SAFE public projects git trees. - safe/jmp/linux-2.6/blob - net/sunrpc/xprt.c

   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 #include <linux/sunrpc/bc_xprt.h>
  50
  51 #include "sunrpc.h"
  52
  53 /*
  54  * Local variables
  55  */
  56
  57 #ifdef RPC_DEBUG
  58 # define RPCDBG_FACILITY        RPCDBG_XPRT
  59 #endif
  60
  61 /*
  62  * Local functions
  63  */
  64 static void     xprt_request_init(struct rpc_task *, struct rpc_xprt *);
  65 static void     xprt_connect_status(struct rpc_task *task);
  66 static int      __xprt_get_cong(struct rpc_xprt *, struct rpc_task *);
  67
  68 static DEFINE_SPINLOCK(xprt_list_lock);
  69 static 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  */
 101 int 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
 119 out:
 120         spin_unlock(&xprt_list_lock);
 121         return result;
 122 }
 123 EXPORT_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  */
 133 int 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
 151 out:
 152         spin_unlock(&xprt_list_lock);
 153         return result;
 154 }
 155 EXPORT_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  */
 165 int 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);
 183 out:
 184         return result;
 185 }
 186 EXPORT_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  */
 196 int 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
 215 out_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 }
 226 EXPORT_SYMBOL_GPL(xprt_reserve_xprt);
 227
 228 static 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  */
 247 int 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);
 266 out_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 }
 276 EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong);
 277
 278 static 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
 288 static 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
 311 out_unlock:
 312         xprt_clear_locked(xprt);
 313 }
 314
 315 static 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         }
 338 out_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  */
 349 void 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 }
 356 EXPORT_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  */
 366 void 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 }
 373 EXPORT_SYMBOL_GPL(xprt_release_xprt_cong);
 374
 375 static 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  */
 386 static 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  */
 406 static 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  */
 422 void xprt_release_rqst_cong(struct rpc_task *task)
 423 {
 424         __xprt_put_cong(task->tk_xprt, task->tk_rqstp);
 425 }
 426 EXPORT_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  */
 435 void 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 }
 458 EXPORT_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  */
 466 void 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 }
 473 EXPORT_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  */
 480 void 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 }
 488 EXPORT_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  */
 496 void 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 }
 509 EXPORT_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  */
 519 void xprt_set_retrans_timeout_def(struct rpc_task *task)
 520 {
 521         task->tk_timeout = task->tk_rqstp->rq_timeout;
 522 }
 523 EXPORT_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  */
 531 void 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 }
 544 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_rtt);
 545
 546 static 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  */
 565 int 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
 597 static 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  */
 612 void 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 }
 620 EXPORT_SYMBOL_GPL(xprt_disconnect_done);
 621
 622 /**
 623  * xprt_force_disconnect - force a transport to disconnect
 624  * @xprt: transport to disconnect
 625  *
 626  */
 627 void 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  */
 650 void 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);
 663 out:
 664         spin_unlock_bh(&xprt->transport_lock);
 665 }
 666
 667 static void
 668 xprt_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;
 681 out_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  */
 690 void 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
 704         if (test_and_clear_bit(XPRT_CLOSE_WAIT, &xprt->state))
 705                 xprt->ops->close(xprt);
 706
 707         if (xprt_connected(xprt))
 708                 xprt_release_write(xprt, task);
 709         else {
 710                 if (task->tk_rqstp)
 711                         task->tk_rqstp->rq_bytes_sent = 0;
 712
 713                 task->tk_timeout = task->tk_rqstp->rq_timeout;
 714                 rpc_sleep_on(&xprt->pending, task, xprt_connect_status);
 715
 716                 if (test_bit(XPRT_CLOSING, &xprt->state))
 717                         return;
 718                 if (xprt_test_and_set_connecting(xprt))
 719                         return;
 720                 xprt->stat.connect_start = jiffies;
 721                 xprt->ops->connect(task);
 722         }
 723 }
 724
 725 static void xprt_connect_status(struct rpc_task *task)
 726 {
 727         struct rpc_xprt *xprt = task->tk_xprt;
 728
 729         if (task->tk_status == 0) {
 730                 xprt->stat.connect_count++;
 731                 xprt->stat.connect_time += (long)jiffies - xprt->stat.connect_start;
 732                 dprintk("RPC: %5u xprt_connect_status: connection established\n",
 733                                 task->tk_pid);
 734                 return;
 735         }
 736
 737         switch (task->tk_status) {
 738         case -EAGAIN:
 739                 dprintk("RPC: %5u xprt_connect_status: retrying\n", task->tk_pid);
 740                 break;
 741         case -ETIMEDOUT:
 742                 dprintk("RPC: %5u xprt_connect_status: connect attempt timed "
 743                                 "out\n", task->tk_pid);
 744                 break;
 745         default:
 746                 dprintk("RPC: %5u xprt_connect_status: error %d connecting to "
 747                                 "server %s\n", task->tk_pid, -task->tk_status,
 748                                 task->tk_client->cl_server);
 749                 xprt_release_write(xprt, task);
 750                 task->tk_status = -EIO;
 751         }
 752 }
 753
 754 /**
 755  * xprt_lookup_rqst - find an RPC request corresponding to an XID
 756  * @xprt: transport on which the original request was transmitted
 757  * @xid: RPC XID of incoming reply
 758  *
 759  */
 760 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
 761 {
 762         struct list_head *pos;
 763
 764         list_for_each(pos, &xprt->recv) {
 765                 struct rpc_rqst *entry = list_entry(pos, struct rpc_rqst, rq_list);
 766                 if (entry->rq_xid == xid)
 767                         return entry;
 768         }
 769
 770         dprintk("RPC:       xprt_lookup_rqst did not find xid %08x\n",
 771                         ntohl(xid));
 772         xprt->stat.bad_xids++;
 773         return NULL;
 774 }
 775 EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
 776
 777 static void xprt_update_rtt(struct rpc_task *task)
 778 {
 779         struct rpc_rqst *req = task->tk_rqstp;
 780         struct rpc_rtt *rtt = task->tk_client->cl_rtt;
 781         unsigned timer = task->tk_msg.rpc_proc->p_timer;
 782
 783         if (timer) {
 784                 if (req->rq_ntrans == 1)
 785                         rpc_update_rtt(rtt, timer, task->tk_rtt);
 786                 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
 787         }
 788 }
 789
 790 /**
 791  * xprt_complete_rqst - called when reply processing is complete
 792  * @task: RPC request that recently completed
 793  * @copied: actual number of bytes received from the transport
 794  *
 795  * Caller holds transport lock.
 796  */
 797 void xprt_complete_rqst(struct rpc_task *task, int copied)
 798 {
 799         struct rpc_rqst *req = task->tk_rqstp;
 800         struct rpc_xprt *xprt = req->rq_xprt;
 801
 802         dprintk("RPC: %5u xid %08x complete (%d bytes received)\n",
 803                         task->tk_pid, ntohl(req->rq_xid), copied);
 804
 805         xprt->stat.recvs++;
 806         task->tk_rtt = (long)jiffies - req->rq_xtime;
 807         if (xprt->ops->timer != NULL)
 808                 xprt_update_rtt(task);
 809
 810         list_del_init(&req->rq_list);
 811         req->rq_private_buf.len = copied;
 812         /* Ensure all writes are done before we update */
 813         /* req->rq_reply_bytes_recvd */
 814         smp_wmb();
 815         req->rq_reply_bytes_recvd = copied;
 816         rpc_wake_up_queued_task(&xprt->pending, task);
 817 }
 818 EXPORT_SYMBOL_GPL(xprt_complete_rqst);
 819
 820 static void xprt_timer(struct rpc_task *task)
 821 {
 822         struct rpc_rqst *req = task->tk_rqstp;
 823         struct rpc_xprt *xprt = req->rq_xprt;
 824
 825         if (task->tk_status != -ETIMEDOUT)
 826                 return;
 827         dprintk("RPC: %5u xprt_timer\n", task->tk_pid);
 828
 829         spin_lock_bh(&xprt->transport_lock);
 830         if (!req->rq_reply_bytes_recvd) {
 831                 if (xprt->ops->timer)
 832                         xprt->ops->timer(task);
 833         } else
 834                 task->tk_status = 0;
 835         spin_unlock_bh(&xprt->transport_lock);
 836 }
 837
 838 static inline int xprt_has_timer(struct rpc_xprt *xprt)
 839 {
 840         return xprt->idle_timeout != 0;
 841 }
 842
 843 /**
 844  * xprt_prepare_transmit - reserve the transport before sending a request
 845  * @task: RPC task about to send a request
 846  *
 847  */
 848 int xprt_prepare_transmit(struct rpc_task *task)
 849 {
 850         struct rpc_rqst *req = task->tk_rqstp;
 851         struct rpc_xprt *xprt = req->rq_xprt;
 852         int err = 0;
 853
 854         dprintk("RPC: %5u xprt_prepare_transmit\n", task->tk_pid);
 855
 856         spin_lock_bh(&xprt->transport_lock);
 857         if (req->rq_reply_bytes_recvd && !req->rq_bytes_sent) {
 858                 err = req->rq_reply_bytes_recvd;
 859                 goto out_unlock;
 860         }
 861         if (!xprt->ops->reserve_xprt(task))
 862                 err = -EAGAIN;
 863 out_unlock:
 864         spin_unlock_bh(&xprt->transport_lock);
 865         return err;
 866 }
 867
 868 void xprt_end_transmit(struct rpc_task *task)
 869 {
 870         xprt_release_write(task->tk_rqstp->rq_xprt, task);
 871 }
 872
 873 /**
 874  * xprt_transmit - send an RPC request on a transport
 875  * @task: controlling RPC task
 876  *
 877  * We have to copy the iovec because sendmsg fiddles with its contents.
 878  */
 879 void xprt_transmit(struct rpc_task *task)
 880 {
 881         struct rpc_rqst *req = task->tk_rqstp;
 882         struct rpc_xprt *xprt = req->rq_xprt;
 883         int status;
 884
 885         dprintk("RPC: %5u xprt_transmit(%u)\n", task->tk_pid, req->rq_slen);
 886
 887         if (!req->rq_reply_bytes_recvd) {
 888                 if (list_empty(&req->rq_list) && rpc_reply_expected(task)) {
 889                         /*
 890                          * Add to the list only if we're expecting a reply
 891                          */
 892                         spin_lock_bh(&xprt->transport_lock);
 893                         /* Update the softirq receive buffer */
 894                         memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
 895                                         sizeof(req->rq_private_buf));
 896                         /* Add request to the receive list */
 897                         list_add_tail(&req->rq_list, &xprt->recv);
 898                         spin_unlock_bh(&xprt->transport_lock);
 899                         xprt_reset_majortimeo(req);
 900                         /* Turn off autodisconnect */
 901                         del_singleshot_timer_sync(&xprt->timer);
 902                 }
 903         } else if (!req->rq_bytes_sent)
 904                 return;
 905
 906         req->rq_connect_cookie = xprt->connect_cookie;
 907         req->rq_xtime = jiffies;
 908         status = xprt->ops->send_request(task);
 909         if (status != 0) {
 910                 task->tk_status = status;
 911                 return;
 912         }
 913
 914         dprintk("RPC: %5u xmit complete\n", task->tk_pid);
 915         spin_lock_bh(&xprt->transport_lock);
 916
 917         xprt->ops->set_retrans_timeout(task);
 918
 919         xprt->stat.sends++;
 920         xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
 921         xprt->stat.bklog_u += xprt->backlog.qlen;
 922
 923         /* Don't race with disconnect */
 924         if (!xprt_connected(xprt))
 925                 task->tk_status = -ENOTCONN;
 926         else if (!req->rq_reply_bytes_recvd && rpc_reply_expected(task)) {
 927                 /*
 928                  * Sleep on the pending queue since
 929                  * we're expecting a reply.
 930                  */
 931                 rpc_sleep_on(&xprt->pending, task, xprt_timer);
 932         }
 933         spin_unlock_bh(&xprt->transport_lock);
 934 }
 935
 936 static void xprt_alloc_slot(struct rpc_task *task)
 937 {
 938         struct rpc_xprt *xprt = task->tk_xprt;
 939
 940         task->tk_status = 0;
 941         if (task->tk_rqstp)
 942                 return;
 943         if (!list_empty(&xprt->free)) {
 944                 struct rpc_rqst *req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
 945                 list_del_init(&req->rq_list);
 946                 task->tk_rqstp = req;
 947                 xprt_request_init(task, xprt);
 948                 return;
 949         }
 950         dprintk("RPC:       waiting for request slot\n");
 951         task->tk_status = -EAGAIN;
 952         task->tk_timeout = 0;
 953         rpc_sleep_on(&xprt->backlog, task, NULL);
 954 }
 955
 956 static void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
 957 {
 958         memset(req, 0, sizeof(*req));   /* mark unused */
 959
 960         spin_lock(&xprt->reserve_lock);
 961         list_add(&req->rq_list, &xprt->free);
 962         rpc_wake_up_next(&xprt->backlog);
 963         spin_unlock(&xprt->reserve_lock);
 964 }
 965
 966 /**
 967  * xprt_reserve - allocate an RPC request slot
 968  * @task: RPC task requesting a slot allocation
 969  *
 970  * If no more slots are available, place the task on the transport's
 971  * backlog queue.
 972  */
 973 void xprt_reserve(struct rpc_task *task)
 974 {
 975         struct rpc_xprt *xprt = task->tk_xprt;
 976
 977         task->tk_status = -EIO;
 978         spin_lock(&xprt->reserve_lock);
 979         xprt_alloc_slot(task);
 980         spin_unlock(&xprt->reserve_lock);
 981 }
 982
 983 static inline __be32 xprt_alloc_xid(struct rpc_xprt *xprt)
 984 {
 985         return xprt->xid++;
 986 }
 987
 988 static inline void xprt_init_xid(struct rpc_xprt *xprt)
 989 {
 990         xprt->xid = net_random();
 991 }
 992
 993 static void xprt_request_init(struct rpc_task *task, struct rpc_xprt *xprt)
 994 {
 995         struct rpc_rqst *req = task->tk_rqstp;
 996
 997         req->rq_timeout = task->tk_client->cl_timeout->to_initval;
 998         req->rq_task    = task;
 999         req->rq_xprt    = xprt;
1000         req->rq_buffer  = NULL;
1001         req->rq_xid     = xprt_alloc_xid(xprt);
1002         req->rq_release_snd_buf = NULL;
1003         xprt_reset_majortimeo(req);
1004         dprintk("RPC: %5u reserved req %p xid %08x\n", task->tk_pid,
1005                         req, ntohl(req->rq_xid));
1006 }
1007
1008 /**
1009  * xprt_release - release an RPC request slot
1010  * @task: task which is finished with the slot
1011  *
1012  */
1013 void xprt_release(struct rpc_task *task)
1014 {
1015         struct rpc_xprt *xprt;
1016         struct rpc_rqst *req;
1017
1018         if (!(req = task->tk_rqstp))
1019                 return;
1020
1021         xprt = req->rq_xprt;
1022         rpc_count_iostats(task);
1023         spin_lock_bh(&xprt->transport_lock);
1024         xprt->ops->release_xprt(xprt, task);
1025         if (xprt->ops->release_request)
1026                 xprt->ops->release_request(task);
1027         if (!list_empty(&req->rq_list))
1028                 list_del(&req->rq_list);
1029         xprt->last_used = jiffies;
1030         if (list_empty(&xprt->recv) && xprt_has_timer(xprt))
1031                 mod_timer(&xprt->timer,
1032                                 xprt->last_used + xprt->idle_timeout);
1033         spin_unlock_bh(&xprt->transport_lock);
1034         if (req->rq_buffer)
1035                 xprt->ops->buf_free(req->rq_buffer);
1036         task->tk_rqstp = NULL;
1037         if (req->rq_release_snd_buf)
1038                 req->rq_release_snd_buf(req);
1039
1040         dprintk("RPC: %5u release request %p\n", task->tk_pid, req);
1041         if (likely(!bc_prealloc(req)))
1042                 xprt_free_slot(xprt, req);
1043         else
1044                 xprt_free_bc_request(req);
1045 }
1046
1047 /**
1048  * xprt_create_transport - create an RPC transport
1049  * @args: rpc transport creation arguments
1050  *
1051  */
1052 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
1053 {
1054         struct rpc_xprt *xprt;
1055         struct rpc_rqst *req;
1056         struct xprt_class *t;
1057
1058         spin_lock(&xprt_list_lock);
1059         list_for_each_entry(t, &xprt_list, list) {
1060                 if (t->ident == args->ident) {
1061                         spin_unlock(&xprt_list_lock);
1062                         goto found;
1063                 }
1064         }
1065         spin_unlock(&xprt_list_lock);
1066         printk(KERN_ERR "RPC: transport (%d) not supported\n", args->ident);
1067         return ERR_PTR(-EIO);
1068
1069 found:
1070         xprt = t->setup(args);
1071         if (IS_ERR(xprt)) {
1072                 dprintk("RPC:       xprt_create_transport: failed, %ld\n",
1073                                 -PTR_ERR(xprt));
1074                 return xprt;
1075         }
1076
1077         kref_init(&xprt->kref);
1078         spin_lock_init(&xprt->transport_lock);
1079         spin_lock_init(&xprt->reserve_lock);
1080
1081         INIT_LIST_HEAD(&xprt->free);
1082         INIT_LIST_HEAD(&xprt->recv);
1083 #if defined(CONFIG_NFS_V4_1)
1084         spin_lock_init(&xprt->bc_pa_lock);
1085         INIT_LIST_HEAD(&xprt->bc_pa_list);
1086 #endif /* CONFIG_NFS_V4_1 */
1087
1088         INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
1089         if (xprt_has_timer(xprt))
1090                 setup_timer(&xprt->timer, xprt_init_autodisconnect,
1091                             (unsigned long)xprt);
1092         else
1093                 init_timer(&xprt->timer);
1094         xprt->last_used = jiffies;
1095         xprt->cwnd = RPC_INITCWND;
1096         xprt->bind_index = 0;
1097
1098         rpc_init_wait_queue(&xprt->binding, "xprt_binding");
1099         rpc_init_wait_queue(&xprt->pending, "xprt_pending");
1100         rpc_init_wait_queue(&xprt->sending, "xprt_sending");
1101         rpc_init_wait_queue(&xprt->resend, "xprt_resend");
1102         rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
1103
1104         /* initialize free list */
1105         for (req = &xprt->slot[xprt->max_reqs-1]; req >= &xprt->slot[0]; req--)
1106                 list_add(&req->rq_list, &xprt->free);
1107
1108         xprt_init_xid(xprt);
1109
1110         dprintk("RPC:       created transport %p with %u slots\n", xprt,
1111                         xprt->max_reqs);
1112         return xprt;
1113 }
1114
1115 /**
1116  * xprt_destroy - destroy an RPC transport, killing off all requests.
1117  * @kref: kref for the transport to destroy
1118  *
1119  */
1120 static void xprt_destroy(struct kref *kref)
1121 {
1122         struct rpc_xprt *xprt = container_of(kref, struct rpc_xprt, kref);
1123
1124         dprintk("RPC:       destroying transport %p\n", xprt);
1125         xprt->shutdown = 1;
1126         del_timer_sync(&xprt->timer);
1127
1128         rpc_destroy_wait_queue(&xprt->binding);
1129         rpc_destroy_wait_queue(&xprt->pending);
1130         rpc_destroy_wait_queue(&xprt->sending);
1131         rpc_destroy_wait_queue(&xprt->resend);
1132         rpc_destroy_wait_queue(&xprt->backlog);
1133         /*
1134          * Tear down transport state and free the rpc_xprt
1135          */
1136         xprt->ops->destroy(xprt);
1137 }
1138
1139 /**
1140  * xprt_put - release a reference to an RPC transport.
1141  * @xprt: pointer to the transport
1142  *
1143  */
1144 void xprt_put(struct rpc_xprt *xprt)
1145 {
1146         kref_put(&xprt->kref, xprt_destroy);
1147 }
1148
1149 /**
1150  * xprt_get - return a reference to an RPC transport.
1151  * @xprt: pointer to the transport
1152  *
1153  */
1154 struct rpc_xprt *xprt_get(struct rpc_xprt *xprt)
1155 {
1156         kref_get(&xprt->kref);
1157         return xprt;
1158 }