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(xprtiod_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_on_wq(xprtiod_workqueue, &xprt->sending,
 299                                 __xprt_lock_write_func, xprt))
 300                 return;
 301         xprt_clear_locked(xprt);
 302 }
 303
 304 static bool __xprt_lock_write_cong_func(struct rpc_task *task, void *data)
 305 {
 306         struct rpc_xprt *xprt = data;
 307         struct rpc_rqst *req;
 308
 309         req = task->tk_rqstp;
 310         if (req == NULL) {
 311                 xprt->snd_task = task;
 312                 return true;
 313         }
 314         if (__xprt_get_cong(xprt, task)) {
 315                 xprt->snd_task = task;
 316                 req->rq_ntrans++;
 317                 return true;
 318         }
 319         return false;
 320 }
 321
 322 static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt)
 323 {
 324         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 325                 return;
 326         if (RPCXPRT_CONGESTED(xprt))
 327                 goto out_unlock;
 328         if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
 329                                 __xprt_lock_write_cong_func, xprt))
 330                 return;
 331 out_unlock:
 332         xprt_clear_locked(xprt);
 333 }
 334
 335 static void xprt_task_clear_bytes_sent(struct rpc_task *task)
 336 {
 337         if (task != NULL) {
 338                 struct rpc_rqst *req = task->tk_rqstp;
 339                 if (req != NULL)
 340                         req->rq_bytes_sent = 0;
 341         }
 342 }
 343
 344 /**
 345  * xprt_release_xprt - allow other requests to use a transport
 346  * @xprt: transport with other tasks potentially waiting
 347  * @task: task that is releasing access to the transport
 348  *
 349  * Note that "task" can be NULL.  No congestion control is provided.
 350  */
 351 void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 352 {
 353         if (xprt->snd_task == task) {
 354                 xprt_task_clear_bytes_sent(task);
 355                 xprt_clear_locked(xprt);
 356                 __xprt_lock_write_next(xprt);
 357         }
 358 }
 359 EXPORT_SYMBOL_GPL(xprt_release_xprt);
 360
 361 /**
 362  * xprt_release_xprt_cong - allow other requests to use a transport
 363  * @xprt: transport with other tasks potentially waiting
 364  * @task: task that is releasing access to the transport
 365  *
 366  * Note that "task" can be NULL.  Another task is awoken to use the
 367  * transport if the transport's congestion window allows it.
 368  */
 369 void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 370 {
 371         if (xprt->snd_task == task) {
 372                 xprt_task_clear_bytes_sent(task);
 373                 xprt_clear_locked(xprt);
 374                 __xprt_lock_write_next_cong(xprt);
 375         }
 376 }
 377 EXPORT_SYMBOL_GPL(xprt_release_xprt_cong);
 378
 379 static inline void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
 380 {
 381         spin_lock_bh(&xprt->transport_lock);
 382         xprt->ops->release_xprt(xprt, task);
 383         spin_unlock_bh(&xprt->transport_lock);
 384 }
 385
 386 /*
 387  * Van Jacobson congestion avoidance. Check if the congestion window
 388  * overflowed. Put the task to sleep if this is the case.
 389  */
 390 static int
 391 __xprt_get_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 392 {
 393         struct rpc_rqst *req = task->tk_rqstp;
 394
 395         if (req->rq_cong)
 396                 return 1;
 397         dprintk("RPC: %5u xprt_cwnd_limited cong = %lu cwnd = %lu\n",
 398                         task->tk_pid, xprt->cong, xprt->cwnd);
 399         if (RPCXPRT_CONGESTED(xprt))
 400                 return 0;
 401         req->rq_cong = 1;
 402         xprt->cong += RPC_CWNDSCALE;
 403         return 1;
 404 }
 405
 406 /*
 407  * Adjust the congestion window, and wake up the next task
 408  * that has been sleeping due to congestion
 409  */
 410 static void
 411 __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
 412 {
 413         if (!req->rq_cong)
 414                 return;
 415         req->rq_cong = 0;
 416         xprt->cong -= RPC_CWNDSCALE;
 417         __xprt_lock_write_next_cong(xprt);
 418 }
 419
 420 /**
 421  * xprt_release_rqst_cong - housekeeping when request is complete
 422  * @task: RPC request that recently completed
 423  *
 424  * Useful for transports that require congestion control.
 425  */
 426 void xprt_release_rqst_cong(struct rpc_task *task)
 427 {
 428         struct rpc_rqst *req = task->tk_rqstp;
 429
 430         __xprt_put_cong(req->rq_xprt, req);
 431 }
 432 EXPORT_SYMBOL_GPL(xprt_release_rqst_cong);
 433
 434 /**
 435  * xprt_adjust_cwnd - adjust transport congestion window
 436  * @xprt: pointer to xprt
 437  * @task: recently completed RPC request used to adjust window
 438  * @result: result code of completed RPC request
 439  *
 440  * The transport code maintains an estimate on the maximum number of out-
 441  * standing RPC requests, using a smoothed version of the congestion
 442  * avoidance implemented in 44BSD. This is basically the Van Jacobson
 443  * congestion algorithm: If a retransmit occurs, the congestion window is
 444  * halved; otherwise, it is incremented by 1/cwnd when
 445  *
 446  *      -       a reply is received and
 447  *      -       a full number of requests are outstanding and
 448  *      -       the congestion window hasn't been updated recently.
 449  */
 450 void xprt_adjust_cwnd(struct rpc_xprt *xprt, struct rpc_task *task, int result)
 451 {
 452         struct rpc_rqst *req = task->tk_rqstp;
 453         unsigned long cwnd = xprt->cwnd;
 454
 455         if (result >= 0 && cwnd <= xprt->cong) {
 456                 /* The (cwnd >> 1) term makes sure
 457                  * the result gets rounded properly. */
 458                 cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
 459                 if (cwnd > RPC_MAXCWND(xprt))
 460                         cwnd = RPC_MAXCWND(xprt);
 461                 __xprt_lock_write_next_cong(xprt);
 462         } else if (result == -ETIMEDOUT) {
 463                 cwnd >>= 1;
 464                 if (cwnd < RPC_CWNDSCALE)
 465                         cwnd = RPC_CWNDSCALE;
 466         }
 467         dprintk("RPC:       cong %ld, cwnd was %ld, now %ld\n",
 468                         xprt->cong, xprt->cwnd, cwnd);
 469         xprt->cwnd = cwnd;
 470         __xprt_put_cong(xprt, req);
 471 }
 472 EXPORT_SYMBOL_GPL(xprt_adjust_cwnd);
 473
 474 /**
 475  * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
 476  * @xprt: transport with waiting tasks
 477  * @status: result code to plant in each task before waking it
 478  *
 479  */
 480 void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
 481 {
 482         if (status < 0)
 483                 rpc_wake_up_status(&xprt->pending, status);
 484         else
 485                 rpc_wake_up(&xprt->pending);
 486 }
 487 EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks);
 488
 489 /**
 490  * xprt_wait_for_buffer_space - wait for transport output buffer to clear
 491  * @task: task to be put to sleep
 492  * @action: function pointer to be executed after wait
 493  *
 494  * Note that we only set the timer for the case of RPC_IS_SOFT(), since
 495  * we don't in general want to force a socket disconnection due to
 496  * an incomplete RPC call transmission.
 497  */
 498 void xprt_wait_for_buffer_space(struct rpc_task *task, rpc_action action)
 499 {
 500         struct rpc_rqst *req = task->tk_rqstp;
 501         struct rpc_xprt *xprt = req->rq_xprt;
 502
 503         task->tk_timeout = RPC_IS_SOFT(task) ? req->rq_timeout : 0;
 504         rpc_sleep_on(&xprt->pending, task, action);
 505 }
 506 EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space);
 507
 508 /**
 509  * xprt_write_space - wake the task waiting for transport output buffer space
 510  * @xprt: transport with waiting tasks
 511  *
 512  * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
 513  */
 514 void xprt_write_space(struct rpc_xprt *xprt)
 515 {
 516         spin_lock_bh(&xprt->transport_lock);
 517         if (xprt->snd_task) {
 518                 dprintk("RPC:       write space: waking waiting task on "
 519                                 "xprt %p\n", xprt);
 520                 rpc_wake_up_queued_task(&xprt->pending, xprt->snd_task);
 521         }
 522         spin_unlock_bh(&xprt->transport_lock);
 523 }
 524 EXPORT_SYMBOL_GPL(xprt_write_space);
 525
 526 /**
 527  * xprt_set_retrans_timeout_def - set a request's retransmit timeout
 528  * @task: task whose timeout is to be set
 529  *
 530  * Set a request's retransmit timeout based on the transport's
 531  * default timeout parameters.  Used by transports that don't adjust
 532  * the retransmit timeout based on round-trip time estimation.
 533  */
 534 void xprt_set_retrans_timeout_def(struct rpc_task *task)
 535 {
 536         task->tk_timeout = task->tk_rqstp->rq_timeout;
 537 }
 538 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_def);
 539
 540 /**
 541  * xprt_set_retrans_timeout_rtt - set a request's retransmit timeout
 542  * @task: task whose timeout is to be set
 543  *
 544  * Set a request's retransmit timeout using the RTT estimator.
 545  */
 546 void xprt_set_retrans_timeout_rtt(struct rpc_task *task)
 547 {
 548         int timer = task->tk_msg.rpc_proc->p_timer;
 549         struct rpc_clnt *clnt = task->tk_client;
 550         struct rpc_rtt *rtt = clnt->cl_rtt;
 551         struct rpc_rqst *req = task->tk_rqstp;
 552         unsigned long max_timeout = clnt->cl_timeout->to_maxval;
 553
 554         task->tk_timeout = rpc_calc_rto(rtt, timer);
 555         task->tk_timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
 556         if (task->tk_timeout > max_timeout || task->tk_timeout == 0)
 557                 task->tk_timeout = max_timeout;
 558 }
 559 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_rtt);
 560
 561 static void xprt_reset_majortimeo(struct rpc_rqst *req)
 562 {
 563         const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
 564
 565         req->rq_majortimeo = req->rq_timeout;
 566         if (to->to_exponential)
 567                 req->rq_majortimeo <<= to->to_retries;
 568         else
 569                 req->rq_majortimeo += to->to_increment * to->to_retries;
 570         if (req->rq_majortimeo > to->to_maxval || req->rq_majortimeo == 0)
 571                 req->rq_majortimeo = to->to_maxval;
 572         req->rq_majortimeo += jiffies;
 573 }
 574
 575 /**
 576  * xprt_adjust_timeout - adjust timeout values for next retransmit
 577  * @req: RPC request containing parameters to use for the adjustment
 578  *
 579  */
 580 int xprt_adjust_timeout(struct rpc_rqst *req)
 581 {
 582         struct rpc_xprt *xprt = req->rq_xprt;
 583         const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
 584         int status = 0;
 585
 586         if (time_before(jiffies, req->rq_majortimeo)) {
 587                 if (to->to_exponential)
 588                         req->rq_timeout <<= 1;
 589                 else
 590                         req->rq_timeout += to->to_increment;
 591                 if (to->to_maxval && req->rq_timeout >= to->to_maxval)
 592                         req->rq_timeout = to->to_maxval;
 593                 req->rq_retries++;
 594         } else {
 595                 req->rq_timeout = to->to_initval;
 596                 req->rq_retries = 0;
 597                 xprt_reset_majortimeo(req);
 598                 /* Reset the RTT counters == "slow start" */
 599                 spin_lock_bh(&xprt->transport_lock);
 600                 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
 601                 spin_unlock_bh(&xprt->transport_lock);
 602                 status = -ETIMEDOUT;
 603         }
 604
 605         if (req->rq_timeout == 0) {
 606                 printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
 607                 req->rq_timeout = 5 * HZ;
 608         }
 609         return status;
 610 }
 611
 612 static void xprt_autoclose(struct work_struct *work)
 613 {
 614         struct rpc_xprt *xprt =
 615                 container_of(work, struct rpc_xprt, task_cleanup);
 616
 617         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 618         xprt->ops->close(xprt);
 619         xprt_release_write(xprt, NULL);
 620         wake_up_bit(&xprt->state, XPRT_LOCKED);
 621 }
 622
 623 /**
 624  * xprt_disconnect_done - mark a transport as disconnected
 625  * @xprt: transport to flag for disconnect
 626  *
 627  */
 628 void xprt_disconnect_done(struct rpc_xprt *xprt)
 629 {
 630         dprintk("RPC:       disconnected transport %p\n", xprt);
 631         spin_lock_bh(&xprt->transport_lock);
 632         xprt_clear_connected(xprt);
 633         xprt_wake_pending_tasks(xprt, -EAGAIN);
 634         spin_unlock_bh(&xprt->transport_lock);
 635 }
 636 EXPORT_SYMBOL_GPL(xprt_disconnect_done);
 637
 638 /**
 639  * xprt_force_disconnect - force a transport to disconnect
 640  * @xprt: transport to disconnect
 641  *
 642  */
 643 void xprt_force_disconnect(struct rpc_xprt *xprt)
 644 {
 645         /* Don't race with the test_bit() in xprt_clear_locked() */
 646         spin_lock_bh(&xprt->transport_lock);
 647         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 648         /* Try to schedule an autoclose RPC call */
 649         if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 650                 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 651         xprt_wake_pending_tasks(xprt, -EAGAIN);
 652         spin_unlock_bh(&xprt->transport_lock);
 653 }
 654
 655 /**
 656  * xprt_conditional_disconnect - force a transport to disconnect
 657  * @xprt: transport to disconnect
 658  * @cookie: 'connection cookie'
 659  *
 660  * This attempts to break the connection if and only if 'cookie' matches
 661  * the current transport 'connection cookie'. It ensures that we don't
 662  * try to break the connection more than once when we need to retransmit
 663  * a batch of RPC requests.
 664  *
 665  */
 666 void xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie)
 667 {
 668         /* Don't race with the test_bit() in xprt_clear_locked() */
 669         spin_lock_bh(&xprt->transport_lock);
 670         if (cookie != xprt->connect_cookie)
 671                 goto out;
 672         if (test_bit(XPRT_CLOSING, &xprt->state))
 673                 goto out;
 674         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 675         /* Try to schedule an autoclose RPC call */
 676         if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 677                 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 678         xprt_wake_pending_tasks(xprt, -EAGAIN);
 679 out:
 680         spin_unlock_bh(&xprt->transport_lock);
 681 }
 682
 683 static bool
 684 xprt_has_timer(const struct rpc_xprt *xprt)
 685 {
 686         return xprt->idle_timeout != 0;
 687 }
 688
 689 static void
 690 xprt_schedule_autodisconnect(struct rpc_xprt *xprt)
 691         __must_hold(&xprt->transport_lock)
 692 {
 693         if (list_empty(&xprt->recv) && xprt_has_timer(xprt))
 694                 mod_timer(&xprt->timer, xprt->last_used + xprt->idle_timeout);
 695 }
 696
 697 static void
 698 xprt_init_autodisconnect(unsigned long data)
 699 {
 700         struct rpc_xprt *xprt = (struct rpc_xprt *)data;
 701
 702         spin_lock(&xprt->transport_lock);
 703         if (!list_empty(&xprt->recv))
 704                 goto out_abort;
 705         /* Reset xprt->last_used to avoid connect/autodisconnect cycling */
 706         xprt->last_used = jiffies;
 707         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 708                 goto out_abort;
 709         spin_unlock(&xprt->transport_lock);
 710         queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 711         return;
 712 out_abort:
 713         spin_unlock(&xprt->transport_lock);
 714 }
 715
 716 bool xprt_lock_connect(struct rpc_xprt *xprt,
 717                 struct rpc_task *task,
 718                 void *cookie)
 719 {
 720         bool ret = false;
 721
 722         spin_lock_bh(&xprt->transport_lock);
 723         if (!test_bit(XPRT_LOCKED, &xprt->state))
 724                 goto out;
 725         if (xprt->snd_task != task)
 726                 goto out;
 727         xprt_task_clear_bytes_sent(task);
 728         xprt->snd_task = cookie;
 729         ret = true;
 730 out:
 731         spin_unlock_bh(&xprt->transport_lock);
 732         return ret;
 733 }
 734
 735 void xprt_unlock_connect(struct rpc_xprt *xprt, void *cookie)
 736 {
 737         spin_lock_bh(&xprt->transport_lock);
 738         if (xprt->snd_task != cookie)
 739                 goto out;
 740         if (!test_bit(XPRT_LOCKED, &xprt->state))
 741                 goto out;
 742         xprt->snd_task =NULL;
 743         xprt->ops->release_xprt(xprt, NULL);
 744         xprt_schedule_autodisconnect(xprt);
 745 out:
 746         spin_unlock_bh(&xprt->transport_lock);
 747         wake_up_bit(&xprt->state, XPRT_LOCKED);
 748 }
 749
 750 /**
 751  * xprt_connect - schedule a transport connect operation
 752  * @task: RPC task that is requesting the connect
 753  *
 754  */
 755 void xprt_connect(struct rpc_task *task)
 756 {
 757         struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
 758
 759         dprintk("RPC: %5u xprt_connect xprt %p %s connected\n", task->tk_pid,
 760                         xprt, (xprt_connected(xprt) ? "is" : "is not"));
 761
 762         if (!xprt_bound(xprt)) {
 763                 task->tk_status = -EAGAIN;
 764                 return;
 765         }
 766         if (!xprt_lock_write(xprt, task))
 767                 return;
 768
 769         if (test_and_clear_bit(XPRT_CLOSE_WAIT, &xprt->state))
 770                 xprt->ops->close(xprt);
 771
 772         if (!xprt_connected(xprt)) {
 773                 task->tk_rqstp->rq_bytes_sent = 0;
 774                 task->tk_timeout = task->tk_rqstp->rq_timeout;
 775                 task->tk_rqstp->rq_connect_cookie = xprt->connect_cookie;
 776                 rpc_sleep_on(&xprt->pending, task, xprt_connect_status);
 777
 778                 if (test_bit(XPRT_CLOSING, &xprt->state))
 779                         return;
 780                 if (xprt_test_and_set_connecting(xprt))
 781                         return;
 782                 xprt->stat.connect_start = jiffies;
 783                 xprt->ops->connect(xprt, task);
 784         }
 785         xprt_release_write(xprt, task);
 786 }
 787
 788 static void xprt_connect_status(struct rpc_task *task)
 789 {
 790         struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
 791
 792         if (task->tk_status == 0) {
 793                 xprt->stat.connect_count++;
 794                 xprt->stat.connect_time += (long)jiffies - xprt->stat.connect_start;
 795                 dprintk("RPC: %5u xprt_connect_status: connection established\n",
 796                                 task->tk_pid);
 797                 return;
 798         }
 799
 800         switch (task->tk_status) {
 801         case -ECONNREFUSED:
 802         case -ECONNRESET:
 803         case -ECONNABORTED:
 804         case -ENETUNREACH:
 805         case -EHOSTUNREACH:
 806         case -EPIPE:
 807         case -EAGAIN:
 808                 dprintk("RPC: %5u xprt_connect_status: retrying\n", task->tk_pid);
 809                 break;
 810         case -ETIMEDOUT:
 811                 dprintk("RPC: %5u xprt_connect_status: connect attempt timed "
 812                                 "out\n", task->tk_pid);
 813                 break;
 814         default:
 815                 dprintk("RPC: %5u xprt_connect_status: error %d connecting to "
 816                                 "server %s\n", task->tk_pid, -task->tk_status,
 817                                 xprt->servername);
 818                 task->tk_status = -EIO;
 819         }
 820 }
 821
 822 /**
 823  * xprt_lookup_rqst - find an RPC request corresponding to an XID
 824  * @xprt: transport on which the original request was transmitted
 825  * @xid: RPC XID of incoming reply
 826  *
 827  */
 828 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
 829 {
 830         struct rpc_rqst *entry;
 831
 832         list_for_each_entry(entry, &xprt->recv, rq_list)
 833                 if (entry->rq_xid == xid) {
 834                         trace_xprt_lookup_rqst(xprt, xid, 0);
 835                         return entry;
 836                 }
 837
 838         dprintk("RPC:       xprt_lookup_rqst did not find xid %08x\n",
 839                         ntohl(xid));
 840         trace_xprt_lookup_rqst(xprt, xid, -ENOENT);
 841         xprt->stat.bad_xids++;
 842         return NULL;
 843 }
 844 EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
 845
 846 static void xprt_update_rtt(struct rpc_task *task)
 847 {
 848         struct rpc_rqst *req = task->tk_rqstp;
 849         struct rpc_rtt *rtt = task->tk_client->cl_rtt;
 850         unsigned int timer = task->tk_msg.rpc_proc->p_timer;
 851         long m = usecs_to_jiffies(ktime_to_us(req->rq_rtt));
 852
 853         if (timer) {
 854                 if (req->rq_ntrans == 1)
 855                         rpc_update_rtt(rtt, timer, m);
 856                 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
 857         }
 858 }
 859
 860 /**
 861  * xprt_complete_rqst - called when reply processing is complete
 862  * @task: RPC request that recently completed
 863  * @copied: actual number of bytes received from the transport
 864  *
 865  * Caller holds transport lock.
 866  */
 867 void xprt_complete_rqst(struct rpc_task *task, int copied)
 868 {
 869         struct rpc_rqst *req = task->tk_rqstp;
 870         struct rpc_xprt *xprt = req->rq_xprt;
 871
 872         dprintk("RPC: %5u xid %08x complete (%d bytes received)\n",
 873                         task->tk_pid, ntohl(req->rq_xid), copied);
 874         trace_xprt_complete_rqst(xprt, req->rq_xid, copied);
 875
 876         xprt->stat.recvs++;
 877         req->rq_rtt = ktime_sub(ktime_get(), req->rq_xtime);
 878         if (xprt->ops->timer != NULL)
 879                 xprt_update_rtt(task);
 880
 881         list_del_init(&req->rq_list);
 882         req->rq_private_buf.len = copied;
 883         /* Ensure all writes are done before we update */
 884         /* req->rq_reply_bytes_recvd */
 885         smp_wmb();
 886         req->rq_reply_bytes_recvd = copied;
 887         rpc_wake_up_queued_task(&xprt->pending, task);
 888 }
 889 EXPORT_SYMBOL_GPL(xprt_complete_rqst);
 890
 891 static void xprt_timer(struct rpc_task *task)
 892 {
 893         struct rpc_rqst *req = task->tk_rqstp;
 894         struct rpc_xprt *xprt = req->rq_xprt;
 895
 896         if (task->tk_status != -ETIMEDOUT)
 897                 return;
 898         dprintk("RPC: %5u xprt_timer\n", task->tk_pid);
 899
 900         if (!req->rq_reply_bytes_recvd) {
 901                 if (xprt->ops->timer)
 902                         xprt->ops->timer(xprt, task);
 903         } else
 904                 task->tk_status = 0;
 905 }
 906
 907 /**
 908  * xprt_prepare_transmit - reserve the transport before sending a request
 909  * @task: RPC task about to send a request
 910  *
 911  */
 912 bool xprt_prepare_transmit(struct rpc_task *task)
 913 {
 914         struct rpc_rqst *req = task->tk_rqstp;
 915         struct rpc_xprt *xprt = req->rq_xprt;
 916         bool ret = false;
 917
 918         dprintk("RPC: %5u xprt_prepare_transmit\n", task->tk_pid);
 919
 920         spin_lock_bh(&xprt->transport_lock);
 921         if (!req->rq_bytes_sent) {
 922                 if (req->rq_reply_bytes_recvd) {
 923                         task->tk_status = req->rq_reply_bytes_recvd;
 924                         goto out_unlock;
 925                 }
 926                 if ((task->tk_flags & RPC_TASK_NO_RETRANS_TIMEOUT)
 927                     && xprt_connected(xprt)
 928                     && req->rq_connect_cookie == xprt->connect_cookie) {
 929                         xprt->ops->set_retrans_timeout(task);
 930                         rpc_sleep_on(&xprt->pending, task, xprt_timer);
 931                         goto out_unlock;
 932                 }
 933         }
 934         if (!xprt->ops->reserve_xprt(xprt, task)) {
 935                 task->tk_status = -EAGAIN;
 936                 goto out_unlock;
 937         }
 938         ret = true;
 939 out_unlock:
 940         spin_unlock_bh(&xprt->transport_lock);
 941         return ret;
 942 }
 943
 944 void xprt_end_transmit(struct rpc_task *task)
 945 {
 946         xprt_release_write(task->tk_rqstp->rq_xprt, task);
 947 }
 948
 949 /**
 950  * xprt_transmit - send an RPC request on a transport
 951  * @task: controlling RPC task
 952  *
 953  * We have to copy the iovec because sendmsg fiddles with its contents.
 954  */
 955 void xprt_transmit(struct rpc_task *task)
 956 {
 957         struct rpc_rqst *req = task->tk_rqstp;
 958         struct rpc_xprt *xprt = req->rq_xprt;
 959         int status, numreqs;
 960
 961         dprintk("RPC: %5u xprt_transmit(%u)\n", task->tk_pid, req->rq_slen);
 962
 963         if (!req->rq_reply_bytes_recvd) {
 964                 if (list_empty(&req->rq_list) && rpc_reply_expected(task)) {
 965                         /*
 966                          * Add to the list only if we're expecting a reply
 967                          */
 968                         spin_lock_bh(&xprt->transport_lock);
 969                         /* Update the softirq receive buffer */
 970                         memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
 971                                         sizeof(req->rq_private_buf));
 972                         /* Add request to the receive list */
 973                         list_add_tail(&req->rq_list, &xprt->recv);
 974                         spin_unlock_bh(&xprt->transport_lock);
 975                         xprt_reset_majortimeo(req);
 976                         /* Turn off autodisconnect */
 977                         del_singleshot_timer_sync(&xprt->timer);
 978                 }
 979         } else if (!req->rq_bytes_sent)
 980                 return;
 981
 982         req->rq_xtime = ktime_get();
 983         status = xprt->ops->send_request(task);
 984         trace_xprt_transmit(xprt, req->rq_xid, status);
 985         if (status != 0) {
 986                 task->tk_status = status;
 987                 return;
 988         }
 989         xprt_inject_disconnect(xprt);
 990
 991         dprintk("RPC: %5u xmit complete\n", task->tk_pid);
 992         task->tk_flags |= RPC_TASK_SENT;
 993         spin_lock_bh(&xprt->transport_lock);
 994
 995         xprt->ops->set_retrans_timeout(task);
 996
 997         numreqs = atomic_read(&xprt->num_reqs);
 998         if (numreqs > xprt->stat.max_slots)
 999                 xprt->stat.max_slots = numreqs;
1000         xprt->stat.sends++;
1001         xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
1002         xprt->stat.bklog_u += xprt->backlog.qlen;
1003         xprt->stat.sending_u += xprt->sending.qlen;
1004         xprt->stat.pending_u += xprt->pending.qlen;
1005
1006         /* Don't race with disconnect */
1007         if (!xprt_connected(xprt))
1008                 task->tk_status = -ENOTCONN;
1009         else {
1010                 /*
1011                  * Sleep on the pending queue since
1012                  * we're expecting a reply.
1013                  */
1014                 if (!req->rq_reply_bytes_recvd && rpc_reply_expected(task))
1015                         rpc_sleep_on(&xprt->pending, task, xprt_timer);
1016                 req->rq_connect_cookie = xprt->connect_cookie;
1017         }
1018         spin_unlock_bh(&xprt->transport_lock);
1019 }
1020
1021 static void xprt_add_backlog(struct rpc_xprt *xprt, struct rpc_task *task)
1022 {
1023         set_bit(XPRT_CONGESTED, &xprt->state);
1024         rpc_sleep_on(&xprt->backlog, task, NULL);
1025 }
1026
1027 static void xprt_wake_up_backlog(struct rpc_xprt *xprt)
1028 {
1029         if (rpc_wake_up_next(&xprt->backlog) == NULL)
1030                 clear_bit(XPRT_CONGESTED, &xprt->state);
1031 }
1032
1033 static bool xprt_throttle_congested(struct rpc_xprt *xprt, struct rpc_task *task)
1034 {
1035         bool ret = false;
1036
1037         if (!test_bit(XPRT_CONGESTED, &xprt->state))
1038                 goto out;
1039         spin_lock(&xprt->reserve_lock);
1040         if (test_bit(XPRT_CONGESTED, &xprt->state)) {
1041                 rpc_sleep_on(&xprt->backlog, task, NULL);
1042                 ret = true;
1043         }
1044         spin_unlock(&xprt->reserve_lock);
1045 out:
1046         return ret;
1047 }
1048
1049 static struct rpc_rqst *xprt_dynamic_alloc_slot(struct rpc_xprt *xprt, gfp_t gfp_flags)
1050 {
1051         struct rpc_rqst *req = ERR_PTR(-EAGAIN);
1052
1053         if (!atomic_add_unless(&xprt->num_reqs, 1, xprt->max_reqs))
1054                 goto out;
1055         req = kzalloc(sizeof(struct rpc_rqst), gfp_flags);
1056         if (req != NULL)
1057                 goto out;
1058         atomic_dec(&xprt->num_reqs);
1059         req = ERR_PTR(-ENOMEM);
1060 out:
1061         return req;
1062 }
1063
1064 static bool xprt_dynamic_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1065 {
1066         if (atomic_add_unless(&xprt->num_reqs, -1, xprt->min_reqs)) {
1067                 kfree(req);
1068                 return true;
1069         }
1070         return false;
1071 }
1072
1073 void xprt_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1074 {
1075         struct rpc_rqst *req;
1076
1077         spin_lock(&xprt->reserve_lock);
1078         if (!list_empty(&xprt->free)) {
1079                 req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
1080                 list_del(&req->rq_list);
1081                 goto out_init_req;
1082         }
1083         req = xprt_dynamic_alloc_slot(xprt, GFP_NOWAIT|__GFP_NOWARN);
1084         if (!IS_ERR(req))
1085                 goto out_init_req;
1086         switch (PTR_ERR(req)) {
1087         case -ENOMEM:
1088                 dprintk("RPC:       dynamic allocation of request slot "
1089                                 "failed! Retrying\n");
1090                 task->tk_status = -ENOMEM;
1091                 break;
1092         case -EAGAIN:
1093                 xprt_add_backlog(xprt, task);
1094                 dprintk("RPC:       waiting for request slot\n");
1095         default:
1096                 task->tk_status = -EAGAIN;
1097         }
1098         spin_unlock(&xprt->reserve_lock);
1099         return;
1100 out_init_req:
1101         task->tk_status = 0;
1102         task->tk_rqstp = req;
1103         xprt_request_init(task, xprt);
1104         spin_unlock(&xprt->reserve_lock);
1105 }
1106 EXPORT_SYMBOL_GPL(xprt_alloc_slot);
1107
1108 void xprt_lock_and_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1109 {
1110         /* Note: grabbing the xprt_lock_write() ensures that we throttle
1111          * new slot allocation if the transport is congested (i.e. when
1112          * reconnecting a stream transport or when out of socket write
1113          * buffer space).
1114          */
1115         if (xprt_lock_write(xprt, task)) {
1116                 xprt_alloc_slot(xprt, task);
1117                 xprt_release_write(xprt, task);
1118         }
1119 }
1120 EXPORT_SYMBOL_GPL(xprt_lock_and_alloc_slot);
1121
1122 static void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1123 {
1124         spin_lock(&xprt->reserve_lock);
1125         if (!xprt_dynamic_free_slot(xprt, req)) {
1126                 memset(req, 0, sizeof(*req));   /* mark unused */
1127                 list_add(&req->rq_list, &xprt->free);
1128         }
1129         xprt_wake_up_backlog(xprt);
1130         spin_unlock(&xprt->reserve_lock);
1131 }
1132
1133 static void xprt_free_all_slots(struct rpc_xprt *xprt)
1134 {
1135         struct rpc_rqst *req;
1136         while (!list_empty(&xprt->free)) {
1137                 req = list_first_entry(&xprt->free, struct rpc_rqst, rq_list);
1138                 list_del(&req->rq_list);
1139                 kfree(req);
1140         }
1141 }
1142
1143 struct rpc_xprt *xprt_alloc(struct net *net, size_t size,
1144                 unsigned int num_prealloc,
1145                 unsigned int max_alloc)
1146 {
1147         struct rpc_xprt *xprt;
1148         struct rpc_rqst *req;
1149         int i;
1150
1151         xprt = kzalloc(size, GFP_KERNEL);
1152         if (xprt == NULL)
1153                 goto out;
1154
1155         xprt_init(xprt, net);
1156
1157         for (i = 0; i < num_prealloc; i++) {
1158                 req = kzalloc(sizeof(struct rpc_rqst), GFP_KERNEL);
1159                 if (!req)
1160                         goto out_free;
1161                 list_add(&req->rq_list, &xprt->free);
1162         }
1163         if (max_alloc > num_prealloc)
1164                 xprt->max_reqs = max_alloc;
1165         else
1166                 xprt->max_reqs = num_prealloc;
1167         xprt->min_reqs = num_prealloc;
1168         atomic_set(&xprt->num_reqs, num_prealloc);
1169
1170         return xprt;
1171
1172 out_free:
1173         xprt_free(xprt);
1174 out:
1175         return NULL;
1176 }
1177 EXPORT_SYMBOL_GPL(xprt_alloc);
1178
1179 void xprt_free(struct rpc_xprt *xprt)
1180 {
1181         put_net(xprt->xprt_net);
1182         xprt_free_all_slots(xprt);
1183         kfree_rcu(xprt, rcu);
1184 }
1185 EXPORT_SYMBOL_GPL(xprt_free);
1186
1187 /**
1188  * xprt_reserve - allocate an RPC request slot
1189  * @task: RPC task requesting a slot allocation
1190  *
1191  * If the transport is marked as being congested, or if no more
1192  * slots are available, place the task on the transport's
1193  * backlog queue.
1194  */
1195 void xprt_reserve(struct rpc_task *task)
1196 {
1197         struct rpc_xprt *xprt = task->tk_xprt;
1198
1199         task->tk_status = 0;
1200         if (task->tk_rqstp != NULL)
1201                 return;
1202
1203         task->tk_timeout = 0;
1204         task->tk_status = -EAGAIN;
1205         if (!xprt_throttle_congested(xprt, task))
1206                 xprt->ops->alloc_slot(xprt, task);
1207 }
1208
1209 /**
1210  * xprt_retry_reserve - allocate an RPC request slot
1211  * @task: RPC task requesting a slot allocation
1212  *
1213  * If no more slots are available, place the task on the transport's
1214  * backlog queue.
1215  * Note that the only difference with xprt_reserve is that we now
1216  * ignore the value of the XPRT_CONGESTED flag.
1217  */
1218 void xprt_retry_reserve(struct rpc_task *task)
1219 {
1220         struct rpc_xprt *xprt = task->tk_xprt;
1221
1222         task->tk_status = 0;
1223         if (task->tk_rqstp != NULL)
1224                 return;
1225
1226         task->tk_timeout = 0;
1227         task->tk_status = -EAGAIN;
1228         xprt->ops->alloc_slot(xprt, task);
1229 }
1230
1231 static inline __be32 xprt_alloc_xid(struct rpc_xprt *xprt)
1232 {
1233         return (__force __be32)xprt->xid++;
1234 }
1235
1236 static inline void xprt_init_xid(struct rpc_xprt *xprt)
1237 {
1238         xprt->xid = prandom_u32();
1239 }
1240
1241 static void xprt_request_init(struct rpc_task *task, struct rpc_xprt *xprt)
1242 {
1243         struct rpc_rqst *req = task->tk_rqstp;
1244
1245         INIT_LIST_HEAD(&req->rq_list);
1246         req->rq_timeout = task->tk_client->cl_timeout->to_initval;
1247         req->rq_task    = task;
1248         req->rq_xprt    = xprt;
1249         req->rq_buffer  = NULL;
1250         req->rq_xid     = xprt_alloc_xid(xprt);
1251         req->rq_connect_cookie = xprt->connect_cookie - 1;
1252         req->rq_bytes_sent = 0;
1253         req->rq_snd_buf.len = 0;
1254         req->rq_snd_buf.buflen = 0;
1255         req->rq_rcv_buf.len = 0;
1256         req->rq_rcv_buf.buflen = 0;
1257         req->rq_release_snd_buf = NULL;
1258         xprt_reset_majortimeo(req);
1259         dprintk("RPC: %5u reserved req %p xid %08x\n", task->tk_pid,
1260                         req, ntohl(req->rq_xid));
1261 }
1262
1263 /**
1264  * xprt_release - release an RPC request slot
1265  * @task: task which is finished with the slot
1266  *
1267  */
1268 void xprt_release(struct rpc_task *task)
1269 {
1270         struct rpc_xprt *xprt;
1271         struct rpc_rqst *req = task->tk_rqstp;
1272
1273         if (req == NULL) {
1274                 if (task->tk_client) {
1275                         xprt = task->tk_xprt;
1276                         if (xprt->snd_task == task)
1277                                 xprt_release_write(xprt, task);
1278                 }
1279                 return;
1280         }
1281
1282         xprt = req->rq_xprt;
1283         if (task->tk_ops->rpc_count_stats != NULL)
1284                 task->tk_ops->rpc_count_stats(task, task->tk_calldata);
1285         else if (task->tk_client)
1286                 rpc_count_iostats(task, task->tk_client->cl_metrics);
1287         spin_lock_bh(&xprt->transport_lock);
1288         xprt->ops->release_xprt(xprt, task);
1289         if (xprt->ops->release_request)
1290                 xprt->ops->release_request(task);
1291         if (!list_empty(&req->rq_list))
1292                 list_del(&req->rq_list);
1293         xprt->last_used = jiffies;
1294         xprt_schedule_autodisconnect(xprt);
1295         spin_unlock_bh(&xprt->transport_lock);
1296         if (req->rq_buffer)
1297                 xprt->ops->buf_free(task);
1298         xprt_inject_disconnect(xprt);
1299         if (req->rq_cred != NULL)
1300                 put_rpccred(req->rq_cred);
1301         task->tk_rqstp = NULL;
1302         if (req->rq_release_snd_buf)
1303                 req->rq_release_snd_buf(req);
1304
1305         dprintk("RPC: %5u release request %p\n", task->tk_pid, req);
1306         if (likely(!bc_prealloc(req)))
1307                 xprt_free_slot(xprt, req);
1308         else
1309                 xprt_free_bc_request(req);
1310 }
1311
1312 static void xprt_init(struct rpc_xprt *xprt, struct net *net)
1313 {
1314         kref_init(&xprt->kref);
1315
1316         spin_lock_init(&xprt->transport_lock);
1317         spin_lock_init(&xprt->reserve_lock);
1318
1319         INIT_LIST_HEAD(&xprt->free);
1320         INIT_LIST_HEAD(&xprt->recv);
1321 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1322         spin_lock_init(&xprt->bc_pa_lock);
1323         INIT_LIST_HEAD(&xprt->bc_pa_list);
1324 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1325         INIT_LIST_HEAD(&xprt->xprt_switch);
1326
1327         xprt->last_used = jiffies;
1328         xprt->cwnd = RPC_INITCWND;
1329         xprt->bind_index = 0;
1330
1331         rpc_init_wait_queue(&xprt->binding, "xprt_binding");
1332         rpc_init_wait_queue(&xprt->pending, "xprt_pending");
1333         rpc_init_priority_wait_queue(&xprt->sending, "xprt_sending");
1334         rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
1335
1336         xprt_init_xid(xprt);
1337
1338         xprt->xprt_net = get_net(net);
1339 }
1340
1341 /**
1342  * xprt_create_transport - create an RPC transport
1343  * @args: rpc transport creation arguments
1344  *
1345  */
1346 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
1347 {
1348         struct rpc_xprt *xprt;
1349         struct xprt_class *t;
1350
1351         spin_lock(&xprt_list_lock);
1352         list_for_each_entry(t, &xprt_list, list) {
1353                 if (t->ident == args->ident) {
1354                         spin_unlock(&xprt_list_lock);
1355                         goto found;
1356                 }
1357         }
1358         spin_unlock(&xprt_list_lock);
1359         dprintk("RPC: transport (%d) not supported\n", args->ident);
1360         return ERR_PTR(-EIO);
1361
1362 found:
1363         xprt = t->setup(args);
1364         if (IS_ERR(xprt)) {
1365                 dprintk("RPC:       xprt_create_transport: failed, %ld\n",
1366                                 -PTR_ERR(xprt));
1367                 goto out;
1368         }
1369         if (args->flags & XPRT_CREATE_NO_IDLE_TIMEOUT)
1370                 xprt->idle_timeout = 0;
1371         INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
1372         if (xprt_has_timer(xprt))
1373                 setup_timer(&xprt->timer, xprt_init_autodisconnect,
1374                             (unsigned long)xprt);
1375         else
1376                 init_timer(&xprt->timer);
1377
1378         if (strlen(args->servername) > RPC_MAXNETNAMELEN) {
1379                 xprt_destroy(xprt);
1380                 return ERR_PTR(-EINVAL);
1381         }
1382         xprt->servername = kstrdup(args->servername, GFP_KERNEL);
1383         if (xprt->servername == NULL) {
1384                 xprt_destroy(xprt);
1385                 return ERR_PTR(-ENOMEM);
1386         }
1387
1388         rpc_xprt_debugfs_register(xprt);
1389
1390         dprintk("RPC:       created transport %p with %u slots\n", xprt,
1391                         xprt->max_reqs);
1392 out:
1393         return xprt;
1394 }
1395
1396 /**
1397  * xprt_destroy - destroy an RPC transport, killing off all requests.
1398  * @xprt: transport to destroy
1399  *
1400  */
1401 static void xprt_destroy(struct rpc_xprt *xprt)
1402 {
1403         dprintk("RPC:       destroying transport %p\n", xprt);
1404
1405         /* Exclude transport connect/disconnect handlers */
1406         wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_UNINTERRUPTIBLE);
1407
1408         del_timer_sync(&xprt->timer);
1409
1410         rpc_xprt_debugfs_unregister(xprt);
1411         rpc_destroy_wait_queue(&xprt->binding);
1412         rpc_destroy_wait_queue(&xprt->pending);
1413         rpc_destroy_wait_queue(&xprt->sending);
1414         rpc_destroy_wait_queue(&xprt->backlog);
1415         cancel_work_sync(&xprt->task_cleanup);
1416         kfree(xprt->servername);
1417         /*
1418          * Tear down transport state and free the rpc_xprt
1419          */
1420         xprt->ops->destroy(xprt);
1421 }
1422
1423 static void xprt_destroy_kref(struct kref *kref)
1424 {
1425         xprt_destroy(container_of(kref, struct rpc_xprt, kref));
1426 }
1427
1428 /**
1429  * xprt_get - return a reference to an RPC transport.
1430  * @xprt: pointer to the transport
1431  *
1432  */
1433 struct rpc_xprt *xprt_get(struct rpc_xprt *xprt)
1434 {
1435         if (xprt != NULL && kref_get_unless_zero(&xprt->kref))
1436                 return xprt;
1437         return NULL;
1438 }
1439 EXPORT_SYMBOL_GPL(xprt_get);
1440
1441 /**
1442  * xprt_put - release a reference to an RPC transport.
1443  * @xprt: pointer to the transport
1444  *
1445  */
1446 void xprt_put(struct rpc_xprt *xprt)
1447 {
1448         if (xprt != NULL)
1449                 kref_put(&xprt->kref, xprt_destroy_kref);
1450 }
1451 EXPORT_SYMBOL_GPL(xprt_put);