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 __be32   xprt_alloc_xid(struct rpc_xprt *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_on_wq(xprtiod_workqueue,
 554                                 &xprt->pending, xprt->snd_task);
 555         }
 556         spin_unlock_bh(&xprt->transport_lock);
 557 }
 558 EXPORT_SYMBOL_GPL(xprt_write_space);
 559
 560 /**
 561  * xprt_set_retrans_timeout_def - set a request's retransmit timeout
 562  * @task: task whose timeout is to be set
 563  *
 564  * Set a request's retransmit timeout based on the transport's
 565  * default timeout parameters.  Used by transports that don't adjust
 566  * the retransmit timeout based on round-trip time estimation.
 567  */
 568 void xprt_set_retrans_timeout_def(struct rpc_task *task)
 569 {
 570         task->tk_timeout = task->tk_rqstp->rq_timeout;
 571 }
 572 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_def);
 573
 574 /**
 575  * xprt_set_retrans_timeout_rtt - set a request's retransmit timeout
 576  * @task: task whose timeout is to be set
 577  *
 578  * Set a request's retransmit timeout using the RTT estimator.
 579  */
 580 void xprt_set_retrans_timeout_rtt(struct rpc_task *task)
 581 {
 582         int timer = task->tk_msg.rpc_proc->p_timer;
 583         struct rpc_clnt *clnt = task->tk_client;
 584         struct rpc_rtt *rtt = clnt->cl_rtt;
 585         struct rpc_rqst *req = task->tk_rqstp;
 586         unsigned long max_timeout = clnt->cl_timeout->to_maxval;
 587
 588         task->tk_timeout = rpc_calc_rto(rtt, timer);
 589         task->tk_timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
 590         if (task->tk_timeout > max_timeout || task->tk_timeout == 0)
 591                 task->tk_timeout = max_timeout;
 592 }
 593 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_rtt);
 594
 595 static void xprt_reset_majortimeo(struct rpc_rqst *req)
 596 {
 597         const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
 598
 599         req->rq_majortimeo = req->rq_timeout;
 600         if (to->to_exponential)
 601                 req->rq_majortimeo <<= to->to_retries;
 602         else
 603                 req->rq_majortimeo += to->to_increment * to->to_retries;
 604         if (req->rq_majortimeo > to->to_maxval || req->rq_majortimeo == 0)
 605                 req->rq_majortimeo = to->to_maxval;
 606         req->rq_majortimeo += jiffies;
 607 }
 608
 609 /**
 610  * xprt_adjust_timeout - adjust timeout values for next retransmit
 611  * @req: RPC request containing parameters to use for the adjustment
 612  *
 613  */
 614 int xprt_adjust_timeout(struct rpc_rqst *req)
 615 {
 616         struct rpc_xprt *xprt = req->rq_xprt;
 617         const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
 618         int status = 0;
 619
 620         if (time_before(jiffies, req->rq_majortimeo)) {
 621                 if (to->to_exponential)
 622                         req->rq_timeout <<= 1;
 623                 else
 624                         req->rq_timeout += to->to_increment;
 625                 if (to->to_maxval && req->rq_timeout >= to->to_maxval)
 626                         req->rq_timeout = to->to_maxval;
 627                 req->rq_retries++;
 628         } else {
 629                 req->rq_timeout = to->to_initval;
 630                 req->rq_retries = 0;
 631                 xprt_reset_majortimeo(req);
 632                 /* Reset the RTT counters == "slow start" */
 633                 spin_lock_bh(&xprt->transport_lock);
 634                 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
 635                 spin_unlock_bh(&xprt->transport_lock);
 636                 status = -ETIMEDOUT;
 637         }
 638
 639         if (req->rq_timeout == 0) {
 640                 printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
 641                 req->rq_timeout = 5 * HZ;
 642         }
 643         return status;
 644 }
 645
 646 static void xprt_autoclose(struct work_struct *work)
 647 {
 648         struct rpc_xprt *xprt =
 649                 container_of(work, struct rpc_xprt, task_cleanup);
 650
 651         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 652         xprt->ops->close(xprt);
 653         xprt_release_write(xprt, NULL);
 654         wake_up_bit(&xprt->state, XPRT_LOCKED);
 655 }
 656
 657 /**
 658  * xprt_disconnect_done - mark a transport as disconnected
 659  * @xprt: transport to flag for disconnect
 660  *
 661  */
 662 void xprt_disconnect_done(struct rpc_xprt *xprt)
 663 {
 664         dprintk("RPC:       disconnected transport %p\n", xprt);
 665         spin_lock_bh(&xprt->transport_lock);
 666         xprt_clear_connected(xprt);
 667         xprt_wake_pending_tasks(xprt, -EAGAIN);
 668         spin_unlock_bh(&xprt->transport_lock);
 669 }
 670 EXPORT_SYMBOL_GPL(xprt_disconnect_done);
 671
 672 /**
 673  * xprt_force_disconnect - force a transport to disconnect
 674  * @xprt: transport to disconnect
 675  *
 676  */
 677 void xprt_force_disconnect(struct rpc_xprt *xprt)
 678 {
 679         /* Don't race with the test_bit() in xprt_clear_locked() */
 680         spin_lock_bh(&xprt->transport_lock);
 681         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 682         /* Try to schedule an autoclose RPC call */
 683         if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 684                 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 685         xprt_wake_pending_tasks(xprt, -EAGAIN);
 686         spin_unlock_bh(&xprt->transport_lock);
 687 }
 688 EXPORT_SYMBOL_GPL(xprt_force_disconnect);
 689
 690 /**
 691  * xprt_conditional_disconnect - force a transport to disconnect
 692  * @xprt: transport to disconnect
 693  * @cookie: 'connection cookie'
 694  *
 695  * This attempts to break the connection if and only if 'cookie' matches
 696  * the current transport 'connection cookie'. It ensures that we don't
 697  * try to break the connection more than once when we need to retransmit
 698  * a batch of RPC requests.
 699  *
 700  */
 701 void xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie)
 702 {
 703         /* Don't race with the test_bit() in xprt_clear_locked() */
 704         spin_lock_bh(&xprt->transport_lock);
 705         if (cookie != xprt->connect_cookie)
 706                 goto out;
 707         if (test_bit(XPRT_CLOSING, &xprt->state))
 708                 goto out;
 709         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 710         /* Try to schedule an autoclose RPC call */
 711         if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 712                 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 713         xprt_wake_pending_tasks(xprt, -EAGAIN);
 714 out:
 715         spin_unlock_bh(&xprt->transport_lock);
 716 }
 717
 718 static bool
 719 xprt_has_timer(const struct rpc_xprt *xprt)
 720 {
 721         return xprt->idle_timeout != 0;
 722 }
 723
 724 static void
 725 xprt_schedule_autodisconnect(struct rpc_xprt *xprt)
 726         __must_hold(&xprt->transport_lock)
 727 {
 728         if (list_empty(&xprt->recv) && xprt_has_timer(xprt))
 729                 mod_timer(&xprt->timer, xprt->last_used + xprt->idle_timeout);
 730 }
 731
 732 static void
 733 xprt_init_autodisconnect(struct timer_list *t)
 734 {
 735         struct rpc_xprt *xprt = from_timer(xprt, t, timer);
 736
 737         spin_lock(&xprt->transport_lock);
 738         if (!list_empty(&xprt->recv))
 739                 goto out_abort;
 740         /* Reset xprt->last_used to avoid connect/autodisconnect cycling */
 741         xprt->last_used = jiffies;
 742         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 743                 goto out_abort;
 744         spin_unlock(&xprt->transport_lock);
 745         queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 746         return;
 747 out_abort:
 748         spin_unlock(&xprt->transport_lock);
 749 }
 750
 751 bool xprt_lock_connect(struct rpc_xprt *xprt,
 752                 struct rpc_task *task,
 753                 void *cookie)
 754 {
 755         bool ret = false;
 756
 757         spin_lock_bh(&xprt->transport_lock);
 758         if (!test_bit(XPRT_LOCKED, &xprt->state))
 759                 goto out;
 760         if (xprt->snd_task != task)
 761                 goto out;
 762         xprt_task_clear_bytes_sent(task);
 763         xprt->snd_task = cookie;
 764         ret = true;
 765 out:
 766         spin_unlock_bh(&xprt->transport_lock);
 767         return ret;
 768 }
 769
 770 void xprt_unlock_connect(struct rpc_xprt *xprt, void *cookie)
 771 {
 772         spin_lock_bh(&xprt->transport_lock);
 773         if (xprt->snd_task != cookie)
 774                 goto out;
 775         if (!test_bit(XPRT_LOCKED, &xprt->state))
 776                 goto out;
 777         xprt->snd_task =NULL;
 778         xprt->ops->release_xprt(xprt, NULL);
 779         xprt_schedule_autodisconnect(xprt);
 780 out:
 781         spin_unlock_bh(&xprt->transport_lock);
 782         wake_up_bit(&xprt->state, XPRT_LOCKED);
 783 }
 784
 785 /**
 786  * xprt_connect - schedule a transport connect operation
 787  * @task: RPC task that is requesting the connect
 788  *
 789  */
 790 void xprt_connect(struct rpc_task *task)
 791 {
 792         struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
 793
 794         dprintk("RPC: %5u xprt_connect xprt %p %s connected\n", task->tk_pid,
 795                         xprt, (xprt_connected(xprt) ? "is" : "is not"));
 796
 797         if (!xprt_bound(xprt)) {
 798                 task->tk_status = -EAGAIN;
 799                 return;
 800         }
 801         if (!xprt_lock_write(xprt, task))
 802                 return;
 803
 804         if (test_and_clear_bit(XPRT_CLOSE_WAIT, &xprt->state))
 805                 xprt->ops->close(xprt);
 806
 807         if (!xprt_connected(xprt)) {
 808                 task->tk_rqstp->rq_bytes_sent = 0;
 809                 task->tk_timeout = task->tk_rqstp->rq_timeout;
 810                 task->tk_rqstp->rq_connect_cookie = xprt->connect_cookie;
 811                 rpc_sleep_on(&xprt->pending, task, xprt_connect_status);
 812
 813                 if (test_bit(XPRT_CLOSING, &xprt->state))
 814                         return;
 815                 if (xprt_test_and_set_connecting(xprt))
 816                         return;
 817                 /* Race breaker */
 818                 if (!xprt_connected(xprt)) {
 819                         xprt->stat.connect_start = jiffies;
 820                         xprt->ops->connect(xprt, task);
 821                 } else {
 822                         xprt_clear_connecting(xprt);
 823                         task->tk_status = 0;
 824                         rpc_wake_up_queued_task(&xprt->pending, task);
 825                 }
 826         }
 827         xprt_release_write(xprt, task);
 828 }
 829
 830 static void xprt_connect_status(struct rpc_task *task)
 831 {
 832         switch (task->tk_status) {
 833         case 0:
 834                 dprintk("RPC: %5u xprt_connect_status: connection established\n",
 835                                 task->tk_pid);
 836                 break;
 837         case -ECONNREFUSED:
 838         case -ECONNRESET:
 839         case -ECONNABORTED:
 840         case -ENETUNREACH:
 841         case -EHOSTUNREACH:
 842         case -EPIPE:
 843         case -EAGAIN:
 844                 dprintk("RPC: %5u xprt_connect_status: retrying\n", task->tk_pid);
 845                 break;
 846         case -ETIMEDOUT:
 847                 dprintk("RPC: %5u xprt_connect_status: connect attempt timed "
 848                                 "out\n", task->tk_pid);
 849                 break;
 850         default:
 851                 dprintk("RPC: %5u xprt_connect_status: error %d connecting to "
 852                                 "server %s\n", task->tk_pid, -task->tk_status,
 853                                 task->tk_rqstp->rq_xprt->servername);
 854                 task->tk_status = -EIO;
 855         }
 856 }
 857
 858 /**
 859  * xprt_lookup_rqst - find an RPC request corresponding to an XID
 860  * @xprt: transport on which the original request was transmitted
 861  * @xid: RPC XID of incoming reply
 862  *
 863  * Caller holds xprt->recv_lock.
 864  */
 865 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
 866 {
 867         struct rpc_rqst *entry;
 868
 869         list_for_each_entry(entry, &xprt->recv, rq_list)
 870                 if (entry->rq_xid == xid) {
 871                         trace_xprt_lookup_rqst(xprt, xid, 0);
 872                         entry->rq_rtt = ktime_sub(ktime_get(), entry->rq_xtime);
 873                         return entry;
 874                 }
 875
 876         dprintk("RPC:       xprt_lookup_rqst did not find xid %08x\n",
 877                         ntohl(xid));
 878         trace_xprt_lookup_rqst(xprt, xid, -ENOENT);
 879         xprt->stat.bad_xids++;
 880         return NULL;
 881 }
 882 EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
 883
 884 /**
 885  * xprt_pin_rqst - Pin a request on the transport receive list
 886  * @req: Request to pin
 887  *
 888  * Caller must ensure this is atomic with the call to xprt_lookup_rqst()
 889  * so should be holding the xprt transport lock.
 890  */
 891 void xprt_pin_rqst(struct rpc_rqst *req)
 892 {
 893         set_bit(RPC_TASK_MSG_RECV, &req->rq_task->tk_runstate);
 894 }
 895 EXPORT_SYMBOL_GPL(xprt_pin_rqst);
 896
 897 /**
 898  * xprt_unpin_rqst - Unpin a request on the transport receive list
 899  * @req: Request to pin
 900  *
 901  * Caller should be holding the xprt transport lock.
 902  */
 903 void xprt_unpin_rqst(struct rpc_rqst *req)
 904 {
 905         struct rpc_task *task = req->rq_task;
 906
 907         clear_bit(RPC_TASK_MSG_RECV, &task->tk_runstate);
 908         if (test_bit(RPC_TASK_MSG_RECV_WAIT, &task->tk_runstate))
 909                 wake_up_bit(&task->tk_runstate, RPC_TASK_MSG_RECV);
 910 }
 911 EXPORT_SYMBOL_GPL(xprt_unpin_rqst);
 912
 913 static void xprt_wait_on_pinned_rqst(struct rpc_rqst *req)
 914 __must_hold(&req->rq_xprt->recv_lock)
 915 {
 916         struct rpc_task *task = req->rq_task;
 917
 918         if (task && test_bit(RPC_TASK_MSG_RECV, &task->tk_runstate)) {
 919                 spin_unlock(&req->rq_xprt->recv_lock);
 920                 set_bit(RPC_TASK_MSG_RECV_WAIT, &task->tk_runstate);
 921                 wait_on_bit(&task->tk_runstate, RPC_TASK_MSG_RECV,
 922                                 TASK_UNINTERRUPTIBLE);
 923                 clear_bit(RPC_TASK_MSG_RECV_WAIT, &task->tk_runstate);
 924                 spin_lock(&req->rq_xprt->recv_lock);
 925         }
 926 }
 927
 928 /**
 929  * xprt_update_rtt - Update RPC RTT statistics
 930  * @task: RPC request that recently completed
 931  *
 932  * Caller holds xprt->recv_lock.
 933  */
 934 void xprt_update_rtt(struct rpc_task *task)
 935 {
 936         struct rpc_rqst *req = task->tk_rqstp;
 937         struct rpc_rtt *rtt = task->tk_client->cl_rtt;
 938         unsigned int timer = task->tk_msg.rpc_proc->p_timer;
 939         long m = usecs_to_jiffies(ktime_to_us(req->rq_rtt));
 940
 941         if (timer) {
 942                 if (req->rq_ntrans == 1)
 943                         rpc_update_rtt(rtt, timer, m);
 944                 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
 945         }
 946 }
 947 EXPORT_SYMBOL_GPL(xprt_update_rtt);
 948
 949 /**
 950  * xprt_complete_rqst - called when reply processing is complete
 951  * @task: RPC request that recently completed
 952  * @copied: actual number of bytes received from the transport
 953  *
 954  * Caller holds xprt->recv_lock.
 955  */
 956 void xprt_complete_rqst(struct rpc_task *task, int copied)
 957 {
 958         struct rpc_rqst *req = task->tk_rqstp;
 959         struct rpc_xprt *xprt = req->rq_xprt;
 960
 961         dprintk("RPC: %5u xid %08x complete (%d bytes received)\n",
 962                         task->tk_pid, ntohl(req->rq_xid), copied);
 963         trace_xprt_complete_rqst(xprt, req->rq_xid, copied);
 964
 965         xprt->stat.recvs++;
 966
 967         list_del_init(&req->rq_list);
 968         req->rq_private_buf.len = copied;
 969         /* Ensure all writes are done before we update */
 970         /* req->rq_reply_bytes_recvd */
 971         smp_wmb();
 972         req->rq_reply_bytes_recvd = copied;
 973         rpc_wake_up_queued_task(&xprt->pending, task);
 974 }
 975 EXPORT_SYMBOL_GPL(xprt_complete_rqst);
 976
 977 static void xprt_timer(struct rpc_task *task)
 978 {
 979         struct rpc_rqst *req = task->tk_rqstp;
 980         struct rpc_xprt *xprt = req->rq_xprt;
 981
 982         if (task->tk_status != -ETIMEDOUT)
 983                 return;
 984
 985         trace_xprt_timer(xprt, req->rq_xid, task->tk_status);
 986         if (!req->rq_reply_bytes_recvd) {
 987                 if (xprt->ops->timer)
 988                         xprt->ops->timer(xprt, task);
 989         } else
 990                 task->tk_status = 0;
 991 }
 992
 993 /**
 994  * xprt_prepare_transmit - reserve the transport before sending a request
 995  * @task: RPC task about to send a request
 996  *
 997  */
 998 bool xprt_prepare_transmit(struct rpc_task *task)
 999 {
1000         struct rpc_rqst *req = task->tk_rqstp;
1001         struct rpc_xprt *xprt = req->rq_xprt;
1002         bool ret = false;
1003
1004         dprintk("RPC: %5u xprt_prepare_transmit\n", task->tk_pid);
1005
1006         spin_lock_bh(&xprt->transport_lock);
1007         if (!req->rq_bytes_sent) {
1008                 if (req->rq_reply_bytes_recvd) {
1009                         task->tk_status = req->rq_reply_bytes_recvd;
1010                         goto out_unlock;
1011                 }
1012                 if ((task->tk_flags & RPC_TASK_NO_RETRANS_TIMEOUT)
1013                     && xprt_connected(xprt)
1014                     && req->rq_connect_cookie == xprt->connect_cookie) {
1015                         xprt->ops->set_retrans_timeout(task);
1016                         rpc_sleep_on(&xprt->pending, task, xprt_timer);
1017                         goto out_unlock;
1018                 }
1019         }
1020         if (!xprt->ops->reserve_xprt(xprt, task)) {
1021                 task->tk_status = -EAGAIN;
1022                 goto out_unlock;
1023         }
1024         ret = true;
1025 out_unlock:
1026         spin_unlock_bh(&xprt->transport_lock);
1027         return ret;
1028 }
1029
1030 void xprt_end_transmit(struct rpc_task *task)
1031 {
1032         xprt_release_write(task->tk_rqstp->rq_xprt, task);
1033 }
1034
1035 /**
1036  * xprt_transmit - send an RPC request on a transport
1037  * @task: controlling RPC task
1038  *
1039  * We have to copy the iovec because sendmsg fiddles with its contents.
1040  */
1041 void xprt_transmit(struct rpc_task *task)
1042 {
1043         struct rpc_rqst *req = task->tk_rqstp;
1044         struct rpc_xprt *xprt = req->rq_xprt;
1045         unsigned int connect_cookie;
1046         int status;
1047
1048         dprintk("RPC: %5u xprt_transmit(%u)\n", task->tk_pid, req->rq_slen);
1049
1050         if (!req->rq_reply_bytes_recvd) {
1051                 if (list_empty(&req->rq_list) && rpc_reply_expected(task)) {
1052                         /*
1053                          * Add to the list only if we're expecting a reply
1054                          */
1055                         /* Update the softirq receive buffer */
1056                         memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
1057                                         sizeof(req->rq_private_buf));
1058                         /* Add request to the receive list */
1059                         spin_lock(&xprt->recv_lock);
1060                         list_add_tail(&req->rq_list, &xprt->recv);
1061                         spin_unlock(&xprt->recv_lock);
1062                         xprt_reset_majortimeo(req);
1063                         /* Turn off autodisconnect */
1064                         del_singleshot_timer_sync(&xprt->timer);
1065                 }
1066         } else if (!req->rq_bytes_sent)
1067                 return;
1068
1069         connect_cookie = xprt->connect_cookie;
1070         status = xprt->ops->send_request(task);
1071         trace_xprt_transmit(xprt, req->rq_xid, status);
1072         if (status != 0) {
1073                 task->tk_status = status;
1074                 return;
1075         }
1076         xprt_inject_disconnect(xprt);
1077
1078         dprintk("RPC: %5u xmit complete\n", task->tk_pid);
1079         task->tk_flags |= RPC_TASK_SENT;
1080         spin_lock_bh(&xprt->transport_lock);
1081
1082         xprt->ops->set_retrans_timeout(task);
1083
1084         xprt->stat.sends++;
1085         xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
1086         xprt->stat.bklog_u += xprt->backlog.qlen;
1087         xprt->stat.sending_u += xprt->sending.qlen;
1088         xprt->stat.pending_u += xprt->pending.qlen;
1089         spin_unlock_bh(&xprt->transport_lock);
1090
1091         req->rq_connect_cookie = connect_cookie;
1092         if (rpc_reply_expected(task) && !READ_ONCE(req->rq_reply_bytes_recvd)) {
1093                 /*
1094                  * Sleep on the pending queue if we're expecting a reply.
1095                  * The spinlock ensures atomicity between the test of
1096                  * req->rq_reply_bytes_recvd, and the call to rpc_sleep_on().
1097                  */
1098                 spin_lock(&xprt->recv_lock);
1099                 if (!req->rq_reply_bytes_recvd) {
1100                         rpc_sleep_on(&xprt->pending, task, xprt_timer);
1101                         /*
1102                          * Send an extra queue wakeup call if the
1103                          * connection was dropped in case the call to
1104                          * rpc_sleep_on() raced.
1105                          */
1106                         if (!xprt_connected(xprt))
1107                                 xprt_wake_pending_tasks(xprt, -ENOTCONN);
1108                 }
1109                 spin_unlock(&xprt->recv_lock);
1110         }
1111 }
1112
1113 static void xprt_add_backlog(struct rpc_xprt *xprt, struct rpc_task *task)
1114 {
1115         set_bit(XPRT_CONGESTED, &xprt->state);
1116         rpc_sleep_on(&xprt->backlog, task, NULL);
1117 }
1118
1119 static void xprt_wake_up_backlog(struct rpc_xprt *xprt)
1120 {
1121         if (rpc_wake_up_next(&xprt->backlog) == NULL)
1122                 clear_bit(XPRT_CONGESTED, &xprt->state);
1123 }
1124
1125 static bool xprt_throttle_congested(struct rpc_xprt *xprt, struct rpc_task *task)
1126 {
1127         bool ret = false;
1128
1129         if (!test_bit(XPRT_CONGESTED, &xprt->state))
1130                 goto out;
1131         spin_lock(&xprt->reserve_lock);
1132         if (test_bit(XPRT_CONGESTED, &xprt->state)) {
1133                 rpc_sleep_on(&xprt->backlog, task, NULL);
1134                 ret = true;
1135         }
1136         spin_unlock(&xprt->reserve_lock);
1137 out:
1138         return ret;
1139 }
1140
1141 static struct rpc_rqst *xprt_dynamic_alloc_slot(struct rpc_xprt *xprt)
1142 {
1143         struct rpc_rqst *req = ERR_PTR(-EAGAIN);
1144
1145         if (xprt->num_reqs >= xprt->max_reqs)
1146                 goto out;
1147         ++xprt->num_reqs;
1148         spin_unlock(&xprt->reserve_lock);
1149         req = kzalloc(sizeof(struct rpc_rqst), GFP_NOFS);
1150         spin_lock(&xprt->reserve_lock);
1151         if (req != NULL)
1152                 goto out;
1153         --xprt->num_reqs;
1154         req = ERR_PTR(-ENOMEM);
1155 out:
1156         return req;
1157 }
1158
1159 static bool xprt_dynamic_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1160 {
1161         if (xprt->num_reqs > xprt->min_reqs) {
1162                 --xprt->num_reqs;
1163                 kfree(req);
1164                 return true;
1165         }
1166         return false;
1167 }
1168
1169 void xprt_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1170 {
1171         struct rpc_rqst *req;
1172
1173         spin_lock(&xprt->reserve_lock);
1174         if (!list_empty(&xprt->free)) {
1175                 req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
1176                 list_del(&req->rq_list);
1177                 goto out_init_req;
1178         }
1179         req = xprt_dynamic_alloc_slot(xprt);
1180         if (!IS_ERR(req))
1181                 goto out_init_req;
1182         switch (PTR_ERR(req)) {
1183         case -ENOMEM:
1184                 dprintk("RPC:       dynamic allocation of request slot "
1185                                 "failed! Retrying\n");
1186                 task->tk_status = -ENOMEM;
1187                 break;
1188         case -EAGAIN:
1189                 xprt_add_backlog(xprt, task);
1190                 dprintk("RPC:       waiting for request slot\n");
1191                 /* fall through */
1192         default:
1193                 task->tk_status = -EAGAIN;
1194         }
1195         spin_unlock(&xprt->reserve_lock);
1196         return;
1197 out_init_req:
1198         xprt->stat.max_slots = max_t(unsigned int, xprt->stat.max_slots,
1199                                      xprt->num_reqs);
1200         spin_unlock(&xprt->reserve_lock);
1201
1202         task->tk_status = 0;
1203         task->tk_rqstp = req;
1204 }
1205 EXPORT_SYMBOL_GPL(xprt_alloc_slot);
1206
1207 void xprt_lock_and_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1208 {
1209         /* Note: grabbing the xprt_lock_write() ensures that we throttle
1210          * new slot allocation if the transport is congested (i.e. when
1211          * reconnecting a stream transport or when out of socket write
1212          * buffer space).
1213          */
1214         if (xprt_lock_write(xprt, task)) {
1215                 xprt_alloc_slot(xprt, task);
1216                 xprt_release_write(xprt, task);
1217         }
1218 }
1219 EXPORT_SYMBOL_GPL(xprt_lock_and_alloc_slot);
1220
1221 void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1222 {
1223         spin_lock(&xprt->reserve_lock);
1224         if (!xprt_dynamic_free_slot(xprt, req)) {
1225                 memset(req, 0, sizeof(*req));   /* mark unused */
1226                 list_add(&req->rq_list, &xprt->free);
1227         }
1228         xprt_wake_up_backlog(xprt);
1229         spin_unlock(&xprt->reserve_lock);
1230 }
1231 EXPORT_SYMBOL_GPL(xprt_free_slot);
1232
1233 static void xprt_free_all_slots(struct rpc_xprt *xprt)
1234 {
1235         struct rpc_rqst *req;
1236         while (!list_empty(&xprt->free)) {
1237                 req = list_first_entry(&xprt->free, struct rpc_rqst, rq_list);
1238                 list_del(&req->rq_list);
1239                 kfree(req);
1240         }
1241 }
1242
1243 struct rpc_xprt *xprt_alloc(struct net *net, size_t size,
1244                 unsigned int num_prealloc,
1245                 unsigned int max_alloc)
1246 {
1247         struct rpc_xprt *xprt;
1248         struct rpc_rqst *req;
1249         int i;
1250
1251         xprt = kzalloc(size, GFP_KERNEL);
1252         if (xprt == NULL)
1253                 goto out;
1254
1255         xprt_init(xprt, net);
1256
1257         for (i = 0; i < num_prealloc; i++) {
1258                 req = kzalloc(sizeof(struct rpc_rqst), GFP_KERNEL);
1259                 if (!req)
1260                         goto out_free;
1261                 list_add(&req->rq_list, &xprt->free);
1262         }
1263         if (max_alloc > num_prealloc)
1264                 xprt->max_reqs = max_alloc;
1265         else
1266                 xprt->max_reqs = num_prealloc;
1267         xprt->min_reqs = num_prealloc;
1268         xprt->num_reqs = num_prealloc;
1269
1270         return xprt;
1271
1272 out_free:
1273         xprt_free(xprt);
1274 out:
1275         return NULL;
1276 }
1277 EXPORT_SYMBOL_GPL(xprt_alloc);
1278
1279 void xprt_free(struct rpc_xprt *xprt)
1280 {
1281         put_net(xprt->xprt_net);
1282         xprt_free_all_slots(xprt);
1283         kfree_rcu(xprt, rcu);
1284 }
1285 EXPORT_SYMBOL_GPL(xprt_free);
1286
1287 static __be32
1288 xprt_alloc_xid(struct rpc_xprt *xprt)
1289 {
1290         __be32 xid;
1291
1292         spin_lock(&xprt->reserve_lock);
1293         xid = (__force __be32)xprt->xid++;
1294         spin_unlock(&xprt->reserve_lock);
1295         return xid;
1296 }
1297
1298 static void
1299 xprt_init_xid(struct rpc_xprt *xprt)
1300 {
1301         xprt->xid = prandom_u32();
1302 }
1303
1304 static void
1305 xprt_request_init(struct rpc_task *task)
1306 {
1307         struct rpc_xprt *xprt = task->tk_xprt;
1308         struct rpc_rqst *req = task->tk_rqstp;
1309
1310         INIT_LIST_HEAD(&req->rq_list);
1311         req->rq_timeout = task->tk_client->cl_timeout->to_initval;
1312         req->rq_task    = task;
1313         req->rq_xprt    = xprt;
1314         req->rq_buffer  = NULL;
1315         req->rq_xid     = xprt_alloc_xid(xprt);
1316         req->rq_connect_cookie = xprt->connect_cookie - 1;
1317         req->rq_bytes_sent = 0;
1318         req->rq_snd_buf.len = 0;
1319         req->rq_snd_buf.buflen = 0;
1320         req->rq_rcv_buf.len = 0;
1321         req->rq_rcv_buf.buflen = 0;
1322         req->rq_release_snd_buf = NULL;
1323         xprt_reset_majortimeo(req);
1324         dprintk("RPC: %5u reserved req %p xid %08x\n", task->tk_pid,
1325                         req, ntohl(req->rq_xid));
1326 }
1327
1328 static void
1329 xprt_do_reserve(struct rpc_xprt *xprt, struct rpc_task *task)
1330 {
1331         xprt->ops->alloc_slot(xprt, task);
1332         if (task->tk_rqstp != NULL)
1333                 xprt_request_init(task);
1334 }
1335
1336 /**
1337  * xprt_reserve - allocate an RPC request slot
1338  * @task: RPC task requesting a slot allocation
1339  *
1340  * If the transport is marked as being congested, or if no more
1341  * slots are available, place the task on the transport's
1342  * backlog queue.
1343  */
1344 void xprt_reserve(struct rpc_task *task)
1345 {
1346         struct rpc_xprt *xprt = task->tk_xprt;
1347
1348         task->tk_status = 0;
1349         if (task->tk_rqstp != NULL)
1350                 return;
1351
1352         task->tk_timeout = 0;
1353         task->tk_status = -EAGAIN;
1354         if (!xprt_throttle_congested(xprt, task))
1355                 xprt_do_reserve(xprt, task);
1356 }
1357
1358 /**
1359  * xprt_retry_reserve - allocate an RPC request slot
1360  * @task: RPC task requesting a slot allocation
1361  *
1362  * If no more slots are available, place the task on the transport's
1363  * backlog queue.
1364  * Note that the only difference with xprt_reserve is that we now
1365  * ignore the value of the XPRT_CONGESTED flag.
1366  */
1367 void xprt_retry_reserve(struct rpc_task *task)
1368 {
1369         struct rpc_xprt *xprt = task->tk_xprt;
1370
1371         task->tk_status = 0;
1372         if (task->tk_rqstp != NULL)
1373                 return;
1374
1375         task->tk_timeout = 0;
1376         task->tk_status = -EAGAIN;
1377         xprt_do_reserve(xprt, task);
1378 }
1379
1380 /**
1381  * xprt_release - release an RPC request slot
1382  * @task: task which is finished with the slot
1383  *
1384  */
1385 void xprt_release(struct rpc_task *task)
1386 {
1387         struct rpc_xprt *xprt;
1388         struct rpc_rqst *req = task->tk_rqstp;
1389
1390         if (req == NULL) {
1391                 if (task->tk_client) {
1392                         xprt = task->tk_xprt;
1393                         if (xprt->snd_task == task)
1394                                 xprt_release_write(xprt, task);
1395                 }
1396                 return;
1397         }
1398
1399         xprt = req->rq_xprt;
1400         if (task->tk_ops->rpc_count_stats != NULL)
1401                 task->tk_ops->rpc_count_stats(task, task->tk_calldata);
1402         else if (task->tk_client)
1403                 rpc_count_iostats(task, task->tk_client->cl_metrics);
1404         spin_lock(&xprt->recv_lock);
1405         if (!list_empty(&req->rq_list)) {
1406                 list_del_init(&req->rq_list);
1407                 xprt_wait_on_pinned_rqst(req);
1408         }
1409         spin_unlock(&xprt->recv_lock);
1410         spin_lock_bh(&xprt->transport_lock);
1411         xprt->ops->release_xprt(xprt, task);
1412         if (xprt->ops->release_request)
1413                 xprt->ops->release_request(task);
1414         xprt->last_used = jiffies;
1415         xprt_schedule_autodisconnect(xprt);
1416         spin_unlock_bh(&xprt->transport_lock);
1417         if (req->rq_buffer)
1418                 xprt->ops->buf_free(task);
1419         xprt_inject_disconnect(xprt);
1420         if (req->rq_cred != NULL)
1421                 put_rpccred(req->rq_cred);
1422         task->tk_rqstp = NULL;
1423         if (req->rq_release_snd_buf)
1424                 req->rq_release_snd_buf(req);
1425
1426         dprintk("RPC: %5u release request %p\n", task->tk_pid, req);
1427         if (likely(!bc_prealloc(req)))
1428                 xprt->ops->free_slot(xprt, req);
1429         else
1430                 xprt_free_bc_request(req);
1431 }
1432
1433 static void xprt_init(struct rpc_xprt *xprt, struct net *net)
1434 {
1435         kref_init(&xprt->kref);
1436
1437         spin_lock_init(&xprt->transport_lock);
1438         spin_lock_init(&xprt->reserve_lock);
1439         spin_lock_init(&xprt->recv_lock);
1440
1441         INIT_LIST_HEAD(&xprt->free);
1442         INIT_LIST_HEAD(&xprt->recv);
1443 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1444         spin_lock_init(&xprt->bc_pa_lock);
1445         INIT_LIST_HEAD(&xprt->bc_pa_list);
1446 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1447         INIT_LIST_HEAD(&xprt->xprt_switch);
1448
1449         xprt->last_used = jiffies;
1450         xprt->cwnd = RPC_INITCWND;
1451         xprt->bind_index = 0;
1452
1453         rpc_init_wait_queue(&xprt->binding, "xprt_binding");
1454         rpc_init_wait_queue(&xprt->pending, "xprt_pending");
1455         rpc_init_priority_wait_queue(&xprt->sending, "xprt_sending");
1456         rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
1457
1458         xprt_init_xid(xprt);
1459
1460         xprt->xprt_net = get_net(net);
1461 }
1462
1463 /**
1464  * xprt_create_transport - create an RPC transport
1465  * @args: rpc transport creation arguments
1466  *
1467  */
1468 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
1469 {
1470         struct rpc_xprt *xprt;
1471         struct xprt_class *t;
1472
1473         spin_lock(&xprt_list_lock);
1474         list_for_each_entry(t, &xprt_list, list) {
1475                 if (t->ident == args->ident) {
1476                         spin_unlock(&xprt_list_lock);
1477                         goto found;
1478                 }
1479         }
1480         spin_unlock(&xprt_list_lock);
1481         dprintk("RPC: transport (%d) not supported\n", args->ident);
1482         return ERR_PTR(-EIO);
1483
1484 found:
1485         xprt = t->setup(args);
1486         if (IS_ERR(xprt)) {
1487                 dprintk("RPC:       xprt_create_transport: failed, %ld\n",
1488                                 -PTR_ERR(xprt));
1489                 goto out;
1490         }
1491         if (args->flags & XPRT_CREATE_NO_IDLE_TIMEOUT)
1492                 xprt->idle_timeout = 0;
1493         INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
1494         if (xprt_has_timer(xprt))
1495                 timer_setup(&xprt->timer, xprt_init_autodisconnect, 0);
1496         else
1497                 timer_setup(&xprt->timer, NULL, 0);
1498
1499         if (strlen(args->servername) > RPC_MAXNETNAMELEN) {
1500                 xprt_destroy(xprt);
1501                 return ERR_PTR(-EINVAL);
1502         }
1503         xprt->servername = kstrdup(args->servername, GFP_KERNEL);
1504         if (xprt->servername == NULL) {
1505                 xprt_destroy(xprt);
1506                 return ERR_PTR(-ENOMEM);
1507         }
1508
1509         rpc_xprt_debugfs_register(xprt);
1510
1511         dprintk("RPC:       created transport %p with %u slots\n", xprt,
1512                         xprt->max_reqs);
1513 out:
1514         return xprt;
1515 }
1516
1517 static void xprt_destroy_cb(struct work_struct *work)
1518 {
1519         struct rpc_xprt *xprt =
1520                 container_of(work, struct rpc_xprt, task_cleanup);
1521
1522         rpc_xprt_debugfs_unregister(xprt);
1523         rpc_destroy_wait_queue(&xprt->binding);
1524         rpc_destroy_wait_queue(&xprt->pending);
1525         rpc_destroy_wait_queue(&xprt->sending);
1526         rpc_destroy_wait_queue(&xprt->backlog);
1527         kfree(xprt->servername);
1528         /*
1529          * Tear down transport state and free the rpc_xprt
1530          */
1531         xprt->ops->destroy(xprt);
1532 }
1533
1534 /**
1535  * xprt_destroy - destroy an RPC transport, killing off all requests.
1536  * @xprt: transport to destroy
1537  *
1538  */
1539 static void xprt_destroy(struct rpc_xprt *xprt)
1540 {
1541         dprintk("RPC:       destroying transport %p\n", xprt);
1542
1543         /*
1544          * Exclude transport connect/disconnect handlers and autoclose
1545          */
1546         wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_UNINTERRUPTIBLE);
1547
1548         /*
1549          * xprt_schedule_autodisconnect() can run after XPRT_LOCKED
1550          * is cleared.  We use ->transport_lock to ensure the mod_timer()
1551          * can only run *before* del_time_sync(), never after.
1552          */
1553         spin_lock_bh(&xprt->transport_lock);
1554         del_timer_sync(&xprt->timer);
1555         spin_unlock_bh(&xprt->transport_lock);
1556
1557         /*
1558          * Destroy sockets etc from the system workqueue so they can
1559          * safely flush receive work running on rpciod.
1560          */
1561         INIT_WORK(&xprt->task_cleanup, xprt_destroy_cb);
1562         schedule_work(&xprt->task_cleanup);
1563 }
1564
1565 static void xprt_destroy_kref(struct kref *kref)
1566 {
1567         xprt_destroy(container_of(kref, struct rpc_xprt, kref));
1568 }
1569
1570 /**
1571  * xprt_get - return a reference to an RPC transport.
1572  * @xprt: pointer to the transport
1573  *
1574  */
1575 struct rpc_xprt *xprt_get(struct rpc_xprt *xprt)
1576 {
1577         if (xprt != NULL && kref_get_unless_zero(&xprt->kref))
1578                 return xprt;
1579         return NULL;
1580 }
1581 EXPORT_SYMBOL_GPL(xprt_get);
1582
1583 /**
1584  * xprt_put - release a reference to an RPC transport.
1585  * @xprt: pointer to the transport
1586  *
1587  */
1588 void xprt_put(struct rpc_xprt *xprt)
1589 {
1590         if (xprt != NULL)
1591                 kref_put(&xprt->kref, xprt_destroy_kref);
1592 }
1593 EXPORT_SYMBOL_GPL(xprt_put);