xprt.c revision 03bf4b707eee06706c9db343dd5c905b7ee47ed2
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, it installs a timer that 16 * is run after the packet's timeout has expired. 17 * - When a packet arrives, the data_ready handler walks the list of 18 * pending requests for that transport. If a matching XID is found, the 19 * caller is woken up, and the timer removed. 20 * - When no reply arrives within the timeout interval, the timer is 21 * fired by the kernel and runs xprt_timer(). It either adjusts the 22 * timeout values (minor timeout) or wakes up the caller with a status 23 * of -ETIMEDOUT. 24 * - When the caller receives a notification from RPC that a reply arrived, 25 * it should release the RPC slot, and process the reply. 26 * If the call timed out, it may choose to retry the operation by 27 * adjusting the initial timeout value, and simply calling rpc_call 28 * again. 29 * 30 * Support for async RPC is done through a set of RPC-specific scheduling 31 * primitives that `transparently' work for processes as well as async 32 * tasks that rely on callbacks. 33 * 34 * Copyright (C) 1995-1997, Olaf Kirch <okir@monad.swb.de> 35 * 36 * Transport switch API copyright (C) 2005, Chuck Lever <cel@netapp.com> 37 */ 38 39#include <linux/module.h> 40 41#include <linux/types.h> 42#include <linux/interrupt.h> 43#include <linux/workqueue.h> 44#include <linux/random.h> 45 46#include <linux/sunrpc/clnt.h> 47 48/* 49 * Local variables 50 */ 51 52#ifdef RPC_DEBUG 53# undef RPC_DEBUG_DATA 54# define RPCDBG_FACILITY RPCDBG_XPRT 55#endif 56 57/* 58 * Local functions 59 */ 60static void xprt_request_init(struct rpc_task *, struct rpc_xprt *); 61static inline void do_xprt_reserve(struct rpc_task *); 62static void xprt_connect_status(struct rpc_task *task); 63static int __xprt_get_cong(struct rpc_xprt *, struct rpc_task *); 64 65/* 66 * The transport code maintains an estimate on the maximum number of out- 67 * standing RPC requests, using a smoothed version of the congestion 68 * avoidance implemented in 44BSD. This is basically the Van Jacobson 69 * congestion algorithm: If a retransmit occurs, the congestion window is 70 * halved; otherwise, it is incremented by 1/cwnd when 71 * 72 * - a reply is received and 73 * - a full number of requests are outstanding and 74 * - the congestion window hasn't been updated recently. 75 */ 76#define RPC_CWNDSHIFT (8U) 77#define RPC_CWNDSCALE (1U << RPC_CWNDSHIFT) 78#define RPC_INITCWND RPC_CWNDSCALE 79#define RPC_MAXCWND(xprt) ((xprt)->max_reqs << RPC_CWNDSHIFT) 80 81#define RPCXPRT_CONGESTED(xprt) ((xprt)->cong >= (xprt)->cwnd) 82 83/** 84 * xprt_reserve_xprt - serialize write access to transports 85 * @task: task that is requesting access to the transport 86 * 87 * This prevents mixing the payload of separate requests, and prevents 88 * transport connects from colliding with writes. No congestion control 89 * is provided. 90 */ 91int xprt_reserve_xprt(struct rpc_task *task) 92{ 93 struct rpc_xprt *xprt = task->tk_xprt; 94 struct rpc_rqst *req = task->tk_rqstp; 95 96 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) { 97 if (task == xprt->snd_task) 98 return 1; 99 if (task == NULL) 100 return 0; 101 goto out_sleep; 102 } 103 xprt->snd_task = task; 104 if (req) { 105 req->rq_bytes_sent = 0; 106 req->rq_ntrans++; 107 } 108 return 1; 109 110out_sleep: 111 dprintk("RPC: %4d failed to lock transport %p\n", 112 task->tk_pid, xprt); 113 task->tk_timeout = 0; 114 task->tk_status = -EAGAIN; 115 if (req && req->rq_ntrans) 116 rpc_sleep_on(&xprt->resend, task, NULL, NULL); 117 else 118 rpc_sleep_on(&xprt->sending, task, NULL, NULL); 119 return 0; 120} 121 122/* 123 * xprt_reserve_xprt_cong - serialize write access to transports 124 * @task: task that is requesting access to the transport 125 * 126 * Same as xprt_reserve_xprt, but Van Jacobson congestion control is 127 * integrated into the decision of whether a request is allowed to be 128 * woken up and given access to the transport. 129 */ 130int xprt_reserve_xprt_cong(struct rpc_task *task) 131{ 132 struct rpc_xprt *xprt = task->tk_xprt; 133 struct rpc_rqst *req = task->tk_rqstp; 134 135 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) { 136 if (task == xprt->snd_task) 137 return 1; 138 goto out_sleep; 139 } 140 if (__xprt_get_cong(xprt, task)) { 141 xprt->snd_task = task; 142 if (req) { 143 req->rq_bytes_sent = 0; 144 req->rq_ntrans++; 145 } 146 return 1; 147 } 148 smp_mb__before_clear_bit(); 149 clear_bit(XPRT_LOCKED, &xprt->state); 150 smp_mb__after_clear_bit(); 151out_sleep: 152 dprintk("RPC: %4d failed to lock transport %p\n", task->tk_pid, xprt); 153 task->tk_timeout = 0; 154 task->tk_status = -EAGAIN; 155 if (req && req->rq_ntrans) 156 rpc_sleep_on(&xprt->resend, task, NULL, NULL); 157 else 158 rpc_sleep_on(&xprt->sending, task, NULL, NULL); 159 return 0; 160} 161 162static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task) 163{ 164 int retval; 165 166 spin_lock_bh(&xprt->transport_lock); 167 retval = xprt->ops->reserve_xprt(task); 168 spin_unlock_bh(&xprt->transport_lock); 169 return retval; 170} 171 172static void __xprt_lock_write_next(struct rpc_xprt *xprt) 173{ 174 struct rpc_task *task; 175 struct rpc_rqst *req; 176 177 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) 178 return; 179 180 task = rpc_wake_up_next(&xprt->resend); 181 if (!task) { 182 task = rpc_wake_up_next(&xprt->sending); 183 if (!task) 184 goto out_unlock; 185 } 186 187 req = task->tk_rqstp; 188 xprt->snd_task = task; 189 if (req) { 190 req->rq_bytes_sent = 0; 191 req->rq_ntrans++; 192 } 193 return; 194 195out_unlock: 196 smp_mb__before_clear_bit(); 197 clear_bit(XPRT_LOCKED, &xprt->state); 198 smp_mb__after_clear_bit(); 199} 200 201static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt) 202{ 203 struct rpc_task *task; 204 205 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) 206 return; 207 if (RPCXPRT_CONGESTED(xprt)) 208 goto out_unlock; 209 task = rpc_wake_up_next(&xprt->resend); 210 if (!task) { 211 task = rpc_wake_up_next(&xprt->sending); 212 if (!task) 213 goto out_unlock; 214 } 215 if (__xprt_get_cong(xprt, task)) { 216 struct rpc_rqst *req = task->tk_rqstp; 217 xprt->snd_task = task; 218 if (req) { 219 req->rq_bytes_sent = 0; 220 req->rq_ntrans++; 221 } 222 return; 223 } 224out_unlock: 225 smp_mb__before_clear_bit(); 226 clear_bit(XPRT_LOCKED, &xprt->state); 227 smp_mb__after_clear_bit(); 228} 229 230/** 231 * xprt_release_xprt - allow other requests to use a transport 232 * @xprt: transport with other tasks potentially waiting 233 * @task: task that is releasing access to the transport 234 * 235 * Note that "task" can be NULL. No congestion control is provided. 236 */ 237void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task) 238{ 239 if (xprt->snd_task == task) { 240 xprt->snd_task = NULL; 241 smp_mb__before_clear_bit(); 242 clear_bit(XPRT_LOCKED, &xprt->state); 243 smp_mb__after_clear_bit(); 244 __xprt_lock_write_next(xprt); 245 } 246} 247 248/** 249 * xprt_release_xprt_cong - allow other requests to use a transport 250 * @xprt: transport with other tasks potentially waiting 251 * @task: task that is releasing access to the transport 252 * 253 * Note that "task" can be NULL. Another task is awoken to use the 254 * transport if the transport's congestion window allows it. 255 */ 256void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task) 257{ 258 if (xprt->snd_task == task) { 259 xprt->snd_task = NULL; 260 smp_mb__before_clear_bit(); 261 clear_bit(XPRT_LOCKED, &xprt->state); 262 smp_mb__after_clear_bit(); 263 __xprt_lock_write_next_cong(xprt); 264 } 265} 266 267static inline void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task) 268{ 269 spin_lock_bh(&xprt->transport_lock); 270 xprt->ops->release_xprt(xprt, task); 271 spin_unlock_bh(&xprt->transport_lock); 272} 273 274/* 275 * Van Jacobson congestion avoidance. Check if the congestion window 276 * overflowed. Put the task to sleep if this is the case. 277 */ 278static int 279__xprt_get_cong(struct rpc_xprt *xprt, struct rpc_task *task) 280{ 281 struct rpc_rqst *req = task->tk_rqstp; 282 283 if (req->rq_cong) 284 return 1; 285 dprintk("RPC: %4d xprt_cwnd_limited cong = %ld cwnd = %ld\n", 286 task->tk_pid, xprt->cong, xprt->cwnd); 287 if (RPCXPRT_CONGESTED(xprt)) 288 return 0; 289 req->rq_cong = 1; 290 xprt->cong += RPC_CWNDSCALE; 291 return 1; 292} 293 294/* 295 * Adjust the congestion window, and wake up the next task 296 * that has been sleeping due to congestion 297 */ 298static void 299__xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req) 300{ 301 if (!req->rq_cong) 302 return; 303 req->rq_cong = 0; 304 xprt->cong -= RPC_CWNDSCALE; 305 __xprt_lock_write_next_cong(xprt); 306} 307 308/** 309 * xprt_release_rqst_cong - housekeeping when request is complete 310 * @task: RPC request that recently completed 311 * 312 * Useful for transports that require congestion control. 313 */ 314void xprt_release_rqst_cong(struct rpc_task *task) 315{ 316 __xprt_put_cong(task->tk_xprt, task->tk_rqstp); 317} 318 319/** 320 * xprt_adjust_cwnd - adjust transport congestion window 321 * @task: recently completed RPC request used to adjust window 322 * @result: result code of completed RPC request 323 * 324 * We use a time-smoothed congestion estimator to avoid heavy oscillation. 325 */ 326void xprt_adjust_cwnd(struct rpc_task *task, int result) 327{ 328 struct rpc_rqst *req = task->tk_rqstp; 329 struct rpc_xprt *xprt = task->tk_xprt; 330 unsigned long cwnd = xprt->cwnd; 331 332 if (result >= 0 && cwnd <= xprt->cong) { 333 /* The (cwnd >> 1) term makes sure 334 * the result gets rounded properly. */ 335 cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd; 336 if (cwnd > RPC_MAXCWND(xprt)) 337 cwnd = RPC_MAXCWND(xprt); 338 __xprt_lock_write_next_cong(xprt); 339 } else if (result == -ETIMEDOUT) { 340 cwnd >>= 1; 341 if (cwnd < RPC_CWNDSCALE) 342 cwnd = RPC_CWNDSCALE; 343 } 344 dprintk("RPC: cong %ld, cwnd was %ld, now %ld\n", 345 xprt->cong, xprt->cwnd, cwnd); 346 xprt->cwnd = cwnd; 347 __xprt_put_cong(xprt, req); 348} 349 350/** 351 * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue 352 * @xprt: transport with waiting tasks 353 * @status: result code to plant in each task before waking it 354 * 355 */ 356void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status) 357{ 358 if (status < 0) 359 rpc_wake_up_status(&xprt->pending, status); 360 else 361 rpc_wake_up(&xprt->pending); 362} 363 364/** 365 * xprt_wait_for_buffer_space - wait for transport output buffer to clear 366 * @task: task to be put to sleep 367 * 368 */ 369void xprt_wait_for_buffer_space(struct rpc_task *task) 370{ 371 struct rpc_rqst *req = task->tk_rqstp; 372 struct rpc_xprt *xprt = req->rq_xprt; 373 374 task->tk_timeout = req->rq_timeout; 375 rpc_sleep_on(&xprt->pending, task, NULL, NULL); 376} 377 378/** 379 * xprt_write_space - wake the task waiting for transport output buffer space 380 * @xprt: transport with waiting tasks 381 * 382 * Can be called in a soft IRQ context, so xprt_write_space never sleeps. 383 */ 384void xprt_write_space(struct rpc_xprt *xprt) 385{ 386 if (unlikely(xprt->shutdown)) 387 return; 388 389 spin_lock_bh(&xprt->transport_lock); 390 if (xprt->snd_task) { 391 dprintk("RPC: write space: waking waiting task on xprt %p\n", 392 xprt); 393 rpc_wake_up_task(xprt->snd_task); 394 } 395 spin_unlock_bh(&xprt->transport_lock); 396} 397 398/** 399 * xprt_set_retrans_timeout_def - set a request's retransmit timeout 400 * @task: task whose timeout is to be set 401 * 402 * Set a request's retransmit timeout based on the transport's 403 * default timeout parameters. Used by transports that don't adjust 404 * the retransmit timeout based on round-trip time estimation. 405 */ 406void xprt_set_retrans_timeout_def(struct rpc_task *task) 407{ 408 task->tk_timeout = task->tk_rqstp->rq_timeout; 409} 410 411/* 412 * xprt_set_retrans_timeout_rtt - set a request's retransmit timeout 413 * @task: task whose timeout is to be set 414 * 415 * Set a request's retransmit timeout using the RTT estimator. 416 */ 417void xprt_set_retrans_timeout_rtt(struct rpc_task *task) 418{ 419 int timer = task->tk_msg.rpc_proc->p_timer; 420 struct rpc_rtt *rtt = task->tk_client->cl_rtt; 421 struct rpc_rqst *req = task->tk_rqstp; 422 unsigned long max_timeout = req->rq_xprt->timeout.to_maxval; 423 424 task->tk_timeout = rpc_calc_rto(rtt, timer); 425 task->tk_timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries; 426 if (task->tk_timeout > max_timeout || task->tk_timeout == 0) 427 task->tk_timeout = max_timeout; 428} 429 430static void xprt_reset_majortimeo(struct rpc_rqst *req) 431{ 432 struct rpc_timeout *to = &req->rq_xprt->timeout; 433 434 req->rq_majortimeo = req->rq_timeout; 435 if (to->to_exponential) 436 req->rq_majortimeo <<= to->to_retries; 437 else 438 req->rq_majortimeo += to->to_increment * to->to_retries; 439 if (req->rq_majortimeo > to->to_maxval || req->rq_majortimeo == 0) 440 req->rq_majortimeo = to->to_maxval; 441 req->rq_majortimeo += jiffies; 442} 443 444/** 445 * xprt_adjust_timeout - adjust timeout values for next retransmit 446 * @req: RPC request containing parameters to use for the adjustment 447 * 448 */ 449int xprt_adjust_timeout(struct rpc_rqst *req) 450{ 451 struct rpc_xprt *xprt = req->rq_xprt; 452 struct rpc_timeout *to = &xprt->timeout; 453 int status = 0; 454 455 if (time_before(jiffies, req->rq_majortimeo)) { 456 if (to->to_exponential) 457 req->rq_timeout <<= 1; 458 else 459 req->rq_timeout += to->to_increment; 460 if (to->to_maxval && req->rq_timeout >= to->to_maxval) 461 req->rq_timeout = to->to_maxval; 462 req->rq_retries++; 463 pprintk("RPC: %lu retrans\n", jiffies); 464 } else { 465 req->rq_timeout = to->to_initval; 466 req->rq_retries = 0; 467 xprt_reset_majortimeo(req); 468 /* Reset the RTT counters == "slow start" */ 469 spin_lock_bh(&xprt->transport_lock); 470 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval); 471 spin_unlock_bh(&xprt->transport_lock); 472 pprintk("RPC: %lu timeout\n", jiffies); 473 status = -ETIMEDOUT; 474 } 475 476 if (req->rq_timeout == 0) { 477 printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n"); 478 req->rq_timeout = 5 * HZ; 479 } 480 return status; 481} 482 483static void xprt_autoclose(void *args) 484{ 485 struct rpc_xprt *xprt = (struct rpc_xprt *)args; 486 487 xprt_disconnect(xprt); 488 xprt->ops->close(xprt); 489 xprt_release_write(xprt, NULL); 490} 491 492/** 493 * xprt_disconnect - mark a transport as disconnected 494 * @xprt: transport to flag for disconnect 495 * 496 */ 497void xprt_disconnect(struct rpc_xprt *xprt) 498{ 499 dprintk("RPC: disconnected transport %p\n", xprt); 500 spin_lock_bh(&xprt->transport_lock); 501 xprt_clear_connected(xprt); 502 xprt_wake_pending_tasks(xprt, -ENOTCONN); 503 spin_unlock_bh(&xprt->transport_lock); 504} 505 506static void 507xprt_init_autodisconnect(unsigned long data) 508{ 509 struct rpc_xprt *xprt = (struct rpc_xprt *)data; 510 511 spin_lock(&xprt->transport_lock); 512 if (!list_empty(&xprt->recv) || xprt->shutdown) 513 goto out_abort; 514 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) 515 goto out_abort; 516 spin_unlock(&xprt->transport_lock); 517 if (xprt_connecting(xprt)) 518 xprt_release_write(xprt, NULL); 519 else 520 schedule_work(&xprt->task_cleanup); 521 return; 522out_abort: 523 spin_unlock(&xprt->transport_lock); 524} 525 526/** 527 * xprt_connect - schedule a transport connect operation 528 * @task: RPC task that is requesting the connect 529 * 530 */ 531void xprt_connect(struct rpc_task *task) 532{ 533 struct rpc_xprt *xprt = task->tk_xprt; 534 535 dprintk("RPC: %4d xprt_connect xprt %p %s connected\n", task->tk_pid, 536 xprt, (xprt_connected(xprt) ? "is" : "is not")); 537 538 if (xprt->shutdown) { 539 task->tk_status = -EIO; 540 return; 541 } 542 if (!xprt->addr.sin_port) { 543 task->tk_status = -EIO; 544 return; 545 } 546 if (!xprt_lock_write(xprt, task)) 547 return; 548 if (xprt_connected(xprt)) 549 xprt_release_write(xprt, task); 550 else { 551 if (task->tk_rqstp) 552 task->tk_rqstp->rq_bytes_sent = 0; 553 554 task->tk_timeout = xprt->connect_timeout; 555 rpc_sleep_on(&xprt->pending, task, xprt_connect_status, NULL); 556 xprt->ops->connect(task); 557 } 558 return; 559} 560 561static void xprt_connect_status(struct rpc_task *task) 562{ 563 struct rpc_xprt *xprt = task->tk_xprt; 564 565 if (task->tk_status >= 0) { 566 dprintk("RPC: %4d xprt_connect_status: connection established\n", 567 task->tk_pid); 568 return; 569 } 570 571 switch (task->tk_status) { 572 case -ECONNREFUSED: 573 case -ECONNRESET: 574 dprintk("RPC: %4d xprt_connect_status: server %s refused connection\n", 575 task->tk_pid, task->tk_client->cl_server); 576 break; 577 case -ENOTCONN: 578 dprintk("RPC: %4d xprt_connect_status: connection broken\n", 579 task->tk_pid); 580 break; 581 case -ETIMEDOUT: 582 dprintk("RPC: %4d xprt_connect_status: connect attempt timed out\n", 583 task->tk_pid); 584 break; 585 default: 586 dprintk("RPC: %4d xprt_connect_status: error %d connecting to server %s\n", 587 task->tk_pid, -task->tk_status, task->tk_client->cl_server); 588 xprt_release_write(xprt, task); 589 task->tk_status = -EIO; 590 return; 591 } 592 593 /* if soft mounted, just cause this RPC to fail */ 594 if (RPC_IS_SOFT(task)) { 595 xprt_release_write(xprt, task); 596 task->tk_status = -EIO; 597 } 598} 599 600/** 601 * xprt_lookup_rqst - find an RPC request corresponding to an XID 602 * @xprt: transport on which the original request was transmitted 603 * @xid: RPC XID of incoming reply 604 * 605 */ 606struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, u32 xid) 607{ 608 struct list_head *pos; 609 struct rpc_rqst *req = NULL; 610 611 list_for_each(pos, &xprt->recv) { 612 struct rpc_rqst *entry = list_entry(pos, struct rpc_rqst, rq_list); 613 if (entry->rq_xid == xid) { 614 req = entry; 615 break; 616 } 617 } 618 return req; 619} 620 621/** 622 * xprt_update_rtt - update an RPC client's RTT state after receiving a reply 623 * @task: RPC request that recently completed 624 * 625 */ 626void xprt_update_rtt(struct rpc_task *task) 627{ 628 struct rpc_rqst *req = task->tk_rqstp; 629 struct rpc_rtt *rtt = task->tk_client->cl_rtt; 630 unsigned timer = task->tk_msg.rpc_proc->p_timer; 631 632 if (timer) { 633 if (req->rq_ntrans == 1) 634 rpc_update_rtt(rtt, timer, 635 (long)jiffies - req->rq_xtime); 636 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1); 637 } 638} 639 640/** 641 * xprt_complete_rqst - called when reply processing is complete 642 * @task: RPC request that recently completed 643 * @copied: actual number of bytes received from the transport 644 * 645 * Caller holds transport lock. 646 */ 647void xprt_complete_rqst(struct rpc_task *task, int copied) 648{ 649 struct rpc_rqst *req = task->tk_rqstp; 650 651 dprintk("RPC: %5u xid %08x complete (%d bytes received)\n", 652 task->tk_pid, ntohl(req->rq_xid), copied); 653 654 list_del_init(&req->rq_list); 655 req->rq_received = req->rq_private_buf.len = copied; 656 rpc_wake_up_task(task); 657} 658 659static void xprt_timer(struct rpc_task *task) 660{ 661 struct rpc_rqst *req = task->tk_rqstp; 662 struct rpc_xprt *xprt = req->rq_xprt; 663 664 dprintk("RPC: %4d xprt_timer\n", task->tk_pid); 665 666 spin_lock(&xprt->transport_lock); 667 if (!req->rq_received) { 668 if (xprt->ops->timer) 669 xprt->ops->timer(task); 670 task->tk_status = -ETIMEDOUT; 671 } 672 task->tk_timeout = 0; 673 rpc_wake_up_task(task); 674 spin_unlock(&xprt->transport_lock); 675} 676 677/** 678 * xprt_prepare_transmit - reserve the transport before sending a request 679 * @task: RPC task about to send a request 680 * 681 */ 682int xprt_prepare_transmit(struct rpc_task *task) 683{ 684 struct rpc_rqst *req = task->tk_rqstp; 685 struct rpc_xprt *xprt = req->rq_xprt; 686 int err = 0; 687 688 dprintk("RPC: %4d xprt_prepare_transmit\n", task->tk_pid); 689 690 if (xprt->shutdown) 691 return -EIO; 692 693 spin_lock_bh(&xprt->transport_lock); 694 if (req->rq_received && !req->rq_bytes_sent) { 695 err = req->rq_received; 696 goto out_unlock; 697 } 698 if (!xprt->ops->reserve_xprt(task)) { 699 err = -EAGAIN; 700 goto out_unlock; 701 } 702 703 if (!xprt_connected(xprt)) { 704 err = -ENOTCONN; 705 goto out_unlock; 706 } 707out_unlock: 708 spin_unlock_bh(&xprt->transport_lock); 709 return err; 710} 711 712/** 713 * xprt_transmit - send an RPC request on a transport 714 * @task: controlling RPC task 715 * 716 * We have to copy the iovec because sendmsg fiddles with its contents. 717 */ 718void xprt_transmit(struct rpc_task *task) 719{ 720 struct rpc_rqst *req = task->tk_rqstp; 721 struct rpc_xprt *xprt = req->rq_xprt; 722 int status; 723 724 dprintk("RPC: %4d xprt_transmit(%u)\n", task->tk_pid, req->rq_slen); 725 726 smp_rmb(); 727 if (!req->rq_received) { 728 if (list_empty(&req->rq_list)) { 729 spin_lock_bh(&xprt->transport_lock); 730 /* Update the softirq receive buffer */ 731 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, 732 sizeof(req->rq_private_buf)); 733 /* Add request to the receive list */ 734 list_add_tail(&req->rq_list, &xprt->recv); 735 spin_unlock_bh(&xprt->transport_lock); 736 xprt_reset_majortimeo(req); 737 /* Turn off autodisconnect */ 738 del_singleshot_timer_sync(&xprt->timer); 739 } 740 } else if (!req->rq_bytes_sent) 741 return; 742 743 status = xprt->ops->send_request(task); 744 if (status == 0) { 745 dprintk("RPC: %4d xmit complete\n", task->tk_pid); 746 spin_lock_bh(&xprt->transport_lock); 747 xprt->ops->set_retrans_timeout(task); 748 /* Don't race with disconnect */ 749 if (!xprt_connected(xprt)) 750 task->tk_status = -ENOTCONN; 751 else if (!req->rq_received) 752 rpc_sleep_on(&xprt->pending, task, NULL, xprt_timer); 753 xprt->ops->release_xprt(xprt, task); 754 spin_unlock_bh(&xprt->transport_lock); 755 return; 756 } 757 758 /* Note: at this point, task->tk_sleeping has not yet been set, 759 * hence there is no danger of the waking up task being put on 760 * schedq, and being picked up by a parallel run of rpciod(). 761 */ 762 task->tk_status = status; 763 764 switch (status) { 765 case -ECONNREFUSED: 766 rpc_sleep_on(&xprt->sending, task, NULL, NULL); 767 case -EAGAIN: 768 case -ENOTCONN: 769 return; 770 default: 771 break; 772 } 773 xprt_release_write(xprt, task); 774 return; 775} 776 777static inline void do_xprt_reserve(struct rpc_task *task) 778{ 779 struct rpc_xprt *xprt = task->tk_xprt; 780 781 task->tk_status = 0; 782 if (task->tk_rqstp) 783 return; 784 if (!list_empty(&xprt->free)) { 785 struct rpc_rqst *req = list_entry(xprt->free.next, struct rpc_rqst, rq_list); 786 list_del_init(&req->rq_list); 787 task->tk_rqstp = req; 788 xprt_request_init(task, xprt); 789 return; 790 } 791 dprintk("RPC: waiting for request slot\n"); 792 task->tk_status = -EAGAIN; 793 task->tk_timeout = 0; 794 rpc_sleep_on(&xprt->backlog, task, NULL, NULL); 795} 796 797/** 798 * xprt_reserve - allocate an RPC request slot 799 * @task: RPC task requesting a slot allocation 800 * 801 * If no more slots are available, place the task on the transport's 802 * backlog queue. 803 */ 804void xprt_reserve(struct rpc_task *task) 805{ 806 struct rpc_xprt *xprt = task->tk_xprt; 807 808 task->tk_status = -EIO; 809 if (!xprt->shutdown) { 810 spin_lock(&xprt->reserve_lock); 811 do_xprt_reserve(task); 812 spin_unlock(&xprt->reserve_lock); 813 } 814} 815 816static inline u32 xprt_alloc_xid(struct rpc_xprt *xprt) 817{ 818 return xprt->xid++; 819} 820 821static inline void xprt_init_xid(struct rpc_xprt *xprt) 822{ 823 get_random_bytes(&xprt->xid, sizeof(xprt->xid)); 824} 825 826static void xprt_request_init(struct rpc_task *task, struct rpc_xprt *xprt) 827{ 828 struct rpc_rqst *req = task->tk_rqstp; 829 830 req->rq_timeout = xprt->timeout.to_initval; 831 req->rq_task = task; 832 req->rq_xprt = xprt; 833 req->rq_xid = xprt_alloc_xid(xprt); 834 dprintk("RPC: %4d reserved req %p xid %08x\n", task->tk_pid, 835 req, ntohl(req->rq_xid)); 836} 837 838/** 839 * xprt_release - release an RPC request slot 840 * @task: task which is finished with the slot 841 * 842 */ 843void xprt_release(struct rpc_task *task) 844{ 845 struct rpc_xprt *xprt = task->tk_xprt; 846 struct rpc_rqst *req; 847 848 if (!(req = task->tk_rqstp)) 849 return; 850 spin_lock_bh(&xprt->transport_lock); 851 xprt->ops->release_xprt(xprt, task); 852 if (xprt->ops->release_request) 853 xprt->ops->release_request(task); 854 if (!list_empty(&req->rq_list)) 855 list_del(&req->rq_list); 856 xprt->last_used = jiffies; 857 if (list_empty(&xprt->recv) && !xprt->shutdown) 858 mod_timer(&xprt->timer, 859 xprt->last_used + xprt->idle_timeout); 860 spin_unlock_bh(&xprt->transport_lock); 861 task->tk_rqstp = NULL; 862 memset(req, 0, sizeof(*req)); /* mark unused */ 863 864 dprintk("RPC: %4d release request %p\n", task->tk_pid, req); 865 866 spin_lock(&xprt->reserve_lock); 867 list_add(&req->rq_list, &xprt->free); 868 rpc_wake_up_next(&xprt->backlog); 869 spin_unlock(&xprt->reserve_lock); 870} 871 872/** 873 * xprt_set_timeout - set constant RPC timeout 874 * @to: RPC timeout parameters to set up 875 * @retr: number of retries 876 * @incr: amount of increase after each retry 877 * 878 */ 879void xprt_set_timeout(struct rpc_timeout *to, unsigned int retr, unsigned long incr) 880{ 881 to->to_initval = 882 to->to_increment = incr; 883 to->to_maxval = to->to_initval + (incr * retr); 884 to->to_retries = retr; 885 to->to_exponential = 0; 886} 887 888static struct rpc_xprt *xprt_setup(int proto, struct sockaddr_in *ap, struct rpc_timeout *to) 889{ 890 int result; 891 struct rpc_xprt *xprt; 892 struct rpc_rqst *req; 893 894 if ((xprt = kmalloc(sizeof(struct rpc_xprt), GFP_KERNEL)) == NULL) 895 return ERR_PTR(-ENOMEM); 896 memset(xprt, 0, sizeof(*xprt)); /* Nnnngh! */ 897 898 xprt->addr = *ap; 899 900 switch (proto) { 901 case IPPROTO_UDP: 902 result = xs_setup_udp(xprt, to); 903 break; 904 case IPPROTO_TCP: 905 result = xs_setup_tcp(xprt, to); 906 break; 907 default: 908 printk(KERN_ERR "RPC: unrecognized transport protocol: %d\n", 909 proto); 910 result = -EIO; 911 break; 912 } 913 if (result) { 914 kfree(xprt); 915 return ERR_PTR(result); 916 } 917 918 spin_lock_init(&xprt->transport_lock); 919 spin_lock_init(&xprt->reserve_lock); 920 921 INIT_LIST_HEAD(&xprt->free); 922 INIT_LIST_HEAD(&xprt->recv); 923 INIT_WORK(&xprt->task_cleanup, xprt_autoclose, xprt); 924 init_timer(&xprt->timer); 925 xprt->timer.function = xprt_init_autodisconnect; 926 xprt->timer.data = (unsigned long) xprt; 927 xprt->last_used = jiffies; 928 xprt->cwnd = RPC_INITCWND; 929 930 rpc_init_wait_queue(&xprt->pending, "xprt_pending"); 931 rpc_init_wait_queue(&xprt->sending, "xprt_sending"); 932 rpc_init_wait_queue(&xprt->resend, "xprt_resend"); 933 rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog"); 934 935 /* initialize free list */ 936 for (req = &xprt->slot[xprt->max_reqs-1]; req >= &xprt->slot[0]; req--) 937 list_add(&req->rq_list, &xprt->free); 938 939 xprt_init_xid(xprt); 940 941 dprintk("RPC: created transport %p with %u slots\n", xprt, 942 xprt->max_reqs); 943 944 return xprt; 945} 946 947/** 948 * xprt_create_proto - create an RPC client transport 949 * @proto: requested transport protocol 950 * @sap: remote peer's address 951 * @to: timeout parameters for new transport 952 * 953 */ 954struct rpc_xprt *xprt_create_proto(int proto, struct sockaddr_in *sap, struct rpc_timeout *to) 955{ 956 struct rpc_xprt *xprt; 957 958 xprt = xprt_setup(proto, sap, to); 959 if (IS_ERR(xprt)) 960 dprintk("RPC: xprt_create_proto failed\n"); 961 else 962 dprintk("RPC: xprt_create_proto created xprt %p\n", xprt); 963 return xprt; 964} 965 966static void xprt_shutdown(struct rpc_xprt *xprt) 967{ 968 xprt->shutdown = 1; 969 rpc_wake_up(&xprt->sending); 970 rpc_wake_up(&xprt->resend); 971 xprt_wake_pending_tasks(xprt, -EIO); 972 rpc_wake_up(&xprt->backlog); 973 del_timer_sync(&xprt->timer); 974} 975 976/** 977 * xprt_destroy - destroy an RPC transport, killing off all requests. 978 * @xprt: transport to destroy 979 * 980 */ 981int xprt_destroy(struct rpc_xprt *xprt) 982{ 983 dprintk("RPC: destroying transport %p\n", xprt); 984 xprt_shutdown(xprt); 985 xprt->ops->destroy(xprt); 986 kfree(xprt); 987 988 return 0; 989} 990