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