net/sunrpc/xprt.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  linux/net/sunrpc/xprt.c
   4  *
   5  *  This is a generic RPC call interface supporting congestion avoidance,
   6  *  and asynchronous calls.
   7  *
   8  *  The interface works like this:
   9  *
  10  *  -   When a process places a call, it allocates a request slot if
  11  *      one is available. Otherwise, it sleeps on the backlog queue
  12  *      (xprt_reserve).
  13  *  -   Next, the caller puts together the RPC message, stuffs it into
  14  *      the request struct, and calls xprt_transmit().
  15  *  -   xprt_transmit sends the message and installs the caller on the
  16  *      transport's wait list. At the same time, if a reply is expected,
  17  *      it installs a timer that is run after the packet's timeout has
  18  *      expired.
  19  *  -   When a packet arrives, the data_ready handler walks the list of
  20  *      pending requests for that transport. If a matching XID is found, the
  21  *      caller is woken up, and the timer removed.
  22  *  -   When no reply arrives within the timeout interval, the timer is
  23  *      fired by the kernel and runs xprt_timer(). It either adjusts the
  24  *      timeout values (minor timeout) or wakes up the caller with a status
  25  *      of -ETIMEDOUT.
  26  *  -   When the caller receives a notification from RPC that a reply arrived,
  27  *      it should release the RPC slot, and process the reply.
  28  *      If the call timed out, it may choose to retry the operation by
  29  *      adjusting the initial timeout value, and simply calling rpc_call
  30  *      again.
  31  *
  32  *  Support for async RPC is done through a set of RPC-specific scheduling
  33  *  primitives that `transparently' work for processes as well as async
  34  *  tasks that rely on callbacks.
  35  *
  36  *  Copyright (C) 1995-1997, Olaf Kirch <okir@monad.swb.de>
  37  *
  38  *  Transport switch API copyright (C) 2005, Chuck Lever <cel@netapp.com>
  39  */
  40
  41 #include <linux/module.h>
  42
  43 #include <linux/types.h>
  44 #include <linux/interrupt.h>
  45 #include <linux/workqueue.h>
  46 #include <linux/net.h>
  47 #include <linux/ktime.h>
  48
  49 #include <linux/sunrpc/clnt.h>
  50 #include <linux/sunrpc/metrics.h>
  51 #include <linux/sunrpc/bc_xprt.h>
  52 #include <linux/rcupdate.h>
  53 #include <linux/sched/mm.h>
  54
  55 #include <trace/events/sunrpc.h>
  56
  57 #include "sunrpc.h"
  58
  59 /*
  60  * Local variables
  61  */
  62
  63 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
  64 # define RPCDBG_FACILITY        RPCDBG_XPRT
  65 #endif
  66
  67 /*
  68  * Local functions
  69  */
  70 static void      xprt_init(struct rpc_xprt *xprt, struct net *net);
  71 static __be32   xprt_alloc_xid(struct rpc_xprt *xprt);
  72 static void      xprt_destroy(struct rpc_xprt *xprt);
  73
  74 static DEFINE_SPINLOCK(xprt_list_lock);
  75 static LIST_HEAD(xprt_list);
  76
  77 static unsigned long xprt_request_timeout(const struct rpc_rqst *req)
  78 {
  79         unsigned long timeout = jiffies + req->rq_timeout;
  80
  81         if (time_before(timeout, req->rq_majortimeo))
  82                 return timeout;
  83         return req->rq_majortimeo;
  84 }
  85
  86 /**
  87  * xprt_register_transport - register a transport implementation
  88  * @transport: transport to register
  89  *
  90  * If a transport implementation is loaded as a kernel module, it can
  91  * call this interface to make itself known to the RPC client.
  92  *
  93  * Returns:
  94  * 0:           transport successfully registered
  95  * -EEXIST:     transport already registered
  96  * -EINVAL:     transport module being unloaded
  97  */
  98 int xprt_register_transport(struct xprt_class *transport)
  99 {
 100         struct xprt_class *t;
 101         int result;
 102
 103         result = -EEXIST;
 104         spin_lock(&xprt_list_lock);
 105         list_for_each_entry(t, &xprt_list, list) {
 106                 /* don't register the same transport class twice */
 107                 if (t->ident == transport->ident)
 108                         goto out;
 109         }
 110
 111         list_add_tail(&transport->list, &xprt_list);
 112         printk(KERN_INFO "RPC: Registered %s transport module.\n",
 113                transport->name);
 114         result = 0;
 115
 116 out:
 117         spin_unlock(&xprt_list_lock);
 118         return result;
 119 }
 120 EXPORT_SYMBOL_GPL(xprt_register_transport);
 121
 122 /**
 123  * xprt_unregister_transport - unregister a transport implementation
 124  * @transport: transport to unregister
 125  *
 126  * Returns:
 127  * 0:           transport successfully unregistered
 128  * -ENOENT:     transport never registered
 129  */
 130 int xprt_unregister_transport(struct xprt_class *transport)
 131 {
 132         struct xprt_class *t;
 133         int result;
 134
 135         result = 0;
 136         spin_lock(&xprt_list_lock);
 137         list_for_each_entry(t, &xprt_list, list) {
 138                 if (t == transport) {
 139                         printk(KERN_INFO
 140                                 "RPC: Unregistered %s transport module.\n",
 141                                 transport->name);
 142                         list_del_init(&transport->list);
 143                         goto out;
 144                 }
 145         }
 146         result = -ENOENT;
 147
 148 out:
 149         spin_unlock(&xprt_list_lock);
 150         return result;
 151 }
 152 EXPORT_SYMBOL_GPL(xprt_unregister_transport);
 153
 154 /**
 155  * xprt_load_transport - load a transport implementation
 156  * @transport_name: transport to load
 157  *
 158  * Returns:
 159  * 0:           transport successfully loaded
 160  * -ENOENT:     transport module not available
 161  */
 162 int xprt_load_transport(const char *transport_name)
 163 {
 164         struct xprt_class *t;
 165         int result;
 166
 167         result = 0;
 168         spin_lock(&xprt_list_lock);
 169         list_for_each_entry(t, &xprt_list, list) {
 170                 if (strcmp(t->name, transport_name) == 0) {
 171                         spin_unlock(&xprt_list_lock);
 172                         goto out;
 173                 }
 174         }
 175         spin_unlock(&xprt_list_lock);
 176         result = request_module("xprt%s", transport_name);
 177 out:
 178         return result;
 179 }
 180 EXPORT_SYMBOL_GPL(xprt_load_transport);
 181
 182 static void xprt_clear_locked(struct rpc_xprt *xprt)
 183 {
 184         xprt->snd_task = NULL;
 185         if (!test_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
 186                 smp_mb__before_atomic();
 187                 clear_bit(XPRT_LOCKED, &xprt->state);
 188                 smp_mb__after_atomic();
 189         } else
 190                 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 191 }
 192
 193 /**
 194  * xprt_reserve_xprt - serialize write access to transports
 195  * @task: task that is requesting access to the transport
 196  * @xprt: pointer to the target transport
 197  *
 198  * This prevents mixing the payload of separate requests, and prevents
 199  * transport connects from colliding with writes.  No congestion control
 200  * is provided.
 201  */
 202 int xprt_reserve_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 203 {
 204         struct rpc_rqst *req = task->tk_rqstp;
 205
 206         if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
 207                 if (task == xprt->snd_task)
 208                         goto out_locked;
 209                 goto out_sleep;
 210         }
 211         if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
 212                 goto out_unlock;
 213         xprt->snd_task = task;
 214
 215 out_locked:
 216         trace_xprt_reserve_xprt(xprt, task);
 217         return 1;
 218
 219 out_unlock:
 220         xprt_clear_locked(xprt);
 221 out_sleep:
 222         task->tk_status = -EAGAIN;
 223         if  (RPC_IS_SOFT(task))
 224                 rpc_sleep_on_timeout(&xprt->sending, task, NULL,
 225                                 xprt_request_timeout(req));
 226         else
 227                 rpc_sleep_on(&xprt->sending, task, NULL);
 228         return 0;
 229 }
 230 EXPORT_SYMBOL_GPL(xprt_reserve_xprt);
 231
 232 static bool
 233 xprt_need_congestion_window_wait(struct rpc_xprt *xprt)
 234 {
 235         return test_bit(XPRT_CWND_WAIT, &xprt->state);
 236 }
 237
 238 static void
 239 xprt_set_congestion_window_wait(struct rpc_xprt *xprt)
 240 {
 241         if (!list_empty(&xprt->xmit_queue)) {
 242                 /* Peek at head of queue to see if it can make progress */
 243                 if (list_first_entry(&xprt->xmit_queue, struct rpc_rqst,
 244                                         rq_xmit)->rq_cong)
 245                         return;
 246         }
 247         set_bit(XPRT_CWND_WAIT, &xprt->state);
 248 }
 249
 250 static void
 251 xprt_test_and_clear_congestion_window_wait(struct rpc_xprt *xprt)
 252 {
 253         if (!RPCXPRT_CONGESTED(xprt))
 254                 clear_bit(XPRT_CWND_WAIT, &xprt->state);
 255 }
 256
 257 /*
 258  * xprt_reserve_xprt_cong - serialize write access to transports
 259  * @task: task that is requesting access to the transport
 260  *
 261  * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
 262  * integrated into the decision of whether a request is allowed to be
 263  * woken up and given access to the transport.
 264  * Note that the lock is only granted if we know there are free slots.
 265  */
 266 int xprt_reserve_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 267 {
 268         struct rpc_rqst *req = task->tk_rqstp;
 269
 270         if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
 271                 if (task == xprt->snd_task)
 272                         goto out_locked;
 273                 goto out_sleep;
 274         }
 275         if (req == NULL) {
 276                 xprt->snd_task = task;
 277                 goto out_locked;
 278         }
 279         if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
 280                 goto out_unlock;
 281         if (!xprt_need_congestion_window_wait(xprt)) {
 282                 xprt->snd_task = task;
 283                 goto out_locked;
 284         }
 285 out_unlock:
 286         xprt_clear_locked(xprt);
 287 out_sleep:
 288         task->tk_status = -EAGAIN;
 289         if (RPC_IS_SOFT(task))
 290                 rpc_sleep_on_timeout(&xprt->sending, task, NULL,
 291                                 xprt_request_timeout(req));
 292         else
 293                 rpc_sleep_on(&xprt->sending, task, NULL);
 294         return 0;
 295 out_locked:
 296         trace_xprt_reserve_cong(xprt, task);
 297         return 1;
 298 }
 299 EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong);
 300
 301 static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
 302 {
 303         int retval;
 304
 305         if (test_bit(XPRT_LOCKED, &xprt->state) && xprt->snd_task == task)
 306                 return 1;
 307         spin_lock(&xprt->transport_lock);
 308         retval = xprt->ops->reserve_xprt(xprt, task);
 309         spin_unlock(&xprt->transport_lock);
 310         return retval;
 311 }
 312
 313 static bool __xprt_lock_write_func(struct rpc_task *task, void *data)
 314 {
 315         struct rpc_xprt *xprt = data;
 316
 317         xprt->snd_task = task;
 318         return true;
 319 }
 320
 321 static void __xprt_lock_write_next(struct rpc_xprt *xprt)
 322 {
 323         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 324                 return;
 325         if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
 326                 goto out_unlock;
 327         if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
 328                                 __xprt_lock_write_func, xprt))
 329                 return;
 330 out_unlock:
 331         xprt_clear_locked(xprt);
 332 }
 333
 334 static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt)
 335 {
 336         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 337                 return;
 338         if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
 339                 goto out_unlock;
 340         if (xprt_need_congestion_window_wait(xprt))
 341                 goto out_unlock;
 342         if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
 343                                 __xprt_lock_write_func, xprt))
 344                 return;
 345 out_unlock:
 346         xprt_clear_locked(xprt);
 347 }
 348
 349 /**
 350  * xprt_release_xprt - allow other requests to use a transport
 351  * @xprt: transport with other tasks potentially waiting
 352  * @task: task that is releasing access to the transport
 353  *
 354  * Note that "task" can be NULL.  No congestion control is provided.
 355  */
 356 void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 357 {
 358         if (xprt->snd_task == task) {
 359                 xprt_clear_locked(xprt);
 360                 __xprt_lock_write_next(xprt);
 361         }
 362         trace_xprt_release_xprt(xprt, task);
 363 }
 364 EXPORT_SYMBOL_GPL(xprt_release_xprt);
 365
 366 /**
 367  * xprt_release_xprt_cong - allow other requests to use a transport
 368  * @xprt: transport with other tasks potentially waiting
 369  * @task: task that is releasing access to the transport
 370  *
 371  * Note that "task" can be NULL.  Another task is awoken to use the
 372  * transport if the transport's congestion window allows it.
 373  */
 374 void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 375 {
 376         if (xprt->snd_task == task) {
 377                 xprt_clear_locked(xprt);
 378                 __xprt_lock_write_next_cong(xprt);
 379         }
 380         trace_xprt_release_cong(xprt, task);
 381 }
 382 EXPORT_SYMBOL_GPL(xprt_release_xprt_cong);
 383
 384 static inline void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
 385 {
 386         if (xprt->snd_task != task)
 387                 return;
 388         spin_lock(&xprt->transport_lock);
 389         xprt->ops->release_xprt(xprt, task);
 390         spin_unlock(&xprt->transport_lock);
 391 }
 392
 393 /*
 394  * Van Jacobson congestion avoidance. Check if the congestion window
 395  * overflowed. Put the task to sleep if this is the case.
 396  */
 397 static int
 398 __xprt_get_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
 399 {
 400         if (req->rq_cong)
 401                 return 1;
 402         trace_xprt_get_cong(xprt, req->rq_task);
 403         if (RPCXPRT_CONGESTED(xprt)) {
 404                 xprt_set_congestion_window_wait(xprt);
 405                 return 0;
 406         }
 407         req->rq_cong = 1;
 408         xprt->cong += RPC_CWNDSCALE;
 409         return 1;
 410 }
 411
 412 /*
 413  * Adjust the congestion window, and wake up the next task
 414  * that has been sleeping due to congestion
 415  */
 416 static void
 417 __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
 418 {
 419         if (!req->rq_cong)
 420                 return;
 421         req->rq_cong = 0;
 422         xprt->cong -= RPC_CWNDSCALE;
 423         xprt_test_and_clear_congestion_window_wait(xprt);
 424         trace_xprt_put_cong(xprt, req->rq_task);
 425         __xprt_lock_write_next_cong(xprt);
 426 }
 427
 428 /**
 429  * xprt_request_get_cong - Request congestion control credits
 430  * @xprt: pointer to transport
 431  * @req: pointer to RPC request
 432  *
 433  * Useful for transports that require congestion control.
 434  */
 435 bool
 436 xprt_request_get_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
 437 {
 438         bool ret = false;
 439
 440         if (req->rq_cong)
 441                 return true;
 442         spin_lock(&xprt->transport_lock);
 443         ret = __xprt_get_cong(xprt, req) != 0;
 444         spin_unlock(&xprt->transport_lock);
 445         return ret;
 446 }
 447 EXPORT_SYMBOL_GPL(xprt_request_get_cong);
 448
 449 /**
 450  * xprt_release_rqst_cong - housekeeping when request is complete
 451  * @task: RPC request that recently completed
 452  *
 453  * Useful for transports that require congestion control.
 454  */
 455 void xprt_release_rqst_cong(struct rpc_task *task)
 456 {
 457         struct rpc_rqst *req = task->tk_rqstp;
 458
 459         __xprt_put_cong(req->rq_xprt, req);
 460 }
 461 EXPORT_SYMBOL_GPL(xprt_release_rqst_cong);
 462
 463 static void xprt_clear_congestion_window_wait_locked(struct rpc_xprt *xprt)
 464 {
 465         if (test_and_clear_bit(XPRT_CWND_WAIT, &xprt->state))
 466                 __xprt_lock_write_next_cong(xprt);
 467 }
 468
 469 /*
 470  * Clear the congestion window wait flag and wake up the next
 471  * entry on xprt->sending
 472  */
 473 static void
 474 xprt_clear_congestion_window_wait(struct rpc_xprt *xprt)
 475 {
 476         if (test_and_clear_bit(XPRT_CWND_WAIT, &xprt->state)) {
 477                 spin_lock(&xprt->transport_lock);
 478                 __xprt_lock_write_next_cong(xprt);
 479                 spin_unlock(&xprt->transport_lock);
 480         }
 481 }
 482
 483 /**
 484  * xprt_adjust_cwnd - adjust transport congestion window
 485  * @xprt: pointer to xprt
 486  * @task: recently completed RPC request used to adjust window
 487  * @result: result code of completed RPC request
 488  *
 489  * The transport code maintains an estimate on the maximum number of out-
 490  * standing RPC requests, using a smoothed version of the congestion
 491  * avoidance implemented in 44BSD. This is basically the Van Jacobson
 492  * congestion algorithm: If a retransmit occurs, the congestion window is
 493  * halved; otherwise, it is incremented by 1/cwnd when
 494  *
 495  *      -       a reply is received and
 496  *      -       a full number of requests are outstanding and
 497  *      -       the congestion window hasn't been updated recently.
 498  */
 499 void xprt_adjust_cwnd(struct rpc_xprt *xprt, struct rpc_task *task, int result)
 500 {
 501         struct rpc_rqst *req = task->tk_rqstp;
 502         unsigned long cwnd = xprt->cwnd;
 503
 504         if (result >= 0 && cwnd <= xprt->cong) {
 505                 /* The (cwnd >> 1) term makes sure
 506                  * the result gets rounded properly. */
 507                 cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
 508                 if (cwnd > RPC_MAXCWND(xprt))
 509                         cwnd = RPC_MAXCWND(xprt);
 510                 __xprt_lock_write_next_cong(xprt);
 511         } else if (result == -ETIMEDOUT) {
 512                 cwnd >>= 1;
 513                 if (cwnd < RPC_CWNDSCALE)
 514                         cwnd = RPC_CWNDSCALE;
 515         }
 516         dprintk("RPC:       cong %ld, cwnd was %ld, now %ld\n",
 517                         xprt->cong, xprt->cwnd, cwnd);
 518         xprt->cwnd = cwnd;
 519         __xprt_put_cong(xprt, req);
 520 }
 521 EXPORT_SYMBOL_GPL(xprt_adjust_cwnd);
 522
 523 /**
 524  * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
 525  * @xprt: transport with waiting tasks
 526  * @status: result code to plant in each task before waking it
 527  *
 528  */
 529 void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
 530 {
 531         if (status < 0)
 532                 rpc_wake_up_status(&xprt->pending, status);
 533         else
 534                 rpc_wake_up(&xprt->pending);
 535 }
 536 EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks);
 537
 538 /**
 539  * xprt_wait_for_buffer_space - wait for transport output buffer to clear
 540  * @xprt: transport
 541  *
 542  * Note that we only set the timer for the case of RPC_IS_SOFT(), since
 543  * we don't in general want to force a socket disconnection due to
 544  * an incomplete RPC call transmission.
 545  */
 546 void xprt_wait_for_buffer_space(struct rpc_xprt *xprt)
 547 {
 548         set_bit(XPRT_WRITE_SPACE, &xprt->state);
 549 }
 550 EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space);
 551
 552 static bool
 553 xprt_clear_write_space_locked(struct rpc_xprt *xprt)
 554 {
 555         if (test_and_clear_bit(XPRT_WRITE_SPACE, &xprt->state)) {
 556                 __xprt_lock_write_next(xprt);
 557                 dprintk("RPC:       write space: waking waiting task on "
 558                                 "xprt %p\n", xprt);
 559                 return true;
 560         }
 561         return false;
 562 }
 563
 564 /**
 565  * xprt_write_space - wake the task waiting for transport output buffer space
 566  * @xprt: transport with waiting tasks
 567  *
 568  * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
 569  */
 570 bool xprt_write_space(struct rpc_xprt *xprt)
 571 {
 572         bool ret;
 573
 574         if (!test_bit(XPRT_WRITE_SPACE, &xprt->state))
 575                 return false;
 576         spin_lock(&xprt->transport_lock);
 577         ret = xprt_clear_write_space_locked(xprt);
 578         spin_unlock(&xprt->transport_lock);
 579         return ret;
 580 }
 581 EXPORT_SYMBOL_GPL(xprt_write_space);
 582
 583 static unsigned long xprt_abs_ktime_to_jiffies(ktime_t abstime)
 584 {
 585         s64 delta = ktime_to_ns(ktime_get() - abstime);
 586         return likely(delta >= 0) ?
 587                 jiffies - nsecs_to_jiffies(delta) :
 588                 jiffies + nsecs_to_jiffies(-delta);
 589 }
 590
 591 static unsigned long xprt_calc_majortimeo(struct rpc_rqst *req)
 592 {
 593         const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
 594         unsigned long majortimeo = req->rq_timeout;
 595
 596         if (to->to_exponential)
 597                 majortimeo <<= to->to_retries;
 598         else
 599                 majortimeo += to->to_increment * to->to_retries;
 600         if (majortimeo > to->to_maxval || majortimeo == 0)
 601                 majortimeo = to->to_maxval;
 602         return majortimeo;
 603 }
 604
 605 static void xprt_reset_majortimeo(struct rpc_rqst *req)
 606 {
 607         req->rq_majortimeo += xprt_calc_majortimeo(req);
 608 }
 609
 610 static void xprt_reset_minortimeo(struct rpc_rqst *req)
 611 {
 612         req->rq_minortimeo += req->rq_timeout;
 613 }
 614
 615 static void xprt_init_majortimeo(struct rpc_task *task, struct rpc_rqst *req)
 616 {
 617         unsigned long time_init;
 618         struct rpc_xprt *xprt = req->rq_xprt;
 619
 620         if (likely(xprt && xprt_connected(xprt)))
 621                 time_init = jiffies;
 622         else
 623                 time_init = xprt_abs_ktime_to_jiffies(task->tk_start);
 624         req->rq_timeout = task->tk_client->cl_timeout->to_initval;
 625         req->rq_majortimeo = time_init + xprt_calc_majortimeo(req);
 626         req->rq_minortimeo = time_init + req->rq_timeout;
 627 }
 628
 629 /**
 630  * xprt_adjust_timeout - adjust timeout values for next retransmit
 631  * @req: RPC request containing parameters to use for the adjustment
 632  *
 633  */
 634 int xprt_adjust_timeout(struct rpc_rqst *req)
 635 {
 636         struct rpc_xprt *xprt = req->rq_xprt;
 637         const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
 638         int status = 0;
 639
 640         if (time_before(jiffies, req->rq_minortimeo))
 641                 return status;
 642         if (time_before(jiffies, req->rq_majortimeo)) {
 643                 if (to->to_exponential)
 644                         req->rq_timeout <<= 1;
 645                 else
 646                         req->rq_timeout += to->to_increment;
 647                 if (to->to_maxval && req->rq_timeout >= to->to_maxval)
 648                         req->rq_timeout = to->to_maxval;
 649                 req->rq_retries++;
 650         } else {
 651                 req->rq_timeout = to->to_initval;
 652                 req->rq_retries = 0;
 653                 xprt_reset_majortimeo(req);
 654                 /* Reset the RTT counters == "slow start" */
 655                 spin_lock(&xprt->transport_lock);
 656                 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
 657                 spin_unlock(&xprt->transport_lock);
 658                 status = -ETIMEDOUT;
 659         }
 660         xprt_reset_minortimeo(req);
 661
 662         if (req->rq_timeout == 0) {
 663                 printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
 664                 req->rq_timeout = 5 * HZ;
 665         }
 666         return status;
 667 }
 668
 669 static void xprt_autoclose(struct work_struct *work)
 670 {
 671         struct rpc_xprt *xprt =
 672                 container_of(work, struct rpc_xprt, task_cleanup);
 673         unsigned int pflags = memalloc_nofs_save();
 674
 675         trace_xprt_disconnect_auto(xprt);
 676         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 677         xprt->ops->close(xprt);
 678         xprt_release_write(xprt, NULL);
 679         wake_up_bit(&xprt->state, XPRT_LOCKED);
 680         memalloc_nofs_restore(pflags);
 681 }
 682
 683 /**
 684  * xprt_disconnect_done - mark a transport as disconnected
 685  * @xprt: transport to flag for disconnect
 686  *
 687  */
 688 void xprt_disconnect_done(struct rpc_xprt *xprt)
 689 {
 690         trace_xprt_disconnect_done(xprt);
 691         spin_lock(&xprt->transport_lock);
 692         xprt_clear_connected(xprt);
 693         xprt_clear_write_space_locked(xprt);
 694         xprt_clear_congestion_window_wait_locked(xprt);
 695         xprt_wake_pending_tasks(xprt, -ENOTCONN);
 696         spin_unlock(&xprt->transport_lock);
 697 }
 698 EXPORT_SYMBOL_GPL(xprt_disconnect_done);
 699
 700 /**
 701  * xprt_force_disconnect - force a transport to disconnect
 702  * @xprt: transport to disconnect
 703  *
 704  */
 705 void xprt_force_disconnect(struct rpc_xprt *xprt)
 706 {
 707         trace_xprt_disconnect_force(xprt);
 708
 709         /* Don't race with the test_bit() in xprt_clear_locked() */
 710         spin_lock(&xprt->transport_lock);
 711         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 712         /* Try to schedule an autoclose RPC call */
 713         if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 714                 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 715         else if (xprt->snd_task)
 716                 rpc_wake_up_queued_task_set_status(&xprt->pending,
 717                                 xprt->snd_task, -ENOTCONN);
 718         spin_unlock(&xprt->transport_lock);
 719 }
 720 EXPORT_SYMBOL_GPL(xprt_force_disconnect);
 721
 722 static unsigned int
 723 xprt_connect_cookie(struct rpc_xprt *xprt)
 724 {
 725         return READ_ONCE(xprt->connect_cookie);
 726 }
 727
 728 static bool
 729 xprt_request_retransmit_after_disconnect(struct rpc_task *task)
 730 {
 731         struct rpc_rqst *req = task->tk_rqstp;
 732         struct rpc_xprt *xprt = req->rq_xprt;
 733
 734         return req->rq_connect_cookie != xprt_connect_cookie(xprt) ||
 735                 !xprt_connected(xprt);
 736 }
 737
 738 /**
 739  * xprt_conditional_disconnect - force a transport to disconnect
 740  * @xprt: transport to disconnect
 741  * @cookie: 'connection cookie'
 742  *
 743  * This attempts to break the connection if and only if 'cookie' matches
 744  * the current transport 'connection cookie'. It ensures that we don't
 745  * try to break the connection more than once when we need to retransmit
 746  * a batch of RPC requests.
 747  *
 748  */
 749 void xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie)
 750 {
 751         /* Don't race with the test_bit() in xprt_clear_locked() */
 752         spin_lock(&xprt->transport_lock);
 753         if (cookie != xprt->connect_cookie)
 754                 goto out;
 755         if (test_bit(XPRT_CLOSING, &xprt->state))
 756                 goto out;
 757         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 758         /* Try to schedule an autoclose RPC call */
 759         if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 760                 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 761         xprt_wake_pending_tasks(xprt, -EAGAIN);
 762 out:
 763         spin_unlock(&xprt->transport_lock);
 764 }
 765
 766 static bool
 767 xprt_has_timer(const struct rpc_xprt *xprt)
 768 {
 769         return xprt->idle_timeout != 0;
 770 }
 771
 772 static void
 773 xprt_schedule_autodisconnect(struct rpc_xprt *xprt)
 774         __must_hold(&xprt->transport_lock)
 775 {
 776         xprt->last_used = jiffies;
 777         if (RB_EMPTY_ROOT(&xprt->recv_queue) && xprt_has_timer(xprt))
 778                 mod_timer(&xprt->timer, xprt->last_used + xprt->idle_timeout);
 779 }
 780
 781 static void
 782 xprt_init_autodisconnect(struct timer_list *t)
 783 {
 784         struct rpc_xprt *xprt = from_timer(xprt, t, timer);
 785
 786         if (!RB_EMPTY_ROOT(&xprt->recv_queue))
 787                 return;
 788         /* Reset xprt->last_used to avoid connect/autodisconnect cycling */
 789         xprt->last_used = jiffies;
 790         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 791                 return;
 792         queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 793 }
 794
 795 bool xprt_lock_connect(struct rpc_xprt *xprt,
 796                 struct rpc_task *task,
 797                 void *cookie)
 798 {
 799         bool ret = false;
 800
 801         spin_lock(&xprt->transport_lock);
 802         if (!test_bit(XPRT_LOCKED, &xprt->state))
 803                 goto out;
 804         if (xprt->snd_task != task)
 805                 goto out;
 806         xprt->snd_task = cookie;
 807         ret = true;
 808 out:
 809         spin_unlock(&xprt->transport_lock);
 810         return ret;
 811 }
 812
 813 void xprt_unlock_connect(struct rpc_xprt *xprt, void *cookie)
 814 {
 815         spin_lock(&xprt->transport_lock);
 816         if (xprt->snd_task != cookie)
 817                 goto out;
 818         if (!test_bit(XPRT_LOCKED, &xprt->state))
 819                 goto out;
 820         xprt->snd_task =NULL;
 821         xprt->ops->release_xprt(xprt, NULL);
 822         xprt_schedule_autodisconnect(xprt);
 823 out:
 824         spin_unlock(&xprt->transport_lock);
 825         wake_up_bit(&xprt->state, XPRT_LOCKED);
 826 }
 827
 828 /**
 829  * xprt_connect - schedule a transport connect operation
 830  * @task: RPC task that is requesting the connect
 831  *
 832  */
 833 void xprt_connect(struct rpc_task *task)
 834 {
 835         struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
 836
 837         dprintk("RPC: %5u xprt_connect xprt %p %s connected\n", task->tk_pid,
 838                         xprt, (xprt_connected(xprt) ? "is" : "is not"));
 839
 840         if (!xprt_bound(xprt)) {
 841                 task->tk_status = -EAGAIN;
 842                 return;
 843         }
 844         if (!xprt_lock_write(xprt, task))
 845                 return;
 846
 847         if (test_and_clear_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
 848                 trace_xprt_disconnect_cleanup(xprt);
 849                 xprt->ops->close(xprt);
 850         }
 851
 852         if (!xprt_connected(xprt)) {
 853                 task->tk_rqstp->rq_connect_cookie = xprt->connect_cookie;
 854                 rpc_sleep_on_timeout(&xprt->pending, task, NULL,
 855                                 xprt_request_timeout(task->tk_rqstp));
 856
 857                 if (test_bit(XPRT_CLOSING, &xprt->state))
 858                         return;
 859                 if (xprt_test_and_set_connecting(xprt))
 860                         return;
 861                 /* Race breaker */
 862                 if (!xprt_connected(xprt)) {
 863                         xprt->stat.connect_start = jiffies;
 864                         xprt->ops->connect(xprt, task);
 865                 } else {
 866                         xprt_clear_connecting(xprt);
 867                         task->tk_status = 0;
 868                         rpc_wake_up_queued_task(&xprt->pending, task);
 869                 }
 870         }
 871         xprt_release_write(xprt, task);
 872 }
 873
 874 /**
 875  * xprt_reconnect_delay - compute the wait before scheduling a connect
 876  * @xprt: transport instance
 877  *
 878  */
 879 unsigned long xprt_reconnect_delay(const struct rpc_xprt *xprt)
 880 {
 881         unsigned long start, now = jiffies;
 882
 883         start = xprt->stat.connect_start + xprt->reestablish_timeout;
 884         if (time_after(start, now))
 885                 return start - now;
 886         return 0;
 887 }
 888 EXPORT_SYMBOL_GPL(xprt_reconnect_delay);
 889
 890 /**
 891  * xprt_reconnect_backoff - compute the new re-establish timeout
 892  * @xprt: transport instance
 893  * @init_to: initial reestablish timeout
 894  *
 895  */
 896 void xprt_reconnect_backoff(struct rpc_xprt *xprt, unsigned long init_to)
 897 {
 898         xprt->reestablish_timeout <<= 1;
 899         if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
 900                 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
 901         if (xprt->reestablish_timeout < init_to)
 902                 xprt->reestablish_timeout = init_to;
 903 }
 904 EXPORT_SYMBOL_GPL(xprt_reconnect_backoff);
 905
 906 enum xprt_xid_rb_cmp {
 907         XID_RB_EQUAL,
 908         XID_RB_LEFT,
 909         XID_RB_RIGHT,
 910 };
 911 static enum xprt_xid_rb_cmp
 912 xprt_xid_cmp(__be32 xid1, __be32 xid2)
 913 {
 914         if (xid1 == xid2)
 915                 return XID_RB_EQUAL;
 916         if ((__force u32)xid1 < (__force u32)xid2)
 917                 return XID_RB_LEFT;
 918         return XID_RB_RIGHT;
 919 }
 920
 921 static struct rpc_rqst *
 922 xprt_request_rb_find(struct rpc_xprt *xprt, __be32 xid)
 923 {
 924         struct rb_node *n = xprt->recv_queue.rb_node;
 925         struct rpc_rqst *req;
 926
 927         while (n != NULL) {
 928                 req = rb_entry(n, struct rpc_rqst, rq_recv);
 929                 switch (xprt_xid_cmp(xid, req->rq_xid)) {
 930                 case XID_RB_LEFT:
 931                         n = n->rb_left;
 932                         break;
 933                 case XID_RB_RIGHT:
 934                         n = n->rb_right;
 935                         break;
 936                 case XID_RB_EQUAL:
 937                         return req;
 938                 }
 939         }
 940         return NULL;
 941 }
 942
 943 static void
 944 xprt_request_rb_insert(struct rpc_xprt *xprt, struct rpc_rqst *new)
 945 {
 946         struct rb_node **p = &xprt->recv_queue.rb_node;
 947         struct rb_node *n = NULL;
 948         struct rpc_rqst *req;
 949
 950         while (*p != NULL) {
 951                 n = *p;
 952                 req = rb_entry(n, struct rpc_rqst, rq_recv);
 953                 switch(xprt_xid_cmp(new->rq_xid, req->rq_xid)) {
 954                 case XID_RB_LEFT:
 955                         p = &n->rb_left;
 956                         break;
 957                 case XID_RB_RIGHT:
 958                         p = &n->rb_right;
 959                         break;
 960                 case XID_RB_EQUAL:
 961                         WARN_ON_ONCE(new != req);
 962                         return;
 963                 }
 964         }
 965         rb_link_node(&new->rq_recv, n, p);
 966         rb_insert_color(&new->rq_recv, &xprt->recv_queue);
 967 }
 968
 969 static void
 970 xprt_request_rb_remove(struct rpc_xprt *xprt, struct rpc_rqst *req)
 971 {
 972         rb_erase(&req->rq_recv, &xprt->recv_queue);
 973 }
 974
 975 /**
 976  * xprt_lookup_rqst - find an RPC request corresponding to an XID
 977  * @xprt: transport on which the original request was transmitted
 978  * @xid: RPC XID of incoming reply
 979  *
 980  * Caller holds xprt->queue_lock.
 981  */
 982 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
 983 {
 984         struct rpc_rqst *entry;
 985
 986         entry = xprt_request_rb_find(xprt, xid);
 987         if (entry != NULL) {
 988                 trace_xprt_lookup_rqst(xprt, xid, 0);
 989                 entry->rq_rtt = ktime_sub(ktime_get(), entry->rq_xtime);
 990                 return entry;
 991         }
 992
 993         dprintk("RPC:       xprt_lookup_rqst did not find xid %08x\n",
 994                         ntohl(xid));
 995         trace_xprt_lookup_rqst(xprt, xid, -ENOENT);
 996         xprt->stat.bad_xids++;
 997         return NULL;
 998 }
 999 EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
1000
1001 static bool
1002 xprt_is_pinned_rqst(struct rpc_rqst *req)
1003 {
1004         return atomic_read(&req->rq_pin) != 0;
1005 }
1006
1007 /**
1008  * xprt_pin_rqst - Pin a request on the transport receive list
1009  * @req: Request to pin
1010  *
1011  * Caller must ensure this is atomic with the call to xprt_lookup_rqst()
1012  * so should be holding xprt->queue_lock.
1013  */
1014 void xprt_pin_rqst(struct rpc_rqst *req)
1015 {
1016         atomic_inc(&req->rq_pin);
1017 }
1018 EXPORT_SYMBOL_GPL(xprt_pin_rqst);
1019
1020 /**
1021  * xprt_unpin_rqst - Unpin a request on the transport receive list
1022  * @req: Request to pin
1023  *
1024  * Caller should be holding xprt->queue_lock.
1025  */
1026 void xprt_unpin_rqst(struct rpc_rqst *req)
1027 {
1028         if (!test_bit(RPC_TASK_MSG_PIN_WAIT, &req->rq_task->tk_runstate)) {
1029                 atomic_dec(&req->rq_pin);
1030                 return;
1031         }
1032         if (atomic_dec_and_test(&req->rq_pin))
1033                 wake_up_var(&req->rq_pin);
1034 }
1035 EXPORT_SYMBOL_GPL(xprt_unpin_rqst);
1036
1037 static void xprt_wait_on_pinned_rqst(struct rpc_rqst *req)
1038 {
1039         wait_var_event(&req->rq_pin, !xprt_is_pinned_rqst(req));
1040 }
1041
1042 static bool
1043 xprt_request_data_received(struct rpc_task *task)
1044 {
1045         return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
1046                 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) != 0;
1047 }
1048
1049 static bool
1050 xprt_request_need_enqueue_receive(struct rpc_task *task, struct rpc_rqst *req)
1051 {
1052         return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
1053                 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) == 0;
1054 }
1055
1056 /**
1057  * xprt_request_enqueue_receive - Add an request to the receive queue
1058  * @task: RPC task
1059  *
1060  */
1061 void
1062 xprt_request_enqueue_receive(struct rpc_task *task)
1063 {
1064         struct rpc_rqst *req = task->tk_rqstp;
1065         struct rpc_xprt *xprt = req->rq_xprt;
1066
1067         if (!xprt_request_need_enqueue_receive(task, req))
1068                 return;
1069
1070         xprt_request_prepare(task->tk_rqstp);
1071         spin_lock(&xprt->queue_lock);
1072
1073         /* Update the softirq receive buffer */
1074         memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
1075                         sizeof(req->rq_private_buf));
1076
1077         /* Add request to the receive list */
1078         xprt_request_rb_insert(xprt, req);
1079         set_bit(RPC_TASK_NEED_RECV, &task->tk_runstate);
1080         spin_unlock(&xprt->queue_lock);
1081
1082         /* Turn off autodisconnect */
1083         del_singleshot_timer_sync(&xprt->timer);
1084 }
1085
1086 /**
1087  * xprt_request_dequeue_receive_locked - Remove a request from the receive queue
1088  * @task: RPC task
1089  *
1090  * Caller must hold xprt->queue_lock.
1091  */
1092 static void
1093 xprt_request_dequeue_receive_locked(struct rpc_task *task)
1094 {
1095         struct rpc_rqst *req = task->tk_rqstp;
1096
1097         if (test_and_clear_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1098                 xprt_request_rb_remove(req->rq_xprt, req);
1099 }
1100
1101 /**
1102  * xprt_update_rtt - Update RPC RTT statistics
1103  * @task: RPC request that recently completed
1104  *
1105  * Caller holds xprt->queue_lock.
1106  */
1107 void xprt_update_rtt(struct rpc_task *task)
1108 {
1109         struct rpc_rqst *req = task->tk_rqstp;
1110         struct rpc_rtt *rtt = task->tk_client->cl_rtt;
1111         unsigned int timer = task->tk_msg.rpc_proc->p_timer;
1112         long m = usecs_to_jiffies(ktime_to_us(req->rq_rtt));
1113
1114         if (timer) {
1115                 if (req->rq_ntrans == 1)
1116                         rpc_update_rtt(rtt, timer, m);
1117                 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
1118         }
1119 }
1120 EXPORT_SYMBOL_GPL(xprt_update_rtt);
1121
1122 /**
1123  * xprt_complete_rqst - called when reply processing is complete
1124  * @task: RPC request that recently completed
1125  * @copied: actual number of bytes received from the transport
1126  *
1127  * Caller holds xprt->queue_lock.
1128  */
1129 void xprt_complete_rqst(struct rpc_task *task, int copied)
1130 {
1131         struct rpc_rqst *req = task->tk_rqstp;
1132         struct rpc_xprt *xprt = req->rq_xprt;
1133
1134         trace_xprt_complete_rqst(xprt, req->rq_xid, copied);
1135
1136         xprt->stat.recvs++;
1137
1138         req->rq_private_buf.len = copied;
1139         /* Ensure all writes are done before we update */
1140         /* req->rq_reply_bytes_recvd */
1141         smp_wmb();
1142         req->rq_reply_bytes_recvd = copied;
1143         xprt_request_dequeue_receive_locked(task);
1144         rpc_wake_up_queued_task(&xprt->pending, task);
1145 }
1146 EXPORT_SYMBOL_GPL(xprt_complete_rqst);
1147
1148 static void xprt_timer(struct rpc_task *task)
1149 {
1150         struct rpc_rqst *req = task->tk_rqstp;
1151         struct rpc_xprt *xprt = req->rq_xprt;
1152
1153         if (task->tk_status != -ETIMEDOUT)
1154                 return;
1155
1156         trace_xprt_timer(xprt, req->rq_xid, task->tk_status);
1157         if (!req->rq_reply_bytes_recvd) {
1158                 if (xprt->ops->timer)
1159                         xprt->ops->timer(xprt, task);
1160         } else
1161                 task->tk_status = 0;
1162 }
1163
1164 /**
1165  * xprt_wait_for_reply_request_def - wait for reply
1166  * @task: pointer to rpc_task
1167  *
1168  * Set a request's retransmit timeout based on the transport's
1169  * default timeout parameters.  Used by transports that don't adjust
1170  * the retransmit timeout based on round-trip time estimation,
1171  * and put the task to sleep on the pending queue.
1172  */
1173 void xprt_wait_for_reply_request_def(struct rpc_task *task)
1174 {
1175         struct rpc_rqst *req = task->tk_rqstp;
1176
1177         rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1178                         xprt_request_timeout(req));
1179 }
1180 EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_def);
1181
1182 /**
1183  * xprt_wait_for_reply_request_rtt - wait for reply using RTT estimator
1184  * @task: pointer to rpc_task
1185  *
1186  * Set a request's retransmit timeout using the RTT estimator,
1187  * and put the task to sleep on the pending queue.
1188  */
1189 void xprt_wait_for_reply_request_rtt(struct rpc_task *task)
1190 {
1191         int timer = task->tk_msg.rpc_proc->p_timer;
1192         struct rpc_clnt *clnt = task->tk_client;
1193         struct rpc_rtt *rtt = clnt->cl_rtt;
1194         struct rpc_rqst *req = task->tk_rqstp;
1195         unsigned long max_timeout = clnt->cl_timeout->to_maxval;
1196         unsigned long timeout;
1197
1198         timeout = rpc_calc_rto(rtt, timer);
1199         timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
1200         if (timeout > max_timeout || timeout == 0)
1201                 timeout = max_timeout;
1202         rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1203                         jiffies + timeout);
1204 }
1205 EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_rtt);
1206
1207 /**
1208  * xprt_request_wait_receive - wait for the reply to an RPC request
1209  * @task: RPC task about to send a request
1210  *
1211  */
1212 void xprt_request_wait_receive(struct rpc_task *task)
1213 {
1214         struct rpc_rqst *req = task->tk_rqstp;
1215         struct rpc_xprt *xprt = req->rq_xprt;
1216
1217         if (!test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1218                 return;
1219         /*
1220          * Sleep on the pending queue if we're expecting a reply.
1221          * The spinlock ensures atomicity between the test of
1222          * req->rq_reply_bytes_recvd, and the call to rpc_sleep_on().
1223          */
1224         spin_lock(&xprt->queue_lock);
1225         if (test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate)) {
1226                 xprt->ops->wait_for_reply_request(task);
1227                 /*
1228                  * Send an extra queue wakeup call if the
1229                  * connection was dropped in case the call to
1230                  * rpc_sleep_on() raced.
1231                  */
1232                 if (xprt_request_retransmit_after_disconnect(task))
1233                         rpc_wake_up_queued_task_set_status(&xprt->pending,
1234                                         task, -ENOTCONN);
1235         }
1236         spin_unlock(&xprt->queue_lock);
1237 }
1238
1239 static bool
1240 xprt_request_need_enqueue_transmit(struct rpc_task *task, struct rpc_rqst *req)
1241 {
1242         return !test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1243 }
1244
1245 /**
1246  * xprt_request_enqueue_transmit - queue a task for transmission
1247  * @task: pointer to rpc_task
1248  *
1249  * Add a task to the transmission queue.
1250  */
1251 void
1252 xprt_request_enqueue_transmit(struct rpc_task *task)
1253 {
1254         struct rpc_rqst *pos, *req = task->tk_rqstp;
1255         struct rpc_xprt *xprt = req->rq_xprt;
1256
1257         if (xprt_request_need_enqueue_transmit(task, req)) {
1258                 req->rq_bytes_sent = 0;
1259                 spin_lock(&xprt->queue_lock);
1260                 /*
1261                  * Requests that carry congestion control credits are added
1262                  * to the head of the list to avoid starvation issues.
1263                  */
1264                 if (req->rq_cong) {
1265                         xprt_clear_congestion_window_wait(xprt);
1266                         list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1267                                 if (pos->rq_cong)
1268                                         continue;
1269                                 /* Note: req is added _before_ pos */
1270                                 list_add_tail(&req->rq_xmit, &pos->rq_xmit);
1271                                 INIT_LIST_HEAD(&req->rq_xmit2);
1272                                 trace_xprt_enq_xmit(task, 1);
1273                                 goto out;
1274                         }
1275                 } else if (RPC_IS_SWAPPER(task)) {
1276                         list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1277                                 if (pos->rq_cong || pos->rq_bytes_sent)
1278                                         continue;
1279                                 if (RPC_IS_SWAPPER(pos->rq_task))
1280                                         continue;
1281                                 /* Note: req is added _before_ pos */
1282                                 list_add_tail(&req->rq_xmit, &pos->rq_xmit);
1283                                 INIT_LIST_HEAD(&req->rq_xmit2);
1284                                 trace_xprt_enq_xmit(task, 2);
1285                                 goto out;
1286                         }
1287                 } else if (!req->rq_seqno) {
1288                         list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1289                                 if (pos->rq_task->tk_owner != task->tk_owner)
1290                                         continue;
1291                                 list_add_tail(&req->rq_xmit2, &pos->rq_xmit2);
1292                                 INIT_LIST_HEAD(&req->rq_xmit);
1293                                 trace_xprt_enq_xmit(task, 3);
1294                                 goto out;
1295                         }
1296                 }
1297                 list_add_tail(&req->rq_xmit, &xprt->xmit_queue);
1298                 INIT_LIST_HEAD(&req->rq_xmit2);
1299                 trace_xprt_enq_xmit(task, 4);
1300 out:
1301                 set_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1302                 spin_unlock(&xprt->queue_lock);
1303         }
1304 }
1305
1306 /**
1307  * xprt_request_dequeue_transmit_locked - remove a task from the transmission queue
1308  * @task: pointer to rpc_task
1309  *
1310  * Remove a task from the transmission queue
1311  * Caller must hold xprt->queue_lock
1312  */
1313 static void
1314 xprt_request_dequeue_transmit_locked(struct rpc_task *task)
1315 {
1316         struct rpc_rqst *req = task->tk_rqstp;
1317
1318         if (!test_and_clear_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1319                 return;
1320         if (!list_empty(&req->rq_xmit)) {
1321                 list_del(&req->rq_xmit);
1322                 if (!list_empty(&req->rq_xmit2)) {
1323                         struct rpc_rqst *next = list_first_entry(&req->rq_xmit2,
1324                                         struct rpc_rqst, rq_xmit2);
1325                         list_del(&req->rq_xmit2);
1326                         list_add_tail(&next->rq_xmit, &next->rq_xprt->xmit_queue);
1327                 }
1328         } else
1329                 list_del(&req->rq_xmit2);
1330 }
1331
1332 /**
1333  * xprt_request_dequeue_transmit - remove a task from the transmission queue
1334  * @task: pointer to rpc_task
1335  *
1336  * Remove a task from the transmission queue
1337  */
1338 static void
1339 xprt_request_dequeue_transmit(struct rpc_task *task)
1340 {
1341         struct rpc_rqst *req = task->tk_rqstp;
1342         struct rpc_xprt *xprt = req->rq_xprt;
1343
1344         spin_lock(&xprt->queue_lock);
1345         xprt_request_dequeue_transmit_locked(task);
1346         spin_unlock(&xprt->queue_lock);
1347 }
1348
1349 /**
1350  * xprt_request_dequeue_xprt - remove a task from the transmit+receive queue
1351  * @task: pointer to rpc_task
1352  *
1353  * Remove a task from the transmit and receive queues, and ensure that
1354  * it is not pinned by the receive work item.
1355  */
1356 void
1357 xprt_request_dequeue_xprt(struct rpc_task *task)
1358 {
1359         struct rpc_rqst *req = task->tk_rqstp;
1360         struct rpc_xprt *xprt = req->rq_xprt;
1361
1362         if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate) ||
1363             test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) ||
1364             xprt_is_pinned_rqst(req)) {
1365                 spin_lock(&xprt->queue_lock);
1366                 xprt_request_dequeue_transmit_locked(task);
1367                 xprt_request_dequeue_receive_locked(task);
1368                 while (xprt_is_pinned_rqst(req)) {
1369                         set_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1370                         spin_unlock(&xprt->queue_lock);
1371                         xprt_wait_on_pinned_rqst(req);
1372                         spin_lock(&xprt->queue_lock);
1373                         clear_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1374                 }
1375                 spin_unlock(&xprt->queue_lock);
1376         }
1377 }
1378
1379 /**
1380  * xprt_request_prepare - prepare an encoded request for transport
1381  * @req: pointer to rpc_rqst
1382  *
1383  * Calls into the transport layer to do whatever is needed to prepare
1384  * the request for transmission or receive.
1385  */
1386 void
1387 xprt_request_prepare(struct rpc_rqst *req)
1388 {
1389         struct rpc_xprt *xprt = req->rq_xprt;
1390
1391         if (xprt->ops->prepare_request)
1392                 xprt->ops->prepare_request(req);
1393 }
1394
1395 /**
1396  * xprt_request_need_retransmit - Test if a task needs retransmission
1397  * @task: pointer to rpc_task
1398  *
1399  * Test for whether a connection breakage requires the task to retransmit
1400  */
1401 bool
1402 xprt_request_need_retransmit(struct rpc_task *task)
1403 {
1404         return xprt_request_retransmit_after_disconnect(task);
1405 }
1406
1407 /**
1408  * xprt_prepare_transmit - reserve the transport before sending a request
1409  * @task: RPC task about to send a request
1410  *
1411  */
1412 bool xprt_prepare_transmit(struct rpc_task *task)
1413 {
1414         struct rpc_rqst *req = task->tk_rqstp;
1415         struct rpc_xprt *xprt = req->rq_xprt;
1416
1417         dprintk("RPC: %5u xprt_prepare_transmit\n", task->tk_pid);
1418
1419         if (!xprt_lock_write(xprt, task)) {
1420                 /* Race breaker: someone may have transmitted us */
1421                 if (!test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1422                         rpc_wake_up_queued_task_set_status(&xprt->sending,
1423                                         task, 0);
1424                 return false;
1425
1426         }
1427         return true;
1428 }
1429
1430 void xprt_end_transmit(struct rpc_task *task)
1431 {
1432         xprt_release_write(task->tk_rqstp->rq_xprt, task);
1433 }
1434
1435 /**
1436  * xprt_request_transmit - send an RPC request on a transport
1437  * @req: pointer to request to transmit
1438  * @snd_task: RPC task that owns the transport lock
1439  *
1440  * This performs the transmission of a single request.
1441  * Note that if the request is not the same as snd_task, then it
1442  * does need to be pinned.
1443  * Returns '0' on success.
1444  */
1445 static int
1446 xprt_request_transmit(struct rpc_rqst *req, struct rpc_task *snd_task)
1447 {
1448         struct rpc_xprt *xprt = req->rq_xprt;
1449         struct rpc_task *task = req->rq_task;
1450         unsigned int connect_cookie;
1451         int is_retrans = RPC_WAS_SENT(task);
1452         int status;
1453
1454         if (!req->rq_bytes_sent) {
1455                 if (xprt_request_data_received(task)) {
1456                         status = 0;
1457                         goto out_dequeue;
1458                 }
1459                 /* Verify that our message lies in the RPCSEC_GSS window */
1460                 if (rpcauth_xmit_need_reencode(task)) {
1461                         status = -EBADMSG;
1462                         goto out_dequeue;
1463                 }
1464                 if (RPC_SIGNALLED(task)) {
1465                         status = -ERESTARTSYS;
1466                         goto out_dequeue;
1467                 }
1468         }
1469
1470         /*
1471          * Update req->rq_ntrans before transmitting to avoid races with
1472          * xprt_update_rtt(), which needs to know that it is recording a
1473          * reply to the first transmission.
1474          */
1475         req->rq_ntrans++;
1476
1477         trace_rpc_xdr_sendto(task, &req->rq_snd_buf);
1478         connect_cookie = xprt->connect_cookie;
1479         status = xprt->ops->send_request(req);
1480         if (status != 0) {
1481                 req->rq_ntrans--;
1482                 trace_xprt_transmit(req, status);
1483                 return status;
1484         }
1485
1486         if (is_retrans)
1487                 task->tk_client->cl_stats->rpcretrans++;
1488
1489         xprt_inject_disconnect(xprt);
1490
1491         task->tk_flags |= RPC_TASK_SENT;
1492         spin_lock(&xprt->transport_lock);
1493
1494         xprt->stat.sends++;
1495         xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
1496         xprt->stat.bklog_u += xprt->backlog.qlen;
1497         xprt->stat.sending_u += xprt->sending.qlen;
1498         xprt->stat.pending_u += xprt->pending.qlen;
1499         spin_unlock(&xprt->transport_lock);
1500
1501         req->rq_connect_cookie = connect_cookie;
1502 out_dequeue:
1503         trace_xprt_transmit(req, status);
1504         xprt_request_dequeue_transmit(task);
1505         rpc_wake_up_queued_task_set_status(&xprt->sending, task, status);
1506         return status;
1507 }
1508
1509 /**
1510  * xprt_transmit - send an RPC request on a transport
1511  * @task: controlling RPC task
1512  *
1513  * Attempts to drain the transmit queue. On exit, either the transport
1514  * signalled an error that needs to be handled before transmission can
1515  * resume, or @task finished transmitting, and detected that it already
1516  * received a reply.
1517  */
1518 void
1519 xprt_transmit(struct rpc_task *task)
1520 {
1521         struct rpc_rqst *next, *req = task->tk_rqstp;
1522         struct rpc_xprt *xprt = req->rq_xprt;
1523         int status;
1524
1525         spin_lock(&xprt->queue_lock);
1526         while (!list_empty(&xprt->xmit_queue)) {
1527                 next = list_first_entry(&xprt->xmit_queue,
1528                                 struct rpc_rqst, rq_xmit);
1529                 xprt_pin_rqst(next);
1530                 spin_unlock(&xprt->queue_lock);
1531                 status = xprt_request_transmit(next, task);
1532                 if (status == -EBADMSG && next != req)
1533                         status = 0;
1534                 cond_resched();
1535                 spin_lock(&xprt->queue_lock);
1536                 xprt_unpin_rqst(next);
1537                 if (status == 0) {
1538                         if (!xprt_request_data_received(task) ||
1539                             test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1540                                 continue;
1541                 } else if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1542                         task->tk_status = status;
1543                 break;
1544         }
1545         spin_unlock(&xprt->queue_lock);
1546 }
1547
1548 static void xprt_add_backlog(struct rpc_xprt *xprt, struct rpc_task *task)
1549 {
1550         set_bit(XPRT_CONGESTED, &xprt->state);
1551         rpc_sleep_on(&xprt->backlog, task, NULL);
1552 }
1553
1554 static void xprt_wake_up_backlog(struct rpc_xprt *xprt)
1555 {
1556         if (rpc_wake_up_next(&xprt->backlog) == NULL)
1557                 clear_bit(XPRT_CONGESTED, &xprt->state);
1558 }
1559
1560 static bool xprt_throttle_congested(struct rpc_xprt *xprt, struct rpc_task *task)
1561 {
1562         bool ret = false;
1563
1564         if (!test_bit(XPRT_CONGESTED, &xprt->state))
1565                 goto out;
1566         spin_lock(&xprt->reserve_lock);
1567         if (test_bit(XPRT_CONGESTED, &xprt->state)) {
1568                 rpc_sleep_on(&xprt->backlog, task, NULL);
1569                 ret = true;
1570         }
1571         spin_unlock(&xprt->reserve_lock);
1572 out:
1573         return ret;
1574 }
1575
1576 static struct rpc_rqst *xprt_dynamic_alloc_slot(struct rpc_xprt *xprt)
1577 {
1578         struct rpc_rqst *req = ERR_PTR(-EAGAIN);
1579
1580         if (xprt->num_reqs >= xprt->max_reqs)
1581                 goto out;
1582         ++xprt->num_reqs;
1583         spin_unlock(&xprt->reserve_lock);
1584         req = kzalloc(sizeof(struct rpc_rqst), GFP_NOFS);
1585         spin_lock(&xprt->reserve_lock);
1586         if (req != NULL)
1587                 goto out;
1588         --xprt->num_reqs;
1589         req = ERR_PTR(-ENOMEM);
1590 out:
1591         return req;
1592 }
1593
1594 static bool xprt_dynamic_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1595 {
1596         if (xprt->num_reqs > xprt->min_reqs) {
1597                 --xprt->num_reqs;
1598                 kfree(req);
1599                 return true;
1600         }
1601         return false;
1602 }
1603
1604 void xprt_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1605 {
1606         struct rpc_rqst *req;
1607
1608         spin_lock(&xprt->reserve_lock);
1609         if (!list_empty(&xprt->free)) {
1610                 req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
1611                 list_del(&req->rq_list);
1612                 goto out_init_req;
1613         }
1614         req = xprt_dynamic_alloc_slot(xprt);
1615         if (!IS_ERR(req))
1616                 goto out_init_req;
1617         switch (PTR_ERR(req)) {
1618         case -ENOMEM:
1619                 dprintk("RPC:       dynamic allocation of request slot "
1620                                 "failed! Retrying\n");
1621                 task->tk_status = -ENOMEM;
1622                 break;
1623         case -EAGAIN:
1624                 xprt_add_backlog(xprt, task);
1625                 dprintk("RPC:       waiting for request slot\n");
1626                 fallthrough;
1627         default:
1628                 task->tk_status = -EAGAIN;
1629         }
1630         spin_unlock(&xprt->reserve_lock);
1631         return;
1632 out_init_req:
1633         xprt->stat.max_slots = max_t(unsigned int, xprt->stat.max_slots,
1634                                      xprt->num_reqs);
1635         spin_unlock(&xprt->reserve_lock);
1636
1637         task->tk_status = 0;
1638         task->tk_rqstp = req;
1639 }
1640 EXPORT_SYMBOL_GPL(xprt_alloc_slot);
1641
1642 void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1643 {
1644         spin_lock(&xprt->reserve_lock);
1645         if (!xprt_dynamic_free_slot(xprt, req)) {
1646                 memset(req, 0, sizeof(*req));   /* mark unused */
1647                 list_add(&req->rq_list, &xprt->free);
1648         }
1649         xprt_wake_up_backlog(xprt);
1650         spin_unlock(&xprt->reserve_lock);
1651 }
1652 EXPORT_SYMBOL_GPL(xprt_free_slot);
1653
1654 static void xprt_free_all_slots(struct rpc_xprt *xprt)
1655 {
1656         struct rpc_rqst *req;
1657         while (!list_empty(&xprt->free)) {
1658                 req = list_first_entry(&xprt->free, struct rpc_rqst, rq_list);
1659                 list_del(&req->rq_list);
1660                 kfree(req);
1661         }
1662 }
1663
1664 struct rpc_xprt *xprt_alloc(struct net *net, size_t size,
1665                 unsigned int num_prealloc,
1666                 unsigned int max_alloc)
1667 {
1668         struct rpc_xprt *xprt;
1669         struct rpc_rqst *req;
1670         int i;
1671
1672         xprt = kzalloc(size, GFP_KERNEL);
1673         if (xprt == NULL)
1674                 goto out;
1675
1676         xprt_init(xprt, net);
1677
1678         for (i = 0; i < num_prealloc; i++) {
1679                 req = kzalloc(sizeof(struct rpc_rqst), GFP_KERNEL);
1680                 if (!req)
1681                         goto out_free;
1682                 list_add(&req->rq_list, &xprt->free);
1683         }
1684         if (max_alloc > num_prealloc)
1685                 xprt->max_reqs = max_alloc;
1686         else
1687                 xprt->max_reqs = num_prealloc;
1688         xprt->min_reqs = num_prealloc;
1689         xprt->num_reqs = num_prealloc;
1690
1691         return xprt;
1692
1693 out_free:
1694         xprt_free(xprt);
1695 out:
1696         return NULL;
1697 }
1698 EXPORT_SYMBOL_GPL(xprt_alloc);
1699
1700 void xprt_free(struct rpc_xprt *xprt)
1701 {
1702         put_net(xprt->xprt_net);
1703         xprt_free_all_slots(xprt);
1704         kfree_rcu(xprt, rcu);
1705 }
1706 EXPORT_SYMBOL_GPL(xprt_free);
1707
1708 static void
1709 xprt_init_connect_cookie(struct rpc_rqst *req, struct rpc_xprt *xprt)
1710 {
1711         req->rq_connect_cookie = xprt_connect_cookie(xprt) - 1;
1712 }
1713
1714 static __be32
1715 xprt_alloc_xid(struct rpc_xprt *xprt)
1716 {
1717         __be32 xid;
1718
1719         spin_lock(&xprt->reserve_lock);
1720         xid = (__force __be32)xprt->xid++;
1721         spin_unlock(&xprt->reserve_lock);
1722         return xid;
1723 }
1724
1725 static void
1726 xprt_init_xid(struct rpc_xprt *xprt)
1727 {
1728         xprt->xid = prandom_u32();
1729 }
1730
1731 static void
1732 xprt_request_init(struct rpc_task *task)
1733 {
1734         struct rpc_xprt *xprt = task->tk_xprt;
1735         struct rpc_rqst *req = task->tk_rqstp;
1736
1737         req->rq_task    = task;
1738         req->rq_xprt    = xprt;
1739         req->rq_buffer  = NULL;
1740         req->rq_xid     = xprt_alloc_xid(xprt);
1741         xprt_init_connect_cookie(req, xprt);
1742         req->rq_snd_buf.len = 0;
1743         req->rq_snd_buf.buflen = 0;
1744         req->rq_rcv_buf.len = 0;
1745         req->rq_rcv_buf.buflen = 0;
1746         req->rq_snd_buf.bvec = NULL;
1747         req->rq_rcv_buf.bvec = NULL;
1748         req->rq_release_snd_buf = NULL;
1749         xprt_init_majortimeo(task, req);
1750         dprintk("RPC: %5u reserved req %p xid %08x\n", task->tk_pid,
1751                         req, ntohl(req->rq_xid));
1752 }
1753
1754 static void
1755 xprt_do_reserve(struct rpc_xprt *xprt, struct rpc_task *task)
1756 {
1757         xprt->ops->alloc_slot(xprt, task);
1758         if (task->tk_rqstp != NULL)
1759                 xprt_request_init(task);
1760 }
1761
1762 /**
1763  * xprt_reserve - allocate an RPC request slot
1764  * @task: RPC task requesting a slot allocation
1765  *
1766  * If the transport is marked as being congested, or if no more
1767  * slots are available, place the task on the transport's
1768  * backlog queue.
1769  */
1770 void xprt_reserve(struct rpc_task *task)
1771 {
1772         struct rpc_xprt *xprt = task->tk_xprt;
1773
1774         task->tk_status = 0;
1775         if (task->tk_rqstp != NULL)
1776                 return;
1777
1778         task->tk_status = -EAGAIN;
1779         if (!xprt_throttle_congested(xprt, task))
1780                 xprt_do_reserve(xprt, task);
1781 }
1782
1783 /**
1784  * xprt_retry_reserve - allocate an RPC request slot
1785  * @task: RPC task requesting a slot allocation
1786  *
1787  * If no more slots are available, place the task on the transport's
1788  * backlog queue.
1789  * Note that the only difference with xprt_reserve is that we now
1790  * ignore the value of the XPRT_CONGESTED flag.
1791  */
1792 void xprt_retry_reserve(struct rpc_task *task)
1793 {
1794         struct rpc_xprt *xprt = task->tk_xprt;
1795
1796         task->tk_status = 0;
1797         if (task->tk_rqstp != NULL)
1798                 return;
1799
1800         task->tk_status = -EAGAIN;
1801         xprt_do_reserve(xprt, task);
1802 }
1803
1804 /**
1805  * xprt_release - release an RPC request slot
1806  * @task: task which is finished with the slot
1807  *
1808  */
1809 void xprt_release(struct rpc_task *task)
1810 {
1811         struct rpc_xprt *xprt;
1812         struct rpc_rqst *req = task->tk_rqstp;
1813
1814         if (req == NULL) {
1815                 if (task->tk_client) {
1816                         xprt = task->tk_xprt;
1817                         xprt_release_write(xprt, task);
1818                 }
1819                 return;
1820         }
1821
1822         xprt = req->rq_xprt;
1823         xprt_request_dequeue_xprt(task);
1824         spin_lock(&xprt->transport_lock);
1825         xprt->ops->release_xprt(xprt, task);
1826         if (xprt->ops->release_request)
1827                 xprt->ops->release_request(task);
1828         xprt_schedule_autodisconnect(xprt);
1829         spin_unlock(&xprt->transport_lock);
1830         if (req->rq_buffer)
1831                 xprt->ops->buf_free(task);
1832         xprt_inject_disconnect(xprt);
1833         xdr_free_bvec(&req->rq_rcv_buf);
1834         xdr_free_bvec(&req->rq_snd_buf);
1835         if (req->rq_cred != NULL)
1836                 put_rpccred(req->rq_cred);
1837         task->tk_rqstp = NULL;
1838         if (req->rq_release_snd_buf)
1839                 req->rq_release_snd_buf(req);
1840
1841         dprintk("RPC: %5u release request %p\n", task->tk_pid, req);
1842         if (likely(!bc_prealloc(req)))
1843                 xprt->ops->free_slot(xprt, req);
1844         else
1845                 xprt_free_bc_request(req);
1846 }
1847
1848 #ifdef CONFIG_SUNRPC_BACKCHANNEL
1849 void
1850 xprt_init_bc_request(struct rpc_rqst *req, struct rpc_task *task)
1851 {
1852         struct xdr_buf *xbufp = &req->rq_snd_buf;
1853
1854         task->tk_rqstp = req;
1855         req->rq_task = task;
1856         xprt_init_connect_cookie(req, req->rq_xprt);
1857         /*
1858          * Set up the xdr_buf length.
1859          * This also indicates that the buffer is XDR encoded already.
1860          */
1861         xbufp->len = xbufp->head[0].iov_len + xbufp->page_len +
1862                 xbufp->tail[0].iov_len;
1863 }
1864 #endif
1865
1866 static void xprt_init(struct rpc_xprt *xprt, struct net *net)
1867 {
1868         kref_init(&xprt->kref);
1869
1870         spin_lock_init(&xprt->transport_lock);
1871         spin_lock_init(&xprt->reserve_lock);
1872         spin_lock_init(&xprt->queue_lock);
1873
1874         INIT_LIST_HEAD(&xprt->free);
1875         xprt->recv_queue = RB_ROOT;
1876         INIT_LIST_HEAD(&xprt->xmit_queue);
1877 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1878         spin_lock_init(&xprt->bc_pa_lock);
1879         INIT_LIST_HEAD(&xprt->bc_pa_list);
1880 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1881         INIT_LIST_HEAD(&xprt->xprt_switch);
1882
1883         xprt->last_used = jiffies;
1884         xprt->cwnd = RPC_INITCWND;
1885         xprt->bind_index = 0;
1886
1887         rpc_init_wait_queue(&xprt->binding, "xprt_binding");
1888         rpc_init_wait_queue(&xprt->pending, "xprt_pending");
1889         rpc_init_wait_queue(&xprt->sending, "xprt_sending");
1890         rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
1891
1892         xprt_init_xid(xprt);
1893
1894         xprt->xprt_net = get_net(net);
1895 }
1896
1897 /**
1898  * xprt_create_transport - create an RPC transport
1899  * @args: rpc transport creation arguments
1900  *
1901  */
1902 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
1903 {
1904         struct rpc_xprt *xprt;
1905         struct xprt_class *t;
1906
1907         spin_lock(&xprt_list_lock);
1908         list_for_each_entry(t, &xprt_list, list) {
1909                 if (t->ident == args->ident) {
1910                         spin_unlock(&xprt_list_lock);
1911                         goto found;
1912                 }
1913         }
1914         spin_unlock(&xprt_list_lock);
1915         dprintk("RPC: transport (%d) not supported\n", args->ident);
1916         return ERR_PTR(-EIO);
1917
1918 found:
1919         xprt = t->setup(args);
1920         if (IS_ERR(xprt))
1921                 goto out;
1922         if (args->flags & XPRT_CREATE_NO_IDLE_TIMEOUT)
1923                 xprt->idle_timeout = 0;
1924         INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
1925         if (xprt_has_timer(xprt))
1926                 timer_setup(&xprt->timer, xprt_init_autodisconnect, 0);
1927         else
1928                 timer_setup(&xprt->timer, NULL, 0);
1929
1930         if (strlen(args->servername) > RPC_MAXNETNAMELEN) {
1931                 xprt_destroy(xprt);
1932                 return ERR_PTR(-EINVAL);
1933         }
1934         xprt->servername = kstrdup(args->servername, GFP_KERNEL);
1935         if (xprt->servername == NULL) {
1936                 xprt_destroy(xprt);
1937                 return ERR_PTR(-ENOMEM);
1938         }
1939
1940         rpc_xprt_debugfs_register(xprt);
1941
1942         trace_xprt_create(xprt);
1943 out:
1944         return xprt;
1945 }
1946
1947 static void xprt_destroy_cb(struct work_struct *work)
1948 {
1949         struct rpc_xprt *xprt =
1950                 container_of(work, struct rpc_xprt, task_cleanup);
1951
1952         trace_xprt_destroy(xprt);
1953
1954         rpc_xprt_debugfs_unregister(xprt);
1955         rpc_destroy_wait_queue(&xprt->binding);
1956         rpc_destroy_wait_queue(&xprt->pending);
1957         rpc_destroy_wait_queue(&xprt->sending);
1958         rpc_destroy_wait_queue(&xprt->backlog);
1959         kfree(xprt->servername);
1960         /*
1961          * Destroy any existing back channel
1962          */
1963         xprt_destroy_backchannel(xprt, UINT_MAX);
1964
1965         /*
1966          * Tear down transport state and free the rpc_xprt
1967          */
1968         xprt->ops->destroy(xprt);
1969 }
1970
1971 /**
1972  * xprt_destroy - destroy an RPC transport, killing off all requests.
1973  * @xprt: transport to destroy
1974  *
1975  */
1976 static void xprt_destroy(struct rpc_xprt *xprt)
1977 {
1978         /*
1979          * Exclude transport connect/disconnect handlers and autoclose
1980          */
1981         wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_UNINTERRUPTIBLE);
1982
1983         del_timer_sync(&xprt->timer);
1984
1985         /*
1986          * Destroy sockets etc from the system workqueue so they can
1987          * safely flush receive work running on rpciod.
1988          */
1989         INIT_WORK(&xprt->task_cleanup, xprt_destroy_cb);
1990         schedule_work(&xprt->task_cleanup);
1991 }
1992
1993 static void xprt_destroy_kref(struct kref *kref)
1994 {
1995         xprt_destroy(container_of(kref, struct rpc_xprt, kref));
1996 }
1997
1998 /**
1999  * xprt_get - return a reference to an RPC transport.
2000  * @xprt: pointer to the transport
2001  *
2002  */
2003 struct rpc_xprt *xprt_get(struct rpc_xprt *xprt)
2004 {
2005         if (xprt != NULL && kref_get_unless_zero(&xprt->kref))
2006                 return xprt;
2007         return NULL;
2008 }
2009 EXPORT_SYMBOL_GPL(xprt_get);
2010
2011 /**
2012  * xprt_put - release a reference to an RPC transport.
2013  * @xprt: pointer to the transport
2014  *
2015  */
2016 void xprt_put(struct rpc_xprt *xprt)
2017 {
2018         if (xprt != NULL)
2019                 kref_put(&xprt->kref, xprt_destroy_kref);
2020 }
2021 EXPORT_SYMBOL_GPL(xprt_put);