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                 clear_bit_unlock(XPRT_LOCKED, &xprt->state);
 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 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         xprt->connect_cookie++;
 739         smp_mb__before_atomic();
 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_schedule_autoclose_locked - Try to schedule an autoclose RPC call
 766  * @xprt: transport to disconnect
 767  */
 768 static void xprt_schedule_autoclose_locked(struct rpc_xprt *xprt)
 769 {
 770         if (test_and_set_bit(XPRT_CLOSE_WAIT, &xprt->state))
 771                 return;
 772         if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 773                 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 774         else if (xprt->snd_task && !test_bit(XPRT_SND_IS_COOKIE, &xprt->state))
 775                 rpc_wake_up_queued_task_set_status(&xprt->pending,
 776                                                    xprt->snd_task, -ENOTCONN);
 777 }
 778
 779 /**
 780  * xprt_force_disconnect - force a transport to disconnect
 781  * @xprt: transport to disconnect
 782  *
 783  */
 784 void xprt_force_disconnect(struct rpc_xprt *xprt)
 785 {
 786         trace_xprt_disconnect_force(xprt);
 787
 788         /* Don't race with the test_bit() in xprt_clear_locked() */
 789         spin_lock(&xprt->transport_lock);
 790         xprt_schedule_autoclose_locked(xprt);
 791         spin_unlock(&xprt->transport_lock);
 792 }
 793 EXPORT_SYMBOL_GPL(xprt_force_disconnect);
 794
 795 static unsigned int
 796 xprt_connect_cookie(struct rpc_xprt *xprt)
 797 {
 798         return READ_ONCE(xprt->connect_cookie);
 799 }
 800
 801 static bool
 802 xprt_request_retransmit_after_disconnect(struct rpc_task *task)
 803 {
 804         struct rpc_rqst *req = task->tk_rqstp;
 805         struct rpc_xprt *xprt = req->rq_xprt;
 806
 807         return req->rq_connect_cookie != xprt_connect_cookie(xprt) ||
 808                 !xprt_connected(xprt);
 809 }
 810
 811 /**
 812  * xprt_conditional_disconnect - force a transport to disconnect
 813  * @xprt: transport to disconnect
 814  * @cookie: 'connection cookie'
 815  *
 816  * This attempts to break the connection if and only if 'cookie' matches
 817  * the current transport 'connection cookie'. It ensures that we don't
 818  * try to break the connection more than once when we need to retransmit
 819  * a batch of RPC requests.
 820  *
 821  */
 822 void xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie)
 823 {
 824         /* Don't race with the test_bit() in xprt_clear_locked() */
 825         spin_lock(&xprt->transport_lock);
 826         if (cookie != xprt->connect_cookie)
 827                 goto out;
 828         if (test_bit(XPRT_CLOSING, &xprt->state))
 829                 goto out;
 830         xprt_schedule_autoclose_locked(xprt);
 831 out:
 832         spin_unlock(&xprt->transport_lock);
 833 }
 834
 835 static bool
 836 xprt_has_timer(const struct rpc_xprt *xprt)
 837 {
 838         return xprt->idle_timeout != 0;
 839 }
 840
 841 static void
 842 xprt_schedule_autodisconnect(struct rpc_xprt *xprt)
 843         __must_hold(&xprt->transport_lock)
 844 {
 845         xprt->last_used = jiffies;
 846         if (RB_EMPTY_ROOT(&xprt->recv_queue) && xprt_has_timer(xprt))
 847                 mod_timer(&xprt->timer, xprt->last_used + xprt->idle_timeout);
 848 }
 849
 850 static void
 851 xprt_init_autodisconnect(struct timer_list *t)
 852 {
 853         struct rpc_xprt *xprt = from_timer(xprt, t, timer);
 854
 855         if (!RB_EMPTY_ROOT(&xprt->recv_queue))
 856                 return;
 857         /* Reset xprt->last_used to avoid connect/autodisconnect cycling */
 858         xprt->last_used = jiffies;
 859         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 860                 return;
 861         queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 862 }
 863
 864 #if IS_ENABLED(CONFIG_FAIL_SUNRPC)
 865 static void xprt_inject_disconnect(struct rpc_xprt *xprt)
 866 {
 867         if (!fail_sunrpc.ignore_client_disconnect &&
 868             should_fail(&fail_sunrpc.attr, 1))
 869                 xprt->ops->inject_disconnect(xprt);
 870 }
 871 #else
 872 static inline void xprt_inject_disconnect(struct rpc_xprt *xprt)
 873 {
 874 }
 875 #endif
 876
 877 bool xprt_lock_connect(struct rpc_xprt *xprt,
 878                 struct rpc_task *task,
 879                 void *cookie)
 880 {
 881         bool ret = false;
 882
 883         spin_lock(&xprt->transport_lock);
 884         if (!test_bit(XPRT_LOCKED, &xprt->state))
 885                 goto out;
 886         if (xprt->snd_task != task)
 887                 goto out;
 888         set_bit(XPRT_SND_IS_COOKIE, &xprt->state);
 889         xprt->snd_task = cookie;
 890         ret = true;
 891 out:
 892         spin_unlock(&xprt->transport_lock);
 893         return ret;
 894 }
 895 EXPORT_SYMBOL_GPL(xprt_lock_connect);
 896
 897 void xprt_unlock_connect(struct rpc_xprt *xprt, void *cookie)
 898 {
 899         spin_lock(&xprt->transport_lock);
 900         if (xprt->snd_task != cookie)
 901                 goto out;
 902         if (!test_bit(XPRT_LOCKED, &xprt->state))
 903                 goto out;
 904         xprt->snd_task =NULL;
 905         clear_bit(XPRT_SND_IS_COOKIE, &xprt->state);
 906         xprt->ops->release_xprt(xprt, NULL);
 907         xprt_schedule_autodisconnect(xprt);
 908 out:
 909         spin_unlock(&xprt->transport_lock);
 910         wake_up_bit(&xprt->state, XPRT_LOCKED);
 911 }
 912 EXPORT_SYMBOL_GPL(xprt_unlock_connect);
 913
 914 /**
 915  * xprt_connect - schedule a transport connect operation
 916  * @task: RPC task that is requesting the connect
 917  *
 918  */
 919 void xprt_connect(struct rpc_task *task)
 920 {
 921         struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
 922
 923         trace_xprt_connect(xprt);
 924
 925         if (!xprt_bound(xprt)) {
 926                 task->tk_status = -EAGAIN;
 927                 return;
 928         }
 929         if (!xprt_lock_write(xprt, task))
 930                 return;
 931
 932         if (test_and_clear_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
 933                 trace_xprt_disconnect_cleanup(xprt);
 934                 xprt->ops->close(xprt);
 935         }
 936
 937         if (!xprt_connected(xprt)) {
 938                 task->tk_rqstp->rq_connect_cookie = xprt->connect_cookie;
 939                 rpc_sleep_on_timeout(&xprt->pending, task, NULL,
 940                                 xprt_request_timeout(task->tk_rqstp));
 941
 942                 if (test_bit(XPRT_CLOSING, &xprt->state))
 943                         return;
 944                 if (xprt_test_and_set_connecting(xprt))
 945                         return;
 946                 /* Race breaker */
 947                 if (!xprt_connected(xprt)) {
 948                         xprt->stat.connect_start = jiffies;
 949                         xprt->ops->connect(xprt, task);
 950                 } else {
 951                         xprt_clear_connecting(xprt);
 952                         task->tk_status = 0;
 953                         rpc_wake_up_queued_task(&xprt->pending, task);
 954                 }
 955         }
 956         xprt_release_write(xprt, task);
 957 }
 958
 959 /**
 960  * xprt_reconnect_delay - compute the wait before scheduling a connect
 961  * @xprt: transport instance
 962  *
 963  */
 964 unsigned long xprt_reconnect_delay(const struct rpc_xprt *xprt)
 965 {
 966         unsigned long start, now = jiffies;
 967
 968         start = xprt->stat.connect_start + xprt->reestablish_timeout;
 969         if (time_after(start, now))
 970                 return start - now;
 971         return 0;
 972 }
 973 EXPORT_SYMBOL_GPL(xprt_reconnect_delay);
 974
 975 /**
 976  * xprt_reconnect_backoff - compute the new re-establish timeout
 977  * @xprt: transport instance
 978  * @init_to: initial reestablish timeout
 979  *
 980  */
 981 void xprt_reconnect_backoff(struct rpc_xprt *xprt, unsigned long init_to)
 982 {
 983         xprt->reestablish_timeout <<= 1;
 984         if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
 985                 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
 986         if (xprt->reestablish_timeout < init_to)
 987                 xprt->reestablish_timeout = init_to;
 988 }
 989 EXPORT_SYMBOL_GPL(xprt_reconnect_backoff);
 990
 991 enum xprt_xid_rb_cmp {
 992         XID_RB_EQUAL,
 993         XID_RB_LEFT,
 994         XID_RB_RIGHT,
 995 };
 996 static enum xprt_xid_rb_cmp
 997 xprt_xid_cmp(__be32 xid1, __be32 xid2)
 998 {
 999         if (xid1 == xid2)
1000                 return XID_RB_EQUAL;
1001         if ((__force u32)xid1 < (__force u32)xid2)
1002                 return XID_RB_LEFT;
1003         return XID_RB_RIGHT;
1004 }
1005
1006 static struct rpc_rqst *
1007 xprt_request_rb_find(struct rpc_xprt *xprt, __be32 xid)
1008 {
1009         struct rb_node *n = xprt->recv_queue.rb_node;
1010         struct rpc_rqst *req;
1011
1012         while (n != NULL) {
1013                 req = rb_entry(n, struct rpc_rqst, rq_recv);
1014                 switch (xprt_xid_cmp(xid, req->rq_xid)) {
1015                 case XID_RB_LEFT:
1016                         n = n->rb_left;
1017                         break;
1018                 case XID_RB_RIGHT:
1019                         n = n->rb_right;
1020                         break;
1021                 case XID_RB_EQUAL:
1022                         return req;
1023                 }
1024         }
1025         return NULL;
1026 }
1027
1028 static void
1029 xprt_request_rb_insert(struct rpc_xprt *xprt, struct rpc_rqst *new)
1030 {
1031         struct rb_node **p = &xprt->recv_queue.rb_node;
1032         struct rb_node *n = NULL;
1033         struct rpc_rqst *req;
1034
1035         while (*p != NULL) {
1036                 n = *p;
1037                 req = rb_entry(n, struct rpc_rqst, rq_recv);
1038                 switch(xprt_xid_cmp(new->rq_xid, req->rq_xid)) {
1039                 case XID_RB_LEFT:
1040                         p = &n->rb_left;
1041                         break;
1042                 case XID_RB_RIGHT:
1043                         p = &n->rb_right;
1044                         break;
1045                 case XID_RB_EQUAL:
1046                         WARN_ON_ONCE(new != req);
1047                         return;
1048                 }
1049         }
1050         rb_link_node(&new->rq_recv, n, p);
1051         rb_insert_color(&new->rq_recv, &xprt->recv_queue);
1052 }
1053
1054 static void
1055 xprt_request_rb_remove(struct rpc_xprt *xprt, struct rpc_rqst *req)
1056 {
1057         rb_erase(&req->rq_recv, &xprt->recv_queue);
1058 }
1059
1060 /**
1061  * xprt_lookup_rqst - find an RPC request corresponding to an XID
1062  * @xprt: transport on which the original request was transmitted
1063  * @xid: RPC XID of incoming reply
1064  *
1065  * Caller holds xprt->queue_lock.
1066  */
1067 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
1068 {
1069         struct rpc_rqst *entry;
1070
1071         entry = xprt_request_rb_find(xprt, xid);
1072         if (entry != NULL) {
1073                 trace_xprt_lookup_rqst(xprt, xid, 0);
1074                 entry->rq_rtt = ktime_sub(ktime_get(), entry->rq_xtime);
1075                 return entry;
1076         }
1077
1078         dprintk("RPC:       xprt_lookup_rqst did not find xid %08x\n",
1079                         ntohl(xid));
1080         trace_xprt_lookup_rqst(xprt, xid, -ENOENT);
1081         xprt->stat.bad_xids++;
1082         return NULL;
1083 }
1084 EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
1085
1086 static bool
1087 xprt_is_pinned_rqst(struct rpc_rqst *req)
1088 {
1089         return atomic_read(&req->rq_pin) != 0;
1090 }
1091
1092 /**
1093  * xprt_pin_rqst - Pin a request on the transport receive list
1094  * @req: Request to pin
1095  *
1096  * Caller must ensure this is atomic with the call to xprt_lookup_rqst()
1097  * so should be holding xprt->queue_lock.
1098  */
1099 void xprt_pin_rqst(struct rpc_rqst *req)
1100 {
1101         atomic_inc(&req->rq_pin);
1102 }
1103 EXPORT_SYMBOL_GPL(xprt_pin_rqst);
1104
1105 /**
1106  * xprt_unpin_rqst - Unpin a request on the transport receive list
1107  * @req: Request to pin
1108  *
1109  * Caller should be holding xprt->queue_lock.
1110  */
1111 void xprt_unpin_rqst(struct rpc_rqst *req)
1112 {
1113         if (!test_bit(RPC_TASK_MSG_PIN_WAIT, &req->rq_task->tk_runstate)) {
1114                 atomic_dec(&req->rq_pin);
1115                 return;
1116         }
1117         if (atomic_dec_and_test(&req->rq_pin))
1118                 wake_up_var(&req->rq_pin);
1119 }
1120 EXPORT_SYMBOL_GPL(xprt_unpin_rqst);
1121
1122 static void xprt_wait_on_pinned_rqst(struct rpc_rqst *req)
1123 {
1124         wait_var_event(&req->rq_pin, !xprt_is_pinned_rqst(req));
1125 }
1126
1127 static bool
1128 xprt_request_data_received(struct rpc_task *task)
1129 {
1130         return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
1131                 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) != 0;
1132 }
1133
1134 static bool
1135 xprt_request_need_enqueue_receive(struct rpc_task *task, struct rpc_rqst *req)
1136 {
1137         return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
1138                 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) == 0;
1139 }
1140
1141 /**
1142  * xprt_request_enqueue_receive - Add an request to the receive queue
1143  * @task: RPC task
1144  *
1145  */
1146 void
1147 xprt_request_enqueue_receive(struct rpc_task *task)
1148 {
1149         struct rpc_rqst *req = task->tk_rqstp;
1150         struct rpc_xprt *xprt = req->rq_xprt;
1151
1152         if (!xprt_request_need_enqueue_receive(task, req))
1153                 return;
1154
1155         xprt_request_prepare(task->tk_rqstp);
1156         spin_lock(&xprt->queue_lock);
1157
1158         /* Update the softirq receive buffer */
1159         memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
1160                         sizeof(req->rq_private_buf));
1161
1162         /* Add request to the receive list */
1163         xprt_request_rb_insert(xprt, req);
1164         set_bit(RPC_TASK_NEED_RECV, &task->tk_runstate);
1165         spin_unlock(&xprt->queue_lock);
1166
1167         /* Turn off autodisconnect */
1168         del_singleshot_timer_sync(&xprt->timer);
1169 }
1170
1171 /**
1172  * xprt_request_dequeue_receive_locked - Remove a request from the receive queue
1173  * @task: RPC task
1174  *
1175  * Caller must hold xprt->queue_lock.
1176  */
1177 static void
1178 xprt_request_dequeue_receive_locked(struct rpc_task *task)
1179 {
1180         struct rpc_rqst *req = task->tk_rqstp;
1181
1182         if (test_and_clear_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1183                 xprt_request_rb_remove(req->rq_xprt, req);
1184 }
1185
1186 /**
1187  * xprt_update_rtt - Update RPC RTT statistics
1188  * @task: RPC request that recently completed
1189  *
1190  * Caller holds xprt->queue_lock.
1191  */
1192 void xprt_update_rtt(struct rpc_task *task)
1193 {
1194         struct rpc_rqst *req = task->tk_rqstp;
1195         struct rpc_rtt *rtt = task->tk_client->cl_rtt;
1196         unsigned int timer = task->tk_msg.rpc_proc->p_timer;
1197         long m = usecs_to_jiffies(ktime_to_us(req->rq_rtt));
1198
1199         if (timer) {
1200                 if (req->rq_ntrans == 1)
1201                         rpc_update_rtt(rtt, timer, m);
1202                 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
1203         }
1204 }
1205 EXPORT_SYMBOL_GPL(xprt_update_rtt);
1206
1207 /**
1208  * xprt_complete_rqst - called when reply processing is complete
1209  * @task: RPC request that recently completed
1210  * @copied: actual number of bytes received from the transport
1211  *
1212  * Caller holds xprt->queue_lock.
1213  */
1214 void xprt_complete_rqst(struct rpc_task *task, int copied)
1215 {
1216         struct rpc_rqst *req = task->tk_rqstp;
1217         struct rpc_xprt *xprt = req->rq_xprt;
1218
1219         xprt->stat.recvs++;
1220
1221         req->rq_private_buf.len = copied;
1222         /* Ensure all writes are done before we update */
1223         /* req->rq_reply_bytes_recvd */
1224         smp_wmb();
1225         req->rq_reply_bytes_recvd = copied;
1226         xprt_request_dequeue_receive_locked(task);
1227         rpc_wake_up_queued_task(&xprt->pending, task);
1228 }
1229 EXPORT_SYMBOL_GPL(xprt_complete_rqst);
1230
1231 static void xprt_timer(struct rpc_task *task)
1232 {
1233         struct rpc_rqst *req = task->tk_rqstp;
1234         struct rpc_xprt *xprt = req->rq_xprt;
1235
1236         if (task->tk_status != -ETIMEDOUT)
1237                 return;
1238
1239         trace_xprt_timer(xprt, req->rq_xid, task->tk_status);
1240         if (!req->rq_reply_bytes_recvd) {
1241                 if (xprt->ops->timer)
1242                         xprt->ops->timer(xprt, task);
1243         } else
1244                 task->tk_status = 0;
1245 }
1246
1247 /**
1248  * xprt_wait_for_reply_request_def - wait for reply
1249  * @task: pointer to rpc_task
1250  *
1251  * Set a request's retransmit timeout based on the transport's
1252  * default timeout parameters.  Used by transports that don't adjust
1253  * the retransmit timeout based on round-trip time estimation,
1254  * and put the task to sleep on the pending queue.
1255  */
1256 void xprt_wait_for_reply_request_def(struct rpc_task *task)
1257 {
1258         struct rpc_rqst *req = task->tk_rqstp;
1259
1260         rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1261                         xprt_request_timeout(req));
1262 }
1263 EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_def);
1264
1265 /**
1266  * xprt_wait_for_reply_request_rtt - wait for reply using RTT estimator
1267  * @task: pointer to rpc_task
1268  *
1269  * Set a request's retransmit timeout using the RTT estimator,
1270  * and put the task to sleep on the pending queue.
1271  */
1272 void xprt_wait_for_reply_request_rtt(struct rpc_task *task)
1273 {
1274         int timer = task->tk_msg.rpc_proc->p_timer;
1275         struct rpc_clnt *clnt = task->tk_client;
1276         struct rpc_rtt *rtt = clnt->cl_rtt;
1277         struct rpc_rqst *req = task->tk_rqstp;
1278         unsigned long max_timeout = clnt->cl_timeout->to_maxval;
1279         unsigned long timeout;
1280
1281         timeout = rpc_calc_rto(rtt, timer);
1282         timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
1283         if (timeout > max_timeout || timeout == 0)
1284                 timeout = max_timeout;
1285         rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1286                         jiffies + timeout);
1287 }
1288 EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_rtt);
1289
1290 /**
1291  * xprt_request_wait_receive - wait for the reply to an RPC request
1292  * @task: RPC task about to send a request
1293  *
1294  */
1295 void xprt_request_wait_receive(struct rpc_task *task)
1296 {
1297         struct rpc_rqst *req = task->tk_rqstp;
1298         struct rpc_xprt *xprt = req->rq_xprt;
1299
1300         if (!test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1301                 return;
1302         /*
1303          * Sleep on the pending queue if we're expecting a reply.
1304          * The spinlock ensures atomicity between the test of
1305          * req->rq_reply_bytes_recvd, and the call to rpc_sleep_on().
1306          */
1307         spin_lock(&xprt->queue_lock);
1308         if (test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate)) {
1309                 xprt->ops->wait_for_reply_request(task);
1310                 /*
1311                  * Send an extra queue wakeup call if the
1312                  * connection was dropped in case the call to
1313                  * rpc_sleep_on() raced.
1314                  */
1315                 if (xprt_request_retransmit_after_disconnect(task))
1316                         rpc_wake_up_queued_task_set_status(&xprt->pending,
1317                                         task, -ENOTCONN);
1318         }
1319         spin_unlock(&xprt->queue_lock);
1320 }
1321
1322 static bool
1323 xprt_request_need_enqueue_transmit(struct rpc_task *task, struct rpc_rqst *req)
1324 {
1325         return !test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1326 }
1327
1328 /**
1329  * xprt_request_enqueue_transmit - queue a task for transmission
1330  * @task: pointer to rpc_task
1331  *
1332  * Add a task to the transmission queue.
1333  */
1334 void
1335 xprt_request_enqueue_transmit(struct rpc_task *task)
1336 {
1337         struct rpc_rqst *pos, *req = task->tk_rqstp;
1338         struct rpc_xprt *xprt = req->rq_xprt;
1339
1340         if (xprt_request_need_enqueue_transmit(task, req)) {
1341                 req->rq_bytes_sent = 0;
1342                 spin_lock(&xprt->queue_lock);
1343                 /*
1344                  * Requests that carry congestion control credits are added
1345                  * to the head of the list to avoid starvation issues.
1346                  */
1347                 if (req->rq_cong) {
1348                         xprt_clear_congestion_window_wait(xprt);
1349                         list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1350                                 if (pos->rq_cong)
1351                                         continue;
1352                                 /* Note: req is added _before_ pos */
1353                                 list_add_tail(&req->rq_xmit, &pos->rq_xmit);
1354                                 INIT_LIST_HEAD(&req->rq_xmit2);
1355                                 goto out;
1356                         }
1357                 } else if (!req->rq_seqno) {
1358                         list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1359                                 if (pos->rq_task->tk_owner != task->tk_owner)
1360                                         continue;
1361                                 list_add_tail(&req->rq_xmit2, &pos->rq_xmit2);
1362                                 INIT_LIST_HEAD(&req->rq_xmit);
1363                                 goto out;
1364                         }
1365                 }
1366                 list_add_tail(&req->rq_xmit, &xprt->xmit_queue);
1367                 INIT_LIST_HEAD(&req->rq_xmit2);
1368 out:
1369                 atomic_long_inc(&xprt->xmit_queuelen);
1370                 set_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1371                 spin_unlock(&xprt->queue_lock);
1372         }
1373 }
1374
1375 /**
1376  * xprt_request_dequeue_transmit_locked - remove a task from the transmission queue
1377  * @task: pointer to rpc_task
1378  *
1379  * Remove a task from the transmission queue
1380  * Caller must hold xprt->queue_lock
1381  */
1382 static void
1383 xprt_request_dequeue_transmit_locked(struct rpc_task *task)
1384 {
1385         struct rpc_rqst *req = task->tk_rqstp;
1386
1387         if (!test_and_clear_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1388                 return;
1389         if (!list_empty(&req->rq_xmit)) {
1390                 list_del(&req->rq_xmit);
1391                 if (!list_empty(&req->rq_xmit2)) {
1392                         struct rpc_rqst *next = list_first_entry(&req->rq_xmit2,
1393                                         struct rpc_rqst, rq_xmit2);
1394                         list_del(&req->rq_xmit2);
1395                         list_add_tail(&next->rq_xmit, &next->rq_xprt->xmit_queue);
1396                 }
1397         } else
1398                 list_del(&req->rq_xmit2);
1399         atomic_long_dec(&req->rq_xprt->xmit_queuelen);
1400 }
1401
1402 /**
1403  * xprt_request_dequeue_transmit - remove a task from the transmission queue
1404  * @task: pointer to rpc_task
1405  *
1406  * Remove a task from the transmission queue
1407  */
1408 static void
1409 xprt_request_dequeue_transmit(struct rpc_task *task)
1410 {
1411         struct rpc_rqst *req = task->tk_rqstp;
1412         struct rpc_xprt *xprt = req->rq_xprt;
1413
1414         spin_lock(&xprt->queue_lock);
1415         xprt_request_dequeue_transmit_locked(task);
1416         spin_unlock(&xprt->queue_lock);
1417 }
1418
1419 /**
1420  * xprt_request_dequeue_xprt - remove a task from the transmit+receive queue
1421  * @task: pointer to rpc_task
1422  *
1423  * Remove a task from the transmit and receive queues, and ensure that
1424  * it is not pinned by the receive work item.
1425  */
1426 void
1427 xprt_request_dequeue_xprt(struct rpc_task *task)
1428 {
1429         struct rpc_rqst *req = task->tk_rqstp;
1430         struct rpc_xprt *xprt = req->rq_xprt;
1431
1432         if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate) ||
1433             test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) ||
1434             xprt_is_pinned_rqst(req)) {
1435                 spin_lock(&xprt->queue_lock);
1436                 xprt_request_dequeue_transmit_locked(task);
1437                 xprt_request_dequeue_receive_locked(task);
1438                 while (xprt_is_pinned_rqst(req)) {
1439                         set_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1440                         spin_unlock(&xprt->queue_lock);
1441                         xprt_wait_on_pinned_rqst(req);
1442                         spin_lock(&xprt->queue_lock);
1443                         clear_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1444                 }
1445                 spin_unlock(&xprt->queue_lock);
1446         }
1447 }
1448
1449 /**
1450  * xprt_request_prepare - prepare an encoded request for transport
1451  * @req: pointer to rpc_rqst
1452  *
1453  * Calls into the transport layer to do whatever is needed to prepare
1454  * the request for transmission or receive.
1455  */
1456 void
1457 xprt_request_prepare(struct rpc_rqst *req)
1458 {
1459         struct rpc_xprt *xprt = req->rq_xprt;
1460
1461         if (xprt->ops->prepare_request)
1462                 xprt->ops->prepare_request(req);
1463 }
1464
1465 /**
1466  * xprt_request_need_retransmit - Test if a task needs retransmission
1467  * @task: pointer to rpc_task
1468  *
1469  * Test for whether a connection breakage requires the task to retransmit
1470  */
1471 bool
1472 xprt_request_need_retransmit(struct rpc_task *task)
1473 {
1474         return xprt_request_retransmit_after_disconnect(task);
1475 }
1476
1477 /**
1478  * xprt_prepare_transmit - reserve the transport before sending a request
1479  * @task: RPC task about to send a request
1480  *
1481  */
1482 bool xprt_prepare_transmit(struct rpc_task *task)
1483 {
1484         struct rpc_rqst *req = task->tk_rqstp;
1485         struct rpc_xprt *xprt = req->rq_xprt;
1486
1487         if (!xprt_lock_write(xprt, task)) {
1488                 /* Race breaker: someone may have transmitted us */
1489                 if (!test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1490                         rpc_wake_up_queued_task_set_status(&xprt->sending,
1491                                         task, 0);
1492                 return false;
1493
1494         }
1495         if (atomic_read(&xprt->swapper))
1496                 /* This will be clear in __rpc_execute */
1497                 current->flags |= PF_MEMALLOC;
1498         return true;
1499 }
1500
1501 void xprt_end_transmit(struct rpc_task *task)
1502 {
1503         struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
1504
1505         xprt_inject_disconnect(xprt);
1506         xprt_release_write(xprt, task);
1507 }
1508
1509 /**
1510  * xprt_request_transmit - send an RPC request on a transport
1511  * @req: pointer to request to transmit
1512  * @snd_task: RPC task that owns the transport lock
1513  *
1514  * This performs the transmission of a single request.
1515  * Note that if the request is not the same as snd_task, then it
1516  * does need to be pinned.
1517  * Returns '0' on success.
1518  */
1519 static int
1520 xprt_request_transmit(struct rpc_rqst *req, struct rpc_task *snd_task)
1521 {
1522         struct rpc_xprt *xprt = req->rq_xprt;
1523         struct rpc_task *task = req->rq_task;
1524         unsigned int connect_cookie;
1525         int is_retrans = RPC_WAS_SENT(task);
1526         int status;
1527
1528         if (!req->rq_bytes_sent) {
1529                 if (xprt_request_data_received(task)) {
1530                         status = 0;
1531                         goto out_dequeue;
1532                 }
1533                 /* Verify that our message lies in the RPCSEC_GSS window */
1534                 if (rpcauth_xmit_need_reencode(task)) {
1535                         status = -EBADMSG;
1536                         goto out_dequeue;
1537                 }
1538                 if (RPC_SIGNALLED(task)) {
1539                         status = -ERESTARTSYS;
1540                         goto out_dequeue;
1541                 }
1542         }
1543
1544         /*
1545          * Update req->rq_ntrans before transmitting to avoid races with
1546          * xprt_update_rtt(), which needs to know that it is recording a
1547          * reply to the first transmission.
1548          */
1549         req->rq_ntrans++;
1550
1551         trace_rpc_xdr_sendto(task, &req->rq_snd_buf);
1552         connect_cookie = xprt->connect_cookie;
1553         status = xprt->ops->send_request(req);
1554         if (status != 0) {
1555                 req->rq_ntrans--;
1556                 trace_xprt_transmit(req, status);
1557                 return status;
1558         }
1559
1560         if (is_retrans) {
1561                 task->tk_client->cl_stats->rpcretrans++;
1562                 trace_xprt_retransmit(req);
1563         }
1564
1565         xprt_inject_disconnect(xprt);
1566
1567         task->tk_flags |= RPC_TASK_SENT;
1568         spin_lock(&xprt->transport_lock);
1569
1570         xprt->stat.sends++;
1571         xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
1572         xprt->stat.bklog_u += xprt->backlog.qlen;
1573         xprt->stat.sending_u += xprt->sending.qlen;
1574         xprt->stat.pending_u += xprt->pending.qlen;
1575         spin_unlock(&xprt->transport_lock);
1576
1577         req->rq_connect_cookie = connect_cookie;
1578 out_dequeue:
1579         trace_xprt_transmit(req, status);
1580         xprt_request_dequeue_transmit(task);
1581         rpc_wake_up_queued_task_set_status(&xprt->sending, task, status);
1582         return status;
1583 }
1584
1585 /**
1586  * xprt_transmit - send an RPC request on a transport
1587  * @task: controlling RPC task
1588  *
1589  * Attempts to drain the transmit queue. On exit, either the transport
1590  * signalled an error that needs to be handled before transmission can
1591  * resume, or @task finished transmitting, and detected that it already
1592  * received a reply.
1593  */
1594 void
1595 xprt_transmit(struct rpc_task *task)
1596 {
1597         struct rpc_rqst *next, *req = task->tk_rqstp;
1598         struct rpc_xprt *xprt = req->rq_xprt;
1599         int status;
1600
1601         spin_lock(&xprt->queue_lock);
1602         for (;;) {
1603                 next = list_first_entry_or_null(&xprt->xmit_queue,
1604                                                 struct rpc_rqst, rq_xmit);
1605                 if (!next)
1606                         break;
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 (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1616                                 task->tk_status = status;
1617                         break;
1618                 }
1619                 /* Was @task transmitted, and has it received a reply? */
1620                 if (xprt_request_data_received(task) &&
1621                     !test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1622                         break;
1623                 cond_resched_lock(&xprt->queue_lock);
1624         }
1625         spin_unlock(&xprt->queue_lock);
1626 }
1627
1628 static void xprt_complete_request_init(struct rpc_task *task)
1629 {
1630         if (task->tk_rqstp)
1631                 xprt_request_init(task);
1632 }
1633
1634 void xprt_add_backlog(struct rpc_xprt *xprt, struct rpc_task *task)
1635 {
1636         set_bit(XPRT_CONGESTED, &xprt->state);
1637         rpc_sleep_on(&xprt->backlog, task, xprt_complete_request_init);
1638 }
1639 EXPORT_SYMBOL_GPL(xprt_add_backlog);
1640
1641 static bool __xprt_set_rq(struct rpc_task *task, void *data)
1642 {
1643         struct rpc_rqst *req = data;
1644
1645         if (task->tk_rqstp == NULL) {
1646                 memset(req, 0, sizeof(*req));   /* mark unused */
1647                 task->tk_rqstp = req;
1648                 return true;
1649         }
1650         return false;
1651 }
1652
1653 bool xprt_wake_up_backlog(struct rpc_xprt *xprt, struct rpc_rqst *req)
1654 {
1655         if (rpc_wake_up_first(&xprt->backlog, __xprt_set_rq, req) == NULL) {
1656                 clear_bit(XPRT_CONGESTED, &xprt->state);
1657                 return false;
1658         }
1659         return true;
1660 }
1661 EXPORT_SYMBOL_GPL(xprt_wake_up_backlog);
1662
1663 static bool xprt_throttle_congested(struct rpc_xprt *xprt, struct rpc_task *task)
1664 {
1665         bool ret = false;
1666
1667         if (!test_bit(XPRT_CONGESTED, &xprt->state))
1668                 goto out;
1669         spin_lock(&xprt->reserve_lock);
1670         if (test_bit(XPRT_CONGESTED, &xprt->state)) {
1671                 xprt_add_backlog(xprt, task);
1672                 ret = true;
1673         }
1674         spin_unlock(&xprt->reserve_lock);
1675 out:
1676         return ret;
1677 }
1678
1679 static struct rpc_rqst *xprt_dynamic_alloc_slot(struct rpc_xprt *xprt)
1680 {
1681         struct rpc_rqst *req = ERR_PTR(-EAGAIN);
1682
1683         if (xprt->num_reqs >= xprt->max_reqs)
1684                 goto out;
1685         ++xprt->num_reqs;
1686         spin_unlock(&xprt->reserve_lock);
1687         req = kzalloc(sizeof(*req), rpc_task_gfp_mask());
1688         spin_lock(&xprt->reserve_lock);
1689         if (req != NULL)
1690                 goto out;
1691         --xprt->num_reqs;
1692         req = ERR_PTR(-ENOMEM);
1693 out:
1694         return req;
1695 }
1696
1697 static bool xprt_dynamic_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1698 {
1699         if (xprt->num_reqs > xprt->min_reqs) {
1700                 --xprt->num_reqs;
1701                 kfree(req);
1702                 return true;
1703         }
1704         return false;
1705 }
1706
1707 void xprt_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1708 {
1709         struct rpc_rqst *req;
1710
1711         spin_lock(&xprt->reserve_lock);
1712         if (!list_empty(&xprt->free)) {
1713                 req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
1714                 list_del(&req->rq_list);
1715                 goto out_init_req;
1716         }
1717         req = xprt_dynamic_alloc_slot(xprt);
1718         if (!IS_ERR(req))
1719                 goto out_init_req;
1720         switch (PTR_ERR(req)) {
1721         case -ENOMEM:
1722                 dprintk("RPC:       dynamic allocation of request slot "
1723                                 "failed! Retrying\n");
1724                 task->tk_status = -ENOMEM;
1725                 break;
1726         case -EAGAIN:
1727                 xprt_add_backlog(xprt, task);
1728                 dprintk("RPC:       waiting for request slot\n");
1729                 fallthrough;
1730         default:
1731                 task->tk_status = -EAGAIN;
1732         }
1733         spin_unlock(&xprt->reserve_lock);
1734         return;
1735 out_init_req:
1736         xprt->stat.max_slots = max_t(unsigned int, xprt->stat.max_slots,
1737                                      xprt->num_reqs);
1738         spin_unlock(&xprt->reserve_lock);
1739
1740         task->tk_status = 0;
1741         task->tk_rqstp = req;
1742 }
1743 EXPORT_SYMBOL_GPL(xprt_alloc_slot);
1744
1745 void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1746 {
1747         spin_lock(&xprt->reserve_lock);
1748         if (!xprt_wake_up_backlog(xprt, req) &&
1749             !xprt_dynamic_free_slot(xprt, req)) {
1750                 memset(req, 0, sizeof(*req));   /* mark unused */
1751                 list_add(&req->rq_list, &xprt->free);
1752         }
1753         spin_unlock(&xprt->reserve_lock);
1754 }
1755 EXPORT_SYMBOL_GPL(xprt_free_slot);
1756
1757 static void xprt_free_all_slots(struct rpc_xprt *xprt)
1758 {
1759         struct rpc_rqst *req;
1760         while (!list_empty(&xprt->free)) {
1761                 req = list_first_entry(&xprt->free, struct rpc_rqst, rq_list);
1762                 list_del(&req->rq_list);
1763                 kfree(req);
1764         }
1765 }
1766
1767 static DEFINE_IDA(rpc_xprt_ids);
1768
1769 void xprt_cleanup_ids(void)
1770 {
1771         ida_destroy(&rpc_xprt_ids);
1772 }
1773
1774 static int xprt_alloc_id(struct rpc_xprt *xprt)
1775 {
1776         int id;
1777
1778         id = ida_simple_get(&rpc_xprt_ids, 0, 0, GFP_KERNEL);
1779         if (id < 0)
1780                 return id;
1781
1782         xprt->id = id;
1783         return 0;
1784 }
1785
1786 static void xprt_free_id(struct rpc_xprt *xprt)
1787 {
1788         ida_simple_remove(&rpc_xprt_ids, xprt->id);
1789 }
1790
1791 struct rpc_xprt *xprt_alloc(struct net *net, size_t size,
1792                 unsigned int num_prealloc,
1793                 unsigned int max_alloc)
1794 {
1795         struct rpc_xprt *xprt;
1796         struct rpc_rqst *req;
1797         int i;
1798
1799         xprt = kzalloc(size, GFP_KERNEL);
1800         if (xprt == NULL)
1801                 goto out;
1802
1803         xprt_alloc_id(xprt);
1804         xprt_init(xprt, net);
1805
1806         for (i = 0; i < num_prealloc; i++) {
1807                 req = kzalloc(sizeof(struct rpc_rqst), GFP_KERNEL);
1808                 if (!req)
1809                         goto out_free;
1810                 list_add(&req->rq_list, &xprt->free);
1811         }
1812         if (max_alloc > num_prealloc)
1813                 xprt->max_reqs = max_alloc;
1814         else
1815                 xprt->max_reqs = num_prealloc;
1816         xprt->min_reqs = num_prealloc;
1817         xprt->num_reqs = num_prealloc;
1818
1819         return xprt;
1820
1821 out_free:
1822         xprt_free(xprt);
1823 out:
1824         return NULL;
1825 }
1826 EXPORT_SYMBOL_GPL(xprt_alloc);
1827
1828 void xprt_free(struct rpc_xprt *xprt)
1829 {
1830         put_net_track(xprt->xprt_net, &xprt->ns_tracker);
1831         xprt_free_all_slots(xprt);
1832         xprt_free_id(xprt);
1833         rpc_sysfs_xprt_destroy(xprt);
1834         kfree_rcu(xprt, rcu);
1835 }
1836 EXPORT_SYMBOL_GPL(xprt_free);
1837
1838 static void
1839 xprt_init_connect_cookie(struct rpc_rqst *req, struct rpc_xprt *xprt)
1840 {
1841         req->rq_connect_cookie = xprt_connect_cookie(xprt) - 1;
1842 }
1843
1844 static __be32
1845 xprt_alloc_xid(struct rpc_xprt *xprt)
1846 {
1847         __be32 xid;
1848
1849         spin_lock(&xprt->reserve_lock);
1850         xid = (__force __be32)xprt->xid++;
1851         spin_unlock(&xprt->reserve_lock);
1852         return xid;
1853 }
1854
1855 static void
1856 xprt_init_xid(struct rpc_xprt *xprt)
1857 {
1858         xprt->xid = prandom_u32();
1859 }
1860
1861 static void
1862 xprt_request_init(struct rpc_task *task)
1863 {
1864         struct rpc_xprt *xprt = task->tk_xprt;
1865         struct rpc_rqst *req = task->tk_rqstp;
1866
1867         req->rq_task    = task;
1868         req->rq_xprt    = xprt;
1869         req->rq_buffer  = NULL;
1870         req->rq_xid     = xprt_alloc_xid(xprt);
1871         xprt_init_connect_cookie(req, xprt);
1872         req->rq_snd_buf.len = 0;
1873         req->rq_snd_buf.buflen = 0;
1874         req->rq_rcv_buf.len = 0;
1875         req->rq_rcv_buf.buflen = 0;
1876         req->rq_snd_buf.bvec = NULL;
1877         req->rq_rcv_buf.bvec = NULL;
1878         req->rq_release_snd_buf = NULL;
1879         xprt_init_majortimeo(task, req);
1880
1881         trace_xprt_reserve(req);
1882 }
1883
1884 static void
1885 xprt_do_reserve(struct rpc_xprt *xprt, struct rpc_task *task)
1886 {
1887         xprt->ops->alloc_slot(xprt, task);
1888         if (task->tk_rqstp != NULL)
1889                 xprt_request_init(task);
1890 }
1891
1892 /**
1893  * xprt_reserve - allocate an RPC request slot
1894  * @task: RPC task requesting a slot allocation
1895  *
1896  * If the transport is marked as being congested, or if no more
1897  * slots are available, place the task on the transport's
1898  * backlog queue.
1899  */
1900 void xprt_reserve(struct rpc_task *task)
1901 {
1902         struct rpc_xprt *xprt = task->tk_xprt;
1903
1904         task->tk_status = 0;
1905         if (task->tk_rqstp != NULL)
1906                 return;
1907
1908         task->tk_status = -EAGAIN;
1909         if (!xprt_throttle_congested(xprt, task))
1910                 xprt_do_reserve(xprt, task);
1911 }
1912
1913 /**
1914  * xprt_retry_reserve - allocate an RPC request slot
1915  * @task: RPC task requesting a slot allocation
1916  *
1917  * If no more slots are available, place the task on the transport's
1918  * backlog queue.
1919  * Note that the only difference with xprt_reserve is that we now
1920  * ignore the value of the XPRT_CONGESTED flag.
1921  */
1922 void xprt_retry_reserve(struct rpc_task *task)
1923 {
1924         struct rpc_xprt *xprt = task->tk_xprt;
1925
1926         task->tk_status = 0;
1927         if (task->tk_rqstp != NULL)
1928                 return;
1929
1930         task->tk_status = -EAGAIN;
1931         xprt_do_reserve(xprt, task);
1932 }
1933
1934 /**
1935  * xprt_release - release an RPC request slot
1936  * @task: task which is finished with the slot
1937  *
1938  */
1939 void xprt_release(struct rpc_task *task)
1940 {
1941         struct rpc_xprt *xprt;
1942         struct rpc_rqst *req = task->tk_rqstp;
1943
1944         if (req == NULL) {
1945                 if (task->tk_client) {
1946                         xprt = task->tk_xprt;
1947                         xprt_release_write(xprt, task);
1948                 }
1949                 return;
1950         }
1951
1952         xprt = req->rq_xprt;
1953         xprt_request_dequeue_xprt(task);
1954         spin_lock(&xprt->transport_lock);
1955         xprt->ops->release_xprt(xprt, task);
1956         if (xprt->ops->release_request)
1957                 xprt->ops->release_request(task);
1958         xprt_schedule_autodisconnect(xprt);
1959         spin_unlock(&xprt->transport_lock);
1960         if (req->rq_buffer)
1961                 xprt->ops->buf_free(task);
1962         xdr_free_bvec(&req->rq_rcv_buf);
1963         xdr_free_bvec(&req->rq_snd_buf);
1964         if (req->rq_cred != NULL)
1965                 put_rpccred(req->rq_cred);
1966         if (req->rq_release_snd_buf)
1967                 req->rq_release_snd_buf(req);
1968
1969         task->tk_rqstp = NULL;
1970         if (likely(!bc_prealloc(req)))
1971                 xprt->ops->free_slot(xprt, req);
1972         else
1973                 xprt_free_bc_request(req);
1974 }
1975
1976 #ifdef CONFIG_SUNRPC_BACKCHANNEL
1977 void
1978 xprt_init_bc_request(struct rpc_rqst *req, struct rpc_task *task)
1979 {
1980         struct xdr_buf *xbufp = &req->rq_snd_buf;
1981
1982         task->tk_rqstp = req;
1983         req->rq_task = task;
1984         xprt_init_connect_cookie(req, req->rq_xprt);
1985         /*
1986          * Set up the xdr_buf length.
1987          * This also indicates that the buffer is XDR encoded already.
1988          */
1989         xbufp->len = xbufp->head[0].iov_len + xbufp->page_len +
1990                 xbufp->tail[0].iov_len;
1991 }
1992 #endif
1993
1994 static void xprt_init(struct rpc_xprt *xprt, struct net *net)
1995 {
1996         kref_init(&xprt->kref);
1997
1998         spin_lock_init(&xprt->transport_lock);
1999         spin_lock_init(&xprt->reserve_lock);
2000         spin_lock_init(&xprt->queue_lock);
2001
2002         INIT_LIST_HEAD(&xprt->free);
2003         xprt->recv_queue = RB_ROOT;
2004         INIT_LIST_HEAD(&xprt->xmit_queue);
2005 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
2006         spin_lock_init(&xprt->bc_pa_lock);
2007         INIT_LIST_HEAD(&xprt->bc_pa_list);
2008 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
2009         INIT_LIST_HEAD(&xprt->xprt_switch);
2010
2011         xprt->last_used = jiffies;
2012         xprt->cwnd = RPC_INITCWND;
2013         xprt->bind_index = 0;
2014
2015         rpc_init_wait_queue(&xprt->binding, "xprt_binding");
2016         rpc_init_wait_queue(&xprt->pending, "xprt_pending");
2017         rpc_init_wait_queue(&xprt->sending, "xprt_sending");
2018         rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
2019
2020         xprt_init_xid(xprt);
2021
2022         xprt->xprt_net = get_net_track(net, &xprt->ns_tracker, GFP_KERNEL);
2023 }
2024
2025 /**
2026  * xprt_create_transport - create an RPC transport
2027  * @args: rpc transport creation arguments
2028  *
2029  */
2030 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
2031 {
2032         struct rpc_xprt *xprt;
2033         const struct xprt_class *t;
2034
2035         t = xprt_class_find_by_ident(args->ident);
2036         if (!t) {
2037                 dprintk("RPC: transport (%d) not supported\n", args->ident);
2038                 return ERR_PTR(-EIO);
2039         }
2040
2041         xprt = t->setup(args);
2042         xprt_class_release(t);
2043
2044         if (IS_ERR(xprt))
2045                 goto out;
2046         if (args->flags & XPRT_CREATE_NO_IDLE_TIMEOUT)
2047                 xprt->idle_timeout = 0;
2048         INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
2049         if (xprt_has_timer(xprt))
2050                 timer_setup(&xprt->timer, xprt_init_autodisconnect, 0);
2051         else
2052                 timer_setup(&xprt->timer, NULL, 0);
2053
2054         if (strlen(args->servername) > RPC_MAXNETNAMELEN) {
2055                 xprt_destroy(xprt);
2056                 return ERR_PTR(-EINVAL);
2057         }
2058         xprt->servername = kstrdup(args->servername, GFP_KERNEL);
2059         if (xprt->servername == NULL) {
2060                 xprt_destroy(xprt);
2061                 return ERR_PTR(-ENOMEM);
2062         }
2063
2064         rpc_xprt_debugfs_register(xprt);
2065
2066         trace_xprt_create(xprt);
2067 out:
2068         return xprt;
2069 }
2070
2071 static void xprt_destroy_cb(struct work_struct *work)
2072 {
2073         struct rpc_xprt *xprt =
2074                 container_of(work, struct rpc_xprt, task_cleanup);
2075
2076         trace_xprt_destroy(xprt);
2077
2078         rpc_xprt_debugfs_unregister(xprt);
2079         rpc_destroy_wait_queue(&xprt->binding);
2080         rpc_destroy_wait_queue(&xprt->pending);
2081         rpc_destroy_wait_queue(&xprt->sending);
2082         rpc_destroy_wait_queue(&xprt->backlog);
2083         kfree(xprt->servername);
2084         /*
2085          * Destroy any existing back channel
2086          */
2087         xprt_destroy_backchannel(xprt, UINT_MAX);
2088
2089         /*
2090          * Tear down transport state and free the rpc_xprt
2091          */
2092         xprt->ops->destroy(xprt);
2093 }
2094
2095 /**
2096  * xprt_destroy - destroy an RPC transport, killing off all requests.
2097  * @xprt: transport to destroy
2098  *
2099  */
2100 static void xprt_destroy(struct rpc_xprt *xprt)
2101 {
2102         /*
2103          * Exclude transport connect/disconnect handlers and autoclose
2104          */
2105         wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_UNINTERRUPTIBLE);
2106
2107         /*
2108          * xprt_schedule_autodisconnect() can run after XPRT_LOCKED
2109          * is cleared.  We use ->transport_lock to ensure the mod_timer()
2110          * can only run *before* del_time_sync(), never after.
2111          */
2112         spin_lock(&xprt->transport_lock);
2113         del_timer_sync(&xprt->timer);
2114         spin_unlock(&xprt->transport_lock);
2115
2116         /*
2117          * Destroy sockets etc from the system workqueue so they can
2118          * safely flush receive work running on rpciod.
2119          */
2120         INIT_WORK(&xprt->task_cleanup, xprt_destroy_cb);
2121         schedule_work(&xprt->task_cleanup);
2122 }
2123
2124 static void xprt_destroy_kref(struct kref *kref)
2125 {
2126         xprt_destroy(container_of(kref, struct rpc_xprt, kref));
2127 }
2128
2129 /**
2130  * xprt_get - return a reference to an RPC transport.
2131  * @xprt: pointer to the transport
2132  *
2133  */
2134 struct rpc_xprt *xprt_get(struct rpc_xprt *xprt)
2135 {
2136         if (xprt != NULL && kref_get_unless_zero(&xprt->kref))
2137                 return xprt;
2138         return NULL;
2139 }
2140 EXPORT_SYMBOL_GPL(xprt_get);
2141
2142 /**
2143  * xprt_put - release a reference to an RPC transport.
2144  * @xprt: pointer to the transport
2145  *
2146  */
2147 void xprt_put(struct rpc_xprt *xprt)
2148 {
2149         if (xprt != NULL)
2150                 kref_put(&xprt->kref, xprt_destroy_kref);
2151 }
2152 EXPORT_SYMBOL_GPL(xprt_put);