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
  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  */
  63 static void     xprt_request_init(struct rpc_task *, struct rpc_xprt *);
  64 static inline void      do_xprt_reserve(struct rpc_task *);
  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         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
 717 static 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  */
 752 struct 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 }
 767 EXPORT_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  */
 774 void 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 }
 787 EXPORT_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  */
 796 void 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 }
 815 EXPORT_SYMBOL_GPL(xprt_complete_rqst);
 816
 817 static 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
 835 /**
 836  * xprt_prepare_transmit - reserve the transport before sending a request
 837  * @task: RPC task about to send a request
 838  *
 839  */
 840 int xprt_prepare_transmit(struct rpc_task *task)
 841 {
 842         struct rpc_rqst *req = task->tk_rqstp;
 843         struct rpc_xprt *xprt = req->rq_xprt;
 844         int err = 0;
 845
 846         dprintk("RPC: %5u xprt_prepare_transmit\n", task->tk_pid);
 847
 848         spin_lock_bh(&xprt->transport_lock);
 849         if (req->rq_reply_bytes_recvd && !req->rq_bytes_sent) {
 850                 err = req->rq_reply_bytes_recvd;
 851                 goto out_unlock;
 852         }
 853         if (!xprt->ops->reserve_xprt(task))
 854                 err = -EAGAIN;
 855 out_unlock:
 856         spin_unlock_bh(&xprt->transport_lock);
 857         return err;
 858 }
 859
 860 void xprt_end_transmit(struct rpc_task *task)
 861 {
 862         xprt_release_write(task->tk_rqstp->rq_xprt, task);
 863 }
 864
 865 /**
 866  * xprt_transmit - send an RPC request on a transport
 867  * @task: controlling RPC task
 868  *
 869  * We have to copy the iovec because sendmsg fiddles with its contents.
 870  */
 871 void xprt_transmit(struct rpc_task *task)
 872 {
 873         struct rpc_rqst *req = task->tk_rqstp;
 874         struct rpc_xprt *xprt = req->rq_xprt;
 875         int status;
 876
 877         dprintk("RPC: %5u xprt_transmit(%u)\n", task->tk_pid, req->rq_slen);
 878
 879         if (!req->rq_reply_bytes_recvd) {
 880                 if (list_empty(&req->rq_list) && rpc_reply_expected(task)) {
 881                         /*
 882                          * Add to the list only if we're expecting a reply
 883                          */
 884                         spin_lock_bh(&xprt->transport_lock);
 885                         /* Update the softirq receive buffer */
 886                         memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
 887                                         sizeof(req->rq_private_buf));
 888                         /* Add request to the receive list */
 889                         list_add_tail(&req->rq_list, &xprt->recv);
 890                         spin_unlock_bh(&xprt->transport_lock);
 891                         xprt_reset_majortimeo(req);
 892                         /* Turn off autodisconnect */
 893                         del_singleshot_timer_sync(&xprt->timer);
 894                 }
 895         } else if (!req->rq_bytes_sent)
 896                 return;
 897
 898         req->rq_connect_cookie = xprt->connect_cookie;
 899         req->rq_xtime = jiffies;
 900         status = xprt->ops->send_request(task);
 901         if (status != 0) {
 902                 task->tk_status = status;
 903                 return;
 904         }
 905
 906         dprintk("RPC: %5u xmit complete\n", task->tk_pid);
 907         spin_lock_bh(&xprt->transport_lock);
 908
 909         xprt->ops->set_retrans_timeout(task);
 910
 911         xprt->stat.sends++;
 912         xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
 913         xprt->stat.bklog_u += xprt->backlog.qlen;
 914
 915         /* Don't race with disconnect */
 916         if (!xprt_connected(xprt))
 917                 task->tk_status = -ENOTCONN;
 918         else if (!req->rq_reply_bytes_recvd && rpc_reply_expected(task)) {
 919                 /*
 920                  * Sleep on the pending queue since
 921                  * we're expecting a reply.
 922                  */
 923                 rpc_sleep_on(&xprt->pending, task, xprt_timer);
 924         }
 925         spin_unlock_bh(&xprt->transport_lock);
 926 }
 927
 928 static inline void do_xprt_reserve(struct rpc_task *task)
 929 {
 930         struct rpc_xprt *xprt = task->tk_xprt;
 931
 932         task->tk_status = 0;
 933         if (task->tk_rqstp)
 934                 return;
 935         if (!list_empty(&xprt->free)) {
 936                 struct rpc_rqst *req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
 937                 list_del_init(&req->rq_list);
 938                 task->tk_rqstp = req;
 939                 xprt_request_init(task, xprt);
 940                 return;
 941         }
 942         dprintk("RPC:       waiting for request slot\n");
 943         task->tk_status = -EAGAIN;
 944         task->tk_timeout = 0;
 945         rpc_sleep_on(&xprt->backlog, task, NULL);
 946 }
 947
 948 /**
 949  * xprt_reserve - allocate an RPC request slot
 950  * @task: RPC task requesting a slot allocation
 951  *
 952  * If no more slots are available, place the task on the transport's
 953  * backlog queue.
 954  */
 955 void xprt_reserve(struct rpc_task *task)
 956 {
 957         struct rpc_xprt *xprt = task->tk_xprt;
 958
 959         task->tk_status = -EIO;
 960         spin_lock(&xprt->reserve_lock);
 961         do_xprt_reserve(task);
 962         spin_unlock(&xprt->reserve_lock);
 963 }
 964
 965 static inline __be32 xprt_alloc_xid(struct rpc_xprt *xprt)
 966 {
 967         return xprt->xid++;
 968 }
 969
 970 static inline void xprt_init_xid(struct rpc_xprt *xprt)
 971 {
 972         xprt->xid = net_random();
 973 }
 974
 975 static void xprt_request_init(struct rpc_task *task, struct rpc_xprt *xprt)
 976 {
 977         struct rpc_rqst *req = task->tk_rqstp;
 978
 979         req->rq_timeout = task->tk_client->cl_timeout->to_initval;
 980         req->rq_task    = task;
 981         req->rq_xprt    = xprt;
 982         req->rq_buffer  = NULL;
 983         req->rq_xid     = xprt_alloc_xid(xprt);
 984         req->rq_release_snd_buf = NULL;
 985         xprt_reset_majortimeo(req);
 986         dprintk("RPC: %5u reserved req %p xid %08x\n", task->tk_pid,
 987                         req, ntohl(req->rq_xid));
 988 }
 989
 990 /**
 991  * xprt_release - release an RPC request slot
 992  * @task: task which is finished with the slot
 993  *
 994  */
 995 void xprt_release(struct rpc_task *task)
 996 {
 997         struct rpc_xprt *xprt;
 998         struct rpc_rqst *req;
 999         int is_bc_request;
1000
1001         if (!(req = task->tk_rqstp))
1002                 return;
1003
1004         /* Preallocated backchannel request? */
1005         is_bc_request = bc_prealloc(req);
1006
1007         xprt = req->rq_xprt;
1008         rpc_count_iostats(task);
1009         spin_lock_bh(&xprt->transport_lock);
1010         xprt->ops->release_xprt(xprt, task);
1011         if (xprt->ops->release_request)
1012                 xprt->ops->release_request(task);
1013         if (!list_empty(&req->rq_list))
1014                 list_del(&req->rq_list);
1015         xprt->last_used = jiffies;
1016         if (list_empty(&xprt->recv))
1017                 mod_timer(&xprt->timer,
1018                                 xprt->last_used + xprt->idle_timeout);
1019         spin_unlock_bh(&xprt->transport_lock);
1020         if (!bc_prealloc(req))
1021                 xprt->ops->buf_free(req->rq_buffer);
1022         task->tk_rqstp = NULL;
1023         if (req->rq_release_snd_buf)
1024                 req->rq_release_snd_buf(req);
1025
1026         /*
1027          * Early exit if this is a backchannel preallocated request.
1028          * There is no need to have it added to the RPC slot list.
1029          */
1030         if (is_bc_request)
1031                 return;
1032
1033         memset(req, 0, sizeof(*req));   /* mark unused */
1034
1035         dprintk("RPC: %5u release request %p\n", task->tk_pid, req);
1036
1037         spin_lock(&xprt->reserve_lock);
1038         list_add(&req->rq_list, &xprt->free);
1039         rpc_wake_up_next(&xprt->backlog);
1040         spin_unlock(&xprt->reserve_lock);
1041 }
1042
1043 /**
1044  * xprt_create_transport - create an RPC transport
1045  * @args: rpc transport creation arguments
1046  *
1047  */
1048 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
1049 {
1050         struct rpc_xprt *xprt;
1051         struct rpc_rqst *req;
1052         struct xprt_class *t;
1053
1054         spin_lock(&xprt_list_lock);
1055         list_for_each_entry(t, &xprt_list, list) {
1056                 if (t->ident == args->ident) {
1057                         spin_unlock(&xprt_list_lock);
1058                         goto found;
1059                 }
1060         }
1061         spin_unlock(&xprt_list_lock);
1062         printk(KERN_ERR "RPC: transport (%d) not supported\n", args->ident);
1063         return ERR_PTR(-EIO);
1064
1065 found:
1066         xprt = t->setup(args);
1067         if (IS_ERR(xprt)) {
1068                 dprintk("RPC:       xprt_create_transport: failed, %ld\n",
1069                                 -PTR_ERR(xprt));
1070                 return xprt;
1071         }
1072
1073         kref_init(&xprt->kref);
1074         spin_lock_init(&xprt->transport_lock);
1075         spin_lock_init(&xprt->reserve_lock);
1076
1077         INIT_LIST_HEAD(&xprt->free);
1078         INIT_LIST_HEAD(&xprt->recv);
1079 #if defined(CONFIG_NFS_V4_1)
1080         spin_lock_init(&xprt->bc_pa_lock);
1081         INIT_LIST_HEAD(&xprt->bc_pa_list);
1082 #endif /* CONFIG_NFS_V4_1 */
1083
1084         INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
1085         setup_timer(&xprt->timer, xprt_init_autodisconnect,
1086                         (unsigned long)xprt);
1087         xprt->last_used = jiffies;
1088         xprt->cwnd = RPC_INITCWND;
1089         xprt->bind_index = 0;
1090
1091         rpc_init_wait_queue(&xprt->binding, "xprt_binding");
1092         rpc_init_wait_queue(&xprt->pending, "xprt_pending");
1093         rpc_init_wait_queue(&xprt->sending, "xprt_sending");
1094         rpc_init_wait_queue(&xprt->resend, "xprt_resend");
1095         rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
1096
1097         /* initialize free list */
1098         for (req = &xprt->slot[xprt->max_reqs-1]; req >= &xprt->slot[0]; req--)
1099                 list_add(&req->rq_list, &xprt->free);
1100
1101         xprt_init_xid(xprt);
1102
1103         dprintk("RPC:       created transport %p with %u slots\n", xprt,
1104                         xprt->max_reqs);
1105
1106         return xprt;
1107 }
1108
1109 /**
1110  * xprt_destroy - destroy an RPC transport, killing off all requests.
1111  * @kref: kref for the transport to destroy
1112  *
1113  */
1114 static void xprt_destroy(struct kref *kref)
1115 {
1116         struct rpc_xprt *xprt = container_of(kref, struct rpc_xprt, kref);
1117
1118         dprintk("RPC:       destroying transport %p\n", xprt);
1119         xprt->shutdown = 1;
1120         del_timer_sync(&xprt->timer);
1121
1122         rpc_destroy_wait_queue(&xprt->binding);
1123         rpc_destroy_wait_queue(&xprt->pending);
1124         rpc_destroy_wait_queue(&xprt->sending);
1125         rpc_destroy_wait_queue(&xprt->resend);
1126         rpc_destroy_wait_queue(&xprt->backlog);
1127         /*
1128          * Tear down transport state and free the rpc_xprt
1129          */
1130         xprt->ops->destroy(xprt);
1131 }
1132
1133 /**
1134  * xprt_put - release a reference to an RPC transport.
1135  * @xprt: pointer to the transport
1136  *
1137  */
1138 void xprt_put(struct rpc_xprt *xprt)
1139 {
1140         kref_put(&xprt->kref, xprt_destroy);
1141 }
1142
1143 /**
1144  * xprt_get - return a reference to an RPC transport.
1145  * @xprt: pointer to the transport
1146  *
1147  */
1148 struct rpc_xprt *xprt_get(struct rpc_xprt *xprt)
1149 {
1150         kref_get(&xprt->kref);
1151         return xprt;
1152 }