net/sunrpc/xprtsock.c

   1 /*
   2  * linux/net/sunrpc/xprtsock.c
   3  *
   4  * Client-side transport implementation for sockets.
   5  *
   6  * TCP callback races fixes (C) 1998 Red Hat
   7  * TCP send fixes (C) 1998 Red Hat
   8  * TCP NFS related read + write fixes
   9  *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
  10  *
  11  * Rewrite of larges part of the code in order to stabilize TCP stuff.
  12  * Fix behaviour when socket buffer is full.
  13  *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
  14  *
  15  * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
  16  *
  17  * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
  18  *   <gilles.quillard@bull.net>
  19  */
  20
  21 #include <linux/types.h>
  22 #include <linux/string.h>
  23 #include <linux/slab.h>
  24 #include <linux/module.h>
  25 #include <linux/capability.h>
  26 #include <linux/pagemap.h>
  27 #include <linux/errno.h>
  28 #include <linux/socket.h>
  29 #include <linux/in.h>
  30 #include <linux/net.h>
  31 #include <linux/mm.h>
  32 #include <linux/un.h>
  33 #include <linux/udp.h>
  34 #include <linux/tcp.h>
  35 #include <linux/sunrpc/clnt.h>
  36 #include <linux/sunrpc/addr.h>
  37 #include <linux/sunrpc/sched.h>
  38 #include <linux/sunrpc/svcsock.h>
  39 #include <linux/sunrpc/xprtsock.h>
  40 #include <linux/file.h>
  41 #ifdef CONFIG_SUNRPC_BACKCHANNEL
  42 #include <linux/sunrpc/bc_xprt.h>
  43 #endif
  44
  45 #include <net/sock.h>
  46 #include <net/checksum.h>
  47 #include <net/udp.h>
  48 #include <net/tcp.h>
  49
  50 #include <trace/events/sunrpc.h>
  51
  52 #include "sunrpc.h"
  53
  54 static void xs_close(struct rpc_xprt *xprt);
  55
  56 /*
  57  * xprtsock tunables
  58  */
  59 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  60 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
  61 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
  62
  63 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  64 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  65
  66 #define XS_TCP_LINGER_TO        (15U * HZ)
  67 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  68
  69 /*
  70  * We can register our own files under /proc/sys/sunrpc by
  71  * calling register_sysctl_table() again.  The files in that
  72  * directory become the union of all files registered there.
  73  *
  74  * We simply need to make sure that we don't collide with
  75  * someone else's file names!
  76  */
  77
  78 #ifdef RPC_DEBUG
  79
  80 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  81 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  82 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
  83 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  84 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  85
  86 static struct ctl_table_header *sunrpc_table_header;
  87
  88 /*
  89  * FIXME: changing the UDP slot table size should also resize the UDP
  90  *        socket buffers for existing UDP transports
  91  */
  92 static struct ctl_table xs_tunables_table[] = {
  93         {
  94                 .procname       = "udp_slot_table_entries",
  95                 .data           = &xprt_udp_slot_table_entries,
  96                 .maxlen         = sizeof(unsigned int),
  97                 .mode           = 0644,
  98                 .proc_handler   = proc_dointvec_minmax,
  99                 .extra1         = &min_slot_table_size,
 100                 .extra2         = &max_slot_table_size
 101         },
 102         {
 103                 .procname       = "tcp_slot_table_entries",
 104                 .data           = &xprt_tcp_slot_table_entries,
 105                 .maxlen         = sizeof(unsigned int),
 106                 .mode           = 0644,
 107                 .proc_handler   = proc_dointvec_minmax,
 108                 .extra1         = &min_slot_table_size,
 109                 .extra2         = &max_slot_table_size
 110         },
 111         {
 112                 .procname       = "tcp_max_slot_table_entries",
 113                 .data           = &xprt_max_tcp_slot_table_entries,
 114                 .maxlen         = sizeof(unsigned int),
 115                 .mode           = 0644,
 116                 .proc_handler   = proc_dointvec_minmax,
 117                 .extra1         = &min_slot_table_size,
 118                 .extra2         = &max_tcp_slot_table_limit
 119         },
 120         {
 121                 .procname       = "min_resvport",
 122                 .data           = &xprt_min_resvport,
 123                 .maxlen         = sizeof(unsigned int),
 124                 .mode           = 0644,
 125                 .proc_handler   = proc_dointvec_minmax,
 126                 .extra1         = &xprt_min_resvport_limit,
 127                 .extra2         = &xprt_max_resvport_limit
 128         },
 129         {
 130                 .procname       = "max_resvport",
 131                 .data           = &xprt_max_resvport,
 132                 .maxlen         = sizeof(unsigned int),
 133                 .mode           = 0644,
 134                 .proc_handler   = proc_dointvec_minmax,
 135                 .extra1         = &xprt_min_resvport_limit,
 136                 .extra2         = &xprt_max_resvport_limit
 137         },
 138         {
 139                 .procname       = "tcp_fin_timeout",
 140                 .data           = &xs_tcp_fin_timeout,
 141                 .maxlen         = sizeof(xs_tcp_fin_timeout),
 142                 .mode           = 0644,
 143                 .proc_handler   = proc_dointvec_jiffies,
 144         },
 145         { },
 146 };
 147
 148 static struct ctl_table sunrpc_table[] = {
 149         {
 150                 .procname       = "sunrpc",
 151                 .mode           = 0555,
 152                 .child          = xs_tunables_table
 153         },
 154         { },
 155 };
 156
 157 #endif
 158
 159 /*
 160  * Wait duration for a reply from the RPC portmapper.
 161  */
 162 #define XS_BIND_TO              (60U * HZ)
 163
 164 /*
 165  * Delay if a UDP socket connect error occurs.  This is most likely some
 166  * kind of resource problem on the local host.
 167  */
 168 #define XS_UDP_REEST_TO         (2U * HZ)
 169
 170 /*
 171  * The reestablish timeout allows clients to delay for a bit before attempting
 172  * to reconnect to a server that just dropped our connection.
 173  *
 174  * We implement an exponential backoff when trying to reestablish a TCP
 175  * transport connection with the server.  Some servers like to drop a TCP
 176  * connection when they are overworked, so we start with a short timeout and
 177  * increase over time if the server is down or not responding.
 178  */
 179 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 180 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
 181
 182 /*
 183  * TCP idle timeout; client drops the transport socket if it is idle
 184  * for this long.  Note that we also timeout UDP sockets to prevent
 185  * holding port numbers when there is no RPC traffic.
 186  */
 187 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 188
 189 #ifdef RPC_DEBUG
 190 # undef  RPC_DEBUG_DATA
 191 # define RPCDBG_FACILITY        RPCDBG_TRANS
 192 #endif
 193
 194 #ifdef RPC_DEBUG_DATA
 195 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 196 {
 197         u8 *buf = (u8 *) packet;
 198         int j;
 199
 200         dprintk("RPC:       %s\n", msg);
 201         for (j = 0; j < count && j < 128; j += 4) {
 202                 if (!(j & 31)) {
 203                         if (j)
 204                                 dprintk("\n");
 205                         dprintk("0x%04x ", j);
 206                 }
 207                 dprintk("%02x%02x%02x%02x ",
 208                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 209         }
 210         dprintk("\n");
 211 }
 212 #else
 213 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 214 {
 215         /* NOP */
 216 }
 217 #endif
 218
 219 struct sock_xprt {
 220         struct rpc_xprt         xprt;
 221
 222         /*
 223          * Network layer
 224          */
 225         struct socket *         sock;
 226         struct sock *           inet;
 227
 228         /*
 229          * State of TCP reply receive
 230          */
 231         __be32                  tcp_fraghdr,
 232                                 tcp_xid,
 233                                 tcp_calldir;
 234
 235         u32                     tcp_offset,
 236                                 tcp_reclen;
 237
 238         unsigned long           tcp_copied,
 239                                 tcp_flags;
 240
 241         /*
 242          * Connection of transports
 243          */
 244         struct delayed_work     connect_worker;
 245         struct sockaddr_storage srcaddr;
 246         unsigned short          srcport;
 247
 248         /*
 249          * UDP socket buffer size parameters
 250          */
 251         size_t                  rcvsize,
 252                                 sndsize;
 253
 254         /*
 255          * Saved socket callback addresses
 256          */
 257         void                    (*old_data_ready)(struct sock *, int);
 258         void                    (*old_state_change)(struct sock *);
 259         void                    (*old_write_space)(struct sock *);
 260         void                    (*old_error_report)(struct sock *);
 261 };
 262
 263 /*
 264  * TCP receive state flags
 265  */
 266 #define TCP_RCV_LAST_FRAG       (1UL << 0)
 267 #define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
 268 #define TCP_RCV_COPY_XID        (1UL << 2)
 269 #define TCP_RCV_COPY_DATA       (1UL << 3)
 270 #define TCP_RCV_READ_CALLDIR    (1UL << 4)
 271 #define TCP_RCV_COPY_CALLDIR    (1UL << 5)
 272
 273 /*
 274  * TCP RPC flags
 275  */
 276 #define TCP_RPC_REPLY           (1UL << 6)
 277
 278 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 279 {
 280         return (struct rpc_xprt *) sk->sk_user_data;
 281 }
 282
 283 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 284 {
 285         return (struct sockaddr *) &xprt->addr;
 286 }
 287
 288 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
 289 {
 290         return (struct sockaddr_un *) &xprt->addr;
 291 }
 292
 293 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 294 {
 295         return (struct sockaddr_in *) &xprt->addr;
 296 }
 297
 298 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 299 {
 300         return (struct sockaddr_in6 *) &xprt->addr;
 301 }
 302
 303 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
 304 {
 305         struct sockaddr *sap = xs_addr(xprt);
 306         struct sockaddr_in6 *sin6;
 307         struct sockaddr_in *sin;
 308         struct sockaddr_un *sun;
 309         char buf[128];
 310
 311         switch (sap->sa_family) {
 312         case AF_LOCAL:
 313                 sun = xs_addr_un(xprt);
 314                 strlcpy(buf, sun->sun_path, sizeof(buf));
 315                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 316                                                 kstrdup(buf, GFP_KERNEL);
 317                 break;
 318         case AF_INET:
 319                 (void)rpc_ntop(sap, buf, sizeof(buf));
 320                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 321                                                 kstrdup(buf, GFP_KERNEL);
 322                 sin = xs_addr_in(xprt);
 323                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 324                 break;
 325         case AF_INET6:
 326                 (void)rpc_ntop(sap, buf, sizeof(buf));
 327                 xprt->address_strings[RPC_DISPLAY_ADDR] =
 328                                                 kstrdup(buf, GFP_KERNEL);
 329                 sin6 = xs_addr_in6(xprt);
 330                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
 331                 break;
 332         default:
 333                 BUG();
 334         }
 335
 336         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 337 }
 338
 339 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
 340 {
 341         struct sockaddr *sap = xs_addr(xprt);
 342         char buf[128];
 343
 344         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 345         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 346
 347         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 348         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 349 }
 350
 351 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
 352                                      const char *protocol,
 353                                      const char *netid)
 354 {
 355         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 356         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 357         xs_format_common_peer_addresses(xprt);
 358         xs_format_common_peer_ports(xprt);
 359 }
 360
 361 static void xs_update_peer_port(struct rpc_xprt *xprt)
 362 {
 363         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 364         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 365
 366         xs_format_common_peer_ports(xprt);
 367 }
 368
 369 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 370 {
 371         unsigned int i;
 372
 373         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 374                 switch (i) {
 375                 case RPC_DISPLAY_PROTO:
 376                 case RPC_DISPLAY_NETID:
 377                         continue;
 378                 default:
 379                         kfree(xprt->address_strings[i]);
 380                 }
 381 }
 382
 383 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 384
 385 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 386 {
 387         struct msghdr msg = {
 388                 .msg_name       = addr,
 389                 .msg_namelen    = addrlen,
 390                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 391         };
 392         struct kvec iov = {
 393                 .iov_base       = vec->iov_base + base,
 394                 .iov_len        = vec->iov_len - base,
 395         };
 396
 397         if (iov.iov_len != 0)
 398                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 399         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 400 }
 401
 402 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy)
 403 {
 404         ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
 405                         int offset, size_t size, int flags);
 406         struct page **ppage;
 407         unsigned int remainder;
 408         int err, sent = 0;
 409
 410         remainder = xdr->page_len - base;
 411         base += xdr->page_base;
 412         ppage = xdr->pages + (base >> PAGE_SHIFT);
 413         base &= ~PAGE_MASK;
 414         do_sendpage = sock->ops->sendpage;
 415         if (!zerocopy)
 416                 do_sendpage = sock_no_sendpage;
 417         for(;;) {
 418                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 419                 int flags = XS_SENDMSG_FLAGS;
 420
 421                 remainder -= len;
 422                 if (remainder != 0 || more)
 423                         flags |= MSG_MORE;
 424                 err = do_sendpage(sock, *ppage, base, len, flags);
 425                 if (remainder == 0 || err != len)
 426                         break;
 427                 sent += err;
 428                 ppage++;
 429                 base = 0;
 430         }
 431         if (sent == 0)
 432                 return err;
 433         if (err > 0)
 434                 sent += err;
 435         return sent;
 436 }
 437
 438 /**
 439  * xs_sendpages - write pages directly to a socket
 440  * @sock: socket to send on
 441  * @addr: UDP only -- address of destination
 442  * @addrlen: UDP only -- length of destination address
 443  * @xdr: buffer containing this request
 444  * @base: starting position in the buffer
 445  * @zerocopy: true if it is safe to use sendpage()
 446  *
 447  */
 448 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy)
 449 {
 450         unsigned int remainder = xdr->len - base;
 451         int err, sent = 0;
 452
 453         if (unlikely(!sock))
 454                 return -ENOTSOCK;
 455
 456         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 457         if (base != 0) {
 458                 addr = NULL;
 459                 addrlen = 0;
 460         }
 461
 462         if (base < xdr->head[0].iov_len || addr != NULL) {
 463                 unsigned int len = xdr->head[0].iov_len - base;
 464                 remainder -= len;
 465                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 466                 if (remainder == 0 || err != len)
 467                         goto out;
 468                 sent += err;
 469                 base = 0;
 470         } else
 471                 base -= xdr->head[0].iov_len;
 472
 473         if (base < xdr->page_len) {
 474                 unsigned int len = xdr->page_len - base;
 475                 remainder -= len;
 476                 err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy);
 477                 if (remainder == 0 || err != len)
 478                         goto out;
 479                 sent += err;
 480                 base = 0;
 481         } else
 482                 base -= xdr->page_len;
 483
 484         if (base >= xdr->tail[0].iov_len)
 485                 return sent;
 486         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 487 out:
 488         if (sent == 0)
 489                 return err;
 490         if (err > 0)
 491                 sent += err;
 492         return sent;
 493 }
 494
 495 static void xs_nospace_callback(struct rpc_task *task)
 496 {
 497         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
 498
 499         transport->inet->sk_write_pending--;
 500         clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 501 }
 502
 503 /**
 504  * xs_nospace - place task on wait queue if transmit was incomplete
 505  * @task: task to put to sleep
 506  *
 507  */
 508 static int xs_nospace(struct rpc_task *task)
 509 {
 510         struct rpc_rqst *req = task->tk_rqstp;
 511         struct rpc_xprt *xprt = req->rq_xprt;
 512         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 513         struct sock *sk = transport->inet;
 514         int ret = -EAGAIN;
 515
 516         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 517                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 518                         req->rq_slen);
 519
 520         /* Protect against races with write_space */
 521         spin_lock_bh(&xprt->transport_lock);
 522
 523         /* Don't race with disconnect */
 524         if (xprt_connected(xprt)) {
 525                 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 526                         /*
 527                          * Notify TCP that we're limited by the application
 528                          * window size
 529                          */
 530                         set_bit(SOCK_NOSPACE, &transport->sock->flags);
 531                         sk->sk_write_pending++;
 532                         /* ...and wait for more buffer space */
 533                         xprt_wait_for_buffer_space(task, xs_nospace_callback);
 534                 }
 535         } else {
 536                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 537                 ret = -ENOTCONN;
 538         }
 539
 540         spin_unlock_bh(&xprt->transport_lock);
 541
 542         /* Race breaker in case memory is freed before above code is called */
 543         sk->sk_write_space(sk);
 544         return ret;
 545 }
 546
 547 /*
 548  * Construct a stream transport record marker in @buf.
 549  */
 550 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
 551 {
 552         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 553         rpc_fraghdr *base = buf->head[0].iov_base;
 554         *base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
 555 }
 556
 557 /**
 558  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
 559  * @task: RPC task that manages the state of an RPC request
 560  *
 561  * Return values:
 562  *        0:    The request has been sent
 563  *   EAGAIN:    The socket was blocked, please call again later to
 564  *              complete the request
 565  * ENOTCONN:    Caller needs to invoke connect logic then call again
 566  *    other:    Some other error occured, the request was not sent
 567  */
 568 static int xs_local_send_request(struct rpc_task *task)
 569 {
 570         struct rpc_rqst *req = task->tk_rqstp;
 571         struct rpc_xprt *xprt = req->rq_xprt;
 572         struct sock_xprt *transport =
 573                                 container_of(xprt, struct sock_xprt, xprt);
 574         struct xdr_buf *xdr = &req->rq_snd_buf;
 575         int status;
 576
 577         xs_encode_stream_record_marker(&req->rq_snd_buf);
 578
 579         xs_pktdump("packet data:",
 580                         req->rq_svec->iov_base, req->rq_svec->iov_len);
 581
 582         status = xs_sendpages(transport->sock, NULL, 0,
 583                                                 xdr, req->rq_bytes_sent, true);
 584         dprintk("RPC:       %s(%u) = %d\n",
 585                         __func__, xdr->len - req->rq_bytes_sent, status);
 586         if (likely(status >= 0)) {
 587                 req->rq_bytes_sent += status;
 588                 req->rq_xmit_bytes_sent += status;
 589                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 590                         req->rq_bytes_sent = 0;
 591                         return 0;
 592                 }
 593                 status = -EAGAIN;
 594         }
 595
 596         switch (status) {
 597         case -EAGAIN:
 598                 status = xs_nospace(task);
 599                 break;
 600         default:
 601                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 602                         -status);
 603         case -EPIPE:
 604                 xs_close(xprt);
 605                 status = -ENOTCONN;
 606         }
 607
 608         return status;
 609 }
 610
 611 /**
 612  * xs_udp_send_request - write an RPC request to a UDP socket
 613  * @task: address of RPC task that manages the state of an RPC request
 614  *
 615  * Return values:
 616  *        0:    The request has been sent
 617  *   EAGAIN:    The socket was blocked, please call again later to
 618  *              complete the request
 619  * ENOTCONN:    Caller needs to invoke connect logic then call again
 620  *    other:    Some other error occurred, the request was not sent
 621  */
 622 static int xs_udp_send_request(struct rpc_task *task)
 623 {
 624         struct rpc_rqst *req = task->tk_rqstp;
 625         struct rpc_xprt *xprt = req->rq_xprt;
 626         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 627         struct xdr_buf *xdr = &req->rq_snd_buf;
 628         int status;
 629
 630         xs_pktdump("packet data:",
 631                                 req->rq_svec->iov_base,
 632                                 req->rq_svec->iov_len);
 633
 634         if (!xprt_bound(xprt))
 635                 return -ENOTCONN;
 636         status = xs_sendpages(transport->sock,
 637                               xs_addr(xprt),
 638                               xprt->addrlen, xdr,
 639                               req->rq_bytes_sent, true);
 640
 641         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 642                         xdr->len - req->rq_bytes_sent, status);
 643
 644         if (status >= 0) {
 645                 req->rq_xmit_bytes_sent += status;
 646                 if (status >= req->rq_slen)
 647                         return 0;
 648                 /* Still some bytes left; set up for a retry later. */
 649                 status = -EAGAIN;
 650         }
 651
 652         switch (status) {
 653         case -ENOTSOCK:
 654                 status = -ENOTCONN;
 655                 /* Should we call xs_close() here? */
 656                 break;
 657         case -EAGAIN:
 658                 status = xs_nospace(task);
 659                 break;
 660         default:
 661                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 662                         -status);
 663         case -ENETUNREACH:
 664         case -EPIPE:
 665         case -ECONNREFUSED:
 666                 /* When the server has died, an ICMP port unreachable message
 667                  * prompts ECONNREFUSED. */
 668                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 669         }
 670
 671         return status;
 672 }
 673
 674 /**
 675  * xs_tcp_shutdown - gracefully shut down a TCP socket
 676  * @xprt: transport
 677  *
 678  * Initiates a graceful shutdown of the TCP socket by calling the
 679  * equivalent of shutdown(SHUT_WR);
 680  */
 681 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
 682 {
 683         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 684         struct socket *sock = transport->sock;
 685
 686         if (sock != NULL) {
 687                 kernel_sock_shutdown(sock, SHUT_WR);
 688                 trace_rpc_socket_shutdown(xprt, sock);
 689         }
 690 }
 691
 692 /**
 693  * xs_tcp_send_request - write an RPC request to a TCP socket
 694  * @task: address of RPC task that manages the state of an RPC request
 695  *
 696  * Return values:
 697  *        0:    The request has been sent
 698  *   EAGAIN:    The socket was blocked, please call again later to
 699  *              complete the request
 700  * ENOTCONN:    Caller needs to invoke connect logic then call again
 701  *    other:    Some other error occurred, the request was not sent
 702  *
 703  * XXX: In the case of soft timeouts, should we eventually give up
 704  *      if sendmsg is not able to make progress?
 705  */
 706 static int xs_tcp_send_request(struct rpc_task *task)
 707 {
 708         struct rpc_rqst *req = task->tk_rqstp;
 709         struct rpc_xprt *xprt = req->rq_xprt;
 710         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 711         struct xdr_buf *xdr = &req->rq_snd_buf;
 712         bool zerocopy = true;
 713         int status;
 714
 715         xs_encode_stream_record_marker(&req->rq_snd_buf);
 716
 717         xs_pktdump("packet data:",
 718                                 req->rq_svec->iov_base,
 719                                 req->rq_svec->iov_len);
 720         /* Don't use zero copy if this is a resend. If the RPC call
 721          * completes while the socket holds a reference to the pages,
 722          * then we may end up resending corrupted data.
 723          */
 724         if (task->tk_flags & RPC_TASK_SENT)
 725                 zerocopy = false;
 726
 727         /* Continue transmitting the packet/record. We must be careful
 728          * to cope with writespace callbacks arriving _after_ we have
 729          * called sendmsg(). */
 730         while (1) {
 731                 status = xs_sendpages(transport->sock,
 732                                         NULL, 0, xdr, req->rq_bytes_sent,
 733                                         zerocopy);
 734
 735                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 736                                 xdr->len - req->rq_bytes_sent, status);
 737
 738                 if (unlikely(status < 0))
 739                         break;
 740
 741                 /* If we've sent the entire packet, immediately
 742                  * reset the count of bytes sent. */
 743                 req->rq_bytes_sent += status;
 744                 req->rq_xmit_bytes_sent += status;
 745                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 746                         req->rq_bytes_sent = 0;
 747                         return 0;
 748                 }
 749
 750                 if (status != 0)
 751                         continue;
 752                 status = -EAGAIN;
 753                 break;
 754         }
 755
 756         switch (status) {
 757         case -ENOTSOCK:
 758                 status = -ENOTCONN;
 759                 /* Should we call xs_close() here? */
 760                 break;
 761         case -EAGAIN:
 762                 status = xs_nospace(task);
 763                 break;
 764         default:
 765                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 766                         -status);
 767         case -ECONNRESET:
 768                 xs_tcp_shutdown(xprt);
 769         case -ECONNREFUSED:
 770         case -ENOTCONN:
 771         case -EPIPE:
 772                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 773         }
 774
 775         return status;
 776 }
 777
 778 /**
 779  * xs_tcp_release_xprt - clean up after a tcp transmission
 780  * @xprt: transport
 781  * @task: rpc task
 782  *
 783  * This cleans up if an error causes us to abort the transmission of a request.
 784  * In this case, the socket may need to be reset in order to avoid confusing
 785  * the server.
 786  */
 787 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 788 {
 789         struct rpc_rqst *req;
 790
 791         if (task != xprt->snd_task)
 792                 return;
 793         if (task == NULL)
 794                 goto out_release;
 795         req = task->tk_rqstp;
 796         if (req == NULL)
 797                 goto out_release;
 798         if (req->rq_bytes_sent == 0)
 799                 goto out_release;
 800         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 801                 goto out_release;
 802         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 803 out_release:
 804         xprt_release_xprt(xprt, task);
 805 }
 806
 807 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 808 {
 809         transport->old_data_ready = sk->sk_data_ready;
 810         transport->old_state_change = sk->sk_state_change;
 811         transport->old_write_space = sk->sk_write_space;
 812         transport->old_error_report = sk->sk_error_report;
 813 }
 814
 815 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 816 {
 817         sk->sk_data_ready = transport->old_data_ready;
 818         sk->sk_state_change = transport->old_state_change;
 819         sk->sk_write_space = transport->old_write_space;
 820         sk->sk_error_report = transport->old_error_report;
 821 }
 822
 823 /**
 824  * xs_error_report - callback to handle TCP socket state errors
 825  * @sk: socket
 826  *
 827  * Note: we don't call sock_error() since there may be a rpc_task
 828  * using the socket, and so we don't want to clear sk->sk_err.
 829  */
 830 static void xs_error_report(struct sock *sk)
 831 {
 832         struct rpc_xprt *xprt;
 833         int err;
 834
 835         read_lock_bh(&sk->sk_callback_lock);
 836         if (!(xprt = xprt_from_sock(sk)))
 837                 goto out;
 838
 839         err = -sk->sk_err;
 840         if (err == 0)
 841                 goto out;
 842         dprintk("RPC:       xs_error_report client %p, error=%d...\n",
 843                         xprt, -err);
 844         trace_rpc_socket_error(xprt, sk->sk_socket, err);
 845         xprt_wake_pending_tasks(xprt, err);
 846  out:
 847         read_unlock_bh(&sk->sk_callback_lock);
 848 }
 849
 850 static void xs_reset_transport(struct sock_xprt *transport)
 851 {
 852         struct socket *sock = transport->sock;
 853         struct sock *sk = transport->inet;
 854
 855         if (sk == NULL)
 856                 return;
 857
 858         transport->srcport = 0;
 859
 860         write_lock_bh(&sk->sk_callback_lock);
 861         transport->inet = NULL;
 862         transport->sock = NULL;
 863
 864         sk->sk_user_data = NULL;
 865
 866         xs_restore_old_callbacks(transport, sk);
 867         write_unlock_bh(&sk->sk_callback_lock);
 868
 869         sk->sk_no_check = 0;
 870
 871         trace_rpc_socket_close(&transport->xprt, sock);
 872         sock_release(sock);
 873 }
 874
 875 /**
 876  * xs_close - close a socket
 877  * @xprt: transport
 878  *
 879  * This is used when all requests are complete; ie, no DRC state remains
 880  * on the server we want to save.
 881  *
 882  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
 883  * xs_reset_transport() zeroing the socket from underneath a writer.
 884  */
 885 static void xs_close(struct rpc_xprt *xprt)
 886 {
 887         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 888
 889         dprintk("RPC:       xs_close xprt %p\n", xprt);
 890
 891         cancel_delayed_work_sync(&transport->connect_worker);
 892
 893         xs_reset_transport(transport);
 894         xprt->reestablish_timeout = 0;
 895
 896         smp_mb__before_clear_bit();
 897         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
 898         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 899         clear_bit(XPRT_CLOSING, &xprt->state);
 900         smp_mb__after_clear_bit();
 901         xprt_disconnect_done(xprt);
 902 }
 903
 904 static void xs_tcp_close(struct rpc_xprt *xprt)
 905 {
 906         if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
 907                 xs_close(xprt);
 908         else
 909                 xs_tcp_shutdown(xprt);
 910 }
 911
 912 /**
 913  * xs_destroy - prepare to shutdown a transport
 914  * @xprt: doomed transport
 915  *
 916  */
 917 static void xs_destroy(struct rpc_xprt *xprt)
 918 {
 919         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 920
 921         xs_close(xprt);
 922         xs_free_peer_addresses(xprt);
 923         xprt_free(xprt);
 924         module_put(THIS_MODULE);
 925 }
 926
 927 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
 928 {
 929         struct xdr_skb_reader desc = {
 930                 .skb            = skb,
 931                 .offset         = sizeof(rpc_fraghdr),
 932                 .count          = skb->len - sizeof(rpc_fraghdr),
 933         };
 934
 935         if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
 936                 return -1;
 937         if (desc.count)
 938                 return -1;
 939         return 0;
 940 }
 941
 942 /**
 943  * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
 944  * @sk: socket with data to read
 945  * @len: how much data to read
 946  *
 947  * Currently this assumes we can read the whole reply in a single gulp.
 948  */
 949 static void xs_local_data_ready(struct sock *sk, int len)
 950 {
 951         struct rpc_task *task;
 952         struct rpc_xprt *xprt;
 953         struct rpc_rqst *rovr;
 954         struct sk_buff *skb;
 955         int err, repsize, copied;
 956         u32 _xid;
 957         __be32 *xp;
 958
 959         read_lock_bh(&sk->sk_callback_lock);
 960         dprintk("RPC:       %s...\n", __func__);
 961         xprt = xprt_from_sock(sk);
 962         if (xprt == NULL)
 963                 goto out;
 964
 965         skb = skb_recv_datagram(sk, 0, 1, &err);
 966         if (skb == NULL)
 967                 goto out;
 968
 969         repsize = skb->len - sizeof(rpc_fraghdr);
 970         if (repsize < 4) {
 971                 dprintk("RPC:       impossible RPC reply size %d\n", repsize);
 972                 goto dropit;
 973         }
 974
 975         /* Copy the XID from the skb... */
 976         xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
 977         if (xp == NULL)
 978                 goto dropit;
 979
 980         /* Look up and lock the request corresponding to the given XID */
 981         spin_lock(&xprt->transport_lock);
 982         rovr = xprt_lookup_rqst(xprt, *xp);
 983         if (!rovr)
 984                 goto out_unlock;
 985         task = rovr->rq_task;
 986
 987         copied = rovr->rq_private_buf.buflen;
 988         if (copied > repsize)
 989                 copied = repsize;
 990
 991         if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 992                 dprintk("RPC:       sk_buff copy failed\n");
 993                 goto out_unlock;
 994         }
 995
 996         xprt_complete_rqst(task, copied);
 997
 998  out_unlock:
 999         spin_unlock(&xprt->transport_lock);
1000  dropit:
1001         skb_free_datagram(sk, skb);
1002  out:
1003         read_unlock_bh(&sk->sk_callback_lock);
1004 }
1005
1006 /**
1007  * xs_udp_data_ready - "data ready" callback for UDP sockets
1008  * @sk: socket with data to read
1009  * @len: how much data to read
1010  *
1011  */
1012 static void xs_udp_data_ready(struct sock *sk, int len)
1013 {
1014         struct rpc_task *task;
1015         struct rpc_xprt *xprt;
1016         struct rpc_rqst *rovr;
1017         struct sk_buff *skb;
1018         int err, repsize, copied;
1019         u32 _xid;
1020         __be32 *xp;
1021
1022         read_lock_bh(&sk->sk_callback_lock);
1023         dprintk("RPC:       xs_udp_data_ready...\n");
1024         if (!(xprt = xprt_from_sock(sk)))
1025                 goto out;
1026
1027         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
1028                 goto out;
1029
1030         repsize = skb->len - sizeof(struct udphdr);
1031         if (repsize < 4) {
1032                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
1033                 goto dropit;
1034         }
1035
1036         /* Copy the XID from the skb... */
1037         xp = skb_header_pointer(skb, sizeof(struct udphdr),
1038                                 sizeof(_xid), &_xid);
1039         if (xp == NULL)
1040                 goto dropit;
1041
1042         /* Look up and lock the request corresponding to the given XID */
1043         spin_lock(&xprt->transport_lock);
1044         rovr = xprt_lookup_rqst(xprt, *xp);
1045         if (!rovr)
1046                 goto out_unlock;
1047         task = rovr->rq_task;
1048
1049         if ((copied = rovr->rq_private_buf.buflen) > repsize)
1050                 copied = repsize;
1051
1052         /* Suck it into the iovec, verify checksum if not done by hw. */
1053         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1054                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1055                 goto out_unlock;
1056         }
1057
1058         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1059
1060         xprt_adjust_cwnd(xprt, task, copied);
1061         xprt_complete_rqst(task, copied);
1062
1063  out_unlock:
1064         spin_unlock(&xprt->transport_lock);
1065  dropit:
1066         skb_free_datagram(sk, skb);
1067  out:
1068         read_unlock_bh(&sk->sk_callback_lock);
1069 }
1070
1071 /*
1072  * Helper function to force a TCP close if the server is sending
1073  * junk and/or it has put us in CLOSE_WAIT
1074  */
1075 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1076 {
1077         set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1078         xprt_force_disconnect(xprt);
1079 }
1080
1081 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1082 {
1083         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1084         size_t len, used;
1085         char *p;
1086
1087         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1088         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1089         used = xdr_skb_read_bits(desc, p, len);
1090         transport->tcp_offset += used;
1091         if (used != len)
1092                 return;
1093
1094         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1095         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1096                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1097         else
1098                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1099         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1100
1101         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1102         transport->tcp_offset = 0;
1103
1104         /* Sanity check of the record length */
1105         if (unlikely(transport->tcp_reclen < 8)) {
1106                 dprintk("RPC:       invalid TCP record fragment length\n");
1107                 xs_tcp_force_close(xprt);
1108                 return;
1109         }
1110         dprintk("RPC:       reading TCP record fragment of length %d\n",
1111                         transport->tcp_reclen);
1112 }
1113
1114 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1115 {
1116         if (transport->tcp_offset == transport->tcp_reclen) {
1117                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1118                 transport->tcp_offset = 0;
1119                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1120                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1121                         transport->tcp_flags |= TCP_RCV_COPY_XID;
1122                         transport->tcp_copied = 0;
1123                 }
1124         }
1125 }
1126
1127 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1128 {
1129         size_t len, used;
1130         char *p;
1131
1132         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1133         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
1134         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1135         used = xdr_skb_read_bits(desc, p, len);
1136         transport->tcp_offset += used;
1137         if (used != len)
1138                 return;
1139         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1140         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1141         transport->tcp_copied = 4;
1142         dprintk("RPC:       reading %s XID %08x\n",
1143                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1144                                                               : "request with",
1145                         ntohl(transport->tcp_xid));
1146         xs_tcp_check_fraghdr(transport);
1147 }
1148
1149 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1150                                        struct xdr_skb_reader *desc)
1151 {
1152         size_t len, used;
1153         u32 offset;
1154         char *p;
1155
1156         /*
1157          * We want transport->tcp_offset to be 8 at the end of this routine
1158          * (4 bytes for the xid and 4 bytes for the call/reply flag).
1159          * When this function is called for the first time,
1160          * transport->tcp_offset is 4 (after having already read the xid).
1161          */
1162         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1163         len = sizeof(transport->tcp_calldir) - offset;
1164         dprintk("RPC:       reading CALL/REPLY flag (%Zu bytes)\n", len);
1165         p = ((char *) &transport->tcp_calldir) + offset;
1166         used = xdr_skb_read_bits(desc, p, len);
1167         transport->tcp_offset += used;
1168         if (used != len)
1169                 return;
1170         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1171         /*
1172          * We don't yet have the XDR buffer, so we will write the calldir
1173          * out after we get the buffer from the 'struct rpc_rqst'
1174          */
1175         switch (ntohl(transport->tcp_calldir)) {
1176         case RPC_REPLY:
1177                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1178                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1179                 transport->tcp_flags |= TCP_RPC_REPLY;
1180                 break;
1181         case RPC_CALL:
1182                 transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1183                 transport->tcp_flags |= TCP_RCV_COPY_DATA;
1184                 transport->tcp_flags &= ~TCP_RPC_REPLY;
1185                 break;
1186         default:
1187                 dprintk("RPC:       invalid request message type\n");
1188                 xs_tcp_force_close(&transport->xprt);
1189         }
1190         xs_tcp_check_fraghdr(transport);
1191 }
1192
1193 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1194                                      struct xdr_skb_reader *desc,
1195                                      struct rpc_rqst *req)
1196 {
1197         struct sock_xprt *transport =
1198                                 container_of(xprt, struct sock_xprt, xprt);
1199         struct xdr_buf *rcvbuf;
1200         size_t len;
1201         ssize_t r;
1202
1203         rcvbuf = &req->rq_private_buf;
1204
1205         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1206                 /*
1207                  * Save the RPC direction in the XDR buffer
1208                  */
1209                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1210                         &transport->tcp_calldir,
1211                         sizeof(transport->tcp_calldir));
1212                 transport->tcp_copied += sizeof(transport->tcp_calldir);
1213                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1214         }
1215
1216         len = desc->count;
1217         if (len > transport->tcp_reclen - transport->tcp_offset) {
1218                 struct xdr_skb_reader my_desc;
1219
1220                 len = transport->tcp_reclen - transport->tcp_offset;
1221                 memcpy(&my_desc, desc, sizeof(my_desc));
1222                 my_desc.count = len;
1223                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1224                                           &my_desc, xdr_skb_read_bits);
1225                 desc->count -= r;
1226                 desc->offset += r;
1227         } else
1228                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1229                                           desc, xdr_skb_read_bits);
1230
1231         if (r > 0) {
1232                 transport->tcp_copied += r;
1233                 transport->tcp_offset += r;
1234         }
1235         if (r != len) {
1236                 /* Error when copying to the receive buffer,
1237                  * usually because we weren't able to allocate
1238                  * additional buffer pages. All we can do now
1239                  * is turn off TCP_RCV_COPY_DATA, so the request
1240                  * will not receive any additional updates,
1241                  * and time out.
1242                  * Any remaining data from this record will
1243                  * be discarded.
1244                  */
1245                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1246                 dprintk("RPC:       XID %08x truncated request\n",
1247                                 ntohl(transport->tcp_xid));
1248                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1249                                 "tcp_offset = %u, tcp_reclen = %u\n",
1250                                 xprt, transport->tcp_copied,
1251                                 transport->tcp_offset, transport->tcp_reclen);
1252                 return;
1253         }
1254
1255         dprintk("RPC:       XID %08x read %Zd bytes\n",
1256                         ntohl(transport->tcp_xid), r);
1257         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1258                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1259                         transport->tcp_offset, transport->tcp_reclen);
1260
1261         if (transport->tcp_copied == req->rq_private_buf.buflen)
1262                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1263         else if (transport->tcp_offset == transport->tcp_reclen) {
1264                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1265                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1266         }
1267 }
1268
1269 /*
1270  * Finds the request corresponding to the RPC xid and invokes the common
1271  * tcp read code to read the data.
1272  */
1273 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1274                                     struct xdr_skb_reader *desc)
1275 {
1276         struct sock_xprt *transport =
1277                                 container_of(xprt, struct sock_xprt, xprt);
1278         struct rpc_rqst *req;
1279
1280         dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1281
1282         /* Find and lock the request corresponding to this xid */
1283         spin_lock(&xprt->transport_lock);
1284         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1285         if (!req) {
1286                 dprintk("RPC:       XID %08x request not found!\n",
1287                                 ntohl(transport->tcp_xid));
1288                 spin_unlock(&xprt->transport_lock);
1289                 return -1;
1290         }
1291
1292         xs_tcp_read_common(xprt, desc, req);
1293
1294         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1295                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1296
1297         spin_unlock(&xprt->transport_lock);
1298         return 0;
1299 }
1300
1301 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1302 /*
1303  * Obtains an rpc_rqst previously allocated and invokes the common
1304  * tcp read code to read the data.  The result is placed in the callback
1305  * queue.
1306  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1307  * connection and return -1.
1308  */
1309 static inline int xs_tcp_read_callback(struct rpc_xprt *xprt,
1310                                        struct xdr_skb_reader *desc)
1311 {
1312         struct sock_xprt *transport =
1313                                 container_of(xprt, struct sock_xprt, xprt);
1314         struct rpc_rqst *req;
1315
1316         req = xprt_alloc_bc_request(xprt);
1317         if (req == NULL) {
1318                 printk(KERN_WARNING "Callback slot table overflowed\n");
1319                 xprt_force_disconnect(xprt);
1320                 return -1;
1321         }
1322
1323         req->rq_xid = transport->tcp_xid;
1324         dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1325         xs_tcp_read_common(xprt, desc, req);
1326
1327         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) {
1328                 struct svc_serv *bc_serv = xprt->bc_serv;
1329
1330                 /*
1331                  * Add callback request to callback list.  The callback
1332                  * service sleeps on the sv_cb_waitq waiting for new
1333                  * requests.  Wake it up after adding enqueing the
1334                  * request.
1335                  */
1336                 dprintk("RPC:       add callback request to list\n");
1337                 spin_lock(&bc_serv->sv_cb_lock);
1338                 list_add(&req->rq_bc_list, &bc_serv->sv_cb_list);
1339                 spin_unlock(&bc_serv->sv_cb_lock);
1340                 wake_up(&bc_serv->sv_cb_waitq);
1341         }
1342
1343         req->rq_private_buf.len = transport->tcp_copied;
1344
1345         return 0;
1346 }
1347
1348 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1349                                         struct xdr_skb_reader *desc)
1350 {
1351         struct sock_xprt *transport =
1352                                 container_of(xprt, struct sock_xprt, xprt);
1353
1354         return (transport->tcp_flags & TCP_RPC_REPLY) ?
1355                 xs_tcp_read_reply(xprt, desc) :
1356                 xs_tcp_read_callback(xprt, desc);
1357 }
1358 #else
1359 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1360                                         struct xdr_skb_reader *desc)
1361 {
1362         return xs_tcp_read_reply(xprt, desc);
1363 }
1364 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1365
1366 /*
1367  * Read data off the transport.  This can be either an RPC_CALL or an
1368  * RPC_REPLY.  Relay the processing to helper functions.
1369  */
1370 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1371                                     struct xdr_skb_reader *desc)
1372 {
1373         struct sock_xprt *transport =
1374                                 container_of(xprt, struct sock_xprt, xprt);
1375
1376         if (_xs_tcp_read_data(xprt, desc) == 0)
1377                 xs_tcp_check_fraghdr(transport);
1378         else {
1379                 /*
1380                  * The transport_lock protects the request handling.
1381                  * There's no need to hold it to update the tcp_flags.
1382                  */
1383                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1384         }
1385 }
1386
1387 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1388 {
1389         size_t len;
1390
1391         len = transport->tcp_reclen - transport->tcp_offset;
1392         if (len > desc->count)
1393                 len = desc->count;
1394         desc->count -= len;
1395         desc->offset += len;
1396         transport->tcp_offset += len;
1397         dprintk("RPC:       discarded %Zu bytes\n", len);
1398         xs_tcp_check_fraghdr(transport);
1399 }
1400
1401 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1402 {
1403         struct rpc_xprt *xprt = rd_desc->arg.data;
1404         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1405         struct xdr_skb_reader desc = {
1406                 .skb    = skb,
1407                 .offset = offset,
1408                 .count  = len,
1409         };
1410
1411         dprintk("RPC:       xs_tcp_data_recv started\n");
1412         do {
1413                 /* Read in a new fragment marker if necessary */
1414                 /* Can we ever really expect to get completely empty fragments? */
1415                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1416                         xs_tcp_read_fraghdr(xprt, &desc);
1417                         continue;
1418                 }
1419                 /* Read in the xid if necessary */
1420                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1421                         xs_tcp_read_xid(transport, &desc);
1422                         continue;
1423                 }
1424                 /* Read in the call/reply flag */
1425                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1426                         xs_tcp_read_calldir(transport, &desc);
1427                         continue;
1428                 }
1429                 /* Read in the request data */
1430                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1431                         xs_tcp_read_data(xprt, &desc);
1432                         continue;
1433                 }
1434                 /* Skip over any trailing bytes on short reads */
1435                 xs_tcp_read_discard(transport, &desc);
1436         } while (desc.count);
1437         dprintk("RPC:       xs_tcp_data_recv done\n");
1438         return len - desc.count;
1439 }
1440
1441 /**
1442  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1443  * @sk: socket with data to read
1444  * @bytes: how much data to read
1445  *
1446  */
1447 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1448 {
1449         struct rpc_xprt *xprt;
1450         read_descriptor_t rd_desc;
1451         int read;
1452
1453         dprintk("RPC:       xs_tcp_data_ready...\n");
1454
1455         read_lock_bh(&sk->sk_callback_lock);
1456         if (!(xprt = xprt_from_sock(sk)))
1457                 goto out;
1458         /* Any data means we had a useful conversation, so
1459          * the we don't need to delay the next reconnect
1460          */
1461         if (xprt->reestablish_timeout)
1462                 xprt->reestablish_timeout = 0;
1463
1464         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1465         rd_desc.arg.data = xprt;
1466         do {
1467                 rd_desc.count = 65536;
1468                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1469         } while (read > 0);
1470 out:
1471         read_unlock_bh(&sk->sk_callback_lock);
1472 }
1473
1474 /*
1475  * Do the equivalent of linger/linger2 handling for dealing with
1476  * broken servers that don't close the socket in a timely
1477  * fashion
1478  */
1479 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1480                 unsigned long timeout)
1481 {
1482         struct sock_xprt *transport;
1483
1484         if (xprt_test_and_set_connecting(xprt))
1485                 return;
1486         set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1487         transport = container_of(xprt, struct sock_xprt, xprt);
1488         queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1489                            timeout);
1490 }
1491
1492 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1493 {
1494         struct sock_xprt *transport;
1495
1496         transport = container_of(xprt, struct sock_xprt, xprt);
1497
1498         if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1499             !cancel_delayed_work(&transport->connect_worker))
1500                 return;
1501         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1502         xprt_clear_connecting(xprt);
1503 }
1504
1505 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1506 {
1507         smp_mb__before_clear_bit();
1508         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1509         clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1510         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1511         clear_bit(XPRT_CLOSING, &xprt->state);
1512         smp_mb__after_clear_bit();
1513 }
1514
1515 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1516 {
1517         xs_sock_reset_connection_flags(xprt);
1518         /* Mark transport as closed and wake up all pending tasks */
1519         xprt_disconnect_done(xprt);
1520 }
1521
1522 /**
1523  * xs_tcp_state_change - callback to handle TCP socket state changes
1524  * @sk: socket whose state has changed
1525  *
1526  */
1527 static void xs_tcp_state_change(struct sock *sk)
1528 {
1529         struct rpc_xprt *xprt;
1530
1531         read_lock_bh(&sk->sk_callback_lock);
1532         if (!(xprt = xprt_from_sock(sk)))
1533                 goto out;
1534         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1535         dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1536                         sk->sk_state, xprt_connected(xprt),
1537                         sock_flag(sk, SOCK_DEAD),
1538                         sock_flag(sk, SOCK_ZAPPED),
1539                         sk->sk_shutdown);
1540
1541         trace_rpc_socket_state_change(xprt, sk->sk_socket);
1542         switch (sk->sk_state) {
1543         case TCP_ESTABLISHED:
1544                 spin_lock(&xprt->transport_lock);
1545                 if (!xprt_test_and_set_connected(xprt)) {
1546                         struct sock_xprt *transport = container_of(xprt,
1547                                         struct sock_xprt, xprt);
1548
1549                         /* Reset TCP record info */
1550                         transport->tcp_offset = 0;
1551                         transport->tcp_reclen = 0;
1552                         transport->tcp_copied = 0;
1553                         transport->tcp_flags =
1554                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1555                         xprt->connect_cookie++;
1556
1557                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1558                 }
1559                 spin_unlock(&xprt->transport_lock);
1560                 break;
1561         case TCP_FIN_WAIT1:
1562                 /* The client initiated a shutdown of the socket */
1563                 xprt->connect_cookie++;
1564                 xprt->reestablish_timeout = 0;
1565                 set_bit(XPRT_CLOSING, &xprt->state);
1566                 smp_mb__before_clear_bit();
1567                 clear_bit(XPRT_CONNECTED, &xprt->state);
1568                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1569                 smp_mb__after_clear_bit();
1570                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1571                 break;
1572         case TCP_CLOSE_WAIT:
1573                 /* The server initiated a shutdown of the socket */
1574                 xprt->connect_cookie++;
1575                 clear_bit(XPRT_CONNECTED, &xprt->state);
1576                 xs_tcp_force_close(xprt);
1577         case TCP_CLOSING:
1578                 /*
1579                  * If the server closed down the connection, make sure that
1580                  * we back off before reconnecting
1581                  */
1582                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1583                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1584                 break;
1585         case TCP_LAST_ACK:
1586                 set_bit(XPRT_CLOSING, &xprt->state);
1587                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1588                 smp_mb__before_clear_bit();
1589                 clear_bit(XPRT_CONNECTED, &xprt->state);
1590                 smp_mb__after_clear_bit();
1591                 break;
1592         case TCP_CLOSE:
1593                 xs_tcp_cancel_linger_timeout(xprt);
1594                 xs_sock_mark_closed(xprt);
1595         }
1596  out:
1597         read_unlock_bh(&sk->sk_callback_lock);
1598 }
1599
1600 static void xs_write_space(struct sock *sk)
1601 {
1602         struct socket *sock;
1603         struct rpc_xprt *xprt;
1604
1605         if (unlikely(!(sock = sk->sk_socket)))
1606                 return;
1607         clear_bit(SOCK_NOSPACE, &sock->flags);
1608
1609         if (unlikely(!(xprt = xprt_from_sock(sk))))
1610                 return;
1611         if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1612                 return;
1613
1614         xprt_write_space(xprt);
1615 }
1616
1617 /**
1618  * xs_udp_write_space - callback invoked when socket buffer space
1619  *                             becomes available
1620  * @sk: socket whose state has changed
1621  *
1622  * Called when more output buffer space is available for this socket.
1623  * We try not to wake our writers until they can make "significant"
1624  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1625  * with a bunch of small requests.
1626  */
1627 static void xs_udp_write_space(struct sock *sk)
1628 {
1629         read_lock_bh(&sk->sk_callback_lock);
1630
1631         /* from net/core/sock.c:sock_def_write_space */
1632         if (sock_writeable(sk))
1633                 xs_write_space(sk);
1634
1635         read_unlock_bh(&sk->sk_callback_lock);
1636 }
1637
1638 /**
1639  * xs_tcp_write_space - callback invoked when socket buffer space
1640  *                             becomes available
1641  * @sk: socket whose state has changed
1642  *
1643  * Called when more output buffer space is available for this socket.
1644  * We try not to wake our writers until they can make "significant"
1645  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1646  * with a bunch of small requests.
1647  */
1648 static void xs_tcp_write_space(struct sock *sk)
1649 {
1650         read_lock_bh(&sk->sk_callback_lock);
1651
1652         /* from net/core/stream.c:sk_stream_write_space */
1653         if (sk_stream_is_writeable(sk))
1654                 xs_write_space(sk);
1655
1656         read_unlock_bh(&sk->sk_callback_lock);
1657 }
1658
1659 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1660 {
1661         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1662         struct sock *sk = transport->inet;
1663
1664         if (transport->rcvsize) {
1665                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1666                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1667         }
1668         if (transport->sndsize) {
1669                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1670                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1671                 sk->sk_write_space(sk);
1672         }
1673 }
1674
1675 /**
1676  * xs_udp_set_buffer_size - set send and receive limits
1677  * @xprt: generic transport
1678  * @sndsize: requested size of send buffer, in bytes
1679  * @rcvsize: requested size of receive buffer, in bytes
1680  *
1681  * Set socket send and receive buffer size limits.
1682  */
1683 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1684 {
1685         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1686
1687         transport->sndsize = 0;
1688         if (sndsize)
1689                 transport->sndsize = sndsize + 1024;
1690         transport->rcvsize = 0;
1691         if (rcvsize)
1692                 transport->rcvsize = rcvsize + 1024;
1693
1694         xs_udp_do_set_buffer_size(xprt);
1695 }
1696
1697 /**
1698  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1699  * @task: task that timed out
1700  *
1701  * Adjust the congestion window after a retransmit timeout has occurred.
1702  */
1703 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1704 {
1705         xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1706 }
1707
1708 static unsigned short xs_get_random_port(void)
1709 {
1710         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1711         unsigned short rand = (unsigned short) prandom_u32() % range;
1712         return rand + xprt_min_resvport;
1713 }
1714
1715 /**
1716  * xs_set_port - reset the port number in the remote endpoint address
1717  * @xprt: generic transport
1718  * @port: new port number
1719  *
1720  */
1721 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1722 {
1723         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1724
1725         rpc_set_port(xs_addr(xprt), port);
1726         xs_update_peer_port(xprt);
1727 }
1728
1729 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1730 {
1731         unsigned short port = transport->srcport;
1732
1733         if (port == 0 && transport->xprt.resvport)
1734                 port = xs_get_random_port();
1735         return port;
1736 }
1737
1738 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1739 {
1740         if (transport->srcport != 0)
1741                 transport->srcport = 0;
1742         if (!transport->xprt.resvport)
1743                 return 0;
1744         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1745                 return xprt_max_resvport;
1746         return --port;
1747 }
1748 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1749 {
1750         struct sockaddr_storage myaddr;
1751         int err, nloop = 0;
1752         unsigned short port = xs_get_srcport(transport);
1753         unsigned short last;
1754
1755         memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1756         do {
1757                 rpc_set_port((struct sockaddr *)&myaddr, port);
1758                 err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1759                                 transport->xprt.addrlen);
1760                 if (port == 0)
1761                         break;
1762                 if (err == 0) {
1763                         transport->srcport = port;
1764                         break;
1765                 }
1766                 last = port;
1767                 port = xs_next_srcport(transport, port);
1768                 if (port > last)
1769                         nloop++;
1770         } while (err == -EADDRINUSE && nloop != 2);
1771
1772         if (myaddr.ss_family == AF_INET)
1773                 dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1774                                 &((struct sockaddr_in *)&myaddr)->sin_addr,
1775                                 port, err ? "failed" : "ok", err);
1776         else
1777                 dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1778                                 &((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1779                                 port, err ? "failed" : "ok", err);
1780         return err;
1781 }
1782
1783 /*
1784  * We don't support autobind on AF_LOCAL sockets
1785  */
1786 static void xs_local_rpcbind(struct rpc_task *task)
1787 {
1788         rcu_read_lock();
1789         xprt_set_bound(rcu_dereference(task->tk_client->cl_xprt));
1790         rcu_read_unlock();
1791 }
1792
1793 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1794 {
1795 }
1796
1797 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1798 static struct lock_class_key xs_key[2];
1799 static struct lock_class_key xs_slock_key[2];
1800
1801 static inline void xs_reclassify_socketu(struct socket *sock)
1802 {
1803         struct sock *sk = sock->sk;
1804
1805         sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1806                 &xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1807 }
1808
1809 static inline void xs_reclassify_socket4(struct socket *sock)
1810 {
1811         struct sock *sk = sock->sk;
1812
1813         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1814                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1815 }
1816
1817 static inline void xs_reclassify_socket6(struct socket *sock)
1818 {
1819         struct sock *sk = sock->sk;
1820
1821         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1822                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1823 }
1824
1825 static inline void xs_reclassify_socket(int family, struct socket *sock)
1826 {
1827         WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1828         if (sock_owned_by_user(sock->sk))
1829                 return;
1830
1831         switch (family) {
1832         case AF_LOCAL:
1833                 xs_reclassify_socketu(sock);
1834                 break;
1835         case AF_INET:
1836                 xs_reclassify_socket4(sock);
1837                 break;
1838         case AF_INET6:
1839                 xs_reclassify_socket6(sock);
1840                 break;
1841         }
1842 }
1843 #else
1844 static inline void xs_reclassify_socketu(struct socket *sock)
1845 {
1846 }
1847
1848 static inline void xs_reclassify_socket4(struct socket *sock)
1849 {
1850 }
1851
1852 static inline void xs_reclassify_socket6(struct socket *sock)
1853 {
1854 }
1855
1856 static inline void xs_reclassify_socket(int family, struct socket *sock)
1857 {
1858 }
1859 #endif
1860
1861 static void xs_dummy_setup_socket(struct work_struct *work)
1862 {
1863 }
1864
1865 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1866                 struct sock_xprt *transport, int family, int type, int protocol)
1867 {
1868         struct socket *sock;
1869         int err;
1870
1871         err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1872         if (err < 0) {
1873                 dprintk("RPC:       can't create %d transport socket (%d).\n",
1874                                 protocol, -err);
1875                 goto out;
1876         }
1877         xs_reclassify_socket(family, sock);
1878
1879         err = xs_bind(transport, sock);
1880         if (err) {
1881                 sock_release(sock);
1882                 goto out;
1883         }
1884
1885         return sock;
1886 out:
1887         return ERR_PTR(err);
1888 }
1889
1890 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1891                                       struct socket *sock)
1892 {
1893         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1894                                                                         xprt);
1895
1896         if (!transport->inet) {
1897                 struct sock *sk = sock->sk;
1898
1899                 write_lock_bh(&sk->sk_callback_lock);
1900
1901                 xs_save_old_callbacks(transport, sk);
1902
1903                 sk->sk_user_data = xprt;
1904                 sk->sk_data_ready = xs_local_data_ready;
1905                 sk->sk_write_space = xs_udp_write_space;
1906                 sk->sk_error_report = xs_error_report;
1907                 sk->sk_allocation = GFP_ATOMIC;
1908
1909                 xprt_clear_connected(xprt);
1910
1911                 /* Reset to new socket */
1912                 transport->sock = sock;
1913                 transport->inet = sk;
1914
1915                 write_unlock_bh(&sk->sk_callback_lock);
1916         }
1917
1918         /* Tell the socket layer to start connecting... */
1919         xprt->stat.connect_count++;
1920         xprt->stat.connect_start = jiffies;
1921         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1922 }
1923
1924 /**
1925  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1926  * @xprt: RPC transport to connect
1927  * @transport: socket transport to connect
1928  * @create_sock: function to create a socket of the correct type
1929  */
1930 static int xs_local_setup_socket(struct sock_xprt *transport)
1931 {
1932         struct rpc_xprt *xprt = &transport->xprt;
1933         struct socket *sock;
1934         int status = -EIO;
1935
1936         current->flags |= PF_FSTRANS;
1937
1938         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1939         status = __sock_create(xprt->xprt_net, AF_LOCAL,
1940                                         SOCK_STREAM, 0, &sock, 1);
1941         if (status < 0) {
1942                 dprintk("RPC:       can't create AF_LOCAL "
1943                         "transport socket (%d).\n", -status);
1944                 goto out;
1945         }
1946         xs_reclassify_socketu(sock);
1947
1948         dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1949                         xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1950
1951         status = xs_local_finish_connecting(xprt, sock);
1952         trace_rpc_socket_connect(xprt, sock, status);
1953         switch (status) {
1954         case 0:
1955                 dprintk("RPC:       xprt %p connected to %s\n",
1956                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1957                 xprt_set_connected(xprt);
1958                 break;
1959         case -ENOENT:
1960                 dprintk("RPC:       xprt %p: socket %s does not exist\n",
1961                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1962                 break;
1963         case -ECONNREFUSED:
1964                 dprintk("RPC:       xprt %p: connection refused for %s\n",
1965                                 xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1966                 break;
1967         default:
1968                 printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1969                                 __func__, -status,
1970                                 xprt->address_strings[RPC_DISPLAY_ADDR]);
1971         }
1972
1973 out:
1974         xprt_clear_connecting(xprt);
1975         xprt_wake_pending_tasks(xprt, status);
1976         current->flags &= ~PF_FSTRANS;
1977         return status;
1978 }
1979
1980 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1981 {
1982         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1983         int ret;
1984
1985          if (RPC_IS_ASYNC(task)) {
1986                 /*
1987                  * We want the AF_LOCAL connect to be resolved in the
1988                  * filesystem namespace of the process making the rpc
1989                  * call.  Thus we connect synchronously.
1990                  *
1991                  * If we want to support asynchronous AF_LOCAL calls,
1992                  * we'll need to figure out how to pass a namespace to
1993                  * connect.
1994                  */
1995                 rpc_exit(task, -ENOTCONN);
1996                 return;
1997         }
1998         ret = xs_local_setup_socket(transport);
1999         if (ret && !RPC_IS_SOFTCONN(task))
2000                 msleep_interruptible(15000);
2001 }
2002
2003 #ifdef CONFIG_SUNRPC_SWAP
2004 static void xs_set_memalloc(struct rpc_xprt *xprt)
2005 {
2006         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2007                         xprt);
2008
2009         if (xprt->swapper)
2010                 sk_set_memalloc(transport->inet);
2011 }
2012
2013 /**
2014  * xs_swapper - Tag this transport as being used for swap.
2015  * @xprt: transport to tag
2016  * @enable: enable/disable
2017  *
2018  */
2019 int xs_swapper(struct rpc_xprt *xprt, int enable)
2020 {
2021         struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2022                         xprt);
2023         int err = 0;
2024
2025         if (enable) {
2026                 xprt->swapper++;
2027                 xs_set_memalloc(xprt);
2028         } else if (xprt->swapper) {
2029                 xprt->swapper--;
2030                 sk_clear_memalloc(transport->inet);
2031         }
2032
2033         return err;
2034 }
2035 EXPORT_SYMBOL_GPL(xs_swapper);
2036 #else
2037 static void xs_set_memalloc(struct rpc_xprt *xprt)
2038 {
2039 }
2040 #endif
2041
2042 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2043 {
2044         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2045
2046         if (!transport->inet) {
2047                 struct sock *sk = sock->sk;
2048
2049                 write_lock_bh(&sk->sk_callback_lock);
2050
2051                 xs_save_old_callbacks(transport, sk);
2052
2053                 sk->sk_user_data = xprt;
2054                 sk->sk_data_ready = xs_udp_data_ready;
2055                 sk->sk_write_space = xs_udp_write_space;
2056                 sk->sk_no_check = UDP_CSUM_NORCV;
2057                 sk->sk_allocation = GFP_ATOMIC;
2058
2059                 xprt_set_connected(xprt);
2060
2061                 /* Reset to new socket */
2062                 transport->sock = sock;
2063                 transport->inet = sk;
2064
2065                 xs_set_memalloc(xprt);
2066
2067                 write_unlock_bh(&sk->sk_callback_lock);
2068         }
2069         xs_udp_do_set_buffer_size(xprt);
2070 }
2071
2072 static void xs_udp_setup_socket(struct work_struct *work)
2073 {
2074         struct sock_xprt *transport =
2075                 container_of(work, struct sock_xprt, connect_worker.work);
2076         struct rpc_xprt *xprt = &transport->xprt;
2077         struct socket *sock = transport->sock;
2078         int status = -EIO;
2079
2080         current->flags |= PF_FSTRANS;
2081
2082         /* Start by resetting any existing state */
2083         xs_reset_transport(transport);
2084         sock = xs_create_sock(xprt, transport,
2085                         xs_addr(xprt)->sa_family, SOCK_DGRAM, IPPROTO_UDP);
2086         if (IS_ERR(sock))
2087                 goto out;
2088
2089         dprintk("RPC:       worker connecting xprt %p via %s to "
2090                                 "%s (port %s)\n", xprt,
2091                         xprt->address_strings[RPC_DISPLAY_PROTO],
2092                         xprt->address_strings[RPC_DISPLAY_ADDR],
2093                         xprt->address_strings[RPC_DISPLAY_PORT]);
2094
2095         xs_udp_finish_connecting(xprt, sock);
2096         trace_rpc_socket_connect(xprt, sock, 0);
2097         status = 0;
2098 out:
2099         xprt_clear_connecting(xprt);
2100         xprt_wake_pending_tasks(xprt, status);
2101         current->flags &= ~PF_FSTRANS;
2102 }
2103
2104 /*
2105  * We need to preserve the port number so the reply cache on the server can
2106  * find our cached RPC replies when we get around to reconnecting.
2107  */
2108 static void xs_abort_connection(struct sock_xprt *transport)
2109 {
2110         int result;
2111         struct sockaddr any;
2112
2113         dprintk("RPC:       disconnecting xprt %p to reuse port\n", transport);
2114
2115         /*
2116          * Disconnect the transport socket by doing a connect operation
2117          * with AF_UNSPEC.  This should return immediately...
2118          */
2119         memset(&any, 0, sizeof(any));
2120         any.sa_family = AF_UNSPEC;
2121         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
2122         trace_rpc_socket_reset_connection(&transport->xprt,
2123                         transport->sock, result);
2124         if (!result)
2125                 xs_sock_reset_connection_flags(&transport->xprt);
2126         dprintk("RPC:       AF_UNSPEC connect return code %d\n", result);
2127 }
2128
2129 static void xs_tcp_reuse_connection(struct sock_xprt *transport)
2130 {
2131         unsigned int state = transport->inet->sk_state;
2132
2133         if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) {
2134                 /* we don't need to abort the connection if the socket
2135                  * hasn't undergone a shutdown
2136                  */
2137                 if (transport->inet->sk_shutdown == 0)
2138                         return;
2139                 dprintk("RPC:       %s: TCP_CLOSEd and sk_shutdown set to %d\n",
2140                                 __func__, transport->inet->sk_shutdown);
2141         }
2142         if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) {
2143                 /* we don't need to abort the connection if the socket
2144                  * hasn't undergone a shutdown
2145                  */
2146                 if (transport->inet->sk_shutdown == 0)
2147                         return;
2148                 dprintk("RPC:       %s: ESTABLISHED/SYN_SENT "
2149                                 "sk_shutdown set to %d\n",
2150                                 __func__, transport->inet->sk_shutdown);
2151         }
2152         xs_abort_connection(transport);
2153 }
2154
2155 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2156 {
2157         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2158         int ret = -ENOTCONN;
2159
2160         if (!transport->inet) {
2161                 struct sock *sk = sock->sk;
2162                 unsigned int keepidle = xprt->timeout->to_initval / HZ;
2163                 unsigned int keepcnt = xprt->timeout->to_retries + 1;
2164                 unsigned int opt_on = 1;
2165
2166                 /* TCP Keepalive options */
2167                 kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2168                                 (char *)&opt_on, sizeof(opt_on));
2169                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2170                                 (char *)&keepidle, sizeof(keepidle));
2171                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2172                                 (char *)&keepidle, sizeof(keepidle));
2173                 kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2174                                 (char *)&keepcnt, sizeof(keepcnt));
2175
2176                 write_lock_bh(&sk->sk_callback_lock);
2177
2178                 xs_save_old_callbacks(transport, sk);
2179
2180                 sk->sk_user_data = xprt;
2181                 sk->sk_data_ready = xs_tcp_data_ready;
2182                 sk->sk_state_change = xs_tcp_state_change;
2183                 sk->sk_write_space = xs_tcp_write_space;
2184                 sk->sk_error_report = xs_error_report;
2185                 sk->sk_allocation = GFP_ATOMIC;
2186
2187                 /* socket options */
2188                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
2189                 sock_reset_flag(sk, SOCK_LINGER);
2190                 tcp_sk(sk)->linger2 = 0;
2191                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2192
2193                 xprt_clear_connected(xprt);
2194
2195                 /* Reset to new socket */
2196                 transport->sock = sock;
2197                 transport->inet = sk;
2198
2199                 write_unlock_bh(&sk->sk_callback_lock);
2200         }
2201
2202         if (!xprt_bound(xprt))
2203                 goto out;
2204
2205         xs_set_memalloc(xprt);
2206
2207         /* Tell the socket layer to start connecting... */
2208         xprt->stat.connect_count++;
2209         xprt->stat.connect_start = jiffies;
2210         ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2211         switch (ret) {
2212         case 0:
2213         case -EINPROGRESS:
2214                 /* SYN_SENT! */
2215                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2216                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2217         }
2218 out:
2219         return ret;
2220 }
2221
2222 /**
2223  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2224  * @xprt: RPC transport to connect
2225  * @transport: socket transport to connect
2226  * @create_sock: function to create a socket of the correct type
2227  *
2228  * Invoked by a work queue tasklet.
2229  */
2230 static void xs_tcp_setup_socket(struct work_struct *work)
2231 {
2232         struct sock_xprt *transport =
2233                 container_of(work, struct sock_xprt, connect_worker.work);
2234         struct socket *sock = transport->sock;
2235         struct rpc_xprt *xprt = &transport->xprt;
2236         int status = -EIO;
2237
2238         current->flags |= PF_FSTRANS;
2239
2240         if (!sock) {
2241                 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
2242                 sock = xs_create_sock(xprt, transport,
2243                                 xs_addr(xprt)->sa_family, SOCK_STREAM, IPPROTO_TCP);
2244                 if (IS_ERR(sock)) {
2245                         status = PTR_ERR(sock);
2246                         goto out;
2247                 }
2248         } else {
2249                 int abort_and_exit;
2250
2251                 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
2252                                 &xprt->state);
2253                 /* "close" the socket, preserving the local port */
2254                 xs_tcp_reuse_connection(transport);
2255
2256                 if (abort_and_exit)
2257                         goto out_eagain;
2258         }
2259
2260         dprintk("RPC:       worker connecting xprt %p via %s to "
2261                                 "%s (port %s)\n", xprt,
2262                         xprt->address_strings[RPC_DISPLAY_PROTO],
2263                         xprt->address_strings[RPC_DISPLAY_ADDR],
2264                         xprt->address_strings[RPC_DISPLAY_PORT]);
2265
2266         status = xs_tcp_finish_connecting(xprt, sock);
2267         trace_rpc_socket_connect(xprt, sock, status);
2268         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2269                         xprt, -status, xprt_connected(xprt),
2270                         sock->sk->sk_state);
2271         switch (status) {
2272         default:
2273                 printk("%s: connect returned unhandled error %d\n",
2274                         __func__, status);
2275         case -EADDRNOTAVAIL:
2276                 /* We're probably in TIME_WAIT. Get rid of existing socket,
2277                  * and retry
2278                  */
2279                 xs_tcp_force_close(xprt);
2280                 break;
2281         case 0:
2282         case -EINPROGRESS:
2283         case -EALREADY:
2284                 xprt_clear_connecting(xprt);
2285                 current->flags &= ~PF_FSTRANS;
2286                 return;
2287         case -EINVAL:
2288                 /* Happens, for instance, if the user specified a link
2289                  * local IPv6 address without a scope-id.
2290                  */
2291         case -ECONNREFUSED:
2292         case -ECONNRESET:
2293         case -ENETUNREACH:
2294                 /* retry with existing socket, after a delay */
2295                 goto out;
2296         }
2297 out_eagain:
2298         status = -EAGAIN;
2299 out:
2300         xprt_clear_connecting(xprt);
2301         xprt_wake_pending_tasks(xprt, status);
2302         current->flags &= ~PF_FSTRANS;
2303 }
2304
2305 /**
2306  * xs_connect - connect a socket to a remote endpoint
2307  * @xprt: pointer to transport structure
2308  * @task: address of RPC task that manages state of connect request
2309  *
2310  * TCP: If the remote end dropped the connection, delay reconnecting.
2311  *
2312  * UDP socket connects are synchronous, but we use a work queue anyway
2313  * to guarantee that even unprivileged user processes can set up a
2314  * socket on a privileged port.
2315  *
2316  * If a UDP socket connect fails, the delay behavior here prevents
2317  * retry floods (hard mounts).
2318  */
2319 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2320 {
2321         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2322
2323         if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
2324                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2325                                 "seconds\n",
2326                                 xprt, xprt->reestablish_timeout / HZ);
2327                 queue_delayed_work(rpciod_workqueue,
2328                                    &transport->connect_worker,
2329                                    xprt->reestablish_timeout);
2330                 xprt->reestablish_timeout <<= 1;
2331                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2332                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2333                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2334                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2335         } else {
2336                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2337                 queue_delayed_work(rpciod_workqueue,
2338                                    &transport->connect_worker, 0);
2339         }
2340 }
2341
2342 /**
2343  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2344  * @xprt: rpc_xprt struct containing statistics
2345  * @seq: output file
2346  *
2347  */
2348 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2349 {
2350         long idle_time = 0;
2351
2352         if (xprt_connected(xprt))
2353                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2354
2355         seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2356                         "%llu %llu %lu %llu %llu\n",
2357                         xprt->stat.bind_count,
2358                         xprt->stat.connect_count,
2359                         xprt->stat.connect_time,
2360                         idle_time,
2361                         xprt->stat.sends,
2362                         xprt->stat.recvs,
2363                         xprt->stat.bad_xids,
2364                         xprt->stat.req_u,
2365                         xprt->stat.bklog_u,
2366                         xprt->stat.max_slots,
2367                         xprt->stat.sending_u,
2368                         xprt->stat.pending_u);
2369 }
2370
2371 /**
2372  * xs_udp_print_stats - display UDP socket-specifc stats
2373  * @xprt: rpc_xprt struct containing statistics
2374  * @seq: output file
2375  *
2376  */
2377 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2378 {
2379         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2380
2381         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2382                         "%lu %llu %llu\n",
2383                         transport->srcport,
2384                         xprt->stat.bind_count,
2385                         xprt->stat.sends,
2386                         xprt->stat.recvs,
2387                         xprt->stat.bad_xids,
2388                         xprt->stat.req_u,
2389                         xprt->stat.bklog_u,
2390                         xprt->stat.max_slots,
2391                         xprt->stat.sending_u,
2392                         xprt->stat.pending_u);
2393 }
2394
2395 /**
2396  * xs_tcp_print_stats - display TCP socket-specifc stats
2397  * @xprt: rpc_xprt struct containing statistics
2398  * @seq: output file
2399  *
2400  */
2401 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2402 {
2403         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2404         long idle_time = 0;
2405
2406         if (xprt_connected(xprt))
2407                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2408
2409         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2410                         "%llu %llu %lu %llu %llu\n",
2411                         transport->srcport,
2412                         xprt->stat.bind_count,
2413                         xprt->stat.connect_count,
2414                         xprt->stat.connect_time,
2415                         idle_time,
2416                         xprt->stat.sends,
2417                         xprt->stat.recvs,
2418                         xprt->stat.bad_xids,
2419                         xprt->stat.req_u,
2420                         xprt->stat.bklog_u,
2421                         xprt->stat.max_slots,
2422                         xprt->stat.sending_u,
2423                         xprt->stat.pending_u);
2424 }
2425
2426 /*
2427  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2428  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2429  * to use the server side send routines.
2430  */
2431 static void *bc_malloc(struct rpc_task *task, size_t size)
2432 {
2433         struct page *page;
2434         struct rpc_buffer *buf;
2435
2436         WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2437         if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2438                 return NULL;
2439
2440         page = alloc_page(GFP_KERNEL);
2441         if (!page)
2442                 return NULL;
2443
2444         buf = page_address(page);
2445         buf->len = PAGE_SIZE;
2446
2447         return buf->data;
2448 }
2449
2450 /*
2451  * Free the space allocated in the bc_alloc routine
2452  */
2453 static void bc_free(void *buffer)
2454 {
2455         struct rpc_buffer *buf;
2456
2457         if (!buffer)
2458                 return;
2459
2460         buf = container_of(buffer, struct rpc_buffer, data);
2461         free_page((unsigned long)buf);
2462 }
2463
2464 /*
2465  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2466  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2467  */
2468 static int bc_sendto(struct rpc_rqst *req)
2469 {
2470         int len;
2471         struct xdr_buf *xbufp = &req->rq_snd_buf;
2472         struct rpc_xprt *xprt = req->rq_xprt;
2473         struct sock_xprt *transport =
2474                                 container_of(xprt, struct sock_xprt, xprt);
2475         struct socket *sock = transport->sock;
2476         unsigned long headoff;
2477         unsigned long tailoff;
2478
2479         xs_encode_stream_record_marker(xbufp);
2480
2481         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2482         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2483         len = svc_send_common(sock, xbufp,
2484                               virt_to_page(xbufp->head[0].iov_base), headoff,
2485                               xbufp->tail[0].iov_base, tailoff);
2486
2487         if (len != xbufp->len) {
2488                 printk(KERN_NOTICE "Error sending entire callback!\n");
2489                 len = -EAGAIN;
2490         }
2491
2492         return len;
2493 }
2494
2495 /*
2496  * The send routine. Borrows from svc_send
2497  */
2498 static int bc_send_request(struct rpc_task *task)
2499 {
2500         struct rpc_rqst *req = task->tk_rqstp;
2501         struct svc_xprt *xprt;
2502         u32                     len;
2503
2504         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2505         /*
2506          * Get the server socket associated with this callback xprt
2507          */
2508         xprt = req->rq_xprt->bc_xprt;
2509
2510         /*
2511          * Grab the mutex to serialize data as the connection is shared
2512          * with the fore channel
2513          */
2514         if (!mutex_trylock(&xprt->xpt_mutex)) {
2515                 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2516                 if (!mutex_trylock(&xprt->xpt_mutex))
2517                         return -EAGAIN;
2518                 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2519         }
2520         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2521                 len = -ENOTCONN;
2522         else
2523                 len = bc_sendto(req);
2524         mutex_unlock(&xprt->xpt_mutex);
2525
2526         if (len > 0)
2527                 len = 0;
2528
2529         return len;
2530 }
2531
2532 /*
2533  * The close routine. Since this is client initiated, we do nothing
2534  */
2535
2536 static void bc_close(struct rpc_xprt *xprt)
2537 {
2538 }
2539
2540 /*
2541  * The xprt destroy routine. Again, because this connection is client
2542  * initiated, we do nothing
2543  */
2544
2545 static void bc_destroy(struct rpc_xprt *xprt)
2546 {
2547 }
2548
2549 static struct rpc_xprt_ops xs_local_ops = {
2550         .reserve_xprt           = xprt_reserve_xprt,
2551         .release_xprt           = xs_tcp_release_xprt,
2552         .alloc_slot             = xprt_alloc_slot,
2553         .rpcbind                = xs_local_rpcbind,
2554         .set_port               = xs_local_set_port,
2555         .connect                = xs_local_connect,
2556         .buf_alloc              = rpc_malloc,
2557         .buf_free               = rpc_free,
2558         .send_request           = xs_local_send_request,
2559         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2560         .close                  = xs_close,
2561         .destroy                = xs_destroy,
2562         .print_stats            = xs_local_print_stats,
2563 };
2564
2565 static struct rpc_xprt_ops xs_udp_ops = {
2566         .set_buffer_size        = xs_udp_set_buffer_size,
2567         .reserve_xprt           = xprt_reserve_xprt_cong,
2568         .release_xprt           = xprt_release_xprt_cong,
2569         .alloc_slot             = xprt_alloc_slot,
2570         .rpcbind                = rpcb_getport_async,
2571         .set_port               = xs_set_port,
2572         .connect                = xs_connect,
2573         .buf_alloc              = rpc_malloc,
2574         .buf_free               = rpc_free,
2575         .send_request           = xs_udp_send_request,
2576         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2577         .timer                  = xs_udp_timer,
2578         .release_request        = xprt_release_rqst_cong,
2579         .close                  = xs_close,
2580         .destroy                = xs_destroy,
2581         .print_stats            = xs_udp_print_stats,
2582 };
2583
2584 static struct rpc_xprt_ops xs_tcp_ops = {
2585         .reserve_xprt           = xprt_reserve_xprt,
2586         .release_xprt           = xs_tcp_release_xprt,
2587         .alloc_slot             = xprt_lock_and_alloc_slot,
2588         .rpcbind                = rpcb_getport_async,
2589         .set_port               = xs_set_port,
2590         .connect                = xs_connect,
2591         .buf_alloc              = rpc_malloc,
2592         .buf_free               = rpc_free,
2593         .send_request           = xs_tcp_send_request,
2594         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2595         .close                  = xs_tcp_close,
2596         .destroy                = xs_destroy,
2597         .print_stats            = xs_tcp_print_stats,
2598 };
2599
2600 /*
2601  * The rpc_xprt_ops for the server backchannel
2602  */
2603
2604 static struct rpc_xprt_ops bc_tcp_ops = {
2605         .reserve_xprt           = xprt_reserve_xprt,
2606         .release_xprt           = xprt_release_xprt,
2607         .alloc_slot             = xprt_alloc_slot,
2608         .buf_alloc              = bc_malloc,
2609         .buf_free               = bc_free,
2610         .send_request           = bc_send_request,
2611         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2612         .close                  = bc_close,
2613         .destroy                = bc_destroy,
2614         .print_stats            = xs_tcp_print_stats,
2615 };
2616
2617 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2618 {
2619         static const struct sockaddr_in sin = {
2620                 .sin_family             = AF_INET,
2621                 .sin_addr.s_addr        = htonl(INADDR_ANY),
2622         };
2623         static const struct sockaddr_in6 sin6 = {
2624                 .sin6_family            = AF_INET6,
2625                 .sin6_addr              = IN6ADDR_ANY_INIT,
2626         };
2627
2628         switch (family) {
2629         case AF_LOCAL:
2630                 break;
2631         case AF_INET:
2632                 memcpy(sap, &sin, sizeof(sin));
2633                 break;
2634         case AF_INET6:
2635                 memcpy(sap, &sin6, sizeof(sin6));
2636                 break;
2637         default:
2638                 dprintk("RPC:       %s: Bad address family\n", __func__);
2639                 return -EAFNOSUPPORT;
2640         }
2641         return 0;
2642 }
2643
2644 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2645                                       unsigned int slot_table_size,
2646                                       unsigned int max_slot_table_size)
2647 {
2648         struct rpc_xprt *xprt;
2649         struct sock_xprt *new;
2650
2651         if (args->addrlen > sizeof(xprt->addr)) {
2652                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2653                 return ERR_PTR(-EBADF);
2654         }
2655
2656         xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2657                         max_slot_table_size);
2658         if (xprt == NULL) {
2659                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2660                                 "rpc_xprt\n");
2661                 return ERR_PTR(-ENOMEM);
2662         }
2663
2664         new = container_of(xprt, struct sock_xprt, xprt);
2665         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2666         xprt->addrlen = args->addrlen;
2667         if (args->srcaddr)
2668                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2669         else {
2670                 int err;
2671                 err = xs_init_anyaddr(args->dstaddr->sa_family,
2672                                         (struct sockaddr *)&new->srcaddr);
2673                 if (err != 0) {
2674                         xprt_free(xprt);
2675                         return ERR_PTR(err);
2676                 }
2677         }
2678
2679         return xprt;
2680 }
2681
2682 static const struct rpc_timeout xs_local_default_timeout = {
2683         .to_initval = 10 * HZ,
2684         .to_maxval = 10 * HZ,
2685         .to_retries = 2,
2686 };
2687
2688 /**
2689  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2690  * @args: rpc transport creation arguments
2691  *
2692  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2693  */
2694 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2695 {
2696         struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2697         struct sock_xprt *transport;
2698         struct rpc_xprt *xprt;
2699         struct rpc_xprt *ret;
2700
2701         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2702                         xprt_max_tcp_slot_table_entries);
2703         if (IS_ERR(xprt))
2704                 return xprt;
2705         transport = container_of(xprt, struct sock_xprt, xprt);
2706
2707         xprt->prot = 0;
2708         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2709         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2710
2711         xprt->bind_timeout = XS_BIND_TO;
2712         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2713         xprt->idle_timeout = XS_IDLE_DISC_TO;
2714
2715         xprt->ops = &xs_local_ops;
2716         xprt->timeout = &xs_local_default_timeout;
2717
2718         INIT_DELAYED_WORK(&transport->connect_worker,
2719                         xs_dummy_setup_socket);
2720
2721         switch (sun->sun_family) {
2722         case AF_LOCAL:
2723                 if (sun->sun_path[0] != '/') {
2724                         dprintk("RPC:       bad AF_LOCAL address: %s\n",
2725                                         sun->sun_path);
2726                         ret = ERR_PTR(-EINVAL);
2727                         goto out_err;
2728                 }
2729                 xprt_set_bound(xprt);
2730                 xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2731                 ret = ERR_PTR(xs_local_setup_socket(transport));
2732                 if (ret)
2733                         goto out_err;
2734                 break;
2735         default:
2736                 ret = ERR_PTR(-EAFNOSUPPORT);
2737                 goto out_err;
2738         }
2739
2740         dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2741                         xprt->address_strings[RPC_DISPLAY_ADDR]);
2742
2743         if (try_module_get(THIS_MODULE))
2744                 return xprt;
2745         ret = ERR_PTR(-EINVAL);
2746 out_err:
2747         xprt_free(xprt);
2748         return ret;
2749 }
2750
2751 static const struct rpc_timeout xs_udp_default_timeout = {
2752         .to_initval = 5 * HZ,
2753         .to_maxval = 30 * HZ,
2754         .to_increment = 5 * HZ,
2755         .to_retries = 5,
2756 };
2757
2758 /**
2759  * xs_setup_udp - Set up transport to use a UDP socket
2760  * @args: rpc transport creation arguments
2761  *
2762  */
2763 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2764 {
2765         struct sockaddr *addr = args->dstaddr;
2766         struct rpc_xprt *xprt;
2767         struct sock_xprt *transport;
2768         struct rpc_xprt *ret;
2769
2770         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2771                         xprt_udp_slot_table_entries);
2772         if (IS_ERR(xprt))
2773                 return xprt;
2774         transport = container_of(xprt, struct sock_xprt, xprt);
2775
2776         xprt->prot = IPPROTO_UDP;
2777         xprt->tsh_size = 0;
2778         /* XXX: header size can vary due to auth type, IPv6, etc. */
2779         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2780
2781         xprt->bind_timeout = XS_BIND_TO;
2782         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2783         xprt->idle_timeout = XS_IDLE_DISC_TO;
2784
2785         xprt->ops = &xs_udp_ops;
2786
2787         xprt->timeout = &xs_udp_default_timeout;
2788
2789         switch (addr->sa_family) {
2790         case AF_INET:
2791                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2792                         xprt_set_bound(xprt);
2793
2794                 INIT_DELAYED_WORK(&transport->connect_worker,
2795                                         xs_udp_setup_socket);
2796                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2797                 break;
2798         case AF_INET6:
2799                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2800                         xprt_set_bound(xprt);
2801
2802                 INIT_DELAYED_WORK(&transport->connect_worker,
2803                                         xs_udp_setup_socket);
2804                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2805                 break;
2806         default:
2807                 ret = ERR_PTR(-EAFNOSUPPORT);
2808                 goto out_err;
2809         }
2810
2811         if (xprt_bound(xprt))
2812                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2813                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2814                                 xprt->address_strings[RPC_DISPLAY_PORT],
2815                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2816         else
2817                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2818                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2819                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2820
2821         if (try_module_get(THIS_MODULE))
2822                 return xprt;
2823         ret = ERR_PTR(-EINVAL);
2824 out_err:
2825         xprt_free(xprt);
2826         return ret;
2827 }
2828
2829 static const struct rpc_timeout xs_tcp_default_timeout = {
2830         .to_initval = 60 * HZ,
2831         .to_maxval = 60 * HZ,
2832         .to_retries = 2,
2833 };
2834
2835 /**
2836  * xs_setup_tcp - Set up transport to use a TCP socket
2837  * @args: rpc transport creation arguments
2838  *
2839  */
2840 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2841 {
2842         struct sockaddr *addr = args->dstaddr;
2843         struct rpc_xprt *xprt;
2844         struct sock_xprt *transport;
2845         struct rpc_xprt *ret;
2846         unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2847
2848         if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2849                 max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2850
2851         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2852                         max_slot_table_size);
2853         if (IS_ERR(xprt))
2854                 return xprt;
2855         transport = container_of(xprt, struct sock_xprt, xprt);
2856
2857         xprt->prot = IPPROTO_TCP;
2858         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2859         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2860
2861         xprt->bind_timeout = XS_BIND_TO;
2862         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2863         xprt->idle_timeout = XS_IDLE_DISC_TO;
2864
2865         xprt->ops = &xs_tcp_ops;
2866         xprt->timeout = &xs_tcp_default_timeout;
2867
2868         switch (addr->sa_family) {
2869         case AF_INET:
2870                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2871                         xprt_set_bound(xprt);
2872
2873                 INIT_DELAYED_WORK(&transport->connect_worker,
2874                                         xs_tcp_setup_socket);
2875                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2876                 break;
2877         case AF_INET6:
2878                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2879                         xprt_set_bound(xprt);
2880
2881                 INIT_DELAYED_WORK(&transport->connect_worker,
2882                                         xs_tcp_setup_socket);
2883                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2884                 break;
2885         default:
2886                 ret = ERR_PTR(-EAFNOSUPPORT);
2887                 goto out_err;
2888         }
2889
2890         if (xprt_bound(xprt))
2891                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2892                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2893                                 xprt->address_strings[RPC_DISPLAY_PORT],
2894                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2895         else
2896                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2897                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2898                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2899
2900
2901         if (try_module_get(THIS_MODULE))
2902                 return xprt;
2903         ret = ERR_PTR(-EINVAL);
2904 out_err:
2905         xprt_free(xprt);
2906         return ret;
2907 }
2908
2909 /**
2910  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2911  * @args: rpc transport creation arguments
2912  *
2913  */
2914 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2915 {
2916         struct sockaddr *addr = args->dstaddr;
2917         struct rpc_xprt *xprt;
2918         struct sock_xprt *transport;
2919         struct svc_sock *bc_sock;
2920         struct rpc_xprt *ret;
2921
2922         if (args->bc_xprt->xpt_bc_xprt) {
2923                 /*
2924                  * This server connection already has a backchannel
2925                  * transport; we can't create a new one, as we wouldn't
2926                  * be able to match replies based on xid any more.  So,
2927                  * reuse the already-existing one:
2928                  */
2929                  return args->bc_xprt->xpt_bc_xprt;
2930         }
2931         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2932                         xprt_tcp_slot_table_entries);
2933         if (IS_ERR(xprt))
2934                 return xprt;
2935         transport = container_of(xprt, struct sock_xprt, xprt);
2936
2937         xprt->prot = IPPROTO_TCP;
2938         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2939         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2940         xprt->timeout = &xs_tcp_default_timeout;
2941
2942         /* backchannel */
2943         xprt_set_bound(xprt);
2944         xprt->bind_timeout = 0;
2945         xprt->reestablish_timeout = 0;
2946         xprt->idle_timeout = 0;
2947
2948         xprt->ops = &bc_tcp_ops;
2949
2950         switch (addr->sa_family) {
2951         case AF_INET:
2952                 xs_format_peer_addresses(xprt, "tcp",
2953                                          RPCBIND_NETID_TCP);
2954                 break;
2955         case AF_INET6:
2956                 xs_format_peer_addresses(xprt, "tcp",
2957                                    RPCBIND_NETID_TCP6);
2958                 break;
2959         default:
2960                 ret = ERR_PTR(-EAFNOSUPPORT);
2961                 goto out_err;
2962         }
2963
2964         dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2965                         xprt->address_strings[RPC_DISPLAY_ADDR],
2966                         xprt->address_strings[RPC_DISPLAY_PORT],
2967                         xprt->address_strings[RPC_DISPLAY_PROTO]);
2968
2969         /*
2970          * Once we've associated a backchannel xprt with a connection,
2971          * we want to keep it around as long as the connection lasts,
2972          * in case we need to start using it for a backchannel again;
2973          * this reference won't be dropped until bc_xprt is destroyed.
2974          */
2975         xprt_get(xprt);
2976         args->bc_xprt->xpt_bc_xprt = xprt;
2977         xprt->bc_xprt = args->bc_xprt;
2978         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2979         transport->sock = bc_sock->sk_sock;
2980         transport->inet = bc_sock->sk_sk;
2981
2982         /*
2983          * Since we don't want connections for the backchannel, we set
2984          * the xprt status to connected
2985          */
2986         xprt_set_connected(xprt);
2987
2988
2989         if (try_module_get(THIS_MODULE))
2990                 return xprt;
2991         xprt_put(xprt);
2992         ret = ERR_PTR(-EINVAL);
2993 out_err:
2994         xprt_free(xprt);
2995         return ret;
2996 }
2997
2998 static struct xprt_class        xs_local_transport = {
2999         .list           = LIST_HEAD_INIT(xs_local_transport.list),
3000         .name           = "named UNIX socket",
3001         .owner          = THIS_MODULE,
3002         .ident          = XPRT_TRANSPORT_LOCAL,
3003         .setup          = xs_setup_local,
3004 };
3005
3006 static struct xprt_class        xs_udp_transport = {
3007         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
3008         .name           = "udp",
3009         .owner          = THIS_MODULE,
3010         .ident          = XPRT_TRANSPORT_UDP,
3011         .setup          = xs_setup_udp,
3012 };
3013
3014 static struct xprt_class        xs_tcp_transport = {
3015         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
3016         .name           = "tcp",
3017         .owner          = THIS_MODULE,
3018         .ident          = XPRT_TRANSPORT_TCP,
3019         .setup          = xs_setup_tcp,
3020 };
3021
3022 static struct xprt_class        xs_bc_tcp_transport = {
3023         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3024         .name           = "tcp NFSv4.1 backchannel",
3025         .owner          = THIS_MODULE,
3026         .ident          = XPRT_TRANSPORT_BC_TCP,
3027         .setup          = xs_setup_bc_tcp,
3028 };
3029
3030 /**
3031  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3032  *
3033  */
3034 int init_socket_xprt(void)
3035 {
3036 #ifdef RPC_DEBUG
3037         if (!sunrpc_table_header)
3038                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
3039 #endif
3040
3041         xprt_register_transport(&xs_local_transport);
3042         xprt_register_transport(&xs_udp_transport);
3043         xprt_register_transport(&xs_tcp_transport);
3044         xprt_register_transport(&xs_bc_tcp_transport);
3045
3046         return 0;
3047 }
3048
3049 /**
3050  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3051  *
3052  */
3053 void cleanup_socket_xprt(void)
3054 {
3055 #ifdef RPC_DEBUG
3056         if (sunrpc_table_header) {
3057                 unregister_sysctl_table(sunrpc_table_header);
3058                 sunrpc_table_header = NULL;
3059         }
3060 #endif
3061
3062         xprt_unregister_transport(&xs_local_transport);
3063         xprt_unregister_transport(&xs_udp_transport);
3064         xprt_unregister_transport(&xs_tcp_transport);
3065         xprt_unregister_transport(&xs_bc_tcp_transport);
3066 }
3067
3068 static int param_set_uint_minmax(const char *val,
3069                 const struct kernel_param *kp,
3070                 unsigned int min, unsigned int max)
3071 {
3072         unsigned long num;
3073         int ret;
3074
3075         if (!val)
3076                 return -EINVAL;
3077         ret = strict_strtoul(val, 0, &num);
3078         if (ret == -EINVAL || num < min || num > max)
3079                 return -EINVAL;
3080         *((unsigned int *)kp->arg) = num;
3081         return 0;
3082 }
3083
3084 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3085 {
3086         return param_set_uint_minmax(val, kp,
3087                         RPC_MIN_RESVPORT,
3088                         RPC_MAX_RESVPORT);
3089 }
3090
3091 static struct kernel_param_ops param_ops_portnr = {
3092         .set = param_set_portnr,
3093         .get = param_get_uint,
3094 };
3095
3096 #define param_check_portnr(name, p) \
3097         __param_check(name, p, unsigned int);
3098
3099 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3100 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3101
3102 static int param_set_slot_table_size(const char *val,
3103                                      const struct kernel_param *kp)
3104 {
3105         return param_set_uint_minmax(val, kp,
3106                         RPC_MIN_SLOT_TABLE,
3107                         RPC_MAX_SLOT_TABLE);
3108 }
3109
3110 static struct kernel_param_ops param_ops_slot_table_size = {
3111         .set = param_set_slot_table_size,
3112         .get = param_get_uint,
3113 };
3114
3115 #define param_check_slot_table_size(name, p) \
3116         __param_check(name, p, unsigned int);
3117
3118 static int param_set_max_slot_table_size(const char *val,
3119                                      const struct kernel_param *kp)
3120 {
3121         return param_set_uint_minmax(val, kp,
3122                         RPC_MIN_SLOT_TABLE,
3123                         RPC_MAX_SLOT_TABLE_LIMIT);
3124 }
3125
3126 static struct kernel_param_ops param_ops_max_slot_table_size = {
3127         .set = param_set_max_slot_table_size,
3128         .get = param_get_uint,
3129 };
3130
3131 #define param_check_max_slot_table_size(name, p) \
3132         __param_check(name, p, unsigned int);
3133
3134 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3135                    slot_table_size, 0644);
3136 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3137                    max_slot_table_size, 0644);
3138 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3139                    slot_table_size, 0644);
3140