net/sunrpc/xprt.c

   1 /*
   2  *  linux/net/sunrpc/xprt.c
   3  *
   4  *  This is a generic RPC call interface supporting congestion avoidance,
   5  *  and asynchronous calls.
   6  *
   7  *  The interface works like this:
   8  *
   9  *  -   When a process places a call, it allocates a request slot if
  10  *      one is available. Otherwise, it sleeps on the backlog queue
  11  *      (xprt_reserve).
  12  *  -   Next, the caller puts together the RPC message, stuffs it into
  13  *      the request struct, and calls xprt_transmit().
  14  *  -   xprt_transmit sends the message and installs the caller on the
  15  *      transport's wait list. At the same time, if a reply is expected,
  16  *      it installs a timer that is run after the packet's timeout has
  17  *      expired.
  18  *  -   When a packet arrives, the data_ready handler walks the list of
  19  *      pending requests for that transport. If a matching XID is found, the
  20  *      caller is woken up, and the timer removed.
  21  *  -   When no reply arrives within the timeout interval, the timer is
  22  *      fired by the kernel and runs xprt_timer(). It either adjusts the
  23  *      timeout values (minor timeout) or wakes up the caller with a status
  24  *      of -ETIMEDOUT.
  25  *  -   When the caller receives a notification from RPC that a reply arrived,
  26  *      it should release the RPC slot, and process the reply.
  27  *      If the call timed out, it may choose to retry the operation by
  28  *      adjusting the initial timeout value, and simply calling rpc_call
  29  *      again.
  30  *
  31  *  Support for async RPC is done through a set of RPC-specific scheduling
  32  *  primitives that `transparently' work for processes as well as async
  33  *  tasks that rely on callbacks.
  34  *
  35  *  Copyright (C) 1995-1997, Olaf Kirch <[email protected]>
  36  *
  37  *  Transport switch API copyright (C) 2005, Chuck Lever <[email protected]>
  38  */
  39
  40 #include <linux/module.h>
  41
  42 #include <linux/types.h>
  43 #include <linux/interrupt.h>
  44 #include <linux/workqueue.h>
  45 #include <linux/net.h>
  46 #include <linux/ktime.h>
  47
  48 #include <linux/sunrpc/clnt.h>
  49 #include <linux/sunrpc/metrics.h>
  50 #include <linux/sunrpc/bc_xprt.h>
  51
  52 #include <trace/events/sunrpc.h>
  53
  54 #include "sunrpc.h"
  55
  56 /*
  57  * Local variables
  58  */
  59
  60 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
  61 # define RPCDBG_FACILITY        RPCDBG_XPRT
  62 #endif
  63
  64 /*
  65  * Local functions
  66  */
  67 static void      xprt_init(struct rpc_xprt *xprt, struct net *net);
  68 static void     xprt_request_init(struct rpc_task *, struct rpc_xprt *);
  69 static void     xprt_connect_status(struct rpc_task *task);
  70 static int      __xprt_get_cong(struct rpc_xprt *, struct rpc_task *);
  71 static void     __xprt_put_cong(struct rpc_xprt *, struct rpc_rqst *);
  72 static void      xprt_destroy(struct rpc_xprt *xprt);
  73
  74 static DEFINE_SPINLOCK(xprt_list_lock);
  75 static LIST_HEAD(xprt_list);
  76
  77 /**
  78  * xprt_register_transport - register a transport implementation
  79  * @transport: transport to register
  80  *
  81  * If a transport implementation is loaded as a kernel module, it can
  82  * call this interface to make itself known to the RPC client.
  83  *
  84  * Returns:
  85  * 0:           transport successfully registered
  86  * -EEXIST:     transport already registered
  87  * -EINVAL:     transport module being unloaded
  88  */
  89 int xprt_register_transport(struct xprt_class *transport)
  90 {
  91         struct xprt_class *t;
  92         int result;
  93
  94         result = -EEXIST;
  95         spin_lock(&xprt_list_lock);
  96         list_for_each_entry(t, &xprt_list, list) {
  97                 /* don't register the same transport class twice */
  98                 if (t->ident == transport->ident)
  99                         goto out;
 100         }
 101
 102         list_add_tail(&transport->list, &xprt_list);
 103         printk(KERN_INFO "RPC: Registered %s transport module.\n",
 104                transport->name);
 105         result = 0;
 106
 107 out:
 108         spin_unlock(&xprt_list_lock);
 109         return result;
 110 }
 111 EXPORT_SYMBOL_GPL(xprt_register_transport);
 112
 113 /**
 114  * xprt_unregister_transport - unregister a transport implementation
 115  * @transport: transport to unregister
 116  *
 117  * Returns:
 118  * 0:           transport successfully unregistered
 119  * -ENOENT:     transport never registered
 120  */
 121 int xprt_unregister_transport(struct xprt_class *transport)
 122 {
 123         struct xprt_class *t;
 124         int result;
 125
 126         result = 0;
 127         spin_lock(&xprt_list_lock);
 128         list_for_each_entry(t, &xprt_list, list) {
 129                 if (t == transport) {
 130                         printk(KERN_INFO
 131                                 "RPC: Unregistered %s transport module.\n",
 132                                 transport->name);
 133                         list_del_init(&transport->list);
 134                         goto out;
 135                 }
 136         }
 137         result = -ENOENT;
 138
 139 out:
 140         spin_unlock(&xprt_list_lock);
 141         return result;
 142 }
 143 EXPORT_SYMBOL_GPL(xprt_unregister_transport);
 144
 145 /**
 146  * xprt_load_transport - load a transport implementation
 147  * @transport_name: transport to load
 148  *
 149  * Returns:
 150  * 0:           transport successfully loaded
 151  * -ENOENT:     transport module not available
 152  */
 153 int xprt_load_transport(const char *transport_name)
 154 {
 155         struct xprt_class *t;
 156         int result;
 157
 158         result = 0;
 159         spin_lock(&xprt_list_lock);
 160         list_for_each_entry(t, &xprt_list, list) {
 161                 if (strcmp(t->name, transport_name) == 0) {
 162                         spin_unlock(&xprt_list_lock);
 163                         goto out;
 164                 }
 165         }
 166         spin_unlock(&xprt_list_lock);
 167         result = request_module("xprt%s", transport_name);
 168 out:
 169         return result;
 170 }
 171 EXPORT_SYMBOL_GPL(xprt_load_transport);
 172
 173 /**
 174  * xprt_reserve_xprt - serialize write access to transports
 175  * @task: task that is requesting access to the transport
 176  * @xprt: pointer to the target transport
 177  *
 178  * This prevents mixing the payload of separate requests, and prevents
 179  * transport connects from colliding with writes.  No congestion control
 180  * is provided.
 181  */
 182 int xprt_reserve_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 183 {
 184         struct rpc_rqst *req = task->tk_rqstp;
 185         int priority;
 186
 187         if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
 188                 if (task == xprt->snd_task)
 189                         return 1;
 190                 goto out_sleep;
 191         }
 192         xprt->snd_task = task;
 193         if (req != NULL)
 194                 req->rq_ntrans++;
 195
 196         return 1;
 197
 198 out_sleep:
 199         dprintk("RPC: %5u failed to lock transport %p\n",
 200                         task->tk_pid, xprt);
 201         task->tk_timeout = 0;
 202         task->tk_status = -EAGAIN;
 203         if (req == NULL)
 204                 priority = RPC_PRIORITY_LOW;
 205         else if (!req->rq_ntrans)
 206                 priority = RPC_PRIORITY_NORMAL;
 207         else
 208                 priority = RPC_PRIORITY_HIGH;
 209         rpc_sleep_on_priority(&xprt->sending, task, NULL, priority);
 210         return 0;
 211 }
 212 EXPORT_SYMBOL_GPL(xprt_reserve_xprt);
 213
 214 static void xprt_clear_locked(struct rpc_xprt *xprt)
 215 {
 216         xprt->snd_task = NULL;
 217         if (!test_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
 218                 smp_mb__before_atomic();
 219                 clear_bit(XPRT_LOCKED, &xprt->state);
 220                 smp_mb__after_atomic();
 221         } else
 222                 queue_work(rpciod_workqueue, &xprt->task_cleanup);
 223 }
 224
 225 /*
 226  * xprt_reserve_xprt_cong - serialize write access to transports
 227  * @task: task that is requesting access to the transport
 228  *
 229  * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
 230  * integrated into the decision of whether a request is allowed to be
 231  * woken up and given access to the transport.
 232  */
 233 int xprt_reserve_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 234 {
 235         struct rpc_rqst *req = task->tk_rqstp;
 236         int priority;
 237
 238         if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
 239                 if (task == xprt->snd_task)
 240                         return 1;
 241                 goto out_sleep;
 242         }
 243         if (req == NULL) {
 244                 xprt->snd_task = task;
 245                 return 1;
 246         }
 247         if (__xprt_get_cong(xprt, task)) {
 248                 xprt->snd_task = task;
 249                 req->rq_ntrans++;
 250                 return 1;
 251         }
 252         xprt_clear_locked(xprt);
 253 out_sleep:
 254         if (req)
 255                 __xprt_put_cong(xprt, req);
 256         dprintk("RPC: %5u failed to lock transport %p\n", task->tk_pid, xprt);
 257         task->tk_timeout = 0;
 258         task->tk_status = -EAGAIN;
 259         if (req == NULL)
 260                 priority = RPC_PRIORITY_LOW;
 261         else if (!req->rq_ntrans)
 262                 priority = RPC_PRIORITY_NORMAL;
 263         else
 264                 priority = RPC_PRIORITY_HIGH;
 265         rpc_sleep_on_priority(&xprt->sending, task, NULL, priority);
 266         return 0;
 267 }
 268 EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong);
 269
 270 static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
 271 {
 272         int retval;
 273
 274         spin_lock_bh(&xprt->transport_lock);
 275         retval = xprt->ops->reserve_xprt(xprt, task);
 276         spin_unlock_bh(&xprt->transport_lock);
 277         return retval;
 278 }
 279
 280 static bool __xprt_lock_write_func(struct rpc_task *task, void *data)
 281 {
 282         struct rpc_xprt *xprt = data;
 283         struct rpc_rqst *req;
 284
 285         req = task->tk_rqstp;
 286         xprt->snd_task = task;
 287         if (req)
 288                 req->rq_ntrans++;
 289         return true;
 290 }
 291
 292 static void __xprt_lock_write_next(struct rpc_xprt *xprt)
 293 {
 294         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 295                 return;
 296
 297         if (rpc_wake_up_first(&xprt->sending, __xprt_lock_write_func, xprt))
 298                 return;
 299         xprt_clear_locked(xprt);
 300 }
 301
 302 static bool __xprt_lock_write_cong_func(struct rpc_task *task, void *data)
 303 {
 304         struct rpc_xprt *xprt = data;
 305         struct rpc_rqst *req;
 306
 307         req = task->tk_rqstp;
 308         if (req == NULL) {
 309                 xprt->snd_task = task;
 310                 return true;
 311         }
 312         if (__xprt_get_cong(xprt, task)) {
 313                 xprt->snd_task = task;
 314                 req->rq_ntrans++;
 315                 return true;
 316         }
 317         return false;
 318 }
 319
 320 static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt)
 321 {
 322         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 323                 return;
 324         if (RPCXPRT_CONGESTED(xprt))
 325                 goto out_unlock;
 326         if (rpc_wake_up_first(&xprt->sending, __xprt_lock_write_cong_func, xprt))
 327                 return;
 328 out_unlock:
 329         xprt_clear_locked(xprt);
 330 }
 331
 332 static void xprt_task_clear_bytes_sent(struct rpc_task *task)
 333 {
 334         if (task != NULL) {
 335                 struct rpc_rqst *req = task->tk_rqstp;
 336                 if (req != NULL)
 337                         req->rq_bytes_sent = 0;
 338         }
 339 }
 340
 341 /**
 342  * xprt_release_xprt - allow other requests to use a transport
 343  * @xprt: transport with other tasks potentially waiting
 344  * @task: task that is releasing access to the transport
 345  *
 346  * Note that "task" can be NULL.  No congestion control is provided.
 347  */
 348 void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 349 {
 350         if (xprt->snd_task == task) {
 351                 xprt_task_clear_bytes_sent(task);
 352                 xprt_clear_locked(xprt);
 353                 __xprt_lock_write_next(xprt);
 354         }
 355 }
 356 EXPORT_SYMBOL_GPL(xprt_release_xprt);
 357
 358 /**
 359  * xprt_release_xprt_cong - allow other requests to use a transport
 360  * @xprt: transport with other tasks potentially waiting
 361  * @task: task that is releasing access to the transport
 362  *
 363  * Note that "task" can be NULL.  Another task is awoken to use the
 364  * transport if the transport's congestion window allows it.
 365  */
 366 void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 367 {
 368         if (xprt->snd_task == task) {
 369                 xprt_task_clear_bytes_sent(task);
 370                 xprt_clear_locked(xprt);
 371                 __xprt_lock_write_next_cong(xprt);
 372         }
 373 }
 374 EXPORT_SYMBOL_GPL(xprt_release_xprt_cong);
 375
 376 static inline void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
 377 {
 378         spin_lock_bh(&xprt->transport_lock);
 379         xprt->ops->release_xprt(xprt, task);
 380         spin_unlock_bh(&xprt->transport_lock);
 381 }
 382
 383 /*
 384  * Van Jacobson congestion avoidance. Check if the congestion window
 385  * overflowed. Put the task to sleep if this is the case.
 386  */
 387 static int
 388 __xprt_get_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 389 {
 390         struct rpc_rqst *req = task->tk_rqstp;
 391
 392         if (req->rq_cong)
 393                 return 1;
 394         dprintk("RPC: %5u xprt_cwnd_limited cong = %lu cwnd = %lu\n",
 395                         task->tk_pid, xprt->cong, xprt->cwnd);
 396         if (RPCXPRT_CONGESTED(xprt))
 397                 return 0;
 398         req->rq_cong = 1;
 399         xprt->cong += RPC_CWNDSCALE;
 400         return 1;
 401 }
 402
 403 /*
 404  * Adjust the congestion window, and wake up the next task
 405  * that has been sleeping due to congestion
 406  */
 407 static void
 408 __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
 409 {
 410         if (!req->rq_cong)
 411                 return;
 412         req->rq_cong = 0;
 413         xprt->cong -= RPC_CWNDSCALE;
 414         __xprt_lock_write_next_cong(xprt);
 415 }
 416
 417 /**
 418  * xprt_release_rqst_cong - housekeeping when request is complete
 419  * @task: RPC request that recently completed
 420  *
 421  * Useful for transports that require congestion control.
 422  */
 423 void xprt_release_rqst_cong(struct rpc_task *task)
 424 {
 425         struct rpc_rqst *req = task->tk_rqstp;
 426
 427         __xprt_put_cong(req->rq_xprt, req);
 428 }
 429 EXPORT_SYMBOL_GPL(xprt_release_rqst_cong);
 430
 431 /**
 432  * xprt_adjust_cwnd - adjust transport congestion window
 433  * @xprt: pointer to xprt
 434  * @task: recently completed RPC request used to adjust window
 435  * @result: result code of completed RPC request
 436  *
 437  * The transport code maintains an estimate on the maximum number of out-
 438  * standing RPC requests, using a smoothed version of the congestion
 439  * avoidance implemented in 44BSD. This is basically the Van Jacobson
 440  * congestion algorithm: If a retransmit occurs, the congestion window is
 441  * halved; otherwise, it is incremented by 1/cwnd when
 442  *
 443  *      -       a reply is received and
 444  *      -       a full number of requests are outstanding and
 445  *      -       the congestion window hasn't been updated recently.
 446  */
 447 void xprt_adjust_cwnd(struct rpc_xprt *xprt, struct rpc_task *task, int result)
 448 {
 449         struct rpc_rqst *req = task->tk_rqstp;
 450         unsigned long cwnd = xprt->cwnd;
 451
 452         if (result >= 0 && cwnd <= xprt->cong) {
 453                 /* The (cwnd >> 1) term makes sure
 454                  * the result gets rounded properly. */
 455                 cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
 456                 if (cwnd > RPC_MAXCWND(xprt))
 457                         cwnd = RPC_MAXCWND(xprt);
 458                 __xprt_lock_write_next_cong(xprt);
 459         } else if (result == -ETIMEDOUT) {
 460                 cwnd >>= 1;
 461                 if (cwnd < RPC_CWNDSCALE)
 462                         cwnd = RPC_CWNDSCALE;
 463         }
 464         dprintk("RPC:       cong %ld, cwnd was %ld, now %ld\n",
 465                         xprt->cong, xprt->cwnd, cwnd);
 466         xprt->cwnd = cwnd;
 467         __xprt_put_cong(xprt, req);
 468 }
 469 EXPORT_SYMBOL_GPL(xprt_adjust_cwnd);
 470
 471 /**
 472  * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
 473  * @xprt: transport with waiting tasks
 474  * @status: result code to plant in each task before waking it
 475  *
 476  */
 477 void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
 478 {
 479         if (status < 0)
 480                 rpc_wake_up_status(&xprt->pending, status);
 481         else
 482                 rpc_wake_up(&xprt->pending);
 483 }
 484 EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks);
 485
 486 /**
 487  * xprt_wait_for_buffer_space - wait for transport output buffer to clear
 488  * @task: task to be put to sleep
 489  * @action: function pointer to be executed after wait
 490  *
 491  * Note that we only set the timer for the case of RPC_IS_SOFT(), since
 492  * we don't in general want to force a socket disconnection due to
 493  * an incomplete RPC call transmission.
 494  */
 495 void xprt_wait_for_buffer_space(struct rpc_task *task, rpc_action action)
 496 {
 497         struct rpc_rqst *req = task->tk_rqstp;
 498         struct rpc_xprt *xprt = req->rq_xprt;
 499
 500         task->tk_timeout = RPC_IS_SOFT(task) ? req->rq_timeout : 0;
 501         rpc_sleep_on(&xprt->pending, task, action);
 502 }
 503 EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space);
 504
 505 /**
 506  * xprt_write_space - wake the task waiting for transport output buffer space
 507  * @xprt: transport with waiting tasks
 508  *
 509  * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
 510  */
 511 void xprt_write_space(struct rpc_xprt *xprt)
 512 {
 513         spin_lock_bh(&xprt->transport_lock);
 514         if (xprt->snd_task) {
 515                 dprintk("RPC:       write space: waking waiting task on "
 516                                 "xprt %p\n", xprt);
 517                 rpc_wake_up_queued_task(&xprt->pending, xprt->snd_task);
 518         }
 519         spin_unlock_bh(&xprt->transport_lock);
 520 }
 521 EXPORT_SYMBOL_GPL(xprt_write_space);
 522
 523 /**
 524  * xprt_set_retrans_timeout_def - set a request's retransmit timeout
 525  * @task: task whose timeout is to be set
 526  *
 527  * Set a request's retransmit timeout based on the transport's
 528  * default timeout parameters.  Used by transports that don't adjust
 529  * the retransmit timeout based on round-trip time estimation.
 530  */
 531 void xprt_set_retrans_timeout_def(struct rpc_task *task)
 532 {
 533         task->tk_timeout = task->tk_rqstp->rq_timeout;
 534 }
 535 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_def);
 536
 537 /**
 538  * xprt_set_retrans_timeout_rtt - set a request's retransmit timeout
 539  * @task: task whose timeout is to be set
 540  *
 541  * Set a request's retransmit timeout using the RTT estimator.
 542  */
 543 void xprt_set_retrans_timeout_rtt(struct rpc_task *task)
 544 {
 545         int timer = task->tk_msg.rpc_proc->p_timer;
 546         struct rpc_clnt *clnt = task->tk_client;
 547         struct rpc_rtt *rtt = clnt->cl_rtt;
 548         struct rpc_rqst *req = task->tk_rqstp;
 549         unsigned long max_timeout = clnt->cl_timeout->to_maxval;
 550
 551         task->tk_timeout = rpc_calc_rto(rtt, timer);
 552         task->tk_timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
 553         if (task->tk_timeout > max_timeout || task->tk_timeout == 0)
 554                 task->tk_timeout = max_timeout;
 555 }
 556 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_rtt);
 557
 558 static void xprt_reset_majortimeo(struct rpc_rqst *req)
 559 {
 560         const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
 561
 562         req->rq_majortimeo = req->rq_timeout;
 563         if (to->to_exponential)
 564                 req->rq_majortimeo <<= to->to_retries;
 565         else
 566                 req->rq_majortimeo += to->to_increment * to->to_retries;
 567         if (req->rq_majortimeo > to->to_maxval || req->rq_majortimeo == 0)
 568                 req->rq_majortimeo = to->to_maxval;
 569         req->rq_majortimeo += jiffies;
 570 }
 571
 572 /**
 573  * xprt_adjust_timeout - adjust timeout values for next retransmit
 574  * @req: RPC request containing parameters to use for the adjustment
 575  *
 576  */
 577 int xprt_adjust_timeout(struct rpc_rqst *req)
 578 {
 579         struct rpc_xprt *xprt = req->rq_xprt;
 580         const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
 581         int status = 0;
 582
 583         if (time_before(jiffies, req->rq_majortimeo)) {
 584                 if (to->to_exponential)
 585                         req->rq_timeout <<= 1;
 586                 else
 587                         req->rq_timeout += to->to_increment;
 588                 if (to->to_maxval && req->rq_timeout >= to->to_maxval)
 589                         req->rq_timeout = to->to_maxval;
 590                 req->rq_retries++;
 591         } else {
 592                 req->rq_timeout = to->to_initval;
 593                 req->rq_retries = 0;
 594                 xprt_reset_majortimeo(req);
 595                 /* Reset the RTT counters == "slow start" */
 596                 spin_lock_bh(&xprt->transport_lock);
 597                 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
 598                 spin_unlock_bh(&xprt->transport_lock);
 599                 status = -ETIMEDOUT;
 600         }
 601
 602         if (req->rq_timeout == 0) {
 603                 printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
 604                 req->rq_timeout = 5 * HZ;
 605         }
 606         return status;
 607 }
 608
 609 static void xprt_autoclose(struct work_struct *work)
 610 {
 611         struct rpc_xprt *xprt =
 612                 container_of(work, struct rpc_xprt, task_cleanup);
 613
 614         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 615         xprt->ops->close(xprt);
 616         xprt_release_write(xprt, NULL);
 617         wake_up_bit(&xprt->state, XPRT_LOCKED);
 618 }
 619
 620 /**
 621  * xprt_disconnect_done - mark a transport as disconnected
 622  * @xprt: transport to flag for disconnect
 623  *
 624  */
 625 void xprt_disconnect_done(struct rpc_xprt *xprt)
 626 {
 627         dprintk("RPC:       disconnected transport %p\n", xprt);
 628         spin_lock_bh(&xprt->transport_lock);
 629         xprt_clear_connected(xprt);
 630         xprt_wake_pending_tasks(xprt, -EAGAIN);
 631         spin_unlock_bh(&xprt->transport_lock);
 632 }
 633 EXPORT_SYMBOL_GPL(xprt_disconnect_done);
 634
 635 /**
 636  * xprt_force_disconnect - force a transport to disconnect
 637  * @xprt: transport to disconnect
 638  *
 639  */
 640 void xprt_force_disconnect(struct rpc_xprt *xprt)
 641 {
 642         /* Don't race with the test_bit() in xprt_clear_locked() */
 643         spin_lock_bh(&xprt->transport_lock);
 644         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 645         /* Try to schedule an autoclose RPC call */
 646         if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 647                 queue_work(rpciod_workqueue, &xprt->task_cleanup);
 648         xprt_wake_pending_tasks(xprt, -EAGAIN);
 649         spin_unlock_bh(&xprt->transport_lock);
 650 }
 651
 652 /**
 653  * xprt_conditional_disconnect - force a transport to disconnect
 654  * @xprt: transport to disconnect
 655  * @cookie: 'connection cookie'
 656  *
 657  * This attempts to break the connection if and only if 'cookie' matches
 658  * the current transport 'connection cookie'. It ensures that we don't
 659  * try to break the connection more than once when we need to retransmit
 660  * a batch of RPC requests.
 661  *
 662  */
 663 void xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie)
 664 {
 665         /* Don't race with the test_bit() in xprt_clear_locked() */
 666         spin_lock_bh(&xprt->transport_lock);
 667         if (cookie != xprt->connect_cookie)
 668                 goto out;
 669         if (test_bit(XPRT_CLOSING, &xprt->state) || !xprt_connected(xprt))
 670                 goto out;
 671         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 672         /* Try to schedule an autoclose RPC call */
 673         if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 674                 queue_work(rpciod_workqueue, &xprt->task_cleanup);
 675         xprt_wake_pending_tasks(xprt, -EAGAIN);
 676 out:
 677         spin_unlock_bh(&xprt->transport_lock);
 678 }
 679
 680 static void
 681 xprt_init_autodisconnect(unsigned long data)
 682 {
 683         struct rpc_xprt *xprt = (struct rpc_xprt *)data;
 684
 685         spin_lock(&xprt->transport_lock);
 686         if (!list_empty(&xprt->recv))
 687                 goto out_abort;
 688         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 689                 goto out_abort;
 690         spin_unlock(&xprt->transport_lock);
 691         queue_work(rpciod_workqueue, &xprt->task_cleanup);
 692         return;
 693 out_abort:
 694         spin_unlock(&xprt->transport_lock);
 695 }
 696
 697 bool xprt_lock_connect(struct rpc_xprt *xprt,
 698                 struct rpc_task *task,
 699                 void *cookie)
 700 {
 701         bool ret = false;
 702
 703         spin_lock_bh(&xprt->transport_lock);
 704         if (!test_bit(XPRT_LOCKED, &xprt->state))
 705                 goto out;
 706         if (xprt->snd_task != task)
 707                 goto out;
 708         xprt_task_clear_bytes_sent(task);
 709         xprt->snd_task = cookie;
 710         ret = true;
 711 out:
 712         spin_unlock_bh(&xprt->transport_lock);
 713         return ret;
 714 }
 715
 716 void xprt_unlock_connect(struct rpc_xprt *xprt, void *cookie)
 717 {
 718         spin_lock_bh(&xprt->transport_lock);
 719         if (xprt->snd_task != cookie)
 720                 goto out;
 721         if (!test_bit(XPRT_LOCKED, &xprt->state))
 722                 goto out;
 723         xprt->snd_task =NULL;
 724         xprt->ops->release_xprt(xprt, NULL);
 725 out:
 726         spin_unlock_bh(&xprt->transport_lock);
 727         wake_up_bit(&xprt->state, XPRT_LOCKED);
 728 }
 729
 730 /**
 731  * xprt_connect - schedule a transport connect operation
 732  * @task: RPC task that is requesting the connect
 733  *
 734  */
 735 void xprt_connect(struct rpc_task *task)
 736 {
 737         struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
 738
 739         dprintk("RPC: %5u xprt_connect xprt %p %s connected\n", task->tk_pid,
 740                         xprt, (xprt_connected(xprt) ? "is" : "is not"));
 741
 742         if (!xprt_bound(xprt)) {
 743                 task->tk_status = -EAGAIN;
 744                 return;
 745         }
 746         if (!xprt_lock_write(xprt, task))
 747                 return;
 748
 749         if (test_and_clear_bit(XPRT_CLOSE_WAIT, &xprt->state))
 750                 xprt->ops->close(xprt);
 751
 752         if (!xprt_connected(xprt)) {
 753                 task->tk_rqstp->rq_bytes_sent = 0;
 754                 task->tk_timeout = task->tk_rqstp->rq_timeout;
 755                 rpc_sleep_on(&xprt->pending, task, xprt_connect_status);
 756
 757                 if (test_bit(XPRT_CLOSING, &xprt->state))
 758                         return;
 759                 if (xprt_test_and_set_connecting(xprt))
 760                         return;
 761                 xprt->stat.connect_start = jiffies;
 762                 xprt->ops->connect(xprt, task);
 763         }
 764         xprt_release_write(xprt, task);
 765 }
 766
 767 static void xprt_connect_status(struct rpc_task *task)
 768 {
 769         struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
 770
 771         if (task->tk_status == 0) {
 772                 xprt->stat.connect_count++;
 773                 xprt->stat.connect_time += (long)jiffies - xprt->stat.connect_start;
 774                 dprintk("RPC: %5u xprt_connect_status: connection established\n",
 775                                 task->tk_pid);
 776                 return;
 777         }
 778
 779         switch (task->tk_status) {
 780         case -ECONNREFUSED:
 781         case -ECONNRESET:
 782         case -ECONNABORTED:
 783         case -ENETUNREACH:
 784         case -EHOSTUNREACH:
 785         case -EPIPE:
 786         case -EAGAIN:
 787                 dprintk("RPC: %5u xprt_connect_status: retrying\n", task->tk_pid);
 788                 break;
 789         case -ETIMEDOUT:
 790                 dprintk("RPC: %5u xprt_connect_status: connect attempt timed "
 791                                 "out\n", task->tk_pid);
 792                 break;
 793         default:
 794                 dprintk("RPC: %5u xprt_connect_status: error %d connecting to "
 795                                 "server %s\n", task->tk_pid, -task->tk_status,
 796                                 xprt->servername);
 797                 task->tk_status = -EIO;
 798         }
 799 }
 800
 801 /**
 802  * xprt_lookup_rqst - find an RPC request corresponding to an XID
 803  * @xprt: transport on which the original request was transmitted
 804  * @xid: RPC XID of incoming reply
 805  *
 806  */
 807 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
 808 {
 809         struct rpc_rqst *entry;
 810
 811         list_for_each_entry(entry, &xprt->recv, rq_list)
 812                 if (entry->rq_xid == xid) {
 813                         trace_xprt_lookup_rqst(xprt, xid, 0);
 814                         return entry;
 815                 }
 816
 817         dprintk("RPC:       xprt_lookup_rqst did not find xid %08x\n",
 818                         ntohl(xid));
 819         trace_xprt_lookup_rqst(xprt, xid, -ENOENT);
 820         xprt->stat.bad_xids++;
 821         return NULL;
 822 }
 823 EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
 824
 825 static void xprt_update_rtt(struct rpc_task *task)
 826 {
 827         struct rpc_rqst *req = task->tk_rqstp;
 828         struct rpc_rtt *rtt = task->tk_client->cl_rtt;
 829         unsigned int timer = task->tk_msg.rpc_proc->p_timer;
 830         long m = usecs_to_jiffies(ktime_to_us(req->rq_rtt));
 831
 832         if (timer) {
 833                 if (req->rq_ntrans == 1)
 834                         rpc_update_rtt(rtt, timer, m);
 835                 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
 836         }
 837 }
 838
 839 /**
 840  * xprt_complete_rqst - called when reply processing is complete
 841  * @task: RPC request that recently completed
 842  * @copied: actual number of bytes received from the transport
 843  *
 844  * Caller holds transport lock.
 845  */
 846 void xprt_complete_rqst(struct rpc_task *task, int copied)
 847 {
 848         struct rpc_rqst *req = task->tk_rqstp;
 849         struct rpc_xprt *xprt = req->rq_xprt;
 850
 851         dprintk("RPC: %5u xid %08x complete (%d bytes received)\n",
 852                         task->tk_pid, ntohl(req->rq_xid), copied);
 853         trace_xprt_complete_rqst(xprt, req->rq_xid, copied);
 854
 855         xprt->stat.recvs++;
 856         req->rq_rtt = ktime_sub(ktime_get(), req->rq_xtime);
 857         if (xprt->ops->timer != NULL)
 858                 xprt_update_rtt(task);
 859
 860         list_del_init(&req->rq_list);
 861         req->rq_private_buf.len = copied;
 862         /* Ensure all writes are done before we update */
 863         /* req->rq_reply_bytes_recvd */
 864         smp_wmb();
 865         req->rq_reply_bytes_recvd = copied;
 866         rpc_wake_up_queued_task(&xprt->pending, task);
 867 }
 868 EXPORT_SYMBOL_GPL(xprt_complete_rqst);
 869
 870 static void xprt_timer(struct rpc_task *task)
 871 {
 872         struct rpc_rqst *req = task->tk_rqstp;
 873         struct rpc_xprt *xprt = req->rq_xprt;
 874
 875         if (task->tk_status != -ETIMEDOUT)
 876                 return;
 877         dprintk("RPC: %5u xprt_timer\n", task->tk_pid);
 878
 879         spin_lock_bh(&xprt->transport_lock);
 880         if (!req->rq_reply_bytes_recvd) {
 881                 if (xprt->ops->timer)
 882                         xprt->ops->timer(xprt, task);
 883         } else
 884                 task->tk_status = 0;
 885         spin_unlock_bh(&xprt->transport_lock);
 886 }
 887
 888 static inline int xprt_has_timer(struct rpc_xprt *xprt)
 889 {
 890         return xprt->idle_timeout != 0;
 891 }
 892
 893 /**
 894  * xprt_prepare_transmit - reserve the transport before sending a request
 895  * @task: RPC task about to send a request
 896  *
 897  */
 898 bool xprt_prepare_transmit(struct rpc_task *task)
 899 {
 900         struct rpc_rqst *req = task->tk_rqstp;
 901         struct rpc_xprt *xprt = req->rq_xprt;
 902         bool ret = false;
 903
 904         dprintk("RPC: %5u xprt_prepare_transmit\n", task->tk_pid);
 905
 906         spin_lock_bh(&xprt->transport_lock);
 907         if (!req->rq_bytes_sent) {
 908                 if (req->rq_reply_bytes_recvd) {
 909                         task->tk_status = req->rq_reply_bytes_recvd;
 910                         goto out_unlock;
 911                 }
 912                 if ((task->tk_flags & RPC_TASK_NO_RETRANS_TIMEOUT)
 913                     && xprt_connected(xprt)
 914                     && req->rq_connect_cookie == xprt->connect_cookie) {
 915                         xprt->ops->set_retrans_timeout(task);
 916                         rpc_sleep_on(&xprt->pending, task, xprt_timer);
 917                         goto out_unlock;
 918                 }
 919         }
 920         if (!xprt->ops->reserve_xprt(xprt, task)) {
 921                 task->tk_status = -EAGAIN;
 922                 goto out_unlock;
 923         }
 924         ret = true;
 925 out_unlock:
 926         spin_unlock_bh(&xprt->transport_lock);
 927         return ret;
 928 }
 929
 930 void xprt_end_transmit(struct rpc_task *task)
 931 {
 932         xprt_release_write(task->tk_rqstp->rq_xprt, task);
 933 }
 934
 935 /**
 936  * xprt_transmit - send an RPC request on a transport
 937  * @task: controlling RPC task
 938  *
 939  * We have to copy the iovec because sendmsg fiddles with its contents.
 940  */
 941 void xprt_transmit(struct rpc_task *task)
 942 {
 943         struct rpc_rqst *req = task->tk_rqstp;
 944         struct rpc_xprt *xprt = req->rq_xprt;
 945         int status, numreqs;
 946
 947         dprintk("RPC: %5u xprt_transmit(%u)\n", task->tk_pid, req->rq_slen);
 948
 949         if (!req->rq_reply_bytes_recvd) {
 950                 if (list_empty(&req->rq_list) && rpc_reply_expected(task)) {
 951                         /*
 952                          * Add to the list only if we're expecting a reply
 953                          */
 954                         spin_lock_bh(&xprt->transport_lock);
 955                         /* Update the softirq receive buffer */
 956                         memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
 957                                         sizeof(req->rq_private_buf));
 958                         /* Add request to the receive list */
 959                         list_add_tail(&req->rq_list, &xprt->recv);
 960                         spin_unlock_bh(&xprt->transport_lock);
 961                         xprt_reset_majortimeo(req);
 962                         /* Turn off autodisconnect */
 963                         del_singleshot_timer_sync(&xprt->timer);
 964                 }
 965         } else if (!req->rq_bytes_sent)
 966                 return;
 967
 968         req->rq_xtime = ktime_get();
 969         status = xprt->ops->send_request(task);
 970         trace_xprt_transmit(xprt, req->rq_xid, status);
 971         if (status != 0) {
 972                 task->tk_status = status;
 973                 return;
 974         }
 975         xprt_inject_disconnect(xprt);
 976
 977         dprintk("RPC: %5u xmit complete\n", task->tk_pid);
 978         task->tk_flags |= RPC_TASK_SENT;
 979         spin_lock_bh(&xprt->transport_lock);
 980
 981         xprt->ops->set_retrans_timeout(task);
 982
 983         numreqs = atomic_read(&xprt->num_reqs);
 984         if (numreqs > xprt->stat.max_slots)
 985                 xprt->stat.max_slots = numreqs;
 986         xprt->stat.sends++;
 987         xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
 988         xprt->stat.bklog_u += xprt->backlog.qlen;
 989         xprt->stat.sending_u += xprt->sending.qlen;
 990         xprt->stat.pending_u += xprt->pending.qlen;
 991
 992         /* Don't race with disconnect */
 993         if (!xprt_connected(xprt))
 994                 task->tk_status = -ENOTCONN;
 995         else {
 996                 /*
 997                  * Sleep on the pending queue since
 998                  * we're expecting a reply.
 999                  */
1000                 if (!req->rq_reply_bytes_recvd && rpc_reply_expected(task))
1001                         rpc_sleep_on(&xprt->pending, task, xprt_timer);
1002                 req->rq_connect_cookie = xprt->connect_cookie;
1003         }
1004         spin_unlock_bh(&xprt->transport_lock);
1005 }
1006
1007 static void xprt_add_backlog(struct rpc_xprt *xprt, struct rpc_task *task)
1008 {
1009         set_bit(XPRT_CONGESTED, &xprt->state);
1010         rpc_sleep_on(&xprt->backlog, task, NULL);
1011 }
1012
1013 static void xprt_wake_up_backlog(struct rpc_xprt *xprt)
1014 {
1015         if (rpc_wake_up_next(&xprt->backlog) == NULL)
1016                 clear_bit(XPRT_CONGESTED, &xprt->state);
1017 }
1018
1019 static bool xprt_throttle_congested(struct rpc_xprt *xprt, struct rpc_task *task)
1020 {
1021         bool ret = false;
1022
1023         if (!test_bit(XPRT_CONGESTED, &xprt->state))
1024                 goto out;
1025         spin_lock(&xprt->reserve_lock);
1026         if (test_bit(XPRT_CONGESTED, &xprt->state)) {
1027                 rpc_sleep_on(&xprt->backlog, task, NULL);
1028                 ret = true;
1029         }
1030         spin_unlock(&xprt->reserve_lock);
1031 out:
1032         return ret;
1033 }
1034
1035 static struct rpc_rqst *xprt_dynamic_alloc_slot(struct rpc_xprt *xprt, gfp_t gfp_flags)
1036 {
1037         struct rpc_rqst *req = ERR_PTR(-EAGAIN);
1038
1039         if (!atomic_add_unless(&xprt->num_reqs, 1, xprt->max_reqs))
1040                 goto out;
1041         req = kzalloc(sizeof(struct rpc_rqst), gfp_flags);
1042         if (req != NULL)
1043                 goto out;
1044         atomic_dec(&xprt->num_reqs);
1045         req = ERR_PTR(-ENOMEM);
1046 out:
1047         return req;
1048 }
1049
1050 static bool xprt_dynamic_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1051 {
1052         if (atomic_add_unless(&xprt->num_reqs, -1, xprt->min_reqs)) {
1053                 kfree(req);
1054                 return true;
1055         }
1056         return false;
1057 }
1058
1059 void xprt_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1060 {
1061         struct rpc_rqst *req;
1062
1063         spin_lock(&xprt->reserve_lock);
1064         if (!list_empty(&xprt->free)) {
1065                 req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
1066                 list_del(&req->rq_list);
1067                 goto out_init_req;
1068         }
1069         req = xprt_dynamic_alloc_slot(xprt, GFP_NOWAIT|__GFP_NOWARN);
1070         if (!IS_ERR(req))
1071                 goto out_init_req;
1072         switch (PTR_ERR(req)) {
1073         case -ENOMEM:
1074                 dprintk("RPC:       dynamic allocation of request slot "
1075                                 "failed! Retrying\n");
1076                 task->tk_status = -ENOMEM;
1077                 break;
1078         case -EAGAIN:
1079                 xprt_add_backlog(xprt, task);
1080                 dprintk("RPC:       waiting for request slot\n");
1081         default:
1082                 task->tk_status = -EAGAIN;
1083         }
1084         spin_unlock(&xprt->reserve_lock);
1085         return;
1086 out_init_req:
1087         task->tk_status = 0;
1088         task->tk_rqstp = req;
1089         xprt_request_init(task, xprt);
1090         spin_unlock(&xprt->reserve_lock);
1091 }
1092 EXPORT_SYMBOL_GPL(xprt_alloc_slot);
1093
1094 void xprt_lock_and_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1095 {
1096         /* Note: grabbing the xprt_lock_write() ensures that we throttle
1097          * new slot allocation if the transport is congested (i.e. when
1098          * reconnecting a stream transport or when out of socket write
1099          * buffer space).
1100          */
1101         if (xprt_lock_write(xprt, task)) {
1102                 xprt_alloc_slot(xprt, task);
1103                 xprt_release_write(xprt, task);
1104         }
1105 }
1106 EXPORT_SYMBOL_GPL(xprt_lock_and_alloc_slot);
1107
1108 static void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1109 {
1110         spin_lock(&xprt->reserve_lock);
1111         if (!xprt_dynamic_free_slot(xprt, req)) {
1112                 memset(req, 0, sizeof(*req));   /* mark unused */
1113                 list_add(&req->rq_list, &xprt->free);
1114         }
1115         xprt_wake_up_backlog(xprt);
1116         spin_unlock(&xprt->reserve_lock);
1117 }
1118
1119 static void xprt_free_all_slots(struct rpc_xprt *xprt)
1120 {
1121         struct rpc_rqst *req;
1122         while (!list_empty(&xprt->free)) {
1123                 req = list_first_entry(&xprt->free, struct rpc_rqst, rq_list);
1124                 list_del(&req->rq_list);
1125                 kfree(req);
1126         }
1127 }
1128
1129 struct rpc_xprt *xprt_alloc(struct net *net, size_t size,
1130                 unsigned int num_prealloc,
1131                 unsigned int max_alloc)
1132 {
1133         struct rpc_xprt *xprt;
1134         struct rpc_rqst *req;
1135         int i;
1136
1137         xprt = kzalloc(size, GFP_KERNEL);
1138         if (xprt == NULL)
1139                 goto out;
1140
1141         xprt_init(xprt, net);
1142
1143         for (i = 0; i < num_prealloc; i++) {
1144                 req = kzalloc(sizeof(struct rpc_rqst), GFP_KERNEL);
1145                 if (!req)
1146                         goto out_free;
1147                 list_add(&req->rq_list, &xprt->free);
1148         }
1149         if (max_alloc > num_prealloc)
1150                 xprt->max_reqs = max_alloc;
1151         else
1152                 xprt->max_reqs = num_prealloc;
1153         xprt->min_reqs = num_prealloc;
1154         atomic_set(&xprt->num_reqs, num_prealloc);
1155
1156         return xprt;
1157
1158 out_free:
1159         xprt_free(xprt);
1160 out:
1161         return NULL;
1162 }
1163 EXPORT_SYMBOL_GPL(xprt_alloc);
1164
1165 void xprt_free(struct rpc_xprt *xprt)
1166 {
1167         put_net(xprt->xprt_net);
1168         xprt_free_all_slots(xprt);
1169         kfree(xprt);
1170 }
1171 EXPORT_SYMBOL_GPL(xprt_free);
1172
1173 /**
1174  * xprt_reserve - allocate an RPC request slot
1175  * @task: RPC task requesting a slot allocation
1176  *
1177  * If the transport is marked as being congested, or if no more
1178  * slots are available, place the task on the transport's
1179  * backlog queue.
1180  */
1181 void xprt_reserve(struct rpc_task *task)
1182 {
1183         struct rpc_xprt *xprt;
1184
1185         task->tk_status = 0;
1186         if (task->tk_rqstp != NULL)
1187                 return;
1188
1189         task->tk_timeout = 0;
1190         task->tk_status = -EAGAIN;
1191         rcu_read_lock();
1192         xprt = rcu_dereference(task->tk_client->cl_xprt);
1193         if (!xprt_throttle_congested(xprt, task))
1194                 xprt->ops->alloc_slot(xprt, task);
1195         rcu_read_unlock();
1196 }
1197
1198 /**
1199  * xprt_retry_reserve - allocate an RPC request slot
1200  * @task: RPC task requesting a slot allocation
1201  *
1202  * If no more slots are available, place the task on the transport's
1203  * backlog queue.
1204  * Note that the only difference with xprt_reserve is that we now
1205  * ignore the value of the XPRT_CONGESTED flag.
1206  */
1207 void xprt_retry_reserve(struct rpc_task *task)
1208 {
1209         struct rpc_xprt *xprt;
1210
1211         task->tk_status = 0;
1212         if (task->tk_rqstp != NULL)
1213                 return;
1214
1215         task->tk_timeout = 0;
1216         task->tk_status = -EAGAIN;
1217         rcu_read_lock();
1218         xprt = rcu_dereference(task->tk_client->cl_xprt);
1219         xprt->ops->alloc_slot(xprt, task);
1220         rcu_read_unlock();
1221 }
1222
1223 static inline __be32 xprt_alloc_xid(struct rpc_xprt *xprt)
1224 {
1225         return (__force __be32)xprt->xid++;
1226 }
1227
1228 static inline void xprt_init_xid(struct rpc_xprt *xprt)
1229 {
1230         xprt->xid = prandom_u32();
1231 }
1232
1233 static void xprt_request_init(struct rpc_task *task, struct rpc_xprt *xprt)
1234 {
1235         struct rpc_rqst *req = task->tk_rqstp;
1236
1237         INIT_LIST_HEAD(&req->rq_list);
1238         req->rq_timeout = task->tk_client->cl_timeout->to_initval;
1239         req->rq_task    = task;
1240         req->rq_xprt    = xprt;
1241         req->rq_buffer  = NULL;
1242         req->rq_xid     = xprt_alloc_xid(xprt);
1243         req->rq_connect_cookie = xprt->connect_cookie - 1;
1244         req->rq_bytes_sent = 0;
1245         req->rq_snd_buf.len = 0;
1246         req->rq_snd_buf.buflen = 0;
1247         req->rq_rcv_buf.len = 0;
1248         req->rq_rcv_buf.buflen = 0;
1249         req->rq_release_snd_buf = NULL;
1250         xprt_reset_majortimeo(req);
1251         dprintk("RPC: %5u reserved req %p xid %08x\n", task->tk_pid,
1252                         req, ntohl(req->rq_xid));
1253 }
1254
1255 /**
1256  * xprt_release - release an RPC request slot
1257  * @task: task which is finished with the slot
1258  *
1259  */
1260 void xprt_release(struct rpc_task *task)
1261 {
1262         struct rpc_xprt *xprt;
1263         struct rpc_rqst *req = task->tk_rqstp;
1264
1265         if (req == NULL) {
1266                 if (task->tk_client) {
1267                         rcu_read_lock();
1268                         xprt = rcu_dereference(task->tk_client->cl_xprt);
1269                         if (xprt->snd_task == task)
1270                                 xprt_release_write(xprt, task);
1271                         rcu_read_unlock();
1272                 }
1273                 return;
1274         }
1275
1276         xprt = req->rq_xprt;
1277         if (task->tk_ops->rpc_count_stats != NULL)
1278                 task->tk_ops->rpc_count_stats(task, task->tk_calldata);
1279         else if (task->tk_client)
1280                 rpc_count_iostats(task, task->tk_client->cl_metrics);
1281         spin_lock_bh(&xprt->transport_lock);
1282         xprt->ops->release_xprt(xprt, task);
1283         if (xprt->ops->release_request)
1284                 xprt->ops->release_request(task);
1285         if (!list_empty(&req->rq_list))
1286                 list_del(&req->rq_list);
1287         xprt->last_used = jiffies;
1288         if (list_empty(&xprt->recv) && xprt_has_timer(xprt))
1289                 mod_timer(&xprt->timer,
1290                                 xprt->last_used + xprt->idle_timeout);
1291         spin_unlock_bh(&xprt->transport_lock);
1292         if (req->rq_buffer)
1293                 xprt->ops->buf_free(req->rq_buffer);
1294         xprt_inject_disconnect(xprt);
1295         if (req->rq_cred != NULL)
1296                 put_rpccred(req->rq_cred);
1297         task->tk_rqstp = NULL;
1298         if (req->rq_release_snd_buf)
1299                 req->rq_release_snd_buf(req);
1300
1301         dprintk("RPC: %5u release request %p\n", task->tk_pid, req);
1302         if (likely(!bc_prealloc(req)))
1303                 xprt_free_slot(xprt, req);
1304         else
1305                 xprt_free_bc_request(req);
1306 }
1307
1308 static void xprt_init(struct rpc_xprt *xprt, struct net *net)
1309 {
1310         atomic_set(&xprt->count, 1);
1311
1312         spin_lock_init(&xprt->transport_lock);
1313         spin_lock_init(&xprt->reserve_lock);
1314
1315         INIT_LIST_HEAD(&xprt->free);
1316         INIT_LIST_HEAD(&xprt->recv);
1317 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1318         spin_lock_init(&xprt->bc_pa_lock);
1319         INIT_LIST_HEAD(&xprt->bc_pa_list);
1320 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1321
1322         xprt->last_used = jiffies;
1323         xprt->cwnd = RPC_INITCWND;
1324         xprt->bind_index = 0;
1325
1326         rpc_init_wait_queue(&xprt->binding, "xprt_binding");
1327         rpc_init_wait_queue(&xprt->pending, "xprt_pending");
1328         rpc_init_priority_wait_queue(&xprt->sending, "xprt_sending");
1329         rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
1330
1331         xprt_init_xid(xprt);
1332
1333         xprt->xprt_net = get_net(net);
1334 }
1335
1336 /**
1337  * xprt_create_transport - create an RPC transport
1338  * @args: rpc transport creation arguments
1339  *
1340  */
1341 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
1342 {
1343         struct rpc_xprt *xprt;
1344         struct xprt_class *t;
1345
1346         spin_lock(&xprt_list_lock);
1347         list_for_each_entry(t, &xprt_list, list) {
1348                 if (t->ident == args->ident) {
1349                         spin_unlock(&xprt_list_lock);
1350                         goto found;
1351                 }
1352         }
1353         spin_unlock(&xprt_list_lock);
1354         dprintk("RPC: transport (%d) not supported\n", args->ident);
1355         return ERR_PTR(-EIO);
1356
1357 found:
1358         xprt = t->setup(args);
1359         if (IS_ERR(xprt)) {
1360                 dprintk("RPC:       xprt_create_transport: failed, %ld\n",
1361                                 -PTR_ERR(xprt));
1362                 goto out;
1363         }
1364         if (args->flags & XPRT_CREATE_NO_IDLE_TIMEOUT)
1365                 xprt->idle_timeout = 0;
1366         INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
1367         if (xprt_has_timer(xprt))
1368                 setup_timer(&xprt->timer, xprt_init_autodisconnect,
1369                             (unsigned long)xprt);
1370         else
1371                 init_timer(&xprt->timer);
1372
1373         if (strlen(args->servername) > RPC_MAXNETNAMELEN) {
1374                 xprt_destroy(xprt);
1375                 return ERR_PTR(-EINVAL);
1376         }
1377         xprt->servername = kstrdup(args->servername, GFP_KERNEL);
1378         if (xprt->servername == NULL) {
1379                 xprt_destroy(xprt);
1380                 return ERR_PTR(-ENOMEM);
1381         }
1382
1383         rpc_xprt_debugfs_register(xprt);
1384
1385         dprintk("RPC:       created transport %p with %u slots\n", xprt,
1386                         xprt->max_reqs);
1387 out:
1388         return xprt;
1389 }
1390
1391 /**
1392  * xprt_destroy - destroy an RPC transport, killing off all requests.
1393  * @xprt: transport to destroy
1394  *
1395  */
1396 static void xprt_destroy(struct rpc_xprt *xprt)
1397 {
1398         dprintk("RPC:       destroying transport %p\n", xprt);
1399
1400         /* Exclude transport connect/disconnect handlers */
1401         wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_UNINTERRUPTIBLE);
1402
1403         del_timer_sync(&xprt->timer);
1404
1405         rpc_xprt_debugfs_unregister(xprt);
1406         rpc_destroy_wait_queue(&xprt->binding);
1407         rpc_destroy_wait_queue(&xprt->pending);
1408         rpc_destroy_wait_queue(&xprt->sending);
1409         rpc_destroy_wait_queue(&xprt->backlog);
1410         cancel_work_sync(&xprt->task_cleanup);
1411         kfree(xprt->servername);
1412         /*
1413          * Tear down transport state and free the rpc_xprt
1414          */
1415         xprt->ops->destroy(xprt);
1416 }
1417
1418 /**
1419  * xprt_put - release a reference to an RPC transport.
1420  * @xprt: pointer to the transport
1421  *
1422  */
1423 void xprt_put(struct rpc_xprt *xprt)
1424 {
1425         if (atomic_dec_and_test(&xprt->count))
1426                 xprt_destroy(xprt);
1427 }