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 #include "sysfs.h"
  59 #include "fail.h"
  60
  61 /*
  62  * Local variables
  63  */
  64
  65 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
  66 # define RPCDBG_FACILITY        RPCDBG_XPRT
  67 #endif
  68
  69 /*
  70  * Local functions
  71  */
  72 static void      xprt_init(struct rpc_xprt *xprt, struct net *net);
  73 static __be32   xprt_alloc_xid(struct rpc_xprt *xprt);
  74 static void      xprt_destroy(struct rpc_xprt *xprt);
  75 static void      xprt_request_init(struct rpc_task *task);
  76
  77 static DEFINE_SPINLOCK(xprt_list_lock);
  78 static LIST_HEAD(xprt_list);
  79
  80 static unsigned long xprt_request_timeout(const struct rpc_rqst *req)
  81 {
  82         unsigned long timeout = jiffies + req->rq_timeout;
  83
  84         if (time_before(timeout, req->rq_majortimeo))
  85                 return timeout;
  86         return req->rq_majortimeo;
  87 }
  88
  89 /**
  90  * xprt_register_transport - register a transport implementation
  91  * @transport: transport to register
  92  *
  93  * If a transport implementation is loaded as a kernel module, it can
  94  * call this interface to make itself known to the RPC client.
  95  *
  96  * Returns:
  97  * 0:           transport successfully registered
  98  * -EEXIST:     transport already registered
  99  * -EINVAL:     transport module being unloaded
 100  */
 101 int xprt_register_transport(struct xprt_class *transport)
 102 {
 103         struct xprt_class *t;
 104         int result;
 105
 106         result = -EEXIST;
 107         spin_lock(&xprt_list_lock);
 108         list_for_each_entry(t, &xprt_list, list) {
 109                 /* don't register the same transport class twice */
 110                 if (t->ident == transport->ident)
 111                         goto out;
 112         }
 113
 114         list_add_tail(&transport->list, &xprt_list);
 115         printk(KERN_INFO "RPC: Registered %s transport module.\n",
 116                transport->name);
 117         result = 0;
 118
 119 out:
 120         spin_unlock(&xprt_list_lock);
 121         return result;
 122 }
 123 EXPORT_SYMBOL_GPL(xprt_register_transport);
 124
 125 /**
 126  * xprt_unregister_transport - unregister a transport implementation
 127  * @transport: transport to unregister
 128  *
 129  * Returns:
 130  * 0:           transport successfully unregistered
 131  * -ENOENT:     transport never registered
 132  */
 133 int xprt_unregister_transport(struct xprt_class *transport)
 134 {
 135         struct xprt_class *t;
 136         int result;
 137
 138         result = 0;
 139         spin_lock(&xprt_list_lock);
 140         list_for_each_entry(t, &xprt_list, list) {
 141                 if (t == transport) {
 142                         printk(KERN_INFO
 143                                 "RPC: Unregistered %s transport module.\n",
 144                                 transport->name);
 145                         list_del_init(&transport->list);
 146                         goto out;
 147                 }
 148         }
 149         result = -ENOENT;
 150
 151 out:
 152         spin_unlock(&xprt_list_lock);
 153         return result;
 154 }
 155 EXPORT_SYMBOL_GPL(xprt_unregister_transport);
 156
 157 static void
 158 xprt_class_release(const struct xprt_class *t)
 159 {
 160         module_put(t->owner);
 161 }
 162
 163 static const struct xprt_class *
 164 xprt_class_find_by_ident_locked(int ident)
 165 {
 166         const struct xprt_class *t;
 167
 168         list_for_each_entry(t, &xprt_list, list) {
 169                 if (t->ident != ident)
 170                         continue;
 171                 if (!try_module_get(t->owner))
 172                         continue;
 173                 return t;
 174         }
 175         return NULL;
 176 }
 177
 178 static const struct xprt_class *
 179 xprt_class_find_by_ident(int ident)
 180 {
 181         const struct xprt_class *t;
 182
 183         spin_lock(&xprt_list_lock);
 184         t = xprt_class_find_by_ident_locked(ident);
 185         spin_unlock(&xprt_list_lock);
 186         return t;
 187 }
 188
 189 static const struct xprt_class *
 190 xprt_class_find_by_netid_locked(const char *netid)
 191 {
 192         const struct xprt_class *t;
 193         unsigned int i;
 194
 195         list_for_each_entry(t, &xprt_list, list) {
 196                 for (i = 0; t->netid[i][0] != '\0'; i++) {
 197                         if (strcmp(t->netid[i], netid) != 0)
 198                                 continue;
 199                         if (!try_module_get(t->owner))
 200                                 continue;
 201                         return t;
 202                 }
 203         }
 204         return NULL;
 205 }
 206
 207 static const struct xprt_class *
 208 xprt_class_find_by_netid(const char *netid)
 209 {
 210         const struct xprt_class *t;
 211
 212         spin_lock(&xprt_list_lock);
 213         t = xprt_class_find_by_netid_locked(netid);
 214         if (!t) {
 215                 spin_unlock(&xprt_list_lock);
 216                 request_module("rpc%s", netid);
 217                 spin_lock(&xprt_list_lock);
 218                 t = xprt_class_find_by_netid_locked(netid);
 219         }
 220         spin_unlock(&xprt_list_lock);
 221         return t;
 222 }
 223
 224 /**
 225  * xprt_find_transport_ident - convert a netid into a transport identifier
 226  * @netid: transport to load
 227  *
 228  * Returns:
 229  * > 0:         transport identifier
 230  * -ENOENT:     transport module not available
 231  */
 232 int xprt_find_transport_ident(const char *netid)
 233 {
 234         const struct xprt_class *t;
 235         int ret;
 236
 237         t = xprt_class_find_by_netid(netid);
 238         if (!t)
 239                 return -ENOENT;
 240         ret = t->ident;
 241         xprt_class_release(t);
 242         return ret;
 243 }
 244 EXPORT_SYMBOL_GPL(xprt_find_transport_ident);
 245
 246 static void xprt_clear_locked(struct rpc_xprt *xprt)
 247 {
 248         xprt->snd_task = NULL;
 249         if (!test_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
 250                 smp_mb__before_atomic();
 251                 clear_bit(XPRT_LOCKED, &xprt->state);
 252                 smp_mb__after_atomic();
 253         } else
 254                 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 255 }
 256
 257 /**
 258  * xprt_reserve_xprt - serialize write access to transports
 259  * @task: task that is requesting access to the transport
 260  * @xprt: pointer to the target transport
 261  *
 262  * This prevents mixing the payload of separate requests, and prevents
 263  * transport connects from colliding with writes.  No congestion control
 264  * is provided.
 265  */
 266 int xprt_reserve_xprt(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 (test_bit(XPRT_WRITE_SPACE, &xprt->state))
 276                 goto out_unlock;
 277         xprt->snd_task = task;
 278
 279 out_locked:
 280         trace_xprt_reserve_xprt(xprt, task);
 281         return 1;
 282
 283 out_unlock:
 284         xprt_clear_locked(xprt);
 285 out_sleep:
 286         task->tk_status = -EAGAIN;
 287         if  (RPC_IS_SOFT(task))
 288                 rpc_sleep_on_timeout(&xprt->sending, task, NULL,
 289                                 xprt_request_timeout(req));
 290         else
 291                 rpc_sleep_on(&xprt->sending, task, NULL);
 292         return 0;
 293 }
 294 EXPORT_SYMBOL_GPL(xprt_reserve_xprt);
 295
 296 static bool
 297 xprt_need_congestion_window_wait(struct rpc_xprt *xprt)
 298 {
 299         return test_bit(XPRT_CWND_WAIT, &xprt->state);
 300 }
 301
 302 static void
 303 xprt_set_congestion_window_wait(struct rpc_xprt *xprt)
 304 {
 305         if (!list_empty(&xprt->xmit_queue)) {
 306                 /* Peek at head of queue to see if it can make progress */
 307                 if (list_first_entry(&xprt->xmit_queue, struct rpc_rqst,
 308                                         rq_xmit)->rq_cong)
 309                         return;
 310         }
 311         set_bit(XPRT_CWND_WAIT, &xprt->state);
 312 }
 313
 314 static void
 315 xprt_test_and_clear_congestion_window_wait(struct rpc_xprt *xprt)
 316 {
 317         if (!RPCXPRT_CONGESTED(xprt))
 318                 clear_bit(XPRT_CWND_WAIT, &xprt->state);
 319 }
 320
 321 /*
 322  * xprt_reserve_xprt_cong - serialize write access to transports
 323  * @task: task that is requesting access to the transport
 324  *
 325  * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
 326  * integrated into the decision of whether a request is allowed to be
 327  * woken up and given access to the transport.
 328  * Note that the lock is only granted if we know there are free slots.
 329  */
 330 int xprt_reserve_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 331 {
 332         struct rpc_rqst *req = task->tk_rqstp;
 333
 334         if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
 335                 if (task == xprt->snd_task)
 336                         goto out_locked;
 337                 goto out_sleep;
 338         }
 339         if (req == NULL) {
 340                 xprt->snd_task = task;
 341                 goto out_locked;
 342         }
 343         if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
 344                 goto out_unlock;
 345         if (!xprt_need_congestion_window_wait(xprt)) {
 346                 xprt->snd_task = task;
 347                 goto out_locked;
 348         }
 349 out_unlock:
 350         xprt_clear_locked(xprt);
 351 out_sleep:
 352         task->tk_status = -EAGAIN;
 353         if (RPC_IS_SOFT(task))
 354                 rpc_sleep_on_timeout(&xprt->sending, task, NULL,
 355                                 xprt_request_timeout(req));
 356         else
 357                 rpc_sleep_on(&xprt->sending, task, NULL);
 358         return 0;
 359 out_locked:
 360         trace_xprt_reserve_cong(xprt, task);
 361         return 1;
 362 }
 363 EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong);
 364
 365 static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
 366 {
 367         int retval;
 368
 369         if (test_bit(XPRT_LOCKED, &xprt->state) && xprt->snd_task == task)
 370                 return 1;
 371         spin_lock(&xprt->transport_lock);
 372         retval = xprt->ops->reserve_xprt(xprt, task);
 373         spin_unlock(&xprt->transport_lock);
 374         return retval;
 375 }
 376
 377 static bool __xprt_lock_write_func(struct rpc_task *task, void *data)
 378 {
 379         struct rpc_xprt *xprt = data;
 380
 381         xprt->snd_task = task;
 382         return true;
 383 }
 384
 385 static void __xprt_lock_write_next(struct rpc_xprt *xprt)
 386 {
 387         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 388                 return;
 389         if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
 390                 goto out_unlock;
 391         if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
 392                                 __xprt_lock_write_func, xprt))
 393                 return;
 394 out_unlock:
 395         xprt_clear_locked(xprt);
 396 }
 397
 398 static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt)
 399 {
 400         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 401                 return;
 402         if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
 403                 goto out_unlock;
 404         if (xprt_need_congestion_window_wait(xprt))
 405                 goto out_unlock;
 406         if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
 407                                 __xprt_lock_write_func, xprt))
 408                 return;
 409 out_unlock:
 410         xprt_clear_locked(xprt);
 411 }
 412
 413 /**
 414  * xprt_release_xprt - allow other requests to use a transport
 415  * @xprt: transport with other tasks potentially waiting
 416  * @task: task that is releasing access to the transport
 417  *
 418  * Note that "task" can be NULL.  No congestion control is provided.
 419  */
 420 void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 421 {
 422         if (xprt->snd_task == task) {
 423                 xprt_clear_locked(xprt);
 424                 __xprt_lock_write_next(xprt);
 425         }
 426         trace_xprt_release_xprt(xprt, task);
 427 }
 428 EXPORT_SYMBOL_GPL(xprt_release_xprt);
 429
 430 /**
 431  * xprt_release_xprt_cong - allow other requests to use a transport
 432  * @xprt: transport with other tasks potentially waiting
 433  * @task: task that is releasing access to the transport
 434  *
 435  * Note that "task" can be NULL.  Another task is awoken to use the
 436  * transport if the transport's congestion window allows it.
 437  */
 438 void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 439 {
 440         if (xprt->snd_task == task) {
 441                 xprt_clear_locked(xprt);
 442                 __xprt_lock_write_next_cong(xprt);
 443         }
 444         trace_xprt_release_cong(xprt, task);
 445 }
 446 EXPORT_SYMBOL_GPL(xprt_release_xprt_cong);
 447
 448 void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
 449 {
 450         if (xprt->snd_task != task)
 451                 return;
 452         spin_lock(&xprt->transport_lock);
 453         xprt->ops->release_xprt(xprt, task);
 454         spin_unlock(&xprt->transport_lock);
 455 }
 456
 457 /*
 458  * Van Jacobson congestion avoidance. Check if the congestion window
 459  * overflowed. Put the task to sleep if this is the case.
 460  */
 461 static int
 462 __xprt_get_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
 463 {
 464         if (req->rq_cong)
 465                 return 1;
 466         trace_xprt_get_cong(xprt, req->rq_task);
 467         if (RPCXPRT_CONGESTED(xprt)) {
 468                 xprt_set_congestion_window_wait(xprt);
 469                 return 0;
 470         }
 471         req->rq_cong = 1;
 472         xprt->cong += RPC_CWNDSCALE;
 473         return 1;
 474 }
 475
 476 /*
 477  * Adjust the congestion window, and wake up the next task
 478  * that has been sleeping due to congestion
 479  */
 480 static void
 481 __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
 482 {
 483         if (!req->rq_cong)
 484                 return;
 485         req->rq_cong = 0;
 486         xprt->cong -= RPC_CWNDSCALE;
 487         xprt_test_and_clear_congestion_window_wait(xprt);
 488         trace_xprt_put_cong(xprt, req->rq_task);
 489         __xprt_lock_write_next_cong(xprt);
 490 }
 491
 492 /**
 493  * xprt_request_get_cong - Request congestion control credits
 494  * @xprt: pointer to transport
 495  * @req: pointer to RPC request
 496  *
 497  * Useful for transports that require congestion control.
 498  */
 499 bool
 500 xprt_request_get_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
 501 {
 502         bool ret = false;
 503
 504         if (req->rq_cong)
 505                 return true;
 506         spin_lock(&xprt->transport_lock);
 507         ret = __xprt_get_cong(xprt, req) != 0;
 508         spin_unlock(&xprt->transport_lock);
 509         return ret;
 510 }
 511 EXPORT_SYMBOL_GPL(xprt_request_get_cong);
 512
 513 /**
 514  * xprt_release_rqst_cong - housekeeping when request is complete
 515  * @task: RPC request that recently completed
 516  *
 517  * Useful for transports that require congestion control.
 518  */
 519 void xprt_release_rqst_cong(struct rpc_task *task)
 520 {
 521         struct rpc_rqst *req = task->tk_rqstp;
 522
 523         __xprt_put_cong(req->rq_xprt, req);
 524 }
 525 EXPORT_SYMBOL_GPL(xprt_release_rqst_cong);
 526
 527 static void xprt_clear_congestion_window_wait_locked(struct rpc_xprt *xprt)
 528 {
 529         if (test_and_clear_bit(XPRT_CWND_WAIT, &xprt->state))
 530                 __xprt_lock_write_next_cong(xprt);
 531 }
 532
 533 /*
 534  * Clear the congestion window wait flag and wake up the next
 535  * entry on xprt->sending
 536  */
 537 static void
 538 xprt_clear_congestion_window_wait(struct rpc_xprt *xprt)
 539 {
 540         if (test_and_clear_bit(XPRT_CWND_WAIT, &xprt->state)) {
 541                 spin_lock(&xprt->transport_lock);
 542                 __xprt_lock_write_next_cong(xprt);
 543                 spin_unlock(&xprt->transport_lock);
 544         }
 545 }
 546
 547 /**
 548  * xprt_adjust_cwnd - adjust transport congestion window
 549  * @xprt: pointer to xprt
 550  * @task: recently completed RPC request used to adjust window
 551  * @result: result code of completed RPC request
 552  *
 553  * The transport code maintains an estimate on the maximum number of out-
 554  * standing RPC requests, using a smoothed version of the congestion
 555  * avoidance implemented in 44BSD. This is basically the Van Jacobson
 556  * congestion algorithm: If a retransmit occurs, the congestion window is
 557  * halved; otherwise, it is incremented by 1/cwnd when
 558  *
 559  *      -       a reply is received and
 560  *      -       a full number of requests are outstanding and
 561  *      -       the congestion window hasn't been updated recently.
 562  */
 563 void xprt_adjust_cwnd(struct rpc_xprt *xprt, struct rpc_task *task, int result)
 564 {
 565         struct rpc_rqst *req = task->tk_rqstp;
 566         unsigned long cwnd = xprt->cwnd;
 567
 568         if (result >= 0 && cwnd <= xprt->cong) {
 569                 /* The (cwnd >> 1) term makes sure
 570                  * the result gets rounded properly. */
 571                 cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
 572                 if (cwnd > RPC_MAXCWND(xprt))
 573                         cwnd = RPC_MAXCWND(xprt);
 574                 __xprt_lock_write_next_cong(xprt);
 575         } else if (result == -ETIMEDOUT) {
 576                 cwnd >>= 1;
 577                 if (cwnd < RPC_CWNDSCALE)
 578                         cwnd = RPC_CWNDSCALE;
 579         }
 580         dprintk("RPC:       cong %ld, cwnd was %ld, now %ld\n",
 581                         xprt->cong, xprt->cwnd, cwnd);
 582         xprt->cwnd = cwnd;
 583         __xprt_put_cong(xprt, req);
 584 }
 585 EXPORT_SYMBOL_GPL(xprt_adjust_cwnd);
 586
 587 /**
 588  * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
 589  * @xprt: transport with waiting tasks
 590  * @status: result code to plant in each task before waking it
 591  *
 592  */
 593 void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
 594 {
 595         if (status < 0)
 596                 rpc_wake_up_status(&xprt->pending, status);
 597         else
 598                 rpc_wake_up(&xprt->pending);
 599 }
 600 EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks);
 601
 602 /**
 603  * xprt_wait_for_buffer_space - wait for transport output buffer to clear
 604  * @xprt: transport
 605  *
 606  * Note that we only set the timer for the case of RPC_IS_SOFT(), since
 607  * we don't in general want to force a socket disconnection due to
 608  * an incomplete RPC call transmission.
 609  */
 610 void xprt_wait_for_buffer_space(struct rpc_xprt *xprt)
 611 {
 612         set_bit(XPRT_WRITE_SPACE, &xprt->state);
 613 }
 614 EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space);
 615
 616 static bool
 617 xprt_clear_write_space_locked(struct rpc_xprt *xprt)
 618 {
 619         if (test_and_clear_bit(XPRT_WRITE_SPACE, &xprt->state)) {
 620                 __xprt_lock_write_next(xprt);
 621                 dprintk("RPC:       write space: waking waiting task on "
 622                                 "xprt %p\n", xprt);
 623                 return true;
 624         }
 625         return false;
 626 }
 627
 628 /**
 629  * xprt_write_space - wake the task waiting for transport output buffer space
 630  * @xprt: transport with waiting tasks
 631  *
 632  * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
 633  */
 634 bool xprt_write_space(struct rpc_xprt *xprt)
 635 {
 636         bool ret;
 637
 638         if (!test_bit(XPRT_WRITE_SPACE, &xprt->state))
 639                 return false;
 640         spin_lock(&xprt->transport_lock);
 641         ret = xprt_clear_write_space_locked(xprt);
 642         spin_unlock(&xprt->transport_lock);
 643         return ret;
 644 }
 645 EXPORT_SYMBOL_GPL(xprt_write_space);
 646
 647 static unsigned long xprt_abs_ktime_to_jiffies(ktime_t abstime)
 648 {
 649         s64 delta = ktime_to_ns(ktime_get() - abstime);
 650         return likely(delta >= 0) ?
 651                 jiffies - nsecs_to_jiffies(delta) :
 652                 jiffies + nsecs_to_jiffies(-delta);
 653 }
 654
 655 static unsigned long xprt_calc_majortimeo(struct rpc_rqst *req)
 656 {
 657         const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
 658         unsigned long majortimeo = req->rq_timeout;
 659
 660         if (to->to_exponential)
 661                 majortimeo <<= to->to_retries;
 662         else
 663                 majortimeo += to->to_increment * to->to_retries;
 664         if (majortimeo > to->to_maxval || majortimeo == 0)
 665                 majortimeo = to->to_maxval;
 666         return majortimeo;
 667 }
 668
 669 static void xprt_reset_majortimeo(struct rpc_rqst *req)
 670 {
 671         req->rq_majortimeo += xprt_calc_majortimeo(req);
 672 }
 673
 674 static void xprt_reset_minortimeo(struct rpc_rqst *req)
 675 {
 676         req->rq_minortimeo += req->rq_timeout;
 677 }
 678
 679 static void xprt_init_majortimeo(struct rpc_task *task, struct rpc_rqst *req)
 680 {
 681         unsigned long time_init;
 682         struct rpc_xprt *xprt = req->rq_xprt;
 683
 684         if (likely(xprt && xprt_connected(xprt)))
 685                 time_init = jiffies;
 686         else
 687                 time_init = xprt_abs_ktime_to_jiffies(task->tk_start);
 688         req->rq_timeout = task->tk_client->cl_timeout->to_initval;
 689         req->rq_majortimeo = time_init + xprt_calc_majortimeo(req);
 690         req->rq_minortimeo = time_init + req->rq_timeout;
 691 }
 692
 693 /**
 694  * xprt_adjust_timeout - adjust timeout values for next retransmit
 695  * @req: RPC request containing parameters to use for the adjustment
 696  *
 697  */
 698 int xprt_adjust_timeout(struct rpc_rqst *req)
 699 {
 700         struct rpc_xprt *xprt = req->rq_xprt;
 701         const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
 702         int status = 0;
 703
 704         if (time_before(jiffies, req->rq_majortimeo)) {
 705                 if (time_before(jiffies, req->rq_minortimeo))
 706                         return status;
 707                 if (to->to_exponential)
 708                         req->rq_timeout <<= 1;
 709                 else
 710                         req->rq_timeout += to->to_increment;
 711                 if (to->to_maxval && req->rq_timeout >= to->to_maxval)
 712                         req->rq_timeout = to->to_maxval;
 713                 req->rq_retries++;
 714         } else {
 715                 req->rq_timeout = to->to_initval;
 716                 req->rq_retries = 0;
 717                 xprt_reset_majortimeo(req);
 718                 /* Reset the RTT counters == "slow start" */
 719                 spin_lock(&xprt->transport_lock);
 720                 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
 721                 spin_unlock(&xprt->transport_lock);
 722                 status = -ETIMEDOUT;
 723         }
 724         xprt_reset_minortimeo(req);
 725
 726         if (req->rq_timeout == 0) {
 727                 printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
 728                 req->rq_timeout = 5 * HZ;
 729         }
 730         return status;
 731 }
 732
 733 static void xprt_autoclose(struct work_struct *work)
 734 {
 735         struct rpc_xprt *xprt =
 736                 container_of(work, struct rpc_xprt, task_cleanup);
 737         unsigned int pflags = memalloc_nofs_save();
 738
 739         trace_xprt_disconnect_auto(xprt);
 740         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 741         xprt->ops->close(xprt);
 742         xprt_release_write(xprt, NULL);
 743         wake_up_bit(&xprt->state, XPRT_LOCKED);
 744         memalloc_nofs_restore(pflags);
 745 }
 746
 747 /**
 748  * xprt_disconnect_done - mark a transport as disconnected
 749  * @xprt: transport to flag for disconnect
 750  *
 751  */
 752 void xprt_disconnect_done(struct rpc_xprt *xprt)
 753 {
 754         trace_xprt_disconnect_done(xprt);
 755         spin_lock(&xprt->transport_lock);
 756         xprt_clear_connected(xprt);
 757         xprt_clear_write_space_locked(xprt);
 758         xprt_clear_congestion_window_wait_locked(xprt);
 759         xprt_wake_pending_tasks(xprt, -ENOTCONN);
 760         spin_unlock(&xprt->transport_lock);
 761 }
 762 EXPORT_SYMBOL_GPL(xprt_disconnect_done);
 763
 764 /**
 765  * xprt_force_disconnect - force a transport to disconnect
 766  * @xprt: transport to disconnect
 767  *
 768  */
 769 void xprt_force_disconnect(struct rpc_xprt *xprt)
 770 {
 771         trace_xprt_disconnect_force(xprt);
 772
 773         /* Don't race with the test_bit() in xprt_clear_locked() */
 774         spin_lock(&xprt->transport_lock);
 775         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 776         /* Try to schedule an autoclose RPC call */
 777         if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 778                 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 779         else if (xprt->snd_task)
 780                 rpc_wake_up_queued_task_set_status(&xprt->pending,
 781                                 xprt->snd_task, -ENOTCONN);
 782         spin_unlock(&xprt->transport_lock);
 783 }
 784 EXPORT_SYMBOL_GPL(xprt_force_disconnect);
 785
 786 static unsigned int
 787 xprt_connect_cookie(struct rpc_xprt *xprt)
 788 {
 789         return READ_ONCE(xprt->connect_cookie);
 790 }
 791
 792 static bool
 793 xprt_request_retransmit_after_disconnect(struct rpc_task *task)
 794 {
 795         struct rpc_rqst *req = task->tk_rqstp;
 796         struct rpc_xprt *xprt = req->rq_xprt;
 797
 798         return req->rq_connect_cookie != xprt_connect_cookie(xprt) ||
 799                 !xprt_connected(xprt);
 800 }
 801
 802 /**
 803  * xprt_conditional_disconnect - force a transport to disconnect
 804  * @xprt: transport to disconnect
 805  * @cookie: 'connection cookie'
 806  *
 807  * This attempts to break the connection if and only if 'cookie' matches
 808  * the current transport 'connection cookie'. It ensures that we don't
 809  * try to break the connection more than once when we need to retransmit
 810  * a batch of RPC requests.
 811  *
 812  */
 813 void xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie)
 814 {
 815         /* Don't race with the test_bit() in xprt_clear_locked() */
 816         spin_lock(&xprt->transport_lock);
 817         if (cookie != xprt->connect_cookie)
 818                 goto out;
 819         if (test_bit(XPRT_CLOSING, &xprt->state))
 820                 goto out;
 821         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 822         /* Try to schedule an autoclose RPC call */
 823         if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 824                 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 825         xprt_wake_pending_tasks(xprt, -EAGAIN);
 826 out:
 827         spin_unlock(&xprt->transport_lock);
 828 }
 829
 830 static bool
 831 xprt_has_timer(const struct rpc_xprt *xprt)
 832 {
 833         return xprt->idle_timeout != 0;
 834 }
 835
 836 static void
 837 xprt_schedule_autodisconnect(struct rpc_xprt *xprt)
 838         __must_hold(&xprt->transport_lock)
 839 {
 840         xprt->last_used = jiffies;
 841         if (RB_EMPTY_ROOT(&xprt->recv_queue) && xprt_has_timer(xprt))
 842                 mod_timer(&xprt->timer, xprt->last_used + xprt->idle_timeout);
 843 }
 844
 845 static void
 846 xprt_init_autodisconnect(struct timer_list *t)
 847 {
 848         struct rpc_xprt *xprt = from_timer(xprt, t, timer);
 849
 850         if (!RB_EMPTY_ROOT(&xprt->recv_queue))
 851                 return;
 852         /* Reset xprt->last_used to avoid connect/autodisconnect cycling */
 853         xprt->last_used = jiffies;
 854         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 855                 return;
 856         queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 857 }
 858
 859 #if IS_ENABLED(CONFIG_FAIL_SUNRPC)
 860 static void xprt_inject_disconnect(struct rpc_xprt *xprt)
 861 {
 862         if (!fail_sunrpc.ignore_client_disconnect &&
 863             should_fail(&fail_sunrpc.attr, 1))
 864                 xprt->ops->inject_disconnect(xprt);
 865 }
 866 #else
 867 static inline void xprt_inject_disconnect(struct rpc_xprt *xprt)
 868 {
 869 }
 870 #endif
 871
 872 bool xprt_lock_connect(struct rpc_xprt *xprt,
 873                 struct rpc_task *task,
 874                 void *cookie)
 875 {
 876         bool ret = false;
 877
 878         spin_lock(&xprt->transport_lock);
 879         if (!test_bit(XPRT_LOCKED, &xprt->state))
 880                 goto out;
 881         if (xprt->snd_task != task)
 882                 goto out;
 883         xprt->snd_task = cookie;
 884         ret = true;
 885 out:
 886         spin_unlock(&xprt->transport_lock);
 887         return ret;
 888 }
 889
 890 void xprt_unlock_connect(struct rpc_xprt *xprt, void *cookie)
 891 {
 892         spin_lock(&xprt->transport_lock);
 893         if (xprt->snd_task != cookie)
 894                 goto out;
 895         if (!test_bit(XPRT_LOCKED, &xprt->state))
 896                 goto out;
 897         xprt->snd_task =NULL;
 898         xprt->ops->release_xprt(xprt, NULL);
 899         xprt_schedule_autodisconnect(xprt);
 900 out:
 901         spin_unlock(&xprt->transport_lock);
 902         wake_up_bit(&xprt->state, XPRT_LOCKED);
 903 }
 904
 905 /**
 906  * xprt_connect - schedule a transport connect operation
 907  * @task: RPC task that is requesting the connect
 908  *
 909  */
 910 void xprt_connect(struct rpc_task *task)
 911 {
 912         struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
 913
 914         trace_xprt_connect(xprt);
 915
 916         if (!xprt_bound(xprt)) {
 917                 task->tk_status = -EAGAIN;
 918                 return;
 919         }
 920         if (!xprt_lock_write(xprt, task))
 921                 return;
 922
 923         if (test_and_clear_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
 924                 trace_xprt_disconnect_cleanup(xprt);
 925                 xprt->ops->close(xprt);
 926         }
 927
 928         if (!xprt_connected(xprt)) {
 929                 task->tk_rqstp->rq_connect_cookie = xprt->connect_cookie;
 930                 rpc_sleep_on_timeout(&xprt->pending, task, NULL,
 931                                 xprt_request_timeout(task->tk_rqstp));
 932
 933                 if (test_bit(XPRT_CLOSING, &xprt->state))
 934                         return;
 935                 if (xprt_test_and_set_connecting(xprt))
 936                         return;
 937                 /* Race breaker */
 938                 if (!xprt_connected(xprt)) {
 939                         xprt->stat.connect_start = jiffies;
 940                         xprt->ops->connect(xprt, task);
 941                 } else {
 942                         xprt_clear_connecting(xprt);
 943                         task->tk_status = 0;
 944                         rpc_wake_up_queued_task(&xprt->pending, task);
 945                 }
 946         }
 947         xprt_release_write(xprt, task);
 948 }
 949
 950 /**
 951  * xprt_reconnect_delay - compute the wait before scheduling a connect
 952  * @xprt: transport instance
 953  *
 954  */
 955 unsigned long xprt_reconnect_delay(const struct rpc_xprt *xprt)
 956 {
 957         unsigned long start, now = jiffies;
 958
 959         start = xprt->stat.connect_start + xprt->reestablish_timeout;
 960         if (time_after(start, now))
 961                 return start - now;
 962         return 0;
 963 }
 964 EXPORT_SYMBOL_GPL(xprt_reconnect_delay);
 965
 966 /**
 967  * xprt_reconnect_backoff - compute the new re-establish timeout
 968  * @xprt: transport instance
 969  * @init_to: initial reestablish timeout
 970  *
 971  */
 972 void xprt_reconnect_backoff(struct rpc_xprt *xprt, unsigned long init_to)
 973 {
 974         xprt->reestablish_timeout <<= 1;
 975         if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
 976                 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
 977         if (xprt->reestablish_timeout < init_to)
 978                 xprt->reestablish_timeout = init_to;
 979 }
 980 EXPORT_SYMBOL_GPL(xprt_reconnect_backoff);
 981
 982 enum xprt_xid_rb_cmp {
 983         XID_RB_EQUAL,
 984         XID_RB_LEFT,
 985         XID_RB_RIGHT,
 986 };
 987 static enum xprt_xid_rb_cmp
 988 xprt_xid_cmp(__be32 xid1, __be32 xid2)
 989 {
 990         if (xid1 == xid2)
 991                 return XID_RB_EQUAL;
 992         if ((__force u32)xid1 < (__force u32)xid2)
 993                 return XID_RB_LEFT;
 994         return XID_RB_RIGHT;
 995 }
 996
 997 static struct rpc_rqst *
 998 xprt_request_rb_find(struct rpc_xprt *xprt, __be32 xid)
 999 {
1000         struct rb_node *n = xprt->recv_queue.rb_node;
1001         struct rpc_rqst *req;
1002
1003         while (n != NULL) {
1004                 req = rb_entry(n, struct rpc_rqst, rq_recv);
1005                 switch (xprt_xid_cmp(xid, req->rq_xid)) {
1006                 case XID_RB_LEFT:
1007                         n = n->rb_left;
1008                         break;
1009                 case XID_RB_RIGHT:
1010                         n = n->rb_right;
1011                         break;
1012                 case XID_RB_EQUAL:
1013                         return req;
1014                 }
1015         }
1016         return NULL;
1017 }
1018
1019 static void
1020 xprt_request_rb_insert(struct rpc_xprt *xprt, struct rpc_rqst *new)
1021 {
1022         struct rb_node **p = &xprt->recv_queue.rb_node;
1023         struct rb_node *n = NULL;
1024         struct rpc_rqst *req;
1025
1026         while (*p != NULL) {
1027                 n = *p;
1028                 req = rb_entry(n, struct rpc_rqst, rq_recv);
1029                 switch(xprt_xid_cmp(new->rq_xid, req->rq_xid)) {
1030                 case XID_RB_LEFT:
1031                         p = &n->rb_left;
1032                         break;
1033                 case XID_RB_RIGHT:
1034                         p = &n->rb_right;
1035                         break;
1036                 case XID_RB_EQUAL:
1037                         WARN_ON_ONCE(new != req);
1038                         return;
1039                 }
1040         }
1041         rb_link_node(&new->rq_recv, n, p);
1042         rb_insert_color(&new->rq_recv, &xprt->recv_queue);
1043 }
1044
1045 static void
1046 xprt_request_rb_remove(struct rpc_xprt *xprt, struct rpc_rqst *req)
1047 {
1048         rb_erase(&req->rq_recv, &xprt->recv_queue);
1049 }
1050
1051 /**
1052  * xprt_lookup_rqst - find an RPC request corresponding to an XID
1053  * @xprt: transport on which the original request was transmitted
1054  * @xid: RPC XID of incoming reply
1055  *
1056  * Caller holds xprt->queue_lock.
1057  */
1058 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
1059 {
1060         struct rpc_rqst *entry;
1061
1062         entry = xprt_request_rb_find(xprt, xid);
1063         if (entry != NULL) {
1064                 trace_xprt_lookup_rqst(xprt, xid, 0);
1065                 entry->rq_rtt = ktime_sub(ktime_get(), entry->rq_xtime);
1066                 return entry;
1067         }
1068
1069         dprintk("RPC:       xprt_lookup_rqst did not find xid %08x\n",
1070                         ntohl(xid));
1071         trace_xprt_lookup_rqst(xprt, xid, -ENOENT);
1072         xprt->stat.bad_xids++;
1073         return NULL;
1074 }
1075 EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
1076
1077 static bool
1078 xprt_is_pinned_rqst(struct rpc_rqst *req)
1079 {
1080         return atomic_read(&req->rq_pin) != 0;
1081 }
1082
1083 /**
1084  * xprt_pin_rqst - Pin a request on the transport receive list
1085  * @req: Request to pin
1086  *
1087  * Caller must ensure this is atomic with the call to xprt_lookup_rqst()
1088  * so should be holding xprt->queue_lock.
1089  */
1090 void xprt_pin_rqst(struct rpc_rqst *req)
1091 {
1092         atomic_inc(&req->rq_pin);
1093 }
1094 EXPORT_SYMBOL_GPL(xprt_pin_rqst);
1095
1096 /**
1097  * xprt_unpin_rqst - Unpin a request on the transport receive list
1098  * @req: Request to pin
1099  *
1100  * Caller should be holding xprt->queue_lock.
1101  */
1102 void xprt_unpin_rqst(struct rpc_rqst *req)
1103 {
1104         if (!test_bit(RPC_TASK_MSG_PIN_WAIT, &req->rq_task->tk_runstate)) {
1105                 atomic_dec(&req->rq_pin);
1106                 return;
1107         }
1108         if (atomic_dec_and_test(&req->rq_pin))
1109                 wake_up_var(&req->rq_pin);
1110 }
1111 EXPORT_SYMBOL_GPL(xprt_unpin_rqst);
1112
1113 static void xprt_wait_on_pinned_rqst(struct rpc_rqst *req)
1114 {
1115         wait_var_event(&req->rq_pin, !xprt_is_pinned_rqst(req));
1116 }
1117
1118 static bool
1119 xprt_request_data_received(struct rpc_task *task)
1120 {
1121         return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
1122                 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) != 0;
1123 }
1124
1125 static bool
1126 xprt_request_need_enqueue_receive(struct rpc_task *task, struct rpc_rqst *req)
1127 {
1128         return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
1129                 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) == 0;
1130 }
1131
1132 /**
1133  * xprt_request_enqueue_receive - Add an request to the receive queue
1134  * @task: RPC task
1135  *
1136  */
1137 void
1138 xprt_request_enqueue_receive(struct rpc_task *task)
1139 {
1140         struct rpc_rqst *req = task->tk_rqstp;
1141         struct rpc_xprt *xprt = req->rq_xprt;
1142
1143         if (!xprt_request_need_enqueue_receive(task, req))
1144                 return;
1145
1146         xprt_request_prepare(task->tk_rqstp);
1147         spin_lock(&xprt->queue_lock);
1148
1149         /* Update the softirq receive buffer */
1150         memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
1151                         sizeof(req->rq_private_buf));
1152
1153         /* Add request to the receive list */
1154         xprt_request_rb_insert(xprt, req);
1155         set_bit(RPC_TASK_NEED_RECV, &task->tk_runstate);
1156         spin_unlock(&xprt->queue_lock);
1157
1158         /* Turn off autodisconnect */
1159         del_singleshot_timer_sync(&xprt->timer);
1160 }
1161
1162 /**
1163  * xprt_request_dequeue_receive_locked - Remove a request from the receive queue
1164  * @task: RPC task
1165  *
1166  * Caller must hold xprt->queue_lock.
1167  */
1168 static void
1169 xprt_request_dequeue_receive_locked(struct rpc_task *task)
1170 {
1171         struct rpc_rqst *req = task->tk_rqstp;
1172
1173         if (test_and_clear_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1174                 xprt_request_rb_remove(req->rq_xprt, req);
1175 }
1176
1177 /**
1178  * xprt_update_rtt - Update RPC RTT statistics
1179  * @task: RPC request that recently completed
1180  *
1181  * Caller holds xprt->queue_lock.
1182  */
1183 void xprt_update_rtt(struct rpc_task *task)
1184 {
1185         struct rpc_rqst *req = task->tk_rqstp;
1186         struct rpc_rtt *rtt = task->tk_client->cl_rtt;
1187         unsigned int timer = task->tk_msg.rpc_proc->p_timer;
1188         long m = usecs_to_jiffies(ktime_to_us(req->rq_rtt));
1189
1190         if (timer) {
1191                 if (req->rq_ntrans == 1)
1192                         rpc_update_rtt(rtt, timer, m);
1193                 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
1194         }
1195 }
1196 EXPORT_SYMBOL_GPL(xprt_update_rtt);
1197
1198 /**
1199  * xprt_complete_rqst - called when reply processing is complete
1200  * @task: RPC request that recently completed
1201  * @copied: actual number of bytes received from the transport
1202  *
1203  * Caller holds xprt->queue_lock.
1204  */
1205 void xprt_complete_rqst(struct rpc_task *task, int copied)
1206 {
1207         struct rpc_rqst *req = task->tk_rqstp;
1208         struct rpc_xprt *xprt = req->rq_xprt;
1209
1210         xprt->stat.recvs++;
1211
1212         req->rq_private_buf.len = copied;
1213         /* Ensure all writes are done before we update */
1214         /* req->rq_reply_bytes_recvd */
1215         smp_wmb();
1216         req->rq_reply_bytes_recvd = copied;
1217         xprt_request_dequeue_receive_locked(task);
1218         rpc_wake_up_queued_task(&xprt->pending, task);
1219 }
1220 EXPORT_SYMBOL_GPL(xprt_complete_rqst);
1221
1222 static void xprt_timer(struct rpc_task *task)
1223 {
1224         struct rpc_rqst *req = task->tk_rqstp;
1225         struct rpc_xprt *xprt = req->rq_xprt;
1226
1227         if (task->tk_status != -ETIMEDOUT)
1228                 return;
1229
1230         trace_xprt_timer(xprt, req->rq_xid, task->tk_status);
1231         if (!req->rq_reply_bytes_recvd) {
1232                 if (xprt->ops->timer)
1233                         xprt->ops->timer(xprt, task);
1234         } else
1235                 task->tk_status = 0;
1236 }
1237
1238 /**
1239  * xprt_wait_for_reply_request_def - wait for reply
1240  * @task: pointer to rpc_task
1241  *
1242  * Set a request's retransmit timeout based on the transport's
1243  * default timeout parameters.  Used by transports that don't adjust
1244  * the retransmit timeout based on round-trip time estimation,
1245  * and put the task to sleep on the pending queue.
1246  */
1247 void xprt_wait_for_reply_request_def(struct rpc_task *task)
1248 {
1249         struct rpc_rqst *req = task->tk_rqstp;
1250
1251         rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1252                         xprt_request_timeout(req));
1253 }
1254 EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_def);
1255
1256 /**
1257  * xprt_wait_for_reply_request_rtt - wait for reply using RTT estimator
1258  * @task: pointer to rpc_task
1259  *
1260  * Set a request's retransmit timeout using the RTT estimator,
1261  * and put the task to sleep on the pending queue.
1262  */
1263 void xprt_wait_for_reply_request_rtt(struct rpc_task *task)
1264 {
1265         int timer = task->tk_msg.rpc_proc->p_timer;
1266         struct rpc_clnt *clnt = task->tk_client;
1267         struct rpc_rtt *rtt = clnt->cl_rtt;
1268         struct rpc_rqst *req = task->tk_rqstp;
1269         unsigned long max_timeout = clnt->cl_timeout->to_maxval;
1270         unsigned long timeout;
1271
1272         timeout = rpc_calc_rto(rtt, timer);
1273         timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
1274         if (timeout > max_timeout || timeout == 0)
1275                 timeout = max_timeout;
1276         rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1277                         jiffies + timeout);
1278 }
1279 EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_rtt);
1280
1281 /**
1282  * xprt_request_wait_receive - wait for the reply to an RPC request
1283  * @task: RPC task about to send a request
1284  *
1285  */
1286 void xprt_request_wait_receive(struct rpc_task *task)
1287 {
1288         struct rpc_rqst *req = task->tk_rqstp;
1289         struct rpc_xprt *xprt = req->rq_xprt;
1290
1291         if (!test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1292                 return;
1293         /*
1294          * Sleep on the pending queue if we're expecting a reply.
1295          * The spinlock ensures atomicity between the test of
1296          * req->rq_reply_bytes_recvd, and the call to rpc_sleep_on().
1297          */
1298         spin_lock(&xprt->queue_lock);
1299         if (test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate)) {
1300                 xprt->ops->wait_for_reply_request(task);
1301                 /*
1302                  * Send an extra queue wakeup call if the
1303                  * connection was dropped in case the call to
1304                  * rpc_sleep_on() raced.
1305                  */
1306                 if (xprt_request_retransmit_after_disconnect(task))
1307                         rpc_wake_up_queued_task_set_status(&xprt->pending,
1308                                         task, -ENOTCONN);
1309         }
1310         spin_unlock(&xprt->queue_lock);
1311 }
1312
1313 static bool
1314 xprt_request_need_enqueue_transmit(struct rpc_task *task, struct rpc_rqst *req)
1315 {
1316         return !test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1317 }
1318
1319 /**
1320  * xprt_request_enqueue_transmit - queue a task for transmission
1321  * @task: pointer to rpc_task
1322  *
1323  * Add a task to the transmission queue.
1324  */
1325 void
1326 xprt_request_enqueue_transmit(struct rpc_task *task)
1327 {
1328         struct rpc_rqst *pos, *req = task->tk_rqstp;
1329         struct rpc_xprt *xprt = req->rq_xprt;
1330
1331         if (xprt_request_need_enqueue_transmit(task, req)) {
1332                 req->rq_bytes_sent = 0;
1333                 spin_lock(&xprt->queue_lock);
1334                 /*
1335                  * Requests that carry congestion control credits are added
1336                  * to the head of the list to avoid starvation issues.
1337                  */
1338                 if (req->rq_cong) {
1339                         xprt_clear_congestion_window_wait(xprt);
1340                         list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1341                                 if (pos->rq_cong)
1342                                         continue;
1343                                 /* Note: req is added _before_ pos */
1344                                 list_add_tail(&req->rq_xmit, &pos->rq_xmit);
1345                                 INIT_LIST_HEAD(&req->rq_xmit2);
1346                                 goto out;
1347                         }
1348                 } else if (RPC_IS_SWAPPER(task)) {
1349                         list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1350                                 if (pos->rq_cong || pos->rq_bytes_sent)
1351                                         continue;
1352                                 if (RPC_IS_SWAPPER(pos->rq_task))
1353                                         continue;
1354                                 /* Note: req is added _before_ pos */
1355                                 list_add_tail(&req->rq_xmit, &pos->rq_xmit);
1356                                 INIT_LIST_HEAD(&req->rq_xmit2);
1357                                 goto out;
1358                         }
1359                 } else if (!req->rq_seqno) {
1360                         list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1361                                 if (pos->rq_task->tk_owner != task->tk_owner)
1362                                         continue;
1363                                 list_add_tail(&req->rq_xmit2, &pos->rq_xmit2);
1364                                 INIT_LIST_HEAD(&req->rq_xmit);
1365                                 goto out;
1366                         }
1367                 }
1368                 list_add_tail(&req->rq_xmit, &xprt->xmit_queue);
1369                 INIT_LIST_HEAD(&req->rq_xmit2);
1370 out:
1371                 atomic_long_inc(&xprt->xmit_queuelen);
1372                 set_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1373                 spin_unlock(&xprt->queue_lock);
1374         }
1375 }
1376
1377 /**
1378  * xprt_request_dequeue_transmit_locked - remove a task from the transmission queue
1379  * @task: pointer to rpc_task
1380  *
1381  * Remove a task from the transmission queue
1382  * Caller must hold xprt->queue_lock
1383  */
1384 static void
1385 xprt_request_dequeue_transmit_locked(struct rpc_task *task)
1386 {
1387         struct rpc_rqst *req = task->tk_rqstp;
1388
1389         if (!test_and_clear_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1390                 return;
1391         if (!list_empty(&req->rq_xmit)) {
1392                 list_del(&req->rq_xmit);
1393                 if (!list_empty(&req->rq_xmit2)) {
1394                         struct rpc_rqst *next = list_first_entry(&req->rq_xmit2,
1395                                         struct rpc_rqst, rq_xmit2);
1396                         list_del(&req->rq_xmit2);
1397                         list_add_tail(&next->rq_xmit, &next->rq_xprt->xmit_queue);
1398                 }
1399         } else
1400                 list_del(&req->rq_xmit2);
1401         atomic_long_dec(&req->rq_xprt->xmit_queuelen);
1402 }
1403
1404 /**
1405  * xprt_request_dequeue_transmit - remove a task from the transmission queue
1406  * @task: pointer to rpc_task
1407  *
1408  * Remove a task from the transmission queue
1409  */
1410 static void
1411 xprt_request_dequeue_transmit(struct rpc_task *task)
1412 {
1413         struct rpc_rqst *req = task->tk_rqstp;
1414         struct rpc_xprt *xprt = req->rq_xprt;
1415
1416         spin_lock(&xprt->queue_lock);
1417         xprt_request_dequeue_transmit_locked(task);
1418         spin_unlock(&xprt->queue_lock);
1419 }
1420
1421 /**
1422  * xprt_request_dequeue_xprt - remove a task from the transmit+receive queue
1423  * @task: pointer to rpc_task
1424  *
1425  * Remove a task from the transmit and receive queues, and ensure that
1426  * it is not pinned by the receive work item.
1427  */
1428 void
1429 xprt_request_dequeue_xprt(struct rpc_task *task)
1430 {
1431         struct rpc_rqst *req = task->tk_rqstp;
1432         struct rpc_xprt *xprt = req->rq_xprt;
1433
1434         if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate) ||
1435             test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) ||
1436             xprt_is_pinned_rqst(req)) {
1437                 spin_lock(&xprt->queue_lock);
1438                 xprt_request_dequeue_transmit_locked(task);
1439                 xprt_request_dequeue_receive_locked(task);
1440                 while (xprt_is_pinned_rqst(req)) {
1441                         set_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1442                         spin_unlock(&xprt->queue_lock);
1443                         xprt_wait_on_pinned_rqst(req);
1444                         spin_lock(&xprt->queue_lock);
1445                         clear_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1446                 }
1447                 spin_unlock(&xprt->queue_lock);
1448         }
1449 }
1450
1451 /**
1452  * xprt_request_prepare - prepare an encoded request for transport
1453  * @req: pointer to rpc_rqst
1454  *
1455  * Calls into the transport layer to do whatever is needed to prepare
1456  * the request for transmission or receive.
1457  */
1458 void
1459 xprt_request_prepare(struct rpc_rqst *req)
1460 {
1461         struct rpc_xprt *xprt = req->rq_xprt;
1462
1463         if (xprt->ops->prepare_request)
1464                 xprt->ops->prepare_request(req);
1465 }
1466
1467 /**
1468  * xprt_request_need_retransmit - Test if a task needs retransmission
1469  * @task: pointer to rpc_task
1470  *
1471  * Test for whether a connection breakage requires the task to retransmit
1472  */
1473 bool
1474 xprt_request_need_retransmit(struct rpc_task *task)
1475 {
1476         return xprt_request_retransmit_after_disconnect(task);
1477 }
1478
1479 /**
1480  * xprt_prepare_transmit - reserve the transport before sending a request
1481  * @task: RPC task about to send a request
1482  *
1483  */
1484 bool xprt_prepare_transmit(struct rpc_task *task)
1485 {
1486         struct rpc_rqst *req = task->tk_rqstp;
1487         struct rpc_xprt *xprt = req->rq_xprt;
1488
1489         if (!xprt_lock_write(xprt, task)) {
1490                 /* Race breaker: someone may have transmitted us */
1491                 if (!test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1492                         rpc_wake_up_queued_task_set_status(&xprt->sending,
1493                                         task, 0);
1494                 return false;
1495
1496         }
1497         return true;
1498 }
1499
1500 void xprt_end_transmit(struct rpc_task *task)
1501 {
1502         struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
1503
1504         xprt_inject_disconnect(xprt);
1505         xprt_release_write(xprt, task);
1506 }
1507
1508 /**
1509  * xprt_request_transmit - send an RPC request on a transport
1510  * @req: pointer to request to transmit
1511  * @snd_task: RPC task that owns the transport lock
1512  *
1513  * This performs the transmission of a single request.
1514  * Note that if the request is not the same as snd_task, then it
1515  * does need to be pinned.
1516  * Returns '0' on success.
1517  */
1518 static int
1519 xprt_request_transmit(struct rpc_rqst *req, struct rpc_task *snd_task)
1520 {
1521         struct rpc_xprt *xprt = req->rq_xprt;
1522         struct rpc_task *task = req->rq_task;
1523         unsigned int connect_cookie;
1524         int is_retrans = RPC_WAS_SENT(task);
1525         int status;
1526
1527         if (!req->rq_bytes_sent) {
1528                 if (xprt_request_data_received(task)) {
1529                         status = 0;
1530                         goto out_dequeue;
1531                 }
1532                 /* Verify that our message lies in the RPCSEC_GSS window */
1533                 if (rpcauth_xmit_need_reencode(task)) {
1534                         status = -EBADMSG;
1535                         goto out_dequeue;
1536                 }
1537                 if (RPC_SIGNALLED(task)) {
1538                         status = -ERESTARTSYS;
1539                         goto out_dequeue;
1540                 }
1541         }
1542
1543         /*
1544          * Update req->rq_ntrans before transmitting to avoid races with
1545          * xprt_update_rtt(), which needs to know that it is recording a
1546          * reply to the first transmission.
1547          */
1548         req->rq_ntrans++;
1549
1550         trace_rpc_xdr_sendto(task, &req->rq_snd_buf);
1551         connect_cookie = xprt->connect_cookie;
1552         status = xprt->ops->send_request(req);
1553         if (status != 0) {
1554                 req->rq_ntrans--;
1555                 trace_xprt_transmit(req, status);
1556                 return status;
1557         }
1558
1559         if (is_retrans) {
1560                 task->tk_client->cl_stats->rpcretrans++;
1561                 trace_xprt_retransmit(req);
1562         }
1563
1564         xprt_inject_disconnect(xprt);
1565
1566         task->tk_flags |= RPC_TASK_SENT;
1567         spin_lock(&xprt->transport_lock);
1568
1569         xprt->stat.sends++;
1570         xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
1571         xprt->stat.bklog_u += xprt->backlog.qlen;
1572         xprt->stat.sending_u += xprt->sending.qlen;
1573         xprt->stat.pending_u += xprt->pending.qlen;
1574         spin_unlock(&xprt->transport_lock);
1575
1576         req->rq_connect_cookie = connect_cookie;
1577 out_dequeue:
1578         trace_xprt_transmit(req, status);
1579         xprt_request_dequeue_transmit(task);
1580         rpc_wake_up_queued_task_set_status(&xprt->sending, task, status);
1581         return status;
1582 }
1583
1584 /**
1585  * xprt_transmit - send an RPC request on a transport
1586  * @task: controlling RPC task
1587  *
1588  * Attempts to drain the transmit queue. On exit, either the transport
1589  * signalled an error that needs to be handled before transmission can
1590  * resume, or @task finished transmitting, and detected that it already
1591  * received a reply.
1592  */
1593 void
1594 xprt_transmit(struct rpc_task *task)
1595 {
1596         struct rpc_rqst *next, *req = task->tk_rqstp;
1597         struct rpc_xprt *xprt = req->rq_xprt;
1598         int counter, status;
1599
1600         spin_lock(&xprt->queue_lock);
1601         counter = 0;
1602         while (!list_empty(&xprt->xmit_queue)) {
1603                 if (++counter == 20)
1604                         break;
1605                 next = list_first_entry(&xprt->xmit_queue,
1606                                 struct rpc_rqst, rq_xmit);
1607                 xprt_pin_rqst(next);
1608                 spin_unlock(&xprt->queue_lock);
1609                 status = xprt_request_transmit(next, task);
1610                 if (status == -EBADMSG && next != req)
1611                         status = 0;
1612                 spin_lock(&xprt->queue_lock);
1613                 xprt_unpin_rqst(next);
1614                 if (status == 0) {
1615                         if (!xprt_request_data_received(task) ||
1616                             test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1617                                 continue;
1618                 } else if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1619                         task->tk_status = status;
1620                 break;
1621         }
1622         spin_unlock(&xprt->queue_lock);
1623 }
1624
1625 static void xprt_complete_request_init(struct rpc_task *task)
1626 {
1627         if (task->tk_rqstp)
1628                 xprt_request_init(task);
1629 }
1630
1631 void xprt_add_backlog(struct rpc_xprt *xprt, struct rpc_task *task)
1632 {
1633         set_bit(XPRT_CONGESTED, &xprt->state);
1634         rpc_sleep_on(&xprt->backlog, task, xprt_complete_request_init);
1635 }
1636 EXPORT_SYMBOL_GPL(xprt_add_backlog);
1637
1638 static bool __xprt_set_rq(struct rpc_task *task, void *data)
1639 {
1640         struct rpc_rqst *req = data;
1641
1642         if (task->tk_rqstp == NULL) {
1643                 memset(req, 0, sizeof(*req));   /* mark unused */
1644                 task->tk_rqstp = req;
1645                 return true;
1646         }
1647         return false;
1648 }
1649
1650 bool xprt_wake_up_backlog(struct rpc_xprt *xprt, struct rpc_rqst *req)
1651 {
1652         if (rpc_wake_up_first(&xprt->backlog, __xprt_set_rq, req) == NULL) {
1653                 clear_bit(XPRT_CONGESTED, &xprt->state);
1654                 return false;
1655         }
1656         return true;
1657 }
1658 EXPORT_SYMBOL_GPL(xprt_wake_up_backlog);
1659
1660 static bool xprt_throttle_congested(struct rpc_xprt *xprt, struct rpc_task *task)
1661 {
1662         bool ret = false;
1663
1664         if (!test_bit(XPRT_CONGESTED, &xprt->state))
1665                 goto out;
1666         spin_lock(&xprt->reserve_lock);
1667         if (test_bit(XPRT_CONGESTED, &xprt->state)) {
1668                 xprt_add_backlog(xprt, task);
1669                 ret = true;
1670         }
1671         spin_unlock(&xprt->reserve_lock);
1672 out:
1673         return ret;
1674 }
1675
1676 static struct rpc_rqst *xprt_dynamic_alloc_slot(struct rpc_xprt *xprt)
1677 {
1678         struct rpc_rqst *req = ERR_PTR(-EAGAIN);
1679
1680         if (xprt->num_reqs >= xprt->max_reqs)
1681                 goto out;
1682         ++xprt->num_reqs;
1683         spin_unlock(&xprt->reserve_lock);
1684         req = kzalloc(sizeof(struct rpc_rqst), GFP_NOFS);
1685         spin_lock(&xprt->reserve_lock);
1686         if (req != NULL)
1687                 goto out;
1688         --xprt->num_reqs;
1689         req = ERR_PTR(-ENOMEM);
1690 out:
1691         return req;
1692 }
1693
1694 static bool xprt_dynamic_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1695 {
1696         if (xprt->num_reqs > xprt->min_reqs) {
1697                 --xprt->num_reqs;
1698                 kfree(req);
1699                 return true;
1700         }
1701         return false;
1702 }
1703
1704 void xprt_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1705 {
1706         struct rpc_rqst *req;
1707
1708         spin_lock(&xprt->reserve_lock);
1709         if (!list_empty(&xprt->free)) {
1710                 req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
1711                 list_del(&req->rq_list);
1712                 goto out_init_req;
1713         }
1714         req = xprt_dynamic_alloc_slot(xprt);
1715         if (!IS_ERR(req))
1716                 goto out_init_req;
1717         switch (PTR_ERR(req)) {
1718         case -ENOMEM:
1719                 dprintk("RPC:       dynamic allocation of request slot "
1720                                 "failed! Retrying\n");
1721                 task->tk_status = -ENOMEM;
1722                 break;
1723         case -EAGAIN:
1724                 xprt_add_backlog(xprt, task);
1725                 dprintk("RPC:       waiting for request slot\n");
1726                 fallthrough;
1727         default:
1728                 task->tk_status = -EAGAIN;
1729         }
1730         spin_unlock(&xprt->reserve_lock);
1731         return;
1732 out_init_req:
1733         xprt->stat.max_slots = max_t(unsigned int, xprt->stat.max_slots,
1734                                      xprt->num_reqs);
1735         spin_unlock(&xprt->reserve_lock);
1736
1737         task->tk_status = 0;
1738         task->tk_rqstp = req;
1739 }
1740 EXPORT_SYMBOL_GPL(xprt_alloc_slot);
1741
1742 void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1743 {
1744         spin_lock(&xprt->reserve_lock);
1745         if (!xprt_wake_up_backlog(xprt, req) &&
1746             !xprt_dynamic_free_slot(xprt, req)) {
1747                 memset(req, 0, sizeof(*req));   /* mark unused */
1748                 list_add(&req->rq_list, &xprt->free);
1749         }
1750         spin_unlock(&xprt->reserve_lock);
1751 }
1752 EXPORT_SYMBOL_GPL(xprt_free_slot);
1753
1754 static void xprt_free_all_slots(struct rpc_xprt *xprt)
1755 {
1756         struct rpc_rqst *req;
1757         while (!list_empty(&xprt->free)) {
1758                 req = list_first_entry(&xprt->free, struct rpc_rqst, rq_list);
1759                 list_del(&req->rq_list);
1760                 kfree(req);
1761         }
1762 }
1763
1764 static DEFINE_IDA(rpc_xprt_ids);
1765
1766 void xprt_cleanup_ids(void)
1767 {
1768         ida_destroy(&rpc_xprt_ids);
1769 }
1770
1771 static int xprt_alloc_id(struct rpc_xprt *xprt)
1772 {
1773         int id;
1774
1775         id = ida_simple_get(&rpc_xprt_ids, 0, 0, GFP_KERNEL);
1776         if (id < 0)
1777                 return id;
1778
1779         xprt->id = id;
1780         return 0;
1781 }
1782
1783 static void xprt_free_id(struct rpc_xprt *xprt)
1784 {
1785         ida_simple_remove(&rpc_xprt_ids, xprt->id);
1786 }
1787
1788 struct rpc_xprt *xprt_alloc(struct net *net, size_t size,
1789                 unsigned int num_prealloc,
1790                 unsigned int max_alloc)
1791 {
1792         struct rpc_xprt *xprt;
1793         struct rpc_rqst *req;
1794         int i;
1795
1796         xprt = kzalloc(size, GFP_KERNEL);
1797         if (xprt == NULL)
1798                 goto out;
1799
1800         xprt_alloc_id(xprt);
1801         xprt_init(xprt, net);
1802
1803         for (i = 0; i < num_prealloc; i++) {
1804                 req = kzalloc(sizeof(struct rpc_rqst), GFP_KERNEL);
1805                 if (!req)
1806                         goto out_free;
1807                 list_add(&req->rq_list, &xprt->free);
1808         }
1809         if (max_alloc > num_prealloc)
1810                 xprt->max_reqs = max_alloc;
1811         else
1812                 xprt->max_reqs = num_prealloc;
1813         xprt->min_reqs = num_prealloc;
1814         xprt->num_reqs = num_prealloc;
1815
1816         return xprt;
1817
1818 out_free:
1819         xprt_free(xprt);
1820 out:
1821         return NULL;
1822 }
1823 EXPORT_SYMBOL_GPL(xprt_alloc);
1824
1825 void xprt_free(struct rpc_xprt *xprt)
1826 {
1827         put_net(xprt->xprt_net);
1828         xprt_free_all_slots(xprt);
1829         xprt_free_id(xprt);
1830         rpc_sysfs_xprt_destroy(xprt);
1831         kfree_rcu(xprt, rcu);
1832 }
1833 EXPORT_SYMBOL_GPL(xprt_free);
1834
1835 static void
1836 xprt_init_connect_cookie(struct rpc_rqst *req, struct rpc_xprt *xprt)
1837 {
1838         req->rq_connect_cookie = xprt_connect_cookie(xprt) - 1;
1839 }
1840
1841 static __be32
1842 xprt_alloc_xid(struct rpc_xprt *xprt)
1843 {
1844         __be32 xid;
1845
1846         spin_lock(&xprt->reserve_lock);
1847         xid = (__force __be32)xprt->xid++;
1848         spin_unlock(&xprt->reserve_lock);
1849         return xid;
1850 }
1851
1852 static void
1853 xprt_init_xid(struct rpc_xprt *xprt)
1854 {
1855         xprt->xid = prandom_u32();
1856 }
1857
1858 static void
1859 xprt_request_init(struct rpc_task *task)
1860 {
1861         struct rpc_xprt *xprt = task->tk_xprt;
1862         struct rpc_rqst *req = task->tk_rqstp;
1863
1864         req->rq_task    = task;
1865         req->rq_xprt    = xprt;
1866         req->rq_buffer  = NULL;
1867         req->rq_xid     = xprt_alloc_xid(xprt);
1868         xprt_init_connect_cookie(req, xprt);
1869         req->rq_snd_buf.len = 0;
1870         req->rq_snd_buf.buflen = 0;
1871         req->rq_rcv_buf.len = 0;
1872         req->rq_rcv_buf.buflen = 0;
1873         req->rq_snd_buf.bvec = NULL;
1874         req->rq_rcv_buf.bvec = NULL;
1875         req->rq_release_snd_buf = NULL;
1876         xprt_init_majortimeo(task, req);
1877
1878         trace_xprt_reserve(req);
1879 }
1880
1881 static void
1882 xprt_do_reserve(struct rpc_xprt *xprt, struct rpc_task *task)
1883 {
1884         xprt->ops->alloc_slot(xprt, task);
1885         if (task->tk_rqstp != NULL)
1886                 xprt_request_init(task);
1887 }
1888
1889 /**
1890  * xprt_reserve - allocate an RPC request slot
1891  * @task: RPC task requesting a slot allocation
1892  *
1893  * If the transport is marked as being congested, or if no more
1894  * slots are available, place the task on the transport's
1895  * backlog queue.
1896  */
1897 void xprt_reserve(struct rpc_task *task)
1898 {
1899         struct rpc_xprt *xprt = task->tk_xprt;
1900
1901         task->tk_status = 0;
1902         if (task->tk_rqstp != NULL)
1903                 return;
1904
1905         task->tk_status = -EAGAIN;
1906         if (!xprt_throttle_congested(xprt, task))
1907                 xprt_do_reserve(xprt, task);
1908 }
1909
1910 /**
1911  * xprt_retry_reserve - allocate an RPC request slot
1912  * @task: RPC task requesting a slot allocation
1913  *
1914  * If no more slots are available, place the task on the transport's
1915  * backlog queue.
1916  * Note that the only difference with xprt_reserve is that we now
1917  * ignore the value of the XPRT_CONGESTED flag.
1918  */
1919 void xprt_retry_reserve(struct rpc_task *task)
1920 {
1921         struct rpc_xprt *xprt = task->tk_xprt;
1922
1923         task->tk_status = 0;
1924         if (task->tk_rqstp != NULL)
1925                 return;
1926
1927         task->tk_status = -EAGAIN;
1928         xprt_do_reserve(xprt, task);
1929 }
1930
1931 /**
1932  * xprt_release - release an RPC request slot
1933  * @task: task which is finished with the slot
1934  *
1935  */
1936 void xprt_release(struct rpc_task *task)
1937 {
1938         struct rpc_xprt *xprt;
1939         struct rpc_rqst *req = task->tk_rqstp;
1940
1941         if (req == NULL) {
1942                 if (task->tk_client) {
1943                         xprt = task->tk_xprt;
1944                         xprt_release_write(xprt, task);
1945                 }
1946                 return;
1947         }
1948
1949         xprt = req->rq_xprt;
1950         xprt_request_dequeue_xprt(task);
1951         spin_lock(&xprt->transport_lock);
1952         xprt->ops->release_xprt(xprt, task);
1953         if (xprt->ops->release_request)
1954                 xprt->ops->release_request(task);
1955         xprt_schedule_autodisconnect(xprt);
1956         spin_unlock(&xprt->transport_lock);
1957         if (req->rq_buffer)
1958                 xprt->ops->buf_free(task);
1959         xdr_free_bvec(&req->rq_rcv_buf);
1960         xdr_free_bvec(&req->rq_snd_buf);
1961         if (req->rq_cred != NULL)
1962                 put_rpccred(req->rq_cred);
1963         if (req->rq_release_snd_buf)
1964                 req->rq_release_snd_buf(req);
1965
1966         task->tk_rqstp = NULL;
1967         if (likely(!bc_prealloc(req)))
1968                 xprt->ops->free_slot(xprt, req);
1969         else
1970                 xprt_free_bc_request(req);
1971 }
1972
1973 #ifdef CONFIG_SUNRPC_BACKCHANNEL
1974 void
1975 xprt_init_bc_request(struct rpc_rqst *req, struct rpc_task *task)
1976 {
1977         struct xdr_buf *xbufp = &req->rq_snd_buf;
1978
1979         task->tk_rqstp = req;
1980         req->rq_task = task;
1981         xprt_init_connect_cookie(req, req->rq_xprt);
1982         /*
1983          * Set up the xdr_buf length.
1984          * This also indicates that the buffer is XDR encoded already.
1985          */
1986         xbufp->len = xbufp->head[0].iov_len + xbufp->page_len +
1987                 xbufp->tail[0].iov_len;
1988 }
1989 #endif
1990
1991 static void xprt_init(struct rpc_xprt *xprt, struct net *net)
1992 {
1993         kref_init(&xprt->kref);
1994
1995         spin_lock_init(&xprt->transport_lock);
1996         spin_lock_init(&xprt->reserve_lock);
1997         spin_lock_init(&xprt->queue_lock);
1998
1999         INIT_LIST_HEAD(&xprt->free);
2000         xprt->recv_queue = RB_ROOT;
2001         INIT_LIST_HEAD(&xprt->xmit_queue);
2002 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
2003         spin_lock_init(&xprt->bc_pa_lock);
2004         INIT_LIST_HEAD(&xprt->bc_pa_list);
2005 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
2006         INIT_LIST_HEAD(&xprt->xprt_switch);
2007
2008         xprt->last_used = jiffies;
2009         xprt->cwnd = RPC_INITCWND;
2010         xprt->bind_index = 0;
2011
2012         rpc_init_wait_queue(&xprt->binding, "xprt_binding");
2013         rpc_init_wait_queue(&xprt->pending, "xprt_pending");
2014         rpc_init_wait_queue(&xprt->sending, "xprt_sending");
2015         rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
2016
2017         xprt_init_xid(xprt);
2018
2019         xprt->xprt_net = get_net(net);
2020 }
2021
2022 /**
2023  * xprt_create_transport - create an RPC transport
2024  * @args: rpc transport creation arguments
2025  *
2026  */
2027 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
2028 {
2029         struct rpc_xprt *xprt;
2030         const struct xprt_class *t;
2031
2032         t = xprt_class_find_by_ident(args->ident);
2033         if (!t) {
2034                 dprintk("RPC: transport (%d) not supported\n", args->ident);
2035                 return ERR_PTR(-EIO);
2036         }
2037
2038         xprt = t->setup(args);
2039         xprt_class_release(t);
2040
2041         if (IS_ERR(xprt))
2042                 goto out;
2043         if (args->flags & XPRT_CREATE_NO_IDLE_TIMEOUT)
2044                 xprt->idle_timeout = 0;
2045         INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
2046         if (xprt_has_timer(xprt))
2047                 timer_setup(&xprt->timer, xprt_init_autodisconnect, 0);
2048         else
2049                 timer_setup(&xprt->timer, NULL, 0);
2050
2051         if (strlen(args->servername) > RPC_MAXNETNAMELEN) {
2052                 xprt_destroy(xprt);
2053                 return ERR_PTR(-EINVAL);
2054         }
2055         xprt->servername = kstrdup(args->servername, GFP_KERNEL);
2056         if (xprt->servername == NULL) {
2057                 xprt_destroy(xprt);
2058                 return ERR_PTR(-ENOMEM);
2059         }
2060
2061         rpc_xprt_debugfs_register(xprt);
2062
2063         trace_xprt_create(xprt);
2064 out:
2065         return xprt;
2066 }
2067
2068 static void xprt_destroy_cb(struct work_struct *work)
2069 {
2070         struct rpc_xprt *xprt =
2071                 container_of(work, struct rpc_xprt, task_cleanup);
2072
2073         trace_xprt_destroy(xprt);
2074
2075         rpc_xprt_debugfs_unregister(xprt);
2076         rpc_destroy_wait_queue(&xprt->binding);
2077         rpc_destroy_wait_queue(&xprt->pending);
2078         rpc_destroy_wait_queue(&xprt->sending);
2079         rpc_destroy_wait_queue(&xprt->backlog);
2080         kfree(xprt->servername);
2081         /*
2082          * Destroy any existing back channel
2083          */
2084         xprt_destroy_backchannel(xprt, UINT_MAX);
2085
2086         /*
2087          * Tear down transport state and free the rpc_xprt
2088          */
2089         xprt->ops->destroy(xprt);
2090 }
2091
2092 /**
2093  * xprt_destroy - destroy an RPC transport, killing off all requests.
2094  * @xprt: transport to destroy
2095  *
2096  */
2097 static void xprt_destroy(struct rpc_xprt *xprt)
2098 {
2099         /*
2100          * Exclude transport connect/disconnect handlers and autoclose
2101          */
2102         wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_UNINTERRUPTIBLE);
2103
2104         del_timer_sync(&xprt->timer);
2105
2106         /*
2107          * Destroy sockets etc from the system workqueue so they can
2108          * safely flush receive work running on rpciod.
2109          */
2110         INIT_WORK(&xprt->task_cleanup, xprt_destroy_cb);
2111         schedule_work(&xprt->task_cleanup);
2112 }
2113
2114 static void xprt_destroy_kref(struct kref *kref)
2115 {
2116         xprt_destroy(container_of(kref, struct rpc_xprt, kref));
2117 }
2118
2119 /**
2120  * xprt_get - return a reference to an RPC transport.
2121  * @xprt: pointer to the transport
2122  *
2123  */
2124 struct rpc_xprt *xprt_get(struct rpc_xprt *xprt)
2125 {
2126         if (xprt != NULL && kref_get_unless_zero(&xprt->kref))
2127                 return xprt;
2128         return NULL;
2129 }
2130 EXPORT_SYMBOL_GPL(xprt_get);
2131
2132 /**
2133  * xprt_put - release a reference to an RPC transport.
2134  * @xprt: pointer to the transport
2135  *
2136  */
2137 void xprt_put(struct rpc_xprt *xprt)
2138 {
2139         if (xprt != NULL)
2140                 kref_put(&xprt->kref, xprt_destroy_kref);
2141 }
2142 EXPORT_SYMBOL_GPL(xprt_put);