net/sunrpc/xprtsock.c

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