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 static int      xprt_request_prepare(struct rpc_rqst *req, struct xdr_buf *buf);
  77
  78 static DEFINE_SPINLOCK(xprt_list_lock);
  79 static LIST_HEAD(xprt_list);
  80
  81 static unsigned long xprt_request_timeout(const struct rpc_rqst *req)
  82 {
  83         unsigned long timeout = jiffies + req->rq_timeout;
  84
  85         if (time_before(timeout, req->rq_majortimeo))
  86                 return timeout;
  87         return req->rq_majortimeo;
  88 }
  89
  90 /**
  91  * xprt_register_transport - register a transport implementation
  92  * @transport: transport to register
  93  *
  94  * If a transport implementation is loaded as a kernel module, it can
  95  * call this interface to make itself known to the RPC client.
  96  *
  97  * Returns:
  98  * 0:           transport successfully registered
  99  * -EEXIST:     transport already registered
 100  * -EINVAL:     transport module being unloaded
 101  */
 102 int xprt_register_transport(struct xprt_class *transport)
 103 {
 104         struct xprt_class *t;
 105         int result;
 106
 107         result = -EEXIST;
 108         spin_lock(&xprt_list_lock);
 109         list_for_each_entry(t, &xprt_list, list) {
 110                 /* don't register the same transport class twice */
 111                 if (t->ident == transport->ident)
 112                         goto out;
 113         }
 114
 115         list_add_tail(&transport->list, &xprt_list);
 116         printk(KERN_INFO "RPC: Registered %s transport module.\n",
 117                transport->name);
 118         result = 0;
 119
 120 out:
 121         spin_unlock(&xprt_list_lock);
 122         return result;
 123 }
 124 EXPORT_SYMBOL_GPL(xprt_register_transport);
 125
 126 /**
 127  * xprt_unregister_transport - unregister a transport implementation
 128  * @transport: transport to unregister
 129  *
 130  * Returns:
 131  * 0:           transport successfully unregistered
 132  * -ENOENT:     transport never registered
 133  */
 134 int xprt_unregister_transport(struct xprt_class *transport)
 135 {
 136         struct xprt_class *t;
 137         int result;
 138
 139         result = 0;
 140         spin_lock(&xprt_list_lock);
 141         list_for_each_entry(t, &xprt_list, list) {
 142                 if (t == transport) {
 143                         printk(KERN_INFO
 144                                 "RPC: Unregistered %s transport module.\n",
 145                                 transport->name);
 146                         list_del_init(&transport->list);
 147                         goto out;
 148                 }
 149         }
 150         result = -ENOENT;
 151
 152 out:
 153         spin_unlock(&xprt_list_lock);
 154         return result;
 155 }
 156 EXPORT_SYMBOL_GPL(xprt_unregister_transport);
 157
 158 static void
 159 xprt_class_release(const struct xprt_class *t)
 160 {
 161         module_put(t->owner);
 162 }
 163
 164 static const struct xprt_class *
 165 xprt_class_find_by_ident_locked(int ident)
 166 {
 167         const struct xprt_class *t;
 168
 169         list_for_each_entry(t, &xprt_list, list) {
 170                 if (t->ident != ident)
 171                         continue;
 172                 if (!try_module_get(t->owner))
 173                         continue;
 174                 return t;
 175         }
 176         return NULL;
 177 }
 178
 179 static const struct xprt_class *
 180 xprt_class_find_by_ident(int ident)
 181 {
 182         const struct xprt_class *t;
 183
 184         spin_lock(&xprt_list_lock);
 185         t = xprt_class_find_by_ident_locked(ident);
 186         spin_unlock(&xprt_list_lock);
 187         return t;
 188 }
 189
 190 static const struct xprt_class *
 191 xprt_class_find_by_netid_locked(const char *netid)
 192 {
 193         const struct xprt_class *t;
 194         unsigned int i;
 195
 196         list_for_each_entry(t, &xprt_list, list) {
 197                 for (i = 0; t->netid[i][0] != '\0'; i++) {
 198                         if (strcmp(t->netid[i], netid) != 0)
 199                                 continue;
 200                         if (!try_module_get(t->owner))
 201                                 continue;
 202                         return t;
 203                 }
 204         }
 205         return NULL;
 206 }
 207
 208 static const struct xprt_class *
 209 xprt_class_find_by_netid(const char *netid)
 210 {
 211         const struct xprt_class *t;
 212
 213         spin_lock(&xprt_list_lock);
 214         t = xprt_class_find_by_netid_locked(netid);
 215         if (!t) {
 216                 spin_unlock(&xprt_list_lock);
 217                 request_module("rpc%s", netid);
 218                 spin_lock(&xprt_list_lock);
 219                 t = xprt_class_find_by_netid_locked(netid);
 220         }
 221         spin_unlock(&xprt_list_lock);
 222         return t;
 223 }
 224
 225 /**
 226  * xprt_find_transport_ident - convert a netid into a transport identifier
 227  * @netid: transport to load
 228  *
 229  * Returns:
 230  * > 0:         transport identifier
 231  * -ENOENT:     transport module not available
 232  */
 233 int xprt_find_transport_ident(const char *netid)
 234 {
 235         const struct xprt_class *t;
 236         int ret;
 237
 238         t = xprt_class_find_by_netid(netid);
 239         if (!t)
 240                 return -ENOENT;
 241         ret = t->ident;
 242         xprt_class_release(t);
 243         return ret;
 244 }
 245 EXPORT_SYMBOL_GPL(xprt_find_transport_ident);
 246
 247 static void xprt_clear_locked(struct rpc_xprt *xprt)
 248 {
 249         xprt->snd_task = NULL;
 250         if (!test_bit(XPRT_CLOSE_WAIT, &xprt->state))
 251                 clear_bit_unlock(XPRT_LOCKED, &xprt->state);
 252         else
 253                 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 254 }
 255
 256 /**
 257  * xprt_reserve_xprt - serialize write access to transports
 258  * @task: task that is requesting access to the transport
 259  * @xprt: pointer to the target transport
 260  *
 261  * This prevents mixing the payload of separate requests, and prevents
 262  * transport connects from colliding with writes.  No congestion control
 263  * is provided.
 264  */
 265 int xprt_reserve_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 266 {
 267         struct rpc_rqst *req = task->tk_rqstp;
 268
 269         if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
 270                 if (task == xprt->snd_task)
 271                         goto out_locked;
 272                 goto out_sleep;
 273         }
 274         if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
 275                 goto out_unlock;
 276         xprt->snd_task = task;
 277
 278 out_locked:
 279         trace_xprt_reserve_xprt(xprt, task);
 280         return 1;
 281
 282 out_unlock:
 283         xprt_clear_locked(xprt);
 284 out_sleep:
 285         task->tk_status = -EAGAIN;
 286         if (RPC_IS_SOFT(task) || RPC_IS_SOFTCONN(task))
 287                 rpc_sleep_on_timeout(&xprt->sending, task, NULL,
 288                                 xprt_request_timeout(req));
 289         else
 290                 rpc_sleep_on(&xprt->sending, task, NULL);
 291         return 0;
 292 }
 293 EXPORT_SYMBOL_GPL(xprt_reserve_xprt);
 294
 295 static bool
 296 xprt_need_congestion_window_wait(struct rpc_xprt *xprt)
 297 {
 298         return test_bit(XPRT_CWND_WAIT, &xprt->state);
 299 }
 300
 301 static void
 302 xprt_set_congestion_window_wait(struct rpc_xprt *xprt)
 303 {
 304         if (!list_empty(&xprt->xmit_queue)) {
 305                 /* Peek at head of queue to see if it can make progress */
 306                 if (list_first_entry(&xprt->xmit_queue, struct rpc_rqst,
 307                                         rq_xmit)->rq_cong)
 308                         return;
 309         }
 310         set_bit(XPRT_CWND_WAIT, &xprt->state);
 311 }
 312
 313 static void
 314 xprt_test_and_clear_congestion_window_wait(struct rpc_xprt *xprt)
 315 {
 316         if (!RPCXPRT_CONGESTED(xprt))
 317                 clear_bit(XPRT_CWND_WAIT, &xprt->state);
 318 }
 319
 320 /*
 321  * xprt_reserve_xprt_cong - serialize write access to transports
 322  * @task: task that is requesting access to the transport
 323  *
 324  * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
 325  * integrated into the decision of whether a request is allowed to be
 326  * woken up and given access to the transport.
 327  * Note that the lock is only granted if we know there are free slots.
 328  */
 329 int xprt_reserve_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 330 {
 331         struct rpc_rqst *req = task->tk_rqstp;
 332
 333         if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
 334                 if (task == xprt->snd_task)
 335                         goto out_locked;
 336                 goto out_sleep;
 337         }
 338         if (req == NULL) {
 339                 xprt->snd_task = task;
 340                 goto out_locked;
 341         }
 342         if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
 343                 goto out_unlock;
 344         if (!xprt_need_congestion_window_wait(xprt)) {
 345                 xprt->snd_task = task;
 346                 goto out_locked;
 347         }
 348 out_unlock:
 349         xprt_clear_locked(xprt);
 350 out_sleep:
 351         task->tk_status = -EAGAIN;
 352         if (RPC_IS_SOFT(task) || RPC_IS_SOFTCONN(task))
 353                 rpc_sleep_on_timeout(&xprt->sending, task, NULL,
 354                                 xprt_request_timeout(req));
 355         else
 356                 rpc_sleep_on(&xprt->sending, task, NULL);
 357         return 0;
 358 out_locked:
 359         trace_xprt_reserve_cong(xprt, task);
 360         return 1;
 361 }
 362 EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong);
 363
 364 static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
 365 {
 366         int retval;
 367
 368         if (test_bit(XPRT_LOCKED, &xprt->state) && xprt->snd_task == task)
 369                 return 1;
 370         spin_lock(&xprt->transport_lock);
 371         retval = xprt->ops->reserve_xprt(xprt, task);
 372         spin_unlock(&xprt->transport_lock);
 373         return retval;
 374 }
 375
 376 static bool __xprt_lock_write_func(struct rpc_task *task, void *data)
 377 {
 378         struct rpc_xprt *xprt = data;
 379
 380         xprt->snd_task = task;
 381         return true;
 382 }
 383
 384 static void __xprt_lock_write_next(struct rpc_xprt *xprt)
 385 {
 386         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 387                 return;
 388         if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
 389                 goto out_unlock;
 390         if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
 391                                 __xprt_lock_write_func, xprt))
 392                 return;
 393 out_unlock:
 394         xprt_clear_locked(xprt);
 395 }
 396
 397 static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt)
 398 {
 399         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 400                 return;
 401         if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
 402                 goto out_unlock;
 403         if (xprt_need_congestion_window_wait(xprt))
 404                 goto out_unlock;
 405         if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
 406                                 __xprt_lock_write_func, xprt))
 407                 return;
 408 out_unlock:
 409         xprt_clear_locked(xprt);
 410 }
 411
 412 /**
 413  * xprt_release_xprt - allow other requests to use a transport
 414  * @xprt: transport with other tasks potentially waiting
 415  * @task: task that is releasing access to the transport
 416  *
 417  * Note that "task" can be NULL.  No congestion control is provided.
 418  */
 419 void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 420 {
 421         if (xprt->snd_task == task) {
 422                 xprt_clear_locked(xprt);
 423                 __xprt_lock_write_next(xprt);
 424         }
 425         trace_xprt_release_xprt(xprt, task);
 426 }
 427 EXPORT_SYMBOL_GPL(xprt_release_xprt);
 428
 429 /**
 430  * xprt_release_xprt_cong - allow other requests to use a transport
 431  * @xprt: transport with other tasks potentially waiting
 432  * @task: task that is releasing access to the transport
 433  *
 434  * Note that "task" can be NULL.  Another task is awoken to use the
 435  * transport if the transport's congestion window allows it.
 436  */
 437 void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 438 {
 439         if (xprt->snd_task == task) {
 440                 xprt_clear_locked(xprt);
 441                 __xprt_lock_write_next_cong(xprt);
 442         }
 443         trace_xprt_release_cong(xprt, task);
 444 }
 445 EXPORT_SYMBOL_GPL(xprt_release_xprt_cong);
 446
 447 void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
 448 {
 449         if (xprt->snd_task != task)
 450                 return;
 451         spin_lock(&xprt->transport_lock);
 452         xprt->ops->release_xprt(xprt, task);
 453         spin_unlock(&xprt->transport_lock);
 454 }
 455
 456 /*
 457  * Van Jacobson congestion avoidance. Check if the congestion window
 458  * overflowed. Put the task to sleep if this is the case.
 459  */
 460 static int
 461 __xprt_get_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
 462 {
 463         if (req->rq_cong)
 464                 return 1;
 465         trace_xprt_get_cong(xprt, req->rq_task);
 466         if (RPCXPRT_CONGESTED(xprt)) {
 467                 xprt_set_congestion_window_wait(xprt);
 468                 return 0;
 469         }
 470         req->rq_cong = 1;
 471         xprt->cong += RPC_CWNDSCALE;
 472         return 1;
 473 }
 474
 475 /*
 476  * Adjust the congestion window, and wake up the next task
 477  * that has been sleeping due to congestion
 478  */
 479 static void
 480 __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
 481 {
 482         if (!req->rq_cong)
 483                 return;
 484         req->rq_cong = 0;
 485         xprt->cong -= RPC_CWNDSCALE;
 486         xprt_test_and_clear_congestion_window_wait(xprt);
 487         trace_xprt_put_cong(xprt, req->rq_task);
 488         __xprt_lock_write_next_cong(xprt);
 489 }
 490
 491 /**
 492  * xprt_request_get_cong - Request congestion control credits
 493  * @xprt: pointer to transport
 494  * @req: pointer to RPC request
 495  *
 496  * Useful for transports that require congestion control.
 497  */
 498 bool
 499 xprt_request_get_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
 500 {
 501         bool ret = false;
 502
 503         if (req->rq_cong)
 504                 return true;
 505         spin_lock(&xprt->transport_lock);
 506         ret = __xprt_get_cong(xprt, req) != 0;
 507         spin_unlock(&xprt->transport_lock);
 508         return ret;
 509 }
 510 EXPORT_SYMBOL_GPL(xprt_request_get_cong);
 511
 512 /**
 513  * xprt_release_rqst_cong - housekeeping when request is complete
 514  * @task: RPC request that recently completed
 515  *
 516  * Useful for transports that require congestion control.
 517  */
 518 void xprt_release_rqst_cong(struct rpc_task *task)
 519 {
 520         struct rpc_rqst *req = task->tk_rqstp;
 521
 522         __xprt_put_cong(req->rq_xprt, req);
 523 }
 524 EXPORT_SYMBOL_GPL(xprt_release_rqst_cong);
 525
 526 static void xprt_clear_congestion_window_wait_locked(struct rpc_xprt *xprt)
 527 {
 528         if (test_and_clear_bit(XPRT_CWND_WAIT, &xprt->state))
 529                 __xprt_lock_write_next_cong(xprt);
 530 }
 531
 532 /*
 533  * Clear the congestion window wait flag and wake up the next
 534  * entry on xprt->sending
 535  */
 536 static void
 537 xprt_clear_congestion_window_wait(struct rpc_xprt *xprt)
 538 {
 539         if (test_and_clear_bit(XPRT_CWND_WAIT, &xprt->state)) {
 540                 spin_lock(&xprt->transport_lock);
 541                 __xprt_lock_write_next_cong(xprt);
 542                 spin_unlock(&xprt->transport_lock);
 543         }
 544 }
 545
 546 /**
 547  * xprt_adjust_cwnd - adjust transport congestion window
 548  * @xprt: pointer to xprt
 549  * @task: recently completed RPC request used to adjust window
 550  * @result: result code of completed RPC request
 551  *
 552  * The transport code maintains an estimate on the maximum number of out-
 553  * standing RPC requests, using a smoothed version of the congestion
 554  * avoidance implemented in 44BSD. This is basically the Van Jacobson
 555  * congestion algorithm: If a retransmit occurs, the congestion window is
 556  * halved; otherwise, it is incremented by 1/cwnd when
 557  *
 558  *      -       a reply is received and
 559  *      -       a full number of requests are outstanding and
 560  *      -       the congestion window hasn't been updated recently.
 561  */
 562 void xprt_adjust_cwnd(struct rpc_xprt *xprt, struct rpc_task *task, int result)
 563 {
 564         struct rpc_rqst *req = task->tk_rqstp;
 565         unsigned long cwnd = xprt->cwnd;
 566
 567         if (result >= 0 && cwnd <= xprt->cong) {
 568                 /* The (cwnd >> 1) term makes sure
 569                  * the result gets rounded properly. */
 570                 cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
 571                 if (cwnd > RPC_MAXCWND(xprt))
 572                         cwnd = RPC_MAXCWND(xprt);
 573                 __xprt_lock_write_next_cong(xprt);
 574         } else if (result == -ETIMEDOUT) {
 575                 cwnd >>= 1;
 576                 if (cwnd < RPC_CWNDSCALE)
 577                         cwnd = RPC_CWNDSCALE;
 578         }
 579         dprintk("RPC:       cong %ld, cwnd was %ld, now %ld\n",
 580                         xprt->cong, xprt->cwnd, cwnd);
 581         xprt->cwnd = cwnd;
 582         __xprt_put_cong(xprt, req);
 583 }
 584 EXPORT_SYMBOL_GPL(xprt_adjust_cwnd);
 585
 586 /**
 587  * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
 588  * @xprt: transport with waiting tasks
 589  * @status: result code to plant in each task before waking it
 590  *
 591  */
 592 void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
 593 {
 594         if (status < 0)
 595                 rpc_wake_up_status(&xprt->pending, status);
 596         else
 597                 rpc_wake_up(&xprt->pending);
 598 }
 599 EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks);
 600
 601 /**
 602  * xprt_wait_for_buffer_space - wait for transport output buffer to clear
 603  * @xprt: transport
 604  *
 605  * Note that we only set the timer for the case of RPC_IS_SOFT(), since
 606  * we don't in general want to force a socket disconnection due to
 607  * an incomplete RPC call transmission.
 608  */
 609 void xprt_wait_for_buffer_space(struct rpc_xprt *xprt)
 610 {
 611         set_bit(XPRT_WRITE_SPACE, &xprt->state);
 612 }
 613 EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space);
 614
 615 static bool
 616 xprt_clear_write_space_locked(struct rpc_xprt *xprt)
 617 {
 618         if (test_and_clear_bit(XPRT_WRITE_SPACE, &xprt->state)) {
 619                 __xprt_lock_write_next(xprt);
 620                 dprintk("RPC:       write space: waking waiting task on "
 621                                 "xprt %p\n", xprt);
 622                 return true;
 623         }
 624         return false;
 625 }
 626
 627 /**
 628  * xprt_write_space - wake the task waiting for transport output buffer space
 629  * @xprt: transport with waiting tasks
 630  *
 631  * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
 632  */
 633 bool xprt_write_space(struct rpc_xprt *xprt)
 634 {
 635         bool ret;
 636
 637         if (!test_bit(XPRT_WRITE_SPACE, &xprt->state))
 638                 return false;
 639         spin_lock(&xprt->transport_lock);
 640         ret = xprt_clear_write_space_locked(xprt);
 641         spin_unlock(&xprt->transport_lock);
 642         return ret;
 643 }
 644 EXPORT_SYMBOL_GPL(xprt_write_space);
 645
 646 static unsigned long xprt_abs_ktime_to_jiffies(ktime_t abstime)
 647 {
 648         s64 delta = ktime_to_ns(ktime_get() - abstime);
 649         return likely(delta >= 0) ?
 650                 jiffies - nsecs_to_jiffies(delta) :
 651                 jiffies + nsecs_to_jiffies(-delta);
 652 }
 653
 654 static unsigned long xprt_calc_majortimeo(struct rpc_rqst *req,
 655                 const struct rpc_timeout *to)
 656 {
 657         unsigned long majortimeo = req->rq_timeout;
 658
 659         if (to->to_exponential)
 660                 majortimeo <<= to->to_retries;
 661         else
 662                 majortimeo += to->to_increment * to->to_retries;
 663         if (majortimeo > to->to_maxval || majortimeo == 0)
 664                 majortimeo = to->to_maxval;
 665         return majortimeo;
 666 }
 667
 668 static void xprt_reset_majortimeo(struct rpc_rqst *req,
 669                 const struct rpc_timeout *to)
 670 {
 671         req->rq_majortimeo += xprt_calc_majortimeo(req, to);
 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                 const struct rpc_timeout *to)
 681 {
 682         unsigned long time_init;
 683         struct rpc_xprt *xprt = req->rq_xprt;
 684
 685         if (likely(xprt && xprt_connected(xprt)))
 686                 time_init = jiffies;
 687         else
 688                 time_init = xprt_abs_ktime_to_jiffies(task->tk_start);
 689
 690         req->rq_timeout = to->to_initval;
 691         req->rq_majortimeo = time_init + xprt_calc_majortimeo(req, to);
 692         req->rq_minortimeo = time_init + req->rq_timeout;
 693 }
 694
 695 /**
 696  * xprt_adjust_timeout - adjust timeout values for next retransmit
 697  * @req: RPC request containing parameters to use for the adjustment
 698  *
 699  */
 700 int xprt_adjust_timeout(struct rpc_rqst *req)
 701 {
 702         struct rpc_xprt *xprt = req->rq_xprt;
 703         const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
 704         int status = 0;
 705
 706         if (time_before(jiffies, req->rq_majortimeo)) {
 707                 if (time_before(jiffies, req->rq_minortimeo))
 708                         return status;
 709                 if (to->to_exponential)
 710                         req->rq_timeout <<= 1;
 711                 else
 712                         req->rq_timeout += to->to_increment;
 713                 if (to->to_maxval && req->rq_timeout >= to->to_maxval)
 714                         req->rq_timeout = to->to_maxval;
 715                 req->rq_retries++;
 716         } else {
 717                 req->rq_timeout = to->to_initval;
 718                 req->rq_retries = 0;
 719                 xprt_reset_majortimeo(req, to);
 720                 /* Reset the RTT counters == "slow start" */
 721                 spin_lock(&xprt->transport_lock);
 722                 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
 723                 spin_unlock(&xprt->transport_lock);
 724                 status = -ETIMEDOUT;
 725         }
 726         xprt_reset_minortimeo(req);
 727
 728         if (req->rq_timeout == 0) {
 729                 printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
 730                 req->rq_timeout = 5 * HZ;
 731         }
 732         return status;
 733 }
 734
 735 static void xprt_autoclose(struct work_struct *work)
 736 {
 737         struct rpc_xprt *xprt =
 738                 container_of(work, struct rpc_xprt, task_cleanup);
 739         unsigned int pflags = memalloc_nofs_save();
 740
 741         trace_xprt_disconnect_auto(xprt);
 742         xprt->connect_cookie++;
 743         smp_mb__before_atomic();
 744         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 745         xprt->ops->close(xprt);
 746         xprt_release_write(xprt, NULL);
 747         wake_up_bit(&xprt->state, XPRT_LOCKED);
 748         memalloc_nofs_restore(pflags);
 749 }
 750
 751 /**
 752  * xprt_disconnect_done - mark a transport as disconnected
 753  * @xprt: transport to flag for disconnect
 754  *
 755  */
 756 void xprt_disconnect_done(struct rpc_xprt *xprt)
 757 {
 758         trace_xprt_disconnect_done(xprt);
 759         spin_lock(&xprt->transport_lock);
 760         xprt_clear_connected(xprt);
 761         xprt_clear_write_space_locked(xprt);
 762         xprt_clear_congestion_window_wait_locked(xprt);
 763         xprt_wake_pending_tasks(xprt, -ENOTCONN);
 764         spin_unlock(&xprt->transport_lock);
 765 }
 766 EXPORT_SYMBOL_GPL(xprt_disconnect_done);
 767
 768 /**
 769  * xprt_schedule_autoclose_locked - Try to schedule an autoclose RPC call
 770  * @xprt: transport to disconnect
 771  */
 772 static void xprt_schedule_autoclose_locked(struct rpc_xprt *xprt)
 773 {
 774         if (test_and_set_bit(XPRT_CLOSE_WAIT, &xprt->state))
 775                 return;
 776         if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 777                 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 778         else if (xprt->snd_task && !test_bit(XPRT_SND_IS_COOKIE, &xprt->state))
 779                 rpc_wake_up_queued_task_set_status(&xprt->pending,
 780                                                    xprt->snd_task, -ENOTCONN);
 781 }
 782
 783 /**
 784  * xprt_force_disconnect - force a transport to disconnect
 785  * @xprt: transport to disconnect
 786  *
 787  */
 788 void xprt_force_disconnect(struct rpc_xprt *xprt)
 789 {
 790         trace_xprt_disconnect_force(xprt);
 791
 792         /* Don't race with the test_bit() in xprt_clear_locked() */
 793         spin_lock(&xprt->transport_lock);
 794         xprt_schedule_autoclose_locked(xprt);
 795         spin_unlock(&xprt->transport_lock);
 796 }
 797 EXPORT_SYMBOL_GPL(xprt_force_disconnect);
 798
 799 static unsigned int
 800 xprt_connect_cookie(struct rpc_xprt *xprt)
 801 {
 802         return READ_ONCE(xprt->connect_cookie);
 803 }
 804
 805 static bool
 806 xprt_request_retransmit_after_disconnect(struct rpc_task *task)
 807 {
 808         struct rpc_rqst *req = task->tk_rqstp;
 809         struct rpc_xprt *xprt = req->rq_xprt;
 810
 811         return req->rq_connect_cookie != xprt_connect_cookie(xprt) ||
 812                 !xprt_connected(xprt);
 813 }
 814
 815 /**
 816  * xprt_conditional_disconnect - force a transport to disconnect
 817  * @xprt: transport to disconnect
 818  * @cookie: 'connection cookie'
 819  *
 820  * This attempts to break the connection if and only if 'cookie' matches
 821  * the current transport 'connection cookie'. It ensures that we don't
 822  * try to break the connection more than once when we need to retransmit
 823  * a batch of RPC requests.
 824  *
 825  */
 826 void xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie)
 827 {
 828         /* Don't race with the test_bit() in xprt_clear_locked() */
 829         spin_lock(&xprt->transport_lock);
 830         if (cookie != xprt->connect_cookie)
 831                 goto out;
 832         if (test_bit(XPRT_CLOSING, &xprt->state))
 833                 goto out;
 834         xprt_schedule_autoclose_locked(xprt);
 835 out:
 836         spin_unlock(&xprt->transport_lock);
 837 }
 838
 839 static bool
 840 xprt_has_timer(const struct rpc_xprt *xprt)
 841 {
 842         return xprt->idle_timeout != 0;
 843 }
 844
 845 static void
 846 xprt_schedule_autodisconnect(struct rpc_xprt *xprt)
 847         __must_hold(&xprt->transport_lock)
 848 {
 849         xprt->last_used = jiffies;
 850         if (RB_EMPTY_ROOT(&xprt->recv_queue) && xprt_has_timer(xprt))
 851                 mod_timer(&xprt->timer, xprt->last_used + xprt->idle_timeout);
 852 }
 853
 854 static void
 855 xprt_init_autodisconnect(struct timer_list *t)
 856 {
 857         struct rpc_xprt *xprt = from_timer(xprt, t, timer);
 858
 859         if (!RB_EMPTY_ROOT(&xprt->recv_queue))
 860                 return;
 861         /* Reset xprt->last_used to avoid connect/autodisconnect cycling */
 862         xprt->last_used = jiffies;
 863         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 864                 return;
 865         queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 866 }
 867
 868 #if IS_ENABLED(CONFIG_FAIL_SUNRPC)
 869 static void xprt_inject_disconnect(struct rpc_xprt *xprt)
 870 {
 871         if (!fail_sunrpc.ignore_client_disconnect &&
 872             should_fail(&fail_sunrpc.attr, 1))
 873                 xprt->ops->inject_disconnect(xprt);
 874 }
 875 #else
 876 static inline void xprt_inject_disconnect(struct rpc_xprt *xprt)
 877 {
 878 }
 879 #endif
 880
 881 bool xprt_lock_connect(struct rpc_xprt *xprt,
 882                 struct rpc_task *task,
 883                 void *cookie)
 884 {
 885         bool ret = false;
 886
 887         spin_lock(&xprt->transport_lock);
 888         if (!test_bit(XPRT_LOCKED, &xprt->state))
 889                 goto out;
 890         if (xprt->snd_task != task)
 891                 goto out;
 892         set_bit(XPRT_SND_IS_COOKIE, &xprt->state);
 893         xprt->snd_task = cookie;
 894         ret = true;
 895 out:
 896         spin_unlock(&xprt->transport_lock);
 897         return ret;
 898 }
 899 EXPORT_SYMBOL_GPL(xprt_lock_connect);
 900
 901 void xprt_unlock_connect(struct rpc_xprt *xprt, void *cookie)
 902 {
 903         spin_lock(&xprt->transport_lock);
 904         if (xprt->snd_task != cookie)
 905                 goto out;
 906         if (!test_bit(XPRT_LOCKED, &xprt->state))
 907                 goto out;
 908         xprt->snd_task =NULL;
 909         clear_bit(XPRT_SND_IS_COOKIE, &xprt->state);
 910         xprt->ops->release_xprt(xprt, NULL);
 911         xprt_schedule_autodisconnect(xprt);
 912 out:
 913         spin_unlock(&xprt->transport_lock);
 914         wake_up_bit(&xprt->state, XPRT_LOCKED);
 915 }
 916 EXPORT_SYMBOL_GPL(xprt_unlock_connect);
 917
 918 /**
 919  * xprt_connect - schedule a transport connect operation
 920  * @task: RPC task that is requesting the connect
 921  *
 922  */
 923 void xprt_connect(struct rpc_task *task)
 924 {
 925         struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
 926
 927         trace_xprt_connect(xprt);
 928
 929         if (!xprt_bound(xprt)) {
 930                 task->tk_status = -EAGAIN;
 931                 return;
 932         }
 933         if (!xprt_lock_write(xprt, task))
 934                 return;
 935
 936         if (!xprt_connected(xprt) && !test_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
 937                 task->tk_rqstp->rq_connect_cookie = xprt->connect_cookie;
 938                 rpc_sleep_on_timeout(&xprt->pending, task, NULL,
 939                                 xprt_request_timeout(task->tk_rqstp));
 940
 941                 if (test_bit(XPRT_CLOSING, &xprt->state))
 942                         return;
 943                 if (xprt_test_and_set_connecting(xprt))
 944                         return;
 945                 /* Race breaker */
 946                 if (!xprt_connected(xprt)) {
 947                         xprt->stat.connect_start = jiffies;
 948                         xprt->ops->connect(xprt, task);
 949                 } else {
 950                         xprt_clear_connecting(xprt);
 951                         task->tk_status = 0;
 952                         rpc_wake_up_queued_task(&xprt->pending, task);
 953                 }
 954         }
 955         xprt_release_write(xprt, task);
 956 }
 957
 958 /**
 959  * xprt_reconnect_delay - compute the wait before scheduling a connect
 960  * @xprt: transport instance
 961  *
 962  */
 963 unsigned long xprt_reconnect_delay(const struct rpc_xprt *xprt)
 964 {
 965         unsigned long start, now = jiffies;
 966
 967         start = xprt->stat.connect_start + xprt->reestablish_timeout;
 968         if (time_after(start, now))
 969                 return start - now;
 970         return 0;
 971 }
 972 EXPORT_SYMBOL_GPL(xprt_reconnect_delay);
 973
 974 /**
 975  * xprt_reconnect_backoff - compute the new re-establish timeout
 976  * @xprt: transport instance
 977  * @init_to: initial reestablish timeout
 978  *
 979  */
 980 void xprt_reconnect_backoff(struct rpc_xprt *xprt, unsigned long init_to)
 981 {
 982         xprt->reestablish_timeout <<= 1;
 983         if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
 984                 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
 985         if (xprt->reestablish_timeout < init_to)
 986                 xprt->reestablish_timeout = init_to;
 987 }
 988 EXPORT_SYMBOL_GPL(xprt_reconnect_backoff);
 989
 990 enum xprt_xid_rb_cmp {
 991         XID_RB_EQUAL,
 992         XID_RB_LEFT,
 993         XID_RB_RIGHT,
 994 };
 995 static enum xprt_xid_rb_cmp
 996 xprt_xid_cmp(__be32 xid1, __be32 xid2)
 997 {
 998         if (xid1 == xid2)
 999                 return XID_RB_EQUAL;
1000         if ((__force u32)xid1 < (__force u32)xid2)
1001                 return XID_RB_LEFT;
1002         return XID_RB_RIGHT;
1003 }
1004
1005 static struct rpc_rqst *
1006 xprt_request_rb_find(struct rpc_xprt *xprt, __be32 xid)
1007 {
1008         struct rb_node *n = xprt->recv_queue.rb_node;
1009         struct rpc_rqst *req;
1010
1011         while (n != NULL) {
1012                 req = rb_entry(n, struct rpc_rqst, rq_recv);
1013                 switch (xprt_xid_cmp(xid, req->rq_xid)) {
1014                 case XID_RB_LEFT:
1015                         n = n->rb_left;
1016                         break;
1017                 case XID_RB_RIGHT:
1018                         n = n->rb_right;
1019                         break;
1020                 case XID_RB_EQUAL:
1021                         return req;
1022                 }
1023         }
1024         return NULL;
1025 }
1026
1027 static void
1028 xprt_request_rb_insert(struct rpc_xprt *xprt, struct rpc_rqst *new)
1029 {
1030         struct rb_node **p = &xprt->recv_queue.rb_node;
1031         struct rb_node *n = NULL;
1032         struct rpc_rqst *req;
1033
1034         while (*p != NULL) {
1035                 n = *p;
1036                 req = rb_entry(n, struct rpc_rqst, rq_recv);
1037                 switch(xprt_xid_cmp(new->rq_xid, req->rq_xid)) {
1038                 case XID_RB_LEFT:
1039                         p = &n->rb_left;
1040                         break;
1041                 case XID_RB_RIGHT:
1042                         p = &n->rb_right;
1043                         break;
1044                 case XID_RB_EQUAL:
1045                         WARN_ON_ONCE(new != req);
1046                         return;
1047                 }
1048         }
1049         rb_link_node(&new->rq_recv, n, p);
1050         rb_insert_color(&new->rq_recv, &xprt->recv_queue);
1051 }
1052
1053 static void
1054 xprt_request_rb_remove(struct rpc_xprt *xprt, struct rpc_rqst *req)
1055 {
1056         rb_erase(&req->rq_recv, &xprt->recv_queue);
1057 }
1058
1059 /**
1060  * xprt_lookup_rqst - find an RPC request corresponding to an XID
1061  * @xprt: transport on which the original request was transmitted
1062  * @xid: RPC XID of incoming reply
1063  *
1064  * Caller holds xprt->queue_lock.
1065  */
1066 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
1067 {
1068         struct rpc_rqst *entry;
1069
1070         entry = xprt_request_rb_find(xprt, xid);
1071         if (entry != NULL) {
1072                 trace_xprt_lookup_rqst(xprt, xid, 0);
1073                 entry->rq_rtt = ktime_sub(ktime_get(), entry->rq_xtime);
1074                 return entry;
1075         }
1076
1077         dprintk("RPC:       xprt_lookup_rqst did not find xid %08x\n",
1078                         ntohl(xid));
1079         trace_xprt_lookup_rqst(xprt, xid, -ENOENT);
1080         xprt->stat.bad_xids++;
1081         return NULL;
1082 }
1083 EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
1084
1085 static bool
1086 xprt_is_pinned_rqst(struct rpc_rqst *req)
1087 {
1088         return atomic_read(&req->rq_pin) != 0;
1089 }
1090
1091 /**
1092  * xprt_pin_rqst - Pin a request on the transport receive list
1093  * @req: Request to pin
1094  *
1095  * Caller must ensure this is atomic with the call to xprt_lookup_rqst()
1096  * so should be holding xprt->queue_lock.
1097  */
1098 void xprt_pin_rqst(struct rpc_rqst *req)
1099 {
1100         atomic_inc(&req->rq_pin);
1101 }
1102 EXPORT_SYMBOL_GPL(xprt_pin_rqst);
1103
1104 /**
1105  * xprt_unpin_rqst - Unpin a request on the transport receive list
1106  * @req: Request to pin
1107  *
1108  * Caller should be holding xprt->queue_lock.
1109  */
1110 void xprt_unpin_rqst(struct rpc_rqst *req)
1111 {
1112         if (!test_bit(RPC_TASK_MSG_PIN_WAIT, &req->rq_task->tk_runstate)) {
1113                 atomic_dec(&req->rq_pin);
1114                 return;
1115         }
1116         if (atomic_dec_and_test(&req->rq_pin))
1117                 wake_up_var(&req->rq_pin);
1118 }
1119 EXPORT_SYMBOL_GPL(xprt_unpin_rqst);
1120
1121 static void xprt_wait_on_pinned_rqst(struct rpc_rqst *req)
1122 {
1123         wait_var_event(&req->rq_pin, !xprt_is_pinned_rqst(req));
1124 }
1125
1126 static bool
1127 xprt_request_data_received(struct rpc_task *task)
1128 {
1129         return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
1130                 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) != 0;
1131 }
1132
1133 static bool
1134 xprt_request_need_enqueue_receive(struct rpc_task *task, struct rpc_rqst *req)
1135 {
1136         return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
1137                 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) == 0;
1138 }
1139
1140 /**
1141  * xprt_request_enqueue_receive - Add an request to the receive queue
1142  * @task: RPC task
1143  *
1144  */
1145 int
1146 xprt_request_enqueue_receive(struct rpc_task *task)
1147 {
1148         struct rpc_rqst *req = task->tk_rqstp;
1149         struct rpc_xprt *xprt = req->rq_xprt;
1150         int ret;
1151
1152         if (!xprt_request_need_enqueue_receive(task, req))
1153                 return 0;
1154
1155         ret = xprt_request_prepare(task->tk_rqstp, &req->rq_rcv_buf);
1156         if (ret)
1157                 return ret;
1158         spin_lock(&xprt->queue_lock);
1159
1160         /* Update the softirq receive buffer */
1161         memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
1162                         sizeof(req->rq_private_buf));
1163
1164         /* Add request to the receive list */
1165         xprt_request_rb_insert(xprt, req);
1166         set_bit(RPC_TASK_NEED_RECV, &task->tk_runstate);
1167         spin_unlock(&xprt->queue_lock);
1168
1169         /* Turn off autodisconnect */
1170         del_timer_sync(&xprt->timer);
1171         return 0;
1172 }
1173
1174 /**
1175  * xprt_request_dequeue_receive_locked - Remove a request from the receive queue
1176  * @task: RPC task
1177  *
1178  * Caller must hold xprt->queue_lock.
1179  */
1180 static void
1181 xprt_request_dequeue_receive_locked(struct rpc_task *task)
1182 {
1183         struct rpc_rqst *req = task->tk_rqstp;
1184
1185         if (test_and_clear_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1186                 xprt_request_rb_remove(req->rq_xprt, req);
1187 }
1188
1189 /**
1190  * xprt_update_rtt - Update RPC RTT statistics
1191  * @task: RPC request that recently completed
1192  *
1193  * Caller holds xprt->queue_lock.
1194  */
1195 void xprt_update_rtt(struct rpc_task *task)
1196 {
1197         struct rpc_rqst *req = task->tk_rqstp;
1198         struct rpc_rtt *rtt = task->tk_client->cl_rtt;
1199         unsigned int timer = task->tk_msg.rpc_proc->p_timer;
1200         long m = usecs_to_jiffies(ktime_to_us(req->rq_rtt));
1201
1202         if (timer) {
1203                 if (req->rq_ntrans == 1)
1204                         rpc_update_rtt(rtt, timer, m);
1205                 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
1206         }
1207 }
1208 EXPORT_SYMBOL_GPL(xprt_update_rtt);
1209
1210 /**
1211  * xprt_complete_rqst - called when reply processing is complete
1212  * @task: RPC request that recently completed
1213  * @copied: actual number of bytes received from the transport
1214  *
1215  * Caller holds xprt->queue_lock.
1216  */
1217 void xprt_complete_rqst(struct rpc_task *task, int copied)
1218 {
1219         struct rpc_rqst *req = task->tk_rqstp;
1220         struct rpc_xprt *xprt = req->rq_xprt;
1221
1222         xprt->stat.recvs++;
1223
1224         xdr_free_bvec(&req->rq_rcv_buf);
1225         req->rq_private_buf.bvec = NULL;
1226         req->rq_private_buf.len = copied;
1227         /* Ensure all writes are done before we update */
1228         /* req->rq_reply_bytes_recvd */
1229         smp_wmb();
1230         req->rq_reply_bytes_recvd = copied;
1231         xprt_request_dequeue_receive_locked(task);
1232         rpc_wake_up_queued_task(&xprt->pending, task);
1233 }
1234 EXPORT_SYMBOL_GPL(xprt_complete_rqst);
1235
1236 static void xprt_timer(struct rpc_task *task)
1237 {
1238         struct rpc_rqst *req = task->tk_rqstp;
1239         struct rpc_xprt *xprt = req->rq_xprt;
1240
1241         if (task->tk_status != -ETIMEDOUT)
1242                 return;
1243
1244         trace_xprt_timer(xprt, req->rq_xid, task->tk_status);
1245         if (!req->rq_reply_bytes_recvd) {
1246                 if (xprt->ops->timer)
1247                         xprt->ops->timer(xprt, task);
1248         } else
1249                 task->tk_status = 0;
1250 }
1251
1252 /**
1253  * xprt_wait_for_reply_request_def - wait for reply
1254  * @task: pointer to rpc_task
1255  *
1256  * Set a request's retransmit timeout based on the transport's
1257  * default timeout parameters.  Used by transports that don't adjust
1258  * the retransmit timeout based on round-trip time estimation,
1259  * and put the task to sleep on the pending queue.
1260  */
1261 void xprt_wait_for_reply_request_def(struct rpc_task *task)
1262 {
1263         struct rpc_rqst *req = task->tk_rqstp;
1264
1265         rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1266                         xprt_request_timeout(req));
1267 }
1268 EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_def);
1269
1270 /**
1271  * xprt_wait_for_reply_request_rtt - wait for reply using RTT estimator
1272  * @task: pointer to rpc_task
1273  *
1274  * Set a request's retransmit timeout using the RTT estimator,
1275  * and put the task to sleep on the pending queue.
1276  */
1277 void xprt_wait_for_reply_request_rtt(struct rpc_task *task)
1278 {
1279         int timer = task->tk_msg.rpc_proc->p_timer;
1280         struct rpc_clnt *clnt = task->tk_client;
1281         struct rpc_rtt *rtt = clnt->cl_rtt;
1282         struct rpc_rqst *req = task->tk_rqstp;
1283         unsigned long max_timeout = clnt->cl_timeout->to_maxval;
1284         unsigned long timeout;
1285
1286         timeout = rpc_calc_rto(rtt, timer);
1287         timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
1288         if (timeout > max_timeout || timeout == 0)
1289                 timeout = max_timeout;
1290         rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1291                         jiffies + timeout);
1292 }
1293 EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_rtt);
1294
1295 /**
1296  * xprt_request_wait_receive - wait for the reply to an RPC request
1297  * @task: RPC task about to send a request
1298  *
1299  */
1300 void xprt_request_wait_receive(struct rpc_task *task)
1301 {
1302         struct rpc_rqst *req = task->tk_rqstp;
1303         struct rpc_xprt *xprt = req->rq_xprt;
1304
1305         if (!test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1306                 return;
1307         /*
1308          * Sleep on the pending queue if we're expecting a reply.
1309          * The spinlock ensures atomicity between the test of
1310          * req->rq_reply_bytes_recvd, and the call to rpc_sleep_on().
1311          */
1312         spin_lock(&xprt->queue_lock);
1313         if (test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate)) {
1314                 xprt->ops->wait_for_reply_request(task);
1315                 /*
1316                  * Send an extra queue wakeup call if the
1317                  * connection was dropped in case the call to
1318                  * rpc_sleep_on() raced.
1319                  */
1320                 if (xprt_request_retransmit_after_disconnect(task))
1321                         rpc_wake_up_queued_task_set_status(&xprt->pending,
1322                                         task, -ENOTCONN);
1323         }
1324         spin_unlock(&xprt->queue_lock);
1325 }
1326
1327 static bool
1328 xprt_request_need_enqueue_transmit(struct rpc_task *task, struct rpc_rqst *req)
1329 {
1330         return !test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1331 }
1332
1333 /**
1334  * xprt_request_enqueue_transmit - queue a task for transmission
1335  * @task: pointer to rpc_task
1336  *
1337  * Add a task to the transmission queue.
1338  */
1339 void
1340 xprt_request_enqueue_transmit(struct rpc_task *task)
1341 {
1342         struct rpc_rqst *pos, *req = task->tk_rqstp;
1343         struct rpc_xprt *xprt = req->rq_xprt;
1344         int ret;
1345
1346         if (xprt_request_need_enqueue_transmit(task, req)) {
1347                 ret = xprt_request_prepare(task->tk_rqstp, &req->rq_snd_buf);
1348                 if (ret) {
1349                         task->tk_status = ret;
1350                         return;
1351                 }
1352                 req->rq_bytes_sent = 0;
1353                 spin_lock(&xprt->queue_lock);
1354                 /*
1355                  * Requests that carry congestion control credits are added
1356                  * to the head of the list to avoid starvation issues.
1357                  */
1358                 if (req->rq_cong) {
1359                         xprt_clear_congestion_window_wait(xprt);
1360                         list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1361                                 if (pos->rq_cong)
1362                                         continue;
1363                                 /* Note: req is added _before_ pos */
1364                                 list_add_tail(&req->rq_xmit, &pos->rq_xmit);
1365                                 INIT_LIST_HEAD(&req->rq_xmit2);
1366                                 goto out;
1367                         }
1368                 } else if (!req->rq_seqno) {
1369                         list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1370                                 if (pos->rq_task->tk_owner != task->tk_owner)
1371                                         continue;
1372                                 list_add_tail(&req->rq_xmit2, &pos->rq_xmit2);
1373                                 INIT_LIST_HEAD(&req->rq_xmit);
1374                                 goto out;
1375                         }
1376                 }
1377                 list_add_tail(&req->rq_xmit, &xprt->xmit_queue);
1378                 INIT_LIST_HEAD(&req->rq_xmit2);
1379 out:
1380                 atomic_long_inc(&xprt->xmit_queuelen);
1381                 set_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1382                 spin_unlock(&xprt->queue_lock);
1383         }
1384 }
1385
1386 /**
1387  * xprt_request_dequeue_transmit_locked - remove a task from the transmission queue
1388  * @task: pointer to rpc_task
1389  *
1390  * Remove a task from the transmission queue
1391  * Caller must hold xprt->queue_lock
1392  */
1393 static void
1394 xprt_request_dequeue_transmit_locked(struct rpc_task *task)
1395 {
1396         struct rpc_rqst *req = task->tk_rqstp;
1397
1398         if (!test_and_clear_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1399                 return;
1400         if (!list_empty(&req->rq_xmit)) {
1401                 list_del(&req->rq_xmit);
1402                 if (!list_empty(&req->rq_xmit2)) {
1403                         struct rpc_rqst *next = list_first_entry(&req->rq_xmit2,
1404                                         struct rpc_rqst, rq_xmit2);
1405                         list_del(&req->rq_xmit2);
1406                         list_add_tail(&next->rq_xmit, &next->rq_xprt->xmit_queue);
1407                 }
1408         } else
1409                 list_del(&req->rq_xmit2);
1410         atomic_long_dec(&req->rq_xprt->xmit_queuelen);
1411         xdr_free_bvec(&req->rq_snd_buf);
1412 }
1413
1414 /**
1415  * xprt_request_dequeue_transmit - remove a task from the transmission queue
1416  * @task: pointer to rpc_task
1417  *
1418  * Remove a task from the transmission queue
1419  */
1420 static void
1421 xprt_request_dequeue_transmit(struct rpc_task *task)
1422 {
1423         struct rpc_rqst *req = task->tk_rqstp;
1424         struct rpc_xprt *xprt = req->rq_xprt;
1425
1426         spin_lock(&xprt->queue_lock);
1427         xprt_request_dequeue_transmit_locked(task);
1428         spin_unlock(&xprt->queue_lock);
1429 }
1430
1431 /**
1432  * xprt_request_dequeue_xprt - remove a task from the transmit+receive queue
1433  * @task: pointer to rpc_task
1434  *
1435  * Remove a task from the transmit and receive queues, and ensure that
1436  * it is not pinned by the receive work item.
1437  */
1438 void
1439 xprt_request_dequeue_xprt(struct rpc_task *task)
1440 {
1441         struct rpc_rqst *req = task->tk_rqstp;
1442         struct rpc_xprt *xprt = req->rq_xprt;
1443
1444         if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate) ||
1445             test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) ||
1446             xprt_is_pinned_rqst(req)) {
1447                 spin_lock(&xprt->queue_lock);
1448                 while (xprt_is_pinned_rqst(req)) {
1449                         set_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1450                         spin_unlock(&xprt->queue_lock);
1451                         xprt_wait_on_pinned_rqst(req);
1452                         spin_lock(&xprt->queue_lock);
1453                         clear_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1454                 }
1455                 xprt_request_dequeue_transmit_locked(task);
1456                 xprt_request_dequeue_receive_locked(task);
1457                 spin_unlock(&xprt->queue_lock);
1458                 xdr_free_bvec(&req->rq_rcv_buf);
1459         }
1460 }
1461
1462 /**
1463  * xprt_request_prepare - prepare an encoded request for transport
1464  * @req: pointer to rpc_rqst
1465  * @buf: pointer to send/rcv xdr_buf
1466  *
1467  * Calls into the transport layer to do whatever is needed to prepare
1468  * the request for transmission or receive.
1469  * Returns error, or zero.
1470  */
1471 static int
1472 xprt_request_prepare(struct rpc_rqst *req, struct xdr_buf *buf)
1473 {
1474         struct rpc_xprt *xprt = req->rq_xprt;
1475
1476         if (xprt->ops->prepare_request)
1477                 return xprt->ops->prepare_request(req, buf);
1478         return 0;
1479 }
1480
1481 /**
1482  * xprt_request_need_retransmit - Test if a task needs retransmission
1483  * @task: pointer to rpc_task
1484  *
1485  * Test for whether a connection breakage requires the task to retransmit
1486  */
1487 bool
1488 xprt_request_need_retransmit(struct rpc_task *task)
1489 {
1490         return xprt_request_retransmit_after_disconnect(task);
1491 }
1492
1493 /**
1494  * xprt_prepare_transmit - reserve the transport before sending a request
1495  * @task: RPC task about to send a request
1496  *
1497  */
1498 bool xprt_prepare_transmit(struct rpc_task *task)
1499 {
1500         struct rpc_rqst *req = task->tk_rqstp;
1501         struct rpc_xprt *xprt = req->rq_xprt;
1502
1503         if (!xprt_lock_write(xprt, task)) {
1504                 /* Race breaker: someone may have transmitted us */
1505                 if (!test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1506                         rpc_wake_up_queued_task_set_status(&xprt->sending,
1507                                         task, 0);
1508                 return false;
1509
1510         }
1511         if (atomic_read(&xprt->swapper))
1512                 /* This will be clear in __rpc_execute */
1513                 current->flags |= PF_MEMALLOC;
1514         return true;
1515 }
1516
1517 void xprt_end_transmit(struct rpc_task *task)
1518 {
1519         struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
1520
1521         xprt_inject_disconnect(xprt);
1522         xprt_release_write(xprt, task);
1523 }
1524
1525 /**
1526  * xprt_request_transmit - send an RPC request on a transport
1527  * @req: pointer to request to transmit
1528  * @snd_task: RPC task that owns the transport lock
1529  *
1530  * This performs the transmission of a single request.
1531  * Note that if the request is not the same as snd_task, then it
1532  * does need to be pinned.
1533  * Returns '0' on success.
1534  */
1535 static int
1536 xprt_request_transmit(struct rpc_rqst *req, struct rpc_task *snd_task)
1537 {
1538         struct rpc_xprt *xprt = req->rq_xprt;
1539         struct rpc_task *task = req->rq_task;
1540         unsigned int connect_cookie;
1541         int is_retrans = RPC_WAS_SENT(task);
1542         int status;
1543
1544         if (!req->rq_bytes_sent) {
1545                 if (xprt_request_data_received(task)) {
1546                         status = 0;
1547                         goto out_dequeue;
1548                 }
1549                 /* Verify that our message lies in the RPCSEC_GSS window */
1550                 if (rpcauth_xmit_need_reencode(task)) {
1551                         status = -EBADMSG;
1552                         goto out_dequeue;
1553                 }
1554                 if (RPC_SIGNALLED(task)) {
1555                         status = -ERESTARTSYS;
1556                         goto out_dequeue;
1557                 }
1558         }
1559
1560         /*
1561          * Update req->rq_ntrans before transmitting to avoid races with
1562          * xprt_update_rtt(), which needs to know that it is recording a
1563          * reply to the first transmission.
1564          */
1565         req->rq_ntrans++;
1566
1567         trace_rpc_xdr_sendto(task, &req->rq_snd_buf);
1568         connect_cookie = xprt->connect_cookie;
1569         status = xprt->ops->send_request(req);
1570         if (status != 0) {
1571                 req->rq_ntrans--;
1572                 trace_xprt_transmit(req, status);
1573                 return status;
1574         }
1575
1576         if (is_retrans) {
1577                 task->tk_client->cl_stats->rpcretrans++;
1578                 trace_xprt_retransmit(req);
1579         }
1580
1581         xprt_inject_disconnect(xprt);
1582
1583         task->tk_flags |= RPC_TASK_SENT;
1584         spin_lock(&xprt->transport_lock);
1585
1586         xprt->stat.sends++;
1587         xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
1588         xprt->stat.bklog_u += xprt->backlog.qlen;
1589         xprt->stat.sending_u += xprt->sending.qlen;
1590         xprt->stat.pending_u += xprt->pending.qlen;
1591         spin_unlock(&xprt->transport_lock);
1592
1593         req->rq_connect_cookie = connect_cookie;
1594 out_dequeue:
1595         trace_xprt_transmit(req, status);
1596         xprt_request_dequeue_transmit(task);
1597         rpc_wake_up_queued_task_set_status(&xprt->sending, task, status);
1598         return status;
1599 }
1600
1601 /**
1602  * xprt_transmit - send an RPC request on a transport
1603  * @task: controlling RPC task
1604  *
1605  * Attempts to drain the transmit queue. On exit, either the transport
1606  * signalled an error that needs to be handled before transmission can
1607  * resume, or @task finished transmitting, and detected that it already
1608  * received a reply.
1609  */
1610 void
1611 xprt_transmit(struct rpc_task *task)
1612 {
1613         struct rpc_rqst *next, *req = task->tk_rqstp;
1614         struct rpc_xprt *xprt = req->rq_xprt;
1615         int status;
1616
1617         spin_lock(&xprt->queue_lock);
1618         for (;;) {
1619                 next = list_first_entry_or_null(&xprt->xmit_queue,
1620                                                 struct rpc_rqst, rq_xmit);
1621                 if (!next)
1622                         break;
1623                 xprt_pin_rqst(next);
1624                 spin_unlock(&xprt->queue_lock);
1625                 status = xprt_request_transmit(next, task);
1626                 if (status == -EBADMSG && next != req)
1627                         status = 0;
1628                 spin_lock(&xprt->queue_lock);
1629                 xprt_unpin_rqst(next);
1630                 if (status < 0) {
1631                         if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1632                                 task->tk_status = status;
1633                         break;
1634                 }
1635                 /* Was @task transmitted, and has it received a reply? */
1636                 if (xprt_request_data_received(task) &&
1637                     !test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1638                         break;
1639                 cond_resched_lock(&xprt->queue_lock);
1640         }
1641         spin_unlock(&xprt->queue_lock);
1642 }
1643
1644 static void xprt_complete_request_init(struct rpc_task *task)
1645 {
1646         if (task->tk_rqstp)
1647                 xprt_request_init(task);
1648 }
1649
1650 void xprt_add_backlog(struct rpc_xprt *xprt, struct rpc_task *task)
1651 {
1652         set_bit(XPRT_CONGESTED, &xprt->state);
1653         rpc_sleep_on(&xprt->backlog, task, xprt_complete_request_init);
1654 }
1655 EXPORT_SYMBOL_GPL(xprt_add_backlog);
1656
1657 static bool __xprt_set_rq(struct rpc_task *task, void *data)
1658 {
1659         struct rpc_rqst *req = data;
1660
1661         if (task->tk_rqstp == NULL) {
1662                 memset(req, 0, sizeof(*req));   /* mark unused */
1663                 task->tk_rqstp = req;
1664                 return true;
1665         }
1666         return false;
1667 }
1668
1669 bool xprt_wake_up_backlog(struct rpc_xprt *xprt, struct rpc_rqst *req)
1670 {
1671         if (rpc_wake_up_first(&xprt->backlog, __xprt_set_rq, req) == NULL) {
1672                 clear_bit(XPRT_CONGESTED, &xprt->state);
1673                 return false;
1674         }
1675         return true;
1676 }
1677 EXPORT_SYMBOL_GPL(xprt_wake_up_backlog);
1678
1679 static bool xprt_throttle_congested(struct rpc_xprt *xprt, struct rpc_task *task)
1680 {
1681         bool ret = false;
1682
1683         if (!test_bit(XPRT_CONGESTED, &xprt->state))
1684                 goto out;
1685         spin_lock(&xprt->reserve_lock);
1686         if (test_bit(XPRT_CONGESTED, &xprt->state)) {
1687                 xprt_add_backlog(xprt, task);
1688                 ret = true;
1689         }
1690         spin_unlock(&xprt->reserve_lock);
1691 out:
1692         return ret;
1693 }
1694
1695 static struct rpc_rqst *xprt_dynamic_alloc_slot(struct rpc_xprt *xprt)
1696 {
1697         struct rpc_rqst *req = ERR_PTR(-EAGAIN);
1698
1699         if (xprt->num_reqs >= xprt->max_reqs)
1700                 goto out;
1701         ++xprt->num_reqs;
1702         spin_unlock(&xprt->reserve_lock);
1703         req = kzalloc(sizeof(*req), rpc_task_gfp_mask());
1704         spin_lock(&xprt->reserve_lock);
1705         if (req != NULL)
1706                 goto out;
1707         --xprt->num_reqs;
1708         req = ERR_PTR(-ENOMEM);
1709 out:
1710         return req;
1711 }
1712
1713 static bool xprt_dynamic_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1714 {
1715         if (xprt->num_reqs > xprt->min_reqs) {
1716                 --xprt->num_reqs;
1717                 kfree(req);
1718                 return true;
1719         }
1720         return false;
1721 }
1722
1723 void xprt_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1724 {
1725         struct rpc_rqst *req;
1726
1727         spin_lock(&xprt->reserve_lock);
1728         if (!list_empty(&xprt->free)) {
1729                 req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
1730                 list_del(&req->rq_list);
1731                 goto out_init_req;
1732         }
1733         req = xprt_dynamic_alloc_slot(xprt);
1734         if (!IS_ERR(req))
1735                 goto out_init_req;
1736         switch (PTR_ERR(req)) {
1737         case -ENOMEM:
1738                 dprintk("RPC:       dynamic allocation of request slot "
1739                                 "failed! Retrying\n");
1740                 task->tk_status = -ENOMEM;
1741                 break;
1742         case -EAGAIN:
1743                 xprt_add_backlog(xprt, task);
1744                 dprintk("RPC:       waiting for request slot\n");
1745                 fallthrough;
1746         default:
1747                 task->tk_status = -EAGAIN;
1748         }
1749         spin_unlock(&xprt->reserve_lock);
1750         return;
1751 out_init_req:
1752         xprt->stat.max_slots = max_t(unsigned int, xprt->stat.max_slots,
1753                                      xprt->num_reqs);
1754         spin_unlock(&xprt->reserve_lock);
1755
1756         task->tk_status = 0;
1757         task->tk_rqstp = req;
1758 }
1759 EXPORT_SYMBOL_GPL(xprt_alloc_slot);
1760
1761 void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1762 {
1763         spin_lock(&xprt->reserve_lock);
1764         if (!xprt_wake_up_backlog(xprt, req) &&
1765             !xprt_dynamic_free_slot(xprt, req)) {
1766                 memset(req, 0, sizeof(*req));   /* mark unused */
1767                 list_add(&req->rq_list, &xprt->free);
1768         }
1769         spin_unlock(&xprt->reserve_lock);
1770 }
1771 EXPORT_SYMBOL_GPL(xprt_free_slot);
1772
1773 static void xprt_free_all_slots(struct rpc_xprt *xprt)
1774 {
1775         struct rpc_rqst *req;
1776         while (!list_empty(&xprt->free)) {
1777                 req = list_first_entry(&xprt->free, struct rpc_rqst, rq_list);
1778                 list_del(&req->rq_list);
1779                 kfree(req);
1780         }
1781 }
1782
1783 static DEFINE_IDA(rpc_xprt_ids);
1784
1785 void xprt_cleanup_ids(void)
1786 {
1787         ida_destroy(&rpc_xprt_ids);
1788 }
1789
1790 static int xprt_alloc_id(struct rpc_xprt *xprt)
1791 {
1792         int id;
1793
1794         id = ida_alloc(&rpc_xprt_ids, GFP_KERNEL);
1795         if (id < 0)
1796                 return id;
1797
1798         xprt->id = id;
1799         return 0;
1800 }
1801
1802 static void xprt_free_id(struct rpc_xprt *xprt)
1803 {
1804         ida_free(&rpc_xprt_ids, xprt->id);
1805 }
1806
1807 struct rpc_xprt *xprt_alloc(struct net *net, size_t size,
1808                 unsigned int num_prealloc,
1809                 unsigned int max_alloc)
1810 {
1811         struct rpc_xprt *xprt;
1812         struct rpc_rqst *req;
1813         int i;
1814
1815         xprt = kzalloc(size, GFP_KERNEL);
1816         if (xprt == NULL)
1817                 goto out;
1818
1819         xprt_alloc_id(xprt);
1820         xprt_init(xprt, net);
1821
1822         for (i = 0; i < num_prealloc; i++) {
1823                 req = kzalloc(sizeof(struct rpc_rqst), GFP_KERNEL);
1824                 if (!req)
1825                         goto out_free;
1826                 list_add(&req->rq_list, &xprt->free);
1827         }
1828         xprt->max_reqs = max_t(unsigned int, max_alloc, num_prealloc);
1829         xprt->min_reqs = num_prealloc;
1830         xprt->num_reqs = num_prealloc;
1831
1832         return xprt;
1833
1834 out_free:
1835         xprt_free(xprt);
1836 out:
1837         return NULL;
1838 }
1839 EXPORT_SYMBOL_GPL(xprt_alloc);
1840
1841 void xprt_free(struct rpc_xprt *xprt)
1842 {
1843         put_net_track(xprt->xprt_net, &xprt->ns_tracker);
1844         xprt_free_all_slots(xprt);
1845         xprt_free_id(xprt);
1846         rpc_sysfs_xprt_destroy(xprt);
1847         kfree_rcu(xprt, rcu);
1848 }
1849 EXPORT_SYMBOL_GPL(xprt_free);
1850
1851 static void
1852 xprt_init_connect_cookie(struct rpc_rqst *req, struct rpc_xprt *xprt)
1853 {
1854         req->rq_connect_cookie = xprt_connect_cookie(xprt) - 1;
1855 }
1856
1857 static __be32
1858 xprt_alloc_xid(struct rpc_xprt *xprt)
1859 {
1860         __be32 xid;
1861
1862         spin_lock(&xprt->reserve_lock);
1863         xid = (__force __be32)xprt->xid++;
1864         spin_unlock(&xprt->reserve_lock);
1865         return xid;
1866 }
1867
1868 static void
1869 xprt_init_xid(struct rpc_xprt *xprt)
1870 {
1871         xprt->xid = get_random_u32();
1872 }
1873
1874 static void
1875 xprt_request_init(struct rpc_task *task)
1876 {
1877         struct rpc_xprt *xprt = task->tk_xprt;
1878         struct rpc_rqst *req = task->tk_rqstp;
1879
1880         req->rq_task    = task;
1881         req->rq_xprt    = xprt;
1882         req->rq_buffer  = NULL;
1883         req->rq_xid     = xprt_alloc_xid(xprt);
1884         xprt_init_connect_cookie(req, xprt);
1885         req->rq_snd_buf.len = 0;
1886         req->rq_snd_buf.buflen = 0;
1887         req->rq_rcv_buf.len = 0;
1888         req->rq_rcv_buf.buflen = 0;
1889         req->rq_snd_buf.bvec = NULL;
1890         req->rq_rcv_buf.bvec = NULL;
1891         req->rq_release_snd_buf = NULL;
1892         xprt_init_majortimeo(task, req, task->tk_client->cl_timeout);
1893
1894         trace_xprt_reserve(req);
1895 }
1896
1897 static void
1898 xprt_do_reserve(struct rpc_xprt *xprt, struct rpc_task *task)
1899 {
1900         xprt->ops->alloc_slot(xprt, task);
1901         if (task->tk_rqstp != NULL)
1902                 xprt_request_init(task);
1903 }
1904
1905 /**
1906  * xprt_reserve - allocate an RPC request slot
1907  * @task: RPC task requesting a slot allocation
1908  *
1909  * If the transport is marked as being congested, or if no more
1910  * slots are available, place the task on the transport's
1911  * backlog queue.
1912  */
1913 void xprt_reserve(struct rpc_task *task)
1914 {
1915         struct rpc_xprt *xprt = task->tk_xprt;
1916
1917         task->tk_status = 0;
1918         if (task->tk_rqstp != NULL)
1919                 return;
1920
1921         task->tk_status = -EAGAIN;
1922         if (!xprt_throttle_congested(xprt, task))
1923                 xprt_do_reserve(xprt, task);
1924 }
1925
1926 /**
1927  * xprt_retry_reserve - allocate an RPC request slot
1928  * @task: RPC task requesting a slot allocation
1929  *
1930  * If no more slots are available, place the task on the transport's
1931  * backlog queue.
1932  * Note that the only difference with xprt_reserve is that we now
1933  * ignore the value of the XPRT_CONGESTED flag.
1934  */
1935 void xprt_retry_reserve(struct rpc_task *task)
1936 {
1937         struct rpc_xprt *xprt = task->tk_xprt;
1938
1939         task->tk_status = 0;
1940         if (task->tk_rqstp != NULL)
1941                 return;
1942
1943         task->tk_status = -EAGAIN;
1944         xprt_do_reserve(xprt, task);
1945 }
1946
1947 /**
1948  * xprt_release - release an RPC request slot
1949  * @task: task which is finished with the slot
1950  *
1951  */
1952 void xprt_release(struct rpc_task *task)
1953 {
1954         struct rpc_xprt *xprt;
1955         struct rpc_rqst *req = task->tk_rqstp;
1956
1957         if (req == NULL) {
1958                 if (task->tk_client) {
1959                         xprt = task->tk_xprt;
1960                         xprt_release_write(xprt, task);
1961                 }
1962                 return;
1963         }
1964
1965         xprt = req->rq_xprt;
1966         xprt_request_dequeue_xprt(task);
1967         spin_lock(&xprt->transport_lock);
1968         xprt->ops->release_xprt(xprt, task);
1969         if (xprt->ops->release_request)
1970                 xprt->ops->release_request(task);
1971         xprt_schedule_autodisconnect(xprt);
1972         spin_unlock(&xprt->transport_lock);
1973         if (req->rq_buffer)
1974                 xprt->ops->buf_free(task);
1975         if (req->rq_cred != NULL)
1976                 put_rpccred(req->rq_cred);
1977         if (req->rq_release_snd_buf)
1978                 req->rq_release_snd_buf(req);
1979
1980         task->tk_rqstp = NULL;
1981         if (likely(!bc_prealloc(req)))
1982                 xprt->ops->free_slot(xprt, req);
1983         else
1984                 xprt_free_bc_request(req);
1985 }
1986
1987 #ifdef CONFIG_SUNRPC_BACKCHANNEL
1988 void
1989 xprt_init_bc_request(struct rpc_rqst *req, struct rpc_task *task,
1990                 const struct rpc_timeout *to)
1991 {
1992         struct xdr_buf *xbufp = &req->rq_snd_buf;
1993
1994         task->tk_rqstp = req;
1995         req->rq_task = task;
1996         xprt_init_connect_cookie(req, req->rq_xprt);
1997         /*
1998          * Set up the xdr_buf length.
1999          * This also indicates that the buffer is XDR encoded already.
2000          */
2001         xbufp->len = xbufp->head[0].iov_len + xbufp->page_len +
2002                 xbufp->tail[0].iov_len;
2003         /*
2004          * Backchannel Replies are sent with !RPC_TASK_SOFT and
2005          * RPC_TASK_NO_RETRANS_TIMEOUT. The major timeout setting
2006          * affects only how long each Reply waits to be sent when
2007          * a transport connection cannot be established.
2008          */
2009         xprt_init_majortimeo(task, req, to);
2010 }
2011 #endif
2012
2013 static void xprt_init(struct rpc_xprt *xprt, struct net *net)
2014 {
2015         kref_init(&xprt->kref);
2016
2017         spin_lock_init(&xprt->transport_lock);
2018         spin_lock_init(&xprt->reserve_lock);
2019         spin_lock_init(&xprt->queue_lock);
2020
2021         INIT_LIST_HEAD(&xprt->free);
2022         xprt->recv_queue = RB_ROOT;
2023         INIT_LIST_HEAD(&xprt->xmit_queue);
2024 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
2025         spin_lock_init(&xprt->bc_pa_lock);
2026         INIT_LIST_HEAD(&xprt->bc_pa_list);
2027 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
2028         INIT_LIST_HEAD(&xprt->xprt_switch);
2029
2030         xprt->last_used = jiffies;
2031         xprt->cwnd = RPC_INITCWND;
2032         xprt->bind_index = 0;
2033
2034         rpc_init_wait_queue(&xprt->binding, "xprt_binding");
2035         rpc_init_wait_queue(&xprt->pending, "xprt_pending");
2036         rpc_init_wait_queue(&xprt->sending, "xprt_sending");
2037         rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
2038
2039         xprt_init_xid(xprt);
2040
2041         xprt->xprt_net = get_net_track(net, &xprt->ns_tracker, GFP_KERNEL);
2042 }
2043
2044 /**
2045  * xprt_create_transport - create an RPC transport
2046  * @args: rpc transport creation arguments
2047  *
2048  */
2049 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
2050 {
2051         struct rpc_xprt *xprt;
2052         const struct xprt_class *t;
2053
2054         t = xprt_class_find_by_ident(args->ident);
2055         if (!t) {
2056                 dprintk("RPC: transport (%d) not supported\n", args->ident);
2057                 return ERR_PTR(-EIO);
2058         }
2059
2060         xprt = t->setup(args);
2061         xprt_class_release(t);
2062
2063         if (IS_ERR(xprt))
2064                 goto out;
2065         if (args->flags & XPRT_CREATE_NO_IDLE_TIMEOUT)
2066                 xprt->idle_timeout = 0;
2067         INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
2068         if (xprt_has_timer(xprt))
2069                 timer_setup(&xprt->timer, xprt_init_autodisconnect, 0);
2070         else
2071                 timer_setup(&xprt->timer, NULL, 0);
2072
2073         if (strlen(args->servername) > RPC_MAXNETNAMELEN) {
2074                 xprt_destroy(xprt);
2075                 return ERR_PTR(-EINVAL);
2076         }
2077         xprt->servername = kstrdup(args->servername, GFP_KERNEL);
2078         if (xprt->servername == NULL) {
2079                 xprt_destroy(xprt);
2080                 return ERR_PTR(-ENOMEM);
2081         }
2082
2083         rpc_xprt_debugfs_register(xprt);
2084
2085         trace_xprt_create(xprt);
2086 out:
2087         return xprt;
2088 }
2089
2090 static void xprt_destroy_cb(struct work_struct *work)
2091 {
2092         struct rpc_xprt *xprt =
2093                 container_of(work, struct rpc_xprt, task_cleanup);
2094
2095         trace_xprt_destroy(xprt);
2096
2097         rpc_xprt_debugfs_unregister(xprt);
2098         rpc_destroy_wait_queue(&xprt->binding);
2099         rpc_destroy_wait_queue(&xprt->pending);
2100         rpc_destroy_wait_queue(&xprt->sending);
2101         rpc_destroy_wait_queue(&xprt->backlog);
2102         kfree(xprt->servername);
2103         /*
2104          * Destroy any existing back channel
2105          */
2106         xprt_destroy_backchannel(xprt, UINT_MAX);
2107
2108         /*
2109          * Tear down transport state and free the rpc_xprt
2110          */
2111         xprt->ops->destroy(xprt);
2112 }
2113
2114 /**
2115  * xprt_destroy - destroy an RPC transport, killing off all requests.
2116  * @xprt: transport to destroy
2117  *
2118  */
2119 static void xprt_destroy(struct rpc_xprt *xprt)
2120 {
2121         /*
2122          * Exclude transport connect/disconnect handlers and autoclose
2123          */
2124         wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_UNINTERRUPTIBLE);
2125
2126         /*
2127          * xprt_schedule_autodisconnect() can run after XPRT_LOCKED
2128          * is cleared.  We use ->transport_lock to ensure the mod_timer()
2129          * can only run *before* del_time_sync(), never after.
2130          */
2131         spin_lock(&xprt->transport_lock);
2132         del_timer_sync(&xprt->timer);
2133         spin_unlock(&xprt->transport_lock);
2134
2135         /*
2136          * Destroy sockets etc from the system workqueue so they can
2137          * safely flush receive work running on rpciod.
2138          */
2139         INIT_WORK(&xprt->task_cleanup, xprt_destroy_cb);
2140         schedule_work(&xprt->task_cleanup);
2141 }
2142
2143 static void xprt_destroy_kref(struct kref *kref)
2144 {
2145         xprt_destroy(container_of(kref, struct rpc_xprt, kref));
2146 }
2147
2148 /**
2149  * xprt_get - return a reference to an RPC transport.
2150  * @xprt: pointer to the transport
2151  *
2152  */
2153 struct rpc_xprt *xprt_get(struct rpc_xprt *xprt)
2154 {
2155         if (xprt != NULL && kref_get_unless_zero(&xprt->kref))
2156                 return xprt;
2157         return NULL;
2158 }
2159 EXPORT_SYMBOL_GPL(xprt_get);
2160
2161 /**
2162  * xprt_put - release a reference to an RPC transport.
2163  * @xprt: pointer to the transport
2164  *
2165  */
2166 void xprt_put(struct rpc_xprt *xprt)
2167 {
2168         if (xprt != NULL)
2169                 kref_put(&xprt->kref, xprt_destroy_kref);
2170 }
2171 EXPORT_SYMBOL_GPL(xprt_put);
2172
2173 void xprt_set_offline_locked(struct rpc_xprt *xprt, struct rpc_xprt_switch *xps)
2174 {
2175         if (!test_and_set_bit(XPRT_OFFLINE, &xprt->state)) {
2176                 spin_lock(&xps->xps_lock);
2177                 xps->xps_nactive--;
2178                 spin_unlock(&xps->xps_lock);
2179         }
2180 }
2181
2182 void xprt_set_online_locked(struct rpc_xprt *xprt, struct rpc_xprt_switch *xps)
2183 {
2184         if (test_and_clear_bit(XPRT_OFFLINE, &xprt->state)) {
2185                 spin_lock(&xps->xps_lock);
2186                 xps->xps_nactive++;
2187                 spin_unlock(&xps->xps_lock);
2188         }
2189 }
2190
2191 void xprt_delete_locked(struct rpc_xprt *xprt, struct rpc_xprt_switch *xps)
2192 {
2193         if (test_and_set_bit(XPRT_REMOVE, &xprt->state))
2194                 return;
2195
2196         xprt_force_disconnect(xprt);
2197         if (!test_bit(XPRT_CONNECTED, &xprt->state))
2198                 return;
2199
2200         if (!xprt->sending.qlen && !xprt->pending.qlen &&
2201             !xprt->backlog.qlen && !atomic_long_read(&xprt->queuelen))
2202                 rpc_xprt_switch_remove_xprt(xps, xprt, true);
2203 }