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