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 <okir@monad.swb.de>
  36  *
  37  *  Transport switch API copyright (C) 2005, Chuck Lever <cel@netapp.com>
  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 #include <linux/rcupdate.h>
  52
  53 #include <trace/events/sunrpc.h>
  54
  55 #include "sunrpc.h"
  56
  57 /*
  58  * Local variables
  59  */
  60
  61 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
  62 # define RPCDBG_FACILITY        RPCDBG_XPRT
  63 #endif
  64
  65 /*
  66  * Local functions
  67  */
  68 static void      xprt_init(struct rpc_xprt *xprt, struct net *net);
  69 static void     xprt_request_init(struct rpc_task *, struct rpc_xprt *);
  70 static void     xprt_connect_status(struct rpc_task *task);
  71 static int      __xprt_get_cong(struct rpc_xprt *, struct rpc_task *);
  72 static void     __xprt_put_cong(struct rpc_xprt *, struct rpc_rqst *);
  73 static void      xprt_destroy(struct rpc_xprt *xprt);
  74
  75 static DEFINE_SPINLOCK(xprt_list_lock);
  76 static LIST_HEAD(xprt_list);
  77
  78 /**
  79  * xprt_register_transport - register a transport implementation
  80  * @transport: transport to register
  81  *
  82  * If a transport implementation is loaded as a kernel module, it can
  83  * call this interface to make itself known to the RPC client.
  84  *
  85  * Returns:
  86  * 0:           transport successfully registered
  87  * -EEXIST:     transport already registered
  88  * -EINVAL:     transport module being unloaded
  89  */
  90 int xprt_register_transport(struct xprt_class *transport)
  91 {
  92         struct xprt_class *t;
  93         int result;
  94
  95         result = -EEXIST;
  96         spin_lock(&xprt_list_lock);
  97         list_for_each_entry(t, &xprt_list, list) {
  98                 /* don't register the same transport class twice */
  99                 if (t->ident == transport->ident)
 100                         goto out;
 101         }
 102
 103         list_add_tail(&transport->list, &xprt_list);
 104         printk(KERN_INFO "RPC: Registered %s transport module.\n",
 105                transport->name);
 106         result = 0;
 107
 108 out:
 109         spin_unlock(&xprt_list_lock);
 110         return result;
 111 }
 112 EXPORT_SYMBOL_GPL(xprt_register_transport);
 113
 114 /**
 115  * xprt_unregister_transport - unregister a transport implementation
 116  * @transport: transport to unregister
 117  *
 118  * Returns:
 119  * 0:           transport successfully unregistered
 120  * -ENOENT:     transport never registered
 121  */
 122 int xprt_unregister_transport(struct xprt_class *transport)
 123 {
 124         struct xprt_class *t;
 125         int result;
 126
 127         result = 0;
 128         spin_lock(&xprt_list_lock);
 129         list_for_each_entry(t, &xprt_list, list) {
 130                 if (t == transport) {
 131                         printk(KERN_INFO
 132                                 "RPC: Unregistered %s transport module.\n",
 133                                 transport->name);
 134                         list_del_init(&transport->list);
 135                         goto out;
 136                 }
 137         }
 138         result = -ENOENT;
 139
 140 out:
 141         spin_unlock(&xprt_list_lock);
 142         return result;
 143 }
 144 EXPORT_SYMBOL_GPL(xprt_unregister_transport);
 145
 146 static void
 147 xprt_class_release(const struct xprt_class *t)
 148 {
 149         module_put(t->owner);
 150 }
 151
 152 static const struct xprt_class *
 153 xprt_class_find_by_netid_locked(const char *netid)
 154 {
 155         const struct xprt_class *t;
 156         unsigned int i;
 157
 158         list_for_each_entry(t, &xprt_list, list) {
 159                 for (i = 0; t->netid[i][0] != '\0'; i++) {
 160                         if (strcmp(t->netid[i], netid) != 0)
 161                                 continue;
 162                         if (!try_module_get(t->owner))
 163                                 continue;
 164                         return t;
 165                 }
 166         }
 167         return NULL;
 168 }
 169
 170 static const struct xprt_class *
 171 xprt_class_find_by_netid(const char *netid)
 172 {
 173         const struct xprt_class *t;
 174
 175         spin_lock(&xprt_list_lock);
 176         t = xprt_class_find_by_netid_locked(netid);
 177         if (!t) {
 178                 spin_unlock(&xprt_list_lock);
 179                 request_module("rpc%s", netid);
 180                 spin_lock(&xprt_list_lock);
 181                 t = xprt_class_find_by_netid_locked(netid);
 182         }
 183         spin_unlock(&xprt_list_lock);
 184         return t;
 185 }
 186
 187 /**
 188  * xprt_load_transport - load a transport implementation
 189  * @netid: transport to load
 190  *
 191  * Returns:
 192  * 0:           transport successfully loaded
 193  * -ENOENT:     transport module not available
 194  */
 195 int xprt_load_transport(const char *netid)
 196 {
 197         const struct xprt_class *t;
 198
 199         t = xprt_class_find_by_netid(netid);
 200         if (!t)
 201                 return -ENOENT;
 202         xprt_class_release(t);
 203         return 0;
 204 }
 205 EXPORT_SYMBOL_GPL(xprt_load_transport);
 206
 207 /**
 208  * xprt_reserve_xprt - serialize write access to transports
 209  * @task: task that is requesting access to the transport
 210  * @xprt: pointer to the target transport
 211  *
 212  * This prevents mixing the payload of separate requests, and prevents
 213  * transport connects from colliding with writes.  No congestion control
 214  * is provided.
 215  */
 216 int xprt_reserve_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 217 {
 218         struct rpc_rqst *req = task->tk_rqstp;
 219         int priority;
 220
 221         if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
 222                 if (task == xprt->snd_task)
 223                         return 1;
 224                 goto out_sleep;
 225         }
 226         xprt->snd_task = task;
 227         if (req != NULL)
 228                 req->rq_ntrans++;
 229
 230         return 1;
 231
 232 out_sleep:
 233         dprintk("RPC: %5u failed to lock transport %p\n",
 234                         task->tk_pid, xprt);
 235         task->tk_timeout = 0;
 236         task->tk_status = -EAGAIN;
 237         if (req == NULL)
 238                 priority = RPC_PRIORITY_LOW;
 239         else if (!req->rq_ntrans)
 240                 priority = RPC_PRIORITY_NORMAL;
 241         else
 242                 priority = RPC_PRIORITY_HIGH;
 243         rpc_sleep_on_priority(&xprt->sending, task, NULL, priority);
 244         return 0;
 245 }
 246 EXPORT_SYMBOL_GPL(xprt_reserve_xprt);
 247
 248 static void xprt_clear_locked(struct rpc_xprt *xprt)
 249 {
 250         xprt->snd_task = NULL;
 251         if (!test_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
 252                 smp_mb__before_atomic();
 253                 clear_bit(XPRT_LOCKED, &xprt->state);
 254                 smp_mb__after_atomic();
 255         } else
 256                 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 257 }
 258
 259 /*
 260  * xprt_reserve_xprt_cong - serialize write access to transports
 261  * @task: task that is requesting access to the transport
 262  *
 263  * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
 264  * integrated into the decision of whether a request is allowed to be
 265  * woken up and given access to the transport.
 266  */
 267 int xprt_reserve_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 268 {
 269         struct rpc_rqst *req = task->tk_rqstp;
 270         int priority;
 271
 272         if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
 273                 if (task == xprt->snd_task)
 274                         return 1;
 275                 goto out_sleep;
 276         }
 277         if (req == NULL) {
 278                 xprt->snd_task = task;
 279                 return 1;
 280         }
 281         if (__xprt_get_cong(xprt, task)) {
 282                 xprt->snd_task = task;
 283                 req->rq_ntrans++;
 284                 return 1;
 285         }
 286         xprt_clear_locked(xprt);
 287 out_sleep:
 288         if (req)
 289                 __xprt_put_cong(xprt, req);
 290         dprintk("RPC: %5u failed to lock transport %p\n", task->tk_pid, xprt);
 291         task->tk_timeout = 0;
 292         task->tk_status = -EAGAIN;
 293         if (req == NULL)
 294                 priority = RPC_PRIORITY_LOW;
 295         else if (!req->rq_ntrans)
 296                 priority = RPC_PRIORITY_NORMAL;
 297         else
 298                 priority = RPC_PRIORITY_HIGH;
 299         rpc_sleep_on_priority(&xprt->sending, task, NULL, priority);
 300         return 0;
 301 }
 302 EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong);
 303
 304 static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
 305 {
 306         int retval;
 307
 308         spin_lock_bh(&xprt->transport_lock);
 309         retval = xprt->ops->reserve_xprt(xprt, task);
 310         spin_unlock_bh(&xprt->transport_lock);
 311         return retval;
 312 }
 313
 314 static bool __xprt_lock_write_func(struct rpc_task *task, void *data)
 315 {
 316         struct rpc_xprt *xprt = data;
 317         struct rpc_rqst *req;
 318
 319         req = task->tk_rqstp;
 320         xprt->snd_task = task;
 321         if (req)
 322                 req->rq_ntrans++;
 323         return true;
 324 }
 325
 326 static void __xprt_lock_write_next(struct rpc_xprt *xprt)
 327 {
 328         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 329                 return;
 330
 331         if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
 332                                 __xprt_lock_write_func, xprt))
 333                 return;
 334         xprt_clear_locked(xprt);
 335 }
 336
 337 static bool __xprt_lock_write_cong_func(struct rpc_task *task, void *data)
 338 {
 339         struct rpc_xprt *xprt = data;
 340         struct rpc_rqst *req;
 341
 342         req = task->tk_rqstp;
 343         if (req == NULL) {
 344                 xprt->snd_task = task;
 345                 return true;
 346         }
 347         if (__xprt_get_cong(xprt, task)) {
 348                 xprt->snd_task = task;
 349                 req->rq_ntrans++;
 350                 return true;
 351         }
 352         return false;
 353 }
 354
 355 static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt)
 356 {
 357         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 358                 return;
 359         if (RPCXPRT_CONGESTED(xprt))
 360                 goto out_unlock;
 361         if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
 362                                 __xprt_lock_write_cong_func, xprt))
 363                 return;
 364 out_unlock:
 365         xprt_clear_locked(xprt);
 366 }
 367
 368 static void xprt_task_clear_bytes_sent(struct rpc_task *task)
 369 {
 370         if (task != NULL) {
 371                 struct rpc_rqst *req = task->tk_rqstp;
 372                 if (req != NULL)
 373                         req->rq_bytes_sent = 0;
 374         }
 375 }
 376
 377 /**
 378  * xprt_release_xprt - allow other requests to use a transport
 379  * @xprt: transport with other tasks potentially waiting
 380  * @task: task that is releasing access to the transport
 381  *
 382  * Note that "task" can be NULL.  No congestion control is provided.
 383  */
 384 void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 385 {
 386         if (xprt->snd_task == task) {
 387                 xprt_task_clear_bytes_sent(task);
 388                 xprt_clear_locked(xprt);
 389                 __xprt_lock_write_next(xprt);
 390         }
 391 }
 392 EXPORT_SYMBOL_GPL(xprt_release_xprt);
 393
 394 /**
 395  * xprt_release_xprt_cong - allow other requests to use a transport
 396  * @xprt: transport with other tasks potentially waiting
 397  * @task: task that is releasing access to the transport
 398  *
 399  * Note that "task" can be NULL.  Another task is awoken to use the
 400  * transport if the transport's congestion window allows it.
 401  */
 402 void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 403 {
 404         if (xprt->snd_task == task) {
 405                 xprt_task_clear_bytes_sent(task);
 406                 xprt_clear_locked(xprt);
 407                 __xprt_lock_write_next_cong(xprt);
 408         }
 409 }
 410 EXPORT_SYMBOL_GPL(xprt_release_xprt_cong);
 411
 412 static inline void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
 413 {
 414         spin_lock_bh(&xprt->transport_lock);
 415         xprt->ops->release_xprt(xprt, task);
 416         spin_unlock_bh(&xprt->transport_lock);
 417 }
 418
 419 /*
 420  * Van Jacobson congestion avoidance. Check if the congestion window
 421  * overflowed. Put the task to sleep if this is the case.
 422  */
 423 static int
 424 __xprt_get_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 425 {
 426         struct rpc_rqst *req = task->tk_rqstp;
 427
 428         if (req->rq_cong)
 429                 return 1;
 430         dprintk("RPC: %5u xprt_cwnd_limited cong = %lu cwnd = %lu\n",
 431                         task->tk_pid, xprt->cong, xprt->cwnd);
 432         if (RPCXPRT_CONGESTED(xprt))
 433                 return 0;
 434         req->rq_cong = 1;
 435         xprt->cong += RPC_CWNDSCALE;
 436         return 1;
 437 }
 438
 439 /*
 440  * Adjust the congestion window, and wake up the next task
 441  * that has been sleeping due to congestion
 442  */
 443 static void
 444 __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
 445 {
 446         if (!req->rq_cong)
 447                 return;
 448         req->rq_cong = 0;
 449         xprt->cong -= RPC_CWNDSCALE;
 450         __xprt_lock_write_next_cong(xprt);
 451 }
 452
 453 /**
 454  * xprt_release_rqst_cong - housekeeping when request is complete
 455  * @task: RPC request that recently completed
 456  *
 457  * Useful for transports that require congestion control.
 458  */
 459 void xprt_release_rqst_cong(struct rpc_task *task)
 460 {
 461         struct rpc_rqst *req = task->tk_rqstp;
 462
 463         __xprt_put_cong(req->rq_xprt, req);
 464 }
 465 EXPORT_SYMBOL_GPL(xprt_release_rqst_cong);
 466
 467 /**
 468  * xprt_adjust_cwnd - adjust transport congestion window
 469  * @xprt: pointer to xprt
 470  * @task: recently completed RPC request used to adjust window
 471  * @result: result code of completed RPC request
 472  *
 473  * The transport code maintains an estimate on the maximum number of out-
 474  * standing RPC requests, using a smoothed version of the congestion
 475  * avoidance implemented in 44BSD. This is basically the Van Jacobson
 476  * congestion algorithm: If a retransmit occurs, the congestion window is
 477  * halved; otherwise, it is incremented by 1/cwnd when
 478  *
 479  *      -       a reply is received and
 480  *      -       a full number of requests are outstanding and
 481  *      -       the congestion window hasn't been updated recently.
 482  */
 483 void xprt_adjust_cwnd(struct rpc_xprt *xprt, struct rpc_task *task, int result)
 484 {
 485         struct rpc_rqst *req = task->tk_rqstp;
 486         unsigned long cwnd = xprt->cwnd;
 487
 488         if (result >= 0 && cwnd <= xprt->cong) {
 489                 /* The (cwnd >> 1) term makes sure
 490                  * the result gets rounded properly. */
 491                 cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
 492                 if (cwnd > RPC_MAXCWND(xprt))
 493                         cwnd = RPC_MAXCWND(xprt);
 494                 __xprt_lock_write_next_cong(xprt);
 495         } else if (result == -ETIMEDOUT) {
 496                 cwnd >>= 1;
 497                 if (cwnd < RPC_CWNDSCALE)
 498                         cwnd = RPC_CWNDSCALE;
 499         }
 500         dprintk("RPC:       cong %ld, cwnd was %ld, now %ld\n",
 501                         xprt->cong, xprt->cwnd, cwnd);
 502         xprt->cwnd = cwnd;
 503         __xprt_put_cong(xprt, req);
 504 }
 505 EXPORT_SYMBOL_GPL(xprt_adjust_cwnd);
 506
 507 /**
 508  * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
 509  * @xprt: transport with waiting tasks
 510  * @status: result code to plant in each task before waking it
 511  *
 512  */
 513 void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
 514 {
 515         if (status < 0)
 516                 rpc_wake_up_status(&xprt->pending, status);
 517         else
 518                 rpc_wake_up(&xprt->pending);
 519 }
 520 EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks);
 521
 522 /**
 523  * xprt_wait_for_buffer_space - wait for transport output buffer to clear
 524  * @task: task to be put to sleep
 525  * @action: function pointer to be executed after wait
 526  *
 527  * Note that we only set the timer for the case of RPC_IS_SOFT(), since
 528  * we don't in general want to force a socket disconnection due to
 529  * an incomplete RPC call transmission.
 530  */
 531 void xprt_wait_for_buffer_space(struct rpc_task *task, rpc_action action)
 532 {
 533         struct rpc_rqst *req = task->tk_rqstp;
 534         struct rpc_xprt *xprt = req->rq_xprt;
 535
 536         task->tk_timeout = RPC_IS_SOFT(task) ? req->rq_timeout : 0;
 537         rpc_sleep_on(&xprt->pending, task, action);
 538 }
 539 EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space);
 540
 541 /**
 542  * xprt_write_space - wake the task waiting for transport output buffer space
 543  * @xprt: transport with waiting tasks
 544  *
 545  * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
 546  */
 547 void xprt_write_space(struct rpc_xprt *xprt)
 548 {
 549         spin_lock_bh(&xprt->transport_lock);
 550         if (xprt->snd_task) {
 551                 dprintk("RPC:       write space: waking waiting task on "
 552                                 "xprt %p\n", xprt);
 553                 rpc_wake_up_queued_task(&xprt->pending, xprt->snd_task);
 554         }
 555         spin_unlock_bh(&xprt->transport_lock);
 556 }
 557 EXPORT_SYMBOL_GPL(xprt_write_space);
 558
 559 /**
 560  * xprt_set_retrans_timeout_def - set a request's retransmit timeout
 561  * @task: task whose timeout is to be set
 562  *
 563  * Set a request's retransmit timeout based on the transport's
 564  * default timeout parameters.  Used by transports that don't adjust
 565  * the retransmit timeout based on round-trip time estimation.
 566  */
 567 void xprt_set_retrans_timeout_def(struct rpc_task *task)
 568 {
 569         task->tk_timeout = task->tk_rqstp->rq_timeout;
 570 }
 571 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_def);
 572
 573 /**
 574  * xprt_set_retrans_timeout_rtt - set a request's retransmit timeout
 575  * @task: task whose timeout is to be set
 576  *
 577  * Set a request's retransmit timeout using the RTT estimator.
 578  */
 579 void xprt_set_retrans_timeout_rtt(struct rpc_task *task)
 580 {
 581         int timer = task->tk_msg.rpc_proc->p_timer;
 582         struct rpc_clnt *clnt = task->tk_client;
 583         struct rpc_rtt *rtt = clnt->cl_rtt;
 584         struct rpc_rqst *req = task->tk_rqstp;
 585         unsigned long max_timeout = clnt->cl_timeout->to_maxval;
 586
 587         task->tk_timeout = rpc_calc_rto(rtt, timer);
 588         task->tk_timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
 589         if (task->tk_timeout > max_timeout || task->tk_timeout == 0)
 590                 task->tk_timeout = max_timeout;
 591 }
 592 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_rtt);
 593
 594 static void xprt_reset_majortimeo(struct rpc_rqst *req)
 595 {
 596         const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
 597
 598         req->rq_majortimeo = req->rq_timeout;
 599         if (to->to_exponential)
 600                 req->rq_majortimeo <<= to->to_retries;
 601         else
 602                 req->rq_majortimeo += to->to_increment * to->to_retries;
 603         if (req->rq_majortimeo > to->to_maxval || req->rq_majortimeo == 0)
 604                 req->rq_majortimeo = to->to_maxval;
 605         req->rq_majortimeo += jiffies;
 606 }
 607
 608 /**
 609  * xprt_adjust_timeout - adjust timeout values for next retransmit
 610  * @req: RPC request containing parameters to use for the adjustment
 611  *
 612  */
 613 int xprt_adjust_timeout(struct rpc_rqst *req)
 614 {
 615         struct rpc_xprt *xprt = req->rq_xprt;
 616         const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
 617         int status = 0;
 618
 619         if (time_before(jiffies, req->rq_majortimeo)) {
 620                 if (to->to_exponential)
 621                         req->rq_timeout <<= 1;
 622                 else
 623                         req->rq_timeout += to->to_increment;
 624                 if (to->to_maxval && req->rq_timeout >= to->to_maxval)
 625                         req->rq_timeout = to->to_maxval;
 626                 req->rq_retries++;
 627         } else {
 628                 req->rq_timeout = to->to_initval;
 629                 req->rq_retries = 0;
 630                 xprt_reset_majortimeo(req);
 631                 /* Reset the RTT counters == "slow start" */
 632                 spin_lock_bh(&xprt->transport_lock);
 633                 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
 634                 spin_unlock_bh(&xprt->transport_lock);
 635                 status = -ETIMEDOUT;
 636         }
 637
 638         if (req->rq_timeout == 0) {
 639                 printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
 640                 req->rq_timeout = 5 * HZ;
 641         }
 642         return status;
 643 }
 644
 645 static void xprt_autoclose(struct work_struct *work)
 646 {
 647         struct rpc_xprt *xprt =
 648                 container_of(work, struct rpc_xprt, task_cleanup);
 649
 650         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 651         xprt->ops->close(xprt);
 652         xprt_release_write(xprt, NULL);
 653         wake_up_bit(&xprt->state, XPRT_LOCKED);
 654 }
 655
 656 /**
 657  * xprt_disconnect_done - mark a transport as disconnected
 658  * @xprt: transport to flag for disconnect
 659  *
 660  */
 661 void xprt_disconnect_done(struct rpc_xprt *xprt)
 662 {
 663         dprintk("RPC:       disconnected transport %p\n", xprt);
 664         spin_lock_bh(&xprt->transport_lock);
 665         xprt_clear_connected(xprt);
 666         xprt_wake_pending_tasks(xprt, -EAGAIN);
 667         spin_unlock_bh(&xprt->transport_lock);
 668 }
 669 EXPORT_SYMBOL_GPL(xprt_disconnect_done);
 670
 671 /**
 672  * xprt_force_disconnect - force a transport to disconnect
 673  * @xprt: transport to disconnect
 674  *
 675  */
 676 void xprt_force_disconnect(struct rpc_xprt *xprt)
 677 {
 678         /* Don't race with the test_bit() in xprt_clear_locked() */
 679         spin_lock_bh(&xprt->transport_lock);
 680         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 681         /* Try to schedule an autoclose RPC call */
 682         if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 683                 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 684         xprt_wake_pending_tasks(xprt, -EAGAIN);
 685         spin_unlock_bh(&xprt->transport_lock);
 686 }
 687 EXPORT_SYMBOL_GPL(xprt_force_disconnect);
 688
 689 /**
 690  * xprt_conditional_disconnect - force a transport to disconnect
 691  * @xprt: transport to disconnect
 692  * @cookie: 'connection cookie'
 693  *
 694  * This attempts to break the connection if and only if 'cookie' matches
 695  * the current transport 'connection cookie'. It ensures that we don't
 696  * try to break the connection more than once when we need to retransmit
 697  * a batch of RPC requests.
 698  *
 699  */
 700 void xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie)
 701 {
 702         /* Don't race with the test_bit() in xprt_clear_locked() */
 703         spin_lock_bh(&xprt->transport_lock);
 704         if (cookie != xprt->connect_cookie)
 705                 goto out;
 706         if (test_bit(XPRT_CLOSING, &xprt->state))
 707                 goto out;
 708         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 709         /* Try to schedule an autoclose RPC call */
 710         if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 711                 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 712         xprt_wake_pending_tasks(xprt, -EAGAIN);
 713 out:
 714         spin_unlock_bh(&xprt->transport_lock);
 715 }
 716
 717 static bool
 718 xprt_has_timer(const struct rpc_xprt *xprt)
 719 {
 720         return xprt->idle_timeout != 0;
 721 }
 722
 723 static void
 724 xprt_schedule_autodisconnect(struct rpc_xprt *xprt)
 725         __must_hold(&xprt->transport_lock)
 726 {
 727         if (list_empty(&xprt->recv) && xprt_has_timer(xprt))
 728                 mod_timer(&xprt->timer, xprt->last_used + xprt->idle_timeout);
 729 }
 730
 731 static void
 732 xprt_init_autodisconnect(unsigned long data)
 733 {
 734         struct rpc_xprt *xprt = (struct rpc_xprt *)data;
 735
 736         spin_lock(&xprt->transport_lock);
 737         if (!list_empty(&xprt->recv))
 738                 goto out_abort;
 739         /* Reset xprt->last_used to avoid connect/autodisconnect cycling */
 740         xprt->last_used = jiffies;
 741         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 742                 goto out_abort;
 743         spin_unlock(&xprt->transport_lock);
 744         queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 745         return;
 746 out_abort:
 747         spin_unlock(&xprt->transport_lock);
 748 }
 749
 750 bool xprt_lock_connect(struct rpc_xprt *xprt,
 751                 struct rpc_task *task,
 752                 void *cookie)
 753 {
 754         bool ret = false;
 755
 756         spin_lock_bh(&xprt->transport_lock);
 757         if (!test_bit(XPRT_LOCKED, &xprt->state))
 758                 goto out;
 759         if (xprt->snd_task != task)
 760                 goto out;
 761         xprt_task_clear_bytes_sent(task);
 762         xprt->snd_task = cookie;
 763         ret = true;
 764 out:
 765         spin_unlock_bh(&xprt->transport_lock);
 766         return ret;
 767 }
 768
 769 void xprt_unlock_connect(struct rpc_xprt *xprt, void *cookie)
 770 {
 771         spin_lock_bh(&xprt->transport_lock);
 772         if (xprt->snd_task != cookie)
 773                 goto out;
 774         if (!test_bit(XPRT_LOCKED, &xprt->state))
 775                 goto out;
 776         xprt->snd_task =NULL;
 777         xprt->ops->release_xprt(xprt, NULL);
 778         xprt_schedule_autodisconnect(xprt);
 779 out:
 780         spin_unlock_bh(&xprt->transport_lock);
 781         wake_up_bit(&xprt->state, XPRT_LOCKED);
 782 }
 783
 784 /**
 785  * xprt_connect - schedule a transport connect operation
 786  * @task: RPC task that is requesting the connect
 787  *
 788  */
 789 void xprt_connect(struct rpc_task *task)
 790 {
 791         struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
 792
 793         dprintk("RPC: %5u xprt_connect xprt %p %s connected\n", task->tk_pid,
 794                         xprt, (xprt_connected(xprt) ? "is" : "is not"));
 795
 796         if (!xprt_bound(xprt)) {
 797                 task->tk_status = -EAGAIN;
 798                 return;
 799         }
 800         if (!xprt_lock_write(xprt, task))
 801                 return;
 802
 803         if (test_and_clear_bit(XPRT_CLOSE_WAIT, &xprt->state))
 804                 xprt->ops->close(xprt);
 805
 806         if (!xprt_connected(xprt)) {
 807                 task->tk_rqstp->rq_bytes_sent = 0;
 808                 task->tk_timeout = task->tk_rqstp->rq_timeout;
 809                 task->tk_rqstp->rq_connect_cookie = xprt->connect_cookie;
 810                 rpc_sleep_on(&xprt->pending, task, xprt_connect_status);
 811
 812                 if (test_bit(XPRT_CLOSING, &xprt->state))
 813                         return;
 814                 if (xprt_test_and_set_connecting(xprt))
 815                         return;
 816                 /* Race breaker */
 817                 if (!xprt_connected(xprt)) {
 818                         xprt->stat.connect_start = jiffies;
 819                         xprt->ops->connect(xprt, task);
 820                 } else {
 821                         xprt_clear_connecting(xprt);
 822                         task->tk_status = 0;
 823                         rpc_wake_up_queued_task(&xprt->pending, task);
 824                 }
 825         }
 826         xprt_release_write(xprt, task);
 827 }
 828
 829 static void xprt_connect_status(struct rpc_task *task)
 830 {
 831         switch (task->tk_status) {
 832         case 0:
 833                 dprintk("RPC: %5u xprt_connect_status: connection established\n",
 834                                 task->tk_pid);
 835                 break;
 836         case -ECONNREFUSED:
 837         case -ECONNRESET:
 838         case -ECONNABORTED:
 839         case -ENETUNREACH:
 840         case -EHOSTUNREACH:
 841         case -EPIPE:
 842         case -EAGAIN:
 843                 dprintk("RPC: %5u xprt_connect_status: retrying\n", task->tk_pid);
 844                 break;
 845         case -ETIMEDOUT:
 846                 dprintk("RPC: %5u xprt_connect_status: connect attempt timed "
 847                                 "out\n", task->tk_pid);
 848                 break;
 849         default:
 850                 dprintk("RPC: %5u xprt_connect_status: error %d connecting to "
 851                                 "server %s\n", task->tk_pid, -task->tk_status,
 852                                 task->tk_rqstp->rq_xprt->servername);
 853                 task->tk_status = -EIO;
 854         }
 855 }
 856
 857 /**
 858  * xprt_lookup_rqst - find an RPC request corresponding to an XID
 859  * @xprt: transport on which the original request was transmitted
 860  * @xid: RPC XID of incoming reply
 861  *
 862  */
 863 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
 864 {
 865         struct rpc_rqst *entry;
 866
 867         list_for_each_entry(entry, &xprt->recv, rq_list)
 868                 if (entry->rq_xid == xid) {
 869                         trace_xprt_lookup_rqst(xprt, xid, 0);
 870                         return entry;
 871                 }
 872
 873         dprintk("RPC:       xprt_lookup_rqst did not find xid %08x\n",
 874                         ntohl(xid));
 875         trace_xprt_lookup_rqst(xprt, xid, -ENOENT);
 876         xprt->stat.bad_xids++;
 877         return NULL;
 878 }
 879 EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
 880
 881 /**
 882  * xprt_pin_rqst - Pin a request on the transport receive list
 883  * @req: Request to pin
 884  *
 885  * Caller must ensure this is atomic with the call to xprt_lookup_rqst()
 886  * so should be holding the xprt transport lock.
 887  */
 888 void xprt_pin_rqst(struct rpc_rqst *req)
 889 {
 890         set_bit(RPC_TASK_MSG_RECV, &req->rq_task->tk_runstate);
 891 }
 892 EXPORT_SYMBOL_GPL(xprt_pin_rqst);
 893
 894 /**
 895  * xprt_unpin_rqst - Unpin a request on the transport receive list
 896  * @req: Request to pin
 897  *
 898  * Caller should be holding the xprt transport lock.
 899  */
 900 void xprt_unpin_rqst(struct rpc_rqst *req)
 901 {
 902         struct rpc_task *task = req->rq_task;
 903
 904         clear_bit(RPC_TASK_MSG_RECV, &task->tk_runstate);
 905         if (test_bit(RPC_TASK_MSG_RECV_WAIT, &task->tk_runstate))
 906                 wake_up_bit(&task->tk_runstate, RPC_TASK_MSG_RECV);
 907 }
 908 EXPORT_SYMBOL_GPL(xprt_unpin_rqst);
 909
 910 static void xprt_wait_on_pinned_rqst(struct rpc_rqst *req)
 911 __must_hold(&req->rq_xprt->recv_lock)
 912 {
 913         struct rpc_task *task = req->rq_task;
 914
 915         if (task && test_bit(RPC_TASK_MSG_RECV, &task->tk_runstate)) {
 916                 spin_unlock(&req->rq_xprt->recv_lock);
 917                 set_bit(RPC_TASK_MSG_RECV_WAIT, &task->tk_runstate);
 918                 wait_on_bit(&task->tk_runstate, RPC_TASK_MSG_RECV,
 919                                 TASK_UNINTERRUPTIBLE);
 920                 clear_bit(RPC_TASK_MSG_RECV_WAIT, &task->tk_runstate);
 921                 spin_lock(&req->rq_xprt->recv_lock);
 922         }
 923 }
 924
 925 static void xprt_update_rtt(struct rpc_task *task)
 926 {
 927         struct rpc_rqst *req = task->tk_rqstp;
 928         struct rpc_rtt *rtt = task->tk_client->cl_rtt;
 929         unsigned int timer = task->tk_msg.rpc_proc->p_timer;
 930         long m = usecs_to_jiffies(ktime_to_us(req->rq_rtt));
 931
 932         if (timer) {
 933                 if (req->rq_ntrans == 1)
 934                         rpc_update_rtt(rtt, timer, m);
 935                 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
 936         }
 937 }
 938
 939 /**
 940  * xprt_complete_rqst - called when reply processing is complete
 941  * @task: RPC request that recently completed
 942  * @copied: actual number of bytes received from the transport
 943  *
 944  * Caller holds transport lock.
 945  */
 946 void xprt_complete_rqst(struct rpc_task *task, int copied)
 947 {
 948         struct rpc_rqst *req = task->tk_rqstp;
 949         struct rpc_xprt *xprt = req->rq_xprt;
 950
 951         dprintk("RPC: %5u xid %08x complete (%d bytes received)\n",
 952                         task->tk_pid, ntohl(req->rq_xid), copied);
 953         trace_xprt_complete_rqst(xprt, req->rq_xid, copied);
 954
 955         xprt->stat.recvs++;
 956         req->rq_rtt = ktime_sub(ktime_get(), req->rq_xtime);
 957         if (xprt->ops->timer != NULL)
 958                 xprt_update_rtt(task);
 959
 960         list_del_init(&req->rq_list);
 961         req->rq_private_buf.len = copied;
 962         /* Ensure all writes are done before we update */
 963         /* req->rq_reply_bytes_recvd */
 964         smp_wmb();
 965         req->rq_reply_bytes_recvd = copied;
 966         rpc_wake_up_queued_task(&xprt->pending, task);
 967 }
 968 EXPORT_SYMBOL_GPL(xprt_complete_rqst);
 969
 970 static void xprt_timer(struct rpc_task *task)
 971 {
 972         struct rpc_rqst *req = task->tk_rqstp;
 973         struct rpc_xprt *xprt = req->rq_xprt;
 974
 975         if (task->tk_status != -ETIMEDOUT)
 976                 return;
 977         dprintk("RPC: %5u xprt_timer\n", task->tk_pid);
 978
 979         if (!req->rq_reply_bytes_recvd) {
 980                 if (xprt->ops->timer)
 981                         xprt->ops->timer(xprt, task);
 982         } else
 983                 task->tk_status = 0;
 984 }
 985
 986 /**
 987  * xprt_prepare_transmit - reserve the transport before sending a request
 988  * @task: RPC task about to send a request
 989  *
 990  */
 991 bool xprt_prepare_transmit(struct rpc_task *task)
 992 {
 993         struct rpc_rqst *req = task->tk_rqstp;
 994         struct rpc_xprt *xprt = req->rq_xprt;
 995         bool ret = false;
 996
 997         dprintk("RPC: %5u xprt_prepare_transmit\n", task->tk_pid);
 998
 999         spin_lock_bh(&xprt->transport_lock);
1000         if (!req->rq_bytes_sent) {
1001                 if (req->rq_reply_bytes_recvd) {
1002                         task->tk_status = req->rq_reply_bytes_recvd;
1003                         goto out_unlock;
1004                 }
1005                 if ((task->tk_flags & RPC_TASK_NO_RETRANS_TIMEOUT)
1006                     && xprt_connected(xprt)
1007                     && req->rq_connect_cookie == xprt->connect_cookie) {
1008                         xprt->ops->set_retrans_timeout(task);
1009                         rpc_sleep_on(&xprt->pending, task, xprt_timer);
1010                         goto out_unlock;
1011                 }
1012         }
1013         if (!xprt->ops->reserve_xprt(xprt, task)) {
1014                 task->tk_status = -EAGAIN;
1015                 goto out_unlock;
1016         }
1017         ret = true;
1018 out_unlock:
1019         spin_unlock_bh(&xprt->transport_lock);
1020         return ret;
1021 }
1022
1023 void xprt_end_transmit(struct rpc_task *task)
1024 {
1025         xprt_release_write(task->tk_rqstp->rq_xprt, task);
1026 }
1027
1028 /**
1029  * xprt_transmit - send an RPC request on a transport
1030  * @task: controlling RPC task
1031  *
1032  * We have to copy the iovec because sendmsg fiddles with its contents.
1033  */
1034 void xprt_transmit(struct rpc_task *task)
1035 {
1036         struct rpc_rqst *req = task->tk_rqstp;
1037         struct rpc_xprt *xprt = req->rq_xprt;
1038         unsigned int connect_cookie;
1039         int status, numreqs;
1040
1041         dprintk("RPC: %5u xprt_transmit(%u)\n", task->tk_pid, req->rq_slen);
1042
1043         if (!req->rq_reply_bytes_recvd) {
1044                 if (list_empty(&req->rq_list) && rpc_reply_expected(task)) {
1045                         /*
1046                          * Add to the list only if we're expecting a reply
1047                          */
1048                         /* Update the softirq receive buffer */
1049                         memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
1050                                         sizeof(req->rq_private_buf));
1051                         /* Add request to the receive list */
1052                         spin_lock(&xprt->recv_lock);
1053                         list_add_tail(&req->rq_list, &xprt->recv);
1054                         spin_unlock(&xprt->recv_lock);
1055                         xprt_reset_majortimeo(req);
1056                         /* Turn off autodisconnect */
1057                         del_singleshot_timer_sync(&xprt->timer);
1058                 }
1059         } else if (!req->rq_bytes_sent)
1060                 return;
1061
1062         connect_cookie = xprt->connect_cookie;
1063         req->rq_xtime = ktime_get();
1064         status = xprt->ops->send_request(task);
1065         trace_xprt_transmit(xprt, req->rq_xid, status);
1066         if (status != 0) {
1067                 task->tk_status = status;
1068                 return;
1069         }
1070         xprt_inject_disconnect(xprt);
1071
1072         dprintk("RPC: %5u xmit complete\n", task->tk_pid);
1073         task->tk_flags |= RPC_TASK_SENT;
1074         spin_lock_bh(&xprt->transport_lock);
1075
1076         xprt->ops->set_retrans_timeout(task);
1077
1078         numreqs = atomic_read(&xprt->num_reqs);
1079         if (numreqs > xprt->stat.max_slots)
1080                 xprt->stat.max_slots = numreqs;
1081         xprt->stat.sends++;
1082         xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
1083         xprt->stat.bklog_u += xprt->backlog.qlen;
1084         xprt->stat.sending_u += xprt->sending.qlen;
1085         xprt->stat.pending_u += xprt->pending.qlen;
1086         spin_unlock_bh(&xprt->transport_lock);
1087
1088         req->rq_connect_cookie = connect_cookie;
1089         if (rpc_reply_expected(task) && !READ_ONCE(req->rq_reply_bytes_recvd)) {
1090                 /*
1091                  * Sleep on the pending queue if we're expecting a reply.
1092                  * The spinlock ensures atomicity between the test of
1093                  * req->rq_reply_bytes_recvd, and the call to rpc_sleep_on().
1094                  */
1095                 spin_lock(&xprt->recv_lock);
1096                 if (!req->rq_reply_bytes_recvd) {
1097                         rpc_sleep_on(&xprt->pending, task, xprt_timer);
1098                         /*
1099                          * Send an extra queue wakeup call if the
1100                          * connection was dropped in case the call to
1101                          * rpc_sleep_on() raced.
1102                          */
1103                         if (!xprt_connected(xprt))
1104                                 xprt_wake_pending_tasks(xprt, -ENOTCONN);
1105                 }
1106                 spin_unlock(&xprt->recv_lock);
1107         }
1108 }
1109
1110 static void xprt_add_backlog(struct rpc_xprt *xprt, struct rpc_task *task)
1111 {
1112         set_bit(XPRT_CONGESTED, &xprt->state);
1113         rpc_sleep_on(&xprt->backlog, task, NULL);
1114 }
1115
1116 static void xprt_wake_up_backlog(struct rpc_xprt *xprt)
1117 {
1118         if (rpc_wake_up_next(&xprt->backlog) == NULL)
1119                 clear_bit(XPRT_CONGESTED, &xprt->state);
1120 }
1121
1122 static bool xprt_throttle_congested(struct rpc_xprt *xprt, struct rpc_task *task)
1123 {
1124         bool ret = false;
1125
1126         if (!test_bit(XPRT_CONGESTED, &xprt->state))
1127                 goto out;
1128         spin_lock(&xprt->reserve_lock);
1129         if (test_bit(XPRT_CONGESTED, &xprt->state)) {
1130                 rpc_sleep_on(&xprt->backlog, task, NULL);
1131                 ret = true;
1132         }
1133         spin_unlock(&xprt->reserve_lock);
1134 out:
1135         return ret;
1136 }
1137
1138 static struct rpc_rqst *xprt_dynamic_alloc_slot(struct rpc_xprt *xprt)
1139 {
1140         struct rpc_rqst *req = ERR_PTR(-EAGAIN);
1141
1142         if (!atomic_add_unless(&xprt->num_reqs, 1, xprt->max_reqs))
1143                 goto out;
1144         spin_unlock(&xprt->reserve_lock);
1145         req = kzalloc(sizeof(struct rpc_rqst), GFP_NOFS);
1146         spin_lock(&xprt->reserve_lock);
1147         if (req != NULL)
1148                 goto out;
1149         atomic_dec(&xprt->num_reqs);
1150         req = ERR_PTR(-ENOMEM);
1151 out:
1152         return req;
1153 }
1154
1155 static bool xprt_dynamic_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1156 {
1157         if (atomic_add_unless(&xprt->num_reqs, -1, xprt->min_reqs)) {
1158                 kfree(req);
1159                 return true;
1160         }
1161         return false;
1162 }
1163
1164 void xprt_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1165 {
1166         struct rpc_rqst *req;
1167
1168         spin_lock(&xprt->reserve_lock);
1169         if (!list_empty(&xprt->free)) {
1170                 req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
1171                 list_del(&req->rq_list);
1172                 goto out_init_req;
1173         }
1174         req = xprt_dynamic_alloc_slot(xprt);
1175         if (!IS_ERR(req))
1176                 goto out_init_req;
1177         switch (PTR_ERR(req)) {
1178         case -ENOMEM:
1179                 dprintk("RPC:       dynamic allocation of request slot "
1180                                 "failed! Retrying\n");
1181                 task->tk_status = -ENOMEM;
1182                 break;
1183         case -EAGAIN:
1184                 xprt_add_backlog(xprt, task);
1185                 dprintk("RPC:       waiting for request slot\n");
1186         default:
1187                 task->tk_status = -EAGAIN;
1188         }
1189         spin_unlock(&xprt->reserve_lock);
1190         return;
1191 out_init_req:
1192         task->tk_status = 0;
1193         task->tk_rqstp = req;
1194         xprt_request_init(task, xprt);
1195         spin_unlock(&xprt->reserve_lock);
1196 }
1197 EXPORT_SYMBOL_GPL(xprt_alloc_slot);
1198
1199 void xprt_lock_and_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1200 {
1201         /* Note: grabbing the xprt_lock_write() ensures that we throttle
1202          * new slot allocation if the transport is congested (i.e. when
1203          * reconnecting a stream transport or when out of socket write
1204          * buffer space).
1205          */
1206         if (xprt_lock_write(xprt, task)) {
1207                 xprt_alloc_slot(xprt, task);
1208                 xprt_release_write(xprt, task);
1209         }
1210 }
1211 EXPORT_SYMBOL_GPL(xprt_lock_and_alloc_slot);
1212
1213 static void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1214 {
1215         spin_lock(&xprt->reserve_lock);
1216         if (!xprt_dynamic_free_slot(xprt, req)) {
1217                 memset(req, 0, sizeof(*req));   /* mark unused */
1218                 list_add(&req->rq_list, &xprt->free);
1219         }
1220         xprt_wake_up_backlog(xprt);
1221         spin_unlock(&xprt->reserve_lock);
1222 }
1223
1224 static void xprt_free_all_slots(struct rpc_xprt *xprt)
1225 {
1226         struct rpc_rqst *req;
1227         while (!list_empty(&xprt->free)) {
1228                 req = list_first_entry(&xprt->free, struct rpc_rqst, rq_list);
1229                 list_del(&req->rq_list);
1230                 kfree(req);
1231         }
1232 }
1233
1234 struct rpc_xprt *xprt_alloc(struct net *net, size_t size,
1235                 unsigned int num_prealloc,
1236                 unsigned int max_alloc)
1237 {
1238         struct rpc_xprt *xprt;
1239         struct rpc_rqst *req;
1240         int i;
1241
1242         xprt = kzalloc(size, GFP_KERNEL);
1243         if (xprt == NULL)
1244                 goto out;
1245
1246         xprt_init(xprt, net);
1247
1248         for (i = 0; i < num_prealloc; i++) {
1249                 req = kzalloc(sizeof(struct rpc_rqst), GFP_KERNEL);
1250                 if (!req)
1251                         goto out_free;
1252                 list_add(&req->rq_list, &xprt->free);
1253         }
1254         if (max_alloc > num_prealloc)
1255                 xprt->max_reqs = max_alloc;
1256         else
1257                 xprt->max_reqs = num_prealloc;
1258         xprt->min_reqs = num_prealloc;
1259         atomic_set(&xprt->num_reqs, num_prealloc);
1260
1261         return xprt;
1262
1263 out_free:
1264         xprt_free(xprt);
1265 out:
1266         return NULL;
1267 }
1268 EXPORT_SYMBOL_GPL(xprt_alloc);
1269
1270 void xprt_free(struct rpc_xprt *xprt)
1271 {
1272         put_net(xprt->xprt_net);
1273         xprt_free_all_slots(xprt);
1274         kfree_rcu(xprt, rcu);
1275 }
1276 EXPORT_SYMBOL_GPL(xprt_free);
1277
1278 /**
1279  * xprt_reserve - allocate an RPC request slot
1280  * @task: RPC task requesting a slot allocation
1281  *
1282  * If the transport is marked as being congested, or if no more
1283  * slots are available, place the task on the transport's
1284  * backlog queue.
1285  */
1286 void xprt_reserve(struct rpc_task *task)
1287 {
1288         struct rpc_xprt *xprt = task->tk_xprt;
1289
1290         task->tk_status = 0;
1291         if (task->tk_rqstp != NULL)
1292                 return;
1293
1294         task->tk_timeout = 0;
1295         task->tk_status = -EAGAIN;
1296         if (!xprt_throttle_congested(xprt, task))
1297                 xprt->ops->alloc_slot(xprt, task);
1298 }
1299
1300 /**
1301  * xprt_retry_reserve - allocate an RPC request slot
1302  * @task: RPC task requesting a slot allocation
1303  *
1304  * If no more slots are available, place the task on the transport's
1305  * backlog queue.
1306  * Note that the only difference with xprt_reserve is that we now
1307  * ignore the value of the XPRT_CONGESTED flag.
1308  */
1309 void xprt_retry_reserve(struct rpc_task *task)
1310 {
1311         struct rpc_xprt *xprt = task->tk_xprt;
1312
1313         task->tk_status = 0;
1314         if (task->tk_rqstp != NULL)
1315                 return;
1316
1317         task->tk_timeout = 0;
1318         task->tk_status = -EAGAIN;
1319         xprt->ops->alloc_slot(xprt, task);
1320 }
1321
1322 static inline __be32 xprt_alloc_xid(struct rpc_xprt *xprt)
1323 {
1324         return (__force __be32)xprt->xid++;
1325 }
1326
1327 static inline void xprt_init_xid(struct rpc_xprt *xprt)
1328 {
1329         xprt->xid = prandom_u32();
1330 }
1331
1332 static void xprt_request_init(struct rpc_task *task, struct rpc_xprt *xprt)
1333 {
1334         struct rpc_rqst *req = task->tk_rqstp;
1335
1336         INIT_LIST_HEAD(&req->rq_list);
1337         req->rq_timeout = task->tk_client->cl_timeout->to_initval;
1338         req->rq_task    = task;
1339         req->rq_xprt    = xprt;
1340         req->rq_buffer  = NULL;
1341         req->rq_xid     = xprt_alloc_xid(xprt);
1342         req->rq_connect_cookie = xprt->connect_cookie - 1;
1343         req->rq_bytes_sent = 0;
1344         req->rq_snd_buf.len = 0;
1345         req->rq_snd_buf.buflen = 0;
1346         req->rq_rcv_buf.len = 0;
1347         req->rq_rcv_buf.buflen = 0;
1348         req->rq_release_snd_buf = NULL;
1349         xprt_reset_majortimeo(req);
1350         dprintk("RPC: %5u reserved req %p xid %08x\n", task->tk_pid,
1351                         req, ntohl(req->rq_xid));
1352 }
1353
1354 /**
1355  * xprt_release - release an RPC request slot
1356  * @task: task which is finished with the slot
1357  *
1358  */
1359 void xprt_release(struct rpc_task *task)
1360 {
1361         struct rpc_xprt *xprt;
1362         struct rpc_rqst *req = task->tk_rqstp;
1363
1364         if (req == NULL) {
1365                 if (task->tk_client) {
1366                         xprt = task->tk_xprt;
1367                         if (xprt->snd_task == task)
1368                                 xprt_release_write(xprt, task);
1369                 }
1370                 return;
1371         }
1372
1373         xprt = req->rq_xprt;
1374         if (task->tk_ops->rpc_count_stats != NULL)
1375                 task->tk_ops->rpc_count_stats(task, task->tk_calldata);
1376         else if (task->tk_client)
1377                 rpc_count_iostats(task, task->tk_client->cl_metrics);
1378         spin_lock(&xprt->recv_lock);
1379         if (!list_empty(&req->rq_list)) {
1380                 list_del_init(&req->rq_list);
1381                 xprt_wait_on_pinned_rqst(req);
1382         }
1383         spin_unlock(&xprt->recv_lock);
1384         spin_lock_bh(&xprt->transport_lock);
1385         xprt->ops->release_xprt(xprt, task);
1386         if (xprt->ops->release_request)
1387                 xprt->ops->release_request(task);
1388         xprt->last_used = jiffies;
1389         xprt_schedule_autodisconnect(xprt);
1390         spin_unlock_bh(&xprt->transport_lock);
1391         if (req->rq_buffer)
1392                 xprt->ops->buf_free(task);
1393         xprt_inject_disconnect(xprt);
1394         if (req->rq_cred != NULL)
1395                 put_rpccred(req->rq_cred);
1396         task->tk_rqstp = NULL;
1397         if (req->rq_release_snd_buf)
1398                 req->rq_release_snd_buf(req);
1399
1400         dprintk("RPC: %5u release request %p\n", task->tk_pid, req);
1401         if (likely(!bc_prealloc(req)))
1402                 xprt_free_slot(xprt, req);
1403         else
1404                 xprt_free_bc_request(req);
1405 }
1406
1407 static void xprt_init(struct rpc_xprt *xprt, struct net *net)
1408 {
1409         kref_init(&xprt->kref);
1410
1411         spin_lock_init(&xprt->transport_lock);
1412         spin_lock_init(&xprt->reserve_lock);
1413         spin_lock_init(&xprt->recv_lock);
1414
1415         INIT_LIST_HEAD(&xprt->free);
1416         INIT_LIST_HEAD(&xprt->recv);
1417 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1418         spin_lock_init(&xprt->bc_pa_lock);
1419         INIT_LIST_HEAD(&xprt->bc_pa_list);
1420 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1421         INIT_LIST_HEAD(&xprt->xprt_switch);
1422
1423         xprt->last_used = jiffies;
1424         xprt->cwnd = RPC_INITCWND;
1425         xprt->bind_index = 0;
1426
1427         rpc_init_wait_queue(&xprt->binding, "xprt_binding");
1428         rpc_init_wait_queue(&xprt->pending, "xprt_pending");
1429         rpc_init_priority_wait_queue(&xprt->sending, "xprt_sending");
1430         rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
1431
1432         xprt_init_xid(xprt);
1433
1434         xprt->xprt_net = get_net(net);
1435 }
1436
1437 /**
1438  * xprt_create_transport - create an RPC transport
1439  * @args: rpc transport creation arguments
1440  *
1441  */
1442 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
1443 {
1444         struct rpc_xprt *xprt;
1445         struct xprt_class *t;
1446
1447         spin_lock(&xprt_list_lock);
1448         list_for_each_entry(t, &xprt_list, list) {
1449                 if (t->ident == args->ident) {
1450                         spin_unlock(&xprt_list_lock);
1451                         goto found;
1452                 }
1453         }
1454         spin_unlock(&xprt_list_lock);
1455         dprintk("RPC: transport (%d) not supported\n", args->ident);
1456         return ERR_PTR(-EIO);
1457
1458 found:
1459         xprt = t->setup(args);
1460         if (IS_ERR(xprt)) {
1461                 dprintk("RPC:       xprt_create_transport: failed, %ld\n",
1462                                 -PTR_ERR(xprt));
1463                 goto out;
1464         }
1465         if (args->flags & XPRT_CREATE_NO_IDLE_TIMEOUT)
1466                 xprt->idle_timeout = 0;
1467         INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
1468         if (xprt_has_timer(xprt))
1469                 setup_timer(&xprt->timer, xprt_init_autodisconnect,
1470                             (unsigned long)xprt);
1471         else
1472                 init_timer(&xprt->timer);
1473
1474         if (strlen(args->servername) > RPC_MAXNETNAMELEN) {
1475                 xprt_destroy(xprt);
1476                 return ERR_PTR(-EINVAL);
1477         }
1478         xprt->servername = kstrdup(args->servername, GFP_KERNEL);
1479         if (xprt->servername == NULL) {
1480                 xprt_destroy(xprt);
1481                 return ERR_PTR(-ENOMEM);
1482         }
1483
1484         rpc_xprt_debugfs_register(xprt);
1485
1486         dprintk("RPC:       created transport %p with %u slots\n", xprt,
1487                         xprt->max_reqs);
1488 out:
1489         return xprt;
1490 }
1491
1492 static void xprt_destroy_cb(struct work_struct *work)
1493 {
1494         struct rpc_xprt *xprt =
1495                 container_of(work, struct rpc_xprt, task_cleanup);
1496
1497         rpc_xprt_debugfs_unregister(xprt);
1498         rpc_destroy_wait_queue(&xprt->binding);
1499         rpc_destroy_wait_queue(&xprt->pending);
1500         rpc_destroy_wait_queue(&xprt->sending);
1501         rpc_destroy_wait_queue(&xprt->backlog);
1502         kfree(xprt->servername);
1503         /*
1504          * Tear down transport state and free the rpc_xprt
1505          */
1506         xprt->ops->destroy(xprt);
1507 }
1508
1509 /**
1510  * xprt_destroy - destroy an RPC transport, killing off all requests.
1511  * @xprt: transport to destroy
1512  *
1513  */
1514 static void xprt_destroy(struct rpc_xprt *xprt)
1515 {
1516         dprintk("RPC:       destroying transport %p\n", xprt);
1517
1518         /*
1519          * Exclude transport connect/disconnect handlers and autoclose
1520          */
1521         wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_UNINTERRUPTIBLE);
1522
1523         del_timer_sync(&xprt->timer);
1524
1525         /*
1526          * Destroy sockets etc from the system workqueue so they can
1527          * safely flush receive work running on rpciod.
1528          */
1529         INIT_WORK(&xprt->task_cleanup, xprt_destroy_cb);
1530         schedule_work(&xprt->task_cleanup);
1531 }
1532
1533 static void xprt_destroy_kref(struct kref *kref)
1534 {
1535         xprt_destroy(container_of(kref, struct rpc_xprt, kref));
1536 }
1537
1538 /**
1539  * xprt_get - return a reference to an RPC transport.
1540  * @xprt: pointer to the transport
1541  *
1542  */
1543 struct rpc_xprt *xprt_get(struct rpc_xprt *xprt)
1544 {
1545         if (xprt != NULL && kref_get_unless_zero(&xprt->kref))
1546                 return xprt;
1547         return NULL;
1548 }
1549 EXPORT_SYMBOL_GPL(xprt_get);
1550
1551 /**
1552  * xprt_put - release a reference to an RPC transport.
1553  * @xprt: pointer to the transport
1554  *
1555  */
1556 void xprt_put(struct rpc_xprt *xprt)
1557 {
1558         if (xprt != NULL)
1559                 kref_put(&xprt->kref, xprt_destroy_kref);
1560 }
1561 EXPORT_SYMBOL_GPL(xprt_put);