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 #include <linux/ktime.h>
  47
  48 #include <linux/sunrpc/clnt.h>
  49 #include <linux/sunrpc/metrics.h>
  50 #include <linux/sunrpc/bc_xprt.h>
  51 #include <linux/rcupdate.h>
  52
  53 #include <trace/events/sunrpc.h>
  54
  55 #include "sunrpc.h"
  56
  57 /*
  58  * Local variables
  59  */
  60
  61 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
  62 # define RPCDBG_FACILITY        RPCDBG_XPRT
  63 #endif
  64
  65 /*
  66  * Local functions
  67  */
  68 static void      xprt_init(struct rpc_xprt *xprt, struct net *net);
  69 static void     xprt_request_init(struct rpc_task *, struct rpc_xprt *);
  70 static void     xprt_connect_status(struct rpc_task *task);
  71 static int      __xprt_get_cong(struct rpc_xprt *, struct rpc_task *);
  72 static void     __xprt_put_cong(struct rpc_xprt *, struct rpc_rqst *);
  73 static void      xprt_destroy(struct rpc_xprt *xprt);
  74
  75 static DEFINE_SPINLOCK(xprt_list_lock);
  76 static LIST_HEAD(xprt_list);
  77
  78 /**
  79  * xprt_register_transport - register a transport implementation
  80  * @transport: transport to register
  81  *
  82  * If a transport implementation is loaded as a kernel module, it can
  83  * call this interface to make itself known to the RPC client.
  84  *
  85  * Returns:
  86  * 0:           transport successfully registered
  87  * -EEXIST:     transport already registered
  88  * -EINVAL:     transport module being unloaded
  89  */
  90 int xprt_register_transport(struct xprt_class *transport)
  91 {
  92         struct xprt_class *t;
  93         int result;
  94
  95         result = -EEXIST;
  96         spin_lock(&xprt_list_lock);
  97         list_for_each_entry(t, &xprt_list, list) {
  98                 /* don't register the same transport class twice */
  99                 if (t->ident == transport->ident)
 100                         goto out;
 101         }
 102
 103         list_add_tail(&transport->list, &xprt_list);
 104         printk(KERN_INFO "RPC: Registered %s transport module.\n",
 105                transport->name);
 106         result = 0;
 107
 108 out:
 109         spin_unlock(&xprt_list_lock);
 110         return result;
 111 }
 112 EXPORT_SYMBOL_GPL(xprt_register_transport);
 113
 114 /**
 115  * xprt_unregister_transport - unregister a transport implementation
 116  * @transport: transport to unregister
 117  *
 118  * Returns:
 119  * 0:           transport successfully unregistered
 120  * -ENOENT:     transport never registered
 121  */
 122 int xprt_unregister_transport(struct xprt_class *transport)
 123 {
 124         struct xprt_class *t;
 125         int result;
 126
 127         result = 0;
 128         spin_lock(&xprt_list_lock);
 129         list_for_each_entry(t, &xprt_list, list) {
 130                 if (t == transport) {
 131                         printk(KERN_INFO
 132                                 "RPC: Unregistered %s transport module.\n",
 133                                 transport->name);
 134                         list_del_init(&transport->list);
 135                         goto out;
 136                 }
 137         }
 138         result = -ENOENT;
 139
 140 out:
 141         spin_unlock(&xprt_list_lock);
 142         return result;
 143 }
 144 EXPORT_SYMBOL_GPL(xprt_unregister_transport);
 145
 146 /**
 147  * xprt_load_transport - load a transport implementation
 148  * @transport_name: transport to load
 149  *
 150  * Returns:
 151  * 0:           transport successfully loaded
 152  * -ENOENT:     transport module not available
 153  */
 154 int xprt_load_transport(const char *transport_name)
 155 {
 156         struct xprt_class *t;
 157         int result;
 158
 159         result = 0;
 160         spin_lock(&xprt_list_lock);
 161         list_for_each_entry(t, &xprt_list, list) {
 162                 if (strcmp(t->name, transport_name) == 0) {
 163                         spin_unlock(&xprt_list_lock);
 164                         goto out;
 165                 }
 166         }
 167         spin_unlock(&xprt_list_lock);
 168         result = request_module("xprt%s", transport_name);
 169 out:
 170         return result;
 171 }
 172 EXPORT_SYMBOL_GPL(xprt_load_transport);
 173
 174 /**
 175  * xprt_reserve_xprt - serialize write access to transports
 176  * @task: task that is requesting access to the transport
 177  * @xprt: pointer to the target transport
 178  *
 179  * This prevents mixing the payload of separate requests, and prevents
 180  * transport connects from colliding with writes.  No congestion control
 181  * is provided.
 182  */
 183 int xprt_reserve_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 184 {
 185         struct rpc_rqst *req = task->tk_rqstp;
 186         int priority;
 187
 188         if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
 189                 if (task == xprt->snd_task)
 190                         return 1;
 191                 goto out_sleep;
 192         }
 193         xprt->snd_task = task;
 194         if (req != NULL)
 195                 req->rq_ntrans++;
 196
 197         return 1;
 198
 199 out_sleep:
 200         dprintk("RPC: %5u failed to lock transport %p\n",
 201                         task->tk_pid, xprt);
 202         task->tk_timeout = 0;
 203         task->tk_status = -EAGAIN;
 204         if (req == NULL)
 205                 priority = RPC_PRIORITY_LOW;
 206         else if (!req->rq_ntrans)
 207                 priority = RPC_PRIORITY_NORMAL;
 208         else
 209                 priority = RPC_PRIORITY_HIGH;
 210         rpc_sleep_on_priority(&xprt->sending, task, NULL, priority);
 211         return 0;
 212 }
 213 EXPORT_SYMBOL_GPL(xprt_reserve_xprt);
 214
 215 static void xprt_clear_locked(struct rpc_xprt *xprt)
 216 {
 217         xprt->snd_task = NULL;
 218         if (!test_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
 219                 smp_mb__before_atomic();
 220                 clear_bit(XPRT_LOCKED, &xprt->state);
 221                 smp_mb__after_atomic();
 222         } else
 223                 queue_work(rpciod_workqueue, &xprt->task_cleanup);
 224 }
 225
 226 /*
 227  * xprt_reserve_xprt_cong - serialize write access to transports
 228  * @task: task that is requesting access to the transport
 229  *
 230  * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
 231  * integrated into the decision of whether a request is allowed to be
 232  * woken up and given access to the transport.
 233  */
 234 int xprt_reserve_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 235 {
 236         struct rpc_rqst *req = task->tk_rqstp;
 237         int priority;
 238
 239         if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
 240                 if (task == xprt->snd_task)
 241                         return 1;
 242                 goto out_sleep;
 243         }
 244         if (req == NULL) {
 245                 xprt->snd_task = task;
 246                 return 1;
 247         }
 248         if (__xprt_get_cong(xprt, task)) {
 249                 xprt->snd_task = task;
 250                 req->rq_ntrans++;
 251                 return 1;
 252         }
 253         xprt_clear_locked(xprt);
 254 out_sleep:
 255         if (req)
 256                 __xprt_put_cong(xprt, req);
 257         dprintk("RPC: %5u failed to lock transport %p\n", task->tk_pid, xprt);
 258         task->tk_timeout = 0;
 259         task->tk_status = -EAGAIN;
 260         if (req == NULL)
 261                 priority = RPC_PRIORITY_LOW;
 262         else if (!req->rq_ntrans)
 263                 priority = RPC_PRIORITY_NORMAL;
 264         else
 265                 priority = RPC_PRIORITY_HIGH;
 266         rpc_sleep_on_priority(&xprt->sending, task, NULL, priority);
 267         return 0;
 268 }
 269 EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong);
 270
 271 static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
 272 {
 273         int retval;
 274
 275         spin_lock_bh(&xprt->transport_lock);
 276         retval = xprt->ops->reserve_xprt(xprt, task);
 277         spin_unlock_bh(&xprt->transport_lock);
 278         return retval;
 279 }
 280
 281 static bool __xprt_lock_write_func(struct rpc_task *task, void *data)
 282 {
 283         struct rpc_xprt *xprt = data;
 284         struct rpc_rqst *req;
 285
 286         req = task->tk_rqstp;
 287         xprt->snd_task = task;
 288         if (req)
 289                 req->rq_ntrans++;
 290         return true;
 291 }
 292
 293 static void __xprt_lock_write_next(struct rpc_xprt *xprt)
 294 {
 295         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 296                 return;
 297
 298         if (rpc_wake_up_first(&xprt->sending, __xprt_lock_write_func, xprt))
 299                 return;
 300         xprt_clear_locked(xprt);
 301 }
 302
 303 static bool __xprt_lock_write_cong_func(struct rpc_task *task, void *data)
 304 {
 305         struct rpc_xprt *xprt = data;
 306         struct rpc_rqst *req;
 307
 308         req = task->tk_rqstp;
 309         if (req == NULL) {
 310                 xprt->snd_task = task;
 311                 return true;
 312         }
 313         if (__xprt_get_cong(xprt, task)) {
 314                 xprt->snd_task = task;
 315                 req->rq_ntrans++;
 316                 return true;
 317         }
 318         return false;
 319 }
 320
 321 static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt)
 322 {
 323         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 324                 return;
 325         if (RPCXPRT_CONGESTED(xprt))
 326                 goto out_unlock;
 327         if (rpc_wake_up_first(&xprt->sending, __xprt_lock_write_cong_func, xprt))
 328                 return;
 329 out_unlock:
 330         xprt_clear_locked(xprt);
 331 }
 332
 333 static void xprt_task_clear_bytes_sent(struct rpc_task *task)
 334 {
 335         if (task != NULL) {
 336                 struct rpc_rqst *req = task->tk_rqstp;
 337                 if (req != NULL)
 338                         req->rq_bytes_sent = 0;
 339         }
 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_task_clear_bytes_sent(task);
 353                 xprt_clear_locked(xprt);
 354                 __xprt_lock_write_next(xprt);
 355         }
 356 }
 357 EXPORT_SYMBOL_GPL(xprt_release_xprt);
 358
 359 /**
 360  * xprt_release_xprt_cong - allow other requests to use a transport
 361  * @xprt: transport with other tasks potentially waiting
 362  * @task: task that is releasing access to the transport
 363  *
 364  * Note that "task" can be NULL.  Another task is awoken to use the
 365  * transport if the transport's congestion window allows it.
 366  */
 367 void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 368 {
 369         if (xprt->snd_task == task) {
 370                 xprt_task_clear_bytes_sent(task);
 371                 xprt_clear_locked(xprt);
 372                 __xprt_lock_write_next_cong(xprt);
 373         }
 374 }
 375 EXPORT_SYMBOL_GPL(xprt_release_xprt_cong);
 376
 377 static inline void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
 378 {
 379         spin_lock_bh(&xprt->transport_lock);
 380         xprt->ops->release_xprt(xprt, task);
 381         spin_unlock_bh(&xprt->transport_lock);
 382 }
 383
 384 /*
 385  * Van Jacobson congestion avoidance. Check if the congestion window
 386  * overflowed. Put the task to sleep if this is the case.
 387  */
 388 static int
 389 __xprt_get_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 390 {
 391         struct rpc_rqst *req = task->tk_rqstp;
 392
 393         if (req->rq_cong)
 394                 return 1;
 395         dprintk("RPC: %5u xprt_cwnd_limited cong = %lu cwnd = %lu\n",
 396                         task->tk_pid, xprt->cong, xprt->cwnd);
 397         if (RPCXPRT_CONGESTED(xprt))
 398                 return 0;
 399         req->rq_cong = 1;
 400         xprt->cong += RPC_CWNDSCALE;
 401         return 1;
 402 }
 403
 404 /*
 405  * Adjust the congestion window, and wake up the next task
 406  * that has been sleeping due to congestion
 407  */
 408 static void
 409 __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
 410 {
 411         if (!req->rq_cong)
 412                 return;
 413         req->rq_cong = 0;
 414         xprt->cong -= RPC_CWNDSCALE;
 415         __xprt_lock_write_next_cong(xprt);
 416 }
 417
 418 /**
 419  * xprt_release_rqst_cong - housekeeping when request is complete
 420  * @task: RPC request that recently completed
 421  *
 422  * Useful for transports that require congestion control.
 423  */
 424 void xprt_release_rqst_cong(struct rpc_task *task)
 425 {
 426         struct rpc_rqst *req = task->tk_rqstp;
 427
 428         __xprt_put_cong(req->rq_xprt, req);
 429 }
 430 EXPORT_SYMBOL_GPL(xprt_release_rqst_cong);
 431
 432 /**
 433  * xprt_adjust_cwnd - adjust transport congestion window
 434  * @xprt: pointer to xprt
 435  * @task: recently completed RPC request used to adjust window
 436  * @result: result code of completed RPC request
 437  *
 438  * The transport code maintains an estimate on the maximum number of out-
 439  * standing RPC requests, using a smoothed version of the congestion
 440  * avoidance implemented in 44BSD. This is basically the Van Jacobson
 441  * congestion algorithm: If a retransmit occurs, the congestion window is
 442  * halved; otherwise, it is incremented by 1/cwnd when
 443  *
 444  *      -       a reply is received and
 445  *      -       a full number of requests are outstanding and
 446  *      -       the congestion window hasn't been updated recently.
 447  */
 448 void xprt_adjust_cwnd(struct rpc_xprt *xprt, struct rpc_task *task, int result)
 449 {
 450         struct rpc_rqst *req = task->tk_rqstp;
 451         unsigned long cwnd = xprt->cwnd;
 452
 453         if (result >= 0 && cwnd <= xprt->cong) {
 454                 /* The (cwnd >> 1) term makes sure
 455                  * the result gets rounded properly. */
 456                 cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
 457                 if (cwnd > RPC_MAXCWND(xprt))
 458                         cwnd = RPC_MAXCWND(xprt);
 459                 __xprt_lock_write_next_cong(xprt);
 460         } else if (result == -ETIMEDOUT) {
 461                 cwnd >>= 1;
 462                 if (cwnd < RPC_CWNDSCALE)
 463                         cwnd = RPC_CWNDSCALE;
 464         }
 465         dprintk("RPC:       cong %ld, cwnd was %ld, now %ld\n",
 466                         xprt->cong, xprt->cwnd, cwnd);
 467         xprt->cwnd = cwnd;
 468         __xprt_put_cong(xprt, req);
 469 }
 470 EXPORT_SYMBOL_GPL(xprt_adjust_cwnd);
 471
 472 /**
 473  * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
 474  * @xprt: transport with waiting tasks
 475  * @status: result code to plant in each task before waking it
 476  *
 477  */
 478 void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
 479 {
 480         if (status < 0)
 481                 rpc_wake_up_status(&xprt->pending, status);
 482         else
 483                 rpc_wake_up(&xprt->pending);
 484 }
 485 EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks);
 486
 487 /**
 488  * xprt_wait_for_buffer_space - wait for transport output buffer to clear
 489  * @task: task to be put to sleep
 490  * @action: function pointer to be executed after wait
 491  *
 492  * Note that we only set the timer for the case of RPC_IS_SOFT(), since
 493  * we don't in general want to force a socket disconnection due to
 494  * an incomplete RPC call transmission.
 495  */
 496 void xprt_wait_for_buffer_space(struct rpc_task *task, rpc_action action)
 497 {
 498         struct rpc_rqst *req = task->tk_rqstp;
 499         struct rpc_xprt *xprt = req->rq_xprt;
 500
 501         task->tk_timeout = RPC_IS_SOFT(task) ? req->rq_timeout : 0;
 502         rpc_sleep_on(&xprt->pending, task, action);
 503 }
 504 EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space);
 505
 506 /**
 507  * xprt_write_space - wake the task waiting for transport output buffer space
 508  * @xprt: transport with waiting tasks
 509  *
 510  * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
 511  */
 512 void xprt_write_space(struct rpc_xprt *xprt)
 513 {
 514         spin_lock_bh(&xprt->transport_lock);
 515         if (xprt->snd_task) {
 516                 dprintk("RPC:       write space: waking waiting task on "
 517                                 "xprt %p\n", xprt);
 518                 rpc_wake_up_queued_task(&xprt->pending, xprt->snd_task);
 519         }
 520         spin_unlock_bh(&xprt->transport_lock);
 521 }
 522 EXPORT_SYMBOL_GPL(xprt_write_space);
 523
 524 /**
 525  * xprt_set_retrans_timeout_def - set a request's retransmit timeout
 526  * @task: task whose timeout is to be set
 527  *
 528  * Set a request's retransmit timeout based on the transport's
 529  * default timeout parameters.  Used by transports that don't adjust
 530  * the retransmit timeout based on round-trip time estimation.
 531  */
 532 void xprt_set_retrans_timeout_def(struct rpc_task *task)
 533 {
 534         task->tk_timeout = task->tk_rqstp->rq_timeout;
 535 }
 536 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_def);
 537
 538 /**
 539  * xprt_set_retrans_timeout_rtt - set a request's retransmit timeout
 540  * @task: task whose timeout is to be set
 541  *
 542  * Set a request's retransmit timeout using the RTT estimator.
 543  */
 544 void xprt_set_retrans_timeout_rtt(struct rpc_task *task)
 545 {
 546         int timer = task->tk_msg.rpc_proc->p_timer;
 547         struct rpc_clnt *clnt = task->tk_client;
 548         struct rpc_rtt *rtt = clnt->cl_rtt;
 549         struct rpc_rqst *req = task->tk_rqstp;
 550         unsigned long max_timeout = clnt->cl_timeout->to_maxval;
 551
 552         task->tk_timeout = rpc_calc_rto(rtt, timer);
 553         task->tk_timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
 554         if (task->tk_timeout > max_timeout || task->tk_timeout == 0)
 555                 task->tk_timeout = max_timeout;
 556 }
 557 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_rtt);
 558
 559 static void xprt_reset_majortimeo(struct rpc_rqst *req)
 560 {
 561         const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
 562
 563         req->rq_majortimeo = req->rq_timeout;
 564         if (to->to_exponential)
 565                 req->rq_majortimeo <<= to->to_retries;
 566         else
 567                 req->rq_majortimeo += to->to_increment * to->to_retries;
 568         if (req->rq_majortimeo > to->to_maxval || req->rq_majortimeo == 0)
 569                 req->rq_majortimeo = to->to_maxval;
 570         req->rq_majortimeo += jiffies;
 571 }
 572
 573 /**
 574  * xprt_adjust_timeout - adjust timeout values for next retransmit
 575  * @req: RPC request containing parameters to use for the adjustment
 576  *
 577  */
 578 int xprt_adjust_timeout(struct rpc_rqst *req)
 579 {
 580         struct rpc_xprt *xprt = req->rq_xprt;
 581         const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
 582         int status = 0;
 583
 584         if (time_before(jiffies, req->rq_majortimeo)) {
 585                 if (to->to_exponential)
 586                         req->rq_timeout <<= 1;
 587                 else
 588                         req->rq_timeout += to->to_increment;
 589                 if (to->to_maxval && req->rq_timeout >= to->to_maxval)
 590                         req->rq_timeout = to->to_maxval;
 591                 req->rq_retries++;
 592         } else {
 593                 req->rq_timeout = to->to_initval;
 594                 req->rq_retries = 0;
 595                 xprt_reset_majortimeo(req);
 596                 /* Reset the RTT counters == "slow start" */
 597                 spin_lock_bh(&xprt->transport_lock);
 598                 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
 599                 spin_unlock_bh(&xprt->transport_lock);
 600                 status = -ETIMEDOUT;
 601         }
 602
 603         if (req->rq_timeout == 0) {
 604                 printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
 605                 req->rq_timeout = 5 * HZ;
 606         }
 607         return status;
 608 }
 609
 610 static void xprt_autoclose(struct work_struct *work)
 611 {
 612         struct rpc_xprt *xprt =
 613                 container_of(work, struct rpc_xprt, task_cleanup);
 614
 615         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 616         xprt->ops->close(xprt);
 617         xprt_release_write(xprt, NULL);
 618         wake_up_bit(&xprt->state, XPRT_LOCKED);
 619 }
 620
 621 /**
 622  * xprt_disconnect_done - mark a transport as disconnected
 623  * @xprt: transport to flag for disconnect
 624  *
 625  */
 626 void xprt_disconnect_done(struct rpc_xprt *xprt)
 627 {
 628         dprintk("RPC:       disconnected transport %p\n", xprt);
 629         spin_lock_bh(&xprt->transport_lock);
 630         xprt_clear_connected(xprt);
 631         xprt_wake_pending_tasks(xprt, -EAGAIN);
 632         spin_unlock_bh(&xprt->transport_lock);
 633 }
 634 EXPORT_SYMBOL_GPL(xprt_disconnect_done);
 635
 636 /**
 637  * xprt_force_disconnect - force a transport to disconnect
 638  * @xprt: transport to disconnect
 639  *
 640  */
 641 void xprt_force_disconnect(struct rpc_xprt *xprt)
 642 {
 643         /* Don't race with the test_bit() in xprt_clear_locked() */
 644         spin_lock_bh(&xprt->transport_lock);
 645         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 646         /* Try to schedule an autoclose RPC call */
 647         if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 648                 queue_work(rpciod_workqueue, &xprt->task_cleanup);
 649         xprt_wake_pending_tasks(xprt, -EAGAIN);
 650         spin_unlock_bh(&xprt->transport_lock);
 651 }
 652
 653 /**
 654  * xprt_conditional_disconnect - force a transport to disconnect
 655  * @xprt: transport to disconnect
 656  * @cookie: 'connection cookie'
 657  *
 658  * This attempts to break the connection if and only if 'cookie' matches
 659  * the current transport 'connection cookie'. It ensures that we don't
 660  * try to break the connection more than once when we need to retransmit
 661  * a batch of RPC requests.
 662  *
 663  */
 664 void xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie)
 665 {
 666         /* Don't race with the test_bit() in xprt_clear_locked() */
 667         spin_lock_bh(&xprt->transport_lock);
 668         if (cookie != xprt->connect_cookie)
 669                 goto out;
 670         if (test_bit(XPRT_CLOSING, &xprt->state) || !xprt_connected(xprt))
 671                 goto out;
 672         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 673         /* Try to schedule an autoclose RPC call */
 674         if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 675                 queue_work(rpciod_workqueue, &xprt->task_cleanup);
 676         xprt_wake_pending_tasks(xprt, -EAGAIN);
 677 out:
 678         spin_unlock_bh(&xprt->transport_lock);
 679 }
 680
 681 static void
 682 xprt_init_autodisconnect(unsigned long data)
 683 {
 684         struct rpc_xprt *xprt = (struct rpc_xprt *)data;
 685
 686         spin_lock(&xprt->transport_lock);
 687         if (!list_empty(&xprt->recv))
 688                 goto out_abort;
 689         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 690                 goto out_abort;
 691         spin_unlock(&xprt->transport_lock);
 692         queue_work(rpciod_workqueue, &xprt->task_cleanup);
 693         return;
 694 out_abort:
 695         spin_unlock(&xprt->transport_lock);
 696 }
 697
 698 bool xprt_lock_connect(struct rpc_xprt *xprt,
 699                 struct rpc_task *task,
 700                 void *cookie)
 701 {
 702         bool ret = false;
 703
 704         spin_lock_bh(&xprt->transport_lock);
 705         if (!test_bit(XPRT_LOCKED, &xprt->state))
 706                 goto out;
 707         if (xprt->snd_task != task)
 708                 goto out;
 709         xprt_task_clear_bytes_sent(task);
 710         xprt->snd_task = cookie;
 711         ret = true;
 712 out:
 713         spin_unlock_bh(&xprt->transport_lock);
 714         return ret;
 715 }
 716
 717 void xprt_unlock_connect(struct rpc_xprt *xprt, void *cookie)
 718 {
 719         spin_lock_bh(&xprt->transport_lock);
 720         if (xprt->snd_task != cookie)
 721                 goto out;
 722         if (!test_bit(XPRT_LOCKED, &xprt->state))
 723                 goto out;
 724         xprt->snd_task =NULL;
 725         xprt->ops->release_xprt(xprt, NULL);
 726 out:
 727         spin_unlock_bh(&xprt->transport_lock);
 728         wake_up_bit(&xprt->state, XPRT_LOCKED);
 729 }
 730
 731 /**
 732  * xprt_connect - schedule a transport connect operation
 733  * @task: RPC task that is requesting the connect
 734  *
 735  */
 736 void xprt_connect(struct rpc_task *task)
 737 {
 738         struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
 739
 740         dprintk("RPC: %5u xprt_connect xprt %p %s connected\n", task->tk_pid,
 741                         xprt, (xprt_connected(xprt) ? "is" : "is not"));
 742
 743         if (!xprt_bound(xprt)) {
 744                 task->tk_status = -EAGAIN;
 745                 return;
 746         }
 747         if (!xprt_lock_write(xprt, task))
 748                 return;
 749
 750         if (test_and_clear_bit(XPRT_CLOSE_WAIT, &xprt->state))
 751                 xprt->ops->close(xprt);
 752
 753         if (!xprt_connected(xprt)) {
 754                 task->tk_rqstp->rq_bytes_sent = 0;
 755                 task->tk_timeout = task->tk_rqstp->rq_timeout;
 756                 rpc_sleep_on(&xprt->pending, task, xprt_connect_status);
 757
 758                 if (test_bit(XPRT_CLOSING, &xprt->state))
 759                         return;
 760                 if (xprt_test_and_set_connecting(xprt))
 761                         return;
 762                 xprt->stat.connect_start = jiffies;
 763                 xprt->ops->connect(xprt, task);
 764         }
 765         xprt_release_write(xprt, task);
 766 }
 767
 768 static void xprt_connect_status(struct rpc_task *task)
 769 {
 770         struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
 771
 772         if (task->tk_status == 0) {
 773                 xprt->stat.connect_count++;
 774                 xprt->stat.connect_time += (long)jiffies - xprt->stat.connect_start;
 775                 dprintk("RPC: %5u xprt_connect_status: connection established\n",
 776                                 task->tk_pid);
 777                 return;
 778         }
 779
 780         switch (task->tk_status) {
 781         case -ECONNREFUSED:
 782         case -ECONNRESET:
 783         case -ECONNABORTED:
 784         case -ENETUNREACH:
 785         case -EHOSTUNREACH:
 786         case -EPIPE:
 787         case -EAGAIN:
 788                 dprintk("RPC: %5u xprt_connect_status: retrying\n", task->tk_pid);
 789                 break;
 790         case -ETIMEDOUT:
 791                 dprintk("RPC: %5u xprt_connect_status: connect attempt timed "
 792                                 "out\n", task->tk_pid);
 793                 break;
 794         default:
 795                 dprintk("RPC: %5u xprt_connect_status: error %d connecting to "
 796                                 "server %s\n", task->tk_pid, -task->tk_status,
 797                                 xprt->servername);
 798                 task->tk_status = -EIO;
 799         }
 800 }
 801
 802 /**
 803  * xprt_lookup_rqst - find an RPC request corresponding to an XID
 804  * @xprt: transport on which the original request was transmitted
 805  * @xid: RPC XID of incoming reply
 806  *
 807  */
 808 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
 809 {
 810         struct rpc_rqst *entry;
 811
 812         list_for_each_entry(entry, &xprt->recv, rq_list)
 813                 if (entry->rq_xid == xid) {
 814                         trace_xprt_lookup_rqst(xprt, xid, 0);
 815                         return entry;
 816                 }
 817
 818         dprintk("RPC:       xprt_lookup_rqst did not find xid %08x\n",
 819                         ntohl(xid));
 820         trace_xprt_lookup_rqst(xprt, xid, -ENOENT);
 821         xprt->stat.bad_xids++;
 822         return NULL;
 823 }
 824 EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
 825
 826 static void xprt_update_rtt(struct rpc_task *task)
 827 {
 828         struct rpc_rqst *req = task->tk_rqstp;
 829         struct rpc_rtt *rtt = task->tk_client->cl_rtt;
 830         unsigned int timer = task->tk_msg.rpc_proc->p_timer;
 831         long m = usecs_to_jiffies(ktime_to_us(req->rq_rtt));
 832
 833         if (timer) {
 834                 if (req->rq_ntrans == 1)
 835                         rpc_update_rtt(rtt, timer, m);
 836                 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
 837         }
 838 }
 839
 840 /**
 841  * xprt_complete_rqst - called when reply processing is complete
 842  * @task: RPC request that recently completed
 843  * @copied: actual number of bytes received from the transport
 844  *
 845  * Caller holds transport lock.
 846  */
 847 void xprt_complete_rqst(struct rpc_task *task, int copied)
 848 {
 849         struct rpc_rqst *req = task->tk_rqstp;
 850         struct rpc_xprt *xprt = req->rq_xprt;
 851
 852         dprintk("RPC: %5u xid %08x complete (%d bytes received)\n",
 853                         task->tk_pid, ntohl(req->rq_xid), copied);
 854         trace_xprt_complete_rqst(xprt, req->rq_xid, copied);
 855
 856         xprt->stat.recvs++;
 857         req->rq_rtt = ktime_sub(ktime_get(), req->rq_xtime);
 858         if (xprt->ops->timer != NULL)
 859                 xprt_update_rtt(task);
 860
 861         list_del_init(&req->rq_list);
 862         req->rq_private_buf.len = copied;
 863         /* Ensure all writes are done before we update */
 864         /* req->rq_reply_bytes_recvd */
 865         smp_wmb();
 866         req->rq_reply_bytes_recvd = copied;
 867         rpc_wake_up_queued_task(&xprt->pending, task);
 868 }
 869 EXPORT_SYMBOL_GPL(xprt_complete_rqst);
 870
 871 static void xprt_timer(struct rpc_task *task)
 872 {
 873         struct rpc_rqst *req = task->tk_rqstp;
 874         struct rpc_xprt *xprt = req->rq_xprt;
 875
 876         if (task->tk_status != -ETIMEDOUT)
 877                 return;
 878         dprintk("RPC: %5u xprt_timer\n", task->tk_pid);
 879
 880         spin_lock_bh(&xprt->transport_lock);
 881         if (!req->rq_reply_bytes_recvd) {
 882                 if (xprt->ops->timer)
 883                         xprt->ops->timer(xprt, task);
 884         } else
 885                 task->tk_status = 0;
 886         spin_unlock_bh(&xprt->transport_lock);
 887 }
 888
 889 static inline int xprt_has_timer(struct rpc_xprt *xprt)
 890 {
 891         return xprt->idle_timeout != 0;
 892 }
 893
 894 /**
 895  * xprt_prepare_transmit - reserve the transport before sending a request
 896  * @task: RPC task about to send a request
 897  *
 898  */
 899 bool xprt_prepare_transmit(struct rpc_task *task)
 900 {
 901         struct rpc_rqst *req = task->tk_rqstp;
 902         struct rpc_xprt *xprt = req->rq_xprt;
 903         bool ret = false;
 904
 905         dprintk("RPC: %5u xprt_prepare_transmit\n", task->tk_pid);
 906
 907         spin_lock_bh(&xprt->transport_lock);
 908         if (!req->rq_bytes_sent) {
 909                 if (req->rq_reply_bytes_recvd) {
 910                         task->tk_status = req->rq_reply_bytes_recvd;
 911                         goto out_unlock;
 912                 }
 913                 if ((task->tk_flags & RPC_TASK_NO_RETRANS_TIMEOUT)
 914                     && xprt_connected(xprt)
 915                     && req->rq_connect_cookie == xprt->connect_cookie) {
 916                         xprt->ops->set_retrans_timeout(task);
 917                         rpc_sleep_on(&xprt->pending, task, xprt_timer);
 918                         goto out_unlock;
 919                 }
 920         }
 921         if (!xprt->ops->reserve_xprt(xprt, task)) {
 922                 task->tk_status = -EAGAIN;
 923                 goto out_unlock;
 924         }
 925         ret = true;
 926 out_unlock:
 927         spin_unlock_bh(&xprt->transport_lock);
 928         return ret;
 929 }
 930
 931 void xprt_end_transmit(struct rpc_task *task)
 932 {
 933         xprt_release_write(task->tk_rqstp->rq_xprt, task);
 934 }
 935
 936 /**
 937  * xprt_transmit - send an RPC request on a transport
 938  * @task: controlling RPC task
 939  *
 940  * We have to copy the iovec because sendmsg fiddles with its contents.
 941  */
 942 void xprt_transmit(struct rpc_task *task)
 943 {
 944         struct rpc_rqst *req = task->tk_rqstp;
 945         struct rpc_xprt *xprt = req->rq_xprt;
 946         int status, numreqs;
 947
 948         dprintk("RPC: %5u xprt_transmit(%u)\n", task->tk_pid, req->rq_slen);
 949
 950         if (!req->rq_reply_bytes_recvd) {
 951                 if (list_empty(&req->rq_list) && rpc_reply_expected(task)) {
 952                         /*
 953                          * Add to the list only if we're expecting a reply
 954                          */
 955                         spin_lock_bh(&xprt->transport_lock);
 956                         /* Update the softirq receive buffer */
 957                         memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
 958                                         sizeof(req->rq_private_buf));
 959                         /* Add request to the receive list */
 960                         list_add_tail(&req->rq_list, &xprt->recv);
 961                         spin_unlock_bh(&xprt->transport_lock);
 962                         xprt_reset_majortimeo(req);
 963                         /* Turn off autodisconnect */
 964                         del_singleshot_timer_sync(&xprt->timer);
 965                 }
 966         } else if (!req->rq_bytes_sent)
 967                 return;
 968
 969         req->rq_xtime = ktime_get();
 970         status = xprt->ops->send_request(task);
 971         trace_xprt_transmit(xprt, req->rq_xid, status);
 972         if (status != 0) {
 973                 task->tk_status = status;
 974                 return;
 975         }
 976         xprt_inject_disconnect(xprt);
 977
 978         dprintk("RPC: %5u xmit complete\n", task->tk_pid);
 979         task->tk_flags |= RPC_TASK_SENT;
 980         spin_lock_bh(&xprt->transport_lock);
 981
 982         xprt->ops->set_retrans_timeout(task);
 983
 984         numreqs = atomic_read(&xprt->num_reqs);
 985         if (numreqs > xprt->stat.max_slots)
 986                 xprt->stat.max_slots = numreqs;
 987         xprt->stat.sends++;
 988         xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
 989         xprt->stat.bklog_u += xprt->backlog.qlen;
 990         xprt->stat.sending_u += xprt->sending.qlen;
 991         xprt->stat.pending_u += xprt->pending.qlen;
 992
 993         /* Don't race with disconnect */
 994         if (!xprt_connected(xprt))
 995                 task->tk_status = -ENOTCONN;
 996         else {
 997                 /*
 998                  * Sleep on the pending queue since
 999                  * we're expecting a reply.
1000                  */
1001                 if (!req->rq_reply_bytes_recvd && rpc_reply_expected(task))
1002                         rpc_sleep_on(&xprt->pending, task, xprt_timer);
1003                 req->rq_connect_cookie = xprt->connect_cookie;
1004         }
1005         spin_unlock_bh(&xprt->transport_lock);
1006 }
1007
1008 static void xprt_add_backlog(struct rpc_xprt *xprt, struct rpc_task *task)
1009 {
1010         set_bit(XPRT_CONGESTED, &xprt->state);
1011         rpc_sleep_on(&xprt->backlog, task, NULL);
1012 }
1013
1014 static void xprt_wake_up_backlog(struct rpc_xprt *xprt)
1015 {
1016         if (rpc_wake_up_next(&xprt->backlog) == NULL)
1017                 clear_bit(XPRT_CONGESTED, &xprt->state);
1018 }
1019
1020 static bool xprt_throttle_congested(struct rpc_xprt *xprt, struct rpc_task *task)
1021 {
1022         bool ret = false;
1023
1024         if (!test_bit(XPRT_CONGESTED, &xprt->state))
1025                 goto out;
1026         spin_lock(&xprt->reserve_lock);
1027         if (test_bit(XPRT_CONGESTED, &xprt->state)) {
1028                 rpc_sleep_on(&xprt->backlog, task, NULL);
1029                 ret = true;
1030         }
1031         spin_unlock(&xprt->reserve_lock);
1032 out:
1033         return ret;
1034 }
1035
1036 static struct rpc_rqst *xprt_dynamic_alloc_slot(struct rpc_xprt *xprt, gfp_t gfp_flags)
1037 {
1038         struct rpc_rqst *req = ERR_PTR(-EAGAIN);
1039
1040         if (!atomic_add_unless(&xprt->num_reqs, 1, xprt->max_reqs))
1041                 goto out;
1042         req = kzalloc(sizeof(struct rpc_rqst), gfp_flags);
1043         if (req != NULL)
1044                 goto out;
1045         atomic_dec(&xprt->num_reqs);
1046         req = ERR_PTR(-ENOMEM);
1047 out:
1048         return req;
1049 }
1050
1051 static bool xprt_dynamic_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1052 {
1053         if (atomic_add_unless(&xprt->num_reqs, -1, xprt->min_reqs)) {
1054                 kfree(req);
1055                 return true;
1056         }
1057         return false;
1058 }
1059
1060 void xprt_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1061 {
1062         struct rpc_rqst *req;
1063
1064         spin_lock(&xprt->reserve_lock);
1065         if (!list_empty(&xprt->free)) {
1066                 req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
1067                 list_del(&req->rq_list);
1068                 goto out_init_req;
1069         }
1070         req = xprt_dynamic_alloc_slot(xprt, GFP_NOWAIT|__GFP_NOWARN);
1071         if (!IS_ERR(req))
1072                 goto out_init_req;
1073         switch (PTR_ERR(req)) {
1074         case -ENOMEM:
1075                 dprintk("RPC:       dynamic allocation of request slot "
1076                                 "failed! Retrying\n");
1077                 task->tk_status = -ENOMEM;
1078                 break;
1079         case -EAGAIN:
1080                 xprt_add_backlog(xprt, task);
1081                 dprintk("RPC:       waiting for request slot\n");
1082         default:
1083                 task->tk_status = -EAGAIN;
1084         }
1085         spin_unlock(&xprt->reserve_lock);
1086         return;
1087 out_init_req:
1088         task->tk_status = 0;
1089         task->tk_rqstp = req;
1090         xprt_request_init(task, xprt);
1091         spin_unlock(&xprt->reserve_lock);
1092 }
1093 EXPORT_SYMBOL_GPL(xprt_alloc_slot);
1094
1095 void xprt_lock_and_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1096 {
1097         /* Note: grabbing the xprt_lock_write() ensures that we throttle
1098          * new slot allocation if the transport is congested (i.e. when
1099          * reconnecting a stream transport or when out of socket write
1100          * buffer space).
1101          */
1102         if (xprt_lock_write(xprt, task)) {
1103                 xprt_alloc_slot(xprt, task);
1104                 xprt_release_write(xprt, task);
1105         }
1106 }
1107 EXPORT_SYMBOL_GPL(xprt_lock_and_alloc_slot);
1108
1109 static void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1110 {
1111         spin_lock(&xprt->reserve_lock);
1112         if (!xprt_dynamic_free_slot(xprt, req)) {
1113                 memset(req, 0, sizeof(*req));   /* mark unused */
1114                 list_add(&req->rq_list, &xprt->free);
1115         }
1116         xprt_wake_up_backlog(xprt);
1117         spin_unlock(&xprt->reserve_lock);
1118 }
1119
1120 static void xprt_free_all_slots(struct rpc_xprt *xprt)
1121 {
1122         struct rpc_rqst *req;
1123         while (!list_empty(&xprt->free)) {
1124                 req = list_first_entry(&xprt->free, struct rpc_rqst, rq_list);
1125                 list_del(&req->rq_list);
1126                 kfree(req);
1127         }
1128 }
1129
1130 struct rpc_xprt *xprt_alloc(struct net *net, size_t size,
1131                 unsigned int num_prealloc,
1132                 unsigned int max_alloc)
1133 {
1134         struct rpc_xprt *xprt;
1135         struct rpc_rqst *req;
1136         int i;
1137
1138         xprt = kzalloc(size, GFP_KERNEL);
1139         if (xprt == NULL)
1140                 goto out;
1141
1142         xprt_init(xprt, net);
1143
1144         for (i = 0; i < num_prealloc; i++) {
1145                 req = kzalloc(sizeof(struct rpc_rqst), GFP_KERNEL);
1146                 if (!req)
1147                         goto out_free;
1148                 list_add(&req->rq_list, &xprt->free);
1149         }
1150         if (max_alloc > num_prealloc)
1151                 xprt->max_reqs = max_alloc;
1152         else
1153                 xprt->max_reqs = num_prealloc;
1154         xprt->min_reqs = num_prealloc;
1155         atomic_set(&xprt->num_reqs, num_prealloc);
1156
1157         return xprt;
1158
1159 out_free:
1160         xprt_free(xprt);
1161 out:
1162         return NULL;
1163 }
1164 EXPORT_SYMBOL_GPL(xprt_alloc);
1165
1166 void xprt_free(struct rpc_xprt *xprt)
1167 {
1168         put_net(xprt->xprt_net);
1169         xprt_free_all_slots(xprt);
1170         kfree_rcu(xprt, rcu);
1171 }
1172 EXPORT_SYMBOL_GPL(xprt_free);
1173
1174 /**
1175  * xprt_reserve - allocate an RPC request slot
1176  * @task: RPC task requesting a slot allocation
1177  *
1178  * If the transport is marked as being congested, or if no more
1179  * slots are available, place the task on the transport's
1180  * backlog queue.
1181  */
1182 void xprt_reserve(struct rpc_task *task)
1183 {
1184         struct rpc_xprt *xprt = task->tk_xprt;
1185
1186         task->tk_status = 0;
1187         if (task->tk_rqstp != NULL)
1188                 return;
1189
1190         task->tk_timeout = 0;
1191         task->tk_status = -EAGAIN;
1192         if (!xprt_throttle_congested(xprt, task))
1193                 xprt->ops->alloc_slot(xprt, task);
1194 }
1195
1196 /**
1197  * xprt_retry_reserve - allocate an RPC request slot
1198  * @task: RPC task requesting a slot allocation
1199  *
1200  * If no more slots are available, place the task on the transport's
1201  * backlog queue.
1202  * Note that the only difference with xprt_reserve is that we now
1203  * ignore the value of the XPRT_CONGESTED flag.
1204  */
1205 void xprt_retry_reserve(struct rpc_task *task)
1206 {
1207         struct rpc_xprt *xprt = task->tk_xprt;
1208
1209         task->tk_status = 0;
1210         if (task->tk_rqstp != NULL)
1211                 return;
1212
1213         task->tk_timeout = 0;
1214         task->tk_status = -EAGAIN;
1215         xprt->ops->alloc_slot(xprt, task);
1216 }
1217
1218 static inline __be32 xprt_alloc_xid(struct rpc_xprt *xprt)
1219 {
1220         return (__force __be32)xprt->xid++;
1221 }
1222
1223 static inline void xprt_init_xid(struct rpc_xprt *xprt)
1224 {
1225         xprt->xid = prandom_u32();
1226 }
1227
1228 static void xprt_request_init(struct rpc_task *task, struct rpc_xprt *xprt)
1229 {
1230         struct rpc_rqst *req = task->tk_rqstp;
1231
1232         INIT_LIST_HEAD(&req->rq_list);
1233         req->rq_timeout = task->tk_client->cl_timeout->to_initval;
1234         req->rq_task    = task;
1235         req->rq_xprt    = xprt;
1236         req->rq_buffer  = NULL;
1237         req->rq_xid     = xprt_alloc_xid(xprt);
1238         req->rq_connect_cookie = xprt->connect_cookie - 1;
1239         req->rq_bytes_sent = 0;
1240         req->rq_snd_buf.len = 0;
1241         req->rq_snd_buf.buflen = 0;
1242         req->rq_rcv_buf.len = 0;
1243         req->rq_rcv_buf.buflen = 0;
1244         req->rq_release_snd_buf = NULL;
1245         xprt_reset_majortimeo(req);
1246         dprintk("RPC: %5u reserved req %p xid %08x\n", task->tk_pid,
1247                         req, ntohl(req->rq_xid));
1248 }
1249
1250 /**
1251  * xprt_release - release an RPC request slot
1252  * @task: task which is finished with the slot
1253  *
1254  */
1255 void xprt_release(struct rpc_task *task)
1256 {
1257         struct rpc_xprt *xprt;
1258         struct rpc_rqst *req = task->tk_rqstp;
1259
1260         if (req == NULL) {
1261                 if (task->tk_client) {
1262                         xprt = task->tk_xprt;
1263                         if (xprt->snd_task == task)
1264                                 xprt_release_write(xprt, task);
1265                 }
1266                 return;
1267         }
1268
1269         xprt = req->rq_xprt;
1270         if (task->tk_ops->rpc_count_stats != NULL)
1271                 task->tk_ops->rpc_count_stats(task, task->tk_calldata);
1272         else if (task->tk_client)
1273                 rpc_count_iostats(task, task->tk_client->cl_metrics);
1274         spin_lock_bh(&xprt->transport_lock);
1275         xprt->ops->release_xprt(xprt, task);
1276         if (xprt->ops->release_request)
1277                 xprt->ops->release_request(task);
1278         if (!list_empty(&req->rq_list))
1279                 list_del(&req->rq_list);
1280         xprt->last_used = jiffies;
1281         if (list_empty(&xprt->recv) && xprt_has_timer(xprt))
1282                 mod_timer(&xprt->timer,
1283                                 xprt->last_used + xprt->idle_timeout);
1284         spin_unlock_bh(&xprt->transport_lock);
1285         if (req->rq_buffer)
1286                 xprt->ops->buf_free(req->rq_buffer);
1287         xprt_inject_disconnect(xprt);
1288         if (req->rq_cred != NULL)
1289                 put_rpccred(req->rq_cred);
1290         task->tk_rqstp = NULL;
1291         if (req->rq_release_snd_buf)
1292                 req->rq_release_snd_buf(req);
1293
1294         dprintk("RPC: %5u release request %p\n", task->tk_pid, req);
1295         if (likely(!bc_prealloc(req)))
1296                 xprt_free_slot(xprt, req);
1297         else
1298                 xprt_free_bc_request(req);
1299 }
1300
1301 static void xprt_init(struct rpc_xprt *xprt, struct net *net)
1302 {
1303         kref_init(&xprt->kref);
1304
1305         spin_lock_init(&xprt->transport_lock);
1306         spin_lock_init(&xprt->reserve_lock);
1307
1308         INIT_LIST_HEAD(&xprt->free);
1309         INIT_LIST_HEAD(&xprt->recv);
1310 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1311         spin_lock_init(&xprt->bc_pa_lock);
1312         INIT_LIST_HEAD(&xprt->bc_pa_list);
1313 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1314         INIT_LIST_HEAD(&xprt->xprt_switch);
1315
1316         xprt->last_used = jiffies;
1317         xprt->cwnd = RPC_INITCWND;
1318         xprt->bind_index = 0;
1319
1320         rpc_init_wait_queue(&xprt->binding, "xprt_binding");
1321         rpc_init_wait_queue(&xprt->pending, "xprt_pending");
1322         rpc_init_priority_wait_queue(&xprt->sending, "xprt_sending");
1323         rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
1324
1325         xprt_init_xid(xprt);
1326
1327         xprt->xprt_net = get_net(net);
1328 }
1329
1330 /**
1331  * xprt_create_transport - create an RPC transport
1332  * @args: rpc transport creation arguments
1333  *
1334  */
1335 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
1336 {
1337         struct rpc_xprt *xprt;
1338         struct xprt_class *t;
1339
1340         spin_lock(&xprt_list_lock);
1341         list_for_each_entry(t, &xprt_list, list) {
1342                 if (t->ident == args->ident) {
1343                         spin_unlock(&xprt_list_lock);
1344                         goto found;
1345                 }
1346         }
1347         spin_unlock(&xprt_list_lock);
1348         dprintk("RPC: transport (%d) not supported\n", args->ident);
1349         return ERR_PTR(-EIO);
1350
1351 found:
1352         xprt = t->setup(args);
1353         if (IS_ERR(xprt)) {
1354                 dprintk("RPC:       xprt_create_transport: failed, %ld\n",
1355                                 -PTR_ERR(xprt));
1356                 goto out;
1357         }
1358         if (args->flags & XPRT_CREATE_NO_IDLE_TIMEOUT)
1359                 xprt->idle_timeout = 0;
1360         INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
1361         if (xprt_has_timer(xprt))
1362                 setup_timer(&xprt->timer, xprt_init_autodisconnect,
1363                             (unsigned long)xprt);
1364         else
1365                 init_timer(&xprt->timer);
1366
1367         if (strlen(args->servername) > RPC_MAXNETNAMELEN) {
1368                 xprt_destroy(xprt);
1369                 return ERR_PTR(-EINVAL);
1370         }
1371         xprt->servername = kstrdup(args->servername, GFP_KERNEL);
1372         if (xprt->servername == NULL) {
1373                 xprt_destroy(xprt);
1374                 return ERR_PTR(-ENOMEM);
1375         }
1376
1377         rpc_xprt_debugfs_register(xprt);
1378
1379         dprintk("RPC:       created transport %p with %u slots\n", xprt,
1380                         xprt->max_reqs);
1381 out:
1382         return xprt;
1383 }
1384
1385 /**
1386  * xprt_destroy - destroy an RPC transport, killing off all requests.
1387  * @xprt: transport to destroy
1388  *
1389  */
1390 static void xprt_destroy(struct rpc_xprt *xprt)
1391 {
1392         dprintk("RPC:       destroying transport %p\n", xprt);
1393
1394         /* Exclude transport connect/disconnect handlers */
1395         wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_UNINTERRUPTIBLE);
1396
1397         del_timer_sync(&xprt->timer);
1398
1399         rpc_xprt_debugfs_unregister(xprt);
1400         rpc_destroy_wait_queue(&xprt->binding);
1401         rpc_destroy_wait_queue(&xprt->pending);
1402         rpc_destroy_wait_queue(&xprt->sending);
1403         rpc_destroy_wait_queue(&xprt->backlog);
1404         cancel_work_sync(&xprt->task_cleanup);
1405         kfree(xprt->servername);
1406         /*
1407          * Tear down transport state and free the rpc_xprt
1408          */
1409         xprt->ops->destroy(xprt);
1410 }
1411
1412 static void xprt_destroy_kref(struct kref *kref)
1413 {
1414         xprt_destroy(container_of(kref, struct rpc_xprt, kref));
1415 }
1416
1417 /**
1418  * xprt_get - return a reference to an RPC transport.
1419  * @xprt: pointer to the transport
1420  *
1421  */
1422 struct rpc_xprt *xprt_get(struct rpc_xprt *xprt)
1423 {
1424         if (xprt != NULL && kref_get_unless_zero(&xprt->kref))
1425                 return xprt;
1426         return NULL;
1427 }
1428 EXPORT_SYMBOL_GPL(xprt_get);
1429
1430 /**
1431  * xprt_put - release a reference to an RPC transport.
1432  * @xprt: pointer to the transport
1433  *
1434  */
1435 void xprt_put(struct rpc_xprt *xprt)
1436 {
1437         if (xprt != NULL)
1438                 kref_put(&xprt->kref, xprt_destroy_kref);
1439 }
1440 EXPORT_SYMBOL_GPL(xprt_put);