xprt.c revision 4cfc7e6019caa3e97d2a81c48c8d575d7b38d751
1/* 2 * linux/net/sunrpc/xprt.c 3 * 4 * This is a generic RPC call interface supporting congestion avoidance, 5 * and asynchronous calls. 6 * 7 * The interface works like this: 8 * 9 * - When a process places a call, it allocates a request slot if 10 * one is available. Otherwise, it sleeps on the backlog queue 11 * (xprt_reserve). 12 * - Next, the caller puts together the RPC message, stuffs it into 13 * the request struct, and calls xprt_transmit(). 14 * - xprt_transmit sends the message and installs the caller on the 15 * transport's wait list. At the same time, if a reply is expected, 16 * it installs a timer that is run after the packet's timeout has 17 * expired. 18 * - When a packet arrives, the data_ready handler walks the list of 19 * pending requests for that transport. If a matching XID is found, the 20 * caller is woken up, and the timer removed. 21 * - When no reply arrives within the timeout interval, the timer is 22 * fired by the kernel and runs xprt_timer(). It either adjusts the 23 * timeout values (minor timeout) or wakes up the caller with a status 24 * of -ETIMEDOUT. 25 * - When the caller receives a notification from RPC that a reply arrived, 26 * it should release the RPC slot, and process the reply. 27 * If the call timed out, it may choose to retry the operation by 28 * adjusting the initial timeout value, and simply calling rpc_call 29 * again. 30 * 31 * Support for async RPC is done through a set of RPC-specific scheduling 32 * primitives that `transparently' work for processes as well as async 33 * tasks that rely on callbacks. 34 * 35 * Copyright (C) 1995-1997, Olaf Kirch <okir@monad.swb.de> 36 * 37 * Transport switch API copyright (C) 2005, Chuck Lever <cel@netapp.com> 38 */ 39 40#include <linux/module.h> 41 42#include <linux/types.h> 43#include <linux/interrupt.h> 44#include <linux/workqueue.h> 45#include <linux/net.h> 46 47#include <linux/sunrpc/clnt.h> 48#include <linux/sunrpc/metrics.h> 49 50#include "sunrpc.h" 51 52/* 53 * Local variables 54 */ 55 56#ifdef RPC_DEBUG 57# define RPCDBG_FACILITY RPCDBG_XPRT 58#endif 59 60/* 61 * Local functions 62 */ 63static void xprt_request_init(struct rpc_task *, struct rpc_xprt *); 64static inline void do_xprt_reserve(struct rpc_task *); 65static void xprt_connect_status(struct rpc_task *task); 66static int __xprt_get_cong(struct rpc_xprt *, struct rpc_task *); 67 68static DEFINE_SPINLOCK(xprt_list_lock); 69static LIST_HEAD(xprt_list); 70 71/* 72 * The transport code maintains an estimate on the maximum number of out- 73 * standing RPC requests, using a smoothed version of the congestion 74 * avoidance implemented in 44BSD. This is basically the Van Jacobson 75 * congestion algorithm: If a retransmit occurs, the congestion window is 76 * halved; otherwise, it is incremented by 1/cwnd when 77 * 78 * - a reply is received and 79 * - a full number of requests are outstanding and 80 * - the congestion window hasn't been updated recently. 81 */ 82#define RPC_CWNDSHIFT (8U) 83#define RPC_CWNDSCALE (1U << RPC_CWNDSHIFT) 84#define RPC_INITCWND RPC_CWNDSCALE 85#define RPC_MAXCWND(xprt) ((xprt)->max_reqs << RPC_CWNDSHIFT) 86 87#define RPCXPRT_CONGESTED(xprt) ((xprt)->cong >= (xprt)->cwnd) 88 89/** 90 * xprt_register_transport - register a transport implementation 91 * @transport: transport to register 92 * 93 * If a transport implementation is loaded as a kernel module, it can 94 * call this interface to make itself known to the RPC client. 95 * 96 * Returns: 97 * 0: transport successfully registered 98 * -EEXIST: transport already registered 99 * -EINVAL: transport module being unloaded 100 */ 101int xprt_register_transport(struct xprt_class *transport) 102{ 103 struct xprt_class *t; 104 int result; 105 106 result = -EEXIST; 107 spin_lock(&xprt_list_lock); 108 list_for_each_entry(t, &xprt_list, list) { 109 /* don't register the same transport class twice */ 110 if (t->ident == transport->ident) 111 goto out; 112 } 113 114 list_add_tail(&transport->list, &xprt_list); 115 printk(KERN_INFO "RPC: Registered %s transport module.\n", 116 transport->name); 117 result = 0; 118 119out: 120 spin_unlock(&xprt_list_lock); 121 return result; 122} 123EXPORT_SYMBOL_GPL(xprt_register_transport); 124 125/** 126 * xprt_unregister_transport - unregister a transport implementation 127 * @transport: transport to unregister 128 * 129 * Returns: 130 * 0: transport successfully unregistered 131 * -ENOENT: transport never registered 132 */ 133int xprt_unregister_transport(struct xprt_class *transport) 134{ 135 struct xprt_class *t; 136 int result; 137 138 result = 0; 139 spin_lock(&xprt_list_lock); 140 list_for_each_entry(t, &xprt_list, list) { 141 if (t == transport) { 142 printk(KERN_INFO 143 "RPC: Unregistered %s transport module.\n", 144 transport->name); 145 list_del_init(&transport->list); 146 goto out; 147 } 148 } 149 result = -ENOENT; 150 151out: 152 spin_unlock(&xprt_list_lock); 153 return result; 154} 155EXPORT_SYMBOL_GPL(xprt_unregister_transport); 156 157/** 158 * xprt_load_transport - load a transport implementation 159 * @transport_name: transport to load 160 * 161 * Returns: 162 * 0: transport successfully loaded 163 * -ENOENT: transport module not available 164 */ 165int xprt_load_transport(const char *transport_name) 166{ 167 struct xprt_class *t; 168 char module_name[sizeof t->name + 5]; 169 int result; 170 171 result = 0; 172 spin_lock(&xprt_list_lock); 173 list_for_each_entry(t, &xprt_list, list) { 174 if (strcmp(t->name, transport_name) == 0) { 175 spin_unlock(&xprt_list_lock); 176 goto out; 177 } 178 } 179 spin_unlock(&xprt_list_lock); 180 strcpy(module_name, "xprt"); 181 strncat(module_name, transport_name, sizeof t->name); 182 result = request_module(module_name); 183out: 184 return result; 185} 186EXPORT_SYMBOL_GPL(xprt_load_transport); 187 188/** 189 * xprt_reserve_xprt - serialize write access to transports 190 * @task: task that is requesting access to the transport 191 * 192 * This prevents mixing the payload of separate requests, and prevents 193 * transport connects from colliding with writes. No congestion control 194 * is provided. 195 */ 196int xprt_reserve_xprt(struct rpc_task *task) 197{ 198 struct rpc_rqst *req = task->tk_rqstp; 199 struct rpc_xprt *xprt = req->rq_xprt; 200 201 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) { 202 if (task == xprt->snd_task) 203 return 1; 204 if (task == NULL) 205 return 0; 206 goto out_sleep; 207 } 208 xprt->snd_task = task; 209 if (req) { 210 req->rq_bytes_sent = 0; 211 req->rq_ntrans++; 212 } 213 return 1; 214 215out_sleep: 216 dprintk("RPC: %5u failed to lock transport %p\n", 217 task->tk_pid, xprt); 218 task->tk_timeout = 0; 219 task->tk_status = -EAGAIN; 220 if (req && req->rq_ntrans) 221 rpc_sleep_on(&xprt->resend, task, NULL); 222 else 223 rpc_sleep_on(&xprt->sending, task, NULL); 224 return 0; 225} 226EXPORT_SYMBOL_GPL(xprt_reserve_xprt); 227 228static void xprt_clear_locked(struct rpc_xprt *xprt) 229{ 230 xprt->snd_task = NULL; 231 if (!test_bit(XPRT_CLOSE_WAIT, &xprt->state) || xprt->shutdown) { 232 smp_mb__before_clear_bit(); 233 clear_bit(XPRT_LOCKED, &xprt->state); 234 smp_mb__after_clear_bit(); 235 } else 236 queue_work(rpciod_workqueue, &xprt->task_cleanup); 237} 238 239/* 240 * xprt_reserve_xprt_cong - serialize write access to transports 241 * @task: task that is requesting access to the transport 242 * 243 * Same as xprt_reserve_xprt, but Van Jacobson congestion control is 244 * integrated into the decision of whether a request is allowed to be 245 * woken up and given access to the transport. 246 */ 247int xprt_reserve_xprt_cong(struct rpc_task *task) 248{ 249 struct rpc_xprt *xprt = task->tk_xprt; 250 struct rpc_rqst *req = task->tk_rqstp; 251 252 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) { 253 if (task == xprt->snd_task) 254 return 1; 255 goto out_sleep; 256 } 257 if (__xprt_get_cong(xprt, task)) { 258 xprt->snd_task = task; 259 if (req) { 260 req->rq_bytes_sent = 0; 261 req->rq_ntrans++; 262 } 263 return 1; 264 } 265 xprt_clear_locked(xprt); 266out_sleep: 267 dprintk("RPC: %5u failed to lock transport %p\n", task->tk_pid, xprt); 268 task->tk_timeout = 0; 269 task->tk_status = -EAGAIN; 270 if (req && req->rq_ntrans) 271 rpc_sleep_on(&xprt->resend, task, NULL); 272 else 273 rpc_sleep_on(&xprt->sending, task, NULL); 274 return 0; 275} 276EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong); 277 278static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task) 279{ 280 int retval; 281 282 spin_lock_bh(&xprt->transport_lock); 283 retval = xprt->ops->reserve_xprt(task); 284 spin_unlock_bh(&xprt->transport_lock); 285 return retval; 286} 287 288static void __xprt_lock_write_next(struct rpc_xprt *xprt) 289{ 290 struct rpc_task *task; 291 struct rpc_rqst *req; 292 293 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) 294 return; 295 296 task = rpc_wake_up_next(&xprt->resend); 297 if (!task) { 298 task = rpc_wake_up_next(&xprt->sending); 299 if (!task) 300 goto out_unlock; 301 } 302 303 req = task->tk_rqstp; 304 xprt->snd_task = task; 305 if (req) { 306 req->rq_bytes_sent = 0; 307 req->rq_ntrans++; 308 } 309 return; 310 311out_unlock: 312 xprt_clear_locked(xprt); 313} 314 315static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt) 316{ 317 struct rpc_task *task; 318 319 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) 320 return; 321 if (RPCXPRT_CONGESTED(xprt)) 322 goto out_unlock; 323 task = rpc_wake_up_next(&xprt->resend); 324 if (!task) { 325 task = rpc_wake_up_next(&xprt->sending); 326 if (!task) 327 goto out_unlock; 328 } 329 if (__xprt_get_cong(xprt, task)) { 330 struct rpc_rqst *req = task->tk_rqstp; 331 xprt->snd_task = task; 332 if (req) { 333 req->rq_bytes_sent = 0; 334 req->rq_ntrans++; 335 } 336 return; 337 } 338out_unlock: 339 xprt_clear_locked(xprt); 340} 341 342/** 343 * xprt_release_xprt - allow other requests to use a transport 344 * @xprt: transport with other tasks potentially waiting 345 * @task: task that is releasing access to the transport 346 * 347 * Note that "task" can be NULL. No congestion control is provided. 348 */ 349void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task) 350{ 351 if (xprt->snd_task == task) { 352 xprt_clear_locked(xprt); 353 __xprt_lock_write_next(xprt); 354 } 355} 356EXPORT_SYMBOL_GPL(xprt_release_xprt); 357 358/** 359 * xprt_release_xprt_cong - allow other requests to use a transport 360 * @xprt: transport with other tasks potentially waiting 361 * @task: task that is releasing access to the transport 362 * 363 * Note that "task" can be NULL. Another task is awoken to use the 364 * transport if the transport's congestion window allows it. 365 */ 366void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task) 367{ 368 if (xprt->snd_task == task) { 369 xprt_clear_locked(xprt); 370 __xprt_lock_write_next_cong(xprt); 371 } 372} 373EXPORT_SYMBOL_GPL(xprt_release_xprt_cong); 374 375static inline void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task) 376{ 377 spin_lock_bh(&xprt->transport_lock); 378 xprt->ops->release_xprt(xprt, task); 379 spin_unlock_bh(&xprt->transport_lock); 380} 381 382/* 383 * Van Jacobson congestion avoidance. Check if the congestion window 384 * overflowed. Put the task to sleep if this is the case. 385 */ 386static int 387__xprt_get_cong(struct rpc_xprt *xprt, struct rpc_task *task) 388{ 389 struct rpc_rqst *req = task->tk_rqstp; 390 391 if (req->rq_cong) 392 return 1; 393 dprintk("RPC: %5u xprt_cwnd_limited cong = %lu cwnd = %lu\n", 394 task->tk_pid, xprt->cong, xprt->cwnd); 395 if (RPCXPRT_CONGESTED(xprt)) 396 return 0; 397 req->rq_cong = 1; 398 xprt->cong += RPC_CWNDSCALE; 399 return 1; 400} 401 402/* 403 * Adjust the congestion window, and wake up the next task 404 * that has been sleeping due to congestion 405 */ 406static void 407__xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req) 408{ 409 if (!req->rq_cong) 410 return; 411 req->rq_cong = 0; 412 xprt->cong -= RPC_CWNDSCALE; 413 __xprt_lock_write_next_cong(xprt); 414} 415 416/** 417 * xprt_release_rqst_cong - housekeeping when request is complete 418 * @task: RPC request that recently completed 419 * 420 * Useful for transports that require congestion control. 421 */ 422void xprt_release_rqst_cong(struct rpc_task *task) 423{ 424 __xprt_put_cong(task->tk_xprt, task->tk_rqstp); 425} 426EXPORT_SYMBOL_GPL(xprt_release_rqst_cong); 427 428/** 429 * xprt_adjust_cwnd - adjust transport congestion window 430 * @task: recently completed RPC request used to adjust window 431 * @result: result code of completed RPC request 432 * 433 * We use a time-smoothed congestion estimator to avoid heavy oscillation. 434 */ 435void xprt_adjust_cwnd(struct rpc_task *task, int result) 436{ 437 struct rpc_rqst *req = task->tk_rqstp; 438 struct rpc_xprt *xprt = task->tk_xprt; 439 unsigned long cwnd = xprt->cwnd; 440 441 if (result >= 0 && cwnd <= xprt->cong) { 442 /* The (cwnd >> 1) term makes sure 443 * the result gets rounded properly. */ 444 cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd; 445 if (cwnd > RPC_MAXCWND(xprt)) 446 cwnd = RPC_MAXCWND(xprt); 447 __xprt_lock_write_next_cong(xprt); 448 } else if (result == -ETIMEDOUT) { 449 cwnd >>= 1; 450 if (cwnd < RPC_CWNDSCALE) 451 cwnd = RPC_CWNDSCALE; 452 } 453 dprintk("RPC: cong %ld, cwnd was %ld, now %ld\n", 454 xprt->cong, xprt->cwnd, cwnd); 455 xprt->cwnd = cwnd; 456 __xprt_put_cong(xprt, req); 457} 458EXPORT_SYMBOL_GPL(xprt_adjust_cwnd); 459 460/** 461 * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue 462 * @xprt: transport with waiting tasks 463 * @status: result code to plant in each task before waking it 464 * 465 */ 466void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status) 467{ 468 if (status < 0) 469 rpc_wake_up_status(&xprt->pending, status); 470 else 471 rpc_wake_up(&xprt->pending); 472} 473EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks); 474 475/** 476 * xprt_wait_for_buffer_space - wait for transport output buffer to clear 477 * @task: task to be put to sleep 478 * @action: function pointer to be executed after wait 479 */ 480void xprt_wait_for_buffer_space(struct rpc_task *task, rpc_action action) 481{ 482 struct rpc_rqst *req = task->tk_rqstp; 483 struct rpc_xprt *xprt = req->rq_xprt; 484 485 task->tk_timeout = req->rq_timeout; 486 rpc_sleep_on(&xprt->pending, task, action); 487} 488EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space); 489 490/** 491 * xprt_write_space - wake the task waiting for transport output buffer space 492 * @xprt: transport with waiting tasks 493 * 494 * Can be called in a soft IRQ context, so xprt_write_space never sleeps. 495 */ 496void xprt_write_space(struct rpc_xprt *xprt) 497{ 498 if (unlikely(xprt->shutdown)) 499 return; 500 501 spin_lock_bh(&xprt->transport_lock); 502 if (xprt->snd_task) { 503 dprintk("RPC: write space: waking waiting task on " 504 "xprt %p\n", xprt); 505 rpc_wake_up_queued_task(&xprt->pending, xprt->snd_task); 506 } 507 spin_unlock_bh(&xprt->transport_lock); 508} 509EXPORT_SYMBOL_GPL(xprt_write_space); 510 511/** 512 * xprt_set_retrans_timeout_def - set a request's retransmit timeout 513 * @task: task whose timeout is to be set 514 * 515 * Set a request's retransmit timeout based on the transport's 516 * default timeout parameters. Used by transports that don't adjust 517 * the retransmit timeout based on round-trip time estimation. 518 */ 519void xprt_set_retrans_timeout_def(struct rpc_task *task) 520{ 521 task->tk_timeout = task->tk_rqstp->rq_timeout; 522} 523EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_def); 524 525/* 526 * xprt_set_retrans_timeout_rtt - set a request's retransmit timeout 527 * @task: task whose timeout is to be set 528 * 529 * Set a request's retransmit timeout using the RTT estimator. 530 */ 531void xprt_set_retrans_timeout_rtt(struct rpc_task *task) 532{ 533 int timer = task->tk_msg.rpc_proc->p_timer; 534 struct rpc_clnt *clnt = task->tk_client; 535 struct rpc_rtt *rtt = clnt->cl_rtt; 536 struct rpc_rqst *req = task->tk_rqstp; 537 unsigned long max_timeout = clnt->cl_timeout->to_maxval; 538 539 task->tk_timeout = rpc_calc_rto(rtt, timer); 540 task->tk_timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries; 541 if (task->tk_timeout > max_timeout || task->tk_timeout == 0) 542 task->tk_timeout = max_timeout; 543} 544EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_rtt); 545 546static void xprt_reset_majortimeo(struct rpc_rqst *req) 547{ 548 const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout; 549 550 req->rq_majortimeo = req->rq_timeout; 551 if (to->to_exponential) 552 req->rq_majortimeo <<= to->to_retries; 553 else 554 req->rq_majortimeo += to->to_increment * to->to_retries; 555 if (req->rq_majortimeo > to->to_maxval || req->rq_majortimeo == 0) 556 req->rq_majortimeo = to->to_maxval; 557 req->rq_majortimeo += jiffies; 558} 559 560/** 561 * xprt_adjust_timeout - adjust timeout values for next retransmit 562 * @req: RPC request containing parameters to use for the adjustment 563 * 564 */ 565int xprt_adjust_timeout(struct rpc_rqst *req) 566{ 567 struct rpc_xprt *xprt = req->rq_xprt; 568 const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout; 569 int status = 0; 570 571 if (time_before(jiffies, req->rq_majortimeo)) { 572 if (to->to_exponential) 573 req->rq_timeout <<= 1; 574 else 575 req->rq_timeout += to->to_increment; 576 if (to->to_maxval && req->rq_timeout >= to->to_maxval) 577 req->rq_timeout = to->to_maxval; 578 req->rq_retries++; 579 } else { 580 req->rq_timeout = to->to_initval; 581 req->rq_retries = 0; 582 xprt_reset_majortimeo(req); 583 /* Reset the RTT counters == "slow start" */ 584 spin_lock_bh(&xprt->transport_lock); 585 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval); 586 spin_unlock_bh(&xprt->transport_lock); 587 status = -ETIMEDOUT; 588 } 589 590 if (req->rq_timeout == 0) { 591 printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n"); 592 req->rq_timeout = 5 * HZ; 593 } 594 return status; 595} 596 597static void xprt_autoclose(struct work_struct *work) 598{ 599 struct rpc_xprt *xprt = 600 container_of(work, struct rpc_xprt, task_cleanup); 601 602 xprt->ops->close(xprt); 603 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 604 xprt_release_write(xprt, NULL); 605} 606 607/** 608 * xprt_disconnect_done - mark a transport as disconnected 609 * @xprt: transport to flag for disconnect 610 * 611 */ 612void xprt_disconnect_done(struct rpc_xprt *xprt) 613{ 614 dprintk("RPC: disconnected transport %p\n", xprt); 615 spin_lock_bh(&xprt->transport_lock); 616 xprt_clear_connected(xprt); 617 xprt_wake_pending_tasks(xprt, -EAGAIN); 618 spin_unlock_bh(&xprt->transport_lock); 619} 620EXPORT_SYMBOL_GPL(xprt_disconnect_done); 621 622/** 623 * xprt_force_disconnect - force a transport to disconnect 624 * @xprt: transport to disconnect 625 * 626 */ 627void xprt_force_disconnect(struct rpc_xprt *xprt) 628{ 629 /* Don't race with the test_bit() in xprt_clear_locked() */ 630 spin_lock_bh(&xprt->transport_lock); 631 set_bit(XPRT_CLOSE_WAIT, &xprt->state); 632 /* Try to schedule an autoclose RPC call */ 633 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0) 634 queue_work(rpciod_workqueue, &xprt->task_cleanup); 635 xprt_wake_pending_tasks(xprt, -EAGAIN); 636 spin_unlock_bh(&xprt->transport_lock); 637} 638 639/** 640 * xprt_conditional_disconnect - force a transport to disconnect 641 * @xprt: transport to disconnect 642 * @cookie: 'connection cookie' 643 * 644 * This attempts to break the connection if and only if 'cookie' matches 645 * the current transport 'connection cookie'. It ensures that we don't 646 * try to break the connection more than once when we need to retransmit 647 * a batch of RPC requests. 648 * 649 */ 650void xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie) 651{ 652 /* Don't race with the test_bit() in xprt_clear_locked() */ 653 spin_lock_bh(&xprt->transport_lock); 654 if (cookie != xprt->connect_cookie) 655 goto out; 656 if (test_bit(XPRT_CLOSING, &xprt->state) || !xprt_connected(xprt)) 657 goto out; 658 set_bit(XPRT_CLOSE_WAIT, &xprt->state); 659 /* Try to schedule an autoclose RPC call */ 660 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0) 661 queue_work(rpciod_workqueue, &xprt->task_cleanup); 662 xprt_wake_pending_tasks(xprt, -EAGAIN); 663out: 664 spin_unlock_bh(&xprt->transport_lock); 665} 666 667static void 668xprt_init_autodisconnect(unsigned long data) 669{ 670 struct rpc_xprt *xprt = (struct rpc_xprt *)data; 671 672 spin_lock(&xprt->transport_lock); 673 if (!list_empty(&xprt->recv) || xprt->shutdown) 674 goto out_abort; 675 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) 676 goto out_abort; 677 spin_unlock(&xprt->transport_lock); 678 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state); 679 queue_work(rpciod_workqueue, &xprt->task_cleanup); 680 return; 681out_abort: 682 spin_unlock(&xprt->transport_lock); 683} 684 685/** 686 * xprt_connect - schedule a transport connect operation 687 * @task: RPC task that is requesting the connect 688 * 689 */ 690void xprt_connect(struct rpc_task *task) 691{ 692 struct rpc_xprt *xprt = task->tk_xprt; 693 694 dprintk("RPC: %5u xprt_connect xprt %p %s connected\n", task->tk_pid, 695 xprt, (xprt_connected(xprt) ? "is" : "is not")); 696 697 if (!xprt_bound(xprt)) { 698 task->tk_status = -EAGAIN; 699 return; 700 } 701 if (!xprt_lock_write(xprt, task)) 702 return; 703 if (xprt_connected(xprt)) 704 xprt_release_write(xprt, task); 705 else { 706 if (task->tk_rqstp) 707 task->tk_rqstp->rq_bytes_sent = 0; 708 709 task->tk_timeout = xprt->connect_timeout; 710 rpc_sleep_on(&xprt->pending, task, xprt_connect_status); 711 xprt->stat.connect_start = jiffies; 712 xprt->ops->connect(task); 713 } 714 return; 715} 716 717static void xprt_connect_status(struct rpc_task *task) 718{ 719 struct rpc_xprt *xprt = task->tk_xprt; 720 721 if (task->tk_status == 0) { 722 xprt->stat.connect_count++; 723 xprt->stat.connect_time += (long)jiffies - xprt->stat.connect_start; 724 dprintk("RPC: %5u xprt_connect_status: connection established\n", 725 task->tk_pid); 726 return; 727 } 728 729 switch (task->tk_status) { 730 case -EAGAIN: 731 dprintk("RPC: %5u xprt_connect_status: retrying\n", task->tk_pid); 732 break; 733 case -ETIMEDOUT: 734 dprintk("RPC: %5u xprt_connect_status: connect attempt timed " 735 "out\n", task->tk_pid); 736 break; 737 default: 738 dprintk("RPC: %5u xprt_connect_status: error %d connecting to " 739 "server %s\n", task->tk_pid, -task->tk_status, 740 task->tk_client->cl_server); 741 xprt_release_write(xprt, task); 742 task->tk_status = -EIO; 743 } 744} 745 746/** 747 * xprt_lookup_rqst - find an RPC request corresponding to an XID 748 * @xprt: transport on which the original request was transmitted 749 * @xid: RPC XID of incoming reply 750 * 751 */ 752struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid) 753{ 754 struct list_head *pos; 755 756 list_for_each(pos, &xprt->recv) { 757 struct rpc_rqst *entry = list_entry(pos, struct rpc_rqst, rq_list); 758 if (entry->rq_xid == xid) 759 return entry; 760 } 761 762 dprintk("RPC: xprt_lookup_rqst did not find xid %08x\n", 763 ntohl(xid)); 764 xprt->stat.bad_xids++; 765 return NULL; 766} 767EXPORT_SYMBOL_GPL(xprt_lookup_rqst); 768 769/** 770 * xprt_update_rtt - update an RPC client's RTT state after receiving a reply 771 * @task: RPC request that recently completed 772 * 773 */ 774void xprt_update_rtt(struct rpc_task *task) 775{ 776 struct rpc_rqst *req = task->tk_rqstp; 777 struct rpc_rtt *rtt = task->tk_client->cl_rtt; 778 unsigned timer = task->tk_msg.rpc_proc->p_timer; 779 780 if (timer) { 781 if (req->rq_ntrans == 1) 782 rpc_update_rtt(rtt, timer, 783 (long)jiffies - req->rq_xtime); 784 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1); 785 } 786} 787EXPORT_SYMBOL_GPL(xprt_update_rtt); 788 789/** 790 * xprt_complete_rqst - called when reply processing is complete 791 * @task: RPC request that recently completed 792 * @copied: actual number of bytes received from the transport 793 * 794 * Caller holds transport lock. 795 */ 796void xprt_complete_rqst(struct rpc_task *task, int copied) 797{ 798 struct rpc_rqst *req = task->tk_rqstp; 799 struct rpc_xprt *xprt = req->rq_xprt; 800 801 dprintk("RPC: %5u xid %08x complete (%d bytes received)\n", 802 task->tk_pid, ntohl(req->rq_xid), copied); 803 804 xprt->stat.recvs++; 805 task->tk_rtt = (long)jiffies - req->rq_xtime; 806 807 list_del_init(&req->rq_list); 808 req->rq_private_buf.len = copied; 809 /* Ensure all writes are done before we update */ 810 /* req->rq_reply_bytes_recvd */ 811 smp_wmb(); 812 req->rq_reply_bytes_recvd = copied; 813 rpc_wake_up_queued_task(&xprt->pending, task); 814} 815EXPORT_SYMBOL_GPL(xprt_complete_rqst); 816 817static void xprt_timer(struct rpc_task *task) 818{ 819 struct rpc_rqst *req = task->tk_rqstp; 820 struct rpc_xprt *xprt = req->rq_xprt; 821 822 if (task->tk_status != -ETIMEDOUT) 823 return; 824 dprintk("RPC: %5u xprt_timer\n", task->tk_pid); 825 826 spin_lock_bh(&xprt->transport_lock); 827 if (!req->rq_reply_bytes_recvd) { 828 if (xprt->ops->timer) 829 xprt->ops->timer(task); 830 } else 831 task->tk_status = 0; 832 spin_unlock_bh(&xprt->transport_lock); 833} 834 835static inline int xprt_has_timer(struct rpc_xprt *xprt) 836{ 837 return xprt->idle_timeout != 0; 838} 839 840/** 841 * xprt_prepare_transmit - reserve the transport before sending a request 842 * @task: RPC task about to send a request 843 * 844 */ 845int xprt_prepare_transmit(struct rpc_task *task) 846{ 847 struct rpc_rqst *req = task->tk_rqstp; 848 struct rpc_xprt *xprt = req->rq_xprt; 849 int err = 0; 850 851 dprintk("RPC: %5u xprt_prepare_transmit\n", task->tk_pid); 852 853 spin_lock_bh(&xprt->transport_lock); 854 if (req->rq_reply_bytes_recvd && !req->rq_bytes_sent) { 855 err = req->rq_reply_bytes_recvd; 856 goto out_unlock; 857 } 858 if (!xprt->ops->reserve_xprt(task)) 859 err = -EAGAIN; 860out_unlock: 861 spin_unlock_bh(&xprt->transport_lock); 862 return err; 863} 864 865void xprt_end_transmit(struct rpc_task *task) 866{ 867 xprt_release_write(task->tk_rqstp->rq_xprt, task); 868} 869 870/** 871 * xprt_transmit - send an RPC request on a transport 872 * @task: controlling RPC task 873 * 874 * We have to copy the iovec because sendmsg fiddles with its contents. 875 */ 876void xprt_transmit(struct rpc_task *task) 877{ 878 struct rpc_rqst *req = task->tk_rqstp; 879 struct rpc_xprt *xprt = req->rq_xprt; 880 int status; 881 882 dprintk("RPC: %5u xprt_transmit(%u)\n", task->tk_pid, req->rq_slen); 883 884 if (!req->rq_reply_bytes_recvd) { 885 if (list_empty(&req->rq_list) && rpc_reply_expected(task)) { 886 /* 887 * Add to the list only if we're expecting a reply 888 */ 889 spin_lock_bh(&xprt->transport_lock); 890 /* Update the softirq receive buffer */ 891 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, 892 sizeof(req->rq_private_buf)); 893 /* Add request to the receive list */ 894 list_add_tail(&req->rq_list, &xprt->recv); 895 spin_unlock_bh(&xprt->transport_lock); 896 xprt_reset_majortimeo(req); 897 /* Turn off autodisconnect */ 898 del_singleshot_timer_sync(&xprt->timer); 899 } 900 } else if (!req->rq_bytes_sent) 901 return; 902 903 req->rq_connect_cookie = xprt->connect_cookie; 904 req->rq_xtime = jiffies; 905 status = xprt->ops->send_request(task); 906 if (status != 0) { 907 task->tk_status = status; 908 return; 909 } 910 911 dprintk("RPC: %5u xmit complete\n", task->tk_pid); 912 spin_lock_bh(&xprt->transport_lock); 913 914 xprt->ops->set_retrans_timeout(task); 915 916 xprt->stat.sends++; 917 xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs; 918 xprt->stat.bklog_u += xprt->backlog.qlen; 919 920 /* Don't race with disconnect */ 921 if (!xprt_connected(xprt)) 922 task->tk_status = -ENOTCONN; 923 else if (!req->rq_reply_bytes_recvd && rpc_reply_expected(task)) { 924 /* 925 * Sleep on the pending queue since 926 * we're expecting a reply. 927 */ 928 rpc_sleep_on(&xprt->pending, task, xprt_timer); 929 } 930 spin_unlock_bh(&xprt->transport_lock); 931} 932 933static inline void do_xprt_reserve(struct rpc_task *task) 934{ 935 struct rpc_xprt *xprt = task->tk_xprt; 936 937 task->tk_status = 0; 938 if (task->tk_rqstp) 939 return; 940 if (!list_empty(&xprt->free)) { 941 struct rpc_rqst *req = list_entry(xprt->free.next, struct rpc_rqst, rq_list); 942 list_del_init(&req->rq_list); 943 task->tk_rqstp = req; 944 xprt_request_init(task, xprt); 945 return; 946 } 947 dprintk("RPC: waiting for request slot\n"); 948 task->tk_status = -EAGAIN; 949 task->tk_timeout = 0; 950 rpc_sleep_on(&xprt->backlog, task, NULL); 951} 952 953/** 954 * xprt_reserve - allocate an RPC request slot 955 * @task: RPC task requesting a slot allocation 956 * 957 * If no more slots are available, place the task on the transport's 958 * backlog queue. 959 */ 960void xprt_reserve(struct rpc_task *task) 961{ 962 struct rpc_xprt *xprt = task->tk_xprt; 963 964 task->tk_status = -EIO; 965 spin_lock(&xprt->reserve_lock); 966 do_xprt_reserve(task); 967 spin_unlock(&xprt->reserve_lock); 968} 969 970static inline __be32 xprt_alloc_xid(struct rpc_xprt *xprt) 971{ 972 return xprt->xid++; 973} 974 975static inline void xprt_init_xid(struct rpc_xprt *xprt) 976{ 977 xprt->xid = net_random(); 978} 979 980static void xprt_request_init(struct rpc_task *task, struct rpc_xprt *xprt) 981{ 982 struct rpc_rqst *req = task->tk_rqstp; 983 984 req->rq_timeout = task->tk_client->cl_timeout->to_initval; 985 req->rq_task = task; 986 req->rq_xprt = xprt; 987 req->rq_buffer = NULL; 988 req->rq_xid = xprt_alloc_xid(xprt); 989 req->rq_release_snd_buf = NULL; 990 xprt_reset_majortimeo(req); 991 dprintk("RPC: %5u reserved req %p xid %08x\n", task->tk_pid, 992 req, ntohl(req->rq_xid)); 993} 994 995/** 996 * xprt_release - release an RPC request slot 997 * @task: task which is finished with the slot 998 * 999 */ 1000void xprt_release(struct rpc_task *task) 1001{ 1002 struct rpc_xprt *xprt; 1003 struct rpc_rqst *req; 1004 int is_bc_request; 1005 1006 if (!(req = task->tk_rqstp)) 1007 return; 1008 1009 /* Preallocated backchannel request? */ 1010 is_bc_request = bc_prealloc(req); 1011 1012 xprt = req->rq_xprt; 1013 rpc_count_iostats(task); 1014 spin_lock_bh(&xprt->transport_lock); 1015 xprt->ops->release_xprt(xprt, task); 1016 if (xprt->ops->release_request) 1017 xprt->ops->release_request(task); 1018 if (!list_empty(&req->rq_list)) 1019 list_del(&req->rq_list); 1020 xprt->last_used = jiffies; 1021 if (list_empty(&xprt->recv) && xprt_has_timer(xprt)) 1022 mod_timer(&xprt->timer, 1023 xprt->last_used + xprt->idle_timeout); 1024 spin_unlock_bh(&xprt->transport_lock); 1025 if (!bc_prealloc(req)) 1026 xprt->ops->buf_free(req->rq_buffer); 1027 task->tk_rqstp = NULL; 1028 if (req->rq_release_snd_buf) 1029 req->rq_release_snd_buf(req); 1030 1031 /* 1032 * Early exit if this is a backchannel preallocated request. 1033 * There is no need to have it added to the RPC slot list. 1034 */ 1035 if (is_bc_request) 1036 return; 1037 1038 memset(req, 0, sizeof(*req)); /* mark unused */ 1039 1040 dprintk("RPC: %5u release request %p\n", task->tk_pid, req); 1041 1042 spin_lock(&xprt->reserve_lock); 1043 list_add(&req->rq_list, &xprt->free); 1044 rpc_wake_up_next(&xprt->backlog); 1045 spin_unlock(&xprt->reserve_lock); 1046} 1047 1048/** 1049 * xprt_create_transport - create an RPC transport 1050 * @args: rpc transport creation arguments 1051 * 1052 */ 1053struct rpc_xprt *xprt_create_transport(struct xprt_create *args) 1054{ 1055 struct rpc_xprt *xprt; 1056 struct rpc_rqst *req; 1057 struct xprt_class *t; 1058 1059 spin_lock(&xprt_list_lock); 1060 list_for_each_entry(t, &xprt_list, list) { 1061 if (t->ident == args->ident) { 1062 spin_unlock(&xprt_list_lock); 1063 goto found; 1064 } 1065 } 1066 spin_unlock(&xprt_list_lock); 1067 printk(KERN_ERR "RPC: transport (%d) not supported\n", args->ident); 1068 return ERR_PTR(-EIO); 1069 1070found: 1071 xprt = t->setup(args); 1072 if (IS_ERR(xprt)) { 1073 dprintk("RPC: xprt_create_transport: failed, %ld\n", 1074 -PTR_ERR(xprt)); 1075 return xprt; 1076 } 1077 1078 kref_init(&xprt->kref); 1079 spin_lock_init(&xprt->transport_lock); 1080 spin_lock_init(&xprt->reserve_lock); 1081 1082 INIT_LIST_HEAD(&xprt->free); 1083 INIT_LIST_HEAD(&xprt->recv); 1084#if defined(CONFIG_NFS_V4_1) 1085 spin_lock_init(&xprt->bc_pa_lock); 1086 INIT_LIST_HEAD(&xprt->bc_pa_list); 1087#endif /* CONFIG_NFS_V4_1 */ 1088 1089 INIT_WORK(&xprt->task_cleanup, xprt_autoclose); 1090 if (xprt_has_timer(xprt)) 1091 setup_timer(&xprt->timer, xprt_init_autodisconnect, 1092 (unsigned long)xprt); 1093 else 1094 init_timer(&xprt->timer); 1095 xprt->last_used = jiffies; 1096 xprt->cwnd = RPC_INITCWND; 1097 xprt->bind_index = 0; 1098 1099 rpc_init_wait_queue(&xprt->binding, "xprt_binding"); 1100 rpc_init_wait_queue(&xprt->pending, "xprt_pending"); 1101 rpc_init_wait_queue(&xprt->sending, "xprt_sending"); 1102 rpc_init_wait_queue(&xprt->resend, "xprt_resend"); 1103 rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog"); 1104 1105 /* initialize free list */ 1106 for (req = &xprt->slot[xprt->max_reqs-1]; req >= &xprt->slot[0]; req--) 1107 list_add(&req->rq_list, &xprt->free); 1108 1109 xprt_init_xid(xprt); 1110 1111 dprintk("RPC: created transport %p with %u slots\n", xprt, 1112 xprt->max_reqs); 1113 return xprt; 1114} 1115 1116/** 1117 * xprt_destroy - destroy an RPC transport, killing off all requests. 1118 * @kref: kref for the transport to destroy 1119 * 1120 */ 1121static void xprt_destroy(struct kref *kref) 1122{ 1123 struct rpc_xprt *xprt = container_of(kref, struct rpc_xprt, kref); 1124 1125 dprintk("RPC: destroying transport %p\n", xprt); 1126 xprt->shutdown = 1; 1127 del_timer_sync(&xprt->timer); 1128 1129 rpc_destroy_wait_queue(&xprt->binding); 1130 rpc_destroy_wait_queue(&xprt->pending); 1131 rpc_destroy_wait_queue(&xprt->sending); 1132 rpc_destroy_wait_queue(&xprt->resend); 1133 rpc_destroy_wait_queue(&xprt->backlog); 1134 /* 1135 * Tear down transport state and free the rpc_xprt 1136 */ 1137 xprt->ops->destroy(xprt); 1138} 1139 1140/** 1141 * xprt_put - release a reference to an RPC transport. 1142 * @xprt: pointer to the transport 1143 * 1144 */ 1145void xprt_put(struct rpc_xprt *xprt) 1146{ 1147 kref_put(&xprt->kref, xprt_destroy); 1148} 1149 1150/** 1151 * xprt_get - return a reference to an RPC transport. 1152 * @xprt: pointer to the transport 1153 * 1154 */ 1155struct rpc_xprt *xprt_get(struct rpc_xprt *xprt) 1156{ 1157 kref_get(&xprt->kref); 1158 return xprt; 1159} 1160