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 static void      xprt_request_init(struct rpc_task *task);
  74
  75 static DEFINE_SPINLOCK(xprt_list_lock);
  76 static LIST_HEAD(xprt_list);
  77
  78 static unsigned long xprt_request_timeout(const struct rpc_rqst *req)
  79 {
  80         unsigned long timeout = jiffies + req->rq_timeout;
  81
  82         if (time_before(timeout, req->rq_majortimeo))
  83                 return timeout;
  84         return req->rq_majortimeo;
  85 }
  86
  87 /**
  88  * xprt_register_transport - register a transport implementation
  89  * @transport: transport to register
  90  *
  91  * If a transport implementation is loaded as a kernel module, it can
  92  * call this interface to make itself known to the RPC client.
  93  *
  94  * Returns:
  95  * 0:           transport successfully registered
  96  * -EEXIST:     transport already registered
  97  * -EINVAL:     transport module being unloaded
  98  */
  99 int xprt_register_transport(struct xprt_class *transport)
 100 {
 101         struct xprt_class *t;
 102         int result;
 103
 104         result = -EEXIST;
 105         spin_lock(&xprt_list_lock);
 106         list_for_each_entry(t, &xprt_list, list) {
 107                 /* don't register the same transport class twice */
 108                 if (t->ident == transport->ident)
 109                         goto out;
 110         }
 111
 112         list_add_tail(&transport->list, &xprt_list);
 113         printk(KERN_INFO "RPC: Registered %s transport module.\n",
 114                transport->name);
 115         result = 0;
 116
 117 out:
 118         spin_unlock(&xprt_list_lock);
 119         return result;
 120 }
 121 EXPORT_SYMBOL_GPL(xprt_register_transport);
 122
 123 /**
 124  * xprt_unregister_transport - unregister a transport implementation
 125  * @transport: transport to unregister
 126  *
 127  * Returns:
 128  * 0:           transport successfully unregistered
 129  * -ENOENT:     transport never registered
 130  */
 131 int xprt_unregister_transport(struct xprt_class *transport)
 132 {
 133         struct xprt_class *t;
 134         int result;
 135
 136         result = 0;
 137         spin_lock(&xprt_list_lock);
 138         list_for_each_entry(t, &xprt_list, list) {
 139                 if (t == transport) {
 140                         printk(KERN_INFO
 141                                 "RPC: Unregistered %s transport module.\n",
 142                                 transport->name);
 143                         list_del_init(&transport->list);
 144                         goto out;
 145                 }
 146         }
 147         result = -ENOENT;
 148
 149 out:
 150         spin_unlock(&xprt_list_lock);
 151         return result;
 152 }
 153 EXPORT_SYMBOL_GPL(xprt_unregister_transport);
 154
 155 static void
 156 xprt_class_release(const struct xprt_class *t)
 157 {
 158         module_put(t->owner);
 159 }
 160
 161 static const struct xprt_class *
 162 xprt_class_find_by_ident_locked(int ident)
 163 {
 164         const struct xprt_class *t;
 165
 166         list_for_each_entry(t, &xprt_list, list) {
 167                 if (t->ident != ident)
 168                         continue;
 169                 if (!try_module_get(t->owner))
 170                         continue;
 171                 return t;
 172         }
 173         return NULL;
 174 }
 175
 176 static const struct xprt_class *
 177 xprt_class_find_by_ident(int ident)
 178 {
 179         const struct xprt_class *t;
 180
 181         spin_lock(&xprt_list_lock);
 182         t = xprt_class_find_by_ident_locked(ident);
 183         spin_unlock(&xprt_list_lock);
 184         return t;
 185 }
 186
 187 static const struct xprt_class *
 188 xprt_class_find_by_netid_locked(const char *netid)
 189 {
 190         const struct xprt_class *t;
 191         unsigned int i;
 192
 193         list_for_each_entry(t, &xprt_list, list) {
 194                 for (i = 0; t->netid[i][0] != '\0'; i++) {
 195                         if (strcmp(t->netid[i], netid) != 0)
 196                                 continue;
 197                         if (!try_module_get(t->owner))
 198                                 continue;
 199                         return t;
 200                 }
 201         }
 202         return NULL;
 203 }
 204
 205 static const struct xprt_class *
 206 xprt_class_find_by_netid(const char *netid)
 207 {
 208         const struct xprt_class *t;
 209
 210         spin_lock(&xprt_list_lock);
 211         t = xprt_class_find_by_netid_locked(netid);
 212         if (!t) {
 213                 spin_unlock(&xprt_list_lock);
 214                 request_module("rpc%s", netid);
 215                 spin_lock(&xprt_list_lock);
 216                 t = xprt_class_find_by_netid_locked(netid);
 217         }
 218         spin_unlock(&xprt_list_lock);
 219         return t;
 220 }
 221
 222 /**
 223  * xprt_find_transport_ident - convert a netid into a transport identifier
 224  * @netid: transport to load
 225  *
 226  * Returns:
 227  * > 0:         transport identifier
 228  * -ENOENT:     transport module not available
 229  */
 230 int xprt_find_transport_ident(const char *netid)
 231 {
 232         const struct xprt_class *t;
 233         int ret;
 234
 235         t = xprt_class_find_by_netid(netid);
 236         if (!t)
 237                 return -ENOENT;
 238         ret = t->ident;
 239         xprt_class_release(t);
 240         return ret;
 241 }
 242 EXPORT_SYMBOL_GPL(xprt_find_transport_ident);
 243
 244 static void xprt_clear_locked(struct rpc_xprt *xprt)
 245 {
 246         xprt->snd_task = NULL;
 247         if (!test_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
 248                 smp_mb__before_atomic();
 249                 clear_bit(XPRT_LOCKED, &xprt->state);
 250                 smp_mb__after_atomic();
 251         } else
 252                 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 253 }
 254
 255 /**
 256  * xprt_reserve_xprt - serialize write access to transports
 257  * @task: task that is requesting access to the transport
 258  * @xprt: pointer to the target transport
 259  *
 260  * This prevents mixing the payload of separate requests, and prevents
 261  * transport connects from colliding with writes.  No congestion control
 262  * is provided.
 263  */
 264 int xprt_reserve_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 265 {
 266         struct rpc_rqst *req = task->tk_rqstp;
 267
 268         if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
 269                 if (task == xprt->snd_task)
 270                         goto out_locked;
 271                 goto out_sleep;
 272         }
 273         if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
 274                 goto out_unlock;
 275         xprt->snd_task = task;
 276
 277 out_locked:
 278         trace_xprt_reserve_xprt(xprt, task);
 279         return 1;
 280
 281 out_unlock:
 282         xprt_clear_locked(xprt);
 283 out_sleep:
 284         task->tk_status = -EAGAIN;
 285         if  (RPC_IS_SOFT(task))
 286                 rpc_sleep_on_timeout(&xprt->sending, task, NULL,
 287                                 xprt_request_timeout(req));
 288         else
 289                 rpc_sleep_on(&xprt->sending, task, NULL);
 290         return 0;
 291 }
 292 EXPORT_SYMBOL_GPL(xprt_reserve_xprt);
 293
 294 static bool
 295 xprt_need_congestion_window_wait(struct rpc_xprt *xprt)
 296 {
 297         return test_bit(XPRT_CWND_WAIT, &xprt->state);
 298 }
 299
 300 static void
 301 xprt_set_congestion_window_wait(struct rpc_xprt *xprt)
 302 {
 303         if (!list_empty(&xprt->xmit_queue)) {
 304                 /* Peek at head of queue to see if it can make progress */
 305                 if (list_first_entry(&xprt->xmit_queue, struct rpc_rqst,
 306                                         rq_xmit)->rq_cong)
 307                         return;
 308         }
 309         set_bit(XPRT_CWND_WAIT, &xprt->state);
 310 }
 311
 312 static void
 313 xprt_test_and_clear_congestion_window_wait(struct rpc_xprt *xprt)
 314 {
 315         if (!RPCXPRT_CONGESTED(xprt))
 316                 clear_bit(XPRT_CWND_WAIT, &xprt->state);
 317 }
 318
 319 /*
 320  * xprt_reserve_xprt_cong - serialize write access to transports
 321  * @task: task that is requesting access to the transport
 322  *
 323  * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
 324  * integrated into the decision of whether a request is allowed to be
 325  * woken up and given access to the transport.
 326  * Note that the lock is only granted if we know there are free slots.
 327  */
 328 int xprt_reserve_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 329 {
 330         struct rpc_rqst *req = task->tk_rqstp;
 331
 332         if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
 333                 if (task == xprt->snd_task)
 334                         goto out_locked;
 335                 goto out_sleep;
 336         }
 337         if (req == NULL) {
 338                 xprt->snd_task = task;
 339                 goto out_locked;
 340         }
 341         if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
 342                 goto out_unlock;
 343         if (!xprt_need_congestion_window_wait(xprt)) {
 344                 xprt->snd_task = task;
 345                 goto out_locked;
 346         }
 347 out_unlock:
 348         xprt_clear_locked(xprt);
 349 out_sleep:
 350         task->tk_status = -EAGAIN;
 351         if (RPC_IS_SOFT(task))
 352                 rpc_sleep_on_timeout(&xprt->sending, task, NULL,
 353                                 xprt_request_timeout(req));
 354         else
 355                 rpc_sleep_on(&xprt->sending, task, NULL);
 356         return 0;
 357 out_locked:
 358         trace_xprt_reserve_cong(xprt, task);
 359         return 1;
 360 }
 361 EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong);
 362
 363 static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
 364 {
 365         int retval;
 366
 367         if (test_bit(XPRT_LOCKED, &xprt->state) && xprt->snd_task == task)
 368                 return 1;
 369         spin_lock(&xprt->transport_lock);
 370         retval = xprt->ops->reserve_xprt(xprt, task);
 371         spin_unlock(&xprt->transport_lock);
 372         return retval;
 373 }
 374
 375 static bool __xprt_lock_write_func(struct rpc_task *task, void *data)
 376 {
 377         struct rpc_xprt *xprt = data;
 378
 379         xprt->snd_task = task;
 380         return true;
 381 }
 382
 383 static void __xprt_lock_write_next(struct rpc_xprt *xprt)
 384 {
 385         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 386                 return;
 387         if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
 388                 goto out_unlock;
 389         if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
 390                                 __xprt_lock_write_func, xprt))
 391                 return;
 392 out_unlock:
 393         xprt_clear_locked(xprt);
 394 }
 395
 396 static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt)
 397 {
 398         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 399                 return;
 400         if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
 401                 goto out_unlock;
 402         if (xprt_need_congestion_window_wait(xprt))
 403                 goto out_unlock;
 404         if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
 405                                 __xprt_lock_write_func, xprt))
 406                 return;
 407 out_unlock:
 408         xprt_clear_locked(xprt);
 409 }
 410
 411 /**
 412  * xprt_release_xprt - allow other requests to use a transport
 413  * @xprt: transport with other tasks potentially waiting
 414  * @task: task that is releasing access to the transport
 415  *
 416  * Note that "task" can be NULL.  No congestion control is provided.
 417  */
 418 void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 419 {
 420         if (xprt->snd_task == task) {
 421                 xprt_clear_locked(xprt);
 422                 __xprt_lock_write_next(xprt);
 423         }
 424         trace_xprt_release_xprt(xprt, task);
 425 }
 426 EXPORT_SYMBOL_GPL(xprt_release_xprt);
 427
 428 /**
 429  * xprt_release_xprt_cong - allow other requests to use a transport
 430  * @xprt: transport with other tasks potentially waiting
 431  * @task: task that is releasing access to the transport
 432  *
 433  * Note that "task" can be NULL.  Another task is awoken to use the
 434  * transport if the transport's congestion window allows it.
 435  */
 436 void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 437 {
 438         if (xprt->snd_task == task) {
 439                 xprt_clear_locked(xprt);
 440                 __xprt_lock_write_next_cong(xprt);
 441         }
 442         trace_xprt_release_cong(xprt, task);
 443 }
 444 EXPORT_SYMBOL_GPL(xprt_release_xprt_cong);
 445
 446 static inline void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
 447 {
 448         if (xprt->snd_task != task)
 449                 return;
 450         spin_lock(&xprt->transport_lock);
 451         xprt->ops->release_xprt(xprt, task);
 452         spin_unlock(&xprt->transport_lock);
 453 }
 454
 455 /*
 456  * Van Jacobson congestion avoidance. Check if the congestion window
 457  * overflowed. Put the task to sleep if this is the case.
 458  */
 459 static int
 460 __xprt_get_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
 461 {
 462         if (req->rq_cong)
 463                 return 1;
 464         trace_xprt_get_cong(xprt, req->rq_task);
 465         if (RPCXPRT_CONGESTED(xprt)) {
 466                 xprt_set_congestion_window_wait(xprt);
 467                 return 0;
 468         }
 469         req->rq_cong = 1;
 470         xprt->cong += RPC_CWNDSCALE;
 471         return 1;
 472 }
 473
 474 /*
 475  * Adjust the congestion window, and wake up the next task
 476  * that has been sleeping due to congestion
 477  */
 478 static void
 479 __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
 480 {
 481         if (!req->rq_cong)
 482                 return;
 483         req->rq_cong = 0;
 484         xprt->cong -= RPC_CWNDSCALE;
 485         xprt_test_and_clear_congestion_window_wait(xprt);
 486         trace_xprt_put_cong(xprt, req->rq_task);
 487         __xprt_lock_write_next_cong(xprt);
 488 }
 489
 490 /**
 491  * xprt_request_get_cong - Request congestion control credits
 492  * @xprt: pointer to transport
 493  * @req: pointer to RPC request
 494  *
 495  * Useful for transports that require congestion control.
 496  */
 497 bool
 498 xprt_request_get_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
 499 {
 500         bool ret = false;
 501
 502         if (req->rq_cong)
 503                 return true;
 504         spin_lock(&xprt->transport_lock);
 505         ret = __xprt_get_cong(xprt, req) != 0;
 506         spin_unlock(&xprt->transport_lock);
 507         return ret;
 508 }
 509 EXPORT_SYMBOL_GPL(xprt_request_get_cong);
 510
 511 /**
 512  * xprt_release_rqst_cong - housekeeping when request is complete
 513  * @task: RPC request that recently completed
 514  *
 515  * Useful for transports that require congestion control.
 516  */
 517 void xprt_release_rqst_cong(struct rpc_task *task)
 518 {
 519         struct rpc_rqst *req = task->tk_rqstp;
 520
 521         __xprt_put_cong(req->rq_xprt, req);
 522 }
 523 EXPORT_SYMBOL_GPL(xprt_release_rqst_cong);
 524
 525 static void xprt_clear_congestion_window_wait_locked(struct rpc_xprt *xprt)
 526 {
 527         if (test_and_clear_bit(XPRT_CWND_WAIT, &xprt->state))
 528                 __xprt_lock_write_next_cong(xprt);
 529 }
 530
 531 /*
 532  * Clear the congestion window wait flag and wake up the next
 533  * entry on xprt->sending
 534  */
 535 static void
 536 xprt_clear_congestion_window_wait(struct rpc_xprt *xprt)
 537 {
 538         if (test_and_clear_bit(XPRT_CWND_WAIT, &xprt->state)) {
 539                 spin_lock(&xprt->transport_lock);
 540                 __xprt_lock_write_next_cong(xprt);
 541                 spin_unlock(&xprt->transport_lock);
 542         }
 543 }
 544
 545 /**
 546  * xprt_adjust_cwnd - adjust transport congestion window
 547  * @xprt: pointer to xprt
 548  * @task: recently completed RPC request used to adjust window
 549  * @result: result code of completed RPC request
 550  *
 551  * The transport code maintains an estimate on the maximum number of out-
 552  * standing RPC requests, using a smoothed version of the congestion
 553  * avoidance implemented in 44BSD. This is basically the Van Jacobson
 554  * congestion algorithm: If a retransmit occurs, the congestion window is
 555  * halved; otherwise, it is incremented by 1/cwnd when
 556  *
 557  *      -       a reply is received and
 558  *      -       a full number of requests are outstanding and
 559  *      -       the congestion window hasn't been updated recently.
 560  */
 561 void xprt_adjust_cwnd(struct rpc_xprt *xprt, struct rpc_task *task, int result)
 562 {
 563         struct rpc_rqst *req = task->tk_rqstp;
 564         unsigned long cwnd = xprt->cwnd;
 565
 566         if (result >= 0 && cwnd <= xprt->cong) {
 567                 /* The (cwnd >> 1) term makes sure
 568                  * the result gets rounded properly. */
 569                 cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
 570                 if (cwnd > RPC_MAXCWND(xprt))
 571                         cwnd = RPC_MAXCWND(xprt);
 572                 __xprt_lock_write_next_cong(xprt);
 573         } else if (result == -ETIMEDOUT) {
 574                 cwnd >>= 1;
 575                 if (cwnd < RPC_CWNDSCALE)
 576                         cwnd = RPC_CWNDSCALE;
 577         }
 578         dprintk("RPC:       cong %ld, cwnd was %ld, now %ld\n",
 579                         xprt->cong, xprt->cwnd, cwnd);
 580         xprt->cwnd = cwnd;
 581         __xprt_put_cong(xprt, req);
 582 }
 583 EXPORT_SYMBOL_GPL(xprt_adjust_cwnd);
 584
 585 /**
 586  * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
 587  * @xprt: transport with waiting tasks
 588  * @status: result code to plant in each task before waking it
 589  *
 590  */
 591 void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
 592 {
 593         if (status < 0)
 594                 rpc_wake_up_status(&xprt->pending, status);
 595         else
 596                 rpc_wake_up(&xprt->pending);
 597 }
 598 EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks);
 599
 600 /**
 601  * xprt_wait_for_buffer_space - wait for transport output buffer to clear
 602  * @xprt: transport
 603  *
 604  * Note that we only set the timer for the case of RPC_IS_SOFT(), since
 605  * we don't in general want to force a socket disconnection due to
 606  * an incomplete RPC call transmission.
 607  */
 608 void xprt_wait_for_buffer_space(struct rpc_xprt *xprt)
 609 {
 610         set_bit(XPRT_WRITE_SPACE, &xprt->state);
 611 }
 612 EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space);
 613
 614 static bool
 615 xprt_clear_write_space_locked(struct rpc_xprt *xprt)
 616 {
 617         if (test_and_clear_bit(XPRT_WRITE_SPACE, &xprt->state)) {
 618                 __xprt_lock_write_next(xprt);
 619                 dprintk("RPC:       write space: waking waiting task on "
 620                                 "xprt %p\n", xprt);
 621                 return true;
 622         }
 623         return false;
 624 }
 625
 626 /**
 627  * xprt_write_space - wake the task waiting for transport output buffer space
 628  * @xprt: transport with waiting tasks
 629  *
 630  * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
 631  */
 632 bool xprt_write_space(struct rpc_xprt *xprt)
 633 {
 634         bool ret;
 635
 636         if (!test_bit(XPRT_WRITE_SPACE, &xprt->state))
 637                 return false;
 638         spin_lock(&xprt->transport_lock);
 639         ret = xprt_clear_write_space_locked(xprt);
 640         spin_unlock(&xprt->transport_lock);
 641         return ret;
 642 }
 643 EXPORT_SYMBOL_GPL(xprt_write_space);
 644
 645 static unsigned long xprt_abs_ktime_to_jiffies(ktime_t abstime)
 646 {
 647         s64 delta = ktime_to_ns(ktime_get() - abstime);
 648         return likely(delta >= 0) ?
 649                 jiffies - nsecs_to_jiffies(delta) :
 650                 jiffies + nsecs_to_jiffies(-delta);
 651 }
 652
 653 static unsigned long xprt_calc_majortimeo(struct rpc_rqst *req)
 654 {
 655         const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
 656         unsigned long majortimeo = req->rq_timeout;
 657
 658         if (to->to_exponential)
 659                 majortimeo <<= to->to_retries;
 660         else
 661                 majortimeo += to->to_increment * to->to_retries;
 662         if (majortimeo > to->to_maxval || majortimeo == 0)
 663                 majortimeo = to->to_maxval;
 664         return majortimeo;
 665 }
 666
 667 static void xprt_reset_majortimeo(struct rpc_rqst *req)
 668 {
 669         req->rq_majortimeo += xprt_calc_majortimeo(req);
 670 }
 671
 672 static void xprt_reset_minortimeo(struct rpc_rqst *req)
 673 {
 674         req->rq_minortimeo += req->rq_timeout;
 675 }
 676
 677 static void xprt_init_majortimeo(struct rpc_task *task, struct rpc_rqst *req)
 678 {
 679         unsigned long time_init;
 680         struct rpc_xprt *xprt = req->rq_xprt;
 681
 682         if (likely(xprt && xprt_connected(xprt)))
 683                 time_init = jiffies;
 684         else
 685                 time_init = xprt_abs_ktime_to_jiffies(task->tk_start);
 686         req->rq_timeout = task->tk_client->cl_timeout->to_initval;
 687         req->rq_majortimeo = time_init + xprt_calc_majortimeo(req);
 688         req->rq_minortimeo = time_init + req->rq_timeout;
 689 }
 690
 691 /**
 692  * xprt_adjust_timeout - adjust timeout values for next retransmit
 693  * @req: RPC request containing parameters to use for the adjustment
 694  *
 695  */
 696 int xprt_adjust_timeout(struct rpc_rqst *req)
 697 {
 698         struct rpc_xprt *xprt = req->rq_xprt;
 699         const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
 700         int status = 0;
 701
 702         if (time_before(jiffies, req->rq_majortimeo)) {
 703                 if (time_before(jiffies, req->rq_minortimeo))
 704                         return status;
 705                 if (to->to_exponential)
 706                         req->rq_timeout <<= 1;
 707                 else
 708                         req->rq_timeout += to->to_increment;
 709                 if (to->to_maxval && req->rq_timeout >= to->to_maxval)
 710                         req->rq_timeout = to->to_maxval;
 711                 req->rq_retries++;
 712         } else {
 713                 req->rq_timeout = to->to_initval;
 714                 req->rq_retries = 0;
 715                 xprt_reset_majortimeo(req);
 716                 /* Reset the RTT counters == "slow start" */
 717                 spin_lock(&xprt->transport_lock);
 718                 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
 719                 spin_unlock(&xprt->transport_lock);
 720                 status = -ETIMEDOUT;
 721         }
 722         xprt_reset_minortimeo(req);
 723
 724         if (req->rq_timeout == 0) {
 725                 printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
 726                 req->rq_timeout = 5 * HZ;
 727         }
 728         return status;
 729 }
 730
 731 static void xprt_autoclose(struct work_struct *work)
 732 {
 733         struct rpc_xprt *xprt =
 734                 container_of(work, struct rpc_xprt, task_cleanup);
 735         unsigned int pflags = memalloc_nofs_save();
 736
 737         trace_xprt_disconnect_auto(xprt);
 738         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 739         xprt->ops->close(xprt);
 740         xprt_release_write(xprt, NULL);
 741         wake_up_bit(&xprt->state, XPRT_LOCKED);
 742         memalloc_nofs_restore(pflags);
 743 }
 744
 745 /**
 746  * xprt_disconnect_done - mark a transport as disconnected
 747  * @xprt: transport to flag for disconnect
 748  *
 749  */
 750 void xprt_disconnect_done(struct rpc_xprt *xprt)
 751 {
 752         trace_xprt_disconnect_done(xprt);
 753         spin_lock(&xprt->transport_lock);
 754         xprt_clear_connected(xprt);
 755         xprt_clear_write_space_locked(xprt);
 756         xprt_clear_congestion_window_wait_locked(xprt);
 757         xprt_wake_pending_tasks(xprt, -ENOTCONN);
 758         spin_unlock(&xprt->transport_lock);
 759 }
 760 EXPORT_SYMBOL_GPL(xprt_disconnect_done);
 761
 762 /**
 763  * xprt_force_disconnect - force a transport to disconnect
 764  * @xprt: transport to disconnect
 765  *
 766  */
 767 void xprt_force_disconnect(struct rpc_xprt *xprt)
 768 {
 769         trace_xprt_disconnect_force(xprt);
 770
 771         /* Don't race with the test_bit() in xprt_clear_locked() */
 772         spin_lock(&xprt->transport_lock);
 773         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 774         /* Try to schedule an autoclose RPC call */
 775         if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 776                 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 777         else if (xprt->snd_task)
 778                 rpc_wake_up_queued_task_set_status(&xprt->pending,
 779                                 xprt->snd_task, -ENOTCONN);
 780         spin_unlock(&xprt->transport_lock);
 781 }
 782 EXPORT_SYMBOL_GPL(xprt_force_disconnect);
 783
 784 static unsigned int
 785 xprt_connect_cookie(struct rpc_xprt *xprt)
 786 {
 787         return READ_ONCE(xprt->connect_cookie);
 788 }
 789
 790 static bool
 791 xprt_request_retransmit_after_disconnect(struct rpc_task *task)
 792 {
 793         struct rpc_rqst *req = task->tk_rqstp;
 794         struct rpc_xprt *xprt = req->rq_xprt;
 795
 796         return req->rq_connect_cookie != xprt_connect_cookie(xprt) ||
 797                 !xprt_connected(xprt);
 798 }
 799
 800 /**
 801  * xprt_conditional_disconnect - force a transport to disconnect
 802  * @xprt: transport to disconnect
 803  * @cookie: 'connection cookie'
 804  *
 805  * This attempts to break the connection if and only if 'cookie' matches
 806  * the current transport 'connection cookie'. It ensures that we don't
 807  * try to break the connection more than once when we need to retransmit
 808  * a batch of RPC requests.
 809  *
 810  */
 811 void xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie)
 812 {
 813         /* Don't race with the test_bit() in xprt_clear_locked() */
 814         spin_lock(&xprt->transport_lock);
 815         if (cookie != xprt->connect_cookie)
 816                 goto out;
 817         if (test_bit(XPRT_CLOSING, &xprt->state))
 818                 goto out;
 819         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 820         /* Try to schedule an autoclose RPC call */
 821         if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 822                 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 823         xprt_wake_pending_tasks(xprt, -EAGAIN);
 824 out:
 825         spin_unlock(&xprt->transport_lock);
 826 }
 827
 828 static bool
 829 xprt_has_timer(const struct rpc_xprt *xprt)
 830 {
 831         return xprt->idle_timeout != 0;
 832 }
 833
 834 static void
 835 xprt_schedule_autodisconnect(struct rpc_xprt *xprt)
 836         __must_hold(&xprt->transport_lock)
 837 {
 838         xprt->last_used = jiffies;
 839         if (RB_EMPTY_ROOT(&xprt->recv_queue) && xprt_has_timer(xprt))
 840                 mod_timer(&xprt->timer, xprt->last_used + xprt->idle_timeout);
 841 }
 842
 843 static void
 844 xprt_init_autodisconnect(struct timer_list *t)
 845 {
 846         struct rpc_xprt *xprt = from_timer(xprt, t, timer);
 847
 848         if (!RB_EMPTY_ROOT(&xprt->recv_queue))
 849                 return;
 850         /* Reset xprt->last_used to avoid connect/autodisconnect cycling */
 851         xprt->last_used = jiffies;
 852         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 853                 return;
 854         queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 855 }
 856
 857 bool xprt_lock_connect(struct rpc_xprt *xprt,
 858                 struct rpc_task *task,
 859                 void *cookie)
 860 {
 861         bool ret = false;
 862
 863         spin_lock(&xprt->transport_lock);
 864         if (!test_bit(XPRT_LOCKED, &xprt->state))
 865                 goto out;
 866         if (xprt->snd_task != task)
 867                 goto out;
 868         xprt->snd_task = cookie;
 869         ret = true;
 870 out:
 871         spin_unlock(&xprt->transport_lock);
 872         return ret;
 873 }
 874
 875 void xprt_unlock_connect(struct rpc_xprt *xprt, void *cookie)
 876 {
 877         spin_lock(&xprt->transport_lock);
 878         if (xprt->snd_task != cookie)
 879                 goto out;
 880         if (!test_bit(XPRT_LOCKED, &xprt->state))
 881                 goto out;
 882         xprt->snd_task =NULL;
 883         xprt->ops->release_xprt(xprt, NULL);
 884         xprt_schedule_autodisconnect(xprt);
 885 out:
 886         spin_unlock(&xprt->transport_lock);
 887         wake_up_bit(&xprt->state, XPRT_LOCKED);
 888 }
 889
 890 /**
 891  * xprt_connect - schedule a transport connect operation
 892  * @task: RPC task that is requesting the connect
 893  *
 894  */
 895 void xprt_connect(struct rpc_task *task)
 896 {
 897         struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
 898
 899         trace_xprt_connect(xprt);
 900
 901         if (!xprt_bound(xprt)) {
 902                 task->tk_status = -EAGAIN;
 903                 return;
 904         }
 905         if (!xprt_lock_write(xprt, task))
 906                 return;
 907
 908         if (test_and_clear_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
 909                 trace_xprt_disconnect_cleanup(xprt);
 910                 xprt->ops->close(xprt);
 911         }
 912
 913         if (!xprt_connected(xprt)) {
 914                 task->tk_rqstp->rq_connect_cookie = xprt->connect_cookie;
 915                 rpc_sleep_on_timeout(&xprt->pending, task, NULL,
 916                                 xprt_request_timeout(task->tk_rqstp));
 917
 918                 if (test_bit(XPRT_CLOSING, &xprt->state))
 919                         return;
 920                 if (xprt_test_and_set_connecting(xprt))
 921                         return;
 922                 /* Race breaker */
 923                 if (!xprt_connected(xprt)) {
 924                         xprt->stat.connect_start = jiffies;
 925                         xprt->ops->connect(xprt, task);
 926                 } else {
 927                         xprt_clear_connecting(xprt);
 928                         task->tk_status = 0;
 929                         rpc_wake_up_queued_task(&xprt->pending, task);
 930                 }
 931         }
 932         xprt_release_write(xprt, task);
 933 }
 934
 935 /**
 936  * xprt_reconnect_delay - compute the wait before scheduling a connect
 937  * @xprt: transport instance
 938  *
 939  */
 940 unsigned long xprt_reconnect_delay(const struct rpc_xprt *xprt)
 941 {
 942         unsigned long start, now = jiffies;
 943
 944         start = xprt->stat.connect_start + xprt->reestablish_timeout;
 945         if (time_after(start, now))
 946                 return start - now;
 947         return 0;
 948 }
 949 EXPORT_SYMBOL_GPL(xprt_reconnect_delay);
 950
 951 /**
 952  * xprt_reconnect_backoff - compute the new re-establish timeout
 953  * @xprt: transport instance
 954  * @init_to: initial reestablish timeout
 955  *
 956  */
 957 void xprt_reconnect_backoff(struct rpc_xprt *xprt, unsigned long init_to)
 958 {
 959         xprt->reestablish_timeout <<= 1;
 960         if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
 961                 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
 962         if (xprt->reestablish_timeout < init_to)
 963                 xprt->reestablish_timeout = init_to;
 964 }
 965 EXPORT_SYMBOL_GPL(xprt_reconnect_backoff);
 966
 967 enum xprt_xid_rb_cmp {
 968         XID_RB_EQUAL,
 969         XID_RB_LEFT,
 970         XID_RB_RIGHT,
 971 };
 972 static enum xprt_xid_rb_cmp
 973 xprt_xid_cmp(__be32 xid1, __be32 xid2)
 974 {
 975         if (xid1 == xid2)
 976                 return XID_RB_EQUAL;
 977         if ((__force u32)xid1 < (__force u32)xid2)
 978                 return XID_RB_LEFT;
 979         return XID_RB_RIGHT;
 980 }
 981
 982 static struct rpc_rqst *
 983 xprt_request_rb_find(struct rpc_xprt *xprt, __be32 xid)
 984 {
 985         struct rb_node *n = xprt->recv_queue.rb_node;
 986         struct rpc_rqst *req;
 987
 988         while (n != NULL) {
 989                 req = rb_entry(n, struct rpc_rqst, rq_recv);
 990                 switch (xprt_xid_cmp(xid, req->rq_xid)) {
 991                 case XID_RB_LEFT:
 992                         n = n->rb_left;
 993                         break;
 994                 case XID_RB_RIGHT:
 995                         n = n->rb_right;
 996                         break;
 997                 case XID_RB_EQUAL:
 998                         return req;
 999                 }
1000         }
1001         return NULL;
1002 }
1003
1004 static void
1005 xprt_request_rb_insert(struct rpc_xprt *xprt, struct rpc_rqst *new)
1006 {
1007         struct rb_node **p = &xprt->recv_queue.rb_node;
1008         struct rb_node *n = NULL;
1009         struct rpc_rqst *req;
1010
1011         while (*p != NULL) {
1012                 n = *p;
1013                 req = rb_entry(n, struct rpc_rqst, rq_recv);
1014                 switch(xprt_xid_cmp(new->rq_xid, req->rq_xid)) {
1015                 case XID_RB_LEFT:
1016                         p = &n->rb_left;
1017                         break;
1018                 case XID_RB_RIGHT:
1019                         p = &n->rb_right;
1020                         break;
1021                 case XID_RB_EQUAL:
1022                         WARN_ON_ONCE(new != req);
1023                         return;
1024                 }
1025         }
1026         rb_link_node(&new->rq_recv, n, p);
1027         rb_insert_color(&new->rq_recv, &xprt->recv_queue);
1028 }
1029
1030 static void
1031 xprt_request_rb_remove(struct rpc_xprt *xprt, struct rpc_rqst *req)
1032 {
1033         rb_erase(&req->rq_recv, &xprt->recv_queue);
1034 }
1035
1036 /**
1037  * xprt_lookup_rqst - find an RPC request corresponding to an XID
1038  * @xprt: transport on which the original request was transmitted
1039  * @xid: RPC XID of incoming reply
1040  *
1041  * Caller holds xprt->queue_lock.
1042  */
1043 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
1044 {
1045         struct rpc_rqst *entry;
1046
1047         entry = xprt_request_rb_find(xprt, xid);
1048         if (entry != NULL) {
1049                 trace_xprt_lookup_rqst(xprt, xid, 0);
1050                 entry->rq_rtt = ktime_sub(ktime_get(), entry->rq_xtime);
1051                 return entry;
1052         }
1053
1054         dprintk("RPC:       xprt_lookup_rqst did not find xid %08x\n",
1055                         ntohl(xid));
1056         trace_xprt_lookup_rqst(xprt, xid, -ENOENT);
1057         xprt->stat.bad_xids++;
1058         return NULL;
1059 }
1060 EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
1061
1062 static bool
1063 xprt_is_pinned_rqst(struct rpc_rqst *req)
1064 {
1065         return atomic_read(&req->rq_pin) != 0;
1066 }
1067
1068 /**
1069  * xprt_pin_rqst - Pin a request on the transport receive list
1070  * @req: Request to pin
1071  *
1072  * Caller must ensure this is atomic with the call to xprt_lookup_rqst()
1073  * so should be holding xprt->queue_lock.
1074  */
1075 void xprt_pin_rqst(struct rpc_rqst *req)
1076 {
1077         atomic_inc(&req->rq_pin);
1078 }
1079 EXPORT_SYMBOL_GPL(xprt_pin_rqst);
1080
1081 /**
1082  * xprt_unpin_rqst - Unpin a request on the transport receive list
1083  * @req: Request to pin
1084  *
1085  * Caller should be holding xprt->queue_lock.
1086  */
1087 void xprt_unpin_rqst(struct rpc_rqst *req)
1088 {
1089         if (!test_bit(RPC_TASK_MSG_PIN_WAIT, &req->rq_task->tk_runstate)) {
1090                 atomic_dec(&req->rq_pin);
1091                 return;
1092         }
1093         if (atomic_dec_and_test(&req->rq_pin))
1094                 wake_up_var(&req->rq_pin);
1095 }
1096 EXPORT_SYMBOL_GPL(xprt_unpin_rqst);
1097
1098 static void xprt_wait_on_pinned_rqst(struct rpc_rqst *req)
1099 {
1100         wait_var_event(&req->rq_pin, !xprt_is_pinned_rqst(req));
1101 }
1102
1103 static bool
1104 xprt_request_data_received(struct rpc_task *task)
1105 {
1106         return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
1107                 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) != 0;
1108 }
1109
1110 static bool
1111 xprt_request_need_enqueue_receive(struct rpc_task *task, struct rpc_rqst *req)
1112 {
1113         return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
1114                 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) == 0;
1115 }
1116
1117 /**
1118  * xprt_request_enqueue_receive - Add an request to the receive queue
1119  * @task: RPC task
1120  *
1121  */
1122 void
1123 xprt_request_enqueue_receive(struct rpc_task *task)
1124 {
1125         struct rpc_rqst *req = task->tk_rqstp;
1126         struct rpc_xprt *xprt = req->rq_xprt;
1127
1128         if (!xprt_request_need_enqueue_receive(task, req))
1129                 return;
1130
1131         xprt_request_prepare(task->tk_rqstp);
1132         spin_lock(&xprt->queue_lock);
1133
1134         /* Update the softirq receive buffer */
1135         memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
1136                         sizeof(req->rq_private_buf));
1137
1138         /* Add request to the receive list */
1139         xprt_request_rb_insert(xprt, req);
1140         set_bit(RPC_TASK_NEED_RECV, &task->tk_runstate);
1141         spin_unlock(&xprt->queue_lock);
1142
1143         /* Turn off autodisconnect */
1144         del_singleshot_timer_sync(&xprt->timer);
1145 }
1146
1147 /**
1148  * xprt_request_dequeue_receive_locked - Remove a request from the receive queue
1149  * @task: RPC task
1150  *
1151  * Caller must hold xprt->queue_lock.
1152  */
1153 static void
1154 xprt_request_dequeue_receive_locked(struct rpc_task *task)
1155 {
1156         struct rpc_rqst *req = task->tk_rqstp;
1157
1158         if (test_and_clear_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1159                 xprt_request_rb_remove(req->rq_xprt, req);
1160 }
1161
1162 /**
1163  * xprt_update_rtt - Update RPC RTT statistics
1164  * @task: RPC request that recently completed
1165  *
1166  * Caller holds xprt->queue_lock.
1167  */
1168 void xprt_update_rtt(struct rpc_task *task)
1169 {
1170         struct rpc_rqst *req = task->tk_rqstp;
1171         struct rpc_rtt *rtt = task->tk_client->cl_rtt;
1172         unsigned int timer = task->tk_msg.rpc_proc->p_timer;
1173         long m = usecs_to_jiffies(ktime_to_us(req->rq_rtt));
1174
1175         if (timer) {
1176                 if (req->rq_ntrans == 1)
1177                         rpc_update_rtt(rtt, timer, m);
1178                 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
1179         }
1180 }
1181 EXPORT_SYMBOL_GPL(xprt_update_rtt);
1182
1183 /**
1184  * xprt_complete_rqst - called when reply processing is complete
1185  * @task: RPC request that recently completed
1186  * @copied: actual number of bytes received from the transport
1187  *
1188  * Caller holds xprt->queue_lock.
1189  */
1190 void xprt_complete_rqst(struct rpc_task *task, int copied)
1191 {
1192         struct rpc_rqst *req = task->tk_rqstp;
1193         struct rpc_xprt *xprt = req->rq_xprt;
1194
1195         xprt->stat.recvs++;
1196
1197         req->rq_private_buf.len = copied;
1198         /* Ensure all writes are done before we update */
1199         /* req->rq_reply_bytes_recvd */
1200         smp_wmb();
1201         req->rq_reply_bytes_recvd = copied;
1202         xprt_request_dequeue_receive_locked(task);
1203         rpc_wake_up_queued_task(&xprt->pending, task);
1204 }
1205 EXPORT_SYMBOL_GPL(xprt_complete_rqst);
1206
1207 static void xprt_timer(struct rpc_task *task)
1208 {
1209         struct rpc_rqst *req = task->tk_rqstp;
1210         struct rpc_xprt *xprt = req->rq_xprt;
1211
1212         if (task->tk_status != -ETIMEDOUT)
1213                 return;
1214
1215         trace_xprt_timer(xprt, req->rq_xid, task->tk_status);
1216         if (!req->rq_reply_bytes_recvd) {
1217                 if (xprt->ops->timer)
1218                         xprt->ops->timer(xprt, task);
1219         } else
1220                 task->tk_status = 0;
1221 }
1222
1223 /**
1224  * xprt_wait_for_reply_request_def - wait for reply
1225  * @task: pointer to rpc_task
1226  *
1227  * Set a request's retransmit timeout based on the transport's
1228  * default timeout parameters.  Used by transports that don't adjust
1229  * the retransmit timeout based on round-trip time estimation,
1230  * and put the task to sleep on the pending queue.
1231  */
1232 void xprt_wait_for_reply_request_def(struct rpc_task *task)
1233 {
1234         struct rpc_rqst *req = task->tk_rqstp;
1235
1236         rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1237                         xprt_request_timeout(req));
1238 }
1239 EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_def);
1240
1241 /**
1242  * xprt_wait_for_reply_request_rtt - wait for reply using RTT estimator
1243  * @task: pointer to rpc_task
1244  *
1245  * Set a request's retransmit timeout using the RTT estimator,
1246  * and put the task to sleep on the pending queue.
1247  */
1248 void xprt_wait_for_reply_request_rtt(struct rpc_task *task)
1249 {
1250         int timer = task->tk_msg.rpc_proc->p_timer;
1251         struct rpc_clnt *clnt = task->tk_client;
1252         struct rpc_rtt *rtt = clnt->cl_rtt;
1253         struct rpc_rqst *req = task->tk_rqstp;
1254         unsigned long max_timeout = clnt->cl_timeout->to_maxval;
1255         unsigned long timeout;
1256
1257         timeout = rpc_calc_rto(rtt, timer);
1258         timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
1259         if (timeout > max_timeout || timeout == 0)
1260                 timeout = max_timeout;
1261         rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1262                         jiffies + timeout);
1263 }
1264 EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_rtt);
1265
1266 /**
1267  * xprt_request_wait_receive - wait for the reply to an RPC request
1268  * @task: RPC task about to send a request
1269  *
1270  */
1271 void xprt_request_wait_receive(struct rpc_task *task)
1272 {
1273         struct rpc_rqst *req = task->tk_rqstp;
1274         struct rpc_xprt *xprt = req->rq_xprt;
1275
1276         if (!test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1277                 return;
1278         /*
1279          * Sleep on the pending queue if we're expecting a reply.
1280          * The spinlock ensures atomicity between the test of
1281          * req->rq_reply_bytes_recvd, and the call to rpc_sleep_on().
1282          */
1283         spin_lock(&xprt->queue_lock);
1284         if (test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate)) {
1285                 xprt->ops->wait_for_reply_request(task);
1286                 /*
1287                  * Send an extra queue wakeup call if the
1288                  * connection was dropped in case the call to
1289                  * rpc_sleep_on() raced.
1290                  */
1291                 if (xprt_request_retransmit_after_disconnect(task))
1292                         rpc_wake_up_queued_task_set_status(&xprt->pending,
1293                                         task, -ENOTCONN);
1294         }
1295         spin_unlock(&xprt->queue_lock);
1296 }
1297
1298 static bool
1299 xprt_request_need_enqueue_transmit(struct rpc_task *task, struct rpc_rqst *req)
1300 {
1301         return !test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1302 }
1303
1304 /**
1305  * xprt_request_enqueue_transmit - queue a task for transmission
1306  * @task: pointer to rpc_task
1307  *
1308  * Add a task to the transmission queue.
1309  */
1310 void
1311 xprt_request_enqueue_transmit(struct rpc_task *task)
1312 {
1313         struct rpc_rqst *pos, *req = task->tk_rqstp;
1314         struct rpc_xprt *xprt = req->rq_xprt;
1315
1316         if (xprt_request_need_enqueue_transmit(task, req)) {
1317                 req->rq_bytes_sent = 0;
1318                 spin_lock(&xprt->queue_lock);
1319                 /*
1320                  * Requests that carry congestion control credits are added
1321                  * to the head of the list to avoid starvation issues.
1322                  */
1323                 if (req->rq_cong) {
1324                         xprt_clear_congestion_window_wait(xprt);
1325                         list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1326                                 if (pos->rq_cong)
1327                                         continue;
1328                                 /* Note: req is added _before_ pos */
1329                                 list_add_tail(&req->rq_xmit, &pos->rq_xmit);
1330                                 INIT_LIST_HEAD(&req->rq_xmit2);
1331                                 goto out;
1332                         }
1333                 } else if (RPC_IS_SWAPPER(task)) {
1334                         list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1335                                 if (pos->rq_cong || pos->rq_bytes_sent)
1336                                         continue;
1337                                 if (RPC_IS_SWAPPER(pos->rq_task))
1338                                         continue;
1339                                 /* Note: req is added _before_ pos */
1340                                 list_add_tail(&req->rq_xmit, &pos->rq_xmit);
1341                                 INIT_LIST_HEAD(&req->rq_xmit2);
1342                                 goto out;
1343                         }
1344                 } else if (!req->rq_seqno) {
1345                         list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1346                                 if (pos->rq_task->tk_owner != task->tk_owner)
1347                                         continue;
1348                                 list_add_tail(&req->rq_xmit2, &pos->rq_xmit2);
1349                                 INIT_LIST_HEAD(&req->rq_xmit);
1350                                 goto out;
1351                         }
1352                 }
1353                 list_add_tail(&req->rq_xmit, &xprt->xmit_queue);
1354                 INIT_LIST_HEAD(&req->rq_xmit2);
1355 out:
1356                 atomic_long_inc(&xprt->xmit_queuelen);
1357                 set_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1358                 spin_unlock(&xprt->queue_lock);
1359         }
1360 }
1361
1362 /**
1363  * xprt_request_dequeue_transmit_locked - remove a task from the transmission queue
1364  * @task: pointer to rpc_task
1365  *
1366  * Remove a task from the transmission queue
1367  * Caller must hold xprt->queue_lock
1368  */
1369 static void
1370 xprt_request_dequeue_transmit_locked(struct rpc_task *task)
1371 {
1372         struct rpc_rqst *req = task->tk_rqstp;
1373
1374         if (!test_and_clear_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1375                 return;
1376         if (!list_empty(&req->rq_xmit)) {
1377                 list_del(&req->rq_xmit);
1378                 if (!list_empty(&req->rq_xmit2)) {
1379                         struct rpc_rqst *next = list_first_entry(&req->rq_xmit2,
1380                                         struct rpc_rqst, rq_xmit2);
1381                         list_del(&req->rq_xmit2);
1382                         list_add_tail(&next->rq_xmit, &next->rq_xprt->xmit_queue);
1383                 }
1384         } else
1385                 list_del(&req->rq_xmit2);
1386         atomic_long_dec(&req->rq_xprt->xmit_queuelen);
1387 }
1388
1389 /**
1390  * xprt_request_dequeue_transmit - remove a task from the transmission queue
1391  * @task: pointer to rpc_task
1392  *
1393  * Remove a task from the transmission queue
1394  */
1395 static void
1396 xprt_request_dequeue_transmit(struct rpc_task *task)
1397 {
1398         struct rpc_rqst *req = task->tk_rqstp;
1399         struct rpc_xprt *xprt = req->rq_xprt;
1400
1401         spin_lock(&xprt->queue_lock);
1402         xprt_request_dequeue_transmit_locked(task);
1403         spin_unlock(&xprt->queue_lock);
1404 }
1405
1406 /**
1407  * xprt_request_dequeue_xprt - remove a task from the transmit+receive queue
1408  * @task: pointer to rpc_task
1409  *
1410  * Remove a task from the transmit and receive queues, and ensure that
1411  * it is not pinned by the receive work item.
1412  */
1413 void
1414 xprt_request_dequeue_xprt(struct rpc_task *task)
1415 {
1416         struct rpc_rqst *req = task->tk_rqstp;
1417         struct rpc_xprt *xprt = req->rq_xprt;
1418
1419         if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate) ||
1420             test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) ||
1421             xprt_is_pinned_rqst(req)) {
1422                 spin_lock(&xprt->queue_lock);
1423                 xprt_request_dequeue_transmit_locked(task);
1424                 xprt_request_dequeue_receive_locked(task);
1425                 while (xprt_is_pinned_rqst(req)) {
1426                         set_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1427                         spin_unlock(&xprt->queue_lock);
1428                         xprt_wait_on_pinned_rqst(req);
1429                         spin_lock(&xprt->queue_lock);
1430                         clear_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1431                 }
1432                 spin_unlock(&xprt->queue_lock);
1433         }
1434 }
1435
1436 /**
1437  * xprt_request_prepare - prepare an encoded request for transport
1438  * @req: pointer to rpc_rqst
1439  *
1440  * Calls into the transport layer to do whatever is needed to prepare
1441  * the request for transmission or receive.
1442  */
1443 void
1444 xprt_request_prepare(struct rpc_rqst *req)
1445 {
1446         struct rpc_xprt *xprt = req->rq_xprt;
1447
1448         if (xprt->ops->prepare_request)
1449                 xprt->ops->prepare_request(req);
1450 }
1451
1452 /**
1453  * xprt_request_need_retransmit - Test if a task needs retransmission
1454  * @task: pointer to rpc_task
1455  *
1456  * Test for whether a connection breakage requires the task to retransmit
1457  */
1458 bool
1459 xprt_request_need_retransmit(struct rpc_task *task)
1460 {
1461         return xprt_request_retransmit_after_disconnect(task);
1462 }
1463
1464 /**
1465  * xprt_prepare_transmit - reserve the transport before sending a request
1466  * @task: RPC task about to send a request
1467  *
1468  */
1469 bool xprt_prepare_transmit(struct rpc_task *task)
1470 {
1471         struct rpc_rqst *req = task->tk_rqstp;
1472         struct rpc_xprt *xprt = req->rq_xprt;
1473
1474         if (!xprt_lock_write(xprt, task)) {
1475                 /* Race breaker: someone may have transmitted us */
1476                 if (!test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1477                         rpc_wake_up_queued_task_set_status(&xprt->sending,
1478                                         task, 0);
1479                 return false;
1480
1481         }
1482         return true;
1483 }
1484
1485 void xprt_end_transmit(struct rpc_task *task)
1486 {
1487         struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
1488
1489         xprt_inject_disconnect(xprt);
1490         xprt_release_write(xprt, task);
1491 }
1492
1493 /**
1494  * xprt_request_transmit - send an RPC request on a transport
1495  * @req: pointer to request to transmit
1496  * @snd_task: RPC task that owns the transport lock
1497  *
1498  * This performs the transmission of a single request.
1499  * Note that if the request is not the same as snd_task, then it
1500  * does need to be pinned.
1501  * Returns '0' on success.
1502  */
1503 static int
1504 xprt_request_transmit(struct rpc_rqst *req, struct rpc_task *snd_task)
1505 {
1506         struct rpc_xprt *xprt = req->rq_xprt;
1507         struct rpc_task *task = req->rq_task;
1508         unsigned int connect_cookie;
1509         int is_retrans = RPC_WAS_SENT(task);
1510         int status;
1511
1512         if (!req->rq_bytes_sent) {
1513                 if (xprt_request_data_received(task)) {
1514                         status = 0;
1515                         goto out_dequeue;
1516                 }
1517                 /* Verify that our message lies in the RPCSEC_GSS window */
1518                 if (rpcauth_xmit_need_reencode(task)) {
1519                         status = -EBADMSG;
1520                         goto out_dequeue;
1521                 }
1522                 if (RPC_SIGNALLED(task)) {
1523                         status = -ERESTARTSYS;
1524                         goto out_dequeue;
1525                 }
1526         }
1527
1528         /*
1529          * Update req->rq_ntrans before transmitting to avoid races with
1530          * xprt_update_rtt(), which needs to know that it is recording a
1531          * reply to the first transmission.
1532          */
1533         req->rq_ntrans++;
1534
1535         trace_rpc_xdr_sendto(task, &req->rq_snd_buf);
1536         connect_cookie = xprt->connect_cookie;
1537         status = xprt->ops->send_request(req);
1538         if (status != 0) {
1539                 req->rq_ntrans--;
1540                 trace_xprt_transmit(req, status);
1541                 return status;
1542         }
1543
1544         if (is_retrans) {
1545                 task->tk_client->cl_stats->rpcretrans++;
1546                 trace_xprt_retransmit(req);
1547         }
1548
1549         xprt_inject_disconnect(xprt);
1550
1551         task->tk_flags |= RPC_TASK_SENT;
1552         spin_lock(&xprt->transport_lock);
1553
1554         xprt->stat.sends++;
1555         xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
1556         xprt->stat.bklog_u += xprt->backlog.qlen;
1557         xprt->stat.sending_u += xprt->sending.qlen;
1558         xprt->stat.pending_u += xprt->pending.qlen;
1559         spin_unlock(&xprt->transport_lock);
1560
1561         req->rq_connect_cookie = connect_cookie;
1562 out_dequeue:
1563         trace_xprt_transmit(req, status);
1564         xprt_request_dequeue_transmit(task);
1565         rpc_wake_up_queued_task_set_status(&xprt->sending, task, status);
1566         return status;
1567 }
1568
1569 /**
1570  * xprt_transmit - send an RPC request on a transport
1571  * @task: controlling RPC task
1572  *
1573  * Attempts to drain the transmit queue. On exit, either the transport
1574  * signalled an error that needs to be handled before transmission can
1575  * resume, or @task finished transmitting, and detected that it already
1576  * received a reply.
1577  */
1578 void
1579 xprt_transmit(struct rpc_task *task)
1580 {
1581         struct rpc_rqst *next, *req = task->tk_rqstp;
1582         struct rpc_xprt *xprt = req->rq_xprt;
1583         int counter, status;
1584
1585         spin_lock(&xprt->queue_lock);
1586         counter = 0;
1587         while (!list_empty(&xprt->xmit_queue)) {
1588                 if (++counter == 20)
1589                         break;
1590                 next = list_first_entry(&xprt->xmit_queue,
1591                                 struct rpc_rqst, rq_xmit);
1592                 xprt_pin_rqst(next);
1593                 spin_unlock(&xprt->queue_lock);
1594                 status = xprt_request_transmit(next, task);
1595                 if (status == -EBADMSG && next != req)
1596                         status = 0;
1597                 spin_lock(&xprt->queue_lock);
1598                 xprt_unpin_rqst(next);
1599                 if (status == 0) {
1600                         if (!xprt_request_data_received(task) ||
1601                             test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1602                                 continue;
1603                 } else if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1604                         task->tk_status = status;
1605                 break;
1606         }
1607         spin_unlock(&xprt->queue_lock);
1608 }
1609
1610 static void xprt_complete_request_init(struct rpc_task *task)
1611 {
1612         if (task->tk_rqstp)
1613                 xprt_request_init(task);
1614 }
1615
1616 void xprt_add_backlog(struct rpc_xprt *xprt, struct rpc_task *task)
1617 {
1618         set_bit(XPRT_CONGESTED, &xprt->state);
1619         rpc_sleep_on(&xprt->backlog, task, xprt_complete_request_init);
1620 }
1621 EXPORT_SYMBOL_GPL(xprt_add_backlog);
1622
1623 static bool __xprt_set_rq(struct rpc_task *task, void *data)
1624 {
1625         struct rpc_rqst *req = data;
1626
1627         if (task->tk_rqstp == NULL) {
1628                 memset(req, 0, sizeof(*req));   /* mark unused */
1629                 task->tk_rqstp = req;
1630                 return true;
1631         }
1632         return false;
1633 }
1634
1635 bool xprt_wake_up_backlog(struct rpc_xprt *xprt, struct rpc_rqst *req)
1636 {
1637         if (rpc_wake_up_first(&xprt->backlog, __xprt_set_rq, req) == NULL) {
1638                 clear_bit(XPRT_CONGESTED, &xprt->state);
1639                 return false;
1640         }
1641         return true;
1642 }
1643 EXPORT_SYMBOL_GPL(xprt_wake_up_backlog);
1644
1645 static bool xprt_throttle_congested(struct rpc_xprt *xprt, struct rpc_task *task)
1646 {
1647         bool ret = false;
1648
1649         if (!test_bit(XPRT_CONGESTED, &xprt->state))
1650                 goto out;
1651         spin_lock(&xprt->reserve_lock);
1652         if (test_bit(XPRT_CONGESTED, &xprt->state)) {
1653                 xprt_add_backlog(xprt, task);
1654                 ret = true;
1655         }
1656         spin_unlock(&xprt->reserve_lock);
1657 out:
1658         return ret;
1659 }
1660
1661 static struct rpc_rqst *xprt_dynamic_alloc_slot(struct rpc_xprt *xprt)
1662 {
1663         struct rpc_rqst *req = ERR_PTR(-EAGAIN);
1664
1665         if (xprt->num_reqs >= xprt->max_reqs)
1666                 goto out;
1667         ++xprt->num_reqs;
1668         spin_unlock(&xprt->reserve_lock);
1669         req = kzalloc(sizeof(struct rpc_rqst), GFP_NOFS);
1670         spin_lock(&xprt->reserve_lock);
1671         if (req != NULL)
1672                 goto out;
1673         --xprt->num_reqs;
1674         req = ERR_PTR(-ENOMEM);
1675 out:
1676         return req;
1677 }
1678
1679 static bool xprt_dynamic_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1680 {
1681         if (xprt->num_reqs > xprt->min_reqs) {
1682                 --xprt->num_reqs;
1683                 kfree(req);
1684                 return true;
1685         }
1686         return false;
1687 }
1688
1689 void xprt_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1690 {
1691         struct rpc_rqst *req;
1692
1693         spin_lock(&xprt->reserve_lock);
1694         if (!list_empty(&xprt->free)) {
1695                 req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
1696                 list_del(&req->rq_list);
1697                 goto out_init_req;
1698         }
1699         req = xprt_dynamic_alloc_slot(xprt);
1700         if (!IS_ERR(req))
1701                 goto out_init_req;
1702         switch (PTR_ERR(req)) {
1703         case -ENOMEM:
1704                 dprintk("RPC:       dynamic allocation of request slot "
1705                                 "failed! Retrying\n");
1706                 task->tk_status = -ENOMEM;
1707                 break;
1708         case -EAGAIN:
1709                 xprt_add_backlog(xprt, task);
1710                 dprintk("RPC:       waiting for request slot\n");
1711                 fallthrough;
1712         default:
1713                 task->tk_status = -EAGAIN;
1714         }
1715         spin_unlock(&xprt->reserve_lock);
1716         return;
1717 out_init_req:
1718         xprt->stat.max_slots = max_t(unsigned int, xprt->stat.max_slots,
1719                                      xprt->num_reqs);
1720         spin_unlock(&xprt->reserve_lock);
1721
1722         task->tk_status = 0;
1723         task->tk_rqstp = req;
1724 }
1725 EXPORT_SYMBOL_GPL(xprt_alloc_slot);
1726
1727 void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1728 {
1729         spin_lock(&xprt->reserve_lock);
1730         if (!xprt_wake_up_backlog(xprt, req) &&
1731             !xprt_dynamic_free_slot(xprt, req)) {
1732                 memset(req, 0, sizeof(*req));   /* mark unused */
1733                 list_add(&req->rq_list, &xprt->free);
1734         }
1735         spin_unlock(&xprt->reserve_lock);
1736 }
1737 EXPORT_SYMBOL_GPL(xprt_free_slot);
1738
1739 static void xprt_free_all_slots(struct rpc_xprt *xprt)
1740 {
1741         struct rpc_rqst *req;
1742         while (!list_empty(&xprt->free)) {
1743                 req = list_first_entry(&xprt->free, struct rpc_rqst, rq_list);
1744                 list_del(&req->rq_list);
1745                 kfree(req);
1746         }
1747 }
1748
1749 struct rpc_xprt *xprt_alloc(struct net *net, size_t size,
1750                 unsigned int num_prealloc,
1751                 unsigned int max_alloc)
1752 {
1753         struct rpc_xprt *xprt;
1754         struct rpc_rqst *req;
1755         int i;
1756
1757         xprt = kzalloc(size, GFP_KERNEL);
1758         if (xprt == NULL)
1759                 goto out;
1760
1761         xprt_init(xprt, net);
1762
1763         for (i = 0; i < num_prealloc; i++) {
1764                 req = kzalloc(sizeof(struct rpc_rqst), GFP_KERNEL);
1765                 if (!req)
1766                         goto out_free;
1767                 list_add(&req->rq_list, &xprt->free);
1768         }
1769         if (max_alloc > num_prealloc)
1770                 xprt->max_reqs = max_alloc;
1771         else
1772                 xprt->max_reqs = num_prealloc;
1773         xprt->min_reqs = num_prealloc;
1774         xprt->num_reqs = num_prealloc;
1775
1776         return xprt;
1777
1778 out_free:
1779         xprt_free(xprt);
1780 out:
1781         return NULL;
1782 }
1783 EXPORT_SYMBOL_GPL(xprt_alloc);
1784
1785 void xprt_free(struct rpc_xprt *xprt)
1786 {
1787         put_net(xprt->xprt_net);
1788         xprt_free_all_slots(xprt);
1789         kfree_rcu(xprt, rcu);
1790 }
1791 EXPORT_SYMBOL_GPL(xprt_free);
1792
1793 static void
1794 xprt_init_connect_cookie(struct rpc_rqst *req, struct rpc_xprt *xprt)
1795 {
1796         req->rq_connect_cookie = xprt_connect_cookie(xprt) - 1;
1797 }
1798
1799 static __be32
1800 xprt_alloc_xid(struct rpc_xprt *xprt)
1801 {
1802         __be32 xid;
1803
1804         spin_lock(&xprt->reserve_lock);
1805         xid = (__force __be32)xprt->xid++;
1806         spin_unlock(&xprt->reserve_lock);
1807         return xid;
1808 }
1809
1810 static void
1811 xprt_init_xid(struct rpc_xprt *xprt)
1812 {
1813         xprt->xid = prandom_u32();
1814 }
1815
1816 static void
1817 xprt_request_init(struct rpc_task *task)
1818 {
1819         struct rpc_xprt *xprt = task->tk_xprt;
1820         struct rpc_rqst *req = task->tk_rqstp;
1821
1822         req->rq_task    = task;
1823         req->rq_xprt    = xprt;
1824         req->rq_buffer  = NULL;
1825         req->rq_xid     = xprt_alloc_xid(xprt);
1826         xprt_init_connect_cookie(req, xprt);
1827         req->rq_snd_buf.len = 0;
1828         req->rq_snd_buf.buflen = 0;
1829         req->rq_rcv_buf.len = 0;
1830         req->rq_rcv_buf.buflen = 0;
1831         req->rq_snd_buf.bvec = NULL;
1832         req->rq_rcv_buf.bvec = NULL;
1833         req->rq_release_snd_buf = NULL;
1834         xprt_init_majortimeo(task, req);
1835
1836         trace_xprt_reserve(req);
1837 }
1838
1839 static void
1840 xprt_do_reserve(struct rpc_xprt *xprt, struct rpc_task *task)
1841 {
1842         xprt->ops->alloc_slot(xprt, task);
1843         if (task->tk_rqstp != NULL)
1844                 xprt_request_init(task);
1845 }
1846
1847 /**
1848  * xprt_reserve - allocate an RPC request slot
1849  * @task: RPC task requesting a slot allocation
1850  *
1851  * If the transport is marked as being congested, or if no more
1852  * slots are available, place the task on the transport's
1853  * backlog queue.
1854  */
1855 void xprt_reserve(struct rpc_task *task)
1856 {
1857         struct rpc_xprt *xprt = task->tk_xprt;
1858
1859         task->tk_status = 0;
1860         if (task->tk_rqstp != NULL)
1861                 return;
1862
1863         task->tk_status = -EAGAIN;
1864         if (!xprt_throttle_congested(xprt, task))
1865                 xprt_do_reserve(xprt, task);
1866 }
1867
1868 /**
1869  * xprt_retry_reserve - allocate an RPC request slot
1870  * @task: RPC task requesting a slot allocation
1871  *
1872  * If no more slots are available, place the task on the transport's
1873  * backlog queue.
1874  * Note that the only difference with xprt_reserve is that we now
1875  * ignore the value of the XPRT_CONGESTED flag.
1876  */
1877 void xprt_retry_reserve(struct rpc_task *task)
1878 {
1879         struct rpc_xprt *xprt = task->tk_xprt;
1880
1881         task->tk_status = 0;
1882         if (task->tk_rqstp != NULL)
1883                 return;
1884
1885         task->tk_status = -EAGAIN;
1886         xprt_do_reserve(xprt, task);
1887 }
1888
1889 /**
1890  * xprt_release - release an RPC request slot
1891  * @task: task which is finished with the slot
1892  *
1893  */
1894 void xprt_release(struct rpc_task *task)
1895 {
1896         struct rpc_xprt *xprt;
1897         struct rpc_rqst *req = task->tk_rqstp;
1898
1899         if (req == NULL) {
1900                 if (task->tk_client) {
1901                         xprt = task->tk_xprt;
1902                         xprt_release_write(xprt, task);
1903                 }
1904                 return;
1905         }
1906
1907         xprt = req->rq_xprt;
1908         xprt_request_dequeue_xprt(task);
1909         spin_lock(&xprt->transport_lock);
1910         xprt->ops->release_xprt(xprt, task);
1911         if (xprt->ops->release_request)
1912                 xprt->ops->release_request(task);
1913         xprt_schedule_autodisconnect(xprt);
1914         spin_unlock(&xprt->transport_lock);
1915         if (req->rq_buffer)
1916                 xprt->ops->buf_free(task);
1917         xdr_free_bvec(&req->rq_rcv_buf);
1918         xdr_free_bvec(&req->rq_snd_buf);
1919         if (req->rq_cred != NULL)
1920                 put_rpccred(req->rq_cred);
1921         if (req->rq_release_snd_buf)
1922                 req->rq_release_snd_buf(req);
1923
1924         task->tk_rqstp = NULL;
1925         if (likely(!bc_prealloc(req)))
1926                 xprt->ops->free_slot(xprt, req);
1927         else
1928                 xprt_free_bc_request(req);
1929 }
1930
1931 #ifdef CONFIG_SUNRPC_BACKCHANNEL
1932 void
1933 xprt_init_bc_request(struct rpc_rqst *req, struct rpc_task *task)
1934 {
1935         struct xdr_buf *xbufp = &req->rq_snd_buf;
1936
1937         task->tk_rqstp = req;
1938         req->rq_task = task;
1939         xprt_init_connect_cookie(req, req->rq_xprt);
1940         /*
1941          * Set up the xdr_buf length.
1942          * This also indicates that the buffer is XDR encoded already.
1943          */
1944         xbufp->len = xbufp->head[0].iov_len + xbufp->page_len +
1945                 xbufp->tail[0].iov_len;
1946 }
1947 #endif
1948
1949 static void xprt_init(struct rpc_xprt *xprt, struct net *net)
1950 {
1951         kref_init(&xprt->kref);
1952
1953         spin_lock_init(&xprt->transport_lock);
1954         spin_lock_init(&xprt->reserve_lock);
1955         spin_lock_init(&xprt->queue_lock);
1956
1957         INIT_LIST_HEAD(&xprt->free);
1958         xprt->recv_queue = RB_ROOT;
1959         INIT_LIST_HEAD(&xprt->xmit_queue);
1960 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1961         spin_lock_init(&xprt->bc_pa_lock);
1962         INIT_LIST_HEAD(&xprt->bc_pa_list);
1963 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1964         INIT_LIST_HEAD(&xprt->xprt_switch);
1965
1966         xprt->last_used = jiffies;
1967         xprt->cwnd = RPC_INITCWND;
1968         xprt->bind_index = 0;
1969
1970         rpc_init_wait_queue(&xprt->binding, "xprt_binding");
1971         rpc_init_wait_queue(&xprt->pending, "xprt_pending");
1972         rpc_init_wait_queue(&xprt->sending, "xprt_sending");
1973         rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
1974
1975         xprt_init_xid(xprt);
1976
1977         xprt->xprt_net = get_net(net);
1978 }
1979
1980 /**
1981  * xprt_create_transport - create an RPC transport
1982  * @args: rpc transport creation arguments
1983  *
1984  */
1985 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
1986 {
1987         struct rpc_xprt *xprt;
1988         const struct xprt_class *t;
1989
1990         t = xprt_class_find_by_ident(args->ident);
1991         if (!t) {
1992                 dprintk("RPC: transport (%d) not supported\n", args->ident);
1993                 return ERR_PTR(-EIO);
1994         }
1995
1996         xprt = t->setup(args);
1997         xprt_class_release(t);
1998
1999         if (IS_ERR(xprt))
2000                 goto out;
2001         if (args->flags & XPRT_CREATE_NO_IDLE_TIMEOUT)
2002                 xprt->idle_timeout = 0;
2003         INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
2004         if (xprt_has_timer(xprt))
2005                 timer_setup(&xprt->timer, xprt_init_autodisconnect, 0);
2006         else
2007                 timer_setup(&xprt->timer, NULL, 0);
2008
2009         if (strlen(args->servername) > RPC_MAXNETNAMELEN) {
2010                 xprt_destroy(xprt);
2011                 return ERR_PTR(-EINVAL);
2012         }
2013         xprt->servername = kstrdup(args->servername, GFP_KERNEL);
2014         if (xprt->servername == NULL) {
2015                 xprt_destroy(xprt);
2016                 return ERR_PTR(-ENOMEM);
2017         }
2018
2019         rpc_xprt_debugfs_register(xprt);
2020
2021         trace_xprt_create(xprt);
2022 out:
2023         return xprt;
2024 }
2025
2026 static void xprt_destroy_cb(struct work_struct *work)
2027 {
2028         struct rpc_xprt *xprt =
2029                 container_of(work, struct rpc_xprt, task_cleanup);
2030
2031         trace_xprt_destroy(xprt);
2032
2033         rpc_xprt_debugfs_unregister(xprt);
2034         rpc_destroy_wait_queue(&xprt->binding);
2035         rpc_destroy_wait_queue(&xprt->pending);
2036         rpc_destroy_wait_queue(&xprt->sending);
2037         rpc_destroy_wait_queue(&xprt->backlog);
2038         kfree(xprt->servername);
2039         /*
2040          * Destroy any existing back channel
2041          */
2042         xprt_destroy_backchannel(xprt, UINT_MAX);
2043
2044         /*
2045          * Tear down transport state and free the rpc_xprt
2046          */
2047         xprt->ops->destroy(xprt);
2048 }
2049
2050 /**
2051  * xprt_destroy - destroy an RPC transport, killing off all requests.
2052  * @xprt: transport to destroy
2053  *
2054  */
2055 static void xprt_destroy(struct rpc_xprt *xprt)
2056 {
2057         /*
2058          * Exclude transport connect/disconnect handlers and autoclose
2059          */
2060         wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_UNINTERRUPTIBLE);
2061
2062         del_timer_sync(&xprt->timer);
2063
2064         /*
2065          * Destroy sockets etc from the system workqueue so they can
2066          * safely flush receive work running on rpciod.
2067          */
2068         INIT_WORK(&xprt->task_cleanup, xprt_destroy_cb);
2069         schedule_work(&xprt->task_cleanup);
2070 }
2071
2072 static void xprt_destroy_kref(struct kref *kref)
2073 {
2074         xprt_destroy(container_of(kref, struct rpc_xprt, kref));
2075 }
2076
2077 /**
2078  * xprt_get - return a reference to an RPC transport.
2079  * @xprt: pointer to the transport
2080  *
2081  */
2082 struct rpc_xprt *xprt_get(struct rpc_xprt *xprt)
2083 {
2084         if (xprt != NULL && kref_get_unless_zero(&xprt->kref))
2085                 return xprt;
2086         return NULL;
2087 }
2088 EXPORT_SYMBOL_GPL(xprt_get);
2089
2090 /**
2091  * xprt_put - release a reference to an RPC transport.
2092  * @xprt: pointer to the transport
2093  *
2094  */
2095 void xprt_put(struct rpc_xprt *xprt)
2096 {
2097         if (xprt != NULL)
2098                 kref_put(&xprt->kref, xprt_destroy_kref);
2099 }
2100 EXPORT_SYMBOL_GPL(xprt_put);