145 } 1146 } 1147 1148 /* 1149 * Handle defer and revisit of requests 1150 */ 1151 1152 static void svc_revisit(struct cache_deferred_req *dreq, int too_many) 1153 { 1154 struct svc_deferred_req *dr = 1155 container_of(dreq, struct svc_deferred_req, handle); 1156 struct svc_xprt *xprt = dr->xprt; 1157 1158 spin_lock(&xprt->xpt_lock); 1159 set_bit(XPT_DEFERRED, &xprt->xpt_flags); 1160 if (too_many || test_bit(XPT_DEAD, &xprt->xpt_flags)) { 1161 spin_unlock(&xprt->xpt_lock); 1162 dprintk("revisit canceled\n"); 1163 svc_xprt_put(xprt); 1164 trace_svc_drop_deferred(dr); 1165 kfree(dr); 1166 return; 1167 } 1168 dprintk("revisit queued\n"); 1169 dr->xprt = NULL; 1170 list_add(&dr->handle.recent, &xprt->xpt_deferred); 1171 spin_unlock(&xprt->xpt_lock); 1172 svc_xprt_enqueue(xprt); 1173 svc_xprt_put(xprt); 1174 } 1175 1176 /* 1177 * Save the request off for later processing. The request buffer looks 1178 * like this: 1179 * 1180 * <xprt-header><rpc-header><rpc-pagelist><rpc-tail> 1181 * 1182 * This code can only handle requests that consist of an xprt-header 1183 * and rpc-header. 1184 */ 1185 static struct cache_deferred_req *svc_defer(struct cache_req *req) 1186 { 1187 struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle); 1188 struct svc_deferred_req *dr; 1189 1190 if (rqstp->rq_arg.page_len || !test_bit(RQ_USEDEFERRAL, &rqstp->rq_flags)) 1191 return NULL; /* if more than a page, give up FIXME */ 1192 if (rqstp->rq_deferred) { 1193 dr = rqstp->rq_deferred; 1194 rqstp->rq_deferred = NULL; 1195 } else { 1196 size_t skip; 1197 size_t size; 1198 /* FIXME maybe discard if size too large */ 1199 size = sizeof(struct svc_deferred_req) + rqstp->rq_arg.len; 1200 dr = kmalloc(size, GFP_KERNEL); 1201 if (dr == NULL) 1202 return NULL; 1203 1204 dr->handle.owner = rqstp->rq_server; 1205 dr->prot = rqstp->rq_prot; 1206 memcpy(&dr->addr, &rqstp->rq_addr, rqstp->rq_addrlen); 1207 dr->addrlen = rqstp->rq_addrlen; 1208 dr->daddr = rqstp->rq_daddr; 1209 dr->argslen = rqstp->rq_arg.len >> 2; 1210 dr->xprt_hlen = rqstp->rq_xprt_hlen; 1211 1212 /* back up head to the start of the buffer and copy */ 1213 skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len; 1214 memcpy(dr->args, rqstp->rq_arg.head[0].iov_base - skip, 1215 dr->argslen << 2); 1216 } 1217 svc_xprt_get(rqstp->rq_xprt); 1218 dr->xprt = rqstp->rq_xprt; 1219 set_bit(RQ_DROPME, &rqstp->rq_flags); 1220 1221 dr->handle.revisit = svc_revisit; 1222 trace_svc_defer(rqstp); 1223 return &dr->handle; 1224 } 1225 1226 /* 1227 * recv data from a deferred request into an active one 1228 */ 1229 static int svc_deferred_recv(struct svc_rqst *rqstp) 1230 { 1231 struct svc_deferred_req *dr = rqstp->rq_deferred; 1232 1233 /* setup iov_base past transport header */ 1234 rqstp->rq_arg.head[0].iov_base = dr->args + (dr->xprt_hlen>>2); 1235 /* The iov_len does not include the transport header bytes */ 1236 rqstp->rq_arg.head[0].iov_len = (dr->argslen<<2) - dr->xprt_hlen; 1237 rqstp->rq_arg.page_len = 0; 1238 /* The rq_arg.len includes the transport header bytes */ 1239 rqstp->rq_arg.len = dr->argslen<<2; 1240 rqstp->rq_prot = dr->prot; 1241 memcpy(&rqstp->rq_addr, &dr->addr, dr->addrlen); 1242 rqstp->rq_addrlen = dr->addrlen; 1243 /* Save off transport header len in case we get deferred again */ 1244 rqstp->rq_xprt_hlen = dr->xprt_hlen; 1245 rqstp->rq_daddr = dr->daddr; 1246 rqstp->rq_respages = rqstp->rq_pages; 1247 return (dr->argslen<<2) - dr->xprt_hlen; 1248 } 1249 1250 1251 static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt) 1252 { 1253 struct svc_deferred_req *dr = NULL; 1254 1255 if (!test_bit(XPT_DEFERRED, &xprt->xpt_flags)) 1256 return NULL; 1257 spin_lock(&xprt->xpt_lock); 1258 if (!list_empty(&xprt->xpt_deferred)) { 1259 dr = list_entry(xprt->xpt_deferred.next, 1260 struct svc_deferred_req, 1261 handle.recent); 1262 list_del_init(&dr->handle.recent); 1263 trace_svc_revisit_deferred(dr); 1264 } else 1265 clear_bit(XPT_DEFERRED, &xprt->xpt_flags); 1266 spin_unlock(&xprt->xpt_lock); 1267 return dr; 1268 } 1269 1270 /** 1271 * svc_find_xprt - find an RPC transport instance 1272 * @serv: pointer to svc_serv to search 1273 * @xcl_name: C string containing transport's class name 1274 * @net: owner net pointer 1275 * @af: Address family of transport's local address 1276 * @port: transport's IP port number 1277 * 1278 * Return the transport instance pointer for the endpoint accepting 1279 * connections/peer traffic from the specified transport class, 1280 * address family and port. 1281 * 1282 * Specifying 0 for the address family or port is effectively a 1283 * wild-card, and will result in matching the first transport in the 1284 * service's list that has a matching class name. 1285 */ 1286 struct svc_xprt *svc_find_xprt(struct svc_serv *serv, const char *xcl_name, 1287 struct net *net, const sa_family_t af, 1288 const unsigned short port) 1289 { 1290 struct svc_xprt *xprt; 1291 struct svc_xprt *found = NULL; 1292 1293 /* Sanity check the args */ 1294 if (serv == NULL || xcl_name == NULL) 1295 return found; 1296 1297 spin_lock_bh(&serv->sv_lock); 1298 list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) { 1299 if (xprt->xpt_net != net) 1300 continue; 1301 if (strcmp(xprt->xpt_class->xcl_name, xcl_name)) 1302 continue; 1303 if (af != AF_UNSPEC && af != xprt->xpt_local.ss_family) 1304 continue; 1305 if (port != 0 && port != svc_xprt_local_port(xprt)) 1306 continue; 1307 found = xprt; 1308 svc_xprt_get(xprt); 1309 break; 1310 } 1311 spin_unlock_bh(&serv->sv_lock); 1312 return found; 1313 } 1314 EXPORT_SYMBOL_GPL(svc_find_xprt); 1315 1316 static int svc_one_xprt_name(const struct svc_xprt *xprt, 1317 char *pos, int remaining) 1318 { 1319 int len; 1320 1321 len = snprintf(pos, remaining, "%s %u\n", 1322 xprt->xpt_class->xcl_name, 1323 svc_xprt_local_port(xprt)); 1324 if (len >= remaining) 1325 return -ENAMETOOLONG; 1326 return len; 1327 } 1328 1329 /** 1330 * svc_xprt_names - format a buffer with a list of transport names 1331 * @serv: pointer to an RPC service 1332 * @buf: pointer to a buffer to be filled in 1333 * @buflen: length of buffer to be filled in 1334 * 1335 * Fills in @buf with a string containing a list of transport names, 1336 * each name terminated with '\n'. 1337 * 1338 * Returns positive length of the filled-in string on success; otherwise 1339 * a negative errno value is returned if an error occurs. 1340 */ 1341 int svc_xprt_names(struct svc_serv *serv, char *buf, const int buflen) 1342 { 1343 struct svc_xprt *xprt; 1344 int len, totlen; 1345 char *pos; 1346 1347 /* Sanity check args */ 1348 if (!serv) 1349 return 0; 1350 1351 spin_lock_bh(&serv->sv_lock); 1352 1353 pos = buf; 1354 totlen = 0; 1355 list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) { 1356 len = svc_one_xprt_name(xprt, pos, buflen - totlen); 1357 if (len < 0) { 1358 *buf = '\0'; 1359 totlen = len; 1360 } 1361 if (len <= 0) 1362 break; 1363 1364 pos += len; 1365 totlen += len; 1366 } 1367 1368 spin_unlock_bh(&serv->sv_lock); 1369 return totlen; 1370 } 1371 EXPORT_SYMBOL_GPL(svc_xprt_names); 1372 1373 1374 /*----------------------------------------------------------------------------*/ 1375 1376 static void *svc_pool_stats_start(struct seq_file *m, loff_t *pos) 1377 { 1378 unsigned int pidx = (unsigned int)*pos; 1379 struct svc_serv *serv = m->private; 1380 1381 dprintk("svc_pool_stats_start, *pidx=%u\n", pidx); 1382 1383 if (!pidx) 1384 return SEQ_START_TOKEN; 1385 return (pidx > serv->sv_nrpools ? NULL : &serv->sv_pools[pidx-1]); 1386 } 1387 1388 static void *svc_pool_stats_next(struct seq_file *m, void *p, loff_t *pos) 1389 { 1390 struct svc_pool *pool = p; 1391 struct svc_serv *serv = m->private; 1392 1393 dprintk("svc_pool_stats_next, *pos=%llu\n", *pos); 1394 1395 if (p == SEQ_START_TOKEN) { 1396 pool = &serv->sv_pools[0]; 1397 } else { 1398 unsigned int pidx = (pool - &serv->sv_pools[0]); 1399 if (pidx < serv->sv_nrpools-1) 1400 pool = &serv->sv_pools[pidx+1]; 1401 else 1402 pool = NULL; 1403 } 1404 ++*pos; 1405 return pool; 1406 } 1407 1408 static void svc_pool_stats_stop(struct seq_file *m, void *p) 1409 { 1410 } 1411 1412 static int svc_pool_stats_show(struct seq_file *m, void *p) 1413 { 1414 struct svc_pool *pool = p; 1415 1416 if (p == SEQ_START_TOKEN) { 1417 seq_puts(m, "# pool packets-arrived sockets-enqueued threads-woken threads-timedout\n"); 1418 return 0; 1419 } 1420 1421 seq_printf(m, "%u %lu %lu %lu %lu\n", 1422 pool->sp_id, 1423 (unsigned long)atomic_long_read(&pool->sp_stats.packets), 1424 pool->sp_stats.sockets_queued, 1425 (unsigned long)atomic_long_read(&pool->sp_stats.threads_woken), 1426 (unsigned long)atomic_long_read(&pool->sp_stats.threads_timedout)); 1427 1428 return 0; 1429 } 1430 1431 static const struct seq_operations svc_pool_stats_seq_ops = { 1432 .start = svc_pool_stats_start, 1433 .next = svc_pool_stats_next, 1434 .stop = svc_pool_stats_stop, 1435 .show = svc_pool_stats_show, 1436 }; 1437 1438 int svc_pool_stats_open(struct svc_serv *serv, struct file *file) 1439 { 1440 int err; 1441 1442 err = seq_open(file, &svc_pool_stats_seq_ops); 1443 if (!err) 1444 ((struct seq_file *) file->private_data)->private = serv; 1445 return err; 1446 } 1447 EXPORT_SYMBOL(svc_pool_stats_open); 1448 1449 /*----------------------------------------------------------------------------*/ 1450